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