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 /*
  26  * A cgroup is freezing if any FREEZING flags are set.  FREEZING_SELF is
  27  * set if "FROZEN" is written to freezer.state cgroupfs file, and cleared
  28  * for "THAWED".  FREEZING_PARENT is set if the parent freezer is FREEZING
  29  * for whatever reason.  IOW, a cgroup has FREEZING_PARENT set if one of
  30  * its ancestors has FREEZING_SELF set.
  31  */
  32 enum freezer_state_flags {
  33         CGROUP_FREEZER_ONLINE   = (1 << 0), /* freezer is fully online */
  34         CGROUP_FREEZING_SELF    = (1 << 1), /* this freezer is freezing */
  35         CGROUP_FREEZING_PARENT  = (1 << 2), /* the parent freezer is freezing */
  36         CGROUP_FROZEN           = (1 << 3), /* this and its descendants frozen */
  37
  38         /* mask for all FREEZING flags */
  39         CGROUP_FREEZING         = CGROUP_FREEZING_SELF | CGROUP_FREEZING_PARENT,
  40 };
  41
  42 struct freezer {
  43         struct cgroup_subsys_state      css;
  44         unsigned int                    state;
  45         spinlock_t                      lock;
  46 };
  47
  48 static inline struct freezer *css_freezer(struct cgroup_subsys_state *css)
  49 {
  50         return css ? container_of(css, struct freezer, css) : NULL;
  51 }
  52
  53 static inline struct freezer *task_freezer(struct task_struct *task)
  54 {
  55         return css_freezer(task_css(task, freezer_cgrp_id));
  56 }
  57
  58 static struct freezer *parent_freezer(struct freezer *freezer)
  59 {
  60         return css_freezer(css_parent(&freezer->css));
  61 }
  62
  63 bool cgroup_freezing(struct task_struct *task)
  64 {
  65         bool ret;
  66
  67         rcu_read_lock();
  68         ret = task_freezer(task)->state & CGROUP_FREEZING;
  69         rcu_read_unlock();
  70
  71         return ret;
  72 }
  73
  74 /*
  75  * cgroups_write_string() limits the size of freezer state strings to
  76  * CGROUP_LOCAL_BUFFER_SIZE
  77  */
  78 static const char *freezer_state_strs(unsigned int state)
  79 {
  80         if (state & CGROUP_FROZEN)
  81                 return "FROZEN";
  82         if (state & CGROUP_FREEZING)
  83                 return "FREEZING";
  84         return "THAWED";
  85 };
  86
  87 static struct cgroup_subsys_state *
  88 freezer_css_alloc(struct cgroup_subsys_state *parent_css)
  89 {
  90         struct freezer *freezer;
  91
  92         freezer = kzalloc(sizeof(struct freezer), GFP_KERNEL);
  93         if (!freezer)
  94                 return ERR_PTR(-ENOMEM);
  95
  96         spin_lock_init(&freezer->lock);
  97         return &freezer->css;
  98 }
  99
 100 /**
 101  * freezer_css_online - commit creation of a freezer css
 102  * @css: css being created
 103  *
 104  * We're committing to creation of @css.  Mark it online and inherit
 105  * parent's freezing state while holding both parent's and our
 106  * freezer->lock.
 107  */
 108 static int freezer_css_online(struct cgroup_subsys_state *css)
 109 {
 110         struct freezer *freezer = css_freezer(css);
 111         struct freezer *parent = parent_freezer(freezer);
 112
 113         /*
 114          * The following double locking and freezing state inheritance
 115          * guarantee that @cgroup can never escape ancestors' freezing
 116          * states.  See css_for_each_descendant_pre() for details.
 117          */
 118         if (parent)
 119                 spin_lock_irq(&parent->lock);
 120         spin_lock_nested(&freezer->lock, SINGLE_DEPTH_NESTING);
 121
 122         freezer->state |= CGROUP_FREEZER_ONLINE;
 123
 124         if (parent && (parent->state & CGROUP_FREEZING)) {
 125                 freezer->state |= CGROUP_FREEZING_PARENT | CGROUP_FROZEN;
 126                 atomic_inc(&system_freezing_cnt);
 127         }
 128
 129         spin_unlock(&freezer->lock);
 130         if (parent)
 131                 spin_unlock_irq(&parent->lock);
 132
 133         return 0;
 134 }
 135
 136 /**
 137  * freezer_css_offline - initiate destruction of a freezer css
 138  * @css: css being destroyed
 139  *
 140  * @css is going away.  Mark it dead and decrement system_freezing_count if
 141  * it was holding one.
 142  */
 143 static void freezer_css_offline(struct cgroup_subsys_state *css)
 144 {
 145         struct freezer *freezer = css_freezer(css);
 146
 147         spin_lock_irq(&freezer->lock);
 148
 149         if (freezer->state & CGROUP_FREEZING)
 150                 atomic_dec(&system_freezing_cnt);
 151
 152         freezer->state = 0;
 153
 154         spin_unlock_irq(&freezer->lock);
 155 }
 156
 157 static void freezer_css_free(struct cgroup_subsys_state *css)
 158 {
 159         kfree(css_freezer(css));
 160 }
 161
 162 /*
 163  * Tasks can be migrated into a different freezer anytime regardless of its
 164  * current state.  freezer_attach() is responsible for making new tasks
 165  * conform to the current state.
 166  *
 167  * Freezer state changes and task migration are synchronized via
 168  * @freezer->lock.  freezer_attach() makes the new tasks conform to the
 169  * current state and all following state changes can see the new tasks.
 170  */
 171 static void freezer_attach(struct cgroup_subsys_state *new_css,
 172                            struct cgroup_taskset *tset)
 173 {
 174         struct freezer *freezer = css_freezer(new_css);
 175         struct task_struct *task;
 176         bool clear_frozen = false;
 177
 178         spin_lock_irq(&freezer->lock);
 179
 180         /*
 181          * Make the new tasks conform to the current state of @new_css.
 182          * For simplicity, when migrating any task to a FROZEN cgroup, we
 183          * revert it to FREEZING and let update_if_frozen() determine the
 184          * correct state later.
 185          *
 186          * Tasks in @tset are on @new_css but may not conform to its
 187          * current state before executing the following - !frozen tasks may
 188          * be visible in a FROZEN cgroup and frozen tasks in a THAWED one.
 189          */
 190         cgroup_taskset_for_each(task, tset) {
 191                 if (!(freezer->state & CGROUP_FREEZING)) {
 192                         __thaw_task(task);
 193                 } else {
 194                         freeze_task(task);
 195                         freezer->state &= ~CGROUP_FROZEN;
 196                         clear_frozen = true;
 197                 }
 198         }
 199
 200         spin_unlock_irq(&freezer->lock);
 201
 202         /*
 203          * Propagate FROZEN clearing upwards.  We may race with
 204          * update_if_frozen(), but as long as both work bottom-up, either
 205          * update_if_frozen() sees child's FROZEN cleared or we clear the
 206          * parent's FROZEN later.  No parent w/ !FROZEN children can be
 207          * left FROZEN.
 208          */
 209         while (clear_frozen && (freezer = parent_freezer(freezer))) {
 210                 spin_lock_irq(&freezer->lock);
 211                 freezer->state &= ~CGROUP_FROZEN;
 212                 clear_frozen = freezer->state & CGROUP_FREEZING;
 213                 spin_unlock_irq(&freezer->lock);
 214         }
 215 }
 216
 217 /**
 218  * freezer_fork - cgroup post fork callback
 219  * @task: a task which has just been forked
 220  *
 221  * @task has just been created and should conform to the current state of
 222  * the cgroup_freezer it belongs to.  This function may race against
 223  * freezer_attach().  Losing to freezer_attach() means that we don't have
 224  * to do anything as freezer_attach() will put @task into the appropriate
 225  * state.
 226  */
 227 static void freezer_fork(struct task_struct *task)
 228 {
 229         struct freezer *freezer;
 230
 231         rcu_read_lock();
 232         freezer = task_freezer(task);
 233
 234         /*
 235          * The root cgroup is non-freezable, so we can skip locking the
 236          * freezer.  This is safe regardless of race with task migration.
 237          * If we didn't race or won, skipping is obviously the right thing
 238          * to do.  If we lost and root is the new cgroup, noop is still the
 239          * right thing to do.
 240          */
 241         if (!parent_freezer(freezer))
 242                 goto out;
 243
 244         /*
 245          * Grab @freezer->lock and freeze @task after verifying @task still
 246          * belongs to @freezer and it's freezing.  The former is for the
 247          * case where we have raced against task migration and lost and
 248          * @task is already in a different cgroup which may not be frozen.
 249          * This isn't strictly necessary as freeze_task() is allowed to be
 250          * called spuriously but let's do it anyway for, if nothing else,
 251          * documentation.
 252          */
 253         spin_lock_irq(&freezer->lock);
 254         if (freezer == task_freezer(task) && (freezer->state & CGROUP_FREEZING))
 255                 freeze_task(task);
 256         spin_unlock_irq(&freezer->lock);
 257 out:
 258         rcu_read_unlock();
 259 }
 260
 261 /**
 262  * update_if_frozen - update whether a cgroup finished freezing
 263  * @css: css of interest
 264  *
 265  * Once FREEZING is initiated, transition to FROZEN is lazily updated by
 266  * calling this function.  If the current state is FREEZING but not FROZEN,
 267  * this function checks whether all tasks of this cgroup and the descendant
 268  * cgroups finished freezing and, if so, sets FROZEN.
 269  *
 270  * The caller is responsible for grabbing RCU read lock and calling
 271  * update_if_frozen() on all descendants prior to invoking this function.
 272  *
 273  * Task states and freezer state might disagree while tasks are being
 274  * migrated into or out of @css, so we can't verify task states against
 275  * @freezer state here.  See freezer_attach() for details.
 276  */
 277 static void update_if_frozen(struct cgroup_subsys_state *css)
 278 {
 279         struct freezer *freezer = css_freezer(css);
 280         struct cgroup_subsys_state *pos;
 281         struct css_task_iter it;
 282         struct task_struct *task;
 283
 284         WARN_ON_ONCE(!rcu_read_lock_held());
 285
 286         spin_lock_irq(&freezer->lock);
 287
 288         if (!(freezer->state & CGROUP_FREEZING) ||
 289             (freezer->state & CGROUP_FROZEN))
 290                 goto out_unlock;
 291
 292         /* are all (live) children frozen? */
 293         css_for_each_child(pos, css) {
 294                 struct freezer *child = css_freezer(pos);
 295
 296                 if ((child->state & CGROUP_FREEZER_ONLINE) &&
 297                     !(child->state & CGROUP_FROZEN))
 298                         goto out_unlock;
 299         }
 300
 301         /* are all tasks frozen? */
 302         css_task_iter_start(css, &it);
 303
 304         while ((task = css_task_iter_next(&it))) {
 305                 if (freezing(task)) {
 306                         /*
 307                          * freezer_should_skip() indicates that the task
 308                          * should be skipped when determining freezing
 309                          * completion.  Consider it frozen in addition to
 310                          * the usual frozen condition.
 311                          */
 312                         if (!frozen(task) && !freezer_should_skip(task))
 313                                 goto out_iter_end;
 314                 }
 315         }
 316
 317         freezer->state |= CGROUP_FROZEN;
 318 out_iter_end:
 319         css_task_iter_end(&it);
 320 out_unlock:
 321         spin_unlock_irq(&freezer->lock);
 322 }
 323
 324 static int freezer_read(struct seq_file *m, void *v)
 325 {
 326         struct cgroup_subsys_state *css = seq_css(m), *pos;
 327
 328         rcu_read_lock();
 329
 330         /* update states bottom-up */
 331         css_for_each_descendant_post(pos, css)
 332                 update_if_frozen(pos);
 333
 334         rcu_read_unlock();
 335
 336         seq_puts(m, freezer_state_strs(css_freezer(css)->state));
 337         seq_putc(m, '\n');
 338         return 0;
 339 }
 340
 341 static void freeze_cgroup(struct freezer *freezer)
 342 {
 343         struct css_task_iter it;
 344         struct task_struct *task;
 345
 346         css_task_iter_start(&freezer->css, &it);
 347         while ((task = css_task_iter_next(&it)))
 348                 freeze_task(task);
 349         css_task_iter_end(&it);
 350 }
 351
 352 static void unfreeze_cgroup(struct freezer *freezer)
 353 {
 354         struct css_task_iter it;
 355         struct task_struct *task;
 356
 357         css_task_iter_start(&freezer->css, &it);
 358         while ((task = css_task_iter_next(&it)))
 359                 __thaw_task(task);
 360         css_task_iter_end(&it);
 361 }
 362
 363 /**
 364  * freezer_apply_state - apply state change to a single cgroup_freezer
 365  * @freezer: freezer to apply state change to
 366  * @freeze: whether to freeze or unfreeze
 367  * @state: CGROUP_FREEZING_* flag to set or clear
 368  *
 369  * Set or clear @state on @cgroup according to @freeze, and perform
 370  * freezing or thawing as necessary.
 371  */
 372 static void freezer_apply_state(struct freezer *freezer, bool freeze,
 373                                 unsigned int state)
 374 {
 375         /* also synchronizes against task migration, see freezer_attach() */
 376         lockdep_assert_held(&freezer->lock);
 377
 378         if (!(freezer->state & CGROUP_FREEZER_ONLINE))
 379                 return;
 380
 381         if (freeze) {
 382                 if (!(freezer->state & CGROUP_FREEZING))
 383                         atomic_inc(&system_freezing_cnt);
 384                 freezer->state |= state;
 385                 freeze_cgroup(freezer);
 386         } else {
 387                 bool was_freezing = freezer->state & CGROUP_FREEZING;
 388
 389                 freezer->state &= ~state;
 390
 391                 if (!(freezer->state & CGROUP_FREEZING)) {
 392                         if (was_freezing)
 393                                 atomic_dec(&system_freezing_cnt);
 394                         freezer->state &= ~CGROUP_FROZEN;
 395                         unfreeze_cgroup(freezer);
 396                 }
 397         }
 398 }
 399
 400 /**
 401  * freezer_change_state - change the freezing state of a cgroup_freezer
 402  * @freezer: freezer of interest
 403  * @freeze: whether to freeze or thaw
 404  *
 405  * Freeze or thaw @freezer according to @freeze.  The operations are
 406  * recursive - all descendants of @freezer will be affected.
 407  */
 408 static void freezer_change_state(struct freezer *freezer, bool freeze)
 409 {
 410         struct cgroup_subsys_state *pos;
 411
 412         /*
 413          * Update all its descendants in pre-order traversal.  Each
 414          * descendant will try to inherit its parent's FREEZING state as
 415          * CGROUP_FREEZING_PARENT.
 416          */
 417         rcu_read_lock();
 418         css_for_each_descendant_pre(pos, &freezer->css) {
 419                 struct freezer *pos_f = css_freezer(pos);
 420                 struct freezer *parent = parent_freezer(pos_f);
 421
 422                 spin_lock_irq(&pos_f->lock);
 423
 424                 if (pos_f == freezer) {
 425                         freezer_apply_state(pos_f, freeze,
 426                                             CGROUP_FREEZING_SELF);
 427                 } else {
 428                         /*
 429                          * Our update to @parent->state is already visible
 430                          * which is all we need.  No need to lock @parent.
 431                          * For more info on synchronization, see
 432                          * freezer_post_create().
 433                          */
 434                         freezer_apply_state(pos_f,
 435                                             parent->state & CGROUP_FREEZING,
 436                                             CGROUP_FREEZING_PARENT);
 437                 }
 438
 439                 spin_unlock_irq(&pos_f->lock);
 440         }
 441         rcu_read_unlock();
 442 }
 443
 444 static int freezer_write(struct cgroup_subsys_state *css, struct cftype *cft,
 445                          char *buffer)
 446 {
 447         bool freeze;
 448
 449         if (strcmp(buffer, freezer_state_strs(0)) == 0)
 450                 freeze = false;
 451         else if (strcmp(buffer, freezer_state_strs(CGROUP_FROZEN)) == 0)
 452                 freeze = true;
 453         else
 454                 return -EINVAL;
 455
 456         freezer_change_state(css_freezer(css), freeze);
 457         return 0;
 458 }
 459
 460 static u64 freezer_self_freezing_read(struct cgroup_subsys_state *css,
 461                                       struct cftype *cft)
 462 {
 463         struct freezer *freezer = css_freezer(css);
 464
 465         return (bool)(freezer->state & CGROUP_FREEZING_SELF);
 466 }
 467
 468 static u64 freezer_parent_freezing_read(struct cgroup_subsys_state *css,
 469                                         struct cftype *cft)
 470 {
 471         struct freezer *freezer = css_freezer(css);
 472
 473         return (bool)(freezer->state & CGROUP_FREEZING_PARENT);
 474 }
 475
 476 static struct cftype files[] = {
 477         {
 478                 .name = "state",
 479                 .flags = CFTYPE_NOT_ON_ROOT,
 480                 .seq_show = freezer_read,
 481                 .write_string = freezer_write,
 482         },
 483         {
 484                 .name = "self_freezing",
 485                 .flags = CFTYPE_NOT_ON_ROOT,
 486                 .read_u64 = freezer_self_freezing_read,
 487         },
 488         {
 489                 .name = "parent_freezing",
 490                 .flags = CFTYPE_NOT_ON_ROOT,
 491                 .read_u64 = freezer_parent_freezing_read,
 492         },
 493         { }     /* terminate */
 494 };
 495
 496 struct cgroup_subsys freezer_cgrp_subsys = {
 497         .css_alloc      = freezer_css_alloc,
 498         .css_online     = freezer_css_online,
 499         .css_offline    = freezer_css_offline,
 500         .css_free       = freezer_css_free,
 501         .attach         = freezer_attach,
 502         .fork           = freezer_fork,
 503         .base_cftypes   = files,
 504 };