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