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/module.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
25enum freezer_state {
26	CGROUP_THAWED = 0,
27	CGROUP_FREEZING,
28	CGROUP_FROZEN,
29};
30
31struct freezer {
32	struct cgroup_subsys_state css;
33	enum freezer_state state;
34	spinlock_t lock; /* protects _writes_ to state */
35};
36
37static inline struct freezer *cgroup_freezer(
38		struct cgroup *cgroup)
39{
40	return container_of(
41		cgroup_subsys_state(cgroup, freezer_subsys_id),
42		struct freezer, css);
43}
44
45static inline struct freezer *task_freezer(struct task_struct *task)
46{
47	return container_of(task_subsys_state(task, freezer_subsys_id),
48			    struct freezer, css);
49}
50
51int cgroup_freezing_or_frozen(struct task_struct *task)
52{
53	struct freezer *freezer;
54	enum freezer_state state;
55
56	task_lock(task);
57	freezer = task_freezer(task);
58	if (!freezer->css.cgroup->parent)
59		state = CGROUP_THAWED; /* root cgroup can't be frozen */
60	else
61		state = freezer->state;
62	task_unlock(task);
63
64	return (state == CGROUP_FREEZING) || (state == CGROUP_FROZEN);
65}
66
67/*
68 * cgroups_write_string() limits the size of freezer state strings to
69 * CGROUP_LOCAL_BUFFER_SIZE
70 */
71static const char *freezer_state_strs[] = {
72	"THAWED",
73	"FREEZING",
74	"FROZEN",
75};
76
77/*
78 * State diagram
79 * Transitions are caused by userspace writes to the freezer.state file.
80 * The values in parenthesis are state labels. The rest are edge labels.
81 *
82 * (THAWED) --FROZEN--> (FREEZING) --FROZEN--> (FROZEN)
83 *    ^ ^                    |                     |
84 *    | \_______THAWED_______/                     |
85 *    \__________________________THAWED____________/
86 */
87
88struct cgroup_subsys freezer_subsys;
89
90/* Locks taken and their ordering
91 * ------------------------------
92 * cgroup_mutex (AKA cgroup_lock)
93 * freezer->lock
94 * css_set_lock
95 * task->alloc_lock (AKA task_lock)
96 * task->sighand->siglock
97 *
98 * cgroup code forces css_set_lock to be taken before task->alloc_lock
99 *
100 * freezer_create(), freezer_destroy():
101 * cgroup_mutex [ by cgroup core ]
102 *
103 * freezer_can_attach():
104 * cgroup_mutex (held by caller of can_attach)
105 *
106 * cgroup_freezing_or_frozen():
107 * task->alloc_lock (to get task's cgroup)
108 *
109 * freezer_fork() (preserving fork() performance means can't take cgroup_mutex):
110 * freezer->lock
111 *  sighand->siglock (if the cgroup is freezing)
112 *
113 * freezer_read():
114 * cgroup_mutex
115 *  freezer->lock
116 *   write_lock css_set_lock (cgroup iterator start)
117 *    task->alloc_lock
118 *   read_lock css_set_lock (cgroup iterator start)
119 *
120 * freezer_write() (freeze):
121 * cgroup_mutex
122 *  freezer->lock
123 *   write_lock css_set_lock (cgroup iterator start)
124 *    task->alloc_lock
125 *   read_lock css_set_lock (cgroup iterator start)
126 *    sighand->siglock (fake signal delivery inside freeze_task())
127 *
128 * freezer_write() (unfreeze):
129 * cgroup_mutex
130 *  freezer->lock
131 *   write_lock css_set_lock (cgroup iterator start)
132 *    task->alloc_lock
133 *   read_lock css_set_lock (cgroup iterator start)
134 *    task->alloc_lock (inside thaw_process(), prevents race with refrigerator())
135 *     sighand->siglock
136 */
137static struct cgroup_subsys_state *freezer_create(struct cgroup_subsys *ss,
138						  struct cgroup *cgroup)
139{
140	struct freezer *freezer;
141
142	freezer = kzalloc(sizeof(struct freezer), GFP_KERNEL);
143	if (!freezer)
144		return ERR_PTR(-ENOMEM);
145
146	spin_lock_init(&freezer->lock);
147	freezer->state = CGROUP_THAWED;
148	return &freezer->css;
149}
150
151static void freezer_destroy(struct cgroup_subsys *ss,
152			    struct cgroup *cgroup)
153{
154	kfree(cgroup_freezer(cgroup));
155}
156
157/* Task is frozen or will freeze immediately when next it gets woken */
158static bool is_task_frozen_enough(struct task_struct *task)
159{
160	return frozen(task) ||
161		(task_is_stopped_or_traced(task) && freezing(task));
162}
163
164/*
165 * The call to cgroup_lock() in the freezer.state write method prevents
166 * a write to that file racing against an attach, and hence the
167 * can_attach() result will remain valid until the attach completes.
168 */
169static int freezer_can_attach(struct cgroup_subsys *ss,
170			      struct cgroup *new_cgroup,
171			      struct task_struct *task, bool threadgroup)
172{
173	struct freezer *freezer;
174
175	/*
176	 * Anything frozen can't move or be moved to/from.
177	 *
178	 * Since orig_freezer->state == FROZEN means that @task has been
179	 * frozen, so it's sufficient to check the latter condition.
180	 */
181
182	if (is_task_frozen_enough(task))
183		return -EBUSY;
184
185	freezer = cgroup_freezer(new_cgroup);
186	if (freezer->state == CGROUP_FROZEN)
187		return -EBUSY;
188
189	if (threadgroup) {
190		struct task_struct *c;
191
192		rcu_read_lock();
193		list_for_each_entry_rcu(c, &task->thread_group, thread_group) {
194			if (is_task_frozen_enough(c)) {
195				rcu_read_unlock();
196				return -EBUSY;
197			}
198		}
199		rcu_read_unlock();
200	}
201
202	return 0;
203}
204
205static void freezer_fork(struct cgroup_subsys *ss, struct task_struct *task)
206{
207	struct freezer *freezer;
208
209	/*
210	 * No lock is needed, since the task isn't on tasklist yet,
211	 * so it can't be moved to another cgroup, which means the
212	 * freezer won't be removed and will be valid during this
213	 * function call.  Nevertheless, apply RCU read-side critical
214	 * section to suppress RCU lockdep false positives.
215	 */
216	rcu_read_lock();
217	freezer = task_freezer(task);
218	rcu_read_unlock();
219
220	/*
221	 * The root cgroup is non-freezable, so we can skip the
222	 * following check.
223	 */
224	if (!freezer->css.cgroup->parent)
225		return;
226
227	spin_lock_irq(&freezer->lock);
228	BUG_ON(freezer->state == CGROUP_FROZEN);
229
230	/* Locking avoids race with FREEZING -> THAWED transitions. */
231	if (freezer->state == CGROUP_FREEZING)
232		freeze_task(task, true);
233	spin_unlock_irq(&freezer->lock);
234}
235
236/*
237 * caller must hold freezer->lock
238 */
239static void update_freezer_state(struct cgroup *cgroup,
240				 struct freezer *freezer)
241{
242	struct cgroup_iter it;
243	struct task_struct *task;
244	unsigned int nfrozen = 0, ntotal = 0;
245
246	cgroup_iter_start(cgroup, &it);
247	while ((task = cgroup_iter_next(cgroup, &it))) {
248		ntotal++;
249		if (is_task_frozen_enough(task))
250			nfrozen++;
251	}
252
253	/*
254	 * Transition to FROZEN when no new tasks can be added ensures
255	 * that we never exist in the FROZEN state while there are unfrozen
256	 * tasks.
257	 */
258	if (nfrozen == ntotal)
259		freezer->state = CGROUP_FROZEN;
260	else if (nfrozen > 0)
261		freezer->state = CGROUP_FREEZING;
262	else
263		freezer->state = CGROUP_THAWED;
264	cgroup_iter_end(cgroup, &it);
265}
266
267static int freezer_read(struct cgroup *cgroup, struct cftype *cft,
268			struct seq_file *m)
269{
270	struct freezer *freezer;
271	enum freezer_state state;
272
273	if (!cgroup_lock_live_group(cgroup))
274		return -ENODEV;
275
276	freezer = cgroup_freezer(cgroup);
277	spin_lock_irq(&freezer->lock);
278	state = freezer->state;
279	if (state == CGROUP_FREEZING) {
280		/* We change from FREEZING to FROZEN lazily if the cgroup was
281		 * only partially frozen when we exitted write. */
282		update_freezer_state(cgroup, freezer);
283		state = freezer->state;
284	}
285	spin_unlock_irq(&freezer->lock);
286	cgroup_unlock();
287
288	seq_puts(m, freezer_state_strs[state]);
289	seq_putc(m, '\n');
290	return 0;
291}
292
293static int try_to_freeze_cgroup(struct cgroup *cgroup, struct freezer *freezer)
294{
295	struct cgroup_iter it;
296	struct task_struct *task;
297	unsigned int num_cant_freeze_now = 0;
298
299	freezer->state = CGROUP_FREEZING;
300	cgroup_iter_start(cgroup, &it);
301	while ((task = cgroup_iter_next(cgroup, &it))) {
302		if (!freeze_task(task, true))
303			continue;
304		if (is_task_frozen_enough(task))
305			continue;
306		if (!freezing(task) && !freezer_should_skip(task))
307			num_cant_freeze_now++;
308	}
309	cgroup_iter_end(cgroup, &it);
310
311	return num_cant_freeze_now ? -EBUSY : 0;
312}
313
314static void unfreeze_cgroup(struct cgroup *cgroup, struct freezer *freezer)
315{
316	struct cgroup_iter it;
317	struct task_struct *task;
318
319	cgroup_iter_start(cgroup, &it);
320	while ((task = cgroup_iter_next(cgroup, &it))) {
321		thaw_process(task);
322	}
323	cgroup_iter_end(cgroup, &it);
324
325	freezer->state = CGROUP_THAWED;
326}
327
328static int freezer_change_state(struct cgroup *cgroup,
329				enum freezer_state goal_state)
330{
331	struct freezer *freezer;
332	int retval = 0;
333
334	freezer = cgroup_freezer(cgroup);
335
336	spin_lock_irq(&freezer->lock);
337
338	update_freezer_state(cgroup, freezer);
339	if (goal_state == freezer->state)
340		goto out;
341
342	switch (goal_state) {
343	case CGROUP_THAWED:
344		unfreeze_cgroup(cgroup, freezer);
345		break;
346	case CGROUP_FROZEN:
347		retval = try_to_freeze_cgroup(cgroup, freezer);
348		break;
349	default:
350		BUG();
351	}
352out:
353	spin_unlock_irq(&freezer->lock);
354
355	return retval;
356}
357
358static int freezer_write(struct cgroup *cgroup,
359			 struct cftype *cft,
360			 const char *buffer)
361{
362	int retval;
363	enum freezer_state goal_state;
364
365	if (strcmp(buffer, freezer_state_strs[CGROUP_THAWED]) == 0)
366		goal_state = CGROUP_THAWED;
367	else if (strcmp(buffer, freezer_state_strs[CGROUP_FROZEN]) == 0)
368		goal_state = CGROUP_FROZEN;
369	else
370		return -EINVAL;
371
372	if (!cgroup_lock_live_group(cgroup))
373		return -ENODEV;
374	retval = freezer_change_state(cgroup, goal_state);
375	cgroup_unlock();
376	return retval;
377}
378
379static struct cftype files[] = {
380	{
381		.name = "state",
382		.read_seq_string = freezer_read,
383		.write_string = freezer_write,
384	},
385};
386
387static int freezer_populate(struct cgroup_subsys *ss, struct cgroup *cgroup)
388{
389	if (!cgroup->parent)
390		return 0;
391	return cgroup_add_files(cgroup, ss, files, ARRAY_SIZE(files));
392}
393
394struct cgroup_subsys freezer_subsys = {
395	.name		= "freezer",
396	.create		= freezer_create,
397	.destroy	= freezer_destroy,
398	.populate	= freezer_populate,
399	.subsys_id	= freezer_subsys_id,
400	.can_attach	= freezer_can_attach,
401	.attach		= NULL,
402	.fork		= freezer_fork,
403	.exit		= NULL,
404};
405