]>
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> |
bd1060a1 | 57 | #include <net/sock.h> |
ddbcc7e8 | 58 | |
ed1777de TH |
59 | #define CREATE_TRACE_POINTS |
60 | #include <trace/events/cgroup.h> | |
61 | ||
8d7e6fb0 TH |
62 | #define CGROUP_FILE_NAME_MAX (MAX_CGROUP_TYPE_NAMELEN + \ |
63 | MAX_CFTYPE_NAME + 2) | |
64 | ||
e25e2cbb TH |
65 | /* |
66 | * cgroup_mutex is the master lock. Any modification to cgroup or its | |
67 | * hierarchy must be performed while holding it. | |
68 | * | |
f0d9a5f1 | 69 | * css_set_lock protects task->cgroups pointer, the list of css_set |
0e1d768f | 70 | * objects, and the chain of tasks off each css_set. |
e25e2cbb | 71 | * |
0e1d768f TH |
72 | * These locks are exported if CONFIG_PROVE_RCU so that accessors in |
73 | * cgroup.h can use them for lockdep annotations. | |
e25e2cbb | 74 | */ |
2219449a | 75 | DEFINE_MUTEX(cgroup_mutex); |
f0d9a5f1 | 76 | DEFINE_SPINLOCK(css_set_lock); |
0a268dbd TH |
77 | |
78 | #ifdef CONFIG_PROVE_RCU | |
0e1d768f | 79 | EXPORT_SYMBOL_GPL(cgroup_mutex); |
f0d9a5f1 | 80 | EXPORT_SYMBOL_GPL(css_set_lock); |
2219449a TH |
81 | #endif |
82 | ||
6fa4918d | 83 | /* |
15a4c835 TH |
84 | * Protects cgroup_idr and css_idr so that IDs can be released without |
85 | * grabbing cgroup_mutex. | |
6fa4918d TH |
86 | */ |
87 | static DEFINE_SPINLOCK(cgroup_idr_lock); | |
88 | ||
34c06254 TH |
89 | /* |
90 | * Protects cgroup_file->kn for !self csses. It synchronizes notifications | |
91 | * against file removal/re-creation across css hiding. | |
92 | */ | |
93 | static DEFINE_SPINLOCK(cgroup_file_kn_lock); | |
94 | ||
1ed13287 TH |
95 | struct percpu_rw_semaphore cgroup_threadgroup_rwsem; |
96 | ||
8353da1f | 97 | #define cgroup_assert_mutex_or_rcu_locked() \ |
f78f5b90 PM |
98 | RCU_LOCKDEP_WARN(!rcu_read_lock_held() && \ |
99 | !lockdep_is_held(&cgroup_mutex), \ | |
8353da1f | 100 | "cgroup_mutex or RCU read lock required"); |
780cd8b3 | 101 | |
e5fca243 TH |
102 | /* |
103 | * cgroup destruction makes heavy use of work items and there can be a lot | |
104 | * of concurrent destructions. Use a separate workqueue so that cgroup | |
105 | * destruction work items don't end up filling up max_active of system_wq | |
106 | * which may lead to deadlock. | |
107 | */ | |
108 | static struct workqueue_struct *cgroup_destroy_wq; | |
109 | ||
3ed80a62 | 110 | /* generate an array of cgroup subsystem pointers */ |
073219e9 | 111 | #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys, |
0a268dbd | 112 | struct cgroup_subsys *cgroup_subsys[] = { |
ddbcc7e8 PM |
113 | #include <linux/cgroup_subsys.h> |
114 | }; | |
073219e9 TH |
115 | #undef SUBSYS |
116 | ||
117 | /* array of cgroup subsystem names */ | |
118 | #define SUBSYS(_x) [_x ## _cgrp_id] = #_x, | |
119 | static const char *cgroup_subsys_name[] = { | |
ddbcc7e8 PM |
120 | #include <linux/cgroup_subsys.h> |
121 | }; | |
073219e9 | 122 | #undef SUBSYS |
ddbcc7e8 | 123 | |
49d1dc4b TH |
124 | /* array of static_keys for cgroup_subsys_enabled() and cgroup_subsys_on_dfl() */ |
125 | #define SUBSYS(_x) \ | |
126 | DEFINE_STATIC_KEY_TRUE(_x ## _cgrp_subsys_enabled_key); \ | |
127 | DEFINE_STATIC_KEY_TRUE(_x ## _cgrp_subsys_on_dfl_key); \ | |
128 | EXPORT_SYMBOL_GPL(_x ## _cgrp_subsys_enabled_key); \ | |
129 | EXPORT_SYMBOL_GPL(_x ## _cgrp_subsys_on_dfl_key); | |
130 | #include <linux/cgroup_subsys.h> | |
131 | #undef SUBSYS | |
132 | ||
133 | #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys_enabled_key, | |
134 | static struct static_key_true *cgroup_subsys_enabled_key[] = { | |
135 | #include <linux/cgroup_subsys.h> | |
136 | }; | |
137 | #undef SUBSYS | |
138 | ||
139 | #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys_on_dfl_key, | |
140 | static struct static_key_true *cgroup_subsys_on_dfl_key[] = { | |
141 | #include <linux/cgroup_subsys.h> | |
142 | }; | |
143 | #undef SUBSYS | |
144 | ||
041cd640 TH |
145 | static DEFINE_PER_CPU(struct cgroup_cpu_stat, cgrp_dfl_root_cpu_stat); |
146 | ||
ddbcc7e8 | 147 | /* |
3dd06ffa | 148 | * The default hierarchy, reserved for the subsystems that are otherwise |
9871bf95 TH |
149 | * unattached - it never has more than a single cgroup, and all tasks are |
150 | * part of that cgroup. | |
ddbcc7e8 | 151 | */ |
041cd640 | 152 | struct cgroup_root cgrp_dfl_root = { .cgrp.cpu_stat = &cgrp_dfl_root_cpu_stat }; |
d0ec4230 | 153 | EXPORT_SYMBOL_GPL(cgrp_dfl_root); |
9871bf95 | 154 | |
a2dd4247 TH |
155 | /* |
156 | * The default hierarchy always exists but is hidden until mounted for the | |
157 | * first time. This is for backward compatibility. | |
158 | */ | |
a7165264 | 159 | static bool cgrp_dfl_visible; |
ddbcc7e8 | 160 | |
5533e011 | 161 | /* some controllers are not supported in the default hierarchy */ |
a7165264 | 162 | static u16 cgrp_dfl_inhibit_ss_mask; |
5533e011 | 163 | |
f6d635ad | 164 | /* some controllers are implicitly enabled on the default hierarchy */ |
b807421a | 165 | static u16 cgrp_dfl_implicit_ss_mask; |
f6d635ad | 166 | |
8cfd8147 TH |
167 | /* some controllers can be threaded on the default hierarchy */ |
168 | static u16 cgrp_dfl_threaded_ss_mask; | |
169 | ||
ddbcc7e8 | 170 | /* The list of hierarchy roots */ |
0a268dbd | 171 | LIST_HEAD(cgroup_roots); |
9871bf95 | 172 | static int cgroup_root_count; |
ddbcc7e8 | 173 | |
3417ae1f | 174 | /* hierarchy ID allocation and mapping, protected by cgroup_mutex */ |
1a574231 | 175 | static DEFINE_IDR(cgroup_hierarchy_idr); |
2c6ab6d2 | 176 | |
794611a1 | 177 | /* |
0cb51d71 TH |
178 | * Assign a monotonically increasing serial number to csses. It guarantees |
179 | * cgroups with bigger numbers are newer than those with smaller numbers. | |
180 | * Also, as csses are always appended to the parent's ->children list, it | |
181 | * guarantees that sibling csses are always sorted in the ascending serial | |
182 | * number order on the list. Protected by cgroup_mutex. | |
794611a1 | 183 | */ |
0cb51d71 | 184 | static u64 css_serial_nr_next = 1; |
794611a1 | 185 | |
cb4a3167 | 186 | /* |
b807421a TH |
187 | * These bitmasks identify subsystems with specific features to avoid |
188 | * having to do iterative checks repeatedly. | |
ddbcc7e8 | 189 | */ |
6e5c8307 TH |
190 | static u16 have_fork_callback __read_mostly; |
191 | static u16 have_exit_callback __read_mostly; | |
d7248a93 | 192 | static u16 have_release_callback __read_mostly; |
b807421a | 193 | static u16 have_canfork_callback __read_mostly; |
ddbcc7e8 | 194 | |
a79a908f AK |
195 | /* cgroup namespace for init task */ |
196 | struct cgroup_namespace init_cgroup_ns = { | |
387ad967 | 197 | .count = REFCOUNT_INIT(2), |
a79a908f AK |
198 | .user_ns = &init_user_ns, |
199 | .ns.ops = &cgroupns_operations, | |
200 | .ns.inum = PROC_CGROUP_INIT_INO, | |
201 | .root_cset = &init_css_set, | |
202 | }; | |
203 | ||
67e9c74b | 204 | static struct file_system_type cgroup2_fs_type; |
d62beb7f | 205 | static struct cftype cgroup_base_files[]; |
628f7cd4 | 206 | |
334c3679 TH |
207 | static int cgroup_apply_control(struct cgroup *cgrp); |
208 | static void cgroup_finalize_control(struct cgroup *cgrp, int ret); | |
3995c283 TH |
209 | static void css_task_iter_skip(struct css_task_iter *it, |
210 | struct task_struct *task); | |
42809dd4 | 211 | static int cgroup_destroy_locked(struct cgroup *cgrp); |
6cd0f5bb TH |
212 | static struct cgroup_subsys_state *css_create(struct cgroup *cgrp, |
213 | struct cgroup_subsys *ss); | |
9d755d33 | 214 | static void css_release(struct percpu_ref *ref); |
f8f22e53 | 215 | static void kill_css(struct cgroup_subsys_state *css); |
4df8dc90 TH |
216 | static int cgroup_addrm_files(struct cgroup_subsys_state *css, |
217 | struct cgroup *cgrp, struct cftype cfts[], | |
2bb566cb | 218 | bool is_add); |
42809dd4 | 219 | |
fc5ed1e9 TH |
220 | /** |
221 | * cgroup_ssid_enabled - cgroup subsys enabled test by subsys ID | |
222 | * @ssid: subsys ID of interest | |
223 | * | |
224 | * cgroup_subsys_enabled() can only be used with literal subsys names which | |
225 | * is fine for individual subsystems but unsuitable for cgroup core. This | |
226 | * is slower static_key_enabled() based test indexed by @ssid. | |
227 | */ | |
0a268dbd | 228 | bool cgroup_ssid_enabled(int ssid) |
fc5ed1e9 | 229 | { |
cfe02a8a AB |
230 | if (CGROUP_SUBSYS_COUNT == 0) |
231 | return false; | |
232 | ||
fc5ed1e9 TH |
233 | return static_key_enabled(cgroup_subsys_enabled_key[ssid]); |
234 | } | |
235 | ||
9e10a130 TH |
236 | /** |
237 | * cgroup_on_dfl - test whether a cgroup is on the default hierarchy | |
238 | * @cgrp: the cgroup of interest | |
239 | * | |
240 | * The default hierarchy is the v2 interface of cgroup and this function | |
241 | * can be used to test whether a cgroup is on the default hierarchy for | |
242 | * cases where a subsystem should behave differnetly depending on the | |
243 | * interface version. | |
244 | * | |
245 | * The set of behaviors which change on the default hierarchy are still | |
246 | * being determined and the mount option is prefixed with __DEVEL__. | |
247 | * | |
248 | * List of changed behaviors: | |
249 | * | |
250 | * - Mount options "noprefix", "xattr", "clone_children", "release_agent" | |
251 | * and "name" are disallowed. | |
252 | * | |
253 | * - When mounting an existing superblock, mount options should match. | |
254 | * | |
255 | * - Remount is disallowed. | |
256 | * | |
257 | * - rename(2) is disallowed. | |
258 | * | |
259 | * - "tasks" is removed. Everything should be at process granularity. Use | |
260 | * "cgroup.procs" instead. | |
261 | * | |
262 | * - "cgroup.procs" is not sorted. pids will be unique unless they got | |
263 | * recycled inbetween reads. | |
264 | * | |
265 | * - "release_agent" and "notify_on_release" are removed. Replacement | |
266 | * notification mechanism will be implemented. | |
267 | * | |
268 | * - "cgroup.clone_children" is removed. | |
269 | * | |
270 | * - "cgroup.subtree_populated" is available. Its value is 0 if the cgroup | |
271 | * and its descendants contain no task; otherwise, 1. The file also | |
272 | * generates kernfs notification which can be monitored through poll and | |
273 | * [di]notify when the value of the file changes. | |
274 | * | |
275 | * - cpuset: tasks will be kept in empty cpusets when hotplug happens and | |
276 | * take masks of ancestors with non-empty cpus/mems, instead of being | |
277 | * moved to an ancestor. | |
278 | * | |
279 | * - cpuset: a task can be moved into an empty cpuset, and again it takes | |
280 | * masks of ancestors. | |
281 | * | |
282 | * - memcg: use_hierarchy is on by default and the cgroup file for the flag | |
283 | * is not created. | |
284 | * | |
285 | * - blkcg: blk-throttle becomes properly hierarchical. | |
286 | * | |
287 | * - debug: disallowed on the default hierarchy. | |
288 | */ | |
0a268dbd | 289 | bool cgroup_on_dfl(const struct cgroup *cgrp) |
9e10a130 TH |
290 | { |
291 | return cgrp->root == &cgrp_dfl_root; | |
292 | } | |
293 | ||
6fa4918d TH |
294 | /* IDR wrappers which synchronize using cgroup_idr_lock */ |
295 | static int cgroup_idr_alloc(struct idr *idr, void *ptr, int start, int end, | |
296 | gfp_t gfp_mask) | |
297 | { | |
298 | int ret; | |
299 | ||
300 | idr_preload(gfp_mask); | |
54504e97 | 301 | spin_lock_bh(&cgroup_idr_lock); |
d0164adc | 302 | ret = idr_alloc(idr, ptr, start, end, gfp_mask & ~__GFP_DIRECT_RECLAIM); |
54504e97 | 303 | spin_unlock_bh(&cgroup_idr_lock); |
6fa4918d TH |
304 | idr_preload_end(); |
305 | return ret; | |
306 | } | |
307 | ||
308 | static void *cgroup_idr_replace(struct idr *idr, void *ptr, int id) | |
309 | { | |
310 | void *ret; | |
311 | ||
54504e97 | 312 | spin_lock_bh(&cgroup_idr_lock); |
6fa4918d | 313 | ret = idr_replace(idr, ptr, id); |
54504e97 | 314 | spin_unlock_bh(&cgroup_idr_lock); |
6fa4918d TH |
315 | return ret; |
316 | } | |
317 | ||
318 | static void cgroup_idr_remove(struct idr *idr, int id) | |
319 | { | |
54504e97 | 320 | spin_lock_bh(&cgroup_idr_lock); |
6fa4918d | 321 | idr_remove(idr, id); |
54504e97 | 322 | spin_unlock_bh(&cgroup_idr_lock); |
6fa4918d TH |
323 | } |
324 | ||
27f26753 | 325 | static bool cgroup_has_tasks(struct cgroup *cgrp) |
d51f39b0 | 326 | { |
27f26753 TH |
327 | return cgrp->nr_populated_csets; |
328 | } | |
d51f39b0 | 329 | |
7a0cf0e7 | 330 | bool cgroup_is_threaded(struct cgroup *cgrp) |
454000ad TH |
331 | { |
332 | return cgrp->dom_cgrp != cgrp; | |
333 | } | |
334 | ||
8cfd8147 TH |
335 | /* can @cgrp host both domain and threaded children? */ |
336 | static bool cgroup_is_mixable(struct cgroup *cgrp) | |
337 | { | |
338 | /* | |
339 | * Root isn't under domain level resource control exempting it from | |
340 | * the no-internal-process constraint, so it can serve as a thread | |
341 | * root and a parent of resource domains at the same time. | |
342 | */ | |
343 | return !cgroup_parent(cgrp); | |
344 | } | |
345 | ||
346 | /* can @cgrp become a thread root? should always be true for a thread root */ | |
347 | static bool cgroup_can_be_thread_root(struct cgroup *cgrp) | |
348 | { | |
349 | /* mixables don't care */ | |
350 | if (cgroup_is_mixable(cgrp)) | |
351 | return true; | |
352 | ||
353 | /* domain roots can't be nested under threaded */ | |
354 | if (cgroup_is_threaded(cgrp)) | |
355 | return false; | |
356 | ||
357 | /* can only have either domain or threaded children */ | |
358 | if (cgrp->nr_populated_domain_children) | |
359 | return false; | |
360 | ||
361 | /* and no domain controllers can be enabled */ | |
362 | if (cgrp->subtree_control & ~cgrp_dfl_threaded_ss_mask) | |
363 | return false; | |
364 | ||
365 | return true; | |
366 | } | |
367 | ||
368 | /* is @cgrp root of a threaded subtree? */ | |
7a0cf0e7 | 369 | bool cgroup_is_thread_root(struct cgroup *cgrp) |
8cfd8147 TH |
370 | { |
371 | /* thread root should be a domain */ | |
372 | if (cgroup_is_threaded(cgrp)) | |
373 | return false; | |
374 | ||
375 | /* a domain w/ threaded children is a thread root */ | |
376 | if (cgrp->nr_threaded_children) | |
377 | return true; | |
378 | ||
379 | /* | |
380 | * A domain which has tasks and explicit threaded controllers | |
381 | * enabled is a thread root. | |
382 | */ | |
383 | if (cgroup_has_tasks(cgrp) && | |
384 | (cgrp->subtree_control & cgrp_dfl_threaded_ss_mask)) | |
385 | return true; | |
386 | ||
387 | return false; | |
388 | } | |
389 | ||
390 | /* a domain which isn't connected to the root w/o brekage can't be used */ | |
391 | static bool cgroup_is_valid_domain(struct cgroup *cgrp) | |
392 | { | |
393 | /* the cgroup itself can be a thread root */ | |
394 | if (cgroup_is_threaded(cgrp)) | |
395 | return false; | |
396 | ||
397 | /* but the ancestors can't be unless mixable */ | |
398 | while ((cgrp = cgroup_parent(cgrp))) { | |
399 | if (!cgroup_is_mixable(cgrp) && cgroup_is_thread_root(cgrp)) | |
400 | return false; | |
401 | if (cgroup_is_threaded(cgrp)) | |
402 | return false; | |
403 | } | |
404 | ||
405 | return true; | |
d51f39b0 TH |
406 | } |
407 | ||
5531dc91 TH |
408 | /* subsystems visibly enabled on a cgroup */ |
409 | static u16 cgroup_control(struct cgroup *cgrp) | |
410 | { | |
411 | struct cgroup *parent = cgroup_parent(cgrp); | |
412 | u16 root_ss_mask = cgrp->root->subsys_mask; | |
413 | ||
8cfd8147 TH |
414 | if (parent) { |
415 | u16 ss_mask = parent->subtree_control; | |
416 | ||
417 | /* threaded cgroups can only have threaded controllers */ | |
418 | if (cgroup_is_threaded(cgrp)) | |
419 | ss_mask &= cgrp_dfl_threaded_ss_mask; | |
420 | return ss_mask; | |
421 | } | |
5531dc91 TH |
422 | |
423 | if (cgroup_on_dfl(cgrp)) | |
f6d635ad TH |
424 | root_ss_mask &= ~(cgrp_dfl_inhibit_ss_mask | |
425 | cgrp_dfl_implicit_ss_mask); | |
5531dc91 TH |
426 | return root_ss_mask; |
427 | } | |
428 | ||
429 | /* subsystems enabled on a cgroup */ | |
430 | static u16 cgroup_ss_mask(struct cgroup *cgrp) | |
431 | { | |
432 | struct cgroup *parent = cgroup_parent(cgrp); | |
433 | ||
8cfd8147 TH |
434 | if (parent) { |
435 | u16 ss_mask = parent->subtree_ss_mask; | |
436 | ||
437 | /* threaded cgroups can only have threaded controllers */ | |
438 | if (cgroup_is_threaded(cgrp)) | |
439 | ss_mask &= cgrp_dfl_threaded_ss_mask; | |
440 | return ss_mask; | |
441 | } | |
5531dc91 TH |
442 | |
443 | return cgrp->root->subsys_mask; | |
444 | } | |
445 | ||
95109b62 TH |
446 | /** |
447 | * cgroup_css - obtain a cgroup's css for the specified subsystem | |
448 | * @cgrp: the cgroup of interest | |
9d800df1 | 449 | * @ss: the subsystem of interest (%NULL returns @cgrp->self) |
95109b62 | 450 | * |
ca8bdcaf TH |
451 | * Return @cgrp's css (cgroup_subsys_state) associated with @ss. This |
452 | * function must be called either under cgroup_mutex or rcu_read_lock() and | |
453 | * the caller is responsible for pinning the returned css if it wants to | |
454 | * keep accessing it outside the said locks. This function may return | |
455 | * %NULL if @cgrp doesn't have @subsys_id enabled. | |
95109b62 TH |
456 | */ |
457 | static struct cgroup_subsys_state *cgroup_css(struct cgroup *cgrp, | |
ca8bdcaf | 458 | struct cgroup_subsys *ss) |
95109b62 | 459 | { |
ca8bdcaf | 460 | if (ss) |
aec25020 | 461 | return rcu_dereference_check(cgrp->subsys[ss->id], |
ace2bee8 | 462 | lockdep_is_held(&cgroup_mutex)); |
ca8bdcaf | 463 | else |
9d800df1 | 464 | return &cgrp->self; |
95109b62 | 465 | } |
42809dd4 | 466 | |
d41bf8c9 TH |
467 | /** |
468 | * cgroup_tryget_css - try to get a cgroup's css for the specified subsystem | |
469 | * @cgrp: the cgroup of interest | |
470 | * @ss: the subsystem of interest | |
471 | * | |
472 | * Find and get @cgrp's css assocaited with @ss. If the css doesn't exist | |
473 | * or is offline, %NULL is returned. | |
474 | */ | |
475 | static struct cgroup_subsys_state *cgroup_tryget_css(struct cgroup *cgrp, | |
476 | struct cgroup_subsys *ss) | |
477 | { | |
478 | struct cgroup_subsys_state *css; | |
479 | ||
480 | rcu_read_lock(); | |
481 | css = cgroup_css(cgrp, ss); | |
482 | if (!css || !css_tryget_online(css)) | |
483 | css = NULL; | |
484 | rcu_read_unlock(); | |
485 | ||
486 | return css; | |
487 | } | |
488 | ||
aec3dfcb TH |
489 | /** |
490 | * cgroup_e_css - obtain a cgroup's effective css for the specified subsystem | |
491 | * @cgrp: the cgroup of interest | |
9d800df1 | 492 | * @ss: the subsystem of interest (%NULL returns @cgrp->self) |
aec3dfcb | 493 | * |
d0f702e6 | 494 | * Similar to cgroup_css() but returns the effective css, which is defined |
aec3dfcb TH |
495 | * as the matching css of the nearest ancestor including self which has @ss |
496 | * enabled. If @ss is associated with the hierarchy @cgrp is on, this | |
497 | * function is guaranteed to return non-NULL css. | |
498 | */ | |
499 | static struct cgroup_subsys_state *cgroup_e_css(struct cgroup *cgrp, | |
500 | struct cgroup_subsys *ss) | |
501 | { | |
502 | lockdep_assert_held(&cgroup_mutex); | |
503 | ||
504 | if (!ss) | |
9d800df1 | 505 | return &cgrp->self; |
aec3dfcb | 506 | |
eeecbd19 TH |
507 | /* |
508 | * This function is used while updating css associations and thus | |
5531dc91 | 509 | * can't test the csses directly. Test ss_mask. |
eeecbd19 | 510 | */ |
5531dc91 | 511 | while (!(cgroup_ss_mask(cgrp) & (1 << ss->id))) { |
d51f39b0 | 512 | cgrp = cgroup_parent(cgrp); |
5531dc91 TH |
513 | if (!cgrp) |
514 | return NULL; | |
515 | } | |
aec3dfcb TH |
516 | |
517 | return cgroup_css(cgrp, ss); | |
95109b62 | 518 | } |
42809dd4 | 519 | |
eeecbd19 TH |
520 | /** |
521 | * cgroup_get_e_css - get a cgroup's effective css for the specified subsystem | |
522 | * @cgrp: the cgroup of interest | |
523 | * @ss: the subsystem of interest | |
524 | * | |
525 | * Find and get the effective css of @cgrp for @ss. The effective css is | |
526 | * defined as the matching css of the nearest ancestor including self which | |
527 | * has @ss enabled. If @ss is not mounted on the hierarchy @cgrp is on, | |
528 | * the root css is returned, so this function always returns a valid css. | |
529 | * The returned css must be put using css_put(). | |
530 | */ | |
531 | struct cgroup_subsys_state *cgroup_get_e_css(struct cgroup *cgrp, | |
532 | struct cgroup_subsys *ss) | |
533 | { | |
534 | struct cgroup_subsys_state *css; | |
535 | ||
536 | rcu_read_lock(); | |
537 | ||
538 | do { | |
539 | css = cgroup_css(cgrp, ss); | |
540 | ||
541 | if (css && css_tryget_online(css)) | |
542 | goto out_unlock; | |
543 | cgrp = cgroup_parent(cgrp); | |
544 | } while (cgrp); | |
545 | ||
546 | css = init_css_set.subsys[ss->id]; | |
547 | css_get(css); | |
548 | out_unlock: | |
549 | rcu_read_unlock(); | |
550 | return css; | |
551 | } | |
552 | ||
a590b90d | 553 | static void cgroup_get_live(struct cgroup *cgrp) |
052c3f3a TH |
554 | { |
555 | WARN_ON_ONCE(cgroup_is_dead(cgrp)); | |
556 | css_get(&cgrp->self); | |
557 | } | |
558 | ||
b4168640 | 559 | struct cgroup_subsys_state *of_css(struct kernfs_open_file *of) |
59f5296b | 560 | { |
2bd59d48 | 561 | struct cgroup *cgrp = of->kn->parent->priv; |
b4168640 | 562 | struct cftype *cft = of_cft(of); |
2bd59d48 TH |
563 | |
564 | /* | |
565 | * This is open and unprotected implementation of cgroup_css(). | |
566 | * seq_css() is only called from a kernfs file operation which has | |
567 | * an active reference on the file. Because all the subsystem | |
568 | * files are drained before a css is disassociated with a cgroup, | |
569 | * the matching css from the cgroup's subsys table is guaranteed to | |
570 | * be and stay valid until the enclosing operation is complete. | |
571 | */ | |
572 | if (cft->ss) | |
573 | return rcu_dereference_raw(cgrp->subsys[cft->ss->id]); | |
574 | else | |
9d800df1 | 575 | return &cgrp->self; |
59f5296b | 576 | } |
b4168640 | 577 | EXPORT_SYMBOL_GPL(of_css); |
59f5296b | 578 | |
1c6727af TH |
579 | /** |
580 | * for_each_css - iterate all css's of a cgroup | |
581 | * @css: the iteration cursor | |
582 | * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end | |
583 | * @cgrp: the target cgroup to iterate css's of | |
584 | * | |
aec3dfcb | 585 | * Should be called under cgroup_[tree_]mutex. |
1c6727af TH |
586 | */ |
587 | #define for_each_css(css, ssid, cgrp) \ | |
588 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \ | |
589 | if (!((css) = rcu_dereference_check( \ | |
590 | (cgrp)->subsys[(ssid)], \ | |
591 | lockdep_is_held(&cgroup_mutex)))) { } \ | |
592 | else | |
593 | ||
aec3dfcb TH |
594 | /** |
595 | * for_each_e_css - iterate all effective css's of a cgroup | |
596 | * @css: the iteration cursor | |
597 | * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end | |
598 | * @cgrp: the target cgroup to iterate css's of | |
599 | * | |
600 | * Should be called under cgroup_[tree_]mutex. | |
601 | */ | |
602 | #define for_each_e_css(css, ssid, cgrp) \ | |
603 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \ | |
604 | if (!((css) = cgroup_e_css(cgrp, cgroup_subsys[(ssid)]))) \ | |
605 | ; \ | |
606 | else | |
607 | ||
cb4a3167 | 608 | /** |
b4e0eeaf | 609 | * do_each_subsys_mask - filter for_each_subsys with a bitmask |
cb4a3167 AS |
610 | * @ss: the iteration cursor |
611 | * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end | |
b4e0eeaf | 612 | * @ss_mask: the bitmask |
cb4a3167 AS |
613 | * |
614 | * The block will only run for cases where the ssid-th bit (1 << ssid) of | |
b4e0eeaf | 615 | * @ss_mask is set. |
cb4a3167 | 616 | */ |
b4e0eeaf TH |
617 | #define do_each_subsys_mask(ss, ssid, ss_mask) do { \ |
618 | unsigned long __ss_mask = (ss_mask); \ | |
619 | if (!CGROUP_SUBSYS_COUNT) { /* to avoid spurious gcc warning */ \ | |
4a705c5c | 620 | (ssid) = 0; \ |
b4e0eeaf TH |
621 | break; \ |
622 | } \ | |
623 | for_each_set_bit(ssid, &__ss_mask, CGROUP_SUBSYS_COUNT) { \ | |
624 | (ss) = cgroup_subsys[ssid]; \ | |
625 | { | |
626 | ||
627 | #define while_each_subsys_mask() \ | |
628 | } \ | |
629 | } \ | |
630 | } while (false) | |
cb4a3167 | 631 | |
f8f22e53 TH |
632 | /* iterate over child cgrps, lock should be held throughout iteration */ |
633 | #define cgroup_for_each_live_child(child, cgrp) \ | |
d5c419b6 | 634 | list_for_each_entry((child), &(cgrp)->self.children, self.sibling) \ |
8353da1f | 635 | if (({ lockdep_assert_held(&cgroup_mutex); \ |
f8f22e53 TH |
636 | cgroup_is_dead(child); })) \ |
637 | ; \ | |
638 | else | |
7ae1bad9 | 639 | |
ce3f1d9d TH |
640 | /* walk live descendants in preorder */ |
641 | #define cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) \ | |
642 | css_for_each_descendant_pre((d_css), cgroup_css((cgrp), NULL)) \ | |
643 | if (({ lockdep_assert_held(&cgroup_mutex); \ | |
644 | (dsct) = (d_css)->cgroup; \ | |
645 | cgroup_is_dead(dsct); })) \ | |
646 | ; \ | |
647 | else | |
648 | ||
649 | /* walk live descendants in postorder */ | |
650 | #define cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) \ | |
651 | css_for_each_descendant_post((d_css), cgroup_css((cgrp), NULL)) \ | |
652 | if (({ lockdep_assert_held(&cgroup_mutex); \ | |
653 | (dsct) = (d_css)->cgroup; \ | |
654 | cgroup_is_dead(dsct); })) \ | |
655 | ; \ | |
656 | else | |
657 | ||
172a2c06 TH |
658 | /* |
659 | * The default css_set - used by init and its children prior to any | |
817929ec PM |
660 | * hierarchies being mounted. It contains a pointer to the root state |
661 | * for each subsystem. Also used to anchor the list of css_sets. Not | |
662 | * reference-counted, to improve performance when child cgroups | |
663 | * haven't been created. | |
664 | */ | |
5024ae29 | 665 | struct css_set init_css_set = { |
4b9502e6 | 666 | .refcount = REFCOUNT_INIT(1), |
454000ad | 667 | .dom_cset = &init_css_set, |
172a2c06 TH |
668 | .tasks = LIST_HEAD_INIT(init_css_set.tasks), |
669 | .mg_tasks = LIST_HEAD_INIT(init_css_set.mg_tasks), | |
798cc880 | 670 | .dying_tasks = LIST_HEAD_INIT(init_css_set.dying_tasks), |
5f617ebb | 671 | .task_iters = LIST_HEAD_INIT(init_css_set.task_iters), |
454000ad | 672 | .threaded_csets = LIST_HEAD_INIT(init_css_set.threaded_csets), |
5f617ebb | 673 | .cgrp_links = LIST_HEAD_INIT(init_css_set.cgrp_links), |
172a2c06 TH |
674 | .mg_preload_node = LIST_HEAD_INIT(init_css_set.mg_preload_node), |
675 | .mg_node = LIST_HEAD_INIT(init_css_set.mg_node), | |
38683148 TH |
676 | |
677 | /* | |
678 | * The following field is re-initialized when this cset gets linked | |
679 | * in cgroup_init(). However, let's initialize the field | |
680 | * statically too so that the default cgroup can be accessed safely | |
681 | * early during boot. | |
682 | */ | |
683 | .dfl_cgrp = &cgrp_dfl_root.cgrp, | |
172a2c06 | 684 | }; |
817929ec | 685 | |
172a2c06 | 686 | static int css_set_count = 1; /* 1 for init_css_set */ |
817929ec | 687 | |
454000ad TH |
688 | static bool css_set_threaded(struct css_set *cset) |
689 | { | |
690 | return cset->dom_cset != cset; | |
691 | } | |
692 | ||
0de0942d TH |
693 | /** |
694 | * css_set_populated - does a css_set contain any tasks? | |
695 | * @cset: target css_set | |
73a7242a WL |
696 | * |
697 | * css_set_populated() should be the same as !!cset->nr_tasks at steady | |
698 | * state. However, css_set_populated() can be called while a task is being | |
699 | * added to or removed from the linked list before the nr_tasks is | |
700 | * properly updated. Hence, we can't just look at ->nr_tasks here. | |
0de0942d TH |
701 | */ |
702 | static bool css_set_populated(struct css_set *cset) | |
703 | { | |
f0d9a5f1 | 704 | lockdep_assert_held(&css_set_lock); |
0de0942d TH |
705 | |
706 | return !list_empty(&cset->tasks) || !list_empty(&cset->mg_tasks); | |
707 | } | |
708 | ||
842b597e | 709 | /** |
788b950c | 710 | * cgroup_update_populated - update the populated count of a cgroup |
842b597e TH |
711 | * @cgrp: the target cgroup |
712 | * @populated: inc or dec populated count | |
713 | * | |
0de0942d | 714 | * One of the css_sets associated with @cgrp is either getting its first |
788b950c TH |
715 | * task or losing the last. Update @cgrp->nr_populated_* accordingly. The |
716 | * count is propagated towards root so that a given cgroup's | |
717 | * nr_populated_children is zero iff none of its descendants contain any | |
718 | * tasks. | |
842b597e | 719 | * |
788b950c TH |
720 | * @cgrp's interface file "cgroup.populated" is zero if both |
721 | * @cgrp->nr_populated_csets and @cgrp->nr_populated_children are zero and | |
722 | * 1 otherwise. When the sum changes from or to zero, userland is notified | |
723 | * that the content of the interface file has changed. This can be used to | |
724 | * detect when @cgrp and its descendants become populated or empty. | |
842b597e TH |
725 | */ |
726 | static void cgroup_update_populated(struct cgroup *cgrp, bool populated) | |
727 | { | |
788b950c TH |
728 | struct cgroup *child = NULL; |
729 | int adj = populated ? 1 : -1; | |
730 | ||
f0d9a5f1 | 731 | lockdep_assert_held(&css_set_lock); |
842b597e TH |
732 | |
733 | do { | |
788b950c | 734 | bool was_populated = cgroup_is_populated(cgrp); |
842b597e | 735 | |
454000ad | 736 | if (!child) { |
788b950c | 737 | cgrp->nr_populated_csets += adj; |
454000ad TH |
738 | } else { |
739 | if (cgroup_is_threaded(child)) | |
740 | cgrp->nr_populated_threaded_children += adj; | |
741 | else | |
742 | cgrp->nr_populated_domain_children += adj; | |
743 | } | |
842b597e | 744 | |
788b950c | 745 | if (was_populated == cgroup_is_populated(cgrp)) |
842b597e TH |
746 | break; |
747 | ||
d62beb7f | 748 | cgroup1_check_for_release(cgrp); |
6f60eade TH |
749 | cgroup_file_notify(&cgrp->events_file); |
750 | ||
788b950c | 751 | child = cgrp; |
d51f39b0 | 752 | cgrp = cgroup_parent(cgrp); |
842b597e TH |
753 | } while (cgrp); |
754 | } | |
755 | ||
0de0942d TH |
756 | /** |
757 | * css_set_update_populated - update populated state of a css_set | |
758 | * @cset: target css_set | |
759 | * @populated: whether @cset is populated or depopulated | |
760 | * | |
761 | * @cset is either getting the first task or losing the last. Update the | |
788b950c | 762 | * populated counters of all associated cgroups accordingly. |
0de0942d TH |
763 | */ |
764 | static void css_set_update_populated(struct css_set *cset, bool populated) | |
765 | { | |
766 | struct cgrp_cset_link *link; | |
767 | ||
f0d9a5f1 | 768 | lockdep_assert_held(&css_set_lock); |
0de0942d TH |
769 | |
770 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) | |
771 | cgroup_update_populated(link->cgrp, populated); | |
772 | } | |
773 | ||
3995c283 TH |
774 | /* |
775 | * @task is leaving, advance task iterators which are pointing to it so | |
776 | * that they can resume at the next position. Advancing an iterator might | |
777 | * remove it from the list, use safe walk. See css_task_iter_skip() for | |
778 | * details. | |
779 | */ | |
780 | static void css_set_skip_task_iters(struct css_set *cset, | |
781 | struct task_struct *task) | |
782 | { | |
783 | struct css_task_iter *it, *pos; | |
784 | ||
785 | list_for_each_entry_safe(it, pos, &cset->task_iters, iters_node) | |
786 | css_task_iter_skip(it, task); | |
787 | } | |
788 | ||
f6d7d049 TH |
789 | /** |
790 | * css_set_move_task - move a task from one css_set to another | |
791 | * @task: task being moved | |
792 | * @from_cset: css_set @task currently belongs to (may be NULL) | |
793 | * @to_cset: new css_set @task is being moved to (may be NULL) | |
794 | * @use_mg_tasks: move to @to_cset->mg_tasks instead of ->tasks | |
795 | * | |
796 | * Move @task from @from_cset to @to_cset. If @task didn't belong to any | |
797 | * css_set, @from_cset can be NULL. If @task is being disassociated | |
798 | * instead of moved, @to_cset can be NULL. | |
799 | * | |
788b950c | 800 | * This function automatically handles populated counter updates and |
ed27b9f7 TH |
801 | * css_task_iter adjustments but the caller is responsible for managing |
802 | * @from_cset and @to_cset's reference counts. | |
f6d7d049 TH |
803 | */ |
804 | static void css_set_move_task(struct task_struct *task, | |
805 | struct css_set *from_cset, struct css_set *to_cset, | |
806 | bool use_mg_tasks) | |
807 | { | |
f0d9a5f1 | 808 | lockdep_assert_held(&css_set_lock); |
f6d7d049 | 809 | |
20b454a6 TH |
810 | if (to_cset && !css_set_populated(to_cset)) |
811 | css_set_update_populated(to_cset, true); | |
812 | ||
f6d7d049 TH |
813 | if (from_cset) { |
814 | WARN_ON_ONCE(list_empty(&task->cg_list)); | |
ed27b9f7 | 815 | |
3995c283 | 816 | css_set_skip_task_iters(from_cset, task); |
f6d7d049 TH |
817 | list_del_init(&task->cg_list); |
818 | if (!css_set_populated(from_cset)) | |
819 | css_set_update_populated(from_cset, false); | |
820 | } else { | |
821 | WARN_ON_ONCE(!list_empty(&task->cg_list)); | |
822 | } | |
823 | ||
824 | if (to_cset) { | |
825 | /* | |
826 | * We are synchronized through cgroup_threadgroup_rwsem | |
827 | * against PF_EXITING setting such that we can't race | |
828 | * against cgroup_exit() changing the css_set to | |
829 | * init_css_set and dropping the old one. | |
830 | */ | |
831 | WARN_ON_ONCE(task->flags & PF_EXITING); | |
832 | ||
f6d7d049 TH |
833 | rcu_assign_pointer(task->cgroups, to_cset); |
834 | list_add_tail(&task->cg_list, use_mg_tasks ? &to_cset->mg_tasks : | |
835 | &to_cset->tasks); | |
836 | } | |
837 | } | |
838 | ||
7717f7ba PM |
839 | /* |
840 | * hash table for cgroup groups. This improves the performance to find | |
841 | * an existing css_set. This hash doesn't (currently) take into | |
842 | * account cgroups in empty hierarchies. | |
843 | */ | |
472b1053 | 844 | #define CSS_SET_HASH_BITS 7 |
0ac801fe | 845 | static DEFINE_HASHTABLE(css_set_table, CSS_SET_HASH_BITS); |
472b1053 | 846 | |
0ac801fe | 847 | static unsigned long css_set_hash(struct cgroup_subsys_state *css[]) |
472b1053 | 848 | { |
0ac801fe | 849 | unsigned long key = 0UL; |
30159ec7 TH |
850 | struct cgroup_subsys *ss; |
851 | int i; | |
472b1053 | 852 | |
30159ec7 | 853 | for_each_subsys(ss, i) |
0ac801fe LZ |
854 | key += (unsigned long)css[i]; |
855 | key = (key >> 16) ^ key; | |
472b1053 | 856 | |
0ac801fe | 857 | return key; |
472b1053 LZ |
858 | } |
859 | ||
dcfe149b | 860 | void put_css_set_locked(struct css_set *cset) |
b4f48b63 | 861 | { |
69d0206c | 862 | struct cgrp_cset_link *link, *tmp_link; |
2d8f243a TH |
863 | struct cgroup_subsys *ss; |
864 | int ssid; | |
5abb8855 | 865 | |
f0d9a5f1 | 866 | lockdep_assert_held(&css_set_lock); |
89c5509b | 867 | |
4b9502e6 | 868 | if (!refcount_dec_and_test(&cset->refcount)) |
146aa1bd | 869 | return; |
81a6a5cd | 870 | |
454000ad TH |
871 | WARN_ON_ONCE(!list_empty(&cset->threaded_csets)); |
872 | ||
53254f90 TH |
873 | /* This css_set is dead. unlink it and release cgroup and css refs */ |
874 | for_each_subsys(ss, ssid) { | |
2d8f243a | 875 | list_del(&cset->e_cset_node[ssid]); |
53254f90 TH |
876 | css_put(cset->subsys[ssid]); |
877 | } | |
5abb8855 | 878 | hash_del(&cset->hlist); |
2c6ab6d2 PM |
879 | css_set_count--; |
880 | ||
69d0206c | 881 | list_for_each_entry_safe(link, tmp_link, &cset->cgrp_links, cgrp_link) { |
69d0206c TH |
882 | list_del(&link->cset_link); |
883 | list_del(&link->cgrp_link); | |
2ceb231b TH |
884 | if (cgroup_parent(link->cgrp)) |
885 | cgroup_put(link->cgrp); | |
2c6ab6d2 | 886 | kfree(link); |
81a6a5cd | 887 | } |
2c6ab6d2 | 888 | |
454000ad TH |
889 | if (css_set_threaded(cset)) { |
890 | list_del(&cset->threaded_csets_node); | |
891 | put_css_set_locked(cset->dom_cset); | |
892 | } | |
893 | ||
5abb8855 | 894 | kfree_rcu(cset, rcu_head); |
b4f48b63 PM |
895 | } |
896 | ||
b326f9d0 | 897 | /** |
7717f7ba | 898 | * compare_css_sets - helper function for find_existing_css_set(). |
5abb8855 TH |
899 | * @cset: candidate css_set being tested |
900 | * @old_cset: existing css_set for a task | |
7717f7ba PM |
901 | * @new_cgrp: cgroup that's being entered by the task |
902 | * @template: desired set of css pointers in css_set (pre-calculated) | |
903 | * | |
6f4b7e63 | 904 | * Returns true if "cset" matches "old_cset" except for the hierarchy |
7717f7ba PM |
905 | * which "new_cgrp" belongs to, for which it should match "new_cgrp". |
906 | */ | |
5abb8855 TH |
907 | static bool compare_css_sets(struct css_set *cset, |
908 | struct css_set *old_cset, | |
7717f7ba PM |
909 | struct cgroup *new_cgrp, |
910 | struct cgroup_subsys_state *template[]) | |
911 | { | |
454000ad | 912 | struct cgroup *new_dfl_cgrp; |
7717f7ba PM |
913 | struct list_head *l1, *l2; |
914 | ||
aec3dfcb TH |
915 | /* |
916 | * On the default hierarchy, there can be csets which are | |
917 | * associated with the same set of cgroups but different csses. | |
918 | * Let's first ensure that csses match. | |
919 | */ | |
920 | if (memcmp(template, cset->subsys, sizeof(cset->subsys))) | |
7717f7ba | 921 | return false; |
7717f7ba | 922 | |
454000ad TH |
923 | |
924 | /* @cset's domain should match the default cgroup's */ | |
925 | if (cgroup_on_dfl(new_cgrp)) | |
926 | new_dfl_cgrp = new_cgrp; | |
927 | else | |
928 | new_dfl_cgrp = old_cset->dfl_cgrp; | |
929 | ||
930 | if (new_dfl_cgrp->dom_cgrp != cset->dom_cset->dfl_cgrp) | |
931 | return false; | |
932 | ||
7717f7ba PM |
933 | /* |
934 | * Compare cgroup pointers in order to distinguish between | |
aec3dfcb TH |
935 | * different cgroups in hierarchies. As different cgroups may |
936 | * share the same effective css, this comparison is always | |
937 | * necessary. | |
7717f7ba | 938 | */ |
69d0206c TH |
939 | l1 = &cset->cgrp_links; |
940 | l2 = &old_cset->cgrp_links; | |
7717f7ba | 941 | while (1) { |
69d0206c | 942 | struct cgrp_cset_link *link1, *link2; |
5abb8855 | 943 | struct cgroup *cgrp1, *cgrp2; |
7717f7ba PM |
944 | |
945 | l1 = l1->next; | |
946 | l2 = l2->next; | |
947 | /* See if we reached the end - both lists are equal length. */ | |
69d0206c TH |
948 | if (l1 == &cset->cgrp_links) { |
949 | BUG_ON(l2 != &old_cset->cgrp_links); | |
7717f7ba PM |
950 | break; |
951 | } else { | |
69d0206c | 952 | BUG_ON(l2 == &old_cset->cgrp_links); |
7717f7ba PM |
953 | } |
954 | /* Locate the cgroups associated with these links. */ | |
69d0206c TH |
955 | link1 = list_entry(l1, struct cgrp_cset_link, cgrp_link); |
956 | link2 = list_entry(l2, struct cgrp_cset_link, cgrp_link); | |
957 | cgrp1 = link1->cgrp; | |
958 | cgrp2 = link2->cgrp; | |
7717f7ba | 959 | /* Hierarchies should be linked in the same order. */ |
5abb8855 | 960 | BUG_ON(cgrp1->root != cgrp2->root); |
7717f7ba PM |
961 | |
962 | /* | |
963 | * If this hierarchy is the hierarchy of the cgroup | |
964 | * that's changing, then we need to check that this | |
965 | * css_set points to the new cgroup; if it's any other | |
966 | * hierarchy, then this css_set should point to the | |
967 | * same cgroup as the old css_set. | |
968 | */ | |
5abb8855 TH |
969 | if (cgrp1->root == new_cgrp->root) { |
970 | if (cgrp1 != new_cgrp) | |
7717f7ba PM |
971 | return false; |
972 | } else { | |
5abb8855 | 973 | if (cgrp1 != cgrp2) |
7717f7ba PM |
974 | return false; |
975 | } | |
976 | } | |
977 | return true; | |
978 | } | |
979 | ||
b326f9d0 TH |
980 | /** |
981 | * find_existing_css_set - init css array and find the matching css_set | |
982 | * @old_cset: the css_set that we're using before the cgroup transition | |
983 | * @cgrp: the cgroup that we're moving into | |
984 | * @template: out param for the new set of csses, should be clear on entry | |
817929ec | 985 | */ |
5abb8855 TH |
986 | static struct css_set *find_existing_css_set(struct css_set *old_cset, |
987 | struct cgroup *cgrp, | |
988 | struct cgroup_subsys_state *template[]) | |
b4f48b63 | 989 | { |
3dd06ffa | 990 | struct cgroup_root *root = cgrp->root; |
30159ec7 | 991 | struct cgroup_subsys *ss; |
5abb8855 | 992 | struct css_set *cset; |
0ac801fe | 993 | unsigned long key; |
b326f9d0 | 994 | int i; |
817929ec | 995 | |
aae8aab4 BB |
996 | /* |
997 | * Build the set of subsystem state objects that we want to see in the | |
998 | * new css_set. while subsystems can change globally, the entries here | |
999 | * won't change, so no need for locking. | |
1000 | */ | |
30159ec7 | 1001 | for_each_subsys(ss, i) { |
f392e51c | 1002 | if (root->subsys_mask & (1UL << i)) { |
aec3dfcb TH |
1003 | /* |
1004 | * @ss is in this hierarchy, so we want the | |
1005 | * effective css from @cgrp. | |
1006 | */ | |
1007 | template[i] = cgroup_e_css(cgrp, ss); | |
817929ec | 1008 | } else { |
aec3dfcb TH |
1009 | /* |
1010 | * @ss is not in this hierarchy, so we don't want | |
1011 | * to change the css. | |
1012 | */ | |
5abb8855 | 1013 | template[i] = old_cset->subsys[i]; |
817929ec PM |
1014 | } |
1015 | } | |
1016 | ||
0ac801fe | 1017 | key = css_set_hash(template); |
5abb8855 TH |
1018 | hash_for_each_possible(css_set_table, cset, hlist, key) { |
1019 | if (!compare_css_sets(cset, old_cset, cgrp, template)) | |
7717f7ba PM |
1020 | continue; |
1021 | ||
1022 | /* This css_set matches what we need */ | |
5abb8855 | 1023 | return cset; |
472b1053 | 1024 | } |
817929ec PM |
1025 | |
1026 | /* No existing cgroup group matched */ | |
1027 | return NULL; | |
1028 | } | |
1029 | ||
69d0206c | 1030 | static void free_cgrp_cset_links(struct list_head *links_to_free) |
36553434 | 1031 | { |
69d0206c | 1032 | struct cgrp_cset_link *link, *tmp_link; |
36553434 | 1033 | |
69d0206c TH |
1034 | list_for_each_entry_safe(link, tmp_link, links_to_free, cset_link) { |
1035 | list_del(&link->cset_link); | |
36553434 LZ |
1036 | kfree(link); |
1037 | } | |
1038 | } | |
1039 | ||
69d0206c TH |
1040 | /** |
1041 | * allocate_cgrp_cset_links - allocate cgrp_cset_links | |
1042 | * @count: the number of links to allocate | |
1043 | * @tmp_links: list_head the allocated links are put on | |
1044 | * | |
1045 | * Allocate @count cgrp_cset_link structures and chain them on @tmp_links | |
1046 | * through ->cset_link. Returns 0 on success or -errno. | |
817929ec | 1047 | */ |
69d0206c | 1048 | static int allocate_cgrp_cset_links(int count, struct list_head *tmp_links) |
817929ec | 1049 | { |
69d0206c | 1050 | struct cgrp_cset_link *link; |
817929ec | 1051 | int i; |
69d0206c TH |
1052 | |
1053 | INIT_LIST_HEAD(tmp_links); | |
1054 | ||
817929ec | 1055 | for (i = 0; i < count; i++) { |
f4f4be2b | 1056 | link = kzalloc(sizeof(*link), GFP_KERNEL); |
817929ec | 1057 | if (!link) { |
69d0206c | 1058 | free_cgrp_cset_links(tmp_links); |
817929ec PM |
1059 | return -ENOMEM; |
1060 | } | |
69d0206c | 1061 | list_add(&link->cset_link, tmp_links); |
817929ec PM |
1062 | } |
1063 | return 0; | |
1064 | } | |
1065 | ||
c12f65d4 LZ |
1066 | /** |
1067 | * link_css_set - a helper function to link a css_set to a cgroup | |
69d0206c | 1068 | * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links() |
5abb8855 | 1069 | * @cset: the css_set to be linked |
c12f65d4 LZ |
1070 | * @cgrp: the destination cgroup |
1071 | */ | |
69d0206c TH |
1072 | static void link_css_set(struct list_head *tmp_links, struct css_set *cset, |
1073 | struct cgroup *cgrp) | |
c12f65d4 | 1074 | { |
69d0206c | 1075 | struct cgrp_cset_link *link; |
c12f65d4 | 1076 | |
69d0206c | 1077 | BUG_ON(list_empty(tmp_links)); |
6803c006 TH |
1078 | |
1079 | if (cgroup_on_dfl(cgrp)) | |
1080 | cset->dfl_cgrp = cgrp; | |
1081 | ||
69d0206c TH |
1082 | link = list_first_entry(tmp_links, struct cgrp_cset_link, cset_link); |
1083 | link->cset = cset; | |
7717f7ba | 1084 | link->cgrp = cgrp; |
842b597e | 1085 | |
7717f7ba | 1086 | /* |
389b9c1b TH |
1087 | * Always add links to the tail of the lists so that the lists are |
1088 | * in choronological order. | |
7717f7ba | 1089 | */ |
389b9c1b | 1090 | list_move_tail(&link->cset_link, &cgrp->cset_links); |
69d0206c | 1091 | list_add_tail(&link->cgrp_link, &cset->cgrp_links); |
2ceb231b TH |
1092 | |
1093 | if (cgroup_parent(cgrp)) | |
a590b90d | 1094 | cgroup_get_live(cgrp); |
c12f65d4 LZ |
1095 | } |
1096 | ||
b326f9d0 TH |
1097 | /** |
1098 | * find_css_set - return a new css_set with one cgroup updated | |
1099 | * @old_cset: the baseline css_set | |
1100 | * @cgrp: the cgroup to be updated | |
1101 | * | |
1102 | * Return a new css_set that's equivalent to @old_cset, but with @cgrp | |
1103 | * substituted into the appropriate hierarchy. | |
817929ec | 1104 | */ |
5abb8855 TH |
1105 | static struct css_set *find_css_set(struct css_set *old_cset, |
1106 | struct cgroup *cgrp) | |
817929ec | 1107 | { |
b326f9d0 | 1108 | struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT] = { }; |
5abb8855 | 1109 | struct css_set *cset; |
69d0206c TH |
1110 | struct list_head tmp_links; |
1111 | struct cgrp_cset_link *link; | |
2d8f243a | 1112 | struct cgroup_subsys *ss; |
0ac801fe | 1113 | unsigned long key; |
2d8f243a | 1114 | int ssid; |
472b1053 | 1115 | |
b326f9d0 TH |
1116 | lockdep_assert_held(&cgroup_mutex); |
1117 | ||
817929ec PM |
1118 | /* First see if we already have a cgroup group that matches |
1119 | * the desired set */ | |
82d6489d | 1120 | spin_lock_irq(&css_set_lock); |
5abb8855 TH |
1121 | cset = find_existing_css_set(old_cset, cgrp, template); |
1122 | if (cset) | |
1123 | get_css_set(cset); | |
82d6489d | 1124 | spin_unlock_irq(&css_set_lock); |
817929ec | 1125 | |
5abb8855 TH |
1126 | if (cset) |
1127 | return cset; | |
817929ec | 1128 | |
f4f4be2b | 1129 | cset = kzalloc(sizeof(*cset), GFP_KERNEL); |
5abb8855 | 1130 | if (!cset) |
817929ec PM |
1131 | return NULL; |
1132 | ||
69d0206c | 1133 | /* Allocate all the cgrp_cset_link objects that we'll need */ |
9871bf95 | 1134 | if (allocate_cgrp_cset_links(cgroup_root_count, &tmp_links) < 0) { |
5abb8855 | 1135 | kfree(cset); |
817929ec PM |
1136 | return NULL; |
1137 | } | |
1138 | ||
4b9502e6 | 1139 | refcount_set(&cset->refcount, 1); |
454000ad | 1140 | cset->dom_cset = cset; |
5abb8855 | 1141 | INIT_LIST_HEAD(&cset->tasks); |
c7561128 | 1142 | INIT_LIST_HEAD(&cset->mg_tasks); |
798cc880 | 1143 | INIT_LIST_HEAD(&cset->dying_tasks); |
ed27b9f7 | 1144 | INIT_LIST_HEAD(&cset->task_iters); |
454000ad | 1145 | INIT_LIST_HEAD(&cset->threaded_csets); |
5abb8855 | 1146 | INIT_HLIST_NODE(&cset->hlist); |
5f617ebb TH |
1147 | INIT_LIST_HEAD(&cset->cgrp_links); |
1148 | INIT_LIST_HEAD(&cset->mg_preload_node); | |
1149 | INIT_LIST_HEAD(&cset->mg_node); | |
817929ec PM |
1150 | |
1151 | /* Copy the set of subsystem state objects generated in | |
1152 | * find_existing_css_set() */ | |
5abb8855 | 1153 | memcpy(cset->subsys, template, sizeof(cset->subsys)); |
817929ec | 1154 | |
82d6489d | 1155 | spin_lock_irq(&css_set_lock); |
817929ec | 1156 | /* Add reference counts and links from the new css_set. */ |
69d0206c | 1157 | list_for_each_entry(link, &old_cset->cgrp_links, cgrp_link) { |
7717f7ba | 1158 | struct cgroup *c = link->cgrp; |
69d0206c | 1159 | |
7717f7ba PM |
1160 | if (c->root == cgrp->root) |
1161 | c = cgrp; | |
69d0206c | 1162 | link_css_set(&tmp_links, cset, c); |
7717f7ba | 1163 | } |
817929ec | 1164 | |
69d0206c | 1165 | BUG_ON(!list_empty(&tmp_links)); |
817929ec | 1166 | |
817929ec | 1167 | css_set_count++; |
472b1053 | 1168 | |
2d8f243a | 1169 | /* Add @cset to the hash table */ |
5abb8855 TH |
1170 | key = css_set_hash(cset->subsys); |
1171 | hash_add(css_set_table, &cset->hlist, key); | |
472b1053 | 1172 | |
53254f90 TH |
1173 | for_each_subsys(ss, ssid) { |
1174 | struct cgroup_subsys_state *css = cset->subsys[ssid]; | |
1175 | ||
2d8f243a | 1176 | list_add_tail(&cset->e_cset_node[ssid], |
53254f90 TH |
1177 | &css->cgroup->e_csets[ssid]); |
1178 | css_get(css); | |
1179 | } | |
2d8f243a | 1180 | |
82d6489d | 1181 | spin_unlock_irq(&css_set_lock); |
817929ec | 1182 | |
454000ad TH |
1183 | /* |
1184 | * If @cset should be threaded, look up the matching dom_cset and | |
1185 | * link them up. We first fully initialize @cset then look for the | |
1186 | * dom_cset. It's simpler this way and safe as @cset is guaranteed | |
1187 | * to stay empty until we return. | |
1188 | */ | |
1189 | if (cgroup_is_threaded(cset->dfl_cgrp)) { | |
1190 | struct css_set *dcset; | |
1191 | ||
1192 | dcset = find_css_set(cset, cset->dfl_cgrp->dom_cgrp); | |
1193 | if (!dcset) { | |
1194 | put_css_set(cset); | |
1195 | return NULL; | |
1196 | } | |
1197 | ||
1198 | spin_lock_irq(&css_set_lock); | |
1199 | cset->dom_cset = dcset; | |
1200 | list_add_tail(&cset->threaded_csets_node, | |
1201 | &dcset->threaded_csets); | |
1202 | spin_unlock_irq(&css_set_lock); | |
1203 | } | |
1204 | ||
5abb8855 | 1205 | return cset; |
b4f48b63 PM |
1206 | } |
1207 | ||
0a268dbd | 1208 | struct cgroup_root *cgroup_root_from_kf(struct kernfs_root *kf_root) |
7717f7ba | 1209 | { |
3dd06ffa | 1210 | struct cgroup *root_cgrp = kf_root->kn->priv; |
2bd59d48 | 1211 | |
3dd06ffa | 1212 | return root_cgrp->root; |
2bd59d48 TH |
1213 | } |
1214 | ||
3dd06ffa | 1215 | static int cgroup_init_root_id(struct cgroup_root *root) |
f2e85d57 TH |
1216 | { |
1217 | int id; | |
1218 | ||
1219 | lockdep_assert_held(&cgroup_mutex); | |
1220 | ||
985ed670 | 1221 | id = idr_alloc_cyclic(&cgroup_hierarchy_idr, root, 0, 0, GFP_KERNEL); |
f2e85d57 TH |
1222 | if (id < 0) |
1223 | return id; | |
1224 | ||
1225 | root->hierarchy_id = id; | |
1226 | return 0; | |
1227 | } | |
1228 | ||
3dd06ffa | 1229 | static void cgroup_exit_root_id(struct cgroup_root *root) |
f2e85d57 TH |
1230 | { |
1231 | lockdep_assert_held(&cgroup_mutex); | |
1232 | ||
8c8a5502 | 1233 | idr_remove(&cgroup_hierarchy_idr, root->hierarchy_id); |
f2e85d57 TH |
1234 | } |
1235 | ||
1592c9b2 | 1236 | void cgroup_free_root(struct cgroup_root *root) |
f2e85d57 TH |
1237 | { |
1238 | if (root) { | |
f2e85d57 TH |
1239 | idr_destroy(&root->cgroup_idr); |
1240 | kfree(root); | |
1241 | } | |
1242 | } | |
1243 | ||
3dd06ffa | 1244 | static void cgroup_destroy_root(struct cgroup_root *root) |
59f5296b | 1245 | { |
3dd06ffa | 1246 | struct cgroup *cgrp = &root->cgrp; |
f2e85d57 | 1247 | struct cgrp_cset_link *link, *tmp_link; |
f2e85d57 | 1248 | |
ed1777de TH |
1249 | trace_cgroup_destroy_root(root); |
1250 | ||
334c3679 | 1251 | cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp); |
f2e85d57 | 1252 | |
776f02fa | 1253 | BUG_ON(atomic_read(&root->nr_cgrps)); |
d5c419b6 | 1254 | BUG_ON(!list_empty(&cgrp->self.children)); |
f2e85d57 | 1255 | |
f2e85d57 | 1256 | /* Rebind all subsystems back to the default hierarchy */ |
334c3679 | 1257 | WARN_ON(rebind_subsystems(&cgrp_dfl_root, root->subsys_mask)); |
7717f7ba | 1258 | |
7717f7ba | 1259 | /* |
f2e85d57 TH |
1260 | * Release all the links from cset_links to this hierarchy's |
1261 | * root cgroup | |
7717f7ba | 1262 | */ |
82d6489d | 1263 | spin_lock_irq(&css_set_lock); |
f2e85d57 TH |
1264 | |
1265 | list_for_each_entry_safe(link, tmp_link, &cgrp->cset_links, cset_link) { | |
1266 | list_del(&link->cset_link); | |
1267 | list_del(&link->cgrp_link); | |
1268 | kfree(link); | |
1269 | } | |
f0d9a5f1 | 1270 | |
82d6489d | 1271 | spin_unlock_irq(&css_set_lock); |
f2e85d57 TH |
1272 | |
1273 | if (!list_empty(&root->root_list)) { | |
1274 | list_del(&root->root_list); | |
1275 | cgroup_root_count--; | |
1276 | } | |
1277 | ||
1278 | cgroup_exit_root_id(root); | |
1279 | ||
1280 | mutex_unlock(&cgroup_mutex); | |
f2e85d57 | 1281 | |
2bd59d48 | 1282 | kernfs_destroy_root(root->kf_root); |
f2e85d57 TH |
1283 | cgroup_free_root(root); |
1284 | } | |
1285 | ||
4f41fc59 SH |
1286 | /* |
1287 | * look up cgroup associated with current task's cgroup namespace on the | |
1288 | * specified hierarchy | |
1289 | */ | |
1290 | static struct cgroup * | |
1291 | current_cgns_cgroup_from_root(struct cgroup_root *root) | |
1292 | { | |
1293 | struct cgroup *res = NULL; | |
1294 | struct css_set *cset; | |
1295 | ||
1296 | lockdep_assert_held(&css_set_lock); | |
1297 | ||
1298 | rcu_read_lock(); | |
1299 | ||
1300 | cset = current->nsproxy->cgroup_ns->root_cset; | |
1301 | if (cset == &init_css_set) { | |
1302 | res = &root->cgrp; | |
1303 | } else { | |
1304 | struct cgrp_cset_link *link; | |
1305 | ||
1306 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) { | |
1307 | struct cgroup *c = link->cgrp; | |
1308 | ||
1309 | if (c->root == root) { | |
1310 | res = c; | |
1311 | break; | |
1312 | } | |
1313 | } | |
1314 | } | |
1315 | rcu_read_unlock(); | |
1316 | ||
1317 | BUG_ON(!res); | |
1318 | return res; | |
1319 | } | |
1320 | ||
ceb6a081 TH |
1321 | /* look up cgroup associated with given css_set on the specified hierarchy */ |
1322 | static struct cgroup *cset_cgroup_from_root(struct css_set *cset, | |
3dd06ffa | 1323 | struct cgroup_root *root) |
7717f7ba | 1324 | { |
7717f7ba PM |
1325 | struct cgroup *res = NULL; |
1326 | ||
96d365e0 | 1327 | lockdep_assert_held(&cgroup_mutex); |
f0d9a5f1 | 1328 | lockdep_assert_held(&css_set_lock); |
96d365e0 | 1329 | |
5abb8855 | 1330 | if (cset == &init_css_set) { |
3dd06ffa | 1331 | res = &root->cgrp; |
13d82fb7 TH |
1332 | } else if (root == &cgrp_dfl_root) { |
1333 | res = cset->dfl_cgrp; | |
7717f7ba | 1334 | } else { |
69d0206c TH |
1335 | struct cgrp_cset_link *link; |
1336 | ||
1337 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) { | |
7717f7ba | 1338 | struct cgroup *c = link->cgrp; |
69d0206c | 1339 | |
7717f7ba PM |
1340 | if (c->root == root) { |
1341 | res = c; | |
1342 | break; | |
1343 | } | |
1344 | } | |
1345 | } | |
96d365e0 | 1346 | |
7717f7ba PM |
1347 | BUG_ON(!res); |
1348 | return res; | |
1349 | } | |
1350 | ||
ddbcc7e8 | 1351 | /* |
ceb6a081 | 1352 | * Return the cgroup for "task" from the given hierarchy. Must be |
f0d9a5f1 | 1353 | * called with cgroup_mutex and css_set_lock held. |
ceb6a081 | 1354 | */ |
0a268dbd TH |
1355 | struct cgroup *task_cgroup_from_root(struct task_struct *task, |
1356 | struct cgroup_root *root) | |
ceb6a081 TH |
1357 | { |
1358 | /* | |
1359 | * No need to lock the task - since we hold cgroup_mutex the | |
1360 | * task can't change groups, so the only thing that can happen | |
1361 | * is that it exits and its css is set back to init_css_set. | |
1362 | */ | |
1363 | return cset_cgroup_from_root(task_css_set(task), root); | |
1364 | } | |
1365 | ||
ddbcc7e8 | 1366 | /* |
ddbcc7e8 PM |
1367 | * A task must hold cgroup_mutex to modify cgroups. |
1368 | * | |
1369 | * Any task can increment and decrement the count field without lock. | |
1370 | * So in general, code holding cgroup_mutex can't rely on the count | |
1371 | * field not changing. However, if the count goes to zero, then only | |
956db3ca | 1372 | * cgroup_attach_task() can increment it again. Because a count of zero |
ddbcc7e8 PM |
1373 | * means that no tasks are currently attached, therefore there is no |
1374 | * way a task attached to that cgroup can fork (the other way to | |
1375 | * increment the count). So code holding cgroup_mutex can safely | |
1376 | * assume that if the count is zero, it will stay zero. Similarly, if | |
1377 | * a task holds cgroup_mutex on a cgroup with zero count, it | |
1378 | * knows that the cgroup won't be removed, as cgroup_rmdir() | |
1379 | * needs that mutex. | |
1380 | * | |
ddbcc7e8 PM |
1381 | * A cgroup can only be deleted if both its 'count' of using tasks |
1382 | * is zero, and its list of 'children' cgroups is empty. Since all | |
1383 | * tasks in the system use _some_ cgroup, and since there is always at | |
3dd06ffa | 1384 | * least one task in the system (init, pid == 1), therefore, root cgroup |
ddbcc7e8 | 1385 | * always has either children cgroups and/or using tasks. So we don't |
3dd06ffa | 1386 | * need a special hack to ensure that root cgroup cannot be deleted. |
ddbcc7e8 PM |
1387 | * |
1388 | * P.S. One more locking exception. RCU is used to guard the | |
956db3ca | 1389 | * update of a tasks cgroup pointer by cgroup_attach_task() |
ddbcc7e8 PM |
1390 | */ |
1391 | ||
2bd59d48 | 1392 | static struct kernfs_syscall_ops cgroup_kf_syscall_ops; |
a424316c | 1393 | |
8d7e6fb0 TH |
1394 | static char *cgroup_file_name(struct cgroup *cgrp, const struct cftype *cft, |
1395 | char *buf) | |
ddbcc7e8 | 1396 | { |
3e1d2eed TH |
1397 | struct cgroup_subsys *ss = cft->ss; |
1398 | ||
8d7e6fb0 TH |
1399 | if (cft->ss && !(cft->flags & CFTYPE_NO_PREFIX) && |
1400 | !(cgrp->root->flags & CGRP_ROOT_NOPREFIX)) | |
1401 | snprintf(buf, CGROUP_FILE_NAME_MAX, "%s.%s", | |
3e1d2eed TH |
1402 | cgroup_on_dfl(cgrp) ? ss->name : ss->legacy_name, |
1403 | cft->name); | |
8d7e6fb0 | 1404 | else |
50034ed4 | 1405 | strlcpy(buf, cft->name, CGROUP_FILE_NAME_MAX); |
8d7e6fb0 | 1406 | return buf; |
ddbcc7e8 PM |
1407 | } |
1408 | ||
f2e85d57 TH |
1409 | /** |
1410 | * cgroup_file_mode - deduce file mode of a control file | |
1411 | * @cft: the control file in question | |
1412 | * | |
7dbdb199 | 1413 | * S_IRUGO for read, S_IWUSR for write. |
f2e85d57 TH |
1414 | */ |
1415 | static umode_t cgroup_file_mode(const struct cftype *cft) | |
65dff759 | 1416 | { |
f2e85d57 | 1417 | umode_t mode = 0; |
65dff759 | 1418 | |
f2e85d57 TH |
1419 | if (cft->read_u64 || cft->read_s64 || cft->seq_show) |
1420 | mode |= S_IRUGO; | |
1421 | ||
7dbdb199 TH |
1422 | if (cft->write_u64 || cft->write_s64 || cft->write) { |
1423 | if (cft->flags & CFTYPE_WORLD_WRITABLE) | |
1424 | mode |= S_IWUGO; | |
1425 | else | |
1426 | mode |= S_IWUSR; | |
1427 | } | |
f2e85d57 TH |
1428 | |
1429 | return mode; | |
65dff759 LZ |
1430 | } |
1431 | ||
af0ba678 | 1432 | /** |
8699b776 | 1433 | * cgroup_calc_subtree_ss_mask - calculate subtree_ss_mask |
0f060deb | 1434 | * @subtree_control: the new subtree_control mask to consider |
5ced2518 | 1435 | * @this_ss_mask: available subsystems |
af0ba678 TH |
1436 | * |
1437 | * On the default hierarchy, a subsystem may request other subsystems to be | |
1438 | * enabled together through its ->depends_on mask. In such cases, more | |
1439 | * subsystems than specified in "cgroup.subtree_control" may be enabled. | |
1440 | * | |
0f060deb | 1441 | * This function calculates which subsystems need to be enabled if |
5ced2518 | 1442 | * @subtree_control is to be applied while restricted to @this_ss_mask. |
af0ba678 | 1443 | */ |
5ced2518 | 1444 | static u16 cgroup_calc_subtree_ss_mask(u16 subtree_control, u16 this_ss_mask) |
667c2491 | 1445 | { |
6e5c8307 | 1446 | u16 cur_ss_mask = subtree_control; |
af0ba678 TH |
1447 | struct cgroup_subsys *ss; |
1448 | int ssid; | |
1449 | ||
1450 | lockdep_assert_held(&cgroup_mutex); | |
1451 | ||
f6d635ad TH |
1452 | cur_ss_mask |= cgrp_dfl_implicit_ss_mask; |
1453 | ||
af0ba678 | 1454 | while (true) { |
6e5c8307 | 1455 | u16 new_ss_mask = cur_ss_mask; |
af0ba678 | 1456 | |
b4e0eeaf | 1457 | do_each_subsys_mask(ss, ssid, cur_ss_mask) { |
a966a4ed | 1458 | new_ss_mask |= ss->depends_on; |
b4e0eeaf | 1459 | } while_each_subsys_mask(); |
af0ba678 TH |
1460 | |
1461 | /* | |
1462 | * Mask out subsystems which aren't available. This can | |
1463 | * happen only if some depended-upon subsystems were bound | |
1464 | * to non-default hierarchies. | |
1465 | */ | |
5ced2518 | 1466 | new_ss_mask &= this_ss_mask; |
af0ba678 TH |
1467 | |
1468 | if (new_ss_mask == cur_ss_mask) | |
1469 | break; | |
1470 | cur_ss_mask = new_ss_mask; | |
1471 | } | |
1472 | ||
0f060deb TH |
1473 | return cur_ss_mask; |
1474 | } | |
1475 | ||
a9746d8d TH |
1476 | /** |
1477 | * cgroup_kn_unlock - unlocking helper for cgroup kernfs methods | |
1478 | * @kn: the kernfs_node being serviced | |
1479 | * | |
1480 | * This helper undoes cgroup_kn_lock_live() and should be invoked before | |
1481 | * the method finishes if locking succeeded. Note that once this function | |
1482 | * returns the cgroup returned by cgroup_kn_lock_live() may become | |
1483 | * inaccessible any time. If the caller intends to continue to access the | |
1484 | * cgroup, it should pin it before invoking this function. | |
1485 | */ | |
0a268dbd | 1486 | void cgroup_kn_unlock(struct kernfs_node *kn) |
ddbcc7e8 | 1487 | { |
a9746d8d TH |
1488 | struct cgroup *cgrp; |
1489 | ||
1490 | if (kernfs_type(kn) == KERNFS_DIR) | |
1491 | cgrp = kn->priv; | |
1492 | else | |
1493 | cgrp = kn->parent->priv; | |
1494 | ||
1495 | mutex_unlock(&cgroup_mutex); | |
a9746d8d TH |
1496 | |
1497 | kernfs_unbreak_active_protection(kn); | |
1498 | cgroup_put(cgrp); | |
ddbcc7e8 PM |
1499 | } |
1500 | ||
a9746d8d TH |
1501 | /** |
1502 | * cgroup_kn_lock_live - locking helper for cgroup kernfs methods | |
1503 | * @kn: the kernfs_node being serviced | |
945ba199 | 1504 | * @drain_offline: perform offline draining on the cgroup |
a9746d8d TH |
1505 | * |
1506 | * This helper is to be used by a cgroup kernfs method currently servicing | |
1507 | * @kn. It breaks the active protection, performs cgroup locking and | |
1508 | * verifies that the associated cgroup is alive. Returns the cgroup if | |
1509 | * alive; otherwise, %NULL. A successful return should be undone by a | |
945ba199 TH |
1510 | * matching cgroup_kn_unlock() invocation. If @drain_offline is %true, the |
1511 | * cgroup is drained of offlining csses before return. | |
a9746d8d TH |
1512 | * |
1513 | * Any cgroup kernfs method implementation which requires locking the | |
1514 | * associated cgroup should use this helper. It avoids nesting cgroup | |
1515 | * locking under kernfs active protection and allows all kernfs operations | |
1516 | * including self-removal. | |
1517 | */ | |
0a268dbd | 1518 | struct cgroup *cgroup_kn_lock_live(struct kernfs_node *kn, bool drain_offline) |
05ef1d7c | 1519 | { |
a9746d8d TH |
1520 | struct cgroup *cgrp; |
1521 | ||
1522 | if (kernfs_type(kn) == KERNFS_DIR) | |
1523 | cgrp = kn->priv; | |
1524 | else | |
1525 | cgrp = kn->parent->priv; | |
05ef1d7c | 1526 | |
2739d3cc | 1527 | /* |
01f6474c | 1528 | * We're gonna grab cgroup_mutex which nests outside kernfs |
a9746d8d TH |
1529 | * active_ref. cgroup liveliness check alone provides enough |
1530 | * protection against removal. Ensure @cgrp stays accessible and | |
1531 | * break the active_ref protection. | |
2739d3cc | 1532 | */ |
aa32362f LZ |
1533 | if (!cgroup_tryget(cgrp)) |
1534 | return NULL; | |
a9746d8d TH |
1535 | kernfs_break_active_protection(kn); |
1536 | ||
945ba199 TH |
1537 | if (drain_offline) |
1538 | cgroup_lock_and_drain_offline(cgrp); | |
1539 | else | |
1540 | mutex_lock(&cgroup_mutex); | |
05ef1d7c | 1541 | |
a9746d8d TH |
1542 | if (!cgroup_is_dead(cgrp)) |
1543 | return cgrp; | |
1544 | ||
1545 | cgroup_kn_unlock(kn); | |
1546 | return NULL; | |
ddbcc7e8 | 1547 | } |
05ef1d7c | 1548 | |
2739d3cc | 1549 | static void cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft) |
05ef1d7c | 1550 | { |
2bd59d48 | 1551 | char name[CGROUP_FILE_NAME_MAX]; |
05ef1d7c | 1552 | |
01f6474c | 1553 | lockdep_assert_held(&cgroup_mutex); |
34c06254 TH |
1554 | |
1555 | if (cft->file_offset) { | |
1556 | struct cgroup_subsys_state *css = cgroup_css(cgrp, cft->ss); | |
1557 | struct cgroup_file *cfile = (void *)css + cft->file_offset; | |
1558 | ||
1559 | spin_lock_irq(&cgroup_file_kn_lock); | |
1560 | cfile->kn = NULL; | |
1561 | spin_unlock_irq(&cgroup_file_kn_lock); | |
1562 | } | |
1563 | ||
2bd59d48 | 1564 | kernfs_remove_by_name(cgrp->kn, cgroup_file_name(cgrp, cft, name)); |
05ef1d7c TH |
1565 | } |
1566 | ||
13af07df | 1567 | /** |
4df8dc90 TH |
1568 | * css_clear_dir - remove subsys files in a cgroup directory |
1569 | * @css: taget css | |
13af07df | 1570 | */ |
334c3679 | 1571 | static void css_clear_dir(struct cgroup_subsys_state *css) |
05ef1d7c | 1572 | { |
334c3679 | 1573 | struct cgroup *cgrp = css->cgroup; |
4df8dc90 | 1574 | struct cftype *cfts; |
05ef1d7c | 1575 | |
88cb04b9 TH |
1576 | if (!(css->flags & CSS_VISIBLE)) |
1577 | return; | |
1578 | ||
1579 | css->flags &= ~CSS_VISIBLE; | |
1580 | ||
4df8dc90 TH |
1581 | list_for_each_entry(cfts, &css->ss->cfts, node) |
1582 | cgroup_addrm_files(css, cgrp, cfts, false); | |
ddbcc7e8 PM |
1583 | } |
1584 | ||
ccdca218 | 1585 | /** |
4df8dc90 TH |
1586 | * css_populate_dir - create subsys files in a cgroup directory |
1587 | * @css: target css | |
ccdca218 TH |
1588 | * |
1589 | * On failure, no file is added. | |
1590 | */ | |
334c3679 | 1591 | static int css_populate_dir(struct cgroup_subsys_state *css) |
ccdca218 | 1592 | { |
334c3679 | 1593 | struct cgroup *cgrp = css->cgroup; |
4df8dc90 TH |
1594 | struct cftype *cfts, *failed_cfts; |
1595 | int ret; | |
ccdca218 | 1596 | |
03970d3c | 1597 | if ((css->flags & CSS_VISIBLE) || !cgrp->kn) |
88cb04b9 TH |
1598 | return 0; |
1599 | ||
4df8dc90 TH |
1600 | if (!css->ss) { |
1601 | if (cgroup_on_dfl(cgrp)) | |
d62beb7f | 1602 | cfts = cgroup_base_files; |
4df8dc90 | 1603 | else |
d62beb7f | 1604 | cfts = cgroup1_base_files; |
ccdca218 | 1605 | |
4df8dc90 TH |
1606 | return cgroup_addrm_files(&cgrp->self, cgrp, cfts, true); |
1607 | } | |
ccdca218 | 1608 | |
4df8dc90 TH |
1609 | list_for_each_entry(cfts, &css->ss->cfts, node) { |
1610 | ret = cgroup_addrm_files(css, cgrp, cfts, true); | |
1611 | if (ret < 0) { | |
1612 | failed_cfts = cfts; | |
1613 | goto err; | |
ccdca218 TH |
1614 | } |
1615 | } | |
88cb04b9 TH |
1616 | |
1617 | css->flags |= CSS_VISIBLE; | |
1618 | ||
ccdca218 TH |
1619 | return 0; |
1620 | err: | |
4df8dc90 TH |
1621 | list_for_each_entry(cfts, &css->ss->cfts, node) { |
1622 | if (cfts == failed_cfts) | |
1623 | break; | |
1624 | cgroup_addrm_files(css, cgrp, cfts, false); | |
1625 | } | |
ccdca218 TH |
1626 | return ret; |
1627 | } | |
1628 | ||
0a268dbd | 1629 | int rebind_subsystems(struct cgroup_root *dst_root, u16 ss_mask) |
ddbcc7e8 | 1630 | { |
1ada4838 | 1631 | struct cgroup *dcgrp = &dst_root->cgrp; |
30159ec7 | 1632 | struct cgroup_subsys *ss; |
2d8f243a | 1633 | int ssid, i, ret; |
ddbcc7e8 | 1634 | |
ace2bee8 | 1635 | lockdep_assert_held(&cgroup_mutex); |
ddbcc7e8 | 1636 | |
b4e0eeaf | 1637 | do_each_subsys_mask(ss, ssid, ss_mask) { |
f6d635ad TH |
1638 | /* |
1639 | * If @ss has non-root csses attached to it, can't move. | |
1640 | * If @ss is an implicit controller, it is exempt from this | |
1641 | * rule and can be stolen. | |
1642 | */ | |
1643 | if (css_next_child(NULL, cgroup_css(&ss->root->cgrp, ss)) && | |
1644 | !ss->implicit_on_dfl) | |
3ed80a62 | 1645 | return -EBUSY; |
1d5be6b2 | 1646 | |
5df36032 | 1647 | /* can't move between two non-dummy roots either */ |
7fd8c565 | 1648 | if (ss->root != &cgrp_dfl_root && dst_root != &cgrp_dfl_root) |
5df36032 | 1649 | return -EBUSY; |
b4e0eeaf | 1650 | } while_each_subsys_mask(); |
ddbcc7e8 | 1651 | |
b4e0eeaf | 1652 | do_each_subsys_mask(ss, ssid, ss_mask) { |
1ada4838 TH |
1653 | struct cgroup_root *src_root = ss->root; |
1654 | struct cgroup *scgrp = &src_root->cgrp; | |
1655 | struct cgroup_subsys_state *css = cgroup_css(scgrp, ss); | |
2d8f243a | 1656 | struct css_set *cset; |
a8a648c4 | 1657 | |
1ada4838 | 1658 | WARN_ON(!css || cgroup_css(dcgrp, ss)); |
a8a648c4 | 1659 | |
334c3679 TH |
1660 | /* disable from the source */ |
1661 | src_root->subsys_mask &= ~(1 << ssid); | |
1662 | WARN_ON(cgroup_apply_control(scgrp)); | |
1663 | cgroup_finalize_control(scgrp, 0); | |
4df8dc90 | 1664 | |
334c3679 | 1665 | /* rebind */ |
1ada4838 TH |
1666 | RCU_INIT_POINTER(scgrp->subsys[ssid], NULL); |
1667 | rcu_assign_pointer(dcgrp->subsys[ssid], css); | |
5df36032 | 1668 | ss->root = dst_root; |
1ada4838 | 1669 | css->cgroup = dcgrp; |
73e80ed8 | 1670 | |
82d6489d | 1671 | spin_lock_irq(&css_set_lock); |
2d8f243a TH |
1672 | hash_for_each(css_set_table, i, cset, hlist) |
1673 | list_move_tail(&cset->e_cset_node[ss->id], | |
1ada4838 | 1674 | &dcgrp->e_csets[ss->id]); |
82d6489d | 1675 | spin_unlock_irq(&css_set_lock); |
2d8f243a | 1676 | |
bd53d617 | 1677 | /* default hierarchy doesn't enable controllers by default */ |
f392e51c | 1678 | dst_root->subsys_mask |= 1 << ssid; |
49d1dc4b TH |
1679 | if (dst_root == &cgrp_dfl_root) { |
1680 | static_branch_enable(cgroup_subsys_on_dfl_key[ssid]); | |
1681 | } else { | |
1ada4838 | 1682 | dcgrp->subtree_control |= 1 << ssid; |
49d1dc4b | 1683 | static_branch_disable(cgroup_subsys_on_dfl_key[ssid]); |
667c2491 | 1684 | } |
a8a648c4 | 1685 | |
334c3679 TH |
1686 | ret = cgroup_apply_control(dcgrp); |
1687 | if (ret) | |
1688 | pr_warn("partial failure to rebind %s controller (err=%d)\n", | |
1689 | ss->name, ret); | |
1690 | ||
5df36032 TH |
1691 | if (ss->bind) |
1692 | ss->bind(css); | |
b4e0eeaf | 1693 | } while_each_subsys_mask(); |
ddbcc7e8 | 1694 | |
1ada4838 | 1695 | kernfs_activate(dcgrp->kn); |
ddbcc7e8 PM |
1696 | return 0; |
1697 | } | |
1698 | ||
1592c9b2 TH |
1699 | int cgroup_show_path(struct seq_file *sf, struct kernfs_node *kf_node, |
1700 | struct kernfs_root *kf_root) | |
4f41fc59 | 1701 | { |
09be4c82 | 1702 | int len = 0; |
4f41fc59 SH |
1703 | char *buf = NULL; |
1704 | struct cgroup_root *kf_cgroot = cgroup_root_from_kf(kf_root); | |
1705 | struct cgroup *ns_cgroup; | |
1706 | ||
1707 | buf = kmalloc(PATH_MAX, GFP_KERNEL); | |
1708 | if (!buf) | |
1709 | return -ENOMEM; | |
1710 | ||
82d6489d | 1711 | spin_lock_irq(&css_set_lock); |
4f41fc59 SH |
1712 | ns_cgroup = current_cgns_cgroup_from_root(kf_cgroot); |
1713 | len = kernfs_path_from_node(kf_node, ns_cgroup->kn, buf, PATH_MAX); | |
82d6489d | 1714 | spin_unlock_irq(&css_set_lock); |
4f41fc59 SH |
1715 | |
1716 | if (len >= PATH_MAX) | |
1717 | len = -ERANGE; | |
1718 | else if (len > 0) { | |
1719 | seq_escape(sf, buf, " \t\n\\"); | |
1720 | len = 0; | |
1721 | } | |
1722 | kfree(buf); | |
1723 | return len; | |
1724 | } | |
1725 | ||
5136f636 TH |
1726 | static int parse_cgroup_root_flags(char *data, unsigned int *root_flags) |
1727 | { | |
1728 | char *token; | |
1729 | ||
1730 | *root_flags = 0; | |
1731 | ||
37ce608b | 1732 | if (!data || *data == '\0') |
5136f636 TH |
1733 | return 0; |
1734 | ||
1735 | while ((token = strsep(&data, ",")) != NULL) { | |
1736 | if (!strcmp(token, "nsdelegate")) { | |
1737 | *root_flags |= CGRP_ROOT_NS_DELEGATE; | |
1738 | continue; | |
1739 | } | |
1740 | ||
1741 | pr_err("cgroup2: unknown option \"%s\"\n", token); | |
1742 | return -EINVAL; | |
1743 | } | |
1744 | ||
1745 | return 0; | |
1746 | } | |
1747 | ||
1748 | static void apply_cgroup_root_flags(unsigned int root_flags) | |
1749 | { | |
1750 | if (current->nsproxy->cgroup_ns == &init_cgroup_ns) { | |
1751 | if (root_flags & CGRP_ROOT_NS_DELEGATE) | |
1752 | cgrp_dfl_root.flags |= CGRP_ROOT_NS_DELEGATE; | |
1753 | else | |
1754 | cgrp_dfl_root.flags &= ~CGRP_ROOT_NS_DELEGATE; | |
1755 | } | |
1756 | } | |
1757 | ||
1758 | static int cgroup_show_options(struct seq_file *seq, struct kernfs_root *kf_root) | |
1759 | { | |
1760 | if (cgrp_dfl_root.flags & CGRP_ROOT_NS_DELEGATE) | |
1761 | seq_puts(seq, ",nsdelegate"); | |
1762 | return 0; | |
1763 | } | |
1764 | ||
2bd59d48 | 1765 | static int cgroup_remount(struct kernfs_root *kf_root, int *flags, char *data) |
ddbcc7e8 | 1766 | { |
5136f636 TH |
1767 | unsigned int root_flags; |
1768 | int ret; | |
1769 | ||
1770 | ret = parse_cgroup_root_flags(data, &root_flags); | |
1771 | if (ret) | |
1772 | return ret; | |
1773 | ||
1774 | apply_cgroup_root_flags(root_flags); | |
1775 | return 0; | |
ddbcc7e8 PM |
1776 | } |
1777 | ||
afeb0f9f TH |
1778 | /* |
1779 | * To reduce the fork() overhead for systems that are not actually using | |
1780 | * their cgroups capability, we don't maintain the lists running through | |
1781 | * each css_set to its tasks until we see the list actually used - in other | |
1782 | * words after the first mount. | |
1783 | */ | |
1784 | static bool use_task_css_set_links __read_mostly; | |
1785 | ||
1786 | static void cgroup_enable_task_cg_lists(void) | |
1787 | { | |
1788 | struct task_struct *p, *g; | |
1789 | ||
82d6489d | 1790 | spin_lock_irq(&css_set_lock); |
afeb0f9f TH |
1791 | |
1792 | if (use_task_css_set_links) | |
1793 | goto out_unlock; | |
1794 | ||
1795 | use_task_css_set_links = true; | |
1796 | ||
1797 | /* | |
1798 | * We need tasklist_lock because RCU is not safe against | |
1799 | * while_each_thread(). Besides, a forking task that has passed | |
1800 | * cgroup_post_fork() without seeing use_task_css_set_links = 1 | |
1801 | * is not guaranteed to have its child immediately visible in the | |
1802 | * tasklist if we walk through it with RCU. | |
1803 | */ | |
1804 | read_lock(&tasklist_lock); | |
1805 | do_each_thread(g, p) { | |
afeb0f9f TH |
1806 | WARN_ON_ONCE(!list_empty(&p->cg_list) || |
1807 | task_css_set(p) != &init_css_set); | |
1808 | ||
1809 | /* | |
1810 | * We should check if the process is exiting, otherwise | |
1811 | * it will race with cgroup_exit() in that the list | |
1812 | * entry won't be deleted though the process has exited. | |
f153ad11 TH |
1813 | * Do it while holding siglock so that we don't end up |
1814 | * racing against cgroup_exit(). | |
82d6489d DBO |
1815 | * |
1816 | * Interrupts were already disabled while acquiring | |
1817 | * the css_set_lock, so we do not need to disable it | |
1818 | * again when acquiring the sighand->siglock here. | |
afeb0f9f | 1819 | */ |
82d6489d | 1820 | spin_lock(&p->sighand->siglock); |
eaf797ab TH |
1821 | if (!(p->flags & PF_EXITING)) { |
1822 | struct css_set *cset = task_css_set(p); | |
1823 | ||
0de0942d TH |
1824 | if (!css_set_populated(cset)) |
1825 | css_set_update_populated(cset, true); | |
389b9c1b | 1826 | list_add_tail(&p->cg_list, &cset->tasks); |
eaf797ab | 1827 | get_css_set(cset); |
73a7242a | 1828 | cset->nr_tasks++; |
eaf797ab | 1829 | } |
82d6489d | 1830 | spin_unlock(&p->sighand->siglock); |
afeb0f9f TH |
1831 | } while_each_thread(g, p); |
1832 | read_unlock(&tasklist_lock); | |
1833 | out_unlock: | |
82d6489d | 1834 | spin_unlock_irq(&css_set_lock); |
afeb0f9f | 1835 | } |
ddbcc7e8 | 1836 | |
cc31edce PM |
1837 | static void init_cgroup_housekeeping(struct cgroup *cgrp) |
1838 | { | |
2d8f243a TH |
1839 | struct cgroup_subsys *ss; |
1840 | int ssid; | |
1841 | ||
d5c419b6 TH |
1842 | INIT_LIST_HEAD(&cgrp->self.sibling); |
1843 | INIT_LIST_HEAD(&cgrp->self.children); | |
69d0206c | 1844 | INIT_LIST_HEAD(&cgrp->cset_links); |
72a8cb30 BB |
1845 | INIT_LIST_HEAD(&cgrp->pidlists); |
1846 | mutex_init(&cgrp->pidlist_mutex); | |
9d800df1 | 1847 | cgrp->self.cgroup = cgrp; |
184faf32 | 1848 | cgrp->self.flags |= CSS_ONLINE; |
454000ad | 1849 | cgrp->dom_cgrp = cgrp; |
1a926e0b RG |
1850 | cgrp->max_descendants = INT_MAX; |
1851 | cgrp->max_depth = INT_MAX; | |
2d8f243a TH |
1852 | |
1853 | for_each_subsys(ss, ssid) | |
1854 | INIT_LIST_HEAD(&cgrp->e_csets[ssid]); | |
f8f22e53 TH |
1855 | |
1856 | init_waitqueue_head(&cgrp->offline_waitq); | |
d62beb7f | 1857 | INIT_WORK(&cgrp->release_agent_work, cgroup1_release_agent); |
cc31edce | 1858 | } |
c6d57f33 | 1859 | |
1592c9b2 | 1860 | void init_cgroup_root(struct cgroup_root *root, struct cgroup_sb_opts *opts) |
ddbcc7e8 | 1861 | { |
3dd06ffa | 1862 | struct cgroup *cgrp = &root->cgrp; |
b0ca5a84 | 1863 | |
ddbcc7e8 | 1864 | INIT_LIST_HEAD(&root->root_list); |
3c9c825b | 1865 | atomic_set(&root->nr_cgrps, 1); |
bd89aabc | 1866 | cgrp->root = root; |
cc31edce | 1867 | init_cgroup_housekeeping(cgrp); |
4e96ee8e | 1868 | idr_init(&root->cgroup_idr); |
c6d57f33 | 1869 | |
c6d57f33 PM |
1870 | root->flags = opts->flags; |
1871 | if (opts->release_agent) | |
50034ed4 | 1872 | strlcpy(root->release_agent_path, opts->release_agent, PATH_MAX); |
c6d57f33 | 1873 | if (opts->name) |
50034ed4 | 1874 | strlcpy(root->name, opts->name, MAX_CGROUP_ROOT_NAMELEN); |
2260e7fc | 1875 | if (opts->cpuset_clone_children) |
3dd06ffa | 1876 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags); |
c6d57f33 PM |
1877 | } |
1878 | ||
9732adc5 | 1879 | int cgroup_setup_root(struct cgroup_root *root, u16 ss_mask, int ref_flags) |
2c6ab6d2 | 1880 | { |
d427dfeb | 1881 | LIST_HEAD(tmp_links); |
3dd06ffa | 1882 | struct cgroup *root_cgrp = &root->cgrp; |
fa069904 | 1883 | struct kernfs_syscall_ops *kf_sops; |
d427dfeb | 1884 | struct css_set *cset; |
d427dfeb | 1885 | int i, ret; |
2c6ab6d2 | 1886 | |
d427dfeb | 1887 | lockdep_assert_held(&cgroup_mutex); |
c6d57f33 | 1888 | |
cf780b7d | 1889 | ret = cgroup_idr_alloc(&root->cgroup_idr, root_cgrp, 1, 2, GFP_KERNEL); |
d427dfeb | 1890 | if (ret < 0) |
2bd59d48 | 1891 | goto out; |
d427dfeb | 1892 | root_cgrp->id = ret; |
b11cfb58 | 1893 | root_cgrp->ancestor_ids[0] = ret; |
c6d57f33 | 1894 | |
9732adc5 ZL |
1895 | ret = percpu_ref_init(&root_cgrp->self.refcnt, css_release, |
1896 | ref_flags, GFP_KERNEL); | |
9d755d33 TH |
1897 | if (ret) |
1898 | goto out; | |
1899 | ||
d427dfeb | 1900 | /* |
f0d9a5f1 | 1901 | * We're accessing css_set_count without locking css_set_lock here, |
d427dfeb | 1902 | * but that's OK - it can only be increased by someone holding |
04313591 TH |
1903 | * cgroup_lock, and that's us. Later rebinding may disable |
1904 | * controllers on the default hierarchy and thus create new csets, | |
1905 | * which can't be more than the existing ones. Allocate 2x. | |
d427dfeb | 1906 | */ |
04313591 | 1907 | ret = allocate_cgrp_cset_links(2 * css_set_count, &tmp_links); |
d427dfeb | 1908 | if (ret) |
9d755d33 | 1909 | goto cancel_ref; |
ddbcc7e8 | 1910 | |
985ed670 | 1911 | ret = cgroup_init_root_id(root); |
ddbcc7e8 | 1912 | if (ret) |
9d755d33 | 1913 | goto cancel_ref; |
ddbcc7e8 | 1914 | |
fa069904 TH |
1915 | kf_sops = root == &cgrp_dfl_root ? |
1916 | &cgroup_kf_syscall_ops : &cgroup1_kf_syscall_ops; | |
1917 | ||
1918 | root->kf_root = kernfs_create_root(kf_sops, | |
aa818825 SL |
1919 | KERNFS_ROOT_CREATE_DEACTIVATED | |
1920 | KERNFS_ROOT_SUPPORT_EXPORTOP, | |
2bd59d48 TH |
1921 | root_cgrp); |
1922 | if (IS_ERR(root->kf_root)) { | |
1923 | ret = PTR_ERR(root->kf_root); | |
1924 | goto exit_root_id; | |
1925 | } | |
1926 | root_cgrp->kn = root->kf_root->kn; | |
ddbcc7e8 | 1927 | |
334c3679 | 1928 | ret = css_populate_dir(&root_cgrp->self); |
d427dfeb | 1929 | if (ret) |
2bd59d48 | 1930 | goto destroy_root; |
ddbcc7e8 | 1931 | |
5df36032 | 1932 | ret = rebind_subsystems(root, ss_mask); |
d427dfeb | 1933 | if (ret) |
2bd59d48 | 1934 | goto destroy_root; |
ddbcc7e8 | 1935 | |
324bda9e AS |
1936 | ret = cgroup_bpf_inherit(root_cgrp); |
1937 | WARN_ON_ONCE(ret); | |
1938 | ||
ed1777de TH |
1939 | trace_cgroup_setup_root(root); |
1940 | ||
d427dfeb TH |
1941 | /* |
1942 | * There must be no failure case after here, since rebinding takes | |
1943 | * care of subsystems' refcounts, which are explicitly dropped in | |
1944 | * the failure exit path. | |
1945 | */ | |
1946 | list_add(&root->root_list, &cgroup_roots); | |
1947 | cgroup_root_count++; | |
0df6a63f | 1948 | |
d427dfeb | 1949 | /* |
3dd06ffa | 1950 | * Link the root cgroup in this hierarchy into all the css_set |
d427dfeb TH |
1951 | * objects. |
1952 | */ | |
82d6489d | 1953 | spin_lock_irq(&css_set_lock); |
0de0942d | 1954 | hash_for_each(css_set_table, i, cset, hlist) { |
d427dfeb | 1955 | link_css_set(&tmp_links, cset, root_cgrp); |
0de0942d TH |
1956 | if (css_set_populated(cset)) |
1957 | cgroup_update_populated(root_cgrp, true); | |
1958 | } | |
82d6489d | 1959 | spin_unlock_irq(&css_set_lock); |
ddbcc7e8 | 1960 | |
d5c419b6 | 1961 | BUG_ON(!list_empty(&root_cgrp->self.children)); |
3c9c825b | 1962 | BUG_ON(atomic_read(&root->nr_cgrps) != 1); |
ddbcc7e8 | 1963 | |
2bd59d48 | 1964 | kernfs_activate(root_cgrp->kn); |
d427dfeb | 1965 | ret = 0; |
2bd59d48 | 1966 | goto out; |
d427dfeb | 1967 | |
2bd59d48 TH |
1968 | destroy_root: |
1969 | kernfs_destroy_root(root->kf_root); | |
1970 | root->kf_root = NULL; | |
1971 | exit_root_id: | |
d427dfeb | 1972 | cgroup_exit_root_id(root); |
9d755d33 | 1973 | cancel_ref: |
9a1049da | 1974 | percpu_ref_exit(&root_cgrp->self.refcnt); |
2bd59d48 | 1975 | out: |
d427dfeb TH |
1976 | free_cgrp_cset_links(&tmp_links); |
1977 | return ret; | |
ddbcc7e8 PM |
1978 | } |
1979 | ||
1592c9b2 TH |
1980 | struct dentry *cgroup_do_mount(struct file_system_type *fs_type, int flags, |
1981 | struct cgroup_root *root, unsigned long magic, | |
1982 | struct cgroup_namespace *ns) | |
ddbcc7e8 | 1983 | { |
2bd59d48 | 1984 | struct dentry *dentry; |
d5d2bfc7 | 1985 | bool new_sb = false; |
ddbcc7e8 | 1986 | |
633feee3 | 1987 | dentry = kernfs_mount(fs_type, flags, root->kf_root, magic, &new_sb); |
ed82571b | 1988 | |
56fde9e0 | 1989 | /* |
633feee3 TH |
1990 | * In non-init cgroup namespace, instead of root cgroup's dentry, |
1991 | * we return the dentry corresponding to the cgroupns->root_cgrp. | |
56fde9e0 | 1992 | */ |
633feee3 TH |
1993 | if (!IS_ERR(dentry) && ns != &init_cgroup_ns) { |
1994 | struct dentry *nsdentry; | |
d5d2bfc7 | 1995 | struct super_block *sb = dentry->d_sb; |
633feee3 | 1996 | struct cgroup *cgrp; |
e37a06f1 | 1997 | |
633feee3 TH |
1998 | mutex_lock(&cgroup_mutex); |
1999 | spin_lock_irq(&css_set_lock); | |
2000 | ||
2001 | cgrp = cset_cgroup_from_root(ns->root_cset, root); | |
2002 | ||
2003 | spin_unlock_irq(&css_set_lock); | |
2004 | mutex_unlock(&cgroup_mutex); | |
2005 | ||
d5d2bfc7 | 2006 | nsdentry = kernfs_node_dentry(cgrp->kn, sb); |
633feee3 | 2007 | dput(dentry); |
d5d2bfc7 AV |
2008 | if (IS_ERR(nsdentry)) |
2009 | deactivate_locked_super(sb); | |
633feee3 | 2010 | dentry = nsdentry; |
67e9c74b TH |
2011 | } |
2012 | ||
d5d2bfc7 | 2013 | if (!new_sb) |
633feee3 TH |
2014 | cgroup_put(&root->cgrp); |
2015 | ||
2016 | return dentry; | |
2017 | } | |
2018 | ||
f7e83571 | 2019 | static struct dentry *cgroup_mount(struct file_system_type *fs_type, |
ddbcc7e8 | 2020 | int flags, const char *unused_dev_name, |
f7e83571 | 2021 | void *data) |
ddbcc7e8 | 2022 | { |
ed82571b | 2023 | struct cgroup_namespace *ns = current->nsproxy->cgroup_ns; |
2bd59d48 | 2024 | struct dentry *dentry; |
5136f636 | 2025 | int ret; |
ddbcc7e8 | 2026 | |
ed82571b SH |
2027 | get_cgroup_ns(ns); |
2028 | ||
2029 | /* Check if the caller has permission to mount. */ | |
2030 | if (!ns_capable(ns->user_ns, CAP_SYS_ADMIN)) { | |
2031 | put_cgroup_ns(ns); | |
2032 | return ERR_PTR(-EPERM); | |
2033 | } | |
2034 | ||
56fde9e0 TH |
2035 | /* |
2036 | * The first time anyone tries to mount a cgroup, enable the list | |
2037 | * linking each css_set to its tasks and fix up all existing tasks. | |
2038 | */ | |
2039 | if (!use_task_css_set_links) | |
2040 | cgroup_enable_task_cg_lists(); | |
e37a06f1 | 2041 | |
633feee3 | 2042 | if (fs_type == &cgroup2_fs_type) { |
5136f636 TH |
2043 | unsigned int root_flags; |
2044 | ||
2045 | ret = parse_cgroup_root_flags(data, &root_flags); | |
2046 | if (ret) { | |
ed82571b | 2047 | put_cgroup_ns(ns); |
5136f636 | 2048 | return ERR_PTR(ret); |
67e9c74b | 2049 | } |
5136f636 | 2050 | |
a7165264 | 2051 | cgrp_dfl_visible = true; |
a590b90d | 2052 | cgroup_get_live(&cgrp_dfl_root.cgrp); |
633feee3 TH |
2053 | |
2054 | dentry = cgroup_do_mount(&cgroup2_fs_type, flags, &cgrp_dfl_root, | |
2055 | CGROUP2_SUPER_MAGIC, ns); | |
5136f636 TH |
2056 | if (!IS_ERR(dentry)) |
2057 | apply_cgroup_root_flags(root_flags); | |
633feee3 TH |
2058 | } else { |
2059 | dentry = cgroup1_mount(&cgroup_fs_type, flags, data, | |
2060 | CGROUP_SUPER_MAGIC, ns); | |
67e9c74b TH |
2061 | } |
2062 | ||
ed82571b | 2063 | put_cgroup_ns(ns); |
2bd59d48 TH |
2064 | return dentry; |
2065 | } | |
8e30e2b8 | 2066 | |
2bd59d48 TH |
2067 | static void cgroup_kill_sb(struct super_block *sb) |
2068 | { | |
2069 | struct kernfs_root *kf_root = kernfs_root_from_sb(sb); | |
3dd06ffa | 2070 | struct cgroup_root *root = cgroup_root_from_kf(kf_root); |
a015edd2 | 2071 | |
970317aa | 2072 | /* |
9d755d33 TH |
2073 | * If @root doesn't have any mounts or children, start killing it. |
2074 | * This prevents new mounts by disabling percpu_ref_tryget_live(). | |
2075 | * cgroup_mount() may wait for @root's release. | |
1f779fb2 LZ |
2076 | * |
2077 | * And don't kill the default root. | |
970317aa | 2078 | */ |
3c606d35 | 2079 | if (!list_empty(&root->cgrp.self.children) || |
1f779fb2 | 2080 | root == &cgrp_dfl_root) |
9d755d33 TH |
2081 | cgroup_put(&root->cgrp); |
2082 | else | |
2083 | percpu_ref_kill(&root->cgrp.self.refcnt); | |
970317aa | 2084 | |
2bd59d48 | 2085 | kernfs_kill_sb(sb); |
ddbcc7e8 | 2086 | } |
970317aa | 2087 | |
0a268dbd | 2088 | struct file_system_type cgroup_fs_type = { |
ddbcc7e8 | 2089 | .name = "cgroup", |
f7e83571 | 2090 | .mount = cgroup_mount, |
ddbcc7e8 | 2091 | .kill_sb = cgroup_kill_sb, |
1c53753e | 2092 | .fs_flags = FS_USERNS_MOUNT, |
ddbcc7e8 | 2093 | }; |
3126121f | 2094 | |
67e9c74b TH |
2095 | static struct file_system_type cgroup2_fs_type = { |
2096 | .name = "cgroup2", | |
2097 | .mount = cgroup_mount, | |
2098 | .kill_sb = cgroup_kill_sb, | |
1c53753e | 2099 | .fs_flags = FS_USERNS_MOUNT, |
67e9c74b | 2100 | }; |
3126121f | 2101 | |
0a268dbd TH |
2102 | int cgroup_path_ns_locked(struct cgroup *cgrp, char *buf, size_t buflen, |
2103 | struct cgroup_namespace *ns) | |
a79a908f AK |
2104 | { |
2105 | struct cgroup *root = cset_cgroup_from_root(ns->root_cset, cgrp->root); | |
a79a908f | 2106 | |
4c737b41 | 2107 | return kernfs_path_from_node(cgrp->kn, root->kn, buf, buflen); |
a79a908f AK |
2108 | } |
2109 | ||
4c737b41 TH |
2110 | int cgroup_path_ns(struct cgroup *cgrp, char *buf, size_t buflen, |
2111 | struct cgroup_namespace *ns) | |
a79a908f | 2112 | { |
4c737b41 | 2113 | int ret; |
a79a908f AK |
2114 | |
2115 | mutex_lock(&cgroup_mutex); | |
82d6489d | 2116 | spin_lock_irq(&css_set_lock); |
a79a908f AK |
2117 | |
2118 | ret = cgroup_path_ns_locked(cgrp, buf, buflen, ns); | |
2119 | ||
82d6489d | 2120 | spin_unlock_irq(&css_set_lock); |
a79a908f AK |
2121 | mutex_unlock(&cgroup_mutex); |
2122 | ||
2123 | return ret; | |
2124 | } | |
2125 | EXPORT_SYMBOL_GPL(cgroup_path_ns); | |
2126 | ||
857a2beb | 2127 | /** |
913ffdb5 | 2128 | * task_cgroup_path - cgroup path of a task in the first cgroup hierarchy |
857a2beb | 2129 | * @task: target task |
857a2beb TH |
2130 | * @buf: the buffer to write the path into |
2131 | * @buflen: the length of the buffer | |
2132 | * | |
913ffdb5 TH |
2133 | * Determine @task's cgroup on the first (the one with the lowest non-zero |
2134 | * hierarchy_id) cgroup hierarchy and copy its path into @buf. This | |
2135 | * function grabs cgroup_mutex and shouldn't be used inside locks used by | |
2136 | * cgroup controller callbacks. | |
2137 | * | |
e61734c5 | 2138 | * Return value is the same as kernfs_path(). |
857a2beb | 2139 | */ |
4c737b41 | 2140 | int task_cgroup_path(struct task_struct *task, char *buf, size_t buflen) |
857a2beb | 2141 | { |
3dd06ffa | 2142 | struct cgroup_root *root; |
913ffdb5 | 2143 | struct cgroup *cgrp; |
e61734c5 | 2144 | int hierarchy_id = 1; |
4c737b41 | 2145 | int ret; |
857a2beb TH |
2146 | |
2147 | mutex_lock(&cgroup_mutex); | |
82d6489d | 2148 | spin_lock_irq(&css_set_lock); |
857a2beb | 2149 | |
913ffdb5 TH |
2150 | root = idr_get_next(&cgroup_hierarchy_idr, &hierarchy_id); |
2151 | ||
857a2beb TH |
2152 | if (root) { |
2153 | cgrp = task_cgroup_from_root(task, root); | |
4c737b41 | 2154 | ret = cgroup_path_ns_locked(cgrp, buf, buflen, &init_cgroup_ns); |
913ffdb5 TH |
2155 | } else { |
2156 | /* if no hierarchy exists, everyone is in "/" */ | |
4c737b41 | 2157 | ret = strlcpy(buf, "/", buflen); |
857a2beb TH |
2158 | } |
2159 | ||
82d6489d | 2160 | spin_unlock_irq(&css_set_lock); |
857a2beb | 2161 | mutex_unlock(&cgroup_mutex); |
4c737b41 | 2162 | return ret; |
857a2beb | 2163 | } |
913ffdb5 | 2164 | EXPORT_SYMBOL_GPL(task_cgroup_path); |
857a2beb | 2165 | |
adaae5dc | 2166 | /** |
e595cd70 | 2167 | * cgroup_migrate_add_task - add a migration target task to a migration context |
adaae5dc | 2168 | * @task: target task |
e595cd70 | 2169 | * @mgctx: target migration context |
adaae5dc | 2170 | * |
e595cd70 TH |
2171 | * Add @task, which is a migration target, to @mgctx->tset. This function |
2172 | * becomes noop if @task doesn't need to be migrated. @task's css_set | |
2173 | * should have been added as a migration source and @task->cg_list will be | |
2174 | * moved from the css_set's tasks list to mg_tasks one. | |
adaae5dc | 2175 | */ |
e595cd70 TH |
2176 | static void cgroup_migrate_add_task(struct task_struct *task, |
2177 | struct cgroup_mgctx *mgctx) | |
adaae5dc TH |
2178 | { |
2179 | struct css_set *cset; | |
2180 | ||
f0d9a5f1 | 2181 | lockdep_assert_held(&css_set_lock); |
adaae5dc TH |
2182 | |
2183 | /* @task either already exited or can't exit until the end */ | |
2184 | if (task->flags & PF_EXITING) | |
2185 | return; | |
2186 | ||
2187 | /* leave @task alone if post_fork() hasn't linked it yet */ | |
2188 | if (list_empty(&task->cg_list)) | |
2189 | return; | |
2190 | ||
2191 | cset = task_css_set(task); | |
2192 | if (!cset->mg_src_cgrp) | |
2193 | return; | |
2194 | ||
61046727 TH |
2195 | mgctx->tset.nr_tasks++; |
2196 | ||
adaae5dc TH |
2197 | list_move_tail(&task->cg_list, &cset->mg_tasks); |
2198 | if (list_empty(&cset->mg_node)) | |
e595cd70 TH |
2199 | list_add_tail(&cset->mg_node, |
2200 | &mgctx->tset.src_csets); | |
adaae5dc | 2201 | if (list_empty(&cset->mg_dst_cset->mg_node)) |
d8ebf519 | 2202 | list_add_tail(&cset->mg_dst_cset->mg_node, |
e595cd70 | 2203 | &mgctx->tset.dst_csets); |
adaae5dc TH |
2204 | } |
2205 | ||
2f7ee569 TH |
2206 | /** |
2207 | * cgroup_taskset_first - reset taskset and return the first task | |
2208 | * @tset: taskset of interest | |
1f7dd3e5 | 2209 | * @dst_cssp: output variable for the destination css |
2f7ee569 TH |
2210 | * |
2211 | * @tset iteration is initialized and the first task is returned. | |
2212 | */ | |
1f7dd3e5 TH |
2213 | struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset, |
2214 | struct cgroup_subsys_state **dst_cssp) | |
2f7ee569 | 2215 | { |
b3dc094e TH |
2216 | tset->cur_cset = list_first_entry(tset->csets, struct css_set, mg_node); |
2217 | tset->cur_task = NULL; | |
2218 | ||
1f7dd3e5 | 2219 | return cgroup_taskset_next(tset, dst_cssp); |
2f7ee569 | 2220 | } |
2f7ee569 TH |
2221 | |
2222 | /** | |
2223 | * cgroup_taskset_next - iterate to the next task in taskset | |
2224 | * @tset: taskset of interest | |
1f7dd3e5 | 2225 | * @dst_cssp: output variable for the destination css |
2f7ee569 TH |
2226 | * |
2227 | * Return the next task in @tset. Iteration must have been initialized | |
2228 | * with cgroup_taskset_first(). | |
2229 | */ | |
1f7dd3e5 TH |
2230 | struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset, |
2231 | struct cgroup_subsys_state **dst_cssp) | |
2f7ee569 | 2232 | { |
b3dc094e TH |
2233 | struct css_set *cset = tset->cur_cset; |
2234 | struct task_struct *task = tset->cur_task; | |
2f7ee569 | 2235 | |
b3dc094e TH |
2236 | while (&cset->mg_node != tset->csets) { |
2237 | if (!task) | |
2238 | task = list_first_entry(&cset->mg_tasks, | |
2239 | struct task_struct, cg_list); | |
2240 | else | |
2241 | task = list_next_entry(task, cg_list); | |
2f7ee569 | 2242 | |
b3dc094e TH |
2243 | if (&task->cg_list != &cset->mg_tasks) { |
2244 | tset->cur_cset = cset; | |
2245 | tset->cur_task = task; | |
1f7dd3e5 TH |
2246 | |
2247 | /* | |
2248 | * This function may be called both before and | |
2249 | * after cgroup_taskset_migrate(). The two cases | |
2250 | * can be distinguished by looking at whether @cset | |
2251 | * has its ->mg_dst_cset set. | |
2252 | */ | |
2253 | if (cset->mg_dst_cset) | |
2254 | *dst_cssp = cset->mg_dst_cset->subsys[tset->ssid]; | |
2255 | else | |
2256 | *dst_cssp = cset->subsys[tset->ssid]; | |
2257 | ||
b3dc094e TH |
2258 | return task; |
2259 | } | |
2f7ee569 | 2260 | |
b3dc094e TH |
2261 | cset = list_next_entry(cset, mg_node); |
2262 | task = NULL; | |
2263 | } | |
2f7ee569 | 2264 | |
b3dc094e | 2265 | return NULL; |
2f7ee569 | 2266 | } |
2f7ee569 | 2267 | |
adaae5dc | 2268 | /** |
37ff9f8f | 2269 | * cgroup_taskset_migrate - migrate a taskset |
e595cd70 | 2270 | * @mgctx: migration context |
adaae5dc | 2271 | * |
e595cd70 | 2272 | * Migrate tasks in @mgctx as setup by migration preparation functions. |
37ff9f8f | 2273 | * This function fails iff one of the ->can_attach callbacks fails and |
e595cd70 TH |
2274 | * guarantees that either all or none of the tasks in @mgctx are migrated. |
2275 | * @mgctx is consumed regardless of success. | |
adaae5dc | 2276 | */ |
bfc2cf6f | 2277 | static int cgroup_migrate_execute(struct cgroup_mgctx *mgctx) |
adaae5dc | 2278 | { |
e595cd70 | 2279 | struct cgroup_taskset *tset = &mgctx->tset; |
37ff9f8f | 2280 | struct cgroup_subsys *ss; |
adaae5dc TH |
2281 | struct task_struct *task, *tmp_task; |
2282 | struct css_set *cset, *tmp_cset; | |
37ff9f8f | 2283 | int ssid, failed_ssid, ret; |
adaae5dc | 2284 | |
adaae5dc | 2285 | /* check that we can legitimately attach to the cgroup */ |
61046727 TH |
2286 | if (tset->nr_tasks) { |
2287 | do_each_subsys_mask(ss, ssid, mgctx->ss_mask) { | |
2288 | if (ss->can_attach) { | |
2289 | tset->ssid = ssid; | |
2290 | ret = ss->can_attach(tset); | |
2291 | if (ret) { | |
2292 | failed_ssid = ssid; | |
2293 | goto out_cancel_attach; | |
2294 | } | |
adaae5dc | 2295 | } |
61046727 TH |
2296 | } while_each_subsys_mask(); |
2297 | } | |
adaae5dc TH |
2298 | |
2299 | /* | |
2300 | * Now that we're guaranteed success, proceed to move all tasks to | |
2301 | * the new cgroup. There are no failure cases after here, so this | |
2302 | * is the commit point. | |
2303 | */ | |
82d6489d | 2304 | spin_lock_irq(&css_set_lock); |
adaae5dc | 2305 | list_for_each_entry(cset, &tset->src_csets, mg_node) { |
f6d7d049 TH |
2306 | list_for_each_entry_safe(task, tmp_task, &cset->mg_tasks, cg_list) { |
2307 | struct css_set *from_cset = task_css_set(task); | |
2308 | struct css_set *to_cset = cset->mg_dst_cset; | |
2309 | ||
2310 | get_css_set(to_cset); | |
73a7242a | 2311 | to_cset->nr_tasks++; |
f6d7d049 TH |
2312 | css_set_move_task(task, from_cset, to_cset, true); |
2313 | put_css_set_locked(from_cset); | |
73a7242a | 2314 | from_cset->nr_tasks--; |
f6d7d049 | 2315 | } |
adaae5dc | 2316 | } |
82d6489d | 2317 | spin_unlock_irq(&css_set_lock); |
adaae5dc TH |
2318 | |
2319 | /* | |
2320 | * Migration is committed, all target tasks are now on dst_csets. | |
2321 | * Nothing is sensitive to fork() after this point. Notify | |
2322 | * controllers that migration is complete. | |
2323 | */ | |
2324 | tset->csets = &tset->dst_csets; | |
2325 | ||
61046727 TH |
2326 | if (tset->nr_tasks) { |
2327 | do_each_subsys_mask(ss, ssid, mgctx->ss_mask) { | |
2328 | if (ss->attach) { | |
2329 | tset->ssid = ssid; | |
2330 | ss->attach(tset); | |
2331 | } | |
2332 | } while_each_subsys_mask(); | |
2333 | } | |
adaae5dc TH |
2334 | |
2335 | ret = 0; | |
2336 | goto out_release_tset; | |
2337 | ||
2338 | out_cancel_attach: | |
61046727 TH |
2339 | if (tset->nr_tasks) { |
2340 | do_each_subsys_mask(ss, ssid, mgctx->ss_mask) { | |
2341 | if (ssid == failed_ssid) | |
2342 | break; | |
2343 | if (ss->cancel_attach) { | |
2344 | tset->ssid = ssid; | |
2345 | ss->cancel_attach(tset); | |
2346 | } | |
2347 | } while_each_subsys_mask(); | |
2348 | } | |
adaae5dc | 2349 | out_release_tset: |
82d6489d | 2350 | spin_lock_irq(&css_set_lock); |
adaae5dc TH |
2351 | list_splice_init(&tset->dst_csets, &tset->src_csets); |
2352 | list_for_each_entry_safe(cset, tmp_cset, &tset->src_csets, mg_node) { | |
2353 | list_splice_tail_init(&cset->mg_tasks, &cset->tasks); | |
2354 | list_del_init(&cset->mg_node); | |
2355 | } | |
82d6489d | 2356 | spin_unlock_irq(&css_set_lock); |
c4fa6c43 WL |
2357 | |
2358 | /* | |
2359 | * Re-initialize the cgroup_taskset structure in case it is reused | |
2360 | * again in another cgroup_migrate_add_task()/cgroup_migrate_execute() | |
2361 | * iteration. | |
2362 | */ | |
2363 | tset->nr_tasks = 0; | |
2364 | tset->csets = &tset->src_csets; | |
adaae5dc TH |
2365 | return ret; |
2366 | } | |
2367 | ||
6c694c88 | 2368 | /** |
8cfd8147 | 2369 | * cgroup_migrate_vet_dst - verify whether a cgroup can be migration destination |
6c694c88 TH |
2370 | * @dst_cgrp: destination cgroup to test |
2371 | * | |
8cfd8147 TH |
2372 | * On the default hierarchy, except for the mixable, (possible) thread root |
2373 | * and threaded cgroups, subtree_control must be zero for migration | |
2374 | * destination cgroups with tasks so that child cgroups don't compete | |
2375 | * against tasks. | |
6c694c88 | 2376 | */ |
8cfd8147 | 2377 | int cgroup_migrate_vet_dst(struct cgroup *dst_cgrp) |
6c694c88 | 2378 | { |
8cfd8147 TH |
2379 | /* v1 doesn't have any restriction */ |
2380 | if (!cgroup_on_dfl(dst_cgrp)) | |
2381 | return 0; | |
2382 | ||
2383 | /* verify @dst_cgrp can host resources */ | |
2384 | if (!cgroup_is_valid_domain(dst_cgrp->dom_cgrp)) | |
2385 | return -EOPNOTSUPP; | |
2386 | ||
2387 | /* mixables don't care */ | |
2388 | if (cgroup_is_mixable(dst_cgrp)) | |
2389 | return 0; | |
2390 | ||
2391 | /* | |
2392 | * If @dst_cgrp is already or can become a thread root or is | |
2393 | * threaded, it doesn't matter. | |
2394 | */ | |
2395 | if (cgroup_can_be_thread_root(dst_cgrp) || cgroup_is_threaded(dst_cgrp)) | |
2396 | return 0; | |
2397 | ||
2398 | /* apply no-internal-process constraint */ | |
2399 | if (dst_cgrp->subtree_control) | |
2400 | return -EBUSY; | |
2401 | ||
2402 | return 0; | |
6c694c88 TH |
2403 | } |
2404 | ||
a043e3b2 | 2405 | /** |
1958d2d5 | 2406 | * cgroup_migrate_finish - cleanup after attach |
e595cd70 | 2407 | * @mgctx: migration context |
74a1166d | 2408 | * |
1958d2d5 TH |
2409 | * Undo cgroup_migrate_add_src() and cgroup_migrate_prepare_dst(). See |
2410 | * those functions for details. | |
74a1166d | 2411 | */ |
e595cd70 | 2412 | void cgroup_migrate_finish(struct cgroup_mgctx *mgctx) |
74a1166d | 2413 | { |
e595cd70 | 2414 | LIST_HEAD(preloaded); |
1958d2d5 | 2415 | struct css_set *cset, *tmp_cset; |
74a1166d | 2416 | |
1958d2d5 TH |
2417 | lockdep_assert_held(&cgroup_mutex); |
2418 | ||
82d6489d | 2419 | spin_lock_irq(&css_set_lock); |
e595cd70 TH |
2420 | |
2421 | list_splice_tail_init(&mgctx->preloaded_src_csets, &preloaded); | |
2422 | list_splice_tail_init(&mgctx->preloaded_dst_csets, &preloaded); | |
2423 | ||
2424 | list_for_each_entry_safe(cset, tmp_cset, &preloaded, mg_preload_node) { | |
1958d2d5 | 2425 | cset->mg_src_cgrp = NULL; |
e4857982 | 2426 | cset->mg_dst_cgrp = NULL; |
1958d2d5 TH |
2427 | cset->mg_dst_cset = NULL; |
2428 | list_del_init(&cset->mg_preload_node); | |
a25eb52e | 2429 | put_css_set_locked(cset); |
1958d2d5 | 2430 | } |
e595cd70 | 2431 | |
82d6489d | 2432 | spin_unlock_irq(&css_set_lock); |
1958d2d5 TH |
2433 | } |
2434 | ||
2435 | /** | |
2436 | * cgroup_migrate_add_src - add a migration source css_set | |
2437 | * @src_cset: the source css_set to add | |
2438 | * @dst_cgrp: the destination cgroup | |
e595cd70 | 2439 | * @mgctx: migration context |
1958d2d5 TH |
2440 | * |
2441 | * Tasks belonging to @src_cset are about to be migrated to @dst_cgrp. Pin | |
e595cd70 | 2442 | * @src_cset and add it to @mgctx->src_csets, which should later be cleaned |
1958d2d5 TH |
2443 | * up by cgroup_migrate_finish(). |
2444 | * | |
1ed13287 TH |
2445 | * This function may be called without holding cgroup_threadgroup_rwsem |
2446 | * even if the target is a process. Threads may be created and destroyed | |
2447 | * but as long as cgroup_mutex is not dropped, no new css_set can be put | |
2448 | * into play and the preloaded css_sets are guaranteed to cover all | |
2449 | * migrations. | |
1958d2d5 | 2450 | */ |
0a268dbd TH |
2451 | void cgroup_migrate_add_src(struct css_set *src_cset, |
2452 | struct cgroup *dst_cgrp, | |
e595cd70 | 2453 | struct cgroup_mgctx *mgctx) |
1958d2d5 TH |
2454 | { |
2455 | struct cgroup *src_cgrp; | |
2456 | ||
2457 | lockdep_assert_held(&cgroup_mutex); | |
f0d9a5f1 | 2458 | lockdep_assert_held(&css_set_lock); |
1958d2d5 | 2459 | |
2b021cbf TH |
2460 | /* |
2461 | * If ->dead, @src_set is associated with one or more dead cgroups | |
2462 | * and doesn't contain any migratable tasks. Ignore it early so | |
2463 | * that the rest of migration path doesn't get confused by it. | |
2464 | */ | |
2465 | if (src_cset->dead) | |
2466 | return; | |
2467 | ||
1958d2d5 TH |
2468 | src_cgrp = cset_cgroup_from_root(src_cset, dst_cgrp->root); |
2469 | ||
1958d2d5 TH |
2470 | if (!list_empty(&src_cset->mg_preload_node)) |
2471 | return; | |
2472 | ||
2473 | WARN_ON(src_cset->mg_src_cgrp); | |
e4857982 | 2474 | WARN_ON(src_cset->mg_dst_cgrp); |
1958d2d5 TH |
2475 | WARN_ON(!list_empty(&src_cset->mg_tasks)); |
2476 | WARN_ON(!list_empty(&src_cset->mg_node)); | |
2477 | ||
2478 | src_cset->mg_src_cgrp = src_cgrp; | |
e4857982 | 2479 | src_cset->mg_dst_cgrp = dst_cgrp; |
1958d2d5 | 2480 | get_css_set(src_cset); |
e595cd70 | 2481 | list_add_tail(&src_cset->mg_preload_node, &mgctx->preloaded_src_csets); |
1958d2d5 TH |
2482 | } |
2483 | ||
2484 | /** | |
2485 | * cgroup_migrate_prepare_dst - prepare destination css_sets for migration | |
e595cd70 | 2486 | * @mgctx: migration context |
1958d2d5 | 2487 | * |
e4857982 | 2488 | * Tasks are about to be moved and all the source css_sets have been |
e595cd70 TH |
2489 | * preloaded to @mgctx->preloaded_src_csets. This function looks up and |
2490 | * pins all destination css_sets, links each to its source, and append them | |
2491 | * to @mgctx->preloaded_dst_csets. | |
1958d2d5 TH |
2492 | * |
2493 | * This function must be called after cgroup_migrate_add_src() has been | |
2494 | * called on each migration source css_set. After migration is performed | |
2495 | * using cgroup_migrate(), cgroup_migrate_finish() must be called on | |
e595cd70 | 2496 | * @mgctx. |
1958d2d5 | 2497 | */ |
e595cd70 | 2498 | int cgroup_migrate_prepare_dst(struct cgroup_mgctx *mgctx) |
1958d2d5 | 2499 | { |
f817de98 | 2500 | struct css_set *src_cset, *tmp_cset; |
1958d2d5 TH |
2501 | |
2502 | lockdep_assert_held(&cgroup_mutex); | |
2503 | ||
2504 | /* look up the dst cset for each src cset and link it to src */ | |
e595cd70 TH |
2505 | list_for_each_entry_safe(src_cset, tmp_cset, &mgctx->preloaded_src_csets, |
2506 | mg_preload_node) { | |
1958d2d5 | 2507 | struct css_set *dst_cset; |
bfc2cf6f TH |
2508 | struct cgroup_subsys *ss; |
2509 | int ssid; | |
1958d2d5 | 2510 | |
e4857982 | 2511 | dst_cset = find_css_set(src_cset, src_cset->mg_dst_cgrp); |
1958d2d5 TH |
2512 | if (!dst_cset) |
2513 | goto err; | |
2514 | ||
2515 | WARN_ON_ONCE(src_cset->mg_dst_cset || dst_cset->mg_dst_cset); | |
f817de98 TH |
2516 | |
2517 | /* | |
2518 | * If src cset equals dst, it's noop. Drop the src. | |
2519 | * cgroup_migrate() will skip the cset too. Note that we | |
2520 | * can't handle src == dst as some nodes are used by both. | |
2521 | */ | |
2522 | if (src_cset == dst_cset) { | |
2523 | src_cset->mg_src_cgrp = NULL; | |
e4857982 | 2524 | src_cset->mg_dst_cgrp = NULL; |
f817de98 | 2525 | list_del_init(&src_cset->mg_preload_node); |
a25eb52e ZL |
2526 | put_css_set(src_cset); |
2527 | put_css_set(dst_cset); | |
f817de98 TH |
2528 | continue; |
2529 | } | |
2530 | ||
1958d2d5 TH |
2531 | src_cset->mg_dst_cset = dst_cset; |
2532 | ||
2533 | if (list_empty(&dst_cset->mg_preload_node)) | |
e595cd70 TH |
2534 | list_add_tail(&dst_cset->mg_preload_node, |
2535 | &mgctx->preloaded_dst_csets); | |
1958d2d5 | 2536 | else |
a25eb52e | 2537 | put_css_set(dst_cset); |
bfc2cf6f TH |
2538 | |
2539 | for_each_subsys(ss, ssid) | |
2540 | if (src_cset->subsys[ssid] != dst_cset->subsys[ssid]) | |
2541 | mgctx->ss_mask |= 1 << ssid; | |
1958d2d5 TH |
2542 | } |
2543 | ||
1958d2d5 TH |
2544 | return 0; |
2545 | err: | |
e595cd70 | 2546 | cgroup_migrate_finish(mgctx); |
1958d2d5 TH |
2547 | return -ENOMEM; |
2548 | } | |
2549 | ||
2550 | /** | |
2551 | * cgroup_migrate - migrate a process or task to a cgroup | |
1958d2d5 TH |
2552 | * @leader: the leader of the process or the task to migrate |
2553 | * @threadgroup: whether @leader points to the whole process or a single task | |
e595cd70 | 2554 | * @mgctx: migration context |
1958d2d5 | 2555 | * |
37ff9f8f TH |
2556 | * Migrate a process or task denoted by @leader. If migrating a process, |
2557 | * the caller must be holding cgroup_threadgroup_rwsem. The caller is also | |
2558 | * responsible for invoking cgroup_migrate_add_src() and | |
1958d2d5 TH |
2559 | * cgroup_migrate_prepare_dst() on the targets before invoking this |
2560 | * function and following up with cgroup_migrate_finish(). | |
2561 | * | |
2562 | * As long as a controller's ->can_attach() doesn't fail, this function is | |
2563 | * guaranteed to succeed. This means that, excluding ->can_attach() | |
2564 | * failure, when migrating multiple targets, the success or failure can be | |
2565 | * decided for all targets by invoking group_migrate_prepare_dst() before | |
2566 | * actually starting migrating. | |
2567 | */ | |
0a268dbd | 2568 | int cgroup_migrate(struct task_struct *leader, bool threadgroup, |
bfc2cf6f | 2569 | struct cgroup_mgctx *mgctx) |
74a1166d | 2570 | { |
adaae5dc | 2571 | struct task_struct *task; |
74a1166d | 2572 | |
fb5d2b4c MSB |
2573 | /* |
2574 | * Prevent freeing of tasks while we take a snapshot. Tasks that are | |
2575 | * already PF_EXITING could be freed from underneath us unless we | |
2576 | * take an rcu_read_lock. | |
2577 | */ | |
82d6489d | 2578 | spin_lock_irq(&css_set_lock); |
fb5d2b4c | 2579 | rcu_read_lock(); |
9db8de37 | 2580 | task = leader; |
74a1166d | 2581 | do { |
e595cd70 | 2582 | cgroup_migrate_add_task(task, mgctx); |
081aa458 LZ |
2583 | if (!threadgroup) |
2584 | break; | |
9db8de37 | 2585 | } while_each_thread(leader, task); |
fb5d2b4c | 2586 | rcu_read_unlock(); |
82d6489d | 2587 | spin_unlock_irq(&css_set_lock); |
74a1166d | 2588 | |
bfc2cf6f | 2589 | return cgroup_migrate_execute(mgctx); |
74a1166d BB |
2590 | } |
2591 | ||
1958d2d5 TH |
2592 | /** |
2593 | * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup | |
2594 | * @dst_cgrp: the cgroup to attach to | |
2595 | * @leader: the task or the leader of the threadgroup to be attached | |
2596 | * @threadgroup: attach the whole threadgroup? | |
2597 | * | |
1ed13287 | 2598 | * Call holding cgroup_mutex and cgroup_threadgroup_rwsem. |
1958d2d5 | 2599 | */ |
0a268dbd TH |
2600 | int cgroup_attach_task(struct cgroup *dst_cgrp, struct task_struct *leader, |
2601 | bool threadgroup) | |
1958d2d5 | 2602 | { |
e595cd70 | 2603 | DEFINE_CGROUP_MGCTX(mgctx); |
1958d2d5 TH |
2604 | struct task_struct *task; |
2605 | int ret; | |
2606 | ||
8cfd8147 TH |
2607 | ret = cgroup_migrate_vet_dst(dst_cgrp); |
2608 | if (ret) | |
2609 | return ret; | |
6c694c88 | 2610 | |
1958d2d5 | 2611 | /* look up all src csets */ |
82d6489d | 2612 | spin_lock_irq(&css_set_lock); |
1958d2d5 TH |
2613 | rcu_read_lock(); |
2614 | task = leader; | |
2615 | do { | |
e595cd70 | 2616 | cgroup_migrate_add_src(task_css_set(task), dst_cgrp, &mgctx); |
1958d2d5 TH |
2617 | if (!threadgroup) |
2618 | break; | |
2619 | } while_each_thread(leader, task); | |
2620 | rcu_read_unlock(); | |
82d6489d | 2621 | spin_unlock_irq(&css_set_lock); |
1958d2d5 TH |
2622 | |
2623 | /* prepare dst csets and commit */ | |
e595cd70 | 2624 | ret = cgroup_migrate_prepare_dst(&mgctx); |
1958d2d5 | 2625 | if (!ret) |
bfc2cf6f | 2626 | ret = cgroup_migrate(leader, threadgroup, &mgctx); |
1958d2d5 | 2627 | |
e595cd70 | 2628 | cgroup_migrate_finish(&mgctx); |
ed1777de TH |
2629 | |
2630 | if (!ret) | |
2631 | trace_cgroup_attach_task(dst_cgrp, leader, threadgroup); | |
2632 | ||
1958d2d5 | 2633 | return ret; |
74a1166d BB |
2634 | } |
2635 | ||
715c809d TH |
2636 | struct task_struct *cgroup_procs_write_start(char *buf, bool threadgroup) |
2637 | __acquires(&cgroup_threadgroup_rwsem) | |
bbcb81d0 | 2638 | { |
bbcb81d0 | 2639 | struct task_struct *tsk; |
acbef755 | 2640 | pid_t pid; |
bbcb81d0 | 2641 | |
acbef755 | 2642 | if (kstrtoint(strstrip(buf), 0, &pid) || pid < 0) |
715c809d | 2643 | return ERR_PTR(-EINVAL); |
74a1166d | 2644 | |
3014dde7 | 2645 | percpu_down_write(&cgroup_threadgroup_rwsem); |
715c809d | 2646 | |
b78949eb | 2647 | rcu_read_lock(); |
bbcb81d0 | 2648 | if (pid) { |
73507f33 | 2649 | tsk = find_task_by_vpid(pid); |
74a1166d | 2650 | if (!tsk) { |
715c809d TH |
2651 | tsk = ERR_PTR(-ESRCH); |
2652 | goto out_unlock_threadgroup; | |
bbcb81d0 | 2653 | } |
dedf22e9 | 2654 | } else { |
b78949eb | 2655 | tsk = current; |
dedf22e9 | 2656 | } |
cd3d0952 TH |
2657 | |
2658 | if (threadgroup) | |
b78949eb | 2659 | tsk = tsk->group_leader; |
c4c27fbd MG |
2660 | |
2661 | /* | |
77f88796 TH |
2662 | * kthreads may acquire PF_NO_SETAFFINITY during initialization. |
2663 | * If userland migrates such a kthread to a non-root cgroup, it can | |
2664 | * become trapped in a cpuset, or RT kthread may be born in a | |
2665 | * cgroup with no rt_runtime allocated. Just say no. | |
c4c27fbd | 2666 | */ |
77f88796 | 2667 | if (tsk->no_cgroup_migration || (tsk->flags & PF_NO_SETAFFINITY)) { |
715c809d TH |
2668 | tsk = ERR_PTR(-EINVAL); |
2669 | goto out_unlock_threadgroup; | |
c4c27fbd MG |
2670 | } |
2671 | ||
b78949eb | 2672 | get_task_struct(tsk); |
715c809d TH |
2673 | goto out_unlock_rcu; |
2674 | ||
2675 | out_unlock_threadgroup: | |
2676 | percpu_up_write(&cgroup_threadgroup_rwsem); | |
2677 | out_unlock_rcu: | |
b78949eb | 2678 | rcu_read_unlock(); |
715c809d TH |
2679 | return tsk; |
2680 | } | |
b78949eb | 2681 | |
715c809d TH |
2682 | void cgroup_procs_write_finish(struct task_struct *task) |
2683 | __releases(&cgroup_threadgroup_rwsem) | |
2684 | { | |
2685 | struct cgroup_subsys *ss; | |
2686 | int ssid; | |
081aa458 | 2687 | |
715c809d TH |
2688 | /* release reference from cgroup_procs_write_start() */ |
2689 | put_task_struct(task); | |
3014dde7 | 2690 | |
3014dde7 | 2691 | percpu_up_write(&cgroup_threadgroup_rwsem); |
5cf1cacb TH |
2692 | for_each_subsys(ss, ssid) |
2693 | if (ss->post_attach) | |
2694 | ss->post_attach(); | |
af351026 PM |
2695 | } |
2696 | ||
6e5c8307 | 2697 | static void cgroup_print_ss_mask(struct seq_file *seq, u16 ss_mask) |
355e0c48 | 2698 | { |
f8f22e53 TH |
2699 | struct cgroup_subsys *ss; |
2700 | bool printed = false; | |
2701 | int ssid; | |
a742c59d | 2702 | |
b4e0eeaf | 2703 | do_each_subsys_mask(ss, ssid, ss_mask) { |
a966a4ed AS |
2704 | if (printed) |
2705 | seq_putc(seq, ' '); | |
2706 | seq_printf(seq, "%s", ss->name); | |
2707 | printed = true; | |
b4e0eeaf | 2708 | } while_each_subsys_mask(); |
f8f22e53 TH |
2709 | if (printed) |
2710 | seq_putc(seq, '\n'); | |
355e0c48 PM |
2711 | } |
2712 | ||
f8f22e53 TH |
2713 | /* show controllers which are enabled from the parent */ |
2714 | static int cgroup_controllers_show(struct seq_file *seq, void *v) | |
ddbcc7e8 | 2715 | { |
f8f22e53 TH |
2716 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
2717 | ||
5531dc91 | 2718 | cgroup_print_ss_mask(seq, cgroup_control(cgrp)); |
f8f22e53 | 2719 | return 0; |
ddbcc7e8 PM |
2720 | } |
2721 | ||
f8f22e53 TH |
2722 | /* show controllers which are enabled for a given cgroup's children */ |
2723 | static int cgroup_subtree_control_show(struct seq_file *seq, void *v) | |
ddbcc7e8 | 2724 | { |
f8f22e53 TH |
2725 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
2726 | ||
667c2491 | 2727 | cgroup_print_ss_mask(seq, cgrp->subtree_control); |
f8f22e53 TH |
2728 | return 0; |
2729 | } | |
2730 | ||
2731 | /** | |
2732 | * cgroup_update_dfl_csses - update css assoc of a subtree in default hierarchy | |
2733 | * @cgrp: root of the subtree to update csses for | |
2734 | * | |
54962604 TH |
2735 | * @cgrp's control masks have changed and its subtree's css associations |
2736 | * need to be updated accordingly. This function looks up all css_sets | |
2737 | * which are attached to the subtree, creates the matching updated css_sets | |
2738 | * and migrates the tasks to the new ones. | |
f8f22e53 TH |
2739 | */ |
2740 | static int cgroup_update_dfl_csses(struct cgroup *cgrp) | |
2741 | { | |
e595cd70 | 2742 | DEFINE_CGROUP_MGCTX(mgctx); |
54962604 TH |
2743 | struct cgroup_subsys_state *d_css; |
2744 | struct cgroup *dsct; | |
f8f22e53 TH |
2745 | struct css_set *src_cset; |
2746 | int ret; | |
2747 | ||
f8f22e53 TH |
2748 | lockdep_assert_held(&cgroup_mutex); |
2749 | ||
3014dde7 TH |
2750 | percpu_down_write(&cgroup_threadgroup_rwsem); |
2751 | ||
f8f22e53 | 2752 | /* look up all csses currently attached to @cgrp's subtree */ |
82d6489d | 2753 | spin_lock_irq(&css_set_lock); |
54962604 | 2754 | cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) { |
f8f22e53 TH |
2755 | struct cgrp_cset_link *link; |
2756 | ||
54962604 | 2757 | list_for_each_entry(link, &dsct->cset_links, cset_link) |
e595cd70 | 2758 | cgroup_migrate_add_src(link->cset, dsct, &mgctx); |
f8f22e53 | 2759 | } |
82d6489d | 2760 | spin_unlock_irq(&css_set_lock); |
f8f22e53 TH |
2761 | |
2762 | /* NULL dst indicates self on default hierarchy */ | |
e595cd70 | 2763 | ret = cgroup_migrate_prepare_dst(&mgctx); |
f8f22e53 TH |
2764 | if (ret) |
2765 | goto out_finish; | |
2766 | ||
82d6489d | 2767 | spin_lock_irq(&css_set_lock); |
e595cd70 | 2768 | list_for_each_entry(src_cset, &mgctx.preloaded_src_csets, mg_preload_node) { |
10265075 | 2769 | struct task_struct *task, *ntask; |
f8f22e53 | 2770 | |
10265075 TH |
2771 | /* all tasks in src_csets need to be migrated */ |
2772 | list_for_each_entry_safe(task, ntask, &src_cset->tasks, cg_list) | |
e595cd70 | 2773 | cgroup_migrate_add_task(task, &mgctx); |
f8f22e53 | 2774 | } |
82d6489d | 2775 | spin_unlock_irq(&css_set_lock); |
f8f22e53 | 2776 | |
bfc2cf6f | 2777 | ret = cgroup_migrate_execute(&mgctx); |
f8f22e53 | 2778 | out_finish: |
e595cd70 | 2779 | cgroup_migrate_finish(&mgctx); |
3014dde7 | 2780 | percpu_up_write(&cgroup_threadgroup_rwsem); |
f8f22e53 TH |
2781 | return ret; |
2782 | } | |
2783 | ||
1b9b96a1 | 2784 | /** |
945ba199 | 2785 | * cgroup_lock_and_drain_offline - lock cgroup_mutex and drain offlined csses |
ce3f1d9d | 2786 | * @cgrp: root of the target subtree |
1b9b96a1 TH |
2787 | * |
2788 | * Because css offlining is asynchronous, userland may try to re-enable a | |
945ba199 TH |
2789 | * controller while the previous css is still around. This function grabs |
2790 | * cgroup_mutex and drains the previous css instances of @cgrp's subtree. | |
1b9b96a1 | 2791 | */ |
0a268dbd | 2792 | void cgroup_lock_and_drain_offline(struct cgroup *cgrp) |
945ba199 | 2793 | __acquires(&cgroup_mutex) |
1b9b96a1 TH |
2794 | { |
2795 | struct cgroup *dsct; | |
ce3f1d9d | 2796 | struct cgroup_subsys_state *d_css; |
1b9b96a1 TH |
2797 | struct cgroup_subsys *ss; |
2798 | int ssid; | |
2799 | ||
945ba199 TH |
2800 | restart: |
2801 | mutex_lock(&cgroup_mutex); | |
1b9b96a1 | 2802 | |
ce3f1d9d | 2803 | cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) { |
1b9b96a1 TH |
2804 | for_each_subsys(ss, ssid) { |
2805 | struct cgroup_subsys_state *css = cgroup_css(dsct, ss); | |
2806 | DEFINE_WAIT(wait); | |
2807 | ||
ce3f1d9d | 2808 | if (!css || !percpu_ref_is_dying(&css->refcnt)) |
1b9b96a1 TH |
2809 | continue; |
2810 | ||
a590b90d | 2811 | cgroup_get_live(dsct); |
1b9b96a1 TH |
2812 | prepare_to_wait(&dsct->offline_waitq, &wait, |
2813 | TASK_UNINTERRUPTIBLE); | |
2814 | ||
2815 | mutex_unlock(&cgroup_mutex); | |
2816 | schedule(); | |
2817 | finish_wait(&dsct->offline_waitq, &wait); | |
1b9b96a1 TH |
2818 | |
2819 | cgroup_put(dsct); | |
945ba199 | 2820 | goto restart; |
1b9b96a1 TH |
2821 | } |
2822 | } | |
1b9b96a1 TH |
2823 | } |
2824 | ||
15a27c36 | 2825 | /** |
efce8f8c | 2826 | * cgroup_save_control - save control masks and dom_cgrp of a subtree |
15a27c36 TH |
2827 | * @cgrp: root of the target subtree |
2828 | * | |
efce8f8c TH |
2829 | * Save ->subtree_control, ->subtree_ss_mask and ->dom_cgrp to the |
2830 | * respective old_ prefixed fields for @cgrp's subtree including @cgrp | |
2831 | * itself. | |
15a27c36 TH |
2832 | */ |
2833 | static void cgroup_save_control(struct cgroup *cgrp) | |
2834 | { | |
2835 | struct cgroup *dsct; | |
2836 | struct cgroup_subsys_state *d_css; | |
2837 | ||
2838 | cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) { | |
2839 | dsct->old_subtree_control = dsct->subtree_control; | |
2840 | dsct->old_subtree_ss_mask = dsct->subtree_ss_mask; | |
efce8f8c | 2841 | dsct->old_dom_cgrp = dsct->dom_cgrp; |
15a27c36 TH |
2842 | } |
2843 | } | |
2844 | ||
2845 | /** | |
2846 | * cgroup_propagate_control - refresh control masks of a subtree | |
2847 | * @cgrp: root of the target subtree | |
2848 | * | |
2849 | * For @cgrp and its subtree, ensure ->subtree_ss_mask matches | |
2850 | * ->subtree_control and propagate controller availability through the | |
2851 | * subtree so that descendants don't have unavailable controllers enabled. | |
2852 | */ | |
2853 | static void cgroup_propagate_control(struct cgroup *cgrp) | |
2854 | { | |
2855 | struct cgroup *dsct; | |
2856 | struct cgroup_subsys_state *d_css; | |
2857 | ||
2858 | cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) { | |
2859 | dsct->subtree_control &= cgroup_control(dsct); | |
5ced2518 TH |
2860 | dsct->subtree_ss_mask = |
2861 | cgroup_calc_subtree_ss_mask(dsct->subtree_control, | |
2862 | cgroup_ss_mask(dsct)); | |
15a27c36 TH |
2863 | } |
2864 | } | |
2865 | ||
2866 | /** | |
efce8f8c | 2867 | * cgroup_restore_control - restore control masks and dom_cgrp of a subtree |
15a27c36 TH |
2868 | * @cgrp: root of the target subtree |
2869 | * | |
efce8f8c TH |
2870 | * Restore ->subtree_control, ->subtree_ss_mask and ->dom_cgrp from the |
2871 | * respective old_ prefixed fields for @cgrp's subtree including @cgrp | |
2872 | * itself. | |
15a27c36 TH |
2873 | */ |
2874 | static void cgroup_restore_control(struct cgroup *cgrp) | |
2875 | { | |
2876 | struct cgroup *dsct; | |
2877 | struct cgroup_subsys_state *d_css; | |
2878 | ||
2879 | cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) { | |
2880 | dsct->subtree_control = dsct->old_subtree_control; | |
2881 | dsct->subtree_ss_mask = dsct->old_subtree_ss_mask; | |
efce8f8c | 2882 | dsct->dom_cgrp = dsct->old_dom_cgrp; |
15a27c36 TH |
2883 | } |
2884 | } | |
2885 | ||
f6d635ad TH |
2886 | static bool css_visible(struct cgroup_subsys_state *css) |
2887 | { | |
2888 | struct cgroup_subsys *ss = css->ss; | |
2889 | struct cgroup *cgrp = css->cgroup; | |
2890 | ||
2891 | if (cgroup_control(cgrp) & (1 << ss->id)) | |
2892 | return true; | |
2893 | if (!(cgroup_ss_mask(cgrp) & (1 << ss->id))) | |
2894 | return false; | |
2895 | return cgroup_on_dfl(cgrp) && ss->implicit_on_dfl; | |
2896 | } | |
2897 | ||
bdb53bd7 TH |
2898 | /** |
2899 | * cgroup_apply_control_enable - enable or show csses according to control | |
ce3f1d9d | 2900 | * @cgrp: root of the target subtree |
bdb53bd7 | 2901 | * |
ce3f1d9d | 2902 | * Walk @cgrp's subtree and create new csses or make the existing ones |
bdb53bd7 TH |
2903 | * visible. A css is created invisible if it's being implicitly enabled |
2904 | * through dependency. An invisible css is made visible when the userland | |
2905 | * explicitly enables it. | |
2906 | * | |
2907 | * Returns 0 on success, -errno on failure. On failure, csses which have | |
2908 | * been processed already aren't cleaned up. The caller is responsible for | |
8a1115ff | 2909 | * cleaning up with cgroup_apply_control_disable(). |
bdb53bd7 TH |
2910 | */ |
2911 | static int cgroup_apply_control_enable(struct cgroup *cgrp) | |
2912 | { | |
2913 | struct cgroup *dsct; | |
ce3f1d9d | 2914 | struct cgroup_subsys_state *d_css; |
bdb53bd7 TH |
2915 | struct cgroup_subsys *ss; |
2916 | int ssid, ret; | |
2917 | ||
ce3f1d9d | 2918 | cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) { |
bdb53bd7 TH |
2919 | for_each_subsys(ss, ssid) { |
2920 | struct cgroup_subsys_state *css = cgroup_css(dsct, ss); | |
2921 | ||
945ba199 TH |
2922 | WARN_ON_ONCE(css && percpu_ref_is_dying(&css->refcnt)); |
2923 | ||
bdb53bd7 TH |
2924 | if (!(cgroup_ss_mask(dsct) & (1 << ss->id))) |
2925 | continue; | |
2926 | ||
2927 | if (!css) { | |
2928 | css = css_create(dsct, ss); | |
2929 | if (IS_ERR(css)) | |
2930 | return PTR_ERR(css); | |
2931 | } | |
2932 | ||
f6d635ad | 2933 | if (css_visible(css)) { |
334c3679 | 2934 | ret = css_populate_dir(css); |
bdb53bd7 TH |
2935 | if (ret) |
2936 | return ret; | |
2937 | } | |
2938 | } | |
2939 | } | |
2940 | ||
2941 | return 0; | |
2942 | } | |
2943 | ||
12b3bb6a TH |
2944 | /** |
2945 | * cgroup_apply_control_disable - kill or hide csses according to control | |
ce3f1d9d | 2946 | * @cgrp: root of the target subtree |
12b3bb6a | 2947 | * |
ce3f1d9d | 2948 | * Walk @cgrp's subtree and kill and hide csses so that they match |
12b3bb6a TH |
2949 | * cgroup_ss_mask() and cgroup_visible_mask(). |
2950 | * | |
2951 | * A css is hidden when the userland requests it to be disabled while other | |
2952 | * subsystems are still depending on it. The css must not actively control | |
2953 | * resources and be in the vanilla state if it's made visible again later. | |
2954 | * Controllers which may be depended upon should provide ->css_reset() for | |
2955 | * this purpose. | |
2956 | */ | |
2957 | static void cgroup_apply_control_disable(struct cgroup *cgrp) | |
2958 | { | |
2959 | struct cgroup *dsct; | |
ce3f1d9d | 2960 | struct cgroup_subsys_state *d_css; |
12b3bb6a TH |
2961 | struct cgroup_subsys *ss; |
2962 | int ssid; | |
2963 | ||
ce3f1d9d | 2964 | cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) { |
12b3bb6a TH |
2965 | for_each_subsys(ss, ssid) { |
2966 | struct cgroup_subsys_state *css = cgroup_css(dsct, ss); | |
2967 | ||
945ba199 TH |
2968 | WARN_ON_ONCE(css && percpu_ref_is_dying(&css->refcnt)); |
2969 | ||
12b3bb6a TH |
2970 | if (!css) |
2971 | continue; | |
2972 | ||
334c3679 TH |
2973 | if (css->parent && |
2974 | !(cgroup_ss_mask(dsct) & (1 << ss->id))) { | |
12b3bb6a | 2975 | kill_css(css); |
f6d635ad | 2976 | } else if (!css_visible(css)) { |
334c3679 | 2977 | css_clear_dir(css); |
12b3bb6a TH |
2978 | if (ss->css_reset) |
2979 | ss->css_reset(css); | |
2980 | } | |
2981 | } | |
2982 | } | |
2983 | } | |
2984 | ||
f7b2814b TH |
2985 | /** |
2986 | * cgroup_apply_control - apply control mask updates to the subtree | |
2987 | * @cgrp: root of the target subtree | |
2988 | * | |
2989 | * subsystems can be enabled and disabled in a subtree using the following | |
2990 | * steps. | |
2991 | * | |
2992 | * 1. Call cgroup_save_control() to stash the current state. | |
2993 | * 2. Update ->subtree_control masks in the subtree as desired. | |
2994 | * 3. Call cgroup_apply_control() to apply the changes. | |
2995 | * 4. Optionally perform other related operations. | |
2996 | * 5. Call cgroup_finalize_control() to finish up. | |
2997 | * | |
2998 | * This function implements step 3 and propagates the mask changes | |
2999 | * throughout @cgrp's subtree, updates csses accordingly and perform | |
3000 | * process migrations. | |
3001 | */ | |
3002 | static int cgroup_apply_control(struct cgroup *cgrp) | |
3003 | { | |
3004 | int ret; | |
3005 | ||
3006 | cgroup_propagate_control(cgrp); | |
3007 | ||
3008 | ret = cgroup_apply_control_enable(cgrp); | |
3009 | if (ret) | |
3010 | return ret; | |
3011 | ||
3012 | /* | |
3013 | * At this point, cgroup_e_css() results reflect the new csses | |
3014 | * making the following cgroup_update_dfl_csses() properly update | |
3015 | * css associations of all tasks in the subtree. | |
3016 | */ | |
3017 | ret = cgroup_update_dfl_csses(cgrp); | |
3018 | if (ret) | |
3019 | return ret; | |
3020 | ||
3021 | return 0; | |
3022 | } | |
3023 | ||
3024 | /** | |
3025 | * cgroup_finalize_control - finalize control mask update | |
3026 | * @cgrp: root of the target subtree | |
3027 | * @ret: the result of the update | |
3028 | * | |
3029 | * Finalize control mask update. See cgroup_apply_control() for more info. | |
3030 | */ | |
3031 | static void cgroup_finalize_control(struct cgroup *cgrp, int ret) | |
3032 | { | |
3033 | if (ret) { | |
3034 | cgroup_restore_control(cgrp); | |
3035 | cgroup_propagate_control(cgrp); | |
3036 | } | |
3037 | ||
3038 | cgroup_apply_control_disable(cgrp); | |
3039 | } | |
3040 | ||
8cfd8147 TH |
3041 | static int cgroup_vet_subtree_control_enable(struct cgroup *cgrp, u16 enable) |
3042 | { | |
3043 | u16 domain_enable = enable & ~cgrp_dfl_threaded_ss_mask; | |
3044 | ||
3045 | /* if nothing is getting enabled, nothing to worry about */ | |
3046 | if (!enable) | |
3047 | return 0; | |
3048 | ||
3049 | /* can @cgrp host any resources? */ | |
3050 | if (!cgroup_is_valid_domain(cgrp->dom_cgrp)) | |
3051 | return -EOPNOTSUPP; | |
3052 | ||
3053 | /* mixables don't care */ | |
3054 | if (cgroup_is_mixable(cgrp)) | |
3055 | return 0; | |
3056 | ||
3057 | if (domain_enable) { | |
3058 | /* can't enable domain controllers inside a thread subtree */ | |
3059 | if (cgroup_is_thread_root(cgrp) || cgroup_is_threaded(cgrp)) | |
3060 | return -EOPNOTSUPP; | |
3061 | } else { | |
3062 | /* | |
3063 | * Threaded controllers can handle internal competitions | |
3064 | * and are always allowed inside a (prospective) thread | |
3065 | * subtree. | |
3066 | */ | |
3067 | if (cgroup_can_be_thread_root(cgrp) || cgroup_is_threaded(cgrp)) | |
3068 | return 0; | |
3069 | } | |
3070 | ||
3071 | /* | |
3072 | * Controllers can't be enabled for a cgroup with tasks to avoid | |
3073 | * child cgroups competing against tasks. | |
3074 | */ | |
3075 | if (cgroup_has_tasks(cgrp)) | |
3076 | return -EBUSY; | |
3077 | ||
3078 | return 0; | |
3079 | } | |
3080 | ||
f8f22e53 | 3081 | /* change the enabled child controllers for a cgroup in the default hierarchy */ |
451af504 TH |
3082 | static ssize_t cgroup_subtree_control_write(struct kernfs_open_file *of, |
3083 | char *buf, size_t nbytes, | |
3084 | loff_t off) | |
f8f22e53 | 3085 | { |
6e5c8307 | 3086 | u16 enable = 0, disable = 0; |
a9746d8d | 3087 | struct cgroup *cgrp, *child; |
f8f22e53 | 3088 | struct cgroup_subsys *ss; |
451af504 | 3089 | char *tok; |
f8f22e53 TH |
3090 | int ssid, ret; |
3091 | ||
3092 | /* | |
d37167ab TH |
3093 | * Parse input - space separated list of subsystem names prefixed |
3094 | * with either + or -. | |
f8f22e53 | 3095 | */ |
451af504 TH |
3096 | buf = strstrip(buf); |
3097 | while ((tok = strsep(&buf, " "))) { | |
d37167ab TH |
3098 | if (tok[0] == '\0') |
3099 | continue; | |
a7165264 | 3100 | do_each_subsys_mask(ss, ssid, ~cgrp_dfl_inhibit_ss_mask) { |
fc5ed1e9 TH |
3101 | if (!cgroup_ssid_enabled(ssid) || |
3102 | strcmp(tok + 1, ss->name)) | |
f8f22e53 TH |
3103 | continue; |
3104 | ||
3105 | if (*tok == '+') { | |
7d331fa9 TH |
3106 | enable |= 1 << ssid; |
3107 | disable &= ~(1 << ssid); | |
f8f22e53 | 3108 | } else if (*tok == '-') { |
7d331fa9 TH |
3109 | disable |= 1 << ssid; |
3110 | enable &= ~(1 << ssid); | |
f8f22e53 TH |
3111 | } else { |
3112 | return -EINVAL; | |
3113 | } | |
3114 | break; | |
b4e0eeaf | 3115 | } while_each_subsys_mask(); |
f8f22e53 TH |
3116 | if (ssid == CGROUP_SUBSYS_COUNT) |
3117 | return -EINVAL; | |
3118 | } | |
3119 | ||
945ba199 | 3120 | cgrp = cgroup_kn_lock_live(of->kn, true); |
a9746d8d TH |
3121 | if (!cgrp) |
3122 | return -ENODEV; | |
f8f22e53 TH |
3123 | |
3124 | for_each_subsys(ss, ssid) { | |
3125 | if (enable & (1 << ssid)) { | |
667c2491 | 3126 | if (cgrp->subtree_control & (1 << ssid)) { |
f8f22e53 TH |
3127 | enable &= ~(1 << ssid); |
3128 | continue; | |
3129 | } | |
3130 | ||
5531dc91 | 3131 | if (!(cgroup_control(cgrp) & (1 << ssid))) { |
c29adf24 TH |
3132 | ret = -ENOENT; |
3133 | goto out_unlock; | |
3134 | } | |
f8f22e53 | 3135 | } else if (disable & (1 << ssid)) { |
667c2491 | 3136 | if (!(cgrp->subtree_control & (1 << ssid))) { |
f8f22e53 TH |
3137 | disable &= ~(1 << ssid); |
3138 | continue; | |
3139 | } | |
3140 | ||
3141 | /* a child has it enabled? */ | |
3142 | cgroup_for_each_live_child(child, cgrp) { | |
667c2491 | 3143 | if (child->subtree_control & (1 << ssid)) { |
f8f22e53 | 3144 | ret = -EBUSY; |
ddab2b6e | 3145 | goto out_unlock; |
f8f22e53 TH |
3146 | } |
3147 | } | |
3148 | } | |
3149 | } | |
3150 | ||
3151 | if (!enable && !disable) { | |
3152 | ret = 0; | |
ddab2b6e | 3153 | goto out_unlock; |
f8f22e53 TH |
3154 | } |
3155 | ||
8cfd8147 TH |
3156 | ret = cgroup_vet_subtree_control_enable(cgrp, enable); |
3157 | if (ret) | |
27f26753 | 3158 | goto out_unlock; |
f8f22e53 | 3159 | |
15a27c36 TH |
3160 | /* save and update control masks and prepare csses */ |
3161 | cgroup_save_control(cgrp); | |
f63070d3 | 3162 | |
15a27c36 TH |
3163 | cgrp->subtree_control |= enable; |
3164 | cgrp->subtree_control &= ~disable; | |
c29adf24 | 3165 | |
f7b2814b | 3166 | ret = cgroup_apply_control(cgrp); |
f7b2814b | 3167 | cgroup_finalize_control(cgrp, ret); |
3c745417 TH |
3168 | if (ret) |
3169 | goto out_unlock; | |
f8f22e53 TH |
3170 | |
3171 | kernfs_activate(cgrp->kn); | |
f8f22e53 | 3172 | out_unlock: |
a9746d8d | 3173 | cgroup_kn_unlock(of->kn); |
451af504 | 3174 | return ret ?: nbytes; |
f8f22e53 TH |
3175 | } |
3176 | ||
c705a00d TH |
3177 | /** |
3178 | * cgroup_enable_threaded - make @cgrp threaded | |
3179 | * @cgrp: the target cgroup | |
3180 | * | |
3181 | * Called when "threaded" is written to the cgroup.type interface file and | |
3182 | * tries to make @cgrp threaded and join the parent's resource domain. | |
3183 | * This function is never called on the root cgroup as cgroup.type doesn't | |
3184 | * exist on it. | |
3185 | */ | |
8cfd8147 TH |
3186 | static int cgroup_enable_threaded(struct cgroup *cgrp) |
3187 | { | |
3188 | struct cgroup *parent = cgroup_parent(cgrp); | |
3189 | struct cgroup *dom_cgrp = parent->dom_cgrp; | |
efce8f8c TH |
3190 | struct cgroup *dsct; |
3191 | struct cgroup_subsys_state *d_css; | |
8cfd8147 TH |
3192 | int ret; |
3193 | ||
3194 | lockdep_assert_held(&cgroup_mutex); | |
3195 | ||
3196 | /* noop if already threaded */ | |
3197 | if (cgroup_is_threaded(cgrp)) | |
3198 | return 0; | |
3199 | ||
7d6dbbe5 TH |
3200 | /* |
3201 | * If @cgroup is populated or has domain controllers enabled, it | |
3202 | * can't be switched. While the below cgroup_can_be_thread_root() | |
3203 | * test can catch the same conditions, that's only when @parent is | |
3204 | * not mixable, so let's check it explicitly. | |
3205 | */ | |
3206 | if (cgroup_is_populated(cgrp) || | |
3207 | cgrp->subtree_control & ~cgrp_dfl_threaded_ss_mask) | |
3208 | return -EOPNOTSUPP; | |
3209 | ||
8cfd8147 TH |
3210 | /* we're joining the parent's domain, ensure its validity */ |
3211 | if (!cgroup_is_valid_domain(dom_cgrp) || | |
3212 | !cgroup_can_be_thread_root(dom_cgrp)) | |
3213 | return -EOPNOTSUPP; | |
3214 | ||
8cfd8147 TH |
3215 | /* |
3216 | * The following shouldn't cause actual migrations and should | |
3217 | * always succeed. | |
3218 | */ | |
3219 | cgroup_save_control(cgrp); | |
3220 | ||
efce8f8c TH |
3221 | cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) |
3222 | if (dsct == cgrp || cgroup_is_threaded(dsct)) | |
3223 | dsct->dom_cgrp = dom_cgrp; | |
3224 | ||
8cfd8147 TH |
3225 | ret = cgroup_apply_control(cgrp); |
3226 | if (!ret) | |
3227 | parent->nr_threaded_children++; | |
8cfd8147 TH |
3228 | |
3229 | cgroup_finalize_control(cgrp, ret); | |
3230 | return ret; | |
3231 | } | |
3232 | ||
3233 | static int cgroup_type_show(struct seq_file *seq, void *v) | |
3234 | { | |
3235 | struct cgroup *cgrp = seq_css(seq)->cgroup; | |
3236 | ||
3237 | if (cgroup_is_threaded(cgrp)) | |
3238 | seq_puts(seq, "threaded\n"); | |
3239 | else if (!cgroup_is_valid_domain(cgrp)) | |
3240 | seq_puts(seq, "domain invalid\n"); | |
3241 | else if (cgroup_is_thread_root(cgrp)) | |
3242 | seq_puts(seq, "domain threaded\n"); | |
3243 | else | |
3244 | seq_puts(seq, "domain\n"); | |
3245 | ||
3246 | return 0; | |
3247 | } | |
3248 | ||
3249 | static ssize_t cgroup_type_write(struct kernfs_open_file *of, char *buf, | |
3250 | size_t nbytes, loff_t off) | |
3251 | { | |
3252 | struct cgroup *cgrp; | |
3253 | int ret; | |
3254 | ||
3255 | /* only switching to threaded mode is supported */ | |
3256 | if (strcmp(strstrip(buf), "threaded")) | |
3257 | return -EINVAL; | |
3258 | ||
3259 | cgrp = cgroup_kn_lock_live(of->kn, false); | |
3260 | if (!cgrp) | |
3261 | return -ENOENT; | |
3262 | ||
3263 | /* threaded can only be enabled */ | |
3264 | ret = cgroup_enable_threaded(cgrp); | |
3265 | ||
3266 | cgroup_kn_unlock(of->kn); | |
3267 | return ret ?: nbytes; | |
3268 | } | |
3269 | ||
1a926e0b RG |
3270 | static int cgroup_max_descendants_show(struct seq_file *seq, void *v) |
3271 | { | |
3272 | struct cgroup *cgrp = seq_css(seq)->cgroup; | |
3273 | int descendants = READ_ONCE(cgrp->max_descendants); | |
3274 | ||
3275 | if (descendants == INT_MAX) | |
3276 | seq_puts(seq, "max\n"); | |
3277 | else | |
3278 | seq_printf(seq, "%d\n", descendants); | |
3279 | ||
3280 | return 0; | |
3281 | } | |
3282 | ||
3283 | static ssize_t cgroup_max_descendants_write(struct kernfs_open_file *of, | |
3284 | char *buf, size_t nbytes, loff_t off) | |
3285 | { | |
3286 | struct cgroup *cgrp; | |
3287 | int descendants; | |
3288 | ssize_t ret; | |
3289 | ||
3290 | buf = strstrip(buf); | |
3291 | if (!strcmp(buf, "max")) { | |
3292 | descendants = INT_MAX; | |
3293 | } else { | |
3294 | ret = kstrtoint(buf, 0, &descendants); | |
3295 | if (ret) | |
3296 | return ret; | |
3297 | } | |
3298 | ||
696b98f2 | 3299 | if (descendants < 0) |
1a926e0b RG |
3300 | return -ERANGE; |
3301 | ||
3302 | cgrp = cgroup_kn_lock_live(of->kn, false); | |
3303 | if (!cgrp) | |
3304 | return -ENOENT; | |
3305 | ||
3306 | cgrp->max_descendants = descendants; | |
3307 | ||
3308 | cgroup_kn_unlock(of->kn); | |
3309 | ||
3310 | return nbytes; | |
3311 | } | |
3312 | ||
3313 | static int cgroup_max_depth_show(struct seq_file *seq, void *v) | |
3314 | { | |
3315 | struct cgroup *cgrp = seq_css(seq)->cgroup; | |
3316 | int depth = READ_ONCE(cgrp->max_depth); | |
3317 | ||
3318 | if (depth == INT_MAX) | |
3319 | seq_puts(seq, "max\n"); | |
3320 | else | |
3321 | seq_printf(seq, "%d\n", depth); | |
3322 | ||
3323 | return 0; | |
3324 | } | |
3325 | ||
3326 | static ssize_t cgroup_max_depth_write(struct kernfs_open_file *of, | |
3327 | char *buf, size_t nbytes, loff_t off) | |
3328 | { | |
3329 | struct cgroup *cgrp; | |
3330 | ssize_t ret; | |
3331 | int depth; | |
3332 | ||
3333 | buf = strstrip(buf); | |
3334 | if (!strcmp(buf, "max")) { | |
3335 | depth = INT_MAX; | |
3336 | } else { | |
3337 | ret = kstrtoint(buf, 0, &depth); | |
3338 | if (ret) | |
3339 | return ret; | |
3340 | } | |
3341 | ||
696b98f2 | 3342 | if (depth < 0) |
1a926e0b RG |
3343 | return -ERANGE; |
3344 | ||
3345 | cgrp = cgroup_kn_lock_live(of->kn, false); | |
3346 | if (!cgrp) | |
3347 | return -ENOENT; | |
3348 | ||
3349 | cgrp->max_depth = depth; | |
3350 | ||
3351 | cgroup_kn_unlock(of->kn); | |
3352 | ||
3353 | return nbytes; | |
3354 | } | |
3355 | ||
4a07c222 | 3356 | static int cgroup_events_show(struct seq_file *seq, void *v) |
842b597e | 3357 | { |
4a07c222 | 3358 | seq_printf(seq, "populated %d\n", |
27bd4dbb | 3359 | cgroup_is_populated(seq_css(seq)->cgroup)); |
842b597e TH |
3360 | return 0; |
3361 | } | |
3362 | ||
3e48930c | 3363 | static int cgroup_stat_show(struct seq_file *seq, void *v) |
ec39225c RG |
3364 | { |
3365 | struct cgroup *cgroup = seq_css(seq)->cgroup; | |
3366 | ||
3367 | seq_printf(seq, "nr_descendants %d\n", | |
3368 | cgroup->nr_descendants); | |
3369 | seq_printf(seq, "nr_dying_descendants %d\n", | |
3370 | cgroup->nr_dying_descendants); | |
3371 | ||
3372 | return 0; | |
3373 | } | |
3374 | ||
d41bf8c9 TH |
3375 | static int __maybe_unused cgroup_extra_stat_show(struct seq_file *seq, |
3376 | struct cgroup *cgrp, int ssid) | |
3377 | { | |
3378 | struct cgroup_subsys *ss = cgroup_subsys[ssid]; | |
3379 | struct cgroup_subsys_state *css; | |
3380 | int ret; | |
3381 | ||
3382 | if (!ss->css_extra_stat_show) | |
3383 | return 0; | |
3384 | ||
3385 | css = cgroup_tryget_css(cgrp, ss); | |
3386 | if (!css) | |
3387 | return 0; | |
3388 | ||
3389 | ret = ss->css_extra_stat_show(seq, css); | |
3390 | css_put(css); | |
3391 | return ret; | |
3392 | } | |
3393 | ||
3394 | static int cpu_stat_show(struct seq_file *seq, void *v) | |
3395 | { | |
c3ba1329 | 3396 | struct cgroup __maybe_unused *cgrp = seq_css(seq)->cgroup; |
d41bf8c9 TH |
3397 | int ret = 0; |
3398 | ||
3399 | cgroup_stat_show_cputime(seq); | |
3400 | #ifdef CONFIG_CGROUP_SCHED | |
3401 | ret = cgroup_extra_stat_show(seq, cgrp, cpu_cgrp_id); | |
3402 | #endif | |
3403 | return ret; | |
3404 | } | |
3405 | ||
e90cbebc TH |
3406 | static int cgroup_file_open(struct kernfs_open_file *of) |
3407 | { | |
3408 | struct cftype *cft = of->kn->priv; | |
3409 | ||
3410 | if (cft->open) | |
3411 | return cft->open(of); | |
3412 | return 0; | |
3413 | } | |
3414 | ||
3415 | static void cgroup_file_release(struct kernfs_open_file *of) | |
3416 | { | |
3417 | struct cftype *cft = of->kn->priv; | |
3418 | ||
3419 | if (cft->release) | |
3420 | cft->release(of); | |
3421 | } | |
3422 | ||
2bd59d48 TH |
3423 | static ssize_t cgroup_file_write(struct kernfs_open_file *of, char *buf, |
3424 | size_t nbytes, loff_t off) | |
355e0c48 | 3425 | { |
5136f636 | 3426 | struct cgroup_namespace *ns = current->nsproxy->cgroup_ns; |
2bd59d48 TH |
3427 | struct cgroup *cgrp = of->kn->parent->priv; |
3428 | struct cftype *cft = of->kn->priv; | |
3429 | struct cgroup_subsys_state *css; | |
a742c59d | 3430 | int ret; |
355e0c48 | 3431 | |
5136f636 TH |
3432 | /* |
3433 | * If namespaces are delegation boundaries, disallow writes to | |
3434 | * files in an non-init namespace root from inside the namespace | |
3435 | * except for the files explicitly marked delegatable - | |
3436 | * cgroup.procs and cgroup.subtree_control. | |
3437 | */ | |
3438 | if ((cgrp->root->flags & CGRP_ROOT_NS_DELEGATE) && | |
3439 | !(cft->flags & CFTYPE_NS_DELEGATABLE) && | |
3440 | ns != &init_cgroup_ns && ns->root_cset->dfl_cgrp == cgrp) | |
3441 | return -EPERM; | |
3442 | ||
b4168640 TH |
3443 | if (cft->write) |
3444 | return cft->write(of, buf, nbytes, off); | |
3445 | ||
2bd59d48 TH |
3446 | /* |
3447 | * kernfs guarantees that a file isn't deleted with operations in | |
3448 | * flight, which means that the matching css is and stays alive and | |
3449 | * doesn't need to be pinned. The RCU locking is not necessary | |
3450 | * either. It's just for the convenience of using cgroup_css(). | |
3451 | */ | |
3452 | rcu_read_lock(); | |
3453 | css = cgroup_css(cgrp, cft->ss); | |
3454 | rcu_read_unlock(); | |
a742c59d | 3455 | |
451af504 | 3456 | if (cft->write_u64) { |
a742c59d TH |
3457 | unsigned long long v; |
3458 | ret = kstrtoull(buf, 0, &v); | |
3459 | if (!ret) | |
3460 | ret = cft->write_u64(css, cft, v); | |
3461 | } else if (cft->write_s64) { | |
3462 | long long v; | |
3463 | ret = kstrtoll(buf, 0, &v); | |
3464 | if (!ret) | |
3465 | ret = cft->write_s64(css, cft, v); | |
e73d2c61 | 3466 | } else { |
a742c59d | 3467 | ret = -EINVAL; |
e73d2c61 | 3468 | } |
2bd59d48 | 3469 | |
a742c59d | 3470 | return ret ?: nbytes; |
355e0c48 PM |
3471 | } |
3472 | ||
6612f05b | 3473 | static void *cgroup_seqfile_start(struct seq_file *seq, loff_t *ppos) |
db3b1497 | 3474 | { |
2bd59d48 | 3475 | return seq_cft(seq)->seq_start(seq, ppos); |
db3b1497 PM |
3476 | } |
3477 | ||
6612f05b | 3478 | static void *cgroup_seqfile_next(struct seq_file *seq, void *v, loff_t *ppos) |
ddbcc7e8 | 3479 | { |
2bd59d48 | 3480 | return seq_cft(seq)->seq_next(seq, v, ppos); |
ddbcc7e8 PM |
3481 | } |
3482 | ||
6612f05b | 3483 | static void cgroup_seqfile_stop(struct seq_file *seq, void *v) |
ddbcc7e8 | 3484 | { |
e90cbebc TH |
3485 | if (seq_cft(seq)->seq_stop) |
3486 | seq_cft(seq)->seq_stop(seq, v); | |
ddbcc7e8 PM |
3487 | } |
3488 | ||
91796569 | 3489 | static int cgroup_seqfile_show(struct seq_file *m, void *arg) |
e73d2c61 | 3490 | { |
7da11279 TH |
3491 | struct cftype *cft = seq_cft(m); |
3492 | struct cgroup_subsys_state *css = seq_css(m); | |
e73d2c61 | 3493 | |
2da8ca82 TH |
3494 | if (cft->seq_show) |
3495 | return cft->seq_show(m, arg); | |
e73d2c61 | 3496 | |
f4c753b7 | 3497 | if (cft->read_u64) |
896f5199 TH |
3498 | seq_printf(m, "%llu\n", cft->read_u64(css, cft)); |
3499 | else if (cft->read_s64) | |
3500 | seq_printf(m, "%lld\n", cft->read_s64(css, cft)); | |
3501 | else | |
3502 | return -EINVAL; | |
3503 | return 0; | |
91796569 PM |
3504 | } |
3505 | ||
2bd59d48 TH |
3506 | static struct kernfs_ops cgroup_kf_single_ops = { |
3507 | .atomic_write_len = PAGE_SIZE, | |
e90cbebc TH |
3508 | .open = cgroup_file_open, |
3509 | .release = cgroup_file_release, | |
2bd59d48 TH |
3510 | .write = cgroup_file_write, |
3511 | .seq_show = cgroup_seqfile_show, | |
91796569 PM |
3512 | }; |
3513 | ||
2bd59d48 TH |
3514 | static struct kernfs_ops cgroup_kf_ops = { |
3515 | .atomic_write_len = PAGE_SIZE, | |
e90cbebc TH |
3516 | .open = cgroup_file_open, |
3517 | .release = cgroup_file_release, | |
2bd59d48 TH |
3518 | .write = cgroup_file_write, |
3519 | .seq_start = cgroup_seqfile_start, | |
3520 | .seq_next = cgroup_seqfile_next, | |
3521 | .seq_stop = cgroup_seqfile_stop, | |
3522 | .seq_show = cgroup_seqfile_show, | |
3523 | }; | |
ddbcc7e8 | 3524 | |
49957f8e TH |
3525 | /* set uid and gid of cgroup dirs and files to that of the creator */ |
3526 | static int cgroup_kn_set_ugid(struct kernfs_node *kn) | |
3527 | { | |
3528 | struct iattr iattr = { .ia_valid = ATTR_UID | ATTR_GID, | |
3529 | .ia_uid = current_fsuid(), | |
3530 | .ia_gid = current_fsgid(), }; | |
3531 | ||
3532 | if (uid_eq(iattr.ia_uid, GLOBAL_ROOT_UID) && | |
3533 | gid_eq(iattr.ia_gid, GLOBAL_ROOT_GID)) | |
3534 | return 0; | |
3535 | ||
3536 | return kernfs_setattr(kn, &iattr); | |
3537 | } | |
3538 | ||
4df8dc90 TH |
3539 | static int cgroup_add_file(struct cgroup_subsys_state *css, struct cgroup *cgrp, |
3540 | struct cftype *cft) | |
ddbcc7e8 | 3541 | { |
8d7e6fb0 | 3542 | char name[CGROUP_FILE_NAME_MAX]; |
2bd59d48 TH |
3543 | struct kernfs_node *kn; |
3544 | struct lock_class_key *key = NULL; | |
49957f8e | 3545 | int ret; |
05ef1d7c | 3546 | |
2bd59d48 TH |
3547 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
3548 | key = &cft->lockdep_key; | |
3549 | #endif | |
3550 | kn = __kernfs_create_file(cgrp->kn, cgroup_file_name(cgrp, cft, name), | |
630d0eb6 DT |
3551 | cgroup_file_mode(cft), |
3552 | GLOBAL_ROOT_UID, GLOBAL_ROOT_GID, | |
3553 | 0, cft->kf_ops, cft, | |
dfeb0750 | 3554 | NULL, key); |
49957f8e TH |
3555 | if (IS_ERR(kn)) |
3556 | return PTR_ERR(kn); | |
3557 | ||
3558 | ret = cgroup_kn_set_ugid(kn); | |
f8f22e53 | 3559 | if (ret) { |
49957f8e | 3560 | kernfs_remove(kn); |
f8f22e53 TH |
3561 | return ret; |
3562 | } | |
3563 | ||
6f60eade TH |
3564 | if (cft->file_offset) { |
3565 | struct cgroup_file *cfile = (void *)css + cft->file_offset; | |
3566 | ||
34c06254 | 3567 | spin_lock_irq(&cgroup_file_kn_lock); |
6f60eade | 3568 | cfile->kn = kn; |
34c06254 | 3569 | spin_unlock_irq(&cgroup_file_kn_lock); |
6f60eade TH |
3570 | } |
3571 | ||
f8f22e53 | 3572 | return 0; |
ddbcc7e8 PM |
3573 | } |
3574 | ||
b1f28d31 TH |
3575 | /** |
3576 | * cgroup_addrm_files - add or remove files to a cgroup directory | |
4df8dc90 TH |
3577 | * @css: the target css |
3578 | * @cgrp: the target cgroup (usually css->cgroup) | |
b1f28d31 TH |
3579 | * @cfts: array of cftypes to be added |
3580 | * @is_add: whether to add or remove | |
3581 | * | |
3582 | * Depending on @is_add, add or remove files defined by @cfts on @cgrp. | |
6732ed85 | 3583 | * For removals, this function never fails. |
b1f28d31 | 3584 | */ |
4df8dc90 TH |
3585 | static int cgroup_addrm_files(struct cgroup_subsys_state *css, |
3586 | struct cgroup *cgrp, struct cftype cfts[], | |
2bb566cb | 3587 | bool is_add) |
ddbcc7e8 | 3588 | { |
6732ed85 | 3589 | struct cftype *cft, *cft_end = NULL; |
b598dde3 | 3590 | int ret = 0; |
b1f28d31 | 3591 | |
01f6474c | 3592 | lockdep_assert_held(&cgroup_mutex); |
db0416b6 | 3593 | |
6732ed85 TH |
3594 | restart: |
3595 | for (cft = cfts; cft != cft_end && cft->name[0] != '\0'; cft++) { | |
f33fddc2 | 3596 | /* does cft->flags tell us to skip this file on @cgrp? */ |
05ebb6e6 | 3597 | if ((cft->flags & __CFTYPE_ONLY_ON_DFL) && !cgroup_on_dfl(cgrp)) |
8cbbf2c9 | 3598 | continue; |
05ebb6e6 | 3599 | if ((cft->flags & __CFTYPE_NOT_ON_DFL) && cgroup_on_dfl(cgrp)) |
873fe09e | 3600 | continue; |
d51f39b0 | 3601 | if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgroup_parent(cgrp)) |
f33fddc2 | 3602 | continue; |
d51f39b0 | 3603 | if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgroup_parent(cgrp)) |
f33fddc2 G |
3604 | continue; |
3605 | ||
2739d3cc | 3606 | if (is_add) { |
4df8dc90 | 3607 | ret = cgroup_add_file(css, cgrp, cft); |
b1f28d31 | 3608 | if (ret) { |
ed3d261b JP |
3609 | pr_warn("%s: failed to add %s, err=%d\n", |
3610 | __func__, cft->name, ret); | |
6732ed85 TH |
3611 | cft_end = cft; |
3612 | is_add = false; | |
3613 | goto restart; | |
b1f28d31 | 3614 | } |
2739d3cc LZ |
3615 | } else { |
3616 | cgroup_rm_file(cgrp, cft); | |
db0416b6 | 3617 | } |
ddbcc7e8 | 3618 | } |
b598dde3 | 3619 | return ret; |
ddbcc7e8 PM |
3620 | } |
3621 | ||
21a2d343 | 3622 | static int cgroup_apply_cftypes(struct cftype *cfts, bool is_add) |
8e3f6541 | 3623 | { |
2bb566cb | 3624 | struct cgroup_subsys *ss = cfts[0].ss; |
3dd06ffa | 3625 | struct cgroup *root = &ss->root->cgrp; |
492eb21b | 3626 | struct cgroup_subsys_state *css; |
9ccece80 | 3627 | int ret = 0; |
8e3f6541 | 3628 | |
01f6474c | 3629 | lockdep_assert_held(&cgroup_mutex); |
e8c82d20 | 3630 | |
e8c82d20 | 3631 | /* add/rm files for all cgroups created before */ |
ca8bdcaf | 3632 | css_for_each_descendant_pre(css, cgroup_css(root, ss)) { |
492eb21b TH |
3633 | struct cgroup *cgrp = css->cgroup; |
3634 | ||
88cb04b9 | 3635 | if (!(css->flags & CSS_VISIBLE)) |
e8c82d20 LZ |
3636 | continue; |
3637 | ||
4df8dc90 | 3638 | ret = cgroup_addrm_files(css, cgrp, cfts, is_add); |
9ccece80 TH |
3639 | if (ret) |
3640 | break; | |
8e3f6541 | 3641 | } |
21a2d343 TH |
3642 | |
3643 | if (is_add && !ret) | |
3644 | kernfs_activate(root->kn); | |
9ccece80 | 3645 | return ret; |
8e3f6541 TH |
3646 | } |
3647 | ||
2da440a2 | 3648 | static void cgroup_exit_cftypes(struct cftype *cfts) |
8e3f6541 | 3649 | { |
2bb566cb | 3650 | struct cftype *cft; |
8e3f6541 | 3651 | |
2bd59d48 TH |
3652 | for (cft = cfts; cft->name[0] != '\0'; cft++) { |
3653 | /* free copy for custom atomic_write_len, see init_cftypes() */ | |
3654 | if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) | |
3655 | kfree(cft->kf_ops); | |
3656 | cft->kf_ops = NULL; | |
2da440a2 | 3657 | cft->ss = NULL; |
a8ddc821 TH |
3658 | |
3659 | /* revert flags set by cgroup core while adding @cfts */ | |
05ebb6e6 | 3660 | cft->flags &= ~(__CFTYPE_ONLY_ON_DFL | __CFTYPE_NOT_ON_DFL); |
2bd59d48 | 3661 | } |
2da440a2 TH |
3662 | } |
3663 | ||
2bd59d48 | 3664 | static int cgroup_init_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
2da440a2 TH |
3665 | { |
3666 | struct cftype *cft; | |
3667 | ||
2bd59d48 TH |
3668 | for (cft = cfts; cft->name[0] != '\0'; cft++) { |
3669 | struct kernfs_ops *kf_ops; | |
3670 | ||
0adb0704 TH |
3671 | WARN_ON(cft->ss || cft->kf_ops); |
3672 | ||
2bd59d48 TH |
3673 | if (cft->seq_start) |
3674 | kf_ops = &cgroup_kf_ops; | |
3675 | else | |
3676 | kf_ops = &cgroup_kf_single_ops; | |
3677 | ||
3678 | /* | |
3679 | * Ugh... if @cft wants a custom max_write_len, we need to | |
3680 | * make a copy of kf_ops to set its atomic_write_len. | |
3681 | */ | |
3682 | if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) { | |
3683 | kf_ops = kmemdup(kf_ops, sizeof(*kf_ops), GFP_KERNEL); | |
3684 | if (!kf_ops) { | |
3685 | cgroup_exit_cftypes(cfts); | |
3686 | return -ENOMEM; | |
3687 | } | |
3688 | kf_ops->atomic_write_len = cft->max_write_len; | |
3689 | } | |
8e3f6541 | 3690 | |
2bd59d48 | 3691 | cft->kf_ops = kf_ops; |
2bb566cb | 3692 | cft->ss = ss; |
2bd59d48 | 3693 | } |
2bb566cb | 3694 | |
2bd59d48 | 3695 | return 0; |
2da440a2 TH |
3696 | } |
3697 | ||
21a2d343 TH |
3698 | static int cgroup_rm_cftypes_locked(struct cftype *cfts) |
3699 | { | |
01f6474c | 3700 | lockdep_assert_held(&cgroup_mutex); |
21a2d343 TH |
3701 | |
3702 | if (!cfts || !cfts[0].ss) | |
3703 | return -ENOENT; | |
3704 | ||
3705 | list_del(&cfts->node); | |
3706 | cgroup_apply_cftypes(cfts, false); | |
3707 | cgroup_exit_cftypes(cfts); | |
3708 | return 0; | |
8e3f6541 | 3709 | } |
8e3f6541 | 3710 | |
79578621 TH |
3711 | /** |
3712 | * cgroup_rm_cftypes - remove an array of cftypes from a subsystem | |
79578621 TH |
3713 | * @cfts: zero-length name terminated array of cftypes |
3714 | * | |
2bb566cb TH |
3715 | * Unregister @cfts. Files described by @cfts are removed from all |
3716 | * existing cgroups and all future cgroups won't have them either. This | |
3717 | * function can be called anytime whether @cfts' subsys is attached or not. | |
79578621 TH |
3718 | * |
3719 | * Returns 0 on successful unregistration, -ENOENT if @cfts is not | |
2bb566cb | 3720 | * registered. |
79578621 | 3721 | */ |
2bb566cb | 3722 | int cgroup_rm_cftypes(struct cftype *cfts) |
79578621 | 3723 | { |
21a2d343 | 3724 | int ret; |
79578621 | 3725 | |
01f6474c | 3726 | mutex_lock(&cgroup_mutex); |
21a2d343 | 3727 | ret = cgroup_rm_cftypes_locked(cfts); |
01f6474c | 3728 | mutex_unlock(&cgroup_mutex); |
21a2d343 | 3729 | return ret; |
80b13586 TH |
3730 | } |
3731 | ||
8e3f6541 TH |
3732 | /** |
3733 | * cgroup_add_cftypes - add an array of cftypes to a subsystem | |
3734 | * @ss: target cgroup subsystem | |
3735 | * @cfts: zero-length name terminated array of cftypes | |
3736 | * | |
3737 | * Register @cfts to @ss. Files described by @cfts are created for all | |
3738 | * existing cgroups to which @ss is attached and all future cgroups will | |
3739 | * have them too. This function can be called anytime whether @ss is | |
3740 | * attached or not. | |
3741 | * | |
3742 | * Returns 0 on successful registration, -errno on failure. Note that this | |
3743 | * function currently returns 0 as long as @cfts registration is successful | |
3744 | * even if some file creation attempts on existing cgroups fail. | |
3745 | */ | |
2cf669a5 | 3746 | static int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
8e3f6541 | 3747 | { |
9ccece80 | 3748 | int ret; |
8e3f6541 | 3749 | |
fc5ed1e9 | 3750 | if (!cgroup_ssid_enabled(ss->id)) |
c731ae1d LZ |
3751 | return 0; |
3752 | ||
dc5736ed LZ |
3753 | if (!cfts || cfts[0].name[0] == '\0') |
3754 | return 0; | |
2bb566cb | 3755 | |
2bd59d48 TH |
3756 | ret = cgroup_init_cftypes(ss, cfts); |
3757 | if (ret) | |
3758 | return ret; | |
79578621 | 3759 | |
01f6474c | 3760 | mutex_lock(&cgroup_mutex); |
21a2d343 | 3761 | |
0adb0704 | 3762 | list_add_tail(&cfts->node, &ss->cfts); |
21a2d343 | 3763 | ret = cgroup_apply_cftypes(cfts, true); |
9ccece80 | 3764 | if (ret) |
21a2d343 | 3765 | cgroup_rm_cftypes_locked(cfts); |
79578621 | 3766 | |
01f6474c | 3767 | mutex_unlock(&cgroup_mutex); |
9ccece80 | 3768 | return ret; |
79578621 TH |
3769 | } |
3770 | ||
a8ddc821 TH |
3771 | /** |
3772 | * cgroup_add_dfl_cftypes - add an array of cftypes for default hierarchy | |
3773 | * @ss: target cgroup subsystem | |
3774 | * @cfts: zero-length name terminated array of cftypes | |
3775 | * | |
3776 | * Similar to cgroup_add_cftypes() but the added files are only used for | |
3777 | * the default hierarchy. | |
3778 | */ | |
3779 | int cgroup_add_dfl_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) | |
3780 | { | |
3781 | struct cftype *cft; | |
3782 | ||
3783 | for (cft = cfts; cft && cft->name[0] != '\0'; cft++) | |
05ebb6e6 | 3784 | cft->flags |= __CFTYPE_ONLY_ON_DFL; |
a8ddc821 TH |
3785 | return cgroup_add_cftypes(ss, cfts); |
3786 | } | |
3787 | ||
3788 | /** | |
3789 | * cgroup_add_legacy_cftypes - add an array of cftypes for legacy hierarchies | |
3790 | * @ss: target cgroup subsystem | |
3791 | * @cfts: zero-length name terminated array of cftypes | |
3792 | * | |
3793 | * Similar to cgroup_add_cftypes() but the added files are only used for | |
3794 | * the legacy hierarchies. | |
3795 | */ | |
2cf669a5 TH |
3796 | int cgroup_add_legacy_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
3797 | { | |
a8ddc821 TH |
3798 | struct cftype *cft; |
3799 | ||
e4b7037c TH |
3800 | for (cft = cfts; cft && cft->name[0] != '\0'; cft++) |
3801 | cft->flags |= __CFTYPE_NOT_ON_DFL; | |
2cf669a5 TH |
3802 | return cgroup_add_cftypes(ss, cfts); |
3803 | } | |
3804 | ||
34c06254 TH |
3805 | /** |
3806 | * cgroup_file_notify - generate a file modified event for a cgroup_file | |
3807 | * @cfile: target cgroup_file | |
3808 | * | |
3809 | * @cfile must have been obtained by setting cftype->file_offset. | |
3810 | */ | |
3811 | void cgroup_file_notify(struct cgroup_file *cfile) | |
3812 | { | |
3813 | unsigned long flags; | |
3814 | ||
3815 | spin_lock_irqsave(&cgroup_file_kn_lock, flags); | |
3816 | if (cfile->kn) | |
3817 | kernfs_notify(cfile->kn); | |
3818 | spin_unlock_irqrestore(&cgroup_file_kn_lock, flags); | |
3819 | } | |
3820 | ||
53fa5261 | 3821 | /** |
492eb21b | 3822 | * css_next_child - find the next child of a given css |
c2931b70 TH |
3823 | * @pos: the current position (%NULL to initiate traversal) |
3824 | * @parent: css whose children to walk | |
53fa5261 | 3825 | * |
c2931b70 | 3826 | * This function returns the next child of @parent and should be called |
87fb54f1 | 3827 | * under either cgroup_mutex or RCU read lock. The only requirement is |
c2931b70 TH |
3828 | * that @parent and @pos are accessible. The next sibling is guaranteed to |
3829 | * be returned regardless of their states. | |
3830 | * | |
3831 | * If a subsystem synchronizes ->css_online() and the start of iteration, a | |
3832 | * css which finished ->css_online() is guaranteed to be visible in the | |
3833 | * future iterations and will stay visible until the last reference is put. | |
3834 | * A css which hasn't finished ->css_online() or already finished | |
3835 | * ->css_offline() may show up during traversal. It's each subsystem's | |
3836 | * responsibility to synchronize against on/offlining. | |
53fa5261 | 3837 | */ |
c2931b70 TH |
3838 | struct cgroup_subsys_state *css_next_child(struct cgroup_subsys_state *pos, |
3839 | struct cgroup_subsys_state *parent) | |
53fa5261 | 3840 | { |
c2931b70 | 3841 | struct cgroup_subsys_state *next; |
53fa5261 | 3842 | |
8353da1f | 3843 | cgroup_assert_mutex_or_rcu_locked(); |
53fa5261 TH |
3844 | |
3845 | /* | |
de3f0341 TH |
3846 | * @pos could already have been unlinked from the sibling list. |
3847 | * Once a cgroup is removed, its ->sibling.next is no longer | |
3848 | * updated when its next sibling changes. CSS_RELEASED is set when | |
3849 | * @pos is taken off list, at which time its next pointer is valid, | |
3850 | * and, as releases are serialized, the one pointed to by the next | |
3851 | * pointer is guaranteed to not have started release yet. This | |
3852 | * implies that if we observe !CSS_RELEASED on @pos in this RCU | |
3853 | * critical section, the one pointed to by its next pointer is | |
3854 | * guaranteed to not have finished its RCU grace period even if we | |
3855 | * have dropped rcu_read_lock() inbetween iterations. | |
3b287a50 | 3856 | * |
de3f0341 TH |
3857 | * If @pos has CSS_RELEASED set, its next pointer can't be |
3858 | * dereferenced; however, as each css is given a monotonically | |
3859 | * increasing unique serial number and always appended to the | |
3860 | * sibling list, the next one can be found by walking the parent's | |
3861 | * children until the first css with higher serial number than | |
3862 | * @pos's. While this path can be slower, it happens iff iteration | |
3863 | * races against release and the race window is very small. | |
53fa5261 | 3864 | */ |
3b287a50 | 3865 | if (!pos) { |
c2931b70 TH |
3866 | next = list_entry_rcu(parent->children.next, struct cgroup_subsys_state, sibling); |
3867 | } else if (likely(!(pos->flags & CSS_RELEASED))) { | |
3868 | next = list_entry_rcu(pos->sibling.next, struct cgroup_subsys_state, sibling); | |
3b287a50 | 3869 | } else { |
c2931b70 | 3870 | list_for_each_entry_rcu(next, &parent->children, sibling) |
3b287a50 TH |
3871 | if (next->serial_nr > pos->serial_nr) |
3872 | break; | |
53fa5261 TH |
3873 | } |
3874 | ||
3b281afb TH |
3875 | /* |
3876 | * @next, if not pointing to the head, can be dereferenced and is | |
c2931b70 | 3877 | * the next sibling. |
3b281afb | 3878 | */ |
c2931b70 TH |
3879 | if (&next->sibling != &parent->children) |
3880 | return next; | |
3b281afb | 3881 | return NULL; |
53fa5261 | 3882 | } |
53fa5261 | 3883 | |
574bd9f7 | 3884 | /** |
492eb21b | 3885 | * css_next_descendant_pre - find the next descendant for pre-order walk |
574bd9f7 | 3886 | * @pos: the current position (%NULL to initiate traversal) |
492eb21b | 3887 | * @root: css whose descendants to walk |
574bd9f7 | 3888 | * |
492eb21b | 3889 | * To be used by css_for_each_descendant_pre(). Find the next descendant |
bd8815a6 TH |
3890 | * to visit for pre-order traversal of @root's descendants. @root is |
3891 | * included in the iteration and the first node to be visited. | |
75501a6d | 3892 | * |
87fb54f1 TH |
3893 | * While this function requires cgroup_mutex or RCU read locking, it |
3894 | * doesn't require the whole traversal to be contained in a single critical | |
3895 | * section. This function will return the correct next descendant as long | |
3896 | * as both @pos and @root are accessible and @pos is a descendant of @root. | |
c2931b70 TH |
3897 | * |
3898 | * If a subsystem synchronizes ->css_online() and the start of iteration, a | |
3899 | * css which finished ->css_online() is guaranteed to be visible in the | |
3900 | * future iterations and will stay visible until the last reference is put. | |
3901 | * A css which hasn't finished ->css_online() or already finished | |
3902 | * ->css_offline() may show up during traversal. It's each subsystem's | |
3903 | * responsibility to synchronize against on/offlining. | |
574bd9f7 | 3904 | */ |
492eb21b TH |
3905 | struct cgroup_subsys_state * |
3906 | css_next_descendant_pre(struct cgroup_subsys_state *pos, | |
3907 | struct cgroup_subsys_state *root) | |
574bd9f7 | 3908 | { |
492eb21b | 3909 | struct cgroup_subsys_state *next; |
574bd9f7 | 3910 | |
8353da1f | 3911 | cgroup_assert_mutex_or_rcu_locked(); |
574bd9f7 | 3912 | |
bd8815a6 | 3913 | /* if first iteration, visit @root */ |
7805d000 | 3914 | if (!pos) |
bd8815a6 | 3915 | return root; |
574bd9f7 TH |
3916 | |
3917 | /* visit the first child if exists */ | |
492eb21b | 3918 | next = css_next_child(NULL, pos); |
574bd9f7 TH |
3919 | if (next) |
3920 | return next; | |
3921 | ||
3922 | /* no child, visit my or the closest ancestor's next sibling */ | |
492eb21b | 3923 | while (pos != root) { |
5c9d535b | 3924 | next = css_next_child(pos, pos->parent); |
75501a6d | 3925 | if (next) |
574bd9f7 | 3926 | return next; |
5c9d535b | 3927 | pos = pos->parent; |
7805d000 | 3928 | } |
574bd9f7 TH |
3929 | |
3930 | return NULL; | |
3931 | } | |
574bd9f7 | 3932 | |
12a9d2fe | 3933 | /** |
492eb21b TH |
3934 | * css_rightmost_descendant - return the rightmost descendant of a css |
3935 | * @pos: css of interest | |
12a9d2fe | 3936 | * |
492eb21b TH |
3937 | * Return the rightmost descendant of @pos. If there's no descendant, @pos |
3938 | * is returned. This can be used during pre-order traversal to skip | |
12a9d2fe | 3939 | * subtree of @pos. |
75501a6d | 3940 | * |
87fb54f1 TH |
3941 | * While this function requires cgroup_mutex or RCU read locking, it |
3942 | * doesn't require the whole traversal to be contained in a single critical | |
3943 | * section. This function will return the correct rightmost descendant as | |
3944 | * long as @pos is accessible. | |
12a9d2fe | 3945 | */ |
492eb21b TH |
3946 | struct cgroup_subsys_state * |
3947 | css_rightmost_descendant(struct cgroup_subsys_state *pos) | |
12a9d2fe | 3948 | { |
492eb21b | 3949 | struct cgroup_subsys_state *last, *tmp; |
12a9d2fe | 3950 | |
8353da1f | 3951 | cgroup_assert_mutex_or_rcu_locked(); |
12a9d2fe TH |
3952 | |
3953 | do { | |
3954 | last = pos; | |
3955 | /* ->prev isn't RCU safe, walk ->next till the end */ | |
3956 | pos = NULL; | |
492eb21b | 3957 | css_for_each_child(tmp, last) |
12a9d2fe TH |
3958 | pos = tmp; |
3959 | } while (pos); | |
3960 | ||
3961 | return last; | |
3962 | } | |
12a9d2fe | 3963 | |
492eb21b TH |
3964 | static struct cgroup_subsys_state * |
3965 | css_leftmost_descendant(struct cgroup_subsys_state *pos) | |
574bd9f7 | 3966 | { |
492eb21b | 3967 | struct cgroup_subsys_state *last; |
574bd9f7 TH |
3968 | |
3969 | do { | |
3970 | last = pos; | |
492eb21b | 3971 | pos = css_next_child(NULL, pos); |
574bd9f7 TH |
3972 | } while (pos); |
3973 | ||
3974 | return last; | |
3975 | } | |
3976 | ||
3977 | /** | |
492eb21b | 3978 | * css_next_descendant_post - find the next descendant for post-order walk |
574bd9f7 | 3979 | * @pos: the current position (%NULL to initiate traversal) |
492eb21b | 3980 | * @root: css whose descendants to walk |
574bd9f7 | 3981 | * |
492eb21b | 3982 | * To be used by css_for_each_descendant_post(). Find the next descendant |
bd8815a6 TH |
3983 | * to visit for post-order traversal of @root's descendants. @root is |
3984 | * included in the iteration and the last node to be visited. | |
75501a6d | 3985 | * |
87fb54f1 TH |
3986 | * While this function requires cgroup_mutex or RCU read locking, it |
3987 | * doesn't require the whole traversal to be contained in a single critical | |
3988 | * section. This function will return the correct next descendant as long | |
3989 | * as both @pos and @cgroup are accessible and @pos is a descendant of | |
3990 | * @cgroup. | |
c2931b70 TH |
3991 | * |
3992 | * If a subsystem synchronizes ->css_online() and the start of iteration, a | |
3993 | * css which finished ->css_online() is guaranteed to be visible in the | |
3994 | * future iterations and will stay visible until the last reference is put. | |
3995 | * A css which hasn't finished ->css_online() or already finished | |
3996 | * ->css_offline() may show up during traversal. It's each subsystem's | |
3997 | * responsibility to synchronize against on/offlining. | |
574bd9f7 | 3998 | */ |
492eb21b TH |
3999 | struct cgroup_subsys_state * |
4000 | css_next_descendant_post(struct cgroup_subsys_state *pos, | |
4001 | struct cgroup_subsys_state *root) | |
574bd9f7 | 4002 | { |
492eb21b | 4003 | struct cgroup_subsys_state *next; |
574bd9f7 | 4004 | |
8353da1f | 4005 | cgroup_assert_mutex_or_rcu_locked(); |
574bd9f7 | 4006 | |
58b79a91 TH |
4007 | /* if first iteration, visit leftmost descendant which may be @root */ |
4008 | if (!pos) | |
4009 | return css_leftmost_descendant(root); | |
574bd9f7 | 4010 | |
bd8815a6 TH |
4011 | /* if we visited @root, we're done */ |
4012 | if (pos == root) | |
4013 | return NULL; | |
4014 | ||
574bd9f7 | 4015 | /* if there's an unvisited sibling, visit its leftmost descendant */ |
5c9d535b | 4016 | next = css_next_child(pos, pos->parent); |
75501a6d | 4017 | if (next) |
492eb21b | 4018 | return css_leftmost_descendant(next); |
574bd9f7 TH |
4019 | |
4020 | /* no sibling left, visit parent */ | |
5c9d535b | 4021 | return pos->parent; |
574bd9f7 | 4022 | } |
574bd9f7 | 4023 | |
f3d46500 TH |
4024 | /** |
4025 | * css_has_online_children - does a css have online children | |
4026 | * @css: the target css | |
4027 | * | |
4028 | * Returns %true if @css has any online children; otherwise, %false. This | |
4029 | * function can be called from any context but the caller is responsible | |
4030 | * for synchronizing against on/offlining as necessary. | |
4031 | */ | |
4032 | bool css_has_online_children(struct cgroup_subsys_state *css) | |
cbc125ef | 4033 | { |
f3d46500 TH |
4034 | struct cgroup_subsys_state *child; |
4035 | bool ret = false; | |
cbc125ef TH |
4036 | |
4037 | rcu_read_lock(); | |
f3d46500 | 4038 | css_for_each_child(child, css) { |
99bae5f9 | 4039 | if (child->flags & CSS_ONLINE) { |
f3d46500 TH |
4040 | ret = true; |
4041 | break; | |
cbc125ef TH |
4042 | } |
4043 | } | |
4044 | rcu_read_unlock(); | |
f3d46500 | 4045 | return ret; |
574bd9f7 | 4046 | } |
574bd9f7 | 4047 | |
450ee0c1 TH |
4048 | static struct css_set *css_task_iter_next_css_set(struct css_task_iter *it) |
4049 | { | |
4050 | struct list_head *l; | |
4051 | struct cgrp_cset_link *link; | |
4052 | struct css_set *cset; | |
4053 | ||
4054 | lockdep_assert_held(&css_set_lock); | |
4055 | ||
4056 | /* find the next threaded cset */ | |
4057 | if (it->tcset_pos) { | |
4058 | l = it->tcset_pos->next; | |
4059 | ||
4060 | if (l != it->tcset_head) { | |
4061 | it->tcset_pos = l; | |
4062 | return container_of(l, struct css_set, | |
4063 | threaded_csets_node); | |
4064 | } | |
4065 | ||
4066 | it->tcset_pos = NULL; | |
4067 | } | |
4068 | ||
4069 | /* find the next cset */ | |
4070 | l = it->cset_pos; | |
4071 | l = l->next; | |
4072 | if (l == it->cset_head) { | |
4073 | it->cset_pos = NULL; | |
4074 | return NULL; | |
4075 | } | |
4076 | ||
4077 | if (it->ss) { | |
4078 | cset = container_of(l, struct css_set, e_cset_node[it->ss->id]); | |
4079 | } else { | |
4080 | link = list_entry(l, struct cgrp_cset_link, cset_link); | |
4081 | cset = link->cset; | |
4082 | } | |
4083 | ||
4084 | it->cset_pos = l; | |
4085 | ||
4086 | /* initialize threaded css_set walking */ | |
4087 | if (it->flags & CSS_TASK_ITER_THREADED) { | |
4088 | if (it->cur_dcset) | |
4089 | put_css_set_locked(it->cur_dcset); | |
4090 | it->cur_dcset = cset; | |
4091 | get_css_set(cset); | |
4092 | ||
4093 | it->tcset_head = &cset->threaded_csets; | |
4094 | it->tcset_pos = &cset->threaded_csets; | |
4095 | } | |
4096 | ||
4097 | return cset; | |
4098 | } | |
4099 | ||
0942eeee | 4100 | /** |
ecb9d535 | 4101 | * css_task_iter_advance_css_set - advance a task itererator to the next css_set |
0942eeee TH |
4102 | * @it: the iterator to advance |
4103 | * | |
4104 | * Advance @it to the next css_set to walk. | |
d515876e | 4105 | */ |
ecb9d535 | 4106 | static void css_task_iter_advance_css_set(struct css_task_iter *it) |
d515876e | 4107 | { |
d515876e TH |
4108 | struct css_set *cset; |
4109 | ||
f0d9a5f1 | 4110 | lockdep_assert_held(&css_set_lock); |
ed27b9f7 | 4111 | |
d515876e TH |
4112 | /* Advance to the next non-empty css_set */ |
4113 | do { | |
450ee0c1 TH |
4114 | cset = css_task_iter_next_css_set(it); |
4115 | if (!cset) { | |
ecb9d535 | 4116 | it->task_pos = NULL; |
d515876e TH |
4117 | return; |
4118 | } | |
5bd3bb19 | 4119 | } while (!css_set_populated(cset) && list_empty(&cset->dying_tasks)); |
c7561128 | 4120 | |
c7561128 | 4121 | if (!list_empty(&cset->tasks)) |
0f0a2b4f | 4122 | it->task_pos = cset->tasks.next; |
798cc880 | 4123 | else if (!list_empty(&cset->mg_tasks)) |
0f0a2b4f | 4124 | it->task_pos = cset->mg_tasks.next; |
798cc880 TH |
4125 | else |
4126 | it->task_pos = cset->dying_tasks.next; | |
0f0a2b4f TH |
4127 | |
4128 | it->tasks_head = &cset->tasks; | |
4129 | it->mg_tasks_head = &cset->mg_tasks; | |
798cc880 | 4130 | it->dying_tasks_head = &cset->dying_tasks; |
ed27b9f7 TH |
4131 | |
4132 | /* | |
4133 | * We don't keep css_sets locked across iteration steps and thus | |
4134 | * need to take steps to ensure that iteration can be resumed after | |
4135 | * the lock is re-acquired. Iteration is performed at two levels - | |
4136 | * css_sets and tasks in them. | |
4137 | * | |
4138 | * Once created, a css_set never leaves its cgroup lists, so a | |
4139 | * pinned css_set is guaranteed to stay put and we can resume | |
4140 | * iteration afterwards. | |
4141 | * | |
4142 | * Tasks may leave @cset across iteration steps. This is resolved | |
4143 | * by registering each iterator with the css_set currently being | |
4144 | * walked and making css_set_move_task() advance iterators whose | |
4145 | * next task is leaving. | |
4146 | */ | |
4147 | if (it->cur_cset) { | |
4148 | list_del(&it->iters_node); | |
4149 | put_css_set_locked(it->cur_cset); | |
4150 | } | |
4151 | get_css_set(cset); | |
4152 | it->cur_cset = cset; | |
4153 | list_add(&it->iters_node, &cset->task_iters); | |
d515876e TH |
4154 | } |
4155 | ||
3995c283 TH |
4156 | static void css_task_iter_skip(struct css_task_iter *it, |
4157 | struct task_struct *task) | |
ecb9d535 | 4158 | { |
3995c283 TH |
4159 | lockdep_assert_held(&css_set_lock); |
4160 | ||
4161 | if (it->task_pos == &task->cg_list) { | |
4162 | it->task_pos = it->task_pos->next; | |
4163 | it->flags |= CSS_TASK_ITER_SKIPPED; | |
4164 | } | |
4165 | } | |
ecb9d535 | 4166 | |
3995c283 TH |
4167 | static void css_task_iter_advance(struct css_task_iter *it) |
4168 | { | |
798cc880 TH |
4169 | struct task_struct *task; |
4170 | ||
f0d9a5f1 | 4171 | lockdep_assert_held(&css_set_lock); |
bc2fb7ed | 4172 | repeat: |
3f080bdc TH |
4173 | if (it->task_pos) { |
4174 | /* | |
4175 | * Advance iterator to find next entry. cset->tasks is | |
4176 | * consumed first and then ->mg_tasks. After ->mg_tasks, | |
4177 | * we move onto the next cset. | |
4178 | */ | |
3995c283 TH |
4179 | if (it->flags & CSS_TASK_ITER_SKIPPED) |
4180 | it->flags &= ~CSS_TASK_ITER_SKIPPED; | |
4181 | else | |
4182 | it->task_pos = it->task_pos->next; | |
ecb9d535 | 4183 | |
3995c283 TH |
4184 | if (it->task_pos == it->tasks_head) |
4185 | it->task_pos = it->mg_tasks_head->next; | |
4186 | if (it->task_pos == it->mg_tasks_head) | |
798cc880 TH |
4187 | it->task_pos = it->dying_tasks_head->next; |
4188 | if (it->task_pos == it->dying_tasks_head) | |
3f080bdc | 4189 | css_task_iter_advance_css_set(it); |
3f080bdc TH |
4190 | } else { |
4191 | /* called from start, proceed to the first cset */ | |
ecb9d535 | 4192 | css_task_iter_advance_css_set(it); |
3f080bdc | 4193 | } |
bc2fb7ed | 4194 | |
798cc880 TH |
4195 | if (!it->task_pos) |
4196 | return; | |
4197 | ||
4198 | task = list_entry(it->task_pos, struct task_struct, cg_list); | |
4199 | ||
4200 | if (it->flags & CSS_TASK_ITER_PROCS) { | |
4201 | /* if PROCS, skip over tasks which aren't group leaders */ | |
4202 | if (!thread_group_leader(task)) | |
4203 | goto repeat; | |
4204 | ||
4205 | /* and dying leaders w/o live member threads */ | |
4206 | if (!atomic_read(&task->signal->live)) | |
4207 | goto repeat; | |
4208 | } else { | |
4209 | /* skip all dying ones */ | |
4210 | if (task->flags & PF_EXITING) | |
4211 | goto repeat; | |
4212 | } | |
ecb9d535 TH |
4213 | } |
4214 | ||
0942eeee | 4215 | /** |
72ec7029 TH |
4216 | * css_task_iter_start - initiate task iteration |
4217 | * @css: the css to walk tasks of | |
bc2fb7ed | 4218 | * @flags: CSS_TASK_ITER_* flags |
0942eeee TH |
4219 | * @it: the task iterator to use |
4220 | * | |
72ec7029 TH |
4221 | * Initiate iteration through the tasks of @css. The caller can call |
4222 | * css_task_iter_next() to walk through the tasks until the function | |
4223 | * returns NULL. On completion of iteration, css_task_iter_end() must be | |
4224 | * called. | |
0942eeee | 4225 | */ |
bc2fb7ed | 4226 | void css_task_iter_start(struct cgroup_subsys_state *css, unsigned int flags, |
72ec7029 | 4227 | struct css_task_iter *it) |
817929ec | 4228 | { |
56fde9e0 TH |
4229 | /* no one should try to iterate before mounting cgroups */ |
4230 | WARN_ON_ONCE(!use_task_css_set_links); | |
31a7df01 | 4231 | |
ed27b9f7 TH |
4232 | memset(it, 0, sizeof(*it)); |
4233 | ||
82d6489d | 4234 | spin_lock_irq(&css_set_lock); |
c59cd3d8 | 4235 | |
3ebb2b6e | 4236 | it->ss = css->ss; |
bc2fb7ed | 4237 | it->flags = flags; |
3ebb2b6e TH |
4238 | |
4239 | if (it->ss) | |
4240 | it->cset_pos = &css->cgroup->e_csets[css->ss->id]; | |
4241 | else | |
4242 | it->cset_pos = &css->cgroup->cset_links; | |
4243 | ||
0f0a2b4f | 4244 | it->cset_head = it->cset_pos; |
c59cd3d8 | 4245 | |
3f080bdc | 4246 | css_task_iter_advance(it); |
ed27b9f7 | 4247 | |
82d6489d | 4248 | spin_unlock_irq(&css_set_lock); |
817929ec PM |
4249 | } |
4250 | ||
0942eeee | 4251 | /** |
72ec7029 | 4252 | * css_task_iter_next - return the next task for the iterator |
0942eeee TH |
4253 | * @it: the task iterator being iterated |
4254 | * | |
4255 | * The "next" function for task iteration. @it should have been | |
72ec7029 TH |
4256 | * initialized via css_task_iter_start(). Returns NULL when the iteration |
4257 | * reaches the end. | |
0942eeee | 4258 | */ |
72ec7029 | 4259 | struct task_struct *css_task_iter_next(struct css_task_iter *it) |
817929ec | 4260 | { |
d5745675 | 4261 | if (it->cur_task) { |
ed27b9f7 | 4262 | put_task_struct(it->cur_task); |
d5745675 TH |
4263 | it->cur_task = NULL; |
4264 | } | |
ed27b9f7 | 4265 | |
82d6489d | 4266 | spin_lock_irq(&css_set_lock); |
ed27b9f7 | 4267 | |
bc44383b TH |
4268 | /* @it may be half-advanced by skips, finish advancing */ |
4269 | if (it->flags & CSS_TASK_ITER_SKIPPED) | |
4270 | css_task_iter_advance(it); | |
4271 | ||
d5745675 TH |
4272 | if (it->task_pos) { |
4273 | it->cur_task = list_entry(it->task_pos, struct task_struct, | |
4274 | cg_list); | |
4275 | get_task_struct(it->cur_task); | |
4276 | css_task_iter_advance(it); | |
4277 | } | |
ed27b9f7 | 4278 | |
82d6489d | 4279 | spin_unlock_irq(&css_set_lock); |
ed27b9f7 TH |
4280 | |
4281 | return it->cur_task; | |
817929ec PM |
4282 | } |
4283 | ||
0942eeee | 4284 | /** |
72ec7029 | 4285 | * css_task_iter_end - finish task iteration |
0942eeee TH |
4286 | * @it: the task iterator to finish |
4287 | * | |
72ec7029 | 4288 | * Finish task iteration started by css_task_iter_start(). |
0942eeee | 4289 | */ |
72ec7029 | 4290 | void css_task_iter_end(struct css_task_iter *it) |
31a7df01 | 4291 | { |
ed27b9f7 | 4292 | if (it->cur_cset) { |
82d6489d | 4293 | spin_lock_irq(&css_set_lock); |
ed27b9f7 TH |
4294 | list_del(&it->iters_node); |
4295 | put_css_set_locked(it->cur_cset); | |
82d6489d | 4296 | spin_unlock_irq(&css_set_lock); |
ed27b9f7 TH |
4297 | } |
4298 | ||
450ee0c1 TH |
4299 | if (it->cur_dcset) |
4300 | put_css_set(it->cur_dcset); | |
4301 | ||
ed27b9f7 TH |
4302 | if (it->cur_task) |
4303 | put_task_struct(it->cur_task); | |
31a7df01 CW |
4304 | } |
4305 | ||
b4b90a8e | 4306 | static void cgroup_procs_release(struct kernfs_open_file *of) |
31a7df01 | 4307 | { |
b4b90a8e TH |
4308 | if (of->priv) { |
4309 | css_task_iter_end(of->priv); | |
4310 | kfree(of->priv); | |
4311 | } | |
4312 | } | |
6c694c88 | 4313 | |
b4b90a8e TH |
4314 | static void *cgroup_procs_next(struct seq_file *s, void *v, loff_t *pos) |
4315 | { | |
4316 | struct kernfs_open_file *of = s->private; | |
4317 | struct css_task_iter *it = of->priv; | |
31a7df01 | 4318 | |
bc2fb7ed | 4319 | return css_task_iter_next(it); |
b4b90a8e | 4320 | } |
31a7df01 | 4321 | |
8cfd8147 TH |
4322 | static void *__cgroup_procs_start(struct seq_file *s, loff_t *pos, |
4323 | unsigned int iter_flags) | |
b4b90a8e TH |
4324 | { |
4325 | struct kernfs_open_file *of = s->private; | |
4326 | struct cgroup *cgrp = seq_css(s)->cgroup; | |
4327 | struct css_task_iter *it = of->priv; | |
8cc99345 | 4328 | |
952aaa12 | 4329 | /* |
b4b90a8e TH |
4330 | * When a seq_file is seeked, it's always traversed sequentially |
4331 | * from position 0, so we can simply keep iterating on !0 *pos. | |
952aaa12 | 4332 | */ |
b4b90a8e TH |
4333 | if (!it) { |
4334 | if (WARN_ON_ONCE((*pos)++)) | |
4335 | return ERR_PTR(-EINVAL); | |
8cc99345 | 4336 | |
b4b90a8e TH |
4337 | it = kzalloc(sizeof(*it), GFP_KERNEL); |
4338 | if (!it) | |
4339 | return ERR_PTR(-ENOMEM); | |
4340 | of->priv = it; | |
450ee0c1 | 4341 | css_task_iter_start(&cgrp->self, iter_flags, it); |
b4b90a8e TH |
4342 | } else if (!(*pos)++) { |
4343 | css_task_iter_end(it); | |
450ee0c1 | 4344 | css_task_iter_start(&cgrp->self, iter_flags, it); |
b4b90a8e | 4345 | } |
bbcb81d0 | 4346 | |
b4b90a8e TH |
4347 | return cgroup_procs_next(s, NULL, NULL); |
4348 | } | |
24528255 | 4349 | |
8cfd8147 TH |
4350 | static void *cgroup_procs_start(struct seq_file *s, loff_t *pos) |
4351 | { | |
4352 | struct cgroup *cgrp = seq_css(s)->cgroup; | |
4353 | ||
4354 | /* | |
4355 | * All processes of a threaded subtree belong to the domain cgroup | |
4356 | * of the subtree. Only threads can be distributed across the | |
4357 | * subtree. Reject reads on cgroup.procs in the subtree proper. | |
4358 | * They're always empty anyway. | |
4359 | */ | |
4360 | if (cgroup_is_threaded(cgrp)) | |
4361 | return ERR_PTR(-EOPNOTSUPP); | |
4362 | ||
4363 | return __cgroup_procs_start(s, pos, CSS_TASK_ITER_PROCS | | |
4364 | CSS_TASK_ITER_THREADED); | |
4365 | } | |
4366 | ||
b4b90a8e | 4367 | static int cgroup_procs_show(struct seq_file *s, void *v) |
bbcb81d0 | 4368 | { |
bc2fb7ed | 4369 | seq_printf(s, "%d\n", task_pid_vnr(v)); |
97978e6d DL |
4370 | return 0; |
4371 | } | |
4372 | ||
715c809d TH |
4373 | static int cgroup_procs_write_permission(struct cgroup *src_cgrp, |
4374 | struct cgroup *dst_cgrp, | |
4375 | struct super_block *sb) | |
4376 | { | |
4377 | struct cgroup_namespace *ns = current->nsproxy->cgroup_ns; | |
4378 | struct cgroup *com_cgrp = src_cgrp; | |
4379 | struct inode *inode; | |
4380 | int ret; | |
4381 | ||
4382 | lockdep_assert_held(&cgroup_mutex); | |
4383 | ||
4384 | /* find the common ancestor */ | |
4385 | while (!cgroup_is_descendant(dst_cgrp, com_cgrp)) | |
4386 | com_cgrp = cgroup_parent(com_cgrp); | |
4387 | ||
4388 | /* %current should be authorized to migrate to the common ancestor */ | |
4389 | inode = kernfs_get_inode(sb, com_cgrp->procs_file.kn); | |
4390 | if (!inode) | |
4391 | return -ENOMEM; | |
4392 | ||
4393 | ret = inode_permission(inode, MAY_WRITE); | |
4394 | iput(inode); | |
4395 | if (ret) | |
4396 | return ret; | |
4397 | ||
4398 | /* | |
4399 | * If namespaces are delegation boundaries, %current must be able | |
4400 | * to see both source and destination cgroups from its namespace. | |
4401 | */ | |
4402 | if ((cgrp_dfl_root.flags & CGRP_ROOT_NS_DELEGATE) && | |
4403 | (!cgroup_is_descendant(src_cgrp, ns->root_cset->dfl_cgrp) || | |
4404 | !cgroup_is_descendant(dst_cgrp, ns->root_cset->dfl_cgrp))) | |
4405 | return -ENOENT; | |
4406 | ||
4407 | return 0; | |
4408 | } | |
4409 | ||
4410 | static ssize_t cgroup_procs_write(struct kernfs_open_file *of, | |
4411 | char *buf, size_t nbytes, loff_t off) | |
4412 | { | |
4413 | struct cgroup *src_cgrp, *dst_cgrp; | |
4414 | struct task_struct *task; | |
4415 | ssize_t ret; | |
4416 | ||
4417 | dst_cgrp = cgroup_kn_lock_live(of->kn, false); | |
4418 | if (!dst_cgrp) | |
4419 | return -ENODEV; | |
4420 | ||
4421 | task = cgroup_procs_write_start(buf, true); | |
4422 | ret = PTR_ERR_OR_ZERO(task); | |
4423 | if (ret) | |
4424 | goto out_unlock; | |
4425 | ||
4426 | /* find the source cgroup */ | |
4427 | spin_lock_irq(&css_set_lock); | |
4428 | src_cgrp = task_cgroup_from_root(task, &cgrp_dfl_root); | |
4429 | spin_unlock_irq(&css_set_lock); | |
4430 | ||
4431 | ret = cgroup_procs_write_permission(src_cgrp, dst_cgrp, | |
4432 | of->file->f_path.dentry->d_sb); | |
4433 | if (ret) | |
4434 | goto out_finish; | |
4435 | ||
4436 | ret = cgroup_attach_task(dst_cgrp, task, true); | |
4437 | ||
4438 | out_finish: | |
4439 | cgroup_procs_write_finish(task); | |
4440 | out_unlock: | |
4441 | cgroup_kn_unlock(of->kn); | |
4442 | ||
4443 | return ret ?: nbytes; | |
4444 | } | |
4445 | ||
8cfd8147 TH |
4446 | static void *cgroup_threads_start(struct seq_file *s, loff_t *pos) |
4447 | { | |
4448 | return __cgroup_procs_start(s, pos, 0); | |
4449 | } | |
4450 | ||
4451 | static ssize_t cgroup_threads_write(struct kernfs_open_file *of, | |
4452 | char *buf, size_t nbytes, loff_t off) | |
4453 | { | |
4454 | struct cgroup *src_cgrp, *dst_cgrp; | |
4455 | struct task_struct *task; | |
4456 | ssize_t ret; | |
4457 | ||
4458 | buf = strstrip(buf); | |
4459 | ||
4460 | dst_cgrp = cgroup_kn_lock_live(of->kn, false); | |
4461 | if (!dst_cgrp) | |
4462 | return -ENODEV; | |
4463 | ||
4464 | task = cgroup_procs_write_start(buf, false); | |
4465 | ret = PTR_ERR_OR_ZERO(task); | |
4466 | if (ret) | |
4467 | goto out_unlock; | |
4468 | ||
4469 | /* find the source cgroup */ | |
4470 | spin_lock_irq(&css_set_lock); | |
4471 | src_cgrp = task_cgroup_from_root(task, &cgrp_dfl_root); | |
4472 | spin_unlock_irq(&css_set_lock); | |
4473 | ||
4474 | /* thread migrations follow the cgroup.procs delegation rule */ | |
4475 | ret = cgroup_procs_write_permission(src_cgrp, dst_cgrp, | |
4476 | of->file->f_path.dentry->d_sb); | |
4477 | if (ret) | |
4478 | goto out_finish; | |
4479 | ||
4480 | /* and must be contained in the same domain */ | |
4481 | ret = -EOPNOTSUPP; | |
4482 | if (src_cgrp->dom_cgrp != dst_cgrp->dom_cgrp) | |
4483 | goto out_finish; | |
4484 | ||
4485 | ret = cgroup_attach_task(dst_cgrp, task, false); | |
4486 | ||
4487 | out_finish: | |
4488 | cgroup_procs_write_finish(task); | |
4489 | out_unlock: | |
4490 | cgroup_kn_unlock(of->kn); | |
4491 | ||
4492 | return ret ?: nbytes; | |
4493 | } | |
4494 | ||
a14c6874 | 4495 | /* cgroup core interface files for the default hierarchy */ |
d62beb7f | 4496 | static struct cftype cgroup_base_files[] = { |
8cfd8147 TH |
4497 | { |
4498 | .name = "cgroup.type", | |
4499 | .flags = CFTYPE_NOT_ON_ROOT, | |
4500 | .seq_show = cgroup_type_show, | |
4501 | .write = cgroup_type_write, | |
4502 | }, | |
81a6a5cd | 4503 | { |
d5c56ced | 4504 | .name = "cgroup.procs", |
5136f636 | 4505 | .flags = CFTYPE_NS_DELEGATABLE, |
6f60eade | 4506 | .file_offset = offsetof(struct cgroup, procs_file), |
b4b90a8e TH |
4507 | .release = cgroup_procs_release, |
4508 | .seq_start = cgroup_procs_start, | |
4509 | .seq_next = cgroup_procs_next, | |
4510 | .seq_show = cgroup_procs_show, | |
acbef755 | 4511 | .write = cgroup_procs_write, |
102a775e | 4512 | }, |
8cfd8147 TH |
4513 | { |
4514 | .name = "cgroup.threads", | |
4f58424d | 4515 | .flags = CFTYPE_NS_DELEGATABLE, |
8cfd8147 TH |
4516 | .release = cgroup_procs_release, |
4517 | .seq_start = cgroup_threads_start, | |
4518 | .seq_next = cgroup_procs_next, | |
4519 | .seq_show = cgroup_procs_show, | |
4520 | .write = cgroup_threads_write, | |
4521 | }, | |
f8f22e53 TH |
4522 | { |
4523 | .name = "cgroup.controllers", | |
f8f22e53 TH |
4524 | .seq_show = cgroup_controllers_show, |
4525 | }, | |
4526 | { | |
4527 | .name = "cgroup.subtree_control", | |
5136f636 | 4528 | .flags = CFTYPE_NS_DELEGATABLE, |
f8f22e53 | 4529 | .seq_show = cgroup_subtree_control_show, |
451af504 | 4530 | .write = cgroup_subtree_control_write, |
f8f22e53 | 4531 | }, |
842b597e | 4532 | { |
4a07c222 | 4533 | .name = "cgroup.events", |
a14c6874 | 4534 | .flags = CFTYPE_NOT_ON_ROOT, |
6f60eade | 4535 | .file_offset = offsetof(struct cgroup, events_file), |
4a07c222 | 4536 | .seq_show = cgroup_events_show, |
842b597e | 4537 | }, |
1a926e0b RG |
4538 | { |
4539 | .name = "cgroup.max.descendants", | |
4540 | .seq_show = cgroup_max_descendants_show, | |
4541 | .write = cgroup_max_descendants_write, | |
4542 | }, | |
4543 | { | |
4544 | .name = "cgroup.max.depth", | |
4545 | .seq_show = cgroup_max_depth_show, | |
4546 | .write = cgroup_max_depth_write, | |
4547 | }, | |
ec39225c RG |
4548 | { |
4549 | .name = "cgroup.stat", | |
3e48930c | 4550 | .seq_show = cgroup_stat_show, |
ec39225c | 4551 | }, |
d41bf8c9 TH |
4552 | { |
4553 | .name = "cpu.stat", | |
4554 | .flags = CFTYPE_NOT_ON_ROOT, | |
4555 | .seq_show = cpu_stat_show, | |
4556 | }, | |
a14c6874 TH |
4557 | { } /* terminate */ |
4558 | }; | |
d5c56ced | 4559 | |
0c21ead1 TH |
4560 | /* |
4561 | * css destruction is four-stage process. | |
4562 | * | |
4563 | * 1. Destruction starts. Killing of the percpu_ref is initiated. | |
4564 | * Implemented in kill_css(). | |
4565 | * | |
4566 | * 2. When the percpu_ref is confirmed to be visible as killed on all CPUs | |
ec903c0c TH |
4567 | * and thus css_tryget_online() is guaranteed to fail, the css can be |
4568 | * offlined by invoking offline_css(). After offlining, the base ref is | |
4569 | * put. Implemented in css_killed_work_fn(). | |
0c21ead1 TH |
4570 | * |
4571 | * 3. When the percpu_ref reaches zero, the only possible remaining | |
4572 | * accessors are inside RCU read sections. css_release() schedules the | |
4573 | * RCU callback. | |
4574 | * | |
4575 | * 4. After the grace period, the css can be freed. Implemented in | |
4576 | * css_free_work_fn(). | |
4577 | * | |
4578 | * It is actually hairier because both step 2 and 4 require process context | |
4579 | * and thus involve punting to css->destroy_work adding two additional | |
4580 | * steps to the already complex sequence. | |
4581 | */ | |
35ef10da | 4582 | static void css_free_work_fn(struct work_struct *work) |
48ddbe19 TH |
4583 | { |
4584 | struct cgroup_subsys_state *css = | |
35ef10da | 4585 | container_of(work, struct cgroup_subsys_state, destroy_work); |
01e58659 | 4586 | struct cgroup_subsys *ss = css->ss; |
0c21ead1 | 4587 | struct cgroup *cgrp = css->cgroup; |
48ddbe19 | 4588 | |
9a1049da TH |
4589 | percpu_ref_exit(&css->refcnt); |
4590 | ||
01e58659 | 4591 | if (ss) { |
9d755d33 | 4592 | /* css free path */ |
8bb5ef79 | 4593 | struct cgroup_subsys_state *parent = css->parent; |
01e58659 VD |
4594 | int id = css->id; |
4595 | ||
01e58659 VD |
4596 | ss->css_free(css); |
4597 | cgroup_idr_remove(&ss->css_idr, id); | |
9d755d33 | 4598 | cgroup_put(cgrp); |
8bb5ef79 TH |
4599 | |
4600 | if (parent) | |
4601 | css_put(parent); | |
9d755d33 TH |
4602 | } else { |
4603 | /* cgroup free path */ | |
4604 | atomic_dec(&cgrp->root->nr_cgrps); | |
d62beb7f | 4605 | cgroup1_pidlist_destroy_all(cgrp); |
971ff493 | 4606 | cancel_work_sync(&cgrp->release_agent_work); |
9d755d33 | 4607 | |
d51f39b0 | 4608 | if (cgroup_parent(cgrp)) { |
9d755d33 TH |
4609 | /* |
4610 | * We get a ref to the parent, and put the ref when | |
4611 | * this cgroup is being freed, so it's guaranteed | |
4612 | * that the parent won't be destroyed before its | |
4613 | * children. | |
4614 | */ | |
d51f39b0 | 4615 | cgroup_put(cgroup_parent(cgrp)); |
9d755d33 | 4616 | kernfs_put(cgrp->kn); |
041cd640 TH |
4617 | if (cgroup_on_dfl(cgrp)) |
4618 | cgroup_stat_exit(cgrp); | |
9d755d33 TH |
4619 | kfree(cgrp); |
4620 | } else { | |
4621 | /* | |
4622 | * This is root cgroup's refcnt reaching zero, | |
4623 | * which indicates that the root should be | |
4624 | * released. | |
4625 | */ | |
4626 | cgroup_destroy_root(cgrp->root); | |
4627 | } | |
4628 | } | |
48ddbe19 TH |
4629 | } |
4630 | ||
0c21ead1 | 4631 | static void css_free_rcu_fn(struct rcu_head *rcu_head) |
d3daf28d TH |
4632 | { |
4633 | struct cgroup_subsys_state *css = | |
0c21ead1 | 4634 | container_of(rcu_head, struct cgroup_subsys_state, rcu_head); |
d3daf28d | 4635 | |
35ef10da | 4636 | INIT_WORK(&css->destroy_work, css_free_work_fn); |
e5fca243 | 4637 | queue_work(cgroup_destroy_wq, &css->destroy_work); |
48ddbe19 TH |
4638 | } |
4639 | ||
25e15d83 | 4640 | static void css_release_work_fn(struct work_struct *work) |
d3daf28d TH |
4641 | { |
4642 | struct cgroup_subsys_state *css = | |
25e15d83 | 4643 | container_of(work, struct cgroup_subsys_state, destroy_work); |
15a4c835 | 4644 | struct cgroup_subsys *ss = css->ss; |
9d755d33 | 4645 | struct cgroup *cgrp = css->cgroup; |
15a4c835 | 4646 | |
1fed1b2e TH |
4647 | mutex_lock(&cgroup_mutex); |
4648 | ||
de3f0341 | 4649 | css->flags |= CSS_RELEASED; |
1fed1b2e TH |
4650 | list_del_rcu(&css->sibling); |
4651 | ||
9d755d33 TH |
4652 | if (ss) { |
4653 | /* css release path */ | |
01e58659 | 4654 | cgroup_idr_replace(&ss->css_idr, NULL, css->id); |
7d172cc8 TH |
4655 | if (ss->css_released) |
4656 | ss->css_released(css); | |
9d755d33 | 4657 | } else { |
0679dee0 RG |
4658 | struct cgroup *tcgrp; |
4659 | ||
9d755d33 | 4660 | /* cgroup release path */ |
ed1777de TH |
4661 | trace_cgroup_release(cgrp); |
4662 | ||
041cd640 TH |
4663 | if (cgroup_on_dfl(cgrp)) |
4664 | cgroup_stat_flush(cgrp); | |
4665 | ||
8d747b13 | 4666 | spin_lock_irq(&css_set_lock); |
0679dee0 RG |
4667 | for (tcgrp = cgroup_parent(cgrp); tcgrp; |
4668 | tcgrp = cgroup_parent(tcgrp)) | |
4669 | tcgrp->nr_dying_descendants--; | |
8d747b13 | 4670 | spin_unlock_irq(&css_set_lock); |
0679dee0 | 4671 | |
9d755d33 TH |
4672 | cgroup_idr_remove(&cgrp->root->cgroup_idr, cgrp->id); |
4673 | cgrp->id = -1; | |
a4189487 LZ |
4674 | |
4675 | /* | |
4676 | * There are two control paths which try to determine | |
4677 | * cgroup from dentry without going through kernfs - | |
4678 | * cgroupstats_build() and css_tryget_online_from_dir(). | |
4679 | * Those are supported by RCU protecting clearing of | |
4680 | * cgrp->kn->priv backpointer. | |
4681 | */ | |
6cd0f5bb TH |
4682 | if (cgrp->kn) |
4683 | RCU_INIT_POINTER(*(void __rcu __force **)&cgrp->kn->priv, | |
4684 | NULL); | |
30070984 DM |
4685 | |
4686 | cgroup_bpf_put(cgrp); | |
9d755d33 | 4687 | } |
d3daf28d | 4688 | |
1fed1b2e TH |
4689 | mutex_unlock(&cgroup_mutex); |
4690 | ||
0c21ead1 | 4691 | call_rcu(&css->rcu_head, css_free_rcu_fn); |
d3daf28d TH |
4692 | } |
4693 | ||
d3daf28d TH |
4694 | static void css_release(struct percpu_ref *ref) |
4695 | { | |
4696 | struct cgroup_subsys_state *css = | |
4697 | container_of(ref, struct cgroup_subsys_state, refcnt); | |
4698 | ||
25e15d83 TH |
4699 | INIT_WORK(&css->destroy_work, css_release_work_fn); |
4700 | queue_work(cgroup_destroy_wq, &css->destroy_work); | |
d3daf28d TH |
4701 | } |
4702 | ||
ddfcadab TH |
4703 | static void init_and_link_css(struct cgroup_subsys_state *css, |
4704 | struct cgroup_subsys *ss, struct cgroup *cgrp) | |
ddbcc7e8 | 4705 | { |
0cb51d71 TH |
4706 | lockdep_assert_held(&cgroup_mutex); |
4707 | ||
a590b90d | 4708 | cgroup_get_live(cgrp); |
ddfcadab | 4709 | |
d5c419b6 | 4710 | memset(css, 0, sizeof(*css)); |
bd89aabc | 4711 | css->cgroup = cgrp; |
72c97e54 | 4712 | css->ss = ss; |
8fa3b8d6 | 4713 | css->id = -1; |
d5c419b6 TH |
4714 | INIT_LIST_HEAD(&css->sibling); |
4715 | INIT_LIST_HEAD(&css->children); | |
0cb51d71 | 4716 | css->serial_nr = css_serial_nr_next++; |
aa226ff4 | 4717 | atomic_set(&css->online_cnt, 0); |
0ae78e0b | 4718 | |
d51f39b0 TH |
4719 | if (cgroup_parent(cgrp)) { |
4720 | css->parent = cgroup_css(cgroup_parent(cgrp), ss); | |
ddfcadab | 4721 | css_get(css->parent); |
ddfcadab | 4722 | } |
48ddbe19 | 4723 | |
ca8bdcaf | 4724 | BUG_ON(cgroup_css(cgrp, ss)); |
ddbcc7e8 PM |
4725 | } |
4726 | ||
2a4ac633 | 4727 | /* invoke ->css_online() on a new CSS and mark it online if successful */ |
623f926b | 4728 | static int online_css(struct cgroup_subsys_state *css) |
a31f2d3f | 4729 | { |
623f926b | 4730 | struct cgroup_subsys *ss = css->ss; |
b1929db4 TH |
4731 | int ret = 0; |
4732 | ||
a31f2d3f TH |
4733 | lockdep_assert_held(&cgroup_mutex); |
4734 | ||
92fb9748 | 4735 | if (ss->css_online) |
eb95419b | 4736 | ret = ss->css_online(css); |
ae7f164a | 4737 | if (!ret) { |
eb95419b | 4738 | css->flags |= CSS_ONLINE; |
aec25020 | 4739 | rcu_assign_pointer(css->cgroup->subsys[ss->id], css); |
aa226ff4 TH |
4740 | |
4741 | atomic_inc(&css->online_cnt); | |
4742 | if (css->parent) | |
4743 | atomic_inc(&css->parent->online_cnt); | |
ae7f164a | 4744 | } |
b1929db4 | 4745 | return ret; |
a31f2d3f TH |
4746 | } |
4747 | ||
2a4ac633 | 4748 | /* if the CSS is online, invoke ->css_offline() on it and mark it offline */ |
623f926b | 4749 | static void offline_css(struct cgroup_subsys_state *css) |
a31f2d3f | 4750 | { |
623f926b | 4751 | struct cgroup_subsys *ss = css->ss; |
a31f2d3f TH |
4752 | |
4753 | lockdep_assert_held(&cgroup_mutex); | |
4754 | ||
4755 | if (!(css->flags & CSS_ONLINE)) | |
4756 | return; | |
4757 | ||
d7eeac19 | 4758 | if (ss->css_offline) |
eb95419b | 4759 | ss->css_offline(css); |
a31f2d3f | 4760 | |
eb95419b | 4761 | css->flags &= ~CSS_ONLINE; |
e3297803 | 4762 | RCU_INIT_POINTER(css->cgroup->subsys[ss->id], NULL); |
f8f22e53 TH |
4763 | |
4764 | wake_up_all(&css->cgroup->offline_waitq); | |
a31f2d3f TH |
4765 | } |
4766 | ||
c81c925a | 4767 | /** |
6cd0f5bb | 4768 | * css_create - create a cgroup_subsys_state |
c81c925a TH |
4769 | * @cgrp: the cgroup new css will be associated with |
4770 | * @ss: the subsys of new css | |
4771 | * | |
4772 | * Create a new css associated with @cgrp - @ss pair. On success, the new | |
6cd0f5bb TH |
4773 | * css is online and installed in @cgrp. This function doesn't create the |
4774 | * interface files. Returns 0 on success, -errno on failure. | |
c81c925a | 4775 | */ |
6cd0f5bb TH |
4776 | static struct cgroup_subsys_state *css_create(struct cgroup *cgrp, |
4777 | struct cgroup_subsys *ss) | |
c81c925a | 4778 | { |
d51f39b0 | 4779 | struct cgroup *parent = cgroup_parent(cgrp); |
1fed1b2e | 4780 | struct cgroup_subsys_state *parent_css = cgroup_css(parent, ss); |
c81c925a TH |
4781 | struct cgroup_subsys_state *css; |
4782 | int err; | |
4783 | ||
c81c925a TH |
4784 | lockdep_assert_held(&cgroup_mutex); |
4785 | ||
1fed1b2e | 4786 | css = ss->css_alloc(parent_css); |
e7e15b87 TH |
4787 | if (!css) |
4788 | css = ERR_PTR(-ENOMEM); | |
c81c925a | 4789 | if (IS_ERR(css)) |
6cd0f5bb | 4790 | return css; |
c81c925a | 4791 | |
ddfcadab | 4792 | init_and_link_css(css, ss, cgrp); |
a2bed820 | 4793 | |
2aad2a86 | 4794 | err = percpu_ref_init(&css->refcnt, css_release, 0, GFP_KERNEL); |
c81c925a | 4795 | if (err) |
3eb59ec6 | 4796 | goto err_free_css; |
c81c925a | 4797 | |
cf780b7d | 4798 | err = cgroup_idr_alloc(&ss->css_idr, NULL, 2, 0, GFP_KERNEL); |
15a4c835 | 4799 | if (err < 0) |
b00c52da | 4800 | goto err_free_css; |
15a4c835 | 4801 | css->id = err; |
c81c925a | 4802 | |
15a4c835 | 4803 | /* @css is ready to be brought online now, make it visible */ |
1fed1b2e | 4804 | list_add_tail_rcu(&css->sibling, &parent_css->children); |
15a4c835 | 4805 | cgroup_idr_replace(&ss->css_idr, css, css->id); |
c81c925a TH |
4806 | |
4807 | err = online_css(css); | |
4808 | if (err) | |
1fed1b2e | 4809 | goto err_list_del; |
94419627 | 4810 | |
c81c925a | 4811 | if (ss->broken_hierarchy && !ss->warned_broken_hierarchy && |
d51f39b0 | 4812 | cgroup_parent(parent)) { |
ed3d261b | 4813 | 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 | 4814 | current->comm, current->pid, ss->name); |
c81c925a | 4815 | if (!strcmp(ss->name, "memory")) |
ed3d261b | 4816 | pr_warn("\"memory\" requires setting use_hierarchy to 1 on the root\n"); |
c81c925a TH |
4817 | ss->warned_broken_hierarchy = true; |
4818 | } | |
4819 | ||
6cd0f5bb | 4820 | return css; |
c81c925a | 4821 | |
1fed1b2e TH |
4822 | err_list_del: |
4823 | list_del_rcu(&css->sibling); | |
3eb59ec6 | 4824 | err_free_css: |
a2bed820 | 4825 | call_rcu(&css->rcu_head, css_free_rcu_fn); |
6cd0f5bb | 4826 | return ERR_PTR(err); |
c81c925a TH |
4827 | } |
4828 | ||
07cd1294 TH |
4829 | /* |
4830 | * The returned cgroup is fully initialized including its control mask, but | |
4831 | * it isn't associated with its kernfs_node and doesn't have the control | |
4832 | * mask applied. | |
4833 | */ | |
a5bca215 | 4834 | static struct cgroup *cgroup_create(struct cgroup *parent) |
ddbcc7e8 | 4835 | { |
a5bca215 | 4836 | struct cgroup_root *root = parent->root; |
a5bca215 TH |
4837 | struct cgroup *cgrp, *tcgrp; |
4838 | int level = parent->level + 1; | |
03970d3c | 4839 | int ret; |
ddbcc7e8 | 4840 | |
0a950f65 | 4841 | /* allocate the cgroup and its ID, 0 is reserved for the root */ |
b11cfb58 TH |
4842 | cgrp = kzalloc(sizeof(*cgrp) + |
4843 | sizeof(cgrp->ancestor_ids[0]) * (level + 1), GFP_KERNEL); | |
a5bca215 TH |
4844 | if (!cgrp) |
4845 | return ERR_PTR(-ENOMEM); | |
0ab02ca8 | 4846 | |
2aad2a86 | 4847 | ret = percpu_ref_init(&cgrp->self.refcnt, css_release, 0, GFP_KERNEL); |
9d755d33 TH |
4848 | if (ret) |
4849 | goto out_free_cgrp; | |
4850 | ||
041cd640 TH |
4851 | if (cgroup_on_dfl(parent)) { |
4852 | ret = cgroup_stat_init(cgrp); | |
4853 | if (ret) | |
4854 | goto out_cancel_ref; | |
4855 | } | |
4856 | ||
0ab02ca8 LZ |
4857 | /* |
4858 | * Temporarily set the pointer to NULL, so idr_find() won't return | |
4859 | * a half-baked cgroup. | |
4860 | */ | |
cf780b7d | 4861 | cgrp->id = cgroup_idr_alloc(&root->cgroup_idr, NULL, 2, 0, GFP_KERNEL); |
0ab02ca8 | 4862 | if (cgrp->id < 0) { |
ba0f4d76 | 4863 | ret = -ENOMEM; |
041cd640 | 4864 | goto out_stat_exit; |
976c06bc TH |
4865 | } |
4866 | ||
cc31edce | 4867 | init_cgroup_housekeeping(cgrp); |
ddbcc7e8 | 4868 | |
9d800df1 | 4869 | cgrp->self.parent = &parent->self; |
ba0f4d76 | 4870 | cgrp->root = root; |
b11cfb58 | 4871 | cgrp->level = level; |
324bda9e AS |
4872 | ret = cgroup_bpf_inherit(cgrp); |
4873 | if (ret) | |
4874 | goto out_idr_free; | |
b11cfb58 | 4875 | |
8d747b13 | 4876 | spin_lock_irq(&css_set_lock); |
0679dee0 | 4877 | for (tcgrp = cgrp; tcgrp; tcgrp = cgroup_parent(tcgrp)) { |
b11cfb58 | 4878 | cgrp->ancestor_ids[tcgrp->level] = tcgrp->id; |
ddbcc7e8 | 4879 | |
0679dee0 RG |
4880 | if (tcgrp != cgrp) |
4881 | tcgrp->nr_descendants++; | |
4882 | } | |
8d747b13 | 4883 | spin_unlock_irq(&css_set_lock); |
0679dee0 | 4884 | |
b6abdb0e LZ |
4885 | if (notify_on_release(parent)) |
4886 | set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); | |
4887 | ||
2260e7fc TH |
4888 | if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &parent->flags)) |
4889 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags); | |
97978e6d | 4890 | |
0cb51d71 | 4891 | cgrp->self.serial_nr = css_serial_nr_next++; |
53fa5261 | 4892 | |
4e139afc | 4893 | /* allocation complete, commit to creation */ |
d5c419b6 | 4894 | list_add_tail_rcu(&cgrp->self.sibling, &cgroup_parent(cgrp)->self.children); |
3c9c825b | 4895 | atomic_inc(&root->nr_cgrps); |
a590b90d | 4896 | cgroup_get_live(parent); |
415cf07a | 4897 | |
0d80255e TH |
4898 | /* |
4899 | * @cgrp is now fully operational. If something fails after this | |
4900 | * point, it'll be released via the normal destruction path. | |
4901 | */ | |
6fa4918d | 4902 | cgroup_idr_replace(&root->cgroup_idr, cgrp, cgrp->id); |
4e96ee8e | 4903 | |
bd53d617 TH |
4904 | /* |
4905 | * On the default hierarchy, a child doesn't automatically inherit | |
667c2491 | 4906 | * subtree_control from the parent. Each is configured manually. |
bd53d617 | 4907 | */ |
03970d3c | 4908 | if (!cgroup_on_dfl(cgrp)) |
5531dc91 | 4909 | cgrp->subtree_control = cgroup_control(cgrp); |
03970d3c TH |
4910 | |
4911 | cgroup_propagate_control(cgrp); | |
4912 | ||
a5bca215 TH |
4913 | return cgrp; |
4914 | ||
324bda9e AS |
4915 | out_idr_free: |
4916 | cgroup_idr_remove(&root->cgroup_idr, cgrp->id); | |
041cd640 TH |
4917 | out_stat_exit: |
4918 | if (cgroup_on_dfl(parent)) | |
4919 | cgroup_stat_exit(cgrp); | |
a5bca215 TH |
4920 | out_cancel_ref: |
4921 | percpu_ref_exit(&cgrp->self.refcnt); | |
4922 | out_free_cgrp: | |
4923 | kfree(cgrp); | |
4924 | return ERR_PTR(ret); | |
a5bca215 TH |
4925 | } |
4926 | ||
1a926e0b RG |
4927 | static bool cgroup_check_hierarchy_limits(struct cgroup *parent) |
4928 | { | |
4929 | struct cgroup *cgroup; | |
4930 | int ret = false; | |
4931 | int level = 1; | |
4932 | ||
4933 | lockdep_assert_held(&cgroup_mutex); | |
4934 | ||
4935 | for (cgroup = parent; cgroup; cgroup = cgroup_parent(cgroup)) { | |
4936 | if (cgroup->nr_descendants >= cgroup->max_descendants) | |
4937 | goto fail; | |
4938 | ||
4939 | if (level > cgroup->max_depth) | |
4940 | goto fail; | |
4941 | ||
4942 | level++; | |
4943 | } | |
4944 | ||
4945 | ret = true; | |
4946 | fail: | |
4947 | return ret; | |
4948 | } | |
4949 | ||
1592c9b2 | 4950 | int cgroup_mkdir(struct kernfs_node *parent_kn, const char *name, umode_t mode) |
a5bca215 TH |
4951 | { |
4952 | struct cgroup *parent, *cgrp; | |
a5bca215 | 4953 | struct kernfs_node *kn; |
03970d3c | 4954 | int ret; |
a5bca215 TH |
4955 | |
4956 | /* do not accept '\n' to prevent making /proc/<pid>/cgroup unparsable */ | |
4957 | if (strchr(name, '\n')) | |
4958 | return -EINVAL; | |
4959 | ||
945ba199 | 4960 | parent = cgroup_kn_lock_live(parent_kn, false); |
a5bca215 TH |
4961 | if (!parent) |
4962 | return -ENODEV; | |
4963 | ||
1a926e0b RG |
4964 | if (!cgroup_check_hierarchy_limits(parent)) { |
4965 | ret = -EAGAIN; | |
4966 | goto out_unlock; | |
4967 | } | |
4968 | ||
a5bca215 TH |
4969 | cgrp = cgroup_create(parent); |
4970 | if (IS_ERR(cgrp)) { | |
4971 | ret = PTR_ERR(cgrp); | |
4972 | goto out_unlock; | |
4973 | } | |
4974 | ||
195e9b6c TH |
4975 | /* create the directory */ |
4976 | kn = kernfs_create_dir(parent->kn, name, mode, cgrp); | |
4977 | if (IS_ERR(kn)) { | |
4978 | ret = PTR_ERR(kn); | |
4979 | goto out_destroy; | |
4980 | } | |
4981 | cgrp->kn = kn; | |
4982 | ||
4983 | /* | |
4984 | * This extra ref will be put in cgroup_free_fn() and guarantees | |
4985 | * that @cgrp->kn is always accessible. | |
4986 | */ | |
4987 | kernfs_get(kn); | |
4988 | ||
4989 | ret = cgroup_kn_set_ugid(kn); | |
4990 | if (ret) | |
4991 | goto out_destroy; | |
4992 | ||
334c3679 | 4993 | ret = css_populate_dir(&cgrp->self); |
195e9b6c TH |
4994 | if (ret) |
4995 | goto out_destroy; | |
4996 | ||
03970d3c TH |
4997 | ret = cgroup_apply_control_enable(cgrp); |
4998 | if (ret) | |
4999 | goto out_destroy; | |
195e9b6c | 5000 | |
ed1777de TH |
5001 | trace_cgroup_mkdir(cgrp); |
5002 | ||
195e9b6c | 5003 | /* let's create and online css's */ |
2bd59d48 | 5004 | kernfs_activate(kn); |
ddbcc7e8 | 5005 | |
ba0f4d76 TH |
5006 | ret = 0; |
5007 | goto out_unlock; | |
ddbcc7e8 | 5008 | |
a5bca215 TH |
5009 | out_destroy: |
5010 | cgroup_destroy_locked(cgrp); | |
ba0f4d76 | 5011 | out_unlock: |
a9746d8d | 5012 | cgroup_kn_unlock(parent_kn); |
ba0f4d76 | 5013 | return ret; |
ddbcc7e8 PM |
5014 | } |
5015 | ||
223dbc38 TH |
5016 | /* |
5017 | * This is called when the refcnt of a css is confirmed to be killed. | |
249f3468 TH |
5018 | * css_tryget_online() is now guaranteed to fail. Tell the subsystem to |
5019 | * initate destruction and put the css ref from kill_css(). | |
223dbc38 TH |
5020 | */ |
5021 | static void css_killed_work_fn(struct work_struct *work) | |
d3daf28d | 5022 | { |
223dbc38 TH |
5023 | struct cgroup_subsys_state *css = |
5024 | container_of(work, struct cgroup_subsys_state, destroy_work); | |
d3daf28d | 5025 | |
f20104de | 5026 | mutex_lock(&cgroup_mutex); |
09a503ea | 5027 | |
aa226ff4 TH |
5028 | do { |
5029 | offline_css(css); | |
5030 | css_put(css); | |
5031 | /* @css can't go away while we're holding cgroup_mutex */ | |
5032 | css = css->parent; | |
5033 | } while (css && atomic_dec_and_test(&css->online_cnt)); | |
5034 | ||
5035 | mutex_unlock(&cgroup_mutex); | |
d3daf28d TH |
5036 | } |
5037 | ||
223dbc38 TH |
5038 | /* css kill confirmation processing requires process context, bounce */ |
5039 | static void css_killed_ref_fn(struct percpu_ref *ref) | |
d3daf28d TH |
5040 | { |
5041 | struct cgroup_subsys_state *css = | |
5042 | container_of(ref, struct cgroup_subsys_state, refcnt); | |
5043 | ||
aa226ff4 TH |
5044 | if (atomic_dec_and_test(&css->online_cnt)) { |
5045 | INIT_WORK(&css->destroy_work, css_killed_work_fn); | |
5046 | queue_work(cgroup_destroy_wq, &css->destroy_work); | |
5047 | } | |
d3daf28d TH |
5048 | } |
5049 | ||
f392e51c TH |
5050 | /** |
5051 | * kill_css - destroy a css | |
5052 | * @css: css to destroy | |
5053 | * | |
5054 | * This function initiates destruction of @css by removing cgroup interface | |
5055 | * files and putting its base reference. ->css_offline() will be invoked | |
ec903c0c TH |
5056 | * asynchronously once css_tryget_online() is guaranteed to fail and when |
5057 | * the reference count reaches zero, @css will be released. | |
f392e51c TH |
5058 | */ |
5059 | static void kill_css(struct cgroup_subsys_state *css) | |
edae0c33 | 5060 | { |
01f6474c | 5061 | lockdep_assert_held(&cgroup_mutex); |
94419627 | 5062 | |
33c35aa4 WL |
5063 | if (css->flags & CSS_DYING) |
5064 | return; | |
5065 | ||
5066 | css->flags |= CSS_DYING; | |
5067 | ||
2bd59d48 TH |
5068 | /* |
5069 | * This must happen before css is disassociated with its cgroup. | |
5070 | * See seq_css() for details. | |
5071 | */ | |
334c3679 | 5072 | css_clear_dir(css); |
3c14f8b4 | 5073 | |
edae0c33 TH |
5074 | /* |
5075 | * Killing would put the base ref, but we need to keep it alive | |
5076 | * until after ->css_offline(). | |
5077 | */ | |
5078 | css_get(css); | |
5079 | ||
5080 | /* | |
5081 | * cgroup core guarantees that, by the time ->css_offline() is | |
5082 | * invoked, no new css reference will be given out via | |
ec903c0c | 5083 | * css_tryget_online(). We can't simply call percpu_ref_kill() and |
edae0c33 TH |
5084 | * proceed to offlining css's because percpu_ref_kill() doesn't |
5085 | * guarantee that the ref is seen as killed on all CPUs on return. | |
5086 | * | |
5087 | * Use percpu_ref_kill_and_confirm() to get notifications as each | |
5088 | * css is confirmed to be seen as killed on all CPUs. | |
5089 | */ | |
5090 | percpu_ref_kill_and_confirm(&css->refcnt, css_killed_ref_fn); | |
d3daf28d TH |
5091 | } |
5092 | ||
5093 | /** | |
5094 | * cgroup_destroy_locked - the first stage of cgroup destruction | |
5095 | * @cgrp: cgroup to be destroyed | |
5096 | * | |
5097 | * css's make use of percpu refcnts whose killing latency shouldn't be | |
5098 | * exposed to userland and are RCU protected. Also, cgroup core needs to | |
ec903c0c TH |
5099 | * guarantee that css_tryget_online() won't succeed by the time |
5100 | * ->css_offline() is invoked. To satisfy all the requirements, | |
5101 | * destruction is implemented in the following two steps. | |
d3daf28d TH |
5102 | * |
5103 | * s1. Verify @cgrp can be destroyed and mark it dying. Remove all | |
5104 | * userland visible parts and start killing the percpu refcnts of | |
5105 | * css's. Set up so that the next stage will be kicked off once all | |
5106 | * the percpu refcnts are confirmed to be killed. | |
5107 | * | |
5108 | * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the | |
5109 | * rest of destruction. Once all cgroup references are gone, the | |
5110 | * cgroup is RCU-freed. | |
5111 | * | |
5112 | * This function implements s1. After this step, @cgrp is gone as far as | |
5113 | * the userland is concerned and a new cgroup with the same name may be | |
5114 | * created. As cgroup doesn't care about the names internally, this | |
5115 | * doesn't cause any problem. | |
5116 | */ | |
42809dd4 TH |
5117 | static int cgroup_destroy_locked(struct cgroup *cgrp) |
5118 | __releases(&cgroup_mutex) __acquires(&cgroup_mutex) | |
ddbcc7e8 | 5119 | { |
0679dee0 | 5120 | struct cgroup *tcgrp, *parent = cgroup_parent(cgrp); |
2bd59d48 | 5121 | struct cgroup_subsys_state *css; |
2b021cbf | 5122 | struct cgrp_cset_link *link; |
1c6727af | 5123 | int ssid; |
ddbcc7e8 | 5124 | |
42809dd4 TH |
5125 | lockdep_assert_held(&cgroup_mutex); |
5126 | ||
91486f61 TH |
5127 | /* |
5128 | * Only migration can raise populated from zero and we're already | |
5129 | * holding cgroup_mutex. | |
5130 | */ | |
5131 | if (cgroup_is_populated(cgrp)) | |
ddbcc7e8 | 5132 | return -EBUSY; |
a043e3b2 | 5133 | |
bb78a92f | 5134 | /* |
d5c419b6 TH |
5135 | * Make sure there's no live children. We can't test emptiness of |
5136 | * ->self.children as dead children linger on it while being | |
5137 | * drained; otherwise, "rmdir parent/child parent" may fail. | |
bb78a92f | 5138 | */ |
f3d46500 | 5139 | if (css_has_online_children(&cgrp->self)) |
bb78a92f HD |
5140 | return -EBUSY; |
5141 | ||
455050d2 | 5142 | /* |
2b021cbf TH |
5143 | * Mark @cgrp and the associated csets dead. The former prevents |
5144 | * further task migration and child creation by disabling | |
5145 | * cgroup_lock_live_group(). The latter makes the csets ignored by | |
5146 | * the migration path. | |
455050d2 | 5147 | */ |
184faf32 | 5148 | cgrp->self.flags &= ~CSS_ONLINE; |
ddbcc7e8 | 5149 | |
82d6489d | 5150 | spin_lock_irq(&css_set_lock); |
2b021cbf TH |
5151 | list_for_each_entry(link, &cgrp->cset_links, cset_link) |
5152 | link->cset->dead = true; | |
82d6489d | 5153 | spin_unlock_irq(&css_set_lock); |
2b021cbf | 5154 | |
249f3468 | 5155 | /* initiate massacre of all css's */ |
1c6727af TH |
5156 | for_each_css(css, ssid, cgrp) |
5157 | kill_css(css); | |
455050d2 | 5158 | |
455050d2 | 5159 | /* |
01f6474c TH |
5160 | * Remove @cgrp directory along with the base files. @cgrp has an |
5161 | * extra ref on its kn. | |
f20104de | 5162 | */ |
01f6474c | 5163 | kernfs_remove(cgrp->kn); |
f20104de | 5164 | |
454000ad TH |
5165 | if (parent && cgroup_is_threaded(cgrp)) |
5166 | parent->nr_threaded_children--; | |
5167 | ||
8d747b13 | 5168 | spin_lock_irq(&css_set_lock); |
0679dee0 RG |
5169 | for (tcgrp = cgroup_parent(cgrp); tcgrp; tcgrp = cgroup_parent(tcgrp)) { |
5170 | tcgrp->nr_descendants--; | |
5171 | tcgrp->nr_dying_descendants++; | |
5172 | } | |
8d747b13 | 5173 | spin_unlock_irq(&css_set_lock); |
0679dee0 | 5174 | |
5a621e6c | 5175 | cgroup1_check_for_release(parent); |
2bd59d48 | 5176 | |
249f3468 | 5177 | /* put the base reference */ |
9d755d33 | 5178 | percpu_ref_kill(&cgrp->self.refcnt); |
455050d2 | 5179 | |
ea15f8cc TH |
5180 | return 0; |
5181 | }; | |
5182 | ||
1592c9b2 | 5183 | int cgroup_rmdir(struct kernfs_node *kn) |
42809dd4 | 5184 | { |
a9746d8d | 5185 | struct cgroup *cgrp; |
2bd59d48 | 5186 | int ret = 0; |
42809dd4 | 5187 | |
945ba199 | 5188 | cgrp = cgroup_kn_lock_live(kn, false); |
a9746d8d TH |
5189 | if (!cgrp) |
5190 | return 0; | |
42809dd4 | 5191 | |
a9746d8d | 5192 | ret = cgroup_destroy_locked(cgrp); |
2bb566cb | 5193 | |
ed1777de TH |
5194 | if (!ret) |
5195 | trace_cgroup_rmdir(cgrp); | |
5196 | ||
a9746d8d | 5197 | cgroup_kn_unlock(kn); |
42809dd4 | 5198 | return ret; |
8e3f6541 TH |
5199 | } |
5200 | ||
2bd59d48 | 5201 | static struct kernfs_syscall_ops cgroup_kf_syscall_ops = { |
5136f636 | 5202 | .show_options = cgroup_show_options, |
2bd59d48 | 5203 | .remount_fs = cgroup_remount, |
2bd59d48 TH |
5204 | .mkdir = cgroup_mkdir, |
5205 | .rmdir = cgroup_rmdir, | |
4f41fc59 | 5206 | .show_path = cgroup_show_path, |
2bd59d48 TH |
5207 | }; |
5208 | ||
15a4c835 | 5209 | static void __init cgroup_init_subsys(struct cgroup_subsys *ss, bool early) |
ddbcc7e8 | 5210 | { |
ddbcc7e8 | 5211 | struct cgroup_subsys_state *css; |
cfe36bde | 5212 | |
a5ae9899 | 5213 | pr_debug("Initializing cgroup subsys %s\n", ss->name); |
ddbcc7e8 | 5214 | |
648bb56d TH |
5215 | mutex_lock(&cgroup_mutex); |
5216 | ||
15a4c835 | 5217 | idr_init(&ss->css_idr); |
0adb0704 | 5218 | INIT_LIST_HEAD(&ss->cfts); |
8e3f6541 | 5219 | |
3dd06ffa TH |
5220 | /* Create the root cgroup state for this subsystem */ |
5221 | ss->root = &cgrp_dfl_root; | |
5222 | css = ss->css_alloc(cgroup_css(&cgrp_dfl_root.cgrp, ss)); | |
ddbcc7e8 PM |
5223 | /* We don't handle early failures gracefully */ |
5224 | BUG_ON(IS_ERR(css)); | |
ddfcadab | 5225 | init_and_link_css(css, ss, &cgrp_dfl_root.cgrp); |
3b514d24 TH |
5226 | |
5227 | /* | |
5228 | * Root csses are never destroyed and we can't initialize | |
5229 | * percpu_ref during early init. Disable refcnting. | |
5230 | */ | |
5231 | css->flags |= CSS_NO_REF; | |
5232 | ||
15a4c835 | 5233 | if (early) { |
9395a450 | 5234 | /* allocation can't be done safely during early init */ |
15a4c835 TH |
5235 | css->id = 1; |
5236 | } else { | |
5237 | css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2, GFP_KERNEL); | |
5238 | BUG_ON(css->id < 0); | |
5239 | } | |
ddbcc7e8 | 5240 | |
e8d55fde | 5241 | /* Update the init_css_set to contain a subsys |
817929ec | 5242 | * pointer to this state - since the subsystem is |
e8d55fde | 5243 | * newly registered, all tasks and hence the |
3dd06ffa | 5244 | * init_css_set is in the subsystem's root cgroup. */ |
aec25020 | 5245 | init_css_set.subsys[ss->id] = css; |
ddbcc7e8 | 5246 | |
cb4a3167 AS |
5247 | have_fork_callback |= (bool)ss->fork << ss->id; |
5248 | have_exit_callback |= (bool)ss->exit << ss->id; | |
d7248a93 | 5249 | have_release_callback |= (bool)ss->release << ss->id; |
7e47682e | 5250 | have_canfork_callback |= (bool)ss->can_fork << ss->id; |
ddbcc7e8 | 5251 | |
e8d55fde LZ |
5252 | /* At system boot, before all subsystems have been |
5253 | * registered, no tasks have been forked, so we don't | |
5254 | * need to invoke fork callbacks here. */ | |
5255 | BUG_ON(!list_empty(&init_task.tasks)); | |
5256 | ||
ae7f164a | 5257 | BUG_ON(online_css(css)); |
a8638030 | 5258 | |
cf5d5941 BB |
5259 | mutex_unlock(&cgroup_mutex); |
5260 | } | |
cf5d5941 | 5261 | |
ddbcc7e8 | 5262 | /** |
a043e3b2 LZ |
5263 | * cgroup_init_early - cgroup initialization at system boot |
5264 | * | |
5265 | * Initialize cgroups at system boot, and initialize any | |
5266 | * subsystems that request early init. | |
ddbcc7e8 PM |
5267 | */ |
5268 | int __init cgroup_init_early(void) | |
5269 | { | |
7b9a6ba5 | 5270 | static struct cgroup_sb_opts __initdata opts; |
30159ec7 | 5271 | struct cgroup_subsys *ss; |
ddbcc7e8 | 5272 | int i; |
30159ec7 | 5273 | |
3dd06ffa | 5274 | init_cgroup_root(&cgrp_dfl_root, &opts); |
3b514d24 TH |
5275 | cgrp_dfl_root.cgrp.self.flags |= CSS_NO_REF; |
5276 | ||
a4ea1cc9 | 5277 | RCU_INIT_POINTER(init_task.cgroups, &init_css_set); |
817929ec | 5278 | |
3ed80a62 | 5279 | for_each_subsys(ss, i) { |
aec25020 | 5280 | WARN(!ss->css_alloc || !ss->css_free || ss->name || ss->id, |
63253ad8 | 5281 | "invalid cgroup_subsys %d:%s css_alloc=%p css_free=%p id:name=%d:%s\n", |
073219e9 | 5282 | i, cgroup_subsys_name[i], ss->css_alloc, ss->css_free, |
aec25020 | 5283 | ss->id, ss->name); |
073219e9 TH |
5284 | WARN(strlen(cgroup_subsys_name[i]) > MAX_CGROUP_TYPE_NAMELEN, |
5285 | "cgroup_subsys_name %s too long\n", cgroup_subsys_name[i]); | |
5286 | ||
aec25020 | 5287 | ss->id = i; |
073219e9 | 5288 | ss->name = cgroup_subsys_name[i]; |
3e1d2eed TH |
5289 | if (!ss->legacy_name) |
5290 | ss->legacy_name = cgroup_subsys_name[i]; | |
ddbcc7e8 PM |
5291 | |
5292 | if (ss->early_init) | |
15a4c835 | 5293 | cgroup_init_subsys(ss, true); |
ddbcc7e8 PM |
5294 | } |
5295 | return 0; | |
5296 | } | |
5297 | ||
6e5c8307 | 5298 | static u16 cgroup_disable_mask __initdata; |
a3e72739 | 5299 | |
ddbcc7e8 | 5300 | /** |
a043e3b2 LZ |
5301 | * cgroup_init - cgroup initialization |
5302 | * | |
5303 | * Register cgroup filesystem and /proc file, and initialize | |
5304 | * any subsystems that didn't request early init. | |
ddbcc7e8 PM |
5305 | */ |
5306 | int __init cgroup_init(void) | |
5307 | { | |
30159ec7 | 5308 | struct cgroup_subsys *ss; |
035f4f51 | 5309 | int ssid; |
ddbcc7e8 | 5310 | |
6e5c8307 | 5311 | BUILD_BUG_ON(CGROUP_SUBSYS_COUNT > 16); |
1ed13287 | 5312 | BUG_ON(percpu_init_rwsem(&cgroup_threadgroup_rwsem)); |
d62beb7f TH |
5313 | BUG_ON(cgroup_init_cftypes(NULL, cgroup_base_files)); |
5314 | BUG_ON(cgroup_init_cftypes(NULL, cgroup1_base_files)); | |
ddbcc7e8 | 5315 | |
041cd640 TH |
5316 | cgroup_stat_boot(); |
5317 | ||
3942a9bd PZ |
5318 | /* |
5319 | * The latency of the synchronize_sched() is too high for cgroups, | |
5320 | * avoid it at the cost of forcing all readers into the slow path. | |
5321 | */ | |
5322 | rcu_sync_enter_start(&cgroup_threadgroup_rwsem.rss); | |
5323 | ||
a79a908f AK |
5324 | get_user_ns(init_cgroup_ns.user_ns); |
5325 | ||
54e7b4eb | 5326 | mutex_lock(&cgroup_mutex); |
54e7b4eb | 5327 | |
2378d8b8 TH |
5328 | /* |
5329 | * Add init_css_set to the hash table so that dfl_root can link to | |
5330 | * it during init. | |
5331 | */ | |
5332 | hash_add(css_set_table, &init_css_set.hlist, | |
5333 | css_set_hash(init_css_set.subsys)); | |
82fe9b0d | 5334 | |
9732adc5 | 5335 | BUG_ON(cgroup_setup_root(&cgrp_dfl_root, 0, 0)); |
4e96ee8e | 5336 | |
54e7b4eb TH |
5337 | mutex_unlock(&cgroup_mutex); |
5338 | ||
172a2c06 | 5339 | for_each_subsys(ss, ssid) { |
15a4c835 TH |
5340 | if (ss->early_init) { |
5341 | struct cgroup_subsys_state *css = | |
5342 | init_css_set.subsys[ss->id]; | |
5343 | ||
5344 | css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2, | |
5345 | GFP_KERNEL); | |
5346 | BUG_ON(css->id < 0); | |
5347 | } else { | |
5348 | cgroup_init_subsys(ss, false); | |
5349 | } | |
172a2c06 | 5350 | |
2d8f243a TH |
5351 | list_add_tail(&init_css_set.e_cset_node[ssid], |
5352 | &cgrp_dfl_root.cgrp.e_csets[ssid]); | |
172a2c06 TH |
5353 | |
5354 | /* | |
c731ae1d LZ |
5355 | * Setting dfl_root subsys_mask needs to consider the |
5356 | * disabled flag and cftype registration needs kmalloc, | |
5357 | * both of which aren't available during early_init. | |
172a2c06 | 5358 | */ |
a3e72739 TH |
5359 | if (cgroup_disable_mask & (1 << ssid)) { |
5360 | static_branch_disable(cgroup_subsys_enabled_key[ssid]); | |
5361 | printk(KERN_INFO "Disabling %s control group subsystem\n", | |
5362 | ss->name); | |
a8ddc821 | 5363 | continue; |
a3e72739 | 5364 | } |
a8ddc821 | 5365 | |
d62beb7f | 5366 | if (cgroup1_ssid_disabled(ssid)) |
223ffb29 JW |
5367 | printk(KERN_INFO "Disabling %s control group subsystem in v1 mounts\n", |
5368 | ss->name); | |
5369 | ||
a8ddc821 TH |
5370 | cgrp_dfl_root.subsys_mask |= 1 << ss->id; |
5371 | ||
8cfd8147 TH |
5372 | /* implicit controllers must be threaded too */ |
5373 | WARN_ON(ss->implicit_on_dfl && !ss->threaded); | |
5374 | ||
f6d635ad TH |
5375 | if (ss->implicit_on_dfl) |
5376 | cgrp_dfl_implicit_ss_mask |= 1 << ss->id; | |
5377 | else if (!ss->dfl_cftypes) | |
a7165264 | 5378 | cgrp_dfl_inhibit_ss_mask |= 1 << ss->id; |
5de4fa13 | 5379 | |
8cfd8147 TH |
5380 | if (ss->threaded) |
5381 | cgrp_dfl_threaded_ss_mask |= 1 << ss->id; | |
5382 | ||
a8ddc821 TH |
5383 | if (ss->dfl_cftypes == ss->legacy_cftypes) { |
5384 | WARN_ON(cgroup_add_cftypes(ss, ss->dfl_cftypes)); | |
5385 | } else { | |
5386 | WARN_ON(cgroup_add_dfl_cftypes(ss, ss->dfl_cftypes)); | |
5387 | WARN_ON(cgroup_add_legacy_cftypes(ss, ss->legacy_cftypes)); | |
c731ae1d | 5388 | } |
295458e6 VD |
5389 | |
5390 | if (ss->bind) | |
5391 | ss->bind(init_css_set.subsys[ssid]); | |
7af608e4 TH |
5392 | |
5393 | mutex_lock(&cgroup_mutex); | |
5394 | css_populate_dir(init_css_set.subsys[ssid]); | |
5395 | mutex_unlock(&cgroup_mutex); | |
676db4af GKH |
5396 | } |
5397 | ||
2378d8b8 TH |
5398 | /* init_css_set.subsys[] has been updated, re-hash */ |
5399 | hash_del(&init_css_set.hlist); | |
5400 | hash_add(css_set_table, &init_css_set.hlist, | |
5401 | css_set_hash(init_css_set.subsys)); | |
5402 | ||
035f4f51 TH |
5403 | WARN_ON(sysfs_create_mount_point(fs_kobj, "cgroup")); |
5404 | WARN_ON(register_filesystem(&cgroup_fs_type)); | |
67e9c74b | 5405 | WARN_ON(register_filesystem(&cgroup2_fs_type)); |
035f4f51 | 5406 | WARN_ON(!proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations)); |
ddbcc7e8 | 5407 | |
2bd59d48 | 5408 | return 0; |
ddbcc7e8 | 5409 | } |
b4f48b63 | 5410 | |
e5fca243 TH |
5411 | static int __init cgroup_wq_init(void) |
5412 | { | |
5413 | /* | |
5414 | * There isn't much point in executing destruction path in | |
5415 | * parallel. Good chunk is serialized with cgroup_mutex anyway. | |
1a11533f | 5416 | * Use 1 for @max_active. |
e5fca243 TH |
5417 | * |
5418 | * We would prefer to do this in cgroup_init() above, but that | |
5419 | * is called before init_workqueues(): so leave this until after. | |
5420 | */ | |
1a11533f | 5421 | cgroup_destroy_wq = alloc_workqueue("cgroup_destroy", 0, 1); |
e5fca243 TH |
5422 | BUG_ON(!cgroup_destroy_wq); |
5423 | return 0; | |
5424 | } | |
5425 | core_initcall(cgroup_wq_init); | |
5426 | ||
69fd5c39 SL |
5427 | void cgroup_path_from_kernfs_id(const union kernfs_node_id *id, |
5428 | char *buf, size_t buflen) | |
5429 | { | |
5430 | struct kernfs_node *kn; | |
5431 | ||
5432 | kn = kernfs_get_node_by_id(cgrp_dfl_root.kf_root, id); | |
5433 | if (!kn) | |
5434 | return; | |
5435 | kernfs_path(kn, buf, buflen); | |
5436 | kernfs_put(kn); | |
5437 | } | |
5438 | ||
a424316c PM |
5439 | /* |
5440 | * proc_cgroup_show() | |
5441 | * - Print task's cgroup paths into seq_file, one line for each hierarchy | |
5442 | * - Used for /proc/<pid>/cgroup. | |
a424316c | 5443 | */ |
006f4ac4 ZL |
5444 | int proc_cgroup_show(struct seq_file *m, struct pid_namespace *ns, |
5445 | struct pid *pid, struct task_struct *tsk) | |
a424316c | 5446 | { |
4c737b41 | 5447 | char *buf; |
a424316c | 5448 | int retval; |
3dd06ffa | 5449 | struct cgroup_root *root; |
a424316c PM |
5450 | |
5451 | retval = -ENOMEM; | |
e61734c5 | 5452 | buf = kmalloc(PATH_MAX, GFP_KERNEL); |
a424316c PM |
5453 | if (!buf) |
5454 | goto out; | |
5455 | ||
a424316c | 5456 | mutex_lock(&cgroup_mutex); |
82d6489d | 5457 | spin_lock_irq(&css_set_lock); |
a424316c | 5458 | |
985ed670 | 5459 | for_each_root(root) { |
a424316c | 5460 | struct cgroup_subsys *ss; |
bd89aabc | 5461 | struct cgroup *cgrp; |
b85d2040 | 5462 | int ssid, count = 0; |
a424316c | 5463 | |
a7165264 | 5464 | if (root == &cgrp_dfl_root && !cgrp_dfl_visible) |
985ed670 TH |
5465 | continue; |
5466 | ||
2c6ab6d2 | 5467 | seq_printf(m, "%d:", root->hierarchy_id); |
d98817d4 TH |
5468 | if (root != &cgrp_dfl_root) |
5469 | for_each_subsys(ss, ssid) | |
5470 | if (root->subsys_mask & (1 << ssid)) | |
5471 | seq_printf(m, "%s%s", count++ ? "," : "", | |
3e1d2eed | 5472 | ss->legacy_name); |
c6d57f33 PM |
5473 | if (strlen(root->name)) |
5474 | seq_printf(m, "%sname=%s", count ? "," : "", | |
5475 | root->name); | |
a424316c | 5476 | seq_putc(m, ':'); |
2e91fa7f | 5477 | |
7717f7ba | 5478 | cgrp = task_cgroup_from_root(tsk, root); |
2e91fa7f TH |
5479 | |
5480 | /* | |
5481 | * On traditional hierarchies, all zombie tasks show up as | |
5482 | * belonging to the root cgroup. On the default hierarchy, | |
5483 | * while a zombie doesn't show up in "cgroup.procs" and | |
5484 | * thus can't be migrated, its /proc/PID/cgroup keeps | |
5485 | * reporting the cgroup it belonged to before exiting. If | |
5486 | * the cgroup is removed before the zombie is reaped, | |
5487 | * " (deleted)" is appended to the cgroup path. | |
5488 | */ | |
5489 | if (cgroup_on_dfl(cgrp) || !(tsk->flags & PF_EXITING)) { | |
4c737b41 | 5490 | retval = cgroup_path_ns_locked(cgrp, buf, PATH_MAX, |
a79a908f | 5491 | current->nsproxy->cgroup_ns); |
e0223003 | 5492 | if (retval >= PATH_MAX) |
2e91fa7f | 5493 | retval = -ENAMETOOLONG; |
e0223003 | 5494 | if (retval < 0) |
2e91fa7f | 5495 | goto out_unlock; |
4c737b41 TH |
5496 | |
5497 | seq_puts(m, buf); | |
2e91fa7f | 5498 | } else { |
4c737b41 | 5499 | seq_puts(m, "/"); |
e61734c5 | 5500 | } |
2e91fa7f | 5501 | |
2e91fa7f TH |
5502 | if (cgroup_on_dfl(cgrp) && cgroup_is_dead(cgrp)) |
5503 | seq_puts(m, " (deleted)\n"); | |
5504 | else | |
5505 | seq_putc(m, '\n'); | |
a424316c PM |
5506 | } |
5507 | ||
006f4ac4 | 5508 | retval = 0; |
a424316c | 5509 | out_unlock: |
82d6489d | 5510 | spin_unlock_irq(&css_set_lock); |
a424316c | 5511 | mutex_unlock(&cgroup_mutex); |
a424316c PM |
5512 | kfree(buf); |
5513 | out: | |
5514 | return retval; | |
5515 | } | |
5516 | ||
b4f48b63 | 5517 | /** |
eaf797ab | 5518 | * cgroup_fork - initialize cgroup related fields during copy_process() |
a043e3b2 | 5519 | * @child: pointer to task_struct of forking parent process. |
b4f48b63 | 5520 | * |
eaf797ab TH |
5521 | * A task is associated with the init_css_set until cgroup_post_fork() |
5522 | * attaches it to the parent's css_set. Empty cg_list indicates that | |
5523 | * @child isn't holding reference to its css_set. | |
b4f48b63 PM |
5524 | */ |
5525 | void cgroup_fork(struct task_struct *child) | |
5526 | { | |
eaf797ab | 5527 | RCU_INIT_POINTER(child->cgroups, &init_css_set); |
817929ec | 5528 | INIT_LIST_HEAD(&child->cg_list); |
b4f48b63 PM |
5529 | } |
5530 | ||
7e47682e AS |
5531 | /** |
5532 | * cgroup_can_fork - called on a new task before the process is exposed | |
5533 | * @child: the task in question. | |
5534 | * | |
5535 | * This calls the subsystem can_fork() callbacks. If the can_fork() callback | |
5536 | * returns an error, the fork aborts with that error code. This allows for | |
5537 | * a cgroup subsystem to conditionally allow or deny new forks. | |
5538 | */ | |
b53202e6 | 5539 | int cgroup_can_fork(struct task_struct *child) |
7e47682e AS |
5540 | { |
5541 | struct cgroup_subsys *ss; | |
5542 | int i, j, ret; | |
5543 | ||
b4e0eeaf | 5544 | do_each_subsys_mask(ss, i, have_canfork_callback) { |
b53202e6 | 5545 | ret = ss->can_fork(child); |
7e47682e AS |
5546 | if (ret) |
5547 | goto out_revert; | |
b4e0eeaf | 5548 | } while_each_subsys_mask(); |
7e47682e AS |
5549 | |
5550 | return 0; | |
5551 | ||
5552 | out_revert: | |
5553 | for_each_subsys(ss, j) { | |
5554 | if (j >= i) | |
5555 | break; | |
5556 | if (ss->cancel_fork) | |
b53202e6 | 5557 | ss->cancel_fork(child); |
7e47682e AS |
5558 | } |
5559 | ||
5560 | return ret; | |
5561 | } | |
5562 | ||
5563 | /** | |
5564 | * cgroup_cancel_fork - called if a fork failed after cgroup_can_fork() | |
5565 | * @child: the task in question | |
5566 | * | |
5567 | * This calls the cancel_fork() callbacks if a fork failed *after* | |
5568 | * cgroup_can_fork() succeded. | |
5569 | */ | |
b53202e6 | 5570 | void cgroup_cancel_fork(struct task_struct *child) |
7e47682e AS |
5571 | { |
5572 | struct cgroup_subsys *ss; | |
5573 | int i; | |
5574 | ||
5575 | for_each_subsys(ss, i) | |
5576 | if (ss->cancel_fork) | |
b53202e6 | 5577 | ss->cancel_fork(child); |
7e47682e AS |
5578 | } |
5579 | ||
817929ec | 5580 | /** |
a043e3b2 LZ |
5581 | * cgroup_post_fork - called on a new task after adding it to the task list |
5582 | * @child: the task in question | |
5583 | * | |
5edee61e TH |
5584 | * Adds the task to the list running through its css_set if necessary and |
5585 | * call the subsystem fork() callbacks. Has to be after the task is | |
5586 | * visible on the task list in case we race with the first call to | |
0942eeee | 5587 | * cgroup_task_iter_start() - to guarantee that the new task ends up on its |
5edee61e | 5588 | * list. |
a043e3b2 | 5589 | */ |
b53202e6 | 5590 | void cgroup_post_fork(struct task_struct *child) |
817929ec | 5591 | { |
30159ec7 | 5592 | struct cgroup_subsys *ss; |
5edee61e TH |
5593 | int i; |
5594 | ||
3ce3230a | 5595 | /* |
251f8c03 | 5596 | * This may race against cgroup_enable_task_cg_lists(). As that |
eaf797ab TH |
5597 | * function sets use_task_css_set_links before grabbing |
5598 | * tasklist_lock and we just went through tasklist_lock to add | |
5599 | * @child, it's guaranteed that either we see the set | |
5600 | * use_task_css_set_links or cgroup_enable_task_cg_lists() sees | |
5601 | * @child during its iteration. | |
5602 | * | |
5603 | * If we won the race, @child is associated with %current's | |
f0d9a5f1 | 5604 | * css_set. Grabbing css_set_lock guarantees both that the |
eaf797ab TH |
5605 | * association is stable, and, on completion of the parent's |
5606 | * migration, @child is visible in the source of migration or | |
5607 | * already in the destination cgroup. This guarantee is necessary | |
5608 | * when implementing operations which need to migrate all tasks of | |
5609 | * a cgroup to another. | |
5610 | * | |
251f8c03 | 5611 | * Note that if we lose to cgroup_enable_task_cg_lists(), @child |
eaf797ab TH |
5612 | * will remain in init_css_set. This is safe because all tasks are |
5613 | * in the init_css_set before cg_links is enabled and there's no | |
5614 | * operation which transfers all tasks out of init_css_set. | |
3ce3230a | 5615 | */ |
817929ec | 5616 | if (use_task_css_set_links) { |
eaf797ab TH |
5617 | struct css_set *cset; |
5618 | ||
82d6489d | 5619 | spin_lock_irq(&css_set_lock); |
0e1d768f | 5620 | cset = task_css_set(current); |
eaf797ab | 5621 | if (list_empty(&child->cg_list)) { |
eaf797ab | 5622 | get_css_set(cset); |
73a7242a | 5623 | cset->nr_tasks++; |
f6d7d049 | 5624 | css_set_move_task(child, NULL, cset, false); |
eaf797ab | 5625 | } |
82d6489d | 5626 | spin_unlock_irq(&css_set_lock); |
817929ec | 5627 | } |
5edee61e TH |
5628 | |
5629 | /* | |
5630 | * Call ss->fork(). This must happen after @child is linked on | |
5631 | * css_set; otherwise, @child might change state between ->fork() | |
5632 | * and addition to css_set. | |
5633 | */ | |
b4e0eeaf | 5634 | do_each_subsys_mask(ss, i, have_fork_callback) { |
b53202e6 | 5635 | ss->fork(child); |
b4e0eeaf | 5636 | } while_each_subsys_mask(); |
817929ec | 5637 | } |
5edee61e | 5638 | |
b4f48b63 PM |
5639 | /** |
5640 | * cgroup_exit - detach cgroup from exiting task | |
5641 | * @tsk: pointer to task_struct of exiting process | |
5642 | * | |
5643 | * Description: Detach cgroup from @tsk and release it. | |
5644 | * | |
5645 | * Note that cgroups marked notify_on_release force every task in | |
5646 | * them to take the global cgroup_mutex mutex when exiting. | |
5647 | * This could impact scaling on very large systems. Be reluctant to | |
5648 | * use notify_on_release cgroups where very high task exit scaling | |
5649 | * is required on large systems. | |
5650 | * | |
0e1d768f TH |
5651 | * We set the exiting tasks cgroup to the root cgroup (top_cgroup). We |
5652 | * call cgroup_exit() while the task is still competent to handle | |
5653 | * notify_on_release(), then leave the task attached to the root cgroup in | |
5654 | * each hierarchy for the remainder of its exit. No need to bother with | |
5655 | * init_css_set refcnting. init_css_set never goes away and we can't race | |
e8604cb4 | 5656 | * with migration path - PF_EXITING is visible to migration path. |
b4f48b63 | 5657 | */ |
1ec41830 | 5658 | void cgroup_exit(struct task_struct *tsk) |
b4f48b63 | 5659 | { |
30159ec7 | 5660 | struct cgroup_subsys *ss; |
5abb8855 | 5661 | struct css_set *cset; |
d41d5a01 | 5662 | int i; |
817929ec PM |
5663 | |
5664 | /* | |
0e1d768f | 5665 | * Unlink from @tsk from its css_set. As migration path can't race |
0de0942d | 5666 | * with us, we can check css_set and cg_list without synchronization. |
817929ec | 5667 | */ |
0de0942d TH |
5668 | cset = task_css_set(tsk); |
5669 | ||
817929ec | 5670 | if (!list_empty(&tsk->cg_list)) { |
82d6489d | 5671 | spin_lock_irq(&css_set_lock); |
f6d7d049 | 5672 | css_set_move_task(tsk, cset, NULL, false); |
798cc880 | 5673 | list_add_tail(&tsk->cg_list, &cset->dying_tasks); |
73a7242a | 5674 | cset->nr_tasks--; |
82d6489d | 5675 | spin_unlock_irq(&css_set_lock); |
2e91fa7f TH |
5676 | } else { |
5677 | get_css_set(cset); | |
817929ec PM |
5678 | } |
5679 | ||
cb4a3167 | 5680 | /* see cgroup_post_fork() for details */ |
b4e0eeaf | 5681 | do_each_subsys_mask(ss, i, have_exit_callback) { |
2e91fa7f | 5682 | ss->exit(tsk); |
b4e0eeaf | 5683 | } while_each_subsys_mask(); |
2e91fa7f | 5684 | } |
30159ec7 | 5685 | |
d7248a93 | 5686 | void cgroup_release(struct task_struct *task) |
2e91fa7f | 5687 | { |
afcf6c8b TH |
5688 | struct cgroup_subsys *ss; |
5689 | int ssid; | |
5690 | ||
d7248a93 ON |
5691 | do_each_subsys_mask(ss, ssid, have_release_callback) { |
5692 | ss->release(task); | |
b4e0eeaf | 5693 | } while_each_subsys_mask(); |
798cc880 TH |
5694 | |
5695 | if (use_task_css_set_links) { | |
5696 | spin_lock_irq(&css_set_lock); | |
5697 | css_set_skip_task_iters(task_css_set(task), task); | |
5698 | list_del_init(&task->cg_list); | |
5699 | spin_unlock_irq(&css_set_lock); | |
5700 | } | |
d7248a93 | 5701 | } |
d41d5a01 | 5702 | |
d7248a93 ON |
5703 | void cgroup_free(struct task_struct *task) |
5704 | { | |
5705 | struct css_set *cset = task_css_set(task); | |
2e91fa7f | 5706 | put_css_set(cset); |
b4f48b63 | 5707 | } |
697f4161 | 5708 | |
8bab8dde PM |
5709 | static int __init cgroup_disable(char *str) |
5710 | { | |
30159ec7 | 5711 | struct cgroup_subsys *ss; |
8bab8dde | 5712 | char *token; |
30159ec7 | 5713 | int i; |
8bab8dde PM |
5714 | |
5715 | while ((token = strsep(&str, ",")) != NULL) { | |
5716 | if (!*token) | |
5717 | continue; | |
be45c900 | 5718 | |
3ed80a62 | 5719 | for_each_subsys(ss, i) { |
3e1d2eed TH |
5720 | if (strcmp(token, ss->name) && |
5721 | strcmp(token, ss->legacy_name)) | |
5722 | continue; | |
a3e72739 | 5723 | cgroup_disable_mask |= 1 << i; |
8bab8dde PM |
5724 | } |
5725 | } | |
5726 | return 1; | |
5727 | } | |
5728 | __setup("cgroup_disable=", cgroup_disable); | |
38460b48 | 5729 | |
b77d7b60 | 5730 | /** |
ec903c0c | 5731 | * css_tryget_online_from_dir - get corresponding css from a cgroup dentry |
35cf0836 TH |
5732 | * @dentry: directory dentry of interest |
5733 | * @ss: subsystem of interest | |
b77d7b60 | 5734 | * |
5a17f543 TH |
5735 | * If @dentry is a directory for a cgroup which has @ss enabled on it, try |
5736 | * to get the corresponding css and return it. If such css doesn't exist | |
5737 | * or can't be pinned, an ERR_PTR value is returned. | |
e5d1367f | 5738 | */ |
ec903c0c TH |
5739 | struct cgroup_subsys_state *css_tryget_online_from_dir(struct dentry *dentry, |
5740 | struct cgroup_subsys *ss) | |
e5d1367f | 5741 | { |
2bd59d48 | 5742 | struct kernfs_node *kn = kernfs_node_from_dentry(dentry); |
f17fc25f | 5743 | struct file_system_type *s_type = dentry->d_sb->s_type; |
2bd59d48 | 5744 | struct cgroup_subsys_state *css = NULL; |
e5d1367f | 5745 | struct cgroup *cgrp; |
e5d1367f | 5746 | |
35cf0836 | 5747 | /* is @dentry a cgroup dir? */ |
f17fc25f TH |
5748 | if ((s_type != &cgroup_fs_type && s_type != &cgroup2_fs_type) || |
5749 | !kn || kernfs_type(kn) != KERNFS_DIR) | |
e5d1367f SE |
5750 | return ERR_PTR(-EBADF); |
5751 | ||
5a17f543 TH |
5752 | rcu_read_lock(); |
5753 | ||
2bd59d48 TH |
5754 | /* |
5755 | * This path doesn't originate from kernfs and @kn could already | |
5756 | * have been or be removed at any point. @kn->priv is RCU | |
a4189487 | 5757 | * protected for this access. See css_release_work_fn() for details. |
2bd59d48 | 5758 | */ |
e0aed7c7 | 5759 | cgrp = rcu_dereference(*(void __rcu __force **)&kn->priv); |
2bd59d48 TH |
5760 | if (cgrp) |
5761 | css = cgroup_css(cgrp, ss); | |
5a17f543 | 5762 | |
ec903c0c | 5763 | if (!css || !css_tryget_online(css)) |
5a17f543 TH |
5764 | css = ERR_PTR(-ENOENT); |
5765 | ||
5766 | rcu_read_unlock(); | |
5767 | return css; | |
e5d1367f | 5768 | } |
e5d1367f | 5769 | |
1cb650b9 LZ |
5770 | /** |
5771 | * css_from_id - lookup css by id | |
5772 | * @id: the cgroup id | |
5773 | * @ss: cgroup subsys to be looked into | |
5774 | * | |
5775 | * Returns the css if there's valid one with @id, otherwise returns NULL. | |
5776 | * Should be called under rcu_read_lock(). | |
5777 | */ | |
5778 | struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss) | |
5779 | { | |
6fa4918d | 5780 | WARN_ON_ONCE(!rcu_read_lock_held()); |
d6ccc55e | 5781 | return idr_find(&ss->css_idr, id); |
e5d1367f SE |
5782 | } |
5783 | ||
16af4396 TH |
5784 | /** |
5785 | * cgroup_get_from_path - lookup and get a cgroup from its default hierarchy path | |
5786 | * @path: path on the default hierarchy | |
5787 | * | |
5788 | * Find the cgroup at @path on the default hierarchy, increment its | |
5789 | * reference count and return it. Returns pointer to the found cgroup on | |
5790 | * success, ERR_PTR(-ENOENT) if @path doens't exist and ERR_PTR(-ENOTDIR) | |
5791 | * if @path points to a non-directory. | |
5792 | */ | |
5793 | struct cgroup *cgroup_get_from_path(const char *path) | |
5794 | { | |
5795 | struct kernfs_node *kn; | |
5796 | struct cgroup *cgrp; | |
5797 | ||
5798 | mutex_lock(&cgroup_mutex); | |
5799 | ||
5800 | kn = kernfs_walk_and_get(cgrp_dfl_root.cgrp.kn, path); | |
5801 | if (kn) { | |
5802 | if (kernfs_type(kn) == KERNFS_DIR) { | |
5803 | cgrp = kn->priv; | |
a590b90d | 5804 | cgroup_get_live(cgrp); |
16af4396 TH |
5805 | } else { |
5806 | cgrp = ERR_PTR(-ENOTDIR); | |
5807 | } | |
5808 | kernfs_put(kn); | |
5809 | } else { | |
5810 | cgrp = ERR_PTR(-ENOENT); | |
5811 | } | |
5812 | ||
5813 | mutex_unlock(&cgroup_mutex); | |
5814 | return cgrp; | |
5815 | } | |
5816 | EXPORT_SYMBOL_GPL(cgroup_get_from_path); | |
5817 | ||
1f3fe7eb MKL |
5818 | /** |
5819 | * cgroup_get_from_fd - get a cgroup pointer from a fd | |
5820 | * @fd: fd obtained by open(cgroup2_dir) | |
5821 | * | |
5822 | * Find the cgroup from a fd which should be obtained | |
5823 | * by opening a cgroup directory. Returns a pointer to the | |
5824 | * cgroup on success. ERR_PTR is returned if the cgroup | |
5825 | * cannot be found. | |
5826 | */ | |
5827 | struct cgroup *cgroup_get_from_fd(int fd) | |
5828 | { | |
5829 | struct cgroup_subsys_state *css; | |
5830 | struct cgroup *cgrp; | |
5831 | struct file *f; | |
5832 | ||
5833 | f = fget_raw(fd); | |
5834 | if (!f) | |
5835 | return ERR_PTR(-EBADF); | |
5836 | ||
5837 | css = css_tryget_online_from_dir(f->f_path.dentry, NULL); | |
5838 | fput(f); | |
5839 | if (IS_ERR(css)) | |
5840 | return ERR_CAST(css); | |
5841 | ||
5842 | cgrp = css->cgroup; | |
5843 | if (!cgroup_on_dfl(cgrp)) { | |
5844 | cgroup_put(cgrp); | |
5845 | return ERR_PTR(-EBADF); | |
5846 | } | |
5847 | ||
5848 | return cgrp; | |
5849 | } | |
5850 | EXPORT_SYMBOL_GPL(cgroup_get_from_fd); | |
5851 | ||
bd1060a1 TH |
5852 | /* |
5853 | * sock->sk_cgrp_data handling. For more info, see sock_cgroup_data | |
5854 | * definition in cgroup-defs.h. | |
5855 | */ | |
5856 | #ifdef CONFIG_SOCK_CGROUP_DATA | |
5857 | ||
5858 | #if defined(CONFIG_CGROUP_NET_PRIO) || defined(CONFIG_CGROUP_NET_CLASSID) | |
5859 | ||
3fa4cc9c | 5860 | DEFINE_SPINLOCK(cgroup_sk_update_lock); |
bd1060a1 TH |
5861 | static bool cgroup_sk_alloc_disabled __read_mostly; |
5862 | ||
5863 | void cgroup_sk_alloc_disable(void) | |
5864 | { | |
5865 | if (cgroup_sk_alloc_disabled) | |
5866 | return; | |
5867 | pr_info("cgroup: disabling cgroup2 socket matching due to net_prio or net_cls activation\n"); | |
5868 | cgroup_sk_alloc_disabled = true; | |
5869 | } | |
5870 | ||
5871 | #else | |
5872 | ||
5873 | #define cgroup_sk_alloc_disabled false | |
5874 | ||
5875 | #endif | |
5876 | ||
5877 | void cgroup_sk_alloc(struct sock_cgroup_data *skcd) | |
5878 | { | |
5879 | if (cgroup_sk_alloc_disabled) | |
5880 | return; | |
5881 | ||
d979a39d JW |
5882 | /* Socket clone path */ |
5883 | if (skcd->val) { | |
a590b90d TH |
5884 | /* |
5885 | * We might be cloning a socket which is left in an empty | |
5886 | * cgroup and the cgroup might have already been rmdir'd. | |
5887 | * Don't use cgroup_get_live(). | |
5888 | */ | |
d979a39d JW |
5889 | cgroup_get(sock_cgroup_ptr(skcd)); |
5890 | return; | |
5891 | } | |
5892 | ||
bd1060a1 TH |
5893 | rcu_read_lock(); |
5894 | ||
5895 | while (true) { | |
5896 | struct css_set *cset; | |
5897 | ||
5898 | cset = task_css_set(current); | |
5899 | if (likely(cgroup_tryget(cset->dfl_cgrp))) { | |
5900 | skcd->val = (unsigned long)cset->dfl_cgrp; | |
5901 | break; | |
5902 | } | |
5903 | cpu_relax(); | |
5904 | } | |
5905 | ||
5906 | rcu_read_unlock(); | |
5907 | } | |
5908 | ||
5909 | void cgroup_sk_free(struct sock_cgroup_data *skcd) | |
5910 | { | |
5911 | cgroup_put(sock_cgroup_ptr(skcd)); | |
5912 | } | |
5913 | ||
5914 | #endif /* CONFIG_SOCK_CGROUP_DATA */ | |
5915 | ||
30070984 | 5916 | #ifdef CONFIG_CGROUP_BPF |
324bda9e AS |
5917 | int cgroup_bpf_attach(struct cgroup *cgrp, struct bpf_prog *prog, |
5918 | enum bpf_attach_type type, u32 flags) | |
5919 | { | |
5920 | int ret; | |
5921 | ||
5922 | mutex_lock(&cgroup_mutex); | |
5923 | ret = __cgroup_bpf_attach(cgrp, prog, type, flags); | |
5924 | mutex_unlock(&cgroup_mutex); | |
5925 | return ret; | |
5926 | } | |
5927 | int cgroup_bpf_detach(struct cgroup *cgrp, struct bpf_prog *prog, | |
5928 | enum bpf_attach_type type, u32 flags) | |
30070984 | 5929 | { |
7f677633 | 5930 | int ret; |
30070984 DM |
5931 | |
5932 | mutex_lock(&cgroup_mutex); | |
324bda9e | 5933 | ret = __cgroup_bpf_detach(cgrp, prog, type, flags); |
30070984 | 5934 | mutex_unlock(&cgroup_mutex); |
7f677633 | 5935 | return ret; |
30070984 | 5936 | } |
468e2f64 AS |
5937 | int cgroup_bpf_query(struct cgroup *cgrp, const union bpf_attr *attr, |
5938 | union bpf_attr __user *uattr) | |
5939 | { | |
5940 | int ret; | |
5941 | ||
5942 | mutex_lock(&cgroup_mutex); | |
5943 | ret = __cgroup_bpf_query(cgrp, attr, uattr); | |
5944 | mutex_unlock(&cgroup_mutex); | |
5945 | return ret; | |
5946 | } | |
30070984 | 5947 | #endif /* CONFIG_CGROUP_BPF */ |
01ee6cfb RG |
5948 | |
5949 | #ifdef CONFIG_SYSFS | |
5950 | static ssize_t show_delegatable_files(struct cftype *files, char *buf, | |
5951 | ssize_t size, const char *prefix) | |
5952 | { | |
5953 | struct cftype *cft; | |
5954 | ssize_t ret = 0; | |
5955 | ||
5956 | for (cft = files; cft && cft->name[0] != '\0'; cft++) { | |
5957 | if (!(cft->flags & CFTYPE_NS_DELEGATABLE)) | |
5958 | continue; | |
5959 | ||
5960 | if (prefix) | |
5961 | ret += snprintf(buf + ret, size - ret, "%s.", prefix); | |
5962 | ||
5963 | ret += snprintf(buf + ret, size - ret, "%s\n", cft->name); | |
5964 | ||
5965 | if (unlikely(ret >= size)) { | |
5966 | WARN_ON(1); | |
5967 | break; | |
5968 | } | |
5969 | } | |
5970 | ||
5971 | return ret; | |
5972 | } | |
5973 | ||
5974 | static ssize_t delegate_show(struct kobject *kobj, struct kobj_attribute *attr, | |
5975 | char *buf) | |
5976 | { | |
5977 | struct cgroup_subsys *ss; | |
5978 | int ssid; | |
5979 | ssize_t ret = 0; | |
5980 | ||
5981 | ret = show_delegatable_files(cgroup_base_files, buf, PAGE_SIZE - ret, | |
5982 | NULL); | |
5983 | ||
5984 | for_each_subsys(ss, ssid) | |
5985 | ret += show_delegatable_files(ss->dfl_cftypes, buf + ret, | |
5986 | PAGE_SIZE - ret, | |
5987 | cgroup_subsys_name[ssid]); | |
5988 | ||
5989 | return ret; | |
5990 | } | |
5991 | static struct kobj_attribute cgroup_delegate_attr = __ATTR_RO(delegate); | |
5992 | ||
5f2e6734 RG |
5993 | static ssize_t features_show(struct kobject *kobj, struct kobj_attribute *attr, |
5994 | char *buf) | |
5995 | { | |
5996 | return snprintf(buf, PAGE_SIZE, "nsdelegate\n"); | |
5997 | } | |
5998 | static struct kobj_attribute cgroup_features_attr = __ATTR_RO(features); | |
5999 | ||
01ee6cfb RG |
6000 | static struct attribute *cgroup_sysfs_attrs[] = { |
6001 | &cgroup_delegate_attr.attr, | |
5f2e6734 | 6002 | &cgroup_features_attr.attr, |
01ee6cfb RG |
6003 | NULL, |
6004 | }; | |
6005 | ||
6006 | static const struct attribute_group cgroup_sysfs_attr_group = { | |
6007 | .attrs = cgroup_sysfs_attrs, | |
6008 | .name = "cgroup", | |
6009 | }; | |
6010 | ||
6011 | static int __init cgroup_sysfs_init(void) | |
6012 | { | |
6013 | return sysfs_create_group(kernel_kobj, &cgroup_sysfs_attr_group); | |
6014 | } | |
6015 | subsys_initcall(cgroup_sysfs_init); | |
6016 | #endif /* CONFIG_SYSFS */ |