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