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(css_parent(&freezer->css));
 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
 76/*
 77 * cgroups_write_string() limits the size of freezer state strings to
 78 * CGROUP_LOCAL_BUFFER_SIZE
 79 */
 80static const char *freezer_state_strs(unsigned int state)
 81{
 82	if (state & CGROUP_FROZEN)
 83		return "FROZEN";
 84	if (state & CGROUP_FREEZING)
 85		return "FREEZING";
 86	return "THAWED";
 87};
 88
 89static struct cgroup_subsys_state *
 90freezer_css_alloc(struct cgroup_subsys_state *parent_css)
 91{
 92	struct freezer *freezer;
 93
 94	freezer = kzalloc(sizeof(struct freezer), GFP_KERNEL);
 95	if (!freezer)
 96		return ERR_PTR(-ENOMEM);
 97
 98	return &freezer->css;
 99}
100
101/**
102 * freezer_css_online - commit creation of a freezer css
103 * @css: css being created
104 *
105 * We're committing to creation of @css.  Mark it online and inherit
106 * parent's freezing state while holding both parent's and our
107 * freezer->lock.
108 */
109static int freezer_css_online(struct cgroup_subsys_state *css)
110{
111	struct freezer *freezer = css_freezer(css);
112	struct freezer *parent = parent_freezer(freezer);
113
114	mutex_lock(&freezer_mutex);
115
116	freezer->state |= CGROUP_FREEZER_ONLINE;
117
118	if (parent && (parent->state & CGROUP_FREEZING)) {
119		freezer->state |= CGROUP_FREEZING_PARENT | CGROUP_FROZEN;
120		atomic_inc(&system_freezing_cnt);
121	}
122
123	mutex_unlock(&freezer_mutex);
124	return 0;
125}
126
127/**
128 * freezer_css_offline - initiate destruction of a freezer css
129 * @css: css being destroyed
130 *
131 * @css is going away.  Mark it dead and decrement system_freezing_count if
132 * it was holding one.
133 */
134static void freezer_css_offline(struct cgroup_subsys_state *css)
135{
136	struct freezer *freezer = css_freezer(css);
137
138	mutex_lock(&freezer_mutex);
139
140	if (freezer->state & CGROUP_FREEZING)
141		atomic_dec(&system_freezing_cnt);
142
143	freezer->state = 0;
144
145	mutex_unlock(&freezer_mutex);
146}
147
148static void freezer_css_free(struct cgroup_subsys_state *css)
149{
150	kfree(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_subsys_state *new_css,
163			   struct cgroup_taskset *tset)
164{
165	struct freezer *freezer = css_freezer(new_css);
166	struct task_struct *task;
167	bool clear_frozen = false;
168
169	mutex_lock(&freezer_mutex);
170
171	/*
172	 * Make the new tasks conform to the current state of @new_css.
173	 * For simplicity, when migrating any task to a FROZEN cgroup, we
174	 * revert it to FREEZING and let update_if_frozen() determine the
175	 * correct state later.
176	 *
177	 * Tasks in @tset are on @new_css but may not conform to its
178	 * current state before executing the following - !frozen tasks may
179	 * be visible in a FROZEN cgroup and frozen tasks in a THAWED one.
180	 */
181	cgroup_taskset_for_each(task, tset) {
182		if (!(freezer->state & CGROUP_FREEZING)) {
183			__thaw_task(task);
184		} else {
185			freeze_task(task);
186			freezer->state &= ~CGROUP_FROZEN;
187			clear_frozen = true;
188		}
189	}
190
191	/* propagate FROZEN clearing upwards */
192	while (clear_frozen && (freezer = parent_freezer(freezer))) {
193		freezer->state &= ~CGROUP_FROZEN;
194		clear_frozen = freezer->state & CGROUP_FREEZING;
195	}
196
197	mutex_unlock(&freezer_mutex);
198}
199
200/**
201 * freezer_fork - cgroup post fork callback
202 * @task: a task which has just been forked
203 *
204 * @task has just been created and should conform to the current state of
205 * the cgroup_freezer it belongs to.  This function may race against
206 * freezer_attach().  Losing to freezer_attach() means that we don't have
207 * to do anything as freezer_attach() will put @task into the appropriate
208 * state.
209 */
210static void freezer_fork(struct task_struct *task)
211{
212	struct freezer *freezer;
213
214	/*
215	 * The root cgroup is non-freezable, so we can skip locking the
216	 * freezer.  This is safe regardless of race with task migration.
217	 * If we didn't race or won, skipping is obviously the right thing
218	 * to do.  If we lost and root is the new cgroup, noop is still the
219	 * right thing to do.
220	 */
221	if (task_css_is_root(task, freezer_cgrp_id))
222		return;
223
224	mutex_lock(&freezer_mutex);
225	rcu_read_lock();
226
227	freezer = task_freezer(task);
228	if (freezer->state & CGROUP_FREEZING)
229		freeze_task(task);
230
231	rcu_read_unlock();
232	mutex_unlock(&freezer_mutex);
233}
234
235/**
236 * update_if_frozen - update whether a cgroup finished freezing
237 * @css: css of interest
238 *
239 * Once FREEZING is initiated, transition to FROZEN is lazily updated by
240 * calling this function.  If the current state is FREEZING but not FROZEN,
241 * this function checks whether all tasks of this cgroup and the descendant
242 * cgroups finished freezing and, if so, sets FROZEN.
243 *
244 * The caller is responsible for grabbing RCU read lock and calling
245 * update_if_frozen() on all descendants prior to invoking this function.
246 *
247 * Task states and freezer state might disagree while tasks are being
248 * migrated into or out of @css, so we can't verify task states against
249 * @freezer state here.  See freezer_attach() for details.
250 */
251static void update_if_frozen(struct cgroup_subsys_state *css)
252{
253	struct freezer *freezer = css_freezer(css);
254	struct cgroup_subsys_state *pos;
255	struct css_task_iter it;
256	struct task_struct *task;
257
258	lockdep_assert_held(&freezer_mutex);
259
260	if (!(freezer->state & CGROUP_FREEZING) ||
261	    (freezer->state & CGROUP_FROZEN))
262		return;
263
264	/* are all (live) children frozen? */
265	rcu_read_lock();
266	css_for_each_child(pos, css) {
267		struct freezer *child = css_freezer(pos);
268
269		if ((child->state & CGROUP_FREEZER_ONLINE) &&
270		    !(child->state & CGROUP_FROZEN)) {
271			rcu_read_unlock();
272			return;
273		}
274	}
275	rcu_read_unlock();
276
277	/* are all tasks frozen? */
278	css_task_iter_start(css, &it);
279
280	while ((task = css_task_iter_next(&it))) {
281		if (freezing(task)) {
282			/*
283			 * freezer_should_skip() indicates that the task
284			 * should be skipped when determining freezing
285			 * completion.  Consider it frozen in addition to
286			 * the usual frozen condition.
287			 */
288			if (!frozen(task) && !freezer_should_skip(task))
289				goto out_iter_end;
290		}
291	}
292
293	freezer->state |= CGROUP_FROZEN;
294out_iter_end:
295	css_task_iter_end(&it);
296}
297
298static int freezer_read(struct seq_file *m, void *v)
299{
300	struct cgroup_subsys_state *css = seq_css(m), *pos;
301
302	mutex_lock(&freezer_mutex);
303	rcu_read_lock();
304
305	/* update states bottom-up */
306	css_for_each_descendant_post(pos, css) {
307		if (!css_tryget(pos))
308			continue;
309		rcu_read_unlock();
310
311		update_if_frozen(pos);
312
313		rcu_read_lock();
314		css_put(pos);
315	}
316
317	rcu_read_unlock();
318	mutex_unlock(&freezer_mutex);
319
320	seq_puts(m, freezer_state_strs(css_freezer(css)->state));
321	seq_putc(m, '\n');
322	return 0;
323}
324
325static void freeze_cgroup(struct freezer *freezer)
326{
327	struct css_task_iter it;
328	struct task_struct *task;
329
330	css_task_iter_start(&freezer->css, &it);
331	while ((task = css_task_iter_next(&it)))
332		freeze_task(task);
333	css_task_iter_end(&it);
334}
335
336static void unfreeze_cgroup(struct freezer *freezer)
337{
338	struct css_task_iter it;
339	struct task_struct *task;
340
341	css_task_iter_start(&freezer->css, &it);
342	while ((task = css_task_iter_next(&it)))
343		__thaw_task(task);
344	css_task_iter_end(&it);
345}
346
347/**
348 * freezer_apply_state - apply state change to a single cgroup_freezer
349 * @freezer: freezer to apply state change to
350 * @freeze: whether to freeze or unfreeze
351 * @state: CGROUP_FREEZING_* flag to set or clear
352 *
353 * Set or clear @state on @cgroup according to @freeze, and perform
354 * freezing or thawing as necessary.
355 */
356static void freezer_apply_state(struct freezer *freezer, bool freeze,
357				unsigned int state)
358{
359	/* also synchronizes against task migration, see freezer_attach() */
360	lockdep_assert_held(&freezer_mutex);
361
362	if (!(freezer->state & CGROUP_FREEZER_ONLINE))
363		return;
364
365	if (freeze) {
366		if (!(freezer->state & CGROUP_FREEZING))
367			atomic_inc(&system_freezing_cnt);
368		freezer->state |= state;
369		freeze_cgroup(freezer);
370	} else {
371		bool was_freezing = freezer->state & CGROUP_FREEZING;
372
373		freezer->state &= ~state;
374
375		if (!(freezer->state & CGROUP_FREEZING)) {
376			if (was_freezing)
377				atomic_dec(&system_freezing_cnt);
378			freezer->state &= ~CGROUP_FROZEN;
379			unfreeze_cgroup(freezer);
380		}
381	}
382}
383
384/**
385 * freezer_change_state - change the freezing state of a cgroup_freezer
386 * @freezer: freezer of interest
387 * @freeze: whether to freeze or thaw
388 *
389 * Freeze or thaw @freezer according to @freeze.  The operations are
390 * recursive - all descendants of @freezer will be affected.
391 */
392static void freezer_change_state(struct freezer *freezer, bool freeze)
393{
394	struct cgroup_subsys_state *pos;
395
396	/*
397	 * Update all its descendants in pre-order traversal.  Each
398	 * descendant will try to inherit its parent's FREEZING state as
399	 * CGROUP_FREEZING_PARENT.
400	 */
401	mutex_lock(&freezer_mutex);
402	rcu_read_lock();
403	css_for_each_descendant_pre(pos, &freezer->css) {
404		struct freezer *pos_f = css_freezer(pos);
405		struct freezer *parent = parent_freezer(pos_f);
406
407		if (!css_tryget(pos))
408			continue;
409		rcu_read_unlock();
410
411		if (pos_f == freezer)
412			freezer_apply_state(pos_f, freeze,
413					    CGROUP_FREEZING_SELF);
414		else
415			freezer_apply_state(pos_f,
416					    parent->state & CGROUP_FREEZING,
417					    CGROUP_FREEZING_PARENT);
418
419		rcu_read_lock();
420		css_put(pos);
421	}
422	rcu_read_unlock();
423	mutex_unlock(&freezer_mutex);
424}
425
426static int freezer_write(struct cgroup_subsys_state *css, struct cftype *cft,
427			 char *buffer)
428{
429	bool freeze;
430
431	if (strcmp(buffer, freezer_state_strs(0)) == 0)
432		freeze = false;
433	else if (strcmp(buffer, freezer_state_strs(CGROUP_FROZEN)) == 0)
434		freeze = true;
435	else
436		return -EINVAL;
437
438	freezer_change_state(css_freezer(css), freeze);
439	return 0;
440}
441
442static 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
450static 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
458static struct cftype files[] = {
459	{
460		.name = "state",
461		.flags = CFTYPE_NOT_ON_ROOT,
462		.seq_show = freezer_read,
463		.write_string = 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
478struct 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	.base_cftypes	= files,
486};