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 | ||
ddbcc7e8 | 145 | /* |
3dd06ffa | 146 | * The default hierarchy, reserved for the subsystems that are otherwise |
9871bf95 TH |
147 | * unattached - it never has more than a single cgroup, and all tasks are |
148 | * part of that cgroup. | |
ddbcc7e8 | 149 | */ |
a2dd4247 | 150 | struct cgroup_root cgrp_dfl_root; |
d0ec4230 | 151 | EXPORT_SYMBOL_GPL(cgrp_dfl_root); |
9871bf95 | 152 | |
a2dd4247 TH |
153 | /* |
154 | * The default hierarchy always exists but is hidden until mounted for the | |
155 | * first time. This is for backward compatibility. | |
156 | */ | |
a7165264 | 157 | static bool cgrp_dfl_visible; |
ddbcc7e8 | 158 | |
5533e011 | 159 | /* some controllers are not supported in the default hierarchy */ |
a7165264 | 160 | static u16 cgrp_dfl_inhibit_ss_mask; |
5533e011 | 161 | |
f6d635ad | 162 | /* some controllers are implicitly enabled on the default hierarchy */ |
b807421a | 163 | static u16 cgrp_dfl_implicit_ss_mask; |
f6d635ad | 164 | |
ddbcc7e8 | 165 | /* The list of hierarchy roots */ |
0a268dbd | 166 | LIST_HEAD(cgroup_roots); |
9871bf95 | 167 | static int cgroup_root_count; |
ddbcc7e8 | 168 | |
3417ae1f | 169 | /* hierarchy ID allocation and mapping, protected by cgroup_mutex */ |
1a574231 | 170 | static DEFINE_IDR(cgroup_hierarchy_idr); |
2c6ab6d2 | 171 | |
794611a1 | 172 | /* |
0cb51d71 TH |
173 | * Assign a monotonically increasing serial number to csses. It guarantees |
174 | * cgroups with bigger numbers are newer than those with smaller numbers. | |
175 | * Also, as csses are always appended to the parent's ->children list, it | |
176 | * guarantees that sibling csses are always sorted in the ascending serial | |
177 | * number order on the list. Protected by cgroup_mutex. | |
794611a1 | 178 | */ |
0cb51d71 | 179 | static u64 css_serial_nr_next = 1; |
794611a1 | 180 | |
cb4a3167 | 181 | /* |
b807421a TH |
182 | * These bitmasks identify subsystems with specific features to avoid |
183 | * having to do iterative checks repeatedly. | |
ddbcc7e8 | 184 | */ |
6e5c8307 TH |
185 | static u16 have_fork_callback __read_mostly; |
186 | static u16 have_exit_callback __read_mostly; | |
187 | static u16 have_free_callback __read_mostly; | |
b807421a | 188 | static u16 have_canfork_callback __read_mostly; |
ddbcc7e8 | 189 | |
a79a908f AK |
190 | /* cgroup namespace for init task */ |
191 | struct cgroup_namespace init_cgroup_ns = { | |
387ad967 | 192 | .count = REFCOUNT_INIT(2), |
a79a908f AK |
193 | .user_ns = &init_user_ns, |
194 | .ns.ops = &cgroupns_operations, | |
195 | .ns.inum = PROC_CGROUP_INIT_INO, | |
196 | .root_cset = &init_css_set, | |
197 | }; | |
198 | ||
67e9c74b | 199 | static struct file_system_type cgroup2_fs_type; |
d62beb7f | 200 | static struct cftype cgroup_base_files[]; |
628f7cd4 | 201 | |
334c3679 TH |
202 | static int cgroup_apply_control(struct cgroup *cgrp); |
203 | static void cgroup_finalize_control(struct cgroup *cgrp, int ret); | |
ed27b9f7 | 204 | static void css_task_iter_advance(struct css_task_iter *it); |
42809dd4 | 205 | static int cgroup_destroy_locked(struct cgroup *cgrp); |
6cd0f5bb TH |
206 | static struct cgroup_subsys_state *css_create(struct cgroup *cgrp, |
207 | struct cgroup_subsys *ss); | |
9d755d33 | 208 | static void css_release(struct percpu_ref *ref); |
f8f22e53 | 209 | static void kill_css(struct cgroup_subsys_state *css); |
4df8dc90 TH |
210 | static int cgroup_addrm_files(struct cgroup_subsys_state *css, |
211 | struct cgroup *cgrp, struct cftype cfts[], | |
2bb566cb | 212 | bool is_add); |
42809dd4 | 213 | |
fc5ed1e9 TH |
214 | /** |
215 | * cgroup_ssid_enabled - cgroup subsys enabled test by subsys ID | |
216 | * @ssid: subsys ID of interest | |
217 | * | |
218 | * cgroup_subsys_enabled() can only be used with literal subsys names which | |
219 | * is fine for individual subsystems but unsuitable for cgroup core. This | |
220 | * is slower static_key_enabled() based test indexed by @ssid. | |
221 | */ | |
0a268dbd | 222 | bool cgroup_ssid_enabled(int ssid) |
fc5ed1e9 | 223 | { |
cfe02a8a AB |
224 | if (CGROUP_SUBSYS_COUNT == 0) |
225 | return false; | |
226 | ||
fc5ed1e9 TH |
227 | return static_key_enabled(cgroup_subsys_enabled_key[ssid]); |
228 | } | |
229 | ||
9e10a130 TH |
230 | /** |
231 | * cgroup_on_dfl - test whether a cgroup is on the default hierarchy | |
232 | * @cgrp: the cgroup of interest | |
233 | * | |
234 | * The default hierarchy is the v2 interface of cgroup and this function | |
235 | * can be used to test whether a cgroup is on the default hierarchy for | |
236 | * cases where a subsystem should behave differnetly depending on the | |
237 | * interface version. | |
238 | * | |
239 | * The set of behaviors which change on the default hierarchy are still | |
240 | * being determined and the mount option is prefixed with __DEVEL__. | |
241 | * | |
242 | * List of changed behaviors: | |
243 | * | |
244 | * - Mount options "noprefix", "xattr", "clone_children", "release_agent" | |
245 | * and "name" are disallowed. | |
246 | * | |
247 | * - When mounting an existing superblock, mount options should match. | |
248 | * | |
249 | * - Remount is disallowed. | |
250 | * | |
251 | * - rename(2) is disallowed. | |
252 | * | |
253 | * - "tasks" is removed. Everything should be at process granularity. Use | |
254 | * "cgroup.procs" instead. | |
255 | * | |
256 | * - "cgroup.procs" is not sorted. pids will be unique unless they got | |
257 | * recycled inbetween reads. | |
258 | * | |
259 | * - "release_agent" and "notify_on_release" are removed. Replacement | |
260 | * notification mechanism will be implemented. | |
261 | * | |
262 | * - "cgroup.clone_children" is removed. | |
263 | * | |
264 | * - "cgroup.subtree_populated" is available. Its value is 0 if the cgroup | |
265 | * and its descendants contain no task; otherwise, 1. The file also | |
266 | * generates kernfs notification which can be monitored through poll and | |
267 | * [di]notify when the value of the file changes. | |
268 | * | |
269 | * - cpuset: tasks will be kept in empty cpusets when hotplug happens and | |
270 | * take masks of ancestors with non-empty cpus/mems, instead of being | |
271 | * moved to an ancestor. | |
272 | * | |
273 | * - cpuset: a task can be moved into an empty cpuset, and again it takes | |
274 | * masks of ancestors. | |
275 | * | |
276 | * - memcg: use_hierarchy is on by default and the cgroup file for the flag | |
277 | * is not created. | |
278 | * | |
279 | * - blkcg: blk-throttle becomes properly hierarchical. | |
280 | * | |
281 | * - debug: disallowed on the default hierarchy. | |
282 | */ | |
0a268dbd | 283 | bool cgroup_on_dfl(const struct cgroup *cgrp) |
9e10a130 TH |
284 | { |
285 | return cgrp->root == &cgrp_dfl_root; | |
286 | } | |
287 | ||
6fa4918d TH |
288 | /* IDR wrappers which synchronize using cgroup_idr_lock */ |
289 | static int cgroup_idr_alloc(struct idr *idr, void *ptr, int start, int end, | |
290 | gfp_t gfp_mask) | |
291 | { | |
292 | int ret; | |
293 | ||
294 | idr_preload(gfp_mask); | |
54504e97 | 295 | spin_lock_bh(&cgroup_idr_lock); |
d0164adc | 296 | ret = idr_alloc(idr, ptr, start, end, gfp_mask & ~__GFP_DIRECT_RECLAIM); |
54504e97 | 297 | spin_unlock_bh(&cgroup_idr_lock); |
6fa4918d TH |
298 | idr_preload_end(); |
299 | return ret; | |
300 | } | |
301 | ||
302 | static void *cgroup_idr_replace(struct idr *idr, void *ptr, int id) | |
303 | { | |
304 | void *ret; | |
305 | ||
54504e97 | 306 | spin_lock_bh(&cgroup_idr_lock); |
6fa4918d | 307 | ret = idr_replace(idr, ptr, id); |
54504e97 | 308 | spin_unlock_bh(&cgroup_idr_lock); |
6fa4918d TH |
309 | return ret; |
310 | } | |
311 | ||
312 | static void cgroup_idr_remove(struct idr *idr, int id) | |
313 | { | |
54504e97 | 314 | spin_lock_bh(&cgroup_idr_lock); |
6fa4918d | 315 | idr_remove(idr, id); |
54504e97 | 316 | spin_unlock_bh(&cgroup_idr_lock); |
6fa4918d TH |
317 | } |
318 | ||
d51f39b0 TH |
319 | static struct cgroup *cgroup_parent(struct cgroup *cgrp) |
320 | { | |
321 | struct cgroup_subsys_state *parent_css = cgrp->self.parent; | |
322 | ||
323 | if (parent_css) | |
324 | return container_of(parent_css, struct cgroup, self); | |
325 | return NULL; | |
326 | } | |
327 | ||
5531dc91 TH |
328 | /* subsystems visibly enabled on a cgroup */ |
329 | static u16 cgroup_control(struct cgroup *cgrp) | |
330 | { | |
331 | struct cgroup *parent = cgroup_parent(cgrp); | |
332 | u16 root_ss_mask = cgrp->root->subsys_mask; | |
333 | ||
334 | if (parent) | |
335 | return parent->subtree_control; | |
336 | ||
337 | if (cgroup_on_dfl(cgrp)) | |
f6d635ad TH |
338 | root_ss_mask &= ~(cgrp_dfl_inhibit_ss_mask | |
339 | cgrp_dfl_implicit_ss_mask); | |
5531dc91 TH |
340 | return root_ss_mask; |
341 | } | |
342 | ||
343 | /* subsystems enabled on a cgroup */ | |
344 | static u16 cgroup_ss_mask(struct cgroup *cgrp) | |
345 | { | |
346 | struct cgroup *parent = cgroup_parent(cgrp); | |
347 | ||
348 | if (parent) | |
349 | return parent->subtree_ss_mask; | |
350 | ||
351 | return cgrp->root->subsys_mask; | |
352 | } | |
353 | ||
95109b62 TH |
354 | /** |
355 | * cgroup_css - obtain a cgroup's css for the specified subsystem | |
356 | * @cgrp: the cgroup of interest | |
9d800df1 | 357 | * @ss: the subsystem of interest (%NULL returns @cgrp->self) |
95109b62 | 358 | * |
ca8bdcaf TH |
359 | * Return @cgrp's css (cgroup_subsys_state) associated with @ss. This |
360 | * function must be called either under cgroup_mutex or rcu_read_lock() and | |
361 | * the caller is responsible for pinning the returned css if it wants to | |
362 | * keep accessing it outside the said locks. This function may return | |
363 | * %NULL if @cgrp doesn't have @subsys_id enabled. | |
95109b62 TH |
364 | */ |
365 | static struct cgroup_subsys_state *cgroup_css(struct cgroup *cgrp, | |
ca8bdcaf | 366 | struct cgroup_subsys *ss) |
95109b62 | 367 | { |
ca8bdcaf | 368 | if (ss) |
aec25020 | 369 | return rcu_dereference_check(cgrp->subsys[ss->id], |
ace2bee8 | 370 | lockdep_is_held(&cgroup_mutex)); |
ca8bdcaf | 371 | else |
9d800df1 | 372 | return &cgrp->self; |
95109b62 | 373 | } |
42809dd4 | 374 | |
aec3dfcb TH |
375 | /** |
376 | * cgroup_e_css - obtain a cgroup's effective css for the specified subsystem | |
377 | * @cgrp: the cgroup of interest | |
9d800df1 | 378 | * @ss: the subsystem of interest (%NULL returns @cgrp->self) |
aec3dfcb | 379 | * |
d0f702e6 | 380 | * Similar to cgroup_css() but returns the effective css, which is defined |
aec3dfcb TH |
381 | * as the matching css of the nearest ancestor including self which has @ss |
382 | * enabled. If @ss is associated with the hierarchy @cgrp is on, this | |
383 | * function is guaranteed to return non-NULL css. | |
384 | */ | |
385 | static struct cgroup_subsys_state *cgroup_e_css(struct cgroup *cgrp, | |
386 | struct cgroup_subsys *ss) | |
387 | { | |
388 | lockdep_assert_held(&cgroup_mutex); | |
389 | ||
390 | if (!ss) | |
9d800df1 | 391 | return &cgrp->self; |
aec3dfcb | 392 | |
eeecbd19 TH |
393 | /* |
394 | * This function is used while updating css associations and thus | |
5531dc91 | 395 | * can't test the csses directly. Test ss_mask. |
eeecbd19 | 396 | */ |
5531dc91 | 397 | while (!(cgroup_ss_mask(cgrp) & (1 << ss->id))) { |
d51f39b0 | 398 | cgrp = cgroup_parent(cgrp); |
5531dc91 TH |
399 | if (!cgrp) |
400 | return NULL; | |
401 | } | |
aec3dfcb TH |
402 | |
403 | return cgroup_css(cgrp, ss); | |
95109b62 | 404 | } |
42809dd4 | 405 | |
eeecbd19 TH |
406 | /** |
407 | * cgroup_get_e_css - get a cgroup's effective css for the specified subsystem | |
408 | * @cgrp: the cgroup of interest | |
409 | * @ss: the subsystem of interest | |
410 | * | |
411 | * Find and get the effective css of @cgrp for @ss. The effective css is | |
412 | * defined as the matching css of the nearest ancestor including self which | |
413 | * has @ss enabled. If @ss is not mounted on the hierarchy @cgrp is on, | |
414 | * the root css is returned, so this function always returns a valid css. | |
415 | * The returned css must be put using css_put(). | |
416 | */ | |
417 | struct cgroup_subsys_state *cgroup_get_e_css(struct cgroup *cgrp, | |
418 | struct cgroup_subsys *ss) | |
419 | { | |
420 | struct cgroup_subsys_state *css; | |
421 | ||
422 | rcu_read_lock(); | |
423 | ||
424 | do { | |
425 | css = cgroup_css(cgrp, ss); | |
426 | ||
427 | if (css && css_tryget_online(css)) | |
428 | goto out_unlock; | |
429 | cgrp = cgroup_parent(cgrp); | |
430 | } while (cgrp); | |
431 | ||
432 | css = init_css_set.subsys[ss->id]; | |
433 | css_get(css); | |
434 | out_unlock: | |
435 | rcu_read_unlock(); | |
436 | return css; | |
437 | } | |
438 | ||
310b4816 | 439 | static void __maybe_unused cgroup_get(struct cgroup *cgrp) |
a590b90d TH |
440 | { |
441 | css_get(&cgrp->self); | |
442 | } | |
443 | ||
444 | static void cgroup_get_live(struct cgroup *cgrp) | |
052c3f3a TH |
445 | { |
446 | WARN_ON_ONCE(cgroup_is_dead(cgrp)); | |
447 | css_get(&cgrp->self); | |
448 | } | |
449 | ||
450 | static bool cgroup_tryget(struct cgroup *cgrp) | |
451 | { | |
452 | return css_tryget(&cgrp->self); | |
453 | } | |
454 | ||
b4168640 | 455 | struct cgroup_subsys_state *of_css(struct kernfs_open_file *of) |
59f5296b | 456 | { |
2bd59d48 | 457 | struct cgroup *cgrp = of->kn->parent->priv; |
b4168640 | 458 | struct cftype *cft = of_cft(of); |
2bd59d48 TH |
459 | |
460 | /* | |
461 | * This is open and unprotected implementation of cgroup_css(). | |
462 | * seq_css() is only called from a kernfs file operation which has | |
463 | * an active reference on the file. Because all the subsystem | |
464 | * files are drained before a css is disassociated with a cgroup, | |
465 | * the matching css from the cgroup's subsys table is guaranteed to | |
466 | * be and stay valid until the enclosing operation is complete. | |
467 | */ | |
468 | if (cft->ss) | |
469 | return rcu_dereference_raw(cgrp->subsys[cft->ss->id]); | |
470 | else | |
9d800df1 | 471 | return &cgrp->self; |
59f5296b | 472 | } |
b4168640 | 473 | EXPORT_SYMBOL_GPL(of_css); |
59f5296b | 474 | |
1c6727af TH |
475 | /** |
476 | * for_each_css - iterate all css's of a cgroup | |
477 | * @css: the iteration cursor | |
478 | * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end | |
479 | * @cgrp: the target cgroup to iterate css's of | |
480 | * | |
aec3dfcb | 481 | * Should be called under cgroup_[tree_]mutex. |
1c6727af TH |
482 | */ |
483 | #define for_each_css(css, ssid, cgrp) \ | |
484 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \ | |
485 | if (!((css) = rcu_dereference_check( \ | |
486 | (cgrp)->subsys[(ssid)], \ | |
487 | lockdep_is_held(&cgroup_mutex)))) { } \ | |
488 | else | |
489 | ||
aec3dfcb TH |
490 | /** |
491 | * for_each_e_css - iterate all effective css's of a cgroup | |
492 | * @css: the iteration cursor | |
493 | * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end | |
494 | * @cgrp: the target cgroup to iterate css's of | |
495 | * | |
496 | * Should be called under cgroup_[tree_]mutex. | |
497 | */ | |
498 | #define for_each_e_css(css, ssid, cgrp) \ | |
499 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \ | |
500 | if (!((css) = cgroup_e_css(cgrp, cgroup_subsys[(ssid)]))) \ | |
501 | ; \ | |
502 | else | |
503 | ||
cb4a3167 | 504 | /** |
b4e0eeaf | 505 | * do_each_subsys_mask - filter for_each_subsys with a bitmask |
cb4a3167 AS |
506 | * @ss: the iteration cursor |
507 | * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end | |
b4e0eeaf | 508 | * @ss_mask: the bitmask |
cb4a3167 AS |
509 | * |
510 | * The block will only run for cases where the ssid-th bit (1 << ssid) of | |
b4e0eeaf | 511 | * @ss_mask is set. |
cb4a3167 | 512 | */ |
b4e0eeaf TH |
513 | #define do_each_subsys_mask(ss, ssid, ss_mask) do { \ |
514 | unsigned long __ss_mask = (ss_mask); \ | |
515 | if (!CGROUP_SUBSYS_COUNT) { /* to avoid spurious gcc warning */ \ | |
4a705c5c | 516 | (ssid) = 0; \ |
b4e0eeaf TH |
517 | break; \ |
518 | } \ | |
519 | for_each_set_bit(ssid, &__ss_mask, CGROUP_SUBSYS_COUNT) { \ | |
520 | (ss) = cgroup_subsys[ssid]; \ | |
521 | { | |
522 | ||
523 | #define while_each_subsys_mask() \ | |
524 | } \ | |
525 | } \ | |
526 | } while (false) | |
cb4a3167 | 527 | |
f8f22e53 TH |
528 | /* iterate over child cgrps, lock should be held throughout iteration */ |
529 | #define cgroup_for_each_live_child(child, cgrp) \ | |
d5c419b6 | 530 | list_for_each_entry((child), &(cgrp)->self.children, self.sibling) \ |
8353da1f | 531 | if (({ lockdep_assert_held(&cgroup_mutex); \ |
f8f22e53 TH |
532 | cgroup_is_dead(child); })) \ |
533 | ; \ | |
534 | else | |
7ae1bad9 | 535 | |
ce3f1d9d TH |
536 | /* walk live descendants in preorder */ |
537 | #define cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) \ | |
538 | css_for_each_descendant_pre((d_css), cgroup_css((cgrp), NULL)) \ | |
539 | if (({ lockdep_assert_held(&cgroup_mutex); \ | |
540 | (dsct) = (d_css)->cgroup; \ | |
541 | cgroup_is_dead(dsct); })) \ | |
542 | ; \ | |
543 | else | |
544 | ||
545 | /* walk live descendants in postorder */ | |
546 | #define cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) \ | |
547 | css_for_each_descendant_post((d_css), cgroup_css((cgrp), NULL)) \ | |
548 | if (({ lockdep_assert_held(&cgroup_mutex); \ | |
549 | (dsct) = (d_css)->cgroup; \ | |
550 | cgroup_is_dead(dsct); })) \ | |
551 | ; \ | |
552 | else | |
553 | ||
172a2c06 TH |
554 | /* |
555 | * The default css_set - used by init and its children prior to any | |
817929ec PM |
556 | * hierarchies being mounted. It contains a pointer to the root state |
557 | * for each subsystem. Also used to anchor the list of css_sets. Not | |
558 | * reference-counted, to improve performance when child cgroups | |
559 | * haven't been created. | |
560 | */ | |
5024ae29 | 561 | struct css_set init_css_set = { |
4b9502e6 | 562 | .refcount = REFCOUNT_INIT(1), |
172a2c06 TH |
563 | .tasks = LIST_HEAD_INIT(init_css_set.tasks), |
564 | .mg_tasks = LIST_HEAD_INIT(init_css_set.mg_tasks), | |
5f617ebb TH |
565 | .task_iters = LIST_HEAD_INIT(init_css_set.task_iters), |
566 | .cgrp_links = LIST_HEAD_INIT(init_css_set.cgrp_links), | |
172a2c06 TH |
567 | .mg_preload_node = LIST_HEAD_INIT(init_css_set.mg_preload_node), |
568 | .mg_node = LIST_HEAD_INIT(init_css_set.mg_node), | |
569 | }; | |
817929ec | 570 | |
172a2c06 | 571 | static int css_set_count = 1; /* 1 for init_css_set */ |
817929ec | 572 | |
0de0942d TH |
573 | /** |
574 | * css_set_populated - does a css_set contain any tasks? | |
575 | * @cset: target css_set | |
576 | */ | |
577 | static bool css_set_populated(struct css_set *cset) | |
578 | { | |
f0d9a5f1 | 579 | lockdep_assert_held(&css_set_lock); |
0de0942d TH |
580 | |
581 | return !list_empty(&cset->tasks) || !list_empty(&cset->mg_tasks); | |
582 | } | |
583 | ||
842b597e TH |
584 | /** |
585 | * cgroup_update_populated - updated populated count of a cgroup | |
586 | * @cgrp: the target cgroup | |
587 | * @populated: inc or dec populated count | |
588 | * | |
0de0942d TH |
589 | * One of the css_sets associated with @cgrp is either getting its first |
590 | * task or losing the last. Update @cgrp->populated_cnt accordingly. The | |
591 | * count is propagated towards root so that a given cgroup's populated_cnt | |
592 | * is zero iff the cgroup and all its descendants don't contain any tasks. | |
842b597e TH |
593 | * |
594 | * @cgrp's interface file "cgroup.populated" is zero if | |
595 | * @cgrp->populated_cnt is zero and 1 otherwise. When @cgrp->populated_cnt | |
596 | * changes from or to zero, userland is notified that the content of the | |
597 | * interface file has changed. This can be used to detect when @cgrp and | |
598 | * its descendants become populated or empty. | |
599 | */ | |
600 | static void cgroup_update_populated(struct cgroup *cgrp, bool populated) | |
601 | { | |
f0d9a5f1 | 602 | lockdep_assert_held(&css_set_lock); |
842b597e TH |
603 | |
604 | do { | |
605 | bool trigger; | |
606 | ||
607 | if (populated) | |
608 | trigger = !cgrp->populated_cnt++; | |
609 | else | |
610 | trigger = !--cgrp->populated_cnt; | |
611 | ||
612 | if (!trigger) | |
613 | break; | |
614 | ||
d62beb7f | 615 | cgroup1_check_for_release(cgrp); |
6f60eade TH |
616 | cgroup_file_notify(&cgrp->events_file); |
617 | ||
d51f39b0 | 618 | cgrp = cgroup_parent(cgrp); |
842b597e TH |
619 | } while (cgrp); |
620 | } | |
621 | ||
0de0942d TH |
622 | /** |
623 | * css_set_update_populated - update populated state of a css_set | |
624 | * @cset: target css_set | |
625 | * @populated: whether @cset is populated or depopulated | |
626 | * | |
627 | * @cset is either getting the first task or losing the last. Update the | |
628 | * ->populated_cnt of all associated cgroups accordingly. | |
629 | */ | |
630 | static void css_set_update_populated(struct css_set *cset, bool populated) | |
631 | { | |
632 | struct cgrp_cset_link *link; | |
633 | ||
f0d9a5f1 | 634 | lockdep_assert_held(&css_set_lock); |
0de0942d TH |
635 | |
636 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) | |
637 | cgroup_update_populated(link->cgrp, populated); | |
638 | } | |
639 | ||
f6d7d049 TH |
640 | /** |
641 | * css_set_move_task - move a task from one css_set to another | |
642 | * @task: task being moved | |
643 | * @from_cset: css_set @task currently belongs to (may be NULL) | |
644 | * @to_cset: new css_set @task is being moved to (may be NULL) | |
645 | * @use_mg_tasks: move to @to_cset->mg_tasks instead of ->tasks | |
646 | * | |
647 | * Move @task from @from_cset to @to_cset. If @task didn't belong to any | |
648 | * css_set, @from_cset can be NULL. If @task is being disassociated | |
649 | * instead of moved, @to_cset can be NULL. | |
650 | * | |
ed27b9f7 TH |
651 | * This function automatically handles populated_cnt updates and |
652 | * css_task_iter adjustments but the caller is responsible for managing | |
653 | * @from_cset and @to_cset's reference counts. | |
f6d7d049 TH |
654 | */ |
655 | static void css_set_move_task(struct task_struct *task, | |
656 | struct css_set *from_cset, struct css_set *to_cset, | |
657 | bool use_mg_tasks) | |
658 | { | |
f0d9a5f1 | 659 | lockdep_assert_held(&css_set_lock); |
f6d7d049 | 660 | |
20b454a6 TH |
661 | if (to_cset && !css_set_populated(to_cset)) |
662 | css_set_update_populated(to_cset, true); | |
663 | ||
f6d7d049 | 664 | if (from_cset) { |
ed27b9f7 TH |
665 | struct css_task_iter *it, *pos; |
666 | ||
f6d7d049 | 667 | WARN_ON_ONCE(list_empty(&task->cg_list)); |
ed27b9f7 TH |
668 | |
669 | /* | |
670 | * @task is leaving, advance task iterators which are | |
671 | * pointing to it so that they can resume at the next | |
672 | * position. Advancing an iterator might remove it from | |
673 | * the list, use safe walk. See css_task_iter_advance*() | |
674 | * for details. | |
675 | */ | |
676 | list_for_each_entry_safe(it, pos, &from_cset->task_iters, | |
677 | iters_node) | |
678 | if (it->task_pos == &task->cg_list) | |
679 | css_task_iter_advance(it); | |
680 | ||
f6d7d049 TH |
681 | list_del_init(&task->cg_list); |
682 | if (!css_set_populated(from_cset)) | |
683 | css_set_update_populated(from_cset, false); | |
684 | } else { | |
685 | WARN_ON_ONCE(!list_empty(&task->cg_list)); | |
686 | } | |
687 | ||
688 | if (to_cset) { | |
689 | /* | |
690 | * We are synchronized through cgroup_threadgroup_rwsem | |
691 | * against PF_EXITING setting such that we can't race | |
692 | * against cgroup_exit() changing the css_set to | |
693 | * init_css_set and dropping the old one. | |
694 | */ | |
695 | WARN_ON_ONCE(task->flags & PF_EXITING); | |
696 | ||
f6d7d049 TH |
697 | rcu_assign_pointer(task->cgroups, to_cset); |
698 | list_add_tail(&task->cg_list, use_mg_tasks ? &to_cset->mg_tasks : | |
699 | &to_cset->tasks); | |
700 | } | |
701 | } | |
702 | ||
7717f7ba PM |
703 | /* |
704 | * hash table for cgroup groups. This improves the performance to find | |
705 | * an existing css_set. This hash doesn't (currently) take into | |
706 | * account cgroups in empty hierarchies. | |
707 | */ | |
472b1053 | 708 | #define CSS_SET_HASH_BITS 7 |
0ac801fe | 709 | static DEFINE_HASHTABLE(css_set_table, CSS_SET_HASH_BITS); |
472b1053 | 710 | |
0ac801fe | 711 | static unsigned long css_set_hash(struct cgroup_subsys_state *css[]) |
472b1053 | 712 | { |
0ac801fe | 713 | unsigned long key = 0UL; |
30159ec7 TH |
714 | struct cgroup_subsys *ss; |
715 | int i; | |
472b1053 | 716 | |
30159ec7 | 717 | for_each_subsys(ss, i) |
0ac801fe LZ |
718 | key += (unsigned long)css[i]; |
719 | key = (key >> 16) ^ key; | |
472b1053 | 720 | |
0ac801fe | 721 | return key; |
472b1053 LZ |
722 | } |
723 | ||
dcfe149b | 724 | void put_css_set_locked(struct css_set *cset) |
b4f48b63 | 725 | { |
69d0206c | 726 | struct cgrp_cset_link *link, *tmp_link; |
2d8f243a TH |
727 | struct cgroup_subsys *ss; |
728 | int ssid; | |
5abb8855 | 729 | |
f0d9a5f1 | 730 | lockdep_assert_held(&css_set_lock); |
89c5509b | 731 | |
4b9502e6 | 732 | if (!refcount_dec_and_test(&cset->refcount)) |
146aa1bd | 733 | return; |
81a6a5cd | 734 | |
53254f90 TH |
735 | /* This css_set is dead. unlink it and release cgroup and css refs */ |
736 | for_each_subsys(ss, ssid) { | |
2d8f243a | 737 | list_del(&cset->e_cset_node[ssid]); |
53254f90 TH |
738 | css_put(cset->subsys[ssid]); |
739 | } | |
5abb8855 | 740 | hash_del(&cset->hlist); |
2c6ab6d2 PM |
741 | css_set_count--; |
742 | ||
69d0206c | 743 | list_for_each_entry_safe(link, tmp_link, &cset->cgrp_links, cgrp_link) { |
69d0206c TH |
744 | list_del(&link->cset_link); |
745 | list_del(&link->cgrp_link); | |
2ceb231b TH |
746 | if (cgroup_parent(link->cgrp)) |
747 | cgroup_put(link->cgrp); | |
2c6ab6d2 | 748 | kfree(link); |
81a6a5cd | 749 | } |
2c6ab6d2 | 750 | |
5abb8855 | 751 | kfree_rcu(cset, rcu_head); |
b4f48b63 PM |
752 | } |
753 | ||
b326f9d0 | 754 | /** |
7717f7ba | 755 | * compare_css_sets - helper function for find_existing_css_set(). |
5abb8855 TH |
756 | * @cset: candidate css_set being tested |
757 | * @old_cset: existing css_set for a task | |
7717f7ba PM |
758 | * @new_cgrp: cgroup that's being entered by the task |
759 | * @template: desired set of css pointers in css_set (pre-calculated) | |
760 | * | |
6f4b7e63 | 761 | * Returns true if "cset" matches "old_cset" except for the hierarchy |
7717f7ba PM |
762 | * which "new_cgrp" belongs to, for which it should match "new_cgrp". |
763 | */ | |
5abb8855 TH |
764 | static bool compare_css_sets(struct css_set *cset, |
765 | struct css_set *old_cset, | |
7717f7ba PM |
766 | struct cgroup *new_cgrp, |
767 | struct cgroup_subsys_state *template[]) | |
768 | { | |
769 | struct list_head *l1, *l2; | |
770 | ||
aec3dfcb TH |
771 | /* |
772 | * On the default hierarchy, there can be csets which are | |
773 | * associated with the same set of cgroups but different csses. | |
774 | * Let's first ensure that csses match. | |
775 | */ | |
776 | if (memcmp(template, cset->subsys, sizeof(cset->subsys))) | |
7717f7ba | 777 | return false; |
7717f7ba PM |
778 | |
779 | /* | |
780 | * Compare cgroup pointers in order to distinguish between | |
aec3dfcb TH |
781 | * different cgroups in hierarchies. As different cgroups may |
782 | * share the same effective css, this comparison is always | |
783 | * necessary. | |
7717f7ba | 784 | */ |
69d0206c TH |
785 | l1 = &cset->cgrp_links; |
786 | l2 = &old_cset->cgrp_links; | |
7717f7ba | 787 | while (1) { |
69d0206c | 788 | struct cgrp_cset_link *link1, *link2; |
5abb8855 | 789 | struct cgroup *cgrp1, *cgrp2; |
7717f7ba PM |
790 | |
791 | l1 = l1->next; | |
792 | l2 = l2->next; | |
793 | /* See if we reached the end - both lists are equal length. */ | |
69d0206c TH |
794 | if (l1 == &cset->cgrp_links) { |
795 | BUG_ON(l2 != &old_cset->cgrp_links); | |
7717f7ba PM |
796 | break; |
797 | } else { | |
69d0206c | 798 | BUG_ON(l2 == &old_cset->cgrp_links); |
7717f7ba PM |
799 | } |
800 | /* Locate the cgroups associated with these links. */ | |
69d0206c TH |
801 | link1 = list_entry(l1, struct cgrp_cset_link, cgrp_link); |
802 | link2 = list_entry(l2, struct cgrp_cset_link, cgrp_link); | |
803 | cgrp1 = link1->cgrp; | |
804 | cgrp2 = link2->cgrp; | |
7717f7ba | 805 | /* Hierarchies should be linked in the same order. */ |
5abb8855 | 806 | BUG_ON(cgrp1->root != cgrp2->root); |
7717f7ba PM |
807 | |
808 | /* | |
809 | * If this hierarchy is the hierarchy of the cgroup | |
810 | * that's changing, then we need to check that this | |
811 | * css_set points to the new cgroup; if it's any other | |
812 | * hierarchy, then this css_set should point to the | |
813 | * same cgroup as the old css_set. | |
814 | */ | |
5abb8855 TH |
815 | if (cgrp1->root == new_cgrp->root) { |
816 | if (cgrp1 != new_cgrp) | |
7717f7ba PM |
817 | return false; |
818 | } else { | |
5abb8855 | 819 | if (cgrp1 != cgrp2) |
7717f7ba PM |
820 | return false; |
821 | } | |
822 | } | |
823 | return true; | |
824 | } | |
825 | ||
b326f9d0 TH |
826 | /** |
827 | * find_existing_css_set - init css array and find the matching css_set | |
828 | * @old_cset: the css_set that we're using before the cgroup transition | |
829 | * @cgrp: the cgroup that we're moving into | |
830 | * @template: out param for the new set of csses, should be clear on entry | |
817929ec | 831 | */ |
5abb8855 TH |
832 | static struct css_set *find_existing_css_set(struct css_set *old_cset, |
833 | struct cgroup *cgrp, | |
834 | struct cgroup_subsys_state *template[]) | |
b4f48b63 | 835 | { |
3dd06ffa | 836 | struct cgroup_root *root = cgrp->root; |
30159ec7 | 837 | struct cgroup_subsys *ss; |
5abb8855 | 838 | struct css_set *cset; |
0ac801fe | 839 | unsigned long key; |
b326f9d0 | 840 | int i; |
817929ec | 841 | |
aae8aab4 BB |
842 | /* |
843 | * Build the set of subsystem state objects that we want to see in the | |
844 | * new css_set. while subsystems can change globally, the entries here | |
845 | * won't change, so no need for locking. | |
846 | */ | |
30159ec7 | 847 | for_each_subsys(ss, i) { |
f392e51c | 848 | if (root->subsys_mask & (1UL << i)) { |
aec3dfcb TH |
849 | /* |
850 | * @ss is in this hierarchy, so we want the | |
851 | * effective css from @cgrp. | |
852 | */ | |
853 | template[i] = cgroup_e_css(cgrp, ss); | |
817929ec | 854 | } else { |
aec3dfcb TH |
855 | /* |
856 | * @ss is not in this hierarchy, so we don't want | |
857 | * to change the css. | |
858 | */ | |
5abb8855 | 859 | template[i] = old_cset->subsys[i]; |
817929ec PM |
860 | } |
861 | } | |
862 | ||
0ac801fe | 863 | key = css_set_hash(template); |
5abb8855 TH |
864 | hash_for_each_possible(css_set_table, cset, hlist, key) { |
865 | if (!compare_css_sets(cset, old_cset, cgrp, template)) | |
7717f7ba PM |
866 | continue; |
867 | ||
868 | /* This css_set matches what we need */ | |
5abb8855 | 869 | return cset; |
472b1053 | 870 | } |
817929ec PM |
871 | |
872 | /* No existing cgroup group matched */ | |
873 | return NULL; | |
874 | } | |
875 | ||
69d0206c | 876 | static void free_cgrp_cset_links(struct list_head *links_to_free) |
36553434 | 877 | { |
69d0206c | 878 | struct cgrp_cset_link *link, *tmp_link; |
36553434 | 879 | |
69d0206c TH |
880 | list_for_each_entry_safe(link, tmp_link, links_to_free, cset_link) { |
881 | list_del(&link->cset_link); | |
36553434 LZ |
882 | kfree(link); |
883 | } | |
884 | } | |
885 | ||
69d0206c TH |
886 | /** |
887 | * allocate_cgrp_cset_links - allocate cgrp_cset_links | |
888 | * @count: the number of links to allocate | |
889 | * @tmp_links: list_head the allocated links are put on | |
890 | * | |
891 | * Allocate @count cgrp_cset_link structures and chain them on @tmp_links | |
892 | * through ->cset_link. Returns 0 on success or -errno. | |
817929ec | 893 | */ |
69d0206c | 894 | static int allocate_cgrp_cset_links(int count, struct list_head *tmp_links) |
817929ec | 895 | { |
69d0206c | 896 | struct cgrp_cset_link *link; |
817929ec | 897 | int i; |
69d0206c TH |
898 | |
899 | INIT_LIST_HEAD(tmp_links); | |
900 | ||
817929ec | 901 | for (i = 0; i < count; i++) { |
f4f4be2b | 902 | link = kzalloc(sizeof(*link), GFP_KERNEL); |
817929ec | 903 | if (!link) { |
69d0206c | 904 | free_cgrp_cset_links(tmp_links); |
817929ec PM |
905 | return -ENOMEM; |
906 | } | |
69d0206c | 907 | list_add(&link->cset_link, tmp_links); |
817929ec PM |
908 | } |
909 | return 0; | |
910 | } | |
911 | ||
c12f65d4 LZ |
912 | /** |
913 | * link_css_set - a helper function to link a css_set to a cgroup | |
69d0206c | 914 | * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links() |
5abb8855 | 915 | * @cset: the css_set to be linked |
c12f65d4 LZ |
916 | * @cgrp: the destination cgroup |
917 | */ | |
69d0206c TH |
918 | static void link_css_set(struct list_head *tmp_links, struct css_set *cset, |
919 | struct cgroup *cgrp) | |
c12f65d4 | 920 | { |
69d0206c | 921 | struct cgrp_cset_link *link; |
c12f65d4 | 922 | |
69d0206c | 923 | BUG_ON(list_empty(tmp_links)); |
6803c006 TH |
924 | |
925 | if (cgroup_on_dfl(cgrp)) | |
926 | cset->dfl_cgrp = cgrp; | |
927 | ||
69d0206c TH |
928 | link = list_first_entry(tmp_links, struct cgrp_cset_link, cset_link); |
929 | link->cset = cset; | |
7717f7ba | 930 | link->cgrp = cgrp; |
842b597e | 931 | |
7717f7ba | 932 | /* |
389b9c1b TH |
933 | * Always add links to the tail of the lists so that the lists are |
934 | * in choronological order. | |
7717f7ba | 935 | */ |
389b9c1b | 936 | list_move_tail(&link->cset_link, &cgrp->cset_links); |
69d0206c | 937 | list_add_tail(&link->cgrp_link, &cset->cgrp_links); |
2ceb231b TH |
938 | |
939 | if (cgroup_parent(cgrp)) | |
a590b90d | 940 | cgroup_get_live(cgrp); |
c12f65d4 LZ |
941 | } |
942 | ||
b326f9d0 TH |
943 | /** |
944 | * find_css_set - return a new css_set with one cgroup updated | |
945 | * @old_cset: the baseline css_set | |
946 | * @cgrp: the cgroup to be updated | |
947 | * | |
948 | * Return a new css_set that's equivalent to @old_cset, but with @cgrp | |
949 | * substituted into the appropriate hierarchy. | |
817929ec | 950 | */ |
5abb8855 TH |
951 | static struct css_set *find_css_set(struct css_set *old_cset, |
952 | struct cgroup *cgrp) | |
817929ec | 953 | { |
b326f9d0 | 954 | struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT] = { }; |
5abb8855 | 955 | struct css_set *cset; |
69d0206c TH |
956 | struct list_head tmp_links; |
957 | struct cgrp_cset_link *link; | |
2d8f243a | 958 | struct cgroup_subsys *ss; |
0ac801fe | 959 | unsigned long key; |
2d8f243a | 960 | int ssid; |
472b1053 | 961 | |
b326f9d0 TH |
962 | lockdep_assert_held(&cgroup_mutex); |
963 | ||
817929ec PM |
964 | /* First see if we already have a cgroup group that matches |
965 | * the desired set */ | |
82d6489d | 966 | spin_lock_irq(&css_set_lock); |
5abb8855 TH |
967 | cset = find_existing_css_set(old_cset, cgrp, template); |
968 | if (cset) | |
969 | get_css_set(cset); | |
82d6489d | 970 | spin_unlock_irq(&css_set_lock); |
817929ec | 971 | |
5abb8855 TH |
972 | if (cset) |
973 | return cset; | |
817929ec | 974 | |
f4f4be2b | 975 | cset = kzalloc(sizeof(*cset), GFP_KERNEL); |
5abb8855 | 976 | if (!cset) |
817929ec PM |
977 | return NULL; |
978 | ||
69d0206c | 979 | /* Allocate all the cgrp_cset_link objects that we'll need */ |
9871bf95 | 980 | if (allocate_cgrp_cset_links(cgroup_root_count, &tmp_links) < 0) { |
5abb8855 | 981 | kfree(cset); |
817929ec PM |
982 | return NULL; |
983 | } | |
984 | ||
4b9502e6 | 985 | refcount_set(&cset->refcount, 1); |
5abb8855 | 986 | INIT_LIST_HEAD(&cset->tasks); |
c7561128 | 987 | INIT_LIST_HEAD(&cset->mg_tasks); |
ed27b9f7 | 988 | INIT_LIST_HEAD(&cset->task_iters); |
5abb8855 | 989 | INIT_HLIST_NODE(&cset->hlist); |
5f617ebb TH |
990 | INIT_LIST_HEAD(&cset->cgrp_links); |
991 | INIT_LIST_HEAD(&cset->mg_preload_node); | |
992 | INIT_LIST_HEAD(&cset->mg_node); | |
817929ec PM |
993 | |
994 | /* Copy the set of subsystem state objects generated in | |
995 | * find_existing_css_set() */ | |
5abb8855 | 996 | memcpy(cset->subsys, template, sizeof(cset->subsys)); |
817929ec | 997 | |
82d6489d | 998 | spin_lock_irq(&css_set_lock); |
817929ec | 999 | /* Add reference counts and links from the new css_set. */ |
69d0206c | 1000 | list_for_each_entry(link, &old_cset->cgrp_links, cgrp_link) { |
7717f7ba | 1001 | struct cgroup *c = link->cgrp; |
69d0206c | 1002 | |
7717f7ba PM |
1003 | if (c->root == cgrp->root) |
1004 | c = cgrp; | |
69d0206c | 1005 | link_css_set(&tmp_links, cset, c); |
7717f7ba | 1006 | } |
817929ec | 1007 | |
69d0206c | 1008 | BUG_ON(!list_empty(&tmp_links)); |
817929ec | 1009 | |
817929ec | 1010 | css_set_count++; |
472b1053 | 1011 | |
2d8f243a | 1012 | /* Add @cset to the hash table */ |
5abb8855 TH |
1013 | key = css_set_hash(cset->subsys); |
1014 | hash_add(css_set_table, &cset->hlist, key); | |
472b1053 | 1015 | |
53254f90 TH |
1016 | for_each_subsys(ss, ssid) { |
1017 | struct cgroup_subsys_state *css = cset->subsys[ssid]; | |
1018 | ||
2d8f243a | 1019 | list_add_tail(&cset->e_cset_node[ssid], |
53254f90 TH |
1020 | &css->cgroup->e_csets[ssid]); |
1021 | css_get(css); | |
1022 | } | |
2d8f243a | 1023 | |
82d6489d | 1024 | spin_unlock_irq(&css_set_lock); |
817929ec | 1025 | |
5abb8855 | 1026 | return cset; |
b4f48b63 PM |
1027 | } |
1028 | ||
0a268dbd | 1029 | struct cgroup_root *cgroup_root_from_kf(struct kernfs_root *kf_root) |
7717f7ba | 1030 | { |
3dd06ffa | 1031 | struct cgroup *root_cgrp = kf_root->kn->priv; |
2bd59d48 | 1032 | |
3dd06ffa | 1033 | return root_cgrp->root; |
2bd59d48 TH |
1034 | } |
1035 | ||
3dd06ffa | 1036 | static int cgroup_init_root_id(struct cgroup_root *root) |
f2e85d57 TH |
1037 | { |
1038 | int id; | |
1039 | ||
1040 | lockdep_assert_held(&cgroup_mutex); | |
1041 | ||
985ed670 | 1042 | id = idr_alloc_cyclic(&cgroup_hierarchy_idr, root, 0, 0, GFP_KERNEL); |
f2e85d57 TH |
1043 | if (id < 0) |
1044 | return id; | |
1045 | ||
1046 | root->hierarchy_id = id; | |
1047 | return 0; | |
1048 | } | |
1049 | ||
3dd06ffa | 1050 | static void cgroup_exit_root_id(struct cgroup_root *root) |
f2e85d57 TH |
1051 | { |
1052 | lockdep_assert_held(&cgroup_mutex); | |
1053 | ||
8c8a5502 | 1054 | idr_remove(&cgroup_hierarchy_idr, root->hierarchy_id); |
f2e85d57 TH |
1055 | } |
1056 | ||
1592c9b2 | 1057 | void cgroup_free_root(struct cgroup_root *root) |
f2e85d57 TH |
1058 | { |
1059 | if (root) { | |
f2e85d57 TH |
1060 | idr_destroy(&root->cgroup_idr); |
1061 | kfree(root); | |
1062 | } | |
1063 | } | |
1064 | ||
3dd06ffa | 1065 | static void cgroup_destroy_root(struct cgroup_root *root) |
59f5296b | 1066 | { |
3dd06ffa | 1067 | struct cgroup *cgrp = &root->cgrp; |
f2e85d57 | 1068 | struct cgrp_cset_link *link, *tmp_link; |
f2e85d57 | 1069 | |
ed1777de TH |
1070 | trace_cgroup_destroy_root(root); |
1071 | ||
334c3679 | 1072 | cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp); |
f2e85d57 | 1073 | |
776f02fa | 1074 | BUG_ON(atomic_read(&root->nr_cgrps)); |
d5c419b6 | 1075 | BUG_ON(!list_empty(&cgrp->self.children)); |
f2e85d57 | 1076 | |
f2e85d57 | 1077 | /* Rebind all subsystems back to the default hierarchy */ |
334c3679 | 1078 | WARN_ON(rebind_subsystems(&cgrp_dfl_root, root->subsys_mask)); |
7717f7ba | 1079 | |
7717f7ba | 1080 | /* |
f2e85d57 TH |
1081 | * Release all the links from cset_links to this hierarchy's |
1082 | * root cgroup | |
7717f7ba | 1083 | */ |
82d6489d | 1084 | spin_lock_irq(&css_set_lock); |
f2e85d57 TH |
1085 | |
1086 | list_for_each_entry_safe(link, tmp_link, &cgrp->cset_links, cset_link) { | |
1087 | list_del(&link->cset_link); | |
1088 | list_del(&link->cgrp_link); | |
1089 | kfree(link); | |
1090 | } | |
f0d9a5f1 | 1091 | |
82d6489d | 1092 | spin_unlock_irq(&css_set_lock); |
f2e85d57 TH |
1093 | |
1094 | if (!list_empty(&root->root_list)) { | |
1095 | list_del(&root->root_list); | |
1096 | cgroup_root_count--; | |
1097 | } | |
1098 | ||
1099 | cgroup_exit_root_id(root); | |
1100 | ||
1101 | mutex_unlock(&cgroup_mutex); | |
f2e85d57 | 1102 | |
2bd59d48 | 1103 | kernfs_destroy_root(root->kf_root); |
f2e85d57 TH |
1104 | cgroup_free_root(root); |
1105 | } | |
1106 | ||
4f41fc59 SH |
1107 | /* |
1108 | * look up cgroup associated with current task's cgroup namespace on the | |
1109 | * specified hierarchy | |
1110 | */ | |
1111 | static struct cgroup * | |
1112 | current_cgns_cgroup_from_root(struct cgroup_root *root) | |
1113 | { | |
1114 | struct cgroup *res = NULL; | |
1115 | struct css_set *cset; | |
1116 | ||
1117 | lockdep_assert_held(&css_set_lock); | |
1118 | ||
1119 | rcu_read_lock(); | |
1120 | ||
1121 | cset = current->nsproxy->cgroup_ns->root_cset; | |
1122 | if (cset == &init_css_set) { | |
1123 | res = &root->cgrp; | |
1124 | } else { | |
1125 | struct cgrp_cset_link *link; | |
1126 | ||
1127 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) { | |
1128 | struct cgroup *c = link->cgrp; | |
1129 | ||
1130 | if (c->root == root) { | |
1131 | res = c; | |
1132 | break; | |
1133 | } | |
1134 | } | |
1135 | } | |
1136 | rcu_read_unlock(); | |
1137 | ||
1138 | BUG_ON(!res); | |
1139 | return res; | |
1140 | } | |
1141 | ||
ceb6a081 TH |
1142 | /* look up cgroup associated with given css_set on the specified hierarchy */ |
1143 | static struct cgroup *cset_cgroup_from_root(struct css_set *cset, | |
3dd06ffa | 1144 | struct cgroup_root *root) |
7717f7ba | 1145 | { |
7717f7ba PM |
1146 | struct cgroup *res = NULL; |
1147 | ||
96d365e0 | 1148 | lockdep_assert_held(&cgroup_mutex); |
f0d9a5f1 | 1149 | lockdep_assert_held(&css_set_lock); |
96d365e0 | 1150 | |
5abb8855 | 1151 | if (cset == &init_css_set) { |
3dd06ffa | 1152 | res = &root->cgrp; |
7717f7ba | 1153 | } else { |
69d0206c TH |
1154 | struct cgrp_cset_link *link; |
1155 | ||
1156 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) { | |
7717f7ba | 1157 | struct cgroup *c = link->cgrp; |
69d0206c | 1158 | |
7717f7ba PM |
1159 | if (c->root == root) { |
1160 | res = c; | |
1161 | break; | |
1162 | } | |
1163 | } | |
1164 | } | |
96d365e0 | 1165 | |
7717f7ba PM |
1166 | BUG_ON(!res); |
1167 | return res; | |
1168 | } | |
1169 | ||
ddbcc7e8 | 1170 | /* |
ceb6a081 | 1171 | * Return the cgroup for "task" from the given hierarchy. Must be |
f0d9a5f1 | 1172 | * called with cgroup_mutex and css_set_lock held. |
ceb6a081 | 1173 | */ |
0a268dbd TH |
1174 | struct cgroup *task_cgroup_from_root(struct task_struct *task, |
1175 | struct cgroup_root *root) | |
ceb6a081 TH |
1176 | { |
1177 | /* | |
1178 | * No need to lock the task - since we hold cgroup_mutex the | |
1179 | * task can't change groups, so the only thing that can happen | |
1180 | * is that it exits and its css is set back to init_css_set. | |
1181 | */ | |
1182 | return cset_cgroup_from_root(task_css_set(task), root); | |
1183 | } | |
1184 | ||
ddbcc7e8 | 1185 | /* |
ddbcc7e8 PM |
1186 | * A task must hold cgroup_mutex to modify cgroups. |
1187 | * | |
1188 | * Any task can increment and decrement the count field without lock. | |
1189 | * So in general, code holding cgroup_mutex can't rely on the count | |
1190 | * field not changing. However, if the count goes to zero, then only | |
956db3ca | 1191 | * cgroup_attach_task() can increment it again. Because a count of zero |
ddbcc7e8 PM |
1192 | * means that no tasks are currently attached, therefore there is no |
1193 | * way a task attached to that cgroup can fork (the other way to | |
1194 | * increment the count). So code holding cgroup_mutex can safely | |
1195 | * assume that if the count is zero, it will stay zero. Similarly, if | |
1196 | * a task holds cgroup_mutex on a cgroup with zero count, it | |
1197 | * knows that the cgroup won't be removed, as cgroup_rmdir() | |
1198 | * needs that mutex. | |
1199 | * | |
ddbcc7e8 PM |
1200 | * A cgroup can only be deleted if both its 'count' of using tasks |
1201 | * is zero, and its list of 'children' cgroups is empty. Since all | |
1202 | * tasks in the system use _some_ cgroup, and since there is always at | |
3dd06ffa | 1203 | * least one task in the system (init, pid == 1), therefore, root cgroup |
ddbcc7e8 | 1204 | * always has either children cgroups and/or using tasks. So we don't |
3dd06ffa | 1205 | * need a special hack to ensure that root cgroup cannot be deleted. |
ddbcc7e8 PM |
1206 | * |
1207 | * P.S. One more locking exception. RCU is used to guard the | |
956db3ca | 1208 | * update of a tasks cgroup pointer by cgroup_attach_task() |
ddbcc7e8 PM |
1209 | */ |
1210 | ||
2bd59d48 | 1211 | static struct kernfs_syscall_ops cgroup_kf_syscall_ops; |
a424316c | 1212 | |
8d7e6fb0 TH |
1213 | static char *cgroup_file_name(struct cgroup *cgrp, const struct cftype *cft, |
1214 | char *buf) | |
ddbcc7e8 | 1215 | { |
3e1d2eed TH |
1216 | struct cgroup_subsys *ss = cft->ss; |
1217 | ||
8d7e6fb0 TH |
1218 | if (cft->ss && !(cft->flags & CFTYPE_NO_PREFIX) && |
1219 | !(cgrp->root->flags & CGRP_ROOT_NOPREFIX)) | |
1220 | snprintf(buf, CGROUP_FILE_NAME_MAX, "%s.%s", | |
3e1d2eed TH |
1221 | cgroup_on_dfl(cgrp) ? ss->name : ss->legacy_name, |
1222 | cft->name); | |
8d7e6fb0 TH |
1223 | else |
1224 | strncpy(buf, cft->name, CGROUP_FILE_NAME_MAX); | |
1225 | return buf; | |
ddbcc7e8 PM |
1226 | } |
1227 | ||
f2e85d57 TH |
1228 | /** |
1229 | * cgroup_file_mode - deduce file mode of a control file | |
1230 | * @cft: the control file in question | |
1231 | * | |
7dbdb199 | 1232 | * S_IRUGO for read, S_IWUSR for write. |
f2e85d57 TH |
1233 | */ |
1234 | static umode_t cgroup_file_mode(const struct cftype *cft) | |
65dff759 | 1235 | { |
f2e85d57 | 1236 | umode_t mode = 0; |
65dff759 | 1237 | |
f2e85d57 TH |
1238 | if (cft->read_u64 || cft->read_s64 || cft->seq_show) |
1239 | mode |= S_IRUGO; | |
1240 | ||
7dbdb199 TH |
1241 | if (cft->write_u64 || cft->write_s64 || cft->write) { |
1242 | if (cft->flags & CFTYPE_WORLD_WRITABLE) | |
1243 | mode |= S_IWUGO; | |
1244 | else | |
1245 | mode |= S_IWUSR; | |
1246 | } | |
f2e85d57 TH |
1247 | |
1248 | return mode; | |
65dff759 LZ |
1249 | } |
1250 | ||
af0ba678 | 1251 | /** |
8699b776 | 1252 | * cgroup_calc_subtree_ss_mask - calculate subtree_ss_mask |
0f060deb | 1253 | * @subtree_control: the new subtree_control mask to consider |
5ced2518 | 1254 | * @this_ss_mask: available subsystems |
af0ba678 TH |
1255 | * |
1256 | * On the default hierarchy, a subsystem may request other subsystems to be | |
1257 | * enabled together through its ->depends_on mask. In such cases, more | |
1258 | * subsystems than specified in "cgroup.subtree_control" may be enabled. | |
1259 | * | |
0f060deb | 1260 | * This function calculates which subsystems need to be enabled if |
5ced2518 | 1261 | * @subtree_control is to be applied while restricted to @this_ss_mask. |
af0ba678 | 1262 | */ |
5ced2518 | 1263 | static u16 cgroup_calc_subtree_ss_mask(u16 subtree_control, u16 this_ss_mask) |
667c2491 | 1264 | { |
6e5c8307 | 1265 | u16 cur_ss_mask = subtree_control; |
af0ba678 TH |
1266 | struct cgroup_subsys *ss; |
1267 | int ssid; | |
1268 | ||
1269 | lockdep_assert_held(&cgroup_mutex); | |
1270 | ||
f6d635ad TH |
1271 | cur_ss_mask |= cgrp_dfl_implicit_ss_mask; |
1272 | ||
af0ba678 | 1273 | while (true) { |
6e5c8307 | 1274 | u16 new_ss_mask = cur_ss_mask; |
af0ba678 | 1275 | |
b4e0eeaf | 1276 | do_each_subsys_mask(ss, ssid, cur_ss_mask) { |
a966a4ed | 1277 | new_ss_mask |= ss->depends_on; |
b4e0eeaf | 1278 | } while_each_subsys_mask(); |
af0ba678 TH |
1279 | |
1280 | /* | |
1281 | * Mask out subsystems which aren't available. This can | |
1282 | * happen only if some depended-upon subsystems were bound | |
1283 | * to non-default hierarchies. | |
1284 | */ | |
5ced2518 | 1285 | new_ss_mask &= this_ss_mask; |
af0ba678 TH |
1286 | |
1287 | if (new_ss_mask == cur_ss_mask) | |
1288 | break; | |
1289 | cur_ss_mask = new_ss_mask; | |
1290 | } | |
1291 | ||
0f060deb TH |
1292 | return cur_ss_mask; |
1293 | } | |
1294 | ||
a9746d8d TH |
1295 | /** |
1296 | * cgroup_kn_unlock - unlocking helper for cgroup kernfs methods | |
1297 | * @kn: the kernfs_node being serviced | |
1298 | * | |
1299 | * This helper undoes cgroup_kn_lock_live() and should be invoked before | |
1300 | * the method finishes if locking succeeded. Note that once this function | |
1301 | * returns the cgroup returned by cgroup_kn_lock_live() may become | |
1302 | * inaccessible any time. If the caller intends to continue to access the | |
1303 | * cgroup, it should pin it before invoking this function. | |
1304 | */ | |
0a268dbd | 1305 | void cgroup_kn_unlock(struct kernfs_node *kn) |
ddbcc7e8 | 1306 | { |
a9746d8d TH |
1307 | struct cgroup *cgrp; |
1308 | ||
1309 | if (kernfs_type(kn) == KERNFS_DIR) | |
1310 | cgrp = kn->priv; | |
1311 | else | |
1312 | cgrp = kn->parent->priv; | |
1313 | ||
1314 | mutex_unlock(&cgroup_mutex); | |
a9746d8d TH |
1315 | |
1316 | kernfs_unbreak_active_protection(kn); | |
1317 | cgroup_put(cgrp); | |
ddbcc7e8 PM |
1318 | } |
1319 | ||
a9746d8d TH |
1320 | /** |
1321 | * cgroup_kn_lock_live - locking helper for cgroup kernfs methods | |
1322 | * @kn: the kernfs_node being serviced | |
945ba199 | 1323 | * @drain_offline: perform offline draining on the cgroup |
a9746d8d TH |
1324 | * |
1325 | * This helper is to be used by a cgroup kernfs method currently servicing | |
1326 | * @kn. It breaks the active protection, performs cgroup locking and | |
1327 | * verifies that the associated cgroup is alive. Returns the cgroup if | |
1328 | * alive; otherwise, %NULL. A successful return should be undone by a | |
945ba199 TH |
1329 | * matching cgroup_kn_unlock() invocation. If @drain_offline is %true, the |
1330 | * cgroup is drained of offlining csses before return. | |
a9746d8d TH |
1331 | * |
1332 | * Any cgroup kernfs method implementation which requires locking the | |
1333 | * associated cgroup should use this helper. It avoids nesting cgroup | |
1334 | * locking under kernfs active protection and allows all kernfs operations | |
1335 | * including self-removal. | |
1336 | */ | |
0a268dbd | 1337 | struct cgroup *cgroup_kn_lock_live(struct kernfs_node *kn, bool drain_offline) |
05ef1d7c | 1338 | { |
a9746d8d TH |
1339 | struct cgroup *cgrp; |
1340 | ||
1341 | if (kernfs_type(kn) == KERNFS_DIR) | |
1342 | cgrp = kn->priv; | |
1343 | else | |
1344 | cgrp = kn->parent->priv; | |
05ef1d7c | 1345 | |
2739d3cc | 1346 | /* |
01f6474c | 1347 | * We're gonna grab cgroup_mutex which nests outside kernfs |
a9746d8d TH |
1348 | * active_ref. cgroup liveliness check alone provides enough |
1349 | * protection against removal. Ensure @cgrp stays accessible and | |
1350 | * break the active_ref protection. | |
2739d3cc | 1351 | */ |
aa32362f LZ |
1352 | if (!cgroup_tryget(cgrp)) |
1353 | return NULL; | |
a9746d8d TH |
1354 | kernfs_break_active_protection(kn); |
1355 | ||
945ba199 TH |
1356 | if (drain_offline) |
1357 | cgroup_lock_and_drain_offline(cgrp); | |
1358 | else | |
1359 | mutex_lock(&cgroup_mutex); | |
05ef1d7c | 1360 | |
a9746d8d TH |
1361 | if (!cgroup_is_dead(cgrp)) |
1362 | return cgrp; | |
1363 | ||
1364 | cgroup_kn_unlock(kn); | |
1365 | return NULL; | |
ddbcc7e8 | 1366 | } |
05ef1d7c | 1367 | |
2739d3cc | 1368 | static void cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft) |
05ef1d7c | 1369 | { |
2bd59d48 | 1370 | char name[CGROUP_FILE_NAME_MAX]; |
05ef1d7c | 1371 | |
01f6474c | 1372 | lockdep_assert_held(&cgroup_mutex); |
34c06254 TH |
1373 | |
1374 | if (cft->file_offset) { | |
1375 | struct cgroup_subsys_state *css = cgroup_css(cgrp, cft->ss); | |
1376 | struct cgroup_file *cfile = (void *)css + cft->file_offset; | |
1377 | ||
1378 | spin_lock_irq(&cgroup_file_kn_lock); | |
1379 | cfile->kn = NULL; | |
1380 | spin_unlock_irq(&cgroup_file_kn_lock); | |
1381 | } | |
1382 | ||
2bd59d48 | 1383 | kernfs_remove_by_name(cgrp->kn, cgroup_file_name(cgrp, cft, name)); |
05ef1d7c TH |
1384 | } |
1385 | ||
13af07df | 1386 | /** |
4df8dc90 TH |
1387 | * css_clear_dir - remove subsys files in a cgroup directory |
1388 | * @css: taget css | |
13af07df | 1389 | */ |
334c3679 | 1390 | static void css_clear_dir(struct cgroup_subsys_state *css) |
05ef1d7c | 1391 | { |
334c3679 | 1392 | struct cgroup *cgrp = css->cgroup; |
4df8dc90 | 1393 | struct cftype *cfts; |
05ef1d7c | 1394 | |
88cb04b9 TH |
1395 | if (!(css->flags & CSS_VISIBLE)) |
1396 | return; | |
1397 | ||
1398 | css->flags &= ~CSS_VISIBLE; | |
1399 | ||
4df8dc90 TH |
1400 | list_for_each_entry(cfts, &css->ss->cfts, node) |
1401 | cgroup_addrm_files(css, cgrp, cfts, false); | |
ddbcc7e8 PM |
1402 | } |
1403 | ||
ccdca218 | 1404 | /** |
4df8dc90 TH |
1405 | * css_populate_dir - create subsys files in a cgroup directory |
1406 | * @css: target css | |
ccdca218 TH |
1407 | * |
1408 | * On failure, no file is added. | |
1409 | */ | |
334c3679 | 1410 | static int css_populate_dir(struct cgroup_subsys_state *css) |
ccdca218 | 1411 | { |
334c3679 | 1412 | struct cgroup *cgrp = css->cgroup; |
4df8dc90 TH |
1413 | struct cftype *cfts, *failed_cfts; |
1414 | int ret; | |
ccdca218 | 1415 | |
03970d3c | 1416 | if ((css->flags & CSS_VISIBLE) || !cgrp->kn) |
88cb04b9 TH |
1417 | return 0; |
1418 | ||
4df8dc90 TH |
1419 | if (!css->ss) { |
1420 | if (cgroup_on_dfl(cgrp)) | |
d62beb7f | 1421 | cfts = cgroup_base_files; |
4df8dc90 | 1422 | else |
d62beb7f | 1423 | cfts = cgroup1_base_files; |
ccdca218 | 1424 | |
4df8dc90 TH |
1425 | return cgroup_addrm_files(&cgrp->self, cgrp, cfts, true); |
1426 | } | |
ccdca218 | 1427 | |
4df8dc90 TH |
1428 | list_for_each_entry(cfts, &css->ss->cfts, node) { |
1429 | ret = cgroup_addrm_files(css, cgrp, cfts, true); | |
1430 | if (ret < 0) { | |
1431 | failed_cfts = cfts; | |
1432 | goto err; | |
ccdca218 TH |
1433 | } |
1434 | } | |
88cb04b9 TH |
1435 | |
1436 | css->flags |= CSS_VISIBLE; | |
1437 | ||
ccdca218 TH |
1438 | return 0; |
1439 | err: | |
4df8dc90 TH |
1440 | list_for_each_entry(cfts, &css->ss->cfts, node) { |
1441 | if (cfts == failed_cfts) | |
1442 | break; | |
1443 | cgroup_addrm_files(css, cgrp, cfts, false); | |
1444 | } | |
ccdca218 TH |
1445 | return ret; |
1446 | } | |
1447 | ||
0a268dbd | 1448 | int rebind_subsystems(struct cgroup_root *dst_root, u16 ss_mask) |
ddbcc7e8 | 1449 | { |
1ada4838 | 1450 | struct cgroup *dcgrp = &dst_root->cgrp; |
30159ec7 | 1451 | struct cgroup_subsys *ss; |
2d8f243a | 1452 | int ssid, i, ret; |
ddbcc7e8 | 1453 | |
ace2bee8 | 1454 | lockdep_assert_held(&cgroup_mutex); |
ddbcc7e8 | 1455 | |
b4e0eeaf | 1456 | do_each_subsys_mask(ss, ssid, ss_mask) { |
f6d635ad TH |
1457 | /* |
1458 | * If @ss has non-root csses attached to it, can't move. | |
1459 | * If @ss is an implicit controller, it is exempt from this | |
1460 | * rule and can be stolen. | |
1461 | */ | |
1462 | if (css_next_child(NULL, cgroup_css(&ss->root->cgrp, ss)) && | |
1463 | !ss->implicit_on_dfl) | |
3ed80a62 | 1464 | return -EBUSY; |
1d5be6b2 | 1465 | |
5df36032 | 1466 | /* can't move between two non-dummy roots either */ |
7fd8c565 | 1467 | if (ss->root != &cgrp_dfl_root && dst_root != &cgrp_dfl_root) |
5df36032 | 1468 | return -EBUSY; |
b4e0eeaf | 1469 | } while_each_subsys_mask(); |
ddbcc7e8 | 1470 | |
b4e0eeaf | 1471 | do_each_subsys_mask(ss, ssid, ss_mask) { |
1ada4838 TH |
1472 | struct cgroup_root *src_root = ss->root; |
1473 | struct cgroup *scgrp = &src_root->cgrp; | |
1474 | struct cgroup_subsys_state *css = cgroup_css(scgrp, ss); | |
2d8f243a | 1475 | struct css_set *cset; |
a8a648c4 | 1476 | |
1ada4838 | 1477 | WARN_ON(!css || cgroup_css(dcgrp, ss)); |
a8a648c4 | 1478 | |
334c3679 TH |
1479 | /* disable from the source */ |
1480 | src_root->subsys_mask &= ~(1 << ssid); | |
1481 | WARN_ON(cgroup_apply_control(scgrp)); | |
1482 | cgroup_finalize_control(scgrp, 0); | |
4df8dc90 | 1483 | |
334c3679 | 1484 | /* rebind */ |
1ada4838 TH |
1485 | RCU_INIT_POINTER(scgrp->subsys[ssid], NULL); |
1486 | rcu_assign_pointer(dcgrp->subsys[ssid], css); | |
5df36032 | 1487 | ss->root = dst_root; |
1ada4838 | 1488 | css->cgroup = dcgrp; |
73e80ed8 | 1489 | |
82d6489d | 1490 | spin_lock_irq(&css_set_lock); |
2d8f243a TH |
1491 | hash_for_each(css_set_table, i, cset, hlist) |
1492 | list_move_tail(&cset->e_cset_node[ss->id], | |
1ada4838 | 1493 | &dcgrp->e_csets[ss->id]); |
82d6489d | 1494 | spin_unlock_irq(&css_set_lock); |
2d8f243a | 1495 | |
bd53d617 | 1496 | /* default hierarchy doesn't enable controllers by default */ |
f392e51c | 1497 | dst_root->subsys_mask |= 1 << ssid; |
49d1dc4b TH |
1498 | if (dst_root == &cgrp_dfl_root) { |
1499 | static_branch_enable(cgroup_subsys_on_dfl_key[ssid]); | |
1500 | } else { | |
1ada4838 | 1501 | dcgrp->subtree_control |= 1 << ssid; |
49d1dc4b | 1502 | static_branch_disable(cgroup_subsys_on_dfl_key[ssid]); |
667c2491 | 1503 | } |
a8a648c4 | 1504 | |
334c3679 TH |
1505 | ret = cgroup_apply_control(dcgrp); |
1506 | if (ret) | |
1507 | pr_warn("partial failure to rebind %s controller (err=%d)\n", | |
1508 | ss->name, ret); | |
1509 | ||
5df36032 TH |
1510 | if (ss->bind) |
1511 | ss->bind(css); | |
b4e0eeaf | 1512 | } while_each_subsys_mask(); |
ddbcc7e8 | 1513 | |
1ada4838 | 1514 | kernfs_activate(dcgrp->kn); |
ddbcc7e8 PM |
1515 | return 0; |
1516 | } | |
1517 | ||
1592c9b2 TH |
1518 | int cgroup_show_path(struct seq_file *sf, struct kernfs_node *kf_node, |
1519 | struct kernfs_root *kf_root) | |
4f41fc59 | 1520 | { |
09be4c82 | 1521 | int len = 0; |
4f41fc59 SH |
1522 | char *buf = NULL; |
1523 | struct cgroup_root *kf_cgroot = cgroup_root_from_kf(kf_root); | |
1524 | struct cgroup *ns_cgroup; | |
1525 | ||
1526 | buf = kmalloc(PATH_MAX, GFP_KERNEL); | |
1527 | if (!buf) | |
1528 | return -ENOMEM; | |
1529 | ||
82d6489d | 1530 | spin_lock_irq(&css_set_lock); |
4f41fc59 SH |
1531 | ns_cgroup = current_cgns_cgroup_from_root(kf_cgroot); |
1532 | len = kernfs_path_from_node(kf_node, ns_cgroup->kn, buf, PATH_MAX); | |
82d6489d | 1533 | spin_unlock_irq(&css_set_lock); |
4f41fc59 SH |
1534 | |
1535 | if (len >= PATH_MAX) | |
1536 | len = -ERANGE; | |
1537 | else if (len > 0) { | |
1538 | seq_escape(sf, buf, " \t\n\\"); | |
1539 | len = 0; | |
1540 | } | |
1541 | kfree(buf); | |
1542 | return len; | |
1543 | } | |
1544 | ||
2bd59d48 | 1545 | static int cgroup_remount(struct kernfs_root *kf_root, int *flags, char *data) |
ddbcc7e8 | 1546 | { |
fa069904 TH |
1547 | pr_err("remount is not allowed\n"); |
1548 | return -EINVAL; | |
ddbcc7e8 PM |
1549 | } |
1550 | ||
afeb0f9f TH |
1551 | /* |
1552 | * To reduce the fork() overhead for systems that are not actually using | |
1553 | * their cgroups capability, we don't maintain the lists running through | |
1554 | * each css_set to its tasks until we see the list actually used - in other | |
1555 | * words after the first mount. | |
1556 | */ | |
1557 | static bool use_task_css_set_links __read_mostly; | |
1558 | ||
1559 | static void cgroup_enable_task_cg_lists(void) | |
1560 | { | |
1561 | struct task_struct *p, *g; | |
1562 | ||
82d6489d | 1563 | spin_lock_irq(&css_set_lock); |
afeb0f9f TH |
1564 | |
1565 | if (use_task_css_set_links) | |
1566 | goto out_unlock; | |
1567 | ||
1568 | use_task_css_set_links = true; | |
1569 | ||
1570 | /* | |
1571 | * We need tasklist_lock because RCU is not safe against | |
1572 | * while_each_thread(). Besides, a forking task that has passed | |
1573 | * cgroup_post_fork() without seeing use_task_css_set_links = 1 | |
1574 | * is not guaranteed to have its child immediately visible in the | |
1575 | * tasklist if we walk through it with RCU. | |
1576 | */ | |
1577 | read_lock(&tasklist_lock); | |
1578 | do_each_thread(g, p) { | |
afeb0f9f TH |
1579 | WARN_ON_ONCE(!list_empty(&p->cg_list) || |
1580 | task_css_set(p) != &init_css_set); | |
1581 | ||
1582 | /* | |
1583 | * We should check if the process is exiting, otherwise | |
1584 | * it will race with cgroup_exit() in that the list | |
1585 | * entry won't be deleted though the process has exited. | |
f153ad11 TH |
1586 | * Do it while holding siglock so that we don't end up |
1587 | * racing against cgroup_exit(). | |
82d6489d DBO |
1588 | * |
1589 | * Interrupts were already disabled while acquiring | |
1590 | * the css_set_lock, so we do not need to disable it | |
1591 | * again when acquiring the sighand->siglock here. | |
afeb0f9f | 1592 | */ |
82d6489d | 1593 | spin_lock(&p->sighand->siglock); |
eaf797ab TH |
1594 | if (!(p->flags & PF_EXITING)) { |
1595 | struct css_set *cset = task_css_set(p); | |
1596 | ||
0de0942d TH |
1597 | if (!css_set_populated(cset)) |
1598 | css_set_update_populated(cset, true); | |
389b9c1b | 1599 | list_add_tail(&p->cg_list, &cset->tasks); |
eaf797ab TH |
1600 | get_css_set(cset); |
1601 | } | |
82d6489d | 1602 | spin_unlock(&p->sighand->siglock); |
afeb0f9f TH |
1603 | } while_each_thread(g, p); |
1604 | read_unlock(&tasklist_lock); | |
1605 | out_unlock: | |
82d6489d | 1606 | spin_unlock_irq(&css_set_lock); |
afeb0f9f | 1607 | } |
ddbcc7e8 | 1608 | |
cc31edce PM |
1609 | static void init_cgroup_housekeeping(struct cgroup *cgrp) |
1610 | { | |
2d8f243a TH |
1611 | struct cgroup_subsys *ss; |
1612 | int ssid; | |
1613 | ||
d5c419b6 TH |
1614 | INIT_LIST_HEAD(&cgrp->self.sibling); |
1615 | INIT_LIST_HEAD(&cgrp->self.children); | |
69d0206c | 1616 | INIT_LIST_HEAD(&cgrp->cset_links); |
72a8cb30 BB |
1617 | INIT_LIST_HEAD(&cgrp->pidlists); |
1618 | mutex_init(&cgrp->pidlist_mutex); | |
9d800df1 | 1619 | cgrp->self.cgroup = cgrp; |
184faf32 | 1620 | cgrp->self.flags |= CSS_ONLINE; |
2d8f243a TH |
1621 | |
1622 | for_each_subsys(ss, ssid) | |
1623 | INIT_LIST_HEAD(&cgrp->e_csets[ssid]); | |
f8f22e53 TH |
1624 | |
1625 | init_waitqueue_head(&cgrp->offline_waitq); | |
d62beb7f | 1626 | INIT_WORK(&cgrp->release_agent_work, cgroup1_release_agent); |
cc31edce | 1627 | } |
c6d57f33 | 1628 | |
1592c9b2 | 1629 | void init_cgroup_root(struct cgroup_root *root, struct cgroup_sb_opts *opts) |
ddbcc7e8 | 1630 | { |
3dd06ffa | 1631 | struct cgroup *cgrp = &root->cgrp; |
b0ca5a84 | 1632 | |
ddbcc7e8 | 1633 | INIT_LIST_HEAD(&root->root_list); |
3c9c825b | 1634 | atomic_set(&root->nr_cgrps, 1); |
bd89aabc | 1635 | cgrp->root = root; |
cc31edce | 1636 | init_cgroup_housekeeping(cgrp); |
4e96ee8e | 1637 | idr_init(&root->cgroup_idr); |
c6d57f33 | 1638 | |
c6d57f33 PM |
1639 | root->flags = opts->flags; |
1640 | if (opts->release_agent) | |
1641 | strcpy(root->release_agent_path, opts->release_agent); | |
1642 | if (opts->name) | |
1643 | strcpy(root->name, opts->name); | |
2260e7fc | 1644 | if (opts->cpuset_clone_children) |
3dd06ffa | 1645 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags); |
c6d57f33 PM |
1646 | } |
1647 | ||
9732adc5 | 1648 | int cgroup_setup_root(struct cgroup_root *root, u16 ss_mask, int ref_flags) |
2c6ab6d2 | 1649 | { |
d427dfeb | 1650 | LIST_HEAD(tmp_links); |
3dd06ffa | 1651 | struct cgroup *root_cgrp = &root->cgrp; |
fa069904 | 1652 | struct kernfs_syscall_ops *kf_sops; |
d427dfeb | 1653 | struct css_set *cset; |
d427dfeb | 1654 | int i, ret; |
2c6ab6d2 | 1655 | |
d427dfeb | 1656 | lockdep_assert_held(&cgroup_mutex); |
c6d57f33 | 1657 | |
cf780b7d | 1658 | ret = cgroup_idr_alloc(&root->cgroup_idr, root_cgrp, 1, 2, GFP_KERNEL); |
d427dfeb | 1659 | if (ret < 0) |
2bd59d48 | 1660 | goto out; |
d427dfeb | 1661 | root_cgrp->id = ret; |
b11cfb58 | 1662 | root_cgrp->ancestor_ids[0] = ret; |
c6d57f33 | 1663 | |
9732adc5 ZL |
1664 | ret = percpu_ref_init(&root_cgrp->self.refcnt, css_release, |
1665 | ref_flags, GFP_KERNEL); | |
9d755d33 TH |
1666 | if (ret) |
1667 | goto out; | |
1668 | ||
d427dfeb | 1669 | /* |
f0d9a5f1 | 1670 | * We're accessing css_set_count without locking css_set_lock here, |
d427dfeb | 1671 | * but that's OK - it can only be increased by someone holding |
04313591 TH |
1672 | * cgroup_lock, and that's us. Later rebinding may disable |
1673 | * controllers on the default hierarchy and thus create new csets, | |
1674 | * which can't be more than the existing ones. Allocate 2x. | |
d427dfeb | 1675 | */ |
04313591 | 1676 | ret = allocate_cgrp_cset_links(2 * css_set_count, &tmp_links); |
d427dfeb | 1677 | if (ret) |
9d755d33 | 1678 | goto cancel_ref; |
ddbcc7e8 | 1679 | |
985ed670 | 1680 | ret = cgroup_init_root_id(root); |
ddbcc7e8 | 1681 | if (ret) |
9d755d33 | 1682 | goto cancel_ref; |
ddbcc7e8 | 1683 | |
fa069904 TH |
1684 | kf_sops = root == &cgrp_dfl_root ? |
1685 | &cgroup_kf_syscall_ops : &cgroup1_kf_syscall_ops; | |
1686 | ||
1687 | root->kf_root = kernfs_create_root(kf_sops, | |
2bd59d48 TH |
1688 | KERNFS_ROOT_CREATE_DEACTIVATED, |
1689 | root_cgrp); | |
1690 | if (IS_ERR(root->kf_root)) { | |
1691 | ret = PTR_ERR(root->kf_root); | |
1692 | goto exit_root_id; | |
1693 | } | |
1694 | root_cgrp->kn = root->kf_root->kn; | |
ddbcc7e8 | 1695 | |
334c3679 | 1696 | ret = css_populate_dir(&root_cgrp->self); |
d427dfeb | 1697 | if (ret) |
2bd59d48 | 1698 | goto destroy_root; |
ddbcc7e8 | 1699 | |
5df36032 | 1700 | ret = rebind_subsystems(root, ss_mask); |
d427dfeb | 1701 | if (ret) |
2bd59d48 | 1702 | goto destroy_root; |
ddbcc7e8 | 1703 | |
ed1777de TH |
1704 | trace_cgroup_setup_root(root); |
1705 | ||
d427dfeb TH |
1706 | /* |
1707 | * There must be no failure case after here, since rebinding takes | |
1708 | * care of subsystems' refcounts, which are explicitly dropped in | |
1709 | * the failure exit path. | |
1710 | */ | |
1711 | list_add(&root->root_list, &cgroup_roots); | |
1712 | cgroup_root_count++; | |
0df6a63f | 1713 | |
d427dfeb | 1714 | /* |
3dd06ffa | 1715 | * Link the root cgroup in this hierarchy into all the css_set |
d427dfeb TH |
1716 | * objects. |
1717 | */ | |
82d6489d | 1718 | spin_lock_irq(&css_set_lock); |
0de0942d | 1719 | hash_for_each(css_set_table, i, cset, hlist) { |
d427dfeb | 1720 | link_css_set(&tmp_links, cset, root_cgrp); |
0de0942d TH |
1721 | if (css_set_populated(cset)) |
1722 | cgroup_update_populated(root_cgrp, true); | |
1723 | } | |
82d6489d | 1724 | spin_unlock_irq(&css_set_lock); |
ddbcc7e8 | 1725 | |
d5c419b6 | 1726 | BUG_ON(!list_empty(&root_cgrp->self.children)); |
3c9c825b | 1727 | BUG_ON(atomic_read(&root->nr_cgrps) != 1); |
ddbcc7e8 | 1728 | |
2bd59d48 | 1729 | kernfs_activate(root_cgrp->kn); |
d427dfeb | 1730 | ret = 0; |
2bd59d48 | 1731 | goto out; |
d427dfeb | 1732 | |
2bd59d48 TH |
1733 | destroy_root: |
1734 | kernfs_destroy_root(root->kf_root); | |
1735 | root->kf_root = NULL; | |
1736 | exit_root_id: | |
d427dfeb | 1737 | cgroup_exit_root_id(root); |
9d755d33 | 1738 | cancel_ref: |
9a1049da | 1739 | percpu_ref_exit(&root_cgrp->self.refcnt); |
2bd59d48 | 1740 | out: |
d427dfeb TH |
1741 | free_cgrp_cset_links(&tmp_links); |
1742 | return ret; | |
ddbcc7e8 PM |
1743 | } |
1744 | ||
1592c9b2 TH |
1745 | struct dentry *cgroup_do_mount(struct file_system_type *fs_type, int flags, |
1746 | struct cgroup_root *root, unsigned long magic, | |
1747 | struct cgroup_namespace *ns) | |
ddbcc7e8 | 1748 | { |
2bd59d48 | 1749 | struct dentry *dentry; |
c6b3d5bc | 1750 | bool new_sb; |
ddbcc7e8 | 1751 | |
633feee3 | 1752 | dentry = kernfs_mount(fs_type, flags, root->kf_root, magic, &new_sb); |
ed82571b | 1753 | |
56fde9e0 | 1754 | /* |
633feee3 TH |
1755 | * In non-init cgroup namespace, instead of root cgroup's dentry, |
1756 | * we return the dentry corresponding to the cgroupns->root_cgrp. | |
56fde9e0 | 1757 | */ |
633feee3 TH |
1758 | if (!IS_ERR(dentry) && ns != &init_cgroup_ns) { |
1759 | struct dentry *nsdentry; | |
1760 | struct cgroup *cgrp; | |
e37a06f1 | 1761 | |
633feee3 TH |
1762 | mutex_lock(&cgroup_mutex); |
1763 | spin_lock_irq(&css_set_lock); | |
1764 | ||
1765 | cgrp = cset_cgroup_from_root(ns->root_cset, root); | |
1766 | ||
1767 | spin_unlock_irq(&css_set_lock); | |
1768 | mutex_unlock(&cgroup_mutex); | |
1769 | ||
1770 | nsdentry = kernfs_node_dentry(cgrp->kn, dentry->d_sb); | |
1771 | dput(dentry); | |
1772 | dentry = nsdentry; | |
67e9c74b TH |
1773 | } |
1774 | ||
633feee3 TH |
1775 | if (IS_ERR(dentry) || !new_sb) |
1776 | cgroup_put(&root->cgrp); | |
1777 | ||
1778 | return dentry; | |
1779 | } | |
1780 | ||
f7e83571 | 1781 | static struct dentry *cgroup_mount(struct file_system_type *fs_type, |
ddbcc7e8 | 1782 | int flags, const char *unused_dev_name, |
f7e83571 | 1783 | void *data) |
ddbcc7e8 | 1784 | { |
ed82571b | 1785 | struct cgroup_namespace *ns = current->nsproxy->cgroup_ns; |
2bd59d48 | 1786 | struct dentry *dentry; |
ddbcc7e8 | 1787 | |
ed82571b SH |
1788 | get_cgroup_ns(ns); |
1789 | ||
1790 | /* Check if the caller has permission to mount. */ | |
1791 | if (!ns_capable(ns->user_ns, CAP_SYS_ADMIN)) { | |
1792 | put_cgroup_ns(ns); | |
1793 | return ERR_PTR(-EPERM); | |
1794 | } | |
1795 | ||
56fde9e0 TH |
1796 | /* |
1797 | * The first time anyone tries to mount a cgroup, enable the list | |
1798 | * linking each css_set to its tasks and fix up all existing tasks. | |
1799 | */ | |
1800 | if (!use_task_css_set_links) | |
1801 | cgroup_enable_task_cg_lists(); | |
e37a06f1 | 1802 | |
633feee3 | 1803 | if (fs_type == &cgroup2_fs_type) { |
67e9c74b TH |
1804 | if (data) { |
1805 | pr_err("cgroup2: unknown option \"%s\"\n", (char *)data); | |
ed82571b | 1806 | put_cgroup_ns(ns); |
67e9c74b TH |
1807 | return ERR_PTR(-EINVAL); |
1808 | } | |
a7165264 | 1809 | cgrp_dfl_visible = true; |
a590b90d | 1810 | cgroup_get_live(&cgrp_dfl_root.cgrp); |
633feee3 TH |
1811 | |
1812 | dentry = cgroup_do_mount(&cgroup2_fs_type, flags, &cgrp_dfl_root, | |
1813 | CGROUP2_SUPER_MAGIC, ns); | |
1814 | } else { | |
1815 | dentry = cgroup1_mount(&cgroup_fs_type, flags, data, | |
1816 | CGROUP_SUPER_MAGIC, ns); | |
67e9c74b TH |
1817 | } |
1818 | ||
ed82571b | 1819 | put_cgroup_ns(ns); |
2bd59d48 TH |
1820 | return dentry; |
1821 | } | |
8e30e2b8 | 1822 | |
2bd59d48 TH |
1823 | static void cgroup_kill_sb(struct super_block *sb) |
1824 | { | |
1825 | struct kernfs_root *kf_root = kernfs_root_from_sb(sb); | |
3dd06ffa | 1826 | struct cgroup_root *root = cgroup_root_from_kf(kf_root); |
a015edd2 | 1827 | |
970317aa | 1828 | /* |
9d755d33 TH |
1829 | * If @root doesn't have any mounts or children, start killing it. |
1830 | * This prevents new mounts by disabling percpu_ref_tryget_live(). | |
1831 | * cgroup_mount() may wait for @root's release. | |
1f779fb2 LZ |
1832 | * |
1833 | * And don't kill the default root. | |
970317aa | 1834 | */ |
3c606d35 | 1835 | if (!list_empty(&root->cgrp.self.children) || |
1f779fb2 | 1836 | root == &cgrp_dfl_root) |
9d755d33 TH |
1837 | cgroup_put(&root->cgrp); |
1838 | else | |
1839 | percpu_ref_kill(&root->cgrp.self.refcnt); | |
970317aa | 1840 | |
2bd59d48 | 1841 | kernfs_kill_sb(sb); |
ddbcc7e8 | 1842 | } |
970317aa | 1843 | |
0a268dbd | 1844 | struct file_system_type cgroup_fs_type = { |
ddbcc7e8 | 1845 | .name = "cgroup", |
f7e83571 | 1846 | .mount = cgroup_mount, |
ddbcc7e8 | 1847 | .kill_sb = cgroup_kill_sb, |
1c53753e | 1848 | .fs_flags = FS_USERNS_MOUNT, |
ddbcc7e8 | 1849 | }; |
3126121f | 1850 | |
67e9c74b TH |
1851 | static struct file_system_type cgroup2_fs_type = { |
1852 | .name = "cgroup2", | |
1853 | .mount = cgroup_mount, | |
1854 | .kill_sb = cgroup_kill_sb, | |
1c53753e | 1855 | .fs_flags = FS_USERNS_MOUNT, |
67e9c74b | 1856 | }; |
3126121f | 1857 | |
0a268dbd TH |
1858 | int cgroup_path_ns_locked(struct cgroup *cgrp, char *buf, size_t buflen, |
1859 | struct cgroup_namespace *ns) | |
a79a908f AK |
1860 | { |
1861 | struct cgroup *root = cset_cgroup_from_root(ns->root_cset, cgrp->root); | |
a79a908f | 1862 | |
4c737b41 | 1863 | return kernfs_path_from_node(cgrp->kn, root->kn, buf, buflen); |
a79a908f AK |
1864 | } |
1865 | ||
4c737b41 TH |
1866 | int cgroup_path_ns(struct cgroup *cgrp, char *buf, size_t buflen, |
1867 | struct cgroup_namespace *ns) | |
a79a908f | 1868 | { |
4c737b41 | 1869 | int ret; |
a79a908f AK |
1870 | |
1871 | mutex_lock(&cgroup_mutex); | |
82d6489d | 1872 | spin_lock_irq(&css_set_lock); |
a79a908f AK |
1873 | |
1874 | ret = cgroup_path_ns_locked(cgrp, buf, buflen, ns); | |
1875 | ||
82d6489d | 1876 | spin_unlock_irq(&css_set_lock); |
a79a908f AK |
1877 | mutex_unlock(&cgroup_mutex); |
1878 | ||
1879 | return ret; | |
1880 | } | |
1881 | EXPORT_SYMBOL_GPL(cgroup_path_ns); | |
1882 | ||
857a2beb | 1883 | /** |
913ffdb5 | 1884 | * task_cgroup_path - cgroup path of a task in the first cgroup hierarchy |
857a2beb | 1885 | * @task: target task |
857a2beb TH |
1886 | * @buf: the buffer to write the path into |
1887 | * @buflen: the length of the buffer | |
1888 | * | |
913ffdb5 TH |
1889 | * Determine @task's cgroup on the first (the one with the lowest non-zero |
1890 | * hierarchy_id) cgroup hierarchy and copy its path into @buf. This | |
1891 | * function grabs cgroup_mutex and shouldn't be used inside locks used by | |
1892 | * cgroup controller callbacks. | |
1893 | * | |
e61734c5 | 1894 | * Return value is the same as kernfs_path(). |
857a2beb | 1895 | */ |
4c737b41 | 1896 | int task_cgroup_path(struct task_struct *task, char *buf, size_t buflen) |
857a2beb | 1897 | { |
3dd06ffa | 1898 | struct cgroup_root *root; |
913ffdb5 | 1899 | struct cgroup *cgrp; |
e61734c5 | 1900 | int hierarchy_id = 1; |
4c737b41 | 1901 | int ret; |
857a2beb TH |
1902 | |
1903 | mutex_lock(&cgroup_mutex); | |
82d6489d | 1904 | spin_lock_irq(&css_set_lock); |
857a2beb | 1905 | |
913ffdb5 TH |
1906 | root = idr_get_next(&cgroup_hierarchy_idr, &hierarchy_id); |
1907 | ||
857a2beb TH |
1908 | if (root) { |
1909 | cgrp = task_cgroup_from_root(task, root); | |
4c737b41 | 1910 | ret = cgroup_path_ns_locked(cgrp, buf, buflen, &init_cgroup_ns); |
913ffdb5 TH |
1911 | } else { |
1912 | /* if no hierarchy exists, everyone is in "/" */ | |
4c737b41 | 1913 | ret = strlcpy(buf, "/", buflen); |
857a2beb TH |
1914 | } |
1915 | ||
82d6489d | 1916 | spin_unlock_irq(&css_set_lock); |
857a2beb | 1917 | mutex_unlock(&cgroup_mutex); |
4c737b41 | 1918 | return ret; |
857a2beb | 1919 | } |
913ffdb5 | 1920 | EXPORT_SYMBOL_GPL(task_cgroup_path); |
857a2beb | 1921 | |
adaae5dc | 1922 | /** |
e595cd70 | 1923 | * cgroup_migrate_add_task - add a migration target task to a migration context |
adaae5dc | 1924 | * @task: target task |
e595cd70 | 1925 | * @mgctx: target migration context |
adaae5dc | 1926 | * |
e595cd70 TH |
1927 | * Add @task, which is a migration target, to @mgctx->tset. This function |
1928 | * becomes noop if @task doesn't need to be migrated. @task's css_set | |
1929 | * should have been added as a migration source and @task->cg_list will be | |
1930 | * moved from the css_set's tasks list to mg_tasks one. | |
adaae5dc | 1931 | */ |
e595cd70 TH |
1932 | static void cgroup_migrate_add_task(struct task_struct *task, |
1933 | struct cgroup_mgctx *mgctx) | |
adaae5dc TH |
1934 | { |
1935 | struct css_set *cset; | |
1936 | ||
f0d9a5f1 | 1937 | lockdep_assert_held(&css_set_lock); |
adaae5dc TH |
1938 | |
1939 | /* @task either already exited or can't exit until the end */ | |
1940 | if (task->flags & PF_EXITING) | |
1941 | return; | |
1942 | ||
1943 | /* leave @task alone if post_fork() hasn't linked it yet */ | |
1944 | if (list_empty(&task->cg_list)) | |
1945 | return; | |
1946 | ||
1947 | cset = task_css_set(task); | |
1948 | if (!cset->mg_src_cgrp) | |
1949 | return; | |
1950 | ||
1951 | list_move_tail(&task->cg_list, &cset->mg_tasks); | |
1952 | if (list_empty(&cset->mg_node)) | |
e595cd70 TH |
1953 | list_add_tail(&cset->mg_node, |
1954 | &mgctx->tset.src_csets); | |
adaae5dc | 1955 | if (list_empty(&cset->mg_dst_cset->mg_node)) |
d8ebf519 | 1956 | list_add_tail(&cset->mg_dst_cset->mg_node, |
e595cd70 | 1957 | &mgctx->tset.dst_csets); |
adaae5dc TH |
1958 | } |
1959 | ||
2f7ee569 TH |
1960 | /** |
1961 | * cgroup_taskset_first - reset taskset and return the first task | |
1962 | * @tset: taskset of interest | |
1f7dd3e5 | 1963 | * @dst_cssp: output variable for the destination css |
2f7ee569 TH |
1964 | * |
1965 | * @tset iteration is initialized and the first task is returned. | |
1966 | */ | |
1f7dd3e5 TH |
1967 | struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset, |
1968 | struct cgroup_subsys_state **dst_cssp) | |
2f7ee569 | 1969 | { |
b3dc094e TH |
1970 | tset->cur_cset = list_first_entry(tset->csets, struct css_set, mg_node); |
1971 | tset->cur_task = NULL; | |
1972 | ||
1f7dd3e5 | 1973 | return cgroup_taskset_next(tset, dst_cssp); |
2f7ee569 | 1974 | } |
2f7ee569 TH |
1975 | |
1976 | /** | |
1977 | * cgroup_taskset_next - iterate to the next task in taskset | |
1978 | * @tset: taskset of interest | |
1f7dd3e5 | 1979 | * @dst_cssp: output variable for the destination css |
2f7ee569 TH |
1980 | * |
1981 | * Return the next task in @tset. Iteration must have been initialized | |
1982 | * with cgroup_taskset_first(). | |
1983 | */ | |
1f7dd3e5 TH |
1984 | struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset, |
1985 | struct cgroup_subsys_state **dst_cssp) | |
2f7ee569 | 1986 | { |
b3dc094e TH |
1987 | struct css_set *cset = tset->cur_cset; |
1988 | struct task_struct *task = tset->cur_task; | |
2f7ee569 | 1989 | |
b3dc094e TH |
1990 | while (&cset->mg_node != tset->csets) { |
1991 | if (!task) | |
1992 | task = list_first_entry(&cset->mg_tasks, | |
1993 | struct task_struct, cg_list); | |
1994 | else | |
1995 | task = list_next_entry(task, cg_list); | |
2f7ee569 | 1996 | |
b3dc094e TH |
1997 | if (&task->cg_list != &cset->mg_tasks) { |
1998 | tset->cur_cset = cset; | |
1999 | tset->cur_task = task; | |
1f7dd3e5 TH |
2000 | |
2001 | /* | |
2002 | * This function may be called both before and | |
2003 | * after cgroup_taskset_migrate(). The two cases | |
2004 | * can be distinguished by looking at whether @cset | |
2005 | * has its ->mg_dst_cset set. | |
2006 | */ | |
2007 | if (cset->mg_dst_cset) | |
2008 | *dst_cssp = cset->mg_dst_cset->subsys[tset->ssid]; | |
2009 | else | |
2010 | *dst_cssp = cset->subsys[tset->ssid]; | |
2011 | ||
b3dc094e TH |
2012 | return task; |
2013 | } | |
2f7ee569 | 2014 | |
b3dc094e TH |
2015 | cset = list_next_entry(cset, mg_node); |
2016 | task = NULL; | |
2017 | } | |
2f7ee569 | 2018 | |
b3dc094e | 2019 | return NULL; |
2f7ee569 | 2020 | } |
2f7ee569 | 2021 | |
adaae5dc | 2022 | /** |
37ff9f8f | 2023 | * cgroup_taskset_migrate - migrate a taskset |
e595cd70 | 2024 | * @mgctx: migration context |
adaae5dc | 2025 | * |
e595cd70 | 2026 | * Migrate tasks in @mgctx as setup by migration preparation functions. |
37ff9f8f | 2027 | * This function fails iff one of the ->can_attach callbacks fails and |
e595cd70 TH |
2028 | * guarantees that either all or none of the tasks in @mgctx are migrated. |
2029 | * @mgctx is consumed regardless of success. | |
adaae5dc | 2030 | */ |
bfc2cf6f | 2031 | static int cgroup_migrate_execute(struct cgroup_mgctx *mgctx) |
adaae5dc | 2032 | { |
e595cd70 | 2033 | struct cgroup_taskset *tset = &mgctx->tset; |
37ff9f8f | 2034 | struct cgroup_subsys *ss; |
adaae5dc TH |
2035 | struct task_struct *task, *tmp_task; |
2036 | struct css_set *cset, *tmp_cset; | |
37ff9f8f | 2037 | int ssid, failed_ssid, ret; |
adaae5dc TH |
2038 | |
2039 | /* methods shouldn't be called if no task is actually migrating */ | |
2040 | if (list_empty(&tset->src_csets)) | |
2041 | return 0; | |
2042 | ||
2043 | /* check that we can legitimately attach to the cgroup */ | |
bfc2cf6f | 2044 | do_each_subsys_mask(ss, ssid, mgctx->ss_mask) { |
37ff9f8f TH |
2045 | if (ss->can_attach) { |
2046 | tset->ssid = ssid; | |
2047 | ret = ss->can_attach(tset); | |
adaae5dc | 2048 | if (ret) { |
37ff9f8f | 2049 | failed_ssid = ssid; |
adaae5dc TH |
2050 | goto out_cancel_attach; |
2051 | } | |
2052 | } | |
37ff9f8f | 2053 | } while_each_subsys_mask(); |
adaae5dc TH |
2054 | |
2055 | /* | |
2056 | * Now that we're guaranteed success, proceed to move all tasks to | |
2057 | * the new cgroup. There are no failure cases after here, so this | |
2058 | * is the commit point. | |
2059 | */ | |
82d6489d | 2060 | spin_lock_irq(&css_set_lock); |
adaae5dc | 2061 | list_for_each_entry(cset, &tset->src_csets, mg_node) { |
f6d7d049 TH |
2062 | list_for_each_entry_safe(task, tmp_task, &cset->mg_tasks, cg_list) { |
2063 | struct css_set *from_cset = task_css_set(task); | |
2064 | struct css_set *to_cset = cset->mg_dst_cset; | |
2065 | ||
2066 | get_css_set(to_cset); | |
2067 | css_set_move_task(task, from_cset, to_cset, true); | |
2068 | put_css_set_locked(from_cset); | |
2069 | } | |
adaae5dc | 2070 | } |
82d6489d | 2071 | spin_unlock_irq(&css_set_lock); |
adaae5dc TH |
2072 | |
2073 | /* | |
2074 | * Migration is committed, all target tasks are now on dst_csets. | |
2075 | * Nothing is sensitive to fork() after this point. Notify | |
2076 | * controllers that migration is complete. | |
2077 | */ | |
2078 | tset->csets = &tset->dst_csets; | |
2079 | ||
bfc2cf6f | 2080 | do_each_subsys_mask(ss, ssid, mgctx->ss_mask) { |
37ff9f8f TH |
2081 | if (ss->attach) { |
2082 | tset->ssid = ssid; | |
2083 | ss->attach(tset); | |
1f7dd3e5 | 2084 | } |
37ff9f8f | 2085 | } while_each_subsys_mask(); |
adaae5dc TH |
2086 | |
2087 | ret = 0; | |
2088 | goto out_release_tset; | |
2089 | ||
2090 | out_cancel_attach: | |
bfc2cf6f | 2091 | do_each_subsys_mask(ss, ssid, mgctx->ss_mask) { |
37ff9f8f | 2092 | if (ssid == failed_ssid) |
adaae5dc | 2093 | break; |
37ff9f8f TH |
2094 | if (ss->cancel_attach) { |
2095 | tset->ssid = ssid; | |
2096 | ss->cancel_attach(tset); | |
1f7dd3e5 | 2097 | } |
37ff9f8f | 2098 | } while_each_subsys_mask(); |
adaae5dc | 2099 | out_release_tset: |
82d6489d | 2100 | spin_lock_irq(&css_set_lock); |
adaae5dc TH |
2101 | list_splice_init(&tset->dst_csets, &tset->src_csets); |
2102 | list_for_each_entry_safe(cset, tmp_cset, &tset->src_csets, mg_node) { | |
2103 | list_splice_tail_init(&cset->mg_tasks, &cset->tasks); | |
2104 | list_del_init(&cset->mg_node); | |
2105 | } | |
82d6489d | 2106 | spin_unlock_irq(&css_set_lock); |
adaae5dc TH |
2107 | return ret; |
2108 | } | |
2109 | ||
6c694c88 TH |
2110 | /** |
2111 | * cgroup_may_migrate_to - verify whether a cgroup can be migration destination | |
2112 | * @dst_cgrp: destination cgroup to test | |
2113 | * | |
2114 | * On the default hierarchy, except for the root, subtree_control must be | |
2115 | * zero for migration destination cgroups with tasks so that child cgroups | |
2116 | * don't compete against tasks. | |
2117 | */ | |
0a268dbd | 2118 | bool cgroup_may_migrate_to(struct cgroup *dst_cgrp) |
6c694c88 TH |
2119 | { |
2120 | return !cgroup_on_dfl(dst_cgrp) || !cgroup_parent(dst_cgrp) || | |
2121 | !dst_cgrp->subtree_control; | |
2122 | } | |
2123 | ||
a043e3b2 | 2124 | /** |
1958d2d5 | 2125 | * cgroup_migrate_finish - cleanup after attach |
e595cd70 | 2126 | * @mgctx: migration context |
74a1166d | 2127 | * |
1958d2d5 TH |
2128 | * Undo cgroup_migrate_add_src() and cgroup_migrate_prepare_dst(). See |
2129 | * those functions for details. | |
74a1166d | 2130 | */ |
e595cd70 | 2131 | void cgroup_migrate_finish(struct cgroup_mgctx *mgctx) |
74a1166d | 2132 | { |
e595cd70 | 2133 | LIST_HEAD(preloaded); |
1958d2d5 | 2134 | struct css_set *cset, *tmp_cset; |
74a1166d | 2135 | |
1958d2d5 TH |
2136 | lockdep_assert_held(&cgroup_mutex); |
2137 | ||
82d6489d | 2138 | spin_lock_irq(&css_set_lock); |
e595cd70 TH |
2139 | |
2140 | list_splice_tail_init(&mgctx->preloaded_src_csets, &preloaded); | |
2141 | list_splice_tail_init(&mgctx->preloaded_dst_csets, &preloaded); | |
2142 | ||
2143 | list_for_each_entry_safe(cset, tmp_cset, &preloaded, mg_preload_node) { | |
1958d2d5 | 2144 | cset->mg_src_cgrp = NULL; |
e4857982 | 2145 | cset->mg_dst_cgrp = NULL; |
1958d2d5 TH |
2146 | cset->mg_dst_cset = NULL; |
2147 | list_del_init(&cset->mg_preload_node); | |
a25eb52e | 2148 | put_css_set_locked(cset); |
1958d2d5 | 2149 | } |
e595cd70 | 2150 | |
82d6489d | 2151 | spin_unlock_irq(&css_set_lock); |
1958d2d5 TH |
2152 | } |
2153 | ||
2154 | /** | |
2155 | * cgroup_migrate_add_src - add a migration source css_set | |
2156 | * @src_cset: the source css_set to add | |
2157 | * @dst_cgrp: the destination cgroup | |
e595cd70 | 2158 | * @mgctx: migration context |
1958d2d5 TH |
2159 | * |
2160 | * Tasks belonging to @src_cset are about to be migrated to @dst_cgrp. Pin | |
e595cd70 | 2161 | * @src_cset and add it to @mgctx->src_csets, which should later be cleaned |
1958d2d5 TH |
2162 | * up by cgroup_migrate_finish(). |
2163 | * | |
1ed13287 TH |
2164 | * This function may be called without holding cgroup_threadgroup_rwsem |
2165 | * even if the target is a process. Threads may be created and destroyed | |
2166 | * but as long as cgroup_mutex is not dropped, no new css_set can be put | |
2167 | * into play and the preloaded css_sets are guaranteed to cover all | |
2168 | * migrations. | |
1958d2d5 | 2169 | */ |
0a268dbd TH |
2170 | void cgroup_migrate_add_src(struct css_set *src_cset, |
2171 | struct cgroup *dst_cgrp, | |
e595cd70 | 2172 | struct cgroup_mgctx *mgctx) |
1958d2d5 TH |
2173 | { |
2174 | struct cgroup *src_cgrp; | |
2175 | ||
2176 | lockdep_assert_held(&cgroup_mutex); | |
f0d9a5f1 | 2177 | lockdep_assert_held(&css_set_lock); |
1958d2d5 | 2178 | |
2b021cbf TH |
2179 | /* |
2180 | * If ->dead, @src_set is associated with one or more dead cgroups | |
2181 | * and doesn't contain any migratable tasks. Ignore it early so | |
2182 | * that the rest of migration path doesn't get confused by it. | |
2183 | */ | |
2184 | if (src_cset->dead) | |
2185 | return; | |
2186 | ||
1958d2d5 TH |
2187 | src_cgrp = cset_cgroup_from_root(src_cset, dst_cgrp->root); |
2188 | ||
1958d2d5 TH |
2189 | if (!list_empty(&src_cset->mg_preload_node)) |
2190 | return; | |
2191 | ||
2192 | WARN_ON(src_cset->mg_src_cgrp); | |
e4857982 | 2193 | WARN_ON(src_cset->mg_dst_cgrp); |
1958d2d5 TH |
2194 | WARN_ON(!list_empty(&src_cset->mg_tasks)); |
2195 | WARN_ON(!list_empty(&src_cset->mg_node)); | |
2196 | ||
2197 | src_cset->mg_src_cgrp = src_cgrp; | |
e4857982 | 2198 | src_cset->mg_dst_cgrp = dst_cgrp; |
1958d2d5 | 2199 | get_css_set(src_cset); |
e595cd70 | 2200 | list_add_tail(&src_cset->mg_preload_node, &mgctx->preloaded_src_csets); |
1958d2d5 TH |
2201 | } |
2202 | ||
2203 | /** | |
2204 | * cgroup_migrate_prepare_dst - prepare destination css_sets for migration | |
e595cd70 | 2205 | * @mgctx: migration context |
1958d2d5 | 2206 | * |
e4857982 | 2207 | * Tasks are about to be moved and all the source css_sets have been |
e595cd70 TH |
2208 | * preloaded to @mgctx->preloaded_src_csets. This function looks up and |
2209 | * pins all destination css_sets, links each to its source, and append them | |
2210 | * to @mgctx->preloaded_dst_csets. | |
1958d2d5 TH |
2211 | * |
2212 | * This function must be called after cgroup_migrate_add_src() has been | |
2213 | * called on each migration source css_set. After migration is performed | |
2214 | * using cgroup_migrate(), cgroup_migrate_finish() must be called on | |
e595cd70 | 2215 | * @mgctx. |
1958d2d5 | 2216 | */ |
e595cd70 | 2217 | int cgroup_migrate_prepare_dst(struct cgroup_mgctx *mgctx) |
1958d2d5 | 2218 | { |
f817de98 | 2219 | struct css_set *src_cset, *tmp_cset; |
1958d2d5 TH |
2220 | |
2221 | lockdep_assert_held(&cgroup_mutex); | |
2222 | ||
2223 | /* look up the dst cset for each src cset and link it to src */ | |
e595cd70 TH |
2224 | list_for_each_entry_safe(src_cset, tmp_cset, &mgctx->preloaded_src_csets, |
2225 | mg_preload_node) { | |
1958d2d5 | 2226 | struct css_set *dst_cset; |
bfc2cf6f TH |
2227 | struct cgroup_subsys *ss; |
2228 | int ssid; | |
1958d2d5 | 2229 | |
e4857982 | 2230 | dst_cset = find_css_set(src_cset, src_cset->mg_dst_cgrp); |
1958d2d5 TH |
2231 | if (!dst_cset) |
2232 | goto err; | |
2233 | ||
2234 | WARN_ON_ONCE(src_cset->mg_dst_cset || dst_cset->mg_dst_cset); | |
f817de98 TH |
2235 | |
2236 | /* | |
2237 | * If src cset equals dst, it's noop. Drop the src. | |
2238 | * cgroup_migrate() will skip the cset too. Note that we | |
2239 | * can't handle src == dst as some nodes are used by both. | |
2240 | */ | |
2241 | if (src_cset == dst_cset) { | |
2242 | src_cset->mg_src_cgrp = NULL; | |
e4857982 | 2243 | src_cset->mg_dst_cgrp = NULL; |
f817de98 | 2244 | list_del_init(&src_cset->mg_preload_node); |
a25eb52e ZL |
2245 | put_css_set(src_cset); |
2246 | put_css_set(dst_cset); | |
f817de98 TH |
2247 | continue; |
2248 | } | |
2249 | ||
1958d2d5 TH |
2250 | src_cset->mg_dst_cset = dst_cset; |
2251 | ||
2252 | if (list_empty(&dst_cset->mg_preload_node)) | |
e595cd70 TH |
2253 | list_add_tail(&dst_cset->mg_preload_node, |
2254 | &mgctx->preloaded_dst_csets); | |
1958d2d5 | 2255 | else |
a25eb52e | 2256 | put_css_set(dst_cset); |
bfc2cf6f TH |
2257 | |
2258 | for_each_subsys(ss, ssid) | |
2259 | if (src_cset->subsys[ssid] != dst_cset->subsys[ssid]) | |
2260 | mgctx->ss_mask |= 1 << ssid; | |
1958d2d5 TH |
2261 | } |
2262 | ||
1958d2d5 TH |
2263 | return 0; |
2264 | err: | |
e595cd70 | 2265 | cgroup_migrate_finish(mgctx); |
1958d2d5 TH |
2266 | return -ENOMEM; |
2267 | } | |
2268 | ||
2269 | /** | |
2270 | * cgroup_migrate - migrate a process or task to a cgroup | |
1958d2d5 TH |
2271 | * @leader: the leader of the process or the task to migrate |
2272 | * @threadgroup: whether @leader points to the whole process or a single task | |
e595cd70 | 2273 | * @mgctx: migration context |
1958d2d5 | 2274 | * |
37ff9f8f TH |
2275 | * Migrate a process or task denoted by @leader. If migrating a process, |
2276 | * the caller must be holding cgroup_threadgroup_rwsem. The caller is also | |
2277 | * responsible for invoking cgroup_migrate_add_src() and | |
1958d2d5 TH |
2278 | * cgroup_migrate_prepare_dst() on the targets before invoking this |
2279 | * function and following up with cgroup_migrate_finish(). | |
2280 | * | |
2281 | * As long as a controller's ->can_attach() doesn't fail, this function is | |
2282 | * guaranteed to succeed. This means that, excluding ->can_attach() | |
2283 | * failure, when migrating multiple targets, the success or failure can be | |
2284 | * decided for all targets by invoking group_migrate_prepare_dst() before | |
2285 | * actually starting migrating. | |
2286 | */ | |
0a268dbd | 2287 | int cgroup_migrate(struct task_struct *leader, bool threadgroup, |
bfc2cf6f | 2288 | struct cgroup_mgctx *mgctx) |
74a1166d | 2289 | { |
adaae5dc | 2290 | struct task_struct *task; |
74a1166d | 2291 | |
fb5d2b4c MSB |
2292 | /* |
2293 | * Prevent freeing of tasks while we take a snapshot. Tasks that are | |
2294 | * already PF_EXITING could be freed from underneath us unless we | |
2295 | * take an rcu_read_lock. | |
2296 | */ | |
82d6489d | 2297 | spin_lock_irq(&css_set_lock); |
fb5d2b4c | 2298 | rcu_read_lock(); |
9db8de37 | 2299 | task = leader; |
74a1166d | 2300 | do { |
e595cd70 | 2301 | cgroup_migrate_add_task(task, mgctx); |
081aa458 LZ |
2302 | if (!threadgroup) |
2303 | break; | |
9db8de37 | 2304 | } while_each_thread(leader, task); |
fb5d2b4c | 2305 | rcu_read_unlock(); |
82d6489d | 2306 | spin_unlock_irq(&css_set_lock); |
74a1166d | 2307 | |
bfc2cf6f | 2308 | return cgroup_migrate_execute(mgctx); |
74a1166d BB |
2309 | } |
2310 | ||
1958d2d5 TH |
2311 | /** |
2312 | * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup | |
2313 | * @dst_cgrp: the cgroup to attach to | |
2314 | * @leader: the task or the leader of the threadgroup to be attached | |
2315 | * @threadgroup: attach the whole threadgroup? | |
2316 | * | |
1ed13287 | 2317 | * Call holding cgroup_mutex and cgroup_threadgroup_rwsem. |
1958d2d5 | 2318 | */ |
0a268dbd TH |
2319 | int cgroup_attach_task(struct cgroup *dst_cgrp, struct task_struct *leader, |
2320 | bool threadgroup) | |
1958d2d5 | 2321 | { |
e595cd70 | 2322 | DEFINE_CGROUP_MGCTX(mgctx); |
1958d2d5 TH |
2323 | struct task_struct *task; |
2324 | int ret; | |
2325 | ||
6c694c88 TH |
2326 | if (!cgroup_may_migrate_to(dst_cgrp)) |
2327 | return -EBUSY; | |
2328 | ||
1958d2d5 | 2329 | /* look up all src csets */ |
82d6489d | 2330 | spin_lock_irq(&css_set_lock); |
1958d2d5 TH |
2331 | rcu_read_lock(); |
2332 | task = leader; | |
2333 | do { | |
e595cd70 | 2334 | cgroup_migrate_add_src(task_css_set(task), dst_cgrp, &mgctx); |
1958d2d5 TH |
2335 | if (!threadgroup) |
2336 | break; | |
2337 | } while_each_thread(leader, task); | |
2338 | rcu_read_unlock(); | |
82d6489d | 2339 | spin_unlock_irq(&css_set_lock); |
1958d2d5 TH |
2340 | |
2341 | /* prepare dst csets and commit */ | |
e595cd70 | 2342 | ret = cgroup_migrate_prepare_dst(&mgctx); |
1958d2d5 | 2343 | if (!ret) |
bfc2cf6f | 2344 | ret = cgroup_migrate(leader, threadgroup, &mgctx); |
1958d2d5 | 2345 | |
e595cd70 | 2346 | cgroup_migrate_finish(&mgctx); |
ed1777de TH |
2347 | |
2348 | if (!ret) | |
2349 | trace_cgroup_attach_task(dst_cgrp, leader, threadgroup); | |
2350 | ||
1958d2d5 | 2351 | return ret; |
74a1166d BB |
2352 | } |
2353 | ||
187fe840 TH |
2354 | static int cgroup_procs_write_permission(struct task_struct *task, |
2355 | struct cgroup *dst_cgrp, | |
2356 | struct kernfs_open_file *of) | |
dedf22e9 | 2357 | { |
dedf22e9 TH |
2358 | int ret = 0; |
2359 | ||
576dd464 | 2360 | if (cgroup_on_dfl(dst_cgrp)) { |
187fe840 TH |
2361 | struct super_block *sb = of->file->f_path.dentry->d_sb; |
2362 | struct cgroup *cgrp; | |
2363 | struct inode *inode; | |
2364 | ||
82d6489d | 2365 | spin_lock_irq(&css_set_lock); |
187fe840 | 2366 | cgrp = task_cgroup_from_root(task, &cgrp_dfl_root); |
82d6489d | 2367 | spin_unlock_irq(&css_set_lock); |
187fe840 TH |
2368 | |
2369 | while (!cgroup_is_descendant(dst_cgrp, cgrp)) | |
2370 | cgrp = cgroup_parent(cgrp); | |
2371 | ||
2372 | ret = -ENOMEM; | |
6f60eade | 2373 | inode = kernfs_get_inode(sb, cgrp->procs_file.kn); |
187fe840 TH |
2374 | if (inode) { |
2375 | ret = inode_permission(inode, MAY_WRITE); | |
2376 | iput(inode); | |
2377 | } | |
576dd464 TH |
2378 | } else { |
2379 | const struct cred *cred = current_cred(); | |
2380 | const struct cred *tcred = get_task_cred(task); | |
2381 | ||
2382 | /* | |
2383 | * even if we're attaching all tasks in the thread group, | |
2384 | * we only need to check permissions on one of them. | |
2385 | */ | |
2386 | if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) && | |
2387 | !uid_eq(cred->euid, tcred->uid) && | |
2388 | !uid_eq(cred->euid, tcred->suid)) | |
2389 | ret = -EACCES; | |
2390 | put_cred(tcred); | |
187fe840 TH |
2391 | } |
2392 | ||
dedf22e9 TH |
2393 | return ret; |
2394 | } | |
2395 | ||
74a1166d BB |
2396 | /* |
2397 | * Find the task_struct of the task to attach by vpid and pass it along to the | |
cd3d0952 | 2398 | * function to attach either it or all tasks in its threadgroup. Will lock |
0e1d768f | 2399 | * cgroup_mutex and threadgroup. |
bbcb81d0 | 2400 | */ |
0a268dbd TH |
2401 | ssize_t __cgroup_procs_write(struct kernfs_open_file *of, char *buf, |
2402 | size_t nbytes, loff_t off, bool threadgroup) | |
bbcb81d0 | 2403 | { |
bbcb81d0 | 2404 | struct task_struct *tsk; |
5cf1cacb | 2405 | struct cgroup_subsys *ss; |
e76ecaee | 2406 | struct cgroup *cgrp; |
acbef755 | 2407 | pid_t pid; |
5cf1cacb | 2408 | int ssid, ret; |
bbcb81d0 | 2409 | |
acbef755 TH |
2410 | if (kstrtoint(strstrip(buf), 0, &pid) || pid < 0) |
2411 | return -EINVAL; | |
2412 | ||
945ba199 | 2413 | cgrp = cgroup_kn_lock_live(of->kn, false); |
e76ecaee | 2414 | if (!cgrp) |
74a1166d BB |
2415 | return -ENODEV; |
2416 | ||
3014dde7 | 2417 | percpu_down_write(&cgroup_threadgroup_rwsem); |
b78949eb | 2418 | rcu_read_lock(); |
bbcb81d0 | 2419 | if (pid) { |
73507f33 | 2420 | tsk = find_task_by_vpid(pid); |
74a1166d | 2421 | if (!tsk) { |
dd4b0a46 | 2422 | ret = -ESRCH; |
3014dde7 | 2423 | goto out_unlock_rcu; |
bbcb81d0 | 2424 | } |
dedf22e9 | 2425 | } else { |
b78949eb | 2426 | tsk = current; |
dedf22e9 | 2427 | } |
cd3d0952 TH |
2428 | |
2429 | if (threadgroup) | |
b78949eb | 2430 | tsk = tsk->group_leader; |
c4c27fbd MG |
2431 | |
2432 | /* | |
77f88796 TH |
2433 | * kthreads may acquire PF_NO_SETAFFINITY during initialization. |
2434 | * If userland migrates such a kthread to a non-root cgroup, it can | |
2435 | * become trapped in a cpuset, or RT kthread may be born in a | |
2436 | * cgroup with no rt_runtime allocated. Just say no. | |
c4c27fbd | 2437 | */ |
77f88796 | 2438 | if (tsk->no_cgroup_migration || (tsk->flags & PF_NO_SETAFFINITY)) { |
c4c27fbd | 2439 | ret = -EINVAL; |
3014dde7 | 2440 | goto out_unlock_rcu; |
c4c27fbd MG |
2441 | } |
2442 | ||
b78949eb MSB |
2443 | get_task_struct(tsk); |
2444 | rcu_read_unlock(); | |
2445 | ||
187fe840 | 2446 | ret = cgroup_procs_write_permission(tsk, cgrp, of); |
dedf22e9 TH |
2447 | if (!ret) |
2448 | ret = cgroup_attach_task(cgrp, tsk, threadgroup); | |
081aa458 | 2449 | |
f9f9e7b7 | 2450 | put_task_struct(tsk); |
3014dde7 TH |
2451 | goto out_unlock_threadgroup; |
2452 | ||
2453 | out_unlock_rcu: | |
2454 | rcu_read_unlock(); | |
2455 | out_unlock_threadgroup: | |
2456 | percpu_up_write(&cgroup_threadgroup_rwsem); | |
5cf1cacb TH |
2457 | for_each_subsys(ss, ssid) |
2458 | if (ss->post_attach) | |
2459 | ss->post_attach(); | |
e76ecaee | 2460 | cgroup_kn_unlock(of->kn); |
acbef755 | 2461 | return ret ?: nbytes; |
bbcb81d0 PM |
2462 | } |
2463 | ||
0a268dbd TH |
2464 | ssize_t cgroup_procs_write(struct kernfs_open_file *of, char *buf, size_t nbytes, |
2465 | loff_t off) | |
af351026 | 2466 | { |
acbef755 | 2467 | return __cgroup_procs_write(of, buf, nbytes, off, true); |
af351026 PM |
2468 | } |
2469 | ||
6e5c8307 | 2470 | static void cgroup_print_ss_mask(struct seq_file *seq, u16 ss_mask) |
355e0c48 | 2471 | { |
f8f22e53 TH |
2472 | struct cgroup_subsys *ss; |
2473 | bool printed = false; | |
2474 | int ssid; | |
a742c59d | 2475 | |
b4e0eeaf | 2476 | do_each_subsys_mask(ss, ssid, ss_mask) { |
a966a4ed AS |
2477 | if (printed) |
2478 | seq_putc(seq, ' '); | |
2479 | seq_printf(seq, "%s", ss->name); | |
2480 | printed = true; | |
b4e0eeaf | 2481 | } while_each_subsys_mask(); |
f8f22e53 TH |
2482 | if (printed) |
2483 | seq_putc(seq, '\n'); | |
355e0c48 PM |
2484 | } |
2485 | ||
f8f22e53 TH |
2486 | /* show controllers which are enabled from the parent */ |
2487 | static int cgroup_controllers_show(struct seq_file *seq, void *v) | |
ddbcc7e8 | 2488 | { |
f8f22e53 TH |
2489 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
2490 | ||
5531dc91 | 2491 | cgroup_print_ss_mask(seq, cgroup_control(cgrp)); |
f8f22e53 | 2492 | return 0; |
ddbcc7e8 PM |
2493 | } |
2494 | ||
f8f22e53 TH |
2495 | /* show controllers which are enabled for a given cgroup's children */ |
2496 | static int cgroup_subtree_control_show(struct seq_file *seq, void *v) | |
ddbcc7e8 | 2497 | { |
f8f22e53 TH |
2498 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
2499 | ||
667c2491 | 2500 | cgroup_print_ss_mask(seq, cgrp->subtree_control); |
f8f22e53 TH |
2501 | return 0; |
2502 | } | |
2503 | ||
2504 | /** | |
2505 | * cgroup_update_dfl_csses - update css assoc of a subtree in default hierarchy | |
2506 | * @cgrp: root of the subtree to update csses for | |
2507 | * | |
54962604 TH |
2508 | * @cgrp's control masks have changed and its subtree's css associations |
2509 | * need to be updated accordingly. This function looks up all css_sets | |
2510 | * which are attached to the subtree, creates the matching updated css_sets | |
2511 | * and migrates the tasks to the new ones. | |
f8f22e53 TH |
2512 | */ |
2513 | static int cgroup_update_dfl_csses(struct cgroup *cgrp) | |
2514 | { | |
e595cd70 | 2515 | DEFINE_CGROUP_MGCTX(mgctx); |
54962604 TH |
2516 | struct cgroup_subsys_state *d_css; |
2517 | struct cgroup *dsct; | |
f8f22e53 TH |
2518 | struct css_set *src_cset; |
2519 | int ret; | |
2520 | ||
f8f22e53 TH |
2521 | lockdep_assert_held(&cgroup_mutex); |
2522 | ||
3014dde7 TH |
2523 | percpu_down_write(&cgroup_threadgroup_rwsem); |
2524 | ||
f8f22e53 | 2525 | /* look up all csses currently attached to @cgrp's subtree */ |
82d6489d | 2526 | spin_lock_irq(&css_set_lock); |
54962604 | 2527 | cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) { |
f8f22e53 TH |
2528 | struct cgrp_cset_link *link; |
2529 | ||
54962604 | 2530 | list_for_each_entry(link, &dsct->cset_links, cset_link) |
e595cd70 | 2531 | cgroup_migrate_add_src(link->cset, dsct, &mgctx); |
f8f22e53 | 2532 | } |
82d6489d | 2533 | spin_unlock_irq(&css_set_lock); |
f8f22e53 TH |
2534 | |
2535 | /* NULL dst indicates self on default hierarchy */ | |
e595cd70 | 2536 | ret = cgroup_migrate_prepare_dst(&mgctx); |
f8f22e53 TH |
2537 | if (ret) |
2538 | goto out_finish; | |
2539 | ||
82d6489d | 2540 | spin_lock_irq(&css_set_lock); |
e595cd70 | 2541 | list_for_each_entry(src_cset, &mgctx.preloaded_src_csets, mg_preload_node) { |
10265075 | 2542 | struct task_struct *task, *ntask; |
f8f22e53 | 2543 | |
10265075 TH |
2544 | /* all tasks in src_csets need to be migrated */ |
2545 | list_for_each_entry_safe(task, ntask, &src_cset->tasks, cg_list) | |
e595cd70 | 2546 | cgroup_migrate_add_task(task, &mgctx); |
f8f22e53 | 2547 | } |
82d6489d | 2548 | spin_unlock_irq(&css_set_lock); |
f8f22e53 | 2549 | |
bfc2cf6f | 2550 | ret = cgroup_migrate_execute(&mgctx); |
f8f22e53 | 2551 | out_finish: |
e595cd70 | 2552 | cgroup_migrate_finish(&mgctx); |
3014dde7 | 2553 | percpu_up_write(&cgroup_threadgroup_rwsem); |
f8f22e53 TH |
2554 | return ret; |
2555 | } | |
2556 | ||
1b9b96a1 | 2557 | /** |
945ba199 | 2558 | * cgroup_lock_and_drain_offline - lock cgroup_mutex and drain offlined csses |
ce3f1d9d | 2559 | * @cgrp: root of the target subtree |
1b9b96a1 TH |
2560 | * |
2561 | * Because css offlining is asynchronous, userland may try to re-enable a | |
945ba199 TH |
2562 | * controller while the previous css is still around. This function grabs |
2563 | * cgroup_mutex and drains the previous css instances of @cgrp's subtree. | |
1b9b96a1 | 2564 | */ |
0a268dbd | 2565 | void cgroup_lock_and_drain_offline(struct cgroup *cgrp) |
945ba199 | 2566 | __acquires(&cgroup_mutex) |
1b9b96a1 TH |
2567 | { |
2568 | struct cgroup *dsct; | |
ce3f1d9d | 2569 | struct cgroup_subsys_state *d_css; |
1b9b96a1 TH |
2570 | struct cgroup_subsys *ss; |
2571 | int ssid; | |
2572 | ||
945ba199 TH |
2573 | restart: |
2574 | mutex_lock(&cgroup_mutex); | |
1b9b96a1 | 2575 | |
ce3f1d9d | 2576 | cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) { |
1b9b96a1 TH |
2577 | for_each_subsys(ss, ssid) { |
2578 | struct cgroup_subsys_state *css = cgroup_css(dsct, ss); | |
2579 | DEFINE_WAIT(wait); | |
2580 | ||
ce3f1d9d | 2581 | if (!css || !percpu_ref_is_dying(&css->refcnt)) |
1b9b96a1 TH |
2582 | continue; |
2583 | ||
a590b90d | 2584 | cgroup_get_live(dsct); |
1b9b96a1 TH |
2585 | prepare_to_wait(&dsct->offline_waitq, &wait, |
2586 | TASK_UNINTERRUPTIBLE); | |
2587 | ||
2588 | mutex_unlock(&cgroup_mutex); | |
2589 | schedule(); | |
2590 | finish_wait(&dsct->offline_waitq, &wait); | |
1b9b96a1 TH |
2591 | |
2592 | cgroup_put(dsct); | |
945ba199 | 2593 | goto restart; |
1b9b96a1 TH |
2594 | } |
2595 | } | |
1b9b96a1 TH |
2596 | } |
2597 | ||
15a27c36 TH |
2598 | /** |
2599 | * cgroup_save_control - save control masks of a subtree | |
2600 | * @cgrp: root of the target subtree | |
2601 | * | |
2602 | * Save ->subtree_control and ->subtree_ss_mask to the respective old_ | |
2603 | * prefixed fields for @cgrp's subtree including @cgrp itself. | |
2604 | */ | |
2605 | static void cgroup_save_control(struct cgroup *cgrp) | |
2606 | { | |
2607 | struct cgroup *dsct; | |
2608 | struct cgroup_subsys_state *d_css; | |
2609 | ||
2610 | cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) { | |
2611 | dsct->old_subtree_control = dsct->subtree_control; | |
2612 | dsct->old_subtree_ss_mask = dsct->subtree_ss_mask; | |
2613 | } | |
2614 | } | |
2615 | ||
2616 | /** | |
2617 | * cgroup_propagate_control - refresh control masks of a subtree | |
2618 | * @cgrp: root of the target subtree | |
2619 | * | |
2620 | * For @cgrp and its subtree, ensure ->subtree_ss_mask matches | |
2621 | * ->subtree_control and propagate controller availability through the | |
2622 | * subtree so that descendants don't have unavailable controllers enabled. | |
2623 | */ | |
2624 | static void cgroup_propagate_control(struct cgroup *cgrp) | |
2625 | { | |
2626 | struct cgroup *dsct; | |
2627 | struct cgroup_subsys_state *d_css; | |
2628 | ||
2629 | cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) { | |
2630 | dsct->subtree_control &= cgroup_control(dsct); | |
5ced2518 TH |
2631 | dsct->subtree_ss_mask = |
2632 | cgroup_calc_subtree_ss_mask(dsct->subtree_control, | |
2633 | cgroup_ss_mask(dsct)); | |
15a27c36 TH |
2634 | } |
2635 | } | |
2636 | ||
2637 | /** | |
2638 | * cgroup_restore_control - restore control masks of a subtree | |
2639 | * @cgrp: root of the target subtree | |
2640 | * | |
2641 | * Restore ->subtree_control and ->subtree_ss_mask from the respective old_ | |
2642 | * prefixed fields for @cgrp's subtree including @cgrp itself. | |
2643 | */ | |
2644 | static void cgroup_restore_control(struct cgroup *cgrp) | |
2645 | { | |
2646 | struct cgroup *dsct; | |
2647 | struct cgroup_subsys_state *d_css; | |
2648 | ||
2649 | cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) { | |
2650 | dsct->subtree_control = dsct->old_subtree_control; | |
2651 | dsct->subtree_ss_mask = dsct->old_subtree_ss_mask; | |
2652 | } | |
2653 | } | |
2654 | ||
f6d635ad TH |
2655 | static bool css_visible(struct cgroup_subsys_state *css) |
2656 | { | |
2657 | struct cgroup_subsys *ss = css->ss; | |
2658 | struct cgroup *cgrp = css->cgroup; | |
2659 | ||
2660 | if (cgroup_control(cgrp) & (1 << ss->id)) | |
2661 | return true; | |
2662 | if (!(cgroup_ss_mask(cgrp) & (1 << ss->id))) | |
2663 | return false; | |
2664 | return cgroup_on_dfl(cgrp) && ss->implicit_on_dfl; | |
2665 | } | |
2666 | ||
bdb53bd7 TH |
2667 | /** |
2668 | * cgroup_apply_control_enable - enable or show csses according to control | |
ce3f1d9d | 2669 | * @cgrp: root of the target subtree |
bdb53bd7 | 2670 | * |
ce3f1d9d | 2671 | * Walk @cgrp's subtree and create new csses or make the existing ones |
bdb53bd7 TH |
2672 | * visible. A css is created invisible if it's being implicitly enabled |
2673 | * through dependency. An invisible css is made visible when the userland | |
2674 | * explicitly enables it. | |
2675 | * | |
2676 | * Returns 0 on success, -errno on failure. On failure, csses which have | |
2677 | * been processed already aren't cleaned up. The caller is responsible for | |
8a1115ff | 2678 | * cleaning up with cgroup_apply_control_disable(). |
bdb53bd7 TH |
2679 | */ |
2680 | static int cgroup_apply_control_enable(struct cgroup *cgrp) | |
2681 | { | |
2682 | struct cgroup *dsct; | |
ce3f1d9d | 2683 | struct cgroup_subsys_state *d_css; |
bdb53bd7 TH |
2684 | struct cgroup_subsys *ss; |
2685 | int ssid, ret; | |
2686 | ||
ce3f1d9d | 2687 | cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) { |
bdb53bd7 TH |
2688 | for_each_subsys(ss, ssid) { |
2689 | struct cgroup_subsys_state *css = cgroup_css(dsct, ss); | |
2690 | ||
945ba199 TH |
2691 | WARN_ON_ONCE(css && percpu_ref_is_dying(&css->refcnt)); |
2692 | ||
bdb53bd7 TH |
2693 | if (!(cgroup_ss_mask(dsct) & (1 << ss->id))) |
2694 | continue; | |
2695 | ||
2696 | if (!css) { | |
2697 | css = css_create(dsct, ss); | |
2698 | if (IS_ERR(css)) | |
2699 | return PTR_ERR(css); | |
2700 | } | |
2701 | ||
f6d635ad | 2702 | if (css_visible(css)) { |
334c3679 | 2703 | ret = css_populate_dir(css); |
bdb53bd7 TH |
2704 | if (ret) |
2705 | return ret; | |
2706 | } | |
2707 | } | |
2708 | } | |
2709 | ||
2710 | return 0; | |
2711 | } | |
2712 | ||
12b3bb6a TH |
2713 | /** |
2714 | * cgroup_apply_control_disable - kill or hide csses according to control | |
ce3f1d9d | 2715 | * @cgrp: root of the target subtree |
12b3bb6a | 2716 | * |
ce3f1d9d | 2717 | * Walk @cgrp's subtree and kill and hide csses so that they match |
12b3bb6a TH |
2718 | * cgroup_ss_mask() and cgroup_visible_mask(). |
2719 | * | |
2720 | * A css is hidden when the userland requests it to be disabled while other | |
2721 | * subsystems are still depending on it. The css must not actively control | |
2722 | * resources and be in the vanilla state if it's made visible again later. | |
2723 | * Controllers which may be depended upon should provide ->css_reset() for | |
2724 | * this purpose. | |
2725 | */ | |
2726 | static void cgroup_apply_control_disable(struct cgroup *cgrp) | |
2727 | { | |
2728 | struct cgroup *dsct; | |
ce3f1d9d | 2729 | struct cgroup_subsys_state *d_css; |
12b3bb6a TH |
2730 | struct cgroup_subsys *ss; |
2731 | int ssid; | |
2732 | ||
ce3f1d9d | 2733 | cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) { |
12b3bb6a TH |
2734 | for_each_subsys(ss, ssid) { |
2735 | struct cgroup_subsys_state *css = cgroup_css(dsct, ss); | |
2736 | ||
945ba199 TH |
2737 | WARN_ON_ONCE(css && percpu_ref_is_dying(&css->refcnt)); |
2738 | ||
12b3bb6a TH |
2739 | if (!css) |
2740 | continue; | |
2741 | ||
334c3679 TH |
2742 | if (css->parent && |
2743 | !(cgroup_ss_mask(dsct) & (1 << ss->id))) { | |
12b3bb6a | 2744 | kill_css(css); |
f6d635ad | 2745 | } else if (!css_visible(css)) { |
334c3679 | 2746 | css_clear_dir(css); |
12b3bb6a TH |
2747 | if (ss->css_reset) |
2748 | ss->css_reset(css); | |
2749 | } | |
2750 | } | |
2751 | } | |
2752 | } | |
2753 | ||
f7b2814b TH |
2754 | /** |
2755 | * cgroup_apply_control - apply control mask updates to the subtree | |
2756 | * @cgrp: root of the target subtree | |
2757 | * | |
2758 | * subsystems can be enabled and disabled in a subtree using the following | |
2759 | * steps. | |
2760 | * | |
2761 | * 1. Call cgroup_save_control() to stash the current state. | |
2762 | * 2. Update ->subtree_control masks in the subtree as desired. | |
2763 | * 3. Call cgroup_apply_control() to apply the changes. | |
2764 | * 4. Optionally perform other related operations. | |
2765 | * 5. Call cgroup_finalize_control() to finish up. | |
2766 | * | |
2767 | * This function implements step 3 and propagates the mask changes | |
2768 | * throughout @cgrp's subtree, updates csses accordingly and perform | |
2769 | * process migrations. | |
2770 | */ | |
2771 | static int cgroup_apply_control(struct cgroup *cgrp) | |
2772 | { | |
2773 | int ret; | |
2774 | ||
2775 | cgroup_propagate_control(cgrp); | |
2776 | ||
2777 | ret = cgroup_apply_control_enable(cgrp); | |
2778 | if (ret) | |
2779 | return ret; | |
2780 | ||
2781 | /* | |
2782 | * At this point, cgroup_e_css() results reflect the new csses | |
2783 | * making the following cgroup_update_dfl_csses() properly update | |
2784 | * css associations of all tasks in the subtree. | |
2785 | */ | |
2786 | ret = cgroup_update_dfl_csses(cgrp); | |
2787 | if (ret) | |
2788 | return ret; | |
2789 | ||
2790 | return 0; | |
2791 | } | |
2792 | ||
2793 | /** | |
2794 | * cgroup_finalize_control - finalize control mask update | |
2795 | * @cgrp: root of the target subtree | |
2796 | * @ret: the result of the update | |
2797 | * | |
2798 | * Finalize control mask update. See cgroup_apply_control() for more info. | |
2799 | */ | |
2800 | static void cgroup_finalize_control(struct cgroup *cgrp, int ret) | |
2801 | { | |
2802 | if (ret) { | |
2803 | cgroup_restore_control(cgrp); | |
2804 | cgroup_propagate_control(cgrp); | |
2805 | } | |
2806 | ||
2807 | cgroup_apply_control_disable(cgrp); | |
2808 | } | |
2809 | ||
f8f22e53 | 2810 | /* change the enabled child controllers for a cgroup in the default hierarchy */ |
451af504 TH |
2811 | static ssize_t cgroup_subtree_control_write(struct kernfs_open_file *of, |
2812 | char *buf, size_t nbytes, | |
2813 | loff_t off) | |
f8f22e53 | 2814 | { |
6e5c8307 | 2815 | u16 enable = 0, disable = 0; |
a9746d8d | 2816 | struct cgroup *cgrp, *child; |
f8f22e53 | 2817 | struct cgroup_subsys *ss; |
451af504 | 2818 | char *tok; |
f8f22e53 TH |
2819 | int ssid, ret; |
2820 | ||
2821 | /* | |
d37167ab TH |
2822 | * Parse input - space separated list of subsystem names prefixed |
2823 | * with either + or -. | |
f8f22e53 | 2824 | */ |
451af504 TH |
2825 | buf = strstrip(buf); |
2826 | while ((tok = strsep(&buf, " "))) { | |
d37167ab TH |
2827 | if (tok[0] == '\0') |
2828 | continue; | |
a7165264 | 2829 | do_each_subsys_mask(ss, ssid, ~cgrp_dfl_inhibit_ss_mask) { |
fc5ed1e9 TH |
2830 | if (!cgroup_ssid_enabled(ssid) || |
2831 | strcmp(tok + 1, ss->name)) | |
f8f22e53 TH |
2832 | continue; |
2833 | ||
2834 | if (*tok == '+') { | |
7d331fa9 TH |
2835 | enable |= 1 << ssid; |
2836 | disable &= ~(1 << ssid); | |
f8f22e53 | 2837 | } else if (*tok == '-') { |
7d331fa9 TH |
2838 | disable |= 1 << ssid; |
2839 | enable &= ~(1 << ssid); | |
f8f22e53 TH |
2840 | } else { |
2841 | return -EINVAL; | |
2842 | } | |
2843 | break; | |
b4e0eeaf | 2844 | } while_each_subsys_mask(); |
f8f22e53 TH |
2845 | if (ssid == CGROUP_SUBSYS_COUNT) |
2846 | return -EINVAL; | |
2847 | } | |
2848 | ||
945ba199 | 2849 | cgrp = cgroup_kn_lock_live(of->kn, true); |
a9746d8d TH |
2850 | if (!cgrp) |
2851 | return -ENODEV; | |
f8f22e53 TH |
2852 | |
2853 | for_each_subsys(ss, ssid) { | |
2854 | if (enable & (1 << ssid)) { | |
667c2491 | 2855 | if (cgrp->subtree_control & (1 << ssid)) { |
f8f22e53 TH |
2856 | enable &= ~(1 << ssid); |
2857 | continue; | |
2858 | } | |
2859 | ||
5531dc91 | 2860 | if (!(cgroup_control(cgrp) & (1 << ssid))) { |
c29adf24 TH |
2861 | ret = -ENOENT; |
2862 | goto out_unlock; | |
2863 | } | |
f8f22e53 | 2864 | } else if (disable & (1 << ssid)) { |
667c2491 | 2865 | if (!(cgrp->subtree_control & (1 << ssid))) { |
f8f22e53 TH |
2866 | disable &= ~(1 << ssid); |
2867 | continue; | |
2868 | } | |
2869 | ||
2870 | /* a child has it enabled? */ | |
2871 | cgroup_for_each_live_child(child, cgrp) { | |
667c2491 | 2872 | if (child->subtree_control & (1 << ssid)) { |
f8f22e53 | 2873 | ret = -EBUSY; |
ddab2b6e | 2874 | goto out_unlock; |
f8f22e53 TH |
2875 | } |
2876 | } | |
2877 | } | |
2878 | } | |
2879 | ||
2880 | if (!enable && !disable) { | |
2881 | ret = 0; | |
ddab2b6e | 2882 | goto out_unlock; |
f8f22e53 TH |
2883 | } |
2884 | ||
2885 | /* | |
667c2491 | 2886 | * Except for the root, subtree_control must be zero for a cgroup |
f8f22e53 TH |
2887 | * with tasks so that child cgroups don't compete against tasks. |
2888 | */ | |
9157056d TH |
2889 | if (enable && cgroup_parent(cgrp)) { |
2890 | struct cgrp_cset_link *link; | |
2891 | ||
2892 | /* | |
2893 | * Because namespaces pin csets too, @cgrp->cset_links | |
2894 | * might not be empty even when @cgrp is empty. Walk and | |
2895 | * verify each cset. | |
2896 | */ | |
2897 | spin_lock_irq(&css_set_lock); | |
2898 | ||
2899 | ret = 0; | |
2900 | list_for_each_entry(link, &cgrp->cset_links, cset_link) { | |
2901 | if (css_set_populated(link->cset)) { | |
2902 | ret = -EBUSY; | |
2903 | break; | |
2904 | } | |
2905 | } | |
2906 | ||
2907 | spin_unlock_irq(&css_set_lock); | |
2908 | ||
2909 | if (ret) | |
2910 | goto out_unlock; | |
f8f22e53 TH |
2911 | } |
2912 | ||
15a27c36 TH |
2913 | /* save and update control masks and prepare csses */ |
2914 | cgroup_save_control(cgrp); | |
f63070d3 | 2915 | |
15a27c36 TH |
2916 | cgrp->subtree_control |= enable; |
2917 | cgrp->subtree_control &= ~disable; | |
c29adf24 | 2918 | |
f7b2814b | 2919 | ret = cgroup_apply_control(cgrp); |
f8f22e53 | 2920 | |
f7b2814b | 2921 | cgroup_finalize_control(cgrp, ret); |
f8f22e53 TH |
2922 | |
2923 | kernfs_activate(cgrp->kn); | |
2924 | ret = 0; | |
2925 | out_unlock: | |
a9746d8d | 2926 | cgroup_kn_unlock(of->kn); |
451af504 | 2927 | return ret ?: nbytes; |
f8f22e53 TH |
2928 | } |
2929 | ||
4a07c222 | 2930 | static int cgroup_events_show(struct seq_file *seq, void *v) |
842b597e | 2931 | { |
4a07c222 | 2932 | seq_printf(seq, "populated %d\n", |
27bd4dbb | 2933 | cgroup_is_populated(seq_css(seq)->cgroup)); |
842b597e TH |
2934 | return 0; |
2935 | } | |
2936 | ||
e90cbebc TH |
2937 | static int cgroup_file_open(struct kernfs_open_file *of) |
2938 | { | |
2939 | struct cftype *cft = of->kn->priv; | |
2940 | ||
2941 | if (cft->open) | |
2942 | return cft->open(of); | |
2943 | return 0; | |
2944 | } | |
2945 | ||
2946 | static void cgroup_file_release(struct kernfs_open_file *of) | |
2947 | { | |
2948 | struct cftype *cft = of->kn->priv; | |
2949 | ||
2950 | if (cft->release) | |
2951 | cft->release(of); | |
2952 | } | |
2953 | ||
2bd59d48 TH |
2954 | static ssize_t cgroup_file_write(struct kernfs_open_file *of, char *buf, |
2955 | size_t nbytes, loff_t off) | |
355e0c48 | 2956 | { |
2bd59d48 TH |
2957 | struct cgroup *cgrp = of->kn->parent->priv; |
2958 | struct cftype *cft = of->kn->priv; | |
2959 | struct cgroup_subsys_state *css; | |
a742c59d | 2960 | int ret; |
355e0c48 | 2961 | |
b4168640 TH |
2962 | if (cft->write) |
2963 | return cft->write(of, buf, nbytes, off); | |
2964 | ||
2bd59d48 TH |
2965 | /* |
2966 | * kernfs guarantees that a file isn't deleted with operations in | |
2967 | * flight, which means that the matching css is and stays alive and | |
2968 | * doesn't need to be pinned. The RCU locking is not necessary | |
2969 | * either. It's just for the convenience of using cgroup_css(). | |
2970 | */ | |
2971 | rcu_read_lock(); | |
2972 | css = cgroup_css(cgrp, cft->ss); | |
2973 | rcu_read_unlock(); | |
a742c59d | 2974 | |
451af504 | 2975 | if (cft->write_u64) { |
a742c59d TH |
2976 | unsigned long long v; |
2977 | ret = kstrtoull(buf, 0, &v); | |
2978 | if (!ret) | |
2979 | ret = cft->write_u64(css, cft, v); | |
2980 | } else if (cft->write_s64) { | |
2981 | long long v; | |
2982 | ret = kstrtoll(buf, 0, &v); | |
2983 | if (!ret) | |
2984 | ret = cft->write_s64(css, cft, v); | |
e73d2c61 | 2985 | } else { |
a742c59d | 2986 | ret = -EINVAL; |
e73d2c61 | 2987 | } |
2bd59d48 | 2988 | |
a742c59d | 2989 | return ret ?: nbytes; |
355e0c48 PM |
2990 | } |
2991 | ||
6612f05b | 2992 | static void *cgroup_seqfile_start(struct seq_file *seq, loff_t *ppos) |
db3b1497 | 2993 | { |
2bd59d48 | 2994 | return seq_cft(seq)->seq_start(seq, ppos); |
db3b1497 PM |
2995 | } |
2996 | ||
6612f05b | 2997 | static void *cgroup_seqfile_next(struct seq_file *seq, void *v, loff_t *ppos) |
ddbcc7e8 | 2998 | { |
2bd59d48 | 2999 | return seq_cft(seq)->seq_next(seq, v, ppos); |
ddbcc7e8 PM |
3000 | } |
3001 | ||
6612f05b | 3002 | static void cgroup_seqfile_stop(struct seq_file *seq, void *v) |
ddbcc7e8 | 3003 | { |
e90cbebc TH |
3004 | if (seq_cft(seq)->seq_stop) |
3005 | seq_cft(seq)->seq_stop(seq, v); | |
ddbcc7e8 PM |
3006 | } |
3007 | ||
91796569 | 3008 | static int cgroup_seqfile_show(struct seq_file *m, void *arg) |
e73d2c61 | 3009 | { |
7da11279 TH |
3010 | struct cftype *cft = seq_cft(m); |
3011 | struct cgroup_subsys_state *css = seq_css(m); | |
e73d2c61 | 3012 | |
2da8ca82 TH |
3013 | if (cft->seq_show) |
3014 | return cft->seq_show(m, arg); | |
e73d2c61 | 3015 | |
f4c753b7 | 3016 | if (cft->read_u64) |
896f5199 TH |
3017 | seq_printf(m, "%llu\n", cft->read_u64(css, cft)); |
3018 | else if (cft->read_s64) | |
3019 | seq_printf(m, "%lld\n", cft->read_s64(css, cft)); | |
3020 | else | |
3021 | return -EINVAL; | |
3022 | return 0; | |
91796569 PM |
3023 | } |
3024 | ||
2bd59d48 TH |
3025 | static struct kernfs_ops cgroup_kf_single_ops = { |
3026 | .atomic_write_len = PAGE_SIZE, | |
e90cbebc TH |
3027 | .open = cgroup_file_open, |
3028 | .release = cgroup_file_release, | |
2bd59d48 TH |
3029 | .write = cgroup_file_write, |
3030 | .seq_show = cgroup_seqfile_show, | |
91796569 PM |
3031 | }; |
3032 | ||
2bd59d48 TH |
3033 | static struct kernfs_ops cgroup_kf_ops = { |
3034 | .atomic_write_len = PAGE_SIZE, | |
e90cbebc TH |
3035 | .open = cgroup_file_open, |
3036 | .release = cgroup_file_release, | |
2bd59d48 TH |
3037 | .write = cgroup_file_write, |
3038 | .seq_start = cgroup_seqfile_start, | |
3039 | .seq_next = cgroup_seqfile_next, | |
3040 | .seq_stop = cgroup_seqfile_stop, | |
3041 | .seq_show = cgroup_seqfile_show, | |
3042 | }; | |
ddbcc7e8 | 3043 | |
49957f8e TH |
3044 | /* set uid and gid of cgroup dirs and files to that of the creator */ |
3045 | static int cgroup_kn_set_ugid(struct kernfs_node *kn) | |
3046 | { | |
3047 | struct iattr iattr = { .ia_valid = ATTR_UID | ATTR_GID, | |
3048 | .ia_uid = current_fsuid(), | |
3049 | .ia_gid = current_fsgid(), }; | |
3050 | ||
3051 | if (uid_eq(iattr.ia_uid, GLOBAL_ROOT_UID) && | |
3052 | gid_eq(iattr.ia_gid, GLOBAL_ROOT_GID)) | |
3053 | return 0; | |
3054 | ||
3055 | return kernfs_setattr(kn, &iattr); | |
3056 | } | |
3057 | ||
4df8dc90 TH |
3058 | static int cgroup_add_file(struct cgroup_subsys_state *css, struct cgroup *cgrp, |
3059 | struct cftype *cft) | |
ddbcc7e8 | 3060 | { |
8d7e6fb0 | 3061 | char name[CGROUP_FILE_NAME_MAX]; |
2bd59d48 TH |
3062 | struct kernfs_node *kn; |
3063 | struct lock_class_key *key = NULL; | |
49957f8e | 3064 | int ret; |
05ef1d7c | 3065 | |
2bd59d48 TH |
3066 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
3067 | key = &cft->lockdep_key; | |
3068 | #endif | |
3069 | kn = __kernfs_create_file(cgrp->kn, cgroup_file_name(cgrp, cft, name), | |
3070 | cgroup_file_mode(cft), 0, cft->kf_ops, cft, | |
dfeb0750 | 3071 | NULL, key); |
49957f8e TH |
3072 | if (IS_ERR(kn)) |
3073 | return PTR_ERR(kn); | |
3074 | ||
3075 | ret = cgroup_kn_set_ugid(kn); | |
f8f22e53 | 3076 | if (ret) { |
49957f8e | 3077 | kernfs_remove(kn); |
f8f22e53 TH |
3078 | return ret; |
3079 | } | |
3080 | ||
6f60eade TH |
3081 | if (cft->file_offset) { |
3082 | struct cgroup_file *cfile = (void *)css + cft->file_offset; | |
3083 | ||
34c06254 | 3084 | spin_lock_irq(&cgroup_file_kn_lock); |
6f60eade | 3085 | cfile->kn = kn; |
34c06254 | 3086 | spin_unlock_irq(&cgroup_file_kn_lock); |
6f60eade TH |
3087 | } |
3088 | ||
f8f22e53 | 3089 | return 0; |
ddbcc7e8 PM |
3090 | } |
3091 | ||
b1f28d31 TH |
3092 | /** |
3093 | * cgroup_addrm_files - add or remove files to a cgroup directory | |
4df8dc90 TH |
3094 | * @css: the target css |
3095 | * @cgrp: the target cgroup (usually css->cgroup) | |
b1f28d31 TH |
3096 | * @cfts: array of cftypes to be added |
3097 | * @is_add: whether to add or remove | |
3098 | * | |
3099 | * Depending on @is_add, add or remove files defined by @cfts on @cgrp. | |
6732ed85 | 3100 | * For removals, this function never fails. |
b1f28d31 | 3101 | */ |
4df8dc90 TH |
3102 | static int cgroup_addrm_files(struct cgroup_subsys_state *css, |
3103 | struct cgroup *cgrp, struct cftype cfts[], | |
2bb566cb | 3104 | bool is_add) |
ddbcc7e8 | 3105 | { |
6732ed85 | 3106 | struct cftype *cft, *cft_end = NULL; |
b598dde3 | 3107 | int ret = 0; |
b1f28d31 | 3108 | |
01f6474c | 3109 | lockdep_assert_held(&cgroup_mutex); |
db0416b6 | 3110 | |
6732ed85 TH |
3111 | restart: |
3112 | for (cft = cfts; cft != cft_end && cft->name[0] != '\0'; cft++) { | |
f33fddc2 | 3113 | /* does cft->flags tell us to skip this file on @cgrp? */ |
05ebb6e6 | 3114 | if ((cft->flags & __CFTYPE_ONLY_ON_DFL) && !cgroup_on_dfl(cgrp)) |
8cbbf2c9 | 3115 | continue; |
05ebb6e6 | 3116 | if ((cft->flags & __CFTYPE_NOT_ON_DFL) && cgroup_on_dfl(cgrp)) |
873fe09e | 3117 | continue; |
d51f39b0 | 3118 | if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgroup_parent(cgrp)) |
f33fddc2 | 3119 | continue; |
d51f39b0 | 3120 | if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgroup_parent(cgrp)) |
f33fddc2 G |
3121 | continue; |
3122 | ||
2739d3cc | 3123 | if (is_add) { |
4df8dc90 | 3124 | ret = cgroup_add_file(css, cgrp, cft); |
b1f28d31 | 3125 | if (ret) { |
ed3d261b JP |
3126 | pr_warn("%s: failed to add %s, err=%d\n", |
3127 | __func__, cft->name, ret); | |
6732ed85 TH |
3128 | cft_end = cft; |
3129 | is_add = false; | |
3130 | goto restart; | |
b1f28d31 | 3131 | } |
2739d3cc LZ |
3132 | } else { |
3133 | cgroup_rm_file(cgrp, cft); | |
db0416b6 | 3134 | } |
ddbcc7e8 | 3135 | } |
b598dde3 | 3136 | return ret; |
ddbcc7e8 PM |
3137 | } |
3138 | ||
21a2d343 | 3139 | static int cgroup_apply_cftypes(struct cftype *cfts, bool is_add) |
8e3f6541 TH |
3140 | { |
3141 | LIST_HEAD(pending); | |
2bb566cb | 3142 | struct cgroup_subsys *ss = cfts[0].ss; |
3dd06ffa | 3143 | struct cgroup *root = &ss->root->cgrp; |
492eb21b | 3144 | struct cgroup_subsys_state *css; |
9ccece80 | 3145 | int ret = 0; |
8e3f6541 | 3146 | |
01f6474c | 3147 | lockdep_assert_held(&cgroup_mutex); |
e8c82d20 | 3148 | |
e8c82d20 | 3149 | /* add/rm files for all cgroups created before */ |
ca8bdcaf | 3150 | css_for_each_descendant_pre(css, cgroup_css(root, ss)) { |
492eb21b TH |
3151 | struct cgroup *cgrp = css->cgroup; |
3152 | ||
88cb04b9 | 3153 | if (!(css->flags & CSS_VISIBLE)) |
e8c82d20 LZ |
3154 | continue; |
3155 | ||
4df8dc90 | 3156 | ret = cgroup_addrm_files(css, cgrp, cfts, is_add); |
9ccece80 TH |
3157 | if (ret) |
3158 | break; | |
8e3f6541 | 3159 | } |
21a2d343 TH |
3160 | |
3161 | if (is_add && !ret) | |
3162 | kernfs_activate(root->kn); | |
9ccece80 | 3163 | return ret; |
8e3f6541 TH |
3164 | } |
3165 | ||
2da440a2 | 3166 | static void cgroup_exit_cftypes(struct cftype *cfts) |
8e3f6541 | 3167 | { |
2bb566cb | 3168 | struct cftype *cft; |
8e3f6541 | 3169 | |
2bd59d48 TH |
3170 | for (cft = cfts; cft->name[0] != '\0'; cft++) { |
3171 | /* free copy for custom atomic_write_len, see init_cftypes() */ | |
3172 | if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) | |
3173 | kfree(cft->kf_ops); | |
3174 | cft->kf_ops = NULL; | |
2da440a2 | 3175 | cft->ss = NULL; |
a8ddc821 TH |
3176 | |
3177 | /* revert flags set by cgroup core while adding @cfts */ | |
05ebb6e6 | 3178 | cft->flags &= ~(__CFTYPE_ONLY_ON_DFL | __CFTYPE_NOT_ON_DFL); |
2bd59d48 | 3179 | } |
2da440a2 TH |
3180 | } |
3181 | ||
2bd59d48 | 3182 | static int cgroup_init_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
2da440a2 TH |
3183 | { |
3184 | struct cftype *cft; | |
3185 | ||
2bd59d48 TH |
3186 | for (cft = cfts; cft->name[0] != '\0'; cft++) { |
3187 | struct kernfs_ops *kf_ops; | |
3188 | ||
0adb0704 TH |
3189 | WARN_ON(cft->ss || cft->kf_ops); |
3190 | ||
2bd59d48 TH |
3191 | if (cft->seq_start) |
3192 | kf_ops = &cgroup_kf_ops; | |
3193 | else | |
3194 | kf_ops = &cgroup_kf_single_ops; | |
3195 | ||
3196 | /* | |
3197 | * Ugh... if @cft wants a custom max_write_len, we need to | |
3198 | * make a copy of kf_ops to set its atomic_write_len. | |
3199 | */ | |
3200 | if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) { | |
3201 | kf_ops = kmemdup(kf_ops, sizeof(*kf_ops), GFP_KERNEL); | |
3202 | if (!kf_ops) { | |
3203 | cgroup_exit_cftypes(cfts); | |
3204 | return -ENOMEM; | |
3205 | } | |
3206 | kf_ops->atomic_write_len = cft->max_write_len; | |
3207 | } | |
8e3f6541 | 3208 | |
2bd59d48 | 3209 | cft->kf_ops = kf_ops; |
2bb566cb | 3210 | cft->ss = ss; |
2bd59d48 | 3211 | } |
2bb566cb | 3212 | |
2bd59d48 | 3213 | return 0; |
2da440a2 TH |
3214 | } |
3215 | ||
21a2d343 TH |
3216 | static int cgroup_rm_cftypes_locked(struct cftype *cfts) |
3217 | { | |
01f6474c | 3218 | lockdep_assert_held(&cgroup_mutex); |
21a2d343 TH |
3219 | |
3220 | if (!cfts || !cfts[0].ss) | |
3221 | return -ENOENT; | |
3222 | ||
3223 | list_del(&cfts->node); | |
3224 | cgroup_apply_cftypes(cfts, false); | |
3225 | cgroup_exit_cftypes(cfts); | |
3226 | return 0; | |
8e3f6541 | 3227 | } |
8e3f6541 | 3228 | |
79578621 TH |
3229 | /** |
3230 | * cgroup_rm_cftypes - remove an array of cftypes from a subsystem | |
79578621 TH |
3231 | * @cfts: zero-length name terminated array of cftypes |
3232 | * | |
2bb566cb TH |
3233 | * Unregister @cfts. Files described by @cfts are removed from all |
3234 | * existing cgroups and all future cgroups won't have them either. This | |
3235 | * function can be called anytime whether @cfts' subsys is attached or not. | |
79578621 TH |
3236 | * |
3237 | * Returns 0 on successful unregistration, -ENOENT if @cfts is not | |
2bb566cb | 3238 | * registered. |
79578621 | 3239 | */ |
2bb566cb | 3240 | int cgroup_rm_cftypes(struct cftype *cfts) |
79578621 | 3241 | { |
21a2d343 | 3242 | int ret; |
79578621 | 3243 | |
01f6474c | 3244 | mutex_lock(&cgroup_mutex); |
21a2d343 | 3245 | ret = cgroup_rm_cftypes_locked(cfts); |
01f6474c | 3246 | mutex_unlock(&cgroup_mutex); |
21a2d343 | 3247 | return ret; |
80b13586 TH |
3248 | } |
3249 | ||
8e3f6541 TH |
3250 | /** |
3251 | * cgroup_add_cftypes - add an array of cftypes to a subsystem | |
3252 | * @ss: target cgroup subsystem | |
3253 | * @cfts: zero-length name terminated array of cftypes | |
3254 | * | |
3255 | * Register @cfts to @ss. Files described by @cfts are created for all | |
3256 | * existing cgroups to which @ss is attached and all future cgroups will | |
3257 | * have them too. This function can be called anytime whether @ss is | |
3258 | * attached or not. | |
3259 | * | |
3260 | * Returns 0 on successful registration, -errno on failure. Note that this | |
3261 | * function currently returns 0 as long as @cfts registration is successful | |
3262 | * even if some file creation attempts on existing cgroups fail. | |
3263 | */ | |
2cf669a5 | 3264 | static int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
8e3f6541 | 3265 | { |
9ccece80 | 3266 | int ret; |
8e3f6541 | 3267 | |
fc5ed1e9 | 3268 | if (!cgroup_ssid_enabled(ss->id)) |
c731ae1d LZ |
3269 | return 0; |
3270 | ||
dc5736ed LZ |
3271 | if (!cfts || cfts[0].name[0] == '\0') |
3272 | return 0; | |
2bb566cb | 3273 | |
2bd59d48 TH |
3274 | ret = cgroup_init_cftypes(ss, cfts); |
3275 | if (ret) | |
3276 | return ret; | |
79578621 | 3277 | |
01f6474c | 3278 | mutex_lock(&cgroup_mutex); |
21a2d343 | 3279 | |
0adb0704 | 3280 | list_add_tail(&cfts->node, &ss->cfts); |
21a2d343 | 3281 | ret = cgroup_apply_cftypes(cfts, true); |
9ccece80 | 3282 | if (ret) |
21a2d343 | 3283 | cgroup_rm_cftypes_locked(cfts); |
79578621 | 3284 | |
01f6474c | 3285 | mutex_unlock(&cgroup_mutex); |
9ccece80 | 3286 | return ret; |
79578621 TH |
3287 | } |
3288 | ||
a8ddc821 TH |
3289 | /** |
3290 | * cgroup_add_dfl_cftypes - add an array of cftypes for default hierarchy | |
3291 | * @ss: target cgroup subsystem | |
3292 | * @cfts: zero-length name terminated array of cftypes | |
3293 | * | |
3294 | * Similar to cgroup_add_cftypes() but the added files are only used for | |
3295 | * the default hierarchy. | |
3296 | */ | |
3297 | int cgroup_add_dfl_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) | |
3298 | { | |
3299 | struct cftype *cft; | |
3300 | ||
3301 | for (cft = cfts; cft && cft->name[0] != '\0'; cft++) | |
05ebb6e6 | 3302 | cft->flags |= __CFTYPE_ONLY_ON_DFL; |
a8ddc821 TH |
3303 | return cgroup_add_cftypes(ss, cfts); |
3304 | } | |
3305 | ||
3306 | /** | |
3307 | * cgroup_add_legacy_cftypes - add an array of cftypes for legacy hierarchies | |
3308 | * @ss: target cgroup subsystem | |
3309 | * @cfts: zero-length name terminated array of cftypes | |
3310 | * | |
3311 | * Similar to cgroup_add_cftypes() but the added files are only used for | |
3312 | * the legacy hierarchies. | |
3313 | */ | |
2cf669a5 TH |
3314 | int cgroup_add_legacy_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
3315 | { | |
a8ddc821 TH |
3316 | struct cftype *cft; |
3317 | ||
e4b7037c TH |
3318 | for (cft = cfts; cft && cft->name[0] != '\0'; cft++) |
3319 | cft->flags |= __CFTYPE_NOT_ON_DFL; | |
2cf669a5 TH |
3320 | return cgroup_add_cftypes(ss, cfts); |
3321 | } | |
3322 | ||
34c06254 TH |
3323 | /** |
3324 | * cgroup_file_notify - generate a file modified event for a cgroup_file | |
3325 | * @cfile: target cgroup_file | |
3326 | * | |
3327 | * @cfile must have been obtained by setting cftype->file_offset. | |
3328 | */ | |
3329 | void cgroup_file_notify(struct cgroup_file *cfile) | |
3330 | { | |
3331 | unsigned long flags; | |
3332 | ||
3333 | spin_lock_irqsave(&cgroup_file_kn_lock, flags); | |
3334 | if (cfile->kn) | |
3335 | kernfs_notify(cfile->kn); | |
3336 | spin_unlock_irqrestore(&cgroup_file_kn_lock, flags); | |
3337 | } | |
3338 | ||
53fa5261 | 3339 | /** |
492eb21b | 3340 | * css_next_child - find the next child of a given css |
c2931b70 TH |
3341 | * @pos: the current position (%NULL to initiate traversal) |
3342 | * @parent: css whose children to walk | |
53fa5261 | 3343 | * |
c2931b70 | 3344 | * This function returns the next child of @parent and should be called |
87fb54f1 | 3345 | * under either cgroup_mutex or RCU read lock. The only requirement is |
c2931b70 TH |
3346 | * that @parent and @pos are accessible. The next sibling is guaranteed to |
3347 | * be returned regardless of their states. | |
3348 | * | |
3349 | * If a subsystem synchronizes ->css_online() and the start of iteration, a | |
3350 | * css which finished ->css_online() is guaranteed to be visible in the | |
3351 | * future iterations and will stay visible until the last reference is put. | |
3352 | * A css which hasn't finished ->css_online() or already finished | |
3353 | * ->css_offline() may show up during traversal. It's each subsystem's | |
3354 | * responsibility to synchronize against on/offlining. | |
53fa5261 | 3355 | */ |
c2931b70 TH |
3356 | struct cgroup_subsys_state *css_next_child(struct cgroup_subsys_state *pos, |
3357 | struct cgroup_subsys_state *parent) | |
53fa5261 | 3358 | { |
c2931b70 | 3359 | struct cgroup_subsys_state *next; |
53fa5261 | 3360 | |
8353da1f | 3361 | cgroup_assert_mutex_or_rcu_locked(); |
53fa5261 TH |
3362 | |
3363 | /* | |
de3f0341 TH |
3364 | * @pos could already have been unlinked from the sibling list. |
3365 | * Once a cgroup is removed, its ->sibling.next is no longer | |
3366 | * updated when its next sibling changes. CSS_RELEASED is set when | |
3367 | * @pos is taken off list, at which time its next pointer is valid, | |
3368 | * and, as releases are serialized, the one pointed to by the next | |
3369 | * pointer is guaranteed to not have started release yet. This | |
3370 | * implies that if we observe !CSS_RELEASED on @pos in this RCU | |
3371 | * critical section, the one pointed to by its next pointer is | |
3372 | * guaranteed to not have finished its RCU grace period even if we | |
3373 | * have dropped rcu_read_lock() inbetween iterations. | |
3b287a50 | 3374 | * |
de3f0341 TH |
3375 | * If @pos has CSS_RELEASED set, its next pointer can't be |
3376 | * dereferenced; however, as each css is given a monotonically | |
3377 | * increasing unique serial number and always appended to the | |
3378 | * sibling list, the next one can be found by walking the parent's | |
3379 | * children until the first css with higher serial number than | |
3380 | * @pos's. While this path can be slower, it happens iff iteration | |
3381 | * races against release and the race window is very small. | |
53fa5261 | 3382 | */ |
3b287a50 | 3383 | if (!pos) { |
c2931b70 TH |
3384 | next = list_entry_rcu(parent->children.next, struct cgroup_subsys_state, sibling); |
3385 | } else if (likely(!(pos->flags & CSS_RELEASED))) { | |
3386 | next = list_entry_rcu(pos->sibling.next, struct cgroup_subsys_state, sibling); | |
3b287a50 | 3387 | } else { |
c2931b70 | 3388 | list_for_each_entry_rcu(next, &parent->children, sibling) |
3b287a50 TH |
3389 | if (next->serial_nr > pos->serial_nr) |
3390 | break; | |
53fa5261 TH |
3391 | } |
3392 | ||
3b281afb TH |
3393 | /* |
3394 | * @next, if not pointing to the head, can be dereferenced and is | |
c2931b70 | 3395 | * the next sibling. |
3b281afb | 3396 | */ |
c2931b70 TH |
3397 | if (&next->sibling != &parent->children) |
3398 | return next; | |
3b281afb | 3399 | return NULL; |
53fa5261 | 3400 | } |
53fa5261 | 3401 | |
574bd9f7 | 3402 | /** |
492eb21b | 3403 | * css_next_descendant_pre - find the next descendant for pre-order walk |
574bd9f7 | 3404 | * @pos: the current position (%NULL to initiate traversal) |
492eb21b | 3405 | * @root: css whose descendants to walk |
574bd9f7 | 3406 | * |
492eb21b | 3407 | * To be used by css_for_each_descendant_pre(). Find the next descendant |
bd8815a6 TH |
3408 | * to visit for pre-order traversal of @root's descendants. @root is |
3409 | * included in the iteration and the first node to be visited. | |
75501a6d | 3410 | * |
87fb54f1 TH |
3411 | * While this function requires cgroup_mutex or RCU read locking, it |
3412 | * doesn't require the whole traversal to be contained in a single critical | |
3413 | * section. This function will return the correct next descendant as long | |
3414 | * as both @pos and @root are accessible and @pos is a descendant of @root. | |
c2931b70 TH |
3415 | * |
3416 | * If a subsystem synchronizes ->css_online() and the start of iteration, a | |
3417 | * css which finished ->css_online() is guaranteed to be visible in the | |
3418 | * future iterations and will stay visible until the last reference is put. | |
3419 | * A css which hasn't finished ->css_online() or already finished | |
3420 | * ->css_offline() may show up during traversal. It's each subsystem's | |
3421 | * responsibility to synchronize against on/offlining. | |
574bd9f7 | 3422 | */ |
492eb21b TH |
3423 | struct cgroup_subsys_state * |
3424 | css_next_descendant_pre(struct cgroup_subsys_state *pos, | |
3425 | struct cgroup_subsys_state *root) | |
574bd9f7 | 3426 | { |
492eb21b | 3427 | struct cgroup_subsys_state *next; |
574bd9f7 | 3428 | |
8353da1f | 3429 | cgroup_assert_mutex_or_rcu_locked(); |
574bd9f7 | 3430 | |
bd8815a6 | 3431 | /* if first iteration, visit @root */ |
7805d000 | 3432 | if (!pos) |
bd8815a6 | 3433 | return root; |
574bd9f7 TH |
3434 | |
3435 | /* visit the first child if exists */ | |
492eb21b | 3436 | next = css_next_child(NULL, pos); |
574bd9f7 TH |
3437 | if (next) |
3438 | return next; | |
3439 | ||
3440 | /* no child, visit my or the closest ancestor's next sibling */ | |
492eb21b | 3441 | while (pos != root) { |
5c9d535b | 3442 | next = css_next_child(pos, pos->parent); |
75501a6d | 3443 | if (next) |
574bd9f7 | 3444 | return next; |
5c9d535b | 3445 | pos = pos->parent; |
7805d000 | 3446 | } |
574bd9f7 TH |
3447 | |
3448 | return NULL; | |
3449 | } | |
574bd9f7 | 3450 | |
12a9d2fe | 3451 | /** |
492eb21b TH |
3452 | * css_rightmost_descendant - return the rightmost descendant of a css |
3453 | * @pos: css of interest | |
12a9d2fe | 3454 | * |
492eb21b TH |
3455 | * Return the rightmost descendant of @pos. If there's no descendant, @pos |
3456 | * is returned. This can be used during pre-order traversal to skip | |
12a9d2fe | 3457 | * subtree of @pos. |
75501a6d | 3458 | * |
87fb54f1 TH |
3459 | * While this function requires cgroup_mutex or RCU read locking, it |
3460 | * doesn't require the whole traversal to be contained in a single critical | |
3461 | * section. This function will return the correct rightmost descendant as | |
3462 | * long as @pos is accessible. | |
12a9d2fe | 3463 | */ |
492eb21b TH |
3464 | struct cgroup_subsys_state * |
3465 | css_rightmost_descendant(struct cgroup_subsys_state *pos) | |
12a9d2fe | 3466 | { |
492eb21b | 3467 | struct cgroup_subsys_state *last, *tmp; |
12a9d2fe | 3468 | |
8353da1f | 3469 | cgroup_assert_mutex_or_rcu_locked(); |
12a9d2fe TH |
3470 | |
3471 | do { | |
3472 | last = pos; | |
3473 | /* ->prev isn't RCU safe, walk ->next till the end */ | |
3474 | pos = NULL; | |
492eb21b | 3475 | css_for_each_child(tmp, last) |
12a9d2fe TH |
3476 | pos = tmp; |
3477 | } while (pos); | |
3478 | ||
3479 | return last; | |
3480 | } | |
12a9d2fe | 3481 | |
492eb21b TH |
3482 | static struct cgroup_subsys_state * |
3483 | css_leftmost_descendant(struct cgroup_subsys_state *pos) | |
574bd9f7 | 3484 | { |
492eb21b | 3485 | struct cgroup_subsys_state *last; |
574bd9f7 TH |
3486 | |
3487 | do { | |
3488 | last = pos; | |
492eb21b | 3489 | pos = css_next_child(NULL, pos); |
574bd9f7 TH |
3490 | } while (pos); |
3491 | ||
3492 | return last; | |
3493 | } | |
3494 | ||
3495 | /** | |
492eb21b | 3496 | * css_next_descendant_post - find the next descendant for post-order walk |
574bd9f7 | 3497 | * @pos: the current position (%NULL to initiate traversal) |
492eb21b | 3498 | * @root: css whose descendants to walk |
574bd9f7 | 3499 | * |
492eb21b | 3500 | * To be used by css_for_each_descendant_post(). Find the next descendant |
bd8815a6 TH |
3501 | * to visit for post-order traversal of @root's descendants. @root is |
3502 | * included in the iteration and the last node to be visited. | |
75501a6d | 3503 | * |
87fb54f1 TH |
3504 | * While this function requires cgroup_mutex or RCU read locking, it |
3505 | * doesn't require the whole traversal to be contained in a single critical | |
3506 | * section. This function will return the correct next descendant as long | |
3507 | * as both @pos and @cgroup are accessible and @pos is a descendant of | |
3508 | * @cgroup. | |
c2931b70 TH |
3509 | * |
3510 | * If a subsystem synchronizes ->css_online() and the start of iteration, a | |
3511 | * css which finished ->css_online() is guaranteed to be visible in the | |
3512 | * future iterations and will stay visible until the last reference is put. | |
3513 | * A css which hasn't finished ->css_online() or already finished | |
3514 | * ->css_offline() may show up during traversal. It's each subsystem's | |
3515 | * responsibility to synchronize against on/offlining. | |
574bd9f7 | 3516 | */ |
492eb21b TH |
3517 | struct cgroup_subsys_state * |
3518 | css_next_descendant_post(struct cgroup_subsys_state *pos, | |
3519 | struct cgroup_subsys_state *root) | |
574bd9f7 | 3520 | { |
492eb21b | 3521 | struct cgroup_subsys_state *next; |
574bd9f7 | 3522 | |
8353da1f | 3523 | cgroup_assert_mutex_or_rcu_locked(); |
574bd9f7 | 3524 | |
58b79a91 TH |
3525 | /* if first iteration, visit leftmost descendant which may be @root */ |
3526 | if (!pos) | |
3527 | return css_leftmost_descendant(root); | |
574bd9f7 | 3528 | |
bd8815a6 TH |
3529 | /* if we visited @root, we're done */ |
3530 | if (pos == root) | |
3531 | return NULL; | |
3532 | ||
574bd9f7 | 3533 | /* if there's an unvisited sibling, visit its leftmost descendant */ |
5c9d535b | 3534 | next = css_next_child(pos, pos->parent); |
75501a6d | 3535 | if (next) |
492eb21b | 3536 | return css_leftmost_descendant(next); |
574bd9f7 TH |
3537 | |
3538 | /* no sibling left, visit parent */ | |
5c9d535b | 3539 | return pos->parent; |
574bd9f7 | 3540 | } |
574bd9f7 | 3541 | |
f3d46500 TH |
3542 | /** |
3543 | * css_has_online_children - does a css have online children | |
3544 | * @css: the target css | |
3545 | * | |
3546 | * Returns %true if @css has any online children; otherwise, %false. This | |
3547 | * function can be called from any context but the caller is responsible | |
3548 | * for synchronizing against on/offlining as necessary. | |
3549 | */ | |
3550 | bool css_has_online_children(struct cgroup_subsys_state *css) | |
cbc125ef | 3551 | { |
f3d46500 TH |
3552 | struct cgroup_subsys_state *child; |
3553 | bool ret = false; | |
cbc125ef TH |
3554 | |
3555 | rcu_read_lock(); | |
f3d46500 | 3556 | css_for_each_child(child, css) { |
99bae5f9 | 3557 | if (child->flags & CSS_ONLINE) { |
f3d46500 TH |
3558 | ret = true; |
3559 | break; | |
cbc125ef TH |
3560 | } |
3561 | } | |
3562 | rcu_read_unlock(); | |
f3d46500 | 3563 | return ret; |
574bd9f7 | 3564 | } |
574bd9f7 | 3565 | |
0942eeee | 3566 | /** |
ecb9d535 | 3567 | * css_task_iter_advance_css_set - advance a task itererator to the next css_set |
0942eeee TH |
3568 | * @it: the iterator to advance |
3569 | * | |
3570 | * Advance @it to the next css_set to walk. | |
d515876e | 3571 | */ |
ecb9d535 | 3572 | static void css_task_iter_advance_css_set(struct css_task_iter *it) |
d515876e | 3573 | { |
0f0a2b4f | 3574 | struct list_head *l = it->cset_pos; |
d515876e TH |
3575 | struct cgrp_cset_link *link; |
3576 | struct css_set *cset; | |
3577 | ||
f0d9a5f1 | 3578 | lockdep_assert_held(&css_set_lock); |
ed27b9f7 | 3579 | |
d515876e TH |
3580 | /* Advance to the next non-empty css_set */ |
3581 | do { | |
3582 | l = l->next; | |
0f0a2b4f TH |
3583 | if (l == it->cset_head) { |
3584 | it->cset_pos = NULL; | |
ecb9d535 | 3585 | it->task_pos = NULL; |
d515876e TH |
3586 | return; |
3587 | } | |
3ebb2b6e TH |
3588 | |
3589 | if (it->ss) { | |
3590 | cset = container_of(l, struct css_set, | |
3591 | e_cset_node[it->ss->id]); | |
3592 | } else { | |
3593 | link = list_entry(l, struct cgrp_cset_link, cset_link); | |
3594 | cset = link->cset; | |
3595 | } | |
0de0942d | 3596 | } while (!css_set_populated(cset)); |
c7561128 | 3597 | |
0f0a2b4f | 3598 | it->cset_pos = l; |
c7561128 TH |
3599 | |
3600 | if (!list_empty(&cset->tasks)) | |
0f0a2b4f | 3601 | it->task_pos = cset->tasks.next; |
c7561128 | 3602 | else |
0f0a2b4f TH |
3603 | it->task_pos = cset->mg_tasks.next; |
3604 | ||
3605 | it->tasks_head = &cset->tasks; | |
3606 | it->mg_tasks_head = &cset->mg_tasks; | |
ed27b9f7 TH |
3607 | |
3608 | /* | |
3609 | * We don't keep css_sets locked across iteration steps and thus | |
3610 | * need to take steps to ensure that iteration can be resumed after | |
3611 | * the lock is re-acquired. Iteration is performed at two levels - | |
3612 | * css_sets and tasks in them. | |
3613 | * | |
3614 | * Once created, a css_set never leaves its cgroup lists, so a | |
3615 | * pinned css_set is guaranteed to stay put and we can resume | |
3616 | * iteration afterwards. | |
3617 | * | |
3618 | * Tasks may leave @cset across iteration steps. This is resolved | |
3619 | * by registering each iterator with the css_set currently being | |
3620 | * walked and making css_set_move_task() advance iterators whose | |
3621 | * next task is leaving. | |
3622 | */ | |
3623 | if (it->cur_cset) { | |
3624 | list_del(&it->iters_node); | |
3625 | put_css_set_locked(it->cur_cset); | |
3626 | } | |
3627 | get_css_set(cset); | |
3628 | it->cur_cset = cset; | |
3629 | list_add(&it->iters_node, &cset->task_iters); | |
d515876e TH |
3630 | } |
3631 | ||
ecb9d535 TH |
3632 | static void css_task_iter_advance(struct css_task_iter *it) |
3633 | { | |
3634 | struct list_head *l = it->task_pos; | |
3635 | ||
f0d9a5f1 | 3636 | lockdep_assert_held(&css_set_lock); |
ecb9d535 TH |
3637 | WARN_ON_ONCE(!l); |
3638 | ||
3639 | /* | |
3640 | * Advance iterator to find next entry. cset->tasks is consumed | |
3641 | * first and then ->mg_tasks. After ->mg_tasks, we move onto the | |
3642 | * next cset. | |
3643 | */ | |
3644 | l = l->next; | |
3645 | ||
3646 | if (l == it->tasks_head) | |
3647 | l = it->mg_tasks_head->next; | |
3648 | ||
3649 | if (l == it->mg_tasks_head) | |
3650 | css_task_iter_advance_css_set(it); | |
3651 | else | |
3652 | it->task_pos = l; | |
3653 | } | |
3654 | ||
0942eeee | 3655 | /** |
72ec7029 TH |
3656 | * css_task_iter_start - initiate task iteration |
3657 | * @css: the css to walk tasks of | |
0942eeee TH |
3658 | * @it: the task iterator to use |
3659 | * | |
72ec7029 TH |
3660 | * Initiate iteration through the tasks of @css. The caller can call |
3661 | * css_task_iter_next() to walk through the tasks until the function | |
3662 | * returns NULL. On completion of iteration, css_task_iter_end() must be | |
3663 | * called. | |
0942eeee | 3664 | */ |
72ec7029 TH |
3665 | void css_task_iter_start(struct cgroup_subsys_state *css, |
3666 | struct css_task_iter *it) | |
817929ec | 3667 | { |
56fde9e0 TH |
3668 | /* no one should try to iterate before mounting cgroups */ |
3669 | WARN_ON_ONCE(!use_task_css_set_links); | |
31a7df01 | 3670 | |
ed27b9f7 TH |
3671 | memset(it, 0, sizeof(*it)); |
3672 | ||
82d6489d | 3673 | spin_lock_irq(&css_set_lock); |
c59cd3d8 | 3674 | |
3ebb2b6e TH |
3675 | it->ss = css->ss; |
3676 | ||
3677 | if (it->ss) | |
3678 | it->cset_pos = &css->cgroup->e_csets[css->ss->id]; | |
3679 | else | |
3680 | it->cset_pos = &css->cgroup->cset_links; | |
3681 | ||
0f0a2b4f | 3682 | it->cset_head = it->cset_pos; |
c59cd3d8 | 3683 | |
ecb9d535 | 3684 | css_task_iter_advance_css_set(it); |
ed27b9f7 | 3685 | |
82d6489d | 3686 | spin_unlock_irq(&css_set_lock); |
817929ec PM |
3687 | } |
3688 | ||
0942eeee | 3689 | /** |
72ec7029 | 3690 | * css_task_iter_next - return the next task for the iterator |
0942eeee TH |
3691 | * @it: the task iterator being iterated |
3692 | * | |
3693 | * The "next" function for task iteration. @it should have been | |
72ec7029 TH |
3694 | * initialized via css_task_iter_start(). Returns NULL when the iteration |
3695 | * reaches the end. | |
0942eeee | 3696 | */ |
72ec7029 | 3697 | struct task_struct *css_task_iter_next(struct css_task_iter *it) |
817929ec | 3698 | { |
d5745675 | 3699 | if (it->cur_task) { |
ed27b9f7 | 3700 | put_task_struct(it->cur_task); |
d5745675 TH |
3701 | it->cur_task = NULL; |
3702 | } | |
ed27b9f7 | 3703 | |
82d6489d | 3704 | spin_lock_irq(&css_set_lock); |
ed27b9f7 | 3705 | |
d5745675 TH |
3706 | if (it->task_pos) { |
3707 | it->cur_task = list_entry(it->task_pos, struct task_struct, | |
3708 | cg_list); | |
3709 | get_task_struct(it->cur_task); | |
3710 | css_task_iter_advance(it); | |
3711 | } | |
ed27b9f7 | 3712 | |
82d6489d | 3713 | spin_unlock_irq(&css_set_lock); |
ed27b9f7 TH |
3714 | |
3715 | return it->cur_task; | |
817929ec PM |
3716 | } |
3717 | ||
0942eeee | 3718 | /** |
72ec7029 | 3719 | * css_task_iter_end - finish task iteration |
0942eeee TH |
3720 | * @it: the task iterator to finish |
3721 | * | |
72ec7029 | 3722 | * Finish task iteration started by css_task_iter_start(). |
0942eeee | 3723 | */ |
72ec7029 | 3724 | void css_task_iter_end(struct css_task_iter *it) |
31a7df01 | 3725 | { |
ed27b9f7 | 3726 | if (it->cur_cset) { |
82d6489d | 3727 | spin_lock_irq(&css_set_lock); |
ed27b9f7 TH |
3728 | list_del(&it->iters_node); |
3729 | put_css_set_locked(it->cur_cset); | |
82d6489d | 3730 | spin_unlock_irq(&css_set_lock); |
ed27b9f7 TH |
3731 | } |
3732 | ||
3733 | if (it->cur_task) | |
3734 | put_task_struct(it->cur_task); | |
31a7df01 CW |
3735 | } |
3736 | ||
b4b90a8e | 3737 | static void cgroup_procs_release(struct kernfs_open_file *of) |
31a7df01 | 3738 | { |
b4b90a8e TH |
3739 | if (of->priv) { |
3740 | css_task_iter_end(of->priv); | |
3741 | kfree(of->priv); | |
3742 | } | |
3743 | } | |
6c694c88 | 3744 | |
b4b90a8e TH |
3745 | static void *cgroup_procs_next(struct seq_file *s, void *v, loff_t *pos) |
3746 | { | |
3747 | struct kernfs_open_file *of = s->private; | |
3748 | struct css_task_iter *it = of->priv; | |
3749 | struct task_struct *task; | |
31a7df01 | 3750 | |
b4b90a8e TH |
3751 | do { |
3752 | task = css_task_iter_next(it); | |
3753 | } while (task && !thread_group_leader(task)); | |
eedd0f4c | 3754 | |
b4b90a8e TH |
3755 | return task; |
3756 | } | |
31a7df01 | 3757 | |
b4b90a8e TH |
3758 | static void *cgroup_procs_start(struct seq_file *s, loff_t *pos) |
3759 | { | |
3760 | struct kernfs_open_file *of = s->private; | |
3761 | struct cgroup *cgrp = seq_css(s)->cgroup; | |
3762 | struct css_task_iter *it = of->priv; | |
8cc99345 | 3763 | |
952aaa12 | 3764 | /* |
b4b90a8e TH |
3765 | * When a seq_file is seeked, it's always traversed sequentially |
3766 | * from position 0, so we can simply keep iterating on !0 *pos. | |
952aaa12 | 3767 | */ |
b4b90a8e TH |
3768 | if (!it) { |
3769 | if (WARN_ON_ONCE((*pos)++)) | |
3770 | return ERR_PTR(-EINVAL); | |
8cc99345 | 3771 | |
b4b90a8e TH |
3772 | it = kzalloc(sizeof(*it), GFP_KERNEL); |
3773 | if (!it) | |
3774 | return ERR_PTR(-ENOMEM); | |
3775 | of->priv = it; | |
3776 | css_task_iter_start(&cgrp->self, it); | |
3777 | } else if (!(*pos)++) { | |
3778 | css_task_iter_end(it); | |
3779 | css_task_iter_start(&cgrp->self, it); | |
3780 | } | |
bbcb81d0 | 3781 | |
b4b90a8e TH |
3782 | return cgroup_procs_next(s, NULL, NULL); |
3783 | } | |
24528255 | 3784 | |
b4b90a8e | 3785 | static int cgroup_procs_show(struct seq_file *s, void *v) |
bbcb81d0 | 3786 | { |
b4b90a8e | 3787 | seq_printf(s, "%d\n", task_tgid_vnr(v)); |
97978e6d DL |
3788 | return 0; |
3789 | } | |
3790 | ||
a14c6874 | 3791 | /* cgroup core interface files for the default hierarchy */ |
d62beb7f | 3792 | static struct cftype cgroup_base_files[] = { |
81a6a5cd | 3793 | { |
d5c56ced | 3794 | .name = "cgroup.procs", |
6f60eade | 3795 | .file_offset = offsetof(struct cgroup, procs_file), |
b4b90a8e TH |
3796 | .release = cgroup_procs_release, |
3797 | .seq_start = cgroup_procs_start, | |
3798 | .seq_next = cgroup_procs_next, | |
3799 | .seq_show = cgroup_procs_show, | |
acbef755 | 3800 | .write = cgroup_procs_write, |
102a775e | 3801 | }, |
f8f22e53 TH |
3802 | { |
3803 | .name = "cgroup.controllers", | |
f8f22e53 TH |
3804 | .seq_show = cgroup_controllers_show, |
3805 | }, | |
3806 | { | |
3807 | .name = "cgroup.subtree_control", | |
f8f22e53 | 3808 | .seq_show = cgroup_subtree_control_show, |
451af504 | 3809 | .write = cgroup_subtree_control_write, |
f8f22e53 | 3810 | }, |
842b597e | 3811 | { |
4a07c222 | 3812 | .name = "cgroup.events", |
a14c6874 | 3813 | .flags = CFTYPE_NOT_ON_ROOT, |
6f60eade | 3814 | .file_offset = offsetof(struct cgroup, events_file), |
4a07c222 | 3815 | .seq_show = cgroup_events_show, |
842b597e | 3816 | }, |
a14c6874 TH |
3817 | { } /* terminate */ |
3818 | }; | |
d5c56ced | 3819 | |
0c21ead1 TH |
3820 | /* |
3821 | * css destruction is four-stage process. | |
3822 | * | |
3823 | * 1. Destruction starts. Killing of the percpu_ref is initiated. | |
3824 | * Implemented in kill_css(). | |
3825 | * | |
3826 | * 2. When the percpu_ref is confirmed to be visible as killed on all CPUs | |
ec903c0c TH |
3827 | * and thus css_tryget_online() is guaranteed to fail, the css can be |
3828 | * offlined by invoking offline_css(). After offlining, the base ref is | |
3829 | * put. Implemented in css_killed_work_fn(). | |
0c21ead1 TH |
3830 | * |
3831 | * 3. When the percpu_ref reaches zero, the only possible remaining | |
3832 | * accessors are inside RCU read sections. css_release() schedules the | |
3833 | * RCU callback. | |
3834 | * | |
3835 | * 4. After the grace period, the css can be freed. Implemented in | |
3836 | * css_free_work_fn(). | |
3837 | * | |
3838 | * It is actually hairier because both step 2 and 4 require process context | |
3839 | * and thus involve punting to css->destroy_work adding two additional | |
3840 | * steps to the already complex sequence. | |
3841 | */ | |
35ef10da | 3842 | static void css_free_work_fn(struct work_struct *work) |
48ddbe19 TH |
3843 | { |
3844 | struct cgroup_subsys_state *css = | |
35ef10da | 3845 | container_of(work, struct cgroup_subsys_state, destroy_work); |
01e58659 | 3846 | struct cgroup_subsys *ss = css->ss; |
0c21ead1 | 3847 | struct cgroup *cgrp = css->cgroup; |
48ddbe19 | 3848 | |
9a1049da TH |
3849 | percpu_ref_exit(&css->refcnt); |
3850 | ||
01e58659 | 3851 | if (ss) { |
9d755d33 | 3852 | /* css free path */ |
8bb5ef79 | 3853 | struct cgroup_subsys_state *parent = css->parent; |
01e58659 VD |
3854 | int id = css->id; |
3855 | ||
01e58659 VD |
3856 | ss->css_free(css); |
3857 | cgroup_idr_remove(&ss->css_idr, id); | |
9d755d33 | 3858 | cgroup_put(cgrp); |
8bb5ef79 TH |
3859 | |
3860 | if (parent) | |
3861 | css_put(parent); | |
9d755d33 TH |
3862 | } else { |
3863 | /* cgroup free path */ | |
3864 | atomic_dec(&cgrp->root->nr_cgrps); | |
d62beb7f | 3865 | cgroup1_pidlist_destroy_all(cgrp); |
971ff493 | 3866 | cancel_work_sync(&cgrp->release_agent_work); |
9d755d33 | 3867 | |
d51f39b0 | 3868 | if (cgroup_parent(cgrp)) { |
9d755d33 TH |
3869 | /* |
3870 | * We get a ref to the parent, and put the ref when | |
3871 | * this cgroup is being freed, so it's guaranteed | |
3872 | * that the parent won't be destroyed before its | |
3873 | * children. | |
3874 | */ | |
d51f39b0 | 3875 | cgroup_put(cgroup_parent(cgrp)); |
9d755d33 TH |
3876 | kernfs_put(cgrp->kn); |
3877 | kfree(cgrp); | |
3878 | } else { | |
3879 | /* | |
3880 | * This is root cgroup's refcnt reaching zero, | |
3881 | * which indicates that the root should be | |
3882 | * released. | |
3883 | */ | |
3884 | cgroup_destroy_root(cgrp->root); | |
3885 | } | |
3886 | } | |
48ddbe19 TH |
3887 | } |
3888 | ||
0c21ead1 | 3889 | static void css_free_rcu_fn(struct rcu_head *rcu_head) |
d3daf28d TH |
3890 | { |
3891 | struct cgroup_subsys_state *css = | |
0c21ead1 | 3892 | container_of(rcu_head, struct cgroup_subsys_state, rcu_head); |
d3daf28d | 3893 | |
35ef10da | 3894 | INIT_WORK(&css->destroy_work, css_free_work_fn); |
e5fca243 | 3895 | queue_work(cgroup_destroy_wq, &css->destroy_work); |
48ddbe19 TH |
3896 | } |
3897 | ||
25e15d83 | 3898 | static void css_release_work_fn(struct work_struct *work) |
d3daf28d TH |
3899 | { |
3900 | struct cgroup_subsys_state *css = | |
25e15d83 | 3901 | container_of(work, struct cgroup_subsys_state, destroy_work); |
15a4c835 | 3902 | struct cgroup_subsys *ss = css->ss; |
9d755d33 | 3903 | struct cgroup *cgrp = css->cgroup; |
15a4c835 | 3904 | |
1fed1b2e TH |
3905 | mutex_lock(&cgroup_mutex); |
3906 | ||
de3f0341 | 3907 | css->flags |= CSS_RELEASED; |
1fed1b2e TH |
3908 | list_del_rcu(&css->sibling); |
3909 | ||
9d755d33 TH |
3910 | if (ss) { |
3911 | /* css release path */ | |
01e58659 | 3912 | cgroup_idr_replace(&ss->css_idr, NULL, css->id); |
7d172cc8 TH |
3913 | if (ss->css_released) |
3914 | ss->css_released(css); | |
9d755d33 TH |
3915 | } else { |
3916 | /* cgroup release path */ | |
ed1777de TH |
3917 | trace_cgroup_release(cgrp); |
3918 | ||
9d755d33 TH |
3919 | cgroup_idr_remove(&cgrp->root->cgroup_idr, cgrp->id); |
3920 | cgrp->id = -1; | |
a4189487 LZ |
3921 | |
3922 | /* | |
3923 | * There are two control paths which try to determine | |
3924 | * cgroup from dentry without going through kernfs - | |
3925 | * cgroupstats_build() and css_tryget_online_from_dir(). | |
3926 | * Those are supported by RCU protecting clearing of | |
3927 | * cgrp->kn->priv backpointer. | |
3928 | */ | |
6cd0f5bb TH |
3929 | if (cgrp->kn) |
3930 | RCU_INIT_POINTER(*(void __rcu __force **)&cgrp->kn->priv, | |
3931 | NULL); | |
30070984 DM |
3932 | |
3933 | cgroup_bpf_put(cgrp); | |
9d755d33 | 3934 | } |
d3daf28d | 3935 | |
1fed1b2e TH |
3936 | mutex_unlock(&cgroup_mutex); |
3937 | ||
0c21ead1 | 3938 | call_rcu(&css->rcu_head, css_free_rcu_fn); |
d3daf28d TH |
3939 | } |
3940 | ||
d3daf28d TH |
3941 | static void css_release(struct percpu_ref *ref) |
3942 | { | |
3943 | struct cgroup_subsys_state *css = | |
3944 | container_of(ref, struct cgroup_subsys_state, refcnt); | |
3945 | ||
25e15d83 TH |
3946 | INIT_WORK(&css->destroy_work, css_release_work_fn); |
3947 | queue_work(cgroup_destroy_wq, &css->destroy_work); | |
d3daf28d TH |
3948 | } |
3949 | ||
ddfcadab TH |
3950 | static void init_and_link_css(struct cgroup_subsys_state *css, |
3951 | struct cgroup_subsys *ss, struct cgroup *cgrp) | |
ddbcc7e8 | 3952 | { |
0cb51d71 TH |
3953 | lockdep_assert_held(&cgroup_mutex); |
3954 | ||
a590b90d | 3955 | cgroup_get_live(cgrp); |
ddfcadab | 3956 | |
d5c419b6 | 3957 | memset(css, 0, sizeof(*css)); |
bd89aabc | 3958 | css->cgroup = cgrp; |
72c97e54 | 3959 | css->ss = ss; |
8fa3b8d6 | 3960 | css->id = -1; |
d5c419b6 TH |
3961 | INIT_LIST_HEAD(&css->sibling); |
3962 | INIT_LIST_HEAD(&css->children); | |
0cb51d71 | 3963 | css->serial_nr = css_serial_nr_next++; |
aa226ff4 | 3964 | atomic_set(&css->online_cnt, 0); |
0ae78e0b | 3965 | |
d51f39b0 TH |
3966 | if (cgroup_parent(cgrp)) { |
3967 | css->parent = cgroup_css(cgroup_parent(cgrp), ss); | |
ddfcadab | 3968 | css_get(css->parent); |
ddfcadab | 3969 | } |
48ddbe19 | 3970 | |
ca8bdcaf | 3971 | BUG_ON(cgroup_css(cgrp, ss)); |
ddbcc7e8 PM |
3972 | } |
3973 | ||
2a4ac633 | 3974 | /* invoke ->css_online() on a new CSS and mark it online if successful */ |
623f926b | 3975 | static int online_css(struct cgroup_subsys_state *css) |
a31f2d3f | 3976 | { |
623f926b | 3977 | struct cgroup_subsys *ss = css->ss; |
b1929db4 TH |
3978 | int ret = 0; |
3979 | ||
a31f2d3f TH |
3980 | lockdep_assert_held(&cgroup_mutex); |
3981 | ||
92fb9748 | 3982 | if (ss->css_online) |
eb95419b | 3983 | ret = ss->css_online(css); |
ae7f164a | 3984 | if (!ret) { |
eb95419b | 3985 | css->flags |= CSS_ONLINE; |
aec25020 | 3986 | rcu_assign_pointer(css->cgroup->subsys[ss->id], css); |
aa226ff4 TH |
3987 | |
3988 | atomic_inc(&css->online_cnt); | |
3989 | if (css->parent) | |
3990 | atomic_inc(&css->parent->online_cnt); | |
ae7f164a | 3991 | } |
b1929db4 | 3992 | return ret; |
a31f2d3f TH |
3993 | } |
3994 | ||
2a4ac633 | 3995 | /* if the CSS is online, invoke ->css_offline() on it and mark it offline */ |
623f926b | 3996 | static void offline_css(struct cgroup_subsys_state *css) |
a31f2d3f | 3997 | { |
623f926b | 3998 | struct cgroup_subsys *ss = css->ss; |
a31f2d3f TH |
3999 | |
4000 | lockdep_assert_held(&cgroup_mutex); | |
4001 | ||
4002 | if (!(css->flags & CSS_ONLINE)) | |
4003 | return; | |
4004 | ||
fa06235b VD |
4005 | if (ss->css_reset) |
4006 | ss->css_reset(css); | |
4007 | ||
d7eeac19 | 4008 | if (ss->css_offline) |
eb95419b | 4009 | ss->css_offline(css); |
a31f2d3f | 4010 | |
eb95419b | 4011 | css->flags &= ~CSS_ONLINE; |
e3297803 | 4012 | RCU_INIT_POINTER(css->cgroup->subsys[ss->id], NULL); |
f8f22e53 TH |
4013 | |
4014 | wake_up_all(&css->cgroup->offline_waitq); | |
a31f2d3f TH |
4015 | } |
4016 | ||
c81c925a | 4017 | /** |
6cd0f5bb | 4018 | * css_create - create a cgroup_subsys_state |
c81c925a TH |
4019 | * @cgrp: the cgroup new css will be associated with |
4020 | * @ss: the subsys of new css | |
4021 | * | |
4022 | * Create a new css associated with @cgrp - @ss pair. On success, the new | |
6cd0f5bb TH |
4023 | * css is online and installed in @cgrp. This function doesn't create the |
4024 | * interface files. Returns 0 on success, -errno on failure. | |
c81c925a | 4025 | */ |
6cd0f5bb TH |
4026 | static struct cgroup_subsys_state *css_create(struct cgroup *cgrp, |
4027 | struct cgroup_subsys *ss) | |
c81c925a | 4028 | { |
d51f39b0 | 4029 | struct cgroup *parent = cgroup_parent(cgrp); |
1fed1b2e | 4030 | struct cgroup_subsys_state *parent_css = cgroup_css(parent, ss); |
c81c925a TH |
4031 | struct cgroup_subsys_state *css; |
4032 | int err; | |
4033 | ||
c81c925a TH |
4034 | lockdep_assert_held(&cgroup_mutex); |
4035 | ||
1fed1b2e | 4036 | css = ss->css_alloc(parent_css); |
e7e15b87 TH |
4037 | if (!css) |
4038 | css = ERR_PTR(-ENOMEM); | |
c81c925a | 4039 | if (IS_ERR(css)) |
6cd0f5bb | 4040 | return css; |
c81c925a | 4041 | |
ddfcadab | 4042 | init_and_link_css(css, ss, cgrp); |
a2bed820 | 4043 | |
2aad2a86 | 4044 | err = percpu_ref_init(&css->refcnt, css_release, 0, GFP_KERNEL); |
c81c925a | 4045 | if (err) |
3eb59ec6 | 4046 | goto err_free_css; |
c81c925a | 4047 | |
cf780b7d | 4048 | err = cgroup_idr_alloc(&ss->css_idr, NULL, 2, 0, GFP_KERNEL); |
15a4c835 | 4049 | if (err < 0) |
b00c52da | 4050 | goto err_free_css; |
15a4c835 | 4051 | css->id = err; |
c81c925a | 4052 | |
15a4c835 | 4053 | /* @css is ready to be brought online now, make it visible */ |
1fed1b2e | 4054 | list_add_tail_rcu(&css->sibling, &parent_css->children); |
15a4c835 | 4055 | cgroup_idr_replace(&ss->css_idr, css, css->id); |
c81c925a TH |
4056 | |
4057 | err = online_css(css); | |
4058 | if (err) | |
1fed1b2e | 4059 | goto err_list_del; |
94419627 | 4060 | |
c81c925a | 4061 | if (ss->broken_hierarchy && !ss->warned_broken_hierarchy && |
d51f39b0 | 4062 | cgroup_parent(parent)) { |
ed3d261b | 4063 | 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 | 4064 | current->comm, current->pid, ss->name); |
c81c925a | 4065 | if (!strcmp(ss->name, "memory")) |
ed3d261b | 4066 | pr_warn("\"memory\" requires setting use_hierarchy to 1 on the root\n"); |
c81c925a TH |
4067 | ss->warned_broken_hierarchy = true; |
4068 | } | |
4069 | ||
6cd0f5bb | 4070 | return css; |
c81c925a | 4071 | |
1fed1b2e TH |
4072 | err_list_del: |
4073 | list_del_rcu(&css->sibling); | |
3eb59ec6 | 4074 | err_free_css: |
a2bed820 | 4075 | call_rcu(&css->rcu_head, css_free_rcu_fn); |
6cd0f5bb | 4076 | return ERR_PTR(err); |
c81c925a TH |
4077 | } |
4078 | ||
07cd1294 TH |
4079 | /* |
4080 | * The returned cgroup is fully initialized including its control mask, but | |
4081 | * it isn't associated with its kernfs_node and doesn't have the control | |
4082 | * mask applied. | |
4083 | */ | |
a5bca215 | 4084 | static struct cgroup *cgroup_create(struct cgroup *parent) |
ddbcc7e8 | 4085 | { |
a5bca215 | 4086 | struct cgroup_root *root = parent->root; |
a5bca215 TH |
4087 | struct cgroup *cgrp, *tcgrp; |
4088 | int level = parent->level + 1; | |
03970d3c | 4089 | int ret; |
ddbcc7e8 | 4090 | |
0a950f65 | 4091 | /* allocate the cgroup and its ID, 0 is reserved for the root */ |
b11cfb58 TH |
4092 | cgrp = kzalloc(sizeof(*cgrp) + |
4093 | sizeof(cgrp->ancestor_ids[0]) * (level + 1), GFP_KERNEL); | |
a5bca215 TH |
4094 | if (!cgrp) |
4095 | return ERR_PTR(-ENOMEM); | |
0ab02ca8 | 4096 | |
2aad2a86 | 4097 | ret = percpu_ref_init(&cgrp->self.refcnt, css_release, 0, GFP_KERNEL); |
9d755d33 TH |
4098 | if (ret) |
4099 | goto out_free_cgrp; | |
4100 | ||
0ab02ca8 LZ |
4101 | /* |
4102 | * Temporarily set the pointer to NULL, so idr_find() won't return | |
4103 | * a half-baked cgroup. | |
4104 | */ | |
cf780b7d | 4105 | cgrp->id = cgroup_idr_alloc(&root->cgroup_idr, NULL, 2, 0, GFP_KERNEL); |
0ab02ca8 | 4106 | if (cgrp->id < 0) { |
ba0f4d76 | 4107 | ret = -ENOMEM; |
9d755d33 | 4108 | goto out_cancel_ref; |
976c06bc TH |
4109 | } |
4110 | ||
cc31edce | 4111 | init_cgroup_housekeeping(cgrp); |
ddbcc7e8 | 4112 | |
9d800df1 | 4113 | cgrp->self.parent = &parent->self; |
ba0f4d76 | 4114 | cgrp->root = root; |
b11cfb58 TH |
4115 | cgrp->level = level; |
4116 | ||
4117 | for (tcgrp = cgrp; tcgrp; tcgrp = cgroup_parent(tcgrp)) | |
4118 | cgrp->ancestor_ids[tcgrp->level] = tcgrp->id; | |
ddbcc7e8 | 4119 | |
b6abdb0e LZ |
4120 | if (notify_on_release(parent)) |
4121 | set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); | |
4122 | ||
2260e7fc TH |
4123 | if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &parent->flags)) |
4124 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags); | |
97978e6d | 4125 | |
0cb51d71 | 4126 | cgrp->self.serial_nr = css_serial_nr_next++; |
53fa5261 | 4127 | |
4e139afc | 4128 | /* allocation complete, commit to creation */ |
d5c419b6 | 4129 | list_add_tail_rcu(&cgrp->self.sibling, &cgroup_parent(cgrp)->self.children); |
3c9c825b | 4130 | atomic_inc(&root->nr_cgrps); |
a590b90d | 4131 | cgroup_get_live(parent); |
415cf07a | 4132 | |
0d80255e TH |
4133 | /* |
4134 | * @cgrp is now fully operational. If something fails after this | |
4135 | * point, it'll be released via the normal destruction path. | |
4136 | */ | |
6fa4918d | 4137 | cgroup_idr_replace(&root->cgroup_idr, cgrp, cgrp->id); |
4e96ee8e | 4138 | |
bd53d617 TH |
4139 | /* |
4140 | * On the default hierarchy, a child doesn't automatically inherit | |
667c2491 | 4141 | * subtree_control from the parent. Each is configured manually. |
bd53d617 | 4142 | */ |
03970d3c | 4143 | if (!cgroup_on_dfl(cgrp)) |
5531dc91 | 4144 | cgrp->subtree_control = cgroup_control(cgrp); |
03970d3c | 4145 | |
30070984 DM |
4146 | if (parent) |
4147 | cgroup_bpf_inherit(cgrp, parent); | |
4148 | ||
03970d3c TH |
4149 | cgroup_propagate_control(cgrp); |
4150 | ||
a5bca215 TH |
4151 | return cgrp; |
4152 | ||
4153 | out_cancel_ref: | |
4154 | percpu_ref_exit(&cgrp->self.refcnt); | |
4155 | out_free_cgrp: | |
4156 | kfree(cgrp); | |
4157 | return ERR_PTR(ret); | |
a5bca215 TH |
4158 | } |
4159 | ||
1592c9b2 | 4160 | int cgroup_mkdir(struct kernfs_node *parent_kn, const char *name, umode_t mode) |
a5bca215 TH |
4161 | { |
4162 | struct cgroup *parent, *cgrp; | |
a5bca215 | 4163 | struct kernfs_node *kn; |
03970d3c | 4164 | int ret; |
a5bca215 TH |
4165 | |
4166 | /* do not accept '\n' to prevent making /proc/<pid>/cgroup unparsable */ | |
4167 | if (strchr(name, '\n')) | |
4168 | return -EINVAL; | |
4169 | ||
945ba199 | 4170 | parent = cgroup_kn_lock_live(parent_kn, false); |
a5bca215 TH |
4171 | if (!parent) |
4172 | return -ENODEV; | |
4173 | ||
4174 | cgrp = cgroup_create(parent); | |
4175 | if (IS_ERR(cgrp)) { | |
4176 | ret = PTR_ERR(cgrp); | |
4177 | goto out_unlock; | |
4178 | } | |
4179 | ||
195e9b6c TH |
4180 | /* create the directory */ |
4181 | kn = kernfs_create_dir(parent->kn, name, mode, cgrp); | |
4182 | if (IS_ERR(kn)) { | |
4183 | ret = PTR_ERR(kn); | |
4184 | goto out_destroy; | |
4185 | } | |
4186 | cgrp->kn = kn; | |
4187 | ||
4188 | /* | |
4189 | * This extra ref will be put in cgroup_free_fn() and guarantees | |
4190 | * that @cgrp->kn is always accessible. | |
4191 | */ | |
4192 | kernfs_get(kn); | |
4193 | ||
4194 | ret = cgroup_kn_set_ugid(kn); | |
4195 | if (ret) | |
4196 | goto out_destroy; | |
4197 | ||
334c3679 | 4198 | ret = css_populate_dir(&cgrp->self); |
195e9b6c TH |
4199 | if (ret) |
4200 | goto out_destroy; | |
4201 | ||
03970d3c TH |
4202 | ret = cgroup_apply_control_enable(cgrp); |
4203 | if (ret) | |
4204 | goto out_destroy; | |
195e9b6c | 4205 | |
ed1777de TH |
4206 | trace_cgroup_mkdir(cgrp); |
4207 | ||
195e9b6c | 4208 | /* let's create and online css's */ |
2bd59d48 | 4209 | kernfs_activate(kn); |
ddbcc7e8 | 4210 | |
ba0f4d76 TH |
4211 | ret = 0; |
4212 | goto out_unlock; | |
ddbcc7e8 | 4213 | |
a5bca215 TH |
4214 | out_destroy: |
4215 | cgroup_destroy_locked(cgrp); | |
ba0f4d76 | 4216 | out_unlock: |
a9746d8d | 4217 | cgroup_kn_unlock(parent_kn); |
ba0f4d76 | 4218 | return ret; |
ddbcc7e8 PM |
4219 | } |
4220 | ||
223dbc38 TH |
4221 | /* |
4222 | * This is called when the refcnt of a css is confirmed to be killed. | |
249f3468 TH |
4223 | * css_tryget_online() is now guaranteed to fail. Tell the subsystem to |
4224 | * initate destruction and put the css ref from kill_css(). | |
223dbc38 TH |
4225 | */ |
4226 | static void css_killed_work_fn(struct work_struct *work) | |
d3daf28d | 4227 | { |
223dbc38 TH |
4228 | struct cgroup_subsys_state *css = |
4229 | container_of(work, struct cgroup_subsys_state, destroy_work); | |
d3daf28d | 4230 | |
f20104de | 4231 | mutex_lock(&cgroup_mutex); |
09a503ea | 4232 | |
aa226ff4 TH |
4233 | do { |
4234 | offline_css(css); | |
4235 | css_put(css); | |
4236 | /* @css can't go away while we're holding cgroup_mutex */ | |
4237 | css = css->parent; | |
4238 | } while (css && atomic_dec_and_test(&css->online_cnt)); | |
4239 | ||
4240 | mutex_unlock(&cgroup_mutex); | |
d3daf28d TH |
4241 | } |
4242 | ||
223dbc38 TH |
4243 | /* css kill confirmation processing requires process context, bounce */ |
4244 | static void css_killed_ref_fn(struct percpu_ref *ref) | |
d3daf28d TH |
4245 | { |
4246 | struct cgroup_subsys_state *css = | |
4247 | container_of(ref, struct cgroup_subsys_state, refcnt); | |
4248 | ||
aa226ff4 TH |
4249 | if (atomic_dec_and_test(&css->online_cnt)) { |
4250 | INIT_WORK(&css->destroy_work, css_killed_work_fn); | |
4251 | queue_work(cgroup_destroy_wq, &css->destroy_work); | |
4252 | } | |
d3daf28d TH |
4253 | } |
4254 | ||
f392e51c TH |
4255 | /** |
4256 | * kill_css - destroy a css | |
4257 | * @css: css to destroy | |
4258 | * | |
4259 | * This function initiates destruction of @css by removing cgroup interface | |
4260 | * files and putting its base reference. ->css_offline() will be invoked | |
ec903c0c TH |
4261 | * asynchronously once css_tryget_online() is guaranteed to fail and when |
4262 | * the reference count reaches zero, @css will be released. | |
f392e51c TH |
4263 | */ |
4264 | static void kill_css(struct cgroup_subsys_state *css) | |
edae0c33 | 4265 | { |
01f6474c | 4266 | lockdep_assert_held(&cgroup_mutex); |
94419627 | 4267 | |
2bd59d48 TH |
4268 | /* |
4269 | * This must happen before css is disassociated with its cgroup. | |
4270 | * See seq_css() for details. | |
4271 | */ | |
334c3679 | 4272 | css_clear_dir(css); |
3c14f8b4 | 4273 | |
edae0c33 TH |
4274 | /* |
4275 | * Killing would put the base ref, but we need to keep it alive | |
4276 | * until after ->css_offline(). | |
4277 | */ | |
4278 | css_get(css); | |
4279 | ||
4280 | /* | |
4281 | * cgroup core guarantees that, by the time ->css_offline() is | |
4282 | * invoked, no new css reference will be given out via | |
ec903c0c | 4283 | * css_tryget_online(). We can't simply call percpu_ref_kill() and |
edae0c33 TH |
4284 | * proceed to offlining css's because percpu_ref_kill() doesn't |
4285 | * guarantee that the ref is seen as killed on all CPUs on return. | |
4286 | * | |
4287 | * Use percpu_ref_kill_and_confirm() to get notifications as each | |
4288 | * css is confirmed to be seen as killed on all CPUs. | |
4289 | */ | |
4290 | percpu_ref_kill_and_confirm(&css->refcnt, css_killed_ref_fn); | |
d3daf28d TH |
4291 | } |
4292 | ||
4293 | /** | |
4294 | * cgroup_destroy_locked - the first stage of cgroup destruction | |
4295 | * @cgrp: cgroup to be destroyed | |
4296 | * | |
4297 | * css's make use of percpu refcnts whose killing latency shouldn't be | |
4298 | * exposed to userland and are RCU protected. Also, cgroup core needs to | |
ec903c0c TH |
4299 | * guarantee that css_tryget_online() won't succeed by the time |
4300 | * ->css_offline() is invoked. To satisfy all the requirements, | |
4301 | * destruction is implemented in the following two steps. | |
d3daf28d TH |
4302 | * |
4303 | * s1. Verify @cgrp can be destroyed and mark it dying. Remove all | |
4304 | * userland visible parts and start killing the percpu refcnts of | |
4305 | * css's. Set up so that the next stage will be kicked off once all | |
4306 | * the percpu refcnts are confirmed to be killed. | |
4307 | * | |
4308 | * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the | |
4309 | * rest of destruction. Once all cgroup references are gone, the | |
4310 | * cgroup is RCU-freed. | |
4311 | * | |
4312 | * This function implements s1. After this step, @cgrp is gone as far as | |
4313 | * the userland is concerned and a new cgroup with the same name may be | |
4314 | * created. As cgroup doesn't care about the names internally, this | |
4315 | * doesn't cause any problem. | |
4316 | */ | |
42809dd4 TH |
4317 | static int cgroup_destroy_locked(struct cgroup *cgrp) |
4318 | __releases(&cgroup_mutex) __acquires(&cgroup_mutex) | |
ddbcc7e8 | 4319 | { |
2bd59d48 | 4320 | struct cgroup_subsys_state *css; |
2b021cbf | 4321 | struct cgrp_cset_link *link; |
1c6727af | 4322 | int ssid; |
ddbcc7e8 | 4323 | |
42809dd4 TH |
4324 | lockdep_assert_held(&cgroup_mutex); |
4325 | ||
91486f61 TH |
4326 | /* |
4327 | * Only migration can raise populated from zero and we're already | |
4328 | * holding cgroup_mutex. | |
4329 | */ | |
4330 | if (cgroup_is_populated(cgrp)) | |
ddbcc7e8 | 4331 | return -EBUSY; |
a043e3b2 | 4332 | |
bb78a92f | 4333 | /* |
d5c419b6 TH |
4334 | * Make sure there's no live children. We can't test emptiness of |
4335 | * ->self.children as dead children linger on it while being | |
4336 | * drained; otherwise, "rmdir parent/child parent" may fail. | |
bb78a92f | 4337 | */ |
f3d46500 | 4338 | if (css_has_online_children(&cgrp->self)) |
bb78a92f HD |
4339 | return -EBUSY; |
4340 | ||
455050d2 | 4341 | /* |
2b021cbf TH |
4342 | * Mark @cgrp and the associated csets dead. The former prevents |
4343 | * further task migration and child creation by disabling | |
4344 | * cgroup_lock_live_group(). The latter makes the csets ignored by | |
4345 | * the migration path. | |
455050d2 | 4346 | */ |
184faf32 | 4347 | cgrp->self.flags &= ~CSS_ONLINE; |
ddbcc7e8 | 4348 | |
82d6489d | 4349 | spin_lock_irq(&css_set_lock); |
2b021cbf TH |
4350 | list_for_each_entry(link, &cgrp->cset_links, cset_link) |
4351 | link->cset->dead = true; | |
82d6489d | 4352 | spin_unlock_irq(&css_set_lock); |
2b021cbf | 4353 | |
249f3468 | 4354 | /* initiate massacre of all css's */ |
1c6727af TH |
4355 | for_each_css(css, ssid, cgrp) |
4356 | kill_css(css); | |
455050d2 | 4357 | |
455050d2 | 4358 | /* |
01f6474c TH |
4359 | * Remove @cgrp directory along with the base files. @cgrp has an |
4360 | * extra ref on its kn. | |
f20104de | 4361 | */ |
01f6474c | 4362 | kernfs_remove(cgrp->kn); |
f20104de | 4363 | |
d62beb7f | 4364 | cgroup1_check_for_release(cgroup_parent(cgrp)); |
2bd59d48 | 4365 | |
249f3468 | 4366 | /* put the base reference */ |
9d755d33 | 4367 | percpu_ref_kill(&cgrp->self.refcnt); |
455050d2 | 4368 | |
ea15f8cc TH |
4369 | return 0; |
4370 | }; | |
4371 | ||
1592c9b2 | 4372 | int cgroup_rmdir(struct kernfs_node *kn) |
42809dd4 | 4373 | { |
a9746d8d | 4374 | struct cgroup *cgrp; |
2bd59d48 | 4375 | int ret = 0; |
42809dd4 | 4376 | |
945ba199 | 4377 | cgrp = cgroup_kn_lock_live(kn, false); |
a9746d8d TH |
4378 | if (!cgrp) |
4379 | return 0; | |
42809dd4 | 4380 | |
a9746d8d | 4381 | ret = cgroup_destroy_locked(cgrp); |
2bb566cb | 4382 | |
ed1777de TH |
4383 | if (!ret) |
4384 | trace_cgroup_rmdir(cgrp); | |
4385 | ||
a9746d8d | 4386 | cgroup_kn_unlock(kn); |
42809dd4 | 4387 | return ret; |
8e3f6541 TH |
4388 | } |
4389 | ||
2bd59d48 TH |
4390 | static struct kernfs_syscall_ops cgroup_kf_syscall_ops = { |
4391 | .remount_fs = cgroup_remount, | |
2bd59d48 TH |
4392 | .mkdir = cgroup_mkdir, |
4393 | .rmdir = cgroup_rmdir, | |
4f41fc59 | 4394 | .show_path = cgroup_show_path, |
2bd59d48 TH |
4395 | }; |
4396 | ||
15a4c835 | 4397 | static void __init cgroup_init_subsys(struct cgroup_subsys *ss, bool early) |
ddbcc7e8 | 4398 | { |
ddbcc7e8 | 4399 | struct cgroup_subsys_state *css; |
cfe36bde | 4400 | |
a5ae9899 | 4401 | pr_debug("Initializing cgroup subsys %s\n", ss->name); |
ddbcc7e8 | 4402 | |
648bb56d TH |
4403 | mutex_lock(&cgroup_mutex); |
4404 | ||
15a4c835 | 4405 | idr_init(&ss->css_idr); |
0adb0704 | 4406 | INIT_LIST_HEAD(&ss->cfts); |
8e3f6541 | 4407 | |
3dd06ffa TH |
4408 | /* Create the root cgroup state for this subsystem */ |
4409 | ss->root = &cgrp_dfl_root; | |
4410 | css = ss->css_alloc(cgroup_css(&cgrp_dfl_root.cgrp, ss)); | |
ddbcc7e8 PM |
4411 | /* We don't handle early failures gracefully */ |
4412 | BUG_ON(IS_ERR(css)); | |
ddfcadab | 4413 | init_and_link_css(css, ss, &cgrp_dfl_root.cgrp); |
3b514d24 TH |
4414 | |
4415 | /* | |
4416 | * Root csses are never destroyed and we can't initialize | |
4417 | * percpu_ref during early init. Disable refcnting. | |
4418 | */ | |
4419 | css->flags |= CSS_NO_REF; | |
4420 | ||
15a4c835 | 4421 | if (early) { |
9395a450 | 4422 | /* allocation can't be done safely during early init */ |
15a4c835 TH |
4423 | css->id = 1; |
4424 | } else { | |
4425 | css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2, GFP_KERNEL); | |
4426 | BUG_ON(css->id < 0); | |
4427 | } | |
ddbcc7e8 | 4428 | |
e8d55fde | 4429 | /* Update the init_css_set to contain a subsys |
817929ec | 4430 | * pointer to this state - since the subsystem is |
e8d55fde | 4431 | * newly registered, all tasks and hence the |
3dd06ffa | 4432 | * init_css_set is in the subsystem's root cgroup. */ |
aec25020 | 4433 | init_css_set.subsys[ss->id] = css; |
ddbcc7e8 | 4434 | |
cb4a3167 AS |
4435 | have_fork_callback |= (bool)ss->fork << ss->id; |
4436 | have_exit_callback |= (bool)ss->exit << ss->id; | |
afcf6c8b | 4437 | have_free_callback |= (bool)ss->free << ss->id; |
7e47682e | 4438 | have_canfork_callback |= (bool)ss->can_fork << ss->id; |
ddbcc7e8 | 4439 | |
e8d55fde LZ |
4440 | /* At system boot, before all subsystems have been |
4441 | * registered, no tasks have been forked, so we don't | |
4442 | * need to invoke fork callbacks here. */ | |
4443 | BUG_ON(!list_empty(&init_task.tasks)); | |
4444 | ||
ae7f164a | 4445 | BUG_ON(online_css(css)); |
a8638030 | 4446 | |
cf5d5941 BB |
4447 | mutex_unlock(&cgroup_mutex); |
4448 | } | |
cf5d5941 | 4449 | |
ddbcc7e8 | 4450 | /** |
a043e3b2 LZ |
4451 | * cgroup_init_early - cgroup initialization at system boot |
4452 | * | |
4453 | * Initialize cgroups at system boot, and initialize any | |
4454 | * subsystems that request early init. | |
ddbcc7e8 PM |
4455 | */ |
4456 | int __init cgroup_init_early(void) | |
4457 | { | |
7b9a6ba5 | 4458 | static struct cgroup_sb_opts __initdata opts; |
30159ec7 | 4459 | struct cgroup_subsys *ss; |
ddbcc7e8 | 4460 | int i; |
30159ec7 | 4461 | |
3dd06ffa | 4462 | init_cgroup_root(&cgrp_dfl_root, &opts); |
3b514d24 TH |
4463 | cgrp_dfl_root.cgrp.self.flags |= CSS_NO_REF; |
4464 | ||
a4ea1cc9 | 4465 | RCU_INIT_POINTER(init_task.cgroups, &init_css_set); |
817929ec | 4466 | |
3ed80a62 | 4467 | for_each_subsys(ss, i) { |
aec25020 | 4468 | WARN(!ss->css_alloc || !ss->css_free || ss->name || ss->id, |
63253ad8 | 4469 | "invalid cgroup_subsys %d:%s css_alloc=%p css_free=%p id:name=%d:%s\n", |
073219e9 | 4470 | i, cgroup_subsys_name[i], ss->css_alloc, ss->css_free, |
aec25020 | 4471 | ss->id, ss->name); |
073219e9 TH |
4472 | WARN(strlen(cgroup_subsys_name[i]) > MAX_CGROUP_TYPE_NAMELEN, |
4473 | "cgroup_subsys_name %s too long\n", cgroup_subsys_name[i]); | |
4474 | ||
aec25020 | 4475 | ss->id = i; |
073219e9 | 4476 | ss->name = cgroup_subsys_name[i]; |
3e1d2eed TH |
4477 | if (!ss->legacy_name) |
4478 | ss->legacy_name = cgroup_subsys_name[i]; | |
ddbcc7e8 PM |
4479 | |
4480 | if (ss->early_init) | |
15a4c835 | 4481 | cgroup_init_subsys(ss, true); |
ddbcc7e8 PM |
4482 | } |
4483 | return 0; | |
4484 | } | |
4485 | ||
6e5c8307 | 4486 | static u16 cgroup_disable_mask __initdata; |
a3e72739 | 4487 | |
ddbcc7e8 | 4488 | /** |
a043e3b2 LZ |
4489 | * cgroup_init - cgroup initialization |
4490 | * | |
4491 | * Register cgroup filesystem and /proc file, and initialize | |
4492 | * any subsystems that didn't request early init. | |
ddbcc7e8 PM |
4493 | */ |
4494 | int __init cgroup_init(void) | |
4495 | { | |
30159ec7 | 4496 | struct cgroup_subsys *ss; |
035f4f51 | 4497 | int ssid; |
ddbcc7e8 | 4498 | |
6e5c8307 | 4499 | BUILD_BUG_ON(CGROUP_SUBSYS_COUNT > 16); |
1ed13287 | 4500 | BUG_ON(percpu_init_rwsem(&cgroup_threadgroup_rwsem)); |
d62beb7f TH |
4501 | BUG_ON(cgroup_init_cftypes(NULL, cgroup_base_files)); |
4502 | BUG_ON(cgroup_init_cftypes(NULL, cgroup1_base_files)); | |
ddbcc7e8 | 4503 | |
3942a9bd PZ |
4504 | /* |
4505 | * The latency of the synchronize_sched() is too high for cgroups, | |
4506 | * avoid it at the cost of forcing all readers into the slow path. | |
4507 | */ | |
4508 | rcu_sync_enter_start(&cgroup_threadgroup_rwsem.rss); | |
4509 | ||
a79a908f AK |
4510 | get_user_ns(init_cgroup_ns.user_ns); |
4511 | ||
54e7b4eb | 4512 | mutex_lock(&cgroup_mutex); |
54e7b4eb | 4513 | |
2378d8b8 TH |
4514 | /* |
4515 | * Add init_css_set to the hash table so that dfl_root can link to | |
4516 | * it during init. | |
4517 | */ | |
4518 | hash_add(css_set_table, &init_css_set.hlist, | |
4519 | css_set_hash(init_css_set.subsys)); | |
82fe9b0d | 4520 | |
9732adc5 | 4521 | BUG_ON(cgroup_setup_root(&cgrp_dfl_root, 0, 0)); |
4e96ee8e | 4522 | |
54e7b4eb TH |
4523 | mutex_unlock(&cgroup_mutex); |
4524 | ||
172a2c06 | 4525 | for_each_subsys(ss, ssid) { |
15a4c835 TH |
4526 | if (ss->early_init) { |
4527 | struct cgroup_subsys_state *css = | |
4528 | init_css_set.subsys[ss->id]; | |
4529 | ||
4530 | css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2, | |
4531 | GFP_KERNEL); | |
4532 | BUG_ON(css->id < 0); | |
4533 | } else { | |
4534 | cgroup_init_subsys(ss, false); | |
4535 | } | |
172a2c06 | 4536 | |
2d8f243a TH |
4537 | list_add_tail(&init_css_set.e_cset_node[ssid], |
4538 | &cgrp_dfl_root.cgrp.e_csets[ssid]); | |
172a2c06 TH |
4539 | |
4540 | /* | |
c731ae1d LZ |
4541 | * Setting dfl_root subsys_mask needs to consider the |
4542 | * disabled flag and cftype registration needs kmalloc, | |
4543 | * both of which aren't available during early_init. | |
172a2c06 | 4544 | */ |
a3e72739 TH |
4545 | if (cgroup_disable_mask & (1 << ssid)) { |
4546 | static_branch_disable(cgroup_subsys_enabled_key[ssid]); | |
4547 | printk(KERN_INFO "Disabling %s control group subsystem\n", | |
4548 | ss->name); | |
a8ddc821 | 4549 | continue; |
a3e72739 | 4550 | } |
a8ddc821 | 4551 | |
d62beb7f | 4552 | if (cgroup1_ssid_disabled(ssid)) |
223ffb29 JW |
4553 | printk(KERN_INFO "Disabling %s control group subsystem in v1 mounts\n", |
4554 | ss->name); | |
4555 | ||
a8ddc821 TH |
4556 | cgrp_dfl_root.subsys_mask |= 1 << ss->id; |
4557 | ||
f6d635ad TH |
4558 | if (ss->implicit_on_dfl) |
4559 | cgrp_dfl_implicit_ss_mask |= 1 << ss->id; | |
4560 | else if (!ss->dfl_cftypes) | |
a7165264 | 4561 | cgrp_dfl_inhibit_ss_mask |= 1 << ss->id; |
5de4fa13 | 4562 | |
a8ddc821 TH |
4563 | if (ss->dfl_cftypes == ss->legacy_cftypes) { |
4564 | WARN_ON(cgroup_add_cftypes(ss, ss->dfl_cftypes)); | |
4565 | } else { | |
4566 | WARN_ON(cgroup_add_dfl_cftypes(ss, ss->dfl_cftypes)); | |
4567 | WARN_ON(cgroup_add_legacy_cftypes(ss, ss->legacy_cftypes)); | |
c731ae1d | 4568 | } |
295458e6 VD |
4569 | |
4570 | if (ss->bind) | |
4571 | ss->bind(init_css_set.subsys[ssid]); | |
676db4af GK |
4572 | } |
4573 | ||
2378d8b8 TH |
4574 | /* init_css_set.subsys[] has been updated, re-hash */ |
4575 | hash_del(&init_css_set.hlist); | |
4576 | hash_add(css_set_table, &init_css_set.hlist, | |
4577 | css_set_hash(init_css_set.subsys)); | |
4578 | ||
035f4f51 TH |
4579 | WARN_ON(sysfs_create_mount_point(fs_kobj, "cgroup")); |
4580 | WARN_ON(register_filesystem(&cgroup_fs_type)); | |
67e9c74b | 4581 | WARN_ON(register_filesystem(&cgroup2_fs_type)); |
035f4f51 | 4582 | WARN_ON(!proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations)); |
ddbcc7e8 | 4583 | |
2bd59d48 | 4584 | return 0; |
ddbcc7e8 | 4585 | } |
b4f48b63 | 4586 | |
e5fca243 TH |
4587 | static int __init cgroup_wq_init(void) |
4588 | { | |
4589 | /* | |
4590 | * There isn't much point in executing destruction path in | |
4591 | * parallel. Good chunk is serialized with cgroup_mutex anyway. | |
1a11533f | 4592 | * Use 1 for @max_active. |
e5fca243 TH |
4593 | * |
4594 | * We would prefer to do this in cgroup_init() above, but that | |
4595 | * is called before init_workqueues(): so leave this until after. | |
4596 | */ | |
1a11533f | 4597 | cgroup_destroy_wq = alloc_workqueue("cgroup_destroy", 0, 1); |
e5fca243 TH |
4598 | BUG_ON(!cgroup_destroy_wq); |
4599 | return 0; | |
4600 | } | |
4601 | core_initcall(cgroup_wq_init); | |
4602 | ||
a424316c PM |
4603 | /* |
4604 | * proc_cgroup_show() | |
4605 | * - Print task's cgroup paths into seq_file, one line for each hierarchy | |
4606 | * - Used for /proc/<pid>/cgroup. | |
a424316c | 4607 | */ |
006f4ac4 ZL |
4608 | int proc_cgroup_show(struct seq_file *m, struct pid_namespace *ns, |
4609 | struct pid *pid, struct task_struct *tsk) | |
a424316c | 4610 | { |
4c737b41 | 4611 | char *buf; |
a424316c | 4612 | int retval; |
3dd06ffa | 4613 | struct cgroup_root *root; |
a424316c PM |
4614 | |
4615 | retval = -ENOMEM; | |
e61734c5 | 4616 | buf = kmalloc(PATH_MAX, GFP_KERNEL); |
a424316c PM |
4617 | if (!buf) |
4618 | goto out; | |
4619 | ||
a424316c | 4620 | mutex_lock(&cgroup_mutex); |
82d6489d | 4621 | spin_lock_irq(&css_set_lock); |
a424316c | 4622 | |
985ed670 | 4623 | for_each_root(root) { |
a424316c | 4624 | struct cgroup_subsys *ss; |
bd89aabc | 4625 | struct cgroup *cgrp; |
b85d2040 | 4626 | int ssid, count = 0; |
a424316c | 4627 | |
a7165264 | 4628 | if (root == &cgrp_dfl_root && !cgrp_dfl_visible) |
985ed670 TH |
4629 | continue; |
4630 | ||
2c6ab6d2 | 4631 | seq_printf(m, "%d:", root->hierarchy_id); |
d98817d4 TH |
4632 | if (root != &cgrp_dfl_root) |
4633 | for_each_subsys(ss, ssid) | |
4634 | if (root->subsys_mask & (1 << ssid)) | |
4635 | seq_printf(m, "%s%s", count++ ? "," : "", | |
3e1d2eed | 4636 | ss->legacy_name); |
c6d57f33 PM |
4637 | if (strlen(root->name)) |
4638 | seq_printf(m, "%sname=%s", count ? "," : "", | |
4639 | root->name); | |
a424316c | 4640 | seq_putc(m, ':'); |
2e91fa7f | 4641 | |
7717f7ba | 4642 | cgrp = task_cgroup_from_root(tsk, root); |
2e91fa7f TH |
4643 | |
4644 | /* | |
4645 | * On traditional hierarchies, all zombie tasks show up as | |
4646 | * belonging to the root cgroup. On the default hierarchy, | |
4647 | * while a zombie doesn't show up in "cgroup.procs" and | |
4648 | * thus can't be migrated, its /proc/PID/cgroup keeps | |
4649 | * reporting the cgroup it belonged to before exiting. If | |
4650 | * the cgroup is removed before the zombie is reaped, | |
4651 | * " (deleted)" is appended to the cgroup path. | |
4652 | */ | |
4653 | if (cgroup_on_dfl(cgrp) || !(tsk->flags & PF_EXITING)) { | |
4c737b41 | 4654 | retval = cgroup_path_ns_locked(cgrp, buf, PATH_MAX, |
a79a908f | 4655 | current->nsproxy->cgroup_ns); |
e0223003 | 4656 | if (retval >= PATH_MAX) |
2e91fa7f | 4657 | retval = -ENAMETOOLONG; |
e0223003 | 4658 | if (retval < 0) |
2e91fa7f | 4659 | goto out_unlock; |
4c737b41 TH |
4660 | |
4661 | seq_puts(m, buf); | |
2e91fa7f | 4662 | } else { |
4c737b41 | 4663 | seq_puts(m, "/"); |
e61734c5 | 4664 | } |
2e91fa7f | 4665 | |
2e91fa7f TH |
4666 | if (cgroup_on_dfl(cgrp) && cgroup_is_dead(cgrp)) |
4667 | seq_puts(m, " (deleted)\n"); | |
4668 | else | |
4669 | seq_putc(m, '\n'); | |
a424316c PM |
4670 | } |
4671 | ||
006f4ac4 | 4672 | retval = 0; |
a424316c | 4673 | out_unlock: |
82d6489d | 4674 | spin_unlock_irq(&css_set_lock); |
a424316c | 4675 | mutex_unlock(&cgroup_mutex); |
a424316c PM |
4676 | kfree(buf); |
4677 | out: | |
4678 | return retval; | |
4679 | } | |
4680 | ||
b4f48b63 | 4681 | /** |
eaf797ab | 4682 | * cgroup_fork - initialize cgroup related fields during copy_process() |
a043e3b2 | 4683 | * @child: pointer to task_struct of forking parent process. |
b4f48b63 | 4684 | * |
eaf797ab TH |
4685 | * A task is associated with the init_css_set until cgroup_post_fork() |
4686 | * attaches it to the parent's css_set. Empty cg_list indicates that | |
4687 | * @child isn't holding reference to its css_set. | |
b4f48b63 PM |
4688 | */ |
4689 | void cgroup_fork(struct task_struct *child) | |
4690 | { | |
eaf797ab | 4691 | RCU_INIT_POINTER(child->cgroups, &init_css_set); |
817929ec | 4692 | INIT_LIST_HEAD(&child->cg_list); |
b4f48b63 PM |
4693 | } |
4694 | ||
7e47682e AS |
4695 | /** |
4696 | * cgroup_can_fork - called on a new task before the process is exposed | |
4697 | * @child: the task in question. | |
4698 | * | |
4699 | * This calls the subsystem can_fork() callbacks. If the can_fork() callback | |
4700 | * returns an error, the fork aborts with that error code. This allows for | |
4701 | * a cgroup subsystem to conditionally allow or deny new forks. | |
4702 | */ | |
b53202e6 | 4703 | int cgroup_can_fork(struct task_struct *child) |
7e47682e AS |
4704 | { |
4705 | struct cgroup_subsys *ss; | |
4706 | int i, j, ret; | |
4707 | ||
b4e0eeaf | 4708 | do_each_subsys_mask(ss, i, have_canfork_callback) { |
b53202e6 | 4709 | ret = ss->can_fork(child); |
7e47682e AS |
4710 | if (ret) |
4711 | goto out_revert; | |
b4e0eeaf | 4712 | } while_each_subsys_mask(); |
7e47682e AS |
4713 | |
4714 | return 0; | |
4715 | ||
4716 | out_revert: | |
4717 | for_each_subsys(ss, j) { | |
4718 | if (j >= i) | |
4719 | break; | |
4720 | if (ss->cancel_fork) | |
b53202e6 | 4721 | ss->cancel_fork(child); |
7e47682e AS |
4722 | } |
4723 | ||
4724 | return ret; | |
4725 | } | |
4726 | ||
4727 | /** | |
4728 | * cgroup_cancel_fork - called if a fork failed after cgroup_can_fork() | |
4729 | * @child: the task in question | |
4730 | * | |
4731 | * This calls the cancel_fork() callbacks if a fork failed *after* | |
4732 | * cgroup_can_fork() succeded. | |
4733 | */ | |
b53202e6 | 4734 | void cgroup_cancel_fork(struct task_struct *child) |
7e47682e AS |
4735 | { |
4736 | struct cgroup_subsys *ss; | |
4737 | int i; | |
4738 | ||
4739 | for_each_subsys(ss, i) | |
4740 | if (ss->cancel_fork) | |
b53202e6 | 4741 | ss->cancel_fork(child); |
7e47682e AS |
4742 | } |
4743 | ||
817929ec | 4744 | /** |
a043e3b2 LZ |
4745 | * cgroup_post_fork - called on a new task after adding it to the task list |
4746 | * @child: the task in question | |
4747 | * | |
5edee61e TH |
4748 | * Adds the task to the list running through its css_set if necessary and |
4749 | * call the subsystem fork() callbacks. Has to be after the task is | |
4750 | * visible on the task list in case we race with the first call to | |
0942eeee | 4751 | * cgroup_task_iter_start() - to guarantee that the new task ends up on its |
5edee61e | 4752 | * list. |
a043e3b2 | 4753 | */ |
b53202e6 | 4754 | void cgroup_post_fork(struct task_struct *child) |
817929ec | 4755 | { |
30159ec7 | 4756 | struct cgroup_subsys *ss; |
5edee61e TH |
4757 | int i; |
4758 | ||
3ce3230a | 4759 | /* |
251f8c03 | 4760 | * This may race against cgroup_enable_task_cg_lists(). As that |
eaf797ab TH |
4761 | * function sets use_task_css_set_links before grabbing |
4762 | * tasklist_lock and we just went through tasklist_lock to add | |
4763 | * @child, it's guaranteed that either we see the set | |
4764 | * use_task_css_set_links or cgroup_enable_task_cg_lists() sees | |
4765 | * @child during its iteration. | |
4766 | * | |
4767 | * If we won the race, @child is associated with %current's | |
f0d9a5f1 | 4768 | * css_set. Grabbing css_set_lock guarantees both that the |
eaf797ab TH |
4769 | * association is stable, and, on completion of the parent's |
4770 | * migration, @child is visible in the source of migration or | |
4771 | * already in the destination cgroup. This guarantee is necessary | |
4772 | * when implementing operations which need to migrate all tasks of | |
4773 | * a cgroup to another. | |
4774 | * | |
251f8c03 | 4775 | * Note that if we lose to cgroup_enable_task_cg_lists(), @child |
eaf797ab TH |
4776 | * will remain in init_css_set. This is safe because all tasks are |
4777 | * in the init_css_set before cg_links is enabled and there's no | |
4778 | * operation which transfers all tasks out of init_css_set. | |
3ce3230a | 4779 | */ |
817929ec | 4780 | if (use_task_css_set_links) { |
eaf797ab TH |
4781 | struct css_set *cset; |
4782 | ||
82d6489d | 4783 | spin_lock_irq(&css_set_lock); |
0e1d768f | 4784 | cset = task_css_set(current); |
eaf797ab | 4785 | if (list_empty(&child->cg_list)) { |
eaf797ab | 4786 | get_css_set(cset); |
f6d7d049 | 4787 | css_set_move_task(child, NULL, cset, false); |
eaf797ab | 4788 | } |
82d6489d | 4789 | spin_unlock_irq(&css_set_lock); |
817929ec | 4790 | } |
5edee61e TH |
4791 | |
4792 | /* | |
4793 | * Call ss->fork(). This must happen after @child is linked on | |
4794 | * css_set; otherwise, @child might change state between ->fork() | |
4795 | * and addition to css_set. | |
4796 | */ | |
b4e0eeaf | 4797 | do_each_subsys_mask(ss, i, have_fork_callback) { |
b53202e6 | 4798 | ss->fork(child); |
b4e0eeaf | 4799 | } while_each_subsys_mask(); |
817929ec | 4800 | } |
5edee61e | 4801 | |
b4f48b63 PM |
4802 | /** |
4803 | * cgroup_exit - detach cgroup from exiting task | |
4804 | * @tsk: pointer to task_struct of exiting process | |
4805 | * | |
4806 | * Description: Detach cgroup from @tsk and release it. | |
4807 | * | |
4808 | * Note that cgroups marked notify_on_release force every task in | |
4809 | * them to take the global cgroup_mutex mutex when exiting. | |
4810 | * This could impact scaling on very large systems. Be reluctant to | |
4811 | * use notify_on_release cgroups where very high task exit scaling | |
4812 | * is required on large systems. | |
4813 | * | |
0e1d768f TH |
4814 | * We set the exiting tasks cgroup to the root cgroup (top_cgroup). We |
4815 | * call cgroup_exit() while the task is still competent to handle | |
4816 | * notify_on_release(), then leave the task attached to the root cgroup in | |
4817 | * each hierarchy for the remainder of its exit. No need to bother with | |
4818 | * init_css_set refcnting. init_css_set never goes away and we can't race | |
e8604cb4 | 4819 | * with migration path - PF_EXITING is visible to migration path. |
b4f48b63 | 4820 | */ |
1ec41830 | 4821 | void cgroup_exit(struct task_struct *tsk) |
b4f48b63 | 4822 | { |
30159ec7 | 4823 | struct cgroup_subsys *ss; |
5abb8855 | 4824 | struct css_set *cset; |
d41d5a01 | 4825 | int i; |
817929ec PM |
4826 | |
4827 | /* | |
0e1d768f | 4828 | * Unlink from @tsk from its css_set. As migration path can't race |
0de0942d | 4829 | * with us, we can check css_set and cg_list without synchronization. |
817929ec | 4830 | */ |
0de0942d TH |
4831 | cset = task_css_set(tsk); |
4832 | ||
817929ec | 4833 | if (!list_empty(&tsk->cg_list)) { |
82d6489d | 4834 | spin_lock_irq(&css_set_lock); |
f6d7d049 | 4835 | css_set_move_task(tsk, cset, NULL, false); |
82d6489d | 4836 | spin_unlock_irq(&css_set_lock); |
2e91fa7f TH |
4837 | } else { |
4838 | get_css_set(cset); | |
817929ec PM |
4839 | } |
4840 | ||
cb4a3167 | 4841 | /* see cgroup_post_fork() for details */ |
b4e0eeaf | 4842 | do_each_subsys_mask(ss, i, have_exit_callback) { |
2e91fa7f | 4843 | ss->exit(tsk); |
b4e0eeaf | 4844 | } while_each_subsys_mask(); |
2e91fa7f | 4845 | } |
30159ec7 | 4846 | |
2e91fa7f TH |
4847 | void cgroup_free(struct task_struct *task) |
4848 | { | |
4849 | struct css_set *cset = task_css_set(task); | |
afcf6c8b TH |
4850 | struct cgroup_subsys *ss; |
4851 | int ssid; | |
4852 | ||
b4e0eeaf | 4853 | do_each_subsys_mask(ss, ssid, have_free_callback) { |
afcf6c8b | 4854 | ss->free(task); |
b4e0eeaf | 4855 | } while_each_subsys_mask(); |
d41d5a01 | 4856 | |
2e91fa7f | 4857 | put_css_set(cset); |
b4f48b63 | 4858 | } |
697f4161 | 4859 | |
8bab8dde PM |
4860 | static int __init cgroup_disable(char *str) |
4861 | { | |
30159ec7 | 4862 | struct cgroup_subsys *ss; |
8bab8dde | 4863 | char *token; |
30159ec7 | 4864 | int i; |
8bab8dde PM |
4865 | |
4866 | while ((token = strsep(&str, ",")) != NULL) { | |
4867 | if (!*token) | |
4868 | continue; | |
be45c900 | 4869 | |
3ed80a62 | 4870 | for_each_subsys(ss, i) { |
3e1d2eed TH |
4871 | if (strcmp(token, ss->name) && |
4872 | strcmp(token, ss->legacy_name)) | |
4873 | continue; | |
a3e72739 | 4874 | cgroup_disable_mask |= 1 << i; |
8bab8dde PM |
4875 | } |
4876 | } | |
4877 | return 1; | |
4878 | } | |
4879 | __setup("cgroup_disable=", cgroup_disable); | |
38460b48 | 4880 | |
b77d7b60 | 4881 | /** |
ec903c0c | 4882 | * css_tryget_online_from_dir - get corresponding css from a cgroup dentry |
35cf0836 TH |
4883 | * @dentry: directory dentry of interest |
4884 | * @ss: subsystem of interest | |
b77d7b60 | 4885 | * |
5a17f543 TH |
4886 | * If @dentry is a directory for a cgroup which has @ss enabled on it, try |
4887 | * to get the corresponding css and return it. If such css doesn't exist | |
4888 | * or can't be pinned, an ERR_PTR value is returned. | |
e5d1367f | 4889 | */ |
ec903c0c TH |
4890 | struct cgroup_subsys_state *css_tryget_online_from_dir(struct dentry *dentry, |
4891 | struct cgroup_subsys *ss) | |
e5d1367f | 4892 | { |
2bd59d48 | 4893 | struct kernfs_node *kn = kernfs_node_from_dentry(dentry); |
f17fc25f | 4894 | struct file_system_type *s_type = dentry->d_sb->s_type; |
2bd59d48 | 4895 | struct cgroup_subsys_state *css = NULL; |
e5d1367f | 4896 | struct cgroup *cgrp; |
e5d1367f | 4897 | |
35cf0836 | 4898 | /* is @dentry a cgroup dir? */ |
f17fc25f TH |
4899 | if ((s_type != &cgroup_fs_type && s_type != &cgroup2_fs_type) || |
4900 | !kn || kernfs_type(kn) != KERNFS_DIR) | |
e5d1367f SE |
4901 | return ERR_PTR(-EBADF); |
4902 | ||
5a17f543 TH |
4903 | rcu_read_lock(); |
4904 | ||
2bd59d48 TH |
4905 | /* |
4906 | * This path doesn't originate from kernfs and @kn could already | |
4907 | * have been or be removed at any point. @kn->priv is RCU | |
a4189487 | 4908 | * protected for this access. See css_release_work_fn() for details. |
2bd59d48 | 4909 | */ |
e0aed7c7 | 4910 | cgrp = rcu_dereference(*(void __rcu __force **)&kn->priv); |
2bd59d48 TH |
4911 | if (cgrp) |
4912 | css = cgroup_css(cgrp, ss); | |
5a17f543 | 4913 | |
ec903c0c | 4914 | if (!css || !css_tryget_online(css)) |
5a17f543 TH |
4915 | css = ERR_PTR(-ENOENT); |
4916 | ||
4917 | rcu_read_unlock(); | |
4918 | return css; | |
e5d1367f | 4919 | } |
e5d1367f | 4920 | |
1cb650b9 LZ |
4921 | /** |
4922 | * css_from_id - lookup css by id | |
4923 | * @id: the cgroup id | |
4924 | * @ss: cgroup subsys to be looked into | |
4925 | * | |
4926 | * Returns the css if there's valid one with @id, otherwise returns NULL. | |
4927 | * Should be called under rcu_read_lock(). | |
4928 | */ | |
4929 | struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss) | |
4930 | { | |
6fa4918d | 4931 | WARN_ON_ONCE(!rcu_read_lock_held()); |
d6ccc55e | 4932 | return idr_find(&ss->css_idr, id); |
e5d1367f SE |
4933 | } |
4934 | ||
16af4396 TH |
4935 | /** |
4936 | * cgroup_get_from_path - lookup and get a cgroup from its default hierarchy path | |
4937 | * @path: path on the default hierarchy | |
4938 | * | |
4939 | * Find the cgroup at @path on the default hierarchy, increment its | |
4940 | * reference count and return it. Returns pointer to the found cgroup on | |
4941 | * success, ERR_PTR(-ENOENT) if @path doens't exist and ERR_PTR(-ENOTDIR) | |
4942 | * if @path points to a non-directory. | |
4943 | */ | |
4944 | struct cgroup *cgroup_get_from_path(const char *path) | |
4945 | { | |
4946 | struct kernfs_node *kn; | |
4947 | struct cgroup *cgrp; | |
4948 | ||
4949 | mutex_lock(&cgroup_mutex); | |
4950 | ||
4951 | kn = kernfs_walk_and_get(cgrp_dfl_root.cgrp.kn, path); | |
4952 | if (kn) { | |
4953 | if (kernfs_type(kn) == KERNFS_DIR) { | |
4954 | cgrp = kn->priv; | |
a590b90d | 4955 | cgroup_get_live(cgrp); |
16af4396 TH |
4956 | } else { |
4957 | cgrp = ERR_PTR(-ENOTDIR); | |
4958 | } | |
4959 | kernfs_put(kn); | |
4960 | } else { | |
4961 | cgrp = ERR_PTR(-ENOENT); | |
4962 | } | |
4963 | ||
4964 | mutex_unlock(&cgroup_mutex); | |
4965 | return cgrp; | |
4966 | } | |
4967 | EXPORT_SYMBOL_GPL(cgroup_get_from_path); | |
4968 | ||
1f3fe7eb MKL |
4969 | /** |
4970 | * cgroup_get_from_fd - get a cgroup pointer from a fd | |
4971 | * @fd: fd obtained by open(cgroup2_dir) | |
4972 | * | |
4973 | * Find the cgroup from a fd which should be obtained | |
4974 | * by opening a cgroup directory. Returns a pointer to the | |
4975 | * cgroup on success. ERR_PTR is returned if the cgroup | |
4976 | * cannot be found. | |
4977 | */ | |
4978 | struct cgroup *cgroup_get_from_fd(int fd) | |
4979 | { | |
4980 | struct cgroup_subsys_state *css; | |
4981 | struct cgroup *cgrp; | |
4982 | struct file *f; | |
4983 | ||
4984 | f = fget_raw(fd); | |
4985 | if (!f) | |
4986 | return ERR_PTR(-EBADF); | |
4987 | ||
4988 | css = css_tryget_online_from_dir(f->f_path.dentry, NULL); | |
4989 | fput(f); | |
4990 | if (IS_ERR(css)) | |
4991 | return ERR_CAST(css); | |
4992 | ||
4993 | cgrp = css->cgroup; | |
4994 | if (!cgroup_on_dfl(cgrp)) { | |
4995 | cgroup_put(cgrp); | |
4996 | return ERR_PTR(-EBADF); | |
4997 | } | |
4998 | ||
4999 | return cgrp; | |
5000 | } | |
5001 | EXPORT_SYMBOL_GPL(cgroup_get_from_fd); | |
5002 | ||
bd1060a1 TH |
5003 | /* |
5004 | * sock->sk_cgrp_data handling. For more info, see sock_cgroup_data | |
5005 | * definition in cgroup-defs.h. | |
5006 | */ | |
5007 | #ifdef CONFIG_SOCK_CGROUP_DATA | |
5008 | ||
5009 | #if defined(CONFIG_CGROUP_NET_PRIO) || defined(CONFIG_CGROUP_NET_CLASSID) | |
5010 | ||
3fa4cc9c | 5011 | DEFINE_SPINLOCK(cgroup_sk_update_lock); |
bd1060a1 TH |
5012 | static bool cgroup_sk_alloc_disabled __read_mostly; |
5013 | ||
5014 | void cgroup_sk_alloc_disable(void) | |
5015 | { | |
5016 | if (cgroup_sk_alloc_disabled) | |
5017 | return; | |
5018 | pr_info("cgroup: disabling cgroup2 socket matching due to net_prio or net_cls activation\n"); | |
5019 | cgroup_sk_alloc_disabled = true; | |
5020 | } | |
5021 | ||
5022 | #else | |
5023 | ||
5024 | #define cgroup_sk_alloc_disabled false | |
5025 | ||
5026 | #endif | |
5027 | ||
5028 | void cgroup_sk_alloc(struct sock_cgroup_data *skcd) | |
5029 | { | |
5030 | if (cgroup_sk_alloc_disabled) | |
5031 | return; | |
5032 | ||
d979a39d JW |
5033 | /* Socket clone path */ |
5034 | if (skcd->val) { | |
a590b90d TH |
5035 | /* |
5036 | * We might be cloning a socket which is left in an empty | |
5037 | * cgroup and the cgroup might have already been rmdir'd. | |
5038 | * Don't use cgroup_get_live(). | |
5039 | */ | |
d979a39d JW |
5040 | cgroup_get(sock_cgroup_ptr(skcd)); |
5041 | return; | |
5042 | } | |
5043 | ||
bd1060a1 TH |
5044 | rcu_read_lock(); |
5045 | ||
5046 | while (true) { | |
5047 | struct css_set *cset; | |
5048 | ||
5049 | cset = task_css_set(current); | |
5050 | if (likely(cgroup_tryget(cset->dfl_cgrp))) { | |
5051 | skcd->val = (unsigned long)cset->dfl_cgrp; | |
5052 | break; | |
5053 | } | |
5054 | cpu_relax(); | |
5055 | } | |
5056 | ||
5057 | rcu_read_unlock(); | |
5058 | } | |
5059 | ||
5060 | void cgroup_sk_free(struct sock_cgroup_data *skcd) | |
5061 | { | |
5062 | cgroup_put(sock_cgroup_ptr(skcd)); | |
5063 | } | |
5064 | ||
5065 | #endif /* CONFIG_SOCK_CGROUP_DATA */ | |
5066 | ||
30070984 | 5067 | #ifdef CONFIG_CGROUP_BPF |
7f677633 AS |
5068 | int cgroup_bpf_update(struct cgroup *cgrp, struct bpf_prog *prog, |
5069 | enum bpf_attach_type type, bool overridable) | |
30070984 DM |
5070 | { |
5071 | struct cgroup *parent = cgroup_parent(cgrp); | |
7f677633 | 5072 | int ret; |
30070984 DM |
5073 | |
5074 | mutex_lock(&cgroup_mutex); | |
7f677633 | 5075 | ret = __cgroup_bpf_update(cgrp, parent, prog, type, overridable); |
30070984 | 5076 | mutex_unlock(&cgroup_mutex); |
7f677633 | 5077 | return ret; |
30070984 DM |
5078 | } |
5079 | #endif /* CONFIG_CGROUP_BPF */ |