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(BlockJob *bjob)
  48 {
  49     Job *job = bjob ? &bjob->job : NULL;
  50     GLOBAL_STATE_CODE();
  51
  52     do {
  53         job = job_next(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(const char *id)
  60 {
  61     Job *job = job_get(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 void block_job_free(Job *job)
  72 {
  73     BlockJob *bjob = container_of(job, BlockJob, job);
  74     GLOBAL_STATE_CODE();
  75
  76     block_job_remove_all_bdrv(bjob);
  77     ratelimit_destroy(&bjob->limit);
  78     error_free(bjob->blocker);
  79 }
  80
  81 static char *child_job_get_parent_desc(BdrvChild *c)
  82 {
  83     BlockJob *job = c->opaque;
  84     return g_strdup_printf("%s job '%s'", job_type_str(&job->job), job->job.id);
  85 }
  86
  87 static void child_job_drained_begin(BdrvChild *c)
  88 {
  89     BlockJob *job = c->opaque;
  90     job_pause(&job->job);
  91 }
  92
  93 static bool child_job_drained_poll(BdrvChild *c)
  94 {
  95     BlockJob *bjob = c->opaque;
  96     Job *job = &bjob->job;
  97     const BlockJobDriver *drv = block_job_driver(bjob);
  98
  99     /* An inactive or completed job doesn't have any pending requests. Jobs
 100      * with !job->busy are either already paused or have a pause point after
 101      * being reentered, so no job driver code will run before they pause. */
 102     if (!job->busy || job_is_completed(job)) {
 103         return false;
 104     }
 105
 106     /* Otherwise, assume that it isn't fully stopped yet, but allow the job to
 107      * override this assumption. */
 108     if (drv->drained_poll) {
 109         return drv->drained_poll(bjob);
 110     } else {
 111         return true;
 112     }
 113 }
 114
 115 static void child_job_drained_end(BdrvChild *c, int *drained_end_counter)
 116 {
 117     BlockJob *job = c->opaque;
 118     job_resume(&job->job);
 119 }
 120
 121 static bool child_job_can_set_aio_ctx(BdrvChild *c, AioContext *ctx,
 122                                       GSList **ignore, Error **errp)
 123 {
 124     BlockJob *job = c->opaque;
 125     GSList *l;
 126
 127     for (l = job->nodes; l; l = l->next) {
 128         BdrvChild *sibling = l->data;
 129         if (!bdrv_child_can_set_aio_context(sibling, ctx, ignore, errp)) {
 130             return false;
 131         }
 132     }
 133     return true;
 134 }
 135
 136 static void child_job_set_aio_ctx(BdrvChild *c, AioContext *ctx,
 137                                   GSList **ignore)
 138 {
 139     BlockJob *job = c->opaque;
 140     GSList *l;
 141
 142     for (l = job->nodes; l; l = l->next) {
 143         BdrvChild *sibling = l->data;
 144         if (g_slist_find(*ignore, sibling)) {
 145             continue;
 146         }
 147         *ignore = g_slist_prepend(*ignore, sibling);
 148         bdrv_set_aio_context_ignore(sibling->bs, ctx, ignore);
 149     }
 150
 151     job->job.aio_context = ctx;
 152 }
 153
 154 static AioContext *child_job_get_parent_aio_context(BdrvChild *c)
 155 {
 156     BlockJob *job = c->opaque;
 157
 158     return job->job.aio_context;
 159 }
 160
 161 static const BdrvChildClass child_job = {
 162     .get_parent_desc    = child_job_get_parent_desc,
 163     .drained_begin      = child_job_drained_begin,
 164     .drained_poll       = child_job_drained_poll,
 165     .drained_end        = child_job_drained_end,
 166     .can_set_aio_ctx    = child_job_can_set_aio_ctx,
 167     .set_aio_ctx        = child_job_set_aio_ctx,
 168     .stay_at_node       = true,
 169     .get_parent_aio_context = child_job_get_parent_aio_context,
 170 };
 171
 172 void block_job_remove_all_bdrv(BlockJob *job)
 173 {
 174     GLOBAL_STATE_CODE();
 175     /*
 176      * bdrv_root_unref_child() may reach child_job_[can_]set_aio_ctx(),
 177      * which will also traverse job->nodes, so consume the list one by
 178      * one to make sure that such a concurrent access does not attempt
 179      * to process an already freed BdrvChild.
 180      */
 181     while (job->nodes) {
 182         GSList *l = job->nodes;
 183         BdrvChild *c = l->data;
 184
 185         job->nodes = l->next;
 186
 187         bdrv_op_unblock_all(c->bs, job->blocker);
 188         bdrv_root_unref_child(c);
 189
 190         g_slist_free_1(l);
 191     }
 192 }
 193
 194 bool block_job_has_bdrv(BlockJob *job, BlockDriverState *bs)
 195 {
 196     GSList *el;
 197     GLOBAL_STATE_CODE();
 198
 199     for (el = job->nodes; el; el = el->next) {
 200         BdrvChild *c = el->data;
 201         if (c->bs == bs) {
 202             return true;
 203         }
 204     }
 205
 206     return false;
 207 }
 208
 209 int block_job_add_bdrv(BlockJob *job, const char *name, BlockDriverState *bs,
 210                        uint64_t perm, uint64_t shared_perm, Error **errp)
 211 {
 212     BdrvChild *c;
 213     bool need_context_ops;
 214     GLOBAL_STATE_CODE();
 215
 216     bdrv_ref(bs);
 217
 218     need_context_ops = bdrv_get_aio_context(bs) != job->job.aio_context;
 219
 220     if (need_context_ops && job->job.aio_context != qemu_get_aio_context()) {
 221         aio_context_release(job->job.aio_context);
 222     }
 223     c = bdrv_root_attach_child(bs, name, &child_job, 0, perm, shared_perm, job,
 224                                errp);
 225     if (need_context_ops && job->job.aio_context != qemu_get_aio_context()) {
 226         aio_context_acquire(job->job.aio_context);
 227     }
 228     if (c == NULL) {
 229         return -EPERM;
 230     }
 231
 232     job->nodes = g_slist_prepend(job->nodes, c);
 233     bdrv_op_block_all(bs, job->blocker);
 234
 235     return 0;
 236 }
 237
 238 static void block_job_on_idle(Notifier *n, void *opaque)
 239 {
 240     aio_wait_kick();
 241 }
 242
 243 bool block_job_is_internal(BlockJob *job)
 244 {
 245     return (job->job.id == NULL);
 246 }
 247
 248 const BlockJobDriver *block_job_driver(BlockJob *job)
 249 {
 250     return container_of(job->job.driver, BlockJobDriver, job_driver);
 251 }
 252
 253 /* Assumes the job_mutex is held */
 254 static bool job_timer_pending(Job *job)
 255 {
 256     return timer_pending(&job->sleep_timer);
 257 }
 258
 259 bool block_job_set_speed(BlockJob *job, int64_t speed, Error **errp)
 260 {
 261     const BlockJobDriver *drv = block_job_driver(job);
 262     int64_t old_speed = job->speed;
 263
 264     GLOBAL_STATE_CODE();
 265
 266     if (job_apply_verb(&job->job, JOB_VERB_SET_SPEED, errp) < 0) {
 267         return false;
 268     }
 269     if (speed < 0) {
 270         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "speed",
 271                    "a non-negative value");
 272         return false;
 273     }
 274
 275     ratelimit_set_speed(&job->limit, speed, BLOCK_JOB_SLICE_TIME);
 276
 277     job->speed = speed;
 278
 279     if (drv->set_speed) {
 280         drv->set_speed(job, speed);
 281     }
 282
 283     if (speed && speed <= old_speed) {
 284         return true;
 285     }
 286
 287     /* kick only if a timer is pending */
 288     job_enter_cond(&job->job, job_timer_pending);
 289
 290     return true;
 291 }
 292
 293 int64_t block_job_ratelimit_get_delay(BlockJob *job, uint64_t n)
 294 {
 295     IO_CODE();
 296     return ratelimit_calculate_delay(&job->limit, n);
 297 }
 298
 299 BlockJobInfo *block_job_query(BlockJob *job, Error **errp)
 300 {
 301     BlockJobInfo *info;
 302     uint64_t progress_current, progress_total;
 303
 304     GLOBAL_STATE_CODE();
 305
 306     if (block_job_is_internal(job)) {
 307         error_setg(errp, "Cannot query QEMU internal jobs");
 308         return NULL;
 309     }
 310
 311     progress_get_snapshot(&job->job.progress, &progress_current,
 312                           &progress_total);
 313
 314     info = g_new0(BlockJobInfo, 1);
 315     info->type      = g_strdup(job_type_str(&job->job));
 316     info->device    = g_strdup(job->job.id);
 317     info->busy      = qatomic_read(&job->job.busy);
 318     info->paused    = job->job.pause_count > 0;
 319     info->offset    = progress_current;
 320     info->len       = progress_total;
 321     info->speed     = job->speed;
 322     info->io_status = job->iostatus;
 323     info->ready     = job_is_ready(&job->job),
 324     info->status    = job->job.status;
 325     info->auto_finalize = job->job.auto_finalize;
 326     info->auto_dismiss  = job->job.auto_dismiss;
 327     if (job->job.ret) {
 328         info->has_error = true;
 329         info->error = job->job.err ?
 330                         g_strdup(error_get_pretty(job->job.err)) :
 331                         g_strdup(strerror(-job->job.ret));
 332     }
 333     return info;
 334 }
 335
 336 static void block_job_iostatus_set_err(BlockJob *job, int error)
 337 {
 338     if (job->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
 339         job->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE :
 340                                           BLOCK_DEVICE_IO_STATUS_FAILED;
 341     }
 342 }
 343
 344 static void block_job_event_cancelled(Notifier *n, void *opaque)
 345 {
 346     BlockJob *job = opaque;
 347     uint64_t progress_current, progress_total;
 348
 349     if (block_job_is_internal(job)) {
 350         return;
 351     }
 352
 353     progress_get_snapshot(&job->job.progress, &progress_current,
 354                           &progress_total);
 355
 356     qapi_event_send_block_job_cancelled(job_type(&job->job),
 357                                         job->job.id,
 358                                         progress_total,
 359                                         progress_current,
 360                                         job->speed);
 361 }
 362
 363 static void block_job_event_completed(Notifier *n, void *opaque)
 364 {
 365     BlockJob *job = opaque;
 366     const char *msg = NULL;
 367     uint64_t progress_current, progress_total;
 368
 369     if (block_job_is_internal(job)) {
 370         return;
 371     }
 372
 373     if (job->job.ret < 0) {
 374         msg = error_get_pretty(job->job.err);
 375     }
 376
 377     progress_get_snapshot(&job->job.progress, &progress_current,
 378                           &progress_total);
 379
 380     qapi_event_send_block_job_completed(job_type(&job->job),
 381                                         job->job.id,
 382                                         progress_total,
 383                                         progress_current,
 384                                         job->speed,
 385                                         !!msg,
 386                                         msg);
 387 }
 388
 389 static void block_job_event_pending(Notifier *n, void *opaque)
 390 {
 391     BlockJob *job = opaque;
 392
 393     if (block_job_is_internal(job)) {
 394         return;
 395     }
 396
 397     qapi_event_send_block_job_pending(job_type(&job->job),
 398                                       job->job.id);
 399 }
 400
 401 static void block_job_event_ready(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_ready(job_type(&job->job),
 414                                     job->job.id,
 415                                     progress_total,
 416                                     progress_current,
 417                                     job->speed);
 418 }
 419
 420
 421 void *block_job_create(const char *job_id, const BlockJobDriver *driver,
 422                        JobTxn *txn, BlockDriverState *bs, uint64_t perm,
 423                        uint64_t shared_perm, int64_t speed, int flags,
 424                        BlockCompletionFunc *cb, void *opaque, Error **errp)
 425 {
 426     BlockJob *job;
 427     int ret;
 428     GLOBAL_STATE_CODE();
 429
 430     if (job_id == NULL && !(flags & JOB_INTERNAL)) {
 431         job_id = bdrv_get_device_name(bs);
 432     }
 433
 434     job = job_create(job_id, &driver->job_driver, txn, bdrv_get_aio_context(bs),
 435                      flags, cb, opaque, errp);
 436     if (job == NULL) {
 437         return NULL;
 438     }
 439
 440     assert(is_block_job(&job->job));
 441     assert(job->job.driver->free == &block_job_free);
 442     assert(job->job.driver->user_resume == &block_job_user_resume);
 443
 444     ratelimit_init(&job->limit);
 445
 446     job->finalize_cancelled_notifier.notify = block_job_event_cancelled;
 447     job->finalize_completed_notifier.notify = block_job_event_completed;
 448     job->pending_notifier.notify = block_job_event_pending;
 449     job->ready_notifier.notify = block_job_event_ready;
 450     job->idle_notifier.notify = block_job_on_idle;
 451
 452     notifier_list_add(&job->job.on_finalize_cancelled,
 453                       &job->finalize_cancelled_notifier);
 454     notifier_list_add(&job->job.on_finalize_completed,
 455                       &job->finalize_completed_notifier);
 456     notifier_list_add(&job->job.on_pending, &job->pending_notifier);
 457     notifier_list_add(&job->job.on_ready, &job->ready_notifier);
 458     notifier_list_add(&job->job.on_idle, &job->idle_notifier);
 459
 460     error_setg(&job->blocker, "block device is in use by block job: %s",
 461                job_type_str(&job->job));
 462
 463     ret = block_job_add_bdrv(job, "main node", bs, perm, shared_perm, errp);
 464     if (ret < 0) {
 465         goto fail;
 466     }
 467
 468     bdrv_op_unblock(bs, BLOCK_OP_TYPE_DATAPLANE, job->blocker);
 469
 470     if (!block_job_set_speed(job, speed, errp)) {
 471         goto fail;
 472     }
 473
 474     return job;
 475
 476 fail:
 477     job_early_fail(&job->job);
 478     return NULL;
 479 }
 480
 481 void block_job_iostatus_reset(BlockJob *job)
 482 {
 483     GLOBAL_STATE_CODE();
 484     if (job->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
 485         return;
 486     }
 487     assert(job->job.user_paused && job->job.pause_count > 0);
 488     job->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
 489 }
 490
 491 void block_job_user_resume(Job *job)
 492 {
 493     BlockJob *bjob = container_of(job, BlockJob, job);
 494     GLOBAL_STATE_CODE();
 495     block_job_iostatus_reset(bjob);
 496 }
 497
 498 BlockErrorAction block_job_error_action(BlockJob *job, BlockdevOnError on_err,
 499                                         int is_read, int error)
 500 {
 501     BlockErrorAction action;
 502     IO_CODE();
 503
 504     switch (on_err) {
 505     case BLOCKDEV_ON_ERROR_ENOSPC:
 506     case BLOCKDEV_ON_ERROR_AUTO:
 507         action = (error == ENOSPC) ?
 508                  BLOCK_ERROR_ACTION_STOP : BLOCK_ERROR_ACTION_REPORT;
 509         break;
 510     case BLOCKDEV_ON_ERROR_STOP:
 511         action = BLOCK_ERROR_ACTION_STOP;
 512         break;
 513     case BLOCKDEV_ON_ERROR_REPORT:
 514         action = BLOCK_ERROR_ACTION_REPORT;
 515         break;
 516     case BLOCKDEV_ON_ERROR_IGNORE:
 517         action = BLOCK_ERROR_ACTION_IGNORE;
 518         break;
 519     default:
 520         abort();
 521     }
 522     if (!block_job_is_internal(job)) {
 523         qapi_event_send_block_job_error(job->job.id,
 524                                         is_read ? IO_OPERATION_TYPE_READ :
 525                                         IO_OPERATION_TYPE_WRITE,
 526                                         action);
 527     }
 528     if (action == BLOCK_ERROR_ACTION_STOP) {
 529         if (!job->job.user_paused) {
 530             job_pause(&job->job);
 531             /* make the pause user visible, which will be resumed from QMP. */
 532             job->job.user_paused = true;
 533         }
 534         block_job_iostatus_set_err(job, error);
 535     }
 536     return action;
 537 }
 538
 539 AioContext *block_job_get_aio_context(BlockJob *job)
 540 {
 541     GLOBAL_STATE_CODE();
 542     return job->job.aio_context;
 543 }