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		static_branch_inc(&freezer_active);
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		static_branch_dec(&freezer_active);
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_taskset *tset)
159{
160	struct task_struct *task;
161	struct cgroup_subsys_state *new_css;
162
163	mutex_lock(&freezer_mutex);
164
165	/*
166	 * Make the new tasks conform to the current state of @new_css.
167	 * For simplicity, when migrating any task to a FROZEN cgroup, we
168	 * revert it to FREEZING and let update_if_frozen() determine the
169	 * correct state later.
170	 *
171	 * Tasks in @tset are on @new_css but may not conform to its
172	 * current state before executing the following - !frozen tasks may
173	 * be visible in a FROZEN cgroup and frozen tasks in a THAWED one.
174	 */
175	cgroup_taskset_for_each(task, new_css, tset) {
176		struct freezer *freezer = css_freezer(new_css);
177
178		if (!(freezer->state & CGROUP_FREEZING)) {
179			__thaw_task(task);
180		} else {
181			freeze_task(task);
182
183			/* clear FROZEN and propagate upwards */
184			while (freezer && (freezer->state & CGROUP_FROZEN)) {
185				freezer->state &= ~CGROUP_FROZEN;
186				freezer = parent_freezer(freezer);
187			}
188		}
189	}
190
191	mutex_unlock(&freezer_mutex);
192}
193
194/**
195 * freezer_fork - cgroup post fork callback
196 * @task: a task which has just been forked
197 *
198 * @task has just been created and should conform to the current state of
199 * the cgroup_freezer it belongs to.  This function may race against
200 * freezer_attach().  Losing to freezer_attach() means that we don't have
201 * to do anything as freezer_attach() will put @task into the appropriate
202 * state.
203 */
204static void freezer_fork(struct task_struct *task)
205{
206	struct freezer *freezer;
207
208	/*
209	 * The root cgroup is non-freezable, so we can skip locking the
210	 * freezer.  This is safe regardless of race with task migration.
211	 * If we didn't race or won, skipping is obviously the right thing
212	 * to do.  If we lost and root is the new cgroup, noop is still the
213	 * right thing to do.
214	 */
215	if (task_css_is_root(task, freezer_cgrp_id))
216		return;
217
218	mutex_lock(&freezer_mutex);
219	rcu_read_lock();
220
221	freezer = task_freezer(task);
222	if (freezer->state & CGROUP_FREEZING)
223		freeze_task(task);
224
225	rcu_read_unlock();
226	mutex_unlock(&freezer_mutex);
227}
228
229/**
230 * update_if_frozen - update whether a cgroup finished freezing
231 * @css: css of interest
232 *
233 * Once FREEZING is initiated, transition to FROZEN is lazily updated by
234 * calling this function.  If the current state is FREEZING but not FROZEN,
235 * this function checks whether all tasks of this cgroup and the descendant
236 * cgroups finished freezing and, if so, sets FROZEN.
237 *
238 * The caller is responsible for grabbing RCU read lock and calling
239 * update_if_frozen() on all descendants prior to invoking this function.
240 *
241 * Task states and freezer state might disagree while tasks are being
242 * migrated into or out of @css, so we can't verify task states against
243 * @freezer state here.  See freezer_attach() for details.
244 */
245static void update_if_frozen(struct cgroup_subsys_state *css)
246{
247	struct freezer *freezer = css_freezer(css);
248	struct cgroup_subsys_state *pos;
249	struct css_task_iter it;
250	struct task_struct *task;
251
252	lockdep_assert_held(&freezer_mutex);
253
254	if (!(freezer->state & CGROUP_FREEZING) ||
255	    (freezer->state & CGROUP_FROZEN))
256		return;
257
258	/* are all (live) children frozen? */
259	rcu_read_lock();
260	css_for_each_child(pos, css) {
261		struct freezer *child = css_freezer(pos);
262
263		if ((child->state & CGROUP_FREEZER_ONLINE) &&
264		    !(child->state & CGROUP_FROZEN)) {
265			rcu_read_unlock();
266			return;
267		}
268	}
269	rcu_read_unlock();
270
271	/* are all tasks frozen? */
272	css_task_iter_start(css, 0, &it);
273
274	while ((task = css_task_iter_next(&it))) {
275		if (freezing(task) && !frozen(task))
276			goto out_iter_end;
277	}
278
279	freezer->state |= CGROUP_FROZEN;
280out_iter_end:
281	css_task_iter_end(&it);
282}
283
284static int freezer_read(struct seq_file *m, void *v)
285{
286	struct cgroup_subsys_state *css = seq_css(m), *pos;
287
288	mutex_lock(&freezer_mutex);
289	rcu_read_lock();
290
291	/* update states bottom-up */
292	css_for_each_descendant_post(pos, css) {
293		if (!css_tryget_online(pos))
294			continue;
295		rcu_read_unlock();
296
297		update_if_frozen(pos);
298
299		rcu_read_lock();
300		css_put(pos);
301	}
302
303	rcu_read_unlock();
304	mutex_unlock(&freezer_mutex);
305
306	seq_puts(m, freezer_state_strs(css_freezer(css)->state));
307	seq_putc(m, '\n');
308	return 0;
309}
310
311static void freeze_cgroup(struct freezer *freezer)
312{
313	struct css_task_iter it;
314	struct task_struct *task;
315
316	css_task_iter_start(&freezer->css, 0, &it);
317	while ((task = css_task_iter_next(&it)))
318		freeze_task(task);
319	css_task_iter_end(&it);
320}
321
322static void unfreeze_cgroup(struct freezer *freezer)
323{
324	struct css_task_iter it;
325	struct task_struct *task;
326
327	css_task_iter_start(&freezer->css, 0, &it);
328	while ((task = css_task_iter_next(&it)))
329		__thaw_task(task);
330	css_task_iter_end(&it);
331}
332
333/**
334 * freezer_apply_state - apply state change to a single cgroup_freezer
335 * @freezer: freezer to apply state change to
336 * @freeze: whether to freeze or unfreeze
337 * @state: CGROUP_FREEZING_* flag to set or clear
338 *
339 * Set or clear @state on @cgroup according to @freeze, and perform
340 * freezing or thawing as necessary.
341 */
342static void freezer_apply_state(struct freezer *freezer, bool freeze,
343				unsigned int state)
344{
345	/* also synchronizes against task migration, see freezer_attach() */
346	lockdep_assert_held(&freezer_mutex);
347
348	if (!(freezer->state & CGROUP_FREEZER_ONLINE))
349		return;
350
351	if (freeze) {
352		if (!(freezer->state & CGROUP_FREEZING))
353			static_branch_inc(&freezer_active);
354		freezer->state |= state;
355		freeze_cgroup(freezer);
356	} else {
357		bool was_freezing = freezer->state & CGROUP_FREEZING;
358
359		freezer->state &= ~state;
360
361		if (!(freezer->state & CGROUP_FREEZING)) {
362			freezer->state &= ~CGROUP_FROZEN;
363			if (was_freezing)
364				static_branch_dec(&freezer_active);
365			unfreeze_cgroup(freezer);
366		}
367	}
368}
369
370/**
371 * freezer_change_state - change the freezing state of a cgroup_freezer
372 * @freezer: freezer of interest
373 * @freeze: whether to freeze or thaw
374 *
375 * Freeze or thaw @freezer according to @freeze.  The operations are
376 * recursive - all descendants of @freezer will be affected.
377 */
378static void freezer_change_state(struct freezer *freezer, bool freeze)
379{
380	struct cgroup_subsys_state *pos;
381
382	/*
383	 * Update all its descendants in pre-order traversal.  Each
384	 * descendant will try to inherit its parent's FREEZING state as
385	 * CGROUP_FREEZING_PARENT.
386	 */
387	mutex_lock(&freezer_mutex);
388	rcu_read_lock();
389	css_for_each_descendant_pre(pos, &freezer->css) {
390		struct freezer *pos_f = css_freezer(pos);
391		struct freezer *parent = parent_freezer(pos_f);
392
393		if (!css_tryget_online(pos))
394			continue;
395		rcu_read_unlock();
396
397		if (pos_f == freezer)
398			freezer_apply_state(pos_f, freeze,
399					    CGROUP_FREEZING_SELF);
400		else
401			freezer_apply_state(pos_f,
402					    parent->state & CGROUP_FREEZING,
403					    CGROUP_FREEZING_PARENT);
404
405		rcu_read_lock();
406		css_put(pos);
407	}
408	rcu_read_unlock();
409	mutex_unlock(&freezer_mutex);
410}
411
412static ssize_t freezer_write(struct kernfs_open_file *of,
413			     char *buf, size_t nbytes, loff_t off)
414{
415	bool freeze;
416
417	buf = strstrip(buf);
418
419	if (strcmp(buf, freezer_state_strs(0)) == 0)
420		freeze = false;
421	else if (strcmp(buf, freezer_state_strs(CGROUP_FROZEN)) == 0)
422		freeze = true;
423	else
424		return -EINVAL;
425
426	freezer_change_state(css_freezer(of_css(of)), freeze);
427	return nbytes;
428}
429
430static u64 freezer_self_freezing_read(struct cgroup_subsys_state *css,
431				      struct cftype *cft)
432{
433	struct freezer *freezer = css_freezer(css);
434
435	return (bool)(freezer->state & CGROUP_FREEZING_SELF);
436}
437
438static u64 freezer_parent_freezing_read(struct cgroup_subsys_state *css,
439					struct cftype *cft)
440{
441	struct freezer *freezer = css_freezer(css);
442
443	return (bool)(freezer->state & CGROUP_FREEZING_PARENT);
444}
445
446static struct cftype files[] = {
447	{
448		.name = "state",
449		.flags = CFTYPE_NOT_ON_ROOT,
450		.seq_show = freezer_read,
451		.write = freezer_write,
452	},
453	{
454		.name = "self_freezing",
455		.flags = CFTYPE_NOT_ON_ROOT,
456		.read_u64 = freezer_self_freezing_read,
457	},
458	{
459		.name = "parent_freezing",
460		.flags = CFTYPE_NOT_ON_ROOT,
461		.read_u64 = freezer_parent_freezing_read,
462	},
463	{ }	/* terminate */
464};
465
466struct cgroup_subsys freezer_cgrp_subsys = {
467	.css_alloc	= freezer_css_alloc,
468	.css_online	= freezer_css_online,
469	.css_offline	= freezer_css_offline,
470	.css_free	= freezer_css_free,
471	.attach		= freezer_attach,
472	.fork		= freezer_fork,
473	.legacy_cftypes	= files,
474};