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 "block/block.h"
  28 #include "block/blockjob_int.h"
  29 #include "block/block_int.h"
  30 #include "block/trace.h"
  31 #include "sysemu/block-backend.h"
  32 #include "qapi/error.h"
  33 #include "qapi/qapi-events-block-core.h"
  34 #include "qapi/qmp/qerror.h"
  35 #include "qemu/coroutine.h"
  36 #include "qemu/main-loop.h"
  37 #include "qemu/timer.h"
  38
  39 static bool is_block_job(Job *job)
  40 {
  41     return job_type(job) == JOB_TYPE_BACKUP ||
  42            job_type(job) == JOB_TYPE_COMMIT ||
  43            job_type(job) == JOB_TYPE_MIRROR ||
  44            job_type(job) == JOB_TYPE_STREAM;
  45 }
  46
  47 BlockJob *block_job_next_locked(BlockJob *bjob)
  48 {
  49     Job *job = bjob ? &bjob->job : NULL;
  50     GLOBAL_STATE_CODE();
  51
  52     do {
  53         job = job_next_locked(job);
  54     } while (job && !is_block_job(job));
  55
  56     return job ? container_of(job, BlockJob, job) : NULL;
  57 }
  58
  59 BlockJob *block_job_next(BlockJob *bjob)
  60 {
  61     JOB_LOCK_GUARD();
  62     return block_job_next_locked(bjob);
  63 }
  64
  65 BlockJob *block_job_get_locked(const char *id)
  66 {
  67     Job *job = job_get_locked(id);
  68     GLOBAL_STATE_CODE();
  69
  70     if (job && is_block_job(job)) {
  71         return container_of(job, BlockJob, job);
  72     } else {
  73         return NULL;
  74     }
  75 }
  76
  77 BlockJob *block_job_get(const char *id)
  78 {
  79     JOB_LOCK_GUARD();
  80     return block_job_get_locked(id);
  81 }
  82
  83 void block_job_free(Job *job)
  84 {
  85     BlockJob *bjob = container_of(job, BlockJob, job);
  86     GLOBAL_STATE_CODE();
  87
  88     block_job_remove_all_bdrv(bjob);
  89     ratelimit_destroy(&bjob->limit);
  90     error_free(bjob->blocker);
  91 }
  92
  93 static char *child_job_get_parent_desc(BdrvChild *c)
  94 {
  95     BlockJob *job = c->opaque;
  96     return g_strdup_printf("%s job '%s'", job_type_str(&job->job), job->job.id);
  97 }
  98
  99 static void child_job_drained_begin(BdrvChild *c)
 100 {
 101     BlockJob *job = c->opaque;
 102     job_pause(&job->job);
 103 }
 104
 105 static bool child_job_drained_poll(BdrvChild *c)
 106 {
 107     BlockJob *bjob = c->opaque;
 108     Job *job = &bjob->job;
 109     const BlockJobDriver *drv = block_job_driver(bjob);
 110
 111     /* An inactive or completed job doesn't have any pending requests. Jobs
 112      * with !job->busy are either already paused or have a pause point after
 113      * being reentered, so no job driver code will run before they pause. */
 114     WITH_JOB_LOCK_GUARD() {
 115         if (!job->busy || job_is_completed_locked(job)) {
 116             return false;
 117         }
 118     }
 119
 120     /* Otherwise, assume that it isn't fully stopped yet, but allow the job to
 121      * override this assumption. */
 122     if (drv->drained_poll) {
 123         return drv->drained_poll(bjob);
 124     } else {
 125         return true;
 126     }
 127 }
 128
 129 static void child_job_drained_end(BdrvChild *c, int *drained_end_counter)
 130 {
 131     BlockJob *job = c->opaque;
 132     job_resume(&job->job);
 133 }
 134
 135 static bool child_job_can_set_aio_ctx(BdrvChild *c, AioContext *ctx,
 136                                       GSList **ignore, Error **errp)
 137 {
 138     BlockJob *job = c->opaque;
 139     GSList *l;
 140
 141     for (l = job->nodes; l; l = l->next) {
 142         BdrvChild *sibling = l->data;
 143         if (!bdrv_child_can_set_aio_context(sibling, ctx, ignore, errp)) {
 144             return false;
 145         }
 146     }
 147     return true;
 148 }
 149
 150 static void child_job_set_aio_ctx(BdrvChild *c, AioContext *ctx,
 151                                   GSList **ignore)
 152 {
 153     BlockJob *job = c->opaque;
 154     GSList *l;
 155
 156     for (l = job->nodes; l; l = l->next) {
 157         BdrvChild *sibling = l->data;
 158         if (g_slist_find(*ignore, sibling)) {
 159             continue;
 160         }
 161         *ignore = g_slist_prepend(*ignore, sibling);
 162         bdrv_set_aio_context_ignore(sibling->bs, ctx, ignore);
 163     }
 164
 165     job->job.aio_context = ctx;
 166 }
 167
 168 static AioContext *child_job_get_parent_aio_context(BdrvChild *c)
 169 {
 170     BlockJob *job = c->opaque;
 171
 172     return job->job.aio_context;
 173 }
 174
 175 static const BdrvChildClass child_job = {
 176     .get_parent_desc    = child_job_get_parent_desc,
 177     .drained_begin      = child_job_drained_begin,
 178     .drained_poll       = child_job_drained_poll,
 179     .drained_end        = child_job_drained_end,
 180     .can_set_aio_ctx    = child_job_can_set_aio_ctx,
 181     .set_aio_ctx        = child_job_set_aio_ctx,
 182     .stay_at_node       = true,
 183     .get_parent_aio_context = child_job_get_parent_aio_context,
 184 };
 185
 186 void block_job_remove_all_bdrv(BlockJob *job)
 187 {
 188     GLOBAL_STATE_CODE();
 189     /*
 190      * bdrv_root_unref_child() may reach child_job_[can_]set_aio_ctx(),
 191      * which will also traverse job->nodes, so consume the list one by
 192      * one to make sure that such a concurrent access does not attempt
 193      * to process an already freed BdrvChild.
 194      */
 195     while (job->nodes) {
 196         GSList *l = job->nodes;
 197         BdrvChild *c = l->data;
 198
 199         job->nodes = l->next;
 200
 201         bdrv_op_unblock_all(c->bs, job->blocker);
 202         bdrv_root_unref_child(c);
 203
 204         g_slist_free_1(l);
 205     }
 206 }
 207
 208 bool block_job_has_bdrv(BlockJob *job, BlockDriverState *bs)
 209 {
 210     GSList *el;
 211     GLOBAL_STATE_CODE();
 212
 213     for (el = job->nodes; el; el = el->next) {
 214         BdrvChild *c = el->data;
 215         if (c->bs == bs) {
 216             return true;
 217         }
 218     }
 219
 220     return false;
 221 }
 222
 223 int block_job_add_bdrv(BlockJob *job, const char *name, BlockDriverState *bs,
 224                        uint64_t perm, uint64_t shared_perm, Error **errp)
 225 {
 226     BdrvChild *c;
 227     bool need_context_ops;
 228     GLOBAL_STATE_CODE();
 229
 230     bdrv_ref(bs);
 231
 232     need_context_ops = bdrv_get_aio_context(bs) != job->job.aio_context;
 233
 234     if (need_context_ops && job->job.aio_context != qemu_get_aio_context()) {
 235         aio_context_release(job->job.aio_context);
 236     }
 237     c = bdrv_root_attach_child(bs, name, &child_job, 0, perm, shared_perm, job,
 238                                errp);
 239     if (need_context_ops && job->job.aio_context != qemu_get_aio_context()) {
 240         aio_context_acquire(job->job.aio_context);
 241     }
 242     if (c == NULL) {
 243         return -EPERM;
 244     }
 245
 246     job->nodes = g_slist_prepend(job->nodes, c);
 247     bdrv_op_block_all(bs, job->blocker);
 248
 249     return 0;
 250 }
 251
 252 static void block_job_on_idle(Notifier *n, void *opaque)
 253 {
 254     aio_wait_kick();
 255 }
 256
 257 bool block_job_is_internal(BlockJob *job)
 258 {
 259     return (job->job.id == NULL);
 260 }
 261
 262 const BlockJobDriver *block_job_driver(BlockJob *job)
 263 {
 264     return container_of(job->job.driver, BlockJobDriver, job_driver);
 265 }
 266
 267 /* Assumes the job_mutex is held */
 268 static bool job_timer_pending(Job *job)
 269 {
 270     return timer_pending(&job->sleep_timer);
 271 }
 272
 273 bool block_job_set_speed_locked(BlockJob *job, int64_t speed, Error **errp)
 274 {
 275     const BlockJobDriver *drv = block_job_driver(job);
 276     int64_t old_speed = job->speed;
 277
 278     GLOBAL_STATE_CODE();
 279
 280     if (job_apply_verb_locked(&job->job, JOB_VERB_SET_SPEED, errp) < 0) {
 281         return false;
 282     }
 283     if (speed < 0) {
 284         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "speed",
 285                    "a non-negative value");
 286         return false;
 287     }
 288
 289     ratelimit_set_speed(&job->limit, speed, BLOCK_JOB_SLICE_TIME);
 290
 291     job->speed = speed;
 292
 293     if (drv->set_speed) {
 294         job_unlock();
 295         drv->set_speed(job, speed);
 296         job_lock();
 297     }
 298
 299     if (speed && speed <= old_speed) {
 300         return true;
 301     }
 302
 303     /* kick only if a timer is pending */
 304     job_enter_cond_locked(&job->job, job_timer_pending);
 305
 306     return true;
 307 }
 308
 309 bool block_job_set_speed(BlockJob *job, int64_t speed, Error **errp)
 310 {
 311     JOB_LOCK_GUARD();
 312     return block_job_set_speed_locked(job, speed, errp);
 313 }
 314
 315 int64_t block_job_ratelimit_get_delay(BlockJob *job, uint64_t n)
 316 {
 317     IO_CODE();
 318     return ratelimit_calculate_delay(&job->limit, n);
 319 }
 320
 321 BlockJobInfo *block_job_query_locked(BlockJob *job, Error **errp)
 322 {
 323     BlockJobInfo *info;
 324     uint64_t progress_current, progress_total;
 325
 326     GLOBAL_STATE_CODE();
 327
 328     if (block_job_is_internal(job)) {
 329         error_setg(errp, "Cannot query QEMU internal jobs");
 330         return NULL;
 331     }
 332
 333     progress_get_snapshot(&job->job.progress, &progress_current,
 334                           &progress_total);
 335
 336     info = g_new0(BlockJobInfo, 1);
 337     info->type      = g_strdup(job_type_str(&job->job));
 338     info->device    = g_strdup(job->job.id);
 339     info->busy      = qatomic_read(&job->job.busy);
 340     info->paused    = job->job.pause_count > 0;
 341     info->offset    = progress_current;
 342     info->len       = progress_total;
 343     info->speed     = job->speed;
 344     info->io_status = job->iostatus;
 345     info->ready     = job_is_ready_locked(&job->job),
 346     info->status    = job->job.status;
 347     info->auto_finalize = job->job.auto_finalize;
 348     info->auto_dismiss  = job->job.auto_dismiss;
 349     if (job->job.ret) {
 350         info->has_error = true;
 351         info->error = job->job.err ?
 352                         g_strdup(error_get_pretty(job->job.err)) :
 353                         g_strdup(strerror(-job->job.ret));
 354     }
 355     return info;
 356 }
 357
 358 BlockJobInfo *block_job_query(BlockJob *job, Error **errp)
 359 {
 360     JOB_LOCK_GUARD();
 361     return block_job_query_locked(job, errp);
 362 }
 363
 364 static void block_job_iostatus_set_err(BlockJob *job, int error)
 365 {
 366     if (job->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
 367         job->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE :
 368                                           BLOCK_DEVICE_IO_STATUS_FAILED;
 369     }
 370 }
 371
 372 static void block_job_event_cancelled(Notifier *n, void *opaque)
 373 {
 374     BlockJob *job = opaque;
 375     uint64_t progress_current, progress_total;
 376
 377     if (block_job_is_internal(job)) {
 378         return;
 379     }
 380
 381     progress_get_snapshot(&job->job.progress, &progress_current,
 382                           &progress_total);
 383
 384     qapi_event_send_block_job_cancelled(job_type(&job->job),
 385                                         job->job.id,
 386                                         progress_total,
 387                                         progress_current,
 388                                         job->speed);
 389 }
 390
 391 static void block_job_event_completed(Notifier *n, void *opaque)
 392 {
 393     BlockJob *job = opaque;
 394     const char *msg = NULL;
 395     uint64_t progress_current, progress_total;
 396
 397     if (block_job_is_internal(job)) {
 398         return;
 399     }
 400
 401     if (job->job.ret < 0) {
 402         msg = error_get_pretty(job->job.err);
 403     }
 404
 405     progress_get_snapshot(&job->job.progress, &progress_current,
 406                           &progress_total);
 407
 408     qapi_event_send_block_job_completed(job_type(&job->job),
 409                                         job->job.id,
 410                                         progress_total,
 411                                         progress_current,
 412                                         job->speed,
 413                                         !!msg,
 414                                         msg);
 415 }
 416
 417 static void block_job_event_pending(Notifier *n, void *opaque)
 418 {
 419     BlockJob *job = opaque;
 420
 421     if (block_job_is_internal(job)) {
 422         return;
 423     }
 424
 425     qapi_event_send_block_job_pending(job_type(&job->job),
 426                                       job->job.id);
 427 }
 428
 429 static void block_job_event_ready(Notifier *n, void *opaque)
 430 {
 431     BlockJob *job = opaque;
 432     uint64_t progress_current, progress_total;
 433
 434     if (block_job_is_internal(job)) {
 435         return;
 436     }
 437
 438     progress_get_snapshot(&job->job.progress, &progress_current,
 439                           &progress_total);
 440
 441     qapi_event_send_block_job_ready(job_type(&job->job),
 442                                     job->job.id,
 443                                     progress_total,
 444                                     progress_current,
 445                                     job->speed);
 446 }
 447
 448
 449 void *block_job_create(const char *job_id, const BlockJobDriver *driver,
 450                        JobTxn *txn, BlockDriverState *bs, uint64_t perm,
 451                        uint64_t shared_perm, int64_t speed, int flags,
 452                        BlockCompletionFunc *cb, void *opaque, Error **errp)
 453 {
 454     BlockJob *job;
 455     int ret;
 456     GLOBAL_STATE_CODE();
 457
 458     if (job_id == NULL && !(flags & JOB_INTERNAL)) {
 459         job_id = bdrv_get_device_name(bs);
 460     }
 461
 462     job = job_create(job_id, &driver->job_driver, txn, bdrv_get_aio_context(bs),
 463                      flags, cb, opaque, errp);
 464     if (job == NULL) {
 465         return NULL;
 466     }
 467
 468     assert(is_block_job(&job->job));
 469     assert(job->job.driver->free == &block_job_free);
 470     assert(job->job.driver->user_resume == &block_job_user_resume);
 471
 472     ratelimit_init(&job->limit);
 473
 474     job->finalize_cancelled_notifier.notify = block_job_event_cancelled;
 475     job->finalize_completed_notifier.notify = block_job_event_completed;
 476     job->pending_notifier.notify = block_job_event_pending;
 477     job->ready_notifier.notify = block_job_event_ready;
 478     job->idle_notifier.notify = block_job_on_idle;
 479
 480     WITH_JOB_LOCK_GUARD() {
 481         notifier_list_add(&job->job.on_finalize_cancelled,
 482                           &job->finalize_cancelled_notifier);
 483         notifier_list_add(&job->job.on_finalize_completed,
 484                           &job->finalize_completed_notifier);
 485         notifier_list_add(&job->job.on_pending, &job->pending_notifier);
 486         notifier_list_add(&job->job.on_ready, &job->ready_notifier);
 487         notifier_list_add(&job->job.on_idle, &job->idle_notifier);
 488     }
 489
 490     error_setg(&job->blocker, "block device is in use by block job: %s",
 491                job_type_str(&job->job));
 492
 493     ret = block_job_add_bdrv(job, "main node", bs, perm, shared_perm, errp);
 494     if (ret < 0) {
 495         goto fail;
 496     }
 497
 498     bdrv_op_unblock(bs, BLOCK_OP_TYPE_DATAPLANE, job->blocker);
 499
 500     if (!block_job_set_speed(job, speed, errp)) {
 501         goto fail;
 502     }
 503
 504     return job;
 505
 506 fail:
 507     job_early_fail(&job->job);
 508     return NULL;
 509 }
 510
 511 void block_job_iostatus_reset_locked(BlockJob *job)
 512 {
 513     GLOBAL_STATE_CODE();
 514     if (job->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
 515         return;
 516     }
 517     assert(job->job.user_paused && job->job.pause_count > 0);
 518     job->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
 519 }
 520
 521 void block_job_iostatus_reset(BlockJob *job)
 522 {
 523     JOB_LOCK_GUARD();
 524     block_job_iostatus_reset_locked(job);
 525 }
 526
 527 void block_job_user_resume(Job *job)
 528 {
 529     BlockJob *bjob = container_of(job, BlockJob, job);
 530     GLOBAL_STATE_CODE();
 531     block_job_iostatus_reset(bjob);
 532 }
 533
 534 BlockErrorAction block_job_error_action(BlockJob *job, BlockdevOnError on_err,
 535                                         int is_read, int error)
 536 {
 537     BlockErrorAction action;
 538     IO_CODE();
 539
 540     switch (on_err) {
 541     case BLOCKDEV_ON_ERROR_ENOSPC:
 542     case BLOCKDEV_ON_ERROR_AUTO:
 543         action = (error == ENOSPC) ?
 544                  BLOCK_ERROR_ACTION_STOP : BLOCK_ERROR_ACTION_REPORT;
 545         break;
 546     case BLOCKDEV_ON_ERROR_STOP:
 547         action = BLOCK_ERROR_ACTION_STOP;
 548         break;
 549     case BLOCKDEV_ON_ERROR_REPORT:
 550         action = BLOCK_ERROR_ACTION_REPORT;
 551         break;
 552     case BLOCKDEV_ON_ERROR_IGNORE:
 553         action = BLOCK_ERROR_ACTION_IGNORE;
 554         break;
 555     default:
 556         abort();
 557     }
 558     if (!block_job_is_internal(job)) {
 559         qapi_event_send_block_job_error(job->job.id,
 560                                         is_read ? IO_OPERATION_TYPE_READ :
 561                                         IO_OPERATION_TYPE_WRITE,
 562                                         action);
 563     }
 564     if (action == BLOCK_ERROR_ACTION_STOP) {
 565         WITH_JOB_LOCK_GUARD() {
 566             if (!job->job.user_paused) {
 567                 job_pause_locked(&job->job);
 568                 /*
 569                  * make the pause user visible, which will be
 570                  * resumed from QMP.
 571                  */
 572                 job->job.user_paused = true;
 573             }
 574         }
 575         block_job_iostatus_set_err(job, error);
 576     }
 577     return action;
 578 }
 579
 580 AioContext *block_job_get_aio_context(BlockJob *job)
 581 {
 582     GLOBAL_STATE_CODE();
 583     return job->job.aio_context;
 584 }