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/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         CGROUP_THAWED = 0,
  26         CGROUP_FREEZING,
  27         CGROUP_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 == CGROUP_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         "THAWED",
  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  * (THAWED) --FROZEN--> (FREEZING) --FROZEN--> (FROZEN)
  79  *    ^ ^                    |                     |
  80  *    | \_______THAWED_______/                     |
  81  *    \__________________________THAWED____________/
  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 = CGROUP_THAWED;
 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 /* Task is frozen or will freeze immediately when next it gets woken */
 149 static bool is_task_frozen_enough(struct task_struct *task)
 150 {
 151         return frozen(task) ||
 152                 (task_is_stopped_or_traced(task) && freezing(task));
 153 }
 154
 155 /*
 156  * The call to cgroup_lock() in the freezer.state write method prevents
 157  * a write to that file racing against an attach, and hence the
 158  * can_attach() result will remain valid until the attach completes.
 159  */
 160 static int freezer_can_attach(struct cgroup_subsys *ss,
 161                               struct cgroup *new_cgroup,
 162                               struct task_struct *task)
 163 {
 164         struct freezer *freezer;
 165
 166         /*
 167          * Anything frozen can't move or be moved to/from.
 168          *
 169          * Since orig_freezer->state == FROZEN means that @task has been
 170          * frozen, so it's sufficient to check the latter condition.
 171          */
 172
 173         if (is_task_frozen_enough(task))
 174                 return -EBUSY;
 175
 176         freezer = cgroup_freezer(new_cgroup);
 177         if (freezer->state == CGROUP_FROZEN)
 178                 return -EBUSY;
 179
 180         return 0;
 181 }
 182
 183 static void freezer_fork(struct cgroup_subsys *ss, struct task_struct *task)
 184 {
 185         struct freezer *freezer;
 186
 187         task_lock(task);
 188         freezer = task_freezer(task);
 189         task_unlock(task);
 190
 191         spin_lock_irq(&freezer->lock);
 192         BUG_ON(freezer->state == CGROUP_FROZEN);
 193
 194         /* Locking avoids race with FREEZING -> THAWED transitions. */
 195         if (freezer->state == CGROUP_FREEZING)
 196                 freeze_task(task, true);
 197         spin_unlock_irq(&freezer->lock);
 198 }
 199
 200 /*
 201  * caller must hold freezer->lock
 202  */
 203 static void update_freezer_state(struct cgroup *cgroup,
 204                                  struct freezer *freezer)
 205 {
 206         struct cgroup_iter it;
 207         struct task_struct *task;
 208         unsigned int nfrozen = 0, ntotal = 0;
 209
 210         cgroup_iter_start(cgroup, &it);
 211         while ((task = cgroup_iter_next(cgroup, &it))) {
 212                 ntotal++;
 213                 if (is_task_frozen_enough(task))
 214                         nfrozen++;
 215         }
 216
 217         /*
 218          * Transition to FROZEN when no new tasks can be added ensures
 219          * that we never exist in the FROZEN state while there are unfrozen
 220          * tasks.
 221          */
 222         if (nfrozen == ntotal)
 223                 freezer->state = CGROUP_FROZEN;
 224         else if (nfrozen > 0)
 225                 freezer->state = CGROUP_FREEZING;
 226         else
 227                 freezer->state = CGROUP_THAWED;
 228         cgroup_iter_end(cgroup, &it);
 229 }
 230
 231 static int freezer_read(struct cgroup *cgroup, struct cftype *cft,
 232                         struct seq_file *m)
 233 {
 234         struct freezer *freezer;
 235         enum freezer_state state;
 236
 237         if (!cgroup_lock_live_group(cgroup))
 238                 return -ENODEV;
 239
 240         freezer = cgroup_freezer(cgroup);
 241         spin_lock_irq(&freezer->lock);
 242         state = freezer->state;
 243         if (state == CGROUP_FREEZING) {
 244                 /* We change from FREEZING to FROZEN lazily if the cgroup was
 245                  * only partially frozen when we exitted write. */
 246                 update_freezer_state(cgroup, freezer);
 247                 state = freezer->state;
 248         }
 249         spin_unlock_irq(&freezer->lock);
 250         cgroup_unlock();
 251
 252         seq_puts(m, freezer_state_strs[state]);
 253         seq_putc(m, '\n');
 254         return 0;
 255 }
 256
 257 static int try_to_freeze_cgroup(struct cgroup *cgroup, struct freezer *freezer)
 258 {
 259         struct cgroup_iter it;
 260         struct task_struct *task;
 261         unsigned int num_cant_freeze_now = 0;
 262
 263         freezer->state = CGROUP_FREEZING;
 264         cgroup_iter_start(cgroup, &it);
 265         while ((task = cgroup_iter_next(cgroup, &it))) {
 266                 if (!freeze_task(task, true))
 267                         continue;
 268                 if (is_task_frozen_enough(task))
 269                         continue;
 270                 if (!freezing(task) && !freezer_should_skip(task))
 271                         num_cant_freeze_now++;
 272         }
 273         cgroup_iter_end(cgroup, &it);
 274
 275         return num_cant_freeze_now ? -EBUSY : 0;
 276 }
 277
 278 static void unfreeze_cgroup(struct cgroup *cgroup, struct freezer *freezer)
 279 {
 280         struct cgroup_iter it;
 281         struct task_struct *task;
 282
 283         cgroup_iter_start(cgroup, &it);
 284         while ((task = cgroup_iter_next(cgroup, &it))) {
 285                 thaw_process(task);
 286         }
 287         cgroup_iter_end(cgroup, &it);
 288
 289         freezer->state = CGROUP_THAWED;
 290 }
 291
 292 static int freezer_change_state(struct cgroup *cgroup,
 293                                 enum freezer_state goal_state)
 294 {
 295         struct freezer *freezer;
 296         int retval = 0;
 297
 298         freezer = cgroup_freezer(cgroup);
 299
 300         spin_lock_irq(&freezer->lock);
 301
 302         update_freezer_state(cgroup, freezer);
 303         if (goal_state == freezer->state)
 304                 goto out;
 305
 306         switch (goal_state) {
 307         case CGROUP_THAWED:
 308                 unfreeze_cgroup(cgroup, freezer);
 309                 break;
 310         case CGROUP_FROZEN:
 311                 retval = try_to_freeze_cgroup(cgroup, freezer);
 312                 break;
 313         default:
 314                 BUG();
 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[CGROUP_THAWED]) == 0)
 330                 goal_state = CGROUP_THAWED;
 331         else if (strcmp(buffer, freezer_state_strs[CGROUP_FROZEN]) == 0)
 332                 goal_state = CGROUP_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 };