]>
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 | |
f6d635ad TH |
189 | /* some controllers are implicitly enabled on the default hierarchy */ |
190 | static unsigned long cgrp_dfl_implicit_ss_mask; | |
191 | ||
ddbcc7e8 PM |
192 | /* The list of hierarchy roots */ |
193 | ||
9871bf95 TH |
194 | static LIST_HEAD(cgroup_roots); |
195 | static int cgroup_root_count; | |
ddbcc7e8 | 196 | |
3417ae1f | 197 | /* hierarchy ID allocation and mapping, protected by cgroup_mutex */ |
1a574231 | 198 | static DEFINE_IDR(cgroup_hierarchy_idr); |
2c6ab6d2 | 199 | |
794611a1 | 200 | /* |
0cb51d71 TH |
201 | * Assign a monotonically increasing serial number to csses. It guarantees |
202 | * cgroups with bigger numbers are newer than those with smaller numbers. | |
203 | * Also, as csses are always appended to the parent's ->children list, it | |
204 | * guarantees that sibling csses are always sorted in the ascending serial | |
205 | * number order on the list. Protected by cgroup_mutex. | |
794611a1 | 206 | */ |
0cb51d71 | 207 | static u64 css_serial_nr_next = 1; |
794611a1 | 208 | |
cb4a3167 AS |
209 | /* |
210 | * These bitmask flags indicate whether tasks in the fork and exit paths have | |
211 | * fork/exit handlers to call. This avoids us having to do extra work in the | |
212 | * fork/exit path to check which subsystems have fork/exit callbacks. | |
ddbcc7e8 | 213 | */ |
6e5c8307 TH |
214 | static u16 have_fork_callback __read_mostly; |
215 | static u16 have_exit_callback __read_mostly; | |
216 | static u16 have_free_callback __read_mostly; | |
ddbcc7e8 | 217 | |
7e47682e | 218 | /* Ditto for the can_fork callback. */ |
6e5c8307 | 219 | static u16 have_canfork_callback __read_mostly; |
7e47682e | 220 | |
67e9c74b | 221 | static struct file_system_type cgroup2_fs_type; |
a14c6874 TH |
222 | static struct cftype cgroup_dfl_base_files[]; |
223 | static struct cftype cgroup_legacy_base_files[]; | |
628f7cd4 | 224 | |
6e5c8307 | 225 | static int rebind_subsystems(struct cgroup_root *dst_root, u16 ss_mask); |
945ba199 | 226 | static void cgroup_lock_and_drain_offline(struct cgroup *cgrp); |
334c3679 TH |
227 | static int cgroup_apply_control(struct cgroup *cgrp); |
228 | static void cgroup_finalize_control(struct cgroup *cgrp, int ret); | |
ed27b9f7 | 229 | static void css_task_iter_advance(struct css_task_iter *it); |
42809dd4 | 230 | static int cgroup_destroy_locked(struct cgroup *cgrp); |
6cd0f5bb TH |
231 | static struct cgroup_subsys_state *css_create(struct cgroup *cgrp, |
232 | struct cgroup_subsys *ss); | |
9d755d33 | 233 | static void css_release(struct percpu_ref *ref); |
f8f22e53 | 234 | static void kill_css(struct cgroup_subsys_state *css); |
4df8dc90 TH |
235 | static int cgroup_addrm_files(struct cgroup_subsys_state *css, |
236 | struct cgroup *cgrp, struct cftype cfts[], | |
2bb566cb | 237 | bool is_add); |
42809dd4 | 238 | |
fc5ed1e9 TH |
239 | /** |
240 | * cgroup_ssid_enabled - cgroup subsys enabled test by subsys ID | |
241 | * @ssid: subsys ID of interest | |
242 | * | |
243 | * cgroup_subsys_enabled() can only be used with literal subsys names which | |
244 | * is fine for individual subsystems but unsuitable for cgroup core. This | |
245 | * is slower static_key_enabled() based test indexed by @ssid. | |
246 | */ | |
247 | static bool cgroup_ssid_enabled(int ssid) | |
248 | { | |
249 | return static_key_enabled(cgroup_subsys_enabled_key[ssid]); | |
250 | } | |
251 | ||
223ffb29 JW |
252 | static bool cgroup_ssid_no_v1(int ssid) |
253 | { | |
254 | return cgroup_no_v1_mask & (1 << ssid); | |
255 | } | |
256 | ||
9e10a130 TH |
257 | /** |
258 | * cgroup_on_dfl - test whether a cgroup is on the default hierarchy | |
259 | * @cgrp: the cgroup of interest | |
260 | * | |
261 | * The default hierarchy is the v2 interface of cgroup and this function | |
262 | * can be used to test whether a cgroup is on the default hierarchy for | |
263 | * cases where a subsystem should behave differnetly depending on the | |
264 | * interface version. | |
265 | * | |
266 | * The set of behaviors which change on the default hierarchy are still | |
267 | * being determined and the mount option is prefixed with __DEVEL__. | |
268 | * | |
269 | * List of changed behaviors: | |
270 | * | |
271 | * - Mount options "noprefix", "xattr", "clone_children", "release_agent" | |
272 | * and "name" are disallowed. | |
273 | * | |
274 | * - When mounting an existing superblock, mount options should match. | |
275 | * | |
276 | * - Remount is disallowed. | |
277 | * | |
278 | * - rename(2) is disallowed. | |
279 | * | |
280 | * - "tasks" is removed. Everything should be at process granularity. Use | |
281 | * "cgroup.procs" instead. | |
282 | * | |
283 | * - "cgroup.procs" is not sorted. pids will be unique unless they got | |
284 | * recycled inbetween reads. | |
285 | * | |
286 | * - "release_agent" and "notify_on_release" are removed. Replacement | |
287 | * notification mechanism will be implemented. | |
288 | * | |
289 | * - "cgroup.clone_children" is removed. | |
290 | * | |
291 | * - "cgroup.subtree_populated" is available. Its value is 0 if the cgroup | |
292 | * and its descendants contain no task; otherwise, 1. The file also | |
293 | * generates kernfs notification which can be monitored through poll and | |
294 | * [di]notify when the value of the file changes. | |
295 | * | |
296 | * - cpuset: tasks will be kept in empty cpusets when hotplug happens and | |
297 | * take masks of ancestors with non-empty cpus/mems, instead of being | |
298 | * moved to an ancestor. | |
299 | * | |
300 | * - cpuset: a task can be moved into an empty cpuset, and again it takes | |
301 | * masks of ancestors. | |
302 | * | |
303 | * - memcg: use_hierarchy is on by default and the cgroup file for the flag | |
304 | * is not created. | |
305 | * | |
306 | * - blkcg: blk-throttle becomes properly hierarchical. | |
307 | * | |
308 | * - debug: disallowed on the default hierarchy. | |
309 | */ | |
310 | static bool cgroup_on_dfl(const struct cgroup *cgrp) | |
311 | { | |
312 | return cgrp->root == &cgrp_dfl_root; | |
313 | } | |
314 | ||
6fa4918d TH |
315 | /* IDR wrappers which synchronize using cgroup_idr_lock */ |
316 | static int cgroup_idr_alloc(struct idr *idr, void *ptr, int start, int end, | |
317 | gfp_t gfp_mask) | |
318 | { | |
319 | int ret; | |
320 | ||
321 | idr_preload(gfp_mask); | |
54504e97 | 322 | spin_lock_bh(&cgroup_idr_lock); |
d0164adc | 323 | ret = idr_alloc(idr, ptr, start, end, gfp_mask & ~__GFP_DIRECT_RECLAIM); |
54504e97 | 324 | spin_unlock_bh(&cgroup_idr_lock); |
6fa4918d TH |
325 | idr_preload_end(); |
326 | return ret; | |
327 | } | |
328 | ||
329 | static void *cgroup_idr_replace(struct idr *idr, void *ptr, int id) | |
330 | { | |
331 | void *ret; | |
332 | ||
54504e97 | 333 | spin_lock_bh(&cgroup_idr_lock); |
6fa4918d | 334 | ret = idr_replace(idr, ptr, id); |
54504e97 | 335 | spin_unlock_bh(&cgroup_idr_lock); |
6fa4918d TH |
336 | return ret; |
337 | } | |
338 | ||
339 | static void cgroup_idr_remove(struct idr *idr, int id) | |
340 | { | |
54504e97 | 341 | spin_lock_bh(&cgroup_idr_lock); |
6fa4918d | 342 | idr_remove(idr, id); |
54504e97 | 343 | spin_unlock_bh(&cgroup_idr_lock); |
6fa4918d TH |
344 | } |
345 | ||
d51f39b0 TH |
346 | static struct cgroup *cgroup_parent(struct cgroup *cgrp) |
347 | { | |
348 | struct cgroup_subsys_state *parent_css = cgrp->self.parent; | |
349 | ||
350 | if (parent_css) | |
351 | return container_of(parent_css, struct cgroup, self); | |
352 | return NULL; | |
353 | } | |
354 | ||
5531dc91 TH |
355 | /* subsystems visibly enabled on a cgroup */ |
356 | static u16 cgroup_control(struct cgroup *cgrp) | |
357 | { | |
358 | struct cgroup *parent = cgroup_parent(cgrp); | |
359 | u16 root_ss_mask = cgrp->root->subsys_mask; | |
360 | ||
361 | if (parent) | |
362 | return parent->subtree_control; | |
363 | ||
364 | if (cgroup_on_dfl(cgrp)) | |
f6d635ad TH |
365 | root_ss_mask &= ~(cgrp_dfl_inhibit_ss_mask | |
366 | cgrp_dfl_implicit_ss_mask); | |
5531dc91 TH |
367 | return root_ss_mask; |
368 | } | |
369 | ||
370 | /* subsystems enabled on a cgroup */ | |
371 | static u16 cgroup_ss_mask(struct cgroup *cgrp) | |
372 | { | |
373 | struct cgroup *parent = cgroup_parent(cgrp); | |
374 | ||
375 | if (parent) | |
376 | return parent->subtree_ss_mask; | |
377 | ||
378 | return cgrp->root->subsys_mask; | |
379 | } | |
380 | ||
95109b62 TH |
381 | /** |
382 | * cgroup_css - obtain a cgroup's css for the specified subsystem | |
383 | * @cgrp: the cgroup of interest | |
9d800df1 | 384 | * @ss: the subsystem of interest (%NULL returns @cgrp->self) |
95109b62 | 385 | * |
ca8bdcaf TH |
386 | * Return @cgrp's css (cgroup_subsys_state) associated with @ss. This |
387 | * function must be called either under cgroup_mutex or rcu_read_lock() and | |
388 | * the caller is responsible for pinning the returned css if it wants to | |
389 | * keep accessing it outside the said locks. This function may return | |
390 | * %NULL if @cgrp doesn't have @subsys_id enabled. | |
95109b62 TH |
391 | */ |
392 | static struct cgroup_subsys_state *cgroup_css(struct cgroup *cgrp, | |
ca8bdcaf | 393 | struct cgroup_subsys *ss) |
95109b62 | 394 | { |
ca8bdcaf | 395 | if (ss) |
aec25020 | 396 | return rcu_dereference_check(cgrp->subsys[ss->id], |
ace2bee8 | 397 | lockdep_is_held(&cgroup_mutex)); |
ca8bdcaf | 398 | else |
9d800df1 | 399 | return &cgrp->self; |
95109b62 | 400 | } |
42809dd4 | 401 | |
aec3dfcb TH |
402 | /** |
403 | * cgroup_e_css - obtain a cgroup's effective css for the specified subsystem | |
404 | * @cgrp: the cgroup of interest | |
9d800df1 | 405 | * @ss: the subsystem of interest (%NULL returns @cgrp->self) |
aec3dfcb | 406 | * |
d0f702e6 | 407 | * Similar to cgroup_css() but returns the effective css, which is defined |
aec3dfcb TH |
408 | * as the matching css of the nearest ancestor including self which has @ss |
409 | * enabled. If @ss is associated with the hierarchy @cgrp is on, this | |
410 | * function is guaranteed to return non-NULL css. | |
411 | */ | |
412 | static struct cgroup_subsys_state *cgroup_e_css(struct cgroup *cgrp, | |
413 | struct cgroup_subsys *ss) | |
414 | { | |
415 | lockdep_assert_held(&cgroup_mutex); | |
416 | ||
417 | if (!ss) | |
9d800df1 | 418 | return &cgrp->self; |
aec3dfcb | 419 | |
eeecbd19 TH |
420 | /* |
421 | * This function is used while updating css associations and thus | |
5531dc91 | 422 | * can't test the csses directly. Test ss_mask. |
eeecbd19 | 423 | */ |
5531dc91 | 424 | while (!(cgroup_ss_mask(cgrp) & (1 << ss->id))) { |
d51f39b0 | 425 | cgrp = cgroup_parent(cgrp); |
5531dc91 TH |
426 | if (!cgrp) |
427 | return NULL; | |
428 | } | |
aec3dfcb TH |
429 | |
430 | return cgroup_css(cgrp, ss); | |
95109b62 | 431 | } |
42809dd4 | 432 | |
eeecbd19 TH |
433 | /** |
434 | * cgroup_get_e_css - get a cgroup's effective css for the specified subsystem | |
435 | * @cgrp: the cgroup of interest | |
436 | * @ss: the subsystem of interest | |
437 | * | |
438 | * Find and get the effective css of @cgrp for @ss. The effective css is | |
439 | * defined as the matching css of the nearest ancestor including self which | |
440 | * has @ss enabled. If @ss is not mounted on the hierarchy @cgrp is on, | |
441 | * the root css is returned, so this function always returns a valid css. | |
442 | * The returned css must be put using css_put(). | |
443 | */ | |
444 | struct cgroup_subsys_state *cgroup_get_e_css(struct cgroup *cgrp, | |
445 | struct cgroup_subsys *ss) | |
446 | { | |
447 | struct cgroup_subsys_state *css; | |
448 | ||
449 | rcu_read_lock(); | |
450 | ||
451 | do { | |
452 | css = cgroup_css(cgrp, ss); | |
453 | ||
454 | if (css && css_tryget_online(css)) | |
455 | goto out_unlock; | |
456 | cgrp = cgroup_parent(cgrp); | |
457 | } while (cgrp); | |
458 | ||
459 | css = init_css_set.subsys[ss->id]; | |
460 | css_get(css); | |
461 | out_unlock: | |
462 | rcu_read_unlock(); | |
463 | return css; | |
464 | } | |
465 | ||
ddbcc7e8 | 466 | /* convenient tests for these bits */ |
54766d4a | 467 | static inline bool cgroup_is_dead(const struct cgroup *cgrp) |
ddbcc7e8 | 468 | { |
184faf32 | 469 | return !(cgrp->self.flags & CSS_ONLINE); |
ddbcc7e8 PM |
470 | } |
471 | ||
052c3f3a TH |
472 | static void cgroup_get(struct cgroup *cgrp) |
473 | { | |
474 | WARN_ON_ONCE(cgroup_is_dead(cgrp)); | |
475 | css_get(&cgrp->self); | |
476 | } | |
477 | ||
478 | static bool cgroup_tryget(struct cgroup *cgrp) | |
479 | { | |
480 | return css_tryget(&cgrp->self); | |
481 | } | |
482 | ||
b4168640 | 483 | struct cgroup_subsys_state *of_css(struct kernfs_open_file *of) |
59f5296b | 484 | { |
2bd59d48 | 485 | struct cgroup *cgrp = of->kn->parent->priv; |
b4168640 | 486 | struct cftype *cft = of_cft(of); |
2bd59d48 TH |
487 | |
488 | /* | |
489 | * This is open and unprotected implementation of cgroup_css(). | |
490 | * seq_css() is only called from a kernfs file operation which has | |
491 | * an active reference on the file. Because all the subsystem | |
492 | * files are drained before a css is disassociated with a cgroup, | |
493 | * the matching css from the cgroup's subsys table is guaranteed to | |
494 | * be and stay valid until the enclosing operation is complete. | |
495 | */ | |
496 | if (cft->ss) | |
497 | return rcu_dereference_raw(cgrp->subsys[cft->ss->id]); | |
498 | else | |
9d800df1 | 499 | return &cgrp->self; |
59f5296b | 500 | } |
b4168640 | 501 | EXPORT_SYMBOL_GPL(of_css); |
59f5296b | 502 | |
e9685a03 | 503 | static int notify_on_release(const struct cgroup *cgrp) |
81a6a5cd | 504 | { |
bd89aabc | 505 | return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); |
81a6a5cd PM |
506 | } |
507 | ||
1c6727af TH |
508 | /** |
509 | * for_each_css - iterate all css's of a cgroup | |
510 | * @css: the iteration cursor | |
511 | * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end | |
512 | * @cgrp: the target cgroup to iterate css's of | |
513 | * | |
aec3dfcb | 514 | * Should be called under cgroup_[tree_]mutex. |
1c6727af TH |
515 | */ |
516 | #define for_each_css(css, ssid, cgrp) \ | |
517 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \ | |
518 | if (!((css) = rcu_dereference_check( \ | |
519 | (cgrp)->subsys[(ssid)], \ | |
520 | lockdep_is_held(&cgroup_mutex)))) { } \ | |
521 | else | |
522 | ||
aec3dfcb TH |
523 | /** |
524 | * for_each_e_css - iterate all effective css's of a cgroup | |
525 | * @css: the iteration cursor | |
526 | * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end | |
527 | * @cgrp: the target cgroup to iterate css's of | |
528 | * | |
529 | * Should be called under cgroup_[tree_]mutex. | |
530 | */ | |
531 | #define for_each_e_css(css, ssid, cgrp) \ | |
532 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \ | |
533 | if (!((css) = cgroup_e_css(cgrp, cgroup_subsys[(ssid)]))) \ | |
534 | ; \ | |
535 | else | |
536 | ||
30159ec7 | 537 | /** |
3ed80a62 | 538 | * for_each_subsys - iterate all enabled cgroup subsystems |
30159ec7 | 539 | * @ss: the iteration cursor |
780cd8b3 | 540 | * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end |
30159ec7 | 541 | */ |
780cd8b3 | 542 | #define for_each_subsys(ss, ssid) \ |
3ed80a62 TH |
543 | for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT && \ |
544 | (((ss) = cgroup_subsys[ssid]) || true); (ssid)++) | |
30159ec7 | 545 | |
cb4a3167 | 546 | /** |
b4e0eeaf | 547 | * do_each_subsys_mask - filter for_each_subsys with a bitmask |
cb4a3167 AS |
548 | * @ss: the iteration cursor |
549 | * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end | |
b4e0eeaf | 550 | * @ss_mask: the bitmask |
cb4a3167 AS |
551 | * |
552 | * The block will only run for cases where the ssid-th bit (1 << ssid) of | |
b4e0eeaf | 553 | * @ss_mask is set. |
cb4a3167 | 554 | */ |
b4e0eeaf TH |
555 | #define do_each_subsys_mask(ss, ssid, ss_mask) do { \ |
556 | unsigned long __ss_mask = (ss_mask); \ | |
557 | if (!CGROUP_SUBSYS_COUNT) { /* to avoid spurious gcc warning */ \ | |
4a705c5c | 558 | (ssid) = 0; \ |
b4e0eeaf TH |
559 | break; \ |
560 | } \ | |
561 | for_each_set_bit(ssid, &__ss_mask, CGROUP_SUBSYS_COUNT) { \ | |
562 | (ss) = cgroup_subsys[ssid]; \ | |
563 | { | |
564 | ||
565 | #define while_each_subsys_mask() \ | |
566 | } \ | |
567 | } \ | |
568 | } while (false) | |
cb4a3167 | 569 | |
985ed670 TH |
570 | /* iterate across the hierarchies */ |
571 | #define for_each_root(root) \ | |
5549c497 | 572 | list_for_each_entry((root), &cgroup_roots, root_list) |
ddbcc7e8 | 573 | |
f8f22e53 TH |
574 | /* iterate over child cgrps, lock should be held throughout iteration */ |
575 | #define cgroup_for_each_live_child(child, cgrp) \ | |
d5c419b6 | 576 | list_for_each_entry((child), &(cgrp)->self.children, self.sibling) \ |
8353da1f | 577 | if (({ lockdep_assert_held(&cgroup_mutex); \ |
f8f22e53 TH |
578 | cgroup_is_dead(child); })) \ |
579 | ; \ | |
580 | else | |
7ae1bad9 | 581 | |
ce3f1d9d TH |
582 | /* walk live descendants in preorder */ |
583 | #define cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) \ | |
584 | css_for_each_descendant_pre((d_css), cgroup_css((cgrp), NULL)) \ | |
585 | if (({ lockdep_assert_held(&cgroup_mutex); \ | |
586 | (dsct) = (d_css)->cgroup; \ | |
587 | cgroup_is_dead(dsct); })) \ | |
588 | ; \ | |
589 | else | |
590 | ||
591 | /* walk live descendants in postorder */ | |
592 | #define cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) \ | |
593 | css_for_each_descendant_post((d_css), cgroup_css((cgrp), NULL)) \ | |
594 | if (({ lockdep_assert_held(&cgroup_mutex); \ | |
595 | (dsct) = (d_css)->cgroup; \ | |
596 | cgroup_is_dead(dsct); })) \ | |
597 | ; \ | |
598 | else | |
599 | ||
81a6a5cd | 600 | static void cgroup_release_agent(struct work_struct *work); |
bd89aabc | 601 | static void check_for_release(struct cgroup *cgrp); |
81a6a5cd | 602 | |
69d0206c TH |
603 | /* |
604 | * A cgroup can be associated with multiple css_sets as different tasks may | |
605 | * belong to different cgroups on different hierarchies. In the other | |
606 | * direction, a css_set is naturally associated with multiple cgroups. | |
607 | * This M:N relationship is represented by the following link structure | |
608 | * which exists for each association and allows traversing the associations | |
609 | * from both sides. | |
610 | */ | |
611 | struct cgrp_cset_link { | |
612 | /* the cgroup and css_set this link associates */ | |
613 | struct cgroup *cgrp; | |
614 | struct css_set *cset; | |
615 | ||
616 | /* list of cgrp_cset_links anchored at cgrp->cset_links */ | |
617 | struct list_head cset_link; | |
618 | ||
619 | /* list of cgrp_cset_links anchored at css_set->cgrp_links */ | |
620 | struct list_head cgrp_link; | |
817929ec PM |
621 | }; |
622 | ||
172a2c06 TH |
623 | /* |
624 | * The default css_set - used by init and its children prior to any | |
817929ec PM |
625 | * hierarchies being mounted. It contains a pointer to the root state |
626 | * for each subsystem. Also used to anchor the list of css_sets. Not | |
627 | * reference-counted, to improve performance when child cgroups | |
628 | * haven't been created. | |
629 | */ | |
5024ae29 | 630 | struct css_set init_css_set = { |
172a2c06 TH |
631 | .refcount = ATOMIC_INIT(1), |
632 | .cgrp_links = LIST_HEAD_INIT(init_css_set.cgrp_links), | |
633 | .tasks = LIST_HEAD_INIT(init_css_set.tasks), | |
634 | .mg_tasks = LIST_HEAD_INIT(init_css_set.mg_tasks), | |
635 | .mg_preload_node = LIST_HEAD_INIT(init_css_set.mg_preload_node), | |
636 | .mg_node = LIST_HEAD_INIT(init_css_set.mg_node), | |
ed27b9f7 | 637 | .task_iters = LIST_HEAD_INIT(init_css_set.task_iters), |
172a2c06 | 638 | }; |
817929ec | 639 | |
172a2c06 | 640 | static int css_set_count = 1; /* 1 for init_css_set */ |
817929ec | 641 | |
0de0942d TH |
642 | /** |
643 | * css_set_populated - does a css_set contain any tasks? | |
644 | * @cset: target css_set | |
645 | */ | |
646 | static bool css_set_populated(struct css_set *cset) | |
647 | { | |
f0d9a5f1 | 648 | lockdep_assert_held(&css_set_lock); |
0de0942d TH |
649 | |
650 | return !list_empty(&cset->tasks) || !list_empty(&cset->mg_tasks); | |
651 | } | |
652 | ||
842b597e TH |
653 | /** |
654 | * cgroup_update_populated - updated populated count of a cgroup | |
655 | * @cgrp: the target cgroup | |
656 | * @populated: inc or dec populated count | |
657 | * | |
0de0942d TH |
658 | * One of the css_sets associated with @cgrp is either getting its first |
659 | * task or losing the last. Update @cgrp->populated_cnt accordingly. The | |
660 | * count is propagated towards root so that a given cgroup's populated_cnt | |
661 | * is zero iff the cgroup and all its descendants don't contain any tasks. | |
842b597e TH |
662 | * |
663 | * @cgrp's interface file "cgroup.populated" is zero if | |
664 | * @cgrp->populated_cnt is zero and 1 otherwise. When @cgrp->populated_cnt | |
665 | * changes from or to zero, userland is notified that the content of the | |
666 | * interface file has changed. This can be used to detect when @cgrp and | |
667 | * its descendants become populated or empty. | |
668 | */ | |
669 | static void cgroup_update_populated(struct cgroup *cgrp, bool populated) | |
670 | { | |
f0d9a5f1 | 671 | lockdep_assert_held(&css_set_lock); |
842b597e TH |
672 | |
673 | do { | |
674 | bool trigger; | |
675 | ||
676 | if (populated) | |
677 | trigger = !cgrp->populated_cnt++; | |
678 | else | |
679 | trigger = !--cgrp->populated_cnt; | |
680 | ||
681 | if (!trigger) | |
682 | break; | |
683 | ||
ad2ed2b3 | 684 | check_for_release(cgrp); |
6f60eade TH |
685 | cgroup_file_notify(&cgrp->events_file); |
686 | ||
d51f39b0 | 687 | cgrp = cgroup_parent(cgrp); |
842b597e TH |
688 | } while (cgrp); |
689 | } | |
690 | ||
0de0942d TH |
691 | /** |
692 | * css_set_update_populated - update populated state of a css_set | |
693 | * @cset: target css_set | |
694 | * @populated: whether @cset is populated or depopulated | |
695 | * | |
696 | * @cset is either getting the first task or losing the last. Update the | |
697 | * ->populated_cnt of all associated cgroups accordingly. | |
698 | */ | |
699 | static void css_set_update_populated(struct css_set *cset, bool populated) | |
700 | { | |
701 | struct cgrp_cset_link *link; | |
702 | ||
f0d9a5f1 | 703 | lockdep_assert_held(&css_set_lock); |
0de0942d TH |
704 | |
705 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) | |
706 | cgroup_update_populated(link->cgrp, populated); | |
707 | } | |
708 | ||
f6d7d049 TH |
709 | /** |
710 | * css_set_move_task - move a task from one css_set to another | |
711 | * @task: task being moved | |
712 | * @from_cset: css_set @task currently belongs to (may be NULL) | |
713 | * @to_cset: new css_set @task is being moved to (may be NULL) | |
714 | * @use_mg_tasks: move to @to_cset->mg_tasks instead of ->tasks | |
715 | * | |
716 | * Move @task from @from_cset to @to_cset. If @task didn't belong to any | |
717 | * css_set, @from_cset can be NULL. If @task is being disassociated | |
718 | * instead of moved, @to_cset can be NULL. | |
719 | * | |
ed27b9f7 TH |
720 | * This function automatically handles populated_cnt updates and |
721 | * css_task_iter adjustments but the caller is responsible for managing | |
722 | * @from_cset and @to_cset's reference counts. | |
f6d7d049 TH |
723 | */ |
724 | static void css_set_move_task(struct task_struct *task, | |
725 | struct css_set *from_cset, struct css_set *to_cset, | |
726 | bool use_mg_tasks) | |
727 | { | |
f0d9a5f1 | 728 | lockdep_assert_held(&css_set_lock); |
f6d7d049 | 729 | |
20b454a6 TH |
730 | if (to_cset && !css_set_populated(to_cset)) |
731 | css_set_update_populated(to_cset, true); | |
732 | ||
f6d7d049 | 733 | if (from_cset) { |
ed27b9f7 TH |
734 | struct css_task_iter *it, *pos; |
735 | ||
f6d7d049 | 736 | WARN_ON_ONCE(list_empty(&task->cg_list)); |
ed27b9f7 TH |
737 | |
738 | /* | |
739 | * @task is leaving, advance task iterators which are | |
740 | * pointing to it so that they can resume at the next | |
741 | * position. Advancing an iterator might remove it from | |
742 | * the list, use safe walk. See css_task_iter_advance*() | |
743 | * for details. | |
744 | */ | |
745 | list_for_each_entry_safe(it, pos, &from_cset->task_iters, | |
746 | iters_node) | |
747 | if (it->task_pos == &task->cg_list) | |
748 | css_task_iter_advance(it); | |
749 | ||
f6d7d049 TH |
750 | list_del_init(&task->cg_list); |
751 | if (!css_set_populated(from_cset)) | |
752 | css_set_update_populated(from_cset, false); | |
753 | } else { | |
754 | WARN_ON_ONCE(!list_empty(&task->cg_list)); | |
755 | } | |
756 | ||
757 | if (to_cset) { | |
758 | /* | |
759 | * We are synchronized through cgroup_threadgroup_rwsem | |
760 | * against PF_EXITING setting such that we can't race | |
761 | * against cgroup_exit() changing the css_set to | |
762 | * init_css_set and dropping the old one. | |
763 | */ | |
764 | WARN_ON_ONCE(task->flags & PF_EXITING); | |
765 | ||
f6d7d049 TH |
766 | rcu_assign_pointer(task->cgroups, to_cset); |
767 | list_add_tail(&task->cg_list, use_mg_tasks ? &to_cset->mg_tasks : | |
768 | &to_cset->tasks); | |
769 | } | |
770 | } | |
771 | ||
7717f7ba PM |
772 | /* |
773 | * hash table for cgroup groups. This improves the performance to find | |
774 | * an existing css_set. This hash doesn't (currently) take into | |
775 | * account cgroups in empty hierarchies. | |
776 | */ | |
472b1053 | 777 | #define CSS_SET_HASH_BITS 7 |
0ac801fe | 778 | static DEFINE_HASHTABLE(css_set_table, CSS_SET_HASH_BITS); |
472b1053 | 779 | |
0ac801fe | 780 | static unsigned long css_set_hash(struct cgroup_subsys_state *css[]) |
472b1053 | 781 | { |
0ac801fe | 782 | unsigned long key = 0UL; |
30159ec7 TH |
783 | struct cgroup_subsys *ss; |
784 | int i; | |
472b1053 | 785 | |
30159ec7 | 786 | for_each_subsys(ss, i) |
0ac801fe LZ |
787 | key += (unsigned long)css[i]; |
788 | key = (key >> 16) ^ key; | |
472b1053 | 789 | |
0ac801fe | 790 | return key; |
472b1053 LZ |
791 | } |
792 | ||
a25eb52e | 793 | static void put_css_set_locked(struct css_set *cset) |
b4f48b63 | 794 | { |
69d0206c | 795 | struct cgrp_cset_link *link, *tmp_link; |
2d8f243a TH |
796 | struct cgroup_subsys *ss; |
797 | int ssid; | |
5abb8855 | 798 | |
f0d9a5f1 | 799 | lockdep_assert_held(&css_set_lock); |
89c5509b TH |
800 | |
801 | if (!atomic_dec_and_test(&cset->refcount)) | |
146aa1bd | 802 | return; |
81a6a5cd | 803 | |
53254f90 TH |
804 | /* This css_set is dead. unlink it and release cgroup and css refs */ |
805 | for_each_subsys(ss, ssid) { | |
2d8f243a | 806 | list_del(&cset->e_cset_node[ssid]); |
53254f90 TH |
807 | css_put(cset->subsys[ssid]); |
808 | } | |
5abb8855 | 809 | hash_del(&cset->hlist); |
2c6ab6d2 PM |
810 | css_set_count--; |
811 | ||
69d0206c | 812 | list_for_each_entry_safe(link, tmp_link, &cset->cgrp_links, cgrp_link) { |
69d0206c TH |
813 | list_del(&link->cset_link); |
814 | list_del(&link->cgrp_link); | |
2ceb231b TH |
815 | if (cgroup_parent(link->cgrp)) |
816 | cgroup_put(link->cgrp); | |
2c6ab6d2 | 817 | kfree(link); |
81a6a5cd | 818 | } |
2c6ab6d2 | 819 | |
5abb8855 | 820 | kfree_rcu(cset, rcu_head); |
b4f48b63 PM |
821 | } |
822 | ||
a25eb52e | 823 | static void put_css_set(struct css_set *cset) |
89c5509b TH |
824 | { |
825 | /* | |
826 | * Ensure that the refcount doesn't hit zero while any readers | |
827 | * can see it. Similar to atomic_dec_and_lock(), but for an | |
828 | * rwlock | |
829 | */ | |
830 | if (atomic_add_unless(&cset->refcount, -1, 1)) | |
831 | return; | |
832 | ||
f0d9a5f1 | 833 | spin_lock_bh(&css_set_lock); |
a25eb52e | 834 | put_css_set_locked(cset); |
f0d9a5f1 | 835 | spin_unlock_bh(&css_set_lock); |
89c5509b TH |
836 | } |
837 | ||
817929ec PM |
838 | /* |
839 | * refcounted get/put for css_set objects | |
840 | */ | |
5abb8855 | 841 | static inline void get_css_set(struct css_set *cset) |
817929ec | 842 | { |
5abb8855 | 843 | atomic_inc(&cset->refcount); |
817929ec PM |
844 | } |
845 | ||
b326f9d0 | 846 | /** |
7717f7ba | 847 | * compare_css_sets - helper function for find_existing_css_set(). |
5abb8855 TH |
848 | * @cset: candidate css_set being tested |
849 | * @old_cset: existing css_set for a task | |
7717f7ba PM |
850 | * @new_cgrp: cgroup that's being entered by the task |
851 | * @template: desired set of css pointers in css_set (pre-calculated) | |
852 | * | |
6f4b7e63 | 853 | * Returns true if "cset" matches "old_cset" except for the hierarchy |
7717f7ba PM |
854 | * which "new_cgrp" belongs to, for which it should match "new_cgrp". |
855 | */ | |
5abb8855 TH |
856 | static bool compare_css_sets(struct css_set *cset, |
857 | struct css_set *old_cset, | |
7717f7ba PM |
858 | struct cgroup *new_cgrp, |
859 | struct cgroup_subsys_state *template[]) | |
860 | { | |
861 | struct list_head *l1, *l2; | |
862 | ||
aec3dfcb TH |
863 | /* |
864 | * On the default hierarchy, there can be csets which are | |
865 | * associated with the same set of cgroups but different csses. | |
866 | * Let's first ensure that csses match. | |
867 | */ | |
868 | if (memcmp(template, cset->subsys, sizeof(cset->subsys))) | |
7717f7ba | 869 | return false; |
7717f7ba PM |
870 | |
871 | /* | |
872 | * Compare cgroup pointers in order to distinguish between | |
aec3dfcb TH |
873 | * different cgroups in hierarchies. As different cgroups may |
874 | * share the same effective css, this comparison is always | |
875 | * necessary. | |
7717f7ba | 876 | */ |
69d0206c TH |
877 | l1 = &cset->cgrp_links; |
878 | l2 = &old_cset->cgrp_links; | |
7717f7ba | 879 | while (1) { |
69d0206c | 880 | struct cgrp_cset_link *link1, *link2; |
5abb8855 | 881 | struct cgroup *cgrp1, *cgrp2; |
7717f7ba PM |
882 | |
883 | l1 = l1->next; | |
884 | l2 = l2->next; | |
885 | /* See if we reached the end - both lists are equal length. */ | |
69d0206c TH |
886 | if (l1 == &cset->cgrp_links) { |
887 | BUG_ON(l2 != &old_cset->cgrp_links); | |
7717f7ba PM |
888 | break; |
889 | } else { | |
69d0206c | 890 | BUG_ON(l2 == &old_cset->cgrp_links); |
7717f7ba PM |
891 | } |
892 | /* Locate the cgroups associated with these links. */ | |
69d0206c TH |
893 | link1 = list_entry(l1, struct cgrp_cset_link, cgrp_link); |
894 | link2 = list_entry(l2, struct cgrp_cset_link, cgrp_link); | |
895 | cgrp1 = link1->cgrp; | |
896 | cgrp2 = link2->cgrp; | |
7717f7ba | 897 | /* Hierarchies should be linked in the same order. */ |
5abb8855 | 898 | BUG_ON(cgrp1->root != cgrp2->root); |
7717f7ba PM |
899 | |
900 | /* | |
901 | * If this hierarchy is the hierarchy of the cgroup | |
902 | * that's changing, then we need to check that this | |
903 | * css_set points to the new cgroup; if it's any other | |
904 | * hierarchy, then this css_set should point to the | |
905 | * same cgroup as the old css_set. | |
906 | */ | |
5abb8855 TH |
907 | if (cgrp1->root == new_cgrp->root) { |
908 | if (cgrp1 != new_cgrp) | |
7717f7ba PM |
909 | return false; |
910 | } else { | |
5abb8855 | 911 | if (cgrp1 != cgrp2) |
7717f7ba PM |
912 | return false; |
913 | } | |
914 | } | |
915 | return true; | |
916 | } | |
917 | ||
b326f9d0 TH |
918 | /** |
919 | * find_existing_css_set - init css array and find the matching css_set | |
920 | * @old_cset: the css_set that we're using before the cgroup transition | |
921 | * @cgrp: the cgroup that we're moving into | |
922 | * @template: out param for the new set of csses, should be clear on entry | |
817929ec | 923 | */ |
5abb8855 TH |
924 | static struct css_set *find_existing_css_set(struct css_set *old_cset, |
925 | struct cgroup *cgrp, | |
926 | struct cgroup_subsys_state *template[]) | |
b4f48b63 | 927 | { |
3dd06ffa | 928 | struct cgroup_root *root = cgrp->root; |
30159ec7 | 929 | struct cgroup_subsys *ss; |
5abb8855 | 930 | struct css_set *cset; |
0ac801fe | 931 | unsigned long key; |
b326f9d0 | 932 | int i; |
817929ec | 933 | |
aae8aab4 BB |
934 | /* |
935 | * Build the set of subsystem state objects that we want to see in the | |
936 | * new css_set. while subsystems can change globally, the entries here | |
937 | * won't change, so no need for locking. | |
938 | */ | |
30159ec7 | 939 | for_each_subsys(ss, i) { |
f392e51c | 940 | if (root->subsys_mask & (1UL << i)) { |
aec3dfcb TH |
941 | /* |
942 | * @ss is in this hierarchy, so we want the | |
943 | * effective css from @cgrp. | |
944 | */ | |
945 | template[i] = cgroup_e_css(cgrp, ss); | |
817929ec | 946 | } else { |
aec3dfcb TH |
947 | /* |
948 | * @ss is not in this hierarchy, so we don't want | |
949 | * to change the css. | |
950 | */ | |
5abb8855 | 951 | template[i] = old_cset->subsys[i]; |
817929ec PM |
952 | } |
953 | } | |
954 | ||
0ac801fe | 955 | key = css_set_hash(template); |
5abb8855 TH |
956 | hash_for_each_possible(css_set_table, cset, hlist, key) { |
957 | if (!compare_css_sets(cset, old_cset, cgrp, template)) | |
7717f7ba PM |
958 | continue; |
959 | ||
960 | /* This css_set matches what we need */ | |
5abb8855 | 961 | return cset; |
472b1053 | 962 | } |
817929ec PM |
963 | |
964 | /* No existing cgroup group matched */ | |
965 | return NULL; | |
966 | } | |
967 | ||
69d0206c | 968 | static void free_cgrp_cset_links(struct list_head *links_to_free) |
36553434 | 969 | { |
69d0206c | 970 | struct cgrp_cset_link *link, *tmp_link; |
36553434 | 971 | |
69d0206c TH |
972 | list_for_each_entry_safe(link, tmp_link, links_to_free, cset_link) { |
973 | list_del(&link->cset_link); | |
36553434 LZ |
974 | kfree(link); |
975 | } | |
976 | } | |
977 | ||
69d0206c TH |
978 | /** |
979 | * allocate_cgrp_cset_links - allocate cgrp_cset_links | |
980 | * @count: the number of links to allocate | |
981 | * @tmp_links: list_head the allocated links are put on | |
982 | * | |
983 | * Allocate @count cgrp_cset_link structures and chain them on @tmp_links | |
984 | * through ->cset_link. Returns 0 on success or -errno. | |
817929ec | 985 | */ |
69d0206c | 986 | static int allocate_cgrp_cset_links(int count, struct list_head *tmp_links) |
817929ec | 987 | { |
69d0206c | 988 | struct cgrp_cset_link *link; |
817929ec | 989 | int i; |
69d0206c TH |
990 | |
991 | INIT_LIST_HEAD(tmp_links); | |
992 | ||
817929ec | 993 | for (i = 0; i < count; i++) { |
f4f4be2b | 994 | link = kzalloc(sizeof(*link), GFP_KERNEL); |
817929ec | 995 | if (!link) { |
69d0206c | 996 | free_cgrp_cset_links(tmp_links); |
817929ec PM |
997 | return -ENOMEM; |
998 | } | |
69d0206c | 999 | list_add(&link->cset_link, tmp_links); |
817929ec PM |
1000 | } |
1001 | return 0; | |
1002 | } | |
1003 | ||
c12f65d4 LZ |
1004 | /** |
1005 | * link_css_set - a helper function to link a css_set to a cgroup | |
69d0206c | 1006 | * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links() |
5abb8855 | 1007 | * @cset: the css_set to be linked |
c12f65d4 LZ |
1008 | * @cgrp: the destination cgroup |
1009 | */ | |
69d0206c TH |
1010 | static void link_css_set(struct list_head *tmp_links, struct css_set *cset, |
1011 | struct cgroup *cgrp) | |
c12f65d4 | 1012 | { |
69d0206c | 1013 | struct cgrp_cset_link *link; |
c12f65d4 | 1014 | |
69d0206c | 1015 | BUG_ON(list_empty(tmp_links)); |
6803c006 TH |
1016 | |
1017 | if (cgroup_on_dfl(cgrp)) | |
1018 | cset->dfl_cgrp = cgrp; | |
1019 | ||
69d0206c TH |
1020 | link = list_first_entry(tmp_links, struct cgrp_cset_link, cset_link); |
1021 | link->cset = cset; | |
7717f7ba | 1022 | link->cgrp = cgrp; |
842b597e | 1023 | |
7717f7ba | 1024 | /* |
389b9c1b TH |
1025 | * Always add links to the tail of the lists so that the lists are |
1026 | * in choronological order. | |
7717f7ba | 1027 | */ |
389b9c1b | 1028 | list_move_tail(&link->cset_link, &cgrp->cset_links); |
69d0206c | 1029 | list_add_tail(&link->cgrp_link, &cset->cgrp_links); |
2ceb231b TH |
1030 | |
1031 | if (cgroup_parent(cgrp)) | |
1032 | cgroup_get(cgrp); | |
c12f65d4 LZ |
1033 | } |
1034 | ||
b326f9d0 TH |
1035 | /** |
1036 | * find_css_set - return a new css_set with one cgroup updated | |
1037 | * @old_cset: the baseline css_set | |
1038 | * @cgrp: the cgroup to be updated | |
1039 | * | |
1040 | * Return a new css_set that's equivalent to @old_cset, but with @cgrp | |
1041 | * substituted into the appropriate hierarchy. | |
817929ec | 1042 | */ |
5abb8855 TH |
1043 | static struct css_set *find_css_set(struct css_set *old_cset, |
1044 | struct cgroup *cgrp) | |
817929ec | 1045 | { |
b326f9d0 | 1046 | struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT] = { }; |
5abb8855 | 1047 | struct css_set *cset; |
69d0206c TH |
1048 | struct list_head tmp_links; |
1049 | struct cgrp_cset_link *link; | |
2d8f243a | 1050 | struct cgroup_subsys *ss; |
0ac801fe | 1051 | unsigned long key; |
2d8f243a | 1052 | int ssid; |
472b1053 | 1053 | |
b326f9d0 TH |
1054 | lockdep_assert_held(&cgroup_mutex); |
1055 | ||
817929ec PM |
1056 | /* First see if we already have a cgroup group that matches |
1057 | * the desired set */ | |
f0d9a5f1 | 1058 | spin_lock_bh(&css_set_lock); |
5abb8855 TH |
1059 | cset = find_existing_css_set(old_cset, cgrp, template); |
1060 | if (cset) | |
1061 | get_css_set(cset); | |
f0d9a5f1 | 1062 | spin_unlock_bh(&css_set_lock); |
817929ec | 1063 | |
5abb8855 TH |
1064 | if (cset) |
1065 | return cset; | |
817929ec | 1066 | |
f4f4be2b | 1067 | cset = kzalloc(sizeof(*cset), GFP_KERNEL); |
5abb8855 | 1068 | if (!cset) |
817929ec PM |
1069 | return NULL; |
1070 | ||
69d0206c | 1071 | /* Allocate all the cgrp_cset_link objects that we'll need */ |
9871bf95 | 1072 | if (allocate_cgrp_cset_links(cgroup_root_count, &tmp_links) < 0) { |
5abb8855 | 1073 | kfree(cset); |
817929ec PM |
1074 | return NULL; |
1075 | } | |
1076 | ||
5abb8855 | 1077 | atomic_set(&cset->refcount, 1); |
69d0206c | 1078 | INIT_LIST_HEAD(&cset->cgrp_links); |
5abb8855 | 1079 | INIT_LIST_HEAD(&cset->tasks); |
c7561128 | 1080 | INIT_LIST_HEAD(&cset->mg_tasks); |
1958d2d5 | 1081 | INIT_LIST_HEAD(&cset->mg_preload_node); |
b3dc094e | 1082 | INIT_LIST_HEAD(&cset->mg_node); |
ed27b9f7 | 1083 | INIT_LIST_HEAD(&cset->task_iters); |
5abb8855 | 1084 | INIT_HLIST_NODE(&cset->hlist); |
817929ec PM |
1085 | |
1086 | /* Copy the set of subsystem state objects generated in | |
1087 | * find_existing_css_set() */ | |
5abb8855 | 1088 | memcpy(cset->subsys, template, sizeof(cset->subsys)); |
817929ec | 1089 | |
f0d9a5f1 | 1090 | spin_lock_bh(&css_set_lock); |
817929ec | 1091 | /* Add reference counts and links from the new css_set. */ |
69d0206c | 1092 | list_for_each_entry(link, &old_cset->cgrp_links, cgrp_link) { |
7717f7ba | 1093 | struct cgroup *c = link->cgrp; |
69d0206c | 1094 | |
7717f7ba PM |
1095 | if (c->root == cgrp->root) |
1096 | c = cgrp; | |
69d0206c | 1097 | link_css_set(&tmp_links, cset, c); |
7717f7ba | 1098 | } |
817929ec | 1099 | |
69d0206c | 1100 | BUG_ON(!list_empty(&tmp_links)); |
817929ec | 1101 | |
817929ec | 1102 | css_set_count++; |
472b1053 | 1103 | |
2d8f243a | 1104 | /* Add @cset to the hash table */ |
5abb8855 TH |
1105 | key = css_set_hash(cset->subsys); |
1106 | hash_add(css_set_table, &cset->hlist, key); | |
472b1053 | 1107 | |
53254f90 TH |
1108 | for_each_subsys(ss, ssid) { |
1109 | struct cgroup_subsys_state *css = cset->subsys[ssid]; | |
1110 | ||
2d8f243a | 1111 | list_add_tail(&cset->e_cset_node[ssid], |
53254f90 TH |
1112 | &css->cgroup->e_csets[ssid]); |
1113 | css_get(css); | |
1114 | } | |
2d8f243a | 1115 | |
f0d9a5f1 | 1116 | spin_unlock_bh(&css_set_lock); |
817929ec | 1117 | |
5abb8855 | 1118 | return cset; |
b4f48b63 PM |
1119 | } |
1120 | ||
3dd06ffa | 1121 | static struct cgroup_root *cgroup_root_from_kf(struct kernfs_root *kf_root) |
7717f7ba | 1122 | { |
3dd06ffa | 1123 | struct cgroup *root_cgrp = kf_root->kn->priv; |
2bd59d48 | 1124 | |
3dd06ffa | 1125 | return root_cgrp->root; |
2bd59d48 TH |
1126 | } |
1127 | ||
3dd06ffa | 1128 | static int cgroup_init_root_id(struct cgroup_root *root) |
f2e85d57 TH |
1129 | { |
1130 | int id; | |
1131 | ||
1132 | lockdep_assert_held(&cgroup_mutex); | |
1133 | ||
985ed670 | 1134 | id = idr_alloc_cyclic(&cgroup_hierarchy_idr, root, 0, 0, GFP_KERNEL); |
f2e85d57 TH |
1135 | if (id < 0) |
1136 | return id; | |
1137 | ||
1138 | root->hierarchy_id = id; | |
1139 | return 0; | |
1140 | } | |
1141 | ||
3dd06ffa | 1142 | static void cgroup_exit_root_id(struct cgroup_root *root) |
f2e85d57 TH |
1143 | { |
1144 | lockdep_assert_held(&cgroup_mutex); | |
1145 | ||
1146 | if (root->hierarchy_id) { | |
1147 | idr_remove(&cgroup_hierarchy_idr, root->hierarchy_id); | |
1148 | root->hierarchy_id = 0; | |
1149 | } | |
1150 | } | |
1151 | ||
3dd06ffa | 1152 | static void cgroup_free_root(struct cgroup_root *root) |
f2e85d57 TH |
1153 | { |
1154 | if (root) { | |
d0f702e6 | 1155 | /* hierarchy ID should already have been released */ |
f2e85d57 TH |
1156 | WARN_ON_ONCE(root->hierarchy_id); |
1157 | ||
1158 | idr_destroy(&root->cgroup_idr); | |
1159 | kfree(root); | |
1160 | } | |
1161 | } | |
1162 | ||
3dd06ffa | 1163 | static void cgroup_destroy_root(struct cgroup_root *root) |
59f5296b | 1164 | { |
3dd06ffa | 1165 | struct cgroup *cgrp = &root->cgrp; |
f2e85d57 | 1166 | struct cgrp_cset_link *link, *tmp_link; |
f2e85d57 | 1167 | |
334c3679 | 1168 | cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp); |
f2e85d57 | 1169 | |
776f02fa | 1170 | BUG_ON(atomic_read(&root->nr_cgrps)); |
d5c419b6 | 1171 | BUG_ON(!list_empty(&cgrp->self.children)); |
f2e85d57 | 1172 | |
f2e85d57 | 1173 | /* Rebind all subsystems back to the default hierarchy */ |
334c3679 | 1174 | WARN_ON(rebind_subsystems(&cgrp_dfl_root, root->subsys_mask)); |
7717f7ba | 1175 | |
7717f7ba | 1176 | /* |
f2e85d57 TH |
1177 | * Release all the links from cset_links to this hierarchy's |
1178 | * root cgroup | |
7717f7ba | 1179 | */ |
f0d9a5f1 | 1180 | spin_lock_bh(&css_set_lock); |
f2e85d57 TH |
1181 | |
1182 | list_for_each_entry_safe(link, tmp_link, &cgrp->cset_links, cset_link) { | |
1183 | list_del(&link->cset_link); | |
1184 | list_del(&link->cgrp_link); | |
1185 | kfree(link); | |
1186 | } | |
f0d9a5f1 TH |
1187 | |
1188 | spin_unlock_bh(&css_set_lock); | |
f2e85d57 TH |
1189 | |
1190 | if (!list_empty(&root->root_list)) { | |
1191 | list_del(&root->root_list); | |
1192 | cgroup_root_count--; | |
1193 | } | |
1194 | ||
1195 | cgroup_exit_root_id(root); | |
1196 | ||
1197 | mutex_unlock(&cgroup_mutex); | |
f2e85d57 | 1198 | |
2bd59d48 | 1199 | kernfs_destroy_root(root->kf_root); |
f2e85d57 TH |
1200 | cgroup_free_root(root); |
1201 | } | |
1202 | ||
ceb6a081 TH |
1203 | /* look up cgroup associated with given css_set on the specified hierarchy */ |
1204 | static struct cgroup *cset_cgroup_from_root(struct css_set *cset, | |
3dd06ffa | 1205 | struct cgroup_root *root) |
7717f7ba | 1206 | { |
7717f7ba PM |
1207 | struct cgroup *res = NULL; |
1208 | ||
96d365e0 | 1209 | lockdep_assert_held(&cgroup_mutex); |
f0d9a5f1 | 1210 | lockdep_assert_held(&css_set_lock); |
96d365e0 | 1211 | |
5abb8855 | 1212 | if (cset == &init_css_set) { |
3dd06ffa | 1213 | res = &root->cgrp; |
7717f7ba | 1214 | } else { |
69d0206c TH |
1215 | struct cgrp_cset_link *link; |
1216 | ||
1217 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) { | |
7717f7ba | 1218 | struct cgroup *c = link->cgrp; |
69d0206c | 1219 | |
7717f7ba PM |
1220 | if (c->root == root) { |
1221 | res = c; | |
1222 | break; | |
1223 | } | |
1224 | } | |
1225 | } | |
96d365e0 | 1226 | |
7717f7ba PM |
1227 | BUG_ON(!res); |
1228 | return res; | |
1229 | } | |
1230 | ||
ddbcc7e8 | 1231 | /* |
ceb6a081 | 1232 | * Return the cgroup for "task" from the given hierarchy. Must be |
f0d9a5f1 | 1233 | * called with cgroup_mutex and css_set_lock held. |
ceb6a081 TH |
1234 | */ |
1235 | static struct cgroup *task_cgroup_from_root(struct task_struct *task, | |
3dd06ffa | 1236 | struct cgroup_root *root) |
ceb6a081 TH |
1237 | { |
1238 | /* | |
1239 | * No need to lock the task - since we hold cgroup_mutex the | |
1240 | * task can't change groups, so the only thing that can happen | |
1241 | * is that it exits and its css is set back to init_css_set. | |
1242 | */ | |
1243 | return cset_cgroup_from_root(task_css_set(task), root); | |
1244 | } | |
1245 | ||
ddbcc7e8 | 1246 | /* |
ddbcc7e8 PM |
1247 | * A task must hold cgroup_mutex to modify cgroups. |
1248 | * | |
1249 | * Any task can increment and decrement the count field without lock. | |
1250 | * So in general, code holding cgroup_mutex can't rely on the count | |
1251 | * field not changing. However, if the count goes to zero, then only | |
956db3ca | 1252 | * cgroup_attach_task() can increment it again. Because a count of zero |
ddbcc7e8 PM |
1253 | * means that no tasks are currently attached, therefore there is no |
1254 | * way a task attached to that cgroup can fork (the other way to | |
1255 | * increment the count). So code holding cgroup_mutex can safely | |
1256 | * assume that if the count is zero, it will stay zero. Similarly, if | |
1257 | * a task holds cgroup_mutex on a cgroup with zero count, it | |
1258 | * knows that the cgroup won't be removed, as cgroup_rmdir() | |
1259 | * needs that mutex. | |
1260 | * | |
ddbcc7e8 PM |
1261 | * A cgroup can only be deleted if both its 'count' of using tasks |
1262 | * is zero, and its list of 'children' cgroups is empty. Since all | |
1263 | * tasks in the system use _some_ cgroup, and since there is always at | |
3dd06ffa | 1264 | * least one task in the system (init, pid == 1), therefore, root cgroup |
ddbcc7e8 | 1265 | * always has either children cgroups and/or using tasks. So we don't |
3dd06ffa | 1266 | * need a special hack to ensure that root cgroup cannot be deleted. |
ddbcc7e8 PM |
1267 | * |
1268 | * P.S. One more locking exception. RCU is used to guard the | |
956db3ca | 1269 | * update of a tasks cgroup pointer by cgroup_attach_task() |
ddbcc7e8 PM |
1270 | */ |
1271 | ||
2bd59d48 | 1272 | static struct kernfs_syscall_ops cgroup_kf_syscall_ops; |
828c0950 | 1273 | static const struct file_operations proc_cgroupstats_operations; |
a424316c | 1274 | |
8d7e6fb0 TH |
1275 | static char *cgroup_file_name(struct cgroup *cgrp, const struct cftype *cft, |
1276 | char *buf) | |
ddbcc7e8 | 1277 | { |
3e1d2eed TH |
1278 | struct cgroup_subsys *ss = cft->ss; |
1279 | ||
8d7e6fb0 TH |
1280 | if (cft->ss && !(cft->flags & CFTYPE_NO_PREFIX) && |
1281 | !(cgrp->root->flags & CGRP_ROOT_NOPREFIX)) | |
1282 | snprintf(buf, CGROUP_FILE_NAME_MAX, "%s.%s", | |
3e1d2eed TH |
1283 | cgroup_on_dfl(cgrp) ? ss->name : ss->legacy_name, |
1284 | cft->name); | |
8d7e6fb0 TH |
1285 | else |
1286 | strncpy(buf, cft->name, CGROUP_FILE_NAME_MAX); | |
1287 | return buf; | |
ddbcc7e8 PM |
1288 | } |
1289 | ||
f2e85d57 TH |
1290 | /** |
1291 | * cgroup_file_mode - deduce file mode of a control file | |
1292 | * @cft: the control file in question | |
1293 | * | |
7dbdb199 | 1294 | * S_IRUGO for read, S_IWUSR for write. |
f2e85d57 TH |
1295 | */ |
1296 | static umode_t cgroup_file_mode(const struct cftype *cft) | |
65dff759 | 1297 | { |
f2e85d57 | 1298 | umode_t mode = 0; |
65dff759 | 1299 | |
f2e85d57 TH |
1300 | if (cft->read_u64 || cft->read_s64 || cft->seq_show) |
1301 | mode |= S_IRUGO; | |
1302 | ||
7dbdb199 TH |
1303 | if (cft->write_u64 || cft->write_s64 || cft->write) { |
1304 | if (cft->flags & CFTYPE_WORLD_WRITABLE) | |
1305 | mode |= S_IWUGO; | |
1306 | else | |
1307 | mode |= S_IWUSR; | |
1308 | } | |
f2e85d57 TH |
1309 | |
1310 | return mode; | |
65dff759 LZ |
1311 | } |
1312 | ||
af0ba678 | 1313 | /** |
8699b776 | 1314 | * cgroup_calc_subtree_ss_mask - calculate subtree_ss_mask |
0f060deb | 1315 | * @subtree_control: the new subtree_control mask to consider |
5ced2518 | 1316 | * @this_ss_mask: available subsystems |
af0ba678 TH |
1317 | * |
1318 | * On the default hierarchy, a subsystem may request other subsystems to be | |
1319 | * enabled together through its ->depends_on mask. In such cases, more | |
1320 | * subsystems than specified in "cgroup.subtree_control" may be enabled. | |
1321 | * | |
0f060deb | 1322 | * This function calculates which subsystems need to be enabled if |
5ced2518 | 1323 | * @subtree_control is to be applied while restricted to @this_ss_mask. |
af0ba678 | 1324 | */ |
5ced2518 | 1325 | static u16 cgroup_calc_subtree_ss_mask(u16 subtree_control, u16 this_ss_mask) |
667c2491 | 1326 | { |
6e5c8307 | 1327 | u16 cur_ss_mask = subtree_control; |
af0ba678 TH |
1328 | struct cgroup_subsys *ss; |
1329 | int ssid; | |
1330 | ||
1331 | lockdep_assert_held(&cgroup_mutex); | |
1332 | ||
f6d635ad TH |
1333 | cur_ss_mask |= cgrp_dfl_implicit_ss_mask; |
1334 | ||
af0ba678 | 1335 | while (true) { |
6e5c8307 | 1336 | u16 new_ss_mask = cur_ss_mask; |
af0ba678 | 1337 | |
b4e0eeaf | 1338 | do_each_subsys_mask(ss, ssid, cur_ss_mask) { |
a966a4ed | 1339 | new_ss_mask |= ss->depends_on; |
b4e0eeaf | 1340 | } while_each_subsys_mask(); |
af0ba678 TH |
1341 | |
1342 | /* | |
1343 | * Mask out subsystems which aren't available. This can | |
1344 | * happen only if some depended-upon subsystems were bound | |
1345 | * to non-default hierarchies. | |
1346 | */ | |
5ced2518 | 1347 | new_ss_mask &= this_ss_mask; |
af0ba678 TH |
1348 | |
1349 | if (new_ss_mask == cur_ss_mask) | |
1350 | break; | |
1351 | cur_ss_mask = new_ss_mask; | |
1352 | } | |
1353 | ||
0f060deb TH |
1354 | return cur_ss_mask; |
1355 | } | |
1356 | ||
a9746d8d TH |
1357 | /** |
1358 | * cgroup_kn_unlock - unlocking helper for cgroup kernfs methods | |
1359 | * @kn: the kernfs_node being serviced | |
1360 | * | |
1361 | * This helper undoes cgroup_kn_lock_live() and should be invoked before | |
1362 | * the method finishes if locking succeeded. Note that once this function | |
1363 | * returns the cgroup returned by cgroup_kn_lock_live() may become | |
1364 | * inaccessible any time. If the caller intends to continue to access the | |
1365 | * cgroup, it should pin it before invoking this function. | |
1366 | */ | |
1367 | static void cgroup_kn_unlock(struct kernfs_node *kn) | |
ddbcc7e8 | 1368 | { |
a9746d8d TH |
1369 | struct cgroup *cgrp; |
1370 | ||
1371 | if (kernfs_type(kn) == KERNFS_DIR) | |
1372 | cgrp = kn->priv; | |
1373 | else | |
1374 | cgrp = kn->parent->priv; | |
1375 | ||
1376 | mutex_unlock(&cgroup_mutex); | |
a9746d8d TH |
1377 | |
1378 | kernfs_unbreak_active_protection(kn); | |
1379 | cgroup_put(cgrp); | |
ddbcc7e8 PM |
1380 | } |
1381 | ||
a9746d8d TH |
1382 | /** |
1383 | * cgroup_kn_lock_live - locking helper for cgroup kernfs methods | |
1384 | * @kn: the kernfs_node being serviced | |
945ba199 | 1385 | * @drain_offline: perform offline draining on the cgroup |
a9746d8d TH |
1386 | * |
1387 | * This helper is to be used by a cgroup kernfs method currently servicing | |
1388 | * @kn. It breaks the active protection, performs cgroup locking and | |
1389 | * verifies that the associated cgroup is alive. Returns the cgroup if | |
1390 | * alive; otherwise, %NULL. A successful return should be undone by a | |
945ba199 TH |
1391 | * matching cgroup_kn_unlock() invocation. If @drain_offline is %true, the |
1392 | * cgroup is drained of offlining csses before return. | |
a9746d8d TH |
1393 | * |
1394 | * Any cgroup kernfs method implementation which requires locking the | |
1395 | * associated cgroup should use this helper. It avoids nesting cgroup | |
1396 | * locking under kernfs active protection and allows all kernfs operations | |
1397 | * including self-removal. | |
1398 | */ | |
945ba199 TH |
1399 | static struct cgroup *cgroup_kn_lock_live(struct kernfs_node *kn, |
1400 | bool drain_offline) | |
05ef1d7c | 1401 | { |
a9746d8d TH |
1402 | struct cgroup *cgrp; |
1403 | ||
1404 | if (kernfs_type(kn) == KERNFS_DIR) | |
1405 | cgrp = kn->priv; | |
1406 | else | |
1407 | cgrp = kn->parent->priv; | |
05ef1d7c | 1408 | |
2739d3cc | 1409 | /* |
01f6474c | 1410 | * We're gonna grab cgroup_mutex which nests outside kernfs |
a9746d8d TH |
1411 | * active_ref. cgroup liveliness check alone provides enough |
1412 | * protection against removal. Ensure @cgrp stays accessible and | |
1413 | * break the active_ref protection. | |
2739d3cc | 1414 | */ |
aa32362f LZ |
1415 | if (!cgroup_tryget(cgrp)) |
1416 | return NULL; | |
a9746d8d TH |
1417 | kernfs_break_active_protection(kn); |
1418 | ||
945ba199 TH |
1419 | if (drain_offline) |
1420 | cgroup_lock_and_drain_offline(cgrp); | |
1421 | else | |
1422 | mutex_lock(&cgroup_mutex); | |
05ef1d7c | 1423 | |
a9746d8d TH |
1424 | if (!cgroup_is_dead(cgrp)) |
1425 | return cgrp; | |
1426 | ||
1427 | cgroup_kn_unlock(kn); | |
1428 | return NULL; | |
ddbcc7e8 | 1429 | } |
05ef1d7c | 1430 | |
2739d3cc | 1431 | static void cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft) |
05ef1d7c | 1432 | { |
2bd59d48 | 1433 | char name[CGROUP_FILE_NAME_MAX]; |
05ef1d7c | 1434 | |
01f6474c | 1435 | lockdep_assert_held(&cgroup_mutex); |
34c06254 TH |
1436 | |
1437 | if (cft->file_offset) { | |
1438 | struct cgroup_subsys_state *css = cgroup_css(cgrp, cft->ss); | |
1439 | struct cgroup_file *cfile = (void *)css + cft->file_offset; | |
1440 | ||
1441 | spin_lock_irq(&cgroup_file_kn_lock); | |
1442 | cfile->kn = NULL; | |
1443 | spin_unlock_irq(&cgroup_file_kn_lock); | |
1444 | } | |
1445 | ||
2bd59d48 | 1446 | kernfs_remove_by_name(cgrp->kn, cgroup_file_name(cgrp, cft, name)); |
05ef1d7c TH |
1447 | } |
1448 | ||
13af07df | 1449 | /** |
4df8dc90 TH |
1450 | * css_clear_dir - remove subsys files in a cgroup directory |
1451 | * @css: taget css | |
13af07df | 1452 | */ |
334c3679 | 1453 | static void css_clear_dir(struct cgroup_subsys_state *css) |
05ef1d7c | 1454 | { |
334c3679 | 1455 | struct cgroup *cgrp = css->cgroup; |
4df8dc90 | 1456 | struct cftype *cfts; |
05ef1d7c | 1457 | |
88cb04b9 TH |
1458 | if (!(css->flags & CSS_VISIBLE)) |
1459 | return; | |
1460 | ||
1461 | css->flags &= ~CSS_VISIBLE; | |
1462 | ||
4df8dc90 TH |
1463 | list_for_each_entry(cfts, &css->ss->cfts, node) |
1464 | cgroup_addrm_files(css, cgrp, cfts, false); | |
ddbcc7e8 PM |
1465 | } |
1466 | ||
ccdca218 | 1467 | /** |
4df8dc90 TH |
1468 | * css_populate_dir - create subsys files in a cgroup directory |
1469 | * @css: target css | |
ccdca218 TH |
1470 | * |
1471 | * On failure, no file is added. | |
1472 | */ | |
334c3679 | 1473 | static int css_populate_dir(struct cgroup_subsys_state *css) |
ccdca218 | 1474 | { |
334c3679 | 1475 | struct cgroup *cgrp = css->cgroup; |
4df8dc90 TH |
1476 | struct cftype *cfts, *failed_cfts; |
1477 | int ret; | |
ccdca218 | 1478 | |
03970d3c | 1479 | if ((css->flags & CSS_VISIBLE) || !cgrp->kn) |
88cb04b9 TH |
1480 | return 0; |
1481 | ||
4df8dc90 TH |
1482 | if (!css->ss) { |
1483 | if (cgroup_on_dfl(cgrp)) | |
1484 | cfts = cgroup_dfl_base_files; | |
1485 | else | |
1486 | cfts = cgroup_legacy_base_files; | |
ccdca218 | 1487 | |
4df8dc90 TH |
1488 | return cgroup_addrm_files(&cgrp->self, cgrp, cfts, true); |
1489 | } | |
ccdca218 | 1490 | |
4df8dc90 TH |
1491 | list_for_each_entry(cfts, &css->ss->cfts, node) { |
1492 | ret = cgroup_addrm_files(css, cgrp, cfts, true); | |
1493 | if (ret < 0) { | |
1494 | failed_cfts = cfts; | |
1495 | goto err; | |
ccdca218 TH |
1496 | } |
1497 | } | |
88cb04b9 TH |
1498 | |
1499 | css->flags |= CSS_VISIBLE; | |
1500 | ||
ccdca218 TH |
1501 | return 0; |
1502 | err: | |
4df8dc90 TH |
1503 | list_for_each_entry(cfts, &css->ss->cfts, node) { |
1504 | if (cfts == failed_cfts) | |
1505 | break; | |
1506 | cgroup_addrm_files(css, cgrp, cfts, false); | |
1507 | } | |
ccdca218 TH |
1508 | return ret; |
1509 | } | |
1510 | ||
6e5c8307 | 1511 | static int rebind_subsystems(struct cgroup_root *dst_root, u16 ss_mask) |
ddbcc7e8 | 1512 | { |
1ada4838 | 1513 | struct cgroup *dcgrp = &dst_root->cgrp; |
30159ec7 | 1514 | struct cgroup_subsys *ss; |
2d8f243a | 1515 | int ssid, i, ret; |
ddbcc7e8 | 1516 | |
ace2bee8 | 1517 | lockdep_assert_held(&cgroup_mutex); |
ddbcc7e8 | 1518 | |
b4e0eeaf | 1519 | do_each_subsys_mask(ss, ssid, ss_mask) { |
f6d635ad TH |
1520 | /* |
1521 | * If @ss has non-root csses attached to it, can't move. | |
1522 | * If @ss is an implicit controller, it is exempt from this | |
1523 | * rule and can be stolen. | |
1524 | */ | |
1525 | if (css_next_child(NULL, cgroup_css(&ss->root->cgrp, ss)) && | |
1526 | !ss->implicit_on_dfl) | |
3ed80a62 | 1527 | return -EBUSY; |
1d5be6b2 | 1528 | |
5df36032 | 1529 | /* can't move between two non-dummy roots either */ |
7fd8c565 | 1530 | if (ss->root != &cgrp_dfl_root && dst_root != &cgrp_dfl_root) |
5df36032 | 1531 | return -EBUSY; |
b4e0eeaf | 1532 | } while_each_subsys_mask(); |
ddbcc7e8 | 1533 | |
b4e0eeaf | 1534 | do_each_subsys_mask(ss, ssid, ss_mask) { |
1ada4838 TH |
1535 | struct cgroup_root *src_root = ss->root; |
1536 | struct cgroup *scgrp = &src_root->cgrp; | |
1537 | struct cgroup_subsys_state *css = cgroup_css(scgrp, ss); | |
2d8f243a | 1538 | struct css_set *cset; |
a8a648c4 | 1539 | |
1ada4838 | 1540 | WARN_ON(!css || cgroup_css(dcgrp, ss)); |
a8a648c4 | 1541 | |
334c3679 TH |
1542 | /* disable from the source */ |
1543 | src_root->subsys_mask &= ~(1 << ssid); | |
1544 | WARN_ON(cgroup_apply_control(scgrp)); | |
1545 | cgroup_finalize_control(scgrp, 0); | |
4df8dc90 | 1546 | |
334c3679 | 1547 | /* rebind */ |
1ada4838 TH |
1548 | RCU_INIT_POINTER(scgrp->subsys[ssid], NULL); |
1549 | rcu_assign_pointer(dcgrp->subsys[ssid], css); | |
5df36032 | 1550 | ss->root = dst_root; |
1ada4838 | 1551 | css->cgroup = dcgrp; |
73e80ed8 | 1552 | |
f0d9a5f1 | 1553 | spin_lock_bh(&css_set_lock); |
2d8f243a TH |
1554 | hash_for_each(css_set_table, i, cset, hlist) |
1555 | list_move_tail(&cset->e_cset_node[ss->id], | |
1ada4838 | 1556 | &dcgrp->e_csets[ss->id]); |
f0d9a5f1 | 1557 | spin_unlock_bh(&css_set_lock); |
2d8f243a | 1558 | |
bd53d617 | 1559 | /* default hierarchy doesn't enable controllers by default */ |
f392e51c | 1560 | dst_root->subsys_mask |= 1 << ssid; |
49d1dc4b TH |
1561 | if (dst_root == &cgrp_dfl_root) { |
1562 | static_branch_enable(cgroup_subsys_on_dfl_key[ssid]); | |
1563 | } else { | |
1ada4838 | 1564 | dcgrp->subtree_control |= 1 << ssid; |
49d1dc4b | 1565 | static_branch_disable(cgroup_subsys_on_dfl_key[ssid]); |
667c2491 | 1566 | } |
a8a648c4 | 1567 | |
334c3679 TH |
1568 | ret = cgroup_apply_control(dcgrp); |
1569 | if (ret) | |
1570 | pr_warn("partial failure to rebind %s controller (err=%d)\n", | |
1571 | ss->name, ret); | |
1572 | ||
5df36032 TH |
1573 | if (ss->bind) |
1574 | ss->bind(css); | |
b4e0eeaf | 1575 | } while_each_subsys_mask(); |
ddbcc7e8 | 1576 | |
1ada4838 | 1577 | kernfs_activate(dcgrp->kn); |
ddbcc7e8 PM |
1578 | return 0; |
1579 | } | |
1580 | ||
2bd59d48 TH |
1581 | static int cgroup_show_options(struct seq_file *seq, |
1582 | struct kernfs_root *kf_root) | |
ddbcc7e8 | 1583 | { |
3dd06ffa | 1584 | struct cgroup_root *root = cgroup_root_from_kf(kf_root); |
ddbcc7e8 | 1585 | struct cgroup_subsys *ss; |
b85d2040 | 1586 | int ssid; |
ddbcc7e8 | 1587 | |
d98817d4 TH |
1588 | if (root != &cgrp_dfl_root) |
1589 | for_each_subsys(ss, ssid) | |
1590 | if (root->subsys_mask & (1 << ssid)) | |
61e57c0c | 1591 | seq_show_option(seq, ss->legacy_name, NULL); |
93438629 | 1592 | if (root->flags & CGRP_ROOT_NOPREFIX) |
ddbcc7e8 | 1593 | seq_puts(seq, ",noprefix"); |
93438629 | 1594 | if (root->flags & CGRP_ROOT_XATTR) |
03b1cde6 | 1595 | seq_puts(seq, ",xattr"); |
69e943b7 TH |
1596 | |
1597 | spin_lock(&release_agent_path_lock); | |
81a6a5cd | 1598 | if (strlen(root->release_agent_path)) |
a068acf2 KC |
1599 | seq_show_option(seq, "release_agent", |
1600 | root->release_agent_path); | |
69e943b7 TH |
1601 | spin_unlock(&release_agent_path_lock); |
1602 | ||
3dd06ffa | 1603 | if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags)) |
97978e6d | 1604 | seq_puts(seq, ",clone_children"); |
c6d57f33 | 1605 | if (strlen(root->name)) |
a068acf2 | 1606 | seq_show_option(seq, "name", root->name); |
ddbcc7e8 PM |
1607 | return 0; |
1608 | } | |
1609 | ||
1610 | struct cgroup_sb_opts { | |
6e5c8307 | 1611 | u16 subsys_mask; |
69dfa00c | 1612 | unsigned int flags; |
81a6a5cd | 1613 | char *release_agent; |
2260e7fc | 1614 | bool cpuset_clone_children; |
c6d57f33 | 1615 | char *name; |
2c6ab6d2 PM |
1616 | /* User explicitly requested empty subsystem */ |
1617 | bool none; | |
ddbcc7e8 PM |
1618 | }; |
1619 | ||
cf5d5941 | 1620 | static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts) |
ddbcc7e8 | 1621 | { |
32a8cf23 DL |
1622 | char *token, *o = data; |
1623 | bool all_ss = false, one_ss = false; | |
6e5c8307 | 1624 | u16 mask = U16_MAX; |
30159ec7 | 1625 | struct cgroup_subsys *ss; |
7b9a6ba5 | 1626 | int nr_opts = 0; |
30159ec7 | 1627 | int i; |
f9ab5b5b LZ |
1628 | |
1629 | #ifdef CONFIG_CPUSETS | |
6e5c8307 | 1630 | mask = ~((u16)1 << cpuset_cgrp_id); |
f9ab5b5b | 1631 | #endif |
ddbcc7e8 | 1632 | |
c6d57f33 | 1633 | memset(opts, 0, sizeof(*opts)); |
ddbcc7e8 PM |
1634 | |
1635 | while ((token = strsep(&o, ",")) != NULL) { | |
7b9a6ba5 TH |
1636 | nr_opts++; |
1637 | ||
ddbcc7e8 PM |
1638 | if (!*token) |
1639 | return -EINVAL; | |
32a8cf23 | 1640 | if (!strcmp(token, "none")) { |
2c6ab6d2 PM |
1641 | /* Explicitly have no subsystems */ |
1642 | opts->none = true; | |
32a8cf23 DL |
1643 | continue; |
1644 | } | |
1645 | if (!strcmp(token, "all")) { | |
1646 | /* Mutually exclusive option 'all' + subsystem name */ | |
1647 | if (one_ss) | |
1648 | return -EINVAL; | |
1649 | all_ss = true; | |
1650 | continue; | |
1651 | } | |
1652 | if (!strcmp(token, "noprefix")) { | |
93438629 | 1653 | opts->flags |= CGRP_ROOT_NOPREFIX; |
32a8cf23 DL |
1654 | continue; |
1655 | } | |
1656 | if (!strcmp(token, "clone_children")) { | |
2260e7fc | 1657 | opts->cpuset_clone_children = true; |
32a8cf23 DL |
1658 | continue; |
1659 | } | |
03b1cde6 | 1660 | if (!strcmp(token, "xattr")) { |
93438629 | 1661 | opts->flags |= CGRP_ROOT_XATTR; |
03b1cde6 AR |
1662 | continue; |
1663 | } | |
32a8cf23 | 1664 | if (!strncmp(token, "release_agent=", 14)) { |
81a6a5cd PM |
1665 | /* Specifying two release agents is forbidden */ |
1666 | if (opts->release_agent) | |
1667 | return -EINVAL; | |
c6d57f33 | 1668 | opts->release_agent = |
e400c285 | 1669 | kstrndup(token + 14, PATH_MAX - 1, GFP_KERNEL); |
81a6a5cd PM |
1670 | if (!opts->release_agent) |
1671 | return -ENOMEM; | |
32a8cf23 DL |
1672 | continue; |
1673 | } | |
1674 | if (!strncmp(token, "name=", 5)) { | |
c6d57f33 PM |
1675 | const char *name = token + 5; |
1676 | /* Can't specify an empty name */ | |
1677 | if (!strlen(name)) | |
1678 | return -EINVAL; | |
1679 | /* Must match [\w.-]+ */ | |
1680 | for (i = 0; i < strlen(name); i++) { | |
1681 | char c = name[i]; | |
1682 | if (isalnum(c)) | |
1683 | continue; | |
1684 | if ((c == '.') || (c == '-') || (c == '_')) | |
1685 | continue; | |
1686 | return -EINVAL; | |
1687 | } | |
1688 | /* Specifying two names is forbidden */ | |
1689 | if (opts->name) | |
1690 | return -EINVAL; | |
1691 | opts->name = kstrndup(name, | |
e400c285 | 1692 | MAX_CGROUP_ROOT_NAMELEN - 1, |
c6d57f33 PM |
1693 | GFP_KERNEL); |
1694 | if (!opts->name) | |
1695 | return -ENOMEM; | |
32a8cf23 DL |
1696 | |
1697 | continue; | |
1698 | } | |
1699 | ||
30159ec7 | 1700 | for_each_subsys(ss, i) { |
3e1d2eed | 1701 | if (strcmp(token, ss->legacy_name)) |
32a8cf23 | 1702 | continue; |
fc5ed1e9 | 1703 | if (!cgroup_ssid_enabled(i)) |
32a8cf23 | 1704 | continue; |
223ffb29 JW |
1705 | if (cgroup_ssid_no_v1(i)) |
1706 | continue; | |
32a8cf23 DL |
1707 | |
1708 | /* Mutually exclusive option 'all' + subsystem name */ | |
1709 | if (all_ss) | |
1710 | return -EINVAL; | |
69dfa00c | 1711 | opts->subsys_mask |= (1 << i); |
32a8cf23 DL |
1712 | one_ss = true; |
1713 | ||
1714 | break; | |
1715 | } | |
1716 | if (i == CGROUP_SUBSYS_COUNT) | |
1717 | return -ENOENT; | |
1718 | } | |
1719 | ||
7b9a6ba5 TH |
1720 | /* |
1721 | * If the 'all' option was specified select all the subsystems, | |
1722 | * otherwise if 'none', 'name=' and a subsystem name options were | |
1723 | * not specified, let's default to 'all' | |
1724 | */ | |
1725 | if (all_ss || (!one_ss && !opts->none && !opts->name)) | |
1726 | for_each_subsys(ss, i) | |
223ffb29 | 1727 | if (cgroup_ssid_enabled(i) && !cgroup_ssid_no_v1(i)) |
7b9a6ba5 TH |
1728 | opts->subsys_mask |= (1 << i); |
1729 | ||
1730 | /* | |
1731 | * We either have to specify by name or by subsystems. (So all | |
1732 | * empty hierarchies must have a name). | |
1733 | */ | |
1734 | if (!opts->subsys_mask && !opts->name) | |
1735 | return -EINVAL; | |
1736 | ||
f9ab5b5b LZ |
1737 | /* |
1738 | * Option noprefix was introduced just for backward compatibility | |
1739 | * with the old cpuset, so we allow noprefix only if mounting just | |
1740 | * the cpuset subsystem. | |
1741 | */ | |
93438629 | 1742 | if ((opts->flags & CGRP_ROOT_NOPREFIX) && (opts->subsys_mask & mask)) |
f9ab5b5b LZ |
1743 | return -EINVAL; |
1744 | ||
2c6ab6d2 | 1745 | /* Can't specify "none" and some subsystems */ |
a1a71b45 | 1746 | if (opts->subsys_mask && opts->none) |
2c6ab6d2 PM |
1747 | return -EINVAL; |
1748 | ||
ddbcc7e8 PM |
1749 | return 0; |
1750 | } | |
1751 | ||
2bd59d48 | 1752 | static int cgroup_remount(struct kernfs_root *kf_root, int *flags, char *data) |
ddbcc7e8 PM |
1753 | { |
1754 | int ret = 0; | |
3dd06ffa | 1755 | struct cgroup_root *root = cgroup_root_from_kf(kf_root); |
ddbcc7e8 | 1756 | struct cgroup_sb_opts opts; |
6e5c8307 | 1757 | u16 added_mask, removed_mask; |
ddbcc7e8 | 1758 | |
aa6ec29b TH |
1759 | if (root == &cgrp_dfl_root) { |
1760 | pr_err("remount is not allowed\n"); | |
873fe09e TH |
1761 | return -EINVAL; |
1762 | } | |
1763 | ||
334c3679 | 1764 | cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp); |
ddbcc7e8 PM |
1765 | |
1766 | /* See what subsystems are wanted */ | |
1767 | ret = parse_cgroupfs_options(data, &opts); | |
1768 | if (ret) | |
1769 | goto out_unlock; | |
1770 | ||
f392e51c | 1771 | if (opts.subsys_mask != root->subsys_mask || opts.release_agent) |
ed3d261b | 1772 | pr_warn("option changes via remount are deprecated (pid=%d comm=%s)\n", |
a2a1f9ea | 1773 | task_tgid_nr(current), current->comm); |
8b5a5a9d | 1774 | |
f392e51c TH |
1775 | added_mask = opts.subsys_mask & ~root->subsys_mask; |
1776 | removed_mask = root->subsys_mask & ~opts.subsys_mask; | |
13af07df | 1777 | |
cf5d5941 | 1778 | /* Don't allow flags or name to change at remount */ |
7450e90b | 1779 | if ((opts.flags ^ root->flags) || |
cf5d5941 | 1780 | (opts.name && strcmp(opts.name, root->name))) { |
69dfa00c | 1781 | pr_err("option or name mismatch, new: 0x%x \"%s\", old: 0x%x \"%s\"\n", |
7450e90b | 1782 | opts.flags, opts.name ?: "", root->flags, root->name); |
c6d57f33 PM |
1783 | ret = -EINVAL; |
1784 | goto out_unlock; | |
1785 | } | |
1786 | ||
f172e67c | 1787 | /* remounting is not allowed for populated hierarchies */ |
d5c419b6 | 1788 | if (!list_empty(&root->cgrp.self.children)) { |
f172e67c | 1789 | ret = -EBUSY; |
0670e08b | 1790 | goto out_unlock; |
cf5d5941 | 1791 | } |
ddbcc7e8 | 1792 | |
5df36032 | 1793 | ret = rebind_subsystems(root, added_mask); |
3126121f | 1794 | if (ret) |
0670e08b | 1795 | goto out_unlock; |
ddbcc7e8 | 1796 | |
334c3679 | 1797 | WARN_ON(rebind_subsystems(&cgrp_dfl_root, removed_mask)); |
5df36032 | 1798 | |
69e943b7 TH |
1799 | if (opts.release_agent) { |
1800 | spin_lock(&release_agent_path_lock); | |
81a6a5cd | 1801 | strcpy(root->release_agent_path, opts.release_agent); |
69e943b7 TH |
1802 | spin_unlock(&release_agent_path_lock); |
1803 | } | |
ddbcc7e8 | 1804 | out_unlock: |
66bdc9cf | 1805 | kfree(opts.release_agent); |
c6d57f33 | 1806 | kfree(opts.name); |
ddbcc7e8 | 1807 | mutex_unlock(&cgroup_mutex); |
ddbcc7e8 PM |
1808 | return ret; |
1809 | } | |
1810 | ||
afeb0f9f TH |
1811 | /* |
1812 | * To reduce the fork() overhead for systems that are not actually using | |
1813 | * their cgroups capability, we don't maintain the lists running through | |
1814 | * each css_set to its tasks until we see the list actually used - in other | |
1815 | * words after the first mount. | |
1816 | */ | |
1817 | static bool use_task_css_set_links __read_mostly; | |
1818 | ||
1819 | static void cgroup_enable_task_cg_lists(void) | |
1820 | { | |
1821 | struct task_struct *p, *g; | |
1822 | ||
f0d9a5f1 | 1823 | spin_lock_bh(&css_set_lock); |
afeb0f9f TH |
1824 | |
1825 | if (use_task_css_set_links) | |
1826 | goto out_unlock; | |
1827 | ||
1828 | use_task_css_set_links = true; | |
1829 | ||
1830 | /* | |
1831 | * We need tasklist_lock because RCU is not safe against | |
1832 | * while_each_thread(). Besides, a forking task that has passed | |
1833 | * cgroup_post_fork() without seeing use_task_css_set_links = 1 | |
1834 | * is not guaranteed to have its child immediately visible in the | |
1835 | * tasklist if we walk through it with RCU. | |
1836 | */ | |
1837 | read_lock(&tasklist_lock); | |
1838 | do_each_thread(g, p) { | |
afeb0f9f TH |
1839 | WARN_ON_ONCE(!list_empty(&p->cg_list) || |
1840 | task_css_set(p) != &init_css_set); | |
1841 | ||
1842 | /* | |
1843 | * We should check if the process is exiting, otherwise | |
1844 | * it will race with cgroup_exit() in that the list | |
1845 | * entry won't be deleted though the process has exited. | |
f153ad11 TH |
1846 | * Do it while holding siglock so that we don't end up |
1847 | * racing against cgroup_exit(). | |
afeb0f9f | 1848 | */ |
f153ad11 | 1849 | spin_lock_irq(&p->sighand->siglock); |
eaf797ab TH |
1850 | if (!(p->flags & PF_EXITING)) { |
1851 | struct css_set *cset = task_css_set(p); | |
1852 | ||
0de0942d TH |
1853 | if (!css_set_populated(cset)) |
1854 | css_set_update_populated(cset, true); | |
389b9c1b | 1855 | list_add_tail(&p->cg_list, &cset->tasks); |
eaf797ab TH |
1856 | get_css_set(cset); |
1857 | } | |
f153ad11 | 1858 | spin_unlock_irq(&p->sighand->siglock); |
afeb0f9f TH |
1859 | } while_each_thread(g, p); |
1860 | read_unlock(&tasklist_lock); | |
1861 | out_unlock: | |
f0d9a5f1 | 1862 | spin_unlock_bh(&css_set_lock); |
afeb0f9f | 1863 | } |
ddbcc7e8 | 1864 | |
cc31edce PM |
1865 | static void init_cgroup_housekeeping(struct cgroup *cgrp) |
1866 | { | |
2d8f243a TH |
1867 | struct cgroup_subsys *ss; |
1868 | int ssid; | |
1869 | ||
d5c419b6 TH |
1870 | INIT_LIST_HEAD(&cgrp->self.sibling); |
1871 | INIT_LIST_HEAD(&cgrp->self.children); | |
69d0206c | 1872 | INIT_LIST_HEAD(&cgrp->cset_links); |
72a8cb30 BB |
1873 | INIT_LIST_HEAD(&cgrp->pidlists); |
1874 | mutex_init(&cgrp->pidlist_mutex); | |
9d800df1 | 1875 | cgrp->self.cgroup = cgrp; |
184faf32 | 1876 | cgrp->self.flags |= CSS_ONLINE; |
2d8f243a TH |
1877 | |
1878 | for_each_subsys(ss, ssid) | |
1879 | INIT_LIST_HEAD(&cgrp->e_csets[ssid]); | |
f8f22e53 TH |
1880 | |
1881 | init_waitqueue_head(&cgrp->offline_waitq); | |
971ff493 | 1882 | INIT_WORK(&cgrp->release_agent_work, cgroup_release_agent); |
cc31edce | 1883 | } |
c6d57f33 | 1884 | |
3dd06ffa | 1885 | static void init_cgroup_root(struct cgroup_root *root, |
172a2c06 | 1886 | struct cgroup_sb_opts *opts) |
ddbcc7e8 | 1887 | { |
3dd06ffa | 1888 | struct cgroup *cgrp = &root->cgrp; |
b0ca5a84 | 1889 | |
ddbcc7e8 | 1890 | INIT_LIST_HEAD(&root->root_list); |
3c9c825b | 1891 | atomic_set(&root->nr_cgrps, 1); |
bd89aabc | 1892 | cgrp->root = root; |
cc31edce | 1893 | init_cgroup_housekeeping(cgrp); |
4e96ee8e | 1894 | idr_init(&root->cgroup_idr); |
c6d57f33 | 1895 | |
c6d57f33 PM |
1896 | root->flags = opts->flags; |
1897 | if (opts->release_agent) | |
1898 | strcpy(root->release_agent_path, opts->release_agent); | |
1899 | if (opts->name) | |
1900 | strcpy(root->name, opts->name); | |
2260e7fc | 1901 | if (opts->cpuset_clone_children) |
3dd06ffa | 1902 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags); |
c6d57f33 PM |
1903 | } |
1904 | ||
6e5c8307 | 1905 | static int cgroup_setup_root(struct cgroup_root *root, u16 ss_mask) |
2c6ab6d2 | 1906 | { |
d427dfeb | 1907 | LIST_HEAD(tmp_links); |
3dd06ffa | 1908 | struct cgroup *root_cgrp = &root->cgrp; |
d427dfeb | 1909 | struct css_set *cset; |
d427dfeb | 1910 | int i, ret; |
2c6ab6d2 | 1911 | |
d427dfeb | 1912 | lockdep_assert_held(&cgroup_mutex); |
c6d57f33 | 1913 | |
cf780b7d | 1914 | ret = cgroup_idr_alloc(&root->cgroup_idr, root_cgrp, 1, 2, GFP_KERNEL); |
d427dfeb | 1915 | if (ret < 0) |
2bd59d48 | 1916 | goto out; |
d427dfeb | 1917 | root_cgrp->id = ret; |
b11cfb58 | 1918 | root_cgrp->ancestor_ids[0] = ret; |
c6d57f33 | 1919 | |
2aad2a86 TH |
1920 | ret = percpu_ref_init(&root_cgrp->self.refcnt, css_release, 0, |
1921 | GFP_KERNEL); | |
9d755d33 TH |
1922 | if (ret) |
1923 | goto out; | |
1924 | ||
d427dfeb | 1925 | /* |
f0d9a5f1 | 1926 | * We're accessing css_set_count without locking css_set_lock here, |
d427dfeb | 1927 | * but that's OK - it can only be increased by someone holding |
04313591 TH |
1928 | * cgroup_lock, and that's us. Later rebinding may disable |
1929 | * controllers on the default hierarchy and thus create new csets, | |
1930 | * which can't be more than the existing ones. Allocate 2x. | |
d427dfeb | 1931 | */ |
04313591 | 1932 | ret = allocate_cgrp_cset_links(2 * css_set_count, &tmp_links); |
d427dfeb | 1933 | if (ret) |
9d755d33 | 1934 | goto cancel_ref; |
ddbcc7e8 | 1935 | |
985ed670 | 1936 | ret = cgroup_init_root_id(root); |
ddbcc7e8 | 1937 | if (ret) |
9d755d33 | 1938 | goto cancel_ref; |
ddbcc7e8 | 1939 | |
2bd59d48 TH |
1940 | root->kf_root = kernfs_create_root(&cgroup_kf_syscall_ops, |
1941 | KERNFS_ROOT_CREATE_DEACTIVATED, | |
1942 | root_cgrp); | |
1943 | if (IS_ERR(root->kf_root)) { | |
1944 | ret = PTR_ERR(root->kf_root); | |
1945 | goto exit_root_id; | |
1946 | } | |
1947 | root_cgrp->kn = root->kf_root->kn; | |
ddbcc7e8 | 1948 | |
334c3679 | 1949 | ret = css_populate_dir(&root_cgrp->self); |
d427dfeb | 1950 | if (ret) |
2bd59d48 | 1951 | goto destroy_root; |
ddbcc7e8 | 1952 | |
5df36032 | 1953 | ret = rebind_subsystems(root, ss_mask); |
d427dfeb | 1954 | if (ret) |
2bd59d48 | 1955 | goto destroy_root; |
ddbcc7e8 | 1956 | |
d427dfeb TH |
1957 | /* |
1958 | * There must be no failure case after here, since rebinding takes | |
1959 | * care of subsystems' refcounts, which are explicitly dropped in | |
1960 | * the failure exit path. | |
1961 | */ | |
1962 | list_add(&root->root_list, &cgroup_roots); | |
1963 | cgroup_root_count++; | |
0df6a63f | 1964 | |
d427dfeb | 1965 | /* |
3dd06ffa | 1966 | * Link the root cgroup in this hierarchy into all the css_set |
d427dfeb TH |
1967 | * objects. |
1968 | */ | |
f0d9a5f1 | 1969 | spin_lock_bh(&css_set_lock); |
0de0942d | 1970 | hash_for_each(css_set_table, i, cset, hlist) { |
d427dfeb | 1971 | link_css_set(&tmp_links, cset, root_cgrp); |
0de0942d TH |
1972 | if (css_set_populated(cset)) |
1973 | cgroup_update_populated(root_cgrp, true); | |
1974 | } | |
f0d9a5f1 | 1975 | spin_unlock_bh(&css_set_lock); |
ddbcc7e8 | 1976 | |
d5c419b6 | 1977 | BUG_ON(!list_empty(&root_cgrp->self.children)); |
3c9c825b | 1978 | BUG_ON(atomic_read(&root->nr_cgrps) != 1); |
ddbcc7e8 | 1979 | |
2bd59d48 | 1980 | kernfs_activate(root_cgrp->kn); |
d427dfeb | 1981 | ret = 0; |
2bd59d48 | 1982 | goto out; |
d427dfeb | 1983 | |
2bd59d48 TH |
1984 | destroy_root: |
1985 | kernfs_destroy_root(root->kf_root); | |
1986 | root->kf_root = NULL; | |
1987 | exit_root_id: | |
d427dfeb | 1988 | cgroup_exit_root_id(root); |
9d755d33 | 1989 | cancel_ref: |
9a1049da | 1990 | percpu_ref_exit(&root_cgrp->self.refcnt); |
2bd59d48 | 1991 | out: |
d427dfeb TH |
1992 | free_cgrp_cset_links(&tmp_links); |
1993 | return ret; | |
ddbcc7e8 PM |
1994 | } |
1995 | ||
f7e83571 | 1996 | static struct dentry *cgroup_mount(struct file_system_type *fs_type, |
ddbcc7e8 | 1997 | int flags, const char *unused_dev_name, |
f7e83571 | 1998 | void *data) |
ddbcc7e8 | 1999 | { |
67e9c74b | 2000 | bool is_v2 = fs_type == &cgroup2_fs_type; |
3a32bd72 | 2001 | struct super_block *pinned_sb = NULL; |
970317aa | 2002 | struct cgroup_subsys *ss; |
3dd06ffa | 2003 | struct cgroup_root *root; |
ddbcc7e8 | 2004 | struct cgroup_sb_opts opts; |
2bd59d48 | 2005 | struct dentry *dentry; |
8e30e2b8 | 2006 | int ret; |
970317aa | 2007 | int i; |
c6b3d5bc | 2008 | bool new_sb; |
ddbcc7e8 | 2009 | |
56fde9e0 TH |
2010 | /* |
2011 | * The first time anyone tries to mount a cgroup, enable the list | |
2012 | * linking each css_set to its tasks and fix up all existing tasks. | |
2013 | */ | |
2014 | if (!use_task_css_set_links) | |
2015 | cgroup_enable_task_cg_lists(); | |
e37a06f1 | 2016 | |
67e9c74b TH |
2017 | if (is_v2) { |
2018 | if (data) { | |
2019 | pr_err("cgroup2: unknown option \"%s\"\n", (char *)data); | |
2020 | return ERR_PTR(-EINVAL); | |
2021 | } | |
a7165264 | 2022 | cgrp_dfl_visible = true; |
67e9c74b TH |
2023 | root = &cgrp_dfl_root; |
2024 | cgroup_get(&root->cgrp); | |
2025 | goto out_mount; | |
2026 | } | |
2027 | ||
334c3679 | 2028 | cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp); |
8e30e2b8 TH |
2029 | |
2030 | /* First find the desired set of subsystems */ | |
ddbcc7e8 | 2031 | ret = parse_cgroupfs_options(data, &opts); |
c6d57f33 | 2032 | if (ret) |
8e30e2b8 | 2033 | goto out_unlock; |
a015edd2 | 2034 | |
970317aa LZ |
2035 | /* |
2036 | * Destruction of cgroup root is asynchronous, so subsystems may | |
2037 | * still be dying after the previous unmount. Let's drain the | |
2038 | * dying subsystems. We just need to ensure that the ones | |
2039 | * unmounted previously finish dying and don't care about new ones | |
2040 | * starting. Testing ref liveliness is good enough. | |
2041 | */ | |
2042 | for_each_subsys(ss, i) { | |
2043 | if (!(opts.subsys_mask & (1 << i)) || | |
2044 | ss->root == &cgrp_dfl_root) | |
2045 | continue; | |
2046 | ||
2047 | if (!percpu_ref_tryget_live(&ss->root->cgrp.self.refcnt)) { | |
2048 | mutex_unlock(&cgroup_mutex); | |
2049 | msleep(10); | |
2050 | ret = restart_syscall(); | |
2051 | goto out_free; | |
2052 | } | |
2053 | cgroup_put(&ss->root->cgrp); | |
2054 | } | |
2055 | ||
985ed670 | 2056 | for_each_root(root) { |
2bd59d48 | 2057 | bool name_match = false; |
3126121f | 2058 | |
3dd06ffa | 2059 | if (root == &cgrp_dfl_root) |
985ed670 | 2060 | continue; |
3126121f | 2061 | |
cf5d5941 | 2062 | /* |
2bd59d48 TH |
2063 | * If we asked for a name then it must match. Also, if |
2064 | * name matches but sybsys_mask doesn't, we should fail. | |
2065 | * Remember whether name matched. | |
cf5d5941 | 2066 | */ |
2bd59d48 TH |
2067 | if (opts.name) { |
2068 | if (strcmp(opts.name, root->name)) | |
2069 | continue; | |
2070 | name_match = true; | |
2071 | } | |
ddbcc7e8 | 2072 | |
c6d57f33 | 2073 | /* |
2bd59d48 TH |
2074 | * If we asked for subsystems (or explicitly for no |
2075 | * subsystems) then they must match. | |
c6d57f33 | 2076 | */ |
2bd59d48 | 2077 | if ((opts.subsys_mask || opts.none) && |
f392e51c | 2078 | (opts.subsys_mask != root->subsys_mask)) { |
2bd59d48 TH |
2079 | if (!name_match) |
2080 | continue; | |
2081 | ret = -EBUSY; | |
2082 | goto out_unlock; | |
2083 | } | |
873fe09e | 2084 | |
7b9a6ba5 TH |
2085 | if (root->flags ^ opts.flags) |
2086 | pr_warn("new mount options do not match the existing superblock, will be ignored\n"); | |
ddbcc7e8 | 2087 | |
776f02fa | 2088 | /* |
3a32bd72 LZ |
2089 | * We want to reuse @root whose lifetime is governed by its |
2090 | * ->cgrp. Let's check whether @root is alive and keep it | |
2091 | * that way. As cgroup_kill_sb() can happen anytime, we | |
2092 | * want to block it by pinning the sb so that @root doesn't | |
2093 | * get killed before mount is complete. | |
2094 | * | |
2095 | * With the sb pinned, tryget_live can reliably indicate | |
2096 | * whether @root can be reused. If it's being killed, | |
2097 | * drain it. We can use wait_queue for the wait but this | |
2098 | * path is super cold. Let's just sleep a bit and retry. | |
776f02fa | 2099 | */ |
3a32bd72 LZ |
2100 | pinned_sb = kernfs_pin_sb(root->kf_root, NULL); |
2101 | if (IS_ERR(pinned_sb) || | |
2102 | !percpu_ref_tryget_live(&root->cgrp.self.refcnt)) { | |
776f02fa | 2103 | mutex_unlock(&cgroup_mutex); |
3a32bd72 LZ |
2104 | if (!IS_ERR_OR_NULL(pinned_sb)) |
2105 | deactivate_super(pinned_sb); | |
776f02fa | 2106 | msleep(10); |
a015edd2 TH |
2107 | ret = restart_syscall(); |
2108 | goto out_free; | |
776f02fa | 2109 | } |
ddbcc7e8 | 2110 | |
776f02fa | 2111 | ret = 0; |
2bd59d48 | 2112 | goto out_unlock; |
ddbcc7e8 | 2113 | } |
ddbcc7e8 | 2114 | |
817929ec | 2115 | /* |
172a2c06 TH |
2116 | * No such thing, create a new one. name= matching without subsys |
2117 | * specification is allowed for already existing hierarchies but we | |
2118 | * can't create new one without subsys specification. | |
817929ec | 2119 | */ |
172a2c06 TH |
2120 | if (!opts.subsys_mask && !opts.none) { |
2121 | ret = -EINVAL; | |
2122 | goto out_unlock; | |
817929ec | 2123 | } |
817929ec | 2124 | |
172a2c06 TH |
2125 | root = kzalloc(sizeof(*root), GFP_KERNEL); |
2126 | if (!root) { | |
2127 | ret = -ENOMEM; | |
2bd59d48 | 2128 | goto out_unlock; |
839ec545 | 2129 | } |
e5f6a860 | 2130 | |
172a2c06 TH |
2131 | init_cgroup_root(root, &opts); |
2132 | ||
35585573 | 2133 | ret = cgroup_setup_root(root, opts.subsys_mask); |
2bd59d48 TH |
2134 | if (ret) |
2135 | cgroup_free_root(root); | |
fa3ca07e | 2136 | |
8e30e2b8 | 2137 | out_unlock: |
ddbcc7e8 | 2138 | mutex_unlock(&cgroup_mutex); |
a015edd2 | 2139 | out_free: |
c6d57f33 PM |
2140 | kfree(opts.release_agent); |
2141 | kfree(opts.name); | |
03b1cde6 | 2142 | |
2bd59d48 | 2143 | if (ret) |
8e30e2b8 | 2144 | return ERR_PTR(ret); |
67e9c74b | 2145 | out_mount: |
c9482a5b | 2146 | dentry = kernfs_mount(fs_type, flags, root->kf_root, |
67e9c74b TH |
2147 | is_v2 ? CGROUP2_SUPER_MAGIC : CGROUP_SUPER_MAGIC, |
2148 | &new_sb); | |
c6b3d5bc | 2149 | if (IS_ERR(dentry) || !new_sb) |
3dd06ffa | 2150 | cgroup_put(&root->cgrp); |
3a32bd72 LZ |
2151 | |
2152 | /* | |
2153 | * If @pinned_sb, we're reusing an existing root and holding an | |
2154 | * extra ref on its sb. Mount is complete. Put the extra ref. | |
2155 | */ | |
2156 | if (pinned_sb) { | |
2157 | WARN_ON(new_sb); | |
2158 | deactivate_super(pinned_sb); | |
2159 | } | |
2160 | ||
2bd59d48 TH |
2161 | return dentry; |
2162 | } | |
2163 | ||
2164 | static void cgroup_kill_sb(struct super_block *sb) | |
2165 | { | |
2166 | struct kernfs_root *kf_root = kernfs_root_from_sb(sb); | |
3dd06ffa | 2167 | struct cgroup_root *root = cgroup_root_from_kf(kf_root); |
2bd59d48 | 2168 | |
9d755d33 TH |
2169 | /* |
2170 | * If @root doesn't have any mounts or children, start killing it. | |
2171 | * This prevents new mounts by disabling percpu_ref_tryget_live(). | |
2172 | * cgroup_mount() may wait for @root's release. | |
1f779fb2 LZ |
2173 | * |
2174 | * And don't kill the default root. | |
9d755d33 | 2175 | */ |
3c606d35 | 2176 | if (!list_empty(&root->cgrp.self.children) || |
1f779fb2 | 2177 | root == &cgrp_dfl_root) |
9d755d33 TH |
2178 | cgroup_put(&root->cgrp); |
2179 | else | |
2180 | percpu_ref_kill(&root->cgrp.self.refcnt); | |
2181 | ||
2bd59d48 | 2182 | kernfs_kill_sb(sb); |
ddbcc7e8 PM |
2183 | } |
2184 | ||
2185 | static struct file_system_type cgroup_fs_type = { | |
2186 | .name = "cgroup", | |
f7e83571 | 2187 | .mount = cgroup_mount, |
ddbcc7e8 PM |
2188 | .kill_sb = cgroup_kill_sb, |
2189 | }; | |
2190 | ||
67e9c74b TH |
2191 | static struct file_system_type cgroup2_fs_type = { |
2192 | .name = "cgroup2", | |
2193 | .mount = cgroup_mount, | |
2194 | .kill_sb = cgroup_kill_sb, | |
2195 | }; | |
2196 | ||
857a2beb | 2197 | /** |
913ffdb5 | 2198 | * task_cgroup_path - cgroup path of a task in the first cgroup hierarchy |
857a2beb | 2199 | * @task: target task |
857a2beb TH |
2200 | * @buf: the buffer to write the path into |
2201 | * @buflen: the length of the buffer | |
2202 | * | |
913ffdb5 TH |
2203 | * Determine @task's cgroup on the first (the one with the lowest non-zero |
2204 | * hierarchy_id) cgroup hierarchy and copy its path into @buf. This | |
2205 | * function grabs cgroup_mutex and shouldn't be used inside locks used by | |
2206 | * cgroup controller callbacks. | |
2207 | * | |
e61734c5 | 2208 | * Return value is the same as kernfs_path(). |
857a2beb | 2209 | */ |
e61734c5 | 2210 | char *task_cgroup_path(struct task_struct *task, char *buf, size_t buflen) |
857a2beb | 2211 | { |
3dd06ffa | 2212 | struct cgroup_root *root; |
913ffdb5 | 2213 | struct cgroup *cgrp; |
e61734c5 TH |
2214 | int hierarchy_id = 1; |
2215 | char *path = NULL; | |
857a2beb TH |
2216 | |
2217 | mutex_lock(&cgroup_mutex); | |
f0d9a5f1 | 2218 | spin_lock_bh(&css_set_lock); |
857a2beb | 2219 | |
913ffdb5 TH |
2220 | root = idr_get_next(&cgroup_hierarchy_idr, &hierarchy_id); |
2221 | ||
857a2beb TH |
2222 | if (root) { |
2223 | cgrp = task_cgroup_from_root(task, root); | |
e61734c5 | 2224 | path = cgroup_path(cgrp, buf, buflen); |
913ffdb5 TH |
2225 | } else { |
2226 | /* if no hierarchy exists, everyone is in "/" */ | |
e61734c5 TH |
2227 | if (strlcpy(buf, "/", buflen) < buflen) |
2228 | path = buf; | |
857a2beb TH |
2229 | } |
2230 | ||
f0d9a5f1 | 2231 | spin_unlock_bh(&css_set_lock); |
857a2beb | 2232 | mutex_unlock(&cgroup_mutex); |
e61734c5 | 2233 | return path; |
857a2beb | 2234 | } |
913ffdb5 | 2235 | EXPORT_SYMBOL_GPL(task_cgroup_path); |
857a2beb | 2236 | |
b3dc094e | 2237 | /* used to track tasks and other necessary states during migration */ |
2f7ee569 | 2238 | struct cgroup_taskset { |
b3dc094e TH |
2239 | /* the src and dst cset list running through cset->mg_node */ |
2240 | struct list_head src_csets; | |
2241 | struct list_head dst_csets; | |
2242 | ||
1f7dd3e5 TH |
2243 | /* the subsys currently being processed */ |
2244 | int ssid; | |
2245 | ||
b3dc094e TH |
2246 | /* |
2247 | * Fields for cgroup_taskset_*() iteration. | |
2248 | * | |
2249 | * Before migration is committed, the target migration tasks are on | |
2250 | * ->mg_tasks of the csets on ->src_csets. After, on ->mg_tasks of | |
2251 | * the csets on ->dst_csets. ->csets point to either ->src_csets | |
2252 | * or ->dst_csets depending on whether migration is committed. | |
2253 | * | |
2254 | * ->cur_csets and ->cur_task point to the current task position | |
2255 | * during iteration. | |
2256 | */ | |
2257 | struct list_head *csets; | |
2258 | struct css_set *cur_cset; | |
2259 | struct task_struct *cur_task; | |
2f7ee569 TH |
2260 | }; |
2261 | ||
adaae5dc TH |
2262 | #define CGROUP_TASKSET_INIT(tset) (struct cgroup_taskset){ \ |
2263 | .src_csets = LIST_HEAD_INIT(tset.src_csets), \ | |
2264 | .dst_csets = LIST_HEAD_INIT(tset.dst_csets), \ | |
2265 | .csets = &tset.src_csets, \ | |
2266 | } | |
2267 | ||
2268 | /** | |
2269 | * cgroup_taskset_add - try to add a migration target task to a taskset | |
2270 | * @task: target task | |
2271 | * @tset: target taskset | |
2272 | * | |
2273 | * Add @task, which is a migration target, to @tset. This function becomes | |
2274 | * noop if @task doesn't need to be migrated. @task's css_set should have | |
2275 | * been added as a migration source and @task->cg_list will be moved from | |
2276 | * the css_set's tasks list to mg_tasks one. | |
2277 | */ | |
2278 | static void cgroup_taskset_add(struct task_struct *task, | |
2279 | struct cgroup_taskset *tset) | |
2280 | { | |
2281 | struct css_set *cset; | |
2282 | ||
f0d9a5f1 | 2283 | lockdep_assert_held(&css_set_lock); |
adaae5dc TH |
2284 | |
2285 | /* @task either already exited or can't exit until the end */ | |
2286 | if (task->flags & PF_EXITING) | |
2287 | return; | |
2288 | ||
2289 | /* leave @task alone if post_fork() hasn't linked it yet */ | |
2290 | if (list_empty(&task->cg_list)) | |
2291 | return; | |
2292 | ||
2293 | cset = task_css_set(task); | |
2294 | if (!cset->mg_src_cgrp) | |
2295 | return; | |
2296 | ||
2297 | list_move_tail(&task->cg_list, &cset->mg_tasks); | |
2298 | if (list_empty(&cset->mg_node)) | |
2299 | list_add_tail(&cset->mg_node, &tset->src_csets); | |
2300 | if (list_empty(&cset->mg_dst_cset->mg_node)) | |
2301 | list_move_tail(&cset->mg_dst_cset->mg_node, | |
2302 | &tset->dst_csets); | |
2303 | } | |
2304 | ||
2f7ee569 TH |
2305 | /** |
2306 | * cgroup_taskset_first - reset taskset and return the first task | |
2307 | * @tset: taskset of interest | |
1f7dd3e5 | 2308 | * @dst_cssp: output variable for the destination css |
2f7ee569 TH |
2309 | * |
2310 | * @tset iteration is initialized and the first task is returned. | |
2311 | */ | |
1f7dd3e5 TH |
2312 | struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset, |
2313 | struct cgroup_subsys_state **dst_cssp) | |
2f7ee569 | 2314 | { |
b3dc094e TH |
2315 | tset->cur_cset = list_first_entry(tset->csets, struct css_set, mg_node); |
2316 | tset->cur_task = NULL; | |
2317 | ||
1f7dd3e5 | 2318 | return cgroup_taskset_next(tset, dst_cssp); |
2f7ee569 | 2319 | } |
2f7ee569 TH |
2320 | |
2321 | /** | |
2322 | * cgroup_taskset_next - iterate to the next task in taskset | |
2323 | * @tset: taskset of interest | |
1f7dd3e5 | 2324 | * @dst_cssp: output variable for the destination css |
2f7ee569 TH |
2325 | * |
2326 | * Return the next task in @tset. Iteration must have been initialized | |
2327 | * with cgroup_taskset_first(). | |
2328 | */ | |
1f7dd3e5 TH |
2329 | struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset, |
2330 | struct cgroup_subsys_state **dst_cssp) | |
2f7ee569 | 2331 | { |
b3dc094e TH |
2332 | struct css_set *cset = tset->cur_cset; |
2333 | struct task_struct *task = tset->cur_task; | |
2f7ee569 | 2334 | |
b3dc094e TH |
2335 | while (&cset->mg_node != tset->csets) { |
2336 | if (!task) | |
2337 | task = list_first_entry(&cset->mg_tasks, | |
2338 | struct task_struct, cg_list); | |
2339 | else | |
2340 | task = list_next_entry(task, cg_list); | |
2f7ee569 | 2341 | |
b3dc094e TH |
2342 | if (&task->cg_list != &cset->mg_tasks) { |
2343 | tset->cur_cset = cset; | |
2344 | tset->cur_task = task; | |
1f7dd3e5 TH |
2345 | |
2346 | /* | |
2347 | * This function may be called both before and | |
2348 | * after cgroup_taskset_migrate(). The two cases | |
2349 | * can be distinguished by looking at whether @cset | |
2350 | * has its ->mg_dst_cset set. | |
2351 | */ | |
2352 | if (cset->mg_dst_cset) | |
2353 | *dst_cssp = cset->mg_dst_cset->subsys[tset->ssid]; | |
2354 | else | |
2355 | *dst_cssp = cset->subsys[tset->ssid]; | |
2356 | ||
b3dc094e TH |
2357 | return task; |
2358 | } | |
2f7ee569 | 2359 | |
b3dc094e TH |
2360 | cset = list_next_entry(cset, mg_node); |
2361 | task = NULL; | |
2362 | } | |
2f7ee569 | 2363 | |
b3dc094e | 2364 | return NULL; |
2f7ee569 | 2365 | } |
2f7ee569 | 2366 | |
adaae5dc | 2367 | /** |
37ff9f8f | 2368 | * cgroup_taskset_migrate - migrate a taskset |
adaae5dc | 2369 | * @tset: taget taskset |
37ff9f8f | 2370 | * @root: cgroup root the migration is taking place on |
adaae5dc | 2371 | * |
37ff9f8f TH |
2372 | * Migrate tasks in @tset as setup by migration preparation functions. |
2373 | * This function fails iff one of the ->can_attach callbacks fails and | |
2374 | * guarantees that either all or none of the tasks in @tset are migrated. | |
2375 | * @tset is consumed regardless of success. | |
adaae5dc TH |
2376 | */ |
2377 | static int cgroup_taskset_migrate(struct cgroup_taskset *tset, | |
37ff9f8f | 2378 | struct cgroup_root *root) |
adaae5dc | 2379 | { |
37ff9f8f | 2380 | struct cgroup_subsys *ss; |
adaae5dc TH |
2381 | struct task_struct *task, *tmp_task; |
2382 | struct css_set *cset, *tmp_cset; | |
37ff9f8f | 2383 | int ssid, failed_ssid, ret; |
adaae5dc TH |
2384 | |
2385 | /* methods shouldn't be called if no task is actually migrating */ | |
2386 | if (list_empty(&tset->src_csets)) | |
2387 | return 0; | |
2388 | ||
2389 | /* check that we can legitimately attach to the cgroup */ | |
37ff9f8f TH |
2390 | do_each_subsys_mask(ss, ssid, root->subsys_mask) { |
2391 | if (ss->can_attach) { | |
2392 | tset->ssid = ssid; | |
2393 | ret = ss->can_attach(tset); | |
adaae5dc | 2394 | if (ret) { |
37ff9f8f | 2395 | failed_ssid = ssid; |
adaae5dc TH |
2396 | goto out_cancel_attach; |
2397 | } | |
2398 | } | |
37ff9f8f | 2399 | } while_each_subsys_mask(); |
adaae5dc TH |
2400 | |
2401 | /* | |
2402 | * Now that we're guaranteed success, proceed to move all tasks to | |
2403 | * the new cgroup. There are no failure cases after here, so this | |
2404 | * is the commit point. | |
2405 | */ | |
f0d9a5f1 | 2406 | spin_lock_bh(&css_set_lock); |
adaae5dc | 2407 | list_for_each_entry(cset, &tset->src_csets, mg_node) { |
f6d7d049 TH |
2408 | list_for_each_entry_safe(task, tmp_task, &cset->mg_tasks, cg_list) { |
2409 | struct css_set *from_cset = task_css_set(task); | |
2410 | struct css_set *to_cset = cset->mg_dst_cset; | |
2411 | ||
2412 | get_css_set(to_cset); | |
2413 | css_set_move_task(task, from_cset, to_cset, true); | |
2414 | put_css_set_locked(from_cset); | |
2415 | } | |
adaae5dc | 2416 | } |
f0d9a5f1 | 2417 | spin_unlock_bh(&css_set_lock); |
adaae5dc TH |
2418 | |
2419 | /* | |
2420 | * Migration is committed, all target tasks are now on dst_csets. | |
2421 | * Nothing is sensitive to fork() after this point. Notify | |
2422 | * controllers that migration is complete. | |
2423 | */ | |
2424 | tset->csets = &tset->dst_csets; | |
2425 | ||
37ff9f8f TH |
2426 | do_each_subsys_mask(ss, ssid, root->subsys_mask) { |
2427 | if (ss->attach) { | |
2428 | tset->ssid = ssid; | |
2429 | ss->attach(tset); | |
1f7dd3e5 | 2430 | } |
37ff9f8f | 2431 | } while_each_subsys_mask(); |
adaae5dc TH |
2432 | |
2433 | ret = 0; | |
2434 | goto out_release_tset; | |
2435 | ||
2436 | out_cancel_attach: | |
37ff9f8f TH |
2437 | do_each_subsys_mask(ss, ssid, root->subsys_mask) { |
2438 | if (ssid == failed_ssid) | |
adaae5dc | 2439 | break; |
37ff9f8f TH |
2440 | if (ss->cancel_attach) { |
2441 | tset->ssid = ssid; | |
2442 | ss->cancel_attach(tset); | |
1f7dd3e5 | 2443 | } |
37ff9f8f | 2444 | } while_each_subsys_mask(); |
adaae5dc | 2445 | out_release_tset: |
f0d9a5f1 | 2446 | spin_lock_bh(&css_set_lock); |
adaae5dc TH |
2447 | list_splice_init(&tset->dst_csets, &tset->src_csets); |
2448 | list_for_each_entry_safe(cset, tmp_cset, &tset->src_csets, mg_node) { | |
2449 | list_splice_tail_init(&cset->mg_tasks, &cset->tasks); | |
2450 | list_del_init(&cset->mg_node); | |
2451 | } | |
f0d9a5f1 | 2452 | spin_unlock_bh(&css_set_lock); |
adaae5dc TH |
2453 | return ret; |
2454 | } | |
2455 | ||
6c694c88 TH |
2456 | /** |
2457 | * cgroup_may_migrate_to - verify whether a cgroup can be migration destination | |
2458 | * @dst_cgrp: destination cgroup to test | |
2459 | * | |
2460 | * On the default hierarchy, except for the root, subtree_control must be | |
2461 | * zero for migration destination cgroups with tasks so that child cgroups | |
2462 | * don't compete against tasks. | |
2463 | */ | |
2464 | static bool cgroup_may_migrate_to(struct cgroup *dst_cgrp) | |
2465 | { | |
2466 | return !cgroup_on_dfl(dst_cgrp) || !cgroup_parent(dst_cgrp) || | |
2467 | !dst_cgrp->subtree_control; | |
2468 | } | |
2469 | ||
a043e3b2 | 2470 | /** |
1958d2d5 TH |
2471 | * cgroup_migrate_finish - cleanup after attach |
2472 | * @preloaded_csets: list of preloaded css_sets | |
74a1166d | 2473 | * |
1958d2d5 TH |
2474 | * Undo cgroup_migrate_add_src() and cgroup_migrate_prepare_dst(). See |
2475 | * those functions for details. | |
74a1166d | 2476 | */ |
1958d2d5 | 2477 | static void cgroup_migrate_finish(struct list_head *preloaded_csets) |
74a1166d | 2478 | { |
1958d2d5 | 2479 | struct css_set *cset, *tmp_cset; |
74a1166d | 2480 | |
1958d2d5 TH |
2481 | lockdep_assert_held(&cgroup_mutex); |
2482 | ||
f0d9a5f1 | 2483 | spin_lock_bh(&css_set_lock); |
1958d2d5 TH |
2484 | list_for_each_entry_safe(cset, tmp_cset, preloaded_csets, mg_preload_node) { |
2485 | cset->mg_src_cgrp = NULL; | |
e4857982 | 2486 | cset->mg_dst_cgrp = NULL; |
1958d2d5 TH |
2487 | cset->mg_dst_cset = NULL; |
2488 | list_del_init(&cset->mg_preload_node); | |
a25eb52e | 2489 | put_css_set_locked(cset); |
1958d2d5 | 2490 | } |
f0d9a5f1 | 2491 | spin_unlock_bh(&css_set_lock); |
1958d2d5 TH |
2492 | } |
2493 | ||
2494 | /** | |
2495 | * cgroup_migrate_add_src - add a migration source css_set | |
2496 | * @src_cset: the source css_set to add | |
2497 | * @dst_cgrp: the destination cgroup | |
2498 | * @preloaded_csets: list of preloaded css_sets | |
2499 | * | |
2500 | * Tasks belonging to @src_cset are about to be migrated to @dst_cgrp. Pin | |
2501 | * @src_cset and add it to @preloaded_csets, which should later be cleaned | |
2502 | * up by cgroup_migrate_finish(). | |
2503 | * | |
1ed13287 TH |
2504 | * This function may be called without holding cgroup_threadgroup_rwsem |
2505 | * even if the target is a process. Threads may be created and destroyed | |
2506 | * but as long as cgroup_mutex is not dropped, no new css_set can be put | |
2507 | * into play and the preloaded css_sets are guaranteed to cover all | |
2508 | * migrations. | |
1958d2d5 TH |
2509 | */ |
2510 | static void cgroup_migrate_add_src(struct css_set *src_cset, | |
2511 | struct cgroup *dst_cgrp, | |
2512 | struct list_head *preloaded_csets) | |
2513 | { | |
2514 | struct cgroup *src_cgrp; | |
2515 | ||
2516 | lockdep_assert_held(&cgroup_mutex); | |
f0d9a5f1 | 2517 | lockdep_assert_held(&css_set_lock); |
1958d2d5 | 2518 | |
2b021cbf TH |
2519 | /* |
2520 | * If ->dead, @src_set is associated with one or more dead cgroups | |
2521 | * and doesn't contain any migratable tasks. Ignore it early so | |
2522 | * that the rest of migration path doesn't get confused by it. | |
2523 | */ | |
2524 | if (src_cset->dead) | |
2525 | return; | |
2526 | ||
1958d2d5 TH |
2527 | src_cgrp = cset_cgroup_from_root(src_cset, dst_cgrp->root); |
2528 | ||
1958d2d5 TH |
2529 | if (!list_empty(&src_cset->mg_preload_node)) |
2530 | return; | |
2531 | ||
2532 | WARN_ON(src_cset->mg_src_cgrp); | |
e4857982 | 2533 | WARN_ON(src_cset->mg_dst_cgrp); |
1958d2d5 TH |
2534 | WARN_ON(!list_empty(&src_cset->mg_tasks)); |
2535 | WARN_ON(!list_empty(&src_cset->mg_node)); | |
2536 | ||
2537 | src_cset->mg_src_cgrp = src_cgrp; | |
e4857982 | 2538 | src_cset->mg_dst_cgrp = dst_cgrp; |
1958d2d5 TH |
2539 | get_css_set(src_cset); |
2540 | list_add(&src_cset->mg_preload_node, preloaded_csets); | |
2541 | } | |
2542 | ||
2543 | /** | |
2544 | * cgroup_migrate_prepare_dst - prepare destination css_sets for migration | |
1958d2d5 TH |
2545 | * @preloaded_csets: list of preloaded source css_sets |
2546 | * | |
e4857982 TH |
2547 | * Tasks are about to be moved and all the source css_sets have been |
2548 | * preloaded to @preloaded_csets. This function looks up and pins all | |
2549 | * destination css_sets, links each to its source, and append them to | |
2550 | * @preloaded_csets. | |
1958d2d5 TH |
2551 | * |
2552 | * This function must be called after cgroup_migrate_add_src() has been | |
2553 | * called on each migration source css_set. After migration is performed | |
2554 | * using cgroup_migrate(), cgroup_migrate_finish() must be called on | |
2555 | * @preloaded_csets. | |
2556 | */ | |
e4857982 | 2557 | static int cgroup_migrate_prepare_dst(struct list_head *preloaded_csets) |
1958d2d5 TH |
2558 | { |
2559 | LIST_HEAD(csets); | |
f817de98 | 2560 | struct css_set *src_cset, *tmp_cset; |
1958d2d5 TH |
2561 | |
2562 | lockdep_assert_held(&cgroup_mutex); | |
2563 | ||
2564 | /* look up the dst cset for each src cset and link it to src */ | |
f817de98 | 2565 | list_for_each_entry_safe(src_cset, tmp_cset, preloaded_csets, mg_preload_node) { |
1958d2d5 TH |
2566 | struct css_set *dst_cset; |
2567 | ||
e4857982 | 2568 | dst_cset = find_css_set(src_cset, src_cset->mg_dst_cgrp); |
1958d2d5 TH |
2569 | if (!dst_cset) |
2570 | goto err; | |
2571 | ||
2572 | WARN_ON_ONCE(src_cset->mg_dst_cset || dst_cset->mg_dst_cset); | |
f817de98 TH |
2573 | |
2574 | /* | |
2575 | * If src cset equals dst, it's noop. Drop the src. | |
2576 | * cgroup_migrate() will skip the cset too. Note that we | |
2577 | * can't handle src == dst as some nodes are used by both. | |
2578 | */ | |
2579 | if (src_cset == dst_cset) { | |
2580 | src_cset->mg_src_cgrp = NULL; | |
e4857982 | 2581 | src_cset->mg_dst_cgrp = NULL; |
f817de98 | 2582 | list_del_init(&src_cset->mg_preload_node); |
a25eb52e ZL |
2583 | put_css_set(src_cset); |
2584 | put_css_set(dst_cset); | |
f817de98 TH |
2585 | continue; |
2586 | } | |
2587 | ||
1958d2d5 TH |
2588 | src_cset->mg_dst_cset = dst_cset; |
2589 | ||
2590 | if (list_empty(&dst_cset->mg_preload_node)) | |
2591 | list_add(&dst_cset->mg_preload_node, &csets); | |
2592 | else | |
a25eb52e | 2593 | put_css_set(dst_cset); |
1958d2d5 TH |
2594 | } |
2595 | ||
f817de98 | 2596 | list_splice_tail(&csets, preloaded_csets); |
1958d2d5 TH |
2597 | return 0; |
2598 | err: | |
2599 | cgroup_migrate_finish(&csets); | |
2600 | return -ENOMEM; | |
2601 | } | |
2602 | ||
2603 | /** | |
2604 | * cgroup_migrate - migrate a process or task to a cgroup | |
1958d2d5 TH |
2605 | * @leader: the leader of the process or the task to migrate |
2606 | * @threadgroup: whether @leader points to the whole process or a single task | |
37ff9f8f | 2607 | * @root: cgroup root migration is taking place on |
1958d2d5 | 2608 | * |
37ff9f8f TH |
2609 | * Migrate a process or task denoted by @leader. If migrating a process, |
2610 | * the caller must be holding cgroup_threadgroup_rwsem. The caller is also | |
2611 | * responsible for invoking cgroup_migrate_add_src() and | |
1958d2d5 TH |
2612 | * cgroup_migrate_prepare_dst() on the targets before invoking this |
2613 | * function and following up with cgroup_migrate_finish(). | |
2614 | * | |
2615 | * As long as a controller's ->can_attach() doesn't fail, this function is | |
2616 | * guaranteed to succeed. This means that, excluding ->can_attach() | |
2617 | * failure, when migrating multiple targets, the success or failure can be | |
2618 | * decided for all targets by invoking group_migrate_prepare_dst() before | |
2619 | * actually starting migrating. | |
2620 | */ | |
9af2ec45 | 2621 | static int cgroup_migrate(struct task_struct *leader, bool threadgroup, |
37ff9f8f | 2622 | struct cgroup_root *root) |
74a1166d | 2623 | { |
adaae5dc TH |
2624 | struct cgroup_taskset tset = CGROUP_TASKSET_INIT(tset); |
2625 | struct task_struct *task; | |
74a1166d | 2626 | |
fb5d2b4c MSB |
2627 | /* |
2628 | * Prevent freeing of tasks while we take a snapshot. Tasks that are | |
2629 | * already PF_EXITING could be freed from underneath us unless we | |
2630 | * take an rcu_read_lock. | |
2631 | */ | |
f0d9a5f1 | 2632 | spin_lock_bh(&css_set_lock); |
fb5d2b4c | 2633 | rcu_read_lock(); |
9db8de37 | 2634 | task = leader; |
74a1166d | 2635 | do { |
adaae5dc | 2636 | cgroup_taskset_add(task, &tset); |
081aa458 LZ |
2637 | if (!threadgroup) |
2638 | break; | |
9db8de37 | 2639 | } while_each_thread(leader, task); |
fb5d2b4c | 2640 | rcu_read_unlock(); |
f0d9a5f1 | 2641 | spin_unlock_bh(&css_set_lock); |
74a1166d | 2642 | |
37ff9f8f | 2643 | return cgroup_taskset_migrate(&tset, root); |
74a1166d BB |
2644 | } |
2645 | ||
1958d2d5 TH |
2646 | /** |
2647 | * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup | |
2648 | * @dst_cgrp: the cgroup to attach to | |
2649 | * @leader: the task or the leader of the threadgroup to be attached | |
2650 | * @threadgroup: attach the whole threadgroup? | |
2651 | * | |
1ed13287 | 2652 | * Call holding cgroup_mutex and cgroup_threadgroup_rwsem. |
1958d2d5 TH |
2653 | */ |
2654 | static int cgroup_attach_task(struct cgroup *dst_cgrp, | |
2655 | struct task_struct *leader, bool threadgroup) | |
2656 | { | |
2657 | LIST_HEAD(preloaded_csets); | |
2658 | struct task_struct *task; | |
2659 | int ret; | |
2660 | ||
6c694c88 TH |
2661 | if (!cgroup_may_migrate_to(dst_cgrp)) |
2662 | return -EBUSY; | |
2663 | ||
1958d2d5 | 2664 | /* look up all src csets */ |
f0d9a5f1 | 2665 | spin_lock_bh(&css_set_lock); |
1958d2d5 TH |
2666 | rcu_read_lock(); |
2667 | task = leader; | |
2668 | do { | |
2669 | cgroup_migrate_add_src(task_css_set(task), dst_cgrp, | |
2670 | &preloaded_csets); | |
2671 | if (!threadgroup) | |
2672 | break; | |
2673 | } while_each_thread(leader, task); | |
2674 | rcu_read_unlock(); | |
f0d9a5f1 | 2675 | spin_unlock_bh(&css_set_lock); |
1958d2d5 TH |
2676 | |
2677 | /* prepare dst csets and commit */ | |
e4857982 | 2678 | ret = cgroup_migrate_prepare_dst(&preloaded_csets); |
1958d2d5 | 2679 | if (!ret) |
37ff9f8f | 2680 | ret = cgroup_migrate(leader, threadgroup, dst_cgrp->root); |
1958d2d5 TH |
2681 | |
2682 | cgroup_migrate_finish(&preloaded_csets); | |
2683 | return ret; | |
74a1166d BB |
2684 | } |
2685 | ||
187fe840 TH |
2686 | static int cgroup_procs_write_permission(struct task_struct *task, |
2687 | struct cgroup *dst_cgrp, | |
2688 | struct kernfs_open_file *of) | |
dedf22e9 TH |
2689 | { |
2690 | const struct cred *cred = current_cred(); | |
2691 | const struct cred *tcred = get_task_cred(task); | |
2692 | int ret = 0; | |
2693 | ||
2694 | /* | |
2695 | * even if we're attaching all tasks in the thread group, we only | |
2696 | * need to check permissions on one of them. | |
2697 | */ | |
2698 | if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) && | |
2699 | !uid_eq(cred->euid, tcred->uid) && | |
2700 | !uid_eq(cred->euid, tcred->suid)) | |
2701 | ret = -EACCES; | |
2702 | ||
187fe840 TH |
2703 | if (!ret && cgroup_on_dfl(dst_cgrp)) { |
2704 | struct super_block *sb = of->file->f_path.dentry->d_sb; | |
2705 | struct cgroup *cgrp; | |
2706 | struct inode *inode; | |
2707 | ||
f0d9a5f1 | 2708 | spin_lock_bh(&css_set_lock); |
187fe840 | 2709 | cgrp = task_cgroup_from_root(task, &cgrp_dfl_root); |
f0d9a5f1 | 2710 | spin_unlock_bh(&css_set_lock); |
187fe840 TH |
2711 | |
2712 | while (!cgroup_is_descendant(dst_cgrp, cgrp)) | |
2713 | cgrp = cgroup_parent(cgrp); | |
2714 | ||
2715 | ret = -ENOMEM; | |
6f60eade | 2716 | inode = kernfs_get_inode(sb, cgrp->procs_file.kn); |
187fe840 TH |
2717 | if (inode) { |
2718 | ret = inode_permission(inode, MAY_WRITE); | |
2719 | iput(inode); | |
2720 | } | |
2721 | } | |
2722 | ||
dedf22e9 TH |
2723 | put_cred(tcred); |
2724 | return ret; | |
2725 | } | |
2726 | ||
74a1166d BB |
2727 | /* |
2728 | * Find the task_struct of the task to attach by vpid and pass it along to the | |
cd3d0952 | 2729 | * function to attach either it or all tasks in its threadgroup. Will lock |
0e1d768f | 2730 | * cgroup_mutex and threadgroup. |
bbcb81d0 | 2731 | */ |
acbef755 TH |
2732 | static ssize_t __cgroup_procs_write(struct kernfs_open_file *of, char *buf, |
2733 | size_t nbytes, loff_t off, bool threadgroup) | |
bbcb81d0 | 2734 | { |
bbcb81d0 | 2735 | struct task_struct *tsk; |
e76ecaee | 2736 | struct cgroup *cgrp; |
acbef755 | 2737 | pid_t pid; |
bbcb81d0 PM |
2738 | int ret; |
2739 | ||
acbef755 TH |
2740 | if (kstrtoint(strstrip(buf), 0, &pid) || pid < 0) |
2741 | return -EINVAL; | |
2742 | ||
945ba199 | 2743 | cgrp = cgroup_kn_lock_live(of->kn, false); |
e76ecaee | 2744 | if (!cgrp) |
74a1166d BB |
2745 | return -ENODEV; |
2746 | ||
3014dde7 | 2747 | percpu_down_write(&cgroup_threadgroup_rwsem); |
b78949eb | 2748 | rcu_read_lock(); |
bbcb81d0 | 2749 | if (pid) { |
73507f33 | 2750 | tsk = find_task_by_vpid(pid); |
74a1166d | 2751 | if (!tsk) { |
dd4b0a46 | 2752 | ret = -ESRCH; |
3014dde7 | 2753 | goto out_unlock_rcu; |
bbcb81d0 | 2754 | } |
dedf22e9 | 2755 | } else { |
b78949eb | 2756 | tsk = current; |
dedf22e9 | 2757 | } |
cd3d0952 TH |
2758 | |
2759 | if (threadgroup) | |
b78949eb | 2760 | tsk = tsk->group_leader; |
c4c27fbd MG |
2761 | |
2762 | /* | |
14a40ffc | 2763 | * Workqueue threads may acquire PF_NO_SETAFFINITY and become |
c4c27fbd MG |
2764 | * trapped in a cpuset, or RT worker may be born in a cgroup |
2765 | * with no rt_runtime allocated. Just say no. | |
2766 | */ | |
14a40ffc | 2767 | if (tsk == kthreadd_task || (tsk->flags & PF_NO_SETAFFINITY)) { |
c4c27fbd | 2768 | ret = -EINVAL; |
3014dde7 | 2769 | goto out_unlock_rcu; |
c4c27fbd MG |
2770 | } |
2771 | ||
b78949eb MSB |
2772 | get_task_struct(tsk); |
2773 | rcu_read_unlock(); | |
2774 | ||
187fe840 | 2775 | ret = cgroup_procs_write_permission(tsk, cgrp, of); |
dedf22e9 TH |
2776 | if (!ret) |
2777 | ret = cgroup_attach_task(cgrp, tsk, threadgroup); | |
081aa458 | 2778 | |
f9f9e7b7 | 2779 | put_task_struct(tsk); |
3014dde7 TH |
2780 | goto out_unlock_threadgroup; |
2781 | ||
2782 | out_unlock_rcu: | |
2783 | rcu_read_unlock(); | |
2784 | out_unlock_threadgroup: | |
2785 | percpu_up_write(&cgroup_threadgroup_rwsem); | |
e76ecaee | 2786 | cgroup_kn_unlock(of->kn); |
e93ad19d | 2787 | cpuset_post_attach_flush(); |
acbef755 | 2788 | return ret ?: nbytes; |
bbcb81d0 PM |
2789 | } |
2790 | ||
7ae1bad9 TH |
2791 | /** |
2792 | * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from' | |
2793 | * @from: attach to all cgroups of a given task | |
2794 | * @tsk: the task to be attached | |
2795 | */ | |
2796 | int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk) | |
2797 | { | |
3dd06ffa | 2798 | struct cgroup_root *root; |
7ae1bad9 TH |
2799 | int retval = 0; |
2800 | ||
47cfcd09 | 2801 | mutex_lock(&cgroup_mutex); |
985ed670 | 2802 | for_each_root(root) { |
96d365e0 TH |
2803 | struct cgroup *from_cgrp; |
2804 | ||
3dd06ffa | 2805 | if (root == &cgrp_dfl_root) |
985ed670 TH |
2806 | continue; |
2807 | ||
f0d9a5f1 | 2808 | spin_lock_bh(&css_set_lock); |
96d365e0 | 2809 | from_cgrp = task_cgroup_from_root(from, root); |
f0d9a5f1 | 2810 | spin_unlock_bh(&css_set_lock); |
7ae1bad9 | 2811 | |
6f4b7e63 | 2812 | retval = cgroup_attach_task(from_cgrp, tsk, false); |
7ae1bad9 TH |
2813 | if (retval) |
2814 | break; | |
2815 | } | |
47cfcd09 | 2816 | mutex_unlock(&cgroup_mutex); |
7ae1bad9 TH |
2817 | |
2818 | return retval; | |
2819 | } | |
2820 | EXPORT_SYMBOL_GPL(cgroup_attach_task_all); | |
2821 | ||
acbef755 TH |
2822 | static ssize_t cgroup_tasks_write(struct kernfs_open_file *of, |
2823 | char *buf, size_t nbytes, loff_t off) | |
74a1166d | 2824 | { |
acbef755 | 2825 | return __cgroup_procs_write(of, buf, nbytes, off, false); |
74a1166d BB |
2826 | } |
2827 | ||
acbef755 TH |
2828 | static ssize_t cgroup_procs_write(struct kernfs_open_file *of, |
2829 | char *buf, size_t nbytes, loff_t off) | |
af351026 | 2830 | { |
acbef755 | 2831 | return __cgroup_procs_write(of, buf, nbytes, off, true); |
af351026 PM |
2832 | } |
2833 | ||
451af504 TH |
2834 | static ssize_t cgroup_release_agent_write(struct kernfs_open_file *of, |
2835 | char *buf, size_t nbytes, loff_t off) | |
e788e066 | 2836 | { |
e76ecaee | 2837 | struct cgroup *cgrp; |
5f469907 | 2838 | |
e76ecaee | 2839 | BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX); |
5f469907 | 2840 | |
945ba199 | 2841 | cgrp = cgroup_kn_lock_live(of->kn, false); |
e76ecaee | 2842 | if (!cgrp) |
e788e066 | 2843 | return -ENODEV; |
69e943b7 | 2844 | spin_lock(&release_agent_path_lock); |
e76ecaee TH |
2845 | strlcpy(cgrp->root->release_agent_path, strstrip(buf), |
2846 | sizeof(cgrp->root->release_agent_path)); | |
69e943b7 | 2847 | spin_unlock(&release_agent_path_lock); |
e76ecaee | 2848 | cgroup_kn_unlock(of->kn); |
451af504 | 2849 | return nbytes; |
e788e066 PM |
2850 | } |
2851 | ||
2da8ca82 | 2852 | static int cgroup_release_agent_show(struct seq_file *seq, void *v) |
e788e066 | 2853 | { |
2da8ca82 | 2854 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
182446d0 | 2855 | |
46cfeb04 | 2856 | spin_lock(&release_agent_path_lock); |
e788e066 | 2857 | seq_puts(seq, cgrp->root->release_agent_path); |
46cfeb04 | 2858 | spin_unlock(&release_agent_path_lock); |
e788e066 | 2859 | seq_putc(seq, '\n'); |
e788e066 PM |
2860 | return 0; |
2861 | } | |
2862 | ||
2da8ca82 | 2863 | static int cgroup_sane_behavior_show(struct seq_file *seq, void *v) |
873fe09e | 2864 | { |
c1d5d42e | 2865 | seq_puts(seq, "0\n"); |
e788e066 PM |
2866 | return 0; |
2867 | } | |
2868 | ||
6e5c8307 | 2869 | static void cgroup_print_ss_mask(struct seq_file *seq, u16 ss_mask) |
355e0c48 | 2870 | { |
f8f22e53 TH |
2871 | struct cgroup_subsys *ss; |
2872 | bool printed = false; | |
2873 | int ssid; | |
a742c59d | 2874 | |
b4e0eeaf | 2875 | do_each_subsys_mask(ss, ssid, ss_mask) { |
a966a4ed AS |
2876 | if (printed) |
2877 | seq_putc(seq, ' '); | |
2878 | seq_printf(seq, "%s", ss->name); | |
2879 | printed = true; | |
b4e0eeaf | 2880 | } while_each_subsys_mask(); |
f8f22e53 TH |
2881 | if (printed) |
2882 | seq_putc(seq, '\n'); | |
355e0c48 PM |
2883 | } |
2884 | ||
f8f22e53 TH |
2885 | /* show controllers which are enabled from the parent */ |
2886 | static int cgroup_controllers_show(struct seq_file *seq, void *v) | |
ddbcc7e8 | 2887 | { |
f8f22e53 TH |
2888 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
2889 | ||
5531dc91 | 2890 | cgroup_print_ss_mask(seq, cgroup_control(cgrp)); |
f8f22e53 | 2891 | return 0; |
ddbcc7e8 PM |
2892 | } |
2893 | ||
f8f22e53 TH |
2894 | /* show controllers which are enabled for a given cgroup's children */ |
2895 | static int cgroup_subtree_control_show(struct seq_file *seq, void *v) | |
ddbcc7e8 | 2896 | { |
f8f22e53 TH |
2897 | struct cgroup *cgrp = seq_css(seq)->cgroup; |
2898 | ||
667c2491 | 2899 | cgroup_print_ss_mask(seq, cgrp->subtree_control); |
f8f22e53 TH |
2900 | return 0; |
2901 | } | |
2902 | ||
2903 | /** | |
2904 | * cgroup_update_dfl_csses - update css assoc of a subtree in default hierarchy | |
2905 | * @cgrp: root of the subtree to update csses for | |
2906 | * | |
54962604 TH |
2907 | * @cgrp's control masks have changed and its subtree's css associations |
2908 | * need to be updated accordingly. This function looks up all css_sets | |
2909 | * which are attached to the subtree, creates the matching updated css_sets | |
2910 | * and migrates the tasks to the new ones. | |
f8f22e53 TH |
2911 | */ |
2912 | static int cgroup_update_dfl_csses(struct cgroup *cgrp) | |
2913 | { | |
2914 | LIST_HEAD(preloaded_csets); | |
10265075 | 2915 | struct cgroup_taskset tset = CGROUP_TASKSET_INIT(tset); |
54962604 TH |
2916 | struct cgroup_subsys_state *d_css; |
2917 | struct cgroup *dsct; | |
f8f22e53 TH |
2918 | struct css_set *src_cset; |
2919 | int ret; | |
2920 | ||
f8f22e53 TH |
2921 | lockdep_assert_held(&cgroup_mutex); |
2922 | ||
3014dde7 TH |
2923 | percpu_down_write(&cgroup_threadgroup_rwsem); |
2924 | ||
f8f22e53 | 2925 | /* look up all csses currently attached to @cgrp's subtree */ |
f0d9a5f1 | 2926 | spin_lock_bh(&css_set_lock); |
54962604 | 2927 | cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) { |
f8f22e53 TH |
2928 | struct cgrp_cset_link *link; |
2929 | ||
54962604 | 2930 | list_for_each_entry(link, &dsct->cset_links, cset_link) |
58cdb1ce | 2931 | cgroup_migrate_add_src(link->cset, dsct, |
f8f22e53 TH |
2932 | &preloaded_csets); |
2933 | } | |
f0d9a5f1 | 2934 | spin_unlock_bh(&css_set_lock); |
f8f22e53 TH |
2935 | |
2936 | /* NULL dst indicates self on default hierarchy */ | |
e4857982 | 2937 | ret = cgroup_migrate_prepare_dst(&preloaded_csets); |
f8f22e53 TH |
2938 | if (ret) |
2939 | goto out_finish; | |
2940 | ||
f0d9a5f1 | 2941 | spin_lock_bh(&css_set_lock); |
f8f22e53 | 2942 | list_for_each_entry(src_cset, &preloaded_csets, mg_preload_node) { |
10265075 | 2943 | struct task_struct *task, *ntask; |
f8f22e53 TH |
2944 | |
2945 | /* src_csets precede dst_csets, break on the first dst_cset */ | |
2946 | if (!src_cset->mg_src_cgrp) | |
2947 | break; | |
2948 | ||
10265075 TH |
2949 | /* all tasks in src_csets need to be migrated */ |
2950 | list_for_each_entry_safe(task, ntask, &src_cset->tasks, cg_list) | |
2951 | cgroup_taskset_add(task, &tset); | |
f8f22e53 | 2952 | } |
f0d9a5f1 | 2953 | spin_unlock_bh(&css_set_lock); |
f8f22e53 | 2954 | |
37ff9f8f | 2955 | ret = cgroup_taskset_migrate(&tset, cgrp->root); |
f8f22e53 TH |
2956 | out_finish: |
2957 | cgroup_migrate_finish(&preloaded_csets); | |
3014dde7 | 2958 | percpu_up_write(&cgroup_threadgroup_rwsem); |
f8f22e53 TH |
2959 | return ret; |
2960 | } | |
2961 | ||
1b9b96a1 | 2962 | /** |
945ba199 | 2963 | * cgroup_lock_and_drain_offline - lock cgroup_mutex and drain offlined csses |
ce3f1d9d | 2964 | * @cgrp: root of the target subtree |
1b9b96a1 TH |
2965 | * |
2966 | * Because css offlining is asynchronous, userland may try to re-enable a | |
945ba199 TH |
2967 | * controller while the previous css is still around. This function grabs |
2968 | * cgroup_mutex and drains the previous css instances of @cgrp's subtree. | |
1b9b96a1 | 2969 | */ |
945ba199 TH |
2970 | static void cgroup_lock_and_drain_offline(struct cgroup *cgrp) |
2971 | __acquires(&cgroup_mutex) | |
1b9b96a1 TH |
2972 | { |
2973 | struct cgroup *dsct; | |
ce3f1d9d | 2974 | struct cgroup_subsys_state *d_css; |
1b9b96a1 TH |
2975 | struct cgroup_subsys *ss; |
2976 | int ssid; | |
2977 | ||
945ba199 TH |
2978 | restart: |
2979 | mutex_lock(&cgroup_mutex); | |
1b9b96a1 | 2980 | |
ce3f1d9d | 2981 | cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) { |
1b9b96a1 TH |
2982 | for_each_subsys(ss, ssid) { |
2983 | struct cgroup_subsys_state *css = cgroup_css(dsct, ss); | |
2984 | DEFINE_WAIT(wait); | |
2985 | ||
ce3f1d9d | 2986 | if (!css || !percpu_ref_is_dying(&css->refcnt)) |
1b9b96a1 TH |
2987 | continue; |
2988 | ||
2989 | cgroup_get(dsct); | |
2990 | prepare_to_wait(&dsct->offline_waitq, &wait, | |
2991 | TASK_UNINTERRUPTIBLE); | |
2992 | ||
2993 | mutex_unlock(&cgroup_mutex); | |
2994 | schedule(); | |
2995 | finish_wait(&dsct->offline_waitq, &wait); | |
1b9b96a1 TH |
2996 | |
2997 | cgroup_put(dsct); | |
945ba199 | 2998 | goto restart; |
1b9b96a1 TH |
2999 | } |
3000 | } | |
1b9b96a1 TH |
3001 | } |
3002 | ||
15a27c36 TH |
3003 | /** |
3004 | * cgroup_save_control - save control masks of a subtree | |
3005 | * @cgrp: root of the target subtree | |
3006 | * | |
3007 | * Save ->subtree_control and ->subtree_ss_mask to the respective old_ | |
3008 | * prefixed fields for @cgrp's subtree including @cgrp itself. | |
3009 | */ | |
3010 | static void cgroup_save_control(struct cgroup *cgrp) | |
3011 | { | |
3012 | struct cgroup *dsct; | |
3013 | struct cgroup_subsys_state *d_css; | |
3014 | ||
3015 | cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) { | |
3016 | dsct->old_subtree_control = dsct->subtree_control; | |
3017 | dsct->old_subtree_ss_mask = dsct->subtree_ss_mask; | |
3018 | } | |
3019 | } | |
3020 | ||
3021 | /** | |
3022 | * cgroup_propagate_control - refresh control masks of a subtree | |
3023 | * @cgrp: root of the target subtree | |
3024 | * | |
3025 | * For @cgrp and its subtree, ensure ->subtree_ss_mask matches | |
3026 | * ->subtree_control and propagate controller availability through the | |
3027 | * subtree so that descendants don't have unavailable controllers enabled. | |
3028 | */ | |
3029 | static void cgroup_propagate_control(struct cgroup *cgrp) | |
3030 | { | |
3031 | struct cgroup *dsct; | |
3032 | struct cgroup_subsys_state *d_css; | |
3033 | ||
3034 | cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) { | |
3035 | dsct->subtree_control &= cgroup_control(dsct); | |
5ced2518 TH |
3036 | dsct->subtree_ss_mask = |
3037 | cgroup_calc_subtree_ss_mask(dsct->subtree_control, | |
3038 | cgroup_ss_mask(dsct)); | |
15a27c36 TH |
3039 | } |
3040 | } | |
3041 | ||
3042 | /** | |
3043 | * cgroup_restore_control - restore control masks of a subtree | |
3044 | * @cgrp: root of the target subtree | |
3045 | * | |
3046 | * Restore ->subtree_control and ->subtree_ss_mask from the respective old_ | |
3047 | * prefixed fields for @cgrp's subtree including @cgrp itself. | |
3048 | */ | |
3049 | static void cgroup_restore_control(struct cgroup *cgrp) | |
3050 | { | |
3051 | struct cgroup *dsct; | |
3052 | struct cgroup_subsys_state *d_css; | |
3053 | ||
3054 | cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) { | |
3055 | dsct->subtree_control = dsct->old_subtree_control; | |
3056 | dsct->subtree_ss_mask = dsct->old_subtree_ss_mask; | |
3057 | } | |
3058 | } | |
3059 | ||
f6d635ad TH |
3060 | static bool css_visible(struct cgroup_subsys_state *css) |
3061 | { | |
3062 | struct cgroup_subsys *ss = css->ss; | |
3063 | struct cgroup *cgrp = css->cgroup; | |
3064 | ||
3065 | if (cgroup_control(cgrp) & (1 << ss->id)) | |
3066 | return true; | |
3067 | if (!(cgroup_ss_mask(cgrp) & (1 << ss->id))) | |
3068 | return false; | |
3069 | return cgroup_on_dfl(cgrp) && ss->implicit_on_dfl; | |
3070 | } | |
3071 | ||
bdb53bd7 TH |
3072 | /** |
3073 | * cgroup_apply_control_enable - enable or show csses according to control | |
ce3f1d9d | 3074 | * @cgrp: root of the target subtree |
bdb53bd7 | 3075 | * |
ce3f1d9d | 3076 | * Walk @cgrp's subtree and create new csses or make the existing ones |
bdb53bd7 TH |
3077 | * visible. A css is created invisible if it's being implicitly enabled |
3078 | * through dependency. An invisible css is made visible when the userland | |
3079 | * explicitly enables it. | |
3080 | * | |
3081 | * Returns 0 on success, -errno on failure. On failure, csses which have | |
3082 | * been processed already aren't cleaned up. The caller is responsible for | |
3083 | * cleaning up with cgroup_apply_control_disble(). | |
3084 | */ | |
3085 | static int cgroup_apply_control_enable(struct cgroup *cgrp) | |
3086 | { | |
3087 | struct cgroup *dsct; | |
ce3f1d9d | 3088 | struct cgroup_subsys_state *d_css; |
bdb53bd7 TH |
3089 | struct cgroup_subsys *ss; |
3090 | int ssid, ret; | |
3091 | ||
ce3f1d9d | 3092 | cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) { |
bdb53bd7 TH |
3093 | for_each_subsys(ss, ssid) { |
3094 | struct cgroup_subsys_state *css = cgroup_css(dsct, ss); | |
3095 | ||
945ba199 TH |
3096 | WARN_ON_ONCE(css && percpu_ref_is_dying(&css->refcnt)); |
3097 | ||
bdb53bd7 TH |
3098 | if (!(cgroup_ss_mask(dsct) & (1 << ss->id))) |
3099 | continue; | |
3100 | ||
3101 | if (!css) { | |
3102 | css = css_create(dsct, ss); | |
3103 | if (IS_ERR(css)) | |
3104 | return PTR_ERR(css); | |
3105 | } | |
3106 | ||
f6d635ad | 3107 | if (css_visible(css)) { |
334c3679 | 3108 | ret = css_populate_dir(css); |
bdb53bd7 TH |
3109 | if (ret) |
3110 | return ret; | |
3111 | } | |
3112 | } | |
3113 | } | |
3114 | ||
3115 | return 0; | |
3116 | } | |
3117 | ||
12b3bb6a TH |
3118 | /** |
3119 | * cgroup_apply_control_disable - kill or hide csses according to control | |
ce3f1d9d | 3120 | * @cgrp: root of the target subtree |
12b3bb6a | 3121 | * |
ce3f1d9d | 3122 | * Walk @cgrp's subtree and kill and hide csses so that they match |
12b3bb6a TH |
3123 | * cgroup_ss_mask() and cgroup_visible_mask(). |
3124 | * | |
3125 | * A css is hidden when the userland requests it to be disabled while other | |
3126 | * subsystems are still depending on it. The css must not actively control | |
3127 | * resources and be in the vanilla state if it's made visible again later. | |
3128 | * Controllers which may be depended upon should provide ->css_reset() for | |
3129 | * this purpose. | |
3130 | */ | |
3131 | static void cgroup_apply_control_disable(struct cgroup *cgrp) | |
3132 | { | |
3133 | struct cgroup *dsct; | |
ce3f1d9d | 3134 | struct cgroup_subsys_state *d_css; |
12b3bb6a TH |
3135 | struct cgroup_subsys *ss; |
3136 | int ssid; | |
3137 | ||
ce3f1d9d | 3138 | cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) { |
12b3bb6a TH |
3139 | for_each_subsys(ss, ssid) { |
3140 | struct cgroup_subsys_state *css = cgroup_css(dsct, ss); | |
3141 | ||
945ba199 TH |
3142 | WARN_ON_ONCE(css && percpu_ref_is_dying(&css->refcnt)); |
3143 | ||
12b3bb6a TH |
3144 | if (!css) |
3145 | continue; | |
3146 | ||
334c3679 TH |
3147 | if (css->parent && |
3148 | !(cgroup_ss_mask(dsct) & (1 << ss->id))) { | |
12b3bb6a | 3149 | kill_css(css); |
f6d635ad | 3150 | } else if (!css_visible(css)) { |
334c3679 | 3151 | css_clear_dir(css); |
12b3bb6a TH |
3152 | if (ss->css_reset) |
3153 | ss->css_reset(css); | |
3154 | } | |
3155 | } | |
3156 | } | |
3157 | } | |
3158 | ||
f7b2814b TH |
3159 | /** |
3160 | * cgroup_apply_control - apply control mask updates to the subtree | |
3161 | * @cgrp: root of the target subtree | |
3162 | * | |
3163 | * subsystems can be enabled and disabled in a subtree using the following | |
3164 | * steps. | |
3165 | * | |
3166 | * 1. Call cgroup_save_control() to stash the current state. | |
3167 | * 2. Update ->subtree_control masks in the subtree as desired. | |
3168 | * 3. Call cgroup_apply_control() to apply the changes. | |
3169 | * 4. Optionally perform other related operations. | |
3170 | * 5. Call cgroup_finalize_control() to finish up. | |
3171 | * | |
3172 | * This function implements step 3 and propagates the mask changes | |
3173 | * throughout @cgrp's subtree, updates csses accordingly and perform | |
3174 | * process migrations. | |
3175 | */ | |
3176 | static int cgroup_apply_control(struct cgroup *cgrp) | |
3177 | { | |
3178 | int ret; | |
3179 | ||
3180 | cgroup_propagate_control(cgrp); | |
3181 | ||
3182 | ret = cgroup_apply_control_enable(cgrp); | |
3183 | if (ret) | |
3184 | return ret; | |
3185 | ||
3186 | /* | |
3187 | * At this point, cgroup_e_css() results reflect the new csses | |
3188 | * making the following cgroup_update_dfl_csses() properly update | |
3189 | * css associations of all tasks in the subtree. | |
3190 | */ | |
3191 | ret = cgroup_update_dfl_csses(cgrp); | |
3192 | if (ret) | |
3193 | return ret; | |
3194 | ||
3195 | return 0; | |
3196 | } | |
3197 | ||
3198 | /** | |
3199 | * cgroup_finalize_control - finalize control mask update | |
3200 | * @cgrp: root of the target subtree | |
3201 | * @ret: the result of the update | |
3202 | * | |
3203 | * Finalize control mask update. See cgroup_apply_control() for more info. | |
3204 | */ | |
3205 | static void cgroup_finalize_control(struct cgroup *cgrp, int ret) | |
3206 | { | |
3207 | if (ret) { | |
3208 | cgroup_restore_control(cgrp); | |
3209 | cgroup_propagate_control(cgrp); | |
3210 | } | |
3211 | ||
3212 | cgroup_apply_control_disable(cgrp); | |
3213 | } | |
3214 | ||
f8f22e53 | 3215 | /* change the enabled child controllers for a cgroup in the default hierarchy */ |
451af504 TH |
3216 | static ssize_t cgroup_subtree_control_write(struct kernfs_open_file *of, |
3217 | char *buf, size_t nbytes, | |
3218 | loff_t off) | |
f8f22e53 | 3219 | { |
6e5c8307 | 3220 | u16 enable = 0, disable = 0; |
a9746d8d | 3221 | struct cgroup *cgrp, *child; |
f8f22e53 | 3222 | struct cgroup_subsys *ss; |
451af504 | 3223 | char *tok; |
f8f22e53 TH |
3224 | int ssid, ret; |
3225 | ||
3226 | /* | |
d37167ab TH |
3227 | * Parse input - space separated list of subsystem names prefixed |
3228 | * with either + or -. | |
f8f22e53 | 3229 | */ |
451af504 TH |
3230 | buf = strstrip(buf); |
3231 | while ((tok = strsep(&buf, " "))) { | |
d37167ab TH |
3232 | if (tok[0] == '\0') |
3233 | continue; | |
a7165264 | 3234 | do_each_subsys_mask(ss, ssid, ~cgrp_dfl_inhibit_ss_mask) { |
fc5ed1e9 TH |
3235 | if (!cgroup_ssid_enabled(ssid) || |
3236 | strcmp(tok + 1, ss->name)) | |
f8f22e53 TH |
3237 | continue; |
3238 | ||
3239 | if (*tok == '+') { | |
7d331fa9 TH |
3240 | enable |= 1 << ssid; |
3241 | disable &= ~(1 << ssid); | |
f8f22e53 | 3242 | } else if (*tok == '-') { |
7d331fa9 TH |
3243 | disable |= 1 << ssid; |
3244 | enable &= ~(1 << ssid); | |
f8f22e53 TH |
3245 | } else { |
3246 | return -EINVAL; | |
3247 | } | |
3248 | break; | |
b4e0eeaf | 3249 | } while_each_subsys_mask(); |
f8f22e53 TH |
3250 | if (ssid == CGROUP_SUBSYS_COUNT) |
3251 | return -EINVAL; | |
3252 | } | |
3253 | ||
945ba199 | 3254 | cgrp = cgroup_kn_lock_live(of->kn, true); |
a9746d8d TH |
3255 | if (!cgrp) |
3256 | return -ENODEV; | |
f8f22e53 TH |
3257 | |
3258 | for_each_subsys(ss, ssid) { | |
3259 | if (enable & (1 << ssid)) { | |
667c2491 | 3260 | if (cgrp->subtree_control & (1 << ssid)) { |
f8f22e53 TH |
3261 | enable &= ~(1 << ssid); |
3262 | continue; | |
3263 | } | |
3264 | ||
5531dc91 | 3265 | if (!(cgroup_control(cgrp) & (1 << ssid))) { |
c29adf24 TH |
3266 | ret = -ENOENT; |
3267 | goto out_unlock; | |
3268 | } | |
f8f22e53 | 3269 | } else if (disable & (1 << ssid)) { |
667c2491 | 3270 | if (!(cgrp->subtree_control & (1 << ssid))) { |
f8f22e53 TH |
3271 | disable &= ~(1 << ssid); |
3272 | continue; | |
3273 | } | |
3274 | ||
3275 | /* a child has it enabled? */ | |
3276 | cgroup_for_each_live_child(child, cgrp) { | |
667c2491 | 3277 | if (child->subtree_control & (1 << ssid)) { |
f8f22e53 | 3278 | ret = -EBUSY; |
ddab2b6e | 3279 | goto out_unlock; |
f8f22e53 TH |
3280 | } |
3281 | } | |
3282 | } | |
3283 | } | |
3284 | ||
3285 | if (!enable && !disable) { | |
3286 | ret = 0; | |
ddab2b6e | 3287 | goto out_unlock; |
f8f22e53 TH |
3288 | } |
3289 | ||
3290 | /* | |
667c2491 | 3291 | * Except for the root, subtree_control must be zero for a cgroup |
f8f22e53 TH |
3292 | * with tasks so that child cgroups don't compete against tasks. |
3293 | */ | |
d51f39b0 | 3294 | if (enable && cgroup_parent(cgrp) && !list_empty(&cgrp->cset_links)) { |
f8f22e53 TH |
3295 | ret = -EBUSY; |
3296 | goto out_unlock; | |
3297 | } | |
3298 | ||
15a27c36 TH |
3299 | /* save and update control masks and prepare csses */ |
3300 | cgroup_save_control(cgrp); | |
f63070d3 | 3301 | |
15a27c36 TH |
3302 | cgrp->subtree_control |= enable; |
3303 | cgrp->subtree_control &= ~disable; | |
c29adf24 | 3304 | |
f7b2814b | 3305 | ret = cgroup_apply_control(cgrp); |
f8f22e53 | 3306 | |
f7b2814b | 3307 | cgroup_finalize_control(cgrp, ret); |
f8f22e53 TH |
3308 | |
3309 | kernfs_activate(cgrp->kn); | |
3310 | ret = 0; | |
3311 | out_unlock: | |
a9746d8d | 3312 | cgroup_kn_unlock(of->kn); |
451af504 | 3313 | return ret ?: nbytes; |
f8f22e53 TH |
3314 | } |
3315 | ||
4a07c222 | 3316 | static int cgroup_events_show(struct seq_file *seq, void *v) |
842b597e | 3317 | { |
4a07c222 | 3318 | seq_printf(seq, "populated %d\n", |
27bd4dbb | 3319 | cgroup_is_populated(seq_css(seq)->cgroup)); |
842b597e TH |
3320 | return 0; |
3321 | } | |
3322 | ||
2bd59d48 TH |
3323 | static ssize_t cgroup_file_write(struct kernfs_open_file *of, char *buf, |
3324 | size_t nbytes, loff_t off) | |
355e0c48 | 3325 | { |
2bd59d48 TH |
3326 | struct cgroup *cgrp = of->kn->parent->priv; |
3327 | struct cftype *cft = of->kn->priv; | |
3328 | struct cgroup_subsys_state *css; | |
a742c59d | 3329 | int ret; |
355e0c48 | 3330 | |
b4168640 TH |
3331 | if (cft->write) |
3332 | return cft->write(of, buf, nbytes, off); | |
3333 | ||
2bd59d48 TH |
3334 | /* |
3335 | * kernfs guarantees that a file isn't deleted with operations in | |
3336 | * flight, which means that the matching css is and stays alive and | |
3337 | * doesn't need to be pinned. The RCU locking is not necessary | |
3338 | * either. It's just for the convenience of using cgroup_css(). | |
3339 | */ | |
3340 | rcu_read_lock(); | |
3341 | css = cgroup_css(cgrp, cft->ss); | |
3342 | rcu_read_unlock(); | |
a742c59d | 3343 | |
451af504 | 3344 | if (cft->write_u64) { |
a742c59d TH |
3345 | unsigned long long v; |
3346 | ret = kstrtoull(buf, 0, &v); | |
3347 | if (!ret) | |
3348 | ret = cft->write_u64(css, cft, v); | |
3349 | } else if (cft->write_s64) { | |
3350 | long long v; | |
3351 | ret = kstrtoll(buf, 0, &v); | |
3352 | if (!ret) | |
3353 | ret = cft->write_s64(css, cft, v); | |
e73d2c61 | 3354 | } else { |
a742c59d | 3355 | ret = -EINVAL; |
e73d2c61 | 3356 | } |
2bd59d48 | 3357 | |
a742c59d | 3358 | return ret ?: nbytes; |
355e0c48 PM |
3359 | } |
3360 | ||
6612f05b | 3361 | static void *cgroup_seqfile_start(struct seq_file *seq, loff_t *ppos) |
db3b1497 | 3362 | { |
2bd59d48 | 3363 | return seq_cft(seq)->seq_start(seq, ppos); |
db3b1497 PM |
3364 | } |
3365 | ||
6612f05b | 3366 | static void *cgroup_seqfile_next(struct seq_file *seq, void *v, loff_t *ppos) |
ddbcc7e8 | 3367 | { |
2bd59d48 | 3368 | return seq_cft(seq)->seq_next(seq, v, ppos); |
ddbcc7e8 PM |
3369 | } |
3370 | ||
6612f05b | 3371 | static void cgroup_seqfile_stop(struct seq_file *seq, void *v) |
ddbcc7e8 | 3372 | { |
2bd59d48 | 3373 | seq_cft(seq)->seq_stop(seq, v); |
ddbcc7e8 PM |
3374 | } |
3375 | ||
91796569 | 3376 | static int cgroup_seqfile_show(struct seq_file *m, void *arg) |
e73d2c61 | 3377 | { |
7da11279 TH |
3378 | struct cftype *cft = seq_cft(m); |
3379 | struct cgroup_subsys_state *css = seq_css(m); | |
e73d2c61 | 3380 | |
2da8ca82 TH |
3381 | if (cft->seq_show) |
3382 | return cft->seq_show(m, arg); | |
e73d2c61 | 3383 | |
f4c753b7 | 3384 | if (cft->read_u64) |
896f5199 TH |
3385 | seq_printf(m, "%llu\n", cft->read_u64(css, cft)); |
3386 | else if (cft->read_s64) | |
3387 | seq_printf(m, "%lld\n", cft->read_s64(css, cft)); | |
3388 | else | |
3389 | return -EINVAL; | |
3390 | return 0; | |
91796569 PM |
3391 | } |
3392 | ||
2bd59d48 TH |
3393 | static struct kernfs_ops cgroup_kf_single_ops = { |
3394 | .atomic_write_len = PAGE_SIZE, | |
3395 | .write = cgroup_file_write, | |
3396 | .seq_show = cgroup_seqfile_show, | |
91796569 PM |
3397 | }; |
3398 | ||
2bd59d48 TH |
3399 | static struct kernfs_ops cgroup_kf_ops = { |
3400 | .atomic_write_len = PAGE_SIZE, | |
3401 | .write = cgroup_file_write, | |
3402 | .seq_start = cgroup_seqfile_start, | |
3403 | .seq_next = cgroup_seqfile_next, | |
3404 | .seq_stop = cgroup_seqfile_stop, | |
3405 | .seq_show = cgroup_seqfile_show, | |
3406 | }; | |
ddbcc7e8 PM |
3407 | |
3408 | /* | |
3409 | * cgroup_rename - Only allow simple rename of directories in place. | |
3410 | */ | |
2bd59d48 TH |
3411 | static int cgroup_rename(struct kernfs_node *kn, struct kernfs_node *new_parent, |
3412 | const char *new_name_str) | |
ddbcc7e8 | 3413 | { |
2bd59d48 | 3414 | struct cgroup *cgrp = kn->priv; |
65dff759 | 3415 | int ret; |
65dff759 | 3416 | |
2bd59d48 | 3417 | if (kernfs_type(kn) != KERNFS_DIR) |
ddbcc7e8 | 3418 | return -ENOTDIR; |
2bd59d48 | 3419 | if (kn->parent != new_parent) |
ddbcc7e8 | 3420 | return -EIO; |
65dff759 | 3421 | |
6db8e85c TH |
3422 | /* |
3423 | * This isn't a proper migration and its usefulness is very | |
aa6ec29b | 3424 | * limited. Disallow on the default hierarchy. |
6db8e85c | 3425 | */ |
aa6ec29b | 3426 | if (cgroup_on_dfl(cgrp)) |
6db8e85c | 3427 | return -EPERM; |
099fca32 | 3428 | |
e1b2dc17 | 3429 | /* |
8353da1f | 3430 | * We're gonna grab cgroup_mutex which nests outside kernfs |
e1b2dc17 | 3431 | * active_ref. kernfs_rename() doesn't require active_ref |
8353da1f | 3432 | * protection. Break them before grabbing cgroup_mutex. |
e1b2dc17 TH |
3433 | */ |
3434 | kernfs_break_active_protection(new_parent); | |
3435 | kernfs_break_active_protection(kn); | |
099fca32 | 3436 | |
2bd59d48 | 3437 | mutex_lock(&cgroup_mutex); |
099fca32 | 3438 | |
2bd59d48 | 3439 | ret = kernfs_rename(kn, new_parent, new_name_str); |
099fca32 | 3440 | |
2bd59d48 | 3441 | mutex_unlock(&cgroup_mutex); |
e1b2dc17 TH |
3442 | |
3443 | kernfs_unbreak_active_protection(kn); | |
3444 | kernfs_unbreak_active_protection(new_parent); | |
2bd59d48 | 3445 | return ret; |
099fca32 LZ |
3446 | } |
3447 | ||
49957f8e TH |
3448 | /* set uid and gid of cgroup dirs and files to that of the creator */ |
3449 | static int cgroup_kn_set_ugid(struct kernfs_node *kn) | |
3450 | { | |
3451 | struct iattr iattr = { .ia_valid = ATTR_UID | ATTR_GID, | |
3452 | .ia_uid = current_fsuid(), | |
3453 | .ia_gid = current_fsgid(), }; | |
3454 | ||
3455 | if (uid_eq(iattr.ia_uid, GLOBAL_ROOT_UID) && | |
3456 | gid_eq(iattr.ia_gid, GLOBAL_ROOT_GID)) | |
3457 | return 0; | |
3458 | ||
3459 | return kernfs_setattr(kn, &iattr); | |
3460 | } | |
3461 | ||
4df8dc90 TH |
3462 | static int cgroup_add_file(struct cgroup_subsys_state *css, struct cgroup *cgrp, |
3463 | struct cftype *cft) | |
ddbcc7e8 | 3464 | { |
8d7e6fb0 | 3465 | char name[CGROUP_FILE_NAME_MAX]; |
2bd59d48 TH |
3466 | struct kernfs_node *kn; |
3467 | struct lock_class_key *key = NULL; | |
49957f8e | 3468 | int ret; |
05ef1d7c | 3469 | |
2bd59d48 TH |
3470 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
3471 | key = &cft->lockdep_key; | |
3472 | #endif | |
3473 | kn = __kernfs_create_file(cgrp->kn, cgroup_file_name(cgrp, cft, name), | |
3474 | cgroup_file_mode(cft), 0, cft->kf_ops, cft, | |
dfeb0750 | 3475 | NULL, key); |
49957f8e TH |
3476 | if (IS_ERR(kn)) |
3477 | return PTR_ERR(kn); | |
3478 | ||
3479 | ret = cgroup_kn_set_ugid(kn); | |
f8f22e53 | 3480 | if (ret) { |
49957f8e | 3481 | kernfs_remove(kn); |
f8f22e53 TH |
3482 | return ret; |
3483 | } | |
3484 | ||
6f60eade TH |
3485 | if (cft->file_offset) { |
3486 | struct cgroup_file *cfile = (void *)css + cft->file_offset; | |
3487 | ||
34c06254 | 3488 | spin_lock_irq(&cgroup_file_kn_lock); |
6f60eade | 3489 | cfile->kn = kn; |
34c06254 | 3490 | spin_unlock_irq(&cgroup_file_kn_lock); |
6f60eade TH |
3491 | } |
3492 | ||
f8f22e53 | 3493 | return 0; |
ddbcc7e8 PM |
3494 | } |
3495 | ||
b1f28d31 TH |
3496 | /** |
3497 | * cgroup_addrm_files - add or remove files to a cgroup directory | |
4df8dc90 TH |
3498 | * @css: the target css |
3499 | * @cgrp: the target cgroup (usually css->cgroup) | |
b1f28d31 TH |
3500 | * @cfts: array of cftypes to be added |
3501 | * @is_add: whether to add or remove | |
3502 | * | |
3503 | * Depending on @is_add, add or remove files defined by @cfts on @cgrp. | |
6732ed85 | 3504 | * For removals, this function never fails. |
b1f28d31 | 3505 | */ |
4df8dc90 TH |
3506 | static int cgroup_addrm_files(struct cgroup_subsys_state *css, |
3507 | struct cgroup *cgrp, struct cftype cfts[], | |
2bb566cb | 3508 | bool is_add) |
ddbcc7e8 | 3509 | { |
6732ed85 | 3510 | struct cftype *cft, *cft_end = NULL; |
b598dde3 | 3511 | int ret = 0; |
b1f28d31 | 3512 | |
01f6474c | 3513 | lockdep_assert_held(&cgroup_mutex); |
db0416b6 | 3514 | |
6732ed85 TH |
3515 | restart: |
3516 | for (cft = cfts; cft != cft_end && cft->name[0] != '\0'; cft++) { | |
f33fddc2 | 3517 | /* does cft->flags tell us to skip this file on @cgrp? */ |
05ebb6e6 | 3518 | if ((cft->flags & __CFTYPE_ONLY_ON_DFL) && !cgroup_on_dfl(cgrp)) |
8cbbf2c9 | 3519 | continue; |
05ebb6e6 | 3520 | if ((cft->flags & __CFTYPE_NOT_ON_DFL) && cgroup_on_dfl(cgrp)) |
873fe09e | 3521 | continue; |
d51f39b0 | 3522 | if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgroup_parent(cgrp)) |
f33fddc2 | 3523 | continue; |
d51f39b0 | 3524 | if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgroup_parent(cgrp)) |
f33fddc2 G |
3525 | continue; |
3526 | ||
2739d3cc | 3527 | if (is_add) { |
4df8dc90 | 3528 | ret = cgroup_add_file(css, cgrp, cft); |
b1f28d31 | 3529 | if (ret) { |
ed3d261b JP |
3530 | pr_warn("%s: failed to add %s, err=%d\n", |
3531 | __func__, cft->name, ret); | |
6732ed85 TH |
3532 | cft_end = cft; |
3533 | is_add = false; | |
3534 | goto restart; | |
b1f28d31 | 3535 | } |
2739d3cc LZ |
3536 | } else { |
3537 | cgroup_rm_file(cgrp, cft); | |
db0416b6 | 3538 | } |
ddbcc7e8 | 3539 | } |
b598dde3 | 3540 | return ret; |
ddbcc7e8 PM |
3541 | } |
3542 | ||
21a2d343 | 3543 | static int cgroup_apply_cftypes(struct cftype *cfts, bool is_add) |
8e3f6541 TH |
3544 | { |
3545 | LIST_HEAD(pending); | |
2bb566cb | 3546 | struct cgroup_subsys *ss = cfts[0].ss; |
3dd06ffa | 3547 | struct cgroup *root = &ss->root->cgrp; |
492eb21b | 3548 | struct cgroup_subsys_state *css; |
9ccece80 | 3549 | int ret = 0; |
8e3f6541 | 3550 | |
01f6474c | 3551 | lockdep_assert_held(&cgroup_mutex); |
e8c82d20 | 3552 | |
e8c82d20 | 3553 | /* add/rm files for all cgroups created before */ |
ca8bdcaf | 3554 | css_for_each_descendant_pre(css, cgroup_css(root, ss)) { |
492eb21b TH |
3555 | struct cgroup *cgrp = css->cgroup; |
3556 | ||
88cb04b9 | 3557 | if (!(css->flags & CSS_VISIBLE)) |
e8c82d20 LZ |
3558 | continue; |
3559 | ||
4df8dc90 | 3560 | ret = cgroup_addrm_files(css, cgrp, cfts, is_add); |
9ccece80 TH |
3561 | if (ret) |
3562 | break; | |
8e3f6541 | 3563 | } |
21a2d343 TH |
3564 | |
3565 | if (is_add && !ret) | |
3566 | kernfs_activate(root->kn); | |
9ccece80 | 3567 | return ret; |
8e3f6541 TH |
3568 | } |
3569 | ||
2da440a2 | 3570 | static void cgroup_exit_cftypes(struct cftype *cfts) |
8e3f6541 | 3571 | { |
2bb566cb | 3572 | struct cftype *cft; |
8e3f6541 | 3573 | |
2bd59d48 TH |
3574 | for (cft = cfts; cft->name[0] != '\0'; cft++) { |
3575 | /* free copy for custom atomic_write_len, see init_cftypes() */ | |
3576 | if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) | |
3577 | kfree(cft->kf_ops); | |
3578 | cft->kf_ops = NULL; | |
2da440a2 | 3579 | cft->ss = NULL; |
a8ddc821 TH |
3580 | |
3581 | /* revert flags set by cgroup core while adding @cfts */ | |
05ebb6e6 | 3582 | cft->flags &= ~(__CFTYPE_ONLY_ON_DFL | __CFTYPE_NOT_ON_DFL); |
2bd59d48 | 3583 | } |
2da440a2 TH |
3584 | } |
3585 | ||
2bd59d48 | 3586 | static int cgroup_init_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
2da440a2 TH |
3587 | { |
3588 | struct cftype *cft; | |
3589 | ||
2bd59d48 TH |
3590 | for (cft = cfts; cft->name[0] != '\0'; cft++) { |
3591 | struct kernfs_ops *kf_ops; | |
3592 | ||
0adb0704 TH |
3593 | WARN_ON(cft->ss || cft->kf_ops); |
3594 | ||
2bd59d48 TH |
3595 | if (cft->seq_start) |
3596 | kf_ops = &cgroup_kf_ops; | |
3597 | else | |
3598 | kf_ops = &cgroup_kf_single_ops; | |
3599 | ||
3600 | /* | |
3601 | * Ugh... if @cft wants a custom max_write_len, we need to | |
3602 | * make a copy of kf_ops to set its atomic_write_len. | |
3603 | */ | |
3604 | if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) { | |
3605 | kf_ops = kmemdup(kf_ops, sizeof(*kf_ops), GFP_KERNEL); | |
3606 | if (!kf_ops) { | |
3607 | cgroup_exit_cftypes(cfts); | |
3608 | return -ENOMEM; | |
3609 | } | |
3610 | kf_ops->atomic_write_len = cft->max_write_len; | |
3611 | } | |
8e3f6541 | 3612 | |
2bd59d48 | 3613 | cft->kf_ops = kf_ops; |
2bb566cb | 3614 | cft->ss = ss; |
2bd59d48 | 3615 | } |
2bb566cb | 3616 | |
2bd59d48 | 3617 | return 0; |
2da440a2 TH |
3618 | } |
3619 | ||
21a2d343 TH |
3620 | static int cgroup_rm_cftypes_locked(struct cftype *cfts) |
3621 | { | |
01f6474c | 3622 | lockdep_assert_held(&cgroup_mutex); |
21a2d343 TH |
3623 | |
3624 | if (!cfts || !cfts[0].ss) | |
3625 | return -ENOENT; | |
3626 | ||
3627 | list_del(&cfts->node); | |
3628 | cgroup_apply_cftypes(cfts, false); | |
3629 | cgroup_exit_cftypes(cfts); | |
3630 | return 0; | |
8e3f6541 | 3631 | } |
8e3f6541 | 3632 | |
79578621 TH |
3633 | /** |
3634 | * cgroup_rm_cftypes - remove an array of cftypes from a subsystem | |
79578621 TH |
3635 | * @cfts: zero-length name terminated array of cftypes |
3636 | * | |
2bb566cb TH |
3637 | * Unregister @cfts. Files described by @cfts are removed from all |
3638 | * existing cgroups and all future cgroups won't have them either. This | |
3639 | * function can be called anytime whether @cfts' subsys is attached or not. | |
79578621 TH |
3640 | * |
3641 | * Returns 0 on successful unregistration, -ENOENT if @cfts is not | |
2bb566cb | 3642 | * registered. |
79578621 | 3643 | */ |
2bb566cb | 3644 | int cgroup_rm_cftypes(struct cftype *cfts) |
79578621 | 3645 | { |
21a2d343 | 3646 | int ret; |
79578621 | 3647 | |
01f6474c | 3648 | mutex_lock(&cgroup_mutex); |
21a2d343 | 3649 | ret = cgroup_rm_cftypes_locked(cfts); |
01f6474c | 3650 | mutex_unlock(&cgroup_mutex); |
21a2d343 | 3651 | return ret; |
80b13586 TH |
3652 | } |
3653 | ||
8e3f6541 TH |
3654 | /** |
3655 | * cgroup_add_cftypes - add an array of cftypes to a subsystem | |
3656 | * @ss: target cgroup subsystem | |
3657 | * @cfts: zero-length name terminated array of cftypes | |
3658 | * | |
3659 | * Register @cfts to @ss. Files described by @cfts are created for all | |
3660 | * existing cgroups to which @ss is attached and all future cgroups will | |
3661 | * have them too. This function can be called anytime whether @ss is | |
3662 | * attached or not. | |
3663 | * | |
3664 | * Returns 0 on successful registration, -errno on failure. Note that this | |
3665 | * function currently returns 0 as long as @cfts registration is successful | |
3666 | * even if some file creation attempts on existing cgroups fail. | |
3667 | */ | |
2cf669a5 | 3668 | static int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
8e3f6541 | 3669 | { |
9ccece80 | 3670 | int ret; |
8e3f6541 | 3671 | |
fc5ed1e9 | 3672 | if (!cgroup_ssid_enabled(ss->id)) |
c731ae1d LZ |
3673 | return 0; |
3674 | ||
dc5736ed LZ |
3675 | if (!cfts || cfts[0].name[0] == '\0') |
3676 | return 0; | |
2bb566cb | 3677 | |
2bd59d48 TH |
3678 | ret = cgroup_init_cftypes(ss, cfts); |
3679 | if (ret) | |
3680 | return ret; | |
79578621 | 3681 | |
01f6474c | 3682 | mutex_lock(&cgroup_mutex); |
21a2d343 | 3683 | |
0adb0704 | 3684 | list_add_tail(&cfts->node, &ss->cfts); |
21a2d343 | 3685 | ret = cgroup_apply_cftypes(cfts, true); |
9ccece80 | 3686 | if (ret) |
21a2d343 | 3687 | cgroup_rm_cftypes_locked(cfts); |
79578621 | 3688 | |
01f6474c | 3689 | mutex_unlock(&cgroup_mutex); |
9ccece80 | 3690 | return ret; |
79578621 TH |
3691 | } |
3692 | ||
a8ddc821 TH |
3693 | /** |
3694 | * cgroup_add_dfl_cftypes - add an array of cftypes for default hierarchy | |
3695 | * @ss: target cgroup subsystem | |
3696 | * @cfts: zero-length name terminated array of cftypes | |
3697 | * | |
3698 | * Similar to cgroup_add_cftypes() but the added files are only used for | |
3699 | * the default hierarchy. | |
3700 | */ | |
3701 | int cgroup_add_dfl_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) | |
3702 | { | |
3703 | struct cftype *cft; | |
3704 | ||
3705 | for (cft = cfts; cft && cft->name[0] != '\0'; cft++) | |
05ebb6e6 | 3706 | cft->flags |= __CFTYPE_ONLY_ON_DFL; |
a8ddc821 TH |
3707 | return cgroup_add_cftypes(ss, cfts); |
3708 | } | |
3709 | ||
3710 | /** | |
3711 | * cgroup_add_legacy_cftypes - add an array of cftypes for legacy hierarchies | |
3712 | * @ss: target cgroup subsystem | |
3713 | * @cfts: zero-length name terminated array of cftypes | |
3714 | * | |
3715 | * Similar to cgroup_add_cftypes() but the added files are only used for | |
3716 | * the legacy hierarchies. | |
3717 | */ | |
2cf669a5 TH |
3718 | int cgroup_add_legacy_cftypes(struct cgroup_subsys *ss, struct cftype *cfts) |
3719 | { | |
a8ddc821 TH |
3720 | struct cftype *cft; |
3721 | ||
e4b7037c TH |
3722 | for (cft = cfts; cft && cft->name[0] != '\0'; cft++) |
3723 | cft->flags |= __CFTYPE_NOT_ON_DFL; | |
2cf669a5 TH |
3724 | return cgroup_add_cftypes(ss, cfts); |
3725 | } | |
3726 | ||
34c06254 TH |
3727 | /** |
3728 | * cgroup_file_notify - generate a file modified event for a cgroup_file | |
3729 | * @cfile: target cgroup_file | |
3730 | * | |
3731 | * @cfile must have been obtained by setting cftype->file_offset. | |
3732 | */ | |
3733 | void cgroup_file_notify(struct cgroup_file *cfile) | |
3734 | { | |
3735 | unsigned long flags; | |
3736 | ||
3737 | spin_lock_irqsave(&cgroup_file_kn_lock, flags); | |
3738 | if (cfile->kn) | |
3739 | kernfs_notify(cfile->kn); | |
3740 | spin_unlock_irqrestore(&cgroup_file_kn_lock, flags); | |
3741 | } | |
3742 | ||
a043e3b2 LZ |
3743 | /** |
3744 | * cgroup_task_count - count the number of tasks in a cgroup. | |
3745 | * @cgrp: the cgroup in question | |
3746 | * | |
3747 | * Return the number of tasks in the cgroup. | |
3748 | */ | |
07bc356e | 3749 | static int cgroup_task_count(const struct cgroup *cgrp) |
bbcb81d0 PM |
3750 | { |
3751 | int count = 0; | |
69d0206c | 3752 | struct cgrp_cset_link *link; |
817929ec | 3753 | |
f0d9a5f1 | 3754 | spin_lock_bh(&css_set_lock); |
69d0206c TH |
3755 | list_for_each_entry(link, &cgrp->cset_links, cset_link) |
3756 | count += atomic_read(&link->cset->refcount); | |
f0d9a5f1 | 3757 | spin_unlock_bh(&css_set_lock); |
bbcb81d0 PM |
3758 | return count; |
3759 | } | |
3760 | ||
53fa5261 | 3761 | /** |
492eb21b | 3762 | * css_next_child - find the next child of a given css |
c2931b70 TH |
3763 | * @pos: the current position (%NULL to initiate traversal) |
3764 | * @parent: css whose children to walk | |
53fa5261 | 3765 | * |
c2931b70 | 3766 | * This function returns the next child of @parent and should be called |
87fb54f1 | 3767 | * under either cgroup_mutex or RCU read lock. The only requirement is |
c2931b70 TH |
3768 | * that @parent and @pos are accessible. The next sibling is guaranteed to |
3769 | * be returned regardless of their states. | |
3770 | * | |
3771 | * If a subsystem synchronizes ->css_online() and the start of iteration, a | |
3772 | * css which finished ->css_online() is guaranteed to be visible in the | |
3773 | * future iterations and will stay visible until the last reference is put. | |
3774 | * A css which hasn't finished ->css_online() or already finished | |
3775 | * ->css_offline() may show up during traversal. It's each subsystem's | |
3776 | * responsibility to synchronize against on/offlining. | |
53fa5261 | 3777 | */ |
c2931b70 TH |
3778 | struct cgroup_subsys_state *css_next_child(struct cgroup_subsys_state *pos, |
3779 | struct cgroup_subsys_state *parent) | |
53fa5261 | 3780 | { |
c2931b70 | 3781 | struct cgroup_subsys_state *next; |
53fa5261 | 3782 | |
8353da1f | 3783 | cgroup_assert_mutex_or_rcu_locked(); |
53fa5261 TH |
3784 | |
3785 | /* | |
de3f0341 TH |
3786 | * @pos could already have been unlinked from the sibling list. |
3787 | * Once a cgroup is removed, its ->sibling.next is no longer | |
3788 | * updated when its next sibling changes. CSS_RELEASED is set when | |
3789 | * @pos is taken off list, at which time its next pointer is valid, | |
3790 | * and, as releases are serialized, the one pointed to by the next | |
3791 | * pointer is guaranteed to not have started release yet. This | |
3792 | * implies that if we observe !CSS_RELEASED on @pos in this RCU | |
3793 | * critical section, the one pointed to by its next pointer is | |
3794 | * guaranteed to not have finished its RCU grace period even if we | |
3795 | * have dropped rcu_read_lock() inbetween iterations. | |
3b287a50 | 3796 | * |
de3f0341 TH |
3797 | * If @pos has CSS_RELEASED set, its next pointer can't be |
3798 | * dereferenced; however, as each css is given a monotonically | |
3799 | * increasing unique serial number and always appended to the | |
3800 | * sibling list, the next one can be found by walking the parent's | |
3801 | * children until the first css with higher serial number than | |
3802 | * @pos's. While this path can be slower, it happens iff iteration | |
3803 | * races against release and the race window is very small. | |
53fa5261 | 3804 | */ |
3b287a50 | 3805 | if (!pos) { |
c2931b70 TH |
3806 | next = list_entry_rcu(parent->children.next, struct cgroup_subsys_state, sibling); |
3807 | } else if (likely(!(pos->flags & CSS_RELEASED))) { | |
3808 | next = list_entry_rcu(pos->sibling.next, struct cgroup_subsys_state, sibling); | |
3b287a50 | 3809 | } else { |
c2931b70 | 3810 | list_for_each_entry_rcu(next, &parent->children, sibling) |
3b287a50 TH |
3811 | if (next->serial_nr > pos->serial_nr) |
3812 | break; | |
53fa5261 TH |
3813 | } |
3814 | ||
3b281afb TH |
3815 | /* |
3816 | * @next, if not pointing to the head, can be dereferenced and is | |
c2931b70 | 3817 | * the next sibling. |
3b281afb | 3818 | */ |
c2931b70 TH |
3819 | if (&next->sibling != &parent->children) |
3820 | return next; | |
3b281afb | 3821 | return NULL; |
53fa5261 | 3822 | } |
53fa5261 | 3823 | |
574bd9f7 | 3824 | /** |
492eb21b | 3825 | * css_next_descendant_pre - find the next descendant for pre-order walk |
574bd9f7 | 3826 | * @pos: the current position (%NULL to initiate traversal) |
492eb21b | 3827 | * @root: css whose descendants to walk |
574bd9f7 | 3828 | * |
492eb21b | 3829 | * To be used by css_for_each_descendant_pre(). Find the next descendant |
bd8815a6 TH |
3830 | * to visit for pre-order traversal of @root's descendants. @root is |
3831 | * included in the iteration and the first node to be visited. | |
75501a6d | 3832 | * |
87fb54f1 TH |
3833 | * While this function requires cgroup_mutex or RCU read locking, it |
3834 | * doesn't require the whole traversal to be contained in a single critical | |
3835 | * section. This function will return the correct next descendant as long | |
3836 | * as both @pos and @root are accessible and @pos is a descendant of @root. | |
c2931b70 TH |
3837 | * |
3838 | * If a subsystem synchronizes ->css_online() and the start of iteration, a | |
3839 | * css which finished ->css_online() is guaranteed to be visible in the | |
3840 | * future iterations and will stay visible until the last reference is put. | |
3841 | * A css which hasn't finished ->css_online() or already finished | |
3842 | * ->css_offline() may show up during traversal. It's each subsystem's | |
3843 | * responsibility to synchronize against on/offlining. | |
574bd9f7 | 3844 | */ |
492eb21b TH |
3845 | struct cgroup_subsys_state * |
3846 | css_next_descendant_pre(struct cgroup_subsys_state *pos, | |
3847 | struct cgroup_subsys_state *root) | |
574bd9f7 | 3848 | { |
492eb21b | 3849 | struct cgroup_subsys_state *next; |
574bd9f7 | 3850 | |
8353da1f | 3851 | cgroup_assert_mutex_or_rcu_locked(); |
574bd9f7 | 3852 | |
bd8815a6 | 3853 | /* if first iteration, visit @root */ |
7805d000 | 3854 | if (!pos) |
bd8815a6 | 3855 | return root; |
574bd9f7 TH |
3856 | |
3857 | /* visit the first child if exists */ | |
492eb21b | 3858 | next = css_next_child(NULL, pos); |
574bd9f7 TH |
3859 | if (next) |
3860 | return next; | |
3861 | ||
3862 | /* no child, visit my or the closest ancestor's next sibling */ | |
492eb21b | 3863 | while (pos != root) { |
5c9d535b | 3864 | next = css_next_child(pos, pos->parent); |
75501a6d | 3865 | if (next) |
574bd9f7 | 3866 | return next; |
5c9d535b | 3867 | pos = pos->parent; |
7805d000 | 3868 | } |
574bd9f7 TH |
3869 | |
3870 | return NULL; | |
3871 | } | |
574bd9f7 | 3872 | |
12a9d2fe | 3873 | /** |
492eb21b TH |
3874 | * css_rightmost_descendant - return the rightmost descendant of a css |
3875 | * @pos: css of interest | |
12a9d2fe | 3876 | * |
492eb21b TH |
3877 | * Return the rightmost descendant of @pos. If there's no descendant, @pos |
3878 | * is returned. This can be used during pre-order traversal to skip | |
12a9d2fe | 3879 | * subtree of @pos. |
75501a6d | 3880 | * |
87fb54f1 TH |
3881 | * While this function requires cgroup_mutex or RCU read locking, it |
3882 | * doesn't require the whole traversal to be contained in a single critical | |
3883 | * section. This function will return the correct rightmost descendant as | |
3884 | * long as @pos is accessible. | |
12a9d2fe | 3885 | */ |
492eb21b TH |
3886 | struct cgroup_subsys_state * |
3887 | css_rightmost_descendant(struct cgroup_subsys_state *pos) | |
12a9d2fe | 3888 | { |
492eb21b | 3889 | struct cgroup_subsys_state *last, *tmp; |
12a9d2fe | 3890 | |
8353da1f | 3891 | cgroup_assert_mutex_or_rcu_locked(); |
12a9d2fe TH |
3892 | |
3893 | do { | |
3894 | last = pos; | |
3895 | /* ->prev isn't RCU safe, walk ->next till the end */ | |
3896 | pos = NULL; | |
492eb21b | 3897 | css_for_each_child(tmp, last) |
12a9d2fe TH |
3898 | pos = tmp; |
3899 | } while (pos); | |
3900 | ||
3901 | return last; | |
3902 | } | |
12a9d2fe | 3903 | |
492eb21b TH |
3904 | static struct cgroup_subsys_state * |
3905 | css_leftmost_descendant(struct cgroup_subsys_state *pos) | |
574bd9f7 | 3906 | { |
492eb21b | 3907 | struct cgroup_subsys_state *last; |
574bd9f7 TH |
3908 | |
3909 | do { | |
3910 | last = pos; | |
492eb21b | 3911 | pos = css_next_child(NULL, pos); |
574bd9f7 TH |
3912 | } while (pos); |
3913 | ||
3914 | return last; | |
3915 | } | |
3916 | ||
3917 | /** | |
492eb21b | 3918 | * css_next_descendant_post - find the next descendant for post-order walk |
574bd9f7 | 3919 | * @pos: the current position (%NULL to initiate traversal) |
492eb21b | 3920 | * @root: css whose descendants to walk |
574bd9f7 | 3921 | * |
492eb21b | 3922 | * To be used by css_for_each_descendant_post(). Find the next descendant |
bd8815a6 TH |
3923 | * to visit for post-order traversal of @root's descendants. @root is |
3924 | * included in the iteration and the last node to be visited. | |
75501a6d | 3925 | * |
87fb54f1 TH |
3926 | * While this function requires cgroup_mutex or RCU read locking, it |
3927 | * doesn't require the whole traversal to be contained in a single critical | |
3928 | * section. This function will return the correct next descendant as long | |
3929 | * as both @pos and @cgroup are accessible and @pos is a descendant of | |
3930 | * @cgroup. | |
c2931b70 TH |
3931 | * |
3932 | * If a subsystem synchronizes ->css_online() and the start of iteration, a | |
3933 | * css which finished ->css_online() is guaranteed to be visible in the | |
3934 | * future iterations and will stay visible until the last reference is put. | |
3935 | * A css which hasn't finished ->css_online() or already finished | |
3936 | * ->css_offline() may show up during traversal. It's each subsystem's | |
3937 | * responsibility to synchronize against on/offlining. | |
574bd9f7 | 3938 | */ |
492eb21b TH |
3939 | struct cgroup_subsys_state * |
3940 | css_next_descendant_post(struct cgroup_subsys_state *pos, | |
3941 | struct cgroup_subsys_state *root) | |
574bd9f7 | 3942 | { |
492eb21b | 3943 | struct cgroup_subsys_state *next; |
574bd9f7 | 3944 | |
8353da1f | 3945 | cgroup_assert_mutex_or_rcu_locked(); |
574bd9f7 | 3946 | |
58b79a91 TH |
3947 | /* if first iteration, visit leftmost descendant which may be @root */ |
3948 | if (!pos) | |
3949 | return css_leftmost_descendant(root); | |
574bd9f7 | 3950 | |
bd8815a6 TH |
3951 | /* if we visited @root, we're done */ |
3952 | if (pos == root) | |
3953 | return NULL; | |
3954 | ||
574bd9f7 | 3955 | /* if there's an unvisited sibling, visit its leftmost descendant */ |
5c9d535b | 3956 | next = css_next_child(pos, pos->parent); |
75501a6d | 3957 | if (next) |
492eb21b | 3958 | return css_leftmost_descendant(next); |
574bd9f7 TH |
3959 | |
3960 | /* no sibling left, visit parent */ | |
5c9d535b | 3961 | return pos->parent; |
574bd9f7 | 3962 | } |
574bd9f7 | 3963 | |
f3d46500 TH |
3964 | /** |
3965 | * css_has_online_children - does a css have online children | |
3966 | * @css: the target css | |
3967 | * | |
3968 | * Returns %true if @css has any online children; otherwise, %false. This | |
3969 | * function can be called from any context but the caller is responsible | |
3970 | * for synchronizing against on/offlining as necessary. | |
3971 | */ | |
3972 | bool css_has_online_children(struct cgroup_subsys_state *css) | |
cbc125ef | 3973 | { |
f3d46500 TH |
3974 | struct cgroup_subsys_state *child; |
3975 | bool ret = false; | |
cbc125ef TH |
3976 | |
3977 | rcu_read_lock(); | |
f3d46500 | 3978 | css_for_each_child(child, css) { |
99bae5f9 | 3979 | if (child->flags & CSS_ONLINE) { |
f3d46500 TH |
3980 | ret = true; |
3981 | break; | |
cbc125ef TH |
3982 | } |
3983 | } | |
3984 | rcu_read_unlock(); | |
f3d46500 | 3985 | return ret; |
574bd9f7 | 3986 | } |
574bd9f7 | 3987 | |
0942eeee | 3988 | /** |
ecb9d535 | 3989 | * css_task_iter_advance_css_set - advance a task itererator to the next css_set |
0942eeee TH |
3990 | * @it: the iterator to advance |
3991 | * | |
3992 | * Advance @it to the next css_set to walk. | |
d515876e | 3993 | */ |
ecb9d535 | 3994 | static void css_task_iter_advance_css_set(struct css_task_iter *it) |
d515876e | 3995 | { |
0f0a2b4f | 3996 | struct list_head *l = it->cset_pos; |
d515876e TH |
3997 | struct cgrp_cset_link *link; |
3998 | struct css_set *cset; | |
3999 | ||
f0d9a5f1 | 4000 | lockdep_assert_held(&css_set_lock); |
ed27b9f7 | 4001 | |
d515876e TH |
4002 | /* Advance to the next non-empty css_set */ |
4003 | do { | |
4004 | l = l->next; | |
0f0a2b4f TH |
4005 | if (l == it->cset_head) { |
4006 | it->cset_pos = NULL; | |
ecb9d535 | 4007 | it->task_pos = NULL; |
d515876e TH |
4008 | return; |
4009 | } | |
3ebb2b6e TH |
4010 | |
4011 | if (it->ss) { | |
4012 | cset = container_of(l, struct css_set, | |
4013 | e_cset_node[it->ss->id]); | |
4014 | } else { | |
4015 | link = list_entry(l, struct cgrp_cset_link, cset_link); | |
4016 | cset = link->cset; | |
4017 | } | |
0de0942d | 4018 | } while (!css_set_populated(cset)); |
c7561128 | 4019 | |
0f0a2b4f | 4020 | it->cset_pos = l; |
c7561128 TH |
4021 | |
4022 | if (!list_empty(&cset->tasks)) | |
0f0a2b4f | 4023 | it->task_pos = cset->tasks.next; |
c7561128 | 4024 | else |
0f0a2b4f TH |
4025 | it->task_pos = cset->mg_tasks.next; |
4026 | ||
4027 | it->tasks_head = &cset->tasks; | |
4028 | it->mg_tasks_head = &cset->mg_tasks; | |
ed27b9f7 TH |
4029 | |
4030 | /* | |
4031 | * We don't keep css_sets locked across iteration steps and thus | |
4032 | * need to take steps to ensure that iteration can be resumed after | |
4033 | * the lock is re-acquired. Iteration is performed at two levels - | |
4034 | * css_sets and tasks in them. | |
4035 | * | |
4036 | * Once created, a css_set never leaves its cgroup lists, so a | |
4037 | * pinned css_set is guaranteed to stay put and we can resume | |
4038 | * iteration afterwards. | |
4039 | * | |
4040 | * Tasks may leave @cset across iteration steps. This is resolved | |
4041 | * by registering each iterator with the css_set currently being | |
4042 | * walked and making css_set_move_task() advance iterators whose | |
4043 | * next task is leaving. | |
4044 | */ | |
4045 | if (it->cur_cset) { | |
4046 | list_del(&it->iters_node); | |
4047 | put_css_set_locked(it->cur_cset); | |
4048 | } | |
4049 | get_css_set(cset); | |
4050 | it->cur_cset = cset; | |
4051 | list_add(&it->iters_node, &cset->task_iters); | |
d515876e TH |
4052 | } |
4053 | ||
ecb9d535 TH |
4054 | static void css_task_iter_advance(struct css_task_iter *it) |
4055 | { | |
4056 | struct list_head *l = it->task_pos; | |
4057 | ||
f0d9a5f1 | 4058 | lockdep_assert_held(&css_set_lock); |
ecb9d535 TH |
4059 | WARN_ON_ONCE(!l); |
4060 | ||
4061 | /* | |
4062 | * Advance iterator to find next entry. cset->tasks is consumed | |
4063 | * first and then ->mg_tasks. After ->mg_tasks, we move onto the | |
4064 | * next cset. | |
4065 | */ | |
4066 | l = l->next; | |
4067 | ||
4068 | if (l == it->tasks_head) | |
4069 | l = it->mg_tasks_head->next; | |
4070 | ||
4071 | if (l == it->mg_tasks_head) | |
4072 | css_task_iter_advance_css_set(it); | |
4073 | else | |
4074 | it->task_pos = l; | |
4075 | } | |
4076 | ||
0942eeee | 4077 | /** |
72ec7029 TH |
4078 | * css_task_iter_start - initiate task iteration |
4079 | * @css: the css to walk tasks of | |
0942eeee TH |
4080 | * @it: the task iterator to use |
4081 | * | |
72ec7029 TH |
4082 | * Initiate iteration through the tasks of @css. The caller can call |
4083 | * css_task_iter_next() to walk through the tasks until the function | |
4084 | * returns NULL. On completion of iteration, css_task_iter_end() must be | |
4085 | * called. | |
0942eeee | 4086 | */ |
72ec7029 TH |
4087 | void css_task_iter_start(struct cgroup_subsys_state *css, |
4088 | struct css_task_iter *it) | |
817929ec | 4089 | { |
56fde9e0 TH |
4090 | /* no one should try to iterate before mounting cgroups */ |
4091 | WARN_ON_ONCE(!use_task_css_set_links); | |
31a7df01 | 4092 | |
ed27b9f7 TH |
4093 | memset(it, 0, sizeof(*it)); |
4094 | ||
f0d9a5f1 | 4095 | spin_lock_bh(&css_set_lock); |
c59cd3d8 | 4096 | |
3ebb2b6e TH |
4097 | it->ss = css->ss; |
4098 | ||
4099 | if (it->ss) | |
4100 | it->cset_pos = &css->cgroup->e_csets[css->ss->id]; | |
4101 | else | |
4102 | it->cset_pos = &css->cgroup->cset_links; | |
4103 | ||
0f0a2b4f | 4104 | it->cset_head = it->cset_pos; |
c59cd3d8 | 4105 | |
ecb9d535 | 4106 | css_task_iter_advance_css_set(it); |
ed27b9f7 | 4107 | |
f0d9a5f1 | 4108 | spin_unlock_bh(&css_set_lock); |
817929ec PM |
4109 | } |
4110 | ||
0942eeee | 4111 | /** |
72ec7029 | 4112 | * css_task_iter_next - return the next task for the iterator |
0942eeee TH |
4113 | * @it: the task iterator being iterated |
4114 | * | |
4115 | * The "next" function for task iteration. @it should have been | |
72ec7029 TH |
4116 | * initialized via css_task_iter_start(). Returns NULL when the iteration |
4117 | * reaches the end. | |
0942eeee | 4118 | */ |
72ec7029 | 4119 | struct task_struct *css_task_iter_next(struct css_task_iter *it) |
817929ec | 4120 | { |
d5745675 | 4121 | if (it->cur_task) { |
ed27b9f7 | 4122 | put_task_struct(it->cur_task); |
d5745675 TH |
4123 | it->cur_task = NULL; |
4124 | } | |
ed27b9f7 | 4125 | |
f0d9a5f1 | 4126 | spin_lock_bh(&css_set_lock); |
ed27b9f7 | 4127 | |
d5745675 TH |
4128 | if (it->task_pos) { |
4129 | it->cur_task = list_entry(it->task_pos, struct task_struct, | |
4130 | cg_list); | |
4131 | get_task_struct(it->cur_task); | |
4132 | css_task_iter_advance(it); | |
4133 | } | |
ed27b9f7 | 4134 | |
f0d9a5f1 | 4135 | spin_unlock_bh(&css_set_lock); |
ed27b9f7 TH |
4136 | |
4137 | return it->cur_task; | |
817929ec PM |
4138 | } |
4139 | ||
0942eeee | 4140 | /** |
72ec7029 | 4141 | * css_task_iter_end - finish task iteration |
0942eeee TH |
4142 | * @it: the task iterator to finish |
4143 | * | |
72ec7029 | 4144 | * Finish task iteration started by css_task_iter_start(). |
0942eeee | 4145 | */ |
72ec7029 | 4146 | void css_task_iter_end(struct css_task_iter *it) |
31a7df01 | 4147 | { |
ed27b9f7 | 4148 | if (it->cur_cset) { |
f0d9a5f1 | 4149 | spin_lock_bh(&css_set_lock); |
ed27b9f7 TH |
4150 | list_del(&it->iters_node); |
4151 | put_css_set_locked(it->cur_cset); | |
f0d9a5f1 | 4152 | spin_unlock_bh(&css_set_lock); |
ed27b9f7 TH |
4153 | } |
4154 | ||
4155 | if (it->cur_task) | |
4156 | put_task_struct(it->cur_task); | |
31a7df01 CW |
4157 | } |
4158 | ||
4159 | /** | |
8cc99345 TH |
4160 | * cgroup_trasnsfer_tasks - move tasks from one cgroup to another |
4161 | * @to: cgroup to which the tasks will be moved | |
4162 | * @from: cgroup in which the tasks currently reside | |
31a7df01 | 4163 | * |
eaf797ab TH |
4164 | * Locking rules between cgroup_post_fork() and the migration path |
4165 | * guarantee that, if a task is forking while being migrated, the new child | |
4166 | * is guaranteed to be either visible in the source cgroup after the | |
4167 | * parent's migration is complete or put into the target cgroup. No task | |
4168 | * can slip out of migration through forking. | |
31a7df01 | 4169 | */ |
8cc99345 | 4170 | int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from) |
31a7df01 | 4171 | { |
952aaa12 TH |
4172 | LIST_HEAD(preloaded_csets); |
4173 | struct cgrp_cset_link *link; | |
72ec7029 | 4174 | struct css_task_iter it; |
e406d1cf | 4175 | struct task_struct *task; |
952aaa12 | 4176 | int ret; |
31a7df01 | 4177 | |
6c694c88 TH |
4178 | if (!cgroup_may_migrate_to(to)) |
4179 | return -EBUSY; | |
4180 | ||
952aaa12 | 4181 | mutex_lock(&cgroup_mutex); |
31a7df01 | 4182 | |
952aaa12 | 4183 | /* all tasks in @from are being moved, all csets are source */ |
f0d9a5f1 | 4184 | spin_lock_bh(&css_set_lock); |
952aaa12 TH |
4185 | list_for_each_entry(link, &from->cset_links, cset_link) |
4186 | cgroup_migrate_add_src(link->cset, to, &preloaded_csets); | |
f0d9a5f1 | 4187 | spin_unlock_bh(&css_set_lock); |
31a7df01 | 4188 | |
e4857982 | 4189 | ret = cgroup_migrate_prepare_dst(&preloaded_csets); |
952aaa12 TH |
4190 | if (ret) |
4191 | goto out_err; | |
8cc99345 | 4192 | |
952aaa12 | 4193 | /* |
2cfa2b19 | 4194 | * Migrate tasks one-by-one until @from is empty. This fails iff |
952aaa12 TH |
4195 | * ->can_attach() fails. |
4196 | */ | |
e406d1cf | 4197 | do { |
9d800df1 | 4198 | css_task_iter_start(&from->self, &it); |
e406d1cf TH |
4199 | task = css_task_iter_next(&it); |
4200 | if (task) | |
4201 | get_task_struct(task); | |
4202 | css_task_iter_end(&it); | |
4203 | ||
4204 | if (task) { | |
37ff9f8f | 4205 | ret = cgroup_migrate(task, false, to->root); |
e406d1cf TH |
4206 | put_task_struct(task); |
4207 | } | |
4208 | } while (task && !ret); | |
952aaa12 TH |
4209 | out_err: |
4210 | cgroup_migrate_finish(&preloaded_csets); | |
47cfcd09 | 4211 | mutex_unlock(&cgroup_mutex); |
e406d1cf | 4212 | return ret; |
8cc99345 TH |
4213 | } |
4214 | ||
bbcb81d0 | 4215 | /* |
102a775e | 4216 | * Stuff for reading the 'tasks'/'procs' files. |
bbcb81d0 PM |
4217 | * |
4218 | * Reading this file can return large amounts of data if a cgroup has | |
4219 | * *lots* of attached tasks. So it may need several calls to read(), | |
4220 | * but we cannot guarantee that the information we produce is correct | |
4221 | * unless we produce it entirely atomically. | |
4222 | * | |
bbcb81d0 | 4223 | */ |
bbcb81d0 | 4224 | |
24528255 LZ |
4225 | /* which pidlist file are we talking about? */ |
4226 | enum cgroup_filetype { | |
4227 | CGROUP_FILE_PROCS, | |
4228 | CGROUP_FILE_TASKS, | |
4229 | }; | |
4230 | ||
4231 | /* | |
4232 | * A pidlist is a list of pids that virtually represents the contents of one | |
4233 | * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists, | |
4234 | * a pair (one each for procs, tasks) for each pid namespace that's relevant | |
4235 | * to the cgroup. | |
4236 | */ | |
4237 | struct cgroup_pidlist { | |
4238 | /* | |
4239 | * used to find which pidlist is wanted. doesn't change as long as | |
4240 | * this particular list stays in the list. | |
4241 | */ | |
4242 | struct { enum cgroup_filetype type; struct pid_namespace *ns; } key; | |
4243 | /* array of xids */ | |
4244 | pid_t *list; | |
4245 | /* how many elements the above list has */ | |
4246 | int length; | |
24528255 LZ |
4247 | /* each of these stored in a list by its cgroup */ |
4248 | struct list_head links; | |
4249 | /* pointer to the cgroup we belong to, for list removal purposes */ | |
4250 | struct cgroup *owner; | |
b1a21367 TH |
4251 | /* for delayed destruction */ |
4252 | struct delayed_work destroy_dwork; | |
24528255 LZ |
4253 | }; |
4254 | ||
d1d9fd33 BB |
4255 | /* |
4256 | * The following two functions "fix" the issue where there are more pids | |
4257 | * than kmalloc will give memory for; in such cases, we use vmalloc/vfree. | |
4258 | * TODO: replace with a kernel-wide solution to this problem | |
4259 | */ | |
4260 | #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2)) | |
4261 | static void *pidlist_allocate(int count) | |
4262 | { | |
4263 | if (PIDLIST_TOO_LARGE(count)) | |
4264 | return vmalloc(count * sizeof(pid_t)); | |
4265 | else | |
4266 | return kmalloc(count * sizeof(pid_t), GFP_KERNEL); | |
4267 | } | |
b1a21367 | 4268 | |
d1d9fd33 BB |
4269 | static void pidlist_free(void *p) |
4270 | { | |
58794514 | 4271 | kvfree(p); |
d1d9fd33 | 4272 | } |
d1d9fd33 | 4273 | |
b1a21367 TH |
4274 | /* |
4275 | * Used to destroy all pidlists lingering waiting for destroy timer. None | |
4276 | * should be left afterwards. | |
4277 | */ | |
4278 | static void cgroup_pidlist_destroy_all(struct cgroup *cgrp) | |
4279 | { | |
4280 | struct cgroup_pidlist *l, *tmp_l; | |
4281 | ||
4282 | mutex_lock(&cgrp->pidlist_mutex); | |
4283 | list_for_each_entry_safe(l, tmp_l, &cgrp->pidlists, links) | |
4284 | mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, 0); | |
4285 | mutex_unlock(&cgrp->pidlist_mutex); | |
4286 | ||
4287 | flush_workqueue(cgroup_pidlist_destroy_wq); | |
4288 | BUG_ON(!list_empty(&cgrp->pidlists)); | |
4289 | } | |
4290 | ||
4291 | static void cgroup_pidlist_destroy_work_fn(struct work_struct *work) | |
4292 | { | |
4293 | struct delayed_work *dwork = to_delayed_work(work); | |
4294 | struct cgroup_pidlist *l = container_of(dwork, struct cgroup_pidlist, | |
4295 | destroy_dwork); | |
4296 | struct cgroup_pidlist *tofree = NULL; | |
4297 | ||
4298 | mutex_lock(&l->owner->pidlist_mutex); | |
b1a21367 TH |
4299 | |
4300 | /* | |
04502365 TH |
4301 | * Destroy iff we didn't get queued again. The state won't change |
4302 | * as destroy_dwork can only be queued while locked. | |
b1a21367 | 4303 | */ |
04502365 | 4304 | if (!delayed_work_pending(dwork)) { |
b1a21367 TH |
4305 | list_del(&l->links); |
4306 | pidlist_free(l->list); | |
4307 | put_pid_ns(l->key.ns); | |
4308 | tofree = l; | |
4309 | } | |
4310 | ||
b1a21367 TH |
4311 | mutex_unlock(&l->owner->pidlist_mutex); |
4312 | kfree(tofree); | |
4313 | } | |
4314 | ||
bbcb81d0 | 4315 | /* |
102a775e | 4316 | * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries |
6ee211ad | 4317 | * Returns the number of unique elements. |
bbcb81d0 | 4318 | */ |
6ee211ad | 4319 | static int pidlist_uniq(pid_t *list, int length) |
bbcb81d0 | 4320 | { |
102a775e | 4321 | int src, dest = 1; |
102a775e BB |
4322 | |
4323 | /* | |
4324 | * we presume the 0th element is unique, so i starts at 1. trivial | |
4325 | * edge cases first; no work needs to be done for either | |
4326 | */ | |
4327 | if (length == 0 || length == 1) | |
4328 | return length; | |
4329 | /* src and dest walk down the list; dest counts unique elements */ | |
4330 | for (src = 1; src < length; src++) { | |
4331 | /* find next unique element */ | |
4332 | while (list[src] == list[src-1]) { | |
4333 | src++; | |
4334 | if (src == length) | |
4335 | goto after; | |
4336 | } | |
4337 | /* dest always points to where the next unique element goes */ | |
4338 | list[dest] = list[src]; | |
4339 | dest++; | |
4340 | } | |
4341 | after: | |
102a775e BB |
4342 | return dest; |
4343 | } | |
4344 | ||
afb2bc14 TH |
4345 | /* |
4346 | * The two pid files - task and cgroup.procs - guaranteed that the result | |
4347 | * is sorted, which forced this whole pidlist fiasco. As pid order is | |
4348 | * different per namespace, each namespace needs differently sorted list, | |
4349 | * making it impossible to use, for example, single rbtree of member tasks | |
4350 | * sorted by task pointer. As pidlists can be fairly large, allocating one | |
4351 | * per open file is dangerous, so cgroup had to implement shared pool of | |
4352 | * pidlists keyed by cgroup and namespace. | |
4353 | * | |
4354 | * All this extra complexity was caused by the original implementation | |
4355 | * committing to an entirely unnecessary property. In the long term, we | |
aa6ec29b TH |
4356 | * want to do away with it. Explicitly scramble sort order if on the |
4357 | * default hierarchy so that no such expectation exists in the new | |
4358 | * interface. | |
afb2bc14 TH |
4359 | * |
4360 | * Scrambling is done by swapping every two consecutive bits, which is | |
4361 | * non-identity one-to-one mapping which disturbs sort order sufficiently. | |
4362 | */ | |
4363 | static pid_t pid_fry(pid_t pid) | |
4364 | { | |
4365 | unsigned a = pid & 0x55555555; | |
4366 | unsigned b = pid & 0xAAAAAAAA; | |
4367 | ||
4368 | return (a << 1) | (b >> 1); | |
4369 | } | |
4370 | ||
4371 | static pid_t cgroup_pid_fry(struct cgroup *cgrp, pid_t pid) | |
4372 | { | |
aa6ec29b | 4373 | if (cgroup_on_dfl(cgrp)) |
afb2bc14 TH |
4374 | return pid_fry(pid); |
4375 | else | |
4376 | return pid; | |
4377 | } | |
4378 | ||
102a775e BB |
4379 | static int cmppid(const void *a, const void *b) |
4380 | { | |
4381 | return *(pid_t *)a - *(pid_t *)b; | |
4382 | } | |
4383 | ||
afb2bc14 TH |
4384 | static int fried_cmppid(const void *a, const void *b) |
4385 | { | |
4386 | return pid_fry(*(pid_t *)a) - pid_fry(*(pid_t *)b); | |
4387 | } | |
4388 | ||
e6b81710 TH |
4389 | static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp, |
4390 | enum cgroup_filetype type) | |
4391 | { | |
4392 | struct cgroup_pidlist *l; | |
4393 | /* don't need task_nsproxy() if we're looking at ourself */ | |
4394 | struct pid_namespace *ns = task_active_pid_ns(current); | |
4395 | ||
4396 | lockdep_assert_held(&cgrp->pidlist_mutex); | |
4397 | ||
4398 | list_for_each_entry(l, &cgrp->pidlists, links) | |
4399 | if (l->key.type == type && l->key.ns == ns) | |
4400 | return l; | |
4401 | return NULL; | |
4402 | } | |
4403 | ||
72a8cb30 BB |
4404 | /* |
4405 | * find the appropriate pidlist for our purpose (given procs vs tasks) | |
4406 | * returns with the lock on that pidlist already held, and takes care | |
4407 | * of the use count, or returns NULL with no locks held if we're out of | |
4408 | * memory. | |
4409 | */ | |
e6b81710 TH |
4410 | static struct cgroup_pidlist *cgroup_pidlist_find_create(struct cgroup *cgrp, |
4411 | enum cgroup_filetype type) | |
72a8cb30 BB |
4412 | { |
4413 | struct cgroup_pidlist *l; | |
b70cc5fd | 4414 | |
e6b81710 TH |
4415 | lockdep_assert_held(&cgrp->pidlist_mutex); |
4416 | ||
4417 | l = cgroup_pidlist_find(cgrp, type); | |
4418 | if (l) | |
4419 | return l; | |
4420 | ||
72a8cb30 | 4421 | /* entry not found; create a new one */ |
f4f4be2b | 4422 | l = kzalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL); |
e6b81710 | 4423 | if (!l) |
72a8cb30 | 4424 | return l; |
e6b81710 | 4425 | |
b1a21367 | 4426 | INIT_DELAYED_WORK(&l->destroy_dwork, cgroup_pidlist_destroy_work_fn); |
72a8cb30 | 4427 | l->key.type = type; |
e6b81710 TH |
4428 | /* don't need task_nsproxy() if we're looking at ourself */ |
4429 | l->key.ns = get_pid_ns(task_active_pid_ns(current)); | |
72a8cb30 BB |
4430 | l->owner = cgrp; |
4431 | list_add(&l->links, &cgrp->pidlists); | |
72a8cb30 BB |
4432 | return l; |
4433 | } | |
4434 | ||
102a775e BB |
4435 | /* |
4436 | * Load a cgroup's pidarray with either procs' tgids or tasks' pids | |
4437 | */ | |
72a8cb30 BB |
4438 | static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type, |
4439 | struct cgroup_pidlist **lp) | |
102a775e BB |
4440 | { |
4441 | pid_t *array; | |
4442 | int length; | |
4443 | int pid, n = 0; /* used for populating the array */ | |
72ec7029 | 4444 | struct css_task_iter it; |
817929ec | 4445 | struct task_struct *tsk; |
102a775e BB |
4446 | struct cgroup_pidlist *l; |
4447 | ||
4bac00d1 TH |
4448 | lockdep_assert_held(&cgrp->pidlist_mutex); |
4449 | ||
102a775e BB |
4450 | /* |
4451 | * If cgroup gets more users after we read count, we won't have | |
4452 | * enough space - tough. This race is indistinguishable to the | |
4453 | * caller from the case that the additional cgroup users didn't | |
4454 | * show up until sometime later on. | |
4455 | */ | |
4456 | length = cgroup_task_count(cgrp); | |
d1d9fd33 | 4457 | array = pidlist_allocate(length); |
102a775e BB |
4458 | if (!array) |
4459 | return -ENOMEM; | |
4460 | /* now, populate the array */ | |
9d800df1 | 4461 | css_task_iter_start(&cgrp->self, &it); |
72ec7029 | 4462 | while ((tsk = css_task_iter_next(&it))) { |
102a775e | 4463 | if (unlikely(n == length)) |
817929ec | 4464 | break; |
102a775e | 4465 | /* get tgid or pid for procs or tasks file respectively */ |
72a8cb30 BB |
4466 | if (type == CGROUP_FILE_PROCS) |
4467 | pid = task_tgid_vnr(tsk); | |
4468 | else | |
4469 | pid = task_pid_vnr(tsk); | |
102a775e BB |
4470 | if (pid > 0) /* make sure to only use valid results */ |
4471 | array[n++] = pid; | |
817929ec | 4472 | } |
72ec7029 | 4473 | css_task_iter_end(&it); |
102a775e BB |
4474 | length = n; |
4475 | /* now sort & (if procs) strip out duplicates */ | |
aa6ec29b | 4476 | if (cgroup_on_dfl(cgrp)) |
afb2bc14 TH |
4477 | sort(array, length, sizeof(pid_t), fried_cmppid, NULL); |
4478 | else | |
4479 | sort(array, length, sizeof(pid_t), cmppid, NULL); | |
72a8cb30 | 4480 | if (type == CGROUP_FILE_PROCS) |
6ee211ad | 4481 | length = pidlist_uniq(array, length); |
e6b81710 | 4482 | |
e6b81710 | 4483 | l = cgroup_pidlist_find_create(cgrp, type); |
72a8cb30 | 4484 | if (!l) { |
d1d9fd33 | 4485 | pidlist_free(array); |
72a8cb30 | 4486 | return -ENOMEM; |
102a775e | 4487 | } |
e6b81710 TH |
4488 | |
4489 | /* store array, freeing old if necessary */ | |
d1d9fd33 | 4490 | pidlist_free(l->list); |
102a775e BB |
4491 | l->list = array; |
4492 | l->length = length; | |
72a8cb30 | 4493 | *lp = l; |
102a775e | 4494 | return 0; |
bbcb81d0 PM |
4495 | } |
4496 | ||
846c7bb0 | 4497 | /** |
a043e3b2 | 4498 | * cgroupstats_build - build and fill cgroupstats |
846c7bb0 BS |
4499 | * @stats: cgroupstats to fill information into |
4500 | * @dentry: A dentry entry belonging to the cgroup for which stats have | |
4501 | * been requested. | |
a043e3b2 LZ |
4502 | * |
4503 | * Build and fill cgroupstats so that taskstats can export it to user | |
4504 | * space. | |
846c7bb0 BS |
4505 | */ |
4506 | int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry) | |
4507 | { | |
2bd59d48 | 4508 | struct kernfs_node *kn = kernfs_node_from_dentry(dentry); |
bd89aabc | 4509 | struct cgroup *cgrp; |
72ec7029 | 4510 | struct css_task_iter it; |
846c7bb0 | 4511 | struct task_struct *tsk; |
33d283be | 4512 | |
2bd59d48 TH |
4513 | /* it should be kernfs_node belonging to cgroupfs and is a directory */ |
4514 | if (dentry->d_sb->s_type != &cgroup_fs_type || !kn || | |
4515 | kernfs_type(kn) != KERNFS_DIR) | |
4516 | return -EINVAL; | |
4517 | ||
bad34660 LZ |
4518 | mutex_lock(&cgroup_mutex); |
4519 | ||
846c7bb0 | 4520 | /* |
2bd59d48 | 4521 | * We aren't being called from kernfs and there's no guarantee on |
ec903c0c | 4522 | * @kn->priv's validity. For this and css_tryget_online_from_dir(), |
2bd59d48 | 4523 | * @kn->priv is RCU safe. Let's do the RCU dancing. |
846c7bb0 | 4524 | */ |
2bd59d48 TH |
4525 | rcu_read_lock(); |
4526 | cgrp = rcu_dereference(kn->priv); | |
bad34660 | 4527 | if (!cgrp || cgroup_is_dead(cgrp)) { |
2bd59d48 | 4528 | rcu_read_unlock(); |
bad34660 | 4529 | mutex_unlock(&cgroup_mutex); |
2bd59d48 TH |
4530 | return -ENOENT; |
4531 | } | |
bad34660 | 4532 | rcu_read_unlock(); |
846c7bb0 | 4533 | |
9d800df1 | 4534 | css_task_iter_start(&cgrp->self, &it); |
72ec7029 | 4535 | while ((tsk = css_task_iter_next(&it))) { |
846c7bb0 BS |
4536 | switch (tsk->state) { |
4537 | case TASK_RUNNING: | |
4538 | stats->nr_running++; | |
4539 | break; | |
4540 | case TASK_INTERRUPTIBLE: | |
4541 | stats->nr_sleeping++; | |
4542 | break; | |
4543 | case TASK_UNINTERRUPTIBLE: | |
4544 | stats->nr_uninterruptible++; | |
4545 | break; | |
4546 | case TASK_STOPPED: | |
4547 | stats->nr_stopped++; | |
4548 | break; | |
4549 | default: | |
4550 | if (delayacct_is_task_waiting_on_io(tsk)) | |
4551 | stats->nr_io_wait++; | |
4552 | break; | |
4553 | } | |
4554 | } | |
72ec7029 | 4555 | css_task_iter_end(&it); |
846c7bb0 | 4556 | |
bad34660 | 4557 | mutex_unlock(&cgroup_mutex); |
2bd59d48 | 4558 | return 0; |
846c7bb0 BS |
4559 | } |
4560 | ||
8f3ff208 | 4561 | |
bbcb81d0 | 4562 | /* |
102a775e | 4563 | * seq_file methods for the tasks/procs files. The seq_file position is the |
cc31edce | 4564 | * next pid to display; the seq_file iterator is a pointer to the pid |
102a775e | 4565 | * in the cgroup->l->list array. |
bbcb81d0 | 4566 | */ |
cc31edce | 4567 | |
102a775e | 4568 | static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos) |
bbcb81d0 | 4569 | { |
cc31edce PM |
4570 | /* |
4571 | * Initially we receive a position value that corresponds to | |
4572 | * one more than the last pid shown (or 0 on the first call or | |
4573 | * after a seek to the start). Use a binary-search to find the | |
4574 | * next pid to display, if any | |
4575 | */ | |
2bd59d48 | 4576 | struct kernfs_open_file *of = s->private; |
7da11279 | 4577 | struct cgroup *cgrp = seq_css(s)->cgroup; |
4bac00d1 | 4578 | struct cgroup_pidlist *l; |
7da11279 | 4579 | enum cgroup_filetype type = seq_cft(s)->private; |
cc31edce | 4580 | int index = 0, pid = *pos; |
4bac00d1 TH |
4581 | int *iter, ret; |
4582 | ||
4583 | mutex_lock(&cgrp->pidlist_mutex); | |
4584 | ||
4585 | /* | |
5d22444f | 4586 | * !NULL @of->priv indicates that this isn't the first start() |
4bac00d1 | 4587 | * after open. If the matching pidlist is around, we can use that. |
5d22444f | 4588 | * Look for it. Note that @of->priv can't be used directly. It |
4bac00d1 TH |
4589 | * could already have been destroyed. |
4590 | */ | |
5d22444f TH |
4591 | if (of->priv) |
4592 | of->priv = cgroup_pidlist_find(cgrp, type); | |
4bac00d1 TH |
4593 | |
4594 | /* | |
4595 | * Either this is the first start() after open or the matching | |
4596 | * pidlist has been destroyed inbetween. Create a new one. | |
4597 | */ | |
5d22444f TH |
4598 | if (!of->priv) { |
4599 | ret = pidlist_array_load(cgrp, type, | |
4600 | (struct cgroup_pidlist **)&of->priv); | |
4bac00d1 TH |
4601 | if (ret) |
4602 | return ERR_PTR(ret); | |
4603 | } | |
5d22444f | 4604 | l = of->priv; |
cc31edce | 4605 | |
cc31edce | 4606 | if (pid) { |
102a775e | 4607 | int end = l->length; |
20777766 | 4608 | |
cc31edce PM |
4609 | while (index < end) { |
4610 | int mid = (index + end) / 2; | |
afb2bc14 | 4611 | if (cgroup_pid_fry(cgrp, l->list[mid]) == pid) { |
cc31edce PM |
4612 | index = mid; |
4613 | break; | |
afb2bc14 | 4614 | } else if (cgroup_pid_fry(cgrp, l->list[mid]) <= pid) |
cc31edce PM |
4615 | index = mid + 1; |
4616 | else | |
4617 | end = mid; | |
4618 | } | |
4619 | } | |
4620 | /* If we're off the end of the array, we're done */ | |
102a775e | 4621 | if (index >= l->length) |
cc31edce PM |
4622 | return NULL; |
4623 | /* Update the abstract position to be the actual pid that we found */ | |
102a775e | 4624 | iter = l->list + index; |
afb2bc14 | 4625 | *pos = cgroup_pid_fry(cgrp, *iter); |
cc31edce PM |
4626 | return iter; |
4627 | } | |
4628 | ||
102a775e | 4629 | static void cgroup_pidlist_stop(struct seq_file *s, void *v) |
cc31edce | 4630 | { |
2bd59d48 | 4631 | struct kernfs_open_file *of = s->private; |
5d22444f | 4632 | struct cgroup_pidlist *l = of->priv; |
62236858 | 4633 | |
5d22444f TH |
4634 | if (l) |
4635 | mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, | |
04502365 | 4636 | CGROUP_PIDLIST_DESTROY_DELAY); |
7da11279 | 4637 | mutex_unlock(&seq_css(s)->cgroup->pidlist_mutex); |
cc31edce PM |
4638 | } |
4639 | ||
102a775e | 4640 | static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos) |
cc31edce | 4641 | { |
2bd59d48 | 4642 | struct kernfs_open_file *of = s->private; |
5d22444f | 4643 | struct cgroup_pidlist *l = of->priv; |
102a775e BB |
4644 | pid_t *p = v; |
4645 | pid_t *end = l->list + l->length; | |
cc31edce PM |
4646 | /* |
4647 | * Advance to the next pid in the array. If this goes off the | |
4648 | * end, we're done | |
4649 | */ | |
4650 | p++; | |
4651 | if (p >= end) { | |
4652 | return NULL; | |
4653 | } else { | |
7da11279 | 4654 | *pos = cgroup_pid_fry(seq_css(s)->cgroup, *p); |
cc31edce PM |
4655 | return p; |
4656 | } | |
4657 | } | |
4658 | ||
102a775e | 4659 | static int cgroup_pidlist_show(struct seq_file *s, void *v) |
cc31edce | 4660 | { |
94ff212d JP |
4661 | seq_printf(s, "%d\n", *(int *)v); |
4662 | ||
4663 | return 0; | |
cc31edce | 4664 | } |
bbcb81d0 | 4665 | |
182446d0 TH |
4666 | static u64 cgroup_read_notify_on_release(struct cgroup_subsys_state *css, |
4667 | struct cftype *cft) | |
81a6a5cd | 4668 | { |
182446d0 | 4669 | return notify_on_release(css->cgroup); |
81a6a5cd PM |
4670 | } |
4671 | ||
182446d0 TH |
4672 | static int cgroup_write_notify_on_release(struct cgroup_subsys_state *css, |
4673 | struct cftype *cft, u64 val) | |
6379c106 | 4674 | { |
6379c106 | 4675 | if (val) |
182446d0 | 4676 | set_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags); |
6379c106 | 4677 | else |
182446d0 | 4678 | clear_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags); |
6379c106 PM |
4679 | return 0; |
4680 | } | |
4681 | ||
182446d0 TH |
4682 | static u64 cgroup_clone_children_read(struct cgroup_subsys_state *css, |
4683 | struct cftype *cft) | |
97978e6d | 4684 | { |
182446d0 | 4685 | return test_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d DL |
4686 | } |
4687 | ||
182446d0 TH |
4688 | static int cgroup_clone_children_write(struct cgroup_subsys_state *css, |
4689 | struct cftype *cft, u64 val) | |
97978e6d DL |
4690 | { |
4691 | if (val) | |
182446d0 | 4692 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d | 4693 | else |
182446d0 | 4694 | clear_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags); |
97978e6d DL |
4695 | return 0; |
4696 | } | |
4697 | ||
a14c6874 TH |
4698 | /* cgroup core interface files for the default hierarchy */ |
4699 | static struct cftype cgroup_dfl_base_files[] = { | |
81a6a5cd | 4700 | { |
d5c56ced | 4701 | .name = "cgroup.procs", |
6f60eade | 4702 | .file_offset = offsetof(struct cgroup, procs_file), |
6612f05b TH |
4703 | .seq_start = cgroup_pidlist_start, |
4704 | .seq_next = cgroup_pidlist_next, | |
4705 | .seq_stop = cgroup_pidlist_stop, | |
4706 | .seq_show = cgroup_pidlist_show, | |
5d22444f | 4707 | .private = CGROUP_FILE_PROCS, |
acbef755 | 4708 | .write = cgroup_procs_write, |
102a775e | 4709 | }, |
f8f22e53 TH |
4710 | { |
4711 | .name = "cgroup.controllers", | |
f8f22e53 TH |
4712 | .seq_show = cgroup_controllers_show, |
4713 | }, | |
4714 | { | |
4715 | .name = "cgroup.subtree_control", | |
f8f22e53 | 4716 | .seq_show = cgroup_subtree_control_show, |
451af504 | 4717 | .write = cgroup_subtree_control_write, |
f8f22e53 | 4718 | }, |
842b597e | 4719 | { |
4a07c222 | 4720 | .name = "cgroup.events", |
a14c6874 | 4721 | .flags = CFTYPE_NOT_ON_ROOT, |
6f60eade | 4722 | .file_offset = offsetof(struct cgroup, events_file), |
4a07c222 | 4723 | .seq_show = cgroup_events_show, |
842b597e | 4724 | }, |
a14c6874 TH |
4725 | { } /* terminate */ |
4726 | }; | |
d5c56ced | 4727 | |
a14c6874 TH |
4728 | /* cgroup core interface files for the legacy hierarchies */ |
4729 | static struct cftype cgroup_legacy_base_files[] = { | |
4730 | { | |
4731 | .name = "cgroup.procs", | |
4732 | .seq_start = cgroup_pidlist_start, | |
4733 | .seq_next = cgroup_pidlist_next, | |
4734 | .seq_stop = cgroup_pidlist_stop, | |
4735 | .seq_show = cgroup_pidlist_show, | |
4736 | .private = CGROUP_FILE_PROCS, | |
4737 | .write = cgroup_procs_write, | |
a14c6874 TH |
4738 | }, |
4739 | { | |
4740 | .name = "cgroup.clone_children", | |
4741 | .read_u64 = cgroup_clone_children_read, | |
4742 | .write_u64 = cgroup_clone_children_write, | |
4743 | }, | |
4744 | { | |
4745 | .name = "cgroup.sane_behavior", | |
4746 | .flags = CFTYPE_ONLY_ON_ROOT, | |
4747 | .seq_show = cgroup_sane_behavior_show, | |
4748 | }, | |
d5c56ced TH |
4749 | { |
4750 | .name = "tasks", | |
6612f05b TH |
4751 | .seq_start = cgroup_pidlist_start, |
4752 | .seq_next = cgroup_pidlist_next, | |
4753 | .seq_stop = cgroup_pidlist_stop, | |
4754 | .seq_show = cgroup_pidlist_show, | |
5d22444f | 4755 | .private = CGROUP_FILE_TASKS, |
acbef755 | 4756 | .write = cgroup_tasks_write, |
d5c56ced TH |
4757 | }, |
4758 | { | |
4759 | .name = "notify_on_release", | |
d5c56ced TH |
4760 | .read_u64 = cgroup_read_notify_on_release, |
4761 | .write_u64 = cgroup_write_notify_on_release, | |
4762 | }, | |
6e6ff25b TH |
4763 | { |
4764 | .name = "release_agent", | |
a14c6874 | 4765 | .flags = CFTYPE_ONLY_ON_ROOT, |
2da8ca82 | 4766 | .seq_show = cgroup_release_agent_show, |
451af504 | 4767 | .write = cgroup_release_agent_write, |
5f469907 | 4768 | .max_write_len = PATH_MAX - 1, |
6e6ff25b | 4769 | }, |
db0416b6 | 4770 | { } /* terminate */ |
bbcb81d0 PM |
4771 | }; |
4772 | ||
0c21ead1 TH |
4773 | /* |
4774 | * css destruction is four-stage process. | |
4775 | * | |
4776 | * 1. Destruction starts. Killing of the percpu_ref is initiated. | |
4777 | * Implemented in kill_css(). | |
4778 | * | |
4779 | * 2. When the percpu_ref is confirmed to be visible as killed on all CPUs | |
ec903c0c TH |
4780 | * and thus css_tryget_online() is guaranteed to fail, the css can be |
4781 | * offlined by invoking offline_css(). After offlining, the base ref is | |
4782 | * put. Implemented in css_killed_work_fn(). | |
0c21ead1 TH |
4783 | * |
4784 | * 3. When the percpu_ref reaches zero, the only possible remaining | |
4785 | * accessors are inside RCU read sections. css_release() schedules the | |
4786 | * RCU callback. | |
4787 | * | |
4788 | * 4. After the grace period, the css can be freed. Implemented in | |
4789 | * css_free_work_fn(). | |
4790 | * | |
4791 | * It is actually hairier because both step 2 and 4 require process context | |
4792 | * and thus involve punting to css->destroy_work adding two additional | |
4793 | * steps to the already complex sequence. | |
4794 | */ | |
35ef10da | 4795 | static void css_free_work_fn(struct work_struct *work) |
48ddbe19 TH |
4796 | { |
4797 | struct cgroup_subsys_state *css = | |
35ef10da | 4798 | container_of(work, struct cgroup_subsys_state, destroy_work); |
01e58659 | 4799 | struct cgroup_subsys *ss = css->ss; |
0c21ead1 | 4800 | struct cgroup *cgrp = css->cgroup; |
48ddbe19 | 4801 | |
9a1049da TH |
4802 | percpu_ref_exit(&css->refcnt); |
4803 | ||
01e58659 | 4804 | if (ss) { |
9d755d33 | 4805 | /* css free path */ |
8bb5ef79 | 4806 | struct cgroup_subsys_state *parent = css->parent; |
01e58659 VD |
4807 | int id = css->id; |
4808 | ||
01e58659 VD |
4809 | ss->css_free(css); |
4810 | cgroup_idr_remove(&ss->css_idr, id); | |
9d755d33 | 4811 | cgroup_put(cgrp); |
8bb5ef79 TH |
4812 | |
4813 | if (parent) | |
4814 | css_put(parent); | |
9d755d33 TH |
4815 | } else { |
4816 | /* cgroup free path */ | |
4817 | atomic_dec(&cgrp->root->nr_cgrps); | |
4818 | cgroup_pidlist_destroy_all(cgrp); | |
971ff493 | 4819 | cancel_work_sync(&cgrp->release_agent_work); |
9d755d33 | 4820 | |
d51f39b0 | 4821 | if (cgroup_parent(cgrp)) { |
9d755d33 TH |
4822 | /* |
4823 | * We get a ref to the parent, and put the ref when | |
4824 | * this cgroup is being freed, so it's guaranteed | |
4825 | * that the parent won't be destroyed before its | |
4826 | * children. | |
4827 | */ | |
d51f39b0 | 4828 | cgroup_put(cgroup_parent(cgrp)); |
9d755d33 TH |
4829 | kernfs_put(cgrp->kn); |
4830 | kfree(cgrp); | |
4831 | } else { | |
4832 | /* | |
4833 | * This is root cgroup's refcnt reaching zero, | |
4834 | * which indicates that the root should be | |
4835 | * released. | |
4836 | */ | |
4837 | cgroup_destroy_root(cgrp->root); | |
4838 | } | |
4839 | } | |
48ddbe19 TH |
4840 | } |
4841 | ||
0c21ead1 | 4842 | static void css_free_rcu_fn(struct rcu_head *rcu_head) |
d3daf28d TH |
4843 | { |
4844 | struct cgroup_subsys_state *css = | |
0c21ead1 | 4845 | container_of(rcu_head, struct cgroup_subsys_state, rcu_head); |
d3daf28d | 4846 | |
35ef10da | 4847 | INIT_WORK(&css->destroy_work, css_free_work_fn); |
e5fca243 | 4848 | queue_work(cgroup_destroy_wq, &css->destroy_work); |
48ddbe19 TH |
4849 | } |
4850 | ||
25e15d83 | 4851 | static void css_release_work_fn(struct work_struct *work) |
d3daf28d TH |
4852 | { |
4853 | struct cgroup_subsys_state *css = | |
25e15d83 | 4854 | container_of(work, struct cgroup_subsys_state, destroy_work); |
15a4c835 | 4855 | struct cgroup_subsys *ss = css->ss; |
9d755d33 | 4856 | struct cgroup *cgrp = css->cgroup; |
15a4c835 | 4857 | |
1fed1b2e TH |
4858 | mutex_lock(&cgroup_mutex); |
4859 | ||
de3f0341 | 4860 | css->flags |= CSS_RELEASED; |
1fed1b2e TH |
4861 | list_del_rcu(&css->sibling); |
4862 | ||
9d755d33 TH |
4863 | if (ss) { |
4864 | /* css release path */ | |
01e58659 | 4865 | cgroup_idr_replace(&ss->css_idr, NULL, css->id); |
7d172cc8 TH |
4866 | if (ss->css_released) |
4867 | ss->css_released(css); | |
9d755d33 TH |
4868 | } else { |
4869 | /* cgroup release path */ | |
9d755d33 TH |
4870 | cgroup_idr_remove(&cgrp->root->cgroup_idr, cgrp->id); |
4871 | cgrp->id = -1; | |
a4189487 LZ |
4872 | |
4873 | /* | |
4874 | * There are two control paths which try to determine | |
4875 | * cgroup from dentry without going through kernfs - | |
4876 | * cgroupstats_build() and css_tryget_online_from_dir(). | |
4877 | * Those are supported by RCU protecting clearing of | |
4878 | * cgrp->kn->priv backpointer. | |
4879 | */ | |
6cd0f5bb TH |
4880 | if (cgrp->kn) |
4881 | RCU_INIT_POINTER(*(void __rcu __force **)&cgrp->kn->priv, | |
4882 | NULL); | |
9d755d33 | 4883 | } |
d3daf28d | 4884 | |
1fed1b2e TH |
4885 | mutex_unlock(&cgroup_mutex); |
4886 | ||
0c21ead1 | 4887 | call_rcu(&css->rcu_head, css_free_rcu_fn); |
d3daf28d TH |
4888 | } |
4889 | ||
d3daf28d TH |
4890 | static void css_release(struct percpu_ref *ref) |
4891 | { | |
4892 | struct cgroup_subsys_state *css = | |
4893 | container_of(ref, struct cgroup_subsys_state, refcnt); | |
4894 | ||
25e15d83 TH |
4895 | INIT_WORK(&css->destroy_work, css_release_work_fn); |
4896 | queue_work(cgroup_destroy_wq, &css->destroy_work); | |
d3daf28d TH |
4897 | } |
4898 | ||
ddfcadab TH |
4899 | static void init_and_link_css(struct cgroup_subsys_state *css, |
4900 | struct cgroup_subsys *ss, struct cgroup *cgrp) | |
ddbcc7e8 | 4901 | { |
0cb51d71 TH |
4902 | lockdep_assert_held(&cgroup_mutex); |
4903 | ||
ddfcadab TH |
4904 | cgroup_get(cgrp); |
4905 | ||
d5c419b6 | 4906 | memset(css, 0, sizeof(*css)); |
bd89aabc | 4907 | css->cgroup = cgrp; |
72c97e54 | 4908 | css->ss = ss; |
d5c419b6 TH |
4909 | INIT_LIST_HEAD(&css->sibling); |
4910 | INIT_LIST_HEAD(&css->children); | |
0cb51d71 | 4911 | css->serial_nr = css_serial_nr_next++; |
aa226ff4 | 4912 | atomic_set(&css->online_cnt, 0); |
0ae78e0b | 4913 | |
d51f39b0 TH |
4914 | if (cgroup_parent(cgrp)) { |
4915 | css->parent = cgroup_css(cgroup_parent(cgrp), ss); | |
ddfcadab | 4916 | css_get(css->parent); |
ddfcadab | 4917 | } |
48ddbe19 | 4918 | |
ca8bdcaf | 4919 | BUG_ON(cgroup_css(cgrp, ss)); |
ddbcc7e8 PM |
4920 | } |
4921 | ||
2a4ac633 | 4922 | /* invoke ->css_online() on a new CSS and mark it online if successful */ |
623f926b | 4923 | static int online_css(struct cgroup_subsys_state *css) |
a31f2d3f | 4924 | { |
623f926b | 4925 | struct cgroup_subsys *ss = css->ss; |
b1929db4 TH |
4926 | int ret = 0; |
4927 | ||
a31f2d3f TH |
4928 | lockdep_assert_held(&cgroup_mutex); |
4929 | ||
92fb9748 | 4930 | if (ss->css_online) |
eb95419b | 4931 | ret = ss->css_online(css); |
ae7f164a | 4932 | if (!ret) { |
eb95419b | 4933 | css->flags |= CSS_ONLINE; |
aec25020 | 4934 | rcu_assign_pointer(css->cgroup->subsys[ss->id], css); |
aa226ff4 TH |
4935 | |
4936 | atomic_inc(&css->online_cnt); | |
4937 | if (css->parent) | |
4938 | atomic_inc(&css->parent->online_cnt); | |
ae7f164a | 4939 | } |
b1929db4 | 4940 | return ret; |
a31f2d3f TH |
4941 | } |
4942 | ||
2a4ac633 | 4943 | /* if the CSS is online, invoke ->css_offline() on it and mark it offline */ |
623f926b | 4944 | static void offline_css(struct cgroup_subsys_state *css) |
a31f2d3f | 4945 | { |
623f926b | 4946 | struct cgroup_subsys *ss = css->ss; |
a31f2d3f TH |
4947 | |
4948 | lockdep_assert_held(&cgroup_mutex); | |
4949 | ||
4950 | if (!(css->flags & CSS_ONLINE)) | |
4951 | return; | |
4952 | ||
fa06235b VD |
4953 | if (ss->css_reset) |
4954 | ss->css_reset(css); | |
4955 | ||
d7eeac19 | 4956 | if (ss->css_offline) |
eb95419b | 4957 | ss->css_offline(css); |
a31f2d3f | 4958 | |
eb95419b | 4959 | css->flags &= ~CSS_ONLINE; |
e3297803 | 4960 | RCU_INIT_POINTER(css->cgroup->subsys[ss->id], NULL); |
f8f22e53 TH |
4961 | |
4962 | wake_up_all(&css->cgroup->offline_waitq); | |
a31f2d3f TH |
4963 | } |
4964 | ||
c81c925a | 4965 | /** |
6cd0f5bb | 4966 | * css_create - create a cgroup_subsys_state |
c81c925a TH |
4967 | * @cgrp: the cgroup new css will be associated with |
4968 | * @ss: the subsys of new css | |
4969 | * | |
4970 | * Create a new css associated with @cgrp - @ss pair. On success, the new | |
6cd0f5bb TH |
4971 | * css is online and installed in @cgrp. This function doesn't create the |
4972 | * interface files. Returns 0 on success, -errno on failure. | |
c81c925a | 4973 | */ |
6cd0f5bb TH |
4974 | static struct cgroup_subsys_state *css_create(struct cgroup *cgrp, |
4975 | struct cgroup_subsys *ss) | |
c81c925a | 4976 | { |
d51f39b0 | 4977 | struct cgroup *parent = cgroup_parent(cgrp); |
1fed1b2e | 4978 | struct cgroup_subsys_state *parent_css = cgroup_css(parent, ss); |
c81c925a TH |
4979 | struct cgroup_subsys_state *css; |
4980 | int err; | |
4981 | ||
c81c925a TH |
4982 | lockdep_assert_held(&cgroup_mutex); |
4983 | ||
1fed1b2e | 4984 | css = ss->css_alloc(parent_css); |
c81c925a | 4985 | if (IS_ERR(css)) |
6cd0f5bb | 4986 | return css; |
c81c925a | 4987 | |
ddfcadab | 4988 | init_and_link_css(css, ss, cgrp); |
a2bed820 | 4989 | |
2aad2a86 | 4990 | err = percpu_ref_init(&css->refcnt, css_release, 0, GFP_KERNEL); |
c81c925a | 4991 | if (err) |
3eb59ec6 | 4992 | goto err_free_css; |
c81c925a | 4993 | |
cf780b7d | 4994 | err = cgroup_idr_alloc(&ss->css_idr, NULL, 2, 0, GFP_KERNEL); |
15a4c835 TH |
4995 | if (err < 0) |
4996 | goto err_free_percpu_ref; | |
4997 | css->id = err; | |
c81c925a | 4998 | |
15a4c835 | 4999 | /* @css is ready to be brought online now, make it visible */ |
1fed1b2e | 5000 | list_add_tail_rcu(&css->sibling, &parent_css->children); |
15a4c835 | 5001 | cgroup_idr_replace(&ss->css_idr, css, css->id); |
c81c925a TH |
5002 | |
5003 | err = online_css(css); | |
5004 | if (err) | |
1fed1b2e | 5005 | goto err_list_del; |
94419627 | 5006 | |
c81c925a | 5007 | if (ss->broken_hierarchy && !ss->warned_broken_hierarchy && |
d51f39b0 | 5008 | cgroup_parent(parent)) { |
ed3d261b | 5009 | 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 | 5010 | current->comm, current->pid, ss->name); |
c81c925a | 5011 | if (!strcmp(ss->name, "memory")) |
ed3d261b | 5012 | pr_warn("\"memory\" requires setting use_hierarchy to 1 on the root\n"); |
c81c925a TH |
5013 | ss->warned_broken_hierarchy = true; |
5014 | } | |
5015 | ||
6cd0f5bb | 5016 | return css; |
c81c925a | 5017 | |
1fed1b2e TH |
5018 | err_list_del: |
5019 | list_del_rcu(&css->sibling); | |
15a4c835 | 5020 | cgroup_idr_remove(&ss->css_idr, css->id); |
3eb59ec6 | 5021 | err_free_percpu_ref: |
9a1049da | 5022 | percpu_ref_exit(&css->refcnt); |
3eb59ec6 | 5023 | err_free_css: |
a2bed820 | 5024 | call_rcu(&css->rcu_head, css_free_rcu_fn); |
6cd0f5bb | 5025 | return ERR_PTR(err); |
c81c925a TH |
5026 | } |
5027 | ||
a5bca215 | 5028 | static struct cgroup *cgroup_create(struct cgroup *parent) |
ddbcc7e8 | 5029 | { |
a5bca215 | 5030 | struct cgroup_root *root = parent->root; |
a5bca215 TH |
5031 | struct cgroup *cgrp, *tcgrp; |
5032 | int level = parent->level + 1; | |
03970d3c | 5033 | int ret; |
ddbcc7e8 | 5034 | |
0a950f65 | 5035 | /* allocate the cgroup and its ID, 0 is reserved for the root */ |
b11cfb58 TH |
5036 | cgrp = kzalloc(sizeof(*cgrp) + |
5037 | sizeof(cgrp->ancestor_ids[0]) * (level + 1), GFP_KERNEL); | |
a5bca215 TH |
5038 | if (!cgrp) |
5039 | return ERR_PTR(-ENOMEM); | |
0ab02ca8 | 5040 | |
2aad2a86 | 5041 | ret = percpu_ref_init(&cgrp->self.refcnt, css_release, 0, GFP_KERNEL); |
9d755d33 TH |
5042 | if (ret) |
5043 | goto out_free_cgrp; | |
5044 | ||
0ab02ca8 LZ |
5045 | /* |
5046 | * Temporarily set the pointer to NULL, so idr_find() won't return | |
5047 | * a half-baked cgroup. | |
5048 | */ | |
cf780b7d | 5049 | cgrp->id = cgroup_idr_alloc(&root->cgroup_idr, NULL, 2, 0, GFP_KERNEL); |
0ab02ca8 | 5050 | if (cgrp->id < 0) { |
ba0f4d76 | 5051 | ret = -ENOMEM; |
9d755d33 | 5052 | goto out_cancel_ref; |
976c06bc TH |
5053 | } |
5054 | ||
cc31edce | 5055 | init_cgroup_housekeeping(cgrp); |
ddbcc7e8 | 5056 | |
9d800df1 | 5057 | cgrp->self.parent = &parent->self; |
ba0f4d76 | 5058 | cgrp->root = root; |
b11cfb58 TH |
5059 | cgrp->level = level; |
5060 | ||
5061 | for (tcgrp = cgrp; tcgrp; tcgrp = cgroup_parent(tcgrp)) | |
5062 | cgrp->ancestor_ids[tcgrp->level] = tcgrp->id; | |
ddbcc7e8 | 5063 | |
b6abdb0e LZ |
5064 | if (notify_on_release(parent)) |
5065 | set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags); | |
5066 | ||
2260e7fc TH |
5067 | if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &parent->flags)) |
5068 | set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags); | |
97978e6d | 5069 | |
0cb51d71 | 5070 | cgrp->self.serial_nr = css_serial_nr_next++; |
53fa5261 | 5071 | |
4e139afc | 5072 | /* allocation complete, commit to creation */ |
d5c419b6 | 5073 | list_add_tail_rcu(&cgrp->self.sibling, &cgroup_parent(cgrp)->self.children); |
3c9c825b | 5074 | atomic_inc(&root->nr_cgrps); |
59f5296b | 5075 | cgroup_get(parent); |
415cf07a | 5076 | |
0d80255e TH |
5077 | /* |
5078 | * @cgrp is now fully operational. If something fails after this | |
5079 | * point, it'll be released via the normal destruction path. | |
5080 | */ | |
6fa4918d | 5081 | cgroup_idr_replace(&root->cgroup_idr, cgrp, cgrp->id); |
4e96ee8e | 5082 | |
bd53d617 TH |
5083 | /* |
5084 | * On the default hierarchy, a child doesn't automatically inherit | |
667c2491 | 5085 | * subtree_control from the parent. Each is configured manually. |
bd53d617 | 5086 | */ |
03970d3c | 5087 | if (!cgroup_on_dfl(cgrp)) |
5531dc91 | 5088 | cgrp->subtree_control = cgroup_control(cgrp); |
03970d3c TH |
5089 | |
5090 | cgroup_propagate_control(cgrp); | |
5091 | ||
5092 | /* @cgrp doesn't have dir yet so the following will only create csses */ | |
5093 | ret = cgroup_apply_control_enable(cgrp); | |
5094 | if (ret) | |
5095 | goto out_destroy; | |
2bd59d48 | 5096 | |
a5bca215 TH |
5097 | return cgrp; |
5098 | ||
5099 | out_cancel_ref: | |
5100 | percpu_ref_exit(&cgrp->self.refcnt); | |
5101 | out_free_cgrp: | |
5102 | kfree(cgrp); | |
5103 | return ERR_PTR(ret); | |
5104 | out_destroy: | |
5105 | cgroup_destroy_locked(cgrp); | |
5106 | return ERR_PTR(ret); | |
5107 | } | |
5108 | ||
5109 | static int cgroup_mkdir(struct kernfs_node *parent_kn, const char *name, | |
5110 | umode_t mode) | |
5111 | { | |
5112 | struct cgroup *parent, *cgrp; | |
a5bca215 | 5113 | struct kernfs_node *kn; |
03970d3c | 5114 | int ret; |
a5bca215 TH |
5115 | |
5116 | /* do not accept '\n' to prevent making /proc/<pid>/cgroup unparsable */ | |
5117 | if (strchr(name, '\n')) | |
5118 | return -EINVAL; | |
5119 | ||
945ba199 | 5120 | parent = cgroup_kn_lock_live(parent_kn, false); |
a5bca215 TH |
5121 | if (!parent) |
5122 | return -ENODEV; | |
5123 | ||
5124 | cgrp = cgroup_create(parent); | |
5125 | if (IS_ERR(cgrp)) { | |
5126 | ret = PTR_ERR(cgrp); | |
5127 | goto out_unlock; | |
5128 | } | |
5129 | ||
195e9b6c TH |
5130 | /* create the directory */ |
5131 | kn = kernfs_create_dir(parent->kn, name, mode, cgrp); | |
5132 | if (IS_ERR(kn)) { | |
5133 | ret = PTR_ERR(kn); | |
5134 | goto out_destroy; | |
5135 | } | |
5136 | cgrp->kn = kn; | |
5137 | ||
5138 | /* | |
5139 | * This extra ref will be put in cgroup_free_fn() and guarantees | |
5140 | * that @cgrp->kn is always accessible. | |
5141 | */ | |
5142 | kernfs_get(kn); | |
5143 | ||
5144 | ret = cgroup_kn_set_ugid(kn); | |
5145 | if (ret) | |
5146 | goto out_destroy; | |
5147 | ||
334c3679 | 5148 | ret = css_populate_dir(&cgrp->self); |
195e9b6c TH |
5149 | if (ret) |
5150 | goto out_destroy; | |
5151 | ||
03970d3c TH |
5152 | ret = cgroup_apply_control_enable(cgrp); |
5153 | if (ret) | |
5154 | goto out_destroy; | |
195e9b6c TH |
5155 | |
5156 | /* let's create and online css's */ | |
2bd59d48 | 5157 | kernfs_activate(kn); |
ddbcc7e8 | 5158 | |
ba0f4d76 TH |
5159 | ret = 0; |
5160 | goto out_unlock; | |
ddbcc7e8 | 5161 | |
a5bca215 TH |
5162 | out_destroy: |
5163 | cgroup_destroy_locked(cgrp); | |
ba0f4d76 | 5164 | out_unlock: |
a9746d8d | 5165 | cgroup_kn_unlock(parent_kn); |
ba0f4d76 | 5166 | return ret; |
ddbcc7e8 PM |
5167 | } |
5168 | ||
223dbc38 TH |
5169 | /* |
5170 | * This is called when the refcnt of a css is confirmed to be killed. | |
249f3468 TH |
5171 | * css_tryget_online() is now guaranteed to fail. Tell the subsystem to |
5172 | * initate destruction and put the css ref from kill_css(). | |
223dbc38 TH |
5173 | */ |
5174 | static void css_killed_work_fn(struct work_struct *work) | |
d3daf28d | 5175 | { |
223dbc38 TH |
5176 | struct cgroup_subsys_state *css = |
5177 | container_of(work, struct cgroup_subsys_state, destroy_work); | |
d3daf28d | 5178 | |
f20104de | 5179 | mutex_lock(&cgroup_mutex); |
09a503ea | 5180 | |
aa226ff4 TH |
5181 | do { |
5182 | offline_css(css); | |
5183 | css_put(css); | |
5184 | /* @css can't go away while we're holding cgroup_mutex */ | |
5185 | css = css->parent; | |
5186 | } while (css && atomic_dec_and_test(&css->online_cnt)); | |
5187 | ||
5188 | mutex_unlock(&cgroup_mutex); | |
d3daf28d TH |
5189 | } |
5190 | ||
223dbc38 TH |
5191 | /* css kill confirmation processing requires process context, bounce */ |
5192 | static void css_killed_ref_fn(struct percpu_ref *ref) | |
d3daf28d TH |
5193 | { |
5194 | struct cgroup_subsys_state *css = | |
5195 | container_of(ref, struct cgroup_subsys_state, refcnt); | |
5196 | ||
aa226ff4 TH |
5197 | if (atomic_dec_and_test(&css->online_cnt)) { |
5198 | INIT_WORK(&css->destroy_work, css_killed_work_fn); | |
5199 | queue_work(cgroup_destroy_wq, &css->destroy_work); | |
5200 | } | |
d3daf28d TH |
5201 | } |
5202 | ||
f392e51c TH |
5203 | /** |
5204 | * kill_css - destroy a css | |
5205 | * @css: css to destroy | |
5206 | * | |
5207 | * This function initiates destruction of @css by removing cgroup interface | |
5208 | * files and putting its base reference. ->css_offline() will be invoked | |
ec903c0c TH |
5209 | * asynchronously once css_tryget_online() is guaranteed to fail and when |
5210 | * the reference count reaches zero, @css will be released. | |
f392e51c TH |
5211 | */ |
5212 | static void kill_css(struct cgroup_subsys_state *css) | |
edae0c33 | 5213 | { |
01f6474c | 5214 | lockdep_assert_held(&cgroup_mutex); |
94419627 | 5215 | |
2bd59d48 TH |
5216 | /* |
5217 | * This must happen before css is disassociated with its cgroup. | |
5218 | * See seq_css() for details. | |
5219 | */ | |
334c3679 | 5220 | css_clear_dir(css); |
3c14f8b4 | 5221 | |
edae0c33 TH |
5222 | /* |
5223 | * Killing would put the base ref, but we need to keep it alive | |
5224 | * until after ->css_offline(). | |
5225 | */ | |
5226 | css_get(css); | |
5227 | ||
5228 | /* | |
5229 | * cgroup core guarantees that, by the time ->css_offline() is | |
5230 | * invoked, no new css reference will be given out via | |
ec903c0c | 5231 | * css_tryget_online(). We can't simply call percpu_ref_kill() and |
edae0c33 TH |
5232 | * proceed to offlining css's because percpu_ref_kill() doesn't |
5233 | * guarantee that the ref is seen as killed on all CPUs on return. | |
5234 | * | |
5235 | * Use percpu_ref_kill_and_confirm() to get notifications as each | |
5236 | * css is confirmed to be seen as killed on all CPUs. | |
5237 | */ | |
5238 | percpu_ref_kill_and_confirm(&css->refcnt, css_killed_ref_fn); | |
d3daf28d TH |
5239 | } |
5240 | ||
5241 | /** | |
5242 | * cgroup_destroy_locked - the first stage of cgroup destruction | |
5243 | * @cgrp: cgroup to be destroyed | |
5244 | * | |
5245 | * css's make use of percpu refcnts whose killing latency shouldn't be | |
5246 | * exposed to userland and are RCU protected. Also, cgroup core needs to | |
ec903c0c TH |
5247 | * guarantee that css_tryget_online() won't succeed by the time |
5248 | * ->css_offline() is invoked. To satisfy all the requirements, | |
5249 | * destruction is implemented in the following two steps. | |
d3daf28d TH |
5250 | * |
5251 | * s1. Verify @cgrp can be destroyed and mark it dying. Remove all | |
5252 | * userland visible parts and start killing the percpu refcnts of | |
5253 | * css's. Set up so that the next stage will be kicked off once all | |
5254 | * the percpu refcnts are confirmed to be killed. | |
5255 | * | |
5256 | * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the | |
5257 | * rest of destruction. Once all cgroup references are gone, the | |
5258 | * cgroup is RCU-freed. | |
5259 | * | |
5260 | * This function implements s1. After this step, @cgrp is gone as far as | |
5261 | * the userland is concerned and a new cgroup with the same name may be | |
5262 | * created. As cgroup doesn't care about the names internally, this | |
5263 | * doesn't cause any problem. | |
5264 | */ | |
42809dd4 TH |
5265 | static int cgroup_destroy_locked(struct cgroup *cgrp) |
5266 | __releases(&cgroup_mutex) __acquires(&cgroup_mutex) | |
ddbcc7e8 | 5267 | { |
2bd59d48 | 5268 | struct cgroup_subsys_state *css; |
2b021cbf | 5269 | struct cgrp_cset_link *link; |
1c6727af | 5270 | int ssid; |
ddbcc7e8 | 5271 | |
42809dd4 TH |
5272 | lockdep_assert_held(&cgroup_mutex); |
5273 | ||
91486f61 TH |
5274 | /* |
5275 | * Only migration can raise populated from zero and we're already | |
5276 | * holding cgroup_mutex. | |
5277 | */ | |
5278 | if (cgroup_is_populated(cgrp)) | |
ddbcc7e8 | 5279 | return -EBUSY; |
a043e3b2 | 5280 | |
bb78a92f | 5281 | /* |
d5c419b6 TH |
5282 | * Make sure there's no live children. We can't test emptiness of |
5283 | * ->self.children as dead children linger on it while being | |
5284 | * drained; otherwise, "rmdir parent/child parent" may fail. | |
bb78a92f | 5285 | */ |
f3d46500 | 5286 | if (css_has_online_children(&cgrp->self)) |
bb78a92f HD |
5287 | return -EBUSY; |
5288 | ||
455050d2 | 5289 | /* |
2b021cbf TH |
5290 | * Mark @cgrp and the associated csets dead. The former prevents |
5291 | * further task migration and child creation by disabling | |
5292 | * cgroup_lock_live_group(). The latter makes the csets ignored by | |
5293 | * the migration path. | |
455050d2 | 5294 | */ |
184faf32 | 5295 | cgrp->self.flags &= ~CSS_ONLINE; |
ddbcc7e8 | 5296 | |
2b021cbf TH |
5297 | spin_lock_bh(&css_set_lock); |
5298 | list_for_each_entry(link, &cgrp->cset_links, cset_link) | |
5299 | link->cset->dead = true; | |
5300 | spin_unlock_bh(&css_set_lock); | |
5301 | ||
249f3468 | 5302 | /* initiate massacre of all css's */ |
1c6727af TH |
5303 | for_each_css(css, ssid, cgrp) |
5304 | kill_css(css); | |
455050d2 | 5305 | |
455050d2 | 5306 | /* |
01f6474c TH |
5307 | * Remove @cgrp directory along with the base files. @cgrp has an |
5308 | * extra ref on its kn. | |
f20104de | 5309 | */ |
01f6474c | 5310 | kernfs_remove(cgrp->kn); |
f20104de | 5311 | |
d51f39b0 | 5312 | check_for_release(cgroup_parent(cgrp)); |
2bd59d48 | 5313 | |
249f3468 | 5314 | /* put the base reference */ |
9d755d33 | 5315 | percpu_ref_kill(&cgrp->self.refcnt); |
455050d2 | 5316 | |
ea15f8cc TH |
5317 | return 0; |
5318 | }; | |
5319 | ||
2bd59d48 | 5320 | static int cgroup_rmdir(struct kernfs_node *kn) |
42809dd4 | 5321 | { |
a9746d8d | 5322 | struct cgroup *cgrp; |
2bd59d48 | 5323 | int ret = 0; |
42809dd4 | 5324 | |
945ba199 | 5325 | cgrp = cgroup_kn_lock_live(kn, false); |
a9746d8d TH |
5326 | if (!cgrp) |
5327 | return 0; | |
42809dd4 | 5328 | |
a9746d8d | 5329 | ret = cgroup_destroy_locked(cgrp); |
2bb566cb | 5330 | |
a9746d8d | 5331 | cgroup_kn_unlock(kn); |
42809dd4 | 5332 | return ret; |
8e3f6541 TH |
5333 | } |
5334 | ||
2bd59d48 TH |
5335 | static struct kernfs_syscall_ops cgroup_kf_syscall_ops = { |
5336 | .remount_fs = cgroup_remount, | |
5337 | .show_options = cgroup_show_options, | |
5338 | .mkdir = cgroup_mkdir, | |
5339 | .rmdir = cgroup_rmdir, | |
5340 | .rename = cgroup_rename, | |
5341 | }; | |
5342 | ||
15a4c835 | 5343 | static void __init cgroup_init_subsys(struct cgroup_subsys *ss, bool early) |
ddbcc7e8 | 5344 | { |
ddbcc7e8 | 5345 | struct cgroup_subsys_state *css; |
cfe36bde | 5346 | |
a5ae9899 | 5347 | pr_debug("Initializing cgroup subsys %s\n", ss->name); |
ddbcc7e8 | 5348 | |
648bb56d TH |
5349 | mutex_lock(&cgroup_mutex); |
5350 | ||
15a4c835 | 5351 | idr_init(&ss->css_idr); |
0adb0704 | 5352 | INIT_LIST_HEAD(&ss->cfts); |
8e3f6541 | 5353 | |
3dd06ffa TH |
5354 | /* Create the root cgroup state for this subsystem */ |
5355 | ss->root = &cgrp_dfl_root; | |
5356 | css = ss->css_alloc(cgroup_css(&cgrp_dfl_root.cgrp, ss)); | |
ddbcc7e8 PM |
5357 | /* We don't handle early failures gracefully */ |
5358 | BUG_ON(IS_ERR(css)); | |
ddfcadab | 5359 | init_and_link_css(css, ss, &cgrp_dfl_root.cgrp); |
3b514d24 TH |
5360 | |
5361 | /* | |
5362 | * Root csses are never destroyed and we can't initialize | |
5363 | * percpu_ref during early init. Disable refcnting. | |
5364 | */ | |
5365 | css->flags |= CSS_NO_REF; | |
5366 | ||
15a4c835 | 5367 | if (early) { |
9395a450 | 5368 | /* allocation can't be done safely during early init */ |
15a4c835 TH |
5369 | css->id = 1; |
5370 | } else { | |
5371 | css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2, GFP_KERNEL); | |
5372 | BUG_ON(css->id < 0); | |
5373 | } | |
ddbcc7e8 | 5374 | |
e8d55fde | 5375 | /* Update the init_css_set to contain a subsys |
817929ec | 5376 | * pointer to this state - since the subsystem is |
e8d55fde | 5377 | * newly registered, all tasks and hence the |
3dd06ffa | 5378 | * init_css_set is in the subsystem's root cgroup. */ |
aec25020 | 5379 | init_css_set.subsys[ss->id] = css; |
ddbcc7e8 | 5380 | |
cb4a3167 AS |
5381 | have_fork_callback |= (bool)ss->fork << ss->id; |
5382 | have_exit_callback |= (bool)ss->exit << ss->id; | |
afcf6c8b | 5383 | have_free_callback |= (bool)ss->free << ss->id; |
7e47682e | 5384 | have_canfork_callback |= (bool)ss->can_fork << ss->id; |
ddbcc7e8 | 5385 | |
e8d55fde LZ |
5386 | /* At system boot, before all subsystems have been |
5387 | * registered, no tasks have been forked, so we don't | |
5388 | * need to invoke fork callbacks here. */ | |
5389 | BUG_ON(!list_empty(&init_task.tasks)); | |
5390 | ||
ae7f164a | 5391 | BUG_ON(online_css(css)); |
a8638030 | 5392 | |
cf5d5941 BB |
5393 | mutex_unlock(&cgroup_mutex); |
5394 | } | |
cf5d5941 | 5395 | |
ddbcc7e8 | 5396 | /** |
a043e3b2 LZ |
5397 | * cgroup_init_early - cgroup initialization at system boot |
5398 | * | |
5399 | * Initialize cgroups at system boot, and initialize any | |
5400 | * subsystems that request early init. | |
ddbcc7e8 PM |
5401 | */ |
5402 | int __init cgroup_init_early(void) | |
5403 | { | |
7b9a6ba5 | 5404 | static struct cgroup_sb_opts __initdata opts; |
30159ec7 | 5405 | struct cgroup_subsys *ss; |
ddbcc7e8 | 5406 | int i; |
30159ec7 | 5407 | |
3dd06ffa | 5408 | init_cgroup_root(&cgrp_dfl_root, &opts); |
3b514d24 TH |
5409 | cgrp_dfl_root.cgrp.self.flags |= CSS_NO_REF; |
5410 | ||
a4ea1cc9 | 5411 | RCU_INIT_POINTER(init_task.cgroups, &init_css_set); |
817929ec | 5412 | |
3ed80a62 | 5413 | for_each_subsys(ss, i) { |
aec25020 | 5414 | WARN(!ss->css_alloc || !ss->css_free || ss->name || ss->id, |
63253ad8 | 5415 | "invalid cgroup_subsys %d:%s css_alloc=%p css_free=%p id:name=%d:%s\n", |
073219e9 | 5416 | i, cgroup_subsys_name[i], ss->css_alloc, ss->css_free, |
aec25020 | 5417 | ss->id, ss->name); |
073219e9 TH |
5418 | WARN(strlen(cgroup_subsys_name[i]) > MAX_CGROUP_TYPE_NAMELEN, |
5419 | "cgroup_subsys_name %s too long\n", cgroup_subsys_name[i]); | |
5420 | ||
aec25020 | 5421 | ss->id = i; |
073219e9 | 5422 | ss->name = cgroup_subsys_name[i]; |
3e1d2eed TH |
5423 | if (!ss->legacy_name) |
5424 | ss->legacy_name = cgroup_subsys_name[i]; | |
ddbcc7e8 PM |
5425 | |
5426 | if (ss->early_init) | |
15a4c835 | 5427 | cgroup_init_subsys(ss, true); |
ddbcc7e8 PM |
5428 | } |
5429 | return 0; | |
5430 | } | |
5431 | ||
6e5c8307 | 5432 | static u16 cgroup_disable_mask __initdata; |
a3e72739 | 5433 | |
ddbcc7e8 | 5434 | /** |
a043e3b2 LZ |
5435 | * cgroup_init - cgroup initialization |
5436 | * | |
5437 | * Register cgroup filesystem and /proc file, and initialize | |
5438 | * any subsystems that didn't request early init. | |
ddbcc7e8 PM |
5439 | */ |
5440 | int __init cgroup_init(void) | |
5441 | { | |
30159ec7 | 5442 | struct cgroup_subsys *ss; |
035f4f51 | 5443 | int ssid; |
ddbcc7e8 | 5444 | |
6e5c8307 | 5445 | BUILD_BUG_ON(CGROUP_SUBSYS_COUNT > 16); |
1ed13287 | 5446 | BUG_ON(percpu_init_rwsem(&cgroup_threadgroup_rwsem)); |
a14c6874 TH |
5447 | BUG_ON(cgroup_init_cftypes(NULL, cgroup_dfl_base_files)); |
5448 | BUG_ON(cgroup_init_cftypes(NULL, cgroup_legacy_base_files)); | |
ddbcc7e8 | 5449 | |
54e7b4eb | 5450 | mutex_lock(&cgroup_mutex); |
54e7b4eb | 5451 | |
2378d8b8 TH |
5452 | /* |
5453 | * Add init_css_set to the hash table so that dfl_root can link to | |
5454 | * it during init. | |
5455 | */ | |
5456 | hash_add(css_set_table, &init_css_set.hlist, | |
5457 | css_set_hash(init_css_set.subsys)); | |
82fe9b0d | 5458 | |
3dd06ffa | 5459 | BUG_ON(cgroup_setup_root(&cgrp_dfl_root, 0)); |
4e96ee8e | 5460 | |
54e7b4eb TH |
5461 | mutex_unlock(&cgroup_mutex); |
5462 | ||
172a2c06 | 5463 | for_each_subsys(ss, ssid) { |
15a4c835 TH |
5464 | if (ss->early_init) { |
5465 | struct cgroup_subsys_state *css = | |
5466 | init_css_set.subsys[ss->id]; | |
5467 | ||
5468 | css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2, | |
5469 | GFP_KERNEL); | |
5470 | BUG_ON(css->id < 0); | |
5471 | } else { | |
5472 | cgroup_init_subsys(ss, false); | |
5473 | } | |
172a2c06 | 5474 | |
2d8f243a TH |
5475 | list_add_tail(&init_css_set.e_cset_node[ssid], |
5476 | &cgrp_dfl_root.cgrp.e_csets[ssid]); | |
172a2c06 TH |
5477 | |
5478 | /* | |
c731ae1d LZ |
5479 | * Setting dfl_root subsys_mask needs to consider the |
5480 | * disabled flag and cftype registration needs kmalloc, | |
5481 | * both of which aren't available during early_init. | |
172a2c06 | 5482 | */ |
a3e72739 TH |
5483 | if (cgroup_disable_mask & (1 << ssid)) { |
5484 | static_branch_disable(cgroup_subsys_enabled_key[ssid]); | |
5485 | printk(KERN_INFO "Disabling %s control group subsystem\n", | |
5486 | ss->name); | |
a8ddc821 | 5487 | continue; |
a3e72739 | 5488 | } |
a8ddc821 | 5489 | |
223ffb29 JW |
5490 | if (cgroup_ssid_no_v1(ssid)) |
5491 | printk(KERN_INFO "Disabling %s control group subsystem in v1 mounts\n", | |
5492 | ss->name); | |
5493 | ||
a8ddc821 TH |
5494 | cgrp_dfl_root.subsys_mask |= 1 << ss->id; |
5495 | ||
f6d635ad TH |
5496 | if (ss->implicit_on_dfl) |
5497 | cgrp_dfl_implicit_ss_mask |= 1 << ss->id; | |
5498 | else if (!ss->dfl_cftypes) | |
a7165264 | 5499 | cgrp_dfl_inhibit_ss_mask |= 1 << ss->id; |
5de4fa13 | 5500 | |
a8ddc821 TH |
5501 | if (ss->dfl_cftypes == ss->legacy_cftypes) { |
5502 | WARN_ON(cgroup_add_cftypes(ss, ss->dfl_cftypes)); | |
5503 | } else { | |
5504 | WARN_ON(cgroup_add_dfl_cftypes(ss, ss->dfl_cftypes)); | |
5505 | WARN_ON(cgroup_add_legacy_cftypes(ss, ss->legacy_cftypes)); | |
c731ae1d | 5506 | } |
295458e6 VD |
5507 | |
5508 | if (ss->bind) | |
5509 | ss->bind(init_css_set.subsys[ssid]); | |
676db4af GKH |
5510 | } |
5511 | ||
2378d8b8 TH |
5512 | /* init_css_set.subsys[] has been updated, re-hash */ |
5513 | hash_del(&init_css_set.hlist); | |
5514 | hash_add(css_set_table, &init_css_set.hlist, | |
5515 | css_set_hash(init_css_set.subsys)); | |
5516 | ||
035f4f51 TH |
5517 | WARN_ON(sysfs_create_mount_point(fs_kobj, "cgroup")); |
5518 | WARN_ON(register_filesystem(&cgroup_fs_type)); | |
67e9c74b | 5519 | WARN_ON(register_filesystem(&cgroup2_fs_type)); |
035f4f51 | 5520 | WARN_ON(!proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations)); |
ddbcc7e8 | 5521 | |
2bd59d48 | 5522 | return 0; |
ddbcc7e8 | 5523 | } |
b4f48b63 | 5524 | |
e5fca243 TH |
5525 | static int __init cgroup_wq_init(void) |
5526 | { | |
5527 | /* | |
5528 | * There isn't much point in executing destruction path in | |
5529 | * parallel. Good chunk is serialized with cgroup_mutex anyway. | |
1a11533f | 5530 | * Use 1 for @max_active. |
e5fca243 TH |
5531 | * |
5532 | * We would prefer to do this in cgroup_init() above, but that | |
5533 | * is called before init_workqueues(): so leave this until after. | |
5534 | */ | |
1a11533f | 5535 | cgroup_destroy_wq = alloc_workqueue("cgroup_destroy", 0, 1); |
e5fca243 | 5536 | BUG_ON(!cgroup_destroy_wq); |
b1a21367 TH |
5537 | |
5538 | /* | |
5539 | * Used to destroy pidlists and separate to serve as flush domain. | |
5540 | * Cap @max_active to 1 too. | |
5541 | */ | |
5542 | cgroup_pidlist_destroy_wq = alloc_workqueue("cgroup_pidlist_destroy", | |
5543 | 0, 1); | |
5544 | BUG_ON(!cgroup_pidlist_destroy_wq); | |
5545 | ||
e5fca243 TH |
5546 | return 0; |
5547 | } | |
5548 | core_initcall(cgroup_wq_init); | |
5549 | ||
a424316c PM |
5550 | /* |
5551 | * proc_cgroup_show() | |
5552 | * - Print task's cgroup paths into seq_file, one line for each hierarchy | |
5553 | * - Used for /proc/<pid>/cgroup. | |
a424316c | 5554 | */ |
006f4ac4 ZL |
5555 | int proc_cgroup_show(struct seq_file *m, struct pid_namespace *ns, |
5556 | struct pid *pid, struct task_struct *tsk) | |
a424316c | 5557 | { |
e61734c5 | 5558 | char *buf, *path; |
a424316c | 5559 | int retval; |
3dd06ffa | 5560 | struct cgroup_root *root; |
a424316c PM |
5561 | |
5562 | retval = -ENOMEM; | |
e61734c5 | 5563 | buf = kmalloc(PATH_MAX, GFP_KERNEL); |
a424316c PM |
5564 | if (!buf) |
5565 | goto out; | |
5566 | ||
a424316c | 5567 | mutex_lock(&cgroup_mutex); |
f0d9a5f1 | 5568 | spin_lock_bh(&css_set_lock); |
a424316c | 5569 | |
985ed670 | 5570 | for_each_root(root) { |
a424316c | 5571 | struct cgroup_subsys *ss; |
bd89aabc | 5572 | struct cgroup *cgrp; |
b85d2040 | 5573 | int ssid, count = 0; |
a424316c | 5574 | |
a7165264 | 5575 | if (root == &cgrp_dfl_root && !cgrp_dfl_visible) |
985ed670 TH |
5576 | continue; |
5577 | ||
2c6ab6d2 | 5578 | seq_printf(m, "%d:", root->hierarchy_id); |
d98817d4 TH |
5579 | if (root != &cgrp_dfl_root) |
5580 | for_each_subsys(ss, ssid) | |
5581 | if (root->subsys_mask & (1 << ssid)) | |
5582 | seq_printf(m, "%s%s", count++ ? "," : "", | |
3e1d2eed | 5583 | ss->legacy_name); |
c6d57f33 PM |
5584 | if (strlen(root->name)) |
5585 | seq_printf(m, "%sname=%s", count ? "," : "", | |
5586 | root->name); | |
a424316c | 5587 | seq_putc(m, ':'); |
2e91fa7f | 5588 | |
7717f7ba | 5589 | cgrp = task_cgroup_from_root(tsk, root); |
2e91fa7f TH |
5590 | |
5591 | /* | |
5592 | * On traditional hierarchies, all zombie tasks show up as | |
5593 | * belonging to the root cgroup. On the default hierarchy, | |
5594 | * while a zombie doesn't show up in "cgroup.procs" and | |
5595 | * thus can't be migrated, its /proc/PID/cgroup keeps | |
5596 | * reporting the cgroup it belonged to before exiting. If | |
5597 | * the cgroup is removed before the zombie is reaped, | |
5598 | * " (deleted)" is appended to the cgroup path. | |
5599 | */ | |
5600 | if (cgroup_on_dfl(cgrp) || !(tsk->flags & PF_EXITING)) { | |
5601 | path = cgroup_path(cgrp, buf, PATH_MAX); | |
5602 | if (!path) { | |
5603 | retval = -ENAMETOOLONG; | |
5604 | goto out_unlock; | |
5605 | } | |
5606 | } else { | |
5607 | path = "/"; | |
e61734c5 | 5608 | } |
2e91fa7f | 5609 | |
e61734c5 | 5610 | seq_puts(m, path); |
2e91fa7f TH |
5611 | |
5612 | if (cgroup_on_dfl(cgrp) && cgroup_is_dead(cgrp)) | |
5613 | seq_puts(m, " (deleted)\n"); | |
5614 | else | |
5615 | seq_putc(m, '\n'); | |
a424316c PM |
5616 | } |
5617 | ||
006f4ac4 | 5618 | retval = 0; |
a424316c | 5619 | out_unlock: |
f0d9a5f1 | 5620 | spin_unlock_bh(&css_set_lock); |
a424316c | 5621 | mutex_unlock(&cgroup_mutex); |
a424316c PM |
5622 | kfree(buf); |
5623 | out: | |
5624 | return retval; | |
5625 | } | |
5626 | ||
a424316c PM |
5627 | /* Display information about each subsystem and each hierarchy */ |
5628 | static int proc_cgroupstats_show(struct seq_file *m, void *v) | |
5629 | { | |
30159ec7 | 5630 | struct cgroup_subsys *ss; |
a424316c | 5631 | int i; |
a424316c | 5632 | |
8bab8dde | 5633 | seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n"); |
aae8aab4 BB |
5634 | /* |
5635 | * ideally we don't want subsystems moving around while we do this. | |
5636 | * cgroup_mutex is also necessary to guarantee an atomic snapshot of | |
5637 | * subsys/hierarchy state. | |
5638 | */ | |
a424316c | 5639 | mutex_lock(&cgroup_mutex); |
30159ec7 TH |
5640 | |
5641 | for_each_subsys(ss, i) | |
2c6ab6d2 | 5642 | seq_printf(m, "%s\t%d\t%d\t%d\n", |
3e1d2eed | 5643 | ss->legacy_name, ss->root->hierarchy_id, |
fc5ed1e9 TH |
5644 | atomic_read(&ss->root->nr_cgrps), |
5645 | cgroup_ssid_enabled(i)); | |
30159ec7 | 5646 | |
a424316c PM |
5647 | mutex_unlock(&cgroup_mutex); |
5648 | return 0; | |
5649 | } | |
5650 | ||
5651 | static int cgroupstats_open(struct inode *inode, struct file *file) | |
5652 | { | |
9dce07f1 | 5653 | return single_open(file, proc_cgroupstats_show, NULL); |
a424316c PM |
5654 | } |
5655 | ||
828c0950 | 5656 | static const struct file_operations proc_cgroupstats_operations = { |
a424316c PM |
5657 | .open = cgroupstats_open, |
5658 | .read = seq_read, | |
5659 | .llseek = seq_lseek, | |
5660 | .release = single_release, | |
5661 | }; | |
5662 | ||
b4f48b63 | 5663 | /** |
eaf797ab | 5664 | * cgroup_fork - initialize cgroup related fields during copy_process() |
a043e3b2 | 5665 | * @child: pointer to task_struct of forking parent process. |
b4f48b63 | 5666 | * |
eaf797ab TH |
5667 | * A task is associated with the init_css_set until cgroup_post_fork() |
5668 | * attaches it to the parent's css_set. Empty cg_list indicates that | |
5669 | * @child isn't holding reference to its css_set. | |
b4f48b63 PM |
5670 | */ |
5671 | void cgroup_fork(struct task_struct *child) | |
5672 | { | |
eaf797ab | 5673 | RCU_INIT_POINTER(child->cgroups, &init_css_set); |
817929ec | 5674 | INIT_LIST_HEAD(&child->cg_list); |
b4f48b63 PM |
5675 | } |
5676 | ||
7e47682e AS |
5677 | /** |
5678 | * cgroup_can_fork - called on a new task before the process is exposed | |
5679 | * @child: the task in question. | |
5680 | * | |
5681 | * This calls the subsystem can_fork() callbacks. If the can_fork() callback | |
5682 | * returns an error, the fork aborts with that error code. This allows for | |
5683 | * a cgroup subsystem to conditionally allow or deny new forks. | |
5684 | */ | |
b53202e6 | 5685 | int cgroup_can_fork(struct task_struct *child) |
7e47682e AS |
5686 | { |
5687 | struct cgroup_subsys *ss; | |
5688 | int i, j, ret; | |
5689 | ||
b4e0eeaf | 5690 | do_each_subsys_mask(ss, i, have_canfork_callback) { |
b53202e6 | 5691 | ret = ss->can_fork(child); |
7e47682e AS |
5692 | if (ret) |
5693 | goto out_revert; | |
b4e0eeaf | 5694 | } while_each_subsys_mask(); |
7e47682e AS |
5695 | |
5696 | return 0; | |
5697 | ||
5698 | out_revert: | |
5699 | for_each_subsys(ss, j) { | |
5700 | if (j >= i) | |
5701 | break; | |
5702 | if (ss->cancel_fork) | |
b53202e6 | 5703 | ss->cancel_fork(child); |
7e47682e AS |
5704 | } |
5705 | ||
5706 | return ret; | |
5707 | } | |
5708 | ||
5709 | /** | |
5710 | * cgroup_cancel_fork - called if a fork failed after cgroup_can_fork() | |
5711 | * @child: the task in question | |
5712 | * | |
5713 | * This calls the cancel_fork() callbacks if a fork failed *after* | |
5714 | * cgroup_can_fork() succeded. | |
5715 | */ | |
b53202e6 | 5716 | void cgroup_cancel_fork(struct task_struct *child) |
7e47682e AS |
5717 | { |
5718 | struct cgroup_subsys *ss; | |
5719 | int i; | |
5720 | ||
5721 | for_each_subsys(ss, i) | |
5722 | if (ss->cancel_fork) | |
b53202e6 | 5723 | ss->cancel_fork(child); |
7e47682e AS |
5724 | } |
5725 | ||
817929ec | 5726 | /** |
a043e3b2 LZ |
5727 | * cgroup_post_fork - called on a new task after adding it to the task list |
5728 | * @child: the task in question | |
5729 | * | |
5edee61e TH |
5730 | * Adds the task to the list running through its css_set if necessary and |
5731 | * call the subsystem fork() callbacks. Has to be after the task is | |
5732 | * visible on the task list in case we race with the first call to | |
0942eeee | 5733 | * cgroup_task_iter_start() - to guarantee that the new task ends up on its |
5edee61e | 5734 | * list. |
a043e3b2 | 5735 | */ |
b53202e6 | 5736 | void cgroup_post_fork(struct task_struct *child) |
817929ec | 5737 | { |
30159ec7 | 5738 | struct cgroup_subsys *ss; |
5edee61e TH |
5739 | int i; |
5740 | ||
3ce3230a | 5741 | /* |
251f8c03 | 5742 | * This may race against cgroup_enable_task_cg_lists(). As that |
eaf797ab TH |
5743 | * function sets use_task_css_set_links before grabbing |
5744 | * tasklist_lock and we just went through tasklist_lock to add | |
5745 | * @child, it's guaranteed that either we see the set | |
5746 | * use_task_css_set_links or cgroup_enable_task_cg_lists() sees | |
5747 | * @child during its iteration. | |
5748 | * | |
5749 | * If we won the race, @child is associated with %current's | |
f0d9a5f1 | 5750 | * css_set. Grabbing css_set_lock guarantees both that the |
eaf797ab TH |
5751 | * association is stable, and, on completion of the parent's |
5752 | * migration, @child is visible in the source of migration or | |
5753 | * already in the destination cgroup. This guarantee is necessary | |
5754 | * when implementing operations which need to migrate all tasks of | |
5755 | * a cgroup to another. | |
5756 | * | |
251f8c03 | 5757 | * Note that if we lose to cgroup_enable_task_cg_lists(), @child |
eaf797ab TH |
5758 | * will remain in init_css_set. This is safe because all tasks are |
5759 | * in the init_css_set before cg_links is enabled and there's no | |
5760 | * operation which transfers all tasks out of init_css_set. | |
3ce3230a | 5761 | */ |
817929ec | 5762 | if (use_task_css_set_links) { |
eaf797ab TH |
5763 | struct css_set *cset; |
5764 | ||
f0d9a5f1 | 5765 | spin_lock_bh(&css_set_lock); |
0e1d768f | 5766 | cset = task_css_set(current); |
eaf797ab | 5767 | if (list_empty(&child->cg_list)) { |
eaf797ab | 5768 | get_css_set(cset); |
f6d7d049 | 5769 | css_set_move_task(child, NULL, cset, false); |
eaf797ab | 5770 | } |
f0d9a5f1 | 5771 | spin_unlock_bh(&css_set_lock); |
817929ec | 5772 | } |
5edee61e TH |
5773 | |
5774 | /* | |
5775 | * Call ss->fork(). This must happen after @child is linked on | |
5776 | * css_set; otherwise, @child might change state between ->fork() | |
5777 | * and addition to css_set. | |
5778 | */ | |
b4e0eeaf | 5779 | do_each_subsys_mask(ss, i, have_fork_callback) { |
b53202e6 | 5780 | ss->fork(child); |
b4e0eeaf | 5781 | } while_each_subsys_mask(); |
817929ec | 5782 | } |
5edee61e | 5783 | |
b4f48b63 PM |
5784 | /** |
5785 | * cgroup_exit - detach cgroup from exiting task | |
5786 | * @tsk: pointer to task_struct of exiting process | |
5787 | * | |
5788 | * Description: Detach cgroup from @tsk and release it. | |
5789 | * | |
5790 | * Note that cgroups marked notify_on_release force every task in | |
5791 | * them to take the global cgroup_mutex mutex when exiting. | |
5792 | * This could impact scaling on very large systems. Be reluctant to | |
5793 | * use notify_on_release cgroups where very high task exit scaling | |
5794 | * is required on large systems. | |
5795 | * | |
0e1d768f TH |
5796 | * We set the exiting tasks cgroup to the root cgroup (top_cgroup). We |
5797 | * call cgroup_exit() while the task is still competent to handle | |
5798 | * notify_on_release(), then leave the task attached to the root cgroup in | |
5799 | * each hierarchy for the remainder of its exit. No need to bother with | |
5800 | * init_css_set refcnting. init_css_set never goes away and we can't race | |
e8604cb4 | 5801 | * with migration path - PF_EXITING is visible to migration path. |
b4f48b63 | 5802 | */ |
1ec41830 | 5803 | void cgroup_exit(struct task_struct *tsk) |
b4f48b63 | 5804 | { |
30159ec7 | 5805 | struct cgroup_subsys *ss; |
5abb8855 | 5806 | struct css_set *cset; |
d41d5a01 | 5807 | int i; |
817929ec PM |
5808 | |
5809 | /* | |
0e1d768f | 5810 | * Unlink from @tsk from its css_set. As migration path can't race |
0de0942d | 5811 | * with us, we can check css_set and cg_list without synchronization. |
817929ec | 5812 | */ |
0de0942d TH |
5813 | cset = task_css_set(tsk); |
5814 | ||
817929ec | 5815 | if (!list_empty(&tsk->cg_list)) { |
f0d9a5f1 | 5816 | spin_lock_bh(&css_set_lock); |
f6d7d049 | 5817 | css_set_move_task(tsk, cset, NULL, false); |
f0d9a5f1 | 5818 | spin_unlock_bh(&css_set_lock); |
2e91fa7f TH |
5819 | } else { |
5820 | get_css_set(cset); | |
817929ec PM |
5821 | } |
5822 | ||
cb4a3167 | 5823 | /* see cgroup_post_fork() for details */ |
b4e0eeaf | 5824 | do_each_subsys_mask(ss, i, have_exit_callback) { |
2e91fa7f | 5825 | ss->exit(tsk); |
b4e0eeaf | 5826 | } while_each_subsys_mask(); |
2e91fa7f | 5827 | } |
30159ec7 | 5828 | |
2e91fa7f TH |
5829 | void cgroup_free(struct task_struct *task) |
5830 | { | |
5831 | struct css_set *cset = task_css_set(task); | |
afcf6c8b TH |
5832 | struct cgroup_subsys *ss; |
5833 | int ssid; | |
5834 | ||
b4e0eeaf | 5835 | do_each_subsys_mask(ss, ssid, have_free_callback) { |
afcf6c8b | 5836 | ss->free(task); |
b4e0eeaf | 5837 | } while_each_subsys_mask(); |
d41d5a01 | 5838 | |
2e91fa7f | 5839 | put_css_set(cset); |
b4f48b63 | 5840 | } |
697f4161 | 5841 | |
bd89aabc | 5842 | static void check_for_release(struct cgroup *cgrp) |
81a6a5cd | 5843 | { |
27bd4dbb | 5844 | if (notify_on_release(cgrp) && !cgroup_is_populated(cgrp) && |
971ff493 ZL |
5845 | !css_has_online_children(&cgrp->self) && !cgroup_is_dead(cgrp)) |
5846 | schedule_work(&cgrp->release_agent_work); | |
81a6a5cd PM |
5847 | } |
5848 | ||
81a6a5cd PM |
5849 | /* |
5850 | * Notify userspace when a cgroup is released, by running the | |
5851 | * configured release agent with the name of the cgroup (path | |
5852 | * relative to the root of cgroup file system) as the argument. | |
5853 | * | |
5854 | * Most likely, this user command will try to rmdir this cgroup. | |
5855 | * | |
5856 | * This races with the possibility that some other task will be | |
5857 | * attached to this cgroup before it is removed, or that some other | |
5858 | * user task will 'mkdir' a child cgroup of this cgroup. That's ok. | |
5859 | * The presumed 'rmdir' will fail quietly if this cgroup is no longer | |
5860 | * unused, and this cgroup will be reprieved from its death sentence, | |
5861 | * to continue to serve a useful existence. Next time it's released, | |
5862 | * we will get notified again, if it still has 'notify_on_release' set. | |
5863 | * | |
5864 | * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which | |
5865 | * means only wait until the task is successfully execve()'d. The | |
5866 | * separate release agent task is forked by call_usermodehelper(), | |
5867 | * then control in this thread returns here, without waiting for the | |
5868 | * release agent task. We don't bother to wait because the caller of | |
5869 | * this routine has no use for the exit status of the release agent | |
5870 | * task, so no sense holding our caller up for that. | |
81a6a5cd | 5871 | */ |
81a6a5cd PM |
5872 | static void cgroup_release_agent(struct work_struct *work) |
5873 | { | |
971ff493 ZL |
5874 | struct cgroup *cgrp = |
5875 | container_of(work, struct cgroup, release_agent_work); | |
5876 | char *pathbuf = NULL, *agentbuf = NULL, *path; | |
5877 | char *argv[3], *envp[3]; | |
5878 | ||
81a6a5cd | 5879 | mutex_lock(&cgroup_mutex); |
971ff493 ZL |
5880 | |
5881 | pathbuf = kmalloc(PATH_MAX, GFP_KERNEL); | |
5882 | agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL); | |
5883 | if (!pathbuf || !agentbuf) | |
5884 | goto out; | |
5885 | ||
5886 | path = cgroup_path(cgrp, pathbuf, PATH_MAX); | |
5887 | if (!path) | |
5888 | goto out; | |
5889 | ||
5890 | argv[0] = agentbuf; | |
5891 | argv[1] = path; | |
5892 | argv[2] = NULL; | |
5893 | ||
5894 | /* minimal command environment */ | |
5895 | envp[0] = "HOME=/"; | |
5896 | envp[1] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin"; | |
5897 | envp[2] = NULL; | |
5898 | ||
81a6a5cd | 5899 | mutex_unlock(&cgroup_mutex); |
971ff493 | 5900 | call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC); |
3e2cd91a | 5901 | goto out_free; |
971ff493 | 5902 | out: |
81a6a5cd | 5903 | mutex_unlock(&cgroup_mutex); |
3e2cd91a | 5904 | out_free: |
971ff493 ZL |
5905 | kfree(agentbuf); |
5906 | kfree(pathbuf); | |
81a6a5cd | 5907 | } |
8bab8dde PM |
5908 | |
5909 | static int __init cgroup_disable(char *str) | |
5910 | { | |
30159ec7 | 5911 | struct cgroup_subsys *ss; |
8bab8dde | 5912 | char *token; |
30159ec7 | 5913 | int i; |
8bab8dde PM |
5914 | |
5915 | while ((token = strsep(&str, ",")) != NULL) { | |
5916 | if (!*token) | |
5917 | continue; | |
be45c900 | 5918 | |
3ed80a62 | 5919 | for_each_subsys(ss, i) { |
3e1d2eed TH |
5920 | if (strcmp(token, ss->name) && |
5921 | strcmp(token, ss->legacy_name)) | |
5922 | continue; | |
a3e72739 | 5923 | cgroup_disable_mask |= 1 << i; |
8bab8dde PM |
5924 | } |
5925 | } | |
5926 | return 1; | |
5927 | } | |
5928 | __setup("cgroup_disable=", cgroup_disable); | |
38460b48 | 5929 | |
223ffb29 JW |
5930 | static int __init cgroup_no_v1(char *str) |
5931 | { | |
5932 | struct cgroup_subsys *ss; | |
5933 | char *token; | |
5934 | int i; | |
5935 | ||
5936 | while ((token = strsep(&str, ",")) != NULL) { | |
5937 | if (!*token) | |
5938 | continue; | |
5939 | ||
5940 | if (!strcmp(token, "all")) { | |
6e5c8307 | 5941 | cgroup_no_v1_mask = U16_MAX; |
223ffb29 JW |
5942 | break; |
5943 | } | |
5944 | ||
5945 | for_each_subsys(ss, i) { | |
5946 | if (strcmp(token, ss->name) && | |
5947 | strcmp(token, ss->legacy_name)) | |
5948 | continue; | |
5949 | ||
5950 | cgroup_no_v1_mask |= 1 << i; | |
5951 | } | |
5952 | } | |
5953 | return 1; | |
5954 | } | |
5955 | __setup("cgroup_no_v1=", cgroup_no_v1); | |
5956 | ||
b77d7b60 | 5957 | /** |
ec903c0c | 5958 | * css_tryget_online_from_dir - get corresponding css from a cgroup dentry |
35cf0836 TH |
5959 | * @dentry: directory dentry of interest |
5960 | * @ss: subsystem of interest | |
b77d7b60 | 5961 | * |
5a17f543 TH |
5962 | * If @dentry is a directory for a cgroup which has @ss enabled on it, try |
5963 | * to get the corresponding css and return it. If such css doesn't exist | |
5964 | * or can't be pinned, an ERR_PTR value is returned. | |
e5d1367f | 5965 | */ |
ec903c0c TH |
5966 | struct cgroup_subsys_state *css_tryget_online_from_dir(struct dentry *dentry, |
5967 | struct cgroup_subsys *ss) | |
e5d1367f | 5968 | { |
2bd59d48 | 5969 | struct kernfs_node *kn = kernfs_node_from_dentry(dentry); |
f17fc25f | 5970 | struct file_system_type *s_type = dentry->d_sb->s_type; |
2bd59d48 | 5971 | struct cgroup_subsys_state *css = NULL; |
e5d1367f | 5972 | struct cgroup *cgrp; |
e5d1367f | 5973 | |
35cf0836 | 5974 | /* is @dentry a cgroup dir? */ |
f17fc25f TH |
5975 | if ((s_type != &cgroup_fs_type && s_type != &cgroup2_fs_type) || |
5976 | !kn || kernfs_type(kn) != KERNFS_DIR) | |
e5d1367f SE |
5977 | return ERR_PTR(-EBADF); |
5978 | ||
5a17f543 TH |
5979 | rcu_read_lock(); |
5980 | ||
2bd59d48 TH |
5981 | /* |
5982 | * This path doesn't originate from kernfs and @kn could already | |
5983 | * have been or be removed at any point. @kn->priv is RCU | |
a4189487 | 5984 | * protected for this access. See css_release_work_fn() for details. |
2bd59d48 TH |
5985 | */ |
5986 | cgrp = rcu_dereference(kn->priv); | |
5987 | if (cgrp) | |
5988 | css = cgroup_css(cgrp, ss); | |
5a17f543 | 5989 | |
ec903c0c | 5990 | if (!css || !css_tryget_online(css)) |
5a17f543 TH |
5991 | css = ERR_PTR(-ENOENT); |
5992 | ||
5993 | rcu_read_unlock(); | |
5994 | return css; | |
e5d1367f | 5995 | } |
e5d1367f | 5996 | |
1cb650b9 LZ |
5997 | /** |
5998 | * css_from_id - lookup css by id | |
5999 | * @id: the cgroup id | |
6000 | * @ss: cgroup subsys to be looked into | |
6001 | * | |
6002 | * Returns the css if there's valid one with @id, otherwise returns NULL. | |
6003 | * Should be called under rcu_read_lock(). | |
6004 | */ | |
6005 | struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss) | |
6006 | { | |
6fa4918d | 6007 | WARN_ON_ONCE(!rcu_read_lock_held()); |
adbe427b | 6008 | return id > 0 ? idr_find(&ss->css_idr, id) : NULL; |
e5d1367f SE |
6009 | } |
6010 | ||
16af4396 TH |
6011 | /** |
6012 | * cgroup_get_from_path - lookup and get a cgroup from its default hierarchy path | |
6013 | * @path: path on the default hierarchy | |
6014 | * | |
6015 | * Find the cgroup at @path on the default hierarchy, increment its | |
6016 | * reference count and return it. Returns pointer to the found cgroup on | |
6017 | * success, ERR_PTR(-ENOENT) if @path doens't exist and ERR_PTR(-ENOTDIR) | |
6018 | * if @path points to a non-directory. | |
6019 | */ | |
6020 | struct cgroup *cgroup_get_from_path(const char *path) | |
6021 | { | |
6022 | struct kernfs_node *kn; | |
6023 | struct cgroup *cgrp; | |
6024 | ||
6025 | mutex_lock(&cgroup_mutex); | |
6026 | ||
6027 | kn = kernfs_walk_and_get(cgrp_dfl_root.cgrp.kn, path); | |
6028 | if (kn) { | |
6029 | if (kernfs_type(kn) == KERNFS_DIR) { | |
6030 | cgrp = kn->priv; | |
6031 | cgroup_get(cgrp); | |
6032 | } else { | |
6033 | cgrp = ERR_PTR(-ENOTDIR); | |
6034 | } | |
6035 | kernfs_put(kn); | |
6036 | } else { | |
6037 | cgrp = ERR_PTR(-ENOENT); | |
6038 | } | |
6039 | ||
6040 | mutex_unlock(&cgroup_mutex); | |
6041 | return cgrp; | |
6042 | } | |
6043 | EXPORT_SYMBOL_GPL(cgroup_get_from_path); | |
6044 | ||
bd1060a1 TH |
6045 | /* |
6046 | * sock->sk_cgrp_data handling. For more info, see sock_cgroup_data | |
6047 | * definition in cgroup-defs.h. | |
6048 | */ | |
6049 | #ifdef CONFIG_SOCK_CGROUP_DATA | |
6050 | ||
6051 | #if defined(CONFIG_CGROUP_NET_PRIO) || defined(CONFIG_CGROUP_NET_CLASSID) | |
6052 | ||
3fa4cc9c | 6053 | DEFINE_SPINLOCK(cgroup_sk_update_lock); |
bd1060a1 TH |
6054 | static bool cgroup_sk_alloc_disabled __read_mostly; |
6055 | ||
6056 | void cgroup_sk_alloc_disable(void) | |
6057 | { | |
6058 | if (cgroup_sk_alloc_disabled) | |
6059 | return; | |
6060 | pr_info("cgroup: disabling cgroup2 socket matching due to net_prio or net_cls activation\n"); | |
6061 | cgroup_sk_alloc_disabled = true; | |
6062 | } | |
6063 | ||
6064 | #else | |
6065 | ||
6066 | #define cgroup_sk_alloc_disabled false | |
6067 | ||
6068 | #endif | |
6069 | ||
6070 | void cgroup_sk_alloc(struct sock_cgroup_data *skcd) | |
6071 | { | |
6072 | if (cgroup_sk_alloc_disabled) | |
6073 | return; | |
6074 | ||
6075 | rcu_read_lock(); | |
6076 | ||
6077 | while (true) { | |
6078 | struct css_set *cset; | |
6079 | ||
6080 | cset = task_css_set(current); | |
6081 | if (likely(cgroup_tryget(cset->dfl_cgrp))) { | |
6082 | skcd->val = (unsigned long)cset->dfl_cgrp; | |
6083 | break; | |
6084 | } | |
6085 | cpu_relax(); | |
6086 | } | |
6087 | ||
6088 | rcu_read_unlock(); | |
6089 | } | |
6090 | ||
6091 | void cgroup_sk_free(struct sock_cgroup_data *skcd) | |
6092 | { | |
6093 | cgroup_put(sock_cgroup_ptr(skcd)); | |
6094 | } | |
6095 | ||
6096 | #endif /* CONFIG_SOCK_CGROUP_DATA */ | |
6097 | ||
fe693435 | 6098 | #ifdef CONFIG_CGROUP_DEBUG |
eb95419b TH |
6099 | static struct cgroup_subsys_state * |
6100 | debug_css_alloc(struct cgroup_subsys_state *parent_css) | |
fe693435 PM |
6101 | { |
6102 | struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL); | |
6103 | ||
6104 | if (!css) | |
6105 | return ERR_PTR(-ENOMEM); | |
6106 | ||
6107 | return css; | |
6108 | } | |
6109 | ||
eb95419b | 6110 | static void debug_css_free(struct cgroup_subsys_state *css) |
fe693435 | 6111 | { |
eb95419b | 6112 | kfree(css); |
fe693435 PM |
6113 | } |
6114 | ||
182446d0 TH |
6115 | static u64 debug_taskcount_read(struct cgroup_subsys_state *css, |
6116 | struct cftype *cft) | |
fe693435 | 6117 | { |
182446d0 | 6118 | return cgroup_task_count(css->cgroup); |
fe693435 PM |
6119 | } |
6120 | ||
182446d0 TH |
6121 | static u64 current_css_set_read(struct cgroup_subsys_state *css, |
6122 | struct cftype *cft) | |
fe693435 PM |
6123 | { |
6124 | return (u64)(unsigned long)current->cgroups; | |
6125 | } | |
6126 | ||
182446d0 | 6127 | static u64 current_css_set_refcount_read(struct cgroup_subsys_state *css, |
03c78cbe | 6128 | struct cftype *cft) |
fe693435 PM |
6129 | { |
6130 | u64 count; | |
6131 | ||
6132 | rcu_read_lock(); | |
a8ad805c | 6133 | count = atomic_read(&task_css_set(current)->refcount); |
fe693435 PM |
6134 | rcu_read_unlock(); |
6135 | return count; | |
6136 | } | |
6137 | ||
2da8ca82 | 6138 | static int current_css_set_cg_links_read(struct seq_file *seq, void *v) |
7717f7ba | 6139 | { |
69d0206c | 6140 | struct cgrp_cset_link *link; |
5abb8855 | 6141 | struct css_set *cset; |
e61734c5 TH |
6142 | char *name_buf; |
6143 | ||
6144 | name_buf = kmalloc(NAME_MAX + 1, GFP_KERNEL); | |
6145 | if (!name_buf) | |
6146 | return -ENOMEM; | |
7717f7ba | 6147 | |
f0d9a5f1 | 6148 | spin_lock_bh(&css_set_lock); |
7717f7ba | 6149 | rcu_read_lock(); |
5abb8855 | 6150 | cset = rcu_dereference(current->cgroups); |
69d0206c | 6151 | list_for_each_entry(link, &cset->cgrp_links, cgrp_link) { |
7717f7ba | 6152 | struct cgroup *c = link->cgrp; |
7717f7ba | 6153 | |
a2dd4247 | 6154 | cgroup_name(c, name_buf, NAME_MAX + 1); |
2c6ab6d2 | 6155 | seq_printf(seq, "Root %d group %s\n", |
a2dd4247 | 6156 | c->root->hierarchy_id, name_buf); |
7717f7ba PM |
6157 | } |
6158 | rcu_read_unlock(); | |
f0d9a5f1 | 6159 | spin_unlock_bh(&css_set_lock); |
e61734c5 | 6160 | kfree(name_buf); |
7717f7ba PM |
6161 | return 0; |
6162 | } | |
6163 | ||
6164 | #define MAX_TASKS_SHOWN_PER_CSS 25 | |
2da8ca82 | 6165 | static int cgroup_css_links_read(struct seq_file *seq, void *v) |
7717f7ba | 6166 | { |
2da8ca82 | 6167 | struct cgroup_subsys_state *css = seq_css(seq); |
69d0206c | 6168 | struct cgrp_cset_link *link; |
7717f7ba | 6169 | |
f0d9a5f1 | 6170 | spin_lock_bh(&css_set_lock); |
182446d0 | 6171 | list_for_each_entry(link, &css->cgroup->cset_links, cset_link) { |
69d0206c | 6172 | struct css_set *cset = link->cset; |
7717f7ba PM |
6173 | struct task_struct *task; |
6174 | int count = 0; | |
c7561128 | 6175 | |
5abb8855 | 6176 | seq_printf(seq, "css_set %p\n", cset); |
c7561128 | 6177 | |
5abb8855 | 6178 | list_for_each_entry(task, &cset->tasks, cg_list) { |
c7561128 TH |
6179 | if (count++ > MAX_TASKS_SHOWN_PER_CSS) |
6180 | goto overflow; | |
6181 | seq_printf(seq, " task %d\n", task_pid_vnr(task)); | |
6182 | } | |
6183 | ||
6184 | list_for_each_entry(task, &cset->mg_tasks, cg_list) { | |
6185 | if (count++ > MAX_TASKS_SHOWN_PER_CSS) | |
6186 | goto overflow; | |
6187 | seq_printf(seq, " task %d\n", task_pid_vnr(task)); | |
7717f7ba | 6188 | } |
c7561128 TH |
6189 | continue; |
6190 | overflow: | |
6191 | seq_puts(seq, " ...\n"); | |
7717f7ba | 6192 | } |
f0d9a5f1 | 6193 | spin_unlock_bh(&css_set_lock); |
7717f7ba PM |
6194 | return 0; |
6195 | } | |
6196 | ||
182446d0 | 6197 | static u64 releasable_read(struct cgroup_subsys_state *css, struct cftype *cft) |
fe693435 | 6198 | { |
27bd4dbb | 6199 | return (!cgroup_is_populated(css->cgroup) && |
a25eb52e | 6200 | !css_has_online_children(&css->cgroup->self)); |
fe693435 PM |
6201 | } |
6202 | ||
6203 | static struct cftype debug_files[] = { | |
fe693435 PM |
6204 | { |
6205 | .name = "taskcount", | |
6206 | .read_u64 = debug_taskcount_read, | |
6207 | }, | |
6208 | ||
6209 | { | |
6210 | .name = "current_css_set", | |
6211 | .read_u64 = current_css_set_read, | |
6212 | }, | |
6213 | ||
6214 | { | |
6215 | .name = "current_css_set_refcount", | |
6216 | .read_u64 = current_css_set_refcount_read, | |
6217 | }, | |
6218 | ||
7717f7ba PM |
6219 | { |
6220 | .name = "current_css_set_cg_links", | |
2da8ca82 | 6221 | .seq_show = current_css_set_cg_links_read, |
7717f7ba PM |
6222 | }, |
6223 | ||
6224 | { | |
6225 | .name = "cgroup_css_links", | |
2da8ca82 | 6226 | .seq_show = cgroup_css_links_read, |
7717f7ba PM |
6227 | }, |
6228 | ||
fe693435 PM |
6229 | { |
6230 | .name = "releasable", | |
6231 | .read_u64 = releasable_read, | |
6232 | }, | |
fe693435 | 6233 | |
4baf6e33 TH |
6234 | { } /* terminate */ |
6235 | }; | |
fe693435 | 6236 | |
073219e9 | 6237 | struct cgroup_subsys debug_cgrp_subsys = { |
92fb9748 TH |
6238 | .css_alloc = debug_css_alloc, |
6239 | .css_free = debug_css_free, | |
5577964e | 6240 | .legacy_cftypes = debug_files, |
fe693435 PM |
6241 | }; |
6242 | #endif /* CONFIG_CGROUP_DEBUG */ |