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 static void block_job_on_idle(Notifier *n, void *opaque)
 254 {
 255     aio_wait_kick();
 256 }
 257
 258 bool block_job_is_internal(BlockJob *job)
 259 {
 260     return (job->job.id == NULL);
 261 }
 262
 263 const BlockJobDriver *block_job_driver(BlockJob *job)
 264 {
 265     return container_of(job->job.driver, BlockJobDriver, job_driver);
 266 }
 267
 268 /* Assumes the job_mutex is held */
 269 static bool job_timer_pending(Job *job)
 270 {
 271     return timer_pending(&job->sleep_timer);
 272 }
 273
 274 bool block_job_set_speed_locked(BlockJob *job, int64_t speed, Error **errp)
 275 {
 276     const BlockJobDriver *drv = block_job_driver(job);
 277     int64_t old_speed = job->speed;
 278
 279     GLOBAL_STATE_CODE();
 280
 281     if (job_apply_verb_locked(&job->job, JOB_VERB_SET_SPEED, errp) < 0) {
 282         return false;
 283     }
 284     if (speed < 0) {
 285         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "speed",
 286                    "a non-negative value");
 287         return false;
 288     }
 289
 290     ratelimit_set_speed(&job->limit, speed, BLOCK_JOB_SLICE_TIME);
 291
 292     job->speed = speed;
 293
 294     if (drv->set_speed) {
 295         job_unlock();
 296         drv->set_speed(job, speed);
 297         job_lock();
 298     }
 299
 300     if (speed && speed <= old_speed) {
 301         return true;
 302     }
 303
 304     /* kick only if a timer is pending */
 305     job_enter_cond_locked(&job->job, job_timer_pending);
 306
 307     return true;
 308 }
 309
 310 bool block_job_set_speed(BlockJob *job, int64_t speed, Error **errp)
 311 {
 312     JOB_LOCK_GUARD();
 313     return block_job_set_speed_locked(job, speed, errp);
 314 }
 315
 316 int64_t block_job_ratelimit_get_delay(BlockJob *job, uint64_t n)
 317 {
 318     IO_CODE();
 319     return ratelimit_calculate_delay(&job->limit, n);
 320 }
 321
 322 BlockJobInfo *block_job_query_locked(BlockJob *job, Error **errp)
 323 {
 324     BlockJobInfo *info;
 325     uint64_t progress_current, progress_total;
 326
 327     GLOBAL_STATE_CODE();
 328
 329     if (block_job_is_internal(job)) {
 330         error_setg(errp, "Cannot query QEMU internal jobs");
 331         return NULL;
 332     }
 333
 334     progress_get_snapshot(&job->job.progress, &progress_current,
 335                           &progress_total);
 336
 337     info = g_new0(BlockJobInfo, 1);
 338     info->type      = g_strdup(job_type_str(&job->job));
 339     info->device    = g_strdup(job->job.id);
 340     info->busy      = qatomic_read(&job->job.busy);
 341     info->paused    = job->job.pause_count > 0;
 342     info->offset    = progress_current;
 343     info->len       = progress_total;
 344     info->speed     = job->speed;
 345     info->io_status = job->iostatus;
 346     info->ready     = job_is_ready_locked(&job->job),
 347     info->status    = job->job.status;
 348     info->auto_finalize = job->job.auto_finalize;
 349     info->auto_dismiss  = job->job.auto_dismiss;
 350     if (job->job.ret) {
 351         info->has_error = true;
 352         info->error = job->job.err ?
 353                         g_strdup(error_get_pretty(job->job.err)) :
 354                         g_strdup(strerror(-job->job.ret));
 355     }
 356     return info;
 357 }
 358
 359 BlockJobInfo *block_job_query(BlockJob *job, Error **errp)
 360 {
 361     JOB_LOCK_GUARD();
 362     return block_job_query_locked(job, errp);
 363 }
 364
 365 static void block_job_iostatus_set_err(BlockJob *job, int error)
 366 {
 367     if (job->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
 368         job->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE :
 369                                           BLOCK_DEVICE_IO_STATUS_FAILED;
 370     }
 371 }
 372
 373 static void block_job_event_cancelled(Notifier *n, void *opaque)
 374 {
 375     BlockJob *job = opaque;
 376     uint64_t progress_current, progress_total;
 377
 378     if (block_job_is_internal(job)) {
 379         return;
 380     }
 381
 382     progress_get_snapshot(&job->job.progress, &progress_current,
 383                           &progress_total);
 384
 385     qapi_event_send_block_job_cancelled(job_type(&job->job),
 386                                         job->job.id,
 387                                         progress_total,
 388                                         progress_current,
 389                                         job->speed);
 390 }
 391
 392 static void block_job_event_completed(Notifier *n, void *opaque)
 393 {
 394     BlockJob *job = opaque;
 395     const char *msg = NULL;
 396     uint64_t progress_current, progress_total;
 397
 398     if (block_job_is_internal(job)) {
 399         return;
 400     }
 401
 402     if (job->job.ret < 0) {
 403         msg = error_get_pretty(job->job.err);
 404     }
 405
 406     progress_get_snapshot(&job->job.progress, &progress_current,
 407                           &progress_total);
 408
 409     qapi_event_send_block_job_completed(job_type(&job->job),
 410                                         job->job.id,
 411                                         progress_total,
 412                                         progress_current,
 413                                         job->speed,
 414                                         !!msg,
 415                                         msg);
 416 }
 417
 418 static void block_job_event_pending(Notifier *n, void *opaque)
 419 {
 420     BlockJob *job = opaque;
 421
 422     if (block_job_is_internal(job)) {
 423         return;
 424     }
 425
 426     qapi_event_send_block_job_pending(job_type(&job->job),
 427                                       job->job.id);
 428 }
 429
 430 static void block_job_event_ready(Notifier *n, void *opaque)
 431 {
 432     BlockJob *job = opaque;
 433     uint64_t progress_current, progress_total;
 434
 435     if (block_job_is_internal(job)) {
 436         return;
 437     }
 438
 439     progress_get_snapshot(&job->job.progress, &progress_current,
 440                           &progress_total);
 441
 442     qapi_event_send_block_job_ready(job_type(&job->job),
 443                                     job->job.id,
 444                                     progress_total,
 445                                     progress_current,
 446                                     job->speed);
 447 }
 448
 449
 450 void *block_job_create(const char *job_id, const BlockJobDriver *driver,
 451                        JobTxn *txn, BlockDriverState *bs, uint64_t perm,
 452                        uint64_t shared_perm, int64_t speed, int flags,
 453                        BlockCompletionFunc *cb, void *opaque, Error **errp)
 454 {
 455     BlockJob *job;
 456     int ret;
 457     GLOBAL_STATE_CODE();
 458
 459     if (job_id == NULL && !(flags & JOB_INTERNAL)) {
 460         job_id = bdrv_get_device_name(bs);
 461     }
 462
 463     job = job_create(job_id, &driver->job_driver, txn, bdrv_get_aio_context(bs),
 464                      flags, cb, opaque, errp);
 465     if (job == NULL) {
 466         return NULL;
 467     }
 468
 469     assert(is_block_job(&job->job));
 470     assert(job->job.driver->free == &block_job_free);
 471     assert(job->job.driver->user_resume == &block_job_user_resume);
 472
 473     ratelimit_init(&job->limit);
 474
 475     job->finalize_cancelled_notifier.notify = block_job_event_cancelled;
 476     job->finalize_completed_notifier.notify = block_job_event_completed;
 477     job->pending_notifier.notify = block_job_event_pending;
 478     job->ready_notifier.notify = block_job_event_ready;
 479     job->idle_notifier.notify = block_job_on_idle;
 480
 481     WITH_JOB_LOCK_GUARD() {
 482         notifier_list_add(&job->job.on_finalize_cancelled,
 483                           &job->finalize_cancelled_notifier);
 484         notifier_list_add(&job->job.on_finalize_completed,
 485                           &job->finalize_completed_notifier);
 486         notifier_list_add(&job->job.on_pending, &job->pending_notifier);
 487         notifier_list_add(&job->job.on_ready, &job->ready_notifier);
 488         notifier_list_add(&job->job.on_idle, &job->idle_notifier);
 489     }
 490
 491     error_setg(&job->blocker, "block device is in use by block job: %s",
 492                job_type_str(&job->job));
 493
 494     ret = block_job_add_bdrv(job, "main node", bs, perm, shared_perm, errp);
 495     if (ret < 0) {
 496         goto fail;
 497     }
 498
 499     bdrv_op_unblock(bs, BLOCK_OP_TYPE_DATAPLANE, job->blocker);
 500
 501     if (!block_job_set_speed(job, speed, errp)) {
 502         goto fail;
 503     }
 504
 505     return job;
 506
 507 fail:
 508     job_early_fail(&job->job);
 509     return NULL;
 510 }
 511
 512 void block_job_iostatus_reset_locked(BlockJob *job)
 513 {
 514     GLOBAL_STATE_CODE();
 515     if (job->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
 516         return;
 517     }
 518     assert(job->job.user_paused && job->job.pause_count > 0);
 519     job->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
 520 }
 521
 522 void block_job_iostatus_reset(BlockJob *job)
 523 {
 524     JOB_LOCK_GUARD();
 525     block_job_iostatus_reset_locked(job);
 526 }
 527
 528 void block_job_user_resume(Job *job)
 529 {
 530     BlockJob *bjob = container_of(job, BlockJob, job);
 531     GLOBAL_STATE_CODE();
 532     block_job_iostatus_reset(bjob);
 533 }
 534
 535 BlockErrorAction block_job_error_action(BlockJob *job, BlockdevOnError on_err,
 536                                         int is_read, int error)
 537 {
 538     BlockErrorAction action;
 539     IO_CODE();
 540
 541     switch (on_err) {
 542     case BLOCKDEV_ON_ERROR_ENOSPC:
 543     case BLOCKDEV_ON_ERROR_AUTO:
 544         action = (error == ENOSPC) ?
 545                  BLOCK_ERROR_ACTION_STOP : BLOCK_ERROR_ACTION_REPORT;
 546         break;
 547     case BLOCKDEV_ON_ERROR_STOP:
 548         action = BLOCK_ERROR_ACTION_STOP;
 549         break;
 550     case BLOCKDEV_ON_ERROR_REPORT:
 551         action = BLOCK_ERROR_ACTION_REPORT;
 552         break;
 553     case BLOCKDEV_ON_ERROR_IGNORE:
 554         action = BLOCK_ERROR_ACTION_IGNORE;
 555         break;
 556     default:
 557         abort();
 558     }
 559     if (!block_job_is_internal(job)) {
 560         qapi_event_send_block_job_error(job->job.id,
 561                                         is_read ? IO_OPERATION_TYPE_READ :
 562                                         IO_OPERATION_TYPE_WRITE,
 563                                         action);
 564     }
 565     if (action == BLOCK_ERROR_ACTION_STOP) {
 566         WITH_JOB_LOCK_GUARD() {
 567             if (!job->job.user_paused) {
 568                 job_pause_locked(&job->job);
 569                 /*
 570                  * make the pause user visible, which will be
 571                  * resumed from QMP.
 572                  */
 573                 job->job.user_paused = true;
 574             }
 575         }
 576         block_job_iostatus_set_err(job, error);
 577     }
 578     return action;
 579 }
 580
 581 AioContext *block_job_get_aio_context(BlockJob *job)
 582 {
 583     GLOBAL_STATE_CODE();
 584     return job->job.aio_context;
 585 }