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 #include <linux/cpu.h>
26
27 /*
28 * A cgroup is freezing if any FREEZING flags are set. FREEZING_SELF is
29 * set if "FROZEN" is written to freezer.state cgroupfs file, and cleared
30 * for "THAWED". FREEZING_PARENT is set if the parent freezer is FREEZING
31 * for whatever reason. IOW, a cgroup has FREEZING_PARENT set if one of
32 * its ancestors has FREEZING_SELF set.
33 */
34 enum freezer_state_flags {
35 CGROUP_FREEZER_ONLINE = (1 << 0), /* freezer is fully online */
36 CGROUP_FREEZING_SELF = (1 << 1), /* this freezer is freezing */
37 CGROUP_FREEZING_PARENT = (1 << 2), /* the parent freezer is freezing */
38 CGROUP_FROZEN = (1 << 3), /* this and its descendants frozen */
39
40 /* mask for all FREEZING flags */
41 CGROUP_FREEZING = CGROUP_FREEZING_SELF | CGROUP_FREEZING_PARENT,
42 };
43
44 struct freezer {
45 struct cgroup_subsys_state css;
46 unsigned int state;
47 };
48
49 static DEFINE_MUTEX(freezer_mutex);
50
51 static inline struct freezer *css_freezer(struct cgroup_subsys_state *css)
52 {
53 return css ? container_of(css, struct freezer, css) : NULL;
54 }
55
56 static inline struct freezer *task_freezer(struct task_struct *task)
57 {
58 return css_freezer(task_css(task, freezer_cgrp_id));
59 }
60
61 static struct freezer *parent_freezer(struct freezer *freezer)
62 {
63 return css_freezer(freezer->css.parent);
64 }
65
66 bool cgroup_freezing(struct task_struct *task)
67 {
68 bool ret;
69 unsigned int state;
70
71 rcu_read_lock();
72 /* Check if the cgroup is still FREEZING, but not FROZEN. The extra
73 * !FROZEN check is required, because the FREEZING bit is not cleared
74 * when the state FROZEN is reached.
75 */
76 state = task_freezer(task)->state;
77 ret = (state & CGROUP_FREEZING) && !(state & CGROUP_FROZEN);
78 rcu_read_unlock();
79
80 return ret;
81 }
82
83 static const char *freezer_state_strs(unsigned int state)
84 {
85 if (state & CGROUP_FROZEN)
86 return "FROZEN";
87 if (state & CGROUP_FREEZING)
88 return "FREEZING";
89 return "THAWED";
90 };
91
92 static struct cgroup_subsys_state *
93 freezer_css_alloc(struct cgroup_subsys_state *parent_css)
94 {
95 struct freezer *freezer;
96
97 freezer = kzalloc(sizeof(struct freezer), GFP_KERNEL);
98 if (!freezer)
99 return ERR_PTR(-ENOMEM);
100
101 return &freezer->css;
102 }
103
104 /**
105 * freezer_css_online - commit creation of a freezer css
106 * @css: css being created
107 *
108 * We're committing to creation of @css. Mark it online and inherit
109 * parent's freezing state while holding cpus read lock and freezer_mutex.
110 */
111 static int freezer_css_online(struct cgroup_subsys_state *css)
112 {
113 struct freezer *freezer = css_freezer(css);
114 struct freezer *parent = parent_freezer(freezer);
115
116 cpus_read_lock();
117 mutex_lock(&freezer_mutex);
118
119 freezer->state |= CGROUP_FREEZER_ONLINE;
120
121 if (parent && (parent->state & CGROUP_FREEZING)) {
122 freezer->state |= CGROUP_FREEZING_PARENT | CGROUP_FROZEN;
123 static_branch_inc_cpuslocked(&freezer_active);
124 }
125
126 mutex_unlock(&freezer_mutex);
127 cpus_read_unlock();
128 return 0;
129 }
130
131 /**
132 * freezer_css_offline - initiate destruction of a freezer css
133 * @css: css being destroyed
134 *
135 * @css is going away. Mark it dead and decrement freezer_active if
136 * it was holding one.
137 */
138 static void freezer_css_offline(struct cgroup_subsys_state *css)
139 {
140 struct freezer *freezer = css_freezer(css);
141
142 cpus_read_lock();
143 mutex_lock(&freezer_mutex);
144
145 if (freezer->state & CGROUP_FREEZING)
146 static_branch_dec_cpuslocked(&freezer_active);
147
148 freezer->state = 0;
149
150 mutex_unlock(&freezer_mutex);
151 cpus_read_unlock();
152 }
153
154 static void freezer_css_free(struct cgroup_subsys_state *css)
155 {
156 kfree(css_freezer(css));
157 }
158
159 /*
160 * Tasks can be migrated into a different freezer anytime regardless of its
161 * current state. freezer_attach() is responsible for making new tasks
162 * conform to the current state.
163 *
164 * Freezer state changes and task migration are synchronized via
165 * @freezer->lock. freezer_attach() makes the new tasks conform to the
166 * current state and all following state changes can see the new tasks.
167 */
168 static void freezer_attach(struct cgroup_taskset *tset)
169 {
170 struct task_struct *task;
171 struct cgroup_subsys_state *new_css;
172
173 mutex_lock(&freezer_mutex);
174
175 /*
176 * Make the new tasks conform to the current state of @new_css.
177 * For simplicity, when migrating any task to a FROZEN cgroup, we
178 * revert it to FREEZING and let update_if_frozen() determine the
179 * correct state later.
180 *
181 * Tasks in @tset are on @new_css but may not conform to its
182 * current state before executing the following - !frozen tasks may
183 * be visible in a FROZEN cgroup and frozen tasks in a THAWED one.
184 */
185 cgroup_taskset_for_each(task, new_css, tset) {
186 struct freezer *freezer = css_freezer(new_css);
187
188 if (!(freezer->state & CGROUP_FREEZING)) {
189 __thaw_task(task);
190 } else {
191 freeze_task(task);
192
193 /* clear FROZEN and propagate upwards */
194 while (freezer && (freezer->state & CGROUP_FROZEN)) {
195 freezer->state &= ~CGROUP_FROZEN;
196 freezer = parent_freezer(freezer);
197 }
198 }
199 }
200
201 mutex_unlock(&freezer_mutex);
202 }
203
204 /**
205 * freezer_fork - cgroup post fork callback
206 * @task: a task which has just been forked
207 *
208 * @task has just been created and should conform to the current state of
209 * the cgroup_freezer it belongs to. This function may race against
210 * freezer_attach(). Losing to freezer_attach() means that we don't have
211 * to do anything as freezer_attach() will put @task into the appropriate
212 * state.
213 */
214 static void freezer_fork(struct task_struct *task)
215 {
216 struct freezer *freezer;
217
218 /*
219 * The root cgroup is non-freezable, so we can skip locking the
220 * freezer. This is safe regardless of race with task migration.
221 * If we didn't race or won, skipping is obviously the right thing
222 * to do. If we lost and root is the new cgroup, noop is still the
223 * right thing to do.
224 */
225 if (task_css_is_root(task, freezer_cgrp_id))
226 return;
227
228 mutex_lock(&freezer_mutex);
229 rcu_read_lock();
230
231 freezer = task_freezer(task);
232 if (freezer->state & CGROUP_FREEZING)
233 freeze_task(task);
234
235 rcu_read_unlock();
236 mutex_unlock(&freezer_mutex);
237 }
238
239 /**
240 * update_if_frozen - update whether a cgroup finished freezing
241 * @css: css of interest
242 *
243 * Once FREEZING is initiated, transition to FROZEN is lazily updated by
244 * calling this function. If the current state is FREEZING but not FROZEN,
245 * this function checks whether all tasks of this cgroup and the descendant
246 * cgroups finished freezing and, if so, sets FROZEN.
247 *
248 * The caller is responsible for grabbing RCU read lock and calling
249 * update_if_frozen() on all descendants prior to invoking this function.
250 *
251 * Task states and freezer state might disagree while tasks are being
252 * migrated into or out of @css, so we can't verify task states against
253 * @freezer state here. See freezer_attach() for details.
254 */
255 static void update_if_frozen(struct cgroup_subsys_state *css)
256 {
257 struct freezer *freezer = css_freezer(css);
258 struct cgroup_subsys_state *pos;
259 struct css_task_iter it;
260 struct task_struct *task;
261
262 lockdep_assert_held(&freezer_mutex);
263
264 if (!(freezer->state & CGROUP_FREEZING) ||
265 (freezer->state & CGROUP_FROZEN))
266 return;
267
268 /* are all (live) children frozen? */
269 rcu_read_lock();
270 css_for_each_child(pos, css) {
271 struct freezer *child = css_freezer(pos);
272
273 if ((child->state & CGROUP_FREEZER_ONLINE) &&
274 !(child->state & CGROUP_FROZEN)) {
275 rcu_read_unlock();
276 return;
277 }
278 }
279 rcu_read_unlock();
280
281 /* are all tasks frozen? */
282 css_task_iter_start(css, 0, &it);
283
284 while ((task = css_task_iter_next(&it))) {
285 if (freezing(task) && !frozen(task))
286 goto out_iter_end;
287 }
288
289 freezer->state |= CGROUP_FROZEN;
290 out_iter_end:
291 css_task_iter_end(&it);
292 }
293
294 static int freezer_read(struct seq_file *m, void *v)
295 {
296 struct cgroup_subsys_state *css = seq_css(m), *pos;
297
298 mutex_lock(&freezer_mutex);
299 rcu_read_lock();
300
301 /* update states bottom-up */
302 css_for_each_descendant_post(pos, css) {
303 if (!css_tryget_online(pos))
304 continue;
305 rcu_read_unlock();
306
307 update_if_frozen(pos);
308
309 rcu_read_lock();
310 css_put(pos);
311 }
312
313 rcu_read_unlock();
314 mutex_unlock(&freezer_mutex);
315
316 seq_puts(m, freezer_state_strs(css_freezer(css)->state));
317 seq_putc(m, '\n');
318 return 0;
319 }
320
321 static void freeze_cgroup(struct freezer *freezer)
322 {
323 struct css_task_iter it;
324 struct task_struct *task;
325
326 css_task_iter_start(&freezer->css, 0, &it);
327 while ((task = css_task_iter_next(&it)))
328 freeze_task(task);
329 css_task_iter_end(&it);
330 }
331
332 static void unfreeze_cgroup(struct freezer *freezer)
333 {
334 struct css_task_iter it;
335 struct task_struct *task;
336
337 css_task_iter_start(&freezer->css, 0, &it);
338 while ((task = css_task_iter_next(&it)))
339 __thaw_task(task);
340 css_task_iter_end(&it);
341 }
342
343 /**
344 * freezer_apply_state - apply state change to a single cgroup_freezer
345 * @freezer: freezer to apply state change to
346 * @freeze: whether to freeze or unfreeze
347 * @state: CGROUP_FREEZING_* flag to set or clear
348 *
349 * Set or clear @state on @cgroup according to @freeze, and perform
350 * freezing or thawing as necessary.
351 */
352 static void freezer_apply_state(struct freezer *freezer, bool freeze,
353 unsigned int state)
354 {
355 /* also synchronizes against task migration, see freezer_attach() */
356 lockdep_assert_held(&freezer_mutex);
357
358 if (!(freezer->state & CGROUP_FREEZER_ONLINE))
359 return;
360
361 if (freeze) {
362 if (!(freezer->state & CGROUP_FREEZING))
363 static_branch_inc_cpuslocked(&freezer_active);
364 freezer->state |= state;
365 freeze_cgroup(freezer);
366 } else {
367 bool was_freezing = freezer->state & CGROUP_FREEZING;
368
369 freezer->state &= ~state;
370
371 if (!(freezer->state & CGROUP_FREEZING)) {
372 freezer->state &= ~CGROUP_FROZEN;
373 if (was_freezing)
374 static_branch_dec_cpuslocked(&freezer_active);
375 unfreeze_cgroup(freezer);
376 }
377 }
378 }
379
380 /**
381 * freezer_change_state - change the freezing state of a cgroup_freezer
382 * @freezer: freezer of interest
383 * @freeze: whether to freeze or thaw
384 *
385 * Freeze or thaw @freezer according to @freeze. The operations are
386 * recursive - all descendants of @freezer will be affected.
387 */
388 static void freezer_change_state(struct freezer *freezer, bool freeze)
389 {
390 struct cgroup_subsys_state *pos;
391
392 cpus_read_lock();
393 /*
394 * Update all its descendants in pre-order traversal. Each
395 * descendant will try to inherit its parent's FREEZING state as
396 * CGROUP_FREEZING_PARENT.
397 */
398 mutex_lock(&freezer_mutex);
399 rcu_read_lock();
400 css_for_each_descendant_pre(pos, &freezer->css) {
401 struct freezer *pos_f = css_freezer(pos);
402 struct freezer *parent = parent_freezer(pos_f);
403
404 if (!css_tryget_online(pos))
405 continue;
406 rcu_read_unlock();
407
408 if (pos_f == freezer)
409 freezer_apply_state(pos_f, freeze,
410 CGROUP_FREEZING_SELF);
411 else
412 freezer_apply_state(pos_f,
413 parent->state & CGROUP_FREEZING,
414 CGROUP_FREEZING_PARENT);
415
416 rcu_read_lock();
417 css_put(pos);
418 }
419 rcu_read_unlock();
420 mutex_unlock(&freezer_mutex);
421 cpus_read_unlock();
422 }
423
424 static ssize_t freezer_write(struct kernfs_open_file *of,
425 char *buf, size_t nbytes, loff_t off)
426 {
427 bool freeze;
428
429 buf = strstrip(buf);
430
431 if (strcmp(buf, freezer_state_strs(0)) == 0)
432 freeze = false;
433 else if (strcmp(buf, freezer_state_strs(CGROUP_FROZEN)) == 0)
434 freeze = true;
435 else
436 return -EINVAL;
437
438 freezer_change_state(css_freezer(of_css(of)), freeze);
439 return nbytes;
440 }
441
442 static u64 freezer_self_freezing_read(struct cgroup_subsys_state *css,
443 struct cftype *cft)
444 {
445 struct freezer *freezer = css_freezer(css);
446
447 return (bool)(freezer->state & CGROUP_FREEZING_SELF);
448 }
449
450 static u64 freezer_parent_freezing_read(struct cgroup_subsys_state *css,
451 struct cftype *cft)
452 {
453 struct freezer *freezer = css_freezer(css);
454
455 return (bool)(freezer->state & CGROUP_FREEZING_PARENT);
456 }
457
458 static struct cftype files[] = {
459 {
460 .name = "state",
461 .flags = CFTYPE_NOT_ON_ROOT,
462 .seq_show = freezer_read,
463 .write = freezer_write,
464 },
465 {
466 .name = "self_freezing",
467 .flags = CFTYPE_NOT_ON_ROOT,
468 .read_u64 = freezer_self_freezing_read,
469 },
470 {
471 .name = "parent_freezing",
472 .flags = CFTYPE_NOT_ON_ROOT,
473 .read_u64 = freezer_parent_freezing_read,
474 },
475 { } /* terminate */
476 };
477
478 struct cgroup_subsys freezer_cgrp_subsys = {
479 .css_alloc = freezer_css_alloc,
480 .css_online = freezer_css_online,
481 .css_offline = freezer_css_offline,
482 .css_free = freezer_css_free,
483 .attach = freezer_attach,
484 .fork = freezer_fork,
485 .legacy_cftypes = files,
486 };
487
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.