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         /*
 188          * No lock is needed, since the task isn't on tasklist yet,
 189          * so it can't be moved to another cgroup, which means the
 190          * freezer won't be removed and will be valid during this
 191          * function call.
 192          */
 193         freezer = task_freezer(task);
 194
 195         /*
 196          * The root cgroup is non-freezable, so we can skip the
 197          * following check.
 198          */
 199         if (!freezer->css.cgroup->parent)
 200                 return;
 201
 202         spin_lock_irq(&freezer->lock);
 203         BUG_ON(freezer->state == CGROUP_FROZEN);
 204
 205         /* Locking avoids race with FREEZING -> THAWED transitions. */
 206         if (freezer->state == CGROUP_FREEZING)
 207                 freeze_task(task, true);
 208         spin_unlock_irq(&freezer->lock);
 209 }
 210
 211 /*
 212  * caller must hold freezer->lock
 213  */
 214 static void update_freezer_state(struct cgroup *cgroup,
 215                                  struct freezer *freezer)
 216 {
 217         struct cgroup_iter it;
 218         struct task_struct *task;
 219         unsigned int nfrozen = 0, ntotal = 0;
 220
 221         cgroup_iter_start(cgroup, &it);
 222         while ((task = cgroup_iter_next(cgroup, &it))) {
 223                 ntotal++;
 224                 if (is_task_frozen_enough(task))
 225                         nfrozen++;
 226         }
 227
 228         /*
 229          * Transition to FROZEN when no new tasks can be added ensures
 230          * that we never exist in the FROZEN state while there are unfrozen
 231          * tasks.
 232          */
 233         if (nfrozen == ntotal)
 234                 freezer->state = CGROUP_FROZEN;
 235         else if (nfrozen > 0)
 236                 freezer->state = CGROUP_FREEZING;
 237         else
 238                 freezer->state = CGROUP_THAWED;
 239         cgroup_iter_end(cgroup, &it);
 240 }
 241
 242 static int freezer_read(struct cgroup *cgroup, struct cftype *cft,
 243                         struct seq_file *m)
 244 {
 245         struct freezer *freezer;
 246         enum freezer_state state;
 247
 248         if (!cgroup_lock_live_group(cgroup))
 249                 return -ENODEV;
 250
 251         freezer = cgroup_freezer(cgroup);
 252         spin_lock_irq(&freezer->lock);
 253         state = freezer->state;
 254         if (state == CGROUP_FREEZING) {
 255                 /* We change from FREEZING to FROZEN lazily if the cgroup was
 256                  * only partially frozen when we exitted write. */
 257                 update_freezer_state(cgroup, freezer);
 258                 state = freezer->state;
 259         }
 260         spin_unlock_irq(&freezer->lock);
 261         cgroup_unlock();
 262
 263         seq_puts(m, freezer_state_strs[state]);
 264         seq_putc(m, '\n');
 265         return 0;
 266 }
 267
 268 static int try_to_freeze_cgroup(struct cgroup *cgroup, struct freezer *freezer)
 269 {
 270         struct cgroup_iter it;
 271         struct task_struct *task;
 272         unsigned int num_cant_freeze_now = 0;
 273
 274         freezer->state = CGROUP_FREEZING;
 275         cgroup_iter_start(cgroup, &it);
 276         while ((task = cgroup_iter_next(cgroup, &it))) {
 277                 if (!freeze_task(task, true))
 278                         continue;
 279                 if (is_task_frozen_enough(task))
 280                         continue;
 281                 if (!freezing(task) && !freezer_should_skip(task))
 282                         num_cant_freeze_now++;
 283         }
 284         cgroup_iter_end(cgroup, &it);
 285
 286         return num_cant_freeze_now ? -EBUSY : 0;
 287 }
 288
 289 static void unfreeze_cgroup(struct cgroup *cgroup, struct freezer *freezer)
 290 {
 291         struct cgroup_iter it;
 292         struct task_struct *task;
 293
 294         cgroup_iter_start(cgroup, &it);
 295         while ((task = cgroup_iter_next(cgroup, &it))) {
 296                 thaw_process(task);
 297         }
 298         cgroup_iter_end(cgroup, &it);
 299
 300         freezer->state = CGROUP_THAWED;
 301 }
 302
 303 static int freezer_change_state(struct cgroup *cgroup,
 304                                 enum freezer_state goal_state)
 305 {
 306         struct freezer *freezer;
 307         int retval = 0;
 308
 309         freezer = cgroup_freezer(cgroup);
 310
 311         spin_lock_irq(&freezer->lock);
 312
 313         update_freezer_state(cgroup, freezer);
 314         if (goal_state == freezer->state)
 315                 goto out;
 316
 317         switch (goal_state) {
 318         case CGROUP_THAWED:
 319                 unfreeze_cgroup(cgroup, freezer);
 320                 break;
 321         case CGROUP_FROZEN:
 322                 retval = try_to_freeze_cgroup(cgroup, freezer);
 323                 break;
 324         default:
 325                 BUG();
 326         }
 327 out:
 328         spin_unlock_irq(&freezer->lock);
 329
 330         return retval;
 331 }
 332
 333 static int freezer_write(struct cgroup *cgroup,
 334                          struct cftype *cft,
 335                          const char *buffer)
 336 {
 337         int retval;
 338         enum freezer_state goal_state;
 339
 340         if (strcmp(buffer, freezer_state_strs[CGROUP_THAWED]) == 0)
 341                 goal_state = CGROUP_THAWED;
 342         else if (strcmp(buffer, freezer_state_strs[CGROUP_FROZEN]) == 0)
 343                 goal_state = CGROUP_FROZEN;
 344         else
 345                 return -EINVAL;
 346
 347         if (!cgroup_lock_live_group(cgroup))
 348                 return -ENODEV;
 349         retval = freezer_change_state(cgroup, goal_state);
 350         cgroup_unlock();
 351         return retval;
 352 }
 353
 354 static struct cftype files[] = {
 355         {
 356                 .name = "state",
 357                 .read_seq_string = freezer_read,
 358                 .write_string = freezer_write,
 359         },
 360 };
 361
 362 static int freezer_populate(struct cgroup_subsys *ss, struct cgroup *cgroup)
 363 {
 364         if (!cgroup->parent)
 365                 return 0;
 366         return cgroup_add_files(cgroup, ss, files, ARRAY_SIZE(files));
 367 }
 368
 369 struct cgroup_subsys freezer_subsys = {
 370         .name           = "freezer",
 371         .create         = freezer_create,
 372         .destroy        = freezer_destroy,
 373         .populate       = freezer_populate,
 374         .subsys_id      = freezer_subsys_id,
 375         .can_attach     = freezer_can_attach,
 376         .attach         = NULL,
 377         .fork           = freezer_fork,
 378         .exit           = NULL,
 379 };