]>
Commit | Line | Data |
---|---|---|
dc52ddc0 MH |
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/cgroup.h> | |
19 | #include <linux/fs.h> | |
20 | #include <linux/uaccess.h> | |
21 | #include <linux/freezer.h> | |
22 | #include <linux/seq_file.h> | |
23 | ||
24 | enum freezer_state { | |
25 | STATE_RUNNING = 0, | |
26 | STATE_FREEZING, | |
27 | STATE_FROZEN, | |
28 | }; | |
29 | ||
30 | struct freezer { | |
31 | struct cgroup_subsys_state css; | |
32 | enum freezer_state state; | |
33 | spinlock_t lock; /* protects _writes_ to state */ | |
34 | }; | |
35 | ||
36 | static inline struct freezer *cgroup_freezer( | |
37 | struct cgroup *cgroup) | |
38 | { | |
39 | return container_of( | |
40 | cgroup_subsys_state(cgroup, freezer_subsys_id), | |
41 | struct freezer, css); | |
42 | } | |
43 | ||
44 | static inline struct freezer *task_freezer(struct task_struct *task) | |
45 | { | |
46 | return container_of(task_subsys_state(task, freezer_subsys_id), | |
47 | struct freezer, css); | |
48 | } | |
49 | ||
50 | int cgroup_frozen(struct task_struct *task) | |
51 | { | |
52 | struct freezer *freezer; | |
53 | enum freezer_state state; | |
54 | ||
55 | task_lock(task); | |
56 | freezer = task_freezer(task); | |
57 | state = freezer->state; | |
58 | task_unlock(task); | |
59 | ||
60 | return state == STATE_FROZEN; | |
61 | } | |
62 | ||
63 | /* | |
64 | * cgroups_write_string() limits the size of freezer state strings to | |
65 | * CGROUP_LOCAL_BUFFER_SIZE | |
66 | */ | |
67 | static const char *freezer_state_strs[] = { | |
68 | "RUNNING", | |
69 | "FREEZING", | |
70 | "FROZEN", | |
71 | }; | |
72 | ||
73 | /* | |
74 | * State diagram | |
75 | * Transitions are caused by userspace writes to the freezer.state file. | |
76 | * The values in parenthesis are state labels. The rest are edge labels. | |
77 | * | |
78 | * (RUNNING) --FROZEN--> (FREEZING) --FROZEN--> (FROZEN) | |
79 | * ^ ^ | | | |
80 | * | \_______RUNNING_______/ | | |
81 | * \_____________________________RUNNING___________/ | |
82 | */ | |
83 | ||
84 | struct cgroup_subsys freezer_subsys; | |
85 | ||
86 | /* Locks taken and their ordering | |
87 | * ------------------------------ | |
88 | * css_set_lock | |
89 | * cgroup_mutex (AKA cgroup_lock) | |
90 | * task->alloc_lock (AKA task_lock) | |
91 | * freezer->lock | |
92 | * task->sighand->siglock | |
93 | * | |
94 | * cgroup code forces css_set_lock to be taken before task->alloc_lock | |
95 | * | |
96 | * freezer_create(), freezer_destroy(): | |
97 | * cgroup_mutex [ by cgroup core ] | |
98 | * | |
99 | * can_attach(): | |
100 | * cgroup_mutex | |
101 | * | |
102 | * cgroup_frozen(): | |
103 | * task->alloc_lock (to get task's cgroup) | |
104 | * | |
105 | * freezer_fork() (preserving fork() performance means can't take cgroup_mutex): | |
106 | * task->alloc_lock (to get task's cgroup) | |
107 | * freezer->lock | |
108 | * sighand->siglock (if the cgroup is freezing) | |
109 | * | |
110 | * freezer_read(): | |
111 | * cgroup_mutex | |
112 | * freezer->lock | |
113 | * read_lock css_set_lock (cgroup iterator start) | |
114 | * | |
115 | * freezer_write() (freeze): | |
116 | * cgroup_mutex | |
117 | * freezer->lock | |
118 | * read_lock css_set_lock (cgroup iterator start) | |
119 | * sighand->siglock | |
120 | * | |
121 | * freezer_write() (unfreeze): | |
122 | * cgroup_mutex | |
123 | * freezer->lock | |
124 | * read_lock css_set_lock (cgroup iterator start) | |
125 | * task->alloc_lock (to prevent races with freeze_task()) | |
126 | * sighand->siglock | |
127 | */ | |
128 | static struct cgroup_subsys_state *freezer_create(struct cgroup_subsys *ss, | |
129 | struct cgroup *cgroup) | |
130 | { | |
131 | struct freezer *freezer; | |
132 | ||
133 | freezer = kzalloc(sizeof(struct freezer), GFP_KERNEL); | |
134 | if (!freezer) | |
135 | return ERR_PTR(-ENOMEM); | |
136 | ||
137 | spin_lock_init(&freezer->lock); | |
138 | freezer->state = STATE_RUNNING; | |
139 | return &freezer->css; | |
140 | } | |
141 | ||
142 | static void freezer_destroy(struct cgroup_subsys *ss, | |
143 | struct cgroup *cgroup) | |
144 | { | |
145 | kfree(cgroup_freezer(cgroup)); | |
146 | } | |
147 | ||
148 | ||
149 | static int freezer_can_attach(struct cgroup_subsys *ss, | |
150 | struct cgroup *new_cgroup, | |
151 | struct task_struct *task) | |
152 | { | |
153 | struct freezer *freezer; | |
154 | int retval = 0; | |
155 | ||
156 | /* | |
157 | * The call to cgroup_lock() in the freezer.state write method prevents | |
158 | * a write to that file racing against an attach, and hence the | |
159 | * can_attach() result will remain valid until the attach completes. | |
160 | */ | |
161 | freezer = cgroup_freezer(new_cgroup); | |
162 | if (freezer->state == STATE_FROZEN) | |
163 | retval = -EBUSY; | |
164 | return retval; | |
165 | } | |
166 | ||
167 | static void freezer_fork(struct cgroup_subsys *ss, struct task_struct *task) | |
168 | { | |
169 | struct freezer *freezer; | |
170 | ||
171 | task_lock(task); | |
172 | freezer = task_freezer(task); | |
173 | task_unlock(task); | |
174 | ||
175 | BUG_ON(freezer->state == STATE_FROZEN); | |
176 | spin_lock_irq(&freezer->lock); | |
177 | /* Locking avoids race with FREEZING -> RUNNING transitions. */ | |
178 | if (freezer->state == STATE_FREEZING) | |
179 | freeze_task(task, true); | |
180 | spin_unlock_irq(&freezer->lock); | |
181 | } | |
182 | ||
183 | /* | |
184 | * caller must hold freezer->lock | |
185 | */ | |
186 | static void check_if_frozen(struct cgroup *cgroup, | |
187 | struct freezer *freezer) | |
188 | { | |
189 | struct cgroup_iter it; | |
190 | struct task_struct *task; | |
191 | unsigned int nfrozen = 0, ntotal = 0; | |
192 | ||
193 | cgroup_iter_start(cgroup, &it); | |
194 | while ((task = cgroup_iter_next(cgroup, &it))) { | |
195 | ntotal++; | |
196 | /* | |
197 | * Task is frozen or will freeze immediately when next it gets | |
198 | * woken | |
199 | */ | |
200 | if (frozen(task) || | |
201 | (task_is_stopped_or_traced(task) && freezing(task))) | |
202 | nfrozen++; | |
203 | } | |
204 | ||
205 | /* | |
206 | * Transition to FROZEN when no new tasks can be added ensures | |
207 | * that we never exist in the FROZEN state while there are unfrozen | |
208 | * tasks. | |
209 | */ | |
210 | if (nfrozen == ntotal) | |
211 | freezer->state = STATE_FROZEN; | |
212 | cgroup_iter_end(cgroup, &it); | |
213 | } | |
214 | ||
215 | static int freezer_read(struct cgroup *cgroup, struct cftype *cft, | |
216 | struct seq_file *m) | |
217 | { | |
218 | struct freezer *freezer; | |
219 | enum freezer_state state; | |
220 | ||
221 | if (!cgroup_lock_live_group(cgroup)) | |
222 | return -ENODEV; | |
223 | ||
224 | freezer = cgroup_freezer(cgroup); | |
225 | spin_lock_irq(&freezer->lock); | |
226 | state = freezer->state; | |
227 | if (state == STATE_FREEZING) { | |
228 | /* We change from FREEZING to FROZEN lazily if the cgroup was | |
229 | * only partially frozen when we exitted write. */ | |
230 | check_if_frozen(cgroup, freezer); | |
231 | state = freezer->state; | |
232 | } | |
233 | spin_unlock_irq(&freezer->lock); | |
234 | cgroup_unlock(); | |
235 | ||
236 | seq_puts(m, freezer_state_strs[state]); | |
237 | seq_putc(m, '\n'); | |
238 | return 0; | |
239 | } | |
240 | ||
241 | static int try_to_freeze_cgroup(struct cgroup *cgroup, struct freezer *freezer) | |
242 | { | |
243 | struct cgroup_iter it; | |
244 | struct task_struct *task; | |
245 | unsigned int num_cant_freeze_now = 0; | |
246 | ||
247 | freezer->state = STATE_FREEZING; | |
248 | cgroup_iter_start(cgroup, &it); | |
249 | while ((task = cgroup_iter_next(cgroup, &it))) { | |
250 | if (!freeze_task(task, true)) | |
251 | continue; | |
252 | if (task_is_stopped_or_traced(task) && freezing(task)) | |
253 | /* | |
254 | * The freeze flag is set so these tasks will | |
255 | * immediately go into the fridge upon waking. | |
256 | */ | |
257 | continue; | |
258 | if (!freezing(task) && !freezer_should_skip(task)) | |
259 | num_cant_freeze_now++; | |
260 | } | |
261 | cgroup_iter_end(cgroup, &it); | |
262 | ||
263 | return num_cant_freeze_now ? -EBUSY : 0; | |
264 | } | |
265 | ||
266 | static int unfreeze_cgroup(struct cgroup *cgroup, struct freezer *freezer) | |
267 | { | |
268 | struct cgroup_iter it; | |
269 | struct task_struct *task; | |
270 | ||
271 | cgroup_iter_start(cgroup, &it); | |
272 | while ((task = cgroup_iter_next(cgroup, &it))) { | |
273 | int do_wake; | |
274 | ||
275 | task_lock(task); | |
276 | do_wake = __thaw_process(task); | |
277 | task_unlock(task); | |
278 | if (do_wake) | |
279 | wake_up_process(task); | |
280 | } | |
281 | cgroup_iter_end(cgroup, &it); | |
282 | freezer->state = STATE_RUNNING; | |
283 | ||
284 | return 0; | |
285 | } | |
286 | ||
287 | static int freezer_change_state(struct cgroup *cgroup, | |
288 | enum freezer_state goal_state) | |
289 | { | |
290 | struct freezer *freezer; | |
291 | int retval = 0; | |
292 | ||
293 | freezer = cgroup_freezer(cgroup); | |
294 | spin_lock_irq(&freezer->lock); | |
295 | check_if_frozen(cgroup, freezer); /* may update freezer->state */ | |
296 | if (goal_state == freezer->state) | |
297 | goto out; | |
298 | switch (freezer->state) { | |
299 | case STATE_RUNNING: | |
300 | retval = try_to_freeze_cgroup(cgroup, freezer); | |
301 | break; | |
302 | case STATE_FREEZING: | |
303 | if (goal_state == STATE_FROZEN) { | |
304 | /* Userspace is retrying after | |
305 | * "/bin/echo FROZEN > freezer.state" returned -EBUSY */ | |
306 | retval = try_to_freeze_cgroup(cgroup, freezer); | |
307 | break; | |
308 | } | |
309 | /* state == FREEZING and goal_state == RUNNING, so unfreeze */ | |
310 | case STATE_FROZEN: | |
311 | retval = unfreeze_cgroup(cgroup, freezer); | |
312 | break; | |
313 | default: | |
314 | break; | |
315 | } | |
316 | out: | |
317 | spin_unlock_irq(&freezer->lock); | |
318 | ||
319 | return retval; | |
320 | } | |
321 | ||
322 | static int freezer_write(struct cgroup *cgroup, | |
323 | struct cftype *cft, | |
324 | const char *buffer) | |
325 | { | |
326 | int retval; | |
327 | enum freezer_state goal_state; | |
328 | ||
329 | if (strcmp(buffer, freezer_state_strs[STATE_RUNNING]) == 0) | |
330 | goal_state = STATE_RUNNING; | |
331 | else if (strcmp(buffer, freezer_state_strs[STATE_FROZEN]) == 0) | |
332 | goal_state = STATE_FROZEN; | |
333 | else | |
334 | return -EIO; | |
335 | ||
336 | if (!cgroup_lock_live_group(cgroup)) | |
337 | return -ENODEV; | |
338 | retval = freezer_change_state(cgroup, goal_state); | |
339 | cgroup_unlock(); | |
340 | return retval; | |
341 | } | |
342 | ||
343 | static struct cftype files[] = { | |
344 | { | |
345 | .name = "state", | |
346 | .read_seq_string = freezer_read, | |
347 | .write_string = freezer_write, | |
348 | }, | |
349 | }; | |
350 | ||
351 | static int freezer_populate(struct cgroup_subsys *ss, struct cgroup *cgroup) | |
352 | { | |
353 | return cgroup_add_files(cgroup, ss, files, ARRAY_SIZE(files)); | |
354 | } | |
355 | ||
356 | struct cgroup_subsys freezer_subsys = { | |
357 | .name = "freezer", | |
358 | .create = freezer_create, | |
359 | .destroy = freezer_destroy, | |
360 | .populate = freezer_populate, | |
361 | .subsys_id = freezer_subsys_id, | |
362 | .can_attach = freezer_can_attach, | |
363 | .attach = NULL, | |
364 | .fork = freezer_fork, | |
365 | .exit = NULL, | |
366 | }; |