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