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/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 int cgroup_freezing_or_frozen(struct task_struct *task)
  52 {
  53         struct freezer *freezer;
  54         enum freezer_state state;
  55
  56         task_lock(task);
  57         freezer = task_freezer(task);
  58         if (!freezer->css.cgroup->parent)
  59                 state = CGROUP_THAWED; /* root cgroup can't be frozen */
  60         else
  61                 state = freezer->state;
  62         task_unlock(task);
  63
  64         return (state == CGROUP_FREEZING) || (state == CGROUP_FROZEN);
  65 }
  66
  67 /*
  68  * cgroups_write_string() limits the size of freezer state strings to
  69  * CGROUP_LOCAL_BUFFER_SIZE
  70  */
  71 static const char *freezer_state_strs[] = {
  72         "THAWED",
  73         "FREEZING",
  74         "FROZEN",
  75 };
  76
  77 /*
  78  * State diagram
  79  * Transitions are caused by userspace writes to the freezer.state file.
  80  * The values in parenthesis are state labels. The rest are edge labels.
  81  *
  82  * (THAWED) --FROZEN--> (FREEZING) --FROZEN--> (FROZEN)
  83  *    ^ ^                    |                     |
  84  *    | \_______THAWED_______/                     |
  85  *    \__________________________THAWED____________/
  86  */
  87
  88 struct cgroup_subsys freezer_subsys;
  89
  90 /* Locks taken and their ordering
  91  * ------------------------------
  92  * css_set_lock
  93  * cgroup_mutex (AKA cgroup_lock)
  94  * task->alloc_lock (AKA task_lock)
  95  * freezer->lock
  96  * task->sighand->siglock
  97  *
  98  * cgroup code forces css_set_lock to be taken before task->alloc_lock
  99  *
 100  * freezer_create(), freezer_destroy():
 101  * cgroup_mutex [ by cgroup core ]
 102  *
 103  * can_attach():
 104  * cgroup_mutex
 105  *
 106  * cgroup_frozen():
 107  * task->alloc_lock (to get task's cgroup)
 108  *
 109  * freezer_fork() (preserving fork() performance means can't take cgroup_mutex):
 110  * task->alloc_lock (to get task's cgroup)
 111  * freezer->lock
 112  *  sighand->siglock (if the cgroup is freezing)
 113  *
 114  * freezer_read():
 115  * cgroup_mutex
 116  *  freezer->lock
 117  *   read_lock css_set_lock (cgroup iterator start)
 118  *
 119  * freezer_write() (freeze):
 120  * cgroup_mutex
 121  *  freezer->lock
 122  *   read_lock css_set_lock (cgroup iterator start)
 123  *    sighand->siglock
 124  *
 125  * freezer_write() (unfreeze):
 126  * cgroup_mutex
 127  *  freezer->lock
 128  *   read_lock css_set_lock (cgroup iterator start)
 129  *    task->alloc_lock (to prevent races with freeze_task())
 130  *     sighand->siglock
 131  */
 132 static struct cgroup_subsys_state *freezer_create(struct cgroup_subsys *ss,
 133                                                   struct cgroup *cgroup)
 134 {
 135         struct freezer *freezer;
 136
 137         freezer = kzalloc(sizeof(struct freezer), GFP_KERNEL);
 138         if (!freezer)
 139                 return ERR_PTR(-ENOMEM);
 140
 141         spin_lock_init(&freezer->lock);
 142         freezer->state = CGROUP_THAWED;
 143         return &freezer->css;
 144 }
 145
 146 static void freezer_destroy(struct cgroup_subsys *ss,
 147                             struct cgroup *cgroup)
 148 {
 149         kfree(cgroup_freezer(cgroup));
 150 }
 151
 152 /* Task is frozen or will freeze immediately when next it gets woken */
 153 static bool is_task_frozen_enough(struct task_struct *task)
 154 {
 155         return frozen(task) ||
 156                 (task_is_stopped_or_traced(task) && freezing(task));
 157 }
 158
 159 /*
 160  * The call to cgroup_lock() in the freezer.state write method prevents
 161  * a write to that file racing against an attach, and hence the
 162  * can_attach() result will remain valid until the attach completes.
 163  */
 164 static int freezer_can_attach(struct cgroup_subsys *ss,
 165                               struct cgroup *new_cgroup,
 166                               struct task_struct *task, bool threadgroup)
 167 {
 168         struct freezer *freezer;
 169
 170         /*
 171          * Anything frozen can't move or be moved to/from.
 172          *
 173          * Since orig_freezer->state == FROZEN means that @task has been
 174          * frozen, so it's sufficient to check the latter condition.
 175          */
 176
 177         if (is_task_frozen_enough(task))
 178                 return -EBUSY;
 179
 180         freezer = cgroup_freezer(new_cgroup);
 181         if (freezer->state == CGROUP_FROZEN)
 182                 return -EBUSY;
 183
 184         if (threadgroup) {
 185                 struct task_struct *c;
 186
 187                 rcu_read_lock();
 188                 list_for_each_entry_rcu(c, &task->thread_group, thread_group) {
 189                         if (is_task_frozen_enough(c)) {
 190                                 rcu_read_unlock();
 191                                 return -EBUSY;
 192                         }
 193                 }
 194                 rcu_read_unlock();
 195         }
 196
 197         return 0;
 198 }
 199
 200 static void freezer_fork(struct cgroup_subsys *ss, struct task_struct *task)
 201 {
 202         struct freezer *freezer;
 203
 204         /*
 205          * No lock is needed, since the task isn't on tasklist yet,
 206          * so it can't be moved to another cgroup, which means the
 207          * freezer won't be removed and will be valid during this
 208          * function call.  Nevertheless, apply RCU read-side critical
 209          * section to suppress RCU lockdep false positives.
 210          */
 211         rcu_read_lock();
 212         freezer = task_freezer(task);
 213         rcu_read_unlock();
 214
 215         /*
 216          * The root cgroup is non-freezable, so we can skip the
 217          * following check.
 218          */
 219         if (!freezer->css.cgroup->parent)
 220                 return;
 221
 222         spin_lock_irq(&freezer->lock);
 223         BUG_ON(freezer->state == CGROUP_FROZEN);
 224
 225         /* Locking avoids race with FREEZING -> THAWED transitions. */
 226         if (freezer->state == CGROUP_FREEZING)
 227                 freeze_task(task, true);
 228         spin_unlock_irq(&freezer->lock);
 229 }
 230
 231 /*
 232  * caller must hold freezer->lock
 233  */
 234 static void update_freezer_state(struct cgroup *cgroup,
 235                                  struct freezer *freezer)
 236 {
 237         struct cgroup_iter it;
 238         struct task_struct *task;
 239         unsigned int nfrozen = 0, ntotal = 0;
 240
 241         cgroup_iter_start(cgroup, &it);
 242         while ((task = cgroup_iter_next(cgroup, &it))) {
 243                 ntotal++;
 244                 if (is_task_frozen_enough(task))
 245                         nfrozen++;
 246         }
 247
 248         /*
 249          * Transition to FROZEN when no new tasks can be added ensures
 250          * that we never exist in the FROZEN state while there are unfrozen
 251          * tasks.
 252          */
 253         if (nfrozen == ntotal)
 254                 freezer->state = CGROUP_FROZEN;
 255         else if (nfrozen > 0)
 256                 freezer->state = CGROUP_FREEZING;
 257         else
 258                 freezer->state = CGROUP_THAWED;
 259         cgroup_iter_end(cgroup, &it);
 260 }
 261
 262 static int freezer_read(struct cgroup *cgroup, struct cftype *cft,
 263                         struct seq_file *m)
 264 {
 265         struct freezer *freezer;
 266         enum freezer_state state;
 267
 268         if (!cgroup_lock_live_group(cgroup))
 269                 return -ENODEV;
 270
 271         freezer = cgroup_freezer(cgroup);
 272         spin_lock_irq(&freezer->lock);
 273         state = freezer->state;
 274         if (state == CGROUP_FREEZING) {
 275                 /* We change from FREEZING to FROZEN lazily if the cgroup was
 276                  * only partially frozen when we exitted write. */
 277                 update_freezer_state(cgroup, freezer);
 278                 state = freezer->state;
 279         }
 280         spin_unlock_irq(&freezer->lock);
 281         cgroup_unlock();
 282
 283         seq_puts(m, freezer_state_strs[state]);
 284         seq_putc(m, '\n');
 285         return 0;
 286 }
 287
 288 static int try_to_freeze_cgroup(struct cgroup *cgroup, struct freezer *freezer)
 289 {
 290         struct cgroup_iter it;
 291         struct task_struct *task;
 292         unsigned int num_cant_freeze_now = 0;
 293
 294         freezer->state = CGROUP_FREEZING;
 295         cgroup_iter_start(cgroup, &it);
 296         while ((task = cgroup_iter_next(cgroup, &it))) {
 297                 if (!freeze_task(task, true))
 298                         continue;
 299                 if (is_task_frozen_enough(task))
 300                         continue;
 301                 if (!freezing(task) && !freezer_should_skip(task))
 302                         num_cant_freeze_now++;
 303         }
 304         cgroup_iter_end(cgroup, &it);
 305
 306         return num_cant_freeze_now ? -EBUSY : 0;
 307 }
 308
 309 static void unfreeze_cgroup(struct cgroup *cgroup, struct freezer *freezer)
 310 {
 311         struct cgroup_iter it;
 312         struct task_struct *task;
 313
 314         cgroup_iter_start(cgroup, &it);
 315         while ((task = cgroup_iter_next(cgroup, &it))) {
 316                 thaw_process(task);
 317         }
 318         cgroup_iter_end(cgroup, &it);
 319
 320         freezer->state = CGROUP_THAWED;
 321 }
 322
 323 static int freezer_change_state(struct cgroup *cgroup,
 324                                 enum freezer_state goal_state)
 325 {
 326         struct freezer *freezer;
 327         int retval = 0;
 328
 329         freezer = cgroup_freezer(cgroup);
 330
 331         spin_lock_irq(&freezer->lock);
 332
 333         update_freezer_state(cgroup, freezer);
 334         if (goal_state == freezer->state)
 335                 goto out;
 336
 337         switch (goal_state) {
 338         case CGROUP_THAWED:
 339                 unfreeze_cgroup(cgroup, freezer);
 340                 break;
 341         case CGROUP_FROZEN:
 342                 retval = try_to_freeze_cgroup(cgroup, freezer);
 343                 break;
 344         default:
 345                 BUG();
 346         }
 347 out:
 348         spin_unlock_irq(&freezer->lock);
 349
 350         return retval;
 351 }
 352
 353 static int freezer_write(struct cgroup *cgroup,
 354                          struct cftype *cft,
 355                          const char *buffer)
 356 {
 357         int retval;
 358         enum freezer_state goal_state;
 359
 360         if (strcmp(buffer, freezer_state_strs[CGROUP_THAWED]) == 0)
 361                 goal_state = CGROUP_THAWED;
 362         else if (strcmp(buffer, freezer_state_strs[CGROUP_FROZEN]) == 0)
 363                 goal_state = CGROUP_FROZEN;
 364         else
 365                 return -EINVAL;
 366
 367         if (!cgroup_lock_live_group(cgroup))
 368                 return -ENODEV;
 369         retval = freezer_change_state(cgroup, goal_state);
 370         cgroup_unlock();
 371         return retval;
 372 }
 373
 374 static struct cftype files[] = {
 375         {
 376                 .name = "state",
 377                 .read_seq_string = freezer_read,
 378                 .write_string = freezer_write,
 379         },
 380 };
 381
 382 static int freezer_populate(struct cgroup_subsys *ss, struct cgroup *cgroup)
 383 {
 384         if (!cgroup->parent)
 385                 return 0;
 386         return cgroup_add_files(cgroup, ss, files, ARRAY_SIZE(files));
 387 }
 388
 389 struct cgroup_subsys freezer_subsys = {
 390         .name           = "freezer",
 391         .create         = freezer_create,
 392         .destroy        = freezer_destroy,
 393         .populate       = freezer_populate,
 394         .subsys_id      = freezer_subsys_id,
 395         .can_attach     = freezer_can_attach,
 396         .attach         = NULL,
 397         .fork           = freezer_fork,
 398         .exit           = NULL,
 399 };