]>
Commit | Line | Data |
---|---|---|
ddbcc7e8 | 1 | /* |
ddbcc7e8 PM |
2 | * Generic process-grouping system. |
3 | * | |
4 | * Based originally on the cpuset system, extracted by Paul Menage | |
5 | * Copyright (C) 2006 Google, Inc | |
6 | * | |
0dea1168 KS |
7 | * Notifications support |
8 | * Copyright (C) 2009 Nokia Corporation | |
9 | * Author: Kirill A. Shutemov | |
10 | * | |
ddbcc7e8 PM |
11 | * Copyright notices from the original cpuset code: |
12 | * -------------------------------------------------- | |
13 | * Copyright (C) 2003 BULL SA. | |
14 | * Copyright (C) 2004-2006 Silicon Graphics, Inc. | |
15 | * | |
16 | * Portions derived from Patrick Mochel's sysfs code. | |
17 | * sysfs is Copyright (c) 2001-3 Patrick Mochel | |
18 | * | |
19 | * 2003-10-10 Written by Simon Derr. | |
20 | * 2003-10-22 Updates by Stephen Hemminger. | |
21 | * 2004 May-July Rework by Paul Jackson. | |
22 | * --------------------------------------------------- | |
23 | * | |
24 | * This file is subject to the terms and conditions of the GNU General Public | |
25 | * License. See the file COPYING in the main directory of the Linux | |
26 | * distribution for more details. | |
27 | */ | |
28 | ||
ed3d261b JP |
29 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
30 | ||
ddbcc7e8 | 31 | #include <linux/cgroup.h> |
2ce9738b | 32 | #include <linux/cred.h> |
c6d57f33 | 33 | #include <linux/ctype.h> |
ddbcc7e8 | 34 | #include <linux/errno.h> |
2ce9738b | 35 | #include <linux/init_task.h> |
ddbcc7e8 PM |
36 | #include <linux/kernel.h> |
37 | #include <linux/list.h> | |
c9482a5b | 38 | #include <linux/magic.h> |
ddbcc7e8 PM |
39 | #include <linux/mm.h> |
40 | #include <linux/mutex.h> | |
41 | #include <linux/mount.h> | |
42 | #include <linux/pagemap.h> | |
a424316c | 43 | #include <linux/proc_fs.h> |
ddbcc7e8 PM |
44 | #include <linux/rcupdate.h> |
45 | #include <linux/sched.h> | |
ddbcc7e8 | 46 | #include <linux/slab.h> |
ddbcc7e8 | 47 | #include <linux/spinlock.h> |
1ed13287 | 48 | #include <linux/percpu-rwsem.h> |
ddbcc7e8 | 49 | #include <linux/string.h> |
bbcb81d0 | 50 | #include <linux/sort.h> |
81a6a5cd | 51 | #include <linux/kmod.h> |
846c7bb0 BS |
52 | #include <linux/delayacct.h> |
53 | #include <linux/cgroupstats.h> | |
0ac801fe | 54 | #include <linux/hashtable.h> |
096b7fe0 | 55 | #include <linux/pid_namespace.h> |
2c6ab6d2 | 56 | #include <linux/idr.h> |
d1d9fd33 | 57 | #include <linux/vmalloc.h> /* TODO: replace with more sophisticated array */ |
c4c27fbd | 58 | #include <linux/kthread.h> |
776f02fa | 59 | #include <linux/delay.h> |
60063497 | 60 | #include <linux/atomic.h> |
e93ad19d | 61 | #include <linux/cpuset.h> |
bd1060a1 | 62 | #include <net/sock.h> |
ddbcc7e8 | 63 | |
b1a21367 TH |
64 | /* |
65 | * pidlists linger the following amount before being destroyed. The goal | |
66 | * is avoiding frequent destruction in the middle of consecutive read calls | |
67 | * Expiring in the middle is a performance problem not a correctness one. | |
68 | * 1 sec should be enough. | |
69 | */ | |
70 | #define CGROUP_PIDLIST_DESTROY_DELAY HZ | |
71 | ||
8d7e6fb0 TH |
72 | #define CGROUP_FILE_NAME_MAX (MAX_CGROUP_TYPE_NAMELEN + \ |
73 | MAX_CFTYPE_NAME + 2) | |
74 | ||
e25e2cbb TH |
75 | /* |
76 | * cgroup_mutex is the master lock. Any modification to cgroup or its | |
77 | * hierarchy must be performed while holding it. | |
78 | * | |
f0d9a5f1 | 79 | * css_set_lock protects task->cgroups pointer, the list of css_set |
0e1d768f | 80 | * objects, and the chain of tasks off each css_set. |
e25e2cbb | 81 | * |
0e1d768f TH |
82 | * These locks are exported if CONFIG_PROVE_RCU so that accessors in |
83 | * cgroup.h can use them for lockdep annotations. | |
e25e2cbb | 84 | */ |
2219449a TH |
85 | #ifdef CONFIG_PROVE_RCU |
86 | DEFINE_MUTEX(cgroup_mutex); | |
f0d9a5f1 | 87 | DEFINE_SPINLOCK(css_set_lock); |
0e1d768f | 88 | EXPORT_SYMBOL_GPL(cgroup_mutex); |
f0d9a5f1 | 89 | EXPORT_SYMBOL_GPL(css_set_lock); |
2219449a | 90 | #else |
81a6a5cd | 91 | static DEFINE_MUTEX(cgroup_mutex); |
f0d9a5f1 | 92 | static DEFINE_SPINLOCK(css_set_lock); |
2219449a TH |
93 | #endif |
94 | ||
6fa4918d | 95 | /* |
15a4c835 TH |
96 | * Protects cgroup_idr and css_idr so that IDs can be released without |
97 | * grabbing cgroup_mutex. | |
6fa4918d TH |
98 | */ |
99 | static DEFINE_SPINLOCK(cgroup_idr_lock); | |
100 | ||
34c06254 TH |
101 | /* |
102 | * Protects cgroup_file->kn for !self csses. It synchronizes notifications | |
103 | * against file removal/re-creation across css hiding. | |
104 | */ | |
105 | static DEFINE_SPINLOCK(cgroup_file_kn_lock); | |
106 | ||
69e943b7 TH |
107 | /* |
108 | * Protects cgroup_subsys->release_agent_path. Modifying it also requires | |
109 | * cgroup_mutex. Reading requires either cgroup_mutex or this spinlock. | |
110 | */ | |
111 | static DEFINE_SPINLOCK(release_agent_path_lock); | |
81a6a5cd | 112 | |
1ed13287 TH |
113 | struct percpu_rw_semaphore cgroup_threadgroup_rwsem; |
114 | ||
8353da1f | 115 | #define cgroup_assert_mutex_or_rcu_locked() \ |
f78f5b90 PM |
116 | RCU_LOCKDEP_WARN(!rcu_read_lock_held() && \ |
117 | !lockdep_is_held(&cgroup_mutex), \ | |
8353da1f | 118 | "cgroup_mutex or RCU read lock required"); |
780cd8b3 | 119 | |
e5fca243 TH |
120 | /* |
121 | * cgroup destruction makes heavy use of work items and there can be a lot | |
122 | * of concurrent destructions. Use a separate workqueue so that cgroup | |
123 | * destruction work items don't end up filling up max_active of system_wq | |
124 | * which may lead to deadlock. | |
125 | */ | |
126 | static struct workqueue_struct *cgroup_destroy_wq; | |
127 | ||
b1a21367 TH |
128 | /* |
129 | * pidlist destructions need to be flushed on cgroup destruction. Use a | |
130 | * separate workqueue as flush domain. | |
131 | */ | |
132 | static struct workqueue_struct *cgroup_pidlist_destroy_wq; | |
133 | ||
3ed80a62 | 134 | /* generate an array of cgroup subsystem pointers */ |
073219e9 | 135 | #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys, |
3ed80a62 | 136 | static struct cgroup_subsys *cgroup_subsys[] = { |
ddbcc7e8 PM |
137 | #include <linux/cgroup_subsys.h> |
138 | }; | |
073219e9 TH |
139 | #undef SUBSYS |
140 | ||
141 | /* array of cgroup subsystem names */ | |
142 | #define SUBSYS(_x) [_x ## _cgrp_id] = #_x, | |
143 | static const char *cgroup_subsys_name[] = { | |
ddbcc7e8 PM |
144 | #include <linux/cgroup_subsys.h> |
145 | }; | |
073219e9 | 146 | #undef SUBSYS |
ddbcc7e8 | 147 | |
49d1dc4b TH |
148 | /* array of static_keys for cgroup_subsys_enabled() and cgroup_subsys_on_dfl() */ |
149 | #define SUBSYS(_x) \ | |
150 | DEFINE_STATIC_KEY_TRUE(_x ## _cgrp_subsys_enabled_key); \ | |
151 | DEFINE_STATIC_KEY_TRUE(_x ## _cgrp_subsys_on_dfl_key); \ | |
152 | EXPORT_SYMBOL_GPL(_x ## _cgrp_subsys_enabled_key); \ | |
153 | EXPORT_SYMBOL_GPL(_x ## _cgrp_subsys_on_dfl_key); | |
154 | #include <linux/cgroup_subsys.h> | |
155 | #undef SUBSYS | |
156 | ||
157 | #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys_enabled_key, | |
158 | static struct static_key_true *cgroup_subsys_enabled_key[] = { | |
159 | #include <linux/cgroup_subsys.h> | |
160 | }; | |
161 | #undef SUBSYS | |
162 | ||
163 | #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys_on_dfl_key, | |
164 | static struct static_key_true *cgroup_subsys_on_dfl_key[] = { | |
165 | #include <linux/cgroup_subsys.h> | |
166 | }; | |
167 | #undef SUBSYS | |
168 | ||
ddbcc7e8 | 169 | /* |
3dd06ffa | 170 | * The default hierarchy, reserved for the subsystems that are otherwise |
9871bf95 TH |
171 | * unattached - it never has more than a single cgroup, and all tasks are |
172 | * part of that cgroup. | |
ddbcc7e8 | 173 | */ |
a2dd4247 | 174 | struct cgroup_root cgrp_dfl_root; |
d0ec4230 | 175 | EXPORT_SYMBOL_GPL(cgrp_dfl_root); |
9871bf95 | 176 | |
a2dd4247 TH |
177 | /* |
178 | * The default hierarchy always exists but is hidden until mounted for the | |
179 | * first time. This is for backward compatibility. | |
180 | */ | |
a7165264 | 181 | static bool cgrp_dfl_visible; |
ddbcc7e8 | 182 | |
223ffb29 | 183 | /* Controllers blocked by the commandline in v1 */ |
6e5c8307 | 184 | static u16 cgroup_no_v1_mask; |
223ffb29 | 185 | |
5533e011 | 186 | /* some controllers are not supported in the default hierarchy */ |
a7165264 | 187 | static u16 cgrp_dfl_inhibit_ss_mask; |
5533e011 | 188 | |
ddbcc7e8 PM |
189 | /* The list of hierarchy roots */ |
190 | ||
9871bf95 TH |
191 | static LIST_HEAD(cgroup_roots); |
192 | static int cgroup_root_count; | |
ddbcc7e8 | 193 | |
3417ae1f | 194 | /* hierarchy ID allocation and mapping, protected by cgroup_mutex */ |
1a574231 | 195 | static DEFINE_IDR(cgroup_hierarchy_idr); |
2c6ab6d2 | 196 | |
794611a1 | 197 | /* |
0cb51d71 TH |
198 | * Assign a monotonically increasing serial number to csses. It guarantees |
199 | * cgroups with bigger numbers are newer than those with smaller numbers. | |
200 | * Also, as csses are always appended to the parent's ->children list, it | |
201 | * guarantees that sibling csses are always sorted in the ascending serial | |
202 | * number order on the list. Protected by cgroup_mutex. | |
794611a1 | 203 | */ |
0cb51d71 | 204 | static u64 css_serial_nr_next = 1; |
794611a1 | 205 | |
cb4a3167 AS |
206 | /* |
207 | * These bitmask flags indicate whether tasks in the fork and exit paths have | |
208 | * fork/exit handlers to call. This avoids us having to do extra work in the | |
209 | * fork/exit path to check which subsystems have fork/exit callbacks. | |
ddbcc7e8 | 210 | */ |
6e5c8307 TH |
211 | static u16 have_fork_callback __read_mostly; |
212 | static u16 have_exit_callback __read_mostly; | |
213 | static u16 have_free_callback __read_mostly; | |
ddbcc7e8 | 214 | |
7e47682e | 215 | /* Ditto for the can_fork callback. */ |
6e5c8307 | 216 | static u16 have_canfork_callback __read_mostly; |
7e47682e | 217 | |
67e9c74b | 218 | static struct file_system_type cgroup2_fs_type; |
a14c6874 TH |
219 | static struct cftype cgroup_dfl_base_files[]; |
220 | static struct cftype cgroup_legacy_base_files[]; | |
628f7cd4 | 221 | |
6e5c8307 | 222 | static int rebind_subsystems(struct cgroup_root *dst_root, u16 ss_mask); |
945ba199 | 223 | static void cgroup_lock_and_drain_offline(struct cgroup *cgrp); |
334c3679 TH |
224 | static int cgroup_apply_control(struct cgroup *cgrp); |
225 | static void cgroup_finalize_control(struct cgroup *cgrp, int ret); | |
ed27b9f7 | 226 | static void css_task_iter_advance(struct css_task_iter *it); |
42809dd4 | 227 | static int cgroup_destroy_locked(struct cgroup *cgrp); |
6cd0f5bb TH |
228 | static struct cgroup_subsys_state *css_create(struct cgroup *cgrp, |
229 | struct cgroup_subsys *ss); | |
9d755d33 | 230 | static void css_release(struct percpu_ref *ref); |
f8f22e53 | 231 | static void kill_css(struct cgroup_subsys_state *css); |
4df8dc90 TH |
232 | static int cgroup_addrm_files(struct cgroup_subsys_state *css, |
233 | struct cgroup *cgrp, struct cftype cfts[], | |
2bb566cb | 234 | bool is_add); |
42809dd4 | 235 | |
fc5ed1e9 TH |
236 | /** |
237 | * cgroup_ssid_enabled - cgroup subsys enabled test by subsys ID | |
238 | * @ssid: subsys ID of interest | |
239 | * | |
240 | * cgroup_subsys_enabled() can only be used with literal subsys names which | |
241 | * is fine for individual subsystems but unsuitable for cgroup core. This | |
242 | * is slower static_key_enabled() based test indexed by @ssid. | |
243 | */ | |
244 | static bool cgroup_ssid_enabled(int ssid) | |
245 | { | |
246 | return static_key_enabled(cgroup_subsys_enabled_key[ssid]); | |
247 | } | |
248 | ||
223ffb29 JW |
249 | static bool cgroup_ssid_no_v1(int ssid) |
250 | { | |
251 | return cgroup_no_v1_mask & (1 << ssid); | |
252 | } | |
253 | ||
9e10a130 TH |
254 | /** |
255 | * cgroup_on_dfl - test whether a cgroup is on the default hierarchy | |
256 | * @cgrp: the cgroup of interest | |
257 | * | |
258 | * The default hierarchy is the v2 interface of cgroup and this function | |
259 | * can be used to test whether a cgroup is on the default hierarchy for | |
260 | * cases where a subsystem should behave differnetly depending on the | |
261 | * interface version. | |
262 | * | |
263 | * The set of behaviors which change on the default hierarchy are still | |
264 | * being determined and the mount option is prefixed with __DEVEL__. | |
265 | * | |
266 | * List of changed behaviors: | |
267 | * | |
268 | * - Mount options "noprefix", "xattr", "clone_children", "release_agent" | |
269 | * and "name" are disallowed. | |
270 | * | |
271 | * - When mounting an existing superblock, mount options should match. | |
272 | * | |
273 | * - Remount is disallowed. | |
274 | * | |
275 | * - rename(2) is disallowed. | |
276 | * | |
277 | * - "tasks" is removed. Everything should be at process granularity. Use | |
278 | * "cgroup.procs" instead. | |
279 | * | |
280 | * - "cgroup.procs" is not sorted. pids will be unique unless they got | |
281 | * recycled inbetween reads. | |
282 | * | |
283 | * - "release_agent" and "notify_on_release" are removed. Replacement | |
284 | * notification mechanism will be implemented. | |
285 | * | |
286 | * - "cgroup.clone_children" is removed. | |
287 | * | |
288 | * - "cgroup.subtree_populated" is available. Its value is 0 if the cgroup | |
289 | * and its descendants contain no task; otherwise, 1. The file also | |
290 | * generates kernfs notification which can be monitored through poll and | |
291 | * [di]notify when the value of the file changes. | |
292 | * | |
293 | * - cpuset: tasks will be kept in empty cpusets when hotplug happens and | |
294 | * take masks of ancestors with non-empty cpus/mems, instead of being | |
295 | * moved to an ancestor. | |
296 | * | |
297 | * - cpuset: a task can be moved into an empty cpuset, and again it takes | |
298 | * masks of ancestors. | |
299 | * | |
300 | * - memcg: use_hierarchy is on by default and the cgroup file for the flag | |
301 | * is not created. | |
302 | * | |
303 | * - blkcg: blk-throttle becomes properly hierarchical. | |
304 | * | |
305 | * - debug: disallowed on the default hierarchy. | |
306 | */ | |
307 | static bool cgroup_on_dfl(const struct cgroup *cgrp) | |
308 | { | |
309 | return cgrp->root == &cgrp_dfl_root; | |
310 | } | |
311 | ||
6fa4918d TH |
312 | /* IDR wrappers which synchronize using cgroup_idr_lock */ |
313 | static int cgroup_idr_alloc(struct idr *idr, void *ptr, int start, int end, | |
314 | gfp_t gfp_mask) | |
315 | { | |
316 | int ret; | |
317 | ||
318 | idr_preload(gfp_mask); | |
54504e97 | 319 | spin_lock_bh(&cgroup_idr_lock); |
d0164adc | 320 | ret = idr_alloc(idr, ptr, start, end, gfp_mask & ~__GFP_DIRECT_RECLAIM); |
54504e97 | 321 | spin_unlock_bh(&cgroup_idr_lock); |
6fa4918d TH |
322 | idr_preload_end(); |
323 | return ret; | |
324 | } | |
325 | ||
326 | static void *cgroup_idr_replace(struct idr *idr, void *ptr, int id) | |
327 | { | |
328 | void *ret; | |
329 | ||
54504e97 | 330 | spin_lock_bh(&cgroup_idr_lock); |
6fa4918d | 331 | ret = idr_replace(idr, ptr, id); |
54504e97 | 332 | spin_unlock_bh(&cgroup_idr_lock); |
6fa4918d TH |
333 | return ret; |
334 | } | |
335 | ||
336 | static void cgroup_idr_remove(struct idr *idr, int id) | |
337 | { | |
54504e97 | 338 | spin_lock_bh(&cgroup_idr_lock); |
6fa4918d | 339 | idr_remove(idr, id); |
54504e97 | 340 | spin_unlock_bh(&cgroup_idr_lock); |
6fa4918d TH |
341 | } |
342 | ||
d51f39b0 TH |
343 | static struct cgroup *cgroup_parent(struct cgroup *cgrp) |
344 | { | |
345 | struct cgroup_subsys_state *parent_css = cgrp->self.parent; | |
346 | ||
347 | if (parent_css) | |
348 | return container_of(parent_css, struct cgroup, self); | |
349 | return NULL; | |
350 | } | |
351 | ||
5531dc91 TH |
352 | /* subsystems visibly enabled on a cgroup */ |
353 | static u16 cgroup_control(struct cgroup *cgrp) | |
354 | { | |
355 | struct cgroup *parent = cgroup_parent(cgrp); | |
356 | u16 root_ss_mask = cgrp->root->subsys_mask; | |
357 | ||
358 | if (parent) | |
359 | return parent->subtree_control; | |
360 | ||
361 | if (cgroup_on_dfl(cgrp)) | |
362 | root_ss_mask &= ~cgrp_dfl_inhibit_ss_mask; | |
363 | ||
364 | return root_ss_mask; | |
365 | } | |
366 | ||
367 | /* subsystems enabled on a cgroup */ | |
368 | static u16 cgroup_ss_mask(struct cgroup *cgrp) | |
369 | { | |
370 | struct cgroup *parent = cgroup_parent(cgrp); | |
371 | ||
372 | if (parent) | |
373 | return parent->subtree_ss_mask; | |
374 | ||
375 | return cgrp->root->subsys_mask; | |
376 | } | |
377 | ||
95109b62 TH |
378 | /** |
379 | * cgroup_css - obtain a cgroup's css for the specified subsystem | |
380 | * @cgrp: the cgroup of interest | |
9d800df1 | 381 | * @ss: the subsystem of interest (%NULL returns @cgrp->self) |
95109b62 | 382 | * |
ca8bdcaf TH |
383 | * Return @cgrp's css (cgroup_subsys_state) associated with @ss. This |
384 | * function must be called either under cgroup_mutex or rcu_read_lock() and | |
385 | * the caller is responsible for pinning the returned css if it wants to | |
386 | * keep accessing it outside the said locks. This function may return | |
387 | * %NULL if @cgrp doesn't have @subsys_id enabled. | |
95109b62 TH |
388 | */ |
389 | static struct cgroup_subsys_state *cgroup_css(struct cgroup *cgrp, | |
ca8bdcaf | 390 | struct cgroup_subsys *ss) |
95109b62 | 391 | { |
ca8bdcaf | 392 | if (ss) |
aec25020 | 393 | return rcu_dereference_check(cgrp->subsys[ss->id], |
ace2bee8 | 394 | lockdep_is_held(&cgroup_mutex)); |
ca8bdcaf | 395 | else |
9d800df1 | 396 | return &cgrp->self; |
95109b62 | 397 | } |
42809dd4 | 398 | |
aec3dfcb TH |
399 | /** |
400 | * cgroup_e_css - obtain a cgroup's effective css for the specified subsystem | |
401 | * @cgrp: the cgroup of interest | |
9d800df1 | 402 | * @ss: the subsystem of interest (%NULL returns @cgrp->self) |
aec3dfcb | 403 | * |
d0f702e6 | 404 | * Similar to cgroup_css() but returns the effective css, which is defined |
aec3dfcb TH |
405 | * as the matching css of the nearest ancestor including self which has @ss |
406 | * enabled. If @ss is associated with the hierarchy @cgrp is on, this | |
407 | * function is guaranteed to return non-NULL css. | |
408 | */ | |
409 | static struct cgroup_subsys_state *cgroup_e_css(struct cgroup *cgrp, | |
410 | struct cgroup_subsys *ss) | |
411 | { | |
412 | lockdep_assert_held(&cgroup_mutex); | |
413 | ||
414 | if (!ss) | |
9d800df1 | 415 | return &cgrp->self; |
aec3dfcb | 416 | |
eeecbd19 TH |
417 | /* |
418 | * This function is used while updating css associations and thus | |
5531dc91 | 419 | * can't test the csses directly. Test ss_mask. |
eeecbd19 | 420 | */ |
5531dc91 | 421 | while (!(cgroup_ss_mask(cgrp) & (1 << ss->id))) { |
d51f39b0 | 422 | cgrp = cgroup_parent(cgrp); |
5531dc91 TH |
423 | if (!cgrp) |
424 | return NULL; | |
425 | } | |
aec3dfcb TH |
426 | |
427 | return cgroup_css(cgrp, ss); | |
95109b62 | 428 | } |
42809dd4 | 429 | |
eeecbd19 TH |
430 | /** |
431 | * cgroup_get_e_css - get a cgroup's effective css for the specified subsystem | |
432 | * @cgrp: the cgroup of interest | |
433 | * @ss: the subsystem of interest | |
434 | * | |
435 | * Find and get the effective css of @cgrp for @ss. The effective css is | |
436 | * defined as the matching css of the nearest ancestor including self which | |
437 | * has @ss enabled. If @ss is not mounted on the hierarchy @cgrp is on, | |
438 | * the root css is returned, so this function always returns a valid css. | |
439 | * The returned css must be put using css_put(). | |
440 | */ | |
441 | struct cgroup_subsys_state *cgroup_get_e_css(struct cgroup *cgrp, | |
442 | struct cgroup_subsys *ss) | |
443 | { | |
444 | struct cgroup_subsys_state *css; | |
445 | ||
446 | rcu_read_lock(); | |
447 | ||
448 | do { | |
449 | css = cgroup_css(cgrp, ss); | |
450 | ||
451 | if (css && css_tryget_online(css)) | |
452 | goto out_unlock; | |
453 | cgrp = cgroup_parent(cgrp); | |
454 | } while (cgrp); | |
455 | ||
456 | css = init_css_set.subsys[ss->id]; | |
457 | css_get(css); | |
458 | out_unlock: | |
459 | rcu_read_unlock(); | |
460 | return css; | |
461 | } | |
462 | ||
ddbcc7e8 | 463 | /* convenient tests for these bits */ |
54766d4a | 464 | static inline bool cgroup_is_dead(const struct cgroup *cgrp) |
ddbcc7e8 | 465 | { |
184faf32 | 466 | return !(cgrp->self.flags & CSS_ONLINE); |
ddbcc7e8 PM |
467 | } |
468 | ||
052c3f3a TH |
469 | static void cgroup_get(struct cgroup *cgrp) |
470 | { | |
471 | WARN_ON_ONCE(cgroup_is_dead(cgrp)); | |
472 | css_get(&cgrp->self); | |
473 | } | |
474 | ||
475 | static bool cgroup_tryget(struct cgroup *cgrp) | |
476 | { | |
477 | return css_tryget(&cgrp->self); | |
478 | } | |
479 | ||
b4168640 | 480 | struct cgroup_subsys_state *of_css(struct kernfs_open_file *of) |
59f5296b | 481 | { |
2bd59d48 | 482 | struct cgroup *cgrp = of->kn->parent->priv; |
b4168640 | 483 | struct cftype *cft = of_cft(of); |
2bd59d48 TH |
484 | |
485 | /* | |
486 | * This is open and unprotected implementation of cgroup_css(). | |
487 | * seq_css() is only called from a kernfs file operation which has | |
488 | * an active reference on the file. Because all the subsystem | |
489 | * files are drained before a css is disassociated with a cgroup, | |
490 | * the matching css from the cgroup's subsys table is guaranteed to | |
491 | * be and stay valid until the enclosing operation is complete. | |
492 | */ | |
493 | if (cft->ss) | |
494 | return rcu_dereference_raw(cgrp->subsys[cft->ss->id]); | |
495 | else | |
9d800df1 | 496 | return &cgrp->self; |
59f5296b | 497 | } |
b4168640 | 498 | EXPORT_SYMBOL_GPL(of_css); |
59f5296b | 499 | |
e9685a03 | 500 | static int notify_on_release(const struct cgroup *cgrp) |
81a6a5cd | 501 | { |
bd89aabc | 502 | return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); |
81a6a5cd PM |
503 | } |
504 | ||
1c6727af TH |
505 | /** |
506 | * for_each_css - iterate all css's of a cgroup | |
507 | * @css: the iteration cursor | |
508 | * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end | |
509 | * @cgrp: the target cgroup to iterate css's of | |
510 | * | |
aec3dfcb | 511 | * Should be called under cgroup_[tree_]mutex. |
1c6727af TH |
512 | */ |
513 | #define for_each_css(css, ssid, cgrp) \ | |
514 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \ | |
515 | if (!((css) = rcu_dereference_check( \ | |
516 | (cgrp)->subsys[(ssid)], \ | |
517 | lockdep_is_held(&cgroup_mutex)))) { } \ | |
518 | else | |
519 | ||
aec3dfcb TH |
520 | /** |
521 | * for_each_e_css - iterate all effective css's of a cgroup | |
522 | * @css: the iteration cursor | |
523 | * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end | |
524 | * @cgrp: the target cgroup to iterate css's of | |
525 | * | |
526 | * Should be called under cgroup_[tree_]mutex. | |
527 | */ | |
528 | #define for_each_e_css(css, ssid, cgrp) \ | |
529 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \ | |
530 | if (!((css) = cgroup_e_css(cgrp, cgroup_subsys[(ssid)]))) \ | |
531 | ; \ | |
532 | else | |
533 | ||
30159ec7 | 534 | /** |
3ed80a62 | 535 | * for_each_subsys - iterate all enabled cgroup subsystems |
30159ec7 | 536 | * @ss: the iteration cursor |
780cd8b3 | 537 | * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end |
30159ec7 | 538 | */ |
780cd8b3 | 539 | #define for_each_subsys(ss, ssid) \ |
3ed80a62 TH |
540 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT && \ |
541 | (((ss) = cgroup_subsys[ssid]) || true); (ssid)++) | |
30159ec7 | 542 | |
cb4a3167 | 543 | /** |
b4e0eeaf | 544 | * do_each_subsys_mask - filter for_each_subsys with a bitmask |
cb4a3167 AS |
545 | * @ss: the iteration cursor |
546 | * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end | |
b4e0eeaf | 547 | * @ss_mask: the bitmask |
cb4a3167 AS |
548 | * |
549 | * The block will only run for cases where the ssid-th bit (1 << ssid) of | |
b4e0eeaf | 550 | * @ss_mask is set. |
cb4a3167 | 551 | */ |
b4e0eeaf TH |
552 | #define do_each_subsys_mask(ss, ssid, ss_mask) do { \ |
553 | unsigned long __ss_mask = (ss_mask); \ | |
554 | if (!CGROUP_SUBSYS_COUNT) { /* to avoid spurious gcc warning */ \ | |
4a705c5c | 555 | (ssid) = 0; \ |
b4e0eeaf TH |
556 | break; \ |
557 | } \ | |
558 | for_each_set_bit(ssid, &__ss_mask, CGROUP_SUBSYS_COUNT) { \ | |
559 | (ss) = cgroup_subsys[ssid]; \ | |
560 | { | |
561 | ||
562 | #define while_each_subsys_mask() \ | |
563 | } \ | |
564 | } \ | |
565 | } while (false) | |
cb4a3167 | 566 | |
985ed670 TH |
567 | /* iterate across the hierarchies */ |
568 | #define for_each_root(root) \ | |
5549c497 | 569 | list_for_each_entry((root), &cgroup_roots, root_list) |
ddbcc7e8 | 570 | |
f8f22e53 TH |
571 | /* iterate over child cgrps, lock should be held throughout iteration */ |
572 | #define cgroup_for_each_live_child(child, cgrp) \ | |
d5c419b6 | 573 | list_for_each_entry((child), &(cgrp)->self.children, self.sibling) \ |
8353da1f | 574 | if (({ lockdep_assert_held(&cgroup_mutex); \ |
f8f22e53 TH |
575 | cgroup_is_dead(child); })) \ |
576 | ; \ | |
577 | else | |
7ae1bad9 | 578 | |
ce3f1d9d TH |
579 | /* walk live descendants in preorder */ |
580 | #define cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) \ | |
581 | css_for_each_descendant_pre((d_css), cgroup_css((cgrp), NULL)) \ | |
582 | if (({ lockdep_assert_held(&cgroup_mutex); \ | |
583 | (dsct) = (d_css)->cgroup; \ | |
584 | cgroup_is_dead(dsct); })) \ | |
585 | ; \ | |
586 | else | |
587 | ||
588 | /* walk live descendants in postorder */ | |
589 | #define cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) \ | |
590 | css_for_each_descendant_post((d_css), cgroup_css((cgrp), NULL)) \ | |
591 | if (({ lockdep_assert_held(&cgroup_mutex); \ | |
592 | (dsct) = (d_css)->cgroup; \ | |
593 | cgroup_is_dead(dsct); })) \ | |
594 | ; \ | |
595 | else | |
596 | ||
81a6a5cd | 597 | static void cgroup_release_agent(struct work_struct *work); |
bd89aabc | 598 | static void check_for_release(struct cgroup *cgrp); |
81a6a5cd | 599 | |
69d0206c TH |
600 | /* |
601 | * A cgroup can be associated with multiple css_sets as different tasks may | |
602 | * belong to different cgroups on different hierarchies. In the other | |
603 | * direction, a css_set is naturally associated with multiple cgroups. | |
604 | * This M:N relationship is represented by the following link structure | |
605 | * which exists for each association and allows traversing the associations | |
606 | * from both sides. | |
607 | */ | |
608 | struct cgrp_cset_link { | |
609 | /* the cgroup and css_set this link associates */ | |
610 | struct cgroup *cgrp; | |
611 | struct css_set *cset; | |
612 | ||
613 | /* list of cgrp_cset_links anchored at cgrp->cset_links */ | |
614 | struct list_head cset_link; | |
615 | ||
616 | /* list of cgrp_cset_links anchored at css_set->cgrp_links */ | |
617 | struct list_head cgrp_link; | |
817929ec PM |
618 | }; |
619 | ||
172a2c06 TH |
620 | /* |
621 | * The default css_set - used by init and its children prior to any | |
817929ec PM |
622 | * hierarchies being mounted. It contains a pointer to the root state |
623 | * for each subsystem. Also used to anchor the list of css_sets. Not | |
624 | * reference-counted, to improve performance when child cgroups | |
625 | * haven't been created. | |
626 | */ | |
5024ae29 | 627 | struct css_set init_css_set = { |
172a2c06 TH |
628 | .refcount = ATOMIC_INIT(1), |
629 | .cgrp_links = LIST_HEAD_INIT(init_css_set.cgrp_links), | |
630 | .tasks = LIST_HEAD_INIT(init_css_set.tasks), | |
631 | .mg_tasks = LIST_HEAD_INIT(init_css_set.mg_tasks), | |
632 | .mg_preload_node = LIST_HEAD_INIT(init_css_set.mg_preload_node), | |
633 | .mg_node = LIST_HEAD_INIT(init_css_set.mg_node), | |
ed27b9f7 | 634 | .task_iters = LIST_HEAD_INIT(init_css_set.task_iters), |
172a2c06 | 635 | }; |
817929ec | 636 | |
172a2c06 | 637 | static int css_set_count = 1; /* 1 for init_css_set */ |
817929ec | 638 | |
0de0942d TH |
639 | /** |
640 | * css_set_populated - does a css_set contain any tasks? | |
641 | * @cset: target css_set | |
642 | */ | |
643 | static bool css_set_populated(struct css_set *cset) | |
644 | { | |
f0d9a5f1 | 645 | lockdep_assert_held(&css_set_lock); |
0de0942d TH |
646 | |
647 | return !list_empty(&cset->tasks) || !list_empty(&cset->mg_tasks); | |
648 | } | |
649 | ||
842b597e TH |
650 | /** |
651 | * cgroup_update_populated - updated populated count of a cgroup | |
652 | * @cgrp: the target cgroup | |
653 | * @populated: inc or dec populated count | |
654 | * | |
0de0942d TH |
655 | * One of the css_sets associated with @cgrp is either getting its first |
656 | * task or losing the last. Update @cgrp->populated_cnt accordingly. The | |
657 | * count is propagated towards root so that a given cgroup's populated_cnt | |
658 | * is zero iff the cgroup and all its descendants don't contain any tasks. | |
842b597e TH |
659 | * |
660 | * @cgrp's interface file "cgroup.populated" is zero if | |
661 | * @cgrp->populated_cnt is zero and 1 otherwise. When @cgrp->populated_cnt | |
662 | * changes from or to zero, userland is notified that the content of the | |
663 | * interface file has changed. This can be used to detect when @cgrp and | |
664 | * its descendants become populated or empty. | |
665 | */ | |
666 | static void cgroup_update_populated(struct cgroup *cgrp, bool populated) | |
667 | { | |
f0d9a5f1 | 668 | lockdep_assert_held(&css_set_lock); |
842b597e TH |
669 | |
670 | do { | |
671 | bool trigger; | |
672 | ||
673 | if (populated) | |
674 | trigger = !cgrp->populated_cnt++; | |
675 | else | |
676 | trigger = !--cgrp->populated_cnt; | |
677 | ||
678 | if (!trigger) | |
679 | break; | |
680 | ||
ad2ed2b3 | 681 | check_for_release(cgrp); |
6f60eade TH |
682 | cgroup_file_notify(&cgrp->events_file); |
683 | ||
d51f39b0 | 684 | cgrp = cgroup_parent(cgrp); |
842b597e TH |
685 | } while (cgrp); |
686 | } | |
687 | ||
0de0942d TH |
688 | /** |
689 | * css_set_update_populated - update populated state of a css_set | |
690 | * @cset: target css_set | |
691 | * @populated: whether @cset is populated or depopulated | |
692 | * | |
693 | * @cset is either getting the first task or losing the last. Update the | |
694 | * ->populated_cnt of all associated cgroups accordingly. | |
695 | */ | |
696 | static void css_set_update_populated(struct css_set *cset, bool populated) | |
697 | { | |
698 | struct cgrp_cset_link *link; | |
699 | ||
f0d9a5f1 | 700 | lockdep_assert_held(&css_set_lock); |
0de0942d TH |
701 | |
702 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) | |
703 | cgroup_update_populated(link->cgrp, populated); | |
704 | } | |
705 | ||
f6d7d049 TH |
706 | /** |
707 | * css_set_move_task - move a task from one css_set to another | |
708 | * @task: task being moved | |
709 | * @from_cset: css_set @task currently belongs to (may be NULL) | |
710 | * @to_cset: new css_set @task is being moved to (may be NULL) | |
711 | * @use_mg_tasks: move to @to_cset->mg_tasks instead of ->tasks | |
712 | * | |
713 | * Move @task from @from_cset to @to_cset. If @task didn't belong to any | |
714 | * css_set, @from_cset can be NULL. If @task is being disassociated | |
715 | * instead of moved, @to_cset can be NULL. | |
716 | * | |
ed27b9f7 TH |
717 | * This function automatically handles populated_cnt updates and |
718 | * css_task_iter adjustments but the caller is responsible for managing | |
719 | * @from_cset and @to_cset's reference counts. | |
f6d7d049 TH |
720 | */ |
721 | static void css_set_move_task(struct task_struct *task, | |
722 | struct css_set *from_cset, struct css_set *to_cset, | |
723 | bool use_mg_tasks) | |
724 | { | |
f0d9a5f1 | 725 | lockdep_assert_held(&css_set_lock); |
f6d7d049 | 726 | |
20b454a6 TH |
727 | if (to_cset && !css_set_populated(to_cset)) |
728 | css_set_update_populated(to_cset, true); | |
729 | ||
f6d7d049 | 730 | if (from_cset) { |
ed27b9f7 TH |
731 | struct css_task_iter *it, *pos; |
732 | ||
f6d7d049 | 733 | WARN_ON_ONCE(list_empty(&task->cg_list)); |
ed27b9f7 TH |
734 | |
735 | /* | |
736 | * @task is leaving, advance task iterators which are | |
737 | * pointing to it so that they can resume at the next | |
738 | * position. Advancing an iterator might remove it from | |
739 | * the list, use safe walk. See css_task_iter_advance*() | |
740 | * for details. | |
741 | */ | |
742 | list_for_each_entry_safe(it, pos, &from_cset->task_iters, | |
743 | iters_node) | |
744 | if (it->task_pos == &task->cg_list) | |
745 | css_task_iter_advance(it); | |
746 | ||
f6d7d049 TH |
747 | list_del_init(&task->cg_list); |
748 | if (!css_set_populated(from_cset)) | |
749 | css_set_update_populated(from_cset, false); | |
750 | } else { | |
751 | WARN_ON_ONCE(!list_empty(&task->cg_list)); | |
752 | } | |
753 | ||
754 | if (to_cset) { | |
755 | /* | |
756 | * We are synchronized through cgroup_threadgroup_rwsem | |
757 | * against PF_EXITING setting such that we can't race | |
758 | * against cgroup_exit() changing the css_set to | |
759 | * init_css_set and dropping the old one. | |
760 | */ | |
761 | WARN_ON_ONCE(task->flags & PF_EXITING); | |
762 | ||
f6d7d049 TH |
763 | rcu_assign_pointer(task->cgroups, to_cset); |
764 | list_add_tail(&task->cg_list, use_mg_tasks ? &to_cset->mg_tasks : | |
765 | &to_cset->tasks); | |
766 | } | |
767 | } | |
768 | ||
7717f7ba PM |
769 | /* |
770 | * hash table for cgroup groups. This improves the performance to find | |
771 | * an existing css_set. This hash doesn't (currently) take into | |
772 | * account cgroups in empty hierarchies. | |
773 | */ | |
472b1053 | 774 | #define CSS_SET_HASH_BITS 7 |
0ac801fe | 775 | static DEFINE_HASHTABLE(css_set_table, CSS_SET_HASH_BITS); |
472b1053 | 776 | |
0ac801fe | 777 | static unsigned long css_set_hash(struct cgroup_subsys_state *css[]) |
472b1053 | 778 | { |
0ac801fe | 779 | unsigned long key = 0UL; |
30159ec7 TH |
780 | struct cgroup_subsys *ss; |
781 | int i; | |
472b1053 | 782 | |
30159ec7 | 783 | for_each_subsys(ss, i) |
0ac801fe LZ |
784 | key += (unsigned long)css[i]; |
785 | key = (key >> 16) ^ key; | |
472b1053 | 786 | |
0ac801fe | 787 | return key; |
472b1053 LZ |
788 | } |
789 | ||
a25eb52e | 790 | static void put_css_set_locked(struct css_set *cset) |
b4f48b63 | 791 | { |
69d0206c | 792 | struct cgrp_cset_link *link, *tmp_link; |
2d8f243a TH |
793 | struct cgroup_subsys *ss; |
794 | int ssid; | |
5abb8855 | 795 | |
f0d9a5f1 | 796 | lockdep_assert_held(&css_set_lock); |
89c5509b TH |
797 | |
798 | if (!atomic_dec_and_test(&cset->refcount)) | |
146aa1bd | 799 | return; |
81a6a5cd | 800 | |
53254f90 TH |
801 | /* This css_set is dead. unlink it and release cgroup and css refs */ |
802 | for_each_subsys(ss, ssid) { | |
2d8f243a | 803 | list_del(&cset->e_cset_node[ssid]); |
53254f90 TH |
804 | css_put(cset->subsys[ssid]); |
805 | } | |
5abb8855 | 806 | hash_del(&cset->hlist); |
2c6ab6d2 PM |
807 | css_set_count--; |
808 | ||
69d0206c | 809 | list_for_each_entry_safe(link, tmp_link, &cset->cgrp_links, cgrp_link) { |
69d0206c TH |
810 | list_del(&link->cset_link); |
811 | list_del(&link->cgrp_link); | |
2ceb231b TH |
812 | if (cgroup_parent(link->cgrp)) |
813 | cgroup_put(link->cgrp); | |
2c6ab6d2 | 814 | kfree(link); |
81a6a5cd | 815 | } |
2c6ab6d2 | 816 | |
5abb8855 | 817 | kfree_rcu(cset, rcu_head); |
b4f48b63 PM |
818 | } |
819 | ||
a25eb52e | 820 | static void put_css_set(struct css_set *cset) |
89c5509b TH |
821 | { |
822 | /* | |
823 | * Ensure that the refcount doesn't hit zero while any readers | |
824 | * can see it. Similar to atomic_dec_and_lock(), but for an | |
825 | * rwlock | |
826 | */ | |
827 | if (atomic_add_unless(&cset->refcount, -1, 1)) | |
828 | return; | |
829 | ||
f0d9a5f1 | 830 | spin_lock_bh(&css_set_lock); |
a25eb52e | 831 | put_css_set_locked(cset); |
f0d9a5f1 | 832 | spin_unlock_bh(&css_set_lock); |
89c5509b TH |
833 | } |
834 | ||
817929ec PM |
835 | /* |
836 | * refcounted get/put for css_set objects | |
837 | */ | |
5abb8855 | 838 | static inline void get_css_set(struct css_set *cset) |
817929ec | 839 | { |
5abb8855 | 840 | atomic_inc(&cset->refcount); |
817929ec PM |
841 | } |
842 | ||
b326f9d0 | 843 | /** |
7717f7ba | 844 | * compare_css_sets - helper function for find_existing_css_set(). |
5abb8855 TH |
845 | * @cset: candidate css_set being tested |
846 | * @old_cset: existing css_set for a task | |
7717f7ba PM |
847 | * @new_cgrp: cgroup that's being entered by the task |
848 | * @template: desired set of css pointers in css_set (pre-calculated) | |
849 | * | |
6f4b7e63 | 850 | * Returns true if "cset" matches "old_cset" except for the hierarchy |
7717f7ba PM |
851 | * which "new_cgrp" belongs to, for which it should match "new_cgrp". |
852 | */ | |
5abb8855 TH |
853 | static bool compare_css_sets(struct css_set *cset, |
854 | struct css_set *old_cset, | |
7717f7ba PM |
855 | struct cgroup *new_cgrp, |
856 | struct cgroup_subsys_state *template[]) | |
857 | { | |
858 | struct list_head *l1, *l2; | |
859 | ||
aec3dfcb TH |
860 | /* |
861 | * On the default hierarchy, there can be csets which are | |
862 | * associated with the same set of cgroups but different csses. | |
863 | * Let's first ensure that csses match. | |
864 | */ | |
865 | if (memcmp(template, cset->subsys, sizeof(cset->subsys))) | |
7717f7ba | 866 | return false; |
7717f7ba PM |
867 | |
868 | /* | |
869 | * Compare cgroup pointers in order to distinguish between | |
aec3dfcb TH |
870 | * different cgroups in hierarchies. As different cgroups may |
871 | * share the same effective css, this comparison is always | |
872 | * necessary. | |
7717f7ba | 873 | */ |
69d0206c TH |
874 | l1 = &cset->cgrp_links; |
875 | l2 = &old_cset->cgrp_links; | |
7717f7ba | 876 | while (1) { |
69d0206c | 877 | struct cgrp_cset_link *link1, *link2; |
5abb8855 | 878 | struct cgroup *cgrp1, *cgrp2; |
7717f7ba PM |
879 | |
880 | l1 = l1->next; | |
881 | l2 = l2->next; | |
882 | /* See if we reached the end - both lists are equal length. */ | |
69d0206c TH |
883 | if (l1 == &cset->cgrp_links) { |
884 | BUG_ON(l2 != &old_cset->cgrp_links); | |
7717f7ba PM |
885 | break; |
886 | } else { | |
69d0206c | 887 | BUG_ON(l2 == &old_cset->cgrp_links); |
7717f7ba PM |
888 | } |
889 | /* Locate the cgroups associated with these links. */ | |
69d0206c TH |
890 | link1 = list_entry(l1, struct cgrp_cset_link, cgrp_link); |
891 | link2 = list_entry(l2, struct cgrp_cset_link, cgrp_link); | |
892 | cgrp1 = link1->cgrp; | |
893 | cgrp2 = link2->cgrp; | |
7717f7ba | 894 | /* Hierarchies should be linked in the same order. */ |
5abb8855 | 895 | BUG_ON(cgrp1->root != cgrp2->root); |
7717f7ba PM |
896 | |
897 | /* | |
898 | * If this hierarchy is the hierarchy of the cgroup | |
899 | * that's changing, then we need to check that this | |
900 | * css_set points to the new cgroup; if it's any other | |
901 | * hierarchy, then this css_set should point to the | |
902 | * same cgroup as the old css_set. | |
903 | */ | |
5abb8855 TH |
904 | if (cgrp1->root == new_cgrp->root) { |
905 | if (cgrp1 != new_cgrp) | |
7717f7ba PM |
906 | return false; |
907 | } else { | |
5abb8855 | 908 | if (cgrp1 != cgrp2) |
7717f7ba PM |
909 | return false; |
910 | } | |
911 | } | |
912 | return true; | |
913 | } | |
914 | ||
b326f9d0 TH |
915 | /** |
916 | * find_existing_css_set - init css array and find the matching css_set | |
917 | * @old_cset: the css_set that we're using before the cgroup transition | |
918 | * @cgrp: the cgroup that we're moving into | |
919 | * @template: out param for the new set of csses, should be clear on entry | |
817929ec | 920 | */ |
5abb8855 TH |
921 | static struct css_set *find_existing_css_set(struct css_set *old_cset, |
922 | struct cgroup *cgrp, | |
923 | struct cgroup_subsys_state *template[]) | |
b4f48b63 | 924 | { |
3dd06ffa | 925 | struct cgroup_root *root = cgrp->root; |
30159ec7 | 926 | struct cgroup_subsys *ss; |
5abb8855 | 927 | struct css_set *cset; |
0ac801fe | 928 | unsigned long key; |
b326f9d0 | 929 | int i; |
817929ec | 930 | |
aae8aab4 BB |
931 | /* |
932 | * Build the set of subsystem state objects that we want to see in the | |
933 | * new css_set. while subsystems can change globally, the entries here | |
934 | * won't change, so no need for locking. | |
935 | */ | |
30159ec7 | 936 | for_each_subsys(ss, i) { |
f392e51c | 937 | if (root->subsys_mask & (1UL << i)) { |
aec3dfcb TH |
938 | /* |
939 | * @ss is in this hierarchy, so we want the | |
940 | * effective css from @cgrp. | |
941 | */ | |
942 | template[i] = cgroup_e_css(cgrp, ss); | |
817929ec | 943 | } else { |
aec3dfcb TH |
944 | /* |
945 | * @ss is not in this hierarchy, so we don't want | |
946 | * to change the css. | |
947 | */ | |
5abb8855 | 948 | template[i] = old_cset->subsys[i]; |
817929ec PM |
949 | } |
950 | } | |
951 | ||
0ac801fe | 952 | key = css_set_hash(template); |
5abb8855 TH |
953 | hash_for_each_possible(css_set_table, cset, hlist, key) { |
954 | if (!compare_css_sets(cset, old_cset, cgrp, template)) | |
7717f7ba PM |
955 | continue; |
956 | ||
957 | /* This css_set matches what we need */ | |
5abb8855 | 958 | return cset; |
472b1053 | 959 | } |
817929ec PM |
960 | |
961 | /* No existing cgroup group matched */ | |
962 | return NULL; | |
963 | } | |
964 | ||
69d0206c | 965 | static void free_cgrp_cset_links(struct list_head *links_to_free) |
36553434 | 966 | { |
69d0206c | 967 | struct cgrp_cset_link *link, *tmp_link; |
36553434 | 968 | |
69d0206c TH |
969 | list_for_each_entry_safe(link, tmp_link, links_to_free, cset_link) { |
970 | list_del(&link->cset_link); | |
36553434 LZ |
971 | kfree(link); |
972 | } | |
973 | } | |
974 | ||
69d0206c TH |
975 | /** |
976 | * allocate_cgrp_cset_links - allocate cgrp_cset_links | |
977 | * @count: the number of links to allocate | |
978 | * @tmp_links: list_head the allocated links are put on | |
979 | * | |
980 | * Allocate @count cgrp_cset_link structures and chain them on @tmp_links | |
981 | * through ->cset_link. Returns 0 on success or -errno. | |
817929ec | 982 | */ |
69d0206c | 983 | static int allocate_cgrp_cset_links(int count, struct list_head *tmp_links) |
817929ec | 984 | { |
69d0206c | 985 | struct cgrp_cset_link *link; |
817929ec | 986 | int i; |
69d0206c TH |
987 | |
988 | INIT_LIST_HEAD(tmp_links); | |
989 | ||
817929ec | 990 | for (i = 0; i < count; i++) { |
f4f4be2b | 991 | link = kzalloc(sizeof(*link), GFP_KERNEL); |
817929ec | 992 | if (!link) { |
69d0206c | 993 | free_cgrp_cset_links(tmp_links); |
817929ec PM |
994 | return -ENOMEM; |
995 | } | |
69d0206c | 996 | list_add(&link->cset_link, tmp_links); |
817929ec PM |
997 | } |
998 | return 0; | |
999 | } | |
1000 | ||
c12f65d4 LZ |
1001 | /** |
1002 | * link_css_set - a helper function to link a css_set to a cgroup | |
69d0206c | 1003 | * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links() |
5abb8855 | 1004 | * @cset: the css_set to be linked |
c12f65d4 LZ |
1005 | * @cgrp: the destination cgroup |
1006 | */ | |
69d0206c TH |
1007 | static void link_css_set(struct list_head *tmp_links, struct css_set *cset, |
1008 | struct cgroup *cgrp) | |
c12f65d4 | 1009 | { |
69d0206c | 1010 | struct cgrp_cset_link *link; |
c12f65d4 | 1011 | |
69d0206c | 1012 | BUG_ON(list_empty(tmp_links)); |
6803c006 TH |
1013 | |
1014 | if (cgroup_on_dfl(cgrp)) | |
1015 | cset->dfl_cgrp = cgrp; | |
1016 | ||
69d0206c TH |
1017 | link = list_first_entry(tmp_links, struct cgrp_cset_link, cset_link); |
1018 | link->cset = cset; | |
7717f7ba | 1019 | link->cgrp = cgrp; |
842b597e | 1020 | |
7717f7ba | 1021 | /* |
389b9c1b TH |
1022 | * Always add links to the tail of the lists so that the lists are |
1023 | * in choronological order. | |
7717f7ba | 1024 | */ |
389b9c1b | 1025 | list_move_tail(&link->cset_link, &cgrp->cset_links); |
69d0206c | 1026 | list_add_tail(&link->cgrp_link, &cset->cgrp_links); |
2ceb231b TH |
1027 | |
1028 | if (cgroup_parent(cgrp)) | |
1029 | cgroup_get(cgrp); | |
c12f65d4 LZ |
1030 | } |
1031 | ||
b326f9d0 TH |
1032 | /** |
1033 | * find_css_set - return a new css_set with one cgroup updated | |
1034 | * @old_cset: the baseline css_set | |
1035 | * @cgrp: the cgroup to be updated | |
1036 | * | |
1037 | * Return a new css_set that's equivalent to @old_cset, but with @cgrp | |
1038 | * substituted into the appropriate hierarchy. | |
817929ec | 1039 | */ |
5abb8855 TH |
1040 | static struct css_set *find_css_set(struct css_set *old_cset, |
1041 | struct cgroup *cgrp) | |
817929ec | 1042 | { |
b326f9d0 | 1043 | struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT] = { }; |
5abb8855 | 1044 | struct css_set *cset; |
69d0206c TH |
1045 | struct list_head tmp_links; |
1046 | struct cgrp_cset_link *link; | |
2d8f243a | 1047 | struct cgroup_subsys *ss; |
0ac801fe | 1048 | unsigned long key; |
2d8f243a | 1049 | int ssid; |
472b1053 | 1050 | |
b326f9d0 TH |
1051 | lockdep_assert_held(&cgroup_mutex); |
1052 | ||
817929ec PM |
1053 | /* First see if we already have a cgroup group that matches |
1054 | * the desired set */ | |
f0d9a5f1 | 1055 | spin_lock_bh(&css_set_lock); |
5abb8855 TH |
1056 | cset = find_existing_css_set(old_cset, cgrp, template); |
1057 | if (cset) | |
1058 | get_css_set(cset); | |
f0d9a5f1 | 1059 | spin_unlock_bh(&css_set_lock); |
817929ec | 1060 | |
5abb8855 TH |
1061 | if (cset) |
1062 | return cset; | |
817929ec | 1063 | |
f4f4be2b | 1064 | cset = kzalloc(sizeof(*cset), GFP_KERNEL); |
5abb8855 | 1065 | if (!cset) |
817929ec PM |
1066 | return NULL; |
1067 | ||
69d0206c | 1068 | /* Allocate all the cgrp_cset_link objects that we'll need */ |
9871bf95 | 1069 | if (allocate_cgrp_cset_links(cgroup_root_count, &tmp_links) < 0) { |
5abb8855 | 1070 | kfree(cset); |
817929ec PM |
1071 | return NULL; |
1072 | } | |
1073 | ||
5abb8855 | 1074 | atomic_set(&cset->refcount, 1); |
69d0206c | 1075 | INIT_LIST_HEAD(&cset->cgrp_links); |
5abb8855 | 1076 | INIT_LIST_HEAD(&cset->tasks); |
c7561128 | 1077 | INIT_LIST_HEAD(&cset->mg_tasks); |
1958d2d5 | 1078 | INIT_LIST_HEAD(&cset->mg_preload_node); |
b3dc094e | 1079 | INIT_LIST_HEAD(&cset->mg_node); |
ed27b9f7 | 1080 | INIT_LIST_HEAD(&cset->task_iters); |
5abb8855 | 1081 | INIT_HLIST_NODE(&cset->hlist); |
817929ec PM |
1082 | |
1083 | /* Copy the set of subsystem state objects generated in | |
1084 | * find_existing_css_set() */ | |
5abb8855 | 1085 | memcpy(cset->subsys, template, sizeof(cset->subsys)); |
817929ec | 1086 | |
f0d9a5f1 | 1087 | spin_lock_bh(&css_set_lock); |
817929ec | 1088 | /* Add reference counts and links from the new css_set. */ |
69d0206c | 1089 | list_for_each_entry(link, &old_cset->cgrp_links, cgrp_link) { |
7717f7ba | 1090 | struct cgroup *c = link->cgrp; |
69d0206c | 1091 | |
7717f7ba PM |
1092 | if (c->root == cgrp->root) |
1093 | c = cgrp; | |
69d0206c | 1094 | link_css_set(&tmp_links, cset, c); |
7717f7ba | 1095 | } |
817929ec | 1096 | |
69d0206c | 1097 | BUG_ON(!list_empty(&tmp_links)); |
817929ec | 1098 | |
817929ec | 1099 | css_set_count++; |
472b1053 | 1100 | |
2d8f243a | 1101 | /* Add @cset to the hash table */ |
5abb8855 TH |
1102 | key = css_set_hash(cset->subsys); |
1103 | hash_add(css_set_table, &cset->hlist, key); | |
472b1053 | 1104 | |
53254f90 TH |
1105 | for_each_subsys(ss, ssid) { |
1106 | struct cgroup_subsys_state *css = cset->subsys[ssid]; | |
1107 | ||
2d8f243a | 1108 | list_add_tail(&cset->e_cset_node[ssid], |
53254f90 TH |
1109 | &css->cgroup->e_csets[ssid]); |
1110 | css_get(css); | |
1111 | } | |
2d8f243a | 1112 | |
f0d9a5f1 | 1113 | spin_unlock_bh(&css_set_lock); |
817929ec | 1114 | |
5abb8855 | 1115 | return cset; |
b4f48b63 PM |
1116 | } |
1117 | ||
3dd06ffa | 1118 | static struct cgroup_root *cgroup_root_from_kf(struct kernfs_root *kf_root) |
7717f7ba | 1119 | { |
3dd06ffa | 1120 | struct cgroup *root_cgrp = kf_root->kn->priv; |
2bd59d48 | 1121 | |
3dd06ffa | 1122 | return root_cgrp->root; |
2bd59d48 TH |
1123 | } |
1124 | ||
3dd06ffa | 1125 | static int cgroup_init_root_id(struct cgroup_root *root) |
f2e85d57 TH |
1126 | { |
1127 | int id; | |
1128 | ||
1129 | lockdep_assert_held(&cgroup_mutex); | |
1130 | ||
985ed670 | 1131 | id = idr_alloc_cyclic(&cgroup_hierarchy_idr, root, 0, 0, GFP_KERNEL); |
f2e85d57 TH |
1132 | if (id < 0) |
1133 | return id; | |
1134 | ||
1135 | root->hierarchy_id = id; | |
1136 | return 0; | |
1137 | } | |
1138 | ||
3dd06ffa | 1139 | static void cgroup_exit_root_id(struct cgroup_root *root) |
f2e85d57 TH |
1140 | { |
1141 | lockdep_assert_held(&cgroup_mutex); | |
1142 | ||
1143 | if (root->hierarchy_id) { | |
1144 | idr_remove(&cgroup_hierarchy_idr, root->hierarchy_id); | |
1145 | root->hierarchy_id = 0; | |
1146 | } | |
1147 | } | |
1148 | ||
3dd06ffa | 1149 | static void cgroup_free_root(struct cgroup_root *root) |
f2e85d57 TH |
1150 | { |
1151 | if (root) { | |
d0f702e6 | 1152 | /* hierarchy ID should already have been released */ |
f2e85d57 TH |
1153 | WARN_ON_ONCE(root->hierarchy_id); |
1154 | ||
1155 | idr_destroy(&root->cgroup_idr); | |
1156 | kfree(root); | |
1157 | } | |
1158 | } | |
1159 | ||
3dd06ffa | 1160 | static void cgroup_destroy_root(struct cgroup_root *root) |
59f5296b | 1161 | { |
3dd06ffa | 1162 | struct cgroup *cgrp = &root->cgrp; |
f2e85d57 | 1163 | struct cgrp_cset_link *link, *tmp_link; |
f2e85d57 | 1164 | |
334c3679 | 1165 | cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp); |
f2e85d57 | 1166 | |
776f02fa | 1167 | BUG_ON(atomic_read(&root->nr_cgrps)); |
d5c419b6 | 1168 | BUG_ON(!list_empty(&cgrp->self.children)); |
f2e85d57 | 1169 | |
f2e85d57 | 1170 | /* Rebind all subsystems back to the default hierarchy */ |
334c3679 | 1171 | WARN_ON(rebind_subsystems(&cgrp_dfl_root, root->subsys_mask)); |
7717f7ba | 1172 | |
7717f7ba | 1173 | /* |
f2e85d57 TH |
1174 | * Release all the links from cset_links to this hierarchy's |
1175 | * root cgroup | |
7717f7ba | 1176 | */ |
f0d9a5f1 | 1177 | spin_lock_bh(&css_set_lock); |
f2e85d57 TH |
1178 | |
1179 | list_for_each_entry_safe(link, tmp_link, &cgrp->cset_links, cset_link) { | |
1180 | list_del(&link->cset_link); | |
1181 | list_del(&link->cgrp_link); | |
1182 | kfree(link); | |
1183 | } | |
f0d9a5f1 TH |
1184 | |
1185 | spin_unlock_bh(&css_set_lock); | |
f2e85d57 TH |
1186 | |
1187 | if (!list_empty(&root->root_list)) { | |
1188 | list_del(&root->root_list); | |
1189 | cgroup_root_count--; | |
1190 | } | |
1191 | ||
1192 | cgroup_exit_root_id(root); | |
1193 | ||
1194 | mutex_unlock(&cgroup_mutex); | |
f2e85d57 | 1195 | |
2bd59d48 | 1196 | kernfs_destroy_root(root->kf_root); |
f2e85d57 TH |
1197 | cgroup_free_root(root); |
1198 | } | |
1199 | ||
ceb6a081 TH |
1200 | /* look up cgroup associated with given css_set on the specified hierarchy */ |
1201 | static struct cgroup *cset_cgroup_from_root(struct css_set *cset, | |
3dd06ffa | 1202 | struct cgroup_root *root) |
7717f7ba | 1203 | { |
7717f7ba PM |
1204 | struct cgroup *res = NULL; |
1205 | ||
96d365e0 | 1206 | lockdep_assert_held(&cgroup_mutex); |
f0d9a5f1 | 1207 | lockdep_assert_held(&css_set_lock); |
96d365e0 | 1208 | |
5abb8855 | 1209 | if (cset == &init_css_set) { |
3dd06ffa | 1210 | res = &root->cgrp; |
7717f7ba | 1211 | } else { |
69d0206c TH |
1212 | struct cgrp_cset_link *link; |
1213 | ||
1214 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) { | |
7717f7ba | 1215 | struct cgroup *c = link->cgrp; |
69d0206c | 1216 | |
7717f7ba PM |
1217 | if (c->root == root) { |
1218 | res = c; | |
1219 | break; | |
1220 | } | |
1221 | } | |
1222 | } | |
96d365e0 | 1223 | |
7717f7ba PM |
1224 | BUG_ON(!res); |
1225 | return res; | |
1226 | } | |
1227 | ||
ddbcc7e8 | 1228 | /* |
ceb6a081 | 1229 | * Return the cgroup for "task" from the given hierarchy. Must be |
f0d9a5f1 | 1230 | * called with cgroup_mutex and css_set_lock held. |
ceb6a081 TH |
1231 | */ |
1232 | static struct cgroup *task_cgroup_from_root(struct task_struct *task, | |
3dd06ffa | 1233 | struct cgroup_root *root) |
ceb6a081 TH |
1234 | { |
1235 | /* | |
1236 | * No need to lock the task - since we hold cgroup_mutex the | |
1237 | * task can't change groups, so the only thing that can happen | |
1238 | * is that it exits and its css is set back to init_css_set. | |
1239 | */ | |
1240 | return cset_cgroup_from_root(task_css_set(task), root); | |
1241 | } | |
1242 | ||
ddbcc7e8 | 1243 | /* |
ddbcc7e8 PM |
1244 | * A task must hold cgroup_mutex to modify cgroups. |
1245 | * | |
1246 | * Any task can increment and decrement the count field without lock. | |
1247 | * So in general, code holding cgroup_mutex can't rely on the count | |
1248 | * field not changing. However, if the count goes to zero, then only | |
956db3ca | 1249 | * cgroup_attach_task() can increment it again. Because a count of zero |
ddbcc7e8 PM |
1250 | * means that no tasks are currently attached, therefore there is no |
1251 | * way a task attached to that cgroup can fork (the other way to | |
1252 | * increment the count). So code holding cgroup_mutex can safely | |
1253 | * assume that if the count is zero, it will stay zero. Similarly, if | |
1254 | * a task holds cgroup_mutex on a cgroup with zero count, it | |
1255 | * knows that the cgroup won't be removed, as cgroup_rmdir() | |
1256 | * needs that mutex. | |
1257 | * | |
ddbcc7e8 PM |
1258 | * A cgroup can only be deleted if both its 'count' of using tasks |
1259 | * is zero, and its list of 'children' cgroups is empty. Since all | |
1260 | * tasks in the system use _some_ cgroup, and since there is always at | |
3dd06ffa | 1261 | * least one task in the system (init, pid == 1), therefore, root cgroup |
ddbcc7e8 | 1262 | * always has either children cgroups and/or using tasks. So we don't |
3dd06ffa | 1263 | * need a special hack to ensure that root cgroup cannot be deleted. |
ddbcc7e8 PM |
1264 | * |
1265 | * P.S. One more locking exception. RCU is used to guard the | |
956db3ca | 1266 | * update of a tasks cgroup pointer by cgroup_attach_task() |
ddbcc7e8 PM |
1267 | */ |
1268 | ||
2bd59d48 | 1269 | static struct kernfs_syscall_ops cgroup_kf_syscall_ops; |
828c0950 | 1270 | static const struct file_operations proc_cgroupstats_operations; |
a424316c | 1271 | |
8d7e6fb0 TH |
1272 | static char *cgroup_file_name(struct cgroup *cgrp, const struct cftype *cft, |
1273 | char *buf) | |
ddbcc7e8 | 1274 | { |
3e1d2eed TH |
1275 | struct cgroup_subsys *ss = cft->ss; |
1276 | ||
8d7e6fb0 TH |
1277 | if (cft->ss && !(cft->flags & CFTYPE_NO_PREFIX) && |
1278 | !(cgrp->root->flags & CGRP_ROOT_NOPREFIX)) | |
1279 | snprintf(buf, CGROUP_FILE_NAME_MAX, "%s.%s", | |
3e1d2eed TH |
1280 | cgroup_on_dfl(cgrp) ? ss->name : ss->legacy_name, |
1281 | cft->name); | |
8d7e6fb0 TH |
1282 | else |
1283 | strncpy(buf, cft->name, CGROUP_FILE_NAME_MAX); | |
1284 | return buf; | |
ddbcc7e8 PM |
1285 | } |
1286 | ||
f2e85d57 TH |
1287 | /** |
1288 | * cgroup_file_mode - deduce file mode of a control file | |
1289 | * @cft: the control file in question | |
1290 | * | |
7dbdb199 | 1291 | * S_IRUGO for read, S_IWUSR for write. |
f2e85d57 TH |
1292 | */ |
1293 | static umode_t cgroup_file_mode(const struct cftype *cft) | |
65dff759 | 1294 | { |
f2e85d57 | 1295 | umode_t mode = 0; |
65dff759 | 1296 | |
f2e85d57 TH |
1297 | if (cft->read_u64 || cft->read_s64 || cft->seq_show) |
1298 | mode |= S_IRUGO; | |
1299 | ||
7dbdb199 TH |
1300 | if (cft->write_u64 || cft->write_s64 || cft->write) { |
1301 | if (cft->flags & CFTYPE_WORLD_WRITABLE) | |
1302 | mode |= S_IWUGO; | |
1303 | else | |
1304 | mode |= S_IWUSR; | |
1305 | } | |
f2e85d57 TH |
1306 | |
1307 | return mode; | |
65dff759 LZ |
1308 | } |
1309 | ||
af0ba678 | 1310 | /** |
8699b776 | 1311 | * cgroup_calc_subtree_ss_mask - calculate subtree_ss_mask |
0f060deb | 1312 | * @subtree_control: the new subtree_control mask to consider |
5ced2518 | 1313 | * @this_ss_mask: available subsystems |
af0ba678 TH |
1314 | * |
1315 | * On the default hierarchy, a subsystem may request other subsystems to be | |
1316 | * enabled together through its ->depends_on mask. In such cases, more | |
1317 | * subsystems than specified in "cgroup.subtree_control" may be enabled. | |
1318 | * | |
0f060deb | 1319 | * This function calculates which subsystems need to be enabled if |
5ced2518 | 1320 | * @subtree_control is to be applied while restricted to @this_ss_mask. |
af0ba678 | 1321 | */ |
5ced2518 | 1322 | static u16 cgroup_calc_subtree_ss_mask(u16 subtree_control, u16 this_ss_mask) |
667c2491 | 1323 | { |
6e5c8307 | 1324 | u16 cur_ss_mask = subtree_control; |
af0ba678 TH |
1325 | struct cgroup_subsys *ss; |
1326 | int ssid; | |
1327 | ||
1328 | lockdep_assert_held(&cgroup_mutex); | |
1329 | ||
af0ba678 | 1330 | while (true) { |
6e5c8307 | 1331 | u16 new_ss_mask = cur_ss_mask; |
af0ba678 | 1332 | |
b4e0eeaf | 1333 | do_each_subsys_mask(ss, ssid, cur_ss_mask) { |
a966a4ed | 1334 | new_ss_mask |= ss->depends_on; |
b4e0eeaf | 1335 | } while_each_subsys_mask(); |
af0ba678 TH |
1336 | |
1337 | /* | |
1338 | * Mask out subsystems which aren't available. This can | |
1339 | * happen only if some depended-upon subsystems were bound | |
1340 | * to non-default hierarchies. | |
1341 | */ | |
5ced2518 | 1342 | new_ss_mask &= this_ss_mask; |
af0ba678 TH |
1343 | |
1344 | if (new_ss_mask == cur_ss_mask) | |
1345 | break; | |
1346 | cur_ss_mask = new_ss_mask; | |
1347 | } | |
1348 | ||
0f060deb TH |
1349 | return cur_ss_mask; |
1350 | } | |
1351 | ||
a9746d8d TH |
1352 | /** |
1353 | * cgroup_kn_unlock - unlocking helper for cgroup kernfs methods | |
1354 | * @kn: the kernfs_node being serviced | |
1355 | * | |
1356 | * This helper undoes cgroup_kn_lock_live() and should be invoked before | |
1357 | * the method finishes if locking succeeded. Note that once this function | |
1358 | * returns the cgroup returned by cgroup_kn_lock_live() may become | |
1359 | * inaccessible any time. If the caller intends to continue to access the | |
1360 | * cgroup, it should pin it before invoking this function. | |
1361 | */ | |
1362 | static void cgroup_kn_unlock(struct kernfs_node *kn) | |
ddbcc7e8 | 1363 | { |
a9746d8d TH |
1364 | struct cgroup *cgrp; |
1365 | ||
1366 | if (kernfs_type(kn) == KERNFS_DIR) | |
1367 | cgrp = kn->priv; | |
1368 | else | |
1369 | cgrp = kn->parent->priv; | |
1370 | ||
1371 | mutex_unlock(&cgroup_mutex); | |
a9746d8d TH |
1372 | |
1373 | kernfs_unbreak_active_protection(kn); | |
1374 | cgroup_put(cgrp); | |
ddbcc7e8 PM |
1375 | } |
1376 | ||
a9746d8d TH |
1377 | /** |
1378 | * cgroup_kn_lock_live - locking helper for cgroup kernfs methods | |
1379 | * @kn: the kernfs_node being serviced | |
945ba199 | 1380 | * @drain_offline: perform offline draining on the cgroup |
a9746d8d TH |
1381 | * |
1382 | * This helper is to be used by a cgroup kernfs method currently servicing | |
1383 | * @kn. It breaks the active protection, performs cgroup locking and | |
1384 | * verifies that the associated cgroup is alive. Returns the cgroup if | |
1385 | * alive; otherwise, %NULL. A successful return should be undone by a | |
945ba199 TH |
1386 | * matching cgroup_kn_unlock() invocation. If @drain_offline is %true, the |
1387 | * cgroup is drained of offlining csses before return. | |
a9746d8d TH |
1388 | * |
1389 | * Any cgroup kernfs method implementation which requires locking the | |
1390 | * associated cgroup should use this helper. It avoids nesting cgroup | |
1391 | * locking under kernfs active protection and allows all kernfs operations | |
1392 | * including self-removal. | |
1393 | */ | |
945ba199 TH |
1394 | static struct cgroup *cgroup_kn_lock_live(struct kernfs_node *kn, |
1395 | bool drain_offline) | |
05ef1d7c | 1396 | { |
a9746d8d TH |
1397 | struct cgroup *cgrp; |
1398 | ||
1399 | if (kernfs_type(kn) == KERNFS_DIR) | |
1400 | cgrp = kn->priv; | |
1401 | else | |
1402 | cgrp = kn->parent->priv; | |
05ef1d7c | 1403 | |
2739d3cc | 1404 | /* |
01f6474c | 1405 | * We're gonna grab cgroup_mutex which nests outside kernfs |
a9746d8d TH |
1406 | * active_ref. cgroup liveliness check alone provides enough |
1407 | * protection against removal. Ensure @cgrp stays accessible and | |
1408 | * break the active_ref protection. | |
2739d3cc | 1409 | */ |
aa32362f LZ |
1410 | if (!cgroup_tryget(cgrp)) |
1411 | return NULL; | |
a9746d8d TH |
1412 | kernfs_break_active_protection(kn); |
1413 | ||
945ba199 TH |
1414 | if (drain_offline) |
1415 | cgroup_lock_and_drain_offline(cgrp); | |
1416 | else | |
1417 | mutex_lock(&cgroup_mutex); | |
05ef1d7c | 1418 | |
a9746d8d TH |
1419 | if (!cgroup_is_dead(cgrp)) |
1420 | return cgrp; | |
1421 | ||
1422 | cgroup_kn_unlock(kn); | |
1423 | return NULL; | |
ddbcc7e8 | 1424 | } |
05ef1d7c | 1425 | |
2739d3cc | 1426 | static void cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft) |
05ef1d7c | 1427 | { |
2bd59d48 | 1428 | char name[CGROUP_FILE_NAME_MAX]; |
05ef1d7c | 1429 | |
01f6474c | 1430 | lockdep_assert_held(&cgroup_mutex); |
34c06254 TH |
1431 | |
1432 | if (cft->file_offset) { | |
1433 | struct cgroup_subsys_state *css = cgroup_css(cgrp, cft->ss); | |
1434 | struct cgroup_file *cfile = (void *)css + cft->file_offset; | |
1435 | ||
1436 | spin_lock_irq(&cgroup_file_kn_lock); | |
1437 | cfile->kn = NULL; | |
1438 | spin_unlock_irq(&cgroup_file_kn_lock); | |
1439 | } | |
1440 | ||
2bd59d48 | 1441 | kernfs_remove_by_name(cgrp->kn, cgroup_file_name(cgrp, cft, name)); |
05ef1d7c TH |
1442 | } |
1443 | ||
13af07df | 1444 | /** |
4df8dc90 TH |
1445 | * css_clear_dir - remove subsys files in a cgroup directory |
1446 | * @css: taget css | |
13af07df | 1447 | */ |
334c3679 | 1448 | static void css_clear_dir(struct cgroup_subsys_state *css) |
05ef1d7c | 1449 | { |
334c3679 | 1450 | struct cgroup *cgrp = css->cgroup; |
4df8dc90 | 1451 | struct cftype *cfts; |
05ef1d7c | 1452 | |
88cb04b9 TH |
1453 | if (!(css->flags & CSS_VISIBLE)) |
1454 | return; | |
1455 | ||
1456 | css->flags &= ~CSS_VISIBLE; | |
1457 | ||
4df8dc90 TH |
1458 | list_for_each_entry(cfts, &css->ss->cfts, node) |
1459 | cgroup_addrm_files(css, cgrp, cfts, false); | |
ddbcc7e8 PM |
1460 | } |
1461 | ||
ccdca218 | 1462 | /** |
4df8dc90 TH |
1463 | * css_populate_dir - create subsys files in a cgroup directory |
1464 | * @css: target css | |
ccdca218 TH |
1465 | * |
1466 | * On failure, no file is added. | |
1467 | */ | |
334c3679 | 1468 | static int css_populate_dir(struct cgroup_subsys_state *css) |
ccdca218 | 1469 | { |
334c3679 | 1470 | struct cgroup *cgrp = css->cgroup; |
4df8dc90 TH |
1471 | struct cftype *cfts, *failed_cfts; |
1472 | int ret; | |
ccdca218 | 1473 | |
03970d3c | 1474 | if ((css->flags & CSS_VISIBLE) || !cgrp->kn) |
88cb04b9 TH |
1475 | return 0; |
1476 | ||
4df8dc90 TH |
1477 | if (!css->ss) { |
1478 | if (cgroup_on_dfl(cgrp)) | |
1479 | cfts = cgroup_dfl_base_files; | |
1480 | else | |
1481 | cfts = cgroup_legacy_base_files; | |
ccdca218 | 1482 | |
4df8dc90 TH |
1483 | return cgroup_addrm_files(&cgrp->self, cgrp, cfts, true); |
1484 | } | |
ccdca218 | 1485 | |
4df8dc90 TH |
1486 | list_for_each_entry(cfts, &css->ss->cfts, node) { |
1487 | ret = cgroup_addrm_files(css, cgrp, cfts, true); | |
1488 | if (ret < 0) { | |
1489 | failed_cfts = cfts; | |
1490 | goto err; | |
ccdca218 TH |
1491 | } |
1492 | } | |
88cb04b9 TH |
1493 | |
1494 | css->flags |= CSS_VISIBLE; | |
1495 | ||
ccdca218 TH |
1496 | return 0; |
1497 | err: | |
4df8dc90 TH |
1498 | list_for_each_entry(cfts, &css->ss->cfts, node) { |
1499 | if (cfts == failed_cfts) | |
1500 | break; | |
1501 | cgroup_addrm_files(css, cgrp, cfts, false); | |
1502 | } | |
ccdca218 TH |
1503 | return ret; |
1504 | } | |
1505 | ||
6e5c8307 | 1506 | static int rebind_subsystems(struct cgroup_root *dst_root, u16 ss_mask) |
ddbcc7e8 | 1507 | { |
1ada4838 | 1508 | struct cgroup *dcgrp = &dst_root->cgrp; |
30159ec7 | 1509 | struct cgroup_subsys *ss; |
2d8f243a | 1510 | int ssid, i, ret; |
ddbcc7e8 | 1511 | |
ace2bee8 | 1512 | lockdep_assert_held(&cgroup_mutex); |
ddbcc7e8 | 1513 | |
b4e0eeaf | 1514 | do_each_subsys_mask(ss, ssid, ss_mask) { |
7fd8c565 TH |
1515 | /* if @ss has non-root csses attached to it, can't move */ |
1516 | if (css_next_child(NULL, cgroup_css(&ss->root->cgrp, ss))) | |
3ed80a62 | 1517 | return -EBUSY; |
1d5be6b2 | 1518 | |
5df36032 | 1519 | /* can't move between two non-dummy roots either */ |
7fd8c565 | 1520 | if (ss->root != &cgrp_dfl_root && dst_root != &cgrp_dfl_root) |
5df36032 | 1521 | return -EBUSY; |
b4e0eeaf | 1522 | } while_each_subsys_mask(); |
ddbcc7e8 | 1523 | |
b4e0eeaf | 1524 | do_each_subsys_mask(ss, ssid, ss_mask) { |
1ada4838 TH |
1525 | struct cgroup_root *src_root = ss->root; |
1526 | struct cgroup *scgrp = &src_root->cgrp; | |
1527 | struct cgroup_subsys_state *css = cgroup_css(scgrp, ss); | |
2d8f243a | 1528 | struct css_set *cset; |
a8a648c4 | 1529 | |
1ada4838 | 1530 | WARN_ON(!css || cgroup_css(dcgrp, ss)); |
a8a648c4 | 1531 | |
334c3679 TH |
1532 | /* disable from the source */ |
1533 | src_root->subsys_mask &= ~(1 << ssid); | |
1534 | WARN_ON(cgroup_apply_control(scgrp)); | |
1535 | cgroup_finalize_control(scgrp, 0); | |
4df8dc90 | 1536 | |
334c3679 | 1537 | /* rebind */ |
1ada4838 TH |
1538 | RCU_INIT_POINTER(scgrp->subsys[ssid], NULL); |
1539 | rcu_assign_pointer(dcgrp->subsys[ssid], css); | |
5df36032 | 1540 | ss->root = dst_root; |
1ada4838 | 1541 | css->cgroup = dcgrp; |
73e80ed8 | 1542 | |
f0d9a5f1 | 1543 | spin_lock_bh(&css_set_lock); |
2d8f243a TH |
1544 | hash_for_each(css_set_table, i, cset, hlist) |
1545 | list_move_tail(&cset->e_cset_node[ss->id], | |
1ada4838 | 1546 | &dcgrp->e_csets[ss->id]); |
f0d9a5f1 | 1547 | spin_unlock_bh(&css_set_lock); |
2d8f243a | 1548 | |
bd53d617 | 1549 | /* default hierarchy doesn't enable controllers by default */ |
f392e51c | 1550 | dst_root->subsys_mask |= 1 << ssid; |
49d1dc4b TH |
1551 | if (dst_root == &cgrp_dfl_root) { |
1552 | static_branch_enable(cgroup_subsys_on_dfl_key[ssid]); | |
1553 | } else { | |
1ada4838 | 1554 | dcgrp->subtree_control |= 1 << ssid; |
49d1dc4b | 1555 | static_branch_disable(cgroup_subsys_on_dfl_key[ssid]); |
667c2491 | 1556 | } |
a8a648c4 | 1557 | |
334c3679 TH |
1558 | ret = cgroup_apply_control(dcgrp); |
1559 | if (ret) | |
1560 | pr_warn("partial failure to rebind %s controller (err=%d)\n", | |
1561 | ss->name, ret); | |
1562 | ||
5df36032 TH |
1563 | if (ss->bind) |
1564 | ss->bind(css); | |
b4e0eeaf | 1565 | } while_each_subsys_mask(); |
ddbcc7e8 | 1566 | |
1ada4838 | 1567 | kernfs_activate(dcgrp->kn); |
ddbcc7e8 PM |
1568 | return 0; |
1569 | } | |
1570 | ||
2bd59d48 TH |
1571 | static int cgroup_show_options(struct seq_file *seq, |
1572 | struct kernfs_root *kf_root) | |
ddbcc7e8 | 1573 | { |
3dd06ffa | 1574 | struct cgroup_root *root = cgroup_root_from_kf(kf_root); |
ddbcc7e8 | 1575 | struct cgroup_subsys *ss; |
b85d2040 | 1576 | int ssid; |
ddbcc7e8 | 1577 | |
d98817d4 TH |
1578 | if (root != &cgrp_dfl_root) |
1579 | for_each_subsys(ss, ssid) | |
1580 | if (root->subsys_mask & (1 << ssid)) | |
61e57c0c | 1581 | seq_show_option(seq, ss->legacy_name, NULL); |
93438629 | 1582 | if (root->flags & CGRP_ROOT_NOPREFIX) |
ddbcc7e8 | 1583 | seq_puts(seq, ",noprefix"); |
93438629 | 1584 | if (root->flags & CGRP_ROOT_XATTR) |
03b1cde6 | 1585 | seq_puts(seq, ",xattr"); |
69e943b7 TH |
1586 | |
1587 | spin_lock(&release_agent_path_lock); | |
81a6a5cd | 1588 | if (strlen(root->release_agent_path)) |
a068acf2 KC |
1589 | seq_show_option(seq, "release_agent", |
1590 | root->release_agent_path); | |
69e943b7 TH |
1591 | spin_unlock(&release_agent_path_lock); |
1592 | ||
3dd06ffa | 1593 | if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags)) |
97978e6d | 1594 | seq_puts(seq, ",clone_children"); |
c6d57f33 | 1595 | if (strlen(root->name)) |
a068acf2 | 1596 | seq_show_option(seq, "name", root->name); |
ddbcc7e8 PM |
1597 | return 0; |
1598 | } | |
1599 | ||
1600 | struct cgroup_sb_opts { | |
6e5c8307 | 1601 | u16 subsys_mask; |
69dfa00c | 1602 | unsigned int flags; |
81a6a5cd | 1603 | char *release_agent; |
2260e7fc | 1604 | bool cpuset_clone_children; |
c6d57f33 | 1605 | char *name; |
2c6ab6d2 PM |
1606 | /* User explicitly requested empty subsystem */ |
1607 | bool none; | |
ddbcc7e8 PM |
1608 | }; |
1609 | ||
cf5d5941 | 1610 | static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts) |
ddbcc7e8 | 1611 | { |
32a8cf23 DL |
1612 | char *token, *o = data; |
1613 | bool all_ss = false, one_ss = false; | |
6e5c8307 | 1614 | u16 mask = U16_MAX; |
30159ec7 | 1615 | struct cgroup_subsys *ss; |
7b9a6ba5 | 1616 | int nr_opts = 0; |
30159ec7 | 1617 | int i; |
f9ab5b5b LZ |
1618 | |
1619 | #ifdef CONFIG_CPUSETS | |
6e5c8307 | 1620 | mask = ~((u16)1 << cpuset_cgrp_id); |
f9ab5b5b | 1621 | #endif |
ddbcc7e8 | 1622 | |
c6d57f33 | 1623 | memset(opts, 0, sizeof(*opts)); |
ddbcc7e8 PM |
1624 | |
1625 | while ((token = strsep(&o, ",")) != NULL) { | |
7b9a6ba5 TH |
1626 | nr_opts++; |
1627 | ||
ddbcc7e8 PM |
1628 | if (!*token) |
1629 | return -EINVAL; | |
32a8cf23 | 1630 | if (!strcmp(token, "none")) { |
2c6ab6d2 PM |
1631 | /* Explicitly have no subsystems */ |
1632 | opts->none = true; | |
32a8cf23 DL |
1633 | continue; |
1634 | } | |
1635 | if (!strcmp(token, "all")) { | |
1636 | /* Mutually exclusive option 'all' + subsystem name */ | |
1637 | if (one_ss) | |
1638 | return -EINVAL; | |
1639 | all_ss = true; | |
1640 | continue; | |
1641 | } | |
1642 | if (!strcmp(token, "noprefix")) { | |
93438629 | 1643 | opts->flags |= CGRP_ROOT_NOPREFIX; |
32a8cf23 DL |
1644 | continue; |
1645 | } | |
1646 | if (!strcmp(token, "clone_children")) { | |
2260e7fc | 1647 | opts->cpuset_clone_children = true; |
32a8cf23 DL |
1648 | continue; |
1649 | } | |
03b1cde6 | 1650 | if (!strcmp(token, "xattr")) { |
93438629 | 1651 | opts->flags |= CGRP_ROOT_XATTR; |
03b1cde6 AR |
1652 | continue; |
1653 | } | |
32a8cf23 | 1654 | if (!strncmp(token, "release_agent=", 14)) { |
81a6a5cd PM |
1655 | /* Specifying two release agents is forbidden */ |
1656 | if (opts->release_agent) | |
1657 | return -EINVAL; | |
c6d57f33 | 1658 | opts->release_agent = |
e400c285 | 1659 | kstrndup(token + 14, PATH_MAX - 1, GFP_KERNEL); |
81a6a5cd PM |
1660 | if (!opts->release_agent) |
1661 | return -ENOMEM; | |
32a8cf23 DL |
1662 | continue; |
1663 | } | |
1664 | if (!strncmp(token, "name=", 5)) { | |
c6d57f33 PM |
1665 | const char *name = token + 5; |
1666 | /* Can't specify an empty name */ | |
1667 | if (!strlen(name)) | |
1668 | return -EINVAL; | |
1669 | /* Must match [\w.-]+ */ | |
1670 | for (i = 0; i < strlen(name); i++) { | |
1671 | char c = name[i]; | |
1672 | if (isalnum(c)) | |
1673 | continue; | |
1674 | if ((c == '.') || (c == '-') || (c == '_')) | |
1675 | continue; | |
1676 | return -EINVAL; | |
1677 | } | |
1678 | /* Specifying two names is forbidden */ | |
1679 | if (opts->name) | |
1680 | return -EINVAL; | |
1681 | opts->name = kstrndup(name, | |
e400c285 | 1682 | MAX_CGROUP_ROOT_NAMELEN - 1, |
c6d57f33 PM |
1683 | GFP_KERNEL); |
1684 | if (!opts->name) | |
1685 | return -ENOMEM; | |
32a8cf23 DL |
1686 | |
1687 | continue; | |
1688 | } | |
1689 | ||
30159ec7 | 1690 | for_each_subsys(ss, i) { |
3e1d2eed | 1691 | if (strcmp(token, ss->legacy_name)) |
32a8cf23 | 1692 | continue; |
fc5ed1e9 | 1693 | if (!cgroup_ssid_enabled(i)) |
32a8cf23 | 1694 | continue; |
223ffb29 JW |
1695 | if (cgroup_ssid_no_v1(i)) |
1696 | continue; | |
32a8cf23 DL |
1697 | |
1698 | /* Mutually exclusive option 'all' + subsystem name */ | |
1699 | if (all_ss) | |
1700 | return -EINVAL; | |
69dfa00c | 1701 | opts->subsys_mask |= (1 << i); |
32a8cf23 DL |
1702 | one_ss = true; |
1703 | ||
1704 | break; | |
1705 | } | |
1706 | if (i == CGROUP_SUBSYS_COUNT) | |
1707 | return -ENOENT; | |
1708 | } | |
1709 | ||
7b9a6ba5 TH |
1710 | /* |
1711 | * If the 'all' option was specified select all the subsystems, | |
1712 | * otherwise if 'none', 'name=' and a subsystem name options were | |
1713 | * not specified, let's default to 'all' | |
1714 | */ | |
1715 | if (all_ss || (!one_ss && !opts->none && !opts->name)) | |
1716 | for_each_subsys(ss, i) | |
223ffb29 | 1717 | if (cgroup_ssid_enabled(i) && !cgroup_ssid_no_v1(i)) |
7b9a6ba5 TH |
1718 | opts->subsys_mask |= (1 << i); |
1719 | ||
1720 | /* | |
1721 | * We either have to specify by name or by subsystems. (So all | |
1722 | * empty hierarchies must have a name). | |
1723 | */ | |
1724 | if (!opts->subsys_mask && !opts->name) | |
1725 | return -EINVAL; | |
1726 | ||
f9ab5b5b LZ |
1727 | /* |
1728 | * Option noprefix was introduced just for backward compatibility | |
1729 | * with the old cpuset, so we allow noprefix only if mounting just | |
1730 | * the cpuset subsystem. | |
1731 | */ | |
93438629 | 1732 | if ((opts->flags & CGRP_ROOT_NOPREFIX) && (opts->subsys_mask & mask)) |
f9ab5b5b LZ |
1733 | return -EINVAL; |
1734 | ||
2c6ab6d2 | 1735 | /* Can't specify "none" and some subsystems */ |
a1a71b45 | 1736 | if (opts->subsys_mask && opts->none) |
2c6ab6d2 PM |
1737 | return -EINVAL; |
1738 | ||
ddbcc7e8 PM |
1739 | return 0; |
1740 | } | |
1741 | ||
2bd59d48 | 1742 | static int cgroup_remount(struct kernfs_root *kf_root, int *flags, char *data) |
ddbcc7e8 PM |
1743 | { |
1744 | int ret = 0; | |
3dd06ffa | 1745 | struct cgroup_root *root = cgroup_root_from_kf(kf_root); |
ddbcc7e8 | 1746 | struct cgroup_sb_opts opts; |
6e5c8307 | 1747 | u16 added_mask, removed_mask; |
ddbcc7e8 | 1748 | |
aa6ec29b TH |
1749 | if (root == &cgrp_dfl_root) { |
1750 | pr_err("remount is not allowed\n"); | |
873fe09e TH |
1751 | return -EINVAL; |
1752 | } | |
1753 | ||
334c3679 | 1754 | cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp); |
ddbcc7e8 PM |
1755 | |
1756 | /* See what subsystems are wanted */ | |
1757 | ret = parse_cgroupfs_options(data, &opts); | |
1758 | if (ret) | |
1759 | goto out_unlock; | |
1760 | ||
f392e51c | 1761 | if (opts.subsys_mask != root->subsys_mask || opts.release_agent) |
ed3d261b | 1762 | pr_warn("option changes via remount are deprecated (pid=%d comm=%s)\n", |
a2a1f9ea | 1763 | task_tgid_nr(current), current->comm); |
8b5a5a9d | 1764 | |
f392e51c TH |
1765 | added_mask = opts.subsys_mask & ~root->subsys_mask; |
1766 | removed_mask = root->subsys_mask & ~opts.subsys_mask; | |
13af07df | 1767 | |
cf5d5941 | 1768 | /* Don't allow flags or name to change at remount */ |
7450e90b | 1769 | if ((opts.flags ^ root->flags) || |
cf5d5941 | 1770 | (opts.name && strcmp(opts.name, root->name))) { |
69dfa00c | 1771 | pr_err("option or name mismatch, new: 0x%x \"%s\", old: 0x%x \"%s\"\n", |
7450e90b | 1772 | opts.flags, opts.name ?: "", root->flags, root->name); |
c6d57f33 PM |
1773 | ret = -EINVAL; |
1774 | goto out_unlock; | |
1775 | } | |
1776 | ||
f172e67c | 1777 | /* remounting is not allowed for populated hierarchies */ |
d5c419b6 | 1778 | if (!list_empty(&root->cgrp.self.children)) { |
f172e67c | 1779 | ret = -EBUSY; |
0670e08b | 1780 | goto out_unlock; |
cf5d5941 | 1781 | } |
ddbcc7e8 | 1782 | |
5df36032 | 1783 | ret = rebind_subsystems(root, added_mask); |
3126121f | 1784 | if (ret) |
0670e08b | 1785 | goto out_unlock; |
ddbcc7e8 | 1786 | |
334c3679 | 1787 | WARN_ON(rebind_subsystems(&cgrp_dfl_root, removed_mask)); |
5df36032 | 1788 | |
69e943b7 TH |
1789 | if (opts.release_agent) { |
1790 | spin_lock(&release_agent_path_lock); | |
81a6a5cd | 1791 | strcpy(root->release_agent_path, opts.release_agent); |
69e943b7 TH |
1792 | spin_unlock(&release_agent_path_lock); |
1793 | } | |
ddbcc7e8 | 1794 | out_unlock: |
66bdc9cf | 1795 | kfree(opts.release_agent); |
c6d57f33 | 1796 | kfree(opts.name); |
ddbcc7e8 | 1797 | mutex_unlock(&cgroup_mutex); |
ddbcc7e8 PM |
1798 | return ret; |
1799 | } | |
1800 | ||
afeb0f9f TH |
1801 | /* |
1802 | * To reduce the fork() overhead for systems that are not actually using | |
1803 | * their cgroups capability, we don't maintain the lists running through | |
1804 | * each css_set to its tasks until we see the list actually used - in other | |
1805 | * words after the first mount. | |
1806 | */ | |
1807 | static bool use_task_css_set_links __read_mostly; | |
1808 | ||
1809 | static void cgroup_enable_task_cg_lists(void) | |
1810 | { | |
1811 | struct task_struct *p, *g; | |
1812 | ||
f0d9a5f1 | 1813 | spin_lock_bh(&css_set_lock); |
afeb0f9f TH |
1814 | |
1815 | if (use_task_css_set_links) | |
1816 | goto out_unlock; | |
1817 | ||
1818 | use_task_css_set_links = true; | |
1819 | ||
1820 | /* | |
1821 | * We need tasklist_lock because RCU is not safe against | |
1822 | * while_each_thread(). Besides, a forking task that has passed | |
1823 | * cgroup_post_fork() without seeing use_task_css_set_links = 1 | |
1824 | * is not guaranteed to have its child immediately visible in the | |
1825 | * tasklist if we walk through it with RCU. | |
1826 | */ | |
1827 | read_lock(&tasklist_lock); | |
1828 | do_each_thread(g, p) { | |
afeb0f9f TH |
1829 | WARN_ON_ONCE(!list_empty(&p->cg_list) || |
1830 | task_css_set(p) != &init_css_set); | |
1831 | ||
1832 | /* | |
1833 | * We should check if the process is exiting, otherwise | |
1834 | * it will race with cgroup_exit() in that the list | |
1835 | * entry won't be deleted though the process has exited. | |
f153ad11 TH |
1836 | * Do it while holding siglock so that we don't end up |
1837 | * racing against cgroup_exit(). | |
afeb0f9f | 1838 | */ |
f153ad11 | 1839 | spin_lock_irq(&p->sighand->siglock); |
eaf797ab TH |
1840 | if (!(p->flags & PF_EXITING)) { |
1841 | struct css_set *cset = task_css_set(p); | |
1842 | ||
0de0942d TH |
1843 | if (!css_set_populated(cset)) |
1844 | css_set_update_populated(cset, true); | |
389b9c1b | 1845 | list_add_tail(&p->cg_list, &cset->tasks); |
eaf797ab TH |
1846 | get_css_set(cset); |
1847 | } | |
f153ad11 | 1848 | spin_unlock_irq(&p->sighand->siglock); |
afeb0f9f TH |
1849 | } while_each_thread(g, p); |
1850 | read_unlock(&tasklist_lock); | |
1851 | out_unlock: | |
f0d9a5f1 | 1852 | spin_unlock_bh(&css_set_lock); |
afeb0f9f | 1853 | } |
ddbcc7e8 | 1854 | |
cc31edce PM |
1855 | static void init_cgroup_housekeeping(struct cgroup *cgrp) |
1856 | { | |
2d8f243a TH |
1857 | struct cgroup_subsys *ss; |
1858 | int ssid; | |
1859 | ||
d5c419b6 TH |
1860 | INIT_LIST_HEAD(&cgrp->self.sibling); |
1861 | INIT_LIST_HEAD(&cgrp->self.children); | |
69d0206c | 1862 | INIT_LIST_HEAD(&cgrp->cset_links); |
72a8cb30 BB |
1863 | INIT_LIST_HEAD(&cgrp->pidlists); |
1864 | mutex_init(&cgrp->pidlist_mutex); | |
9d800df1 | 1865 | cgrp->self.cgroup = cgrp; |
184faf32 | 1866 | cgrp->self.flags |= CSS_ONLINE; |
2d8f243a TH |
1867 | |
1868 | for_each_subsys(ss, ssid) | |
1869 | INIT_LIST_HEAD(&cgrp->e_csets[ssid]); | |
f8f22e53 TH |
1870 | |
1871 | init_waitqueue_head(&cgrp->offline_waitq); | |
971ff493 | 1872 | INIT_WORK(&cgrp->release_agent_work, cgroup_release_agent); |
cc31edce | 1873 | } |
c6d57f33 | 1874 | |
3dd06ffa | 1875 | static void init_cgroup_root(struct cgroup_root *root, |
172a2c06 | 1876 | struct cgroup_sb_opts *opts) |
ddbcc7e8 | 1877 | { |
3dd06ffa | 1878 | struct cgroup *cgrp = &root->cgrp; |
b0ca5a84 | 1879 | |
ddbcc7e8 | 1880 | INIT_LIST_HEAD(&root->root_list); |
3c9c825b | 1881 | atomic_set(&root->nr_cgrps, 1); |
bd89aabc | 1882 | cgrp->root = root; |
cc31edce | 1883 | init_cgroup_housekeeping(cgrp); |
4e96ee8e | 1884 | idr_init(&root->cgroup_idr); |
c6d57f33 | 1885 | |
c6d57f33 PM |
1886 | root->flags = opts->flags; |
1887 | if (opts->release_agent) | |
1888 | strcpy(root->release_agent_path, opts->release_agent); | |
1889 | if (opts->name) | |
1890 | strcpy(root->name, opts->name); | |
2260e7fc | 1891 | if (opts->cpuset_clone_children) |
3dd06ffa | 1892 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags); |
c6d57f33 PM |
1893 | } |
1894 | ||
6e5c8307 | 1895 | static int cgroup_setup_root(struct cgroup_root *root, u16 ss_mask) |
2c6ab6d2 | 1896 | { |
d427dfeb | 1897 | LIST_HEAD(tmp_links); |
3dd06ffa | 1898 | struct cgroup *root_cgrp = &root->cgrp; |
d427dfeb | 1899 | struct css_set *cset; |
d427dfeb | 1900 | int i, ret; |
2c6ab6d2 | 1901 | |
d427dfeb | 1902 | lockdep_assert_held(&cgroup_mutex); |
c6d57f33 | 1903 | |
cf780b7d | 1904 | ret = cgroup_idr_alloc(&root->cgroup_idr, root_cgrp, 1, 2, GFP_KERNEL); |
d427dfeb | 1905 | if (ret < 0) |
2bd59d48 | 1906 | goto out; |
d427dfeb | 1907 | root_cgrp->id = ret; |
b11cfb58 | 1908 | root_cgrp->ancestor_ids[0] = ret; |
c6d57f33 | 1909 | |
2aad2a86 TH |
1910 | ret = percpu_ref_init(&root_cgrp->self.refcnt, css_release, 0, |
1911 | GFP_KERNEL); | |
9d755d33 TH |
1912 | if (ret) |
1913 | goto out; | |
1914 | ||
d427dfeb | 1915 | /* |
f0d9a5f1 | 1916 | * We're accessing css_set_count without locking css_set_lock here, |
d427dfeb | 1917 | * but that's OK - it can only be increased by someone holding |
04313591 TH |
1918 | * cgroup_lock, and that's us. Later rebinding may disable |
1919 | * controllers on the default hierarchy and thus create new csets, | |
1920 | * which can't be more than the existing ones. Allocate 2x. | |
d427dfeb | 1921 | */ |
04313591 | 1922 | ret = allocate_cgrp_cset_links(2 * css_set_count, &tmp_links); |
d427dfeb | 1923 | if (ret) |
9d755d33 | 1924 | goto cancel_ref; |
ddbcc7e8 | 1925 | |
985ed670 | 1926 | ret = cgroup_init_root_id(root); |
ddbcc7e8 | 1927 | if (ret) |
9d755d33 | 1928 | goto cancel_ref; |
ddbcc7e8 | 1929 | |
2bd59d48 TH |
1930 | root->kf_root = kernfs_create_root(&cgroup_kf_syscall_ops, |
1931 | KERNFS_ROOT_CREATE_DEACTIVATED, | |
1932 | root_cgrp); | |
1933 | if (IS_ERR(root->kf_root)) { | |
1934 | ret = PTR_ERR(root->kf_root); | |
1935 | goto exit_root_id; | |
1936 | } | |
1937 | root_cgrp->kn = root->kf_root->kn; | |
ddbcc7e8 | 1938 | |
334c3679 | 1939 | ret = css_populate_dir(&root_cgrp->self); |
d427dfeb | 1940 | if (ret) |
2bd59d48 | 1941 | goto destroy_root; |
ddbcc7e8 | 1942 | |
5df36032 | 1943 | ret = rebind_subsystems(root, ss_mask); |
d427dfeb | 1944 | if (ret) |
2bd59d48 | 1945 | goto destroy_root; |
ddbcc7e8 | 1946 | |
d427dfeb TH |
1947 | /* |
1948 | * There must be no failure case after here, since rebinding takes | |
1949 | * care of subsystems' refcounts, which are explicitly dropped in | |
1950 | * the failure exit path. | |
1951 | */ | |
1952 | list_add(&root->root_list, &cgroup_roots); | |
1953 | cgroup_root_count++; | |
0df6a63f | 1954 | |
d427dfeb | 1955 | /* |
3dd06ffa | 1956 | * Link the root cgroup in this hierarchy into all the css_set |
d427dfeb TH |
1957 | * objects. |
1958 | */ | |
f0d9a5f1 | 1959 | spin_lock_bh(&css_set_lock); |
0de0942d | 1960 | hash_for_each(css_set_table, i, cset, hlist) { |
d427dfeb | 1961 | link_css_set(&tmp_links, cset, root_cgrp); |
0de0942d TH |
1962 | if (css_set_populated(cset)) |
1963 | cgroup_update_populated(root_cgrp, true); | |
1964 | } | |
f0d9a5f1 | 1965 | spin_unlock_bh(&css_set_lock); |
ddbcc7e8 | 1966 | |
d5c419b6 | 1967 | BUG_ON(!list_empty(&root_cgrp->self.children)); |
3c9c825b | 1968 | BUG_ON(atomic_read(&root->nr_cgrps) != 1); |
ddbcc7e8 | 1969 | |
2bd59d48 | 1970 | kernfs_activate(root_cgrp->kn); |
d427dfeb | 1971 | ret = 0; |
2bd59d48 | 1972 | goto out; |
d427dfeb | 1973 | |
2bd59d48 TH |
1974 | destroy_root: |
1975 | kernfs_destroy_root(root->kf_root); | |
1976 | root->kf_root = NULL; | |
1977 | exit_root_id: | |
d427dfeb | 1978 | cgroup_exit_root_id(root); |
9d755d33 | 1979 | cancel_ref: |
9a1049da | 1980 | percpu_ref_exit(&root_cgrp->self.refcnt); |
2bd59d48 | 1981 | out: |
d427dfeb TH |
1982 | free_cgrp_cset_links(&tmp_links); |
1983 | return ret; | |
ddbcc7e8 PM |
1984 | } |
1985 | ||
f7e83571 | 1986 | static struct dentry *cgroup_mount(struct file_system_type *fs_type, |
ddbcc7e8 | 1987 | int flags, const char *unused_dev_name, |
f7e83571 | 1988 | void *data) |
ddbcc7e8 | 1989 | { |
67e9c74b | 1990 | bool is_v2 = fs_type == &cgroup2_fs_type; |
3a32bd72 | 1991 | struct super_block *pinned_sb = NULL; |
970317aa | 1992 | struct cgroup_subsys *ss; |
3dd06ffa | 1993 | struct cgroup_root *root; |
ddbcc7e8 | 1994 | struct cgroup_sb_opts opts; |
2bd59d48 | 1995 | struct dentry *dentry; |
8e30e2b8 | 1996 | int ret; |
970317aa | 1997 | int i; |
c6b3d5bc | 1998 | bool new_sb; |
ddbcc7e8 | 1999 | |
56fde9e0 TH |
2000 | /* |
2001 | * The first time anyone tries to mount a cgroup, enable the list | |
2002 | * linking each css_set to its tasks and fix up all existing tasks. | |
2003 | */ | |
2004 | if (!use_task_css_set_links) | |
2005 | cgroup_enable_task_cg_lists(); | |
e37a06f1 | 2006 | |
67e9c74b TH |
2007 | if (is_v2) { |
2008 | if (data) { | |
2009 | pr_err("cgroup2: unknown option \"%s\"\n", (char *)data); | |
2010 | return ERR_PTR(-EINVAL); | |
2011 | } | |
a7165264 | 2012 | cgrp_dfl_visible = true; |
67e9c74b TH |
2013 | root = &cgrp_dfl_root; |
2014 | cgroup_get(&root->cgrp); | |
2015 | goto out_mount; | |
2016 | } | |
2017 | ||
334c3679 | 2018 | cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp); |
8e30e2b8 TH |
2019 | |
2020 | /* First find the desired set of subsystems */ | |
ddbcc7e8 | 2021 | ret = parse_cgroupfs_options(data, &opts); |
c6d57f33 | 2022 | if (ret) |
8e30e2b8 | 2023 | goto out_unlock; |
a015edd2 | 2024 | |
970317aa LZ |
2025 | /* |
2026 | * Destruction of cgroup root is asynchronous, so subsystems may | |
2027 | * still be dying after the previous unmount. Let's drain the | |
2028 | * dying subsystems. We just need to ensure that the ones | |
2029 | * unmounted previously finish dying and don't care about new ones | |
2030 | * starting. Testing ref liveliness is good enough. | |
2031 | */ | |
2032 | for_each_subsys(ss, i) { | |
2033 | if (!(opts.subsys_mask & (1 << i)) || | |
2034 | ss->root == &cgrp_dfl_root) | |
2035 | continue; | |
2036 | ||
2037 | if (!percpu_ref_tryget_live(&ss->root->cgrp.self.refcnt)) { | |
2038 | mutex_unlock(&cgroup_mutex); | |
2039 | msleep(10); | |
2040 | ret = restart_syscall(); | |
2041 | goto out_free; | |
2042 | } | |
2043 | cgroup_put(&ss->root->cgrp); | |
2044 | } | |
2045 | ||
985ed670 | 2046 | for_each_root(root) { |
2bd59d48 | 2047 | bool name_match = false; |
3126121f | 2048 | |
3dd06ffa | 2049 | if (root == &cgrp_dfl_root) |
985ed670 | 2050 | continue; |
3126121f | 2051 | |
cf5d5941 | 2052 | /* |
2bd59d48 TH |
2053 | * If we asked for a name then it must match. Also, if |
2054 | * name matches but sybsys_mask doesn't, we should fail. | |
2055 | * Remember whether name matched. | |
cf5d5941 | 2056 | */ |
2bd59d48 TH |
2057 | if (opts.name) { |
2058 | if (strcmp(opts.name, root->name)) | |
2059 | continue; | |
2060 | name_match = true; | |
2061 | } | |
ddbcc7e8 | 2062 | |
c6d57f33 | 2063 | /* |
2bd59d48 TH |
2064 | * If we asked for subsystems (or explicitly for no |
2065 | * subsystems) then they must match. | |
c6d57f33 | 2066 | */ |
2bd59d48 | 2067 | if ((opts.subsys_mask || opts.none) && |
f392e51c | 2068 | (opts.subsys_mask != root->subsys_mask)) { |
2bd59d48 TH |
2069 | if (!name_match) |
2070 | continue; | |
2071 | ret = -EBUSY; | |
2072 | goto out_unlock; | |
2073 | } | |
873fe09e | 2074 | |
7b9a6ba5 TH |
2075 | if (root->flags ^ opts.flags) |
2076 | pr_warn("new mount options do not match the existing superblock, will be ignored\n"); | |
ddbcc7e8 | 2077 | |
776f02fa | 2078 | /* |
3a32bd72 LZ |
2079 | * We want to reuse @root whose lifetime is governed by its |
2080 | * ->cgrp. Let's check whether @root is alive and keep it | |
2081 | * that way. As cgroup_kill_sb() can happen anytime, we | |
2082 | * want to block it by pinning the sb so that @root doesn't | |
2083 | * get killed before mount is complete. | |
2084 | * | |
2085 | * With the sb pinned, tryget_live can reliably indicate | |
2086 | * whether @root can be reused. If it's being killed, | |
2087 | * drain it. We can use wait_queue for the wait but this | |
2088 | * path is super cold. Let's just sleep a bit and retry. | |
776f02fa | 2089 | */ |
3a32bd72 LZ |
2090 | pinned_sb = kernfs_pin_sb(root->kf_root, NULL); |
2091 | if (IS_ERR(pinned_sb) || | |
2092 | !percpu_ref_tryget_live(&root->cgrp.self.refcnt)) { | |
776f02fa | 2093 | mutex_unlock(&cgroup_mutex); |
3a32bd72 LZ |
2094 | if (!IS_ERR_OR_NULL(pinned_sb)) |
2095 | deactivate_super(pinned_sb); | |
776f02fa | 2096 | msleep(10); |
a015edd2 TH |
2097 | ret = restart_syscall(); |
2098 | goto out_free; | |
776f02fa | 2099 | } |
ddbcc7e8 | 2100 | |
776f02fa | 2101 | ret = 0; |
2bd59d48 | 2102 | goto out_unlock; |
ddbcc7e8 | 2103 | } |
ddbcc7e8 | 2104 | |
817929ec | 2105 | /* |
172a2c06 TH |
2106 | * No such thing, create a new one. name= matching without subsys |
2107 | * specification is allowed for already existing hierarchies but we | |
2108 | * can't create new one without subsys specification. | |
817929ec | 2109 | */ |
172a2c06 TH |
2110 | if (!opts.subsys_mask && !opts.none) { |
2111 | ret = -EINVAL; | |
2112 | goto out_unlock; | |
817929ec | 2113 | } |
817929ec | 2114 | |
172a2c06 TH |
2115 | root = kzalloc(sizeof(*root), GFP_KERNEL); |
2116 | if (!root) { | |
2117 | ret = -ENOMEM; | |
2bd59d48 | 2118 | goto out_unlock; |
839ec545 | 2119 | } |
e5f6a860 | 2120 | |
172a2c06 TH |
2121 | init_cgroup_root(root, &opts); |
2122 | ||
35585573 | 2123 | ret = cgroup_setup_root(root, opts.subsys_mask); |
2bd59d48 TH |
2124 | if (ret) |
2125 | cgroup_free_root(root); | |
fa3ca07e | 2126 | |
8e30e2b8 | 2127 | out_unlock: |
ddbcc7e8 | 2128 | mutex_unlock(&cgroup_mutex); |
a015edd2 | 2129 | out_free: |
c6d57f33 PM |
2130 | kfree(opts.release_agent); |
2131 | kfree(opts.name); | |
03b1cde6 | 2132 | |
2bd59d48 | 2133 | if (ret) |
8e30e2b8 | 2134 | return ERR_PTR(ret); |
67e9c74b | 2135 | out_mount: |
c9482a5b | 2136 | dentry = kernfs_mount(fs_type, flags, root->kf_root, |
67e9c74b TH |
2137 | is_v2 ? CGROUP2_SUPER_MAGIC : CGROUP_SUPER_MAGIC, |
2138 | &new_sb); | |
c6b3d5bc | 2139 | if (IS_ERR(dentry) || !new_sb) |
3dd06ffa | 2140 | cgroup_put(&root->cgrp); |
3a32bd72 LZ |
2141 | |
2142 | /* | |
2143 | * If @pinned_sb, we're reusing an existing root and holding an | |
2144 | * extra ref on its sb. Mount is complete. Put the extra ref. | |
2145 | */ | |
2146 | if (pinned_sb) { | |
2147 | WARN_ON(new_sb); | |
2148 | deactivate_super(pinned_sb); | |
2149 | } | |
2150 | ||
2bd59d48 TH |
2151 | return dentry; |
2152 | } | |
2153 | ||
2154 | static void cgroup_kill_sb(struct super_block *sb) | |
2155 | { | |
2156 | struct kernfs_root *kf_root = kernfs_root_from_sb(sb); | |
3dd06ffa | 2157 | struct cgroup_root *root = cgroup_root_from_kf(kf_root); |
2bd59d48 | 2158 | |
9d755d33 TH |
2159 | /* |
2160 | * If @root doesn't have any mounts or children, start killing it. | |
2161 | * This prevents new mounts by disabling percpu_ref_tryget_live(). | |
2162 | * cgroup_mount() may wait for @root's release. | |
1f779fb2 LZ |
2163 | * |
2164 | * And don't kill the default root. | |
9d755d33 | 2165 | */ |
3c606d35 | 2166 | if (!list_empty(&root->cgrp.self.children) || |
1f779fb2 | 2167 | root == &cgrp_dfl_root) |
9d755d33 TH |
2168 | cgroup_put(&root->cgrp); |
2169 | else | |
2170 | percpu_ref_kill(&root->cgrp.self.refcnt); | |
2171 | ||
2bd59d48 | 2172 | kernfs_kill_sb(sb); |
ddbcc7e8 PM |
2173 | } |
2174 | ||
2175 | static struct file_system_type cgroup_fs_type = { | |
2176 | .name = "cgroup", | |
f7e83571 | 2177 | .mount = cgroup_mount, |
ddbcc7e8 PM |
2178 | .kill_sb = cgroup_kill_sb, |
2179 | }; | |
2180 | ||
67e9c74b TH |
2181 | static struct file_system_type cgroup2_fs_type = { |
2182 | .name = "cgroup2", | |
2183 | .mount = cgroup_mount, | |
2184 | .kill_sb = cgroup_kill_sb, | |
2185 | }; | |
2186 | ||
857a2beb | 2187 | /** |
913ffdb5 | 2188 | * task_cgroup_path - cgroup path of a task in the first cgroup hierarchy |
857a2beb | 2189 | * @task: target task |
857a2beb TH |
2190 | * @buf: the buffer to write the path into |
2191 | * @buflen: the length of the buffer | |
2192 | * | |
913ffdb5 TH |
2193 | * Determine @task's cgroup on the first (the one with the lowest non-zero |
2194 | * hierarchy_id) cgroup hierarchy and copy its path into @buf. This | |
2195 | * function grabs cgroup_mutex and shouldn't be used inside locks used by | |
2196 | * cgroup controller callbacks. | |
2197 | * | |
e61734c5 | 2198 | * Return value is the same as kernfs_path(). |
857a2beb | 2199 | */ |
e61734c5 | 2200 | char *task_cgroup_path(struct task_struct *task, char *buf, size_t buflen) |
857a2beb | 2201 | { |
3dd06ffa | 2202 | struct cgroup_root *root; |
913ffdb5 | 2203 | struct cgroup *cgrp; |
e61734c5 TH |
2204 | int hierarchy_id = 1; |
2205 | char *path = NULL; | |
857a2beb TH |
2206 | |
2207 | mutex_lock(&cgroup_mutex); | |
f0d9a5f1 | 2208 | spin_lock_bh(&css_set_lock); |
857a2beb | 2209 | |
913ffdb5 TH |
2210 | root = idr_get_next(&cgroup_hierarchy_idr, &hierarchy_id); |
2211 | ||
857a2beb TH |
2212 | if (root) { |
2213 | cgrp = task_cgroup_from_root(task, root); | |
e61734c5 | 2214 | path = cgroup_path(cgrp, buf, buflen); |
913ffdb5 TH |
2215 | } else { |
2216 | /* if no hierarchy exists, everyone is in "/" */ | |
e61734c5 TH |
2217 | if (strlcpy(buf, "/", buflen) < buflen) |
2218 | path = buf; | |
857a2beb TH |
2219 | } |
2220 | ||
f0d9a5f1 | 2221 | spin_unlock_bh(&css_set_lock); |
857a2beb | 2222 | mutex_unlock(&cgroup_mutex); |
e61734c5 | 2223 | return path; |
857a2beb | 2224 | } |
913ffdb5 | 2225 | EXPORT_SYMBOL_GPL(task_cgroup_path); |
857a2beb | 2226 | |
b3dc094e | 2227 | /* used to track tasks and other necessary states during migration */ |
2f7ee569 | 2228 | struct cgroup_taskset { |
b3dc094e TH |
2229 | /* the src and dst cset list running through cset->mg_node */ |
2230 | struct list_head src_csets; | |
2231 | struct list_head dst_csets; | |
2232 | ||
1f7dd3e5 TH |
2233 | /* the subsys currently being processed */ |
2234 | int ssid; | |
2235 | ||
b3dc094e TH |
2236 | /* |
2237 | * Fields for cgroup_taskset_*() iteration. | |
2238 | * | |
2239 | * Before migration is committed, the target migration tasks are on | |
2240 | * ->mg_tasks of the csets on ->src_csets. After, on ->mg_tasks of | |
2241 | * the csets on ->dst_csets. ->csets point to either ->src_csets | |
2242 | * or ->dst_csets depending on whether migration is committed. | |
2243 | * | |
2244 | * ->cur_csets and ->cur_task point to the current task position | |
2245 | * during iteration. | |
2246 | */ | |
2247 | struct list_head *csets; | |
2248 | struct css_set *cur_cset; | |
2249 | struct task_struct *cur_task; | |
2f7ee569 TH |
2250 | }; |
2251 | ||
adaae5dc TH |
2252 | #define CGROUP_TASKSET_INIT(tset) (struct cgroup_taskset){ \ |
2253 | .src_csets = LIST_HEAD_INIT(tset.src_csets), \ | |
2254 | .dst_csets = LIST_HEAD_INIT(tset.dst_csets), \ | |
2255 | .csets = &tset.src_csets, \ | |
2256 | } | |
2257 | ||
2258 | /** | |
2259 | * cgroup_taskset_add - try to add a migration target task to a taskset | |
2260 | * @task: target task | |
2261 | * @tset: target taskset | |
2262 | * | |
2263 | * Add @task, which is a migration target, to @tset. This function becomes | |
2264 | * noop if @task doesn't need to be migrated. @task's css_set should have | |
2265 | * been added as a migration source and @task->cg_list will be moved from | |
2266 | * the css_set's tasks list to mg_tasks one. | |
2267 | */ | |
2268 | static void cgroup_taskset_add(struct task_struct *task, | |
2269 | struct cgroup_taskset *tset) | |
2270 | { | |
2271 | struct css_set *cset; | |
2272 | ||
f0d9a5f1 | 2273 | lockdep_assert_held(&css_set_lock); |
adaae5dc TH |
2274 | |
2275 | /* @task either already exited or can't exit until the end */ | |
2276 | if (task->flags & PF_EXITING) | |
2277 | return; | |
2278 | ||
2279 | /* leave @task alone if post_fork() hasn't linked it yet */ | |
2280 | if (list_empty(&task->cg_list)) | |
2281 | return; | |
2282 | ||
2283 | cset = task_css_set(task); | |
2284 | if (!cset->mg_src_cgrp) | |
2285 | return; | |
2286 | ||
2287 | list_move_tail(&task->cg_list, &cset->mg_tasks); | |
2288 | if (list_empty(&cset->mg_node)) | |
2289 | list_add_tail(&cset->mg_node, &tset->src_csets); | |
2290 | if (list_empty(&cset->mg_dst_cset->mg_node)) | |
2291 | list_move_tail(&cset->mg_dst_cset->mg_node, | |
2292 | &tset->dst_csets); | |
2293 | } | |
2294 | ||
2f7ee569 TH |
2295 | /** |
2296 | * cgroup_taskset_first - reset taskset and return the first task | |
2297 | * @tset: taskset of interest | |
1f7dd3e5 | 2298 | * @dst_cssp: output variable for the destination css |
2f7ee569 TH |
2299 | * |
2300 | * @tset iteration is initialized and the first task is returned. | |
2301 | */ | |
1f7dd3e5 TH |
2302 | struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset, |
2303 | struct cgroup_subsys_state **dst_cssp) | |
2f7ee569 | 2304 | { |
b3dc094e TH |
2305 | tset->cur_cset = list_first_entry(tset->csets, struct css_set, mg_node); |
2306 | tset->cur_task = NULL; | |
2307 | ||
1f7dd3e5 | 2308 | return cgroup_taskset_next(tset, dst_cssp); |
2f7ee569 | 2309 | } |
2f7ee569 TH |
2310 | |
2311 | /** | |
2312 | * cgroup_taskset_next - iterate to the next task in taskset | |
2313 | * @tset: taskset of interest | |
1f7dd3e5 | 2314 | * @dst_cssp: output variable for the destination css |
2f7ee569 TH |
2315 | * |
2316 | * Return the next task in @tset. Iteration must have been initialized | |
2317 | * with cgroup_taskset_first(). | |
2318 | */ | |
1f7dd3e5 TH |
2319 | struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset, |
2320 | struct cgroup_subsys_state **dst_cssp) | |
2f7ee569 | 2321 | { |
b3dc094e TH |
2322 | struct css_set *cset = tset->cur_cset; |
2323 | struct task_struct *task = tset->cur_task; | |
2f7ee569 | 2324 | |
b3dc094e TH |
2325 | while (&cset->mg_node != tset->csets) { |
2326 | if (!task) | |
2327 | task = list_first_entry(&cset->mg_tasks, | |
2328 | struct task_struct, cg_list); | |
2329 | else | |
2330 | task = list_next_entry(task, cg_list); | |
2f7ee569 | 2331 | |
b3dc094e TH |
2332 | if (&task->cg_list != &cset->mg_tasks) { |
2333 | tset->cur_cset = cset; | |
2334 | tset->cur_task = task; | |
1f7dd3e5 TH |
2335 | |
2336 | /* | |
2337 | * This function may be called both before and | |
2338 | * after cgroup_taskset_migrate(). The two cases | |
2339 | * can be distinguished by looking at whether @cset | |
2340 | * has its ->mg_dst_cset set. | |
2341 | */ | |
2342 | if (cset->mg_dst_cset) | |
2343 | *dst_cssp = cset->mg_dst_cset->subsys[tset->ssid]; | |
2344 | else | |
2345 | *dst_cssp = cset->subsys[tset->ssid]; | |
2346 | ||
b3dc094e TH |
2347 | return task; |
2348 | } | |
2f7ee569 | 2349 | |
b3dc094e TH |
2350 | cset = list_next_entry(cset, mg_node); |
2351 | task = NULL; | |
2352 | } | |
2f7ee569 | 2353 | |
b3dc094e | 2354 | return NULL; |
2f7ee569 | 2355 | } |
2f7ee569 | 2356 | |
adaae5dc TH |
2357 | /** |
2358 | * cgroup_taskset_migrate - migrate a taskset to a cgroup | |
2359 | * @tset: taget taskset | |
2360 | * @dst_cgrp: destination cgroup | |
2361 | * | |
2362 | * Migrate tasks in @tset to @dst_cgrp. This function fails iff one of the | |
2363 | * ->can_attach callbacks fails and guarantees that either all or none of | |
2364 | * the tasks in @tset are migrated. @tset is consumed regardless of | |
2365 | * success. | |
2366 | */ | |
2367 | static int cgroup_taskset_migrate(struct cgroup_taskset *tset, | |
2368 | struct cgroup *dst_cgrp) | |
2369 | { | |
2370 | struct cgroup_subsys_state *css, *failed_css = NULL; | |
2371 | struct task_struct *task, *tmp_task; | |
2372 | struct css_set *cset, *tmp_cset; | |
2373 | int i, ret; | |
2374 | ||
2375 | /* methods shouldn't be called if no task is actually migrating */ | |
2376 | if (list_empty(&tset->src_csets)) | |
2377 | return 0; | |
2378 | ||
2379 | /* check that we can legitimately attach to the cgroup */ | |
2380 | for_each_e_css(css, i, dst_cgrp) { | |
2381 | if (css->ss->can_attach) { | |
1f7dd3e5 TH |
2382 | tset->ssid = i; |
2383 | ret = css->ss->can_attach(tset); | |
adaae5dc TH |
2384 | if (ret) { |
2385 | failed_css = css; | |
2386 | goto out_cancel_attach; | |
2387 | } | |
2388 | } | |
2389 | } | |
2390 | ||
2391 | /* | |
2392 | * Now that we're guaranteed success, proceed to move all tasks to | |
2393 | * the new cgroup. There are no failure cases after here, so this | |
2394 | * is the commit point. | |
2395 | */ | |
f0d9a5f1 | 2396 | spin_lock_bh(&css_set_lock); |
adaae5dc | 2397 | list_for_each_entry(cset, &tset->src_csets, mg_node) { |
f6d7d049 TH |
2398 | list_for_each_entry_safe(task, tmp_task, &cset->mg_tasks, cg_list) { |
2399 | struct css_set *from_cset = task_css_set(task); | |
2400 | struct css_set *to_cset = cset->mg_dst_cset; | |
2401 | ||
2402 | get_css_set(to_cset); | |
2403 | css_set_move_task(task, from_cset, to_cset, true); | |
2404 | put_css_set_locked(from_cset); | |
2405 | } | |
adaae5dc | 2406 | } |
f0d9a5f1 | 2407 | spin_unlock_bh(&css_set_lock); |
adaae5dc TH |
2408 | |
2409 | /* | |
2410 | * Migration is committed, all target tasks are now on dst_csets. | |
2411 | * Nothing is sensitive to fork() after this point. Notify | |
2412 | * controllers that migration is complete. | |
2413 | */ | |
2414 | tset->csets = &tset->dst_csets; | |
2415 | ||
1f7dd3e5 TH |
2416 | for_each_e_css(css, i, dst_cgrp) { |
2417 | if (css->ss->attach) { | |
2418 | tset->ssid = i; | |
2419 | css->ss->attach(tset); | |
2420 | } | |
2421 | } | |
adaae5dc TH |
2422 | |
2423 | ret = 0; | |
2424 | goto out_release_tset; | |
2425 | ||
2426 | out_cancel_attach: | |
2427 | for_each_e_css(css, i, dst_cgrp) { | |
2428 | if (css == failed_css) | |
2429 | break; | |
1f7dd3e5 TH |
2430 | if (css->ss->cancel_attach) { |
2431 | tset->ssid = i; | |
2432 | css->ss->cancel_attach(tset); | |
2433 | } | |
adaae5dc TH |
2434 | } |
2435 | out_release_tset: | |
f0d9a5f1 | 2436 | spin_lock_bh(&css_set_lock); |
adaae5dc TH |
2437 | list_splice_init(&tset->dst_csets, &tset->src_csets); |
2438 | list_for_each_entry_safe(cset, tmp_cset, &tset->src_csets, mg_node) { | |
2439 | list_splice_tail_init(&cset->mg_tasks, &cset->tasks); | |
2440 | list_del_init(&cset->mg_node); | |
2441 | } | |
f0d9a5f1 | 2442 | spin_unlock_bh(&css_set_lock); |
adaae5dc TH |
2443 | return ret; |
2444 | } | |
2445 | ||
a043e3b2 | 2446 | /** |
1958d2d5 TH |
2447 | * cgroup_migrate_finish - cleanup after attach |
2448 | * @preloaded_csets: list of preloaded css_sets | |
74a1166d | 2449 | * |
1958d2d5 TH |
2450 | * Undo cgroup_migrate_add_src() and cgroup_migrate_prepare_dst(). See |
2451 | * those functions for details. | |
74a1166d | 2452 | */ |
1958d2d5 | 2453 | static void cgroup_migrate_finish(struct list_head *preloaded_csets) |
74a1166d | 2454 | { |
1958d2d5 | 2455 | struct css_set *cset, *tmp_cset; |
74a1166d | 2456 | |
1958d2d5 TH |
2457 | lockdep_assert_held(&cgroup_mutex); |
2458 | ||
f0d9a5f1 | 2459 | spin_lock_bh(&css_set_lock); |
1958d2d5 TH |
2460 | list_for_each_entry_safe(cset, tmp_cset, preloaded_csets, mg_preload_node) { |
2461 | cset->mg_src_cgrp = NULL; | |
2462 | cset->mg_dst_cset = NULL; | |
2463 | list_del_init(&cset->mg_preload_node); | |
a25eb52e | 2464 | put_css_set_locked(cset); |
1958d2d5 | 2465 | } |
f0d9a5f1 | 2466 | spin_unlock_bh(&css_set_lock); |
1958d2d5 TH |
2467 | } |
2468 | ||
2469 | /** | |
2470 | * cgroup_migrate_add_src - add a migration source css_set | |
2471 | * @src_cset: the source css_set to add | |
2472 | * @dst_cgrp: the destination cgroup | |
2473 | * @preloaded_csets: list of preloaded css_sets | |
2474 | * | |
2475 | * Tasks belonging to @src_cset are about to be migrated to @dst_cgrp. Pin | |
2476 | * @src_cset and add it to @preloaded_csets, which should later be cleaned | |
2477 | * up by cgroup_migrate_finish(). | |
2478 | * | |
1ed13287 TH |
2479 | * This function may be called without holding cgroup_threadgroup_rwsem |
2480 | * even if the target is a process. Threads may be created and destroyed | |
2481 | * but as long as cgroup_mutex is not dropped, no new css_set can be put | |
2482 | * into play and the preloaded css_sets are guaranteed to cover all | |
2483 | * migrations. | |
1958d2d5 TH |
2484 | */ |
2485 | static void cgroup_migrate_add_src(struct css_set *src_cset, | |
2486 | struct cgroup *dst_cgrp, | |
2487 | struct list_head *preloaded_csets) | |
2488 | { | |
2489 | struct cgroup *src_cgrp; | |
2490 | ||
2491 | lockdep_assert_held(&cgroup_mutex); | |
f0d9a5f1 | 2492 | lockdep_assert_held(&css_set_lock); |
1958d2d5 TH |
2493 | |
2494 | src_cgrp = cset_cgroup_from_root(src_cset, dst_cgrp->root); | |
2495 | ||
1958d2d5 TH |
2496 | if (!list_empty(&src_cset->mg_preload_node)) |
2497 | return; | |
2498 | ||
2499 | WARN_ON(src_cset->mg_src_cgrp); | |
2500 | WARN_ON(!list_empty(&src_cset->mg_tasks)); | |
2501 | WARN_ON(!list_empty(&src_cset->mg_node)); | |
2502 | ||
2503 | src_cset->mg_src_cgrp = src_cgrp; | |
2504 | get_css_set(src_cset); | |
2505 | list_add(&src_cset->mg_preload_node, preloaded_csets); | |
2506 | } | |
2507 | ||
2508 | /** | |
2509 | * cgroup_migrate_prepare_dst - prepare destination css_sets for migration | |
f817de98 | 2510 | * @dst_cgrp: the destination cgroup (may be %NULL) |
1958d2d5 TH |
2511 | * @preloaded_csets: list of preloaded source css_sets |
2512 | * | |
2513 | * Tasks are about to be moved to @dst_cgrp and all the source css_sets | |
2514 | * have been preloaded to @preloaded_csets. This function looks up and | |
f817de98 TH |
2515 | * pins all destination css_sets, links each to its source, and append them |
2516 | * to @preloaded_csets. If @dst_cgrp is %NULL, the destination of each | |
2517 | * source css_set is assumed to be its cgroup on the default hierarchy. | |
1958d2d5 TH |
2518 | * |
2519 | * This function must be called after cgroup_migrate_add_src() has been | |
2520 | * called on each migration source css_set. After migration is performed | |
2521 | * using cgroup_migrate(), cgroup_migrate_finish() must be called on | |
2522 | * @preloaded_csets. | |
2523 | */ | |
2524 | static int cgroup_migrate_prepare_dst(struct cgroup *dst_cgrp, | |
2525 | struct list_head *preloaded_csets) | |
2526 | { | |
2527 | LIST_HEAD(csets); | |
f817de98 | 2528 | struct css_set *src_cset, *tmp_cset; |
1958d2d5 TH |
2529 | |
2530 | lockdep_assert_held(&cgroup_mutex); | |
2531 | ||
f8f22e53 | 2532 | /* |
62716ea0 | 2533 | * Except for the root, subtree_control must be zero for a cgroup |
f8f22e53 TH |
2534 | * with tasks so that child cgroups don't compete against tasks. |
2535 | */ | |
d51f39b0 | 2536 | if (dst_cgrp && cgroup_on_dfl(dst_cgrp) && cgroup_parent(dst_cgrp) && |
62716ea0 | 2537 | dst_cgrp->subtree_control) |
f8f22e53 TH |
2538 | return -EBUSY; |
2539 | ||
1958d2d5 | 2540 | /* look up the dst cset for each src cset and link it to src */ |
f817de98 | 2541 | list_for_each_entry_safe(src_cset, tmp_cset, preloaded_csets, mg_preload_node) { |
1958d2d5 TH |
2542 | struct css_set *dst_cset; |
2543 | ||
f817de98 TH |
2544 | dst_cset = find_css_set(src_cset, |
2545 | dst_cgrp ?: src_cset->dfl_cgrp); | |
1958d2d5 TH |
2546 | if (!dst_cset) |
2547 | goto err; | |
2548 | ||
2549 | WARN_ON_ONCE(src_cset->mg_dst_cset || dst_cset->mg_dst_cset); | |
f817de98 TH |
2550 | |
2551 | /* | |
2552 | * If src cset equals dst, it's noop. Drop the src. | |
2553 | * cgroup_migrate() will skip the cset too. Note that we | |
2554 | * can't handle src == dst as some nodes are used by both. | |
2555 | */ | |
2556 | if (src_cset == dst_cset) { | |
2557 | src_cset->mg_src_cgrp = NULL; | |
2558 | list_del_init(&src_cset->mg_preload_node); | |
a25eb52e ZL |
2559 | put_css_set(src_cset); |
2560 | put_css_set(dst_cset); | |
f817de98 TH |
2561 | continue; |
2562 | } | |
2563 | ||
1958d2d5 TH |
2564 | src_cset->mg_dst_cset = dst_cset; |
2565 | ||
2566 | if (list_empty(&dst_cset->mg_preload_node)) | |
2567 | list_add(&dst_cset->mg_preload_node, &csets); | |
2568 | else | |
a25eb52e | 2569 | put_css_set(dst_cset); |
1958d2d5 TH |
2570 | } |
2571 | ||
f817de98 | 2572 | list_splice_tail(&csets, preloaded_csets); |
1958d2d5 TH |
2573 | return 0; |
2574 | err: | |
2575 | cgroup_migrate_finish(&csets); | |
2576 | return -ENOMEM; | |
2577 | } | |
2578 | ||
2579 | /** | |
2580 | * cgroup_migrate - migrate a process or task to a cgroup | |
1958d2d5 TH |
2581 | * @leader: the leader of the process or the task to migrate |
2582 | * @threadgroup: whether @leader points to the whole process or a single task | |
9af2ec45 | 2583 | * @cgrp: the destination cgroup |
1958d2d5 TH |
2584 | * |
2585 | * Migrate a process or task denoted by @leader to @cgrp. If migrating a | |
1ed13287 | 2586 | * process, the caller must be holding cgroup_threadgroup_rwsem. The |
1958d2d5 TH |
2587 | * caller is also responsible for invoking cgroup_migrate_add_src() and |
2588 | * cgroup_migrate_prepare_dst() on the targets before invoking this | |
2589 | * function and following up with cgroup_migrate_finish(). | |
2590 | * | |
2591 | * As long as a controller's ->can_attach() doesn't fail, this function is | |
2592 | * guaranteed to succeed. This means that, excluding ->can_attach() | |
2593 | * failure, when migrating multiple targets, the success or failure can be | |
2594 | * decided for all targets by invoking group_migrate_prepare_dst() before | |
2595 | * actually starting migrating. | |
2596 | */ | |
9af2ec45 TH |
2597 | static int cgroup_migrate(struct task_struct *leader, bool threadgroup, |
2598 | struct cgroup *cgrp) | |
74a1166d | 2599 | { |
adaae5dc TH |
2600 | struct cgroup_taskset tset = CGROUP_TASKSET_INIT(tset); |
2601 | struct task_struct *task; | |
74a1166d | 2602 | |
fb5d2b4c MSB |
2603 | /* |
2604 | * Prevent freeing of tasks while we take a snapshot. Tasks that are | |
2605 | * already PF_EXITING could be freed from underneath us unless we | |
2606 | * take an rcu_read_lock. | |
2607 | */ | |
f0d9a5f1 | 2608 | spin_lock_bh(&css_set_lock); |
fb5d2b4c | 2609 | rcu_read_lock(); |
9db8de37 | 2610 | task = leader; |
74a1166d | 2611 | do { |
adaae5dc | 2612 | cgroup_taskset_add(task, &tset); |
081aa458 LZ |
2613 | if (!threadgroup) |
2614 | break; | |
9db8de37 | 2615 | } while_each_thread(leader, task); |
fb5d2b4c | 2616 | rcu_read_unlock(); |
f0d9a5f1 | 2617 | spin_unlock_bh(&css_set_lock); |
74a1166d | 2618 | |
adaae5dc | 2619 | return cgroup_taskset_migrate(&tset, cgrp); |
74a1166d BB |
2620 | } |
2621 | ||
1958d2d5 TH |
2622 | /** |
2623 | * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup | |
2624 | * @dst_cgrp: the cgroup to attach to | |
2625 | * @leader: the task or the leader of the threadgroup to be attached | |
2626 | * @threadgroup: attach the whole threadgroup? | |
2627 | * | |
1ed13287 | 2628 | * Call holding cgroup_mutex and cgroup_threadgroup_rwsem. |
1958d2d5 TH |
2629 | */ |
2630 | static int cgroup_attach_task(struct cgroup *dst_cgrp, | |
2631 | struct task_struct *leader, bool threadgroup) | |
2632 | { | |
2633 | LIST_HEAD(preloaded_csets); | |
2634 | struct task_struct *task; | |
2635 | int ret; | |
2636 | ||
2637 | /* look up all src csets */ | |
f0d9a5f1 | 2638 | spin_lock_bh(&css_set_lock); |
1958d2d5 TH |
2639 | rcu_read_lock(); |
2640 | task = leader; | |
2641 | do { | |
2642 | cgroup_migrate_add_src(task_css_set(task), dst_cgrp, | |
2643 | &preloaded_csets); | |
2644 | if (!threadgroup) | |
2645 | break; | |
2646 | } while_each_thread(leader, task); | |
2647 | rcu_read_unlock(); | |
f0d9a5f1 | 2648 | spin_unlock_bh(&css_set_lock); |
1958d2d5 TH |
2649 | |
2650 | /* prepare dst csets and commit */ | |
2651 | ret = cgroup_migrate_prepare_dst(dst_cgrp, &preloaded_csets); | |
2652 | if (!ret) | |
9af2ec45 | 2653 | ret = cgroup_migrate(leader, threadgroup, dst_cgrp); |
1958d2d5 TH |
2654 | |
2655 | cgroup_migrate_finish(&preloaded_csets); | |
2656 | return ret; | |
74a1166d BB |
2657 | } |
2658 | ||
187fe840 TH |
2659 | static int cgroup_procs_write_permission(struct task_struct *task, |
2660 | struct cgroup *dst_cgrp, | |
2661 | struct kernfs_open_file *of) | |
dedf22e9 TH |
2662 | { |
2663 | const struct cred *cred = current_cred(); | |
2664 | const struct cred *tcred = get_task_cred(task); | |
2665 | int ret = 0; | |
2666 | ||
2667 | /* | |
2668 | * even if we're attaching all tasks in the thread group, we only | |
2669 | * need to check permissions on one of them. | |
2670 | */ | |
2671 | if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) && | |
2672 | !uid_eq(cred->euid, tcred->uid) && | |
2673 | !uid_eq(cred->euid, tcred->suid)) | |
2674 | ret = -EACCES; | |
2675 | ||
187fe840 TH |
2676 | if (!ret && cgroup_on_dfl(dst_cgrp)) { |
2677 | struct super_block *sb = of->file->f_path.dentry->d_sb; | |
2678 | struct cgroup *cgrp; | |
2679 | struct inode *inode; | |
2680 | ||
f0d9a5f1 | 2681 | spin_lock_bh(&css_set_lock); |
187fe840 | 2682 | cgrp = task_cgroup_from_root(task, &cgrp_dfl_root); |
f0d9a5f1 | 2683 | spin_unlock_bh(&css_set_lock); |
187fe840 TH |
2684 | |
2685 | while (!cgroup_is_descendant(dst_cgrp, cgrp)) | |
2686 | cgrp = cgroup_parent(cgrp); | |
2687 | ||
2688 | ret = -ENOMEM; | |
6f60eade | 2689 | inode = kernfs_get_inode(sb, cgrp->procs_file.kn); |
187fe840 TH |
2690 | if (inode) { |
2691 | ret = inode_permission(inode, MAY_WRITE); | |
2692 | iput(inode); | |
2693 | } | |
2694 | } | |
2695 | ||
dedf22e9 TH |
2696 | put_cred(tcred); |
2697 | return ret; | |
2698 | } | |
2699 | ||
74a1166d BB |
2700 | /* |
2701 | * Find the task_struct of the task to attach by vpid and pass it along to the | |
cd3d0952 | 2702 | * function to attach either it or all tasks in its threadgroup. Will lock |
0e1d768f | 2703 | * cgroup_mutex and threadgroup. |
bbcb81d0 | 2704 | */ |
acbef755 TH |
2705 | static ssize_t __cgroup_procs_write(struct kernfs_open_file *of, char *buf, |
2706 | size_t nbytes, loff_t off, bool threadgroup) | |
bbcb81d0 | 2707 | { |
bbcb81d0 | 2708 | struct task_struct *tsk; |
e76ecaee | 2709 | struct cgroup *cgrp; |
acbef755 | 2710 | pid_t pid; |
bbcb81d0 PM |
2711 | int ret; |
2712 | ||
acbef755 TH |
2713 | if (kstrtoint(strstrip(buf), 0, &pid) || pid < 0) |
2714 | return -EINVAL; | |
2715 | ||
945ba199 | 2716 | cgrp = cgroup_kn_lock_live(of->kn, false); |
e76ecaee | 2717 | if (!cgrp) |
74a1166d BB |
2718 | return -ENODEV; |
2719 | ||
3014dde7 | 2720 | percpu_down_write(&cgroup_threadgroup_rwsem); |
b78949eb | 2721 | rcu_read_lock(); |
bbcb81d0 | 2722 | if (pid) { |
73507f33 | 2723 | tsk = find_task_by_vpid(pid); |
74a1166d | 2724 | if (!tsk) { |
dd4b0a46 | 2725 | ret = -ESRCH; |
3014dde7 | 2726 | goto out_unlock_rcu; |
bbcb81d0 | 2727 | } |
dedf22e9 | 2728 | } else { |
b78949eb | 2729 | tsk = current; |
dedf22e9 | 2730 | } |
cd3d0952 TH |
2731 | |
2732 | if (threadgroup) | |
b78949eb | 2733 | tsk = tsk->group_leader; |
c4c27fbd MG |
2734 | |
2735 | /* | |
14a40ffc | 2736 | * Workqueue threads may acquire PF_NO_SETAFFINITY and become |
c4c27fbd MG |
2737 | * trapped in a cpuset, or RT worker may be born in a cgroup |
2738 | * with no rt_runtime allocated. Just say no. | |
2739 | */ | |
14a40ffc | 2740 | if (tsk == kthreadd_task || (tsk->flags & PF_NO_SETAFFINITY)) { |
c4c27fbd | 2741 | ret = -EINVAL; |
3014dde7 | 2742 | goto out_unlock_rcu; |
c4c27fbd MG |
2743 | } |
2744 | ||
b78949eb MSB |
2745 | get_task_struct(tsk); |
2746 | rcu_read_unlock(); | |
2747 | ||
187fe840 | 2748 | ret = cgroup_procs_write_permission(tsk, cgrp, of); |
dedf22e9 TH |
2749 | if (!ret) |
2750 | ret = cgroup_attach_task(cgrp, tsk, threadgroup); | |
081aa458 | 2751 | |
f9f9e7b7 | 2752 | put_task_struct(tsk); |
3014dde7 TH |
2753 | goto out_unlock_threadgroup; |
2754 | ||
2755 | out_unlock_rcu: | |
2756 | rcu_read_unlock(); | |
2757 | out_unlock_threadgroup: | |
2758 | percpu_up_write(&cgroup_threadgroup_rwsem); | |
e76ecaee | 2759 | cgroup_kn_unlock(of->kn); |
e93ad19d | 2760 | cpuset_post_attach_flush(); |
acbef755 | 2761 | return ret ?: nbytes; |
bbcb81d0 PM |
2762 | } |
2763 | ||
7ae1bad9 TH |
2764 | /** |
2765 | * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from' | |
2766 | * @from: attach to all cgroups of a given task | |
2767 | * @tsk: the task to be attached | |
2768 | */ | |
2769 | int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk) | |
2770 | { | |
3dd06ffa | 2771 | struct cgroup_root *root; |
7ae1bad9 TH |
2772 | int retval = 0; |
2773 | ||
47cfcd09 | 2774 | mutex_lock(&cgroup_mutex); |
985ed670 | 2775 | for_each_root(root) { |
96d365e0 TH |
2776 | struct cgroup *from_cgrp; |
2777 | ||
3dd06ffa | 2778 | if (root == &cgrp_dfl_root) |
985ed670 TH |
2779 | continue; |
2780 | ||
f0d9a5f1 | 2781 | spin_lock_bh(&css_set_lock); |
96d365e0 | 2782 | from_cgrp = task_cgroup_from_root(from, root); |
f0d9a5f1 | 2783 | spin_unlock_bh(&css_set_lock); |
7ae1bad9 | 2784 | |
6f4b7e63 | 2785 | retval = cgroup_attach_task(from_cgrp, tsk, false); |
7ae1bad9 TH |
2786 | if (retval) |
2787 | break; | |
2788 | } | |
47cfcd09 | 2789 | mutex_unlock(&cgroup_mutex); |
7ae1bad9 TH |
2790 | |
2791 | return retval; | |
2792 | } | |
2793 | EXPORT_SYMBOL_GPL(cgroup_attach_task_all); | |
2794 | ||
acbef755 TH |
2795 | static ssize_t cgroup_tasks_write(struct kernfs_open_file *of, |
2796 | char *buf, size_t nbytes, loff_t off) | |
74a1166d | 2797 | { |
acbef755 | 2798 | return __cgroup_procs_write(of, buf, nbytes, off, false); |
74a1166d BB |
2799 | } |
2800 | ||
acbef755 TH |
2801 | static ssize_t cgroup_procs_write(struct kernfs_open_file *of, |
2802 | char *buf, size_t nbytes, loff_t off) | |
af351026 | 2803 | { |
acbef755 | 2804 | return __cgroup_procs_write(of, buf, nbytes, off, true); |
af351026 PM |
2805 | } |
2806 | ||
451af504 TH |
2807 | static ssize_t cgroup_release_agent_write(struct kernfs_open_file *of, |
2808 | char *buf, size_t nbytes, loff_t off) | |
e788e066 | 2809 | { |
e76ecaee | 2810 | struct cgroup *cgrp; |
5f469907 | 2811 | |
e76ecaee | 2812 | BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX); |
5f469907 | 2813 | |
945ba199 | 2814 | cgrp = cgroup_kn_lock_live(of->kn, false); |
e76ecaee | 2815 | if (!cgrp) |
e788e066 | 2816 | return -ENODEV; |
69e943b7 | 2817 | spin_lock(&release_agent_path_lock); |
e76ecaee TH |
2818 | strlcpy(cgrp->root->release_agent_path, strstrip(buf), |
2819 | sizeof(cgrp->root->release_agent_path)); | |
69e943b7 | 2820 | spin_unlock(&release_agent_path_lock); |
e76ecaee | 2821 | cgroup_kn_unlock(of->kn); |
451af504 | 2822 | return nbytes; |
e788e066 PM |
2823 | } |
2824 | ||
2da8ca82 | 2825 | static int cgroup_release_agent_show(struct seq_file *seq, void *v) |
e788e066 | 2826 | { |
2da8ca82 | 2827 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
182446d0 | 2828 | |
46cfeb04 | 2829 | spin_lock(&release_agent_path_lock); |
e788e066 | 2830 | seq_puts(seq, cgrp->root->release_agent_path); |
46cfeb04 | 2831 | spin_unlock(&release_agent_path_lock); |
e788e066 | 2832 | seq_putc(seq, '\n'); |
e788e066 PM |
2833 | return 0; |
2834 | } | |
2835 | ||
2da8ca82 | 2836 | static int cgroup_sane_behavior_show(struct seq_file *seq, void *v) |
873fe09e | 2837 | { |
c1d5d42e | 2838 | seq_puts(seq, "0\n"); |
e788e066 PM |
2839 | return 0; |
2840 | } | |
2841 | ||
6e5c8307 | 2842 | static void cgroup_print_ss_mask(struct seq_file *seq, u16 ss_mask) |
355e0c48 | 2843 | { |
f8f22e53 TH |
2844 | struct cgroup_subsys *ss; |
2845 | bool printed = false; | |
2846 | int ssid; | |
a742c59d | 2847 | |
b4e0eeaf | 2848 | do_each_subsys_mask(ss, ssid, ss_mask) { |
a966a4ed AS |
2849 | if (printed) |
2850 | seq_putc(seq, ' '); | |
2851 | seq_printf(seq, "%s", ss->name); | |
2852 | printed = true; | |
b4e0eeaf | 2853 | } while_each_subsys_mask(); |
f8f22e53 TH |
2854 | if (printed) |
2855 | seq_putc(seq, '\n'); | |
355e0c48 PM |
2856 | } |
2857 | ||
f8f22e53 TH |
2858 | /* show controllers which are enabled from the parent */ |
2859 | static int cgroup_controllers_show(struct seq_file *seq, void *v) | |
ddbcc7e8 | 2860 | { |
f8f22e53 TH |
2861 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
2862 | ||
5531dc91 | 2863 | cgroup_print_ss_mask(seq, cgroup_control(cgrp)); |
f8f22e53 | 2864 | return 0; |
ddbcc7e8 PM |
2865 | } |
2866 | ||
f8f22e53 TH |
2867 | /* show controllers which are enabled for a given cgroup's children */ |
2868 | static int cgroup_subtree_control_show(struct seq_file *seq, void *v) | |
ddbcc7e8 | 2869 | { |
f8f22e53 TH |
2870 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
2871 | ||
667c2491 | 2872 | cgroup_print_ss_mask(seq, cgrp->subtree_control); |
f8f22e53 TH |
2873 | return 0; |
2874 | } | |
2875 | ||
2876 | /** | |
2877 | * cgroup_update_dfl_csses - update css assoc of a subtree in default hierarchy | |
2878 | * @cgrp: root of the subtree to update csses for | |
2879 | * | |
54962604 TH |
2880 | * @cgrp's control masks have changed and its subtree's css associations |
2881 | * need to be updated accordingly. This function looks up all css_sets | |
2882 | * which are attached to the subtree, creates the matching updated css_sets | |
2883 | * and migrates the tasks to the new ones. | |
f8f22e53 TH |
2884 | */ |
2885 | static int cgroup_update_dfl_csses(struct cgroup *cgrp) | |
2886 | { | |
2887 | LIST_HEAD(preloaded_csets); | |
10265075 | 2888 | struct cgroup_taskset tset = CGROUP_TASKSET_INIT(tset); |
54962604 TH |
2889 | struct cgroup_subsys_state *d_css; |
2890 | struct cgroup *dsct; | |
f8f22e53 TH |
2891 | struct css_set *src_cset; |
2892 | int ret; | |
2893 | ||
f8f22e53 TH |
2894 | lockdep_assert_held(&cgroup_mutex); |
2895 | ||
3014dde7 TH |
2896 | percpu_down_write(&cgroup_threadgroup_rwsem); |
2897 | ||
f8f22e53 | 2898 | /* look up all csses currently attached to @cgrp's subtree */ |
f0d9a5f1 | 2899 | spin_lock_bh(&css_set_lock); |
54962604 | 2900 | cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) { |
f8f22e53 TH |
2901 | struct cgrp_cset_link *link; |
2902 | ||
54962604 | 2903 | list_for_each_entry(link, &dsct->cset_links, cset_link) |
58cdb1ce | 2904 | cgroup_migrate_add_src(link->cset, dsct, |
f8f22e53 TH |
2905 | &preloaded_csets); |
2906 | } | |
f0d9a5f1 | 2907 | spin_unlock_bh(&css_set_lock); |
f8f22e53 TH |
2908 | |
2909 | /* NULL dst indicates self on default hierarchy */ | |
2910 | ret = cgroup_migrate_prepare_dst(NULL, &preloaded_csets); | |
2911 | if (ret) | |
2912 | goto out_finish; | |
2913 | ||
f0d9a5f1 | 2914 | spin_lock_bh(&css_set_lock); |
f8f22e53 | 2915 | list_for_each_entry(src_cset, &preloaded_csets, mg_preload_node) { |
10265075 | 2916 | struct task_struct *task, *ntask; |
f8f22e53 TH |
2917 | |
2918 | /* src_csets precede dst_csets, break on the first dst_cset */ | |
2919 | if (!src_cset->mg_src_cgrp) | |
2920 | break; | |
2921 | ||
10265075 TH |
2922 | /* all tasks in src_csets need to be migrated */ |
2923 | list_for_each_entry_safe(task, ntask, &src_cset->tasks, cg_list) | |
2924 | cgroup_taskset_add(task, &tset); | |
f8f22e53 | 2925 | } |
f0d9a5f1 | 2926 | spin_unlock_bh(&css_set_lock); |
f8f22e53 | 2927 | |
10265075 | 2928 | ret = cgroup_taskset_migrate(&tset, cgrp); |
f8f22e53 TH |
2929 | out_finish: |
2930 | cgroup_migrate_finish(&preloaded_csets); | |
3014dde7 | 2931 | percpu_up_write(&cgroup_threadgroup_rwsem); |
f8f22e53 TH |
2932 | return ret; |
2933 | } | |
2934 | ||
1b9b96a1 | 2935 | /** |
945ba199 | 2936 | * cgroup_lock_and_drain_offline - lock cgroup_mutex and drain offlined csses |
ce3f1d9d | 2937 | * @cgrp: root of the target subtree |
1b9b96a1 TH |
2938 | * |
2939 | * Because css offlining is asynchronous, userland may try to re-enable a | |
945ba199 TH |
2940 | * controller while the previous css is still around. This function grabs |
2941 | * cgroup_mutex and drains the previous css instances of @cgrp's subtree. | |
1b9b96a1 | 2942 | */ |
945ba199 TH |
2943 | static void cgroup_lock_and_drain_offline(struct cgroup *cgrp) |
2944 | __acquires(&cgroup_mutex) | |
1b9b96a1 TH |
2945 | { |
2946 | struct cgroup *dsct; | |
ce3f1d9d | 2947 | struct cgroup_subsys_state *d_css; |
1b9b96a1 TH |
2948 | struct cgroup_subsys *ss; |
2949 | int ssid; | |
2950 | ||
945ba199 TH |
2951 | restart: |
2952 | mutex_lock(&cgroup_mutex); | |
1b9b96a1 | 2953 | |
ce3f1d9d | 2954 | cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) { |
1b9b96a1 TH |
2955 | for_each_subsys(ss, ssid) { |
2956 | struct cgroup_subsys_state *css = cgroup_css(dsct, ss); | |
2957 | DEFINE_WAIT(wait); | |
2958 | ||
ce3f1d9d | 2959 | if (!css || !percpu_ref_is_dying(&css->refcnt)) |
1b9b96a1 TH |
2960 | continue; |
2961 | ||
2962 | cgroup_get(dsct); | |
2963 | prepare_to_wait(&dsct->offline_waitq, &wait, | |
2964 | TASK_UNINTERRUPTIBLE); | |
2965 | ||
2966 | mutex_unlock(&cgroup_mutex); | |
2967 | schedule(); | |
2968 | finish_wait(&dsct->offline_waitq, &wait); | |
1b9b96a1 TH |
2969 | |
2970 | cgroup_put(dsct); | |
945ba199 | 2971 | goto restart; |
1b9b96a1 TH |
2972 | } |
2973 | } | |
1b9b96a1 TH |
2974 | } |
2975 | ||
15a27c36 TH |
2976 | /** |
2977 | * cgroup_save_control - save control masks of a subtree | |
2978 | * @cgrp: root of the target subtree | |
2979 | * | |
2980 | * Save ->subtree_control and ->subtree_ss_mask to the respective old_ | |
2981 | * prefixed fields for @cgrp's subtree including @cgrp itself. | |
2982 | */ | |
2983 | static void cgroup_save_control(struct cgroup *cgrp) | |
2984 | { | |
2985 | struct cgroup *dsct; | |
2986 | struct cgroup_subsys_state *d_css; | |
2987 | ||
2988 | cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) { | |
2989 | dsct->old_subtree_control = dsct->subtree_control; | |
2990 | dsct->old_subtree_ss_mask = dsct->subtree_ss_mask; | |
2991 | } | |
2992 | } | |
2993 | ||
2994 | /** | |
2995 | * cgroup_propagate_control - refresh control masks of a subtree | |
2996 | * @cgrp: root of the target subtree | |
2997 | * | |
2998 | * For @cgrp and its subtree, ensure ->subtree_ss_mask matches | |
2999 | * ->subtree_control and propagate controller availability through the | |
3000 | * subtree so that descendants don't have unavailable controllers enabled. | |
3001 | */ | |
3002 | static void cgroup_propagate_control(struct cgroup *cgrp) | |
3003 | { | |
3004 | struct cgroup *dsct; | |
3005 | struct cgroup_subsys_state *d_css; | |
3006 | ||
3007 | cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) { | |
3008 | dsct->subtree_control &= cgroup_control(dsct); | |
5ced2518 TH |
3009 | dsct->subtree_ss_mask = |
3010 | cgroup_calc_subtree_ss_mask(dsct->subtree_control, | |
3011 | cgroup_ss_mask(dsct)); | |
15a27c36 TH |
3012 | } |
3013 | } | |
3014 | ||
3015 | /** | |
3016 | * cgroup_restore_control - restore control masks of a subtree | |
3017 | * @cgrp: root of the target subtree | |
3018 | * | |
3019 | * Restore ->subtree_control and ->subtree_ss_mask from the respective old_ | |
3020 | * prefixed fields for @cgrp's subtree including @cgrp itself. | |
3021 | */ | |
3022 | static void cgroup_restore_control(struct cgroup *cgrp) | |
3023 | { | |
3024 | struct cgroup *dsct; | |
3025 | struct cgroup_subsys_state *d_css; | |
3026 | ||
3027 | cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) { | |
3028 | dsct->subtree_control = dsct->old_subtree_control; | |
3029 | dsct->subtree_ss_mask = dsct->old_subtree_ss_mask; | |
3030 | } | |
3031 | } | |
3032 | ||
bdb53bd7 TH |
3033 | /** |
3034 | * cgroup_apply_control_enable - enable or show csses according to control | |
ce3f1d9d | 3035 | * @cgrp: root of the target subtree |
bdb53bd7 | 3036 | * |
ce3f1d9d | 3037 | * Walk @cgrp's subtree and create new csses or make the existing ones |
bdb53bd7 TH |
3038 | * visible. A css is created invisible if it's being implicitly enabled |
3039 | * through dependency. An invisible css is made visible when the userland | |
3040 | * explicitly enables it. | |
3041 | * | |
3042 | * Returns 0 on success, -errno on failure. On failure, csses which have | |
3043 | * been processed already aren't cleaned up. The caller is responsible for | |
3044 | * cleaning up with cgroup_apply_control_disble(). | |
3045 | */ | |
3046 | static int cgroup_apply_control_enable(struct cgroup *cgrp) | |
3047 | { | |
3048 | struct cgroup *dsct; | |
ce3f1d9d | 3049 | struct cgroup_subsys_state *d_css; |
bdb53bd7 TH |
3050 | struct cgroup_subsys *ss; |
3051 | int ssid, ret; | |
3052 | ||
ce3f1d9d | 3053 | cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) { |
bdb53bd7 TH |
3054 | for_each_subsys(ss, ssid) { |
3055 | struct cgroup_subsys_state *css = cgroup_css(dsct, ss); | |
3056 | ||
945ba199 TH |
3057 | WARN_ON_ONCE(css && percpu_ref_is_dying(&css->refcnt)); |
3058 | ||
bdb53bd7 TH |
3059 | if (!(cgroup_ss_mask(dsct) & (1 << ss->id))) |
3060 | continue; | |
3061 | ||
3062 | if (!css) { | |
3063 | css = css_create(dsct, ss); | |
3064 | if (IS_ERR(css)) | |
3065 | return PTR_ERR(css); | |
3066 | } | |
3067 | ||
3068 | if (cgroup_control(dsct) & (1 << ss->id)) { | |
334c3679 | 3069 | ret = css_populate_dir(css); |
bdb53bd7 TH |
3070 | if (ret) |
3071 | return ret; | |
3072 | } | |
3073 | } | |
3074 | } | |
3075 | ||
3076 | return 0; | |
3077 | } | |
3078 | ||
12b3bb6a TH |
3079 | /** |
3080 | * cgroup_apply_control_disable - kill or hide csses according to control | |
ce3f1d9d | 3081 | * @cgrp: root of the target subtree |
12b3bb6a | 3082 | * |
ce3f1d9d | 3083 | * Walk @cgrp's subtree and kill and hide csses so that they match |
12b3bb6a TH |
3084 | * cgroup_ss_mask() and cgroup_visible_mask(). |
3085 | * | |
3086 | * A css is hidden when the userland requests it to be disabled while other | |
3087 | * subsystems are still depending on it. The css must not actively control | |
3088 | * resources and be in the vanilla state if it's made visible again later. | |
3089 | * Controllers which may be depended upon should provide ->css_reset() for | |
3090 | * this purpose. | |
3091 | */ | |
3092 | static void cgroup_apply_control_disable(struct cgroup *cgrp) | |
3093 | { | |
3094 | struct cgroup *dsct; | |
ce3f1d9d | 3095 | struct cgroup_subsys_state *d_css; |
12b3bb6a TH |
3096 | struct cgroup_subsys *ss; |
3097 | int ssid; | |
3098 | ||
ce3f1d9d | 3099 | cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) { |
12b3bb6a TH |
3100 | for_each_subsys(ss, ssid) { |
3101 | struct cgroup_subsys_state *css = cgroup_css(dsct, ss); | |
3102 | ||
945ba199 TH |
3103 | WARN_ON_ONCE(css && percpu_ref_is_dying(&css->refcnt)); |
3104 | ||
12b3bb6a TH |
3105 | if (!css) |
3106 | continue; | |
3107 | ||
334c3679 TH |
3108 | if (css->parent && |
3109 | !(cgroup_ss_mask(dsct) & (1 << ss->id))) { | |
12b3bb6a TH |
3110 | kill_css(css); |
3111 | } else if (!(cgroup_control(dsct) & (1 << ss->id))) { | |
334c3679 | 3112 | css_clear_dir(css); |
12b3bb6a TH |
3113 | if (ss->css_reset) |
3114 | ss->css_reset(css); | |
3115 | } | |
3116 | } | |
3117 | } | |
3118 | } | |
3119 | ||
f7b2814b TH |
3120 | /** |
3121 | * cgroup_apply_control - apply control mask updates to the subtree | |
3122 | * @cgrp: root of the target subtree | |
3123 | * | |
3124 | * subsystems can be enabled and disabled in a subtree using the following | |
3125 | * steps. | |
3126 | * | |
3127 | * 1. Call cgroup_save_control() to stash the current state. | |
3128 | * 2. Update ->subtree_control masks in the subtree as desired. | |
3129 | * 3. Call cgroup_apply_control() to apply the changes. | |
3130 | * 4. Optionally perform other related operations. | |
3131 | * 5. Call cgroup_finalize_control() to finish up. | |
3132 | * | |
3133 | * This function implements step 3 and propagates the mask changes | |
3134 | * throughout @cgrp's subtree, updates csses accordingly and perform | |
3135 | * process migrations. | |
3136 | */ | |
3137 | static int cgroup_apply_control(struct cgroup *cgrp) | |
3138 | { | |
3139 | int ret; | |
3140 | ||
3141 | cgroup_propagate_control(cgrp); | |
3142 | ||
3143 | ret = cgroup_apply_control_enable(cgrp); | |
3144 | if (ret) | |
3145 | return ret; | |
3146 | ||
3147 | /* | |
3148 | * At this point, cgroup_e_css() results reflect the new csses | |
3149 | * making the following cgroup_update_dfl_csses() properly update | |
3150 | * css associations of all tasks in the subtree. | |
3151 | */ | |
3152 | ret = cgroup_update_dfl_csses(cgrp); | |
3153 | if (ret) | |
3154 | return ret; | |
3155 | ||
3156 | return 0; | |
3157 | } | |
3158 | ||
3159 | /** | |
3160 | * cgroup_finalize_control - finalize control mask update | |
3161 | * @cgrp: root of the target subtree | |
3162 | * @ret: the result of the update | |
3163 | * | |
3164 | * Finalize control mask update. See cgroup_apply_control() for more info. | |
3165 | */ | |
3166 | static void cgroup_finalize_control(struct cgroup *cgrp, int ret) | |
3167 | { | |
3168 | if (ret) { | |
3169 | cgroup_restore_control(cgrp); | |
3170 | cgroup_propagate_control(cgrp); | |
3171 | } | |
3172 | ||
3173 | cgroup_apply_control_disable(cgrp); | |
3174 | } | |
3175 | ||
f8f22e53 | 3176 | /* change the enabled child controllers for a cgroup in the default hierarchy */ |
451af504 TH |
3177 | static ssize_t cgroup_subtree_control_write(struct kernfs_open_file *of, |
3178 | char *buf, size_t nbytes, | |
3179 | loff_t off) | |
f8f22e53 | 3180 | { |
6e5c8307 | 3181 | u16 enable = 0, disable = 0; |
a9746d8d | 3182 | struct cgroup *cgrp, *child; |
f8f22e53 | 3183 | struct cgroup_subsys *ss; |
451af504 | 3184 | char *tok; |
f8f22e53 TH |
3185 | int ssid, ret; |
3186 | ||
3187 | /* | |
d37167ab TH |
3188 | * Parse input - space separated list of subsystem names prefixed |
3189 | * with either + or -. | |
f8f22e53 | 3190 | */ |
451af504 TH |
3191 | buf = strstrip(buf); |
3192 | while ((tok = strsep(&buf, " "))) { | |
d37167ab TH |
3193 | if (tok[0] == '\0') |
3194 | continue; | |
a7165264 | 3195 | do_each_subsys_mask(ss, ssid, ~cgrp_dfl_inhibit_ss_mask) { |
fc5ed1e9 TH |
3196 | if (!cgroup_ssid_enabled(ssid) || |
3197 | strcmp(tok + 1, ss->name)) | |
f8f22e53 TH |
3198 | continue; |
3199 | ||
3200 | if (*tok == '+') { | |
7d331fa9 TH |
3201 | enable |= 1 << ssid; |
3202 | disable &= ~(1 << ssid); | |
f8f22e53 | 3203 | } else if (*tok == '-') { |
7d331fa9 TH |
3204 | disable |= 1 << ssid; |
3205 | enable &= ~(1 << ssid); | |
f8f22e53 TH |
3206 | } else { |
3207 | return -EINVAL; | |
3208 | } | |
3209 | break; | |
b4e0eeaf | 3210 | } while_each_subsys_mask(); |
f8f22e53 TH |
3211 | if (ssid == CGROUP_SUBSYS_COUNT) |
3212 | return -EINVAL; | |
3213 | } | |
3214 | ||
945ba199 | 3215 | cgrp = cgroup_kn_lock_live(of->kn, true); |
a9746d8d TH |
3216 | if (!cgrp) |
3217 | return -ENODEV; | |
f8f22e53 TH |
3218 | |
3219 | for_each_subsys(ss, ssid) { | |
3220 | if (enable & (1 << ssid)) { | |
667c2491 | 3221 | if (cgrp->subtree_control & (1 << ssid)) { |
f8f22e53 TH |
3222 | enable &= ~(1 << ssid); |
3223 | continue; | |
3224 | } | |
3225 | ||
5531dc91 | 3226 | if (!(cgroup_control(cgrp) & (1 << ssid))) { |
c29adf24 TH |
3227 | ret = -ENOENT; |
3228 | goto out_unlock; | |
3229 | } | |
f8f22e53 | 3230 | } else if (disable & (1 << ssid)) { |
667c2491 | 3231 | if (!(cgrp->subtree_control & (1 << ssid))) { |
f8f22e53 TH |
3232 | disable &= ~(1 << ssid); |
3233 | continue; | |
3234 | } | |
3235 | ||
3236 | /* a child has it enabled? */ | |
3237 | cgroup_for_each_live_child(child, cgrp) { | |
667c2491 | 3238 | if (child->subtree_control & (1 << ssid)) { |
f8f22e53 | 3239 | ret = -EBUSY; |
ddab2b6e | 3240 | goto out_unlock; |
f8f22e53 TH |
3241 | } |
3242 | } | |
3243 | } | |
3244 | } | |
3245 | ||
3246 | if (!enable && !disable) { | |
3247 | ret = 0; | |
ddab2b6e | 3248 | goto out_unlock; |
f8f22e53 TH |
3249 | } |
3250 | ||
3251 | /* | |
667c2491 | 3252 | * Except for the root, subtree_control must be zero for a cgroup |
f8f22e53 TH |
3253 | * with tasks so that child cgroups don't compete against tasks. |
3254 | */ | |
d51f39b0 | 3255 | if (enable && cgroup_parent(cgrp) && !list_empty(&cgrp->cset_links)) { |
f8f22e53 TH |
3256 | ret = -EBUSY; |
3257 | goto out_unlock; | |
3258 | } | |
3259 | ||
15a27c36 TH |
3260 | /* save and update control masks and prepare csses */ |
3261 | cgroup_save_control(cgrp); | |
f63070d3 | 3262 | |
15a27c36 TH |
3263 | cgrp->subtree_control |= enable; |
3264 | cgrp->subtree_control &= ~disable; | |
c29adf24 | 3265 | |
f7b2814b | 3266 | ret = cgroup_apply_control(cgrp); |
f8f22e53 | 3267 | |
f7b2814b | 3268 | cgroup_finalize_control(cgrp, ret); |
f8f22e53 TH |
3269 | |
3270 | kernfs_activate(cgrp->kn); | |
3271 | ret = 0; | |
3272 | out_unlock: | |
a9746d8d | 3273 | cgroup_kn_unlock(of->kn); |
451af504 | 3274 | return ret ?: nbytes; |
f8f22e53 TH |
3275 | } |
3276 | ||
4a07c222 | 3277 | static int cgroup_events_show(struct seq_file *seq, void *v) |
842b597e | 3278 | { |
4a07c222 | 3279 | seq_printf(seq, "populated %d\n", |
27bd4dbb | 3280 | cgroup_is_populated(seq_css(seq)->cgroup)); |
842b597e TH |
3281 | return 0; |
3282 | } | |
3283 | ||
2bd59d48 TH |
3284 | static ssize_t cgroup_file_write(struct kernfs_open_file *of, char *buf, |
3285 | size_t nbytes, loff_t off) | |
355e0c48 | 3286 | { |
2bd59d48 TH |
3287 | struct cgroup *cgrp = of->kn->parent->priv; |
3288 | struct cftype *cft = of->kn->priv; | |
3289 | struct cgroup_subsys_state *css; | |
a742c59d | 3290 | int ret; |
355e0c48 | 3291 | |
b4168640 TH |
3292 | if (cft->write) |
3293 | return cft->write(of, buf, nbytes, off); | |
3294 | ||
2bd59d48 TH |
3295 | /* |
3296 | * kernfs guarantees that a file isn't deleted with operations in | |
3297 | * flight, which means that the matching css is and stays alive and | |
3298 | * doesn't need to be pinned. The RCU locking is not necessary | |
3299 | * either. It's just for the convenience of using cgroup_css(). | |
3300 | */ | |
3301 | rcu_read_lock(); | |
3302 | css = cgroup_css(cgrp, cft->ss); | |
3303 | rcu_read_unlock(); | |
a742c59d | 3304 | |
451af504 | 3305 | if (cft->write_u64) { |
a742c59d TH |
3306 | unsigned long long v; |
3307 | ret = kstrtoull(buf, 0, &v); | |
3308 | if (!ret) | |
3309 | ret = cft->write_u64(css, cft, v); | |
3310 | } else if (cft->write_s64) { | |
3311 | long long v; | |
3312 | ret = kstrtoll(buf, 0, &v); | |
3313 | if (!ret) | |
3314 | ret = cft->write_s64(css, cft, v); | |
e73d2c61 | 3315 | } else { |
a742c59d | 3316 | ret = -EINVAL; |
e73d2c61 | 3317 | } |
2bd59d48 | 3318 | |
a742c59d | 3319 | return ret ?: nbytes; |
355e0c48 PM |
3320 | } |
3321 | ||
6612f05b | 3322 | static void *cgroup_seqfile_start(struct seq_file *seq, loff_t *ppos) |
db3b1497 | 3323 | { |
2bd59d48 | 3324 | return seq_cft(seq)->seq_start(seq, ppos); |
db3b1497 PM |
3325 | } |
3326 | ||
6612f05b | 3327 | static void *cgroup_seqfile_next(struct seq_file *seq, void *v, loff_t *ppos) |
ddbcc7e8 | 3328 | { |
2bd59d48 | 3329 | return seq_cft(seq)->seq_next(seq, v, ppos); |
ddbcc7e8 PM |
3330 | } |
3331 | ||
6612f05b | 3332 | static void cgroup_seqfile_stop(struct seq_file *seq, void *v) |
ddbcc7e8 | 3333 | { |
2bd59d48 | 3334 | seq_cft(seq)->seq_stop(seq, v); |
ddbcc7e8 PM |
3335 | } |
3336 | ||
91796569 | 3337 | static int cgroup_seqfile_show(struct seq_file *m, void *arg) |
e73d2c61 | 3338 | { |
7da11279 TH |
3339 | struct cftype *cft = seq_cft(m); |
3340 | struct cgroup_subsys_state *css = seq_css(m); | |
e73d2c61 | 3341 | |
2da8ca82 TH |
3342 | if (cft->seq_show) |
3343 | return cft->seq_show(m, arg); | |
e73d2c61 | 3344 | |
f4c753b7 | 3345 | if (cft->read_u64) |
896f5199 TH |
3346 | seq_printf(m, "%llu\n", cft->read_u64(css, cft)); |
3347 | else if (cft->read_s64) | |
3348 | seq_printf(m, "%lld\n", cft->read_s64(css, cft)); | |
3349 | else | |
3350 | return -EINVAL; | |
3351 | return 0; | |
91796569 PM |
3352 | } |
3353 | ||
2bd59d48 TH |
3354 | static struct kernfs_ops cgroup_kf_single_ops = { |
3355 | .atomic_write_len = PAGE_SIZE, | |
3356 | .write = cgroup_file_write, | |
3357 | .seq_show = cgroup_seqfile_show, | |
91796569 PM |
3358 | }; |
3359 | ||
2bd59d48 TH |
3360 | static struct kernfs_ops cgroup_kf_ops = { |
3361 | .atomic_write_len = PAGE_SIZE, | |
3362 | .write = cgroup_file_write, | |
3363 | .seq_start = cgroup_seqfile_start, | |
3364 | .seq_next = cgroup_seqfile_next, | |
3365 | .seq_stop = cgroup_seqfile_stop, | |
3366 | .seq_show = cgroup_seqfile_show, | |
3367 | }; | |
ddbcc7e8 PM |
3368 | |
3369 | /* | |
3370 | * cgroup_rename - Only allow simple rename of directories in place. | |
3371 | */ | |
2bd59d48 TH |
3372 | static int cgroup_rename(struct kernfs_node *kn, struct kernfs_node *new_parent, |
3373 | const char *new_name_str) | |
ddbcc7e8 | 3374 | { |
2bd59d48 | 3375 | struct cgroup *cgrp = kn->priv; |
65dff759 | 3376 | int ret; |
65dff759 | 3377 | |
2bd59d48 | 3378 | if (kernfs_type(kn) != KERNFS_DIR) |
ddbcc7e8 | 3379 | return -ENOTDIR; |
2bd59d48 | 3380 | if (kn->parent != new_parent) |
ddbcc7e8 | 3381 | return -EIO; |
65dff759 | 3382 | |
6db8e85c TH |
3383 | /* |
3384 | * This isn't a proper migration and its usefulness is very | |
aa6ec29b | 3385 | * limited. Disallow on the default hierarchy. |
6db8e85c | 3386 | */ |
aa6ec29b | 3387 | if (cgroup_on_dfl(cgrp)) |
6db8e85c | 3388 | return -EPERM; |
099fca32 | 3389 | |
e1b2dc17 | 3390 | /* |
8353da1f | 3391 | * We're gonna grab cgroup_mutex which nests outside kernfs |
e1b2dc17 | 3392 | * active_ref. kernfs_rename() doesn't require active_ref |
8353da1f | 3393 | * protection. Break them before grabbing cgroup_mutex. |
e1b2dc17 TH |
3394 | */ |
3395 | kernfs_break_active_protection(new_parent); | |
3396 | kernfs_break_active_protection(kn); | |
099fca32 | 3397 | |
2bd59d48 | 3398 | mutex_lock(&cgroup_mutex); |
099fca32 | 3399 | |
2bd59d48 | 3400 | ret = kernfs_rename(kn, new_parent, new_name_str); |
099fca32 | 3401 | |
2bd59d48 | 3402 | mutex_unlock(&cgroup_mutex); |
e1b2dc17 TH |
3403 | |
3404 | kernfs_unbreak_active_protection(kn); | |
3405 | kernfs_unbreak_active_protection(new_parent); | |
2bd59d48 | 3406 | return ret; |
099fca32 LZ |
3407 | } |
3408 | ||
49957f8e TH |
3409 | /* set uid and gid of cgroup dirs and files to that of the creator */ |
3410 | static int cgroup_kn_set_ugid(struct kernfs_node *kn) | |
3411 | { | |
3412 | struct iattr iattr = { .ia_valid = ATTR_UID | ATTR_GID, | |
3413 | .ia_uid = current_fsuid(), | |
3414 | .ia_gid = current_fsgid(), }; | |
3415 | ||
3416 | if (uid_eq(iattr.ia_uid, GLOBAL_ROOT_UID) && | |
3417 | gid_eq(iattr.ia_gid, GLOBAL_ROOT_GID)) | |
3418 | return 0; | |
3419 | ||
3420 | return kernfs_setattr(kn, &iattr); | |
3421 | } | |
3422 | ||
4df8dc90 TH |
3423 | static int cgroup_add_file(struct cgroup_subsys_state *css, struct cgroup *cgrp, |
3424 | struct cftype *cft) | |
ddbcc7e8 | 3425 | { |
8d7e6fb0 | 3426 | char name[CGROUP_FILE_NAME_MAX]; |
2bd59d48 TH |
3427 | struct kernfs_node *kn; |
3428 | struct lock_class_key *key = NULL; | |
49957f8e | 3429 | int ret; |
05ef1d7c | 3430 | |
2bd59d48 TH |
3431 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
3432 | key = &cft->lockdep_key; | |
3433 | #endif | |
3434 | kn = __kernfs_create_file(cgrp->kn, cgroup_file_name(cgrp, cft, name), | |
3435 | cgroup_file_mode(cft), 0, cft->kf_ops, cft, | |
dfeb0750 | 3436 | NULL, key); |
49957f8e TH |
3437 | if (IS_ERR(kn)) |
3438 | return PTR_ERR(kn); | |
3439 | ||
3440 | ret = cgroup_kn_set_ugid(kn); | |
f8f22e53 | 3441 | if (ret) { |
49957f8e | 3442 | kernfs_remove(kn); |
f8f22e53 TH |
3443 | return ret; |
3444 | } | |
3445 | ||
6f60eade TH |
3446 | if (cft->file_offset) { |
3447 | struct cgroup_file *cfile = (void *)css + cft->file_offset; | |
3448 | ||
34c06254 | 3449 | spin_lock_irq(&cgroup_file_kn_lock); |
6f60eade | 3450 | cfile->kn = kn; |
34c06254 | 3451 | spin_unlock_irq(&cgroup_file_kn_lock); |
6f60eade TH |
3452 | } |
3453 | ||
f8f22e53 | 3454 | return 0; |
ddbcc7e8 PM |
3455 | } |
3456 | ||
b1f28d31 TH |
3457 | /** |
3458 | * cgroup_addrm_files - add or remove files to a cgroup directory | |
4df8dc90 TH |
3459 | * @css: the target css |
3460 | * @cgrp: the target cgroup (usually css->cgroup) | |
b1f28d31 TH |
3461 | * @cfts: array of cftypes to be added |
3462 | * @is_add: whether to add or remove | |
3463 | * | |
3464 | * Depending on @is_add, add or remove files defined by @cfts on @cgrp. | |
6732ed85 | 3465 | * For removals, this function never fails. |
b1f28d31 | 3466 | */ |
4df8dc90 TH |
3467 | static int cgroup_addrm_files(struct cgroup_subsys_state *css, |
3468 | struct cgroup *cgrp, struct cftype cfts[], | |
2bb566cb | 3469 | bool is_add) |
ddbcc7e8 | 3470 | { |
6732ed85 | 3471 | struct cftype *cft, *cft_end = NULL; |
b598dde3 | 3472 | int ret = 0; |
b1f28d31 | 3473 | |
01f6474c | 3474 | lockdep_assert_held(&cgroup_mutex); |
db0416b6 | 3475 | |
6732ed85 TH |
3476 | restart: |
3477 | for (cft = cfts; cft != cft_end && cft->name[0] != '\0'; cft++) { | |
f33fddc2 | 3478 | /* does cft->flags tell us to skip this file on @cgrp? */ |
05ebb6e6 | 3479 | if ((cft->flags & __CFTYPE_ONLY_ON_DFL) && !cgroup_on_dfl(cgrp)) |
8cbbf2c9 | 3480 | continue; |
05ebb6e6 | 3481 | if ((cft->flags & __CFTYPE_NOT_ON_DFL) && cgroup_on_dfl(cgrp)) |
873fe09e | 3482 | continue; |
d51f39b0 | 3483 | if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgroup_parent(cgrp)) |
f33fddc2 | 3484 | continue; |
d51f39b0 | 3485 | if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgroup_parent(cgrp)) |
f33fddc2 G |
3486 | continue; |
3487 | ||
2739d3cc | 3488 | if (is_add) { |
4df8dc90 | 3489 | ret = cgroup_add_file(css, cgrp, cft); |
b1f28d31 | 3490 | if (ret) { |
ed3d261b JP |
3491 | pr_warn("%s: failed to add %s, err=%d\n", |
3492 | __func__, cft->name, ret); | |
6732ed85 TH |
3493 | cft_end = cft; |
3494 | is_add = false; | |
3495 | goto restart; | |
b1f28d31 | 3496 | } |
2739d3cc LZ |
3497 | } else { |
3498 | cgroup_rm_file(cgrp, cft); | |
db0416b6 | 3499 | } |
ddbcc7e8 | 3500 | } |
b598dde3 | 3501 | return ret; |
ddbcc7e8 PM |
3502 | } |
3503 | ||
21a2d343 | 3504 | static int cgroup_apply_cftypes(struct cftype *cfts, bool is_add) |
8e3f6541 TH |
3505 | { |
3506 | LIST_HEAD(pending); | |
2bb566cb | 3507 | struct cgroup_subsys *ss = cfts[0].ss; |
3dd06ffa | 3508 | struct cgroup *root = &ss->root->cgrp; |
492eb21b | 3509 | struct cgroup_subsys_state *css; |
9ccece80 | 3510 | int ret = 0; |
8e3f6541 | 3511 | |
01f6474c | 3512 | lockdep_assert_held(&cgroup_mutex); |
e8c82d20 | 3513 | |
e8c82d20 | 3514 | /* add/rm files for all cgroups created before */ |
ca8bdcaf | 3515 | css_for_each_descendant_pre(css, cgroup_css(root, ss)) { |
492eb21b TH |
3516 | struct cgroup *cgrp = css->cgroup; |
3517 | ||
88cb04b9 | 3518 | if (!(css->flags & CSS_VISIBLE)) |
e8c82d20 LZ |
3519 | continue; |
3520 | ||
4df8dc90 | 3521 | ret = cgroup_addrm_files(css, cgrp, cfts, is_add); |
9ccece80 TH |
3522 | if (ret) |
3523 | break; | |
8e3f6541 | 3524 | } |
21a2d343 TH |
3525 | |
3526 | if (is_add && !ret) | |
3527 | kernfs_activate(root->kn); | |
9ccece80 | 3528 | return ret; |
8e3f6541 TH |
3529 | } |
3530 | ||
2da440a2 | 3531 | static void cgroup_exit_cftypes(struct cftype *cfts) |
8e3f6541 | 3532 | { |
2bb566cb | 3533 | struct cftype *cft; |
8e3f6541 | 3534 | |
2bd59d48 TH |
3535 | for (cft = cfts; cft->name[0] != '\0'; cft++) { |
3536 | /* free copy for custom atomic_write_len, see init_cftypes() */ | |
3537 | if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) | |
3538 | kfree(cft->kf_ops); | |
3539 | cft->kf_ops = NULL; | |
2da440a2 | 3540 | cft->ss = NULL; |
a8ddc821 TH |
3541 | |
3542 | /* revert flags set by cgroup core while adding @cfts */ | |
05ebb6e6 | 3543 | cft->flags &= ~(__CFTYPE_ONLY_ON_DFL | __CFTYPE_NOT_ON_DFL); |
2bd59d48 | 3544 | } |
2da440a2 TH |
3545 | } |
3546 | ||
2bd59d48 | 3547 | static int cgroup_init_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
2da440a2 TH |
3548 | { |
3549 | struct cftype *cft; | |
3550 | ||
2bd59d48 TH |
3551 | for (cft = cfts; cft->name[0] != '\0'; cft++) { |
3552 | struct kernfs_ops *kf_ops; | |
3553 | ||
0adb0704 TH |
3554 | WARN_ON(cft->ss || cft->kf_ops); |
3555 | ||
2bd59d48 TH |
3556 | if (cft->seq_start) |
3557 | kf_ops = &cgroup_kf_ops; | |
3558 | else | |
3559 | kf_ops = &cgroup_kf_single_ops; | |
3560 | ||
3561 | /* | |
3562 | * Ugh... if @cft wants a custom max_write_len, we need to | |
3563 | * make a copy of kf_ops to set its atomic_write_len. | |
3564 | */ | |
3565 | if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) { | |
3566 | kf_ops = kmemdup(kf_ops, sizeof(*kf_ops), GFP_KERNEL); | |
3567 | if (!kf_ops) { | |
3568 | cgroup_exit_cftypes(cfts); | |
3569 | return -ENOMEM; | |
3570 | } | |
3571 | kf_ops->atomic_write_len = cft->max_write_len; | |
3572 | } | |
8e3f6541 | 3573 | |
2bd59d48 | 3574 | cft->kf_ops = kf_ops; |
2bb566cb | 3575 | cft->ss = ss; |
2bd59d48 | 3576 | } |
2bb566cb | 3577 | |
2bd59d48 | 3578 | return 0; |
2da440a2 TH |
3579 | } |
3580 | ||
21a2d343 TH |
3581 | static int cgroup_rm_cftypes_locked(struct cftype *cfts) |
3582 | { | |
01f6474c | 3583 | lockdep_assert_held(&cgroup_mutex); |
21a2d343 TH |
3584 | |
3585 | if (!cfts || !cfts[0].ss) | |
3586 | return -ENOENT; | |
3587 | ||
3588 | list_del(&cfts->node); | |
3589 | cgroup_apply_cftypes(cfts, false); | |
3590 | cgroup_exit_cftypes(cfts); | |
3591 | return 0; | |
8e3f6541 | 3592 | } |
8e3f6541 | 3593 | |
79578621 TH |
3594 | /** |
3595 | * cgroup_rm_cftypes - remove an array of cftypes from a subsystem | |
79578621 TH |
3596 | * @cfts: zero-length name terminated array of cftypes |
3597 | * | |
2bb566cb TH |
3598 | * Unregister @cfts. Files described by @cfts are removed from all |
3599 | * existing cgroups and all future cgroups won't have them either. This | |
3600 | * function can be called anytime whether @cfts' subsys is attached or not. | |
79578621 TH |
3601 | * |
3602 | * Returns 0 on successful unregistration, -ENOENT if @cfts is not | |
2bb566cb | 3603 | * registered. |
79578621 | 3604 | */ |
2bb566cb | 3605 | int cgroup_rm_cftypes(struct cftype *cfts) |
79578621 | 3606 | { |
21a2d343 | 3607 | int ret; |
79578621 | 3608 | |
01f6474c | 3609 | mutex_lock(&cgroup_mutex); |
21a2d343 | 3610 | ret = cgroup_rm_cftypes_locked(cfts); |
01f6474c | 3611 | mutex_unlock(&cgroup_mutex); |
21a2d343 | 3612 | return ret; |
80b13586 TH |
3613 | } |
3614 | ||
8e3f6541 TH |
3615 | /** |
3616 | * cgroup_add_cftypes - add an array of cftypes to a subsystem | |
3617 | * @ss: target cgroup subsystem | |
3618 | * @cfts: zero-length name terminated array of cftypes | |
3619 | * | |
3620 | * Register @cfts to @ss. Files described by @cfts are created for all | |
3621 | * existing cgroups to which @ss is attached and all future cgroups will | |
3622 | * have them too. This function can be called anytime whether @ss is | |
3623 | * attached or not. | |
3624 | * | |
3625 | * Returns 0 on successful registration, -errno on failure. Note that this | |
3626 | * function currently returns 0 as long as @cfts registration is successful | |
3627 | * even if some file creation attempts on existing cgroups fail. | |
3628 | */ | |
2cf669a5 | 3629 | static int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
8e3f6541 | 3630 | { |
9ccece80 | 3631 | int ret; |
8e3f6541 | 3632 | |
fc5ed1e9 | 3633 | if (!cgroup_ssid_enabled(ss->id)) |
c731ae1d LZ |
3634 | return 0; |
3635 | ||
dc5736ed LZ |
3636 | if (!cfts || cfts[0].name[0] == '\0') |
3637 | return 0; | |
2bb566cb | 3638 | |
2bd59d48 TH |
3639 | ret = cgroup_init_cftypes(ss, cfts); |
3640 | if (ret) | |
3641 | return ret; | |
79578621 | 3642 | |
01f6474c | 3643 | mutex_lock(&cgroup_mutex); |
21a2d343 | 3644 | |
0adb0704 | 3645 | list_add_tail(&cfts->node, &ss->cfts); |
21a2d343 | 3646 | ret = cgroup_apply_cftypes(cfts, true); |
9ccece80 | 3647 | if (ret) |
21a2d343 | 3648 | cgroup_rm_cftypes_locked(cfts); |
79578621 | 3649 | |
01f6474c | 3650 | mutex_unlock(&cgroup_mutex); |
9ccece80 | 3651 | return ret; |
79578621 TH |
3652 | } |
3653 | ||
a8ddc821 TH |
3654 | /** |
3655 | * cgroup_add_dfl_cftypes - add an array of cftypes for default hierarchy | |
3656 | * @ss: target cgroup subsystem | |
3657 | * @cfts: zero-length name terminated array of cftypes | |
3658 | * | |
3659 | * Similar to cgroup_add_cftypes() but the added files are only used for | |
3660 | * the default hierarchy. | |
3661 | */ | |
3662 | int cgroup_add_dfl_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) | |
3663 | { | |
3664 | struct cftype *cft; | |
3665 | ||
3666 | for (cft = cfts; cft && cft->name[0] != '\0'; cft++) | |
05ebb6e6 | 3667 | cft->flags |= __CFTYPE_ONLY_ON_DFL; |
a8ddc821 TH |
3668 | return cgroup_add_cftypes(ss, cfts); |
3669 | } | |
3670 | ||
3671 | /** | |
3672 | * cgroup_add_legacy_cftypes - add an array of cftypes for legacy hierarchies | |
3673 | * @ss: target cgroup subsystem | |
3674 | * @cfts: zero-length name terminated array of cftypes | |
3675 | * | |
3676 | * Similar to cgroup_add_cftypes() but the added files are only used for | |
3677 | * the legacy hierarchies. | |
3678 | */ | |
2cf669a5 TH |
3679 | int cgroup_add_legacy_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
3680 | { | |
a8ddc821 TH |
3681 | struct cftype *cft; |
3682 | ||
e4b7037c TH |
3683 | for (cft = cfts; cft && cft->name[0] != '\0'; cft++) |
3684 | cft->flags |= __CFTYPE_NOT_ON_DFL; | |
2cf669a5 TH |
3685 | return cgroup_add_cftypes(ss, cfts); |
3686 | } | |
3687 | ||
34c06254 TH |
3688 | /** |
3689 | * cgroup_file_notify - generate a file modified event for a cgroup_file | |
3690 | * @cfile: target cgroup_file | |
3691 | * | |
3692 | * @cfile must have been obtained by setting cftype->file_offset. | |
3693 | */ | |
3694 | void cgroup_file_notify(struct cgroup_file *cfile) | |
3695 | { | |
3696 | unsigned long flags; | |
3697 | ||
3698 | spin_lock_irqsave(&cgroup_file_kn_lock, flags); | |
3699 | if (cfile->kn) | |
3700 | kernfs_notify(cfile->kn); | |
3701 | spin_unlock_irqrestore(&cgroup_file_kn_lock, flags); | |
3702 | } | |
3703 | ||
a043e3b2 LZ |
3704 | /** |
3705 | * cgroup_task_count - count the number of tasks in a cgroup. | |
3706 | * @cgrp: the cgroup in question | |
3707 | * | |
3708 | * Return the number of tasks in the cgroup. | |
3709 | */ | |
07bc356e | 3710 | static int cgroup_task_count(const struct cgroup *cgrp) |
bbcb81d0 PM |
3711 | { |
3712 | int count = 0; | |
69d0206c | 3713 | struct cgrp_cset_link *link; |
817929ec | 3714 | |
f0d9a5f1 | 3715 | spin_lock_bh(&css_set_lock); |
69d0206c TH |
3716 | list_for_each_entry(link, &cgrp->cset_links, cset_link) |
3717 | count += atomic_read(&link->cset->refcount); | |
f0d9a5f1 | 3718 | spin_unlock_bh(&css_set_lock); |
bbcb81d0 PM |
3719 | return count; |
3720 | } | |
3721 | ||
53fa5261 | 3722 | /** |
492eb21b | 3723 | * css_next_child - find the next child of a given css |
c2931b70 TH |
3724 | * @pos: the current position (%NULL to initiate traversal) |
3725 | * @parent: css whose children to walk | |
53fa5261 | 3726 | * |
c2931b70 | 3727 | * This function returns the next child of @parent and should be called |
87fb54f1 | 3728 | * under either cgroup_mutex or RCU read lock. The only requirement is |
c2931b70 TH |
3729 | * that @parent and @pos are accessible. The next sibling is guaranteed to |
3730 | * be returned regardless of their states. | |
3731 | * | |
3732 | * If a subsystem synchronizes ->css_online() and the start of iteration, a | |
3733 | * css which finished ->css_online() is guaranteed to be visible in the | |
3734 | * future iterations and will stay visible until the last reference is put. | |
3735 | * A css which hasn't finished ->css_online() or already finished | |
3736 | * ->css_offline() may show up during traversal. It's each subsystem's | |
3737 | * responsibility to synchronize against on/offlining. | |
53fa5261 | 3738 | */ |
c2931b70 TH |
3739 | struct cgroup_subsys_state *css_next_child(struct cgroup_subsys_state *pos, |
3740 | struct cgroup_subsys_state *parent) | |
53fa5261 | 3741 | { |
c2931b70 | 3742 | struct cgroup_subsys_state *next; |
53fa5261 | 3743 | |
8353da1f | 3744 | cgroup_assert_mutex_or_rcu_locked(); |
53fa5261 TH |
3745 | |
3746 | /* | |
de3f0341 TH |
3747 | * @pos could already have been unlinked from the sibling list. |
3748 | * Once a cgroup is removed, its ->sibling.next is no longer | |
3749 | * updated when its next sibling changes. CSS_RELEASED is set when | |
3750 | * @pos is taken off list, at which time its next pointer is valid, | |
3751 | * and, as releases are serialized, the one pointed to by the next | |
3752 | * pointer is guaranteed to not have started release yet. This | |
3753 | * implies that if we observe !CSS_RELEASED on @pos in this RCU | |
3754 | * critical section, the one pointed to by its next pointer is | |
3755 | * guaranteed to not have finished its RCU grace period even if we | |
3756 | * have dropped rcu_read_lock() inbetween iterations. | |
3b287a50 | 3757 | * |
de3f0341 TH |
3758 | * If @pos has CSS_RELEASED set, its next pointer can't be |
3759 | * dereferenced; however, as each css is given a monotonically | |
3760 | * increasing unique serial number and always appended to the | |
3761 | * sibling list, the next one can be found by walking the parent's | |
3762 | * children until the first css with higher serial number than | |
3763 | * @pos's. While this path can be slower, it happens iff iteration | |
3764 | * races against release and the race window is very small. | |
53fa5261 | 3765 | */ |
3b287a50 | 3766 | if (!pos) { |
c2931b70 TH |
3767 | next = list_entry_rcu(parent->children.next, struct cgroup_subsys_state, sibling); |
3768 | } else if (likely(!(pos->flags & CSS_RELEASED))) { | |
3769 | next = list_entry_rcu(pos->sibling.next, struct cgroup_subsys_state, sibling); | |
3b287a50 | 3770 | } else { |
c2931b70 | 3771 | list_for_each_entry_rcu(next, &parent->children, sibling) |
3b287a50 TH |
3772 | if (next->serial_nr > pos->serial_nr) |
3773 | break; | |
53fa5261 TH |
3774 | } |
3775 | ||
3b281afb TH |
3776 | /* |
3777 | * @next, if not pointing to the head, can be dereferenced and is | |
c2931b70 | 3778 | * the next sibling. |
3b281afb | 3779 | */ |
c2931b70 TH |
3780 | if (&next->sibling != &parent->children) |
3781 | return next; | |
3b281afb | 3782 | return NULL; |
53fa5261 | 3783 | } |
53fa5261 | 3784 | |
574bd9f7 | 3785 | /** |
492eb21b | 3786 | * css_next_descendant_pre - find the next descendant for pre-order walk |
574bd9f7 | 3787 | * @pos: the current position (%NULL to initiate traversal) |
492eb21b | 3788 | * @root: css whose descendants to walk |
574bd9f7 | 3789 | * |
492eb21b | 3790 | * To be used by css_for_each_descendant_pre(). Find the next descendant |
bd8815a6 TH |
3791 | * to visit for pre-order traversal of @root's descendants. @root is |
3792 | * included in the iteration and the first node to be visited. | |
75501a6d | 3793 | * |
87fb54f1 TH |
3794 | * While this function requires cgroup_mutex or RCU read locking, it |
3795 | * doesn't require the whole traversal to be contained in a single critical | |
3796 | * section. This function will return the correct next descendant as long | |
3797 | * as both @pos and @root are accessible and @pos is a descendant of @root. | |
c2931b70 TH |
3798 | * |
3799 | * If a subsystem synchronizes ->css_online() and the start of iteration, a | |
3800 | * css which finished ->css_online() is guaranteed to be visible in the | |
3801 | * future iterations and will stay visible until the last reference is put. | |
3802 | * A css which hasn't finished ->css_online() or already finished | |
3803 | * ->css_offline() may show up during traversal. It's each subsystem's | |
3804 | * responsibility to synchronize against on/offlining. | |
574bd9f7 | 3805 | */ |
492eb21b TH |
3806 | struct cgroup_subsys_state * |
3807 | css_next_descendant_pre(struct cgroup_subsys_state *pos, | |
3808 | struct cgroup_subsys_state *root) | |
574bd9f7 | 3809 | { |
492eb21b | 3810 | struct cgroup_subsys_state *next; |
574bd9f7 | 3811 | |
8353da1f | 3812 | cgroup_assert_mutex_or_rcu_locked(); |
574bd9f7 | 3813 | |
bd8815a6 | 3814 | /* if first iteration, visit @root */ |
7805d000 | 3815 | if (!pos) |
bd8815a6 | 3816 | return root; |
574bd9f7 TH |
3817 | |
3818 | /* visit the first child if exists */ | |
492eb21b | 3819 | next = css_next_child(NULL, pos); |
574bd9f7 TH |
3820 | if (next) |
3821 | return next; | |
3822 | ||
3823 | /* no child, visit my or the closest ancestor's next sibling */ | |
492eb21b | 3824 | while (pos != root) { |
5c9d535b | 3825 | next = css_next_child(pos, pos->parent); |
75501a6d | 3826 | if (next) |
574bd9f7 | 3827 | return next; |
5c9d535b | 3828 | pos = pos->parent; |
7805d000 | 3829 | } |
574bd9f7 TH |
3830 | |
3831 | return NULL; | |
3832 | } | |
574bd9f7 | 3833 | |
12a9d2fe | 3834 | /** |
492eb21b TH |
3835 | * css_rightmost_descendant - return the rightmost descendant of a css |
3836 | * @pos: css of interest | |
12a9d2fe | 3837 | * |
492eb21b TH |
3838 | * Return the rightmost descendant of @pos. If there's no descendant, @pos |
3839 | * is returned. This can be used during pre-order traversal to skip | |
12a9d2fe | 3840 | * subtree of @pos. |
75501a6d | 3841 | * |
87fb54f1 TH |
3842 | * While this function requires cgroup_mutex or RCU read locking, it |
3843 | * doesn't require the whole traversal to be contained in a single critical | |
3844 | * section. This function will return the correct rightmost descendant as | |
3845 | * long as @pos is accessible. | |
12a9d2fe | 3846 | */ |
492eb21b TH |
3847 | struct cgroup_subsys_state * |
3848 | css_rightmost_descendant(struct cgroup_subsys_state *pos) | |
12a9d2fe | 3849 | { |
492eb21b | 3850 | struct cgroup_subsys_state *last, *tmp; |
12a9d2fe | 3851 | |
8353da1f | 3852 | cgroup_assert_mutex_or_rcu_locked(); |
12a9d2fe TH |
3853 | |
3854 | do { | |
3855 | last = pos; | |
3856 | /* ->prev isn't RCU safe, walk ->next till the end */ | |
3857 | pos = NULL; | |
492eb21b | 3858 | css_for_each_child(tmp, last) |
12a9d2fe TH |
3859 | pos = tmp; |
3860 | } while (pos); | |
3861 | ||
3862 | return last; | |
3863 | } | |
12a9d2fe | 3864 | |
492eb21b TH |
3865 | static struct cgroup_subsys_state * |
3866 | css_leftmost_descendant(struct cgroup_subsys_state *pos) | |
574bd9f7 | 3867 | { |
492eb21b | 3868 | struct cgroup_subsys_state *last; |
574bd9f7 TH |
3869 | |
3870 | do { | |
3871 | last = pos; | |
492eb21b | 3872 | pos = css_next_child(NULL, pos); |
574bd9f7 TH |
3873 | } while (pos); |
3874 | ||
3875 | return last; | |
3876 | } | |
3877 | ||
3878 | /** | |
492eb21b | 3879 | * css_next_descendant_post - find the next descendant for post-order walk |
574bd9f7 | 3880 | * @pos: the current position (%NULL to initiate traversal) |
492eb21b | 3881 | * @root: css whose descendants to walk |
574bd9f7 | 3882 | * |
492eb21b | 3883 | * To be used by css_for_each_descendant_post(). Find the next descendant |
bd8815a6 TH |
3884 | * to visit for post-order traversal of @root's descendants. @root is |
3885 | * included in the iteration and the last node to be visited. | |
75501a6d | 3886 | * |
87fb54f1 TH |
3887 | * While this function requires cgroup_mutex or RCU read locking, it |
3888 | * doesn't require the whole traversal to be contained in a single critical | |
3889 | * section. This function will return the correct next descendant as long | |
3890 | * as both @pos and @cgroup are accessible and @pos is a descendant of | |
3891 | * @cgroup. | |
c2931b70 TH |
3892 | * |
3893 | * If a subsystem synchronizes ->css_online() and the start of iteration, a | |
3894 | * css which finished ->css_online() is guaranteed to be visible in the | |
3895 | * future iterations and will stay visible until the last reference is put. | |
3896 | * A css which hasn't finished ->css_online() or already finished | |
3897 | * ->css_offline() may show up during traversal. It's each subsystem's | |
3898 | * responsibility to synchronize against on/offlining. | |
574bd9f7 | 3899 | */ |
492eb21b TH |
3900 | struct cgroup_subsys_state * |
3901 | css_next_descendant_post(struct cgroup_subsys_state *pos, | |
3902 | struct cgroup_subsys_state *root) | |
574bd9f7 | 3903 | { |
492eb21b | 3904 | struct cgroup_subsys_state *next; |
574bd9f7 | 3905 | |
8353da1f | 3906 | cgroup_assert_mutex_or_rcu_locked(); |
574bd9f7 | 3907 | |
58b79a91 TH |
3908 | /* if first iteration, visit leftmost descendant which may be @root */ |
3909 | if (!pos) | |
3910 | return css_leftmost_descendant(root); | |
574bd9f7 | 3911 | |
bd8815a6 TH |
3912 | /* if we visited @root, we're done */ |
3913 | if (pos == root) | |
3914 | return NULL; | |
3915 | ||
574bd9f7 | 3916 | /* if there's an unvisited sibling, visit its leftmost descendant */ |
5c9d535b | 3917 | next = css_next_child(pos, pos->parent); |
75501a6d | 3918 | if (next) |
492eb21b | 3919 | return css_leftmost_descendant(next); |
574bd9f7 TH |
3920 | |
3921 | /* no sibling left, visit parent */ | |
5c9d535b | 3922 | return pos->parent; |
574bd9f7 | 3923 | } |
574bd9f7 | 3924 | |
f3d46500 TH |
3925 | /** |
3926 | * css_has_online_children - does a css have online children | |
3927 | * @css: the target css | |
3928 | * | |
3929 | * Returns %true if @css has any online children; otherwise, %false. This | |
3930 | * function can be called from any context but the caller is responsible | |
3931 | * for synchronizing against on/offlining as necessary. | |
3932 | */ | |
3933 | bool css_has_online_children(struct cgroup_subsys_state *css) | |
cbc125ef | 3934 | { |
f3d46500 TH |
3935 | struct cgroup_subsys_state *child; |
3936 | bool ret = false; | |
cbc125ef TH |
3937 | |
3938 | rcu_read_lock(); | |
f3d46500 | 3939 | css_for_each_child(child, css) { |
99bae5f9 | 3940 | if (child->flags & CSS_ONLINE) { |
f3d46500 TH |
3941 | ret = true; |
3942 | break; | |
cbc125ef TH |
3943 | } |
3944 | } | |
3945 | rcu_read_unlock(); | |
f3d46500 | 3946 | return ret; |
574bd9f7 | 3947 | } |
574bd9f7 | 3948 | |
0942eeee | 3949 | /** |
ecb9d535 | 3950 | * css_task_iter_advance_css_set - advance a task itererator to the next css_set |
0942eeee TH |
3951 | * @it: the iterator to advance |
3952 | * | |
3953 | * Advance @it to the next css_set to walk. | |
d515876e | 3954 | */ |
ecb9d535 | 3955 | static void css_task_iter_advance_css_set(struct css_task_iter *it) |
d515876e | 3956 | { |
0f0a2b4f | 3957 | struct list_head *l = it->cset_pos; |
d515876e TH |
3958 | struct cgrp_cset_link *link; |
3959 | struct css_set *cset; | |
3960 | ||
f0d9a5f1 | 3961 | lockdep_assert_held(&css_set_lock); |
ed27b9f7 | 3962 | |
d515876e TH |
3963 | /* Advance to the next non-empty css_set */ |
3964 | do { | |
3965 | l = l->next; | |
0f0a2b4f TH |
3966 | if (l == it->cset_head) { |
3967 | it->cset_pos = NULL; | |
ecb9d535 | 3968 | it->task_pos = NULL; |
d515876e TH |
3969 | return; |
3970 | } | |
3ebb2b6e TH |
3971 | |
3972 | if (it->ss) { | |
3973 | cset = container_of(l, struct css_set, | |
3974 | e_cset_node[it->ss->id]); | |
3975 | } else { | |
3976 | link = list_entry(l, struct cgrp_cset_link, cset_link); | |
3977 | cset = link->cset; | |
3978 | } | |
0de0942d | 3979 | } while (!css_set_populated(cset)); |
c7561128 | 3980 | |
0f0a2b4f | 3981 | it->cset_pos = l; |
c7561128 TH |
3982 | |
3983 | if (!list_empty(&cset->tasks)) | |
0f0a2b4f | 3984 | it->task_pos = cset->tasks.next; |
c7561128 | 3985 | else |
0f0a2b4f TH |
3986 | it->task_pos = cset->mg_tasks.next; |
3987 | ||
3988 | it->tasks_head = &cset->tasks; | |
3989 | it->mg_tasks_head = &cset->mg_tasks; | |
ed27b9f7 TH |
3990 | |
3991 | /* | |
3992 | * We don't keep css_sets locked across iteration steps and thus | |
3993 | * need to take steps to ensure that iteration can be resumed after | |
3994 | * the lock is re-acquired. Iteration is performed at two levels - | |
3995 | * css_sets and tasks in them. | |
3996 | * | |
3997 | * Once created, a css_set never leaves its cgroup lists, so a | |
3998 | * pinned css_set is guaranteed to stay put and we can resume | |
3999 | * iteration afterwards. | |
4000 | * | |
4001 | * Tasks may leave @cset across iteration steps. This is resolved | |
4002 | * by registering each iterator with the css_set currently being | |
4003 | * walked and making css_set_move_task() advance iterators whose | |
4004 | * next task is leaving. | |
4005 | */ | |
4006 | if (it->cur_cset) { | |
4007 | list_del(&it->iters_node); | |
4008 | put_css_set_locked(it->cur_cset); | |
4009 | } | |
4010 | get_css_set(cset); | |
4011 | it->cur_cset = cset; | |
4012 | list_add(&it->iters_node, &cset->task_iters); | |
d515876e TH |
4013 | } |
4014 | ||
ecb9d535 TH |
4015 | static void css_task_iter_advance(struct css_task_iter *it) |
4016 | { | |
4017 | struct list_head *l = it->task_pos; | |
4018 | ||
f0d9a5f1 | 4019 | lockdep_assert_held(&css_set_lock); |
ecb9d535 TH |
4020 | WARN_ON_ONCE(!l); |
4021 | ||
4022 | /* | |
4023 | * Advance iterator to find next entry. cset->tasks is consumed | |
4024 | * first and then ->mg_tasks. After ->mg_tasks, we move onto the | |
4025 | * next cset. | |
4026 | */ | |
4027 | l = l->next; | |
4028 | ||
4029 | if (l == it->tasks_head) | |
4030 | l = it->mg_tasks_head->next; | |
4031 | ||
4032 | if (l == it->mg_tasks_head) | |
4033 | css_task_iter_advance_css_set(it); | |
4034 | else | |
4035 | it->task_pos = l; | |
4036 | } | |
4037 | ||
0942eeee | 4038 | /** |
72ec7029 TH |
4039 | * css_task_iter_start - initiate task iteration |
4040 | * @css: the css to walk tasks of | |
0942eeee TH |
4041 | * @it: the task iterator to use |
4042 | * | |
72ec7029 TH |
4043 | * Initiate iteration through the tasks of @css. The caller can call |
4044 | * css_task_iter_next() to walk through the tasks until the function | |
4045 | * returns NULL. On completion of iteration, css_task_iter_end() must be | |
4046 | * called. | |
0942eeee | 4047 | */ |
72ec7029 TH |
4048 | void css_task_iter_start(struct cgroup_subsys_state *css, |
4049 | struct css_task_iter *it) | |
817929ec | 4050 | { |
56fde9e0 TH |
4051 | /* no one should try to iterate before mounting cgroups */ |
4052 | WARN_ON_ONCE(!use_task_css_set_links); | |
31a7df01 | 4053 | |
ed27b9f7 TH |
4054 | memset(it, 0, sizeof(*it)); |
4055 | ||
f0d9a5f1 | 4056 | spin_lock_bh(&css_set_lock); |
c59cd3d8 | 4057 | |
3ebb2b6e TH |
4058 | it->ss = css->ss; |
4059 | ||
4060 | if (it->ss) | |
4061 | it->cset_pos = &css->cgroup->e_csets[css->ss->id]; | |
4062 | else | |
4063 | it->cset_pos = &css->cgroup->cset_links; | |
4064 | ||
0f0a2b4f | 4065 | it->cset_head = it->cset_pos; |
c59cd3d8 | 4066 | |
ecb9d535 | 4067 | css_task_iter_advance_css_set(it); |
ed27b9f7 | 4068 | |
f0d9a5f1 | 4069 | spin_unlock_bh(&css_set_lock); |
817929ec PM |
4070 | } |
4071 | ||
0942eeee | 4072 | /** |
72ec7029 | 4073 | * css_task_iter_next - return the next task for the iterator |
0942eeee TH |
4074 | * @it: the task iterator being iterated |
4075 | * | |
4076 | * The "next" function for task iteration. @it should have been | |
72ec7029 TH |
4077 | * initialized via css_task_iter_start(). Returns NULL when the iteration |
4078 | * reaches the end. | |
0942eeee | 4079 | */ |
72ec7029 | 4080 | struct task_struct *css_task_iter_next(struct css_task_iter *it) |
817929ec | 4081 | { |
d5745675 | 4082 | if (it->cur_task) { |
ed27b9f7 | 4083 | put_task_struct(it->cur_task); |
d5745675 TH |
4084 | it->cur_task = NULL; |
4085 | } | |
ed27b9f7 | 4086 | |
f0d9a5f1 | 4087 | spin_lock_bh(&css_set_lock); |
ed27b9f7 | 4088 | |
d5745675 TH |
4089 | if (it->task_pos) { |
4090 | it->cur_task = list_entry(it->task_pos, struct task_struct, | |
4091 | cg_list); | |
4092 | get_task_struct(it->cur_task); | |
4093 | css_task_iter_advance(it); | |
4094 | } | |
ed27b9f7 | 4095 | |
f0d9a5f1 | 4096 | spin_unlock_bh(&css_set_lock); |
ed27b9f7 TH |
4097 | |
4098 | return it->cur_task; | |
817929ec PM |
4099 | } |
4100 | ||
0942eeee | 4101 | /** |
72ec7029 | 4102 | * css_task_iter_end - finish task iteration |
0942eeee TH |
4103 | * @it: the task iterator to finish |
4104 | * | |
72ec7029 | 4105 | * Finish task iteration started by css_task_iter_start(). |
0942eeee | 4106 | */ |
72ec7029 | 4107 | void css_task_iter_end(struct css_task_iter *it) |
31a7df01 | 4108 | { |
ed27b9f7 | 4109 | if (it->cur_cset) { |
f0d9a5f1 | 4110 | spin_lock_bh(&css_set_lock); |
ed27b9f7 TH |
4111 | list_del(&it->iters_node); |
4112 | put_css_set_locked(it->cur_cset); | |
f0d9a5f1 | 4113 | spin_unlock_bh(&css_set_lock); |
ed27b9f7 TH |
4114 | } |
4115 | ||
4116 | if (it->cur_task) | |
4117 | put_task_struct(it->cur_task); | |
31a7df01 CW |
4118 | } |
4119 | ||
4120 | /** | |
8cc99345 TH |
4121 | * cgroup_trasnsfer_tasks - move tasks from one cgroup to another |
4122 | * @to: cgroup to which the tasks will be moved | |
4123 | * @from: cgroup in which the tasks currently reside | |
31a7df01 | 4124 | * |
eaf797ab TH |
4125 | * Locking rules between cgroup_post_fork() and the migration path |
4126 | * guarantee that, if a task is forking while being migrated, the new child | |
4127 | * is guaranteed to be either visible in the source cgroup after the | |
4128 | * parent's migration is complete or put into the target cgroup. No task | |
4129 | * can slip out of migration through forking. | |
31a7df01 | 4130 | */ |
8cc99345 | 4131 | int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from) |
31a7df01 | 4132 | { |
952aaa12 TH |
4133 | LIST_HEAD(preloaded_csets); |
4134 | struct cgrp_cset_link *link; | |
72ec7029 | 4135 | struct css_task_iter it; |
e406d1cf | 4136 | struct task_struct *task; |
952aaa12 | 4137 | int ret; |
31a7df01 | 4138 | |
952aaa12 | 4139 | mutex_lock(&cgroup_mutex); |
31a7df01 | 4140 | |
952aaa12 | 4141 | /* all tasks in @from are being moved, all csets are source */ |
f0d9a5f1 | 4142 | spin_lock_bh(&css_set_lock); |
952aaa12 TH |
4143 | list_for_each_entry(link, &from->cset_links, cset_link) |
4144 | cgroup_migrate_add_src(link->cset, to, &preloaded_csets); | |
f0d9a5f1 | 4145 | spin_unlock_bh(&css_set_lock); |
31a7df01 | 4146 | |
952aaa12 TH |
4147 | ret = cgroup_migrate_prepare_dst(to, &preloaded_csets); |
4148 | if (ret) | |
4149 | goto out_err; | |
8cc99345 | 4150 | |
952aaa12 | 4151 | /* |
2cfa2b19 | 4152 | * Migrate tasks one-by-one until @from is empty. This fails iff |
952aaa12 TH |
4153 | * ->can_attach() fails. |
4154 | */ | |
e406d1cf | 4155 | do { |
9d800df1 | 4156 | css_task_iter_start(&from->self, &it); |
e406d1cf TH |
4157 | task = css_task_iter_next(&it); |
4158 | if (task) | |
4159 | get_task_struct(task); | |
4160 | css_task_iter_end(&it); | |
4161 | ||
4162 | if (task) { | |
9af2ec45 | 4163 | ret = cgroup_migrate(task, false, to); |
e406d1cf TH |
4164 | put_task_struct(task); |
4165 | } | |
4166 | } while (task && !ret); | |
952aaa12 TH |
4167 | out_err: |
4168 | cgroup_migrate_finish(&preloaded_csets); | |
47cfcd09 | 4169 | mutex_unlock(&cgroup_mutex); |
e406d1cf | 4170 | return ret; |
8cc99345 TH |
4171 | } |
4172 | ||
bbcb81d0 | 4173 | /* |
102a775e | 4174 | * Stuff for reading the 'tasks'/'procs' files. |
bbcb81d0 PM |
4175 | * |
4176 | * Reading this file can return large amounts of data if a cgroup has | |
4177 | * *lots* of attached tasks. So it may need several calls to read(), | |
4178 | * but we cannot guarantee that the information we produce is correct | |
4179 | * unless we produce it entirely atomically. | |
4180 | * | |
bbcb81d0 | 4181 | */ |
bbcb81d0 | 4182 | |
24528255 LZ |
4183 | /* which pidlist file are we talking about? */ |
4184 | enum cgroup_filetype { | |
4185 | CGROUP_FILE_PROCS, | |
4186 | CGROUP_FILE_TASKS, | |
4187 | }; | |
4188 | ||
4189 | /* | |
4190 | * A pidlist is a list of pids that virtually represents the contents of one | |
4191 | * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists, | |
4192 | * a pair (one each for procs, tasks) for each pid namespace that's relevant | |
4193 | * to the cgroup. | |
4194 | */ | |
4195 | struct cgroup_pidlist { | |
4196 | /* | |
4197 | * used to find which pidlist is wanted. doesn't change as long as | |
4198 | * this particular list stays in the list. | |
4199 | */ | |
4200 | struct { enum cgroup_filetype type; struct pid_namespace *ns; } key; | |
4201 | /* array of xids */ | |
4202 | pid_t *list; | |
4203 | /* how many elements the above list has */ | |
4204 | int length; | |
24528255 LZ |
4205 | /* each of these stored in a list by its cgroup */ |
4206 | struct list_head links; | |
4207 | /* pointer to the cgroup we belong to, for list removal purposes */ | |
4208 | struct cgroup *owner; | |
b1a21367 TH |
4209 | /* for delayed destruction */ |
4210 | struct delayed_work destroy_dwork; | |
24528255 LZ |
4211 | }; |
4212 | ||
d1d9fd33 BB |
4213 | /* |
4214 | * The following two functions "fix" the issue where there are more pids | |
4215 | * than kmalloc will give memory for; in such cases, we use vmalloc/vfree. | |
4216 | * TODO: replace with a kernel-wide solution to this problem | |
4217 | */ | |
4218 | #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2)) | |
4219 | static void *pidlist_allocate(int count) | |
4220 | { | |
4221 | if (PIDLIST_TOO_LARGE(count)) | |
4222 | return vmalloc(count * sizeof(pid_t)); | |
4223 | else | |
4224 | return kmalloc(count * sizeof(pid_t), GFP_KERNEL); | |
4225 | } | |
b1a21367 | 4226 | |
d1d9fd33 BB |
4227 | static void pidlist_free(void *p) |
4228 | { | |
58794514 | 4229 | kvfree(p); |
d1d9fd33 | 4230 | } |
d1d9fd33 | 4231 | |
b1a21367 TH |
4232 | /* |
4233 | * Used to destroy all pidlists lingering waiting for destroy timer. None | |
4234 | * should be left afterwards. | |
4235 | */ | |
4236 | static void cgroup_pidlist_destroy_all(struct cgroup *cgrp) | |
4237 | { | |
4238 | struct cgroup_pidlist *l, *tmp_l; | |
4239 | ||
4240 | mutex_lock(&cgrp->pidlist_mutex); | |
4241 | list_for_each_entry_safe(l, tmp_l, &cgrp->pidlists, links) | |
4242 | mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, 0); | |
4243 | mutex_unlock(&cgrp->pidlist_mutex); | |
4244 | ||
4245 | flush_workqueue(cgroup_pidlist_destroy_wq); | |
4246 | BUG_ON(!list_empty(&cgrp->pidlists)); | |
4247 | } | |
4248 | ||
4249 | static void cgroup_pidlist_destroy_work_fn(struct work_struct *work) | |
4250 | { | |
4251 | struct delayed_work *dwork = to_delayed_work(work); | |
4252 | struct cgroup_pidlist *l = container_of(dwork, struct cgroup_pidlist, | |
4253 | destroy_dwork); | |
4254 | struct cgroup_pidlist *tofree = NULL; | |
4255 | ||
4256 | mutex_lock(&l->owner->pidlist_mutex); | |
b1a21367 TH |
4257 | |
4258 | /* | |
04502365 TH |
4259 | * Destroy iff we didn't get queued again. The state won't change |
4260 | * as destroy_dwork can only be queued while locked. | |
b1a21367 | 4261 | */ |
04502365 | 4262 | if (!delayed_work_pending(dwork)) { |
b1a21367 TH |
4263 | list_del(&l->links); |
4264 | pidlist_free(l->list); | |
4265 | put_pid_ns(l->key.ns); | |
4266 | tofree = l; | |
4267 | } | |
4268 | ||
b1a21367 TH |
4269 | mutex_unlock(&l->owner->pidlist_mutex); |
4270 | kfree(tofree); | |
4271 | } | |
4272 | ||
bbcb81d0 | 4273 | /* |
102a775e | 4274 | * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries |
6ee211ad | 4275 | * Returns the number of unique elements. |
bbcb81d0 | 4276 | */ |
6ee211ad | 4277 | static int pidlist_uniq(pid_t *list, int length) |
bbcb81d0 | 4278 | { |
102a775e | 4279 | int src, dest = 1; |
102a775e BB |
4280 | |
4281 | /* | |
4282 | * we presume the 0th element is unique, so i starts at 1. trivial | |
4283 | * edge cases first; no work needs to be done for either | |
4284 | */ | |
4285 | if (length == 0 || length == 1) | |
4286 | return length; | |
4287 | /* src and dest walk down the list; dest counts unique elements */ | |
4288 | for (src = 1; src < length; src++) { | |
4289 | /* find next unique element */ | |
4290 | while (list[src] == list[src-1]) { | |
4291 | src++; | |
4292 | if (src == length) | |
4293 | goto after; | |
4294 | } | |
4295 | /* dest always points to where the next unique element goes */ | |
4296 | list[dest] = list[src]; | |
4297 | dest++; | |
4298 | } | |
4299 | after: | |
102a775e BB |
4300 | return dest; |
4301 | } | |
4302 | ||
afb2bc14 TH |
4303 | /* |
4304 | * The two pid files - task and cgroup.procs - guaranteed that the result | |
4305 | * is sorted, which forced this whole pidlist fiasco. As pid order is | |
4306 | * different per namespace, each namespace needs differently sorted list, | |
4307 | * making it impossible to use, for example, single rbtree of member tasks | |
4308 | * sorted by task pointer. As pidlists can be fairly large, allocating one | |
4309 | * per open file is dangerous, so cgroup had to implement shared pool of | |
4310 | * pidlists keyed by cgroup and namespace. | |
4311 | * | |
4312 | * All this extra complexity was caused by the original implementation | |
4313 | * committing to an entirely unnecessary property. In the long term, we | |
aa6ec29b TH |
4314 | * want to do away with it. Explicitly scramble sort order if on the |
4315 | * default hierarchy so that no such expectation exists in the new | |
4316 | * interface. | |
afb2bc14 TH |
4317 | * |
4318 | * Scrambling is done by swapping every two consecutive bits, which is | |
4319 | * non-identity one-to-one mapping which disturbs sort order sufficiently. | |
4320 | */ | |
4321 | static pid_t pid_fry(pid_t pid) | |
4322 | { | |
4323 | unsigned a = pid & 0x55555555; | |
4324 | unsigned b = pid & 0xAAAAAAAA; | |
4325 | ||
4326 | return (a << 1) | (b >> 1); | |
4327 | } | |
4328 | ||
4329 | static pid_t cgroup_pid_fry(struct cgroup *cgrp, pid_t pid) | |
4330 | { | |
aa6ec29b | 4331 | if (cgroup_on_dfl(cgrp)) |
afb2bc14 TH |
4332 | return pid_fry(pid); |
4333 | else | |
4334 | return pid; | |
4335 | } | |
4336 | ||
102a775e BB |
4337 | static int cmppid(const void *a, const void *b) |
4338 | { | |
4339 | return *(pid_t *)a - *(pid_t *)b; | |
4340 | } | |
4341 | ||
afb2bc14 TH |
4342 | static int fried_cmppid(const void *a, const void *b) |
4343 | { | |
4344 | return pid_fry(*(pid_t *)a) - pid_fry(*(pid_t *)b); | |
4345 | } | |
4346 | ||
e6b81710 TH |
4347 | static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp, |
4348 | enum cgroup_filetype type) | |
4349 | { | |
4350 | struct cgroup_pidlist *l; | |
4351 | /* don't need task_nsproxy() if we're looking at ourself */ | |
4352 | struct pid_namespace *ns = task_active_pid_ns(current); | |
4353 | ||
4354 | lockdep_assert_held(&cgrp->pidlist_mutex); | |
4355 | ||
4356 | list_for_each_entry(l, &cgrp->pidlists, links) | |
4357 | if (l->key.type == type && l->key.ns == ns) | |
4358 | return l; | |
4359 | return NULL; | |
4360 | } | |
4361 | ||
72a8cb30 BB |
4362 | /* |
4363 | * find the appropriate pidlist for our purpose (given procs vs tasks) | |
4364 | * returns with the lock on that pidlist already held, and takes care | |
4365 | * of the use count, or returns NULL with no locks held if we're out of | |
4366 | * memory. | |
4367 | */ | |
e6b81710 TH |
4368 | static struct cgroup_pidlist *cgroup_pidlist_find_create(struct cgroup *cgrp, |
4369 | enum cgroup_filetype type) | |
72a8cb30 BB |
4370 | { |
4371 | struct cgroup_pidlist *l; | |
b70cc5fd | 4372 | |
e6b81710 TH |
4373 | lockdep_assert_held(&cgrp->pidlist_mutex); |
4374 | ||
4375 | l = cgroup_pidlist_find(cgrp, type); | |
4376 | if (l) | |
4377 | return l; | |
4378 | ||
72a8cb30 | 4379 | /* entry not found; create a new one */ |
f4f4be2b | 4380 | l = kzalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL); |
e6b81710 | 4381 | if (!l) |
72a8cb30 | 4382 | return l; |
e6b81710 | 4383 | |
b1a21367 | 4384 | INIT_DELAYED_WORK(&l->destroy_dwork, cgroup_pidlist_destroy_work_fn); |
72a8cb30 | 4385 | l->key.type = type; |
e6b81710 TH |
4386 | /* don't need task_nsproxy() if we're looking at ourself */ |
4387 | l->key.ns = get_pid_ns(task_active_pid_ns(current)); | |
72a8cb30 BB |
4388 | l->owner = cgrp; |
4389 | list_add(&l->links, &cgrp->pidlists); | |
72a8cb30 BB |
4390 | return l; |
4391 | } | |
4392 | ||
102a775e BB |
4393 | /* |
4394 | * Load a cgroup's pidarray with either procs' tgids or tasks' pids | |
4395 | */ | |
72a8cb30 BB |
4396 | static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type, |
4397 | struct cgroup_pidlist **lp) | |
102a775e BB |
4398 | { |
4399 | pid_t *array; | |
4400 | int length; | |
4401 | int pid, n = 0; /* used for populating the array */ | |
72ec7029 | 4402 | struct css_task_iter it; |
817929ec | 4403 | struct task_struct *tsk; |
102a775e BB |
4404 | struct cgroup_pidlist *l; |
4405 | ||
4bac00d1 TH |
4406 | lockdep_assert_held(&cgrp->pidlist_mutex); |
4407 | ||
102a775e BB |
4408 | /* |
4409 | * If cgroup gets more users after we read count, we won't have | |
4410 | * enough space - tough. This race is indistinguishable to the | |
4411 | * caller from the case that the additional cgroup users didn't | |
4412 | * show up until sometime later on. | |
4413 | */ | |
4414 | length = cgroup_task_count(cgrp); | |
d1d9fd33 | 4415 | array = pidlist_allocate(length); |
102a775e BB |
4416 | if (!array) |
4417 | return -ENOMEM; | |
4418 | /* now, populate the array */ | |
9d800df1 | 4419 | css_task_iter_start(&cgrp->self, &it); |
72ec7029 | 4420 | while ((tsk = css_task_iter_next(&it))) { |
102a775e | 4421 | if (unlikely(n == length)) |
817929ec | 4422 | break; |
102a775e | 4423 | /* get tgid or pid for procs or tasks file respectively */ |
72a8cb30 BB |
4424 | if (type == CGROUP_FILE_PROCS) |
4425 | pid = task_tgid_vnr(tsk); | |
4426 | else | |
4427 | pid = task_pid_vnr(tsk); | |
102a775e BB |
4428 | if (pid > 0) /* make sure to only use valid results */ |
4429 | array[n++] = pid; | |
817929ec | 4430 | } |
72ec7029 | 4431 | css_task_iter_end(&it); |
102a775e BB |
4432 | length = n; |
4433 | /* now sort & (if procs) strip out duplicates */ | |
aa6ec29b | 4434 | if (cgroup_on_dfl(cgrp)) |
afb2bc14 TH |
4435 | sort(array, length, sizeof(pid_t), fried_cmppid, NULL); |
4436 | else | |
4437 | sort(array, length, sizeof(pid_t), cmppid, NULL); | |
72a8cb30 | 4438 | if (type == CGROUP_FILE_PROCS) |
6ee211ad | 4439 | length = pidlist_uniq(array, length); |
e6b81710 | 4440 | |
e6b81710 | 4441 | l = cgroup_pidlist_find_create(cgrp, type); |
72a8cb30 | 4442 | if (!l) { |
d1d9fd33 | 4443 | pidlist_free(array); |
72a8cb30 | 4444 | return -ENOMEM; |
102a775e | 4445 | } |
e6b81710 TH |
4446 | |
4447 | /* store array, freeing old if necessary */ | |
d1d9fd33 | 4448 | pidlist_free(l->list); |
102a775e BB |
4449 | l->list = array; |
4450 | l->length = length; | |
72a8cb30 | 4451 | *lp = l; |
102a775e | 4452 | return 0; |
bbcb81d0 PM |
4453 | } |
4454 | ||
846c7bb0 | 4455 | /** |
a043e3b2 | 4456 | * cgroupstats_build - build and fill cgroupstats |
846c7bb0 BS |
4457 | * @stats: cgroupstats to fill information into |
4458 | * @dentry: A dentry entry belonging to the cgroup for which stats have | |
4459 | * been requested. | |
a043e3b2 LZ |
4460 | * |
4461 | * Build and fill cgroupstats so that taskstats can export it to user | |
4462 | * space. | |
846c7bb0 BS |
4463 | */ |
4464 | int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry) | |
4465 | { | |
2bd59d48 | 4466 | struct kernfs_node *kn = kernfs_node_from_dentry(dentry); |
bd89aabc | 4467 | struct cgroup *cgrp; |
72ec7029 | 4468 | struct css_task_iter it; |
846c7bb0 | 4469 | struct task_struct *tsk; |
33d283be | 4470 | |
2bd59d48 TH |
4471 | /* it should be kernfs_node belonging to cgroupfs and is a directory */ |
4472 | if (dentry->d_sb->s_type != &cgroup_fs_type || !kn || | |
4473 | kernfs_type(kn) != KERNFS_DIR) | |
4474 | return -EINVAL; | |
4475 | ||
bad34660 LZ |
4476 | mutex_lock(&cgroup_mutex); |
4477 | ||
846c7bb0 | 4478 | /* |
2bd59d48 | 4479 | * We aren't being called from kernfs and there's no guarantee on |
ec903c0c | 4480 | * @kn->priv's validity. For this and css_tryget_online_from_dir(), |
2bd59d48 | 4481 | * @kn->priv is RCU safe. Let's do the RCU dancing. |
846c7bb0 | 4482 | */ |
2bd59d48 TH |
4483 | rcu_read_lock(); |
4484 | cgrp = rcu_dereference(kn->priv); | |
bad34660 | 4485 | if (!cgrp || cgroup_is_dead(cgrp)) { |
2bd59d48 | 4486 | rcu_read_unlock(); |
bad34660 | 4487 | mutex_unlock(&cgroup_mutex); |
2bd59d48 TH |
4488 | return -ENOENT; |
4489 | } | |
bad34660 | 4490 | rcu_read_unlock(); |
846c7bb0 | 4491 | |
9d800df1 | 4492 | css_task_iter_start(&cgrp->self, &it); |
72ec7029 | 4493 | while ((tsk = css_task_iter_next(&it))) { |
846c7bb0 BS |
4494 | switch (tsk->state) { |
4495 | case TASK_RUNNING: | |
4496 | stats->nr_running++; | |
4497 | break; | |
4498 | case TASK_INTERRUPTIBLE: | |
4499 | stats->nr_sleeping++; | |
4500 | break; | |
4501 | case TASK_UNINTERRUPTIBLE: | |
4502 | stats->nr_uninterruptible++; | |
4503 | break; | |
4504 | case TASK_STOPPED: | |
4505 | stats->nr_stopped++; | |
4506 | break; | |
4507 | default: | |
4508 | if (delayacct_is_task_waiting_on_io(tsk)) | |
4509 | stats->nr_io_wait++; | |
4510 | break; | |
4511 | } | |
4512 | } | |
72ec7029 | 4513 | css_task_iter_end(&it); |
846c7bb0 | 4514 | |
bad34660 | 4515 | mutex_unlock(&cgroup_mutex); |
2bd59d48 | 4516 | return 0; |
846c7bb0 BS |
4517 | } |
4518 | ||
8f3ff208 | 4519 | |
bbcb81d0 | 4520 | /* |
102a775e | 4521 | * seq_file methods for the tasks/procs files. The seq_file position is the |
cc31edce | 4522 | * next pid to display; the seq_file iterator is a pointer to the pid |
102a775e | 4523 | * in the cgroup->l->list array. |
bbcb81d0 | 4524 | */ |
cc31edce | 4525 | |
102a775e | 4526 | static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos) |
bbcb81d0 | 4527 | { |
cc31edce PM |
4528 | /* |
4529 | * Initially we receive a position value that corresponds to | |
4530 | * one more than the last pid shown (or 0 on the first call or | |
4531 | * after a seek to the start). Use a binary-search to find the | |
4532 | * next pid to display, if any | |
4533 | */ | |
2bd59d48 | 4534 | struct kernfs_open_file *of = s->private; |
7da11279 | 4535 | struct cgroup *cgrp = seq_css(s)->cgroup; |
4bac00d1 | 4536 | struct cgroup_pidlist *l; |
7da11279 | 4537 | enum cgroup_filetype type = seq_cft(s)->private; |
cc31edce | 4538 | int index = 0, pid = *pos; |
4bac00d1 TH |
4539 | int *iter, ret; |
4540 | ||
4541 | mutex_lock(&cgrp->pidlist_mutex); | |
4542 | ||
4543 | /* | |
5d22444f | 4544 | * !NULL @of->priv indicates that this isn't the first start() |
4bac00d1 | 4545 | * after open. If the matching pidlist is around, we can use that. |
5d22444f | 4546 | * Look for it. Note that @of->priv can't be used directly. It |
4bac00d1 TH |
4547 | * could already have been destroyed. |
4548 | */ | |
5d22444f TH |
4549 | if (of->priv) |
4550 | of->priv = cgroup_pidlist_find(cgrp, type); | |
4bac00d1 TH |
4551 | |
4552 | /* | |
4553 | * Either this is the first start() after open or the matching | |
4554 | * pidlist has been destroyed inbetween. Create a new one. | |
4555 | */ | |
5d22444f TH |
4556 | if (!of->priv) { |
4557 | ret = pidlist_array_load(cgrp, type, | |
4558 | (struct cgroup_pidlist **)&of->priv); | |
4bac00d1 TH |
4559 | if (ret) |
4560 | return ERR_PTR(ret); | |
4561 | } | |
5d22444f | 4562 | l = of->priv; |
cc31edce | 4563 | |
cc31edce | 4564 | if (pid) { |
102a775e | 4565 | int end = l->length; |
20777766 | 4566 | |
cc31edce PM |
4567 | while (index < end) { |
4568 | int mid = (index + end) / 2; | |
afb2bc14 | 4569 | if (cgroup_pid_fry(cgrp, l->list[mid]) == pid) { |
cc31edce PM |
4570 | index = mid; |
4571 | break; | |
afb2bc14 | 4572 | } else if (cgroup_pid_fry(cgrp, l->list[mid]) <= pid) |
cc31edce PM |
4573 | index = mid + 1; |
4574 | else | |
4575 | end = mid; | |
4576 | } | |
4577 | } | |
4578 | /* If we're off the end of the array, we're done */ | |
102a775e | 4579 | if (index >= l->length) |
cc31edce PM |
4580 | return NULL; |
4581 | /* Update the abstract position to be the actual pid that we found */ | |
102a775e | 4582 | iter = l->list + index; |
afb2bc14 | 4583 | *pos = cgroup_pid_fry(cgrp, *iter); |
cc31edce PM |
4584 | return iter; |
4585 | } | |
4586 | ||
102a775e | 4587 | static void cgroup_pidlist_stop(struct seq_file *s, void *v) |
cc31edce | 4588 | { |
2bd59d48 | 4589 | struct kernfs_open_file *of = s->private; |
5d22444f | 4590 | struct cgroup_pidlist *l = of->priv; |
62236858 | 4591 | |
5d22444f TH |
4592 | if (l) |
4593 | mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, | |
04502365 | 4594 | CGROUP_PIDLIST_DESTROY_DELAY); |
7da11279 | 4595 | mutex_unlock(&seq_css(s)->cgroup->pidlist_mutex); |
cc31edce PM |
4596 | } |
4597 | ||
102a775e | 4598 | static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos) |
cc31edce | 4599 | { |
2bd59d48 | 4600 | struct kernfs_open_file *of = s->private; |
5d22444f | 4601 | struct cgroup_pidlist *l = of->priv; |
102a775e BB |
4602 | pid_t *p = v; |
4603 | pid_t *end = l->list + l->length; | |
cc31edce PM |
4604 | /* |
4605 | * Advance to the next pid in the array. If this goes off the | |
4606 | * end, we're done | |
4607 | */ | |
4608 | p++; | |
4609 | if (p >= end) { | |
4610 | return NULL; | |
4611 | } else { | |
7da11279 | 4612 | *pos = cgroup_pid_fry(seq_css(s)->cgroup, *p); |
cc31edce PM |
4613 | return p; |
4614 | } | |
4615 | } | |
4616 | ||
102a775e | 4617 | static int cgroup_pidlist_show(struct seq_file *s, void *v) |
cc31edce | 4618 | { |
94ff212d JP |
4619 | seq_printf(s, "%d\n", *(int *)v); |
4620 | ||
4621 | return 0; | |
cc31edce | 4622 | } |
bbcb81d0 | 4623 | |
182446d0 TH |
4624 | static u64 cgroup_read_notify_on_release(struct cgroup_subsys_state *css, |
4625 | struct cftype *cft) | |
81a6a5cd | 4626 | { |
182446d0 | 4627 | return notify_on_release(css->cgroup); |
81a6a5cd PM |
4628 | } |
4629 | ||
182446d0 TH |
4630 | static int cgroup_write_notify_on_release(struct cgroup_subsys_state *css, |
4631 | struct cftype *cft, u64 val) | |
6379c106 | 4632 | { |
6379c106 | 4633 | if (val) |
182446d0 | 4634 | set_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags); |
6379c106 | 4635 | else |
182446d0 | 4636 | clear_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags); |
6379c106 PM |
4637 | return 0; |
4638 | } | |
4639 | ||
182446d0 TH |
4640 | static u64 cgroup_clone_children_read(struct cgroup_subsys_state *css, |
4641 | struct cftype *cft) | |
97978e6d | 4642 | { |
182446d0 | 4643 | return test_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d DL |
4644 | } |
4645 | ||
182446d0 TH |
4646 | static int cgroup_clone_children_write(struct cgroup_subsys_state *css, |
4647 | struct cftype *cft, u64 val) | |
97978e6d DL |
4648 | { |
4649 | if (val) | |
182446d0 | 4650 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d | 4651 | else |
182446d0 | 4652 | clear_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d DL |
4653 | return 0; |
4654 | } | |
4655 | ||
a14c6874 TH |
4656 | /* cgroup core interface files for the default hierarchy */ |
4657 | static struct cftype cgroup_dfl_base_files[] = { | |
81a6a5cd | 4658 | { |
d5c56ced | 4659 | .name = "cgroup.procs", |
6f60eade | 4660 | .file_offset = offsetof(struct cgroup, procs_file), |
6612f05b TH |
4661 | .seq_start = cgroup_pidlist_start, |
4662 | .seq_next = cgroup_pidlist_next, | |
4663 | .seq_stop = cgroup_pidlist_stop, | |
4664 | .seq_show = cgroup_pidlist_show, | |
5d22444f | 4665 | .private = CGROUP_FILE_PROCS, |
acbef755 | 4666 | .write = cgroup_procs_write, |
102a775e | 4667 | }, |
f8f22e53 TH |
4668 | { |
4669 | .name = "cgroup.controllers", | |
f8f22e53 TH |
4670 | .seq_show = cgroup_controllers_show, |
4671 | }, | |
4672 | { | |
4673 | .name = "cgroup.subtree_control", | |
f8f22e53 | 4674 | .seq_show = cgroup_subtree_control_show, |
451af504 | 4675 | .write = cgroup_subtree_control_write, |
f8f22e53 | 4676 | }, |
842b597e | 4677 | { |
4a07c222 | 4678 | .name = "cgroup.events", |
a14c6874 | 4679 | .flags = CFTYPE_NOT_ON_ROOT, |
6f60eade | 4680 | .file_offset = offsetof(struct cgroup, events_file), |
4a07c222 | 4681 | .seq_show = cgroup_events_show, |
842b597e | 4682 | }, |
a14c6874 TH |
4683 | { } /* terminate */ |
4684 | }; | |
d5c56ced | 4685 | |
a14c6874 TH |
4686 | /* cgroup core interface files for the legacy hierarchies */ |
4687 | static struct cftype cgroup_legacy_base_files[] = { | |
4688 | { | |
4689 | .name = "cgroup.procs", | |
4690 | .seq_start = cgroup_pidlist_start, | |
4691 | .seq_next = cgroup_pidlist_next, | |
4692 | .seq_stop = cgroup_pidlist_stop, | |
4693 | .seq_show = cgroup_pidlist_show, | |
4694 | .private = CGROUP_FILE_PROCS, | |
4695 | .write = cgroup_procs_write, | |
a14c6874 TH |
4696 | }, |
4697 | { | |
4698 | .name = "cgroup.clone_children", | |
4699 | .read_u64 = cgroup_clone_children_read, | |
4700 | .write_u64 = cgroup_clone_children_write, | |
4701 | }, | |
4702 | { | |
4703 | .name = "cgroup.sane_behavior", | |
4704 | .flags = CFTYPE_ONLY_ON_ROOT, | |
4705 | .seq_show = cgroup_sane_behavior_show, | |
4706 | }, | |
d5c56ced TH |
4707 | { |
4708 | .name = "tasks", | |
6612f05b TH |
4709 | .seq_start = cgroup_pidlist_start, |
4710 | .seq_next = cgroup_pidlist_next, | |
4711 | .seq_stop = cgroup_pidlist_stop, | |
4712 | .seq_show = cgroup_pidlist_show, | |
5d22444f | 4713 | .private = CGROUP_FILE_TASKS, |
acbef755 | 4714 | .write = cgroup_tasks_write, |
d5c56ced TH |
4715 | }, |
4716 | { | |
4717 | .name = "notify_on_release", | |
d5c56ced TH |
4718 | .read_u64 = cgroup_read_notify_on_release, |
4719 | .write_u64 = cgroup_write_notify_on_release, | |
4720 | }, | |
6e6ff25b TH |
4721 | { |
4722 | .name = "release_agent", | |
a14c6874 | 4723 | .flags = CFTYPE_ONLY_ON_ROOT, |
2da8ca82 | 4724 | .seq_show = cgroup_release_agent_show, |
451af504 | 4725 | .write = cgroup_release_agent_write, |
5f469907 | 4726 | .max_write_len = PATH_MAX - 1, |
6e6ff25b | 4727 | }, |
db0416b6 | 4728 | { } /* terminate */ |
bbcb81d0 PM |
4729 | }; |
4730 | ||
0c21ead1 TH |
4731 | /* |
4732 | * css destruction is four-stage process. | |
4733 | * | |
4734 | * 1. Destruction starts. Killing of the percpu_ref is initiated. | |
4735 | * Implemented in kill_css(). | |
4736 | * | |
4737 | * 2. When the percpu_ref is confirmed to be visible as killed on all CPUs | |
ec903c0c TH |
4738 | * and thus css_tryget_online() is guaranteed to fail, the css can be |
4739 | * offlined by invoking offline_css(). After offlining, the base ref is | |
4740 | * put. Implemented in css_killed_work_fn(). | |
0c21ead1 TH |
4741 | * |
4742 | * 3. When the percpu_ref reaches zero, the only possible remaining | |
4743 | * accessors are inside RCU read sections. css_release() schedules the | |
4744 | * RCU callback. | |
4745 | * | |
4746 | * 4. After the grace period, the css can be freed. Implemented in | |
4747 | * css_free_work_fn(). | |
4748 | * | |
4749 | * It is actually hairier because both step 2 and 4 require process context | |
4750 | * and thus involve punting to css->destroy_work adding two additional | |
4751 | * steps to the already complex sequence. | |
4752 | */ | |
35ef10da | 4753 | static void css_free_work_fn(struct work_struct *work) |
48ddbe19 TH |
4754 | { |
4755 | struct cgroup_subsys_state *css = | |
35ef10da | 4756 | container_of(work, struct cgroup_subsys_state, destroy_work); |
01e58659 | 4757 | struct cgroup_subsys *ss = css->ss; |
0c21ead1 | 4758 | struct cgroup *cgrp = css->cgroup; |
48ddbe19 | 4759 | |
9a1049da TH |
4760 | percpu_ref_exit(&css->refcnt); |
4761 | ||
01e58659 | 4762 | if (ss) { |
9d755d33 | 4763 | /* css free path */ |
8bb5ef79 | 4764 | struct cgroup_subsys_state *parent = css->parent; |
01e58659 VD |
4765 | int id = css->id; |
4766 | ||
01e58659 VD |
4767 | ss->css_free(css); |
4768 | cgroup_idr_remove(&ss->css_idr, id); | |
9d755d33 | 4769 | cgroup_put(cgrp); |
8bb5ef79 TH |
4770 | |
4771 | if (parent) | |
4772 | css_put(parent); | |
9d755d33 TH |
4773 | } else { |
4774 | /* cgroup free path */ | |
4775 | atomic_dec(&cgrp->root->nr_cgrps); | |
4776 | cgroup_pidlist_destroy_all(cgrp); | |
971ff493 | 4777 | cancel_work_sync(&cgrp->release_agent_work); |
9d755d33 | 4778 | |
d51f39b0 | 4779 | if (cgroup_parent(cgrp)) { |
9d755d33 TH |
4780 | /* |
4781 | * We get a ref to the parent, and put the ref when | |
4782 | * this cgroup is being freed, so it's guaranteed | |
4783 | * that the parent won't be destroyed before its | |
4784 | * children. | |
4785 | */ | |
d51f39b0 | 4786 | cgroup_put(cgroup_parent(cgrp)); |
9d755d33 TH |
4787 | kernfs_put(cgrp->kn); |
4788 | kfree(cgrp); | |
4789 | } else { | |
4790 | /* | |
4791 | * This is root cgroup's refcnt reaching zero, | |
4792 | * which indicates that the root should be | |
4793 | * released. | |
4794 | */ | |
4795 | cgroup_destroy_root(cgrp->root); | |
4796 | } | |
4797 | } | |
48ddbe19 TH |
4798 | } |
4799 | ||
0c21ead1 | 4800 | static void css_free_rcu_fn(struct rcu_head *rcu_head) |
d3daf28d TH |
4801 | { |
4802 | struct cgroup_subsys_state *css = | |
0c21ead1 | 4803 | container_of(rcu_head, struct cgroup_subsys_state, rcu_head); |
d3daf28d | 4804 | |
35ef10da | 4805 | INIT_WORK(&css->destroy_work, css_free_work_fn); |
e5fca243 | 4806 | queue_work(cgroup_destroy_wq, &css->destroy_work); |
48ddbe19 TH |
4807 | } |
4808 | ||
25e15d83 | 4809 | static void css_release_work_fn(struct work_struct *work) |
d3daf28d TH |
4810 | { |
4811 | struct cgroup_subsys_state *css = | |
25e15d83 | 4812 | container_of(work, struct cgroup_subsys_state, destroy_work); |
15a4c835 | 4813 | struct cgroup_subsys *ss = css->ss; |
9d755d33 | 4814 | struct cgroup *cgrp = css->cgroup; |
15a4c835 | 4815 | |
1fed1b2e TH |
4816 | mutex_lock(&cgroup_mutex); |
4817 | ||
de3f0341 | 4818 | css->flags |= CSS_RELEASED; |
1fed1b2e TH |
4819 | list_del_rcu(&css->sibling); |
4820 | ||
9d755d33 TH |
4821 | if (ss) { |
4822 | /* css release path */ | |
01e58659 | 4823 | cgroup_idr_replace(&ss->css_idr, NULL, css->id); |
7d172cc8 TH |
4824 | if (ss->css_released) |
4825 | ss->css_released(css); | |
9d755d33 TH |
4826 | } else { |
4827 | /* cgroup release path */ | |
9d755d33 TH |
4828 | cgroup_idr_remove(&cgrp->root->cgroup_idr, cgrp->id); |
4829 | cgrp->id = -1; | |
a4189487 LZ |
4830 | |
4831 | /* | |
4832 | * There are two control paths which try to determine | |
4833 | * cgroup from dentry without going through kernfs - | |
4834 | * cgroupstats_build() and css_tryget_online_from_dir(). | |
4835 | * Those are supported by RCU protecting clearing of | |
4836 | * cgrp->kn->priv backpointer. | |
4837 | */ | |
6cd0f5bb TH |
4838 | if (cgrp->kn) |
4839 | RCU_INIT_POINTER(*(void __rcu __force **)&cgrp->kn->priv, | |
4840 | NULL); | |
9d755d33 | 4841 | } |
d3daf28d | 4842 | |
1fed1b2e TH |
4843 | mutex_unlock(&cgroup_mutex); |
4844 | ||
0c21ead1 | 4845 | call_rcu(&css->rcu_head, css_free_rcu_fn); |
d3daf28d TH |
4846 | } |
4847 | ||
d3daf28d TH |
4848 | static void css_release(struct percpu_ref *ref) |
4849 | { | |
4850 | struct cgroup_subsys_state *css = | |
4851 | container_of(ref, struct cgroup_subsys_state, refcnt); | |
4852 | ||
25e15d83 TH |
4853 | INIT_WORK(&css->destroy_work, css_release_work_fn); |
4854 | queue_work(cgroup_destroy_wq, &css->destroy_work); | |
d3daf28d TH |
4855 | } |
4856 | ||
ddfcadab TH |
4857 | static void init_and_link_css(struct cgroup_subsys_state *css, |
4858 | struct cgroup_subsys *ss, struct cgroup *cgrp) | |
ddbcc7e8 | 4859 | { |
0cb51d71 TH |
4860 | lockdep_assert_held(&cgroup_mutex); |
4861 | ||
ddfcadab TH |
4862 | cgroup_get(cgrp); |
4863 | ||
d5c419b6 | 4864 | memset(css, 0, sizeof(*css)); |
bd89aabc | 4865 | css->cgroup = cgrp; |
72c97e54 | 4866 | css->ss = ss; |
d5c419b6 TH |
4867 | INIT_LIST_HEAD(&css->sibling); |
4868 | INIT_LIST_HEAD(&css->children); | |
0cb51d71 | 4869 | css->serial_nr = css_serial_nr_next++; |
aa226ff4 | 4870 | atomic_set(&css->online_cnt, 0); |
0ae78e0b | 4871 | |
d51f39b0 TH |
4872 | if (cgroup_parent(cgrp)) { |
4873 | css->parent = cgroup_css(cgroup_parent(cgrp), ss); | |
ddfcadab | 4874 | css_get(css->parent); |
ddfcadab | 4875 | } |
48ddbe19 | 4876 | |
ca8bdcaf | 4877 | BUG_ON(cgroup_css(cgrp, ss)); |
ddbcc7e8 PM |
4878 | } |
4879 | ||
2a4ac633 | 4880 | /* invoke ->css_online() on a new CSS and mark it online if successful */ |
623f926b | 4881 | static int online_css(struct cgroup_subsys_state *css) |
a31f2d3f | 4882 | { |
623f926b | 4883 | struct cgroup_subsys *ss = css->ss; |
b1929db4 TH |
4884 | int ret = 0; |
4885 | ||
a31f2d3f TH |
4886 | lockdep_assert_held(&cgroup_mutex); |
4887 | ||
92fb9748 | 4888 | if (ss->css_online) |
eb95419b | 4889 | ret = ss->css_online(css); |
ae7f164a | 4890 | if (!ret) { |
eb95419b | 4891 | css->flags |= CSS_ONLINE; |
aec25020 | 4892 | rcu_assign_pointer(css->cgroup->subsys[ss->id], css); |
aa226ff4 TH |
4893 | |
4894 | atomic_inc(&css->online_cnt); | |
4895 | if (css->parent) | |
4896 | atomic_inc(&css->parent->online_cnt); | |
ae7f164a | 4897 | } |
b1929db4 | 4898 | return ret; |
a31f2d3f TH |
4899 | } |
4900 | ||
2a4ac633 | 4901 | /* if the CSS is online, invoke ->css_offline() on it and mark it offline */ |
623f926b | 4902 | static void offline_css(struct cgroup_subsys_state *css) |
a31f2d3f | 4903 | { |
623f926b | 4904 | struct cgroup_subsys *ss = css->ss; |
a31f2d3f TH |
4905 | |
4906 | lockdep_assert_held(&cgroup_mutex); | |
4907 | ||
4908 | if (!(css->flags & CSS_ONLINE)) | |
4909 | return; | |
4910 | ||
fa06235b VD |
4911 | if (ss->css_reset) |
4912 | ss->css_reset(css); | |
4913 | ||
d7eeac19 | 4914 | if (ss->css_offline) |
eb95419b | 4915 | ss->css_offline(css); |
a31f2d3f | 4916 | |
eb95419b | 4917 | css->flags &= ~CSS_ONLINE; |
e3297803 | 4918 | RCU_INIT_POINTER(css->cgroup->subsys[ss->id], NULL); |
f8f22e53 TH |
4919 | |
4920 | wake_up_all(&css->cgroup->offline_waitq); | |
a31f2d3f TH |
4921 | } |
4922 | ||
c81c925a | 4923 | /** |
6cd0f5bb | 4924 | * css_create - create a cgroup_subsys_state |
c81c925a TH |
4925 | * @cgrp: the cgroup new css will be associated with |
4926 | * @ss: the subsys of new css | |
4927 | * | |
4928 | * Create a new css associated with @cgrp - @ss pair. On success, the new | |
6cd0f5bb TH |
4929 | * css is online and installed in @cgrp. This function doesn't create the |
4930 | * interface files. Returns 0 on success, -errno on failure. | |
c81c925a | 4931 | */ |
6cd0f5bb TH |
4932 | static struct cgroup_subsys_state *css_create(struct cgroup *cgrp, |
4933 | struct cgroup_subsys *ss) | |
c81c925a | 4934 | { |
d51f39b0 | 4935 | struct cgroup *parent = cgroup_parent(cgrp); |
1fed1b2e | 4936 | struct cgroup_subsys_state *parent_css = cgroup_css(parent, ss); |
c81c925a TH |
4937 | struct cgroup_subsys_state *css; |
4938 | int err; | |
4939 | ||
c81c925a TH |
4940 | lockdep_assert_held(&cgroup_mutex); |
4941 | ||
1fed1b2e | 4942 | css = ss->css_alloc(parent_css); |
c81c925a | 4943 | if (IS_ERR(css)) |
6cd0f5bb | 4944 | return css; |
c81c925a | 4945 | |
ddfcadab | 4946 | init_and_link_css(css, ss, cgrp); |
a2bed820 | 4947 | |
2aad2a86 | 4948 | err = percpu_ref_init(&css->refcnt, css_release, 0, GFP_KERNEL); |
c81c925a | 4949 | if (err) |
3eb59ec6 | 4950 | goto err_free_css; |
c81c925a | 4951 | |
cf780b7d | 4952 | err = cgroup_idr_alloc(&ss->css_idr, NULL, 2, 0, GFP_KERNEL); |
15a4c835 TH |
4953 | if (err < 0) |
4954 | goto err_free_percpu_ref; | |
4955 | css->id = err; | |
c81c925a | 4956 | |
15a4c835 | 4957 | /* @css is ready to be brought online now, make it visible */ |
1fed1b2e | 4958 | list_add_tail_rcu(&css->sibling, &parent_css->children); |
15a4c835 | 4959 | cgroup_idr_replace(&ss->css_idr, css, css->id); |
c81c925a TH |
4960 | |
4961 | err = online_css(css); | |
4962 | if (err) | |
1fed1b2e | 4963 | goto err_list_del; |
94419627 | 4964 | |
c81c925a | 4965 | if (ss->broken_hierarchy && !ss->warned_broken_hierarchy && |
d51f39b0 | 4966 | cgroup_parent(parent)) { |
ed3d261b | 4967 | 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 | 4968 | current->comm, current->pid, ss->name); |
c81c925a | 4969 | if (!strcmp(ss->name, "memory")) |
ed3d261b | 4970 | pr_warn("\"memory\" requires setting use_hierarchy to 1 on the root\n"); |
c81c925a TH |
4971 | ss->warned_broken_hierarchy = true; |
4972 | } | |
4973 | ||
6cd0f5bb | 4974 | return css; |
c81c925a | 4975 | |
1fed1b2e TH |
4976 | err_list_del: |
4977 | list_del_rcu(&css->sibling); | |
15a4c835 | 4978 | cgroup_idr_remove(&ss->css_idr, css->id); |
3eb59ec6 | 4979 | err_free_percpu_ref: |
9a1049da | 4980 | percpu_ref_exit(&css->refcnt); |
3eb59ec6 | 4981 | err_free_css: |
a2bed820 | 4982 | call_rcu(&css->rcu_head, css_free_rcu_fn); |
6cd0f5bb | 4983 | return ERR_PTR(err); |
c81c925a TH |
4984 | } |
4985 | ||
a5bca215 | 4986 | static struct cgroup *cgroup_create(struct cgroup *parent) |
ddbcc7e8 | 4987 | { |
a5bca215 | 4988 | struct cgroup_root *root = parent->root; |
a5bca215 TH |
4989 | struct cgroup *cgrp, *tcgrp; |
4990 | int level = parent->level + 1; | |
03970d3c | 4991 | int ret; |
ddbcc7e8 | 4992 | |
0a950f65 | 4993 | /* allocate the cgroup and its ID, 0 is reserved for the root */ |
b11cfb58 TH |
4994 | cgrp = kzalloc(sizeof(*cgrp) + |
4995 | sizeof(cgrp->ancestor_ids[0]) * (level + 1), GFP_KERNEL); | |
a5bca215 TH |
4996 | if (!cgrp) |
4997 | return ERR_PTR(-ENOMEM); | |
0ab02ca8 | 4998 | |
2aad2a86 | 4999 | ret = percpu_ref_init(&cgrp->self.refcnt, css_release, 0, GFP_KERNEL); |
9d755d33 TH |
5000 | if (ret) |
5001 | goto out_free_cgrp; | |
5002 | ||
0ab02ca8 LZ |
5003 | /* |
5004 | * Temporarily set the pointer to NULL, so idr_find() won't return | |
5005 | * a half-baked cgroup. | |
5006 | */ | |
cf780b7d | 5007 | cgrp->id = cgroup_idr_alloc(&root->cgroup_idr, NULL, 2, 0, GFP_KERNEL); |
0ab02ca8 | 5008 | if (cgrp->id < 0) { |
ba0f4d76 | 5009 | ret = -ENOMEM; |
9d755d33 | 5010 | goto out_cancel_ref; |
976c06bc TH |
5011 | } |
5012 | ||
cc31edce | 5013 | init_cgroup_housekeeping(cgrp); |
ddbcc7e8 | 5014 | |
9d800df1 | 5015 | cgrp->self.parent = &parent->self; |
ba0f4d76 | 5016 | cgrp->root = root; |
b11cfb58 TH |
5017 | cgrp->level = level; |
5018 | ||
5019 | for (tcgrp = cgrp; tcgrp; tcgrp = cgroup_parent(tcgrp)) | |
5020 | cgrp->ancestor_ids[tcgrp->level] = tcgrp->id; | |
ddbcc7e8 | 5021 | |
b6abdb0e LZ |
5022 | if (notify_on_release(parent)) |
5023 | set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); | |
5024 | ||
2260e7fc TH |
5025 | if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &parent->flags)) |
5026 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags); | |
97978e6d | 5027 | |
0cb51d71 | 5028 | cgrp->self.serial_nr = css_serial_nr_next++; |
53fa5261 | 5029 | |
4e139afc | 5030 | /* allocation complete, commit to creation */ |
d5c419b6 | 5031 | list_add_tail_rcu(&cgrp->self.sibling, &cgroup_parent(cgrp)->self.children); |
3c9c825b | 5032 | atomic_inc(&root->nr_cgrps); |
59f5296b | 5033 | cgroup_get(parent); |
415cf07a | 5034 | |
0d80255e TH |
5035 | /* |
5036 | * @cgrp is now fully operational. If something fails after this | |
5037 | * point, it'll be released via the normal destruction path. | |
5038 | */ | |
6fa4918d | 5039 | cgroup_idr_replace(&root->cgroup_idr, cgrp, cgrp->id); |
4e96ee8e | 5040 | |
bd53d617 TH |
5041 | /* |
5042 | * On the default hierarchy, a child doesn't automatically inherit | |
667c2491 | 5043 | * subtree_control from the parent. Each is configured manually. |
bd53d617 | 5044 | */ |
03970d3c | 5045 | if (!cgroup_on_dfl(cgrp)) |
5531dc91 | 5046 | cgrp->subtree_control = cgroup_control(cgrp); |
03970d3c TH |
5047 | |
5048 | cgroup_propagate_control(cgrp); | |
5049 | ||
5050 | /* @cgrp doesn't have dir yet so the following will only create csses */ | |
5051 | ret = cgroup_apply_control_enable(cgrp); | |
5052 | if (ret) | |
5053 | goto out_destroy; | |
2bd59d48 | 5054 | |
a5bca215 TH |
5055 | return cgrp; |
5056 | ||
5057 | out_cancel_ref: | |
5058 | percpu_ref_exit(&cgrp->self.refcnt); | |
5059 | out_free_cgrp: | |
5060 | kfree(cgrp); | |
5061 | return ERR_PTR(ret); | |
5062 | out_destroy: | |
5063 | cgroup_destroy_locked(cgrp); | |
5064 | return ERR_PTR(ret); | |
5065 | } | |
5066 | ||
5067 | static int cgroup_mkdir(struct kernfs_node *parent_kn, const char *name, | |
5068 | umode_t mode) | |
5069 | { | |
5070 | struct cgroup *parent, *cgrp; | |
a5bca215 | 5071 | struct kernfs_node *kn; |
03970d3c | 5072 | int ret; |
a5bca215 TH |
5073 | |
5074 | /* do not accept '\n' to prevent making /proc/<pid>/cgroup unparsable */ | |
5075 | if (strchr(name, '\n')) | |
5076 | return -EINVAL; | |
5077 | ||
945ba199 | 5078 | parent = cgroup_kn_lock_live(parent_kn, false); |
a5bca215 TH |
5079 | if (!parent) |
5080 | return -ENODEV; | |
5081 | ||
5082 | cgrp = cgroup_create(parent); | |
5083 | if (IS_ERR(cgrp)) { | |
5084 | ret = PTR_ERR(cgrp); | |
5085 | goto out_unlock; | |
5086 | } | |
5087 | ||
195e9b6c TH |
5088 | /* create the directory */ |
5089 | kn = kernfs_create_dir(parent->kn, name, mode, cgrp); | |
5090 | if (IS_ERR(kn)) { | |
5091 | ret = PTR_ERR(kn); | |
5092 | goto out_destroy; | |
5093 | } | |
5094 | cgrp->kn = kn; | |
5095 | ||
5096 | /* | |
5097 | * This extra ref will be put in cgroup_free_fn() and guarantees | |
5098 | * that @cgrp->kn is always accessible. | |
5099 | */ | |
5100 | kernfs_get(kn); | |
5101 | ||
5102 | ret = cgroup_kn_set_ugid(kn); | |
5103 | if (ret) | |
5104 | goto out_destroy; | |
5105 | ||
334c3679 | 5106 | ret = css_populate_dir(&cgrp->self); |
195e9b6c TH |
5107 | if (ret) |
5108 | goto out_destroy; | |
5109 | ||
03970d3c TH |
5110 | ret = cgroup_apply_control_enable(cgrp); |
5111 | if (ret) | |
5112 | goto out_destroy; | |
195e9b6c TH |
5113 | |
5114 | /* let's create and online css's */ | |
2bd59d48 | 5115 | kernfs_activate(kn); |
ddbcc7e8 | 5116 | |
ba0f4d76 TH |
5117 | ret = 0; |
5118 | goto out_unlock; | |
ddbcc7e8 | 5119 | |
a5bca215 TH |
5120 | out_destroy: |
5121 | cgroup_destroy_locked(cgrp); | |
ba0f4d76 | 5122 | out_unlock: |
a9746d8d | 5123 | cgroup_kn_unlock(parent_kn); |
ba0f4d76 | 5124 | return ret; |
ddbcc7e8 PM |
5125 | } |
5126 | ||
223dbc38 TH |
5127 | /* |
5128 | * This is called when the refcnt of a css is confirmed to be killed. | |
249f3468 TH |
5129 | * css_tryget_online() is now guaranteed to fail. Tell the subsystem to |
5130 | * initate destruction and put the css ref from kill_css(). | |
223dbc38 TH |
5131 | */ |
5132 | static void css_killed_work_fn(struct work_struct *work) | |
d3daf28d | 5133 | { |
223dbc38 TH |
5134 | struct cgroup_subsys_state *css = |
5135 | container_of(work, struct cgroup_subsys_state, destroy_work); | |
d3daf28d | 5136 | |
f20104de | 5137 | mutex_lock(&cgroup_mutex); |
09a503ea | 5138 | |
aa226ff4 TH |
5139 | do { |
5140 | offline_css(css); | |
5141 | css_put(css); | |
5142 | /* @css can't go away while we're holding cgroup_mutex */ | |
5143 | css = css->parent; | |
5144 | } while (css && atomic_dec_and_test(&css->online_cnt)); | |
5145 | ||
5146 | mutex_unlock(&cgroup_mutex); | |
d3daf28d TH |
5147 | } |
5148 | ||
223dbc38 TH |
5149 | /* css kill confirmation processing requires process context, bounce */ |
5150 | static void css_killed_ref_fn(struct percpu_ref *ref) | |
d3daf28d TH |
5151 | { |
5152 | struct cgroup_subsys_state *css = | |
5153 | container_of(ref, struct cgroup_subsys_state, refcnt); | |
5154 | ||
aa226ff4 TH |
5155 | if (atomic_dec_and_test(&css->online_cnt)) { |
5156 | INIT_WORK(&css->destroy_work, css_killed_work_fn); | |
5157 | queue_work(cgroup_destroy_wq, &css->destroy_work); | |
5158 | } | |
d3daf28d TH |
5159 | } |
5160 | ||
f392e51c TH |
5161 | /** |
5162 | * kill_css - destroy a css | |
5163 | * @css: css to destroy | |
5164 | * | |
5165 | * This function initiates destruction of @css by removing cgroup interface | |
5166 | * files and putting its base reference. ->css_offline() will be invoked | |
ec903c0c TH |
5167 | * asynchronously once css_tryget_online() is guaranteed to fail and when |
5168 | * the reference count reaches zero, @css will be released. | |
f392e51c TH |
5169 | */ |
5170 | static void kill_css(struct cgroup_subsys_state *css) | |
edae0c33 | 5171 | { |
01f6474c | 5172 | lockdep_assert_held(&cgroup_mutex); |
94419627 | 5173 | |
2bd59d48 TH |
5174 | /* |
5175 | * This must happen before css is disassociated with its cgroup. | |
5176 | * See seq_css() for details. | |
5177 | */ | |
334c3679 | 5178 | css_clear_dir(css); |
3c14f8b4 | 5179 | |
edae0c33 TH |
5180 | /* |
5181 | * Killing would put the base ref, but we need to keep it alive | |
5182 | * until after ->css_offline(). | |
5183 | */ | |
5184 | css_get(css); | |
5185 | ||
5186 | /* | |
5187 | * cgroup core guarantees that, by the time ->css_offline() is | |
5188 | * invoked, no new css reference will be given out via | |
ec903c0c | 5189 | * css_tryget_online(). We can't simply call percpu_ref_kill() and |
edae0c33 TH |
5190 | * proceed to offlining css's because percpu_ref_kill() doesn't |
5191 | * guarantee that the ref is seen as killed on all CPUs on return. | |
5192 | * | |
5193 | * Use percpu_ref_kill_and_confirm() to get notifications as each | |
5194 | * css is confirmed to be seen as killed on all CPUs. | |
5195 | */ | |
5196 | percpu_ref_kill_and_confirm(&css->refcnt, css_killed_ref_fn); | |
d3daf28d TH |
5197 | } |
5198 | ||
5199 | /** | |
5200 | * cgroup_destroy_locked - the first stage of cgroup destruction | |
5201 | * @cgrp: cgroup to be destroyed | |
5202 | * | |
5203 | * css's make use of percpu refcnts whose killing latency shouldn't be | |
5204 | * exposed to userland and are RCU protected. Also, cgroup core needs to | |
ec903c0c TH |
5205 | * guarantee that css_tryget_online() won't succeed by the time |
5206 | * ->css_offline() is invoked. To satisfy all the requirements, | |
5207 | * destruction is implemented in the following two steps. | |
d3daf28d TH |
5208 | * |
5209 | * s1. Verify @cgrp can be destroyed and mark it dying. Remove all | |
5210 | * userland visible parts and start killing the percpu refcnts of | |
5211 | * css's. Set up so that the next stage will be kicked off once all | |
5212 | * the percpu refcnts are confirmed to be killed. | |
5213 | * | |
5214 | * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the | |
5215 | * rest of destruction. Once all cgroup references are gone, the | |
5216 | * cgroup is RCU-freed. | |
5217 | * | |
5218 | * This function implements s1. After this step, @cgrp is gone as far as | |
5219 | * the userland is concerned and a new cgroup with the same name may be | |
5220 | * created. As cgroup doesn't care about the names internally, this | |
5221 | * doesn't cause any problem. | |
5222 | */ | |
42809dd4 TH |
5223 | static int cgroup_destroy_locked(struct cgroup *cgrp) |
5224 | __releases(&cgroup_mutex) __acquires(&cgroup_mutex) | |
ddbcc7e8 | 5225 | { |
2bd59d48 | 5226 | struct cgroup_subsys_state *css; |
1c6727af | 5227 | int ssid; |
ddbcc7e8 | 5228 | |
42809dd4 TH |
5229 | lockdep_assert_held(&cgroup_mutex); |
5230 | ||
91486f61 TH |
5231 | /* |
5232 | * Only migration can raise populated from zero and we're already | |
5233 | * holding cgroup_mutex. | |
5234 | */ | |
5235 | if (cgroup_is_populated(cgrp)) | |
ddbcc7e8 | 5236 | return -EBUSY; |
a043e3b2 | 5237 | |
bb78a92f | 5238 | /* |
d5c419b6 TH |
5239 | * Make sure there's no live children. We can't test emptiness of |
5240 | * ->self.children as dead children linger on it while being | |
5241 | * drained; otherwise, "rmdir parent/child parent" may fail. | |
bb78a92f | 5242 | */ |
f3d46500 | 5243 | if (css_has_online_children(&cgrp->self)) |
bb78a92f HD |
5244 | return -EBUSY; |
5245 | ||
455050d2 TH |
5246 | /* |
5247 | * Mark @cgrp dead. This prevents further task migration and child | |
de3f0341 | 5248 | * creation by disabling cgroup_lock_live_group(). |
455050d2 | 5249 | */ |
184faf32 | 5250 | cgrp->self.flags &= ~CSS_ONLINE; |
ddbcc7e8 | 5251 | |
249f3468 | 5252 | /* initiate massacre of all css's */ |
1c6727af TH |
5253 | for_each_css(css, ssid, cgrp) |
5254 | kill_css(css); | |
455050d2 | 5255 | |
455050d2 | 5256 | /* |
01f6474c TH |
5257 | * Remove @cgrp directory along with the base files. @cgrp has an |
5258 | * extra ref on its kn. | |
f20104de | 5259 | */ |
01f6474c | 5260 | kernfs_remove(cgrp->kn); |
f20104de | 5261 | |
d51f39b0 | 5262 | check_for_release(cgroup_parent(cgrp)); |
2bd59d48 | 5263 | |
249f3468 | 5264 | /* put the base reference */ |
9d755d33 | 5265 | percpu_ref_kill(&cgrp->self.refcnt); |
455050d2 | 5266 | |
ea15f8cc TH |
5267 | return 0; |
5268 | }; | |
5269 | ||
2bd59d48 | 5270 | static int cgroup_rmdir(struct kernfs_node *kn) |
42809dd4 | 5271 | { |
a9746d8d | 5272 | struct cgroup *cgrp; |
2bd59d48 | 5273 | int ret = 0; |
42809dd4 | 5274 | |
945ba199 | 5275 | cgrp = cgroup_kn_lock_live(kn, false); |
a9746d8d TH |
5276 | if (!cgrp) |
5277 | return 0; | |
42809dd4 | 5278 | |
a9746d8d | 5279 | ret = cgroup_destroy_locked(cgrp); |
2bb566cb | 5280 | |
a9746d8d | 5281 | cgroup_kn_unlock(kn); |
42809dd4 | 5282 | return ret; |
8e3f6541 TH |
5283 | } |
5284 | ||
2bd59d48 TH |
5285 | static struct kernfs_syscall_ops cgroup_kf_syscall_ops = { |
5286 | .remount_fs = cgroup_remount, | |
5287 | .show_options = cgroup_show_options, | |
5288 | .mkdir = cgroup_mkdir, | |
5289 | .rmdir = cgroup_rmdir, | |
5290 | .rename = cgroup_rename, | |
5291 | }; | |
5292 | ||
15a4c835 | 5293 | static void __init cgroup_init_subsys(struct cgroup_subsys *ss, bool early) |
ddbcc7e8 | 5294 | { |
ddbcc7e8 | 5295 | struct cgroup_subsys_state *css; |
cfe36bde | 5296 | |
a5ae9899 | 5297 | pr_debug("Initializing cgroup subsys %s\n", ss->name); |
ddbcc7e8 | 5298 | |
648bb56d TH |
5299 | mutex_lock(&cgroup_mutex); |
5300 | ||
15a4c835 | 5301 | idr_init(&ss->css_idr); |
0adb0704 | 5302 | INIT_LIST_HEAD(&ss->cfts); |
8e3f6541 | 5303 | |
3dd06ffa TH |
5304 | /* Create the root cgroup state for this subsystem */ |
5305 | ss->root = &cgrp_dfl_root; | |
5306 | css = ss->css_alloc(cgroup_css(&cgrp_dfl_root.cgrp, ss)); | |
ddbcc7e8 PM |
5307 | /* We don't handle early failures gracefully */ |
5308 | BUG_ON(IS_ERR(css)); | |
ddfcadab | 5309 | init_and_link_css(css, ss, &cgrp_dfl_root.cgrp); |
3b514d24 TH |
5310 | |
5311 | /* | |
5312 | * Root csses are never destroyed and we can't initialize | |
5313 | * percpu_ref during early init. Disable refcnting. | |
5314 | */ | |
5315 | css->flags |= CSS_NO_REF; | |
5316 | ||
15a4c835 | 5317 | if (early) { |
9395a450 | 5318 | /* allocation can't be done safely during early init */ |
15a4c835 TH |
5319 | css->id = 1; |
5320 | } else { | |
5321 | css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2, GFP_KERNEL); | |
5322 | BUG_ON(css->id < 0); | |
5323 | } | |
ddbcc7e8 | 5324 | |
e8d55fde | 5325 | /* Update the init_css_set to contain a subsys |
817929ec | 5326 | * pointer to this state - since the subsystem is |
e8d55fde | 5327 | * newly registered, all tasks and hence the |
3dd06ffa | 5328 | * init_css_set is in the subsystem's root cgroup. */ |
aec25020 | 5329 | init_css_set.subsys[ss->id] = css; |
ddbcc7e8 | 5330 | |
cb4a3167 AS |
5331 | have_fork_callback |= (bool)ss->fork << ss->id; |
5332 | have_exit_callback |= (bool)ss->exit << ss->id; | |
afcf6c8b | 5333 | have_free_callback |= (bool)ss->free << ss->id; |
7e47682e | 5334 | have_canfork_callback |= (bool)ss->can_fork << ss->id; |
ddbcc7e8 | 5335 | |
e8d55fde LZ |
5336 | /* At system boot, before all subsystems have been |
5337 | * registered, no tasks have been forked, so we don't | |
5338 | * need to invoke fork callbacks here. */ | |
5339 | BUG_ON(!list_empty(&init_task.tasks)); | |
5340 | ||
ae7f164a | 5341 | BUG_ON(online_css(css)); |
a8638030 | 5342 | |
cf5d5941 BB |
5343 | mutex_unlock(&cgroup_mutex); |
5344 | } | |
cf5d5941 | 5345 | |
ddbcc7e8 | 5346 | /** |
a043e3b2 LZ |
5347 | * cgroup_init_early - cgroup initialization at system boot |
5348 | * | |
5349 | * Initialize cgroups at system boot, and initialize any | |
5350 | * subsystems that request early init. | |
ddbcc7e8 PM |
5351 | */ |
5352 | int __init cgroup_init_early(void) | |
5353 | { | |
7b9a6ba5 | 5354 | static struct cgroup_sb_opts __initdata opts; |
30159ec7 | 5355 | struct cgroup_subsys *ss; |
ddbcc7e8 | 5356 | int i; |
30159ec7 | 5357 | |
3dd06ffa | 5358 | init_cgroup_root(&cgrp_dfl_root, &opts); |
3b514d24 TH |
5359 | cgrp_dfl_root.cgrp.self.flags |= CSS_NO_REF; |
5360 | ||
a4ea1cc9 | 5361 | RCU_INIT_POINTER(init_task.cgroups, &init_css_set); |
817929ec | 5362 | |
3ed80a62 | 5363 | for_each_subsys(ss, i) { |
aec25020 | 5364 | WARN(!ss->css_alloc || !ss->css_free || ss->name || ss->id, |
63253ad8 | 5365 | "invalid cgroup_subsys %d:%s css_alloc=%p css_free=%p id:name=%d:%s\n", |
073219e9 | 5366 | i, cgroup_subsys_name[i], ss->css_alloc, ss->css_free, |
aec25020 | 5367 | ss->id, ss->name); |
073219e9 TH |
5368 | WARN(strlen(cgroup_subsys_name[i]) > MAX_CGROUP_TYPE_NAMELEN, |
5369 | "cgroup_subsys_name %s too long\n", cgroup_subsys_name[i]); | |
5370 | ||
aec25020 | 5371 | ss->id = i; |
073219e9 | 5372 | ss->name = cgroup_subsys_name[i]; |
3e1d2eed TH |
5373 | if (!ss->legacy_name) |
5374 | ss->legacy_name = cgroup_subsys_name[i]; | |
ddbcc7e8 PM |
5375 | |
5376 | if (ss->early_init) | |
15a4c835 | 5377 | cgroup_init_subsys(ss, true); |
ddbcc7e8 PM |
5378 | } |
5379 | return 0; | |
5380 | } | |
5381 | ||
6e5c8307 | 5382 | static u16 cgroup_disable_mask __initdata; |
a3e72739 | 5383 | |
ddbcc7e8 | 5384 | /** |
a043e3b2 LZ |
5385 | * cgroup_init - cgroup initialization |
5386 | * | |
5387 | * Register cgroup filesystem and /proc file, and initialize | |
5388 | * any subsystems that didn't request early init. | |
ddbcc7e8 PM |
5389 | */ |
5390 | int __init cgroup_init(void) | |
5391 | { | |
30159ec7 | 5392 | struct cgroup_subsys *ss; |
035f4f51 | 5393 | int ssid; |
ddbcc7e8 | 5394 | |
6e5c8307 | 5395 | BUILD_BUG_ON(CGROUP_SUBSYS_COUNT > 16); |
1ed13287 | 5396 | BUG_ON(percpu_init_rwsem(&cgroup_threadgroup_rwsem)); |
a14c6874 TH |
5397 | BUG_ON(cgroup_init_cftypes(NULL, cgroup_dfl_base_files)); |
5398 | BUG_ON(cgroup_init_cftypes(NULL, cgroup_legacy_base_files)); | |
ddbcc7e8 | 5399 | |
54e7b4eb | 5400 | mutex_lock(&cgroup_mutex); |
54e7b4eb | 5401 | |
2378d8b8 TH |
5402 | /* |
5403 | * Add init_css_set to the hash table so that dfl_root can link to | |
5404 | * it during init. | |
5405 | */ | |
5406 | hash_add(css_set_table, &init_css_set.hlist, | |
5407 | css_set_hash(init_css_set.subsys)); | |
82fe9b0d | 5408 | |
3dd06ffa | 5409 | BUG_ON(cgroup_setup_root(&cgrp_dfl_root, 0)); |
4e96ee8e | 5410 | |
54e7b4eb TH |
5411 | mutex_unlock(&cgroup_mutex); |
5412 | ||
172a2c06 | 5413 | for_each_subsys(ss, ssid) { |
15a4c835 TH |
5414 | if (ss->early_init) { |
5415 | struct cgroup_subsys_state *css = | |
5416 | init_css_set.subsys[ss->id]; | |
5417 | ||
5418 | css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2, | |
5419 | GFP_KERNEL); | |
5420 | BUG_ON(css->id < 0); | |
5421 | } else { | |
5422 | cgroup_init_subsys(ss, false); | |
5423 | } | |
172a2c06 | 5424 | |
2d8f243a TH |
5425 | list_add_tail(&init_css_set.e_cset_node[ssid], |
5426 | &cgrp_dfl_root.cgrp.e_csets[ssid]); | |
172a2c06 TH |
5427 | |
5428 | /* | |
c731ae1d LZ |
5429 | * Setting dfl_root subsys_mask needs to consider the |
5430 | * disabled flag and cftype registration needs kmalloc, | |
5431 | * both of which aren't available during early_init. | |
172a2c06 | 5432 | */ |
a3e72739 TH |
5433 | if (cgroup_disable_mask & (1 << ssid)) { |
5434 | static_branch_disable(cgroup_subsys_enabled_key[ssid]); | |
5435 | printk(KERN_INFO "Disabling %s control group subsystem\n", | |
5436 | ss->name); | |
a8ddc821 | 5437 | continue; |
a3e72739 | 5438 | } |
a8ddc821 | 5439 | |
223ffb29 JW |
5440 | if (cgroup_ssid_no_v1(ssid)) |
5441 | printk(KERN_INFO "Disabling %s control group subsystem in v1 mounts\n", | |
5442 | ss->name); | |
5443 | ||
a8ddc821 TH |
5444 | cgrp_dfl_root.subsys_mask |= 1 << ss->id; |
5445 | ||
5de4fa13 | 5446 | if (!ss->dfl_cftypes) |
a7165264 | 5447 | cgrp_dfl_inhibit_ss_mask |= 1 << ss->id; |
5de4fa13 | 5448 | |
a8ddc821 TH |
5449 | if (ss->dfl_cftypes == ss->legacy_cftypes) { |
5450 | WARN_ON(cgroup_add_cftypes(ss, ss->dfl_cftypes)); | |
5451 | } else { | |
5452 | WARN_ON(cgroup_add_dfl_cftypes(ss, ss->dfl_cftypes)); | |
5453 | WARN_ON(cgroup_add_legacy_cftypes(ss, ss->legacy_cftypes)); | |
c731ae1d | 5454 | } |
295458e6 VD |
5455 | |
5456 | if (ss->bind) | |
5457 | ss->bind(init_css_set.subsys[ssid]); | |
676db4af GKH |
5458 | } |
5459 | ||
2378d8b8 TH |
5460 | /* init_css_set.subsys[] has been updated, re-hash */ |
5461 | hash_del(&init_css_set.hlist); | |
5462 | hash_add(css_set_table, &init_css_set.hlist, | |
5463 | css_set_hash(init_css_set.subsys)); | |
5464 | ||
035f4f51 TH |
5465 | WARN_ON(sysfs_create_mount_point(fs_kobj, "cgroup")); |
5466 | WARN_ON(register_filesystem(&cgroup_fs_type)); | |
67e9c74b | 5467 | WARN_ON(register_filesystem(&cgroup2_fs_type)); |
035f4f51 | 5468 | WARN_ON(!proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations)); |
ddbcc7e8 | 5469 | |
2bd59d48 | 5470 | return 0; |
ddbcc7e8 | 5471 | } |
b4f48b63 | 5472 | |
e5fca243 TH |
5473 | static int __init cgroup_wq_init(void) |
5474 | { | |
5475 | /* | |
5476 | * There isn't much point in executing destruction path in | |
5477 | * parallel. Good chunk is serialized with cgroup_mutex anyway. | |
1a11533f | 5478 | * Use 1 for @max_active. |
e5fca243 TH |
5479 | * |
5480 | * We would prefer to do this in cgroup_init() above, but that | |
5481 | * is called before init_workqueues(): so leave this until after. | |
5482 | */ | |
1a11533f | 5483 | cgroup_destroy_wq = alloc_workqueue("cgroup_destroy", 0, 1); |
e5fca243 | 5484 | BUG_ON(!cgroup_destroy_wq); |
b1a21367 TH |
5485 | |
5486 | /* | |
5487 | * Used to destroy pidlists and separate to serve as flush domain. | |
5488 | * Cap @max_active to 1 too. | |
5489 | */ | |
5490 | cgroup_pidlist_destroy_wq = alloc_workqueue("cgroup_pidlist_destroy", | |
5491 | 0, 1); | |
5492 | BUG_ON(!cgroup_pidlist_destroy_wq); | |
5493 | ||
e5fca243 TH |
5494 | return 0; |
5495 | } | |
5496 | core_initcall(cgroup_wq_init); | |
5497 | ||
a424316c PM |
5498 | /* |
5499 | * proc_cgroup_show() | |
5500 | * - Print task's cgroup paths into seq_file, one line for each hierarchy | |
5501 | * - Used for /proc/<pid>/cgroup. | |
a424316c | 5502 | */ |
006f4ac4 ZL |
5503 | int proc_cgroup_show(struct seq_file *m, struct pid_namespace *ns, |
5504 | struct pid *pid, struct task_struct *tsk) | |
a424316c | 5505 | { |
e61734c5 | 5506 | char *buf, *path; |
a424316c | 5507 | int retval; |
3dd06ffa | 5508 | struct cgroup_root *root; |
a424316c PM |
5509 | |
5510 | retval = -ENOMEM; | |
e61734c5 | 5511 | buf = kmalloc(PATH_MAX, GFP_KERNEL); |
a424316c PM |
5512 | if (!buf) |
5513 | goto out; | |
5514 | ||
a424316c | 5515 | mutex_lock(&cgroup_mutex); |
f0d9a5f1 | 5516 | spin_lock_bh(&css_set_lock); |
a424316c | 5517 | |
985ed670 | 5518 | for_each_root(root) { |
a424316c | 5519 | struct cgroup_subsys *ss; |
bd89aabc | 5520 | struct cgroup *cgrp; |
b85d2040 | 5521 | int ssid, count = 0; |
a424316c | 5522 | |
a7165264 | 5523 | if (root == &cgrp_dfl_root && !cgrp_dfl_visible) |
985ed670 TH |
5524 | continue; |
5525 | ||
2c6ab6d2 | 5526 | seq_printf(m, "%d:", root->hierarchy_id); |
d98817d4 TH |
5527 | if (root != &cgrp_dfl_root) |
5528 | for_each_subsys(ss, ssid) | |
5529 | if (root->subsys_mask & (1 << ssid)) | |
5530 | seq_printf(m, "%s%s", count++ ? "," : "", | |
3e1d2eed | 5531 | ss->legacy_name); |
c6d57f33 PM |
5532 | if (strlen(root->name)) |
5533 | seq_printf(m, "%sname=%s", count ? "," : "", | |
5534 | root->name); | |
a424316c | 5535 | seq_putc(m, ':'); |
2e91fa7f | 5536 | |
7717f7ba | 5537 | cgrp = task_cgroup_from_root(tsk, root); |
2e91fa7f TH |
5538 | |
5539 | /* | |
5540 | * On traditional hierarchies, all zombie tasks show up as | |
5541 | * belonging to the root cgroup. On the default hierarchy, | |
5542 | * while a zombie doesn't show up in "cgroup.procs" and | |
5543 | * thus can't be migrated, its /proc/PID/cgroup keeps | |
5544 | * reporting the cgroup it belonged to before exiting. If | |
5545 | * the cgroup is removed before the zombie is reaped, | |
5546 | * " (deleted)" is appended to the cgroup path. | |
5547 | */ | |
5548 | if (cgroup_on_dfl(cgrp) || !(tsk->flags & PF_EXITING)) { | |
5549 | path = cgroup_path(cgrp, buf, PATH_MAX); | |
5550 | if (!path) { | |
5551 | retval = -ENAMETOOLONG; | |
5552 | goto out_unlock; | |
5553 | } | |
5554 | } else { | |
5555 | path = "/"; | |
e61734c5 | 5556 | } |
2e91fa7f | 5557 | |
e61734c5 | 5558 | seq_puts(m, path); |
2e91fa7f TH |
5559 | |
5560 | if (cgroup_on_dfl(cgrp) && cgroup_is_dead(cgrp)) | |
5561 | seq_puts(m, " (deleted)\n"); | |
5562 | else | |
5563 | seq_putc(m, '\n'); | |
a424316c PM |
5564 | } |
5565 | ||
006f4ac4 | 5566 | retval = 0; |
a424316c | 5567 | out_unlock: |
f0d9a5f1 | 5568 | spin_unlock_bh(&css_set_lock); |
a424316c | 5569 | mutex_unlock(&cgroup_mutex); |
a424316c PM |
5570 | kfree(buf); |
5571 | out: | |
5572 | return retval; | |
5573 | } | |
5574 | ||
a424316c PM |
5575 | /* Display information about each subsystem and each hierarchy */ |
5576 | static int proc_cgroupstats_show(struct seq_file *m, void *v) | |
5577 | { | |
30159ec7 | 5578 | struct cgroup_subsys *ss; |
a424316c | 5579 | int i; |
a424316c | 5580 | |
8bab8dde | 5581 | seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n"); |
aae8aab4 BB |
5582 | /* |
5583 | * ideally we don't want subsystems moving around while we do this. | |
5584 | * cgroup_mutex is also necessary to guarantee an atomic snapshot of | |
5585 | * subsys/hierarchy state. | |
5586 | */ | |
a424316c | 5587 | mutex_lock(&cgroup_mutex); |
30159ec7 TH |
5588 | |
5589 | for_each_subsys(ss, i) | |
2c6ab6d2 | 5590 | seq_printf(m, "%s\t%d\t%d\t%d\n", |
3e1d2eed | 5591 | ss->legacy_name, ss->root->hierarchy_id, |
fc5ed1e9 TH |
5592 | atomic_read(&ss->root->nr_cgrps), |
5593 | cgroup_ssid_enabled(i)); | |
30159ec7 | 5594 | |
a424316c PM |
5595 | mutex_unlock(&cgroup_mutex); |
5596 | return 0; | |
5597 | } | |
5598 | ||
5599 | static int cgroupstats_open(struct inode *inode, struct file *file) | |
5600 | { | |
9dce07f1 | 5601 | return single_open(file, proc_cgroupstats_show, NULL); |
a424316c PM |
5602 | } |
5603 | ||
828c0950 | 5604 | static const struct file_operations proc_cgroupstats_operations = { |
a424316c PM |
5605 | .open = cgroupstats_open, |
5606 | .read = seq_read, | |
5607 | .llseek = seq_lseek, | |
5608 | .release = single_release, | |
5609 | }; | |
5610 | ||
b4f48b63 | 5611 | /** |
eaf797ab | 5612 | * cgroup_fork - initialize cgroup related fields during copy_process() |
a043e3b2 | 5613 | * @child: pointer to task_struct of forking parent process. |
b4f48b63 | 5614 | * |
eaf797ab TH |
5615 | * A task is associated with the init_css_set until cgroup_post_fork() |
5616 | * attaches it to the parent's css_set. Empty cg_list indicates that | |
5617 | * @child isn't holding reference to its css_set. | |
b4f48b63 PM |
5618 | */ |
5619 | void cgroup_fork(struct task_struct *child) | |
5620 | { | |
eaf797ab | 5621 | RCU_INIT_POINTER(child->cgroups, &init_css_set); |
817929ec | 5622 | INIT_LIST_HEAD(&child->cg_list); |
b4f48b63 PM |
5623 | } |
5624 | ||
7e47682e AS |
5625 | /** |
5626 | * cgroup_can_fork - called on a new task before the process is exposed | |
5627 | * @child: the task in question. | |
5628 | * | |
5629 | * This calls the subsystem can_fork() callbacks. If the can_fork() callback | |
5630 | * returns an error, the fork aborts with that error code. This allows for | |
5631 | * a cgroup subsystem to conditionally allow or deny new forks. | |
5632 | */ | |
b53202e6 | 5633 | int cgroup_can_fork(struct task_struct *child) |
7e47682e AS |
5634 | { |
5635 | struct cgroup_subsys *ss; | |
5636 | int i, j, ret; | |
5637 | ||
b4e0eeaf | 5638 | do_each_subsys_mask(ss, i, have_canfork_callback) { |
b53202e6 | 5639 | ret = ss->can_fork(child); |
7e47682e AS |
5640 | if (ret) |
5641 | goto out_revert; | |
b4e0eeaf | 5642 | } while_each_subsys_mask(); |
7e47682e AS |
5643 | |
5644 | return 0; | |
5645 | ||
5646 | out_revert: | |
5647 | for_each_subsys(ss, j) { | |
5648 | if (j >= i) | |
5649 | break; | |
5650 | if (ss->cancel_fork) | |
b53202e6 | 5651 | ss->cancel_fork(child); |
7e47682e AS |
5652 | } |
5653 | ||
5654 | return ret; | |
5655 | } | |
5656 | ||
5657 | /** | |
5658 | * cgroup_cancel_fork - called if a fork failed after cgroup_can_fork() | |
5659 | * @child: the task in question | |
5660 | * | |
5661 | * This calls the cancel_fork() callbacks if a fork failed *after* | |
5662 | * cgroup_can_fork() succeded. | |
5663 | */ | |
b53202e6 | 5664 | void cgroup_cancel_fork(struct task_struct *child) |
7e47682e AS |
5665 | { |
5666 | struct cgroup_subsys *ss; | |
5667 | int i; | |
5668 | ||
5669 | for_each_subsys(ss, i) | |
5670 | if (ss->cancel_fork) | |
b53202e6 | 5671 | ss->cancel_fork(child); |
7e47682e AS |
5672 | } |
5673 | ||
817929ec | 5674 | /** |
a043e3b2 LZ |
5675 | * cgroup_post_fork - called on a new task after adding it to the task list |
5676 | * @child: the task in question | |
5677 | * | |
5edee61e TH |
5678 | * Adds the task to the list running through its css_set if necessary and |
5679 | * call the subsystem fork() callbacks. Has to be after the task is | |
5680 | * visible on the task list in case we race with the first call to | |
0942eeee | 5681 | * cgroup_task_iter_start() - to guarantee that the new task ends up on its |
5edee61e | 5682 | * list. |
a043e3b2 | 5683 | */ |
b53202e6 | 5684 | void cgroup_post_fork(struct task_struct *child) |
817929ec | 5685 | { |
30159ec7 | 5686 | struct cgroup_subsys *ss; |
5edee61e TH |
5687 | int i; |
5688 | ||
3ce3230a | 5689 | /* |
251f8c03 | 5690 | * This may race against cgroup_enable_task_cg_lists(). As that |
eaf797ab TH |
5691 | * function sets use_task_css_set_links before grabbing |
5692 | * tasklist_lock and we just went through tasklist_lock to add | |
5693 | * @child, it's guaranteed that either we see the set | |
5694 | * use_task_css_set_links or cgroup_enable_task_cg_lists() sees | |
5695 | * @child during its iteration. | |
5696 | * | |
5697 | * If we won the race, @child is associated with %current's | |
f0d9a5f1 | 5698 | * css_set. Grabbing css_set_lock guarantees both that the |
eaf797ab TH |
5699 | * association is stable, and, on completion of the parent's |
5700 | * migration, @child is visible in the source of migration or | |
5701 | * already in the destination cgroup. This guarantee is necessary | |
5702 | * when implementing operations which need to migrate all tasks of | |
5703 | * a cgroup to another. | |
5704 | * | |
251f8c03 | 5705 | * Note that if we lose to cgroup_enable_task_cg_lists(), @child |
eaf797ab TH |
5706 | * will remain in init_css_set. This is safe because all tasks are |
5707 | * in the init_css_set before cg_links is enabled and there's no | |
5708 | * operation which transfers all tasks out of init_css_set. | |
3ce3230a | 5709 | */ |
817929ec | 5710 | if (use_task_css_set_links) { |
eaf797ab TH |
5711 | struct css_set *cset; |
5712 | ||
f0d9a5f1 | 5713 | spin_lock_bh(&css_set_lock); |
0e1d768f | 5714 | cset = task_css_set(current); |
eaf797ab | 5715 | if (list_empty(&child->cg_list)) { |
eaf797ab | 5716 | get_css_set(cset); |
f6d7d049 | 5717 | css_set_move_task(child, NULL, cset, false); |
eaf797ab | 5718 | } |
f0d9a5f1 | 5719 | spin_unlock_bh(&css_set_lock); |
817929ec | 5720 | } |
5edee61e TH |
5721 | |
5722 | /* | |
5723 | * Call ss->fork(). This must happen after @child is linked on | |
5724 | * css_set; otherwise, @child might change state between ->fork() | |
5725 | * and addition to css_set. | |
5726 | */ | |
b4e0eeaf | 5727 | do_each_subsys_mask(ss, i, have_fork_callback) { |
b53202e6 | 5728 | ss->fork(child); |
b4e0eeaf | 5729 | } while_each_subsys_mask(); |
817929ec | 5730 | } |
5edee61e | 5731 | |
b4f48b63 PM |
5732 | /** |
5733 | * cgroup_exit - detach cgroup from exiting task | |
5734 | * @tsk: pointer to task_struct of exiting process | |
5735 | * | |
5736 | * Description: Detach cgroup from @tsk and release it. | |
5737 | * | |
5738 | * Note that cgroups marked notify_on_release force every task in | |
5739 | * them to take the global cgroup_mutex mutex when exiting. | |
5740 | * This could impact scaling on very large systems. Be reluctant to | |
5741 | * use notify_on_release cgroups where very high task exit scaling | |
5742 | * is required on large systems. | |
5743 | * | |
0e1d768f TH |
5744 | * We set the exiting tasks cgroup to the root cgroup (top_cgroup). We |
5745 | * call cgroup_exit() while the task is still competent to handle | |
5746 | * notify_on_release(), then leave the task attached to the root cgroup in | |
5747 | * each hierarchy for the remainder of its exit. No need to bother with | |
5748 | * init_css_set refcnting. init_css_set never goes away and we can't race | |
e8604cb4 | 5749 | * with migration path - PF_EXITING is visible to migration path. |
b4f48b63 | 5750 | */ |
1ec41830 | 5751 | void cgroup_exit(struct task_struct *tsk) |
b4f48b63 | 5752 | { |
30159ec7 | 5753 | struct cgroup_subsys *ss; |
5abb8855 | 5754 | struct css_set *cset; |
d41d5a01 | 5755 | int i; |
817929ec PM |
5756 | |
5757 | /* | |
0e1d768f | 5758 | * Unlink from @tsk from its css_set. As migration path can't race |
0de0942d | 5759 | * with us, we can check css_set and cg_list without synchronization. |
817929ec | 5760 | */ |
0de0942d TH |
5761 | cset = task_css_set(tsk); |
5762 | ||
817929ec | 5763 | if (!list_empty(&tsk->cg_list)) { |
f0d9a5f1 | 5764 | spin_lock_bh(&css_set_lock); |
f6d7d049 | 5765 | css_set_move_task(tsk, cset, NULL, false); |
f0d9a5f1 | 5766 | spin_unlock_bh(&css_set_lock); |
2e91fa7f TH |
5767 | } else { |
5768 | get_css_set(cset); | |
817929ec PM |
5769 | } |
5770 | ||
cb4a3167 | 5771 | /* see cgroup_post_fork() for details */ |
b4e0eeaf | 5772 | do_each_subsys_mask(ss, i, have_exit_callback) { |
2e91fa7f | 5773 | ss->exit(tsk); |
b4e0eeaf | 5774 | } while_each_subsys_mask(); |
2e91fa7f | 5775 | } |
30159ec7 | 5776 | |
2e91fa7f TH |
5777 | void cgroup_free(struct task_struct *task) |
5778 | { | |
5779 | struct css_set *cset = task_css_set(task); | |
afcf6c8b TH |
5780 | struct cgroup_subsys *ss; |
5781 | int ssid; | |
5782 | ||
b4e0eeaf | 5783 | do_each_subsys_mask(ss, ssid, have_free_callback) { |
afcf6c8b | 5784 | ss->free(task); |
b4e0eeaf | 5785 | } while_each_subsys_mask(); |
d41d5a01 | 5786 | |
2e91fa7f | 5787 | put_css_set(cset); |
b4f48b63 | 5788 | } |
697f4161 | 5789 | |
bd89aabc | 5790 | static void check_for_release(struct cgroup *cgrp) |
81a6a5cd | 5791 | { |
27bd4dbb | 5792 | if (notify_on_release(cgrp) && !cgroup_is_populated(cgrp) && |
971ff493 ZL |
5793 | !css_has_online_children(&cgrp->self) && !cgroup_is_dead(cgrp)) |
5794 | schedule_work(&cgrp->release_agent_work); | |
81a6a5cd PM |
5795 | } |
5796 | ||
81a6a5cd PM |
5797 | /* |
5798 | * Notify userspace when a cgroup is released, by running the | |
5799 | * configured release agent with the name of the cgroup (path | |
5800 | * relative to the root of cgroup file system) as the argument. | |
5801 | * | |
5802 | * Most likely, this user command will try to rmdir this cgroup. | |
5803 | * | |
5804 | * This races with the possibility that some other task will be | |
5805 | * attached to this cgroup before it is removed, or that some other | |
5806 | * user task will 'mkdir' a child cgroup of this cgroup. That's ok. | |
5807 | * The presumed 'rmdir' will fail quietly if this cgroup is no longer | |
5808 | * unused, and this cgroup will be reprieved from its death sentence, | |
5809 | * to continue to serve a useful existence. Next time it's released, | |
5810 | * we will get notified again, if it still has 'notify_on_release' set. | |
5811 | * | |
5812 | * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which | |
5813 | * means only wait until the task is successfully execve()'d. The | |
5814 | * separate release agent task is forked by call_usermodehelper(), | |
5815 | * then control in this thread returns here, without waiting for the | |
5816 | * release agent task. We don't bother to wait because the caller of | |
5817 | * this routine has no use for the exit status of the release agent | |
5818 | * task, so no sense holding our caller up for that. | |
81a6a5cd | 5819 | */ |
81a6a5cd PM |
5820 | static void cgroup_release_agent(struct work_struct *work) |
5821 | { | |
971ff493 ZL |
5822 | struct cgroup *cgrp = |
5823 | container_of(work, struct cgroup, release_agent_work); | |
5824 | char *pathbuf = NULL, *agentbuf = NULL, *path; | |
5825 | char *argv[3], *envp[3]; | |
5826 | ||
81a6a5cd | 5827 | mutex_lock(&cgroup_mutex); |
971ff493 ZL |
5828 | |
5829 | pathbuf = kmalloc(PATH_MAX, GFP_KERNEL); | |
5830 | agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL); | |
5831 | if (!pathbuf || !agentbuf) | |
5832 | goto out; | |
5833 | ||
5834 | path = cgroup_path(cgrp, pathbuf, PATH_MAX); | |
5835 | if (!path) | |
5836 | goto out; | |
5837 | ||
5838 | argv[0] = agentbuf; | |
5839 | argv[1] = path; | |
5840 | argv[2] = NULL; | |
5841 | ||
5842 | /* minimal command environment */ | |
5843 | envp[0] = "HOME=/"; | |
5844 | envp[1] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin"; | |
5845 | envp[2] = NULL; | |
5846 | ||
81a6a5cd | 5847 | mutex_unlock(&cgroup_mutex); |
971ff493 | 5848 | call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC); |
3e2cd91a | 5849 | goto out_free; |
971ff493 | 5850 | out: |
81a6a5cd | 5851 | mutex_unlock(&cgroup_mutex); |
3e2cd91a | 5852 | out_free: |
971ff493 ZL |
5853 | kfree(agentbuf); |
5854 | kfree(pathbuf); | |
81a6a5cd | 5855 | } |
8bab8dde PM |
5856 | |
5857 | static int __init cgroup_disable(char *str) | |
5858 | { | |
30159ec7 | 5859 | struct cgroup_subsys *ss; |
8bab8dde | 5860 | char *token; |
30159ec7 | 5861 | int i; |
8bab8dde PM |
5862 | |
5863 | while ((token = strsep(&str, ",")) != NULL) { | |
5864 | if (!*token) | |
5865 | continue; | |
be45c900 | 5866 | |
3ed80a62 | 5867 | for_each_subsys(ss, i) { |
3e1d2eed TH |
5868 | if (strcmp(token, ss->name) && |
5869 | strcmp(token, ss->legacy_name)) | |
5870 | continue; | |
a3e72739 | 5871 | cgroup_disable_mask |= 1 << i; |
8bab8dde PM |
5872 | } |
5873 | } | |
5874 | return 1; | |
5875 | } | |
5876 | __setup("cgroup_disable=", cgroup_disable); | |
38460b48 | 5877 | |
223ffb29 JW |
5878 | static int __init cgroup_no_v1(char *str) |
5879 | { | |
5880 | struct cgroup_subsys *ss; | |
5881 | char *token; | |
5882 | int i; | |
5883 | ||
5884 | while ((token = strsep(&str, ",")) != NULL) { | |
5885 | if (!*token) | |
5886 | continue; | |
5887 | ||
5888 | if (!strcmp(token, "all")) { | |
6e5c8307 | 5889 | cgroup_no_v1_mask = U16_MAX; |
223ffb29 JW |
5890 | break; |
5891 | } | |
5892 | ||
5893 | for_each_subsys(ss, i) { | |
5894 | if (strcmp(token, ss->name) && | |
5895 | strcmp(token, ss->legacy_name)) | |
5896 | continue; | |
5897 | ||
5898 | cgroup_no_v1_mask |= 1 << i; | |
5899 | } | |
5900 | } | |
5901 | return 1; | |
5902 | } | |
5903 | __setup("cgroup_no_v1=", cgroup_no_v1); | |
5904 | ||
b77d7b60 | 5905 | /** |
ec903c0c | 5906 | * css_tryget_online_from_dir - get corresponding css from a cgroup dentry |
35cf0836 TH |
5907 | * @dentry: directory dentry of interest |
5908 | * @ss: subsystem of interest | |
b77d7b60 | 5909 | * |
5a17f543 TH |
5910 | * If @dentry is a directory for a cgroup which has @ss enabled on it, try |
5911 | * to get the corresponding css and return it. If such css doesn't exist | |
5912 | * or can't be pinned, an ERR_PTR value is returned. | |
e5d1367f | 5913 | */ |
ec903c0c TH |
5914 | struct cgroup_subsys_state *css_tryget_online_from_dir(struct dentry *dentry, |
5915 | struct cgroup_subsys *ss) | |
e5d1367f | 5916 | { |
2bd59d48 | 5917 | struct kernfs_node *kn = kernfs_node_from_dentry(dentry); |
f17fc25f | 5918 | struct file_system_type *s_type = dentry->d_sb->s_type; |
2bd59d48 | 5919 | struct cgroup_subsys_state *css = NULL; |
e5d1367f | 5920 | struct cgroup *cgrp; |
e5d1367f | 5921 | |
35cf0836 | 5922 | /* is @dentry a cgroup dir? */ |
f17fc25f TH |
5923 | if ((s_type != &cgroup_fs_type && s_type != &cgroup2_fs_type) || |
5924 | !kn || kernfs_type(kn) != KERNFS_DIR) | |
e5d1367f SE |
5925 | return ERR_PTR(-EBADF); |
5926 | ||
5a17f543 TH |
5927 | rcu_read_lock(); |
5928 | ||
2bd59d48 TH |
5929 | /* |
5930 | * This path doesn't originate from kernfs and @kn could already | |
5931 | * have been or be removed at any point. @kn->priv is RCU | |
a4189487 | 5932 | * protected for this access. See css_release_work_fn() for details. |
2bd59d48 TH |
5933 | */ |
5934 | cgrp = rcu_dereference(kn->priv); | |
5935 | if (cgrp) | |
5936 | css = cgroup_css(cgrp, ss); | |
5a17f543 | 5937 | |
ec903c0c | 5938 | if (!css || !css_tryget_online(css)) |
5a17f543 TH |
5939 | css = ERR_PTR(-ENOENT); |
5940 | ||
5941 | rcu_read_unlock(); | |
5942 | return css; | |
e5d1367f | 5943 | } |
e5d1367f | 5944 | |
1cb650b9 LZ |
5945 | /** |
5946 | * css_from_id - lookup css by id | |
5947 | * @id: the cgroup id | |
5948 | * @ss: cgroup subsys to be looked into | |
5949 | * | |
5950 | * Returns the css if there's valid one with @id, otherwise returns NULL. | |
5951 | * Should be called under rcu_read_lock(). | |
5952 | */ | |
5953 | struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss) | |
5954 | { | |
6fa4918d | 5955 | WARN_ON_ONCE(!rcu_read_lock_held()); |
adbe427b | 5956 | return id > 0 ? idr_find(&ss->css_idr, id) : NULL; |
e5d1367f SE |
5957 | } |
5958 | ||
16af4396 TH |
5959 | /** |
5960 | * cgroup_get_from_path - lookup and get a cgroup from its default hierarchy path | |
5961 | * @path: path on the default hierarchy | |
5962 | * | |
5963 | * Find the cgroup at @path on the default hierarchy, increment its | |
5964 | * reference count and return it. Returns pointer to the found cgroup on | |
5965 | * success, ERR_PTR(-ENOENT) if @path doens't exist and ERR_PTR(-ENOTDIR) | |
5966 | * if @path points to a non-directory. | |
5967 | */ | |
5968 | struct cgroup *cgroup_get_from_path(const char *path) | |
5969 | { | |
5970 | struct kernfs_node *kn; | |
5971 | struct cgroup *cgrp; | |
5972 | ||
5973 | mutex_lock(&cgroup_mutex); | |
5974 | ||
5975 | kn = kernfs_walk_and_get(cgrp_dfl_root.cgrp.kn, path); | |
5976 | if (kn) { | |
5977 | if (kernfs_type(kn) == KERNFS_DIR) { | |
5978 | cgrp = kn->priv; | |
5979 | cgroup_get(cgrp); | |
5980 | } else { | |
5981 | cgrp = ERR_PTR(-ENOTDIR); | |
5982 | } | |
5983 | kernfs_put(kn); | |
5984 | } else { | |
5985 | cgrp = ERR_PTR(-ENOENT); | |
5986 | } | |
5987 | ||
5988 | mutex_unlock(&cgroup_mutex); | |
5989 | return cgrp; | |
5990 | } | |
5991 | EXPORT_SYMBOL_GPL(cgroup_get_from_path); | |
5992 | ||
bd1060a1 TH |
5993 | /* |
5994 | * sock->sk_cgrp_data handling. For more info, see sock_cgroup_data | |
5995 | * definition in cgroup-defs.h. | |
5996 | */ | |
5997 | #ifdef CONFIG_SOCK_CGROUP_DATA | |
5998 | ||
5999 | #if defined(CONFIG_CGROUP_NET_PRIO) || defined(CONFIG_CGROUP_NET_CLASSID) | |
6000 | ||
3fa4cc9c | 6001 | DEFINE_SPINLOCK(cgroup_sk_update_lock); |
bd1060a1 TH |
6002 | static bool cgroup_sk_alloc_disabled __read_mostly; |
6003 | ||
6004 | void cgroup_sk_alloc_disable(void) | |
6005 | { | |
6006 | if (cgroup_sk_alloc_disabled) | |
6007 | return; | |
6008 | pr_info("cgroup: disabling cgroup2 socket matching due to net_prio or net_cls activation\n"); | |
6009 | cgroup_sk_alloc_disabled = true; | |
6010 | } | |
6011 | ||
6012 | #else | |
6013 | ||
6014 | #define cgroup_sk_alloc_disabled false | |
6015 | ||
6016 | #endif | |
6017 | ||
6018 | void cgroup_sk_alloc(struct sock_cgroup_data *skcd) | |
6019 | { | |
6020 | if (cgroup_sk_alloc_disabled) | |
6021 | return; | |
6022 | ||
6023 | rcu_read_lock(); | |
6024 | ||
6025 | while (true) { | |
6026 | struct css_set *cset; | |
6027 | ||
6028 | cset = task_css_set(current); | |
6029 | if (likely(cgroup_tryget(cset->dfl_cgrp))) { | |
6030 | skcd->val = (unsigned long)cset->dfl_cgrp; | |
6031 | break; | |
6032 | } | |
6033 | cpu_relax(); | |
6034 | } | |
6035 | ||
6036 | rcu_read_unlock(); | |
6037 | } | |
6038 | ||
6039 | void cgroup_sk_free(struct sock_cgroup_data *skcd) | |
6040 | { | |
6041 | cgroup_put(sock_cgroup_ptr(skcd)); | |
6042 | } | |
6043 | ||
6044 | #endif /* CONFIG_SOCK_CGROUP_DATA */ | |
6045 | ||
fe693435 | 6046 | #ifdef CONFIG_CGROUP_DEBUG |
eb95419b TH |
6047 | static struct cgroup_subsys_state * |
6048 | debug_css_alloc(struct cgroup_subsys_state *parent_css) | |
fe693435 PM |
6049 | { |
6050 | struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL); | |
6051 | ||
6052 | if (!css) | |
6053 | return ERR_PTR(-ENOMEM); | |
6054 | ||
6055 | return css; | |
6056 | } | |
6057 | ||
eb95419b | 6058 | static void debug_css_free(struct cgroup_subsys_state *css) |
fe693435 | 6059 | { |
eb95419b | 6060 | kfree(css); |
fe693435 PM |
6061 | } |
6062 | ||
182446d0 TH |
6063 | static u64 debug_taskcount_read(struct cgroup_subsys_state *css, |
6064 | struct cftype *cft) | |
fe693435 | 6065 | { |
182446d0 | 6066 | return cgroup_task_count(css->cgroup); |
fe693435 PM |
6067 | } |
6068 | ||
182446d0 TH |
6069 | static u64 current_css_set_read(struct cgroup_subsys_state *css, |
6070 | struct cftype *cft) | |
fe693435 PM |
6071 | { |
6072 | return (u64)(unsigned long)current->cgroups; | |
6073 | } | |
6074 | ||
182446d0 | 6075 | static u64 current_css_set_refcount_read(struct cgroup_subsys_state *css, |
03c78cbe | 6076 | struct cftype *cft) |
fe693435 PM |
6077 | { |
6078 | u64 count; | |
6079 | ||
6080 | rcu_read_lock(); | |
a8ad805c | 6081 | count = atomic_read(&task_css_set(current)->refcount); |
fe693435 PM |
6082 | rcu_read_unlock(); |
6083 | return count; | |
6084 | } | |
6085 | ||
2da8ca82 | 6086 | static int current_css_set_cg_links_read(struct seq_file *seq, void *v) |
7717f7ba | 6087 | { |
69d0206c | 6088 | struct cgrp_cset_link *link; |
5abb8855 | 6089 | struct css_set *cset; |
e61734c5 TH |
6090 | char *name_buf; |
6091 | ||
6092 | name_buf = kmalloc(NAME_MAX + 1, GFP_KERNEL); | |
6093 | if (!name_buf) | |
6094 | return -ENOMEM; | |
7717f7ba | 6095 | |
f0d9a5f1 | 6096 | spin_lock_bh(&css_set_lock); |
7717f7ba | 6097 | rcu_read_lock(); |
5abb8855 | 6098 | cset = rcu_dereference(current->cgroups); |
69d0206c | 6099 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) { |
7717f7ba | 6100 | struct cgroup *c = link->cgrp; |
7717f7ba | 6101 | |
a2dd4247 | 6102 | cgroup_name(c, name_buf, NAME_MAX + 1); |
2c6ab6d2 | 6103 | seq_printf(seq, "Root %d group %s\n", |
a2dd4247 | 6104 | c->root->hierarchy_id, name_buf); |
7717f7ba PM |
6105 | } |
6106 | rcu_read_unlock(); | |
f0d9a5f1 | 6107 | spin_unlock_bh(&css_set_lock); |
e61734c5 | 6108 | kfree(name_buf); |
7717f7ba PM |
6109 | return 0; |
6110 | } | |
6111 | ||
6112 | #define MAX_TASKS_SHOWN_PER_CSS 25 | |
2da8ca82 | 6113 | static int cgroup_css_links_read(struct seq_file *seq, void *v) |
7717f7ba | 6114 | { |
2da8ca82 | 6115 | struct cgroup_subsys_state *css = seq_css(seq); |
69d0206c | 6116 | struct cgrp_cset_link *link; |
7717f7ba | 6117 | |
f0d9a5f1 | 6118 | spin_lock_bh(&css_set_lock); |
182446d0 | 6119 | list_for_each_entry(link, &css->cgroup->cset_links, cset_link) { |
69d0206c | 6120 | struct css_set *cset = link->cset; |
7717f7ba PM |
6121 | struct task_struct *task; |
6122 | int count = 0; | |
c7561128 | 6123 | |
5abb8855 | 6124 | seq_printf(seq, "css_set %p\n", cset); |
c7561128 | 6125 | |
5abb8855 | 6126 | list_for_each_entry(task, &cset->tasks, cg_list) { |
c7561128 TH |
6127 | if (count++ > MAX_TASKS_SHOWN_PER_CSS) |
6128 | goto overflow; | |
6129 | seq_printf(seq, " task %d\n", task_pid_vnr(task)); | |
6130 | } | |
6131 | ||
6132 | list_for_each_entry(task, &cset->mg_tasks, cg_list) { | |
6133 | if (count++ > MAX_TASKS_SHOWN_PER_CSS) | |
6134 | goto overflow; | |
6135 | seq_printf(seq, " task %d\n", task_pid_vnr(task)); | |
7717f7ba | 6136 | } |
c7561128 TH |
6137 | continue; |
6138 | overflow: | |
6139 | seq_puts(seq, " ...\n"); | |
7717f7ba | 6140 | } |
f0d9a5f1 | 6141 | spin_unlock_bh(&css_set_lock); |
7717f7ba PM |
6142 | return 0; |
6143 | } | |
6144 | ||
182446d0 | 6145 | static u64 releasable_read(struct cgroup_subsys_state *css, struct cftype *cft) |
fe693435 | 6146 | { |
27bd4dbb | 6147 | return (!cgroup_is_populated(css->cgroup) && |
a25eb52e | 6148 | !css_has_online_children(&css->cgroup->self)); |
fe693435 PM |
6149 | } |
6150 | ||
6151 | static struct cftype debug_files[] = { | |
fe693435 PM |
6152 | { |
6153 | .name = "taskcount", | |
6154 | .read_u64 = debug_taskcount_read, | |
6155 | }, | |
6156 | ||
6157 | { | |
6158 | .name = "current_css_set", | |
6159 | .read_u64 = current_css_set_read, | |
6160 | }, | |
6161 | ||
6162 | { | |
6163 | .name = "current_css_set_refcount", | |
6164 | .read_u64 = current_css_set_refcount_read, | |
6165 | }, | |
6166 | ||
7717f7ba PM |
6167 | { |
6168 | .name = "current_css_set_cg_links", | |
2da8ca82 | 6169 | .seq_show = current_css_set_cg_links_read, |
7717f7ba PM |
6170 | }, |
6171 | ||
6172 | { | |
6173 | .name = "cgroup_css_links", | |
2da8ca82 | 6174 | .seq_show = cgroup_css_links_read, |
7717f7ba PM |
6175 | }, |
6176 | ||
fe693435 PM |
6177 | { |
6178 | .name = "releasable", | |
6179 | .read_u64 = releasable_read, | |
6180 | }, | |
fe693435 | 6181 | |
4baf6e33 TH |
6182 | { } /* terminate */ |
6183 | }; | |
fe693435 | 6184 | |
073219e9 | 6185 | struct cgroup_subsys debug_cgrp_subsys = { |
92fb9748 TH |
6186 | .css_alloc = debug_css_alloc, |
6187 | .css_free = debug_css_free, | |
5577964e | 6188 | .legacy_cftypes = debug_files, |
fe693435 PM |
6189 | }; |
6190 | #endif /* CONFIG_CGROUP_DEBUG */ |