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 static bool child_job_can_set_aio_ctx(BdrvChild *c, AioContext *ctx,
 130                                       GSList **ignore, Error **errp)
 131 {
 132     BlockJob *job = c->opaque;
 133     GSList *l;
 134
 135     for (l = job->nodes; l; l = l->next) {
 136         BdrvChild *sibling = l->data;
 137         if (!bdrv_child_can_set_aio_context(sibling, ctx, ignore, errp)) {
 138             return false;
 139         }
 140     }
 141     return true;
 142 }
 143
 144 static void child_job_set_aio_ctx(BdrvChild *c, AioContext *ctx,
 145                                   GSList **ignore)
 146 {
 147     BlockJob *job = c->opaque;
 148     GSList *l;
 149
 150     for (l = job->nodes; l; l = l->next) {
 151         BdrvChild *sibling = l->data;
 152         if (g_slist_find(*ignore, sibling)) {
 153             continue;
 154         }
 155         *ignore = g_slist_prepend(*ignore, sibling);
 156         bdrv_set_aio_context_ignore(sibling->bs, ctx, ignore);
 157     }
 158
 159     job_set_aio_context(&job->job, ctx);
 160 }
 161
 162 static AioContext *child_job_get_parent_aio_context(BdrvChild *c)
 163 {
 164     BlockJob *job = c->opaque;
 165     GLOBAL_STATE_CODE();
 166
 167     return job->job.aio_context;
 168 }
 169
 170 static const BdrvChildClass child_job = {
 171     .get_parent_desc    = child_job_get_parent_desc,
 172     .drained_begin      = child_job_drained_begin,
 173     .drained_poll       = child_job_drained_poll,
 174     .drained_end        = child_job_drained_end,
 175     .can_set_aio_ctx    = child_job_can_set_aio_ctx,
 176     .set_aio_ctx        = child_job_set_aio_ctx,
 177     .stay_at_node       = true,
 178     .get_parent_aio_context = child_job_get_parent_aio_context,
 179 };
 180
 181 void block_job_remove_all_bdrv(BlockJob *job)
 182 {
 183     GLOBAL_STATE_CODE();
 184     /*
 185      * bdrv_root_unref_child() may reach child_job_[can_]set_aio_ctx(),
 186      * which will also traverse job->nodes, so consume the list one by
 187      * one to make sure that such a concurrent access does not attempt
 188      * to process an already freed BdrvChild.
 189      */
 190     while (job->nodes) {
 191         GSList *l = job->nodes;
 192         BdrvChild *c = l->data;
 193
 194         job->nodes = l->next;
 195
 196         bdrv_op_unblock_all(c->bs, job->blocker);
 197         bdrv_root_unref_child(c);
 198
 199         g_slist_free_1(l);
 200     }
 201 }
 202
 203 bool block_job_has_bdrv(BlockJob *job, BlockDriverState *bs)
 204 {
 205     GSList *el;
 206     GLOBAL_STATE_CODE();
 207
 208     for (el = job->nodes; el; el = el->next) {
 209         BdrvChild *c = el->data;
 210         if (c->bs == bs) {
 211             return true;
 212         }
 213     }
 214
 215     return false;
 216 }
 217
 218 int block_job_add_bdrv(BlockJob *job, const char *name, BlockDriverState *bs,
 219                        uint64_t perm, uint64_t shared_perm, Error **errp)
 220 {
 221     BdrvChild *c;
 222     bool need_context_ops;
 223     GLOBAL_STATE_CODE();
 224
 225     bdrv_ref(bs);
 226
 227     need_context_ops = bdrv_get_aio_context(bs) != job->job.aio_context;
 228
 229     if (need_context_ops && job->job.aio_context != qemu_get_aio_context()) {
 230         aio_context_release(job->job.aio_context);
 231     }
 232     c = bdrv_root_attach_child(bs, name, &child_job, 0, perm, shared_perm, job,
 233                                errp);
 234     if (need_context_ops && job->job.aio_context != qemu_get_aio_context()) {
 235         aio_context_acquire(job->job.aio_context);
 236     }
 237     if (c == NULL) {
 238         return -EPERM;
 239     }
 240
 241     job->nodes = g_slist_prepend(job->nodes, c);
 242     bdrv_op_block_all(bs, job->blocker);
 243
 244     return 0;
 245 }
 246
 247 /* Called with job_mutex lock held. */
 248 static void block_job_on_idle_locked(Notifier *n, void *opaque)
 249 {
 250     aio_wait_kick();
 251 }
 252
 253 bool block_job_is_internal(BlockJob *job)
 254 {
 255     return (job->job.id == NULL);
 256 }
 257
 258 const BlockJobDriver *block_job_driver(BlockJob *job)
 259 {
 260     return container_of(job->job.driver, BlockJobDriver, job_driver);
 261 }
 262
 263 /* Assumes the job_mutex is held */
 264 static bool job_timer_pending(Job *job)
 265 {
 266     return timer_pending(&job->sleep_timer);
 267 }
 268
 269 bool block_job_set_speed_locked(BlockJob *job, int64_t speed, Error **errp)
 270 {
 271     const BlockJobDriver *drv = block_job_driver(job);
 272     int64_t old_speed = job->speed;
 273
 274     GLOBAL_STATE_CODE();
 275
 276     if (job_apply_verb_locked(&job->job, JOB_VERB_SET_SPEED, errp) < 0) {
 277         return false;
 278     }
 279     if (speed < 0) {
 280         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "speed",
 281                    "a non-negative value");
 282         return false;
 283     }
 284
 285     ratelimit_set_speed(&job->limit, speed, BLOCK_JOB_SLICE_TIME);
 286
 287     job->speed = speed;
 288
 289     if (drv->set_speed) {
 290         job_unlock();
 291         drv->set_speed(job, speed);
 292         job_lock();
 293     }
 294
 295     if (speed && speed <= old_speed) {
 296         return true;
 297     }
 298
 299     /* kick only if a timer is pending */
 300     job_enter_cond_locked(&job->job, job_timer_pending);
 301
 302     return true;
 303 }
 304
 305 static bool block_job_set_speed(BlockJob *job, int64_t speed, Error **errp)
 306 {
 307     JOB_LOCK_GUARD();
 308     return block_job_set_speed_locked(job, speed, errp);
 309 }
 310
 311 int64_t block_job_ratelimit_get_delay(BlockJob *job, uint64_t n)
 312 {
 313     IO_CODE();
 314     return ratelimit_calculate_delay(&job->limit, n);
 315 }
 316
 317 BlockJobInfo *block_job_query_locked(BlockJob *job, Error **errp)
 318 {
 319     BlockJobInfo *info;
 320     uint64_t progress_current, progress_total;
 321
 322     GLOBAL_STATE_CODE();
 323
 324     if (block_job_is_internal(job)) {
 325         error_setg(errp, "Cannot query QEMU internal jobs");
 326         return NULL;
 327     }
 328
 329     progress_get_snapshot(&job->job.progress, &progress_current,
 330                           &progress_total);
 331
 332     info = g_new0(BlockJobInfo, 1);
 333     info->type      = g_strdup(job_type_str(&job->job));
 334     info->device    = g_strdup(job->job.id);
 335     info->busy      = job->job.busy;
 336     info->paused    = job->job.pause_count > 0;
 337     info->offset    = progress_current;
 338     info->len       = progress_total;
 339     info->speed     = job->speed;
 340     info->io_status = job->iostatus;
 341     info->ready     = job_is_ready_locked(&job->job),
 342     info->status    = job->job.status;
 343     info->auto_finalize = job->job.auto_finalize;
 344     info->auto_dismiss  = job->job.auto_dismiss;
 345     if (job->job.ret) {
 346         info->has_error = true;
 347         info->error = job->job.err ?
 348                         g_strdup(error_get_pretty(job->job.err)) :
 349                         g_strdup(strerror(-job->job.ret));
 350     }
 351     return info;
 352 }
 353
 354 /* Called with job lock held */
 355 static void block_job_iostatus_set_err_locked(BlockJob *job, int error)
 356 {
 357     if (job->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
 358         job->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE :
 359                                           BLOCK_DEVICE_IO_STATUS_FAILED;
 360     }
 361 }
 362
 363 /* Called with job_mutex lock held. */
 364 static void block_job_event_cancelled_locked(Notifier *n, void *opaque)
 365 {
 366     BlockJob *job = opaque;
 367     uint64_t progress_current, progress_total;
 368
 369     if (block_job_is_internal(job)) {
 370         return;
 371     }
 372
 373     progress_get_snapshot(&job->job.progress, &progress_current,
 374                           &progress_total);
 375
 376     qapi_event_send_block_job_cancelled(job_type(&job->job),
 377                                         job->job.id,
 378                                         progress_total,
 379                                         progress_current,
 380                                         job->speed);
 381 }
 382
 383 /* Called with job_mutex lock held. */
 384 static void block_job_event_completed_locked(Notifier *n, void *opaque)
 385 {
 386     BlockJob *job = opaque;
 387     const char *msg = NULL;
 388     uint64_t progress_current, progress_total;
 389
 390     if (block_job_is_internal(job)) {
 391         return;
 392     }
 393
 394     if (job->job.ret < 0) {
 395         msg = error_get_pretty(job->job.err);
 396     }
 397
 398     progress_get_snapshot(&job->job.progress, &progress_current,
 399                           &progress_total);
 400
 401     qapi_event_send_block_job_completed(job_type(&job->job),
 402                                         job->job.id,
 403                                         progress_total,
 404                                         progress_current,
 405                                         job->speed,
 406                                         !!msg,
 407                                         msg);
 408 }
 409
 410 /* Called with job_mutex lock held. */
 411 static void block_job_event_pending_locked(Notifier *n, void *opaque)
 412 {
 413     BlockJob *job = opaque;
 414
 415     if (block_job_is_internal(job)) {
 416         return;
 417     }
 418
 419     qapi_event_send_block_job_pending(job_type(&job->job),
 420                                       job->job.id);
 421 }
 422
 423 /* Called with job_mutex lock held. */
 424 static void block_job_event_ready_locked(Notifier *n, void *opaque)
 425 {
 426     BlockJob *job = opaque;
 427     uint64_t progress_current, progress_total;
 428
 429     if (block_job_is_internal(job)) {
 430         return;
 431     }
 432
 433     progress_get_snapshot(&job->job.progress, &progress_current,
 434                           &progress_total);
 435
 436     qapi_event_send_block_job_ready(job_type(&job->job),
 437                                     job->job.id,
 438                                     progress_total,
 439                                     progress_current,
 440                                     job->speed);
 441 }
 442
 443
 444 void *block_job_create(const char *job_id, const BlockJobDriver *driver,
 445                        JobTxn *txn, BlockDriverState *bs, uint64_t perm,
 446                        uint64_t shared_perm, int64_t speed, int flags,
 447                        BlockCompletionFunc *cb, void *opaque, Error **errp)
 448 {
 449     BlockJob *job;
 450     int ret;
 451     GLOBAL_STATE_CODE();
 452
 453     if (job_id == NULL && !(flags & JOB_INTERNAL)) {
 454         job_id = bdrv_get_device_name(bs);
 455     }
 456
 457     job = job_create(job_id, &driver->job_driver, txn, bdrv_get_aio_context(bs),
 458                      flags, cb, opaque, errp);
 459     if (job == NULL) {
 460         return NULL;
 461     }
 462
 463     assert(is_block_job(&job->job));
 464     assert(job->job.driver->free == &block_job_free);
 465     assert(job->job.driver->user_resume == &block_job_user_resume);
 466
 467     ratelimit_init(&job->limit);
 468
 469     job->finalize_cancelled_notifier.notify = block_job_event_cancelled_locked;
 470     job->finalize_completed_notifier.notify = block_job_event_completed_locked;
 471     job->pending_notifier.notify = block_job_event_pending_locked;
 472     job->ready_notifier.notify = block_job_event_ready_locked;
 473     job->idle_notifier.notify = block_job_on_idle_locked;
 474
 475     WITH_JOB_LOCK_GUARD() {
 476         notifier_list_add(&job->job.on_finalize_cancelled,
 477                           &job->finalize_cancelled_notifier);
 478         notifier_list_add(&job->job.on_finalize_completed,
 479                           &job->finalize_completed_notifier);
 480         notifier_list_add(&job->job.on_pending, &job->pending_notifier);
 481         notifier_list_add(&job->job.on_ready, &job->ready_notifier);
 482         notifier_list_add(&job->job.on_idle, &job->idle_notifier);
 483     }
 484
 485     error_setg(&job->blocker, "block device is in use by block job: %s",
 486                job_type_str(&job->job));
 487
 488     ret = block_job_add_bdrv(job, "main node", bs, perm, shared_perm, errp);
 489     if (ret < 0) {
 490         goto fail;
 491     }
 492
 493     bdrv_op_unblock(bs, BLOCK_OP_TYPE_DATAPLANE, job->blocker);
 494
 495     if (!block_job_set_speed(job, speed, errp)) {
 496         goto fail;
 497     }
 498
 499     return job;
 500
 501 fail:
 502     job_early_fail(&job->job);
 503     return NULL;
 504 }
 505
 506 void block_job_iostatus_reset_locked(BlockJob *job)
 507 {
 508     GLOBAL_STATE_CODE();
 509     if (job->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
 510         return;
 511     }
 512     assert(job->job.user_paused && job->job.pause_count > 0);
 513     job->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
 514 }
 515
 516 static void block_job_iostatus_reset(BlockJob *job)
 517 {
 518     JOB_LOCK_GUARD();
 519     block_job_iostatus_reset_locked(job);
 520 }
 521
 522 void block_job_user_resume(Job *job)
 523 {
 524     BlockJob *bjob = container_of(job, BlockJob, job);
 525     GLOBAL_STATE_CODE();
 526     block_job_iostatus_reset(bjob);
 527 }
 528
 529 BlockErrorAction block_job_error_action(BlockJob *job, BlockdevOnError on_err,
 530                                         int is_read, int error)
 531 {
 532     BlockErrorAction action;
 533     IO_CODE();
 534
 535     switch (on_err) {
 536     case BLOCKDEV_ON_ERROR_ENOSPC:
 537     case BLOCKDEV_ON_ERROR_AUTO:
 538         action = (error == ENOSPC) ?
 539                  BLOCK_ERROR_ACTION_STOP : BLOCK_ERROR_ACTION_REPORT;
 540         break;
 541     case BLOCKDEV_ON_ERROR_STOP:
 542         action = BLOCK_ERROR_ACTION_STOP;
 543         break;
 544     case BLOCKDEV_ON_ERROR_REPORT:
 545         action = BLOCK_ERROR_ACTION_REPORT;
 546         break;
 547     case BLOCKDEV_ON_ERROR_IGNORE:
 548         action = BLOCK_ERROR_ACTION_IGNORE;
 549         break;
 550     default:
 551         abort();
 552     }
 553     if (!block_job_is_internal(job)) {
 554         qapi_event_send_block_job_error(job->job.id,
 555                                         is_read ? IO_OPERATION_TYPE_READ :
 556                                         IO_OPERATION_TYPE_WRITE,
 557                                         action);
 558     }
 559     if (action == BLOCK_ERROR_ACTION_STOP) {
 560         WITH_JOB_LOCK_GUARD() {
 561             if (!job->job.user_paused) {
 562                 job_pause_locked(&job->job);
 563                 /*
 564                  * make the pause user visible, which will be
 565                  * resumed from QMP.
 566                  */
 567                 job->job.user_paused = true;
 568             }
 569             block_job_iostatus_set_err_locked(job, error);
 570         }
 571     }
 572     return action;
 573 }
 574
 575 AioContext *block_job_get_aio_context(BlockJob *job)
 576 {
 577     GLOBAL_STATE_CODE();
 578     return job->job.aio_context;
 579 }