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