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 "block/block.h"
  29 #include "block/blockjob_int.h"
  30 #include "block/block_int.h"
  31 #include "block/trace.h"
  32 #include "sysemu/block-backend.h"
  33 #include "qapi/error.h"
  34 #include "qapi/qapi-events-block-core.h"
  35 #include "qapi/qmp/qerror.h"
  36 #include "qemu/coroutine.h"
  37 #include "qemu/timer.h"
  38
  39 /*
  40  * The block job API is composed of two categories of functions.
  41  *
  42  * The first includes functions used by the monitor.  The monitor is
  43  * peculiar in that it accesses the block job list with block_job_get, and
  44  * therefore needs consistency across block_job_get and the actual operation
  45  * (e.g. block_job_set_speed).  The consistency is achieved with
  46  * aio_context_acquire/release.  These functions are declared in blockjob.h.
  47  *
  48  * The second includes functions used by the block job drivers and sometimes
  49  * by the core block layer.  These do not care about locking, because the
  50  * whole coroutine runs under the AioContext lock, and are declared in
  51  * blockjob_int.h.
  52  */
  53
  54 static bool is_block_job(Job *job)
  55 {
  56     return job_type(job) == JOB_TYPE_BACKUP ||
  57            job_type(job) == JOB_TYPE_COMMIT ||
  58            job_type(job) == JOB_TYPE_MIRROR ||
  59            job_type(job) == JOB_TYPE_STREAM;
  60 }
  61
  62 BlockJob *block_job_next(BlockJob *bjob)
  63 {
  64     Job *job = bjob ? &bjob->job : NULL;
  65
  66     do {
  67         job = job_next(job);
  68     } while (job && !is_block_job(job));
  69
  70     return job ? container_of(job, BlockJob, job) : NULL;
  71 }
  72
  73 BlockJob *block_job_get(const char *id)
  74 {
  75     Job *job = job_get(id);
  76
  77     if (job && is_block_job(job)) {
  78         return container_of(job, BlockJob, job);
  79     } else {
  80         return NULL;
  81     }
  82 }
  83
  84 static void block_job_attached_aio_context(AioContext *new_context,
  85                                            void *opaque);
  86 static void block_job_detach_aio_context(void *opaque);
  87
  88 void block_job_free(Job *job)
  89 {
  90     BlockJob *bjob = container_of(job, BlockJob, job);
  91     BlockDriverState *bs = blk_bs(bjob->blk);
  92
  93     bs->job = NULL;
  94     block_job_remove_all_bdrv(bjob);
  95     blk_remove_aio_context_notifier(bjob->blk,
  96                                     block_job_attached_aio_context,
  97                                     block_job_detach_aio_context, bjob);
  98     blk_unref(bjob->blk);
  99     error_free(bjob->blocker);
 100 }
 101
 102 static void block_job_attached_aio_context(AioContext *new_context,
 103                                            void *opaque)
 104 {
 105     BlockJob *job = opaque;
 106
 107     job->job.aio_context = new_context;
 108     if (job->driver->attached_aio_context) {
 109         job->driver->attached_aio_context(job, new_context);
 110     }
 111
 112     job_resume(&job->job);
 113 }
 114
 115 void block_job_drain(Job *job)
 116 {
 117     BlockJob *bjob = container_of(job, BlockJob, job);
 118
 119     blk_drain(bjob->blk);
 120     if (bjob->driver->drain) {
 121         bjob->driver->drain(bjob);
 122     }
 123 }
 124
 125 static void block_job_detach_aio_context(void *opaque)
 126 {
 127     BlockJob *job = opaque;
 128
 129     /* In case the job terminates during aio_poll()... */
 130     job_ref(&job->job);
 131
 132     job_pause(&job->job);
 133
 134     while (!job->job.paused && !job_is_completed(&job->job)) {
 135         job_drain(&job->job);
 136     }
 137
 138     job->job.aio_context = NULL;
 139     job_unref(&job->job);
 140 }
 141
 142 static char *child_job_get_parent_desc(BdrvChild *c)
 143 {
 144     BlockJob *job = c->opaque;
 145     return g_strdup_printf("%s job '%s'", job_type_str(&job->job), job->job.id);
 146 }
 147
 148 static void child_job_drained_begin(BdrvChild *c)
 149 {
 150     BlockJob *job = c->opaque;
 151     job_pause(&job->job);
 152 }
 153
 154 static void child_job_drained_end(BdrvChild *c)
 155 {
 156     BlockJob *job = c->opaque;
 157     job_resume(&job->job);
 158 }
 159
 160 static const BdrvChildRole child_job = {
 161     .get_parent_desc    = child_job_get_parent_desc,
 162     .drained_begin      = child_job_drained_begin,
 163     .drained_end        = child_job_drained_end,
 164     .stay_at_node       = true,
 165 };
 166
 167 void block_job_remove_all_bdrv(BlockJob *job)
 168 {
 169     GSList *l;
 170     for (l = job->nodes; l; l = l->next) {
 171         BdrvChild *c = l->data;
 172         bdrv_op_unblock_all(c->bs, job->blocker);
 173         bdrv_root_unref_child(c);
 174     }
 175     g_slist_free(job->nodes);
 176     job->nodes = NULL;
 177 }
 178
 179 int block_job_add_bdrv(BlockJob *job, const char *name, BlockDriverState *bs,
 180                        uint64_t perm, uint64_t shared_perm, Error **errp)
 181 {
 182     BdrvChild *c;
 183
 184     c = bdrv_root_attach_child(bs, name, &child_job, perm, shared_perm,
 185                                job, errp);
 186     if (c == NULL) {
 187         return -EPERM;
 188     }
 189
 190     job->nodes = g_slist_prepend(job->nodes, c);
 191     bdrv_ref(bs);
 192     bdrv_op_block_all(bs, job->blocker);
 193
 194     return 0;
 195 }
 196
 197 bool block_job_is_internal(BlockJob *job)
 198 {
 199     return (job->job.id == NULL);
 200 }
 201
 202 const BlockJobDriver *block_job_driver(BlockJob *job)
 203 {
 204     return job->driver;
 205 }
 206
 207 /* Assumes the job_mutex is held */
 208 static bool job_timer_pending(Job *job)
 209 {
 210     return timer_pending(&job->sleep_timer);
 211 }
 212
 213 void block_job_set_speed(BlockJob *job, int64_t speed, Error **errp)
 214 {
 215     int64_t old_speed = job->speed;
 216
 217     if (job_apply_verb(&job->job, JOB_VERB_SET_SPEED, errp)) {
 218         return;
 219     }
 220     if (speed < 0) {
 221         error_setg(errp, QERR_INVALID_PARAMETER, "speed");
 222         return;
 223     }
 224
 225     ratelimit_set_speed(&job->limit, speed, BLOCK_JOB_SLICE_TIME);
 226
 227     job->speed = speed;
 228     if (speed && speed <= old_speed) {
 229         return;
 230     }
 231
 232     /* kick only if a timer is pending */
 233     job_enter_cond(&job->job, job_timer_pending);
 234 }
 235
 236 int64_t block_job_ratelimit_get_delay(BlockJob *job, uint64_t n)
 237 {
 238     if (!job->speed) {
 239         return 0;
 240     }
 241
 242     return ratelimit_calculate_delay(&job->limit, n);
 243 }
 244
 245 void block_job_dismiss(BlockJob **jobptr, Error **errp)
 246 {
 247     BlockJob *job = *jobptr;
 248     /* similarly to _complete, this is QMP-interface only. */
 249     assert(job->job.id);
 250     if (job_apply_verb(&job->job, JOB_VERB_DISMISS, errp)) {
 251         return;
 252     }
 253
 254     job_do_dismiss(&job->job);
 255     *jobptr = NULL;
 256 }
 257
 258 void block_job_cancel(BlockJob *job, bool force)
 259 {
 260     if (job->job.status == JOB_STATUS_CONCLUDED) {
 261         job_do_dismiss(&job->job);
 262         return;
 263     }
 264     job_cancel_async(&job->job, force);
 265     if (!job_started(&job->job)) {
 266         block_job_completed(job, -ECANCELED);
 267     } else if (job->job.deferred_to_main_loop) {
 268         job_completed_txn_abort(&job->job);
 269     } else {
 270         block_job_enter(job);
 271     }
 272 }
 273
 274 void block_job_user_cancel(BlockJob *job, bool force, Error **errp)
 275 {
 276     if (job_apply_verb(&job->job, JOB_VERB_CANCEL, errp)) {
 277         return;
 278     }
 279     block_job_cancel(job, force);
 280 }
 281
 282 /* A wrapper around block_job_cancel() taking an Error ** parameter so it may be
 283  * used with job_finish_sync() without the need for (rather nasty) function
 284  * pointer casts there. */
 285 static void block_job_cancel_err(Job *job, Error **errp)
 286 {
 287     BlockJob *bjob = container_of(job, BlockJob, job);
 288     assert(is_block_job(job));
 289     block_job_cancel(bjob, false);
 290 }
 291
 292 int block_job_cancel_sync(BlockJob *job)
 293 {
 294     return job_finish_sync(&job->job, &block_job_cancel_err, NULL);
 295 }
 296
 297 void block_job_cancel_sync_all(void)
 298 {
 299     BlockJob *job;
 300     AioContext *aio_context;
 301
 302     while ((job = block_job_next(NULL))) {
 303         aio_context = blk_get_aio_context(job->blk);
 304         aio_context_acquire(aio_context);
 305         block_job_cancel_sync(job);
 306         aio_context_release(aio_context);
 307     }
 308 }
 309
 310 int block_job_complete_sync(BlockJob *job, Error **errp)
 311 {
 312     return job_finish_sync(&job->job, job_complete, errp);
 313 }
 314
 315 void block_job_progress_update(BlockJob *job, uint64_t done)
 316 {
 317     job->offset += done;
 318 }
 319
 320 void block_job_progress_set_remaining(BlockJob *job, uint64_t remaining)
 321 {
 322     job->len = job->offset + remaining;
 323 }
 324
 325 BlockJobInfo *block_job_query(BlockJob *job, Error **errp)
 326 {
 327     BlockJobInfo *info;
 328
 329     if (block_job_is_internal(job)) {
 330         error_setg(errp, "Cannot query QEMU internal jobs");
 331         return NULL;
 332     }
 333     info = g_new0(BlockJobInfo, 1);
 334     info->type      = g_strdup(job_type_str(&job->job));
 335     info->device    = g_strdup(job->job.id);
 336     info->len       = job->len;
 337     info->busy      = atomic_read(&job->job.busy);
 338     info->paused    = job->job.pause_count > 0;
 339     info->offset    = job->offset;
 340     info->speed     = job->speed;
 341     info->io_status = job->iostatus;
 342     info->ready     = job->ready;
 343     info->status    = job->job.status;
 344     info->auto_finalize = job->job.auto_finalize;
 345     info->auto_dismiss  = job->job.auto_dismiss;
 346     info->has_error = job->job.ret != 0;
 347     info->error     = job->job.ret ? g_strdup(strerror(-job->job.ret)) : NULL;
 348     return info;
 349 }
 350
 351 static void block_job_iostatus_set_err(BlockJob *job, int error)
 352 {
 353     if (job->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
 354         job->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE :
 355                                           BLOCK_DEVICE_IO_STATUS_FAILED;
 356     }
 357 }
 358
 359 static void block_job_event_cancelled(Notifier *n, void *opaque)
 360 {
 361     BlockJob *job = opaque;
 362
 363     if (block_job_is_internal(job)) {
 364         return;
 365     }
 366
 367     qapi_event_send_block_job_cancelled(job_type(&job->job),
 368                                         job->job.id,
 369                                         job->len,
 370                                         job->offset,
 371                                         job->speed,
 372                                         &error_abort);
 373 }
 374
 375 static void block_job_event_completed(Notifier *n, void *opaque)
 376 {
 377     BlockJob *job = opaque;
 378     const char *msg = NULL;
 379
 380     if (block_job_is_internal(job)) {
 381         return;
 382     }
 383
 384     if (job->job.ret < 0) {
 385         msg = strerror(-job->job.ret);
 386     }
 387
 388     qapi_event_send_block_job_completed(job_type(&job->job),
 389                                         job->job.id,
 390                                         job->len,
 391                                         job->offset,
 392                                         job->speed,
 393                                         !!msg,
 394                                         msg,
 395                                         &error_abort);
 396 }
 397
 398 static void block_job_event_pending(Notifier *n, void *opaque)
 399 {
 400     BlockJob *job = opaque;
 401
 402     if (block_job_is_internal(job)) {
 403         return;
 404     }
 405
 406     qapi_event_send_block_job_pending(job_type(&job->job),
 407                                       job->job.id,
 408                                       &error_abort);
 409 }
 410
 411 /*
 412  * API for block job drivers and the block layer.  These functions are
 413  * declared in blockjob_int.h.
 414  */
 415
 416 void *block_job_create(const char *job_id, const BlockJobDriver *driver,
 417                        JobTxn *txn, BlockDriverState *bs, uint64_t perm,
 418                        uint64_t shared_perm, int64_t speed, int flags,
 419                        BlockCompletionFunc *cb, void *opaque, Error **errp)
 420 {
 421     BlockBackend *blk;
 422     BlockJob *job;
 423     int ret;
 424
 425     if (bs->job) {
 426         error_setg(errp, QERR_DEVICE_IN_USE, bdrv_get_device_name(bs));
 427         return NULL;
 428     }
 429
 430     if (job_id == NULL && !(flags & JOB_INTERNAL)) {
 431         job_id = bdrv_get_device_name(bs);
 432     }
 433
 434     blk = blk_new(perm, shared_perm);
 435     ret = blk_insert_bs(blk, bs, errp);
 436     if (ret < 0) {
 437         blk_unref(blk);
 438         return NULL;
 439     }
 440
 441     job = job_create(job_id, &driver->job_driver, txn, blk_get_aio_context(blk),
 442                      flags, cb, opaque, errp);
 443     if (job == NULL) {
 444         blk_unref(blk);
 445         return NULL;
 446     }
 447
 448     assert(is_block_job(&job->job));
 449     assert(job->job.driver->free == &block_job_free);
 450     assert(job->job.driver->user_resume == &block_job_user_resume);
 451     assert(job->job.driver->drain == &block_job_drain);
 452
 453     job->driver        = driver;
 454     job->blk           = blk;
 455
 456     job->finalize_cancelled_notifier.notify = block_job_event_cancelled;
 457     job->finalize_completed_notifier.notify = block_job_event_completed;
 458     job->pending_notifier.notify = block_job_event_pending;
 459
 460     notifier_list_add(&job->job.on_finalize_cancelled,
 461                       &job->finalize_cancelled_notifier);
 462     notifier_list_add(&job->job.on_finalize_completed,
 463                       &job->finalize_completed_notifier);
 464     notifier_list_add(&job->job.on_pending, &job->pending_notifier);
 465
 466     error_setg(&job->blocker, "block device is in use by block job: %s",
 467                job_type_str(&job->job));
 468     block_job_add_bdrv(job, "main node", bs, 0, BLK_PERM_ALL, &error_abort);
 469     bs->job = job;
 470
 471     bdrv_op_unblock(bs, BLOCK_OP_TYPE_DATAPLANE, job->blocker);
 472
 473     blk_add_aio_context_notifier(blk, block_job_attached_aio_context,
 474                                  block_job_detach_aio_context, job);
 475
 476     /* Only set speed when necessary to avoid NotSupported error */
 477     if (speed != 0) {
 478         Error *local_err = NULL;
 479
 480         block_job_set_speed(job, speed, &local_err);
 481         if (local_err) {
 482             job_early_fail(&job->job);
 483             error_propagate(errp, local_err);
 484             return NULL;
 485         }
 486     }
 487
 488     return job;
 489 }
 490
 491 void block_job_completed(BlockJob *job, int ret)
 492 {
 493     assert(job && job->job.txn && !job_is_completed(&job->job));
 494     assert(blk_bs(job->blk)->job == job);
 495     job->job.ret = ret;
 496     job_update_rc(&job->job);
 497     trace_block_job_completed(job, ret, job->job.ret);
 498     if (job->job.ret) {
 499         job_completed_txn_abort(&job->job);
 500     } else {
 501         job_completed_txn_success(&job->job);
 502     }
 503 }
 504
 505 void block_job_enter(BlockJob *job)
 506 {
 507     job_enter_cond(&job->job, NULL);
 508 }
 509
 510 void block_job_yield(BlockJob *job)
 511 {
 512     assert(job->job.busy);
 513
 514     /* Check cancellation *before* setting busy = false, too!  */
 515     if (job_is_cancelled(&job->job)) {
 516         return;
 517     }
 518
 519     if (!job_should_pause(&job->job)) {
 520         job_do_yield(&job->job, -1);
 521     }
 522
 523     job_pause_point(&job->job);
 524 }
 525
 526 void block_job_iostatus_reset(BlockJob *job)
 527 {
 528     if (job->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
 529         return;
 530     }
 531     assert(job->job.user_paused && job->job.pause_count > 0);
 532     job->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
 533 }
 534
 535 void block_job_user_resume(Job *job)
 536 {
 537     BlockJob *bjob = container_of(job, BlockJob, job);
 538     block_job_iostatus_reset(bjob);
 539 }
 540
 541 void block_job_event_ready(BlockJob *job)
 542 {
 543     job_state_transition(&job->job, JOB_STATUS_READY);
 544     job->ready = true;
 545
 546     if (block_job_is_internal(job)) {
 547         return;
 548     }
 549
 550     qapi_event_send_block_job_ready(job_type(&job->job),
 551                                     job->job.id,
 552                                     job->len,
 553                                     job->offset,
 554                                     job->speed, &error_abort);
 555 }
 556
 557 BlockErrorAction block_job_error_action(BlockJob *job, BlockdevOnError on_err,
 558                                         int is_read, int error)
 559 {
 560     BlockErrorAction action;
 561
 562     switch (on_err) {
 563     case BLOCKDEV_ON_ERROR_ENOSPC:
 564     case BLOCKDEV_ON_ERROR_AUTO:
 565         action = (error == ENOSPC) ?
 566                  BLOCK_ERROR_ACTION_STOP : BLOCK_ERROR_ACTION_REPORT;
 567         break;
 568     case BLOCKDEV_ON_ERROR_STOP:
 569         action = BLOCK_ERROR_ACTION_STOP;
 570         break;
 571     case BLOCKDEV_ON_ERROR_REPORT:
 572         action = BLOCK_ERROR_ACTION_REPORT;
 573         break;
 574     case BLOCKDEV_ON_ERROR_IGNORE:
 575         action = BLOCK_ERROR_ACTION_IGNORE;
 576         break;
 577     default:
 578         abort();
 579     }
 580     if (!block_job_is_internal(job)) {
 581         qapi_event_send_block_job_error(job->job.id,
 582                                         is_read ? IO_OPERATION_TYPE_READ :
 583                                         IO_OPERATION_TYPE_WRITE,
 584                                         action, &error_abort);
 585     }
 586     if (action == BLOCK_ERROR_ACTION_STOP) {
 587         job_pause(&job->job);
 588         /* make the pause user visible, which will be resumed from QMP. */
 589         job->job.user_paused = true;
 590         block_job_iostatus_set_err(job, error);
 591     }
 592     return action;
 593 }