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_subsys *ss,
 132                                                   struct cgroup *cgroup)
 133 {
 134         struct freezer *freezer;
 135
 136         freezer = kzalloc(sizeof(struct freezer), GFP_KERNEL);
 137         if (!freezer)
 138                 return ERR_PTR(-ENOMEM);
 139
 140         spin_lock_init(&freezer->lock);
 141         freezer->state = CGROUP_THAWED;
 142         return &freezer->css;
 143 }
 144
 145 static void freezer_destroy(struct cgroup_subsys *ss,
 146                             struct cgroup *cgroup)
 147 {
 148         struct freezer *freezer = cgroup_freezer(cgroup);
 149
 150         if (freezer->state != CGROUP_THAWED)
 151                 atomic_dec(&system_freezing_cnt);
 152         kfree(freezer);
 153 }
 154
 155 /* task is frozen or will freeze immediately when next it gets woken */
 156 static bool is_task_frozen_enough(struct task_struct *task)
 157 {
 158         return frozen(task) ||
 159                 (task_is_stopped_or_traced(task) && freezing(task));
 160 }
 161
 162 /*
 163  * The call to cgroup_lock() in the freezer.state write method prevents
 164  * a write to that file racing against an attach, and hence the
 165  * can_attach() result will remain valid until the attach completes.
 166  */
 167 static int freezer_can_attach(struct cgroup_subsys *ss,
 168                               struct cgroup *new_cgroup,
 169                               struct task_struct *task)
 170 {
 171         struct freezer *freezer;
 172
 173         /*
 174          * Anything frozen can't move or be moved to/from.
 175          */
 176
 177         freezer = cgroup_freezer(new_cgroup);
 178         if (freezer->state != CGROUP_THAWED)
 179                 return -EBUSY;
 180
 181         return 0;
 182 }
 183
 184 static int freezer_can_attach_task(struct cgroup *cgrp, struct task_struct *tsk)
 185 {
 186         return cgroup_freezing(tsk) ? -EBUSY : 0;
 187 }
 188
 189 static void freezer_fork(struct cgroup_subsys *ss, struct task_struct *task)
 190 {
 191         struct freezer *freezer;
 192
 193         /*
 194          * No lock is needed, since the task isn't on tasklist yet,
 195          * so it can't be moved to another cgroup, which means the
 196          * freezer won't be removed and will be valid during this
 197          * function call.  Nevertheless, apply RCU read-side critical
 198          * section to suppress RCU lockdep false positives.
 199          */
 200         rcu_read_lock();
 201         freezer = task_freezer(task);
 202         rcu_read_unlock();
 203
 204         /*
 205          * The root cgroup is non-freezable, so we can skip the
 206          * following check.
 207          */
 208         if (!freezer->css.cgroup->parent)
 209                 return;
 210
 211         spin_lock_irq(&freezer->lock);
 212         BUG_ON(freezer->state == CGROUP_FROZEN);
 213
 214         /* Locking avoids race with FREEZING -> THAWED transitions. */
 215         if (freezer->state == CGROUP_FREEZING)
 216                 freeze_task(task);
 217         spin_unlock_irq(&freezer->lock);
 218 }
 219
 220 /*
 221  * caller must hold freezer->lock
 222  */
 223 static void update_if_frozen(struct cgroup *cgroup,
 224                                  struct freezer *freezer)
 225 {
 226         struct cgroup_iter it;
 227         struct task_struct *task;
 228         unsigned int nfrozen = 0, ntotal = 0;
 229         enum freezer_state old_state = freezer->state;
 230
 231         cgroup_iter_start(cgroup, &it);
 232         while ((task = cgroup_iter_next(cgroup, &it))) {
 233                 ntotal++;
 234                 if (freezing(task) && is_task_frozen_enough(task))
 235                         nfrozen++;
 236         }
 237
 238         if (old_state == CGROUP_THAWED) {
 239                 BUG_ON(nfrozen > 0);
 240         } else if (old_state == CGROUP_FREEZING) {
 241                 if (nfrozen == ntotal)
 242                         freezer->state = CGROUP_FROZEN;
 243         } else { /* old_state == CGROUP_FROZEN */
 244                 BUG_ON(nfrozen != ntotal);
 245         }
 246
 247         cgroup_iter_end(cgroup, &it);
 248 }
 249
 250 static int freezer_read(struct cgroup *cgroup, struct cftype *cft,
 251                         struct seq_file *m)
 252 {
 253         struct freezer *freezer;
 254         enum freezer_state state;
 255
 256         if (!cgroup_lock_live_group(cgroup))
 257                 return -ENODEV;
 258
 259         freezer = cgroup_freezer(cgroup);
 260         spin_lock_irq(&freezer->lock);
 261         state = freezer->state;
 262         if (state == CGROUP_FREEZING) {
 263                 /* We change from FREEZING to FROZEN lazily if the cgroup was
 264                  * only partially frozen when we exitted write. */
 265                 update_if_frozen(cgroup, freezer);
 266                 state = freezer->state;
 267         }
 268         spin_unlock_irq(&freezer->lock);
 269         cgroup_unlock();
 270
 271         seq_puts(m, freezer_state_strs[state]);
 272         seq_putc(m, '\n');
 273         return 0;
 274 }
 275
 276 static int try_to_freeze_cgroup(struct cgroup *cgroup, struct freezer *freezer)
 277 {
 278         struct cgroup_iter it;
 279         struct task_struct *task;
 280         unsigned int num_cant_freeze_now = 0;
 281
 282         cgroup_iter_start(cgroup, &it);
 283         while ((task = cgroup_iter_next(cgroup, &it))) {
 284                 if (!freeze_task(task))
 285                         continue;
 286                 if (is_task_frozen_enough(task))
 287                         continue;
 288                 if (!freezing(task) && !freezer_should_skip(task))
 289                         num_cant_freeze_now++;
 290         }
 291         cgroup_iter_end(cgroup, &it);
 292
 293         return num_cant_freeze_now ? -EBUSY : 0;
 294 }
 295
 296 static void unfreeze_cgroup(struct cgroup *cgroup, struct freezer *freezer)
 297 {
 298         struct cgroup_iter it;
 299         struct task_struct *task;
 300
 301         cgroup_iter_start(cgroup, &it);
 302         while ((task = cgroup_iter_next(cgroup, &it)))
 303                 __thaw_task(task);
 304         cgroup_iter_end(cgroup, &it);
 305 }
 306
 307 static int freezer_change_state(struct cgroup *cgroup,
 308                                 enum freezer_state goal_state)
 309 {
 310         struct freezer *freezer;
 311         int retval = 0;
 312
 313         freezer = cgroup_freezer(cgroup);
 314
 315         spin_lock_irq(&freezer->lock);
 316
 317         update_if_frozen(cgroup, freezer);
 318
 319         switch (goal_state) {
 320         case CGROUP_THAWED:
 321                 if (freezer->state != CGROUP_THAWED)
 322                         atomic_dec(&system_freezing_cnt);
 323                 freezer->state = CGROUP_THAWED;
 324                 unfreeze_cgroup(cgroup, freezer);
 325                 break;
 326         case CGROUP_FROZEN:
 327                 if (freezer->state == CGROUP_THAWED)
 328                         atomic_inc(&system_freezing_cnt);
 329                 freezer->state = CGROUP_FREEZING;
 330                 retval = try_to_freeze_cgroup(cgroup, freezer);
 331                 break;
 332         default:
 333                 BUG();
 334         }
 335
 336         spin_unlock_irq(&freezer->lock);
 337
 338         return retval;
 339 }
 340
 341 static int freezer_write(struct cgroup *cgroup,
 342                          struct cftype *cft,
 343                          const char *buffer)
 344 {
 345         int retval;
 346         enum freezer_state goal_state;
 347
 348         if (strcmp(buffer, freezer_state_strs[CGROUP_THAWED]) == 0)
 349                 goal_state = CGROUP_THAWED;
 350         else if (strcmp(buffer, freezer_state_strs[CGROUP_FROZEN]) == 0)
 351                 goal_state = CGROUP_FROZEN;
 352         else
 353                 return -EINVAL;
 354
 355         if (!cgroup_lock_live_group(cgroup))
 356                 return -ENODEV;
 357         retval = freezer_change_state(cgroup, goal_state);
 358         cgroup_unlock();
 359         return retval;
 360 }
 361
 362 static struct cftype files[] = {
 363         {
 364                 .name = "state",
 365                 .read_seq_string = freezer_read,
 366                 .write_string = freezer_write,
 367         },
 368 };
 369
 370 static int freezer_populate(struct cgroup_subsys *ss, struct cgroup *cgroup)
 371 {
 372         if (!cgroup->parent)
 373                 return 0;
 374         return cgroup_add_files(cgroup, ss, files, ARRAY_SIZE(files));
 375 }
 376
 377 struct cgroup_subsys freezer_subsys = {
 378         .name           = "freezer",
 379         .create         = freezer_create,
 380         .destroy        = freezer_destroy,
 381         .populate       = freezer_populate,
 382         .subsys_id      = freezer_subsys_id,
 383         .can_attach     = freezer_can_attach,
 384         .can_attach_task = freezer_can_attach_task,
 385         .pre_attach     = NULL,
 386         .attach_task    = NULL,
 387         .attach         = NULL,
 388         .fork           = freezer_fork,
 389         .exit           = NULL,
 390 };