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/aio-wait.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/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_get_locked(const char *id)
  60 {
  61     Job *job = job_get_locked(id);
  62     GLOBAL_STATE_CODE();
  63
  64     if (job && is_block_job(job)) {
  65         return container_of(job, BlockJob, job);
  66     } else {
  67         return NULL;
  68     }
  69 }
  70
  71 BlockJob *block_job_get(const char *id)
  72 {
  73     JOB_LOCK_GUARD();
  74     return block_job_get_locked(id);
  75 }
  76
  77 void block_job_free(Job *job)
  78 {
  79     BlockJob *bjob = container_of(job, BlockJob, job);
  80     GLOBAL_STATE_CODE();
  81
  82     block_job_remove_all_bdrv(bjob);
  83     ratelimit_destroy(&bjob->limit);
  84     error_free(bjob->blocker);
  85 }
  86
  87 static char *child_job_get_parent_desc(BdrvChild *c)
  88 {
  89     BlockJob *job = c->opaque;
  90     return g_strdup_printf("%s job '%s'", job_type_str(&job->job), job->job.id);
  91 }
  92
  93 static void child_job_drained_begin(BdrvChild *c)
  94 {
  95     BlockJob *job = c->opaque;
  96     job_pause(&job->job);
  97 }
  98
  99 static bool child_job_drained_poll(BdrvChild *c)
 100 {
 101     BlockJob *bjob = c->opaque;
 102     Job *job = &bjob->job;
 103     const BlockJobDriver *drv = block_job_driver(bjob);
 104
 105     /* An inactive or completed job doesn't have any pending requests. Jobs
 106      * with !job->busy are either already paused or have a pause point after
 107      * being reentered, so no job driver code will run before they pause. */
 108     WITH_JOB_LOCK_GUARD() {
 109         if (!job->busy || job_is_completed_locked(job)) {
 110             return false;
 111         }
 112     }
 113
 114     /* Otherwise, assume that it isn't fully stopped yet, but allow the job to
 115      * override this assumption. */
 116     if (drv->drained_poll) {
 117         return drv->drained_poll(bjob);
 118     } else {
 119         return true;
 120     }
 121 }
 122
 123 static void child_job_drained_end(BdrvChild *c)
 124 {
 125     BlockJob *job = c->opaque;
 126     job_resume(&job->job);
 127 }
 128
 129 typedef struct BdrvStateChildJobContext {
 130     AioContext *new_ctx;
 131     BlockJob *job;
 132 } BdrvStateChildJobContext;
 133
 134 static void child_job_set_aio_ctx_commit(void *opaque)
 135 {
 136     BdrvStateChildJobContext *s = opaque;
 137     BlockJob *job = s->job;
 138
 139     job_set_aio_context(&job->job, s->new_ctx);
 140 }
 141
 142 static TransactionActionDrv change_child_job_context = {
 143     .commit = child_job_set_aio_ctx_commit,
 144     .clean = g_free,
 145 };
 146
 147 static bool child_job_change_aio_ctx(BdrvChild *c, AioContext *ctx,
 148                                      GHashTable *visited, Transaction *tran,
 149                                      Error **errp)
 150 {
 151     BlockJob *job = c->opaque;
 152     BdrvStateChildJobContext *s;
 153     GSList *l;
 154
 155     for (l = job->nodes; l; l = l->next) {
 156         BdrvChild *sibling = l->data;
 157         if (!bdrv_child_change_aio_context(sibling, ctx, visited,
 158                                            tran, errp)) {
 159             return false;
 160         }
 161     }
 162
 163     s = g_new(BdrvStateChildJobContext, 1);
 164     *s = (BdrvStateChildJobContext) {
 165         .new_ctx = ctx,
 166         .job = job,
 167     };
 168
 169     tran_add(tran, &change_child_job_context, s);
 170     return true;
 171 }
 172
 173 static AioContext *child_job_get_parent_aio_context(BdrvChild *c)
 174 {
 175     BlockJob *job = c->opaque;
 176     IO_CODE();
 177     JOB_LOCK_GUARD();
 178
 179     return job->job.aio_context;
 180 }
 181
 182 static const BdrvChildClass child_job = {
 183     .get_parent_desc    = child_job_get_parent_desc,
 184     .drained_begin      = child_job_drained_begin,
 185     .drained_poll       = child_job_drained_poll,
 186     .drained_end        = child_job_drained_end,
 187     .change_aio_ctx     = child_job_change_aio_ctx,
 188     .stay_at_node       = true,
 189     .get_parent_aio_context = child_job_get_parent_aio_context,
 190 };
 191
 192 void block_job_remove_all_bdrv(BlockJob *job)
 193 {
 194     GLOBAL_STATE_CODE();
 195     /*
 196      * bdrv_root_unref_child() may reach child_job_[can_]set_aio_ctx(),
 197      * which will also traverse job->nodes, so consume the list one by
 198      * one to make sure that such a concurrent access does not attempt
 199      * to process an already freed BdrvChild.
 200      */
 201     bdrv_graph_wrlock(NULL);
 202     while (job->nodes) {
 203         GSList *l = job->nodes;
 204         BdrvChild *c = l->data;
 205
 206         job->nodes = l->next;
 207
 208         bdrv_op_unblock_all(c->bs, job->blocker);
 209         bdrv_root_unref_child(c);
 210
 211         g_slist_free_1(l);
 212     }
 213     bdrv_graph_wrunlock();
 214 }
 215
 216 bool block_job_has_bdrv(BlockJob *job, BlockDriverState *bs)
 217 {
 218     GSList *el;
 219     GLOBAL_STATE_CODE();
 220
 221     for (el = job->nodes; el; el = el->next) {
 222         BdrvChild *c = el->data;
 223         if (c->bs == bs) {
 224             return true;
 225         }
 226     }
 227
 228     return false;
 229 }
 230
 231 int block_job_add_bdrv(BlockJob *job, const char *name, BlockDriverState *bs,
 232                        uint64_t perm, uint64_t shared_perm, Error **errp)
 233 {
 234     BdrvChild *c;
 235     AioContext *ctx = bdrv_get_aio_context(bs);
 236     bool need_context_ops;
 237     GLOBAL_STATE_CODE();
 238
 239     bdrv_ref(bs);
 240
 241     need_context_ops = ctx != job->job.aio_context;
 242
 243     if (need_context_ops) {
 244         if (job->job.aio_context != qemu_get_aio_context()) {
 245             aio_context_release(job->job.aio_context);
 246         }
 247         aio_context_acquire(ctx);
 248     }
 249     c = bdrv_root_attach_child(bs, name, &child_job, 0, perm, shared_perm, job,
 250                                errp);
 251     if (need_context_ops) {
 252         aio_context_release(ctx);
 253         if (job->job.aio_context != qemu_get_aio_context()) {
 254             aio_context_acquire(job->job.aio_context);
 255         }
 256     }
 257     if (c == NULL) {
 258         return -EPERM;
 259     }
 260
 261     job->nodes = g_slist_prepend(job->nodes, c);
 262     bdrv_op_block_all(bs, job->blocker);
 263
 264     return 0;
 265 }
 266
 267 /* Called with job_mutex lock held. */
 268 static void block_job_on_idle_locked(Notifier *n, void *opaque)
 269 {
 270     aio_wait_kick();
 271 }
 272
 273 bool block_job_is_internal(BlockJob *job)
 274 {
 275     return (job->job.id == NULL);
 276 }
 277
 278 const BlockJobDriver *block_job_driver(BlockJob *job)
 279 {
 280     return container_of(job->job.driver, BlockJobDriver, job_driver);
 281 }
 282
 283 /* Assumes the job_mutex is held */
 284 static bool job_timer_pending(Job *job)
 285 {
 286     return timer_pending(&job->sleep_timer);
 287 }
 288
 289 bool block_job_set_speed_locked(BlockJob *job, int64_t speed, Error **errp)
 290 {
 291     const BlockJobDriver *drv = block_job_driver(job);
 292     int64_t old_speed = job->speed;
 293
 294     GLOBAL_STATE_CODE();
 295
 296     if (job_apply_verb_locked(&job->job, JOB_VERB_SET_SPEED, errp) < 0) {
 297         return false;
 298     }
 299     if (speed < 0) {
 300         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "speed",
 301                    "a non-negative value");
 302         return false;
 303     }
 304
 305     ratelimit_set_speed(&job->limit, speed, BLOCK_JOB_SLICE_TIME);
 306
 307     job->speed = speed;
 308
 309     if (drv->set_speed) {
 310         job_unlock();
 311         drv->set_speed(job, speed);
 312         job_lock();
 313     }
 314
 315     if (speed && speed <= old_speed) {
 316         return true;
 317     }
 318
 319     /* kick only if a timer is pending */
 320     job_enter_cond_locked(&job->job, job_timer_pending);
 321
 322     return true;
 323 }
 324
 325 static bool block_job_set_speed(BlockJob *job, int64_t speed, Error **errp)
 326 {
 327     JOB_LOCK_GUARD();
 328     return block_job_set_speed_locked(job, speed, errp);
 329 }
 330
 331 void block_job_ratelimit_processed_bytes(BlockJob *job, uint64_t n)
 332 {
 333     IO_CODE();
 334     ratelimit_calculate_delay(&job->limit, n);
 335 }
 336
 337 void block_job_ratelimit_sleep(BlockJob *job)
 338 {
 339     uint64_t delay_ns;
 340
 341     /*
 342      * Sleep at least once. If the job is reentered early, keep waiting until
 343      * we've waited for the full time that is necessary to keep the job at the
 344      * right speed.
 345      *
 346      * Make sure to recalculate the delay after each (possibly interrupted)
 347      * sleep because the speed can change while the job has yielded.
 348      */
 349     do {
 350         delay_ns = ratelimit_calculate_delay(&job->limit, 0);
 351         job_sleep_ns(&job->job, delay_ns);
 352     } while (delay_ns && !job_is_cancelled(&job->job));
 353 }
 354
 355 BlockJobInfo *block_job_query_locked(BlockJob *job, Error **errp)
 356 {
 357     BlockJobInfo *info;
 358     uint64_t progress_current, progress_total;
 359
 360     GLOBAL_STATE_CODE();
 361
 362     if (block_job_is_internal(job)) {
 363         error_setg(errp, "Cannot query QEMU internal jobs");
 364         return NULL;
 365     }
 366
 367     progress_get_snapshot(&job->job.progress, &progress_current,
 368                           &progress_total);
 369
 370     info = g_new0(BlockJobInfo, 1);
 371     info->type      = g_strdup(job_type_str(&job->job));
 372     info->device    = g_strdup(job->job.id);
 373     info->busy      = job->job.busy;
 374     info->paused    = job->job.pause_count > 0;
 375     info->offset    = progress_current;
 376     info->len       = progress_total;
 377     info->speed     = job->speed;
 378     info->io_status = job->iostatus;
 379     info->ready     = job_is_ready_locked(&job->job),
 380     info->status    = job->job.status;
 381     info->auto_finalize = job->job.auto_finalize;
 382     info->auto_dismiss  = job->job.auto_dismiss;
 383     if (job->job.ret) {
 384         info->error = job->job.err ?
 385                         g_strdup(error_get_pretty(job->job.err)) :
 386                         g_strdup(strerror(-job->job.ret));
 387     }
 388     return info;
 389 }
 390
 391 /* Called with job lock held */
 392 static void block_job_iostatus_set_err_locked(BlockJob *job, int error)
 393 {
 394     if (job->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
 395         job->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE :
 396                                           BLOCK_DEVICE_IO_STATUS_FAILED;
 397     }
 398 }
 399
 400 /* Called with job_mutex lock held. */
 401 static void block_job_event_cancelled_locked(Notifier *n, void *opaque)
 402 {
 403     BlockJob *job = opaque;
 404     uint64_t progress_current, progress_total;
 405
 406     if (block_job_is_internal(job)) {
 407         return;
 408     }
 409
 410     progress_get_snapshot(&job->job.progress, &progress_current,
 411                           &progress_total);
 412
 413     qapi_event_send_block_job_cancelled(job_type(&job->job),
 414                                         job->job.id,
 415                                         progress_total,
 416                                         progress_current,
 417                                         job->speed);
 418 }
 419
 420 /* Called with job_mutex lock held. */
 421 static void block_job_event_completed_locked(Notifier *n, void *opaque)
 422 {
 423     BlockJob *job = opaque;
 424     const char *msg = NULL;
 425     uint64_t progress_current, progress_total;
 426
 427     if (block_job_is_internal(job)) {
 428         return;
 429     }
 430
 431     if (job->job.ret < 0) {
 432         msg = error_get_pretty(job->job.err);
 433     }
 434
 435     progress_get_snapshot(&job->job.progress, &progress_current,
 436                           &progress_total);
 437
 438     qapi_event_send_block_job_completed(job_type(&job->job),
 439                                         job->job.id,
 440                                         progress_total,
 441                                         progress_current,
 442                                         job->speed,
 443                                         msg);
 444 }
 445
 446 /* Called with job_mutex lock held. */
 447 static void block_job_event_pending_locked(Notifier *n, void *opaque)
 448 {
 449     BlockJob *job = opaque;
 450
 451     if (block_job_is_internal(job)) {
 452         return;
 453     }
 454
 455     qapi_event_send_block_job_pending(job_type(&job->job),
 456                                       job->job.id);
 457 }
 458
 459 /* Called with job_mutex lock held. */
 460 static void block_job_event_ready_locked(Notifier *n, void *opaque)
 461 {
 462     BlockJob *job = opaque;
 463     uint64_t progress_current, progress_total;
 464
 465     if (block_job_is_internal(job)) {
 466         return;
 467     }
 468
 469     progress_get_snapshot(&job->job.progress, &progress_current,
 470                           &progress_total);
 471
 472     qapi_event_send_block_job_ready(job_type(&job->job),
 473                                     job->job.id,
 474                                     progress_total,
 475                                     progress_current,
 476                                     job->speed);
 477 }
 478
 479
 480 void *block_job_create(const char *job_id, const BlockJobDriver *driver,
 481                        JobTxn *txn, BlockDriverState *bs, uint64_t perm,
 482                        uint64_t shared_perm, int64_t speed, int flags,
 483                        BlockCompletionFunc *cb, void *opaque, Error **errp)
 484 {
 485     BlockJob *job;
 486     int ret;
 487     GLOBAL_STATE_CODE();
 488     GRAPH_RDLOCK_GUARD_MAINLOOP();
 489
 490     if (job_id == NULL && !(flags & JOB_INTERNAL)) {
 491         job_id = bdrv_get_device_name(bs);
 492     }
 493
 494     job = job_create(job_id, &driver->job_driver, txn, bdrv_get_aio_context(bs),
 495                      flags, cb, opaque, errp);
 496     if (job == NULL) {
 497         return NULL;
 498     }
 499
 500     assert(is_block_job(&job->job));
 501     assert(job->job.driver->free == &block_job_free);
 502     assert(job->job.driver->user_resume == &block_job_user_resume);
 503
 504     ratelimit_init(&job->limit);
 505
 506     job->finalize_cancelled_notifier.notify = block_job_event_cancelled_locked;
 507     job->finalize_completed_notifier.notify = block_job_event_completed_locked;
 508     job->pending_notifier.notify = block_job_event_pending_locked;
 509     job->ready_notifier.notify = block_job_event_ready_locked;
 510     job->idle_notifier.notify = block_job_on_idle_locked;
 511
 512     WITH_JOB_LOCK_GUARD() {
 513         notifier_list_add(&job->job.on_finalize_cancelled,
 514                           &job->finalize_cancelled_notifier);
 515         notifier_list_add(&job->job.on_finalize_completed,
 516                           &job->finalize_completed_notifier);
 517         notifier_list_add(&job->job.on_pending, &job->pending_notifier);
 518         notifier_list_add(&job->job.on_ready, &job->ready_notifier);
 519         notifier_list_add(&job->job.on_idle, &job->idle_notifier);
 520     }
 521
 522     error_setg(&job->blocker, "block device is in use by block job: %s",
 523                job_type_str(&job->job));
 524
 525     ret = block_job_add_bdrv(job, "main node", bs, perm, shared_perm, errp);
 526     if (ret < 0) {
 527         goto fail;
 528     }
 529
 530     bdrv_op_unblock(bs, BLOCK_OP_TYPE_DATAPLANE, job->blocker);
 531
 532     if (!block_job_set_speed(job, speed, errp)) {
 533         goto fail;
 534     }
 535
 536     return job;
 537
 538 fail:
 539     job_early_fail(&job->job);
 540     return NULL;
 541 }
 542
 543 void block_job_iostatus_reset_locked(BlockJob *job)
 544 {
 545     GLOBAL_STATE_CODE();
 546     if (job->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
 547         return;
 548     }
 549     assert(job->job.user_paused && job->job.pause_count > 0);
 550     job->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
 551 }
 552
 553 static void block_job_iostatus_reset(BlockJob *job)
 554 {
 555     JOB_LOCK_GUARD();
 556     block_job_iostatus_reset_locked(job);
 557 }
 558
 559 void block_job_user_resume(Job *job)
 560 {
 561     BlockJob *bjob = container_of(job, BlockJob, job);
 562     GLOBAL_STATE_CODE();
 563     block_job_iostatus_reset(bjob);
 564 }
 565
 566 BlockErrorAction block_job_error_action(BlockJob *job, BlockdevOnError on_err,
 567                                         int is_read, int error)
 568 {
 569     BlockErrorAction action;
 570     IO_CODE();
 571
 572     switch (on_err) {
 573     case BLOCKDEV_ON_ERROR_ENOSPC:
 574     case BLOCKDEV_ON_ERROR_AUTO:
 575         action = (error == ENOSPC) ?
 576                  BLOCK_ERROR_ACTION_STOP : BLOCK_ERROR_ACTION_REPORT;
 577         break;
 578     case BLOCKDEV_ON_ERROR_STOP:
 579         action = BLOCK_ERROR_ACTION_STOP;
 580         break;
 581     case BLOCKDEV_ON_ERROR_REPORT:
 582         action = BLOCK_ERROR_ACTION_REPORT;
 583         break;
 584     case BLOCKDEV_ON_ERROR_IGNORE:
 585         action = BLOCK_ERROR_ACTION_IGNORE;
 586         break;
 587     default:
 588         abort();
 589     }
 590     if (!block_job_is_internal(job)) {
 591         qapi_event_send_block_job_error(job->job.id,
 592                                         is_read ? IO_OPERATION_TYPE_READ :
 593                                         IO_OPERATION_TYPE_WRITE,
 594                                         action);
 595     }
 596     if (action == BLOCK_ERROR_ACTION_STOP) {
 597         WITH_JOB_LOCK_GUARD() {
 598             if (!job->job.user_paused) {
 599                 job_pause_locked(&job->job);
 600                 /*
 601                  * make the pause user visible, which will be
 602                  * resumed from QMP.
 603                  */
 604                 job->job.user_paused = true;
 605             }
 606             block_job_iostatus_set_err_locked(job, error);
 607         }
 608     }
 609     return action;
 610 }
 611
 612 AioContext *block_job_get_aio_context(BlockJob *job)
 613 {
 614     GLOBAL_STATE_CODE();
 615     return job->job.aio_context;
 616 }