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 "trace.h"
  29 #include "block/block.h"
  30 #include "block/blockjob.h"
  31 #include "block/block_int.h"
  32 #include "sysemu/block-backend.h"
  33 #include "qapi/qmp/qerror.h"
  34 #include "qapi/qmp/qjson.h"
  35 #include "qemu/coroutine.h"
  36 #include "qemu/id.h"
  37 #include "qmp-commands.h"
  38 #include "qemu/timer.h"
  39 #include "qapi-event.h"
  40
  41 /* Transactional group of block jobs */
  42 struct BlockJobTxn {
  43
  44     /* Is this txn being cancelled? */
  45     bool aborting;
  46
  47     /* List of jobs */
  48     QLIST_HEAD(, BlockJob) jobs;
  49
  50     /* Reference count */
  51     int refcnt;
  52 };
  53
  54 static QLIST_HEAD(, BlockJob) block_jobs = QLIST_HEAD_INITIALIZER(block_jobs);
  55
  56 BlockJob *block_job_next(BlockJob *job)
  57 {
  58     if (!job) {
  59         return QLIST_FIRST(&block_jobs);
  60     }
  61     return QLIST_NEXT(job, job_list);
  62 }
  63
  64 BlockJob *block_job_get(const char *id)
  65 {
  66     BlockJob *job;
  67
  68     QLIST_FOREACH(job, &block_jobs, job_list) {
  69         if (job->id && !strcmp(id, job->id)) {
  70             return job;
  71         }
  72     }
  73
  74     return NULL;
  75 }
  76
  77 static void block_job_attached_aio_context(AioContext *new_context,
  78                                            void *opaque)
  79 {
  80     BlockJob *job = opaque;
  81
  82     if (job->driver->attached_aio_context) {
  83         job->driver->attached_aio_context(job, new_context);
  84     }
  85
  86     block_job_resume(job);
  87 }
  88
  89 static void block_job_drain(BlockJob *job)
  90 {
  91     /* If job is !job->busy this kicks it into the next pause point. */
  92     block_job_enter(job);
  93
  94     blk_drain(job->blk);
  95     if (job->driver->drain) {
  96         job->driver->drain(job);
  97     }
  98 }
  99
 100 static void block_job_detach_aio_context(void *opaque)
 101 {
 102     BlockJob *job = opaque;
 103
 104     /* In case the job terminates during aio_poll()... */
 105     block_job_ref(job);
 106
 107     block_job_pause(job);
 108
 109     while (!job->paused && !job->completed) {
 110         block_job_drain(job);
 111     }
 112
 113     block_job_unref(job);
 114 }
 115
 116 void block_job_add_bdrv(BlockJob *job, BlockDriverState *bs)
 117 {
 118     job->nodes = g_slist_prepend(job->nodes, bs);
 119     bdrv_ref(bs);
 120     bdrv_op_block_all(bs, job->blocker);
 121 }
 122
 123 void *block_job_create(const char *job_id, const BlockJobDriver *driver,
 124                        BlockDriverState *bs, int64_t speed,
 125                        BlockCompletionFunc *cb, void *opaque, Error **errp)
 126 {
 127     BlockBackend *blk;
 128     BlockJob *job;
 129
 130     assert(cb);
 131     if (bs->job) {
 132         error_setg(errp, QERR_DEVICE_IN_USE, bdrv_get_device_name(bs));
 133         return NULL;
 134     }
 135
 136     if (job_id == NULL) {
 137         job_id = bdrv_get_device_name(bs);
 138         if (!*job_id) {
 139             error_setg(errp, "An explicit job ID is required for this node");
 140             return NULL;
 141         }
 142     }
 143
 144     if (!id_wellformed(job_id)) {
 145         error_setg(errp, "Invalid job ID '%s'", job_id);
 146         return NULL;
 147     }
 148
 149     if (block_job_get(job_id)) {
 150         error_setg(errp, "Job ID '%s' already in use", job_id);
 151         return NULL;
 152     }
 153
 154     blk = blk_new();
 155     blk_insert_bs(blk, bs);
 156
 157     job = g_malloc0(driver->instance_size);
 158     error_setg(&job->blocker, "block device is in use by block job: %s",
 159                BlockJobType_lookup[driver->job_type]);
 160     block_job_add_bdrv(job, bs);
 161     bdrv_op_unblock(bs, BLOCK_OP_TYPE_DATAPLANE, job->blocker);
 162
 163     job->driver        = driver;
 164     job->id            = g_strdup(job_id);
 165     job->blk           = blk;
 166     job->cb            = cb;
 167     job->opaque        = opaque;
 168     job->busy          = true;
 169     job->refcnt        = 1;
 170     bs->job = job;
 171
 172     QLIST_INSERT_HEAD(&block_jobs, job, job_list);
 173
 174     blk_add_aio_context_notifier(blk, block_job_attached_aio_context,
 175                                  block_job_detach_aio_context, job);
 176
 177     /* Only set speed when necessary to avoid NotSupported error */
 178     if (speed != 0) {
 179         Error *local_err = NULL;
 180
 181         block_job_set_speed(job, speed, &local_err);
 182         if (local_err) {
 183             block_job_unref(job);
 184             error_propagate(errp, local_err);
 185             return NULL;
 186         }
 187     }
 188     return job;
 189 }
 190
 191 bool block_job_is_internal(BlockJob *job)
 192 {
 193     return (job->id == NULL);
 194 }
 195
 196 void block_job_ref(BlockJob *job)
 197 {
 198     ++job->refcnt;
 199 }
 200
 201 void block_job_unref(BlockJob *job)
 202 {
 203     if (--job->refcnt == 0) {
 204         GSList *l;
 205         BlockDriverState *bs = blk_bs(job->blk);
 206         bs->job = NULL;
 207         for (l = job->nodes; l; l = l->next) {
 208             bs = l->data;
 209             bdrv_op_unblock_all(bs, job->blocker);
 210             bdrv_unref(bs);
 211         }
 212         g_slist_free(job->nodes);
 213         blk_remove_aio_context_notifier(job->blk,
 214                                         block_job_attached_aio_context,
 215                                         block_job_detach_aio_context, job);
 216         blk_unref(job->blk);
 217         error_free(job->blocker);
 218         g_free(job->id);
 219         QLIST_REMOVE(job, job_list);
 220         g_free(job);
 221     }
 222 }
 223
 224 static void block_job_completed_single(BlockJob *job)
 225 {
 226     if (!job->ret) {
 227         if (job->driver->commit) {
 228             job->driver->commit(job);
 229         }
 230     } else {
 231         if (job->driver->abort) {
 232             job->driver->abort(job);
 233         }
 234     }
 235     job->cb(job->opaque, job->ret);
 236     if (job->txn) {
 237         block_job_txn_unref(job->txn);
 238     }
 239     block_job_unref(job);
 240 }
 241
 242 static void block_job_completed_txn_abort(BlockJob *job)
 243 {
 244     AioContext *ctx;
 245     BlockJobTxn *txn = job->txn;
 246     BlockJob *other_job, *next;
 247
 248     if (txn->aborting) {
 249         /*
 250          * We are cancelled by another job, which will handle everything.
 251          */
 252         return;
 253     }
 254     txn->aborting = true;
 255     /* We are the first failed job. Cancel other jobs. */
 256     QLIST_FOREACH(other_job, &txn->jobs, txn_list) {
 257         ctx = blk_get_aio_context(other_job->blk);
 258         aio_context_acquire(ctx);
 259     }
 260     QLIST_FOREACH(other_job, &txn->jobs, txn_list) {
 261         if (other_job == job || other_job->completed) {
 262             /* Other jobs are "effectively" cancelled by us, set the status for
 263              * them; this job, however, may or may not be cancelled, depending
 264              * on the caller, so leave it. */
 265             if (other_job != job) {
 266                 other_job->cancelled = true;
 267             }
 268             continue;
 269         }
 270         block_job_cancel_sync(other_job);
 271         assert(other_job->completed);
 272     }
 273     QLIST_FOREACH_SAFE(other_job, &txn->jobs, txn_list, next) {
 274         ctx = blk_get_aio_context(other_job->blk);
 275         block_job_completed_single(other_job);
 276         aio_context_release(ctx);
 277     }
 278 }
 279
 280 static void block_job_completed_txn_success(BlockJob *job)
 281 {
 282     AioContext *ctx;
 283     BlockJobTxn *txn = job->txn;
 284     BlockJob *other_job, *next;
 285     /*
 286      * Successful completion, see if there are other running jobs in this
 287      * txn.
 288      */
 289     QLIST_FOREACH(other_job, &txn->jobs, txn_list) {
 290         if (!other_job->completed) {
 291             return;
 292         }
 293     }
 294     /* We are the last completed job, commit the transaction. */
 295     QLIST_FOREACH_SAFE(other_job, &txn->jobs, txn_list, next) {
 296         ctx = blk_get_aio_context(other_job->blk);
 297         aio_context_acquire(ctx);
 298         assert(other_job->ret == 0);
 299         block_job_completed_single(other_job);
 300         aio_context_release(ctx);
 301     }
 302 }
 303
 304 void block_job_completed(BlockJob *job, int ret)
 305 {
 306     assert(blk_bs(job->blk)->job == job);
 307     assert(!job->completed);
 308     job->completed = true;
 309     job->ret = ret;
 310     if (!job->txn) {
 311         block_job_completed_single(job);
 312     } else if (ret < 0 || block_job_is_cancelled(job)) {
 313         block_job_completed_txn_abort(job);
 314     } else {
 315         block_job_completed_txn_success(job);
 316     }
 317 }
 318
 319 void block_job_set_speed(BlockJob *job, int64_t speed, Error **errp)
 320 {
 321     Error *local_err = NULL;
 322
 323     if (!job->driver->set_speed) {
 324         error_setg(errp, QERR_UNSUPPORTED);
 325         return;
 326     }
 327     job->driver->set_speed(job, speed, &local_err);
 328     if (local_err) {
 329         error_propagate(errp, local_err);
 330         return;
 331     }
 332
 333     job->speed = speed;
 334 }
 335
 336 void block_job_complete(BlockJob *job, Error **errp)
 337 {
 338     /* Should not be reachable via external interface for internal jobs */
 339     assert(job->id);
 340     if (job->pause_count || job->cancelled || !job->driver->complete) {
 341         error_setg(errp, "The active block job '%s' cannot be completed",
 342                    job->id);
 343         return;
 344     }
 345
 346     job->driver->complete(job, errp);
 347 }
 348
 349 void block_job_pause(BlockJob *job)
 350 {
 351     job->pause_count++;
 352 }
 353
 354 static bool block_job_should_pause(BlockJob *job)
 355 {
 356     return job->pause_count > 0;
 357 }
 358
 359 void coroutine_fn block_job_pause_point(BlockJob *job)
 360 {
 361     if (!block_job_should_pause(job)) {
 362         return;
 363     }
 364     if (block_job_is_cancelled(job)) {
 365         return;
 366     }
 367
 368     if (job->driver->pause) {
 369         job->driver->pause(job);
 370     }
 371
 372     if (block_job_should_pause(job) && !block_job_is_cancelled(job)) {
 373         job->paused = true;
 374         job->busy = false;
 375         qemu_coroutine_yield(); /* wait for block_job_resume() */
 376         job->busy = true;
 377         job->paused = false;
 378     }
 379
 380     if (job->driver->resume) {
 381         job->driver->resume(job);
 382     }
 383 }
 384
 385 void block_job_resume(BlockJob *job)
 386 {
 387     assert(job->pause_count > 0);
 388     job->pause_count--;
 389     if (job->pause_count) {
 390         return;
 391     }
 392     block_job_enter(job);
 393 }
 394
 395 void block_job_enter(BlockJob *job)
 396 {
 397     if (job->co && !job->busy) {
 398         qemu_coroutine_enter(job->co);
 399     }
 400 }
 401
 402 void block_job_cancel(BlockJob *job)
 403 {
 404     job->cancelled = true;
 405     block_job_iostatus_reset(job);
 406     block_job_enter(job);
 407 }
 408
 409 bool block_job_is_cancelled(BlockJob *job)
 410 {
 411     return job->cancelled;
 412 }
 413
 414 void block_job_iostatus_reset(BlockJob *job)
 415 {
 416     job->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
 417     if (job->driver->iostatus_reset) {
 418         job->driver->iostatus_reset(job);
 419     }
 420 }
 421
 422 static int block_job_finish_sync(BlockJob *job,
 423                                  void (*finish)(BlockJob *, Error **errp),
 424                                  Error **errp)
 425 {
 426     Error *local_err = NULL;
 427     int ret;
 428
 429     assert(blk_bs(job->blk)->job == job);
 430
 431     block_job_ref(job);
 432
 433     finish(job, &local_err);
 434     if (local_err) {
 435         error_propagate(errp, local_err);
 436         block_job_unref(job);
 437         return -EBUSY;
 438     }
 439     /* block_job_drain calls block_job_enter, and it should be enough to
 440      * induce progress until the job completes or moves to the main thread.
 441     */
 442     while (!job->deferred_to_main_loop && !job->completed) {
 443         block_job_drain(job);
 444     }
 445     while (!job->completed) {
 446         aio_poll(qemu_get_aio_context(), true);
 447     }
 448     ret = (job->cancelled && job->ret == 0) ? -ECANCELED : job->ret;
 449     block_job_unref(job);
 450     return ret;
 451 }
 452
 453 /* A wrapper around block_job_cancel() taking an Error ** parameter so it may be
 454  * used with block_job_finish_sync() without the need for (rather nasty)
 455  * function pointer casts there. */
 456 static void block_job_cancel_err(BlockJob *job, Error **errp)
 457 {
 458     block_job_cancel(job);
 459 }
 460
 461 int block_job_cancel_sync(BlockJob *job)
 462 {
 463     return block_job_finish_sync(job, &block_job_cancel_err, NULL);
 464 }
 465
 466 void block_job_cancel_sync_all(void)
 467 {
 468     BlockJob *job;
 469     AioContext *aio_context;
 470
 471     while ((job = QLIST_FIRST(&block_jobs))) {
 472         aio_context = blk_get_aio_context(job->blk);
 473         aio_context_acquire(aio_context);
 474         block_job_cancel_sync(job);
 475         aio_context_release(aio_context);
 476     }
 477 }
 478
 479 int block_job_complete_sync(BlockJob *job, Error **errp)
 480 {
 481     return block_job_finish_sync(job, &block_job_complete, errp);
 482 }
 483
 484 void block_job_sleep_ns(BlockJob *job, QEMUClockType type, int64_t ns)
 485 {
 486     assert(job->busy);
 487
 488     /* Check cancellation *before* setting busy = false, too!  */
 489     if (block_job_is_cancelled(job)) {
 490         return;
 491     }
 492
 493     job->busy = false;
 494     if (!block_job_should_pause(job)) {
 495         co_aio_sleep_ns(blk_get_aio_context(job->blk), type, ns);
 496     }
 497     job->busy = true;
 498
 499     block_job_pause_point(job);
 500 }
 501
 502 void block_job_yield(BlockJob *job)
 503 {
 504     assert(job->busy);
 505
 506     /* Check cancellation *before* setting busy = false, too!  */
 507     if (block_job_is_cancelled(job)) {
 508         return;
 509     }
 510
 511     job->busy = false;
 512     if (!block_job_should_pause(job)) {
 513         qemu_coroutine_yield();
 514     }
 515     job->busy = true;
 516
 517     block_job_pause_point(job);
 518 }
 519
 520 BlockJobInfo *block_job_query(BlockJob *job, Error **errp)
 521 {
 522     BlockJobInfo *info;
 523
 524     if (block_job_is_internal(job)) {
 525         error_setg(errp, "Cannot query QEMU internal jobs");
 526         return NULL;
 527     }
 528     info = g_new0(BlockJobInfo, 1);
 529     info->type      = g_strdup(BlockJobType_lookup[job->driver->job_type]);
 530     info->device    = g_strdup(job->id);
 531     info->len       = job->len;
 532     info->busy      = job->busy;
 533     info->paused    = job->pause_count > 0;
 534     info->offset    = job->offset;
 535     info->speed     = job->speed;
 536     info->io_status = job->iostatus;
 537     info->ready     = job->ready;
 538     return info;
 539 }
 540
 541 static void block_job_iostatus_set_err(BlockJob *job, int error)
 542 {
 543     if (job->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
 544         job->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE :
 545                                           BLOCK_DEVICE_IO_STATUS_FAILED;
 546     }
 547 }
 548
 549 void block_job_event_cancelled(BlockJob *job)
 550 {
 551     if (block_job_is_internal(job)) {
 552         return;
 553     }
 554
 555     qapi_event_send_block_job_cancelled(job->driver->job_type,
 556                                         job->id,
 557                                         job->len,
 558                                         job->offset,
 559                                         job->speed,
 560                                         &error_abort);
 561 }
 562
 563 void block_job_event_completed(BlockJob *job, const char *msg)
 564 {
 565     if (block_job_is_internal(job)) {
 566         return;
 567     }
 568
 569     qapi_event_send_block_job_completed(job->driver->job_type,
 570                                         job->id,
 571                                         job->len,
 572                                         job->offset,
 573                                         job->speed,
 574                                         !!msg,
 575                                         msg,
 576                                         &error_abort);
 577 }
 578
 579 void block_job_event_ready(BlockJob *job)
 580 {
 581     job->ready = true;
 582
 583     if (block_job_is_internal(job)) {
 584         return;
 585     }
 586
 587     qapi_event_send_block_job_ready(job->driver->job_type,
 588                                     job->id,
 589                                     job->len,
 590                                     job->offset,
 591                                     job->speed, &error_abort);
 592 }
 593
 594 BlockErrorAction block_job_error_action(BlockJob *job, BlockdevOnError on_err,
 595                                         int is_read, int error)
 596 {
 597     BlockErrorAction action;
 598
 599     switch (on_err) {
 600     case BLOCKDEV_ON_ERROR_ENOSPC:
 601     case BLOCKDEV_ON_ERROR_AUTO:
 602         action = (error == ENOSPC) ?
 603                  BLOCK_ERROR_ACTION_STOP : BLOCK_ERROR_ACTION_REPORT;
 604         break;
 605     case BLOCKDEV_ON_ERROR_STOP:
 606         action = BLOCK_ERROR_ACTION_STOP;
 607         break;
 608     case BLOCKDEV_ON_ERROR_REPORT:
 609         action = BLOCK_ERROR_ACTION_REPORT;
 610         break;
 611     case BLOCKDEV_ON_ERROR_IGNORE:
 612         action = BLOCK_ERROR_ACTION_IGNORE;
 613         break;
 614     default:
 615         abort();
 616     }
 617     if (!block_job_is_internal(job)) {
 618         qapi_event_send_block_job_error(job->id,
 619                                         is_read ? IO_OPERATION_TYPE_READ :
 620                                         IO_OPERATION_TYPE_WRITE,
 621                                         action, &error_abort);
 622     }
 623     if (action == BLOCK_ERROR_ACTION_STOP) {
 624         /* make the pause user visible, which will be resumed from QMP. */
 625         job->user_paused = true;
 626         block_job_pause(job);
 627         block_job_iostatus_set_err(job, error);
 628     }
 629     return action;
 630 }
 631
 632 typedef struct {
 633     BlockJob *job;
 634     AioContext *aio_context;
 635     BlockJobDeferToMainLoopFn *fn;
 636     void *opaque;
 637 } BlockJobDeferToMainLoopData;
 638
 639 static void block_job_defer_to_main_loop_bh(void *opaque)
 640 {
 641     BlockJobDeferToMainLoopData *data = opaque;
 642     AioContext *aio_context;
 643
 644     /* Prevent race with block_job_defer_to_main_loop() */
 645     aio_context_acquire(data->aio_context);
 646
 647     /* Fetch BDS AioContext again, in case it has changed */
 648     aio_context = blk_get_aio_context(data->job->blk);
 649     aio_context_acquire(aio_context);
 650
 651     data->job->deferred_to_main_loop = false;
 652     data->fn(data->job, data->opaque);
 653
 654     aio_context_release(aio_context);
 655
 656     aio_context_release(data->aio_context);
 657
 658     g_free(data);
 659 }
 660
 661 void block_job_defer_to_main_loop(BlockJob *job,
 662                                   BlockJobDeferToMainLoopFn *fn,
 663                                   void *opaque)
 664 {
 665     BlockJobDeferToMainLoopData *data = g_malloc(sizeof(*data));
 666     data->job = job;
 667     data->aio_context = blk_get_aio_context(job->blk);
 668     data->fn = fn;
 669     data->opaque = opaque;
 670     job->deferred_to_main_loop = true;
 671
 672     aio_bh_schedule_oneshot(qemu_get_aio_context(),
 673                             block_job_defer_to_main_loop_bh, data);
 674 }
 675
 676 BlockJobTxn *block_job_txn_new(void)
 677 {
 678     BlockJobTxn *txn = g_new0(BlockJobTxn, 1);
 679     QLIST_INIT(&txn->jobs);
 680     txn->refcnt = 1;
 681     return txn;
 682 }
 683
 684 static void block_job_txn_ref(BlockJobTxn *txn)
 685 {
 686     txn->refcnt++;
 687 }
 688
 689 void block_job_txn_unref(BlockJobTxn *txn)
 690 {
 691     if (txn && --txn->refcnt == 0) {
 692         g_free(txn);
 693     }
 694 }
 695
 696 void block_job_txn_add_job(BlockJobTxn *txn, BlockJob *job)
 697 {
 698     if (!txn) {
 699         return;
 700     }
 701
 702     assert(!job->txn);
 703     job->txn = txn;
 704
 705     QLIST_INSERT_HEAD(&txn->jobs, job, txn_list);
 706     block_job_txn_ref(txn);
 707 }