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 /*
 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 cgroup_taskset *tset)
 163 {
 164         struct freezer *freezer;
 165         struct task_struct *task;
 166
 167         /*
 168          * Anything frozen can't move or be moved to/from.
 169          */
 170         cgroup_taskset_for_each(task, new_cgroup, tset)
 171                 if (cgroup_freezing(task))
 172                         return -EBUSY;
 173
 174         freezer = cgroup_freezer(new_cgroup);
 175         if (freezer->state != CGROUP_THAWED)
 176                 return -EBUSY;
 177
 178         return 0;
 179 }
 180
 181 static void freezer_fork(struct cgroup_subsys *ss, struct task_struct *task)
 182 {
 183         struct freezer *freezer;
 184
 185         /*
 186          * No lock is needed, since the task isn't on tasklist yet,
 187          * so it can't be moved to another cgroup, which means the
 188          * freezer won't be removed and will be valid during this
 189          * function call.  Nevertheless, apply RCU read-side critical
 190          * section to suppress RCU lockdep false positives.
 191          */
 192         rcu_read_lock();
 193         freezer = task_freezer(task);
 194         rcu_read_unlock();
 195
 196         /*
 197          * The root cgroup is non-freezable, so we can skip the
 198          * following check.
 199          */
 200         if (!freezer->css.cgroup->parent)
 201                 return;
 202
 203         spin_lock_irq(&freezer->lock);
 204         BUG_ON(freezer->state == CGROUP_FROZEN);
 205
 206         /* Locking avoids race with FREEZING -> THAWED transitions. */
 207         if (freezer->state == CGROUP_FREEZING)
 208                 freeze_task(task);
 209         spin_unlock_irq(&freezer->lock);
 210 }
 211
 212 /*
 213  * caller must hold freezer->lock
 214  */
 215 static void update_if_frozen(struct cgroup *cgroup,
 216                                  struct freezer *freezer)
 217 {
 218         struct cgroup_iter it;
 219         struct task_struct *task;
 220         unsigned int nfrozen = 0, ntotal = 0;
 221         enum freezer_state old_state = freezer->state;
 222
 223         cgroup_iter_start(cgroup, &it);
 224         while ((task = cgroup_iter_next(cgroup, &it))) {
 225                 ntotal++;
 226                 if (freezing(task) && frozen(task))
 227                         nfrozen++;
 228         }
 229
 230         if (old_state == CGROUP_THAWED) {
 231                 BUG_ON(nfrozen > 0);
 232         } else if (old_state == CGROUP_FREEZING) {
 233                 if (nfrozen == ntotal)
 234                         freezer->state = CGROUP_FROZEN;
 235         } else { /* old_state == CGROUP_FROZEN */
 236                 BUG_ON(nfrozen != ntotal);
 237         }
 238
 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_if_frozen(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         cgroup_iter_start(cgroup, &it);
 275         while ((task = cgroup_iter_next(cgroup, &it))) {
 276                 if (!freeze_task(task))
 277                         continue;
 278                 if (frozen(task))
 279                         continue;
 280                 if (!freezing(task) && !freezer_should_skip(task))
 281                         num_cant_freeze_now++;
 282         }
 283         cgroup_iter_end(cgroup, &it);
 284
 285         return num_cant_freeze_now ? -EBUSY : 0;
 286 }
 287
 288 static void unfreeze_cgroup(struct cgroup *cgroup, struct freezer *freezer)
 289 {
 290         struct cgroup_iter it;
 291         struct task_struct *task;
 292
 293         cgroup_iter_start(cgroup, &it);
 294         while ((task = cgroup_iter_next(cgroup, &it)))
 295                 __thaw_task(task);
 296         cgroup_iter_end(cgroup, &it);
 297 }
 298
 299 static int freezer_change_state(struct cgroup *cgroup,
 300                                 enum freezer_state goal_state)
 301 {
 302         struct freezer *freezer;
 303         int retval = 0;
 304
 305         freezer = cgroup_freezer(cgroup);
 306
 307         spin_lock_irq(&freezer->lock);
 308
 309         update_if_frozen(cgroup, freezer);
 310
 311         switch (goal_state) {
 312         case CGROUP_THAWED:
 313                 if (freezer->state != CGROUP_THAWED)
 314                         atomic_dec(&system_freezing_cnt);
 315                 freezer->state = CGROUP_THAWED;
 316                 unfreeze_cgroup(cgroup, freezer);
 317                 break;
 318         case CGROUP_FROZEN:
 319                 if (freezer->state == CGROUP_THAWED)
 320                         atomic_inc(&system_freezing_cnt);
 321                 freezer->state = CGROUP_FREEZING;
 322                 retval = try_to_freeze_cgroup(cgroup, freezer);
 323                 break;
 324         default:
 325                 BUG();
 326         }
 327
 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         .fork           = freezer_fork,
 377 };