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> |
e34a98d5 | 57 | #include <linux/fs_parser.h> |
d4ff749b | 58 | #include <linux/sched/cputime.h> |
2ce7135a | 59 | #include <linux/psi.h> |
bd1060a1 | 60 | #include <net/sock.h> |
ddbcc7e8 | 61 | |
ed1777de TH |
62 | #define CREATE_TRACE_POINTS |
63 | #include <trace/events/cgroup.h> | |
64 | ||
8d7e6fb0 TH |
65 | #define CGROUP_FILE_NAME_MAX (MAX_CGROUP_TYPE_NAMELEN + \ |
66 | MAX_CFTYPE_NAME + 2) | |
b12e3583 TH |
67 | /* let's not notify more than 100 times per second */ |
68 | #define CGROUP_FILE_NOTIFY_MIN_INTV DIV_ROUND_UP(HZ, 100) | |
8d7e6fb0 | 69 | |
e25e2cbb TH |
70 | /* |
71 | * cgroup_mutex is the master lock. Any modification to cgroup or its | |
72 | * hierarchy must be performed while holding it. | |
73 | * | |
f0d9a5f1 | 74 | * css_set_lock protects task->cgroups pointer, the list of css_set |
0e1d768f | 75 | * objects, and the chain of tasks off each css_set. |
e25e2cbb | 76 | * |
0e1d768f TH |
77 | * These locks are exported if CONFIG_PROVE_RCU so that accessors in |
78 | * cgroup.h can use them for lockdep annotations. | |
e25e2cbb | 79 | */ |
2219449a | 80 | DEFINE_MUTEX(cgroup_mutex); |
f0d9a5f1 | 81 | DEFINE_SPINLOCK(css_set_lock); |
0a268dbd TH |
82 | |
83 | #ifdef CONFIG_PROVE_RCU | |
0e1d768f | 84 | EXPORT_SYMBOL_GPL(cgroup_mutex); |
f0d9a5f1 | 85 | EXPORT_SYMBOL_GPL(css_set_lock); |
2219449a TH |
86 | #endif |
87 | ||
e4f8d81c SRV |
88 | DEFINE_SPINLOCK(trace_cgroup_path_lock); |
89 | char trace_cgroup_path[TRACE_CGROUP_PATH_LEN]; | |
5cf8114d | 90 | bool cgroup_debug __read_mostly; |
e4f8d81c | 91 | |
6fa4918d | 92 | /* |
15a4c835 TH |
93 | * Protects cgroup_idr and css_idr so that IDs can be released without |
94 | * grabbing cgroup_mutex. | |
6fa4918d TH |
95 | */ |
96 | static DEFINE_SPINLOCK(cgroup_idr_lock); | |
97 | ||
34c06254 TH |
98 | /* |
99 | * Protects cgroup_file->kn for !self csses. It synchronizes notifications | |
100 | * against file removal/re-creation across css hiding. | |
101 | */ | |
102 | static DEFINE_SPINLOCK(cgroup_file_kn_lock); | |
103 | ||
3f2947b7 | 104 | DEFINE_PERCPU_RWSEM(cgroup_threadgroup_rwsem); |
1ed13287 | 105 | |
8353da1f | 106 | #define cgroup_assert_mutex_or_rcu_locked() \ |
f78f5b90 PM |
107 | RCU_LOCKDEP_WARN(!rcu_read_lock_held() && \ |
108 | !lockdep_is_held(&cgroup_mutex), \ | |
8353da1f | 109 | "cgroup_mutex or RCU read lock required"); |
780cd8b3 | 110 | |
e5fca243 TH |
111 | /* |
112 | * cgroup destruction makes heavy use of work items and there can be a lot | |
113 | * of concurrent destructions. Use a separate workqueue so that cgroup | |
114 | * destruction work items don't end up filling up max_active of system_wq | |
115 | * which may lead to deadlock. | |
116 | */ | |
117 | static struct workqueue_struct *cgroup_destroy_wq; | |
118 | ||
3ed80a62 | 119 | /* generate an array of cgroup subsystem pointers */ |
073219e9 | 120 | #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys, |
0a268dbd | 121 | struct cgroup_subsys *cgroup_subsys[] = { |
ddbcc7e8 PM |
122 | #include <linux/cgroup_subsys.h> |
123 | }; | |
073219e9 TH |
124 | #undef SUBSYS |
125 | ||
126 | /* array of cgroup subsystem names */ | |
127 | #define SUBSYS(_x) [_x ## _cgrp_id] = #_x, | |
128 | static const char *cgroup_subsys_name[] = { | |
ddbcc7e8 PM |
129 | #include <linux/cgroup_subsys.h> |
130 | }; | |
073219e9 | 131 | #undef SUBSYS |
ddbcc7e8 | 132 | |
49d1dc4b TH |
133 | /* array of static_keys for cgroup_subsys_enabled() and cgroup_subsys_on_dfl() */ |
134 | #define SUBSYS(_x) \ | |
135 | DEFINE_STATIC_KEY_TRUE(_x ## _cgrp_subsys_enabled_key); \ | |
136 | DEFINE_STATIC_KEY_TRUE(_x ## _cgrp_subsys_on_dfl_key); \ | |
137 | EXPORT_SYMBOL_GPL(_x ## _cgrp_subsys_enabled_key); \ | |
138 | EXPORT_SYMBOL_GPL(_x ## _cgrp_subsys_on_dfl_key); | |
139 | #include <linux/cgroup_subsys.h> | |
140 | #undef SUBSYS | |
141 | ||
142 | #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys_enabled_key, | |
143 | static struct static_key_true *cgroup_subsys_enabled_key[] = { | |
144 | #include <linux/cgroup_subsys.h> | |
145 | }; | |
146 | #undef SUBSYS | |
147 | ||
148 | #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys_on_dfl_key, | |
149 | static struct static_key_true *cgroup_subsys_on_dfl_key[] = { | |
150 | #include <linux/cgroup_subsys.h> | |
151 | }; | |
152 | #undef SUBSYS | |
153 | ||
c58632b3 | 154 | static DEFINE_PER_CPU(struct cgroup_rstat_cpu, cgrp_dfl_root_rstat_cpu); |
041cd640 | 155 | |
ddbcc7e8 | 156 | /* |
3dd06ffa | 157 | * The default hierarchy, reserved for the subsystems that are otherwise |
9871bf95 TH |
158 | * unattached - it never has more than a single cgroup, and all tasks are |
159 | * part of that cgroup. | |
ddbcc7e8 | 160 | */ |
c58632b3 | 161 | struct cgroup_root cgrp_dfl_root = { .cgrp.rstat_cpu = &cgrp_dfl_root_rstat_cpu }; |
d0ec4230 | 162 | EXPORT_SYMBOL_GPL(cgrp_dfl_root); |
9871bf95 | 163 | |
a2dd4247 TH |
164 | /* |
165 | * The default hierarchy always exists but is hidden until mounted for the | |
166 | * first time. This is for backward compatibility. | |
167 | */ | |
a7165264 | 168 | static bool cgrp_dfl_visible; |
ddbcc7e8 | 169 | |
5533e011 | 170 | /* some controllers are not supported in the default hierarchy */ |
a7165264 | 171 | static u16 cgrp_dfl_inhibit_ss_mask; |
5533e011 | 172 | |
f6d635ad | 173 | /* some controllers are implicitly enabled on the default hierarchy */ |
b807421a | 174 | static u16 cgrp_dfl_implicit_ss_mask; |
f6d635ad | 175 | |
8cfd8147 TH |
176 | /* some controllers can be threaded on the default hierarchy */ |
177 | static u16 cgrp_dfl_threaded_ss_mask; | |
178 | ||
ddbcc7e8 | 179 | /* The list of hierarchy roots */ |
0a268dbd | 180 | LIST_HEAD(cgroup_roots); |
9871bf95 | 181 | static int cgroup_root_count; |
ddbcc7e8 | 182 | |
3417ae1f | 183 | /* hierarchy ID allocation and mapping, protected by cgroup_mutex */ |
1a574231 | 184 | static DEFINE_IDR(cgroup_hierarchy_idr); |
2c6ab6d2 | 185 | |
794611a1 | 186 | /* |
0cb51d71 TH |
187 | * Assign a monotonically increasing serial number to csses. It guarantees |
188 | * cgroups with bigger numbers are newer than those with smaller numbers. | |
189 | * Also, as csses are always appended to the parent's ->children list, it | |
190 | * guarantees that sibling csses are always sorted in the ascending serial | |
191 | * number order on the list. Protected by cgroup_mutex. | |
794611a1 | 192 | */ |
0cb51d71 | 193 | static u64 css_serial_nr_next = 1; |
794611a1 | 194 | |
cb4a3167 | 195 | /* |
b807421a TH |
196 | * These bitmasks identify subsystems with specific features to avoid |
197 | * having to do iterative checks repeatedly. | |
ddbcc7e8 | 198 | */ |
6e5c8307 TH |
199 | static u16 have_fork_callback __read_mostly; |
200 | static u16 have_exit_callback __read_mostly; | |
51bee5ab | 201 | static u16 have_release_callback __read_mostly; |
b807421a | 202 | static u16 have_canfork_callback __read_mostly; |
ddbcc7e8 | 203 | |
a79a908f AK |
204 | /* cgroup namespace for init task */ |
205 | struct cgroup_namespace init_cgroup_ns = { | |
387ad967 | 206 | .count = REFCOUNT_INIT(2), |
a79a908f AK |
207 | .user_ns = &init_user_ns, |
208 | .ns.ops = &cgroupns_operations, | |
209 | .ns.inum = PROC_CGROUP_INIT_INO, | |
210 | .root_cset = &init_css_set, | |
211 | }; | |
212 | ||
67e9c74b | 213 | static struct file_system_type cgroup2_fs_type; |
d62beb7f | 214 | static struct cftype cgroup_base_files[]; |
628f7cd4 | 215 | |
334c3679 TH |
216 | static int cgroup_apply_control(struct cgroup *cgrp); |
217 | static void cgroup_finalize_control(struct cgroup *cgrp, int ret); | |
b636fd38 TH |
218 | static void css_task_iter_skip(struct css_task_iter *it, |
219 | struct task_struct *task); | |
42809dd4 | 220 | static int cgroup_destroy_locked(struct cgroup *cgrp); |
6cd0f5bb TH |
221 | static struct cgroup_subsys_state *css_create(struct cgroup *cgrp, |
222 | struct cgroup_subsys *ss); | |
9d755d33 | 223 | static void css_release(struct percpu_ref *ref); |
f8f22e53 | 224 | static void kill_css(struct cgroup_subsys_state *css); |
4df8dc90 TH |
225 | static int cgroup_addrm_files(struct cgroup_subsys_state *css, |
226 | struct cgroup *cgrp, struct cftype cfts[], | |
2bb566cb | 227 | bool is_add); |
42809dd4 | 228 | |
fc5ed1e9 TH |
229 | /** |
230 | * cgroup_ssid_enabled - cgroup subsys enabled test by subsys ID | |
231 | * @ssid: subsys ID of interest | |
232 | * | |
233 | * cgroup_subsys_enabled() can only be used with literal subsys names which | |
234 | * is fine for individual subsystems but unsuitable for cgroup core. This | |
235 | * is slower static_key_enabled() based test indexed by @ssid. | |
236 | */ | |
0a268dbd | 237 | bool cgroup_ssid_enabled(int ssid) |
fc5ed1e9 | 238 | { |
cfe02a8a AB |
239 | if (CGROUP_SUBSYS_COUNT == 0) |
240 | return false; | |
241 | ||
fc5ed1e9 TH |
242 | return static_key_enabled(cgroup_subsys_enabled_key[ssid]); |
243 | } | |
244 | ||
9e10a130 TH |
245 | /** |
246 | * cgroup_on_dfl - test whether a cgroup is on the default hierarchy | |
247 | * @cgrp: the cgroup of interest | |
248 | * | |
249 | * The default hierarchy is the v2 interface of cgroup and this function | |
250 | * can be used to test whether a cgroup is on the default hierarchy for | |
251 | * cases where a subsystem should behave differnetly depending on the | |
252 | * interface version. | |
253 | * | |
254 | * The set of behaviors which change on the default hierarchy are still | |
255 | * being determined and the mount option is prefixed with __DEVEL__. | |
256 | * | |
257 | * List of changed behaviors: | |
258 | * | |
259 | * - Mount options "noprefix", "xattr", "clone_children", "release_agent" | |
260 | * and "name" are disallowed. | |
261 | * | |
262 | * - When mounting an existing superblock, mount options should match. | |
263 | * | |
264 | * - Remount is disallowed. | |
265 | * | |
266 | * - rename(2) is disallowed. | |
267 | * | |
268 | * - "tasks" is removed. Everything should be at process granularity. Use | |
269 | * "cgroup.procs" instead. | |
270 | * | |
271 | * - "cgroup.procs" is not sorted. pids will be unique unless they got | |
272 | * recycled inbetween reads. | |
273 | * | |
274 | * - "release_agent" and "notify_on_release" are removed. Replacement | |
275 | * notification mechanism will be implemented. | |
276 | * | |
277 | * - "cgroup.clone_children" is removed. | |
278 | * | |
279 | * - "cgroup.subtree_populated" is available. Its value is 0 if the cgroup | |
280 | * and its descendants contain no task; otherwise, 1. The file also | |
281 | * generates kernfs notification which can be monitored through poll and | |
282 | * [di]notify when the value of the file changes. | |
283 | * | |
284 | * - cpuset: tasks will be kept in empty cpusets when hotplug happens and | |
285 | * take masks of ancestors with non-empty cpus/mems, instead of being | |
286 | * moved to an ancestor. | |
287 | * | |
288 | * - cpuset: a task can be moved into an empty cpuset, and again it takes | |
289 | * masks of ancestors. | |
290 | * | |
291 | * - memcg: use_hierarchy is on by default and the cgroup file for the flag | |
292 | * is not created. | |
293 | * | |
294 | * - blkcg: blk-throttle becomes properly hierarchical. | |
295 | * | |
296 | * - debug: disallowed on the default hierarchy. | |
297 | */ | |
0a268dbd | 298 | bool cgroup_on_dfl(const struct cgroup *cgrp) |
9e10a130 TH |
299 | { |
300 | return cgrp->root == &cgrp_dfl_root; | |
301 | } | |
302 | ||
6fa4918d TH |
303 | /* IDR wrappers which synchronize using cgroup_idr_lock */ |
304 | static int cgroup_idr_alloc(struct idr *idr, void *ptr, int start, int end, | |
305 | gfp_t gfp_mask) | |
306 | { | |
307 | int ret; | |
308 | ||
309 | idr_preload(gfp_mask); | |
54504e97 | 310 | spin_lock_bh(&cgroup_idr_lock); |
d0164adc | 311 | ret = idr_alloc(idr, ptr, start, end, gfp_mask & ~__GFP_DIRECT_RECLAIM); |
54504e97 | 312 | spin_unlock_bh(&cgroup_idr_lock); |
6fa4918d TH |
313 | idr_preload_end(); |
314 | return ret; | |
315 | } | |
316 | ||
317 | static void *cgroup_idr_replace(struct idr *idr, void *ptr, int id) | |
318 | { | |
319 | void *ret; | |
320 | ||
54504e97 | 321 | spin_lock_bh(&cgroup_idr_lock); |
6fa4918d | 322 | ret = idr_replace(idr, ptr, id); |
54504e97 | 323 | spin_unlock_bh(&cgroup_idr_lock); |
6fa4918d TH |
324 | return ret; |
325 | } | |
326 | ||
327 | static void cgroup_idr_remove(struct idr *idr, int id) | |
328 | { | |
54504e97 | 329 | spin_lock_bh(&cgroup_idr_lock); |
6fa4918d | 330 | idr_remove(idr, id); |
54504e97 | 331 | spin_unlock_bh(&cgroup_idr_lock); |
6fa4918d TH |
332 | } |
333 | ||
27f26753 | 334 | static bool cgroup_has_tasks(struct cgroup *cgrp) |
d51f39b0 | 335 | { |
27f26753 TH |
336 | return cgrp->nr_populated_csets; |
337 | } | |
d51f39b0 | 338 | |
7a0cf0e7 | 339 | bool cgroup_is_threaded(struct cgroup *cgrp) |
454000ad TH |
340 | { |
341 | return cgrp->dom_cgrp != cgrp; | |
342 | } | |
343 | ||
8cfd8147 TH |
344 | /* can @cgrp host both domain and threaded children? */ |
345 | static bool cgroup_is_mixable(struct cgroup *cgrp) | |
346 | { | |
347 | /* | |
348 | * Root isn't under domain level resource control exempting it from | |
349 | * the no-internal-process constraint, so it can serve as a thread | |
350 | * root and a parent of resource domains at the same time. | |
351 | */ | |
352 | return !cgroup_parent(cgrp); | |
353 | } | |
354 | ||
355 | /* can @cgrp become a thread root? should always be true for a thread root */ | |
356 | static bool cgroup_can_be_thread_root(struct cgroup *cgrp) | |
357 | { | |
358 | /* mixables don't care */ | |
359 | if (cgroup_is_mixable(cgrp)) | |
360 | return true; | |
361 | ||
362 | /* domain roots can't be nested under threaded */ | |
363 | if (cgroup_is_threaded(cgrp)) | |
364 | return false; | |
365 | ||
366 | /* can only have either domain or threaded children */ | |
367 | if (cgrp->nr_populated_domain_children) | |
368 | return false; | |
369 | ||
370 | /* and no domain controllers can be enabled */ | |
371 | if (cgrp->subtree_control & ~cgrp_dfl_threaded_ss_mask) | |
372 | return false; | |
373 | ||
374 | return true; | |
375 | } | |
376 | ||
377 | /* is @cgrp root of a threaded subtree? */ | |
7a0cf0e7 | 378 | bool cgroup_is_thread_root(struct cgroup *cgrp) |
8cfd8147 TH |
379 | { |
380 | /* thread root should be a domain */ | |
381 | if (cgroup_is_threaded(cgrp)) | |
382 | return false; | |
383 | ||
384 | /* a domain w/ threaded children is a thread root */ | |
385 | if (cgrp->nr_threaded_children) | |
386 | return true; | |
387 | ||
388 | /* | |
389 | * A domain which has tasks and explicit threaded controllers | |
390 | * enabled is a thread root. | |
391 | */ | |
392 | if (cgroup_has_tasks(cgrp) && | |
393 | (cgrp->subtree_control & cgrp_dfl_threaded_ss_mask)) | |
394 | return true; | |
395 | ||
396 | return false; | |
397 | } | |
398 | ||
399 | /* a domain which isn't connected to the root w/o brekage can't be used */ | |
400 | static bool cgroup_is_valid_domain(struct cgroup *cgrp) | |
401 | { | |
402 | /* the cgroup itself can be a thread root */ | |
403 | if (cgroup_is_threaded(cgrp)) | |
404 | return false; | |
405 | ||
406 | /* but the ancestors can't be unless mixable */ | |
407 | while ((cgrp = cgroup_parent(cgrp))) { | |
408 | if (!cgroup_is_mixable(cgrp) && cgroup_is_thread_root(cgrp)) | |
409 | return false; | |
410 | if (cgroup_is_threaded(cgrp)) | |
411 | return false; | |
412 | } | |
413 | ||
414 | return true; | |
d51f39b0 TH |
415 | } |
416 | ||
5531dc91 TH |
417 | /* subsystems visibly enabled on a cgroup */ |
418 | static u16 cgroup_control(struct cgroup *cgrp) | |
419 | { | |
420 | struct cgroup *parent = cgroup_parent(cgrp); | |
421 | u16 root_ss_mask = cgrp->root->subsys_mask; | |
422 | ||
8cfd8147 TH |
423 | if (parent) { |
424 | u16 ss_mask = parent->subtree_control; | |
425 | ||
426 | /* threaded cgroups can only have threaded controllers */ | |
427 | if (cgroup_is_threaded(cgrp)) | |
428 | ss_mask &= cgrp_dfl_threaded_ss_mask; | |
429 | return ss_mask; | |
430 | } | |
5531dc91 TH |
431 | |
432 | if (cgroup_on_dfl(cgrp)) | |
f6d635ad TH |
433 | root_ss_mask &= ~(cgrp_dfl_inhibit_ss_mask | |
434 | cgrp_dfl_implicit_ss_mask); | |
5531dc91 TH |
435 | return root_ss_mask; |
436 | } | |
437 | ||
438 | /* subsystems enabled on a cgroup */ | |
439 | static u16 cgroup_ss_mask(struct cgroup *cgrp) | |
440 | { | |
441 | struct cgroup *parent = cgroup_parent(cgrp); | |
442 | ||
8cfd8147 TH |
443 | if (parent) { |
444 | u16 ss_mask = parent->subtree_ss_mask; | |
445 | ||
446 | /* threaded cgroups can only have threaded controllers */ | |
447 | if (cgroup_is_threaded(cgrp)) | |
448 | ss_mask &= cgrp_dfl_threaded_ss_mask; | |
449 | return ss_mask; | |
450 | } | |
5531dc91 TH |
451 | |
452 | return cgrp->root->subsys_mask; | |
453 | } | |
454 | ||
95109b62 TH |
455 | /** |
456 | * cgroup_css - obtain a cgroup's css for the specified subsystem | |
457 | * @cgrp: the cgroup of interest | |
9d800df1 | 458 | * @ss: the subsystem of interest (%NULL returns @cgrp->self) |
95109b62 | 459 | * |
ca8bdcaf TH |
460 | * Return @cgrp's css (cgroup_subsys_state) associated with @ss. This |
461 | * function must be called either under cgroup_mutex or rcu_read_lock() and | |
462 | * the caller is responsible for pinning the returned css if it wants to | |
463 | * keep accessing it outside the said locks. This function may return | |
464 | * %NULL if @cgrp doesn't have @subsys_id enabled. | |
95109b62 TH |
465 | */ |
466 | static struct cgroup_subsys_state *cgroup_css(struct cgroup *cgrp, | |
ca8bdcaf | 467 | struct cgroup_subsys *ss) |
95109b62 | 468 | { |
ca8bdcaf | 469 | if (ss) |
aec25020 | 470 | return rcu_dereference_check(cgrp->subsys[ss->id], |
ace2bee8 | 471 | lockdep_is_held(&cgroup_mutex)); |
ca8bdcaf | 472 | else |
9d800df1 | 473 | return &cgrp->self; |
95109b62 | 474 | } |
42809dd4 | 475 | |
d41bf8c9 TH |
476 | /** |
477 | * cgroup_tryget_css - try to get a cgroup's css for the specified subsystem | |
478 | * @cgrp: the cgroup of interest | |
479 | * @ss: the subsystem of interest | |
480 | * | |
481 | * Find and get @cgrp's css assocaited with @ss. If the css doesn't exist | |
482 | * or is offline, %NULL is returned. | |
483 | */ | |
484 | static struct cgroup_subsys_state *cgroup_tryget_css(struct cgroup *cgrp, | |
485 | struct cgroup_subsys *ss) | |
486 | { | |
487 | struct cgroup_subsys_state *css; | |
488 | ||
489 | rcu_read_lock(); | |
490 | css = cgroup_css(cgrp, ss); | |
a581563f | 491 | if (css && !css_tryget_online(css)) |
d41bf8c9 TH |
492 | css = NULL; |
493 | rcu_read_unlock(); | |
494 | ||
495 | return css; | |
496 | } | |
497 | ||
aec3dfcb | 498 | /** |
fc5a828b | 499 | * cgroup_e_css_by_mask - obtain a cgroup's effective css for the specified ss |
aec3dfcb | 500 | * @cgrp: the cgroup of interest |
9d800df1 | 501 | * @ss: the subsystem of interest (%NULL returns @cgrp->self) |
aec3dfcb | 502 | * |
d0f702e6 | 503 | * Similar to cgroup_css() but returns the effective css, which is defined |
aec3dfcb TH |
504 | * as the matching css of the nearest ancestor including self which has @ss |
505 | * enabled. If @ss is associated with the hierarchy @cgrp is on, this | |
506 | * function is guaranteed to return non-NULL css. | |
507 | */ | |
fc5a828b DZ |
508 | static struct cgroup_subsys_state *cgroup_e_css_by_mask(struct cgroup *cgrp, |
509 | struct cgroup_subsys *ss) | |
aec3dfcb TH |
510 | { |
511 | lockdep_assert_held(&cgroup_mutex); | |
512 | ||
513 | if (!ss) | |
9d800df1 | 514 | return &cgrp->self; |
aec3dfcb | 515 | |
eeecbd19 TH |
516 | /* |
517 | * This function is used while updating css associations and thus | |
5531dc91 | 518 | * can't test the csses directly. Test ss_mask. |
eeecbd19 | 519 | */ |
5531dc91 | 520 | while (!(cgroup_ss_mask(cgrp) & (1 << ss->id))) { |
d51f39b0 | 521 | cgrp = cgroup_parent(cgrp); |
5531dc91 TH |
522 | if (!cgrp) |
523 | return NULL; | |
524 | } | |
aec3dfcb TH |
525 | |
526 | return cgroup_css(cgrp, ss); | |
95109b62 | 527 | } |
42809dd4 | 528 | |
fc5a828b DZ |
529 | /** |
530 | * cgroup_e_css - obtain a cgroup's effective css for the specified subsystem | |
531 | * @cgrp: the cgroup of interest | |
532 | * @ss: the subsystem of interest | |
533 | * | |
534 | * Find and get the effective css of @cgrp for @ss. The effective css is | |
535 | * defined as the matching css of the nearest ancestor including self which | |
536 | * has @ss enabled. If @ss is not mounted on the hierarchy @cgrp is on, | |
537 | * the root css is returned, so this function always returns a valid css. | |
538 | * | |
539 | * The returned css is not guaranteed to be online, and therefore it is the | |
540 | * callers responsiblity to tryget a reference for it. | |
541 | */ | |
542 | struct cgroup_subsys_state *cgroup_e_css(struct cgroup *cgrp, | |
543 | struct cgroup_subsys *ss) | |
544 | { | |
545 | struct cgroup_subsys_state *css; | |
546 | ||
547 | do { | |
548 | css = cgroup_css(cgrp, ss); | |
549 | ||
550 | if (css) | |
551 | return css; | |
552 | cgrp = cgroup_parent(cgrp); | |
553 | } while (cgrp); | |
554 | ||
555 | return init_css_set.subsys[ss->id]; | |
556 | } | |
557 | ||
eeecbd19 TH |
558 | /** |
559 | * cgroup_get_e_css - get a cgroup's effective css for the specified subsystem | |
560 | * @cgrp: the cgroup of interest | |
561 | * @ss: the subsystem of interest | |
562 | * | |
563 | * Find and get the effective css of @cgrp for @ss. The effective css is | |
564 | * defined as the matching css of the nearest ancestor including self which | |
565 | * has @ss enabled. If @ss is not mounted on the hierarchy @cgrp is on, | |
566 | * the root css is returned, so this function always returns a valid css. | |
567 | * The returned css must be put using css_put(). | |
568 | */ | |
569 | struct cgroup_subsys_state *cgroup_get_e_css(struct cgroup *cgrp, | |
570 | struct cgroup_subsys *ss) | |
571 | { | |
572 | struct cgroup_subsys_state *css; | |
573 | ||
574 | rcu_read_lock(); | |
575 | ||
576 | do { | |
577 | css = cgroup_css(cgrp, ss); | |
578 | ||
579 | if (css && css_tryget_online(css)) | |
580 | goto out_unlock; | |
581 | cgrp = cgroup_parent(cgrp); | |
582 | } while (cgrp); | |
583 | ||
584 | css = init_css_set.subsys[ss->id]; | |
585 | css_get(css); | |
586 | out_unlock: | |
587 | rcu_read_unlock(); | |
588 | return css; | |
589 | } | |
590 | ||
a590b90d | 591 | static void cgroup_get_live(struct cgroup *cgrp) |
052c3f3a TH |
592 | { |
593 | WARN_ON_ONCE(cgroup_is_dead(cgrp)); | |
594 | css_get(&cgrp->self); | |
595 | } | |
596 | ||
aade7f9e RG |
597 | /** |
598 | * __cgroup_task_count - count the number of tasks in a cgroup. The caller | |
599 | * is responsible for taking the css_set_lock. | |
600 | * @cgrp: the cgroup in question | |
601 | */ | |
602 | int __cgroup_task_count(const struct cgroup *cgrp) | |
603 | { | |
604 | int count = 0; | |
605 | struct cgrp_cset_link *link; | |
606 | ||
607 | lockdep_assert_held(&css_set_lock); | |
608 | ||
609 | list_for_each_entry(link, &cgrp->cset_links, cset_link) | |
610 | count += link->cset->nr_tasks; | |
611 | ||
612 | return count; | |
613 | } | |
614 | ||
615 | /** | |
616 | * cgroup_task_count - count the number of tasks in a cgroup. | |
617 | * @cgrp: the cgroup in question | |
618 | */ | |
619 | int cgroup_task_count(const struct cgroup *cgrp) | |
620 | { | |
621 | int count; | |
622 | ||
623 | spin_lock_irq(&css_set_lock); | |
624 | count = __cgroup_task_count(cgrp); | |
625 | spin_unlock_irq(&css_set_lock); | |
626 | ||
627 | return count; | |
628 | } | |
629 | ||
b4168640 | 630 | struct cgroup_subsys_state *of_css(struct kernfs_open_file *of) |
59f5296b | 631 | { |
2bd59d48 | 632 | struct cgroup *cgrp = of->kn->parent->priv; |
b4168640 | 633 | struct cftype *cft = of_cft(of); |
2bd59d48 TH |
634 | |
635 | /* | |
636 | * This is open and unprotected implementation of cgroup_css(). | |
637 | * seq_css() is only called from a kernfs file operation which has | |
638 | * an active reference on the file. Because all the subsystem | |
639 | * files are drained before a css is disassociated with a cgroup, | |
640 | * the matching css from the cgroup's subsys table is guaranteed to | |
641 | * be and stay valid until the enclosing operation is complete. | |
642 | */ | |
643 | if (cft->ss) | |
644 | return rcu_dereference_raw(cgrp->subsys[cft->ss->id]); | |
645 | else | |
9d800df1 | 646 | return &cgrp->self; |
59f5296b | 647 | } |
b4168640 | 648 | EXPORT_SYMBOL_GPL(of_css); |
59f5296b | 649 | |
1c6727af TH |
650 | /** |
651 | * for_each_css - iterate all css's of a cgroup | |
652 | * @css: the iteration cursor | |
653 | * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end | |
654 | * @cgrp: the target cgroup to iterate css's of | |
655 | * | |
aec3dfcb | 656 | * Should be called under cgroup_[tree_]mutex. |
1c6727af TH |
657 | */ |
658 | #define for_each_css(css, ssid, cgrp) \ | |
659 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \ | |
660 | if (!((css) = rcu_dereference_check( \ | |
661 | (cgrp)->subsys[(ssid)], \ | |
662 | lockdep_is_held(&cgroup_mutex)))) { } \ | |
663 | else | |
664 | ||
aec3dfcb TH |
665 | /** |
666 | * for_each_e_css - iterate all effective css's of a cgroup | |
667 | * @css: the iteration cursor | |
668 | * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end | |
669 | * @cgrp: the target cgroup to iterate css's of | |
670 | * | |
671 | * Should be called under cgroup_[tree_]mutex. | |
672 | */ | |
fc5a828b DZ |
673 | #define for_each_e_css(css, ssid, cgrp) \ |
674 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \ | |
675 | if (!((css) = cgroup_e_css_by_mask(cgrp, \ | |
676 | cgroup_subsys[(ssid)]))) \ | |
677 | ; \ | |
aec3dfcb TH |
678 | else |
679 | ||
cb4a3167 | 680 | /** |
b4e0eeaf | 681 | * do_each_subsys_mask - filter for_each_subsys with a bitmask |
cb4a3167 AS |
682 | * @ss: the iteration cursor |
683 | * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end | |
b4e0eeaf | 684 | * @ss_mask: the bitmask |
cb4a3167 AS |
685 | * |
686 | * The block will only run for cases where the ssid-th bit (1 << ssid) of | |
b4e0eeaf | 687 | * @ss_mask is set. |
cb4a3167 | 688 | */ |
b4e0eeaf TH |
689 | #define do_each_subsys_mask(ss, ssid, ss_mask) do { \ |
690 | unsigned long __ss_mask = (ss_mask); \ | |
691 | if (!CGROUP_SUBSYS_COUNT) { /* to avoid spurious gcc warning */ \ | |
4a705c5c | 692 | (ssid) = 0; \ |
b4e0eeaf TH |
693 | break; \ |
694 | } \ | |
695 | for_each_set_bit(ssid, &__ss_mask, CGROUP_SUBSYS_COUNT) { \ | |
696 | (ss) = cgroup_subsys[ssid]; \ | |
697 | { | |
698 | ||
699 | #define while_each_subsys_mask() \ | |
700 | } \ | |
701 | } \ | |
702 | } while (false) | |
cb4a3167 | 703 | |
f8f22e53 TH |
704 | /* iterate over child cgrps, lock should be held throughout iteration */ |
705 | #define cgroup_for_each_live_child(child, cgrp) \ | |
d5c419b6 | 706 | list_for_each_entry((child), &(cgrp)->self.children, self.sibling) \ |
8353da1f | 707 | if (({ lockdep_assert_held(&cgroup_mutex); \ |
f8f22e53 TH |
708 | cgroup_is_dead(child); })) \ |
709 | ; \ | |
710 | else | |
7ae1bad9 | 711 | |
ce3f1d9d TH |
712 | /* walk live descendants in preorder */ |
713 | #define cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) \ | |
714 | css_for_each_descendant_pre((d_css), cgroup_css((cgrp), NULL)) \ | |
715 | if (({ lockdep_assert_held(&cgroup_mutex); \ | |
716 | (dsct) = (d_css)->cgroup; \ | |
717 | cgroup_is_dead(dsct); })) \ | |
718 | ; \ | |
719 | else | |
720 | ||
721 | /* walk live descendants in postorder */ | |
722 | #define cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) \ | |
723 | css_for_each_descendant_post((d_css), cgroup_css((cgrp), NULL)) \ | |
724 | if (({ lockdep_assert_held(&cgroup_mutex); \ | |
725 | (dsct) = (d_css)->cgroup; \ | |
726 | cgroup_is_dead(dsct); })) \ | |
727 | ; \ | |
728 | else | |
729 | ||
172a2c06 TH |
730 | /* |
731 | * The default css_set - used by init and its children prior to any | |
817929ec PM |
732 | * hierarchies being mounted. It contains a pointer to the root state |
733 | * for each subsystem. Also used to anchor the list of css_sets. Not | |
734 | * reference-counted, to improve performance when child cgroups | |
735 | * haven't been created. | |
736 | */ | |
5024ae29 | 737 | struct css_set init_css_set = { |
4b9502e6 | 738 | .refcount = REFCOUNT_INIT(1), |
454000ad | 739 | .dom_cset = &init_css_set, |
172a2c06 TH |
740 | .tasks = LIST_HEAD_INIT(init_css_set.tasks), |
741 | .mg_tasks = LIST_HEAD_INIT(init_css_set.mg_tasks), | |
c03cd773 | 742 | .dying_tasks = LIST_HEAD_INIT(init_css_set.dying_tasks), |
5f617ebb | 743 | .task_iters = LIST_HEAD_INIT(init_css_set.task_iters), |
454000ad | 744 | .threaded_csets = LIST_HEAD_INIT(init_css_set.threaded_csets), |
5f617ebb | 745 | .cgrp_links = LIST_HEAD_INIT(init_css_set.cgrp_links), |
172a2c06 TH |
746 | .mg_preload_node = LIST_HEAD_INIT(init_css_set.mg_preload_node), |
747 | .mg_node = LIST_HEAD_INIT(init_css_set.mg_node), | |
38683148 TH |
748 | |
749 | /* | |
750 | * The following field is re-initialized when this cset gets linked | |
751 | * in cgroup_init(). However, let's initialize the field | |
752 | * statically too so that the default cgroup can be accessed safely | |
753 | * early during boot. | |
754 | */ | |
755 | .dfl_cgrp = &cgrp_dfl_root.cgrp, | |
172a2c06 | 756 | }; |
817929ec | 757 | |
172a2c06 | 758 | static int css_set_count = 1; /* 1 for init_css_set */ |
817929ec | 759 | |
454000ad TH |
760 | static bool css_set_threaded(struct css_set *cset) |
761 | { | |
762 | return cset->dom_cset != cset; | |
763 | } | |
764 | ||
0de0942d TH |
765 | /** |
766 | * css_set_populated - does a css_set contain any tasks? | |
767 | * @cset: target css_set | |
73a7242a WL |
768 | * |
769 | * css_set_populated() should be the same as !!cset->nr_tasks at steady | |
770 | * state. However, css_set_populated() can be called while a task is being | |
771 | * added to or removed from the linked list before the nr_tasks is | |
772 | * properly updated. Hence, we can't just look at ->nr_tasks here. | |
0de0942d TH |
773 | */ |
774 | static bool css_set_populated(struct css_set *cset) | |
775 | { | |
f0d9a5f1 | 776 | lockdep_assert_held(&css_set_lock); |
0de0942d TH |
777 | |
778 | return !list_empty(&cset->tasks) || !list_empty(&cset->mg_tasks); | |
779 | } | |
780 | ||
842b597e | 781 | /** |
788b950c | 782 | * cgroup_update_populated - update the populated count of a cgroup |
842b597e TH |
783 | * @cgrp: the target cgroup |
784 | * @populated: inc or dec populated count | |
785 | * | |
0de0942d | 786 | * One of the css_sets associated with @cgrp is either getting its first |
788b950c TH |
787 | * task or losing the last. Update @cgrp->nr_populated_* accordingly. The |
788 | * count is propagated towards root so that a given cgroup's | |
789 | * nr_populated_children is zero iff none of its descendants contain any | |
790 | * tasks. | |
842b597e | 791 | * |
788b950c TH |
792 | * @cgrp's interface file "cgroup.populated" is zero if both |
793 | * @cgrp->nr_populated_csets and @cgrp->nr_populated_children are zero and | |
794 | * 1 otherwise. When the sum changes from or to zero, userland is notified | |
795 | * that the content of the interface file has changed. This can be used to | |
796 | * detect when @cgrp and its descendants become populated or empty. | |
842b597e TH |
797 | */ |
798 | static void cgroup_update_populated(struct cgroup *cgrp, bool populated) | |
799 | { | |
788b950c TH |
800 | struct cgroup *child = NULL; |
801 | int adj = populated ? 1 : -1; | |
802 | ||
f0d9a5f1 | 803 | lockdep_assert_held(&css_set_lock); |
842b597e TH |
804 | |
805 | do { | |
788b950c | 806 | bool was_populated = cgroup_is_populated(cgrp); |
842b597e | 807 | |
454000ad | 808 | if (!child) { |
788b950c | 809 | cgrp->nr_populated_csets += adj; |
454000ad TH |
810 | } else { |
811 | if (cgroup_is_threaded(child)) | |
812 | cgrp->nr_populated_threaded_children += adj; | |
813 | else | |
814 | cgrp->nr_populated_domain_children += adj; | |
815 | } | |
842b597e | 816 | |
788b950c | 817 | if (was_populated == cgroup_is_populated(cgrp)) |
842b597e TH |
818 | break; |
819 | ||
d62beb7f | 820 | cgroup1_check_for_release(cgrp); |
4c476d8c RG |
821 | TRACE_CGROUP_PATH(notify_populated, cgrp, |
822 | cgroup_is_populated(cgrp)); | |
6f60eade TH |
823 | cgroup_file_notify(&cgrp->events_file); |
824 | ||
788b950c | 825 | child = cgrp; |
d51f39b0 | 826 | cgrp = cgroup_parent(cgrp); |
842b597e TH |
827 | } while (cgrp); |
828 | } | |
829 | ||
0de0942d TH |
830 | /** |
831 | * css_set_update_populated - update populated state of a css_set | |
832 | * @cset: target css_set | |
833 | * @populated: whether @cset is populated or depopulated | |
834 | * | |
835 | * @cset is either getting the first task or losing the last. Update the | |
788b950c | 836 | * populated counters of all associated cgroups accordingly. |
0de0942d TH |
837 | */ |
838 | static void css_set_update_populated(struct css_set *cset, bool populated) | |
839 | { | |
840 | struct cgrp_cset_link *link; | |
841 | ||
f0d9a5f1 | 842 | lockdep_assert_held(&css_set_lock); |
0de0942d TH |
843 | |
844 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) | |
845 | cgroup_update_populated(link->cgrp, populated); | |
846 | } | |
847 | ||
b636fd38 TH |
848 | /* |
849 | * @task is leaving, advance task iterators which are pointing to it so | |
850 | * that they can resume at the next position. Advancing an iterator might | |
851 | * remove it from the list, use safe walk. See css_task_iter_skip() for | |
852 | * details. | |
853 | */ | |
854 | static void css_set_skip_task_iters(struct css_set *cset, | |
855 | struct task_struct *task) | |
856 | { | |
857 | struct css_task_iter *it, *pos; | |
858 | ||
859 | list_for_each_entry_safe(it, pos, &cset->task_iters, iters_node) | |
860 | css_task_iter_skip(it, task); | |
861 | } | |
862 | ||
f6d7d049 TH |
863 | /** |
864 | * css_set_move_task - move a task from one css_set to another | |
865 | * @task: task being moved | |
866 | * @from_cset: css_set @task currently belongs to (may be NULL) | |
867 | * @to_cset: new css_set @task is being moved to (may be NULL) | |
868 | * @use_mg_tasks: move to @to_cset->mg_tasks instead of ->tasks | |
869 | * | |
870 | * Move @task from @from_cset to @to_cset. If @task didn't belong to any | |
871 | * css_set, @from_cset can be NULL. If @task is being disassociated | |
872 | * instead of moved, @to_cset can be NULL. | |
873 | * | |
788b950c | 874 | * This function automatically handles populated counter updates and |
ed27b9f7 TH |
875 | * css_task_iter adjustments but the caller is responsible for managing |
876 | * @from_cset and @to_cset's reference counts. | |
f6d7d049 TH |
877 | */ |
878 | static void css_set_move_task(struct task_struct *task, | |
879 | struct css_set *from_cset, struct css_set *to_cset, | |
880 | bool use_mg_tasks) | |
881 | { | |
f0d9a5f1 | 882 | lockdep_assert_held(&css_set_lock); |
f6d7d049 | 883 | |
20b454a6 TH |
884 | if (to_cset && !css_set_populated(to_cset)) |
885 | css_set_update_populated(to_cset, true); | |
886 | ||
f6d7d049 TH |
887 | if (from_cset) { |
888 | WARN_ON_ONCE(list_empty(&task->cg_list)); | |
ed27b9f7 | 889 | |
b636fd38 | 890 | css_set_skip_task_iters(from_cset, task); |
f6d7d049 TH |
891 | list_del_init(&task->cg_list); |
892 | if (!css_set_populated(from_cset)) | |
893 | css_set_update_populated(from_cset, false); | |
894 | } else { | |
895 | WARN_ON_ONCE(!list_empty(&task->cg_list)); | |
896 | } | |
897 | ||
898 | if (to_cset) { | |
899 | /* | |
900 | * We are synchronized through cgroup_threadgroup_rwsem | |
901 | * against PF_EXITING setting such that we can't race | |
e7c7b1d8 | 902 | * against cgroup_exit()/cgroup_free() dropping the css_set. |
f6d7d049 TH |
903 | */ |
904 | WARN_ON_ONCE(task->flags & PF_EXITING); | |
905 | ||
2ce7135a | 906 | cgroup_move_task(task, to_cset); |
f6d7d049 TH |
907 | list_add_tail(&task->cg_list, use_mg_tasks ? &to_cset->mg_tasks : |
908 | &to_cset->tasks); | |
909 | } | |
910 | } | |
911 | ||
7717f7ba PM |
912 | /* |
913 | * hash table for cgroup groups. This improves the performance to find | |
914 | * an existing css_set. This hash doesn't (currently) take into | |
915 | * account cgroups in empty hierarchies. | |
916 | */ | |
472b1053 | 917 | #define CSS_SET_HASH_BITS 7 |
0ac801fe | 918 | static DEFINE_HASHTABLE(css_set_table, CSS_SET_HASH_BITS); |
472b1053 | 919 | |
0ac801fe | 920 | static unsigned long css_set_hash(struct cgroup_subsys_state *css[]) |
472b1053 | 921 | { |
0ac801fe | 922 | unsigned long key = 0UL; |
30159ec7 TH |
923 | struct cgroup_subsys *ss; |
924 | int i; | |
472b1053 | 925 | |
30159ec7 | 926 | for_each_subsys(ss, i) |
0ac801fe LZ |
927 | key += (unsigned long)css[i]; |
928 | key = (key >> 16) ^ key; | |
472b1053 | 929 | |
0ac801fe | 930 | return key; |
472b1053 LZ |
931 | } |
932 | ||
dcfe149b | 933 | void put_css_set_locked(struct css_set *cset) |
b4f48b63 | 934 | { |
69d0206c | 935 | struct cgrp_cset_link *link, *tmp_link; |
2d8f243a TH |
936 | struct cgroup_subsys *ss; |
937 | int ssid; | |
5abb8855 | 938 | |
f0d9a5f1 | 939 | lockdep_assert_held(&css_set_lock); |
89c5509b | 940 | |
4b9502e6 | 941 | if (!refcount_dec_and_test(&cset->refcount)) |
146aa1bd | 942 | return; |
81a6a5cd | 943 | |
454000ad TH |
944 | WARN_ON_ONCE(!list_empty(&cset->threaded_csets)); |
945 | ||
53254f90 TH |
946 | /* This css_set is dead. unlink it and release cgroup and css refs */ |
947 | for_each_subsys(ss, ssid) { | |
2d8f243a | 948 | list_del(&cset->e_cset_node[ssid]); |
53254f90 TH |
949 | css_put(cset->subsys[ssid]); |
950 | } | |
5abb8855 | 951 | hash_del(&cset->hlist); |
2c6ab6d2 PM |
952 | css_set_count--; |
953 | ||
69d0206c | 954 | list_for_each_entry_safe(link, tmp_link, &cset->cgrp_links, cgrp_link) { |
69d0206c TH |
955 | list_del(&link->cset_link); |
956 | list_del(&link->cgrp_link); | |
2ceb231b TH |
957 | if (cgroup_parent(link->cgrp)) |
958 | cgroup_put(link->cgrp); | |
2c6ab6d2 | 959 | kfree(link); |
81a6a5cd | 960 | } |
2c6ab6d2 | 961 | |
454000ad TH |
962 | if (css_set_threaded(cset)) { |
963 | list_del(&cset->threaded_csets_node); | |
964 | put_css_set_locked(cset->dom_cset); | |
965 | } | |
966 | ||
5abb8855 | 967 | kfree_rcu(cset, rcu_head); |
b4f48b63 PM |
968 | } |
969 | ||
b326f9d0 | 970 | /** |
7717f7ba | 971 | * compare_css_sets - helper function for find_existing_css_set(). |
5abb8855 TH |
972 | * @cset: candidate css_set being tested |
973 | * @old_cset: existing css_set for a task | |
7717f7ba PM |
974 | * @new_cgrp: cgroup that's being entered by the task |
975 | * @template: desired set of css pointers in css_set (pre-calculated) | |
976 | * | |
6f4b7e63 | 977 | * Returns true if "cset" matches "old_cset" except for the hierarchy |
7717f7ba PM |
978 | * which "new_cgrp" belongs to, for which it should match "new_cgrp". |
979 | */ | |
5abb8855 TH |
980 | static bool compare_css_sets(struct css_set *cset, |
981 | struct css_set *old_cset, | |
7717f7ba PM |
982 | struct cgroup *new_cgrp, |
983 | struct cgroup_subsys_state *template[]) | |
984 | { | |
454000ad | 985 | struct cgroup *new_dfl_cgrp; |
7717f7ba PM |
986 | struct list_head *l1, *l2; |
987 | ||
aec3dfcb TH |
988 | /* |
989 | * On the default hierarchy, there can be csets which are | |
990 | * associated with the same set of cgroups but different csses. | |
991 | * Let's first ensure that csses match. | |
992 | */ | |
993 | if (memcmp(template, cset->subsys, sizeof(cset->subsys))) | |
7717f7ba | 994 | return false; |
7717f7ba | 995 | |
454000ad TH |
996 | |
997 | /* @cset's domain should match the default cgroup's */ | |
998 | if (cgroup_on_dfl(new_cgrp)) | |
999 | new_dfl_cgrp = new_cgrp; | |
1000 | else | |
1001 | new_dfl_cgrp = old_cset->dfl_cgrp; | |
1002 | ||
1003 | if (new_dfl_cgrp->dom_cgrp != cset->dom_cset->dfl_cgrp) | |
1004 | return false; | |
1005 | ||
7717f7ba PM |
1006 | /* |
1007 | * Compare cgroup pointers in order to distinguish between | |
aec3dfcb TH |
1008 | * different cgroups in hierarchies. As different cgroups may |
1009 | * share the same effective css, this comparison is always | |
1010 | * necessary. | |
7717f7ba | 1011 | */ |
69d0206c TH |
1012 | l1 = &cset->cgrp_links; |
1013 | l2 = &old_cset->cgrp_links; | |
7717f7ba | 1014 | while (1) { |
69d0206c | 1015 | struct cgrp_cset_link *link1, *link2; |
5abb8855 | 1016 | struct cgroup *cgrp1, *cgrp2; |
7717f7ba PM |
1017 | |
1018 | l1 = l1->next; | |
1019 | l2 = l2->next; | |
1020 | /* See if we reached the end - both lists are equal length. */ | |
69d0206c TH |
1021 | if (l1 == &cset->cgrp_links) { |
1022 | BUG_ON(l2 != &old_cset->cgrp_links); | |
7717f7ba PM |
1023 | break; |
1024 | } else { | |
69d0206c | 1025 | BUG_ON(l2 == &old_cset->cgrp_links); |
7717f7ba PM |
1026 | } |
1027 | /* Locate the cgroups associated with these links. */ | |
69d0206c TH |
1028 | link1 = list_entry(l1, struct cgrp_cset_link, cgrp_link); |
1029 | link2 = list_entry(l2, struct cgrp_cset_link, cgrp_link); | |
1030 | cgrp1 = link1->cgrp; | |
1031 | cgrp2 = link2->cgrp; | |
7717f7ba | 1032 | /* Hierarchies should be linked in the same order. */ |
5abb8855 | 1033 | BUG_ON(cgrp1->root != cgrp2->root); |
7717f7ba PM |
1034 | |
1035 | /* | |
1036 | * If this hierarchy is the hierarchy of the cgroup | |
1037 | * that's changing, then we need to check that this | |
1038 | * css_set points to the new cgroup; if it's any other | |
1039 | * hierarchy, then this css_set should point to the | |
1040 | * same cgroup as the old css_set. | |
1041 | */ | |
5abb8855 TH |
1042 | if (cgrp1->root == new_cgrp->root) { |
1043 | if (cgrp1 != new_cgrp) | |
7717f7ba PM |
1044 | return false; |
1045 | } else { | |
5abb8855 | 1046 | if (cgrp1 != cgrp2) |
7717f7ba PM |
1047 | return false; |
1048 | } | |
1049 | } | |
1050 | return true; | |
1051 | } | |
1052 | ||
b326f9d0 TH |
1053 | /** |
1054 | * find_existing_css_set - init css array and find the matching css_set | |
1055 | * @old_cset: the css_set that we're using before the cgroup transition | |
1056 | * @cgrp: the cgroup that we're moving into | |
1057 | * @template: out param for the new set of csses, should be clear on entry | |
817929ec | 1058 | */ |
5abb8855 TH |
1059 | static struct css_set *find_existing_css_set(struct css_set *old_cset, |
1060 | struct cgroup *cgrp, | |
1061 | struct cgroup_subsys_state *template[]) | |
b4f48b63 | 1062 | { |
3dd06ffa | 1063 | struct cgroup_root *root = cgrp->root; |
30159ec7 | 1064 | struct cgroup_subsys *ss; |
5abb8855 | 1065 | struct css_set *cset; |
0ac801fe | 1066 | unsigned long key; |
b326f9d0 | 1067 | int i; |
817929ec | 1068 | |
aae8aab4 BB |
1069 | /* |
1070 | * Build the set of subsystem state objects that we want to see in the | |
1071 | * new css_set. while subsystems can change globally, the entries here | |
1072 | * won't change, so no need for locking. | |
1073 | */ | |
30159ec7 | 1074 | for_each_subsys(ss, i) { |
f392e51c | 1075 | if (root->subsys_mask & (1UL << i)) { |
aec3dfcb TH |
1076 | /* |
1077 | * @ss is in this hierarchy, so we want the | |
1078 | * effective css from @cgrp. | |
1079 | */ | |
fc5a828b | 1080 | template[i] = cgroup_e_css_by_mask(cgrp, ss); |
817929ec | 1081 | } else { |
aec3dfcb TH |
1082 | /* |
1083 | * @ss is not in this hierarchy, so we don't want | |
1084 | * to change the css. | |
1085 | */ | |
5abb8855 | 1086 | template[i] = old_cset->subsys[i]; |
817929ec PM |
1087 | } |
1088 | } | |
1089 | ||
0ac801fe | 1090 | key = css_set_hash(template); |
5abb8855 TH |
1091 | hash_for_each_possible(css_set_table, cset, hlist, key) { |
1092 | if (!compare_css_sets(cset, old_cset, cgrp, template)) | |
7717f7ba PM |
1093 | continue; |
1094 | ||
1095 | /* This css_set matches what we need */ | |
5abb8855 | 1096 | return cset; |
472b1053 | 1097 | } |
817929ec PM |
1098 | |
1099 | /* No existing cgroup group matched */ | |
1100 | return NULL; | |
1101 | } | |
1102 | ||
69d0206c | 1103 | static void free_cgrp_cset_links(struct list_head *links_to_free) |
36553434 | 1104 | { |
69d0206c | 1105 | struct cgrp_cset_link *link, *tmp_link; |
36553434 | 1106 | |
69d0206c TH |
1107 | list_for_each_entry_safe(link, tmp_link, links_to_free, cset_link) { |
1108 | list_del(&link->cset_link); | |
36553434 LZ |
1109 | kfree(link); |
1110 | } | |
1111 | } | |
1112 | ||
69d0206c TH |
1113 | /** |
1114 | * allocate_cgrp_cset_links - allocate cgrp_cset_links | |
1115 | * @count: the number of links to allocate | |
1116 | * @tmp_links: list_head the allocated links are put on | |
1117 | * | |
1118 | * Allocate @count cgrp_cset_link structures and chain them on @tmp_links | |
1119 | * through ->cset_link. Returns 0 on success or -errno. | |
817929ec | 1120 | */ |
69d0206c | 1121 | static int allocate_cgrp_cset_links(int count, struct list_head *tmp_links) |
817929ec | 1122 | { |
69d0206c | 1123 | struct cgrp_cset_link *link; |
817929ec | 1124 | int i; |
69d0206c TH |
1125 | |
1126 | INIT_LIST_HEAD(tmp_links); | |
1127 | ||
817929ec | 1128 | for (i = 0; i < count; i++) { |
f4f4be2b | 1129 | link = kzalloc(sizeof(*link), GFP_KERNEL); |
817929ec | 1130 | if (!link) { |
69d0206c | 1131 | free_cgrp_cset_links(tmp_links); |
817929ec PM |
1132 | return -ENOMEM; |
1133 | } | |
69d0206c | 1134 | list_add(&link->cset_link, tmp_links); |
817929ec PM |
1135 | } |
1136 | return 0; | |
1137 | } | |
1138 | ||
c12f65d4 LZ |
1139 | /** |
1140 | * link_css_set - a helper function to link a css_set to a cgroup | |
69d0206c | 1141 | * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links() |
5abb8855 | 1142 | * @cset: the css_set to be linked |
c12f65d4 LZ |
1143 | * @cgrp: the destination cgroup |
1144 | */ | |
69d0206c TH |
1145 | static void link_css_set(struct list_head *tmp_links, struct css_set *cset, |
1146 | struct cgroup *cgrp) | |
c12f65d4 | 1147 | { |
69d0206c | 1148 | struct cgrp_cset_link *link; |
c12f65d4 | 1149 | |
69d0206c | 1150 | BUG_ON(list_empty(tmp_links)); |
6803c006 TH |
1151 | |
1152 | if (cgroup_on_dfl(cgrp)) | |
1153 | cset->dfl_cgrp = cgrp; | |
1154 | ||
69d0206c TH |
1155 | link = list_first_entry(tmp_links, struct cgrp_cset_link, cset_link); |
1156 | link->cset = cset; | |
7717f7ba | 1157 | link->cgrp = cgrp; |
842b597e | 1158 | |
7717f7ba | 1159 | /* |
389b9c1b TH |
1160 | * Always add links to the tail of the lists so that the lists are |
1161 | * in choronological order. | |
7717f7ba | 1162 | */ |
389b9c1b | 1163 | list_move_tail(&link->cset_link, &cgrp->cset_links); |
69d0206c | 1164 | list_add_tail(&link->cgrp_link, &cset->cgrp_links); |
2ceb231b TH |
1165 | |
1166 | if (cgroup_parent(cgrp)) | |
a590b90d | 1167 | cgroup_get_live(cgrp); |
c12f65d4 LZ |
1168 | } |
1169 | ||
b326f9d0 TH |
1170 | /** |
1171 | * find_css_set - return a new css_set with one cgroup updated | |
1172 | * @old_cset: the baseline css_set | |
1173 | * @cgrp: the cgroup to be updated | |
1174 | * | |
1175 | * Return a new css_set that's equivalent to @old_cset, but with @cgrp | |
1176 | * substituted into the appropriate hierarchy. | |
817929ec | 1177 | */ |
5abb8855 TH |
1178 | static struct css_set *find_css_set(struct css_set *old_cset, |
1179 | struct cgroup *cgrp) | |
817929ec | 1180 | { |
b326f9d0 | 1181 | struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT] = { }; |
5abb8855 | 1182 | struct css_set *cset; |
69d0206c TH |
1183 | struct list_head tmp_links; |
1184 | struct cgrp_cset_link *link; | |
2d8f243a | 1185 | struct cgroup_subsys *ss; |
0ac801fe | 1186 | unsigned long key; |
2d8f243a | 1187 | int ssid; |
472b1053 | 1188 | |
b326f9d0 TH |
1189 | lockdep_assert_held(&cgroup_mutex); |
1190 | ||
817929ec PM |
1191 | /* First see if we already have a cgroup group that matches |
1192 | * the desired set */ | |
82d6489d | 1193 | spin_lock_irq(&css_set_lock); |
5abb8855 TH |
1194 | cset = find_existing_css_set(old_cset, cgrp, template); |
1195 | if (cset) | |
1196 | get_css_set(cset); | |
82d6489d | 1197 | spin_unlock_irq(&css_set_lock); |
817929ec | 1198 | |
5abb8855 TH |
1199 | if (cset) |
1200 | return cset; | |
817929ec | 1201 | |
f4f4be2b | 1202 | cset = kzalloc(sizeof(*cset), GFP_KERNEL); |
5abb8855 | 1203 | if (!cset) |
817929ec PM |
1204 | return NULL; |
1205 | ||
69d0206c | 1206 | /* Allocate all the cgrp_cset_link objects that we'll need */ |
9871bf95 | 1207 | if (allocate_cgrp_cset_links(cgroup_root_count, &tmp_links) < 0) { |
5abb8855 | 1208 | kfree(cset); |
817929ec PM |
1209 | return NULL; |
1210 | } | |
1211 | ||
4b9502e6 | 1212 | refcount_set(&cset->refcount, 1); |
454000ad | 1213 | cset->dom_cset = cset; |
5abb8855 | 1214 | INIT_LIST_HEAD(&cset->tasks); |
c7561128 | 1215 | INIT_LIST_HEAD(&cset->mg_tasks); |
c03cd773 | 1216 | INIT_LIST_HEAD(&cset->dying_tasks); |
ed27b9f7 | 1217 | INIT_LIST_HEAD(&cset->task_iters); |
454000ad | 1218 | INIT_LIST_HEAD(&cset->threaded_csets); |
5abb8855 | 1219 | INIT_HLIST_NODE(&cset->hlist); |
5f617ebb TH |
1220 | INIT_LIST_HEAD(&cset->cgrp_links); |
1221 | INIT_LIST_HEAD(&cset->mg_preload_node); | |
1222 | INIT_LIST_HEAD(&cset->mg_node); | |
817929ec PM |
1223 | |
1224 | /* Copy the set of subsystem state objects generated in | |
1225 | * find_existing_css_set() */ | |
5abb8855 | 1226 | memcpy(cset->subsys, template, sizeof(cset->subsys)); |
817929ec | 1227 | |
82d6489d | 1228 | spin_lock_irq(&css_set_lock); |
817929ec | 1229 | /* Add reference counts and links from the new css_set. */ |
69d0206c | 1230 | list_for_each_entry(link, &old_cset->cgrp_links, cgrp_link) { |
7717f7ba | 1231 | struct cgroup *c = link->cgrp; |
69d0206c | 1232 | |
7717f7ba PM |
1233 | if (c->root == cgrp->root) |
1234 | c = cgrp; | |
69d0206c | 1235 | link_css_set(&tmp_links, cset, c); |
7717f7ba | 1236 | } |
817929ec | 1237 | |
69d0206c | 1238 | BUG_ON(!list_empty(&tmp_links)); |
817929ec | 1239 | |
817929ec | 1240 | css_set_count++; |
472b1053 | 1241 | |
2d8f243a | 1242 | /* Add @cset to the hash table */ |
5abb8855 TH |
1243 | key = css_set_hash(cset->subsys); |
1244 | hash_add(css_set_table, &cset->hlist, key); | |
472b1053 | 1245 | |
53254f90 TH |
1246 | for_each_subsys(ss, ssid) { |
1247 | struct cgroup_subsys_state *css = cset->subsys[ssid]; | |
1248 | ||
2d8f243a | 1249 | list_add_tail(&cset->e_cset_node[ssid], |
53254f90 TH |
1250 | &css->cgroup->e_csets[ssid]); |
1251 | css_get(css); | |
1252 | } | |
2d8f243a | 1253 | |
82d6489d | 1254 | spin_unlock_irq(&css_set_lock); |
817929ec | 1255 | |
454000ad TH |
1256 | /* |
1257 | * If @cset should be threaded, look up the matching dom_cset and | |
1258 | * link them up. We first fully initialize @cset then look for the | |
1259 | * dom_cset. It's simpler this way and safe as @cset is guaranteed | |
1260 | * to stay empty until we return. | |
1261 | */ | |
1262 | if (cgroup_is_threaded(cset->dfl_cgrp)) { | |
1263 | struct css_set *dcset; | |
1264 | ||
1265 | dcset = find_css_set(cset, cset->dfl_cgrp->dom_cgrp); | |
1266 | if (!dcset) { | |
1267 | put_css_set(cset); | |
1268 | return NULL; | |
1269 | } | |
1270 | ||
1271 | spin_lock_irq(&css_set_lock); | |
1272 | cset->dom_cset = dcset; | |
1273 | list_add_tail(&cset->threaded_csets_node, | |
1274 | &dcset->threaded_csets); | |
1275 | spin_unlock_irq(&css_set_lock); | |
1276 | } | |
1277 | ||
5abb8855 | 1278 | return cset; |
b4f48b63 PM |
1279 | } |
1280 | ||
0a268dbd | 1281 | struct cgroup_root *cgroup_root_from_kf(struct kernfs_root *kf_root) |
7717f7ba | 1282 | { |
3dd06ffa | 1283 | struct cgroup *root_cgrp = kf_root->kn->priv; |
2bd59d48 | 1284 | |
3dd06ffa | 1285 | return root_cgrp->root; |
2bd59d48 TH |
1286 | } |
1287 | ||
3dd06ffa | 1288 | static int cgroup_init_root_id(struct cgroup_root *root) |
f2e85d57 TH |
1289 | { |
1290 | int id; | |
1291 | ||
1292 | lockdep_assert_held(&cgroup_mutex); | |
1293 | ||
985ed670 | 1294 | id = idr_alloc_cyclic(&cgroup_hierarchy_idr, root, 0, 0, GFP_KERNEL); |
f2e85d57 TH |
1295 | if (id < 0) |
1296 | return id; | |
1297 | ||
1298 | root->hierarchy_id = id; | |
1299 | return 0; | |
1300 | } | |
1301 | ||
3dd06ffa | 1302 | static void cgroup_exit_root_id(struct cgroup_root *root) |
f2e85d57 TH |
1303 | { |
1304 | lockdep_assert_held(&cgroup_mutex); | |
1305 | ||
8c8a5502 | 1306 | idr_remove(&cgroup_hierarchy_idr, root->hierarchy_id); |
f2e85d57 TH |
1307 | } |
1308 | ||
1592c9b2 | 1309 | void cgroup_free_root(struct cgroup_root *root) |
f2e85d57 | 1310 | { |
74321038 | 1311 | kfree(root); |
f2e85d57 TH |
1312 | } |
1313 | ||
3dd06ffa | 1314 | static void cgroup_destroy_root(struct cgroup_root *root) |
59f5296b | 1315 | { |
3dd06ffa | 1316 | struct cgroup *cgrp = &root->cgrp; |
f2e85d57 | 1317 | struct cgrp_cset_link *link, *tmp_link; |
f2e85d57 | 1318 | |
ed1777de TH |
1319 | trace_cgroup_destroy_root(root); |
1320 | ||
334c3679 | 1321 | cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp); |
f2e85d57 | 1322 | |
776f02fa | 1323 | BUG_ON(atomic_read(&root->nr_cgrps)); |
d5c419b6 | 1324 | BUG_ON(!list_empty(&cgrp->self.children)); |
f2e85d57 | 1325 | |
f2e85d57 | 1326 | /* Rebind all subsystems back to the default hierarchy */ |
334c3679 | 1327 | WARN_ON(rebind_subsystems(&cgrp_dfl_root, root->subsys_mask)); |
7717f7ba | 1328 | |
7717f7ba | 1329 | /* |
f2e85d57 TH |
1330 | * Release all the links from cset_links to this hierarchy's |
1331 | * root cgroup | |
7717f7ba | 1332 | */ |
82d6489d | 1333 | spin_lock_irq(&css_set_lock); |
f2e85d57 TH |
1334 | |
1335 | list_for_each_entry_safe(link, tmp_link, &cgrp->cset_links, cset_link) { | |
1336 | list_del(&link->cset_link); | |
1337 | list_del(&link->cgrp_link); | |
1338 | kfree(link); | |
1339 | } | |
f0d9a5f1 | 1340 | |
82d6489d | 1341 | spin_unlock_irq(&css_set_lock); |
f2e85d57 TH |
1342 | |
1343 | if (!list_empty(&root->root_list)) { | |
1344 | list_del(&root->root_list); | |
1345 | cgroup_root_count--; | |
1346 | } | |
1347 | ||
1348 | cgroup_exit_root_id(root); | |
1349 | ||
1350 | mutex_unlock(&cgroup_mutex); | |
f2e85d57 | 1351 | |
2bd59d48 | 1352 | kernfs_destroy_root(root->kf_root); |
f2e85d57 TH |
1353 | cgroup_free_root(root); |
1354 | } | |
1355 | ||
4f41fc59 SH |
1356 | /* |
1357 | * look up cgroup associated with current task's cgroup namespace on the | |
1358 | * specified hierarchy | |
1359 | */ | |
1360 | static struct cgroup * | |
1361 | current_cgns_cgroup_from_root(struct cgroup_root *root) | |
1362 | { | |
1363 | struct cgroup *res = NULL; | |
1364 | struct css_set *cset; | |
1365 | ||
1366 | lockdep_assert_held(&css_set_lock); | |
1367 | ||
1368 | rcu_read_lock(); | |
1369 | ||
1370 | cset = current->nsproxy->cgroup_ns->root_cset; | |
1371 | if (cset == &init_css_set) { | |
1372 | res = &root->cgrp; | |
61e867fd ML |
1373 | } else if (root == &cgrp_dfl_root) { |
1374 | res = cset->dfl_cgrp; | |
4f41fc59 SH |
1375 | } else { |
1376 | struct cgrp_cset_link *link; | |
1377 | ||
1378 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) { | |
1379 | struct cgroup *c = link->cgrp; | |
1380 | ||
1381 | if (c->root == root) { | |
1382 | res = c; | |
1383 | break; | |
1384 | } | |
1385 | } | |
1386 | } | |
1387 | rcu_read_unlock(); | |
1388 | ||
1389 | BUG_ON(!res); | |
1390 | return res; | |
1391 | } | |
1392 | ||
ceb6a081 TH |
1393 | /* look up cgroup associated with given css_set on the specified hierarchy */ |
1394 | static struct cgroup *cset_cgroup_from_root(struct css_set *cset, | |
3dd06ffa | 1395 | struct cgroup_root *root) |
7717f7ba | 1396 | { |
7717f7ba PM |
1397 | struct cgroup *res = NULL; |
1398 | ||
96d365e0 | 1399 | lockdep_assert_held(&cgroup_mutex); |
f0d9a5f1 | 1400 | lockdep_assert_held(&css_set_lock); |
96d365e0 | 1401 | |
5abb8855 | 1402 | if (cset == &init_css_set) { |
3dd06ffa | 1403 | res = &root->cgrp; |
13d82fb7 TH |
1404 | } else if (root == &cgrp_dfl_root) { |
1405 | res = cset->dfl_cgrp; | |
7717f7ba | 1406 | } else { |
69d0206c TH |
1407 | struct cgrp_cset_link *link; |
1408 | ||
1409 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) { | |
7717f7ba | 1410 | struct cgroup *c = link->cgrp; |
69d0206c | 1411 | |
7717f7ba PM |
1412 | if (c->root == root) { |
1413 | res = c; | |
1414 | break; | |
1415 | } | |
1416 | } | |
1417 | } | |
96d365e0 | 1418 | |
7717f7ba PM |
1419 | BUG_ON(!res); |
1420 | return res; | |
1421 | } | |
1422 | ||
ddbcc7e8 | 1423 | /* |
ceb6a081 | 1424 | * Return the cgroup for "task" from the given hierarchy. Must be |
f0d9a5f1 | 1425 | * called with cgroup_mutex and css_set_lock held. |
ceb6a081 | 1426 | */ |
0a268dbd TH |
1427 | struct cgroup *task_cgroup_from_root(struct task_struct *task, |
1428 | struct cgroup_root *root) | |
ceb6a081 TH |
1429 | { |
1430 | /* | |
e7c7b1d8 MK |
1431 | * No need to lock the task - since we hold css_set_lock the |
1432 | * task can't change groups. | |
ceb6a081 TH |
1433 | */ |
1434 | return cset_cgroup_from_root(task_css_set(task), root); | |
1435 | } | |
1436 | ||
ddbcc7e8 | 1437 | /* |
ddbcc7e8 PM |
1438 | * A task must hold cgroup_mutex to modify cgroups. |
1439 | * | |
1440 | * Any task can increment and decrement the count field without lock. | |
1441 | * So in general, code holding cgroup_mutex can't rely on the count | |
1442 | * field not changing. However, if the count goes to zero, then only | |
956db3ca | 1443 | * cgroup_attach_task() can increment it again. Because a count of zero |
ddbcc7e8 PM |
1444 | * means that no tasks are currently attached, therefore there is no |
1445 | * way a task attached to that cgroup can fork (the other way to | |
1446 | * increment the count). So code holding cgroup_mutex can safely | |
1447 | * assume that if the count is zero, it will stay zero. Similarly, if | |
1448 | * a task holds cgroup_mutex on a cgroup with zero count, it | |
1449 | * knows that the cgroup won't be removed, as cgroup_rmdir() | |
1450 | * needs that mutex. | |
1451 | * | |
ddbcc7e8 PM |
1452 | * A cgroup can only be deleted if both its 'count' of using tasks |
1453 | * is zero, and its list of 'children' cgroups is empty. Since all | |
1454 | * tasks in the system use _some_ cgroup, and since there is always at | |
3dd06ffa | 1455 | * least one task in the system (init, pid == 1), therefore, root cgroup |
ddbcc7e8 | 1456 | * always has either children cgroups and/or using tasks. So we don't |
3dd06ffa | 1457 | * need a special hack to ensure that root cgroup cannot be deleted. |
ddbcc7e8 PM |
1458 | * |
1459 | * P.S. One more locking exception. RCU is used to guard the | |
956db3ca | 1460 | * update of a tasks cgroup pointer by cgroup_attach_task() |
ddbcc7e8 PM |
1461 | */ |
1462 | ||
2bd59d48 | 1463 | static struct kernfs_syscall_ops cgroup_kf_syscall_ops; |
a424316c | 1464 | |
cf892988 JA |
1465 | static char *cgroup_file_name(struct cgroup *cgrp, const struct cftype *cft, |
1466 | char *buf) | |
ddbcc7e8 | 1467 | { |
3e1d2eed TH |
1468 | struct cgroup_subsys *ss = cft->ss; |
1469 | ||
8d7e6fb0 | 1470 | if (cft->ss && !(cft->flags & CFTYPE_NO_PREFIX) && |
c1bbd933 TH |
1471 | !(cgrp->root->flags & CGRP_ROOT_NOPREFIX)) { |
1472 | const char *dbg = (cft->flags & CFTYPE_DEBUG) ? ".__DEBUG__." : ""; | |
1473 | ||
1474 | snprintf(buf, CGROUP_FILE_NAME_MAX, "%s%s.%s", | |
1475 | dbg, cgroup_on_dfl(cgrp) ? ss->name : ss->legacy_name, | |
cf892988 | 1476 | cft->name); |
c1bbd933 | 1477 | } else { |
cf892988 | 1478 | strscpy(buf, cft->name, CGROUP_FILE_NAME_MAX); |
c1bbd933 | 1479 | } |
8d7e6fb0 | 1480 | return buf; |
ddbcc7e8 PM |
1481 | } |
1482 | ||
f2e85d57 TH |
1483 | /** |
1484 | * cgroup_file_mode - deduce file mode of a control file | |
1485 | * @cft: the control file in question | |
1486 | * | |
7dbdb199 | 1487 | * S_IRUGO for read, S_IWUSR for write. |
f2e85d57 TH |
1488 | */ |
1489 | static umode_t cgroup_file_mode(const struct cftype *cft) | |
65dff759 | 1490 | { |
f2e85d57 | 1491 | umode_t mode = 0; |
65dff759 | 1492 | |
f2e85d57 TH |
1493 | if (cft->read_u64 || cft->read_s64 || cft->seq_show) |
1494 | mode |= S_IRUGO; | |
1495 | ||
7dbdb199 TH |
1496 | if (cft->write_u64 || cft->write_s64 || cft->write) { |
1497 | if (cft->flags & CFTYPE_WORLD_WRITABLE) | |
1498 | mode |= S_IWUGO; | |
1499 | else | |
1500 | mode |= S_IWUSR; | |
1501 | } | |
f2e85d57 TH |
1502 | |
1503 | return mode; | |
65dff759 LZ |
1504 | } |
1505 | ||
af0ba678 | 1506 | /** |
8699b776 | 1507 | * cgroup_calc_subtree_ss_mask - calculate subtree_ss_mask |
0f060deb | 1508 | * @subtree_control: the new subtree_control mask to consider |
5ced2518 | 1509 | * @this_ss_mask: available subsystems |
af0ba678 TH |
1510 | * |
1511 | * On the default hierarchy, a subsystem may request other subsystems to be | |
1512 | * enabled together through its ->depends_on mask. In such cases, more | |
1513 | * subsystems than specified in "cgroup.subtree_control" may be enabled. | |
1514 | * | |
0f060deb | 1515 | * This function calculates which subsystems need to be enabled if |
5ced2518 | 1516 | * @subtree_control is to be applied while restricted to @this_ss_mask. |
af0ba678 | 1517 | */ |
5ced2518 | 1518 | static u16 cgroup_calc_subtree_ss_mask(u16 subtree_control, u16 this_ss_mask) |
667c2491 | 1519 | { |
6e5c8307 | 1520 | u16 cur_ss_mask = subtree_control; |
af0ba678 TH |
1521 | struct cgroup_subsys *ss; |
1522 | int ssid; | |
1523 | ||
1524 | lockdep_assert_held(&cgroup_mutex); | |
1525 | ||
f6d635ad TH |
1526 | cur_ss_mask |= cgrp_dfl_implicit_ss_mask; |
1527 | ||
af0ba678 | 1528 | while (true) { |
6e5c8307 | 1529 | u16 new_ss_mask = cur_ss_mask; |
af0ba678 | 1530 | |
b4e0eeaf | 1531 | do_each_subsys_mask(ss, ssid, cur_ss_mask) { |
a966a4ed | 1532 | new_ss_mask |= ss->depends_on; |
b4e0eeaf | 1533 | } while_each_subsys_mask(); |
af0ba678 TH |
1534 | |
1535 | /* | |
1536 | * Mask out subsystems which aren't available. This can | |
1537 | * happen only if some depended-upon subsystems were bound | |
1538 | * to non-default hierarchies. | |
1539 | */ | |
5ced2518 | 1540 | new_ss_mask &= this_ss_mask; |
af0ba678 TH |
1541 | |
1542 | if (new_ss_mask == cur_ss_mask) | |
1543 | break; | |
1544 | cur_ss_mask = new_ss_mask; | |
1545 | } | |
1546 | ||
0f060deb TH |
1547 | return cur_ss_mask; |
1548 | } | |
1549 | ||
a9746d8d TH |
1550 | /** |
1551 | * cgroup_kn_unlock - unlocking helper for cgroup kernfs methods | |
1552 | * @kn: the kernfs_node being serviced | |
1553 | * | |
1554 | * This helper undoes cgroup_kn_lock_live() and should be invoked before | |
1555 | * the method finishes if locking succeeded. Note that once this function | |
1556 | * returns the cgroup returned by cgroup_kn_lock_live() may become | |
1557 | * inaccessible any time. If the caller intends to continue to access the | |
1558 | * cgroup, it should pin it before invoking this function. | |
1559 | */ | |
0a268dbd | 1560 | void cgroup_kn_unlock(struct kernfs_node *kn) |
ddbcc7e8 | 1561 | { |
a9746d8d TH |
1562 | struct cgroup *cgrp; |
1563 | ||
1564 | if (kernfs_type(kn) == KERNFS_DIR) | |
1565 | cgrp = kn->priv; | |
1566 | else | |
1567 | cgrp = kn->parent->priv; | |
1568 | ||
1569 | mutex_unlock(&cgroup_mutex); | |
a9746d8d TH |
1570 | |
1571 | kernfs_unbreak_active_protection(kn); | |
1572 | cgroup_put(cgrp); | |
ddbcc7e8 PM |
1573 | } |
1574 | ||
a9746d8d TH |
1575 | /** |
1576 | * cgroup_kn_lock_live - locking helper for cgroup kernfs methods | |
1577 | * @kn: the kernfs_node being serviced | |
945ba199 | 1578 | * @drain_offline: perform offline draining on the cgroup |
a9746d8d TH |
1579 | * |
1580 | * This helper is to be used by a cgroup kernfs method currently servicing | |
1581 | * @kn. It breaks the active protection, performs cgroup locking and | |
1582 | * verifies that the associated cgroup is alive. Returns the cgroup if | |
1583 | * alive; otherwise, %NULL. A successful return should be undone by a | |
945ba199 TH |
1584 | * matching cgroup_kn_unlock() invocation. If @drain_offline is %true, the |
1585 | * cgroup is drained of offlining csses before return. | |
a9746d8d TH |
1586 | * |
1587 | * Any cgroup kernfs method implementation which requires locking the | |
1588 | * associated cgroup should use this helper. It avoids nesting cgroup | |
1589 | * locking under kernfs active protection and allows all kernfs operations | |
1590 | * including self-removal. | |
1591 | */ | |
0a268dbd | 1592 | struct cgroup *cgroup_kn_lock_live(struct kernfs_node *kn, bool drain_offline) |
05ef1d7c | 1593 | { |
a9746d8d TH |
1594 | struct cgroup *cgrp; |
1595 | ||
1596 | if (kernfs_type(kn) == KERNFS_DIR) | |
1597 | cgrp = kn->priv; | |
1598 | else | |
1599 | cgrp = kn->parent->priv; | |
05ef1d7c | 1600 | |
2739d3cc | 1601 | /* |
01f6474c | 1602 | * We're gonna grab cgroup_mutex which nests outside kernfs |
a9746d8d TH |
1603 | * active_ref. cgroup liveliness check alone provides enough |
1604 | * protection against removal. Ensure @cgrp stays accessible and | |
1605 | * break the active_ref protection. | |
2739d3cc | 1606 | */ |
aa32362f LZ |
1607 | if (!cgroup_tryget(cgrp)) |
1608 | return NULL; | |
a9746d8d TH |
1609 | kernfs_break_active_protection(kn); |
1610 | ||
945ba199 TH |
1611 | if (drain_offline) |
1612 | cgroup_lock_and_drain_offline(cgrp); | |
1613 | else | |
1614 | mutex_lock(&cgroup_mutex); | |
05ef1d7c | 1615 | |
a9746d8d TH |
1616 | if (!cgroup_is_dead(cgrp)) |
1617 | return cgrp; | |
1618 | ||
1619 | cgroup_kn_unlock(kn); | |
1620 | return NULL; | |
ddbcc7e8 | 1621 | } |
05ef1d7c | 1622 | |
2739d3cc | 1623 | static void cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft) |
05ef1d7c | 1624 | { |
2bd59d48 | 1625 | char name[CGROUP_FILE_NAME_MAX]; |
05ef1d7c | 1626 | |
01f6474c | 1627 | lockdep_assert_held(&cgroup_mutex); |
34c06254 TH |
1628 | |
1629 | if (cft->file_offset) { | |
1630 | struct cgroup_subsys_state *css = cgroup_css(cgrp, cft->ss); | |
1631 | struct cgroup_file *cfile = (void *)css + cft->file_offset; | |
1632 | ||
1633 | spin_lock_irq(&cgroup_file_kn_lock); | |
1634 | cfile->kn = NULL; | |
1635 | spin_unlock_irq(&cgroup_file_kn_lock); | |
b12e3583 TH |
1636 | |
1637 | del_timer_sync(&cfile->notify_timer); | |
34c06254 TH |
1638 | } |
1639 | ||
2bd59d48 | 1640 | kernfs_remove_by_name(cgrp->kn, cgroup_file_name(cgrp, cft, name)); |
05ef1d7c TH |
1641 | } |
1642 | ||
13af07df | 1643 | /** |
4df8dc90 TH |
1644 | * css_clear_dir - remove subsys files in a cgroup directory |
1645 | * @css: taget css | |
13af07df | 1646 | */ |
334c3679 | 1647 | static void css_clear_dir(struct cgroup_subsys_state *css) |
05ef1d7c | 1648 | { |
334c3679 | 1649 | struct cgroup *cgrp = css->cgroup; |
4df8dc90 | 1650 | struct cftype *cfts; |
05ef1d7c | 1651 | |
88cb04b9 TH |
1652 | if (!(css->flags & CSS_VISIBLE)) |
1653 | return; | |
1654 | ||
1655 | css->flags &= ~CSS_VISIBLE; | |
1656 | ||
5faaf05f TH |
1657 | if (!css->ss) { |
1658 | if (cgroup_on_dfl(cgrp)) | |
1659 | cfts = cgroup_base_files; | |
1660 | else | |
1661 | cfts = cgroup1_base_files; | |
1662 | ||
4df8dc90 | 1663 | cgroup_addrm_files(css, cgrp, cfts, false); |
5faaf05f TH |
1664 | } else { |
1665 | list_for_each_entry(cfts, &css->ss->cfts, node) | |
1666 | cgroup_addrm_files(css, cgrp, cfts, false); | |
1667 | } | |
ddbcc7e8 PM |
1668 | } |
1669 | ||
ccdca218 | 1670 | /** |
4df8dc90 TH |
1671 | * css_populate_dir - create subsys files in a cgroup directory |
1672 | * @css: target css | |
ccdca218 TH |
1673 | * |
1674 | * On failure, no file is added. | |
1675 | */ | |
334c3679 | 1676 | static int css_populate_dir(struct cgroup_subsys_state *css) |
ccdca218 | 1677 | { |
334c3679 | 1678 | struct cgroup *cgrp = css->cgroup; |
4df8dc90 TH |
1679 | struct cftype *cfts, *failed_cfts; |
1680 | int ret; | |
ccdca218 | 1681 | |
03970d3c | 1682 | if ((css->flags & CSS_VISIBLE) || !cgrp->kn) |
88cb04b9 TH |
1683 | return 0; |
1684 | ||
4df8dc90 TH |
1685 | if (!css->ss) { |
1686 | if (cgroup_on_dfl(cgrp)) | |
d62beb7f | 1687 | cfts = cgroup_base_files; |
4df8dc90 | 1688 | else |
d62beb7f | 1689 | cfts = cgroup1_base_files; |
ccdca218 | 1690 | |
5faaf05f TH |
1691 | ret = cgroup_addrm_files(&cgrp->self, cgrp, cfts, true); |
1692 | if (ret < 0) | |
1693 | return ret; | |
1694 | } else { | |
1695 | list_for_each_entry(cfts, &css->ss->cfts, node) { | |
1696 | ret = cgroup_addrm_files(css, cgrp, cfts, true); | |
1697 | if (ret < 0) { | |
1698 | failed_cfts = cfts; | |
1699 | goto err; | |
1700 | } | |
ccdca218 TH |
1701 | } |
1702 | } | |
88cb04b9 TH |
1703 | |
1704 | css->flags |= CSS_VISIBLE; | |
1705 | ||
ccdca218 TH |
1706 | return 0; |
1707 | err: | |
4df8dc90 TH |
1708 | list_for_each_entry(cfts, &css->ss->cfts, node) { |
1709 | if (cfts == failed_cfts) | |
1710 | break; | |
1711 | cgroup_addrm_files(css, cgrp, cfts, false); | |
1712 | } | |
ccdca218 TH |
1713 | return ret; |
1714 | } | |
1715 | ||
0a268dbd | 1716 | int rebind_subsystems(struct cgroup_root *dst_root, u16 ss_mask) |
ddbcc7e8 | 1717 | { |
1ada4838 | 1718 | struct cgroup *dcgrp = &dst_root->cgrp; |
30159ec7 | 1719 | struct cgroup_subsys *ss; |
2d8f243a | 1720 | int ssid, i, ret; |
ddbcc7e8 | 1721 | |
ace2bee8 | 1722 | lockdep_assert_held(&cgroup_mutex); |
ddbcc7e8 | 1723 | |
b4e0eeaf | 1724 | do_each_subsys_mask(ss, ssid, ss_mask) { |
f6d635ad TH |
1725 | /* |
1726 | * If @ss has non-root csses attached to it, can't move. | |
1727 | * If @ss is an implicit controller, it is exempt from this | |
1728 | * rule and can be stolen. | |
1729 | */ | |
1730 | if (css_next_child(NULL, cgroup_css(&ss->root->cgrp, ss)) && | |
1731 | !ss->implicit_on_dfl) | |
3ed80a62 | 1732 | return -EBUSY; |
1d5be6b2 | 1733 | |
5df36032 | 1734 | /* can't move between two non-dummy roots either */ |
7fd8c565 | 1735 | if (ss->root != &cgrp_dfl_root && dst_root != &cgrp_dfl_root) |
5df36032 | 1736 | return -EBUSY; |
b4e0eeaf | 1737 | } while_each_subsys_mask(); |
ddbcc7e8 | 1738 | |
b4e0eeaf | 1739 | do_each_subsys_mask(ss, ssid, ss_mask) { |
1ada4838 TH |
1740 | struct cgroup_root *src_root = ss->root; |
1741 | struct cgroup *scgrp = &src_root->cgrp; | |
1742 | struct cgroup_subsys_state *css = cgroup_css(scgrp, ss); | |
2d8f243a | 1743 | struct css_set *cset; |
a8a648c4 | 1744 | |
1ada4838 | 1745 | WARN_ON(!css || cgroup_css(dcgrp, ss)); |
a8a648c4 | 1746 | |
334c3679 TH |
1747 | /* disable from the source */ |
1748 | src_root->subsys_mask &= ~(1 << ssid); | |
1749 | WARN_ON(cgroup_apply_control(scgrp)); | |
1750 | cgroup_finalize_control(scgrp, 0); | |
4df8dc90 | 1751 | |
334c3679 | 1752 | /* rebind */ |
1ada4838 TH |
1753 | RCU_INIT_POINTER(scgrp->subsys[ssid], NULL); |
1754 | rcu_assign_pointer(dcgrp->subsys[ssid], css); | |
5df36032 | 1755 | ss->root = dst_root; |
1ada4838 | 1756 | css->cgroup = dcgrp; |
73e80ed8 | 1757 | |
82d6489d | 1758 | spin_lock_irq(&css_set_lock); |
2d8f243a TH |
1759 | hash_for_each(css_set_table, i, cset, hlist) |
1760 | list_move_tail(&cset->e_cset_node[ss->id], | |
1ada4838 | 1761 | &dcgrp->e_csets[ss->id]); |
82d6489d | 1762 | spin_unlock_irq(&css_set_lock); |
2d8f243a | 1763 | |
bd53d617 | 1764 | /* default hierarchy doesn't enable controllers by default */ |
f392e51c | 1765 | dst_root->subsys_mask |= 1 << ssid; |
49d1dc4b TH |
1766 | if (dst_root == &cgrp_dfl_root) { |
1767 | static_branch_enable(cgroup_subsys_on_dfl_key[ssid]); | |
1768 | } else { | |
1ada4838 | 1769 | dcgrp->subtree_control |= 1 << ssid; |
49d1dc4b | 1770 | static_branch_disable(cgroup_subsys_on_dfl_key[ssid]); |
667c2491 | 1771 | } |
a8a648c4 | 1772 | |
334c3679 TH |
1773 | ret = cgroup_apply_control(dcgrp); |
1774 | if (ret) | |
1775 | pr_warn("partial failure to rebind %s controller (err=%d)\n", | |
1776 | ss->name, ret); | |
1777 | ||
5df36032 TH |
1778 | if (ss->bind) |
1779 | ss->bind(css); | |
b4e0eeaf | 1780 | } while_each_subsys_mask(); |
ddbcc7e8 | 1781 | |
1ada4838 | 1782 | kernfs_activate(dcgrp->kn); |
ddbcc7e8 PM |
1783 | return 0; |
1784 | } | |
1785 | ||
1592c9b2 TH |
1786 | int cgroup_show_path(struct seq_file *sf, struct kernfs_node *kf_node, |
1787 | struct kernfs_root *kf_root) | |
4f41fc59 | 1788 | { |
09be4c82 | 1789 | int len = 0; |
4f41fc59 SH |
1790 | char *buf = NULL; |
1791 | struct cgroup_root *kf_cgroot = cgroup_root_from_kf(kf_root); | |
1792 | struct cgroup *ns_cgroup; | |
1793 | ||
1794 | buf = kmalloc(PATH_MAX, GFP_KERNEL); | |
1795 | if (!buf) | |
1796 | return -ENOMEM; | |
1797 | ||
82d6489d | 1798 | spin_lock_irq(&css_set_lock); |
4f41fc59 SH |
1799 | ns_cgroup = current_cgns_cgroup_from_root(kf_cgroot); |
1800 | len = kernfs_path_from_node(kf_node, ns_cgroup->kn, buf, PATH_MAX); | |
82d6489d | 1801 | spin_unlock_irq(&css_set_lock); |
4f41fc59 SH |
1802 | |
1803 | if (len >= PATH_MAX) | |
1804 | len = -ERANGE; | |
1805 | else if (len > 0) { | |
1806 | seq_escape(sf, buf, " \t\n\\"); | |
1807 | len = 0; | |
1808 | } | |
1809 | kfree(buf); | |
1810 | return len; | |
1811 | } | |
1812 | ||
e34a98d5 AV |
1813 | enum cgroup2_param { |
1814 | Opt_nsdelegate, | |
9852ae3f | 1815 | Opt_memory_localevents, |
e34a98d5 AV |
1816 | nr__cgroup2_params |
1817 | }; | |
5136f636 | 1818 | |
d7167b14 | 1819 | static const struct fs_parameter_spec cgroup2_fs_parameters[] = { |
9852ae3f CD |
1820 | fsparam_flag("nsdelegate", Opt_nsdelegate), |
1821 | fsparam_flag("memory_localevents", Opt_memory_localevents), | |
e34a98d5 AV |
1822 | {} |
1823 | }; | |
5136f636 | 1824 | |
e34a98d5 AV |
1825 | static int cgroup2_parse_param(struct fs_context *fc, struct fs_parameter *param) |
1826 | { | |
1827 | struct cgroup_fs_context *ctx = cgroup_fc2context(fc); | |
1828 | struct fs_parse_result result; | |
1829 | int opt; | |
5136f636 | 1830 | |
d7167b14 | 1831 | opt = fs_parse(fc, cgroup2_fs_parameters, param, &result); |
e34a98d5 AV |
1832 | if (opt < 0) |
1833 | return opt; | |
5136f636 | 1834 | |
e34a98d5 AV |
1835 | switch (opt) { |
1836 | case Opt_nsdelegate: | |
1837 | ctx->flags |= CGRP_ROOT_NS_DELEGATE; | |
1838 | return 0; | |
9852ae3f CD |
1839 | case Opt_memory_localevents: |
1840 | ctx->flags |= CGRP_ROOT_MEMORY_LOCAL_EVENTS; | |
1841 | return 0; | |
e34a98d5 AV |
1842 | } |
1843 | return -EINVAL; | |
5136f636 TH |
1844 | } |
1845 | ||
1846 | static void apply_cgroup_root_flags(unsigned int root_flags) | |
1847 | { | |
1848 | if (current->nsproxy->cgroup_ns == &init_cgroup_ns) { | |
1849 | if (root_flags & CGRP_ROOT_NS_DELEGATE) | |
1850 | cgrp_dfl_root.flags |= CGRP_ROOT_NS_DELEGATE; | |
1851 | else | |
1852 | cgrp_dfl_root.flags &= ~CGRP_ROOT_NS_DELEGATE; | |
9852ae3f CD |
1853 | |
1854 | if (root_flags & CGRP_ROOT_MEMORY_LOCAL_EVENTS) | |
1855 | cgrp_dfl_root.flags |= CGRP_ROOT_MEMORY_LOCAL_EVENTS; | |
1856 | else | |
1857 | cgrp_dfl_root.flags &= ~CGRP_ROOT_MEMORY_LOCAL_EVENTS; | |
5136f636 TH |
1858 | } |
1859 | } | |
1860 | ||
1861 | static int cgroup_show_options(struct seq_file *seq, struct kernfs_root *kf_root) | |
1862 | { | |
1863 | if (cgrp_dfl_root.flags & CGRP_ROOT_NS_DELEGATE) | |
1864 | seq_puts(seq, ",nsdelegate"); | |
9852ae3f CD |
1865 | if (cgrp_dfl_root.flags & CGRP_ROOT_MEMORY_LOCAL_EVENTS) |
1866 | seq_puts(seq, ",memory_localevents"); | |
5136f636 TH |
1867 | return 0; |
1868 | } | |
1869 | ||
90129625 | 1870 | static int cgroup_reconfigure(struct fs_context *fc) |
ddbcc7e8 | 1871 | { |
90129625 | 1872 | struct cgroup_fs_context *ctx = cgroup_fc2context(fc); |
5136f636 | 1873 | |
f5dfb531 | 1874 | apply_cgroup_root_flags(ctx->flags); |
5136f636 | 1875 | return 0; |
ddbcc7e8 PM |
1876 | } |
1877 | ||
cc31edce PM |
1878 | static void init_cgroup_housekeeping(struct cgroup *cgrp) |
1879 | { | |
2d8f243a TH |
1880 | struct cgroup_subsys *ss; |
1881 | int ssid; | |
1882 | ||
d5c419b6 TH |
1883 | INIT_LIST_HEAD(&cgrp->self.sibling); |
1884 | INIT_LIST_HEAD(&cgrp->self.children); | |
69d0206c | 1885 | INIT_LIST_HEAD(&cgrp->cset_links); |
72a8cb30 BB |
1886 | INIT_LIST_HEAD(&cgrp->pidlists); |
1887 | mutex_init(&cgrp->pidlist_mutex); | |
9d800df1 | 1888 | cgrp->self.cgroup = cgrp; |
184faf32 | 1889 | cgrp->self.flags |= CSS_ONLINE; |
454000ad | 1890 | cgrp->dom_cgrp = cgrp; |
1a926e0b RG |
1891 | cgrp->max_descendants = INT_MAX; |
1892 | cgrp->max_depth = INT_MAX; | |
8f53470b | 1893 | INIT_LIST_HEAD(&cgrp->rstat_css_list); |
d4ff749b | 1894 | prev_cputime_init(&cgrp->prev_cputime); |
2d8f243a TH |
1895 | |
1896 | for_each_subsys(ss, ssid) | |
1897 | INIT_LIST_HEAD(&cgrp->e_csets[ssid]); | |
f8f22e53 TH |
1898 | |
1899 | init_waitqueue_head(&cgrp->offline_waitq); | |
d62beb7f | 1900 | INIT_WORK(&cgrp->release_agent_work, cgroup1_release_agent); |
cc31edce | 1901 | } |
c6d57f33 | 1902 | |
cf6299b1 | 1903 | void init_cgroup_root(struct cgroup_fs_context *ctx) |
ddbcc7e8 | 1904 | { |
cf6299b1 | 1905 | struct cgroup_root *root = ctx->root; |
3dd06ffa | 1906 | struct cgroup *cgrp = &root->cgrp; |
b0ca5a84 | 1907 | |
ddbcc7e8 | 1908 | INIT_LIST_HEAD(&root->root_list); |
3c9c825b | 1909 | atomic_set(&root->nr_cgrps, 1); |
bd89aabc | 1910 | cgrp->root = root; |
cc31edce | 1911 | init_cgroup_housekeeping(cgrp); |
c6d57f33 | 1912 | |
f5dfb531 AV |
1913 | root->flags = ctx->flags; |
1914 | if (ctx->release_agent) | |
1915 | strscpy(root->release_agent_path, ctx->release_agent, PATH_MAX); | |
1916 | if (ctx->name) | |
1917 | strscpy(root->name, ctx->name, MAX_CGROUP_ROOT_NAMELEN); | |
1918 | if (ctx->cpuset_clone_children) | |
3dd06ffa | 1919 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags); |
c6d57f33 PM |
1920 | } |
1921 | ||
35ac1184 | 1922 | int cgroup_setup_root(struct cgroup_root *root, u16 ss_mask) |
2c6ab6d2 | 1923 | { |
d427dfeb | 1924 | LIST_HEAD(tmp_links); |
3dd06ffa | 1925 | struct cgroup *root_cgrp = &root->cgrp; |
fa069904 | 1926 | struct kernfs_syscall_ops *kf_sops; |
d427dfeb | 1927 | struct css_set *cset; |
d427dfeb | 1928 | int i, ret; |
2c6ab6d2 | 1929 | |
d427dfeb | 1930 | lockdep_assert_held(&cgroup_mutex); |
c6d57f33 | 1931 | |
9732adc5 | 1932 | ret = percpu_ref_init(&root_cgrp->self.refcnt, css_release, |
35ac1184 | 1933 | 0, GFP_KERNEL); |
9d755d33 TH |
1934 | if (ret) |
1935 | goto out; | |
1936 | ||
d427dfeb | 1937 | /* |
f0d9a5f1 | 1938 | * We're accessing css_set_count without locking css_set_lock here, |
d427dfeb | 1939 | * but that's OK - it can only be increased by someone holding |
04313591 TH |
1940 | * cgroup_lock, and that's us. Later rebinding may disable |
1941 | * controllers on the default hierarchy and thus create new csets, | |
1942 | * which can't be more than the existing ones. Allocate 2x. | |
d427dfeb | 1943 | */ |
04313591 | 1944 | ret = allocate_cgrp_cset_links(2 * css_set_count, &tmp_links); |
d427dfeb | 1945 | if (ret) |
9d755d33 | 1946 | goto cancel_ref; |
ddbcc7e8 | 1947 | |
985ed670 | 1948 | ret = cgroup_init_root_id(root); |
ddbcc7e8 | 1949 | if (ret) |
9d755d33 | 1950 | goto cancel_ref; |
ddbcc7e8 | 1951 | |
fa069904 TH |
1952 | kf_sops = root == &cgrp_dfl_root ? |
1953 | &cgroup_kf_syscall_ops : &cgroup1_kf_syscall_ops; | |
1954 | ||
1955 | root->kf_root = kernfs_create_root(kf_sops, | |
aa818825 SL |
1956 | KERNFS_ROOT_CREATE_DEACTIVATED | |
1957 | KERNFS_ROOT_SUPPORT_EXPORTOP, | |
2bd59d48 TH |
1958 | root_cgrp); |
1959 | if (IS_ERR(root->kf_root)) { | |
1960 | ret = PTR_ERR(root->kf_root); | |
1961 | goto exit_root_id; | |
1962 | } | |
1963 | root_cgrp->kn = root->kf_root->kn; | |
d7495343 | 1964 | WARN_ON_ONCE(cgroup_ino(root_cgrp) != 1); |
74321038 | 1965 | root_cgrp->ancestor_ids[0] = cgroup_id(root_cgrp); |
ddbcc7e8 | 1966 | |
334c3679 | 1967 | ret = css_populate_dir(&root_cgrp->self); |
d427dfeb | 1968 | if (ret) |
2bd59d48 | 1969 | goto destroy_root; |
ddbcc7e8 | 1970 | |
5df36032 | 1971 | ret = rebind_subsystems(root, ss_mask); |
d427dfeb | 1972 | if (ret) |
2bd59d48 | 1973 | goto destroy_root; |
ddbcc7e8 | 1974 | |
324bda9e AS |
1975 | ret = cgroup_bpf_inherit(root_cgrp); |
1976 | WARN_ON_ONCE(ret); | |
1977 | ||
ed1777de TH |
1978 | trace_cgroup_setup_root(root); |
1979 | ||
d427dfeb TH |
1980 | /* |
1981 | * There must be no failure case after here, since rebinding takes | |
1982 | * care of subsystems' refcounts, which are explicitly dropped in | |
1983 | * the failure exit path. | |
1984 | */ | |
1985 | list_add(&root->root_list, &cgroup_roots); | |
1986 | cgroup_root_count++; | |
0df6a63f | 1987 | |
d427dfeb | 1988 | /* |
3dd06ffa | 1989 | * Link the root cgroup in this hierarchy into all the css_set |
d427dfeb TH |
1990 | * objects. |
1991 | */ | |
82d6489d | 1992 | spin_lock_irq(&css_set_lock); |
0de0942d | 1993 | hash_for_each(css_set_table, i, cset, hlist) { |
d427dfeb | 1994 | link_css_set(&tmp_links, cset, root_cgrp); |
0de0942d TH |
1995 | if (css_set_populated(cset)) |
1996 | cgroup_update_populated(root_cgrp, true); | |
1997 | } | |
82d6489d | 1998 | spin_unlock_irq(&css_set_lock); |
ddbcc7e8 | 1999 | |
d5c419b6 | 2000 | BUG_ON(!list_empty(&root_cgrp->self.children)); |
3c9c825b | 2001 | BUG_ON(atomic_read(&root->nr_cgrps) != 1); |
ddbcc7e8 | 2002 | |
2bd59d48 | 2003 | kernfs_activate(root_cgrp->kn); |
d427dfeb | 2004 | ret = 0; |
2bd59d48 | 2005 | goto out; |
d427dfeb | 2006 | |
2bd59d48 TH |
2007 | destroy_root: |
2008 | kernfs_destroy_root(root->kf_root); | |
2009 | root->kf_root = NULL; | |
2010 | exit_root_id: | |
d427dfeb | 2011 | cgroup_exit_root_id(root); |
9d755d33 | 2012 | cancel_ref: |
9a1049da | 2013 | percpu_ref_exit(&root_cgrp->self.refcnt); |
2bd59d48 | 2014 | out: |
d427dfeb TH |
2015 | free_cgrp_cset_links(&tmp_links); |
2016 | return ret; | |
ddbcc7e8 PM |
2017 | } |
2018 | ||
cca8f327 | 2019 | int cgroup_do_get_tree(struct fs_context *fc) |
ddbcc7e8 | 2020 | { |
71d883c3 | 2021 | struct cgroup_fs_context *ctx = cgroup_fc2context(fc); |
23bf1b6b | 2022 | int ret; |
ddbcc7e8 | 2023 | |
23bf1b6b | 2024 | ctx->kfc.root = ctx->root->kf_root; |
cca8f327 | 2025 | if (fc->fs_type == &cgroup2_fs_type) |
23bf1b6b | 2026 | ctx->kfc.magic = CGROUP2_SUPER_MAGIC; |
cca8f327 | 2027 | else |
23bf1b6b DH |
2028 | ctx->kfc.magic = CGROUP_SUPER_MAGIC; |
2029 | ret = kernfs_get_tree(fc); | |
ed82571b | 2030 | |
56fde9e0 | 2031 | /* |
633feee3 TH |
2032 | * In non-init cgroup namespace, instead of root cgroup's dentry, |
2033 | * we return the dentry corresponding to the cgroupns->root_cgrp. | |
56fde9e0 | 2034 | */ |
cca8f327 | 2035 | if (!ret && ctx->ns != &init_cgroup_ns) { |
633feee3 | 2036 | struct dentry *nsdentry; |
71d883c3 | 2037 | struct super_block *sb = fc->root->d_sb; |
633feee3 | 2038 | struct cgroup *cgrp; |
e37a06f1 | 2039 | |
633feee3 TH |
2040 | mutex_lock(&cgroup_mutex); |
2041 | spin_lock_irq(&css_set_lock); | |
2042 | ||
cca8f327 | 2043 | cgrp = cset_cgroup_from_root(ctx->ns->root_cset, ctx->root); |
633feee3 TH |
2044 | |
2045 | spin_unlock_irq(&css_set_lock); | |
2046 | mutex_unlock(&cgroup_mutex); | |
2047 | ||
399504e2 | 2048 | nsdentry = kernfs_node_dentry(cgrp->kn, sb); |
71d883c3 | 2049 | dput(fc->root); |
71d883c3 | 2050 | if (IS_ERR(nsdentry)) { |
399504e2 | 2051 | deactivate_locked_super(sb); |
630faf81 AV |
2052 | ret = PTR_ERR(nsdentry); |
2053 | nsdentry = NULL; | |
71d883c3 | 2054 | } |
630faf81 | 2055 | fc->root = nsdentry; |
67e9c74b TH |
2056 | } |
2057 | ||
23bf1b6b | 2058 | if (!ctx->kfc.new_sb_created) |
71d883c3 | 2059 | cgroup_put(&ctx->root->cgrp); |
633feee3 | 2060 | |
71d883c3 | 2061 | return ret; |
633feee3 TH |
2062 | } |
2063 | ||
90129625 AV |
2064 | /* |
2065 | * Destroy a cgroup filesystem context. | |
2066 | */ | |
2067 | static void cgroup_fs_context_free(struct fs_context *fc) | |
ddbcc7e8 | 2068 | { |
90129625 AV |
2069 | struct cgroup_fs_context *ctx = cgroup_fc2context(fc); |
2070 | ||
f5dfb531 AV |
2071 | kfree(ctx->name); |
2072 | kfree(ctx->release_agent); | |
cca8f327 | 2073 | put_cgroup_ns(ctx->ns); |
23bf1b6b | 2074 | kernfs_free_fs_context(fc); |
90129625 AV |
2075 | kfree(ctx); |
2076 | } | |
2077 | ||
90129625 | 2078 | static int cgroup_get_tree(struct fs_context *fc) |
ddbcc7e8 | 2079 | { |
90129625 | 2080 | struct cgroup_fs_context *ctx = cgroup_fc2context(fc); |
5136f636 | 2081 | int ret; |
ddbcc7e8 | 2082 | |
90129625 AV |
2083 | cgrp_dfl_visible = true; |
2084 | cgroup_get_live(&cgrp_dfl_root.cgrp); | |
cf6299b1 | 2085 | ctx->root = &cgrp_dfl_root; |
ed82571b | 2086 | |
cca8f327 | 2087 | ret = cgroup_do_get_tree(fc); |
71d883c3 AV |
2088 | if (!ret) |
2089 | apply_cgroup_root_flags(ctx->flags); | |
2090 | return ret; | |
90129625 AV |
2091 | } |
2092 | ||
90129625 AV |
2093 | static const struct fs_context_operations cgroup_fs_context_ops = { |
2094 | .free = cgroup_fs_context_free, | |
e34a98d5 | 2095 | .parse_param = cgroup2_parse_param, |
90129625 AV |
2096 | .get_tree = cgroup_get_tree, |
2097 | .reconfigure = cgroup_reconfigure, | |
2098 | }; | |
2099 | ||
2100 | static const struct fs_context_operations cgroup1_fs_context_ops = { | |
2101 | .free = cgroup_fs_context_free, | |
8d2451f4 | 2102 | .parse_param = cgroup1_parse_param, |
90129625 AV |
2103 | .get_tree = cgroup1_get_tree, |
2104 | .reconfigure = cgroup1_reconfigure, | |
2105 | }; | |
2106 | ||
2107 | /* | |
23bf1b6b DH |
2108 | * Initialise the cgroup filesystem creation/reconfiguration context. Notably, |
2109 | * we select the namespace we're going to use. | |
90129625 AV |
2110 | */ |
2111 | static int cgroup_init_fs_context(struct fs_context *fc) | |
2112 | { | |
2113 | struct cgroup_fs_context *ctx; | |
2114 | ||
2115 | ctx = kzalloc(sizeof(struct cgroup_fs_context), GFP_KERNEL); | |
2116 | if (!ctx) | |
2117 | return -ENOMEM; | |
ed82571b | 2118 | |
cca8f327 AV |
2119 | ctx->ns = current->nsproxy->cgroup_ns; |
2120 | get_cgroup_ns(ctx->ns); | |
23bf1b6b | 2121 | fc->fs_private = &ctx->kfc; |
90129625 AV |
2122 | if (fc->fs_type == &cgroup2_fs_type) |
2123 | fc->ops = &cgroup_fs_context_ops; | |
2124 | else | |
2125 | fc->ops = &cgroup1_fs_context_ops; | |
f7a99451 | 2126 | put_user_ns(fc->user_ns); |
23bf1b6b DH |
2127 | fc->user_ns = get_user_ns(ctx->ns->user_ns); |
2128 | fc->global = true; | |
90129625 | 2129 | return 0; |
2bd59d48 | 2130 | } |
8e30e2b8 | 2131 | |
2bd59d48 TH |
2132 | static void cgroup_kill_sb(struct super_block *sb) |
2133 | { | |
2134 | struct kernfs_root *kf_root = kernfs_root_from_sb(sb); | |
3dd06ffa | 2135 | struct cgroup_root *root = cgroup_root_from_kf(kf_root); |
a015edd2 | 2136 | |
970317aa | 2137 | /* |
35ac1184 | 2138 | * If @root doesn't have any children, start killing it. |
9d755d33 TH |
2139 | * This prevents new mounts by disabling percpu_ref_tryget_live(). |
2140 | * cgroup_mount() may wait for @root's release. | |
1f779fb2 LZ |
2141 | * |
2142 | * And don't kill the default root. | |
970317aa | 2143 | */ |
35ac1184 AV |
2144 | if (list_empty(&root->cgrp.self.children) && root != &cgrp_dfl_root && |
2145 | !percpu_ref_is_dying(&root->cgrp.self.refcnt)) | |
9d755d33 | 2146 | percpu_ref_kill(&root->cgrp.self.refcnt); |
35ac1184 | 2147 | cgroup_put(&root->cgrp); |
2bd59d48 | 2148 | kernfs_kill_sb(sb); |
ddbcc7e8 | 2149 | } |
970317aa | 2150 | |
0a268dbd | 2151 | struct file_system_type cgroup_fs_type = { |
8d2451f4 AV |
2152 | .name = "cgroup", |
2153 | .init_fs_context = cgroup_init_fs_context, | |
d7167b14 | 2154 | .parameters = cgroup1_fs_parameters, |
8d2451f4 AV |
2155 | .kill_sb = cgroup_kill_sb, |
2156 | .fs_flags = FS_USERNS_MOUNT, | |
ddbcc7e8 | 2157 | }; |
3126121f | 2158 | |
67e9c74b | 2159 | static struct file_system_type cgroup2_fs_type = { |
e34a98d5 AV |
2160 | .name = "cgroup2", |
2161 | .init_fs_context = cgroup_init_fs_context, | |
d7167b14 | 2162 | .parameters = cgroup2_fs_parameters, |
e34a98d5 AV |
2163 | .kill_sb = cgroup_kill_sb, |
2164 | .fs_flags = FS_USERNS_MOUNT, | |
67e9c74b | 2165 | }; |
3126121f | 2166 | |
d5f68d33 AV |
2167 | #ifdef CONFIG_CPUSETS |
2168 | static const struct fs_context_operations cpuset_fs_context_ops = { | |
2169 | .get_tree = cgroup1_get_tree, | |
2170 | .free = cgroup_fs_context_free, | |
2171 | }; | |
2172 | ||
2173 | /* | |
2174 | * This is ugly, but preserves the userspace API for existing cpuset | |
2175 | * users. If someone tries to mount the "cpuset" filesystem, we | |
2176 | * silently switch it to mount "cgroup" instead | |
2177 | */ | |
2178 | static int cpuset_init_fs_context(struct fs_context *fc) | |
2179 | { | |
2180 | char *agent = kstrdup("/sbin/cpuset_release_agent", GFP_USER); | |
2181 | struct cgroup_fs_context *ctx; | |
2182 | int err; | |
2183 | ||
2184 | err = cgroup_init_fs_context(fc); | |
2185 | if (err) { | |
2186 | kfree(agent); | |
2187 | return err; | |
2188 | } | |
2189 | ||
2190 | fc->ops = &cpuset_fs_context_ops; | |
2191 | ||
2192 | ctx = cgroup_fc2context(fc); | |
2193 | ctx->subsys_mask = 1 << cpuset_cgrp_id; | |
2194 | ctx->flags |= CGRP_ROOT_NOPREFIX; | |
2195 | ctx->release_agent = agent; | |
2196 | ||
2197 | get_filesystem(&cgroup_fs_type); | |
2198 | put_filesystem(fc->fs_type); | |
2199 | fc->fs_type = &cgroup_fs_type; | |
2200 | ||
2201 | return 0; | |
2202 | } | |
2203 | ||
2204 | static struct file_system_type cpuset_fs_type = { | |
2205 | .name = "cpuset", | |
2206 | .init_fs_context = cpuset_init_fs_context, | |
2207 | .fs_flags = FS_USERNS_MOUNT, | |
2208 | }; | |
2209 | #endif | |
2210 | ||
0a268dbd TH |
2211 | int cgroup_path_ns_locked(struct cgroup *cgrp, char *buf, size_t buflen, |
2212 | struct cgroup_namespace *ns) | |
a79a908f AK |
2213 | { |
2214 | struct cgroup *root = cset_cgroup_from_root(ns->root_cset, cgrp->root); | |
a79a908f | 2215 | |
4c737b41 | 2216 | return kernfs_path_from_node(cgrp->kn, root->kn, buf, buflen); |
a79a908f AK |
2217 | } |
2218 | ||
4c737b41 TH |
2219 | int cgroup_path_ns(struct cgroup *cgrp, char *buf, size_t buflen, |
2220 | struct cgroup_namespace *ns) | |
a79a908f | 2221 | { |
4c737b41 | 2222 | int ret; |
a79a908f AK |
2223 | |
2224 | mutex_lock(&cgroup_mutex); | |
82d6489d | 2225 | spin_lock_irq(&css_set_lock); |
a79a908f AK |
2226 | |
2227 | ret = cgroup_path_ns_locked(cgrp, buf, buflen, ns); | |
2228 | ||
82d6489d | 2229 | spin_unlock_irq(&css_set_lock); |
a79a908f AK |
2230 | mutex_unlock(&cgroup_mutex); |
2231 | ||
2232 | return ret; | |
2233 | } | |
2234 | EXPORT_SYMBOL_GPL(cgroup_path_ns); | |
2235 | ||
857a2beb | 2236 | /** |
913ffdb5 | 2237 | * task_cgroup_path - cgroup path of a task in the first cgroup hierarchy |
857a2beb | 2238 | * @task: target task |
857a2beb TH |
2239 | * @buf: the buffer to write the path into |
2240 | * @buflen: the length of the buffer | |
2241 | * | |
913ffdb5 TH |
2242 | * Determine @task's cgroup on the first (the one with the lowest non-zero |
2243 | * hierarchy_id) cgroup hierarchy and copy its path into @buf. This | |
2244 | * function grabs cgroup_mutex and shouldn't be used inside locks used by | |
2245 | * cgroup controller callbacks. | |
2246 | * | |
e61734c5 | 2247 | * Return value is the same as kernfs_path(). |
857a2beb | 2248 | */ |
4c737b41 | 2249 | int task_cgroup_path(struct task_struct *task, char *buf, size_t buflen) |
857a2beb | 2250 | { |
3dd06ffa | 2251 | struct cgroup_root *root; |
913ffdb5 | 2252 | struct cgroup *cgrp; |
e61734c5 | 2253 | int hierarchy_id = 1; |
4c737b41 | 2254 | int ret; |
857a2beb TH |
2255 | |
2256 | mutex_lock(&cgroup_mutex); | |
82d6489d | 2257 | spin_lock_irq(&css_set_lock); |
857a2beb | 2258 | |
913ffdb5 TH |
2259 | root = idr_get_next(&cgroup_hierarchy_idr, &hierarchy_id); |
2260 | ||
857a2beb TH |
2261 | if (root) { |
2262 | cgrp = task_cgroup_from_root(task, root); | |
4c737b41 | 2263 | ret = cgroup_path_ns_locked(cgrp, buf, buflen, &init_cgroup_ns); |
913ffdb5 TH |
2264 | } else { |
2265 | /* if no hierarchy exists, everyone is in "/" */ | |
4c737b41 | 2266 | ret = strlcpy(buf, "/", buflen); |
857a2beb TH |
2267 | } |
2268 | ||
82d6489d | 2269 | spin_unlock_irq(&css_set_lock); |
857a2beb | 2270 | mutex_unlock(&cgroup_mutex); |
4c737b41 | 2271 | return ret; |
857a2beb | 2272 | } |
913ffdb5 | 2273 | EXPORT_SYMBOL_GPL(task_cgroup_path); |
857a2beb | 2274 | |
adaae5dc | 2275 | /** |
e595cd70 | 2276 | * cgroup_migrate_add_task - add a migration target task to a migration context |
adaae5dc | 2277 | * @task: target task |
e595cd70 | 2278 | * @mgctx: target migration context |
adaae5dc | 2279 | * |
e595cd70 TH |
2280 | * Add @task, which is a migration target, to @mgctx->tset. This function |
2281 | * becomes noop if @task doesn't need to be migrated. @task's css_set | |
2282 | * should have been added as a migration source and @task->cg_list will be | |
2283 | * moved from the css_set's tasks list to mg_tasks one. | |
adaae5dc | 2284 | */ |
e595cd70 TH |
2285 | static void cgroup_migrate_add_task(struct task_struct *task, |
2286 | struct cgroup_mgctx *mgctx) | |
adaae5dc TH |
2287 | { |
2288 | struct css_set *cset; | |
2289 | ||
f0d9a5f1 | 2290 | lockdep_assert_held(&css_set_lock); |
adaae5dc TH |
2291 | |
2292 | /* @task either already exited or can't exit until the end */ | |
2293 | if (task->flags & PF_EXITING) | |
2294 | return; | |
2295 | ||
5153faac TH |
2296 | /* cgroup_threadgroup_rwsem protects racing against forks */ |
2297 | WARN_ON_ONCE(list_empty(&task->cg_list)); | |
adaae5dc TH |
2298 | |
2299 | cset = task_css_set(task); | |
2300 | if (!cset->mg_src_cgrp) | |
2301 | return; | |
2302 | ||
61046727 TH |
2303 | mgctx->tset.nr_tasks++; |
2304 | ||
adaae5dc TH |
2305 | list_move_tail(&task->cg_list, &cset->mg_tasks); |
2306 | if (list_empty(&cset->mg_node)) | |
e595cd70 TH |
2307 | list_add_tail(&cset->mg_node, |
2308 | &mgctx->tset.src_csets); | |
adaae5dc | 2309 | if (list_empty(&cset->mg_dst_cset->mg_node)) |
d8ebf519 | 2310 | list_add_tail(&cset->mg_dst_cset->mg_node, |
e595cd70 | 2311 | &mgctx->tset.dst_csets); |
adaae5dc TH |
2312 | } |
2313 | ||
2f7ee569 TH |
2314 | /** |
2315 | * cgroup_taskset_first - reset taskset and return the first task | |
2316 | * @tset: taskset of interest | |
1f7dd3e5 | 2317 | * @dst_cssp: output variable for the destination css |
2f7ee569 TH |
2318 | * |
2319 | * @tset iteration is initialized and the first task is returned. | |
2320 | */ | |
1f7dd3e5 TH |
2321 | struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset, |
2322 | struct cgroup_subsys_state **dst_cssp) | |
2f7ee569 | 2323 | { |
b3dc094e TH |
2324 | tset->cur_cset = list_first_entry(tset->csets, struct css_set, mg_node); |
2325 | tset->cur_task = NULL; | |
2326 | ||
1f7dd3e5 | 2327 | return cgroup_taskset_next(tset, dst_cssp); |
2f7ee569 | 2328 | } |
2f7ee569 TH |
2329 | |
2330 | /** | |
2331 | * cgroup_taskset_next - iterate to the next task in taskset | |
2332 | * @tset: taskset of interest | |
1f7dd3e5 | 2333 | * @dst_cssp: output variable for the destination css |
2f7ee569 TH |
2334 | * |
2335 | * Return the next task in @tset. Iteration must have been initialized | |
2336 | * with cgroup_taskset_first(). | |
2337 | */ | |
1f7dd3e5 TH |
2338 | struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset, |
2339 | struct cgroup_subsys_state **dst_cssp) | |
2f7ee569 | 2340 | { |
b3dc094e TH |
2341 | struct css_set *cset = tset->cur_cset; |
2342 | struct task_struct *task = tset->cur_task; | |
2f7ee569 | 2343 | |
b3dc094e TH |
2344 | while (&cset->mg_node != tset->csets) { |
2345 | if (!task) | |
2346 | task = list_first_entry(&cset->mg_tasks, | |
2347 | struct task_struct, cg_list); | |
2348 | else | |
2349 | task = list_next_entry(task, cg_list); | |
2f7ee569 | 2350 | |
b3dc094e TH |
2351 | if (&task->cg_list != &cset->mg_tasks) { |
2352 | tset->cur_cset = cset; | |
2353 | tset->cur_task = task; | |
1f7dd3e5 TH |
2354 | |
2355 | /* | |
2356 | * This function may be called both before and | |
2357 | * after cgroup_taskset_migrate(). The two cases | |
2358 | * can be distinguished by looking at whether @cset | |
2359 | * has its ->mg_dst_cset set. | |
2360 | */ | |
2361 | if (cset->mg_dst_cset) | |
2362 | *dst_cssp = cset->mg_dst_cset->subsys[tset->ssid]; | |
2363 | else | |
2364 | *dst_cssp = cset->subsys[tset->ssid]; | |
2365 | ||
b3dc094e TH |
2366 | return task; |
2367 | } | |
2f7ee569 | 2368 | |
b3dc094e TH |
2369 | cset = list_next_entry(cset, mg_node); |
2370 | task = NULL; | |
2371 | } | |
2f7ee569 | 2372 | |
b3dc094e | 2373 | return NULL; |
2f7ee569 | 2374 | } |
2f7ee569 | 2375 | |
adaae5dc | 2376 | /** |
37ff9f8f | 2377 | * cgroup_taskset_migrate - migrate a taskset |
e595cd70 | 2378 | * @mgctx: migration context |
adaae5dc | 2379 | * |
e595cd70 | 2380 | * Migrate tasks in @mgctx as setup by migration preparation functions. |
37ff9f8f | 2381 | * This function fails iff one of the ->can_attach callbacks fails and |
e595cd70 TH |
2382 | * guarantees that either all or none of the tasks in @mgctx are migrated. |
2383 | * @mgctx is consumed regardless of success. | |
adaae5dc | 2384 | */ |
bfc2cf6f | 2385 | static int cgroup_migrate_execute(struct cgroup_mgctx *mgctx) |
adaae5dc | 2386 | { |
e595cd70 | 2387 | struct cgroup_taskset *tset = &mgctx->tset; |
37ff9f8f | 2388 | struct cgroup_subsys *ss; |
adaae5dc TH |
2389 | struct task_struct *task, *tmp_task; |
2390 | struct css_set *cset, *tmp_cset; | |
37ff9f8f | 2391 | int ssid, failed_ssid, ret; |
adaae5dc | 2392 | |
adaae5dc | 2393 | /* check that we can legitimately attach to the cgroup */ |
61046727 TH |
2394 | if (tset->nr_tasks) { |
2395 | do_each_subsys_mask(ss, ssid, mgctx->ss_mask) { | |
2396 | if (ss->can_attach) { | |
2397 | tset->ssid = ssid; | |
2398 | ret = ss->can_attach(tset); | |
2399 | if (ret) { | |
2400 | failed_ssid = ssid; | |
2401 | goto out_cancel_attach; | |
2402 | } | |
adaae5dc | 2403 | } |
61046727 TH |
2404 | } while_each_subsys_mask(); |
2405 | } | |
adaae5dc TH |
2406 | |
2407 | /* | |
2408 | * Now that we're guaranteed success, proceed to move all tasks to | |
2409 | * the new cgroup. There are no failure cases after here, so this | |
2410 | * is the commit point. | |
2411 | */ | |
82d6489d | 2412 | spin_lock_irq(&css_set_lock); |
adaae5dc | 2413 | list_for_each_entry(cset, &tset->src_csets, mg_node) { |
f6d7d049 TH |
2414 | list_for_each_entry_safe(task, tmp_task, &cset->mg_tasks, cg_list) { |
2415 | struct css_set *from_cset = task_css_set(task); | |
2416 | struct css_set *to_cset = cset->mg_dst_cset; | |
2417 | ||
2418 | get_css_set(to_cset); | |
73a7242a | 2419 | to_cset->nr_tasks++; |
f6d7d049 | 2420 | css_set_move_task(task, from_cset, to_cset, true); |
73a7242a | 2421 | from_cset->nr_tasks--; |
76f969e8 RG |
2422 | /* |
2423 | * If the source or destination cgroup is frozen, | |
2424 | * the task might require to change its state. | |
2425 | */ | |
2426 | cgroup_freezer_migrate_task(task, from_cset->dfl_cgrp, | |
2427 | to_cset->dfl_cgrp); | |
2428 | put_css_set_locked(from_cset); | |
2429 | ||
f6d7d049 | 2430 | } |
adaae5dc | 2431 | } |
82d6489d | 2432 | spin_unlock_irq(&css_set_lock); |
adaae5dc TH |
2433 | |
2434 | /* | |
2435 | * Migration is committed, all target tasks are now on dst_csets. | |
2436 | * Nothing is sensitive to fork() after this point. Notify | |
2437 | * controllers that migration is complete. | |
2438 | */ | |
2439 | tset->csets = &tset->dst_csets; | |
2440 | ||
61046727 TH |
2441 | if (tset->nr_tasks) { |
2442 | do_each_subsys_mask(ss, ssid, mgctx->ss_mask) { | |
2443 | if (ss->attach) { | |
2444 | tset->ssid = ssid; | |
2445 | ss->attach(tset); | |
2446 | } | |
2447 | } while_each_subsys_mask(); | |
2448 | } | |
adaae5dc TH |
2449 | |
2450 | ret = 0; | |
2451 | goto out_release_tset; | |
2452 | ||
2453 | out_cancel_attach: | |
61046727 TH |
2454 | if (tset->nr_tasks) { |
2455 | do_each_subsys_mask(ss, ssid, mgctx->ss_mask) { | |
2456 | if (ssid == failed_ssid) | |
2457 | break; | |
2458 | if (ss->cancel_attach) { | |
2459 | tset->ssid = ssid; | |
2460 | ss->cancel_attach(tset); | |
2461 | } | |
2462 | } while_each_subsys_mask(); | |
2463 | } | |
adaae5dc | 2464 | out_release_tset: |
82d6489d | 2465 | spin_lock_irq(&css_set_lock); |
adaae5dc TH |
2466 | list_splice_init(&tset->dst_csets, &tset->src_csets); |
2467 | list_for_each_entry_safe(cset, tmp_cset, &tset->src_csets, mg_node) { | |
2468 | list_splice_tail_init(&cset->mg_tasks, &cset->tasks); | |
2469 | list_del_init(&cset->mg_node); | |
2470 | } | |
82d6489d | 2471 | spin_unlock_irq(&css_set_lock); |
c4fa6c43 WL |
2472 | |
2473 | /* | |
2474 | * Re-initialize the cgroup_taskset structure in case it is reused | |
2475 | * again in another cgroup_migrate_add_task()/cgroup_migrate_execute() | |
2476 | * iteration. | |
2477 | */ | |
2478 | tset->nr_tasks = 0; | |
2479 | tset->csets = &tset->src_csets; | |
adaae5dc TH |
2480 | return ret; |
2481 | } | |
2482 | ||
6c694c88 | 2483 | /** |
8cfd8147 | 2484 | * cgroup_migrate_vet_dst - verify whether a cgroup can be migration destination |
6c694c88 TH |
2485 | * @dst_cgrp: destination cgroup to test |
2486 | * | |
8cfd8147 TH |
2487 | * On the default hierarchy, except for the mixable, (possible) thread root |
2488 | * and threaded cgroups, subtree_control must be zero for migration | |
2489 | * destination cgroups with tasks so that child cgroups don't compete | |
2490 | * against tasks. | |
6c694c88 | 2491 | */ |
8cfd8147 | 2492 | int cgroup_migrate_vet_dst(struct cgroup *dst_cgrp) |
6c694c88 | 2493 | { |
8cfd8147 TH |
2494 | /* v1 doesn't have any restriction */ |
2495 | if (!cgroup_on_dfl(dst_cgrp)) | |
2496 | return 0; | |
2497 | ||
2498 | /* verify @dst_cgrp can host resources */ | |
2499 | if (!cgroup_is_valid_domain(dst_cgrp->dom_cgrp)) | |
2500 | return -EOPNOTSUPP; | |
2501 | ||
2502 | /* mixables don't care */ | |
2503 | if (cgroup_is_mixable(dst_cgrp)) | |
2504 | return 0; | |
2505 | ||
2506 | /* | |
2507 | * If @dst_cgrp is already or can become a thread root or is | |
2508 | * threaded, it doesn't matter. | |
2509 | */ | |
2510 | if (cgroup_can_be_thread_root(dst_cgrp) || cgroup_is_threaded(dst_cgrp)) | |
2511 | return 0; | |
2512 | ||
2513 | /* apply no-internal-process constraint */ | |
2514 | if (dst_cgrp->subtree_control) | |
2515 | return -EBUSY; | |
2516 | ||
2517 | return 0; | |
6c694c88 TH |
2518 | } |
2519 | ||
a043e3b2 | 2520 | /** |
1958d2d5 | 2521 | * cgroup_migrate_finish - cleanup after attach |
e595cd70 | 2522 | * @mgctx: migration context |
74a1166d | 2523 | * |
1958d2d5 TH |
2524 | * Undo cgroup_migrate_add_src() and cgroup_migrate_prepare_dst(). See |
2525 | * those functions for details. | |
74a1166d | 2526 | */ |
e595cd70 | 2527 | void cgroup_migrate_finish(struct cgroup_mgctx *mgctx) |
74a1166d | 2528 | { |
e595cd70 | 2529 | LIST_HEAD(preloaded); |
1958d2d5 | 2530 | struct css_set *cset, *tmp_cset; |
74a1166d | 2531 | |
1958d2d5 TH |
2532 | lockdep_assert_held(&cgroup_mutex); |
2533 | ||
82d6489d | 2534 | spin_lock_irq(&css_set_lock); |
e595cd70 TH |
2535 | |
2536 | list_splice_tail_init(&mgctx->preloaded_src_csets, &preloaded); | |
2537 | list_splice_tail_init(&mgctx->preloaded_dst_csets, &preloaded); | |
2538 | ||
2539 | list_for_each_entry_safe(cset, tmp_cset, &preloaded, mg_preload_node) { | |
1958d2d5 | 2540 | cset->mg_src_cgrp = NULL; |
e4857982 | 2541 | cset->mg_dst_cgrp = NULL; |
1958d2d5 TH |
2542 | cset->mg_dst_cset = NULL; |
2543 | list_del_init(&cset->mg_preload_node); | |
a25eb52e | 2544 | put_css_set_locked(cset); |
1958d2d5 | 2545 | } |
e595cd70 | 2546 | |
82d6489d | 2547 | spin_unlock_irq(&css_set_lock); |
1958d2d5 TH |
2548 | } |
2549 | ||
2550 | /** | |
2551 | * cgroup_migrate_add_src - add a migration source css_set | |
2552 | * @src_cset: the source css_set to add | |
2553 | * @dst_cgrp: the destination cgroup | |
e595cd70 | 2554 | * @mgctx: migration context |
1958d2d5 TH |
2555 | * |
2556 | * Tasks belonging to @src_cset are about to be migrated to @dst_cgrp. Pin | |
e595cd70 | 2557 | * @src_cset and add it to @mgctx->src_csets, which should later be cleaned |
1958d2d5 TH |
2558 | * up by cgroup_migrate_finish(). |
2559 | * | |
1ed13287 TH |
2560 | * This function may be called without holding cgroup_threadgroup_rwsem |
2561 | * even if the target is a process. Threads may be created and destroyed | |
2562 | * but as long as cgroup_mutex is not dropped, no new css_set can be put | |
2563 | * into play and the preloaded css_sets are guaranteed to cover all | |
2564 | * migrations. | |
1958d2d5 | 2565 | */ |
0a268dbd TH |
2566 | void cgroup_migrate_add_src(struct css_set *src_cset, |
2567 | struct cgroup *dst_cgrp, | |
e595cd70 | 2568 | struct cgroup_mgctx *mgctx) |
1958d2d5 TH |
2569 | { |
2570 | struct cgroup *src_cgrp; | |
2571 | ||
2572 | lockdep_assert_held(&cgroup_mutex); | |
f0d9a5f1 | 2573 | lockdep_assert_held(&css_set_lock); |
1958d2d5 | 2574 | |
2b021cbf TH |
2575 | /* |
2576 | * If ->dead, @src_set is associated with one or more dead cgroups | |
2577 | * and doesn't contain any migratable tasks. Ignore it early so | |
2578 | * that the rest of migration path doesn't get confused by it. | |
2579 | */ | |
2580 | if (src_cset->dead) | |
2581 | return; | |
2582 | ||
1958d2d5 TH |
2583 | src_cgrp = cset_cgroup_from_root(src_cset, dst_cgrp->root); |
2584 | ||
1958d2d5 TH |
2585 | if (!list_empty(&src_cset->mg_preload_node)) |
2586 | return; | |
2587 | ||
2588 | WARN_ON(src_cset->mg_src_cgrp); | |
e4857982 | 2589 | WARN_ON(src_cset->mg_dst_cgrp); |
1958d2d5 TH |
2590 | WARN_ON(!list_empty(&src_cset->mg_tasks)); |
2591 | WARN_ON(!list_empty(&src_cset->mg_node)); | |
2592 | ||
2593 | src_cset->mg_src_cgrp = src_cgrp; | |
e4857982 | 2594 | src_cset->mg_dst_cgrp = dst_cgrp; |
1958d2d5 | 2595 | get_css_set(src_cset); |
e595cd70 | 2596 | list_add_tail(&src_cset->mg_preload_node, &mgctx->preloaded_src_csets); |
1958d2d5 TH |
2597 | } |
2598 | ||
2599 | /** | |
2600 | * cgroup_migrate_prepare_dst - prepare destination css_sets for migration | |
e595cd70 | 2601 | * @mgctx: migration context |
1958d2d5 | 2602 | * |
e4857982 | 2603 | * Tasks are about to be moved and all the source css_sets have been |
e595cd70 TH |
2604 | * preloaded to @mgctx->preloaded_src_csets. This function looks up and |
2605 | * pins all destination css_sets, links each to its source, and append them | |
2606 | * to @mgctx->preloaded_dst_csets. | |
1958d2d5 TH |
2607 | * |
2608 | * This function must be called after cgroup_migrate_add_src() has been | |
2609 | * called on each migration source css_set. After migration is performed | |
2610 | * using cgroup_migrate(), cgroup_migrate_finish() must be called on | |
e595cd70 | 2611 | * @mgctx. |
1958d2d5 | 2612 | */ |
e595cd70 | 2613 | int cgroup_migrate_prepare_dst(struct cgroup_mgctx *mgctx) |
1958d2d5 | 2614 | { |
f817de98 | 2615 | struct css_set *src_cset, *tmp_cset; |
1958d2d5 TH |
2616 | |
2617 | lockdep_assert_held(&cgroup_mutex); | |
2618 | ||
2619 | /* look up the dst cset for each src cset and link it to src */ | |
e595cd70 TH |
2620 | list_for_each_entry_safe(src_cset, tmp_cset, &mgctx->preloaded_src_csets, |
2621 | mg_preload_node) { | |
1958d2d5 | 2622 | struct css_set *dst_cset; |
bfc2cf6f TH |
2623 | struct cgroup_subsys *ss; |
2624 | int ssid; | |
1958d2d5 | 2625 | |
e4857982 | 2626 | dst_cset = find_css_set(src_cset, src_cset->mg_dst_cgrp); |
1958d2d5 | 2627 | if (!dst_cset) |
d6e486ee | 2628 | return -ENOMEM; |
1958d2d5 TH |
2629 | |
2630 | WARN_ON_ONCE(src_cset->mg_dst_cset || dst_cset->mg_dst_cset); | |
f817de98 TH |
2631 | |
2632 | /* | |
2633 | * If src cset equals dst, it's noop. Drop the src. | |
2634 | * cgroup_migrate() will skip the cset too. Note that we | |
2635 | * can't handle src == dst as some nodes are used by both. | |
2636 | */ | |
2637 | if (src_cset == dst_cset) { | |
2638 | src_cset->mg_src_cgrp = NULL; | |
e4857982 | 2639 | src_cset->mg_dst_cgrp = NULL; |
f817de98 | 2640 | list_del_init(&src_cset->mg_preload_node); |
a25eb52e ZL |
2641 | put_css_set(src_cset); |
2642 | put_css_set(dst_cset); | |
f817de98 TH |
2643 | continue; |
2644 | } | |
2645 | ||
1958d2d5 TH |
2646 | src_cset->mg_dst_cset = dst_cset; |
2647 | ||
2648 | if (list_empty(&dst_cset->mg_preload_node)) | |
e595cd70 TH |
2649 | list_add_tail(&dst_cset->mg_preload_node, |
2650 | &mgctx->preloaded_dst_csets); | |
1958d2d5 | 2651 | else |
a25eb52e | 2652 | put_css_set(dst_cset); |
bfc2cf6f TH |
2653 | |
2654 | for_each_subsys(ss, ssid) | |
2655 | if (src_cset->subsys[ssid] != dst_cset->subsys[ssid]) | |
2656 | mgctx->ss_mask |= 1 << ssid; | |
1958d2d5 TH |
2657 | } |
2658 | ||
1958d2d5 | 2659 | return 0; |
1958d2d5 TH |
2660 | } |
2661 | ||
2662 | /** | |
2663 | * cgroup_migrate - migrate a process or task to a cgroup | |
1958d2d5 TH |
2664 | * @leader: the leader of the process or the task to migrate |
2665 | * @threadgroup: whether @leader points to the whole process or a single task | |
e595cd70 | 2666 | * @mgctx: migration context |
1958d2d5 | 2667 | * |
37ff9f8f TH |
2668 | * Migrate a process or task denoted by @leader. If migrating a process, |
2669 | * the caller must be holding cgroup_threadgroup_rwsem. The caller is also | |
2670 | * responsible for invoking cgroup_migrate_add_src() and | |
1958d2d5 TH |
2671 | * cgroup_migrate_prepare_dst() on the targets before invoking this |
2672 | * function and following up with cgroup_migrate_finish(). | |
2673 | * | |
2674 | * As long as a controller's ->can_attach() doesn't fail, this function is | |
2675 | * guaranteed to succeed. This means that, excluding ->can_attach() | |
2676 | * failure, when migrating multiple targets, the success or failure can be | |
2677 | * decided for all targets by invoking group_migrate_prepare_dst() before | |
2678 | * actually starting migrating. | |
2679 | */ | |
0a268dbd | 2680 | int cgroup_migrate(struct task_struct *leader, bool threadgroup, |
bfc2cf6f | 2681 | struct cgroup_mgctx *mgctx) |
74a1166d | 2682 | { |
adaae5dc | 2683 | struct task_struct *task; |
74a1166d | 2684 | |
fb5d2b4c MSB |
2685 | /* |
2686 | * Prevent freeing of tasks while we take a snapshot. Tasks that are | |
2687 | * already PF_EXITING could be freed from underneath us unless we | |
2688 | * take an rcu_read_lock. | |
2689 | */ | |
82d6489d | 2690 | spin_lock_irq(&css_set_lock); |
fb5d2b4c | 2691 | rcu_read_lock(); |
9db8de37 | 2692 | task = leader; |
74a1166d | 2693 | do { |
e595cd70 | 2694 | cgroup_migrate_add_task(task, mgctx); |
081aa458 LZ |
2695 | if (!threadgroup) |
2696 | break; | |
9db8de37 | 2697 | } while_each_thread(leader, task); |
fb5d2b4c | 2698 | rcu_read_unlock(); |
82d6489d | 2699 | spin_unlock_irq(&css_set_lock); |
74a1166d | 2700 | |
bfc2cf6f | 2701 | return cgroup_migrate_execute(mgctx); |
74a1166d BB |
2702 | } |
2703 | ||
1958d2d5 TH |
2704 | /** |
2705 | * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup | |
2706 | * @dst_cgrp: the cgroup to attach to | |
2707 | * @leader: the task or the leader of the threadgroup to be attached | |
2708 | * @threadgroup: attach the whole threadgroup? | |
2709 | * | |
1ed13287 | 2710 | * Call holding cgroup_mutex and cgroup_threadgroup_rwsem. |
1958d2d5 | 2711 | */ |
0a268dbd TH |
2712 | int cgroup_attach_task(struct cgroup *dst_cgrp, struct task_struct *leader, |
2713 | bool threadgroup) | |
1958d2d5 | 2714 | { |
e595cd70 | 2715 | DEFINE_CGROUP_MGCTX(mgctx); |
1958d2d5 TH |
2716 | struct task_struct *task; |
2717 | int ret; | |
2718 | ||
8cfd8147 TH |
2719 | ret = cgroup_migrate_vet_dst(dst_cgrp); |
2720 | if (ret) | |
2721 | return ret; | |
6c694c88 | 2722 | |
1958d2d5 | 2723 | /* look up all src csets */ |
82d6489d | 2724 | spin_lock_irq(&css_set_lock); |
1958d2d5 TH |
2725 | rcu_read_lock(); |
2726 | task = leader; | |
2727 | do { | |
e595cd70 | 2728 | cgroup_migrate_add_src(task_css_set(task), dst_cgrp, &mgctx); |
1958d2d5 TH |
2729 | if (!threadgroup) |
2730 | break; | |
2731 | } while_each_thread(leader, task); | |
2732 | rcu_read_unlock(); | |
82d6489d | 2733 | spin_unlock_irq(&css_set_lock); |
1958d2d5 TH |
2734 | |
2735 | /* prepare dst csets and commit */ | |
e595cd70 | 2736 | ret = cgroup_migrate_prepare_dst(&mgctx); |
1958d2d5 | 2737 | if (!ret) |
bfc2cf6f | 2738 | ret = cgroup_migrate(leader, threadgroup, &mgctx); |
1958d2d5 | 2739 | |
e595cd70 | 2740 | cgroup_migrate_finish(&mgctx); |
ed1777de TH |
2741 | |
2742 | if (!ret) | |
e4f8d81c | 2743 | TRACE_CGROUP_PATH(attach_task, dst_cgrp, leader, threadgroup); |
ed1777de | 2744 | |
1958d2d5 | 2745 | return ret; |
74a1166d BB |
2746 | } |
2747 | ||
9a3284fa MK |
2748 | struct task_struct *cgroup_procs_write_start(char *buf, bool threadgroup, |
2749 | bool *locked) | |
715c809d | 2750 | __acquires(&cgroup_threadgroup_rwsem) |
bbcb81d0 | 2751 | { |
bbcb81d0 | 2752 | struct task_struct *tsk; |
acbef755 | 2753 | pid_t pid; |
bbcb81d0 | 2754 | |
acbef755 | 2755 | if (kstrtoint(strstrip(buf), 0, &pid) || pid < 0) |
715c809d | 2756 | return ERR_PTR(-EINVAL); |
74a1166d | 2757 | |
9a3284fa MK |
2758 | /* |
2759 | * If we migrate a single thread, we don't care about threadgroup | |
2760 | * stability. If the thread is `current`, it won't exit(2) under our | |
2761 | * hands or change PID through exec(2). We exclude | |
2762 | * cgroup_update_dfl_csses and other cgroup_{proc,thread}s_write | |
2763 | * callers by cgroup_mutex. | |
2764 | * Therefore, we can skip the global lock. | |
2765 | */ | |
2766 | lockdep_assert_held(&cgroup_mutex); | |
2767 | if (pid || threadgroup) { | |
2768 | percpu_down_write(&cgroup_threadgroup_rwsem); | |
2769 | *locked = true; | |
2770 | } else { | |
2771 | *locked = false; | |
2772 | } | |
715c809d | 2773 | |
b78949eb | 2774 | rcu_read_lock(); |
bbcb81d0 | 2775 | if (pid) { |
73507f33 | 2776 | tsk = find_task_by_vpid(pid); |
74a1166d | 2777 | if (!tsk) { |
715c809d TH |
2778 | tsk = ERR_PTR(-ESRCH); |
2779 | goto out_unlock_threadgroup; | |
bbcb81d0 | 2780 | } |
dedf22e9 | 2781 | } else { |
b78949eb | 2782 | tsk = current; |
dedf22e9 | 2783 | } |
cd3d0952 TH |
2784 | |
2785 | if (threadgroup) | |
b78949eb | 2786 | tsk = tsk->group_leader; |
c4c27fbd MG |
2787 | |
2788 | /* | |
77f88796 TH |
2789 | * kthreads may acquire PF_NO_SETAFFINITY during initialization. |
2790 | * If userland migrates such a kthread to a non-root cgroup, it can | |
2791 | * become trapped in a cpuset, or RT kthread may be born in a | |
2792 | * cgroup with no rt_runtime allocated. Just say no. | |
c4c27fbd | 2793 | */ |
77f88796 | 2794 | if (tsk->no_cgroup_migration || (tsk->flags & PF_NO_SETAFFINITY)) { |
715c809d TH |
2795 | tsk = ERR_PTR(-EINVAL); |
2796 | goto out_unlock_threadgroup; | |
c4c27fbd MG |
2797 | } |
2798 | ||
b78949eb | 2799 | get_task_struct(tsk); |
715c809d TH |
2800 | goto out_unlock_rcu; |
2801 | ||
2802 | out_unlock_threadgroup: | |
9a3284fa MK |
2803 | if (*locked) { |
2804 | percpu_up_write(&cgroup_threadgroup_rwsem); | |
2805 | *locked = false; | |
2806 | } | |
715c809d | 2807 | out_unlock_rcu: |
b78949eb | 2808 | rcu_read_unlock(); |
715c809d TH |
2809 | return tsk; |
2810 | } | |
b78949eb | 2811 | |
9a3284fa | 2812 | void cgroup_procs_write_finish(struct task_struct *task, bool locked) |
715c809d TH |
2813 | __releases(&cgroup_threadgroup_rwsem) |
2814 | { | |
2815 | struct cgroup_subsys *ss; | |
2816 | int ssid; | |
081aa458 | 2817 | |
715c809d TH |
2818 | /* release reference from cgroup_procs_write_start() */ |
2819 | put_task_struct(task); | |
3014dde7 | 2820 | |
9a3284fa MK |
2821 | if (locked) |
2822 | percpu_up_write(&cgroup_threadgroup_rwsem); | |
5cf1cacb TH |
2823 | for_each_subsys(ss, ssid) |
2824 | if (ss->post_attach) | |
2825 | ss->post_attach(); | |
af351026 PM |
2826 | } |
2827 | ||
6e5c8307 | 2828 | static void cgroup_print_ss_mask(struct seq_file *seq, u16 ss_mask) |
355e0c48 | 2829 | { |
f8f22e53 TH |
2830 | struct cgroup_subsys *ss; |
2831 | bool printed = false; | |
2832 | int ssid; | |
a742c59d | 2833 | |
b4e0eeaf | 2834 | do_each_subsys_mask(ss, ssid, ss_mask) { |
a966a4ed AS |
2835 | if (printed) |
2836 | seq_putc(seq, ' '); | |
85db0023 | 2837 | seq_puts(seq, ss->name); |
a966a4ed | 2838 | printed = true; |
b4e0eeaf | 2839 | } while_each_subsys_mask(); |
f8f22e53 TH |
2840 | if (printed) |
2841 | seq_putc(seq, '\n'); | |
355e0c48 PM |
2842 | } |
2843 | ||
f8f22e53 TH |
2844 | /* show controllers which are enabled from the parent */ |
2845 | static int cgroup_controllers_show(struct seq_file *seq, void *v) | |
ddbcc7e8 | 2846 | { |
f8f22e53 TH |
2847 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
2848 | ||
5531dc91 | 2849 | cgroup_print_ss_mask(seq, cgroup_control(cgrp)); |
f8f22e53 | 2850 | return 0; |
ddbcc7e8 PM |
2851 | } |
2852 | ||
f8f22e53 TH |
2853 | /* show controllers which are enabled for a given cgroup's children */ |
2854 | static int cgroup_subtree_control_show(struct seq_file *seq, void *v) | |
ddbcc7e8 | 2855 | { |
f8f22e53 TH |
2856 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
2857 | ||
667c2491 | 2858 | cgroup_print_ss_mask(seq, cgrp->subtree_control); |
f8f22e53 TH |
2859 | return 0; |
2860 | } | |
2861 | ||
2862 | /** | |
2863 | * cgroup_update_dfl_csses - update css assoc of a subtree in default hierarchy | |
2864 | * @cgrp: root of the subtree to update csses for | |
2865 | * | |
54962604 TH |
2866 | * @cgrp's control masks have changed and its subtree's css associations |
2867 | * need to be updated accordingly. This function looks up all css_sets | |
2868 | * which are attached to the subtree, creates the matching updated css_sets | |
2869 | * and migrates the tasks to the new ones. | |
f8f22e53 TH |
2870 | */ |
2871 | static int cgroup_update_dfl_csses(struct cgroup *cgrp) | |
2872 | { | |
e595cd70 | 2873 | DEFINE_CGROUP_MGCTX(mgctx); |
54962604 TH |
2874 | struct cgroup_subsys_state *d_css; |
2875 | struct cgroup *dsct; | |
f8f22e53 TH |
2876 | struct css_set *src_cset; |
2877 | int ret; | |
2878 | ||
f8f22e53 TH |
2879 | lockdep_assert_held(&cgroup_mutex); |
2880 | ||
3014dde7 TH |
2881 | percpu_down_write(&cgroup_threadgroup_rwsem); |
2882 | ||
f8f22e53 | 2883 | /* look up all csses currently attached to @cgrp's subtree */ |
82d6489d | 2884 | spin_lock_irq(&css_set_lock); |
54962604 | 2885 | cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) { |
f8f22e53 TH |
2886 | struct cgrp_cset_link *link; |
2887 | ||
54962604 | 2888 | list_for_each_entry(link, &dsct->cset_links, cset_link) |
e595cd70 | 2889 | cgroup_migrate_add_src(link->cset, dsct, &mgctx); |
f8f22e53 | 2890 | } |
82d6489d | 2891 | spin_unlock_irq(&css_set_lock); |
f8f22e53 TH |
2892 | |
2893 | /* NULL dst indicates self on default hierarchy */ | |
e595cd70 | 2894 | ret = cgroup_migrate_prepare_dst(&mgctx); |
f8f22e53 TH |
2895 | if (ret) |
2896 | goto out_finish; | |
2897 | ||
82d6489d | 2898 | spin_lock_irq(&css_set_lock); |
e595cd70 | 2899 | list_for_each_entry(src_cset, &mgctx.preloaded_src_csets, mg_preload_node) { |
10265075 | 2900 | struct task_struct *task, *ntask; |
f8f22e53 | 2901 | |
10265075 TH |
2902 | /* all tasks in src_csets need to be migrated */ |
2903 | list_for_each_entry_safe(task, ntask, &src_cset->tasks, cg_list) | |
e595cd70 | 2904 | cgroup_migrate_add_task(task, &mgctx); |
f8f22e53 | 2905 | } |
82d6489d | 2906 | spin_unlock_irq(&css_set_lock); |
f8f22e53 | 2907 | |
bfc2cf6f | 2908 | ret = cgroup_migrate_execute(&mgctx); |
f8f22e53 | 2909 | out_finish: |
e595cd70 | 2910 | cgroup_migrate_finish(&mgctx); |
3014dde7 | 2911 | percpu_up_write(&cgroup_threadgroup_rwsem); |
f8f22e53 TH |
2912 | return ret; |
2913 | } | |
2914 | ||
1b9b96a1 | 2915 | /** |
945ba199 | 2916 | * cgroup_lock_and_drain_offline - lock cgroup_mutex and drain offlined csses |
ce3f1d9d | 2917 | * @cgrp: root of the target subtree |
1b9b96a1 TH |
2918 | * |
2919 | * Because css offlining is asynchronous, userland may try to re-enable a | |
945ba199 TH |
2920 | * controller while the previous css is still around. This function grabs |
2921 | * cgroup_mutex and drains the previous css instances of @cgrp's subtree. | |
1b9b96a1 | 2922 | */ |
0a268dbd | 2923 | void cgroup_lock_and_drain_offline(struct cgroup *cgrp) |
945ba199 | 2924 | __acquires(&cgroup_mutex) |
1b9b96a1 TH |
2925 | { |
2926 | struct cgroup *dsct; | |
ce3f1d9d | 2927 | struct cgroup_subsys_state *d_css; |
1b9b96a1 TH |
2928 | struct cgroup_subsys *ss; |
2929 | int ssid; | |
2930 | ||
945ba199 TH |
2931 | restart: |
2932 | mutex_lock(&cgroup_mutex); | |
1b9b96a1 | 2933 | |
ce3f1d9d | 2934 | cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) { |
1b9b96a1 TH |
2935 | for_each_subsys(ss, ssid) { |
2936 | struct cgroup_subsys_state *css = cgroup_css(dsct, ss); | |
2937 | DEFINE_WAIT(wait); | |
2938 | ||
ce3f1d9d | 2939 | if (!css || !percpu_ref_is_dying(&css->refcnt)) |
1b9b96a1 TH |
2940 | continue; |
2941 | ||
a590b90d | 2942 | cgroup_get_live(dsct); |
1b9b96a1 TH |
2943 | prepare_to_wait(&dsct->offline_waitq, &wait, |
2944 | TASK_UNINTERRUPTIBLE); | |
2945 | ||
2946 | mutex_unlock(&cgroup_mutex); | |
2947 | schedule(); | |
2948 | finish_wait(&dsct->offline_waitq, &wait); | |
1b9b96a1 TH |
2949 | |
2950 | cgroup_put(dsct); | |
945ba199 | 2951 | goto restart; |
1b9b96a1 TH |
2952 | } |
2953 | } | |
1b9b96a1 TH |
2954 | } |
2955 | ||
15a27c36 | 2956 | /** |
479adb89 | 2957 | * cgroup_save_control - save control masks and dom_cgrp of a subtree |
15a27c36 TH |
2958 | * @cgrp: root of the target subtree |
2959 | * | |
479adb89 TH |
2960 | * Save ->subtree_control, ->subtree_ss_mask and ->dom_cgrp to the |
2961 | * respective old_ prefixed fields for @cgrp's subtree including @cgrp | |
2962 | * itself. | |
15a27c36 TH |
2963 | */ |
2964 | static void cgroup_save_control(struct cgroup *cgrp) | |
2965 | { | |
2966 | struct cgroup *dsct; | |
2967 | struct cgroup_subsys_state *d_css; | |
2968 | ||
2969 | cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) { | |
2970 | dsct->old_subtree_control = dsct->subtree_control; | |
2971 | dsct->old_subtree_ss_mask = dsct->subtree_ss_mask; | |
479adb89 | 2972 | dsct->old_dom_cgrp = dsct->dom_cgrp; |
15a27c36 TH |
2973 | } |
2974 | } | |
2975 | ||
2976 | /** | |
2977 | * cgroup_propagate_control - refresh control masks of a subtree | |
2978 | * @cgrp: root of the target subtree | |
2979 | * | |
2980 | * For @cgrp and its subtree, ensure ->subtree_ss_mask matches | |
2981 | * ->subtree_control and propagate controller availability through the | |
2982 | * subtree so that descendants don't have unavailable controllers enabled. | |
2983 | */ | |
2984 | static void cgroup_propagate_control(struct cgroup *cgrp) | |
2985 | { | |
2986 | struct cgroup *dsct; | |
2987 | struct cgroup_subsys_state *d_css; | |
2988 | ||
2989 | cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) { | |
2990 | dsct->subtree_control &= cgroup_control(dsct); | |
5ced2518 TH |
2991 | dsct->subtree_ss_mask = |
2992 | cgroup_calc_subtree_ss_mask(dsct->subtree_control, | |
2993 | cgroup_ss_mask(dsct)); | |
15a27c36 TH |
2994 | } |
2995 | } | |
2996 | ||
2997 | /** | |
479adb89 | 2998 | * cgroup_restore_control - restore control masks and dom_cgrp of a subtree |
15a27c36 TH |
2999 | * @cgrp: root of the target subtree |
3000 | * | |
479adb89 TH |
3001 | * Restore ->subtree_control, ->subtree_ss_mask and ->dom_cgrp from the |
3002 | * respective old_ prefixed fields for @cgrp's subtree including @cgrp | |
3003 | * itself. | |
15a27c36 TH |
3004 | */ |
3005 | static void cgroup_restore_control(struct cgroup *cgrp) | |
3006 | { | |
3007 | struct cgroup *dsct; | |
3008 | struct cgroup_subsys_state *d_css; | |
3009 | ||
3010 | cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) { | |
3011 | dsct->subtree_control = dsct->old_subtree_control; | |
3012 | dsct->subtree_ss_mask = dsct->old_subtree_ss_mask; | |
479adb89 | 3013 | dsct->dom_cgrp = dsct->old_dom_cgrp; |
15a27c36 TH |
3014 | } |
3015 | } | |
3016 | ||
f6d635ad TH |
3017 | static bool css_visible(struct cgroup_subsys_state *css) |
3018 | { | |
3019 | struct cgroup_subsys *ss = css->ss; | |
3020 | struct cgroup *cgrp = css->cgroup; | |
3021 | ||
3022 | if (cgroup_control(cgrp) & (1 << ss->id)) | |
3023 | return true; | |
3024 | if (!(cgroup_ss_mask(cgrp) & (1 << ss->id))) | |
3025 | return false; | |
3026 | return cgroup_on_dfl(cgrp) && ss->implicit_on_dfl; | |
3027 | } | |
3028 | ||
bdb53bd7 TH |
3029 | /** |
3030 | * cgroup_apply_control_enable - enable or show csses according to control | |
ce3f1d9d | 3031 | * @cgrp: root of the target subtree |
bdb53bd7 | 3032 | * |
ce3f1d9d | 3033 | * Walk @cgrp's subtree and create new csses or make the existing ones |
bdb53bd7 TH |
3034 | * visible. A css is created invisible if it's being implicitly enabled |
3035 | * through dependency. An invisible css is made visible when the userland | |
3036 | * explicitly enables it. | |
3037 | * | |
3038 | * Returns 0 on success, -errno on failure. On failure, csses which have | |
3039 | * been processed already aren't cleaned up. The caller is responsible for | |
8a1115ff | 3040 | * cleaning up with cgroup_apply_control_disable(). |
bdb53bd7 TH |
3041 | */ |
3042 | static int cgroup_apply_control_enable(struct cgroup *cgrp) | |
3043 | { | |
3044 | struct cgroup *dsct; | |
ce3f1d9d | 3045 | struct cgroup_subsys_state *d_css; |
bdb53bd7 TH |
3046 | struct cgroup_subsys *ss; |
3047 | int ssid, ret; | |
3048 | ||
ce3f1d9d | 3049 | cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) { |
bdb53bd7 TH |
3050 | for_each_subsys(ss, ssid) { |
3051 | struct cgroup_subsys_state *css = cgroup_css(dsct, ss); | |
3052 | ||
945ba199 TH |
3053 | WARN_ON_ONCE(css && percpu_ref_is_dying(&css->refcnt)); |
3054 | ||
bdb53bd7 TH |
3055 | if (!(cgroup_ss_mask(dsct) & (1 << ss->id))) |
3056 | continue; | |
3057 | ||
3058 | if (!css) { | |
3059 | css = css_create(dsct, ss); | |
3060 | if (IS_ERR(css)) | |
3061 | return PTR_ERR(css); | |
3062 | } | |
3063 | ||
f6d635ad | 3064 | if (css_visible(css)) { |
334c3679 | 3065 | ret = css_populate_dir(css); |
bdb53bd7 TH |
3066 | if (ret) |
3067 | return ret; | |
3068 | } | |
3069 | } | |
3070 | } | |
3071 | ||
3072 | return 0; | |
3073 | } | |
3074 | ||
12b3bb6a TH |
3075 | /** |
3076 | * cgroup_apply_control_disable - kill or hide csses according to control | |
ce3f1d9d | 3077 | * @cgrp: root of the target subtree |
12b3bb6a | 3078 | * |
ce3f1d9d | 3079 | * Walk @cgrp's subtree and kill and hide csses so that they match |
12b3bb6a TH |
3080 | * cgroup_ss_mask() and cgroup_visible_mask(). |
3081 | * | |
3082 | * A css is hidden when the userland requests it to be disabled while other | |
3083 | * subsystems are still depending on it. The css must not actively control | |
3084 | * resources and be in the vanilla state if it's made visible again later. | |
3085 | * Controllers which may be depended upon should provide ->css_reset() for | |
3086 | * this purpose. | |
3087 | */ | |
3088 | static void cgroup_apply_control_disable(struct cgroup *cgrp) | |
3089 | { | |
3090 | struct cgroup *dsct; | |
ce3f1d9d | 3091 | struct cgroup_subsys_state *d_css; |
12b3bb6a TH |
3092 | struct cgroup_subsys *ss; |
3093 | int ssid; | |
3094 | ||
ce3f1d9d | 3095 | cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) { |
12b3bb6a TH |
3096 | for_each_subsys(ss, ssid) { |
3097 | struct cgroup_subsys_state *css = cgroup_css(dsct, ss); | |
3098 | ||
945ba199 TH |
3099 | WARN_ON_ONCE(css && percpu_ref_is_dying(&css->refcnt)); |
3100 | ||
12b3bb6a TH |
3101 | if (!css) |
3102 | continue; | |
3103 | ||
334c3679 TH |
3104 | if (css->parent && |
3105 | !(cgroup_ss_mask(dsct) & (1 << ss->id))) { | |
12b3bb6a | 3106 | kill_css(css); |
f6d635ad | 3107 | } else if (!css_visible(css)) { |
334c3679 | 3108 | css_clear_dir(css); |
12b3bb6a TH |
3109 | if (ss->css_reset) |
3110 | ss->css_reset(css); | |
3111 | } | |
3112 | } | |
3113 | } | |
3114 | } | |
3115 | ||
f7b2814b TH |
3116 | /** |
3117 | * cgroup_apply_control - apply control mask updates to the subtree | |
3118 | * @cgrp: root of the target subtree | |
3119 | * | |
3120 | * subsystems can be enabled and disabled in a subtree using the following | |
3121 | * steps. | |
3122 | * | |
3123 | * 1. Call cgroup_save_control() to stash the current state. | |
3124 | * 2. Update ->subtree_control masks in the subtree as desired. | |
3125 | * 3. Call cgroup_apply_control() to apply the changes. | |
3126 | * 4. Optionally perform other related operations. | |
3127 | * 5. Call cgroup_finalize_control() to finish up. | |
3128 | * | |
3129 | * This function implements step 3 and propagates the mask changes | |
3130 | * throughout @cgrp's subtree, updates csses accordingly and perform | |
3131 | * process migrations. | |
3132 | */ | |
3133 | static int cgroup_apply_control(struct cgroup *cgrp) | |
3134 | { | |
3135 | int ret; | |
3136 | ||
3137 | cgroup_propagate_control(cgrp); | |
3138 | ||
3139 | ret = cgroup_apply_control_enable(cgrp); | |
3140 | if (ret) | |
3141 | return ret; | |
3142 | ||
3143 | /* | |
fc5a828b | 3144 | * At this point, cgroup_e_css_by_mask() results reflect the new csses |
f7b2814b TH |
3145 | * making the following cgroup_update_dfl_csses() properly update |
3146 | * css associations of all tasks in the subtree. | |
3147 | */ | |
3148 | ret = cgroup_update_dfl_csses(cgrp); | |
3149 | if (ret) | |
3150 | return ret; | |
3151 | ||
3152 | return 0; | |
3153 | } | |
3154 | ||
3155 | /** | |
3156 | * cgroup_finalize_control - finalize control mask update | |
3157 | * @cgrp: root of the target subtree | |
3158 | * @ret: the result of the update | |
3159 | * | |
3160 | * Finalize control mask update. See cgroup_apply_control() for more info. | |
3161 | */ | |
3162 | static void cgroup_finalize_control(struct cgroup *cgrp, int ret) | |
3163 | { | |
3164 | if (ret) { | |
3165 | cgroup_restore_control(cgrp); | |
3166 | cgroup_propagate_control(cgrp); | |
3167 | } | |
3168 | ||
3169 | cgroup_apply_control_disable(cgrp); | |
3170 | } | |
3171 | ||
8cfd8147 TH |
3172 | static int cgroup_vet_subtree_control_enable(struct cgroup *cgrp, u16 enable) |
3173 | { | |
3174 | u16 domain_enable = enable & ~cgrp_dfl_threaded_ss_mask; | |
3175 | ||
3176 | /* if nothing is getting enabled, nothing to worry about */ | |
3177 | if (!enable) | |
3178 | return 0; | |
3179 | ||
3180 | /* can @cgrp host any resources? */ | |
3181 | if (!cgroup_is_valid_domain(cgrp->dom_cgrp)) | |
3182 | return -EOPNOTSUPP; | |
3183 | ||
3184 | /* mixables don't care */ | |
3185 | if (cgroup_is_mixable(cgrp)) | |
3186 | return 0; | |
3187 | ||
3188 | if (domain_enable) { | |
3189 | /* can't enable domain controllers inside a thread subtree */ | |
3190 | if (cgroup_is_thread_root(cgrp) || cgroup_is_threaded(cgrp)) | |
3191 | return -EOPNOTSUPP; | |
3192 | } else { | |
3193 | /* | |
3194 | * Threaded controllers can handle internal competitions | |
3195 | * and are always allowed inside a (prospective) thread | |
3196 | * subtree. | |
3197 | */ | |
3198 | if (cgroup_can_be_thread_root(cgrp) || cgroup_is_threaded(cgrp)) | |
3199 | return 0; | |
3200 | } | |
3201 | ||
3202 | /* | |
3203 | * Controllers can't be enabled for a cgroup with tasks to avoid | |
3204 | * child cgroups competing against tasks. | |
3205 | */ | |
3206 | if (cgroup_has_tasks(cgrp)) | |
3207 | return -EBUSY; | |
3208 | ||
3209 | return 0; | |
3210 | } | |
3211 | ||
f8f22e53 | 3212 | /* change the enabled child controllers for a cgroup in the default hierarchy */ |
451af504 TH |
3213 | static ssize_t cgroup_subtree_control_write(struct kernfs_open_file *of, |
3214 | char *buf, size_t nbytes, | |
3215 | loff_t off) | |
f8f22e53 | 3216 | { |
6e5c8307 | 3217 | u16 enable = 0, disable = 0; |
a9746d8d | 3218 | struct cgroup *cgrp, *child; |
f8f22e53 | 3219 | struct cgroup_subsys *ss; |
451af504 | 3220 | char *tok; |
f8f22e53 TH |
3221 | int ssid, ret; |
3222 | ||
3223 | /* | |
d37167ab TH |
3224 | * Parse input - space separated list of subsystem names prefixed |
3225 | * with either + or -. | |
f8f22e53 | 3226 | */ |
451af504 TH |
3227 | buf = strstrip(buf); |
3228 | while ((tok = strsep(&buf, " "))) { | |
d37167ab TH |
3229 | if (tok[0] == '\0') |
3230 | continue; | |
a7165264 | 3231 | do_each_subsys_mask(ss, ssid, ~cgrp_dfl_inhibit_ss_mask) { |
fc5ed1e9 TH |
3232 | if (!cgroup_ssid_enabled(ssid) || |
3233 | strcmp(tok + 1, ss->name)) | |
f8f22e53 TH |
3234 | continue; |
3235 | ||
3236 | if (*tok == '+') { | |
7d331fa9 TH |
3237 | enable |= 1 << ssid; |
3238 | disable &= ~(1 << ssid); | |
f8f22e53 | 3239 | } else if (*tok == '-') { |
7d331fa9 TH |
3240 | disable |= 1 << ssid; |
3241 | enable &= ~(1 << ssid); | |
f8f22e53 TH |
3242 | } else { |
3243 | return -EINVAL; | |
3244 | } | |
3245 | break; | |
b4e0eeaf | 3246 | } while_each_subsys_mask(); |
f8f22e53 TH |
3247 | if (ssid == CGROUP_SUBSYS_COUNT) |
3248 | return -EINVAL; | |
3249 | } | |
3250 | ||
945ba199 | 3251 | cgrp = cgroup_kn_lock_live(of->kn, true); |
a9746d8d TH |
3252 | if (!cgrp) |
3253 | return -ENODEV; | |
f8f22e53 TH |
3254 | |
3255 | for_each_subsys(ss, ssid) { | |
3256 | if (enable & (1 << ssid)) { | |
667c2491 | 3257 | if (cgrp->subtree_control & (1 << ssid)) { |
f8f22e53 TH |
3258 | enable &= ~(1 << ssid); |
3259 | continue; | |
3260 | } | |
3261 | ||
5531dc91 | 3262 | if (!(cgroup_control(cgrp) & (1 << ssid))) { |
c29adf24 TH |
3263 | ret = -ENOENT; |
3264 | goto out_unlock; | |
3265 | } | |
f8f22e53 | 3266 | } else if (disable & (1 << ssid)) { |
667c2491 | 3267 | if (!(cgrp->subtree_control & (1 << ssid))) { |
f8f22e53 TH |
3268 | disable &= ~(1 << ssid); |
3269 | continue; | |
3270 | } | |
3271 | ||
3272 | /* a child has it enabled? */ | |
3273 | cgroup_for_each_live_child(child, cgrp) { | |
667c2491 | 3274 | if (child->subtree_control & (1 << ssid)) { |
f8f22e53 | 3275 | ret = -EBUSY; |
ddab2b6e | 3276 | goto out_unlock; |
f8f22e53 TH |
3277 | } |
3278 | } | |
3279 | } | |
3280 | } | |
3281 | ||
3282 | if (!enable && !disable) { | |
3283 | ret = 0; | |
ddab2b6e | 3284 | goto out_unlock; |
f8f22e53 TH |
3285 | } |
3286 | ||
8cfd8147 TH |
3287 | ret = cgroup_vet_subtree_control_enable(cgrp, enable); |
3288 | if (ret) | |
27f26753 | 3289 | goto out_unlock; |
f8f22e53 | 3290 | |
15a27c36 TH |
3291 | /* save and update control masks and prepare csses */ |
3292 | cgroup_save_control(cgrp); | |
f63070d3 | 3293 | |
15a27c36 TH |
3294 | cgrp->subtree_control |= enable; |
3295 | cgrp->subtree_control &= ~disable; | |
c29adf24 | 3296 | |
f7b2814b | 3297 | ret = cgroup_apply_control(cgrp); |
f7b2814b | 3298 | cgroup_finalize_control(cgrp, ret); |
3c745417 TH |
3299 | if (ret) |
3300 | goto out_unlock; | |
f8f22e53 TH |
3301 | |
3302 | kernfs_activate(cgrp->kn); | |
f8f22e53 | 3303 | out_unlock: |
a9746d8d | 3304 | cgroup_kn_unlock(of->kn); |
451af504 | 3305 | return ret ?: nbytes; |
f8f22e53 TH |
3306 | } |
3307 | ||
c705a00d TH |
3308 | /** |
3309 | * cgroup_enable_threaded - make @cgrp threaded | |
3310 | * @cgrp: the target cgroup | |
3311 | * | |
3312 | * Called when "threaded" is written to the cgroup.type interface file and | |
3313 | * tries to make @cgrp threaded and join the parent's resource domain. | |
3314 | * This function is never called on the root cgroup as cgroup.type doesn't | |
3315 | * exist on it. | |
3316 | */ | |
8cfd8147 TH |
3317 | static int cgroup_enable_threaded(struct cgroup *cgrp) |
3318 | { | |
3319 | struct cgroup *parent = cgroup_parent(cgrp); | |
3320 | struct cgroup *dom_cgrp = parent->dom_cgrp; | |
479adb89 TH |
3321 | struct cgroup *dsct; |
3322 | struct cgroup_subsys_state *d_css; | |
8cfd8147 TH |
3323 | int ret; |
3324 | ||
3325 | lockdep_assert_held(&cgroup_mutex); | |
3326 | ||
3327 | /* noop if already threaded */ | |
3328 | if (cgroup_is_threaded(cgrp)) | |
3329 | return 0; | |
3330 | ||
d1897c95 TH |
3331 | /* |
3332 | * If @cgroup is populated or has domain controllers enabled, it | |
3333 | * can't be switched. While the below cgroup_can_be_thread_root() | |
3334 | * test can catch the same conditions, that's only when @parent is | |
3335 | * not mixable, so let's check it explicitly. | |
3336 | */ | |
3337 | if (cgroup_is_populated(cgrp) || | |
3338 | cgrp->subtree_control & ~cgrp_dfl_threaded_ss_mask) | |
3339 | return -EOPNOTSUPP; | |
3340 | ||
8cfd8147 TH |
3341 | /* we're joining the parent's domain, ensure its validity */ |
3342 | if (!cgroup_is_valid_domain(dom_cgrp) || | |
3343 | !cgroup_can_be_thread_root(dom_cgrp)) | |
3344 | return -EOPNOTSUPP; | |
3345 | ||
8cfd8147 TH |
3346 | /* |
3347 | * The following shouldn't cause actual migrations and should | |
3348 | * always succeed. | |
3349 | */ | |
3350 | cgroup_save_control(cgrp); | |
3351 | ||
479adb89 TH |
3352 | cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) |
3353 | if (dsct == cgrp || cgroup_is_threaded(dsct)) | |
3354 | dsct->dom_cgrp = dom_cgrp; | |
3355 | ||
8cfd8147 TH |
3356 | ret = cgroup_apply_control(cgrp); |
3357 | if (!ret) | |
3358 | parent->nr_threaded_children++; | |
8cfd8147 TH |
3359 | |
3360 | cgroup_finalize_control(cgrp, ret); | |
3361 | return ret; | |
3362 | } | |
3363 | ||
3364 | static int cgroup_type_show(struct seq_file *seq, void *v) | |
3365 | { | |
3366 | struct cgroup *cgrp = seq_css(seq)->cgroup; | |
3367 | ||
3368 | if (cgroup_is_threaded(cgrp)) | |
3369 | seq_puts(seq, "threaded\n"); | |
3370 | else if (!cgroup_is_valid_domain(cgrp)) | |
3371 | seq_puts(seq, "domain invalid\n"); | |
3372 | else if (cgroup_is_thread_root(cgrp)) | |
3373 | seq_puts(seq, "domain threaded\n"); | |
3374 | else | |
3375 | seq_puts(seq, "domain\n"); | |
3376 | ||
3377 | return 0; | |
3378 | } | |
3379 | ||
3380 | static ssize_t cgroup_type_write(struct kernfs_open_file *of, char *buf, | |
3381 | size_t nbytes, loff_t off) | |
3382 | { | |
3383 | struct cgroup *cgrp; | |
3384 | int ret; | |
3385 | ||
3386 | /* only switching to threaded mode is supported */ | |
3387 | if (strcmp(strstrip(buf), "threaded")) | |
3388 | return -EINVAL; | |
3389 | ||
3390 | cgrp = cgroup_kn_lock_live(of->kn, false); | |
3391 | if (!cgrp) | |
3392 | return -ENOENT; | |
3393 | ||
3394 | /* threaded can only be enabled */ | |
3395 | ret = cgroup_enable_threaded(cgrp); | |
3396 | ||
3397 | cgroup_kn_unlock(of->kn); | |
3398 | return ret ?: nbytes; | |
3399 | } | |
3400 | ||
1a926e0b RG |
3401 | static int cgroup_max_descendants_show(struct seq_file *seq, void *v) |
3402 | { | |
3403 | struct cgroup *cgrp = seq_css(seq)->cgroup; | |
3404 | int descendants = READ_ONCE(cgrp->max_descendants); | |
3405 | ||
3406 | if (descendants == INT_MAX) | |
3407 | seq_puts(seq, "max\n"); | |
3408 | else | |
3409 | seq_printf(seq, "%d\n", descendants); | |
3410 | ||
3411 | return 0; | |
3412 | } | |
3413 | ||
3414 | static ssize_t cgroup_max_descendants_write(struct kernfs_open_file *of, | |
3415 | char *buf, size_t nbytes, loff_t off) | |
3416 | { | |
3417 | struct cgroup *cgrp; | |
3418 | int descendants; | |
3419 | ssize_t ret; | |
3420 | ||
3421 | buf = strstrip(buf); | |
3422 | if (!strcmp(buf, "max")) { | |
3423 | descendants = INT_MAX; | |
3424 | } else { | |
3425 | ret = kstrtoint(buf, 0, &descendants); | |
3426 | if (ret) | |
3427 | return ret; | |
3428 | } | |
3429 | ||
696b98f2 | 3430 | if (descendants < 0) |
1a926e0b RG |
3431 | return -ERANGE; |
3432 | ||
3433 | cgrp = cgroup_kn_lock_live(of->kn, false); | |
3434 | if (!cgrp) | |
3435 | return -ENOENT; | |
3436 | ||
3437 | cgrp->max_descendants = descendants; | |
3438 | ||
3439 | cgroup_kn_unlock(of->kn); | |
3440 | ||
3441 | return nbytes; | |
3442 | } | |
3443 | ||
3444 | static int cgroup_max_depth_show(struct seq_file *seq, void *v) | |
3445 | { | |
3446 | struct cgroup *cgrp = seq_css(seq)->cgroup; | |
3447 | int depth = READ_ONCE(cgrp->max_depth); | |
3448 | ||
3449 | if (depth == INT_MAX) | |
3450 | seq_puts(seq, "max\n"); | |
3451 | else | |
3452 | seq_printf(seq, "%d\n", depth); | |
3453 | ||
3454 | return 0; | |
3455 | } | |
3456 | ||
3457 | static ssize_t cgroup_max_depth_write(struct kernfs_open_file *of, | |
3458 | char *buf, size_t nbytes, loff_t off) | |
3459 | { | |
3460 | struct cgroup *cgrp; | |
3461 | ssize_t ret; | |
3462 | int depth; | |
3463 | ||
3464 | buf = strstrip(buf); | |
3465 | if (!strcmp(buf, "max")) { | |
3466 | depth = INT_MAX; | |
3467 | } else { | |
3468 | ret = kstrtoint(buf, 0, &depth); | |
3469 | if (ret) | |
3470 | return ret; | |
3471 | } | |
3472 | ||
696b98f2 | 3473 | if (depth < 0) |
1a926e0b RG |
3474 | return -ERANGE; |
3475 | ||
3476 | cgrp = cgroup_kn_lock_live(of->kn, false); | |
3477 | if (!cgrp) | |
3478 | return -ENOENT; | |
3479 | ||
3480 | cgrp->max_depth = depth; | |
3481 | ||
3482 | cgroup_kn_unlock(of->kn); | |
3483 | ||
3484 | return nbytes; | |
3485 | } | |
3486 | ||
4a07c222 | 3487 | static int cgroup_events_show(struct seq_file *seq, void *v) |
842b597e | 3488 | { |
76f969e8 RG |
3489 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
3490 | ||
3491 | seq_printf(seq, "populated %d\n", cgroup_is_populated(cgrp)); | |
3492 | seq_printf(seq, "frozen %d\n", test_bit(CGRP_FROZEN, &cgrp->flags)); | |
3493 | ||
842b597e TH |
3494 | return 0; |
3495 | } | |
3496 | ||
3e48930c | 3497 | static int cgroup_stat_show(struct seq_file *seq, void *v) |
ec39225c RG |
3498 | { |
3499 | struct cgroup *cgroup = seq_css(seq)->cgroup; | |
3500 | ||
3501 | seq_printf(seq, "nr_descendants %d\n", | |
3502 | cgroup->nr_descendants); | |
3503 | seq_printf(seq, "nr_dying_descendants %d\n", | |
3504 | cgroup->nr_dying_descendants); | |
3505 | ||
3506 | return 0; | |
3507 | } | |
3508 | ||
d41bf8c9 TH |
3509 | static int __maybe_unused cgroup_extra_stat_show(struct seq_file *seq, |
3510 | struct cgroup *cgrp, int ssid) | |
3511 | { | |
3512 | struct cgroup_subsys *ss = cgroup_subsys[ssid]; | |
3513 | struct cgroup_subsys_state *css; | |
3514 | int ret; | |
3515 | ||
3516 | if (!ss->css_extra_stat_show) | |
3517 | return 0; | |
3518 | ||
3519 | css = cgroup_tryget_css(cgrp, ss); | |
3520 | if (!css) | |
3521 | return 0; | |
3522 | ||
3523 | ret = ss->css_extra_stat_show(seq, css); | |
3524 | css_put(css); | |
3525 | return ret; | |
3526 | } | |
3527 | ||
3528 | static int cpu_stat_show(struct seq_file *seq, void *v) | |
3529 | { | |
c3ba1329 | 3530 | struct cgroup __maybe_unused *cgrp = seq_css(seq)->cgroup; |
d41bf8c9 TH |
3531 | int ret = 0; |
3532 | ||
d4ff749b | 3533 | cgroup_base_stat_cputime_show(seq); |
d41bf8c9 TH |
3534 | #ifdef CONFIG_CGROUP_SCHED |
3535 | ret = cgroup_extra_stat_show(seq, cgrp, cpu_cgrp_id); | |
3536 | #endif | |
3537 | return ret; | |
3538 | } | |
3539 | ||
2ce7135a JW |
3540 | #ifdef CONFIG_PSI |
3541 | static int cgroup_io_pressure_show(struct seq_file *seq, void *v) | |
3542 | { | |
74321038 TH |
3543 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
3544 | struct psi_group *psi = cgroup_id(cgrp) == 1 ? &psi_system : &cgrp->psi; | |
df5ba5be DS |
3545 | |
3546 | return psi_show(seq, psi, PSI_IO); | |
2ce7135a JW |
3547 | } |
3548 | static int cgroup_memory_pressure_show(struct seq_file *seq, void *v) | |
3549 | { | |
74321038 TH |
3550 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
3551 | struct psi_group *psi = cgroup_id(cgrp) == 1 ? &psi_system : &cgrp->psi; | |
df5ba5be DS |
3552 | |
3553 | return psi_show(seq, psi, PSI_MEM); | |
2ce7135a JW |
3554 | } |
3555 | static int cgroup_cpu_pressure_show(struct seq_file *seq, void *v) | |
3556 | { | |
74321038 TH |
3557 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
3558 | struct psi_group *psi = cgroup_id(cgrp) == 1 ? &psi_system : &cgrp->psi; | |
df5ba5be DS |
3559 | |
3560 | return psi_show(seq, psi, PSI_CPU); | |
2ce7135a | 3561 | } |
0e94682b SB |
3562 | |
3563 | static ssize_t cgroup_pressure_write(struct kernfs_open_file *of, char *buf, | |
3564 | size_t nbytes, enum psi_res res) | |
3565 | { | |
3566 | struct psi_trigger *new; | |
3567 | struct cgroup *cgrp; | |
3568 | ||
3569 | cgrp = cgroup_kn_lock_live(of->kn, false); | |
3570 | if (!cgrp) | |
3571 | return -ENODEV; | |
3572 | ||
3573 | cgroup_get(cgrp); | |
3574 | cgroup_kn_unlock(of->kn); | |
3575 | ||
3576 | new = psi_trigger_create(&cgrp->psi, buf, nbytes, res); | |
3577 | if (IS_ERR(new)) { | |
3578 | cgroup_put(cgrp); | |
3579 | return PTR_ERR(new); | |
3580 | } | |
3581 | ||
3582 | psi_trigger_replace(&of->priv, new); | |
3583 | ||
3584 | cgroup_put(cgrp); | |
3585 | ||
3586 | return nbytes; | |
3587 | } | |
3588 | ||
3589 | static ssize_t cgroup_io_pressure_write(struct kernfs_open_file *of, | |
3590 | char *buf, size_t nbytes, | |
3591 | loff_t off) | |
3592 | { | |
3593 | return cgroup_pressure_write(of, buf, nbytes, PSI_IO); | |
3594 | } | |
3595 | ||
3596 | static ssize_t cgroup_memory_pressure_write(struct kernfs_open_file *of, | |
3597 | char *buf, size_t nbytes, | |
3598 | loff_t off) | |
3599 | { | |
3600 | return cgroup_pressure_write(of, buf, nbytes, PSI_MEM); | |
3601 | } | |
3602 | ||
3603 | static ssize_t cgroup_cpu_pressure_write(struct kernfs_open_file *of, | |
3604 | char *buf, size_t nbytes, | |
3605 | loff_t off) | |
3606 | { | |
3607 | return cgroup_pressure_write(of, buf, nbytes, PSI_CPU); | |
3608 | } | |
3609 | ||
3610 | static __poll_t cgroup_pressure_poll(struct kernfs_open_file *of, | |
3611 | poll_table *pt) | |
3612 | { | |
3613 | return psi_trigger_poll(&of->priv, of->file, pt); | |
3614 | } | |
3615 | ||
3616 | static void cgroup_pressure_release(struct kernfs_open_file *of) | |
3617 | { | |
3618 | psi_trigger_replace(&of->priv, NULL); | |
3619 | } | |
3620 | #endif /* CONFIG_PSI */ | |
2ce7135a | 3621 | |
76f969e8 RG |
3622 | static int cgroup_freeze_show(struct seq_file *seq, void *v) |
3623 | { | |
3624 | struct cgroup *cgrp = seq_css(seq)->cgroup; | |
3625 | ||
3626 | seq_printf(seq, "%d\n", cgrp->freezer.freeze); | |
3627 | ||
3628 | return 0; | |
3629 | } | |
3630 | ||
3631 | static ssize_t cgroup_freeze_write(struct kernfs_open_file *of, | |
3632 | char *buf, size_t nbytes, loff_t off) | |
3633 | { | |
3634 | struct cgroup *cgrp; | |
3635 | ssize_t ret; | |
3636 | int freeze; | |
3637 | ||
3638 | ret = kstrtoint(strstrip(buf), 0, &freeze); | |
3639 | if (ret) | |
3640 | return ret; | |
3641 | ||
3642 | if (freeze < 0 || freeze > 1) | |
3643 | return -ERANGE; | |
3644 | ||
3645 | cgrp = cgroup_kn_lock_live(of->kn, false); | |
3646 | if (!cgrp) | |
3647 | return -ENOENT; | |
3648 | ||
3649 | cgroup_freeze(cgrp, freeze); | |
3650 | ||
3651 | cgroup_kn_unlock(of->kn); | |
3652 | ||
3653 | return nbytes; | |
3654 | } | |
3655 | ||
e90cbebc TH |
3656 | static int cgroup_file_open(struct kernfs_open_file *of) |
3657 | { | |
3658 | struct cftype *cft = of->kn->priv; | |
3659 | ||
3660 | if (cft->open) | |
3661 | return cft->open(of); | |
3662 | return 0; | |
3663 | } | |
3664 | ||
3665 | static void cgroup_file_release(struct kernfs_open_file *of) | |
3666 | { | |
3667 | struct cftype *cft = of->kn->priv; | |
3668 | ||
3669 | if (cft->release) | |
3670 | cft->release(of); | |
3671 | } | |
3672 | ||
2bd59d48 TH |
3673 | static ssize_t cgroup_file_write(struct kernfs_open_file *of, char *buf, |
3674 | size_t nbytes, loff_t off) | |
355e0c48 | 3675 | { |
5136f636 | 3676 | struct cgroup_namespace *ns = current->nsproxy->cgroup_ns; |
2bd59d48 TH |
3677 | struct cgroup *cgrp = of->kn->parent->priv; |
3678 | struct cftype *cft = of->kn->priv; | |
3679 | struct cgroup_subsys_state *css; | |
a742c59d | 3680 | int ret; |
355e0c48 | 3681 | |
5136f636 TH |
3682 | /* |
3683 | * If namespaces are delegation boundaries, disallow writes to | |
3684 | * files in an non-init namespace root from inside the namespace | |
3685 | * except for the files explicitly marked delegatable - | |
3686 | * cgroup.procs and cgroup.subtree_control. | |
3687 | */ | |
3688 | if ((cgrp->root->flags & CGRP_ROOT_NS_DELEGATE) && | |
3689 | !(cft->flags & CFTYPE_NS_DELEGATABLE) && | |
3690 | ns != &init_cgroup_ns && ns->root_cset->dfl_cgrp == cgrp) | |
3691 | return -EPERM; | |
3692 | ||
b4168640 TH |
3693 | if (cft->write) |
3694 | return cft->write(of, buf, nbytes, off); | |
3695 | ||
2bd59d48 TH |
3696 | /* |
3697 | * kernfs guarantees that a file isn't deleted with operations in | |
3698 | * flight, which means that the matching css is and stays alive and | |
3699 | * doesn't need to be pinned. The RCU locking is not necessary | |
3700 | * either. It's just for the convenience of using cgroup_css(). | |
3701 | */ | |
3702 | rcu_read_lock(); | |
3703 | css = cgroup_css(cgrp, cft->ss); | |
3704 | rcu_read_unlock(); | |
a742c59d | 3705 | |
451af504 | 3706 | if (cft->write_u64) { |
a742c59d TH |
3707 | unsigned long long v; |
3708 | ret = kstrtoull(buf, 0, &v); | |
3709 | if (!ret) | |
3710 | ret = cft->write_u64(css, cft, v); | |
3711 | } else if (cft->write_s64) { | |
3712 | long long v; | |
3713 | ret = kstrtoll(buf, 0, &v); | |
3714 | if (!ret) | |
3715 | ret = cft->write_s64(css, cft, v); | |
e73d2c61 | 3716 | } else { |
a742c59d | 3717 | ret = -EINVAL; |
e73d2c61 | 3718 | } |
2bd59d48 | 3719 | |
a742c59d | 3720 | return ret ?: nbytes; |
355e0c48 PM |
3721 | } |
3722 | ||
dc50537b JW |
3723 | static __poll_t cgroup_file_poll(struct kernfs_open_file *of, poll_table *pt) |
3724 | { | |
3725 | struct cftype *cft = of->kn->priv; | |
3726 | ||
3727 | if (cft->poll) | |
3728 | return cft->poll(of, pt); | |
3729 | ||
3730 | return kernfs_generic_poll(of, pt); | |
3731 | } | |
3732 | ||
6612f05b | 3733 | static void *cgroup_seqfile_start(struct seq_file *seq, loff_t *ppos) |
db3b1497 | 3734 | { |
2bd59d48 | 3735 | return seq_cft(seq)->seq_start(seq, ppos); |
db3b1497 PM |
3736 | } |
3737 | ||
6612f05b | 3738 | static void *cgroup_seqfile_next(struct seq_file *seq, void *v, loff_t *ppos) |
ddbcc7e8 | 3739 | { |
2bd59d48 | 3740 | return seq_cft(seq)->seq_next(seq, v, ppos); |
ddbcc7e8 PM |
3741 | } |
3742 | ||
6612f05b | 3743 | static void cgroup_seqfile_stop(struct seq_file *seq, void *v) |
ddbcc7e8 | 3744 | { |
e90cbebc TH |
3745 | if (seq_cft(seq)->seq_stop) |
3746 | seq_cft(seq)->seq_stop(seq, v); | |
ddbcc7e8 PM |
3747 | } |
3748 | ||
91796569 | 3749 | static int cgroup_seqfile_show(struct seq_file *m, void *arg) |
e73d2c61 | 3750 | { |
7da11279 TH |
3751 | struct cftype *cft = seq_cft(m); |
3752 | struct cgroup_subsys_state *css = seq_css(m); | |
e73d2c61 | 3753 | |
2da8ca82 TH |
3754 | if (cft->seq_show) |
3755 | return cft->seq_show(m, arg); | |
e73d2c61 | 3756 | |
f4c753b7 | 3757 | if (cft->read_u64) |
896f5199 TH |
3758 | seq_printf(m, "%llu\n", cft->read_u64(css, cft)); |
3759 | else if (cft->read_s64) | |
3760 | seq_printf(m, "%lld\n", cft->read_s64(css, cft)); | |
3761 | else | |
3762 | return -EINVAL; | |
3763 | return 0; | |
91796569 PM |
3764 | } |
3765 | ||
2bd59d48 TH |
3766 | static struct kernfs_ops cgroup_kf_single_ops = { |
3767 | .atomic_write_len = PAGE_SIZE, | |
e90cbebc TH |
3768 | .open = cgroup_file_open, |
3769 | .release = cgroup_file_release, | |
2bd59d48 | 3770 | .write = cgroup_file_write, |
dc50537b | 3771 | .poll = cgroup_file_poll, |
2bd59d48 | 3772 | .seq_show = cgroup_seqfile_show, |
91796569 PM |
3773 | }; |
3774 | ||
2bd59d48 TH |
3775 | static struct kernfs_ops cgroup_kf_ops = { |
3776 | .atomic_write_len = PAGE_SIZE, | |
e90cbebc TH |
3777 | .open = cgroup_file_open, |
3778 | .release = cgroup_file_release, | |
2bd59d48 | 3779 | .write = cgroup_file_write, |
dc50537b | 3780 | .poll = cgroup_file_poll, |
2bd59d48 TH |
3781 | .seq_start = cgroup_seqfile_start, |
3782 | .seq_next = cgroup_seqfile_next, | |
3783 | .seq_stop = cgroup_seqfile_stop, | |
3784 | .seq_show = cgroup_seqfile_show, | |
3785 | }; | |
ddbcc7e8 | 3786 | |
49957f8e TH |
3787 | /* set uid and gid of cgroup dirs and files to that of the creator */ |
3788 | static int cgroup_kn_set_ugid(struct kernfs_node *kn) | |
3789 | { | |
3790 | struct iattr iattr = { .ia_valid = ATTR_UID | ATTR_GID, | |
3791 | .ia_uid = current_fsuid(), | |
3792 | .ia_gid = current_fsgid(), }; | |
3793 | ||
3794 | if (uid_eq(iattr.ia_uid, GLOBAL_ROOT_UID) && | |
3795 | gid_eq(iattr.ia_gid, GLOBAL_ROOT_GID)) | |
3796 | return 0; | |
3797 | ||
3798 | return kernfs_setattr(kn, &iattr); | |
3799 | } | |
3800 | ||
b12e3583 TH |
3801 | static void cgroup_file_notify_timer(struct timer_list *timer) |
3802 | { | |
3803 | cgroup_file_notify(container_of(timer, struct cgroup_file, | |
3804 | notify_timer)); | |
3805 | } | |
3806 | ||
4df8dc90 TH |
3807 | static int cgroup_add_file(struct cgroup_subsys_state *css, struct cgroup *cgrp, |
3808 | struct cftype *cft) | |
ddbcc7e8 | 3809 | { |
8d7e6fb0 | 3810 | char name[CGROUP_FILE_NAME_MAX]; |
2bd59d48 TH |
3811 | struct kernfs_node *kn; |
3812 | struct lock_class_key *key = NULL; | |
49957f8e | 3813 | int ret; |
05ef1d7c | 3814 | |
2bd59d48 TH |
3815 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
3816 | key = &cft->lockdep_key; | |
3817 | #endif | |
3818 | kn = __kernfs_create_file(cgrp->kn, cgroup_file_name(cgrp, cft, name), | |
488dee96 DT |
3819 | cgroup_file_mode(cft), |
3820 | GLOBAL_ROOT_UID, GLOBAL_ROOT_GID, | |
3821 | 0, cft->kf_ops, cft, | |
dfeb0750 | 3822 | NULL, key); |
49957f8e TH |
3823 | if (IS_ERR(kn)) |
3824 | return PTR_ERR(kn); | |
3825 | ||
3826 | ret = cgroup_kn_set_ugid(kn); | |
f8f22e53 | 3827 | if (ret) { |
49957f8e | 3828 | kernfs_remove(kn); |
f8f22e53 TH |
3829 | return ret; |
3830 | } | |
3831 | ||
6f60eade TH |
3832 | if (cft->file_offset) { |
3833 | struct cgroup_file *cfile = (void *)css + cft->file_offset; | |
3834 | ||
b12e3583 TH |
3835 | timer_setup(&cfile->notify_timer, cgroup_file_notify_timer, 0); |
3836 | ||
34c06254 | 3837 | spin_lock_irq(&cgroup_file_kn_lock); |
6f60eade | 3838 | cfile->kn = kn; |
34c06254 | 3839 | spin_unlock_irq(&cgroup_file_kn_lock); |
6f60eade TH |
3840 | } |
3841 | ||
f8f22e53 | 3842 | return 0; |
ddbcc7e8 PM |
3843 | } |
3844 | ||
b1f28d31 TH |
3845 | /** |
3846 | * cgroup_addrm_files - add or remove files to a cgroup directory | |
4df8dc90 TH |
3847 | * @css: the target css |
3848 | * @cgrp: the target cgroup (usually css->cgroup) | |
b1f28d31 TH |
3849 | * @cfts: array of cftypes to be added |
3850 | * @is_add: whether to add or remove | |
3851 | * | |
3852 | * Depending on @is_add, add or remove files defined by @cfts on @cgrp. | |
6732ed85 | 3853 | * For removals, this function never fails. |
b1f28d31 | 3854 | */ |
4df8dc90 TH |
3855 | static int cgroup_addrm_files(struct cgroup_subsys_state *css, |
3856 | struct cgroup *cgrp, struct cftype cfts[], | |
2bb566cb | 3857 | bool is_add) |
ddbcc7e8 | 3858 | { |
6732ed85 | 3859 | struct cftype *cft, *cft_end = NULL; |
b598dde3 | 3860 | int ret = 0; |
b1f28d31 | 3861 | |
01f6474c | 3862 | lockdep_assert_held(&cgroup_mutex); |
db0416b6 | 3863 | |
6732ed85 TH |
3864 | restart: |
3865 | for (cft = cfts; cft != cft_end && cft->name[0] != '\0'; cft++) { | |
f33fddc2 | 3866 | /* does cft->flags tell us to skip this file on @cgrp? */ |
05ebb6e6 | 3867 | if ((cft->flags & __CFTYPE_ONLY_ON_DFL) && !cgroup_on_dfl(cgrp)) |
8cbbf2c9 | 3868 | continue; |
05ebb6e6 | 3869 | if ((cft->flags & __CFTYPE_NOT_ON_DFL) && cgroup_on_dfl(cgrp)) |
873fe09e | 3870 | continue; |
d51f39b0 | 3871 | if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgroup_parent(cgrp)) |
f33fddc2 | 3872 | continue; |
d51f39b0 | 3873 | if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgroup_parent(cgrp)) |
f33fddc2 | 3874 | continue; |
5cf8114d WL |
3875 | if ((cft->flags & CFTYPE_DEBUG) && !cgroup_debug) |
3876 | continue; | |
2739d3cc | 3877 | if (is_add) { |
4df8dc90 | 3878 | ret = cgroup_add_file(css, cgrp, cft); |
b1f28d31 | 3879 | if (ret) { |
ed3d261b JP |
3880 | pr_warn("%s: failed to add %s, err=%d\n", |
3881 | __func__, cft->name, ret); | |
6732ed85 TH |
3882 | cft_end = cft; |
3883 | is_add = false; | |
3884 | goto restart; | |
b1f28d31 | 3885 | } |
2739d3cc LZ |
3886 | } else { |
3887 | cgroup_rm_file(cgrp, cft); | |
db0416b6 | 3888 | } |
ddbcc7e8 | 3889 | } |
b598dde3 | 3890 | return ret; |
ddbcc7e8 PM |
3891 | } |
3892 | ||
21a2d343 | 3893 | static int cgroup_apply_cftypes(struct cftype *cfts, bool is_add) |
8e3f6541 | 3894 | { |
2bb566cb | 3895 | struct cgroup_subsys *ss = cfts[0].ss; |
3dd06ffa | 3896 | struct cgroup *root = &ss->root->cgrp; |
492eb21b | 3897 | struct cgroup_subsys_state *css; |
9ccece80 | 3898 | int ret = 0; |
8e3f6541 | 3899 | |
01f6474c | 3900 | lockdep_assert_held(&cgroup_mutex); |
e8c82d20 | 3901 | |
e8c82d20 | 3902 | /* add/rm files for all cgroups created before */ |
ca8bdcaf | 3903 | css_for_each_descendant_pre(css, cgroup_css(root, ss)) { |
492eb21b TH |
3904 | struct cgroup *cgrp = css->cgroup; |
3905 | ||
88cb04b9 | 3906 | if (!(css->flags & CSS_VISIBLE)) |
e8c82d20 LZ |
3907 | continue; |
3908 | ||
4df8dc90 | 3909 | ret = cgroup_addrm_files(css, cgrp, cfts, is_add); |
9ccece80 TH |
3910 | if (ret) |
3911 | break; | |
8e3f6541 | 3912 | } |
21a2d343 TH |
3913 | |
3914 | if (is_add && !ret) | |
3915 | kernfs_activate(root->kn); | |
9ccece80 | 3916 | return ret; |
8e3f6541 TH |
3917 | } |
3918 | ||
2da440a2 | 3919 | static void cgroup_exit_cftypes(struct cftype *cfts) |
8e3f6541 | 3920 | { |
2bb566cb | 3921 | struct cftype *cft; |
8e3f6541 | 3922 | |
2bd59d48 TH |
3923 | for (cft = cfts; cft->name[0] != '\0'; cft++) { |
3924 | /* free copy for custom atomic_write_len, see init_cftypes() */ | |
3925 | if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) | |
3926 | kfree(cft->kf_ops); | |
3927 | cft->kf_ops = NULL; | |
2da440a2 | 3928 | cft->ss = NULL; |
a8ddc821 TH |
3929 | |
3930 | /* revert flags set by cgroup core while adding @cfts */ | |
05ebb6e6 | 3931 | cft->flags &= ~(__CFTYPE_ONLY_ON_DFL | __CFTYPE_NOT_ON_DFL); |
2bd59d48 | 3932 | } |
2da440a2 TH |
3933 | } |
3934 | ||
2bd59d48 | 3935 | static int cgroup_init_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
2da440a2 TH |
3936 | { |
3937 | struct cftype *cft; | |
3938 | ||
2bd59d48 TH |
3939 | for (cft = cfts; cft->name[0] != '\0'; cft++) { |
3940 | struct kernfs_ops *kf_ops; | |
3941 | ||
0adb0704 TH |
3942 | WARN_ON(cft->ss || cft->kf_ops); |
3943 | ||
2bd59d48 TH |
3944 | if (cft->seq_start) |
3945 | kf_ops = &cgroup_kf_ops; | |
3946 | else | |
3947 | kf_ops = &cgroup_kf_single_ops; | |
3948 | ||
3949 | /* | |
3950 | * Ugh... if @cft wants a custom max_write_len, we need to | |
3951 | * make a copy of kf_ops to set its atomic_write_len. | |
3952 | */ | |
3953 | if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) { | |
3954 | kf_ops = kmemdup(kf_ops, sizeof(*kf_ops), GFP_KERNEL); | |
3955 | if (!kf_ops) { | |
3956 | cgroup_exit_cftypes(cfts); | |
3957 | return -ENOMEM; | |
3958 | } | |
3959 | kf_ops->atomic_write_len = cft->max_write_len; | |
3960 | } | |
8e3f6541 | 3961 | |
2bd59d48 | 3962 | cft->kf_ops = kf_ops; |
2bb566cb | 3963 | cft->ss = ss; |
2bd59d48 | 3964 | } |
2bb566cb | 3965 | |
2bd59d48 | 3966 | return 0; |
2da440a2 TH |
3967 | } |
3968 | ||
21a2d343 TH |
3969 | static int cgroup_rm_cftypes_locked(struct cftype *cfts) |
3970 | { | |
01f6474c | 3971 | lockdep_assert_held(&cgroup_mutex); |
21a2d343 TH |
3972 | |
3973 | if (!cfts || !cfts[0].ss) | |
3974 | return -ENOENT; | |
3975 | ||
3976 | list_del(&cfts->node); | |
3977 | cgroup_apply_cftypes(cfts, false); | |
3978 | cgroup_exit_cftypes(cfts); | |
3979 | return 0; | |
8e3f6541 | 3980 | } |
8e3f6541 | 3981 | |
79578621 TH |
3982 | /** |
3983 | * cgroup_rm_cftypes - remove an array of cftypes from a subsystem | |
79578621 TH |
3984 | * @cfts: zero-length name terminated array of cftypes |
3985 | * | |
2bb566cb TH |
3986 | * Unregister @cfts. Files described by @cfts are removed from all |
3987 | * existing cgroups and all future cgroups won't have them either. This | |
3988 | * function can be called anytime whether @cfts' subsys is attached or not. | |
79578621 TH |
3989 | * |
3990 | * Returns 0 on successful unregistration, -ENOENT if @cfts is not | |
2bb566cb | 3991 | * registered. |
79578621 | 3992 | */ |
2bb566cb | 3993 | int cgroup_rm_cftypes(struct cftype *cfts) |
79578621 | 3994 | { |
21a2d343 | 3995 | int ret; |
79578621 | 3996 | |
01f6474c | 3997 | mutex_lock(&cgroup_mutex); |
21a2d343 | 3998 | ret = cgroup_rm_cftypes_locked(cfts); |
01f6474c | 3999 | mutex_unlock(&cgroup_mutex); |
21a2d343 | 4000 | return ret; |
80b13586 TH |
4001 | } |
4002 | ||
8e3f6541 TH |
4003 | /** |
4004 | * cgroup_add_cftypes - add an array of cftypes to a subsystem | |
4005 | * @ss: target cgroup subsystem | |
4006 | * @cfts: zero-length name terminated array of cftypes | |
4007 | * | |
4008 | * Register @cfts to @ss. Files described by @cfts are created for all | |
4009 | * existing cgroups to which @ss is attached and all future cgroups will | |
4010 | * have them too. This function can be called anytime whether @ss is | |
4011 | * attached or not. | |
4012 | * | |
4013 | * Returns 0 on successful registration, -errno on failure. Note that this | |
4014 | * function currently returns 0 as long as @cfts registration is successful | |
4015 | * even if some file creation attempts on existing cgroups fail. | |
4016 | */ | |
2cf669a5 | 4017 | static int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
8e3f6541 | 4018 | { |
9ccece80 | 4019 | int ret; |
8e3f6541 | 4020 | |
fc5ed1e9 | 4021 | if (!cgroup_ssid_enabled(ss->id)) |
c731ae1d LZ |
4022 | return 0; |
4023 | ||
dc5736ed LZ |
4024 | if (!cfts || cfts[0].name[0] == '\0') |
4025 | return 0; | |
2bb566cb | 4026 | |
2bd59d48 TH |
4027 | ret = cgroup_init_cftypes(ss, cfts); |
4028 | if (ret) | |
4029 | return ret; | |
79578621 | 4030 | |
01f6474c | 4031 | mutex_lock(&cgroup_mutex); |
21a2d343 | 4032 | |
0adb0704 | 4033 | list_add_tail(&cfts->node, &ss->cfts); |
21a2d343 | 4034 | ret = cgroup_apply_cftypes(cfts, true); |
9ccece80 | 4035 | if (ret) |
21a2d343 | 4036 | cgroup_rm_cftypes_locked(cfts); |
79578621 | 4037 | |
01f6474c | 4038 | mutex_unlock(&cgroup_mutex); |
9ccece80 | 4039 | return ret; |
79578621 TH |
4040 | } |
4041 | ||
a8ddc821 TH |
4042 | /** |
4043 | * cgroup_add_dfl_cftypes - add an array of cftypes for default hierarchy | |
4044 | * @ss: target cgroup subsystem | |
4045 | * @cfts: zero-length name terminated array of cftypes | |
4046 | * | |
4047 | * Similar to cgroup_add_cftypes() but the added files are only used for | |
4048 | * the default hierarchy. | |
4049 | */ | |
4050 | int cgroup_add_dfl_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) | |
4051 | { | |
4052 | struct cftype *cft; | |
4053 | ||
4054 | for (cft = cfts; cft && cft->name[0] != '\0'; cft++) | |
05ebb6e6 | 4055 | cft->flags |= __CFTYPE_ONLY_ON_DFL; |
a8ddc821 TH |
4056 | return cgroup_add_cftypes(ss, cfts); |
4057 | } | |
4058 | ||
4059 | /** | |
4060 | * cgroup_add_legacy_cftypes - add an array of cftypes for legacy hierarchies | |
4061 | * @ss: target cgroup subsystem | |
4062 | * @cfts: zero-length name terminated array of cftypes | |
4063 | * | |
4064 | * Similar to cgroup_add_cftypes() but the added files are only used for | |
4065 | * the legacy hierarchies. | |
4066 | */ | |
2cf669a5 TH |
4067 | int cgroup_add_legacy_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
4068 | { | |
a8ddc821 TH |
4069 | struct cftype *cft; |
4070 | ||
e4b7037c TH |
4071 | for (cft = cfts; cft && cft->name[0] != '\0'; cft++) |
4072 | cft->flags |= __CFTYPE_NOT_ON_DFL; | |
2cf669a5 TH |
4073 | return cgroup_add_cftypes(ss, cfts); |
4074 | } | |
4075 | ||
34c06254 TH |
4076 | /** |
4077 | * cgroup_file_notify - generate a file modified event for a cgroup_file | |
4078 | * @cfile: target cgroup_file | |
4079 | * | |
4080 | * @cfile must have been obtained by setting cftype->file_offset. | |
4081 | */ | |
4082 | void cgroup_file_notify(struct cgroup_file *cfile) | |
4083 | { | |
4084 | unsigned long flags; | |
4085 | ||
4086 | spin_lock_irqsave(&cgroup_file_kn_lock, flags); | |
b12e3583 TH |
4087 | if (cfile->kn) { |
4088 | unsigned long last = cfile->notified_at; | |
4089 | unsigned long next = last + CGROUP_FILE_NOTIFY_MIN_INTV; | |
4090 | ||
4091 | if (time_in_range(jiffies, last, next)) { | |
4092 | timer_reduce(&cfile->notify_timer, next); | |
4093 | } else { | |
4094 | kernfs_notify(cfile->kn); | |
4095 | cfile->notified_at = jiffies; | |
4096 | } | |
4097 | } | |
34c06254 TH |
4098 | spin_unlock_irqrestore(&cgroup_file_kn_lock, flags); |
4099 | } | |
4100 | ||
53fa5261 | 4101 | /** |
492eb21b | 4102 | * css_next_child - find the next child of a given css |
c2931b70 TH |
4103 | * @pos: the current position (%NULL to initiate traversal) |
4104 | * @parent: css whose children to walk | |
53fa5261 | 4105 | * |
c2931b70 | 4106 | * This function returns the next child of @parent and should be called |
87fb54f1 | 4107 | * under either cgroup_mutex or RCU read lock. The only requirement is |
c2931b70 TH |
4108 | * that @parent and @pos are accessible. The next sibling is guaranteed to |
4109 | * be returned regardless of their states. | |
4110 | * | |
4111 | * If a subsystem synchronizes ->css_online() and the start of iteration, a | |
4112 | * css which finished ->css_online() is guaranteed to be visible in the | |
4113 | * future iterations and will stay visible until the last reference is put. | |
4114 | * A css which hasn't finished ->css_online() or already finished | |
4115 | * ->css_offline() may show up during traversal. It's each subsystem's | |
4116 | * responsibility to synchronize against on/offlining. | |
53fa5261 | 4117 | */ |
c2931b70 TH |
4118 | struct cgroup_subsys_state *css_next_child(struct cgroup_subsys_state *pos, |
4119 | struct cgroup_subsys_state *parent) | |
53fa5261 | 4120 | { |
c2931b70 | 4121 | struct cgroup_subsys_state *next; |
53fa5261 | 4122 | |
8353da1f | 4123 | cgroup_assert_mutex_or_rcu_locked(); |
53fa5261 TH |
4124 | |
4125 | /* | |
de3f0341 TH |
4126 | * @pos could already have been unlinked from the sibling list. |
4127 | * Once a cgroup is removed, its ->sibling.next is no longer | |
4128 | * updated when its next sibling changes. CSS_RELEASED is set when | |
4129 | * @pos is taken off list, at which time its next pointer is valid, | |
4130 | * and, as releases are serialized, the one pointed to by the next | |
4131 | * pointer is guaranteed to not have started release yet. This | |
4132 | * implies that if we observe !CSS_RELEASED on @pos in this RCU | |
4133 | * critical section, the one pointed to by its next pointer is | |
4134 | * guaranteed to not have finished its RCU grace period even if we | |
4135 | * have dropped rcu_read_lock() inbetween iterations. | |
3b287a50 | 4136 | * |
de3f0341 TH |
4137 | * If @pos has CSS_RELEASED set, its next pointer can't be |
4138 | * dereferenced; however, as each css is given a monotonically | |
4139 | * increasing unique serial number and always appended to the | |
4140 | * sibling list, the next one can be found by walking the parent's | |
4141 | * children until the first css with higher serial number than | |
4142 | * @pos's. While this path can be slower, it happens iff iteration | |
4143 | * races against release and the race window is very small. | |
53fa5261 | 4144 | */ |
3b287a50 | 4145 | if (!pos) { |
c2931b70 TH |
4146 | next = list_entry_rcu(parent->children.next, struct cgroup_subsys_state, sibling); |
4147 | } else if (likely(!(pos->flags & CSS_RELEASED))) { | |
4148 | next = list_entry_rcu(pos->sibling.next, struct cgroup_subsys_state, sibling); | |
3b287a50 | 4149 | } else { |
c2931b70 | 4150 | list_for_each_entry_rcu(next, &parent->children, sibling) |
3b287a50 TH |
4151 | if (next->serial_nr > pos->serial_nr) |
4152 | break; | |
53fa5261 TH |
4153 | } |
4154 | ||
3b281afb TH |
4155 | /* |
4156 | * @next, if not pointing to the head, can be dereferenced and is | |
c2931b70 | 4157 | * the next sibling. |
3b281afb | 4158 | */ |
c2931b70 TH |
4159 | if (&next->sibling != &parent->children) |
4160 | return next; | |
3b281afb | 4161 | return NULL; |
53fa5261 | 4162 | } |
53fa5261 | 4163 | |
574bd9f7 | 4164 | /** |
492eb21b | 4165 | * css_next_descendant_pre - find the next descendant for pre-order walk |
574bd9f7 | 4166 | * @pos: the current position (%NULL to initiate traversal) |
492eb21b | 4167 | * @root: css whose descendants to walk |
574bd9f7 | 4168 | * |
492eb21b | 4169 | * To be used by css_for_each_descendant_pre(). Find the next descendant |
bd8815a6 TH |
4170 | * to visit for pre-order traversal of @root's descendants. @root is |
4171 | * included in the iteration and the first node to be visited. | |
75501a6d | 4172 | * |
87fb54f1 TH |
4173 | * While this function requires cgroup_mutex or RCU read locking, it |
4174 | * doesn't require the whole traversal to be contained in a single critical | |
4175 | * section. This function will return the correct next descendant as long | |
4176 | * as both @pos and @root are accessible and @pos is a descendant of @root. | |
c2931b70 TH |
4177 | * |
4178 | * If a subsystem synchronizes ->css_online() and the start of iteration, a | |
4179 | * css which finished ->css_online() is guaranteed to be visible in the | |
4180 | * future iterations and will stay visible until the last reference is put. | |
4181 | * A css which hasn't finished ->css_online() or already finished | |
4182 | * ->css_offline() may show up during traversal. It's each subsystem's | |
4183 | * responsibility to synchronize against on/offlining. | |
574bd9f7 | 4184 | */ |
492eb21b TH |
4185 | struct cgroup_subsys_state * |
4186 | css_next_descendant_pre(struct cgroup_subsys_state *pos, | |
4187 | struct cgroup_subsys_state *root) | |
574bd9f7 | 4188 | { |
492eb21b | 4189 | struct cgroup_subsys_state *next; |
574bd9f7 | 4190 | |
8353da1f | 4191 | cgroup_assert_mutex_or_rcu_locked(); |
574bd9f7 | 4192 | |
bd8815a6 | 4193 | /* if first iteration, visit @root */ |
7805d000 | 4194 | if (!pos) |
bd8815a6 | 4195 | return root; |
574bd9f7 TH |
4196 | |
4197 | /* visit the first child if exists */ | |
492eb21b | 4198 | next = css_next_child(NULL, pos); |
574bd9f7 TH |
4199 | if (next) |
4200 | return next; | |
4201 | ||
4202 | /* no child, visit my or the closest ancestor's next sibling */ | |
492eb21b | 4203 | while (pos != root) { |
5c9d535b | 4204 | next = css_next_child(pos, pos->parent); |
75501a6d | 4205 | if (next) |
574bd9f7 | 4206 | return next; |
5c9d535b | 4207 | pos = pos->parent; |
7805d000 | 4208 | } |
574bd9f7 TH |
4209 | |
4210 | return NULL; | |
4211 | } | |
474a2800 | 4212 | EXPORT_SYMBOL_GPL(css_next_descendant_pre); |
574bd9f7 | 4213 | |
12a9d2fe | 4214 | /** |
492eb21b TH |
4215 | * css_rightmost_descendant - return the rightmost descendant of a css |
4216 | * @pos: css of interest | |
12a9d2fe | 4217 | * |
492eb21b TH |
4218 | * Return the rightmost descendant of @pos. If there's no descendant, @pos |
4219 | * is returned. This can be used during pre-order traversal to skip | |
12a9d2fe | 4220 | * subtree of @pos. |
75501a6d | 4221 | * |
87fb54f1 TH |
4222 | * While this function requires cgroup_mutex or RCU read locking, it |
4223 | * doesn't require the whole traversal to be contained in a single critical | |
4224 | * section. This function will return the correct rightmost descendant as | |
4225 | * long as @pos is accessible. | |
12a9d2fe | 4226 | */ |
492eb21b TH |
4227 | struct cgroup_subsys_state * |
4228 | css_rightmost_descendant(struct cgroup_subsys_state *pos) | |
12a9d2fe | 4229 | { |
492eb21b | 4230 | struct cgroup_subsys_state *last, *tmp; |
12a9d2fe | 4231 | |
8353da1f | 4232 | cgroup_assert_mutex_or_rcu_locked(); |
12a9d2fe TH |
4233 | |
4234 | do { | |
4235 | last = pos; | |
4236 | /* ->prev isn't RCU safe, walk ->next till the end */ | |
4237 | pos = NULL; | |
492eb21b | 4238 | css_for_each_child(tmp, last) |
12a9d2fe TH |
4239 | pos = tmp; |
4240 | } while (pos); | |
4241 | ||
4242 | return last; | |
4243 | } | |
12a9d2fe | 4244 | |
492eb21b TH |
4245 | static struct cgroup_subsys_state * |
4246 | css_leftmost_descendant(struct cgroup_subsys_state *pos) | |
574bd9f7 | 4247 | { |
492eb21b | 4248 | struct cgroup_subsys_state *last; |
574bd9f7 TH |
4249 | |
4250 | do { | |
4251 | last = pos; | |
492eb21b | 4252 | pos = css_next_child(NULL, pos); |
574bd9f7 TH |
4253 | } while (pos); |
4254 | ||
4255 | return last; | |
4256 | } | |
4257 | ||
4258 | /** | |
492eb21b | 4259 | * css_next_descendant_post - find the next descendant for post-order walk |
574bd9f7 | 4260 | * @pos: the current position (%NULL to initiate traversal) |
492eb21b | 4261 | * @root: css whose descendants to walk |
574bd9f7 | 4262 | * |
492eb21b | 4263 | * To be used by css_for_each_descendant_post(). Find the next descendant |
bd8815a6 TH |
4264 | * to visit for post-order traversal of @root's descendants. @root is |
4265 | * included in the iteration and the last node to be visited. | |
75501a6d | 4266 | * |
87fb54f1 TH |
4267 | * While this function requires cgroup_mutex or RCU read locking, it |
4268 | * doesn't require the whole traversal to be contained in a single critical | |
4269 | * section. This function will return the correct next descendant as long | |
4270 | * as both @pos and @cgroup are accessible and @pos is a descendant of | |
4271 | * @cgroup. | |
c2931b70 TH |
4272 | * |
4273 | * If a subsystem synchronizes ->css_online() and the start of iteration, a | |
4274 | * css which finished ->css_online() is guaranteed to be visible in the | |
4275 | * future iterations and will stay visible until the last reference is put. | |
4276 | * A css which hasn't finished ->css_online() or already finished | |
4277 | * ->css_offline() may show up during traversal. It's each subsystem's | |
4278 | * responsibility to synchronize against on/offlining. | |
574bd9f7 | 4279 | */ |
492eb21b TH |
4280 | struct cgroup_subsys_state * |
4281 | css_next_descendant_post(struct cgroup_subsys_state *pos, | |
4282 | struct cgroup_subsys_state *root) | |
574bd9f7 | 4283 | { |
492eb21b | 4284 | struct cgroup_subsys_state *next; |
574bd9f7 | 4285 | |
8353da1f | 4286 | cgroup_assert_mutex_or_rcu_locked(); |
574bd9f7 | 4287 | |
58b79a91 TH |
4288 | /* if first iteration, visit leftmost descendant which may be @root */ |
4289 | if (!pos) | |
4290 | return css_leftmost_descendant(root); | |
574bd9f7 | 4291 | |
bd8815a6 TH |
4292 | /* if we visited @root, we're done */ |
4293 | if (pos == root) | |
4294 | return NULL; | |
4295 | ||
574bd9f7 | 4296 | /* if there's an unvisited sibling, visit its leftmost descendant */ |
5c9d535b | 4297 | next = css_next_child(pos, pos->parent); |
75501a6d | 4298 | if (next) |
492eb21b | 4299 | return css_leftmost_descendant(next); |
574bd9f7 TH |
4300 | |
4301 | /* no sibling left, visit parent */ | |
5c9d535b | 4302 | return pos->parent; |
574bd9f7 | 4303 | } |
574bd9f7 | 4304 | |
f3d46500 TH |
4305 | /** |
4306 | * css_has_online_children - does a css have online children | |
4307 | * @css: the target css | |
4308 | * | |
4309 | * Returns %true if @css has any online children; otherwise, %false. This | |
4310 | * function can be called from any context but the caller is responsible | |
4311 | * for synchronizing against on/offlining as necessary. | |
4312 | */ | |
4313 | bool css_has_online_children(struct cgroup_subsys_state *css) | |
cbc125ef | 4314 | { |
f3d46500 TH |
4315 | struct cgroup_subsys_state *child; |
4316 | bool ret = false; | |
cbc125ef TH |
4317 | |
4318 | rcu_read_lock(); | |
f3d46500 | 4319 | css_for_each_child(child, css) { |
99bae5f9 | 4320 | if (child->flags & CSS_ONLINE) { |
f3d46500 TH |
4321 | ret = true; |
4322 | break; | |
cbc125ef TH |
4323 | } |
4324 | } | |
4325 | rcu_read_unlock(); | |
f3d46500 | 4326 | return ret; |
574bd9f7 | 4327 | } |
574bd9f7 | 4328 | |
450ee0c1 TH |
4329 | static struct css_set *css_task_iter_next_css_set(struct css_task_iter *it) |
4330 | { | |
4331 | struct list_head *l; | |
4332 | struct cgrp_cset_link *link; | |
4333 | struct css_set *cset; | |
4334 | ||
4335 | lockdep_assert_held(&css_set_lock); | |
4336 | ||
4337 | /* find the next threaded cset */ | |
4338 | if (it->tcset_pos) { | |
4339 | l = it->tcset_pos->next; | |
4340 | ||
4341 | if (l != it->tcset_head) { | |
4342 | it->tcset_pos = l; | |
4343 | return container_of(l, struct css_set, | |
4344 | threaded_csets_node); | |
4345 | } | |
4346 | ||
4347 | it->tcset_pos = NULL; | |
4348 | } | |
4349 | ||
4350 | /* find the next cset */ | |
4351 | l = it->cset_pos; | |
4352 | l = l->next; | |
4353 | if (l == it->cset_head) { | |
4354 | it->cset_pos = NULL; | |
4355 | return NULL; | |
4356 | } | |
4357 | ||
4358 | if (it->ss) { | |
4359 | cset = container_of(l, struct css_set, e_cset_node[it->ss->id]); | |
4360 | } else { | |
4361 | link = list_entry(l, struct cgrp_cset_link, cset_link); | |
4362 | cset = link->cset; | |
4363 | } | |
4364 | ||
4365 | it->cset_pos = l; | |
4366 | ||
4367 | /* initialize threaded css_set walking */ | |
4368 | if (it->flags & CSS_TASK_ITER_THREADED) { | |
4369 | if (it->cur_dcset) | |
4370 | put_css_set_locked(it->cur_dcset); | |
4371 | it->cur_dcset = cset; | |
4372 | get_css_set(cset); | |
4373 | ||
4374 | it->tcset_head = &cset->threaded_csets; | |
4375 | it->tcset_pos = &cset->threaded_csets; | |
4376 | } | |
4377 | ||
4378 | return cset; | |
4379 | } | |
4380 | ||
0942eeee | 4381 | /** |
ecb9d535 | 4382 | * css_task_iter_advance_css_set - advance a task itererator to the next css_set |
0942eeee TH |
4383 | * @it: the iterator to advance |
4384 | * | |
4385 | * Advance @it to the next css_set to walk. | |
d515876e | 4386 | */ |
ecb9d535 | 4387 | static void css_task_iter_advance_css_set(struct css_task_iter *it) |
d515876e | 4388 | { |
d515876e TH |
4389 | struct css_set *cset; |
4390 | ||
f0d9a5f1 | 4391 | lockdep_assert_held(&css_set_lock); |
ed27b9f7 | 4392 | |
d515876e TH |
4393 | /* Advance to the next non-empty css_set */ |
4394 | do { | |
450ee0c1 TH |
4395 | cset = css_task_iter_next_css_set(it); |
4396 | if (!cset) { | |
ecb9d535 | 4397 | it->task_pos = NULL; |
d515876e TH |
4398 | return; |
4399 | } | |
c596687a | 4400 | } while (!css_set_populated(cset) && list_empty(&cset->dying_tasks)); |
c7561128 | 4401 | |
c7561128 | 4402 | if (!list_empty(&cset->tasks)) |
0f0a2b4f | 4403 | it->task_pos = cset->tasks.next; |
c03cd773 | 4404 | else if (!list_empty(&cset->mg_tasks)) |
0f0a2b4f | 4405 | it->task_pos = cset->mg_tasks.next; |
c03cd773 TH |
4406 | else |
4407 | it->task_pos = cset->dying_tasks.next; | |
0f0a2b4f TH |
4408 | |
4409 | it->tasks_head = &cset->tasks; | |
4410 | it->mg_tasks_head = &cset->mg_tasks; | |
c03cd773 | 4411 | it->dying_tasks_head = &cset->dying_tasks; |
ed27b9f7 TH |
4412 | |
4413 | /* | |
4414 | * We don't keep css_sets locked across iteration steps and thus | |
4415 | * need to take steps to ensure that iteration can be resumed after | |
4416 | * the lock is re-acquired. Iteration is performed at two levels - | |
4417 | * css_sets and tasks in them. | |
4418 | * | |
4419 | * Once created, a css_set never leaves its cgroup lists, so a | |
4420 | * pinned css_set is guaranteed to stay put and we can resume | |
4421 | * iteration afterwards. | |
4422 | * | |
4423 | * Tasks may leave @cset across iteration steps. This is resolved | |
4424 | * by registering each iterator with the css_set currently being | |
4425 | * walked and making css_set_move_task() advance iterators whose | |
4426 | * next task is leaving. | |
4427 | */ | |
4428 | if (it->cur_cset) { | |
4429 | list_del(&it->iters_node); | |
4430 | put_css_set_locked(it->cur_cset); | |
4431 | } | |
4432 | get_css_set(cset); | |
4433 | it->cur_cset = cset; | |
4434 | list_add(&it->iters_node, &cset->task_iters); | |
d515876e TH |
4435 | } |
4436 | ||
b636fd38 TH |
4437 | static void css_task_iter_skip(struct css_task_iter *it, |
4438 | struct task_struct *task) | |
ecb9d535 | 4439 | { |
b636fd38 TH |
4440 | lockdep_assert_held(&css_set_lock); |
4441 | ||
4442 | if (it->task_pos == &task->cg_list) { | |
4443 | it->task_pos = it->task_pos->next; | |
4444 | it->flags |= CSS_TASK_ITER_SKIPPED; | |
4445 | } | |
4446 | } | |
ecb9d535 | 4447 | |
ecb9d535 TH |
4448 | static void css_task_iter_advance(struct css_task_iter *it) |
4449 | { | |
c03cd773 | 4450 | struct task_struct *task; |
ecb9d535 | 4451 | |
f0d9a5f1 | 4452 | lockdep_assert_held(&css_set_lock); |
bc2fb7ed | 4453 | repeat: |
e9d81a1b TH |
4454 | if (it->task_pos) { |
4455 | /* | |
4456 | * Advance iterator to find next entry. cset->tasks is | |
4457 | * consumed first and then ->mg_tasks. After ->mg_tasks, | |
4458 | * we move onto the next cset. | |
4459 | */ | |
b636fd38 TH |
4460 | if (it->flags & CSS_TASK_ITER_SKIPPED) |
4461 | it->flags &= ~CSS_TASK_ITER_SKIPPED; | |
4462 | else | |
4463 | it->task_pos = it->task_pos->next; | |
ecb9d535 | 4464 | |
b636fd38 TH |
4465 | if (it->task_pos == it->tasks_head) |
4466 | it->task_pos = it->mg_tasks_head->next; | |
4467 | if (it->task_pos == it->mg_tasks_head) | |
c03cd773 TH |
4468 | it->task_pos = it->dying_tasks_head->next; |
4469 | if (it->task_pos == it->dying_tasks_head) | |
e9d81a1b | 4470 | css_task_iter_advance_css_set(it); |
e9d81a1b TH |
4471 | } else { |
4472 | /* called from start, proceed to the first cset */ | |
ecb9d535 | 4473 | css_task_iter_advance_css_set(it); |
e9d81a1b | 4474 | } |
bc2fb7ed | 4475 | |
c03cd773 TH |
4476 | if (!it->task_pos) |
4477 | return; | |
4478 | ||
4479 | task = list_entry(it->task_pos, struct task_struct, cg_list); | |
4480 | ||
4481 | if (it->flags & CSS_TASK_ITER_PROCS) { | |
4482 | /* if PROCS, skip over tasks which aren't group leaders */ | |
4483 | if (!thread_group_leader(task)) | |
4484 | goto repeat; | |
4485 | ||
4486 | /* and dying leaders w/o live member threads */ | |
4487 | if (!atomic_read(&task->signal->live)) | |
4488 | goto repeat; | |
4489 | } else { | |
4490 | /* skip all dying ones */ | |
4491 | if (task->flags & PF_EXITING) | |
4492 | goto repeat; | |
4493 | } | |
ecb9d535 TH |
4494 | } |
4495 | ||
0942eeee | 4496 | /** |
72ec7029 TH |
4497 | * css_task_iter_start - initiate task iteration |
4498 | * @css: the css to walk tasks of | |
bc2fb7ed | 4499 | * @flags: CSS_TASK_ITER_* flags |
0942eeee TH |
4500 | * @it: the task iterator to use |
4501 | * | |
72ec7029 TH |
4502 | * Initiate iteration through the tasks of @css. The caller can call |
4503 | * css_task_iter_next() to walk through the tasks until the function | |
4504 | * returns NULL. On completion of iteration, css_task_iter_end() must be | |
4505 | * called. | |
0942eeee | 4506 | */ |
bc2fb7ed | 4507 | void css_task_iter_start(struct cgroup_subsys_state *css, unsigned int flags, |
72ec7029 | 4508 | struct css_task_iter *it) |
817929ec | 4509 | { |
ed27b9f7 TH |
4510 | memset(it, 0, sizeof(*it)); |
4511 | ||
82d6489d | 4512 | spin_lock_irq(&css_set_lock); |
c59cd3d8 | 4513 | |
3ebb2b6e | 4514 | it->ss = css->ss; |
bc2fb7ed | 4515 | it->flags = flags; |
3ebb2b6e TH |
4516 | |
4517 | if (it->ss) | |
4518 | it->cset_pos = &css->cgroup->e_csets[css->ss->id]; | |
4519 | else | |
4520 | it->cset_pos = &css->cgroup->cset_links; | |
4521 | ||
0f0a2b4f | 4522 | it->cset_head = it->cset_pos; |
c59cd3d8 | 4523 | |
e9d81a1b | 4524 | css_task_iter_advance(it); |
ed27b9f7 | 4525 | |
82d6489d | 4526 | spin_unlock_irq(&css_set_lock); |
817929ec PM |
4527 | } |
4528 | ||
0942eeee | 4529 | /** |
72ec7029 | 4530 | * css_task_iter_next - return the next task for the iterator |
0942eeee TH |
4531 | * @it: the task iterator being iterated |
4532 | * | |
4533 | * The "next" function for task iteration. @it should have been | |
72ec7029 TH |
4534 | * initialized via css_task_iter_start(). Returns NULL when the iteration |
4535 | * reaches the end. | |
0942eeee | 4536 | */ |
72ec7029 | 4537 | struct task_struct *css_task_iter_next(struct css_task_iter *it) |
817929ec | 4538 | { |
d5745675 | 4539 | if (it->cur_task) { |
ed27b9f7 | 4540 | put_task_struct(it->cur_task); |
d5745675 TH |
4541 | it->cur_task = NULL; |
4542 | } | |
ed27b9f7 | 4543 | |
82d6489d | 4544 | spin_lock_irq(&css_set_lock); |
ed27b9f7 | 4545 | |
cee0c33c TH |
4546 | /* @it may be half-advanced by skips, finish advancing */ |
4547 | if (it->flags & CSS_TASK_ITER_SKIPPED) | |
4548 | css_task_iter_advance(it); | |
4549 | ||
d5745675 TH |
4550 | if (it->task_pos) { |
4551 | it->cur_task = list_entry(it->task_pos, struct task_struct, | |
4552 | cg_list); | |
4553 | get_task_struct(it->cur_task); | |
4554 | css_task_iter_advance(it); | |
4555 | } | |
ed27b9f7 | 4556 | |
82d6489d | 4557 | spin_unlock_irq(&css_set_lock); |
ed27b9f7 TH |
4558 | |
4559 | return it->cur_task; | |
817929ec PM |
4560 | } |
4561 | ||
0942eeee | 4562 | /** |
72ec7029 | 4563 | * css_task_iter_end - finish task iteration |
0942eeee TH |
4564 | * @it: the task iterator to finish |
4565 | * | |
72ec7029 | 4566 | * Finish task iteration started by css_task_iter_start(). |
0942eeee | 4567 | */ |
72ec7029 | 4568 | void css_task_iter_end(struct css_task_iter *it) |
31a7df01 | 4569 | { |
ed27b9f7 | 4570 | if (it->cur_cset) { |
82d6489d | 4571 | spin_lock_irq(&css_set_lock); |
ed27b9f7 TH |
4572 | list_del(&it->iters_node); |
4573 | put_css_set_locked(it->cur_cset); | |
82d6489d | 4574 | spin_unlock_irq(&css_set_lock); |
ed27b9f7 TH |
4575 | } |
4576 | ||
450ee0c1 TH |
4577 | if (it->cur_dcset) |
4578 | put_css_set(it->cur_dcset); | |
4579 | ||
ed27b9f7 TH |
4580 | if (it->cur_task) |
4581 | put_task_struct(it->cur_task); | |
31a7df01 CW |
4582 | } |
4583 | ||
b4b90a8e | 4584 | static void cgroup_procs_release(struct kernfs_open_file *of) |
31a7df01 | 4585 | { |
b4b90a8e TH |
4586 | if (of->priv) { |
4587 | css_task_iter_end(of->priv); | |
4588 | kfree(of->priv); | |
4589 | } | |
4590 | } | |
6c694c88 | 4591 | |
b4b90a8e TH |
4592 | static void *cgroup_procs_next(struct seq_file *s, void *v, loff_t *pos) |
4593 | { | |
4594 | struct kernfs_open_file *of = s->private; | |
4595 | struct css_task_iter *it = of->priv; | |
31a7df01 | 4596 | |
bc2fb7ed | 4597 | return css_task_iter_next(it); |
b4b90a8e | 4598 | } |
31a7df01 | 4599 | |
8cfd8147 TH |
4600 | static void *__cgroup_procs_start(struct seq_file *s, loff_t *pos, |
4601 | unsigned int iter_flags) | |
b4b90a8e TH |
4602 | { |
4603 | struct kernfs_open_file *of = s->private; | |
4604 | struct cgroup *cgrp = seq_css(s)->cgroup; | |
4605 | struct css_task_iter *it = of->priv; | |
8cc99345 | 4606 | |
952aaa12 | 4607 | /* |
b4b90a8e TH |
4608 | * When a seq_file is seeked, it's always traversed sequentially |
4609 | * from position 0, so we can simply keep iterating on !0 *pos. | |
952aaa12 | 4610 | */ |
b4b90a8e TH |
4611 | if (!it) { |
4612 | if (WARN_ON_ONCE((*pos)++)) | |
4613 | return ERR_PTR(-EINVAL); | |
8cc99345 | 4614 | |
b4b90a8e TH |
4615 | it = kzalloc(sizeof(*it), GFP_KERNEL); |
4616 | if (!it) | |
4617 | return ERR_PTR(-ENOMEM); | |
4618 | of->priv = it; | |
450ee0c1 | 4619 | css_task_iter_start(&cgrp->self, iter_flags, it); |
b4b90a8e TH |
4620 | } else if (!(*pos)++) { |
4621 | css_task_iter_end(it); | |
450ee0c1 | 4622 | css_task_iter_start(&cgrp->self, iter_flags, it); |
b4b90a8e | 4623 | } |
bbcb81d0 | 4624 | |
b4b90a8e TH |
4625 | return cgroup_procs_next(s, NULL, NULL); |
4626 | } | |
24528255 | 4627 | |
8cfd8147 TH |
4628 | static void *cgroup_procs_start(struct seq_file *s, loff_t *pos) |
4629 | { | |
4630 | struct cgroup *cgrp = seq_css(s)->cgroup; | |
4631 | ||
4632 | /* | |
4633 | * All processes of a threaded subtree belong to the domain cgroup | |
4634 | * of the subtree. Only threads can be distributed across the | |
4635 | * subtree. Reject reads on cgroup.procs in the subtree proper. | |
4636 | * They're always empty anyway. | |
4637 | */ | |
4638 | if (cgroup_is_threaded(cgrp)) | |
4639 | return ERR_PTR(-EOPNOTSUPP); | |
4640 | ||
4641 | return __cgroup_procs_start(s, pos, CSS_TASK_ITER_PROCS | | |
4642 | CSS_TASK_ITER_THREADED); | |
4643 | } | |
4644 | ||
b4b90a8e | 4645 | static int cgroup_procs_show(struct seq_file *s, void *v) |
bbcb81d0 | 4646 | { |
bc2fb7ed | 4647 | seq_printf(s, "%d\n", task_pid_vnr(v)); |
97978e6d DL |
4648 | return 0; |
4649 | } | |
4650 | ||
715c809d TH |
4651 | static int cgroup_procs_write_permission(struct cgroup *src_cgrp, |
4652 | struct cgroup *dst_cgrp, | |
4653 | struct super_block *sb) | |
4654 | { | |
4655 | struct cgroup_namespace *ns = current->nsproxy->cgroup_ns; | |
4656 | struct cgroup *com_cgrp = src_cgrp; | |
4657 | struct inode *inode; | |
4658 | int ret; | |
4659 | ||
4660 | lockdep_assert_held(&cgroup_mutex); | |
4661 | ||
4662 | /* find the common ancestor */ | |
4663 | while (!cgroup_is_descendant(dst_cgrp, com_cgrp)) | |
4664 | com_cgrp = cgroup_parent(com_cgrp); | |
4665 | ||
4666 | /* %current should be authorized to migrate to the common ancestor */ | |
4667 | inode = kernfs_get_inode(sb, com_cgrp->procs_file.kn); | |
4668 | if (!inode) | |
4669 | return -ENOMEM; | |
4670 | ||
4671 | ret = inode_permission(inode, MAY_WRITE); | |
4672 | iput(inode); | |
4673 | if (ret) | |
4674 | return ret; | |
4675 | ||
4676 | /* | |
4677 | * If namespaces are delegation boundaries, %current must be able | |
4678 | * to see both source and destination cgroups from its namespace. | |
4679 | */ | |
4680 | if ((cgrp_dfl_root.flags & CGRP_ROOT_NS_DELEGATE) && | |
4681 | (!cgroup_is_descendant(src_cgrp, ns->root_cset->dfl_cgrp) || | |
4682 | !cgroup_is_descendant(dst_cgrp, ns->root_cset->dfl_cgrp))) | |
4683 | return -ENOENT; | |
4684 | ||
4685 | return 0; | |
4686 | } | |
4687 | ||
4688 | static ssize_t cgroup_procs_write(struct kernfs_open_file *of, | |
4689 | char *buf, size_t nbytes, loff_t off) | |
4690 | { | |
4691 | struct cgroup *src_cgrp, *dst_cgrp; | |
4692 | struct task_struct *task; | |
4693 | ssize_t ret; | |
9a3284fa | 4694 | bool locked; |
715c809d TH |
4695 | |
4696 | dst_cgrp = cgroup_kn_lock_live(of->kn, false); | |
4697 | if (!dst_cgrp) | |
4698 | return -ENODEV; | |
4699 | ||
9a3284fa | 4700 | task = cgroup_procs_write_start(buf, true, &locked); |
715c809d TH |
4701 | ret = PTR_ERR_OR_ZERO(task); |
4702 | if (ret) | |
4703 | goto out_unlock; | |
4704 | ||
4705 | /* find the source cgroup */ | |
4706 | spin_lock_irq(&css_set_lock); | |
4707 | src_cgrp = task_cgroup_from_root(task, &cgrp_dfl_root); | |
4708 | spin_unlock_irq(&css_set_lock); | |
4709 | ||
4710 | ret = cgroup_procs_write_permission(src_cgrp, dst_cgrp, | |
4711 | of->file->f_path.dentry->d_sb); | |
4712 | if (ret) | |
4713 | goto out_finish; | |
4714 | ||
4715 | ret = cgroup_attach_task(dst_cgrp, task, true); | |
4716 | ||
4717 | out_finish: | |
9a3284fa | 4718 | cgroup_procs_write_finish(task, locked); |
715c809d TH |
4719 | out_unlock: |
4720 | cgroup_kn_unlock(of->kn); | |
4721 | ||
4722 | return ret ?: nbytes; | |
4723 | } | |
4724 | ||
8cfd8147 TH |
4725 | static void *cgroup_threads_start(struct seq_file *s, loff_t *pos) |
4726 | { | |
4727 | return __cgroup_procs_start(s, pos, 0); | |
4728 | } | |
4729 | ||
4730 | static ssize_t cgroup_threads_write(struct kernfs_open_file *of, | |
4731 | char *buf, size_t nbytes, loff_t off) | |
4732 | { | |
4733 | struct cgroup *src_cgrp, *dst_cgrp; | |
4734 | struct task_struct *task; | |
4735 | ssize_t ret; | |
9a3284fa | 4736 | bool locked; |
8cfd8147 TH |
4737 | |
4738 | buf = strstrip(buf); | |
4739 | ||
4740 | dst_cgrp = cgroup_kn_lock_live(of->kn, false); | |
4741 | if (!dst_cgrp) | |
4742 | return -ENODEV; | |
4743 | ||
9a3284fa | 4744 | task = cgroup_procs_write_start(buf, false, &locked); |
8cfd8147 TH |
4745 | ret = PTR_ERR_OR_ZERO(task); |
4746 | if (ret) | |
4747 | goto out_unlock; | |
4748 | ||
4749 | /* find the source cgroup */ | |
4750 | spin_lock_irq(&css_set_lock); | |
4751 | src_cgrp = task_cgroup_from_root(task, &cgrp_dfl_root); | |
4752 | spin_unlock_irq(&css_set_lock); | |
4753 | ||
4754 | /* thread migrations follow the cgroup.procs delegation rule */ | |
4755 | ret = cgroup_procs_write_permission(src_cgrp, dst_cgrp, | |
4756 | of->file->f_path.dentry->d_sb); | |
4757 | if (ret) | |
4758 | goto out_finish; | |
4759 | ||
4760 | /* and must be contained in the same domain */ | |
4761 | ret = -EOPNOTSUPP; | |
4762 | if (src_cgrp->dom_cgrp != dst_cgrp->dom_cgrp) | |
4763 | goto out_finish; | |
4764 | ||
4765 | ret = cgroup_attach_task(dst_cgrp, task, false); | |
4766 | ||
4767 | out_finish: | |
9a3284fa | 4768 | cgroup_procs_write_finish(task, locked); |
8cfd8147 TH |
4769 | out_unlock: |
4770 | cgroup_kn_unlock(of->kn); | |
4771 | ||
4772 | return ret ?: nbytes; | |
4773 | } | |
4774 | ||
a14c6874 | 4775 | /* cgroup core interface files for the default hierarchy */ |
d62beb7f | 4776 | static struct cftype cgroup_base_files[] = { |
8cfd8147 TH |
4777 | { |
4778 | .name = "cgroup.type", | |
4779 | .flags = CFTYPE_NOT_ON_ROOT, | |
4780 | .seq_show = cgroup_type_show, | |
4781 | .write = cgroup_type_write, | |
4782 | }, | |
81a6a5cd | 4783 | { |
d5c56ced | 4784 | .name = "cgroup.procs", |
5136f636 | 4785 | .flags = CFTYPE_NS_DELEGATABLE, |
6f60eade | 4786 | .file_offset = offsetof(struct cgroup, procs_file), |
b4b90a8e TH |
4787 | .release = cgroup_procs_release, |
4788 | .seq_start = cgroup_procs_start, | |
4789 | .seq_next = cgroup_procs_next, | |
4790 | .seq_show = cgroup_procs_show, | |
acbef755 | 4791 | .write = cgroup_procs_write, |
102a775e | 4792 | }, |
8cfd8147 TH |
4793 | { |
4794 | .name = "cgroup.threads", | |
4f58424d | 4795 | .flags = CFTYPE_NS_DELEGATABLE, |
8cfd8147 TH |
4796 | .release = cgroup_procs_release, |
4797 | .seq_start = cgroup_threads_start, | |
4798 | .seq_next = cgroup_procs_next, | |
4799 | .seq_show = cgroup_procs_show, | |
4800 | .write = cgroup_threads_write, | |
4801 | }, | |
f8f22e53 TH |
4802 | { |
4803 | .name = "cgroup.controllers", | |
f8f22e53 TH |
4804 | .seq_show = cgroup_controllers_show, |
4805 | }, | |
4806 | { | |
4807 | .name = "cgroup.subtree_control", | |
5136f636 | 4808 | .flags = CFTYPE_NS_DELEGATABLE, |
f8f22e53 | 4809 | .seq_show = cgroup_subtree_control_show, |
451af504 | 4810 | .write = cgroup_subtree_control_write, |
f8f22e53 | 4811 | }, |
842b597e | 4812 | { |
4a07c222 | 4813 | .name = "cgroup.events", |
a14c6874 | 4814 | .flags = CFTYPE_NOT_ON_ROOT, |
6f60eade | 4815 | .file_offset = offsetof(struct cgroup, events_file), |
4a07c222 | 4816 | .seq_show = cgroup_events_show, |
842b597e | 4817 | }, |
1a926e0b RG |
4818 | { |
4819 | .name = "cgroup.max.descendants", | |
4820 | .seq_show = cgroup_max_descendants_show, | |
4821 | .write = cgroup_max_descendants_write, | |
4822 | }, | |
4823 | { | |
4824 | .name = "cgroup.max.depth", | |
4825 | .seq_show = cgroup_max_depth_show, | |
4826 | .write = cgroup_max_depth_write, | |
4827 | }, | |
ec39225c RG |
4828 | { |
4829 | .name = "cgroup.stat", | |
3e48930c | 4830 | .seq_show = cgroup_stat_show, |
ec39225c | 4831 | }, |
76f969e8 RG |
4832 | { |
4833 | .name = "cgroup.freeze", | |
4834 | .flags = CFTYPE_NOT_ON_ROOT, | |
4835 | .seq_show = cgroup_freeze_show, | |
4836 | .write = cgroup_freeze_write, | |
4837 | }, | |
d41bf8c9 TH |
4838 | { |
4839 | .name = "cpu.stat", | |
4840 | .flags = CFTYPE_NOT_ON_ROOT, | |
4841 | .seq_show = cpu_stat_show, | |
4842 | }, | |
2ce7135a JW |
4843 | #ifdef CONFIG_PSI |
4844 | { | |
4845 | .name = "io.pressure", | |
2ce7135a | 4846 | .seq_show = cgroup_io_pressure_show, |
0e94682b SB |
4847 | .write = cgroup_io_pressure_write, |
4848 | .poll = cgroup_pressure_poll, | |
4849 | .release = cgroup_pressure_release, | |
2ce7135a JW |
4850 | }, |
4851 | { | |
4852 | .name = "memory.pressure", | |
2ce7135a | 4853 | .seq_show = cgroup_memory_pressure_show, |
0e94682b SB |
4854 | .write = cgroup_memory_pressure_write, |
4855 | .poll = cgroup_pressure_poll, | |
4856 | .release = cgroup_pressure_release, | |
2ce7135a JW |
4857 | }, |
4858 | { | |
4859 | .name = "cpu.pressure", | |
2ce7135a | 4860 | .seq_show = cgroup_cpu_pressure_show, |
0e94682b SB |
4861 | .write = cgroup_cpu_pressure_write, |
4862 | .poll = cgroup_pressure_poll, | |
4863 | .release = cgroup_pressure_release, | |
2ce7135a | 4864 | }, |
0e94682b | 4865 | #endif /* CONFIG_PSI */ |
a14c6874 TH |
4866 | { } /* terminate */ |
4867 | }; | |
d5c56ced | 4868 | |
0c21ead1 TH |
4869 | /* |
4870 | * css destruction is four-stage process. | |
4871 | * | |
4872 | * 1. Destruction starts. Killing of the percpu_ref is initiated. | |
4873 | * Implemented in kill_css(). | |
4874 | * | |
4875 | * 2. When the percpu_ref is confirmed to be visible as killed on all CPUs | |
ec903c0c TH |
4876 | * and thus css_tryget_online() is guaranteed to fail, the css can be |
4877 | * offlined by invoking offline_css(). After offlining, the base ref is | |
4878 | * put. Implemented in css_killed_work_fn(). | |
0c21ead1 TH |
4879 | * |
4880 | * 3. When the percpu_ref reaches zero, the only possible remaining | |
4881 | * accessors are inside RCU read sections. css_release() schedules the | |
4882 | * RCU callback. | |
4883 | * | |
4884 | * 4. After the grace period, the css can be freed. Implemented in | |
4885 | * css_free_work_fn(). | |
4886 | * | |
4887 | * It is actually hairier because both step 2 and 4 require process context | |
4888 | * and thus involve punting to css->destroy_work adding two additional | |
4889 | * steps to the already complex sequence. | |
4890 | */ | |
8f36aaec | 4891 | static void css_free_rwork_fn(struct work_struct *work) |
48ddbe19 | 4892 | { |
8f36aaec TH |
4893 | struct cgroup_subsys_state *css = container_of(to_rcu_work(work), |
4894 | struct cgroup_subsys_state, destroy_rwork); | |
01e58659 | 4895 | struct cgroup_subsys *ss = css->ss; |
0c21ead1 | 4896 | struct cgroup *cgrp = css->cgroup; |
48ddbe19 | 4897 | |
9a1049da TH |
4898 | percpu_ref_exit(&css->refcnt); |
4899 | ||
01e58659 | 4900 | if (ss) { |
9d755d33 | 4901 | /* css free path */ |
8bb5ef79 | 4902 | struct cgroup_subsys_state *parent = css->parent; |
01e58659 VD |
4903 | int id = css->id; |
4904 | ||
01e58659 VD |
4905 | ss->css_free(css); |
4906 | cgroup_idr_remove(&ss->css_idr, id); | |
9d755d33 | 4907 | cgroup_put(cgrp); |
8bb5ef79 TH |
4908 | |
4909 | if (parent) | |
4910 | css_put(parent); | |
9d755d33 TH |
4911 | } else { |
4912 | /* cgroup free path */ | |
4913 | atomic_dec(&cgrp->root->nr_cgrps); | |
d62beb7f | 4914 | cgroup1_pidlist_destroy_all(cgrp); |
971ff493 | 4915 | cancel_work_sync(&cgrp->release_agent_work); |
9d755d33 | 4916 | |
d51f39b0 | 4917 | if (cgroup_parent(cgrp)) { |
9d755d33 TH |
4918 | /* |
4919 | * We get a ref to the parent, and put the ref when | |
4920 | * this cgroup is being freed, so it's guaranteed | |
4921 | * that the parent won't be destroyed before its | |
4922 | * children. | |
4923 | */ | |
d51f39b0 | 4924 | cgroup_put(cgroup_parent(cgrp)); |
9d755d33 | 4925 | kernfs_put(cgrp->kn); |
2ce7135a | 4926 | psi_cgroup_free(cgrp); |
041cd640 | 4927 | if (cgroup_on_dfl(cgrp)) |
c58632b3 | 4928 | cgroup_rstat_exit(cgrp); |
9d755d33 TH |
4929 | kfree(cgrp); |
4930 | } else { | |
4931 | /* | |
4932 | * This is root cgroup's refcnt reaching zero, | |
4933 | * which indicates that the root should be | |
4934 | * released. | |
4935 | */ | |
4936 | cgroup_destroy_root(cgrp->root); | |
4937 | } | |
4938 | } | |
48ddbe19 TH |
4939 | } |
4940 | ||
25e15d83 | 4941 | static void css_release_work_fn(struct work_struct *work) |
d3daf28d TH |
4942 | { |
4943 | struct cgroup_subsys_state *css = | |
25e15d83 | 4944 | container_of(work, struct cgroup_subsys_state, destroy_work); |
15a4c835 | 4945 | struct cgroup_subsys *ss = css->ss; |
9d755d33 | 4946 | struct cgroup *cgrp = css->cgroup; |
15a4c835 | 4947 | |
1fed1b2e TH |
4948 | mutex_lock(&cgroup_mutex); |
4949 | ||
de3f0341 | 4950 | css->flags |= CSS_RELEASED; |
1fed1b2e TH |
4951 | list_del_rcu(&css->sibling); |
4952 | ||
9d755d33 TH |
4953 | if (ss) { |
4954 | /* css release path */ | |
8f53470b TH |
4955 | if (!list_empty(&css->rstat_css_node)) { |
4956 | cgroup_rstat_flush(cgrp); | |
4957 | list_del_rcu(&css->rstat_css_node); | |
4958 | } | |
4959 | ||
01e58659 | 4960 | cgroup_idr_replace(&ss->css_idr, NULL, css->id); |
7d172cc8 TH |
4961 | if (ss->css_released) |
4962 | ss->css_released(css); | |
9d755d33 | 4963 | } else { |
0679dee0 RG |
4964 | struct cgroup *tcgrp; |
4965 | ||
9d755d33 | 4966 | /* cgroup release path */ |
e4f8d81c | 4967 | TRACE_CGROUP_PATH(release, cgrp); |
ed1777de | 4968 | |
041cd640 | 4969 | if (cgroup_on_dfl(cgrp)) |
c58632b3 | 4970 | cgroup_rstat_flush(cgrp); |
041cd640 | 4971 | |
4dcabece | 4972 | spin_lock_irq(&css_set_lock); |
0679dee0 RG |
4973 | for (tcgrp = cgroup_parent(cgrp); tcgrp; |
4974 | tcgrp = cgroup_parent(tcgrp)) | |
4975 | tcgrp->nr_dying_descendants--; | |
4dcabece | 4976 | spin_unlock_irq(&css_set_lock); |
0679dee0 | 4977 | |
a4189487 LZ |
4978 | /* |
4979 | * There are two control paths which try to determine | |
4980 | * cgroup from dentry without going through kernfs - | |
4981 | * cgroupstats_build() and css_tryget_online_from_dir(). | |
4982 | * Those are supported by RCU protecting clearing of | |
4983 | * cgrp->kn->priv backpointer. | |
4984 | */ | |
6cd0f5bb TH |
4985 | if (cgrp->kn) |
4986 | RCU_INIT_POINTER(*(void __rcu __force **)&cgrp->kn->priv, | |
4987 | NULL); | |
9d755d33 | 4988 | } |
d3daf28d | 4989 | |
1fed1b2e TH |
4990 | mutex_unlock(&cgroup_mutex); |
4991 | ||
8f36aaec TH |
4992 | INIT_RCU_WORK(&css->destroy_rwork, css_free_rwork_fn); |
4993 | queue_rcu_work(cgroup_destroy_wq, &css->destroy_rwork); | |
d3daf28d TH |
4994 | } |
4995 | ||
d3daf28d TH |
4996 | static void css_release(struct percpu_ref *ref) |
4997 | { | |
4998 | struct cgroup_subsys_state *css = | |
4999 | container_of(ref, struct cgroup_subsys_state, refcnt); | |
5000 | ||
25e15d83 TH |
5001 | INIT_WORK(&css->destroy_work, css_release_work_fn); |
5002 | queue_work(cgroup_destroy_wq, &css->destroy_work); | |
d3daf28d TH |
5003 | } |
5004 | ||
ddfcadab TH |
5005 | static void init_and_link_css(struct cgroup_subsys_state *css, |
5006 | struct cgroup_subsys *ss, struct cgroup *cgrp) | |
ddbcc7e8 | 5007 | { |
0cb51d71 TH |
5008 | lockdep_assert_held(&cgroup_mutex); |
5009 | ||
a590b90d | 5010 | cgroup_get_live(cgrp); |
ddfcadab | 5011 | |
d5c419b6 | 5012 | memset(css, 0, sizeof(*css)); |
bd89aabc | 5013 | css->cgroup = cgrp; |
72c97e54 | 5014 | css->ss = ss; |
8fa3b8d6 | 5015 | css->id = -1; |
d5c419b6 TH |
5016 | INIT_LIST_HEAD(&css->sibling); |
5017 | INIT_LIST_HEAD(&css->children); | |
8f53470b | 5018 | INIT_LIST_HEAD(&css->rstat_css_node); |
0cb51d71 | 5019 | css->serial_nr = css_serial_nr_next++; |
aa226ff4 | 5020 | atomic_set(&css->online_cnt, 0); |
0ae78e0b | 5021 | |
d51f39b0 TH |
5022 | if (cgroup_parent(cgrp)) { |
5023 | css->parent = cgroup_css(cgroup_parent(cgrp), ss); | |
ddfcadab | 5024 | css_get(css->parent); |
ddfcadab | 5025 | } |
48ddbe19 | 5026 | |
8f53470b TH |
5027 | if (cgroup_on_dfl(cgrp) && ss->css_rstat_flush) |
5028 | list_add_rcu(&css->rstat_css_node, &cgrp->rstat_css_list); | |
5029 | ||
ca8bdcaf | 5030 | BUG_ON(cgroup_css(cgrp, ss)); |
ddbcc7e8 PM |
5031 | } |
5032 | ||
2a4ac633 | 5033 | /* invoke ->css_online() on a new CSS and mark it online if successful */ |
623f926b | 5034 | static int online_css(struct cgroup_subsys_state *css) |
a31f2d3f | 5035 | { |
623f926b | 5036 | struct cgroup_subsys *ss = css->ss; |
b1929db4 TH |
5037 | int ret = 0; |
5038 | ||
a31f2d3f TH |
5039 | lockdep_assert_held(&cgroup_mutex); |
5040 | ||
92fb9748 | 5041 | if (ss->css_online) |
eb95419b | 5042 | ret = ss->css_online(css); |
ae7f164a | 5043 | if (!ret) { |
eb95419b | 5044 | css->flags |= CSS_ONLINE; |
aec25020 | 5045 | rcu_assign_pointer(css->cgroup->subsys[ss->id], css); |
aa226ff4 TH |
5046 | |
5047 | atomic_inc(&css->online_cnt); | |
5048 | if (css->parent) | |
5049 | atomic_inc(&css->parent->online_cnt); | |
ae7f164a | 5050 | } |
b1929db4 | 5051 | return ret; |
a31f2d3f TH |
5052 | } |
5053 | ||
2a4ac633 | 5054 | /* if the CSS is online, invoke ->css_offline() on it and mark it offline */ |
623f926b | 5055 | static void offline_css(struct cgroup_subsys_state *css) |
a31f2d3f | 5056 | { |
623f926b | 5057 | struct cgroup_subsys *ss = css->ss; |
a31f2d3f TH |
5058 | |
5059 | lockdep_assert_held(&cgroup_mutex); | |
5060 | ||
5061 | if (!(css->flags & CSS_ONLINE)) | |
5062 | return; | |
5063 | ||
d7eeac19 | 5064 | if (ss->css_offline) |
eb95419b | 5065 | ss->css_offline(css); |
a31f2d3f | 5066 | |
eb95419b | 5067 | css->flags &= ~CSS_ONLINE; |
e3297803 | 5068 | RCU_INIT_POINTER(css->cgroup->subsys[ss->id], NULL); |
f8f22e53 TH |
5069 | |
5070 | wake_up_all(&css->cgroup->offline_waitq); | |
a31f2d3f TH |
5071 | } |
5072 | ||
c81c925a | 5073 | /** |
6cd0f5bb | 5074 | * css_create - create a cgroup_subsys_state |
c81c925a TH |
5075 | * @cgrp: the cgroup new css will be associated with |
5076 | * @ss: the subsys of new css | |
5077 | * | |
5078 | * Create a new css associated with @cgrp - @ss pair. On success, the new | |
6cd0f5bb TH |
5079 | * css is online and installed in @cgrp. This function doesn't create the |
5080 | * interface files. Returns 0 on success, -errno on failure. | |
c81c925a | 5081 | */ |
6cd0f5bb TH |
5082 | static struct cgroup_subsys_state *css_create(struct cgroup *cgrp, |
5083 | struct cgroup_subsys *ss) | |
c81c925a | 5084 | { |
d51f39b0 | 5085 | struct cgroup *parent = cgroup_parent(cgrp); |
1fed1b2e | 5086 | struct cgroup_subsys_state *parent_css = cgroup_css(parent, ss); |
c81c925a TH |
5087 | struct cgroup_subsys_state *css; |
5088 | int err; | |
5089 | ||
c81c925a TH |
5090 | lockdep_assert_held(&cgroup_mutex); |
5091 | ||
1fed1b2e | 5092 | css = ss->css_alloc(parent_css); |
e7e15b87 TH |
5093 | if (!css) |
5094 | css = ERR_PTR(-ENOMEM); | |
c81c925a | 5095 | if (IS_ERR(css)) |
6cd0f5bb | 5096 | return css; |
c81c925a | 5097 | |
ddfcadab | 5098 | init_and_link_css(css, ss, cgrp); |
a2bed820 | 5099 | |
2aad2a86 | 5100 | err = percpu_ref_init(&css->refcnt, css_release, 0, GFP_KERNEL); |
c81c925a | 5101 | if (err) |
3eb59ec6 | 5102 | goto err_free_css; |
c81c925a | 5103 | |
cf780b7d | 5104 | err = cgroup_idr_alloc(&ss->css_idr, NULL, 2, 0, GFP_KERNEL); |
15a4c835 | 5105 | if (err < 0) |
b00c52da | 5106 | goto err_free_css; |
15a4c835 | 5107 | css->id = err; |
c81c925a | 5108 | |
15a4c835 | 5109 | /* @css is ready to be brought online now, make it visible */ |
1fed1b2e | 5110 | list_add_tail_rcu(&css->sibling, &parent_css->children); |
15a4c835 | 5111 | cgroup_idr_replace(&ss->css_idr, css, css->id); |
c81c925a TH |
5112 | |
5113 | err = online_css(css); | |
5114 | if (err) | |
1fed1b2e | 5115 | goto err_list_del; |
94419627 | 5116 | |
c81c925a | 5117 | if (ss->broken_hierarchy && !ss->warned_broken_hierarchy && |
d51f39b0 | 5118 | cgroup_parent(parent)) { |
ed3d261b | 5119 | 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 | 5120 | current->comm, current->pid, ss->name); |
c81c925a | 5121 | if (!strcmp(ss->name, "memory")) |
ed3d261b | 5122 | pr_warn("\"memory\" requires setting use_hierarchy to 1 on the root\n"); |
c81c925a TH |
5123 | ss->warned_broken_hierarchy = true; |
5124 | } | |
5125 | ||
6cd0f5bb | 5126 | return css; |
c81c925a | 5127 | |
1fed1b2e TH |
5128 | err_list_del: |
5129 | list_del_rcu(&css->sibling); | |
3eb59ec6 | 5130 | err_free_css: |
8f53470b | 5131 | list_del_rcu(&css->rstat_css_node); |
8f36aaec TH |
5132 | INIT_RCU_WORK(&css->destroy_rwork, css_free_rwork_fn); |
5133 | queue_rcu_work(cgroup_destroy_wq, &css->destroy_rwork); | |
6cd0f5bb | 5134 | return ERR_PTR(err); |
c81c925a TH |
5135 | } |
5136 | ||
07cd1294 TH |
5137 | /* |
5138 | * The returned cgroup is fully initialized including its control mask, but | |
5139 | * it isn't associated with its kernfs_node and doesn't have the control | |
5140 | * mask applied. | |
5141 | */ | |
74321038 TH |
5142 | static struct cgroup *cgroup_create(struct cgroup *parent, const char *name, |
5143 | umode_t mode) | |
ddbcc7e8 | 5144 | { |
a5bca215 | 5145 | struct cgroup_root *root = parent->root; |
a5bca215 | 5146 | struct cgroup *cgrp, *tcgrp; |
74321038 | 5147 | struct kernfs_node *kn; |
a5bca215 | 5148 | int level = parent->level + 1; |
03970d3c | 5149 | int ret; |
ddbcc7e8 | 5150 | |
0a950f65 | 5151 | /* allocate the cgroup and its ID, 0 is reserved for the root */ |
acafe7e3 KC |
5152 | cgrp = kzalloc(struct_size(cgrp, ancestor_ids, (level + 1)), |
5153 | GFP_KERNEL); | |
a5bca215 TH |
5154 | if (!cgrp) |
5155 | return ERR_PTR(-ENOMEM); | |
0ab02ca8 | 5156 | |
2aad2a86 | 5157 | ret = percpu_ref_init(&cgrp->self.refcnt, css_release, 0, GFP_KERNEL); |
9d755d33 TH |
5158 | if (ret) |
5159 | goto out_free_cgrp; | |
5160 | ||
041cd640 | 5161 | if (cgroup_on_dfl(parent)) { |
c58632b3 | 5162 | ret = cgroup_rstat_init(cgrp); |
041cd640 TH |
5163 | if (ret) |
5164 | goto out_cancel_ref; | |
5165 | } | |
5166 | ||
74321038 TH |
5167 | /* create the directory */ |
5168 | kn = kernfs_create_dir(parent->kn, name, mode, cgrp); | |
5169 | if (IS_ERR(kn)) { | |
5170 | ret = PTR_ERR(kn); | |
041cd640 | 5171 | goto out_stat_exit; |
976c06bc | 5172 | } |
74321038 | 5173 | cgrp->kn = kn; |
976c06bc | 5174 | |
cc31edce | 5175 | init_cgroup_housekeeping(cgrp); |
ddbcc7e8 | 5176 | |
9d800df1 | 5177 | cgrp->self.parent = &parent->self; |
ba0f4d76 | 5178 | cgrp->root = root; |
b11cfb58 | 5179 | cgrp->level = level; |
2ce7135a JW |
5180 | |
5181 | ret = psi_cgroup_alloc(cgrp); | |
324bda9e | 5182 | if (ret) |
74321038 | 5183 | goto out_kernfs_remove; |
b11cfb58 | 5184 | |
2ce7135a JW |
5185 | ret = cgroup_bpf_inherit(cgrp); |
5186 | if (ret) | |
5187 | goto out_psi_free; | |
5188 | ||
76f969e8 RG |
5189 | /* |
5190 | * New cgroup inherits effective freeze counter, and | |
5191 | * if the parent has to be frozen, the child has too. | |
5192 | */ | |
5193 | cgrp->freezer.e_freeze = parent->freezer.e_freeze; | |
97a61369 RG |
5194 | if (cgrp->freezer.e_freeze) { |
5195 | /* | |
5196 | * Set the CGRP_FREEZE flag, so when a process will be | |
5197 | * attached to the child cgroup, it will become frozen. | |
5198 | * At this point the new cgroup is unpopulated, so we can | |
5199 | * consider it frozen immediately. | |
5200 | */ | |
5201 | set_bit(CGRP_FREEZE, &cgrp->flags); | |
76f969e8 | 5202 | set_bit(CGRP_FROZEN, &cgrp->flags); |
97a61369 | 5203 | } |
76f969e8 | 5204 | |
4dcabece | 5205 | spin_lock_irq(&css_set_lock); |
0679dee0 | 5206 | for (tcgrp = cgrp; tcgrp; tcgrp = cgroup_parent(tcgrp)) { |
74321038 | 5207 | cgrp->ancestor_ids[tcgrp->level] = cgroup_id(tcgrp); |
ddbcc7e8 | 5208 | |
76f969e8 | 5209 | if (tcgrp != cgrp) { |
0679dee0 | 5210 | tcgrp->nr_descendants++; |
76f969e8 RG |
5211 | |
5212 | /* | |
5213 | * If the new cgroup is frozen, all ancestor cgroups | |
5214 | * get a new frozen descendant, but their state can't | |
5215 | * change because of this. | |
5216 | */ | |
5217 | if (cgrp->freezer.e_freeze) | |
5218 | tcgrp->freezer.nr_frozen_descendants++; | |
5219 | } | |
0679dee0 | 5220 | } |
4dcabece | 5221 | spin_unlock_irq(&css_set_lock); |
0679dee0 | 5222 | |
b6abdb0e LZ |
5223 | if (notify_on_release(parent)) |
5224 | set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); | |
5225 | ||
2260e7fc TH |
5226 | if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &parent->flags)) |
5227 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags); | |
97978e6d | 5228 | |
0cb51d71 | 5229 | cgrp->self.serial_nr = css_serial_nr_next++; |
53fa5261 | 5230 | |
4e139afc | 5231 | /* allocation complete, commit to creation */ |
d5c419b6 | 5232 | list_add_tail_rcu(&cgrp->self.sibling, &cgroup_parent(cgrp)->self.children); |
3c9c825b | 5233 | atomic_inc(&root->nr_cgrps); |
a590b90d | 5234 | cgroup_get_live(parent); |
415cf07a | 5235 | |
bd53d617 TH |
5236 | /* |
5237 | * On the default hierarchy, a child doesn't automatically inherit | |
667c2491 | 5238 | * subtree_control from the parent. Each is configured manually. |
bd53d617 | 5239 | */ |
03970d3c | 5240 | if (!cgroup_on_dfl(cgrp)) |
5531dc91 | 5241 | cgrp->subtree_control = cgroup_control(cgrp); |
03970d3c TH |
5242 | |
5243 | cgroup_propagate_control(cgrp); | |
5244 | ||
a5bca215 TH |
5245 | return cgrp; |
5246 | ||
2ce7135a JW |
5247 | out_psi_free: |
5248 | psi_cgroup_free(cgrp); | |
74321038 TH |
5249 | out_kernfs_remove: |
5250 | kernfs_remove(cgrp->kn); | |
041cd640 TH |
5251 | out_stat_exit: |
5252 | if (cgroup_on_dfl(parent)) | |
c58632b3 | 5253 | cgroup_rstat_exit(cgrp); |
a5bca215 TH |
5254 | out_cancel_ref: |
5255 | percpu_ref_exit(&cgrp->self.refcnt); | |
5256 | out_free_cgrp: | |
5257 | kfree(cgrp); | |
5258 | return ERR_PTR(ret); | |
a5bca215 TH |
5259 | } |
5260 | ||
1a926e0b RG |
5261 | static bool cgroup_check_hierarchy_limits(struct cgroup *parent) |
5262 | { | |
5263 | struct cgroup *cgroup; | |
5264 | int ret = false; | |
5265 | int level = 1; | |
5266 | ||
5267 | lockdep_assert_held(&cgroup_mutex); | |
5268 | ||
5269 | for (cgroup = parent; cgroup; cgroup = cgroup_parent(cgroup)) { | |
5270 | if (cgroup->nr_descendants >= cgroup->max_descendants) | |
5271 | goto fail; | |
5272 | ||
5273 | if (level > cgroup->max_depth) | |
5274 | goto fail; | |
5275 | ||
5276 | level++; | |
5277 | } | |
5278 | ||
5279 | ret = true; | |
5280 | fail: | |
5281 | return ret; | |
5282 | } | |
5283 | ||
1592c9b2 | 5284 | int cgroup_mkdir(struct kernfs_node *parent_kn, const char *name, umode_t mode) |
a5bca215 TH |
5285 | { |
5286 | struct cgroup *parent, *cgrp; | |
03970d3c | 5287 | int ret; |
a5bca215 TH |
5288 | |
5289 | /* do not accept '\n' to prevent making /proc/<pid>/cgroup unparsable */ | |
5290 | if (strchr(name, '\n')) | |
5291 | return -EINVAL; | |
5292 | ||
945ba199 | 5293 | parent = cgroup_kn_lock_live(parent_kn, false); |
a5bca215 TH |
5294 | if (!parent) |
5295 | return -ENODEV; | |
5296 | ||
1a926e0b RG |
5297 | if (!cgroup_check_hierarchy_limits(parent)) { |
5298 | ret = -EAGAIN; | |
5299 | goto out_unlock; | |
5300 | } | |
5301 | ||
74321038 | 5302 | cgrp = cgroup_create(parent, name, mode); |
a5bca215 TH |
5303 | if (IS_ERR(cgrp)) { |
5304 | ret = PTR_ERR(cgrp); | |
5305 | goto out_unlock; | |
5306 | } | |
5307 | ||
195e9b6c TH |
5308 | /* |
5309 | * This extra ref will be put in cgroup_free_fn() and guarantees | |
5310 | * that @cgrp->kn is always accessible. | |
5311 | */ | |
74321038 | 5312 | kernfs_get(cgrp->kn); |
195e9b6c | 5313 | |
74321038 | 5314 | ret = cgroup_kn_set_ugid(cgrp->kn); |
195e9b6c TH |
5315 | if (ret) |
5316 | goto out_destroy; | |
5317 | ||
334c3679 | 5318 | ret = css_populate_dir(&cgrp->self); |
195e9b6c TH |
5319 | if (ret) |
5320 | goto out_destroy; | |
5321 | ||
03970d3c TH |
5322 | ret = cgroup_apply_control_enable(cgrp); |
5323 | if (ret) | |
5324 | goto out_destroy; | |
195e9b6c | 5325 | |
e4f8d81c | 5326 | TRACE_CGROUP_PATH(mkdir, cgrp); |
ed1777de | 5327 | |
195e9b6c | 5328 | /* let's create and online css's */ |
74321038 | 5329 | kernfs_activate(cgrp->kn); |
ddbcc7e8 | 5330 | |
ba0f4d76 TH |
5331 | ret = 0; |
5332 | goto out_unlock; | |
ddbcc7e8 | 5333 | |
a5bca215 TH |
5334 | out_destroy: |
5335 | cgroup_destroy_locked(cgrp); | |
ba0f4d76 | 5336 | out_unlock: |
a9746d8d | 5337 | cgroup_kn_unlock(parent_kn); |
ba0f4d76 | 5338 | return ret; |
ddbcc7e8 PM |
5339 | } |
5340 | ||
223dbc38 TH |
5341 | /* |
5342 | * This is called when the refcnt of a css is confirmed to be killed. | |
249f3468 TH |
5343 | * css_tryget_online() is now guaranteed to fail. Tell the subsystem to |
5344 | * initate destruction and put the css ref from kill_css(). | |
223dbc38 TH |
5345 | */ |
5346 | static void css_killed_work_fn(struct work_struct *work) | |
d3daf28d | 5347 | { |
223dbc38 TH |
5348 | struct cgroup_subsys_state *css = |
5349 | container_of(work, struct cgroup_subsys_state, destroy_work); | |
d3daf28d | 5350 | |
f20104de | 5351 | mutex_lock(&cgroup_mutex); |
09a503ea | 5352 | |
aa226ff4 TH |
5353 | do { |
5354 | offline_css(css); | |
5355 | css_put(css); | |
5356 | /* @css can't go away while we're holding cgroup_mutex */ | |
5357 | css = css->parent; | |
5358 | } while (css && atomic_dec_and_test(&css->online_cnt)); | |
5359 | ||
5360 | mutex_unlock(&cgroup_mutex); | |
d3daf28d TH |
5361 | } |
5362 | ||
223dbc38 TH |
5363 | /* css kill confirmation processing requires process context, bounce */ |
5364 | static void css_killed_ref_fn(struct percpu_ref *ref) | |
d3daf28d TH |
5365 | { |
5366 | struct cgroup_subsys_state *css = | |
5367 | container_of(ref, struct cgroup_subsys_state, refcnt); | |
5368 | ||
aa226ff4 TH |
5369 | if (atomic_dec_and_test(&css->online_cnt)) { |
5370 | INIT_WORK(&css->destroy_work, css_killed_work_fn); | |
5371 | queue_work(cgroup_destroy_wq, &css->destroy_work); | |
5372 | } | |
d3daf28d TH |
5373 | } |
5374 | ||
f392e51c TH |
5375 | /** |
5376 | * kill_css - destroy a css | |
5377 | * @css: css to destroy | |
5378 | * | |
5379 | * This function initiates destruction of @css by removing cgroup interface | |
5380 | * files and putting its base reference. ->css_offline() will be invoked | |
ec903c0c TH |
5381 | * asynchronously once css_tryget_online() is guaranteed to fail and when |
5382 | * the reference count reaches zero, @css will be released. | |
f392e51c TH |
5383 | */ |
5384 | static void kill_css(struct cgroup_subsys_state *css) | |
edae0c33 | 5385 | { |
01f6474c | 5386 | lockdep_assert_held(&cgroup_mutex); |
94419627 | 5387 | |
33c35aa4 WL |
5388 | if (css->flags & CSS_DYING) |
5389 | return; | |
5390 | ||
5391 | css->flags |= CSS_DYING; | |
5392 | ||
2bd59d48 TH |
5393 | /* |
5394 | * This must happen before css is disassociated with its cgroup. | |
5395 | * See seq_css() for details. | |
5396 | */ | |
334c3679 | 5397 | css_clear_dir(css); |
3c14f8b4 | 5398 | |
edae0c33 TH |
5399 | /* |
5400 | * Killing would put the base ref, but we need to keep it alive | |
5401 | * until after ->css_offline(). | |
5402 | */ | |
5403 | css_get(css); | |
5404 | ||
5405 | /* | |
5406 | * cgroup core guarantees that, by the time ->css_offline() is | |
5407 | * invoked, no new css reference will be given out via | |
ec903c0c | 5408 | * css_tryget_online(). We can't simply call percpu_ref_kill() and |
edae0c33 TH |
5409 | * proceed to offlining css's because percpu_ref_kill() doesn't |
5410 | * guarantee that the ref is seen as killed on all CPUs on return. | |
5411 | * | |
5412 | * Use percpu_ref_kill_and_confirm() to get notifications as each | |
5413 | * css is confirmed to be seen as killed on all CPUs. | |
5414 | */ | |
5415 | percpu_ref_kill_and_confirm(&css->refcnt, css_killed_ref_fn); | |
d3daf28d TH |
5416 | } |
5417 | ||
5418 | /** | |
5419 | * cgroup_destroy_locked - the first stage of cgroup destruction | |
5420 | * @cgrp: cgroup to be destroyed | |
5421 | * | |
5422 | * css's make use of percpu refcnts whose killing latency shouldn't be | |
5423 | * exposed to userland and are RCU protected. Also, cgroup core needs to | |
ec903c0c TH |
5424 | * guarantee that css_tryget_online() won't succeed by the time |
5425 | * ->css_offline() is invoked. To satisfy all the requirements, | |
5426 | * destruction is implemented in the following two steps. | |
d3daf28d TH |
5427 | * |
5428 | * s1. Verify @cgrp can be destroyed and mark it dying. Remove all | |
5429 | * userland visible parts and start killing the percpu refcnts of | |
5430 | * css's. Set up so that the next stage will be kicked off once all | |
5431 | * the percpu refcnts are confirmed to be killed. | |
5432 | * | |
5433 | * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the | |
5434 | * rest of destruction. Once all cgroup references are gone, the | |
5435 | * cgroup is RCU-freed. | |
5436 | * | |
5437 | * This function implements s1. After this step, @cgrp is gone as far as | |
5438 | * the userland is concerned and a new cgroup with the same name may be | |
5439 | * created. As cgroup doesn't care about the names internally, this | |
5440 | * doesn't cause any problem. | |
5441 | */ | |
42809dd4 TH |
5442 | static int cgroup_destroy_locked(struct cgroup *cgrp) |
5443 | __releases(&cgroup_mutex) __acquires(&cgroup_mutex) | |
ddbcc7e8 | 5444 | { |
0679dee0 | 5445 | struct cgroup *tcgrp, *parent = cgroup_parent(cgrp); |
2bd59d48 | 5446 | struct cgroup_subsys_state *css; |
2b021cbf | 5447 | struct cgrp_cset_link *link; |
1c6727af | 5448 | int ssid; |
ddbcc7e8 | 5449 | |
42809dd4 TH |
5450 | lockdep_assert_held(&cgroup_mutex); |
5451 | ||
91486f61 TH |
5452 | /* |
5453 | * Only migration can raise populated from zero and we're already | |
5454 | * holding cgroup_mutex. | |
5455 | */ | |
5456 | if (cgroup_is_populated(cgrp)) | |
ddbcc7e8 | 5457 | return -EBUSY; |
a043e3b2 | 5458 | |
bb78a92f | 5459 | /* |
d5c419b6 TH |
5460 | * Make sure there's no live children. We can't test emptiness of |
5461 | * ->self.children as dead children linger on it while being | |
5462 | * drained; otherwise, "rmdir parent/child parent" may fail. | |
bb78a92f | 5463 | */ |
f3d46500 | 5464 | if (css_has_online_children(&cgrp->self)) |
bb78a92f HD |
5465 | return -EBUSY; |
5466 | ||
455050d2 | 5467 | /* |
2b021cbf TH |
5468 | * Mark @cgrp and the associated csets dead. The former prevents |
5469 | * further task migration and child creation by disabling | |
5470 | * cgroup_lock_live_group(). The latter makes the csets ignored by | |
5471 | * the migration path. | |
455050d2 | 5472 | */ |
184faf32 | 5473 | cgrp->self.flags &= ~CSS_ONLINE; |
ddbcc7e8 | 5474 | |
82d6489d | 5475 | spin_lock_irq(&css_set_lock); |
2b021cbf TH |
5476 | list_for_each_entry(link, &cgrp->cset_links, cset_link) |
5477 | link->cset->dead = true; | |
82d6489d | 5478 | spin_unlock_irq(&css_set_lock); |
2b021cbf | 5479 | |
249f3468 | 5480 | /* initiate massacre of all css's */ |
1c6727af TH |
5481 | for_each_css(css, ssid, cgrp) |
5482 | kill_css(css); | |
455050d2 | 5483 | |
5faaf05f TH |
5484 | /* clear and remove @cgrp dir, @cgrp has an extra ref on its kn */ |
5485 | css_clear_dir(&cgrp->self); | |
01f6474c | 5486 | kernfs_remove(cgrp->kn); |
f20104de | 5487 | |
454000ad TH |
5488 | if (parent && cgroup_is_threaded(cgrp)) |
5489 | parent->nr_threaded_children--; | |
5490 | ||
4dcabece | 5491 | spin_lock_irq(&css_set_lock); |
0679dee0 RG |
5492 | for (tcgrp = cgroup_parent(cgrp); tcgrp; tcgrp = cgroup_parent(tcgrp)) { |
5493 | tcgrp->nr_descendants--; | |
5494 | tcgrp->nr_dying_descendants++; | |
76f969e8 RG |
5495 | /* |
5496 | * If the dying cgroup is frozen, decrease frozen descendants | |
5497 | * counters of ancestor cgroups. | |
5498 | */ | |
5499 | if (test_bit(CGRP_FROZEN, &cgrp->flags)) | |
5500 | tcgrp->freezer.nr_frozen_descendants--; | |
0679dee0 | 5501 | } |
4dcabece | 5502 | spin_unlock_irq(&css_set_lock); |
0679dee0 | 5503 | |
5a621e6c | 5504 | cgroup1_check_for_release(parent); |
2bd59d48 | 5505 | |
4bfc0bb2 RG |
5506 | cgroup_bpf_offline(cgrp); |
5507 | ||
249f3468 | 5508 | /* put the base reference */ |
9d755d33 | 5509 | percpu_ref_kill(&cgrp->self.refcnt); |
455050d2 | 5510 | |
ea15f8cc TH |
5511 | return 0; |
5512 | }; | |
5513 | ||
1592c9b2 | 5514 | int cgroup_rmdir(struct kernfs_node *kn) |
42809dd4 | 5515 | { |
a9746d8d | 5516 | struct cgroup *cgrp; |
2bd59d48 | 5517 | int ret = 0; |
42809dd4 | 5518 | |
945ba199 | 5519 | cgrp = cgroup_kn_lock_live(kn, false); |
a9746d8d TH |
5520 | if (!cgrp) |
5521 | return 0; | |
42809dd4 | 5522 | |
a9746d8d | 5523 | ret = cgroup_destroy_locked(cgrp); |
ed1777de | 5524 | if (!ret) |
e4f8d81c | 5525 | TRACE_CGROUP_PATH(rmdir, cgrp); |
ed1777de | 5526 | |
a9746d8d | 5527 | cgroup_kn_unlock(kn); |
42809dd4 | 5528 | return ret; |
8e3f6541 TH |
5529 | } |
5530 | ||
2bd59d48 | 5531 | static struct kernfs_syscall_ops cgroup_kf_syscall_ops = { |
5136f636 | 5532 | .show_options = cgroup_show_options, |
2bd59d48 TH |
5533 | .mkdir = cgroup_mkdir, |
5534 | .rmdir = cgroup_rmdir, | |
4f41fc59 | 5535 | .show_path = cgroup_show_path, |
2bd59d48 TH |
5536 | }; |
5537 | ||
15a4c835 | 5538 | static void __init cgroup_init_subsys(struct cgroup_subsys *ss, bool early) |
ddbcc7e8 | 5539 | { |
ddbcc7e8 | 5540 | struct cgroup_subsys_state *css; |
cfe36bde | 5541 | |
a5ae9899 | 5542 | pr_debug("Initializing cgroup subsys %s\n", ss->name); |
ddbcc7e8 | 5543 | |
648bb56d TH |
5544 | mutex_lock(&cgroup_mutex); |
5545 | ||
15a4c835 | 5546 | idr_init(&ss->css_idr); |
0adb0704 | 5547 | INIT_LIST_HEAD(&ss->cfts); |
8e3f6541 | 5548 | |
3dd06ffa TH |
5549 | /* Create the root cgroup state for this subsystem */ |
5550 | ss->root = &cgrp_dfl_root; | |
5551 | css = ss->css_alloc(cgroup_css(&cgrp_dfl_root.cgrp, ss)); | |
ddbcc7e8 PM |
5552 | /* We don't handle early failures gracefully */ |
5553 | BUG_ON(IS_ERR(css)); | |
ddfcadab | 5554 | init_and_link_css(css, ss, &cgrp_dfl_root.cgrp); |
3b514d24 TH |
5555 | |
5556 | /* | |
5557 | * Root csses are never destroyed and we can't initialize | |
5558 | * percpu_ref during early init. Disable refcnting. | |
5559 | */ | |
5560 | css->flags |= CSS_NO_REF; | |
5561 | ||
15a4c835 | 5562 | if (early) { |
9395a450 | 5563 | /* allocation can't be done safely during early init */ |
15a4c835 TH |
5564 | css->id = 1; |
5565 | } else { | |
5566 | css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2, GFP_KERNEL); | |
5567 | BUG_ON(css->id < 0); | |
5568 | } | |
ddbcc7e8 | 5569 | |
e8d55fde | 5570 | /* Update the init_css_set to contain a subsys |
817929ec | 5571 | * pointer to this state - since the subsystem is |
e8d55fde | 5572 | * newly registered, all tasks and hence the |
3dd06ffa | 5573 | * init_css_set is in the subsystem's root cgroup. */ |
aec25020 | 5574 | init_css_set.subsys[ss->id] = css; |
ddbcc7e8 | 5575 | |
cb4a3167 AS |
5576 | have_fork_callback |= (bool)ss->fork << ss->id; |
5577 | have_exit_callback |= (bool)ss->exit << ss->id; | |
51bee5ab | 5578 | have_release_callback |= (bool)ss->release << ss->id; |
7e47682e | 5579 | have_canfork_callback |= (bool)ss->can_fork << ss->id; |
ddbcc7e8 | 5580 | |
e8d55fde LZ |
5581 | /* At system boot, before all subsystems have been |
5582 | * registered, no tasks have been forked, so we don't | |
5583 | * need to invoke fork callbacks here. */ | |
5584 | BUG_ON(!list_empty(&init_task.tasks)); | |
5585 | ||
ae7f164a | 5586 | BUG_ON(online_css(css)); |
a8638030 | 5587 | |
cf5d5941 BB |
5588 | mutex_unlock(&cgroup_mutex); |
5589 | } | |
cf5d5941 | 5590 | |
ddbcc7e8 | 5591 | /** |
a043e3b2 LZ |
5592 | * cgroup_init_early - cgroup initialization at system boot |
5593 | * | |
5594 | * Initialize cgroups at system boot, and initialize any | |
5595 | * subsystems that request early init. | |
ddbcc7e8 PM |
5596 | */ |
5597 | int __init cgroup_init_early(void) | |
5598 | { | |
f5dfb531 | 5599 | static struct cgroup_fs_context __initdata ctx; |
30159ec7 | 5600 | struct cgroup_subsys *ss; |
ddbcc7e8 | 5601 | int i; |
30159ec7 | 5602 | |
cf6299b1 AV |
5603 | ctx.root = &cgrp_dfl_root; |
5604 | init_cgroup_root(&ctx); | |
3b514d24 TH |
5605 | cgrp_dfl_root.cgrp.self.flags |= CSS_NO_REF; |
5606 | ||
a4ea1cc9 | 5607 | RCU_INIT_POINTER(init_task.cgroups, &init_css_set); |
817929ec | 5608 | |
3ed80a62 | 5609 | for_each_subsys(ss, i) { |
aec25020 | 5610 | WARN(!ss->css_alloc || !ss->css_free || ss->name || ss->id, |
63253ad8 | 5611 | "invalid cgroup_subsys %d:%s css_alloc=%p css_free=%p id:name=%d:%s\n", |
073219e9 | 5612 | i, cgroup_subsys_name[i], ss->css_alloc, ss->css_free, |
aec25020 | 5613 | ss->id, ss->name); |
073219e9 TH |
5614 | WARN(strlen(cgroup_subsys_name[i]) > MAX_CGROUP_TYPE_NAMELEN, |
5615 | "cgroup_subsys_name %s too long\n", cgroup_subsys_name[i]); | |
5616 | ||
aec25020 | 5617 | ss->id = i; |
073219e9 | 5618 | ss->name = cgroup_subsys_name[i]; |
3e1d2eed TH |
5619 | if (!ss->legacy_name) |
5620 | ss->legacy_name = cgroup_subsys_name[i]; | |
ddbcc7e8 PM |
5621 | |
5622 | if (ss->early_init) | |
15a4c835 | 5623 | cgroup_init_subsys(ss, true); |
ddbcc7e8 PM |
5624 | } |
5625 | return 0; | |
5626 | } | |
5627 | ||
6e5c8307 | 5628 | static u16 cgroup_disable_mask __initdata; |
a3e72739 | 5629 | |
ddbcc7e8 | 5630 | /** |
a043e3b2 LZ |
5631 | * cgroup_init - cgroup initialization |
5632 | * | |
5633 | * Register cgroup filesystem and /proc file, and initialize | |
5634 | * any subsystems that didn't request early init. | |
ddbcc7e8 PM |
5635 | */ |
5636 | int __init cgroup_init(void) | |
5637 | { | |
30159ec7 | 5638 | struct cgroup_subsys *ss; |
035f4f51 | 5639 | int ssid; |
ddbcc7e8 | 5640 | |
6e5c8307 | 5641 | BUILD_BUG_ON(CGROUP_SUBSYS_COUNT > 16); |
d62beb7f TH |
5642 | BUG_ON(cgroup_init_cftypes(NULL, cgroup_base_files)); |
5643 | BUG_ON(cgroup_init_cftypes(NULL, cgroup1_base_files)); | |
ddbcc7e8 | 5644 | |
c58632b3 | 5645 | cgroup_rstat_boot(); |
041cd640 | 5646 | |
3942a9bd | 5647 | /* |
2af3024c | 5648 | * The latency of the synchronize_rcu() is too high for cgroups, |
3942a9bd PZ |
5649 | * avoid it at the cost of forcing all readers into the slow path. |
5650 | */ | |
5651 | rcu_sync_enter_start(&cgroup_threadgroup_rwsem.rss); | |
5652 | ||
a79a908f AK |
5653 | get_user_ns(init_cgroup_ns.user_ns); |
5654 | ||
54e7b4eb | 5655 | mutex_lock(&cgroup_mutex); |
54e7b4eb | 5656 | |
2378d8b8 TH |
5657 | /* |
5658 | * Add init_css_set to the hash table so that dfl_root can link to | |
5659 | * it during init. | |
5660 | */ | |
5661 | hash_add(css_set_table, &init_css_set.hlist, | |
5662 | css_set_hash(init_css_set.subsys)); | |
82fe9b0d | 5663 | |
35ac1184 | 5664 | BUG_ON(cgroup_setup_root(&cgrp_dfl_root, 0)); |
4e96ee8e | 5665 | |
54e7b4eb TH |
5666 | mutex_unlock(&cgroup_mutex); |
5667 | ||
172a2c06 | 5668 | for_each_subsys(ss, ssid) { |
15a4c835 TH |
5669 | if (ss->early_init) { |
5670 | struct cgroup_subsys_state *css = | |
5671 | init_css_set.subsys[ss->id]; | |
5672 | ||
5673 | css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2, | |
5674 | GFP_KERNEL); | |
5675 | BUG_ON(css->id < 0); | |
5676 | } else { | |
5677 | cgroup_init_subsys(ss, false); | |
5678 | } | |
172a2c06 | 5679 | |
2d8f243a TH |
5680 | list_add_tail(&init_css_set.e_cset_node[ssid], |
5681 | &cgrp_dfl_root.cgrp.e_csets[ssid]); | |
172a2c06 TH |
5682 | |
5683 | /* | |
c731ae1d LZ |
5684 | * Setting dfl_root subsys_mask needs to consider the |
5685 | * disabled flag and cftype registration needs kmalloc, | |
5686 | * both of which aren't available during early_init. | |
172a2c06 | 5687 | */ |
a3e72739 TH |
5688 | if (cgroup_disable_mask & (1 << ssid)) { |
5689 | static_branch_disable(cgroup_subsys_enabled_key[ssid]); | |
5690 | printk(KERN_INFO "Disabling %s control group subsystem\n", | |
5691 | ss->name); | |
a8ddc821 | 5692 | continue; |
a3e72739 | 5693 | } |
a8ddc821 | 5694 | |
d62beb7f | 5695 | if (cgroup1_ssid_disabled(ssid)) |
223ffb29 JW |
5696 | printk(KERN_INFO "Disabling %s control group subsystem in v1 mounts\n", |
5697 | ss->name); | |
5698 | ||
a8ddc821 TH |
5699 | cgrp_dfl_root.subsys_mask |= 1 << ss->id; |
5700 | ||
8cfd8147 TH |
5701 | /* implicit controllers must be threaded too */ |
5702 | WARN_ON(ss->implicit_on_dfl && !ss->threaded); | |
5703 | ||
f6d635ad TH |
5704 | if (ss->implicit_on_dfl) |
5705 | cgrp_dfl_implicit_ss_mask |= 1 << ss->id; | |
5706 | else if (!ss->dfl_cftypes) | |
a7165264 | 5707 | cgrp_dfl_inhibit_ss_mask |= 1 << ss->id; |
5de4fa13 | 5708 | |
8cfd8147 TH |
5709 | if (ss->threaded) |
5710 | cgrp_dfl_threaded_ss_mask |= 1 << ss->id; | |
5711 | ||
a8ddc821 TH |
5712 | if (ss->dfl_cftypes == ss->legacy_cftypes) { |
5713 | WARN_ON(cgroup_add_cftypes(ss, ss->dfl_cftypes)); | |
5714 | } else { | |
5715 | WARN_ON(cgroup_add_dfl_cftypes(ss, ss->dfl_cftypes)); | |
5716 | WARN_ON(cgroup_add_legacy_cftypes(ss, ss->legacy_cftypes)); | |
c731ae1d | 5717 | } |
295458e6 VD |
5718 | |
5719 | if (ss->bind) | |
5720 | ss->bind(init_css_set.subsys[ssid]); | |
7af608e4 TH |
5721 | |
5722 | mutex_lock(&cgroup_mutex); | |
5723 | css_populate_dir(init_css_set.subsys[ssid]); | |
5724 | mutex_unlock(&cgroup_mutex); | |
676db4af GK |
5725 | } |
5726 | ||
2378d8b8 TH |
5727 | /* init_css_set.subsys[] has been updated, re-hash */ |
5728 | hash_del(&init_css_set.hlist); | |
5729 | hash_add(css_set_table, &init_css_set.hlist, | |
5730 | css_set_hash(init_css_set.subsys)); | |
5731 | ||
035f4f51 TH |
5732 | WARN_ON(sysfs_create_mount_point(fs_kobj, "cgroup")); |
5733 | WARN_ON(register_filesystem(&cgroup_fs_type)); | |
67e9c74b | 5734 | WARN_ON(register_filesystem(&cgroup2_fs_type)); |
3f3942ac | 5735 | WARN_ON(!proc_create_single("cgroups", 0, NULL, proc_cgroupstats_show)); |
d5f68d33 AV |
5736 | #ifdef CONFIG_CPUSETS |
5737 | WARN_ON(register_filesystem(&cpuset_fs_type)); | |
5738 | #endif | |
ddbcc7e8 | 5739 | |
2bd59d48 | 5740 | return 0; |
ddbcc7e8 | 5741 | } |
b4f48b63 | 5742 | |
e5fca243 TH |
5743 | static int __init cgroup_wq_init(void) |
5744 | { | |
5745 | /* | |
5746 | * There isn't much point in executing destruction path in | |
5747 | * parallel. Good chunk is serialized with cgroup_mutex anyway. | |
1a11533f | 5748 | * Use 1 for @max_active. |
e5fca243 TH |
5749 | * |
5750 | * We would prefer to do this in cgroup_init() above, but that | |
5751 | * is called before init_workqueues(): so leave this until after. | |
5752 | */ | |
1a11533f | 5753 | cgroup_destroy_wq = alloc_workqueue("cgroup_destroy", 0, 1); |
e5fca243 TH |
5754 | BUG_ON(!cgroup_destroy_wq); |
5755 | return 0; | |
5756 | } | |
5757 | core_initcall(cgroup_wq_init); | |
5758 | ||
67c0496e | 5759 | void cgroup_path_from_kernfs_id(u64 id, char *buf, size_t buflen) |
69fd5c39 SL |
5760 | { |
5761 | struct kernfs_node *kn; | |
5762 | ||
fe0f726c | 5763 | kn = kernfs_find_and_get_node_by_id(cgrp_dfl_root.kf_root, id); |
69fd5c39 SL |
5764 | if (!kn) |
5765 | return; | |
5766 | kernfs_path(kn, buf, buflen); | |
5767 | kernfs_put(kn); | |
5768 | } | |
5769 | ||
a424316c PM |
5770 | /* |
5771 | * proc_cgroup_show() | |
5772 | * - Print task's cgroup paths into seq_file, one line for each hierarchy | |
5773 | * - Used for /proc/<pid>/cgroup. | |
a424316c | 5774 | */ |
006f4ac4 ZL |
5775 | int proc_cgroup_show(struct seq_file *m, struct pid_namespace *ns, |
5776 | struct pid *pid, struct task_struct *tsk) | |
a424316c | 5777 | { |
4c737b41 | 5778 | char *buf; |
a424316c | 5779 | int retval; |
3dd06ffa | 5780 | struct cgroup_root *root; |
a424316c PM |
5781 | |
5782 | retval = -ENOMEM; | |
e61734c5 | 5783 | buf = kmalloc(PATH_MAX, GFP_KERNEL); |
a424316c PM |
5784 | if (!buf) |
5785 | goto out; | |
5786 | ||
a424316c | 5787 | mutex_lock(&cgroup_mutex); |
82d6489d | 5788 | spin_lock_irq(&css_set_lock); |
a424316c | 5789 | |
985ed670 | 5790 | for_each_root(root) { |
a424316c | 5791 | struct cgroup_subsys *ss; |
bd89aabc | 5792 | struct cgroup *cgrp; |
b85d2040 | 5793 | int ssid, count = 0; |
a424316c | 5794 | |
a7165264 | 5795 | if (root == &cgrp_dfl_root && !cgrp_dfl_visible) |
985ed670 TH |
5796 | continue; |
5797 | ||
2c6ab6d2 | 5798 | seq_printf(m, "%d:", root->hierarchy_id); |
d98817d4 TH |
5799 | if (root != &cgrp_dfl_root) |
5800 | for_each_subsys(ss, ssid) | |
5801 | if (root->subsys_mask & (1 << ssid)) | |
5802 | seq_printf(m, "%s%s", count++ ? "," : "", | |
3e1d2eed | 5803 | ss->legacy_name); |
c6d57f33 PM |
5804 | if (strlen(root->name)) |
5805 | seq_printf(m, "%sname=%s", count ? "," : "", | |
5806 | root->name); | |
a424316c | 5807 | seq_putc(m, ':'); |
2e91fa7f | 5808 | |
7717f7ba | 5809 | cgrp = task_cgroup_from_root(tsk, root); |
2e91fa7f TH |
5810 | |
5811 | /* | |
5812 | * On traditional hierarchies, all zombie tasks show up as | |
5813 | * belonging to the root cgroup. On the default hierarchy, | |
5814 | * while a zombie doesn't show up in "cgroup.procs" and | |
5815 | * thus can't be migrated, its /proc/PID/cgroup keeps | |
5816 | * reporting the cgroup it belonged to before exiting. If | |
5817 | * the cgroup is removed before the zombie is reaped, | |
5818 | * " (deleted)" is appended to the cgroup path. | |
5819 | */ | |
5820 | if (cgroup_on_dfl(cgrp) || !(tsk->flags & PF_EXITING)) { | |
4c737b41 | 5821 | retval = cgroup_path_ns_locked(cgrp, buf, PATH_MAX, |
a79a908f | 5822 | current->nsproxy->cgroup_ns); |
e0223003 | 5823 | if (retval >= PATH_MAX) |
2e91fa7f | 5824 | retval = -ENAMETOOLONG; |
e0223003 | 5825 | if (retval < 0) |
2e91fa7f | 5826 | goto out_unlock; |
4c737b41 TH |
5827 | |
5828 | seq_puts(m, buf); | |
2e91fa7f | 5829 | } else { |
4c737b41 | 5830 | seq_puts(m, "/"); |
e61734c5 | 5831 | } |
2e91fa7f | 5832 | |
2e91fa7f TH |
5833 | if (cgroup_on_dfl(cgrp) && cgroup_is_dead(cgrp)) |
5834 | seq_puts(m, " (deleted)\n"); | |
5835 | else | |
5836 | seq_putc(m, '\n'); | |
a424316c PM |
5837 | } |
5838 | ||
006f4ac4 | 5839 | retval = 0; |
a424316c | 5840 | out_unlock: |
82d6489d | 5841 | spin_unlock_irq(&css_set_lock); |
a424316c | 5842 | mutex_unlock(&cgroup_mutex); |
a424316c PM |
5843 | kfree(buf); |
5844 | out: | |
5845 | return retval; | |
5846 | } | |
5847 | ||
b4f48b63 | 5848 | /** |
eaf797ab | 5849 | * cgroup_fork - initialize cgroup related fields during copy_process() |
a043e3b2 | 5850 | * @child: pointer to task_struct of forking parent process. |
b4f48b63 | 5851 | * |
eaf797ab TH |
5852 | * A task is associated with the init_css_set until cgroup_post_fork() |
5853 | * attaches it to the parent's css_set. Empty cg_list indicates that | |
5854 | * @child isn't holding reference to its css_set. | |
b4f48b63 PM |
5855 | */ |
5856 | void cgroup_fork(struct task_struct *child) | |
5857 | { | |
eaf797ab | 5858 | RCU_INIT_POINTER(child->cgroups, &init_css_set); |
817929ec | 5859 | INIT_LIST_HEAD(&child->cg_list); |
b4f48b63 PM |
5860 | } |
5861 | ||
7e47682e AS |
5862 | /** |
5863 | * cgroup_can_fork - called on a new task before the process is exposed | |
5864 | * @child: the task in question. | |
5865 | * | |
5866 | * This calls the subsystem can_fork() callbacks. If the can_fork() callback | |
5867 | * returns an error, the fork aborts with that error code. This allows for | |
5868 | * a cgroup subsystem to conditionally allow or deny new forks. | |
5869 | */ | |
b53202e6 | 5870 | int cgroup_can_fork(struct task_struct *child) |
7e47682e AS |
5871 | { |
5872 | struct cgroup_subsys *ss; | |
5873 | int i, j, ret; | |
5874 | ||
b4e0eeaf | 5875 | do_each_subsys_mask(ss, i, have_canfork_callback) { |
b53202e6 | 5876 | ret = ss->can_fork(child); |
7e47682e AS |
5877 | if (ret) |
5878 | goto out_revert; | |
b4e0eeaf | 5879 | } while_each_subsys_mask(); |
7e47682e AS |
5880 | |
5881 | return 0; | |
5882 | ||
5883 | out_revert: | |
5884 | for_each_subsys(ss, j) { | |
5885 | if (j >= i) | |
5886 | break; | |
5887 | if (ss->cancel_fork) | |
b53202e6 | 5888 | ss->cancel_fork(child); |
7e47682e AS |
5889 | } |
5890 | ||
5891 | return ret; | |
5892 | } | |
5893 | ||
5894 | /** | |
5895 | * cgroup_cancel_fork - called if a fork failed after cgroup_can_fork() | |
5896 | * @child: the task in question | |
5897 | * | |
5898 | * This calls the cancel_fork() callbacks if a fork failed *after* | |
5899 | * cgroup_can_fork() succeded. | |
5900 | */ | |
b53202e6 | 5901 | void cgroup_cancel_fork(struct task_struct *child) |
7e47682e AS |
5902 | { |
5903 | struct cgroup_subsys *ss; | |
5904 | int i; | |
5905 | ||
5906 | for_each_subsys(ss, i) | |
5907 | if (ss->cancel_fork) | |
b53202e6 | 5908 | ss->cancel_fork(child); |
7e47682e AS |
5909 | } |
5910 | ||
817929ec | 5911 | /** |
a043e3b2 LZ |
5912 | * cgroup_post_fork - called on a new task after adding it to the task list |
5913 | * @child: the task in question | |
5914 | * | |
5edee61e TH |
5915 | * Adds the task to the list running through its css_set if necessary and |
5916 | * call the subsystem fork() callbacks. Has to be after the task is | |
5917 | * visible on the task list in case we race with the first call to | |
0942eeee | 5918 | * cgroup_task_iter_start() - to guarantee that the new task ends up on its |
5edee61e | 5919 | * list. |
a043e3b2 | 5920 | */ |
b53202e6 | 5921 | void cgroup_post_fork(struct task_struct *child) |
817929ec | 5922 | { |
30159ec7 | 5923 | struct cgroup_subsys *ss; |
5153faac | 5924 | struct css_set *cset; |
5edee61e TH |
5925 | int i; |
5926 | ||
5153faac TH |
5927 | spin_lock_irq(&css_set_lock); |
5928 | ||
5929 | WARN_ON_ONCE(!list_empty(&child->cg_list)); | |
5930 | cset = task_css_set(current); /* current is @child's parent */ | |
5931 | get_css_set(cset); | |
5932 | cset->nr_tasks++; | |
5933 | css_set_move_task(child, NULL, cset, false); | |
5934 | ||
3ce3230a | 5935 | /* |
5153faac TH |
5936 | * If the cgroup has to be frozen, the new task has too. Let's set |
5937 | * the JOBCTL_TRAP_FREEZE jobctl bit to get the task into the | |
5938 | * frozen state. | |
3ce3230a | 5939 | */ |
5153faac TH |
5940 | if (unlikely(cgroup_task_freeze(child))) { |
5941 | spin_lock(&child->sighand->siglock); | |
5942 | WARN_ON_ONCE(child->frozen); | |
5943 | child->jobctl |= JOBCTL_TRAP_FREEZE; | |
5944 | spin_unlock(&child->sighand->siglock); | |
76f969e8 RG |
5945 | |
5946 | /* | |
5153faac TH |
5947 | * Calling cgroup_update_frozen() isn't required here, |
5948 | * because it will be called anyway a bit later from | |
5949 | * do_freezer_trap(). So we avoid cgroup's transient switch | |
5950 | * from the frozen state and back. | |
76f969e8 | 5951 | */ |
817929ec | 5952 | } |
5edee61e | 5953 | |
5153faac TH |
5954 | spin_unlock_irq(&css_set_lock); |
5955 | ||
5edee61e TH |
5956 | /* |
5957 | * Call ss->fork(). This must happen after @child is linked on | |
5958 | * css_set; otherwise, @child might change state between ->fork() | |
5959 | * and addition to css_set. | |
5960 | */ | |
b4e0eeaf | 5961 | do_each_subsys_mask(ss, i, have_fork_callback) { |
b53202e6 | 5962 | ss->fork(child); |
b4e0eeaf | 5963 | } while_each_subsys_mask(); |
817929ec | 5964 | } |
5edee61e | 5965 | |
b4f48b63 PM |
5966 | /** |
5967 | * cgroup_exit - detach cgroup from exiting task | |
5968 | * @tsk: pointer to task_struct of exiting process | |
5969 | * | |
e7c7b1d8 | 5970 | * Description: Detach cgroup from @tsk. |
b4f48b63 | 5971 | * |
b4f48b63 | 5972 | */ |
1ec41830 | 5973 | void cgroup_exit(struct task_struct *tsk) |
b4f48b63 | 5974 | { |
30159ec7 | 5975 | struct cgroup_subsys *ss; |
5abb8855 | 5976 | struct css_set *cset; |
d41d5a01 | 5977 | int i; |
817929ec | 5978 | |
5153faac | 5979 | spin_lock_irq(&css_set_lock); |
0de0942d | 5980 | |
5153faac TH |
5981 | WARN_ON_ONCE(list_empty(&tsk->cg_list)); |
5982 | cset = task_css_set(tsk); | |
5983 | css_set_move_task(tsk, cset, NULL, false); | |
5984 | list_add_tail(&tsk->cg_list, &cset->dying_tasks); | |
5985 | cset->nr_tasks--; | |
76f969e8 | 5986 | |
5153faac TH |
5987 | WARN_ON_ONCE(cgroup_task_frozen(tsk)); |
5988 | if (unlikely(cgroup_task_freeze(tsk))) | |
5989 | cgroup_update_frozen(task_dfl_cgroup(tsk)); | |
76f969e8 | 5990 | |
5153faac | 5991 | spin_unlock_irq(&css_set_lock); |
817929ec | 5992 | |
cb4a3167 | 5993 | /* see cgroup_post_fork() for details */ |
b4e0eeaf | 5994 | do_each_subsys_mask(ss, i, have_exit_callback) { |
2e91fa7f | 5995 | ss->exit(tsk); |
b4e0eeaf | 5996 | } while_each_subsys_mask(); |
2e91fa7f | 5997 | } |
30159ec7 | 5998 | |
51bee5ab | 5999 | void cgroup_release(struct task_struct *task) |
2e91fa7f | 6000 | { |
afcf6c8b TH |
6001 | struct cgroup_subsys *ss; |
6002 | int ssid; | |
6003 | ||
51bee5ab ON |
6004 | do_each_subsys_mask(ss, ssid, have_release_callback) { |
6005 | ss->release(task); | |
b4e0eeaf | 6006 | } while_each_subsys_mask(); |
c03cd773 | 6007 | |
5153faac TH |
6008 | spin_lock_irq(&css_set_lock); |
6009 | css_set_skip_task_iters(task_css_set(task), task); | |
6010 | list_del_init(&task->cg_list); | |
6011 | spin_unlock_irq(&css_set_lock); | |
51bee5ab | 6012 | } |
d41d5a01 | 6013 | |
51bee5ab ON |
6014 | void cgroup_free(struct task_struct *task) |
6015 | { | |
6016 | struct css_set *cset = task_css_set(task); | |
2e91fa7f | 6017 | put_css_set(cset); |
b4f48b63 | 6018 | } |
697f4161 | 6019 | |
8bab8dde PM |
6020 | static int __init cgroup_disable(char *str) |
6021 | { | |
30159ec7 | 6022 | struct cgroup_subsys *ss; |
8bab8dde | 6023 | char *token; |
30159ec7 | 6024 | int i; |
8bab8dde PM |
6025 | |
6026 | while ((token = strsep(&str, ",")) != NULL) { | |
6027 | if (!*token) | |
6028 | continue; | |
be45c900 | 6029 | |
3ed80a62 | 6030 | for_each_subsys(ss, i) { |
3e1d2eed TH |
6031 | if (strcmp(token, ss->name) && |
6032 | strcmp(token, ss->legacy_name)) | |
6033 | continue; | |
a3e72739 | 6034 | cgroup_disable_mask |= 1 << i; |
8bab8dde PM |
6035 | } |
6036 | } | |
6037 | return 1; | |
6038 | } | |
6039 | __setup("cgroup_disable=", cgroup_disable); | |
38460b48 | 6040 | |
5cf8114d WL |
6041 | void __init __weak enable_debug_cgroup(void) { } |
6042 | ||
6043 | static int __init enable_cgroup_debug(char *str) | |
6044 | { | |
6045 | cgroup_debug = true; | |
6046 | enable_debug_cgroup(); | |
6047 | return 1; | |
6048 | } | |
6049 | __setup("cgroup_debug", enable_cgroup_debug); | |
6050 | ||
b77d7b60 | 6051 | /** |
ec903c0c | 6052 | * css_tryget_online_from_dir - get corresponding css from a cgroup dentry |
35cf0836 TH |
6053 | * @dentry: directory dentry of interest |
6054 | * @ss: subsystem of interest | |
b77d7b60 | 6055 | * |
5a17f543 TH |
6056 | * If @dentry is a directory for a cgroup which has @ss enabled on it, try |
6057 | * to get the corresponding css and return it. If such css doesn't exist | |
6058 | * or can't be pinned, an ERR_PTR value is returned. | |
e5d1367f | 6059 | */ |
ec903c0c TH |
6060 | struct cgroup_subsys_state *css_tryget_online_from_dir(struct dentry *dentry, |
6061 | struct cgroup_subsys *ss) | |
e5d1367f | 6062 | { |
2bd59d48 | 6063 | struct kernfs_node *kn = kernfs_node_from_dentry(dentry); |
f17fc25f | 6064 | struct file_system_type *s_type = dentry->d_sb->s_type; |
2bd59d48 | 6065 | struct cgroup_subsys_state *css = NULL; |
e5d1367f | 6066 | struct cgroup *cgrp; |
e5d1367f | 6067 | |
35cf0836 | 6068 | /* is @dentry a cgroup dir? */ |
f17fc25f TH |
6069 | if ((s_type != &cgroup_fs_type && s_type != &cgroup2_fs_type) || |
6070 | !kn || kernfs_type(kn) != KERNFS_DIR) | |
e5d1367f SE |
6071 | return ERR_PTR(-EBADF); |
6072 | ||
5a17f543 TH |
6073 | rcu_read_lock(); |
6074 | ||
2bd59d48 TH |
6075 | /* |
6076 | * This path doesn't originate from kernfs and @kn could already | |
6077 | * have been or be removed at any point. @kn->priv is RCU | |
a4189487 | 6078 | * protected for this access. See css_release_work_fn() for details. |
2bd59d48 | 6079 | */ |
e0aed7c7 | 6080 | cgrp = rcu_dereference(*(void __rcu __force **)&kn->priv); |
2bd59d48 TH |
6081 | if (cgrp) |
6082 | css = cgroup_css(cgrp, ss); | |
5a17f543 | 6083 | |
ec903c0c | 6084 | if (!css || !css_tryget_online(css)) |
5a17f543 TH |
6085 | css = ERR_PTR(-ENOENT); |
6086 | ||
6087 | rcu_read_unlock(); | |
6088 | return css; | |
e5d1367f | 6089 | } |
e5d1367f | 6090 | |
1cb650b9 LZ |
6091 | /** |
6092 | * css_from_id - lookup css by id | |
6093 | * @id: the cgroup id | |
6094 | * @ss: cgroup subsys to be looked into | |
6095 | * | |
6096 | * Returns the css if there's valid one with @id, otherwise returns NULL. | |
6097 | * Should be called under rcu_read_lock(). | |
6098 | */ | |
6099 | struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss) | |
6100 | { | |
6fa4918d | 6101 | WARN_ON_ONCE(!rcu_read_lock_held()); |
d6ccc55e | 6102 | return idr_find(&ss->css_idr, id); |
e5d1367f SE |
6103 | } |
6104 | ||
16af4396 TH |
6105 | /** |
6106 | * cgroup_get_from_path - lookup and get a cgroup from its default hierarchy path | |
6107 | * @path: path on the default hierarchy | |
6108 | * | |
6109 | * Find the cgroup at @path on the default hierarchy, increment its | |
6110 | * reference count and return it. Returns pointer to the found cgroup on | |
6111 | * success, ERR_PTR(-ENOENT) if @path doens't exist and ERR_PTR(-ENOTDIR) | |
6112 | * if @path points to a non-directory. | |
6113 | */ | |
6114 | struct cgroup *cgroup_get_from_path(const char *path) | |
6115 | { | |
6116 | struct kernfs_node *kn; | |
6117 | struct cgroup *cgrp; | |
6118 | ||
6119 | mutex_lock(&cgroup_mutex); | |
6120 | ||
6121 | kn = kernfs_walk_and_get(cgrp_dfl_root.cgrp.kn, path); | |
6122 | if (kn) { | |
6123 | if (kernfs_type(kn) == KERNFS_DIR) { | |
6124 | cgrp = kn->priv; | |
a590b90d | 6125 | cgroup_get_live(cgrp); |
16af4396 TH |
6126 | } else { |
6127 | cgrp = ERR_PTR(-ENOTDIR); | |
6128 | } | |
6129 | kernfs_put(kn); | |
6130 | } else { | |
6131 | cgrp = ERR_PTR(-ENOENT); | |
6132 | } | |
6133 | ||
6134 | mutex_unlock(&cgroup_mutex); | |
6135 | return cgrp; | |
6136 | } | |
6137 | EXPORT_SYMBOL_GPL(cgroup_get_from_path); | |
6138 | ||
1f3fe7eb MKL |
6139 | /** |
6140 | * cgroup_get_from_fd - get a cgroup pointer from a fd | |
6141 | * @fd: fd obtained by open(cgroup2_dir) | |
6142 | * | |
6143 | * Find the cgroup from a fd which should be obtained | |
6144 | * by opening a cgroup directory. Returns a pointer to the | |
6145 | * cgroup on success. ERR_PTR is returned if the cgroup | |
6146 | * cannot be found. | |
6147 | */ | |
6148 | struct cgroup *cgroup_get_from_fd(int fd) | |
6149 | { | |
6150 | struct cgroup_subsys_state *css; | |
6151 | struct cgroup *cgrp; | |
6152 | struct file *f; | |
6153 | ||
6154 | f = fget_raw(fd); | |
6155 | if (!f) | |
6156 | return ERR_PTR(-EBADF); | |
6157 | ||
6158 | css = css_tryget_online_from_dir(f->f_path.dentry, NULL); | |
6159 | fput(f); | |
6160 | if (IS_ERR(css)) | |
6161 | return ERR_CAST(css); | |
6162 | ||
6163 | cgrp = css->cgroup; | |
6164 | if (!cgroup_on_dfl(cgrp)) { | |
6165 | cgroup_put(cgrp); | |
6166 | return ERR_PTR(-EBADF); | |
6167 | } | |
6168 | ||
6169 | return cgrp; | |
6170 | } | |
6171 | EXPORT_SYMBOL_GPL(cgroup_get_from_fd); | |
6172 | ||
38cf3a68 TH |
6173 | static u64 power_of_ten(int power) |
6174 | { | |
6175 | u64 v = 1; | |
6176 | while (power--) | |
6177 | v *= 10; | |
6178 | return v; | |
6179 | } | |
6180 | ||
6181 | /** | |
6182 | * cgroup_parse_float - parse a floating number | |
6183 | * @input: input string | |
6184 | * @dec_shift: number of decimal digits to shift | |
6185 | * @v: output | |
6186 | * | |
6187 | * Parse a decimal floating point number in @input and store the result in | |
6188 | * @v with decimal point right shifted @dec_shift times. For example, if | |
6189 | * @input is "12.3456" and @dec_shift is 3, *@v will be set to 12345. | |
6190 | * Returns 0 on success, -errno otherwise. | |
6191 | * | |
6192 | * There's nothing cgroup specific about this function except that it's | |
6193 | * currently the only user. | |
6194 | */ | |
6195 | int cgroup_parse_float(const char *input, unsigned dec_shift, s64 *v) | |
6196 | { | |
6197 | s64 whole, frac = 0; | |
6198 | int fstart = 0, fend = 0, flen; | |
6199 | ||
6200 | if (!sscanf(input, "%lld.%n%lld%n", &whole, &fstart, &frac, &fend)) | |
6201 | return -EINVAL; | |
6202 | if (frac < 0) | |
6203 | return -EINVAL; | |
6204 | ||
6205 | flen = fend > fstart ? fend - fstart : 0; | |
6206 | if (flen < dec_shift) | |
6207 | frac *= power_of_ten(dec_shift - flen); | |
6208 | else | |
6209 | frac = DIV_ROUND_CLOSEST_ULL(frac, power_of_ten(flen - dec_shift)); | |
6210 | ||
6211 | *v = whole * power_of_ten(dec_shift) + frac; | |
6212 | return 0; | |
6213 | } | |
6214 | ||
bd1060a1 TH |
6215 | /* |
6216 | * sock->sk_cgrp_data handling. For more info, see sock_cgroup_data | |
6217 | * definition in cgroup-defs.h. | |
6218 | */ | |
6219 | #ifdef CONFIG_SOCK_CGROUP_DATA | |
6220 | ||
6221 | #if defined(CONFIG_CGROUP_NET_PRIO) || defined(CONFIG_CGROUP_NET_CLASSID) | |
6222 | ||
3fa4cc9c | 6223 | DEFINE_SPINLOCK(cgroup_sk_update_lock); |
bd1060a1 TH |
6224 | static bool cgroup_sk_alloc_disabled __read_mostly; |
6225 | ||
6226 | void cgroup_sk_alloc_disable(void) | |
6227 | { | |
6228 | if (cgroup_sk_alloc_disabled) | |
6229 | return; | |
6230 | pr_info("cgroup: disabling cgroup2 socket matching due to net_prio or net_cls activation\n"); | |
6231 | cgroup_sk_alloc_disabled = true; | |
6232 | } | |
6233 | ||
6234 | #else | |
6235 | ||
6236 | #define cgroup_sk_alloc_disabled false | |
6237 | ||
6238 | #endif | |
6239 | ||
6240 | void cgroup_sk_alloc(struct sock_cgroup_data *skcd) | |
6241 | { | |
6242 | if (cgroup_sk_alloc_disabled) | |
6243 | return; | |
6244 | ||
d979a39d JW |
6245 | /* Socket clone path */ |
6246 | if (skcd->val) { | |
a590b90d TH |
6247 | /* |
6248 | * We might be cloning a socket which is left in an empty | |
6249 | * cgroup and the cgroup might have already been rmdir'd. | |
6250 | * Don't use cgroup_get_live(). | |
6251 | */ | |
d979a39d | 6252 | cgroup_get(sock_cgroup_ptr(skcd)); |
4bfc0bb2 | 6253 | cgroup_bpf_get(sock_cgroup_ptr(skcd)); |
d979a39d JW |
6254 | return; |
6255 | } | |
6256 | ||
bd1060a1 TH |
6257 | rcu_read_lock(); |
6258 | ||
6259 | while (true) { | |
6260 | struct css_set *cset; | |
6261 | ||
6262 | cset = task_css_set(current); | |
6263 | if (likely(cgroup_tryget(cset->dfl_cgrp))) { | |
6264 | skcd->val = (unsigned long)cset->dfl_cgrp; | |
4bfc0bb2 | 6265 | cgroup_bpf_get(cset->dfl_cgrp); |
bd1060a1 TH |
6266 | break; |
6267 | } | |
6268 | cpu_relax(); | |
6269 | } | |
6270 | ||
6271 | rcu_read_unlock(); | |
6272 | } | |
6273 | ||
6274 | void cgroup_sk_free(struct sock_cgroup_data *skcd) | |
6275 | { | |
4bfc0bb2 RG |
6276 | struct cgroup *cgrp = sock_cgroup_ptr(skcd); |
6277 | ||
6278 | cgroup_bpf_put(cgrp); | |
6279 | cgroup_put(cgrp); | |
bd1060a1 TH |
6280 | } |
6281 | ||
6282 | #endif /* CONFIG_SOCK_CGROUP_DATA */ | |
6283 | ||
30070984 | 6284 | #ifdef CONFIG_CGROUP_BPF |
324bda9e AS |
6285 | int cgroup_bpf_attach(struct cgroup *cgrp, struct bpf_prog *prog, |
6286 | enum bpf_attach_type type, u32 flags) | |
6287 | { | |
6288 | int ret; | |
6289 | ||
6290 | mutex_lock(&cgroup_mutex); | |
6291 | ret = __cgroup_bpf_attach(cgrp, prog, type, flags); | |
6292 | mutex_unlock(&cgroup_mutex); | |
6293 | return ret; | |
6294 | } | |
6295 | int cgroup_bpf_detach(struct cgroup *cgrp, struct bpf_prog *prog, | |
6296 | enum bpf_attach_type type, u32 flags) | |
30070984 | 6297 | { |
7f677633 | 6298 | int ret; |
30070984 DM |
6299 | |
6300 | mutex_lock(&cgroup_mutex); | |
1832f4ef | 6301 | ret = __cgroup_bpf_detach(cgrp, prog, type); |
30070984 | 6302 | mutex_unlock(&cgroup_mutex); |
7f677633 | 6303 | return ret; |
30070984 | 6304 | } |
468e2f64 AS |
6305 | int cgroup_bpf_query(struct cgroup *cgrp, const union bpf_attr *attr, |
6306 | union bpf_attr __user *uattr) | |
6307 | { | |
6308 | int ret; | |
6309 | ||
6310 | mutex_lock(&cgroup_mutex); | |
6311 | ret = __cgroup_bpf_query(cgrp, attr, uattr); | |
6312 | mutex_unlock(&cgroup_mutex); | |
6313 | return ret; | |
6314 | } | |
30070984 | 6315 | #endif /* CONFIG_CGROUP_BPF */ |
01ee6cfb RG |
6316 | |
6317 | #ifdef CONFIG_SYSFS | |
6318 | static ssize_t show_delegatable_files(struct cftype *files, char *buf, | |
6319 | ssize_t size, const char *prefix) | |
6320 | { | |
6321 | struct cftype *cft; | |
6322 | ssize_t ret = 0; | |
6323 | ||
6324 | for (cft = files; cft && cft->name[0] != '\0'; cft++) { | |
6325 | if (!(cft->flags & CFTYPE_NS_DELEGATABLE)) | |
6326 | continue; | |
6327 | ||
6328 | if (prefix) | |
6329 | ret += snprintf(buf + ret, size - ret, "%s.", prefix); | |
6330 | ||
6331 | ret += snprintf(buf + ret, size - ret, "%s\n", cft->name); | |
6332 | ||
4d9ebbe2 | 6333 | if (WARN_ON(ret >= size)) |
01ee6cfb | 6334 | break; |
01ee6cfb RG |
6335 | } |
6336 | ||
6337 | return ret; | |
6338 | } | |
6339 | ||
6340 | static ssize_t delegate_show(struct kobject *kobj, struct kobj_attribute *attr, | |
6341 | char *buf) | |
6342 | { | |
6343 | struct cgroup_subsys *ss; | |
6344 | int ssid; | |
6345 | ssize_t ret = 0; | |
6346 | ||
6347 | ret = show_delegatable_files(cgroup_base_files, buf, PAGE_SIZE - ret, | |
6348 | NULL); | |
6349 | ||
6350 | for_each_subsys(ss, ssid) | |
6351 | ret += show_delegatable_files(ss->dfl_cftypes, buf + ret, | |
6352 | PAGE_SIZE - ret, | |
6353 | cgroup_subsys_name[ssid]); | |
6354 | ||
6355 | return ret; | |
6356 | } | |
6357 | static struct kobj_attribute cgroup_delegate_attr = __ATTR_RO(delegate); | |
6358 | ||
5f2e6734 RG |
6359 | static ssize_t features_show(struct kobject *kobj, struct kobj_attribute *attr, |
6360 | char *buf) | |
6361 | { | |
9852ae3f | 6362 | return snprintf(buf, PAGE_SIZE, "nsdelegate\nmemory_localevents\n"); |
5f2e6734 RG |
6363 | } |
6364 | static struct kobj_attribute cgroup_features_attr = __ATTR_RO(features); | |
6365 | ||
01ee6cfb RG |
6366 | static struct attribute *cgroup_sysfs_attrs[] = { |
6367 | &cgroup_delegate_attr.attr, | |
5f2e6734 | 6368 | &cgroup_features_attr.attr, |
01ee6cfb RG |
6369 | NULL, |
6370 | }; | |
6371 | ||
6372 | static const struct attribute_group cgroup_sysfs_attr_group = { | |
6373 | .attrs = cgroup_sysfs_attrs, | |
6374 | .name = "cgroup", | |
6375 | }; | |
6376 | ||
6377 | static int __init cgroup_sysfs_init(void) | |
6378 | { | |
6379 | return sysfs_create_group(kernel_kobj, &cgroup_sysfs_attr_group); | |
6380 | } | |
6381 | subsys_initcall(cgroup_sysfs_init); | |
a5e112e6 | 6382 | |
01ee6cfb | 6383 | #endif /* CONFIG_SYSFS */ |