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 "qemu-common.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/coroutine.h"
  37 #include "qemu/timer.h"
  38
  39 static void block_job_event_cancelled(BlockJob *job);
  40 static void block_job_event_completed(BlockJob *job, const char *msg);
  41 static int block_job_event_pending(BlockJob *job);
  42
  43 /* Transactional group of block jobs */
  44 struct BlockJobTxn {
  45
  46     /* Is this txn being cancelled? */
  47     bool aborting;
  48
  49     /* List of jobs */
  50     QLIST_HEAD(, BlockJob) jobs;
  51
  52     /* Reference count */
  53     int refcnt;
  54 };
  55
  56 /*
  57  * The block job API is composed of two categories of functions.
  58  *
  59  * The first includes functions used by the monitor.  The monitor is
  60  * peculiar in that it accesses the block job list with block_job_get, and
  61  * therefore needs consistency across block_job_get and the actual operation
  62  * (e.g. block_job_set_speed).  The consistency is achieved with
  63  * aio_context_acquire/release.  These functions are declared in blockjob.h.
  64  *
  65  * The second includes functions used by the block job drivers and sometimes
  66  * by the core block layer.  These do not care about locking, because the
  67  * whole coroutine runs under the AioContext lock, and are declared in
  68  * blockjob_int.h.
  69  */
  70
  71 static bool is_block_job(Job *job)
  72 {
  73     return job_type(job) == JOB_TYPE_BACKUP ||
  74            job_type(job) == JOB_TYPE_COMMIT ||
  75            job_type(job) == JOB_TYPE_MIRROR ||
  76            job_type(job) == JOB_TYPE_STREAM;
  77 }
  78
  79 BlockJob *block_job_next(BlockJob *bjob)
  80 {
  81     Job *job = bjob ? &bjob->job : NULL;
  82
  83     do {
  84         job = job_next(job);
  85     } while (job && !is_block_job(job));
  86
  87     return job ? container_of(job, BlockJob, job) : NULL;
  88 }
  89
  90 BlockJob *block_job_get(const char *id)
  91 {
  92     Job *job = job_get(id);
  93
  94     if (job && is_block_job(job)) {
  95         return container_of(job, BlockJob, job);
  96     } else {
  97         return NULL;
  98     }
  99 }
 100
 101 BlockJobTxn *block_job_txn_new(void)
 102 {
 103     BlockJobTxn *txn = g_new0(BlockJobTxn, 1);
 104     QLIST_INIT(&txn->jobs);
 105     txn->refcnt = 1;
 106     return txn;
 107 }
 108
 109 static void block_job_txn_ref(BlockJobTxn *txn)
 110 {
 111     txn->refcnt++;
 112 }
 113
 114 void block_job_txn_unref(BlockJobTxn *txn)
 115 {
 116     if (txn && --txn->refcnt == 0) {
 117         g_free(txn);
 118     }
 119 }
 120
 121 void block_job_txn_add_job(BlockJobTxn *txn, BlockJob *job)
 122 {
 123     if (!txn) {
 124         return;
 125     }
 126
 127     assert(!job->txn);
 128     job->txn = txn;
 129
 130     QLIST_INSERT_HEAD(&txn->jobs, job, txn_list);
 131     block_job_txn_ref(txn);
 132 }
 133
 134 static void block_job_txn_del_job(BlockJob *job)
 135 {
 136     if (job->txn) {
 137         QLIST_REMOVE(job, txn_list);
 138         block_job_txn_unref(job->txn);
 139         job->txn = NULL;
 140     }
 141 }
 142
 143 static void block_job_attached_aio_context(AioContext *new_context,
 144                                            void *opaque);
 145 static void block_job_detach_aio_context(void *opaque);
 146
 147 void block_job_free(Job *job)
 148 {
 149     BlockJob *bjob = container_of(job, BlockJob, job);
 150     BlockDriverState *bs = blk_bs(bjob->blk);
 151
 152     assert(!bjob->txn);
 153
 154     bs->job = NULL;
 155     block_job_remove_all_bdrv(bjob);
 156     blk_remove_aio_context_notifier(bjob->blk,
 157                                     block_job_attached_aio_context,
 158                                     block_job_detach_aio_context, bjob);
 159     blk_unref(bjob->blk);
 160     error_free(bjob->blocker);
 161 }
 162
 163 static void block_job_attached_aio_context(AioContext *new_context,
 164                                            void *opaque)
 165 {
 166     BlockJob *job = opaque;
 167
 168     job->job.aio_context = new_context;
 169     if (job->driver->attached_aio_context) {
 170         job->driver->attached_aio_context(job, new_context);
 171     }
 172
 173     job_resume(&job->job);
 174 }
 175
 176 static void block_job_drain(BlockJob *job)
 177 {
 178     /* If job is !job->job.busy this kicks it into the next pause point. */
 179     block_job_enter(job);
 180
 181     blk_drain(job->blk);
 182     if (job->driver->drain) {
 183         job->driver->drain(job);
 184     }
 185 }
 186
 187 static void block_job_detach_aio_context(void *opaque)
 188 {
 189     BlockJob *job = opaque;
 190
 191     /* In case the job terminates during aio_poll()... */
 192     job_ref(&job->job);
 193
 194     job_pause(&job->job);
 195
 196     while (!job->job.paused && !job->completed) {
 197         block_job_drain(job);
 198     }
 199
 200     job->job.aio_context = NULL;
 201     job_unref(&job->job);
 202 }
 203
 204 static char *child_job_get_parent_desc(BdrvChild *c)
 205 {
 206     BlockJob *job = c->opaque;
 207     return g_strdup_printf("%s job '%s'", job_type_str(&job->job), job->job.id);
 208 }
 209
 210 static void child_job_drained_begin(BdrvChild *c)
 211 {
 212     BlockJob *job = c->opaque;
 213     job_pause(&job->job);
 214 }
 215
 216 static void child_job_drained_end(BdrvChild *c)
 217 {
 218     BlockJob *job = c->opaque;
 219     job_resume(&job->job);
 220 }
 221
 222 static const BdrvChildRole child_job = {
 223     .get_parent_desc    = child_job_get_parent_desc,
 224     .drained_begin      = child_job_drained_begin,
 225     .drained_end        = child_job_drained_end,
 226     .stay_at_node       = true,
 227 };
 228
 229 void block_job_remove_all_bdrv(BlockJob *job)
 230 {
 231     GSList *l;
 232     for (l = job->nodes; l; l = l->next) {
 233         BdrvChild *c = l->data;
 234         bdrv_op_unblock_all(c->bs, job->blocker);
 235         bdrv_root_unref_child(c);
 236     }
 237     g_slist_free(job->nodes);
 238     job->nodes = NULL;
 239 }
 240
 241 int block_job_add_bdrv(BlockJob *job, const char *name, BlockDriverState *bs,
 242                        uint64_t perm, uint64_t shared_perm, Error **errp)
 243 {
 244     BdrvChild *c;
 245
 246     c = bdrv_root_attach_child(bs, name, &child_job, perm, shared_perm,
 247                                job, errp);
 248     if (c == NULL) {
 249         return -EPERM;
 250     }
 251
 252     job->nodes = g_slist_prepend(job->nodes, c);
 253     bdrv_ref(bs);
 254     bdrv_op_block_all(bs, job->blocker);
 255
 256     return 0;
 257 }
 258
 259 bool block_job_is_internal(BlockJob *job)
 260 {
 261     return (job->job.id == NULL);
 262 }
 263
 264 const BlockJobDriver *block_job_driver(BlockJob *job)
 265 {
 266     return job->driver;
 267 }
 268
 269 static void block_job_decommission(BlockJob *job)
 270 {
 271     assert(job);
 272     job->completed = true;
 273     job->job.busy = false;
 274     job->job.paused = false;
 275     job->job.deferred_to_main_loop = true;
 276     block_job_txn_del_job(job);
 277     job_state_transition(&job->job, JOB_STATUS_NULL);
 278     job_unref(&job->job);
 279 }
 280
 281 static void block_job_do_dismiss(BlockJob *job)
 282 {
 283     block_job_decommission(job);
 284 }
 285
 286 static void block_job_conclude(BlockJob *job)
 287 {
 288     job_state_transition(&job->job, JOB_STATUS_CONCLUDED);
 289     if (job->auto_dismiss || !job_started(&job->job)) {
 290         block_job_do_dismiss(job);
 291     }
 292 }
 293
 294 static void block_job_update_rc(BlockJob *job)
 295 {
 296     if (!job->ret && job_is_cancelled(&job->job)) {
 297         job->ret = -ECANCELED;
 298     }
 299     if (job->ret) {
 300         job_state_transition(&job->job, JOB_STATUS_ABORTING);
 301     }
 302 }
 303
 304 static int block_job_prepare(BlockJob *job)
 305 {
 306     if (job->ret == 0 && job->driver->prepare) {
 307         job->ret = job->driver->prepare(job);
 308     }
 309     return job->ret;
 310 }
 311
 312 static void block_job_commit(BlockJob *job)
 313 {
 314     assert(!job->ret);
 315     if (job->driver->commit) {
 316         job->driver->commit(job);
 317     }
 318 }
 319
 320 static void block_job_abort(BlockJob *job)
 321 {
 322     assert(job->ret);
 323     if (job->driver->abort) {
 324         job->driver->abort(job);
 325     }
 326 }
 327
 328 static void block_job_clean(BlockJob *job)
 329 {
 330     if (job->driver->clean) {
 331         job->driver->clean(job);
 332     }
 333 }
 334
 335 static int block_job_finalize_single(BlockJob *job)
 336 {
 337     assert(job->completed);
 338
 339     /* Ensure abort is called for late-transactional failures */
 340     block_job_update_rc(job);
 341
 342     if (!job->ret) {
 343         block_job_commit(job);
 344     } else {
 345         block_job_abort(job);
 346     }
 347     block_job_clean(job);
 348
 349     if (job->cb) {
 350         job->cb(job->opaque, job->ret);
 351     }
 352
 353     /* Emit events only if we actually started */
 354     if (job_started(&job->job)) {
 355         if (job_is_cancelled(&job->job)) {
 356             block_job_event_cancelled(job);
 357         } else {
 358             const char *msg = NULL;
 359             if (job->ret < 0) {
 360                 msg = strerror(-job->ret);
 361             }
 362             block_job_event_completed(job, msg);
 363         }
 364     }
 365
 366     block_job_txn_del_job(job);
 367     block_job_conclude(job);
 368     return 0;
 369 }
 370
 371 static void block_job_cancel_async(BlockJob *job, bool force)
 372 {
 373     if (job->iostatus != BLOCK_DEVICE_IO_STATUS_OK) {
 374         block_job_iostatus_reset(job);
 375     }
 376     if (job->job.user_paused) {
 377         /* Do not call block_job_enter here, the caller will handle it.  */
 378         job->job.user_paused = false;
 379         job->job.pause_count--;
 380     }
 381     job->job.cancelled = true;
 382     /* To prevent 'force == false' overriding a previous 'force == true' */
 383     job->force |= force;
 384 }
 385
 386 static int block_job_txn_apply(BlockJobTxn *txn, int fn(BlockJob *), bool lock)
 387 {
 388     AioContext *ctx;
 389     BlockJob *job, *next;
 390     int rc = 0;
 391
 392     QLIST_FOREACH_SAFE(job, &txn->jobs, txn_list, next) {
 393         if (lock) {
 394             ctx = blk_get_aio_context(job->blk);
 395             aio_context_acquire(ctx);
 396         }
 397         rc = fn(job);
 398         if (lock) {
 399             aio_context_release(ctx);
 400         }
 401         if (rc) {
 402             break;
 403         }
 404     }
 405     return rc;
 406 }
 407
 408 static int block_job_finish_sync(BlockJob *job,
 409                                  void (*finish)(BlockJob *, Error **errp),
 410                                  Error **errp)
 411 {
 412     Error *local_err = NULL;
 413     int ret;
 414
 415     assert(blk_bs(job->blk)->job == job);
 416
 417     job_ref(&job->job);
 418
 419     if (finish) {
 420         finish(job, &local_err);
 421     }
 422     if (local_err) {
 423         error_propagate(errp, local_err);
 424         job_unref(&job->job);
 425         return -EBUSY;
 426     }
 427     /* block_job_drain calls block_job_enter, and it should be enough to
 428      * induce progress until the job completes or moves to the main thread.
 429     */
 430     while (!job->job.deferred_to_main_loop && !job->completed) {
 431         block_job_drain(job);
 432     }
 433     while (!job->completed) {
 434         aio_poll(qemu_get_aio_context(), true);
 435     }
 436     ret = (job_is_cancelled(&job->job) && job->ret == 0)
 437           ? -ECANCELED : job->ret;
 438     job_unref(&job->job);
 439     return ret;
 440 }
 441
 442 static void block_job_completed_txn_abort(BlockJob *job)
 443 {
 444     AioContext *ctx;
 445     BlockJobTxn *txn = job->txn;
 446     BlockJob *other_job;
 447
 448     if (txn->aborting) {
 449         /*
 450          * We are cancelled by another job, which will handle everything.
 451          */
 452         return;
 453     }
 454     txn->aborting = true;
 455     block_job_txn_ref(txn);
 456
 457     /* We are the first failed job. Cancel other jobs. */
 458     QLIST_FOREACH(other_job, &txn->jobs, txn_list) {
 459         ctx = blk_get_aio_context(other_job->blk);
 460         aio_context_acquire(ctx);
 461     }
 462
 463     /* Other jobs are effectively cancelled by us, set the status for
 464      * them; this job, however, may or may not be cancelled, depending
 465      * on the caller, so leave it. */
 466     QLIST_FOREACH(other_job, &txn->jobs, txn_list) {
 467         if (other_job != job) {
 468             block_job_cancel_async(other_job, false);
 469         }
 470     }
 471     while (!QLIST_EMPTY(&txn->jobs)) {
 472         other_job = QLIST_FIRST(&txn->jobs);
 473         ctx = blk_get_aio_context(other_job->blk);
 474         if (!other_job->completed) {
 475             assert(job_is_cancelled(&other_job->job));
 476             block_job_finish_sync(other_job, NULL, NULL);
 477         }
 478         block_job_finalize_single(other_job);
 479         aio_context_release(ctx);
 480     }
 481
 482     block_job_txn_unref(txn);
 483 }
 484
 485 static int block_job_needs_finalize(BlockJob *job)
 486 {
 487     return !job->auto_finalize;
 488 }
 489
 490 static void block_job_do_finalize(BlockJob *job)
 491 {
 492     int rc;
 493     assert(job && job->txn);
 494
 495     /* prepare the transaction to complete */
 496     rc = block_job_txn_apply(job->txn, block_job_prepare, true);
 497     if (rc) {
 498         block_job_completed_txn_abort(job);
 499     } else {
 500         block_job_txn_apply(job->txn, block_job_finalize_single, true);
 501     }
 502 }
 503
 504 static void block_job_completed_txn_success(BlockJob *job)
 505 {
 506     BlockJobTxn *txn = job->txn;
 507     BlockJob *other_job;
 508
 509     job_state_transition(&job->job, JOB_STATUS_WAITING);
 510
 511     /*
 512      * Successful completion, see if there are other running jobs in this
 513      * txn.
 514      */
 515     QLIST_FOREACH(other_job, &txn->jobs, txn_list) {
 516         if (!other_job->completed) {
 517             return;
 518         }
 519         assert(other_job->ret == 0);
 520     }
 521
 522     block_job_txn_apply(txn, block_job_event_pending, false);
 523
 524     /* If no jobs need manual finalization, automatically do so */
 525     if (block_job_txn_apply(txn, block_job_needs_finalize, false) == 0) {
 526         block_job_do_finalize(job);
 527     }
 528 }
 529
 530 /* Assumes the job_mutex is held */
 531 static bool job_timer_pending(Job *job)
 532 {
 533     return timer_pending(&job->sleep_timer);
 534 }
 535
 536 void block_job_set_speed(BlockJob *job, int64_t speed, Error **errp)
 537 {
 538     int64_t old_speed = job->speed;
 539
 540     if (job_apply_verb(&job->job, JOB_VERB_SET_SPEED, errp)) {
 541         return;
 542     }
 543     if (speed < 0) {
 544         error_setg(errp, QERR_INVALID_PARAMETER, "speed");
 545         return;
 546     }
 547
 548     ratelimit_set_speed(&job->limit, speed, BLOCK_JOB_SLICE_TIME);
 549
 550     job->speed = speed;
 551     if (speed && speed <= old_speed) {
 552         return;
 553     }
 554
 555     /* kick only if a timer is pending */
 556     job_enter_cond(&job->job, job_timer_pending);
 557 }
 558
 559 int64_t block_job_ratelimit_get_delay(BlockJob *job, uint64_t n)
 560 {
 561     if (!job->speed) {
 562         return 0;
 563     }
 564
 565     return ratelimit_calculate_delay(&job->limit, n);
 566 }
 567
 568 void block_job_complete(BlockJob *job, Error **errp)
 569 {
 570     /* Should not be reachable via external interface for internal jobs */
 571     assert(job->job.id);
 572     if (job_apply_verb(&job->job, JOB_VERB_COMPLETE, errp)) {
 573         return;
 574     }
 575     if (job->job.pause_count || job_is_cancelled(&job->job) ||
 576         !job->driver->complete)
 577     {
 578         error_setg(errp, "The active block job '%s' cannot be completed",
 579                    job->job.id);
 580         return;
 581     }
 582
 583     job->driver->complete(job, errp);
 584 }
 585
 586 void block_job_finalize(BlockJob *job, Error **errp)
 587 {
 588     assert(job && job->job.id);
 589     if (job_apply_verb(&job->job, JOB_VERB_FINALIZE, errp)) {
 590         return;
 591     }
 592     block_job_do_finalize(job);
 593 }
 594
 595 void block_job_dismiss(BlockJob **jobptr, Error **errp)
 596 {
 597     BlockJob *job = *jobptr;
 598     /* similarly to _complete, this is QMP-interface only. */
 599     assert(job->job.id);
 600     if (job_apply_verb(&job->job, JOB_VERB_DISMISS, errp)) {
 601         return;
 602     }
 603
 604     block_job_do_dismiss(job);
 605     *jobptr = NULL;
 606 }
 607
 608 void block_job_cancel(BlockJob *job, bool force)
 609 {
 610     if (job->job.status == JOB_STATUS_CONCLUDED) {
 611         block_job_do_dismiss(job);
 612         return;
 613     }
 614     block_job_cancel_async(job, force);
 615     if (!job_started(&job->job)) {
 616         block_job_completed(job, -ECANCELED);
 617     } else if (job->job.deferred_to_main_loop) {
 618         block_job_completed_txn_abort(job);
 619     } else {
 620         block_job_enter(job);
 621     }
 622 }
 623
 624 void block_job_user_cancel(BlockJob *job, bool force, Error **errp)
 625 {
 626     if (job_apply_verb(&job->job, JOB_VERB_CANCEL, errp)) {
 627         return;
 628     }
 629     block_job_cancel(job, force);
 630 }
 631
 632 /* A wrapper around block_job_cancel() taking an Error ** parameter so it may be
 633  * used with block_job_finish_sync() without the need for (rather nasty)
 634  * function pointer casts there. */
 635 static void block_job_cancel_err(BlockJob *job, Error **errp)
 636 {
 637     block_job_cancel(job, false);
 638 }
 639
 640 int block_job_cancel_sync(BlockJob *job)
 641 {
 642     return block_job_finish_sync(job, &block_job_cancel_err, NULL);
 643 }
 644
 645 void block_job_cancel_sync_all(void)
 646 {
 647     BlockJob *job;
 648     AioContext *aio_context;
 649
 650     while ((job = block_job_next(NULL))) {
 651         aio_context = blk_get_aio_context(job->blk);
 652         aio_context_acquire(aio_context);
 653         block_job_cancel_sync(job);
 654         aio_context_release(aio_context);
 655     }
 656 }
 657
 658 int block_job_complete_sync(BlockJob *job, Error **errp)
 659 {
 660     return block_job_finish_sync(job, &block_job_complete, errp);
 661 }
 662
 663 void block_job_progress_update(BlockJob *job, uint64_t done)
 664 {
 665     job->offset += done;
 666 }
 667
 668 void block_job_progress_set_remaining(BlockJob *job, uint64_t remaining)
 669 {
 670     job->len = job->offset + remaining;
 671 }
 672
 673 BlockJobInfo *block_job_query(BlockJob *job, Error **errp)
 674 {
 675     BlockJobInfo *info;
 676
 677     if (block_job_is_internal(job)) {
 678         error_setg(errp, "Cannot query QEMU internal jobs");
 679         return NULL;
 680     }
 681     info = g_new0(BlockJobInfo, 1);
 682     info->type      = g_strdup(job_type_str(&job->job));
 683     info->device    = g_strdup(job->job.id);
 684     info->len       = job->len;
 685     info->busy      = atomic_read(&job->job.busy);
 686     info->paused    = job->job.pause_count > 0;
 687     info->offset    = job->offset;
 688     info->speed     = job->speed;
 689     info->io_status = job->iostatus;
 690     info->ready     = job->ready;
 691     info->status    = job->job.status;
 692     info->auto_finalize = job->auto_finalize;
 693     info->auto_dismiss  = job->auto_dismiss;
 694     info->has_error = job->ret != 0;
 695     info->error     = job->ret ? g_strdup(strerror(-job->ret)) : NULL;
 696     return info;
 697 }
 698
 699 static void block_job_iostatus_set_err(BlockJob *job, int error)
 700 {
 701     if (job->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
 702         job->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE :
 703                                           BLOCK_DEVICE_IO_STATUS_FAILED;
 704     }
 705 }
 706
 707 static void block_job_event_cancelled(BlockJob *job)
 708 {
 709     if (block_job_is_internal(job)) {
 710         return;
 711     }
 712
 713     qapi_event_send_block_job_cancelled(job_type(&job->job),
 714                                         job->job.id,
 715                                         job->len,
 716                                         job->offset,
 717                                         job->speed,
 718                                         &error_abort);
 719 }
 720
 721 static void block_job_event_completed(BlockJob *job, const char *msg)
 722 {
 723     if (block_job_is_internal(job)) {
 724         return;
 725     }
 726
 727     qapi_event_send_block_job_completed(job_type(&job->job),
 728                                         job->job.id,
 729                                         job->len,
 730                                         job->offset,
 731                                         job->speed,
 732                                         !!msg,
 733                                         msg,
 734                                         &error_abort);
 735 }
 736
 737 static int block_job_event_pending(BlockJob *job)
 738 {
 739     job_state_transition(&job->job, JOB_STATUS_PENDING);
 740     if (!job->auto_finalize && !block_job_is_internal(job)) {
 741         qapi_event_send_block_job_pending(job_type(&job->job),
 742                                           job->job.id,
 743                                           &error_abort);
 744     }
 745     return 0;
 746 }
 747
 748 /*
 749  * API for block job drivers and the block layer.  These functions are
 750  * declared in blockjob_int.h.
 751  */
 752
 753 void *block_job_create(const char *job_id, const BlockJobDriver *driver,
 754                        BlockJobTxn *txn, BlockDriverState *bs, uint64_t perm,
 755                        uint64_t shared_perm, int64_t speed, int flags,
 756                        BlockCompletionFunc *cb, void *opaque, Error **errp)
 757 {
 758     BlockBackend *blk;
 759     BlockJob *job;
 760     int ret;
 761
 762     if (bs->job) {
 763         error_setg(errp, QERR_DEVICE_IN_USE, bdrv_get_device_name(bs));
 764         return NULL;
 765     }
 766
 767     if (job_id == NULL && !(flags & BLOCK_JOB_INTERNAL)) {
 768         job_id = bdrv_get_device_name(bs);
 769         if (!*job_id) {
 770             error_setg(errp, "An explicit job ID is required for this node");
 771             return NULL;
 772         }
 773     }
 774
 775     if (job_id) {
 776         if (flags & BLOCK_JOB_INTERNAL) {
 777             error_setg(errp, "Cannot specify job ID for internal block job");
 778             return NULL;
 779         }
 780     }
 781
 782     blk = blk_new(perm, shared_perm);
 783     ret = blk_insert_bs(blk, bs, errp);
 784     if (ret < 0) {
 785         blk_unref(blk);
 786         return NULL;
 787     }
 788
 789     job = job_create(job_id, &driver->job_driver, blk_get_aio_context(blk),
 790                      errp);
 791     if (job == NULL) {
 792         blk_unref(blk);
 793         return NULL;
 794     }
 795
 796     assert(is_block_job(&job->job));
 797     assert(job->job.driver->free == &block_job_free);
 798     assert(job->job.driver->user_resume == &block_job_user_resume);
 799
 800     job->driver        = driver;
 801     job->blk           = blk;
 802     job->cb            = cb;
 803     job->opaque        = opaque;
 804     job->auto_finalize = !(flags & BLOCK_JOB_MANUAL_FINALIZE);
 805     job->auto_dismiss  = !(flags & BLOCK_JOB_MANUAL_DISMISS);
 806
 807     error_setg(&job->blocker, "block device is in use by block job: %s",
 808                job_type_str(&job->job));
 809     block_job_add_bdrv(job, "main node", bs, 0, BLK_PERM_ALL, &error_abort);
 810     bs->job = job;
 811
 812     bdrv_op_unblock(bs, BLOCK_OP_TYPE_DATAPLANE, job->blocker);
 813
 814     blk_add_aio_context_notifier(blk, block_job_attached_aio_context,
 815                                  block_job_detach_aio_context, job);
 816
 817     /* Only set speed when necessary to avoid NotSupported error */
 818     if (speed != 0) {
 819         Error *local_err = NULL;
 820
 821         block_job_set_speed(job, speed, &local_err);
 822         if (local_err) {
 823             block_job_early_fail(job);
 824             error_propagate(errp, local_err);
 825             return NULL;
 826         }
 827     }
 828
 829     /* Single jobs are modeled as single-job transactions for sake of
 830      * consolidating the job management logic */
 831     if (!txn) {
 832         txn = block_job_txn_new();
 833         block_job_txn_add_job(txn, job);
 834         block_job_txn_unref(txn);
 835     } else {
 836         block_job_txn_add_job(txn, job);
 837     }
 838
 839     return job;
 840 }
 841
 842 void block_job_early_fail(BlockJob *job)
 843 {
 844     assert(job->job.status == JOB_STATUS_CREATED);
 845     block_job_decommission(job);
 846 }
 847
 848 void block_job_completed(BlockJob *job, int ret)
 849 {
 850     assert(job && job->txn && !job->completed);
 851     assert(blk_bs(job->blk)->job == job);
 852     job->completed = true;
 853     job->ret = ret;
 854     block_job_update_rc(job);
 855     trace_block_job_completed(job, ret, job->ret);
 856     if (job->ret) {
 857         block_job_completed_txn_abort(job);
 858     } else {
 859         block_job_completed_txn_success(job);
 860     }
 861 }
 862
 863 void block_job_enter(BlockJob *job)
 864 {
 865     job_enter_cond(&job->job, NULL);
 866 }
 867
 868 void block_job_yield(BlockJob *job)
 869 {
 870     assert(job->job.busy);
 871
 872     /* Check cancellation *before* setting busy = false, too!  */
 873     if (job_is_cancelled(&job->job)) {
 874         return;
 875     }
 876
 877     if (!job_should_pause(&job->job)) {
 878         job_do_yield(&job->job, -1);
 879     }
 880
 881     job_pause_point(&job->job);
 882 }
 883
 884 void block_job_iostatus_reset(BlockJob *job)
 885 {
 886     if (job->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
 887         return;
 888     }
 889     assert(job->job.user_paused && job->job.pause_count > 0);
 890     job->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
 891 }
 892
 893 void block_job_user_resume(Job *job)
 894 {
 895     BlockJob *bjob = container_of(job, BlockJob, job);
 896     block_job_iostatus_reset(bjob);
 897 }
 898
 899 void block_job_event_ready(BlockJob *job)
 900 {
 901     job_state_transition(&job->job, JOB_STATUS_READY);
 902     job->ready = true;
 903
 904     if (block_job_is_internal(job)) {
 905         return;
 906     }
 907
 908     qapi_event_send_block_job_ready(job_type(&job->job),
 909                                     job->job.id,
 910                                     job->len,
 911                                     job->offset,
 912                                     job->speed, &error_abort);
 913 }
 914
 915 BlockErrorAction block_job_error_action(BlockJob *job, BlockdevOnError on_err,
 916                                         int is_read, int error)
 917 {
 918     BlockErrorAction action;
 919
 920     switch (on_err) {
 921     case BLOCKDEV_ON_ERROR_ENOSPC:
 922     case BLOCKDEV_ON_ERROR_AUTO:
 923         action = (error == ENOSPC) ?
 924                  BLOCK_ERROR_ACTION_STOP : BLOCK_ERROR_ACTION_REPORT;
 925         break;
 926     case BLOCKDEV_ON_ERROR_STOP:
 927         action = BLOCK_ERROR_ACTION_STOP;
 928         break;
 929     case BLOCKDEV_ON_ERROR_REPORT:
 930         action = BLOCK_ERROR_ACTION_REPORT;
 931         break;
 932     case BLOCKDEV_ON_ERROR_IGNORE:
 933         action = BLOCK_ERROR_ACTION_IGNORE;
 934         break;
 935     default:
 936         abort();
 937     }
 938     if (!block_job_is_internal(job)) {
 939         qapi_event_send_block_job_error(job->job.id,
 940                                         is_read ? IO_OPERATION_TYPE_READ :
 941                                         IO_OPERATION_TYPE_WRITE,
 942                                         action, &error_abort);
 943     }
 944     if (action == BLOCK_ERROR_ACTION_STOP) {
 945         job_pause(&job->job);
 946         /* make the pause user visible, which will be resumed from QMP. */
 947         job->job.user_paused = true;
 948         block_job_iostatus_set_err(job, error);
 949     }
 950     return action;
 951 }