blockjob.c

   1 /*
   2  * QEMU System Emulator block driver
   3  *
   4  * Copyright (c) 2011 IBM Corp.
   5  * Copyright (c) 2012 Red Hat, Inc.
   6  *
   7  * Permission is hereby granted, free of charge, to any person obtaining a copy
   8  * of this software and associated documentation files (the "Software"), to deal
   9  * in the Software without restriction, including without limitation the rights
  10  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
  11  * copies of the Software, and to permit persons to whom the Software is
  12  * furnished to do so, subject to the following conditions:
  13  *
  14  * The above copyright notice and this permission notice shall be included in
  15  * all copies or substantial portions of the Software.
  16  *
  17  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  18  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  19  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
  20  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  21  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  22  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
  23  * THE SOFTWARE.
  24  */
  25
  26 #include "qemu/osdep.h"
  27 #include "block/aio-wait.h"
  28 #include "block/block.h"
  29 #include "block/blockjob_int.h"
  30 #include "block/block_int.h"
  31 #include "block/trace.h"
  32 #include "sysemu/block-backend.h"
  33 #include "qapi/error.h"
  34 #include "qapi/qapi-events-block-core.h"
  35 #include "qapi/qmp/qerror.h"
  36 #include "qemu/main-loop.h"
  37 #include "qemu/timer.h"
  38
  39 static bool is_block_job(Job *job)
  40 {
  41     return job_type(job) == JOB_TYPE_BACKUP ||
  42            job_type(job) == JOB_TYPE_COMMIT ||
  43            job_type(job) == JOB_TYPE_MIRROR ||
  44            job_type(job) == JOB_TYPE_STREAM;
  45 }
  46
  47 BlockJob *block_job_next_locked(BlockJob *bjob)
  48 {
  49     Job *job = bjob ? &bjob->job : NULL;
  50     GLOBAL_STATE_CODE();
  51
  52     do {
  53         job = job_next_locked(job);
  54     } while (job && !is_block_job(job));
  55
  56     return job ? container_of(job, BlockJob, job) : NULL;
  57 }
  58
  59 BlockJob *block_job_get_locked(const char *id)
  60 {
  61     Job *job = job_get_locked(id);
  62     GLOBAL_STATE_CODE();
  63
  64     if (job && is_block_job(job)) {
  65         return container_of(job, BlockJob, job);
  66     } else {
  67         return NULL;
  68     }
  69 }
  70
  71 BlockJob *block_job_get(const char *id)
  72 {
  73     JOB_LOCK_GUARD();
  74     return block_job_get_locked(id);
  75 }
  76
  77 void block_job_free(Job *job)
  78 {
  79     BlockJob *bjob = container_of(job, BlockJob, job);
  80     GLOBAL_STATE_CODE();
  81
  82     block_job_remove_all_bdrv(bjob);
  83     ratelimit_destroy(&bjob->limit);
  84     error_free(bjob->blocker);
  85 }
  86
  87 static char *child_job_get_parent_desc(BdrvChild *c)
  88 {
  89     BlockJob *job = c->opaque;
  90     return g_strdup_printf("%s job '%s'", job_type_str(&job->job), job->job.id);
  91 }
  92
  93 static void child_job_drained_begin(BdrvChild *c)
  94 {
  95     BlockJob *job = c->opaque;
  96     job_pause(&job->job);
  97 }
  98
  99 static bool child_job_drained_poll(BdrvChild *c)
 100 {
 101     BlockJob *bjob = c->opaque;
 102     Job *job = &bjob->job;
 103     const BlockJobDriver *drv = block_job_driver(bjob);
 104
 105     /* An inactive or completed job doesn't have any pending requests. Jobs
 106      * with !job->busy are either already paused or have a pause point after
 107      * being reentered, so no job driver code will run before they pause. */
 108     WITH_JOB_LOCK_GUARD() {
 109         if (!job->busy || job_is_completed_locked(job)) {
 110             return false;
 111         }
 112     }
 113
 114     /* Otherwise, assume that it isn't fully stopped yet, but allow the job to
 115      * override this assumption. */
 116     if (drv->drained_poll) {
 117         return drv->drained_poll(bjob);
 118     } else {
 119         return true;
 120     }
 121 }
 122
 123 static void child_job_drained_end(BdrvChild *c)
 124 {
 125     BlockJob *job = c->opaque;
 126     job_resume(&job->job);
 127 }
 128
 129 typedef struct BdrvStateChildJobContext {
 130     AioContext *new_ctx;
 131     BlockJob *job;
 132 } BdrvStateChildJobContext;
 133
 134 static void child_job_set_aio_ctx_commit(void *opaque)
 135 {
 136     BdrvStateChildJobContext *s = opaque;
 137     BlockJob *job = s->job;
 138
 139     job_set_aio_context(&job->job, s->new_ctx);
 140 }
 141
 142 static TransactionActionDrv change_child_job_context = {
 143     .commit = child_job_set_aio_ctx_commit,
 144     .clean = g_free,
 145 };
 146
 147 static bool child_job_change_aio_ctx(BdrvChild *c, AioContext *ctx,
 148                                      GHashTable *visited, Transaction *tran,
 149                                      Error **errp)
 150 {
 151     BlockJob *job = c->opaque;
 152     BdrvStateChildJobContext *s;
 153     GSList *l;
 154
 155     for (l = job->nodes; l; l = l->next) {
 156         BdrvChild *sibling = l->data;
 157         if (!bdrv_child_change_aio_context(sibling, ctx, visited,
 158                                            tran, errp)) {
 159             return false;
 160         }
 161     }
 162
 163     s = g_new(BdrvStateChildJobContext, 1);
 164     *s = (BdrvStateChildJobContext) {
 165         .new_ctx = ctx,
 166         .job = job,
 167     };
 168
 169     tran_add(tran, &change_child_job_context, s);
 170     return true;
 171 }
 172
 173 static AioContext *child_job_get_parent_aio_context(BdrvChild *c)
 174 {
 175     BlockJob *job = c->opaque;
 176     IO_CODE();
 177     JOB_LOCK_GUARD();
 178
 179     return job->job.aio_context;
 180 }
 181
 182 static const BdrvChildClass child_job = {
 183     .get_parent_desc    = child_job_get_parent_desc,
 184     .drained_begin      = child_job_drained_begin,
 185     .drained_poll       = child_job_drained_poll,
 186     .drained_end        = child_job_drained_end,
 187     .change_aio_ctx     = child_job_change_aio_ctx,
 188     .stay_at_node       = true,
 189     .get_parent_aio_context = child_job_get_parent_aio_context,
 190 };
 191
 192 void block_job_remove_all_bdrv(BlockJob *job)
 193 {
 194     GLOBAL_STATE_CODE();
 195     /*
 196      * bdrv_root_unref_child() may reach child_job_[can_]set_aio_ctx(),
 197      * which will also traverse job->nodes, so consume the list one by
 198      * one to make sure that such a concurrent access does not attempt
 199      * to process an already freed BdrvChild.
 200      */
 201     while (job->nodes) {
 202         GSList *l = job->nodes;
 203         BdrvChild *c = l->data;
 204
 205         job->nodes = l->next;
 206
 207         bdrv_op_unblock_all(c->bs, job->blocker);
 208         bdrv_root_unref_child(c);
 209
 210         g_slist_free_1(l);
 211     }
 212 }
 213
 214 bool block_job_has_bdrv(BlockJob *job, BlockDriverState *bs)
 215 {
 216     GSList *el;
 217     GLOBAL_STATE_CODE();
 218
 219     for (el = job->nodes; el; el = el->next) {
 220         BdrvChild *c = el->data;
 221         if (c->bs == bs) {
 222             return true;
 223         }
 224     }
 225
 226     return false;
 227 }
 228
 229 int block_job_add_bdrv(BlockJob *job, const char *name, BlockDriverState *bs,
 230                        uint64_t perm, uint64_t shared_perm, Error **errp)
 231 {
 232     BdrvChild *c;
 233     bool need_context_ops;
 234     GLOBAL_STATE_CODE();
 235
 236     bdrv_ref(bs);
 237
 238     need_context_ops = bdrv_get_aio_context(bs) != job->job.aio_context;
 239
 240     if (need_context_ops && job->job.aio_context != qemu_get_aio_context()) {
 241         aio_context_release(job->job.aio_context);
 242     }
 243     c = bdrv_root_attach_child(bs, name, &child_job, 0, perm, shared_perm, job,
 244                                errp);
 245     if (need_context_ops && job->job.aio_context != qemu_get_aio_context()) {
 246         aio_context_acquire(job->job.aio_context);
 247     }
 248     if (c == NULL) {
 249         return -EPERM;
 250     }
 251
 252     job->nodes = g_slist_prepend(job->nodes, c);
 253     bdrv_op_block_all(bs, job->blocker);
 254
 255     return 0;
 256 }
 257
 258 /* Called with job_mutex lock held. */
 259 static void block_job_on_idle_locked(Notifier *n, void *opaque)
 260 {
 261     aio_wait_kick();
 262 }
 263
 264 bool block_job_is_internal(BlockJob *job)
 265 {
 266     return (job->job.id == NULL);
 267 }
 268
 269 const BlockJobDriver *block_job_driver(BlockJob *job)
 270 {
 271     return container_of(job->job.driver, BlockJobDriver, job_driver);
 272 }
 273
 274 /* Assumes the job_mutex is held */
 275 static bool job_timer_pending(Job *job)
 276 {
 277     return timer_pending(&job->sleep_timer);
 278 }
 279
 280 bool block_job_set_speed_locked(BlockJob *job, int64_t speed, Error **errp)
 281 {
 282     const BlockJobDriver *drv = block_job_driver(job);
 283     int64_t old_speed = job->speed;
 284
 285     GLOBAL_STATE_CODE();
 286
 287     if (job_apply_verb_locked(&job->job, JOB_VERB_SET_SPEED, errp) < 0) {
 288         return false;
 289     }
 290     if (speed < 0) {
 291         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "speed",
 292                    "a non-negative value");
 293         return false;
 294     }
 295
 296     ratelimit_set_speed(&job->limit, speed, BLOCK_JOB_SLICE_TIME);
 297
 298     job->speed = speed;
 299
 300     if (drv->set_speed) {
 301         job_unlock();
 302         drv->set_speed(job, speed);
 303         job_lock();
 304     }
 305
 306     if (speed && speed <= old_speed) {
 307         return true;
 308     }
 309
 310     /* kick only if a timer is pending */
 311     job_enter_cond_locked(&job->job, job_timer_pending);
 312
 313     return true;
 314 }
 315
 316 static bool block_job_set_speed(BlockJob *job, int64_t speed, Error **errp)
 317 {
 318     JOB_LOCK_GUARD();
 319     return block_job_set_speed_locked(job, speed, errp);
 320 }
 321
 322 void block_job_ratelimit_processed_bytes(BlockJob *job, uint64_t n)
 323 {
 324     IO_CODE();
 325     ratelimit_calculate_delay(&job->limit, n);
 326 }
 327
 328 void block_job_ratelimit_sleep(BlockJob *job)
 329 {
 330     uint64_t delay_ns;
 331
 332     /*
 333      * Sleep at least once. If the job is reentered early, keep waiting until
 334      * we've waited for the full time that is necessary to keep the job at the
 335      * right speed.
 336      *
 337      * Make sure to recalculate the delay after each (possibly interrupted)
 338      * sleep because the speed can change while the job has yielded.
 339      */
 340     do {
 341         delay_ns = ratelimit_calculate_delay(&job->limit, 0);
 342         job_sleep_ns(&job->job, delay_ns);
 343     } while (delay_ns && !job_is_cancelled(&job->job));
 344 }
 345
 346 BlockJobInfo *block_job_query_locked(BlockJob *job, Error **errp)
 347 {
 348     BlockJobInfo *info;
 349     uint64_t progress_current, progress_total;
 350
 351     GLOBAL_STATE_CODE();
 352
 353     if (block_job_is_internal(job)) {
 354         error_setg(errp, "Cannot query QEMU internal jobs");
 355         return NULL;
 356     }
 357
 358     progress_get_snapshot(&job->job.progress, &progress_current,
 359                           &progress_total);
 360
 361     info = g_new0(BlockJobInfo, 1);
 362     info->type      = g_strdup(job_type_str(&job->job));
 363     info->device    = g_strdup(job->job.id);
 364     info->busy      = job->job.busy;
 365     info->paused    = job->job.pause_count > 0;
 366     info->offset    = progress_current;
 367     info->len       = progress_total;
 368     info->speed     = job->speed;
 369     info->io_status = job->iostatus;
 370     info->ready     = job_is_ready_locked(&job->job),
 371     info->status    = job->job.status;
 372     info->auto_finalize = job->job.auto_finalize;
 373     info->auto_dismiss  = job->job.auto_dismiss;
 374     if (job->job.ret) {
 375         info->error = job->job.err ?
 376                         g_strdup(error_get_pretty(job->job.err)) :
 377                         g_strdup(strerror(-job->job.ret));
 378     }
 379     return info;
 380 }
 381
 382 /* Called with job lock held */
 383 static void block_job_iostatus_set_err_locked(BlockJob *job, int error)
 384 {
 385     if (job->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
 386         job->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE :
 387                                           BLOCK_DEVICE_IO_STATUS_FAILED;
 388     }
 389 }
 390
 391 /* Called with job_mutex lock held. */
 392 static void block_job_event_cancelled_locked(Notifier *n, void *opaque)
 393 {
 394     BlockJob *job = opaque;
 395     uint64_t progress_current, progress_total;
 396
 397     if (block_job_is_internal(job)) {
 398         return;
 399     }
 400
 401     progress_get_snapshot(&job->job.progress, &progress_current,
 402                           &progress_total);
 403
 404     qapi_event_send_block_job_cancelled(job_type(&job->job),
 405                                         job->job.id,
 406                                         progress_total,
 407                                         progress_current,
 408                                         job->speed);
 409 }
 410
 411 /* Called with job_mutex lock held. */
 412 static void block_job_event_completed_locked(Notifier *n, void *opaque)
 413 {
 414     BlockJob *job = opaque;
 415     const char *msg = NULL;
 416     uint64_t progress_current, progress_total;
 417
 418     if (block_job_is_internal(job)) {
 419         return;
 420     }
 421
 422     if (job->job.ret < 0) {
 423         msg = error_get_pretty(job->job.err);
 424     }
 425
 426     progress_get_snapshot(&job->job.progress, &progress_current,
 427                           &progress_total);
 428
 429     qapi_event_send_block_job_completed(job_type(&job->job),
 430                                         job->job.id,
 431                                         progress_total,
 432                                         progress_current,
 433                                         job->speed,
 434                                         msg);
 435 }
 436
 437 /* Called with job_mutex lock held. */
 438 static void block_job_event_pending_locked(Notifier *n, void *opaque)
 439 {
 440     BlockJob *job = opaque;
 441
 442     if (block_job_is_internal(job)) {
 443         return;
 444     }
 445
 446     qapi_event_send_block_job_pending(job_type(&job->job),
 447                                       job->job.id);
 448 }
 449
 450 /* Called with job_mutex lock held. */
 451 static void block_job_event_ready_locked(Notifier *n, void *opaque)
 452 {
 453     BlockJob *job = opaque;
 454     uint64_t progress_current, progress_total;
 455
 456     if (block_job_is_internal(job)) {
 457         return;
 458     }
 459
 460     progress_get_snapshot(&job->job.progress, &progress_current,
 461                           &progress_total);
 462
 463     qapi_event_send_block_job_ready(job_type(&job->job),
 464                                     job->job.id,
 465                                     progress_total,
 466                                     progress_current,
 467                                     job->speed);
 468 }
 469
 470
 471 void *block_job_create(const char *job_id, const BlockJobDriver *driver,
 472                        JobTxn *txn, BlockDriverState *bs, uint64_t perm,
 473                        uint64_t shared_perm, int64_t speed, int flags,
 474                        BlockCompletionFunc *cb, void *opaque, Error **errp)
 475 {
 476     BlockJob *job;
 477     int ret;
 478     GLOBAL_STATE_CODE();
 479
 480     if (job_id == NULL && !(flags & JOB_INTERNAL)) {
 481         job_id = bdrv_get_device_name(bs);
 482     }
 483
 484     job = job_create(job_id, &driver->job_driver, txn, bdrv_get_aio_context(bs),
 485                      flags, cb, opaque, errp);
 486     if (job == NULL) {
 487         return NULL;
 488     }
 489
 490     assert(is_block_job(&job->job));
 491     assert(job->job.driver->free == &block_job_free);
 492     assert(job->job.driver->user_resume == &block_job_user_resume);
 493
 494     ratelimit_init(&job->limit);
 495
 496     job->finalize_cancelled_notifier.notify = block_job_event_cancelled_locked;
 497     job->finalize_completed_notifier.notify = block_job_event_completed_locked;
 498     job->pending_notifier.notify = block_job_event_pending_locked;
 499     job->ready_notifier.notify = block_job_event_ready_locked;
 500     job->idle_notifier.notify = block_job_on_idle_locked;
 501
 502     WITH_JOB_LOCK_GUARD() {
 503         notifier_list_add(&job->job.on_finalize_cancelled,
 504                           &job->finalize_cancelled_notifier);
 505         notifier_list_add(&job->job.on_finalize_completed,
 506                           &job->finalize_completed_notifier);
 507         notifier_list_add(&job->job.on_pending, &job->pending_notifier);
 508         notifier_list_add(&job->job.on_ready, &job->ready_notifier);
 509         notifier_list_add(&job->job.on_idle, &job->idle_notifier);
 510     }
 511
 512     error_setg(&job->blocker, "block device is in use by block job: %s",
 513                job_type_str(&job->job));
 514
 515     ret = block_job_add_bdrv(job, "main node", bs, perm, shared_perm, errp);
 516     if (ret < 0) {
 517         goto fail;
 518     }
 519
 520     bdrv_op_unblock(bs, BLOCK_OP_TYPE_DATAPLANE, job->blocker);
 521
 522     if (!block_job_set_speed(job, speed, errp)) {
 523         goto fail;
 524     }
 525
 526     return job;
 527
 528 fail:
 529     job_early_fail(&job->job);
 530     return NULL;
 531 }
 532
 533 void block_job_iostatus_reset_locked(BlockJob *job)
 534 {
 535     GLOBAL_STATE_CODE();
 536     if (job->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
 537         return;
 538     }
 539     assert(job->job.user_paused && job->job.pause_count > 0);
 540     job->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
 541 }
 542
 543 static void block_job_iostatus_reset(BlockJob *job)
 544 {
 545     JOB_LOCK_GUARD();
 546     block_job_iostatus_reset_locked(job);
 547 }
 548
 549 void block_job_user_resume(Job *job)
 550 {
 551     BlockJob *bjob = container_of(job, BlockJob, job);
 552     GLOBAL_STATE_CODE();
 553     block_job_iostatus_reset(bjob);
 554 }
 555
 556 BlockErrorAction block_job_error_action(BlockJob *job, BlockdevOnError on_err,
 557                                         int is_read, int error)
 558 {
 559     BlockErrorAction action;
 560     IO_CODE();
 561
 562     switch (on_err) {
 563     case BLOCKDEV_ON_ERROR_ENOSPC:
 564     case BLOCKDEV_ON_ERROR_AUTO:
 565         action = (error == ENOSPC) ?
 566                  BLOCK_ERROR_ACTION_STOP : BLOCK_ERROR_ACTION_REPORT;
 567         break;
 568     case BLOCKDEV_ON_ERROR_STOP:
 569         action = BLOCK_ERROR_ACTION_STOP;
 570         break;
 571     case BLOCKDEV_ON_ERROR_REPORT:
 572         action = BLOCK_ERROR_ACTION_REPORT;
 573         break;
 574     case BLOCKDEV_ON_ERROR_IGNORE:
 575         action = BLOCK_ERROR_ACTION_IGNORE;
 576         break;
 577     default:
 578         abort();
 579     }
 580     if (!block_job_is_internal(job)) {
 581         qapi_event_send_block_job_error(job->job.id,
 582                                         is_read ? IO_OPERATION_TYPE_READ :
 583                                         IO_OPERATION_TYPE_WRITE,
 584                                         action);
 585     }
 586     if (action == BLOCK_ERROR_ACTION_STOP) {
 587         WITH_JOB_LOCK_GUARD() {
 588             if (!job->job.user_paused) {
 589                 job_pause_locked(&job->job);
 590                 /*
 591                  * make the pause user visible, which will be
 592                  * resumed from QMP.
 593                  */
 594                 job->job.user_paused = true;
 595             }
 596             block_job_iostatus_set_err_locked(job, error);
 597         }
 598     }
 599     return action;
 600 }
 601
 602 AioContext *block_job_get_aio_context(BlockJob *job)
 603 {
 604     GLOBAL_STATE_CODE();
 605     return job->job.aio_context;
 606 }