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_int.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 static void block_job_event_cancelled(BlockJob *job);
  42 static void block_job_event_completed(BlockJob *job, const char *msg);
  43
  44 /* Transactional group of block jobs */
  45 struct BlockJobTxn {
  46
  47     /* Is this txn being cancelled? */
  48     bool aborting;
  49
  50     /* List of jobs */
  51     QLIST_HEAD(, BlockJob) jobs;
  52
  53     /* Reference count */
  54     int refcnt;
  55 };
  56
  57 static QLIST_HEAD(, BlockJob) block_jobs = QLIST_HEAD_INITIALIZER(block_jobs);
  58
  59 BlockJob *block_job_next(BlockJob *job)
  60 {
  61     if (!job) {
  62         return QLIST_FIRST(&block_jobs);
  63     }
  64     return QLIST_NEXT(job, job_list);
  65 }
  66
  67 BlockJob *block_job_get(const char *id)
  68 {
  69     BlockJob *job;
  70
  71     QLIST_FOREACH(job, &block_jobs, job_list) {
  72         if (job->id && !strcmp(id, job->id)) {
  73             return job;
  74         }
  75     }
  76
  77     return NULL;
  78 }
  79
  80 static void block_job_attached_aio_context(AioContext *new_context,
  81                                            void *opaque)
  82 {
  83     BlockJob *job = opaque;
  84
  85     if (job->driver->attached_aio_context) {
  86         job->driver->attached_aio_context(job, new_context);
  87     }
  88
  89     block_job_resume(job);
  90 }
  91
  92 static void block_job_drain(BlockJob *job)
  93 {
  94     /* If job is !job->busy this kicks it into the next pause point. */
  95     block_job_enter(job);
  96
  97     blk_drain(job->blk);
  98     if (job->driver->drain) {
  99         job->driver->drain(job);
 100     }
 101 }
 102
 103 static void block_job_detach_aio_context(void *opaque)
 104 {
 105     BlockJob *job = opaque;
 106
 107     /* In case the job terminates during aio_poll()... */
 108     block_job_ref(job);
 109
 110     block_job_pause(job);
 111
 112     while (!job->paused && !job->completed) {
 113         block_job_drain(job);
 114     }
 115
 116     block_job_unref(job);
 117 }
 118
 119 void block_job_add_bdrv(BlockJob *job, BlockDriverState *bs)
 120 {
 121     job->nodes = g_slist_prepend(job->nodes, bs);
 122     bdrv_ref(bs);
 123     bdrv_op_block_all(bs, job->blocker);
 124 }
 125
 126 void *block_job_create(const char *job_id, const BlockJobDriver *driver,
 127                        BlockDriverState *bs, int64_t speed, int flags,
 128                        BlockCompletionFunc *cb, void *opaque, Error **errp)
 129 {
 130     BlockBackend *blk;
 131     BlockJob *job;
 132
 133     if (bs->job) {
 134         error_setg(errp, QERR_DEVICE_IN_USE, bdrv_get_device_name(bs));
 135         return NULL;
 136     }
 137
 138     if (job_id == NULL && !(flags & BLOCK_JOB_INTERNAL)) {
 139         job_id = bdrv_get_device_name(bs);
 140         if (!*job_id) {
 141             error_setg(errp, "An explicit job ID is required for this node");
 142             return NULL;
 143         }
 144     }
 145
 146     if (job_id) {
 147         if (flags & BLOCK_JOB_INTERNAL) {
 148             error_setg(errp, "Cannot specify job ID for internal block job");
 149             return NULL;
 150         }
 151
 152         if (!id_wellformed(job_id)) {
 153             error_setg(errp, "Invalid job ID '%s'", job_id);
 154             return NULL;
 155         }
 156
 157         if (block_job_get(job_id)) {
 158             error_setg(errp, "Job ID '%s' already in use", job_id);
 159             return NULL;
 160         }
 161     }
 162
 163     blk = blk_new();
 164     blk_insert_bs(blk, bs);
 165
 166     job = g_malloc0(driver->instance_size);
 167     error_setg(&job->blocker, "block device is in use by block job: %s",
 168                BlockJobType_lookup[driver->job_type]);
 169     block_job_add_bdrv(job, bs);
 170     bdrv_op_unblock(bs, BLOCK_OP_TYPE_DATAPLANE, job->blocker);
 171
 172     job->driver        = driver;
 173     job->id            = g_strdup(job_id);
 174     job->blk           = blk;
 175     job->cb            = cb;
 176     job->opaque        = opaque;
 177     job->busy          = true;
 178     job->refcnt        = 1;
 179     bs->job = job;
 180
 181     QLIST_INSERT_HEAD(&block_jobs, job, job_list);
 182
 183     blk_add_aio_context_notifier(blk, block_job_attached_aio_context,
 184                                  block_job_detach_aio_context, job);
 185
 186     /* Only set speed when necessary to avoid NotSupported error */
 187     if (speed != 0) {
 188         Error *local_err = NULL;
 189
 190         block_job_set_speed(job, speed, &local_err);
 191         if (local_err) {
 192             block_job_unref(job);
 193             error_propagate(errp, local_err);
 194             return NULL;
 195         }
 196     }
 197     return job;
 198 }
 199
 200 bool block_job_is_internal(BlockJob *job)
 201 {
 202     return (job->id == NULL);
 203 }
 204
 205 void block_job_ref(BlockJob *job)
 206 {
 207     ++job->refcnt;
 208 }
 209
 210 void block_job_unref(BlockJob *job)
 211 {
 212     if (--job->refcnt == 0) {
 213         GSList *l;
 214         BlockDriverState *bs = blk_bs(job->blk);
 215         bs->job = NULL;
 216         for (l = job->nodes; l; l = l->next) {
 217             bs = l->data;
 218             bdrv_op_unblock_all(bs, job->blocker);
 219             bdrv_unref(bs);
 220         }
 221         g_slist_free(job->nodes);
 222         blk_remove_aio_context_notifier(job->blk,
 223                                         block_job_attached_aio_context,
 224                                         block_job_detach_aio_context, job);
 225         blk_unref(job->blk);
 226         error_free(job->blocker);
 227         g_free(job->id);
 228         QLIST_REMOVE(job, job_list);
 229         g_free(job);
 230     }
 231 }
 232
 233 static void block_job_completed_single(BlockJob *job)
 234 {
 235     if (!job->ret) {
 236         if (job->driver->commit) {
 237             job->driver->commit(job);
 238         }
 239     } else {
 240         if (job->driver->abort) {
 241             job->driver->abort(job);
 242         }
 243     }
 244
 245     if (job->cb) {
 246         job->cb(job->opaque, job->ret);
 247     }
 248     if (block_job_is_cancelled(job)) {
 249         block_job_event_cancelled(job);
 250     } else {
 251         const char *msg = NULL;
 252         if (job->ret < 0) {
 253             msg = strerror(-job->ret);
 254         }
 255         block_job_event_completed(job, msg);
 256     }
 257
 258     if (job->txn) {
 259         block_job_txn_unref(job->txn);
 260     }
 261     block_job_unref(job);
 262 }
 263
 264 static void block_job_completed_txn_abort(BlockJob *job)
 265 {
 266     AioContext *ctx;
 267     BlockJobTxn *txn = job->txn;
 268     BlockJob *other_job, *next;
 269
 270     if (txn->aborting) {
 271         /*
 272          * We are cancelled by another job, which will handle everything.
 273          */
 274         return;
 275     }
 276     txn->aborting = true;
 277     /* We are the first failed job. Cancel other jobs. */
 278     QLIST_FOREACH(other_job, &txn->jobs, txn_list) {
 279         ctx = blk_get_aio_context(other_job->blk);
 280         aio_context_acquire(ctx);
 281     }
 282     QLIST_FOREACH(other_job, &txn->jobs, txn_list) {
 283         if (other_job == job || other_job->completed) {
 284             /* Other jobs are "effectively" cancelled by us, set the status for
 285              * them; this job, however, may or may not be cancelled, depending
 286              * on the caller, so leave it. */
 287             if (other_job != job) {
 288                 other_job->cancelled = true;
 289             }
 290             continue;
 291         }
 292         block_job_cancel_sync(other_job);
 293         assert(other_job->completed);
 294     }
 295     QLIST_FOREACH_SAFE(other_job, &txn->jobs, txn_list, next) {
 296         ctx = blk_get_aio_context(other_job->blk);
 297         block_job_completed_single(other_job);
 298         aio_context_release(ctx);
 299     }
 300 }
 301
 302 static void block_job_completed_txn_success(BlockJob *job)
 303 {
 304     AioContext *ctx;
 305     BlockJobTxn *txn = job->txn;
 306     BlockJob *other_job, *next;
 307     /*
 308      * Successful completion, see if there are other running jobs in this
 309      * txn.
 310      */
 311     QLIST_FOREACH(other_job, &txn->jobs, txn_list) {
 312         if (!other_job->completed) {
 313             return;
 314         }
 315     }
 316     /* We are the last completed job, commit the transaction. */
 317     QLIST_FOREACH_SAFE(other_job, &txn->jobs, txn_list, next) {
 318         ctx = blk_get_aio_context(other_job->blk);
 319         aio_context_acquire(ctx);
 320         assert(other_job->ret == 0);
 321         block_job_completed_single(other_job);
 322         aio_context_release(ctx);
 323     }
 324 }
 325
 326 void block_job_completed(BlockJob *job, int ret)
 327 {
 328     assert(blk_bs(job->blk)->job == job);
 329     assert(!job->completed);
 330     job->completed = true;
 331     job->ret = ret;
 332     if (!job->txn) {
 333         block_job_completed_single(job);
 334     } else if (ret < 0 || block_job_is_cancelled(job)) {
 335         block_job_completed_txn_abort(job);
 336     } else {
 337         block_job_completed_txn_success(job);
 338     }
 339 }
 340
 341 void block_job_set_speed(BlockJob *job, int64_t speed, Error **errp)
 342 {
 343     Error *local_err = NULL;
 344
 345     if (!job->driver->set_speed) {
 346         error_setg(errp, QERR_UNSUPPORTED);
 347         return;
 348     }
 349     job->driver->set_speed(job, speed, &local_err);
 350     if (local_err) {
 351         error_propagate(errp, local_err);
 352         return;
 353     }
 354
 355     job->speed = speed;
 356 }
 357
 358 void block_job_complete(BlockJob *job, Error **errp)
 359 {
 360     /* Should not be reachable via external interface for internal jobs */
 361     assert(job->id);
 362     if (job->pause_count || job->cancelled || !job->driver->complete) {
 363         error_setg(errp, "The active block job '%s' cannot be completed",
 364                    job->id);
 365         return;
 366     }
 367
 368     job->driver->complete(job, errp);
 369 }
 370
 371 void block_job_pause(BlockJob *job)
 372 {
 373     job->pause_count++;
 374 }
 375
 376 void block_job_user_pause(BlockJob *job)
 377 {
 378     job->user_paused = true;
 379     block_job_pause(job);
 380 }
 381
 382 static bool block_job_should_pause(BlockJob *job)
 383 {
 384     return job->pause_count > 0;
 385 }
 386
 387 bool block_job_user_paused(BlockJob *job)
 388 {
 389     return job ? job->user_paused : 0;
 390 }
 391
 392 void coroutine_fn block_job_pause_point(BlockJob *job)
 393 {
 394     if (!block_job_should_pause(job)) {
 395         return;
 396     }
 397     if (block_job_is_cancelled(job)) {
 398         return;
 399     }
 400
 401     if (job->driver->pause) {
 402         job->driver->pause(job);
 403     }
 404
 405     if (block_job_should_pause(job) && !block_job_is_cancelled(job)) {
 406         job->paused = true;
 407         job->busy = false;
 408         qemu_coroutine_yield(); /* wait for block_job_resume() */
 409         job->busy = true;
 410         job->paused = false;
 411     }
 412
 413     if (job->driver->resume) {
 414         job->driver->resume(job);
 415     }
 416 }
 417
 418 void block_job_resume(BlockJob *job)
 419 {
 420     assert(job->pause_count > 0);
 421     job->pause_count--;
 422     if (job->pause_count) {
 423         return;
 424     }
 425     block_job_enter(job);
 426 }
 427
 428 void block_job_user_resume(BlockJob *job)
 429 {
 430     if (job && job->user_paused && job->pause_count > 0) {
 431         job->user_paused = false;
 432         block_job_resume(job);
 433     }
 434 }
 435
 436 void block_job_enter(BlockJob *job)
 437 {
 438     if (job->co && !job->busy) {
 439         qemu_coroutine_enter(job->co);
 440     }
 441 }
 442
 443 void block_job_cancel(BlockJob *job)
 444 {
 445     job->cancelled = true;
 446     block_job_iostatus_reset(job);
 447     block_job_enter(job);
 448 }
 449
 450 bool block_job_is_cancelled(BlockJob *job)
 451 {
 452     return job->cancelled;
 453 }
 454
 455 void block_job_iostatus_reset(BlockJob *job)
 456 {
 457     job->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
 458     if (job->driver->iostatus_reset) {
 459         job->driver->iostatus_reset(job);
 460     }
 461 }
 462
 463 static int block_job_finish_sync(BlockJob *job,
 464                                  void (*finish)(BlockJob *, Error **errp),
 465                                  Error **errp)
 466 {
 467     Error *local_err = NULL;
 468     int ret;
 469
 470     assert(blk_bs(job->blk)->job == job);
 471
 472     block_job_ref(job);
 473
 474     finish(job, &local_err);
 475     if (local_err) {
 476         error_propagate(errp, local_err);
 477         block_job_unref(job);
 478         return -EBUSY;
 479     }
 480     /* block_job_drain calls block_job_enter, and it should be enough to
 481      * induce progress until the job completes or moves to the main thread.
 482     */
 483     while (!job->deferred_to_main_loop && !job->completed) {
 484         block_job_drain(job);
 485     }
 486     while (!job->completed) {
 487         aio_poll(qemu_get_aio_context(), true);
 488     }
 489     ret = (job->cancelled && job->ret == 0) ? -ECANCELED : job->ret;
 490     block_job_unref(job);
 491     return ret;
 492 }
 493
 494 /* A wrapper around block_job_cancel() taking an Error ** parameter so it may be
 495  * used with block_job_finish_sync() without the need for (rather nasty)
 496  * function pointer casts there. */
 497 static void block_job_cancel_err(BlockJob *job, Error **errp)
 498 {
 499     block_job_cancel(job);
 500 }
 501
 502 int block_job_cancel_sync(BlockJob *job)
 503 {
 504     return block_job_finish_sync(job, &block_job_cancel_err, NULL);
 505 }
 506
 507 void block_job_cancel_sync_all(void)
 508 {
 509     BlockJob *job;
 510     AioContext *aio_context;
 511
 512     while ((job = QLIST_FIRST(&block_jobs))) {
 513         aio_context = blk_get_aio_context(job->blk);
 514         aio_context_acquire(aio_context);
 515         block_job_cancel_sync(job);
 516         aio_context_release(aio_context);
 517     }
 518 }
 519
 520 int block_job_complete_sync(BlockJob *job, Error **errp)
 521 {
 522     return block_job_finish_sync(job, &block_job_complete, errp);
 523 }
 524
 525 void block_job_sleep_ns(BlockJob *job, QEMUClockType type, int64_t ns)
 526 {
 527     assert(job->busy);
 528
 529     /* Check cancellation *before* setting busy = false, too!  */
 530     if (block_job_is_cancelled(job)) {
 531         return;
 532     }
 533
 534     job->busy = false;
 535     if (!block_job_should_pause(job)) {
 536         co_aio_sleep_ns(blk_get_aio_context(job->blk), type, ns);
 537     }
 538     job->busy = true;
 539
 540     block_job_pause_point(job);
 541 }
 542
 543 void block_job_yield(BlockJob *job)
 544 {
 545     assert(job->busy);
 546
 547     /* Check cancellation *before* setting busy = false, too!  */
 548     if (block_job_is_cancelled(job)) {
 549         return;
 550     }
 551
 552     job->busy = false;
 553     if (!block_job_should_pause(job)) {
 554         qemu_coroutine_yield();
 555     }
 556     job->busy = true;
 557
 558     block_job_pause_point(job);
 559 }
 560
 561 BlockJobInfo *block_job_query(BlockJob *job, Error **errp)
 562 {
 563     BlockJobInfo *info;
 564
 565     if (block_job_is_internal(job)) {
 566         error_setg(errp, "Cannot query QEMU internal jobs");
 567         return NULL;
 568     }
 569     info = g_new0(BlockJobInfo, 1);
 570     info->type      = g_strdup(BlockJobType_lookup[job->driver->job_type]);
 571     info->device    = g_strdup(job->id);
 572     info->len       = job->len;
 573     info->busy      = job->busy;
 574     info->paused    = job->pause_count > 0;
 575     info->offset    = job->offset;
 576     info->speed     = job->speed;
 577     info->io_status = job->iostatus;
 578     info->ready     = job->ready;
 579     return info;
 580 }
 581
 582 static void block_job_iostatus_set_err(BlockJob *job, int error)
 583 {
 584     if (job->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
 585         job->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE :
 586                                           BLOCK_DEVICE_IO_STATUS_FAILED;
 587     }
 588 }
 589
 590 static void block_job_event_cancelled(BlockJob *job)
 591 {
 592     if (block_job_is_internal(job)) {
 593         return;
 594     }
 595
 596     qapi_event_send_block_job_cancelled(job->driver->job_type,
 597                                         job->id,
 598                                         job->len,
 599                                         job->offset,
 600                                         job->speed,
 601                                         &error_abort);
 602 }
 603
 604 static void block_job_event_completed(BlockJob *job, const char *msg)
 605 {
 606     if (block_job_is_internal(job)) {
 607         return;
 608     }
 609
 610     qapi_event_send_block_job_completed(job->driver->job_type,
 611                                         job->id,
 612                                         job->len,
 613                                         job->offset,
 614                                         job->speed,
 615                                         !!msg,
 616                                         msg,
 617                                         &error_abort);
 618 }
 619
 620 void block_job_event_ready(BlockJob *job)
 621 {
 622     job->ready = true;
 623
 624     if (block_job_is_internal(job)) {
 625         return;
 626     }
 627
 628     qapi_event_send_block_job_ready(job->driver->job_type,
 629                                     job->id,
 630                                     job->len,
 631                                     job->offset,
 632                                     job->speed, &error_abort);
 633 }
 634
 635 BlockErrorAction block_job_error_action(BlockJob *job, BlockdevOnError on_err,
 636                                         int is_read, int error)
 637 {
 638     BlockErrorAction action;
 639
 640     switch (on_err) {
 641     case BLOCKDEV_ON_ERROR_ENOSPC:
 642     case BLOCKDEV_ON_ERROR_AUTO:
 643         action = (error == ENOSPC) ?
 644                  BLOCK_ERROR_ACTION_STOP : BLOCK_ERROR_ACTION_REPORT;
 645         break;
 646     case BLOCKDEV_ON_ERROR_STOP:
 647         action = BLOCK_ERROR_ACTION_STOP;
 648         break;
 649     case BLOCKDEV_ON_ERROR_REPORT:
 650         action = BLOCK_ERROR_ACTION_REPORT;
 651         break;
 652     case BLOCKDEV_ON_ERROR_IGNORE:
 653         action = BLOCK_ERROR_ACTION_IGNORE;
 654         break;
 655     default:
 656         abort();
 657     }
 658     if (!block_job_is_internal(job)) {
 659         qapi_event_send_block_job_error(job->id,
 660                                         is_read ? IO_OPERATION_TYPE_READ :
 661                                         IO_OPERATION_TYPE_WRITE,
 662                                         action, &error_abort);
 663     }
 664     if (action == BLOCK_ERROR_ACTION_STOP) {
 665         /* make the pause user visible, which will be resumed from QMP. */
 666         block_job_user_pause(job);
 667         block_job_iostatus_set_err(job, error);
 668     }
 669     return action;
 670 }
 671
 672 typedef struct {
 673     BlockJob *job;
 674     AioContext *aio_context;
 675     BlockJobDeferToMainLoopFn *fn;
 676     void *opaque;
 677 } BlockJobDeferToMainLoopData;
 678
 679 static void block_job_defer_to_main_loop_bh(void *opaque)
 680 {
 681     BlockJobDeferToMainLoopData *data = opaque;
 682     AioContext *aio_context;
 683
 684     /* Prevent race with block_job_defer_to_main_loop() */
 685     aio_context_acquire(data->aio_context);
 686
 687     /* Fetch BDS AioContext again, in case it has changed */
 688     aio_context = blk_get_aio_context(data->job->blk);
 689     aio_context_acquire(aio_context);
 690
 691     data->job->deferred_to_main_loop = false;
 692     data->fn(data->job, data->opaque);
 693
 694     aio_context_release(aio_context);
 695
 696     aio_context_release(data->aio_context);
 697
 698     g_free(data);
 699 }
 700
 701 void block_job_defer_to_main_loop(BlockJob *job,
 702                                   BlockJobDeferToMainLoopFn *fn,
 703                                   void *opaque)
 704 {
 705     BlockJobDeferToMainLoopData *data = g_malloc(sizeof(*data));
 706     data->job = job;
 707     data->aio_context = blk_get_aio_context(job->blk);
 708     data->fn = fn;
 709     data->opaque = opaque;
 710     job->deferred_to_main_loop = true;
 711
 712     aio_bh_schedule_oneshot(qemu_get_aio_context(),
 713                             block_job_defer_to_main_loop_bh, data);
 714 }
 715
 716 BlockJobTxn *block_job_txn_new(void)
 717 {
 718     BlockJobTxn *txn = g_new0(BlockJobTxn, 1);
 719     QLIST_INIT(&txn->jobs);
 720     txn->refcnt = 1;
 721     return txn;
 722 }
 723
 724 static void block_job_txn_ref(BlockJobTxn *txn)
 725 {
 726     txn->refcnt++;
 727 }
 728
 729 void block_job_txn_unref(BlockJobTxn *txn)
 730 {
 731     if (txn && --txn->refcnt == 0) {
 732         g_free(txn);
 733     }
 734 }
 735
 736 void block_job_txn_add_job(BlockJobTxn *txn, BlockJob *job)
 737 {
 738     if (!txn) {
 739         return;
 740     }
 741
 742     assert(!job->txn);
 743     job->txn = txn;
 744
 745     QLIST_INSERT_HEAD(&txn->jobs, job, txn_list);
 746     block_job_txn_ref(txn);
 747 }