ARM: tegra11: clock: Fix boot over-clocking warnings
[linux-2.6.git] / kernel / cgroup_freezer.c
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
25 enum freezer_state {
26         CGROUP_THAWED = 0,
27         CGROUP_FREEZING,
28         CGROUP_FROZEN,
29 };
30
31 struct freezer {
32         struct cgroup_subsys_state css;
33         enum freezer_state state;
34         spinlock_t lock; /* protects _writes_ to state */
35 };
36
37 static 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
45 static 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
51 bool cgroup_freezing(struct task_struct *task)
52 {
53         enum freezer_state state;
54         bool ret;
55
56         rcu_read_lock();
57         state = task_freezer(task)->state;
58         ret = state == CGROUP_FREEZING || state == CGROUP_FROZEN;
59         rcu_read_unlock();
60
61         return ret;
62 }
63
64 /*
65  * cgroups_write_string() limits the size of freezer state strings to
66  * CGROUP_LOCAL_BUFFER_SIZE
67  */
68 static const char *freezer_state_strs[] = {
69         "THAWED",
70         "FREEZING",
71         "FROZEN",
72 };
73
74 /*
75  * State diagram
76  * Transitions are caused by userspace writes to the freezer.state file.
77  * The values in parenthesis are state labels. The rest are edge labels.
78  *
79  * (THAWED) --FROZEN--> (FREEZING) --FROZEN--> (FROZEN)
80  *    ^ ^                    |                     |
81  *    | \_______THAWED_______/                     |
82  *    \__________________________THAWED____________/
83  */
84
85 struct cgroup_subsys freezer_subsys;
86
87 /* Locks taken and their ordering
88  * ------------------------------
89  * cgroup_mutex (AKA cgroup_lock)
90  * freezer->lock
91  * css_set_lock
92  * task->alloc_lock (AKA task_lock)
93  * task->sighand->siglock
94  *
95  * cgroup code forces css_set_lock to be taken before task->alloc_lock
96  *
97  * freezer_create(), freezer_destroy():
98  * cgroup_mutex [ by cgroup core ]
99  *
100  * freezer_can_attach():
101  * cgroup_mutex (held by caller of can_attach)
102  *
103  * freezer_fork() (preserving fork() performance means can't take cgroup_mutex):
104  * freezer->lock
105  *  sighand->siglock (if the cgroup is freezing)
106  *
107  * freezer_read():
108  * cgroup_mutex
109  *  freezer->lock
110  *   write_lock css_set_lock (cgroup iterator start)
111  *    task->alloc_lock
112  *   read_lock css_set_lock (cgroup iterator start)
113  *
114  * freezer_write() (freeze):
115  * cgroup_mutex
116  *  freezer->lock
117  *   write_lock css_set_lock (cgroup iterator start)
118  *    task->alloc_lock
119  *   read_lock css_set_lock (cgroup iterator start)
120  *    sighand->siglock (fake signal delivery inside freeze_task())
121  *
122  * freezer_write() (unfreeze):
123  * cgroup_mutex
124  *  freezer->lock
125  *   write_lock css_set_lock (cgroup iterator start)
126  *    task->alloc_lock
127  *   read_lock css_set_lock (cgroup iterator start)
128  *    task->alloc_lock (inside __thaw_task(), prevents race with refrigerator())
129  *     sighand->siglock
130  */
131 static struct cgroup_subsys_state *freezer_create(struct cgroup *cgroup)
132 {
133         struct freezer *freezer;
134
135         freezer = kzalloc(sizeof(struct freezer), GFP_KERNEL);
136         if (!freezer)
137                 return ERR_PTR(-ENOMEM);
138
139         spin_lock_init(&freezer->lock);
140         freezer->state = CGROUP_THAWED;
141         return &freezer->css;
142 }
143
144 static void freezer_destroy(struct cgroup *cgroup)
145 {
146         struct freezer *freezer = cgroup_freezer(cgroup);
147
148         if (freezer->state != CGROUP_THAWED)
149                 atomic_dec(&system_freezing_cnt);
150         kfree(freezer);
151 }
152
153 /* task is frozen or will freeze immediately when next it gets woken */
154 static bool is_task_frozen_enough(struct task_struct *task)
155 {
156         return frozen(task) ||
157                 (task_is_stopped_or_traced(task) && freezing(task));
158 }
159
160 /*
161  * The call to cgroup_lock() in the freezer.state write method prevents
162  * a write to that file racing against an attach, and hence the
163  * can_attach() result will remain valid until the attach completes.
164  */
165 static int freezer_can_attach(struct cgroup *new_cgroup,
166                               struct cgroup_taskset *tset)
167 {
168         struct freezer *freezer;
169         struct task_struct *task;
170
171         if ((current != task) && (!capable(CAP_SYS_ADMIN))) {
172                 const struct cred *cred = current_cred(), *tcred;
173
174                 tcred = __task_cred(task);
175                 if (cred->euid != tcred->uid && cred->euid != tcred->suid)
176                         return -EPERM;
177         }
178
179         /*
180          * Anything frozen can't move or be moved to/from.
181          */
182         cgroup_taskset_for_each(task, new_cgroup, tset)
183                 if (cgroup_freezing(task))
184                         return -EBUSY;
185
186         freezer = cgroup_freezer(new_cgroup);
187         if (freezer->state != CGROUP_THAWED)
188                 return -EBUSY;
189
190         return 0;
191 }
192
193 static void freezer_fork(struct task_struct *task)
194 {
195         struct freezer *freezer;
196
197         /*
198          * No lock is needed, since the task isn't on tasklist yet,
199          * so it can't be moved to another cgroup, which means the
200          * freezer won't be removed and will be valid during this
201          * function call.  Nevertheless, apply RCU read-side critical
202          * section to suppress RCU lockdep false positives.
203          */
204         rcu_read_lock();
205         freezer = task_freezer(task);
206         rcu_read_unlock();
207
208         /*
209          * The root cgroup is non-freezable, so we can skip the
210          * following check.
211          */
212         if (!freezer->css.cgroup->parent)
213                 return;
214
215         spin_lock_irq(&freezer->lock);
216         BUG_ON(freezer->state == CGROUP_FROZEN);
217
218         /* Locking avoids race with FREEZING -> THAWED transitions. */
219         if (freezer->state == CGROUP_FREEZING)
220                 freeze_task(task);
221         spin_unlock_irq(&freezer->lock);
222 }
223
224 /*
225  * caller must hold freezer->lock
226  */
227 static void update_if_frozen(struct cgroup *cgroup,
228                                  struct freezer *freezer)
229 {
230         struct cgroup_iter it;
231         struct task_struct *task;
232         unsigned int nfrozen = 0, ntotal = 0;
233         enum freezer_state old_state = freezer->state;
234
235         cgroup_iter_start(cgroup, &it);
236         while ((task = cgroup_iter_next(cgroup, &it))) {
237                 ntotal++;
238                 if (freezing(task) && is_task_frozen_enough(task))
239                         nfrozen++;
240         }
241
242         if (old_state == CGROUP_THAWED) {
243                 BUG_ON(nfrozen > 0);
244         } else if (old_state == CGROUP_FREEZING) {
245                 if (nfrozen == ntotal)
246                         freezer->state = CGROUP_FROZEN;
247         } else { /* old_state == CGROUP_FROZEN */
248                 BUG_ON(nfrozen != ntotal);
249         }
250
251         cgroup_iter_end(cgroup, &it);
252 }
253
254 static int freezer_read(struct cgroup *cgroup, struct cftype *cft,
255                         struct seq_file *m)
256 {
257         struct freezer *freezer;
258         enum freezer_state state;
259
260         if (!cgroup_lock_live_group(cgroup))
261                 return -ENODEV;
262
263         freezer = cgroup_freezer(cgroup);
264         spin_lock_irq(&freezer->lock);
265         state = freezer->state;
266         if (state == CGROUP_FREEZING) {
267                 /* We change from FREEZING to FROZEN lazily if the cgroup was
268                  * only partially frozen when we exitted write. */
269                 update_if_frozen(cgroup, freezer);
270                 state = freezer->state;
271         }
272         spin_unlock_irq(&freezer->lock);
273         cgroup_unlock();
274
275         seq_puts(m, freezer_state_strs[state]);
276         seq_putc(m, '\n');
277         return 0;
278 }
279
280 static int try_to_freeze_cgroup(struct cgroup *cgroup, struct freezer *freezer)
281 {
282         struct cgroup_iter it;
283         struct task_struct *task;
284         unsigned int num_cant_freeze_now = 0;
285
286         cgroup_iter_start(cgroup, &it);
287         while ((task = cgroup_iter_next(cgroup, &it))) {
288                 if (!freeze_task(task))
289                         continue;
290                 if (is_task_frozen_enough(task))
291                         continue;
292                 if (!freezing(task) && !freezer_should_skip(task))
293                         num_cant_freeze_now++;
294         }
295         cgroup_iter_end(cgroup, &it);
296
297         return num_cant_freeze_now ? -EBUSY : 0;
298 }
299
300 static void unfreeze_cgroup(struct cgroup *cgroup, struct freezer *freezer)
301 {
302         struct cgroup_iter it;
303         struct task_struct *task;
304
305         cgroup_iter_start(cgroup, &it);
306         while ((task = cgroup_iter_next(cgroup, &it)))
307                 __thaw_task(task);
308         cgroup_iter_end(cgroup, &it);
309 }
310
311 static int freezer_change_state(struct cgroup *cgroup,
312                                 enum freezer_state goal_state)
313 {
314         struct freezer *freezer;
315         int retval = 0;
316
317         freezer = cgroup_freezer(cgroup);
318
319         spin_lock_irq(&freezer->lock);
320
321         update_if_frozen(cgroup, freezer);
322
323         switch (goal_state) {
324         case CGROUP_THAWED:
325                 if (freezer->state != CGROUP_THAWED)
326                         atomic_dec(&system_freezing_cnt);
327                 freezer->state = CGROUP_THAWED;
328                 unfreeze_cgroup(cgroup, freezer);
329                 break;
330         case CGROUP_FROZEN:
331                 if (freezer->state == CGROUP_THAWED)
332                         atomic_inc(&system_freezing_cnt);
333                 freezer->state = CGROUP_FREEZING;
334                 retval = try_to_freeze_cgroup(cgroup, freezer);
335                 break;
336         default:
337                 BUG();
338         }
339
340         spin_unlock_irq(&freezer->lock);
341
342         return retval;
343 }
344
345 static int freezer_write(struct cgroup *cgroup,
346                          struct cftype *cft,
347                          const char *buffer)
348 {
349         int retval;
350         enum freezer_state goal_state;
351
352         if (strcmp(buffer, freezer_state_strs[CGROUP_THAWED]) == 0)
353                 goal_state = CGROUP_THAWED;
354         else if (strcmp(buffer, freezer_state_strs[CGROUP_FROZEN]) == 0)
355                 goal_state = CGROUP_FROZEN;
356         else
357                 return -EINVAL;
358
359         if (!cgroup_lock_live_group(cgroup))
360                 return -ENODEV;
361         retval = freezer_change_state(cgroup, goal_state);
362         cgroup_unlock();
363         return retval;
364 }
365
366 static struct cftype files[] = {
367         {
368                 .name = "state",
369                 .read_seq_string = freezer_read,
370                 .write_string = freezer_write,
371         },
372 };
373
374 static int freezer_populate(struct cgroup_subsys *ss, struct cgroup *cgroup)
375 {
376         if (!cgroup->parent)
377                 return 0;
378         return cgroup_add_files(cgroup, ss, files, ARRAY_SIZE(files));
379 }
380
381 struct cgroup_subsys freezer_subsys = {
382         .name           = "freezer",
383         .create         = freezer_create,
384         .destroy        = freezer_destroy,
385         .populate       = freezer_populate,
386         .subsys_id      = freezer_subsys_id,
387         .can_attach     = freezer_can_attach,
388         .fork           = freezer_fork,
389 };