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     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     /*
 187      * bdrv_root_unref_child() may reach child_job_[can_]set_aio_ctx(),
 188      * which will also traverse job->nodes, so consume the list one by
 189      * one to make sure that such a concurrent access does not attempt
 190      * to process an already freed BdrvChild.
 191      */
 192     while (job->nodes) {
 193         GSList *l = job->nodes;
 194         BdrvChild *c = l->data;
 195
 196         job->nodes = l->next;
 197
 198         bdrv_op_unblock_all(c->bs, job->blocker);
 199         bdrv_root_unref_child(c);
 200
 201         g_slist_free_1(l);
 202     }
 203 }
 204
 205 bool block_job_has_bdrv(BlockJob *job, BlockDriverState *bs)
 206 {
 207     GSList *el;
 208
 209     for (el = job->nodes; el; el = el->next) {
 210         BdrvChild *c = el->data;
 211         if (c->bs == bs) {
 212             return true;
 213         }
 214     }
 215
 216     return false;
 217 }
 218
 219 int block_job_add_bdrv(BlockJob *job, const char *name, BlockDriverState *bs,
 220                        uint64_t perm, uint64_t shared_perm, Error **errp)
 221 {
 222     BdrvChild *c;
 223     bool need_context_ops;
 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 static void block_job_on_idle(Notifier *n, void *opaque)
 248 {
 249     aio_wait_kick();
 250 }
 251
 252 bool block_job_is_internal(BlockJob *job)
 253 {
 254     return (job->job.id == NULL);
 255 }
 256
 257 const BlockJobDriver *block_job_driver(BlockJob *job)
 258 {
 259     return container_of(job->job.driver, BlockJobDriver, job_driver);
 260 }
 261
 262 /* Assumes the job_mutex is held */
 263 static bool job_timer_pending(Job *job)
 264 {
 265     return timer_pending(&job->sleep_timer);
 266 }
 267
 268 bool block_job_set_speed(BlockJob *job, int64_t speed, Error **errp)
 269 {
 270     const BlockJobDriver *drv = block_job_driver(job);
 271     int64_t old_speed = job->speed;
 272
 273     if (job_apply_verb(&job->job, JOB_VERB_SET_SPEED, errp) < 0) {
 274         return false;
 275     }
 276     if (speed < 0) {
 277         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "speed",
 278                    "a non-negative value");
 279         return false;
 280     }
 281
 282     ratelimit_set_speed(&job->limit, speed, BLOCK_JOB_SLICE_TIME);
 283
 284     job->speed = speed;
 285
 286     if (drv->set_speed) {
 287         drv->set_speed(job, speed);
 288     }
 289
 290     if (speed && speed <= old_speed) {
 291         return true;
 292     }
 293
 294     /* kick only if a timer is pending */
 295     job_enter_cond(&job->job, job_timer_pending);
 296
 297     return true;
 298 }
 299
 300 int64_t block_job_ratelimit_get_delay(BlockJob *job, uint64_t n)
 301 {
 302     return ratelimit_calculate_delay(&job->limit, n);
 303 }
 304
 305 BlockJobInfo *block_job_query(BlockJob *job, Error **errp)
 306 {
 307     BlockJobInfo *info;
 308     uint64_t progress_current, progress_total;
 309
 310     if (block_job_is_internal(job)) {
 311         error_setg(errp, "Cannot query QEMU internal jobs");
 312         return NULL;
 313     }
 314
 315     progress_get_snapshot(&job->job.progress, &progress_current,
 316                           &progress_total);
 317
 318     info = g_new0(BlockJobInfo, 1);
 319     info->type      = g_strdup(job_type_str(&job->job));
 320     info->device    = g_strdup(job->job.id);
 321     info->busy      = qatomic_read(&job->job.busy);
 322     info->paused    = job->job.pause_count > 0;
 323     info->offset    = progress_current;
 324     info->len       = progress_total;
 325     info->speed     = job->speed;
 326     info->io_status = job->iostatus;
 327     info->ready     = job_is_ready(&job->job),
 328     info->status    = job->job.status;
 329     info->auto_finalize = job->job.auto_finalize;
 330     info->auto_dismiss  = job->job.auto_dismiss;
 331     if (job->job.ret) {
 332         info->has_error = true;
 333         info->error = job->job.err ?
 334                         g_strdup(error_get_pretty(job->job.err)) :
 335                         g_strdup(strerror(-job->job.ret));
 336     }
 337     return info;
 338 }
 339
 340 static void block_job_iostatus_set_err(BlockJob *job, int error)
 341 {
 342     if (job->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
 343         job->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE :
 344                                           BLOCK_DEVICE_IO_STATUS_FAILED;
 345     }
 346 }
 347
 348 static void block_job_event_cancelled(Notifier *n, void *opaque)
 349 {
 350     BlockJob *job = opaque;
 351     uint64_t progress_current, progress_total;
 352
 353     if (block_job_is_internal(job)) {
 354         return;
 355     }
 356
 357     progress_get_snapshot(&job->job.progress, &progress_current,
 358                           &progress_total);
 359
 360     qapi_event_send_block_job_cancelled(job_type(&job->job),
 361                                         job->job.id,
 362                                         progress_total,
 363                                         progress_current,
 364                                         job->speed);
 365 }
 366
 367 static void block_job_event_completed(Notifier *n, void *opaque)
 368 {
 369     BlockJob *job = opaque;
 370     const char *msg = NULL;
 371     uint64_t progress_current, progress_total;
 372
 373     if (block_job_is_internal(job)) {
 374         return;
 375     }
 376
 377     if (job->job.ret < 0) {
 378         msg = error_get_pretty(job->job.err);
 379     }
 380
 381     progress_get_snapshot(&job->job.progress, &progress_current,
 382                           &progress_total);
 383
 384     qapi_event_send_block_job_completed(job_type(&job->job),
 385                                         job->job.id,
 386                                         progress_total,
 387                                         progress_current,
 388                                         job->speed,
 389                                         !!msg,
 390                                         msg);
 391 }
 392
 393 static void block_job_event_pending(Notifier *n, void *opaque)
 394 {
 395     BlockJob *job = opaque;
 396
 397     if (block_job_is_internal(job)) {
 398         return;
 399     }
 400
 401     qapi_event_send_block_job_pending(job_type(&job->job),
 402                                       job->job.id);
 403 }
 404
 405 static void block_job_event_ready(Notifier *n, void *opaque)
 406 {
 407     BlockJob *job = opaque;
 408     uint64_t progress_current, progress_total;
 409
 410     if (block_job_is_internal(job)) {
 411         return;
 412     }
 413
 414     progress_get_snapshot(&job->job.progress, &progress_current,
 415                           &progress_total);
 416
 417     qapi_event_send_block_job_ready(job_type(&job->job),
 418                                     job->job.id,
 419                                     progress_total,
 420                                     progress_current,
 421                                     job->speed);
 422 }
 423
 424
 425 /*
 426  * API for block job drivers and the block layer.  These functions are
 427  * declared in blockjob_int.h.
 428  */
 429
 430 void *block_job_create(const char *job_id, const BlockJobDriver *driver,
 431                        JobTxn *txn, BlockDriverState *bs, uint64_t perm,
 432                        uint64_t shared_perm, int64_t speed, int flags,
 433                        BlockCompletionFunc *cb, void *opaque, Error **errp)
 434 {
 435     BlockJob *job;
 436     int ret;
 437
 438     if (job_id == NULL && !(flags & JOB_INTERNAL)) {
 439         job_id = bdrv_get_device_name(bs);
 440     }
 441
 442     job = job_create(job_id, &driver->job_driver, txn, bdrv_get_aio_context(bs),
 443                      flags, cb, opaque, errp);
 444     if (job == NULL) {
 445         return NULL;
 446     }
 447
 448     assert(is_block_job(&job->job));
 449     assert(job->job.driver->free == &block_job_free);
 450     assert(job->job.driver->user_resume == &block_job_user_resume);
 451
 452     ratelimit_init(&job->limit);
 453
 454     job->finalize_cancelled_notifier.notify = block_job_event_cancelled;
 455     job->finalize_completed_notifier.notify = block_job_event_completed;
 456     job->pending_notifier.notify = block_job_event_pending;
 457     job->ready_notifier.notify = block_job_event_ready;
 458     job->idle_notifier.notify = block_job_on_idle;
 459
 460     notifier_list_add(&job->job.on_finalize_cancelled,
 461                       &job->finalize_cancelled_notifier);
 462     notifier_list_add(&job->job.on_finalize_completed,
 463                       &job->finalize_completed_notifier);
 464     notifier_list_add(&job->job.on_pending, &job->pending_notifier);
 465     notifier_list_add(&job->job.on_ready, &job->ready_notifier);
 466     notifier_list_add(&job->job.on_idle, &job->idle_notifier);
 467
 468     error_setg(&job->blocker, "block device is in use by block job: %s",
 469                job_type_str(&job->job));
 470
 471     ret = block_job_add_bdrv(job, "main node", bs, perm, shared_perm, errp);
 472     if (ret < 0) {
 473         goto fail;
 474     }
 475
 476     bdrv_op_unblock(bs, BLOCK_OP_TYPE_DATAPLANE, job->blocker);
 477
 478     if (!block_job_set_speed(job, speed, errp)) {
 479         goto fail;
 480     }
 481
 482     return job;
 483
 484 fail:
 485     job_early_fail(&job->job);
 486     return NULL;
 487 }
 488
 489 void block_job_iostatus_reset(BlockJob *job)
 490 {
 491     if (job->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
 492         return;
 493     }
 494     assert(job->job.user_paused && job->job.pause_count > 0);
 495     job->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
 496 }
 497
 498 void block_job_user_resume(Job *job)
 499 {
 500     BlockJob *bjob = container_of(job, BlockJob, job);
 501     block_job_iostatus_reset(bjob);
 502 }
 503
 504 BlockErrorAction block_job_error_action(BlockJob *job, BlockdevOnError on_err,
 505                                         int is_read, int error)
 506 {
 507     BlockErrorAction action;
 508
 509     switch (on_err) {
 510     case BLOCKDEV_ON_ERROR_ENOSPC:
 511     case BLOCKDEV_ON_ERROR_AUTO:
 512         action = (error == ENOSPC) ?
 513                  BLOCK_ERROR_ACTION_STOP : BLOCK_ERROR_ACTION_REPORT;
 514         break;
 515     case BLOCKDEV_ON_ERROR_STOP:
 516         action = BLOCK_ERROR_ACTION_STOP;
 517         break;
 518     case BLOCKDEV_ON_ERROR_REPORT:
 519         action = BLOCK_ERROR_ACTION_REPORT;
 520         break;
 521     case BLOCKDEV_ON_ERROR_IGNORE:
 522         action = BLOCK_ERROR_ACTION_IGNORE;
 523         break;
 524     default:
 525         abort();
 526     }
 527     if (!block_job_is_internal(job)) {
 528         qapi_event_send_block_job_error(job->job.id,
 529                                         is_read ? IO_OPERATION_TYPE_READ :
 530                                         IO_OPERATION_TYPE_WRITE,
 531                                         action);
 532     }
 533     if (action == BLOCK_ERROR_ACTION_STOP) {
 534         if (!job->job.user_paused) {
 535             job_pause(&job->job);
 536             /* make the pause user visible, which will be resumed from QMP. */
 537             job->job.user_paused = true;
 538         }
 539         block_job_iostatus_set_err(job, error);
 540     }
 541     return action;
 542 }
 543
 544 AioContext *block_job_get_aio_context(BlockJob *job)
 545 {
 546     return job->job.aio_context;
 547 }