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