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/qmp/qerror.h"
  33 #include "qapi/qmp/qjson.h"
  34 #include "qemu/coroutine.h"
  35 #include "qemu/id.h"
  36 #include "qmp-commands.h"
  37 #include "qemu/timer.h"
  38 #include "qapi-event.h"
  39
  40 /* Right now, this mutex is only needed to synchronize accesses to job->busy
  41  * and job->sleep_timer, such as concurrent calls to block_job_do_yield and
  42  * block_job_enter. */
  43 static QemuMutex block_job_mutex;
  44
  45 static void block_job_lock(void)
  46 {
  47     qemu_mutex_lock(&block_job_mutex);
  48 }
  49
  50 static void block_job_unlock(void)
  51 {
  52     qemu_mutex_unlock(&block_job_mutex);
  53 }
  54
  55 static void __attribute__((__constructor__)) block_job_init(void)
  56 {
  57     qemu_mutex_init(&block_job_mutex);
  58 }
  59
  60 static void block_job_event_cancelled(BlockJob *job);
  61 static void block_job_event_completed(BlockJob *job, const char *msg);
  62 static void block_job_enter_cond(BlockJob *job, bool(*fn)(BlockJob *job));
  63
  64 /* Transactional group of block jobs */
  65 struct BlockJobTxn {
  66
  67     /* Is this txn being cancelled? */
  68     bool aborting;
  69
  70     /* List of jobs */
  71     QLIST_HEAD(, BlockJob) jobs;
  72
  73     /* Reference count */
  74     int refcnt;
  75 };
  76
  77 static QLIST_HEAD(, BlockJob) block_jobs = QLIST_HEAD_INITIALIZER(block_jobs);
  78
  79 /*
  80  * The block job API is composed of two categories of functions.
  81  *
  82  * The first includes functions used by the monitor.  The monitor is
  83  * peculiar in that it accesses the block job list with block_job_get, and
  84  * therefore needs consistency across block_job_get and the actual operation
  85  * (e.g. block_job_set_speed).  The consistency is achieved with
  86  * aio_context_acquire/release.  These functions are declared in blockjob.h.
  87  *
  88  * The second includes functions used by the block job drivers and sometimes
  89  * by the core block layer.  These do not care about locking, because the
  90  * whole coroutine runs under the AioContext lock, and are declared in
  91  * blockjob_int.h.
  92  */
  93
  94 BlockJob *block_job_next(BlockJob *job)
  95 {
  96     if (!job) {
  97         return QLIST_FIRST(&block_jobs);
  98     }
  99     return QLIST_NEXT(job, job_list);
 100 }
 101
 102 BlockJob *block_job_get(const char *id)
 103 {
 104     BlockJob *job;
 105
 106     QLIST_FOREACH(job, &block_jobs, job_list) {
 107         if (job->id && !strcmp(id, job->id)) {
 108             return job;
 109         }
 110     }
 111
 112     return NULL;
 113 }
 114
 115 BlockJobTxn *block_job_txn_new(void)
 116 {
 117     BlockJobTxn *txn = g_new0(BlockJobTxn, 1);
 118     QLIST_INIT(&txn->jobs);
 119     txn->refcnt = 1;
 120     return txn;
 121 }
 122
 123 static void block_job_txn_ref(BlockJobTxn *txn)
 124 {
 125     txn->refcnt++;
 126 }
 127
 128 void block_job_txn_unref(BlockJobTxn *txn)
 129 {
 130     if (txn && --txn->refcnt == 0) {
 131         g_free(txn);
 132     }
 133 }
 134
 135 void block_job_txn_add_job(BlockJobTxn *txn, BlockJob *job)
 136 {
 137     if (!txn) {
 138         return;
 139     }
 140
 141     assert(!job->txn);
 142     job->txn = txn;
 143
 144     QLIST_INSERT_HEAD(&txn->jobs, job, txn_list);
 145     block_job_txn_ref(txn);
 146 }
 147
 148 static void block_job_pause(BlockJob *job)
 149 {
 150     job->pause_count++;
 151 }
 152
 153 static void block_job_resume(BlockJob *job)
 154 {
 155     assert(job->pause_count > 0);
 156     job->pause_count--;
 157     if (job->pause_count) {
 158         return;
 159     }
 160     block_job_enter(job);
 161 }
 162
 163 void block_job_ref(BlockJob *job)
 164 {
 165     ++job->refcnt;
 166 }
 167
 168 static void block_job_attached_aio_context(AioContext *new_context,
 169                                            void *opaque);
 170 static void block_job_detach_aio_context(void *opaque);
 171
 172 void block_job_unref(BlockJob *job)
 173 {
 174     if (--job->refcnt == 0) {
 175         BlockDriverState *bs = blk_bs(job->blk);
 176         QLIST_REMOVE(job, job_list);
 177         bs->job = NULL;
 178         block_job_remove_all_bdrv(job);
 179         blk_remove_aio_context_notifier(job->blk,
 180                                         block_job_attached_aio_context,
 181                                         block_job_detach_aio_context, job);
 182         blk_unref(job->blk);
 183         error_free(job->blocker);
 184         g_free(job->id);
 185         assert(!timer_pending(&job->sleep_timer));
 186         g_free(job);
 187     }
 188 }
 189
 190 static void block_job_attached_aio_context(AioContext *new_context,
 191                                            void *opaque)
 192 {
 193     BlockJob *job = opaque;
 194
 195     if (job->driver->attached_aio_context) {
 196         job->driver->attached_aio_context(job, new_context);
 197     }
 198
 199     block_job_resume(job);
 200 }
 201
 202 static void block_job_drain(BlockJob *job)
 203 {
 204     /* If job is !job->busy this kicks it into the next pause point. */
 205     block_job_enter(job);
 206
 207     blk_drain(job->blk);
 208     if (job->driver->drain) {
 209         job->driver->drain(job);
 210     }
 211 }
 212
 213 static void block_job_detach_aio_context(void *opaque)
 214 {
 215     BlockJob *job = opaque;
 216
 217     /* In case the job terminates during aio_poll()... */
 218     block_job_ref(job);
 219
 220     block_job_pause(job);
 221
 222     while (!job->paused && !job->completed) {
 223         block_job_drain(job);
 224     }
 225
 226     block_job_unref(job);
 227 }
 228
 229 static char *child_job_get_parent_desc(BdrvChild *c)
 230 {
 231     BlockJob *job = c->opaque;
 232     return g_strdup_printf("%s job '%s'",
 233                            BlockJobType_str(job->driver->job_type),
 234                            job->id);
 235 }
 236
 237 static const BdrvChildRole child_job = {
 238     .get_parent_desc    = child_job_get_parent_desc,
 239     .stay_at_node       = true,
 240 };
 241
 242 static void block_job_drained_begin(void *opaque)
 243 {
 244     BlockJob *job = opaque;
 245     block_job_pause(job);
 246 }
 247
 248 static void block_job_drained_end(void *opaque)
 249 {
 250     BlockJob *job = opaque;
 251     block_job_resume(job);
 252 }
 253
 254 static const BlockDevOps block_job_dev_ops = {
 255     .drained_begin = block_job_drained_begin,
 256     .drained_end = block_job_drained_end,
 257 };
 258
 259 void block_job_remove_all_bdrv(BlockJob *job)
 260 {
 261     GSList *l;
 262     for (l = job->nodes; l; l = l->next) {
 263         BdrvChild *c = l->data;
 264         bdrv_op_unblock_all(c->bs, job->blocker);
 265         bdrv_root_unref_child(c);
 266     }
 267     g_slist_free(job->nodes);
 268     job->nodes = NULL;
 269 }
 270
 271 int block_job_add_bdrv(BlockJob *job, const char *name, BlockDriverState *bs,
 272                        uint64_t perm, uint64_t shared_perm, Error **errp)
 273 {
 274     BdrvChild *c;
 275
 276     c = bdrv_root_attach_child(bs, name, &child_job, perm, shared_perm,
 277                                job, errp);
 278     if (c == NULL) {
 279         return -EPERM;
 280     }
 281
 282     job->nodes = g_slist_prepend(job->nodes, c);
 283     bdrv_ref(bs);
 284     bdrv_op_block_all(bs, job->blocker);
 285
 286     return 0;
 287 }
 288
 289 bool block_job_is_internal(BlockJob *job)
 290 {
 291     return (job->id == NULL);
 292 }
 293
 294 static bool block_job_started(BlockJob *job)
 295 {
 296     return job->co;
 297 }
 298
 299 /**
 300  * All jobs must allow a pause point before entering their job proper. This
 301  * ensures that jobs can be paused prior to being started, then resumed later.
 302  */
 303 static void coroutine_fn block_job_co_entry(void *opaque)
 304 {
 305     BlockJob *job = opaque;
 306
 307     assert(job && job->driver && job->driver->start);
 308     block_job_pause_point(job);
 309     job->driver->start(job);
 310 }
 311
 312 static void block_job_sleep_timer_cb(void *opaque)
 313 {
 314     BlockJob *job = opaque;
 315
 316     block_job_enter(job);
 317 }
 318
 319 void block_job_start(BlockJob *job)
 320 {
 321     assert(job && !block_job_started(job) && job->paused &&
 322            job->driver && job->driver->start);
 323     job->co = qemu_coroutine_create(block_job_co_entry, job);
 324     job->pause_count--;
 325     job->busy = true;
 326     job->paused = false;
 327     bdrv_coroutine_enter(blk_bs(job->blk), job->co);
 328 }
 329
 330 static void block_job_completed_single(BlockJob *job)
 331 {
 332     assert(job->completed);
 333
 334     if (!job->ret) {
 335         if (job->driver->commit) {
 336             job->driver->commit(job);
 337         }
 338     } else {
 339         if (job->driver->abort) {
 340             job->driver->abort(job);
 341         }
 342     }
 343     if (job->driver->clean) {
 344         job->driver->clean(job);
 345     }
 346
 347     if (job->cb) {
 348         job->cb(job->opaque, job->ret);
 349     }
 350
 351     /* Emit events only if we actually started */
 352     if (block_job_started(job)) {
 353         if (block_job_is_cancelled(job)) {
 354             block_job_event_cancelled(job);
 355         } else {
 356             const char *msg = NULL;
 357             if (job->ret < 0) {
 358                 msg = strerror(-job->ret);
 359             }
 360             block_job_event_completed(job, msg);
 361         }
 362     }
 363
 364     if (job->txn) {
 365         QLIST_REMOVE(job, txn_list);
 366         block_job_txn_unref(job->txn);
 367     }
 368     block_job_unref(job);
 369 }
 370
 371 static void block_job_cancel_async(BlockJob *job)
 372 {
 373     if (job->iostatus != BLOCK_DEVICE_IO_STATUS_OK) {
 374         block_job_iostatus_reset(job);
 375     }
 376     if (job->user_paused) {
 377         /* Do not call block_job_enter here, the caller will handle it.  */
 378         job->user_paused = false;
 379         job->pause_count--;
 380     }
 381     job->cancelled = true;
 382 }
 383
 384 static int block_job_finish_sync(BlockJob *job,
 385                                  void (*finish)(BlockJob *, Error **errp),
 386                                  Error **errp)
 387 {
 388     Error *local_err = NULL;
 389     int ret;
 390
 391     assert(blk_bs(job->blk)->job == job);
 392
 393     block_job_ref(job);
 394
 395     if (finish) {
 396         finish(job, &local_err);
 397     }
 398     if (local_err) {
 399         error_propagate(errp, local_err);
 400         block_job_unref(job);
 401         return -EBUSY;
 402     }
 403     /* block_job_drain calls block_job_enter, and it should be enough to
 404      * induce progress until the job completes or moves to the main thread.
 405     */
 406     while (!job->deferred_to_main_loop && !job->completed) {
 407         block_job_drain(job);
 408     }
 409     while (!job->completed) {
 410         aio_poll(qemu_get_aio_context(), true);
 411     }
 412     ret = (job->cancelled && job->ret == 0) ? -ECANCELED : job->ret;
 413     block_job_unref(job);
 414     return ret;
 415 }
 416
 417 static void block_job_completed_txn_abort(BlockJob *job)
 418 {
 419     AioContext *ctx;
 420     BlockJobTxn *txn = job->txn;
 421     BlockJob *other_job;
 422
 423     if (txn->aborting) {
 424         /*
 425          * We are cancelled by another job, which will handle everything.
 426          */
 427         return;
 428     }
 429     txn->aborting = true;
 430     block_job_txn_ref(txn);
 431
 432     /* We are the first failed job. Cancel other jobs. */
 433     QLIST_FOREACH(other_job, &txn->jobs, txn_list) {
 434         ctx = blk_get_aio_context(other_job->blk);
 435         aio_context_acquire(ctx);
 436     }
 437
 438     /* Other jobs are effectively cancelled by us, set the status for
 439      * them; this job, however, may or may not be cancelled, depending
 440      * on the caller, so leave it. */
 441     QLIST_FOREACH(other_job, &txn->jobs, txn_list) {
 442         if (other_job != job) {
 443             block_job_cancel_async(other_job);
 444         }
 445     }
 446     while (!QLIST_EMPTY(&txn->jobs)) {
 447         other_job = QLIST_FIRST(&txn->jobs);
 448         ctx = blk_get_aio_context(other_job->blk);
 449         if (!other_job->completed) {
 450             assert(other_job->cancelled);
 451             block_job_finish_sync(other_job, NULL, NULL);
 452         }
 453         block_job_completed_single(other_job);
 454         aio_context_release(ctx);
 455     }
 456
 457     block_job_txn_unref(txn);
 458 }
 459
 460 static void block_job_completed_txn_success(BlockJob *job)
 461 {
 462     AioContext *ctx;
 463     BlockJobTxn *txn = job->txn;
 464     BlockJob *other_job, *next;
 465     /*
 466      * Successful completion, see if there are other running jobs in this
 467      * txn.
 468      */
 469     QLIST_FOREACH(other_job, &txn->jobs, txn_list) {
 470         if (!other_job->completed) {
 471             return;
 472         }
 473     }
 474     /* We are the last completed job, commit the transaction. */
 475     QLIST_FOREACH_SAFE(other_job, &txn->jobs, txn_list, next) {
 476         ctx = blk_get_aio_context(other_job->blk);
 477         aio_context_acquire(ctx);
 478         assert(other_job->ret == 0);
 479         block_job_completed_single(other_job);
 480         aio_context_release(ctx);
 481     }
 482 }
 483
 484 /* Assumes the block_job_mutex is held */
 485 static bool block_job_timer_pending(BlockJob *job)
 486 {
 487     return timer_pending(&job->sleep_timer);
 488 }
 489
 490 void block_job_set_speed(BlockJob *job, int64_t speed, Error **errp)
 491 {
 492     Error *local_err = NULL;
 493     int64_t old_speed = job->speed;
 494
 495     if (!job->driver->set_speed) {
 496         error_setg(errp, QERR_UNSUPPORTED);
 497         return;
 498     }
 499     job->driver->set_speed(job, speed, &local_err);
 500     if (local_err) {
 501         error_propagate(errp, local_err);
 502         return;
 503     }
 504
 505     job->speed = speed;
 506     if (speed <= old_speed) {
 507         return;
 508     }
 509
 510     /* kick only if a timer is pending */
 511     block_job_enter_cond(job, block_job_timer_pending);
 512 }
 513
 514 void block_job_complete(BlockJob *job, Error **errp)
 515 {
 516     /* Should not be reachable via external interface for internal jobs */
 517     assert(job->id);
 518     if (job->pause_count || job->cancelled ||
 519         !block_job_started(job) || !job->driver->complete) {
 520         error_setg(errp, "The active block job '%s' cannot be completed",
 521                    job->id);
 522         return;
 523     }
 524
 525     job->driver->complete(job, errp);
 526 }
 527
 528 void block_job_user_pause(BlockJob *job)
 529 {
 530     job->user_paused = true;
 531     block_job_pause(job);
 532 }
 533
 534 bool block_job_user_paused(BlockJob *job)
 535 {
 536     return job->user_paused;
 537 }
 538
 539 void block_job_user_resume(BlockJob *job)
 540 {
 541     if (job && job->user_paused && job->pause_count > 0) {
 542         block_job_iostatus_reset(job);
 543         job->user_paused = false;
 544         block_job_resume(job);
 545     }
 546 }
 547
 548 void block_job_cancel(BlockJob *job)
 549 {
 550     if (block_job_started(job)) {
 551         block_job_cancel_async(job);
 552         block_job_enter(job);
 553     } else {
 554         block_job_completed(job, -ECANCELED);
 555     }
 556 }
 557
 558 /* A wrapper around block_job_cancel() taking an Error ** parameter so it may be
 559  * used with block_job_finish_sync() without the need for (rather nasty)
 560  * function pointer casts there. */
 561 static void block_job_cancel_err(BlockJob *job, Error **errp)
 562 {
 563     block_job_cancel(job);
 564 }
 565
 566 int block_job_cancel_sync(BlockJob *job)
 567 {
 568     return block_job_finish_sync(job, &block_job_cancel_err, NULL);
 569 }
 570
 571 void block_job_cancel_sync_all(void)
 572 {
 573     BlockJob *job;
 574     AioContext *aio_context;
 575
 576     while ((job = QLIST_FIRST(&block_jobs))) {
 577         aio_context = blk_get_aio_context(job->blk);
 578         aio_context_acquire(aio_context);
 579         block_job_cancel_sync(job);
 580         aio_context_release(aio_context);
 581     }
 582 }
 583
 584 int block_job_complete_sync(BlockJob *job, Error **errp)
 585 {
 586     return block_job_finish_sync(job, &block_job_complete, errp);
 587 }
 588
 589 BlockJobInfo *block_job_query(BlockJob *job, Error **errp)
 590 {
 591     BlockJobInfo *info;
 592
 593     if (block_job_is_internal(job)) {
 594         error_setg(errp, "Cannot query QEMU internal jobs");
 595         return NULL;
 596     }
 597     info = g_new0(BlockJobInfo, 1);
 598     info->type      = g_strdup(BlockJobType_str(job->driver->job_type));
 599     info->device    = g_strdup(job->id);
 600     info->len       = job->len;
 601     info->busy      = atomic_read(&job->busy);
 602     info->paused    = job->pause_count > 0;
 603     info->offset    = job->offset;
 604     info->speed     = job->speed;
 605     info->io_status = job->iostatus;
 606     info->ready     = job->ready;
 607     return info;
 608 }
 609
 610 static void block_job_iostatus_set_err(BlockJob *job, int error)
 611 {
 612     if (job->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
 613         job->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE :
 614                                           BLOCK_DEVICE_IO_STATUS_FAILED;
 615     }
 616 }
 617
 618 static void block_job_event_cancelled(BlockJob *job)
 619 {
 620     if (block_job_is_internal(job)) {
 621         return;
 622     }
 623
 624     qapi_event_send_block_job_cancelled(job->driver->job_type,
 625                                         job->id,
 626                                         job->len,
 627                                         job->offset,
 628                                         job->speed,
 629                                         &error_abort);
 630 }
 631
 632 static void block_job_event_completed(BlockJob *job, const char *msg)
 633 {
 634     if (block_job_is_internal(job)) {
 635         return;
 636     }
 637
 638     qapi_event_send_block_job_completed(job->driver->job_type,
 639                                         job->id,
 640                                         job->len,
 641                                         job->offset,
 642                                         job->speed,
 643                                         !!msg,
 644                                         msg,
 645                                         &error_abort);
 646 }
 647
 648 /*
 649  * API for block job drivers and the block layer.  These functions are
 650  * declared in blockjob_int.h.
 651  */
 652
 653 void *block_job_create(const char *job_id, const BlockJobDriver *driver,
 654                        BlockDriverState *bs, uint64_t perm,
 655                        uint64_t shared_perm, int64_t speed, int flags,
 656                        BlockCompletionFunc *cb, void *opaque, Error **errp)
 657 {
 658     BlockBackend *blk;
 659     BlockJob *job;
 660     int ret;
 661
 662     if (bs->job) {
 663         error_setg(errp, QERR_DEVICE_IN_USE, bdrv_get_device_name(bs));
 664         return NULL;
 665     }
 666
 667     if (job_id == NULL && !(flags & BLOCK_JOB_INTERNAL)) {
 668         job_id = bdrv_get_device_name(bs);
 669         if (!*job_id) {
 670             error_setg(errp, "An explicit job ID is required for this node");
 671             return NULL;
 672         }
 673     }
 674
 675     if (job_id) {
 676         if (flags & BLOCK_JOB_INTERNAL) {
 677             error_setg(errp, "Cannot specify job ID for internal block job");
 678             return NULL;
 679         }
 680
 681         if (!id_wellformed(job_id)) {
 682             error_setg(errp, "Invalid job ID '%s'", job_id);
 683             return NULL;
 684         }
 685
 686         if (block_job_get(job_id)) {
 687             error_setg(errp, "Job ID '%s' already in use", job_id);
 688             return NULL;
 689         }
 690     }
 691
 692     blk = blk_new(perm, shared_perm);
 693     ret = blk_insert_bs(blk, bs, errp);
 694     if (ret < 0) {
 695         blk_unref(blk);
 696         return NULL;
 697     }
 698
 699     job = g_malloc0(driver->instance_size);
 700     job->driver        = driver;
 701     job->id            = g_strdup(job_id);
 702     job->blk           = blk;
 703     job->cb            = cb;
 704     job->opaque        = opaque;
 705     job->busy          = false;
 706     job->paused        = true;
 707     job->pause_count   = 1;
 708     job->refcnt        = 1;
 709     aio_timer_init(qemu_get_aio_context(), &job->sleep_timer,
 710                    QEMU_CLOCK_REALTIME, SCALE_NS,
 711                    block_job_sleep_timer_cb, job);
 712
 713     error_setg(&job->blocker, "block device is in use by block job: %s",
 714                BlockJobType_str(driver->job_type));
 715     block_job_add_bdrv(job, "main node", bs, 0, BLK_PERM_ALL, &error_abort);
 716     bs->job = job;
 717
 718     blk_set_dev_ops(blk, &block_job_dev_ops, job);
 719     bdrv_op_unblock(bs, BLOCK_OP_TYPE_DATAPLANE, job->blocker);
 720
 721     QLIST_INSERT_HEAD(&block_jobs, job, job_list);
 722
 723     blk_add_aio_context_notifier(blk, block_job_attached_aio_context,
 724                                  block_job_detach_aio_context, job);
 725
 726     /* Only set speed when necessary to avoid NotSupported error */
 727     if (speed != 0) {
 728         Error *local_err = NULL;
 729
 730         block_job_set_speed(job, speed, &local_err);
 731         if (local_err) {
 732             block_job_unref(job);
 733             error_propagate(errp, local_err);
 734             return NULL;
 735         }
 736     }
 737     return job;
 738 }
 739
 740 void block_job_pause_all(void)
 741 {
 742     BlockJob *job = NULL;
 743     while ((job = block_job_next(job))) {
 744         AioContext *aio_context = blk_get_aio_context(job->blk);
 745
 746         aio_context_acquire(aio_context);
 747         block_job_ref(job);
 748         block_job_pause(job);
 749         aio_context_release(aio_context);
 750     }
 751 }
 752
 753 void block_job_early_fail(BlockJob *job)
 754 {
 755     block_job_unref(job);
 756 }
 757
 758 void block_job_completed(BlockJob *job, int ret)
 759 {
 760     assert(blk_bs(job->blk)->job == job);
 761     assert(!job->completed);
 762     job->completed = true;
 763     job->ret = ret;
 764     if (!job->txn) {
 765         block_job_completed_single(job);
 766     } else if (ret < 0 || block_job_is_cancelled(job)) {
 767         block_job_completed_txn_abort(job);
 768     } else {
 769         block_job_completed_txn_success(job);
 770     }
 771 }
 772
 773 static bool block_job_should_pause(BlockJob *job)
 774 {
 775     return job->pause_count > 0;
 776 }
 777
 778 /* Yield, and schedule a timer to reenter the coroutine after @ns nanoseconds.
 779  * Reentering the job coroutine with block_job_enter() before the timer has
 780  * expired is allowed and cancels the timer.
 781  *
 782  * If @ns is (uint64_t) -1, no timer is scheduled and block_job_enter() must be
 783  * called explicitly. */
 784 static void block_job_do_yield(BlockJob *job, uint64_t ns)
 785 {
 786     block_job_lock();
 787     if (ns != -1) {
 788         timer_mod(&job->sleep_timer, ns);
 789     }
 790     job->busy = false;
 791     block_job_unlock();
 792     qemu_coroutine_yield();
 793
 794     /* Set by block_job_enter before re-entering the coroutine.  */
 795     assert(job->busy);
 796 }
 797
 798 void coroutine_fn block_job_pause_point(BlockJob *job)
 799 {
 800     assert(job && block_job_started(job));
 801
 802     if (!block_job_should_pause(job)) {
 803         return;
 804     }
 805     if (block_job_is_cancelled(job)) {
 806         return;
 807     }
 808
 809     if (job->driver->pause) {
 810         job->driver->pause(job);
 811     }
 812
 813     if (block_job_should_pause(job) && !block_job_is_cancelled(job)) {
 814         job->paused = true;
 815         block_job_do_yield(job, -1);
 816         job->paused = false;
 817     }
 818
 819     if (job->driver->resume) {
 820         job->driver->resume(job);
 821     }
 822 }
 823
 824 void block_job_resume_all(void)
 825 {
 826     BlockJob *job, *next;
 827
 828     QLIST_FOREACH_SAFE(job, &block_jobs, job_list, next) {
 829         AioContext *aio_context = blk_get_aio_context(job->blk);
 830
 831         aio_context_acquire(aio_context);
 832         block_job_resume(job);
 833         block_job_unref(job);
 834         aio_context_release(aio_context);
 835     }
 836 }
 837
 838 /*
 839  * Conditionally enter a block_job pending a call to fn() while
 840  * under the block_job_lock critical section.
 841  */
 842 static void block_job_enter_cond(BlockJob *job, bool(*fn)(BlockJob *job))
 843 {
 844     if (!block_job_started(job)) {
 845         return;
 846     }
 847     if (job->deferred_to_main_loop) {
 848         return;
 849     }
 850
 851     block_job_lock();
 852     if (job->busy) {
 853         block_job_unlock();
 854         return;
 855     }
 856
 857     if (fn && !fn(job)) {
 858         block_job_unlock();
 859         return;
 860     }
 861
 862     assert(!job->deferred_to_main_loop);
 863     timer_del(&job->sleep_timer);
 864     job->busy = true;
 865     block_job_unlock();
 866     aio_co_wake(job->co);
 867 }
 868
 869 void block_job_enter(BlockJob *job)
 870 {
 871     block_job_enter_cond(job, NULL);
 872 }
 873
 874 bool block_job_is_cancelled(BlockJob *job)
 875 {
 876     return job->cancelled;
 877 }
 878
 879 void block_job_sleep_ns(BlockJob *job, int64_t ns)
 880 {
 881     assert(job->busy);
 882
 883     /* Check cancellation *before* setting busy = false, too!  */
 884     if (block_job_is_cancelled(job)) {
 885         return;
 886     }
 887
 888     if (!block_job_should_pause(job)) {
 889         block_job_do_yield(job, qemu_clock_get_ns(QEMU_CLOCK_REALTIME) + ns);
 890     }
 891
 892     block_job_pause_point(job);
 893 }
 894
 895 void block_job_yield(BlockJob *job)
 896 {
 897     assert(job->busy);
 898
 899     /* Check cancellation *before* setting busy = false, too!  */
 900     if (block_job_is_cancelled(job)) {
 901         return;
 902     }
 903
 904     if (!block_job_should_pause(job)) {
 905         block_job_do_yield(job, -1);
 906     }
 907
 908     block_job_pause_point(job);
 909 }
 910
 911 void block_job_iostatus_reset(BlockJob *job)
 912 {
 913     if (job->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
 914         return;
 915     }
 916     assert(job->user_paused && job->pause_count > 0);
 917     job->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
 918 }
 919
 920 void block_job_event_ready(BlockJob *job)
 921 {
 922     job->ready = true;
 923
 924     if (block_job_is_internal(job)) {
 925         return;
 926     }
 927
 928     qapi_event_send_block_job_ready(job->driver->job_type,
 929                                     job->id,
 930                                     job->len,
 931                                     job->offset,
 932                                     job->speed, &error_abort);
 933 }
 934
 935 BlockErrorAction block_job_error_action(BlockJob *job, BlockdevOnError on_err,
 936                                         int is_read, int error)
 937 {
 938     BlockErrorAction action;
 939
 940     switch (on_err) {
 941     case BLOCKDEV_ON_ERROR_ENOSPC:
 942     case BLOCKDEV_ON_ERROR_AUTO:
 943         action = (error == ENOSPC) ?
 944                  BLOCK_ERROR_ACTION_STOP : BLOCK_ERROR_ACTION_REPORT;
 945         break;
 946     case BLOCKDEV_ON_ERROR_STOP:
 947         action = BLOCK_ERROR_ACTION_STOP;
 948         break;
 949     case BLOCKDEV_ON_ERROR_REPORT:
 950         action = BLOCK_ERROR_ACTION_REPORT;
 951         break;
 952     case BLOCKDEV_ON_ERROR_IGNORE:
 953         action = BLOCK_ERROR_ACTION_IGNORE;
 954         break;
 955     default:
 956         abort();
 957     }
 958     if (!block_job_is_internal(job)) {
 959         qapi_event_send_block_job_error(job->id,
 960                                         is_read ? IO_OPERATION_TYPE_READ :
 961                                         IO_OPERATION_TYPE_WRITE,
 962                                         action, &error_abort);
 963     }
 964     if (action == BLOCK_ERROR_ACTION_STOP) {
 965         /* make the pause user visible, which will be resumed from QMP. */
 966         block_job_user_pause(job);
 967         block_job_iostatus_set_err(job, error);
 968     }
 969     return action;
 970 }
 971
 972 typedef struct {
 973     BlockJob *job;
 974     AioContext *aio_context;
 975     BlockJobDeferToMainLoopFn *fn;
 976     void *opaque;
 977 } BlockJobDeferToMainLoopData;
 978
 979 static void block_job_defer_to_main_loop_bh(void *opaque)
 980 {
 981     BlockJobDeferToMainLoopData *data = opaque;
 982     AioContext *aio_context;
 983
 984     /* Prevent race with block_job_defer_to_main_loop() */
 985     aio_context_acquire(data->aio_context);
 986
 987     /* Fetch BDS AioContext again, in case it has changed */
 988     aio_context = blk_get_aio_context(data->job->blk);
 989     if (aio_context != data->aio_context) {
 990         aio_context_acquire(aio_context);
 991     }
 992
 993     data->fn(data->job, data->opaque);
 994
 995     if (aio_context != data->aio_context) {
 996         aio_context_release(aio_context);
 997     }
 998
 999     aio_context_release(data->aio_context);
1000
1001     g_free(data);
1002 }
1003
1004 void block_job_defer_to_main_loop(BlockJob *job,
1005                                   BlockJobDeferToMainLoopFn *fn,
1006                                   void *opaque)
1007 {
1008     BlockJobDeferToMainLoopData *data = g_malloc(sizeof(*data));
1009     data->job = job;
1010     data->aio_context = blk_get_aio_context(job->blk);
1011     data->fn = fn;
1012     data->opaque = opaque;
1013     job->deferred_to_main_loop = true;
1014
1015     aio_bh_schedule_oneshot(qemu_get_aio_context(),
1016                             block_job_defer_to_main_loop_bh, data);
1017 }