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> |
a79a908f AK |
62 | #include <linux/proc_ns.h> |
63 | #include <linux/nsproxy.h> | |
64 | #include <linux/proc_ns.h> | |
bd1060a1 | 65 | #include <net/sock.h> |
ddbcc7e8 | 66 | |
b1a21367 TH |
67 | /* |
68 | * pidlists linger the following amount before being destroyed. The goal | |
69 | * is avoiding frequent destruction in the middle of consecutive read calls | |
70 | * Expiring in the middle is a performance problem not a correctness one. | |
71 | * 1 sec should be enough. | |
72 | */ | |
73 | #define CGROUP_PIDLIST_DESTROY_DELAY HZ | |
74 | ||
8d7e6fb0 TH |
75 | #define CGROUP_FILE_NAME_MAX (MAX_CGROUP_TYPE_NAMELEN + \ |
76 | MAX_CFTYPE_NAME + 2) | |
77 | ||
e25e2cbb TH |
78 | /* |
79 | * cgroup_mutex is the master lock. Any modification to cgroup or its | |
80 | * hierarchy must be performed while holding it. | |
81 | * | |
f0d9a5f1 | 82 | * css_set_lock protects task->cgroups pointer, the list of css_set |
0e1d768f | 83 | * objects, and the chain of tasks off each css_set. |
e25e2cbb | 84 | * |
0e1d768f TH |
85 | * These locks are exported if CONFIG_PROVE_RCU so that accessors in |
86 | * cgroup.h can use them for lockdep annotations. | |
e25e2cbb | 87 | */ |
2219449a TH |
88 | #ifdef CONFIG_PROVE_RCU |
89 | DEFINE_MUTEX(cgroup_mutex); | |
f0d9a5f1 | 90 | DEFINE_SPINLOCK(css_set_lock); |
0e1d768f | 91 | EXPORT_SYMBOL_GPL(cgroup_mutex); |
f0d9a5f1 | 92 | EXPORT_SYMBOL_GPL(css_set_lock); |
2219449a | 93 | #else |
81a6a5cd | 94 | static DEFINE_MUTEX(cgroup_mutex); |
f0d9a5f1 | 95 | static DEFINE_SPINLOCK(css_set_lock); |
2219449a TH |
96 | #endif |
97 | ||
6fa4918d | 98 | /* |
15a4c835 TH |
99 | * Protects cgroup_idr and css_idr so that IDs can be released without |
100 | * grabbing cgroup_mutex. | |
6fa4918d TH |
101 | */ |
102 | static DEFINE_SPINLOCK(cgroup_idr_lock); | |
103 | ||
34c06254 TH |
104 | /* |
105 | * Protects cgroup_file->kn for !self csses. It synchronizes notifications | |
106 | * against file removal/re-creation across css hiding. | |
107 | */ | |
108 | static DEFINE_SPINLOCK(cgroup_file_kn_lock); | |
109 | ||
69e943b7 TH |
110 | /* |
111 | * Protects cgroup_subsys->release_agent_path. Modifying it also requires | |
112 | * cgroup_mutex. Reading requires either cgroup_mutex or this spinlock. | |
113 | */ | |
114 | static DEFINE_SPINLOCK(release_agent_path_lock); | |
81a6a5cd | 115 | |
1ed13287 TH |
116 | struct percpu_rw_semaphore cgroup_threadgroup_rwsem; |
117 | ||
8353da1f | 118 | #define cgroup_assert_mutex_or_rcu_locked() \ |
f78f5b90 PM |
119 | RCU_LOCKDEP_WARN(!rcu_read_lock_held() && \ |
120 | !lockdep_is_held(&cgroup_mutex), \ | |
8353da1f | 121 | "cgroup_mutex or RCU read lock required"); |
780cd8b3 | 122 | |
e5fca243 TH |
123 | /* |
124 | * cgroup destruction makes heavy use of work items and there can be a lot | |
125 | * of concurrent destructions. Use a separate workqueue so that cgroup | |
126 | * destruction work items don't end up filling up max_active of system_wq | |
127 | * which may lead to deadlock. | |
128 | */ | |
129 | static struct workqueue_struct *cgroup_destroy_wq; | |
130 | ||
b1a21367 TH |
131 | /* |
132 | * pidlist destructions need to be flushed on cgroup destruction. Use a | |
133 | * separate workqueue as flush domain. | |
134 | */ | |
135 | static struct workqueue_struct *cgroup_pidlist_destroy_wq; | |
136 | ||
3ed80a62 | 137 | /* generate an array of cgroup subsystem pointers */ |
073219e9 | 138 | #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys, |
3ed80a62 | 139 | static struct cgroup_subsys *cgroup_subsys[] = { |
ddbcc7e8 PM |
140 | #include <linux/cgroup_subsys.h> |
141 | }; | |
073219e9 TH |
142 | #undef SUBSYS |
143 | ||
144 | /* array of cgroup subsystem names */ | |
145 | #define SUBSYS(_x) [_x ## _cgrp_id] = #_x, | |
146 | static const char *cgroup_subsys_name[] = { | |
ddbcc7e8 PM |
147 | #include <linux/cgroup_subsys.h> |
148 | }; | |
073219e9 | 149 | #undef SUBSYS |
ddbcc7e8 | 150 | |
49d1dc4b TH |
151 | /* array of static_keys for cgroup_subsys_enabled() and cgroup_subsys_on_dfl() */ |
152 | #define SUBSYS(_x) \ | |
153 | DEFINE_STATIC_KEY_TRUE(_x ## _cgrp_subsys_enabled_key); \ | |
154 | DEFINE_STATIC_KEY_TRUE(_x ## _cgrp_subsys_on_dfl_key); \ | |
155 | EXPORT_SYMBOL_GPL(_x ## _cgrp_subsys_enabled_key); \ | |
156 | EXPORT_SYMBOL_GPL(_x ## _cgrp_subsys_on_dfl_key); | |
157 | #include <linux/cgroup_subsys.h> | |
158 | #undef SUBSYS | |
159 | ||
160 | #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys_enabled_key, | |
161 | static struct static_key_true *cgroup_subsys_enabled_key[] = { | |
162 | #include <linux/cgroup_subsys.h> | |
163 | }; | |
164 | #undef SUBSYS | |
165 | ||
166 | #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys_on_dfl_key, | |
167 | static struct static_key_true *cgroup_subsys_on_dfl_key[] = { | |
168 | #include <linux/cgroup_subsys.h> | |
169 | }; | |
170 | #undef SUBSYS | |
171 | ||
ddbcc7e8 | 172 | /* |
3dd06ffa | 173 | * The default hierarchy, reserved for the subsystems that are otherwise |
9871bf95 TH |
174 | * unattached - it never has more than a single cgroup, and all tasks are |
175 | * part of that cgroup. | |
ddbcc7e8 | 176 | */ |
a2dd4247 | 177 | struct cgroup_root cgrp_dfl_root; |
d0ec4230 | 178 | EXPORT_SYMBOL_GPL(cgrp_dfl_root); |
9871bf95 | 179 | |
a2dd4247 TH |
180 | /* |
181 | * The default hierarchy always exists but is hidden until mounted for the | |
182 | * first time. This is for backward compatibility. | |
183 | */ | |
184 | static bool cgrp_dfl_root_visible; | |
ddbcc7e8 | 185 | |
223ffb29 JW |
186 | /* Controllers blocked by the commandline in v1 */ |
187 | static unsigned long cgroup_no_v1_mask; | |
188 | ||
5533e011 | 189 | /* some controllers are not supported in the default hierarchy */ |
8ab456ac | 190 | static unsigned long cgrp_dfl_root_inhibit_ss_mask; |
5533e011 | 191 | |
ddbcc7e8 PM |
192 | /* The list of hierarchy roots */ |
193 | ||
9871bf95 TH |
194 | static LIST_HEAD(cgroup_roots); |
195 | static int cgroup_root_count; | |
ddbcc7e8 | 196 | |
3417ae1f | 197 | /* hierarchy ID allocation and mapping, protected by cgroup_mutex */ |
1a574231 | 198 | static DEFINE_IDR(cgroup_hierarchy_idr); |
2c6ab6d2 | 199 | |
794611a1 | 200 | /* |
0cb51d71 TH |
201 | * Assign a monotonically increasing serial number to csses. It guarantees |
202 | * cgroups with bigger numbers are newer than those with smaller numbers. | |
203 | * Also, as csses are always appended to the parent's ->children list, it | |
204 | * guarantees that sibling csses are always sorted in the ascending serial | |
205 | * number order on the list. Protected by cgroup_mutex. | |
794611a1 | 206 | */ |
0cb51d71 | 207 | static u64 css_serial_nr_next = 1; |
794611a1 | 208 | |
cb4a3167 AS |
209 | /* |
210 | * These bitmask flags indicate whether tasks in the fork and exit paths have | |
211 | * fork/exit handlers to call. This avoids us having to do extra work in the | |
212 | * fork/exit path to check which subsystems have fork/exit callbacks. | |
ddbcc7e8 | 213 | */ |
cb4a3167 AS |
214 | static unsigned long have_fork_callback __read_mostly; |
215 | static unsigned long have_exit_callback __read_mostly; | |
afcf6c8b | 216 | static unsigned long have_free_callback __read_mostly; |
ddbcc7e8 | 217 | |
a79a908f AK |
218 | /* cgroup namespace for init task */ |
219 | struct cgroup_namespace init_cgroup_ns = { | |
220 | .count = { .counter = 2, }, | |
221 | .user_ns = &init_user_ns, | |
222 | .ns.ops = &cgroupns_operations, | |
223 | .ns.inum = PROC_CGROUP_INIT_INO, | |
224 | .root_cset = &init_css_set, | |
225 | }; | |
226 | ||
7e47682e AS |
227 | /* Ditto for the can_fork callback. */ |
228 | static unsigned long have_canfork_callback __read_mostly; | |
229 | ||
67e9c74b | 230 | static struct file_system_type cgroup2_fs_type; |
a14c6874 TH |
231 | static struct cftype cgroup_dfl_base_files[]; |
232 | static struct cftype cgroup_legacy_base_files[]; | |
628f7cd4 | 233 | |
3dd06ffa | 234 | static int rebind_subsystems(struct cgroup_root *dst_root, |
8ab456ac | 235 | unsigned long ss_mask); |
ed27b9f7 | 236 | static void css_task_iter_advance(struct css_task_iter *it); |
42809dd4 | 237 | static int cgroup_destroy_locked(struct cgroup *cgrp); |
f63070d3 TH |
238 | static int create_css(struct cgroup *cgrp, struct cgroup_subsys *ss, |
239 | bool visible); | |
9d755d33 | 240 | static void css_release(struct percpu_ref *ref); |
f8f22e53 | 241 | static void kill_css(struct cgroup_subsys_state *css); |
4df8dc90 TH |
242 | static int cgroup_addrm_files(struct cgroup_subsys_state *css, |
243 | struct cgroup *cgrp, struct cftype cfts[], | |
2bb566cb | 244 | bool is_add); |
42809dd4 | 245 | |
fc5ed1e9 TH |
246 | /** |
247 | * cgroup_ssid_enabled - cgroup subsys enabled test by subsys ID | |
248 | * @ssid: subsys ID of interest | |
249 | * | |
250 | * cgroup_subsys_enabled() can only be used with literal subsys names which | |
251 | * is fine for individual subsystems but unsuitable for cgroup core. This | |
252 | * is slower static_key_enabled() based test indexed by @ssid. | |
253 | */ | |
254 | static bool cgroup_ssid_enabled(int ssid) | |
255 | { | |
256 | return static_key_enabled(cgroup_subsys_enabled_key[ssid]); | |
257 | } | |
258 | ||
223ffb29 JW |
259 | static bool cgroup_ssid_no_v1(int ssid) |
260 | { | |
261 | return cgroup_no_v1_mask & (1 << ssid); | |
262 | } | |
263 | ||
9e10a130 TH |
264 | /** |
265 | * cgroup_on_dfl - test whether a cgroup is on the default hierarchy | |
266 | * @cgrp: the cgroup of interest | |
267 | * | |
268 | * The default hierarchy is the v2 interface of cgroup and this function | |
269 | * can be used to test whether a cgroup is on the default hierarchy for | |
270 | * cases where a subsystem should behave differnetly depending on the | |
271 | * interface version. | |
272 | * | |
273 | * The set of behaviors which change on the default hierarchy are still | |
274 | * being determined and the mount option is prefixed with __DEVEL__. | |
275 | * | |
276 | * List of changed behaviors: | |
277 | * | |
278 | * - Mount options "noprefix", "xattr", "clone_children", "release_agent" | |
279 | * and "name" are disallowed. | |
280 | * | |
281 | * - When mounting an existing superblock, mount options should match. | |
282 | * | |
283 | * - Remount is disallowed. | |
284 | * | |
285 | * - rename(2) is disallowed. | |
286 | * | |
287 | * - "tasks" is removed. Everything should be at process granularity. Use | |
288 | * "cgroup.procs" instead. | |
289 | * | |
290 | * - "cgroup.procs" is not sorted. pids will be unique unless they got | |
291 | * recycled inbetween reads. | |
292 | * | |
293 | * - "release_agent" and "notify_on_release" are removed. Replacement | |
294 | * notification mechanism will be implemented. | |
295 | * | |
296 | * - "cgroup.clone_children" is removed. | |
297 | * | |
298 | * - "cgroup.subtree_populated" is available. Its value is 0 if the cgroup | |
299 | * and its descendants contain no task; otherwise, 1. The file also | |
300 | * generates kernfs notification which can be monitored through poll and | |
301 | * [di]notify when the value of the file changes. | |
302 | * | |
303 | * - cpuset: tasks will be kept in empty cpusets when hotplug happens and | |
304 | * take masks of ancestors with non-empty cpus/mems, instead of being | |
305 | * moved to an ancestor. | |
306 | * | |
307 | * - cpuset: a task can be moved into an empty cpuset, and again it takes | |
308 | * masks of ancestors. | |
309 | * | |
310 | * - memcg: use_hierarchy is on by default and the cgroup file for the flag | |
311 | * is not created. | |
312 | * | |
313 | * - blkcg: blk-throttle becomes properly hierarchical. | |
314 | * | |
315 | * - debug: disallowed on the default hierarchy. | |
316 | */ | |
317 | static bool cgroup_on_dfl(const struct cgroup *cgrp) | |
318 | { | |
319 | return cgrp->root == &cgrp_dfl_root; | |
320 | } | |
321 | ||
6fa4918d TH |
322 | /* IDR wrappers which synchronize using cgroup_idr_lock */ |
323 | static int cgroup_idr_alloc(struct idr *idr, void *ptr, int start, int end, | |
324 | gfp_t gfp_mask) | |
325 | { | |
326 | int ret; | |
327 | ||
328 | idr_preload(gfp_mask); | |
54504e97 | 329 | spin_lock_bh(&cgroup_idr_lock); |
d0164adc | 330 | ret = idr_alloc(idr, ptr, start, end, gfp_mask & ~__GFP_DIRECT_RECLAIM); |
54504e97 | 331 | spin_unlock_bh(&cgroup_idr_lock); |
6fa4918d TH |
332 | idr_preload_end(); |
333 | return ret; | |
334 | } | |
335 | ||
336 | static void *cgroup_idr_replace(struct idr *idr, void *ptr, int id) | |
337 | { | |
338 | void *ret; | |
339 | ||
54504e97 | 340 | spin_lock_bh(&cgroup_idr_lock); |
6fa4918d | 341 | ret = idr_replace(idr, ptr, id); |
54504e97 | 342 | spin_unlock_bh(&cgroup_idr_lock); |
6fa4918d TH |
343 | return ret; |
344 | } | |
345 | ||
346 | static void cgroup_idr_remove(struct idr *idr, int id) | |
347 | { | |
54504e97 | 348 | spin_lock_bh(&cgroup_idr_lock); |
6fa4918d | 349 | idr_remove(idr, id); |
54504e97 | 350 | spin_unlock_bh(&cgroup_idr_lock); |
6fa4918d TH |
351 | } |
352 | ||
d51f39b0 TH |
353 | static struct cgroup *cgroup_parent(struct cgroup *cgrp) |
354 | { | |
355 | struct cgroup_subsys_state *parent_css = cgrp->self.parent; | |
356 | ||
357 | if (parent_css) | |
358 | return container_of(parent_css, struct cgroup, self); | |
359 | return NULL; | |
360 | } | |
361 | ||
95109b62 TH |
362 | /** |
363 | * cgroup_css - obtain a cgroup's css for the specified subsystem | |
364 | * @cgrp: the cgroup of interest | |
9d800df1 | 365 | * @ss: the subsystem of interest (%NULL returns @cgrp->self) |
95109b62 | 366 | * |
ca8bdcaf TH |
367 | * Return @cgrp's css (cgroup_subsys_state) associated with @ss. This |
368 | * function must be called either under cgroup_mutex or rcu_read_lock() and | |
369 | * the caller is responsible for pinning the returned css if it wants to | |
370 | * keep accessing it outside the said locks. This function may return | |
371 | * %NULL if @cgrp doesn't have @subsys_id enabled. | |
95109b62 TH |
372 | */ |
373 | static struct cgroup_subsys_state *cgroup_css(struct cgroup *cgrp, | |
ca8bdcaf | 374 | struct cgroup_subsys *ss) |
95109b62 | 375 | { |
ca8bdcaf | 376 | if (ss) |
aec25020 | 377 | return rcu_dereference_check(cgrp->subsys[ss->id], |
ace2bee8 | 378 | lockdep_is_held(&cgroup_mutex)); |
ca8bdcaf | 379 | else |
9d800df1 | 380 | return &cgrp->self; |
95109b62 | 381 | } |
42809dd4 | 382 | |
aec3dfcb TH |
383 | /** |
384 | * cgroup_e_css - obtain a cgroup's effective css for the specified subsystem | |
385 | * @cgrp: the cgroup of interest | |
9d800df1 | 386 | * @ss: the subsystem of interest (%NULL returns @cgrp->self) |
aec3dfcb | 387 | * |
d0f702e6 | 388 | * Similar to cgroup_css() but returns the effective css, which is defined |
aec3dfcb TH |
389 | * as the matching css of the nearest ancestor including self which has @ss |
390 | * enabled. If @ss is associated with the hierarchy @cgrp is on, this | |
391 | * function is guaranteed to return non-NULL css. | |
392 | */ | |
393 | static struct cgroup_subsys_state *cgroup_e_css(struct cgroup *cgrp, | |
394 | struct cgroup_subsys *ss) | |
395 | { | |
396 | lockdep_assert_held(&cgroup_mutex); | |
397 | ||
398 | if (!ss) | |
9d800df1 | 399 | return &cgrp->self; |
aec3dfcb TH |
400 | |
401 | if (!(cgrp->root->subsys_mask & (1 << ss->id))) | |
402 | return NULL; | |
403 | ||
eeecbd19 TH |
404 | /* |
405 | * This function is used while updating css associations and thus | |
406 | * can't test the csses directly. Use ->child_subsys_mask. | |
407 | */ | |
d51f39b0 TH |
408 | while (cgroup_parent(cgrp) && |
409 | !(cgroup_parent(cgrp)->child_subsys_mask & (1 << ss->id))) | |
410 | cgrp = cgroup_parent(cgrp); | |
aec3dfcb TH |
411 | |
412 | return cgroup_css(cgrp, ss); | |
95109b62 | 413 | } |
42809dd4 | 414 | |
eeecbd19 TH |
415 | /** |
416 | * cgroup_get_e_css - get a cgroup's effective css for the specified subsystem | |
417 | * @cgrp: the cgroup of interest | |
418 | * @ss: the subsystem of interest | |
419 | * | |
420 | * Find and get the effective css of @cgrp for @ss. The effective css is | |
421 | * defined as the matching css of the nearest ancestor including self which | |
422 | * has @ss enabled. If @ss is not mounted on the hierarchy @cgrp is on, | |
423 | * the root css is returned, so this function always returns a valid css. | |
424 | * The returned css must be put using css_put(). | |
425 | */ | |
426 | struct cgroup_subsys_state *cgroup_get_e_css(struct cgroup *cgrp, | |
427 | struct cgroup_subsys *ss) | |
428 | { | |
429 | struct cgroup_subsys_state *css; | |
430 | ||
431 | rcu_read_lock(); | |
432 | ||
433 | do { | |
434 | css = cgroup_css(cgrp, ss); | |
435 | ||
436 | if (css && css_tryget_online(css)) | |
437 | goto out_unlock; | |
438 | cgrp = cgroup_parent(cgrp); | |
439 | } while (cgrp); | |
440 | ||
441 | css = init_css_set.subsys[ss->id]; | |
442 | css_get(css); | |
443 | out_unlock: | |
444 | rcu_read_unlock(); | |
445 | return css; | |
446 | } | |
447 | ||
ddbcc7e8 | 448 | /* convenient tests for these bits */ |
54766d4a | 449 | static inline bool cgroup_is_dead(const struct cgroup *cgrp) |
ddbcc7e8 | 450 | { |
184faf32 | 451 | return !(cgrp->self.flags & CSS_ONLINE); |
ddbcc7e8 PM |
452 | } |
453 | ||
052c3f3a TH |
454 | static void cgroup_get(struct cgroup *cgrp) |
455 | { | |
456 | WARN_ON_ONCE(cgroup_is_dead(cgrp)); | |
457 | css_get(&cgrp->self); | |
458 | } | |
459 | ||
460 | static bool cgroup_tryget(struct cgroup *cgrp) | |
461 | { | |
462 | return css_tryget(&cgrp->self); | |
463 | } | |
464 | ||
b4168640 | 465 | struct cgroup_subsys_state *of_css(struct kernfs_open_file *of) |
59f5296b | 466 | { |
2bd59d48 | 467 | struct cgroup *cgrp = of->kn->parent->priv; |
b4168640 | 468 | struct cftype *cft = of_cft(of); |
2bd59d48 TH |
469 | |
470 | /* | |
471 | * This is open and unprotected implementation of cgroup_css(). | |
472 | * seq_css() is only called from a kernfs file operation which has | |
473 | * an active reference on the file. Because all the subsystem | |
474 | * files are drained before a css is disassociated with a cgroup, | |
475 | * the matching css from the cgroup's subsys table is guaranteed to | |
476 | * be and stay valid until the enclosing operation is complete. | |
477 | */ | |
478 | if (cft->ss) | |
479 | return rcu_dereference_raw(cgrp->subsys[cft->ss->id]); | |
480 | else | |
9d800df1 | 481 | return &cgrp->self; |
59f5296b | 482 | } |
b4168640 | 483 | EXPORT_SYMBOL_GPL(of_css); |
59f5296b | 484 | |
e9685a03 | 485 | static int notify_on_release(const struct cgroup *cgrp) |
81a6a5cd | 486 | { |
bd89aabc | 487 | return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); |
81a6a5cd PM |
488 | } |
489 | ||
1c6727af TH |
490 | /** |
491 | * for_each_css - iterate all css's of a cgroup | |
492 | * @css: the iteration cursor | |
493 | * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end | |
494 | * @cgrp: the target cgroup to iterate css's of | |
495 | * | |
aec3dfcb | 496 | * Should be called under cgroup_[tree_]mutex. |
1c6727af TH |
497 | */ |
498 | #define for_each_css(css, ssid, cgrp) \ | |
499 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \ | |
500 | if (!((css) = rcu_dereference_check( \ | |
501 | (cgrp)->subsys[(ssid)], \ | |
502 | lockdep_is_held(&cgroup_mutex)))) { } \ | |
503 | else | |
504 | ||
aec3dfcb TH |
505 | /** |
506 | * for_each_e_css - iterate all effective css's of a cgroup | |
507 | * @css: the iteration cursor | |
508 | * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end | |
509 | * @cgrp: the target cgroup to iterate css's of | |
510 | * | |
511 | * Should be called under cgroup_[tree_]mutex. | |
512 | */ | |
513 | #define for_each_e_css(css, ssid, cgrp) \ | |
514 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \ | |
515 | if (!((css) = cgroup_e_css(cgrp, cgroup_subsys[(ssid)]))) \ | |
516 | ; \ | |
517 | else | |
518 | ||
30159ec7 | 519 | /** |
3ed80a62 | 520 | * for_each_subsys - iterate all enabled cgroup subsystems |
30159ec7 | 521 | * @ss: the iteration cursor |
780cd8b3 | 522 | * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end |
30159ec7 | 523 | */ |
780cd8b3 | 524 | #define for_each_subsys(ss, ssid) \ |
3ed80a62 TH |
525 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT && \ |
526 | (((ss) = cgroup_subsys[ssid]) || true); (ssid)++) | |
30159ec7 | 527 | |
cb4a3167 AS |
528 | /** |
529 | * for_each_subsys_which - filter for_each_subsys with a bitmask | |
530 | * @ss: the iteration cursor | |
531 | * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end | |
532 | * @ss_maskp: a pointer to the bitmask | |
533 | * | |
534 | * The block will only run for cases where the ssid-th bit (1 << ssid) of | |
535 | * mask is set to 1. | |
536 | */ | |
537 | #define for_each_subsys_which(ss, ssid, ss_maskp) \ | |
538 | if (!CGROUP_SUBSYS_COUNT) /* to avoid spurious gcc warning */ \ | |
4a705c5c | 539 | (ssid) = 0; \ |
cb4a3167 AS |
540 | else \ |
541 | for_each_set_bit(ssid, ss_maskp, CGROUP_SUBSYS_COUNT) \ | |
542 | if (((ss) = cgroup_subsys[ssid]) && false) \ | |
543 | break; \ | |
544 | else | |
545 | ||
985ed670 TH |
546 | /* iterate across the hierarchies */ |
547 | #define for_each_root(root) \ | |
5549c497 | 548 | list_for_each_entry((root), &cgroup_roots, root_list) |
ddbcc7e8 | 549 | |
f8f22e53 TH |
550 | /* iterate over child cgrps, lock should be held throughout iteration */ |
551 | #define cgroup_for_each_live_child(child, cgrp) \ | |
d5c419b6 | 552 | list_for_each_entry((child), &(cgrp)->self.children, self.sibling) \ |
8353da1f | 553 | if (({ lockdep_assert_held(&cgroup_mutex); \ |
f8f22e53 TH |
554 | cgroup_is_dead(child); })) \ |
555 | ; \ | |
556 | else | |
7ae1bad9 | 557 | |
81a6a5cd | 558 | static void cgroup_release_agent(struct work_struct *work); |
bd89aabc | 559 | static void check_for_release(struct cgroup *cgrp); |
81a6a5cd | 560 | |
69d0206c TH |
561 | /* |
562 | * A cgroup can be associated with multiple css_sets as different tasks may | |
563 | * belong to different cgroups on different hierarchies. In the other | |
564 | * direction, a css_set is naturally associated with multiple cgroups. | |
565 | * This M:N relationship is represented by the following link structure | |
566 | * which exists for each association and allows traversing the associations | |
567 | * from both sides. | |
568 | */ | |
569 | struct cgrp_cset_link { | |
570 | /* the cgroup and css_set this link associates */ | |
571 | struct cgroup *cgrp; | |
572 | struct css_set *cset; | |
573 | ||
574 | /* list of cgrp_cset_links anchored at cgrp->cset_links */ | |
575 | struct list_head cset_link; | |
576 | ||
577 | /* list of cgrp_cset_links anchored at css_set->cgrp_links */ | |
578 | struct list_head cgrp_link; | |
817929ec PM |
579 | }; |
580 | ||
172a2c06 TH |
581 | /* |
582 | * The default css_set - used by init and its children prior to any | |
817929ec PM |
583 | * hierarchies being mounted. It contains a pointer to the root state |
584 | * for each subsystem. Also used to anchor the list of css_sets. Not | |
585 | * reference-counted, to improve performance when child cgroups | |
586 | * haven't been created. | |
587 | */ | |
5024ae29 | 588 | struct css_set init_css_set = { |
172a2c06 TH |
589 | .refcount = ATOMIC_INIT(1), |
590 | .cgrp_links = LIST_HEAD_INIT(init_css_set.cgrp_links), | |
591 | .tasks = LIST_HEAD_INIT(init_css_set.tasks), | |
592 | .mg_tasks = LIST_HEAD_INIT(init_css_set.mg_tasks), | |
593 | .mg_preload_node = LIST_HEAD_INIT(init_css_set.mg_preload_node), | |
594 | .mg_node = LIST_HEAD_INIT(init_css_set.mg_node), | |
ed27b9f7 | 595 | .task_iters = LIST_HEAD_INIT(init_css_set.task_iters), |
172a2c06 | 596 | }; |
817929ec | 597 | |
172a2c06 | 598 | static int css_set_count = 1; /* 1 for init_css_set */ |
817929ec | 599 | |
0de0942d TH |
600 | /** |
601 | * css_set_populated - does a css_set contain any tasks? | |
602 | * @cset: target css_set | |
603 | */ | |
604 | static bool css_set_populated(struct css_set *cset) | |
605 | { | |
f0d9a5f1 | 606 | lockdep_assert_held(&css_set_lock); |
0de0942d TH |
607 | |
608 | return !list_empty(&cset->tasks) || !list_empty(&cset->mg_tasks); | |
609 | } | |
610 | ||
842b597e TH |
611 | /** |
612 | * cgroup_update_populated - updated populated count of a cgroup | |
613 | * @cgrp: the target cgroup | |
614 | * @populated: inc or dec populated count | |
615 | * | |
0de0942d TH |
616 | * One of the css_sets associated with @cgrp is either getting its first |
617 | * task or losing the last. Update @cgrp->populated_cnt accordingly. The | |
618 | * count is propagated towards root so that a given cgroup's populated_cnt | |
619 | * is zero iff the cgroup and all its descendants don't contain any tasks. | |
842b597e TH |
620 | * |
621 | * @cgrp's interface file "cgroup.populated" is zero if | |
622 | * @cgrp->populated_cnt is zero and 1 otherwise. When @cgrp->populated_cnt | |
623 | * changes from or to zero, userland is notified that the content of the | |
624 | * interface file has changed. This can be used to detect when @cgrp and | |
625 | * its descendants become populated or empty. | |
626 | */ | |
627 | static void cgroup_update_populated(struct cgroup *cgrp, bool populated) | |
628 | { | |
f0d9a5f1 | 629 | lockdep_assert_held(&css_set_lock); |
842b597e TH |
630 | |
631 | do { | |
632 | bool trigger; | |
633 | ||
634 | if (populated) | |
635 | trigger = !cgrp->populated_cnt++; | |
636 | else | |
637 | trigger = !--cgrp->populated_cnt; | |
638 | ||
639 | if (!trigger) | |
640 | break; | |
641 | ||
ad2ed2b3 | 642 | check_for_release(cgrp); |
6f60eade TH |
643 | cgroup_file_notify(&cgrp->events_file); |
644 | ||
d51f39b0 | 645 | cgrp = cgroup_parent(cgrp); |
842b597e TH |
646 | } while (cgrp); |
647 | } | |
648 | ||
0de0942d TH |
649 | /** |
650 | * css_set_update_populated - update populated state of a css_set | |
651 | * @cset: target css_set | |
652 | * @populated: whether @cset is populated or depopulated | |
653 | * | |
654 | * @cset is either getting the first task or losing the last. Update the | |
655 | * ->populated_cnt of all associated cgroups accordingly. | |
656 | */ | |
657 | static void css_set_update_populated(struct css_set *cset, bool populated) | |
658 | { | |
659 | struct cgrp_cset_link *link; | |
660 | ||
f0d9a5f1 | 661 | lockdep_assert_held(&css_set_lock); |
0de0942d TH |
662 | |
663 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) | |
664 | cgroup_update_populated(link->cgrp, populated); | |
665 | } | |
666 | ||
f6d7d049 TH |
667 | /** |
668 | * css_set_move_task - move a task from one css_set to another | |
669 | * @task: task being moved | |
670 | * @from_cset: css_set @task currently belongs to (may be NULL) | |
671 | * @to_cset: new css_set @task is being moved to (may be NULL) | |
672 | * @use_mg_tasks: move to @to_cset->mg_tasks instead of ->tasks | |
673 | * | |
674 | * Move @task from @from_cset to @to_cset. If @task didn't belong to any | |
675 | * css_set, @from_cset can be NULL. If @task is being disassociated | |
676 | * instead of moved, @to_cset can be NULL. | |
677 | * | |
ed27b9f7 TH |
678 | * This function automatically handles populated_cnt updates and |
679 | * css_task_iter adjustments but the caller is responsible for managing | |
680 | * @from_cset and @to_cset's reference counts. | |
f6d7d049 TH |
681 | */ |
682 | static void css_set_move_task(struct task_struct *task, | |
683 | struct css_set *from_cset, struct css_set *to_cset, | |
684 | bool use_mg_tasks) | |
685 | { | |
f0d9a5f1 | 686 | lockdep_assert_held(&css_set_lock); |
f6d7d049 TH |
687 | |
688 | if (from_cset) { | |
ed27b9f7 TH |
689 | struct css_task_iter *it, *pos; |
690 | ||
f6d7d049 | 691 | WARN_ON_ONCE(list_empty(&task->cg_list)); |
ed27b9f7 TH |
692 | |
693 | /* | |
694 | * @task is leaving, advance task iterators which are | |
695 | * pointing to it so that they can resume at the next | |
696 | * position. Advancing an iterator might remove it from | |
697 | * the list, use safe walk. See css_task_iter_advance*() | |
698 | * for details. | |
699 | */ | |
700 | list_for_each_entry_safe(it, pos, &from_cset->task_iters, | |
701 | iters_node) | |
702 | if (it->task_pos == &task->cg_list) | |
703 | css_task_iter_advance(it); | |
704 | ||
f6d7d049 TH |
705 | list_del_init(&task->cg_list); |
706 | if (!css_set_populated(from_cset)) | |
707 | css_set_update_populated(from_cset, false); | |
708 | } else { | |
709 | WARN_ON_ONCE(!list_empty(&task->cg_list)); | |
710 | } | |
711 | ||
712 | if (to_cset) { | |
713 | /* | |
714 | * We are synchronized through cgroup_threadgroup_rwsem | |
715 | * against PF_EXITING setting such that we can't race | |
716 | * against cgroup_exit() changing the css_set to | |
717 | * init_css_set and dropping the old one. | |
718 | */ | |
719 | WARN_ON_ONCE(task->flags & PF_EXITING); | |
720 | ||
721 | if (!css_set_populated(to_cset)) | |
722 | css_set_update_populated(to_cset, true); | |
723 | rcu_assign_pointer(task->cgroups, to_cset); | |
724 | list_add_tail(&task->cg_list, use_mg_tasks ? &to_cset->mg_tasks : | |
725 | &to_cset->tasks); | |
726 | } | |
727 | } | |
728 | ||
7717f7ba PM |
729 | /* |
730 | * hash table for cgroup groups. This improves the performance to find | |
731 | * an existing css_set. This hash doesn't (currently) take into | |
732 | * account cgroups in empty hierarchies. | |
733 | */ | |
472b1053 | 734 | #define CSS_SET_HASH_BITS 7 |
0ac801fe | 735 | static DEFINE_HASHTABLE(css_set_table, CSS_SET_HASH_BITS); |
472b1053 | 736 | |
0ac801fe | 737 | static unsigned long css_set_hash(struct cgroup_subsys_state *css[]) |
472b1053 | 738 | { |
0ac801fe | 739 | unsigned long key = 0UL; |
30159ec7 TH |
740 | struct cgroup_subsys *ss; |
741 | int i; | |
472b1053 | 742 | |
30159ec7 | 743 | for_each_subsys(ss, i) |
0ac801fe LZ |
744 | key += (unsigned long)css[i]; |
745 | key = (key >> 16) ^ key; | |
472b1053 | 746 | |
0ac801fe | 747 | return key; |
472b1053 LZ |
748 | } |
749 | ||
a25eb52e | 750 | static void put_css_set_locked(struct css_set *cset) |
b4f48b63 | 751 | { |
69d0206c | 752 | struct cgrp_cset_link *link, *tmp_link; |
2d8f243a TH |
753 | struct cgroup_subsys *ss; |
754 | int ssid; | |
5abb8855 | 755 | |
f0d9a5f1 | 756 | lockdep_assert_held(&css_set_lock); |
89c5509b TH |
757 | |
758 | if (!atomic_dec_and_test(&cset->refcount)) | |
146aa1bd | 759 | return; |
81a6a5cd | 760 | |
53254f90 TH |
761 | /* This css_set is dead. unlink it and release cgroup and css refs */ |
762 | for_each_subsys(ss, ssid) { | |
2d8f243a | 763 | list_del(&cset->e_cset_node[ssid]); |
53254f90 TH |
764 | css_put(cset->subsys[ssid]); |
765 | } | |
5abb8855 | 766 | hash_del(&cset->hlist); |
2c6ab6d2 PM |
767 | css_set_count--; |
768 | ||
69d0206c | 769 | list_for_each_entry_safe(link, tmp_link, &cset->cgrp_links, cgrp_link) { |
69d0206c TH |
770 | list_del(&link->cset_link); |
771 | list_del(&link->cgrp_link); | |
2ceb231b TH |
772 | if (cgroup_parent(link->cgrp)) |
773 | cgroup_put(link->cgrp); | |
2c6ab6d2 | 774 | kfree(link); |
81a6a5cd | 775 | } |
2c6ab6d2 | 776 | |
5abb8855 | 777 | kfree_rcu(cset, rcu_head); |
b4f48b63 PM |
778 | } |
779 | ||
a25eb52e | 780 | static void put_css_set(struct css_set *cset) |
89c5509b TH |
781 | { |
782 | /* | |
783 | * Ensure that the refcount doesn't hit zero while any readers | |
784 | * can see it. Similar to atomic_dec_and_lock(), but for an | |
785 | * rwlock | |
786 | */ | |
787 | if (atomic_add_unless(&cset->refcount, -1, 1)) | |
788 | return; | |
789 | ||
f0d9a5f1 | 790 | spin_lock_bh(&css_set_lock); |
a25eb52e | 791 | put_css_set_locked(cset); |
f0d9a5f1 | 792 | spin_unlock_bh(&css_set_lock); |
89c5509b TH |
793 | } |
794 | ||
817929ec PM |
795 | /* |
796 | * refcounted get/put for css_set objects | |
797 | */ | |
5abb8855 | 798 | static inline void get_css_set(struct css_set *cset) |
817929ec | 799 | { |
5abb8855 | 800 | atomic_inc(&cset->refcount); |
817929ec PM |
801 | } |
802 | ||
b326f9d0 | 803 | /** |
7717f7ba | 804 | * compare_css_sets - helper function for find_existing_css_set(). |
5abb8855 TH |
805 | * @cset: candidate css_set being tested |
806 | * @old_cset: existing css_set for a task | |
7717f7ba PM |
807 | * @new_cgrp: cgroup that's being entered by the task |
808 | * @template: desired set of css pointers in css_set (pre-calculated) | |
809 | * | |
6f4b7e63 | 810 | * Returns true if "cset" matches "old_cset" except for the hierarchy |
7717f7ba PM |
811 | * which "new_cgrp" belongs to, for which it should match "new_cgrp". |
812 | */ | |
5abb8855 TH |
813 | static bool compare_css_sets(struct css_set *cset, |
814 | struct css_set *old_cset, | |
7717f7ba PM |
815 | struct cgroup *new_cgrp, |
816 | struct cgroup_subsys_state *template[]) | |
817 | { | |
818 | struct list_head *l1, *l2; | |
819 | ||
aec3dfcb TH |
820 | /* |
821 | * On the default hierarchy, there can be csets which are | |
822 | * associated with the same set of cgroups but different csses. | |
823 | * Let's first ensure that csses match. | |
824 | */ | |
825 | if (memcmp(template, cset->subsys, sizeof(cset->subsys))) | |
7717f7ba | 826 | return false; |
7717f7ba PM |
827 | |
828 | /* | |
829 | * Compare cgroup pointers in order to distinguish between | |
aec3dfcb TH |
830 | * different cgroups in hierarchies. As different cgroups may |
831 | * share the same effective css, this comparison is always | |
832 | * necessary. | |
7717f7ba | 833 | */ |
69d0206c TH |
834 | l1 = &cset->cgrp_links; |
835 | l2 = &old_cset->cgrp_links; | |
7717f7ba | 836 | while (1) { |
69d0206c | 837 | struct cgrp_cset_link *link1, *link2; |
5abb8855 | 838 | struct cgroup *cgrp1, *cgrp2; |
7717f7ba PM |
839 | |
840 | l1 = l1->next; | |
841 | l2 = l2->next; | |
842 | /* See if we reached the end - both lists are equal length. */ | |
69d0206c TH |
843 | if (l1 == &cset->cgrp_links) { |
844 | BUG_ON(l2 != &old_cset->cgrp_links); | |
7717f7ba PM |
845 | break; |
846 | } else { | |
69d0206c | 847 | BUG_ON(l2 == &old_cset->cgrp_links); |
7717f7ba PM |
848 | } |
849 | /* Locate the cgroups associated with these links. */ | |
69d0206c TH |
850 | link1 = list_entry(l1, struct cgrp_cset_link, cgrp_link); |
851 | link2 = list_entry(l2, struct cgrp_cset_link, cgrp_link); | |
852 | cgrp1 = link1->cgrp; | |
853 | cgrp2 = link2->cgrp; | |
7717f7ba | 854 | /* Hierarchies should be linked in the same order. */ |
5abb8855 | 855 | BUG_ON(cgrp1->root != cgrp2->root); |
7717f7ba PM |
856 | |
857 | /* | |
858 | * If this hierarchy is the hierarchy of the cgroup | |
859 | * that's changing, then we need to check that this | |
860 | * css_set points to the new cgroup; if it's any other | |
861 | * hierarchy, then this css_set should point to the | |
862 | * same cgroup as the old css_set. | |
863 | */ | |
5abb8855 TH |
864 | if (cgrp1->root == new_cgrp->root) { |
865 | if (cgrp1 != new_cgrp) | |
7717f7ba PM |
866 | return false; |
867 | } else { | |
5abb8855 | 868 | if (cgrp1 != cgrp2) |
7717f7ba PM |
869 | return false; |
870 | } | |
871 | } | |
872 | return true; | |
873 | } | |
874 | ||
b326f9d0 TH |
875 | /** |
876 | * find_existing_css_set - init css array and find the matching css_set | |
877 | * @old_cset: the css_set that we're using before the cgroup transition | |
878 | * @cgrp: the cgroup that we're moving into | |
879 | * @template: out param for the new set of csses, should be clear on entry | |
817929ec | 880 | */ |
5abb8855 TH |
881 | static struct css_set *find_existing_css_set(struct css_set *old_cset, |
882 | struct cgroup *cgrp, | |
883 | struct cgroup_subsys_state *template[]) | |
b4f48b63 | 884 | { |
3dd06ffa | 885 | struct cgroup_root *root = cgrp->root; |
30159ec7 | 886 | struct cgroup_subsys *ss; |
5abb8855 | 887 | struct css_set *cset; |
0ac801fe | 888 | unsigned long key; |
b326f9d0 | 889 | int i; |
817929ec | 890 | |
aae8aab4 BB |
891 | /* |
892 | * Build the set of subsystem state objects that we want to see in the | |
893 | * new css_set. while subsystems can change globally, the entries here | |
894 | * won't change, so no need for locking. | |
895 | */ | |
30159ec7 | 896 | for_each_subsys(ss, i) { |
f392e51c | 897 | if (root->subsys_mask & (1UL << i)) { |
aec3dfcb TH |
898 | /* |
899 | * @ss is in this hierarchy, so we want the | |
900 | * effective css from @cgrp. | |
901 | */ | |
902 | template[i] = cgroup_e_css(cgrp, ss); | |
817929ec | 903 | } else { |
aec3dfcb TH |
904 | /* |
905 | * @ss is not in this hierarchy, so we don't want | |
906 | * to change the css. | |
907 | */ | |
5abb8855 | 908 | template[i] = old_cset->subsys[i]; |
817929ec PM |
909 | } |
910 | } | |
911 | ||
0ac801fe | 912 | key = css_set_hash(template); |
5abb8855 TH |
913 | hash_for_each_possible(css_set_table, cset, hlist, key) { |
914 | if (!compare_css_sets(cset, old_cset, cgrp, template)) | |
7717f7ba PM |
915 | continue; |
916 | ||
917 | /* This css_set matches what we need */ | |
5abb8855 | 918 | return cset; |
472b1053 | 919 | } |
817929ec PM |
920 | |
921 | /* No existing cgroup group matched */ | |
922 | return NULL; | |
923 | } | |
924 | ||
69d0206c | 925 | static void free_cgrp_cset_links(struct list_head *links_to_free) |
36553434 | 926 | { |
69d0206c | 927 | struct cgrp_cset_link *link, *tmp_link; |
36553434 | 928 | |
69d0206c TH |
929 | list_for_each_entry_safe(link, tmp_link, links_to_free, cset_link) { |
930 | list_del(&link->cset_link); | |
36553434 LZ |
931 | kfree(link); |
932 | } | |
933 | } | |
934 | ||
69d0206c TH |
935 | /** |
936 | * allocate_cgrp_cset_links - allocate cgrp_cset_links | |
937 | * @count: the number of links to allocate | |
938 | * @tmp_links: list_head the allocated links are put on | |
939 | * | |
940 | * Allocate @count cgrp_cset_link structures and chain them on @tmp_links | |
941 | * through ->cset_link. Returns 0 on success or -errno. | |
817929ec | 942 | */ |
69d0206c | 943 | static int allocate_cgrp_cset_links(int count, struct list_head *tmp_links) |
817929ec | 944 | { |
69d0206c | 945 | struct cgrp_cset_link *link; |
817929ec | 946 | int i; |
69d0206c TH |
947 | |
948 | INIT_LIST_HEAD(tmp_links); | |
949 | ||
817929ec | 950 | for (i = 0; i < count; i++) { |
f4f4be2b | 951 | link = kzalloc(sizeof(*link), GFP_KERNEL); |
817929ec | 952 | if (!link) { |
69d0206c | 953 | free_cgrp_cset_links(tmp_links); |
817929ec PM |
954 | return -ENOMEM; |
955 | } | |
69d0206c | 956 | list_add(&link->cset_link, tmp_links); |
817929ec PM |
957 | } |
958 | return 0; | |
959 | } | |
960 | ||
c12f65d4 LZ |
961 | /** |
962 | * link_css_set - a helper function to link a css_set to a cgroup | |
69d0206c | 963 | * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links() |
5abb8855 | 964 | * @cset: the css_set to be linked |
c12f65d4 LZ |
965 | * @cgrp: the destination cgroup |
966 | */ | |
69d0206c TH |
967 | static void link_css_set(struct list_head *tmp_links, struct css_set *cset, |
968 | struct cgroup *cgrp) | |
c12f65d4 | 969 | { |
69d0206c | 970 | struct cgrp_cset_link *link; |
c12f65d4 | 971 | |
69d0206c | 972 | BUG_ON(list_empty(tmp_links)); |
6803c006 TH |
973 | |
974 | if (cgroup_on_dfl(cgrp)) | |
975 | cset->dfl_cgrp = cgrp; | |
976 | ||
69d0206c TH |
977 | link = list_first_entry(tmp_links, struct cgrp_cset_link, cset_link); |
978 | link->cset = cset; | |
7717f7ba | 979 | link->cgrp = cgrp; |
842b597e | 980 | |
7717f7ba | 981 | /* |
389b9c1b TH |
982 | * Always add links to the tail of the lists so that the lists are |
983 | * in choronological order. | |
7717f7ba | 984 | */ |
389b9c1b | 985 | list_move_tail(&link->cset_link, &cgrp->cset_links); |
69d0206c | 986 | list_add_tail(&link->cgrp_link, &cset->cgrp_links); |
2ceb231b TH |
987 | |
988 | if (cgroup_parent(cgrp)) | |
989 | cgroup_get(cgrp); | |
c12f65d4 LZ |
990 | } |
991 | ||
b326f9d0 TH |
992 | /** |
993 | * find_css_set - return a new css_set with one cgroup updated | |
994 | * @old_cset: the baseline css_set | |
995 | * @cgrp: the cgroup to be updated | |
996 | * | |
997 | * Return a new css_set that's equivalent to @old_cset, but with @cgrp | |
998 | * substituted into the appropriate hierarchy. | |
817929ec | 999 | */ |
5abb8855 TH |
1000 | static struct css_set *find_css_set(struct css_set *old_cset, |
1001 | struct cgroup *cgrp) | |
817929ec | 1002 | { |
b326f9d0 | 1003 | struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT] = { }; |
5abb8855 | 1004 | struct css_set *cset; |
69d0206c TH |
1005 | struct list_head tmp_links; |
1006 | struct cgrp_cset_link *link; | |
2d8f243a | 1007 | struct cgroup_subsys *ss; |
0ac801fe | 1008 | unsigned long key; |
2d8f243a | 1009 | int ssid; |
472b1053 | 1010 | |
b326f9d0 TH |
1011 | lockdep_assert_held(&cgroup_mutex); |
1012 | ||
817929ec PM |
1013 | /* First see if we already have a cgroup group that matches |
1014 | * the desired set */ | |
f0d9a5f1 | 1015 | spin_lock_bh(&css_set_lock); |
5abb8855 TH |
1016 | cset = find_existing_css_set(old_cset, cgrp, template); |
1017 | if (cset) | |
1018 | get_css_set(cset); | |
f0d9a5f1 | 1019 | spin_unlock_bh(&css_set_lock); |
817929ec | 1020 | |
5abb8855 TH |
1021 | if (cset) |
1022 | return cset; | |
817929ec | 1023 | |
f4f4be2b | 1024 | cset = kzalloc(sizeof(*cset), GFP_KERNEL); |
5abb8855 | 1025 | if (!cset) |
817929ec PM |
1026 | return NULL; |
1027 | ||
69d0206c | 1028 | /* Allocate all the cgrp_cset_link objects that we'll need */ |
9871bf95 | 1029 | if (allocate_cgrp_cset_links(cgroup_root_count, &tmp_links) < 0) { |
5abb8855 | 1030 | kfree(cset); |
817929ec PM |
1031 | return NULL; |
1032 | } | |
1033 | ||
5abb8855 | 1034 | atomic_set(&cset->refcount, 1); |
69d0206c | 1035 | INIT_LIST_HEAD(&cset->cgrp_links); |
5abb8855 | 1036 | INIT_LIST_HEAD(&cset->tasks); |
c7561128 | 1037 | INIT_LIST_HEAD(&cset->mg_tasks); |
1958d2d5 | 1038 | INIT_LIST_HEAD(&cset->mg_preload_node); |
b3dc094e | 1039 | INIT_LIST_HEAD(&cset->mg_node); |
ed27b9f7 | 1040 | INIT_LIST_HEAD(&cset->task_iters); |
5abb8855 | 1041 | INIT_HLIST_NODE(&cset->hlist); |
817929ec PM |
1042 | |
1043 | /* Copy the set of subsystem state objects generated in | |
1044 | * find_existing_css_set() */ | |
5abb8855 | 1045 | memcpy(cset->subsys, template, sizeof(cset->subsys)); |
817929ec | 1046 | |
f0d9a5f1 | 1047 | spin_lock_bh(&css_set_lock); |
817929ec | 1048 | /* Add reference counts and links from the new css_set. */ |
69d0206c | 1049 | list_for_each_entry(link, &old_cset->cgrp_links, cgrp_link) { |
7717f7ba | 1050 | struct cgroup *c = link->cgrp; |
69d0206c | 1051 | |
7717f7ba PM |
1052 | if (c->root == cgrp->root) |
1053 | c = cgrp; | |
69d0206c | 1054 | link_css_set(&tmp_links, cset, c); |
7717f7ba | 1055 | } |
817929ec | 1056 | |
69d0206c | 1057 | BUG_ON(!list_empty(&tmp_links)); |
817929ec | 1058 | |
817929ec | 1059 | css_set_count++; |
472b1053 | 1060 | |
2d8f243a | 1061 | /* Add @cset to the hash table */ |
5abb8855 TH |
1062 | key = css_set_hash(cset->subsys); |
1063 | hash_add(css_set_table, &cset->hlist, key); | |
472b1053 | 1064 | |
53254f90 TH |
1065 | for_each_subsys(ss, ssid) { |
1066 | struct cgroup_subsys_state *css = cset->subsys[ssid]; | |
1067 | ||
2d8f243a | 1068 | list_add_tail(&cset->e_cset_node[ssid], |
53254f90 TH |
1069 | &css->cgroup->e_csets[ssid]); |
1070 | css_get(css); | |
1071 | } | |
2d8f243a | 1072 | |
f0d9a5f1 | 1073 | spin_unlock_bh(&css_set_lock); |
817929ec | 1074 | |
5abb8855 | 1075 | return cset; |
b4f48b63 PM |
1076 | } |
1077 | ||
3dd06ffa | 1078 | static struct cgroup_root *cgroup_root_from_kf(struct kernfs_root *kf_root) |
7717f7ba | 1079 | { |
3dd06ffa | 1080 | struct cgroup *root_cgrp = kf_root->kn->priv; |
2bd59d48 | 1081 | |
3dd06ffa | 1082 | return root_cgrp->root; |
2bd59d48 TH |
1083 | } |
1084 | ||
3dd06ffa | 1085 | static int cgroup_init_root_id(struct cgroup_root *root) |
f2e85d57 TH |
1086 | { |
1087 | int id; | |
1088 | ||
1089 | lockdep_assert_held(&cgroup_mutex); | |
1090 | ||
985ed670 | 1091 | id = idr_alloc_cyclic(&cgroup_hierarchy_idr, root, 0, 0, GFP_KERNEL); |
f2e85d57 TH |
1092 | if (id < 0) |
1093 | return id; | |
1094 | ||
1095 | root->hierarchy_id = id; | |
1096 | return 0; | |
1097 | } | |
1098 | ||
3dd06ffa | 1099 | static void cgroup_exit_root_id(struct cgroup_root *root) |
f2e85d57 TH |
1100 | { |
1101 | lockdep_assert_held(&cgroup_mutex); | |
1102 | ||
1103 | if (root->hierarchy_id) { | |
1104 | idr_remove(&cgroup_hierarchy_idr, root->hierarchy_id); | |
1105 | root->hierarchy_id = 0; | |
1106 | } | |
1107 | } | |
1108 | ||
3dd06ffa | 1109 | static void cgroup_free_root(struct cgroup_root *root) |
f2e85d57 TH |
1110 | { |
1111 | if (root) { | |
d0f702e6 | 1112 | /* hierarchy ID should already have been released */ |
f2e85d57 TH |
1113 | WARN_ON_ONCE(root->hierarchy_id); |
1114 | ||
1115 | idr_destroy(&root->cgroup_idr); | |
1116 | kfree(root); | |
1117 | } | |
1118 | } | |
1119 | ||
3dd06ffa | 1120 | static void cgroup_destroy_root(struct cgroup_root *root) |
59f5296b | 1121 | { |
3dd06ffa | 1122 | struct cgroup *cgrp = &root->cgrp; |
f2e85d57 | 1123 | struct cgrp_cset_link *link, *tmp_link; |
f2e85d57 | 1124 | |
2bd59d48 | 1125 | mutex_lock(&cgroup_mutex); |
f2e85d57 | 1126 | |
776f02fa | 1127 | BUG_ON(atomic_read(&root->nr_cgrps)); |
d5c419b6 | 1128 | BUG_ON(!list_empty(&cgrp->self.children)); |
f2e85d57 | 1129 | |
f2e85d57 | 1130 | /* Rebind all subsystems back to the default hierarchy */ |
f392e51c | 1131 | rebind_subsystems(&cgrp_dfl_root, root->subsys_mask); |
7717f7ba | 1132 | |
7717f7ba | 1133 | /* |
f2e85d57 TH |
1134 | * Release all the links from cset_links to this hierarchy's |
1135 | * root cgroup | |
7717f7ba | 1136 | */ |
f0d9a5f1 | 1137 | spin_lock_bh(&css_set_lock); |
f2e85d57 TH |
1138 | |
1139 | list_for_each_entry_safe(link, tmp_link, &cgrp->cset_links, cset_link) { | |
1140 | list_del(&link->cset_link); | |
1141 | list_del(&link->cgrp_link); | |
1142 | kfree(link); | |
1143 | } | |
f0d9a5f1 TH |
1144 | |
1145 | spin_unlock_bh(&css_set_lock); | |
f2e85d57 TH |
1146 | |
1147 | if (!list_empty(&root->root_list)) { | |
1148 | list_del(&root->root_list); | |
1149 | cgroup_root_count--; | |
1150 | } | |
1151 | ||
1152 | cgroup_exit_root_id(root); | |
1153 | ||
1154 | mutex_unlock(&cgroup_mutex); | |
f2e85d57 | 1155 | |
2bd59d48 | 1156 | kernfs_destroy_root(root->kf_root); |
f2e85d57 TH |
1157 | cgroup_free_root(root); |
1158 | } | |
1159 | ||
ceb6a081 TH |
1160 | /* look up cgroup associated with given css_set on the specified hierarchy */ |
1161 | static struct cgroup *cset_cgroup_from_root(struct css_set *cset, | |
3dd06ffa | 1162 | struct cgroup_root *root) |
7717f7ba | 1163 | { |
7717f7ba PM |
1164 | struct cgroup *res = NULL; |
1165 | ||
96d365e0 | 1166 | lockdep_assert_held(&cgroup_mutex); |
f0d9a5f1 | 1167 | lockdep_assert_held(&css_set_lock); |
96d365e0 | 1168 | |
5abb8855 | 1169 | if (cset == &init_css_set) { |
3dd06ffa | 1170 | res = &root->cgrp; |
7717f7ba | 1171 | } else { |
69d0206c TH |
1172 | struct cgrp_cset_link *link; |
1173 | ||
1174 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) { | |
7717f7ba | 1175 | struct cgroup *c = link->cgrp; |
69d0206c | 1176 | |
7717f7ba PM |
1177 | if (c->root == root) { |
1178 | res = c; | |
1179 | break; | |
1180 | } | |
1181 | } | |
1182 | } | |
96d365e0 | 1183 | |
7717f7ba PM |
1184 | BUG_ON(!res); |
1185 | return res; | |
1186 | } | |
1187 | ||
ddbcc7e8 | 1188 | /* |
ceb6a081 | 1189 | * Return the cgroup for "task" from the given hierarchy. Must be |
f0d9a5f1 | 1190 | * called with cgroup_mutex and css_set_lock held. |
ceb6a081 TH |
1191 | */ |
1192 | static struct cgroup *task_cgroup_from_root(struct task_struct *task, | |
3dd06ffa | 1193 | struct cgroup_root *root) |
ceb6a081 TH |
1194 | { |
1195 | /* | |
1196 | * No need to lock the task - since we hold cgroup_mutex the | |
1197 | * task can't change groups, so the only thing that can happen | |
1198 | * is that it exits and its css is set back to init_css_set. | |
1199 | */ | |
1200 | return cset_cgroup_from_root(task_css_set(task), root); | |
1201 | } | |
1202 | ||
ddbcc7e8 | 1203 | /* |
ddbcc7e8 PM |
1204 | * A task must hold cgroup_mutex to modify cgroups. |
1205 | * | |
1206 | * Any task can increment and decrement the count field without lock. | |
1207 | * So in general, code holding cgroup_mutex can't rely on the count | |
1208 | * field not changing. However, if the count goes to zero, then only | |
956db3ca | 1209 | * cgroup_attach_task() can increment it again. Because a count of zero |
ddbcc7e8 PM |
1210 | * means that no tasks are currently attached, therefore there is no |
1211 | * way a task attached to that cgroup can fork (the other way to | |
1212 | * increment the count). So code holding cgroup_mutex can safely | |
1213 | * assume that if the count is zero, it will stay zero. Similarly, if | |
1214 | * a task holds cgroup_mutex on a cgroup with zero count, it | |
1215 | * knows that the cgroup won't be removed, as cgroup_rmdir() | |
1216 | * needs that mutex. | |
1217 | * | |
ddbcc7e8 PM |
1218 | * A cgroup can only be deleted if both its 'count' of using tasks |
1219 | * is zero, and its list of 'children' cgroups is empty. Since all | |
1220 | * tasks in the system use _some_ cgroup, and since there is always at | |
3dd06ffa | 1221 | * least one task in the system (init, pid == 1), therefore, root cgroup |
ddbcc7e8 | 1222 | * always has either children cgroups and/or using tasks. So we don't |
3dd06ffa | 1223 | * need a special hack to ensure that root cgroup cannot be deleted. |
ddbcc7e8 PM |
1224 | * |
1225 | * P.S. One more locking exception. RCU is used to guard the | |
956db3ca | 1226 | * update of a tasks cgroup pointer by cgroup_attach_task() |
ddbcc7e8 PM |
1227 | */ |
1228 | ||
2bd59d48 | 1229 | static struct kernfs_syscall_ops cgroup_kf_syscall_ops; |
828c0950 | 1230 | static const struct file_operations proc_cgroupstats_operations; |
a424316c | 1231 | |
8d7e6fb0 TH |
1232 | static char *cgroup_file_name(struct cgroup *cgrp, const struct cftype *cft, |
1233 | char *buf) | |
ddbcc7e8 | 1234 | { |
3e1d2eed TH |
1235 | struct cgroup_subsys *ss = cft->ss; |
1236 | ||
8d7e6fb0 TH |
1237 | if (cft->ss && !(cft->flags & CFTYPE_NO_PREFIX) && |
1238 | !(cgrp->root->flags & CGRP_ROOT_NOPREFIX)) | |
1239 | snprintf(buf, CGROUP_FILE_NAME_MAX, "%s.%s", | |
3e1d2eed TH |
1240 | cgroup_on_dfl(cgrp) ? ss->name : ss->legacy_name, |
1241 | cft->name); | |
8d7e6fb0 TH |
1242 | else |
1243 | strncpy(buf, cft->name, CGROUP_FILE_NAME_MAX); | |
1244 | return buf; | |
ddbcc7e8 PM |
1245 | } |
1246 | ||
f2e85d57 TH |
1247 | /** |
1248 | * cgroup_file_mode - deduce file mode of a control file | |
1249 | * @cft: the control file in question | |
1250 | * | |
7dbdb199 | 1251 | * S_IRUGO for read, S_IWUSR for write. |
f2e85d57 TH |
1252 | */ |
1253 | static umode_t cgroup_file_mode(const struct cftype *cft) | |
65dff759 | 1254 | { |
f2e85d57 | 1255 | umode_t mode = 0; |
65dff759 | 1256 | |
f2e85d57 TH |
1257 | if (cft->read_u64 || cft->read_s64 || cft->seq_show) |
1258 | mode |= S_IRUGO; | |
1259 | ||
7dbdb199 TH |
1260 | if (cft->write_u64 || cft->write_s64 || cft->write) { |
1261 | if (cft->flags & CFTYPE_WORLD_WRITABLE) | |
1262 | mode |= S_IWUGO; | |
1263 | else | |
1264 | mode |= S_IWUSR; | |
1265 | } | |
f2e85d57 TH |
1266 | |
1267 | return mode; | |
65dff759 LZ |
1268 | } |
1269 | ||
af0ba678 | 1270 | /** |
0f060deb | 1271 | * cgroup_calc_child_subsys_mask - calculate child_subsys_mask |
af0ba678 | 1272 | * @cgrp: the target cgroup |
0f060deb | 1273 | * @subtree_control: the new subtree_control mask to consider |
af0ba678 TH |
1274 | * |
1275 | * On the default hierarchy, a subsystem may request other subsystems to be | |
1276 | * enabled together through its ->depends_on mask. In such cases, more | |
1277 | * subsystems than specified in "cgroup.subtree_control" may be enabled. | |
1278 | * | |
0f060deb TH |
1279 | * This function calculates which subsystems need to be enabled if |
1280 | * @subtree_control is to be applied to @cgrp. The returned mask is always | |
1281 | * a superset of @subtree_control and follows the usual hierarchy rules. | |
af0ba678 | 1282 | */ |
8ab456ac AS |
1283 | static unsigned long cgroup_calc_child_subsys_mask(struct cgroup *cgrp, |
1284 | unsigned long subtree_control) | |
667c2491 | 1285 | { |
af0ba678 | 1286 | struct cgroup *parent = cgroup_parent(cgrp); |
8ab456ac | 1287 | unsigned long cur_ss_mask = subtree_control; |
af0ba678 TH |
1288 | struct cgroup_subsys *ss; |
1289 | int ssid; | |
1290 | ||
1291 | lockdep_assert_held(&cgroup_mutex); | |
1292 | ||
0f060deb TH |
1293 | if (!cgroup_on_dfl(cgrp)) |
1294 | return cur_ss_mask; | |
af0ba678 TH |
1295 | |
1296 | while (true) { | |
8ab456ac | 1297 | unsigned long new_ss_mask = cur_ss_mask; |
af0ba678 | 1298 | |
a966a4ed AS |
1299 | for_each_subsys_which(ss, ssid, &cur_ss_mask) |
1300 | new_ss_mask |= ss->depends_on; | |
af0ba678 TH |
1301 | |
1302 | /* | |
1303 | * Mask out subsystems which aren't available. This can | |
1304 | * happen only if some depended-upon subsystems were bound | |
1305 | * to non-default hierarchies. | |
1306 | */ | |
1307 | if (parent) | |
1308 | new_ss_mask &= parent->child_subsys_mask; | |
1309 | else | |
1310 | new_ss_mask &= cgrp->root->subsys_mask; | |
1311 | ||
1312 | if (new_ss_mask == cur_ss_mask) | |
1313 | break; | |
1314 | cur_ss_mask = new_ss_mask; | |
1315 | } | |
1316 | ||
0f060deb TH |
1317 | return cur_ss_mask; |
1318 | } | |
1319 | ||
1320 | /** | |
1321 | * cgroup_refresh_child_subsys_mask - update child_subsys_mask | |
1322 | * @cgrp: the target cgroup | |
1323 | * | |
1324 | * Update @cgrp->child_subsys_mask according to the current | |
1325 | * @cgrp->subtree_control using cgroup_calc_child_subsys_mask(). | |
1326 | */ | |
1327 | static void cgroup_refresh_child_subsys_mask(struct cgroup *cgrp) | |
1328 | { | |
1329 | cgrp->child_subsys_mask = | |
1330 | cgroup_calc_child_subsys_mask(cgrp, cgrp->subtree_control); | |
667c2491 TH |
1331 | } |
1332 | ||
a9746d8d TH |
1333 | /** |
1334 | * cgroup_kn_unlock - unlocking helper for cgroup kernfs methods | |
1335 | * @kn: the kernfs_node being serviced | |
1336 | * | |
1337 | * This helper undoes cgroup_kn_lock_live() and should be invoked before | |
1338 | * the method finishes if locking succeeded. Note that once this function | |
1339 | * returns the cgroup returned by cgroup_kn_lock_live() may become | |
1340 | * inaccessible any time. If the caller intends to continue to access the | |
1341 | * cgroup, it should pin it before invoking this function. | |
1342 | */ | |
1343 | static void cgroup_kn_unlock(struct kernfs_node *kn) | |
ddbcc7e8 | 1344 | { |
a9746d8d TH |
1345 | struct cgroup *cgrp; |
1346 | ||
1347 | if (kernfs_type(kn) == KERNFS_DIR) | |
1348 | cgrp = kn->priv; | |
1349 | else | |
1350 | cgrp = kn->parent->priv; | |
1351 | ||
1352 | mutex_unlock(&cgroup_mutex); | |
a9746d8d TH |
1353 | |
1354 | kernfs_unbreak_active_protection(kn); | |
1355 | cgroup_put(cgrp); | |
ddbcc7e8 PM |
1356 | } |
1357 | ||
a9746d8d TH |
1358 | /** |
1359 | * cgroup_kn_lock_live - locking helper for cgroup kernfs methods | |
1360 | * @kn: the kernfs_node being serviced | |
1361 | * | |
1362 | * This helper is to be used by a cgroup kernfs method currently servicing | |
1363 | * @kn. It breaks the active protection, performs cgroup locking and | |
1364 | * verifies that the associated cgroup is alive. Returns the cgroup if | |
1365 | * alive; otherwise, %NULL. A successful return should be undone by a | |
1366 | * matching cgroup_kn_unlock() invocation. | |
1367 | * | |
1368 | * Any cgroup kernfs method implementation which requires locking the | |
1369 | * associated cgroup should use this helper. It avoids nesting cgroup | |
1370 | * locking under kernfs active protection and allows all kernfs operations | |
1371 | * including self-removal. | |
1372 | */ | |
1373 | static struct cgroup *cgroup_kn_lock_live(struct kernfs_node *kn) | |
05ef1d7c | 1374 | { |
a9746d8d TH |
1375 | struct cgroup *cgrp; |
1376 | ||
1377 | if (kernfs_type(kn) == KERNFS_DIR) | |
1378 | cgrp = kn->priv; | |
1379 | else | |
1380 | cgrp = kn->parent->priv; | |
05ef1d7c | 1381 | |
2739d3cc | 1382 | /* |
01f6474c | 1383 | * We're gonna grab cgroup_mutex which nests outside kernfs |
a9746d8d TH |
1384 | * active_ref. cgroup liveliness check alone provides enough |
1385 | * protection against removal. Ensure @cgrp stays accessible and | |
1386 | * break the active_ref protection. | |
2739d3cc | 1387 | */ |
aa32362f LZ |
1388 | if (!cgroup_tryget(cgrp)) |
1389 | return NULL; | |
a9746d8d TH |
1390 | kernfs_break_active_protection(kn); |
1391 | ||
2bd59d48 | 1392 | mutex_lock(&cgroup_mutex); |
05ef1d7c | 1393 | |
a9746d8d TH |
1394 | if (!cgroup_is_dead(cgrp)) |
1395 | return cgrp; | |
1396 | ||
1397 | cgroup_kn_unlock(kn); | |
1398 | return NULL; | |
ddbcc7e8 | 1399 | } |
05ef1d7c | 1400 | |
2739d3cc | 1401 | static void cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft) |
05ef1d7c | 1402 | { |
2bd59d48 | 1403 | char name[CGROUP_FILE_NAME_MAX]; |
05ef1d7c | 1404 | |
01f6474c | 1405 | lockdep_assert_held(&cgroup_mutex); |
34c06254 TH |
1406 | |
1407 | if (cft->file_offset) { | |
1408 | struct cgroup_subsys_state *css = cgroup_css(cgrp, cft->ss); | |
1409 | struct cgroup_file *cfile = (void *)css + cft->file_offset; | |
1410 | ||
1411 | spin_lock_irq(&cgroup_file_kn_lock); | |
1412 | cfile->kn = NULL; | |
1413 | spin_unlock_irq(&cgroup_file_kn_lock); | |
1414 | } | |
1415 | ||
2bd59d48 | 1416 | kernfs_remove_by_name(cgrp->kn, cgroup_file_name(cgrp, cft, name)); |
05ef1d7c TH |
1417 | } |
1418 | ||
13af07df | 1419 | /** |
4df8dc90 TH |
1420 | * css_clear_dir - remove subsys files in a cgroup directory |
1421 | * @css: taget css | |
1422 | * @cgrp_override: specify if target cgroup is different from css->cgroup | |
13af07df | 1423 | */ |
4df8dc90 TH |
1424 | static void css_clear_dir(struct cgroup_subsys_state *css, |
1425 | struct cgroup *cgrp_override) | |
05ef1d7c | 1426 | { |
4df8dc90 TH |
1427 | struct cgroup *cgrp = cgrp_override ?: css->cgroup; |
1428 | struct cftype *cfts; | |
05ef1d7c | 1429 | |
4df8dc90 TH |
1430 | list_for_each_entry(cfts, &css->ss->cfts, node) |
1431 | cgroup_addrm_files(css, cgrp, cfts, false); | |
ddbcc7e8 PM |
1432 | } |
1433 | ||
ccdca218 | 1434 | /** |
4df8dc90 TH |
1435 | * css_populate_dir - create subsys files in a cgroup directory |
1436 | * @css: target css | |
1437 | * @cgrp_overried: specify if target cgroup is different from css->cgroup | |
ccdca218 TH |
1438 | * |
1439 | * On failure, no file is added. | |
1440 | */ | |
4df8dc90 TH |
1441 | static int css_populate_dir(struct cgroup_subsys_state *css, |
1442 | struct cgroup *cgrp_override) | |
ccdca218 | 1443 | { |
4df8dc90 TH |
1444 | struct cgroup *cgrp = cgrp_override ?: css->cgroup; |
1445 | struct cftype *cfts, *failed_cfts; | |
1446 | int ret; | |
ccdca218 | 1447 | |
4df8dc90 TH |
1448 | if (!css->ss) { |
1449 | if (cgroup_on_dfl(cgrp)) | |
1450 | cfts = cgroup_dfl_base_files; | |
1451 | else | |
1452 | cfts = cgroup_legacy_base_files; | |
ccdca218 | 1453 | |
4df8dc90 TH |
1454 | return cgroup_addrm_files(&cgrp->self, cgrp, cfts, true); |
1455 | } | |
ccdca218 | 1456 | |
4df8dc90 TH |
1457 | list_for_each_entry(cfts, &css->ss->cfts, node) { |
1458 | ret = cgroup_addrm_files(css, cgrp, cfts, true); | |
1459 | if (ret < 0) { | |
1460 | failed_cfts = cfts; | |
1461 | goto err; | |
ccdca218 TH |
1462 | } |
1463 | } | |
1464 | return 0; | |
1465 | err: | |
4df8dc90 TH |
1466 | list_for_each_entry(cfts, &css->ss->cfts, node) { |
1467 | if (cfts == failed_cfts) | |
1468 | break; | |
1469 | cgroup_addrm_files(css, cgrp, cfts, false); | |
1470 | } | |
ccdca218 TH |
1471 | return ret; |
1472 | } | |
1473 | ||
8ab456ac AS |
1474 | static int rebind_subsystems(struct cgroup_root *dst_root, |
1475 | unsigned long ss_mask) | |
ddbcc7e8 | 1476 | { |
1ada4838 | 1477 | struct cgroup *dcgrp = &dst_root->cgrp; |
30159ec7 | 1478 | struct cgroup_subsys *ss; |
8ab456ac | 1479 | unsigned long tmp_ss_mask; |
2d8f243a | 1480 | int ssid, i, ret; |
ddbcc7e8 | 1481 | |
ace2bee8 | 1482 | lockdep_assert_held(&cgroup_mutex); |
ddbcc7e8 | 1483 | |
a966a4ed | 1484 | for_each_subsys_which(ss, ssid, &ss_mask) { |
7fd8c565 TH |
1485 | /* if @ss has non-root csses attached to it, can't move */ |
1486 | if (css_next_child(NULL, cgroup_css(&ss->root->cgrp, ss))) | |
3ed80a62 | 1487 | return -EBUSY; |
1d5be6b2 | 1488 | |
5df36032 | 1489 | /* can't move between two non-dummy roots either */ |
7fd8c565 | 1490 | if (ss->root != &cgrp_dfl_root && dst_root != &cgrp_dfl_root) |
5df36032 | 1491 | return -EBUSY; |
ddbcc7e8 PM |
1492 | } |
1493 | ||
5533e011 TH |
1494 | /* skip creating root files on dfl_root for inhibited subsystems */ |
1495 | tmp_ss_mask = ss_mask; | |
1496 | if (dst_root == &cgrp_dfl_root) | |
1497 | tmp_ss_mask &= ~cgrp_dfl_root_inhibit_ss_mask; | |
1498 | ||
4df8dc90 TH |
1499 | for_each_subsys_which(ss, ssid, &tmp_ss_mask) { |
1500 | struct cgroup *scgrp = &ss->root->cgrp; | |
1501 | int tssid; | |
1502 | ||
1503 | ret = css_populate_dir(cgroup_css(scgrp, ss), dcgrp); | |
1504 | if (!ret) | |
1505 | continue; | |
ddbcc7e8 | 1506 | |
a2dd4247 TH |
1507 | /* |
1508 | * Rebinding back to the default root is not allowed to | |
1509 | * fail. Using both default and non-default roots should | |
1510 | * be rare. Moving subsystems back and forth even more so. | |
1511 | * Just warn about it and continue. | |
1512 | */ | |
4df8dc90 TH |
1513 | if (dst_root == &cgrp_dfl_root) { |
1514 | if (cgrp_dfl_root_visible) { | |
1515 | pr_warn("failed to create files (%d) while rebinding 0x%lx to default root\n", | |
1516 | ret, ss_mask); | |
1517 | pr_warn("you may retry by moving them to a different hierarchy and unbinding\n"); | |
1518 | } | |
1519 | continue; | |
a2dd4247 | 1520 | } |
4df8dc90 TH |
1521 | |
1522 | for_each_subsys_which(ss, tssid, &tmp_ss_mask) { | |
1523 | if (tssid == ssid) | |
1524 | break; | |
1525 | css_clear_dir(cgroup_css(scgrp, ss), dcgrp); | |
1526 | } | |
1527 | return ret; | |
5df36032 | 1528 | } |
3126121f TH |
1529 | |
1530 | /* | |
1531 | * Nothing can fail from this point on. Remove files for the | |
1532 | * removed subsystems and rebind each subsystem. | |
1533 | */ | |
a966a4ed | 1534 | for_each_subsys_which(ss, ssid, &ss_mask) { |
1ada4838 TH |
1535 | struct cgroup_root *src_root = ss->root; |
1536 | struct cgroup *scgrp = &src_root->cgrp; | |
1537 | struct cgroup_subsys_state *css = cgroup_css(scgrp, ss); | |
2d8f243a | 1538 | struct css_set *cset; |
a8a648c4 | 1539 | |
1ada4838 | 1540 | WARN_ON(!css || cgroup_css(dcgrp, ss)); |
a8a648c4 | 1541 | |
4df8dc90 TH |
1542 | css_clear_dir(css, NULL); |
1543 | ||
1ada4838 TH |
1544 | RCU_INIT_POINTER(scgrp->subsys[ssid], NULL); |
1545 | rcu_assign_pointer(dcgrp->subsys[ssid], css); | |
5df36032 | 1546 | ss->root = dst_root; |
1ada4838 | 1547 | css->cgroup = dcgrp; |
73e80ed8 | 1548 | |
f0d9a5f1 | 1549 | spin_lock_bh(&css_set_lock); |
2d8f243a TH |
1550 | hash_for_each(css_set_table, i, cset, hlist) |
1551 | list_move_tail(&cset->e_cset_node[ss->id], | |
1ada4838 | 1552 | &dcgrp->e_csets[ss->id]); |
f0d9a5f1 | 1553 | spin_unlock_bh(&css_set_lock); |
2d8f243a | 1554 | |
f392e51c | 1555 | src_root->subsys_mask &= ~(1 << ssid); |
1ada4838 TH |
1556 | scgrp->subtree_control &= ~(1 << ssid); |
1557 | cgroup_refresh_child_subsys_mask(scgrp); | |
f392e51c | 1558 | |
bd53d617 | 1559 | /* default hierarchy doesn't enable controllers by default */ |
f392e51c | 1560 | dst_root->subsys_mask |= 1 << ssid; |
49d1dc4b TH |
1561 | if (dst_root == &cgrp_dfl_root) { |
1562 | static_branch_enable(cgroup_subsys_on_dfl_key[ssid]); | |
1563 | } else { | |
1ada4838 TH |
1564 | dcgrp->subtree_control |= 1 << ssid; |
1565 | cgroup_refresh_child_subsys_mask(dcgrp); | |
49d1dc4b | 1566 | static_branch_disable(cgroup_subsys_on_dfl_key[ssid]); |
667c2491 | 1567 | } |
a8a648c4 | 1568 | |
5df36032 TH |
1569 | if (ss->bind) |
1570 | ss->bind(css); | |
ddbcc7e8 | 1571 | } |
ddbcc7e8 | 1572 | |
1ada4838 | 1573 | kernfs_activate(dcgrp->kn); |
ddbcc7e8 PM |
1574 | return 0; |
1575 | } | |
1576 | ||
2bd59d48 TH |
1577 | static int cgroup_show_options(struct seq_file *seq, |
1578 | struct kernfs_root *kf_root) | |
ddbcc7e8 | 1579 | { |
3dd06ffa | 1580 | struct cgroup_root *root = cgroup_root_from_kf(kf_root); |
ddbcc7e8 | 1581 | struct cgroup_subsys *ss; |
b85d2040 | 1582 | int ssid; |
ddbcc7e8 | 1583 | |
d98817d4 TH |
1584 | if (root != &cgrp_dfl_root) |
1585 | for_each_subsys(ss, ssid) | |
1586 | if (root->subsys_mask & (1 << ssid)) | |
61e57c0c | 1587 | seq_show_option(seq, ss->legacy_name, NULL); |
93438629 | 1588 | if (root->flags & CGRP_ROOT_NOPREFIX) |
ddbcc7e8 | 1589 | seq_puts(seq, ",noprefix"); |
93438629 | 1590 | if (root->flags & CGRP_ROOT_XATTR) |
03b1cde6 | 1591 | seq_puts(seq, ",xattr"); |
69e943b7 TH |
1592 | |
1593 | spin_lock(&release_agent_path_lock); | |
81a6a5cd | 1594 | if (strlen(root->release_agent_path)) |
a068acf2 KC |
1595 | seq_show_option(seq, "release_agent", |
1596 | root->release_agent_path); | |
69e943b7 TH |
1597 | spin_unlock(&release_agent_path_lock); |
1598 | ||
3dd06ffa | 1599 | if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags)) |
97978e6d | 1600 | seq_puts(seq, ",clone_children"); |
c6d57f33 | 1601 | if (strlen(root->name)) |
a068acf2 | 1602 | seq_show_option(seq, "name", root->name); |
ddbcc7e8 PM |
1603 | return 0; |
1604 | } | |
1605 | ||
1606 | struct cgroup_sb_opts { | |
8ab456ac | 1607 | unsigned long subsys_mask; |
69dfa00c | 1608 | unsigned int flags; |
81a6a5cd | 1609 | char *release_agent; |
2260e7fc | 1610 | bool cpuset_clone_children; |
c6d57f33 | 1611 | char *name; |
2c6ab6d2 PM |
1612 | /* User explicitly requested empty subsystem */ |
1613 | bool none; | |
ddbcc7e8 PM |
1614 | }; |
1615 | ||
cf5d5941 | 1616 | static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts) |
ddbcc7e8 | 1617 | { |
32a8cf23 DL |
1618 | char *token, *o = data; |
1619 | bool all_ss = false, one_ss = false; | |
8ab456ac | 1620 | unsigned long mask = -1UL; |
30159ec7 | 1621 | struct cgroup_subsys *ss; |
7b9a6ba5 | 1622 | int nr_opts = 0; |
30159ec7 | 1623 | int i; |
f9ab5b5b LZ |
1624 | |
1625 | #ifdef CONFIG_CPUSETS | |
69dfa00c | 1626 | mask = ~(1U << cpuset_cgrp_id); |
f9ab5b5b | 1627 | #endif |
ddbcc7e8 | 1628 | |
c6d57f33 | 1629 | memset(opts, 0, sizeof(*opts)); |
ddbcc7e8 PM |
1630 | |
1631 | while ((token = strsep(&o, ",")) != NULL) { | |
7b9a6ba5 TH |
1632 | nr_opts++; |
1633 | ||
ddbcc7e8 PM |
1634 | if (!*token) |
1635 | return -EINVAL; | |
32a8cf23 | 1636 | if (!strcmp(token, "none")) { |
2c6ab6d2 PM |
1637 | /* Explicitly have no subsystems */ |
1638 | opts->none = true; | |
32a8cf23 DL |
1639 | continue; |
1640 | } | |
1641 | if (!strcmp(token, "all")) { | |
1642 | /* Mutually exclusive option 'all' + subsystem name */ | |
1643 | if (one_ss) | |
1644 | return -EINVAL; | |
1645 | all_ss = true; | |
1646 | continue; | |
1647 | } | |
1648 | if (!strcmp(token, "noprefix")) { | |
93438629 | 1649 | opts->flags |= CGRP_ROOT_NOPREFIX; |
32a8cf23 DL |
1650 | continue; |
1651 | } | |
1652 | if (!strcmp(token, "clone_children")) { | |
2260e7fc | 1653 | opts->cpuset_clone_children = true; |
32a8cf23 DL |
1654 | continue; |
1655 | } | |
03b1cde6 | 1656 | if (!strcmp(token, "xattr")) { |
93438629 | 1657 | opts->flags |= CGRP_ROOT_XATTR; |
03b1cde6 AR |
1658 | continue; |
1659 | } | |
32a8cf23 | 1660 | if (!strncmp(token, "release_agent=", 14)) { |
81a6a5cd PM |
1661 | /* Specifying two release agents is forbidden */ |
1662 | if (opts->release_agent) | |
1663 | return -EINVAL; | |
c6d57f33 | 1664 | opts->release_agent = |
e400c285 | 1665 | kstrndup(token + 14, PATH_MAX - 1, GFP_KERNEL); |
81a6a5cd PM |
1666 | if (!opts->release_agent) |
1667 | return -ENOMEM; | |
32a8cf23 DL |
1668 | continue; |
1669 | } | |
1670 | if (!strncmp(token, "name=", 5)) { | |
c6d57f33 PM |
1671 | const char *name = token + 5; |
1672 | /* Can't specify an empty name */ | |
1673 | if (!strlen(name)) | |
1674 | return -EINVAL; | |
1675 | /* Must match [\w.-]+ */ | |
1676 | for (i = 0; i < strlen(name); i++) { | |
1677 | char c = name[i]; | |
1678 | if (isalnum(c)) | |
1679 | continue; | |
1680 | if ((c == '.') || (c == '-') || (c == '_')) | |
1681 | continue; | |
1682 | return -EINVAL; | |
1683 | } | |
1684 | /* Specifying two names is forbidden */ | |
1685 | if (opts->name) | |
1686 | return -EINVAL; | |
1687 | opts->name = kstrndup(name, | |
e400c285 | 1688 | MAX_CGROUP_ROOT_NAMELEN - 1, |
c6d57f33 PM |
1689 | GFP_KERNEL); |
1690 | if (!opts->name) | |
1691 | return -ENOMEM; | |
32a8cf23 DL |
1692 | |
1693 | continue; | |
1694 | } | |
1695 | ||
30159ec7 | 1696 | for_each_subsys(ss, i) { |
3e1d2eed | 1697 | if (strcmp(token, ss->legacy_name)) |
32a8cf23 | 1698 | continue; |
fc5ed1e9 | 1699 | if (!cgroup_ssid_enabled(i)) |
32a8cf23 | 1700 | continue; |
223ffb29 JW |
1701 | if (cgroup_ssid_no_v1(i)) |
1702 | continue; | |
32a8cf23 DL |
1703 | |
1704 | /* Mutually exclusive option 'all' + subsystem name */ | |
1705 | if (all_ss) | |
1706 | return -EINVAL; | |
69dfa00c | 1707 | opts->subsys_mask |= (1 << i); |
32a8cf23 DL |
1708 | one_ss = true; |
1709 | ||
1710 | break; | |
1711 | } | |
1712 | if (i == CGROUP_SUBSYS_COUNT) | |
1713 | return -ENOENT; | |
1714 | } | |
1715 | ||
7b9a6ba5 TH |
1716 | /* |
1717 | * If the 'all' option was specified select all the subsystems, | |
1718 | * otherwise if 'none', 'name=' and a subsystem name options were | |
1719 | * not specified, let's default to 'all' | |
1720 | */ | |
1721 | if (all_ss || (!one_ss && !opts->none && !opts->name)) | |
1722 | for_each_subsys(ss, i) | |
223ffb29 | 1723 | if (cgroup_ssid_enabled(i) && !cgroup_ssid_no_v1(i)) |
7b9a6ba5 TH |
1724 | opts->subsys_mask |= (1 << i); |
1725 | ||
1726 | /* | |
1727 | * We either have to specify by name or by subsystems. (So all | |
1728 | * empty hierarchies must have a name). | |
1729 | */ | |
1730 | if (!opts->subsys_mask && !opts->name) | |
1731 | return -EINVAL; | |
1732 | ||
f9ab5b5b LZ |
1733 | /* |
1734 | * Option noprefix was introduced just for backward compatibility | |
1735 | * with the old cpuset, so we allow noprefix only if mounting just | |
1736 | * the cpuset subsystem. | |
1737 | */ | |
93438629 | 1738 | if ((opts->flags & CGRP_ROOT_NOPREFIX) && (opts->subsys_mask & mask)) |
f9ab5b5b LZ |
1739 | return -EINVAL; |
1740 | ||
2c6ab6d2 | 1741 | /* Can't specify "none" and some subsystems */ |
a1a71b45 | 1742 | if (opts->subsys_mask && opts->none) |
2c6ab6d2 PM |
1743 | return -EINVAL; |
1744 | ||
ddbcc7e8 PM |
1745 | return 0; |
1746 | } | |
1747 | ||
2bd59d48 | 1748 | static int cgroup_remount(struct kernfs_root *kf_root, int *flags, char *data) |
ddbcc7e8 PM |
1749 | { |
1750 | int ret = 0; | |
3dd06ffa | 1751 | struct cgroup_root *root = cgroup_root_from_kf(kf_root); |
ddbcc7e8 | 1752 | struct cgroup_sb_opts opts; |
8ab456ac | 1753 | unsigned long added_mask, removed_mask; |
ddbcc7e8 | 1754 | |
aa6ec29b TH |
1755 | if (root == &cgrp_dfl_root) { |
1756 | pr_err("remount is not allowed\n"); | |
873fe09e TH |
1757 | return -EINVAL; |
1758 | } | |
1759 | ||
ddbcc7e8 PM |
1760 | mutex_lock(&cgroup_mutex); |
1761 | ||
1762 | /* See what subsystems are wanted */ | |
1763 | ret = parse_cgroupfs_options(data, &opts); | |
1764 | if (ret) | |
1765 | goto out_unlock; | |
1766 | ||
f392e51c | 1767 | if (opts.subsys_mask != root->subsys_mask || opts.release_agent) |
ed3d261b | 1768 | pr_warn("option changes via remount are deprecated (pid=%d comm=%s)\n", |
a2a1f9ea | 1769 | task_tgid_nr(current), current->comm); |
8b5a5a9d | 1770 | |
f392e51c TH |
1771 | added_mask = opts.subsys_mask & ~root->subsys_mask; |
1772 | removed_mask = root->subsys_mask & ~opts.subsys_mask; | |
13af07df | 1773 | |
cf5d5941 | 1774 | /* Don't allow flags or name to change at remount */ |
7450e90b | 1775 | if ((opts.flags ^ root->flags) || |
cf5d5941 | 1776 | (opts.name && strcmp(opts.name, root->name))) { |
69dfa00c | 1777 | pr_err("option or name mismatch, new: 0x%x \"%s\", old: 0x%x \"%s\"\n", |
7450e90b | 1778 | opts.flags, opts.name ?: "", root->flags, root->name); |
c6d57f33 PM |
1779 | ret = -EINVAL; |
1780 | goto out_unlock; | |
1781 | } | |
1782 | ||
f172e67c | 1783 | /* remounting is not allowed for populated hierarchies */ |
d5c419b6 | 1784 | if (!list_empty(&root->cgrp.self.children)) { |
f172e67c | 1785 | ret = -EBUSY; |
0670e08b | 1786 | goto out_unlock; |
cf5d5941 | 1787 | } |
ddbcc7e8 | 1788 | |
5df36032 | 1789 | ret = rebind_subsystems(root, added_mask); |
3126121f | 1790 | if (ret) |
0670e08b | 1791 | goto out_unlock; |
ddbcc7e8 | 1792 | |
3dd06ffa | 1793 | rebind_subsystems(&cgrp_dfl_root, removed_mask); |
5df36032 | 1794 | |
69e943b7 TH |
1795 | if (opts.release_agent) { |
1796 | spin_lock(&release_agent_path_lock); | |
81a6a5cd | 1797 | strcpy(root->release_agent_path, opts.release_agent); |
69e943b7 TH |
1798 | spin_unlock(&release_agent_path_lock); |
1799 | } | |
ddbcc7e8 | 1800 | out_unlock: |
66bdc9cf | 1801 | kfree(opts.release_agent); |
c6d57f33 | 1802 | kfree(opts.name); |
ddbcc7e8 | 1803 | mutex_unlock(&cgroup_mutex); |
ddbcc7e8 PM |
1804 | return ret; |
1805 | } | |
1806 | ||
afeb0f9f TH |
1807 | /* |
1808 | * To reduce the fork() overhead for systems that are not actually using | |
1809 | * their cgroups capability, we don't maintain the lists running through | |
1810 | * each css_set to its tasks until we see the list actually used - in other | |
1811 | * words after the first mount. | |
1812 | */ | |
1813 | static bool use_task_css_set_links __read_mostly; | |
1814 | ||
1815 | static void cgroup_enable_task_cg_lists(void) | |
1816 | { | |
1817 | struct task_struct *p, *g; | |
1818 | ||
f0d9a5f1 | 1819 | spin_lock_bh(&css_set_lock); |
afeb0f9f TH |
1820 | |
1821 | if (use_task_css_set_links) | |
1822 | goto out_unlock; | |
1823 | ||
1824 | use_task_css_set_links = true; | |
1825 | ||
1826 | /* | |
1827 | * We need tasklist_lock because RCU is not safe against | |
1828 | * while_each_thread(). Besides, a forking task that has passed | |
1829 | * cgroup_post_fork() without seeing use_task_css_set_links = 1 | |
1830 | * is not guaranteed to have its child immediately visible in the | |
1831 | * tasklist if we walk through it with RCU. | |
1832 | */ | |
1833 | read_lock(&tasklist_lock); | |
1834 | do_each_thread(g, p) { | |
afeb0f9f TH |
1835 | WARN_ON_ONCE(!list_empty(&p->cg_list) || |
1836 | task_css_set(p) != &init_css_set); | |
1837 | ||
1838 | /* | |
1839 | * We should check if the process is exiting, otherwise | |
1840 | * it will race with cgroup_exit() in that the list | |
1841 | * entry won't be deleted though the process has exited. | |
f153ad11 TH |
1842 | * Do it while holding siglock so that we don't end up |
1843 | * racing against cgroup_exit(). | |
afeb0f9f | 1844 | */ |
f153ad11 | 1845 | spin_lock_irq(&p->sighand->siglock); |
eaf797ab TH |
1846 | if (!(p->flags & PF_EXITING)) { |
1847 | struct css_set *cset = task_css_set(p); | |
1848 | ||
0de0942d TH |
1849 | if (!css_set_populated(cset)) |
1850 | css_set_update_populated(cset, true); | |
389b9c1b | 1851 | list_add_tail(&p->cg_list, &cset->tasks); |
eaf797ab TH |
1852 | get_css_set(cset); |
1853 | } | |
f153ad11 | 1854 | spin_unlock_irq(&p->sighand->siglock); |
afeb0f9f TH |
1855 | } while_each_thread(g, p); |
1856 | read_unlock(&tasklist_lock); | |
1857 | out_unlock: | |
f0d9a5f1 | 1858 | spin_unlock_bh(&css_set_lock); |
afeb0f9f | 1859 | } |
ddbcc7e8 | 1860 | |
cc31edce PM |
1861 | static void init_cgroup_housekeeping(struct cgroup *cgrp) |
1862 | { | |
2d8f243a TH |
1863 | struct cgroup_subsys *ss; |
1864 | int ssid; | |
1865 | ||
d5c419b6 TH |
1866 | INIT_LIST_HEAD(&cgrp->self.sibling); |
1867 | INIT_LIST_HEAD(&cgrp->self.children); | |
69d0206c | 1868 | INIT_LIST_HEAD(&cgrp->cset_links); |
72a8cb30 BB |
1869 | INIT_LIST_HEAD(&cgrp->pidlists); |
1870 | mutex_init(&cgrp->pidlist_mutex); | |
9d800df1 | 1871 | cgrp->self.cgroup = cgrp; |
184faf32 | 1872 | cgrp->self.flags |= CSS_ONLINE; |
2d8f243a TH |
1873 | |
1874 | for_each_subsys(ss, ssid) | |
1875 | INIT_LIST_HEAD(&cgrp->e_csets[ssid]); | |
f8f22e53 TH |
1876 | |
1877 | init_waitqueue_head(&cgrp->offline_waitq); | |
971ff493 | 1878 | INIT_WORK(&cgrp->release_agent_work, cgroup_release_agent); |
cc31edce | 1879 | } |
c6d57f33 | 1880 | |
3dd06ffa | 1881 | static void init_cgroup_root(struct cgroup_root *root, |
172a2c06 | 1882 | struct cgroup_sb_opts *opts) |
ddbcc7e8 | 1883 | { |
3dd06ffa | 1884 | struct cgroup *cgrp = &root->cgrp; |
b0ca5a84 | 1885 | |
ddbcc7e8 | 1886 | INIT_LIST_HEAD(&root->root_list); |
3c9c825b | 1887 | atomic_set(&root->nr_cgrps, 1); |
bd89aabc | 1888 | cgrp->root = root; |
cc31edce | 1889 | init_cgroup_housekeeping(cgrp); |
4e96ee8e | 1890 | idr_init(&root->cgroup_idr); |
c6d57f33 | 1891 | |
c6d57f33 PM |
1892 | root->flags = opts->flags; |
1893 | if (opts->release_agent) | |
1894 | strcpy(root->release_agent_path, opts->release_agent); | |
1895 | if (opts->name) | |
1896 | strcpy(root->name, opts->name); | |
2260e7fc | 1897 | if (opts->cpuset_clone_children) |
3dd06ffa | 1898 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags); |
c6d57f33 PM |
1899 | } |
1900 | ||
8ab456ac | 1901 | static int cgroup_setup_root(struct cgroup_root *root, unsigned long ss_mask) |
2c6ab6d2 | 1902 | { |
d427dfeb | 1903 | LIST_HEAD(tmp_links); |
3dd06ffa | 1904 | struct cgroup *root_cgrp = &root->cgrp; |
d427dfeb | 1905 | struct css_set *cset; |
d427dfeb | 1906 | int i, ret; |
2c6ab6d2 | 1907 | |
d427dfeb | 1908 | lockdep_assert_held(&cgroup_mutex); |
c6d57f33 | 1909 | |
cf780b7d | 1910 | ret = cgroup_idr_alloc(&root->cgroup_idr, root_cgrp, 1, 2, GFP_KERNEL); |
d427dfeb | 1911 | if (ret < 0) |
2bd59d48 | 1912 | goto out; |
d427dfeb | 1913 | root_cgrp->id = ret; |
b11cfb58 | 1914 | root_cgrp->ancestor_ids[0] = ret; |
c6d57f33 | 1915 | |
2aad2a86 TH |
1916 | ret = percpu_ref_init(&root_cgrp->self.refcnt, css_release, 0, |
1917 | GFP_KERNEL); | |
9d755d33 TH |
1918 | if (ret) |
1919 | goto out; | |
1920 | ||
d427dfeb | 1921 | /* |
f0d9a5f1 | 1922 | * We're accessing css_set_count without locking css_set_lock here, |
d427dfeb TH |
1923 | * but that's OK - it can only be increased by someone holding |
1924 | * cgroup_lock, and that's us. The worst that can happen is that we | |
1925 | * have some link structures left over | |
1926 | */ | |
1927 | ret = allocate_cgrp_cset_links(css_set_count, &tmp_links); | |
1928 | if (ret) | |
9d755d33 | 1929 | goto cancel_ref; |
ddbcc7e8 | 1930 | |
985ed670 | 1931 | ret = cgroup_init_root_id(root); |
ddbcc7e8 | 1932 | if (ret) |
9d755d33 | 1933 | goto cancel_ref; |
ddbcc7e8 | 1934 | |
2bd59d48 TH |
1935 | root->kf_root = kernfs_create_root(&cgroup_kf_syscall_ops, |
1936 | KERNFS_ROOT_CREATE_DEACTIVATED, | |
1937 | root_cgrp); | |
1938 | if (IS_ERR(root->kf_root)) { | |
1939 | ret = PTR_ERR(root->kf_root); | |
1940 | goto exit_root_id; | |
1941 | } | |
1942 | root_cgrp->kn = root->kf_root->kn; | |
ddbcc7e8 | 1943 | |
4df8dc90 | 1944 | ret = css_populate_dir(&root_cgrp->self, NULL); |
d427dfeb | 1945 | if (ret) |
2bd59d48 | 1946 | goto destroy_root; |
ddbcc7e8 | 1947 | |
5df36032 | 1948 | ret = rebind_subsystems(root, ss_mask); |
d427dfeb | 1949 | if (ret) |
2bd59d48 | 1950 | goto destroy_root; |
ddbcc7e8 | 1951 | |
d427dfeb TH |
1952 | /* |
1953 | * There must be no failure case after here, since rebinding takes | |
1954 | * care of subsystems' refcounts, which are explicitly dropped in | |
1955 | * the failure exit path. | |
1956 | */ | |
1957 | list_add(&root->root_list, &cgroup_roots); | |
1958 | cgroup_root_count++; | |
0df6a63f | 1959 | |
d427dfeb | 1960 | /* |
3dd06ffa | 1961 | * Link the root cgroup in this hierarchy into all the css_set |
d427dfeb TH |
1962 | * objects. |
1963 | */ | |
f0d9a5f1 | 1964 | spin_lock_bh(&css_set_lock); |
0de0942d | 1965 | hash_for_each(css_set_table, i, cset, hlist) { |
d427dfeb | 1966 | link_css_set(&tmp_links, cset, root_cgrp); |
0de0942d TH |
1967 | if (css_set_populated(cset)) |
1968 | cgroup_update_populated(root_cgrp, true); | |
1969 | } | |
f0d9a5f1 | 1970 | spin_unlock_bh(&css_set_lock); |
ddbcc7e8 | 1971 | |
d5c419b6 | 1972 | BUG_ON(!list_empty(&root_cgrp->self.children)); |
3c9c825b | 1973 | BUG_ON(atomic_read(&root->nr_cgrps) != 1); |
ddbcc7e8 | 1974 | |
2bd59d48 | 1975 | kernfs_activate(root_cgrp->kn); |
d427dfeb | 1976 | ret = 0; |
2bd59d48 | 1977 | goto out; |
d427dfeb | 1978 | |
2bd59d48 TH |
1979 | destroy_root: |
1980 | kernfs_destroy_root(root->kf_root); | |
1981 | root->kf_root = NULL; | |
1982 | exit_root_id: | |
d427dfeb | 1983 | cgroup_exit_root_id(root); |
9d755d33 | 1984 | cancel_ref: |
9a1049da | 1985 | percpu_ref_exit(&root_cgrp->self.refcnt); |
2bd59d48 | 1986 | out: |
d427dfeb TH |
1987 | free_cgrp_cset_links(&tmp_links); |
1988 | return ret; | |
ddbcc7e8 PM |
1989 | } |
1990 | ||
f7e83571 | 1991 | static struct dentry *cgroup_mount(struct file_system_type *fs_type, |
ddbcc7e8 | 1992 | int flags, const char *unused_dev_name, |
f7e83571 | 1993 | void *data) |
ddbcc7e8 | 1994 | { |
67e9c74b | 1995 | bool is_v2 = fs_type == &cgroup2_fs_type; |
3a32bd72 | 1996 | struct super_block *pinned_sb = NULL; |
970317aa | 1997 | struct cgroup_subsys *ss; |
3dd06ffa | 1998 | struct cgroup_root *root; |
ddbcc7e8 | 1999 | struct cgroup_sb_opts opts; |
2bd59d48 | 2000 | struct dentry *dentry; |
8e30e2b8 | 2001 | int ret; |
970317aa | 2002 | int i; |
c6b3d5bc | 2003 | bool new_sb; |
ddbcc7e8 | 2004 | |
56fde9e0 TH |
2005 | /* |
2006 | * The first time anyone tries to mount a cgroup, enable the list | |
2007 | * linking each css_set to its tasks and fix up all existing tasks. | |
2008 | */ | |
2009 | if (!use_task_css_set_links) | |
2010 | cgroup_enable_task_cg_lists(); | |
e37a06f1 | 2011 | |
67e9c74b TH |
2012 | if (is_v2) { |
2013 | if (data) { | |
2014 | pr_err("cgroup2: unknown option \"%s\"\n", (char *)data); | |
2015 | return ERR_PTR(-EINVAL); | |
2016 | } | |
2017 | cgrp_dfl_root_visible = true; | |
2018 | root = &cgrp_dfl_root; | |
2019 | cgroup_get(&root->cgrp); | |
2020 | goto out_mount; | |
2021 | } | |
2022 | ||
aae8aab4 | 2023 | mutex_lock(&cgroup_mutex); |
8e30e2b8 TH |
2024 | |
2025 | /* First find the desired set of subsystems */ | |
ddbcc7e8 | 2026 | ret = parse_cgroupfs_options(data, &opts); |
c6d57f33 | 2027 | if (ret) |
8e30e2b8 | 2028 | goto out_unlock; |
a015edd2 | 2029 | |
970317aa LZ |
2030 | /* |
2031 | * Destruction of cgroup root is asynchronous, so subsystems may | |
2032 | * still be dying after the previous unmount. Let's drain the | |
2033 | * dying subsystems. We just need to ensure that the ones | |
2034 | * unmounted previously finish dying and don't care about new ones | |
2035 | * starting. Testing ref liveliness is good enough. | |
2036 | */ | |
2037 | for_each_subsys(ss, i) { | |
2038 | if (!(opts.subsys_mask & (1 << i)) || | |
2039 | ss->root == &cgrp_dfl_root) | |
2040 | continue; | |
2041 | ||
2042 | if (!percpu_ref_tryget_live(&ss->root->cgrp.self.refcnt)) { | |
2043 | mutex_unlock(&cgroup_mutex); | |
2044 | msleep(10); | |
2045 | ret = restart_syscall(); | |
2046 | goto out_free; | |
2047 | } | |
2048 | cgroup_put(&ss->root->cgrp); | |
2049 | } | |
2050 | ||
985ed670 | 2051 | for_each_root(root) { |
2bd59d48 | 2052 | bool name_match = false; |
3126121f | 2053 | |
3dd06ffa | 2054 | if (root == &cgrp_dfl_root) |
985ed670 | 2055 | continue; |
3126121f | 2056 | |
cf5d5941 | 2057 | /* |
2bd59d48 TH |
2058 | * If we asked for a name then it must match. Also, if |
2059 | * name matches but sybsys_mask doesn't, we should fail. | |
2060 | * Remember whether name matched. | |
cf5d5941 | 2061 | */ |
2bd59d48 TH |
2062 | if (opts.name) { |
2063 | if (strcmp(opts.name, root->name)) | |
2064 | continue; | |
2065 | name_match = true; | |
2066 | } | |
ddbcc7e8 | 2067 | |
c6d57f33 | 2068 | /* |
2bd59d48 TH |
2069 | * If we asked for subsystems (or explicitly for no |
2070 | * subsystems) then they must match. | |
c6d57f33 | 2071 | */ |
2bd59d48 | 2072 | if ((opts.subsys_mask || opts.none) && |
f392e51c | 2073 | (opts.subsys_mask != root->subsys_mask)) { |
2bd59d48 TH |
2074 | if (!name_match) |
2075 | continue; | |
2076 | ret = -EBUSY; | |
2077 | goto out_unlock; | |
2078 | } | |
873fe09e | 2079 | |
7b9a6ba5 TH |
2080 | if (root->flags ^ opts.flags) |
2081 | pr_warn("new mount options do not match the existing superblock, will be ignored\n"); | |
ddbcc7e8 | 2082 | |
776f02fa | 2083 | /* |
3a32bd72 LZ |
2084 | * We want to reuse @root whose lifetime is governed by its |
2085 | * ->cgrp. Let's check whether @root is alive and keep it | |
2086 | * that way. As cgroup_kill_sb() can happen anytime, we | |
2087 | * want to block it by pinning the sb so that @root doesn't | |
2088 | * get killed before mount is complete. | |
2089 | * | |
2090 | * With the sb pinned, tryget_live can reliably indicate | |
2091 | * whether @root can be reused. If it's being killed, | |
2092 | * drain it. We can use wait_queue for the wait but this | |
2093 | * path is super cold. Let's just sleep a bit and retry. | |
776f02fa | 2094 | */ |
3a32bd72 LZ |
2095 | pinned_sb = kernfs_pin_sb(root->kf_root, NULL); |
2096 | if (IS_ERR(pinned_sb) || | |
2097 | !percpu_ref_tryget_live(&root->cgrp.self.refcnt)) { | |
776f02fa | 2098 | mutex_unlock(&cgroup_mutex); |
3a32bd72 LZ |
2099 | if (!IS_ERR_OR_NULL(pinned_sb)) |
2100 | deactivate_super(pinned_sb); | |
776f02fa | 2101 | msleep(10); |
a015edd2 TH |
2102 | ret = restart_syscall(); |
2103 | goto out_free; | |
776f02fa | 2104 | } |
ddbcc7e8 | 2105 | |
776f02fa | 2106 | ret = 0; |
2bd59d48 | 2107 | goto out_unlock; |
ddbcc7e8 | 2108 | } |
ddbcc7e8 | 2109 | |
817929ec | 2110 | /* |
172a2c06 TH |
2111 | * No such thing, create a new one. name= matching without subsys |
2112 | * specification is allowed for already existing hierarchies but we | |
2113 | * can't create new one without subsys specification. | |
817929ec | 2114 | */ |
172a2c06 TH |
2115 | if (!opts.subsys_mask && !opts.none) { |
2116 | ret = -EINVAL; | |
2117 | goto out_unlock; | |
817929ec | 2118 | } |
817929ec | 2119 | |
172a2c06 TH |
2120 | root = kzalloc(sizeof(*root), GFP_KERNEL); |
2121 | if (!root) { | |
2122 | ret = -ENOMEM; | |
2bd59d48 | 2123 | goto out_unlock; |
839ec545 | 2124 | } |
e5f6a860 | 2125 | |
172a2c06 TH |
2126 | init_cgroup_root(root, &opts); |
2127 | ||
35585573 | 2128 | ret = cgroup_setup_root(root, opts.subsys_mask); |
2bd59d48 TH |
2129 | if (ret) |
2130 | cgroup_free_root(root); | |
fa3ca07e | 2131 | |
8e30e2b8 | 2132 | out_unlock: |
ddbcc7e8 | 2133 | mutex_unlock(&cgroup_mutex); |
a015edd2 | 2134 | out_free: |
c6d57f33 PM |
2135 | kfree(opts.release_agent); |
2136 | kfree(opts.name); | |
03b1cde6 | 2137 | |
2bd59d48 | 2138 | if (ret) |
8e30e2b8 | 2139 | return ERR_PTR(ret); |
67e9c74b | 2140 | out_mount: |
c9482a5b | 2141 | dentry = kernfs_mount(fs_type, flags, root->kf_root, |
67e9c74b TH |
2142 | is_v2 ? CGROUP2_SUPER_MAGIC : CGROUP_SUPER_MAGIC, |
2143 | &new_sb); | |
c6b3d5bc | 2144 | if (IS_ERR(dentry) || !new_sb) |
3dd06ffa | 2145 | cgroup_put(&root->cgrp); |
3a32bd72 LZ |
2146 | |
2147 | /* | |
2148 | * If @pinned_sb, we're reusing an existing root and holding an | |
2149 | * extra ref on its sb. Mount is complete. Put the extra ref. | |
2150 | */ | |
2151 | if (pinned_sb) { | |
2152 | WARN_ON(new_sb); | |
2153 | deactivate_super(pinned_sb); | |
2154 | } | |
2155 | ||
2bd59d48 TH |
2156 | return dentry; |
2157 | } | |
2158 | ||
2159 | static void cgroup_kill_sb(struct super_block *sb) | |
2160 | { | |
2161 | struct kernfs_root *kf_root = kernfs_root_from_sb(sb); | |
3dd06ffa | 2162 | struct cgroup_root *root = cgroup_root_from_kf(kf_root); |
2bd59d48 | 2163 | |
9d755d33 TH |
2164 | /* |
2165 | * If @root doesn't have any mounts or children, start killing it. | |
2166 | * This prevents new mounts by disabling percpu_ref_tryget_live(). | |
2167 | * cgroup_mount() may wait for @root's release. | |
1f779fb2 LZ |
2168 | * |
2169 | * And don't kill the default root. | |
9d755d33 | 2170 | */ |
3c606d35 | 2171 | if (!list_empty(&root->cgrp.self.children) || |
1f779fb2 | 2172 | root == &cgrp_dfl_root) |
9d755d33 TH |
2173 | cgroup_put(&root->cgrp); |
2174 | else | |
2175 | percpu_ref_kill(&root->cgrp.self.refcnt); | |
2176 | ||
2bd59d48 | 2177 | kernfs_kill_sb(sb); |
ddbcc7e8 PM |
2178 | } |
2179 | ||
2180 | static struct file_system_type cgroup_fs_type = { | |
2181 | .name = "cgroup", | |
f7e83571 | 2182 | .mount = cgroup_mount, |
ddbcc7e8 PM |
2183 | .kill_sb = cgroup_kill_sb, |
2184 | }; | |
2185 | ||
67e9c74b TH |
2186 | static struct file_system_type cgroup2_fs_type = { |
2187 | .name = "cgroup2", | |
2188 | .mount = cgroup_mount, | |
2189 | .kill_sb = cgroup_kill_sb, | |
2190 | }; | |
2191 | ||
a79a908f AK |
2192 | static char *cgroup_path_ns_locked(struct cgroup *cgrp, char *buf, size_t buflen, |
2193 | struct cgroup_namespace *ns) | |
2194 | { | |
2195 | struct cgroup *root = cset_cgroup_from_root(ns->root_cset, cgrp->root); | |
2196 | int ret; | |
2197 | ||
2198 | ret = kernfs_path_from_node(cgrp->kn, root->kn, buf, buflen); | |
2199 | if (ret < 0 || ret >= buflen) | |
2200 | return NULL; | |
2201 | return buf; | |
2202 | } | |
2203 | ||
2204 | char *cgroup_path_ns(struct cgroup *cgrp, char *buf, size_t buflen, | |
2205 | struct cgroup_namespace *ns) | |
2206 | { | |
2207 | char *ret; | |
2208 | ||
2209 | mutex_lock(&cgroup_mutex); | |
2210 | spin_lock_bh(&css_set_lock); | |
2211 | ||
2212 | ret = cgroup_path_ns_locked(cgrp, buf, buflen, ns); | |
2213 | ||
2214 | spin_unlock_bh(&css_set_lock); | |
2215 | mutex_unlock(&cgroup_mutex); | |
2216 | ||
2217 | return ret; | |
2218 | } | |
2219 | EXPORT_SYMBOL_GPL(cgroup_path_ns); | |
2220 | ||
857a2beb | 2221 | /** |
913ffdb5 | 2222 | * task_cgroup_path - cgroup path of a task in the first cgroup hierarchy |
857a2beb | 2223 | * @task: target task |
857a2beb TH |
2224 | * @buf: the buffer to write the path into |
2225 | * @buflen: the length of the buffer | |
2226 | * | |
913ffdb5 TH |
2227 | * Determine @task's cgroup on the first (the one with the lowest non-zero |
2228 | * hierarchy_id) cgroup hierarchy and copy its path into @buf. This | |
2229 | * function grabs cgroup_mutex and shouldn't be used inside locks used by | |
2230 | * cgroup controller callbacks. | |
2231 | * | |
e61734c5 | 2232 | * Return value is the same as kernfs_path(). |
857a2beb | 2233 | */ |
e61734c5 | 2234 | char *task_cgroup_path(struct task_struct *task, char *buf, size_t buflen) |
857a2beb | 2235 | { |
3dd06ffa | 2236 | struct cgroup_root *root; |
913ffdb5 | 2237 | struct cgroup *cgrp; |
e61734c5 TH |
2238 | int hierarchy_id = 1; |
2239 | char *path = NULL; | |
857a2beb TH |
2240 | |
2241 | mutex_lock(&cgroup_mutex); | |
f0d9a5f1 | 2242 | spin_lock_bh(&css_set_lock); |
857a2beb | 2243 | |
913ffdb5 TH |
2244 | root = idr_get_next(&cgroup_hierarchy_idr, &hierarchy_id); |
2245 | ||
857a2beb TH |
2246 | if (root) { |
2247 | cgrp = task_cgroup_from_root(task, root); | |
a79a908f | 2248 | path = cgroup_path_ns_locked(cgrp, buf, buflen, &init_cgroup_ns); |
913ffdb5 TH |
2249 | } else { |
2250 | /* if no hierarchy exists, everyone is in "/" */ | |
e61734c5 TH |
2251 | if (strlcpy(buf, "/", buflen) < buflen) |
2252 | path = buf; | |
857a2beb TH |
2253 | } |
2254 | ||
f0d9a5f1 | 2255 | spin_unlock_bh(&css_set_lock); |
857a2beb | 2256 | mutex_unlock(&cgroup_mutex); |
e61734c5 | 2257 | return path; |
857a2beb | 2258 | } |
913ffdb5 | 2259 | EXPORT_SYMBOL_GPL(task_cgroup_path); |
857a2beb | 2260 | |
b3dc094e | 2261 | /* used to track tasks and other necessary states during migration */ |
2f7ee569 | 2262 | struct cgroup_taskset { |
b3dc094e TH |
2263 | /* the src and dst cset list running through cset->mg_node */ |
2264 | struct list_head src_csets; | |
2265 | struct list_head dst_csets; | |
2266 | ||
1f7dd3e5 TH |
2267 | /* the subsys currently being processed */ |
2268 | int ssid; | |
2269 | ||
b3dc094e TH |
2270 | /* |
2271 | * Fields for cgroup_taskset_*() iteration. | |
2272 | * | |
2273 | * Before migration is committed, the target migration tasks are on | |
2274 | * ->mg_tasks of the csets on ->src_csets. After, on ->mg_tasks of | |
2275 | * the csets on ->dst_csets. ->csets point to either ->src_csets | |
2276 | * or ->dst_csets depending on whether migration is committed. | |
2277 | * | |
2278 | * ->cur_csets and ->cur_task point to the current task position | |
2279 | * during iteration. | |
2280 | */ | |
2281 | struct list_head *csets; | |
2282 | struct css_set *cur_cset; | |
2283 | struct task_struct *cur_task; | |
2f7ee569 TH |
2284 | }; |
2285 | ||
adaae5dc TH |
2286 | #define CGROUP_TASKSET_INIT(tset) (struct cgroup_taskset){ \ |
2287 | .src_csets = LIST_HEAD_INIT(tset.src_csets), \ | |
2288 | .dst_csets = LIST_HEAD_INIT(tset.dst_csets), \ | |
2289 | .csets = &tset.src_csets, \ | |
2290 | } | |
2291 | ||
2292 | /** | |
2293 | * cgroup_taskset_add - try to add a migration target task to a taskset | |
2294 | * @task: target task | |
2295 | * @tset: target taskset | |
2296 | * | |
2297 | * Add @task, which is a migration target, to @tset. This function becomes | |
2298 | * noop if @task doesn't need to be migrated. @task's css_set should have | |
2299 | * been added as a migration source and @task->cg_list will be moved from | |
2300 | * the css_set's tasks list to mg_tasks one. | |
2301 | */ | |
2302 | static void cgroup_taskset_add(struct task_struct *task, | |
2303 | struct cgroup_taskset *tset) | |
2304 | { | |
2305 | struct css_set *cset; | |
2306 | ||
f0d9a5f1 | 2307 | lockdep_assert_held(&css_set_lock); |
adaae5dc TH |
2308 | |
2309 | /* @task either already exited or can't exit until the end */ | |
2310 | if (task->flags & PF_EXITING) | |
2311 | return; | |
2312 | ||
2313 | /* leave @task alone if post_fork() hasn't linked it yet */ | |
2314 | if (list_empty(&task->cg_list)) | |
2315 | return; | |
2316 | ||
2317 | cset = task_css_set(task); | |
2318 | if (!cset->mg_src_cgrp) | |
2319 | return; | |
2320 | ||
2321 | list_move_tail(&task->cg_list, &cset->mg_tasks); | |
2322 | if (list_empty(&cset->mg_node)) | |
2323 | list_add_tail(&cset->mg_node, &tset->src_csets); | |
2324 | if (list_empty(&cset->mg_dst_cset->mg_node)) | |
2325 | list_move_tail(&cset->mg_dst_cset->mg_node, | |
2326 | &tset->dst_csets); | |
2327 | } | |
2328 | ||
2f7ee569 TH |
2329 | /** |
2330 | * cgroup_taskset_first - reset taskset and return the first task | |
2331 | * @tset: taskset of interest | |
1f7dd3e5 | 2332 | * @dst_cssp: output variable for the destination css |
2f7ee569 TH |
2333 | * |
2334 | * @tset iteration is initialized and the first task is returned. | |
2335 | */ | |
1f7dd3e5 TH |
2336 | struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset, |
2337 | struct cgroup_subsys_state **dst_cssp) | |
2f7ee569 | 2338 | { |
b3dc094e TH |
2339 | tset->cur_cset = list_first_entry(tset->csets, struct css_set, mg_node); |
2340 | tset->cur_task = NULL; | |
2341 | ||
1f7dd3e5 | 2342 | return cgroup_taskset_next(tset, dst_cssp); |
2f7ee569 | 2343 | } |
2f7ee569 TH |
2344 | |
2345 | /** | |
2346 | * cgroup_taskset_next - iterate to the next task in taskset | |
2347 | * @tset: taskset of interest | |
1f7dd3e5 | 2348 | * @dst_cssp: output variable for the destination css |
2f7ee569 TH |
2349 | * |
2350 | * Return the next task in @tset. Iteration must have been initialized | |
2351 | * with cgroup_taskset_first(). | |
2352 | */ | |
1f7dd3e5 TH |
2353 | struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset, |
2354 | struct cgroup_subsys_state **dst_cssp) | |
2f7ee569 | 2355 | { |
b3dc094e TH |
2356 | struct css_set *cset = tset->cur_cset; |
2357 | struct task_struct *task = tset->cur_task; | |
2f7ee569 | 2358 | |
b3dc094e TH |
2359 | while (&cset->mg_node != tset->csets) { |
2360 | if (!task) | |
2361 | task = list_first_entry(&cset->mg_tasks, | |
2362 | struct task_struct, cg_list); | |
2363 | else | |
2364 | task = list_next_entry(task, cg_list); | |
2f7ee569 | 2365 | |
b3dc094e TH |
2366 | if (&task->cg_list != &cset->mg_tasks) { |
2367 | tset->cur_cset = cset; | |
2368 | tset->cur_task = task; | |
1f7dd3e5 TH |
2369 | |
2370 | /* | |
2371 | * This function may be called both before and | |
2372 | * after cgroup_taskset_migrate(). The two cases | |
2373 | * can be distinguished by looking at whether @cset | |
2374 | * has its ->mg_dst_cset set. | |
2375 | */ | |
2376 | if (cset->mg_dst_cset) | |
2377 | *dst_cssp = cset->mg_dst_cset->subsys[tset->ssid]; | |
2378 | else | |
2379 | *dst_cssp = cset->subsys[tset->ssid]; | |
2380 | ||
b3dc094e TH |
2381 | return task; |
2382 | } | |
2f7ee569 | 2383 | |
b3dc094e TH |
2384 | cset = list_next_entry(cset, mg_node); |
2385 | task = NULL; | |
2386 | } | |
2f7ee569 | 2387 | |
b3dc094e | 2388 | return NULL; |
2f7ee569 | 2389 | } |
2f7ee569 | 2390 | |
adaae5dc TH |
2391 | /** |
2392 | * cgroup_taskset_migrate - migrate a taskset to a cgroup | |
2393 | * @tset: taget taskset | |
2394 | * @dst_cgrp: destination cgroup | |
2395 | * | |
2396 | * Migrate tasks in @tset to @dst_cgrp. This function fails iff one of the | |
2397 | * ->can_attach callbacks fails and guarantees that either all or none of | |
2398 | * the tasks in @tset are migrated. @tset is consumed regardless of | |
2399 | * success. | |
2400 | */ | |
2401 | static int cgroup_taskset_migrate(struct cgroup_taskset *tset, | |
2402 | struct cgroup *dst_cgrp) | |
2403 | { | |
2404 | struct cgroup_subsys_state *css, *failed_css = NULL; | |
2405 | struct task_struct *task, *tmp_task; | |
2406 | struct css_set *cset, *tmp_cset; | |
2407 | int i, ret; | |
2408 | ||
2409 | /* methods shouldn't be called if no task is actually migrating */ | |
2410 | if (list_empty(&tset->src_csets)) | |
2411 | return 0; | |
2412 | ||
2413 | /* check that we can legitimately attach to the cgroup */ | |
2414 | for_each_e_css(css, i, dst_cgrp) { | |
2415 | if (css->ss->can_attach) { | |
1f7dd3e5 TH |
2416 | tset->ssid = i; |
2417 | ret = css->ss->can_attach(tset); | |
adaae5dc TH |
2418 | if (ret) { |
2419 | failed_css = css; | |
2420 | goto out_cancel_attach; | |
2421 | } | |
2422 | } | |
2423 | } | |
2424 | ||
2425 | /* | |
2426 | * Now that we're guaranteed success, proceed to move all tasks to | |
2427 | * the new cgroup. There are no failure cases after here, so this | |
2428 | * is the commit point. | |
2429 | */ | |
f0d9a5f1 | 2430 | spin_lock_bh(&css_set_lock); |
adaae5dc | 2431 | list_for_each_entry(cset, &tset->src_csets, mg_node) { |
f6d7d049 TH |
2432 | list_for_each_entry_safe(task, tmp_task, &cset->mg_tasks, cg_list) { |
2433 | struct css_set *from_cset = task_css_set(task); | |
2434 | struct css_set *to_cset = cset->mg_dst_cset; | |
2435 | ||
2436 | get_css_set(to_cset); | |
2437 | css_set_move_task(task, from_cset, to_cset, true); | |
2438 | put_css_set_locked(from_cset); | |
2439 | } | |
adaae5dc | 2440 | } |
f0d9a5f1 | 2441 | spin_unlock_bh(&css_set_lock); |
adaae5dc TH |
2442 | |
2443 | /* | |
2444 | * Migration is committed, all target tasks are now on dst_csets. | |
2445 | * Nothing is sensitive to fork() after this point. Notify | |
2446 | * controllers that migration is complete. | |
2447 | */ | |
2448 | tset->csets = &tset->dst_csets; | |
2449 | ||
1f7dd3e5 TH |
2450 | for_each_e_css(css, i, dst_cgrp) { |
2451 | if (css->ss->attach) { | |
2452 | tset->ssid = i; | |
2453 | css->ss->attach(tset); | |
2454 | } | |
2455 | } | |
adaae5dc TH |
2456 | |
2457 | ret = 0; | |
2458 | goto out_release_tset; | |
2459 | ||
2460 | out_cancel_attach: | |
2461 | for_each_e_css(css, i, dst_cgrp) { | |
2462 | if (css == failed_css) | |
2463 | break; | |
1f7dd3e5 TH |
2464 | if (css->ss->cancel_attach) { |
2465 | tset->ssid = i; | |
2466 | css->ss->cancel_attach(tset); | |
2467 | } | |
adaae5dc TH |
2468 | } |
2469 | out_release_tset: | |
f0d9a5f1 | 2470 | spin_lock_bh(&css_set_lock); |
adaae5dc TH |
2471 | list_splice_init(&tset->dst_csets, &tset->src_csets); |
2472 | list_for_each_entry_safe(cset, tmp_cset, &tset->src_csets, mg_node) { | |
2473 | list_splice_tail_init(&cset->mg_tasks, &cset->tasks); | |
2474 | list_del_init(&cset->mg_node); | |
2475 | } | |
f0d9a5f1 | 2476 | spin_unlock_bh(&css_set_lock); |
adaae5dc TH |
2477 | return ret; |
2478 | } | |
2479 | ||
a043e3b2 | 2480 | /** |
1958d2d5 TH |
2481 | * cgroup_migrate_finish - cleanup after attach |
2482 | * @preloaded_csets: list of preloaded css_sets | |
74a1166d | 2483 | * |
1958d2d5 TH |
2484 | * Undo cgroup_migrate_add_src() and cgroup_migrate_prepare_dst(). See |
2485 | * those functions for details. | |
74a1166d | 2486 | */ |
1958d2d5 | 2487 | static void cgroup_migrate_finish(struct list_head *preloaded_csets) |
74a1166d | 2488 | { |
1958d2d5 | 2489 | struct css_set *cset, *tmp_cset; |
74a1166d | 2490 | |
1958d2d5 TH |
2491 | lockdep_assert_held(&cgroup_mutex); |
2492 | ||
f0d9a5f1 | 2493 | spin_lock_bh(&css_set_lock); |
1958d2d5 TH |
2494 | list_for_each_entry_safe(cset, tmp_cset, preloaded_csets, mg_preload_node) { |
2495 | cset->mg_src_cgrp = NULL; | |
2496 | cset->mg_dst_cset = NULL; | |
2497 | list_del_init(&cset->mg_preload_node); | |
a25eb52e | 2498 | put_css_set_locked(cset); |
1958d2d5 | 2499 | } |
f0d9a5f1 | 2500 | spin_unlock_bh(&css_set_lock); |
1958d2d5 TH |
2501 | } |
2502 | ||
2503 | /** | |
2504 | * cgroup_migrate_add_src - add a migration source css_set | |
2505 | * @src_cset: the source css_set to add | |
2506 | * @dst_cgrp: the destination cgroup | |
2507 | * @preloaded_csets: list of preloaded css_sets | |
2508 | * | |
2509 | * Tasks belonging to @src_cset are about to be migrated to @dst_cgrp. Pin | |
2510 | * @src_cset and add it to @preloaded_csets, which should later be cleaned | |
2511 | * up by cgroup_migrate_finish(). | |
2512 | * | |
1ed13287 TH |
2513 | * This function may be called without holding cgroup_threadgroup_rwsem |
2514 | * even if the target is a process. Threads may be created and destroyed | |
2515 | * but as long as cgroup_mutex is not dropped, no new css_set can be put | |
2516 | * into play and the preloaded css_sets are guaranteed to cover all | |
2517 | * migrations. | |
1958d2d5 TH |
2518 | */ |
2519 | static void cgroup_migrate_add_src(struct css_set *src_cset, | |
2520 | struct cgroup *dst_cgrp, | |
2521 | struct list_head *preloaded_csets) | |
2522 | { | |
2523 | struct cgroup *src_cgrp; | |
2524 | ||
2525 | lockdep_assert_held(&cgroup_mutex); | |
f0d9a5f1 | 2526 | lockdep_assert_held(&css_set_lock); |
1958d2d5 TH |
2527 | |
2528 | src_cgrp = cset_cgroup_from_root(src_cset, dst_cgrp->root); | |
2529 | ||
1958d2d5 TH |
2530 | if (!list_empty(&src_cset->mg_preload_node)) |
2531 | return; | |
2532 | ||
2533 | WARN_ON(src_cset->mg_src_cgrp); | |
2534 | WARN_ON(!list_empty(&src_cset->mg_tasks)); | |
2535 | WARN_ON(!list_empty(&src_cset->mg_node)); | |
2536 | ||
2537 | src_cset->mg_src_cgrp = src_cgrp; | |
2538 | get_css_set(src_cset); | |
2539 | list_add(&src_cset->mg_preload_node, preloaded_csets); | |
2540 | } | |
2541 | ||
2542 | /** | |
2543 | * cgroup_migrate_prepare_dst - prepare destination css_sets for migration | |
f817de98 | 2544 | * @dst_cgrp: the destination cgroup (may be %NULL) |
1958d2d5 TH |
2545 | * @preloaded_csets: list of preloaded source css_sets |
2546 | * | |
2547 | * Tasks are about to be moved to @dst_cgrp and all the source css_sets | |
2548 | * have been preloaded to @preloaded_csets. This function looks up and | |
f817de98 TH |
2549 | * pins all destination css_sets, links each to its source, and append them |
2550 | * to @preloaded_csets. If @dst_cgrp is %NULL, the destination of each | |
2551 | * source css_set is assumed to be its cgroup on the default hierarchy. | |
1958d2d5 TH |
2552 | * |
2553 | * This function must be called after cgroup_migrate_add_src() has been | |
2554 | * called on each migration source css_set. After migration is performed | |
2555 | * using cgroup_migrate(), cgroup_migrate_finish() must be called on | |
2556 | * @preloaded_csets. | |
2557 | */ | |
2558 | static int cgroup_migrate_prepare_dst(struct cgroup *dst_cgrp, | |
2559 | struct list_head *preloaded_csets) | |
2560 | { | |
2561 | LIST_HEAD(csets); | |
f817de98 | 2562 | struct css_set *src_cset, *tmp_cset; |
1958d2d5 TH |
2563 | |
2564 | lockdep_assert_held(&cgroup_mutex); | |
2565 | ||
f8f22e53 TH |
2566 | /* |
2567 | * Except for the root, child_subsys_mask must be zero for a cgroup | |
2568 | * with tasks so that child cgroups don't compete against tasks. | |
2569 | */ | |
d51f39b0 | 2570 | if (dst_cgrp && cgroup_on_dfl(dst_cgrp) && cgroup_parent(dst_cgrp) && |
f8f22e53 TH |
2571 | dst_cgrp->child_subsys_mask) |
2572 | return -EBUSY; | |
2573 | ||
1958d2d5 | 2574 | /* look up the dst cset for each src cset and link it to src */ |
f817de98 | 2575 | list_for_each_entry_safe(src_cset, tmp_cset, preloaded_csets, mg_preload_node) { |
1958d2d5 TH |
2576 | struct css_set *dst_cset; |
2577 | ||
f817de98 TH |
2578 | dst_cset = find_css_set(src_cset, |
2579 | dst_cgrp ?: src_cset->dfl_cgrp); | |
1958d2d5 TH |
2580 | if (!dst_cset) |
2581 | goto err; | |
2582 | ||
2583 | WARN_ON_ONCE(src_cset->mg_dst_cset || dst_cset->mg_dst_cset); | |
f817de98 TH |
2584 | |
2585 | /* | |
2586 | * If src cset equals dst, it's noop. Drop the src. | |
2587 | * cgroup_migrate() will skip the cset too. Note that we | |
2588 | * can't handle src == dst as some nodes are used by both. | |
2589 | */ | |
2590 | if (src_cset == dst_cset) { | |
2591 | src_cset->mg_src_cgrp = NULL; | |
2592 | list_del_init(&src_cset->mg_preload_node); | |
a25eb52e ZL |
2593 | put_css_set(src_cset); |
2594 | put_css_set(dst_cset); | |
f817de98 TH |
2595 | continue; |
2596 | } | |
2597 | ||
1958d2d5 TH |
2598 | src_cset->mg_dst_cset = dst_cset; |
2599 | ||
2600 | if (list_empty(&dst_cset->mg_preload_node)) | |
2601 | list_add(&dst_cset->mg_preload_node, &csets); | |
2602 | else | |
a25eb52e | 2603 | put_css_set(dst_cset); |
1958d2d5 TH |
2604 | } |
2605 | ||
f817de98 | 2606 | list_splice_tail(&csets, preloaded_csets); |
1958d2d5 TH |
2607 | return 0; |
2608 | err: | |
2609 | cgroup_migrate_finish(&csets); | |
2610 | return -ENOMEM; | |
2611 | } | |
2612 | ||
2613 | /** | |
2614 | * cgroup_migrate - migrate a process or task to a cgroup | |
1958d2d5 TH |
2615 | * @leader: the leader of the process or the task to migrate |
2616 | * @threadgroup: whether @leader points to the whole process or a single task | |
9af2ec45 | 2617 | * @cgrp: the destination cgroup |
1958d2d5 TH |
2618 | * |
2619 | * Migrate a process or task denoted by @leader to @cgrp. If migrating a | |
1ed13287 | 2620 | * process, the caller must be holding cgroup_threadgroup_rwsem. The |
1958d2d5 TH |
2621 | * caller is also responsible for invoking cgroup_migrate_add_src() and |
2622 | * cgroup_migrate_prepare_dst() on the targets before invoking this | |
2623 | * function and following up with cgroup_migrate_finish(). | |
2624 | * | |
2625 | * As long as a controller's ->can_attach() doesn't fail, this function is | |
2626 | * guaranteed to succeed. This means that, excluding ->can_attach() | |
2627 | * failure, when migrating multiple targets, the success or failure can be | |
2628 | * decided for all targets by invoking group_migrate_prepare_dst() before | |
2629 | * actually starting migrating. | |
2630 | */ | |
9af2ec45 TH |
2631 | static int cgroup_migrate(struct task_struct *leader, bool threadgroup, |
2632 | struct cgroup *cgrp) | |
74a1166d | 2633 | { |
adaae5dc TH |
2634 | struct cgroup_taskset tset = CGROUP_TASKSET_INIT(tset); |
2635 | struct task_struct *task; | |
74a1166d | 2636 | |
fb5d2b4c MSB |
2637 | /* |
2638 | * Prevent freeing of tasks while we take a snapshot. Tasks that are | |
2639 | * already PF_EXITING could be freed from underneath us unless we | |
2640 | * take an rcu_read_lock. | |
2641 | */ | |
f0d9a5f1 | 2642 | spin_lock_bh(&css_set_lock); |
fb5d2b4c | 2643 | rcu_read_lock(); |
9db8de37 | 2644 | task = leader; |
74a1166d | 2645 | do { |
adaae5dc | 2646 | cgroup_taskset_add(task, &tset); |
081aa458 LZ |
2647 | if (!threadgroup) |
2648 | break; | |
9db8de37 | 2649 | } while_each_thread(leader, task); |
fb5d2b4c | 2650 | rcu_read_unlock(); |
f0d9a5f1 | 2651 | spin_unlock_bh(&css_set_lock); |
74a1166d | 2652 | |
adaae5dc | 2653 | return cgroup_taskset_migrate(&tset, cgrp); |
74a1166d BB |
2654 | } |
2655 | ||
1958d2d5 TH |
2656 | /** |
2657 | * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup | |
2658 | * @dst_cgrp: the cgroup to attach to | |
2659 | * @leader: the task or the leader of the threadgroup to be attached | |
2660 | * @threadgroup: attach the whole threadgroup? | |
2661 | * | |
1ed13287 | 2662 | * Call holding cgroup_mutex and cgroup_threadgroup_rwsem. |
1958d2d5 TH |
2663 | */ |
2664 | static int cgroup_attach_task(struct cgroup *dst_cgrp, | |
2665 | struct task_struct *leader, bool threadgroup) | |
2666 | { | |
2667 | LIST_HEAD(preloaded_csets); | |
2668 | struct task_struct *task; | |
2669 | int ret; | |
2670 | ||
2671 | /* look up all src csets */ | |
f0d9a5f1 | 2672 | spin_lock_bh(&css_set_lock); |
1958d2d5 TH |
2673 | rcu_read_lock(); |
2674 | task = leader; | |
2675 | do { | |
2676 | cgroup_migrate_add_src(task_css_set(task), dst_cgrp, | |
2677 | &preloaded_csets); | |
2678 | if (!threadgroup) | |
2679 | break; | |
2680 | } while_each_thread(leader, task); | |
2681 | rcu_read_unlock(); | |
f0d9a5f1 | 2682 | spin_unlock_bh(&css_set_lock); |
1958d2d5 TH |
2683 | |
2684 | /* prepare dst csets and commit */ | |
2685 | ret = cgroup_migrate_prepare_dst(dst_cgrp, &preloaded_csets); | |
2686 | if (!ret) | |
9af2ec45 | 2687 | ret = cgroup_migrate(leader, threadgroup, dst_cgrp); |
1958d2d5 TH |
2688 | |
2689 | cgroup_migrate_finish(&preloaded_csets); | |
2690 | return ret; | |
74a1166d BB |
2691 | } |
2692 | ||
187fe840 TH |
2693 | static int cgroup_procs_write_permission(struct task_struct *task, |
2694 | struct cgroup *dst_cgrp, | |
2695 | struct kernfs_open_file *of) | |
dedf22e9 TH |
2696 | { |
2697 | const struct cred *cred = current_cred(); | |
2698 | const struct cred *tcred = get_task_cred(task); | |
2699 | int ret = 0; | |
2700 | ||
2701 | /* | |
2702 | * even if we're attaching all tasks in the thread group, we only | |
2703 | * need to check permissions on one of them. | |
2704 | */ | |
2705 | if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) && | |
2706 | !uid_eq(cred->euid, tcred->uid) && | |
2707 | !uid_eq(cred->euid, tcred->suid)) | |
2708 | ret = -EACCES; | |
2709 | ||
187fe840 TH |
2710 | if (!ret && cgroup_on_dfl(dst_cgrp)) { |
2711 | struct super_block *sb = of->file->f_path.dentry->d_sb; | |
2712 | struct cgroup *cgrp; | |
2713 | struct inode *inode; | |
2714 | ||
f0d9a5f1 | 2715 | spin_lock_bh(&css_set_lock); |
187fe840 | 2716 | cgrp = task_cgroup_from_root(task, &cgrp_dfl_root); |
f0d9a5f1 | 2717 | spin_unlock_bh(&css_set_lock); |
187fe840 TH |
2718 | |
2719 | while (!cgroup_is_descendant(dst_cgrp, cgrp)) | |
2720 | cgrp = cgroup_parent(cgrp); | |
2721 | ||
2722 | ret = -ENOMEM; | |
6f60eade | 2723 | inode = kernfs_get_inode(sb, cgrp->procs_file.kn); |
187fe840 TH |
2724 | if (inode) { |
2725 | ret = inode_permission(inode, MAY_WRITE); | |
2726 | iput(inode); | |
2727 | } | |
2728 | } | |
2729 | ||
dedf22e9 TH |
2730 | put_cred(tcred); |
2731 | return ret; | |
2732 | } | |
2733 | ||
74a1166d BB |
2734 | /* |
2735 | * Find the task_struct of the task to attach by vpid and pass it along to the | |
cd3d0952 | 2736 | * function to attach either it or all tasks in its threadgroup. Will lock |
0e1d768f | 2737 | * cgroup_mutex and threadgroup. |
bbcb81d0 | 2738 | */ |
acbef755 TH |
2739 | static ssize_t __cgroup_procs_write(struct kernfs_open_file *of, char *buf, |
2740 | size_t nbytes, loff_t off, bool threadgroup) | |
bbcb81d0 | 2741 | { |
bbcb81d0 | 2742 | struct task_struct *tsk; |
e76ecaee | 2743 | struct cgroup *cgrp; |
acbef755 | 2744 | pid_t pid; |
bbcb81d0 PM |
2745 | int ret; |
2746 | ||
acbef755 TH |
2747 | if (kstrtoint(strstrip(buf), 0, &pid) || pid < 0) |
2748 | return -EINVAL; | |
2749 | ||
e76ecaee TH |
2750 | cgrp = cgroup_kn_lock_live(of->kn); |
2751 | if (!cgrp) | |
74a1166d BB |
2752 | return -ENODEV; |
2753 | ||
3014dde7 | 2754 | percpu_down_write(&cgroup_threadgroup_rwsem); |
b78949eb | 2755 | rcu_read_lock(); |
bbcb81d0 | 2756 | if (pid) { |
73507f33 | 2757 | tsk = find_task_by_vpid(pid); |
74a1166d | 2758 | if (!tsk) { |
dd4b0a46 | 2759 | ret = -ESRCH; |
3014dde7 | 2760 | goto out_unlock_rcu; |
bbcb81d0 | 2761 | } |
dedf22e9 | 2762 | } else { |
b78949eb | 2763 | tsk = current; |
dedf22e9 | 2764 | } |
cd3d0952 TH |
2765 | |
2766 | if (threadgroup) | |
b78949eb | 2767 | tsk = tsk->group_leader; |
c4c27fbd MG |
2768 | |
2769 | /* | |
14a40ffc | 2770 | * Workqueue threads may acquire PF_NO_SETAFFINITY and become |
c4c27fbd MG |
2771 | * trapped in a cpuset, or RT worker may be born in a cgroup |
2772 | * with no rt_runtime allocated. Just say no. | |
2773 | */ | |
14a40ffc | 2774 | if (tsk == kthreadd_task || (tsk->flags & PF_NO_SETAFFINITY)) { |
c4c27fbd | 2775 | ret = -EINVAL; |
3014dde7 | 2776 | goto out_unlock_rcu; |
c4c27fbd MG |
2777 | } |
2778 | ||
b78949eb MSB |
2779 | get_task_struct(tsk); |
2780 | rcu_read_unlock(); | |
2781 | ||
187fe840 | 2782 | ret = cgroup_procs_write_permission(tsk, cgrp, of); |
dedf22e9 TH |
2783 | if (!ret) |
2784 | ret = cgroup_attach_task(cgrp, tsk, threadgroup); | |
081aa458 | 2785 | |
f9f9e7b7 | 2786 | put_task_struct(tsk); |
3014dde7 TH |
2787 | goto out_unlock_threadgroup; |
2788 | ||
2789 | out_unlock_rcu: | |
2790 | rcu_read_unlock(); | |
2791 | out_unlock_threadgroup: | |
2792 | percpu_up_write(&cgroup_threadgroup_rwsem); | |
e76ecaee | 2793 | cgroup_kn_unlock(of->kn); |
e93ad19d | 2794 | cpuset_post_attach_flush(); |
acbef755 | 2795 | return ret ?: nbytes; |
bbcb81d0 PM |
2796 | } |
2797 | ||
7ae1bad9 TH |
2798 | /** |
2799 | * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from' | |
2800 | * @from: attach to all cgroups of a given task | |
2801 | * @tsk: the task to be attached | |
2802 | */ | |
2803 | int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk) | |
2804 | { | |
3dd06ffa | 2805 | struct cgroup_root *root; |
7ae1bad9 TH |
2806 | int retval = 0; |
2807 | ||
47cfcd09 | 2808 | mutex_lock(&cgroup_mutex); |
985ed670 | 2809 | for_each_root(root) { |
96d365e0 TH |
2810 | struct cgroup *from_cgrp; |
2811 | ||
3dd06ffa | 2812 | if (root == &cgrp_dfl_root) |
985ed670 TH |
2813 | continue; |
2814 | ||
f0d9a5f1 | 2815 | spin_lock_bh(&css_set_lock); |
96d365e0 | 2816 | from_cgrp = task_cgroup_from_root(from, root); |
f0d9a5f1 | 2817 | spin_unlock_bh(&css_set_lock); |
7ae1bad9 | 2818 | |
6f4b7e63 | 2819 | retval = cgroup_attach_task(from_cgrp, tsk, false); |
7ae1bad9 TH |
2820 | if (retval) |
2821 | break; | |
2822 | } | |
47cfcd09 | 2823 | mutex_unlock(&cgroup_mutex); |
7ae1bad9 TH |
2824 | |
2825 | return retval; | |
2826 | } | |
2827 | EXPORT_SYMBOL_GPL(cgroup_attach_task_all); | |
2828 | ||
acbef755 TH |
2829 | static ssize_t cgroup_tasks_write(struct kernfs_open_file *of, |
2830 | char *buf, size_t nbytes, loff_t off) | |
74a1166d | 2831 | { |
acbef755 | 2832 | return __cgroup_procs_write(of, buf, nbytes, off, false); |
74a1166d BB |
2833 | } |
2834 | ||
acbef755 TH |
2835 | static ssize_t cgroup_procs_write(struct kernfs_open_file *of, |
2836 | char *buf, size_t nbytes, loff_t off) | |
af351026 | 2837 | { |
acbef755 | 2838 | return __cgroup_procs_write(of, buf, nbytes, off, true); |
af351026 PM |
2839 | } |
2840 | ||
451af504 TH |
2841 | static ssize_t cgroup_release_agent_write(struct kernfs_open_file *of, |
2842 | char *buf, size_t nbytes, loff_t off) | |
e788e066 | 2843 | { |
e76ecaee | 2844 | struct cgroup *cgrp; |
5f469907 | 2845 | |
e76ecaee | 2846 | BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX); |
5f469907 | 2847 | |
e76ecaee TH |
2848 | cgrp = cgroup_kn_lock_live(of->kn); |
2849 | if (!cgrp) | |
e788e066 | 2850 | return -ENODEV; |
69e943b7 | 2851 | spin_lock(&release_agent_path_lock); |
e76ecaee TH |
2852 | strlcpy(cgrp->root->release_agent_path, strstrip(buf), |
2853 | sizeof(cgrp->root->release_agent_path)); | |
69e943b7 | 2854 | spin_unlock(&release_agent_path_lock); |
e76ecaee | 2855 | cgroup_kn_unlock(of->kn); |
451af504 | 2856 | return nbytes; |
e788e066 PM |
2857 | } |
2858 | ||
2da8ca82 | 2859 | static int cgroup_release_agent_show(struct seq_file *seq, void *v) |
e788e066 | 2860 | { |
2da8ca82 | 2861 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
182446d0 | 2862 | |
46cfeb04 | 2863 | spin_lock(&release_agent_path_lock); |
e788e066 | 2864 | seq_puts(seq, cgrp->root->release_agent_path); |
46cfeb04 | 2865 | spin_unlock(&release_agent_path_lock); |
e788e066 | 2866 | seq_putc(seq, '\n'); |
e788e066 PM |
2867 | return 0; |
2868 | } | |
2869 | ||
2da8ca82 | 2870 | static int cgroup_sane_behavior_show(struct seq_file *seq, void *v) |
873fe09e | 2871 | { |
c1d5d42e | 2872 | seq_puts(seq, "0\n"); |
e788e066 PM |
2873 | return 0; |
2874 | } | |
2875 | ||
8ab456ac | 2876 | static void cgroup_print_ss_mask(struct seq_file *seq, unsigned long ss_mask) |
355e0c48 | 2877 | { |
f8f22e53 TH |
2878 | struct cgroup_subsys *ss; |
2879 | bool printed = false; | |
2880 | int ssid; | |
a742c59d | 2881 | |
a966a4ed AS |
2882 | for_each_subsys_which(ss, ssid, &ss_mask) { |
2883 | if (printed) | |
2884 | seq_putc(seq, ' '); | |
2885 | seq_printf(seq, "%s", ss->name); | |
2886 | printed = true; | |
e73d2c61 | 2887 | } |
f8f22e53 TH |
2888 | if (printed) |
2889 | seq_putc(seq, '\n'); | |
355e0c48 PM |
2890 | } |
2891 | ||
f8f22e53 TH |
2892 | /* show controllers which are currently attached to the default hierarchy */ |
2893 | static int cgroup_root_controllers_show(struct seq_file *seq, void *v) | |
db3b1497 | 2894 | { |
f8f22e53 TH |
2895 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
2896 | ||
5533e011 TH |
2897 | cgroup_print_ss_mask(seq, cgrp->root->subsys_mask & |
2898 | ~cgrp_dfl_root_inhibit_ss_mask); | |
f8f22e53 | 2899 | return 0; |
db3b1497 PM |
2900 | } |
2901 | ||
f8f22e53 TH |
2902 | /* show controllers which are enabled from the parent */ |
2903 | static int cgroup_controllers_show(struct seq_file *seq, void *v) | |
ddbcc7e8 | 2904 | { |
f8f22e53 TH |
2905 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
2906 | ||
667c2491 | 2907 | cgroup_print_ss_mask(seq, cgroup_parent(cgrp)->subtree_control); |
f8f22e53 | 2908 | return 0; |
ddbcc7e8 PM |
2909 | } |
2910 | ||
f8f22e53 TH |
2911 | /* show controllers which are enabled for a given cgroup's children */ |
2912 | static int cgroup_subtree_control_show(struct seq_file *seq, void *v) | |
ddbcc7e8 | 2913 | { |
f8f22e53 TH |
2914 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
2915 | ||
667c2491 | 2916 | cgroup_print_ss_mask(seq, cgrp->subtree_control); |
f8f22e53 TH |
2917 | return 0; |
2918 | } | |
2919 | ||
2920 | /** | |
2921 | * cgroup_update_dfl_csses - update css assoc of a subtree in default hierarchy | |
2922 | * @cgrp: root of the subtree to update csses for | |
2923 | * | |
2924 | * @cgrp's child_subsys_mask has changed and its subtree's (self excluded) | |
2925 | * css associations need to be updated accordingly. This function looks up | |
2926 | * all css_sets which are attached to the subtree, creates the matching | |
2927 | * updated css_sets and migrates the tasks to the new ones. | |
2928 | */ | |
2929 | static int cgroup_update_dfl_csses(struct cgroup *cgrp) | |
2930 | { | |
2931 | LIST_HEAD(preloaded_csets); | |
10265075 | 2932 | struct cgroup_taskset tset = CGROUP_TASKSET_INIT(tset); |
f8f22e53 TH |
2933 | struct cgroup_subsys_state *css; |
2934 | struct css_set *src_cset; | |
2935 | int ret; | |
2936 | ||
f8f22e53 TH |
2937 | lockdep_assert_held(&cgroup_mutex); |
2938 | ||
3014dde7 TH |
2939 | percpu_down_write(&cgroup_threadgroup_rwsem); |
2940 | ||
f8f22e53 | 2941 | /* look up all csses currently attached to @cgrp's subtree */ |
f0d9a5f1 | 2942 | spin_lock_bh(&css_set_lock); |
f8f22e53 TH |
2943 | css_for_each_descendant_pre(css, cgroup_css(cgrp, NULL)) { |
2944 | struct cgrp_cset_link *link; | |
2945 | ||
2946 | /* self is not affected by child_subsys_mask change */ | |
2947 | if (css->cgroup == cgrp) | |
2948 | continue; | |
2949 | ||
2950 | list_for_each_entry(link, &css->cgroup->cset_links, cset_link) | |
2951 | cgroup_migrate_add_src(link->cset, cgrp, | |
2952 | &preloaded_csets); | |
2953 | } | |
f0d9a5f1 | 2954 | spin_unlock_bh(&css_set_lock); |
f8f22e53 TH |
2955 | |
2956 | /* NULL dst indicates self on default hierarchy */ | |
2957 | ret = cgroup_migrate_prepare_dst(NULL, &preloaded_csets); | |
2958 | if (ret) | |
2959 | goto out_finish; | |
2960 | ||
f0d9a5f1 | 2961 | spin_lock_bh(&css_set_lock); |
f8f22e53 | 2962 | list_for_each_entry(src_cset, &preloaded_csets, mg_preload_node) { |
10265075 | 2963 | struct task_struct *task, *ntask; |
f8f22e53 TH |
2964 | |
2965 | /* src_csets precede dst_csets, break on the first dst_cset */ | |
2966 | if (!src_cset->mg_src_cgrp) | |
2967 | break; | |
2968 | ||
10265075 TH |
2969 | /* all tasks in src_csets need to be migrated */ |
2970 | list_for_each_entry_safe(task, ntask, &src_cset->tasks, cg_list) | |
2971 | cgroup_taskset_add(task, &tset); | |
f8f22e53 | 2972 | } |
f0d9a5f1 | 2973 | spin_unlock_bh(&css_set_lock); |
f8f22e53 | 2974 | |
10265075 | 2975 | ret = cgroup_taskset_migrate(&tset, cgrp); |
f8f22e53 TH |
2976 | out_finish: |
2977 | cgroup_migrate_finish(&preloaded_csets); | |
3014dde7 | 2978 | percpu_up_write(&cgroup_threadgroup_rwsem); |
f8f22e53 TH |
2979 | return ret; |
2980 | } | |
2981 | ||
2982 | /* change the enabled child controllers for a cgroup in the default hierarchy */ | |
451af504 TH |
2983 | static ssize_t cgroup_subtree_control_write(struct kernfs_open_file *of, |
2984 | char *buf, size_t nbytes, | |
2985 | loff_t off) | |
f8f22e53 | 2986 | { |
8ab456ac AS |
2987 | unsigned long enable = 0, disable = 0; |
2988 | unsigned long css_enable, css_disable, old_sc, new_sc, old_ss, new_ss; | |
a9746d8d | 2989 | struct cgroup *cgrp, *child; |
f8f22e53 | 2990 | struct cgroup_subsys *ss; |
451af504 | 2991 | char *tok; |
f8f22e53 TH |
2992 | int ssid, ret; |
2993 | ||
2994 | /* | |
d37167ab TH |
2995 | * Parse input - space separated list of subsystem names prefixed |
2996 | * with either + or -. | |
f8f22e53 | 2997 | */ |
451af504 TH |
2998 | buf = strstrip(buf); |
2999 | while ((tok = strsep(&buf, " "))) { | |
a966a4ed AS |
3000 | unsigned long tmp_ss_mask = ~cgrp_dfl_root_inhibit_ss_mask; |
3001 | ||
d37167ab TH |
3002 | if (tok[0] == '\0') |
3003 | continue; | |
a966a4ed | 3004 | for_each_subsys_which(ss, ssid, &tmp_ss_mask) { |
fc5ed1e9 TH |
3005 | if (!cgroup_ssid_enabled(ssid) || |
3006 | strcmp(tok + 1, ss->name)) | |
f8f22e53 TH |
3007 | continue; |
3008 | ||
3009 | if (*tok == '+') { | |
7d331fa9 TH |
3010 | enable |= 1 << ssid; |
3011 | disable &= ~(1 << ssid); | |
f8f22e53 | 3012 | } else if (*tok == '-') { |
7d331fa9 TH |
3013 | disable |= 1 << ssid; |
3014 | enable &= ~(1 << ssid); | |
f8f22e53 TH |
3015 | } else { |
3016 | return -EINVAL; | |
3017 | } | |
3018 | break; | |
3019 | } | |
3020 | if (ssid == CGROUP_SUBSYS_COUNT) | |
3021 | return -EINVAL; | |
3022 | } | |
3023 | ||
a9746d8d TH |
3024 | cgrp = cgroup_kn_lock_live(of->kn); |
3025 | if (!cgrp) | |
3026 | return -ENODEV; | |
f8f22e53 TH |
3027 | |
3028 | for_each_subsys(ss, ssid) { | |
3029 | if (enable & (1 << ssid)) { | |
667c2491 | 3030 | if (cgrp->subtree_control & (1 << ssid)) { |
f8f22e53 TH |
3031 | enable &= ~(1 << ssid); |
3032 | continue; | |
3033 | } | |
3034 | ||
c29adf24 TH |
3035 | /* unavailable or not enabled on the parent? */ |
3036 | if (!(cgrp_dfl_root.subsys_mask & (1 << ssid)) || | |
3037 | (cgroup_parent(cgrp) && | |
667c2491 | 3038 | !(cgroup_parent(cgrp)->subtree_control & (1 << ssid)))) { |
c29adf24 TH |
3039 | ret = -ENOENT; |
3040 | goto out_unlock; | |
3041 | } | |
f8f22e53 | 3042 | } else if (disable & (1 << ssid)) { |
667c2491 | 3043 | if (!(cgrp->subtree_control & (1 << ssid))) { |
f8f22e53 TH |
3044 | disable &= ~(1 << ssid); |
3045 | continue; | |
3046 | } | |
3047 | ||
3048 | /* a child has it enabled? */ | |
3049 | cgroup_for_each_live_child(child, cgrp) { | |
667c2491 | 3050 | if (child->subtree_control & (1 << ssid)) { |
f8f22e53 | 3051 | ret = -EBUSY; |
ddab2b6e | 3052 | goto out_unlock; |
f8f22e53 TH |
3053 | } |
3054 | } | |
3055 | } | |
3056 | } | |
3057 | ||
3058 | if (!enable && !disable) { | |
3059 | ret = 0; | |
ddab2b6e | 3060 | goto out_unlock; |
f8f22e53 TH |
3061 | } |
3062 | ||
3063 | /* | |
667c2491 | 3064 | * Except for the root, subtree_control must be zero for a cgroup |
f8f22e53 TH |
3065 | * with tasks so that child cgroups don't compete against tasks. |
3066 | */ | |
d51f39b0 | 3067 | if (enable && cgroup_parent(cgrp) && !list_empty(&cgrp->cset_links)) { |
f8f22e53 TH |
3068 | ret = -EBUSY; |
3069 | goto out_unlock; | |
3070 | } | |
3071 | ||
3072 | /* | |
f63070d3 TH |
3073 | * Update subsys masks and calculate what needs to be done. More |
3074 | * subsystems than specified may need to be enabled or disabled | |
3075 | * depending on subsystem dependencies. | |
3076 | */ | |
755bf5ee TH |
3077 | old_sc = cgrp->subtree_control; |
3078 | old_ss = cgrp->child_subsys_mask; | |
3079 | new_sc = (old_sc | enable) & ~disable; | |
3080 | new_ss = cgroup_calc_child_subsys_mask(cgrp, new_sc); | |
f63070d3 | 3081 | |
755bf5ee TH |
3082 | css_enable = ~old_ss & new_ss; |
3083 | css_disable = old_ss & ~new_ss; | |
f63070d3 TH |
3084 | enable |= css_enable; |
3085 | disable |= css_disable; | |
c29adf24 | 3086 | |
db6e3053 TH |
3087 | /* |
3088 | * Because css offlining is asynchronous, userland might try to | |
3089 | * re-enable the same controller while the previous instance is | |
3090 | * still around. In such cases, wait till it's gone using | |
3091 | * offline_waitq. | |
3092 | */ | |
a966a4ed | 3093 | for_each_subsys_which(ss, ssid, &css_enable) { |
db6e3053 TH |
3094 | cgroup_for_each_live_child(child, cgrp) { |
3095 | DEFINE_WAIT(wait); | |
3096 | ||
3097 | if (!cgroup_css(child, ss)) | |
3098 | continue; | |
3099 | ||
3100 | cgroup_get(child); | |
3101 | prepare_to_wait(&child->offline_waitq, &wait, | |
3102 | TASK_UNINTERRUPTIBLE); | |
3103 | cgroup_kn_unlock(of->kn); | |
3104 | schedule(); | |
3105 | finish_wait(&child->offline_waitq, &wait); | |
3106 | cgroup_put(child); | |
3107 | ||
3108 | return restart_syscall(); | |
3109 | } | |
3110 | } | |
3111 | ||
755bf5ee TH |
3112 | cgrp->subtree_control = new_sc; |
3113 | cgrp->child_subsys_mask = new_ss; | |
3114 | ||
f63070d3 TH |
3115 | /* |
3116 | * Create new csses or make the existing ones visible. A css is | |
3117 | * created invisible if it's being implicitly enabled through | |
3118 | * dependency. An invisible css is made visible when the userland | |
3119 | * explicitly enables it. | |
f8f22e53 TH |
3120 | */ |
3121 | for_each_subsys(ss, ssid) { | |
3122 | if (!(enable & (1 << ssid))) | |
3123 | continue; | |
3124 | ||
3125 | cgroup_for_each_live_child(child, cgrp) { | |
f63070d3 TH |
3126 | if (css_enable & (1 << ssid)) |
3127 | ret = create_css(child, ss, | |
3128 | cgrp->subtree_control & (1 << ssid)); | |
3129 | else | |
4df8dc90 TH |
3130 | ret = css_populate_dir(cgroup_css(child, ss), |
3131 | NULL); | |
f8f22e53 TH |
3132 | if (ret) |
3133 | goto err_undo_css; | |
3134 | } | |
3135 | } | |
3136 | ||
c29adf24 TH |
3137 | /* |
3138 | * At this point, cgroup_e_css() results reflect the new csses | |
3139 | * making the following cgroup_update_dfl_csses() properly update | |
3140 | * css associations of all tasks in the subtree. | |
3141 | */ | |
f8f22e53 TH |
3142 | ret = cgroup_update_dfl_csses(cgrp); |
3143 | if (ret) | |
3144 | goto err_undo_css; | |
3145 | ||
f63070d3 TH |
3146 | /* |
3147 | * All tasks are migrated out of disabled csses. Kill or hide | |
3148 | * them. A css is hidden when the userland requests it to be | |
b4536f0c TH |
3149 | * disabled while other subsystems are still depending on it. The |
3150 | * css must not actively control resources and be in the vanilla | |
3151 | * state if it's made visible again later. Controllers which may | |
3152 | * be depended upon should provide ->css_reset() for this purpose. | |
f63070d3 | 3153 | */ |
f8f22e53 TH |
3154 | for_each_subsys(ss, ssid) { |
3155 | if (!(disable & (1 << ssid))) | |
3156 | continue; | |
3157 | ||
f63070d3 | 3158 | cgroup_for_each_live_child(child, cgrp) { |
b4536f0c TH |
3159 | struct cgroup_subsys_state *css = cgroup_css(child, ss); |
3160 | ||
3161 | if (css_disable & (1 << ssid)) { | |
3162 | kill_css(css); | |
3163 | } else { | |
4df8dc90 | 3164 | css_clear_dir(css, NULL); |
b4536f0c TH |
3165 | if (ss->css_reset) |
3166 | ss->css_reset(css); | |
3167 | } | |
f63070d3 | 3168 | } |
f8f22e53 TH |
3169 | } |
3170 | ||
56c807ba TH |
3171 | /* |
3172 | * The effective csses of all the descendants (excluding @cgrp) may | |
3173 | * have changed. Subsystems can optionally subscribe to this event | |
3174 | * by implementing ->css_e_css_changed() which is invoked if any of | |
3175 | * the effective csses seen from the css's cgroup may have changed. | |
3176 | */ | |
3177 | for_each_subsys(ss, ssid) { | |
3178 | struct cgroup_subsys_state *this_css = cgroup_css(cgrp, ss); | |
3179 | struct cgroup_subsys_state *css; | |
3180 | ||
3181 | if (!ss->css_e_css_changed || !this_css) | |
3182 | continue; | |
3183 | ||
3184 | css_for_each_descendant_pre(css, this_css) | |
3185 | if (css != this_css) | |
3186 | ss->css_e_css_changed(css); | |
3187 | } | |
3188 | ||
f8f22e53 TH |
3189 | kernfs_activate(cgrp->kn); |
3190 | ret = 0; | |
3191 | out_unlock: | |
a9746d8d | 3192 | cgroup_kn_unlock(of->kn); |
451af504 | 3193 | return ret ?: nbytes; |
f8f22e53 TH |
3194 | |
3195 | err_undo_css: | |
755bf5ee TH |
3196 | cgrp->subtree_control = old_sc; |
3197 | cgrp->child_subsys_mask = old_ss; | |
f8f22e53 TH |
3198 | |
3199 | for_each_subsys(ss, ssid) { | |
3200 | if (!(enable & (1 << ssid))) | |
3201 | continue; | |
3202 | ||
3203 | cgroup_for_each_live_child(child, cgrp) { | |
3204 | struct cgroup_subsys_state *css = cgroup_css(child, ss); | |
f63070d3 TH |
3205 | |
3206 | if (!css) | |
3207 | continue; | |
3208 | ||
3209 | if (css_enable & (1 << ssid)) | |
f8f22e53 | 3210 | kill_css(css); |
f63070d3 | 3211 | else |
4df8dc90 | 3212 | css_clear_dir(css, NULL); |
f8f22e53 TH |
3213 | } |
3214 | } | |
3215 | goto out_unlock; | |
3216 | } | |
3217 | ||
4a07c222 | 3218 | static int cgroup_events_show(struct seq_file *seq, void *v) |
842b597e | 3219 | { |
4a07c222 | 3220 | seq_printf(seq, "populated %d\n", |
27bd4dbb | 3221 | cgroup_is_populated(seq_css(seq)->cgroup)); |
842b597e TH |
3222 | return 0; |
3223 | } | |
3224 | ||
2bd59d48 TH |
3225 | static ssize_t cgroup_file_write(struct kernfs_open_file *of, char *buf, |
3226 | size_t nbytes, loff_t off) | |
355e0c48 | 3227 | { |
2bd59d48 TH |
3228 | struct cgroup *cgrp = of->kn->parent->priv; |
3229 | struct cftype *cft = of->kn->priv; | |
3230 | struct cgroup_subsys_state *css; | |
a742c59d | 3231 | int ret; |
355e0c48 | 3232 | |
b4168640 TH |
3233 | if (cft->write) |
3234 | return cft->write(of, buf, nbytes, off); | |
3235 | ||
2bd59d48 TH |
3236 | /* |
3237 | * kernfs guarantees that a file isn't deleted with operations in | |
3238 | * flight, which means that the matching css is and stays alive and | |
3239 | * doesn't need to be pinned. The RCU locking is not necessary | |
3240 | * either. It's just for the convenience of using cgroup_css(). | |
3241 | */ | |
3242 | rcu_read_lock(); | |
3243 | css = cgroup_css(cgrp, cft->ss); | |
3244 | rcu_read_unlock(); | |
a742c59d | 3245 | |
451af504 | 3246 | if (cft->write_u64) { |
a742c59d TH |
3247 | unsigned long long v; |
3248 | ret = kstrtoull(buf, 0, &v); | |
3249 | if (!ret) | |
3250 | ret = cft->write_u64(css, cft, v); | |
3251 | } else if (cft->write_s64) { | |
3252 | long long v; | |
3253 | ret = kstrtoll(buf, 0, &v); | |
3254 | if (!ret) | |
3255 | ret = cft->write_s64(css, cft, v); | |
e73d2c61 | 3256 | } else { |
a742c59d | 3257 | ret = -EINVAL; |
e73d2c61 | 3258 | } |
2bd59d48 | 3259 | |
a742c59d | 3260 | return ret ?: nbytes; |
355e0c48 PM |
3261 | } |
3262 | ||
6612f05b | 3263 | static void *cgroup_seqfile_start(struct seq_file *seq, loff_t *ppos) |
db3b1497 | 3264 | { |
2bd59d48 | 3265 | return seq_cft(seq)->seq_start(seq, ppos); |
db3b1497 PM |
3266 | } |
3267 | ||
6612f05b | 3268 | static void *cgroup_seqfile_next(struct seq_file *seq, void *v, loff_t *ppos) |
ddbcc7e8 | 3269 | { |
2bd59d48 | 3270 | return seq_cft(seq)->seq_next(seq, v, ppos); |
ddbcc7e8 PM |
3271 | } |
3272 | ||
6612f05b | 3273 | static void cgroup_seqfile_stop(struct seq_file *seq, void *v) |
ddbcc7e8 | 3274 | { |
2bd59d48 | 3275 | seq_cft(seq)->seq_stop(seq, v); |
ddbcc7e8 PM |
3276 | } |
3277 | ||
91796569 | 3278 | static int cgroup_seqfile_show(struct seq_file *m, void *arg) |
e73d2c61 | 3279 | { |
7da11279 TH |
3280 | struct cftype *cft = seq_cft(m); |
3281 | struct cgroup_subsys_state *css = seq_css(m); | |
e73d2c61 | 3282 | |
2da8ca82 TH |
3283 | if (cft->seq_show) |
3284 | return cft->seq_show(m, arg); | |
e73d2c61 | 3285 | |
f4c753b7 | 3286 | if (cft->read_u64) |
896f5199 TH |
3287 | seq_printf(m, "%llu\n", cft->read_u64(css, cft)); |
3288 | else if (cft->read_s64) | |
3289 | seq_printf(m, "%lld\n", cft->read_s64(css, cft)); | |
3290 | else | |
3291 | return -EINVAL; | |
3292 | return 0; | |
91796569 PM |
3293 | } |
3294 | ||
2bd59d48 TH |
3295 | static struct kernfs_ops cgroup_kf_single_ops = { |
3296 | .atomic_write_len = PAGE_SIZE, | |
3297 | .write = cgroup_file_write, | |
3298 | .seq_show = cgroup_seqfile_show, | |
91796569 PM |
3299 | }; |
3300 | ||
2bd59d48 TH |
3301 | static struct kernfs_ops cgroup_kf_ops = { |
3302 | .atomic_write_len = PAGE_SIZE, | |
3303 | .write = cgroup_file_write, | |
3304 | .seq_start = cgroup_seqfile_start, | |
3305 | .seq_next = cgroup_seqfile_next, | |
3306 | .seq_stop = cgroup_seqfile_stop, | |
3307 | .seq_show = cgroup_seqfile_show, | |
3308 | }; | |
ddbcc7e8 PM |
3309 | |
3310 | /* | |
3311 | * cgroup_rename - Only allow simple rename of directories in place. | |
3312 | */ | |
2bd59d48 TH |
3313 | static int cgroup_rename(struct kernfs_node *kn, struct kernfs_node *new_parent, |
3314 | const char *new_name_str) | |
ddbcc7e8 | 3315 | { |
2bd59d48 | 3316 | struct cgroup *cgrp = kn->priv; |
65dff759 | 3317 | int ret; |
65dff759 | 3318 | |
2bd59d48 | 3319 | if (kernfs_type(kn) != KERNFS_DIR) |
ddbcc7e8 | 3320 | return -ENOTDIR; |
2bd59d48 | 3321 | if (kn->parent != new_parent) |
ddbcc7e8 | 3322 | return -EIO; |
65dff759 | 3323 | |
6db8e85c TH |
3324 | /* |
3325 | * This isn't a proper migration and its usefulness is very | |
aa6ec29b | 3326 | * limited. Disallow on the default hierarchy. |
6db8e85c | 3327 | */ |
aa6ec29b | 3328 | if (cgroup_on_dfl(cgrp)) |
6db8e85c | 3329 | return -EPERM; |
099fca32 | 3330 | |
e1b2dc17 | 3331 | /* |
8353da1f | 3332 | * We're gonna grab cgroup_mutex which nests outside kernfs |
e1b2dc17 | 3333 | * active_ref. kernfs_rename() doesn't require active_ref |
8353da1f | 3334 | * protection. Break them before grabbing cgroup_mutex. |
e1b2dc17 TH |
3335 | */ |
3336 | kernfs_break_active_protection(new_parent); | |
3337 | kernfs_break_active_protection(kn); | |
099fca32 | 3338 | |
2bd59d48 | 3339 | mutex_lock(&cgroup_mutex); |
099fca32 | 3340 | |
2bd59d48 | 3341 | ret = kernfs_rename(kn, new_parent, new_name_str); |
099fca32 | 3342 | |
2bd59d48 | 3343 | mutex_unlock(&cgroup_mutex); |
e1b2dc17 TH |
3344 | |
3345 | kernfs_unbreak_active_protection(kn); | |
3346 | kernfs_unbreak_active_protection(new_parent); | |
2bd59d48 | 3347 | return ret; |
099fca32 LZ |
3348 | } |
3349 | ||
49957f8e TH |
3350 | /* set uid and gid of cgroup dirs and files to that of the creator */ |
3351 | static int cgroup_kn_set_ugid(struct kernfs_node *kn) | |
3352 | { | |
3353 | struct iattr iattr = { .ia_valid = ATTR_UID | ATTR_GID, | |
3354 | .ia_uid = current_fsuid(), | |
3355 | .ia_gid = current_fsgid(), }; | |
3356 | ||
3357 | if (uid_eq(iattr.ia_uid, GLOBAL_ROOT_UID) && | |
3358 | gid_eq(iattr.ia_gid, GLOBAL_ROOT_GID)) | |
3359 | return 0; | |
3360 | ||
3361 | return kernfs_setattr(kn, &iattr); | |
3362 | } | |
3363 | ||
4df8dc90 TH |
3364 | static int cgroup_add_file(struct cgroup_subsys_state *css, struct cgroup *cgrp, |
3365 | struct cftype *cft) | |
ddbcc7e8 | 3366 | { |
8d7e6fb0 | 3367 | char name[CGROUP_FILE_NAME_MAX]; |
2bd59d48 TH |
3368 | struct kernfs_node *kn; |
3369 | struct lock_class_key *key = NULL; | |
49957f8e | 3370 | int ret; |
05ef1d7c | 3371 | |
2bd59d48 TH |
3372 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
3373 | key = &cft->lockdep_key; | |
3374 | #endif | |
3375 | kn = __kernfs_create_file(cgrp->kn, cgroup_file_name(cgrp, cft, name), | |
3376 | cgroup_file_mode(cft), 0, cft->kf_ops, cft, | |
dfeb0750 | 3377 | NULL, key); |
49957f8e TH |
3378 | if (IS_ERR(kn)) |
3379 | return PTR_ERR(kn); | |
3380 | ||
3381 | ret = cgroup_kn_set_ugid(kn); | |
f8f22e53 | 3382 | if (ret) { |
49957f8e | 3383 | kernfs_remove(kn); |
f8f22e53 TH |
3384 | return ret; |
3385 | } | |
3386 | ||
6f60eade TH |
3387 | if (cft->file_offset) { |
3388 | struct cgroup_file *cfile = (void *)css + cft->file_offset; | |
3389 | ||
34c06254 | 3390 | spin_lock_irq(&cgroup_file_kn_lock); |
6f60eade | 3391 | cfile->kn = kn; |
34c06254 | 3392 | spin_unlock_irq(&cgroup_file_kn_lock); |
6f60eade TH |
3393 | } |
3394 | ||
f8f22e53 | 3395 | return 0; |
ddbcc7e8 PM |
3396 | } |
3397 | ||
b1f28d31 TH |
3398 | /** |
3399 | * cgroup_addrm_files - add or remove files to a cgroup directory | |
4df8dc90 TH |
3400 | * @css: the target css |
3401 | * @cgrp: the target cgroup (usually css->cgroup) | |
b1f28d31 TH |
3402 | * @cfts: array of cftypes to be added |
3403 | * @is_add: whether to add or remove | |
3404 | * | |
3405 | * Depending on @is_add, add or remove files defined by @cfts on @cgrp. | |
6732ed85 | 3406 | * For removals, this function never fails. |
b1f28d31 | 3407 | */ |
4df8dc90 TH |
3408 | static int cgroup_addrm_files(struct cgroup_subsys_state *css, |
3409 | struct cgroup *cgrp, struct cftype cfts[], | |
2bb566cb | 3410 | bool is_add) |
ddbcc7e8 | 3411 | { |
6732ed85 | 3412 | struct cftype *cft, *cft_end = NULL; |
b1f28d31 TH |
3413 | int ret; |
3414 | ||
01f6474c | 3415 | lockdep_assert_held(&cgroup_mutex); |
db0416b6 | 3416 | |
6732ed85 TH |
3417 | restart: |
3418 | for (cft = cfts; cft != cft_end && cft->name[0] != '\0'; cft++) { | |
f33fddc2 | 3419 | /* does cft->flags tell us to skip this file on @cgrp? */ |
05ebb6e6 | 3420 | if ((cft->flags & __CFTYPE_ONLY_ON_DFL) && !cgroup_on_dfl(cgrp)) |
8cbbf2c9 | 3421 | continue; |
05ebb6e6 | 3422 | if ((cft->flags & __CFTYPE_NOT_ON_DFL) && cgroup_on_dfl(cgrp)) |
873fe09e | 3423 | continue; |
d51f39b0 | 3424 | if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgroup_parent(cgrp)) |
f33fddc2 | 3425 | continue; |
d51f39b0 | 3426 | if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgroup_parent(cgrp)) |
f33fddc2 G |
3427 | continue; |
3428 | ||
2739d3cc | 3429 | if (is_add) { |
4df8dc90 | 3430 | ret = cgroup_add_file(css, cgrp, cft); |
b1f28d31 | 3431 | if (ret) { |
ed3d261b JP |
3432 | pr_warn("%s: failed to add %s, err=%d\n", |
3433 | __func__, cft->name, ret); | |
6732ed85 TH |
3434 | cft_end = cft; |
3435 | is_add = false; | |
3436 | goto restart; | |
b1f28d31 | 3437 | } |
2739d3cc LZ |
3438 | } else { |
3439 | cgroup_rm_file(cgrp, cft); | |
db0416b6 | 3440 | } |
ddbcc7e8 | 3441 | } |
b1f28d31 | 3442 | return 0; |
ddbcc7e8 PM |
3443 | } |
3444 | ||
21a2d343 | 3445 | static int cgroup_apply_cftypes(struct cftype *cfts, bool is_add) |
8e3f6541 TH |
3446 | { |
3447 | LIST_HEAD(pending); | |
2bb566cb | 3448 | struct cgroup_subsys *ss = cfts[0].ss; |
3dd06ffa | 3449 | struct cgroup *root = &ss->root->cgrp; |
492eb21b | 3450 | struct cgroup_subsys_state *css; |
9ccece80 | 3451 | int ret = 0; |
8e3f6541 | 3452 | |
01f6474c | 3453 | lockdep_assert_held(&cgroup_mutex); |
e8c82d20 | 3454 | |
e8c82d20 | 3455 | /* add/rm files for all cgroups created before */ |
ca8bdcaf | 3456 | css_for_each_descendant_pre(css, cgroup_css(root, ss)) { |
492eb21b TH |
3457 | struct cgroup *cgrp = css->cgroup; |
3458 | ||
e8c82d20 LZ |
3459 | if (cgroup_is_dead(cgrp)) |
3460 | continue; | |
3461 | ||
4df8dc90 | 3462 | ret = cgroup_addrm_files(css, cgrp, cfts, is_add); |
9ccece80 TH |
3463 | if (ret) |
3464 | break; | |
8e3f6541 | 3465 | } |
21a2d343 TH |
3466 | |
3467 | if (is_add && !ret) | |
3468 | kernfs_activate(root->kn); | |
9ccece80 | 3469 | return ret; |
8e3f6541 TH |
3470 | } |
3471 | ||
2da440a2 | 3472 | static void cgroup_exit_cftypes(struct cftype *cfts) |
8e3f6541 | 3473 | { |
2bb566cb | 3474 | struct cftype *cft; |
8e3f6541 | 3475 | |
2bd59d48 TH |
3476 | for (cft = cfts; cft->name[0] != '\0'; cft++) { |
3477 | /* free copy for custom atomic_write_len, see init_cftypes() */ | |
3478 | if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) | |
3479 | kfree(cft->kf_ops); | |
3480 | cft->kf_ops = NULL; | |
2da440a2 | 3481 | cft->ss = NULL; |
a8ddc821 TH |
3482 | |
3483 | /* revert flags set by cgroup core while adding @cfts */ | |
05ebb6e6 | 3484 | cft->flags &= ~(__CFTYPE_ONLY_ON_DFL | __CFTYPE_NOT_ON_DFL); |
2bd59d48 | 3485 | } |
2da440a2 TH |
3486 | } |
3487 | ||
2bd59d48 | 3488 | static int cgroup_init_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
2da440a2 TH |
3489 | { |
3490 | struct cftype *cft; | |
3491 | ||
2bd59d48 TH |
3492 | for (cft = cfts; cft->name[0] != '\0'; cft++) { |
3493 | struct kernfs_ops *kf_ops; | |
3494 | ||
0adb0704 TH |
3495 | WARN_ON(cft->ss || cft->kf_ops); |
3496 | ||
2bd59d48 TH |
3497 | if (cft->seq_start) |
3498 | kf_ops = &cgroup_kf_ops; | |
3499 | else | |
3500 | kf_ops = &cgroup_kf_single_ops; | |
3501 | ||
3502 | /* | |
3503 | * Ugh... if @cft wants a custom max_write_len, we need to | |
3504 | * make a copy of kf_ops to set its atomic_write_len. | |
3505 | */ | |
3506 | if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) { | |
3507 | kf_ops = kmemdup(kf_ops, sizeof(*kf_ops), GFP_KERNEL); | |
3508 | if (!kf_ops) { | |
3509 | cgroup_exit_cftypes(cfts); | |
3510 | return -ENOMEM; | |
3511 | } | |
3512 | kf_ops->atomic_write_len = cft->max_write_len; | |
3513 | } | |
8e3f6541 | 3514 | |
2bd59d48 | 3515 | cft->kf_ops = kf_ops; |
2bb566cb | 3516 | cft->ss = ss; |
2bd59d48 | 3517 | } |
2bb566cb | 3518 | |
2bd59d48 | 3519 | return 0; |
2da440a2 TH |
3520 | } |
3521 | ||
21a2d343 TH |
3522 | static int cgroup_rm_cftypes_locked(struct cftype *cfts) |
3523 | { | |
01f6474c | 3524 | lockdep_assert_held(&cgroup_mutex); |
21a2d343 TH |
3525 | |
3526 | if (!cfts || !cfts[0].ss) | |
3527 | return -ENOENT; | |
3528 | ||
3529 | list_del(&cfts->node); | |
3530 | cgroup_apply_cftypes(cfts, false); | |
3531 | cgroup_exit_cftypes(cfts); | |
3532 | return 0; | |
8e3f6541 | 3533 | } |
8e3f6541 | 3534 | |
79578621 TH |
3535 | /** |
3536 | * cgroup_rm_cftypes - remove an array of cftypes from a subsystem | |
79578621 TH |
3537 | * @cfts: zero-length name terminated array of cftypes |
3538 | * | |
2bb566cb TH |
3539 | * Unregister @cfts. Files described by @cfts are removed from all |
3540 | * existing cgroups and all future cgroups won't have them either. This | |
3541 | * function can be called anytime whether @cfts' subsys is attached or not. | |
79578621 TH |
3542 | * |
3543 | * Returns 0 on successful unregistration, -ENOENT if @cfts is not | |
2bb566cb | 3544 | * registered. |
79578621 | 3545 | */ |
2bb566cb | 3546 | int cgroup_rm_cftypes(struct cftype *cfts) |
79578621 | 3547 | { |
21a2d343 | 3548 | int ret; |
79578621 | 3549 | |
01f6474c | 3550 | mutex_lock(&cgroup_mutex); |
21a2d343 | 3551 | ret = cgroup_rm_cftypes_locked(cfts); |
01f6474c | 3552 | mutex_unlock(&cgroup_mutex); |
21a2d343 | 3553 | return ret; |
80b13586 TH |
3554 | } |
3555 | ||
8e3f6541 TH |
3556 | /** |
3557 | * cgroup_add_cftypes - add an array of cftypes to a subsystem | |
3558 | * @ss: target cgroup subsystem | |
3559 | * @cfts: zero-length name terminated array of cftypes | |
3560 | * | |
3561 | * Register @cfts to @ss. Files described by @cfts are created for all | |
3562 | * existing cgroups to which @ss is attached and all future cgroups will | |
3563 | * have them too. This function can be called anytime whether @ss is | |
3564 | * attached or not. | |
3565 | * | |
3566 | * Returns 0 on successful registration, -errno on failure. Note that this | |
3567 | * function currently returns 0 as long as @cfts registration is successful | |
3568 | * even if some file creation attempts on existing cgroups fail. | |
3569 | */ | |
2cf669a5 | 3570 | static int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
8e3f6541 | 3571 | { |
9ccece80 | 3572 | int ret; |
8e3f6541 | 3573 | |
fc5ed1e9 | 3574 | if (!cgroup_ssid_enabled(ss->id)) |
c731ae1d LZ |
3575 | return 0; |
3576 | ||
dc5736ed LZ |
3577 | if (!cfts || cfts[0].name[0] == '\0') |
3578 | return 0; | |
2bb566cb | 3579 | |
2bd59d48 TH |
3580 | ret = cgroup_init_cftypes(ss, cfts); |
3581 | if (ret) | |
3582 | return ret; | |
79578621 | 3583 | |
01f6474c | 3584 | mutex_lock(&cgroup_mutex); |
21a2d343 | 3585 | |
0adb0704 | 3586 | list_add_tail(&cfts->node, &ss->cfts); |
21a2d343 | 3587 | ret = cgroup_apply_cftypes(cfts, true); |
9ccece80 | 3588 | if (ret) |
21a2d343 | 3589 | cgroup_rm_cftypes_locked(cfts); |
79578621 | 3590 | |
01f6474c | 3591 | mutex_unlock(&cgroup_mutex); |
9ccece80 | 3592 | return ret; |
79578621 TH |
3593 | } |
3594 | ||
a8ddc821 TH |
3595 | /** |
3596 | * cgroup_add_dfl_cftypes - add an array of cftypes for default hierarchy | |
3597 | * @ss: target cgroup subsystem | |
3598 | * @cfts: zero-length name terminated array of cftypes | |
3599 | * | |
3600 | * Similar to cgroup_add_cftypes() but the added files are only used for | |
3601 | * the default hierarchy. | |
3602 | */ | |
3603 | int cgroup_add_dfl_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) | |
3604 | { | |
3605 | struct cftype *cft; | |
3606 | ||
3607 | for (cft = cfts; cft && cft->name[0] != '\0'; cft++) | |
05ebb6e6 | 3608 | cft->flags |= __CFTYPE_ONLY_ON_DFL; |
a8ddc821 TH |
3609 | return cgroup_add_cftypes(ss, cfts); |
3610 | } | |
3611 | ||
3612 | /** | |
3613 | * cgroup_add_legacy_cftypes - add an array of cftypes for legacy hierarchies | |
3614 | * @ss: target cgroup subsystem | |
3615 | * @cfts: zero-length name terminated array of cftypes | |
3616 | * | |
3617 | * Similar to cgroup_add_cftypes() but the added files are only used for | |
3618 | * the legacy hierarchies. | |
3619 | */ | |
2cf669a5 TH |
3620 | int cgroup_add_legacy_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
3621 | { | |
a8ddc821 TH |
3622 | struct cftype *cft; |
3623 | ||
e4b7037c TH |
3624 | for (cft = cfts; cft && cft->name[0] != '\0'; cft++) |
3625 | cft->flags |= __CFTYPE_NOT_ON_DFL; | |
2cf669a5 TH |
3626 | return cgroup_add_cftypes(ss, cfts); |
3627 | } | |
3628 | ||
34c06254 TH |
3629 | /** |
3630 | * cgroup_file_notify - generate a file modified event for a cgroup_file | |
3631 | * @cfile: target cgroup_file | |
3632 | * | |
3633 | * @cfile must have been obtained by setting cftype->file_offset. | |
3634 | */ | |
3635 | void cgroup_file_notify(struct cgroup_file *cfile) | |
3636 | { | |
3637 | unsigned long flags; | |
3638 | ||
3639 | spin_lock_irqsave(&cgroup_file_kn_lock, flags); | |
3640 | if (cfile->kn) | |
3641 | kernfs_notify(cfile->kn); | |
3642 | spin_unlock_irqrestore(&cgroup_file_kn_lock, flags); | |
3643 | } | |
3644 | ||
a043e3b2 LZ |
3645 | /** |
3646 | * cgroup_task_count - count the number of tasks in a cgroup. | |
3647 | * @cgrp: the cgroup in question | |
3648 | * | |
3649 | * Return the number of tasks in the cgroup. | |
3650 | */ | |
07bc356e | 3651 | static int cgroup_task_count(const struct cgroup *cgrp) |
bbcb81d0 PM |
3652 | { |
3653 | int count = 0; | |
69d0206c | 3654 | struct cgrp_cset_link *link; |
817929ec | 3655 | |
f0d9a5f1 | 3656 | spin_lock_bh(&css_set_lock); |
69d0206c TH |
3657 | list_for_each_entry(link, &cgrp->cset_links, cset_link) |
3658 | count += atomic_read(&link->cset->refcount); | |
f0d9a5f1 | 3659 | spin_unlock_bh(&css_set_lock); |
bbcb81d0 PM |
3660 | return count; |
3661 | } | |
3662 | ||
53fa5261 | 3663 | /** |
492eb21b | 3664 | * css_next_child - find the next child of a given css |
c2931b70 TH |
3665 | * @pos: the current position (%NULL to initiate traversal) |
3666 | * @parent: css whose children to walk | |
53fa5261 | 3667 | * |
c2931b70 | 3668 | * This function returns the next child of @parent and should be called |
87fb54f1 | 3669 | * under either cgroup_mutex or RCU read lock. The only requirement is |
c2931b70 TH |
3670 | * that @parent and @pos are accessible. The next sibling is guaranteed to |
3671 | * be returned regardless of their states. | |
3672 | * | |
3673 | * If a subsystem synchronizes ->css_online() and the start of iteration, a | |
3674 | * css which finished ->css_online() is guaranteed to be visible in the | |
3675 | * future iterations and will stay visible until the last reference is put. | |
3676 | * A css which hasn't finished ->css_online() or already finished | |
3677 | * ->css_offline() may show up during traversal. It's each subsystem's | |
3678 | * responsibility to synchronize against on/offlining. | |
53fa5261 | 3679 | */ |
c2931b70 TH |
3680 | struct cgroup_subsys_state *css_next_child(struct cgroup_subsys_state *pos, |
3681 | struct cgroup_subsys_state *parent) | |
53fa5261 | 3682 | { |
c2931b70 | 3683 | struct cgroup_subsys_state *next; |
53fa5261 | 3684 | |
8353da1f | 3685 | cgroup_assert_mutex_or_rcu_locked(); |
53fa5261 TH |
3686 | |
3687 | /* | |
de3f0341 TH |
3688 | * @pos could already have been unlinked from the sibling list. |
3689 | * Once a cgroup is removed, its ->sibling.next is no longer | |
3690 | * updated when its next sibling changes. CSS_RELEASED is set when | |
3691 | * @pos is taken off list, at which time its next pointer is valid, | |
3692 | * and, as releases are serialized, the one pointed to by the next | |
3693 | * pointer is guaranteed to not have started release yet. This | |
3694 | * implies that if we observe !CSS_RELEASED on @pos in this RCU | |
3695 | * critical section, the one pointed to by its next pointer is | |
3696 | * guaranteed to not have finished its RCU grace period even if we | |
3697 | * have dropped rcu_read_lock() inbetween iterations. | |
3b287a50 | 3698 | * |
de3f0341 TH |
3699 | * If @pos has CSS_RELEASED set, its next pointer can't be |
3700 | * dereferenced; however, as each css is given a monotonically | |
3701 | * increasing unique serial number and always appended to the | |
3702 | * sibling list, the next one can be found by walking the parent's | |
3703 | * children until the first css with higher serial number than | |
3704 | * @pos's. While this path can be slower, it happens iff iteration | |
3705 | * races against release and the race window is very small. | |
53fa5261 | 3706 | */ |
3b287a50 | 3707 | if (!pos) { |
c2931b70 TH |
3708 | next = list_entry_rcu(parent->children.next, struct cgroup_subsys_state, sibling); |
3709 | } else if (likely(!(pos->flags & CSS_RELEASED))) { | |
3710 | next = list_entry_rcu(pos->sibling.next, struct cgroup_subsys_state, sibling); | |
3b287a50 | 3711 | } else { |
c2931b70 | 3712 | list_for_each_entry_rcu(next, &parent->children, sibling) |
3b287a50 TH |
3713 | if (next->serial_nr > pos->serial_nr) |
3714 | break; | |
53fa5261 TH |
3715 | } |
3716 | ||
3b281afb TH |
3717 | /* |
3718 | * @next, if not pointing to the head, can be dereferenced and is | |
c2931b70 | 3719 | * the next sibling. |
3b281afb | 3720 | */ |
c2931b70 TH |
3721 | if (&next->sibling != &parent->children) |
3722 | return next; | |
3b281afb | 3723 | return NULL; |
53fa5261 | 3724 | } |
53fa5261 | 3725 | |
574bd9f7 | 3726 | /** |
492eb21b | 3727 | * css_next_descendant_pre - find the next descendant for pre-order walk |
574bd9f7 | 3728 | * @pos: the current position (%NULL to initiate traversal) |
492eb21b | 3729 | * @root: css whose descendants to walk |
574bd9f7 | 3730 | * |
492eb21b | 3731 | * To be used by css_for_each_descendant_pre(). Find the next descendant |
bd8815a6 TH |
3732 | * to visit for pre-order traversal of @root's descendants. @root is |
3733 | * included in the iteration and the first node to be visited. | |
75501a6d | 3734 | * |
87fb54f1 TH |
3735 | * While this function requires cgroup_mutex or RCU read locking, it |
3736 | * doesn't require the whole traversal to be contained in a single critical | |
3737 | * section. This function will return the correct next descendant as long | |
3738 | * as both @pos and @root are accessible and @pos is a descendant of @root. | |
c2931b70 TH |
3739 | * |
3740 | * If a subsystem synchronizes ->css_online() and the start of iteration, a | |
3741 | * css which finished ->css_online() is guaranteed to be visible in the | |
3742 | * future iterations and will stay visible until the last reference is put. | |
3743 | * A css which hasn't finished ->css_online() or already finished | |
3744 | * ->css_offline() may show up during traversal. It's each subsystem's | |
3745 | * responsibility to synchronize against on/offlining. | |
574bd9f7 | 3746 | */ |
492eb21b TH |
3747 | struct cgroup_subsys_state * |
3748 | css_next_descendant_pre(struct cgroup_subsys_state *pos, | |
3749 | struct cgroup_subsys_state *root) | |
574bd9f7 | 3750 | { |
492eb21b | 3751 | struct cgroup_subsys_state *next; |
574bd9f7 | 3752 | |
8353da1f | 3753 | cgroup_assert_mutex_or_rcu_locked(); |
574bd9f7 | 3754 | |
bd8815a6 | 3755 | /* if first iteration, visit @root */ |
7805d000 | 3756 | if (!pos) |
bd8815a6 | 3757 | return root; |
574bd9f7 TH |
3758 | |
3759 | /* visit the first child if exists */ | |
492eb21b | 3760 | next = css_next_child(NULL, pos); |
574bd9f7 TH |
3761 | if (next) |
3762 | return next; | |
3763 | ||
3764 | /* no child, visit my or the closest ancestor's next sibling */ | |
492eb21b | 3765 | while (pos != root) { |
5c9d535b | 3766 | next = css_next_child(pos, pos->parent); |
75501a6d | 3767 | if (next) |
574bd9f7 | 3768 | return next; |
5c9d535b | 3769 | pos = pos->parent; |
7805d000 | 3770 | } |
574bd9f7 TH |
3771 | |
3772 | return NULL; | |
3773 | } | |
574bd9f7 | 3774 | |
12a9d2fe | 3775 | /** |
492eb21b TH |
3776 | * css_rightmost_descendant - return the rightmost descendant of a css |
3777 | * @pos: css of interest | |
12a9d2fe | 3778 | * |
492eb21b TH |
3779 | * Return the rightmost descendant of @pos. If there's no descendant, @pos |
3780 | * is returned. This can be used during pre-order traversal to skip | |
12a9d2fe | 3781 | * subtree of @pos. |
75501a6d | 3782 | * |
87fb54f1 TH |
3783 | * While this function requires cgroup_mutex or RCU read locking, it |
3784 | * doesn't require the whole traversal to be contained in a single critical | |
3785 | * section. This function will return the correct rightmost descendant as | |
3786 | * long as @pos is accessible. | |
12a9d2fe | 3787 | */ |
492eb21b TH |
3788 | struct cgroup_subsys_state * |
3789 | css_rightmost_descendant(struct cgroup_subsys_state *pos) | |
12a9d2fe | 3790 | { |
492eb21b | 3791 | struct cgroup_subsys_state *last, *tmp; |
12a9d2fe | 3792 | |
8353da1f | 3793 | cgroup_assert_mutex_or_rcu_locked(); |
12a9d2fe TH |
3794 | |
3795 | do { | |
3796 | last = pos; | |
3797 | /* ->prev isn't RCU safe, walk ->next till the end */ | |
3798 | pos = NULL; | |
492eb21b | 3799 | css_for_each_child(tmp, last) |
12a9d2fe TH |
3800 | pos = tmp; |
3801 | } while (pos); | |
3802 | ||
3803 | return last; | |
3804 | } | |
12a9d2fe | 3805 | |
492eb21b TH |
3806 | static struct cgroup_subsys_state * |
3807 | css_leftmost_descendant(struct cgroup_subsys_state *pos) | |
574bd9f7 | 3808 | { |
492eb21b | 3809 | struct cgroup_subsys_state *last; |
574bd9f7 TH |
3810 | |
3811 | do { | |
3812 | last = pos; | |
492eb21b | 3813 | pos = css_next_child(NULL, pos); |
574bd9f7 TH |
3814 | } while (pos); |
3815 | ||
3816 | return last; | |
3817 | } | |
3818 | ||
3819 | /** | |
492eb21b | 3820 | * css_next_descendant_post - find the next descendant for post-order walk |
574bd9f7 | 3821 | * @pos: the current position (%NULL to initiate traversal) |
492eb21b | 3822 | * @root: css whose descendants to walk |
574bd9f7 | 3823 | * |
492eb21b | 3824 | * To be used by css_for_each_descendant_post(). Find the next descendant |
bd8815a6 TH |
3825 | * to visit for post-order traversal of @root's descendants. @root is |
3826 | * included in the iteration and the last node to be visited. | |
75501a6d | 3827 | * |
87fb54f1 TH |
3828 | * While this function requires cgroup_mutex or RCU read locking, it |
3829 | * doesn't require the whole traversal to be contained in a single critical | |
3830 | * section. This function will return the correct next descendant as long | |
3831 | * as both @pos and @cgroup are accessible and @pos is a descendant of | |
3832 | * @cgroup. | |
c2931b70 TH |
3833 | * |
3834 | * If a subsystem synchronizes ->css_online() and the start of iteration, a | |
3835 | * css which finished ->css_online() is guaranteed to be visible in the | |
3836 | * future iterations and will stay visible until the last reference is put. | |
3837 | * A css which hasn't finished ->css_online() or already finished | |
3838 | * ->css_offline() may show up during traversal. It's each subsystem's | |
3839 | * responsibility to synchronize against on/offlining. | |
574bd9f7 | 3840 | */ |
492eb21b TH |
3841 | struct cgroup_subsys_state * |
3842 | css_next_descendant_post(struct cgroup_subsys_state *pos, | |
3843 | struct cgroup_subsys_state *root) | |
574bd9f7 | 3844 | { |
492eb21b | 3845 | struct cgroup_subsys_state *next; |
574bd9f7 | 3846 | |
8353da1f | 3847 | cgroup_assert_mutex_or_rcu_locked(); |
574bd9f7 | 3848 | |
58b79a91 TH |
3849 | /* if first iteration, visit leftmost descendant which may be @root */ |
3850 | if (!pos) | |
3851 | return css_leftmost_descendant(root); | |
574bd9f7 | 3852 | |
bd8815a6 TH |
3853 | /* if we visited @root, we're done */ |
3854 | if (pos == root) | |
3855 | return NULL; | |
3856 | ||
574bd9f7 | 3857 | /* if there's an unvisited sibling, visit its leftmost descendant */ |
5c9d535b | 3858 | next = css_next_child(pos, pos->parent); |
75501a6d | 3859 | if (next) |
492eb21b | 3860 | return css_leftmost_descendant(next); |
574bd9f7 TH |
3861 | |
3862 | /* no sibling left, visit parent */ | |
5c9d535b | 3863 | return pos->parent; |
574bd9f7 | 3864 | } |
574bd9f7 | 3865 | |
f3d46500 TH |
3866 | /** |
3867 | * css_has_online_children - does a css have online children | |
3868 | * @css: the target css | |
3869 | * | |
3870 | * Returns %true if @css has any online children; otherwise, %false. This | |
3871 | * function can be called from any context but the caller is responsible | |
3872 | * for synchronizing against on/offlining as necessary. | |
3873 | */ | |
3874 | bool css_has_online_children(struct cgroup_subsys_state *css) | |
cbc125ef | 3875 | { |
f3d46500 TH |
3876 | struct cgroup_subsys_state *child; |
3877 | bool ret = false; | |
cbc125ef TH |
3878 | |
3879 | rcu_read_lock(); | |
f3d46500 | 3880 | css_for_each_child(child, css) { |
99bae5f9 | 3881 | if (child->flags & CSS_ONLINE) { |
f3d46500 TH |
3882 | ret = true; |
3883 | break; | |
cbc125ef TH |
3884 | } |
3885 | } | |
3886 | rcu_read_unlock(); | |
f3d46500 | 3887 | return ret; |
574bd9f7 | 3888 | } |
574bd9f7 | 3889 | |
0942eeee | 3890 | /** |
ecb9d535 | 3891 | * css_task_iter_advance_css_set - advance a task itererator to the next css_set |
0942eeee TH |
3892 | * @it: the iterator to advance |
3893 | * | |
3894 | * Advance @it to the next css_set to walk. | |
d515876e | 3895 | */ |
ecb9d535 | 3896 | static void css_task_iter_advance_css_set(struct css_task_iter *it) |
d515876e | 3897 | { |
0f0a2b4f | 3898 | struct list_head *l = it->cset_pos; |
d515876e TH |
3899 | struct cgrp_cset_link *link; |
3900 | struct css_set *cset; | |
3901 | ||
f0d9a5f1 | 3902 | lockdep_assert_held(&css_set_lock); |
ed27b9f7 | 3903 | |
d515876e TH |
3904 | /* Advance to the next non-empty css_set */ |
3905 | do { | |
3906 | l = l->next; | |
0f0a2b4f TH |
3907 | if (l == it->cset_head) { |
3908 | it->cset_pos = NULL; | |
ecb9d535 | 3909 | it->task_pos = NULL; |
d515876e TH |
3910 | return; |
3911 | } | |
3ebb2b6e TH |
3912 | |
3913 | if (it->ss) { | |
3914 | cset = container_of(l, struct css_set, | |
3915 | e_cset_node[it->ss->id]); | |
3916 | } else { | |
3917 | link = list_entry(l, struct cgrp_cset_link, cset_link); | |
3918 | cset = link->cset; | |
3919 | } | |
0de0942d | 3920 | } while (!css_set_populated(cset)); |
c7561128 | 3921 | |
0f0a2b4f | 3922 | it->cset_pos = l; |
c7561128 TH |
3923 | |
3924 | if (!list_empty(&cset->tasks)) | |
0f0a2b4f | 3925 | it->task_pos = cset->tasks.next; |
c7561128 | 3926 | else |
0f0a2b4f TH |
3927 | it->task_pos = cset->mg_tasks.next; |
3928 | ||
3929 | it->tasks_head = &cset->tasks; | |
3930 | it->mg_tasks_head = &cset->mg_tasks; | |
ed27b9f7 TH |
3931 | |
3932 | /* | |
3933 | * We don't keep css_sets locked across iteration steps and thus | |
3934 | * need to take steps to ensure that iteration can be resumed after | |
3935 | * the lock is re-acquired. Iteration is performed at two levels - | |
3936 | * css_sets and tasks in them. | |
3937 | * | |
3938 | * Once created, a css_set never leaves its cgroup lists, so a | |
3939 | * pinned css_set is guaranteed to stay put and we can resume | |
3940 | * iteration afterwards. | |
3941 | * | |
3942 | * Tasks may leave @cset across iteration steps. This is resolved | |
3943 | * by registering each iterator with the css_set currently being | |
3944 | * walked and making css_set_move_task() advance iterators whose | |
3945 | * next task is leaving. | |
3946 | */ | |
3947 | if (it->cur_cset) { | |
3948 | list_del(&it->iters_node); | |
3949 | put_css_set_locked(it->cur_cset); | |
3950 | } | |
3951 | get_css_set(cset); | |
3952 | it->cur_cset = cset; | |
3953 | list_add(&it->iters_node, &cset->task_iters); | |
d515876e TH |
3954 | } |
3955 | ||
ecb9d535 TH |
3956 | static void css_task_iter_advance(struct css_task_iter *it) |
3957 | { | |
3958 | struct list_head *l = it->task_pos; | |
3959 | ||
f0d9a5f1 | 3960 | lockdep_assert_held(&css_set_lock); |
ecb9d535 TH |
3961 | WARN_ON_ONCE(!l); |
3962 | ||
3963 | /* | |
3964 | * Advance iterator to find next entry. cset->tasks is consumed | |
3965 | * first and then ->mg_tasks. After ->mg_tasks, we move onto the | |
3966 | * next cset. | |
3967 | */ | |
3968 | l = l->next; | |
3969 | ||
3970 | if (l == it->tasks_head) | |
3971 | l = it->mg_tasks_head->next; | |
3972 | ||
3973 | if (l == it->mg_tasks_head) | |
3974 | css_task_iter_advance_css_set(it); | |
3975 | else | |
3976 | it->task_pos = l; | |
3977 | } | |
3978 | ||
0942eeee | 3979 | /** |
72ec7029 TH |
3980 | * css_task_iter_start - initiate task iteration |
3981 | * @css: the css to walk tasks of | |
0942eeee TH |
3982 | * @it: the task iterator to use |
3983 | * | |
72ec7029 TH |
3984 | * Initiate iteration through the tasks of @css. The caller can call |
3985 | * css_task_iter_next() to walk through the tasks until the function | |
3986 | * returns NULL. On completion of iteration, css_task_iter_end() must be | |
3987 | * called. | |
0942eeee | 3988 | */ |
72ec7029 TH |
3989 | void css_task_iter_start(struct cgroup_subsys_state *css, |
3990 | struct css_task_iter *it) | |
817929ec | 3991 | { |
56fde9e0 TH |
3992 | /* no one should try to iterate before mounting cgroups */ |
3993 | WARN_ON_ONCE(!use_task_css_set_links); | |
31a7df01 | 3994 | |
ed27b9f7 TH |
3995 | memset(it, 0, sizeof(*it)); |
3996 | ||
f0d9a5f1 | 3997 | spin_lock_bh(&css_set_lock); |
c59cd3d8 | 3998 | |
3ebb2b6e TH |
3999 | it->ss = css->ss; |
4000 | ||
4001 | if (it->ss) | |
4002 | it->cset_pos = &css->cgroup->e_csets[css->ss->id]; | |
4003 | else | |
4004 | it->cset_pos = &css->cgroup->cset_links; | |
4005 | ||
0f0a2b4f | 4006 | it->cset_head = it->cset_pos; |
c59cd3d8 | 4007 | |
ecb9d535 | 4008 | css_task_iter_advance_css_set(it); |
ed27b9f7 | 4009 | |
f0d9a5f1 | 4010 | spin_unlock_bh(&css_set_lock); |
817929ec PM |
4011 | } |
4012 | ||
0942eeee | 4013 | /** |
72ec7029 | 4014 | * css_task_iter_next - return the next task for the iterator |
0942eeee TH |
4015 | * @it: the task iterator being iterated |
4016 | * | |
4017 | * The "next" function for task iteration. @it should have been | |
72ec7029 TH |
4018 | * initialized via css_task_iter_start(). Returns NULL when the iteration |
4019 | * reaches the end. | |
0942eeee | 4020 | */ |
72ec7029 | 4021 | struct task_struct *css_task_iter_next(struct css_task_iter *it) |
817929ec | 4022 | { |
d5745675 | 4023 | if (it->cur_task) { |
ed27b9f7 | 4024 | put_task_struct(it->cur_task); |
d5745675 TH |
4025 | it->cur_task = NULL; |
4026 | } | |
ed27b9f7 | 4027 | |
f0d9a5f1 | 4028 | spin_lock_bh(&css_set_lock); |
ed27b9f7 | 4029 | |
d5745675 TH |
4030 | if (it->task_pos) { |
4031 | it->cur_task = list_entry(it->task_pos, struct task_struct, | |
4032 | cg_list); | |
4033 | get_task_struct(it->cur_task); | |
4034 | css_task_iter_advance(it); | |
4035 | } | |
ed27b9f7 | 4036 | |
f0d9a5f1 | 4037 | spin_unlock_bh(&css_set_lock); |
ed27b9f7 TH |
4038 | |
4039 | return it->cur_task; | |
817929ec PM |
4040 | } |
4041 | ||
0942eeee | 4042 | /** |
72ec7029 | 4043 | * css_task_iter_end - finish task iteration |
0942eeee TH |
4044 | * @it: the task iterator to finish |
4045 | * | |
72ec7029 | 4046 | * Finish task iteration started by css_task_iter_start(). |
0942eeee | 4047 | */ |
72ec7029 | 4048 | void css_task_iter_end(struct css_task_iter *it) |
31a7df01 | 4049 | { |
ed27b9f7 | 4050 | if (it->cur_cset) { |
f0d9a5f1 | 4051 | spin_lock_bh(&css_set_lock); |
ed27b9f7 TH |
4052 | list_del(&it->iters_node); |
4053 | put_css_set_locked(it->cur_cset); | |
f0d9a5f1 | 4054 | spin_unlock_bh(&css_set_lock); |
ed27b9f7 TH |
4055 | } |
4056 | ||
4057 | if (it->cur_task) | |
4058 | put_task_struct(it->cur_task); | |
31a7df01 CW |
4059 | } |
4060 | ||
4061 | /** | |
8cc99345 TH |
4062 | * cgroup_trasnsfer_tasks - move tasks from one cgroup to another |
4063 | * @to: cgroup to which the tasks will be moved | |
4064 | * @from: cgroup in which the tasks currently reside | |
31a7df01 | 4065 | * |
eaf797ab TH |
4066 | * Locking rules between cgroup_post_fork() and the migration path |
4067 | * guarantee that, if a task is forking while being migrated, the new child | |
4068 | * is guaranteed to be either visible in the source cgroup after the | |
4069 | * parent's migration is complete or put into the target cgroup. No task | |
4070 | * can slip out of migration through forking. | |
31a7df01 | 4071 | */ |
8cc99345 | 4072 | int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from) |
31a7df01 | 4073 | { |
952aaa12 TH |
4074 | LIST_HEAD(preloaded_csets); |
4075 | struct cgrp_cset_link *link; | |
72ec7029 | 4076 | struct css_task_iter it; |
e406d1cf | 4077 | struct task_struct *task; |
952aaa12 | 4078 | int ret; |
31a7df01 | 4079 | |
952aaa12 | 4080 | mutex_lock(&cgroup_mutex); |
31a7df01 | 4081 | |
952aaa12 | 4082 | /* all tasks in @from are being moved, all csets are source */ |
f0d9a5f1 | 4083 | spin_lock_bh(&css_set_lock); |
952aaa12 TH |
4084 | list_for_each_entry(link, &from->cset_links, cset_link) |
4085 | cgroup_migrate_add_src(link->cset, to, &preloaded_csets); | |
f0d9a5f1 | 4086 | spin_unlock_bh(&css_set_lock); |
31a7df01 | 4087 | |
952aaa12 TH |
4088 | ret = cgroup_migrate_prepare_dst(to, &preloaded_csets); |
4089 | if (ret) | |
4090 | goto out_err; | |
8cc99345 | 4091 | |
952aaa12 | 4092 | /* |
2cfa2b19 | 4093 | * Migrate tasks one-by-one until @from is empty. This fails iff |
952aaa12 TH |
4094 | * ->can_attach() fails. |
4095 | */ | |
e406d1cf | 4096 | do { |
9d800df1 | 4097 | css_task_iter_start(&from->self, &it); |
e406d1cf TH |
4098 | task = css_task_iter_next(&it); |
4099 | if (task) | |
4100 | get_task_struct(task); | |
4101 | css_task_iter_end(&it); | |
4102 | ||
4103 | if (task) { | |
9af2ec45 | 4104 | ret = cgroup_migrate(task, false, to); |
e406d1cf TH |
4105 | put_task_struct(task); |
4106 | } | |
4107 | } while (task && !ret); | |
952aaa12 TH |
4108 | out_err: |
4109 | cgroup_migrate_finish(&preloaded_csets); | |
47cfcd09 | 4110 | mutex_unlock(&cgroup_mutex); |
e406d1cf | 4111 | return ret; |
8cc99345 TH |
4112 | } |
4113 | ||
bbcb81d0 | 4114 | /* |
102a775e | 4115 | * Stuff for reading the 'tasks'/'procs' files. |
bbcb81d0 PM |
4116 | * |
4117 | * Reading this file can return large amounts of data if a cgroup has | |
4118 | * *lots* of attached tasks. So it may need several calls to read(), | |
4119 | * but we cannot guarantee that the information we produce is correct | |
4120 | * unless we produce it entirely atomically. | |
4121 | * | |
bbcb81d0 | 4122 | */ |
bbcb81d0 | 4123 | |
24528255 LZ |
4124 | /* which pidlist file are we talking about? */ |
4125 | enum cgroup_filetype { | |
4126 | CGROUP_FILE_PROCS, | |
4127 | CGROUP_FILE_TASKS, | |
4128 | }; | |
4129 | ||
4130 | /* | |
4131 | * A pidlist is a list of pids that virtually represents the contents of one | |
4132 | * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists, | |
4133 | * a pair (one each for procs, tasks) for each pid namespace that's relevant | |
4134 | * to the cgroup. | |
4135 | */ | |
4136 | struct cgroup_pidlist { | |
4137 | /* | |
4138 | * used to find which pidlist is wanted. doesn't change as long as | |
4139 | * this particular list stays in the list. | |
4140 | */ | |
4141 | struct { enum cgroup_filetype type; struct pid_namespace *ns; } key; | |
4142 | /* array of xids */ | |
4143 | pid_t *list; | |
4144 | /* how many elements the above list has */ | |
4145 | int length; | |
24528255 LZ |
4146 | /* each of these stored in a list by its cgroup */ |
4147 | struct list_head links; | |
4148 | /* pointer to the cgroup we belong to, for list removal purposes */ | |
4149 | struct cgroup *owner; | |
b1a21367 TH |
4150 | /* for delayed destruction */ |
4151 | struct delayed_work destroy_dwork; | |
24528255 LZ |
4152 | }; |
4153 | ||
d1d9fd33 BB |
4154 | /* |
4155 | * The following two functions "fix" the issue where there are more pids | |
4156 | * than kmalloc will give memory for; in such cases, we use vmalloc/vfree. | |
4157 | * TODO: replace with a kernel-wide solution to this problem | |
4158 | */ | |
4159 | #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2)) | |
4160 | static void *pidlist_allocate(int count) | |
4161 | { | |
4162 | if (PIDLIST_TOO_LARGE(count)) | |
4163 | return vmalloc(count * sizeof(pid_t)); | |
4164 | else | |
4165 | return kmalloc(count * sizeof(pid_t), GFP_KERNEL); | |
4166 | } | |
b1a21367 | 4167 | |
d1d9fd33 BB |
4168 | static void pidlist_free(void *p) |
4169 | { | |
58794514 | 4170 | kvfree(p); |
d1d9fd33 | 4171 | } |
d1d9fd33 | 4172 | |
b1a21367 TH |
4173 | /* |
4174 | * Used to destroy all pidlists lingering waiting for destroy timer. None | |
4175 | * should be left afterwards. | |
4176 | */ | |
4177 | static void cgroup_pidlist_destroy_all(struct cgroup *cgrp) | |
4178 | { | |
4179 | struct cgroup_pidlist *l, *tmp_l; | |
4180 | ||
4181 | mutex_lock(&cgrp->pidlist_mutex); | |
4182 | list_for_each_entry_safe(l, tmp_l, &cgrp->pidlists, links) | |
4183 | mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, 0); | |
4184 | mutex_unlock(&cgrp->pidlist_mutex); | |
4185 | ||
4186 | flush_workqueue(cgroup_pidlist_destroy_wq); | |
4187 | BUG_ON(!list_empty(&cgrp->pidlists)); | |
4188 | } | |
4189 | ||
4190 | static void cgroup_pidlist_destroy_work_fn(struct work_struct *work) | |
4191 | { | |
4192 | struct delayed_work *dwork = to_delayed_work(work); | |
4193 | struct cgroup_pidlist *l = container_of(dwork, struct cgroup_pidlist, | |
4194 | destroy_dwork); | |
4195 | struct cgroup_pidlist *tofree = NULL; | |
4196 | ||
4197 | mutex_lock(&l->owner->pidlist_mutex); | |
b1a21367 TH |
4198 | |
4199 | /* | |
04502365 TH |
4200 | * Destroy iff we didn't get queued again. The state won't change |
4201 | * as destroy_dwork can only be queued while locked. | |
b1a21367 | 4202 | */ |
04502365 | 4203 | if (!delayed_work_pending(dwork)) { |
b1a21367 TH |
4204 | list_del(&l->links); |
4205 | pidlist_free(l->list); | |
4206 | put_pid_ns(l->key.ns); | |
4207 | tofree = l; | |
4208 | } | |
4209 | ||
b1a21367 TH |
4210 | mutex_unlock(&l->owner->pidlist_mutex); |
4211 | kfree(tofree); | |
4212 | } | |
4213 | ||
bbcb81d0 | 4214 | /* |
102a775e | 4215 | * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries |
6ee211ad | 4216 | * Returns the number of unique elements. |
bbcb81d0 | 4217 | */ |
6ee211ad | 4218 | static int pidlist_uniq(pid_t *list, int length) |
bbcb81d0 | 4219 | { |
102a775e | 4220 | int src, dest = 1; |
102a775e BB |
4221 | |
4222 | /* | |
4223 | * we presume the 0th element is unique, so i starts at 1. trivial | |
4224 | * edge cases first; no work needs to be done for either | |
4225 | */ | |
4226 | if (length == 0 || length == 1) | |
4227 | return length; | |
4228 | /* src and dest walk down the list; dest counts unique elements */ | |
4229 | for (src = 1; src < length; src++) { | |
4230 | /* find next unique element */ | |
4231 | while (list[src] == list[src-1]) { | |
4232 | src++; | |
4233 | if (src == length) | |
4234 | goto after; | |
4235 | } | |
4236 | /* dest always points to where the next unique element goes */ | |
4237 | list[dest] = list[src]; | |
4238 | dest++; | |
4239 | } | |
4240 | after: | |
102a775e BB |
4241 | return dest; |
4242 | } | |
4243 | ||
afb2bc14 TH |
4244 | /* |
4245 | * The two pid files - task and cgroup.procs - guaranteed that the result | |
4246 | * is sorted, which forced this whole pidlist fiasco. As pid order is | |
4247 | * different per namespace, each namespace needs differently sorted list, | |
4248 | * making it impossible to use, for example, single rbtree of member tasks | |
4249 | * sorted by task pointer. As pidlists can be fairly large, allocating one | |
4250 | * per open file is dangerous, so cgroup had to implement shared pool of | |
4251 | * pidlists keyed by cgroup and namespace. | |
4252 | * | |
4253 | * All this extra complexity was caused by the original implementation | |
4254 | * committing to an entirely unnecessary property. In the long term, we | |
aa6ec29b TH |
4255 | * want to do away with it. Explicitly scramble sort order if on the |
4256 | * default hierarchy so that no such expectation exists in the new | |
4257 | * interface. | |
afb2bc14 TH |
4258 | * |
4259 | * Scrambling is done by swapping every two consecutive bits, which is | |
4260 | * non-identity one-to-one mapping which disturbs sort order sufficiently. | |
4261 | */ | |
4262 | static pid_t pid_fry(pid_t pid) | |
4263 | { | |
4264 | unsigned a = pid & 0x55555555; | |
4265 | unsigned b = pid & 0xAAAAAAAA; | |
4266 | ||
4267 | return (a << 1) | (b >> 1); | |
4268 | } | |
4269 | ||
4270 | static pid_t cgroup_pid_fry(struct cgroup *cgrp, pid_t pid) | |
4271 | { | |
aa6ec29b | 4272 | if (cgroup_on_dfl(cgrp)) |
afb2bc14 TH |
4273 | return pid_fry(pid); |
4274 | else | |
4275 | return pid; | |
4276 | } | |
4277 | ||
102a775e BB |
4278 | static int cmppid(const void *a, const void *b) |
4279 | { | |
4280 | return *(pid_t *)a - *(pid_t *)b; | |
4281 | } | |
4282 | ||
afb2bc14 TH |
4283 | static int fried_cmppid(const void *a, const void *b) |
4284 | { | |
4285 | return pid_fry(*(pid_t *)a) - pid_fry(*(pid_t *)b); | |
4286 | } | |
4287 | ||
e6b81710 TH |
4288 | static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp, |
4289 | enum cgroup_filetype type) | |
4290 | { | |
4291 | struct cgroup_pidlist *l; | |
4292 | /* don't need task_nsproxy() if we're looking at ourself */ | |
4293 | struct pid_namespace *ns = task_active_pid_ns(current); | |
4294 | ||
4295 | lockdep_assert_held(&cgrp->pidlist_mutex); | |
4296 | ||
4297 | list_for_each_entry(l, &cgrp->pidlists, links) | |
4298 | if (l->key.type == type && l->key.ns == ns) | |
4299 | return l; | |
4300 | return NULL; | |
4301 | } | |
4302 | ||
72a8cb30 BB |
4303 | /* |
4304 | * find the appropriate pidlist for our purpose (given procs vs tasks) | |
4305 | * returns with the lock on that pidlist already held, and takes care | |
4306 | * of the use count, or returns NULL with no locks held if we're out of | |
4307 | * memory. | |
4308 | */ | |
e6b81710 TH |
4309 | static struct cgroup_pidlist *cgroup_pidlist_find_create(struct cgroup *cgrp, |
4310 | enum cgroup_filetype type) | |
72a8cb30 BB |
4311 | { |
4312 | struct cgroup_pidlist *l; | |
b70cc5fd | 4313 | |
e6b81710 TH |
4314 | lockdep_assert_held(&cgrp->pidlist_mutex); |
4315 | ||
4316 | l = cgroup_pidlist_find(cgrp, type); | |
4317 | if (l) | |
4318 | return l; | |
4319 | ||
72a8cb30 | 4320 | /* entry not found; create a new one */ |
f4f4be2b | 4321 | l = kzalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL); |
e6b81710 | 4322 | if (!l) |
72a8cb30 | 4323 | return l; |
e6b81710 | 4324 | |
b1a21367 | 4325 | INIT_DELAYED_WORK(&l->destroy_dwork, cgroup_pidlist_destroy_work_fn); |
72a8cb30 | 4326 | l->key.type = type; |
e6b81710 TH |
4327 | /* don't need task_nsproxy() if we're looking at ourself */ |
4328 | l->key.ns = get_pid_ns(task_active_pid_ns(current)); | |
72a8cb30 BB |
4329 | l->owner = cgrp; |
4330 | list_add(&l->links, &cgrp->pidlists); | |
72a8cb30 BB |
4331 | return l; |
4332 | } | |
4333 | ||
102a775e BB |
4334 | /* |
4335 | * Load a cgroup's pidarray with either procs' tgids or tasks' pids | |
4336 | */ | |
72a8cb30 BB |
4337 | static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type, |
4338 | struct cgroup_pidlist **lp) | |
102a775e BB |
4339 | { |
4340 | pid_t *array; | |
4341 | int length; | |
4342 | int pid, n = 0; /* used for populating the array */ | |
72ec7029 | 4343 | struct css_task_iter it; |
817929ec | 4344 | struct task_struct *tsk; |
102a775e BB |
4345 | struct cgroup_pidlist *l; |
4346 | ||
4bac00d1 TH |
4347 | lockdep_assert_held(&cgrp->pidlist_mutex); |
4348 | ||
102a775e BB |
4349 | /* |
4350 | * If cgroup gets more users after we read count, we won't have | |
4351 | * enough space - tough. This race is indistinguishable to the | |
4352 | * caller from the case that the additional cgroup users didn't | |
4353 | * show up until sometime later on. | |
4354 | */ | |
4355 | length = cgroup_task_count(cgrp); | |
d1d9fd33 | 4356 | array = pidlist_allocate(length); |
102a775e BB |
4357 | if (!array) |
4358 | return -ENOMEM; | |
4359 | /* now, populate the array */ | |
9d800df1 | 4360 | css_task_iter_start(&cgrp->self, &it); |
72ec7029 | 4361 | while ((tsk = css_task_iter_next(&it))) { |
102a775e | 4362 | if (unlikely(n == length)) |
817929ec | 4363 | break; |
102a775e | 4364 | /* get tgid or pid for procs or tasks file respectively */ |
72a8cb30 BB |
4365 | if (type == CGROUP_FILE_PROCS) |
4366 | pid = task_tgid_vnr(tsk); | |
4367 | else | |
4368 | pid = task_pid_vnr(tsk); | |
102a775e BB |
4369 | if (pid > 0) /* make sure to only use valid results */ |
4370 | array[n++] = pid; | |
817929ec | 4371 | } |
72ec7029 | 4372 | css_task_iter_end(&it); |
102a775e BB |
4373 | length = n; |
4374 | /* now sort & (if procs) strip out duplicates */ | |
aa6ec29b | 4375 | if (cgroup_on_dfl(cgrp)) |
afb2bc14 TH |
4376 | sort(array, length, sizeof(pid_t), fried_cmppid, NULL); |
4377 | else | |
4378 | sort(array, length, sizeof(pid_t), cmppid, NULL); | |
72a8cb30 | 4379 | if (type == CGROUP_FILE_PROCS) |
6ee211ad | 4380 | length = pidlist_uniq(array, length); |
e6b81710 | 4381 | |
e6b81710 | 4382 | l = cgroup_pidlist_find_create(cgrp, type); |
72a8cb30 | 4383 | if (!l) { |
d1d9fd33 | 4384 | pidlist_free(array); |
72a8cb30 | 4385 | return -ENOMEM; |
102a775e | 4386 | } |
e6b81710 TH |
4387 | |
4388 | /* store array, freeing old if necessary */ | |
d1d9fd33 | 4389 | pidlist_free(l->list); |
102a775e BB |
4390 | l->list = array; |
4391 | l->length = length; | |
72a8cb30 | 4392 | *lp = l; |
102a775e | 4393 | return 0; |
bbcb81d0 PM |
4394 | } |
4395 | ||
846c7bb0 | 4396 | /** |
a043e3b2 | 4397 | * cgroupstats_build - build and fill cgroupstats |
846c7bb0 BS |
4398 | * @stats: cgroupstats to fill information into |
4399 | * @dentry: A dentry entry belonging to the cgroup for which stats have | |
4400 | * been requested. | |
a043e3b2 LZ |
4401 | * |
4402 | * Build and fill cgroupstats so that taskstats can export it to user | |
4403 | * space. | |
846c7bb0 BS |
4404 | */ |
4405 | int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry) | |
4406 | { | |
2bd59d48 | 4407 | struct kernfs_node *kn = kernfs_node_from_dentry(dentry); |
bd89aabc | 4408 | struct cgroup *cgrp; |
72ec7029 | 4409 | struct css_task_iter it; |
846c7bb0 | 4410 | struct task_struct *tsk; |
33d283be | 4411 | |
2bd59d48 TH |
4412 | /* it should be kernfs_node belonging to cgroupfs and is a directory */ |
4413 | if (dentry->d_sb->s_type != &cgroup_fs_type || !kn || | |
4414 | kernfs_type(kn) != KERNFS_DIR) | |
4415 | return -EINVAL; | |
4416 | ||
bad34660 LZ |
4417 | mutex_lock(&cgroup_mutex); |
4418 | ||
846c7bb0 | 4419 | /* |
2bd59d48 | 4420 | * We aren't being called from kernfs and there's no guarantee on |
ec903c0c | 4421 | * @kn->priv's validity. For this and css_tryget_online_from_dir(), |
2bd59d48 | 4422 | * @kn->priv is RCU safe. Let's do the RCU dancing. |
846c7bb0 | 4423 | */ |
2bd59d48 TH |
4424 | rcu_read_lock(); |
4425 | cgrp = rcu_dereference(kn->priv); | |
bad34660 | 4426 | if (!cgrp || cgroup_is_dead(cgrp)) { |
2bd59d48 | 4427 | rcu_read_unlock(); |
bad34660 | 4428 | mutex_unlock(&cgroup_mutex); |
2bd59d48 TH |
4429 | return -ENOENT; |
4430 | } | |
bad34660 | 4431 | rcu_read_unlock(); |
846c7bb0 | 4432 | |
9d800df1 | 4433 | css_task_iter_start(&cgrp->self, &it); |
72ec7029 | 4434 | while ((tsk = css_task_iter_next(&it))) { |
846c7bb0 BS |
4435 | switch (tsk->state) { |
4436 | case TASK_RUNNING: | |
4437 | stats->nr_running++; | |
4438 | break; | |
4439 | case TASK_INTERRUPTIBLE: | |
4440 | stats->nr_sleeping++; | |
4441 | break; | |
4442 | case TASK_UNINTERRUPTIBLE: | |
4443 | stats->nr_uninterruptible++; | |
4444 | break; | |
4445 | case TASK_STOPPED: | |
4446 | stats->nr_stopped++; | |
4447 | break; | |
4448 | default: | |
4449 | if (delayacct_is_task_waiting_on_io(tsk)) | |
4450 | stats->nr_io_wait++; | |
4451 | break; | |
4452 | } | |
4453 | } | |
72ec7029 | 4454 | css_task_iter_end(&it); |
846c7bb0 | 4455 | |
bad34660 | 4456 | mutex_unlock(&cgroup_mutex); |
2bd59d48 | 4457 | return 0; |
846c7bb0 BS |
4458 | } |
4459 | ||
8f3ff208 | 4460 | |
bbcb81d0 | 4461 | /* |
102a775e | 4462 | * seq_file methods for the tasks/procs files. The seq_file position is the |
cc31edce | 4463 | * next pid to display; the seq_file iterator is a pointer to the pid |
102a775e | 4464 | * in the cgroup->l->list array. |
bbcb81d0 | 4465 | */ |
cc31edce | 4466 | |
102a775e | 4467 | static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos) |
bbcb81d0 | 4468 | { |
cc31edce PM |
4469 | /* |
4470 | * Initially we receive a position value that corresponds to | |
4471 | * one more than the last pid shown (or 0 on the first call or | |
4472 | * after a seek to the start). Use a binary-search to find the | |
4473 | * next pid to display, if any | |
4474 | */ | |
2bd59d48 | 4475 | struct kernfs_open_file *of = s->private; |
7da11279 | 4476 | struct cgroup *cgrp = seq_css(s)->cgroup; |
4bac00d1 | 4477 | struct cgroup_pidlist *l; |
7da11279 | 4478 | enum cgroup_filetype type = seq_cft(s)->private; |
cc31edce | 4479 | int index = 0, pid = *pos; |
4bac00d1 TH |
4480 | int *iter, ret; |
4481 | ||
4482 | mutex_lock(&cgrp->pidlist_mutex); | |
4483 | ||
4484 | /* | |
5d22444f | 4485 | * !NULL @of->priv indicates that this isn't the first start() |
4bac00d1 | 4486 | * after open. If the matching pidlist is around, we can use that. |
5d22444f | 4487 | * Look for it. Note that @of->priv can't be used directly. It |
4bac00d1 TH |
4488 | * could already have been destroyed. |
4489 | */ | |
5d22444f TH |
4490 | if (of->priv) |
4491 | of->priv = cgroup_pidlist_find(cgrp, type); | |
4bac00d1 TH |
4492 | |
4493 | /* | |
4494 | * Either this is the first start() after open or the matching | |
4495 | * pidlist has been destroyed inbetween. Create a new one. | |
4496 | */ | |
5d22444f TH |
4497 | if (!of->priv) { |
4498 | ret = pidlist_array_load(cgrp, type, | |
4499 | (struct cgroup_pidlist **)&of->priv); | |
4bac00d1 TH |
4500 | if (ret) |
4501 | return ERR_PTR(ret); | |
4502 | } | |
5d22444f | 4503 | l = of->priv; |
cc31edce | 4504 | |
cc31edce | 4505 | if (pid) { |
102a775e | 4506 | int end = l->length; |
20777766 | 4507 | |
cc31edce PM |
4508 | while (index < end) { |
4509 | int mid = (index + end) / 2; | |
afb2bc14 | 4510 | if (cgroup_pid_fry(cgrp, l->list[mid]) == pid) { |
cc31edce PM |
4511 | index = mid; |
4512 | break; | |
afb2bc14 | 4513 | } else if (cgroup_pid_fry(cgrp, l->list[mid]) <= pid) |
cc31edce PM |
4514 | index = mid + 1; |
4515 | else | |
4516 | end = mid; | |
4517 | } | |
4518 | } | |
4519 | /* If we're off the end of the array, we're done */ | |
102a775e | 4520 | if (index >= l->length) |
cc31edce PM |
4521 | return NULL; |
4522 | /* Update the abstract position to be the actual pid that we found */ | |
102a775e | 4523 | iter = l->list + index; |
afb2bc14 | 4524 | *pos = cgroup_pid_fry(cgrp, *iter); |
cc31edce PM |
4525 | return iter; |
4526 | } | |
4527 | ||
102a775e | 4528 | static void cgroup_pidlist_stop(struct seq_file *s, void *v) |
cc31edce | 4529 | { |
2bd59d48 | 4530 | struct kernfs_open_file *of = s->private; |
5d22444f | 4531 | struct cgroup_pidlist *l = of->priv; |
62236858 | 4532 | |
5d22444f TH |
4533 | if (l) |
4534 | mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, | |
04502365 | 4535 | CGROUP_PIDLIST_DESTROY_DELAY); |
7da11279 | 4536 | mutex_unlock(&seq_css(s)->cgroup->pidlist_mutex); |
cc31edce PM |
4537 | } |
4538 | ||
102a775e | 4539 | static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos) |
cc31edce | 4540 | { |
2bd59d48 | 4541 | struct kernfs_open_file *of = s->private; |
5d22444f | 4542 | struct cgroup_pidlist *l = of->priv; |
102a775e BB |
4543 | pid_t *p = v; |
4544 | pid_t *end = l->list + l->length; | |
cc31edce PM |
4545 | /* |
4546 | * Advance to the next pid in the array. If this goes off the | |
4547 | * end, we're done | |
4548 | */ | |
4549 | p++; | |
4550 | if (p >= end) { | |
4551 | return NULL; | |
4552 | } else { | |
7da11279 | 4553 | *pos = cgroup_pid_fry(seq_css(s)->cgroup, *p); |
cc31edce PM |
4554 | return p; |
4555 | } | |
4556 | } | |
4557 | ||
102a775e | 4558 | static int cgroup_pidlist_show(struct seq_file *s, void *v) |
cc31edce | 4559 | { |
94ff212d JP |
4560 | seq_printf(s, "%d\n", *(int *)v); |
4561 | ||
4562 | return 0; | |
cc31edce | 4563 | } |
bbcb81d0 | 4564 | |
182446d0 TH |
4565 | static u64 cgroup_read_notify_on_release(struct cgroup_subsys_state *css, |
4566 | struct cftype *cft) | |
81a6a5cd | 4567 | { |
182446d0 | 4568 | return notify_on_release(css->cgroup); |
81a6a5cd PM |
4569 | } |
4570 | ||
182446d0 TH |
4571 | static int cgroup_write_notify_on_release(struct cgroup_subsys_state *css, |
4572 | struct cftype *cft, u64 val) | |
6379c106 | 4573 | { |
6379c106 | 4574 | if (val) |
182446d0 | 4575 | set_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags); |
6379c106 | 4576 | else |
182446d0 | 4577 | clear_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags); |
6379c106 PM |
4578 | return 0; |
4579 | } | |
4580 | ||
182446d0 TH |
4581 | static u64 cgroup_clone_children_read(struct cgroup_subsys_state *css, |
4582 | struct cftype *cft) | |
97978e6d | 4583 | { |
182446d0 | 4584 | return test_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d DL |
4585 | } |
4586 | ||
182446d0 TH |
4587 | static int cgroup_clone_children_write(struct cgroup_subsys_state *css, |
4588 | struct cftype *cft, u64 val) | |
97978e6d DL |
4589 | { |
4590 | if (val) | |
182446d0 | 4591 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d | 4592 | else |
182446d0 | 4593 | clear_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d DL |
4594 | return 0; |
4595 | } | |
4596 | ||
a14c6874 TH |
4597 | /* cgroup core interface files for the default hierarchy */ |
4598 | static struct cftype cgroup_dfl_base_files[] = { | |
81a6a5cd | 4599 | { |
d5c56ced | 4600 | .name = "cgroup.procs", |
6f60eade | 4601 | .file_offset = offsetof(struct cgroup, procs_file), |
6612f05b TH |
4602 | .seq_start = cgroup_pidlist_start, |
4603 | .seq_next = cgroup_pidlist_next, | |
4604 | .seq_stop = cgroup_pidlist_stop, | |
4605 | .seq_show = cgroup_pidlist_show, | |
5d22444f | 4606 | .private = CGROUP_FILE_PROCS, |
acbef755 | 4607 | .write = cgroup_procs_write, |
102a775e | 4608 | }, |
f8f22e53 TH |
4609 | { |
4610 | .name = "cgroup.controllers", | |
a14c6874 | 4611 | .flags = CFTYPE_ONLY_ON_ROOT, |
f8f22e53 TH |
4612 | .seq_show = cgroup_root_controllers_show, |
4613 | }, | |
4614 | { | |
4615 | .name = "cgroup.controllers", | |
a14c6874 | 4616 | .flags = CFTYPE_NOT_ON_ROOT, |
f8f22e53 TH |
4617 | .seq_show = cgroup_controllers_show, |
4618 | }, | |
4619 | { | |
4620 | .name = "cgroup.subtree_control", | |
f8f22e53 | 4621 | .seq_show = cgroup_subtree_control_show, |
451af504 | 4622 | .write = cgroup_subtree_control_write, |
f8f22e53 | 4623 | }, |
842b597e | 4624 | { |
4a07c222 | 4625 | .name = "cgroup.events", |
a14c6874 | 4626 | .flags = CFTYPE_NOT_ON_ROOT, |
6f60eade | 4627 | .file_offset = offsetof(struct cgroup, events_file), |
4a07c222 | 4628 | .seq_show = cgroup_events_show, |
842b597e | 4629 | }, |
a14c6874 TH |
4630 | { } /* terminate */ |
4631 | }; | |
d5c56ced | 4632 | |
a14c6874 TH |
4633 | /* cgroup core interface files for the legacy hierarchies */ |
4634 | static struct cftype cgroup_legacy_base_files[] = { | |
4635 | { | |
4636 | .name = "cgroup.procs", | |
4637 | .seq_start = cgroup_pidlist_start, | |
4638 | .seq_next = cgroup_pidlist_next, | |
4639 | .seq_stop = cgroup_pidlist_stop, | |
4640 | .seq_show = cgroup_pidlist_show, | |
4641 | .private = CGROUP_FILE_PROCS, | |
4642 | .write = cgroup_procs_write, | |
a14c6874 TH |
4643 | }, |
4644 | { | |
4645 | .name = "cgroup.clone_children", | |
4646 | .read_u64 = cgroup_clone_children_read, | |
4647 | .write_u64 = cgroup_clone_children_write, | |
4648 | }, | |
4649 | { | |
4650 | .name = "cgroup.sane_behavior", | |
4651 | .flags = CFTYPE_ONLY_ON_ROOT, | |
4652 | .seq_show = cgroup_sane_behavior_show, | |
4653 | }, | |
d5c56ced TH |
4654 | { |
4655 | .name = "tasks", | |
6612f05b TH |
4656 | .seq_start = cgroup_pidlist_start, |
4657 | .seq_next = cgroup_pidlist_next, | |
4658 | .seq_stop = cgroup_pidlist_stop, | |
4659 | .seq_show = cgroup_pidlist_show, | |
5d22444f | 4660 | .private = CGROUP_FILE_TASKS, |
acbef755 | 4661 | .write = cgroup_tasks_write, |
d5c56ced TH |
4662 | }, |
4663 | { | |
4664 | .name = "notify_on_release", | |
d5c56ced TH |
4665 | .read_u64 = cgroup_read_notify_on_release, |
4666 | .write_u64 = cgroup_write_notify_on_release, | |
4667 | }, | |
6e6ff25b TH |
4668 | { |
4669 | .name = "release_agent", | |
a14c6874 | 4670 | .flags = CFTYPE_ONLY_ON_ROOT, |
2da8ca82 | 4671 | .seq_show = cgroup_release_agent_show, |
451af504 | 4672 | .write = cgroup_release_agent_write, |
5f469907 | 4673 | .max_write_len = PATH_MAX - 1, |
6e6ff25b | 4674 | }, |
db0416b6 | 4675 | { } /* terminate */ |
bbcb81d0 PM |
4676 | }; |
4677 | ||
0c21ead1 TH |
4678 | /* |
4679 | * css destruction is four-stage process. | |
4680 | * | |
4681 | * 1. Destruction starts. Killing of the percpu_ref is initiated. | |
4682 | * Implemented in kill_css(). | |
4683 | * | |
4684 | * 2. When the percpu_ref is confirmed to be visible as killed on all CPUs | |
ec903c0c TH |
4685 | * and thus css_tryget_online() is guaranteed to fail, the css can be |
4686 | * offlined by invoking offline_css(). After offlining, the base ref is | |
4687 | * put. Implemented in css_killed_work_fn(). | |
0c21ead1 TH |
4688 | * |
4689 | * 3. When the percpu_ref reaches zero, the only possible remaining | |
4690 | * accessors are inside RCU read sections. css_release() schedules the | |
4691 | * RCU callback. | |
4692 | * | |
4693 | * 4. After the grace period, the css can be freed. Implemented in | |
4694 | * css_free_work_fn(). | |
4695 | * | |
4696 | * It is actually hairier because both step 2 and 4 require process context | |
4697 | * and thus involve punting to css->destroy_work adding two additional | |
4698 | * steps to the already complex sequence. | |
4699 | */ | |
35ef10da | 4700 | static void css_free_work_fn(struct work_struct *work) |
48ddbe19 TH |
4701 | { |
4702 | struct cgroup_subsys_state *css = | |
35ef10da | 4703 | container_of(work, struct cgroup_subsys_state, destroy_work); |
01e58659 | 4704 | struct cgroup_subsys *ss = css->ss; |
0c21ead1 | 4705 | struct cgroup *cgrp = css->cgroup; |
48ddbe19 | 4706 | |
9a1049da TH |
4707 | percpu_ref_exit(&css->refcnt); |
4708 | ||
01e58659 | 4709 | if (ss) { |
9d755d33 | 4710 | /* css free path */ |
8bb5ef79 | 4711 | struct cgroup_subsys_state *parent = css->parent; |
01e58659 VD |
4712 | int id = css->id; |
4713 | ||
01e58659 VD |
4714 | ss->css_free(css); |
4715 | cgroup_idr_remove(&ss->css_idr, id); | |
9d755d33 | 4716 | cgroup_put(cgrp); |
8bb5ef79 TH |
4717 | |
4718 | if (parent) | |
4719 | css_put(parent); | |
9d755d33 TH |
4720 | } else { |
4721 | /* cgroup free path */ | |
4722 | atomic_dec(&cgrp->root->nr_cgrps); | |
4723 | cgroup_pidlist_destroy_all(cgrp); | |
971ff493 | 4724 | cancel_work_sync(&cgrp->release_agent_work); |
9d755d33 | 4725 | |
d51f39b0 | 4726 | if (cgroup_parent(cgrp)) { |
9d755d33 TH |
4727 | /* |
4728 | * We get a ref to the parent, and put the ref when | |
4729 | * this cgroup is being freed, so it's guaranteed | |
4730 | * that the parent won't be destroyed before its | |
4731 | * children. | |
4732 | */ | |
d51f39b0 | 4733 | cgroup_put(cgroup_parent(cgrp)); |
9d755d33 TH |
4734 | kernfs_put(cgrp->kn); |
4735 | kfree(cgrp); | |
4736 | } else { | |
4737 | /* | |
4738 | * This is root cgroup's refcnt reaching zero, | |
4739 | * which indicates that the root should be | |
4740 | * released. | |
4741 | */ | |
4742 | cgroup_destroy_root(cgrp->root); | |
4743 | } | |
4744 | } | |
48ddbe19 TH |
4745 | } |
4746 | ||
0c21ead1 | 4747 | static void css_free_rcu_fn(struct rcu_head *rcu_head) |
d3daf28d TH |
4748 | { |
4749 | struct cgroup_subsys_state *css = | |
0c21ead1 | 4750 | container_of(rcu_head, struct cgroup_subsys_state, rcu_head); |
d3daf28d | 4751 | |
35ef10da | 4752 | INIT_WORK(&css->destroy_work, css_free_work_fn); |
e5fca243 | 4753 | queue_work(cgroup_destroy_wq, &css->destroy_work); |
48ddbe19 TH |
4754 | } |
4755 | ||
25e15d83 | 4756 | static void css_release_work_fn(struct work_struct *work) |
d3daf28d TH |
4757 | { |
4758 | struct cgroup_subsys_state *css = | |
25e15d83 | 4759 | container_of(work, struct cgroup_subsys_state, destroy_work); |
15a4c835 | 4760 | struct cgroup_subsys *ss = css->ss; |
9d755d33 | 4761 | struct cgroup *cgrp = css->cgroup; |
15a4c835 | 4762 | |
1fed1b2e TH |
4763 | mutex_lock(&cgroup_mutex); |
4764 | ||
de3f0341 | 4765 | css->flags |= CSS_RELEASED; |
1fed1b2e TH |
4766 | list_del_rcu(&css->sibling); |
4767 | ||
9d755d33 TH |
4768 | if (ss) { |
4769 | /* css release path */ | |
01e58659 | 4770 | cgroup_idr_replace(&ss->css_idr, NULL, css->id); |
7d172cc8 TH |
4771 | if (ss->css_released) |
4772 | ss->css_released(css); | |
9d755d33 TH |
4773 | } else { |
4774 | /* cgroup release path */ | |
9d755d33 TH |
4775 | cgroup_idr_remove(&cgrp->root->cgroup_idr, cgrp->id); |
4776 | cgrp->id = -1; | |
a4189487 LZ |
4777 | |
4778 | /* | |
4779 | * There are two control paths which try to determine | |
4780 | * cgroup from dentry without going through kernfs - | |
4781 | * cgroupstats_build() and css_tryget_online_from_dir(). | |
4782 | * Those are supported by RCU protecting clearing of | |
4783 | * cgrp->kn->priv backpointer. | |
4784 | */ | |
4785 | RCU_INIT_POINTER(*(void __rcu __force **)&cgrp->kn->priv, NULL); | |
9d755d33 | 4786 | } |
d3daf28d | 4787 | |
1fed1b2e TH |
4788 | mutex_unlock(&cgroup_mutex); |
4789 | ||
0c21ead1 | 4790 | call_rcu(&css->rcu_head, css_free_rcu_fn); |
d3daf28d TH |
4791 | } |
4792 | ||
d3daf28d TH |
4793 | static void css_release(struct percpu_ref *ref) |
4794 | { | |
4795 | struct cgroup_subsys_state *css = | |
4796 | container_of(ref, struct cgroup_subsys_state, refcnt); | |
4797 | ||
25e15d83 TH |
4798 | INIT_WORK(&css->destroy_work, css_release_work_fn); |
4799 | queue_work(cgroup_destroy_wq, &css->destroy_work); | |
d3daf28d TH |
4800 | } |
4801 | ||
ddfcadab TH |
4802 | static void init_and_link_css(struct cgroup_subsys_state *css, |
4803 | struct cgroup_subsys *ss, struct cgroup *cgrp) | |
ddbcc7e8 | 4804 | { |
0cb51d71 TH |
4805 | lockdep_assert_held(&cgroup_mutex); |
4806 | ||
ddfcadab TH |
4807 | cgroup_get(cgrp); |
4808 | ||
d5c419b6 | 4809 | memset(css, 0, sizeof(*css)); |
bd89aabc | 4810 | css->cgroup = cgrp; |
72c97e54 | 4811 | css->ss = ss; |
d5c419b6 TH |
4812 | INIT_LIST_HEAD(&css->sibling); |
4813 | INIT_LIST_HEAD(&css->children); | |
0cb51d71 | 4814 | css->serial_nr = css_serial_nr_next++; |
aa226ff4 | 4815 | atomic_set(&css->online_cnt, 0); |
0ae78e0b | 4816 | |
d51f39b0 TH |
4817 | if (cgroup_parent(cgrp)) { |
4818 | css->parent = cgroup_css(cgroup_parent(cgrp), ss); | |
ddfcadab | 4819 | css_get(css->parent); |
ddfcadab | 4820 | } |
48ddbe19 | 4821 | |
ca8bdcaf | 4822 | BUG_ON(cgroup_css(cgrp, ss)); |
ddbcc7e8 PM |
4823 | } |
4824 | ||
2a4ac633 | 4825 | /* invoke ->css_online() on a new CSS and mark it online if successful */ |
623f926b | 4826 | static int online_css(struct cgroup_subsys_state *css) |
a31f2d3f | 4827 | { |
623f926b | 4828 | struct cgroup_subsys *ss = css->ss; |
b1929db4 TH |
4829 | int ret = 0; |
4830 | ||
a31f2d3f TH |
4831 | lockdep_assert_held(&cgroup_mutex); |
4832 | ||
92fb9748 | 4833 | if (ss->css_online) |
eb95419b | 4834 | ret = ss->css_online(css); |
ae7f164a | 4835 | if (!ret) { |
eb95419b | 4836 | css->flags |= CSS_ONLINE; |
aec25020 | 4837 | rcu_assign_pointer(css->cgroup->subsys[ss->id], css); |
aa226ff4 TH |
4838 | |
4839 | atomic_inc(&css->online_cnt); | |
4840 | if (css->parent) | |
4841 | atomic_inc(&css->parent->online_cnt); | |
ae7f164a | 4842 | } |
b1929db4 | 4843 | return ret; |
a31f2d3f TH |
4844 | } |
4845 | ||
2a4ac633 | 4846 | /* if the CSS is online, invoke ->css_offline() on it and mark it offline */ |
623f926b | 4847 | static void offline_css(struct cgroup_subsys_state *css) |
a31f2d3f | 4848 | { |
623f926b | 4849 | struct cgroup_subsys *ss = css->ss; |
a31f2d3f TH |
4850 | |
4851 | lockdep_assert_held(&cgroup_mutex); | |
4852 | ||
4853 | if (!(css->flags & CSS_ONLINE)) | |
4854 | return; | |
4855 | ||
d7eeac19 | 4856 | if (ss->css_offline) |
eb95419b | 4857 | ss->css_offline(css); |
a31f2d3f | 4858 | |
eb95419b | 4859 | css->flags &= ~CSS_ONLINE; |
e3297803 | 4860 | RCU_INIT_POINTER(css->cgroup->subsys[ss->id], NULL); |
f8f22e53 TH |
4861 | |
4862 | wake_up_all(&css->cgroup->offline_waitq); | |
a31f2d3f TH |
4863 | } |
4864 | ||
c81c925a TH |
4865 | /** |
4866 | * create_css - create a cgroup_subsys_state | |
4867 | * @cgrp: the cgroup new css will be associated with | |
4868 | * @ss: the subsys of new css | |
f63070d3 | 4869 | * @visible: whether to create control knobs for the new css or not |
c81c925a TH |
4870 | * |
4871 | * Create a new css associated with @cgrp - @ss pair. On success, the new | |
f63070d3 TH |
4872 | * css is online and installed in @cgrp with all interface files created if |
4873 | * @visible. Returns 0 on success, -errno on failure. | |
c81c925a | 4874 | */ |
f63070d3 TH |
4875 | static int create_css(struct cgroup *cgrp, struct cgroup_subsys *ss, |
4876 | bool visible) | |
c81c925a | 4877 | { |
d51f39b0 | 4878 | struct cgroup *parent = cgroup_parent(cgrp); |
1fed1b2e | 4879 | struct cgroup_subsys_state *parent_css = cgroup_css(parent, ss); |
c81c925a TH |
4880 | struct cgroup_subsys_state *css; |
4881 | int err; | |
4882 | ||
c81c925a TH |
4883 | lockdep_assert_held(&cgroup_mutex); |
4884 | ||
1fed1b2e | 4885 | css = ss->css_alloc(parent_css); |
c81c925a TH |
4886 | if (IS_ERR(css)) |
4887 | return PTR_ERR(css); | |
4888 | ||
ddfcadab | 4889 | init_and_link_css(css, ss, cgrp); |
a2bed820 | 4890 | |
2aad2a86 | 4891 | err = percpu_ref_init(&css->refcnt, css_release, 0, GFP_KERNEL); |
c81c925a | 4892 | if (err) |
3eb59ec6 | 4893 | goto err_free_css; |
c81c925a | 4894 | |
cf780b7d | 4895 | err = cgroup_idr_alloc(&ss->css_idr, NULL, 2, 0, GFP_KERNEL); |
15a4c835 TH |
4896 | if (err < 0) |
4897 | goto err_free_percpu_ref; | |
4898 | css->id = err; | |
c81c925a | 4899 | |
f63070d3 | 4900 | if (visible) { |
4df8dc90 | 4901 | err = css_populate_dir(css, NULL); |
f63070d3 TH |
4902 | if (err) |
4903 | goto err_free_id; | |
4904 | } | |
15a4c835 TH |
4905 | |
4906 | /* @css is ready to be brought online now, make it visible */ | |
1fed1b2e | 4907 | list_add_tail_rcu(&css->sibling, &parent_css->children); |
15a4c835 | 4908 | cgroup_idr_replace(&ss->css_idr, css, css->id); |
c81c925a TH |
4909 | |
4910 | err = online_css(css); | |
4911 | if (err) | |
1fed1b2e | 4912 | goto err_list_del; |
94419627 | 4913 | |
c81c925a | 4914 | if (ss->broken_hierarchy && !ss->warned_broken_hierarchy && |
d51f39b0 | 4915 | cgroup_parent(parent)) { |
ed3d261b | 4916 | 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 | 4917 | current->comm, current->pid, ss->name); |
c81c925a | 4918 | if (!strcmp(ss->name, "memory")) |
ed3d261b | 4919 | pr_warn("\"memory\" requires setting use_hierarchy to 1 on the root\n"); |
c81c925a TH |
4920 | ss->warned_broken_hierarchy = true; |
4921 | } | |
4922 | ||
4923 | return 0; | |
4924 | ||
1fed1b2e TH |
4925 | err_list_del: |
4926 | list_del_rcu(&css->sibling); | |
4df8dc90 | 4927 | css_clear_dir(css, NULL); |
15a4c835 TH |
4928 | err_free_id: |
4929 | cgroup_idr_remove(&ss->css_idr, css->id); | |
3eb59ec6 | 4930 | err_free_percpu_ref: |
9a1049da | 4931 | percpu_ref_exit(&css->refcnt); |
3eb59ec6 | 4932 | err_free_css: |
a2bed820 | 4933 | call_rcu(&css->rcu_head, css_free_rcu_fn); |
c81c925a TH |
4934 | return err; |
4935 | } | |
4936 | ||
b3bfd983 TH |
4937 | static int cgroup_mkdir(struct kernfs_node *parent_kn, const char *name, |
4938 | umode_t mode) | |
ddbcc7e8 | 4939 | { |
b11cfb58 | 4940 | struct cgroup *parent, *cgrp, *tcgrp; |
a9746d8d | 4941 | struct cgroup_root *root; |
ddbcc7e8 | 4942 | struct cgroup_subsys *ss; |
2bd59d48 | 4943 | struct kernfs_node *kn; |
b11cfb58 | 4944 | int level, ssid, ret; |
ddbcc7e8 | 4945 | |
71b1fb5c AC |
4946 | /* Do not accept '\n' to prevent making /proc/<pid>/cgroup unparsable. |
4947 | */ | |
4948 | if (strchr(name, '\n')) | |
4949 | return -EINVAL; | |
4950 | ||
a9746d8d TH |
4951 | parent = cgroup_kn_lock_live(parent_kn); |
4952 | if (!parent) | |
4953 | return -ENODEV; | |
4954 | root = parent->root; | |
b11cfb58 | 4955 | level = parent->level + 1; |
ddbcc7e8 | 4956 | |
0a950f65 | 4957 | /* allocate the cgroup and its ID, 0 is reserved for the root */ |
b11cfb58 TH |
4958 | cgrp = kzalloc(sizeof(*cgrp) + |
4959 | sizeof(cgrp->ancestor_ids[0]) * (level + 1), GFP_KERNEL); | |
ba0f4d76 TH |
4960 | if (!cgrp) { |
4961 | ret = -ENOMEM; | |
4962 | goto out_unlock; | |
0ab02ca8 LZ |
4963 | } |
4964 | ||
2aad2a86 | 4965 | ret = percpu_ref_init(&cgrp->self.refcnt, css_release, 0, GFP_KERNEL); |
9d755d33 TH |
4966 | if (ret) |
4967 | goto out_free_cgrp; | |
4968 | ||
0ab02ca8 LZ |
4969 | /* |
4970 | * Temporarily set the pointer to NULL, so idr_find() won't return | |
4971 | * a half-baked cgroup. | |
4972 | */ | |
cf780b7d | 4973 | cgrp->id = cgroup_idr_alloc(&root->cgroup_idr, NULL, 2, 0, GFP_KERNEL); |
0ab02ca8 | 4974 | if (cgrp->id < 0) { |
ba0f4d76 | 4975 | ret = -ENOMEM; |
9d755d33 | 4976 | goto out_cancel_ref; |
976c06bc TH |
4977 | } |
4978 | ||
cc31edce | 4979 | init_cgroup_housekeeping(cgrp); |
ddbcc7e8 | 4980 | |
9d800df1 | 4981 | cgrp->self.parent = &parent->self; |
ba0f4d76 | 4982 | cgrp->root = root; |
b11cfb58 TH |
4983 | cgrp->level = level; |
4984 | ||
4985 | for (tcgrp = cgrp; tcgrp; tcgrp = cgroup_parent(tcgrp)) | |
4986 | cgrp->ancestor_ids[tcgrp->level] = tcgrp->id; | |
ddbcc7e8 | 4987 | |
b6abdb0e LZ |
4988 | if (notify_on_release(parent)) |
4989 | set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); | |
4990 | ||
2260e7fc TH |
4991 | if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &parent->flags)) |
4992 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags); | |
97978e6d | 4993 | |
2bd59d48 | 4994 | /* create the directory */ |
e61734c5 | 4995 | kn = kernfs_create_dir(parent->kn, name, mode, cgrp); |
2bd59d48 | 4996 | if (IS_ERR(kn)) { |
ba0f4d76 TH |
4997 | ret = PTR_ERR(kn); |
4998 | goto out_free_id; | |
2bd59d48 TH |
4999 | } |
5000 | cgrp->kn = kn; | |
ddbcc7e8 | 5001 | |
4e139afc | 5002 | /* |
6f30558f TH |
5003 | * This extra ref will be put in cgroup_free_fn() and guarantees |
5004 | * that @cgrp->kn is always accessible. | |
4e139afc | 5005 | */ |
6f30558f | 5006 | kernfs_get(kn); |
ddbcc7e8 | 5007 | |
0cb51d71 | 5008 | cgrp->self.serial_nr = css_serial_nr_next++; |
53fa5261 | 5009 | |
4e139afc | 5010 | /* allocation complete, commit to creation */ |
d5c419b6 | 5011 | list_add_tail_rcu(&cgrp->self.sibling, &cgroup_parent(cgrp)->self.children); |
3c9c825b | 5012 | atomic_inc(&root->nr_cgrps); |
59f5296b | 5013 | cgroup_get(parent); |
415cf07a | 5014 | |
0d80255e TH |
5015 | /* |
5016 | * @cgrp is now fully operational. If something fails after this | |
5017 | * point, it'll be released via the normal destruction path. | |
5018 | */ | |
6fa4918d | 5019 | cgroup_idr_replace(&root->cgroup_idr, cgrp, cgrp->id); |
4e96ee8e | 5020 | |
ba0f4d76 TH |
5021 | ret = cgroup_kn_set_ugid(kn); |
5022 | if (ret) | |
5023 | goto out_destroy; | |
49957f8e | 5024 | |
4df8dc90 | 5025 | ret = css_populate_dir(&cgrp->self, NULL); |
ba0f4d76 TH |
5026 | if (ret) |
5027 | goto out_destroy; | |
628f7cd4 | 5028 | |
9d403e99 | 5029 | /* let's create and online css's */ |
b85d2040 | 5030 | for_each_subsys(ss, ssid) { |
f392e51c | 5031 | if (parent->child_subsys_mask & (1 << ssid)) { |
f63070d3 TH |
5032 | ret = create_css(cgrp, ss, |
5033 | parent->subtree_control & (1 << ssid)); | |
ba0f4d76 TH |
5034 | if (ret) |
5035 | goto out_destroy; | |
b85d2040 | 5036 | } |
a8638030 | 5037 | } |
ddbcc7e8 | 5038 | |
bd53d617 TH |
5039 | /* |
5040 | * On the default hierarchy, a child doesn't automatically inherit | |
667c2491 | 5041 | * subtree_control from the parent. Each is configured manually. |
bd53d617 | 5042 | */ |
667c2491 TH |
5043 | if (!cgroup_on_dfl(cgrp)) { |
5044 | cgrp->subtree_control = parent->subtree_control; | |
5045 | cgroup_refresh_child_subsys_mask(cgrp); | |
5046 | } | |
2bd59d48 | 5047 | |
2bd59d48 | 5048 | kernfs_activate(kn); |
ddbcc7e8 | 5049 | |
ba0f4d76 TH |
5050 | ret = 0; |
5051 | goto out_unlock; | |
ddbcc7e8 | 5052 | |
ba0f4d76 | 5053 | out_free_id: |
6fa4918d | 5054 | cgroup_idr_remove(&root->cgroup_idr, cgrp->id); |
9d755d33 | 5055 | out_cancel_ref: |
9a1049da | 5056 | percpu_ref_exit(&cgrp->self.refcnt); |
ba0f4d76 | 5057 | out_free_cgrp: |
bd89aabc | 5058 | kfree(cgrp); |
ba0f4d76 | 5059 | out_unlock: |
a9746d8d | 5060 | cgroup_kn_unlock(parent_kn); |
ba0f4d76 | 5061 | return ret; |
4b8b47eb | 5062 | |
ba0f4d76 | 5063 | out_destroy: |
4b8b47eb | 5064 | cgroup_destroy_locked(cgrp); |
ba0f4d76 | 5065 | goto out_unlock; |
ddbcc7e8 PM |
5066 | } |
5067 | ||
223dbc38 TH |
5068 | /* |
5069 | * This is called when the refcnt of a css is confirmed to be killed. | |
249f3468 TH |
5070 | * css_tryget_online() is now guaranteed to fail. Tell the subsystem to |
5071 | * initate destruction and put the css ref from kill_css(). | |
223dbc38 TH |
5072 | */ |
5073 | static void css_killed_work_fn(struct work_struct *work) | |
d3daf28d | 5074 | { |
223dbc38 TH |
5075 | struct cgroup_subsys_state *css = |
5076 | container_of(work, struct cgroup_subsys_state, destroy_work); | |
d3daf28d | 5077 | |
f20104de | 5078 | mutex_lock(&cgroup_mutex); |
09a503ea | 5079 | |
aa226ff4 TH |
5080 | do { |
5081 | offline_css(css); | |
5082 | css_put(css); | |
5083 | /* @css can't go away while we're holding cgroup_mutex */ | |
5084 | css = css->parent; | |
5085 | } while (css && atomic_dec_and_test(&css->online_cnt)); | |
5086 | ||
5087 | mutex_unlock(&cgroup_mutex); | |
d3daf28d TH |
5088 | } |
5089 | ||
223dbc38 TH |
5090 | /* css kill confirmation processing requires process context, bounce */ |
5091 | static void css_killed_ref_fn(struct percpu_ref *ref) | |
d3daf28d TH |
5092 | { |
5093 | struct cgroup_subsys_state *css = | |
5094 | container_of(ref, struct cgroup_subsys_state, refcnt); | |
5095 | ||
aa226ff4 TH |
5096 | if (atomic_dec_and_test(&css->online_cnt)) { |
5097 | INIT_WORK(&css->destroy_work, css_killed_work_fn); | |
5098 | queue_work(cgroup_destroy_wq, &css->destroy_work); | |
5099 | } | |
d3daf28d TH |
5100 | } |
5101 | ||
f392e51c TH |
5102 | /** |
5103 | * kill_css - destroy a css | |
5104 | * @css: css to destroy | |
5105 | * | |
5106 | * This function initiates destruction of @css by removing cgroup interface | |
5107 | * files and putting its base reference. ->css_offline() will be invoked | |
ec903c0c TH |
5108 | * asynchronously once css_tryget_online() is guaranteed to fail and when |
5109 | * the reference count reaches zero, @css will be released. | |
f392e51c TH |
5110 | */ |
5111 | static void kill_css(struct cgroup_subsys_state *css) | |
edae0c33 | 5112 | { |
01f6474c | 5113 | lockdep_assert_held(&cgroup_mutex); |
94419627 | 5114 | |
2bd59d48 TH |
5115 | /* |
5116 | * This must happen before css is disassociated with its cgroup. | |
5117 | * See seq_css() for details. | |
5118 | */ | |
4df8dc90 | 5119 | css_clear_dir(css, NULL); |
3c14f8b4 | 5120 | |
edae0c33 TH |
5121 | /* |
5122 | * Killing would put the base ref, but we need to keep it alive | |
5123 | * until after ->css_offline(). | |
5124 | */ | |
5125 | css_get(css); | |
5126 | ||
5127 | /* | |
5128 | * cgroup core guarantees that, by the time ->css_offline() is | |
5129 | * invoked, no new css reference will be given out via | |
ec903c0c | 5130 | * css_tryget_online(). We can't simply call percpu_ref_kill() and |
edae0c33 TH |
5131 | * proceed to offlining css's because percpu_ref_kill() doesn't |
5132 | * guarantee that the ref is seen as killed on all CPUs on return. | |
5133 | * | |
5134 | * Use percpu_ref_kill_and_confirm() to get notifications as each | |
5135 | * css is confirmed to be seen as killed on all CPUs. | |
5136 | */ | |
5137 | percpu_ref_kill_and_confirm(&css->refcnt, css_killed_ref_fn); | |
d3daf28d TH |
5138 | } |
5139 | ||
5140 | /** | |
5141 | * cgroup_destroy_locked - the first stage of cgroup destruction | |
5142 | * @cgrp: cgroup to be destroyed | |
5143 | * | |
5144 | * css's make use of percpu refcnts whose killing latency shouldn't be | |
5145 | * exposed to userland and are RCU protected. Also, cgroup core needs to | |
ec903c0c TH |
5146 | * guarantee that css_tryget_online() won't succeed by the time |
5147 | * ->css_offline() is invoked. To satisfy all the requirements, | |
5148 | * destruction is implemented in the following two steps. | |
d3daf28d TH |
5149 | * |
5150 | * s1. Verify @cgrp can be destroyed and mark it dying. Remove all | |
5151 | * userland visible parts and start killing the percpu refcnts of | |
5152 | * css's. Set up so that the next stage will be kicked off once all | |
5153 | * the percpu refcnts are confirmed to be killed. | |
5154 | * | |
5155 | * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the | |
5156 | * rest of destruction. Once all cgroup references are gone, the | |
5157 | * cgroup is RCU-freed. | |
5158 | * | |
5159 | * This function implements s1. After this step, @cgrp is gone as far as | |
5160 | * the userland is concerned and a new cgroup with the same name may be | |
5161 | * created. As cgroup doesn't care about the names internally, this | |
5162 | * doesn't cause any problem. | |
5163 | */ | |
42809dd4 TH |
5164 | static int cgroup_destroy_locked(struct cgroup *cgrp) |
5165 | __releases(&cgroup_mutex) __acquires(&cgroup_mutex) | |
ddbcc7e8 | 5166 | { |
2bd59d48 | 5167 | struct cgroup_subsys_state *css; |
1c6727af | 5168 | int ssid; |
ddbcc7e8 | 5169 | |
42809dd4 TH |
5170 | lockdep_assert_held(&cgroup_mutex); |
5171 | ||
91486f61 TH |
5172 | /* |
5173 | * Only migration can raise populated from zero and we're already | |
5174 | * holding cgroup_mutex. | |
5175 | */ | |
5176 | if (cgroup_is_populated(cgrp)) | |
ddbcc7e8 | 5177 | return -EBUSY; |
a043e3b2 | 5178 | |
bb78a92f | 5179 | /* |
d5c419b6 TH |
5180 | * Make sure there's no live children. We can't test emptiness of |
5181 | * ->self.children as dead children linger on it while being | |
5182 | * drained; otherwise, "rmdir parent/child parent" may fail. | |
bb78a92f | 5183 | */ |
f3d46500 | 5184 | if (css_has_online_children(&cgrp->self)) |
bb78a92f HD |
5185 | return -EBUSY; |
5186 | ||
455050d2 TH |
5187 | /* |
5188 | * Mark @cgrp dead. This prevents further task migration and child | |
de3f0341 | 5189 | * creation by disabling cgroup_lock_live_group(). |
455050d2 | 5190 | */ |
184faf32 | 5191 | cgrp->self.flags &= ~CSS_ONLINE; |
ddbcc7e8 | 5192 | |
249f3468 | 5193 | /* initiate massacre of all css's */ |
1c6727af TH |
5194 | for_each_css(css, ssid, cgrp) |
5195 | kill_css(css); | |
455050d2 | 5196 | |
455050d2 | 5197 | /* |
01f6474c TH |
5198 | * Remove @cgrp directory along with the base files. @cgrp has an |
5199 | * extra ref on its kn. | |
f20104de | 5200 | */ |
01f6474c | 5201 | kernfs_remove(cgrp->kn); |
f20104de | 5202 | |
d51f39b0 | 5203 | check_for_release(cgroup_parent(cgrp)); |
2bd59d48 | 5204 | |
249f3468 | 5205 | /* put the base reference */ |
9d755d33 | 5206 | percpu_ref_kill(&cgrp->self.refcnt); |
455050d2 | 5207 | |
ea15f8cc TH |
5208 | return 0; |
5209 | }; | |
5210 | ||
2bd59d48 | 5211 | static int cgroup_rmdir(struct kernfs_node *kn) |
42809dd4 | 5212 | { |
a9746d8d | 5213 | struct cgroup *cgrp; |
2bd59d48 | 5214 | int ret = 0; |
42809dd4 | 5215 | |
a9746d8d TH |
5216 | cgrp = cgroup_kn_lock_live(kn); |
5217 | if (!cgrp) | |
5218 | return 0; | |
42809dd4 | 5219 | |
a9746d8d | 5220 | ret = cgroup_destroy_locked(cgrp); |
2bb566cb | 5221 | |
a9746d8d | 5222 | cgroup_kn_unlock(kn); |
42809dd4 | 5223 | return ret; |
8e3f6541 TH |
5224 | } |
5225 | ||
2bd59d48 TH |
5226 | static struct kernfs_syscall_ops cgroup_kf_syscall_ops = { |
5227 | .remount_fs = cgroup_remount, | |
5228 | .show_options = cgroup_show_options, | |
5229 | .mkdir = cgroup_mkdir, | |
5230 | .rmdir = cgroup_rmdir, | |
5231 | .rename = cgroup_rename, | |
5232 | }; | |
5233 | ||
15a4c835 | 5234 | static void __init cgroup_init_subsys(struct cgroup_subsys *ss, bool early) |
ddbcc7e8 | 5235 | { |
ddbcc7e8 | 5236 | struct cgroup_subsys_state *css; |
cfe36bde | 5237 | |
a5ae9899 | 5238 | pr_debug("Initializing cgroup subsys %s\n", ss->name); |
ddbcc7e8 | 5239 | |
648bb56d TH |
5240 | mutex_lock(&cgroup_mutex); |
5241 | ||
15a4c835 | 5242 | idr_init(&ss->css_idr); |
0adb0704 | 5243 | INIT_LIST_HEAD(&ss->cfts); |
8e3f6541 | 5244 | |
3dd06ffa TH |
5245 | /* Create the root cgroup state for this subsystem */ |
5246 | ss->root = &cgrp_dfl_root; | |
5247 | css = ss->css_alloc(cgroup_css(&cgrp_dfl_root.cgrp, ss)); | |
ddbcc7e8 PM |
5248 | /* We don't handle early failures gracefully */ |
5249 | BUG_ON(IS_ERR(css)); | |
ddfcadab | 5250 | init_and_link_css(css, ss, &cgrp_dfl_root.cgrp); |
3b514d24 TH |
5251 | |
5252 | /* | |
5253 | * Root csses are never destroyed and we can't initialize | |
5254 | * percpu_ref during early init. Disable refcnting. | |
5255 | */ | |
5256 | css->flags |= CSS_NO_REF; | |
5257 | ||
15a4c835 | 5258 | if (early) { |
9395a450 | 5259 | /* allocation can't be done safely during early init */ |
15a4c835 TH |
5260 | css->id = 1; |
5261 | } else { | |
5262 | css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2, GFP_KERNEL); | |
5263 | BUG_ON(css->id < 0); | |
5264 | } | |
ddbcc7e8 | 5265 | |
e8d55fde | 5266 | /* Update the init_css_set to contain a subsys |
817929ec | 5267 | * pointer to this state - since the subsystem is |
e8d55fde | 5268 | * newly registered, all tasks and hence the |
3dd06ffa | 5269 | * init_css_set is in the subsystem's root cgroup. */ |
aec25020 | 5270 | init_css_set.subsys[ss->id] = css; |
ddbcc7e8 | 5271 | |
cb4a3167 AS |
5272 | have_fork_callback |= (bool)ss->fork << ss->id; |
5273 | have_exit_callback |= (bool)ss->exit << ss->id; | |
afcf6c8b | 5274 | have_free_callback |= (bool)ss->free << ss->id; |
7e47682e | 5275 | have_canfork_callback |= (bool)ss->can_fork << ss->id; |
ddbcc7e8 | 5276 | |
e8d55fde LZ |
5277 | /* At system boot, before all subsystems have been |
5278 | * registered, no tasks have been forked, so we don't | |
5279 | * need to invoke fork callbacks here. */ | |
5280 | BUG_ON(!list_empty(&init_task.tasks)); | |
5281 | ||
ae7f164a | 5282 | BUG_ON(online_css(css)); |
a8638030 | 5283 | |
cf5d5941 BB |
5284 | mutex_unlock(&cgroup_mutex); |
5285 | } | |
cf5d5941 | 5286 | |
ddbcc7e8 | 5287 | /** |
a043e3b2 LZ |
5288 | * cgroup_init_early - cgroup initialization at system boot |
5289 | * | |
5290 | * Initialize cgroups at system boot, and initialize any | |
5291 | * subsystems that request early init. | |
ddbcc7e8 PM |
5292 | */ |
5293 | int __init cgroup_init_early(void) | |
5294 | { | |
7b9a6ba5 | 5295 | static struct cgroup_sb_opts __initdata opts; |
30159ec7 | 5296 | struct cgroup_subsys *ss; |
ddbcc7e8 | 5297 | int i; |
30159ec7 | 5298 | |
3dd06ffa | 5299 | init_cgroup_root(&cgrp_dfl_root, &opts); |
3b514d24 TH |
5300 | cgrp_dfl_root.cgrp.self.flags |= CSS_NO_REF; |
5301 | ||
a4ea1cc9 | 5302 | RCU_INIT_POINTER(init_task.cgroups, &init_css_set); |
817929ec | 5303 | |
3ed80a62 | 5304 | for_each_subsys(ss, i) { |
aec25020 | 5305 | WARN(!ss->css_alloc || !ss->css_free || ss->name || ss->id, |
073219e9 TH |
5306 | "invalid cgroup_subsys %d:%s css_alloc=%p css_free=%p name:id=%d:%s\n", |
5307 | i, cgroup_subsys_name[i], ss->css_alloc, ss->css_free, | |
aec25020 | 5308 | ss->id, ss->name); |
073219e9 TH |
5309 | WARN(strlen(cgroup_subsys_name[i]) > MAX_CGROUP_TYPE_NAMELEN, |
5310 | "cgroup_subsys_name %s too long\n", cgroup_subsys_name[i]); | |
5311 | ||
aec25020 | 5312 | ss->id = i; |
073219e9 | 5313 | ss->name = cgroup_subsys_name[i]; |
3e1d2eed TH |
5314 | if (!ss->legacy_name) |
5315 | ss->legacy_name = cgroup_subsys_name[i]; | |
ddbcc7e8 PM |
5316 | |
5317 | if (ss->early_init) | |
15a4c835 | 5318 | cgroup_init_subsys(ss, true); |
ddbcc7e8 PM |
5319 | } |
5320 | return 0; | |
5321 | } | |
5322 | ||
a3e72739 TH |
5323 | static unsigned long cgroup_disable_mask __initdata; |
5324 | ||
ddbcc7e8 | 5325 | /** |
a043e3b2 LZ |
5326 | * cgroup_init - cgroup initialization |
5327 | * | |
5328 | * Register cgroup filesystem and /proc file, and initialize | |
5329 | * any subsystems that didn't request early init. | |
ddbcc7e8 PM |
5330 | */ |
5331 | int __init cgroup_init(void) | |
5332 | { | |
30159ec7 | 5333 | struct cgroup_subsys *ss; |
0ac801fe | 5334 | unsigned long key; |
035f4f51 | 5335 | int ssid; |
ddbcc7e8 | 5336 | |
1ed13287 | 5337 | BUG_ON(percpu_init_rwsem(&cgroup_threadgroup_rwsem)); |
a14c6874 TH |
5338 | BUG_ON(cgroup_init_cftypes(NULL, cgroup_dfl_base_files)); |
5339 | BUG_ON(cgroup_init_cftypes(NULL, cgroup_legacy_base_files)); | |
ddbcc7e8 | 5340 | |
a79a908f AK |
5341 | get_user_ns(init_cgroup_ns.user_ns); |
5342 | ||
54e7b4eb | 5343 | mutex_lock(&cgroup_mutex); |
54e7b4eb | 5344 | |
82fe9b0d TH |
5345 | /* Add init_css_set to the hash table */ |
5346 | key = css_set_hash(init_css_set.subsys); | |
5347 | hash_add(css_set_table, &init_css_set.hlist, key); | |
5348 | ||
3dd06ffa | 5349 | BUG_ON(cgroup_setup_root(&cgrp_dfl_root, 0)); |
4e96ee8e | 5350 | |
54e7b4eb TH |
5351 | mutex_unlock(&cgroup_mutex); |
5352 | ||
172a2c06 | 5353 | for_each_subsys(ss, ssid) { |
15a4c835 TH |
5354 | if (ss->early_init) { |
5355 | struct cgroup_subsys_state *css = | |
5356 | init_css_set.subsys[ss->id]; | |
5357 | ||
5358 | css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2, | |
5359 | GFP_KERNEL); | |
5360 | BUG_ON(css->id < 0); | |
5361 | } else { | |
5362 | cgroup_init_subsys(ss, false); | |
5363 | } | |
172a2c06 | 5364 | |
2d8f243a TH |
5365 | list_add_tail(&init_css_set.e_cset_node[ssid], |
5366 | &cgrp_dfl_root.cgrp.e_csets[ssid]); | |
172a2c06 TH |
5367 | |
5368 | /* | |
c731ae1d LZ |
5369 | * Setting dfl_root subsys_mask needs to consider the |
5370 | * disabled flag and cftype registration needs kmalloc, | |
5371 | * both of which aren't available during early_init. | |
172a2c06 | 5372 | */ |
a3e72739 TH |
5373 | if (cgroup_disable_mask & (1 << ssid)) { |
5374 | static_branch_disable(cgroup_subsys_enabled_key[ssid]); | |
5375 | printk(KERN_INFO "Disabling %s control group subsystem\n", | |
5376 | ss->name); | |
a8ddc821 | 5377 | continue; |
a3e72739 | 5378 | } |
a8ddc821 | 5379 | |
223ffb29 JW |
5380 | if (cgroup_ssid_no_v1(ssid)) |
5381 | printk(KERN_INFO "Disabling %s control group subsystem in v1 mounts\n", | |
5382 | ss->name); | |
5383 | ||
a8ddc821 TH |
5384 | cgrp_dfl_root.subsys_mask |= 1 << ss->id; |
5385 | ||
5de4fa13 TH |
5386 | if (!ss->dfl_cftypes) |
5387 | cgrp_dfl_root_inhibit_ss_mask |= 1 << ss->id; | |
5388 | ||
a8ddc821 TH |
5389 | if (ss->dfl_cftypes == ss->legacy_cftypes) { |
5390 | WARN_ON(cgroup_add_cftypes(ss, ss->dfl_cftypes)); | |
5391 | } else { | |
5392 | WARN_ON(cgroup_add_dfl_cftypes(ss, ss->dfl_cftypes)); | |
5393 | WARN_ON(cgroup_add_legacy_cftypes(ss, ss->legacy_cftypes)); | |
c731ae1d | 5394 | } |
295458e6 VD |
5395 | |
5396 | if (ss->bind) | |
5397 | ss->bind(init_css_set.subsys[ssid]); | |
676db4af GK |
5398 | } |
5399 | ||
035f4f51 TH |
5400 | WARN_ON(sysfs_create_mount_point(fs_kobj, "cgroup")); |
5401 | WARN_ON(register_filesystem(&cgroup_fs_type)); | |
67e9c74b | 5402 | WARN_ON(register_filesystem(&cgroup2_fs_type)); |
035f4f51 | 5403 | WARN_ON(!proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations)); |
ddbcc7e8 | 5404 | |
2bd59d48 | 5405 | return 0; |
ddbcc7e8 | 5406 | } |
b4f48b63 | 5407 | |
e5fca243 TH |
5408 | static int __init cgroup_wq_init(void) |
5409 | { | |
5410 | /* | |
5411 | * There isn't much point in executing destruction path in | |
5412 | * parallel. Good chunk is serialized with cgroup_mutex anyway. | |
1a11533f | 5413 | * Use 1 for @max_active. |
e5fca243 TH |
5414 | * |
5415 | * We would prefer to do this in cgroup_init() above, but that | |
5416 | * is called before init_workqueues(): so leave this until after. | |
5417 | */ | |
1a11533f | 5418 | cgroup_destroy_wq = alloc_workqueue("cgroup_destroy", 0, 1); |
e5fca243 | 5419 | BUG_ON(!cgroup_destroy_wq); |
b1a21367 TH |
5420 | |
5421 | /* | |
5422 | * Used to destroy pidlists and separate to serve as flush domain. | |
5423 | * Cap @max_active to 1 too. | |
5424 | */ | |
5425 | cgroup_pidlist_destroy_wq = alloc_workqueue("cgroup_pidlist_destroy", | |
5426 | 0, 1); | |
5427 | BUG_ON(!cgroup_pidlist_destroy_wq); | |
5428 | ||
e5fca243 TH |
5429 | return 0; |
5430 | } | |
5431 | core_initcall(cgroup_wq_init); | |
5432 | ||
a424316c PM |
5433 | /* |
5434 | * proc_cgroup_show() | |
5435 | * - Print task's cgroup paths into seq_file, one line for each hierarchy | |
5436 | * - Used for /proc/<pid>/cgroup. | |
a424316c | 5437 | */ |
006f4ac4 ZL |
5438 | int proc_cgroup_show(struct seq_file *m, struct pid_namespace *ns, |
5439 | struct pid *pid, struct task_struct *tsk) | |
a424316c | 5440 | { |
e61734c5 | 5441 | char *buf, *path; |
a424316c | 5442 | int retval; |
3dd06ffa | 5443 | struct cgroup_root *root; |
a424316c PM |
5444 | |
5445 | retval = -ENOMEM; | |
e61734c5 | 5446 | buf = kmalloc(PATH_MAX, GFP_KERNEL); |
a424316c PM |
5447 | if (!buf) |
5448 | goto out; | |
5449 | ||
a424316c | 5450 | mutex_lock(&cgroup_mutex); |
f0d9a5f1 | 5451 | spin_lock_bh(&css_set_lock); |
a424316c | 5452 | |
985ed670 | 5453 | for_each_root(root) { |
a424316c | 5454 | struct cgroup_subsys *ss; |
bd89aabc | 5455 | struct cgroup *cgrp; |
b85d2040 | 5456 | int ssid, count = 0; |
a424316c | 5457 | |
a2dd4247 | 5458 | if (root == &cgrp_dfl_root && !cgrp_dfl_root_visible) |
985ed670 TH |
5459 | continue; |
5460 | ||
2c6ab6d2 | 5461 | seq_printf(m, "%d:", root->hierarchy_id); |
d98817d4 TH |
5462 | if (root != &cgrp_dfl_root) |
5463 | for_each_subsys(ss, ssid) | |
5464 | if (root->subsys_mask & (1 << ssid)) | |
5465 | seq_printf(m, "%s%s", count++ ? "," : "", | |
3e1d2eed | 5466 | ss->legacy_name); |
c6d57f33 PM |
5467 | if (strlen(root->name)) |
5468 | seq_printf(m, "%sname=%s", count ? "," : "", | |
5469 | root->name); | |
a424316c | 5470 | seq_putc(m, ':'); |
2e91fa7f | 5471 | |
7717f7ba | 5472 | cgrp = task_cgroup_from_root(tsk, root); |
2e91fa7f TH |
5473 | |
5474 | /* | |
5475 | * On traditional hierarchies, all zombie tasks show up as | |
5476 | * belonging to the root cgroup. On the default hierarchy, | |
5477 | * while a zombie doesn't show up in "cgroup.procs" and | |
5478 | * thus can't be migrated, its /proc/PID/cgroup keeps | |
5479 | * reporting the cgroup it belonged to before exiting. If | |
5480 | * the cgroup is removed before the zombie is reaped, | |
5481 | * " (deleted)" is appended to the cgroup path. | |
5482 | */ | |
5483 | if (cgroup_on_dfl(cgrp) || !(tsk->flags & PF_EXITING)) { | |
a79a908f AK |
5484 | path = cgroup_path_ns_locked(cgrp, buf, PATH_MAX, |
5485 | current->nsproxy->cgroup_ns); | |
2e91fa7f TH |
5486 | if (!path) { |
5487 | retval = -ENAMETOOLONG; | |
5488 | goto out_unlock; | |
5489 | } | |
5490 | } else { | |
5491 | path = "/"; | |
e61734c5 | 5492 | } |
2e91fa7f | 5493 | |
e61734c5 | 5494 | seq_puts(m, path); |
2e91fa7f TH |
5495 | |
5496 | if (cgroup_on_dfl(cgrp) && cgroup_is_dead(cgrp)) | |
5497 | seq_puts(m, " (deleted)\n"); | |
5498 | else | |
5499 | seq_putc(m, '\n'); | |
a424316c PM |
5500 | } |
5501 | ||
006f4ac4 | 5502 | retval = 0; |
a424316c | 5503 | out_unlock: |
f0d9a5f1 | 5504 | spin_unlock_bh(&css_set_lock); |
a424316c | 5505 | mutex_unlock(&cgroup_mutex); |
a424316c PM |
5506 | kfree(buf); |
5507 | out: | |
5508 | return retval; | |
5509 | } | |
5510 | ||
a424316c PM |
5511 | /* Display information about each subsystem and each hierarchy */ |
5512 | static int proc_cgroupstats_show(struct seq_file *m, void *v) | |
5513 | { | |
30159ec7 | 5514 | struct cgroup_subsys *ss; |
a424316c | 5515 | int i; |
a424316c | 5516 | |
8bab8dde | 5517 | seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n"); |
aae8aab4 BB |
5518 | /* |
5519 | * ideally we don't want subsystems moving around while we do this. | |
5520 | * cgroup_mutex is also necessary to guarantee an atomic snapshot of | |
5521 | * subsys/hierarchy state. | |
5522 | */ | |
a424316c | 5523 | mutex_lock(&cgroup_mutex); |
30159ec7 TH |
5524 | |
5525 | for_each_subsys(ss, i) | |
2c6ab6d2 | 5526 | seq_printf(m, "%s\t%d\t%d\t%d\n", |
3e1d2eed | 5527 | ss->legacy_name, ss->root->hierarchy_id, |
fc5ed1e9 TH |
5528 | atomic_read(&ss->root->nr_cgrps), |
5529 | cgroup_ssid_enabled(i)); | |
30159ec7 | 5530 | |
a424316c PM |
5531 | mutex_unlock(&cgroup_mutex); |
5532 | return 0; | |
5533 | } | |
5534 | ||
5535 | static int cgroupstats_open(struct inode *inode, struct file *file) | |
5536 | { | |
9dce07f1 | 5537 | return single_open(file, proc_cgroupstats_show, NULL); |
a424316c PM |
5538 | } |
5539 | ||
828c0950 | 5540 | static const struct file_operations proc_cgroupstats_operations = { |
a424316c PM |
5541 | .open = cgroupstats_open, |
5542 | .read = seq_read, | |
5543 | .llseek = seq_lseek, | |
5544 | .release = single_release, | |
5545 | }; | |
5546 | ||
b4f48b63 | 5547 | /** |
eaf797ab | 5548 | * cgroup_fork - initialize cgroup related fields during copy_process() |
a043e3b2 | 5549 | * @child: pointer to task_struct of forking parent process. |
b4f48b63 | 5550 | * |
eaf797ab TH |
5551 | * A task is associated with the init_css_set until cgroup_post_fork() |
5552 | * attaches it to the parent's css_set. Empty cg_list indicates that | |
5553 | * @child isn't holding reference to its css_set. | |
b4f48b63 PM |
5554 | */ |
5555 | void cgroup_fork(struct task_struct *child) | |
5556 | { | |
eaf797ab | 5557 | RCU_INIT_POINTER(child->cgroups, &init_css_set); |
817929ec | 5558 | INIT_LIST_HEAD(&child->cg_list); |
b4f48b63 PM |
5559 | } |
5560 | ||
7e47682e AS |
5561 | /** |
5562 | * cgroup_can_fork - called on a new task before the process is exposed | |
5563 | * @child: the task in question. | |
5564 | * | |
5565 | * This calls the subsystem can_fork() callbacks. If the can_fork() callback | |
5566 | * returns an error, the fork aborts with that error code. This allows for | |
5567 | * a cgroup subsystem to conditionally allow or deny new forks. | |
5568 | */ | |
b53202e6 | 5569 | int cgroup_can_fork(struct task_struct *child) |
7e47682e AS |
5570 | { |
5571 | struct cgroup_subsys *ss; | |
5572 | int i, j, ret; | |
5573 | ||
5574 | for_each_subsys_which(ss, i, &have_canfork_callback) { | |
b53202e6 | 5575 | ret = ss->can_fork(child); |
7e47682e AS |
5576 | if (ret) |
5577 | goto out_revert; | |
5578 | } | |
5579 | ||
5580 | return 0; | |
5581 | ||
5582 | out_revert: | |
5583 | for_each_subsys(ss, j) { | |
5584 | if (j >= i) | |
5585 | break; | |
5586 | if (ss->cancel_fork) | |
b53202e6 | 5587 | ss->cancel_fork(child); |
7e47682e AS |
5588 | } |
5589 | ||
5590 | return ret; | |
5591 | } | |
5592 | ||
5593 | /** | |
5594 | * cgroup_cancel_fork - called if a fork failed after cgroup_can_fork() | |
5595 | * @child: the task in question | |
5596 | * | |
5597 | * This calls the cancel_fork() callbacks if a fork failed *after* | |
5598 | * cgroup_can_fork() succeded. | |
5599 | */ | |
b53202e6 | 5600 | void cgroup_cancel_fork(struct task_struct *child) |
7e47682e AS |
5601 | { |
5602 | struct cgroup_subsys *ss; | |
5603 | int i; | |
5604 | ||
5605 | for_each_subsys(ss, i) | |
5606 | if (ss->cancel_fork) | |
b53202e6 | 5607 | ss->cancel_fork(child); |
7e47682e AS |
5608 | } |
5609 | ||
817929ec | 5610 | /** |
a043e3b2 LZ |
5611 | * cgroup_post_fork - called on a new task after adding it to the task list |
5612 | * @child: the task in question | |
5613 | * | |
5edee61e TH |
5614 | * Adds the task to the list running through its css_set if necessary and |
5615 | * call the subsystem fork() callbacks. Has to be after the task is | |
5616 | * visible on the task list in case we race with the first call to | |
0942eeee | 5617 | * cgroup_task_iter_start() - to guarantee that the new task ends up on its |
5edee61e | 5618 | * list. |
a043e3b2 | 5619 | */ |
b53202e6 | 5620 | void cgroup_post_fork(struct task_struct *child) |
817929ec | 5621 | { |
30159ec7 | 5622 | struct cgroup_subsys *ss; |
5edee61e TH |
5623 | int i; |
5624 | ||
3ce3230a | 5625 | /* |
251f8c03 | 5626 | * This may race against cgroup_enable_task_cg_lists(). As that |
eaf797ab TH |
5627 | * function sets use_task_css_set_links before grabbing |
5628 | * tasklist_lock and we just went through tasklist_lock to add | |
5629 | * @child, it's guaranteed that either we see the set | |
5630 | * use_task_css_set_links or cgroup_enable_task_cg_lists() sees | |
5631 | * @child during its iteration. | |
5632 | * | |
5633 | * If we won the race, @child is associated with %current's | |
f0d9a5f1 | 5634 | * css_set. Grabbing css_set_lock guarantees both that the |
eaf797ab TH |
5635 | * association is stable, and, on completion of the parent's |
5636 | * migration, @child is visible in the source of migration or | |
5637 | * already in the destination cgroup. This guarantee is necessary | |
5638 | * when implementing operations which need to migrate all tasks of | |
5639 | * a cgroup to another. | |
5640 | * | |
251f8c03 | 5641 | * Note that if we lose to cgroup_enable_task_cg_lists(), @child |
eaf797ab TH |
5642 | * will remain in init_css_set. This is safe because all tasks are |
5643 | * in the init_css_set before cg_links is enabled and there's no | |
5644 | * operation which transfers all tasks out of init_css_set. | |
3ce3230a | 5645 | */ |
817929ec | 5646 | if (use_task_css_set_links) { |
eaf797ab TH |
5647 | struct css_set *cset; |
5648 | ||
f0d9a5f1 | 5649 | spin_lock_bh(&css_set_lock); |
0e1d768f | 5650 | cset = task_css_set(current); |
eaf797ab | 5651 | if (list_empty(&child->cg_list)) { |
eaf797ab | 5652 | get_css_set(cset); |
f6d7d049 | 5653 | css_set_move_task(child, NULL, cset, false); |
eaf797ab | 5654 | } |
f0d9a5f1 | 5655 | spin_unlock_bh(&css_set_lock); |
817929ec | 5656 | } |
5edee61e TH |
5657 | |
5658 | /* | |
5659 | * Call ss->fork(). This must happen after @child is linked on | |
5660 | * css_set; otherwise, @child might change state between ->fork() | |
5661 | * and addition to css_set. | |
5662 | */ | |
cb4a3167 | 5663 | for_each_subsys_which(ss, i, &have_fork_callback) |
b53202e6 | 5664 | ss->fork(child); |
817929ec | 5665 | } |
5edee61e | 5666 | |
b4f48b63 PM |
5667 | /** |
5668 | * cgroup_exit - detach cgroup from exiting task | |
5669 | * @tsk: pointer to task_struct of exiting process | |
5670 | * | |
5671 | * Description: Detach cgroup from @tsk and release it. | |
5672 | * | |
5673 | * Note that cgroups marked notify_on_release force every task in | |
5674 | * them to take the global cgroup_mutex mutex when exiting. | |
5675 | * This could impact scaling on very large systems. Be reluctant to | |
5676 | * use notify_on_release cgroups where very high task exit scaling | |
5677 | * is required on large systems. | |
5678 | * | |
0e1d768f TH |
5679 | * We set the exiting tasks cgroup to the root cgroup (top_cgroup). We |
5680 | * call cgroup_exit() while the task is still competent to handle | |
5681 | * notify_on_release(), then leave the task attached to the root cgroup in | |
5682 | * each hierarchy for the remainder of its exit. No need to bother with | |
5683 | * init_css_set refcnting. init_css_set never goes away and we can't race | |
e8604cb4 | 5684 | * with migration path - PF_EXITING is visible to migration path. |
b4f48b63 | 5685 | */ |
1ec41830 | 5686 | void cgroup_exit(struct task_struct *tsk) |
b4f48b63 | 5687 | { |
30159ec7 | 5688 | struct cgroup_subsys *ss; |
5abb8855 | 5689 | struct css_set *cset; |
d41d5a01 | 5690 | int i; |
817929ec PM |
5691 | |
5692 | /* | |
0e1d768f | 5693 | * Unlink from @tsk from its css_set. As migration path can't race |
0de0942d | 5694 | * with us, we can check css_set and cg_list without synchronization. |
817929ec | 5695 | */ |
0de0942d TH |
5696 | cset = task_css_set(tsk); |
5697 | ||
817929ec | 5698 | if (!list_empty(&tsk->cg_list)) { |
f0d9a5f1 | 5699 | spin_lock_bh(&css_set_lock); |
f6d7d049 | 5700 | css_set_move_task(tsk, cset, NULL, false); |
f0d9a5f1 | 5701 | spin_unlock_bh(&css_set_lock); |
2e91fa7f TH |
5702 | } else { |
5703 | get_css_set(cset); | |
817929ec PM |
5704 | } |
5705 | ||
cb4a3167 | 5706 | /* see cgroup_post_fork() for details */ |
2e91fa7f TH |
5707 | for_each_subsys_which(ss, i, &have_exit_callback) |
5708 | ss->exit(tsk); | |
5709 | } | |
30159ec7 | 5710 | |
2e91fa7f TH |
5711 | void cgroup_free(struct task_struct *task) |
5712 | { | |
5713 | struct css_set *cset = task_css_set(task); | |
afcf6c8b TH |
5714 | struct cgroup_subsys *ss; |
5715 | int ssid; | |
5716 | ||
5717 | for_each_subsys_which(ss, ssid, &have_free_callback) | |
5718 | ss->free(task); | |
d41d5a01 | 5719 | |
2e91fa7f | 5720 | put_css_set(cset); |
b4f48b63 | 5721 | } |
697f4161 | 5722 | |
bd89aabc | 5723 | static void check_for_release(struct cgroup *cgrp) |
81a6a5cd | 5724 | { |
27bd4dbb | 5725 | if (notify_on_release(cgrp) && !cgroup_is_populated(cgrp) && |
971ff493 ZL |
5726 | !css_has_online_children(&cgrp->self) && !cgroup_is_dead(cgrp)) |
5727 | schedule_work(&cgrp->release_agent_work); | |
81a6a5cd PM |
5728 | } |
5729 | ||
81a6a5cd PM |
5730 | /* |
5731 | * Notify userspace when a cgroup is released, by running the | |
5732 | * configured release agent with the name of the cgroup (path | |
5733 | * relative to the root of cgroup file system) as the argument. | |
5734 | * | |
5735 | * Most likely, this user command will try to rmdir this cgroup. | |
5736 | * | |
5737 | * This races with the possibility that some other task will be | |
5738 | * attached to this cgroup before it is removed, or that some other | |
5739 | * user task will 'mkdir' a child cgroup of this cgroup. That's ok. | |
5740 | * The presumed 'rmdir' will fail quietly if this cgroup is no longer | |
5741 | * unused, and this cgroup will be reprieved from its death sentence, | |
5742 | * to continue to serve a useful existence. Next time it's released, | |
5743 | * we will get notified again, if it still has 'notify_on_release' set. | |
5744 | * | |
5745 | * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which | |
5746 | * means only wait until the task is successfully execve()'d. The | |
5747 | * separate release agent task is forked by call_usermodehelper(), | |
5748 | * then control in this thread returns here, without waiting for the | |
5749 | * release agent task. We don't bother to wait because the caller of | |
5750 | * this routine has no use for the exit status of the release agent | |
5751 | * task, so no sense holding our caller up for that. | |
81a6a5cd | 5752 | */ |
81a6a5cd PM |
5753 | static void cgroup_release_agent(struct work_struct *work) |
5754 | { | |
971ff493 ZL |
5755 | struct cgroup *cgrp = |
5756 | container_of(work, struct cgroup, release_agent_work); | |
5757 | char *pathbuf = NULL, *agentbuf = NULL, *path; | |
5758 | char *argv[3], *envp[3]; | |
5759 | ||
81a6a5cd | 5760 | mutex_lock(&cgroup_mutex); |
971ff493 ZL |
5761 | |
5762 | pathbuf = kmalloc(PATH_MAX, GFP_KERNEL); | |
5763 | agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL); | |
5764 | if (!pathbuf || !agentbuf) | |
5765 | goto out; | |
5766 | ||
a79a908f AK |
5767 | spin_lock_bh(&css_set_lock); |
5768 | path = cgroup_path_ns_locked(cgrp, pathbuf, PATH_MAX, &init_cgroup_ns); | |
5769 | spin_unlock_bh(&css_set_lock); | |
971ff493 ZL |
5770 | if (!path) |
5771 | goto out; | |
5772 | ||
5773 | argv[0] = agentbuf; | |
5774 | argv[1] = path; | |
5775 | argv[2] = NULL; | |
5776 | ||
5777 | /* minimal command environment */ | |
5778 | envp[0] = "HOME=/"; | |
5779 | envp[1] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin"; | |
5780 | envp[2] = NULL; | |
5781 | ||
81a6a5cd | 5782 | mutex_unlock(&cgroup_mutex); |
971ff493 | 5783 | call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC); |
3e2cd91a | 5784 | goto out_free; |
971ff493 | 5785 | out: |
81a6a5cd | 5786 | mutex_unlock(&cgroup_mutex); |
3e2cd91a | 5787 | out_free: |
971ff493 ZL |
5788 | kfree(agentbuf); |
5789 | kfree(pathbuf); | |
81a6a5cd | 5790 | } |
8bab8dde PM |
5791 | |
5792 | static int __init cgroup_disable(char *str) | |
5793 | { | |
30159ec7 | 5794 | struct cgroup_subsys *ss; |
8bab8dde | 5795 | char *token; |
30159ec7 | 5796 | int i; |
8bab8dde PM |
5797 | |
5798 | while ((token = strsep(&str, ",")) != NULL) { | |
5799 | if (!*token) | |
5800 | continue; | |
be45c900 | 5801 | |
3ed80a62 | 5802 | for_each_subsys(ss, i) { |
3e1d2eed TH |
5803 | if (strcmp(token, ss->name) && |
5804 | strcmp(token, ss->legacy_name)) | |
5805 | continue; | |
a3e72739 | 5806 | cgroup_disable_mask |= 1 << i; |
8bab8dde PM |
5807 | } |
5808 | } | |
5809 | return 1; | |
5810 | } | |
5811 | __setup("cgroup_disable=", cgroup_disable); | |
38460b48 | 5812 | |
223ffb29 JW |
5813 | static int __init cgroup_no_v1(char *str) |
5814 | { | |
5815 | struct cgroup_subsys *ss; | |
5816 | char *token; | |
5817 | int i; | |
5818 | ||
5819 | while ((token = strsep(&str, ",")) != NULL) { | |
5820 | if (!*token) | |
5821 | continue; | |
5822 | ||
5823 | if (!strcmp(token, "all")) { | |
5824 | cgroup_no_v1_mask = ~0UL; | |
5825 | break; | |
5826 | } | |
5827 | ||
5828 | for_each_subsys(ss, i) { | |
5829 | if (strcmp(token, ss->name) && | |
5830 | strcmp(token, ss->legacy_name)) | |
5831 | continue; | |
5832 | ||
5833 | cgroup_no_v1_mask |= 1 << i; | |
5834 | } | |
5835 | } | |
5836 | return 1; | |
5837 | } | |
5838 | __setup("cgroup_no_v1=", cgroup_no_v1); | |
5839 | ||
b77d7b60 | 5840 | /** |
ec903c0c | 5841 | * css_tryget_online_from_dir - get corresponding css from a cgroup dentry |
35cf0836 TH |
5842 | * @dentry: directory dentry of interest |
5843 | * @ss: subsystem of interest | |
b77d7b60 | 5844 | * |
5a17f543 TH |
5845 | * If @dentry is a directory for a cgroup which has @ss enabled on it, try |
5846 | * to get the corresponding css and return it. If such css doesn't exist | |
5847 | * or can't be pinned, an ERR_PTR value is returned. | |
e5d1367f | 5848 | */ |
ec903c0c TH |
5849 | struct cgroup_subsys_state *css_tryget_online_from_dir(struct dentry *dentry, |
5850 | struct cgroup_subsys *ss) | |
e5d1367f | 5851 | { |
2bd59d48 TH |
5852 | struct kernfs_node *kn = kernfs_node_from_dentry(dentry); |
5853 | struct cgroup_subsys_state *css = NULL; | |
e5d1367f | 5854 | struct cgroup *cgrp; |
e5d1367f | 5855 | |
35cf0836 | 5856 | /* is @dentry a cgroup dir? */ |
2bd59d48 TH |
5857 | if (dentry->d_sb->s_type != &cgroup_fs_type || !kn || |
5858 | kernfs_type(kn) != KERNFS_DIR) | |
e5d1367f SE |
5859 | return ERR_PTR(-EBADF); |
5860 | ||
5a17f543 TH |
5861 | rcu_read_lock(); |
5862 | ||
2bd59d48 TH |
5863 | /* |
5864 | * This path doesn't originate from kernfs and @kn could already | |
5865 | * have been or be removed at any point. @kn->priv is RCU | |
a4189487 | 5866 | * protected for this access. See css_release_work_fn() for details. |
2bd59d48 TH |
5867 | */ |
5868 | cgrp = rcu_dereference(kn->priv); | |
5869 | if (cgrp) | |
5870 | css = cgroup_css(cgrp, ss); | |
5a17f543 | 5871 | |
ec903c0c | 5872 | if (!css || !css_tryget_online(css)) |
5a17f543 TH |
5873 | css = ERR_PTR(-ENOENT); |
5874 | ||
5875 | rcu_read_unlock(); | |
5876 | return css; | |
e5d1367f | 5877 | } |
e5d1367f | 5878 | |
1cb650b9 LZ |
5879 | /** |
5880 | * css_from_id - lookup css by id | |
5881 | * @id: the cgroup id | |
5882 | * @ss: cgroup subsys to be looked into | |
5883 | * | |
5884 | * Returns the css if there's valid one with @id, otherwise returns NULL. | |
5885 | * Should be called under rcu_read_lock(). | |
5886 | */ | |
5887 | struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss) | |
5888 | { | |
6fa4918d | 5889 | WARN_ON_ONCE(!rcu_read_lock_held()); |
adbe427b | 5890 | return id > 0 ? idr_find(&ss->css_idr, id) : NULL; |
e5d1367f SE |
5891 | } |
5892 | ||
16af4396 TH |
5893 | /** |
5894 | * cgroup_get_from_path - lookup and get a cgroup from its default hierarchy path | |
5895 | * @path: path on the default hierarchy | |
5896 | * | |
5897 | * Find the cgroup at @path on the default hierarchy, increment its | |
5898 | * reference count and return it. Returns pointer to the found cgroup on | |
5899 | * success, ERR_PTR(-ENOENT) if @path doens't exist and ERR_PTR(-ENOTDIR) | |
5900 | * if @path points to a non-directory. | |
5901 | */ | |
5902 | struct cgroup *cgroup_get_from_path(const char *path) | |
5903 | { | |
5904 | struct kernfs_node *kn; | |
5905 | struct cgroup *cgrp; | |
5906 | ||
5907 | mutex_lock(&cgroup_mutex); | |
5908 | ||
5909 | kn = kernfs_walk_and_get(cgrp_dfl_root.cgrp.kn, path); | |
5910 | if (kn) { | |
5911 | if (kernfs_type(kn) == KERNFS_DIR) { | |
5912 | cgrp = kn->priv; | |
5913 | cgroup_get(cgrp); | |
5914 | } else { | |
5915 | cgrp = ERR_PTR(-ENOTDIR); | |
5916 | } | |
5917 | kernfs_put(kn); | |
5918 | } else { | |
5919 | cgrp = ERR_PTR(-ENOENT); | |
5920 | } | |
5921 | ||
5922 | mutex_unlock(&cgroup_mutex); | |
5923 | return cgrp; | |
5924 | } | |
5925 | EXPORT_SYMBOL_GPL(cgroup_get_from_path); | |
5926 | ||
bd1060a1 TH |
5927 | /* |
5928 | * sock->sk_cgrp_data handling. For more info, see sock_cgroup_data | |
5929 | * definition in cgroup-defs.h. | |
5930 | */ | |
5931 | #ifdef CONFIG_SOCK_CGROUP_DATA | |
5932 | ||
5933 | #if defined(CONFIG_CGROUP_NET_PRIO) || defined(CONFIG_CGROUP_NET_CLASSID) | |
5934 | ||
3fa4cc9c | 5935 | DEFINE_SPINLOCK(cgroup_sk_update_lock); |
bd1060a1 TH |
5936 | static bool cgroup_sk_alloc_disabled __read_mostly; |
5937 | ||
5938 | void cgroup_sk_alloc_disable(void) | |
5939 | { | |
5940 | if (cgroup_sk_alloc_disabled) | |
5941 | return; | |
5942 | pr_info("cgroup: disabling cgroup2 socket matching due to net_prio or net_cls activation\n"); | |
5943 | cgroup_sk_alloc_disabled = true; | |
5944 | } | |
5945 | ||
5946 | #else | |
5947 | ||
5948 | #define cgroup_sk_alloc_disabled false | |
5949 | ||
5950 | #endif | |
5951 | ||
5952 | void cgroup_sk_alloc(struct sock_cgroup_data *skcd) | |
5953 | { | |
5954 | if (cgroup_sk_alloc_disabled) | |
5955 | return; | |
5956 | ||
5957 | rcu_read_lock(); | |
5958 | ||
5959 | while (true) { | |
5960 | struct css_set *cset; | |
5961 | ||
5962 | cset = task_css_set(current); | |
5963 | if (likely(cgroup_tryget(cset->dfl_cgrp))) { | |
5964 | skcd->val = (unsigned long)cset->dfl_cgrp; | |
5965 | break; | |
5966 | } | |
5967 | cpu_relax(); | |
5968 | } | |
5969 | ||
5970 | rcu_read_unlock(); | |
5971 | } | |
5972 | ||
5973 | void cgroup_sk_free(struct sock_cgroup_data *skcd) | |
5974 | { | |
5975 | cgroup_put(sock_cgroup_ptr(skcd)); | |
5976 | } | |
5977 | ||
5978 | #endif /* CONFIG_SOCK_CGROUP_DATA */ | |
5979 | ||
a79a908f AK |
5980 | /* cgroup namespaces */ |
5981 | ||
5982 | static struct cgroup_namespace *alloc_cgroup_ns(void) | |
5983 | { | |
5984 | struct cgroup_namespace *new_ns; | |
5985 | int ret; | |
5986 | ||
5987 | new_ns = kzalloc(sizeof(struct cgroup_namespace), GFP_KERNEL); | |
5988 | if (!new_ns) | |
5989 | return ERR_PTR(-ENOMEM); | |
5990 | ret = ns_alloc_inum(&new_ns->ns); | |
5991 | if (ret) { | |
5992 | kfree(new_ns); | |
5993 | return ERR_PTR(ret); | |
5994 | } | |
5995 | atomic_set(&new_ns->count, 1); | |
5996 | new_ns->ns.ops = &cgroupns_operations; | |
5997 | return new_ns; | |
5998 | } | |
5999 | ||
6000 | void free_cgroup_ns(struct cgroup_namespace *ns) | |
6001 | { | |
6002 | put_css_set(ns->root_cset); | |
6003 | put_user_ns(ns->user_ns); | |
6004 | ns_free_inum(&ns->ns); | |
6005 | kfree(ns); | |
6006 | } | |
6007 | EXPORT_SYMBOL(free_cgroup_ns); | |
6008 | ||
6009 | struct cgroup_namespace *copy_cgroup_ns(unsigned long flags, | |
6010 | struct user_namespace *user_ns, | |
6011 | struct cgroup_namespace *old_ns) | |
6012 | { | |
6013 | struct cgroup_namespace *new_ns = NULL; | |
6014 | struct css_set *cset = NULL; | |
6015 | int err; | |
6016 | ||
6017 | BUG_ON(!old_ns); | |
6018 | ||
6019 | if (!(flags & CLONE_NEWCGROUP)) { | |
6020 | get_cgroup_ns(old_ns); | |
6021 | return old_ns; | |
6022 | } | |
6023 | ||
6024 | /* Allow only sysadmin to create cgroup namespace. */ | |
6025 | err = -EPERM; | |
6026 | if (!ns_capable(user_ns, CAP_SYS_ADMIN)) | |
6027 | goto err_out; | |
6028 | ||
6029 | mutex_lock(&cgroup_mutex); | |
6030 | spin_lock_bh(&css_set_lock); | |
6031 | ||
6032 | cset = task_css_set(current); | |
6033 | get_css_set(cset); | |
6034 | ||
6035 | spin_unlock_bh(&css_set_lock); | |
6036 | mutex_unlock(&cgroup_mutex); | |
6037 | ||
6038 | err = -ENOMEM; | |
6039 | new_ns = alloc_cgroup_ns(); | |
6040 | if (!new_ns) | |
6041 | goto err_out; | |
6042 | ||
6043 | new_ns->user_ns = get_user_ns(user_ns); | |
6044 | new_ns->root_cset = cset; | |
6045 | ||
6046 | return new_ns; | |
6047 | ||
6048 | err_out: | |
6049 | if (cset) | |
6050 | put_css_set(cset); | |
6051 | kfree(new_ns); | |
6052 | return ERR_PTR(err); | |
6053 | } | |
6054 | ||
6055 | static inline struct cgroup_namespace *to_cg_ns(struct ns_common *ns) | |
6056 | { | |
6057 | return container_of(ns, struct cgroup_namespace, ns); | |
6058 | } | |
6059 | ||
6060 | static int cgroupns_install(struct nsproxy *nsproxy, void *ns) | |
6061 | { | |
6062 | pr_info("setns not supported for cgroup namespace"); | |
6063 | return -EINVAL; | |
6064 | } | |
6065 | ||
6066 | static struct ns_common *cgroupns_get(struct task_struct *task) | |
6067 | { | |
6068 | struct cgroup_namespace *ns = NULL; | |
6069 | struct nsproxy *nsproxy; | |
6070 | ||
6071 | task_lock(task); | |
6072 | nsproxy = task->nsproxy; | |
6073 | if (nsproxy) { | |
6074 | ns = nsproxy->cgroup_ns; | |
6075 | get_cgroup_ns(ns); | |
6076 | } | |
6077 | task_unlock(task); | |
6078 | ||
6079 | return ns ? &ns->ns : NULL; | |
6080 | } | |
6081 | ||
6082 | static void cgroupns_put(struct ns_common *ns) | |
6083 | { | |
6084 | put_cgroup_ns(to_cg_ns(ns)); | |
6085 | } | |
6086 | ||
6087 | const struct proc_ns_operations cgroupns_operations = { | |
6088 | .name = "cgroup", | |
6089 | .type = CLONE_NEWCGROUP, | |
6090 | .get = cgroupns_get, | |
6091 | .put = cgroupns_put, | |
6092 | .install = cgroupns_install, | |
6093 | }; | |
6094 | ||
6095 | static __init int cgroup_namespaces_init(void) | |
6096 | { | |
6097 | return 0; | |
6098 | } | |
6099 | subsys_initcall(cgroup_namespaces_init); | |
6100 | ||
fe693435 | 6101 | #ifdef CONFIG_CGROUP_DEBUG |
eb95419b TH |
6102 | static struct cgroup_subsys_state * |
6103 | debug_css_alloc(struct cgroup_subsys_state *parent_css) | |
fe693435 PM |
6104 | { |
6105 | struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL); | |
6106 | ||
6107 | if (!css) | |
6108 | return ERR_PTR(-ENOMEM); | |
6109 | ||
6110 | return css; | |
6111 | } | |
6112 | ||
eb95419b | 6113 | static void debug_css_free(struct cgroup_subsys_state *css) |
fe693435 | 6114 | { |
eb95419b | 6115 | kfree(css); |
fe693435 PM |
6116 | } |
6117 | ||
182446d0 TH |
6118 | static u64 debug_taskcount_read(struct cgroup_subsys_state *css, |
6119 | struct cftype *cft) | |
fe693435 | 6120 | { |
182446d0 | 6121 | return cgroup_task_count(css->cgroup); |
fe693435 PM |
6122 | } |
6123 | ||
182446d0 TH |
6124 | static u64 current_css_set_read(struct cgroup_subsys_state *css, |
6125 | struct cftype *cft) | |
fe693435 PM |
6126 | { |
6127 | return (u64)(unsigned long)current->cgroups; | |
6128 | } | |
6129 | ||
182446d0 | 6130 | static u64 current_css_set_refcount_read(struct cgroup_subsys_state *css, |
03c78cbe | 6131 | struct cftype *cft) |
fe693435 PM |
6132 | { |
6133 | u64 count; | |
6134 | ||
6135 | rcu_read_lock(); | |
a8ad805c | 6136 | count = atomic_read(&task_css_set(current)->refcount); |
fe693435 PM |
6137 | rcu_read_unlock(); |
6138 | return count; | |
6139 | } | |
6140 | ||
2da8ca82 | 6141 | static int current_css_set_cg_links_read(struct seq_file *seq, void *v) |
7717f7ba | 6142 | { |
69d0206c | 6143 | struct cgrp_cset_link *link; |
5abb8855 | 6144 | struct css_set *cset; |
e61734c5 TH |
6145 | char *name_buf; |
6146 | ||
6147 | name_buf = kmalloc(NAME_MAX + 1, GFP_KERNEL); | |
6148 | if (!name_buf) | |
6149 | return -ENOMEM; | |
7717f7ba | 6150 | |
f0d9a5f1 | 6151 | spin_lock_bh(&css_set_lock); |
7717f7ba | 6152 | rcu_read_lock(); |
5abb8855 | 6153 | cset = rcu_dereference(current->cgroups); |
69d0206c | 6154 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) { |
7717f7ba | 6155 | struct cgroup *c = link->cgrp; |
7717f7ba | 6156 | |
a2dd4247 | 6157 | cgroup_name(c, name_buf, NAME_MAX + 1); |
2c6ab6d2 | 6158 | seq_printf(seq, "Root %d group %s\n", |
a2dd4247 | 6159 | c->root->hierarchy_id, name_buf); |
7717f7ba PM |
6160 | } |
6161 | rcu_read_unlock(); | |
f0d9a5f1 | 6162 | spin_unlock_bh(&css_set_lock); |
e61734c5 | 6163 | kfree(name_buf); |
7717f7ba PM |
6164 | return 0; |
6165 | } | |
6166 | ||
6167 | #define MAX_TASKS_SHOWN_PER_CSS 25 | |
2da8ca82 | 6168 | static int cgroup_css_links_read(struct seq_file *seq, void *v) |
7717f7ba | 6169 | { |
2da8ca82 | 6170 | struct cgroup_subsys_state *css = seq_css(seq); |
69d0206c | 6171 | struct cgrp_cset_link *link; |
7717f7ba | 6172 | |
f0d9a5f1 | 6173 | spin_lock_bh(&css_set_lock); |
182446d0 | 6174 | list_for_each_entry(link, &css->cgroup->cset_links, cset_link) { |
69d0206c | 6175 | struct css_set *cset = link->cset; |
7717f7ba PM |
6176 | struct task_struct *task; |
6177 | int count = 0; | |
c7561128 | 6178 | |
5abb8855 | 6179 | seq_printf(seq, "css_set %p\n", cset); |
c7561128 | 6180 | |
5abb8855 | 6181 | list_for_each_entry(task, &cset->tasks, cg_list) { |
c7561128 TH |
6182 | if (count++ > MAX_TASKS_SHOWN_PER_CSS) |
6183 | goto overflow; | |
6184 | seq_printf(seq, " task %d\n", task_pid_vnr(task)); | |
6185 | } | |
6186 | ||
6187 | list_for_each_entry(task, &cset->mg_tasks, cg_list) { | |
6188 | if (count++ > MAX_TASKS_SHOWN_PER_CSS) | |
6189 | goto overflow; | |
6190 | seq_printf(seq, " task %d\n", task_pid_vnr(task)); | |
7717f7ba | 6191 | } |
c7561128 TH |
6192 | continue; |
6193 | overflow: | |
6194 | seq_puts(seq, " ...\n"); | |
7717f7ba | 6195 | } |
f0d9a5f1 | 6196 | spin_unlock_bh(&css_set_lock); |
7717f7ba PM |
6197 | return 0; |
6198 | } | |
6199 | ||
182446d0 | 6200 | static u64 releasable_read(struct cgroup_subsys_state *css, struct cftype *cft) |
fe693435 | 6201 | { |
27bd4dbb | 6202 | return (!cgroup_is_populated(css->cgroup) && |
a25eb52e | 6203 | !css_has_online_children(&css->cgroup->self)); |
fe693435 PM |
6204 | } |
6205 | ||
6206 | static struct cftype debug_files[] = { | |
fe693435 PM |
6207 | { |
6208 | .name = "taskcount", | |
6209 | .read_u64 = debug_taskcount_read, | |
6210 | }, | |
6211 | ||
6212 | { | |
6213 | .name = "current_css_set", | |
6214 | .read_u64 = current_css_set_read, | |
6215 | }, | |
6216 | ||
6217 | { | |
6218 | .name = "current_css_set_refcount", | |
6219 | .read_u64 = current_css_set_refcount_read, | |
6220 | }, | |
6221 | ||
7717f7ba PM |
6222 | { |
6223 | .name = "current_css_set_cg_links", | |
2da8ca82 | 6224 | .seq_show = current_css_set_cg_links_read, |
7717f7ba PM |
6225 | }, |
6226 | ||
6227 | { | |
6228 | .name = "cgroup_css_links", | |
2da8ca82 | 6229 | .seq_show = cgroup_css_links_read, |
7717f7ba PM |
6230 | }, |
6231 | ||
fe693435 PM |
6232 | { |
6233 | .name = "releasable", | |
6234 | .read_u64 = releasable_read, | |
6235 | }, | |
fe693435 | 6236 | |
4baf6e33 TH |
6237 | { } /* terminate */ |
6238 | }; | |
fe693435 | 6239 | |
073219e9 | 6240 | struct cgroup_subsys debug_cgrp_subsys = { |
92fb9748 TH |
6241 | .css_alloc = debug_css_alloc, |
6242 | .css_free = debug_css_free, | |
5577964e | 6243 | .legacy_cftypes = debug_files, |
fe693435 PM |
6244 | }; |
6245 | #endif /* CONFIG_CGROUP_DEBUG */ |