blob: 1b72d56edce5e2d9253c2c03f203e81aa76bd16c
1 | /* |
2 | * cgroup_freezer.c - control group freezer subsystem |
3 | * |
4 | * Copyright IBM Corporation, 2007 |
5 | * |
6 | * Author : Cedric Le Goater <clg@fr.ibm.com> |
7 | * |
8 | * This program is free software; you can redistribute it and/or modify it |
9 | * under the terms of version 2.1 of the GNU Lesser General Public License |
10 | * as published by the Free Software Foundation. |
11 | * |
12 | * This program is distributed in the hope that it would be useful, but |
13 | * WITHOUT ANY WARRANTY; without even the implied warranty of |
14 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. |
15 | */ |
16 | |
17 | #include <linux/export.h> |
18 | #include <linux/slab.h> |
19 | #include <linux/cgroup.h> |
20 | #include <linux/fs.h> |
21 | #include <linux/uaccess.h> |
22 | #include <linux/freezer.h> |
23 | #include <linux/seq_file.h> |
24 | #include <linux/mutex.h> |
25 | |
26 | /* |
27 | * A cgroup is freezing if any FREEZING flags are set. FREEZING_SELF is |
28 | * set if "FROZEN" is written to freezer.state cgroupfs file, and cleared |
29 | * for "THAWED". FREEZING_PARENT is set if the parent freezer is FREEZING |
30 | * for whatever reason. IOW, a cgroup has FREEZING_PARENT set if one of |
31 | * its ancestors has FREEZING_SELF set. |
32 | */ |
33 | enum freezer_state_flags { |
34 | CGROUP_FREEZER_ONLINE = (1 << 0), /* freezer is fully online */ |
35 | CGROUP_FREEZING_SELF = (1 << 1), /* this freezer is freezing */ |
36 | CGROUP_FREEZING_PARENT = (1 << 2), /* the parent freezer is freezing */ |
37 | CGROUP_FROZEN = (1 << 3), /* this and its descendants frozen */ |
38 | |
39 | /* mask for all FREEZING flags */ |
40 | CGROUP_FREEZING = CGROUP_FREEZING_SELF | CGROUP_FREEZING_PARENT, |
41 | }; |
42 | |
43 | struct freezer { |
44 | struct cgroup_subsys_state css; |
45 | unsigned int state; |
46 | }; |
47 | |
48 | static DEFINE_MUTEX(freezer_mutex); |
49 | |
50 | static inline struct freezer *css_freezer(struct cgroup_subsys_state *css) |
51 | { |
52 | return css ? container_of(css, struct freezer, css) : NULL; |
53 | } |
54 | |
55 | static inline struct freezer *task_freezer(struct task_struct *task) |
56 | { |
57 | return css_freezer(task_css(task, freezer_cgrp_id)); |
58 | } |
59 | |
60 | static struct freezer *parent_freezer(struct freezer *freezer) |
61 | { |
62 | return css_freezer(freezer->css.parent); |
63 | } |
64 | |
65 | bool cgroup_freezing(struct task_struct *task) |
66 | { |
67 | bool ret; |
68 | |
69 | rcu_read_lock(); |
70 | ret = task_freezer(task)->state & CGROUP_FREEZING; |
71 | rcu_read_unlock(); |
72 | |
73 | return ret; |
74 | } |
75 | |
76 | static const char *freezer_state_strs(unsigned int state) |
77 | { |
78 | if (state & CGROUP_FROZEN) |
79 | return "FROZEN"; |
80 | if (state & CGROUP_FREEZING) |
81 | return "FREEZING"; |
82 | return "THAWED"; |
83 | }; |
84 | |
85 | static struct cgroup_subsys_state * |
86 | freezer_css_alloc(struct cgroup_subsys_state *parent_css) |
87 | { |
88 | struct freezer *freezer; |
89 | |
90 | freezer = kzalloc(sizeof(struct freezer), GFP_KERNEL); |
91 | if (!freezer) |
92 | return ERR_PTR(-ENOMEM); |
93 | |
94 | return &freezer->css; |
95 | } |
96 | |
97 | /** |
98 | * freezer_css_online - commit creation of a freezer css |
99 | * @css: css being created |
100 | * |
101 | * We're committing to creation of @css. Mark it online and inherit |
102 | * parent's freezing state while holding both parent's and our |
103 | * freezer->lock. |
104 | */ |
105 | static int freezer_css_online(struct cgroup_subsys_state *css) |
106 | { |
107 | struct freezer *freezer = css_freezer(css); |
108 | struct freezer *parent = parent_freezer(freezer); |
109 | |
110 | mutex_lock(&freezer_mutex); |
111 | |
112 | freezer->state |= CGROUP_FREEZER_ONLINE; |
113 | |
114 | if (parent && (parent->state & CGROUP_FREEZING)) { |
115 | freezer->state |= CGROUP_FREEZING_PARENT | CGROUP_FROZEN; |
116 | atomic_inc(&system_freezing_cnt); |
117 | } |
118 | |
119 | mutex_unlock(&freezer_mutex); |
120 | return 0; |
121 | } |
122 | |
123 | /** |
124 | * freezer_css_offline - initiate destruction of a freezer css |
125 | * @css: css being destroyed |
126 | * |
127 | * @css is going away. Mark it dead and decrement system_freezing_count if |
128 | * it was holding one. |
129 | */ |
130 | static void freezer_css_offline(struct cgroup_subsys_state *css) |
131 | { |
132 | struct freezer *freezer = css_freezer(css); |
133 | |
134 | mutex_lock(&freezer_mutex); |
135 | |
136 | if (freezer->state & CGROUP_FREEZING) |
137 | atomic_dec(&system_freezing_cnt); |
138 | |
139 | freezer->state = 0; |
140 | |
141 | mutex_unlock(&freezer_mutex); |
142 | } |
143 | |
144 | static void freezer_css_free(struct cgroup_subsys_state *css) |
145 | { |
146 | kfree(css_freezer(css)); |
147 | } |
148 | |
149 | /* |
150 | * Tasks can be migrated into a different freezer anytime regardless of its |
151 | * current state. freezer_attach() is responsible for making new tasks |
152 | * conform to the current state. |
153 | * |
154 | * Freezer state changes and task migration are synchronized via |
155 | * @freezer->lock. freezer_attach() makes the new tasks conform to the |
156 | * current state and all following state changes can see the new tasks. |
157 | */ |
158 | static void freezer_attach(struct cgroup_taskset *tset) |
159 | { |
160 | struct task_struct *task; |
161 | struct cgroup_subsys_state *new_css; |
162 | |
163 | mutex_lock(&freezer_mutex); |
164 | |
165 | /* |
166 | * Make the new tasks conform to the current state of @new_css. |
167 | * For simplicity, when migrating any task to a FROZEN cgroup, we |
168 | * revert it to FREEZING and let update_if_frozen() determine the |
169 | * correct state later. |
170 | * |
171 | * Tasks in @tset are on @new_css but may not conform to its |
172 | * current state before executing the following - !frozen tasks may |
173 | * be visible in a FROZEN cgroup and frozen tasks in a THAWED one. |
174 | */ |
175 | cgroup_taskset_for_each(task, new_css, tset) { |
176 | struct freezer *freezer = css_freezer(new_css); |
177 | |
178 | if (!(freezer->state & CGROUP_FREEZING)) { |
179 | __thaw_task(task); |
180 | } else { |
181 | freeze_task(task); |
182 | /* clear FROZEN and propagate upwards */ |
183 | while (freezer && (freezer->state & CGROUP_FROZEN)) { |
184 | freezer->state &= ~CGROUP_FROZEN; |
185 | freezer = parent_freezer(freezer); |
186 | } |
187 | } |
188 | } |
189 | |
190 | mutex_unlock(&freezer_mutex); |
191 | } |
192 | |
193 | /** |
194 | * freezer_fork - cgroup post fork callback |
195 | * @task: a task which has just been forked |
196 | * |
197 | * @task has just been created and should conform to the current state of |
198 | * the cgroup_freezer it belongs to. This function may race against |
199 | * freezer_attach(). Losing to freezer_attach() means that we don't have |
200 | * to do anything as freezer_attach() will put @task into the appropriate |
201 | * state. |
202 | */ |
203 | static void freezer_fork(struct task_struct *task) |
204 | { |
205 | struct freezer *freezer; |
206 | |
207 | /* |
208 | * The root cgroup is non-freezable, so we can skip locking the |
209 | * freezer. This is safe regardless of race with task migration. |
210 | * If we didn't race or won, skipping is obviously the right thing |
211 | * to do. If we lost and root is the new cgroup, noop is still the |
212 | * right thing to do. |
213 | */ |
214 | if (task_css_is_root(task, freezer_cgrp_id)) |
215 | return; |
216 | |
217 | mutex_lock(&freezer_mutex); |
218 | rcu_read_lock(); |
219 | |
220 | freezer = task_freezer(task); |
221 | if (freezer->state & CGROUP_FREEZING) |
222 | freeze_task(task); |
223 | |
224 | rcu_read_unlock(); |
225 | mutex_unlock(&freezer_mutex); |
226 | } |
227 | |
228 | /** |
229 | * update_if_frozen - update whether a cgroup finished freezing |
230 | * @css: css of interest |
231 | * |
232 | * Once FREEZING is initiated, transition to FROZEN is lazily updated by |
233 | * calling this function. If the current state is FREEZING but not FROZEN, |
234 | * this function checks whether all tasks of this cgroup and the descendant |
235 | * cgroups finished freezing and, if so, sets FROZEN. |
236 | * |
237 | * The caller is responsible for grabbing RCU read lock and calling |
238 | * update_if_frozen() on all descendants prior to invoking this function. |
239 | * |
240 | * Task states and freezer state might disagree while tasks are being |
241 | * migrated into or out of @css, so we can't verify task states against |
242 | * @freezer state here. See freezer_attach() for details. |
243 | */ |
244 | static void update_if_frozen(struct cgroup_subsys_state *css) |
245 | { |
246 | struct freezer *freezer = css_freezer(css); |
247 | struct cgroup_subsys_state *pos; |
248 | struct css_task_iter it; |
249 | struct task_struct *task; |
250 | |
251 | lockdep_assert_held(&freezer_mutex); |
252 | |
253 | if (!(freezer->state & CGROUP_FREEZING) || |
254 | (freezer->state & CGROUP_FROZEN)) |
255 | return; |
256 | |
257 | /* are all (live) children frozen? */ |
258 | rcu_read_lock(); |
259 | css_for_each_child(pos, css) { |
260 | struct freezer *child = css_freezer(pos); |
261 | |
262 | if ((child->state & CGROUP_FREEZER_ONLINE) && |
263 | !(child->state & CGROUP_FROZEN)) { |
264 | rcu_read_unlock(); |
265 | return; |
266 | } |
267 | } |
268 | rcu_read_unlock(); |
269 | |
270 | /* are all tasks frozen? */ |
271 | css_task_iter_start(css, &it); |
272 | |
273 | while ((task = css_task_iter_next(&it))) { |
274 | if (freezing(task)) { |
275 | /* |
276 | * freezer_should_skip() indicates that the task |
277 | * should be skipped when determining freezing |
278 | * completion. Consider it frozen in addition to |
279 | * the usual frozen condition. |
280 | */ |
281 | if (!frozen(task) && !freezer_should_skip(task)) |
282 | goto out_iter_end; |
283 | } |
284 | } |
285 | |
286 | freezer->state |= CGROUP_FROZEN; |
287 | out_iter_end: |
288 | css_task_iter_end(&it); |
289 | } |
290 | |
291 | static int freezer_read(struct seq_file *m, void *v) |
292 | { |
293 | struct cgroup_subsys_state *css = seq_css(m), *pos; |
294 | |
295 | mutex_lock(&freezer_mutex); |
296 | rcu_read_lock(); |
297 | |
298 | /* update states bottom-up */ |
299 | css_for_each_descendant_post(pos, css) { |
300 | if (!css_tryget_online(pos)) |
301 | continue; |
302 | rcu_read_unlock(); |
303 | |
304 | update_if_frozen(pos); |
305 | |
306 | rcu_read_lock(); |
307 | css_put(pos); |
308 | } |
309 | |
310 | rcu_read_unlock(); |
311 | mutex_unlock(&freezer_mutex); |
312 | |
313 | seq_puts(m, freezer_state_strs(css_freezer(css)->state)); |
314 | seq_putc(m, '\n'); |
315 | return 0; |
316 | } |
317 | |
318 | static void freeze_cgroup(struct freezer *freezer) |
319 | { |
320 | struct css_task_iter it; |
321 | struct task_struct *task; |
322 | |
323 | css_task_iter_start(&freezer->css, &it); |
324 | while ((task = css_task_iter_next(&it))) |
325 | freeze_task(task); |
326 | css_task_iter_end(&it); |
327 | } |
328 | |
329 | static void unfreeze_cgroup(struct freezer *freezer) |
330 | { |
331 | struct css_task_iter it; |
332 | struct task_struct *task; |
333 | |
334 | css_task_iter_start(&freezer->css, &it); |
335 | while ((task = css_task_iter_next(&it))) |
336 | __thaw_task(task); |
337 | css_task_iter_end(&it); |
338 | } |
339 | |
340 | /** |
341 | * freezer_apply_state - apply state change to a single cgroup_freezer |
342 | * @freezer: freezer to apply state change to |
343 | * @freeze: whether to freeze or unfreeze |
344 | * @state: CGROUP_FREEZING_* flag to set or clear |
345 | * |
346 | * Set or clear @state on @cgroup according to @freeze, and perform |
347 | * freezing or thawing as necessary. |
348 | */ |
349 | static void freezer_apply_state(struct freezer *freezer, bool freeze, |
350 | unsigned int state) |
351 | { |
352 | /* also synchronizes against task migration, see freezer_attach() */ |
353 | lockdep_assert_held(&freezer_mutex); |
354 | |
355 | if (!(freezer->state & CGROUP_FREEZER_ONLINE)) |
356 | return; |
357 | |
358 | if (freeze) { |
359 | if (!(freezer->state & CGROUP_FREEZING)) |
360 | atomic_inc(&system_freezing_cnt); |
361 | freezer->state |= state; |
362 | freeze_cgroup(freezer); |
363 | } else { |
364 | bool was_freezing = freezer->state & CGROUP_FREEZING; |
365 | |
366 | freezer->state &= ~state; |
367 | |
368 | if (!(freezer->state & CGROUP_FREEZING)) { |
369 | if (was_freezing) |
370 | atomic_dec(&system_freezing_cnt); |
371 | freezer->state &= ~CGROUP_FROZEN; |
372 | unfreeze_cgroup(freezer); |
373 | } |
374 | } |
375 | } |
376 | |
377 | /** |
378 | * freezer_change_state - change the freezing state of a cgroup_freezer |
379 | * @freezer: freezer of interest |
380 | * @freeze: whether to freeze or thaw |
381 | * |
382 | * Freeze or thaw @freezer according to @freeze. The operations are |
383 | * recursive - all descendants of @freezer will be affected. |
384 | */ |
385 | static void freezer_change_state(struct freezer *freezer, bool freeze) |
386 | { |
387 | struct cgroup_subsys_state *pos; |
388 | |
389 | /* |
390 | * Update all its descendants in pre-order traversal. Each |
391 | * descendant will try to inherit its parent's FREEZING state as |
392 | * CGROUP_FREEZING_PARENT. |
393 | */ |
394 | mutex_lock(&freezer_mutex); |
395 | rcu_read_lock(); |
396 | css_for_each_descendant_pre(pos, &freezer->css) { |
397 | struct freezer *pos_f = css_freezer(pos); |
398 | struct freezer *parent = parent_freezer(pos_f); |
399 | |
400 | if (!css_tryget_online(pos)) |
401 | continue; |
402 | rcu_read_unlock(); |
403 | |
404 | if (pos_f == freezer) |
405 | freezer_apply_state(pos_f, freeze, |
406 | CGROUP_FREEZING_SELF); |
407 | else |
408 | freezer_apply_state(pos_f, |
409 | parent->state & CGROUP_FREEZING, |
410 | CGROUP_FREEZING_PARENT); |
411 | |
412 | rcu_read_lock(); |
413 | css_put(pos); |
414 | } |
415 | rcu_read_unlock(); |
416 | mutex_unlock(&freezer_mutex); |
417 | } |
418 | |
419 | static ssize_t freezer_write(struct kernfs_open_file *of, |
420 | char *buf, size_t nbytes, loff_t off) |
421 | { |
422 | bool freeze; |
423 | |
424 | buf = strstrip(buf); |
425 | |
426 | if (strcmp(buf, freezer_state_strs(0)) == 0) |
427 | freeze = false; |
428 | else if (strcmp(buf, freezer_state_strs(CGROUP_FROZEN)) == 0) |
429 | freeze = true; |
430 | else |
431 | return -EINVAL; |
432 | |
433 | freezer_change_state(css_freezer(of_css(of)), freeze); |
434 | return nbytes; |
435 | } |
436 | |
437 | static u64 freezer_self_freezing_read(struct cgroup_subsys_state *css, |
438 | struct cftype *cft) |
439 | { |
440 | struct freezer *freezer = css_freezer(css); |
441 | |
442 | return (bool)(freezer->state & CGROUP_FREEZING_SELF); |
443 | } |
444 | |
445 | static u64 freezer_parent_freezing_read(struct cgroup_subsys_state *css, |
446 | struct cftype *cft) |
447 | { |
448 | struct freezer *freezer = css_freezer(css); |
449 | |
450 | return (bool)(freezer->state & CGROUP_FREEZING_PARENT); |
451 | } |
452 | |
453 | static struct cftype files[] = { |
454 | { |
455 | .name = "state", |
456 | .flags = CFTYPE_NOT_ON_ROOT, |
457 | .seq_show = freezer_read, |
458 | .write = freezer_write, |
459 | }, |
460 | { |
461 | .name = "self_freezing", |
462 | .flags = CFTYPE_NOT_ON_ROOT, |
463 | .read_u64 = freezer_self_freezing_read, |
464 | }, |
465 | { |
466 | .name = "parent_freezing", |
467 | .flags = CFTYPE_NOT_ON_ROOT, |
468 | .read_u64 = freezer_parent_freezing_read, |
469 | }, |
470 | { } /* terminate */ |
471 | }; |
472 | |
473 | struct cgroup_subsys freezer_cgrp_subsys = { |
474 | .css_alloc = freezer_css_alloc, |
475 | .css_online = freezer_css_online, |
476 | .css_offline = freezer_css_offline, |
477 | .css_free = freezer_css_free, |
478 | .attach = freezer_attach, |
479 | .fork = freezer_fork, |
480 | .legacy_cftypes = files, |
481 | }; |
482 |