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 */
33enum 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
43struct freezer {
44	struct cgroup_subsys_state	css;
45	unsigned int			state;
46};
47
48static DEFINE_MUTEX(freezer_mutex);
49
50static inline struct freezer *css_freezer(struct cgroup_subsys_state *css)
51{
52	return css ? container_of(css, struct freezer, css) : NULL;
53}
54
55static inline struct freezer *task_freezer(struct task_struct *task)
56{
57	return css_freezer(task_css(task, freezer_cgrp_id));
58}
59
60static struct freezer *parent_freezer(struct freezer *freezer)
61{
62	return css_freezer(freezer->css.parent);
63}
64
65bool 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
76static 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
85static struct cgroup_subsys_state *
86freezer_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 */
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	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 */
130static 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
144static 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 */
158static void freezer_attach(struct cgroup_subsys_state *new_css,
159			   struct cgroup_taskset *tset)
160{
161	struct freezer *freezer = css_freezer(new_css);
162	struct task_struct *task;
163	bool clear_frozen = false;
164
165	mutex_lock(&freezer_mutex);
166
167	/*
168	 * Make the new tasks conform to the current state of @new_css.
169	 * For simplicity, when migrating any task to a FROZEN cgroup, we
170	 * revert it to FREEZING and let update_if_frozen() determine the
171	 * correct state later.
172	 *
173	 * Tasks in @tset are on @new_css but may not conform to its
174	 * current state before executing the following - !frozen tasks may
175	 * be visible in a FROZEN cgroup and frozen tasks in a THAWED one.
176	 */
177	cgroup_taskset_for_each(task, tset) {
178		if (!(freezer->state & CGROUP_FREEZING)) {
179			__thaw_task(task);
180		} else {
181			freeze_task(task);
182			freezer->state &= ~CGROUP_FROZEN;
183			clear_frozen = true;
184		}
185	}
186
187	/* propagate FROZEN clearing upwards */
188	while (clear_frozen && (freezer = parent_freezer(freezer))) {
189		freezer->state &= ~CGROUP_FROZEN;
190		clear_frozen = freezer->state & CGROUP_FREEZING;
191	}
192
193	mutex_unlock(&freezer_mutex);
194}
195
196/**
197 * freezer_fork - cgroup post fork callback
198 * @task: a task which has just been forked
199 *
200 * @task has just been created and should conform to the current state of
201 * the cgroup_freezer it belongs to.  This function may race against
202 * freezer_attach().  Losing to freezer_attach() means that we don't have
203 * to do anything as freezer_attach() will put @task into the appropriate
204 * state.
205 */
206static void freezer_fork(struct task_struct *task)
207{
208	struct freezer *freezer;
209
210	/*
211	 * The root cgroup is non-freezable, so we can skip locking the
212	 * freezer.  This is safe regardless of race with task migration.
213	 * If we didn't race or won, skipping is obviously the right thing
214	 * to do.  If we lost and root is the new cgroup, noop is still the
215	 * right thing to do.
216	 */
217	if (task_css_is_root(task, freezer_cgrp_id))
218		return;
219
220	mutex_lock(&freezer_mutex);
221	rcu_read_lock();
222
223	freezer = task_freezer(task);
224	if (freezer->state & CGROUP_FREEZING)
225		freeze_task(task);
226
227	rcu_read_unlock();
228	mutex_unlock(&freezer_mutex);
229}
230
231/**
232 * update_if_frozen - update whether a cgroup finished freezing
233 * @css: css of interest
234 *
235 * Once FREEZING is initiated, transition to FROZEN is lazily updated by
236 * calling this function.  If the current state is FREEZING but not FROZEN,
237 * this function checks whether all tasks of this cgroup and the descendant
238 * cgroups finished freezing and, if so, sets FROZEN.
239 *
240 * The caller is responsible for grabbing RCU read lock and calling
241 * update_if_frozen() on all descendants prior to invoking this function.
242 *
243 * Task states and freezer state might disagree while tasks are being
244 * migrated into or out of @css, so we can't verify task states against
245 * @freezer state here.  See freezer_attach() for details.
246 */
247static void update_if_frozen(struct cgroup_subsys_state *css)
248{
249	struct freezer *freezer = css_freezer(css);
250	struct cgroup_subsys_state *pos;
251	struct css_task_iter it;
252	struct task_struct *task;
253
254	lockdep_assert_held(&freezer_mutex);
255
256	if (!(freezer->state & CGROUP_FREEZING) ||
257	    (freezer->state & CGROUP_FROZEN))
258		return;
259
260	/* are all (live) children frozen? */
261	rcu_read_lock();
262	css_for_each_child(pos, css) {
263		struct freezer *child = css_freezer(pos);
264
265		if ((child->state & CGROUP_FREEZER_ONLINE) &&
266		    !(child->state & CGROUP_FROZEN)) {
267			rcu_read_unlock();
268			return;
269		}
270	}
271	rcu_read_unlock();
272
273	/* are all tasks frozen? */
274	css_task_iter_start(css, &it);
275
276	while ((task = css_task_iter_next(&it))) {
277		if (freezing(task)) {
278			/*
279			 * freezer_should_skip() indicates that the task
280			 * should be skipped when determining freezing
281			 * completion.  Consider it frozen in addition to
282			 * the usual frozen condition.
283			 */
284			if (!frozen(task) && !freezer_should_skip(task))
285				goto out_iter_end;
286		}
287	}
288
289	freezer->state |= CGROUP_FROZEN;
290out_iter_end:
291	css_task_iter_end(&it);
292}
293
294static 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
321static 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, &it);
327	while ((task = css_task_iter_next(&it)))
328		freeze_task(task);
329	css_task_iter_end(&it);
330}
331
332static 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, &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 */
352static 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			atomic_inc(&system_freezing_cnt);
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			if (was_freezing)
373				atomic_dec(&system_freezing_cnt);
374			freezer->state &= ~CGROUP_FROZEN;
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 */
388static void freezer_change_state(struct freezer *freezer, bool freeze)
389{
390	struct cgroup_subsys_state *pos;
391
392	/*
393	 * Update all its descendants in pre-order traversal.  Each
394	 * descendant will try to inherit its parent's FREEZING state as
395	 * CGROUP_FREEZING_PARENT.
396	 */
397	mutex_lock(&freezer_mutex);
398	rcu_read_lock();
399	css_for_each_descendant_pre(pos, &freezer->css) {
400		struct freezer *pos_f = css_freezer(pos);
401		struct freezer *parent = parent_freezer(pos_f);
402
403		if (!css_tryget_online(pos))
404			continue;
405		rcu_read_unlock();
406
407		if (pos_f == freezer)
408			freezer_apply_state(pos_f, freeze,
409					    CGROUP_FREEZING_SELF);
410		else
411			freezer_apply_state(pos_f,
412					    parent->state & CGROUP_FREEZING,
413					    CGROUP_FREEZING_PARENT);
414
415		rcu_read_lock();
416		css_put(pos);
417	}
418	rcu_read_unlock();
419	mutex_unlock(&freezer_mutex);
420}
421
422static ssize_t freezer_write(struct kernfs_open_file *of,
423			     char *buf, size_t nbytes, loff_t off)
424{
425	bool freeze;
426
427	buf = strstrip(buf);
428
429	if (strcmp(buf, freezer_state_strs(0)) == 0)
430		freeze = false;
431	else if (strcmp(buf, freezer_state_strs(CGROUP_FROZEN)) == 0)
432		freeze = true;
433	else
434		return -EINVAL;
435
436	freezer_change_state(css_freezer(of_css(of)), freeze);
437	return nbytes;
438}
439
440static u64 freezer_self_freezing_read(struct cgroup_subsys_state *css,
441				      struct cftype *cft)
442{
443	struct freezer *freezer = css_freezer(css);
444
445	return (bool)(freezer->state & CGROUP_FREEZING_SELF);
446}
447
448static u64 freezer_parent_freezing_read(struct cgroup_subsys_state *css,
449					struct cftype *cft)
450{
451	struct freezer *freezer = css_freezer(css);
452
453	return (bool)(freezer->state & CGROUP_FREEZING_PARENT);
454}
455
456static struct cftype files[] = {
457	{
458		.name = "state",
459		.flags = CFTYPE_NOT_ON_ROOT,
460		.seq_show = freezer_read,
461		.write = freezer_write,
462	},
463	{
464		.name = "self_freezing",
465		.flags = CFTYPE_NOT_ON_ROOT,
466		.read_u64 = freezer_self_freezing_read,
467	},
468	{
469		.name = "parent_freezing",
470		.flags = CFTYPE_NOT_ON_ROOT,
471		.read_u64 = freezer_parent_freezing_read,
472	},
473	{ }	/* terminate */
474};
475
476struct cgroup_subsys freezer_cgrp_subsys = {
477	.css_alloc	= freezer_css_alloc,
478	.css_online	= freezer_css_online,
479	.css_offline	= freezer_css_offline,
480	.css_free	= freezer_css_free,
481	.attach		= freezer_attach,
482	.fork		= freezer_fork,
483	.legacy_cftypes	= files,
484};
485