Commit | Line | Data |
---|---|---|
ddbcc7e8 | 1 | /* |
ddbcc7e8 PM |
2 | * Generic process-grouping system. |
3 | * | |
4 | * Based originally on the cpuset system, extracted by Paul Menage | |
5 | * Copyright (C) 2006 Google, Inc | |
6 | * | |
0dea1168 KS |
7 | * Notifications support |
8 | * Copyright (C) 2009 Nokia Corporation | |
9 | * Author: Kirill A. Shutemov | |
10 | * | |
ddbcc7e8 PM |
11 | * Copyright notices from the original cpuset code: |
12 | * -------------------------------------------------- | |
13 | * Copyright (C) 2003 BULL SA. | |
14 | * Copyright (C) 2004-2006 Silicon Graphics, Inc. | |
15 | * | |
16 | * Portions derived from Patrick Mochel's sysfs code. | |
17 | * sysfs is Copyright (c) 2001-3 Patrick Mochel | |
18 | * | |
19 | * 2003-10-10 Written by Simon Derr. | |
20 | * 2003-10-22 Updates by Stephen Hemminger. | |
21 | * 2004 May-July Rework by Paul Jackson. | |
22 | * --------------------------------------------------- | |
23 | * | |
24 | * This file is subject to the terms and conditions of the GNU General Public | |
25 | * License. See the file COPYING in the main directory of the Linux | |
26 | * distribution for more details. | |
27 | */ | |
28 | ||
ed3d261b JP |
29 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
30 | ||
ddbcc7e8 | 31 | #include <linux/cgroup.h> |
2ce9738b | 32 | #include <linux/cred.h> |
c6d57f33 | 33 | #include <linux/ctype.h> |
ddbcc7e8 | 34 | #include <linux/errno.h> |
2ce9738b | 35 | #include <linux/init_task.h> |
ddbcc7e8 PM |
36 | #include <linux/kernel.h> |
37 | #include <linux/list.h> | |
c9482a5b | 38 | #include <linux/magic.h> |
ddbcc7e8 PM |
39 | #include <linux/mm.h> |
40 | #include <linux/mutex.h> | |
41 | #include <linux/mount.h> | |
42 | #include <linux/pagemap.h> | |
a424316c | 43 | #include <linux/proc_fs.h> |
ddbcc7e8 PM |
44 | #include <linux/rcupdate.h> |
45 | #include <linux/sched.h> | |
ddbcc7e8 | 46 | #include <linux/slab.h> |
ddbcc7e8 | 47 | #include <linux/spinlock.h> |
1ed13287 | 48 | #include <linux/percpu-rwsem.h> |
ddbcc7e8 | 49 | #include <linux/string.h> |
bbcb81d0 | 50 | #include <linux/sort.h> |
81a6a5cd | 51 | #include <linux/kmod.h> |
846c7bb0 BS |
52 | #include <linux/delayacct.h> |
53 | #include <linux/cgroupstats.h> | |
0ac801fe | 54 | #include <linux/hashtable.h> |
096b7fe0 | 55 | #include <linux/pid_namespace.h> |
2c6ab6d2 | 56 | #include <linux/idr.h> |
d1d9fd33 | 57 | #include <linux/vmalloc.h> /* TODO: replace with more sophisticated array */ |
c4c27fbd | 58 | #include <linux/kthread.h> |
776f02fa | 59 | #include <linux/delay.h> |
60063497 | 60 | #include <linux/atomic.h> |
e93ad19d | 61 | #include <linux/cpuset.h> |
bd1060a1 | 62 | #include <net/sock.h> |
ddbcc7e8 | 63 | |
b1a21367 TH |
64 | /* |
65 | * pidlists linger the following amount before being destroyed. The goal | |
66 | * is avoiding frequent destruction in the middle of consecutive read calls | |
67 | * Expiring in the middle is a performance problem not a correctness one. | |
68 | * 1 sec should be enough. | |
69 | */ | |
70 | #define CGROUP_PIDLIST_DESTROY_DELAY HZ | |
71 | ||
8d7e6fb0 TH |
72 | #define CGROUP_FILE_NAME_MAX (MAX_CGROUP_TYPE_NAMELEN + \ |
73 | MAX_CFTYPE_NAME + 2) | |
74 | ||
e25e2cbb TH |
75 | /* |
76 | * cgroup_mutex is the master lock. Any modification to cgroup or its | |
77 | * hierarchy must be performed while holding it. | |
78 | * | |
f0d9a5f1 | 79 | * css_set_lock protects task->cgroups pointer, the list of css_set |
0e1d768f | 80 | * objects, and the chain of tasks off each css_set. |
e25e2cbb | 81 | * |
0e1d768f TH |
82 | * These locks are exported if CONFIG_PROVE_RCU so that accessors in |
83 | * cgroup.h can use them for lockdep annotations. | |
e25e2cbb | 84 | */ |
2219449a TH |
85 | #ifdef CONFIG_PROVE_RCU |
86 | DEFINE_MUTEX(cgroup_mutex); | |
f0d9a5f1 | 87 | DEFINE_SPINLOCK(css_set_lock); |
0e1d768f | 88 | EXPORT_SYMBOL_GPL(cgroup_mutex); |
f0d9a5f1 | 89 | EXPORT_SYMBOL_GPL(css_set_lock); |
2219449a | 90 | #else |
81a6a5cd | 91 | static DEFINE_MUTEX(cgroup_mutex); |
f0d9a5f1 | 92 | static DEFINE_SPINLOCK(css_set_lock); |
2219449a TH |
93 | #endif |
94 | ||
6fa4918d | 95 | /* |
15a4c835 TH |
96 | * Protects cgroup_idr and css_idr so that IDs can be released without |
97 | * grabbing cgroup_mutex. | |
6fa4918d TH |
98 | */ |
99 | static DEFINE_SPINLOCK(cgroup_idr_lock); | |
100 | ||
34c06254 TH |
101 | /* |
102 | * Protects cgroup_file->kn for !self csses. It synchronizes notifications | |
103 | * against file removal/re-creation across css hiding. | |
104 | */ | |
105 | static DEFINE_SPINLOCK(cgroup_file_kn_lock); | |
106 | ||
69e943b7 TH |
107 | /* |
108 | * Protects cgroup_subsys->release_agent_path. Modifying it also requires | |
109 | * cgroup_mutex. Reading requires either cgroup_mutex or this spinlock. | |
110 | */ | |
111 | static DEFINE_SPINLOCK(release_agent_path_lock); | |
81a6a5cd | 112 | |
1ed13287 TH |
113 | struct percpu_rw_semaphore cgroup_threadgroup_rwsem; |
114 | ||
8353da1f | 115 | #define cgroup_assert_mutex_or_rcu_locked() \ |
f78f5b90 PM |
116 | RCU_LOCKDEP_WARN(!rcu_read_lock_held() && \ |
117 | !lockdep_is_held(&cgroup_mutex), \ | |
8353da1f | 118 | "cgroup_mutex or RCU read lock required"); |
780cd8b3 | 119 | |
e5fca243 TH |
120 | /* |
121 | * cgroup destruction makes heavy use of work items and there can be a lot | |
122 | * of concurrent destructions. Use a separate workqueue so that cgroup | |
123 | * destruction work items don't end up filling up max_active of system_wq | |
124 | * which may lead to deadlock. | |
125 | */ | |
126 | static struct workqueue_struct *cgroup_destroy_wq; | |
127 | ||
b1a21367 TH |
128 | /* |
129 | * pidlist destructions need to be flushed on cgroup destruction. Use a | |
130 | * separate workqueue as flush domain. | |
131 | */ | |
132 | static struct workqueue_struct *cgroup_pidlist_destroy_wq; | |
133 | ||
3ed80a62 | 134 | /* generate an array of cgroup subsystem pointers */ |
073219e9 | 135 | #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys, |
3ed80a62 | 136 | static struct cgroup_subsys *cgroup_subsys[] = { |
ddbcc7e8 PM |
137 | #include <linux/cgroup_subsys.h> |
138 | }; | |
073219e9 TH |
139 | #undef SUBSYS |
140 | ||
141 | /* array of cgroup subsystem names */ | |
142 | #define SUBSYS(_x) [_x ## _cgrp_id] = #_x, | |
143 | static const char *cgroup_subsys_name[] = { | |
ddbcc7e8 PM |
144 | #include <linux/cgroup_subsys.h> |
145 | }; | |
073219e9 | 146 | #undef SUBSYS |
ddbcc7e8 | 147 | |
49d1dc4b TH |
148 | /* array of static_keys for cgroup_subsys_enabled() and cgroup_subsys_on_dfl() */ |
149 | #define SUBSYS(_x) \ | |
150 | DEFINE_STATIC_KEY_TRUE(_x ## _cgrp_subsys_enabled_key); \ | |
151 | DEFINE_STATIC_KEY_TRUE(_x ## _cgrp_subsys_on_dfl_key); \ | |
152 | EXPORT_SYMBOL_GPL(_x ## _cgrp_subsys_enabled_key); \ | |
153 | EXPORT_SYMBOL_GPL(_x ## _cgrp_subsys_on_dfl_key); | |
154 | #include <linux/cgroup_subsys.h> | |
155 | #undef SUBSYS | |
156 | ||
157 | #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys_enabled_key, | |
158 | static struct static_key_true *cgroup_subsys_enabled_key[] = { | |
159 | #include <linux/cgroup_subsys.h> | |
160 | }; | |
161 | #undef SUBSYS | |
162 | ||
163 | #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys_on_dfl_key, | |
164 | static struct static_key_true *cgroup_subsys_on_dfl_key[] = { | |
165 | #include <linux/cgroup_subsys.h> | |
166 | }; | |
167 | #undef SUBSYS | |
168 | ||
ddbcc7e8 | 169 | /* |
3dd06ffa | 170 | * The default hierarchy, reserved for the subsystems that are otherwise |
9871bf95 TH |
171 | * unattached - it never has more than a single cgroup, and all tasks are |
172 | * part of that cgroup. | |
ddbcc7e8 | 173 | */ |
a2dd4247 | 174 | struct cgroup_root cgrp_dfl_root; |
d0ec4230 | 175 | EXPORT_SYMBOL_GPL(cgrp_dfl_root); |
9871bf95 | 176 | |
a2dd4247 TH |
177 | /* |
178 | * The default hierarchy always exists but is hidden until mounted for the | |
179 | * first time. This is for backward compatibility. | |
180 | */ | |
a7165264 | 181 | static bool cgrp_dfl_visible; |
ddbcc7e8 | 182 | |
223ffb29 | 183 | /* Controllers blocked by the commandline in v1 */ |
6e5c8307 | 184 | static u16 cgroup_no_v1_mask; |
223ffb29 | 185 | |
5533e011 | 186 | /* some controllers are not supported in the default hierarchy */ |
a7165264 | 187 | static u16 cgrp_dfl_inhibit_ss_mask; |
5533e011 | 188 | |
ddbcc7e8 PM |
189 | /* The list of hierarchy roots */ |
190 | ||
9871bf95 TH |
191 | static LIST_HEAD(cgroup_roots); |
192 | static int cgroup_root_count; | |
ddbcc7e8 | 193 | |
3417ae1f | 194 | /* hierarchy ID allocation and mapping, protected by cgroup_mutex */ |
1a574231 | 195 | static DEFINE_IDR(cgroup_hierarchy_idr); |
2c6ab6d2 | 196 | |
794611a1 | 197 | /* |
0cb51d71 TH |
198 | * Assign a monotonically increasing serial number to csses. It guarantees |
199 | * cgroups with bigger numbers are newer than those with smaller numbers. | |
200 | * Also, as csses are always appended to the parent's ->children list, it | |
201 | * guarantees that sibling csses are always sorted in the ascending serial | |
202 | * number order on the list. Protected by cgroup_mutex. | |
794611a1 | 203 | */ |
0cb51d71 | 204 | static u64 css_serial_nr_next = 1; |
794611a1 | 205 | |
cb4a3167 AS |
206 | /* |
207 | * These bitmask flags indicate whether tasks in the fork and exit paths have | |
208 | * fork/exit handlers to call. This avoids us having to do extra work in the | |
209 | * fork/exit path to check which subsystems have fork/exit callbacks. | |
ddbcc7e8 | 210 | */ |
6e5c8307 TH |
211 | static u16 have_fork_callback __read_mostly; |
212 | static u16 have_exit_callback __read_mostly; | |
213 | static u16 have_free_callback __read_mostly; | |
ddbcc7e8 | 214 | |
7e47682e | 215 | /* Ditto for the can_fork callback. */ |
6e5c8307 | 216 | static u16 have_canfork_callback __read_mostly; |
7e47682e | 217 | |
67e9c74b | 218 | static struct file_system_type cgroup2_fs_type; |
a14c6874 TH |
219 | static struct cftype cgroup_dfl_base_files[]; |
220 | static struct cftype cgroup_legacy_base_files[]; | |
628f7cd4 | 221 | |
6e5c8307 | 222 | static int rebind_subsystems(struct cgroup_root *dst_root, u16 ss_mask); |
ed27b9f7 | 223 | static void css_task_iter_advance(struct css_task_iter *it); |
42809dd4 | 224 | static int cgroup_destroy_locked(struct cgroup *cgrp); |
6cd0f5bb TH |
225 | static struct cgroup_subsys_state *css_create(struct cgroup *cgrp, |
226 | struct cgroup_subsys *ss); | |
9d755d33 | 227 | static void css_release(struct percpu_ref *ref); |
f8f22e53 | 228 | static void kill_css(struct cgroup_subsys_state *css); |
4df8dc90 TH |
229 | static int cgroup_addrm_files(struct cgroup_subsys_state *css, |
230 | struct cgroup *cgrp, struct cftype cfts[], | |
2bb566cb | 231 | bool is_add); |
42809dd4 | 232 | |
fc5ed1e9 TH |
233 | /** |
234 | * cgroup_ssid_enabled - cgroup subsys enabled test by subsys ID | |
235 | * @ssid: subsys ID of interest | |
236 | * | |
237 | * cgroup_subsys_enabled() can only be used with literal subsys names which | |
238 | * is fine for individual subsystems but unsuitable for cgroup core. This | |
239 | * is slower static_key_enabled() based test indexed by @ssid. | |
240 | */ | |
241 | static bool cgroup_ssid_enabled(int ssid) | |
242 | { | |
243 | return static_key_enabled(cgroup_subsys_enabled_key[ssid]); | |
244 | } | |
245 | ||
223ffb29 JW |
246 | static bool cgroup_ssid_no_v1(int ssid) |
247 | { | |
248 | return cgroup_no_v1_mask & (1 << ssid); | |
249 | } | |
250 | ||
9e10a130 TH |
251 | /** |
252 | * cgroup_on_dfl - test whether a cgroup is on the default hierarchy | |
253 | * @cgrp: the cgroup of interest | |
254 | * | |
255 | * The default hierarchy is the v2 interface of cgroup and this function | |
256 | * can be used to test whether a cgroup is on the default hierarchy for | |
257 | * cases where a subsystem should behave differnetly depending on the | |
258 | * interface version. | |
259 | * | |
260 | * The set of behaviors which change on the default hierarchy are still | |
261 | * being determined and the mount option is prefixed with __DEVEL__. | |
262 | * | |
263 | * List of changed behaviors: | |
264 | * | |
265 | * - Mount options "noprefix", "xattr", "clone_children", "release_agent" | |
266 | * and "name" are disallowed. | |
267 | * | |
268 | * - When mounting an existing superblock, mount options should match. | |
269 | * | |
270 | * - Remount is disallowed. | |
271 | * | |
272 | * - rename(2) is disallowed. | |
273 | * | |
274 | * - "tasks" is removed. Everything should be at process granularity. Use | |
275 | * "cgroup.procs" instead. | |
276 | * | |
277 | * - "cgroup.procs" is not sorted. pids will be unique unless they got | |
278 | * recycled inbetween reads. | |
279 | * | |
280 | * - "release_agent" and "notify_on_release" are removed. Replacement | |
281 | * notification mechanism will be implemented. | |
282 | * | |
283 | * - "cgroup.clone_children" is removed. | |
284 | * | |
285 | * - "cgroup.subtree_populated" is available. Its value is 0 if the cgroup | |
286 | * and its descendants contain no task; otherwise, 1. The file also | |
287 | * generates kernfs notification which can be monitored through poll and | |
288 | * [di]notify when the value of the file changes. | |
289 | * | |
290 | * - cpuset: tasks will be kept in empty cpusets when hotplug happens and | |
291 | * take masks of ancestors with non-empty cpus/mems, instead of being | |
292 | * moved to an ancestor. | |
293 | * | |
294 | * - cpuset: a task can be moved into an empty cpuset, and again it takes | |
295 | * masks of ancestors. | |
296 | * | |
297 | * - memcg: use_hierarchy is on by default and the cgroup file for the flag | |
298 | * is not created. | |
299 | * | |
300 | * - blkcg: blk-throttle becomes properly hierarchical. | |
301 | * | |
302 | * - debug: disallowed on the default hierarchy. | |
303 | */ | |
304 | static bool cgroup_on_dfl(const struct cgroup *cgrp) | |
305 | { | |
306 | return cgrp->root == &cgrp_dfl_root; | |
307 | } | |
308 | ||
6fa4918d TH |
309 | /* IDR wrappers which synchronize using cgroup_idr_lock */ |
310 | static int cgroup_idr_alloc(struct idr *idr, void *ptr, int start, int end, | |
311 | gfp_t gfp_mask) | |
312 | { | |
313 | int ret; | |
314 | ||
315 | idr_preload(gfp_mask); | |
54504e97 | 316 | spin_lock_bh(&cgroup_idr_lock); |
d0164adc | 317 | ret = idr_alloc(idr, ptr, start, end, gfp_mask & ~__GFP_DIRECT_RECLAIM); |
54504e97 | 318 | spin_unlock_bh(&cgroup_idr_lock); |
6fa4918d TH |
319 | idr_preload_end(); |
320 | return ret; | |
321 | } | |
322 | ||
323 | static void *cgroup_idr_replace(struct idr *idr, void *ptr, int id) | |
324 | { | |
325 | void *ret; | |
326 | ||
54504e97 | 327 | spin_lock_bh(&cgroup_idr_lock); |
6fa4918d | 328 | ret = idr_replace(idr, ptr, id); |
54504e97 | 329 | spin_unlock_bh(&cgroup_idr_lock); |
6fa4918d TH |
330 | return ret; |
331 | } | |
332 | ||
333 | static void cgroup_idr_remove(struct idr *idr, int id) | |
334 | { | |
54504e97 | 335 | spin_lock_bh(&cgroup_idr_lock); |
6fa4918d | 336 | idr_remove(idr, id); |
54504e97 | 337 | spin_unlock_bh(&cgroup_idr_lock); |
6fa4918d TH |
338 | } |
339 | ||
d51f39b0 TH |
340 | static struct cgroup *cgroup_parent(struct cgroup *cgrp) |
341 | { | |
342 | struct cgroup_subsys_state *parent_css = cgrp->self.parent; | |
343 | ||
344 | if (parent_css) | |
345 | return container_of(parent_css, struct cgroup, self); | |
346 | return NULL; | |
347 | } | |
348 | ||
95109b62 TH |
349 | /** |
350 | * cgroup_css - obtain a cgroup's css for the specified subsystem | |
351 | * @cgrp: the cgroup of interest | |
9d800df1 | 352 | * @ss: the subsystem of interest (%NULL returns @cgrp->self) |
95109b62 | 353 | * |
ca8bdcaf TH |
354 | * Return @cgrp's css (cgroup_subsys_state) associated with @ss. This |
355 | * function must be called either under cgroup_mutex or rcu_read_lock() and | |
356 | * the caller is responsible for pinning the returned css if it wants to | |
357 | * keep accessing it outside the said locks. This function may return | |
358 | * %NULL if @cgrp doesn't have @subsys_id enabled. | |
95109b62 TH |
359 | */ |
360 | static struct cgroup_subsys_state *cgroup_css(struct cgroup *cgrp, | |
ca8bdcaf | 361 | struct cgroup_subsys *ss) |
95109b62 | 362 | { |
ca8bdcaf | 363 | if (ss) |
aec25020 | 364 | return rcu_dereference_check(cgrp->subsys[ss->id], |
ace2bee8 | 365 | lockdep_is_held(&cgroup_mutex)); |
ca8bdcaf | 366 | else |
9d800df1 | 367 | return &cgrp->self; |
95109b62 | 368 | } |
42809dd4 | 369 | |
aec3dfcb TH |
370 | /** |
371 | * cgroup_e_css - obtain a cgroup's effective css for the specified subsystem | |
372 | * @cgrp: the cgroup of interest | |
9d800df1 | 373 | * @ss: the subsystem of interest (%NULL returns @cgrp->self) |
aec3dfcb | 374 | * |
d0f702e6 | 375 | * Similar to cgroup_css() but returns the effective css, which is defined |
aec3dfcb TH |
376 | * as the matching css of the nearest ancestor including self which has @ss |
377 | * enabled. If @ss is associated with the hierarchy @cgrp is on, this | |
378 | * function is guaranteed to return non-NULL css. | |
379 | */ | |
380 | static struct cgroup_subsys_state *cgroup_e_css(struct cgroup *cgrp, | |
381 | struct cgroup_subsys *ss) | |
382 | { | |
383 | lockdep_assert_held(&cgroup_mutex); | |
384 | ||
385 | if (!ss) | |
9d800df1 | 386 | return &cgrp->self; |
aec3dfcb TH |
387 | |
388 | if (!(cgrp->root->subsys_mask & (1 << ss->id))) | |
389 | return NULL; | |
390 | ||
eeecbd19 TH |
391 | /* |
392 | * This function is used while updating css associations and thus | |
8699b776 | 393 | * can't test the csses directly. Use ->subtree_ss_mask. |
eeecbd19 | 394 | */ |
d51f39b0 | 395 | while (cgroup_parent(cgrp) && |
8699b776 | 396 | !(cgroup_parent(cgrp)->subtree_ss_mask & (1 << ss->id))) |
d51f39b0 | 397 | cgrp = cgroup_parent(cgrp); |
aec3dfcb TH |
398 | |
399 | return cgroup_css(cgrp, ss); | |
95109b62 | 400 | } |
42809dd4 | 401 | |
eeecbd19 TH |
402 | /** |
403 | * cgroup_get_e_css - get a cgroup's effective css for the specified subsystem | |
404 | * @cgrp: the cgroup of interest | |
405 | * @ss: the subsystem of interest | |
406 | * | |
407 | * Find and get the effective css of @cgrp for @ss. The effective css is | |
408 | * defined as the matching css of the nearest ancestor including self which | |
409 | * has @ss enabled. If @ss is not mounted on the hierarchy @cgrp is on, | |
410 | * the root css is returned, so this function always returns a valid css. | |
411 | * The returned css must be put using css_put(). | |
412 | */ | |
413 | struct cgroup_subsys_state *cgroup_get_e_css(struct cgroup *cgrp, | |
414 | struct cgroup_subsys *ss) | |
415 | { | |
416 | struct cgroup_subsys_state *css; | |
417 | ||
418 | rcu_read_lock(); | |
419 | ||
420 | do { | |
421 | css = cgroup_css(cgrp, ss); | |
422 | ||
423 | if (css && css_tryget_online(css)) | |
424 | goto out_unlock; | |
425 | cgrp = cgroup_parent(cgrp); | |
426 | } while (cgrp); | |
427 | ||
428 | css = init_css_set.subsys[ss->id]; | |
429 | css_get(css); | |
430 | out_unlock: | |
431 | rcu_read_unlock(); | |
432 | return css; | |
433 | } | |
434 | ||
ddbcc7e8 | 435 | /* convenient tests for these bits */ |
54766d4a | 436 | static inline bool cgroup_is_dead(const struct cgroup *cgrp) |
ddbcc7e8 | 437 | { |
184faf32 | 438 | return !(cgrp->self.flags & CSS_ONLINE); |
ddbcc7e8 PM |
439 | } |
440 | ||
052c3f3a TH |
441 | static void cgroup_get(struct cgroup *cgrp) |
442 | { | |
443 | WARN_ON_ONCE(cgroup_is_dead(cgrp)); | |
444 | css_get(&cgrp->self); | |
445 | } | |
446 | ||
447 | static bool cgroup_tryget(struct cgroup *cgrp) | |
448 | { | |
449 | return css_tryget(&cgrp->self); | |
450 | } | |
451 | ||
b4168640 | 452 | struct cgroup_subsys_state *of_css(struct kernfs_open_file *of) |
59f5296b | 453 | { |
2bd59d48 | 454 | struct cgroup *cgrp = of->kn->parent->priv; |
b4168640 | 455 | struct cftype *cft = of_cft(of); |
2bd59d48 TH |
456 | |
457 | /* | |
458 | * This is open and unprotected implementation of cgroup_css(). | |
459 | * seq_css() is only called from a kernfs file operation which has | |
460 | * an active reference on the file. Because all the subsystem | |
461 | * files are drained before a css is disassociated with a cgroup, | |
462 | * the matching css from the cgroup's subsys table is guaranteed to | |
463 | * be and stay valid until the enclosing operation is complete. | |
464 | */ | |
465 | if (cft->ss) | |
466 | return rcu_dereference_raw(cgrp->subsys[cft->ss->id]); | |
467 | else | |
9d800df1 | 468 | return &cgrp->self; |
59f5296b | 469 | } |
b4168640 | 470 | EXPORT_SYMBOL_GPL(of_css); |
59f5296b | 471 | |
e9685a03 | 472 | static int notify_on_release(const struct cgroup *cgrp) |
81a6a5cd | 473 | { |
bd89aabc | 474 | return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); |
81a6a5cd PM |
475 | } |
476 | ||
1c6727af TH |
477 | /** |
478 | * for_each_css - iterate all css's of a cgroup | |
479 | * @css: the iteration cursor | |
480 | * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end | |
481 | * @cgrp: the target cgroup to iterate css's of | |
482 | * | |
aec3dfcb | 483 | * Should be called under cgroup_[tree_]mutex. |
1c6727af TH |
484 | */ |
485 | #define for_each_css(css, ssid, cgrp) \ | |
486 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \ | |
487 | if (!((css) = rcu_dereference_check( \ | |
488 | (cgrp)->subsys[(ssid)], \ | |
489 | lockdep_is_held(&cgroup_mutex)))) { } \ | |
490 | else | |
491 | ||
aec3dfcb TH |
492 | /** |
493 | * for_each_e_css - iterate all effective css's of a cgroup | |
494 | * @css: the iteration cursor | |
495 | * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end | |
496 | * @cgrp: the target cgroup to iterate css's of | |
497 | * | |
498 | * Should be called under cgroup_[tree_]mutex. | |
499 | */ | |
500 | #define for_each_e_css(css, ssid, cgrp) \ | |
501 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \ | |
502 | if (!((css) = cgroup_e_css(cgrp, cgroup_subsys[(ssid)]))) \ | |
503 | ; \ | |
504 | else | |
505 | ||
30159ec7 | 506 | /** |
3ed80a62 | 507 | * for_each_subsys - iterate all enabled cgroup subsystems |
30159ec7 | 508 | * @ss: the iteration cursor |
780cd8b3 | 509 | * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end |
30159ec7 | 510 | */ |
780cd8b3 | 511 | #define for_each_subsys(ss, ssid) \ |
3ed80a62 TH |
512 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT && \ |
513 | (((ss) = cgroup_subsys[ssid]) || true); (ssid)++) | |
30159ec7 | 514 | |
cb4a3167 | 515 | /** |
b4e0eeaf | 516 | * do_each_subsys_mask - filter for_each_subsys with a bitmask |
cb4a3167 AS |
517 | * @ss: the iteration cursor |
518 | * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end | |
b4e0eeaf | 519 | * @ss_mask: the bitmask |
cb4a3167 AS |
520 | * |
521 | * The block will only run for cases where the ssid-th bit (1 << ssid) of | |
b4e0eeaf | 522 | * @ss_mask is set. |
cb4a3167 | 523 | */ |
b4e0eeaf TH |
524 | #define do_each_subsys_mask(ss, ssid, ss_mask) do { \ |
525 | unsigned long __ss_mask = (ss_mask); \ | |
526 | if (!CGROUP_SUBSYS_COUNT) { /* to avoid spurious gcc warning */ \ | |
4a705c5c | 527 | (ssid) = 0; \ |
b4e0eeaf TH |
528 | break; \ |
529 | } \ | |
530 | for_each_set_bit(ssid, &__ss_mask, CGROUP_SUBSYS_COUNT) { \ | |
531 | (ss) = cgroup_subsys[ssid]; \ | |
532 | { | |
533 | ||
534 | #define while_each_subsys_mask() \ | |
535 | } \ | |
536 | } \ | |
537 | } while (false) | |
cb4a3167 | 538 | |
985ed670 TH |
539 | /* iterate across the hierarchies */ |
540 | #define for_each_root(root) \ | |
5549c497 | 541 | list_for_each_entry((root), &cgroup_roots, root_list) |
ddbcc7e8 | 542 | |
f8f22e53 TH |
543 | /* iterate over child cgrps, lock should be held throughout iteration */ |
544 | #define cgroup_for_each_live_child(child, cgrp) \ | |
d5c419b6 | 545 | list_for_each_entry((child), &(cgrp)->self.children, self.sibling) \ |
8353da1f | 546 | if (({ lockdep_assert_held(&cgroup_mutex); \ |
f8f22e53 TH |
547 | cgroup_is_dead(child); })) \ |
548 | ; \ | |
549 | else | |
7ae1bad9 | 550 | |
81a6a5cd | 551 | static void cgroup_release_agent(struct work_struct *work); |
bd89aabc | 552 | static void check_for_release(struct cgroup *cgrp); |
81a6a5cd | 553 | |
69d0206c TH |
554 | /* |
555 | * A cgroup can be associated with multiple css_sets as different tasks may | |
556 | * belong to different cgroups on different hierarchies. In the other | |
557 | * direction, a css_set is naturally associated with multiple cgroups. | |
558 | * This M:N relationship is represented by the following link structure | |
559 | * which exists for each association and allows traversing the associations | |
560 | * from both sides. | |
561 | */ | |
562 | struct cgrp_cset_link { | |
563 | /* the cgroup and css_set this link associates */ | |
564 | struct cgroup *cgrp; | |
565 | struct css_set *cset; | |
566 | ||
567 | /* list of cgrp_cset_links anchored at cgrp->cset_links */ | |
568 | struct list_head cset_link; | |
569 | ||
570 | /* list of cgrp_cset_links anchored at css_set->cgrp_links */ | |
571 | struct list_head cgrp_link; | |
817929ec PM |
572 | }; |
573 | ||
172a2c06 TH |
574 | /* |
575 | * The default css_set - used by init and its children prior to any | |
817929ec PM |
576 | * hierarchies being mounted. It contains a pointer to the root state |
577 | * for each subsystem. Also used to anchor the list of css_sets. Not | |
578 | * reference-counted, to improve performance when child cgroups | |
579 | * haven't been created. | |
580 | */ | |
5024ae29 | 581 | struct css_set init_css_set = { |
172a2c06 TH |
582 | .refcount = ATOMIC_INIT(1), |
583 | .cgrp_links = LIST_HEAD_INIT(init_css_set.cgrp_links), | |
584 | .tasks = LIST_HEAD_INIT(init_css_set.tasks), | |
585 | .mg_tasks = LIST_HEAD_INIT(init_css_set.mg_tasks), | |
586 | .mg_preload_node = LIST_HEAD_INIT(init_css_set.mg_preload_node), | |
587 | .mg_node = LIST_HEAD_INIT(init_css_set.mg_node), | |
ed27b9f7 | 588 | .task_iters = LIST_HEAD_INIT(init_css_set.task_iters), |
172a2c06 | 589 | }; |
817929ec | 590 | |
172a2c06 | 591 | static int css_set_count = 1; /* 1 for init_css_set */ |
817929ec | 592 | |
0de0942d TH |
593 | /** |
594 | * css_set_populated - does a css_set contain any tasks? | |
595 | * @cset: target css_set | |
596 | */ | |
597 | static bool css_set_populated(struct css_set *cset) | |
598 | { | |
f0d9a5f1 | 599 | lockdep_assert_held(&css_set_lock); |
0de0942d TH |
600 | |
601 | return !list_empty(&cset->tasks) || !list_empty(&cset->mg_tasks); | |
602 | } | |
603 | ||
842b597e TH |
604 | /** |
605 | * cgroup_update_populated - updated populated count of a cgroup | |
606 | * @cgrp: the target cgroup | |
607 | * @populated: inc or dec populated count | |
608 | * | |
0de0942d TH |
609 | * One of the css_sets associated with @cgrp is either getting its first |
610 | * task or losing the last. Update @cgrp->populated_cnt accordingly. The | |
611 | * count is propagated towards root so that a given cgroup's populated_cnt | |
612 | * is zero iff the cgroup and all its descendants don't contain any tasks. | |
842b597e TH |
613 | * |
614 | * @cgrp's interface file "cgroup.populated" is zero if | |
615 | * @cgrp->populated_cnt is zero and 1 otherwise. When @cgrp->populated_cnt | |
616 | * changes from or to zero, userland is notified that the content of the | |
617 | * interface file has changed. This can be used to detect when @cgrp and | |
618 | * its descendants become populated or empty. | |
619 | */ | |
620 | static void cgroup_update_populated(struct cgroup *cgrp, bool populated) | |
621 | { | |
f0d9a5f1 | 622 | lockdep_assert_held(&css_set_lock); |
842b597e TH |
623 | |
624 | do { | |
625 | bool trigger; | |
626 | ||
627 | if (populated) | |
628 | trigger = !cgrp->populated_cnt++; | |
629 | else | |
630 | trigger = !--cgrp->populated_cnt; | |
631 | ||
632 | if (!trigger) | |
633 | break; | |
634 | ||
ad2ed2b3 | 635 | check_for_release(cgrp); |
6f60eade TH |
636 | cgroup_file_notify(&cgrp->events_file); |
637 | ||
d51f39b0 | 638 | cgrp = cgroup_parent(cgrp); |
842b597e TH |
639 | } while (cgrp); |
640 | } | |
641 | ||
0de0942d TH |
642 | /** |
643 | * css_set_update_populated - update populated state of a css_set | |
644 | * @cset: target css_set | |
645 | * @populated: whether @cset is populated or depopulated | |
646 | * | |
647 | * @cset is either getting the first task or losing the last. Update the | |
648 | * ->populated_cnt of all associated cgroups accordingly. | |
649 | */ | |
650 | static void css_set_update_populated(struct css_set *cset, bool populated) | |
651 | { | |
652 | struct cgrp_cset_link *link; | |
653 | ||
f0d9a5f1 | 654 | lockdep_assert_held(&css_set_lock); |
0de0942d TH |
655 | |
656 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) | |
657 | cgroup_update_populated(link->cgrp, populated); | |
658 | } | |
659 | ||
f6d7d049 TH |
660 | /** |
661 | * css_set_move_task - move a task from one css_set to another | |
662 | * @task: task being moved | |
663 | * @from_cset: css_set @task currently belongs to (may be NULL) | |
664 | * @to_cset: new css_set @task is being moved to (may be NULL) | |
665 | * @use_mg_tasks: move to @to_cset->mg_tasks instead of ->tasks | |
666 | * | |
667 | * Move @task from @from_cset to @to_cset. If @task didn't belong to any | |
668 | * css_set, @from_cset can be NULL. If @task is being disassociated | |
669 | * instead of moved, @to_cset can be NULL. | |
670 | * | |
ed27b9f7 TH |
671 | * This function automatically handles populated_cnt updates and |
672 | * css_task_iter adjustments but the caller is responsible for managing | |
673 | * @from_cset and @to_cset's reference counts. | |
f6d7d049 TH |
674 | */ |
675 | static void css_set_move_task(struct task_struct *task, | |
676 | struct css_set *from_cset, struct css_set *to_cset, | |
677 | bool use_mg_tasks) | |
678 | { | |
f0d9a5f1 | 679 | lockdep_assert_held(&css_set_lock); |
f6d7d049 | 680 | |
20b454a6 TH |
681 | if (to_cset && !css_set_populated(to_cset)) |
682 | css_set_update_populated(to_cset, true); | |
683 | ||
f6d7d049 | 684 | if (from_cset) { |
ed27b9f7 TH |
685 | struct css_task_iter *it, *pos; |
686 | ||
f6d7d049 | 687 | WARN_ON_ONCE(list_empty(&task->cg_list)); |
ed27b9f7 TH |
688 | |
689 | /* | |
690 | * @task is leaving, advance task iterators which are | |
691 | * pointing to it so that they can resume at the next | |
692 | * position. Advancing an iterator might remove it from | |
693 | * the list, use safe walk. See css_task_iter_advance*() | |
694 | * for details. | |
695 | */ | |
696 | list_for_each_entry_safe(it, pos, &from_cset->task_iters, | |
697 | iters_node) | |
698 | if (it->task_pos == &task->cg_list) | |
699 | css_task_iter_advance(it); | |
700 | ||
f6d7d049 TH |
701 | list_del_init(&task->cg_list); |
702 | if (!css_set_populated(from_cset)) | |
703 | css_set_update_populated(from_cset, false); | |
704 | } else { | |
705 | WARN_ON_ONCE(!list_empty(&task->cg_list)); | |
706 | } | |
707 | ||
708 | if (to_cset) { | |
709 | /* | |
710 | * We are synchronized through cgroup_threadgroup_rwsem | |
711 | * against PF_EXITING setting such that we can't race | |
712 | * against cgroup_exit() changing the css_set to | |
713 | * init_css_set and dropping the old one. | |
714 | */ | |
715 | WARN_ON_ONCE(task->flags & PF_EXITING); | |
716 | ||
f6d7d049 TH |
717 | rcu_assign_pointer(task->cgroups, to_cset); |
718 | list_add_tail(&task->cg_list, use_mg_tasks ? &to_cset->mg_tasks : | |
719 | &to_cset->tasks); | |
720 | } | |
721 | } | |
722 | ||
7717f7ba PM |
723 | /* |
724 | * hash table for cgroup groups. This improves the performance to find | |
725 | * an existing css_set. This hash doesn't (currently) take into | |
726 | * account cgroups in empty hierarchies. | |
727 | */ | |
472b1053 | 728 | #define CSS_SET_HASH_BITS 7 |
0ac801fe | 729 | static DEFINE_HASHTABLE(css_set_table, CSS_SET_HASH_BITS); |
472b1053 | 730 | |
0ac801fe | 731 | static unsigned long css_set_hash(struct cgroup_subsys_state *css[]) |
472b1053 | 732 | { |
0ac801fe | 733 | unsigned long key = 0UL; |
30159ec7 TH |
734 | struct cgroup_subsys *ss; |
735 | int i; | |
472b1053 | 736 | |
30159ec7 | 737 | for_each_subsys(ss, i) |
0ac801fe LZ |
738 | key += (unsigned long)css[i]; |
739 | key = (key >> 16) ^ key; | |
472b1053 | 740 | |
0ac801fe | 741 | return key; |
472b1053 LZ |
742 | } |
743 | ||
a25eb52e | 744 | static void put_css_set_locked(struct css_set *cset) |
b4f48b63 | 745 | { |
69d0206c | 746 | struct cgrp_cset_link *link, *tmp_link; |
2d8f243a TH |
747 | struct cgroup_subsys *ss; |
748 | int ssid; | |
5abb8855 | 749 | |
f0d9a5f1 | 750 | lockdep_assert_held(&css_set_lock); |
89c5509b TH |
751 | |
752 | if (!atomic_dec_and_test(&cset->refcount)) | |
146aa1bd | 753 | return; |
81a6a5cd | 754 | |
53254f90 TH |
755 | /* This css_set is dead. unlink it and release cgroup and css refs */ |
756 | for_each_subsys(ss, ssid) { | |
2d8f243a | 757 | list_del(&cset->e_cset_node[ssid]); |
53254f90 TH |
758 | css_put(cset->subsys[ssid]); |
759 | } | |
5abb8855 | 760 | hash_del(&cset->hlist); |
2c6ab6d2 PM |
761 | css_set_count--; |
762 | ||
69d0206c | 763 | list_for_each_entry_safe(link, tmp_link, &cset->cgrp_links, cgrp_link) { |
69d0206c TH |
764 | list_del(&link->cset_link); |
765 | list_del(&link->cgrp_link); | |
2ceb231b TH |
766 | if (cgroup_parent(link->cgrp)) |
767 | cgroup_put(link->cgrp); | |
2c6ab6d2 | 768 | kfree(link); |
81a6a5cd | 769 | } |
2c6ab6d2 | 770 | |
5abb8855 | 771 | kfree_rcu(cset, rcu_head); |
b4f48b63 PM |
772 | } |
773 | ||
a25eb52e | 774 | static void put_css_set(struct css_set *cset) |
89c5509b TH |
775 | { |
776 | /* | |
777 | * Ensure that the refcount doesn't hit zero while any readers | |
778 | * can see it. Similar to atomic_dec_and_lock(), but for an | |
779 | * rwlock | |
780 | */ | |
781 | if (atomic_add_unless(&cset->refcount, -1, 1)) | |
782 | return; | |
783 | ||
f0d9a5f1 | 784 | spin_lock_bh(&css_set_lock); |
a25eb52e | 785 | put_css_set_locked(cset); |
f0d9a5f1 | 786 | spin_unlock_bh(&css_set_lock); |
89c5509b TH |
787 | } |
788 | ||
817929ec PM |
789 | /* |
790 | * refcounted get/put for css_set objects | |
791 | */ | |
5abb8855 | 792 | static inline void get_css_set(struct css_set *cset) |
817929ec | 793 | { |
5abb8855 | 794 | atomic_inc(&cset->refcount); |
817929ec PM |
795 | } |
796 | ||
b326f9d0 | 797 | /** |
7717f7ba | 798 | * compare_css_sets - helper function for find_existing_css_set(). |
5abb8855 TH |
799 | * @cset: candidate css_set being tested |
800 | * @old_cset: existing css_set for a task | |
7717f7ba PM |
801 | * @new_cgrp: cgroup that's being entered by the task |
802 | * @template: desired set of css pointers in css_set (pre-calculated) | |
803 | * | |
6f4b7e63 | 804 | * Returns true if "cset" matches "old_cset" except for the hierarchy |
7717f7ba PM |
805 | * which "new_cgrp" belongs to, for which it should match "new_cgrp". |
806 | */ | |
5abb8855 TH |
807 | static bool compare_css_sets(struct css_set *cset, |
808 | struct css_set *old_cset, | |
7717f7ba PM |
809 | struct cgroup *new_cgrp, |
810 | struct cgroup_subsys_state *template[]) | |
811 | { | |
812 | struct list_head *l1, *l2; | |
813 | ||
aec3dfcb TH |
814 | /* |
815 | * On the default hierarchy, there can be csets which are | |
816 | * associated with the same set of cgroups but different csses. | |
817 | * Let's first ensure that csses match. | |
818 | */ | |
819 | if (memcmp(template, cset->subsys, sizeof(cset->subsys))) | |
7717f7ba | 820 | return false; |
7717f7ba PM |
821 | |
822 | /* | |
823 | * Compare cgroup pointers in order to distinguish between | |
aec3dfcb TH |
824 | * different cgroups in hierarchies. As different cgroups may |
825 | * share the same effective css, this comparison is always | |
826 | * necessary. | |
7717f7ba | 827 | */ |
69d0206c TH |
828 | l1 = &cset->cgrp_links; |
829 | l2 = &old_cset->cgrp_links; | |
7717f7ba | 830 | while (1) { |
69d0206c | 831 | struct cgrp_cset_link *link1, *link2; |
5abb8855 | 832 | struct cgroup *cgrp1, *cgrp2; |
7717f7ba PM |
833 | |
834 | l1 = l1->next; | |
835 | l2 = l2->next; | |
836 | /* See if we reached the end - both lists are equal length. */ | |
69d0206c TH |
837 | if (l1 == &cset->cgrp_links) { |
838 | BUG_ON(l2 != &old_cset->cgrp_links); | |
7717f7ba PM |
839 | break; |
840 | } else { | |
69d0206c | 841 | BUG_ON(l2 == &old_cset->cgrp_links); |
7717f7ba PM |
842 | } |
843 | /* Locate the cgroups associated with these links. */ | |
69d0206c TH |
844 | link1 = list_entry(l1, struct cgrp_cset_link, cgrp_link); |
845 | link2 = list_entry(l2, struct cgrp_cset_link, cgrp_link); | |
846 | cgrp1 = link1->cgrp; | |
847 | cgrp2 = link2->cgrp; | |
7717f7ba | 848 | /* Hierarchies should be linked in the same order. */ |
5abb8855 | 849 | BUG_ON(cgrp1->root != cgrp2->root); |
7717f7ba PM |
850 | |
851 | /* | |
852 | * If this hierarchy is the hierarchy of the cgroup | |
853 | * that's changing, then we need to check that this | |
854 | * css_set points to the new cgroup; if it's any other | |
855 | * hierarchy, then this css_set should point to the | |
856 | * same cgroup as the old css_set. | |
857 | */ | |
5abb8855 TH |
858 | if (cgrp1->root == new_cgrp->root) { |
859 | if (cgrp1 != new_cgrp) | |
7717f7ba PM |
860 | return false; |
861 | } else { | |
5abb8855 | 862 | if (cgrp1 != cgrp2) |
7717f7ba PM |
863 | return false; |
864 | } | |
865 | } | |
866 | return true; | |
867 | } | |
868 | ||
b326f9d0 TH |
869 | /** |
870 | * find_existing_css_set - init css array and find the matching css_set | |
871 | * @old_cset: the css_set that we're using before the cgroup transition | |
872 | * @cgrp: the cgroup that we're moving into | |
873 | * @template: out param for the new set of csses, should be clear on entry | |
817929ec | 874 | */ |
5abb8855 TH |
875 | static struct css_set *find_existing_css_set(struct css_set *old_cset, |
876 | struct cgroup *cgrp, | |
877 | struct cgroup_subsys_state *template[]) | |
b4f48b63 | 878 | { |
3dd06ffa | 879 | struct cgroup_root *root = cgrp->root; |
30159ec7 | 880 | struct cgroup_subsys *ss; |
5abb8855 | 881 | struct css_set *cset; |
0ac801fe | 882 | unsigned long key; |
b326f9d0 | 883 | int i; |
817929ec | 884 | |
aae8aab4 BB |
885 | /* |
886 | * Build the set of subsystem state objects that we want to see in the | |
887 | * new css_set. while subsystems can change globally, the entries here | |
888 | * won't change, so no need for locking. | |
889 | */ | |
30159ec7 | 890 | for_each_subsys(ss, i) { |
f392e51c | 891 | if (root->subsys_mask & (1UL << i)) { |
aec3dfcb TH |
892 | /* |
893 | * @ss is in this hierarchy, so we want the | |
894 | * effective css from @cgrp. | |
895 | */ | |
896 | template[i] = cgroup_e_css(cgrp, ss); | |
817929ec | 897 | } else { |
aec3dfcb TH |
898 | /* |
899 | * @ss is not in this hierarchy, so we don't want | |
900 | * to change the css. | |
901 | */ | |
5abb8855 | 902 | template[i] = old_cset->subsys[i]; |
817929ec PM |
903 | } |
904 | } | |
905 | ||
0ac801fe | 906 | key = css_set_hash(template); |
5abb8855 TH |
907 | hash_for_each_possible(css_set_table, cset, hlist, key) { |
908 | if (!compare_css_sets(cset, old_cset, cgrp, template)) | |
7717f7ba PM |
909 | continue; |
910 | ||
911 | /* This css_set matches what we need */ | |
5abb8855 | 912 | return cset; |
472b1053 | 913 | } |
817929ec PM |
914 | |
915 | /* No existing cgroup group matched */ | |
916 | return NULL; | |
917 | } | |
918 | ||
69d0206c | 919 | static void free_cgrp_cset_links(struct list_head *links_to_free) |
36553434 | 920 | { |
69d0206c | 921 | struct cgrp_cset_link *link, *tmp_link; |
36553434 | 922 | |
69d0206c TH |
923 | list_for_each_entry_safe(link, tmp_link, links_to_free, cset_link) { |
924 | list_del(&link->cset_link); | |
36553434 LZ |
925 | kfree(link); |
926 | } | |
927 | } | |
928 | ||
69d0206c TH |
929 | /** |
930 | * allocate_cgrp_cset_links - allocate cgrp_cset_links | |
931 | * @count: the number of links to allocate | |
932 | * @tmp_links: list_head the allocated links are put on | |
933 | * | |
934 | * Allocate @count cgrp_cset_link structures and chain them on @tmp_links | |
935 | * through ->cset_link. Returns 0 on success or -errno. | |
817929ec | 936 | */ |
69d0206c | 937 | static int allocate_cgrp_cset_links(int count, struct list_head *tmp_links) |
817929ec | 938 | { |
69d0206c | 939 | struct cgrp_cset_link *link; |
817929ec | 940 | int i; |
69d0206c TH |
941 | |
942 | INIT_LIST_HEAD(tmp_links); | |
943 | ||
817929ec | 944 | for (i = 0; i < count; i++) { |
f4f4be2b | 945 | link = kzalloc(sizeof(*link), GFP_KERNEL); |
817929ec | 946 | if (!link) { |
69d0206c | 947 | free_cgrp_cset_links(tmp_links); |
817929ec PM |
948 | return -ENOMEM; |
949 | } | |
69d0206c | 950 | list_add(&link->cset_link, tmp_links); |
817929ec PM |
951 | } |
952 | return 0; | |
953 | } | |
954 | ||
c12f65d4 LZ |
955 | /** |
956 | * link_css_set - a helper function to link a css_set to a cgroup | |
69d0206c | 957 | * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links() |
5abb8855 | 958 | * @cset: the css_set to be linked |
c12f65d4 LZ |
959 | * @cgrp: the destination cgroup |
960 | */ | |
69d0206c TH |
961 | static void link_css_set(struct list_head *tmp_links, struct css_set *cset, |
962 | struct cgroup *cgrp) | |
c12f65d4 | 963 | { |
69d0206c | 964 | struct cgrp_cset_link *link; |
c12f65d4 | 965 | |
69d0206c | 966 | BUG_ON(list_empty(tmp_links)); |
6803c006 TH |
967 | |
968 | if (cgroup_on_dfl(cgrp)) | |
969 | cset->dfl_cgrp = cgrp; | |
970 | ||
69d0206c TH |
971 | link = list_first_entry(tmp_links, struct cgrp_cset_link, cset_link); |
972 | link->cset = cset; | |
7717f7ba | 973 | link->cgrp = cgrp; |
842b597e | 974 | |
7717f7ba | 975 | /* |
389b9c1b TH |
976 | * Always add links to the tail of the lists so that the lists are |
977 | * in choronological order. | |
7717f7ba | 978 | */ |
389b9c1b | 979 | list_move_tail(&link->cset_link, &cgrp->cset_links); |
69d0206c | 980 | list_add_tail(&link->cgrp_link, &cset->cgrp_links); |
2ceb231b TH |
981 | |
982 | if (cgroup_parent(cgrp)) | |
983 | cgroup_get(cgrp); | |
c12f65d4 LZ |
984 | } |
985 | ||
b326f9d0 TH |
986 | /** |
987 | * find_css_set - return a new css_set with one cgroup updated | |
988 | * @old_cset: the baseline css_set | |
989 | * @cgrp: the cgroup to be updated | |
990 | * | |
991 | * Return a new css_set that's equivalent to @old_cset, but with @cgrp | |
992 | * substituted into the appropriate hierarchy. | |
817929ec | 993 | */ |
5abb8855 TH |
994 | static struct css_set *find_css_set(struct css_set *old_cset, |
995 | struct cgroup *cgrp) | |
817929ec | 996 | { |
b326f9d0 | 997 | struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT] = { }; |
5abb8855 | 998 | struct css_set *cset; |
69d0206c TH |
999 | struct list_head tmp_links; |
1000 | struct cgrp_cset_link *link; | |
2d8f243a | 1001 | struct cgroup_subsys *ss; |
0ac801fe | 1002 | unsigned long key; |
2d8f243a | 1003 | int ssid; |
472b1053 | 1004 | |
b326f9d0 TH |
1005 | lockdep_assert_held(&cgroup_mutex); |
1006 | ||
817929ec PM |
1007 | /* First see if we already have a cgroup group that matches |
1008 | * the desired set */ | |
f0d9a5f1 | 1009 | spin_lock_bh(&css_set_lock); |
5abb8855 TH |
1010 | cset = find_existing_css_set(old_cset, cgrp, template); |
1011 | if (cset) | |
1012 | get_css_set(cset); | |
f0d9a5f1 | 1013 | spin_unlock_bh(&css_set_lock); |
817929ec | 1014 | |
5abb8855 TH |
1015 | if (cset) |
1016 | return cset; | |
817929ec | 1017 | |
f4f4be2b | 1018 | cset = kzalloc(sizeof(*cset), GFP_KERNEL); |
5abb8855 | 1019 | if (!cset) |
817929ec PM |
1020 | return NULL; |
1021 | ||
69d0206c | 1022 | /* Allocate all the cgrp_cset_link objects that we'll need */ |
9871bf95 | 1023 | if (allocate_cgrp_cset_links(cgroup_root_count, &tmp_links) < 0) { |
5abb8855 | 1024 | kfree(cset); |
817929ec PM |
1025 | return NULL; |
1026 | } | |
1027 | ||
5abb8855 | 1028 | atomic_set(&cset->refcount, 1); |
69d0206c | 1029 | INIT_LIST_HEAD(&cset->cgrp_links); |
5abb8855 | 1030 | INIT_LIST_HEAD(&cset->tasks); |
c7561128 | 1031 | INIT_LIST_HEAD(&cset->mg_tasks); |
1958d2d5 | 1032 | INIT_LIST_HEAD(&cset->mg_preload_node); |
b3dc094e | 1033 | INIT_LIST_HEAD(&cset->mg_node); |
ed27b9f7 | 1034 | INIT_LIST_HEAD(&cset->task_iters); |
5abb8855 | 1035 | INIT_HLIST_NODE(&cset->hlist); |
817929ec PM |
1036 | |
1037 | /* Copy the set of subsystem state objects generated in | |
1038 | * find_existing_css_set() */ | |
5abb8855 | 1039 | memcpy(cset->subsys, template, sizeof(cset->subsys)); |
817929ec | 1040 | |
f0d9a5f1 | 1041 | spin_lock_bh(&css_set_lock); |
817929ec | 1042 | /* Add reference counts and links from the new css_set. */ |
69d0206c | 1043 | list_for_each_entry(link, &old_cset->cgrp_links, cgrp_link) { |
7717f7ba | 1044 | struct cgroup *c = link->cgrp; |
69d0206c | 1045 | |
7717f7ba PM |
1046 | if (c->root == cgrp->root) |
1047 | c = cgrp; | |
69d0206c | 1048 | link_css_set(&tmp_links, cset, c); |
7717f7ba | 1049 | } |
817929ec | 1050 | |
69d0206c | 1051 | BUG_ON(!list_empty(&tmp_links)); |
817929ec | 1052 | |
817929ec | 1053 | css_set_count++; |
472b1053 | 1054 | |
2d8f243a | 1055 | /* Add @cset to the hash table */ |
5abb8855 TH |
1056 | key = css_set_hash(cset->subsys); |
1057 | hash_add(css_set_table, &cset->hlist, key); | |
472b1053 | 1058 | |
53254f90 TH |
1059 | for_each_subsys(ss, ssid) { |
1060 | struct cgroup_subsys_state *css = cset->subsys[ssid]; | |
1061 | ||
2d8f243a | 1062 | list_add_tail(&cset->e_cset_node[ssid], |
53254f90 TH |
1063 | &css->cgroup->e_csets[ssid]); |
1064 | css_get(css); | |
1065 | } | |
2d8f243a | 1066 | |
f0d9a5f1 | 1067 | spin_unlock_bh(&css_set_lock); |
817929ec | 1068 | |
5abb8855 | 1069 | return cset; |
b4f48b63 PM |
1070 | } |
1071 | ||
3dd06ffa | 1072 | static struct cgroup_root *cgroup_root_from_kf(struct kernfs_root *kf_root) |
7717f7ba | 1073 | { |
3dd06ffa | 1074 | struct cgroup *root_cgrp = kf_root->kn->priv; |
2bd59d48 | 1075 | |
3dd06ffa | 1076 | return root_cgrp->root; |
2bd59d48 TH |
1077 | } |
1078 | ||
3dd06ffa | 1079 | static int cgroup_init_root_id(struct cgroup_root *root) |
f2e85d57 TH |
1080 | { |
1081 | int id; | |
1082 | ||
1083 | lockdep_assert_held(&cgroup_mutex); | |
1084 | ||
985ed670 | 1085 | id = idr_alloc_cyclic(&cgroup_hierarchy_idr, root, 0, 0, GFP_KERNEL); |
f2e85d57 TH |
1086 | if (id < 0) |
1087 | return id; | |
1088 | ||
1089 | root->hierarchy_id = id; | |
1090 | return 0; | |
1091 | } | |
1092 | ||
3dd06ffa | 1093 | static void cgroup_exit_root_id(struct cgroup_root *root) |
f2e85d57 TH |
1094 | { |
1095 | lockdep_assert_held(&cgroup_mutex); | |
1096 | ||
1097 | if (root->hierarchy_id) { | |
1098 | idr_remove(&cgroup_hierarchy_idr, root->hierarchy_id); | |
1099 | root->hierarchy_id = 0; | |
1100 | } | |
1101 | } | |
1102 | ||
3dd06ffa | 1103 | static void cgroup_free_root(struct cgroup_root *root) |
f2e85d57 TH |
1104 | { |
1105 | if (root) { | |
d0f702e6 | 1106 | /* hierarchy ID should already have been released */ |
f2e85d57 TH |
1107 | WARN_ON_ONCE(root->hierarchy_id); |
1108 | ||
1109 | idr_destroy(&root->cgroup_idr); | |
1110 | kfree(root); | |
1111 | } | |
1112 | } | |
1113 | ||
3dd06ffa | 1114 | static void cgroup_destroy_root(struct cgroup_root *root) |
59f5296b | 1115 | { |
3dd06ffa | 1116 | struct cgroup *cgrp = &root->cgrp; |
f2e85d57 | 1117 | struct cgrp_cset_link *link, *tmp_link; |
f2e85d57 | 1118 | |
2bd59d48 | 1119 | mutex_lock(&cgroup_mutex); |
f2e85d57 | 1120 | |
776f02fa | 1121 | BUG_ON(atomic_read(&root->nr_cgrps)); |
d5c419b6 | 1122 | BUG_ON(!list_empty(&cgrp->self.children)); |
f2e85d57 | 1123 | |
f2e85d57 | 1124 | /* Rebind all subsystems back to the default hierarchy */ |
f392e51c | 1125 | rebind_subsystems(&cgrp_dfl_root, root->subsys_mask); |
7717f7ba | 1126 | |
7717f7ba | 1127 | /* |
f2e85d57 TH |
1128 | * Release all the links from cset_links to this hierarchy's |
1129 | * root cgroup | |
7717f7ba | 1130 | */ |
f0d9a5f1 | 1131 | spin_lock_bh(&css_set_lock); |
f2e85d57 TH |
1132 | |
1133 | list_for_each_entry_safe(link, tmp_link, &cgrp->cset_links, cset_link) { | |
1134 | list_del(&link->cset_link); | |
1135 | list_del(&link->cgrp_link); | |
1136 | kfree(link); | |
1137 | } | |
f0d9a5f1 TH |
1138 | |
1139 | spin_unlock_bh(&css_set_lock); | |
f2e85d57 TH |
1140 | |
1141 | if (!list_empty(&root->root_list)) { | |
1142 | list_del(&root->root_list); | |
1143 | cgroup_root_count--; | |
1144 | } | |
1145 | ||
1146 | cgroup_exit_root_id(root); | |
1147 | ||
1148 | mutex_unlock(&cgroup_mutex); | |
f2e85d57 | 1149 | |
2bd59d48 | 1150 | kernfs_destroy_root(root->kf_root); |
f2e85d57 TH |
1151 | cgroup_free_root(root); |
1152 | } | |
1153 | ||
ceb6a081 TH |
1154 | /* look up cgroup associated with given css_set on the specified hierarchy */ |
1155 | static struct cgroup *cset_cgroup_from_root(struct css_set *cset, | |
3dd06ffa | 1156 | struct cgroup_root *root) |
7717f7ba | 1157 | { |
7717f7ba PM |
1158 | struct cgroup *res = NULL; |
1159 | ||
96d365e0 | 1160 | lockdep_assert_held(&cgroup_mutex); |
f0d9a5f1 | 1161 | lockdep_assert_held(&css_set_lock); |
96d365e0 | 1162 | |
5abb8855 | 1163 | if (cset == &init_css_set) { |
3dd06ffa | 1164 | res = &root->cgrp; |
7717f7ba | 1165 | } else { |
69d0206c TH |
1166 | struct cgrp_cset_link *link; |
1167 | ||
1168 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) { | |
7717f7ba | 1169 | struct cgroup *c = link->cgrp; |
69d0206c | 1170 | |
7717f7ba PM |
1171 | if (c->root == root) { |
1172 | res = c; | |
1173 | break; | |
1174 | } | |
1175 | } | |
1176 | } | |
96d365e0 | 1177 | |
7717f7ba PM |
1178 | BUG_ON(!res); |
1179 | return res; | |
1180 | } | |
1181 | ||
ddbcc7e8 | 1182 | /* |
ceb6a081 | 1183 | * Return the cgroup for "task" from the given hierarchy. Must be |
f0d9a5f1 | 1184 | * called with cgroup_mutex and css_set_lock held. |
ceb6a081 TH |
1185 | */ |
1186 | static struct cgroup *task_cgroup_from_root(struct task_struct *task, | |
3dd06ffa | 1187 | struct cgroup_root *root) |
ceb6a081 TH |
1188 | { |
1189 | /* | |
1190 | * No need to lock the task - since we hold cgroup_mutex the | |
1191 | * task can't change groups, so the only thing that can happen | |
1192 | * is that it exits and its css is set back to init_css_set. | |
1193 | */ | |
1194 | return cset_cgroup_from_root(task_css_set(task), root); | |
1195 | } | |
1196 | ||
ddbcc7e8 | 1197 | /* |
ddbcc7e8 PM |
1198 | * A task must hold cgroup_mutex to modify cgroups. |
1199 | * | |
1200 | * Any task can increment and decrement the count field without lock. | |
1201 | * So in general, code holding cgroup_mutex can't rely on the count | |
1202 | * field not changing. However, if the count goes to zero, then only | |
956db3ca | 1203 | * cgroup_attach_task() can increment it again. Because a count of zero |
ddbcc7e8 PM |
1204 | * means that no tasks are currently attached, therefore there is no |
1205 | * way a task attached to that cgroup can fork (the other way to | |
1206 | * increment the count). So code holding cgroup_mutex can safely | |
1207 | * assume that if the count is zero, it will stay zero. Similarly, if | |
1208 | * a task holds cgroup_mutex on a cgroup with zero count, it | |
1209 | * knows that the cgroup won't be removed, as cgroup_rmdir() | |
1210 | * needs that mutex. | |
1211 | * | |
ddbcc7e8 PM |
1212 | * A cgroup can only be deleted if both its 'count' of using tasks |
1213 | * is zero, and its list of 'children' cgroups is empty. Since all | |
1214 | * tasks in the system use _some_ cgroup, and since there is always at | |
3dd06ffa | 1215 | * least one task in the system (init, pid == 1), therefore, root cgroup |
ddbcc7e8 | 1216 | * always has either children cgroups and/or using tasks. So we don't |
3dd06ffa | 1217 | * need a special hack to ensure that root cgroup cannot be deleted. |
ddbcc7e8 PM |
1218 | * |
1219 | * P.S. One more locking exception. RCU is used to guard the | |
956db3ca | 1220 | * update of a tasks cgroup pointer by cgroup_attach_task() |
ddbcc7e8 PM |
1221 | */ |
1222 | ||
2bd59d48 | 1223 | static struct kernfs_syscall_ops cgroup_kf_syscall_ops; |
828c0950 | 1224 | static const struct file_operations proc_cgroupstats_operations; |
a424316c | 1225 | |
8d7e6fb0 TH |
1226 | static char *cgroup_file_name(struct cgroup *cgrp, const struct cftype *cft, |
1227 | char *buf) | |
ddbcc7e8 | 1228 | { |
3e1d2eed TH |
1229 | struct cgroup_subsys *ss = cft->ss; |
1230 | ||
8d7e6fb0 TH |
1231 | if (cft->ss && !(cft->flags & CFTYPE_NO_PREFIX) && |
1232 | !(cgrp->root->flags & CGRP_ROOT_NOPREFIX)) | |
1233 | snprintf(buf, CGROUP_FILE_NAME_MAX, "%s.%s", | |
3e1d2eed TH |
1234 | cgroup_on_dfl(cgrp) ? ss->name : ss->legacy_name, |
1235 | cft->name); | |
8d7e6fb0 TH |
1236 | else |
1237 | strncpy(buf, cft->name, CGROUP_FILE_NAME_MAX); | |
1238 | return buf; | |
ddbcc7e8 PM |
1239 | } |
1240 | ||
f2e85d57 TH |
1241 | /** |
1242 | * cgroup_file_mode - deduce file mode of a control file | |
1243 | * @cft: the control file in question | |
1244 | * | |
7dbdb199 | 1245 | * S_IRUGO for read, S_IWUSR for write. |
f2e85d57 TH |
1246 | */ |
1247 | static umode_t cgroup_file_mode(const struct cftype *cft) | |
65dff759 | 1248 | { |
f2e85d57 | 1249 | umode_t mode = 0; |
65dff759 | 1250 | |
f2e85d57 TH |
1251 | if (cft->read_u64 || cft->read_s64 || cft->seq_show) |
1252 | mode |= S_IRUGO; | |
1253 | ||
7dbdb199 TH |
1254 | if (cft->write_u64 || cft->write_s64 || cft->write) { |
1255 | if (cft->flags & CFTYPE_WORLD_WRITABLE) | |
1256 | mode |= S_IWUGO; | |
1257 | else | |
1258 | mode |= S_IWUSR; | |
1259 | } | |
f2e85d57 TH |
1260 | |
1261 | return mode; | |
65dff759 LZ |
1262 | } |
1263 | ||
af0ba678 | 1264 | /** |
8699b776 | 1265 | * cgroup_calc_subtree_ss_mask - calculate subtree_ss_mask |
af0ba678 | 1266 | * @cgrp: the target cgroup |
0f060deb | 1267 | * @subtree_control: the new subtree_control mask to consider |
af0ba678 TH |
1268 | * |
1269 | * On the default hierarchy, a subsystem may request other subsystems to be | |
1270 | * enabled together through its ->depends_on mask. In such cases, more | |
1271 | * subsystems than specified in "cgroup.subtree_control" may be enabled. | |
1272 | * | |
0f060deb TH |
1273 | * This function calculates which subsystems need to be enabled if |
1274 | * @subtree_control is to be applied to @cgrp. The returned mask is always | |
1275 | * a superset of @subtree_control and follows the usual hierarchy rules. | |
af0ba678 | 1276 | */ |
6e5c8307 | 1277 | static u16 cgroup_calc_subtree_ss_mask(struct cgroup *cgrp, u16 subtree_control) |
667c2491 | 1278 | { |
af0ba678 | 1279 | struct cgroup *parent = cgroup_parent(cgrp); |
6e5c8307 | 1280 | u16 cur_ss_mask = subtree_control; |
af0ba678 TH |
1281 | struct cgroup_subsys *ss; |
1282 | int ssid; | |
1283 | ||
1284 | lockdep_assert_held(&cgroup_mutex); | |
1285 | ||
0f060deb TH |
1286 | if (!cgroup_on_dfl(cgrp)) |
1287 | return cur_ss_mask; | |
af0ba678 TH |
1288 | |
1289 | while (true) { | |
6e5c8307 | 1290 | u16 new_ss_mask = cur_ss_mask; |
af0ba678 | 1291 | |
b4e0eeaf | 1292 | do_each_subsys_mask(ss, ssid, cur_ss_mask) { |
a966a4ed | 1293 | new_ss_mask |= ss->depends_on; |
b4e0eeaf | 1294 | } while_each_subsys_mask(); |
af0ba678 TH |
1295 | |
1296 | /* | |
1297 | * Mask out subsystems which aren't available. This can | |
1298 | * happen only if some depended-upon subsystems were bound | |
1299 | * to non-default hierarchies. | |
1300 | */ | |
1301 | if (parent) | |
8699b776 | 1302 | new_ss_mask &= parent->subtree_ss_mask; |
af0ba678 TH |
1303 | else |
1304 | new_ss_mask &= cgrp->root->subsys_mask; | |
1305 | ||
1306 | if (new_ss_mask == cur_ss_mask) | |
1307 | break; | |
1308 | cur_ss_mask = new_ss_mask; | |
1309 | } | |
1310 | ||
0f060deb TH |
1311 | return cur_ss_mask; |
1312 | } | |
1313 | ||
1314 | /** | |
8699b776 | 1315 | * cgroup_refresh_subtree_ss_mask - update subtree_ss_mask |
0f060deb TH |
1316 | * @cgrp: the target cgroup |
1317 | * | |
8699b776 TH |
1318 | * Update @cgrp->subtree_ss_mask according to the current |
1319 | * @cgrp->subtree_control using cgroup_calc_subtree_ss_mask(). | |
0f060deb | 1320 | */ |
8699b776 | 1321 | static void cgroup_refresh_subtree_ss_mask(struct cgroup *cgrp) |
0f060deb | 1322 | { |
8699b776 TH |
1323 | cgrp->subtree_ss_mask = |
1324 | cgroup_calc_subtree_ss_mask(cgrp, cgrp->subtree_control); | |
667c2491 TH |
1325 | } |
1326 | ||
a9746d8d TH |
1327 | /** |
1328 | * cgroup_kn_unlock - unlocking helper for cgroup kernfs methods | |
1329 | * @kn: the kernfs_node being serviced | |
1330 | * | |
1331 | * This helper undoes cgroup_kn_lock_live() and should be invoked before | |
1332 | * the method finishes if locking succeeded. Note that once this function | |
1333 | * returns the cgroup returned by cgroup_kn_lock_live() may become | |
1334 | * inaccessible any time. If the caller intends to continue to access the | |
1335 | * cgroup, it should pin it before invoking this function. | |
1336 | */ | |
1337 | static void cgroup_kn_unlock(struct kernfs_node *kn) | |
ddbcc7e8 | 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; | |
1345 | ||
1346 | mutex_unlock(&cgroup_mutex); | |
a9746d8d TH |
1347 | |
1348 | kernfs_unbreak_active_protection(kn); | |
1349 | cgroup_put(cgrp); | |
ddbcc7e8 PM |
1350 | } |
1351 | ||
a9746d8d TH |
1352 | /** |
1353 | * cgroup_kn_lock_live - locking helper for cgroup kernfs methods | |
1354 | * @kn: the kernfs_node being serviced | |
1355 | * | |
1356 | * This helper is to be used by a cgroup kernfs method currently servicing | |
1357 | * @kn. It breaks the active protection, performs cgroup locking and | |
1358 | * verifies that the associated cgroup is alive. Returns the cgroup if | |
1359 | * alive; otherwise, %NULL. A successful return should be undone by a | |
1360 | * matching cgroup_kn_unlock() invocation. | |
1361 | * | |
1362 | * Any cgroup kernfs method implementation which requires locking the | |
1363 | * associated cgroup should use this helper. It avoids nesting cgroup | |
1364 | * locking under kernfs active protection and allows all kernfs operations | |
1365 | * including self-removal. | |
1366 | */ | |
1367 | static struct cgroup *cgroup_kn_lock_live(struct kernfs_node *kn) | |
05ef1d7c | 1368 | { |
a9746d8d TH |
1369 | struct cgroup *cgrp; |
1370 | ||
1371 | if (kernfs_type(kn) == KERNFS_DIR) | |
1372 | cgrp = kn->priv; | |
1373 | else | |
1374 | cgrp = kn->parent->priv; | |
05ef1d7c | 1375 | |
2739d3cc | 1376 | /* |
01f6474c | 1377 | * We're gonna grab cgroup_mutex which nests outside kernfs |
a9746d8d TH |
1378 | * active_ref. cgroup liveliness check alone provides enough |
1379 | * protection against removal. Ensure @cgrp stays accessible and | |
1380 | * break the active_ref protection. | |
2739d3cc | 1381 | */ |
aa32362f LZ |
1382 | if (!cgroup_tryget(cgrp)) |
1383 | return NULL; | |
a9746d8d TH |
1384 | kernfs_break_active_protection(kn); |
1385 | ||
2bd59d48 | 1386 | mutex_lock(&cgroup_mutex); |
05ef1d7c | 1387 | |
a9746d8d TH |
1388 | if (!cgroup_is_dead(cgrp)) |
1389 | return cgrp; | |
1390 | ||
1391 | cgroup_kn_unlock(kn); | |
1392 | return NULL; | |
ddbcc7e8 | 1393 | } |
05ef1d7c | 1394 | |
2739d3cc | 1395 | static void cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft) |
05ef1d7c | 1396 | { |
2bd59d48 | 1397 | char name[CGROUP_FILE_NAME_MAX]; |
05ef1d7c | 1398 | |
01f6474c | 1399 | lockdep_assert_held(&cgroup_mutex); |
34c06254 TH |
1400 | |
1401 | if (cft->file_offset) { | |
1402 | struct cgroup_subsys_state *css = cgroup_css(cgrp, cft->ss); | |
1403 | struct cgroup_file *cfile = (void *)css + cft->file_offset; | |
1404 | ||
1405 | spin_lock_irq(&cgroup_file_kn_lock); | |
1406 | cfile->kn = NULL; | |
1407 | spin_unlock_irq(&cgroup_file_kn_lock); | |
1408 | } | |
1409 | ||
2bd59d48 | 1410 | kernfs_remove_by_name(cgrp->kn, cgroup_file_name(cgrp, cft, name)); |
05ef1d7c TH |
1411 | } |
1412 | ||
13af07df | 1413 | /** |
4df8dc90 TH |
1414 | * css_clear_dir - remove subsys files in a cgroup directory |
1415 | * @css: taget css | |
1416 | * @cgrp_override: specify if target cgroup is different from css->cgroup | |
13af07df | 1417 | */ |
4df8dc90 TH |
1418 | static void css_clear_dir(struct cgroup_subsys_state *css, |
1419 | struct cgroup *cgrp_override) | |
05ef1d7c | 1420 | { |
4df8dc90 TH |
1421 | struct cgroup *cgrp = cgrp_override ?: css->cgroup; |
1422 | struct cftype *cfts; | |
05ef1d7c | 1423 | |
4df8dc90 TH |
1424 | list_for_each_entry(cfts, &css->ss->cfts, node) |
1425 | cgroup_addrm_files(css, cgrp, cfts, false); | |
ddbcc7e8 PM |
1426 | } |
1427 | ||
ccdca218 | 1428 | /** |
4df8dc90 TH |
1429 | * css_populate_dir - create subsys files in a cgroup directory |
1430 | * @css: target css | |
1431 | * @cgrp_overried: specify if target cgroup is different from css->cgroup | |
ccdca218 TH |
1432 | * |
1433 | * On failure, no file is added. | |
1434 | */ | |
4df8dc90 TH |
1435 | static int css_populate_dir(struct cgroup_subsys_state *css, |
1436 | struct cgroup *cgrp_override) | |
ccdca218 | 1437 | { |
4df8dc90 TH |
1438 | struct cgroup *cgrp = cgrp_override ?: css->cgroup; |
1439 | struct cftype *cfts, *failed_cfts; | |
1440 | int ret; | |
ccdca218 | 1441 | |
4df8dc90 TH |
1442 | if (!css->ss) { |
1443 | if (cgroup_on_dfl(cgrp)) | |
1444 | cfts = cgroup_dfl_base_files; | |
1445 | else | |
1446 | cfts = cgroup_legacy_base_files; | |
ccdca218 | 1447 | |
4df8dc90 TH |
1448 | return cgroup_addrm_files(&cgrp->self, cgrp, cfts, true); |
1449 | } | |
ccdca218 | 1450 | |
4df8dc90 TH |
1451 | list_for_each_entry(cfts, &css->ss->cfts, node) { |
1452 | ret = cgroup_addrm_files(css, cgrp, cfts, true); | |
1453 | if (ret < 0) { | |
1454 | failed_cfts = cfts; | |
1455 | goto err; | |
ccdca218 TH |
1456 | } |
1457 | } | |
1458 | return 0; | |
1459 | err: | |
4df8dc90 TH |
1460 | list_for_each_entry(cfts, &css->ss->cfts, node) { |
1461 | if (cfts == failed_cfts) | |
1462 | break; | |
1463 | cgroup_addrm_files(css, cgrp, cfts, false); | |
1464 | } | |
ccdca218 TH |
1465 | return ret; |
1466 | } | |
1467 | ||
6e5c8307 | 1468 | static int rebind_subsystems(struct cgroup_root *dst_root, u16 ss_mask) |
ddbcc7e8 | 1469 | { |
1ada4838 | 1470 | struct cgroup *dcgrp = &dst_root->cgrp; |
30159ec7 | 1471 | struct cgroup_subsys *ss; |
6e5c8307 | 1472 | u16 tmp_ss_mask; |
2d8f243a | 1473 | int ssid, i, ret; |
ddbcc7e8 | 1474 | |
ace2bee8 | 1475 | lockdep_assert_held(&cgroup_mutex); |
ddbcc7e8 | 1476 | |
b4e0eeaf | 1477 | do_each_subsys_mask(ss, ssid, ss_mask) { |
7fd8c565 TH |
1478 | /* if @ss has non-root csses attached to it, can't move */ |
1479 | if (css_next_child(NULL, cgroup_css(&ss->root->cgrp, ss))) | |
3ed80a62 | 1480 | return -EBUSY; |
1d5be6b2 | 1481 | |
5df36032 | 1482 | /* can't move between two non-dummy roots either */ |
7fd8c565 | 1483 | if (ss->root != &cgrp_dfl_root && dst_root != &cgrp_dfl_root) |
5df36032 | 1484 | return -EBUSY; |
b4e0eeaf | 1485 | } while_each_subsys_mask(); |
ddbcc7e8 | 1486 | |
5533e011 TH |
1487 | /* skip creating root files on dfl_root for inhibited subsystems */ |
1488 | tmp_ss_mask = ss_mask; | |
1489 | if (dst_root == &cgrp_dfl_root) | |
a7165264 | 1490 | tmp_ss_mask &= ~cgrp_dfl_inhibit_ss_mask; |
5533e011 | 1491 | |
b4e0eeaf | 1492 | do_each_subsys_mask(ss, ssid, tmp_ss_mask) { |
4df8dc90 TH |
1493 | struct cgroup *scgrp = &ss->root->cgrp; |
1494 | int tssid; | |
1495 | ||
1496 | ret = css_populate_dir(cgroup_css(scgrp, ss), dcgrp); | |
1497 | if (!ret) | |
1498 | continue; | |
ddbcc7e8 | 1499 | |
a2dd4247 TH |
1500 | /* |
1501 | * Rebinding back to the default root is not allowed to | |
1502 | * fail. Using both default and non-default roots should | |
1503 | * be rare. Moving subsystems back and forth even more so. | |
1504 | * Just warn about it and continue. | |
1505 | */ | |
4df8dc90 | 1506 | if (dst_root == &cgrp_dfl_root) { |
a7165264 | 1507 | if (cgrp_dfl_visible) { |
6e5c8307 | 1508 | pr_warn("failed to create files (%d) while rebinding 0x%x to default root\n", |
4df8dc90 TH |
1509 | ret, ss_mask); |
1510 | pr_warn("you may retry by moving them to a different hierarchy and unbinding\n"); | |
1511 | } | |
1512 | continue; | |
a2dd4247 | 1513 | } |
4df8dc90 | 1514 | |
b4e0eeaf | 1515 | do_each_subsys_mask(ss, tssid, tmp_ss_mask) { |
4df8dc90 TH |
1516 | if (tssid == ssid) |
1517 | break; | |
1518 | css_clear_dir(cgroup_css(scgrp, ss), dcgrp); | |
b4e0eeaf | 1519 | } while_each_subsys_mask(); |
4df8dc90 | 1520 | return ret; |
b4e0eeaf | 1521 | } while_each_subsys_mask(); |
3126121f TH |
1522 | |
1523 | /* | |
1524 | * Nothing can fail from this point on. Remove files for the | |
1525 | * removed subsystems and rebind each subsystem. | |
1526 | */ | |
b4e0eeaf | 1527 | do_each_subsys_mask(ss, ssid, ss_mask) { |
1ada4838 TH |
1528 | struct cgroup_root *src_root = ss->root; |
1529 | struct cgroup *scgrp = &src_root->cgrp; | |
1530 | struct cgroup_subsys_state *css = cgroup_css(scgrp, ss); | |
2d8f243a | 1531 | struct css_set *cset; |
a8a648c4 | 1532 | |
1ada4838 | 1533 | WARN_ON(!css || cgroup_css(dcgrp, ss)); |
a8a648c4 | 1534 | |
4df8dc90 TH |
1535 | css_clear_dir(css, NULL); |
1536 | ||
1ada4838 TH |
1537 | RCU_INIT_POINTER(scgrp->subsys[ssid], NULL); |
1538 | rcu_assign_pointer(dcgrp->subsys[ssid], css); | |
5df36032 | 1539 | ss->root = dst_root; |
1ada4838 | 1540 | css->cgroup = dcgrp; |
73e80ed8 | 1541 | |
f0d9a5f1 | 1542 | spin_lock_bh(&css_set_lock); |
2d8f243a TH |
1543 | hash_for_each(css_set_table, i, cset, hlist) |
1544 | list_move_tail(&cset->e_cset_node[ss->id], | |
1ada4838 | 1545 | &dcgrp->e_csets[ss->id]); |
f0d9a5f1 | 1546 | spin_unlock_bh(&css_set_lock); |
2d8f243a | 1547 | |
f392e51c | 1548 | src_root->subsys_mask &= ~(1 << ssid); |
1ada4838 | 1549 | scgrp->subtree_control &= ~(1 << ssid); |
8699b776 | 1550 | cgroup_refresh_subtree_ss_mask(scgrp); |
f392e51c | 1551 | |
bd53d617 | 1552 | /* default hierarchy doesn't enable controllers by default */ |
f392e51c | 1553 | dst_root->subsys_mask |= 1 << ssid; |
49d1dc4b TH |
1554 | if (dst_root == &cgrp_dfl_root) { |
1555 | static_branch_enable(cgroup_subsys_on_dfl_key[ssid]); | |
1556 | } else { | |
1ada4838 | 1557 | dcgrp->subtree_control |= 1 << ssid; |
8699b776 | 1558 | cgroup_refresh_subtree_ss_mask(dcgrp); |
49d1dc4b | 1559 | static_branch_disable(cgroup_subsys_on_dfl_key[ssid]); |
667c2491 | 1560 | } |
a8a648c4 | 1561 | |
5df36032 TH |
1562 | if (ss->bind) |
1563 | ss->bind(css); | |
b4e0eeaf | 1564 | } while_each_subsys_mask(); |
ddbcc7e8 | 1565 | |
1ada4838 | 1566 | kernfs_activate(dcgrp->kn); |
ddbcc7e8 PM |
1567 | return 0; |
1568 | } | |
1569 | ||
2bd59d48 TH |
1570 | static int cgroup_show_options(struct seq_file *seq, |
1571 | struct kernfs_root *kf_root) | |
ddbcc7e8 | 1572 | { |
3dd06ffa | 1573 | struct cgroup_root *root = cgroup_root_from_kf(kf_root); |
ddbcc7e8 | 1574 | struct cgroup_subsys *ss; |
b85d2040 | 1575 | int ssid; |
ddbcc7e8 | 1576 | |
d98817d4 TH |
1577 | if (root != &cgrp_dfl_root) |
1578 | for_each_subsys(ss, ssid) | |
1579 | if (root->subsys_mask & (1 << ssid)) | |
61e57c0c | 1580 | seq_show_option(seq, ss->legacy_name, NULL); |
93438629 | 1581 | if (root->flags & CGRP_ROOT_NOPREFIX) |
ddbcc7e8 | 1582 | seq_puts(seq, ",noprefix"); |
93438629 | 1583 | if (root->flags & CGRP_ROOT_XATTR) |
03b1cde6 | 1584 | seq_puts(seq, ",xattr"); |
69e943b7 TH |
1585 | |
1586 | spin_lock(&release_agent_path_lock); | |
81a6a5cd | 1587 | if (strlen(root->release_agent_path)) |
a068acf2 KC |
1588 | seq_show_option(seq, "release_agent", |
1589 | root->release_agent_path); | |
69e943b7 TH |
1590 | spin_unlock(&release_agent_path_lock); |
1591 | ||
3dd06ffa | 1592 | if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags)) |
97978e6d | 1593 | seq_puts(seq, ",clone_children"); |
c6d57f33 | 1594 | if (strlen(root->name)) |
a068acf2 | 1595 | seq_show_option(seq, "name", root->name); |
ddbcc7e8 PM |
1596 | return 0; |
1597 | } | |
1598 | ||
1599 | struct cgroup_sb_opts { | |
6e5c8307 | 1600 | u16 subsys_mask; |
69dfa00c | 1601 | unsigned int flags; |
81a6a5cd | 1602 | char *release_agent; |
2260e7fc | 1603 | bool cpuset_clone_children; |
c6d57f33 | 1604 | char *name; |
2c6ab6d2 PM |
1605 | /* User explicitly requested empty subsystem */ |
1606 | bool none; | |
ddbcc7e8 PM |
1607 | }; |
1608 | ||
cf5d5941 | 1609 | static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts) |
ddbcc7e8 | 1610 | { |
32a8cf23 DL |
1611 | char *token, *o = data; |
1612 | bool all_ss = false, one_ss = false; | |
6e5c8307 | 1613 | u16 mask = U16_MAX; |
30159ec7 | 1614 | struct cgroup_subsys *ss; |
7b9a6ba5 | 1615 | int nr_opts = 0; |
30159ec7 | 1616 | int i; |
f9ab5b5b LZ |
1617 | |
1618 | #ifdef CONFIG_CPUSETS | |
6e5c8307 | 1619 | mask = ~((u16)1 << cpuset_cgrp_id); |
f9ab5b5b | 1620 | #endif |
ddbcc7e8 | 1621 | |
c6d57f33 | 1622 | memset(opts, 0, sizeof(*opts)); |
ddbcc7e8 PM |
1623 | |
1624 | while ((token = strsep(&o, ",")) != NULL) { | |
7b9a6ba5 TH |
1625 | nr_opts++; |
1626 | ||
ddbcc7e8 PM |
1627 | if (!*token) |
1628 | return -EINVAL; | |
32a8cf23 | 1629 | if (!strcmp(token, "none")) { |
2c6ab6d2 PM |
1630 | /* Explicitly have no subsystems */ |
1631 | opts->none = true; | |
32a8cf23 DL |
1632 | continue; |
1633 | } | |
1634 | if (!strcmp(token, "all")) { | |
1635 | /* Mutually exclusive option 'all' + subsystem name */ | |
1636 | if (one_ss) | |
1637 | return -EINVAL; | |
1638 | all_ss = true; | |
1639 | continue; | |
1640 | } | |
1641 | if (!strcmp(token, "noprefix")) { | |
93438629 | 1642 | opts->flags |= CGRP_ROOT_NOPREFIX; |
32a8cf23 DL |
1643 | continue; |
1644 | } | |
1645 | if (!strcmp(token, "clone_children")) { | |
2260e7fc | 1646 | opts->cpuset_clone_children = true; |
32a8cf23 DL |
1647 | continue; |
1648 | } | |
03b1cde6 | 1649 | if (!strcmp(token, "xattr")) { |
93438629 | 1650 | opts->flags |= CGRP_ROOT_XATTR; |
03b1cde6 AR |
1651 | continue; |
1652 | } | |
32a8cf23 | 1653 | if (!strncmp(token, "release_agent=", 14)) { |
81a6a5cd PM |
1654 | /* Specifying two release agents is forbidden */ |
1655 | if (opts->release_agent) | |
1656 | return -EINVAL; | |
c6d57f33 | 1657 | opts->release_agent = |
e400c285 | 1658 | kstrndup(token + 14, PATH_MAX - 1, GFP_KERNEL); |
81a6a5cd PM |
1659 | if (!opts->release_agent) |
1660 | return -ENOMEM; | |
32a8cf23 DL |
1661 | continue; |
1662 | } | |
1663 | if (!strncmp(token, "name=", 5)) { | |
c6d57f33 PM |
1664 | const char *name = token + 5; |
1665 | /* Can't specify an empty name */ | |
1666 | if (!strlen(name)) | |
1667 | return -EINVAL; | |
1668 | /* Must match [\w.-]+ */ | |
1669 | for (i = 0; i < strlen(name); i++) { | |
1670 | char c = name[i]; | |
1671 | if (isalnum(c)) | |
1672 | continue; | |
1673 | if ((c == '.') || (c == '-') || (c == '_')) | |
1674 | continue; | |
1675 | return -EINVAL; | |
1676 | } | |
1677 | /* Specifying two names is forbidden */ | |
1678 | if (opts->name) | |
1679 | return -EINVAL; | |
1680 | opts->name = kstrndup(name, | |
e400c285 | 1681 | MAX_CGROUP_ROOT_NAMELEN - 1, |
c6d57f33 PM |
1682 | GFP_KERNEL); |
1683 | if (!opts->name) | |
1684 | return -ENOMEM; | |
32a8cf23 DL |
1685 | |
1686 | continue; | |
1687 | } | |
1688 | ||
30159ec7 | 1689 | for_each_subsys(ss, i) { |
3e1d2eed | 1690 | if (strcmp(token, ss->legacy_name)) |
32a8cf23 | 1691 | continue; |
fc5ed1e9 | 1692 | if (!cgroup_ssid_enabled(i)) |
32a8cf23 | 1693 | continue; |
223ffb29 JW |
1694 | if (cgroup_ssid_no_v1(i)) |
1695 | continue; | |
32a8cf23 DL |
1696 | |
1697 | /* Mutually exclusive option 'all' + subsystem name */ | |
1698 | if (all_ss) | |
1699 | return -EINVAL; | |
69dfa00c | 1700 | opts->subsys_mask |= (1 << i); |
32a8cf23 DL |
1701 | one_ss = true; |
1702 | ||
1703 | break; | |
1704 | } | |
1705 | if (i == CGROUP_SUBSYS_COUNT) | |
1706 | return -ENOENT; | |
1707 | } | |
1708 | ||
7b9a6ba5 TH |
1709 | /* |
1710 | * If the 'all' option was specified select all the subsystems, | |
1711 | * otherwise if 'none', 'name=' and a subsystem name options were | |
1712 | * not specified, let's default to 'all' | |
1713 | */ | |
1714 | if (all_ss || (!one_ss && !opts->none && !opts->name)) | |
1715 | for_each_subsys(ss, i) | |
223ffb29 | 1716 | if (cgroup_ssid_enabled(i) && !cgroup_ssid_no_v1(i)) |
7b9a6ba5 TH |
1717 | opts->subsys_mask |= (1 << i); |
1718 | ||
1719 | /* | |
1720 | * We either have to specify by name or by subsystems. (So all | |
1721 | * empty hierarchies must have a name). | |
1722 | */ | |
1723 | if (!opts->subsys_mask && !opts->name) | |
1724 | return -EINVAL; | |
1725 | ||
f9ab5b5b LZ |
1726 | /* |
1727 | * Option noprefix was introduced just for backward compatibility | |
1728 | * with the old cpuset, so we allow noprefix only if mounting just | |
1729 | * the cpuset subsystem. | |
1730 | */ | |
93438629 | 1731 | if ((opts->flags & CGRP_ROOT_NOPREFIX) && (opts->subsys_mask & mask)) |
f9ab5b5b LZ |
1732 | return -EINVAL; |
1733 | ||
2c6ab6d2 | 1734 | /* Can't specify "none" and some subsystems */ |
a1a71b45 | 1735 | if (opts->subsys_mask && opts->none) |
2c6ab6d2 PM |
1736 | return -EINVAL; |
1737 | ||
ddbcc7e8 PM |
1738 | return 0; |
1739 | } | |
1740 | ||
2bd59d48 | 1741 | static int cgroup_remount(struct kernfs_root *kf_root, int *flags, char *data) |
ddbcc7e8 PM |
1742 | { |
1743 | int ret = 0; | |
3dd06ffa | 1744 | struct cgroup_root *root = cgroup_root_from_kf(kf_root); |
ddbcc7e8 | 1745 | struct cgroup_sb_opts opts; |
6e5c8307 | 1746 | u16 added_mask, removed_mask; |
ddbcc7e8 | 1747 | |
aa6ec29b TH |
1748 | if (root == &cgrp_dfl_root) { |
1749 | pr_err("remount is not allowed\n"); | |
873fe09e TH |
1750 | return -EINVAL; |
1751 | } | |
1752 | ||
ddbcc7e8 PM |
1753 | mutex_lock(&cgroup_mutex); |
1754 | ||
1755 | /* See what subsystems are wanted */ | |
1756 | ret = parse_cgroupfs_options(data, &opts); | |
1757 | if (ret) | |
1758 | goto out_unlock; | |
1759 | ||
f392e51c | 1760 | if (opts.subsys_mask != root->subsys_mask || opts.release_agent) |
ed3d261b | 1761 | pr_warn("option changes via remount are deprecated (pid=%d comm=%s)\n", |
a2a1f9ea | 1762 | task_tgid_nr(current), current->comm); |
8b5a5a9d | 1763 | |
f392e51c TH |
1764 | added_mask = opts.subsys_mask & ~root->subsys_mask; |
1765 | removed_mask = root->subsys_mask & ~opts.subsys_mask; | |
13af07df | 1766 | |
cf5d5941 | 1767 | /* Don't allow flags or name to change at remount */ |
7450e90b | 1768 | if ((opts.flags ^ root->flags) || |
cf5d5941 | 1769 | (opts.name && strcmp(opts.name, root->name))) { |
69dfa00c | 1770 | pr_err("option or name mismatch, new: 0x%x \"%s\", old: 0x%x \"%s\"\n", |
7450e90b | 1771 | opts.flags, opts.name ?: "", root->flags, root->name); |
c6d57f33 PM |
1772 | ret = -EINVAL; |
1773 | goto out_unlock; | |
1774 | } | |
1775 | ||
f172e67c | 1776 | /* remounting is not allowed for populated hierarchies */ |
d5c419b6 | 1777 | if (!list_empty(&root->cgrp.self.children)) { |
f172e67c | 1778 | ret = -EBUSY; |
0670e08b | 1779 | goto out_unlock; |
cf5d5941 | 1780 | } |
ddbcc7e8 | 1781 | |
5df36032 | 1782 | ret = rebind_subsystems(root, added_mask); |
3126121f | 1783 | if (ret) |
0670e08b | 1784 | goto out_unlock; |
ddbcc7e8 | 1785 | |
3dd06ffa | 1786 | rebind_subsystems(&cgrp_dfl_root, removed_mask); |
5df36032 | 1787 | |
69e943b7 TH |
1788 | if (opts.release_agent) { |
1789 | spin_lock(&release_agent_path_lock); | |
81a6a5cd | 1790 | strcpy(root->release_agent_path, opts.release_agent); |
69e943b7 TH |
1791 | spin_unlock(&release_agent_path_lock); |
1792 | } | |
ddbcc7e8 | 1793 | out_unlock: |
66bdc9cf | 1794 | kfree(opts.release_agent); |
c6d57f33 | 1795 | kfree(opts.name); |
ddbcc7e8 | 1796 | mutex_unlock(&cgroup_mutex); |
ddbcc7e8 PM |
1797 | return ret; |
1798 | } | |
1799 | ||
afeb0f9f TH |
1800 | /* |
1801 | * To reduce the fork() overhead for systems that are not actually using | |
1802 | * their cgroups capability, we don't maintain the lists running through | |
1803 | * each css_set to its tasks until we see the list actually used - in other | |
1804 | * words after the first mount. | |
1805 | */ | |
1806 | static bool use_task_css_set_links __read_mostly; | |
1807 | ||
1808 | static void cgroup_enable_task_cg_lists(void) | |
1809 | { | |
1810 | struct task_struct *p, *g; | |
1811 | ||
f0d9a5f1 | 1812 | spin_lock_bh(&css_set_lock); |
afeb0f9f TH |
1813 | |
1814 | if (use_task_css_set_links) | |
1815 | goto out_unlock; | |
1816 | ||
1817 | use_task_css_set_links = true; | |
1818 | ||
1819 | /* | |
1820 | * We need tasklist_lock because RCU is not safe against | |
1821 | * while_each_thread(). Besides, a forking task that has passed | |
1822 | * cgroup_post_fork() without seeing use_task_css_set_links = 1 | |
1823 | * is not guaranteed to have its child immediately visible in the | |
1824 | * tasklist if we walk through it with RCU. | |
1825 | */ | |
1826 | read_lock(&tasklist_lock); | |
1827 | do_each_thread(g, p) { | |
afeb0f9f TH |
1828 | WARN_ON_ONCE(!list_empty(&p->cg_list) || |
1829 | task_css_set(p) != &init_css_set); | |
1830 | ||
1831 | /* | |
1832 | * We should check if the process is exiting, otherwise | |
1833 | * it will race with cgroup_exit() in that the list | |
1834 | * entry won't be deleted though the process has exited. | |
f153ad11 TH |
1835 | * Do it while holding siglock so that we don't end up |
1836 | * racing against cgroup_exit(). | |
afeb0f9f | 1837 | */ |
f153ad11 | 1838 | spin_lock_irq(&p->sighand->siglock); |
eaf797ab TH |
1839 | if (!(p->flags & PF_EXITING)) { |
1840 | struct css_set *cset = task_css_set(p); | |
1841 | ||
0de0942d TH |
1842 | if (!css_set_populated(cset)) |
1843 | css_set_update_populated(cset, true); | |
389b9c1b | 1844 | list_add_tail(&p->cg_list, &cset->tasks); |
eaf797ab TH |
1845 | get_css_set(cset); |
1846 | } | |
f153ad11 | 1847 | spin_unlock_irq(&p->sighand->siglock); |
afeb0f9f TH |
1848 | } while_each_thread(g, p); |
1849 | read_unlock(&tasklist_lock); | |
1850 | out_unlock: | |
f0d9a5f1 | 1851 | spin_unlock_bh(&css_set_lock); |
afeb0f9f | 1852 | } |
ddbcc7e8 | 1853 | |
cc31edce PM |
1854 | static void init_cgroup_housekeeping(struct cgroup *cgrp) |
1855 | { | |
2d8f243a TH |
1856 | struct cgroup_subsys *ss; |
1857 | int ssid; | |
1858 | ||
d5c419b6 TH |
1859 | INIT_LIST_HEAD(&cgrp->self.sibling); |
1860 | INIT_LIST_HEAD(&cgrp->self.children); | |
69d0206c | 1861 | INIT_LIST_HEAD(&cgrp->cset_links); |
72a8cb30 BB |
1862 | INIT_LIST_HEAD(&cgrp->pidlists); |
1863 | mutex_init(&cgrp->pidlist_mutex); | |
9d800df1 | 1864 | cgrp->self.cgroup = cgrp; |
184faf32 | 1865 | cgrp->self.flags |= CSS_ONLINE; |
2d8f243a TH |
1866 | |
1867 | for_each_subsys(ss, ssid) | |
1868 | INIT_LIST_HEAD(&cgrp->e_csets[ssid]); | |
f8f22e53 TH |
1869 | |
1870 | init_waitqueue_head(&cgrp->offline_waitq); | |
971ff493 | 1871 | INIT_WORK(&cgrp->release_agent_work, cgroup_release_agent); |
cc31edce | 1872 | } |
c6d57f33 | 1873 | |
3dd06ffa | 1874 | static void init_cgroup_root(struct cgroup_root *root, |
172a2c06 | 1875 | struct cgroup_sb_opts *opts) |
ddbcc7e8 | 1876 | { |
3dd06ffa | 1877 | struct cgroup *cgrp = &root->cgrp; |
b0ca5a84 | 1878 | |
ddbcc7e8 | 1879 | INIT_LIST_HEAD(&root->root_list); |
3c9c825b | 1880 | atomic_set(&root->nr_cgrps, 1); |
bd89aabc | 1881 | cgrp->root = root; |
cc31edce | 1882 | init_cgroup_housekeeping(cgrp); |
4e96ee8e | 1883 | idr_init(&root->cgroup_idr); |
c6d57f33 | 1884 | |
c6d57f33 PM |
1885 | root->flags = opts->flags; |
1886 | if (opts->release_agent) | |
1887 | strcpy(root->release_agent_path, opts->release_agent); | |
1888 | if (opts->name) | |
1889 | strcpy(root->name, opts->name); | |
2260e7fc | 1890 | if (opts->cpuset_clone_children) |
3dd06ffa | 1891 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags); |
c6d57f33 PM |
1892 | } |
1893 | ||
6e5c8307 | 1894 | static int cgroup_setup_root(struct cgroup_root *root, u16 ss_mask) |
2c6ab6d2 | 1895 | { |
d427dfeb | 1896 | LIST_HEAD(tmp_links); |
3dd06ffa | 1897 | struct cgroup *root_cgrp = &root->cgrp; |
d427dfeb | 1898 | struct css_set *cset; |
d427dfeb | 1899 | int i, ret; |
2c6ab6d2 | 1900 | |
d427dfeb | 1901 | lockdep_assert_held(&cgroup_mutex); |
c6d57f33 | 1902 | |
cf780b7d | 1903 | ret = cgroup_idr_alloc(&root->cgroup_idr, root_cgrp, 1, 2, GFP_KERNEL); |
d427dfeb | 1904 | if (ret < 0) |
2bd59d48 | 1905 | goto out; |
d427dfeb | 1906 | root_cgrp->id = ret; |
b11cfb58 | 1907 | root_cgrp->ancestor_ids[0] = ret; |
c6d57f33 | 1908 | |
2aad2a86 TH |
1909 | ret = percpu_ref_init(&root_cgrp->self.refcnt, css_release, 0, |
1910 | GFP_KERNEL); | |
9d755d33 TH |
1911 | if (ret) |
1912 | goto out; | |
1913 | ||
d427dfeb | 1914 | /* |
f0d9a5f1 | 1915 | * We're accessing css_set_count without locking css_set_lock here, |
d427dfeb TH |
1916 | * but that's OK - it can only be increased by someone holding |
1917 | * cgroup_lock, and that's us. The worst that can happen is that we | |
1918 | * have some link structures left over | |
1919 | */ | |
1920 | ret = allocate_cgrp_cset_links(css_set_count, &tmp_links); | |
1921 | if (ret) | |
9d755d33 | 1922 | goto cancel_ref; |
ddbcc7e8 | 1923 | |
985ed670 | 1924 | ret = cgroup_init_root_id(root); |
ddbcc7e8 | 1925 | if (ret) |
9d755d33 | 1926 | goto cancel_ref; |
ddbcc7e8 | 1927 | |
2bd59d48 TH |
1928 | root->kf_root = kernfs_create_root(&cgroup_kf_syscall_ops, |
1929 | KERNFS_ROOT_CREATE_DEACTIVATED, | |
1930 | root_cgrp); | |
1931 | if (IS_ERR(root->kf_root)) { | |
1932 | ret = PTR_ERR(root->kf_root); | |
1933 | goto exit_root_id; | |
1934 | } | |
1935 | root_cgrp->kn = root->kf_root->kn; | |
ddbcc7e8 | 1936 | |
4df8dc90 | 1937 | ret = css_populate_dir(&root_cgrp->self, NULL); |
d427dfeb | 1938 | if (ret) |
2bd59d48 | 1939 | goto destroy_root; |
ddbcc7e8 | 1940 | |
5df36032 | 1941 | ret = rebind_subsystems(root, ss_mask); |
d427dfeb | 1942 | if (ret) |
2bd59d48 | 1943 | goto destroy_root; |
ddbcc7e8 | 1944 | |
d427dfeb TH |
1945 | /* |
1946 | * There must be no failure case after here, since rebinding takes | |
1947 | * care of subsystems' refcounts, which are explicitly dropped in | |
1948 | * the failure exit path. | |
1949 | */ | |
1950 | list_add(&root->root_list, &cgroup_roots); | |
1951 | cgroup_root_count++; | |
0df6a63f | 1952 | |
d427dfeb | 1953 | /* |
3dd06ffa | 1954 | * Link the root cgroup in this hierarchy into all the css_set |
d427dfeb TH |
1955 | * objects. |
1956 | */ | |
f0d9a5f1 | 1957 | spin_lock_bh(&css_set_lock); |
0de0942d | 1958 | hash_for_each(css_set_table, i, cset, hlist) { |
d427dfeb | 1959 | link_css_set(&tmp_links, cset, root_cgrp); |
0de0942d TH |
1960 | if (css_set_populated(cset)) |
1961 | cgroup_update_populated(root_cgrp, true); | |
1962 | } | |
f0d9a5f1 | 1963 | spin_unlock_bh(&css_set_lock); |
ddbcc7e8 | 1964 | |
d5c419b6 | 1965 | BUG_ON(!list_empty(&root_cgrp->self.children)); |
3c9c825b | 1966 | BUG_ON(atomic_read(&root->nr_cgrps) != 1); |
ddbcc7e8 | 1967 | |
2bd59d48 | 1968 | kernfs_activate(root_cgrp->kn); |
d427dfeb | 1969 | ret = 0; |
2bd59d48 | 1970 | goto out; |
d427dfeb | 1971 | |
2bd59d48 TH |
1972 | destroy_root: |
1973 | kernfs_destroy_root(root->kf_root); | |
1974 | root->kf_root = NULL; | |
1975 | exit_root_id: | |
d427dfeb | 1976 | cgroup_exit_root_id(root); |
9d755d33 | 1977 | cancel_ref: |
9a1049da | 1978 | percpu_ref_exit(&root_cgrp->self.refcnt); |
2bd59d48 | 1979 | out: |
d427dfeb TH |
1980 | free_cgrp_cset_links(&tmp_links); |
1981 | return ret; | |
ddbcc7e8 PM |
1982 | } |
1983 | ||
f7e83571 | 1984 | static struct dentry *cgroup_mount(struct file_system_type *fs_type, |
ddbcc7e8 | 1985 | int flags, const char *unused_dev_name, |
f7e83571 | 1986 | void *data) |
ddbcc7e8 | 1987 | { |
67e9c74b | 1988 | bool is_v2 = fs_type == &cgroup2_fs_type; |
3a32bd72 | 1989 | struct super_block *pinned_sb = NULL; |
970317aa | 1990 | struct cgroup_subsys *ss; |
3dd06ffa | 1991 | struct cgroup_root *root; |
ddbcc7e8 | 1992 | struct cgroup_sb_opts opts; |
2bd59d48 | 1993 | struct dentry *dentry; |
8e30e2b8 | 1994 | int ret; |
970317aa | 1995 | int i; |
c6b3d5bc | 1996 | bool new_sb; |
ddbcc7e8 | 1997 | |
56fde9e0 TH |
1998 | /* |
1999 | * The first time anyone tries to mount a cgroup, enable the list | |
2000 | * linking each css_set to its tasks and fix up all existing tasks. | |
2001 | */ | |
2002 | if (!use_task_css_set_links) | |
2003 | cgroup_enable_task_cg_lists(); | |
e37a06f1 | 2004 | |
67e9c74b TH |
2005 | if (is_v2) { |
2006 | if (data) { | |
2007 | pr_err("cgroup2: unknown option \"%s\"\n", (char *)data); | |
2008 | return ERR_PTR(-EINVAL); | |
2009 | } | |
a7165264 | 2010 | cgrp_dfl_visible = true; |
67e9c74b TH |
2011 | root = &cgrp_dfl_root; |
2012 | cgroup_get(&root->cgrp); | |
2013 | goto out_mount; | |
2014 | } | |
2015 | ||
aae8aab4 | 2016 | mutex_lock(&cgroup_mutex); |
8e30e2b8 TH |
2017 | |
2018 | /* First find the desired set of subsystems */ | |
ddbcc7e8 | 2019 | ret = parse_cgroupfs_options(data, &opts); |
c6d57f33 | 2020 | if (ret) |
8e30e2b8 | 2021 | goto out_unlock; |
a015edd2 | 2022 | |
970317aa LZ |
2023 | /* |
2024 | * Destruction of cgroup root is asynchronous, so subsystems may | |
2025 | * still be dying after the previous unmount. Let's drain the | |
2026 | * dying subsystems. We just need to ensure that the ones | |
2027 | * unmounted previously finish dying and don't care about new ones | |
2028 | * starting. Testing ref liveliness is good enough. | |
2029 | */ | |
2030 | for_each_subsys(ss, i) { | |
2031 | if (!(opts.subsys_mask & (1 << i)) || | |
2032 | ss->root == &cgrp_dfl_root) | |
2033 | continue; | |
2034 | ||
2035 | if (!percpu_ref_tryget_live(&ss->root->cgrp.self.refcnt)) { | |
2036 | mutex_unlock(&cgroup_mutex); | |
2037 | msleep(10); | |
2038 | ret = restart_syscall(); | |
2039 | goto out_free; | |
2040 | } | |
2041 | cgroup_put(&ss->root->cgrp); | |
2042 | } | |
2043 | ||
985ed670 | 2044 | for_each_root(root) { |
2bd59d48 | 2045 | bool name_match = false; |
3126121f | 2046 | |
3dd06ffa | 2047 | if (root == &cgrp_dfl_root) |
985ed670 | 2048 | continue; |
3126121f | 2049 | |
cf5d5941 | 2050 | /* |
2bd59d48 TH |
2051 | * If we asked for a name then it must match. Also, if |
2052 | * name matches but sybsys_mask doesn't, we should fail. | |
2053 | * Remember whether name matched. | |
cf5d5941 | 2054 | */ |
2bd59d48 TH |
2055 | if (opts.name) { |
2056 | if (strcmp(opts.name, root->name)) | |
2057 | continue; | |
2058 | name_match = true; | |
2059 | } | |
ddbcc7e8 | 2060 | |
c6d57f33 | 2061 | /* |
2bd59d48 TH |
2062 | * If we asked for subsystems (or explicitly for no |
2063 | * subsystems) then they must match. | |
c6d57f33 | 2064 | */ |
2bd59d48 | 2065 | if ((opts.subsys_mask || opts.none) && |
f392e51c | 2066 | (opts.subsys_mask != root->subsys_mask)) { |
2bd59d48 TH |
2067 | if (!name_match) |
2068 | continue; | |
2069 | ret = -EBUSY; | |
2070 | goto out_unlock; | |
2071 | } | |
873fe09e | 2072 | |
7b9a6ba5 TH |
2073 | if (root->flags ^ opts.flags) |
2074 | pr_warn("new mount options do not match the existing superblock, will be ignored\n"); | |
ddbcc7e8 | 2075 | |
776f02fa | 2076 | /* |
3a32bd72 LZ |
2077 | * We want to reuse @root whose lifetime is governed by its |
2078 | * ->cgrp. Let's check whether @root is alive and keep it | |
2079 | * that way. As cgroup_kill_sb() can happen anytime, we | |
2080 | * want to block it by pinning the sb so that @root doesn't | |
2081 | * get killed before mount is complete. | |
2082 | * | |
2083 | * With the sb pinned, tryget_live can reliably indicate | |
2084 | * whether @root can be reused. If it's being killed, | |
2085 | * drain it. We can use wait_queue for the wait but this | |
2086 | * path is super cold. Let's just sleep a bit and retry. | |
776f02fa | 2087 | */ |
3a32bd72 LZ |
2088 | pinned_sb = kernfs_pin_sb(root->kf_root, NULL); |
2089 | if (IS_ERR(pinned_sb) || | |
2090 | !percpu_ref_tryget_live(&root->cgrp.self.refcnt)) { | |
776f02fa | 2091 | mutex_unlock(&cgroup_mutex); |
3a32bd72 LZ |
2092 | if (!IS_ERR_OR_NULL(pinned_sb)) |
2093 | deactivate_super(pinned_sb); | |
776f02fa | 2094 | msleep(10); |
a015edd2 TH |
2095 | ret = restart_syscall(); |
2096 | goto out_free; | |
776f02fa | 2097 | } |
ddbcc7e8 | 2098 | |
776f02fa | 2099 | ret = 0; |
2bd59d48 | 2100 | goto out_unlock; |
ddbcc7e8 | 2101 | } |
ddbcc7e8 | 2102 | |
817929ec | 2103 | /* |
172a2c06 TH |
2104 | * No such thing, create a new one. name= matching without subsys |
2105 | * specification is allowed for already existing hierarchies but we | |
2106 | * can't create new one without subsys specification. | |
817929ec | 2107 | */ |
172a2c06 TH |
2108 | if (!opts.subsys_mask && !opts.none) { |
2109 | ret = -EINVAL; | |
2110 | goto out_unlock; | |
817929ec | 2111 | } |
817929ec | 2112 | |
172a2c06 TH |
2113 | root = kzalloc(sizeof(*root), GFP_KERNEL); |
2114 | if (!root) { | |
2115 | ret = -ENOMEM; | |
2bd59d48 | 2116 | goto out_unlock; |
839ec545 | 2117 | } |
e5f6a860 | 2118 | |
172a2c06 TH |
2119 | init_cgroup_root(root, &opts); |
2120 | ||
35585573 | 2121 | ret = cgroup_setup_root(root, opts.subsys_mask); |
2bd59d48 TH |
2122 | if (ret) |
2123 | cgroup_free_root(root); | |
fa3ca07e | 2124 | |
8e30e2b8 | 2125 | out_unlock: |
ddbcc7e8 | 2126 | mutex_unlock(&cgroup_mutex); |
a015edd2 | 2127 | out_free: |
c6d57f33 PM |
2128 | kfree(opts.release_agent); |
2129 | kfree(opts.name); | |
03b1cde6 | 2130 | |
2bd59d48 | 2131 | if (ret) |
8e30e2b8 | 2132 | return ERR_PTR(ret); |
67e9c74b | 2133 | out_mount: |
c9482a5b | 2134 | dentry = kernfs_mount(fs_type, flags, root->kf_root, |
67e9c74b TH |
2135 | is_v2 ? CGROUP2_SUPER_MAGIC : CGROUP_SUPER_MAGIC, |
2136 | &new_sb); | |
c6b3d5bc | 2137 | if (IS_ERR(dentry) || !new_sb) |
3dd06ffa | 2138 | cgroup_put(&root->cgrp); |
3a32bd72 LZ |
2139 | |
2140 | /* | |
2141 | * If @pinned_sb, we're reusing an existing root and holding an | |
2142 | * extra ref on its sb. Mount is complete. Put the extra ref. | |
2143 | */ | |
2144 | if (pinned_sb) { | |
2145 | WARN_ON(new_sb); | |
2146 | deactivate_super(pinned_sb); | |
2147 | } | |
2148 | ||
2bd59d48 TH |
2149 | return dentry; |
2150 | } | |
2151 | ||
2152 | static void cgroup_kill_sb(struct super_block *sb) | |
2153 | { | |
2154 | struct kernfs_root *kf_root = kernfs_root_from_sb(sb); | |
3dd06ffa | 2155 | struct cgroup_root *root = cgroup_root_from_kf(kf_root); |
2bd59d48 | 2156 | |
9d755d33 TH |
2157 | /* |
2158 | * If @root doesn't have any mounts or children, start killing it. | |
2159 | * This prevents new mounts by disabling percpu_ref_tryget_live(). | |
2160 | * cgroup_mount() may wait for @root's release. | |
1f779fb2 LZ |
2161 | * |
2162 | * And don't kill the default root. | |
9d755d33 | 2163 | */ |
3c606d35 | 2164 | if (!list_empty(&root->cgrp.self.children) || |
1f779fb2 | 2165 | root == &cgrp_dfl_root) |
9d755d33 TH |
2166 | cgroup_put(&root->cgrp); |
2167 | else | |
2168 | percpu_ref_kill(&root->cgrp.self.refcnt); | |
2169 | ||
2bd59d48 | 2170 | kernfs_kill_sb(sb); |
ddbcc7e8 PM |
2171 | } |
2172 | ||
2173 | static struct file_system_type cgroup_fs_type = { | |
2174 | .name = "cgroup", | |
f7e83571 | 2175 | .mount = cgroup_mount, |
ddbcc7e8 PM |
2176 | .kill_sb = cgroup_kill_sb, |
2177 | }; | |
2178 | ||
67e9c74b TH |
2179 | static struct file_system_type cgroup2_fs_type = { |
2180 | .name = "cgroup2", | |
2181 | .mount = cgroup_mount, | |
2182 | .kill_sb = cgroup_kill_sb, | |
2183 | }; | |
2184 | ||
857a2beb | 2185 | /** |
913ffdb5 | 2186 | * task_cgroup_path - cgroup path of a task in the first cgroup hierarchy |
857a2beb | 2187 | * @task: target task |
857a2beb TH |
2188 | * @buf: the buffer to write the path into |
2189 | * @buflen: the length of the buffer | |
2190 | * | |
913ffdb5 TH |
2191 | * Determine @task's cgroup on the first (the one with the lowest non-zero |
2192 | * hierarchy_id) cgroup hierarchy and copy its path into @buf. This | |
2193 | * function grabs cgroup_mutex and shouldn't be used inside locks used by | |
2194 | * cgroup controller callbacks. | |
2195 | * | |
e61734c5 | 2196 | * Return value is the same as kernfs_path(). |
857a2beb | 2197 | */ |
e61734c5 | 2198 | char *task_cgroup_path(struct task_struct *task, char *buf, size_t buflen) |
857a2beb | 2199 | { |
3dd06ffa | 2200 | struct cgroup_root *root; |
913ffdb5 | 2201 | struct cgroup *cgrp; |
e61734c5 TH |
2202 | int hierarchy_id = 1; |
2203 | char *path = NULL; | |
857a2beb TH |
2204 | |
2205 | mutex_lock(&cgroup_mutex); | |
f0d9a5f1 | 2206 | spin_lock_bh(&css_set_lock); |
857a2beb | 2207 | |
913ffdb5 TH |
2208 | root = idr_get_next(&cgroup_hierarchy_idr, &hierarchy_id); |
2209 | ||
857a2beb TH |
2210 | if (root) { |
2211 | cgrp = task_cgroup_from_root(task, root); | |
e61734c5 | 2212 | path = cgroup_path(cgrp, buf, buflen); |
913ffdb5 TH |
2213 | } else { |
2214 | /* if no hierarchy exists, everyone is in "/" */ | |
e61734c5 TH |
2215 | if (strlcpy(buf, "/", buflen) < buflen) |
2216 | path = buf; | |
857a2beb TH |
2217 | } |
2218 | ||
f0d9a5f1 | 2219 | spin_unlock_bh(&css_set_lock); |
857a2beb | 2220 | mutex_unlock(&cgroup_mutex); |
e61734c5 | 2221 | return path; |
857a2beb | 2222 | } |
913ffdb5 | 2223 | EXPORT_SYMBOL_GPL(task_cgroup_path); |
857a2beb | 2224 | |
b3dc094e | 2225 | /* used to track tasks and other necessary states during migration */ |
2f7ee569 | 2226 | struct cgroup_taskset { |
b3dc094e TH |
2227 | /* the src and dst cset list running through cset->mg_node */ |
2228 | struct list_head src_csets; | |
2229 | struct list_head dst_csets; | |
2230 | ||
1f7dd3e5 TH |
2231 | /* the subsys currently being processed */ |
2232 | int ssid; | |
2233 | ||
b3dc094e TH |
2234 | /* |
2235 | * Fields for cgroup_taskset_*() iteration. | |
2236 | * | |
2237 | * Before migration is committed, the target migration tasks are on | |
2238 | * ->mg_tasks of the csets on ->src_csets. After, on ->mg_tasks of | |
2239 | * the csets on ->dst_csets. ->csets point to either ->src_csets | |
2240 | * or ->dst_csets depending on whether migration is committed. | |
2241 | * | |
2242 | * ->cur_csets and ->cur_task point to the current task position | |
2243 | * during iteration. | |
2244 | */ | |
2245 | struct list_head *csets; | |
2246 | struct css_set *cur_cset; | |
2247 | struct task_struct *cur_task; | |
2f7ee569 TH |
2248 | }; |
2249 | ||
adaae5dc TH |
2250 | #define CGROUP_TASKSET_INIT(tset) (struct cgroup_taskset){ \ |
2251 | .src_csets = LIST_HEAD_INIT(tset.src_csets), \ | |
2252 | .dst_csets = LIST_HEAD_INIT(tset.dst_csets), \ | |
2253 | .csets = &tset.src_csets, \ | |
2254 | } | |
2255 | ||
2256 | /** | |
2257 | * cgroup_taskset_add - try to add a migration target task to a taskset | |
2258 | * @task: target task | |
2259 | * @tset: target taskset | |
2260 | * | |
2261 | * Add @task, which is a migration target, to @tset. This function becomes | |
2262 | * noop if @task doesn't need to be migrated. @task's css_set should have | |
2263 | * been added as a migration source and @task->cg_list will be moved from | |
2264 | * the css_set's tasks list to mg_tasks one. | |
2265 | */ | |
2266 | static void cgroup_taskset_add(struct task_struct *task, | |
2267 | struct cgroup_taskset *tset) | |
2268 | { | |
2269 | struct css_set *cset; | |
2270 | ||
f0d9a5f1 | 2271 | lockdep_assert_held(&css_set_lock); |
adaae5dc TH |
2272 | |
2273 | /* @task either already exited or can't exit until the end */ | |
2274 | if (task->flags & PF_EXITING) | |
2275 | return; | |
2276 | ||
2277 | /* leave @task alone if post_fork() hasn't linked it yet */ | |
2278 | if (list_empty(&task->cg_list)) | |
2279 | return; | |
2280 | ||
2281 | cset = task_css_set(task); | |
2282 | if (!cset->mg_src_cgrp) | |
2283 | return; | |
2284 | ||
2285 | list_move_tail(&task->cg_list, &cset->mg_tasks); | |
2286 | if (list_empty(&cset->mg_node)) | |
2287 | list_add_tail(&cset->mg_node, &tset->src_csets); | |
2288 | if (list_empty(&cset->mg_dst_cset->mg_node)) | |
2289 | list_move_tail(&cset->mg_dst_cset->mg_node, | |
2290 | &tset->dst_csets); | |
2291 | } | |
2292 | ||
2f7ee569 TH |
2293 | /** |
2294 | * cgroup_taskset_first - reset taskset and return the first task | |
2295 | * @tset: taskset of interest | |
1f7dd3e5 | 2296 | * @dst_cssp: output variable for the destination css |
2f7ee569 TH |
2297 | * |
2298 | * @tset iteration is initialized and the first task is returned. | |
2299 | */ | |
1f7dd3e5 TH |
2300 | struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset, |
2301 | struct cgroup_subsys_state **dst_cssp) | |
2f7ee569 | 2302 | { |
b3dc094e TH |
2303 | tset->cur_cset = list_first_entry(tset->csets, struct css_set, mg_node); |
2304 | tset->cur_task = NULL; | |
2305 | ||
1f7dd3e5 | 2306 | return cgroup_taskset_next(tset, dst_cssp); |
2f7ee569 | 2307 | } |
2f7ee569 TH |
2308 | |
2309 | /** | |
2310 | * cgroup_taskset_next - iterate to the next task in taskset | |
2311 | * @tset: taskset of interest | |
1f7dd3e5 | 2312 | * @dst_cssp: output variable for the destination css |
2f7ee569 TH |
2313 | * |
2314 | * Return the next task in @tset. Iteration must have been initialized | |
2315 | * with cgroup_taskset_first(). | |
2316 | */ | |
1f7dd3e5 TH |
2317 | struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset, |
2318 | struct cgroup_subsys_state **dst_cssp) | |
2f7ee569 | 2319 | { |
b3dc094e TH |
2320 | struct css_set *cset = tset->cur_cset; |
2321 | struct task_struct *task = tset->cur_task; | |
2f7ee569 | 2322 | |
b3dc094e TH |
2323 | while (&cset->mg_node != tset->csets) { |
2324 | if (!task) | |
2325 | task = list_first_entry(&cset->mg_tasks, | |
2326 | struct task_struct, cg_list); | |
2327 | else | |
2328 | task = list_next_entry(task, cg_list); | |
2f7ee569 | 2329 | |
b3dc094e TH |
2330 | if (&task->cg_list != &cset->mg_tasks) { |
2331 | tset->cur_cset = cset; | |
2332 | tset->cur_task = task; | |
1f7dd3e5 TH |
2333 | |
2334 | /* | |
2335 | * This function may be called both before and | |
2336 | * after cgroup_taskset_migrate(). The two cases | |
2337 | * can be distinguished by looking at whether @cset | |
2338 | * has its ->mg_dst_cset set. | |
2339 | */ | |
2340 | if (cset->mg_dst_cset) | |
2341 | *dst_cssp = cset->mg_dst_cset->subsys[tset->ssid]; | |
2342 | else | |
2343 | *dst_cssp = cset->subsys[tset->ssid]; | |
2344 | ||
b3dc094e TH |
2345 | return task; |
2346 | } | |
2f7ee569 | 2347 | |
b3dc094e TH |
2348 | cset = list_next_entry(cset, mg_node); |
2349 | task = NULL; | |
2350 | } | |
2f7ee569 | 2351 | |
b3dc094e | 2352 | return NULL; |
2f7ee569 | 2353 | } |
2f7ee569 | 2354 | |
adaae5dc TH |
2355 | /** |
2356 | * cgroup_taskset_migrate - migrate a taskset to a cgroup | |
2357 | * @tset: taget taskset | |
2358 | * @dst_cgrp: destination cgroup | |
2359 | * | |
2360 | * Migrate tasks in @tset to @dst_cgrp. This function fails iff one of the | |
2361 | * ->can_attach callbacks fails and guarantees that either all or none of | |
2362 | * the tasks in @tset are migrated. @tset is consumed regardless of | |
2363 | * success. | |
2364 | */ | |
2365 | static int cgroup_taskset_migrate(struct cgroup_taskset *tset, | |
2366 | struct cgroup *dst_cgrp) | |
2367 | { | |
2368 | struct cgroup_subsys_state *css, *failed_css = NULL; | |
2369 | struct task_struct *task, *tmp_task; | |
2370 | struct css_set *cset, *tmp_cset; | |
2371 | int i, ret; | |
2372 | ||
2373 | /* methods shouldn't be called if no task is actually migrating */ | |
2374 | if (list_empty(&tset->src_csets)) | |
2375 | return 0; | |
2376 | ||
2377 | /* check that we can legitimately attach to the cgroup */ | |
2378 | for_each_e_css(css, i, dst_cgrp) { | |
2379 | if (css->ss->can_attach) { | |
1f7dd3e5 TH |
2380 | tset->ssid = i; |
2381 | ret = css->ss->can_attach(tset); | |
adaae5dc TH |
2382 | if (ret) { |
2383 | failed_css = css; | |
2384 | goto out_cancel_attach; | |
2385 | } | |
2386 | } | |
2387 | } | |
2388 | ||
2389 | /* | |
2390 | * Now that we're guaranteed success, proceed to move all tasks to | |
2391 | * the new cgroup. There are no failure cases after here, so this | |
2392 | * is the commit point. | |
2393 | */ | |
f0d9a5f1 | 2394 | spin_lock_bh(&css_set_lock); |
adaae5dc | 2395 | list_for_each_entry(cset, &tset->src_csets, mg_node) { |
f6d7d049 TH |
2396 | list_for_each_entry_safe(task, tmp_task, &cset->mg_tasks, cg_list) { |
2397 | struct css_set *from_cset = task_css_set(task); | |
2398 | struct css_set *to_cset = cset->mg_dst_cset; | |
2399 | ||
2400 | get_css_set(to_cset); | |
2401 | css_set_move_task(task, from_cset, to_cset, true); | |
2402 | put_css_set_locked(from_cset); | |
2403 | } | |
adaae5dc | 2404 | } |
f0d9a5f1 | 2405 | spin_unlock_bh(&css_set_lock); |
adaae5dc TH |
2406 | |
2407 | /* | |
2408 | * Migration is committed, all target tasks are now on dst_csets. | |
2409 | * Nothing is sensitive to fork() after this point. Notify | |
2410 | * controllers that migration is complete. | |
2411 | */ | |
2412 | tset->csets = &tset->dst_csets; | |
2413 | ||
1f7dd3e5 TH |
2414 | for_each_e_css(css, i, dst_cgrp) { |
2415 | if (css->ss->attach) { | |
2416 | tset->ssid = i; | |
2417 | css->ss->attach(tset); | |
2418 | } | |
2419 | } | |
adaae5dc TH |
2420 | |
2421 | ret = 0; | |
2422 | goto out_release_tset; | |
2423 | ||
2424 | out_cancel_attach: | |
2425 | for_each_e_css(css, i, dst_cgrp) { | |
2426 | if (css == failed_css) | |
2427 | break; | |
1f7dd3e5 TH |
2428 | if (css->ss->cancel_attach) { |
2429 | tset->ssid = i; | |
2430 | css->ss->cancel_attach(tset); | |
2431 | } | |
adaae5dc TH |
2432 | } |
2433 | out_release_tset: | |
f0d9a5f1 | 2434 | spin_lock_bh(&css_set_lock); |
adaae5dc TH |
2435 | list_splice_init(&tset->dst_csets, &tset->src_csets); |
2436 | list_for_each_entry_safe(cset, tmp_cset, &tset->src_csets, mg_node) { | |
2437 | list_splice_tail_init(&cset->mg_tasks, &cset->tasks); | |
2438 | list_del_init(&cset->mg_node); | |
2439 | } | |
f0d9a5f1 | 2440 | spin_unlock_bh(&css_set_lock); |
adaae5dc TH |
2441 | return ret; |
2442 | } | |
2443 | ||
a043e3b2 | 2444 | /** |
1958d2d5 TH |
2445 | * cgroup_migrate_finish - cleanup after attach |
2446 | * @preloaded_csets: list of preloaded css_sets | |
74a1166d | 2447 | * |
1958d2d5 TH |
2448 | * Undo cgroup_migrate_add_src() and cgroup_migrate_prepare_dst(). See |
2449 | * those functions for details. | |
74a1166d | 2450 | */ |
1958d2d5 | 2451 | static void cgroup_migrate_finish(struct list_head *preloaded_csets) |
74a1166d | 2452 | { |
1958d2d5 | 2453 | struct css_set *cset, *tmp_cset; |
74a1166d | 2454 | |
1958d2d5 TH |
2455 | lockdep_assert_held(&cgroup_mutex); |
2456 | ||
f0d9a5f1 | 2457 | spin_lock_bh(&css_set_lock); |
1958d2d5 TH |
2458 | list_for_each_entry_safe(cset, tmp_cset, preloaded_csets, mg_preload_node) { |
2459 | cset->mg_src_cgrp = NULL; | |
2460 | cset->mg_dst_cset = NULL; | |
2461 | list_del_init(&cset->mg_preload_node); | |
a25eb52e | 2462 | put_css_set_locked(cset); |
1958d2d5 | 2463 | } |
f0d9a5f1 | 2464 | spin_unlock_bh(&css_set_lock); |
1958d2d5 TH |
2465 | } |
2466 | ||
2467 | /** | |
2468 | * cgroup_migrate_add_src - add a migration source css_set | |
2469 | * @src_cset: the source css_set to add | |
2470 | * @dst_cgrp: the destination cgroup | |
2471 | * @preloaded_csets: list of preloaded css_sets | |
2472 | * | |
2473 | * Tasks belonging to @src_cset are about to be migrated to @dst_cgrp. Pin | |
2474 | * @src_cset and add it to @preloaded_csets, which should later be cleaned | |
2475 | * up by cgroup_migrate_finish(). | |
2476 | * | |
1ed13287 TH |
2477 | * This function may be called without holding cgroup_threadgroup_rwsem |
2478 | * even if the target is a process. Threads may be created and destroyed | |
2479 | * but as long as cgroup_mutex is not dropped, no new css_set can be put | |
2480 | * into play and the preloaded css_sets are guaranteed to cover all | |
2481 | * migrations. | |
1958d2d5 TH |
2482 | */ |
2483 | static void cgroup_migrate_add_src(struct css_set *src_cset, | |
2484 | struct cgroup *dst_cgrp, | |
2485 | struct list_head *preloaded_csets) | |
2486 | { | |
2487 | struct cgroup *src_cgrp; | |
2488 | ||
2489 | lockdep_assert_held(&cgroup_mutex); | |
f0d9a5f1 | 2490 | lockdep_assert_held(&css_set_lock); |
1958d2d5 TH |
2491 | |
2492 | src_cgrp = cset_cgroup_from_root(src_cset, dst_cgrp->root); | |
2493 | ||
1958d2d5 TH |
2494 | if (!list_empty(&src_cset->mg_preload_node)) |
2495 | return; | |
2496 | ||
2497 | WARN_ON(src_cset->mg_src_cgrp); | |
2498 | WARN_ON(!list_empty(&src_cset->mg_tasks)); | |
2499 | WARN_ON(!list_empty(&src_cset->mg_node)); | |
2500 | ||
2501 | src_cset->mg_src_cgrp = src_cgrp; | |
2502 | get_css_set(src_cset); | |
2503 | list_add(&src_cset->mg_preload_node, preloaded_csets); | |
2504 | } | |
2505 | ||
2506 | /** | |
2507 | * cgroup_migrate_prepare_dst - prepare destination css_sets for migration | |
f817de98 | 2508 | * @dst_cgrp: the destination cgroup (may be %NULL) |
1958d2d5 TH |
2509 | * @preloaded_csets: list of preloaded source css_sets |
2510 | * | |
2511 | * Tasks are about to be moved to @dst_cgrp and all the source css_sets | |
2512 | * have been preloaded to @preloaded_csets. This function looks up and | |
f817de98 TH |
2513 | * pins all destination css_sets, links each to its source, and append them |
2514 | * to @preloaded_csets. If @dst_cgrp is %NULL, the destination of each | |
2515 | * source css_set is assumed to be its cgroup on the default hierarchy. | |
1958d2d5 TH |
2516 | * |
2517 | * This function must be called after cgroup_migrate_add_src() has been | |
2518 | * called on each migration source css_set. After migration is performed | |
2519 | * using cgroup_migrate(), cgroup_migrate_finish() must be called on | |
2520 | * @preloaded_csets. | |
2521 | */ | |
2522 | static int cgroup_migrate_prepare_dst(struct cgroup *dst_cgrp, | |
2523 | struct list_head *preloaded_csets) | |
2524 | { | |
2525 | LIST_HEAD(csets); | |
f817de98 | 2526 | struct css_set *src_cset, *tmp_cset; |
1958d2d5 TH |
2527 | |
2528 | lockdep_assert_held(&cgroup_mutex); | |
2529 | ||
f8f22e53 | 2530 | /* |
62716ea0 | 2531 | * Except for the root, subtree_control must be zero for a cgroup |
f8f22e53 TH |
2532 | * with tasks so that child cgroups don't compete against tasks. |
2533 | */ | |
d51f39b0 | 2534 | if (dst_cgrp && cgroup_on_dfl(dst_cgrp) && cgroup_parent(dst_cgrp) && |
62716ea0 | 2535 | dst_cgrp->subtree_control) |
f8f22e53 TH |
2536 | return -EBUSY; |
2537 | ||
1958d2d5 | 2538 | /* look up the dst cset for each src cset and link it to src */ |
f817de98 | 2539 | list_for_each_entry_safe(src_cset, tmp_cset, preloaded_csets, mg_preload_node) { |
1958d2d5 TH |
2540 | struct css_set *dst_cset; |
2541 | ||
f817de98 TH |
2542 | dst_cset = find_css_set(src_cset, |
2543 | dst_cgrp ?: src_cset->dfl_cgrp); | |
1958d2d5 TH |
2544 | if (!dst_cset) |
2545 | goto err; | |
2546 | ||
2547 | WARN_ON_ONCE(src_cset->mg_dst_cset || dst_cset->mg_dst_cset); | |
f817de98 TH |
2548 | |
2549 | /* | |
2550 | * If src cset equals dst, it's noop. Drop the src. | |
2551 | * cgroup_migrate() will skip the cset too. Note that we | |
2552 | * can't handle src == dst as some nodes are used by both. | |
2553 | */ | |
2554 | if (src_cset == dst_cset) { | |
2555 | src_cset->mg_src_cgrp = NULL; | |
2556 | list_del_init(&src_cset->mg_preload_node); | |
a25eb52e ZL |
2557 | put_css_set(src_cset); |
2558 | put_css_set(dst_cset); | |
f817de98 TH |
2559 | continue; |
2560 | } | |
2561 | ||
1958d2d5 TH |
2562 | src_cset->mg_dst_cset = dst_cset; |
2563 | ||
2564 | if (list_empty(&dst_cset->mg_preload_node)) | |
2565 | list_add(&dst_cset->mg_preload_node, &csets); | |
2566 | else | |
a25eb52e | 2567 | put_css_set(dst_cset); |
1958d2d5 TH |
2568 | } |
2569 | ||
f817de98 | 2570 | list_splice_tail(&csets, preloaded_csets); |
1958d2d5 TH |
2571 | return 0; |
2572 | err: | |
2573 | cgroup_migrate_finish(&csets); | |
2574 | return -ENOMEM; | |
2575 | } | |
2576 | ||
2577 | /** | |
2578 | * cgroup_migrate - migrate a process or task to a cgroup | |
1958d2d5 TH |
2579 | * @leader: the leader of the process or the task to migrate |
2580 | * @threadgroup: whether @leader points to the whole process or a single task | |
9af2ec45 | 2581 | * @cgrp: the destination cgroup |
1958d2d5 TH |
2582 | * |
2583 | * Migrate a process or task denoted by @leader to @cgrp. If migrating a | |
1ed13287 | 2584 | * process, the caller must be holding cgroup_threadgroup_rwsem. The |
1958d2d5 TH |
2585 | * caller is also responsible for invoking cgroup_migrate_add_src() and |
2586 | * cgroup_migrate_prepare_dst() on the targets before invoking this | |
2587 | * function and following up with cgroup_migrate_finish(). | |
2588 | * | |
2589 | * As long as a controller's ->can_attach() doesn't fail, this function is | |
2590 | * guaranteed to succeed. This means that, excluding ->can_attach() | |
2591 | * failure, when migrating multiple targets, the success or failure can be | |
2592 | * decided for all targets by invoking group_migrate_prepare_dst() before | |
2593 | * actually starting migrating. | |
2594 | */ | |
9af2ec45 TH |
2595 | static int cgroup_migrate(struct task_struct *leader, bool threadgroup, |
2596 | struct cgroup *cgrp) | |
74a1166d | 2597 | { |
adaae5dc TH |
2598 | struct cgroup_taskset tset = CGROUP_TASKSET_INIT(tset); |
2599 | struct task_struct *task; | |
74a1166d | 2600 | |
fb5d2b4c MSB |
2601 | /* |
2602 | * Prevent freeing of tasks while we take a snapshot. Tasks that are | |
2603 | * already PF_EXITING could be freed from underneath us unless we | |
2604 | * take an rcu_read_lock. | |
2605 | */ | |
f0d9a5f1 | 2606 | spin_lock_bh(&css_set_lock); |
fb5d2b4c | 2607 | rcu_read_lock(); |
9db8de37 | 2608 | task = leader; |
74a1166d | 2609 | do { |
adaae5dc | 2610 | cgroup_taskset_add(task, &tset); |
081aa458 LZ |
2611 | if (!threadgroup) |
2612 | break; | |
9db8de37 | 2613 | } while_each_thread(leader, task); |
fb5d2b4c | 2614 | rcu_read_unlock(); |
f0d9a5f1 | 2615 | spin_unlock_bh(&css_set_lock); |
74a1166d | 2616 | |
adaae5dc | 2617 | return cgroup_taskset_migrate(&tset, cgrp); |
74a1166d BB |
2618 | } |
2619 | ||
1958d2d5 TH |
2620 | /** |
2621 | * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup | |
2622 | * @dst_cgrp: the cgroup to attach to | |
2623 | * @leader: the task or the leader of the threadgroup to be attached | |
2624 | * @threadgroup: attach the whole threadgroup? | |
2625 | * | |
1ed13287 | 2626 | * Call holding cgroup_mutex and cgroup_threadgroup_rwsem. |
1958d2d5 TH |
2627 | */ |
2628 | static int cgroup_attach_task(struct cgroup *dst_cgrp, | |
2629 | struct task_struct *leader, bool threadgroup) | |
2630 | { | |
2631 | LIST_HEAD(preloaded_csets); | |
2632 | struct task_struct *task; | |
2633 | int ret; | |
2634 | ||
2635 | /* look up all src csets */ | |
f0d9a5f1 | 2636 | spin_lock_bh(&css_set_lock); |
1958d2d5 TH |
2637 | rcu_read_lock(); |
2638 | task = leader; | |
2639 | do { | |
2640 | cgroup_migrate_add_src(task_css_set(task), dst_cgrp, | |
2641 | &preloaded_csets); | |
2642 | if (!threadgroup) | |
2643 | break; | |
2644 | } while_each_thread(leader, task); | |
2645 | rcu_read_unlock(); | |
f0d9a5f1 | 2646 | spin_unlock_bh(&css_set_lock); |
1958d2d5 TH |
2647 | |
2648 | /* prepare dst csets and commit */ | |
2649 | ret = cgroup_migrate_prepare_dst(dst_cgrp, &preloaded_csets); | |
2650 | if (!ret) | |
9af2ec45 | 2651 | ret = cgroup_migrate(leader, threadgroup, dst_cgrp); |
1958d2d5 TH |
2652 | |
2653 | cgroup_migrate_finish(&preloaded_csets); | |
2654 | return ret; | |
74a1166d BB |
2655 | } |
2656 | ||
187fe840 TH |
2657 | static int cgroup_procs_write_permission(struct task_struct *task, |
2658 | struct cgroup *dst_cgrp, | |
2659 | struct kernfs_open_file *of) | |
dedf22e9 TH |
2660 | { |
2661 | const struct cred *cred = current_cred(); | |
2662 | const struct cred *tcred = get_task_cred(task); | |
2663 | int ret = 0; | |
2664 | ||
2665 | /* | |
2666 | * even if we're attaching all tasks in the thread group, we only | |
2667 | * need to check permissions on one of them. | |
2668 | */ | |
2669 | if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) && | |
2670 | !uid_eq(cred->euid, tcred->uid) && | |
2671 | !uid_eq(cred->euid, tcred->suid)) | |
2672 | ret = -EACCES; | |
2673 | ||
187fe840 TH |
2674 | if (!ret && cgroup_on_dfl(dst_cgrp)) { |
2675 | struct super_block *sb = of->file->f_path.dentry->d_sb; | |
2676 | struct cgroup *cgrp; | |
2677 | struct inode *inode; | |
2678 | ||
f0d9a5f1 | 2679 | spin_lock_bh(&css_set_lock); |
187fe840 | 2680 | cgrp = task_cgroup_from_root(task, &cgrp_dfl_root); |
f0d9a5f1 | 2681 | spin_unlock_bh(&css_set_lock); |
187fe840 TH |
2682 | |
2683 | while (!cgroup_is_descendant(dst_cgrp, cgrp)) | |
2684 | cgrp = cgroup_parent(cgrp); | |
2685 | ||
2686 | ret = -ENOMEM; | |
6f60eade | 2687 | inode = kernfs_get_inode(sb, cgrp->procs_file.kn); |
187fe840 TH |
2688 | if (inode) { |
2689 | ret = inode_permission(inode, MAY_WRITE); | |
2690 | iput(inode); | |
2691 | } | |
2692 | } | |
2693 | ||
dedf22e9 TH |
2694 | put_cred(tcred); |
2695 | return ret; | |
2696 | } | |
2697 | ||
74a1166d BB |
2698 | /* |
2699 | * Find the task_struct of the task to attach by vpid and pass it along to the | |
cd3d0952 | 2700 | * function to attach either it or all tasks in its threadgroup. Will lock |
0e1d768f | 2701 | * cgroup_mutex and threadgroup. |
bbcb81d0 | 2702 | */ |
acbef755 TH |
2703 | static ssize_t __cgroup_procs_write(struct kernfs_open_file *of, char *buf, |
2704 | size_t nbytes, loff_t off, bool threadgroup) | |
bbcb81d0 | 2705 | { |
bbcb81d0 | 2706 | struct task_struct *tsk; |
e76ecaee | 2707 | struct cgroup *cgrp; |
acbef755 | 2708 | pid_t pid; |
bbcb81d0 PM |
2709 | int ret; |
2710 | ||
acbef755 TH |
2711 | if (kstrtoint(strstrip(buf), 0, &pid) || pid < 0) |
2712 | return -EINVAL; | |
2713 | ||
e76ecaee TH |
2714 | cgrp = cgroup_kn_lock_live(of->kn); |
2715 | if (!cgrp) | |
74a1166d BB |
2716 | return -ENODEV; |
2717 | ||
3014dde7 | 2718 | percpu_down_write(&cgroup_threadgroup_rwsem); |
b78949eb | 2719 | rcu_read_lock(); |
bbcb81d0 | 2720 | if (pid) { |
73507f33 | 2721 | tsk = find_task_by_vpid(pid); |
74a1166d | 2722 | if (!tsk) { |
dd4b0a46 | 2723 | ret = -ESRCH; |
3014dde7 | 2724 | goto out_unlock_rcu; |
bbcb81d0 | 2725 | } |
dedf22e9 | 2726 | } else { |
b78949eb | 2727 | tsk = current; |
dedf22e9 | 2728 | } |
cd3d0952 TH |
2729 | |
2730 | if (threadgroup) | |
b78949eb | 2731 | tsk = tsk->group_leader; |
c4c27fbd MG |
2732 | |
2733 | /* | |
14a40ffc | 2734 | * Workqueue threads may acquire PF_NO_SETAFFINITY and become |
c4c27fbd MG |
2735 | * trapped in a cpuset, or RT worker may be born in a cgroup |
2736 | * with no rt_runtime allocated. Just say no. | |
2737 | */ | |
14a40ffc | 2738 | if (tsk == kthreadd_task || (tsk->flags & PF_NO_SETAFFINITY)) { |
c4c27fbd | 2739 | ret = -EINVAL; |
3014dde7 | 2740 | goto out_unlock_rcu; |
c4c27fbd MG |
2741 | } |
2742 | ||
b78949eb MSB |
2743 | get_task_struct(tsk); |
2744 | rcu_read_unlock(); | |
2745 | ||
187fe840 | 2746 | ret = cgroup_procs_write_permission(tsk, cgrp, of); |
dedf22e9 TH |
2747 | if (!ret) |
2748 | ret = cgroup_attach_task(cgrp, tsk, threadgroup); | |
081aa458 | 2749 | |
f9f9e7b7 | 2750 | put_task_struct(tsk); |
3014dde7 TH |
2751 | goto out_unlock_threadgroup; |
2752 | ||
2753 | out_unlock_rcu: | |
2754 | rcu_read_unlock(); | |
2755 | out_unlock_threadgroup: | |
2756 | percpu_up_write(&cgroup_threadgroup_rwsem); | |
e76ecaee | 2757 | cgroup_kn_unlock(of->kn); |
e93ad19d | 2758 | cpuset_post_attach_flush(); |
acbef755 | 2759 | return ret ?: nbytes; |
bbcb81d0 PM |
2760 | } |
2761 | ||
7ae1bad9 TH |
2762 | /** |
2763 | * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from' | |
2764 | * @from: attach to all cgroups of a given task | |
2765 | * @tsk: the task to be attached | |
2766 | */ | |
2767 | int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk) | |
2768 | { | |
3dd06ffa | 2769 | struct cgroup_root *root; |
7ae1bad9 TH |
2770 | int retval = 0; |
2771 | ||
47cfcd09 | 2772 | mutex_lock(&cgroup_mutex); |
985ed670 | 2773 | for_each_root(root) { |
96d365e0 TH |
2774 | struct cgroup *from_cgrp; |
2775 | ||
3dd06ffa | 2776 | if (root == &cgrp_dfl_root) |
985ed670 TH |
2777 | continue; |
2778 | ||
f0d9a5f1 | 2779 | spin_lock_bh(&css_set_lock); |
96d365e0 | 2780 | from_cgrp = task_cgroup_from_root(from, root); |
f0d9a5f1 | 2781 | spin_unlock_bh(&css_set_lock); |
7ae1bad9 | 2782 | |
6f4b7e63 | 2783 | retval = cgroup_attach_task(from_cgrp, tsk, false); |
7ae1bad9 TH |
2784 | if (retval) |
2785 | break; | |
2786 | } | |
47cfcd09 | 2787 | mutex_unlock(&cgroup_mutex); |
7ae1bad9 TH |
2788 | |
2789 | return retval; | |
2790 | } | |
2791 | EXPORT_SYMBOL_GPL(cgroup_attach_task_all); | |
2792 | ||
acbef755 TH |
2793 | static ssize_t cgroup_tasks_write(struct kernfs_open_file *of, |
2794 | char *buf, size_t nbytes, loff_t off) | |
74a1166d | 2795 | { |
acbef755 | 2796 | return __cgroup_procs_write(of, buf, nbytes, off, false); |
74a1166d BB |
2797 | } |
2798 | ||
acbef755 TH |
2799 | static ssize_t cgroup_procs_write(struct kernfs_open_file *of, |
2800 | char *buf, size_t nbytes, loff_t off) | |
af351026 | 2801 | { |
acbef755 | 2802 | return __cgroup_procs_write(of, buf, nbytes, off, true); |
af351026 PM |
2803 | } |
2804 | ||
451af504 TH |
2805 | static ssize_t cgroup_release_agent_write(struct kernfs_open_file *of, |
2806 | char *buf, size_t nbytes, loff_t off) | |
e788e066 | 2807 | { |
e76ecaee | 2808 | struct cgroup *cgrp; |
5f469907 | 2809 | |
e76ecaee | 2810 | BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX); |
5f469907 | 2811 | |
e76ecaee TH |
2812 | cgrp = cgroup_kn_lock_live(of->kn); |
2813 | if (!cgrp) | |
e788e066 | 2814 | return -ENODEV; |
69e943b7 | 2815 | spin_lock(&release_agent_path_lock); |
e76ecaee TH |
2816 | strlcpy(cgrp->root->release_agent_path, strstrip(buf), |
2817 | sizeof(cgrp->root->release_agent_path)); | |
69e943b7 | 2818 | spin_unlock(&release_agent_path_lock); |
e76ecaee | 2819 | cgroup_kn_unlock(of->kn); |
451af504 | 2820 | return nbytes; |
e788e066 PM |
2821 | } |
2822 | ||
2da8ca82 | 2823 | static int cgroup_release_agent_show(struct seq_file *seq, void *v) |
e788e066 | 2824 | { |
2da8ca82 | 2825 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
182446d0 | 2826 | |
46cfeb04 | 2827 | spin_lock(&release_agent_path_lock); |
e788e066 | 2828 | seq_puts(seq, cgrp->root->release_agent_path); |
46cfeb04 | 2829 | spin_unlock(&release_agent_path_lock); |
e788e066 | 2830 | seq_putc(seq, '\n'); |
e788e066 PM |
2831 | return 0; |
2832 | } | |
2833 | ||
2da8ca82 | 2834 | static int cgroup_sane_behavior_show(struct seq_file *seq, void *v) |
873fe09e | 2835 | { |
c1d5d42e | 2836 | seq_puts(seq, "0\n"); |
e788e066 PM |
2837 | return 0; |
2838 | } | |
2839 | ||
6e5c8307 | 2840 | static void cgroup_print_ss_mask(struct seq_file *seq, u16 ss_mask) |
355e0c48 | 2841 | { |
f8f22e53 TH |
2842 | struct cgroup_subsys *ss; |
2843 | bool printed = false; | |
2844 | int ssid; | |
a742c59d | 2845 | |
b4e0eeaf | 2846 | do_each_subsys_mask(ss, ssid, ss_mask) { |
a966a4ed AS |
2847 | if (printed) |
2848 | seq_putc(seq, ' '); | |
2849 | seq_printf(seq, "%s", ss->name); | |
2850 | printed = true; | |
b4e0eeaf | 2851 | } while_each_subsys_mask(); |
f8f22e53 TH |
2852 | if (printed) |
2853 | seq_putc(seq, '\n'); | |
355e0c48 PM |
2854 | } |
2855 | ||
f8f22e53 TH |
2856 | /* show controllers which are currently attached to the default hierarchy */ |
2857 | static int cgroup_root_controllers_show(struct seq_file *seq, void *v) | |
db3b1497 | 2858 | { |
f8f22e53 TH |
2859 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
2860 | ||
5533e011 | 2861 | cgroup_print_ss_mask(seq, cgrp->root->subsys_mask & |
a7165264 | 2862 | ~cgrp_dfl_inhibit_ss_mask); |
f8f22e53 | 2863 | return 0; |
db3b1497 PM |
2864 | } |
2865 | ||
f8f22e53 TH |
2866 | /* show controllers which are enabled from the parent */ |
2867 | static int cgroup_controllers_show(struct seq_file *seq, void *v) | |
ddbcc7e8 | 2868 | { |
f8f22e53 TH |
2869 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
2870 | ||
667c2491 | 2871 | cgroup_print_ss_mask(seq, cgroup_parent(cgrp)->subtree_control); |
f8f22e53 | 2872 | return 0; |
ddbcc7e8 PM |
2873 | } |
2874 | ||
f8f22e53 TH |
2875 | /* show controllers which are enabled for a given cgroup's children */ |
2876 | static int cgroup_subtree_control_show(struct seq_file *seq, void *v) | |
ddbcc7e8 | 2877 | { |
f8f22e53 TH |
2878 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
2879 | ||
667c2491 | 2880 | cgroup_print_ss_mask(seq, cgrp->subtree_control); |
f8f22e53 TH |
2881 | return 0; |
2882 | } | |
2883 | ||
2884 | /** | |
2885 | * cgroup_update_dfl_csses - update css assoc of a subtree in default hierarchy | |
2886 | * @cgrp: root of the subtree to update csses for | |
2887 | * | |
8699b776 | 2888 | * @cgrp's subtree_ss_mask has changed and its subtree's (self excluded) |
f8f22e53 TH |
2889 | * css associations need to be updated accordingly. This function looks up |
2890 | * all css_sets which are attached to the subtree, creates the matching | |
2891 | * updated css_sets and migrates the tasks to the new ones. | |
2892 | */ | |
2893 | static int cgroup_update_dfl_csses(struct cgroup *cgrp) | |
2894 | { | |
2895 | LIST_HEAD(preloaded_csets); | |
10265075 | 2896 | struct cgroup_taskset tset = CGROUP_TASKSET_INIT(tset); |
f8f22e53 TH |
2897 | struct cgroup_subsys_state *css; |
2898 | struct css_set *src_cset; | |
2899 | int ret; | |
2900 | ||
f8f22e53 TH |
2901 | lockdep_assert_held(&cgroup_mutex); |
2902 | ||
3014dde7 TH |
2903 | percpu_down_write(&cgroup_threadgroup_rwsem); |
2904 | ||
f8f22e53 | 2905 | /* look up all csses currently attached to @cgrp's subtree */ |
f0d9a5f1 | 2906 | spin_lock_bh(&css_set_lock); |
f8f22e53 TH |
2907 | css_for_each_descendant_pre(css, cgroup_css(cgrp, NULL)) { |
2908 | struct cgrp_cset_link *link; | |
2909 | ||
8699b776 | 2910 | /* self is not affected by subtree_ss_mask change */ |
f8f22e53 TH |
2911 | if (css->cgroup == cgrp) |
2912 | continue; | |
2913 | ||
2914 | list_for_each_entry(link, &css->cgroup->cset_links, cset_link) | |
2915 | cgroup_migrate_add_src(link->cset, cgrp, | |
2916 | &preloaded_csets); | |
2917 | } | |
f0d9a5f1 | 2918 | spin_unlock_bh(&css_set_lock); |
f8f22e53 TH |
2919 | |
2920 | /* NULL dst indicates self on default hierarchy */ | |
2921 | ret = cgroup_migrate_prepare_dst(NULL, &preloaded_csets); | |
2922 | if (ret) | |
2923 | goto out_finish; | |
2924 | ||
f0d9a5f1 | 2925 | spin_lock_bh(&css_set_lock); |
f8f22e53 | 2926 | list_for_each_entry(src_cset, &preloaded_csets, mg_preload_node) { |
10265075 | 2927 | struct task_struct *task, *ntask; |
f8f22e53 TH |
2928 | |
2929 | /* src_csets precede dst_csets, break on the first dst_cset */ | |
2930 | if (!src_cset->mg_src_cgrp) | |
2931 | break; | |
2932 | ||
10265075 TH |
2933 | /* all tasks in src_csets need to be migrated */ |
2934 | list_for_each_entry_safe(task, ntask, &src_cset->tasks, cg_list) | |
2935 | cgroup_taskset_add(task, &tset); | |
f8f22e53 | 2936 | } |
f0d9a5f1 | 2937 | spin_unlock_bh(&css_set_lock); |
f8f22e53 | 2938 | |
10265075 | 2939 | ret = cgroup_taskset_migrate(&tset, cgrp); |
f8f22e53 TH |
2940 | out_finish: |
2941 | cgroup_migrate_finish(&preloaded_csets); | |
3014dde7 | 2942 | percpu_up_write(&cgroup_threadgroup_rwsem); |
f8f22e53 TH |
2943 | return ret; |
2944 | } | |
2945 | ||
2946 | /* change the enabled child controllers for a cgroup in the default hierarchy */ | |
451af504 TH |
2947 | static ssize_t cgroup_subtree_control_write(struct kernfs_open_file *of, |
2948 | char *buf, size_t nbytes, | |
2949 | loff_t off) | |
f8f22e53 | 2950 | { |
6e5c8307 TH |
2951 | u16 enable = 0, disable = 0; |
2952 | u16 css_enable, css_disable, old_sc, new_sc, old_ss, new_ss; | |
a9746d8d | 2953 | struct cgroup *cgrp, *child; |
f8f22e53 | 2954 | struct cgroup_subsys *ss; |
451af504 | 2955 | char *tok; |
f8f22e53 TH |
2956 | int ssid, ret; |
2957 | ||
2958 | /* | |
d37167ab TH |
2959 | * Parse input - space separated list of subsystem names prefixed |
2960 | * with either + or -. | |
f8f22e53 | 2961 | */ |
451af504 TH |
2962 | buf = strstrip(buf); |
2963 | while ((tok = strsep(&buf, " "))) { | |
d37167ab TH |
2964 | if (tok[0] == '\0') |
2965 | continue; | |
a7165264 | 2966 | do_each_subsys_mask(ss, ssid, ~cgrp_dfl_inhibit_ss_mask) { |
fc5ed1e9 TH |
2967 | if (!cgroup_ssid_enabled(ssid) || |
2968 | strcmp(tok + 1, ss->name)) | |
f8f22e53 TH |
2969 | continue; |
2970 | ||
2971 | if (*tok == '+') { | |
7d331fa9 TH |
2972 | enable |= 1 << ssid; |
2973 | disable &= ~(1 << ssid); | |
f8f22e53 | 2974 | } else if (*tok == '-') { |
7d331fa9 TH |
2975 | disable |= 1 << ssid; |
2976 | enable &= ~(1 << ssid); | |
f8f22e53 TH |
2977 | } else { |
2978 | return -EINVAL; | |
2979 | } | |
2980 | break; | |
b4e0eeaf | 2981 | } while_each_subsys_mask(); |
f8f22e53 TH |
2982 | if (ssid == CGROUP_SUBSYS_COUNT) |
2983 | return -EINVAL; | |
2984 | } | |
2985 | ||
a9746d8d TH |
2986 | cgrp = cgroup_kn_lock_live(of->kn); |
2987 | if (!cgrp) | |
2988 | return -ENODEV; | |
f8f22e53 TH |
2989 | |
2990 | for_each_subsys(ss, ssid) { | |
2991 | if (enable & (1 << ssid)) { | |
667c2491 | 2992 | if (cgrp->subtree_control & (1 << ssid)) { |
f8f22e53 TH |
2993 | enable &= ~(1 << ssid); |
2994 | continue; | |
2995 | } | |
2996 | ||
c29adf24 TH |
2997 | /* unavailable or not enabled on the parent? */ |
2998 | if (!(cgrp_dfl_root.subsys_mask & (1 << ssid)) || | |
2999 | (cgroup_parent(cgrp) && | |
667c2491 | 3000 | !(cgroup_parent(cgrp)->subtree_control & (1 << ssid)))) { |
c29adf24 TH |
3001 | ret = -ENOENT; |
3002 | goto out_unlock; | |
3003 | } | |
f8f22e53 | 3004 | } else if (disable & (1 << ssid)) { |
667c2491 | 3005 | if (!(cgrp->subtree_control & (1 << ssid))) { |
f8f22e53 TH |
3006 | disable &= ~(1 << ssid); |
3007 | continue; | |
3008 | } | |
3009 | ||
3010 | /* a child has it enabled? */ | |
3011 | cgroup_for_each_live_child(child, cgrp) { | |
667c2491 | 3012 | if (child->subtree_control & (1 << ssid)) { |
f8f22e53 | 3013 | ret = -EBUSY; |
ddab2b6e | 3014 | goto out_unlock; |
f8f22e53 TH |
3015 | } |
3016 | } | |
3017 | } | |
3018 | } | |
3019 | ||
3020 | if (!enable && !disable) { | |
3021 | ret = 0; | |
ddab2b6e | 3022 | goto out_unlock; |
f8f22e53 TH |
3023 | } |
3024 | ||
3025 | /* | |
667c2491 | 3026 | * Except for the root, subtree_control must be zero for a cgroup |
f8f22e53 TH |
3027 | * with tasks so that child cgroups don't compete against tasks. |
3028 | */ | |
d51f39b0 | 3029 | if (enable && cgroup_parent(cgrp) && !list_empty(&cgrp->cset_links)) { |
f8f22e53 TH |
3030 | ret = -EBUSY; |
3031 | goto out_unlock; | |
3032 | } | |
3033 | ||
3034 | /* | |
f63070d3 TH |
3035 | * Update subsys masks and calculate what needs to be done. More |
3036 | * subsystems than specified may need to be enabled or disabled | |
3037 | * depending on subsystem dependencies. | |
3038 | */ | |
755bf5ee | 3039 | old_sc = cgrp->subtree_control; |
8699b776 | 3040 | old_ss = cgrp->subtree_ss_mask; |
755bf5ee | 3041 | new_sc = (old_sc | enable) & ~disable; |
8699b776 | 3042 | new_ss = cgroup_calc_subtree_ss_mask(cgrp, new_sc); |
f63070d3 | 3043 | |
755bf5ee TH |
3044 | css_enable = ~old_ss & new_ss; |
3045 | css_disable = old_ss & ~new_ss; | |
f63070d3 TH |
3046 | enable |= css_enable; |
3047 | disable |= css_disable; | |
c29adf24 | 3048 | |
db6e3053 TH |
3049 | /* |
3050 | * Because css offlining is asynchronous, userland might try to | |
3051 | * re-enable the same controller while the previous instance is | |
3052 | * still around. In such cases, wait till it's gone using | |
3053 | * offline_waitq. | |
3054 | */ | |
b4e0eeaf | 3055 | do_each_subsys_mask(ss, ssid, css_enable) { |
db6e3053 TH |
3056 | cgroup_for_each_live_child(child, cgrp) { |
3057 | DEFINE_WAIT(wait); | |
3058 | ||
3059 | if (!cgroup_css(child, ss)) | |
3060 | continue; | |
3061 | ||
3062 | cgroup_get(child); | |
3063 | prepare_to_wait(&child->offline_waitq, &wait, | |
3064 | TASK_UNINTERRUPTIBLE); | |
3065 | cgroup_kn_unlock(of->kn); | |
3066 | schedule(); | |
3067 | finish_wait(&child->offline_waitq, &wait); | |
3068 | cgroup_put(child); | |
3069 | ||
3070 | return restart_syscall(); | |
3071 | } | |
b4e0eeaf | 3072 | } while_each_subsys_mask(); |
db6e3053 | 3073 | |
755bf5ee | 3074 | cgrp->subtree_control = new_sc; |
8699b776 | 3075 | cgrp->subtree_ss_mask = new_ss; |
755bf5ee | 3076 | |
f63070d3 TH |
3077 | /* |
3078 | * Create new csses or make the existing ones visible. A css is | |
3079 | * created invisible if it's being implicitly enabled through | |
3080 | * dependency. An invisible css is made visible when the userland | |
3081 | * explicitly enables it. | |
f8f22e53 | 3082 | */ |
996cd1fb | 3083 | do_each_subsys_mask(ss, ssid, enable) { |
f8f22e53 | 3084 | cgroup_for_each_live_child(child, cgrp) { |
6cd0f5bb TH |
3085 | if (css_enable & (1 << ssid)) { |
3086 | struct cgroup_subsys_state *css; | |
3087 | ||
3088 | css = css_create(child, ss); | |
3089 | if (IS_ERR(css)) { | |
3090 | ret = PTR_ERR(css); | |
3091 | goto err_undo_css; | |
3092 | } | |
3093 | ||
3094 | if (cgrp->subtree_control & (1 << ssid)) { | |
3095 | ret = css_populate_dir(css, NULL); | |
3096 | if (ret) | |
3097 | goto err_undo_css; | |
3098 | } | |
3099 | } else { | |
4df8dc90 TH |
3100 | ret = css_populate_dir(cgroup_css(child, ss), |
3101 | NULL); | |
6cd0f5bb TH |
3102 | if (ret) |
3103 | goto err_undo_css; | |
3104 | } | |
f8f22e53 | 3105 | } |
996cd1fb | 3106 | } while_each_subsys_mask(); |
f8f22e53 | 3107 | |
c29adf24 TH |
3108 | /* |
3109 | * At this point, cgroup_e_css() results reflect the new csses | |
3110 | * making the following cgroup_update_dfl_csses() properly update | |
3111 | * css associations of all tasks in the subtree. | |
3112 | */ | |
f8f22e53 TH |
3113 | ret = cgroup_update_dfl_csses(cgrp); |
3114 | if (ret) | |
3115 | goto err_undo_css; | |
3116 | ||
f63070d3 TH |
3117 | /* |
3118 | * All tasks are migrated out of disabled csses. Kill or hide | |
3119 | * them. A css is hidden when the userland requests it to be | |
b4536f0c TH |
3120 | * disabled while other subsystems are still depending on it. The |
3121 | * css must not actively control resources and be in the vanilla | |
3122 | * state if it's made visible again later. Controllers which may | |
3123 | * be depended upon should provide ->css_reset() for this purpose. | |
f63070d3 | 3124 | */ |
996cd1fb | 3125 | do_each_subsys_mask(ss, ssid, disable) { |
f63070d3 | 3126 | cgroup_for_each_live_child(child, cgrp) { |
b4536f0c TH |
3127 | struct cgroup_subsys_state *css = cgroup_css(child, ss); |
3128 | ||
3129 | if (css_disable & (1 << ssid)) { | |
3130 | kill_css(css); | |
3131 | } else { | |
4df8dc90 | 3132 | css_clear_dir(css, NULL); |
b4536f0c TH |
3133 | if (ss->css_reset) |
3134 | ss->css_reset(css); | |
3135 | } | |
f63070d3 | 3136 | } |
996cd1fb | 3137 | } while_each_subsys_mask(); |
f8f22e53 TH |
3138 | |
3139 | kernfs_activate(cgrp->kn); | |
3140 | ret = 0; | |
3141 | out_unlock: | |
a9746d8d | 3142 | cgroup_kn_unlock(of->kn); |
451af504 | 3143 | return ret ?: nbytes; |
f8f22e53 TH |
3144 | |
3145 | err_undo_css: | |
755bf5ee | 3146 | cgrp->subtree_control = old_sc; |
8699b776 | 3147 | cgrp->subtree_ss_mask = old_ss; |
f8f22e53 | 3148 | |
996cd1fb | 3149 | do_each_subsys_mask(ss, ssid, enable) { |
f8f22e53 TH |
3150 | cgroup_for_each_live_child(child, cgrp) { |
3151 | struct cgroup_subsys_state *css = cgroup_css(child, ss); | |
f63070d3 TH |
3152 | |
3153 | if (!css) | |
3154 | continue; | |
3155 | ||
3156 | if (css_enable & (1 << ssid)) | |
f8f22e53 | 3157 | kill_css(css); |
f63070d3 | 3158 | else |
4df8dc90 | 3159 | css_clear_dir(css, NULL); |
f8f22e53 | 3160 | } |
996cd1fb | 3161 | } while_each_subsys_mask(); |
f8f22e53 TH |
3162 | goto out_unlock; |
3163 | } | |
3164 | ||
4a07c222 | 3165 | static int cgroup_events_show(struct seq_file *seq, void *v) |
842b597e | 3166 | { |
4a07c222 | 3167 | seq_printf(seq, "populated %d\n", |
27bd4dbb | 3168 | cgroup_is_populated(seq_css(seq)->cgroup)); |
842b597e TH |
3169 | return 0; |
3170 | } | |
3171 | ||
2bd59d48 TH |
3172 | static ssize_t cgroup_file_write(struct kernfs_open_file *of, char *buf, |
3173 | size_t nbytes, loff_t off) | |
355e0c48 | 3174 | { |
2bd59d48 TH |
3175 | struct cgroup *cgrp = of->kn->parent->priv; |
3176 | struct cftype *cft = of->kn->priv; | |
3177 | struct cgroup_subsys_state *css; | |
a742c59d | 3178 | int ret; |
355e0c48 | 3179 | |
b4168640 TH |
3180 | if (cft->write) |
3181 | return cft->write(of, buf, nbytes, off); | |
3182 | ||
2bd59d48 TH |
3183 | /* |
3184 | * kernfs guarantees that a file isn't deleted with operations in | |
3185 | * flight, which means that the matching css is and stays alive and | |
3186 | * doesn't need to be pinned. The RCU locking is not necessary | |
3187 | * either. It's just for the convenience of using cgroup_css(). | |
3188 | */ | |
3189 | rcu_read_lock(); | |
3190 | css = cgroup_css(cgrp, cft->ss); | |
3191 | rcu_read_unlock(); | |
a742c59d | 3192 | |
451af504 | 3193 | if (cft->write_u64) { |
a742c59d TH |
3194 | unsigned long long v; |
3195 | ret = kstrtoull(buf, 0, &v); | |
3196 | if (!ret) | |
3197 | ret = cft->write_u64(css, cft, v); | |
3198 | } else if (cft->write_s64) { | |
3199 | long long v; | |
3200 | ret = kstrtoll(buf, 0, &v); | |
3201 | if (!ret) | |
3202 | ret = cft->write_s64(css, cft, v); | |
e73d2c61 | 3203 | } else { |
a742c59d | 3204 | ret = -EINVAL; |
e73d2c61 | 3205 | } |
2bd59d48 | 3206 | |
a742c59d | 3207 | return ret ?: nbytes; |
355e0c48 PM |
3208 | } |
3209 | ||
6612f05b | 3210 | static void *cgroup_seqfile_start(struct seq_file *seq, loff_t *ppos) |
db3b1497 | 3211 | { |
2bd59d48 | 3212 | return seq_cft(seq)->seq_start(seq, ppos); |
db3b1497 PM |
3213 | } |
3214 | ||
6612f05b | 3215 | static void *cgroup_seqfile_next(struct seq_file *seq, void *v, loff_t *ppos) |
ddbcc7e8 | 3216 | { |
2bd59d48 | 3217 | return seq_cft(seq)->seq_next(seq, v, ppos); |
ddbcc7e8 PM |
3218 | } |
3219 | ||
6612f05b | 3220 | static void cgroup_seqfile_stop(struct seq_file *seq, void *v) |
ddbcc7e8 | 3221 | { |
2bd59d48 | 3222 | seq_cft(seq)->seq_stop(seq, v); |
ddbcc7e8 PM |
3223 | } |
3224 | ||
91796569 | 3225 | static int cgroup_seqfile_show(struct seq_file *m, void *arg) |
e73d2c61 | 3226 | { |
7da11279 TH |
3227 | struct cftype *cft = seq_cft(m); |
3228 | struct cgroup_subsys_state *css = seq_css(m); | |
e73d2c61 | 3229 | |
2da8ca82 TH |
3230 | if (cft->seq_show) |
3231 | return cft->seq_show(m, arg); | |
e73d2c61 | 3232 | |
f4c753b7 | 3233 | if (cft->read_u64) |
896f5199 TH |
3234 | seq_printf(m, "%llu\n", cft->read_u64(css, cft)); |
3235 | else if (cft->read_s64) | |
3236 | seq_printf(m, "%lld\n", cft->read_s64(css, cft)); | |
3237 | else | |
3238 | return -EINVAL; | |
3239 | return 0; | |
91796569 PM |
3240 | } |
3241 | ||
2bd59d48 TH |
3242 | static struct kernfs_ops cgroup_kf_single_ops = { |
3243 | .atomic_write_len = PAGE_SIZE, | |
3244 | .write = cgroup_file_write, | |
3245 | .seq_show = cgroup_seqfile_show, | |
91796569 PM |
3246 | }; |
3247 | ||
2bd59d48 TH |
3248 | static struct kernfs_ops cgroup_kf_ops = { |
3249 | .atomic_write_len = PAGE_SIZE, | |
3250 | .write = cgroup_file_write, | |
3251 | .seq_start = cgroup_seqfile_start, | |
3252 | .seq_next = cgroup_seqfile_next, | |
3253 | .seq_stop = cgroup_seqfile_stop, | |
3254 | .seq_show = cgroup_seqfile_show, | |
3255 | }; | |
ddbcc7e8 PM |
3256 | |
3257 | /* | |
3258 | * cgroup_rename - Only allow simple rename of directories in place. | |
3259 | */ | |
2bd59d48 TH |
3260 | static int cgroup_rename(struct kernfs_node *kn, struct kernfs_node *new_parent, |
3261 | const char *new_name_str) | |
ddbcc7e8 | 3262 | { |
2bd59d48 | 3263 | struct cgroup *cgrp = kn->priv; |
65dff759 | 3264 | int ret; |
65dff759 | 3265 | |
2bd59d48 | 3266 | if (kernfs_type(kn) != KERNFS_DIR) |
ddbcc7e8 | 3267 | return -ENOTDIR; |
2bd59d48 | 3268 | if (kn->parent != new_parent) |
ddbcc7e8 | 3269 | return -EIO; |
65dff759 | 3270 | |
6db8e85c TH |
3271 | /* |
3272 | * This isn't a proper migration and its usefulness is very | |
aa6ec29b | 3273 | * limited. Disallow on the default hierarchy. |
6db8e85c | 3274 | */ |
aa6ec29b | 3275 | if (cgroup_on_dfl(cgrp)) |
6db8e85c | 3276 | return -EPERM; |
099fca32 | 3277 | |
e1b2dc17 | 3278 | /* |
8353da1f | 3279 | * We're gonna grab cgroup_mutex which nests outside kernfs |
e1b2dc17 | 3280 | * active_ref. kernfs_rename() doesn't require active_ref |
8353da1f | 3281 | * protection. Break them before grabbing cgroup_mutex. |
e1b2dc17 TH |
3282 | */ |
3283 | kernfs_break_active_protection(new_parent); | |
3284 | kernfs_break_active_protection(kn); | |
099fca32 | 3285 | |
2bd59d48 | 3286 | mutex_lock(&cgroup_mutex); |
099fca32 | 3287 | |
2bd59d48 | 3288 | ret = kernfs_rename(kn, new_parent, new_name_str); |
099fca32 | 3289 | |
2bd59d48 | 3290 | mutex_unlock(&cgroup_mutex); |
e1b2dc17 TH |
3291 | |
3292 | kernfs_unbreak_active_protection(kn); | |
3293 | kernfs_unbreak_active_protection(new_parent); | |
2bd59d48 | 3294 | return ret; |
099fca32 LZ |
3295 | } |
3296 | ||
49957f8e TH |
3297 | /* set uid and gid of cgroup dirs and files to that of the creator */ |
3298 | static int cgroup_kn_set_ugid(struct kernfs_node *kn) | |
3299 | { | |
3300 | struct iattr iattr = { .ia_valid = ATTR_UID | ATTR_GID, | |
3301 | .ia_uid = current_fsuid(), | |
3302 | .ia_gid = current_fsgid(), }; | |
3303 | ||
3304 | if (uid_eq(iattr.ia_uid, GLOBAL_ROOT_UID) && | |
3305 | gid_eq(iattr.ia_gid, GLOBAL_ROOT_GID)) | |
3306 | return 0; | |
3307 | ||
3308 | return kernfs_setattr(kn, &iattr); | |
3309 | } | |
3310 | ||
4df8dc90 TH |
3311 | static int cgroup_add_file(struct cgroup_subsys_state *css, struct cgroup *cgrp, |
3312 | struct cftype *cft) | |
ddbcc7e8 | 3313 | { |
8d7e6fb0 | 3314 | char name[CGROUP_FILE_NAME_MAX]; |
2bd59d48 TH |
3315 | struct kernfs_node *kn; |
3316 | struct lock_class_key *key = NULL; | |
49957f8e | 3317 | int ret; |
05ef1d7c | 3318 | |
2bd59d48 TH |
3319 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
3320 | key = &cft->lockdep_key; | |
3321 | #endif | |
3322 | kn = __kernfs_create_file(cgrp->kn, cgroup_file_name(cgrp, cft, name), | |
3323 | cgroup_file_mode(cft), 0, cft->kf_ops, cft, | |
dfeb0750 | 3324 | NULL, key); |
49957f8e TH |
3325 | if (IS_ERR(kn)) |
3326 | return PTR_ERR(kn); | |
3327 | ||
3328 | ret = cgroup_kn_set_ugid(kn); | |
f8f22e53 | 3329 | if (ret) { |
49957f8e | 3330 | kernfs_remove(kn); |
f8f22e53 TH |
3331 | return ret; |
3332 | } | |
3333 | ||
6f60eade TH |
3334 | if (cft->file_offset) { |
3335 | struct cgroup_file *cfile = (void *)css + cft->file_offset; | |
3336 | ||
34c06254 | 3337 | spin_lock_irq(&cgroup_file_kn_lock); |
6f60eade | 3338 | cfile->kn = kn; |
34c06254 | 3339 | spin_unlock_irq(&cgroup_file_kn_lock); |
6f60eade TH |
3340 | } |
3341 | ||
f8f22e53 | 3342 | return 0; |
ddbcc7e8 PM |
3343 | } |
3344 | ||
b1f28d31 TH |
3345 | /** |
3346 | * cgroup_addrm_files - add or remove files to a cgroup directory | |
4df8dc90 TH |
3347 | * @css: the target css |
3348 | * @cgrp: the target cgroup (usually css->cgroup) | |
b1f28d31 TH |
3349 | * @cfts: array of cftypes to be added |
3350 | * @is_add: whether to add or remove | |
3351 | * | |
3352 | * Depending on @is_add, add or remove files defined by @cfts on @cgrp. | |
6732ed85 | 3353 | * For removals, this function never fails. |
b1f28d31 | 3354 | */ |
4df8dc90 TH |
3355 | static int cgroup_addrm_files(struct cgroup_subsys_state *css, |
3356 | struct cgroup *cgrp, struct cftype cfts[], | |
2bb566cb | 3357 | bool is_add) |
ddbcc7e8 | 3358 | { |
6732ed85 | 3359 | struct cftype *cft, *cft_end = NULL; |
b598dde3 | 3360 | int ret = 0; |
b1f28d31 | 3361 | |
01f6474c | 3362 | lockdep_assert_held(&cgroup_mutex); |
db0416b6 | 3363 | |
6732ed85 TH |
3364 | restart: |
3365 | for (cft = cfts; cft != cft_end && cft->name[0] != '\0'; cft++) { | |
f33fddc2 | 3366 | /* does cft->flags tell us to skip this file on @cgrp? */ |
05ebb6e6 | 3367 | if ((cft->flags & __CFTYPE_ONLY_ON_DFL) && !cgroup_on_dfl(cgrp)) |
8cbbf2c9 | 3368 | continue; |
05ebb6e6 | 3369 | if ((cft->flags & __CFTYPE_NOT_ON_DFL) && cgroup_on_dfl(cgrp)) |
873fe09e | 3370 | continue; |
d51f39b0 | 3371 | if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgroup_parent(cgrp)) |
f33fddc2 | 3372 | continue; |
d51f39b0 | 3373 | if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgroup_parent(cgrp)) |
f33fddc2 G |
3374 | continue; |
3375 | ||
2739d3cc | 3376 | if (is_add) { |
4df8dc90 | 3377 | ret = cgroup_add_file(css, cgrp, cft); |
b1f28d31 | 3378 | if (ret) { |
ed3d261b JP |
3379 | pr_warn("%s: failed to add %s, err=%d\n", |
3380 | __func__, cft->name, ret); | |
6732ed85 TH |
3381 | cft_end = cft; |
3382 | is_add = false; | |
3383 | goto restart; | |
b1f28d31 | 3384 | } |
2739d3cc LZ |
3385 | } else { |
3386 | cgroup_rm_file(cgrp, cft); | |
db0416b6 | 3387 | } |
ddbcc7e8 | 3388 | } |
b598dde3 | 3389 | return ret; |
ddbcc7e8 PM |
3390 | } |
3391 | ||
21a2d343 | 3392 | static int cgroup_apply_cftypes(struct cftype *cfts, bool is_add) |
8e3f6541 TH |
3393 | { |
3394 | LIST_HEAD(pending); | |
2bb566cb | 3395 | struct cgroup_subsys *ss = cfts[0].ss; |
3dd06ffa | 3396 | struct cgroup *root = &ss->root->cgrp; |
492eb21b | 3397 | struct cgroup_subsys_state *css; |
9ccece80 | 3398 | int ret = 0; |
8e3f6541 | 3399 | |
01f6474c | 3400 | lockdep_assert_held(&cgroup_mutex); |
e8c82d20 | 3401 | |
e8c82d20 | 3402 | /* add/rm files for all cgroups created before */ |
ca8bdcaf | 3403 | css_for_each_descendant_pre(css, cgroup_css(root, ss)) { |
492eb21b TH |
3404 | struct cgroup *cgrp = css->cgroup; |
3405 | ||
e8c82d20 LZ |
3406 | if (cgroup_is_dead(cgrp)) |
3407 | continue; | |
3408 | ||
4df8dc90 | 3409 | ret = cgroup_addrm_files(css, cgrp, cfts, is_add); |
9ccece80 TH |
3410 | if (ret) |
3411 | break; | |
8e3f6541 | 3412 | } |
21a2d343 TH |
3413 | |
3414 | if (is_add && !ret) | |
3415 | kernfs_activate(root->kn); | |
9ccece80 | 3416 | return ret; |
8e3f6541 TH |
3417 | } |
3418 | ||
2da440a2 | 3419 | static void cgroup_exit_cftypes(struct cftype *cfts) |
8e3f6541 | 3420 | { |
2bb566cb | 3421 | struct cftype *cft; |
8e3f6541 | 3422 | |
2bd59d48 TH |
3423 | for (cft = cfts; cft->name[0] != '\0'; cft++) { |
3424 | /* free copy for custom atomic_write_len, see init_cftypes() */ | |
3425 | if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) | |
3426 | kfree(cft->kf_ops); | |
3427 | cft->kf_ops = NULL; | |
2da440a2 | 3428 | cft->ss = NULL; |
a8ddc821 TH |
3429 | |
3430 | /* revert flags set by cgroup core while adding @cfts */ | |
05ebb6e6 | 3431 | cft->flags &= ~(__CFTYPE_ONLY_ON_DFL | __CFTYPE_NOT_ON_DFL); |
2bd59d48 | 3432 | } |
2da440a2 TH |
3433 | } |
3434 | ||
2bd59d48 | 3435 | static int cgroup_init_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
2da440a2 TH |
3436 | { |
3437 | struct cftype *cft; | |
3438 | ||
2bd59d48 TH |
3439 | for (cft = cfts; cft->name[0] != '\0'; cft++) { |
3440 | struct kernfs_ops *kf_ops; | |
3441 | ||
0adb0704 TH |
3442 | WARN_ON(cft->ss || cft->kf_ops); |
3443 | ||
2bd59d48 TH |
3444 | if (cft->seq_start) |
3445 | kf_ops = &cgroup_kf_ops; | |
3446 | else | |
3447 | kf_ops = &cgroup_kf_single_ops; | |
3448 | ||
3449 | /* | |
3450 | * Ugh... if @cft wants a custom max_write_len, we need to | |
3451 | * make a copy of kf_ops to set its atomic_write_len. | |
3452 | */ | |
3453 | if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) { | |
3454 | kf_ops = kmemdup(kf_ops, sizeof(*kf_ops), GFP_KERNEL); | |
3455 | if (!kf_ops) { | |
3456 | cgroup_exit_cftypes(cfts); | |
3457 | return -ENOMEM; | |
3458 | } | |
3459 | kf_ops->atomic_write_len = cft->max_write_len; | |
3460 | } | |
8e3f6541 | 3461 | |
2bd59d48 | 3462 | cft->kf_ops = kf_ops; |
2bb566cb | 3463 | cft->ss = ss; |
2bd59d48 | 3464 | } |
2bb566cb | 3465 | |
2bd59d48 | 3466 | return 0; |
2da440a2 TH |
3467 | } |
3468 | ||
21a2d343 TH |
3469 | static int cgroup_rm_cftypes_locked(struct cftype *cfts) |
3470 | { | |
01f6474c | 3471 | lockdep_assert_held(&cgroup_mutex); |
21a2d343 TH |
3472 | |
3473 | if (!cfts || !cfts[0].ss) | |
3474 | return -ENOENT; | |
3475 | ||
3476 | list_del(&cfts->node); | |
3477 | cgroup_apply_cftypes(cfts, false); | |
3478 | cgroup_exit_cftypes(cfts); | |
3479 | return 0; | |
8e3f6541 | 3480 | } |
8e3f6541 | 3481 | |
79578621 TH |
3482 | /** |
3483 | * cgroup_rm_cftypes - remove an array of cftypes from a subsystem | |
79578621 TH |
3484 | * @cfts: zero-length name terminated array of cftypes |
3485 | * | |
2bb566cb TH |
3486 | * Unregister @cfts. Files described by @cfts are removed from all |
3487 | * existing cgroups and all future cgroups won't have them either. This | |
3488 | * function can be called anytime whether @cfts' subsys is attached or not. | |
79578621 TH |
3489 | * |
3490 | * Returns 0 on successful unregistration, -ENOENT if @cfts is not | |
2bb566cb | 3491 | * registered. |
79578621 | 3492 | */ |
2bb566cb | 3493 | int cgroup_rm_cftypes(struct cftype *cfts) |
79578621 | 3494 | { |
21a2d343 | 3495 | int ret; |
79578621 | 3496 | |
01f6474c | 3497 | mutex_lock(&cgroup_mutex); |
21a2d343 | 3498 | ret = cgroup_rm_cftypes_locked(cfts); |
01f6474c | 3499 | mutex_unlock(&cgroup_mutex); |
21a2d343 | 3500 | return ret; |
80b13586 TH |
3501 | } |
3502 | ||
8e3f6541 TH |
3503 | /** |
3504 | * cgroup_add_cftypes - add an array of cftypes to a subsystem | |
3505 | * @ss: target cgroup subsystem | |
3506 | * @cfts: zero-length name terminated array of cftypes | |
3507 | * | |
3508 | * Register @cfts to @ss. Files described by @cfts are created for all | |
3509 | * existing cgroups to which @ss is attached and all future cgroups will | |
3510 | * have them too. This function can be called anytime whether @ss is | |
3511 | * attached or not. | |
3512 | * | |
3513 | * Returns 0 on successful registration, -errno on failure. Note that this | |
3514 | * function currently returns 0 as long as @cfts registration is successful | |
3515 | * even if some file creation attempts on existing cgroups fail. | |
3516 | */ | |
2cf669a5 | 3517 | static int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
8e3f6541 | 3518 | { |
9ccece80 | 3519 | int ret; |
8e3f6541 | 3520 | |
fc5ed1e9 | 3521 | if (!cgroup_ssid_enabled(ss->id)) |
c731ae1d LZ |
3522 | return 0; |
3523 | ||
dc5736ed LZ |
3524 | if (!cfts || cfts[0].name[0] == '\0') |
3525 | return 0; | |
2bb566cb | 3526 | |
2bd59d48 TH |
3527 | ret = cgroup_init_cftypes(ss, cfts); |
3528 | if (ret) | |
3529 | return ret; | |
79578621 | 3530 | |
01f6474c | 3531 | mutex_lock(&cgroup_mutex); |
21a2d343 | 3532 | |
0adb0704 | 3533 | list_add_tail(&cfts->node, &ss->cfts); |
21a2d343 | 3534 | ret = cgroup_apply_cftypes(cfts, true); |
9ccece80 | 3535 | if (ret) |
21a2d343 | 3536 | cgroup_rm_cftypes_locked(cfts); |
79578621 | 3537 | |
01f6474c | 3538 | mutex_unlock(&cgroup_mutex); |
9ccece80 | 3539 | return ret; |
79578621 TH |
3540 | } |
3541 | ||
a8ddc821 TH |
3542 | /** |
3543 | * cgroup_add_dfl_cftypes - add an array of cftypes for default hierarchy | |
3544 | * @ss: target cgroup subsystem | |
3545 | * @cfts: zero-length name terminated array of cftypes | |
3546 | * | |
3547 | * Similar to cgroup_add_cftypes() but the added files are only used for | |
3548 | * the default hierarchy. | |
3549 | */ | |
3550 | int cgroup_add_dfl_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) | |
3551 | { | |
3552 | struct cftype *cft; | |
3553 | ||
3554 | for (cft = cfts; cft && cft->name[0] != '\0'; cft++) | |
05ebb6e6 | 3555 | cft->flags |= __CFTYPE_ONLY_ON_DFL; |
a8ddc821 TH |
3556 | return cgroup_add_cftypes(ss, cfts); |
3557 | } | |
3558 | ||
3559 | /** | |
3560 | * cgroup_add_legacy_cftypes - add an array of cftypes for legacy hierarchies | |
3561 | * @ss: target cgroup subsystem | |
3562 | * @cfts: zero-length name terminated array of cftypes | |
3563 | * | |
3564 | * Similar to cgroup_add_cftypes() but the added files are only used for | |
3565 | * the legacy hierarchies. | |
3566 | */ | |
2cf669a5 TH |
3567 | int cgroup_add_legacy_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
3568 | { | |
a8ddc821 TH |
3569 | struct cftype *cft; |
3570 | ||
e4b7037c TH |
3571 | for (cft = cfts; cft && cft->name[0] != '\0'; cft++) |
3572 | cft->flags |= __CFTYPE_NOT_ON_DFL; | |
2cf669a5 TH |
3573 | return cgroup_add_cftypes(ss, cfts); |
3574 | } | |
3575 | ||
34c06254 TH |
3576 | /** |
3577 | * cgroup_file_notify - generate a file modified event for a cgroup_file | |
3578 | * @cfile: target cgroup_file | |
3579 | * | |
3580 | * @cfile must have been obtained by setting cftype->file_offset. | |
3581 | */ | |
3582 | void cgroup_file_notify(struct cgroup_file *cfile) | |
3583 | { | |
3584 | unsigned long flags; | |
3585 | ||
3586 | spin_lock_irqsave(&cgroup_file_kn_lock, flags); | |
3587 | if (cfile->kn) | |
3588 | kernfs_notify(cfile->kn); | |
3589 | spin_unlock_irqrestore(&cgroup_file_kn_lock, flags); | |
3590 | } | |
3591 | ||
a043e3b2 LZ |
3592 | /** |
3593 | * cgroup_task_count - count the number of tasks in a cgroup. | |
3594 | * @cgrp: the cgroup in question | |
3595 | * | |
3596 | * Return the number of tasks in the cgroup. | |
3597 | */ | |
07bc356e | 3598 | static int cgroup_task_count(const struct cgroup *cgrp) |
bbcb81d0 PM |
3599 | { |
3600 | int count = 0; | |
69d0206c | 3601 | struct cgrp_cset_link *link; |
817929ec | 3602 | |
f0d9a5f1 | 3603 | spin_lock_bh(&css_set_lock); |
69d0206c TH |
3604 | list_for_each_entry(link, &cgrp->cset_links, cset_link) |
3605 | count += atomic_read(&link->cset->refcount); | |
f0d9a5f1 | 3606 | spin_unlock_bh(&css_set_lock); |
bbcb81d0 PM |
3607 | return count; |
3608 | } | |
3609 | ||
53fa5261 | 3610 | /** |
492eb21b | 3611 | * css_next_child - find the next child of a given css |
c2931b70 TH |
3612 | * @pos: the current position (%NULL to initiate traversal) |
3613 | * @parent: css whose children to walk | |
53fa5261 | 3614 | * |
c2931b70 | 3615 | * This function returns the next child of @parent and should be called |
87fb54f1 | 3616 | * under either cgroup_mutex or RCU read lock. The only requirement is |
c2931b70 TH |
3617 | * that @parent and @pos are accessible. The next sibling is guaranteed to |
3618 | * be returned regardless of their states. | |
3619 | * | |
3620 | * If a subsystem synchronizes ->css_online() and the start of iteration, a | |
3621 | * css which finished ->css_online() is guaranteed to be visible in the | |
3622 | * future iterations and will stay visible until the last reference is put. | |
3623 | * A css which hasn't finished ->css_online() or already finished | |
3624 | * ->css_offline() may show up during traversal. It's each subsystem's | |
3625 | * responsibility to synchronize against on/offlining. | |
53fa5261 | 3626 | */ |
c2931b70 TH |
3627 | struct cgroup_subsys_state *css_next_child(struct cgroup_subsys_state *pos, |
3628 | struct cgroup_subsys_state *parent) | |
53fa5261 | 3629 | { |
c2931b70 | 3630 | struct cgroup_subsys_state *next; |
53fa5261 | 3631 | |
8353da1f | 3632 | cgroup_assert_mutex_or_rcu_locked(); |
53fa5261 TH |
3633 | |
3634 | /* | |
de3f0341 TH |
3635 | * @pos could already have been unlinked from the sibling list. |
3636 | * Once a cgroup is removed, its ->sibling.next is no longer | |
3637 | * updated when its next sibling changes. CSS_RELEASED is set when | |
3638 | * @pos is taken off list, at which time its next pointer is valid, | |
3639 | * and, as releases are serialized, the one pointed to by the next | |
3640 | * pointer is guaranteed to not have started release yet. This | |
3641 | * implies that if we observe !CSS_RELEASED on @pos in this RCU | |
3642 | * critical section, the one pointed to by its next pointer is | |
3643 | * guaranteed to not have finished its RCU grace period even if we | |
3644 | * have dropped rcu_read_lock() inbetween iterations. | |
3b287a50 | 3645 | * |
de3f0341 TH |
3646 | * If @pos has CSS_RELEASED set, its next pointer can't be |
3647 | * dereferenced; however, as each css is given a monotonically | |
3648 | * increasing unique serial number and always appended to the | |
3649 | * sibling list, the next one can be found by walking the parent's | |
3650 | * children until the first css with higher serial number than | |
3651 | * @pos's. While this path can be slower, it happens iff iteration | |
3652 | * races against release and the race window is very small. | |
53fa5261 | 3653 | */ |
3b287a50 | 3654 | if (!pos) { |
c2931b70 TH |
3655 | next = list_entry_rcu(parent->children.next, struct cgroup_subsys_state, sibling); |
3656 | } else if (likely(!(pos->flags & CSS_RELEASED))) { | |
3657 | next = list_entry_rcu(pos->sibling.next, struct cgroup_subsys_state, sibling); | |
3b287a50 | 3658 | } else { |
c2931b70 | 3659 | list_for_each_entry_rcu(next, &parent->children, sibling) |
3b287a50 TH |
3660 | if (next->serial_nr > pos->serial_nr) |
3661 | break; | |
53fa5261 TH |
3662 | } |
3663 | ||
3b281afb TH |
3664 | /* |
3665 | * @next, if not pointing to the head, can be dereferenced and is | |
c2931b70 | 3666 | * the next sibling. |
3b281afb | 3667 | */ |
c2931b70 TH |
3668 | if (&next->sibling != &parent->children) |
3669 | return next; | |
3b281afb | 3670 | return NULL; |
53fa5261 | 3671 | } |
53fa5261 | 3672 | |
574bd9f7 | 3673 | /** |
492eb21b | 3674 | * css_next_descendant_pre - find the next descendant for pre-order walk |
574bd9f7 | 3675 | * @pos: the current position (%NULL to initiate traversal) |
492eb21b | 3676 | * @root: css whose descendants to walk |
574bd9f7 | 3677 | * |
492eb21b | 3678 | * To be used by css_for_each_descendant_pre(). Find the next descendant |
bd8815a6 TH |
3679 | * to visit for pre-order traversal of @root's descendants. @root is |
3680 | * included in the iteration and the first node to be visited. | |
75501a6d | 3681 | * |
87fb54f1 TH |
3682 | * While this function requires cgroup_mutex or RCU read locking, it |
3683 | * doesn't require the whole traversal to be contained in a single critical | |
3684 | * section. This function will return the correct next descendant as long | |
3685 | * as both @pos and @root are accessible and @pos is a descendant of @root. | |
c2931b70 TH |
3686 | * |
3687 | * If a subsystem synchronizes ->css_online() and the start of iteration, a | |
3688 | * css which finished ->css_online() is guaranteed to be visible in the | |
3689 | * future iterations and will stay visible until the last reference is put. | |
3690 | * A css which hasn't finished ->css_online() or already finished | |
3691 | * ->css_offline() may show up during traversal. It's each subsystem's | |
3692 | * responsibility to synchronize against on/offlining. | |
574bd9f7 | 3693 | */ |
492eb21b TH |
3694 | struct cgroup_subsys_state * |
3695 | css_next_descendant_pre(struct cgroup_subsys_state *pos, | |
3696 | struct cgroup_subsys_state *root) | |
574bd9f7 | 3697 | { |
492eb21b | 3698 | struct cgroup_subsys_state *next; |
574bd9f7 | 3699 | |
8353da1f | 3700 | cgroup_assert_mutex_or_rcu_locked(); |
574bd9f7 | 3701 | |
bd8815a6 | 3702 | /* if first iteration, visit @root */ |
7805d000 | 3703 | if (!pos) |
bd8815a6 | 3704 | return root; |
574bd9f7 TH |
3705 | |
3706 | /* visit the first child if exists */ | |
492eb21b | 3707 | next = css_next_child(NULL, pos); |
574bd9f7 TH |
3708 | if (next) |
3709 | return next; | |
3710 | ||
3711 | /* no child, visit my or the closest ancestor's next sibling */ | |
492eb21b | 3712 | while (pos != root) { |
5c9d535b | 3713 | next = css_next_child(pos, pos->parent); |
75501a6d | 3714 | if (next) |
574bd9f7 | 3715 | return next; |
5c9d535b | 3716 | pos = pos->parent; |
7805d000 | 3717 | } |
574bd9f7 TH |
3718 | |
3719 | return NULL; | |
3720 | } | |
574bd9f7 | 3721 | |
12a9d2fe | 3722 | /** |
492eb21b TH |
3723 | * css_rightmost_descendant - return the rightmost descendant of a css |
3724 | * @pos: css of interest | |
12a9d2fe | 3725 | * |
492eb21b TH |
3726 | * Return the rightmost descendant of @pos. If there's no descendant, @pos |
3727 | * is returned. This can be used during pre-order traversal to skip | |
12a9d2fe | 3728 | * subtree of @pos. |
75501a6d | 3729 | * |
87fb54f1 TH |
3730 | * While this function requires cgroup_mutex or RCU read locking, it |
3731 | * doesn't require the whole traversal to be contained in a single critical | |
3732 | * section. This function will return the correct rightmost descendant as | |
3733 | * long as @pos is accessible. | |
12a9d2fe | 3734 | */ |
492eb21b TH |
3735 | struct cgroup_subsys_state * |
3736 | css_rightmost_descendant(struct cgroup_subsys_state *pos) | |
12a9d2fe | 3737 | { |
492eb21b | 3738 | struct cgroup_subsys_state *last, *tmp; |
12a9d2fe | 3739 | |
8353da1f | 3740 | cgroup_assert_mutex_or_rcu_locked(); |
12a9d2fe TH |
3741 | |
3742 | do { | |
3743 | last = pos; | |
3744 | /* ->prev isn't RCU safe, walk ->next till the end */ | |
3745 | pos = NULL; | |
492eb21b | 3746 | css_for_each_child(tmp, last) |
12a9d2fe TH |
3747 | pos = tmp; |
3748 | } while (pos); | |
3749 | ||
3750 | return last; | |
3751 | } | |
12a9d2fe | 3752 | |
492eb21b TH |
3753 | static struct cgroup_subsys_state * |
3754 | css_leftmost_descendant(struct cgroup_subsys_state *pos) | |
574bd9f7 | 3755 | { |
492eb21b | 3756 | struct cgroup_subsys_state *last; |
574bd9f7 TH |
3757 | |
3758 | do { | |
3759 | last = pos; | |
492eb21b | 3760 | pos = css_next_child(NULL, pos); |
574bd9f7 TH |
3761 | } while (pos); |
3762 | ||
3763 | return last; | |
3764 | } | |
3765 | ||
3766 | /** | |
492eb21b | 3767 | * css_next_descendant_post - find the next descendant for post-order walk |
574bd9f7 | 3768 | * @pos: the current position (%NULL to initiate traversal) |
492eb21b | 3769 | * @root: css whose descendants to walk |
574bd9f7 | 3770 | * |
492eb21b | 3771 | * To be used by css_for_each_descendant_post(). Find the next descendant |
bd8815a6 TH |
3772 | * to visit for post-order traversal of @root's descendants. @root is |
3773 | * included in the iteration and the last node to be visited. | |
75501a6d | 3774 | * |
87fb54f1 TH |
3775 | * While this function requires cgroup_mutex or RCU read locking, it |
3776 | * doesn't require the whole traversal to be contained in a single critical | |
3777 | * section. This function will return the correct next descendant as long | |
3778 | * as both @pos and @cgroup are accessible and @pos is a descendant of | |
3779 | * @cgroup. | |
c2931b70 TH |
3780 | * |
3781 | * If a subsystem synchronizes ->css_online() and the start of iteration, a | |
3782 | * css which finished ->css_online() is guaranteed to be visible in the | |
3783 | * future iterations and will stay visible until the last reference is put. | |
3784 | * A css which hasn't finished ->css_online() or already finished | |
3785 | * ->css_offline() may show up during traversal. It's each subsystem's | |
3786 | * responsibility to synchronize against on/offlining. | |
574bd9f7 | 3787 | */ |
492eb21b TH |
3788 | struct cgroup_subsys_state * |
3789 | css_next_descendant_post(struct cgroup_subsys_state *pos, | |
3790 | struct cgroup_subsys_state *root) | |
574bd9f7 | 3791 | { |
492eb21b | 3792 | struct cgroup_subsys_state *next; |
574bd9f7 | 3793 | |
8353da1f | 3794 | cgroup_assert_mutex_or_rcu_locked(); |
574bd9f7 | 3795 | |
58b79a91 TH |
3796 | /* if first iteration, visit leftmost descendant which may be @root */ |
3797 | if (!pos) | |
3798 | return css_leftmost_descendant(root); | |
574bd9f7 | 3799 | |
bd8815a6 TH |
3800 | /* if we visited @root, we're done */ |
3801 | if (pos == root) | |
3802 | return NULL; | |
3803 | ||
574bd9f7 | 3804 | /* if there's an unvisited sibling, visit its leftmost descendant */ |
5c9d535b | 3805 | next = css_next_child(pos, pos->parent); |
75501a6d | 3806 | if (next) |
492eb21b | 3807 | return css_leftmost_descendant(next); |
574bd9f7 TH |
3808 | |
3809 | /* no sibling left, visit parent */ | |
5c9d535b | 3810 | return pos->parent; |
574bd9f7 | 3811 | } |
574bd9f7 | 3812 | |
f3d46500 TH |
3813 | /** |
3814 | * css_has_online_children - does a css have online children | |
3815 | * @css: the target css | |
3816 | * | |
3817 | * Returns %true if @css has any online children; otherwise, %false. This | |
3818 | * function can be called from any context but the caller is responsible | |
3819 | * for synchronizing against on/offlining as necessary. | |
3820 | */ | |
3821 | bool css_has_online_children(struct cgroup_subsys_state *css) | |
cbc125ef | 3822 | { |
f3d46500 TH |
3823 | struct cgroup_subsys_state *child; |
3824 | bool ret = false; | |
cbc125ef TH |
3825 | |
3826 | rcu_read_lock(); | |
f3d46500 | 3827 | css_for_each_child(child, css) { |
99bae5f9 | 3828 | if (child->flags & CSS_ONLINE) { |
f3d46500 TH |
3829 | ret = true; |
3830 | break; | |
cbc125ef TH |
3831 | } |
3832 | } | |
3833 | rcu_read_unlock(); | |
f3d46500 | 3834 | return ret; |
574bd9f7 | 3835 | } |
574bd9f7 | 3836 | |
0942eeee | 3837 | /** |
ecb9d535 | 3838 | * css_task_iter_advance_css_set - advance a task itererator to the next css_set |
0942eeee TH |
3839 | * @it: the iterator to advance |
3840 | * | |
3841 | * Advance @it to the next css_set to walk. | |
d515876e | 3842 | */ |
ecb9d535 | 3843 | static void css_task_iter_advance_css_set(struct css_task_iter *it) |
d515876e | 3844 | { |
0f0a2b4f | 3845 | struct list_head *l = it->cset_pos; |
d515876e TH |
3846 | struct cgrp_cset_link *link; |
3847 | struct css_set *cset; | |
3848 | ||
f0d9a5f1 | 3849 | lockdep_assert_held(&css_set_lock); |
ed27b9f7 | 3850 | |
d515876e TH |
3851 | /* Advance to the next non-empty css_set */ |
3852 | do { | |
3853 | l = l->next; | |
0f0a2b4f TH |
3854 | if (l == it->cset_head) { |
3855 | it->cset_pos = NULL; | |
ecb9d535 | 3856 | it->task_pos = NULL; |
d515876e TH |
3857 | return; |
3858 | } | |
3ebb2b6e TH |
3859 | |
3860 | if (it->ss) { | |
3861 | cset = container_of(l, struct css_set, | |
3862 | e_cset_node[it->ss->id]); | |
3863 | } else { | |
3864 | link = list_entry(l, struct cgrp_cset_link, cset_link); | |
3865 | cset = link->cset; | |
3866 | } | |
0de0942d | 3867 | } while (!css_set_populated(cset)); |
c7561128 | 3868 | |
0f0a2b4f | 3869 | it->cset_pos = l; |
c7561128 TH |
3870 | |
3871 | if (!list_empty(&cset->tasks)) | |
0f0a2b4f | 3872 | it->task_pos = cset->tasks.next; |
c7561128 | 3873 | else |
0f0a2b4f TH |
3874 | it->task_pos = cset->mg_tasks.next; |
3875 | ||
3876 | it->tasks_head = &cset->tasks; | |
3877 | it->mg_tasks_head = &cset->mg_tasks; | |
ed27b9f7 TH |
3878 | |
3879 | /* | |
3880 | * We don't keep css_sets locked across iteration steps and thus | |
3881 | * need to take steps to ensure that iteration can be resumed after | |
3882 | * the lock is re-acquired. Iteration is performed at two levels - | |
3883 | * css_sets and tasks in them. | |
3884 | * | |
3885 | * Once created, a css_set never leaves its cgroup lists, so a | |
3886 | * pinned css_set is guaranteed to stay put and we can resume | |
3887 | * iteration afterwards. | |
3888 | * | |
3889 | * Tasks may leave @cset across iteration steps. This is resolved | |
3890 | * by registering each iterator with the css_set currently being | |
3891 | * walked and making css_set_move_task() advance iterators whose | |
3892 | * next task is leaving. | |
3893 | */ | |
3894 | if (it->cur_cset) { | |
3895 | list_del(&it->iters_node); | |
3896 | put_css_set_locked(it->cur_cset); | |
3897 | } | |
3898 | get_css_set(cset); | |
3899 | it->cur_cset = cset; | |
3900 | list_add(&it->iters_node, &cset->task_iters); | |
d515876e TH |
3901 | } |
3902 | ||
ecb9d535 TH |
3903 | static void css_task_iter_advance(struct css_task_iter *it) |
3904 | { | |
3905 | struct list_head *l = it->task_pos; | |
3906 | ||
f0d9a5f1 | 3907 | lockdep_assert_held(&css_set_lock); |
ecb9d535 TH |
3908 | WARN_ON_ONCE(!l); |
3909 | ||
3910 | /* | |
3911 | * Advance iterator to find next entry. cset->tasks is consumed | |
3912 | * first and then ->mg_tasks. After ->mg_tasks, we move onto the | |
3913 | * next cset. | |
3914 | */ | |
3915 | l = l->next; | |
3916 | ||
3917 | if (l == it->tasks_head) | |
3918 | l = it->mg_tasks_head->next; | |
3919 | ||
3920 | if (l == it->mg_tasks_head) | |
3921 | css_task_iter_advance_css_set(it); | |
3922 | else | |
3923 | it->task_pos = l; | |
3924 | } | |
3925 | ||
0942eeee | 3926 | /** |
72ec7029 TH |
3927 | * css_task_iter_start - initiate task iteration |
3928 | * @css: the css to walk tasks of | |
0942eeee TH |
3929 | * @it: the task iterator to use |
3930 | * | |
72ec7029 TH |
3931 | * Initiate iteration through the tasks of @css. The caller can call |
3932 | * css_task_iter_next() to walk through the tasks until the function | |
3933 | * returns NULL. On completion of iteration, css_task_iter_end() must be | |
3934 | * called. | |
0942eeee | 3935 | */ |
72ec7029 TH |
3936 | void css_task_iter_start(struct cgroup_subsys_state *css, |
3937 | struct css_task_iter *it) | |
817929ec | 3938 | { |
56fde9e0 TH |
3939 | /* no one should try to iterate before mounting cgroups */ |
3940 | WARN_ON_ONCE(!use_task_css_set_links); | |
31a7df01 | 3941 | |
ed27b9f7 TH |
3942 | memset(it, 0, sizeof(*it)); |
3943 | ||
f0d9a5f1 | 3944 | spin_lock_bh(&css_set_lock); |
c59cd3d8 | 3945 | |
3ebb2b6e TH |
3946 | it->ss = css->ss; |
3947 | ||
3948 | if (it->ss) | |
3949 | it->cset_pos = &css->cgroup->e_csets[css->ss->id]; | |
3950 | else | |
3951 | it->cset_pos = &css->cgroup->cset_links; | |
3952 | ||
0f0a2b4f | 3953 | it->cset_head = it->cset_pos; |
c59cd3d8 | 3954 | |
ecb9d535 | 3955 | css_task_iter_advance_css_set(it); |
ed27b9f7 | 3956 | |
f0d9a5f1 | 3957 | spin_unlock_bh(&css_set_lock); |
817929ec PM |
3958 | } |
3959 | ||
0942eeee | 3960 | /** |
72ec7029 | 3961 | * css_task_iter_next - return the next task for the iterator |
0942eeee TH |
3962 | * @it: the task iterator being iterated |
3963 | * | |
3964 | * The "next" function for task iteration. @it should have been | |
72ec7029 TH |
3965 | * initialized via css_task_iter_start(). Returns NULL when the iteration |
3966 | * reaches the end. | |
0942eeee | 3967 | */ |
72ec7029 | 3968 | struct task_struct *css_task_iter_next(struct css_task_iter *it) |
817929ec | 3969 | { |
d5745675 | 3970 | if (it->cur_task) { |
ed27b9f7 | 3971 | put_task_struct(it->cur_task); |
d5745675 TH |
3972 | it->cur_task = NULL; |
3973 | } | |
ed27b9f7 | 3974 | |
f0d9a5f1 | 3975 | spin_lock_bh(&css_set_lock); |
ed27b9f7 | 3976 | |
d5745675 TH |
3977 | if (it->task_pos) { |
3978 | it->cur_task = list_entry(it->task_pos, struct task_struct, | |
3979 | cg_list); | |
3980 | get_task_struct(it->cur_task); | |
3981 | css_task_iter_advance(it); | |
3982 | } | |
ed27b9f7 | 3983 | |
f0d9a5f1 | 3984 | spin_unlock_bh(&css_set_lock); |
ed27b9f7 TH |
3985 | |
3986 | return it->cur_task; | |
817929ec PM |
3987 | } |
3988 | ||
0942eeee | 3989 | /** |
72ec7029 | 3990 | * css_task_iter_end - finish task iteration |
0942eeee TH |
3991 | * @it: the task iterator to finish |
3992 | * | |
72ec7029 | 3993 | * Finish task iteration started by css_task_iter_start(). |
0942eeee | 3994 | */ |
72ec7029 | 3995 | void css_task_iter_end(struct css_task_iter *it) |
31a7df01 | 3996 | { |
ed27b9f7 | 3997 | if (it->cur_cset) { |
f0d9a5f1 | 3998 | spin_lock_bh(&css_set_lock); |
ed27b9f7 TH |
3999 | list_del(&it->iters_node); |
4000 | put_css_set_locked(it->cur_cset); | |
f0d9a5f1 | 4001 | spin_unlock_bh(&css_set_lock); |
ed27b9f7 TH |
4002 | } |
4003 | ||
4004 | if (it->cur_task) | |
4005 | put_task_struct(it->cur_task); | |
31a7df01 CW |
4006 | } |
4007 | ||
4008 | /** | |
8cc99345 TH |
4009 | * cgroup_trasnsfer_tasks - move tasks from one cgroup to another |
4010 | * @to: cgroup to which the tasks will be moved | |
4011 | * @from: cgroup in which the tasks currently reside | |
31a7df01 | 4012 | * |
eaf797ab TH |
4013 | * Locking rules between cgroup_post_fork() and the migration path |
4014 | * guarantee that, if a task is forking while being migrated, the new child | |
4015 | * is guaranteed to be either visible in the source cgroup after the | |
4016 | * parent's migration is complete or put into the target cgroup. No task | |
4017 | * can slip out of migration through forking. | |
31a7df01 | 4018 | */ |
8cc99345 | 4019 | int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from) |
31a7df01 | 4020 | { |
952aaa12 TH |
4021 | LIST_HEAD(preloaded_csets); |
4022 | struct cgrp_cset_link *link; | |
72ec7029 | 4023 | struct css_task_iter it; |
e406d1cf | 4024 | struct task_struct *task; |
952aaa12 | 4025 | int ret; |
31a7df01 | 4026 | |
952aaa12 | 4027 | mutex_lock(&cgroup_mutex); |
31a7df01 | 4028 | |
952aaa12 | 4029 | /* all tasks in @from are being moved, all csets are source */ |
f0d9a5f1 | 4030 | spin_lock_bh(&css_set_lock); |
952aaa12 TH |
4031 | list_for_each_entry(link, &from->cset_links, cset_link) |
4032 | cgroup_migrate_add_src(link->cset, to, &preloaded_csets); | |
f0d9a5f1 | 4033 | spin_unlock_bh(&css_set_lock); |
31a7df01 | 4034 | |
952aaa12 TH |
4035 | ret = cgroup_migrate_prepare_dst(to, &preloaded_csets); |
4036 | if (ret) | |
4037 | goto out_err; | |
8cc99345 | 4038 | |
952aaa12 | 4039 | /* |
2cfa2b19 | 4040 | * Migrate tasks one-by-one until @from is empty. This fails iff |
952aaa12 TH |
4041 | * ->can_attach() fails. |
4042 | */ | |
e406d1cf | 4043 | do { |
9d800df1 | 4044 | css_task_iter_start(&from->self, &it); |
e406d1cf TH |
4045 | task = css_task_iter_next(&it); |
4046 | if (task) | |
4047 | get_task_struct(task); | |
4048 | css_task_iter_end(&it); | |
4049 | ||
4050 | if (task) { | |
9af2ec45 | 4051 | ret = cgroup_migrate(task, false, to); |
e406d1cf TH |
4052 | put_task_struct(task); |
4053 | } | |
4054 | } while (task && !ret); | |
952aaa12 TH |
4055 | out_err: |
4056 | cgroup_migrate_finish(&preloaded_csets); | |
47cfcd09 | 4057 | mutex_unlock(&cgroup_mutex); |
e406d1cf | 4058 | return ret; |
8cc99345 TH |
4059 | } |
4060 | ||
bbcb81d0 | 4061 | /* |
102a775e | 4062 | * Stuff for reading the 'tasks'/'procs' files. |
bbcb81d0 PM |
4063 | * |
4064 | * Reading this file can return large amounts of data if a cgroup has | |
4065 | * *lots* of attached tasks. So it may need several calls to read(), | |
4066 | * but we cannot guarantee that the information we produce is correct | |
4067 | * unless we produce it entirely atomically. | |
4068 | * | |
bbcb81d0 | 4069 | */ |
bbcb81d0 | 4070 | |
24528255 LZ |
4071 | /* which pidlist file are we talking about? */ |
4072 | enum cgroup_filetype { | |
4073 | CGROUP_FILE_PROCS, | |
4074 | CGROUP_FILE_TASKS, | |
4075 | }; | |
4076 | ||
4077 | /* | |
4078 | * A pidlist is a list of pids that virtually represents the contents of one | |
4079 | * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists, | |
4080 | * a pair (one each for procs, tasks) for each pid namespace that's relevant | |
4081 | * to the cgroup. | |
4082 | */ | |
4083 | struct cgroup_pidlist { | |
4084 | /* | |
4085 | * used to find which pidlist is wanted. doesn't change as long as | |
4086 | * this particular list stays in the list. | |
4087 | */ | |
4088 | struct { enum cgroup_filetype type; struct pid_namespace *ns; } key; | |
4089 | /* array of xids */ | |
4090 | pid_t *list; | |
4091 | /* how many elements the above list has */ | |
4092 | int length; | |
24528255 LZ |
4093 | /* each of these stored in a list by its cgroup */ |
4094 | struct list_head links; | |
4095 | /* pointer to the cgroup we belong to, for list removal purposes */ | |
4096 | struct cgroup *owner; | |
b1a21367 TH |
4097 | /* for delayed destruction */ |
4098 | struct delayed_work destroy_dwork; | |
24528255 LZ |
4099 | }; |
4100 | ||
d1d9fd33 BB |
4101 | /* |
4102 | * The following two functions "fix" the issue where there are more pids | |
4103 | * than kmalloc will give memory for; in such cases, we use vmalloc/vfree. | |
4104 | * TODO: replace with a kernel-wide solution to this problem | |
4105 | */ | |
4106 | #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2)) | |
4107 | static void *pidlist_allocate(int count) | |
4108 | { | |
4109 | if (PIDLIST_TOO_LARGE(count)) | |
4110 | return vmalloc(count * sizeof(pid_t)); | |
4111 | else | |
4112 | return kmalloc(count * sizeof(pid_t), GFP_KERNEL); | |
4113 | } | |
b1a21367 | 4114 | |
d1d9fd33 BB |
4115 | static void pidlist_free(void *p) |
4116 | { | |
58794514 | 4117 | kvfree(p); |
d1d9fd33 | 4118 | } |
d1d9fd33 | 4119 | |
b1a21367 TH |
4120 | /* |
4121 | * Used to destroy all pidlists lingering waiting for destroy timer. None | |
4122 | * should be left afterwards. | |
4123 | */ | |
4124 | static void cgroup_pidlist_destroy_all(struct cgroup *cgrp) | |
4125 | { | |
4126 | struct cgroup_pidlist *l, *tmp_l; | |
4127 | ||
4128 | mutex_lock(&cgrp->pidlist_mutex); | |
4129 | list_for_each_entry_safe(l, tmp_l, &cgrp->pidlists, links) | |
4130 | mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, 0); | |
4131 | mutex_unlock(&cgrp->pidlist_mutex); | |
4132 | ||
4133 | flush_workqueue(cgroup_pidlist_destroy_wq); | |
4134 | BUG_ON(!list_empty(&cgrp->pidlists)); | |
4135 | } | |
4136 | ||
4137 | static void cgroup_pidlist_destroy_work_fn(struct work_struct *work) | |
4138 | { | |
4139 | struct delayed_work *dwork = to_delayed_work(work); | |
4140 | struct cgroup_pidlist *l = container_of(dwork, struct cgroup_pidlist, | |
4141 | destroy_dwork); | |
4142 | struct cgroup_pidlist *tofree = NULL; | |
4143 | ||
4144 | mutex_lock(&l->owner->pidlist_mutex); | |
b1a21367 TH |
4145 | |
4146 | /* | |
04502365 TH |
4147 | * Destroy iff we didn't get queued again. The state won't change |
4148 | * as destroy_dwork can only be queued while locked. | |
b1a21367 | 4149 | */ |
04502365 | 4150 | if (!delayed_work_pending(dwork)) { |
b1a21367 TH |
4151 | list_del(&l->links); |
4152 | pidlist_free(l->list); | |
4153 | put_pid_ns(l->key.ns); | |
4154 | tofree = l; | |
4155 | } | |
4156 | ||
b1a21367 TH |
4157 | mutex_unlock(&l->owner->pidlist_mutex); |
4158 | kfree(tofree); | |
4159 | } | |
4160 | ||
bbcb81d0 | 4161 | /* |
102a775e | 4162 | * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries |
6ee211ad | 4163 | * Returns the number of unique elements. |
bbcb81d0 | 4164 | */ |
6ee211ad | 4165 | static int pidlist_uniq(pid_t *list, int length) |
bbcb81d0 | 4166 | { |
102a775e | 4167 | int src, dest = 1; |
102a775e BB |
4168 | |
4169 | /* | |
4170 | * we presume the 0th element is unique, so i starts at 1. trivial | |
4171 | * edge cases first; no work needs to be done for either | |
4172 | */ | |
4173 | if (length == 0 || length == 1) | |
4174 | return length; | |
4175 | /* src and dest walk down the list; dest counts unique elements */ | |
4176 | for (src = 1; src < length; src++) { | |
4177 | /* find next unique element */ | |
4178 | while (list[src] == list[src-1]) { | |
4179 | src++; | |
4180 | if (src == length) | |
4181 | goto after; | |
4182 | } | |
4183 | /* dest always points to where the next unique element goes */ | |
4184 | list[dest] = list[src]; | |
4185 | dest++; | |
4186 | } | |
4187 | after: | |
102a775e BB |
4188 | return dest; |
4189 | } | |
4190 | ||
afb2bc14 TH |
4191 | /* |
4192 | * The two pid files - task and cgroup.procs - guaranteed that the result | |
4193 | * is sorted, which forced this whole pidlist fiasco. As pid order is | |
4194 | * different per namespace, each namespace needs differently sorted list, | |
4195 | * making it impossible to use, for example, single rbtree of member tasks | |
4196 | * sorted by task pointer. As pidlists can be fairly large, allocating one | |
4197 | * per open file is dangerous, so cgroup had to implement shared pool of | |
4198 | * pidlists keyed by cgroup and namespace. | |
4199 | * | |
4200 | * All this extra complexity was caused by the original implementation | |
4201 | * committing to an entirely unnecessary property. In the long term, we | |
aa6ec29b TH |
4202 | * want to do away with it. Explicitly scramble sort order if on the |
4203 | * default hierarchy so that no such expectation exists in the new | |
4204 | * interface. | |
afb2bc14 TH |
4205 | * |
4206 | * Scrambling is done by swapping every two consecutive bits, which is | |
4207 | * non-identity one-to-one mapping which disturbs sort order sufficiently. | |
4208 | */ | |
4209 | static pid_t pid_fry(pid_t pid) | |
4210 | { | |
4211 | unsigned a = pid & 0x55555555; | |
4212 | unsigned b = pid & 0xAAAAAAAA; | |
4213 | ||
4214 | return (a << 1) | (b >> 1); | |
4215 | } | |
4216 | ||
4217 | static pid_t cgroup_pid_fry(struct cgroup *cgrp, pid_t pid) | |
4218 | { | |
aa6ec29b | 4219 | if (cgroup_on_dfl(cgrp)) |
afb2bc14 TH |
4220 | return pid_fry(pid); |
4221 | else | |
4222 | return pid; | |
4223 | } | |
4224 | ||
102a775e BB |
4225 | static int cmppid(const void *a, const void *b) |
4226 | { | |
4227 | return *(pid_t *)a - *(pid_t *)b; | |
4228 | } | |
4229 | ||
afb2bc14 TH |
4230 | static int fried_cmppid(const void *a, const void *b) |
4231 | { | |
4232 | return pid_fry(*(pid_t *)a) - pid_fry(*(pid_t *)b); | |
4233 | } | |
4234 | ||
e6b81710 TH |
4235 | static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp, |
4236 | enum cgroup_filetype type) | |
4237 | { | |
4238 | struct cgroup_pidlist *l; | |
4239 | /* don't need task_nsproxy() if we're looking at ourself */ | |
4240 | struct pid_namespace *ns = task_active_pid_ns(current); | |
4241 | ||
4242 | lockdep_assert_held(&cgrp->pidlist_mutex); | |
4243 | ||
4244 | list_for_each_entry(l, &cgrp->pidlists, links) | |
4245 | if (l->key.type == type && l->key.ns == ns) | |
4246 | return l; | |
4247 | return NULL; | |
4248 | } | |
4249 | ||
72a8cb30 BB |
4250 | /* |
4251 | * find the appropriate pidlist for our purpose (given procs vs tasks) | |
4252 | * returns with the lock on that pidlist already held, and takes care | |
4253 | * of the use count, or returns NULL with no locks held if we're out of | |
4254 | * memory. | |
4255 | */ | |
e6b81710 TH |
4256 | static struct cgroup_pidlist *cgroup_pidlist_find_create(struct cgroup *cgrp, |
4257 | enum cgroup_filetype type) | |
72a8cb30 BB |
4258 | { |
4259 | struct cgroup_pidlist *l; | |
b70cc5fd | 4260 | |
e6b81710 TH |
4261 | lockdep_assert_held(&cgrp->pidlist_mutex); |
4262 | ||
4263 | l = cgroup_pidlist_find(cgrp, type); | |
4264 | if (l) | |
4265 | return l; | |
4266 | ||
72a8cb30 | 4267 | /* entry not found; create a new one */ |
f4f4be2b | 4268 | l = kzalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL); |
e6b81710 | 4269 | if (!l) |
72a8cb30 | 4270 | return l; |
e6b81710 | 4271 | |
b1a21367 | 4272 | INIT_DELAYED_WORK(&l->destroy_dwork, cgroup_pidlist_destroy_work_fn); |
72a8cb30 | 4273 | l->key.type = type; |
e6b81710 TH |
4274 | /* don't need task_nsproxy() if we're looking at ourself */ |
4275 | l->key.ns = get_pid_ns(task_active_pid_ns(current)); | |
72a8cb30 BB |
4276 | l->owner = cgrp; |
4277 | list_add(&l->links, &cgrp->pidlists); | |
72a8cb30 BB |
4278 | return l; |
4279 | } | |
4280 | ||
102a775e BB |
4281 | /* |
4282 | * Load a cgroup's pidarray with either procs' tgids or tasks' pids | |
4283 | */ | |
72a8cb30 BB |
4284 | static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type, |
4285 | struct cgroup_pidlist **lp) | |
102a775e BB |
4286 | { |
4287 | pid_t *array; | |
4288 | int length; | |
4289 | int pid, n = 0; /* used for populating the array */ | |
72ec7029 | 4290 | struct css_task_iter it; |
817929ec | 4291 | struct task_struct *tsk; |
102a775e BB |
4292 | struct cgroup_pidlist *l; |
4293 | ||
4bac00d1 TH |
4294 | lockdep_assert_held(&cgrp->pidlist_mutex); |
4295 | ||
102a775e BB |
4296 | /* |
4297 | * If cgroup gets more users after we read count, we won't have | |
4298 | * enough space - tough. This race is indistinguishable to the | |
4299 | * caller from the case that the additional cgroup users didn't | |
4300 | * show up until sometime later on. | |
4301 | */ | |
4302 | length = cgroup_task_count(cgrp); | |
d1d9fd33 | 4303 | array = pidlist_allocate(length); |
102a775e BB |
4304 | if (!array) |
4305 | return -ENOMEM; | |
4306 | /* now, populate the array */ | |
9d800df1 | 4307 | css_task_iter_start(&cgrp->self, &it); |
72ec7029 | 4308 | while ((tsk = css_task_iter_next(&it))) { |
102a775e | 4309 | if (unlikely(n == length)) |
817929ec | 4310 | break; |
102a775e | 4311 | /* get tgid or pid for procs or tasks file respectively */ |
72a8cb30 BB |
4312 | if (type == CGROUP_FILE_PROCS) |
4313 | pid = task_tgid_vnr(tsk); | |
4314 | else | |
4315 | pid = task_pid_vnr(tsk); | |
102a775e BB |
4316 | if (pid > 0) /* make sure to only use valid results */ |
4317 | array[n++] = pid; | |
817929ec | 4318 | } |
72ec7029 | 4319 | css_task_iter_end(&it); |
102a775e BB |
4320 | length = n; |
4321 | /* now sort & (if procs) strip out duplicates */ | |
aa6ec29b | 4322 | if (cgroup_on_dfl(cgrp)) |
afb2bc14 TH |
4323 | sort(array, length, sizeof(pid_t), fried_cmppid, NULL); |
4324 | else | |
4325 | sort(array, length, sizeof(pid_t), cmppid, NULL); | |
72a8cb30 | 4326 | if (type == CGROUP_FILE_PROCS) |
6ee211ad | 4327 | length = pidlist_uniq(array, length); |
e6b81710 | 4328 | |
e6b81710 | 4329 | l = cgroup_pidlist_find_create(cgrp, type); |
72a8cb30 | 4330 | if (!l) { |
d1d9fd33 | 4331 | pidlist_free(array); |
72a8cb30 | 4332 | return -ENOMEM; |
102a775e | 4333 | } |
e6b81710 TH |
4334 | |
4335 | /* store array, freeing old if necessary */ | |
d1d9fd33 | 4336 | pidlist_free(l->list); |
102a775e BB |
4337 | l->list = array; |
4338 | l->length = length; | |
72a8cb30 | 4339 | *lp = l; |
102a775e | 4340 | return 0; |
bbcb81d0 PM |
4341 | } |
4342 | ||
846c7bb0 | 4343 | /** |
a043e3b2 | 4344 | * cgroupstats_build - build and fill cgroupstats |
846c7bb0 BS |
4345 | * @stats: cgroupstats to fill information into |
4346 | * @dentry: A dentry entry belonging to the cgroup for which stats have | |
4347 | * been requested. | |
a043e3b2 LZ |
4348 | * |
4349 | * Build and fill cgroupstats so that taskstats can export it to user | |
4350 | * space. | |
846c7bb0 BS |
4351 | */ |
4352 | int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry) | |
4353 | { | |
2bd59d48 | 4354 | struct kernfs_node *kn = kernfs_node_from_dentry(dentry); |
bd89aabc | 4355 | struct cgroup *cgrp; |
72ec7029 | 4356 | struct css_task_iter it; |
846c7bb0 | 4357 | struct task_struct *tsk; |
33d283be | 4358 | |
2bd59d48 TH |
4359 | /* it should be kernfs_node belonging to cgroupfs and is a directory */ |
4360 | if (dentry->d_sb->s_type != &cgroup_fs_type || !kn || | |
4361 | kernfs_type(kn) != KERNFS_DIR) | |
4362 | return -EINVAL; | |
4363 | ||
bad34660 LZ |
4364 | mutex_lock(&cgroup_mutex); |
4365 | ||
846c7bb0 | 4366 | /* |
2bd59d48 | 4367 | * We aren't being called from kernfs and there's no guarantee on |
ec903c0c | 4368 | * @kn->priv's validity. For this and css_tryget_online_from_dir(), |
2bd59d48 | 4369 | * @kn->priv is RCU safe. Let's do the RCU dancing. |
846c7bb0 | 4370 | */ |
2bd59d48 TH |
4371 | rcu_read_lock(); |
4372 | cgrp = rcu_dereference(kn->priv); | |
bad34660 | 4373 | if (!cgrp || cgroup_is_dead(cgrp)) { |
2bd59d48 | 4374 | rcu_read_unlock(); |
bad34660 | 4375 | mutex_unlock(&cgroup_mutex); |
2bd59d48 TH |
4376 | return -ENOENT; |
4377 | } | |
bad34660 | 4378 | rcu_read_unlock(); |
846c7bb0 | 4379 | |
9d800df1 | 4380 | css_task_iter_start(&cgrp->self, &it); |
72ec7029 | 4381 | while ((tsk = css_task_iter_next(&it))) { |
846c7bb0 BS |
4382 | switch (tsk->state) { |
4383 | case TASK_RUNNING: | |
4384 | stats->nr_running++; | |
4385 | break; | |
4386 | case TASK_INTERRUPTIBLE: | |
4387 | stats->nr_sleeping++; | |
4388 | break; | |
4389 | case TASK_UNINTERRUPTIBLE: | |
4390 | stats->nr_uninterruptible++; | |
4391 | break; | |
4392 | case TASK_STOPPED: | |
4393 | stats->nr_stopped++; | |
4394 | break; | |
4395 | default: | |
4396 | if (delayacct_is_task_waiting_on_io(tsk)) | |
4397 | stats->nr_io_wait++; | |
4398 | break; | |
4399 | } | |
4400 | } | |
72ec7029 | 4401 | css_task_iter_end(&it); |
846c7bb0 | 4402 | |
bad34660 | 4403 | mutex_unlock(&cgroup_mutex); |
2bd59d48 | 4404 | return 0; |
846c7bb0 BS |
4405 | } |
4406 | ||
8f3ff208 | 4407 | |
bbcb81d0 | 4408 | /* |
102a775e | 4409 | * seq_file methods for the tasks/procs files. The seq_file position is the |
cc31edce | 4410 | * next pid to display; the seq_file iterator is a pointer to the pid |
102a775e | 4411 | * in the cgroup->l->list array. |
bbcb81d0 | 4412 | */ |
cc31edce | 4413 | |
102a775e | 4414 | static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos) |
bbcb81d0 | 4415 | { |
cc31edce PM |
4416 | /* |
4417 | * Initially we receive a position value that corresponds to | |
4418 | * one more than the last pid shown (or 0 on the first call or | |
4419 | * after a seek to the start). Use a binary-search to find the | |
4420 | * next pid to display, if any | |
4421 | */ | |
2bd59d48 | 4422 | struct kernfs_open_file *of = s->private; |
7da11279 | 4423 | struct cgroup *cgrp = seq_css(s)->cgroup; |
4bac00d1 | 4424 | struct cgroup_pidlist *l; |
7da11279 | 4425 | enum cgroup_filetype type = seq_cft(s)->private; |
cc31edce | 4426 | int index = 0, pid = *pos; |
4bac00d1 TH |
4427 | int *iter, ret; |
4428 | ||
4429 | mutex_lock(&cgrp->pidlist_mutex); | |
4430 | ||
4431 | /* | |
5d22444f | 4432 | * !NULL @of->priv indicates that this isn't the first start() |
4bac00d1 | 4433 | * after open. If the matching pidlist is around, we can use that. |
5d22444f | 4434 | * Look for it. Note that @of->priv can't be used directly. It |
4bac00d1 TH |
4435 | * could already have been destroyed. |
4436 | */ | |
5d22444f TH |
4437 | if (of->priv) |
4438 | of->priv = cgroup_pidlist_find(cgrp, type); | |
4bac00d1 TH |
4439 | |
4440 | /* | |
4441 | * Either this is the first start() after open or the matching | |
4442 | * pidlist has been destroyed inbetween. Create a new one. | |
4443 | */ | |
5d22444f TH |
4444 | if (!of->priv) { |
4445 | ret = pidlist_array_load(cgrp, type, | |
4446 | (struct cgroup_pidlist **)&of->priv); | |
4bac00d1 TH |
4447 | if (ret) |
4448 | return ERR_PTR(ret); | |
4449 | } | |
5d22444f | 4450 | l = of->priv; |
cc31edce | 4451 | |
cc31edce | 4452 | if (pid) { |
102a775e | 4453 | int end = l->length; |
20777766 | 4454 | |
cc31edce PM |
4455 | while (index < end) { |
4456 | int mid = (index + end) / 2; | |
afb2bc14 | 4457 | if (cgroup_pid_fry(cgrp, l->list[mid]) == pid) { |
cc31edce PM |
4458 | index = mid; |
4459 | break; | |
afb2bc14 | 4460 | } else if (cgroup_pid_fry(cgrp, l->list[mid]) <= pid) |
cc31edce PM |
4461 | index = mid + 1; |
4462 | else | |
4463 | end = mid; | |
4464 | } | |
4465 | } | |
4466 | /* If we're off the end of the array, we're done */ | |
102a775e | 4467 | if (index >= l->length) |
cc31edce PM |
4468 | return NULL; |
4469 | /* Update the abstract position to be the actual pid that we found */ | |
102a775e | 4470 | iter = l->list + index; |
afb2bc14 | 4471 | *pos = cgroup_pid_fry(cgrp, *iter); |
cc31edce PM |
4472 | return iter; |
4473 | } | |
4474 | ||
102a775e | 4475 | static void cgroup_pidlist_stop(struct seq_file *s, void *v) |
cc31edce | 4476 | { |
2bd59d48 | 4477 | struct kernfs_open_file *of = s->private; |
5d22444f | 4478 | struct cgroup_pidlist *l = of->priv; |
62236858 | 4479 | |
5d22444f TH |
4480 | if (l) |
4481 | mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, | |
04502365 | 4482 | CGROUP_PIDLIST_DESTROY_DELAY); |
7da11279 | 4483 | mutex_unlock(&seq_css(s)->cgroup->pidlist_mutex); |
cc31edce PM |
4484 | } |
4485 | ||
102a775e | 4486 | static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos) |
cc31edce | 4487 | { |
2bd59d48 | 4488 | struct kernfs_open_file *of = s->private; |
5d22444f | 4489 | struct cgroup_pidlist *l = of->priv; |
102a775e BB |
4490 | pid_t *p = v; |
4491 | pid_t *end = l->list + l->length; | |
cc31edce PM |
4492 | /* |
4493 | * Advance to the next pid in the array. If this goes off the | |
4494 | * end, we're done | |
4495 | */ | |
4496 | p++; | |
4497 | if (p >= end) { | |
4498 | return NULL; | |
4499 | } else { | |
7da11279 | 4500 | *pos = cgroup_pid_fry(seq_css(s)->cgroup, *p); |
cc31edce PM |
4501 | return p; |
4502 | } | |
4503 | } | |
4504 | ||
102a775e | 4505 | static int cgroup_pidlist_show(struct seq_file *s, void *v) |
cc31edce | 4506 | { |
94ff212d JP |
4507 | seq_printf(s, "%d\n", *(int *)v); |
4508 | ||
4509 | return 0; | |
cc31edce | 4510 | } |
bbcb81d0 | 4511 | |
182446d0 TH |
4512 | static u64 cgroup_read_notify_on_release(struct cgroup_subsys_state *css, |
4513 | struct cftype *cft) | |
81a6a5cd | 4514 | { |
182446d0 | 4515 | return notify_on_release(css->cgroup); |
81a6a5cd PM |
4516 | } |
4517 | ||
182446d0 TH |
4518 | static int cgroup_write_notify_on_release(struct cgroup_subsys_state *css, |
4519 | struct cftype *cft, u64 val) | |
6379c106 | 4520 | { |
6379c106 | 4521 | if (val) |
182446d0 | 4522 | set_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags); |
6379c106 | 4523 | else |
182446d0 | 4524 | clear_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags); |
6379c106 PM |
4525 | return 0; |
4526 | } | |
4527 | ||
182446d0 TH |
4528 | static u64 cgroup_clone_children_read(struct cgroup_subsys_state *css, |
4529 | struct cftype *cft) | |
97978e6d | 4530 | { |
182446d0 | 4531 | return test_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d DL |
4532 | } |
4533 | ||
182446d0 TH |
4534 | static int cgroup_clone_children_write(struct cgroup_subsys_state *css, |
4535 | struct cftype *cft, u64 val) | |
97978e6d DL |
4536 | { |
4537 | if (val) | |
182446d0 | 4538 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d | 4539 | else |
182446d0 | 4540 | clear_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d DL |
4541 | return 0; |
4542 | } | |
4543 | ||
a14c6874 TH |
4544 | /* cgroup core interface files for the default hierarchy */ |
4545 | static struct cftype cgroup_dfl_base_files[] = { | |
81a6a5cd | 4546 | { |
d5c56ced | 4547 | .name = "cgroup.procs", |
6f60eade | 4548 | .file_offset = offsetof(struct cgroup, procs_file), |
6612f05b TH |
4549 | .seq_start = cgroup_pidlist_start, |
4550 | .seq_next = cgroup_pidlist_next, | |
4551 | .seq_stop = cgroup_pidlist_stop, | |
4552 | .seq_show = cgroup_pidlist_show, | |
5d22444f | 4553 | .private = CGROUP_FILE_PROCS, |
acbef755 | 4554 | .write = cgroup_procs_write, |
102a775e | 4555 | }, |
f8f22e53 TH |
4556 | { |
4557 | .name = "cgroup.controllers", | |
a14c6874 | 4558 | .flags = CFTYPE_ONLY_ON_ROOT, |
f8f22e53 TH |
4559 | .seq_show = cgroup_root_controllers_show, |
4560 | }, | |
4561 | { | |
4562 | .name = "cgroup.controllers", | |
a14c6874 | 4563 | .flags = CFTYPE_NOT_ON_ROOT, |
f8f22e53 TH |
4564 | .seq_show = cgroup_controllers_show, |
4565 | }, | |
4566 | { | |
4567 | .name = "cgroup.subtree_control", | |
f8f22e53 | 4568 | .seq_show = cgroup_subtree_control_show, |
451af504 | 4569 | .write = cgroup_subtree_control_write, |
f8f22e53 | 4570 | }, |
842b597e | 4571 | { |
4a07c222 | 4572 | .name = "cgroup.events", |
a14c6874 | 4573 | .flags = CFTYPE_NOT_ON_ROOT, |
6f60eade | 4574 | .file_offset = offsetof(struct cgroup, events_file), |
4a07c222 | 4575 | .seq_show = cgroup_events_show, |
842b597e | 4576 | }, |
a14c6874 TH |
4577 | { } /* terminate */ |
4578 | }; | |
d5c56ced | 4579 | |
a14c6874 TH |
4580 | /* cgroup core interface files for the legacy hierarchies */ |
4581 | static struct cftype cgroup_legacy_base_files[] = { | |
4582 | { | |
4583 | .name = "cgroup.procs", | |
4584 | .seq_start = cgroup_pidlist_start, | |
4585 | .seq_next = cgroup_pidlist_next, | |
4586 | .seq_stop = cgroup_pidlist_stop, | |
4587 | .seq_show = cgroup_pidlist_show, | |
4588 | .private = CGROUP_FILE_PROCS, | |
4589 | .write = cgroup_procs_write, | |
a14c6874 TH |
4590 | }, |
4591 | { | |
4592 | .name = "cgroup.clone_children", | |
4593 | .read_u64 = cgroup_clone_children_read, | |
4594 | .write_u64 = cgroup_clone_children_write, | |
4595 | }, | |
4596 | { | |
4597 | .name = "cgroup.sane_behavior", | |
4598 | .flags = CFTYPE_ONLY_ON_ROOT, | |
4599 | .seq_show = cgroup_sane_behavior_show, | |
4600 | }, | |
d5c56ced TH |
4601 | { |
4602 | .name = "tasks", | |
6612f05b TH |
4603 | .seq_start = cgroup_pidlist_start, |
4604 | .seq_next = cgroup_pidlist_next, | |
4605 | .seq_stop = cgroup_pidlist_stop, | |
4606 | .seq_show = cgroup_pidlist_show, | |
5d22444f | 4607 | .private = CGROUP_FILE_TASKS, |
acbef755 | 4608 | .write = cgroup_tasks_write, |
d5c56ced TH |
4609 | }, |
4610 | { | |
4611 | .name = "notify_on_release", | |
d5c56ced TH |
4612 | .read_u64 = cgroup_read_notify_on_release, |
4613 | .write_u64 = cgroup_write_notify_on_release, | |
4614 | }, | |
6e6ff25b TH |
4615 | { |
4616 | .name = "release_agent", | |
a14c6874 | 4617 | .flags = CFTYPE_ONLY_ON_ROOT, |
2da8ca82 | 4618 | .seq_show = cgroup_release_agent_show, |
451af504 | 4619 | .write = cgroup_release_agent_write, |
5f469907 | 4620 | .max_write_len = PATH_MAX - 1, |
6e6ff25b | 4621 | }, |
db0416b6 | 4622 | { } /* terminate */ |
bbcb81d0 PM |
4623 | }; |
4624 | ||
0c21ead1 TH |
4625 | /* |
4626 | * css destruction is four-stage process. | |
4627 | * | |
4628 | * 1. Destruction starts. Killing of the percpu_ref is initiated. | |
4629 | * Implemented in kill_css(). | |
4630 | * | |
4631 | * 2. When the percpu_ref is confirmed to be visible as killed on all CPUs | |
ec903c0c TH |
4632 | * and thus css_tryget_online() is guaranteed to fail, the css can be |
4633 | * offlined by invoking offline_css(). After offlining, the base ref is | |
4634 | * put. Implemented in css_killed_work_fn(). | |
0c21ead1 TH |
4635 | * |
4636 | * 3. When the percpu_ref reaches zero, the only possible remaining | |
4637 | * accessors are inside RCU read sections. css_release() schedules the | |
4638 | * RCU callback. | |
4639 | * | |
4640 | * 4. After the grace period, the css can be freed. Implemented in | |
4641 | * css_free_work_fn(). | |
4642 | * | |
4643 | * It is actually hairier because both step 2 and 4 require process context | |
4644 | * and thus involve punting to css->destroy_work adding two additional | |
4645 | * steps to the already complex sequence. | |
4646 | */ | |
35ef10da | 4647 | static void css_free_work_fn(struct work_struct *work) |
48ddbe19 TH |
4648 | { |
4649 | struct cgroup_subsys_state *css = | |
35ef10da | 4650 | container_of(work, struct cgroup_subsys_state, destroy_work); |
01e58659 | 4651 | struct cgroup_subsys *ss = css->ss; |
0c21ead1 | 4652 | struct cgroup *cgrp = css->cgroup; |
48ddbe19 | 4653 | |
9a1049da TH |
4654 | percpu_ref_exit(&css->refcnt); |
4655 | ||
01e58659 | 4656 | if (ss) { |
9d755d33 | 4657 | /* css free path */ |
8bb5ef79 | 4658 | struct cgroup_subsys_state *parent = css->parent; |
01e58659 VD |
4659 | int id = css->id; |
4660 | ||
01e58659 VD |
4661 | ss->css_free(css); |
4662 | cgroup_idr_remove(&ss->css_idr, id); | |
9d755d33 | 4663 | cgroup_put(cgrp); |
8bb5ef79 TH |
4664 | |
4665 | if (parent) | |
4666 | css_put(parent); | |
9d755d33 TH |
4667 | } else { |
4668 | /* cgroup free path */ | |
4669 | atomic_dec(&cgrp->root->nr_cgrps); | |
4670 | cgroup_pidlist_destroy_all(cgrp); | |
971ff493 | 4671 | cancel_work_sync(&cgrp->release_agent_work); |
9d755d33 | 4672 | |
d51f39b0 | 4673 | if (cgroup_parent(cgrp)) { |
9d755d33 TH |
4674 | /* |
4675 | * We get a ref to the parent, and put the ref when | |
4676 | * this cgroup is being freed, so it's guaranteed | |
4677 | * that the parent won't be destroyed before its | |
4678 | * children. | |
4679 | */ | |
d51f39b0 | 4680 | cgroup_put(cgroup_parent(cgrp)); |
9d755d33 TH |
4681 | kernfs_put(cgrp->kn); |
4682 | kfree(cgrp); | |
4683 | } else { | |
4684 | /* | |
4685 | * This is root cgroup's refcnt reaching zero, | |
4686 | * which indicates that the root should be | |
4687 | * released. | |
4688 | */ | |
4689 | cgroup_destroy_root(cgrp->root); | |
4690 | } | |
4691 | } | |
48ddbe19 TH |
4692 | } |
4693 | ||
0c21ead1 | 4694 | static void css_free_rcu_fn(struct rcu_head *rcu_head) |
d3daf28d TH |
4695 | { |
4696 | struct cgroup_subsys_state *css = | |
0c21ead1 | 4697 | container_of(rcu_head, struct cgroup_subsys_state, rcu_head); |
d3daf28d | 4698 | |
35ef10da | 4699 | INIT_WORK(&css->destroy_work, css_free_work_fn); |
e5fca243 | 4700 | queue_work(cgroup_destroy_wq, &css->destroy_work); |
48ddbe19 TH |
4701 | } |
4702 | ||
25e15d83 | 4703 | static void css_release_work_fn(struct work_struct *work) |
d3daf28d TH |
4704 | { |
4705 | struct cgroup_subsys_state *css = | |
25e15d83 | 4706 | container_of(work, struct cgroup_subsys_state, destroy_work); |
15a4c835 | 4707 | struct cgroup_subsys *ss = css->ss; |
9d755d33 | 4708 | struct cgroup *cgrp = css->cgroup; |
15a4c835 | 4709 | |
1fed1b2e TH |
4710 | mutex_lock(&cgroup_mutex); |
4711 | ||
de3f0341 | 4712 | css->flags |= CSS_RELEASED; |
1fed1b2e TH |
4713 | list_del_rcu(&css->sibling); |
4714 | ||
9d755d33 TH |
4715 | if (ss) { |
4716 | /* css release path */ | |
01e58659 | 4717 | cgroup_idr_replace(&ss->css_idr, NULL, css->id); |
7d172cc8 TH |
4718 | if (ss->css_released) |
4719 | ss->css_released(css); | |
9d755d33 TH |
4720 | } else { |
4721 | /* cgroup release path */ | |
9d755d33 TH |
4722 | cgroup_idr_remove(&cgrp->root->cgroup_idr, cgrp->id); |
4723 | cgrp->id = -1; | |
a4189487 LZ |
4724 | |
4725 | /* | |
4726 | * There are two control paths which try to determine | |
4727 | * cgroup from dentry without going through kernfs - | |
4728 | * cgroupstats_build() and css_tryget_online_from_dir(). | |
4729 | * Those are supported by RCU protecting clearing of | |
4730 | * cgrp->kn->priv backpointer. | |
4731 | */ | |
6cd0f5bb TH |
4732 | if (cgrp->kn) |
4733 | RCU_INIT_POINTER(*(void __rcu __force **)&cgrp->kn->priv, | |
4734 | NULL); | |
9d755d33 | 4735 | } |
d3daf28d | 4736 | |
1fed1b2e TH |
4737 | mutex_unlock(&cgroup_mutex); |
4738 | ||
0c21ead1 | 4739 | call_rcu(&css->rcu_head, css_free_rcu_fn); |
d3daf28d TH |
4740 | } |
4741 | ||
d3daf28d TH |
4742 | static void css_release(struct percpu_ref *ref) |
4743 | { | |
4744 | struct cgroup_subsys_state *css = | |
4745 | container_of(ref, struct cgroup_subsys_state, refcnt); | |
4746 | ||
25e15d83 TH |
4747 | INIT_WORK(&css->destroy_work, css_release_work_fn); |
4748 | queue_work(cgroup_destroy_wq, &css->destroy_work); | |
d3daf28d TH |
4749 | } |
4750 | ||
ddfcadab TH |
4751 | static void init_and_link_css(struct cgroup_subsys_state *css, |
4752 | struct cgroup_subsys *ss, struct cgroup *cgrp) | |
ddbcc7e8 | 4753 | { |
0cb51d71 TH |
4754 | lockdep_assert_held(&cgroup_mutex); |
4755 | ||
ddfcadab TH |
4756 | cgroup_get(cgrp); |
4757 | ||
d5c419b6 | 4758 | memset(css, 0, sizeof(*css)); |
bd89aabc | 4759 | css->cgroup = cgrp; |
72c97e54 | 4760 | css->ss = ss; |
d5c419b6 TH |
4761 | INIT_LIST_HEAD(&css->sibling); |
4762 | INIT_LIST_HEAD(&css->children); | |
0cb51d71 | 4763 | css->serial_nr = css_serial_nr_next++; |
aa226ff4 | 4764 | atomic_set(&css->online_cnt, 0); |
0ae78e0b | 4765 | |
d51f39b0 TH |
4766 | if (cgroup_parent(cgrp)) { |
4767 | css->parent = cgroup_css(cgroup_parent(cgrp), ss); | |
ddfcadab | 4768 | css_get(css->parent); |
ddfcadab | 4769 | } |
48ddbe19 | 4770 | |
ca8bdcaf | 4771 | BUG_ON(cgroup_css(cgrp, ss)); |
ddbcc7e8 PM |
4772 | } |
4773 | ||
2a4ac633 | 4774 | /* invoke ->css_online() on a new CSS and mark it online if successful */ |
623f926b | 4775 | static int online_css(struct cgroup_subsys_state *css) |
a31f2d3f | 4776 | { |
623f926b | 4777 | struct cgroup_subsys *ss = css->ss; |
b1929db4 TH |
4778 | int ret = 0; |
4779 | ||
a31f2d3f TH |
4780 | lockdep_assert_held(&cgroup_mutex); |
4781 | ||
92fb9748 | 4782 | if (ss->css_online) |
eb95419b | 4783 | ret = ss->css_online(css); |
ae7f164a | 4784 | if (!ret) { |
eb95419b | 4785 | css->flags |= CSS_ONLINE; |
aec25020 | 4786 | rcu_assign_pointer(css->cgroup->subsys[ss->id], css); |
aa226ff4 TH |
4787 | |
4788 | atomic_inc(&css->online_cnt); | |
4789 | if (css->parent) | |
4790 | atomic_inc(&css->parent->online_cnt); | |
ae7f164a | 4791 | } |
b1929db4 | 4792 | return ret; |
a31f2d3f TH |
4793 | } |
4794 | ||
2a4ac633 | 4795 | /* if the CSS is online, invoke ->css_offline() on it and mark it offline */ |
623f926b | 4796 | static void offline_css(struct cgroup_subsys_state *css) |
a31f2d3f | 4797 | { |
623f926b | 4798 | struct cgroup_subsys *ss = css->ss; |
a31f2d3f TH |
4799 | |
4800 | lockdep_assert_held(&cgroup_mutex); | |
4801 | ||
4802 | if (!(css->flags & CSS_ONLINE)) | |
4803 | return; | |
4804 | ||
fa06235b VD |
4805 | if (ss->css_reset) |
4806 | ss->css_reset(css); | |
4807 | ||
d7eeac19 | 4808 | if (ss->css_offline) |
eb95419b | 4809 | ss->css_offline(css); |
a31f2d3f | 4810 | |
eb95419b | 4811 | css->flags &= ~CSS_ONLINE; |
e3297803 | 4812 | RCU_INIT_POINTER(css->cgroup->subsys[ss->id], NULL); |
f8f22e53 TH |
4813 | |
4814 | wake_up_all(&css->cgroup->offline_waitq); | |
a31f2d3f TH |
4815 | } |
4816 | ||
c81c925a | 4817 | /** |
6cd0f5bb | 4818 | * css_create - create a cgroup_subsys_state |
c81c925a TH |
4819 | * @cgrp: the cgroup new css will be associated with |
4820 | * @ss: the subsys of new css | |
4821 | * | |
4822 | * Create a new css associated with @cgrp - @ss pair. On success, the new | |
6cd0f5bb TH |
4823 | * css is online and installed in @cgrp. This function doesn't create the |
4824 | * interface files. Returns 0 on success, -errno on failure. | |
c81c925a | 4825 | */ |
6cd0f5bb TH |
4826 | static struct cgroup_subsys_state *css_create(struct cgroup *cgrp, |
4827 | struct cgroup_subsys *ss) | |
c81c925a | 4828 | { |
d51f39b0 | 4829 | struct cgroup *parent = cgroup_parent(cgrp); |
1fed1b2e | 4830 | struct cgroup_subsys_state *parent_css = cgroup_css(parent, ss); |
c81c925a TH |
4831 | struct cgroup_subsys_state *css; |
4832 | int err; | |
4833 | ||
c81c925a TH |
4834 | lockdep_assert_held(&cgroup_mutex); |
4835 | ||
1fed1b2e | 4836 | css = ss->css_alloc(parent_css); |
c81c925a | 4837 | if (IS_ERR(css)) |
6cd0f5bb | 4838 | return css; |
c81c925a | 4839 | |
ddfcadab | 4840 | init_and_link_css(css, ss, cgrp); |
a2bed820 | 4841 | |
2aad2a86 | 4842 | err = percpu_ref_init(&css->refcnt, css_release, 0, GFP_KERNEL); |
c81c925a | 4843 | if (err) |
3eb59ec6 | 4844 | goto err_free_css; |
c81c925a | 4845 | |
cf780b7d | 4846 | err = cgroup_idr_alloc(&ss->css_idr, NULL, 2, 0, GFP_KERNEL); |
15a4c835 TH |
4847 | if (err < 0) |
4848 | goto err_free_percpu_ref; | |
4849 | css->id = err; | |
c81c925a | 4850 | |
15a4c835 | 4851 | /* @css is ready to be brought online now, make it visible */ |
1fed1b2e | 4852 | list_add_tail_rcu(&css->sibling, &parent_css->children); |
15a4c835 | 4853 | cgroup_idr_replace(&ss->css_idr, css, css->id); |
c81c925a TH |
4854 | |
4855 | err = online_css(css); | |
4856 | if (err) | |
1fed1b2e | 4857 | goto err_list_del; |
94419627 | 4858 | |
c81c925a | 4859 | if (ss->broken_hierarchy && !ss->warned_broken_hierarchy && |
d51f39b0 | 4860 | cgroup_parent(parent)) { |
ed3d261b | 4861 | 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 | 4862 | current->comm, current->pid, ss->name); |
c81c925a | 4863 | if (!strcmp(ss->name, "memory")) |
ed3d261b | 4864 | pr_warn("\"memory\" requires setting use_hierarchy to 1 on the root\n"); |
c81c925a TH |
4865 | ss->warned_broken_hierarchy = true; |
4866 | } | |
4867 | ||
6cd0f5bb | 4868 | return css; |
c81c925a | 4869 | |
1fed1b2e TH |
4870 | err_list_del: |
4871 | list_del_rcu(&css->sibling); | |
15a4c835 | 4872 | cgroup_idr_remove(&ss->css_idr, css->id); |
3eb59ec6 | 4873 | err_free_percpu_ref: |
9a1049da | 4874 | percpu_ref_exit(&css->refcnt); |
3eb59ec6 | 4875 | err_free_css: |
a2bed820 | 4876 | call_rcu(&css->rcu_head, css_free_rcu_fn); |
6cd0f5bb | 4877 | return ERR_PTR(err); |
c81c925a TH |
4878 | } |
4879 | ||
b3bfd983 TH |
4880 | static int cgroup_mkdir(struct kernfs_node *parent_kn, const char *name, |
4881 | umode_t mode) | |
ddbcc7e8 | 4882 | { |
b11cfb58 | 4883 | struct cgroup *parent, *cgrp, *tcgrp; |
a9746d8d | 4884 | struct cgroup_root *root; |
ddbcc7e8 | 4885 | struct cgroup_subsys *ss; |
2bd59d48 | 4886 | struct kernfs_node *kn; |
b11cfb58 | 4887 | int level, ssid, ret; |
ddbcc7e8 | 4888 | |
71b1fb5c AC |
4889 | /* Do not accept '\n' to prevent making /proc/<pid>/cgroup unparsable. |
4890 | */ | |
4891 | if (strchr(name, '\n')) | |
4892 | return -EINVAL; | |
4893 | ||
a9746d8d TH |
4894 | parent = cgroup_kn_lock_live(parent_kn); |
4895 | if (!parent) | |
4896 | return -ENODEV; | |
4897 | root = parent->root; | |
b11cfb58 | 4898 | level = parent->level + 1; |
ddbcc7e8 | 4899 | |
0a950f65 | 4900 | /* allocate the cgroup and its ID, 0 is reserved for the root */ |
b11cfb58 TH |
4901 | cgrp = kzalloc(sizeof(*cgrp) + |
4902 | sizeof(cgrp->ancestor_ids[0]) * (level + 1), GFP_KERNEL); | |
ba0f4d76 TH |
4903 | if (!cgrp) { |
4904 | ret = -ENOMEM; | |
4905 | goto out_unlock; | |
0ab02ca8 LZ |
4906 | } |
4907 | ||
2aad2a86 | 4908 | ret = percpu_ref_init(&cgrp->self.refcnt, css_release, 0, GFP_KERNEL); |
9d755d33 TH |
4909 | if (ret) |
4910 | goto out_free_cgrp; | |
4911 | ||
0ab02ca8 LZ |
4912 | /* |
4913 | * Temporarily set the pointer to NULL, so idr_find() won't return | |
4914 | * a half-baked cgroup. | |
4915 | */ | |
cf780b7d | 4916 | cgrp->id = cgroup_idr_alloc(&root->cgroup_idr, NULL, 2, 0, GFP_KERNEL); |
0ab02ca8 | 4917 | if (cgrp->id < 0) { |
ba0f4d76 | 4918 | ret = -ENOMEM; |
9d755d33 | 4919 | goto out_cancel_ref; |
976c06bc TH |
4920 | } |
4921 | ||
cc31edce | 4922 | init_cgroup_housekeeping(cgrp); |
ddbcc7e8 | 4923 | |
9d800df1 | 4924 | cgrp->self.parent = &parent->self; |
ba0f4d76 | 4925 | cgrp->root = root; |
b11cfb58 TH |
4926 | cgrp->level = level; |
4927 | ||
4928 | for (tcgrp = cgrp; tcgrp; tcgrp = cgroup_parent(tcgrp)) | |
4929 | cgrp->ancestor_ids[tcgrp->level] = tcgrp->id; | |
ddbcc7e8 | 4930 | |
b6abdb0e LZ |
4931 | if (notify_on_release(parent)) |
4932 | set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); | |
4933 | ||
2260e7fc TH |
4934 | if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &parent->flags)) |
4935 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags); | |
97978e6d | 4936 | |
2bd59d48 | 4937 | /* create the directory */ |
e61734c5 | 4938 | kn = kernfs_create_dir(parent->kn, name, mode, cgrp); |
2bd59d48 | 4939 | if (IS_ERR(kn)) { |
ba0f4d76 TH |
4940 | ret = PTR_ERR(kn); |
4941 | goto out_free_id; | |
2bd59d48 TH |
4942 | } |
4943 | cgrp->kn = kn; | |
ddbcc7e8 | 4944 | |
4e139afc | 4945 | /* |
6f30558f TH |
4946 | * This extra ref will be put in cgroup_free_fn() and guarantees |
4947 | * that @cgrp->kn is always accessible. | |
4e139afc | 4948 | */ |
6f30558f | 4949 | kernfs_get(kn); |
ddbcc7e8 | 4950 | |
0cb51d71 | 4951 | cgrp->self.serial_nr = css_serial_nr_next++; |
53fa5261 | 4952 | |
4e139afc | 4953 | /* allocation complete, commit to creation */ |
d5c419b6 | 4954 | list_add_tail_rcu(&cgrp->self.sibling, &cgroup_parent(cgrp)->self.children); |
3c9c825b | 4955 | atomic_inc(&root->nr_cgrps); |
59f5296b | 4956 | cgroup_get(parent); |
415cf07a | 4957 | |
0d80255e TH |
4958 | /* |
4959 | * @cgrp is now fully operational. If something fails after this | |
4960 | * point, it'll be released via the normal destruction path. | |
4961 | */ | |
6fa4918d | 4962 | cgroup_idr_replace(&root->cgroup_idr, cgrp, cgrp->id); |
4e96ee8e | 4963 | |
ba0f4d76 TH |
4964 | ret = cgroup_kn_set_ugid(kn); |
4965 | if (ret) | |
4966 | goto out_destroy; | |
49957f8e | 4967 | |
4df8dc90 | 4968 | ret = css_populate_dir(&cgrp->self, NULL); |
ba0f4d76 TH |
4969 | if (ret) |
4970 | goto out_destroy; | |
628f7cd4 | 4971 | |
9d403e99 | 4972 | /* let's create and online css's */ |
996cd1fb | 4973 | do_each_subsys_mask(ss, ssid, parent->subtree_ss_mask) { |
6cd0f5bb TH |
4974 | struct cgroup_subsys_state *css; |
4975 | ||
4976 | css = css_create(cgrp, ss); | |
4977 | if (IS_ERR(css)) { | |
4978 | ret = PTR_ERR(css); | |
996cd1fb | 4979 | goto out_destroy; |
6cd0f5bb TH |
4980 | } |
4981 | ||
4982 | if (parent->subtree_control & (1 << ssid)) { | |
4983 | ret = css_populate_dir(css, NULL); | |
4984 | if (ret) | |
4985 | goto out_destroy; | |
4986 | } | |
996cd1fb | 4987 | } while_each_subsys_mask(); |
ddbcc7e8 | 4988 | |
bd53d617 TH |
4989 | /* |
4990 | * On the default hierarchy, a child doesn't automatically inherit | |
667c2491 | 4991 | * subtree_control from the parent. Each is configured manually. |
bd53d617 | 4992 | */ |
667c2491 TH |
4993 | if (!cgroup_on_dfl(cgrp)) { |
4994 | cgrp->subtree_control = parent->subtree_control; | |
8699b776 | 4995 | cgroup_refresh_subtree_ss_mask(cgrp); |
667c2491 | 4996 | } |
2bd59d48 | 4997 | |
2bd59d48 | 4998 | kernfs_activate(kn); |
ddbcc7e8 | 4999 | |
ba0f4d76 TH |
5000 | ret = 0; |
5001 | goto out_unlock; | |
ddbcc7e8 | 5002 | |
ba0f4d76 | 5003 | out_free_id: |
6fa4918d | 5004 | cgroup_idr_remove(&root->cgroup_idr, cgrp->id); |
9d755d33 | 5005 | out_cancel_ref: |
9a1049da | 5006 | percpu_ref_exit(&cgrp->self.refcnt); |
ba0f4d76 | 5007 | out_free_cgrp: |
bd89aabc | 5008 | kfree(cgrp); |
ba0f4d76 | 5009 | out_unlock: |
a9746d8d | 5010 | cgroup_kn_unlock(parent_kn); |
ba0f4d76 | 5011 | return ret; |
4b8b47eb | 5012 | |
ba0f4d76 | 5013 | out_destroy: |
4b8b47eb | 5014 | cgroup_destroy_locked(cgrp); |
ba0f4d76 | 5015 | goto out_unlock; |
ddbcc7e8 PM |
5016 | } |
5017 | ||
223dbc38 TH |
5018 | /* |
5019 | * This is called when the refcnt of a css is confirmed to be killed. | |
249f3468 TH |
5020 | * css_tryget_online() is now guaranteed to fail. Tell the subsystem to |
5021 | * initate destruction and put the css ref from kill_css(). | |
223dbc38 TH |
5022 | */ |
5023 | static void css_killed_work_fn(struct work_struct *work) | |
d3daf28d | 5024 | { |
223dbc38 TH |
5025 | struct cgroup_subsys_state *css = |
5026 | container_of(work, struct cgroup_subsys_state, destroy_work); | |
d3daf28d | 5027 | |
f20104de | 5028 | mutex_lock(&cgroup_mutex); |
09a503ea | 5029 | |
aa226ff4 TH |
5030 | do { |
5031 | offline_css(css); | |
5032 | css_put(css); | |
5033 | /* @css can't go away while we're holding cgroup_mutex */ | |
5034 | css = css->parent; | |
5035 | } while (css && atomic_dec_and_test(&css->online_cnt)); | |
5036 | ||
5037 | mutex_unlock(&cgroup_mutex); | |
d3daf28d TH |
5038 | } |
5039 | ||
223dbc38 TH |
5040 | /* css kill confirmation processing requires process context, bounce */ |
5041 | static void css_killed_ref_fn(struct percpu_ref *ref) | |
d3daf28d TH |
5042 | { |
5043 | struct cgroup_subsys_state *css = | |
5044 | container_of(ref, struct cgroup_subsys_state, refcnt); | |
5045 | ||
aa226ff4 TH |
5046 | if (atomic_dec_and_test(&css->online_cnt)) { |
5047 | INIT_WORK(&css->destroy_work, css_killed_work_fn); | |
5048 | queue_work(cgroup_destroy_wq, &css->destroy_work); | |
5049 | } | |
d3daf28d TH |
5050 | } |
5051 | ||
f392e51c TH |
5052 | /** |
5053 | * kill_css - destroy a css | |
5054 | * @css: css to destroy | |
5055 | * | |
5056 | * This function initiates destruction of @css by removing cgroup interface | |
5057 | * files and putting its base reference. ->css_offline() will be invoked | |
ec903c0c TH |
5058 | * asynchronously once css_tryget_online() is guaranteed to fail and when |
5059 | * the reference count reaches zero, @css will be released. | |
f392e51c TH |
5060 | */ |
5061 | static void kill_css(struct cgroup_subsys_state *css) | |
edae0c33 | 5062 | { |
01f6474c | 5063 | lockdep_assert_held(&cgroup_mutex); |
94419627 | 5064 | |
2bd59d48 TH |
5065 | /* |
5066 | * This must happen before css is disassociated with its cgroup. | |
5067 | * See seq_css() for details. | |
5068 | */ | |
4df8dc90 | 5069 | css_clear_dir(css, NULL); |
3c14f8b4 | 5070 | |
edae0c33 TH |
5071 | /* |
5072 | * Killing would put the base ref, but we need to keep it alive | |
5073 | * until after ->css_offline(). | |
5074 | */ | |
5075 | css_get(css); | |
5076 | ||
5077 | /* | |
5078 | * cgroup core guarantees that, by the time ->css_offline() is | |
5079 | * invoked, no new css reference will be given out via | |
ec903c0c | 5080 | * css_tryget_online(). We can't simply call percpu_ref_kill() and |
edae0c33 TH |
5081 | * proceed to offlining css's because percpu_ref_kill() doesn't |
5082 | * guarantee that the ref is seen as killed on all CPUs on return. | |
5083 | * | |
5084 | * Use percpu_ref_kill_and_confirm() to get notifications as each | |
5085 | * css is confirmed to be seen as killed on all CPUs. | |
5086 | */ | |
5087 | percpu_ref_kill_and_confirm(&css->refcnt, css_killed_ref_fn); | |
d3daf28d TH |
5088 | } |
5089 | ||
5090 | /** | |
5091 | * cgroup_destroy_locked - the first stage of cgroup destruction | |
5092 | * @cgrp: cgroup to be destroyed | |
5093 | * | |
5094 | * css's make use of percpu refcnts whose killing latency shouldn't be | |
5095 | * exposed to userland and are RCU protected. Also, cgroup core needs to | |
ec903c0c TH |
5096 | * guarantee that css_tryget_online() won't succeed by the time |
5097 | * ->css_offline() is invoked. To satisfy all the requirements, | |
5098 | * destruction is implemented in the following two steps. | |
d3daf28d TH |
5099 | * |
5100 | * s1. Verify @cgrp can be destroyed and mark it dying. Remove all | |
5101 | * userland visible parts and start killing the percpu refcnts of | |
5102 | * css's. Set up so that the next stage will be kicked off once all | |
5103 | * the percpu refcnts are confirmed to be killed. | |
5104 | * | |
5105 | * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the | |
5106 | * rest of destruction. Once all cgroup references are gone, the | |
5107 | * cgroup is RCU-freed. | |
5108 | * | |
5109 | * This function implements s1. After this step, @cgrp is gone as far as | |
5110 | * the userland is concerned and a new cgroup with the same name may be | |
5111 | * created. As cgroup doesn't care about the names internally, this | |
5112 | * doesn't cause any problem. | |
5113 | */ | |
42809dd4 TH |
5114 | static int cgroup_destroy_locked(struct cgroup *cgrp) |
5115 | __releases(&cgroup_mutex) __acquires(&cgroup_mutex) | |
ddbcc7e8 | 5116 | { |
2bd59d48 | 5117 | struct cgroup_subsys_state *css; |
1c6727af | 5118 | int ssid; |
ddbcc7e8 | 5119 | |
42809dd4 TH |
5120 | lockdep_assert_held(&cgroup_mutex); |
5121 | ||
91486f61 TH |
5122 | /* |
5123 | * Only migration can raise populated from zero and we're already | |
5124 | * holding cgroup_mutex. | |
5125 | */ | |
5126 | if (cgroup_is_populated(cgrp)) | |
ddbcc7e8 | 5127 | return -EBUSY; |
a043e3b2 | 5128 | |
bb78a92f | 5129 | /* |
d5c419b6 TH |
5130 | * Make sure there's no live children. We can't test emptiness of |
5131 | * ->self.children as dead children linger on it while being | |
5132 | * drained; otherwise, "rmdir parent/child parent" may fail. | |
bb78a92f | 5133 | */ |
f3d46500 | 5134 | if (css_has_online_children(&cgrp->self)) |
bb78a92f HD |
5135 | return -EBUSY; |
5136 | ||
455050d2 TH |
5137 | /* |
5138 | * Mark @cgrp dead. This prevents further task migration and child | |
de3f0341 | 5139 | * creation by disabling cgroup_lock_live_group(). |
455050d2 | 5140 | */ |
184faf32 | 5141 | cgrp->self.flags &= ~CSS_ONLINE; |
ddbcc7e8 | 5142 | |
249f3468 | 5143 | /* initiate massacre of all css's */ |
1c6727af TH |
5144 | for_each_css(css, ssid, cgrp) |
5145 | kill_css(css); | |
455050d2 | 5146 | |
455050d2 | 5147 | /* |
01f6474c TH |
5148 | * Remove @cgrp directory along with the base files. @cgrp has an |
5149 | * extra ref on its kn. | |
f20104de | 5150 | */ |
01f6474c | 5151 | kernfs_remove(cgrp->kn); |
f20104de | 5152 | |
d51f39b0 | 5153 | check_for_release(cgroup_parent(cgrp)); |
2bd59d48 | 5154 | |
249f3468 | 5155 | /* put the base reference */ |
9d755d33 | 5156 | percpu_ref_kill(&cgrp->self.refcnt); |
455050d2 | 5157 | |
ea15f8cc TH |
5158 | return 0; |
5159 | }; | |
5160 | ||
2bd59d48 | 5161 | static int cgroup_rmdir(struct kernfs_node *kn) |
42809dd4 | 5162 | { |
a9746d8d | 5163 | struct cgroup *cgrp; |
2bd59d48 | 5164 | int ret = 0; |
42809dd4 | 5165 | |
a9746d8d TH |
5166 | cgrp = cgroup_kn_lock_live(kn); |
5167 | if (!cgrp) | |
5168 | return 0; | |
42809dd4 | 5169 | |
a9746d8d | 5170 | ret = cgroup_destroy_locked(cgrp); |
2bb566cb | 5171 | |
a9746d8d | 5172 | cgroup_kn_unlock(kn); |
42809dd4 | 5173 | return ret; |
8e3f6541 TH |
5174 | } |
5175 | ||
2bd59d48 TH |
5176 | static struct kernfs_syscall_ops cgroup_kf_syscall_ops = { |
5177 | .remount_fs = cgroup_remount, | |
5178 | .show_options = cgroup_show_options, | |
5179 | .mkdir = cgroup_mkdir, | |
5180 | .rmdir = cgroup_rmdir, | |
5181 | .rename = cgroup_rename, | |
5182 | }; | |
5183 | ||
15a4c835 | 5184 | static void __init cgroup_init_subsys(struct cgroup_subsys *ss, bool early) |
ddbcc7e8 | 5185 | { |
ddbcc7e8 | 5186 | struct cgroup_subsys_state *css; |
cfe36bde | 5187 | |
a5ae9899 | 5188 | pr_debug("Initializing cgroup subsys %s\n", ss->name); |
ddbcc7e8 | 5189 | |
648bb56d TH |
5190 | mutex_lock(&cgroup_mutex); |
5191 | ||
15a4c835 | 5192 | idr_init(&ss->css_idr); |
0adb0704 | 5193 | INIT_LIST_HEAD(&ss->cfts); |
8e3f6541 | 5194 | |
3dd06ffa TH |
5195 | /* Create the root cgroup state for this subsystem */ |
5196 | ss->root = &cgrp_dfl_root; | |
5197 | css = ss->css_alloc(cgroup_css(&cgrp_dfl_root.cgrp, ss)); | |
ddbcc7e8 PM |
5198 | /* We don't handle early failures gracefully */ |
5199 | BUG_ON(IS_ERR(css)); | |
ddfcadab | 5200 | init_and_link_css(css, ss, &cgrp_dfl_root.cgrp); |
3b514d24 TH |
5201 | |
5202 | /* | |
5203 | * Root csses are never destroyed and we can't initialize | |
5204 | * percpu_ref during early init. Disable refcnting. | |
5205 | */ | |
5206 | css->flags |= CSS_NO_REF; | |
5207 | ||
15a4c835 | 5208 | if (early) { |
9395a450 | 5209 | /* allocation can't be done safely during early init */ |
15a4c835 TH |
5210 | css->id = 1; |
5211 | } else { | |
5212 | css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2, GFP_KERNEL); | |
5213 | BUG_ON(css->id < 0); | |
5214 | } | |
ddbcc7e8 | 5215 | |
e8d55fde | 5216 | /* Update the init_css_set to contain a subsys |
817929ec | 5217 | * pointer to this state - since the subsystem is |
e8d55fde | 5218 | * newly registered, all tasks and hence the |
3dd06ffa | 5219 | * init_css_set is in the subsystem's root cgroup. */ |
aec25020 | 5220 | init_css_set.subsys[ss->id] = css; |
ddbcc7e8 | 5221 | |
cb4a3167 AS |
5222 | have_fork_callback |= (bool)ss->fork << ss->id; |
5223 | have_exit_callback |= (bool)ss->exit << ss->id; | |
afcf6c8b | 5224 | have_free_callback |= (bool)ss->free << ss->id; |
7e47682e | 5225 | have_canfork_callback |= (bool)ss->can_fork << ss->id; |
ddbcc7e8 | 5226 | |
e8d55fde LZ |
5227 | /* At system boot, before all subsystems have been |
5228 | * registered, no tasks have been forked, so we don't | |
5229 | * need to invoke fork callbacks here. */ | |
5230 | BUG_ON(!list_empty(&init_task.tasks)); | |
5231 | ||
ae7f164a | 5232 | BUG_ON(online_css(css)); |
a8638030 | 5233 | |
cf5d5941 BB |
5234 | mutex_unlock(&cgroup_mutex); |
5235 | } | |
cf5d5941 | 5236 | |
ddbcc7e8 | 5237 | /** |
a043e3b2 LZ |
5238 | * cgroup_init_early - cgroup initialization at system boot |
5239 | * | |
5240 | * Initialize cgroups at system boot, and initialize any | |
5241 | * subsystems that request early init. | |
ddbcc7e8 PM |
5242 | */ |
5243 | int __init cgroup_init_early(void) | |
5244 | { | |
7b9a6ba5 | 5245 | static struct cgroup_sb_opts __initdata opts; |
30159ec7 | 5246 | struct cgroup_subsys *ss; |
ddbcc7e8 | 5247 | int i; |
30159ec7 | 5248 | |
3dd06ffa | 5249 | init_cgroup_root(&cgrp_dfl_root, &opts); |
3b514d24 TH |
5250 | cgrp_dfl_root.cgrp.self.flags |= CSS_NO_REF; |
5251 | ||
a4ea1cc9 | 5252 | RCU_INIT_POINTER(init_task.cgroups, &init_css_set); |
817929ec | 5253 | |
3ed80a62 | 5254 | for_each_subsys(ss, i) { |
aec25020 | 5255 | WARN(!ss->css_alloc || !ss->css_free || ss->name || ss->id, |
63253ad8 | 5256 | "invalid cgroup_subsys %d:%s css_alloc=%p css_free=%p id:name=%d:%s\n", |
073219e9 | 5257 | i, cgroup_subsys_name[i], ss->css_alloc, ss->css_free, |
aec25020 | 5258 | ss->id, ss->name); |
073219e9 TH |
5259 | WARN(strlen(cgroup_subsys_name[i]) > MAX_CGROUP_TYPE_NAMELEN, |
5260 | "cgroup_subsys_name %s too long\n", cgroup_subsys_name[i]); | |
5261 | ||
aec25020 | 5262 | ss->id = i; |
073219e9 | 5263 | ss->name = cgroup_subsys_name[i]; |
3e1d2eed TH |
5264 | if (!ss->legacy_name) |
5265 | ss->legacy_name = cgroup_subsys_name[i]; | |
ddbcc7e8 PM |
5266 | |
5267 | if (ss->early_init) | |
15a4c835 | 5268 | cgroup_init_subsys(ss, true); |
ddbcc7e8 PM |
5269 | } |
5270 | return 0; | |
5271 | } | |
5272 | ||
6e5c8307 | 5273 | static u16 cgroup_disable_mask __initdata; |
a3e72739 | 5274 | |
ddbcc7e8 | 5275 | /** |
a043e3b2 LZ |
5276 | * cgroup_init - cgroup initialization |
5277 | * | |
5278 | * Register cgroup filesystem and /proc file, and initialize | |
5279 | * any subsystems that didn't request early init. | |
ddbcc7e8 PM |
5280 | */ |
5281 | int __init cgroup_init(void) | |
5282 | { | |
30159ec7 | 5283 | struct cgroup_subsys *ss; |
035f4f51 | 5284 | int ssid; |
ddbcc7e8 | 5285 | |
6e5c8307 | 5286 | BUILD_BUG_ON(CGROUP_SUBSYS_COUNT > 16); |
1ed13287 | 5287 | BUG_ON(percpu_init_rwsem(&cgroup_threadgroup_rwsem)); |
a14c6874 TH |
5288 | BUG_ON(cgroup_init_cftypes(NULL, cgroup_dfl_base_files)); |
5289 | BUG_ON(cgroup_init_cftypes(NULL, cgroup_legacy_base_files)); | |
ddbcc7e8 | 5290 | |
54e7b4eb | 5291 | mutex_lock(&cgroup_mutex); |
54e7b4eb | 5292 | |
2378d8b8 TH |
5293 | /* |
5294 | * Add init_css_set to the hash table so that dfl_root can link to | |
5295 | * it during init. | |
5296 | */ | |
5297 | hash_add(css_set_table, &init_css_set.hlist, | |
5298 | css_set_hash(init_css_set.subsys)); | |
82fe9b0d | 5299 | |
3dd06ffa | 5300 | BUG_ON(cgroup_setup_root(&cgrp_dfl_root, 0)); |
4e96ee8e | 5301 | |
54e7b4eb TH |
5302 | mutex_unlock(&cgroup_mutex); |
5303 | ||
172a2c06 | 5304 | for_each_subsys(ss, ssid) { |
15a4c835 TH |
5305 | if (ss->early_init) { |
5306 | struct cgroup_subsys_state *css = | |
5307 | init_css_set.subsys[ss->id]; | |
5308 | ||
5309 | css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2, | |
5310 | GFP_KERNEL); | |
5311 | BUG_ON(css->id < 0); | |
5312 | } else { | |
5313 | cgroup_init_subsys(ss, false); | |
5314 | } | |
172a2c06 | 5315 | |
2d8f243a TH |
5316 | list_add_tail(&init_css_set.e_cset_node[ssid], |
5317 | &cgrp_dfl_root.cgrp.e_csets[ssid]); | |
172a2c06 TH |
5318 | |
5319 | /* | |
c731ae1d LZ |
5320 | * Setting dfl_root subsys_mask needs to consider the |
5321 | * disabled flag and cftype registration needs kmalloc, | |
5322 | * both of which aren't available during early_init. | |
172a2c06 | 5323 | */ |
a3e72739 TH |
5324 | if (cgroup_disable_mask & (1 << ssid)) { |
5325 | static_branch_disable(cgroup_subsys_enabled_key[ssid]); | |
5326 | printk(KERN_INFO "Disabling %s control group subsystem\n", | |
5327 | ss->name); | |
a8ddc821 | 5328 | continue; |
a3e72739 | 5329 | } |
a8ddc821 | 5330 | |
223ffb29 JW |
5331 | if (cgroup_ssid_no_v1(ssid)) |
5332 | printk(KERN_INFO "Disabling %s control group subsystem in v1 mounts\n", | |
5333 | ss->name); | |
5334 | ||
a8ddc821 TH |
5335 | cgrp_dfl_root.subsys_mask |= 1 << ss->id; |
5336 | ||
5de4fa13 | 5337 | if (!ss->dfl_cftypes) |
a7165264 | 5338 | cgrp_dfl_inhibit_ss_mask |= 1 << ss->id; |
5de4fa13 | 5339 | |
a8ddc821 TH |
5340 | if (ss->dfl_cftypes == ss->legacy_cftypes) { |
5341 | WARN_ON(cgroup_add_cftypes(ss, ss->dfl_cftypes)); | |
5342 | } else { | |
5343 | WARN_ON(cgroup_add_dfl_cftypes(ss, ss->dfl_cftypes)); | |
5344 | WARN_ON(cgroup_add_legacy_cftypes(ss, ss->legacy_cftypes)); | |
c731ae1d | 5345 | } |
295458e6 VD |
5346 | |
5347 | if (ss->bind) | |
5348 | ss->bind(init_css_set.subsys[ssid]); | |
676db4af GK |
5349 | } |
5350 | ||
2378d8b8 TH |
5351 | /* init_css_set.subsys[] has been updated, re-hash */ |
5352 | hash_del(&init_css_set.hlist); | |
5353 | hash_add(css_set_table, &init_css_set.hlist, | |
5354 | css_set_hash(init_css_set.subsys)); | |
5355 | ||
035f4f51 TH |
5356 | WARN_ON(sysfs_create_mount_point(fs_kobj, "cgroup")); |
5357 | WARN_ON(register_filesystem(&cgroup_fs_type)); | |
67e9c74b | 5358 | WARN_ON(register_filesystem(&cgroup2_fs_type)); |
035f4f51 | 5359 | WARN_ON(!proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations)); |
ddbcc7e8 | 5360 | |
2bd59d48 | 5361 | return 0; |
ddbcc7e8 | 5362 | } |
b4f48b63 | 5363 | |
e5fca243 TH |
5364 | static int __init cgroup_wq_init(void) |
5365 | { | |
5366 | /* | |
5367 | * There isn't much point in executing destruction path in | |
5368 | * parallel. Good chunk is serialized with cgroup_mutex anyway. | |
1a11533f | 5369 | * Use 1 for @max_active. |
e5fca243 TH |
5370 | * |
5371 | * We would prefer to do this in cgroup_init() above, but that | |
5372 | * is called before init_workqueues(): so leave this until after. | |
5373 | */ | |
1a11533f | 5374 | cgroup_destroy_wq = alloc_workqueue("cgroup_destroy", 0, 1); |
e5fca243 | 5375 | BUG_ON(!cgroup_destroy_wq); |
b1a21367 TH |
5376 | |
5377 | /* | |
5378 | * Used to destroy pidlists and separate to serve as flush domain. | |
5379 | * Cap @max_active to 1 too. | |
5380 | */ | |
5381 | cgroup_pidlist_destroy_wq = alloc_workqueue("cgroup_pidlist_destroy", | |
5382 | 0, 1); | |
5383 | BUG_ON(!cgroup_pidlist_destroy_wq); | |
5384 | ||
e5fca243 TH |
5385 | return 0; |
5386 | } | |
5387 | core_initcall(cgroup_wq_init); | |
5388 | ||
a424316c PM |
5389 | /* |
5390 | * proc_cgroup_show() | |
5391 | * - Print task's cgroup paths into seq_file, one line for each hierarchy | |
5392 | * - Used for /proc/<pid>/cgroup. | |
a424316c | 5393 | */ |
006f4ac4 ZL |
5394 | int proc_cgroup_show(struct seq_file *m, struct pid_namespace *ns, |
5395 | struct pid *pid, struct task_struct *tsk) | |
a424316c | 5396 | { |
e61734c5 | 5397 | char *buf, *path; |
a424316c | 5398 | int retval; |
3dd06ffa | 5399 | struct cgroup_root *root; |
a424316c PM |
5400 | |
5401 | retval = -ENOMEM; | |
e61734c5 | 5402 | buf = kmalloc(PATH_MAX, GFP_KERNEL); |
a424316c PM |
5403 | if (!buf) |
5404 | goto out; | |
5405 | ||
a424316c | 5406 | mutex_lock(&cgroup_mutex); |
f0d9a5f1 | 5407 | spin_lock_bh(&css_set_lock); |
a424316c | 5408 | |
985ed670 | 5409 | for_each_root(root) { |
a424316c | 5410 | struct cgroup_subsys *ss; |
bd89aabc | 5411 | struct cgroup *cgrp; |
b85d2040 | 5412 | int ssid, count = 0; |
a424316c | 5413 | |
a7165264 | 5414 | if (root == &cgrp_dfl_root && !cgrp_dfl_visible) |
985ed670 TH |
5415 | continue; |
5416 | ||
2c6ab6d2 | 5417 | seq_printf(m, "%d:", root->hierarchy_id); |
d98817d4 TH |
5418 | if (root != &cgrp_dfl_root) |
5419 | for_each_subsys(ss, ssid) | |
5420 | if (root->subsys_mask & (1 << ssid)) | |
5421 | seq_printf(m, "%s%s", count++ ? "," : "", | |
3e1d2eed | 5422 | ss->legacy_name); |
c6d57f33 PM |
5423 | if (strlen(root->name)) |
5424 | seq_printf(m, "%sname=%s", count ? "," : "", | |
5425 | root->name); | |
a424316c | 5426 | seq_putc(m, ':'); |
2e91fa7f | 5427 | |
7717f7ba | 5428 | cgrp = task_cgroup_from_root(tsk, root); |
2e91fa7f TH |
5429 | |
5430 | /* | |
5431 | * On traditional hierarchies, all zombie tasks show up as | |
5432 | * belonging to the root cgroup. On the default hierarchy, | |
5433 | * while a zombie doesn't show up in "cgroup.procs" and | |
5434 | * thus can't be migrated, its /proc/PID/cgroup keeps | |
5435 | * reporting the cgroup it belonged to before exiting. If | |
5436 | * the cgroup is removed before the zombie is reaped, | |
5437 | * " (deleted)" is appended to the cgroup path. | |
5438 | */ | |
5439 | if (cgroup_on_dfl(cgrp) || !(tsk->flags & PF_EXITING)) { | |
5440 | path = cgroup_path(cgrp, buf, PATH_MAX); | |
5441 | if (!path) { | |
5442 | retval = -ENAMETOOLONG; | |
5443 | goto out_unlock; | |
5444 | } | |
5445 | } else { | |
5446 | path = "/"; | |
e61734c5 | 5447 | } |
2e91fa7f | 5448 | |
e61734c5 | 5449 | seq_puts(m, path); |
2e91fa7f TH |
5450 | |
5451 | if (cgroup_on_dfl(cgrp) && cgroup_is_dead(cgrp)) | |
5452 | seq_puts(m, " (deleted)\n"); | |
5453 | else | |
5454 | seq_putc(m, '\n'); | |
a424316c PM |
5455 | } |
5456 | ||
006f4ac4 | 5457 | retval = 0; |
a424316c | 5458 | out_unlock: |
f0d9a5f1 | 5459 | spin_unlock_bh(&css_set_lock); |
a424316c | 5460 | mutex_unlock(&cgroup_mutex); |
a424316c PM |
5461 | kfree(buf); |
5462 | out: | |
5463 | return retval; | |
5464 | } | |
5465 | ||
a424316c PM |
5466 | /* Display information about each subsystem and each hierarchy */ |
5467 | static int proc_cgroupstats_show(struct seq_file *m, void *v) | |
5468 | { | |
30159ec7 | 5469 | struct cgroup_subsys *ss; |
a424316c | 5470 | int i; |
a424316c | 5471 | |
8bab8dde | 5472 | seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n"); |
aae8aab4 BB |
5473 | /* |
5474 | * ideally we don't want subsystems moving around while we do this. | |
5475 | * cgroup_mutex is also necessary to guarantee an atomic snapshot of | |
5476 | * subsys/hierarchy state. | |
5477 | */ | |
a424316c | 5478 | mutex_lock(&cgroup_mutex); |
30159ec7 TH |
5479 | |
5480 | for_each_subsys(ss, i) | |
2c6ab6d2 | 5481 | seq_printf(m, "%s\t%d\t%d\t%d\n", |
3e1d2eed | 5482 | ss->legacy_name, ss->root->hierarchy_id, |
fc5ed1e9 TH |
5483 | atomic_read(&ss->root->nr_cgrps), |
5484 | cgroup_ssid_enabled(i)); | |
30159ec7 | 5485 | |
a424316c PM |
5486 | mutex_unlock(&cgroup_mutex); |
5487 | return 0; | |
5488 | } | |
5489 | ||
5490 | static int cgroupstats_open(struct inode *inode, struct file *file) | |
5491 | { | |
9dce07f1 | 5492 | return single_open(file, proc_cgroupstats_show, NULL); |
a424316c PM |
5493 | } |
5494 | ||
828c0950 | 5495 | static const struct file_operations proc_cgroupstats_operations = { |
a424316c PM |
5496 | .open = cgroupstats_open, |
5497 | .read = seq_read, | |
5498 | .llseek = seq_lseek, | |
5499 | .release = single_release, | |
5500 | }; | |
5501 | ||
b4f48b63 | 5502 | /** |
eaf797ab | 5503 | * cgroup_fork - initialize cgroup related fields during copy_process() |
a043e3b2 | 5504 | * @child: pointer to task_struct of forking parent process. |
b4f48b63 | 5505 | * |
eaf797ab TH |
5506 | * A task is associated with the init_css_set until cgroup_post_fork() |
5507 | * attaches it to the parent's css_set. Empty cg_list indicates that | |
5508 | * @child isn't holding reference to its css_set. | |
b4f48b63 PM |
5509 | */ |
5510 | void cgroup_fork(struct task_struct *child) | |
5511 | { | |
eaf797ab | 5512 | RCU_INIT_POINTER(child->cgroups, &init_css_set); |
817929ec | 5513 | INIT_LIST_HEAD(&child->cg_list); |
b4f48b63 PM |
5514 | } |
5515 | ||
7e47682e AS |
5516 | /** |
5517 | * cgroup_can_fork - called on a new task before the process is exposed | |
5518 | * @child: the task in question. | |
5519 | * | |
5520 | * This calls the subsystem can_fork() callbacks. If the can_fork() callback | |
5521 | * returns an error, the fork aborts with that error code. This allows for | |
5522 | * a cgroup subsystem to conditionally allow or deny new forks. | |
5523 | */ | |
b53202e6 | 5524 | int cgroup_can_fork(struct task_struct *child) |
7e47682e AS |
5525 | { |
5526 | struct cgroup_subsys *ss; | |
5527 | int i, j, ret; | |
5528 | ||
b4e0eeaf | 5529 | do_each_subsys_mask(ss, i, have_canfork_callback) { |
b53202e6 | 5530 | ret = ss->can_fork(child); |
7e47682e AS |
5531 | if (ret) |
5532 | goto out_revert; | |
b4e0eeaf | 5533 | } while_each_subsys_mask(); |
7e47682e AS |
5534 | |
5535 | return 0; | |
5536 | ||
5537 | out_revert: | |
5538 | for_each_subsys(ss, j) { | |
5539 | if (j >= i) | |
5540 | break; | |
5541 | if (ss->cancel_fork) | |
b53202e6 | 5542 | ss->cancel_fork(child); |
7e47682e AS |
5543 | } |
5544 | ||
5545 | return ret; | |
5546 | } | |
5547 | ||
5548 | /** | |
5549 | * cgroup_cancel_fork - called if a fork failed after cgroup_can_fork() | |
5550 | * @child: the task in question | |
5551 | * | |
5552 | * This calls the cancel_fork() callbacks if a fork failed *after* | |
5553 | * cgroup_can_fork() succeded. | |
5554 | */ | |
b53202e6 | 5555 | void cgroup_cancel_fork(struct task_struct *child) |
7e47682e AS |
5556 | { |
5557 | struct cgroup_subsys *ss; | |
5558 | int i; | |
5559 | ||
5560 | for_each_subsys(ss, i) | |
5561 | if (ss->cancel_fork) | |
b53202e6 | 5562 | ss->cancel_fork(child); |
7e47682e AS |
5563 | } |
5564 | ||
817929ec | 5565 | /** |
a043e3b2 LZ |
5566 | * cgroup_post_fork - called on a new task after adding it to the task list |
5567 | * @child: the task in question | |
5568 | * | |
5edee61e TH |
5569 | * Adds the task to the list running through its css_set if necessary and |
5570 | * call the subsystem fork() callbacks. Has to be after the task is | |
5571 | * visible on the task list in case we race with the first call to | |
0942eeee | 5572 | * cgroup_task_iter_start() - to guarantee that the new task ends up on its |
5edee61e | 5573 | * list. |
a043e3b2 | 5574 | */ |
b53202e6 | 5575 | void cgroup_post_fork(struct task_struct *child) |
817929ec | 5576 | { |
30159ec7 | 5577 | struct cgroup_subsys *ss; |
5edee61e TH |
5578 | int i; |
5579 | ||
3ce3230a | 5580 | /* |
251f8c03 | 5581 | * This may race against cgroup_enable_task_cg_lists(). As that |
eaf797ab TH |
5582 | * function sets use_task_css_set_links before grabbing |
5583 | * tasklist_lock and we just went through tasklist_lock to add | |
5584 | * @child, it's guaranteed that either we see the set | |
5585 | * use_task_css_set_links or cgroup_enable_task_cg_lists() sees | |
5586 | * @child during its iteration. | |
5587 | * | |
5588 | * If we won the race, @child is associated with %current's | |
f0d9a5f1 | 5589 | * css_set. Grabbing css_set_lock guarantees both that the |
eaf797ab TH |
5590 | * association is stable, and, on completion of the parent's |
5591 | * migration, @child is visible in the source of migration or | |
5592 | * already in the destination cgroup. This guarantee is necessary | |
5593 | * when implementing operations which need to migrate all tasks of | |
5594 | * a cgroup to another. | |
5595 | * | |
251f8c03 | 5596 | * Note that if we lose to cgroup_enable_task_cg_lists(), @child |
eaf797ab TH |
5597 | * will remain in init_css_set. This is safe because all tasks are |
5598 | * in the init_css_set before cg_links is enabled and there's no | |
5599 | * operation which transfers all tasks out of init_css_set. | |
3ce3230a | 5600 | */ |
817929ec | 5601 | if (use_task_css_set_links) { |
eaf797ab TH |
5602 | struct css_set *cset; |
5603 | ||
f0d9a5f1 | 5604 | spin_lock_bh(&css_set_lock); |
0e1d768f | 5605 | cset = task_css_set(current); |
eaf797ab | 5606 | if (list_empty(&child->cg_list)) { |
eaf797ab | 5607 | get_css_set(cset); |
f6d7d049 | 5608 | css_set_move_task(child, NULL, cset, false); |
eaf797ab | 5609 | } |
f0d9a5f1 | 5610 | spin_unlock_bh(&css_set_lock); |
817929ec | 5611 | } |
5edee61e TH |
5612 | |
5613 | /* | |
5614 | * Call ss->fork(). This must happen after @child is linked on | |
5615 | * css_set; otherwise, @child might change state between ->fork() | |
5616 | * and addition to css_set. | |
5617 | */ | |
b4e0eeaf | 5618 | do_each_subsys_mask(ss, i, have_fork_callback) { |
b53202e6 | 5619 | ss->fork(child); |
b4e0eeaf | 5620 | } while_each_subsys_mask(); |
817929ec | 5621 | } |
5edee61e | 5622 | |
b4f48b63 PM |
5623 | /** |
5624 | * cgroup_exit - detach cgroup from exiting task | |
5625 | * @tsk: pointer to task_struct of exiting process | |
5626 | * | |
5627 | * Description: Detach cgroup from @tsk and release it. | |
5628 | * | |
5629 | * Note that cgroups marked notify_on_release force every task in | |
5630 | * them to take the global cgroup_mutex mutex when exiting. | |
5631 | * This could impact scaling on very large systems. Be reluctant to | |
5632 | * use notify_on_release cgroups where very high task exit scaling | |
5633 | * is required on large systems. | |
5634 | * | |
0e1d768f TH |
5635 | * We set the exiting tasks cgroup to the root cgroup (top_cgroup). We |
5636 | * call cgroup_exit() while the task is still competent to handle | |
5637 | * notify_on_release(), then leave the task attached to the root cgroup in | |
5638 | * each hierarchy for the remainder of its exit. No need to bother with | |
5639 | * init_css_set refcnting. init_css_set never goes away and we can't race | |
e8604cb4 | 5640 | * with migration path - PF_EXITING is visible to migration path. |
b4f48b63 | 5641 | */ |
1ec41830 | 5642 | void cgroup_exit(struct task_struct *tsk) |
b4f48b63 | 5643 | { |
30159ec7 | 5644 | struct cgroup_subsys *ss; |
5abb8855 | 5645 | struct css_set *cset; |
d41d5a01 | 5646 | int i; |
817929ec PM |
5647 | |
5648 | /* | |
0e1d768f | 5649 | * Unlink from @tsk from its css_set. As migration path can't race |
0de0942d | 5650 | * with us, we can check css_set and cg_list without synchronization. |
817929ec | 5651 | */ |
0de0942d TH |
5652 | cset = task_css_set(tsk); |
5653 | ||
817929ec | 5654 | if (!list_empty(&tsk->cg_list)) { |
f0d9a5f1 | 5655 | spin_lock_bh(&css_set_lock); |
f6d7d049 | 5656 | css_set_move_task(tsk, cset, NULL, false); |
f0d9a5f1 | 5657 | spin_unlock_bh(&css_set_lock); |
2e91fa7f TH |
5658 | } else { |
5659 | get_css_set(cset); | |
817929ec PM |
5660 | } |
5661 | ||
cb4a3167 | 5662 | /* see cgroup_post_fork() for details */ |
b4e0eeaf | 5663 | do_each_subsys_mask(ss, i, have_exit_callback) { |
2e91fa7f | 5664 | ss->exit(tsk); |
b4e0eeaf | 5665 | } while_each_subsys_mask(); |
2e91fa7f | 5666 | } |
30159ec7 | 5667 | |
2e91fa7f TH |
5668 | void cgroup_free(struct task_struct *task) |
5669 | { | |
5670 | struct css_set *cset = task_css_set(task); | |
afcf6c8b TH |
5671 | struct cgroup_subsys *ss; |
5672 | int ssid; | |
5673 | ||
b4e0eeaf | 5674 | do_each_subsys_mask(ss, ssid, have_free_callback) { |
afcf6c8b | 5675 | ss->free(task); |
b4e0eeaf | 5676 | } while_each_subsys_mask(); |
d41d5a01 | 5677 | |
2e91fa7f | 5678 | put_css_set(cset); |
b4f48b63 | 5679 | } |
697f4161 | 5680 | |
bd89aabc | 5681 | static void check_for_release(struct cgroup *cgrp) |
81a6a5cd | 5682 | { |
27bd4dbb | 5683 | if (notify_on_release(cgrp) && !cgroup_is_populated(cgrp) && |
971ff493 ZL |
5684 | !css_has_online_children(&cgrp->self) && !cgroup_is_dead(cgrp)) |
5685 | schedule_work(&cgrp->release_agent_work); | |
81a6a5cd PM |
5686 | } |
5687 | ||
81a6a5cd PM |
5688 | /* |
5689 | * Notify userspace when a cgroup is released, by running the | |
5690 | * configured release agent with the name of the cgroup (path | |
5691 | * relative to the root of cgroup file system) as the argument. | |
5692 | * | |
5693 | * Most likely, this user command will try to rmdir this cgroup. | |
5694 | * | |
5695 | * This races with the possibility that some other task will be | |
5696 | * attached to this cgroup before it is removed, or that some other | |
5697 | * user task will 'mkdir' a child cgroup of this cgroup. That's ok. | |
5698 | * The presumed 'rmdir' will fail quietly if this cgroup is no longer | |
5699 | * unused, and this cgroup will be reprieved from its death sentence, | |
5700 | * to continue to serve a useful existence. Next time it's released, | |
5701 | * we will get notified again, if it still has 'notify_on_release' set. | |
5702 | * | |
5703 | * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which | |
5704 | * means only wait until the task is successfully execve()'d. The | |
5705 | * separate release agent task is forked by call_usermodehelper(), | |
5706 | * then control in this thread returns here, without waiting for the | |
5707 | * release agent task. We don't bother to wait because the caller of | |
5708 | * this routine has no use for the exit status of the release agent | |
5709 | * task, so no sense holding our caller up for that. | |
81a6a5cd | 5710 | */ |
81a6a5cd PM |
5711 | static void cgroup_release_agent(struct work_struct *work) |
5712 | { | |
971ff493 ZL |
5713 | struct cgroup *cgrp = |
5714 | container_of(work, struct cgroup, release_agent_work); | |
5715 | char *pathbuf = NULL, *agentbuf = NULL, *path; | |
5716 | char *argv[3], *envp[3]; | |
5717 | ||
81a6a5cd | 5718 | mutex_lock(&cgroup_mutex); |
971ff493 ZL |
5719 | |
5720 | pathbuf = kmalloc(PATH_MAX, GFP_KERNEL); | |
5721 | agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL); | |
5722 | if (!pathbuf || !agentbuf) | |
5723 | goto out; | |
5724 | ||
5725 | path = cgroup_path(cgrp, pathbuf, PATH_MAX); | |
5726 | if (!path) | |
5727 | goto out; | |
5728 | ||
5729 | argv[0] = agentbuf; | |
5730 | argv[1] = path; | |
5731 | argv[2] = NULL; | |
5732 | ||
5733 | /* minimal command environment */ | |
5734 | envp[0] = "HOME=/"; | |
5735 | envp[1] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin"; | |
5736 | envp[2] = NULL; | |
5737 | ||
81a6a5cd | 5738 | mutex_unlock(&cgroup_mutex); |
971ff493 | 5739 | call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC); |
3e2cd91a | 5740 | goto out_free; |
971ff493 | 5741 | out: |
81a6a5cd | 5742 | mutex_unlock(&cgroup_mutex); |
3e2cd91a | 5743 | out_free: |
971ff493 ZL |
5744 | kfree(agentbuf); |
5745 | kfree(pathbuf); | |
81a6a5cd | 5746 | } |
8bab8dde PM |
5747 | |
5748 | static int __init cgroup_disable(char *str) | |
5749 | { | |
30159ec7 | 5750 | struct cgroup_subsys *ss; |
8bab8dde | 5751 | char *token; |
30159ec7 | 5752 | int i; |
8bab8dde PM |
5753 | |
5754 | while ((token = strsep(&str, ",")) != NULL) { | |
5755 | if (!*token) | |
5756 | continue; | |
be45c900 | 5757 | |
3ed80a62 | 5758 | for_each_subsys(ss, i) { |
3e1d2eed TH |
5759 | if (strcmp(token, ss->name) && |
5760 | strcmp(token, ss->legacy_name)) | |
5761 | continue; | |
a3e72739 | 5762 | cgroup_disable_mask |= 1 << i; |
8bab8dde PM |
5763 | } |
5764 | } | |
5765 | return 1; | |
5766 | } | |
5767 | __setup("cgroup_disable=", cgroup_disable); | |
38460b48 | 5768 | |
223ffb29 JW |
5769 | static int __init cgroup_no_v1(char *str) |
5770 | { | |
5771 | struct cgroup_subsys *ss; | |
5772 | char *token; | |
5773 | int i; | |
5774 | ||
5775 | while ((token = strsep(&str, ",")) != NULL) { | |
5776 | if (!*token) | |
5777 | continue; | |
5778 | ||
5779 | if (!strcmp(token, "all")) { | |
6e5c8307 | 5780 | cgroup_no_v1_mask = U16_MAX; |
223ffb29 JW |
5781 | break; |
5782 | } | |
5783 | ||
5784 | for_each_subsys(ss, i) { | |
5785 | if (strcmp(token, ss->name) && | |
5786 | strcmp(token, ss->legacy_name)) | |
5787 | continue; | |
5788 | ||
5789 | cgroup_no_v1_mask |= 1 << i; | |
5790 | } | |
5791 | } | |
5792 | return 1; | |
5793 | } | |
5794 | __setup("cgroup_no_v1=", cgroup_no_v1); | |
5795 | ||
b77d7b60 | 5796 | /** |
ec903c0c | 5797 | * css_tryget_online_from_dir - get corresponding css from a cgroup dentry |
35cf0836 TH |
5798 | * @dentry: directory dentry of interest |
5799 | * @ss: subsystem of interest | |
b77d7b60 | 5800 | * |
5a17f543 TH |
5801 | * If @dentry is a directory for a cgroup which has @ss enabled on it, try |
5802 | * to get the corresponding css and return it. If such css doesn't exist | |
5803 | * or can't be pinned, an ERR_PTR value is returned. | |
e5d1367f | 5804 | */ |
ec903c0c TH |
5805 | struct cgroup_subsys_state *css_tryget_online_from_dir(struct dentry *dentry, |
5806 | struct cgroup_subsys *ss) | |
e5d1367f | 5807 | { |
2bd59d48 | 5808 | struct kernfs_node *kn = kernfs_node_from_dentry(dentry); |
f17fc25f | 5809 | struct file_system_type *s_type = dentry->d_sb->s_type; |
2bd59d48 | 5810 | struct cgroup_subsys_state *css = NULL; |
e5d1367f | 5811 | struct cgroup *cgrp; |
e5d1367f | 5812 | |
35cf0836 | 5813 | /* is @dentry a cgroup dir? */ |
f17fc25f TH |
5814 | if ((s_type != &cgroup_fs_type && s_type != &cgroup2_fs_type) || |
5815 | !kn || kernfs_type(kn) != KERNFS_DIR) | |
e5d1367f SE |
5816 | return ERR_PTR(-EBADF); |
5817 | ||
5a17f543 TH |
5818 | rcu_read_lock(); |
5819 | ||
2bd59d48 TH |
5820 | /* |
5821 | * This path doesn't originate from kernfs and @kn could already | |
5822 | * have been or be removed at any point. @kn->priv is RCU | |
a4189487 | 5823 | * protected for this access. See css_release_work_fn() for details. |
2bd59d48 TH |
5824 | */ |
5825 | cgrp = rcu_dereference(kn->priv); | |
5826 | if (cgrp) | |
5827 | css = cgroup_css(cgrp, ss); | |
5a17f543 | 5828 | |
ec903c0c | 5829 | if (!css || !css_tryget_online(css)) |
5a17f543 TH |
5830 | css = ERR_PTR(-ENOENT); |
5831 | ||
5832 | rcu_read_unlock(); | |
5833 | return css; | |
e5d1367f | 5834 | } |
e5d1367f | 5835 | |
1cb650b9 LZ |
5836 | /** |
5837 | * css_from_id - lookup css by id | |
5838 | * @id: the cgroup id | |
5839 | * @ss: cgroup subsys to be looked into | |
5840 | * | |
5841 | * Returns the css if there's valid one with @id, otherwise returns NULL. | |
5842 | * Should be called under rcu_read_lock(). | |
5843 | */ | |
5844 | struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss) | |
5845 | { | |
6fa4918d | 5846 | WARN_ON_ONCE(!rcu_read_lock_held()); |
adbe427b | 5847 | return id > 0 ? idr_find(&ss->css_idr, id) : NULL; |
e5d1367f SE |
5848 | } |
5849 | ||
16af4396 TH |
5850 | /** |
5851 | * cgroup_get_from_path - lookup and get a cgroup from its default hierarchy path | |
5852 | * @path: path on the default hierarchy | |
5853 | * | |
5854 | * Find the cgroup at @path on the default hierarchy, increment its | |
5855 | * reference count and return it. Returns pointer to the found cgroup on | |
5856 | * success, ERR_PTR(-ENOENT) if @path doens't exist and ERR_PTR(-ENOTDIR) | |
5857 | * if @path points to a non-directory. | |
5858 | */ | |
5859 | struct cgroup *cgroup_get_from_path(const char *path) | |
5860 | { | |
5861 | struct kernfs_node *kn; | |
5862 | struct cgroup *cgrp; | |
5863 | ||
5864 | mutex_lock(&cgroup_mutex); | |
5865 | ||
5866 | kn = kernfs_walk_and_get(cgrp_dfl_root.cgrp.kn, path); | |
5867 | if (kn) { | |
5868 | if (kernfs_type(kn) == KERNFS_DIR) { | |
5869 | cgrp = kn->priv; | |
5870 | cgroup_get(cgrp); | |
5871 | } else { | |
5872 | cgrp = ERR_PTR(-ENOTDIR); | |
5873 | } | |
5874 | kernfs_put(kn); | |
5875 | } else { | |
5876 | cgrp = ERR_PTR(-ENOENT); | |
5877 | } | |
5878 | ||
5879 | mutex_unlock(&cgroup_mutex); | |
5880 | return cgrp; | |
5881 | } | |
5882 | EXPORT_SYMBOL_GPL(cgroup_get_from_path); | |
5883 | ||
bd1060a1 TH |
5884 | /* |
5885 | * sock->sk_cgrp_data handling. For more info, see sock_cgroup_data | |
5886 | * definition in cgroup-defs.h. | |
5887 | */ | |
5888 | #ifdef CONFIG_SOCK_CGROUP_DATA | |
5889 | ||
5890 | #if defined(CONFIG_CGROUP_NET_PRIO) || defined(CONFIG_CGROUP_NET_CLASSID) | |
5891 | ||
3fa4cc9c | 5892 | DEFINE_SPINLOCK(cgroup_sk_update_lock); |
bd1060a1 TH |
5893 | static bool cgroup_sk_alloc_disabled __read_mostly; |
5894 | ||
5895 | void cgroup_sk_alloc_disable(void) | |
5896 | { | |
5897 | if (cgroup_sk_alloc_disabled) | |
5898 | return; | |
5899 | pr_info("cgroup: disabling cgroup2 socket matching due to net_prio or net_cls activation\n"); | |
5900 | cgroup_sk_alloc_disabled = true; | |
5901 | } | |
5902 | ||
5903 | #else | |
5904 | ||
5905 | #define cgroup_sk_alloc_disabled false | |
5906 | ||
5907 | #endif | |
5908 | ||
5909 | void cgroup_sk_alloc(struct sock_cgroup_data *skcd) | |
5910 | { | |
5911 | if (cgroup_sk_alloc_disabled) | |
5912 | return; | |
5913 | ||
5914 | rcu_read_lock(); | |
5915 | ||
5916 | while (true) { | |
5917 | struct css_set *cset; | |
5918 | ||
5919 | cset = task_css_set(current); | |
5920 | if (likely(cgroup_tryget(cset->dfl_cgrp))) { | |
5921 | skcd->val = (unsigned long)cset->dfl_cgrp; | |
5922 | break; | |
5923 | } | |
5924 | cpu_relax(); | |
5925 | } | |
5926 | ||
5927 | rcu_read_unlock(); | |
5928 | } | |
5929 | ||
5930 | void cgroup_sk_free(struct sock_cgroup_data *skcd) | |
5931 | { | |
5932 | cgroup_put(sock_cgroup_ptr(skcd)); | |
5933 | } | |
5934 | ||
5935 | #endif /* CONFIG_SOCK_CGROUP_DATA */ | |
5936 | ||
fe693435 | 5937 | #ifdef CONFIG_CGROUP_DEBUG |
eb95419b TH |
5938 | static struct cgroup_subsys_state * |
5939 | debug_css_alloc(struct cgroup_subsys_state *parent_css) | |
fe693435 PM |
5940 | { |
5941 | struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL); | |
5942 | ||
5943 | if (!css) | |
5944 | return ERR_PTR(-ENOMEM); | |
5945 | ||
5946 | return css; | |
5947 | } | |
5948 | ||
eb95419b | 5949 | static void debug_css_free(struct cgroup_subsys_state *css) |
fe693435 | 5950 | { |
eb95419b | 5951 | kfree(css); |
fe693435 PM |
5952 | } |
5953 | ||
182446d0 TH |
5954 | static u64 debug_taskcount_read(struct cgroup_subsys_state *css, |
5955 | struct cftype *cft) | |
fe693435 | 5956 | { |
182446d0 | 5957 | return cgroup_task_count(css->cgroup); |
fe693435 PM |
5958 | } |
5959 | ||
182446d0 TH |
5960 | static u64 current_css_set_read(struct cgroup_subsys_state *css, |
5961 | struct cftype *cft) | |
fe693435 PM |
5962 | { |
5963 | return (u64)(unsigned long)current->cgroups; | |
5964 | } | |
5965 | ||
182446d0 | 5966 | static u64 current_css_set_refcount_read(struct cgroup_subsys_state *css, |
03c78cbe | 5967 | struct cftype *cft) |
fe693435 PM |
5968 | { |
5969 | u64 count; | |
5970 | ||
5971 | rcu_read_lock(); | |
a8ad805c | 5972 | count = atomic_read(&task_css_set(current)->refcount); |
fe693435 PM |
5973 | rcu_read_unlock(); |
5974 | return count; | |
5975 | } | |
5976 | ||
2da8ca82 | 5977 | static int current_css_set_cg_links_read(struct seq_file *seq, void *v) |
7717f7ba | 5978 | { |
69d0206c | 5979 | struct cgrp_cset_link *link; |
5abb8855 | 5980 | struct css_set *cset; |
e61734c5 TH |
5981 | char *name_buf; |
5982 | ||
5983 | name_buf = kmalloc(NAME_MAX + 1, GFP_KERNEL); | |
5984 | if (!name_buf) | |
5985 | return -ENOMEM; | |
7717f7ba | 5986 | |
f0d9a5f1 | 5987 | spin_lock_bh(&css_set_lock); |
7717f7ba | 5988 | rcu_read_lock(); |
5abb8855 | 5989 | cset = rcu_dereference(current->cgroups); |
69d0206c | 5990 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) { |
7717f7ba | 5991 | struct cgroup *c = link->cgrp; |
7717f7ba | 5992 | |
a2dd4247 | 5993 | cgroup_name(c, name_buf, NAME_MAX + 1); |
2c6ab6d2 | 5994 | seq_printf(seq, "Root %d group %s\n", |
a2dd4247 | 5995 | c->root->hierarchy_id, name_buf); |
7717f7ba PM |
5996 | } |
5997 | rcu_read_unlock(); | |
f0d9a5f1 | 5998 | spin_unlock_bh(&css_set_lock); |
e61734c5 | 5999 | kfree(name_buf); |
7717f7ba PM |
6000 | return 0; |
6001 | } | |
6002 | ||
6003 | #define MAX_TASKS_SHOWN_PER_CSS 25 | |
2da8ca82 | 6004 | static int cgroup_css_links_read(struct seq_file *seq, void *v) |
7717f7ba | 6005 | { |
2da8ca82 | 6006 | struct cgroup_subsys_state *css = seq_css(seq); |
69d0206c | 6007 | struct cgrp_cset_link *link; |
7717f7ba | 6008 | |
f0d9a5f1 | 6009 | spin_lock_bh(&css_set_lock); |
182446d0 | 6010 | list_for_each_entry(link, &css->cgroup->cset_links, cset_link) { |
69d0206c | 6011 | struct css_set *cset = link->cset; |
7717f7ba PM |
6012 | struct task_struct *task; |
6013 | int count = 0; | |
c7561128 | 6014 | |
5abb8855 | 6015 | seq_printf(seq, "css_set %p\n", cset); |
c7561128 | 6016 | |
5abb8855 | 6017 | list_for_each_entry(task, &cset->tasks, cg_list) { |
c7561128 TH |
6018 | if (count++ > MAX_TASKS_SHOWN_PER_CSS) |
6019 | goto overflow; | |
6020 | seq_printf(seq, " task %d\n", task_pid_vnr(task)); | |
6021 | } | |
6022 | ||
6023 | list_for_each_entry(task, &cset->mg_tasks, cg_list) { | |
6024 | if (count++ > MAX_TASKS_SHOWN_PER_CSS) | |
6025 | goto overflow; | |
6026 | seq_printf(seq, " task %d\n", task_pid_vnr(task)); | |
7717f7ba | 6027 | } |
c7561128 TH |
6028 | continue; |
6029 | overflow: | |
6030 | seq_puts(seq, " ...\n"); | |
7717f7ba | 6031 | } |
f0d9a5f1 | 6032 | spin_unlock_bh(&css_set_lock); |
7717f7ba PM |
6033 | return 0; |
6034 | } | |
6035 | ||
182446d0 | 6036 | static u64 releasable_read(struct cgroup_subsys_state *css, struct cftype *cft) |
fe693435 | 6037 | { |
27bd4dbb | 6038 | return (!cgroup_is_populated(css->cgroup) && |
a25eb52e | 6039 | !css_has_online_children(&css->cgroup->self)); |
fe693435 PM |
6040 | } |
6041 | ||
6042 | static struct cftype debug_files[] = { | |
fe693435 PM |
6043 | { |
6044 | .name = "taskcount", | |
6045 | .read_u64 = debug_taskcount_read, | |
6046 | }, | |
6047 | ||
6048 | { | |
6049 | .name = "current_css_set", | |
6050 | .read_u64 = current_css_set_read, | |
6051 | }, | |
6052 | ||
6053 | { | |
6054 | .name = "current_css_set_refcount", | |
6055 | .read_u64 = current_css_set_refcount_read, | |
6056 | }, | |
6057 | ||
7717f7ba PM |
6058 | { |
6059 | .name = "current_css_set_cg_links", | |
2da8ca82 | 6060 | .seq_show = current_css_set_cg_links_read, |
7717f7ba PM |
6061 | }, |
6062 | ||
6063 | { | |
6064 | .name = "cgroup_css_links", | |
2da8ca82 | 6065 | .seq_show = cgroup_css_links_read, |
7717f7ba PM |
6066 | }, |
6067 | ||
fe693435 PM |
6068 | { |
6069 | .name = "releasable", | |
6070 | .read_u64 = releasable_read, | |
6071 | }, | |
fe693435 | 6072 | |
4baf6e33 TH |
6073 | { } /* terminate */ |
6074 | }; | |
fe693435 | 6075 | |
073219e9 | 6076 | struct cgroup_subsys debug_cgrp_subsys = { |
92fb9748 TH |
6077 | .css_alloc = debug_css_alloc, |
6078 | .css_free = debug_css_free, | |
5577964e | 6079 | .legacy_cftypes = debug_files, |
fe693435 PM |
6080 | }; |
6081 | #endif /* CONFIG_CGROUP_DEBUG */ |