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 */
34enum 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
44struct freezer {
45 struct cgroup_subsys_state css;
46 unsigned int state;
47};
48
49static DEFINE_MUTEX(freezer_mutex);
50
51static inline struct freezer *css_freezer(struct cgroup_subsys_state *css)
52{
53 return css ? container_of(css, struct freezer, css) : NULL;
54}
55
56static inline struct freezer *task_freezer(struct task_struct *task)
57{
58 return css_freezer(css: task_css(task, subsys_id: freezer_cgrp_id));
59}
60
61static struct freezer *parent_freezer(struct freezer *freezer)
62{
63 return css_freezer(css: freezer->css.parent);
64}
65
66bool cgroup_freezing(struct task_struct *task)
67{
68 bool ret;
69
70 rcu_read_lock();
71 ret = task_freezer(task)->state & CGROUP_FREEZING;
72 rcu_read_unlock();
73
74 return ret;
75}
76
77static const char *freezer_state_strs(unsigned int state)
78{
79 if (state & CGROUP_FROZEN)
80 return "FROZEN";
81 if (state & CGROUP_FREEZING)
82 return "FREEZING";
83 return "THAWED";
84};
85
86static struct cgroup_subsys_state *
87freezer_css_alloc(struct cgroup_subsys_state *parent_css)
88{
89 struct freezer *freezer;
90
91 freezer = kzalloc(sizeof(struct freezer), GFP_KERNEL);
92 if (!freezer)
93 return ERR_PTR(error: -ENOMEM);
94
95 return &freezer->css;
96}
97
98/**
99 * freezer_css_online - commit creation of a freezer css
100 * @css: css being created
101 *
102 * We're committing to creation of @css. Mark it online and inherit
103 * parent's freezing state while holding cpus read lock and freezer_mutex.
104 */
105static 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 cpus_read_lock();
111 mutex_lock(lock: &freezer_mutex);
112
113 freezer->state |= CGROUP_FREEZER_ONLINE;
114
115 if (parent && (parent->state & CGROUP_FREEZING)) {
116 freezer->state |= CGROUP_FREEZING_PARENT | CGROUP_FROZEN;
117 static_branch_inc_cpuslocked(&freezer_active);
118 }
119
120 mutex_unlock(lock: &freezer_mutex);
121 cpus_read_unlock();
122 return 0;
123}
124
125/**
126 * freezer_css_offline - initiate destruction of a freezer css
127 * @css: css being destroyed
128 *
129 * @css is going away. Mark it dead and decrement freezer_active if
130 * it was holding one.
131 */
132static void freezer_css_offline(struct cgroup_subsys_state *css)
133{
134 struct freezer *freezer = css_freezer(css);
135
136 cpus_read_lock();
137 mutex_lock(lock: &freezer_mutex);
138
139 if (freezer->state & CGROUP_FREEZING)
140 static_branch_dec_cpuslocked(&freezer_active);
141
142 freezer->state = 0;
143
144 mutex_unlock(lock: &freezer_mutex);
145 cpus_read_unlock();
146}
147
148static void freezer_css_free(struct cgroup_subsys_state *css)
149{
150 kfree(objp: css_freezer(css));
151}
152
153/*
154 * Tasks can be migrated into a different freezer anytime regardless of its
155 * current state. freezer_attach() is responsible for making new tasks
156 * conform to the current state.
157 *
158 * Freezer state changes and task migration are synchronized via
159 * @freezer->lock. freezer_attach() makes the new tasks conform to the
160 * current state and all following state changes can see the new tasks.
161 */
162static void freezer_attach(struct cgroup_taskset *tset)
163{
164 struct task_struct *task;
165 struct cgroup_subsys_state *new_css;
166
167 mutex_lock(lock: &freezer_mutex);
168
169 /*
170 * Make the new tasks conform to the current state of @new_css.
171 * For simplicity, when migrating any task to a FROZEN cgroup, we
172 * revert it to FREEZING and let update_if_frozen() determine the
173 * correct state later.
174 *
175 * Tasks in @tset are on @new_css but may not conform to its
176 * current state before executing the following - !frozen tasks may
177 * be visible in a FROZEN cgroup and frozen tasks in a THAWED one.
178 */
179 cgroup_taskset_for_each(task, new_css, tset) {
180 struct freezer *freezer = css_freezer(css: new_css);
181
182 if (!(freezer->state & CGROUP_FREEZING)) {
183 __thaw_task(t: task);
184 } else {
185 /* clear FROZEN and propagate upwards */
186 while (freezer && (freezer->state & CGROUP_FROZEN)) {
187 freezer->state &= ~CGROUP_FROZEN;
188 freezer = parent_freezer(freezer);
189 }
190 freeze_task(p: task);
191 }
192 }
193
194 mutex_unlock(lock: &freezer_mutex);
195}
196
197/**
198 * freezer_fork - cgroup post fork callback
199 * @task: a task which has just been forked
200 *
201 * @task has just been created and should conform to the current state of
202 * the cgroup_freezer it belongs to. This function may race against
203 * freezer_attach(). Losing to freezer_attach() means that we don't have
204 * to do anything as freezer_attach() will put @task into the appropriate
205 * state.
206 */
207static void freezer_fork(struct task_struct *task)
208{
209 struct freezer *freezer;
210
211 /*
212 * The root cgroup is non-freezable, so we can skip locking the
213 * freezer. This is safe regardless of race with task migration.
214 * If we didn't race or won, skipping is obviously the right thing
215 * to do. If we lost and root is the new cgroup, noop is still the
216 * right thing to do.
217 */
218 if (task_css_is_root(task, subsys_id: freezer_cgrp_id))
219 return;
220
221 mutex_lock(lock: &freezer_mutex);
222 rcu_read_lock();
223
224 freezer = task_freezer(task);
225 if (freezer->state & CGROUP_FREEZING)
226 freeze_task(p: task);
227
228 rcu_read_unlock();
229 mutex_unlock(lock: &freezer_mutex);
230}
231
232/**
233 * update_if_frozen - update whether a cgroup finished freezing
234 * @css: css of interest
235 *
236 * Once FREEZING is initiated, transition to FROZEN is lazily updated by
237 * calling this function. If the current state is FREEZING but not FROZEN,
238 * this function checks whether all tasks of this cgroup and the descendant
239 * cgroups finished freezing and, if so, sets FROZEN.
240 *
241 * The caller is responsible for grabbing RCU read lock and calling
242 * update_if_frozen() on all descendants prior to invoking this function.
243 *
244 * Task states and freezer state might disagree while tasks are being
245 * migrated into or out of @css, so we can't verify task states against
246 * @freezer state here. See freezer_attach() for details.
247 */
248static void update_if_frozen(struct cgroup_subsys_state *css)
249{
250 struct freezer *freezer = css_freezer(css);
251 struct cgroup_subsys_state *pos;
252 struct css_task_iter it;
253 struct task_struct *task;
254
255 lockdep_assert_held(&freezer_mutex);
256
257 if (!(freezer->state & CGROUP_FREEZING) ||
258 (freezer->state & CGROUP_FROZEN))
259 return;
260
261 /* are all (live) children frozen? */
262 rcu_read_lock();
263 css_for_each_child(pos, css) {
264 struct freezer *child = css_freezer(css: pos);
265
266 if ((child->state & CGROUP_FREEZER_ONLINE) &&
267 !(child->state & CGROUP_FROZEN)) {
268 rcu_read_unlock();
269 return;
270 }
271 }
272 rcu_read_unlock();
273
274 /* are all tasks frozen? */
275 css_task_iter_start(css, flags: 0, it: &it);
276
277 while ((task = css_task_iter_next(it: &it))) {
278 if (freezing(p: task) && !frozen(p: task))
279 goto out_iter_end;
280 }
281
282 freezer->state |= CGROUP_FROZEN;
283out_iter_end:
284 css_task_iter_end(it: &it);
285}
286
287static int freezer_read(struct seq_file *m, void *v)
288{
289 struct cgroup_subsys_state *css = seq_css(seq: m), *pos;
290
291 mutex_lock(lock: &freezer_mutex);
292 rcu_read_lock();
293
294 /* update states bottom-up */
295 css_for_each_descendant_post(pos, css) {
296 if (!css_tryget_online(css: pos))
297 continue;
298 rcu_read_unlock();
299
300 update_if_frozen(css: pos);
301
302 rcu_read_lock();
303 css_put(css: pos);
304 }
305
306 rcu_read_unlock();
307 mutex_unlock(lock: &freezer_mutex);
308
309 seq_puts(m, s: freezer_state_strs(state: css_freezer(css)->state));
310 seq_putc(m, c: '\n');
311 return 0;
312}
313
314static void freeze_cgroup(struct freezer *freezer)
315{
316 struct css_task_iter it;
317 struct task_struct *task;
318
319 css_task_iter_start(css: &freezer->css, flags: 0, it: &it);
320 while ((task = css_task_iter_next(it: &it)))
321 freeze_task(p: task);
322 css_task_iter_end(it: &it);
323}
324
325static void unfreeze_cgroup(struct freezer *freezer)
326{
327 struct css_task_iter it;
328 struct task_struct *task;
329
330 css_task_iter_start(css: &freezer->css, flags: 0, it: &it);
331 while ((task = css_task_iter_next(it: &it)))
332 __thaw_task(t: task);
333 css_task_iter_end(it: &it);
334}
335
336/**
337 * freezer_apply_state - apply state change to a single cgroup_freezer
338 * @freezer: freezer to apply state change to
339 * @freeze: whether to freeze or unfreeze
340 * @state: CGROUP_FREEZING_* flag to set or clear
341 *
342 * Set or clear @state on @cgroup according to @freeze, and perform
343 * freezing or thawing as necessary.
344 */
345static void freezer_apply_state(struct freezer *freezer, bool freeze,
346 unsigned int state)
347{
348 /* also synchronizes against task migration, see freezer_attach() */
349 lockdep_assert_held(&freezer_mutex);
350
351 if (!(freezer->state & CGROUP_FREEZER_ONLINE))
352 return;
353
354 if (freeze) {
355 if (!(freezer->state & CGROUP_FREEZING))
356 static_branch_inc_cpuslocked(&freezer_active);
357 freezer->state |= state;
358 freeze_cgroup(freezer);
359 } else {
360 bool was_freezing = freezer->state & CGROUP_FREEZING;
361
362 freezer->state &= ~state;
363
364 if (!(freezer->state & CGROUP_FREEZING)) {
365 freezer->state &= ~CGROUP_FROZEN;
366 if (was_freezing)
367 static_branch_dec_cpuslocked(&freezer_active);
368 unfreeze_cgroup(freezer);
369 }
370 }
371}
372
373/**
374 * freezer_change_state - change the freezing state of a cgroup_freezer
375 * @freezer: freezer of interest
376 * @freeze: whether to freeze or thaw
377 *
378 * Freeze or thaw @freezer according to @freeze. The operations are
379 * recursive - all descendants of @freezer will be affected.
380 */
381static void freezer_change_state(struct freezer *freezer, bool freeze)
382{
383 struct cgroup_subsys_state *pos;
384
385 cpus_read_lock();
386 /*
387 * Update all its descendants in pre-order traversal. Each
388 * descendant will try to inherit its parent's FREEZING state as
389 * CGROUP_FREEZING_PARENT.
390 */
391 mutex_lock(lock: &freezer_mutex);
392 rcu_read_lock();
393 css_for_each_descendant_pre(pos, &freezer->css) {
394 struct freezer *pos_f = css_freezer(css: pos);
395 struct freezer *parent = parent_freezer(freezer: pos_f);
396
397 if (!css_tryget_online(css: pos))
398 continue;
399 rcu_read_unlock();
400
401 if (pos_f == freezer)
402 freezer_apply_state(freezer: pos_f, freeze,
403 state: CGROUP_FREEZING_SELF);
404 else
405 freezer_apply_state(freezer: pos_f,
406 freeze: parent->state & CGROUP_FREEZING,
407 state: CGROUP_FREEZING_PARENT);
408
409 rcu_read_lock();
410 css_put(css: pos);
411 }
412 rcu_read_unlock();
413 mutex_unlock(lock: &freezer_mutex);
414 cpus_read_unlock();
415}
416
417static ssize_t freezer_write(struct kernfs_open_file *of,
418 char *buf, size_t nbytes, loff_t off)
419{
420 bool freeze;
421
422 buf = strstrip(str: buf);
423
424 if (strcmp(buf, freezer_state_strs(state: 0)) == 0)
425 freeze = false;
426 else if (strcmp(buf, freezer_state_strs(state: CGROUP_FROZEN)) == 0) {
427 pr_info_once("Freezing with imperfect legacy cgroup freezer. "
428 "See cgroup.freeze of cgroup v2\n");
429 freeze = true;
430 } else
431 return -EINVAL;
432
433 freezer_change_state(freezer: css_freezer(css: of_css(of)), freeze);
434 return nbytes;
435}
436
437static 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
445static 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
453static 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
473struct 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