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_set_aio_context(&job->job, ctx);
 166 }
 167
 168 static AioContext *child_job_get_parent_aio_context(BdrvChild *c)
 169 {
 170     BlockJob *job = c->opaque;
 171     GLOBAL_STATE_CODE();
 172
 173     return job->job.aio_context;
 174 }
 175
 176 static const BdrvChildClass child_job = {
 177     .get_parent_desc    = child_job_get_parent_desc,
 178     .drained_begin      = child_job_drained_begin,
 179     .drained_poll       = child_job_drained_poll,
 180     .drained_end        = child_job_drained_end,
 181     .can_set_aio_ctx    = child_job_can_set_aio_ctx,
 182     .set_aio_ctx        = child_job_set_aio_ctx,
 183     .stay_at_node       = true,
 184     .get_parent_aio_context = child_job_get_parent_aio_context,
 185 };
 186
 187 void block_job_remove_all_bdrv(BlockJob *job)
 188 {
 189     GLOBAL_STATE_CODE();
 190     /*
 191      * bdrv_root_unref_child() may reach child_job_[can_]set_aio_ctx(),
 192      * which will also traverse job->nodes, so consume the list one by
 193      * one to make sure that such a concurrent access does not attempt
 194      * to process an already freed BdrvChild.
 195      */
 196     while (job->nodes) {
 197         GSList *l = job->nodes;
 198         BdrvChild *c = l->data;
 199
 200         job->nodes = l->next;
 201
 202         bdrv_op_unblock_all(c->bs, job->blocker);
 203         bdrv_root_unref_child(c);
 204
 205         g_slist_free_1(l);
 206     }
 207 }
 208
 209 bool block_job_has_bdrv(BlockJob *job, BlockDriverState *bs)
 210 {
 211     GSList *el;
 212     GLOBAL_STATE_CODE();
 213
 214     for (el = job->nodes; el; el = el->next) {
 215         BdrvChild *c = el->data;
 216         if (c->bs == bs) {
 217             return true;
 218         }
 219     }
 220
 221     return false;
 222 }
 223
 224 int block_job_add_bdrv(BlockJob *job, const char *name, BlockDriverState *bs,
 225                        uint64_t perm, uint64_t shared_perm, Error **errp)
 226 {
 227     BdrvChild *c;
 228     bool need_context_ops;
 229     GLOBAL_STATE_CODE();
 230
 231     bdrv_ref(bs);
 232
 233     need_context_ops = bdrv_get_aio_context(bs) != job->job.aio_context;
 234
 235     if (need_context_ops && job->job.aio_context != qemu_get_aio_context()) {
 236         aio_context_release(job->job.aio_context);
 237     }
 238     c = bdrv_root_attach_child(bs, name, &child_job, 0, perm, shared_perm, job,
 239                                errp);
 240     if (need_context_ops && job->job.aio_context != qemu_get_aio_context()) {
 241         aio_context_acquire(job->job.aio_context);
 242     }
 243     if (c == NULL) {
 244         return -EPERM;
 245     }
 246
 247     job->nodes = g_slist_prepend(job->nodes, c);
 248     bdrv_op_block_all(bs, job->blocker);
 249
 250     return 0;
 251 }
 252
 253 /* Called with job_mutex lock held. */
 254 static void block_job_on_idle_locked(Notifier *n, void *opaque)
 255 {
 256     aio_wait_kick();
 257 }
 258
 259 bool block_job_is_internal(BlockJob *job)
 260 {
 261     return (job->job.id == NULL);
 262 }
 263
 264 const BlockJobDriver *block_job_driver(BlockJob *job)
 265 {
 266     return container_of(job->job.driver, BlockJobDriver, job_driver);
 267 }
 268
 269 /* Assumes the job_mutex is held */
 270 static bool job_timer_pending(Job *job)
 271 {
 272     return timer_pending(&job->sleep_timer);
 273 }
 274
 275 bool block_job_set_speed_locked(BlockJob *job, int64_t speed, Error **errp)
 276 {
 277     const BlockJobDriver *drv = block_job_driver(job);
 278     int64_t old_speed = job->speed;
 279
 280     GLOBAL_STATE_CODE();
 281
 282     if (job_apply_verb_locked(&job->job, JOB_VERB_SET_SPEED, errp) < 0) {
 283         return false;
 284     }
 285     if (speed < 0) {
 286         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "speed",
 287                    "a non-negative value");
 288         return false;
 289     }
 290
 291     ratelimit_set_speed(&job->limit, speed, BLOCK_JOB_SLICE_TIME);
 292
 293     job->speed = speed;
 294
 295     if (drv->set_speed) {
 296         job_unlock();
 297         drv->set_speed(job, speed);
 298         job_lock();
 299     }
 300
 301     if (speed && speed <= old_speed) {
 302         return true;
 303     }
 304
 305     /* kick only if a timer is pending */
 306     job_enter_cond_locked(&job->job, job_timer_pending);
 307
 308     return true;
 309 }
 310
 311 bool block_job_set_speed(BlockJob *job, int64_t speed, Error **errp)
 312 {
 313     JOB_LOCK_GUARD();
 314     return block_job_set_speed_locked(job, speed, errp);
 315 }
 316
 317 int64_t block_job_ratelimit_get_delay(BlockJob *job, uint64_t n)
 318 {
 319     IO_CODE();
 320     return ratelimit_calculate_delay(&job->limit, n);
 321 }
 322
 323 BlockJobInfo *block_job_query_locked(BlockJob *job, Error **errp)
 324 {
 325     BlockJobInfo *info;
 326     uint64_t progress_current, progress_total;
 327
 328     GLOBAL_STATE_CODE();
 329
 330     if (block_job_is_internal(job)) {
 331         error_setg(errp, "Cannot query QEMU internal jobs");
 332         return NULL;
 333     }
 334
 335     progress_get_snapshot(&job->job.progress, &progress_current,
 336                           &progress_total);
 337
 338     info = g_new0(BlockJobInfo, 1);
 339     info->type      = g_strdup(job_type_str(&job->job));
 340     info->device    = g_strdup(job->job.id);
 341     info->busy      = qatomic_read(&job->job.busy);
 342     info->paused    = job->job.pause_count > 0;
 343     info->offset    = progress_current;
 344     info->len       = progress_total;
 345     info->speed     = job->speed;
 346     info->io_status = job->iostatus;
 347     info->ready     = job_is_ready_locked(&job->job),
 348     info->status    = job->job.status;
 349     info->auto_finalize = job->job.auto_finalize;
 350     info->auto_dismiss  = job->job.auto_dismiss;
 351     if (job->job.ret) {
 352         info->has_error = true;
 353         info->error = job->job.err ?
 354                         g_strdup(error_get_pretty(job->job.err)) :
 355                         g_strdup(strerror(-job->job.ret));
 356     }
 357     return info;
 358 }
 359
 360 BlockJobInfo *block_job_query(BlockJob *job, Error **errp)
 361 {
 362     JOB_LOCK_GUARD();
 363     return block_job_query_locked(job, errp);
 364 }
 365
 366 /* Called with job lock held */
 367 static void block_job_iostatus_set_err_locked(BlockJob *job, int error)
 368 {
 369     if (job->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
 370         job->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE :
 371                                           BLOCK_DEVICE_IO_STATUS_FAILED;
 372     }
 373 }
 374
 375 /* Called with job_mutex lock held. */
 376 static void block_job_event_cancelled_locked(Notifier *n, void *opaque)
 377 {
 378     BlockJob *job = opaque;
 379     uint64_t progress_current, progress_total;
 380
 381     if (block_job_is_internal(job)) {
 382         return;
 383     }
 384
 385     progress_get_snapshot(&job->job.progress, &progress_current,
 386                           &progress_total);
 387
 388     qapi_event_send_block_job_cancelled(job_type(&job->job),
 389                                         job->job.id,
 390                                         progress_total,
 391                                         progress_current,
 392                                         job->speed);
 393 }
 394
 395 /* Called with job_mutex lock held. */
 396 static void block_job_event_completed_locked(Notifier *n, void *opaque)
 397 {
 398     BlockJob *job = opaque;
 399     const char *msg = NULL;
 400     uint64_t progress_current, progress_total;
 401
 402     if (block_job_is_internal(job)) {
 403         return;
 404     }
 405
 406     if (job->job.ret < 0) {
 407         msg = error_get_pretty(job->job.err);
 408     }
 409
 410     progress_get_snapshot(&job->job.progress, &progress_current,
 411                           &progress_total);
 412
 413     qapi_event_send_block_job_completed(job_type(&job->job),
 414                                         job->job.id,
 415                                         progress_total,
 416                                         progress_current,
 417                                         job->speed,
 418                                         !!msg,
 419                                         msg);
 420 }
 421
 422 /* Called with job_mutex lock held. */
 423 static void block_job_event_pending_locked(Notifier *n, void *opaque)
 424 {
 425     BlockJob *job = opaque;
 426
 427     if (block_job_is_internal(job)) {
 428         return;
 429     }
 430
 431     qapi_event_send_block_job_pending(job_type(&job->job),
 432                                       job->job.id);
 433 }
 434
 435 /* Called with job_mutex lock held. */
 436 static void block_job_event_ready_locked(Notifier *n, void *opaque)
 437 {
 438     BlockJob *job = opaque;
 439     uint64_t progress_current, progress_total;
 440
 441     if (block_job_is_internal(job)) {
 442         return;
 443     }
 444
 445     progress_get_snapshot(&job->job.progress, &progress_current,
 446                           &progress_total);
 447
 448     qapi_event_send_block_job_ready(job_type(&job->job),
 449                                     job->job.id,
 450                                     progress_total,
 451                                     progress_current,
 452                                     job->speed);
 453 }
 454
 455
 456 void *block_job_create(const char *job_id, const BlockJobDriver *driver,
 457                        JobTxn *txn, BlockDriverState *bs, uint64_t perm,
 458                        uint64_t shared_perm, int64_t speed, int flags,
 459                        BlockCompletionFunc *cb, void *opaque, Error **errp)
 460 {
 461     BlockJob *job;
 462     int ret;
 463     GLOBAL_STATE_CODE();
 464
 465     if (job_id == NULL && !(flags & JOB_INTERNAL)) {
 466         job_id = bdrv_get_device_name(bs);
 467     }
 468
 469     job = job_create(job_id, &driver->job_driver, txn, bdrv_get_aio_context(bs),
 470                      flags, cb, opaque, errp);
 471     if (job == NULL) {
 472         return NULL;
 473     }
 474
 475     assert(is_block_job(&job->job));
 476     assert(job->job.driver->free == &block_job_free);
 477     assert(job->job.driver->user_resume == &block_job_user_resume);
 478
 479     ratelimit_init(&job->limit);
 480
 481     job->finalize_cancelled_notifier.notify = block_job_event_cancelled_locked;
 482     job->finalize_completed_notifier.notify = block_job_event_completed_locked;
 483     job->pending_notifier.notify = block_job_event_pending_locked;
 484     job->ready_notifier.notify = block_job_event_ready_locked;
 485     job->idle_notifier.notify = block_job_on_idle_locked;
 486
 487     WITH_JOB_LOCK_GUARD() {
 488         notifier_list_add(&job->job.on_finalize_cancelled,
 489                           &job->finalize_cancelled_notifier);
 490         notifier_list_add(&job->job.on_finalize_completed,
 491                           &job->finalize_completed_notifier);
 492         notifier_list_add(&job->job.on_pending, &job->pending_notifier);
 493         notifier_list_add(&job->job.on_ready, &job->ready_notifier);
 494         notifier_list_add(&job->job.on_idle, &job->idle_notifier);
 495     }
 496
 497     error_setg(&job->blocker, "block device is in use by block job: %s",
 498                job_type_str(&job->job));
 499
 500     ret = block_job_add_bdrv(job, "main node", bs, perm, shared_perm, errp);
 501     if (ret < 0) {
 502         goto fail;
 503     }
 504
 505     bdrv_op_unblock(bs, BLOCK_OP_TYPE_DATAPLANE, job->blocker);
 506
 507     if (!block_job_set_speed(job, speed, errp)) {
 508         goto fail;
 509     }
 510
 511     return job;
 512
 513 fail:
 514     job_early_fail(&job->job);
 515     return NULL;
 516 }
 517
 518 void block_job_iostatus_reset_locked(BlockJob *job)
 519 {
 520     GLOBAL_STATE_CODE();
 521     if (job->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
 522         return;
 523     }
 524     assert(job->job.user_paused && job->job.pause_count > 0);
 525     job->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
 526 }
 527
 528 void block_job_iostatus_reset(BlockJob *job)
 529 {
 530     JOB_LOCK_GUARD();
 531     block_job_iostatus_reset_locked(job);
 532 }
 533
 534 void block_job_user_resume(Job *job)
 535 {
 536     BlockJob *bjob = container_of(job, BlockJob, job);
 537     GLOBAL_STATE_CODE();
 538     block_job_iostatus_reset(bjob);
 539 }
 540
 541 BlockErrorAction block_job_error_action(BlockJob *job, BlockdevOnError on_err,
 542                                         int is_read, int error)
 543 {
 544     BlockErrorAction action;
 545     IO_CODE();
 546
 547     switch (on_err) {
 548     case BLOCKDEV_ON_ERROR_ENOSPC:
 549     case BLOCKDEV_ON_ERROR_AUTO:
 550         action = (error == ENOSPC) ?
 551                  BLOCK_ERROR_ACTION_STOP : BLOCK_ERROR_ACTION_REPORT;
 552         break;
 553     case BLOCKDEV_ON_ERROR_STOP:
 554         action = BLOCK_ERROR_ACTION_STOP;
 555         break;
 556     case BLOCKDEV_ON_ERROR_REPORT:
 557         action = BLOCK_ERROR_ACTION_REPORT;
 558         break;
 559     case BLOCKDEV_ON_ERROR_IGNORE:
 560         action = BLOCK_ERROR_ACTION_IGNORE;
 561         break;
 562     default:
 563         abort();
 564     }
 565     if (!block_job_is_internal(job)) {
 566         qapi_event_send_block_job_error(job->job.id,
 567                                         is_read ? IO_OPERATION_TYPE_READ :
 568                                         IO_OPERATION_TYPE_WRITE,
 569                                         action);
 570     }
 571     if (action == BLOCK_ERROR_ACTION_STOP) {
 572         WITH_JOB_LOCK_GUARD() {
 573             if (!job->job.user_paused) {
 574                 job_pause_locked(&job->job);
 575                 /*
 576                  * make the pause user visible, which will be
 577                  * resumed from QMP.
 578                  */
 579                 job->job.user_paused = true;
 580             }
 581             block_job_iostatus_set_err_locked(job, error);
 582         }
 583     }
 584     return action;
 585 }
 586
 587 AioContext *block_job_get_aio_context(BlockJob *job)
 588 {
 589     GLOBAL_STATE_CODE();
 590     return job->job.aio_context;
 591 }