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