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