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     QLIST_REMOVE(job, txn_list);
 361     block_job_txn_unref(job->txn);
 362     block_job_unref(job);
 363 }
 364
 365 static void block_job_cancel_async(BlockJob *job)
 366 {
 367     if (job->iostatus != BLOCK_DEVICE_IO_STATUS_OK) {
 368         block_job_iostatus_reset(job);
 369     }
 370     if (job->user_paused) {
 371         /* Do not call block_job_enter here, the caller will handle it.  */
 372         job->user_paused = false;
 373         job->pause_count--;
 374     }
 375     job->cancelled = true;
 376 }
 377
 378 static int block_job_finish_sync(BlockJob *job,
 379                                  void (*finish)(BlockJob *, Error **errp),
 380                                  Error **errp)
 381 {
 382     Error *local_err = NULL;
 383     int ret;
 384
 385     assert(blk_bs(job->blk)->job == job);
 386
 387     block_job_ref(job);
 388
 389     if (finish) {
 390         finish(job, &local_err);
 391     }
 392     if (local_err) {
 393         error_propagate(errp, local_err);
 394         block_job_unref(job);
 395         return -EBUSY;
 396     }
 397     /* block_job_drain calls block_job_enter, and it should be enough to
 398      * induce progress until the job completes or moves to the main thread.
 399     */
 400     while (!job->deferred_to_main_loop && !job->completed) {
 401         block_job_drain(job);
 402     }
 403     while (!job->completed) {
 404         aio_poll(qemu_get_aio_context(), true);
 405     }
 406     ret = (job->cancelled && job->ret == 0) ? -ECANCELED : job->ret;
 407     block_job_unref(job);
 408     return ret;
 409 }
 410
 411 static void block_job_completed_txn_abort(BlockJob *job)
 412 {
 413     AioContext *ctx;
 414     BlockJobTxn *txn = job->txn;
 415     BlockJob *other_job;
 416
 417     if (txn->aborting) {
 418         /*
 419          * We are cancelled by another job, which will handle everything.
 420          */
 421         return;
 422     }
 423     txn->aborting = true;
 424     block_job_txn_ref(txn);
 425
 426     /* We are the first failed job. Cancel other jobs. */
 427     QLIST_FOREACH(other_job, &txn->jobs, txn_list) {
 428         ctx = blk_get_aio_context(other_job->blk);
 429         aio_context_acquire(ctx);
 430     }
 431
 432     /* Other jobs are effectively cancelled by us, set the status for
 433      * them; this job, however, may or may not be cancelled, depending
 434      * on the caller, so leave it. */
 435     QLIST_FOREACH(other_job, &txn->jobs, txn_list) {
 436         if (other_job != job) {
 437             block_job_cancel_async(other_job);
 438         }
 439     }
 440     while (!QLIST_EMPTY(&txn->jobs)) {
 441         other_job = QLIST_FIRST(&txn->jobs);
 442         ctx = blk_get_aio_context(other_job->blk);
 443         if (!other_job->completed) {
 444             assert(other_job->cancelled);
 445             block_job_finish_sync(other_job, NULL, NULL);
 446         }
 447         block_job_completed_single(other_job);
 448         aio_context_release(ctx);
 449     }
 450
 451     block_job_txn_unref(txn);
 452 }
 453
 454 static void block_job_completed_txn_success(BlockJob *job)
 455 {
 456     AioContext *ctx;
 457     BlockJobTxn *txn = job->txn;
 458     BlockJob *other_job, *next;
 459     /*
 460      * Successful completion, see if there are other running jobs in this
 461      * txn.
 462      */
 463     QLIST_FOREACH(other_job, &txn->jobs, txn_list) {
 464         if (!other_job->completed) {
 465             return;
 466         }
 467     }
 468     /* We are the last completed job, commit the transaction. */
 469     QLIST_FOREACH_SAFE(other_job, &txn->jobs, txn_list, next) {
 470         ctx = blk_get_aio_context(other_job->blk);
 471         aio_context_acquire(ctx);
 472         assert(other_job->ret == 0);
 473         block_job_completed_single(other_job);
 474         aio_context_release(ctx);
 475     }
 476 }
 477
 478 /* Assumes the block_job_mutex is held */
 479 static bool block_job_timer_pending(BlockJob *job)
 480 {
 481     return timer_pending(&job->sleep_timer);
 482 }
 483
 484 void block_job_set_speed(BlockJob *job, int64_t speed, Error **errp)
 485 {
 486     Error *local_err = NULL;
 487     int64_t old_speed = job->speed;
 488
 489     if (!job->driver->set_speed) {
 490         error_setg(errp, QERR_UNSUPPORTED);
 491         return;
 492     }
 493     job->driver->set_speed(job, speed, &local_err);
 494     if (local_err) {
 495         error_propagate(errp, local_err);
 496         return;
 497     }
 498
 499     job->speed = speed;
 500     if (speed && speed <= old_speed) {
 501         return;
 502     }
 503
 504     /* kick only if a timer is pending */
 505     block_job_enter_cond(job, block_job_timer_pending);
 506 }
 507
 508 void block_job_complete(BlockJob *job, Error **errp)
 509 {
 510     /* Should not be reachable via external interface for internal jobs */
 511     assert(job->id);
 512     if (job->pause_count || job->cancelled ||
 513         !block_job_started(job) || !job->driver->complete) {
 514         error_setg(errp, "The active block job '%s' cannot be completed",
 515                    job->id);
 516         return;
 517     }
 518
 519     job->driver->complete(job, errp);
 520 }
 521
 522 void block_job_user_pause(BlockJob *job)
 523 {
 524     job->user_paused = true;
 525     block_job_pause(job);
 526 }
 527
 528 bool block_job_user_paused(BlockJob *job)
 529 {
 530     return job->user_paused;
 531 }
 532
 533 void block_job_user_resume(BlockJob *job)
 534 {
 535     if (job && job->user_paused && job->pause_count > 0) {
 536         block_job_iostatus_reset(job);
 537         job->user_paused = false;
 538         block_job_resume(job);
 539     }
 540 }
 541
 542 void block_job_cancel(BlockJob *job)
 543 {
 544     if (block_job_started(job)) {
 545         block_job_cancel_async(job);
 546         block_job_enter(job);
 547     } else {
 548         block_job_completed(job, -ECANCELED);
 549     }
 550 }
 551
 552 /* A wrapper around block_job_cancel() taking an Error ** parameter so it may be
 553  * used with block_job_finish_sync() without the need for (rather nasty)
 554  * function pointer casts there. */
 555 static void block_job_cancel_err(BlockJob *job, Error **errp)
 556 {
 557     block_job_cancel(job);
 558 }
 559
 560 int block_job_cancel_sync(BlockJob *job)
 561 {
 562     return block_job_finish_sync(job, &block_job_cancel_err, NULL);
 563 }
 564
 565 void block_job_cancel_sync_all(void)
 566 {
 567     BlockJob *job;
 568     AioContext *aio_context;
 569
 570     while ((job = QLIST_FIRST(&block_jobs))) {
 571         aio_context = blk_get_aio_context(job->blk);
 572         aio_context_acquire(aio_context);
 573         block_job_cancel_sync(job);
 574         aio_context_release(aio_context);
 575     }
 576 }
 577
 578 int block_job_complete_sync(BlockJob *job, Error **errp)
 579 {
 580     return block_job_finish_sync(job, &block_job_complete, errp);
 581 }
 582
 583 BlockJobInfo *block_job_query(BlockJob *job, Error **errp)
 584 {
 585     BlockJobInfo *info;
 586
 587     if (block_job_is_internal(job)) {
 588         error_setg(errp, "Cannot query QEMU internal jobs");
 589         return NULL;
 590     }
 591     info = g_new0(BlockJobInfo, 1);
 592     info->type      = g_strdup(BlockJobType_str(job->driver->job_type));
 593     info->device    = g_strdup(job->id);
 594     info->len       = job->len;
 595     info->busy      = atomic_read(&job->busy);
 596     info->paused    = job->pause_count > 0;
 597     info->offset    = job->offset;
 598     info->speed     = job->speed;
 599     info->io_status = job->iostatus;
 600     info->ready     = job->ready;
 601     return info;
 602 }
 603
 604 static void block_job_iostatus_set_err(BlockJob *job, int error)
 605 {
 606     if (job->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
 607         job->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE :
 608                                           BLOCK_DEVICE_IO_STATUS_FAILED;
 609     }
 610 }
 611
 612 static void block_job_event_cancelled(BlockJob *job)
 613 {
 614     if (block_job_is_internal(job)) {
 615         return;
 616     }
 617
 618     qapi_event_send_block_job_cancelled(job->driver->job_type,
 619                                         job->id,
 620                                         job->len,
 621                                         job->offset,
 622                                         job->speed,
 623                                         &error_abort);
 624 }
 625
 626 static void block_job_event_completed(BlockJob *job, const char *msg)
 627 {
 628     if (block_job_is_internal(job)) {
 629         return;
 630     }
 631
 632     qapi_event_send_block_job_completed(job->driver->job_type,
 633                                         job->id,
 634                                         job->len,
 635                                         job->offset,
 636                                         job->speed,
 637                                         !!msg,
 638                                         msg,
 639                                         &error_abort);
 640 }
 641
 642 /*
 643  * API for block job drivers and the block layer.  These functions are
 644  * declared in blockjob_int.h.
 645  */
 646
 647 void *block_job_create(const char *job_id, const BlockJobDriver *driver,
 648                        BlockJobTxn *txn, BlockDriverState *bs, uint64_t perm,
 649                        uint64_t shared_perm, int64_t speed, int flags,
 650                        BlockCompletionFunc *cb, void *opaque, Error **errp)
 651 {
 652     BlockBackend *blk;
 653     BlockJob *job;
 654     int ret;
 655
 656     if (bs->job) {
 657         error_setg(errp, QERR_DEVICE_IN_USE, bdrv_get_device_name(bs));
 658         return NULL;
 659     }
 660
 661     if (job_id == NULL && !(flags & BLOCK_JOB_INTERNAL)) {
 662         job_id = bdrv_get_device_name(bs);
 663         if (!*job_id) {
 664             error_setg(errp, "An explicit job ID is required for this node");
 665             return NULL;
 666         }
 667     }
 668
 669     if (job_id) {
 670         if (flags & BLOCK_JOB_INTERNAL) {
 671             error_setg(errp, "Cannot specify job ID for internal block job");
 672             return NULL;
 673         }
 674
 675         if (!id_wellformed(job_id)) {
 676             error_setg(errp, "Invalid job ID '%s'", job_id);
 677             return NULL;
 678         }
 679
 680         if (block_job_get(job_id)) {
 681             error_setg(errp, "Job ID '%s' already in use", job_id);
 682             return NULL;
 683         }
 684     }
 685
 686     blk = blk_new(perm, shared_perm);
 687     ret = blk_insert_bs(blk, bs, errp);
 688     if (ret < 0) {
 689         blk_unref(blk);
 690         return NULL;
 691     }
 692
 693     job = g_malloc0(driver->instance_size);
 694     job->driver        = driver;
 695     job->id            = g_strdup(job_id);
 696     job->blk           = blk;
 697     job->cb            = cb;
 698     job->opaque        = opaque;
 699     job->busy          = false;
 700     job->paused        = true;
 701     job->pause_count   = 1;
 702     job->refcnt        = 1;
 703     aio_timer_init(qemu_get_aio_context(), &job->sleep_timer,
 704                    QEMU_CLOCK_REALTIME, SCALE_NS,
 705                    block_job_sleep_timer_cb, job);
 706
 707     error_setg(&job->blocker, "block device is in use by block job: %s",
 708                BlockJobType_str(driver->job_type));
 709     block_job_add_bdrv(job, "main node", bs, 0, BLK_PERM_ALL, &error_abort);
 710     bs->job = job;
 711
 712     bdrv_op_unblock(bs, BLOCK_OP_TYPE_DATAPLANE, job->blocker);
 713
 714     QLIST_INSERT_HEAD(&block_jobs, job, job_list);
 715
 716     blk_add_aio_context_notifier(blk, block_job_attached_aio_context,
 717                                  block_job_detach_aio_context, job);
 718
 719     /* Only set speed when necessary to avoid NotSupported error */
 720     if (speed != 0) {
 721         Error *local_err = NULL;
 722
 723         block_job_set_speed(job, speed, &local_err);
 724         if (local_err) {
 725             block_job_unref(job);
 726             error_propagate(errp, local_err);
 727             return NULL;
 728         }
 729     }
 730
 731     /* Single jobs are modeled as single-job transactions for sake of
 732      * consolidating the job management logic */
 733     if (!txn) {
 734         txn = block_job_txn_new();
 735         block_job_txn_add_job(txn, job);
 736         block_job_txn_unref(txn);
 737     } else {
 738         block_job_txn_add_job(txn, job);
 739     }
 740
 741     return job;
 742 }
 743
 744 void block_job_pause_all(void)
 745 {
 746     BlockJob *job = NULL;
 747     while ((job = block_job_next(job))) {
 748         AioContext *aio_context = blk_get_aio_context(job->blk);
 749
 750         aio_context_acquire(aio_context);
 751         block_job_ref(job);
 752         block_job_pause(job);
 753         aio_context_release(aio_context);
 754     }
 755 }
 756
 757 void block_job_early_fail(BlockJob *job)
 758 {
 759     block_job_unref(job);
 760 }
 761
 762 void block_job_completed(BlockJob *job, int ret)
 763 {
 764     assert(job && job->txn && !job->completed);
 765     assert(blk_bs(job->blk)->job == job);
 766     job->completed = true;
 767     job->ret = ret;
 768     if (ret < 0 || block_job_is_cancelled(job)) {
 769         block_job_completed_txn_abort(job);
 770     } else {
 771         block_job_completed_txn_success(job);
 772     }
 773 }
 774
 775 static bool block_job_should_pause(BlockJob *job)
 776 {
 777     return job->pause_count > 0;
 778 }
 779
 780 /* Yield, and schedule a timer to reenter the coroutine after @ns nanoseconds.
 781  * Reentering the job coroutine with block_job_enter() before the timer has
 782  * expired is allowed and cancels the timer.
 783  *
 784  * If @ns is (uint64_t) -1, no timer is scheduled and block_job_enter() must be
 785  * called explicitly. */
 786 static void block_job_do_yield(BlockJob *job, uint64_t ns)
 787 {
 788     block_job_lock();
 789     if (ns != -1) {
 790         timer_mod(&job->sleep_timer, ns);
 791     }
 792     job->busy = false;
 793     block_job_unlock();
 794     qemu_coroutine_yield();
 795
 796     /* Set by block_job_enter before re-entering the coroutine.  */
 797     assert(job->busy);
 798 }
 799
 800 void coroutine_fn block_job_pause_point(BlockJob *job)
 801 {
 802     assert(job && block_job_started(job));
 803
 804     if (!block_job_should_pause(job)) {
 805         return;
 806     }
 807     if (block_job_is_cancelled(job)) {
 808         return;
 809     }
 810
 811     if (job->driver->pause) {
 812         job->driver->pause(job);
 813     }
 814
 815     if (block_job_should_pause(job) && !block_job_is_cancelled(job)) {
 816         job->paused = true;
 817         block_job_do_yield(job, -1);
 818         job->paused = false;
 819     }
 820
 821     if (job->driver->resume) {
 822         job->driver->resume(job);
 823     }
 824 }
 825
 826 void block_job_resume_all(void)
 827 {
 828     BlockJob *job, *next;
 829
 830     QLIST_FOREACH_SAFE(job, &block_jobs, job_list, next) {
 831         AioContext *aio_context = blk_get_aio_context(job->blk);
 832
 833         aio_context_acquire(aio_context);
 834         block_job_resume(job);
 835         block_job_unref(job);
 836         aio_context_release(aio_context);
 837     }
 838 }
 839
 840 /*
 841  * Conditionally enter a block_job pending a call to fn() while
 842  * under the block_job_lock critical section.
 843  */
 844 static void block_job_enter_cond(BlockJob *job, bool(*fn)(BlockJob *job))
 845 {
 846     if (!block_job_started(job)) {
 847         return;
 848     }
 849     if (job->deferred_to_main_loop) {
 850         return;
 851     }
 852
 853     block_job_lock();
 854     if (job->busy) {
 855         block_job_unlock();
 856         return;
 857     }
 858
 859     if (fn && !fn(job)) {
 860         block_job_unlock();
 861         return;
 862     }
 863
 864     assert(!job->deferred_to_main_loop);
 865     timer_del(&job->sleep_timer);
 866     job->busy = true;
 867     block_job_unlock();
 868     aio_co_wake(job->co);
 869 }
 870
 871 void block_job_enter(BlockJob *job)
 872 {
 873     block_job_enter_cond(job, NULL);
 874 }
 875
 876 bool block_job_is_cancelled(BlockJob *job)
 877 {
 878     return job->cancelled;
 879 }
 880
 881 void block_job_sleep_ns(BlockJob *job, int64_t ns)
 882 {
 883     assert(job->busy);
 884
 885     /* Check cancellation *before* setting busy = false, too!  */
 886     if (block_job_is_cancelled(job)) {
 887         return;
 888     }
 889
 890     if (!block_job_should_pause(job)) {
 891         block_job_do_yield(job, qemu_clock_get_ns(QEMU_CLOCK_REALTIME) + ns);
 892     }
 893
 894     block_job_pause_point(job);
 895 }
 896
 897 void block_job_yield(BlockJob *job)
 898 {
 899     assert(job->busy);
 900
 901     /* Check cancellation *before* setting busy = false, too!  */
 902     if (block_job_is_cancelled(job)) {
 903         return;
 904     }
 905
 906     if (!block_job_should_pause(job)) {
 907         block_job_do_yield(job, -1);
 908     }
 909
 910     block_job_pause_point(job);
 911 }
 912
 913 void block_job_iostatus_reset(BlockJob *job)
 914 {
 915     if (job->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
 916         return;
 917     }
 918     assert(job->user_paused && job->pause_count > 0);
 919     job->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
 920 }
 921
 922 void block_job_event_ready(BlockJob *job)
 923 {
 924     job->ready = true;
 925
 926     if (block_job_is_internal(job)) {
 927         return;
 928     }
 929
 930     qapi_event_send_block_job_ready(job->driver->job_type,
 931                                     job->id,
 932                                     job->len,
 933                                     job->offset,
 934                                     job->speed, &error_abort);
 935 }
 936
 937 BlockErrorAction block_job_error_action(BlockJob *job, BlockdevOnError on_err,
 938                                         int is_read, int error)
 939 {
 940     BlockErrorAction action;
 941
 942     switch (on_err) {
 943     case BLOCKDEV_ON_ERROR_ENOSPC:
 944     case BLOCKDEV_ON_ERROR_AUTO:
 945         action = (error == ENOSPC) ?
 946                  BLOCK_ERROR_ACTION_STOP : BLOCK_ERROR_ACTION_REPORT;
 947         break;
 948     case BLOCKDEV_ON_ERROR_STOP:
 949         action = BLOCK_ERROR_ACTION_STOP;
 950         break;
 951     case BLOCKDEV_ON_ERROR_REPORT:
 952         action = BLOCK_ERROR_ACTION_REPORT;
 953         break;
 954     case BLOCKDEV_ON_ERROR_IGNORE:
 955         action = BLOCK_ERROR_ACTION_IGNORE;
 956         break;
 957     default:
 958         abort();
 959     }
 960     if (!block_job_is_internal(job)) {
 961         qapi_event_send_block_job_error(job->id,
 962                                         is_read ? IO_OPERATION_TYPE_READ :
 963                                         IO_OPERATION_TYPE_WRITE,
 964                                         action, &error_abort);
 965     }
 966     if (action == BLOCK_ERROR_ACTION_STOP) {
 967         /* make the pause user visible, which will be resumed from QMP. */
 968         block_job_user_pause(job);
 969         block_job_iostatus_set_err(job, error);
 970     }
 971     return action;
 972 }
 973
 974 typedef struct {
 975     BlockJob *job;
 976     AioContext *aio_context;
 977     BlockJobDeferToMainLoopFn *fn;
 978     void *opaque;
 979 } BlockJobDeferToMainLoopData;
 980
 981 static void block_job_defer_to_main_loop_bh(void *opaque)
 982 {
 983     BlockJobDeferToMainLoopData *data = opaque;
 984     AioContext *aio_context;
 985
 986     /* Prevent race with block_job_defer_to_main_loop() */
 987     aio_context_acquire(data->aio_context);
 988
 989     /* Fetch BDS AioContext again, in case it has changed */
 990     aio_context = blk_get_aio_context(data->job->blk);
 991     if (aio_context != data->aio_context) {
 992         aio_context_acquire(aio_context);
 993     }
 994
 995     data->fn(data->job, data->opaque);
 996
 997     if (aio_context != data->aio_context) {
 998         aio_context_release(aio_context);
 999     }
1000
1001     aio_context_release(data->aio_context);
1002
1003     g_free(data);
1004 }
1005
1006 void block_job_defer_to_main_loop(BlockJob *job,
1007                                   BlockJobDeferToMainLoopFn *fn,
1008                                   void *opaque)
1009 {
1010     BlockJobDeferToMainLoopData *data = g_malloc(sizeof(*data));
1011     data->job = job;
1012     data->aio_context = blk_get_aio_context(job->blk);
1013     data->fn = fn;
1014     data->opaque = opaque;
1015     job->deferred_to_main_loop = true;
1016
1017     aio_bh_schedule_oneshot(qemu_get_aio_context(),
1018                             block_job_defer_to_main_loop_bh, data);
1019 }