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