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