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, int flags,
 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 && !(flags & BLOCK_JOB_INTERNAL)) {
 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 (job_id) {
 145         if (flags & BLOCK_JOB_INTERNAL) {
 146             error_setg(errp, "Cannot specify job ID for internal block job");
 147             return NULL;
 148         }
 149
 150         if (!id_wellformed(job_id)) {
 151             error_setg(errp, "Invalid job ID '%s'", job_id);
 152             return NULL;
 153         }
 154
 155         if (block_job_get(job_id)) {
 156             error_setg(errp, "Job ID '%s' already in use", job_id);
 157             return NULL;
 158         }
 159     }
 160
 161     blk = blk_new();
 162     blk_insert_bs(blk, bs);
 163
 164     job = g_malloc0(driver->instance_size);
 165     error_setg(&job->blocker, "block device is in use by block job: %s",
 166                BlockJobType_lookup[driver->job_type]);
 167     block_job_add_bdrv(job, bs);
 168     bdrv_op_unblock(bs, BLOCK_OP_TYPE_DATAPLANE, job->blocker);
 169
 170     job->driver        = driver;
 171     job->id            = g_strdup(job_id);
 172     job->blk           = blk;
 173     job->cb            = cb;
 174     job->opaque        = opaque;
 175     job->busy          = true;
 176     job->refcnt        = 1;
 177     bs->job = job;
 178
 179     QLIST_INSERT_HEAD(&block_jobs, job, job_list);
 180
 181     blk_add_aio_context_notifier(blk, block_job_attached_aio_context,
 182                                  block_job_detach_aio_context, job);
 183
 184     /* Only set speed when necessary to avoid NotSupported error */
 185     if (speed != 0) {
 186         Error *local_err = NULL;
 187
 188         block_job_set_speed(job, speed, &local_err);
 189         if (local_err) {
 190             block_job_unref(job);
 191             error_propagate(errp, local_err);
 192             return NULL;
 193         }
 194     }
 195     return job;
 196 }
 197
 198 bool block_job_is_internal(BlockJob *job)
 199 {
 200     return (job->id == NULL);
 201 }
 202
 203 void block_job_ref(BlockJob *job)
 204 {
 205     ++job->refcnt;
 206 }
 207
 208 void block_job_unref(BlockJob *job)
 209 {
 210     if (--job->refcnt == 0) {
 211         GSList *l;
 212         BlockDriverState *bs = blk_bs(job->blk);
 213         bs->job = NULL;
 214         for (l = job->nodes; l; l = l->next) {
 215             bs = l->data;
 216             bdrv_op_unblock_all(bs, job->blocker);
 217             bdrv_unref(bs);
 218         }
 219         g_slist_free(job->nodes);
 220         blk_remove_aio_context_notifier(job->blk,
 221                                         block_job_attached_aio_context,
 222                                         block_job_detach_aio_context, job);
 223         blk_unref(job->blk);
 224         error_free(job->blocker);
 225         g_free(job->id);
 226         QLIST_REMOVE(job, job_list);
 227         g_free(job);
 228     }
 229 }
 230
 231 static void block_job_completed_single(BlockJob *job)
 232 {
 233     if (!job->ret) {
 234         if (job->driver->commit) {
 235             job->driver->commit(job);
 236         }
 237     } else {
 238         if (job->driver->abort) {
 239             job->driver->abort(job);
 240         }
 241     }
 242     job->cb(job->opaque, job->ret);
 243     if (job->txn) {
 244         block_job_txn_unref(job->txn);
 245     }
 246     block_job_unref(job);
 247 }
 248
 249 static void block_job_completed_txn_abort(BlockJob *job)
 250 {
 251     AioContext *ctx;
 252     BlockJobTxn *txn = job->txn;
 253     BlockJob *other_job, *next;
 254
 255     if (txn->aborting) {
 256         /*
 257          * We are cancelled by another job, which will handle everything.
 258          */
 259         return;
 260     }
 261     txn->aborting = true;
 262     /* We are the first failed job. Cancel other jobs. */
 263     QLIST_FOREACH(other_job, &txn->jobs, txn_list) {
 264         ctx = blk_get_aio_context(other_job->blk);
 265         aio_context_acquire(ctx);
 266     }
 267     QLIST_FOREACH(other_job, &txn->jobs, txn_list) {
 268         if (other_job == job || other_job->completed) {
 269             /* Other jobs are "effectively" cancelled by us, set the status for
 270              * them; this job, however, may or may not be cancelled, depending
 271              * on the caller, so leave it. */
 272             if (other_job != job) {
 273                 other_job->cancelled = true;
 274             }
 275             continue;
 276         }
 277         block_job_cancel_sync(other_job);
 278         assert(other_job->completed);
 279     }
 280     QLIST_FOREACH_SAFE(other_job, &txn->jobs, txn_list, next) {
 281         ctx = blk_get_aio_context(other_job->blk);
 282         block_job_completed_single(other_job);
 283         aio_context_release(ctx);
 284     }
 285 }
 286
 287 static void block_job_completed_txn_success(BlockJob *job)
 288 {
 289     AioContext *ctx;
 290     BlockJobTxn *txn = job->txn;
 291     BlockJob *other_job, *next;
 292     /*
 293      * Successful completion, see if there are other running jobs in this
 294      * txn.
 295      */
 296     QLIST_FOREACH(other_job, &txn->jobs, txn_list) {
 297         if (!other_job->completed) {
 298             return;
 299         }
 300     }
 301     /* We are the last completed job, commit the transaction. */
 302     QLIST_FOREACH_SAFE(other_job, &txn->jobs, txn_list, next) {
 303         ctx = blk_get_aio_context(other_job->blk);
 304         aio_context_acquire(ctx);
 305         assert(other_job->ret == 0);
 306         block_job_completed_single(other_job);
 307         aio_context_release(ctx);
 308     }
 309 }
 310
 311 void block_job_completed(BlockJob *job, int ret)
 312 {
 313     assert(blk_bs(job->blk)->job == job);
 314     assert(!job->completed);
 315     job->completed = true;
 316     job->ret = ret;
 317     if (!job->txn) {
 318         block_job_completed_single(job);
 319     } else if (ret < 0 || block_job_is_cancelled(job)) {
 320         block_job_completed_txn_abort(job);
 321     } else {
 322         block_job_completed_txn_success(job);
 323     }
 324 }
 325
 326 void block_job_set_speed(BlockJob *job, int64_t speed, Error **errp)
 327 {
 328     Error *local_err = NULL;
 329
 330     if (!job->driver->set_speed) {
 331         error_setg(errp, QERR_UNSUPPORTED);
 332         return;
 333     }
 334     job->driver->set_speed(job, speed, &local_err);
 335     if (local_err) {
 336         error_propagate(errp, local_err);
 337         return;
 338     }
 339
 340     job->speed = speed;
 341 }
 342
 343 void block_job_complete(BlockJob *job, Error **errp)
 344 {
 345     /* Should not be reachable via external interface for internal jobs */
 346     assert(job->id);
 347     if (job->pause_count || job->cancelled || !job->driver->complete) {
 348         error_setg(errp, "The active block job '%s' cannot be completed",
 349                    job->id);
 350         return;
 351     }
 352
 353     job->driver->complete(job, errp);
 354 }
 355
 356 void block_job_pause(BlockJob *job)
 357 {
 358     job->pause_count++;
 359 }
 360
 361 static bool block_job_should_pause(BlockJob *job)
 362 {
 363     return job->pause_count > 0;
 364 }
 365
 366 void coroutine_fn block_job_pause_point(BlockJob *job)
 367 {
 368     if (!block_job_should_pause(job)) {
 369         return;
 370     }
 371     if (block_job_is_cancelled(job)) {
 372         return;
 373     }
 374
 375     if (job->driver->pause) {
 376         job->driver->pause(job);
 377     }
 378
 379     if (block_job_should_pause(job) && !block_job_is_cancelled(job)) {
 380         job->paused = true;
 381         job->busy = false;
 382         qemu_coroutine_yield(); /* wait for block_job_resume() */
 383         job->busy = true;
 384         job->paused = false;
 385     }
 386
 387     if (job->driver->resume) {
 388         job->driver->resume(job);
 389     }
 390 }
 391
 392 void block_job_resume(BlockJob *job)
 393 {
 394     assert(job->pause_count > 0);
 395     job->pause_count--;
 396     if (job->pause_count) {
 397         return;
 398     }
 399     block_job_enter(job);
 400 }
 401
 402 void block_job_enter(BlockJob *job)
 403 {
 404     if (job->co && !job->busy) {
 405         qemu_coroutine_enter(job->co);
 406     }
 407 }
 408
 409 void block_job_cancel(BlockJob *job)
 410 {
 411     job->cancelled = true;
 412     block_job_iostatus_reset(job);
 413     block_job_enter(job);
 414 }
 415
 416 bool block_job_is_cancelled(BlockJob *job)
 417 {
 418     return job->cancelled;
 419 }
 420
 421 void block_job_iostatus_reset(BlockJob *job)
 422 {
 423     job->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
 424     if (job->driver->iostatus_reset) {
 425         job->driver->iostatus_reset(job);
 426     }
 427 }
 428
 429 static int block_job_finish_sync(BlockJob *job,
 430                                  void (*finish)(BlockJob *, Error **errp),
 431                                  Error **errp)
 432 {
 433     Error *local_err = NULL;
 434     int ret;
 435
 436     assert(blk_bs(job->blk)->job == job);
 437
 438     block_job_ref(job);
 439
 440     finish(job, &local_err);
 441     if (local_err) {
 442         error_propagate(errp, local_err);
 443         block_job_unref(job);
 444         return -EBUSY;
 445     }
 446     /* block_job_drain calls block_job_enter, and it should be enough to
 447      * induce progress until the job completes or moves to the main thread.
 448     */
 449     while (!job->deferred_to_main_loop && !job->completed) {
 450         block_job_drain(job);
 451     }
 452     while (!job->completed) {
 453         aio_poll(qemu_get_aio_context(), true);
 454     }
 455     ret = (job->cancelled && job->ret == 0) ? -ECANCELED : job->ret;
 456     block_job_unref(job);
 457     return ret;
 458 }
 459
 460 /* A wrapper around block_job_cancel() taking an Error ** parameter so it may be
 461  * used with block_job_finish_sync() without the need for (rather nasty)
 462  * function pointer casts there. */
 463 static void block_job_cancel_err(BlockJob *job, Error **errp)
 464 {
 465     block_job_cancel(job);
 466 }
 467
 468 int block_job_cancel_sync(BlockJob *job)
 469 {
 470     return block_job_finish_sync(job, &block_job_cancel_err, NULL);
 471 }
 472
 473 void block_job_cancel_sync_all(void)
 474 {
 475     BlockJob *job;
 476     AioContext *aio_context;
 477
 478     while ((job = QLIST_FIRST(&block_jobs))) {
 479         aio_context = blk_get_aio_context(job->blk);
 480         aio_context_acquire(aio_context);
 481         block_job_cancel_sync(job);
 482         aio_context_release(aio_context);
 483     }
 484 }
 485
 486 int block_job_complete_sync(BlockJob *job, Error **errp)
 487 {
 488     return block_job_finish_sync(job, &block_job_complete, errp);
 489 }
 490
 491 void block_job_sleep_ns(BlockJob *job, QEMUClockType type, int64_t ns)
 492 {
 493     assert(job->busy);
 494
 495     /* Check cancellation *before* setting busy = false, too!  */
 496     if (block_job_is_cancelled(job)) {
 497         return;
 498     }
 499
 500     job->busy = false;
 501     if (!block_job_should_pause(job)) {
 502         co_aio_sleep_ns(blk_get_aio_context(job->blk), type, ns);
 503     }
 504     job->busy = true;
 505
 506     block_job_pause_point(job);
 507 }
 508
 509 void block_job_yield(BlockJob *job)
 510 {
 511     assert(job->busy);
 512
 513     /* Check cancellation *before* setting busy = false, too!  */
 514     if (block_job_is_cancelled(job)) {
 515         return;
 516     }
 517
 518     job->busy = false;
 519     if (!block_job_should_pause(job)) {
 520         qemu_coroutine_yield();
 521     }
 522     job->busy = true;
 523
 524     block_job_pause_point(job);
 525 }
 526
 527 BlockJobInfo *block_job_query(BlockJob *job, Error **errp)
 528 {
 529     BlockJobInfo *info;
 530
 531     if (block_job_is_internal(job)) {
 532         error_setg(errp, "Cannot query QEMU internal jobs");
 533         return NULL;
 534     }
 535     info = g_new0(BlockJobInfo, 1);
 536     info->type      = g_strdup(BlockJobType_lookup[job->driver->job_type]);
 537     info->device    = g_strdup(job->id);
 538     info->len       = job->len;
 539     info->busy      = job->busy;
 540     info->paused    = job->pause_count > 0;
 541     info->offset    = job->offset;
 542     info->speed     = job->speed;
 543     info->io_status = job->iostatus;
 544     info->ready     = job->ready;
 545     return info;
 546 }
 547
 548 static void block_job_iostatus_set_err(BlockJob *job, int error)
 549 {
 550     if (job->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
 551         job->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE :
 552                                           BLOCK_DEVICE_IO_STATUS_FAILED;
 553     }
 554 }
 555
 556 void block_job_event_cancelled(BlockJob *job)
 557 {
 558     if (block_job_is_internal(job)) {
 559         return;
 560     }
 561
 562     qapi_event_send_block_job_cancelled(job->driver->job_type,
 563                                         job->id,
 564                                         job->len,
 565                                         job->offset,
 566                                         job->speed,
 567                                         &error_abort);
 568 }
 569
 570 void block_job_event_completed(BlockJob *job, const char *msg)
 571 {
 572     if (block_job_is_internal(job)) {
 573         return;
 574     }
 575
 576     qapi_event_send_block_job_completed(job->driver->job_type,
 577                                         job->id,
 578                                         job->len,
 579                                         job->offset,
 580                                         job->speed,
 581                                         !!msg,
 582                                         msg,
 583                                         &error_abort);
 584 }
 585
 586 void block_job_event_ready(BlockJob *job)
 587 {
 588     job->ready = true;
 589
 590     if (block_job_is_internal(job)) {
 591         return;
 592     }
 593
 594     qapi_event_send_block_job_ready(job->driver->job_type,
 595                                     job->id,
 596                                     job->len,
 597                                     job->offset,
 598                                     job->speed, &error_abort);
 599 }
 600
 601 BlockErrorAction block_job_error_action(BlockJob *job, BlockdevOnError on_err,
 602                                         int is_read, int error)
 603 {
 604     BlockErrorAction action;
 605
 606     switch (on_err) {
 607     case BLOCKDEV_ON_ERROR_ENOSPC:
 608     case BLOCKDEV_ON_ERROR_AUTO:
 609         action = (error == ENOSPC) ?
 610                  BLOCK_ERROR_ACTION_STOP : BLOCK_ERROR_ACTION_REPORT;
 611         break;
 612     case BLOCKDEV_ON_ERROR_STOP:
 613         action = BLOCK_ERROR_ACTION_STOP;
 614         break;
 615     case BLOCKDEV_ON_ERROR_REPORT:
 616         action = BLOCK_ERROR_ACTION_REPORT;
 617         break;
 618     case BLOCKDEV_ON_ERROR_IGNORE:
 619         action = BLOCK_ERROR_ACTION_IGNORE;
 620         break;
 621     default:
 622         abort();
 623     }
 624     if (!block_job_is_internal(job)) {
 625         qapi_event_send_block_job_error(job->id,
 626                                         is_read ? IO_OPERATION_TYPE_READ :
 627                                         IO_OPERATION_TYPE_WRITE,
 628                                         action, &error_abort);
 629     }
 630     if (action == BLOCK_ERROR_ACTION_STOP) {
 631         /* make the pause user visible, which will be resumed from QMP. */
 632         job->user_paused = true;
 633         block_job_pause(job);
 634         block_job_iostatus_set_err(job, error);
 635     }
 636     return action;
 637 }
 638
 639 typedef struct {
 640     BlockJob *job;
 641     AioContext *aio_context;
 642     BlockJobDeferToMainLoopFn *fn;
 643     void *opaque;
 644 } BlockJobDeferToMainLoopData;
 645
 646 static void block_job_defer_to_main_loop_bh(void *opaque)
 647 {
 648     BlockJobDeferToMainLoopData *data = opaque;
 649     AioContext *aio_context;
 650
 651     /* Prevent race with block_job_defer_to_main_loop() */
 652     aio_context_acquire(data->aio_context);
 653
 654     /* Fetch BDS AioContext again, in case it has changed */
 655     aio_context = blk_get_aio_context(data->job->blk);
 656     aio_context_acquire(aio_context);
 657
 658     data->job->deferred_to_main_loop = false;
 659     data->fn(data->job, data->opaque);
 660
 661     aio_context_release(aio_context);
 662
 663     aio_context_release(data->aio_context);
 664
 665     g_free(data);
 666 }
 667
 668 void block_job_defer_to_main_loop(BlockJob *job,
 669                                   BlockJobDeferToMainLoopFn *fn,
 670                                   void *opaque)
 671 {
 672     BlockJobDeferToMainLoopData *data = g_malloc(sizeof(*data));
 673     data->job = job;
 674     data->aio_context = blk_get_aio_context(job->blk);
 675     data->fn = fn;
 676     data->opaque = opaque;
 677     job->deferred_to_main_loop = true;
 678
 679     aio_bh_schedule_oneshot(qemu_get_aio_context(),
 680                             block_job_defer_to_main_loop_bh, data);
 681 }
 682
 683 BlockJobTxn *block_job_txn_new(void)
 684 {
 685     BlockJobTxn *txn = g_new0(BlockJobTxn, 1);
 686     QLIST_INIT(&txn->jobs);
 687     txn->refcnt = 1;
 688     return txn;
 689 }
 690
 691 static void block_job_txn_ref(BlockJobTxn *txn)
 692 {
 693     txn->refcnt++;
 694 }
 695
 696 void block_job_txn_unref(BlockJobTxn *txn)
 697 {
 698     if (txn && --txn->refcnt == 0) {
 699         g_free(txn);
 700     }
 701 }
 702
 703 void block_job_txn_add_job(BlockJobTxn *txn, BlockJob *job)
 704 {
 705     if (!txn) {
 706         return;
 707     }
 708
 709     assert(!job->txn);
 710     job->txn = txn;
 711
 712     QLIST_INSERT_HEAD(&txn->jobs, job, txn_list);
 713     block_job_txn_ref(txn);
 714 }