]>
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> |
96d365e0 | 48 | #include <linux/rwsem.h> |
1ed13287 | 49 | #include <linux/percpu-rwsem.h> |
ddbcc7e8 | 50 | #include <linux/string.h> |
bbcb81d0 | 51 | #include <linux/sort.h> |
81a6a5cd | 52 | #include <linux/kmod.h> |
846c7bb0 BS |
53 | #include <linux/delayacct.h> |
54 | #include <linux/cgroupstats.h> | |
0ac801fe | 55 | #include <linux/hashtable.h> |
096b7fe0 | 56 | #include <linux/pid_namespace.h> |
2c6ab6d2 | 57 | #include <linux/idr.h> |
d1d9fd33 | 58 | #include <linux/vmalloc.h> /* TODO: replace with more sophisticated array */ |
c4c27fbd | 59 | #include <linux/kthread.h> |
776f02fa | 60 | #include <linux/delay.h> |
846c7bb0 | 61 | |
60063497 | 62 | #include <linux/atomic.h> |
ddbcc7e8 | 63 | |
b1a21367 TH |
64 | /* |
65 | * pidlists linger the following amount before being destroyed. The goal | |
66 | * is avoiding frequent destruction in the middle of consecutive read calls | |
67 | * Expiring in the middle is a performance problem not a correctness one. | |
68 | * 1 sec should be enough. | |
69 | */ | |
70 | #define CGROUP_PIDLIST_DESTROY_DELAY HZ | |
71 | ||
8d7e6fb0 TH |
72 | #define CGROUP_FILE_NAME_MAX (MAX_CGROUP_TYPE_NAMELEN + \ |
73 | MAX_CFTYPE_NAME + 2) | |
74 | ||
e25e2cbb TH |
75 | /* |
76 | * cgroup_mutex is the master lock. Any modification to cgroup or its | |
77 | * hierarchy must be performed while holding it. | |
78 | * | |
0e1d768f TH |
79 | * css_set_rwsem protects task->cgroups pointer, the list of css_set |
80 | * objects, and the chain of tasks off each css_set. | |
e25e2cbb | 81 | * |
0e1d768f TH |
82 | * These locks are exported if CONFIG_PROVE_RCU so that accessors in |
83 | * cgroup.h can use them for lockdep annotations. | |
e25e2cbb | 84 | */ |
2219449a TH |
85 | #ifdef CONFIG_PROVE_RCU |
86 | DEFINE_MUTEX(cgroup_mutex); | |
0e1d768f TH |
87 | DECLARE_RWSEM(css_set_rwsem); |
88 | EXPORT_SYMBOL_GPL(cgroup_mutex); | |
89 | EXPORT_SYMBOL_GPL(css_set_rwsem); | |
2219449a | 90 | #else |
81a6a5cd | 91 | static DEFINE_MUTEX(cgroup_mutex); |
0e1d768f | 92 | static DECLARE_RWSEM(css_set_rwsem); |
2219449a TH |
93 | #endif |
94 | ||
6fa4918d | 95 | /* |
15a4c835 TH |
96 | * Protects cgroup_idr and css_idr so that IDs can be released without |
97 | * grabbing cgroup_mutex. | |
6fa4918d TH |
98 | */ |
99 | static DEFINE_SPINLOCK(cgroup_idr_lock); | |
100 | ||
69e943b7 TH |
101 | /* |
102 | * Protects cgroup_subsys->release_agent_path. Modifying it also requires | |
103 | * cgroup_mutex. Reading requires either cgroup_mutex or this spinlock. | |
104 | */ | |
105 | static DEFINE_SPINLOCK(release_agent_path_lock); | |
81a6a5cd | 106 | |
1ed13287 TH |
107 | struct percpu_rw_semaphore cgroup_threadgroup_rwsem; |
108 | ||
8353da1f | 109 | #define cgroup_assert_mutex_or_rcu_locked() \ |
f78f5b90 PM |
110 | RCU_LOCKDEP_WARN(!rcu_read_lock_held() && \ |
111 | !lockdep_is_held(&cgroup_mutex), \ | |
8353da1f | 112 | "cgroup_mutex or RCU read lock required"); |
780cd8b3 | 113 | |
e5fca243 TH |
114 | /* |
115 | * cgroup destruction makes heavy use of work items and there can be a lot | |
116 | * of concurrent destructions. Use a separate workqueue so that cgroup | |
117 | * destruction work items don't end up filling up max_active of system_wq | |
118 | * which may lead to deadlock. | |
119 | */ | |
120 | static struct workqueue_struct *cgroup_destroy_wq; | |
121 | ||
b1a21367 TH |
122 | /* |
123 | * pidlist destructions need to be flushed on cgroup destruction. Use a | |
124 | * separate workqueue as flush domain. | |
125 | */ | |
126 | static struct workqueue_struct *cgroup_pidlist_destroy_wq; | |
127 | ||
3ed80a62 | 128 | /* generate an array of cgroup subsystem pointers */ |
073219e9 | 129 | #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys, |
3ed80a62 | 130 | static struct cgroup_subsys *cgroup_subsys[] = { |
ddbcc7e8 PM |
131 | #include <linux/cgroup_subsys.h> |
132 | }; | |
073219e9 TH |
133 | #undef SUBSYS |
134 | ||
135 | /* array of cgroup subsystem names */ | |
136 | #define SUBSYS(_x) [_x ## _cgrp_id] = #_x, | |
137 | static const char *cgroup_subsys_name[] = { | |
ddbcc7e8 PM |
138 | #include <linux/cgroup_subsys.h> |
139 | }; | |
073219e9 | 140 | #undef SUBSYS |
ddbcc7e8 | 141 | |
49d1dc4b TH |
142 | /* array of static_keys for cgroup_subsys_enabled() and cgroup_subsys_on_dfl() */ |
143 | #define SUBSYS(_x) \ | |
144 | DEFINE_STATIC_KEY_TRUE(_x ## _cgrp_subsys_enabled_key); \ | |
145 | DEFINE_STATIC_KEY_TRUE(_x ## _cgrp_subsys_on_dfl_key); \ | |
146 | EXPORT_SYMBOL_GPL(_x ## _cgrp_subsys_enabled_key); \ | |
147 | EXPORT_SYMBOL_GPL(_x ## _cgrp_subsys_on_dfl_key); | |
148 | #include <linux/cgroup_subsys.h> | |
149 | #undef SUBSYS | |
150 | ||
151 | #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys_enabled_key, | |
152 | static struct static_key_true *cgroup_subsys_enabled_key[] = { | |
153 | #include <linux/cgroup_subsys.h> | |
154 | }; | |
155 | #undef SUBSYS | |
156 | ||
157 | #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys_on_dfl_key, | |
158 | static struct static_key_true *cgroup_subsys_on_dfl_key[] = { | |
159 | #include <linux/cgroup_subsys.h> | |
160 | }; | |
161 | #undef SUBSYS | |
162 | ||
ddbcc7e8 | 163 | /* |
3dd06ffa | 164 | * The default hierarchy, reserved for the subsystems that are otherwise |
9871bf95 TH |
165 | * unattached - it never has more than a single cgroup, and all tasks are |
166 | * part of that cgroup. | |
ddbcc7e8 | 167 | */ |
a2dd4247 | 168 | struct cgroup_root cgrp_dfl_root; |
d0ec4230 | 169 | EXPORT_SYMBOL_GPL(cgrp_dfl_root); |
9871bf95 | 170 | |
a2dd4247 TH |
171 | /* |
172 | * The default hierarchy always exists but is hidden until mounted for the | |
173 | * first time. This is for backward compatibility. | |
174 | */ | |
175 | static bool cgrp_dfl_root_visible; | |
ddbcc7e8 | 176 | |
a8ddc821 TH |
177 | /* |
178 | * Set by the boot param of the same name and makes subsystems with NULL | |
179 | * ->dfl_files to use ->legacy_files on the default hierarchy. | |
180 | */ | |
181 | static bool cgroup_legacy_files_on_dfl; | |
182 | ||
5533e011 | 183 | /* some controllers are not supported in the default hierarchy */ |
8ab456ac | 184 | static unsigned long cgrp_dfl_root_inhibit_ss_mask; |
5533e011 | 185 | |
ddbcc7e8 PM |
186 | /* The list of hierarchy roots */ |
187 | ||
9871bf95 TH |
188 | static LIST_HEAD(cgroup_roots); |
189 | static int cgroup_root_count; | |
ddbcc7e8 | 190 | |
3417ae1f | 191 | /* hierarchy ID allocation and mapping, protected by cgroup_mutex */ |
1a574231 | 192 | static DEFINE_IDR(cgroup_hierarchy_idr); |
2c6ab6d2 | 193 | |
794611a1 | 194 | /* |
0cb51d71 TH |
195 | * Assign a monotonically increasing serial number to csses. It guarantees |
196 | * cgroups with bigger numbers are newer than those with smaller numbers. | |
197 | * Also, as csses are always appended to the parent's ->children list, it | |
198 | * guarantees that sibling csses are always sorted in the ascending serial | |
199 | * number order on the list. Protected by cgroup_mutex. | |
794611a1 | 200 | */ |
0cb51d71 | 201 | static u64 css_serial_nr_next = 1; |
794611a1 | 202 | |
cb4a3167 AS |
203 | /* |
204 | * These bitmask flags indicate whether tasks in the fork and exit paths have | |
205 | * fork/exit handlers to call. This avoids us having to do extra work in the | |
206 | * fork/exit path to check which subsystems have fork/exit callbacks. | |
ddbcc7e8 | 207 | */ |
cb4a3167 AS |
208 | static unsigned long have_fork_callback __read_mostly; |
209 | static unsigned long have_exit_callback __read_mostly; | |
ddbcc7e8 | 210 | |
7e47682e AS |
211 | /* Ditto for the can_fork callback. */ |
212 | static unsigned long have_canfork_callback __read_mostly; | |
213 | ||
a14c6874 TH |
214 | static struct cftype cgroup_dfl_base_files[]; |
215 | static struct cftype cgroup_legacy_base_files[]; | |
628f7cd4 | 216 | |
3dd06ffa | 217 | static int rebind_subsystems(struct cgroup_root *dst_root, |
8ab456ac | 218 | unsigned long ss_mask); |
42809dd4 | 219 | static int cgroup_destroy_locked(struct cgroup *cgrp); |
f63070d3 TH |
220 | static int create_css(struct cgroup *cgrp, struct cgroup_subsys *ss, |
221 | bool visible); | |
9d755d33 | 222 | static void css_release(struct percpu_ref *ref); |
f8f22e53 | 223 | static void kill_css(struct cgroup_subsys_state *css); |
4df8dc90 TH |
224 | static int cgroup_addrm_files(struct cgroup_subsys_state *css, |
225 | struct cgroup *cgrp, struct cftype cfts[], | |
2bb566cb | 226 | bool is_add); |
42809dd4 | 227 | |
fc5ed1e9 TH |
228 | /** |
229 | * cgroup_ssid_enabled - cgroup subsys enabled test by subsys ID | |
230 | * @ssid: subsys ID of interest | |
231 | * | |
232 | * cgroup_subsys_enabled() can only be used with literal subsys names which | |
233 | * is fine for individual subsystems but unsuitable for cgroup core. This | |
234 | * is slower static_key_enabled() based test indexed by @ssid. | |
235 | */ | |
236 | static bool cgroup_ssid_enabled(int ssid) | |
237 | { | |
238 | return static_key_enabled(cgroup_subsys_enabled_key[ssid]); | |
239 | } | |
240 | ||
9e10a130 TH |
241 | /** |
242 | * cgroup_on_dfl - test whether a cgroup is on the default hierarchy | |
243 | * @cgrp: the cgroup of interest | |
244 | * | |
245 | * The default hierarchy is the v2 interface of cgroup and this function | |
246 | * can be used to test whether a cgroup is on the default hierarchy for | |
247 | * cases where a subsystem should behave differnetly depending on the | |
248 | * interface version. | |
249 | * | |
250 | * The set of behaviors which change on the default hierarchy are still | |
251 | * being determined and the mount option is prefixed with __DEVEL__. | |
252 | * | |
253 | * List of changed behaviors: | |
254 | * | |
255 | * - Mount options "noprefix", "xattr", "clone_children", "release_agent" | |
256 | * and "name" are disallowed. | |
257 | * | |
258 | * - When mounting an existing superblock, mount options should match. | |
259 | * | |
260 | * - Remount is disallowed. | |
261 | * | |
262 | * - rename(2) is disallowed. | |
263 | * | |
264 | * - "tasks" is removed. Everything should be at process granularity. Use | |
265 | * "cgroup.procs" instead. | |
266 | * | |
267 | * - "cgroup.procs" is not sorted. pids will be unique unless they got | |
268 | * recycled inbetween reads. | |
269 | * | |
270 | * - "release_agent" and "notify_on_release" are removed. Replacement | |
271 | * notification mechanism will be implemented. | |
272 | * | |
273 | * - "cgroup.clone_children" is removed. | |
274 | * | |
275 | * - "cgroup.subtree_populated" is available. Its value is 0 if the cgroup | |
276 | * and its descendants contain no task; otherwise, 1. The file also | |
277 | * generates kernfs notification which can be monitored through poll and | |
278 | * [di]notify when the value of the file changes. | |
279 | * | |
280 | * - cpuset: tasks will be kept in empty cpusets when hotplug happens and | |
281 | * take masks of ancestors with non-empty cpus/mems, instead of being | |
282 | * moved to an ancestor. | |
283 | * | |
284 | * - cpuset: a task can be moved into an empty cpuset, and again it takes | |
285 | * masks of ancestors. | |
286 | * | |
287 | * - memcg: use_hierarchy is on by default and the cgroup file for the flag | |
288 | * is not created. | |
289 | * | |
290 | * - blkcg: blk-throttle becomes properly hierarchical. | |
291 | * | |
292 | * - debug: disallowed on the default hierarchy. | |
293 | */ | |
294 | static bool cgroup_on_dfl(const struct cgroup *cgrp) | |
295 | { | |
296 | return cgrp->root == &cgrp_dfl_root; | |
297 | } | |
298 | ||
6fa4918d TH |
299 | /* IDR wrappers which synchronize using cgroup_idr_lock */ |
300 | static int cgroup_idr_alloc(struct idr *idr, void *ptr, int start, int end, | |
301 | gfp_t gfp_mask) | |
302 | { | |
303 | int ret; | |
304 | ||
305 | idr_preload(gfp_mask); | |
54504e97 | 306 | spin_lock_bh(&cgroup_idr_lock); |
cf780b7d | 307 | ret = idr_alloc(idr, ptr, start, end, gfp_mask & ~__GFP_WAIT); |
54504e97 | 308 | spin_unlock_bh(&cgroup_idr_lock); |
6fa4918d TH |
309 | idr_preload_end(); |
310 | return ret; | |
311 | } | |
312 | ||
313 | static void *cgroup_idr_replace(struct idr *idr, void *ptr, int id) | |
314 | { | |
315 | void *ret; | |
316 | ||
54504e97 | 317 | spin_lock_bh(&cgroup_idr_lock); |
6fa4918d | 318 | ret = idr_replace(idr, ptr, id); |
54504e97 | 319 | spin_unlock_bh(&cgroup_idr_lock); |
6fa4918d TH |
320 | return ret; |
321 | } | |
322 | ||
323 | static void cgroup_idr_remove(struct idr *idr, int id) | |
324 | { | |
54504e97 | 325 | spin_lock_bh(&cgroup_idr_lock); |
6fa4918d | 326 | idr_remove(idr, id); |
54504e97 | 327 | spin_unlock_bh(&cgroup_idr_lock); |
6fa4918d TH |
328 | } |
329 | ||
d51f39b0 TH |
330 | static struct cgroup *cgroup_parent(struct cgroup *cgrp) |
331 | { | |
332 | struct cgroup_subsys_state *parent_css = cgrp->self.parent; | |
333 | ||
334 | if (parent_css) | |
335 | return container_of(parent_css, struct cgroup, self); | |
336 | return NULL; | |
337 | } | |
338 | ||
95109b62 TH |
339 | /** |
340 | * cgroup_css - obtain a cgroup's css for the specified subsystem | |
341 | * @cgrp: the cgroup of interest | |
9d800df1 | 342 | * @ss: the subsystem of interest (%NULL returns @cgrp->self) |
95109b62 | 343 | * |
ca8bdcaf TH |
344 | * Return @cgrp's css (cgroup_subsys_state) associated with @ss. This |
345 | * function must be called either under cgroup_mutex or rcu_read_lock() and | |
346 | * the caller is responsible for pinning the returned css if it wants to | |
347 | * keep accessing it outside the said locks. This function may return | |
348 | * %NULL if @cgrp doesn't have @subsys_id enabled. | |
95109b62 TH |
349 | */ |
350 | static struct cgroup_subsys_state *cgroup_css(struct cgroup *cgrp, | |
ca8bdcaf | 351 | struct cgroup_subsys *ss) |
95109b62 | 352 | { |
ca8bdcaf | 353 | if (ss) |
aec25020 | 354 | return rcu_dereference_check(cgrp->subsys[ss->id], |
ace2bee8 | 355 | lockdep_is_held(&cgroup_mutex)); |
ca8bdcaf | 356 | else |
9d800df1 | 357 | return &cgrp->self; |
95109b62 | 358 | } |
42809dd4 | 359 | |
aec3dfcb TH |
360 | /** |
361 | * cgroup_e_css - obtain a cgroup's effective css for the specified subsystem | |
362 | * @cgrp: the cgroup of interest | |
9d800df1 | 363 | * @ss: the subsystem of interest (%NULL returns @cgrp->self) |
aec3dfcb | 364 | * |
d0f702e6 | 365 | * Similar to cgroup_css() but returns the effective css, which is defined |
aec3dfcb TH |
366 | * as the matching css of the nearest ancestor including self which has @ss |
367 | * enabled. If @ss is associated with the hierarchy @cgrp is on, this | |
368 | * function is guaranteed to return non-NULL css. | |
369 | */ | |
370 | static struct cgroup_subsys_state *cgroup_e_css(struct cgroup *cgrp, | |
371 | struct cgroup_subsys *ss) | |
372 | { | |
373 | lockdep_assert_held(&cgroup_mutex); | |
374 | ||
375 | if (!ss) | |
9d800df1 | 376 | return &cgrp->self; |
aec3dfcb TH |
377 | |
378 | if (!(cgrp->root->subsys_mask & (1 << ss->id))) | |
379 | return NULL; | |
380 | ||
eeecbd19 TH |
381 | /* |
382 | * This function is used while updating css associations and thus | |
383 | * can't test the csses directly. Use ->child_subsys_mask. | |
384 | */ | |
d51f39b0 TH |
385 | while (cgroup_parent(cgrp) && |
386 | !(cgroup_parent(cgrp)->child_subsys_mask & (1 << ss->id))) | |
387 | cgrp = cgroup_parent(cgrp); | |
aec3dfcb TH |
388 | |
389 | return cgroup_css(cgrp, ss); | |
95109b62 | 390 | } |
42809dd4 | 391 | |
eeecbd19 TH |
392 | /** |
393 | * cgroup_get_e_css - get a cgroup's effective css for the specified subsystem | |
394 | * @cgrp: the cgroup of interest | |
395 | * @ss: the subsystem of interest | |
396 | * | |
397 | * Find and get the effective css of @cgrp for @ss. The effective css is | |
398 | * defined as the matching css of the nearest ancestor including self which | |
399 | * has @ss enabled. If @ss is not mounted on the hierarchy @cgrp is on, | |
400 | * the root css is returned, so this function always returns a valid css. | |
401 | * The returned css must be put using css_put(). | |
402 | */ | |
403 | struct cgroup_subsys_state *cgroup_get_e_css(struct cgroup *cgrp, | |
404 | struct cgroup_subsys *ss) | |
405 | { | |
406 | struct cgroup_subsys_state *css; | |
407 | ||
408 | rcu_read_lock(); | |
409 | ||
410 | do { | |
411 | css = cgroup_css(cgrp, ss); | |
412 | ||
413 | if (css && css_tryget_online(css)) | |
414 | goto out_unlock; | |
415 | cgrp = cgroup_parent(cgrp); | |
416 | } while (cgrp); | |
417 | ||
418 | css = init_css_set.subsys[ss->id]; | |
419 | css_get(css); | |
420 | out_unlock: | |
421 | rcu_read_unlock(); | |
422 | return css; | |
423 | } | |
424 | ||
ddbcc7e8 | 425 | /* convenient tests for these bits */ |
54766d4a | 426 | static inline bool cgroup_is_dead(const struct cgroup *cgrp) |
ddbcc7e8 | 427 | { |
184faf32 | 428 | return !(cgrp->self.flags & CSS_ONLINE); |
ddbcc7e8 PM |
429 | } |
430 | ||
b4168640 | 431 | struct cgroup_subsys_state *of_css(struct kernfs_open_file *of) |
59f5296b | 432 | { |
2bd59d48 | 433 | struct cgroup *cgrp = of->kn->parent->priv; |
b4168640 | 434 | struct cftype *cft = of_cft(of); |
2bd59d48 TH |
435 | |
436 | /* | |
437 | * This is open and unprotected implementation of cgroup_css(). | |
438 | * seq_css() is only called from a kernfs file operation which has | |
439 | * an active reference on the file. Because all the subsystem | |
440 | * files are drained before a css is disassociated with a cgroup, | |
441 | * the matching css from the cgroup's subsys table is guaranteed to | |
442 | * be and stay valid until the enclosing operation is complete. | |
443 | */ | |
444 | if (cft->ss) | |
445 | return rcu_dereference_raw(cgrp->subsys[cft->ss->id]); | |
446 | else | |
9d800df1 | 447 | return &cgrp->self; |
59f5296b | 448 | } |
b4168640 | 449 | EXPORT_SYMBOL_GPL(of_css); |
59f5296b | 450 | |
78574cf9 LZ |
451 | /** |
452 | * cgroup_is_descendant - test ancestry | |
453 | * @cgrp: the cgroup to be tested | |
454 | * @ancestor: possible ancestor of @cgrp | |
455 | * | |
456 | * Test whether @cgrp is a descendant of @ancestor. It also returns %true | |
457 | * if @cgrp == @ancestor. This function is safe to call as long as @cgrp | |
458 | * and @ancestor are accessible. | |
459 | */ | |
460 | bool cgroup_is_descendant(struct cgroup *cgrp, struct cgroup *ancestor) | |
461 | { | |
462 | while (cgrp) { | |
463 | if (cgrp == ancestor) | |
464 | return true; | |
d51f39b0 | 465 | cgrp = cgroup_parent(cgrp); |
78574cf9 LZ |
466 | } |
467 | return false; | |
468 | } | |
ddbcc7e8 | 469 | |
e9685a03 | 470 | static int notify_on_release(const struct cgroup *cgrp) |
81a6a5cd | 471 | { |
bd89aabc | 472 | return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); |
81a6a5cd PM |
473 | } |
474 | ||
1c6727af TH |
475 | /** |
476 | * for_each_css - iterate all css's of a cgroup | |
477 | * @css: the iteration cursor | |
478 | * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end | |
479 | * @cgrp: the target cgroup to iterate css's of | |
480 | * | |
aec3dfcb | 481 | * Should be called under cgroup_[tree_]mutex. |
1c6727af TH |
482 | */ |
483 | #define for_each_css(css, ssid, cgrp) \ | |
484 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \ | |
485 | if (!((css) = rcu_dereference_check( \ | |
486 | (cgrp)->subsys[(ssid)], \ | |
487 | lockdep_is_held(&cgroup_mutex)))) { } \ | |
488 | else | |
489 | ||
aec3dfcb TH |
490 | /** |
491 | * for_each_e_css - iterate all effective css's of a cgroup | |
492 | * @css: the iteration cursor | |
493 | * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end | |
494 | * @cgrp: the target cgroup to iterate css's of | |
495 | * | |
496 | * Should be called under cgroup_[tree_]mutex. | |
497 | */ | |
498 | #define for_each_e_css(css, ssid, cgrp) \ | |
499 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \ | |
500 | if (!((css) = cgroup_e_css(cgrp, cgroup_subsys[(ssid)]))) \ | |
501 | ; \ | |
502 | else | |
503 | ||
30159ec7 | 504 | /** |
3ed80a62 | 505 | * for_each_subsys - iterate all enabled cgroup subsystems |
30159ec7 | 506 | * @ss: the iteration cursor |
780cd8b3 | 507 | * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end |
30159ec7 | 508 | */ |
780cd8b3 | 509 | #define for_each_subsys(ss, ssid) \ |
3ed80a62 TH |
510 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT && \ |
511 | (((ss) = cgroup_subsys[ssid]) || true); (ssid)++) | |
30159ec7 | 512 | |
cb4a3167 AS |
513 | /** |
514 | * for_each_subsys_which - filter for_each_subsys with a bitmask | |
515 | * @ss: the iteration cursor | |
516 | * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end | |
517 | * @ss_maskp: a pointer to the bitmask | |
518 | * | |
519 | * The block will only run for cases where the ssid-th bit (1 << ssid) of | |
520 | * mask is set to 1. | |
521 | */ | |
522 | #define for_each_subsys_which(ss, ssid, ss_maskp) \ | |
523 | if (!CGROUP_SUBSYS_COUNT) /* to avoid spurious gcc warning */ \ | |
4a705c5c | 524 | (ssid) = 0; \ |
cb4a3167 AS |
525 | else \ |
526 | for_each_set_bit(ssid, ss_maskp, CGROUP_SUBSYS_COUNT) \ | |
527 | if (((ss) = cgroup_subsys[ssid]) && false) \ | |
528 | break; \ | |
529 | else | |
530 | ||
985ed670 TH |
531 | /* iterate across the hierarchies */ |
532 | #define for_each_root(root) \ | |
5549c497 | 533 | list_for_each_entry((root), &cgroup_roots, root_list) |
ddbcc7e8 | 534 | |
f8f22e53 TH |
535 | /* iterate over child cgrps, lock should be held throughout iteration */ |
536 | #define cgroup_for_each_live_child(child, cgrp) \ | |
d5c419b6 | 537 | list_for_each_entry((child), &(cgrp)->self.children, self.sibling) \ |
8353da1f | 538 | if (({ lockdep_assert_held(&cgroup_mutex); \ |
f8f22e53 TH |
539 | cgroup_is_dead(child); })) \ |
540 | ; \ | |
541 | else | |
7ae1bad9 | 542 | |
81a6a5cd | 543 | static void cgroup_release_agent(struct work_struct *work); |
bd89aabc | 544 | static void check_for_release(struct cgroup *cgrp); |
81a6a5cd | 545 | |
69d0206c TH |
546 | /* |
547 | * A cgroup can be associated with multiple css_sets as different tasks may | |
548 | * belong to different cgroups on different hierarchies. In the other | |
549 | * direction, a css_set is naturally associated with multiple cgroups. | |
550 | * This M:N relationship is represented by the following link structure | |
551 | * which exists for each association and allows traversing the associations | |
552 | * from both sides. | |
553 | */ | |
554 | struct cgrp_cset_link { | |
555 | /* the cgroup and css_set this link associates */ | |
556 | struct cgroup *cgrp; | |
557 | struct css_set *cset; | |
558 | ||
559 | /* list of cgrp_cset_links anchored at cgrp->cset_links */ | |
560 | struct list_head cset_link; | |
561 | ||
562 | /* list of cgrp_cset_links anchored at css_set->cgrp_links */ | |
563 | struct list_head cgrp_link; | |
817929ec PM |
564 | }; |
565 | ||
172a2c06 TH |
566 | /* |
567 | * The default css_set - used by init and its children prior to any | |
817929ec PM |
568 | * hierarchies being mounted. It contains a pointer to the root state |
569 | * for each subsystem. Also used to anchor the list of css_sets. Not | |
570 | * reference-counted, to improve performance when child cgroups | |
571 | * haven't been created. | |
572 | */ | |
5024ae29 | 573 | struct css_set init_css_set = { |
172a2c06 TH |
574 | .refcount = ATOMIC_INIT(1), |
575 | .cgrp_links = LIST_HEAD_INIT(init_css_set.cgrp_links), | |
576 | .tasks = LIST_HEAD_INIT(init_css_set.tasks), | |
577 | .mg_tasks = LIST_HEAD_INIT(init_css_set.mg_tasks), | |
578 | .mg_preload_node = LIST_HEAD_INIT(init_css_set.mg_preload_node), | |
579 | .mg_node = LIST_HEAD_INIT(init_css_set.mg_node), | |
580 | }; | |
817929ec | 581 | |
172a2c06 | 582 | static int css_set_count = 1; /* 1 for init_css_set */ |
817929ec | 583 | |
0de0942d TH |
584 | /** |
585 | * css_set_populated - does a css_set contain any tasks? | |
586 | * @cset: target css_set | |
587 | */ | |
588 | static bool css_set_populated(struct css_set *cset) | |
589 | { | |
590 | lockdep_assert_held(&css_set_rwsem); | |
591 | ||
592 | return !list_empty(&cset->tasks) || !list_empty(&cset->mg_tasks); | |
593 | } | |
594 | ||
842b597e TH |
595 | /** |
596 | * cgroup_update_populated - updated populated count of a cgroup | |
597 | * @cgrp: the target cgroup | |
598 | * @populated: inc or dec populated count | |
599 | * | |
0de0942d TH |
600 | * One of the css_sets associated with @cgrp is either getting its first |
601 | * task or losing the last. Update @cgrp->populated_cnt accordingly. The | |
602 | * count is propagated towards root so that a given cgroup's populated_cnt | |
603 | * is zero iff the cgroup and all its descendants don't contain any tasks. | |
842b597e TH |
604 | * |
605 | * @cgrp's interface file "cgroup.populated" is zero if | |
606 | * @cgrp->populated_cnt is zero and 1 otherwise. When @cgrp->populated_cnt | |
607 | * changes from or to zero, userland is notified that the content of the | |
608 | * interface file has changed. This can be used to detect when @cgrp and | |
609 | * its descendants become populated or empty. | |
610 | */ | |
611 | static void cgroup_update_populated(struct cgroup *cgrp, bool populated) | |
612 | { | |
613 | lockdep_assert_held(&css_set_rwsem); | |
614 | ||
615 | do { | |
616 | bool trigger; | |
617 | ||
618 | if (populated) | |
619 | trigger = !cgrp->populated_cnt++; | |
620 | else | |
621 | trigger = !--cgrp->populated_cnt; | |
622 | ||
623 | if (!trigger) | |
624 | break; | |
625 | ||
6f60eade TH |
626 | cgroup_file_notify(&cgrp->events_file); |
627 | ||
d51f39b0 | 628 | cgrp = cgroup_parent(cgrp); |
842b597e TH |
629 | } while (cgrp); |
630 | } | |
631 | ||
0de0942d TH |
632 | /** |
633 | * css_set_update_populated - update populated state of a css_set | |
634 | * @cset: target css_set | |
635 | * @populated: whether @cset is populated or depopulated | |
636 | * | |
637 | * @cset is either getting the first task or losing the last. Update the | |
638 | * ->populated_cnt of all associated cgroups accordingly. | |
639 | */ | |
640 | static void css_set_update_populated(struct css_set *cset, bool populated) | |
641 | { | |
642 | struct cgrp_cset_link *link; | |
643 | ||
644 | lockdep_assert_held(&css_set_rwsem); | |
645 | ||
646 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) | |
647 | cgroup_update_populated(link->cgrp, populated); | |
648 | } | |
649 | ||
7717f7ba PM |
650 | /* |
651 | * hash table for cgroup groups. This improves the performance to find | |
652 | * an existing css_set. This hash doesn't (currently) take into | |
653 | * account cgroups in empty hierarchies. | |
654 | */ | |
472b1053 | 655 | #define CSS_SET_HASH_BITS 7 |
0ac801fe | 656 | static DEFINE_HASHTABLE(css_set_table, CSS_SET_HASH_BITS); |
472b1053 | 657 | |
0ac801fe | 658 | static unsigned long css_set_hash(struct cgroup_subsys_state *css[]) |
472b1053 | 659 | { |
0ac801fe | 660 | unsigned long key = 0UL; |
30159ec7 TH |
661 | struct cgroup_subsys *ss; |
662 | int i; | |
472b1053 | 663 | |
30159ec7 | 664 | for_each_subsys(ss, i) |
0ac801fe LZ |
665 | key += (unsigned long)css[i]; |
666 | key = (key >> 16) ^ key; | |
472b1053 | 667 | |
0ac801fe | 668 | return key; |
472b1053 LZ |
669 | } |
670 | ||
a25eb52e | 671 | static void put_css_set_locked(struct css_set *cset) |
b4f48b63 | 672 | { |
69d0206c | 673 | struct cgrp_cset_link *link, *tmp_link; |
2d8f243a TH |
674 | struct cgroup_subsys *ss; |
675 | int ssid; | |
5abb8855 | 676 | |
89c5509b TH |
677 | lockdep_assert_held(&css_set_rwsem); |
678 | ||
679 | if (!atomic_dec_and_test(&cset->refcount)) | |
146aa1bd | 680 | return; |
81a6a5cd | 681 | |
2c6ab6d2 | 682 | /* This css_set is dead. unlink it and release cgroup refcounts */ |
2d8f243a TH |
683 | for_each_subsys(ss, ssid) |
684 | list_del(&cset->e_cset_node[ssid]); | |
5abb8855 | 685 | hash_del(&cset->hlist); |
2c6ab6d2 PM |
686 | css_set_count--; |
687 | ||
69d0206c | 688 | list_for_each_entry_safe(link, tmp_link, &cset->cgrp_links, cgrp_link) { |
2c6ab6d2 | 689 | struct cgroup *cgrp = link->cgrp; |
5abb8855 | 690 | |
69d0206c TH |
691 | list_del(&link->cset_link); |
692 | list_del(&link->cgrp_link); | |
71b5707e | 693 | |
96d365e0 | 694 | /* @cgrp can't go away while we're holding css_set_rwsem */ |
0de0942d | 695 | if (list_empty(&cgrp->cset_links)) |
a25eb52e | 696 | check_for_release(cgrp); |
2c6ab6d2 PM |
697 | |
698 | kfree(link); | |
81a6a5cd | 699 | } |
2c6ab6d2 | 700 | |
5abb8855 | 701 | kfree_rcu(cset, rcu_head); |
b4f48b63 PM |
702 | } |
703 | ||
a25eb52e | 704 | static void put_css_set(struct css_set *cset) |
89c5509b TH |
705 | { |
706 | /* | |
707 | * Ensure that the refcount doesn't hit zero while any readers | |
708 | * can see it. Similar to atomic_dec_and_lock(), but for an | |
709 | * rwlock | |
710 | */ | |
711 | if (atomic_add_unless(&cset->refcount, -1, 1)) | |
712 | return; | |
713 | ||
714 | down_write(&css_set_rwsem); | |
a25eb52e | 715 | put_css_set_locked(cset); |
89c5509b TH |
716 | up_write(&css_set_rwsem); |
717 | } | |
718 | ||
817929ec PM |
719 | /* |
720 | * refcounted get/put for css_set objects | |
721 | */ | |
5abb8855 | 722 | static inline void get_css_set(struct css_set *cset) |
817929ec | 723 | { |
5abb8855 | 724 | atomic_inc(&cset->refcount); |
817929ec PM |
725 | } |
726 | ||
b326f9d0 | 727 | /** |
7717f7ba | 728 | * compare_css_sets - helper function for find_existing_css_set(). |
5abb8855 TH |
729 | * @cset: candidate css_set being tested |
730 | * @old_cset: existing css_set for a task | |
7717f7ba PM |
731 | * @new_cgrp: cgroup that's being entered by the task |
732 | * @template: desired set of css pointers in css_set (pre-calculated) | |
733 | * | |
6f4b7e63 | 734 | * Returns true if "cset" matches "old_cset" except for the hierarchy |
7717f7ba PM |
735 | * which "new_cgrp" belongs to, for which it should match "new_cgrp". |
736 | */ | |
5abb8855 TH |
737 | static bool compare_css_sets(struct css_set *cset, |
738 | struct css_set *old_cset, | |
7717f7ba PM |
739 | struct cgroup *new_cgrp, |
740 | struct cgroup_subsys_state *template[]) | |
741 | { | |
742 | struct list_head *l1, *l2; | |
743 | ||
aec3dfcb TH |
744 | /* |
745 | * On the default hierarchy, there can be csets which are | |
746 | * associated with the same set of cgroups but different csses. | |
747 | * Let's first ensure that csses match. | |
748 | */ | |
749 | if (memcmp(template, cset->subsys, sizeof(cset->subsys))) | |
7717f7ba | 750 | return false; |
7717f7ba PM |
751 | |
752 | /* | |
753 | * Compare cgroup pointers in order to distinguish between | |
aec3dfcb TH |
754 | * different cgroups in hierarchies. As different cgroups may |
755 | * share the same effective css, this comparison is always | |
756 | * necessary. | |
7717f7ba | 757 | */ |
69d0206c TH |
758 | l1 = &cset->cgrp_links; |
759 | l2 = &old_cset->cgrp_links; | |
7717f7ba | 760 | while (1) { |
69d0206c | 761 | struct cgrp_cset_link *link1, *link2; |
5abb8855 | 762 | struct cgroup *cgrp1, *cgrp2; |
7717f7ba PM |
763 | |
764 | l1 = l1->next; | |
765 | l2 = l2->next; | |
766 | /* See if we reached the end - both lists are equal length. */ | |
69d0206c TH |
767 | if (l1 == &cset->cgrp_links) { |
768 | BUG_ON(l2 != &old_cset->cgrp_links); | |
7717f7ba PM |
769 | break; |
770 | } else { | |
69d0206c | 771 | BUG_ON(l2 == &old_cset->cgrp_links); |
7717f7ba PM |
772 | } |
773 | /* Locate the cgroups associated with these links. */ | |
69d0206c TH |
774 | link1 = list_entry(l1, struct cgrp_cset_link, cgrp_link); |
775 | link2 = list_entry(l2, struct cgrp_cset_link, cgrp_link); | |
776 | cgrp1 = link1->cgrp; | |
777 | cgrp2 = link2->cgrp; | |
7717f7ba | 778 | /* Hierarchies should be linked in the same order. */ |
5abb8855 | 779 | BUG_ON(cgrp1->root != cgrp2->root); |
7717f7ba PM |
780 | |
781 | /* | |
782 | * If this hierarchy is the hierarchy of the cgroup | |
783 | * that's changing, then we need to check that this | |
784 | * css_set points to the new cgroup; if it's any other | |
785 | * hierarchy, then this css_set should point to the | |
786 | * same cgroup as the old css_set. | |
787 | */ | |
5abb8855 TH |
788 | if (cgrp1->root == new_cgrp->root) { |
789 | if (cgrp1 != new_cgrp) | |
7717f7ba PM |
790 | return false; |
791 | } else { | |
5abb8855 | 792 | if (cgrp1 != cgrp2) |
7717f7ba PM |
793 | return false; |
794 | } | |
795 | } | |
796 | return true; | |
797 | } | |
798 | ||
b326f9d0 TH |
799 | /** |
800 | * find_existing_css_set - init css array and find the matching css_set | |
801 | * @old_cset: the css_set that we're using before the cgroup transition | |
802 | * @cgrp: the cgroup that we're moving into | |
803 | * @template: out param for the new set of csses, should be clear on entry | |
817929ec | 804 | */ |
5abb8855 TH |
805 | static struct css_set *find_existing_css_set(struct css_set *old_cset, |
806 | struct cgroup *cgrp, | |
807 | struct cgroup_subsys_state *template[]) | |
b4f48b63 | 808 | { |
3dd06ffa | 809 | struct cgroup_root *root = cgrp->root; |
30159ec7 | 810 | struct cgroup_subsys *ss; |
5abb8855 | 811 | struct css_set *cset; |
0ac801fe | 812 | unsigned long key; |
b326f9d0 | 813 | int i; |
817929ec | 814 | |
aae8aab4 BB |
815 | /* |
816 | * Build the set of subsystem state objects that we want to see in the | |
817 | * new css_set. while subsystems can change globally, the entries here | |
818 | * won't change, so no need for locking. | |
819 | */ | |
30159ec7 | 820 | for_each_subsys(ss, i) { |
f392e51c | 821 | if (root->subsys_mask & (1UL << i)) { |
aec3dfcb TH |
822 | /* |
823 | * @ss is in this hierarchy, so we want the | |
824 | * effective css from @cgrp. | |
825 | */ | |
826 | template[i] = cgroup_e_css(cgrp, ss); | |
817929ec | 827 | } else { |
aec3dfcb TH |
828 | /* |
829 | * @ss is not in this hierarchy, so we don't want | |
830 | * to change the css. | |
831 | */ | |
5abb8855 | 832 | template[i] = old_cset->subsys[i]; |
817929ec PM |
833 | } |
834 | } | |
835 | ||
0ac801fe | 836 | key = css_set_hash(template); |
5abb8855 TH |
837 | hash_for_each_possible(css_set_table, cset, hlist, key) { |
838 | if (!compare_css_sets(cset, old_cset, cgrp, template)) | |
7717f7ba PM |
839 | continue; |
840 | ||
841 | /* This css_set matches what we need */ | |
5abb8855 | 842 | return cset; |
472b1053 | 843 | } |
817929ec PM |
844 | |
845 | /* No existing cgroup group matched */ | |
846 | return NULL; | |
847 | } | |
848 | ||
69d0206c | 849 | static void free_cgrp_cset_links(struct list_head *links_to_free) |
36553434 | 850 | { |
69d0206c | 851 | struct cgrp_cset_link *link, *tmp_link; |
36553434 | 852 | |
69d0206c TH |
853 | list_for_each_entry_safe(link, tmp_link, links_to_free, cset_link) { |
854 | list_del(&link->cset_link); | |
36553434 LZ |
855 | kfree(link); |
856 | } | |
857 | } | |
858 | ||
69d0206c TH |
859 | /** |
860 | * allocate_cgrp_cset_links - allocate cgrp_cset_links | |
861 | * @count: the number of links to allocate | |
862 | * @tmp_links: list_head the allocated links are put on | |
863 | * | |
864 | * Allocate @count cgrp_cset_link structures and chain them on @tmp_links | |
865 | * through ->cset_link. Returns 0 on success or -errno. | |
817929ec | 866 | */ |
69d0206c | 867 | static int allocate_cgrp_cset_links(int count, struct list_head *tmp_links) |
817929ec | 868 | { |
69d0206c | 869 | struct cgrp_cset_link *link; |
817929ec | 870 | int i; |
69d0206c TH |
871 | |
872 | INIT_LIST_HEAD(tmp_links); | |
873 | ||
817929ec | 874 | for (i = 0; i < count; i++) { |
f4f4be2b | 875 | link = kzalloc(sizeof(*link), GFP_KERNEL); |
817929ec | 876 | if (!link) { |
69d0206c | 877 | free_cgrp_cset_links(tmp_links); |
817929ec PM |
878 | return -ENOMEM; |
879 | } | |
69d0206c | 880 | list_add(&link->cset_link, tmp_links); |
817929ec PM |
881 | } |
882 | return 0; | |
883 | } | |
884 | ||
c12f65d4 LZ |
885 | /** |
886 | * link_css_set - a helper function to link a css_set to a cgroup | |
69d0206c | 887 | * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links() |
5abb8855 | 888 | * @cset: the css_set to be linked |
c12f65d4 LZ |
889 | * @cgrp: the destination cgroup |
890 | */ | |
69d0206c TH |
891 | static void link_css_set(struct list_head *tmp_links, struct css_set *cset, |
892 | struct cgroup *cgrp) | |
c12f65d4 | 893 | { |
69d0206c | 894 | struct cgrp_cset_link *link; |
c12f65d4 | 895 | |
69d0206c | 896 | BUG_ON(list_empty(tmp_links)); |
6803c006 TH |
897 | |
898 | if (cgroup_on_dfl(cgrp)) | |
899 | cset->dfl_cgrp = cgrp; | |
900 | ||
69d0206c TH |
901 | link = list_first_entry(tmp_links, struct cgrp_cset_link, cset_link); |
902 | link->cset = cset; | |
7717f7ba | 903 | link->cgrp = cgrp; |
842b597e | 904 | |
69d0206c | 905 | list_move(&link->cset_link, &cgrp->cset_links); |
842b597e | 906 | |
7717f7ba PM |
907 | /* |
908 | * Always add links to the tail of the list so that the list | |
909 | * is sorted by order of hierarchy creation | |
910 | */ | |
69d0206c | 911 | list_add_tail(&link->cgrp_link, &cset->cgrp_links); |
c12f65d4 LZ |
912 | } |
913 | ||
b326f9d0 TH |
914 | /** |
915 | * find_css_set - return a new css_set with one cgroup updated | |
916 | * @old_cset: the baseline css_set | |
917 | * @cgrp: the cgroup to be updated | |
918 | * | |
919 | * Return a new css_set that's equivalent to @old_cset, but with @cgrp | |
920 | * substituted into the appropriate hierarchy. | |
817929ec | 921 | */ |
5abb8855 TH |
922 | static struct css_set *find_css_set(struct css_set *old_cset, |
923 | struct cgroup *cgrp) | |
817929ec | 924 | { |
b326f9d0 | 925 | struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT] = { }; |
5abb8855 | 926 | struct css_set *cset; |
69d0206c TH |
927 | struct list_head tmp_links; |
928 | struct cgrp_cset_link *link; | |
2d8f243a | 929 | struct cgroup_subsys *ss; |
0ac801fe | 930 | unsigned long key; |
2d8f243a | 931 | int ssid; |
472b1053 | 932 | |
b326f9d0 TH |
933 | lockdep_assert_held(&cgroup_mutex); |
934 | ||
817929ec PM |
935 | /* First see if we already have a cgroup group that matches |
936 | * the desired set */ | |
96d365e0 | 937 | down_read(&css_set_rwsem); |
5abb8855 TH |
938 | cset = find_existing_css_set(old_cset, cgrp, template); |
939 | if (cset) | |
940 | get_css_set(cset); | |
96d365e0 | 941 | up_read(&css_set_rwsem); |
817929ec | 942 | |
5abb8855 TH |
943 | if (cset) |
944 | return cset; | |
817929ec | 945 | |
f4f4be2b | 946 | cset = kzalloc(sizeof(*cset), GFP_KERNEL); |
5abb8855 | 947 | if (!cset) |
817929ec PM |
948 | return NULL; |
949 | ||
69d0206c | 950 | /* Allocate all the cgrp_cset_link objects that we'll need */ |
9871bf95 | 951 | if (allocate_cgrp_cset_links(cgroup_root_count, &tmp_links) < 0) { |
5abb8855 | 952 | kfree(cset); |
817929ec PM |
953 | return NULL; |
954 | } | |
955 | ||
5abb8855 | 956 | atomic_set(&cset->refcount, 1); |
69d0206c | 957 | INIT_LIST_HEAD(&cset->cgrp_links); |
5abb8855 | 958 | INIT_LIST_HEAD(&cset->tasks); |
c7561128 | 959 | INIT_LIST_HEAD(&cset->mg_tasks); |
1958d2d5 | 960 | INIT_LIST_HEAD(&cset->mg_preload_node); |
b3dc094e | 961 | INIT_LIST_HEAD(&cset->mg_node); |
5abb8855 | 962 | INIT_HLIST_NODE(&cset->hlist); |
817929ec PM |
963 | |
964 | /* Copy the set of subsystem state objects generated in | |
965 | * find_existing_css_set() */ | |
5abb8855 | 966 | memcpy(cset->subsys, template, sizeof(cset->subsys)); |
817929ec | 967 | |
96d365e0 | 968 | down_write(&css_set_rwsem); |
817929ec | 969 | /* Add reference counts and links from the new css_set. */ |
69d0206c | 970 | list_for_each_entry(link, &old_cset->cgrp_links, cgrp_link) { |
7717f7ba | 971 | struct cgroup *c = link->cgrp; |
69d0206c | 972 | |
7717f7ba PM |
973 | if (c->root == cgrp->root) |
974 | c = cgrp; | |
69d0206c | 975 | link_css_set(&tmp_links, cset, c); |
7717f7ba | 976 | } |
817929ec | 977 | |
69d0206c | 978 | BUG_ON(!list_empty(&tmp_links)); |
817929ec | 979 | |
817929ec | 980 | css_set_count++; |
472b1053 | 981 | |
2d8f243a | 982 | /* Add @cset to the hash table */ |
5abb8855 TH |
983 | key = css_set_hash(cset->subsys); |
984 | hash_add(css_set_table, &cset->hlist, key); | |
472b1053 | 985 | |
2d8f243a TH |
986 | for_each_subsys(ss, ssid) |
987 | list_add_tail(&cset->e_cset_node[ssid], | |
988 | &cset->subsys[ssid]->cgroup->e_csets[ssid]); | |
989 | ||
96d365e0 | 990 | up_write(&css_set_rwsem); |
817929ec | 991 | |
5abb8855 | 992 | return cset; |
b4f48b63 PM |
993 | } |
994 | ||
3dd06ffa | 995 | static struct cgroup_root *cgroup_root_from_kf(struct kernfs_root *kf_root) |
7717f7ba | 996 | { |
3dd06ffa | 997 | struct cgroup *root_cgrp = kf_root->kn->priv; |
2bd59d48 | 998 | |
3dd06ffa | 999 | return root_cgrp->root; |
2bd59d48 TH |
1000 | } |
1001 | ||
3dd06ffa | 1002 | static int cgroup_init_root_id(struct cgroup_root *root) |
f2e85d57 TH |
1003 | { |
1004 | int id; | |
1005 | ||
1006 | lockdep_assert_held(&cgroup_mutex); | |
1007 | ||
985ed670 | 1008 | id = idr_alloc_cyclic(&cgroup_hierarchy_idr, root, 0, 0, GFP_KERNEL); |
f2e85d57 TH |
1009 | if (id < 0) |
1010 | return id; | |
1011 | ||
1012 | root->hierarchy_id = id; | |
1013 | return 0; | |
1014 | } | |
1015 | ||
3dd06ffa | 1016 | static void cgroup_exit_root_id(struct cgroup_root *root) |
f2e85d57 TH |
1017 | { |
1018 | lockdep_assert_held(&cgroup_mutex); | |
1019 | ||
1020 | if (root->hierarchy_id) { | |
1021 | idr_remove(&cgroup_hierarchy_idr, root->hierarchy_id); | |
1022 | root->hierarchy_id = 0; | |
1023 | } | |
1024 | } | |
1025 | ||
3dd06ffa | 1026 | static void cgroup_free_root(struct cgroup_root *root) |
f2e85d57 TH |
1027 | { |
1028 | if (root) { | |
d0f702e6 | 1029 | /* hierarchy ID should already have been released */ |
f2e85d57 TH |
1030 | WARN_ON_ONCE(root->hierarchy_id); |
1031 | ||
1032 | idr_destroy(&root->cgroup_idr); | |
1033 | kfree(root); | |
1034 | } | |
1035 | } | |
1036 | ||
3dd06ffa | 1037 | static void cgroup_destroy_root(struct cgroup_root *root) |
59f5296b | 1038 | { |
3dd06ffa | 1039 | struct cgroup *cgrp = &root->cgrp; |
f2e85d57 | 1040 | struct cgrp_cset_link *link, *tmp_link; |
f2e85d57 | 1041 | |
2bd59d48 | 1042 | mutex_lock(&cgroup_mutex); |
f2e85d57 | 1043 | |
776f02fa | 1044 | BUG_ON(atomic_read(&root->nr_cgrps)); |
d5c419b6 | 1045 | BUG_ON(!list_empty(&cgrp->self.children)); |
f2e85d57 | 1046 | |
f2e85d57 | 1047 | /* Rebind all subsystems back to the default hierarchy */ |
f392e51c | 1048 | rebind_subsystems(&cgrp_dfl_root, root->subsys_mask); |
7717f7ba | 1049 | |
7717f7ba | 1050 | /* |
f2e85d57 TH |
1051 | * Release all the links from cset_links to this hierarchy's |
1052 | * root cgroup | |
7717f7ba | 1053 | */ |
96d365e0 | 1054 | down_write(&css_set_rwsem); |
f2e85d57 TH |
1055 | |
1056 | list_for_each_entry_safe(link, tmp_link, &cgrp->cset_links, cset_link) { | |
1057 | list_del(&link->cset_link); | |
1058 | list_del(&link->cgrp_link); | |
1059 | kfree(link); | |
1060 | } | |
96d365e0 | 1061 | up_write(&css_set_rwsem); |
f2e85d57 TH |
1062 | |
1063 | if (!list_empty(&root->root_list)) { | |
1064 | list_del(&root->root_list); | |
1065 | cgroup_root_count--; | |
1066 | } | |
1067 | ||
1068 | cgroup_exit_root_id(root); | |
1069 | ||
1070 | mutex_unlock(&cgroup_mutex); | |
f2e85d57 | 1071 | |
2bd59d48 | 1072 | kernfs_destroy_root(root->kf_root); |
f2e85d57 TH |
1073 | cgroup_free_root(root); |
1074 | } | |
1075 | ||
ceb6a081 TH |
1076 | /* look up cgroup associated with given css_set on the specified hierarchy */ |
1077 | static struct cgroup *cset_cgroup_from_root(struct css_set *cset, | |
3dd06ffa | 1078 | struct cgroup_root *root) |
7717f7ba | 1079 | { |
7717f7ba PM |
1080 | struct cgroup *res = NULL; |
1081 | ||
96d365e0 TH |
1082 | lockdep_assert_held(&cgroup_mutex); |
1083 | lockdep_assert_held(&css_set_rwsem); | |
1084 | ||
5abb8855 | 1085 | if (cset == &init_css_set) { |
3dd06ffa | 1086 | res = &root->cgrp; |
7717f7ba | 1087 | } else { |
69d0206c TH |
1088 | struct cgrp_cset_link *link; |
1089 | ||
1090 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) { | |
7717f7ba | 1091 | struct cgroup *c = link->cgrp; |
69d0206c | 1092 | |
7717f7ba PM |
1093 | if (c->root == root) { |
1094 | res = c; | |
1095 | break; | |
1096 | } | |
1097 | } | |
1098 | } | |
96d365e0 | 1099 | |
7717f7ba PM |
1100 | BUG_ON(!res); |
1101 | return res; | |
1102 | } | |
1103 | ||
ddbcc7e8 | 1104 | /* |
ceb6a081 TH |
1105 | * Return the cgroup for "task" from the given hierarchy. Must be |
1106 | * called with cgroup_mutex and css_set_rwsem held. | |
1107 | */ | |
1108 | static struct cgroup *task_cgroup_from_root(struct task_struct *task, | |
3dd06ffa | 1109 | struct cgroup_root *root) |
ceb6a081 TH |
1110 | { |
1111 | /* | |
1112 | * No need to lock the task - since we hold cgroup_mutex the | |
1113 | * task can't change groups, so the only thing that can happen | |
1114 | * is that it exits and its css is set back to init_css_set. | |
1115 | */ | |
1116 | return cset_cgroup_from_root(task_css_set(task), root); | |
1117 | } | |
1118 | ||
ddbcc7e8 | 1119 | /* |
ddbcc7e8 PM |
1120 | * A task must hold cgroup_mutex to modify cgroups. |
1121 | * | |
1122 | * Any task can increment and decrement the count field without lock. | |
1123 | * So in general, code holding cgroup_mutex can't rely on the count | |
1124 | * field not changing. However, if the count goes to zero, then only | |
956db3ca | 1125 | * cgroup_attach_task() can increment it again. Because a count of zero |
ddbcc7e8 PM |
1126 | * means that no tasks are currently attached, therefore there is no |
1127 | * way a task attached to that cgroup can fork (the other way to | |
1128 | * increment the count). So code holding cgroup_mutex can safely | |
1129 | * assume that if the count is zero, it will stay zero. Similarly, if | |
1130 | * a task holds cgroup_mutex on a cgroup with zero count, it | |
1131 | * knows that the cgroup won't be removed, as cgroup_rmdir() | |
1132 | * needs that mutex. | |
1133 | * | |
ddbcc7e8 PM |
1134 | * A cgroup can only be deleted if both its 'count' of using tasks |
1135 | * is zero, and its list of 'children' cgroups is empty. Since all | |
1136 | * tasks in the system use _some_ cgroup, and since there is always at | |
3dd06ffa | 1137 | * least one task in the system (init, pid == 1), therefore, root cgroup |
ddbcc7e8 | 1138 | * always has either children cgroups and/or using tasks. So we don't |
3dd06ffa | 1139 | * need a special hack to ensure that root cgroup cannot be deleted. |
ddbcc7e8 PM |
1140 | * |
1141 | * P.S. One more locking exception. RCU is used to guard the | |
956db3ca | 1142 | * update of a tasks cgroup pointer by cgroup_attach_task() |
ddbcc7e8 PM |
1143 | */ |
1144 | ||
2bd59d48 | 1145 | static struct kernfs_syscall_ops cgroup_kf_syscall_ops; |
828c0950 | 1146 | static const struct file_operations proc_cgroupstats_operations; |
a424316c | 1147 | |
8d7e6fb0 TH |
1148 | static char *cgroup_file_name(struct cgroup *cgrp, const struct cftype *cft, |
1149 | char *buf) | |
ddbcc7e8 | 1150 | { |
3e1d2eed TH |
1151 | struct cgroup_subsys *ss = cft->ss; |
1152 | ||
8d7e6fb0 TH |
1153 | if (cft->ss && !(cft->flags & CFTYPE_NO_PREFIX) && |
1154 | !(cgrp->root->flags & CGRP_ROOT_NOPREFIX)) | |
1155 | snprintf(buf, CGROUP_FILE_NAME_MAX, "%s.%s", | |
3e1d2eed TH |
1156 | cgroup_on_dfl(cgrp) ? ss->name : ss->legacy_name, |
1157 | cft->name); | |
8d7e6fb0 TH |
1158 | else |
1159 | strncpy(buf, cft->name, CGROUP_FILE_NAME_MAX); | |
1160 | return buf; | |
ddbcc7e8 PM |
1161 | } |
1162 | ||
f2e85d57 TH |
1163 | /** |
1164 | * cgroup_file_mode - deduce file mode of a control file | |
1165 | * @cft: the control file in question | |
1166 | * | |
7dbdb199 | 1167 | * S_IRUGO for read, S_IWUSR for write. |
f2e85d57 TH |
1168 | */ |
1169 | static umode_t cgroup_file_mode(const struct cftype *cft) | |
65dff759 | 1170 | { |
f2e85d57 | 1171 | umode_t mode = 0; |
65dff759 | 1172 | |
f2e85d57 TH |
1173 | if (cft->read_u64 || cft->read_s64 || cft->seq_show) |
1174 | mode |= S_IRUGO; | |
1175 | ||
7dbdb199 TH |
1176 | if (cft->write_u64 || cft->write_s64 || cft->write) { |
1177 | if (cft->flags & CFTYPE_WORLD_WRITABLE) | |
1178 | mode |= S_IWUGO; | |
1179 | else | |
1180 | mode |= S_IWUSR; | |
1181 | } | |
f2e85d57 TH |
1182 | |
1183 | return mode; | |
65dff759 LZ |
1184 | } |
1185 | ||
59f5296b | 1186 | static void cgroup_get(struct cgroup *cgrp) |
be445626 | 1187 | { |
2bd59d48 | 1188 | WARN_ON_ONCE(cgroup_is_dead(cgrp)); |
9d755d33 | 1189 | css_get(&cgrp->self); |
be445626 LZ |
1190 | } |
1191 | ||
aa32362f LZ |
1192 | static bool cgroup_tryget(struct cgroup *cgrp) |
1193 | { | |
1194 | return css_tryget(&cgrp->self); | |
1195 | } | |
1196 | ||
59f5296b | 1197 | static void cgroup_put(struct cgroup *cgrp) |
be445626 | 1198 | { |
9d755d33 | 1199 | css_put(&cgrp->self); |
be445626 LZ |
1200 | } |
1201 | ||
af0ba678 | 1202 | /** |
0f060deb | 1203 | * cgroup_calc_child_subsys_mask - calculate child_subsys_mask |
af0ba678 | 1204 | * @cgrp: the target cgroup |
0f060deb | 1205 | * @subtree_control: the new subtree_control mask to consider |
af0ba678 TH |
1206 | * |
1207 | * On the default hierarchy, a subsystem may request other subsystems to be | |
1208 | * enabled together through its ->depends_on mask. In such cases, more | |
1209 | * subsystems than specified in "cgroup.subtree_control" may be enabled. | |
1210 | * | |
0f060deb TH |
1211 | * This function calculates which subsystems need to be enabled if |
1212 | * @subtree_control is to be applied to @cgrp. The returned mask is always | |
1213 | * a superset of @subtree_control and follows the usual hierarchy rules. | |
af0ba678 | 1214 | */ |
8ab456ac AS |
1215 | static unsigned long cgroup_calc_child_subsys_mask(struct cgroup *cgrp, |
1216 | unsigned long subtree_control) | |
667c2491 | 1217 | { |
af0ba678 | 1218 | struct cgroup *parent = cgroup_parent(cgrp); |
8ab456ac | 1219 | unsigned long cur_ss_mask = subtree_control; |
af0ba678 TH |
1220 | struct cgroup_subsys *ss; |
1221 | int ssid; | |
1222 | ||
1223 | lockdep_assert_held(&cgroup_mutex); | |
1224 | ||
0f060deb TH |
1225 | if (!cgroup_on_dfl(cgrp)) |
1226 | return cur_ss_mask; | |
af0ba678 TH |
1227 | |
1228 | while (true) { | |
8ab456ac | 1229 | unsigned long new_ss_mask = cur_ss_mask; |
af0ba678 | 1230 | |
a966a4ed AS |
1231 | for_each_subsys_which(ss, ssid, &cur_ss_mask) |
1232 | new_ss_mask |= ss->depends_on; | |
af0ba678 TH |
1233 | |
1234 | /* | |
1235 | * Mask out subsystems which aren't available. This can | |
1236 | * happen only if some depended-upon subsystems were bound | |
1237 | * to non-default hierarchies. | |
1238 | */ | |
1239 | if (parent) | |
1240 | new_ss_mask &= parent->child_subsys_mask; | |
1241 | else | |
1242 | new_ss_mask &= cgrp->root->subsys_mask; | |
1243 | ||
1244 | if (new_ss_mask == cur_ss_mask) | |
1245 | break; | |
1246 | cur_ss_mask = new_ss_mask; | |
1247 | } | |
1248 | ||
0f060deb TH |
1249 | return cur_ss_mask; |
1250 | } | |
1251 | ||
1252 | /** | |
1253 | * cgroup_refresh_child_subsys_mask - update child_subsys_mask | |
1254 | * @cgrp: the target cgroup | |
1255 | * | |
1256 | * Update @cgrp->child_subsys_mask according to the current | |
1257 | * @cgrp->subtree_control using cgroup_calc_child_subsys_mask(). | |
1258 | */ | |
1259 | static void cgroup_refresh_child_subsys_mask(struct cgroup *cgrp) | |
1260 | { | |
1261 | cgrp->child_subsys_mask = | |
1262 | cgroup_calc_child_subsys_mask(cgrp, cgrp->subtree_control); | |
667c2491 TH |
1263 | } |
1264 | ||
a9746d8d TH |
1265 | /** |
1266 | * cgroup_kn_unlock - unlocking helper for cgroup kernfs methods | |
1267 | * @kn: the kernfs_node being serviced | |
1268 | * | |
1269 | * This helper undoes cgroup_kn_lock_live() and should be invoked before | |
1270 | * the method finishes if locking succeeded. Note that once this function | |
1271 | * returns the cgroup returned by cgroup_kn_lock_live() may become | |
1272 | * inaccessible any time. If the caller intends to continue to access the | |
1273 | * cgroup, it should pin it before invoking this function. | |
1274 | */ | |
1275 | static void cgroup_kn_unlock(struct kernfs_node *kn) | |
ddbcc7e8 | 1276 | { |
a9746d8d TH |
1277 | struct cgroup *cgrp; |
1278 | ||
1279 | if (kernfs_type(kn) == KERNFS_DIR) | |
1280 | cgrp = kn->priv; | |
1281 | else | |
1282 | cgrp = kn->parent->priv; | |
1283 | ||
1284 | mutex_unlock(&cgroup_mutex); | |
a9746d8d TH |
1285 | |
1286 | kernfs_unbreak_active_protection(kn); | |
1287 | cgroup_put(cgrp); | |
ddbcc7e8 PM |
1288 | } |
1289 | ||
a9746d8d TH |
1290 | /** |
1291 | * cgroup_kn_lock_live - locking helper for cgroup kernfs methods | |
1292 | * @kn: the kernfs_node being serviced | |
1293 | * | |
1294 | * This helper is to be used by a cgroup kernfs method currently servicing | |
1295 | * @kn. It breaks the active protection, performs cgroup locking and | |
1296 | * verifies that the associated cgroup is alive. Returns the cgroup if | |
1297 | * alive; otherwise, %NULL. A successful return should be undone by a | |
1298 | * matching cgroup_kn_unlock() invocation. | |
1299 | * | |
1300 | * Any cgroup kernfs method implementation which requires locking the | |
1301 | * associated cgroup should use this helper. It avoids nesting cgroup | |
1302 | * locking under kernfs active protection and allows all kernfs operations | |
1303 | * including self-removal. | |
1304 | */ | |
1305 | static struct cgroup *cgroup_kn_lock_live(struct kernfs_node *kn) | |
05ef1d7c | 1306 | { |
a9746d8d TH |
1307 | struct cgroup *cgrp; |
1308 | ||
1309 | if (kernfs_type(kn) == KERNFS_DIR) | |
1310 | cgrp = kn->priv; | |
1311 | else | |
1312 | cgrp = kn->parent->priv; | |
05ef1d7c | 1313 | |
2739d3cc | 1314 | /* |
01f6474c | 1315 | * We're gonna grab cgroup_mutex which nests outside kernfs |
a9746d8d TH |
1316 | * active_ref. cgroup liveliness check alone provides enough |
1317 | * protection against removal. Ensure @cgrp stays accessible and | |
1318 | * break the active_ref protection. | |
2739d3cc | 1319 | */ |
aa32362f LZ |
1320 | if (!cgroup_tryget(cgrp)) |
1321 | return NULL; | |
a9746d8d TH |
1322 | kernfs_break_active_protection(kn); |
1323 | ||
2bd59d48 | 1324 | mutex_lock(&cgroup_mutex); |
05ef1d7c | 1325 | |
a9746d8d TH |
1326 | if (!cgroup_is_dead(cgrp)) |
1327 | return cgrp; | |
1328 | ||
1329 | cgroup_kn_unlock(kn); | |
1330 | return NULL; | |
ddbcc7e8 | 1331 | } |
05ef1d7c | 1332 | |
2739d3cc | 1333 | static void cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft) |
05ef1d7c | 1334 | { |
2bd59d48 | 1335 | char name[CGROUP_FILE_NAME_MAX]; |
05ef1d7c | 1336 | |
01f6474c | 1337 | lockdep_assert_held(&cgroup_mutex); |
2bd59d48 | 1338 | kernfs_remove_by_name(cgrp->kn, cgroup_file_name(cgrp, cft, name)); |
05ef1d7c TH |
1339 | } |
1340 | ||
13af07df | 1341 | /** |
4df8dc90 TH |
1342 | * css_clear_dir - remove subsys files in a cgroup directory |
1343 | * @css: taget css | |
1344 | * @cgrp_override: specify if target cgroup is different from css->cgroup | |
13af07df | 1345 | */ |
4df8dc90 TH |
1346 | static void css_clear_dir(struct cgroup_subsys_state *css, |
1347 | struct cgroup *cgrp_override) | |
05ef1d7c | 1348 | { |
4df8dc90 TH |
1349 | struct cgroup *cgrp = cgrp_override ?: css->cgroup; |
1350 | struct cftype *cfts; | |
05ef1d7c | 1351 | |
4df8dc90 TH |
1352 | list_for_each_entry(cfts, &css->ss->cfts, node) |
1353 | cgroup_addrm_files(css, cgrp, cfts, false); | |
ddbcc7e8 PM |
1354 | } |
1355 | ||
ccdca218 | 1356 | /** |
4df8dc90 TH |
1357 | * css_populate_dir - create subsys files in a cgroup directory |
1358 | * @css: target css | |
1359 | * @cgrp_overried: specify if target cgroup is different from css->cgroup | |
ccdca218 TH |
1360 | * |
1361 | * On failure, no file is added. | |
1362 | */ | |
4df8dc90 TH |
1363 | static int css_populate_dir(struct cgroup_subsys_state *css, |
1364 | struct cgroup *cgrp_override) | |
ccdca218 | 1365 | { |
4df8dc90 TH |
1366 | struct cgroup *cgrp = cgrp_override ?: css->cgroup; |
1367 | struct cftype *cfts, *failed_cfts; | |
1368 | int ret; | |
ccdca218 | 1369 | |
4df8dc90 TH |
1370 | if (!css->ss) { |
1371 | if (cgroup_on_dfl(cgrp)) | |
1372 | cfts = cgroup_dfl_base_files; | |
1373 | else | |
1374 | cfts = cgroup_legacy_base_files; | |
ccdca218 | 1375 | |
4df8dc90 TH |
1376 | return cgroup_addrm_files(&cgrp->self, cgrp, cfts, true); |
1377 | } | |
ccdca218 | 1378 | |
4df8dc90 TH |
1379 | list_for_each_entry(cfts, &css->ss->cfts, node) { |
1380 | ret = cgroup_addrm_files(css, cgrp, cfts, true); | |
1381 | if (ret < 0) { | |
1382 | failed_cfts = cfts; | |
1383 | goto err; | |
ccdca218 TH |
1384 | } |
1385 | } | |
1386 | return 0; | |
1387 | err: | |
4df8dc90 TH |
1388 | list_for_each_entry(cfts, &css->ss->cfts, node) { |
1389 | if (cfts == failed_cfts) | |
1390 | break; | |
1391 | cgroup_addrm_files(css, cgrp, cfts, false); | |
1392 | } | |
ccdca218 TH |
1393 | return ret; |
1394 | } | |
1395 | ||
8ab456ac AS |
1396 | static int rebind_subsystems(struct cgroup_root *dst_root, |
1397 | unsigned long ss_mask) | |
ddbcc7e8 | 1398 | { |
1ada4838 | 1399 | struct cgroup *dcgrp = &dst_root->cgrp; |
30159ec7 | 1400 | struct cgroup_subsys *ss; |
8ab456ac | 1401 | unsigned long tmp_ss_mask; |
2d8f243a | 1402 | int ssid, i, ret; |
ddbcc7e8 | 1403 | |
ace2bee8 | 1404 | lockdep_assert_held(&cgroup_mutex); |
ddbcc7e8 | 1405 | |
a966a4ed | 1406 | for_each_subsys_which(ss, ssid, &ss_mask) { |
7fd8c565 TH |
1407 | /* if @ss has non-root csses attached to it, can't move */ |
1408 | if (css_next_child(NULL, cgroup_css(&ss->root->cgrp, ss))) | |
3ed80a62 | 1409 | return -EBUSY; |
1d5be6b2 | 1410 | |
5df36032 | 1411 | /* can't move between two non-dummy roots either */ |
7fd8c565 | 1412 | if (ss->root != &cgrp_dfl_root && dst_root != &cgrp_dfl_root) |
5df36032 | 1413 | return -EBUSY; |
ddbcc7e8 PM |
1414 | } |
1415 | ||
5533e011 TH |
1416 | /* skip creating root files on dfl_root for inhibited subsystems */ |
1417 | tmp_ss_mask = ss_mask; | |
1418 | if (dst_root == &cgrp_dfl_root) | |
1419 | tmp_ss_mask &= ~cgrp_dfl_root_inhibit_ss_mask; | |
1420 | ||
4df8dc90 TH |
1421 | for_each_subsys_which(ss, ssid, &tmp_ss_mask) { |
1422 | struct cgroup *scgrp = &ss->root->cgrp; | |
1423 | int tssid; | |
1424 | ||
1425 | ret = css_populate_dir(cgroup_css(scgrp, ss), dcgrp); | |
1426 | if (!ret) | |
1427 | continue; | |
ddbcc7e8 | 1428 | |
a2dd4247 TH |
1429 | /* |
1430 | * Rebinding back to the default root is not allowed to | |
1431 | * fail. Using both default and non-default roots should | |
1432 | * be rare. Moving subsystems back and forth even more so. | |
1433 | * Just warn about it and continue. | |
1434 | */ | |
4df8dc90 TH |
1435 | if (dst_root == &cgrp_dfl_root) { |
1436 | if (cgrp_dfl_root_visible) { | |
1437 | pr_warn("failed to create files (%d) while rebinding 0x%lx to default root\n", | |
1438 | ret, ss_mask); | |
1439 | pr_warn("you may retry by moving them to a different hierarchy and unbinding\n"); | |
1440 | } | |
1441 | continue; | |
a2dd4247 | 1442 | } |
4df8dc90 TH |
1443 | |
1444 | for_each_subsys_which(ss, tssid, &tmp_ss_mask) { | |
1445 | if (tssid == ssid) | |
1446 | break; | |
1447 | css_clear_dir(cgroup_css(scgrp, ss), dcgrp); | |
1448 | } | |
1449 | return ret; | |
5df36032 | 1450 | } |
3126121f TH |
1451 | |
1452 | /* | |
1453 | * Nothing can fail from this point on. Remove files for the | |
1454 | * removed subsystems and rebind each subsystem. | |
1455 | */ | |
a966a4ed | 1456 | for_each_subsys_which(ss, ssid, &ss_mask) { |
1ada4838 TH |
1457 | struct cgroup_root *src_root = ss->root; |
1458 | struct cgroup *scgrp = &src_root->cgrp; | |
1459 | struct cgroup_subsys_state *css = cgroup_css(scgrp, ss); | |
2d8f243a | 1460 | struct css_set *cset; |
a8a648c4 | 1461 | |
1ada4838 | 1462 | WARN_ON(!css || cgroup_css(dcgrp, ss)); |
a8a648c4 | 1463 | |
4df8dc90 TH |
1464 | css_clear_dir(css, NULL); |
1465 | ||
1ada4838 TH |
1466 | RCU_INIT_POINTER(scgrp->subsys[ssid], NULL); |
1467 | rcu_assign_pointer(dcgrp->subsys[ssid], css); | |
5df36032 | 1468 | ss->root = dst_root; |
1ada4838 | 1469 | css->cgroup = dcgrp; |
73e80ed8 | 1470 | |
2d8f243a TH |
1471 | down_write(&css_set_rwsem); |
1472 | hash_for_each(css_set_table, i, cset, hlist) | |
1473 | list_move_tail(&cset->e_cset_node[ss->id], | |
1ada4838 | 1474 | &dcgrp->e_csets[ss->id]); |
2d8f243a TH |
1475 | up_write(&css_set_rwsem); |
1476 | ||
f392e51c | 1477 | src_root->subsys_mask &= ~(1 << ssid); |
1ada4838 TH |
1478 | scgrp->subtree_control &= ~(1 << ssid); |
1479 | cgroup_refresh_child_subsys_mask(scgrp); | |
f392e51c | 1480 | |
bd53d617 | 1481 | /* default hierarchy doesn't enable controllers by default */ |
f392e51c | 1482 | dst_root->subsys_mask |= 1 << ssid; |
49d1dc4b TH |
1483 | if (dst_root == &cgrp_dfl_root) { |
1484 | static_branch_enable(cgroup_subsys_on_dfl_key[ssid]); | |
1485 | } else { | |
1ada4838 TH |
1486 | dcgrp->subtree_control |= 1 << ssid; |
1487 | cgroup_refresh_child_subsys_mask(dcgrp); | |
49d1dc4b | 1488 | static_branch_disable(cgroup_subsys_on_dfl_key[ssid]); |
667c2491 | 1489 | } |
a8a648c4 | 1490 | |
5df36032 TH |
1491 | if (ss->bind) |
1492 | ss->bind(css); | |
ddbcc7e8 | 1493 | } |
ddbcc7e8 | 1494 | |
1ada4838 | 1495 | kernfs_activate(dcgrp->kn); |
ddbcc7e8 PM |
1496 | return 0; |
1497 | } | |
1498 | ||
2bd59d48 TH |
1499 | static int cgroup_show_options(struct seq_file *seq, |
1500 | struct kernfs_root *kf_root) | |
ddbcc7e8 | 1501 | { |
3dd06ffa | 1502 | struct cgroup_root *root = cgroup_root_from_kf(kf_root); |
ddbcc7e8 | 1503 | struct cgroup_subsys *ss; |
b85d2040 | 1504 | int ssid; |
ddbcc7e8 | 1505 | |
d98817d4 TH |
1506 | if (root != &cgrp_dfl_root) |
1507 | for_each_subsys(ss, ssid) | |
1508 | if (root->subsys_mask & (1 << ssid)) | |
61e57c0c | 1509 | seq_show_option(seq, ss->legacy_name, NULL); |
93438629 | 1510 | if (root->flags & CGRP_ROOT_NOPREFIX) |
ddbcc7e8 | 1511 | seq_puts(seq, ",noprefix"); |
93438629 | 1512 | if (root->flags & CGRP_ROOT_XATTR) |
03b1cde6 | 1513 | seq_puts(seq, ",xattr"); |
69e943b7 TH |
1514 | |
1515 | spin_lock(&release_agent_path_lock); | |
81a6a5cd | 1516 | if (strlen(root->release_agent_path)) |
a068acf2 KC |
1517 | seq_show_option(seq, "release_agent", |
1518 | root->release_agent_path); | |
69e943b7 TH |
1519 | spin_unlock(&release_agent_path_lock); |
1520 | ||
3dd06ffa | 1521 | if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags)) |
97978e6d | 1522 | seq_puts(seq, ",clone_children"); |
c6d57f33 | 1523 | if (strlen(root->name)) |
a068acf2 | 1524 | seq_show_option(seq, "name", root->name); |
ddbcc7e8 PM |
1525 | return 0; |
1526 | } | |
1527 | ||
1528 | struct cgroup_sb_opts { | |
8ab456ac | 1529 | unsigned long subsys_mask; |
69dfa00c | 1530 | unsigned int flags; |
81a6a5cd | 1531 | char *release_agent; |
2260e7fc | 1532 | bool cpuset_clone_children; |
c6d57f33 | 1533 | char *name; |
2c6ab6d2 PM |
1534 | /* User explicitly requested empty subsystem */ |
1535 | bool none; | |
ddbcc7e8 PM |
1536 | }; |
1537 | ||
cf5d5941 | 1538 | static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts) |
ddbcc7e8 | 1539 | { |
32a8cf23 DL |
1540 | char *token, *o = data; |
1541 | bool all_ss = false, one_ss = false; | |
8ab456ac | 1542 | unsigned long mask = -1UL; |
30159ec7 | 1543 | struct cgroup_subsys *ss; |
7b9a6ba5 | 1544 | int nr_opts = 0; |
30159ec7 | 1545 | int i; |
f9ab5b5b LZ |
1546 | |
1547 | #ifdef CONFIG_CPUSETS | |
69dfa00c | 1548 | mask = ~(1U << cpuset_cgrp_id); |
f9ab5b5b | 1549 | #endif |
ddbcc7e8 | 1550 | |
c6d57f33 | 1551 | memset(opts, 0, sizeof(*opts)); |
ddbcc7e8 PM |
1552 | |
1553 | while ((token = strsep(&o, ",")) != NULL) { | |
7b9a6ba5 TH |
1554 | nr_opts++; |
1555 | ||
ddbcc7e8 PM |
1556 | if (!*token) |
1557 | return -EINVAL; | |
32a8cf23 | 1558 | if (!strcmp(token, "none")) { |
2c6ab6d2 PM |
1559 | /* Explicitly have no subsystems */ |
1560 | opts->none = true; | |
32a8cf23 DL |
1561 | continue; |
1562 | } | |
1563 | if (!strcmp(token, "all")) { | |
1564 | /* Mutually exclusive option 'all' + subsystem name */ | |
1565 | if (one_ss) | |
1566 | return -EINVAL; | |
1567 | all_ss = true; | |
1568 | continue; | |
1569 | } | |
873fe09e TH |
1570 | if (!strcmp(token, "__DEVEL__sane_behavior")) { |
1571 | opts->flags |= CGRP_ROOT_SANE_BEHAVIOR; | |
1572 | continue; | |
1573 | } | |
32a8cf23 | 1574 | if (!strcmp(token, "noprefix")) { |
93438629 | 1575 | opts->flags |= CGRP_ROOT_NOPREFIX; |
32a8cf23 DL |
1576 | continue; |
1577 | } | |
1578 | if (!strcmp(token, "clone_children")) { | |
2260e7fc | 1579 | opts->cpuset_clone_children = true; |
32a8cf23 DL |
1580 | continue; |
1581 | } | |
03b1cde6 | 1582 | if (!strcmp(token, "xattr")) { |
93438629 | 1583 | opts->flags |= CGRP_ROOT_XATTR; |
03b1cde6 AR |
1584 | continue; |
1585 | } | |
32a8cf23 | 1586 | if (!strncmp(token, "release_agent=", 14)) { |
81a6a5cd PM |
1587 | /* Specifying two release agents is forbidden */ |
1588 | if (opts->release_agent) | |
1589 | return -EINVAL; | |
c6d57f33 | 1590 | opts->release_agent = |
e400c285 | 1591 | kstrndup(token + 14, PATH_MAX - 1, GFP_KERNEL); |
81a6a5cd PM |
1592 | if (!opts->release_agent) |
1593 | return -ENOMEM; | |
32a8cf23 DL |
1594 | continue; |
1595 | } | |
1596 | if (!strncmp(token, "name=", 5)) { | |
c6d57f33 PM |
1597 | const char *name = token + 5; |
1598 | /* Can't specify an empty name */ | |
1599 | if (!strlen(name)) | |
1600 | return -EINVAL; | |
1601 | /* Must match [\w.-]+ */ | |
1602 | for (i = 0; i < strlen(name); i++) { | |
1603 | char c = name[i]; | |
1604 | if (isalnum(c)) | |
1605 | continue; | |
1606 | if ((c == '.') || (c == '-') || (c == '_')) | |
1607 | continue; | |
1608 | return -EINVAL; | |
1609 | } | |
1610 | /* Specifying two names is forbidden */ | |
1611 | if (opts->name) | |
1612 | return -EINVAL; | |
1613 | opts->name = kstrndup(name, | |
e400c285 | 1614 | MAX_CGROUP_ROOT_NAMELEN - 1, |
c6d57f33 PM |
1615 | GFP_KERNEL); |
1616 | if (!opts->name) | |
1617 | return -ENOMEM; | |
32a8cf23 DL |
1618 | |
1619 | continue; | |
1620 | } | |
1621 | ||
30159ec7 | 1622 | for_each_subsys(ss, i) { |
3e1d2eed | 1623 | if (strcmp(token, ss->legacy_name)) |
32a8cf23 | 1624 | continue; |
fc5ed1e9 | 1625 | if (!cgroup_ssid_enabled(i)) |
32a8cf23 DL |
1626 | continue; |
1627 | ||
1628 | /* Mutually exclusive option 'all' + subsystem name */ | |
1629 | if (all_ss) | |
1630 | return -EINVAL; | |
69dfa00c | 1631 | opts->subsys_mask |= (1 << i); |
32a8cf23 DL |
1632 | one_ss = true; |
1633 | ||
1634 | break; | |
1635 | } | |
1636 | if (i == CGROUP_SUBSYS_COUNT) | |
1637 | return -ENOENT; | |
1638 | } | |
1639 | ||
873fe09e | 1640 | if (opts->flags & CGRP_ROOT_SANE_BEHAVIOR) { |
ed3d261b | 1641 | pr_warn("sane_behavior: this is still under development and its behaviors will change, proceed at your own risk\n"); |
7b9a6ba5 TH |
1642 | if (nr_opts != 1) { |
1643 | pr_err("sane_behavior: no other mount options allowed\n"); | |
873fe09e TH |
1644 | return -EINVAL; |
1645 | } | |
7b9a6ba5 | 1646 | return 0; |
873fe09e TH |
1647 | } |
1648 | ||
7b9a6ba5 TH |
1649 | /* |
1650 | * If the 'all' option was specified select all the subsystems, | |
1651 | * otherwise if 'none', 'name=' and a subsystem name options were | |
1652 | * not specified, let's default to 'all' | |
1653 | */ | |
1654 | if (all_ss || (!one_ss && !opts->none && !opts->name)) | |
1655 | for_each_subsys(ss, i) | |
fc5ed1e9 | 1656 | if (cgroup_ssid_enabled(i)) |
7b9a6ba5 TH |
1657 | opts->subsys_mask |= (1 << i); |
1658 | ||
1659 | /* | |
1660 | * We either have to specify by name or by subsystems. (So all | |
1661 | * empty hierarchies must have a name). | |
1662 | */ | |
1663 | if (!opts->subsys_mask && !opts->name) | |
1664 | return -EINVAL; | |
1665 | ||
f9ab5b5b LZ |
1666 | /* |
1667 | * Option noprefix was introduced just for backward compatibility | |
1668 | * with the old cpuset, so we allow noprefix only if mounting just | |
1669 | * the cpuset subsystem. | |
1670 | */ | |
93438629 | 1671 | if ((opts->flags & CGRP_ROOT_NOPREFIX) && (opts->subsys_mask & mask)) |
f9ab5b5b LZ |
1672 | return -EINVAL; |
1673 | ||
2c6ab6d2 | 1674 | /* Can't specify "none" and some subsystems */ |
a1a71b45 | 1675 | if (opts->subsys_mask && opts->none) |
2c6ab6d2 PM |
1676 | return -EINVAL; |
1677 | ||
ddbcc7e8 PM |
1678 | return 0; |
1679 | } | |
1680 | ||
2bd59d48 | 1681 | static int cgroup_remount(struct kernfs_root *kf_root, int *flags, char *data) |
ddbcc7e8 PM |
1682 | { |
1683 | int ret = 0; | |
3dd06ffa | 1684 | struct cgroup_root *root = cgroup_root_from_kf(kf_root); |
ddbcc7e8 | 1685 | struct cgroup_sb_opts opts; |
8ab456ac | 1686 | unsigned long added_mask, removed_mask; |
ddbcc7e8 | 1687 | |
aa6ec29b TH |
1688 | if (root == &cgrp_dfl_root) { |
1689 | pr_err("remount is not allowed\n"); | |
873fe09e TH |
1690 | return -EINVAL; |
1691 | } | |
1692 | ||
ddbcc7e8 PM |
1693 | mutex_lock(&cgroup_mutex); |
1694 | ||
1695 | /* See what subsystems are wanted */ | |
1696 | ret = parse_cgroupfs_options(data, &opts); | |
1697 | if (ret) | |
1698 | goto out_unlock; | |
1699 | ||
f392e51c | 1700 | if (opts.subsys_mask != root->subsys_mask || opts.release_agent) |
ed3d261b | 1701 | pr_warn("option changes via remount are deprecated (pid=%d comm=%s)\n", |
a2a1f9ea | 1702 | task_tgid_nr(current), current->comm); |
8b5a5a9d | 1703 | |
f392e51c TH |
1704 | added_mask = opts.subsys_mask & ~root->subsys_mask; |
1705 | removed_mask = root->subsys_mask & ~opts.subsys_mask; | |
13af07df | 1706 | |
cf5d5941 | 1707 | /* Don't allow flags or name to change at remount */ |
7450e90b | 1708 | if ((opts.flags ^ root->flags) || |
cf5d5941 | 1709 | (opts.name && strcmp(opts.name, root->name))) { |
69dfa00c | 1710 | pr_err("option or name mismatch, new: 0x%x \"%s\", old: 0x%x \"%s\"\n", |
7450e90b | 1711 | opts.flags, opts.name ?: "", root->flags, root->name); |
c6d57f33 PM |
1712 | ret = -EINVAL; |
1713 | goto out_unlock; | |
1714 | } | |
1715 | ||
f172e67c | 1716 | /* remounting is not allowed for populated hierarchies */ |
d5c419b6 | 1717 | if (!list_empty(&root->cgrp.self.children)) { |
f172e67c | 1718 | ret = -EBUSY; |
0670e08b | 1719 | goto out_unlock; |
cf5d5941 | 1720 | } |
ddbcc7e8 | 1721 | |
5df36032 | 1722 | ret = rebind_subsystems(root, added_mask); |
3126121f | 1723 | if (ret) |
0670e08b | 1724 | goto out_unlock; |
ddbcc7e8 | 1725 | |
3dd06ffa | 1726 | rebind_subsystems(&cgrp_dfl_root, removed_mask); |
5df36032 | 1727 | |
69e943b7 TH |
1728 | if (opts.release_agent) { |
1729 | spin_lock(&release_agent_path_lock); | |
81a6a5cd | 1730 | strcpy(root->release_agent_path, opts.release_agent); |
69e943b7 TH |
1731 | spin_unlock(&release_agent_path_lock); |
1732 | } | |
ddbcc7e8 | 1733 | out_unlock: |
66bdc9cf | 1734 | kfree(opts.release_agent); |
c6d57f33 | 1735 | kfree(opts.name); |
ddbcc7e8 | 1736 | mutex_unlock(&cgroup_mutex); |
ddbcc7e8 PM |
1737 | return ret; |
1738 | } | |
1739 | ||
afeb0f9f TH |
1740 | /* |
1741 | * To reduce the fork() overhead for systems that are not actually using | |
1742 | * their cgroups capability, we don't maintain the lists running through | |
1743 | * each css_set to its tasks until we see the list actually used - in other | |
1744 | * words after the first mount. | |
1745 | */ | |
1746 | static bool use_task_css_set_links __read_mostly; | |
1747 | ||
1748 | static void cgroup_enable_task_cg_lists(void) | |
1749 | { | |
1750 | struct task_struct *p, *g; | |
1751 | ||
96d365e0 | 1752 | down_write(&css_set_rwsem); |
afeb0f9f TH |
1753 | |
1754 | if (use_task_css_set_links) | |
1755 | goto out_unlock; | |
1756 | ||
1757 | use_task_css_set_links = true; | |
1758 | ||
1759 | /* | |
1760 | * We need tasklist_lock because RCU is not safe against | |
1761 | * while_each_thread(). Besides, a forking task that has passed | |
1762 | * cgroup_post_fork() without seeing use_task_css_set_links = 1 | |
1763 | * is not guaranteed to have its child immediately visible in the | |
1764 | * tasklist if we walk through it with RCU. | |
1765 | */ | |
1766 | read_lock(&tasklist_lock); | |
1767 | do_each_thread(g, p) { | |
afeb0f9f TH |
1768 | WARN_ON_ONCE(!list_empty(&p->cg_list) || |
1769 | task_css_set(p) != &init_css_set); | |
1770 | ||
1771 | /* | |
1772 | * We should check if the process is exiting, otherwise | |
1773 | * it will race with cgroup_exit() in that the list | |
1774 | * entry won't be deleted though the process has exited. | |
f153ad11 TH |
1775 | * Do it while holding siglock so that we don't end up |
1776 | * racing against cgroup_exit(). | |
afeb0f9f | 1777 | */ |
f153ad11 | 1778 | spin_lock_irq(&p->sighand->siglock); |
eaf797ab TH |
1779 | if (!(p->flags & PF_EXITING)) { |
1780 | struct css_set *cset = task_css_set(p); | |
1781 | ||
0de0942d TH |
1782 | if (!css_set_populated(cset)) |
1783 | css_set_update_populated(cset, true); | |
eaf797ab TH |
1784 | list_add(&p->cg_list, &cset->tasks); |
1785 | get_css_set(cset); | |
1786 | } | |
f153ad11 | 1787 | spin_unlock_irq(&p->sighand->siglock); |
afeb0f9f TH |
1788 | } while_each_thread(g, p); |
1789 | read_unlock(&tasklist_lock); | |
1790 | out_unlock: | |
96d365e0 | 1791 | up_write(&css_set_rwsem); |
afeb0f9f | 1792 | } |
ddbcc7e8 | 1793 | |
cc31edce PM |
1794 | static void init_cgroup_housekeeping(struct cgroup *cgrp) |
1795 | { | |
2d8f243a TH |
1796 | struct cgroup_subsys *ss; |
1797 | int ssid; | |
1798 | ||
d5c419b6 TH |
1799 | INIT_LIST_HEAD(&cgrp->self.sibling); |
1800 | INIT_LIST_HEAD(&cgrp->self.children); | |
6f60eade | 1801 | INIT_LIST_HEAD(&cgrp->self.files); |
69d0206c | 1802 | INIT_LIST_HEAD(&cgrp->cset_links); |
72a8cb30 BB |
1803 | INIT_LIST_HEAD(&cgrp->pidlists); |
1804 | mutex_init(&cgrp->pidlist_mutex); | |
9d800df1 | 1805 | cgrp->self.cgroup = cgrp; |
184faf32 | 1806 | cgrp->self.flags |= CSS_ONLINE; |
2d8f243a TH |
1807 | |
1808 | for_each_subsys(ss, ssid) | |
1809 | INIT_LIST_HEAD(&cgrp->e_csets[ssid]); | |
f8f22e53 TH |
1810 | |
1811 | init_waitqueue_head(&cgrp->offline_waitq); | |
971ff493 | 1812 | INIT_WORK(&cgrp->release_agent_work, cgroup_release_agent); |
cc31edce | 1813 | } |
c6d57f33 | 1814 | |
3dd06ffa | 1815 | static void init_cgroup_root(struct cgroup_root *root, |
172a2c06 | 1816 | struct cgroup_sb_opts *opts) |
ddbcc7e8 | 1817 | { |
3dd06ffa | 1818 | struct cgroup *cgrp = &root->cgrp; |
b0ca5a84 | 1819 | |
ddbcc7e8 | 1820 | INIT_LIST_HEAD(&root->root_list); |
3c9c825b | 1821 | atomic_set(&root->nr_cgrps, 1); |
bd89aabc | 1822 | cgrp->root = root; |
cc31edce | 1823 | init_cgroup_housekeeping(cgrp); |
4e96ee8e | 1824 | idr_init(&root->cgroup_idr); |
c6d57f33 | 1825 | |
c6d57f33 PM |
1826 | root->flags = opts->flags; |
1827 | if (opts->release_agent) | |
1828 | strcpy(root->release_agent_path, opts->release_agent); | |
1829 | if (opts->name) | |
1830 | strcpy(root->name, opts->name); | |
2260e7fc | 1831 | if (opts->cpuset_clone_children) |
3dd06ffa | 1832 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags); |
c6d57f33 PM |
1833 | } |
1834 | ||
8ab456ac | 1835 | static int cgroup_setup_root(struct cgroup_root *root, unsigned long ss_mask) |
2c6ab6d2 | 1836 | { |
d427dfeb | 1837 | LIST_HEAD(tmp_links); |
3dd06ffa | 1838 | struct cgroup *root_cgrp = &root->cgrp; |
d427dfeb | 1839 | struct css_set *cset; |
d427dfeb | 1840 | int i, ret; |
2c6ab6d2 | 1841 | |
d427dfeb | 1842 | lockdep_assert_held(&cgroup_mutex); |
c6d57f33 | 1843 | |
cf780b7d | 1844 | ret = cgroup_idr_alloc(&root->cgroup_idr, root_cgrp, 1, 2, GFP_KERNEL); |
d427dfeb | 1845 | if (ret < 0) |
2bd59d48 | 1846 | goto out; |
d427dfeb | 1847 | root_cgrp->id = ret; |
c6d57f33 | 1848 | |
2aad2a86 TH |
1849 | ret = percpu_ref_init(&root_cgrp->self.refcnt, css_release, 0, |
1850 | GFP_KERNEL); | |
9d755d33 TH |
1851 | if (ret) |
1852 | goto out; | |
1853 | ||
d427dfeb | 1854 | /* |
96d365e0 | 1855 | * We're accessing css_set_count without locking css_set_rwsem here, |
d427dfeb TH |
1856 | * but that's OK - it can only be increased by someone holding |
1857 | * cgroup_lock, and that's us. The worst that can happen is that we | |
1858 | * have some link structures left over | |
1859 | */ | |
1860 | ret = allocate_cgrp_cset_links(css_set_count, &tmp_links); | |
1861 | if (ret) | |
9d755d33 | 1862 | goto cancel_ref; |
ddbcc7e8 | 1863 | |
985ed670 | 1864 | ret = cgroup_init_root_id(root); |
ddbcc7e8 | 1865 | if (ret) |
9d755d33 | 1866 | goto cancel_ref; |
ddbcc7e8 | 1867 | |
2bd59d48 TH |
1868 | root->kf_root = kernfs_create_root(&cgroup_kf_syscall_ops, |
1869 | KERNFS_ROOT_CREATE_DEACTIVATED, | |
1870 | root_cgrp); | |
1871 | if (IS_ERR(root->kf_root)) { | |
1872 | ret = PTR_ERR(root->kf_root); | |
1873 | goto exit_root_id; | |
1874 | } | |
1875 | root_cgrp->kn = root->kf_root->kn; | |
ddbcc7e8 | 1876 | |
4df8dc90 | 1877 | ret = css_populate_dir(&root_cgrp->self, NULL); |
d427dfeb | 1878 | if (ret) |
2bd59d48 | 1879 | goto destroy_root; |
ddbcc7e8 | 1880 | |
5df36032 | 1881 | ret = rebind_subsystems(root, ss_mask); |
d427dfeb | 1882 | if (ret) |
2bd59d48 | 1883 | goto destroy_root; |
ddbcc7e8 | 1884 | |
d427dfeb TH |
1885 | /* |
1886 | * There must be no failure case after here, since rebinding takes | |
1887 | * care of subsystems' refcounts, which are explicitly dropped in | |
1888 | * the failure exit path. | |
1889 | */ | |
1890 | list_add(&root->root_list, &cgroup_roots); | |
1891 | cgroup_root_count++; | |
0df6a63f | 1892 | |
d427dfeb | 1893 | /* |
3dd06ffa | 1894 | * Link the root cgroup in this hierarchy into all the css_set |
d427dfeb TH |
1895 | * objects. |
1896 | */ | |
96d365e0 | 1897 | down_write(&css_set_rwsem); |
0de0942d | 1898 | hash_for_each(css_set_table, i, cset, hlist) { |
d427dfeb | 1899 | link_css_set(&tmp_links, cset, root_cgrp); |
0de0942d TH |
1900 | if (css_set_populated(cset)) |
1901 | cgroup_update_populated(root_cgrp, true); | |
1902 | } | |
96d365e0 | 1903 | up_write(&css_set_rwsem); |
ddbcc7e8 | 1904 | |
d5c419b6 | 1905 | BUG_ON(!list_empty(&root_cgrp->self.children)); |
3c9c825b | 1906 | BUG_ON(atomic_read(&root->nr_cgrps) != 1); |
ddbcc7e8 | 1907 | |
2bd59d48 | 1908 | kernfs_activate(root_cgrp->kn); |
d427dfeb | 1909 | ret = 0; |
2bd59d48 | 1910 | goto out; |
d427dfeb | 1911 | |
2bd59d48 TH |
1912 | destroy_root: |
1913 | kernfs_destroy_root(root->kf_root); | |
1914 | root->kf_root = NULL; | |
1915 | exit_root_id: | |
d427dfeb | 1916 | cgroup_exit_root_id(root); |
9d755d33 | 1917 | cancel_ref: |
9a1049da | 1918 | percpu_ref_exit(&root_cgrp->self.refcnt); |
2bd59d48 | 1919 | out: |
d427dfeb TH |
1920 | free_cgrp_cset_links(&tmp_links); |
1921 | return ret; | |
ddbcc7e8 PM |
1922 | } |
1923 | ||
f7e83571 | 1924 | static struct dentry *cgroup_mount(struct file_system_type *fs_type, |
ddbcc7e8 | 1925 | int flags, const char *unused_dev_name, |
f7e83571 | 1926 | void *data) |
ddbcc7e8 | 1927 | { |
3a32bd72 | 1928 | struct super_block *pinned_sb = NULL; |
970317aa | 1929 | struct cgroup_subsys *ss; |
3dd06ffa | 1930 | struct cgroup_root *root; |
ddbcc7e8 | 1931 | struct cgroup_sb_opts opts; |
2bd59d48 | 1932 | struct dentry *dentry; |
8e30e2b8 | 1933 | int ret; |
970317aa | 1934 | int i; |
c6b3d5bc | 1935 | bool new_sb; |
ddbcc7e8 | 1936 | |
56fde9e0 TH |
1937 | /* |
1938 | * The first time anyone tries to mount a cgroup, enable the list | |
1939 | * linking each css_set to its tasks and fix up all existing tasks. | |
1940 | */ | |
1941 | if (!use_task_css_set_links) | |
1942 | cgroup_enable_task_cg_lists(); | |
e37a06f1 | 1943 | |
aae8aab4 | 1944 | mutex_lock(&cgroup_mutex); |
8e30e2b8 TH |
1945 | |
1946 | /* First find the desired set of subsystems */ | |
ddbcc7e8 | 1947 | ret = parse_cgroupfs_options(data, &opts); |
c6d57f33 | 1948 | if (ret) |
8e30e2b8 | 1949 | goto out_unlock; |
a015edd2 | 1950 | |
2bd59d48 | 1951 | /* look for a matching existing root */ |
7b9a6ba5 | 1952 | if (opts.flags & CGRP_ROOT_SANE_BEHAVIOR) { |
a2dd4247 TH |
1953 | cgrp_dfl_root_visible = true; |
1954 | root = &cgrp_dfl_root; | |
1955 | cgroup_get(&root->cgrp); | |
1956 | ret = 0; | |
1957 | goto out_unlock; | |
ddbcc7e8 PM |
1958 | } |
1959 | ||
970317aa LZ |
1960 | /* |
1961 | * Destruction of cgroup root is asynchronous, so subsystems may | |
1962 | * still be dying after the previous unmount. Let's drain the | |
1963 | * dying subsystems. We just need to ensure that the ones | |
1964 | * unmounted previously finish dying and don't care about new ones | |
1965 | * starting. Testing ref liveliness is good enough. | |
1966 | */ | |
1967 | for_each_subsys(ss, i) { | |
1968 | if (!(opts.subsys_mask & (1 << i)) || | |
1969 | ss->root == &cgrp_dfl_root) | |
1970 | continue; | |
1971 | ||
1972 | if (!percpu_ref_tryget_live(&ss->root->cgrp.self.refcnt)) { | |
1973 | mutex_unlock(&cgroup_mutex); | |
1974 | msleep(10); | |
1975 | ret = restart_syscall(); | |
1976 | goto out_free; | |
1977 | } | |
1978 | cgroup_put(&ss->root->cgrp); | |
1979 | } | |
1980 | ||
985ed670 | 1981 | for_each_root(root) { |
2bd59d48 | 1982 | bool name_match = false; |
3126121f | 1983 | |
3dd06ffa | 1984 | if (root == &cgrp_dfl_root) |
985ed670 | 1985 | continue; |
3126121f | 1986 | |
cf5d5941 | 1987 | /* |
2bd59d48 TH |
1988 | * If we asked for a name then it must match. Also, if |
1989 | * name matches but sybsys_mask doesn't, we should fail. | |
1990 | * Remember whether name matched. | |
cf5d5941 | 1991 | */ |
2bd59d48 TH |
1992 | if (opts.name) { |
1993 | if (strcmp(opts.name, root->name)) | |
1994 | continue; | |
1995 | name_match = true; | |
1996 | } | |
ddbcc7e8 | 1997 | |
c6d57f33 | 1998 | /* |
2bd59d48 TH |
1999 | * If we asked for subsystems (or explicitly for no |
2000 | * subsystems) then they must match. | |
c6d57f33 | 2001 | */ |
2bd59d48 | 2002 | if ((opts.subsys_mask || opts.none) && |
f392e51c | 2003 | (opts.subsys_mask != root->subsys_mask)) { |
2bd59d48 TH |
2004 | if (!name_match) |
2005 | continue; | |
2006 | ret = -EBUSY; | |
2007 | goto out_unlock; | |
2008 | } | |
873fe09e | 2009 | |
7b9a6ba5 TH |
2010 | if (root->flags ^ opts.flags) |
2011 | pr_warn("new mount options do not match the existing superblock, will be ignored\n"); | |
ddbcc7e8 | 2012 | |
776f02fa | 2013 | /* |
3a32bd72 LZ |
2014 | * We want to reuse @root whose lifetime is governed by its |
2015 | * ->cgrp. Let's check whether @root is alive and keep it | |
2016 | * that way. As cgroup_kill_sb() can happen anytime, we | |
2017 | * want to block it by pinning the sb so that @root doesn't | |
2018 | * get killed before mount is complete. | |
2019 | * | |
2020 | * With the sb pinned, tryget_live can reliably indicate | |
2021 | * whether @root can be reused. If it's being killed, | |
2022 | * drain it. We can use wait_queue for the wait but this | |
2023 | * path is super cold. Let's just sleep a bit and retry. | |
776f02fa | 2024 | */ |
3a32bd72 LZ |
2025 | pinned_sb = kernfs_pin_sb(root->kf_root, NULL); |
2026 | if (IS_ERR(pinned_sb) || | |
2027 | !percpu_ref_tryget_live(&root->cgrp.self.refcnt)) { | |
776f02fa | 2028 | mutex_unlock(&cgroup_mutex); |
3a32bd72 LZ |
2029 | if (!IS_ERR_OR_NULL(pinned_sb)) |
2030 | deactivate_super(pinned_sb); | |
776f02fa | 2031 | msleep(10); |
a015edd2 TH |
2032 | ret = restart_syscall(); |
2033 | goto out_free; | |
776f02fa | 2034 | } |
ddbcc7e8 | 2035 | |
776f02fa | 2036 | ret = 0; |
2bd59d48 | 2037 | goto out_unlock; |
ddbcc7e8 | 2038 | } |
ddbcc7e8 | 2039 | |
817929ec | 2040 | /* |
172a2c06 TH |
2041 | * No such thing, create a new one. name= matching without subsys |
2042 | * specification is allowed for already existing hierarchies but we | |
2043 | * can't create new one without subsys specification. | |
817929ec | 2044 | */ |
172a2c06 TH |
2045 | if (!opts.subsys_mask && !opts.none) { |
2046 | ret = -EINVAL; | |
2047 | goto out_unlock; | |
817929ec | 2048 | } |
817929ec | 2049 | |
172a2c06 TH |
2050 | root = kzalloc(sizeof(*root), GFP_KERNEL); |
2051 | if (!root) { | |
2052 | ret = -ENOMEM; | |
2bd59d48 | 2053 | goto out_unlock; |
839ec545 | 2054 | } |
e5f6a860 | 2055 | |
172a2c06 TH |
2056 | init_cgroup_root(root, &opts); |
2057 | ||
35585573 | 2058 | ret = cgroup_setup_root(root, opts.subsys_mask); |
2bd59d48 TH |
2059 | if (ret) |
2060 | cgroup_free_root(root); | |
fa3ca07e | 2061 | |
8e30e2b8 | 2062 | out_unlock: |
ddbcc7e8 | 2063 | mutex_unlock(&cgroup_mutex); |
a015edd2 | 2064 | out_free: |
c6d57f33 PM |
2065 | kfree(opts.release_agent); |
2066 | kfree(opts.name); | |
03b1cde6 | 2067 | |
2bd59d48 | 2068 | if (ret) |
8e30e2b8 | 2069 | return ERR_PTR(ret); |
2bd59d48 | 2070 | |
c9482a5b JZ |
2071 | dentry = kernfs_mount(fs_type, flags, root->kf_root, |
2072 | CGROUP_SUPER_MAGIC, &new_sb); | |
c6b3d5bc | 2073 | if (IS_ERR(dentry) || !new_sb) |
3dd06ffa | 2074 | cgroup_put(&root->cgrp); |
3a32bd72 LZ |
2075 | |
2076 | /* | |
2077 | * If @pinned_sb, we're reusing an existing root and holding an | |
2078 | * extra ref on its sb. Mount is complete. Put the extra ref. | |
2079 | */ | |
2080 | if (pinned_sb) { | |
2081 | WARN_ON(new_sb); | |
2082 | deactivate_super(pinned_sb); | |
2083 | } | |
2084 | ||
2bd59d48 TH |
2085 | return dentry; |
2086 | } | |
2087 | ||
2088 | static void cgroup_kill_sb(struct super_block *sb) | |
2089 | { | |
2090 | struct kernfs_root *kf_root = kernfs_root_from_sb(sb); | |
3dd06ffa | 2091 | struct cgroup_root *root = cgroup_root_from_kf(kf_root); |
2bd59d48 | 2092 | |
9d755d33 TH |
2093 | /* |
2094 | * If @root doesn't have any mounts or children, start killing it. | |
2095 | * This prevents new mounts by disabling percpu_ref_tryget_live(). | |
2096 | * cgroup_mount() may wait for @root's release. | |
1f779fb2 LZ |
2097 | * |
2098 | * And don't kill the default root. | |
9d755d33 | 2099 | */ |
3c606d35 | 2100 | if (!list_empty(&root->cgrp.self.children) || |
1f779fb2 | 2101 | root == &cgrp_dfl_root) |
9d755d33 TH |
2102 | cgroup_put(&root->cgrp); |
2103 | else | |
2104 | percpu_ref_kill(&root->cgrp.self.refcnt); | |
2105 | ||
2bd59d48 | 2106 | kernfs_kill_sb(sb); |
ddbcc7e8 PM |
2107 | } |
2108 | ||
2109 | static struct file_system_type cgroup_fs_type = { | |
2110 | .name = "cgroup", | |
f7e83571 | 2111 | .mount = cgroup_mount, |
ddbcc7e8 PM |
2112 | .kill_sb = cgroup_kill_sb, |
2113 | }; | |
2114 | ||
857a2beb | 2115 | /** |
913ffdb5 | 2116 | * task_cgroup_path - cgroup path of a task in the first cgroup hierarchy |
857a2beb | 2117 | * @task: target task |
857a2beb TH |
2118 | * @buf: the buffer to write the path into |
2119 | * @buflen: the length of the buffer | |
2120 | * | |
913ffdb5 TH |
2121 | * Determine @task's cgroup on the first (the one with the lowest non-zero |
2122 | * hierarchy_id) cgroup hierarchy and copy its path into @buf. This | |
2123 | * function grabs cgroup_mutex and shouldn't be used inside locks used by | |
2124 | * cgroup controller callbacks. | |
2125 | * | |
e61734c5 | 2126 | * Return value is the same as kernfs_path(). |
857a2beb | 2127 | */ |
e61734c5 | 2128 | char *task_cgroup_path(struct task_struct *task, char *buf, size_t buflen) |
857a2beb | 2129 | { |
3dd06ffa | 2130 | struct cgroup_root *root; |
913ffdb5 | 2131 | struct cgroup *cgrp; |
e61734c5 TH |
2132 | int hierarchy_id = 1; |
2133 | char *path = NULL; | |
857a2beb TH |
2134 | |
2135 | mutex_lock(&cgroup_mutex); | |
96d365e0 | 2136 | down_read(&css_set_rwsem); |
857a2beb | 2137 | |
913ffdb5 TH |
2138 | root = idr_get_next(&cgroup_hierarchy_idr, &hierarchy_id); |
2139 | ||
857a2beb TH |
2140 | if (root) { |
2141 | cgrp = task_cgroup_from_root(task, root); | |
e61734c5 | 2142 | path = cgroup_path(cgrp, buf, buflen); |
913ffdb5 TH |
2143 | } else { |
2144 | /* if no hierarchy exists, everyone is in "/" */ | |
e61734c5 TH |
2145 | if (strlcpy(buf, "/", buflen) < buflen) |
2146 | path = buf; | |
857a2beb TH |
2147 | } |
2148 | ||
96d365e0 | 2149 | up_read(&css_set_rwsem); |
857a2beb | 2150 | mutex_unlock(&cgroup_mutex); |
e61734c5 | 2151 | return path; |
857a2beb | 2152 | } |
913ffdb5 | 2153 | EXPORT_SYMBOL_GPL(task_cgroup_path); |
857a2beb | 2154 | |
b3dc094e | 2155 | /* used to track tasks and other necessary states during migration */ |
2f7ee569 | 2156 | struct cgroup_taskset { |
b3dc094e TH |
2157 | /* the src and dst cset list running through cset->mg_node */ |
2158 | struct list_head src_csets; | |
2159 | struct list_head dst_csets; | |
2160 | ||
2161 | /* | |
2162 | * Fields for cgroup_taskset_*() iteration. | |
2163 | * | |
2164 | * Before migration is committed, the target migration tasks are on | |
2165 | * ->mg_tasks of the csets on ->src_csets. After, on ->mg_tasks of | |
2166 | * the csets on ->dst_csets. ->csets point to either ->src_csets | |
2167 | * or ->dst_csets depending on whether migration is committed. | |
2168 | * | |
2169 | * ->cur_csets and ->cur_task point to the current task position | |
2170 | * during iteration. | |
2171 | */ | |
2172 | struct list_head *csets; | |
2173 | struct css_set *cur_cset; | |
2174 | struct task_struct *cur_task; | |
2f7ee569 TH |
2175 | }; |
2176 | ||
adaae5dc TH |
2177 | #define CGROUP_TASKSET_INIT(tset) (struct cgroup_taskset){ \ |
2178 | .src_csets = LIST_HEAD_INIT(tset.src_csets), \ | |
2179 | .dst_csets = LIST_HEAD_INIT(tset.dst_csets), \ | |
2180 | .csets = &tset.src_csets, \ | |
2181 | } | |
2182 | ||
2183 | /** | |
2184 | * cgroup_taskset_add - try to add a migration target task to a taskset | |
2185 | * @task: target task | |
2186 | * @tset: target taskset | |
2187 | * | |
2188 | * Add @task, which is a migration target, to @tset. This function becomes | |
2189 | * noop if @task doesn't need to be migrated. @task's css_set should have | |
2190 | * been added as a migration source and @task->cg_list will be moved from | |
2191 | * the css_set's tasks list to mg_tasks one. | |
2192 | */ | |
2193 | static void cgroup_taskset_add(struct task_struct *task, | |
2194 | struct cgroup_taskset *tset) | |
2195 | { | |
2196 | struct css_set *cset; | |
2197 | ||
2198 | lockdep_assert_held(&css_set_rwsem); | |
2199 | ||
2200 | /* @task either already exited or can't exit until the end */ | |
2201 | if (task->flags & PF_EXITING) | |
2202 | return; | |
2203 | ||
2204 | /* leave @task alone if post_fork() hasn't linked it yet */ | |
2205 | if (list_empty(&task->cg_list)) | |
2206 | return; | |
2207 | ||
2208 | cset = task_css_set(task); | |
2209 | if (!cset->mg_src_cgrp) | |
2210 | return; | |
2211 | ||
2212 | list_move_tail(&task->cg_list, &cset->mg_tasks); | |
2213 | if (list_empty(&cset->mg_node)) | |
2214 | list_add_tail(&cset->mg_node, &tset->src_csets); | |
2215 | if (list_empty(&cset->mg_dst_cset->mg_node)) | |
2216 | list_move_tail(&cset->mg_dst_cset->mg_node, | |
2217 | &tset->dst_csets); | |
2218 | } | |
2219 | ||
2f7ee569 TH |
2220 | /** |
2221 | * cgroup_taskset_first - reset taskset and return the first task | |
2222 | * @tset: taskset of interest | |
2223 | * | |
2224 | * @tset iteration is initialized and the first task is returned. | |
2225 | */ | |
2226 | struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset) | |
2227 | { | |
b3dc094e TH |
2228 | tset->cur_cset = list_first_entry(tset->csets, struct css_set, mg_node); |
2229 | tset->cur_task = NULL; | |
2230 | ||
2231 | return cgroup_taskset_next(tset); | |
2f7ee569 | 2232 | } |
2f7ee569 TH |
2233 | |
2234 | /** | |
2235 | * cgroup_taskset_next - iterate to the next task in taskset | |
2236 | * @tset: taskset of interest | |
2237 | * | |
2238 | * Return the next task in @tset. Iteration must have been initialized | |
2239 | * with cgroup_taskset_first(). | |
2240 | */ | |
2241 | struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset) | |
2242 | { | |
b3dc094e TH |
2243 | struct css_set *cset = tset->cur_cset; |
2244 | struct task_struct *task = tset->cur_task; | |
2f7ee569 | 2245 | |
b3dc094e TH |
2246 | while (&cset->mg_node != tset->csets) { |
2247 | if (!task) | |
2248 | task = list_first_entry(&cset->mg_tasks, | |
2249 | struct task_struct, cg_list); | |
2250 | else | |
2251 | task = list_next_entry(task, cg_list); | |
2f7ee569 | 2252 | |
b3dc094e TH |
2253 | if (&task->cg_list != &cset->mg_tasks) { |
2254 | tset->cur_cset = cset; | |
2255 | tset->cur_task = task; | |
2256 | return task; | |
2257 | } | |
2f7ee569 | 2258 | |
b3dc094e TH |
2259 | cset = list_next_entry(cset, mg_node); |
2260 | task = NULL; | |
2261 | } | |
2f7ee569 | 2262 | |
b3dc094e | 2263 | return NULL; |
2f7ee569 | 2264 | } |
2f7ee569 | 2265 | |
cb0f1fe9 | 2266 | /** |
74a1166d | 2267 | * cgroup_task_migrate - move a task from one cgroup to another. |
cb0f1fe9 TH |
2268 | * @tsk: the task being migrated |
2269 | * @new_cset: the new css_set @tsk is being attached to | |
74a1166d | 2270 | * |
cb0f1fe9 | 2271 | * Must be called with cgroup_mutex, threadgroup and css_set_rwsem locked. |
74a1166d | 2272 | */ |
b309e5b7 | 2273 | static void cgroup_task_migrate(struct task_struct *tsk, |
5abb8855 | 2274 | struct css_set *new_cset) |
74a1166d | 2275 | { |
5abb8855 | 2276 | struct css_set *old_cset; |
74a1166d | 2277 | |
cb0f1fe9 TH |
2278 | lockdep_assert_held(&cgroup_mutex); |
2279 | lockdep_assert_held(&css_set_rwsem); | |
2280 | ||
74a1166d | 2281 | /* |
1ed13287 TH |
2282 | * We are synchronized through cgroup_threadgroup_rwsem against |
2283 | * PF_EXITING setting such that we can't race against cgroup_exit() | |
2284 | * changing the css_set to init_css_set and dropping the old one. | |
74a1166d | 2285 | */ |
c84cdf75 | 2286 | WARN_ON_ONCE(tsk->flags & PF_EXITING); |
a8ad805c | 2287 | old_cset = task_css_set(tsk); |
74a1166d | 2288 | |
0de0942d TH |
2289 | if (!css_set_populated(new_cset)) |
2290 | css_set_update_populated(new_cset, true); | |
2291 | ||
b3dc094e | 2292 | get_css_set(new_cset); |
5abb8855 | 2293 | rcu_assign_pointer(tsk->cgroups, new_cset); |
1b9aba49 | 2294 | list_move_tail(&tsk->cg_list, &new_cset->mg_tasks); |
74a1166d | 2295 | |
0de0942d TH |
2296 | if (!css_set_populated(old_cset)) |
2297 | css_set_update_populated(old_cset, false); | |
2298 | ||
74a1166d | 2299 | /* |
5abb8855 TH |
2300 | * We just gained a reference on old_cset by taking it from the |
2301 | * task. As trading it for new_cset is protected by cgroup_mutex, | |
2302 | * we're safe to drop it here; it will be freed under RCU. | |
74a1166d | 2303 | */ |
a25eb52e | 2304 | put_css_set_locked(old_cset); |
74a1166d BB |
2305 | } |
2306 | ||
adaae5dc TH |
2307 | /** |
2308 | * cgroup_taskset_migrate - migrate a taskset to a cgroup | |
2309 | * @tset: taget taskset | |
2310 | * @dst_cgrp: destination cgroup | |
2311 | * | |
2312 | * Migrate tasks in @tset to @dst_cgrp. This function fails iff one of the | |
2313 | * ->can_attach callbacks fails and guarantees that either all or none of | |
2314 | * the tasks in @tset are migrated. @tset is consumed regardless of | |
2315 | * success. | |
2316 | */ | |
2317 | static int cgroup_taskset_migrate(struct cgroup_taskset *tset, | |
2318 | struct cgroup *dst_cgrp) | |
2319 | { | |
2320 | struct cgroup_subsys_state *css, *failed_css = NULL; | |
2321 | struct task_struct *task, *tmp_task; | |
2322 | struct css_set *cset, *tmp_cset; | |
2323 | int i, ret; | |
2324 | ||
2325 | /* methods shouldn't be called if no task is actually migrating */ | |
2326 | if (list_empty(&tset->src_csets)) | |
2327 | return 0; | |
2328 | ||
2329 | /* check that we can legitimately attach to the cgroup */ | |
2330 | for_each_e_css(css, i, dst_cgrp) { | |
2331 | if (css->ss->can_attach) { | |
2332 | ret = css->ss->can_attach(css, tset); | |
2333 | if (ret) { | |
2334 | failed_css = css; | |
2335 | goto out_cancel_attach; | |
2336 | } | |
2337 | } | |
2338 | } | |
2339 | ||
2340 | /* | |
2341 | * Now that we're guaranteed success, proceed to move all tasks to | |
2342 | * the new cgroup. There are no failure cases after here, so this | |
2343 | * is the commit point. | |
2344 | */ | |
2345 | down_write(&css_set_rwsem); | |
2346 | list_for_each_entry(cset, &tset->src_csets, mg_node) { | |
2347 | list_for_each_entry_safe(task, tmp_task, &cset->mg_tasks, cg_list) | |
b309e5b7 | 2348 | cgroup_task_migrate(task, cset->mg_dst_cset); |
adaae5dc TH |
2349 | } |
2350 | up_write(&css_set_rwsem); | |
2351 | ||
2352 | /* | |
2353 | * Migration is committed, all target tasks are now on dst_csets. | |
2354 | * Nothing is sensitive to fork() after this point. Notify | |
2355 | * controllers that migration is complete. | |
2356 | */ | |
2357 | tset->csets = &tset->dst_csets; | |
2358 | ||
2359 | for_each_e_css(css, i, dst_cgrp) | |
2360 | if (css->ss->attach) | |
2361 | css->ss->attach(css, tset); | |
2362 | ||
2363 | ret = 0; | |
2364 | goto out_release_tset; | |
2365 | ||
2366 | out_cancel_attach: | |
2367 | for_each_e_css(css, i, dst_cgrp) { | |
2368 | if (css == failed_css) | |
2369 | break; | |
2370 | if (css->ss->cancel_attach) | |
2371 | css->ss->cancel_attach(css, tset); | |
2372 | } | |
2373 | out_release_tset: | |
2374 | down_write(&css_set_rwsem); | |
2375 | list_splice_init(&tset->dst_csets, &tset->src_csets); | |
2376 | list_for_each_entry_safe(cset, tmp_cset, &tset->src_csets, mg_node) { | |
2377 | list_splice_tail_init(&cset->mg_tasks, &cset->tasks); | |
2378 | list_del_init(&cset->mg_node); | |
2379 | } | |
2380 | up_write(&css_set_rwsem); | |
2381 | return ret; | |
2382 | } | |
2383 | ||
a043e3b2 | 2384 | /** |
1958d2d5 TH |
2385 | * cgroup_migrate_finish - cleanup after attach |
2386 | * @preloaded_csets: list of preloaded css_sets | |
74a1166d | 2387 | * |
1958d2d5 TH |
2388 | * Undo cgroup_migrate_add_src() and cgroup_migrate_prepare_dst(). See |
2389 | * those functions for details. | |
74a1166d | 2390 | */ |
1958d2d5 | 2391 | static void cgroup_migrate_finish(struct list_head *preloaded_csets) |
74a1166d | 2392 | { |
1958d2d5 | 2393 | struct css_set *cset, *tmp_cset; |
74a1166d | 2394 | |
1958d2d5 TH |
2395 | lockdep_assert_held(&cgroup_mutex); |
2396 | ||
2397 | down_write(&css_set_rwsem); | |
2398 | list_for_each_entry_safe(cset, tmp_cset, preloaded_csets, mg_preload_node) { | |
2399 | cset->mg_src_cgrp = NULL; | |
2400 | cset->mg_dst_cset = NULL; | |
2401 | list_del_init(&cset->mg_preload_node); | |
a25eb52e | 2402 | put_css_set_locked(cset); |
1958d2d5 TH |
2403 | } |
2404 | up_write(&css_set_rwsem); | |
2405 | } | |
2406 | ||
2407 | /** | |
2408 | * cgroup_migrate_add_src - add a migration source css_set | |
2409 | * @src_cset: the source css_set to add | |
2410 | * @dst_cgrp: the destination cgroup | |
2411 | * @preloaded_csets: list of preloaded css_sets | |
2412 | * | |
2413 | * Tasks belonging to @src_cset are about to be migrated to @dst_cgrp. Pin | |
2414 | * @src_cset and add it to @preloaded_csets, which should later be cleaned | |
2415 | * up by cgroup_migrate_finish(). | |
2416 | * | |
1ed13287 TH |
2417 | * This function may be called without holding cgroup_threadgroup_rwsem |
2418 | * even if the target is a process. Threads may be created and destroyed | |
2419 | * but as long as cgroup_mutex is not dropped, no new css_set can be put | |
2420 | * into play and the preloaded css_sets are guaranteed to cover all | |
2421 | * migrations. | |
1958d2d5 TH |
2422 | */ |
2423 | static void cgroup_migrate_add_src(struct css_set *src_cset, | |
2424 | struct cgroup *dst_cgrp, | |
2425 | struct list_head *preloaded_csets) | |
2426 | { | |
2427 | struct cgroup *src_cgrp; | |
2428 | ||
2429 | lockdep_assert_held(&cgroup_mutex); | |
2430 | lockdep_assert_held(&css_set_rwsem); | |
2431 | ||
2432 | src_cgrp = cset_cgroup_from_root(src_cset, dst_cgrp->root); | |
2433 | ||
1958d2d5 TH |
2434 | if (!list_empty(&src_cset->mg_preload_node)) |
2435 | return; | |
2436 | ||
2437 | WARN_ON(src_cset->mg_src_cgrp); | |
2438 | WARN_ON(!list_empty(&src_cset->mg_tasks)); | |
2439 | WARN_ON(!list_empty(&src_cset->mg_node)); | |
2440 | ||
2441 | src_cset->mg_src_cgrp = src_cgrp; | |
2442 | get_css_set(src_cset); | |
2443 | list_add(&src_cset->mg_preload_node, preloaded_csets); | |
2444 | } | |
2445 | ||
2446 | /** | |
2447 | * cgroup_migrate_prepare_dst - prepare destination css_sets for migration | |
f817de98 | 2448 | * @dst_cgrp: the destination cgroup (may be %NULL) |
1958d2d5 TH |
2449 | * @preloaded_csets: list of preloaded source css_sets |
2450 | * | |
2451 | * Tasks are about to be moved to @dst_cgrp and all the source css_sets | |
2452 | * have been preloaded to @preloaded_csets. This function looks up and | |
f817de98 TH |
2453 | * pins all destination css_sets, links each to its source, and append them |
2454 | * to @preloaded_csets. If @dst_cgrp is %NULL, the destination of each | |
2455 | * source css_set is assumed to be its cgroup on the default hierarchy. | |
1958d2d5 TH |
2456 | * |
2457 | * This function must be called after cgroup_migrate_add_src() has been | |
2458 | * called on each migration source css_set. After migration is performed | |
2459 | * using cgroup_migrate(), cgroup_migrate_finish() must be called on | |
2460 | * @preloaded_csets. | |
2461 | */ | |
2462 | static int cgroup_migrate_prepare_dst(struct cgroup *dst_cgrp, | |
2463 | struct list_head *preloaded_csets) | |
2464 | { | |
2465 | LIST_HEAD(csets); | |
f817de98 | 2466 | struct css_set *src_cset, *tmp_cset; |
1958d2d5 TH |
2467 | |
2468 | lockdep_assert_held(&cgroup_mutex); | |
2469 | ||
f8f22e53 TH |
2470 | /* |
2471 | * Except for the root, child_subsys_mask must be zero for a cgroup | |
2472 | * with tasks so that child cgroups don't compete against tasks. | |
2473 | */ | |
d51f39b0 | 2474 | if (dst_cgrp && cgroup_on_dfl(dst_cgrp) && cgroup_parent(dst_cgrp) && |
f8f22e53 TH |
2475 | dst_cgrp->child_subsys_mask) |
2476 | return -EBUSY; | |
2477 | ||
1958d2d5 | 2478 | /* look up the dst cset for each src cset and link it to src */ |
f817de98 | 2479 | list_for_each_entry_safe(src_cset, tmp_cset, preloaded_csets, mg_preload_node) { |
1958d2d5 TH |
2480 | struct css_set *dst_cset; |
2481 | ||
f817de98 TH |
2482 | dst_cset = find_css_set(src_cset, |
2483 | dst_cgrp ?: src_cset->dfl_cgrp); | |
1958d2d5 TH |
2484 | if (!dst_cset) |
2485 | goto err; | |
2486 | ||
2487 | WARN_ON_ONCE(src_cset->mg_dst_cset || dst_cset->mg_dst_cset); | |
f817de98 TH |
2488 | |
2489 | /* | |
2490 | * If src cset equals dst, it's noop. Drop the src. | |
2491 | * cgroup_migrate() will skip the cset too. Note that we | |
2492 | * can't handle src == dst as some nodes are used by both. | |
2493 | */ | |
2494 | if (src_cset == dst_cset) { | |
2495 | src_cset->mg_src_cgrp = NULL; | |
2496 | list_del_init(&src_cset->mg_preload_node); | |
a25eb52e ZL |
2497 | put_css_set(src_cset); |
2498 | put_css_set(dst_cset); | |
f817de98 TH |
2499 | continue; |
2500 | } | |
2501 | ||
1958d2d5 TH |
2502 | src_cset->mg_dst_cset = dst_cset; |
2503 | ||
2504 | if (list_empty(&dst_cset->mg_preload_node)) | |
2505 | list_add(&dst_cset->mg_preload_node, &csets); | |
2506 | else | |
a25eb52e | 2507 | put_css_set(dst_cset); |
1958d2d5 TH |
2508 | } |
2509 | ||
f817de98 | 2510 | list_splice_tail(&csets, preloaded_csets); |
1958d2d5 TH |
2511 | return 0; |
2512 | err: | |
2513 | cgroup_migrate_finish(&csets); | |
2514 | return -ENOMEM; | |
2515 | } | |
2516 | ||
2517 | /** | |
2518 | * cgroup_migrate - migrate a process or task to a cgroup | |
1958d2d5 TH |
2519 | * @leader: the leader of the process or the task to migrate |
2520 | * @threadgroup: whether @leader points to the whole process or a single task | |
9af2ec45 | 2521 | * @cgrp: the destination cgroup |
1958d2d5 TH |
2522 | * |
2523 | * Migrate a process or task denoted by @leader to @cgrp. If migrating a | |
1ed13287 | 2524 | * process, the caller must be holding cgroup_threadgroup_rwsem. The |
1958d2d5 TH |
2525 | * caller is also responsible for invoking cgroup_migrate_add_src() and |
2526 | * cgroup_migrate_prepare_dst() on the targets before invoking this | |
2527 | * function and following up with cgroup_migrate_finish(). | |
2528 | * | |
2529 | * As long as a controller's ->can_attach() doesn't fail, this function is | |
2530 | * guaranteed to succeed. This means that, excluding ->can_attach() | |
2531 | * failure, when migrating multiple targets, the success or failure can be | |
2532 | * decided for all targets by invoking group_migrate_prepare_dst() before | |
2533 | * actually starting migrating. | |
2534 | */ | |
9af2ec45 TH |
2535 | static int cgroup_migrate(struct task_struct *leader, bool threadgroup, |
2536 | struct cgroup *cgrp) | |
74a1166d | 2537 | { |
adaae5dc TH |
2538 | struct cgroup_taskset tset = CGROUP_TASKSET_INIT(tset); |
2539 | struct task_struct *task; | |
74a1166d | 2540 | |
fb5d2b4c MSB |
2541 | /* |
2542 | * Prevent freeing of tasks while we take a snapshot. Tasks that are | |
2543 | * already PF_EXITING could be freed from underneath us unless we | |
2544 | * take an rcu_read_lock. | |
2545 | */ | |
b3dc094e | 2546 | down_write(&css_set_rwsem); |
fb5d2b4c | 2547 | rcu_read_lock(); |
9db8de37 | 2548 | task = leader; |
74a1166d | 2549 | do { |
adaae5dc | 2550 | cgroup_taskset_add(task, &tset); |
081aa458 LZ |
2551 | if (!threadgroup) |
2552 | break; | |
9db8de37 | 2553 | } while_each_thread(leader, task); |
fb5d2b4c | 2554 | rcu_read_unlock(); |
b3dc094e | 2555 | up_write(&css_set_rwsem); |
74a1166d | 2556 | |
adaae5dc | 2557 | return cgroup_taskset_migrate(&tset, cgrp); |
74a1166d BB |
2558 | } |
2559 | ||
1958d2d5 TH |
2560 | /** |
2561 | * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup | |
2562 | * @dst_cgrp: the cgroup to attach to | |
2563 | * @leader: the task or the leader of the threadgroup to be attached | |
2564 | * @threadgroup: attach the whole threadgroup? | |
2565 | * | |
1ed13287 | 2566 | * Call holding cgroup_mutex and cgroup_threadgroup_rwsem. |
1958d2d5 TH |
2567 | */ |
2568 | static int cgroup_attach_task(struct cgroup *dst_cgrp, | |
2569 | struct task_struct *leader, bool threadgroup) | |
2570 | { | |
2571 | LIST_HEAD(preloaded_csets); | |
2572 | struct task_struct *task; | |
2573 | int ret; | |
2574 | ||
2575 | /* look up all src csets */ | |
2576 | down_read(&css_set_rwsem); | |
2577 | rcu_read_lock(); | |
2578 | task = leader; | |
2579 | do { | |
2580 | cgroup_migrate_add_src(task_css_set(task), dst_cgrp, | |
2581 | &preloaded_csets); | |
2582 | if (!threadgroup) | |
2583 | break; | |
2584 | } while_each_thread(leader, task); | |
2585 | rcu_read_unlock(); | |
2586 | up_read(&css_set_rwsem); | |
2587 | ||
2588 | /* prepare dst csets and commit */ | |
2589 | ret = cgroup_migrate_prepare_dst(dst_cgrp, &preloaded_csets); | |
2590 | if (!ret) | |
9af2ec45 | 2591 | ret = cgroup_migrate(leader, threadgroup, dst_cgrp); |
1958d2d5 TH |
2592 | |
2593 | cgroup_migrate_finish(&preloaded_csets); | |
2594 | return ret; | |
74a1166d BB |
2595 | } |
2596 | ||
187fe840 TH |
2597 | static int cgroup_procs_write_permission(struct task_struct *task, |
2598 | struct cgroup *dst_cgrp, | |
2599 | struct kernfs_open_file *of) | |
dedf22e9 TH |
2600 | { |
2601 | const struct cred *cred = current_cred(); | |
2602 | const struct cred *tcred = get_task_cred(task); | |
2603 | int ret = 0; | |
2604 | ||
2605 | /* | |
2606 | * even if we're attaching all tasks in the thread group, we only | |
2607 | * need to check permissions on one of them. | |
2608 | */ | |
2609 | if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) && | |
2610 | !uid_eq(cred->euid, tcred->uid) && | |
2611 | !uid_eq(cred->euid, tcred->suid)) | |
2612 | ret = -EACCES; | |
2613 | ||
187fe840 TH |
2614 | if (!ret && cgroup_on_dfl(dst_cgrp)) { |
2615 | struct super_block *sb = of->file->f_path.dentry->d_sb; | |
2616 | struct cgroup *cgrp; | |
2617 | struct inode *inode; | |
2618 | ||
2619 | down_read(&css_set_rwsem); | |
2620 | cgrp = task_cgroup_from_root(task, &cgrp_dfl_root); | |
2621 | up_read(&css_set_rwsem); | |
2622 | ||
2623 | while (!cgroup_is_descendant(dst_cgrp, cgrp)) | |
2624 | cgrp = cgroup_parent(cgrp); | |
2625 | ||
2626 | ret = -ENOMEM; | |
6f60eade | 2627 | inode = kernfs_get_inode(sb, cgrp->procs_file.kn); |
187fe840 TH |
2628 | if (inode) { |
2629 | ret = inode_permission(inode, MAY_WRITE); | |
2630 | iput(inode); | |
2631 | } | |
2632 | } | |
2633 | ||
dedf22e9 TH |
2634 | put_cred(tcred); |
2635 | return ret; | |
2636 | } | |
2637 | ||
74a1166d BB |
2638 | /* |
2639 | * Find the task_struct of the task to attach by vpid and pass it along to the | |
cd3d0952 | 2640 | * function to attach either it or all tasks in its threadgroup. Will lock |
0e1d768f | 2641 | * cgroup_mutex and threadgroup. |
bbcb81d0 | 2642 | */ |
acbef755 TH |
2643 | static ssize_t __cgroup_procs_write(struct kernfs_open_file *of, char *buf, |
2644 | size_t nbytes, loff_t off, bool threadgroup) | |
bbcb81d0 | 2645 | { |
bbcb81d0 | 2646 | struct task_struct *tsk; |
e76ecaee | 2647 | struct cgroup *cgrp; |
acbef755 | 2648 | pid_t pid; |
bbcb81d0 PM |
2649 | int ret; |
2650 | ||
acbef755 TH |
2651 | if (kstrtoint(strstrip(buf), 0, &pid) || pid < 0) |
2652 | return -EINVAL; | |
2653 | ||
e76ecaee TH |
2654 | cgrp = cgroup_kn_lock_live(of->kn); |
2655 | if (!cgrp) | |
74a1166d BB |
2656 | return -ENODEV; |
2657 | ||
3014dde7 | 2658 | percpu_down_write(&cgroup_threadgroup_rwsem); |
b78949eb | 2659 | rcu_read_lock(); |
bbcb81d0 | 2660 | if (pid) { |
73507f33 | 2661 | tsk = find_task_by_vpid(pid); |
74a1166d | 2662 | if (!tsk) { |
dd4b0a46 | 2663 | ret = -ESRCH; |
3014dde7 | 2664 | goto out_unlock_rcu; |
bbcb81d0 | 2665 | } |
dedf22e9 | 2666 | } else { |
b78949eb | 2667 | tsk = current; |
dedf22e9 | 2668 | } |
cd3d0952 TH |
2669 | |
2670 | if (threadgroup) | |
b78949eb | 2671 | tsk = tsk->group_leader; |
c4c27fbd MG |
2672 | |
2673 | /* | |
14a40ffc | 2674 | * Workqueue threads may acquire PF_NO_SETAFFINITY and become |
c4c27fbd MG |
2675 | * trapped in a cpuset, or RT worker may be born in a cgroup |
2676 | * with no rt_runtime allocated. Just say no. | |
2677 | */ | |
14a40ffc | 2678 | if (tsk == kthreadd_task || (tsk->flags & PF_NO_SETAFFINITY)) { |
c4c27fbd | 2679 | ret = -EINVAL; |
3014dde7 | 2680 | goto out_unlock_rcu; |
c4c27fbd MG |
2681 | } |
2682 | ||
b78949eb MSB |
2683 | get_task_struct(tsk); |
2684 | rcu_read_unlock(); | |
2685 | ||
187fe840 | 2686 | ret = cgroup_procs_write_permission(tsk, cgrp, of); |
dedf22e9 TH |
2687 | if (!ret) |
2688 | ret = cgroup_attach_task(cgrp, tsk, threadgroup); | |
081aa458 | 2689 | |
f9f9e7b7 | 2690 | put_task_struct(tsk); |
3014dde7 TH |
2691 | goto out_unlock_threadgroup; |
2692 | ||
2693 | out_unlock_rcu: | |
2694 | rcu_read_unlock(); | |
2695 | out_unlock_threadgroup: | |
2696 | percpu_up_write(&cgroup_threadgroup_rwsem); | |
e76ecaee | 2697 | cgroup_kn_unlock(of->kn); |
acbef755 | 2698 | return ret ?: nbytes; |
bbcb81d0 PM |
2699 | } |
2700 | ||
7ae1bad9 TH |
2701 | /** |
2702 | * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from' | |
2703 | * @from: attach to all cgroups of a given task | |
2704 | * @tsk: the task to be attached | |
2705 | */ | |
2706 | int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk) | |
2707 | { | |
3dd06ffa | 2708 | struct cgroup_root *root; |
7ae1bad9 TH |
2709 | int retval = 0; |
2710 | ||
47cfcd09 | 2711 | mutex_lock(&cgroup_mutex); |
985ed670 | 2712 | for_each_root(root) { |
96d365e0 TH |
2713 | struct cgroup *from_cgrp; |
2714 | ||
3dd06ffa | 2715 | if (root == &cgrp_dfl_root) |
985ed670 TH |
2716 | continue; |
2717 | ||
96d365e0 TH |
2718 | down_read(&css_set_rwsem); |
2719 | from_cgrp = task_cgroup_from_root(from, root); | |
2720 | up_read(&css_set_rwsem); | |
7ae1bad9 | 2721 | |
6f4b7e63 | 2722 | retval = cgroup_attach_task(from_cgrp, tsk, false); |
7ae1bad9 TH |
2723 | if (retval) |
2724 | break; | |
2725 | } | |
47cfcd09 | 2726 | mutex_unlock(&cgroup_mutex); |
7ae1bad9 TH |
2727 | |
2728 | return retval; | |
2729 | } | |
2730 | EXPORT_SYMBOL_GPL(cgroup_attach_task_all); | |
2731 | ||
acbef755 TH |
2732 | static ssize_t cgroup_tasks_write(struct kernfs_open_file *of, |
2733 | char *buf, size_t nbytes, loff_t off) | |
74a1166d | 2734 | { |
acbef755 | 2735 | return __cgroup_procs_write(of, buf, nbytes, off, false); |
74a1166d BB |
2736 | } |
2737 | ||
acbef755 TH |
2738 | static ssize_t cgroup_procs_write(struct kernfs_open_file *of, |
2739 | char *buf, size_t nbytes, loff_t off) | |
af351026 | 2740 | { |
acbef755 | 2741 | return __cgroup_procs_write(of, buf, nbytes, off, true); |
af351026 PM |
2742 | } |
2743 | ||
451af504 TH |
2744 | static ssize_t cgroup_release_agent_write(struct kernfs_open_file *of, |
2745 | char *buf, size_t nbytes, loff_t off) | |
e788e066 | 2746 | { |
e76ecaee | 2747 | struct cgroup *cgrp; |
5f469907 | 2748 | |
e76ecaee | 2749 | BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX); |
5f469907 | 2750 | |
e76ecaee TH |
2751 | cgrp = cgroup_kn_lock_live(of->kn); |
2752 | if (!cgrp) | |
e788e066 | 2753 | return -ENODEV; |
69e943b7 | 2754 | spin_lock(&release_agent_path_lock); |
e76ecaee TH |
2755 | strlcpy(cgrp->root->release_agent_path, strstrip(buf), |
2756 | sizeof(cgrp->root->release_agent_path)); | |
69e943b7 | 2757 | spin_unlock(&release_agent_path_lock); |
e76ecaee | 2758 | cgroup_kn_unlock(of->kn); |
451af504 | 2759 | return nbytes; |
e788e066 PM |
2760 | } |
2761 | ||
2da8ca82 | 2762 | static int cgroup_release_agent_show(struct seq_file *seq, void *v) |
e788e066 | 2763 | { |
2da8ca82 | 2764 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
182446d0 | 2765 | |
46cfeb04 | 2766 | spin_lock(&release_agent_path_lock); |
e788e066 | 2767 | seq_puts(seq, cgrp->root->release_agent_path); |
46cfeb04 | 2768 | spin_unlock(&release_agent_path_lock); |
e788e066 | 2769 | seq_putc(seq, '\n'); |
e788e066 PM |
2770 | return 0; |
2771 | } | |
2772 | ||
2da8ca82 | 2773 | static int cgroup_sane_behavior_show(struct seq_file *seq, void *v) |
873fe09e | 2774 | { |
c1d5d42e | 2775 | seq_puts(seq, "0\n"); |
e788e066 PM |
2776 | return 0; |
2777 | } | |
2778 | ||
8ab456ac | 2779 | static void cgroup_print_ss_mask(struct seq_file *seq, unsigned long ss_mask) |
355e0c48 | 2780 | { |
f8f22e53 TH |
2781 | struct cgroup_subsys *ss; |
2782 | bool printed = false; | |
2783 | int ssid; | |
a742c59d | 2784 | |
a966a4ed AS |
2785 | for_each_subsys_which(ss, ssid, &ss_mask) { |
2786 | if (printed) | |
2787 | seq_putc(seq, ' '); | |
2788 | seq_printf(seq, "%s", ss->name); | |
2789 | printed = true; | |
e73d2c61 | 2790 | } |
f8f22e53 TH |
2791 | if (printed) |
2792 | seq_putc(seq, '\n'); | |
355e0c48 PM |
2793 | } |
2794 | ||
f8f22e53 TH |
2795 | /* show controllers which are currently attached to the default hierarchy */ |
2796 | static int cgroup_root_controllers_show(struct seq_file *seq, void *v) | |
db3b1497 | 2797 | { |
f8f22e53 TH |
2798 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
2799 | ||
5533e011 TH |
2800 | cgroup_print_ss_mask(seq, cgrp->root->subsys_mask & |
2801 | ~cgrp_dfl_root_inhibit_ss_mask); | |
f8f22e53 | 2802 | return 0; |
db3b1497 PM |
2803 | } |
2804 | ||
f8f22e53 TH |
2805 | /* show controllers which are enabled from the parent */ |
2806 | static int cgroup_controllers_show(struct seq_file *seq, void *v) | |
ddbcc7e8 | 2807 | { |
f8f22e53 TH |
2808 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
2809 | ||
667c2491 | 2810 | cgroup_print_ss_mask(seq, cgroup_parent(cgrp)->subtree_control); |
f8f22e53 | 2811 | return 0; |
ddbcc7e8 PM |
2812 | } |
2813 | ||
f8f22e53 TH |
2814 | /* show controllers which are enabled for a given cgroup's children */ |
2815 | static int cgroup_subtree_control_show(struct seq_file *seq, void *v) | |
ddbcc7e8 | 2816 | { |
f8f22e53 TH |
2817 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
2818 | ||
667c2491 | 2819 | cgroup_print_ss_mask(seq, cgrp->subtree_control); |
f8f22e53 TH |
2820 | return 0; |
2821 | } | |
2822 | ||
2823 | /** | |
2824 | * cgroup_update_dfl_csses - update css assoc of a subtree in default hierarchy | |
2825 | * @cgrp: root of the subtree to update csses for | |
2826 | * | |
2827 | * @cgrp's child_subsys_mask has changed and its subtree's (self excluded) | |
2828 | * css associations need to be updated accordingly. This function looks up | |
2829 | * all css_sets which are attached to the subtree, creates the matching | |
2830 | * updated css_sets and migrates the tasks to the new ones. | |
2831 | */ | |
2832 | static int cgroup_update_dfl_csses(struct cgroup *cgrp) | |
2833 | { | |
2834 | LIST_HEAD(preloaded_csets); | |
10265075 | 2835 | struct cgroup_taskset tset = CGROUP_TASKSET_INIT(tset); |
f8f22e53 TH |
2836 | struct cgroup_subsys_state *css; |
2837 | struct css_set *src_cset; | |
2838 | int ret; | |
2839 | ||
f8f22e53 TH |
2840 | lockdep_assert_held(&cgroup_mutex); |
2841 | ||
3014dde7 TH |
2842 | percpu_down_write(&cgroup_threadgroup_rwsem); |
2843 | ||
f8f22e53 TH |
2844 | /* look up all csses currently attached to @cgrp's subtree */ |
2845 | down_read(&css_set_rwsem); | |
2846 | css_for_each_descendant_pre(css, cgroup_css(cgrp, NULL)) { | |
2847 | struct cgrp_cset_link *link; | |
2848 | ||
2849 | /* self is not affected by child_subsys_mask change */ | |
2850 | if (css->cgroup == cgrp) | |
2851 | continue; | |
2852 | ||
2853 | list_for_each_entry(link, &css->cgroup->cset_links, cset_link) | |
2854 | cgroup_migrate_add_src(link->cset, cgrp, | |
2855 | &preloaded_csets); | |
2856 | } | |
2857 | up_read(&css_set_rwsem); | |
2858 | ||
2859 | /* NULL dst indicates self on default hierarchy */ | |
2860 | ret = cgroup_migrate_prepare_dst(NULL, &preloaded_csets); | |
2861 | if (ret) | |
2862 | goto out_finish; | |
2863 | ||
10265075 | 2864 | down_write(&css_set_rwsem); |
f8f22e53 | 2865 | list_for_each_entry(src_cset, &preloaded_csets, mg_preload_node) { |
10265075 | 2866 | struct task_struct *task, *ntask; |
f8f22e53 TH |
2867 | |
2868 | /* src_csets precede dst_csets, break on the first dst_cset */ | |
2869 | if (!src_cset->mg_src_cgrp) | |
2870 | break; | |
2871 | ||
10265075 TH |
2872 | /* all tasks in src_csets need to be migrated */ |
2873 | list_for_each_entry_safe(task, ntask, &src_cset->tasks, cg_list) | |
2874 | cgroup_taskset_add(task, &tset); | |
f8f22e53 | 2875 | } |
10265075 | 2876 | up_write(&css_set_rwsem); |
f8f22e53 | 2877 | |
10265075 | 2878 | ret = cgroup_taskset_migrate(&tset, cgrp); |
f8f22e53 TH |
2879 | out_finish: |
2880 | cgroup_migrate_finish(&preloaded_csets); | |
3014dde7 | 2881 | percpu_up_write(&cgroup_threadgroup_rwsem); |
f8f22e53 TH |
2882 | return ret; |
2883 | } | |
2884 | ||
2885 | /* change the enabled child controllers for a cgroup in the default hierarchy */ | |
451af504 TH |
2886 | static ssize_t cgroup_subtree_control_write(struct kernfs_open_file *of, |
2887 | char *buf, size_t nbytes, | |
2888 | loff_t off) | |
f8f22e53 | 2889 | { |
8ab456ac AS |
2890 | unsigned long enable = 0, disable = 0; |
2891 | unsigned long css_enable, css_disable, old_sc, new_sc, old_ss, new_ss; | |
a9746d8d | 2892 | struct cgroup *cgrp, *child; |
f8f22e53 | 2893 | struct cgroup_subsys *ss; |
451af504 | 2894 | char *tok; |
f8f22e53 TH |
2895 | int ssid, ret; |
2896 | ||
2897 | /* | |
d37167ab TH |
2898 | * Parse input - space separated list of subsystem names prefixed |
2899 | * with either + or -. | |
f8f22e53 | 2900 | */ |
451af504 TH |
2901 | buf = strstrip(buf); |
2902 | while ((tok = strsep(&buf, " "))) { | |
a966a4ed AS |
2903 | unsigned long tmp_ss_mask = ~cgrp_dfl_root_inhibit_ss_mask; |
2904 | ||
d37167ab TH |
2905 | if (tok[0] == '\0') |
2906 | continue; | |
a966a4ed | 2907 | for_each_subsys_which(ss, ssid, &tmp_ss_mask) { |
fc5ed1e9 TH |
2908 | if (!cgroup_ssid_enabled(ssid) || |
2909 | strcmp(tok + 1, ss->name)) | |
f8f22e53 TH |
2910 | continue; |
2911 | ||
2912 | if (*tok == '+') { | |
7d331fa9 TH |
2913 | enable |= 1 << ssid; |
2914 | disable &= ~(1 << ssid); | |
f8f22e53 | 2915 | } else if (*tok == '-') { |
7d331fa9 TH |
2916 | disable |= 1 << ssid; |
2917 | enable &= ~(1 << ssid); | |
f8f22e53 TH |
2918 | } else { |
2919 | return -EINVAL; | |
2920 | } | |
2921 | break; | |
2922 | } | |
2923 | if (ssid == CGROUP_SUBSYS_COUNT) | |
2924 | return -EINVAL; | |
2925 | } | |
2926 | ||
a9746d8d TH |
2927 | cgrp = cgroup_kn_lock_live(of->kn); |
2928 | if (!cgrp) | |
2929 | return -ENODEV; | |
f8f22e53 TH |
2930 | |
2931 | for_each_subsys(ss, ssid) { | |
2932 | if (enable & (1 << ssid)) { | |
667c2491 | 2933 | if (cgrp->subtree_control & (1 << ssid)) { |
f8f22e53 TH |
2934 | enable &= ~(1 << ssid); |
2935 | continue; | |
2936 | } | |
2937 | ||
c29adf24 TH |
2938 | /* unavailable or not enabled on the parent? */ |
2939 | if (!(cgrp_dfl_root.subsys_mask & (1 << ssid)) || | |
2940 | (cgroup_parent(cgrp) && | |
667c2491 | 2941 | !(cgroup_parent(cgrp)->subtree_control & (1 << ssid)))) { |
c29adf24 TH |
2942 | ret = -ENOENT; |
2943 | goto out_unlock; | |
2944 | } | |
f8f22e53 | 2945 | } else if (disable & (1 << ssid)) { |
667c2491 | 2946 | if (!(cgrp->subtree_control & (1 << ssid))) { |
f8f22e53 TH |
2947 | disable &= ~(1 << ssid); |
2948 | continue; | |
2949 | } | |
2950 | ||
2951 | /* a child has it enabled? */ | |
2952 | cgroup_for_each_live_child(child, cgrp) { | |
667c2491 | 2953 | if (child->subtree_control & (1 << ssid)) { |
f8f22e53 | 2954 | ret = -EBUSY; |
ddab2b6e | 2955 | goto out_unlock; |
f8f22e53 TH |
2956 | } |
2957 | } | |
2958 | } | |
2959 | } | |
2960 | ||
2961 | if (!enable && !disable) { | |
2962 | ret = 0; | |
ddab2b6e | 2963 | goto out_unlock; |
f8f22e53 TH |
2964 | } |
2965 | ||
2966 | /* | |
667c2491 | 2967 | * Except for the root, subtree_control must be zero for a cgroup |
f8f22e53 TH |
2968 | * with tasks so that child cgroups don't compete against tasks. |
2969 | */ | |
d51f39b0 | 2970 | if (enable && cgroup_parent(cgrp) && !list_empty(&cgrp->cset_links)) { |
f8f22e53 TH |
2971 | ret = -EBUSY; |
2972 | goto out_unlock; | |
2973 | } | |
2974 | ||
2975 | /* | |
f63070d3 TH |
2976 | * Update subsys masks and calculate what needs to be done. More |
2977 | * subsystems than specified may need to be enabled or disabled | |
2978 | * depending on subsystem dependencies. | |
2979 | */ | |
755bf5ee TH |
2980 | old_sc = cgrp->subtree_control; |
2981 | old_ss = cgrp->child_subsys_mask; | |
2982 | new_sc = (old_sc | enable) & ~disable; | |
2983 | new_ss = cgroup_calc_child_subsys_mask(cgrp, new_sc); | |
f63070d3 | 2984 | |
755bf5ee TH |
2985 | css_enable = ~old_ss & new_ss; |
2986 | css_disable = old_ss & ~new_ss; | |
f63070d3 TH |
2987 | enable |= css_enable; |
2988 | disable |= css_disable; | |
c29adf24 | 2989 | |
db6e3053 TH |
2990 | /* |
2991 | * Because css offlining is asynchronous, userland might try to | |
2992 | * re-enable the same controller while the previous instance is | |
2993 | * still around. In such cases, wait till it's gone using | |
2994 | * offline_waitq. | |
2995 | */ | |
a966a4ed | 2996 | for_each_subsys_which(ss, ssid, &css_enable) { |
db6e3053 TH |
2997 | cgroup_for_each_live_child(child, cgrp) { |
2998 | DEFINE_WAIT(wait); | |
2999 | ||
3000 | if (!cgroup_css(child, ss)) | |
3001 | continue; | |
3002 | ||
3003 | cgroup_get(child); | |
3004 | prepare_to_wait(&child->offline_waitq, &wait, | |
3005 | TASK_UNINTERRUPTIBLE); | |
3006 | cgroup_kn_unlock(of->kn); | |
3007 | schedule(); | |
3008 | finish_wait(&child->offline_waitq, &wait); | |
3009 | cgroup_put(child); | |
3010 | ||
3011 | return restart_syscall(); | |
3012 | } | |
3013 | } | |
3014 | ||
755bf5ee TH |
3015 | cgrp->subtree_control = new_sc; |
3016 | cgrp->child_subsys_mask = new_ss; | |
3017 | ||
f63070d3 TH |
3018 | /* |
3019 | * Create new csses or make the existing ones visible. A css is | |
3020 | * created invisible if it's being implicitly enabled through | |
3021 | * dependency. An invisible css is made visible when the userland | |
3022 | * explicitly enables it. | |
f8f22e53 TH |
3023 | */ |
3024 | for_each_subsys(ss, ssid) { | |
3025 | if (!(enable & (1 << ssid))) | |
3026 | continue; | |
3027 | ||
3028 | cgroup_for_each_live_child(child, cgrp) { | |
f63070d3 TH |
3029 | if (css_enable & (1 << ssid)) |
3030 | ret = create_css(child, ss, | |
3031 | cgrp->subtree_control & (1 << ssid)); | |
3032 | else | |
4df8dc90 TH |
3033 | ret = css_populate_dir(cgroup_css(child, ss), |
3034 | NULL); | |
f8f22e53 TH |
3035 | if (ret) |
3036 | goto err_undo_css; | |
3037 | } | |
3038 | } | |
3039 | ||
c29adf24 TH |
3040 | /* |
3041 | * At this point, cgroup_e_css() results reflect the new csses | |
3042 | * making the following cgroup_update_dfl_csses() properly update | |
3043 | * css associations of all tasks in the subtree. | |
3044 | */ | |
f8f22e53 TH |
3045 | ret = cgroup_update_dfl_csses(cgrp); |
3046 | if (ret) | |
3047 | goto err_undo_css; | |
3048 | ||
f63070d3 TH |
3049 | /* |
3050 | * All tasks are migrated out of disabled csses. Kill or hide | |
3051 | * them. A css is hidden when the userland requests it to be | |
b4536f0c TH |
3052 | * disabled while other subsystems are still depending on it. The |
3053 | * css must not actively control resources and be in the vanilla | |
3054 | * state if it's made visible again later. Controllers which may | |
3055 | * be depended upon should provide ->css_reset() for this purpose. | |
f63070d3 | 3056 | */ |
f8f22e53 TH |
3057 | for_each_subsys(ss, ssid) { |
3058 | if (!(disable & (1 << ssid))) | |
3059 | continue; | |
3060 | ||
f63070d3 | 3061 | cgroup_for_each_live_child(child, cgrp) { |
b4536f0c TH |
3062 | struct cgroup_subsys_state *css = cgroup_css(child, ss); |
3063 | ||
3064 | if (css_disable & (1 << ssid)) { | |
3065 | kill_css(css); | |
3066 | } else { | |
4df8dc90 | 3067 | css_clear_dir(css, NULL); |
b4536f0c TH |
3068 | if (ss->css_reset) |
3069 | ss->css_reset(css); | |
3070 | } | |
f63070d3 | 3071 | } |
f8f22e53 TH |
3072 | } |
3073 | ||
56c807ba TH |
3074 | /* |
3075 | * The effective csses of all the descendants (excluding @cgrp) may | |
3076 | * have changed. Subsystems can optionally subscribe to this event | |
3077 | * by implementing ->css_e_css_changed() which is invoked if any of | |
3078 | * the effective csses seen from the css's cgroup may have changed. | |
3079 | */ | |
3080 | for_each_subsys(ss, ssid) { | |
3081 | struct cgroup_subsys_state *this_css = cgroup_css(cgrp, ss); | |
3082 | struct cgroup_subsys_state *css; | |
3083 | ||
3084 | if (!ss->css_e_css_changed || !this_css) | |
3085 | continue; | |
3086 | ||
3087 | css_for_each_descendant_pre(css, this_css) | |
3088 | if (css != this_css) | |
3089 | ss->css_e_css_changed(css); | |
3090 | } | |
3091 | ||
f8f22e53 TH |
3092 | kernfs_activate(cgrp->kn); |
3093 | ret = 0; | |
3094 | out_unlock: | |
a9746d8d | 3095 | cgroup_kn_unlock(of->kn); |
451af504 | 3096 | return ret ?: nbytes; |
f8f22e53 TH |
3097 | |
3098 | err_undo_css: | |
755bf5ee TH |
3099 | cgrp->subtree_control = old_sc; |
3100 | cgrp->child_subsys_mask = old_ss; | |
f8f22e53 TH |
3101 | |
3102 | for_each_subsys(ss, ssid) { | |
3103 | if (!(enable & (1 << ssid))) | |
3104 | continue; | |
3105 | ||
3106 | cgroup_for_each_live_child(child, cgrp) { | |
3107 | struct cgroup_subsys_state *css = cgroup_css(child, ss); | |
f63070d3 TH |
3108 | |
3109 | if (!css) | |
3110 | continue; | |
3111 | ||
3112 | if (css_enable & (1 << ssid)) | |
f8f22e53 | 3113 | kill_css(css); |
f63070d3 | 3114 | else |
4df8dc90 | 3115 | css_clear_dir(css, NULL); |
f8f22e53 TH |
3116 | } |
3117 | } | |
3118 | goto out_unlock; | |
3119 | } | |
3120 | ||
4a07c222 | 3121 | static int cgroup_events_show(struct seq_file *seq, void *v) |
842b597e | 3122 | { |
4a07c222 TH |
3123 | seq_printf(seq, "populated %d\n", |
3124 | (bool)seq_css(seq)->cgroup->populated_cnt); | |
842b597e TH |
3125 | return 0; |
3126 | } | |
3127 | ||
2bd59d48 TH |
3128 | static ssize_t cgroup_file_write(struct kernfs_open_file *of, char *buf, |
3129 | size_t nbytes, loff_t off) | |
355e0c48 | 3130 | { |
2bd59d48 TH |
3131 | struct cgroup *cgrp = of->kn->parent->priv; |
3132 | struct cftype *cft = of->kn->priv; | |
3133 | struct cgroup_subsys_state *css; | |
a742c59d | 3134 | int ret; |
355e0c48 | 3135 | |
b4168640 TH |
3136 | if (cft->write) |
3137 | return cft->write(of, buf, nbytes, off); | |
3138 | ||
2bd59d48 TH |
3139 | /* |
3140 | * kernfs guarantees that a file isn't deleted with operations in | |
3141 | * flight, which means that the matching css is and stays alive and | |
3142 | * doesn't need to be pinned. The RCU locking is not necessary | |
3143 | * either. It's just for the convenience of using cgroup_css(). | |
3144 | */ | |
3145 | rcu_read_lock(); | |
3146 | css = cgroup_css(cgrp, cft->ss); | |
3147 | rcu_read_unlock(); | |
a742c59d | 3148 | |
451af504 | 3149 | if (cft->write_u64) { |
a742c59d TH |
3150 | unsigned long long v; |
3151 | ret = kstrtoull(buf, 0, &v); | |
3152 | if (!ret) | |
3153 | ret = cft->write_u64(css, cft, v); | |
3154 | } else if (cft->write_s64) { | |
3155 | long long v; | |
3156 | ret = kstrtoll(buf, 0, &v); | |
3157 | if (!ret) | |
3158 | ret = cft->write_s64(css, cft, v); | |
e73d2c61 | 3159 | } else { |
a742c59d | 3160 | ret = -EINVAL; |
e73d2c61 | 3161 | } |
2bd59d48 | 3162 | |
a742c59d | 3163 | return ret ?: nbytes; |
355e0c48 PM |
3164 | } |
3165 | ||
6612f05b | 3166 | static void *cgroup_seqfile_start(struct seq_file *seq, loff_t *ppos) |
db3b1497 | 3167 | { |
2bd59d48 | 3168 | return seq_cft(seq)->seq_start(seq, ppos); |
db3b1497 PM |
3169 | } |
3170 | ||
6612f05b | 3171 | static void *cgroup_seqfile_next(struct seq_file *seq, void *v, loff_t *ppos) |
ddbcc7e8 | 3172 | { |
2bd59d48 | 3173 | return seq_cft(seq)->seq_next(seq, v, ppos); |
ddbcc7e8 PM |
3174 | } |
3175 | ||
6612f05b | 3176 | static void cgroup_seqfile_stop(struct seq_file *seq, void *v) |
ddbcc7e8 | 3177 | { |
2bd59d48 | 3178 | seq_cft(seq)->seq_stop(seq, v); |
ddbcc7e8 PM |
3179 | } |
3180 | ||
91796569 | 3181 | static int cgroup_seqfile_show(struct seq_file *m, void *arg) |
e73d2c61 | 3182 | { |
7da11279 TH |
3183 | struct cftype *cft = seq_cft(m); |
3184 | struct cgroup_subsys_state *css = seq_css(m); | |
e73d2c61 | 3185 | |
2da8ca82 TH |
3186 | if (cft->seq_show) |
3187 | return cft->seq_show(m, arg); | |
e73d2c61 | 3188 | |
f4c753b7 | 3189 | if (cft->read_u64) |
896f5199 TH |
3190 | seq_printf(m, "%llu\n", cft->read_u64(css, cft)); |
3191 | else if (cft->read_s64) | |
3192 | seq_printf(m, "%lld\n", cft->read_s64(css, cft)); | |
3193 | else | |
3194 | return -EINVAL; | |
3195 | return 0; | |
91796569 PM |
3196 | } |
3197 | ||
2bd59d48 TH |
3198 | static struct kernfs_ops cgroup_kf_single_ops = { |
3199 | .atomic_write_len = PAGE_SIZE, | |
3200 | .write = cgroup_file_write, | |
3201 | .seq_show = cgroup_seqfile_show, | |
91796569 PM |
3202 | }; |
3203 | ||
2bd59d48 TH |
3204 | static struct kernfs_ops cgroup_kf_ops = { |
3205 | .atomic_write_len = PAGE_SIZE, | |
3206 | .write = cgroup_file_write, | |
3207 | .seq_start = cgroup_seqfile_start, | |
3208 | .seq_next = cgroup_seqfile_next, | |
3209 | .seq_stop = cgroup_seqfile_stop, | |
3210 | .seq_show = cgroup_seqfile_show, | |
3211 | }; | |
ddbcc7e8 PM |
3212 | |
3213 | /* | |
3214 | * cgroup_rename - Only allow simple rename of directories in place. | |
3215 | */ | |
2bd59d48 TH |
3216 | static int cgroup_rename(struct kernfs_node *kn, struct kernfs_node *new_parent, |
3217 | const char *new_name_str) | |
ddbcc7e8 | 3218 | { |
2bd59d48 | 3219 | struct cgroup *cgrp = kn->priv; |
65dff759 | 3220 | int ret; |
65dff759 | 3221 | |
2bd59d48 | 3222 | if (kernfs_type(kn) != KERNFS_DIR) |
ddbcc7e8 | 3223 | return -ENOTDIR; |
2bd59d48 | 3224 | if (kn->parent != new_parent) |
ddbcc7e8 | 3225 | return -EIO; |
65dff759 | 3226 | |
6db8e85c TH |
3227 | /* |
3228 | * This isn't a proper migration and its usefulness is very | |
aa6ec29b | 3229 | * limited. Disallow on the default hierarchy. |
6db8e85c | 3230 | */ |
aa6ec29b | 3231 | if (cgroup_on_dfl(cgrp)) |
6db8e85c | 3232 | return -EPERM; |
099fca32 | 3233 | |
e1b2dc17 | 3234 | /* |
8353da1f | 3235 | * We're gonna grab cgroup_mutex which nests outside kernfs |
e1b2dc17 | 3236 | * active_ref. kernfs_rename() doesn't require active_ref |
8353da1f | 3237 | * protection. Break them before grabbing cgroup_mutex. |
e1b2dc17 TH |
3238 | */ |
3239 | kernfs_break_active_protection(new_parent); | |
3240 | kernfs_break_active_protection(kn); | |
099fca32 | 3241 | |
2bd59d48 | 3242 | mutex_lock(&cgroup_mutex); |
099fca32 | 3243 | |
2bd59d48 | 3244 | ret = kernfs_rename(kn, new_parent, new_name_str); |
099fca32 | 3245 | |
2bd59d48 | 3246 | mutex_unlock(&cgroup_mutex); |
e1b2dc17 TH |
3247 | |
3248 | kernfs_unbreak_active_protection(kn); | |
3249 | kernfs_unbreak_active_protection(new_parent); | |
2bd59d48 | 3250 | return ret; |
099fca32 LZ |
3251 | } |
3252 | ||
49957f8e TH |
3253 | /* set uid and gid of cgroup dirs and files to that of the creator */ |
3254 | static int cgroup_kn_set_ugid(struct kernfs_node *kn) | |
3255 | { | |
3256 | struct iattr iattr = { .ia_valid = ATTR_UID | ATTR_GID, | |
3257 | .ia_uid = current_fsuid(), | |
3258 | .ia_gid = current_fsgid(), }; | |
3259 | ||
3260 | if (uid_eq(iattr.ia_uid, GLOBAL_ROOT_UID) && | |
3261 | gid_eq(iattr.ia_gid, GLOBAL_ROOT_GID)) | |
3262 | return 0; | |
3263 | ||
3264 | return kernfs_setattr(kn, &iattr); | |
3265 | } | |
3266 | ||
4df8dc90 TH |
3267 | static int cgroup_add_file(struct cgroup_subsys_state *css, struct cgroup *cgrp, |
3268 | struct cftype *cft) | |
ddbcc7e8 | 3269 | { |
8d7e6fb0 | 3270 | char name[CGROUP_FILE_NAME_MAX]; |
2bd59d48 TH |
3271 | struct kernfs_node *kn; |
3272 | struct lock_class_key *key = NULL; | |
49957f8e | 3273 | int ret; |
05ef1d7c | 3274 | |
2bd59d48 TH |
3275 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
3276 | key = &cft->lockdep_key; | |
3277 | #endif | |
3278 | kn = __kernfs_create_file(cgrp->kn, cgroup_file_name(cgrp, cft, name), | |
3279 | cgroup_file_mode(cft), 0, cft->kf_ops, cft, | |
dfeb0750 | 3280 | NULL, key); |
49957f8e TH |
3281 | if (IS_ERR(kn)) |
3282 | return PTR_ERR(kn); | |
3283 | ||
3284 | ret = cgroup_kn_set_ugid(kn); | |
f8f22e53 | 3285 | if (ret) { |
49957f8e | 3286 | kernfs_remove(kn); |
f8f22e53 TH |
3287 | return ret; |
3288 | } | |
3289 | ||
6f60eade TH |
3290 | if (cft->file_offset) { |
3291 | struct cgroup_file *cfile = (void *)css + cft->file_offset; | |
3292 | ||
3293 | kernfs_get(kn); | |
3294 | cfile->kn = kn; | |
3295 | list_add(&cfile->node, &css->files); | |
3296 | } | |
3297 | ||
f8f22e53 | 3298 | return 0; |
ddbcc7e8 PM |
3299 | } |
3300 | ||
b1f28d31 TH |
3301 | /** |
3302 | * cgroup_addrm_files - add or remove files to a cgroup directory | |
4df8dc90 TH |
3303 | * @css: the target css |
3304 | * @cgrp: the target cgroup (usually css->cgroup) | |
b1f28d31 TH |
3305 | * @cfts: array of cftypes to be added |
3306 | * @is_add: whether to add or remove | |
3307 | * | |
3308 | * Depending on @is_add, add or remove files defined by @cfts on @cgrp. | |
6732ed85 | 3309 | * For removals, this function never fails. |
b1f28d31 | 3310 | */ |
4df8dc90 TH |
3311 | static int cgroup_addrm_files(struct cgroup_subsys_state *css, |
3312 | struct cgroup *cgrp, struct cftype cfts[], | |
2bb566cb | 3313 | bool is_add) |
ddbcc7e8 | 3314 | { |
6732ed85 | 3315 | struct cftype *cft, *cft_end = NULL; |
b1f28d31 TH |
3316 | int ret; |
3317 | ||
01f6474c | 3318 | lockdep_assert_held(&cgroup_mutex); |
db0416b6 | 3319 | |
6732ed85 TH |
3320 | restart: |
3321 | for (cft = cfts; cft != cft_end && cft->name[0] != '\0'; cft++) { | |
f33fddc2 | 3322 | /* does cft->flags tell us to skip this file on @cgrp? */ |
05ebb6e6 | 3323 | if ((cft->flags & __CFTYPE_ONLY_ON_DFL) && !cgroup_on_dfl(cgrp)) |
8cbbf2c9 | 3324 | continue; |
05ebb6e6 | 3325 | if ((cft->flags & __CFTYPE_NOT_ON_DFL) && cgroup_on_dfl(cgrp)) |
873fe09e | 3326 | continue; |
d51f39b0 | 3327 | if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgroup_parent(cgrp)) |
f33fddc2 | 3328 | continue; |
d51f39b0 | 3329 | if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgroup_parent(cgrp)) |
f33fddc2 G |
3330 | continue; |
3331 | ||
2739d3cc | 3332 | if (is_add) { |
4df8dc90 | 3333 | ret = cgroup_add_file(css, cgrp, cft); |
b1f28d31 | 3334 | if (ret) { |
ed3d261b JP |
3335 | pr_warn("%s: failed to add %s, err=%d\n", |
3336 | __func__, cft->name, ret); | |
6732ed85 TH |
3337 | cft_end = cft; |
3338 | is_add = false; | |
3339 | goto restart; | |
b1f28d31 | 3340 | } |
2739d3cc LZ |
3341 | } else { |
3342 | cgroup_rm_file(cgrp, cft); | |
db0416b6 | 3343 | } |
ddbcc7e8 | 3344 | } |
b1f28d31 | 3345 | return 0; |
ddbcc7e8 PM |
3346 | } |
3347 | ||
21a2d343 | 3348 | static int cgroup_apply_cftypes(struct cftype *cfts, bool is_add) |
8e3f6541 TH |
3349 | { |
3350 | LIST_HEAD(pending); | |
2bb566cb | 3351 | struct cgroup_subsys *ss = cfts[0].ss; |
3dd06ffa | 3352 | struct cgroup *root = &ss->root->cgrp; |
492eb21b | 3353 | struct cgroup_subsys_state *css; |
9ccece80 | 3354 | int ret = 0; |
8e3f6541 | 3355 | |
01f6474c | 3356 | lockdep_assert_held(&cgroup_mutex); |
e8c82d20 | 3357 | |
e8c82d20 | 3358 | /* add/rm files for all cgroups created before */ |
ca8bdcaf | 3359 | css_for_each_descendant_pre(css, cgroup_css(root, ss)) { |
492eb21b TH |
3360 | struct cgroup *cgrp = css->cgroup; |
3361 | ||
e8c82d20 LZ |
3362 | if (cgroup_is_dead(cgrp)) |
3363 | continue; | |
3364 | ||
4df8dc90 | 3365 | ret = cgroup_addrm_files(css, cgrp, cfts, is_add); |
9ccece80 TH |
3366 | if (ret) |
3367 | break; | |
8e3f6541 | 3368 | } |
21a2d343 TH |
3369 | |
3370 | if (is_add && !ret) | |
3371 | kernfs_activate(root->kn); | |
9ccece80 | 3372 | return ret; |
8e3f6541 TH |
3373 | } |
3374 | ||
2da440a2 | 3375 | static void cgroup_exit_cftypes(struct cftype *cfts) |
8e3f6541 | 3376 | { |
2bb566cb | 3377 | struct cftype *cft; |
8e3f6541 | 3378 | |
2bd59d48 TH |
3379 | for (cft = cfts; cft->name[0] != '\0'; cft++) { |
3380 | /* free copy for custom atomic_write_len, see init_cftypes() */ | |
3381 | if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) | |
3382 | kfree(cft->kf_ops); | |
3383 | cft->kf_ops = NULL; | |
2da440a2 | 3384 | cft->ss = NULL; |
a8ddc821 TH |
3385 | |
3386 | /* revert flags set by cgroup core while adding @cfts */ | |
05ebb6e6 | 3387 | cft->flags &= ~(__CFTYPE_ONLY_ON_DFL | __CFTYPE_NOT_ON_DFL); |
2bd59d48 | 3388 | } |
2da440a2 TH |
3389 | } |
3390 | ||
2bd59d48 | 3391 | static int cgroup_init_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
2da440a2 TH |
3392 | { |
3393 | struct cftype *cft; | |
3394 | ||
2bd59d48 TH |
3395 | for (cft = cfts; cft->name[0] != '\0'; cft++) { |
3396 | struct kernfs_ops *kf_ops; | |
3397 | ||
0adb0704 TH |
3398 | WARN_ON(cft->ss || cft->kf_ops); |
3399 | ||
2bd59d48 TH |
3400 | if (cft->seq_start) |
3401 | kf_ops = &cgroup_kf_ops; | |
3402 | else | |
3403 | kf_ops = &cgroup_kf_single_ops; | |
3404 | ||
3405 | /* | |
3406 | * Ugh... if @cft wants a custom max_write_len, we need to | |
3407 | * make a copy of kf_ops to set its atomic_write_len. | |
3408 | */ | |
3409 | if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) { | |
3410 | kf_ops = kmemdup(kf_ops, sizeof(*kf_ops), GFP_KERNEL); | |
3411 | if (!kf_ops) { | |
3412 | cgroup_exit_cftypes(cfts); | |
3413 | return -ENOMEM; | |
3414 | } | |
3415 | kf_ops->atomic_write_len = cft->max_write_len; | |
3416 | } | |
8e3f6541 | 3417 | |
2bd59d48 | 3418 | cft->kf_ops = kf_ops; |
2bb566cb | 3419 | cft->ss = ss; |
2bd59d48 | 3420 | } |
2bb566cb | 3421 | |
2bd59d48 | 3422 | return 0; |
2da440a2 TH |
3423 | } |
3424 | ||
21a2d343 TH |
3425 | static int cgroup_rm_cftypes_locked(struct cftype *cfts) |
3426 | { | |
01f6474c | 3427 | lockdep_assert_held(&cgroup_mutex); |
21a2d343 TH |
3428 | |
3429 | if (!cfts || !cfts[0].ss) | |
3430 | return -ENOENT; | |
3431 | ||
3432 | list_del(&cfts->node); | |
3433 | cgroup_apply_cftypes(cfts, false); | |
3434 | cgroup_exit_cftypes(cfts); | |
3435 | return 0; | |
8e3f6541 | 3436 | } |
8e3f6541 | 3437 | |
79578621 TH |
3438 | /** |
3439 | * cgroup_rm_cftypes - remove an array of cftypes from a subsystem | |
79578621 TH |
3440 | * @cfts: zero-length name terminated array of cftypes |
3441 | * | |
2bb566cb TH |
3442 | * Unregister @cfts. Files described by @cfts are removed from all |
3443 | * existing cgroups and all future cgroups won't have them either. This | |
3444 | * function can be called anytime whether @cfts' subsys is attached or not. | |
79578621 TH |
3445 | * |
3446 | * Returns 0 on successful unregistration, -ENOENT if @cfts is not | |
2bb566cb | 3447 | * registered. |
79578621 | 3448 | */ |
2bb566cb | 3449 | int cgroup_rm_cftypes(struct cftype *cfts) |
79578621 | 3450 | { |
21a2d343 | 3451 | int ret; |
79578621 | 3452 | |
01f6474c | 3453 | mutex_lock(&cgroup_mutex); |
21a2d343 | 3454 | ret = cgroup_rm_cftypes_locked(cfts); |
01f6474c | 3455 | mutex_unlock(&cgroup_mutex); |
21a2d343 | 3456 | return ret; |
80b13586 TH |
3457 | } |
3458 | ||
8e3f6541 TH |
3459 | /** |
3460 | * cgroup_add_cftypes - add an array of cftypes to a subsystem | |
3461 | * @ss: target cgroup subsystem | |
3462 | * @cfts: zero-length name terminated array of cftypes | |
3463 | * | |
3464 | * Register @cfts to @ss. Files described by @cfts are created for all | |
3465 | * existing cgroups to which @ss is attached and all future cgroups will | |
3466 | * have them too. This function can be called anytime whether @ss is | |
3467 | * attached or not. | |
3468 | * | |
3469 | * Returns 0 on successful registration, -errno on failure. Note that this | |
3470 | * function currently returns 0 as long as @cfts registration is successful | |
3471 | * even if some file creation attempts on existing cgroups fail. | |
3472 | */ | |
2cf669a5 | 3473 | static int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
8e3f6541 | 3474 | { |
9ccece80 | 3475 | int ret; |
8e3f6541 | 3476 | |
fc5ed1e9 | 3477 | if (!cgroup_ssid_enabled(ss->id)) |
c731ae1d LZ |
3478 | return 0; |
3479 | ||
dc5736ed LZ |
3480 | if (!cfts || cfts[0].name[0] == '\0') |
3481 | return 0; | |
2bb566cb | 3482 | |
2bd59d48 TH |
3483 | ret = cgroup_init_cftypes(ss, cfts); |
3484 | if (ret) | |
3485 | return ret; | |
79578621 | 3486 | |
01f6474c | 3487 | mutex_lock(&cgroup_mutex); |
21a2d343 | 3488 | |
0adb0704 | 3489 | list_add_tail(&cfts->node, &ss->cfts); |
21a2d343 | 3490 | ret = cgroup_apply_cftypes(cfts, true); |
9ccece80 | 3491 | if (ret) |
21a2d343 | 3492 | cgroup_rm_cftypes_locked(cfts); |
79578621 | 3493 | |
01f6474c | 3494 | mutex_unlock(&cgroup_mutex); |
9ccece80 | 3495 | return ret; |
79578621 TH |
3496 | } |
3497 | ||
a8ddc821 TH |
3498 | /** |
3499 | * cgroup_add_dfl_cftypes - add an array of cftypes for default hierarchy | |
3500 | * @ss: target cgroup subsystem | |
3501 | * @cfts: zero-length name terminated array of cftypes | |
3502 | * | |
3503 | * Similar to cgroup_add_cftypes() but the added files are only used for | |
3504 | * the default hierarchy. | |
3505 | */ | |
3506 | int cgroup_add_dfl_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) | |
3507 | { | |
3508 | struct cftype *cft; | |
3509 | ||
3510 | for (cft = cfts; cft && cft->name[0] != '\0'; cft++) | |
05ebb6e6 | 3511 | cft->flags |= __CFTYPE_ONLY_ON_DFL; |
a8ddc821 TH |
3512 | return cgroup_add_cftypes(ss, cfts); |
3513 | } | |
3514 | ||
3515 | /** | |
3516 | * cgroup_add_legacy_cftypes - add an array of cftypes for legacy hierarchies | |
3517 | * @ss: target cgroup subsystem | |
3518 | * @cfts: zero-length name terminated array of cftypes | |
3519 | * | |
3520 | * Similar to cgroup_add_cftypes() but the added files are only used for | |
3521 | * the legacy hierarchies. | |
3522 | */ | |
2cf669a5 TH |
3523 | int cgroup_add_legacy_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
3524 | { | |
a8ddc821 TH |
3525 | struct cftype *cft; |
3526 | ||
fa8137be VG |
3527 | /* |
3528 | * If legacy_flies_on_dfl, we want to show the legacy files on the | |
3529 | * dfl hierarchy but iff the target subsystem hasn't been updated | |
3530 | * for the dfl hierarchy yet. | |
3531 | */ | |
3532 | if (!cgroup_legacy_files_on_dfl || | |
3533 | ss->dfl_cftypes != ss->legacy_cftypes) { | |
3534 | for (cft = cfts; cft && cft->name[0] != '\0'; cft++) | |
3535 | cft->flags |= __CFTYPE_NOT_ON_DFL; | |
3536 | } | |
3537 | ||
2cf669a5 TH |
3538 | return cgroup_add_cftypes(ss, cfts); |
3539 | } | |
3540 | ||
a043e3b2 LZ |
3541 | /** |
3542 | * cgroup_task_count - count the number of tasks in a cgroup. | |
3543 | * @cgrp: the cgroup in question | |
3544 | * | |
3545 | * Return the number of tasks in the cgroup. | |
3546 | */ | |
07bc356e | 3547 | static int cgroup_task_count(const struct cgroup *cgrp) |
bbcb81d0 PM |
3548 | { |
3549 | int count = 0; | |
69d0206c | 3550 | struct cgrp_cset_link *link; |
817929ec | 3551 | |
96d365e0 | 3552 | down_read(&css_set_rwsem); |
69d0206c TH |
3553 | list_for_each_entry(link, &cgrp->cset_links, cset_link) |
3554 | count += atomic_read(&link->cset->refcount); | |
96d365e0 | 3555 | up_read(&css_set_rwsem); |
bbcb81d0 PM |
3556 | return count; |
3557 | } | |
3558 | ||
53fa5261 | 3559 | /** |
492eb21b | 3560 | * css_next_child - find the next child of a given css |
c2931b70 TH |
3561 | * @pos: the current position (%NULL to initiate traversal) |
3562 | * @parent: css whose children to walk | |
53fa5261 | 3563 | * |
c2931b70 | 3564 | * This function returns the next child of @parent and should be called |
87fb54f1 | 3565 | * under either cgroup_mutex or RCU read lock. The only requirement is |
c2931b70 TH |
3566 | * that @parent and @pos are accessible. The next sibling is guaranteed to |
3567 | * be returned regardless of their states. | |
3568 | * | |
3569 | * If a subsystem synchronizes ->css_online() and the start of iteration, a | |
3570 | * css which finished ->css_online() is guaranteed to be visible in the | |
3571 | * future iterations and will stay visible until the last reference is put. | |
3572 | * A css which hasn't finished ->css_online() or already finished | |
3573 | * ->css_offline() may show up during traversal. It's each subsystem's | |
3574 | * responsibility to synchronize against on/offlining. | |
53fa5261 | 3575 | */ |
c2931b70 TH |
3576 | struct cgroup_subsys_state *css_next_child(struct cgroup_subsys_state *pos, |
3577 | struct cgroup_subsys_state *parent) | |
53fa5261 | 3578 | { |
c2931b70 | 3579 | struct cgroup_subsys_state *next; |
53fa5261 | 3580 | |
8353da1f | 3581 | cgroup_assert_mutex_or_rcu_locked(); |
53fa5261 TH |
3582 | |
3583 | /* | |
de3f0341 TH |
3584 | * @pos could already have been unlinked from the sibling list. |
3585 | * Once a cgroup is removed, its ->sibling.next is no longer | |
3586 | * updated when its next sibling changes. CSS_RELEASED is set when | |
3587 | * @pos is taken off list, at which time its next pointer is valid, | |
3588 | * and, as releases are serialized, the one pointed to by the next | |
3589 | * pointer is guaranteed to not have started release yet. This | |
3590 | * implies that if we observe !CSS_RELEASED on @pos in this RCU | |
3591 | * critical section, the one pointed to by its next pointer is | |
3592 | * guaranteed to not have finished its RCU grace period even if we | |
3593 | * have dropped rcu_read_lock() inbetween iterations. | |
3b287a50 | 3594 | * |
de3f0341 TH |
3595 | * If @pos has CSS_RELEASED set, its next pointer can't be |
3596 | * dereferenced; however, as each css is given a monotonically | |
3597 | * increasing unique serial number and always appended to the | |
3598 | * sibling list, the next one can be found by walking the parent's | |
3599 | * children until the first css with higher serial number than | |
3600 | * @pos's. While this path can be slower, it happens iff iteration | |
3601 | * races against release and the race window is very small. | |
53fa5261 | 3602 | */ |
3b287a50 | 3603 | if (!pos) { |
c2931b70 TH |
3604 | next = list_entry_rcu(parent->children.next, struct cgroup_subsys_state, sibling); |
3605 | } else if (likely(!(pos->flags & CSS_RELEASED))) { | |
3606 | next = list_entry_rcu(pos->sibling.next, struct cgroup_subsys_state, sibling); | |
3b287a50 | 3607 | } else { |
c2931b70 | 3608 | list_for_each_entry_rcu(next, &parent->children, sibling) |
3b287a50 TH |
3609 | if (next->serial_nr > pos->serial_nr) |
3610 | break; | |
53fa5261 TH |
3611 | } |
3612 | ||
3b281afb TH |
3613 | /* |
3614 | * @next, if not pointing to the head, can be dereferenced and is | |
c2931b70 | 3615 | * the next sibling. |
3b281afb | 3616 | */ |
c2931b70 TH |
3617 | if (&next->sibling != &parent->children) |
3618 | return next; | |
3b281afb | 3619 | return NULL; |
53fa5261 | 3620 | } |
53fa5261 | 3621 | |
574bd9f7 | 3622 | /** |
492eb21b | 3623 | * css_next_descendant_pre - find the next descendant for pre-order walk |
574bd9f7 | 3624 | * @pos: the current position (%NULL to initiate traversal) |
492eb21b | 3625 | * @root: css whose descendants to walk |
574bd9f7 | 3626 | * |
492eb21b | 3627 | * To be used by css_for_each_descendant_pre(). Find the next descendant |
bd8815a6 TH |
3628 | * to visit for pre-order traversal of @root's descendants. @root is |
3629 | * included in the iteration and the first node to be visited. | |
75501a6d | 3630 | * |
87fb54f1 TH |
3631 | * While this function requires cgroup_mutex or RCU read locking, it |
3632 | * doesn't require the whole traversal to be contained in a single critical | |
3633 | * section. This function will return the correct next descendant as long | |
3634 | * as both @pos and @root are accessible and @pos is a descendant of @root. | |
c2931b70 TH |
3635 | * |
3636 | * If a subsystem synchronizes ->css_online() and the start of iteration, a | |
3637 | * css which finished ->css_online() is guaranteed to be visible in the | |
3638 | * future iterations and will stay visible until the last reference is put. | |
3639 | * A css which hasn't finished ->css_online() or already finished | |
3640 | * ->css_offline() may show up during traversal. It's each subsystem's | |
3641 | * responsibility to synchronize against on/offlining. | |
574bd9f7 | 3642 | */ |
492eb21b TH |
3643 | struct cgroup_subsys_state * |
3644 | css_next_descendant_pre(struct cgroup_subsys_state *pos, | |
3645 | struct cgroup_subsys_state *root) | |
574bd9f7 | 3646 | { |
492eb21b | 3647 | struct cgroup_subsys_state *next; |
574bd9f7 | 3648 | |
8353da1f | 3649 | cgroup_assert_mutex_or_rcu_locked(); |
574bd9f7 | 3650 | |
bd8815a6 | 3651 | /* if first iteration, visit @root */ |
7805d000 | 3652 | if (!pos) |
bd8815a6 | 3653 | return root; |
574bd9f7 TH |
3654 | |
3655 | /* visit the first child if exists */ | |
492eb21b | 3656 | next = css_next_child(NULL, pos); |
574bd9f7 TH |
3657 | if (next) |
3658 | return next; | |
3659 | ||
3660 | /* no child, visit my or the closest ancestor's next sibling */ | |
492eb21b | 3661 | while (pos != root) { |
5c9d535b | 3662 | next = css_next_child(pos, pos->parent); |
75501a6d | 3663 | if (next) |
574bd9f7 | 3664 | return next; |
5c9d535b | 3665 | pos = pos->parent; |
7805d000 | 3666 | } |
574bd9f7 TH |
3667 | |
3668 | return NULL; | |
3669 | } | |
574bd9f7 | 3670 | |
12a9d2fe | 3671 | /** |
492eb21b TH |
3672 | * css_rightmost_descendant - return the rightmost descendant of a css |
3673 | * @pos: css of interest | |
12a9d2fe | 3674 | * |
492eb21b TH |
3675 | * Return the rightmost descendant of @pos. If there's no descendant, @pos |
3676 | * is returned. This can be used during pre-order traversal to skip | |
12a9d2fe | 3677 | * subtree of @pos. |
75501a6d | 3678 | * |
87fb54f1 TH |
3679 | * While this function requires cgroup_mutex or RCU read locking, it |
3680 | * doesn't require the whole traversal to be contained in a single critical | |
3681 | * section. This function will return the correct rightmost descendant as | |
3682 | * long as @pos is accessible. | |
12a9d2fe | 3683 | */ |
492eb21b TH |
3684 | struct cgroup_subsys_state * |
3685 | css_rightmost_descendant(struct cgroup_subsys_state *pos) | |
12a9d2fe | 3686 | { |
492eb21b | 3687 | struct cgroup_subsys_state *last, *tmp; |
12a9d2fe | 3688 | |
8353da1f | 3689 | cgroup_assert_mutex_or_rcu_locked(); |
12a9d2fe TH |
3690 | |
3691 | do { | |
3692 | last = pos; | |
3693 | /* ->prev isn't RCU safe, walk ->next till the end */ | |
3694 | pos = NULL; | |
492eb21b | 3695 | css_for_each_child(tmp, last) |
12a9d2fe TH |
3696 | pos = tmp; |
3697 | } while (pos); | |
3698 | ||
3699 | return last; | |
3700 | } | |
12a9d2fe | 3701 | |
492eb21b TH |
3702 | static struct cgroup_subsys_state * |
3703 | css_leftmost_descendant(struct cgroup_subsys_state *pos) | |
574bd9f7 | 3704 | { |
492eb21b | 3705 | struct cgroup_subsys_state *last; |
574bd9f7 TH |
3706 | |
3707 | do { | |
3708 | last = pos; | |
492eb21b | 3709 | pos = css_next_child(NULL, pos); |
574bd9f7 TH |
3710 | } while (pos); |
3711 | ||
3712 | return last; | |
3713 | } | |
3714 | ||
3715 | /** | |
492eb21b | 3716 | * css_next_descendant_post - find the next descendant for post-order walk |
574bd9f7 | 3717 | * @pos: the current position (%NULL to initiate traversal) |
492eb21b | 3718 | * @root: css whose descendants to walk |
574bd9f7 | 3719 | * |
492eb21b | 3720 | * To be used by css_for_each_descendant_post(). Find the next descendant |
bd8815a6 TH |
3721 | * to visit for post-order traversal of @root's descendants. @root is |
3722 | * included in the iteration and the last node to be visited. | |
75501a6d | 3723 | * |
87fb54f1 TH |
3724 | * While this function requires cgroup_mutex or RCU read locking, it |
3725 | * doesn't require the whole traversal to be contained in a single critical | |
3726 | * section. This function will return the correct next descendant as long | |
3727 | * as both @pos and @cgroup are accessible and @pos is a descendant of | |
3728 | * @cgroup. | |
c2931b70 TH |
3729 | * |
3730 | * If a subsystem synchronizes ->css_online() and the start of iteration, a | |
3731 | * css which finished ->css_online() is guaranteed to be visible in the | |
3732 | * future iterations and will stay visible until the last reference is put. | |
3733 | * A css which hasn't finished ->css_online() or already finished | |
3734 | * ->css_offline() may show up during traversal. It's each subsystem's | |
3735 | * responsibility to synchronize against on/offlining. | |
574bd9f7 | 3736 | */ |
492eb21b TH |
3737 | struct cgroup_subsys_state * |
3738 | css_next_descendant_post(struct cgroup_subsys_state *pos, | |
3739 | struct cgroup_subsys_state *root) | |
574bd9f7 | 3740 | { |
492eb21b | 3741 | struct cgroup_subsys_state *next; |
574bd9f7 | 3742 | |
8353da1f | 3743 | cgroup_assert_mutex_or_rcu_locked(); |
574bd9f7 | 3744 | |
58b79a91 TH |
3745 | /* if first iteration, visit leftmost descendant which may be @root */ |
3746 | if (!pos) | |
3747 | return css_leftmost_descendant(root); | |
574bd9f7 | 3748 | |
bd8815a6 TH |
3749 | /* if we visited @root, we're done */ |
3750 | if (pos == root) | |
3751 | return NULL; | |
3752 | ||
574bd9f7 | 3753 | /* if there's an unvisited sibling, visit its leftmost descendant */ |
5c9d535b | 3754 | next = css_next_child(pos, pos->parent); |
75501a6d | 3755 | if (next) |
492eb21b | 3756 | return css_leftmost_descendant(next); |
574bd9f7 TH |
3757 | |
3758 | /* no sibling left, visit parent */ | |
5c9d535b | 3759 | return pos->parent; |
574bd9f7 | 3760 | } |
574bd9f7 | 3761 | |
f3d46500 TH |
3762 | /** |
3763 | * css_has_online_children - does a css have online children | |
3764 | * @css: the target css | |
3765 | * | |
3766 | * Returns %true if @css has any online children; otherwise, %false. This | |
3767 | * function can be called from any context but the caller is responsible | |
3768 | * for synchronizing against on/offlining as necessary. | |
3769 | */ | |
3770 | bool css_has_online_children(struct cgroup_subsys_state *css) | |
cbc125ef | 3771 | { |
f3d46500 TH |
3772 | struct cgroup_subsys_state *child; |
3773 | bool ret = false; | |
cbc125ef TH |
3774 | |
3775 | rcu_read_lock(); | |
f3d46500 | 3776 | css_for_each_child(child, css) { |
99bae5f9 | 3777 | if (child->flags & CSS_ONLINE) { |
f3d46500 TH |
3778 | ret = true; |
3779 | break; | |
cbc125ef TH |
3780 | } |
3781 | } | |
3782 | rcu_read_unlock(); | |
f3d46500 | 3783 | return ret; |
574bd9f7 | 3784 | } |
574bd9f7 | 3785 | |
0942eeee | 3786 | /** |
72ec7029 | 3787 | * css_advance_task_iter - advance a task itererator to the next css_set |
0942eeee TH |
3788 | * @it: the iterator to advance |
3789 | * | |
3790 | * Advance @it to the next css_set to walk. | |
d515876e | 3791 | */ |
72ec7029 | 3792 | static void css_advance_task_iter(struct css_task_iter *it) |
d515876e | 3793 | { |
0f0a2b4f | 3794 | struct list_head *l = it->cset_pos; |
d515876e TH |
3795 | struct cgrp_cset_link *link; |
3796 | struct css_set *cset; | |
3797 | ||
3798 | /* Advance to the next non-empty css_set */ | |
3799 | do { | |
3800 | l = l->next; | |
0f0a2b4f TH |
3801 | if (l == it->cset_head) { |
3802 | it->cset_pos = NULL; | |
d515876e TH |
3803 | return; |
3804 | } | |
3ebb2b6e TH |
3805 | |
3806 | if (it->ss) { | |
3807 | cset = container_of(l, struct css_set, | |
3808 | e_cset_node[it->ss->id]); | |
3809 | } else { | |
3810 | link = list_entry(l, struct cgrp_cset_link, cset_link); | |
3811 | cset = link->cset; | |
3812 | } | |
0de0942d | 3813 | } while (!css_set_populated(cset)); |
c7561128 | 3814 | |
0f0a2b4f | 3815 | it->cset_pos = l; |
c7561128 TH |
3816 | |
3817 | if (!list_empty(&cset->tasks)) | |
0f0a2b4f | 3818 | it->task_pos = cset->tasks.next; |
c7561128 | 3819 | else |
0f0a2b4f TH |
3820 | it->task_pos = cset->mg_tasks.next; |
3821 | ||
3822 | it->tasks_head = &cset->tasks; | |
3823 | it->mg_tasks_head = &cset->mg_tasks; | |
d515876e TH |
3824 | } |
3825 | ||
0942eeee | 3826 | /** |
72ec7029 TH |
3827 | * css_task_iter_start - initiate task iteration |
3828 | * @css: the css to walk tasks of | |
0942eeee TH |
3829 | * @it: the task iterator to use |
3830 | * | |
72ec7029 TH |
3831 | * Initiate iteration through the tasks of @css. The caller can call |
3832 | * css_task_iter_next() to walk through the tasks until the function | |
3833 | * returns NULL. On completion of iteration, css_task_iter_end() must be | |
3834 | * called. | |
0942eeee TH |
3835 | * |
3836 | * Note that this function acquires a lock which is released when the | |
3837 | * iteration finishes. The caller can't sleep while iteration is in | |
3838 | * progress. | |
3839 | */ | |
72ec7029 TH |
3840 | void css_task_iter_start(struct cgroup_subsys_state *css, |
3841 | struct css_task_iter *it) | |
96d365e0 | 3842 | __acquires(css_set_rwsem) |
817929ec | 3843 | { |
56fde9e0 TH |
3844 | /* no one should try to iterate before mounting cgroups */ |
3845 | WARN_ON_ONCE(!use_task_css_set_links); | |
31a7df01 | 3846 | |
96d365e0 | 3847 | down_read(&css_set_rwsem); |
c59cd3d8 | 3848 | |
3ebb2b6e TH |
3849 | it->ss = css->ss; |
3850 | ||
3851 | if (it->ss) | |
3852 | it->cset_pos = &css->cgroup->e_csets[css->ss->id]; | |
3853 | else | |
3854 | it->cset_pos = &css->cgroup->cset_links; | |
3855 | ||
0f0a2b4f | 3856 | it->cset_head = it->cset_pos; |
c59cd3d8 | 3857 | |
72ec7029 | 3858 | css_advance_task_iter(it); |
817929ec PM |
3859 | } |
3860 | ||
0942eeee | 3861 | /** |
72ec7029 | 3862 | * css_task_iter_next - return the next task for the iterator |
0942eeee TH |
3863 | * @it: the task iterator being iterated |
3864 | * | |
3865 | * The "next" function for task iteration. @it should have been | |
72ec7029 TH |
3866 | * initialized via css_task_iter_start(). Returns NULL when the iteration |
3867 | * reaches the end. | |
0942eeee | 3868 | */ |
72ec7029 | 3869 | struct task_struct *css_task_iter_next(struct css_task_iter *it) |
817929ec PM |
3870 | { |
3871 | struct task_struct *res; | |
0f0a2b4f | 3872 | struct list_head *l = it->task_pos; |
817929ec PM |
3873 | |
3874 | /* If the iterator cg is NULL, we have no tasks */ | |
0f0a2b4f | 3875 | if (!it->cset_pos) |
817929ec PM |
3876 | return NULL; |
3877 | res = list_entry(l, struct task_struct, cg_list); | |
c7561128 TH |
3878 | |
3879 | /* | |
3880 | * Advance iterator to find next entry. cset->tasks is consumed | |
3881 | * first and then ->mg_tasks. After ->mg_tasks, we move onto the | |
3882 | * next cset. | |
3883 | */ | |
817929ec | 3884 | l = l->next; |
c7561128 | 3885 | |
0f0a2b4f TH |
3886 | if (l == it->tasks_head) |
3887 | l = it->mg_tasks_head->next; | |
c7561128 | 3888 | |
0f0a2b4f | 3889 | if (l == it->mg_tasks_head) |
72ec7029 | 3890 | css_advance_task_iter(it); |
c7561128 | 3891 | else |
0f0a2b4f | 3892 | it->task_pos = l; |
c7561128 | 3893 | |
817929ec PM |
3894 | return res; |
3895 | } | |
3896 | ||
0942eeee | 3897 | /** |
72ec7029 | 3898 | * css_task_iter_end - finish task iteration |
0942eeee TH |
3899 | * @it: the task iterator to finish |
3900 | * | |
72ec7029 | 3901 | * Finish task iteration started by css_task_iter_start(). |
0942eeee | 3902 | */ |
72ec7029 | 3903 | void css_task_iter_end(struct css_task_iter *it) |
96d365e0 | 3904 | __releases(css_set_rwsem) |
31a7df01 | 3905 | { |
96d365e0 | 3906 | up_read(&css_set_rwsem); |
31a7df01 CW |
3907 | } |
3908 | ||
3909 | /** | |
8cc99345 TH |
3910 | * cgroup_trasnsfer_tasks - move tasks from one cgroup to another |
3911 | * @to: cgroup to which the tasks will be moved | |
3912 | * @from: cgroup in which the tasks currently reside | |
31a7df01 | 3913 | * |
eaf797ab TH |
3914 | * Locking rules between cgroup_post_fork() and the migration path |
3915 | * guarantee that, if a task is forking while being migrated, the new child | |
3916 | * is guaranteed to be either visible in the source cgroup after the | |
3917 | * parent's migration is complete or put into the target cgroup. No task | |
3918 | * can slip out of migration through forking. | |
31a7df01 | 3919 | */ |
8cc99345 | 3920 | int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from) |
31a7df01 | 3921 | { |
952aaa12 TH |
3922 | LIST_HEAD(preloaded_csets); |
3923 | struct cgrp_cset_link *link; | |
72ec7029 | 3924 | struct css_task_iter it; |
e406d1cf | 3925 | struct task_struct *task; |
952aaa12 | 3926 | int ret; |
31a7df01 | 3927 | |
952aaa12 | 3928 | mutex_lock(&cgroup_mutex); |
31a7df01 | 3929 | |
952aaa12 TH |
3930 | /* all tasks in @from are being moved, all csets are source */ |
3931 | down_read(&css_set_rwsem); | |
3932 | list_for_each_entry(link, &from->cset_links, cset_link) | |
3933 | cgroup_migrate_add_src(link->cset, to, &preloaded_csets); | |
3934 | up_read(&css_set_rwsem); | |
31a7df01 | 3935 | |
952aaa12 TH |
3936 | ret = cgroup_migrate_prepare_dst(to, &preloaded_csets); |
3937 | if (ret) | |
3938 | goto out_err; | |
8cc99345 | 3939 | |
952aaa12 TH |
3940 | /* |
3941 | * Migrate tasks one-by-one until @form is empty. This fails iff | |
3942 | * ->can_attach() fails. | |
3943 | */ | |
e406d1cf | 3944 | do { |
9d800df1 | 3945 | css_task_iter_start(&from->self, &it); |
e406d1cf TH |
3946 | task = css_task_iter_next(&it); |
3947 | if (task) | |
3948 | get_task_struct(task); | |
3949 | css_task_iter_end(&it); | |
3950 | ||
3951 | if (task) { | |
9af2ec45 | 3952 | ret = cgroup_migrate(task, false, to); |
e406d1cf TH |
3953 | put_task_struct(task); |
3954 | } | |
3955 | } while (task && !ret); | |
952aaa12 TH |
3956 | out_err: |
3957 | cgroup_migrate_finish(&preloaded_csets); | |
47cfcd09 | 3958 | mutex_unlock(&cgroup_mutex); |
e406d1cf | 3959 | return ret; |
8cc99345 TH |
3960 | } |
3961 | ||
bbcb81d0 | 3962 | /* |
102a775e | 3963 | * Stuff for reading the 'tasks'/'procs' files. |
bbcb81d0 PM |
3964 | * |
3965 | * Reading this file can return large amounts of data if a cgroup has | |
3966 | * *lots* of attached tasks. So it may need several calls to read(), | |
3967 | * but we cannot guarantee that the information we produce is correct | |
3968 | * unless we produce it entirely atomically. | |
3969 | * | |
bbcb81d0 | 3970 | */ |
bbcb81d0 | 3971 | |
24528255 LZ |
3972 | /* which pidlist file are we talking about? */ |
3973 | enum cgroup_filetype { | |
3974 | CGROUP_FILE_PROCS, | |
3975 | CGROUP_FILE_TASKS, | |
3976 | }; | |
3977 | ||
3978 | /* | |
3979 | * A pidlist is a list of pids that virtually represents the contents of one | |
3980 | * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists, | |
3981 | * a pair (one each for procs, tasks) for each pid namespace that's relevant | |
3982 | * to the cgroup. | |
3983 | */ | |
3984 | struct cgroup_pidlist { | |
3985 | /* | |
3986 | * used to find which pidlist is wanted. doesn't change as long as | |
3987 | * this particular list stays in the list. | |
3988 | */ | |
3989 | struct { enum cgroup_filetype type; struct pid_namespace *ns; } key; | |
3990 | /* array of xids */ | |
3991 | pid_t *list; | |
3992 | /* how many elements the above list has */ | |
3993 | int length; | |
24528255 LZ |
3994 | /* each of these stored in a list by its cgroup */ |
3995 | struct list_head links; | |
3996 | /* pointer to the cgroup we belong to, for list removal purposes */ | |
3997 | struct cgroup *owner; | |
b1a21367 TH |
3998 | /* for delayed destruction */ |
3999 | struct delayed_work destroy_dwork; | |
24528255 LZ |
4000 | }; |
4001 | ||
d1d9fd33 BB |
4002 | /* |
4003 | * The following two functions "fix" the issue where there are more pids | |
4004 | * than kmalloc will give memory for; in such cases, we use vmalloc/vfree. | |
4005 | * TODO: replace with a kernel-wide solution to this problem | |
4006 | */ | |
4007 | #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2)) | |
4008 | static void *pidlist_allocate(int count) | |
4009 | { | |
4010 | if (PIDLIST_TOO_LARGE(count)) | |
4011 | return vmalloc(count * sizeof(pid_t)); | |
4012 | else | |
4013 | return kmalloc(count * sizeof(pid_t), GFP_KERNEL); | |
4014 | } | |
b1a21367 | 4015 | |
d1d9fd33 BB |
4016 | static void pidlist_free(void *p) |
4017 | { | |
58794514 | 4018 | kvfree(p); |
d1d9fd33 | 4019 | } |
d1d9fd33 | 4020 | |
b1a21367 TH |
4021 | /* |
4022 | * Used to destroy all pidlists lingering waiting for destroy timer. None | |
4023 | * should be left afterwards. | |
4024 | */ | |
4025 | static void cgroup_pidlist_destroy_all(struct cgroup *cgrp) | |
4026 | { | |
4027 | struct cgroup_pidlist *l, *tmp_l; | |
4028 | ||
4029 | mutex_lock(&cgrp->pidlist_mutex); | |
4030 | list_for_each_entry_safe(l, tmp_l, &cgrp->pidlists, links) | |
4031 | mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, 0); | |
4032 | mutex_unlock(&cgrp->pidlist_mutex); | |
4033 | ||
4034 | flush_workqueue(cgroup_pidlist_destroy_wq); | |
4035 | BUG_ON(!list_empty(&cgrp->pidlists)); | |
4036 | } | |
4037 | ||
4038 | static void cgroup_pidlist_destroy_work_fn(struct work_struct *work) | |
4039 | { | |
4040 | struct delayed_work *dwork = to_delayed_work(work); | |
4041 | struct cgroup_pidlist *l = container_of(dwork, struct cgroup_pidlist, | |
4042 | destroy_dwork); | |
4043 | struct cgroup_pidlist *tofree = NULL; | |
4044 | ||
4045 | mutex_lock(&l->owner->pidlist_mutex); | |
b1a21367 TH |
4046 | |
4047 | /* | |
04502365 TH |
4048 | * Destroy iff we didn't get queued again. The state won't change |
4049 | * as destroy_dwork can only be queued while locked. | |
b1a21367 | 4050 | */ |
04502365 | 4051 | if (!delayed_work_pending(dwork)) { |
b1a21367 TH |
4052 | list_del(&l->links); |
4053 | pidlist_free(l->list); | |
4054 | put_pid_ns(l->key.ns); | |
4055 | tofree = l; | |
4056 | } | |
4057 | ||
b1a21367 TH |
4058 | mutex_unlock(&l->owner->pidlist_mutex); |
4059 | kfree(tofree); | |
4060 | } | |
4061 | ||
bbcb81d0 | 4062 | /* |
102a775e | 4063 | * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries |
6ee211ad | 4064 | * Returns the number of unique elements. |
bbcb81d0 | 4065 | */ |
6ee211ad | 4066 | static int pidlist_uniq(pid_t *list, int length) |
bbcb81d0 | 4067 | { |
102a775e | 4068 | int src, dest = 1; |
102a775e BB |
4069 | |
4070 | /* | |
4071 | * we presume the 0th element is unique, so i starts at 1. trivial | |
4072 | * edge cases first; no work needs to be done for either | |
4073 | */ | |
4074 | if (length == 0 || length == 1) | |
4075 | return length; | |
4076 | /* src and dest walk down the list; dest counts unique elements */ | |
4077 | for (src = 1; src < length; src++) { | |
4078 | /* find next unique element */ | |
4079 | while (list[src] == list[src-1]) { | |
4080 | src++; | |
4081 | if (src == length) | |
4082 | goto after; | |
4083 | } | |
4084 | /* dest always points to where the next unique element goes */ | |
4085 | list[dest] = list[src]; | |
4086 | dest++; | |
4087 | } | |
4088 | after: | |
102a775e BB |
4089 | return dest; |
4090 | } | |
4091 | ||
afb2bc14 TH |
4092 | /* |
4093 | * The two pid files - task and cgroup.procs - guaranteed that the result | |
4094 | * is sorted, which forced this whole pidlist fiasco. As pid order is | |
4095 | * different per namespace, each namespace needs differently sorted list, | |
4096 | * making it impossible to use, for example, single rbtree of member tasks | |
4097 | * sorted by task pointer. As pidlists can be fairly large, allocating one | |
4098 | * per open file is dangerous, so cgroup had to implement shared pool of | |
4099 | * pidlists keyed by cgroup and namespace. | |
4100 | * | |
4101 | * All this extra complexity was caused by the original implementation | |
4102 | * committing to an entirely unnecessary property. In the long term, we | |
aa6ec29b TH |
4103 | * want to do away with it. Explicitly scramble sort order if on the |
4104 | * default hierarchy so that no such expectation exists in the new | |
4105 | * interface. | |
afb2bc14 TH |
4106 | * |
4107 | * Scrambling is done by swapping every two consecutive bits, which is | |
4108 | * non-identity one-to-one mapping which disturbs sort order sufficiently. | |
4109 | */ | |
4110 | static pid_t pid_fry(pid_t pid) | |
4111 | { | |
4112 | unsigned a = pid & 0x55555555; | |
4113 | unsigned b = pid & 0xAAAAAAAA; | |
4114 | ||
4115 | return (a << 1) | (b >> 1); | |
4116 | } | |
4117 | ||
4118 | static pid_t cgroup_pid_fry(struct cgroup *cgrp, pid_t pid) | |
4119 | { | |
aa6ec29b | 4120 | if (cgroup_on_dfl(cgrp)) |
afb2bc14 TH |
4121 | return pid_fry(pid); |
4122 | else | |
4123 | return pid; | |
4124 | } | |
4125 | ||
102a775e BB |
4126 | static int cmppid(const void *a, const void *b) |
4127 | { | |
4128 | return *(pid_t *)a - *(pid_t *)b; | |
4129 | } | |
4130 | ||
afb2bc14 TH |
4131 | static int fried_cmppid(const void *a, const void *b) |
4132 | { | |
4133 | return pid_fry(*(pid_t *)a) - pid_fry(*(pid_t *)b); | |
4134 | } | |
4135 | ||
e6b81710 TH |
4136 | static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp, |
4137 | enum cgroup_filetype type) | |
4138 | { | |
4139 | struct cgroup_pidlist *l; | |
4140 | /* don't need task_nsproxy() if we're looking at ourself */ | |
4141 | struct pid_namespace *ns = task_active_pid_ns(current); | |
4142 | ||
4143 | lockdep_assert_held(&cgrp->pidlist_mutex); | |
4144 | ||
4145 | list_for_each_entry(l, &cgrp->pidlists, links) | |
4146 | if (l->key.type == type && l->key.ns == ns) | |
4147 | return l; | |
4148 | return NULL; | |
4149 | } | |
4150 | ||
72a8cb30 BB |
4151 | /* |
4152 | * find the appropriate pidlist for our purpose (given procs vs tasks) | |
4153 | * returns with the lock on that pidlist already held, and takes care | |
4154 | * of the use count, or returns NULL with no locks held if we're out of | |
4155 | * memory. | |
4156 | */ | |
e6b81710 TH |
4157 | static struct cgroup_pidlist *cgroup_pidlist_find_create(struct cgroup *cgrp, |
4158 | enum cgroup_filetype type) | |
72a8cb30 BB |
4159 | { |
4160 | struct cgroup_pidlist *l; | |
b70cc5fd | 4161 | |
e6b81710 TH |
4162 | lockdep_assert_held(&cgrp->pidlist_mutex); |
4163 | ||
4164 | l = cgroup_pidlist_find(cgrp, type); | |
4165 | if (l) | |
4166 | return l; | |
4167 | ||
72a8cb30 | 4168 | /* entry not found; create a new one */ |
f4f4be2b | 4169 | l = kzalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL); |
e6b81710 | 4170 | if (!l) |
72a8cb30 | 4171 | return l; |
e6b81710 | 4172 | |
b1a21367 | 4173 | INIT_DELAYED_WORK(&l->destroy_dwork, cgroup_pidlist_destroy_work_fn); |
72a8cb30 | 4174 | l->key.type = type; |
e6b81710 TH |
4175 | /* don't need task_nsproxy() if we're looking at ourself */ |
4176 | l->key.ns = get_pid_ns(task_active_pid_ns(current)); | |
72a8cb30 BB |
4177 | l->owner = cgrp; |
4178 | list_add(&l->links, &cgrp->pidlists); | |
72a8cb30 BB |
4179 | return l; |
4180 | } | |
4181 | ||
102a775e BB |
4182 | /* |
4183 | * Load a cgroup's pidarray with either procs' tgids or tasks' pids | |
4184 | */ | |
72a8cb30 BB |
4185 | static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type, |
4186 | struct cgroup_pidlist **lp) | |
102a775e BB |
4187 | { |
4188 | pid_t *array; | |
4189 | int length; | |
4190 | int pid, n = 0; /* used for populating the array */ | |
72ec7029 | 4191 | struct css_task_iter it; |
817929ec | 4192 | struct task_struct *tsk; |
102a775e BB |
4193 | struct cgroup_pidlist *l; |
4194 | ||
4bac00d1 TH |
4195 | lockdep_assert_held(&cgrp->pidlist_mutex); |
4196 | ||
102a775e BB |
4197 | /* |
4198 | * If cgroup gets more users after we read count, we won't have | |
4199 | * enough space - tough. This race is indistinguishable to the | |
4200 | * caller from the case that the additional cgroup users didn't | |
4201 | * show up until sometime later on. | |
4202 | */ | |
4203 | length = cgroup_task_count(cgrp); | |
d1d9fd33 | 4204 | array = pidlist_allocate(length); |
102a775e BB |
4205 | if (!array) |
4206 | return -ENOMEM; | |
4207 | /* now, populate the array */ | |
9d800df1 | 4208 | css_task_iter_start(&cgrp->self, &it); |
72ec7029 | 4209 | while ((tsk = css_task_iter_next(&it))) { |
102a775e | 4210 | if (unlikely(n == length)) |
817929ec | 4211 | break; |
102a775e | 4212 | /* get tgid or pid for procs or tasks file respectively */ |
72a8cb30 BB |
4213 | if (type == CGROUP_FILE_PROCS) |
4214 | pid = task_tgid_vnr(tsk); | |
4215 | else | |
4216 | pid = task_pid_vnr(tsk); | |
102a775e BB |
4217 | if (pid > 0) /* make sure to only use valid results */ |
4218 | array[n++] = pid; | |
817929ec | 4219 | } |
72ec7029 | 4220 | css_task_iter_end(&it); |
102a775e BB |
4221 | length = n; |
4222 | /* now sort & (if procs) strip out duplicates */ | |
aa6ec29b | 4223 | if (cgroup_on_dfl(cgrp)) |
afb2bc14 TH |
4224 | sort(array, length, sizeof(pid_t), fried_cmppid, NULL); |
4225 | else | |
4226 | sort(array, length, sizeof(pid_t), cmppid, NULL); | |
72a8cb30 | 4227 | if (type == CGROUP_FILE_PROCS) |
6ee211ad | 4228 | length = pidlist_uniq(array, length); |
e6b81710 | 4229 | |
e6b81710 | 4230 | l = cgroup_pidlist_find_create(cgrp, type); |
72a8cb30 | 4231 | if (!l) { |
d1d9fd33 | 4232 | pidlist_free(array); |
72a8cb30 | 4233 | return -ENOMEM; |
102a775e | 4234 | } |
e6b81710 TH |
4235 | |
4236 | /* store array, freeing old if necessary */ | |
d1d9fd33 | 4237 | pidlist_free(l->list); |
102a775e BB |
4238 | l->list = array; |
4239 | l->length = length; | |
72a8cb30 | 4240 | *lp = l; |
102a775e | 4241 | return 0; |
bbcb81d0 PM |
4242 | } |
4243 | ||
846c7bb0 | 4244 | /** |
a043e3b2 | 4245 | * cgroupstats_build - build and fill cgroupstats |
846c7bb0 BS |
4246 | * @stats: cgroupstats to fill information into |
4247 | * @dentry: A dentry entry belonging to the cgroup for which stats have | |
4248 | * been requested. | |
a043e3b2 LZ |
4249 | * |
4250 | * Build and fill cgroupstats so that taskstats can export it to user | |
4251 | * space. | |
846c7bb0 BS |
4252 | */ |
4253 | int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry) | |
4254 | { | |
2bd59d48 | 4255 | struct kernfs_node *kn = kernfs_node_from_dentry(dentry); |
bd89aabc | 4256 | struct cgroup *cgrp; |
72ec7029 | 4257 | struct css_task_iter it; |
846c7bb0 | 4258 | struct task_struct *tsk; |
33d283be | 4259 | |
2bd59d48 TH |
4260 | /* it should be kernfs_node belonging to cgroupfs and is a directory */ |
4261 | if (dentry->d_sb->s_type != &cgroup_fs_type || !kn || | |
4262 | kernfs_type(kn) != KERNFS_DIR) | |
4263 | return -EINVAL; | |
4264 | ||
bad34660 LZ |
4265 | mutex_lock(&cgroup_mutex); |
4266 | ||
846c7bb0 | 4267 | /* |
2bd59d48 | 4268 | * We aren't being called from kernfs and there's no guarantee on |
ec903c0c | 4269 | * @kn->priv's validity. For this and css_tryget_online_from_dir(), |
2bd59d48 | 4270 | * @kn->priv is RCU safe. Let's do the RCU dancing. |
846c7bb0 | 4271 | */ |
2bd59d48 TH |
4272 | rcu_read_lock(); |
4273 | cgrp = rcu_dereference(kn->priv); | |
bad34660 | 4274 | if (!cgrp || cgroup_is_dead(cgrp)) { |
2bd59d48 | 4275 | rcu_read_unlock(); |
bad34660 | 4276 | mutex_unlock(&cgroup_mutex); |
2bd59d48 TH |
4277 | return -ENOENT; |
4278 | } | |
bad34660 | 4279 | rcu_read_unlock(); |
846c7bb0 | 4280 | |
9d800df1 | 4281 | css_task_iter_start(&cgrp->self, &it); |
72ec7029 | 4282 | while ((tsk = css_task_iter_next(&it))) { |
846c7bb0 BS |
4283 | switch (tsk->state) { |
4284 | case TASK_RUNNING: | |
4285 | stats->nr_running++; | |
4286 | break; | |
4287 | case TASK_INTERRUPTIBLE: | |
4288 | stats->nr_sleeping++; | |
4289 | break; | |
4290 | case TASK_UNINTERRUPTIBLE: | |
4291 | stats->nr_uninterruptible++; | |
4292 | break; | |
4293 | case TASK_STOPPED: | |
4294 | stats->nr_stopped++; | |
4295 | break; | |
4296 | default: | |
4297 | if (delayacct_is_task_waiting_on_io(tsk)) | |
4298 | stats->nr_io_wait++; | |
4299 | break; | |
4300 | } | |
4301 | } | |
72ec7029 | 4302 | css_task_iter_end(&it); |
846c7bb0 | 4303 | |
bad34660 | 4304 | mutex_unlock(&cgroup_mutex); |
2bd59d48 | 4305 | return 0; |
846c7bb0 BS |
4306 | } |
4307 | ||
8f3ff208 | 4308 | |
bbcb81d0 | 4309 | /* |
102a775e | 4310 | * seq_file methods for the tasks/procs files. The seq_file position is the |
cc31edce | 4311 | * next pid to display; the seq_file iterator is a pointer to the pid |
102a775e | 4312 | * in the cgroup->l->list array. |
bbcb81d0 | 4313 | */ |
cc31edce | 4314 | |
102a775e | 4315 | static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos) |
bbcb81d0 | 4316 | { |
cc31edce PM |
4317 | /* |
4318 | * Initially we receive a position value that corresponds to | |
4319 | * one more than the last pid shown (or 0 on the first call or | |
4320 | * after a seek to the start). Use a binary-search to find the | |
4321 | * next pid to display, if any | |
4322 | */ | |
2bd59d48 | 4323 | struct kernfs_open_file *of = s->private; |
7da11279 | 4324 | struct cgroup *cgrp = seq_css(s)->cgroup; |
4bac00d1 | 4325 | struct cgroup_pidlist *l; |
7da11279 | 4326 | enum cgroup_filetype type = seq_cft(s)->private; |
cc31edce | 4327 | int index = 0, pid = *pos; |
4bac00d1 TH |
4328 | int *iter, ret; |
4329 | ||
4330 | mutex_lock(&cgrp->pidlist_mutex); | |
4331 | ||
4332 | /* | |
5d22444f | 4333 | * !NULL @of->priv indicates that this isn't the first start() |
4bac00d1 | 4334 | * after open. If the matching pidlist is around, we can use that. |
5d22444f | 4335 | * Look for it. Note that @of->priv can't be used directly. It |
4bac00d1 TH |
4336 | * could already have been destroyed. |
4337 | */ | |
5d22444f TH |
4338 | if (of->priv) |
4339 | of->priv = cgroup_pidlist_find(cgrp, type); | |
4bac00d1 TH |
4340 | |
4341 | /* | |
4342 | * Either this is the first start() after open or the matching | |
4343 | * pidlist has been destroyed inbetween. Create a new one. | |
4344 | */ | |
5d22444f TH |
4345 | if (!of->priv) { |
4346 | ret = pidlist_array_load(cgrp, type, | |
4347 | (struct cgroup_pidlist **)&of->priv); | |
4bac00d1 TH |
4348 | if (ret) |
4349 | return ERR_PTR(ret); | |
4350 | } | |
5d22444f | 4351 | l = of->priv; |
cc31edce | 4352 | |
cc31edce | 4353 | if (pid) { |
102a775e | 4354 | int end = l->length; |
20777766 | 4355 | |
cc31edce PM |
4356 | while (index < end) { |
4357 | int mid = (index + end) / 2; | |
afb2bc14 | 4358 | if (cgroup_pid_fry(cgrp, l->list[mid]) == pid) { |
cc31edce PM |
4359 | index = mid; |
4360 | break; | |
afb2bc14 | 4361 | } else if (cgroup_pid_fry(cgrp, l->list[mid]) <= pid) |
cc31edce PM |
4362 | index = mid + 1; |
4363 | else | |
4364 | end = mid; | |
4365 | } | |
4366 | } | |
4367 | /* If we're off the end of the array, we're done */ | |
102a775e | 4368 | if (index >= l->length) |
cc31edce PM |
4369 | return NULL; |
4370 | /* Update the abstract position to be the actual pid that we found */ | |
102a775e | 4371 | iter = l->list + index; |
afb2bc14 | 4372 | *pos = cgroup_pid_fry(cgrp, *iter); |
cc31edce PM |
4373 | return iter; |
4374 | } | |
4375 | ||
102a775e | 4376 | static void cgroup_pidlist_stop(struct seq_file *s, void *v) |
cc31edce | 4377 | { |
2bd59d48 | 4378 | struct kernfs_open_file *of = s->private; |
5d22444f | 4379 | struct cgroup_pidlist *l = of->priv; |
62236858 | 4380 | |
5d22444f TH |
4381 | if (l) |
4382 | mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, | |
04502365 | 4383 | CGROUP_PIDLIST_DESTROY_DELAY); |
7da11279 | 4384 | mutex_unlock(&seq_css(s)->cgroup->pidlist_mutex); |
cc31edce PM |
4385 | } |
4386 | ||
102a775e | 4387 | static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos) |
cc31edce | 4388 | { |
2bd59d48 | 4389 | struct kernfs_open_file *of = s->private; |
5d22444f | 4390 | struct cgroup_pidlist *l = of->priv; |
102a775e BB |
4391 | pid_t *p = v; |
4392 | pid_t *end = l->list + l->length; | |
cc31edce PM |
4393 | /* |
4394 | * Advance to the next pid in the array. If this goes off the | |
4395 | * end, we're done | |
4396 | */ | |
4397 | p++; | |
4398 | if (p >= end) { | |
4399 | return NULL; | |
4400 | } else { | |
7da11279 | 4401 | *pos = cgroup_pid_fry(seq_css(s)->cgroup, *p); |
cc31edce PM |
4402 | return p; |
4403 | } | |
4404 | } | |
4405 | ||
102a775e | 4406 | static int cgroup_pidlist_show(struct seq_file *s, void *v) |
cc31edce | 4407 | { |
94ff212d JP |
4408 | seq_printf(s, "%d\n", *(int *)v); |
4409 | ||
4410 | return 0; | |
cc31edce | 4411 | } |
bbcb81d0 | 4412 | |
182446d0 TH |
4413 | static u64 cgroup_read_notify_on_release(struct cgroup_subsys_state *css, |
4414 | struct cftype *cft) | |
81a6a5cd | 4415 | { |
182446d0 | 4416 | return notify_on_release(css->cgroup); |
81a6a5cd PM |
4417 | } |
4418 | ||
182446d0 TH |
4419 | static int cgroup_write_notify_on_release(struct cgroup_subsys_state *css, |
4420 | struct cftype *cft, u64 val) | |
6379c106 | 4421 | { |
6379c106 | 4422 | if (val) |
182446d0 | 4423 | set_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags); |
6379c106 | 4424 | else |
182446d0 | 4425 | clear_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags); |
6379c106 PM |
4426 | return 0; |
4427 | } | |
4428 | ||
182446d0 TH |
4429 | static u64 cgroup_clone_children_read(struct cgroup_subsys_state *css, |
4430 | struct cftype *cft) | |
97978e6d | 4431 | { |
182446d0 | 4432 | return test_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d DL |
4433 | } |
4434 | ||
182446d0 TH |
4435 | static int cgroup_clone_children_write(struct cgroup_subsys_state *css, |
4436 | struct cftype *cft, u64 val) | |
97978e6d DL |
4437 | { |
4438 | if (val) | |
182446d0 | 4439 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d | 4440 | else |
182446d0 | 4441 | clear_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d DL |
4442 | return 0; |
4443 | } | |
4444 | ||
a14c6874 TH |
4445 | /* cgroup core interface files for the default hierarchy */ |
4446 | static struct cftype cgroup_dfl_base_files[] = { | |
81a6a5cd | 4447 | { |
d5c56ced | 4448 | .name = "cgroup.procs", |
6f60eade | 4449 | .file_offset = offsetof(struct cgroup, procs_file), |
6612f05b TH |
4450 | .seq_start = cgroup_pidlist_start, |
4451 | .seq_next = cgroup_pidlist_next, | |
4452 | .seq_stop = cgroup_pidlist_stop, | |
4453 | .seq_show = cgroup_pidlist_show, | |
5d22444f | 4454 | .private = CGROUP_FILE_PROCS, |
acbef755 | 4455 | .write = cgroup_procs_write, |
102a775e | 4456 | }, |
f8f22e53 TH |
4457 | { |
4458 | .name = "cgroup.controllers", | |
a14c6874 | 4459 | .flags = CFTYPE_ONLY_ON_ROOT, |
f8f22e53 TH |
4460 | .seq_show = cgroup_root_controllers_show, |
4461 | }, | |
4462 | { | |
4463 | .name = "cgroup.controllers", | |
a14c6874 | 4464 | .flags = CFTYPE_NOT_ON_ROOT, |
f8f22e53 TH |
4465 | .seq_show = cgroup_controllers_show, |
4466 | }, | |
4467 | { | |
4468 | .name = "cgroup.subtree_control", | |
f8f22e53 | 4469 | .seq_show = cgroup_subtree_control_show, |
451af504 | 4470 | .write = cgroup_subtree_control_write, |
f8f22e53 | 4471 | }, |
842b597e | 4472 | { |
4a07c222 | 4473 | .name = "cgroup.events", |
a14c6874 | 4474 | .flags = CFTYPE_NOT_ON_ROOT, |
6f60eade | 4475 | .file_offset = offsetof(struct cgroup, events_file), |
4a07c222 | 4476 | .seq_show = cgroup_events_show, |
842b597e | 4477 | }, |
a14c6874 TH |
4478 | { } /* terminate */ |
4479 | }; | |
d5c56ced | 4480 | |
a14c6874 TH |
4481 | /* cgroup core interface files for the legacy hierarchies */ |
4482 | static struct cftype cgroup_legacy_base_files[] = { | |
4483 | { | |
4484 | .name = "cgroup.procs", | |
4485 | .seq_start = cgroup_pidlist_start, | |
4486 | .seq_next = cgroup_pidlist_next, | |
4487 | .seq_stop = cgroup_pidlist_stop, | |
4488 | .seq_show = cgroup_pidlist_show, | |
4489 | .private = CGROUP_FILE_PROCS, | |
4490 | .write = cgroup_procs_write, | |
a14c6874 TH |
4491 | }, |
4492 | { | |
4493 | .name = "cgroup.clone_children", | |
4494 | .read_u64 = cgroup_clone_children_read, | |
4495 | .write_u64 = cgroup_clone_children_write, | |
4496 | }, | |
4497 | { | |
4498 | .name = "cgroup.sane_behavior", | |
4499 | .flags = CFTYPE_ONLY_ON_ROOT, | |
4500 | .seq_show = cgroup_sane_behavior_show, | |
4501 | }, | |
d5c56ced TH |
4502 | { |
4503 | .name = "tasks", | |
6612f05b TH |
4504 | .seq_start = cgroup_pidlist_start, |
4505 | .seq_next = cgroup_pidlist_next, | |
4506 | .seq_stop = cgroup_pidlist_stop, | |
4507 | .seq_show = cgroup_pidlist_show, | |
5d22444f | 4508 | .private = CGROUP_FILE_TASKS, |
acbef755 | 4509 | .write = cgroup_tasks_write, |
d5c56ced TH |
4510 | }, |
4511 | { | |
4512 | .name = "notify_on_release", | |
d5c56ced TH |
4513 | .read_u64 = cgroup_read_notify_on_release, |
4514 | .write_u64 = cgroup_write_notify_on_release, | |
4515 | }, | |
6e6ff25b TH |
4516 | { |
4517 | .name = "release_agent", | |
a14c6874 | 4518 | .flags = CFTYPE_ONLY_ON_ROOT, |
2da8ca82 | 4519 | .seq_show = cgroup_release_agent_show, |
451af504 | 4520 | .write = cgroup_release_agent_write, |
5f469907 | 4521 | .max_write_len = PATH_MAX - 1, |
6e6ff25b | 4522 | }, |
db0416b6 | 4523 | { } /* terminate */ |
bbcb81d0 PM |
4524 | }; |
4525 | ||
0c21ead1 TH |
4526 | /* |
4527 | * css destruction is four-stage process. | |
4528 | * | |
4529 | * 1. Destruction starts. Killing of the percpu_ref is initiated. | |
4530 | * Implemented in kill_css(). | |
4531 | * | |
4532 | * 2. When the percpu_ref is confirmed to be visible as killed on all CPUs | |
ec903c0c TH |
4533 | * and thus css_tryget_online() is guaranteed to fail, the css can be |
4534 | * offlined by invoking offline_css(). After offlining, the base ref is | |
4535 | * put. Implemented in css_killed_work_fn(). | |
0c21ead1 TH |
4536 | * |
4537 | * 3. When the percpu_ref reaches zero, the only possible remaining | |
4538 | * accessors are inside RCU read sections. css_release() schedules the | |
4539 | * RCU callback. | |
4540 | * | |
4541 | * 4. After the grace period, the css can be freed. Implemented in | |
4542 | * css_free_work_fn(). | |
4543 | * | |
4544 | * It is actually hairier because both step 2 and 4 require process context | |
4545 | * and thus involve punting to css->destroy_work adding two additional | |
4546 | * steps to the already complex sequence. | |
4547 | */ | |
35ef10da | 4548 | static void css_free_work_fn(struct work_struct *work) |
48ddbe19 TH |
4549 | { |
4550 | struct cgroup_subsys_state *css = | |
35ef10da | 4551 | container_of(work, struct cgroup_subsys_state, destroy_work); |
01e58659 | 4552 | struct cgroup_subsys *ss = css->ss; |
0c21ead1 | 4553 | struct cgroup *cgrp = css->cgroup; |
6f60eade | 4554 | struct cgroup_file *cfile; |
48ddbe19 | 4555 | |
9a1049da TH |
4556 | percpu_ref_exit(&css->refcnt); |
4557 | ||
6f60eade TH |
4558 | list_for_each_entry(cfile, &css->files, node) |
4559 | kernfs_put(cfile->kn); | |
4560 | ||
01e58659 | 4561 | if (ss) { |
9d755d33 | 4562 | /* css free path */ |
01e58659 VD |
4563 | int id = css->id; |
4564 | ||
9d755d33 TH |
4565 | if (css->parent) |
4566 | css_put(css->parent); | |
0ae78e0b | 4567 | |
01e58659 VD |
4568 | ss->css_free(css); |
4569 | cgroup_idr_remove(&ss->css_idr, id); | |
9d755d33 TH |
4570 | cgroup_put(cgrp); |
4571 | } else { | |
4572 | /* cgroup free path */ | |
4573 | atomic_dec(&cgrp->root->nr_cgrps); | |
4574 | cgroup_pidlist_destroy_all(cgrp); | |
971ff493 | 4575 | cancel_work_sync(&cgrp->release_agent_work); |
9d755d33 | 4576 | |
d51f39b0 | 4577 | if (cgroup_parent(cgrp)) { |
9d755d33 TH |
4578 | /* |
4579 | * We get a ref to the parent, and put the ref when | |
4580 | * this cgroup is being freed, so it's guaranteed | |
4581 | * that the parent won't be destroyed before its | |
4582 | * children. | |
4583 | */ | |
d51f39b0 | 4584 | cgroup_put(cgroup_parent(cgrp)); |
9d755d33 TH |
4585 | kernfs_put(cgrp->kn); |
4586 | kfree(cgrp); | |
4587 | } else { | |
4588 | /* | |
4589 | * This is root cgroup's refcnt reaching zero, | |
4590 | * which indicates that the root should be | |
4591 | * released. | |
4592 | */ | |
4593 | cgroup_destroy_root(cgrp->root); | |
4594 | } | |
4595 | } | |
48ddbe19 TH |
4596 | } |
4597 | ||
0c21ead1 | 4598 | static void css_free_rcu_fn(struct rcu_head *rcu_head) |
d3daf28d TH |
4599 | { |
4600 | struct cgroup_subsys_state *css = | |
0c21ead1 | 4601 | container_of(rcu_head, struct cgroup_subsys_state, rcu_head); |
d3daf28d | 4602 | |
35ef10da | 4603 | INIT_WORK(&css->destroy_work, css_free_work_fn); |
e5fca243 | 4604 | queue_work(cgroup_destroy_wq, &css->destroy_work); |
48ddbe19 TH |
4605 | } |
4606 | ||
25e15d83 | 4607 | static void css_release_work_fn(struct work_struct *work) |
d3daf28d TH |
4608 | { |
4609 | struct cgroup_subsys_state *css = | |
25e15d83 | 4610 | container_of(work, struct cgroup_subsys_state, destroy_work); |
15a4c835 | 4611 | struct cgroup_subsys *ss = css->ss; |
9d755d33 | 4612 | struct cgroup *cgrp = css->cgroup; |
15a4c835 | 4613 | |
1fed1b2e TH |
4614 | mutex_lock(&cgroup_mutex); |
4615 | ||
de3f0341 | 4616 | css->flags |= CSS_RELEASED; |
1fed1b2e TH |
4617 | list_del_rcu(&css->sibling); |
4618 | ||
9d755d33 TH |
4619 | if (ss) { |
4620 | /* css release path */ | |
01e58659 | 4621 | cgroup_idr_replace(&ss->css_idr, NULL, css->id); |
7d172cc8 TH |
4622 | if (ss->css_released) |
4623 | ss->css_released(css); | |
9d755d33 TH |
4624 | } else { |
4625 | /* cgroup release path */ | |
9d755d33 TH |
4626 | cgroup_idr_remove(&cgrp->root->cgroup_idr, cgrp->id); |
4627 | cgrp->id = -1; | |
a4189487 LZ |
4628 | |
4629 | /* | |
4630 | * There are two control paths which try to determine | |
4631 | * cgroup from dentry without going through kernfs - | |
4632 | * cgroupstats_build() and css_tryget_online_from_dir(). | |
4633 | * Those are supported by RCU protecting clearing of | |
4634 | * cgrp->kn->priv backpointer. | |
4635 | */ | |
4636 | RCU_INIT_POINTER(*(void __rcu __force **)&cgrp->kn->priv, NULL); | |
9d755d33 | 4637 | } |
d3daf28d | 4638 | |
1fed1b2e TH |
4639 | mutex_unlock(&cgroup_mutex); |
4640 | ||
0c21ead1 | 4641 | call_rcu(&css->rcu_head, css_free_rcu_fn); |
d3daf28d TH |
4642 | } |
4643 | ||
d3daf28d TH |
4644 | static void css_release(struct percpu_ref *ref) |
4645 | { | |
4646 | struct cgroup_subsys_state *css = | |
4647 | container_of(ref, struct cgroup_subsys_state, refcnt); | |
4648 | ||
25e15d83 TH |
4649 | INIT_WORK(&css->destroy_work, css_release_work_fn); |
4650 | queue_work(cgroup_destroy_wq, &css->destroy_work); | |
d3daf28d TH |
4651 | } |
4652 | ||
ddfcadab TH |
4653 | static void init_and_link_css(struct cgroup_subsys_state *css, |
4654 | struct cgroup_subsys *ss, struct cgroup *cgrp) | |
ddbcc7e8 | 4655 | { |
0cb51d71 TH |
4656 | lockdep_assert_held(&cgroup_mutex); |
4657 | ||
ddfcadab TH |
4658 | cgroup_get(cgrp); |
4659 | ||
d5c419b6 | 4660 | memset(css, 0, sizeof(*css)); |
bd89aabc | 4661 | css->cgroup = cgrp; |
72c97e54 | 4662 | css->ss = ss; |
d5c419b6 TH |
4663 | INIT_LIST_HEAD(&css->sibling); |
4664 | INIT_LIST_HEAD(&css->children); | |
6f60eade | 4665 | INIT_LIST_HEAD(&css->files); |
0cb51d71 | 4666 | css->serial_nr = css_serial_nr_next++; |
0ae78e0b | 4667 | |
d51f39b0 TH |
4668 | if (cgroup_parent(cgrp)) { |
4669 | css->parent = cgroup_css(cgroup_parent(cgrp), ss); | |
ddfcadab | 4670 | css_get(css->parent); |
ddfcadab | 4671 | } |
48ddbe19 | 4672 | |
ca8bdcaf | 4673 | BUG_ON(cgroup_css(cgrp, ss)); |
ddbcc7e8 PM |
4674 | } |
4675 | ||
2a4ac633 | 4676 | /* invoke ->css_online() on a new CSS and mark it online if successful */ |
623f926b | 4677 | static int online_css(struct cgroup_subsys_state *css) |
a31f2d3f | 4678 | { |
623f926b | 4679 | struct cgroup_subsys *ss = css->ss; |
b1929db4 TH |
4680 | int ret = 0; |
4681 | ||
a31f2d3f TH |
4682 | lockdep_assert_held(&cgroup_mutex); |
4683 | ||
92fb9748 | 4684 | if (ss->css_online) |
eb95419b | 4685 | ret = ss->css_online(css); |
ae7f164a | 4686 | if (!ret) { |
eb95419b | 4687 | css->flags |= CSS_ONLINE; |
aec25020 | 4688 | rcu_assign_pointer(css->cgroup->subsys[ss->id], css); |
ae7f164a | 4689 | } |
b1929db4 | 4690 | return ret; |
a31f2d3f TH |
4691 | } |
4692 | ||
2a4ac633 | 4693 | /* if the CSS is online, invoke ->css_offline() on it and mark it offline */ |
623f926b | 4694 | static void offline_css(struct cgroup_subsys_state *css) |
a31f2d3f | 4695 | { |
623f926b | 4696 | struct cgroup_subsys *ss = css->ss; |
a31f2d3f TH |
4697 | |
4698 | lockdep_assert_held(&cgroup_mutex); | |
4699 | ||
4700 | if (!(css->flags & CSS_ONLINE)) | |
4701 | return; | |
4702 | ||
d7eeac19 | 4703 | if (ss->css_offline) |
eb95419b | 4704 | ss->css_offline(css); |
a31f2d3f | 4705 | |
eb95419b | 4706 | css->flags &= ~CSS_ONLINE; |
e3297803 | 4707 | RCU_INIT_POINTER(css->cgroup->subsys[ss->id], NULL); |
f8f22e53 TH |
4708 | |
4709 | wake_up_all(&css->cgroup->offline_waitq); | |
a31f2d3f TH |
4710 | } |
4711 | ||
c81c925a TH |
4712 | /** |
4713 | * create_css - create a cgroup_subsys_state | |
4714 | * @cgrp: the cgroup new css will be associated with | |
4715 | * @ss: the subsys of new css | |
f63070d3 | 4716 | * @visible: whether to create control knobs for the new css or not |
c81c925a TH |
4717 | * |
4718 | * Create a new css associated with @cgrp - @ss pair. On success, the new | |
f63070d3 TH |
4719 | * css is online and installed in @cgrp with all interface files created if |
4720 | * @visible. Returns 0 on success, -errno on failure. | |
c81c925a | 4721 | */ |
f63070d3 TH |
4722 | static int create_css(struct cgroup *cgrp, struct cgroup_subsys *ss, |
4723 | bool visible) | |
c81c925a | 4724 | { |
d51f39b0 | 4725 | struct cgroup *parent = cgroup_parent(cgrp); |
1fed1b2e | 4726 | struct cgroup_subsys_state *parent_css = cgroup_css(parent, ss); |
c81c925a TH |
4727 | struct cgroup_subsys_state *css; |
4728 | int err; | |
4729 | ||
c81c925a TH |
4730 | lockdep_assert_held(&cgroup_mutex); |
4731 | ||
1fed1b2e | 4732 | css = ss->css_alloc(parent_css); |
c81c925a TH |
4733 | if (IS_ERR(css)) |
4734 | return PTR_ERR(css); | |
4735 | ||
ddfcadab | 4736 | init_and_link_css(css, ss, cgrp); |
a2bed820 | 4737 | |
2aad2a86 | 4738 | err = percpu_ref_init(&css->refcnt, css_release, 0, GFP_KERNEL); |
c81c925a | 4739 | if (err) |
3eb59ec6 | 4740 | goto err_free_css; |
c81c925a | 4741 | |
cf780b7d | 4742 | err = cgroup_idr_alloc(&ss->css_idr, NULL, 2, 0, GFP_KERNEL); |
15a4c835 TH |
4743 | if (err < 0) |
4744 | goto err_free_percpu_ref; | |
4745 | css->id = err; | |
c81c925a | 4746 | |
f63070d3 | 4747 | if (visible) { |
4df8dc90 | 4748 | err = css_populate_dir(css, NULL); |
f63070d3 TH |
4749 | if (err) |
4750 | goto err_free_id; | |
4751 | } | |
15a4c835 TH |
4752 | |
4753 | /* @css is ready to be brought online now, make it visible */ | |
1fed1b2e | 4754 | list_add_tail_rcu(&css->sibling, &parent_css->children); |
15a4c835 | 4755 | cgroup_idr_replace(&ss->css_idr, css, css->id); |
c81c925a TH |
4756 | |
4757 | err = online_css(css); | |
4758 | if (err) | |
1fed1b2e | 4759 | goto err_list_del; |
94419627 | 4760 | |
c81c925a | 4761 | if (ss->broken_hierarchy && !ss->warned_broken_hierarchy && |
d51f39b0 | 4762 | cgroup_parent(parent)) { |
ed3d261b | 4763 | 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 | 4764 | current->comm, current->pid, ss->name); |
c81c925a | 4765 | if (!strcmp(ss->name, "memory")) |
ed3d261b | 4766 | pr_warn("\"memory\" requires setting use_hierarchy to 1 on the root\n"); |
c81c925a TH |
4767 | ss->warned_broken_hierarchy = true; |
4768 | } | |
4769 | ||
4770 | return 0; | |
4771 | ||
1fed1b2e TH |
4772 | err_list_del: |
4773 | list_del_rcu(&css->sibling); | |
4df8dc90 | 4774 | css_clear_dir(css, NULL); |
15a4c835 TH |
4775 | err_free_id: |
4776 | cgroup_idr_remove(&ss->css_idr, css->id); | |
3eb59ec6 | 4777 | err_free_percpu_ref: |
9a1049da | 4778 | percpu_ref_exit(&css->refcnt); |
3eb59ec6 | 4779 | err_free_css: |
a2bed820 | 4780 | call_rcu(&css->rcu_head, css_free_rcu_fn); |
c81c925a TH |
4781 | return err; |
4782 | } | |
4783 | ||
b3bfd983 TH |
4784 | static int cgroup_mkdir(struct kernfs_node *parent_kn, const char *name, |
4785 | umode_t mode) | |
ddbcc7e8 | 4786 | { |
a9746d8d TH |
4787 | struct cgroup *parent, *cgrp; |
4788 | struct cgroup_root *root; | |
ddbcc7e8 | 4789 | struct cgroup_subsys *ss; |
2bd59d48 | 4790 | struct kernfs_node *kn; |
b3bfd983 | 4791 | int ssid, ret; |
ddbcc7e8 | 4792 | |
71b1fb5c AC |
4793 | /* Do not accept '\n' to prevent making /proc/<pid>/cgroup unparsable. |
4794 | */ | |
4795 | if (strchr(name, '\n')) | |
4796 | return -EINVAL; | |
4797 | ||
a9746d8d TH |
4798 | parent = cgroup_kn_lock_live(parent_kn); |
4799 | if (!parent) | |
4800 | return -ENODEV; | |
4801 | root = parent->root; | |
ddbcc7e8 | 4802 | |
0a950f65 | 4803 | /* allocate the cgroup and its ID, 0 is reserved for the root */ |
bd89aabc | 4804 | cgrp = kzalloc(sizeof(*cgrp), GFP_KERNEL); |
ba0f4d76 TH |
4805 | if (!cgrp) { |
4806 | ret = -ENOMEM; | |
4807 | goto out_unlock; | |
0ab02ca8 LZ |
4808 | } |
4809 | ||
2aad2a86 | 4810 | ret = percpu_ref_init(&cgrp->self.refcnt, css_release, 0, GFP_KERNEL); |
9d755d33 TH |
4811 | if (ret) |
4812 | goto out_free_cgrp; | |
4813 | ||
0ab02ca8 LZ |
4814 | /* |
4815 | * Temporarily set the pointer to NULL, so idr_find() won't return | |
4816 | * a half-baked cgroup. | |
4817 | */ | |
cf780b7d | 4818 | cgrp->id = cgroup_idr_alloc(&root->cgroup_idr, NULL, 2, 0, GFP_KERNEL); |
0ab02ca8 | 4819 | if (cgrp->id < 0) { |
ba0f4d76 | 4820 | ret = -ENOMEM; |
9d755d33 | 4821 | goto out_cancel_ref; |
976c06bc TH |
4822 | } |
4823 | ||
cc31edce | 4824 | init_cgroup_housekeeping(cgrp); |
ddbcc7e8 | 4825 | |
9d800df1 | 4826 | cgrp->self.parent = &parent->self; |
ba0f4d76 | 4827 | cgrp->root = root; |
ddbcc7e8 | 4828 | |
b6abdb0e LZ |
4829 | if (notify_on_release(parent)) |
4830 | set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); | |
4831 | ||
2260e7fc TH |
4832 | if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &parent->flags)) |
4833 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags); | |
97978e6d | 4834 | |
2bd59d48 | 4835 | /* create the directory */ |
e61734c5 | 4836 | kn = kernfs_create_dir(parent->kn, name, mode, cgrp); |
2bd59d48 | 4837 | if (IS_ERR(kn)) { |
ba0f4d76 TH |
4838 | ret = PTR_ERR(kn); |
4839 | goto out_free_id; | |
2bd59d48 TH |
4840 | } |
4841 | cgrp->kn = kn; | |
ddbcc7e8 | 4842 | |
4e139afc | 4843 | /* |
6f30558f TH |
4844 | * This extra ref will be put in cgroup_free_fn() and guarantees |
4845 | * that @cgrp->kn is always accessible. | |
4e139afc | 4846 | */ |
6f30558f | 4847 | kernfs_get(kn); |
ddbcc7e8 | 4848 | |
0cb51d71 | 4849 | cgrp->self.serial_nr = css_serial_nr_next++; |
53fa5261 | 4850 | |
4e139afc | 4851 | /* allocation complete, commit to creation */ |
d5c419b6 | 4852 | list_add_tail_rcu(&cgrp->self.sibling, &cgroup_parent(cgrp)->self.children); |
3c9c825b | 4853 | atomic_inc(&root->nr_cgrps); |
59f5296b | 4854 | cgroup_get(parent); |
415cf07a | 4855 | |
0d80255e TH |
4856 | /* |
4857 | * @cgrp is now fully operational. If something fails after this | |
4858 | * point, it'll be released via the normal destruction path. | |
4859 | */ | |
6fa4918d | 4860 | cgroup_idr_replace(&root->cgroup_idr, cgrp, cgrp->id); |
4e96ee8e | 4861 | |
ba0f4d76 TH |
4862 | ret = cgroup_kn_set_ugid(kn); |
4863 | if (ret) | |
4864 | goto out_destroy; | |
49957f8e | 4865 | |
4df8dc90 | 4866 | ret = css_populate_dir(&cgrp->self, NULL); |
ba0f4d76 TH |
4867 | if (ret) |
4868 | goto out_destroy; | |
628f7cd4 | 4869 | |
9d403e99 | 4870 | /* let's create and online css's */ |
b85d2040 | 4871 | for_each_subsys(ss, ssid) { |
f392e51c | 4872 | if (parent->child_subsys_mask & (1 << ssid)) { |
f63070d3 TH |
4873 | ret = create_css(cgrp, ss, |
4874 | parent->subtree_control & (1 << ssid)); | |
ba0f4d76 TH |
4875 | if (ret) |
4876 | goto out_destroy; | |
b85d2040 | 4877 | } |
a8638030 | 4878 | } |
ddbcc7e8 | 4879 | |
bd53d617 TH |
4880 | /* |
4881 | * On the default hierarchy, a child doesn't automatically inherit | |
667c2491 | 4882 | * subtree_control from the parent. Each is configured manually. |
bd53d617 | 4883 | */ |
667c2491 TH |
4884 | if (!cgroup_on_dfl(cgrp)) { |
4885 | cgrp->subtree_control = parent->subtree_control; | |
4886 | cgroup_refresh_child_subsys_mask(cgrp); | |
4887 | } | |
2bd59d48 | 4888 | |
2bd59d48 | 4889 | kernfs_activate(kn); |
ddbcc7e8 | 4890 | |
ba0f4d76 TH |
4891 | ret = 0; |
4892 | goto out_unlock; | |
ddbcc7e8 | 4893 | |
ba0f4d76 | 4894 | out_free_id: |
6fa4918d | 4895 | cgroup_idr_remove(&root->cgroup_idr, cgrp->id); |
9d755d33 | 4896 | out_cancel_ref: |
9a1049da | 4897 | percpu_ref_exit(&cgrp->self.refcnt); |
ba0f4d76 | 4898 | out_free_cgrp: |
bd89aabc | 4899 | kfree(cgrp); |
ba0f4d76 | 4900 | out_unlock: |
a9746d8d | 4901 | cgroup_kn_unlock(parent_kn); |
ba0f4d76 | 4902 | return ret; |
4b8b47eb | 4903 | |
ba0f4d76 | 4904 | out_destroy: |
4b8b47eb | 4905 | cgroup_destroy_locked(cgrp); |
ba0f4d76 | 4906 | goto out_unlock; |
ddbcc7e8 PM |
4907 | } |
4908 | ||
223dbc38 TH |
4909 | /* |
4910 | * This is called when the refcnt of a css is confirmed to be killed. | |
249f3468 TH |
4911 | * css_tryget_online() is now guaranteed to fail. Tell the subsystem to |
4912 | * initate destruction and put the css ref from kill_css(). | |
223dbc38 TH |
4913 | */ |
4914 | static void css_killed_work_fn(struct work_struct *work) | |
d3daf28d | 4915 | { |
223dbc38 TH |
4916 | struct cgroup_subsys_state *css = |
4917 | container_of(work, struct cgroup_subsys_state, destroy_work); | |
d3daf28d | 4918 | |
f20104de | 4919 | mutex_lock(&cgroup_mutex); |
09a503ea | 4920 | offline_css(css); |
f20104de | 4921 | mutex_unlock(&cgroup_mutex); |
09a503ea | 4922 | |
09a503ea | 4923 | css_put(css); |
d3daf28d TH |
4924 | } |
4925 | ||
223dbc38 TH |
4926 | /* css kill confirmation processing requires process context, bounce */ |
4927 | static void css_killed_ref_fn(struct percpu_ref *ref) | |
d3daf28d TH |
4928 | { |
4929 | struct cgroup_subsys_state *css = | |
4930 | container_of(ref, struct cgroup_subsys_state, refcnt); | |
4931 | ||
223dbc38 | 4932 | INIT_WORK(&css->destroy_work, css_killed_work_fn); |
e5fca243 | 4933 | queue_work(cgroup_destroy_wq, &css->destroy_work); |
d3daf28d TH |
4934 | } |
4935 | ||
f392e51c TH |
4936 | /** |
4937 | * kill_css - destroy a css | |
4938 | * @css: css to destroy | |
4939 | * | |
4940 | * This function initiates destruction of @css by removing cgroup interface | |
4941 | * files and putting its base reference. ->css_offline() will be invoked | |
ec903c0c TH |
4942 | * asynchronously once css_tryget_online() is guaranteed to fail and when |
4943 | * the reference count reaches zero, @css will be released. | |
f392e51c TH |
4944 | */ |
4945 | static void kill_css(struct cgroup_subsys_state *css) | |
edae0c33 | 4946 | { |
01f6474c | 4947 | lockdep_assert_held(&cgroup_mutex); |
94419627 | 4948 | |
2bd59d48 TH |
4949 | /* |
4950 | * This must happen before css is disassociated with its cgroup. | |
4951 | * See seq_css() for details. | |
4952 | */ | |
4df8dc90 | 4953 | css_clear_dir(css, NULL); |
3c14f8b4 | 4954 | |
edae0c33 TH |
4955 | /* |
4956 | * Killing would put the base ref, but we need to keep it alive | |
4957 | * until after ->css_offline(). | |
4958 | */ | |
4959 | css_get(css); | |
4960 | ||
4961 | /* | |
4962 | * cgroup core guarantees that, by the time ->css_offline() is | |
4963 | * invoked, no new css reference will be given out via | |
ec903c0c | 4964 | * css_tryget_online(). We can't simply call percpu_ref_kill() and |
edae0c33 TH |
4965 | * proceed to offlining css's because percpu_ref_kill() doesn't |
4966 | * guarantee that the ref is seen as killed on all CPUs on return. | |
4967 | * | |
4968 | * Use percpu_ref_kill_and_confirm() to get notifications as each | |
4969 | * css is confirmed to be seen as killed on all CPUs. | |
4970 | */ | |
4971 | percpu_ref_kill_and_confirm(&css->refcnt, css_killed_ref_fn); | |
d3daf28d TH |
4972 | } |
4973 | ||
4974 | /** | |
4975 | * cgroup_destroy_locked - the first stage of cgroup destruction | |
4976 | * @cgrp: cgroup to be destroyed | |
4977 | * | |
4978 | * css's make use of percpu refcnts whose killing latency shouldn't be | |
4979 | * exposed to userland and are RCU protected. Also, cgroup core needs to | |
ec903c0c TH |
4980 | * guarantee that css_tryget_online() won't succeed by the time |
4981 | * ->css_offline() is invoked. To satisfy all the requirements, | |
4982 | * destruction is implemented in the following two steps. | |
d3daf28d TH |
4983 | * |
4984 | * s1. Verify @cgrp can be destroyed and mark it dying. Remove all | |
4985 | * userland visible parts and start killing the percpu refcnts of | |
4986 | * css's. Set up so that the next stage will be kicked off once all | |
4987 | * the percpu refcnts are confirmed to be killed. | |
4988 | * | |
4989 | * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the | |
4990 | * rest of destruction. Once all cgroup references are gone, the | |
4991 | * cgroup is RCU-freed. | |
4992 | * | |
4993 | * This function implements s1. After this step, @cgrp is gone as far as | |
4994 | * the userland is concerned and a new cgroup with the same name may be | |
4995 | * created. As cgroup doesn't care about the names internally, this | |
4996 | * doesn't cause any problem. | |
4997 | */ | |
42809dd4 TH |
4998 | static int cgroup_destroy_locked(struct cgroup *cgrp) |
4999 | __releases(&cgroup_mutex) __acquires(&cgroup_mutex) | |
ddbcc7e8 | 5000 | { |
2bd59d48 | 5001 | struct cgroup_subsys_state *css; |
ddd69148 | 5002 | bool empty; |
1c6727af | 5003 | int ssid; |
ddbcc7e8 | 5004 | |
42809dd4 TH |
5005 | lockdep_assert_held(&cgroup_mutex); |
5006 | ||
ddd69148 | 5007 | /* |
96d365e0 | 5008 | * css_set_rwsem synchronizes access to ->cset_links and prevents |
89c5509b | 5009 | * @cgrp from being removed while put_css_set() is in progress. |
ddd69148 | 5010 | */ |
96d365e0 | 5011 | down_read(&css_set_rwsem); |
bb78a92f | 5012 | empty = list_empty(&cgrp->cset_links); |
96d365e0 | 5013 | up_read(&css_set_rwsem); |
ddd69148 | 5014 | if (!empty) |
ddbcc7e8 | 5015 | return -EBUSY; |
a043e3b2 | 5016 | |
bb78a92f | 5017 | /* |
d5c419b6 TH |
5018 | * Make sure there's no live children. We can't test emptiness of |
5019 | * ->self.children as dead children linger on it while being | |
5020 | * drained; otherwise, "rmdir parent/child parent" may fail. | |
bb78a92f | 5021 | */ |
f3d46500 | 5022 | if (css_has_online_children(&cgrp->self)) |
bb78a92f HD |
5023 | return -EBUSY; |
5024 | ||
455050d2 TH |
5025 | /* |
5026 | * Mark @cgrp dead. This prevents further task migration and child | |
de3f0341 | 5027 | * creation by disabling cgroup_lock_live_group(). |
455050d2 | 5028 | */ |
184faf32 | 5029 | cgrp->self.flags &= ~CSS_ONLINE; |
ddbcc7e8 | 5030 | |
249f3468 | 5031 | /* initiate massacre of all css's */ |
1c6727af TH |
5032 | for_each_css(css, ssid, cgrp) |
5033 | kill_css(css); | |
455050d2 | 5034 | |
455050d2 | 5035 | /* |
01f6474c TH |
5036 | * Remove @cgrp directory along with the base files. @cgrp has an |
5037 | * extra ref on its kn. | |
f20104de | 5038 | */ |
01f6474c | 5039 | kernfs_remove(cgrp->kn); |
f20104de | 5040 | |
d51f39b0 | 5041 | check_for_release(cgroup_parent(cgrp)); |
2bd59d48 | 5042 | |
249f3468 | 5043 | /* put the base reference */ |
9d755d33 | 5044 | percpu_ref_kill(&cgrp->self.refcnt); |
455050d2 | 5045 | |
ea15f8cc TH |
5046 | return 0; |
5047 | }; | |
5048 | ||
2bd59d48 | 5049 | static int cgroup_rmdir(struct kernfs_node *kn) |
42809dd4 | 5050 | { |
a9746d8d | 5051 | struct cgroup *cgrp; |
2bd59d48 | 5052 | int ret = 0; |
42809dd4 | 5053 | |
a9746d8d TH |
5054 | cgrp = cgroup_kn_lock_live(kn); |
5055 | if (!cgrp) | |
5056 | return 0; | |
42809dd4 | 5057 | |
a9746d8d | 5058 | ret = cgroup_destroy_locked(cgrp); |
2bb566cb | 5059 | |
a9746d8d | 5060 | cgroup_kn_unlock(kn); |
42809dd4 | 5061 | return ret; |
8e3f6541 TH |
5062 | } |
5063 | ||
2bd59d48 TH |
5064 | static struct kernfs_syscall_ops cgroup_kf_syscall_ops = { |
5065 | .remount_fs = cgroup_remount, | |
5066 | .show_options = cgroup_show_options, | |
5067 | .mkdir = cgroup_mkdir, | |
5068 | .rmdir = cgroup_rmdir, | |
5069 | .rename = cgroup_rename, | |
5070 | }; | |
5071 | ||
15a4c835 | 5072 | static void __init cgroup_init_subsys(struct cgroup_subsys *ss, bool early) |
ddbcc7e8 | 5073 | { |
ddbcc7e8 | 5074 | struct cgroup_subsys_state *css; |
cfe36bde DC |
5075 | |
5076 | printk(KERN_INFO "Initializing cgroup subsys %s\n", ss->name); | |
ddbcc7e8 | 5077 | |
648bb56d TH |
5078 | mutex_lock(&cgroup_mutex); |
5079 | ||
15a4c835 | 5080 | idr_init(&ss->css_idr); |
0adb0704 | 5081 | INIT_LIST_HEAD(&ss->cfts); |
8e3f6541 | 5082 | |
3dd06ffa TH |
5083 | /* Create the root cgroup state for this subsystem */ |
5084 | ss->root = &cgrp_dfl_root; | |
5085 | css = ss->css_alloc(cgroup_css(&cgrp_dfl_root.cgrp, ss)); | |
ddbcc7e8 PM |
5086 | /* We don't handle early failures gracefully */ |
5087 | BUG_ON(IS_ERR(css)); | |
ddfcadab | 5088 | init_and_link_css(css, ss, &cgrp_dfl_root.cgrp); |
3b514d24 TH |
5089 | |
5090 | /* | |
5091 | * Root csses are never destroyed and we can't initialize | |
5092 | * percpu_ref during early init. Disable refcnting. | |
5093 | */ | |
5094 | css->flags |= CSS_NO_REF; | |
5095 | ||
15a4c835 | 5096 | if (early) { |
9395a450 | 5097 | /* allocation can't be done safely during early init */ |
15a4c835 TH |
5098 | css->id = 1; |
5099 | } else { | |
5100 | css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2, GFP_KERNEL); | |
5101 | BUG_ON(css->id < 0); | |
5102 | } | |
ddbcc7e8 | 5103 | |
e8d55fde | 5104 | /* Update the init_css_set to contain a subsys |
817929ec | 5105 | * pointer to this state - since the subsystem is |
e8d55fde | 5106 | * newly registered, all tasks and hence the |
3dd06ffa | 5107 | * init_css_set is in the subsystem's root cgroup. */ |
aec25020 | 5108 | init_css_set.subsys[ss->id] = css; |
ddbcc7e8 | 5109 | |
cb4a3167 AS |
5110 | have_fork_callback |= (bool)ss->fork << ss->id; |
5111 | have_exit_callback |= (bool)ss->exit << ss->id; | |
7e47682e | 5112 | have_canfork_callback |= (bool)ss->can_fork << ss->id; |
ddbcc7e8 | 5113 | |
e8d55fde LZ |
5114 | /* At system boot, before all subsystems have been |
5115 | * registered, no tasks have been forked, so we don't | |
5116 | * need to invoke fork callbacks here. */ | |
5117 | BUG_ON(!list_empty(&init_task.tasks)); | |
5118 | ||
ae7f164a | 5119 | BUG_ON(online_css(css)); |
a8638030 | 5120 | |
cf5d5941 BB |
5121 | mutex_unlock(&cgroup_mutex); |
5122 | } | |
cf5d5941 | 5123 | |
ddbcc7e8 | 5124 | /** |
a043e3b2 LZ |
5125 | * cgroup_init_early - cgroup initialization at system boot |
5126 | * | |
5127 | * Initialize cgroups at system boot, and initialize any | |
5128 | * subsystems that request early init. | |
ddbcc7e8 PM |
5129 | */ |
5130 | int __init cgroup_init_early(void) | |
5131 | { | |
7b9a6ba5 | 5132 | static struct cgroup_sb_opts __initdata opts; |
30159ec7 | 5133 | struct cgroup_subsys *ss; |
ddbcc7e8 | 5134 | int i; |
30159ec7 | 5135 | |
3dd06ffa | 5136 | init_cgroup_root(&cgrp_dfl_root, &opts); |
3b514d24 TH |
5137 | cgrp_dfl_root.cgrp.self.flags |= CSS_NO_REF; |
5138 | ||
a4ea1cc9 | 5139 | RCU_INIT_POINTER(init_task.cgroups, &init_css_set); |
817929ec | 5140 | |
3ed80a62 | 5141 | for_each_subsys(ss, i) { |
aec25020 | 5142 | WARN(!ss->css_alloc || !ss->css_free || ss->name || ss->id, |
073219e9 TH |
5143 | "invalid cgroup_subsys %d:%s css_alloc=%p css_free=%p name:id=%d:%s\n", |
5144 | i, cgroup_subsys_name[i], ss->css_alloc, ss->css_free, | |
aec25020 | 5145 | ss->id, ss->name); |
073219e9 TH |
5146 | WARN(strlen(cgroup_subsys_name[i]) > MAX_CGROUP_TYPE_NAMELEN, |
5147 | "cgroup_subsys_name %s too long\n", cgroup_subsys_name[i]); | |
5148 | ||
aec25020 | 5149 | ss->id = i; |
073219e9 | 5150 | ss->name = cgroup_subsys_name[i]; |
3e1d2eed TH |
5151 | if (!ss->legacy_name) |
5152 | ss->legacy_name = cgroup_subsys_name[i]; | |
ddbcc7e8 PM |
5153 | |
5154 | if (ss->early_init) | |
15a4c835 | 5155 | cgroup_init_subsys(ss, true); |
ddbcc7e8 PM |
5156 | } |
5157 | return 0; | |
5158 | } | |
5159 | ||
a3e72739 TH |
5160 | static unsigned long cgroup_disable_mask __initdata; |
5161 | ||
ddbcc7e8 | 5162 | /** |
a043e3b2 LZ |
5163 | * cgroup_init - cgroup initialization |
5164 | * | |
5165 | * Register cgroup filesystem and /proc file, and initialize | |
5166 | * any subsystems that didn't request early init. | |
ddbcc7e8 PM |
5167 | */ |
5168 | int __init cgroup_init(void) | |
5169 | { | |
30159ec7 | 5170 | struct cgroup_subsys *ss; |
0ac801fe | 5171 | unsigned long key; |
172a2c06 | 5172 | int ssid, err; |
ddbcc7e8 | 5173 | |
1ed13287 | 5174 | BUG_ON(percpu_init_rwsem(&cgroup_threadgroup_rwsem)); |
a14c6874 TH |
5175 | BUG_ON(cgroup_init_cftypes(NULL, cgroup_dfl_base_files)); |
5176 | BUG_ON(cgroup_init_cftypes(NULL, cgroup_legacy_base_files)); | |
ddbcc7e8 | 5177 | |
54e7b4eb | 5178 | mutex_lock(&cgroup_mutex); |
54e7b4eb | 5179 | |
82fe9b0d TH |
5180 | /* Add init_css_set to the hash table */ |
5181 | key = css_set_hash(init_css_set.subsys); | |
5182 | hash_add(css_set_table, &init_css_set.hlist, key); | |
5183 | ||
3dd06ffa | 5184 | BUG_ON(cgroup_setup_root(&cgrp_dfl_root, 0)); |
4e96ee8e | 5185 | |
54e7b4eb TH |
5186 | mutex_unlock(&cgroup_mutex); |
5187 | ||
172a2c06 | 5188 | for_each_subsys(ss, ssid) { |
15a4c835 TH |
5189 | if (ss->early_init) { |
5190 | struct cgroup_subsys_state *css = | |
5191 | init_css_set.subsys[ss->id]; | |
5192 | ||
5193 | css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2, | |
5194 | GFP_KERNEL); | |
5195 | BUG_ON(css->id < 0); | |
5196 | } else { | |
5197 | cgroup_init_subsys(ss, false); | |
5198 | } | |
172a2c06 | 5199 | |
2d8f243a TH |
5200 | list_add_tail(&init_css_set.e_cset_node[ssid], |
5201 | &cgrp_dfl_root.cgrp.e_csets[ssid]); | |
172a2c06 TH |
5202 | |
5203 | /* | |
c731ae1d LZ |
5204 | * Setting dfl_root subsys_mask needs to consider the |
5205 | * disabled flag and cftype registration needs kmalloc, | |
5206 | * both of which aren't available during early_init. | |
172a2c06 | 5207 | */ |
a3e72739 TH |
5208 | if (cgroup_disable_mask & (1 << ssid)) { |
5209 | static_branch_disable(cgroup_subsys_enabled_key[ssid]); | |
5210 | printk(KERN_INFO "Disabling %s control group subsystem\n", | |
5211 | ss->name); | |
a8ddc821 | 5212 | continue; |
a3e72739 | 5213 | } |
a8ddc821 TH |
5214 | |
5215 | cgrp_dfl_root.subsys_mask |= 1 << ss->id; | |
5216 | ||
5217 | if (cgroup_legacy_files_on_dfl && !ss->dfl_cftypes) | |
5218 | ss->dfl_cftypes = ss->legacy_cftypes; | |
5219 | ||
5de4fa13 TH |
5220 | if (!ss->dfl_cftypes) |
5221 | cgrp_dfl_root_inhibit_ss_mask |= 1 << ss->id; | |
5222 | ||
a8ddc821 TH |
5223 | if (ss->dfl_cftypes == ss->legacy_cftypes) { |
5224 | WARN_ON(cgroup_add_cftypes(ss, ss->dfl_cftypes)); | |
5225 | } else { | |
5226 | WARN_ON(cgroup_add_dfl_cftypes(ss, ss->dfl_cftypes)); | |
5227 | WARN_ON(cgroup_add_legacy_cftypes(ss, ss->legacy_cftypes)); | |
c731ae1d | 5228 | } |
295458e6 VD |
5229 | |
5230 | if (ss->bind) | |
5231 | ss->bind(init_css_set.subsys[ssid]); | |
676db4af GKH |
5232 | } |
5233 | ||
f9bb4882 EB |
5234 | err = sysfs_create_mount_point(fs_kobj, "cgroup"); |
5235 | if (err) | |
5236 | return err; | |
676db4af | 5237 | |
ddbcc7e8 | 5238 | err = register_filesystem(&cgroup_fs_type); |
676db4af | 5239 | if (err < 0) { |
f9bb4882 | 5240 | sysfs_remove_mount_point(fs_kobj, "cgroup"); |
2bd59d48 | 5241 | return err; |
676db4af | 5242 | } |
ddbcc7e8 | 5243 | |
46ae220b | 5244 | proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations); |
2bd59d48 | 5245 | return 0; |
ddbcc7e8 | 5246 | } |
b4f48b63 | 5247 | |
e5fca243 TH |
5248 | static int __init cgroup_wq_init(void) |
5249 | { | |
5250 | /* | |
5251 | * There isn't much point in executing destruction path in | |
5252 | * parallel. Good chunk is serialized with cgroup_mutex anyway. | |
1a11533f | 5253 | * Use 1 for @max_active. |
e5fca243 TH |
5254 | * |
5255 | * We would prefer to do this in cgroup_init() above, but that | |
5256 | * is called before init_workqueues(): so leave this until after. | |
5257 | */ | |
1a11533f | 5258 | cgroup_destroy_wq = alloc_workqueue("cgroup_destroy", 0, 1); |
e5fca243 | 5259 | BUG_ON(!cgroup_destroy_wq); |
b1a21367 TH |
5260 | |
5261 | /* | |
5262 | * Used to destroy pidlists and separate to serve as flush domain. | |
5263 | * Cap @max_active to 1 too. | |
5264 | */ | |
5265 | cgroup_pidlist_destroy_wq = alloc_workqueue("cgroup_pidlist_destroy", | |
5266 | 0, 1); | |
5267 | BUG_ON(!cgroup_pidlist_destroy_wq); | |
5268 | ||
e5fca243 TH |
5269 | return 0; |
5270 | } | |
5271 | core_initcall(cgroup_wq_init); | |
5272 | ||
a424316c PM |
5273 | /* |
5274 | * proc_cgroup_show() | |
5275 | * - Print task's cgroup paths into seq_file, one line for each hierarchy | |
5276 | * - Used for /proc/<pid>/cgroup. | |
a424316c | 5277 | */ |
006f4ac4 ZL |
5278 | int proc_cgroup_show(struct seq_file *m, struct pid_namespace *ns, |
5279 | struct pid *pid, struct task_struct *tsk) | |
a424316c | 5280 | { |
e61734c5 | 5281 | char *buf, *path; |
a424316c | 5282 | int retval; |
3dd06ffa | 5283 | struct cgroup_root *root; |
a424316c PM |
5284 | |
5285 | retval = -ENOMEM; | |
e61734c5 | 5286 | buf = kmalloc(PATH_MAX, GFP_KERNEL); |
a424316c PM |
5287 | if (!buf) |
5288 | goto out; | |
5289 | ||
a424316c | 5290 | mutex_lock(&cgroup_mutex); |
96d365e0 | 5291 | down_read(&css_set_rwsem); |
a424316c | 5292 | |
985ed670 | 5293 | for_each_root(root) { |
a424316c | 5294 | struct cgroup_subsys *ss; |
bd89aabc | 5295 | struct cgroup *cgrp; |
b85d2040 | 5296 | int ssid, count = 0; |
a424316c | 5297 | |
a2dd4247 | 5298 | if (root == &cgrp_dfl_root && !cgrp_dfl_root_visible) |
985ed670 TH |
5299 | continue; |
5300 | ||
2c6ab6d2 | 5301 | seq_printf(m, "%d:", root->hierarchy_id); |
d98817d4 TH |
5302 | if (root != &cgrp_dfl_root) |
5303 | for_each_subsys(ss, ssid) | |
5304 | if (root->subsys_mask & (1 << ssid)) | |
5305 | seq_printf(m, "%s%s", count++ ? "," : "", | |
3e1d2eed | 5306 | ss->legacy_name); |
c6d57f33 PM |
5307 | if (strlen(root->name)) |
5308 | seq_printf(m, "%sname=%s", count ? "," : "", | |
5309 | root->name); | |
a424316c | 5310 | seq_putc(m, ':'); |
7717f7ba | 5311 | cgrp = task_cgroup_from_root(tsk, root); |
e61734c5 TH |
5312 | path = cgroup_path(cgrp, buf, PATH_MAX); |
5313 | if (!path) { | |
5314 | retval = -ENAMETOOLONG; | |
a424316c | 5315 | goto out_unlock; |
e61734c5 TH |
5316 | } |
5317 | seq_puts(m, path); | |
a424316c PM |
5318 | seq_putc(m, '\n'); |
5319 | } | |
5320 | ||
006f4ac4 | 5321 | retval = 0; |
a424316c | 5322 | out_unlock: |
96d365e0 | 5323 | up_read(&css_set_rwsem); |
a424316c | 5324 | mutex_unlock(&cgroup_mutex); |
a424316c PM |
5325 | kfree(buf); |
5326 | out: | |
5327 | return retval; | |
5328 | } | |
5329 | ||
a424316c PM |
5330 | /* Display information about each subsystem and each hierarchy */ |
5331 | static int proc_cgroupstats_show(struct seq_file *m, void *v) | |
5332 | { | |
30159ec7 | 5333 | struct cgroup_subsys *ss; |
a424316c | 5334 | int i; |
a424316c | 5335 | |
8bab8dde | 5336 | seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n"); |
aae8aab4 BB |
5337 | /* |
5338 | * ideally we don't want subsystems moving around while we do this. | |
5339 | * cgroup_mutex is also necessary to guarantee an atomic snapshot of | |
5340 | * subsys/hierarchy state. | |
5341 | */ | |
a424316c | 5342 | mutex_lock(&cgroup_mutex); |
30159ec7 TH |
5343 | |
5344 | for_each_subsys(ss, i) | |
2c6ab6d2 | 5345 | seq_printf(m, "%s\t%d\t%d\t%d\n", |
3e1d2eed | 5346 | ss->legacy_name, ss->root->hierarchy_id, |
fc5ed1e9 TH |
5347 | atomic_read(&ss->root->nr_cgrps), |
5348 | cgroup_ssid_enabled(i)); | |
30159ec7 | 5349 | |
a424316c PM |
5350 | mutex_unlock(&cgroup_mutex); |
5351 | return 0; | |
5352 | } | |
5353 | ||
5354 | static int cgroupstats_open(struct inode *inode, struct file *file) | |
5355 | { | |
9dce07f1 | 5356 | return single_open(file, proc_cgroupstats_show, NULL); |
a424316c PM |
5357 | } |
5358 | ||
828c0950 | 5359 | static const struct file_operations proc_cgroupstats_operations = { |
a424316c PM |
5360 | .open = cgroupstats_open, |
5361 | .read = seq_read, | |
5362 | .llseek = seq_lseek, | |
5363 | .release = single_release, | |
5364 | }; | |
5365 | ||
7e47682e AS |
5366 | static void **subsys_canfork_priv_p(void *ss_priv[CGROUP_CANFORK_COUNT], int i) |
5367 | { | |
5368 | if (CGROUP_CANFORK_START <= i && i < CGROUP_CANFORK_END) | |
5369 | return &ss_priv[i - CGROUP_CANFORK_START]; | |
5370 | return NULL; | |
5371 | } | |
5372 | ||
5373 | static void *subsys_canfork_priv(void *ss_priv[CGROUP_CANFORK_COUNT], int i) | |
5374 | { | |
5375 | void **private = subsys_canfork_priv_p(ss_priv, i); | |
5376 | return private ? *private : NULL; | |
5377 | } | |
5378 | ||
b4f48b63 | 5379 | /** |
eaf797ab | 5380 | * cgroup_fork - initialize cgroup related fields during copy_process() |
a043e3b2 | 5381 | * @child: pointer to task_struct of forking parent process. |
b4f48b63 | 5382 | * |
eaf797ab TH |
5383 | * A task is associated with the init_css_set until cgroup_post_fork() |
5384 | * attaches it to the parent's css_set. Empty cg_list indicates that | |
5385 | * @child isn't holding reference to its css_set. | |
b4f48b63 PM |
5386 | */ |
5387 | void cgroup_fork(struct task_struct *child) | |
5388 | { | |
eaf797ab | 5389 | RCU_INIT_POINTER(child->cgroups, &init_css_set); |
817929ec | 5390 | INIT_LIST_HEAD(&child->cg_list); |
b4f48b63 PM |
5391 | } |
5392 | ||
7e47682e AS |
5393 | /** |
5394 | * cgroup_can_fork - called on a new task before the process is exposed | |
5395 | * @child: the task in question. | |
5396 | * | |
5397 | * This calls the subsystem can_fork() callbacks. If the can_fork() callback | |
5398 | * returns an error, the fork aborts with that error code. This allows for | |
5399 | * a cgroup subsystem to conditionally allow or deny new forks. | |
5400 | */ | |
5401 | int cgroup_can_fork(struct task_struct *child, | |
5402 | void *ss_priv[CGROUP_CANFORK_COUNT]) | |
5403 | { | |
5404 | struct cgroup_subsys *ss; | |
5405 | int i, j, ret; | |
5406 | ||
5407 | for_each_subsys_which(ss, i, &have_canfork_callback) { | |
5408 | ret = ss->can_fork(child, subsys_canfork_priv_p(ss_priv, i)); | |
5409 | if (ret) | |
5410 | goto out_revert; | |
5411 | } | |
5412 | ||
5413 | return 0; | |
5414 | ||
5415 | out_revert: | |
5416 | for_each_subsys(ss, j) { | |
5417 | if (j >= i) | |
5418 | break; | |
5419 | if (ss->cancel_fork) | |
5420 | ss->cancel_fork(child, subsys_canfork_priv(ss_priv, j)); | |
5421 | } | |
5422 | ||
5423 | return ret; | |
5424 | } | |
5425 | ||
5426 | /** | |
5427 | * cgroup_cancel_fork - called if a fork failed after cgroup_can_fork() | |
5428 | * @child: the task in question | |
5429 | * | |
5430 | * This calls the cancel_fork() callbacks if a fork failed *after* | |
5431 | * cgroup_can_fork() succeded. | |
5432 | */ | |
5433 | void cgroup_cancel_fork(struct task_struct *child, | |
5434 | void *ss_priv[CGROUP_CANFORK_COUNT]) | |
5435 | { | |
5436 | struct cgroup_subsys *ss; | |
5437 | int i; | |
5438 | ||
5439 | for_each_subsys(ss, i) | |
5440 | if (ss->cancel_fork) | |
5441 | ss->cancel_fork(child, subsys_canfork_priv(ss_priv, i)); | |
5442 | } | |
5443 | ||
817929ec | 5444 | /** |
a043e3b2 LZ |
5445 | * cgroup_post_fork - called on a new task after adding it to the task list |
5446 | * @child: the task in question | |
5447 | * | |
5edee61e TH |
5448 | * Adds the task to the list running through its css_set if necessary and |
5449 | * call the subsystem fork() callbacks. Has to be after the task is | |
5450 | * visible on the task list in case we race with the first call to | |
0942eeee | 5451 | * cgroup_task_iter_start() - to guarantee that the new task ends up on its |
5edee61e | 5452 | * list. |
a043e3b2 | 5453 | */ |
7e47682e AS |
5454 | void cgroup_post_fork(struct task_struct *child, |
5455 | void *old_ss_priv[CGROUP_CANFORK_COUNT]) | |
817929ec | 5456 | { |
30159ec7 | 5457 | struct cgroup_subsys *ss; |
5edee61e TH |
5458 | int i; |
5459 | ||
3ce3230a | 5460 | /* |
251f8c03 | 5461 | * This may race against cgroup_enable_task_cg_lists(). As that |
eaf797ab TH |
5462 | * function sets use_task_css_set_links before grabbing |
5463 | * tasklist_lock and we just went through tasklist_lock to add | |
5464 | * @child, it's guaranteed that either we see the set | |
5465 | * use_task_css_set_links or cgroup_enable_task_cg_lists() sees | |
5466 | * @child during its iteration. | |
5467 | * | |
5468 | * If we won the race, @child is associated with %current's | |
5469 | * css_set. Grabbing css_set_rwsem guarantees both that the | |
5470 | * association is stable, and, on completion of the parent's | |
5471 | * migration, @child is visible in the source of migration or | |
5472 | * already in the destination cgroup. This guarantee is necessary | |
5473 | * when implementing operations which need to migrate all tasks of | |
5474 | * a cgroup to another. | |
5475 | * | |
251f8c03 | 5476 | * Note that if we lose to cgroup_enable_task_cg_lists(), @child |
eaf797ab TH |
5477 | * will remain in init_css_set. This is safe because all tasks are |
5478 | * in the init_css_set before cg_links is enabled and there's no | |
5479 | * operation which transfers all tasks out of init_css_set. | |
3ce3230a | 5480 | */ |
817929ec | 5481 | if (use_task_css_set_links) { |
eaf797ab TH |
5482 | struct css_set *cset; |
5483 | ||
96d365e0 | 5484 | down_write(&css_set_rwsem); |
0e1d768f | 5485 | cset = task_css_set(current); |
eaf797ab TH |
5486 | if (list_empty(&child->cg_list)) { |
5487 | rcu_assign_pointer(child->cgroups, cset); | |
5488 | list_add(&child->cg_list, &cset->tasks); | |
5489 | get_css_set(cset); | |
5490 | } | |
96d365e0 | 5491 | up_write(&css_set_rwsem); |
817929ec | 5492 | } |
5edee61e TH |
5493 | |
5494 | /* | |
5495 | * Call ss->fork(). This must happen after @child is linked on | |
5496 | * css_set; otherwise, @child might change state between ->fork() | |
5497 | * and addition to css_set. | |
5498 | */ | |
cb4a3167 | 5499 | for_each_subsys_which(ss, i, &have_fork_callback) |
7e47682e | 5500 | ss->fork(child, subsys_canfork_priv(old_ss_priv, i)); |
817929ec | 5501 | } |
5edee61e | 5502 | |
b4f48b63 PM |
5503 | /** |
5504 | * cgroup_exit - detach cgroup from exiting task | |
5505 | * @tsk: pointer to task_struct of exiting process | |
5506 | * | |
5507 | * Description: Detach cgroup from @tsk and release it. | |
5508 | * | |
5509 | * Note that cgroups marked notify_on_release force every task in | |
5510 | * them to take the global cgroup_mutex mutex when exiting. | |
5511 | * This could impact scaling on very large systems. Be reluctant to | |
5512 | * use notify_on_release cgroups where very high task exit scaling | |
5513 | * is required on large systems. | |
5514 | * | |
0e1d768f TH |
5515 | * We set the exiting tasks cgroup to the root cgroup (top_cgroup). We |
5516 | * call cgroup_exit() while the task is still competent to handle | |
5517 | * notify_on_release(), then leave the task attached to the root cgroup in | |
5518 | * each hierarchy for the remainder of its exit. No need to bother with | |
5519 | * init_css_set refcnting. init_css_set never goes away and we can't race | |
e8604cb4 | 5520 | * with migration path - PF_EXITING is visible to migration path. |
b4f48b63 | 5521 | */ |
1ec41830 | 5522 | void cgroup_exit(struct task_struct *tsk) |
b4f48b63 | 5523 | { |
30159ec7 | 5524 | struct cgroup_subsys *ss; |
5abb8855 | 5525 | struct css_set *cset; |
eaf797ab | 5526 | bool put_cset = false; |
d41d5a01 | 5527 | int i; |
817929ec PM |
5528 | |
5529 | /* | |
0e1d768f | 5530 | * Unlink from @tsk from its css_set. As migration path can't race |
0de0942d | 5531 | * with us, we can check css_set and cg_list without synchronization. |
817929ec | 5532 | */ |
0de0942d TH |
5533 | cset = task_css_set(tsk); |
5534 | ||
817929ec | 5535 | if (!list_empty(&tsk->cg_list)) { |
96d365e0 | 5536 | down_write(&css_set_rwsem); |
0e1d768f | 5537 | list_del_init(&tsk->cg_list); |
0de0942d TH |
5538 | if (!css_set_populated(cset)) |
5539 | css_set_update_populated(cset, false); | |
96d365e0 | 5540 | up_write(&css_set_rwsem); |
0e1d768f | 5541 | put_cset = true; |
817929ec PM |
5542 | } |
5543 | ||
b4f48b63 | 5544 | /* Reassign the task to the init_css_set. */ |
a8ad805c | 5545 | RCU_INIT_POINTER(tsk->cgroups, &init_css_set); |
d41d5a01 | 5546 | |
cb4a3167 AS |
5547 | /* see cgroup_post_fork() for details */ |
5548 | for_each_subsys_which(ss, i, &have_exit_callback) { | |
5549 | struct cgroup_subsys_state *old_css = cset->subsys[i]; | |
5550 | struct cgroup_subsys_state *css = task_css(tsk, i); | |
30159ec7 | 5551 | |
cb4a3167 | 5552 | ss->exit(css, old_css, tsk); |
d41d5a01 | 5553 | } |
d41d5a01 | 5554 | |
eaf797ab | 5555 | if (put_cset) |
a25eb52e | 5556 | put_css_set(cset); |
b4f48b63 | 5557 | } |
697f4161 | 5558 | |
bd89aabc | 5559 | static void check_for_release(struct cgroup *cgrp) |
81a6a5cd | 5560 | { |
a25eb52e | 5561 | if (notify_on_release(cgrp) && !cgroup_has_tasks(cgrp) && |
971ff493 ZL |
5562 | !css_has_online_children(&cgrp->self) && !cgroup_is_dead(cgrp)) |
5563 | schedule_work(&cgrp->release_agent_work); | |
81a6a5cd PM |
5564 | } |
5565 | ||
81a6a5cd PM |
5566 | /* |
5567 | * Notify userspace when a cgroup is released, by running the | |
5568 | * configured release agent with the name of the cgroup (path | |
5569 | * relative to the root of cgroup file system) as the argument. | |
5570 | * | |
5571 | * Most likely, this user command will try to rmdir this cgroup. | |
5572 | * | |
5573 | * This races with the possibility that some other task will be | |
5574 | * attached to this cgroup before it is removed, or that some other | |
5575 | * user task will 'mkdir' a child cgroup of this cgroup. That's ok. | |
5576 | * The presumed 'rmdir' will fail quietly if this cgroup is no longer | |
5577 | * unused, and this cgroup will be reprieved from its death sentence, | |
5578 | * to continue to serve a useful existence. Next time it's released, | |
5579 | * we will get notified again, if it still has 'notify_on_release' set. | |
5580 | * | |
5581 | * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which | |
5582 | * means only wait until the task is successfully execve()'d. The | |
5583 | * separate release agent task is forked by call_usermodehelper(), | |
5584 | * then control in this thread returns here, without waiting for the | |
5585 | * release agent task. We don't bother to wait because the caller of | |
5586 | * this routine has no use for the exit status of the release agent | |
5587 | * task, so no sense holding our caller up for that. | |
81a6a5cd | 5588 | */ |
81a6a5cd PM |
5589 | static void cgroup_release_agent(struct work_struct *work) |
5590 | { | |
971ff493 ZL |
5591 | struct cgroup *cgrp = |
5592 | container_of(work, struct cgroup, release_agent_work); | |
5593 | char *pathbuf = NULL, *agentbuf = NULL, *path; | |
5594 | char *argv[3], *envp[3]; | |
5595 | ||
81a6a5cd | 5596 | mutex_lock(&cgroup_mutex); |
971ff493 ZL |
5597 | |
5598 | pathbuf = kmalloc(PATH_MAX, GFP_KERNEL); | |
5599 | agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL); | |
5600 | if (!pathbuf || !agentbuf) | |
5601 | goto out; | |
5602 | ||
5603 | path = cgroup_path(cgrp, pathbuf, PATH_MAX); | |
5604 | if (!path) | |
5605 | goto out; | |
5606 | ||
5607 | argv[0] = agentbuf; | |
5608 | argv[1] = path; | |
5609 | argv[2] = NULL; | |
5610 | ||
5611 | /* minimal command environment */ | |
5612 | envp[0] = "HOME=/"; | |
5613 | envp[1] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin"; | |
5614 | envp[2] = NULL; | |
5615 | ||
81a6a5cd | 5616 | mutex_unlock(&cgroup_mutex); |
971ff493 | 5617 | call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC); |
3e2cd91a | 5618 | goto out_free; |
971ff493 | 5619 | out: |
81a6a5cd | 5620 | mutex_unlock(&cgroup_mutex); |
3e2cd91a | 5621 | out_free: |
971ff493 ZL |
5622 | kfree(agentbuf); |
5623 | kfree(pathbuf); | |
81a6a5cd | 5624 | } |
8bab8dde PM |
5625 | |
5626 | static int __init cgroup_disable(char *str) | |
5627 | { | |
30159ec7 | 5628 | struct cgroup_subsys *ss; |
8bab8dde | 5629 | char *token; |
30159ec7 | 5630 | int i; |
8bab8dde PM |
5631 | |
5632 | while ((token = strsep(&str, ",")) != NULL) { | |
5633 | if (!*token) | |
5634 | continue; | |
be45c900 | 5635 | |
3ed80a62 | 5636 | for_each_subsys(ss, i) { |
3e1d2eed TH |
5637 | if (strcmp(token, ss->name) && |
5638 | strcmp(token, ss->legacy_name)) | |
5639 | continue; | |
a3e72739 | 5640 | cgroup_disable_mask |= 1 << i; |
8bab8dde PM |
5641 | } |
5642 | } | |
5643 | return 1; | |
5644 | } | |
5645 | __setup("cgroup_disable=", cgroup_disable); | |
38460b48 | 5646 | |
a8ddc821 TH |
5647 | static int __init cgroup_set_legacy_files_on_dfl(char *str) |
5648 | { | |
5649 | printk("cgroup: using legacy files on the default hierarchy\n"); | |
5650 | cgroup_legacy_files_on_dfl = true; | |
5651 | return 0; | |
5652 | } | |
5653 | __setup("cgroup__DEVEL__legacy_files_on_dfl", cgroup_set_legacy_files_on_dfl); | |
5654 | ||
b77d7b60 | 5655 | /** |
ec903c0c | 5656 | * css_tryget_online_from_dir - get corresponding css from a cgroup dentry |
35cf0836 TH |
5657 | * @dentry: directory dentry of interest |
5658 | * @ss: subsystem of interest | |
b77d7b60 | 5659 | * |
5a17f543 TH |
5660 | * If @dentry is a directory for a cgroup which has @ss enabled on it, try |
5661 | * to get the corresponding css and return it. If such css doesn't exist | |
5662 | * or can't be pinned, an ERR_PTR value is returned. | |
e5d1367f | 5663 | */ |
ec903c0c TH |
5664 | struct cgroup_subsys_state *css_tryget_online_from_dir(struct dentry *dentry, |
5665 | struct cgroup_subsys *ss) | |
e5d1367f | 5666 | { |
2bd59d48 TH |
5667 | struct kernfs_node *kn = kernfs_node_from_dentry(dentry); |
5668 | struct cgroup_subsys_state *css = NULL; | |
e5d1367f | 5669 | struct cgroup *cgrp; |
e5d1367f | 5670 | |
35cf0836 | 5671 | /* is @dentry a cgroup dir? */ |
2bd59d48 TH |
5672 | if (dentry->d_sb->s_type != &cgroup_fs_type || !kn || |
5673 | kernfs_type(kn) != KERNFS_DIR) | |
e5d1367f SE |
5674 | return ERR_PTR(-EBADF); |
5675 | ||
5a17f543 TH |
5676 | rcu_read_lock(); |
5677 | ||
2bd59d48 TH |
5678 | /* |
5679 | * This path doesn't originate from kernfs and @kn could already | |
5680 | * have been or be removed at any point. @kn->priv is RCU | |
a4189487 | 5681 | * protected for this access. See css_release_work_fn() for details. |
2bd59d48 TH |
5682 | */ |
5683 | cgrp = rcu_dereference(kn->priv); | |
5684 | if (cgrp) | |
5685 | css = cgroup_css(cgrp, ss); | |
5a17f543 | 5686 | |
ec903c0c | 5687 | if (!css || !css_tryget_online(css)) |
5a17f543 TH |
5688 | css = ERR_PTR(-ENOENT); |
5689 | ||
5690 | rcu_read_unlock(); | |
5691 | return css; | |
e5d1367f | 5692 | } |
e5d1367f | 5693 | |
1cb650b9 LZ |
5694 | /** |
5695 | * css_from_id - lookup css by id | |
5696 | * @id: the cgroup id | |
5697 | * @ss: cgroup subsys to be looked into | |
5698 | * | |
5699 | * Returns the css if there's valid one with @id, otherwise returns NULL. | |
5700 | * Should be called under rcu_read_lock(). | |
5701 | */ | |
5702 | struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss) | |
5703 | { | |
6fa4918d | 5704 | WARN_ON_ONCE(!rcu_read_lock_held()); |
adbe427b | 5705 | return id > 0 ? idr_find(&ss->css_idr, id) : NULL; |
e5d1367f SE |
5706 | } |
5707 | ||
fe693435 | 5708 | #ifdef CONFIG_CGROUP_DEBUG |
eb95419b TH |
5709 | static struct cgroup_subsys_state * |
5710 | debug_css_alloc(struct cgroup_subsys_state *parent_css) | |
fe693435 PM |
5711 | { |
5712 | struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL); | |
5713 | ||
5714 | if (!css) | |
5715 | return ERR_PTR(-ENOMEM); | |
5716 | ||
5717 | return css; | |
5718 | } | |
5719 | ||
eb95419b | 5720 | static void debug_css_free(struct cgroup_subsys_state *css) |
fe693435 | 5721 | { |
eb95419b | 5722 | kfree(css); |
fe693435 PM |
5723 | } |
5724 | ||
182446d0 TH |
5725 | static u64 debug_taskcount_read(struct cgroup_subsys_state *css, |
5726 | struct cftype *cft) | |
fe693435 | 5727 | { |
182446d0 | 5728 | return cgroup_task_count(css->cgroup); |
fe693435 PM |
5729 | } |
5730 | ||
182446d0 TH |
5731 | static u64 current_css_set_read(struct cgroup_subsys_state *css, |
5732 | struct cftype *cft) | |
fe693435 PM |
5733 | { |
5734 | return (u64)(unsigned long)current->cgroups; | |
5735 | } | |
5736 | ||
182446d0 | 5737 | static u64 current_css_set_refcount_read(struct cgroup_subsys_state *css, |
03c78cbe | 5738 | struct cftype *cft) |
fe693435 PM |
5739 | { |
5740 | u64 count; | |
5741 | ||
5742 | rcu_read_lock(); | |
a8ad805c | 5743 | count = atomic_read(&task_css_set(current)->refcount); |
fe693435 PM |
5744 | rcu_read_unlock(); |
5745 | return count; | |
5746 | } | |
5747 | ||
2da8ca82 | 5748 | static int current_css_set_cg_links_read(struct seq_file *seq, void *v) |
7717f7ba | 5749 | { |
69d0206c | 5750 | struct cgrp_cset_link *link; |
5abb8855 | 5751 | struct css_set *cset; |
e61734c5 TH |
5752 | char *name_buf; |
5753 | ||
5754 | name_buf = kmalloc(NAME_MAX + 1, GFP_KERNEL); | |
5755 | if (!name_buf) | |
5756 | return -ENOMEM; | |
7717f7ba | 5757 | |
96d365e0 | 5758 | down_read(&css_set_rwsem); |
7717f7ba | 5759 | rcu_read_lock(); |
5abb8855 | 5760 | cset = rcu_dereference(current->cgroups); |
69d0206c | 5761 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) { |
7717f7ba | 5762 | struct cgroup *c = link->cgrp; |
7717f7ba | 5763 | |
a2dd4247 | 5764 | cgroup_name(c, name_buf, NAME_MAX + 1); |
2c6ab6d2 | 5765 | seq_printf(seq, "Root %d group %s\n", |
a2dd4247 | 5766 | c->root->hierarchy_id, name_buf); |
7717f7ba PM |
5767 | } |
5768 | rcu_read_unlock(); | |
96d365e0 | 5769 | up_read(&css_set_rwsem); |
e61734c5 | 5770 | kfree(name_buf); |
7717f7ba PM |
5771 | return 0; |
5772 | } | |
5773 | ||
5774 | #define MAX_TASKS_SHOWN_PER_CSS 25 | |
2da8ca82 | 5775 | static int cgroup_css_links_read(struct seq_file *seq, void *v) |
7717f7ba | 5776 | { |
2da8ca82 | 5777 | struct cgroup_subsys_state *css = seq_css(seq); |
69d0206c | 5778 | struct cgrp_cset_link *link; |
7717f7ba | 5779 | |
96d365e0 | 5780 | down_read(&css_set_rwsem); |
182446d0 | 5781 | list_for_each_entry(link, &css->cgroup->cset_links, cset_link) { |
69d0206c | 5782 | struct css_set *cset = link->cset; |
7717f7ba PM |
5783 | struct task_struct *task; |
5784 | int count = 0; | |
c7561128 | 5785 | |
5abb8855 | 5786 | seq_printf(seq, "css_set %p\n", cset); |
c7561128 | 5787 | |
5abb8855 | 5788 | list_for_each_entry(task, &cset->tasks, cg_list) { |
c7561128 TH |
5789 | if (count++ > MAX_TASKS_SHOWN_PER_CSS) |
5790 | goto overflow; | |
5791 | seq_printf(seq, " task %d\n", task_pid_vnr(task)); | |
5792 | } | |
5793 | ||
5794 | list_for_each_entry(task, &cset->mg_tasks, cg_list) { | |
5795 | if (count++ > MAX_TASKS_SHOWN_PER_CSS) | |
5796 | goto overflow; | |
5797 | seq_printf(seq, " task %d\n", task_pid_vnr(task)); | |
7717f7ba | 5798 | } |
c7561128 TH |
5799 | continue; |
5800 | overflow: | |
5801 | seq_puts(seq, " ...\n"); | |
7717f7ba | 5802 | } |
96d365e0 | 5803 | up_read(&css_set_rwsem); |
7717f7ba PM |
5804 | return 0; |
5805 | } | |
5806 | ||
182446d0 | 5807 | static u64 releasable_read(struct cgroup_subsys_state *css, struct cftype *cft) |
fe693435 | 5808 | { |
a25eb52e ZL |
5809 | return (!cgroup_has_tasks(css->cgroup) && |
5810 | !css_has_online_children(&css->cgroup->self)); | |
fe693435 PM |
5811 | } |
5812 | ||
5813 | static struct cftype debug_files[] = { | |
fe693435 PM |
5814 | { |
5815 | .name = "taskcount", | |
5816 | .read_u64 = debug_taskcount_read, | |
5817 | }, | |
5818 | ||
5819 | { | |
5820 | .name = "current_css_set", | |
5821 | .read_u64 = current_css_set_read, | |
5822 | }, | |
5823 | ||
5824 | { | |
5825 | .name = "current_css_set_refcount", | |
5826 | .read_u64 = current_css_set_refcount_read, | |
5827 | }, | |
5828 | ||
7717f7ba PM |
5829 | { |
5830 | .name = "current_css_set_cg_links", | |
2da8ca82 | 5831 | .seq_show = current_css_set_cg_links_read, |
7717f7ba PM |
5832 | }, |
5833 | ||
5834 | { | |
5835 | .name = "cgroup_css_links", | |
2da8ca82 | 5836 | .seq_show = cgroup_css_links_read, |
7717f7ba PM |
5837 | }, |
5838 | ||
fe693435 PM |
5839 | { |
5840 | .name = "releasable", | |
5841 | .read_u64 = releasable_read, | |
5842 | }, | |
fe693435 | 5843 | |
4baf6e33 TH |
5844 | { } /* terminate */ |
5845 | }; | |
fe693435 | 5846 | |
073219e9 | 5847 | struct cgroup_subsys debug_cgrp_subsys = { |
92fb9748 TH |
5848 | .css_alloc = debug_css_alloc, |
5849 | .css_free = debug_css_free, | |
5577964e | 5850 | .legacy_cftypes = debug_files, |
fe693435 PM |
5851 | }; |
5852 | #endif /* CONFIG_CGROUP_DEBUG */ |