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