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 "sysemu/block-backend.h"
  32 #include "qapi/error.h"
  33 #include "qapi/qapi-events-block-core.h"
  34 #include "qapi/qmp/qerror.h"
  35 #include "qemu/coroutine.h"
  36 #include "qemu/id.h"
  37 #include "qemu/timer.h"
  38
  39 /* Right now, this mutex is only needed to synchronize accesses to job->busy
  40  * and job->sleep_timer, such as concurrent calls to block_job_do_yield and
  41  * block_job_enter. */
  42 static QemuMutex block_job_mutex;
  43
  44 static void block_job_lock(void)
  45 {
  46     qemu_mutex_lock(&block_job_mutex);
  47 }
  48
  49 static void block_job_unlock(void)
  50 {
  51     qemu_mutex_unlock(&block_job_mutex);
  52 }
  53
  54 static void __attribute__((__constructor__)) block_job_init(void)
  55 {
  56     qemu_mutex_init(&block_job_mutex);
  57 }
  58
  59 static void block_job_event_cancelled(BlockJob *job);
  60 static void block_job_event_completed(BlockJob *job, const char *msg);
  61 static void block_job_enter_cond(BlockJob *job, bool(*fn)(BlockJob *job));
  62
  63 /* Transactional group of block jobs */
  64 struct BlockJobTxn {
  65
  66     /* Is this txn being cancelled? */
  67     bool aborting;
  68
  69     /* List of jobs */
  70     QLIST_HEAD(, BlockJob) jobs;
  71
  72     /* Reference count */
  73     int refcnt;
  74 };
  75
  76 static QLIST_HEAD(, BlockJob) block_jobs = QLIST_HEAD_INITIALIZER(block_jobs);
  77
  78 /*
  79  * The block job API is composed of two categories of functions.
  80  *
  81  * The first includes functions used by the monitor.  The monitor is
  82  * peculiar in that it accesses the block job list with block_job_get, and
  83  * therefore needs consistency across block_job_get and the actual operation
  84  * (e.g. block_job_set_speed).  The consistency is achieved with
  85  * aio_context_acquire/release.  These functions are declared in blockjob.h.
  86  *
  87  * The second includes functions used by the block job drivers and sometimes
  88  * by the core block layer.  These do not care about locking, because the
  89  * whole coroutine runs under the AioContext lock, and are declared in
  90  * blockjob_int.h.
  91  */
  92
  93 BlockJob *block_job_next(BlockJob *job)
  94 {
  95     if (!job) {
  96         return QLIST_FIRST(&block_jobs);
  97     }
  98     return QLIST_NEXT(job, job_list);
  99 }
 100
 101 BlockJob *block_job_get(const char *id)
 102 {
 103     BlockJob *job;
 104
 105     QLIST_FOREACH(job, &block_jobs, job_list) {
 106         if (job->id && !strcmp(id, job->id)) {
 107             return job;
 108         }
 109     }
 110
 111     return NULL;
 112 }
 113
 114 BlockJobTxn *block_job_txn_new(void)
 115 {
 116     BlockJobTxn *txn = g_new0(BlockJobTxn, 1);
 117     QLIST_INIT(&txn->jobs);
 118     txn->refcnt = 1;
 119     return txn;
 120 }
 121
 122 static void block_job_txn_ref(BlockJobTxn *txn)
 123 {
 124     txn->refcnt++;
 125 }
 126
 127 void block_job_txn_unref(BlockJobTxn *txn)
 128 {
 129     if (txn && --txn->refcnt == 0) {
 130         g_free(txn);
 131     }
 132 }
 133
 134 void block_job_txn_add_job(BlockJobTxn *txn, BlockJob *job)
 135 {
 136     if (!txn) {
 137         return;
 138     }
 139
 140     assert(!job->txn);
 141     job->txn = txn;
 142
 143     QLIST_INSERT_HEAD(&txn->jobs, job, txn_list);
 144     block_job_txn_ref(txn);
 145 }
 146
 147 static void block_job_pause(BlockJob *job)
 148 {
 149     job->pause_count++;
 150 }
 151
 152 static void block_job_resume(BlockJob *job)
 153 {
 154     assert(job->pause_count > 0);
 155     job->pause_count--;
 156     if (job->pause_count) {
 157         return;
 158     }
 159     block_job_enter(job);
 160 }
 161
 162 void block_job_ref(BlockJob *job)
 163 {
 164     ++job->refcnt;
 165 }
 166
 167 static void block_job_attached_aio_context(AioContext *new_context,
 168                                            void *opaque);
 169 static void block_job_detach_aio_context(void *opaque);
 170
 171 void block_job_unref(BlockJob *job)
 172 {
 173     if (--job->refcnt == 0) {
 174         BlockDriverState *bs = blk_bs(job->blk);
 175         QLIST_REMOVE(job, job_list);
 176         bs->job = NULL;
 177         block_job_remove_all_bdrv(job);
 178         blk_remove_aio_context_notifier(job->blk,
 179                                         block_job_attached_aio_context,
 180                                         block_job_detach_aio_context, job);
 181         blk_unref(job->blk);
 182         error_free(job->blocker);
 183         g_free(job->id);
 184         assert(!timer_pending(&job->sleep_timer));
 185         g_free(job);
 186     }
 187 }
 188
 189 static void block_job_attached_aio_context(AioContext *new_context,
 190                                            void *opaque)
 191 {
 192     BlockJob *job = opaque;
 193
 194     if (job->driver->attached_aio_context) {
 195         job->driver->attached_aio_context(job, new_context);
 196     }
 197
 198     block_job_resume(job);
 199 }
 200
 201 static void block_job_drain(BlockJob *job)
 202 {
 203     /* If job is !job->busy this kicks it into the next pause point. */
 204     block_job_enter(job);
 205
 206     blk_drain(job->blk);
 207     if (job->driver->drain) {
 208         job->driver->drain(job);
 209     }
 210 }
 211
 212 static void block_job_detach_aio_context(void *opaque)
 213 {
 214     BlockJob *job = opaque;
 215
 216     /* In case the job terminates during aio_poll()... */
 217     block_job_ref(job);
 218
 219     block_job_pause(job);
 220
 221     while (!job->paused && !job->completed) {
 222         block_job_drain(job);
 223     }
 224
 225     block_job_unref(job);
 226 }
 227
 228 static char *child_job_get_parent_desc(BdrvChild *c)
 229 {
 230     BlockJob *job = c->opaque;
 231     return g_strdup_printf("%s job '%s'",
 232                            BlockJobType_str(job->driver->job_type),
 233                            job->id);
 234 }
 235
 236 static void child_job_drained_begin(BdrvChild *c)
 237 {
 238     BlockJob *job = c->opaque;
 239     block_job_pause(job);
 240 }
 241
 242 static void child_job_drained_end(BdrvChild *c)
 243 {
 244     BlockJob *job = c->opaque;
 245     block_job_resume(job);
 246 }
 247
 248 static const BdrvChildRole child_job = {
 249     .get_parent_desc    = child_job_get_parent_desc,
 250     .drained_begin      = child_job_drained_begin,
 251     .drained_end        = child_job_drained_end,
 252     .stay_at_node       = true,
 253 };
 254
 255 void block_job_remove_all_bdrv(BlockJob *job)
 256 {
 257     GSList *l;
 258     for (l = job->nodes; l; l = l->next) {
 259         BdrvChild *c = l->data;
 260         bdrv_op_unblock_all(c->bs, job->blocker);
 261         bdrv_root_unref_child(c);
 262     }
 263     g_slist_free(job->nodes);
 264     job->nodes = NULL;
 265 }
 266
 267 int block_job_add_bdrv(BlockJob *job, const char *name, BlockDriverState *bs,
 268                        uint64_t perm, uint64_t shared_perm, Error **errp)
 269 {
 270     BdrvChild *c;
 271
 272     c = bdrv_root_attach_child(bs, name, &child_job, perm, shared_perm,
 273                                job, errp);
 274     if (c == NULL) {
 275         return -EPERM;
 276     }
 277
 278     job->nodes = g_slist_prepend(job->nodes, c);
 279     bdrv_ref(bs);
 280     bdrv_op_block_all(bs, job->blocker);
 281
 282     return 0;
 283 }
 284
 285 bool block_job_is_internal(BlockJob *job)
 286 {
 287     return (job->id == NULL);
 288 }
 289
 290 static bool block_job_started(BlockJob *job)
 291 {
 292     return job->co;
 293 }
 294
 295 /**
 296  * All jobs must allow a pause point before entering their job proper. This
 297  * ensures that jobs can be paused prior to being started, then resumed later.
 298  */
 299 static void coroutine_fn block_job_co_entry(void *opaque)
 300 {
 301     BlockJob *job = opaque;
 302
 303     assert(job && job->driver && job->driver->start);
 304     block_job_pause_point(job);
 305     job->driver->start(job);
 306 }
 307
 308 static void block_job_sleep_timer_cb(void *opaque)
 309 {
 310     BlockJob *job = opaque;
 311
 312     block_job_enter(job);
 313 }
 314
 315 void block_job_start(BlockJob *job)
 316 {
 317     assert(job && !block_job_started(job) && job->paused &&
 318            job->driver && job->driver->start);
 319     job->co = qemu_coroutine_create(block_job_co_entry, job);
 320     job->pause_count--;
 321     job->busy = true;
 322     job->paused = false;
 323     bdrv_coroutine_enter(blk_bs(job->blk), job->co);
 324 }
 325
 326 static void block_job_completed_single(BlockJob *job)
 327 {
 328     assert(job->completed);
 329
 330     if (!job->ret) {
 331         if (job->driver->commit) {
 332             job->driver->commit(job);
 333         }
 334     } else {
 335         if (job->driver->abort) {
 336             job->driver->abort(job);
 337         }
 338     }
 339     if (job->driver->clean) {
 340         job->driver->clean(job);
 341     }
 342
 343     if (job->cb) {
 344         job->cb(job->opaque, job->ret);
 345     }
 346
 347     /* Emit events only if we actually started */
 348     if (block_job_started(job)) {
 349         if (block_job_is_cancelled(job)) {
 350             block_job_event_cancelled(job);
 351         } else {
 352             const char *msg = NULL;
 353             if (job->ret < 0) {
 354                 msg = strerror(-job->ret);
 355             }
 356             block_job_event_completed(job, msg);
 357         }
 358     }
 359
 360     if (job->txn) {
 361         QLIST_REMOVE(job, txn_list);
 362         block_job_txn_unref(job->txn);
 363     }
 364     block_job_unref(job);
 365 }
 366
 367 static void block_job_cancel_async(BlockJob *job)
 368 {
 369     if (job->iostatus != BLOCK_DEVICE_IO_STATUS_OK) {
 370         block_job_iostatus_reset(job);
 371     }
 372     if (job->user_paused) {
 373         /* Do not call block_job_enter here, the caller will handle it.  */
 374         job->user_paused = false;
 375         job->pause_count--;
 376     }
 377     job->cancelled = true;
 378 }
 379
 380 static int block_job_finish_sync(BlockJob *job,
 381                                  void (*finish)(BlockJob *, Error **errp),
 382                                  Error **errp)
 383 {
 384     Error *local_err = NULL;
 385     int ret;
 386
 387     assert(blk_bs(job->blk)->job == job);
 388
 389     block_job_ref(job);
 390
 391     if (finish) {
 392         finish(job, &local_err);
 393     }
 394     if (local_err) {
 395         error_propagate(errp, local_err);
 396         block_job_unref(job);
 397         return -EBUSY;
 398     }
 399     /* block_job_drain calls block_job_enter, and it should be enough to
 400      * induce progress until the job completes or moves to the main thread.
 401     */
 402     while (!job->deferred_to_main_loop && !job->completed) {
 403         block_job_drain(job);
 404     }
 405     while (!job->completed) {
 406         aio_poll(qemu_get_aio_context(), true);
 407     }
 408     ret = (job->cancelled && job->ret == 0) ? -ECANCELED : job->ret;
 409     block_job_unref(job);
 410     return ret;
 411 }
 412
 413 static void block_job_completed_txn_abort(BlockJob *job)
 414 {
 415     AioContext *ctx;
 416     BlockJobTxn *txn = job->txn;
 417     BlockJob *other_job;
 418
 419     if (txn->aborting) {
 420         /*
 421          * We are cancelled by another job, which will handle everything.
 422          */
 423         return;
 424     }
 425     txn->aborting = true;
 426     block_job_txn_ref(txn);
 427
 428     /* We are the first failed job. Cancel other jobs. */
 429     QLIST_FOREACH(other_job, &txn->jobs, txn_list) {
 430         ctx = blk_get_aio_context(other_job->blk);
 431         aio_context_acquire(ctx);
 432     }
 433
 434     /* Other jobs are effectively cancelled by us, set the status for
 435      * them; this job, however, may or may not be cancelled, depending
 436      * on the caller, so leave it. */
 437     QLIST_FOREACH(other_job, &txn->jobs, txn_list) {
 438         if (other_job != job) {
 439             block_job_cancel_async(other_job);
 440         }
 441     }
 442     while (!QLIST_EMPTY(&txn->jobs)) {
 443         other_job = QLIST_FIRST(&txn->jobs);
 444         ctx = blk_get_aio_context(other_job->blk);
 445         if (!other_job->completed) {
 446             assert(other_job->cancelled);
 447             block_job_finish_sync(other_job, NULL, NULL);
 448         }
 449         block_job_completed_single(other_job);
 450         aio_context_release(ctx);
 451     }
 452
 453     block_job_txn_unref(txn);
 454 }
 455
 456 static void block_job_completed_txn_success(BlockJob *job)
 457 {
 458     AioContext *ctx;
 459     BlockJobTxn *txn = job->txn;
 460     BlockJob *other_job, *next;
 461     /*
 462      * Successful completion, see if there are other running jobs in this
 463      * txn.
 464      */
 465     QLIST_FOREACH(other_job, &txn->jobs, txn_list) {
 466         if (!other_job->completed) {
 467             return;
 468         }
 469     }
 470     /* We are the last completed job, commit the transaction. */
 471     QLIST_FOREACH_SAFE(other_job, &txn->jobs, txn_list, next) {
 472         ctx = blk_get_aio_context(other_job->blk);
 473         aio_context_acquire(ctx);
 474         assert(other_job->ret == 0);
 475         block_job_completed_single(other_job);
 476         aio_context_release(ctx);
 477     }
 478 }
 479
 480 /* Assumes the block_job_mutex is held */
 481 static bool block_job_timer_pending(BlockJob *job)
 482 {
 483     return timer_pending(&job->sleep_timer);
 484 }
 485
 486 void block_job_set_speed(BlockJob *job, int64_t speed, Error **errp)
 487 {
 488     Error *local_err = NULL;
 489     int64_t old_speed = job->speed;
 490
 491     if (!job->driver->set_speed) {
 492         error_setg(errp, QERR_UNSUPPORTED);
 493         return;
 494     }
 495     job->driver->set_speed(job, speed, &local_err);
 496     if (local_err) {
 497         error_propagate(errp, local_err);
 498         return;
 499     }
 500
 501     job->speed = speed;
 502     if (speed <= old_speed) {
 503         return;
 504     }
 505
 506     /* kick only if a timer is pending */
 507     block_job_enter_cond(job, block_job_timer_pending);
 508 }
 509
 510 void block_job_complete(BlockJob *job, Error **errp)
 511 {
 512     /* Should not be reachable via external interface for internal jobs */
 513     assert(job->id);
 514     if (job->pause_count || job->cancelled ||
 515         !block_job_started(job) || !job->driver->complete) {
 516         error_setg(errp, "The active block job '%s' cannot be completed",
 517                    job->id);
 518         return;
 519     }
 520
 521     job->driver->complete(job, errp);
 522 }
 523
 524 void block_job_user_pause(BlockJob *job)
 525 {
 526     job->user_paused = true;
 527     block_job_pause(job);
 528 }
 529
 530 bool block_job_user_paused(BlockJob *job)
 531 {
 532     return job->user_paused;
 533 }
 534
 535 void block_job_user_resume(BlockJob *job)
 536 {
 537     if (job && job->user_paused && job->pause_count > 0) {
 538         block_job_iostatus_reset(job);
 539         job->user_paused = false;
 540         block_job_resume(job);
 541     }
 542 }
 543
 544 void block_job_cancel(BlockJob *job)
 545 {
 546     if (block_job_started(job)) {
 547         block_job_cancel_async(job);
 548         block_job_enter(job);
 549     } else {
 550         block_job_completed(job, -ECANCELED);
 551     }
 552 }
 553
 554 /* A wrapper around block_job_cancel() taking an Error ** parameter so it may be
 555  * used with block_job_finish_sync() without the need for (rather nasty)
 556  * function pointer casts there. */
 557 static void block_job_cancel_err(BlockJob *job, Error **errp)
 558 {
 559     block_job_cancel(job);
 560 }
 561
 562 int block_job_cancel_sync(BlockJob *job)
 563 {
 564     return block_job_finish_sync(job, &block_job_cancel_err, NULL);
 565 }
 566
 567 void block_job_cancel_sync_all(void)
 568 {
 569     BlockJob *job;
 570     AioContext *aio_context;
 571
 572     while ((job = QLIST_FIRST(&block_jobs))) {
 573         aio_context = blk_get_aio_context(job->blk);
 574         aio_context_acquire(aio_context);
 575         block_job_cancel_sync(job);
 576         aio_context_release(aio_context);
 577     }
 578 }
 579
 580 int block_job_complete_sync(BlockJob *job, Error **errp)
 581 {
 582     return block_job_finish_sync(job, &block_job_complete, errp);
 583 }
 584
 585 BlockJobInfo *block_job_query(BlockJob *job, Error **errp)
 586 {
 587     BlockJobInfo *info;
 588
 589     if (block_job_is_internal(job)) {
 590         error_setg(errp, "Cannot query QEMU internal jobs");
 591         return NULL;
 592     }
 593     info = g_new0(BlockJobInfo, 1);
 594     info->type      = g_strdup(BlockJobType_str(job->driver->job_type));
 595     info->device    = g_strdup(job->id);
 596     info->len       = job->len;
 597     info->busy      = atomic_read(&job->busy);
 598     info->paused    = job->pause_count > 0;
 599     info->offset    = job->offset;
 600     info->speed     = job->speed;
 601     info->io_status = job->iostatus;
 602     info->ready     = job->ready;
 603     return info;
 604 }
 605
 606 static void block_job_iostatus_set_err(BlockJob *job, int error)
 607 {
 608     if (job->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
 609         job->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE :
 610                                           BLOCK_DEVICE_IO_STATUS_FAILED;
 611     }
 612 }
 613
 614 static void block_job_event_cancelled(BlockJob *job)
 615 {
 616     if (block_job_is_internal(job)) {
 617         return;
 618     }
 619
 620     qapi_event_send_block_job_cancelled(job->driver->job_type,
 621                                         job->id,
 622                                         job->len,
 623                                         job->offset,
 624                                         job->speed,
 625                                         &error_abort);
 626 }
 627
 628 static void block_job_event_completed(BlockJob *job, const char *msg)
 629 {
 630     if (block_job_is_internal(job)) {
 631         return;
 632     }
 633
 634     qapi_event_send_block_job_completed(job->driver->job_type,
 635                                         job->id,
 636                                         job->len,
 637                                         job->offset,
 638                                         job->speed,
 639                                         !!msg,
 640                                         msg,
 641                                         &error_abort);
 642 }
 643
 644 /*
 645  * API for block job drivers and the block layer.  These functions are
 646  * declared in blockjob_int.h.
 647  */
 648
 649 void *block_job_create(const char *job_id, const BlockJobDriver *driver,
 650                        BlockDriverState *bs, uint64_t perm,
 651                        uint64_t shared_perm, int64_t speed, int flags,
 652                        BlockCompletionFunc *cb, void *opaque, Error **errp)
 653 {
 654     BlockBackend *blk;
 655     BlockJob *job;
 656     int ret;
 657
 658     if (bs->job) {
 659         error_setg(errp, QERR_DEVICE_IN_USE, bdrv_get_device_name(bs));
 660         return NULL;
 661     }
 662
 663     if (job_id == NULL && !(flags & BLOCK_JOB_INTERNAL)) {
 664         job_id = bdrv_get_device_name(bs);
 665         if (!*job_id) {
 666             error_setg(errp, "An explicit job ID is required for this node");
 667             return NULL;
 668         }
 669     }
 670
 671     if (job_id) {
 672         if (flags & BLOCK_JOB_INTERNAL) {
 673             error_setg(errp, "Cannot specify job ID for internal block job");
 674             return NULL;
 675         }
 676
 677         if (!id_wellformed(job_id)) {
 678             error_setg(errp, "Invalid job ID '%s'", job_id);
 679             return NULL;
 680         }
 681
 682         if (block_job_get(job_id)) {
 683             error_setg(errp, "Job ID '%s' already in use", job_id);
 684             return NULL;
 685         }
 686     }
 687
 688     blk = blk_new(perm, shared_perm);
 689     ret = blk_insert_bs(blk, bs, errp);
 690     if (ret < 0) {
 691         blk_unref(blk);
 692         return NULL;
 693     }
 694
 695     job = g_malloc0(driver->instance_size);
 696     job->driver        = driver;
 697     job->id            = g_strdup(job_id);
 698     job->blk           = blk;
 699     job->cb            = cb;
 700     job->opaque        = opaque;
 701     job->busy          = false;
 702     job->paused        = true;
 703     job->pause_count   = 1;
 704     job->refcnt        = 1;
 705     aio_timer_init(qemu_get_aio_context(), &job->sleep_timer,
 706                    QEMU_CLOCK_REALTIME, SCALE_NS,
 707                    block_job_sleep_timer_cb, job);
 708
 709     error_setg(&job->blocker, "block device is in use by block job: %s",
 710                BlockJobType_str(driver->job_type));
 711     block_job_add_bdrv(job, "main node", bs, 0, BLK_PERM_ALL, &error_abort);
 712     bs->job = job;
 713
 714     bdrv_op_unblock(bs, BLOCK_OP_TYPE_DATAPLANE, job->blocker);
 715
 716     QLIST_INSERT_HEAD(&block_jobs, job, job_list);
 717
 718     blk_add_aio_context_notifier(blk, block_job_attached_aio_context,
 719                                  block_job_detach_aio_context, job);
 720
 721     /* Only set speed when necessary to avoid NotSupported error */
 722     if (speed != 0) {
 723         Error *local_err = NULL;
 724
 725         block_job_set_speed(job, speed, &local_err);
 726         if (local_err) {
 727             block_job_unref(job);
 728             error_propagate(errp, local_err);
 729             return NULL;
 730         }
 731     }
 732     return job;
 733 }
 734
 735 void block_job_pause_all(void)
 736 {
 737     BlockJob *job = NULL;
 738     while ((job = block_job_next(job))) {
 739         AioContext *aio_context = blk_get_aio_context(job->blk);
 740
 741         aio_context_acquire(aio_context);
 742         block_job_ref(job);
 743         block_job_pause(job);
 744         aio_context_release(aio_context);
 745     }
 746 }
 747
 748 void block_job_early_fail(BlockJob *job)
 749 {
 750     block_job_unref(job);
 751 }
 752
 753 void block_job_completed(BlockJob *job, int ret)
 754 {
 755     assert(blk_bs(job->blk)->job == job);
 756     assert(!job->completed);
 757     job->completed = true;
 758     job->ret = ret;
 759     if (!job->txn) {
 760         block_job_completed_single(job);
 761     } else if (ret < 0 || block_job_is_cancelled(job)) {
 762         block_job_completed_txn_abort(job);
 763     } else {
 764         block_job_completed_txn_success(job);
 765     }
 766 }
 767
 768 static bool block_job_should_pause(BlockJob *job)
 769 {
 770     return job->pause_count > 0;
 771 }
 772
 773 /* Yield, and schedule a timer to reenter the coroutine after @ns nanoseconds.
 774  * Reentering the job coroutine with block_job_enter() before the timer has
 775  * expired is allowed and cancels the timer.
 776  *
 777  * If @ns is (uint64_t) -1, no timer is scheduled and block_job_enter() must be
 778  * called explicitly. */
 779 static void block_job_do_yield(BlockJob *job, uint64_t ns)
 780 {
 781     block_job_lock();
 782     if (ns != -1) {
 783         timer_mod(&job->sleep_timer, ns);
 784     }
 785     job->busy = false;
 786     block_job_unlock();
 787     qemu_coroutine_yield();
 788
 789     /* Set by block_job_enter before re-entering the coroutine.  */
 790     assert(job->busy);
 791 }
 792
 793 void coroutine_fn block_job_pause_point(BlockJob *job)
 794 {
 795     assert(job && block_job_started(job));
 796
 797     if (!block_job_should_pause(job)) {
 798         return;
 799     }
 800     if (block_job_is_cancelled(job)) {
 801         return;
 802     }
 803
 804     if (job->driver->pause) {
 805         job->driver->pause(job);
 806     }
 807
 808     if (block_job_should_pause(job) && !block_job_is_cancelled(job)) {
 809         job->paused = true;
 810         block_job_do_yield(job, -1);
 811         job->paused = false;
 812     }
 813
 814     if (job->driver->resume) {
 815         job->driver->resume(job);
 816     }
 817 }
 818
 819 void block_job_resume_all(void)
 820 {
 821     BlockJob *job, *next;
 822
 823     QLIST_FOREACH_SAFE(job, &block_jobs, job_list, next) {
 824         AioContext *aio_context = blk_get_aio_context(job->blk);
 825
 826         aio_context_acquire(aio_context);
 827         block_job_resume(job);
 828         block_job_unref(job);
 829         aio_context_release(aio_context);
 830     }
 831 }
 832
 833 /*
 834  * Conditionally enter a block_job pending a call to fn() while
 835  * under the block_job_lock critical section.
 836  */
 837 static void block_job_enter_cond(BlockJob *job, bool(*fn)(BlockJob *job))
 838 {
 839     if (!block_job_started(job)) {
 840         return;
 841     }
 842     if (job->deferred_to_main_loop) {
 843         return;
 844     }
 845
 846     block_job_lock();
 847     if (job->busy) {
 848         block_job_unlock();
 849         return;
 850     }
 851
 852     if (fn && !fn(job)) {
 853         block_job_unlock();
 854         return;
 855     }
 856
 857     assert(!job->deferred_to_main_loop);
 858     timer_del(&job->sleep_timer);
 859     job->busy = true;
 860     block_job_unlock();
 861     aio_co_wake(job->co);
 862 }
 863
 864 void block_job_enter(BlockJob *job)
 865 {
 866     block_job_enter_cond(job, NULL);
 867 }
 868
 869 bool block_job_is_cancelled(BlockJob *job)
 870 {
 871     return job->cancelled;
 872 }
 873
 874 void block_job_sleep_ns(BlockJob *job, int64_t ns)
 875 {
 876     assert(job->busy);
 877
 878     /* Check cancellation *before* setting busy = false, too!  */
 879     if (block_job_is_cancelled(job)) {
 880         return;
 881     }
 882
 883     if (!block_job_should_pause(job)) {
 884         block_job_do_yield(job, qemu_clock_get_ns(QEMU_CLOCK_REALTIME) + ns);
 885     }
 886
 887     block_job_pause_point(job);
 888 }
 889
 890 void block_job_yield(BlockJob *job)
 891 {
 892     assert(job->busy);
 893
 894     /* Check cancellation *before* setting busy = false, too!  */
 895     if (block_job_is_cancelled(job)) {
 896         return;
 897     }
 898
 899     if (!block_job_should_pause(job)) {
 900         block_job_do_yield(job, -1);
 901     }
 902
 903     block_job_pause_point(job);
 904 }
 905
 906 void block_job_iostatus_reset(BlockJob *job)
 907 {
 908     if (job->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
 909         return;
 910     }
 911     assert(job->user_paused && job->pause_count > 0);
 912     job->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
 913 }
 914
 915 void block_job_event_ready(BlockJob *job)
 916 {
 917     job->ready = true;
 918
 919     if (block_job_is_internal(job)) {
 920         return;
 921     }
 922
 923     qapi_event_send_block_job_ready(job->driver->job_type,
 924                                     job->id,
 925                                     job->len,
 926                                     job->offset,
 927                                     job->speed, &error_abort);
 928 }
 929
 930 BlockErrorAction block_job_error_action(BlockJob *job, BlockdevOnError on_err,
 931                                         int is_read, int error)
 932 {
 933     BlockErrorAction action;
 934
 935     switch (on_err) {
 936     case BLOCKDEV_ON_ERROR_ENOSPC:
 937     case BLOCKDEV_ON_ERROR_AUTO:
 938         action = (error == ENOSPC) ?
 939                  BLOCK_ERROR_ACTION_STOP : BLOCK_ERROR_ACTION_REPORT;
 940         break;
 941     case BLOCKDEV_ON_ERROR_STOP:
 942         action = BLOCK_ERROR_ACTION_STOP;
 943         break;
 944     case BLOCKDEV_ON_ERROR_REPORT:
 945         action = BLOCK_ERROR_ACTION_REPORT;
 946         break;
 947     case BLOCKDEV_ON_ERROR_IGNORE:
 948         action = BLOCK_ERROR_ACTION_IGNORE;
 949         break;
 950     default:
 951         abort();
 952     }
 953     if (!block_job_is_internal(job)) {
 954         qapi_event_send_block_job_error(job->id,
 955                                         is_read ? IO_OPERATION_TYPE_READ :
 956                                         IO_OPERATION_TYPE_WRITE,
 957                                         action, &error_abort);
 958     }
 959     if (action == BLOCK_ERROR_ACTION_STOP) {
 960         /* make the pause user visible, which will be resumed from QMP. */
 961         block_job_user_pause(job);
 962         block_job_iostatus_set_err(job, error);
 963     }
 964     return action;
 965 }
 966
 967 typedef struct {
 968     BlockJob *job;
 969     AioContext *aio_context;
 970     BlockJobDeferToMainLoopFn *fn;
 971     void *opaque;
 972 } BlockJobDeferToMainLoopData;
 973
 974 static void block_job_defer_to_main_loop_bh(void *opaque)
 975 {
 976     BlockJobDeferToMainLoopData *data = opaque;
 977     AioContext *aio_context;
 978
 979     /* Prevent race with block_job_defer_to_main_loop() */
 980     aio_context_acquire(data->aio_context);
 981
 982     /* Fetch BDS AioContext again, in case it has changed */
 983     aio_context = blk_get_aio_context(data->job->blk);
 984     if (aio_context != data->aio_context) {
 985         aio_context_acquire(aio_context);
 986     }
 987
 988     data->fn(data->job, data->opaque);
 989
 990     if (aio_context != data->aio_context) {
 991         aio_context_release(aio_context);
 992     }
 993
 994     aio_context_release(data->aio_context);
 995
 996     g_free(data);
 997 }
 998
 999 void block_job_defer_to_main_loop(BlockJob *job,
1000                                   BlockJobDeferToMainLoopFn *fn,
1001                                   void *opaque)
1002 {
1003     BlockJobDeferToMainLoopData *data = g_malloc(sizeof(*data));
1004     data->job = job;
1005     data->aio_context = blk_get_aio_context(job->blk);
1006     data->fn = fn;
1007     data->opaque = opaque;
1008     job->deferred_to_main_loop = true;
1009
1010     aio_bh_schedule_oneshot(qemu_get_aio_context(),
1011                             block_job_defer_to_main_loop_bh, data);
1012 }