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 /*
  40  * The block job API is composed of two categories of functions.
  41  *
  42  * The first includes functions used by the monitor.  The monitor is
  43  * peculiar in that it accesses the block job list with block_job_get, and
  44  * therefore needs consistency across block_job_get and the actual operation
  45  * (e.g. block_job_set_speed).  The consistency is achieved with
  46  * aio_context_acquire/release.  These functions are declared in blockjob.h.
  47  *
  48  * The second includes functions used by the block job drivers and sometimes
  49  * by the core block layer.  These do not care about locking, because the
  50  * whole coroutine runs under the AioContext lock, and are declared in
  51  * blockjob_int.h.
  52  */
  53
  54 static bool is_block_job(Job *job)
  55 {
  56     return job_type(job) == JOB_TYPE_BACKUP ||
  57            job_type(job) == JOB_TYPE_COMMIT ||
  58            job_type(job) == JOB_TYPE_MIRROR ||
  59            job_type(job) == JOB_TYPE_STREAM;
  60 }
  61
  62 BlockJob *block_job_next(BlockJob *bjob)
  63 {
  64     Job *job = bjob ? &bjob->job : NULL;
  65
  66     do {
  67         job = job_next(job);
  68     } while (job && !is_block_job(job));
  69
  70     return job ? container_of(job, BlockJob, job) : NULL;
  71 }
  72
  73 BlockJob *block_job_get(const char *id)
  74 {
  75     Job *job = job_get(id);
  76
  77     if (job && is_block_job(job)) {
  78         return container_of(job, BlockJob, job);
  79     } else {
  80         return NULL;
  81     }
  82 }
  83
  84 void block_job_free(Job *job)
  85 {
  86     BlockJob *bjob = container_of(job, BlockJob, job);
  87
  88     block_job_remove_all_bdrv(bjob);
  89     blk_unref(bjob->blk);
  90     ratelimit_destroy(&bjob->limit);
  91     error_free(bjob->blocker);
  92 }
  93
  94 static char *child_job_get_parent_desc(BdrvChild *c)
  95 {
  96     BlockJob *job = c->opaque;
  97     return g_strdup_printf("%s job '%s'", job_type_str(&job->job), job->job.id);
  98 }
  99
 100 static void child_job_drained_begin(BdrvChild *c)
 101 {
 102     BlockJob *job = c->opaque;
 103     job_pause(&job->job);
 104 }
 105
 106 static bool child_job_drained_poll(BdrvChild *c)
 107 {
 108     BlockJob *bjob = c->opaque;
 109     Job *job = &bjob->job;
 110     const BlockJobDriver *drv = block_job_driver(bjob);
 111
 112     /* An inactive or completed job doesn't have any pending requests. Jobs
 113      * with !job->busy are either already paused or have a pause point after
 114      * being reentered, so no job driver code will run before they pause. */
 115     if (!job->busy || job_is_completed(job)) {
 116         return false;
 117     }
 118
 119     /* Otherwise, assume that it isn't fully stopped yet, but allow the job to
 120      * override this assumption. */
 121     if (drv->drained_poll) {
 122         return drv->drained_poll(bjob);
 123     } else {
 124         return true;
 125     }
 126 }
 127
 128 static void child_job_drained_end(BdrvChild *c, int *drained_end_counter)
 129 {
 130     BlockJob *job = c->opaque;
 131     job_resume(&job->job);
 132 }
 133
 134 static bool child_job_can_set_aio_ctx(BdrvChild *c, AioContext *ctx,
 135                                       GSList **ignore, Error **errp)
 136 {
 137     BlockJob *job = c->opaque;
 138     GSList *l;
 139
 140     for (l = job->nodes; l; l = l->next) {
 141         BdrvChild *sibling = l->data;
 142         if (!bdrv_child_can_set_aio_context(sibling, ctx, ignore, errp)) {
 143             return false;
 144         }
 145     }
 146     return true;
 147 }
 148
 149 static void child_job_set_aio_ctx(BdrvChild *c, AioContext *ctx,
 150                                   GSList **ignore)
 151 {
 152     BlockJob *job = c->opaque;
 153     GSList *l;
 154
 155     for (l = job->nodes; l; l = l->next) {
 156         BdrvChild *sibling = l->data;
 157         if (g_slist_find(*ignore, sibling)) {
 158             continue;
 159         }
 160         *ignore = g_slist_prepend(*ignore, sibling);
 161         bdrv_set_aio_context_ignore(sibling->bs, ctx, ignore);
 162     }
 163
 164     job->job.aio_context = ctx;
 165 }
 166
 167 static AioContext *child_job_get_parent_aio_context(BdrvChild *c)
 168 {
 169     BlockJob *job = c->opaque;
 170
 171     return job->job.aio_context;
 172 }
 173
 174 static const BdrvChildClass child_job = {
 175     .get_parent_desc    = child_job_get_parent_desc,
 176     .drained_begin      = child_job_drained_begin,
 177     .drained_poll       = child_job_drained_poll,
 178     .drained_end        = child_job_drained_end,
 179     .can_set_aio_ctx    = child_job_can_set_aio_ctx,
 180     .set_aio_ctx        = child_job_set_aio_ctx,
 181     .stay_at_node       = true,
 182     .get_parent_aio_context = child_job_get_parent_aio_context,
 183 };
 184
 185 void block_job_remove_all_bdrv(BlockJob *job)
 186 {
 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
 210     for (el = job->nodes; el; el = el->next) {
 211         BdrvChild *c = el->data;
 212         if (c->bs == bs) {
 213             return true;
 214         }
 215     }
 216
 217     return false;
 218 }
 219
 220 int block_job_add_bdrv(BlockJob *job, const char *name, BlockDriverState *bs,
 221                        uint64_t perm, uint64_t shared_perm, Error **errp)
 222 {
 223     BdrvChild *c;
 224     bool need_context_ops;
 225
 226     bdrv_ref(bs);
 227
 228     need_context_ops = bdrv_get_aio_context(bs) != job->job.aio_context;
 229
 230     if (need_context_ops && job->job.aio_context != qemu_get_aio_context()) {
 231         aio_context_release(job->job.aio_context);
 232     }
 233     c = bdrv_root_attach_child(bs, name, &child_job, 0, perm, shared_perm, job,
 234                                errp);
 235     if (need_context_ops && job->job.aio_context != qemu_get_aio_context()) {
 236         aio_context_acquire(job->job.aio_context);
 237     }
 238     if (c == NULL) {
 239         return -EPERM;
 240     }
 241
 242     job->nodes = g_slist_prepend(job->nodes, c);
 243     bdrv_op_block_all(bs, job->blocker);
 244
 245     return 0;
 246 }
 247
 248 static void block_job_on_idle(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(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     if (job_apply_verb(&job->job, JOB_VERB_SET_SPEED, errp) < 0) {
 275         return false;
 276     }
 277     if (speed < 0) {
 278         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "speed",
 279                    "a non-negative value");
 280         return false;
 281     }
 282
 283     ratelimit_set_speed(&job->limit, speed, BLOCK_JOB_SLICE_TIME);
 284
 285     job->speed = speed;
 286
 287     if (drv->set_speed) {
 288         drv->set_speed(job, speed);
 289     }
 290
 291     if (speed && speed <= old_speed) {
 292         return true;
 293     }
 294
 295     /* kick only if a timer is pending */
 296     job_enter_cond(&job->job, job_timer_pending);
 297
 298     return true;
 299 }
 300
 301 int64_t block_job_ratelimit_get_delay(BlockJob *job, uint64_t n)
 302 {
 303     return ratelimit_calculate_delay(&job->limit, n);
 304 }
 305
 306 BlockJobInfo *block_job_query(BlockJob *job, Error **errp)
 307 {
 308     BlockJobInfo *info;
 309     uint64_t progress_current, progress_total;
 310
 311     if (block_job_is_internal(job)) {
 312         error_setg(errp, "Cannot query QEMU internal jobs");
 313         return NULL;
 314     }
 315
 316     progress_get_snapshot(&job->job.progress, &progress_current,
 317                           &progress_total);
 318
 319     info = g_new0(BlockJobInfo, 1);
 320     info->type      = g_strdup(job_type_str(&job->job));
 321     info->device    = g_strdup(job->job.id);
 322     info->busy      = qatomic_read(&job->job.busy);
 323     info->paused    = job->job.pause_count > 0;
 324     info->offset    = progress_current;
 325     info->len       = progress_total;
 326     info->speed     = job->speed;
 327     info->io_status = job->iostatus;
 328     info->ready     = job_is_ready(&job->job),
 329     info->status    = job->job.status;
 330     info->auto_finalize = job->job.auto_finalize;
 331     info->auto_dismiss  = job->job.auto_dismiss;
 332     if (job->job.ret) {
 333         info->has_error = true;
 334         info->error = job->job.err ?
 335                         g_strdup(error_get_pretty(job->job.err)) :
 336                         g_strdup(strerror(-job->job.ret));
 337     }
 338     return info;
 339 }
 340
 341 static void block_job_iostatus_set_err(BlockJob *job, int error)
 342 {
 343     if (job->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
 344         job->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE :
 345                                           BLOCK_DEVICE_IO_STATUS_FAILED;
 346     }
 347 }
 348
 349 static void block_job_event_cancelled(Notifier *n, void *opaque)
 350 {
 351     BlockJob *job = opaque;
 352     uint64_t progress_current, progress_total;
 353
 354     if (block_job_is_internal(job)) {
 355         return;
 356     }
 357
 358     progress_get_snapshot(&job->job.progress, &progress_current,
 359                           &progress_total);
 360
 361     qapi_event_send_block_job_cancelled(job_type(&job->job),
 362                                         job->job.id,
 363                                         progress_total,
 364                                         progress_current,
 365                                         job->speed);
 366 }
 367
 368 static void block_job_event_completed(Notifier *n, void *opaque)
 369 {
 370     BlockJob *job = opaque;
 371     const char *msg = NULL;
 372     uint64_t progress_current, progress_total;
 373
 374     if (block_job_is_internal(job)) {
 375         return;
 376     }
 377
 378     if (job->job.ret < 0) {
 379         msg = error_get_pretty(job->job.err);
 380     }
 381
 382     progress_get_snapshot(&job->job.progress, &progress_current,
 383                           &progress_total);
 384
 385     qapi_event_send_block_job_completed(job_type(&job->job),
 386                                         job->job.id,
 387                                         progress_total,
 388                                         progress_current,
 389                                         job->speed,
 390                                         !!msg,
 391                                         msg);
 392 }
 393
 394 static void block_job_event_pending(Notifier *n, void *opaque)
 395 {
 396     BlockJob *job = opaque;
 397
 398     if (block_job_is_internal(job)) {
 399         return;
 400     }
 401
 402     qapi_event_send_block_job_pending(job_type(&job->job),
 403                                       job->job.id);
 404 }
 405
 406 static void block_job_event_ready(Notifier *n, void *opaque)
 407 {
 408     BlockJob *job = opaque;
 409     uint64_t progress_current, progress_total;
 410
 411     if (block_job_is_internal(job)) {
 412         return;
 413     }
 414
 415     progress_get_snapshot(&job->job.progress, &progress_current,
 416                           &progress_total);
 417
 418     qapi_event_send_block_job_ready(job_type(&job->job),
 419                                     job->job.id,
 420                                     progress_total,
 421                                     progress_current,
 422                                     job->speed);
 423 }
 424
 425
 426 /*
 427  * API for block job drivers and the block layer.  These functions are
 428  * declared in blockjob_int.h.
 429  */
 430
 431 void *block_job_create(const char *job_id, const BlockJobDriver *driver,
 432                        JobTxn *txn, BlockDriverState *bs, uint64_t perm,
 433                        uint64_t shared_perm, int64_t speed, int flags,
 434                        BlockCompletionFunc *cb, void *opaque, Error **errp)
 435 {
 436     BlockBackend *blk;
 437     BlockJob *job;
 438
 439     if (job_id == NULL && !(flags & JOB_INTERNAL)) {
 440         job_id = bdrv_get_device_name(bs);
 441     }
 442
 443     blk = blk_new_with_bs(bs, perm, shared_perm, errp);
 444     if (!blk) {
 445         return NULL;
 446     }
 447
 448     job = job_create(job_id, &driver->job_driver, txn, blk_get_aio_context(blk),
 449                      flags, cb, opaque, errp);
 450     if (job == NULL) {
 451         blk_unref(blk);
 452         return NULL;
 453     }
 454
 455     assert(is_block_job(&job->job));
 456     assert(job->job.driver->free == &block_job_free);
 457     assert(job->job.driver->user_resume == &block_job_user_resume);
 458
 459     ratelimit_init(&job->limit);
 460
 461     job->blk = blk;
 462
 463     job->finalize_cancelled_notifier.notify = block_job_event_cancelled;
 464     job->finalize_completed_notifier.notify = block_job_event_completed;
 465     job->pending_notifier.notify = block_job_event_pending;
 466     job->ready_notifier.notify = block_job_event_ready;
 467     job->idle_notifier.notify = block_job_on_idle;
 468
 469     notifier_list_add(&job->job.on_finalize_cancelled,
 470                       &job->finalize_cancelled_notifier);
 471     notifier_list_add(&job->job.on_finalize_completed,
 472                       &job->finalize_completed_notifier);
 473     notifier_list_add(&job->job.on_pending, &job->pending_notifier);
 474     notifier_list_add(&job->job.on_ready, &job->ready_notifier);
 475     notifier_list_add(&job->job.on_idle, &job->idle_notifier);
 476
 477     error_setg(&job->blocker, "block device is in use by block job: %s",
 478                job_type_str(&job->job));
 479     block_job_add_bdrv(job, "main node", bs, 0, BLK_PERM_ALL, &error_abort);
 480
 481     bdrv_op_unblock(bs, BLOCK_OP_TYPE_DATAPLANE, job->blocker);
 482
 483     /* Disable request queuing in the BlockBackend to avoid deadlocks on drain:
 484      * The job reports that it's busy until it reaches a pause point. */
 485     blk_set_disable_request_queuing(blk, true);
 486     blk_set_allow_aio_context_change(blk, true);
 487
 488     if (!block_job_set_speed(job, speed, errp)) {
 489         job_early_fail(&job->job);
 490         return NULL;
 491     }
 492
 493     return job;
 494 }
 495
 496 void block_job_iostatus_reset(BlockJob *job)
 497 {
 498     if (job->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
 499         return;
 500     }
 501     assert(job->job.user_paused && job->job.pause_count > 0);
 502     job->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
 503 }
 504
 505 void block_job_user_resume(Job *job)
 506 {
 507     BlockJob *bjob = container_of(job, BlockJob, job);
 508     block_job_iostatus_reset(bjob);
 509 }
 510
 511 BlockErrorAction block_job_error_action(BlockJob *job, BlockdevOnError on_err,
 512                                         int is_read, int error)
 513 {
 514     BlockErrorAction action;
 515
 516     switch (on_err) {
 517     case BLOCKDEV_ON_ERROR_ENOSPC:
 518     case BLOCKDEV_ON_ERROR_AUTO:
 519         action = (error == ENOSPC) ?
 520                  BLOCK_ERROR_ACTION_STOP : BLOCK_ERROR_ACTION_REPORT;
 521         break;
 522     case BLOCKDEV_ON_ERROR_STOP:
 523         action = BLOCK_ERROR_ACTION_STOP;
 524         break;
 525     case BLOCKDEV_ON_ERROR_REPORT:
 526         action = BLOCK_ERROR_ACTION_REPORT;
 527         break;
 528     case BLOCKDEV_ON_ERROR_IGNORE:
 529         action = BLOCK_ERROR_ACTION_IGNORE;
 530         break;
 531     default:
 532         abort();
 533     }
 534     if (!block_job_is_internal(job)) {
 535         qapi_event_send_block_job_error(job->job.id,
 536                                         is_read ? IO_OPERATION_TYPE_READ :
 537                                         IO_OPERATION_TYPE_WRITE,
 538                                         action);
 539     }
 540     if (action == BLOCK_ERROR_ACTION_STOP) {
 541         if (!job->job.user_paused) {
 542             job_pause(&job->job);
 543             /* make the pause user visible, which will be resumed from QMP. */
 544             job->job.user_paused = true;
 545         }
 546         block_job_iostatus_set_err(job, error);
 547     }
 548     return action;
 549 }