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