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     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 const BdrvChildRole child_job = {
 167     .get_parent_desc    = child_job_get_parent_desc,
 168     .drained_begin      = child_job_drained_begin,
 169     .drained_poll       = child_job_drained_poll,
 170     .drained_end        = child_job_drained_end,
 171     .can_set_aio_ctx    = child_job_can_set_aio_ctx,
 172     .set_aio_ctx        = child_job_set_aio_ctx,
 173     .stay_at_node       = true,
 174 };
 175
 176 void block_job_remove_all_bdrv(BlockJob *job)
 177 {
 178     GSList *l;
 179     for (l = job->nodes; l; l = l->next) {
 180         BdrvChild *c = l->data;
 181         bdrv_op_unblock_all(c->bs, job->blocker);
 182         bdrv_root_unref_child(c);
 183     }
 184     g_slist_free(job->nodes);
 185     job->nodes = NULL;
 186 }
 187
 188 bool block_job_has_bdrv(BlockJob *job, BlockDriverState *bs)
 189 {
 190     GSList *el;
 191
 192     for (el = job->nodes; el; el = el->next) {
 193         BdrvChild *c = el->data;
 194         if (c->bs == bs) {
 195             return true;
 196         }
 197     }
 198
 199     return false;
 200 }
 201
 202 int block_job_add_bdrv(BlockJob *job, const char *name, BlockDriverState *bs,
 203                        uint64_t perm, uint64_t shared_perm, Error **errp)
 204 {
 205     BdrvChild *c;
 206
 207     bdrv_ref(bs);
 208     if (job->job.aio_context != qemu_get_aio_context()) {
 209         aio_context_release(job->job.aio_context);
 210     }
 211     c = bdrv_root_attach_child(bs, name, &child_job, job->job.aio_context,
 212                                perm, shared_perm, job, errp);
 213     if (job->job.aio_context != qemu_get_aio_context()) {
 214         aio_context_acquire(job->job.aio_context);
 215     }
 216     if (c == NULL) {
 217         return -EPERM;
 218     }
 219
 220     job->nodes = g_slist_prepend(job->nodes, c);
 221     bdrv_op_block_all(bs, job->blocker);
 222
 223     return 0;
 224 }
 225
 226 static void block_job_on_idle(Notifier *n, void *opaque)
 227 {
 228     aio_wait_kick();
 229 }
 230
 231 bool block_job_is_internal(BlockJob *job)
 232 {
 233     return (job->job.id == NULL);
 234 }
 235
 236 const BlockJobDriver *block_job_driver(BlockJob *job)
 237 {
 238     return container_of(job->job.driver, BlockJobDriver, job_driver);
 239 }
 240
 241 /* Assumes the job_mutex is held */
 242 static bool job_timer_pending(Job *job)
 243 {
 244     return timer_pending(&job->sleep_timer);
 245 }
 246
 247 void block_job_set_speed(BlockJob *job, int64_t speed, Error **errp)
 248 {
 249     int64_t old_speed = job->speed;
 250
 251     if (job_apply_verb(&job->job, JOB_VERB_SET_SPEED, errp)) {
 252         return;
 253     }
 254     if (speed < 0) {
 255         error_setg(errp, QERR_INVALID_PARAMETER, "speed");
 256         return;
 257     }
 258
 259     ratelimit_set_speed(&job->limit, speed, BLOCK_JOB_SLICE_TIME);
 260
 261     job->speed = speed;
 262     if (speed && speed <= old_speed) {
 263         return;
 264     }
 265
 266     /* kick only if a timer is pending */
 267     job_enter_cond(&job->job, job_timer_pending);
 268 }
 269
 270 int64_t block_job_ratelimit_get_delay(BlockJob *job, uint64_t n)
 271 {
 272     if (!job->speed) {
 273         return 0;
 274     }
 275
 276     return ratelimit_calculate_delay(&job->limit, n);
 277 }
 278
 279 BlockJobInfo *block_job_query(BlockJob *job, Error **errp)
 280 {
 281     BlockJobInfo *info;
 282
 283     if (block_job_is_internal(job)) {
 284         error_setg(errp, "Cannot query QEMU internal jobs");
 285         return NULL;
 286     }
 287     info = g_new0(BlockJobInfo, 1);
 288     info->type      = g_strdup(job_type_str(&job->job));
 289     info->device    = g_strdup(job->job.id);
 290     info->busy      = atomic_read(&job->job.busy);
 291     info->paused    = job->job.pause_count > 0;
 292     info->offset    = job->job.progress_current;
 293     info->len       = job->job.progress_total;
 294     info->speed     = job->speed;
 295     info->io_status = job->iostatus;
 296     info->ready     = job_is_ready(&job->job),
 297     info->status    = job->job.status;
 298     info->auto_finalize = job->job.auto_finalize;
 299     info->auto_dismiss  = job->job.auto_dismiss;
 300     info->has_error = job->job.ret != 0;
 301     info->error     = job->job.ret ? g_strdup(strerror(-job->job.ret)) : NULL;
 302     return info;
 303 }
 304
 305 static void block_job_iostatus_set_err(BlockJob *job, int error)
 306 {
 307     if (job->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
 308         job->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE :
 309                                           BLOCK_DEVICE_IO_STATUS_FAILED;
 310     }
 311 }
 312
 313 static void block_job_event_cancelled(Notifier *n, void *opaque)
 314 {
 315     BlockJob *job = opaque;
 316
 317     if (block_job_is_internal(job)) {
 318         return;
 319     }
 320
 321     qapi_event_send_block_job_cancelled(job_type(&job->job),
 322                                         job->job.id,
 323                                         job->job.progress_total,
 324                                         job->job.progress_current,
 325                                         job->speed);
 326 }
 327
 328 static void block_job_event_completed(Notifier *n, void *opaque)
 329 {
 330     BlockJob *job = opaque;
 331     const char *msg = NULL;
 332
 333     if (block_job_is_internal(job)) {
 334         return;
 335     }
 336
 337     if (job->job.ret < 0) {
 338         msg = strerror(-job->job.ret);
 339     }
 340
 341     qapi_event_send_block_job_completed(job_type(&job->job),
 342                                         job->job.id,
 343                                         job->job.progress_total,
 344                                         job->job.progress_current,
 345                                         job->speed,
 346                                         !!msg,
 347                                         msg);
 348 }
 349
 350 static void block_job_event_pending(Notifier *n, void *opaque)
 351 {
 352     BlockJob *job = opaque;
 353
 354     if (block_job_is_internal(job)) {
 355         return;
 356     }
 357
 358     qapi_event_send_block_job_pending(job_type(&job->job),
 359                                       job->job.id);
 360 }
 361
 362 static void block_job_event_ready(Notifier *n, void *opaque)
 363 {
 364     BlockJob *job = opaque;
 365
 366     if (block_job_is_internal(job)) {
 367         return;
 368     }
 369
 370     qapi_event_send_block_job_ready(job_type(&job->job),
 371                                     job->job.id,
 372                                     job->job.progress_total,
 373                                     job->job.progress_current,
 374                                     job->speed);
 375 }
 376
 377
 378 /*
 379  * API for block job drivers and the block layer.  These functions are
 380  * declared in blockjob_int.h.
 381  */
 382
 383 void *block_job_create(const char *job_id, const BlockJobDriver *driver,
 384                        JobTxn *txn, BlockDriverState *bs, uint64_t perm,
 385                        uint64_t shared_perm, int64_t speed, int flags,
 386                        BlockCompletionFunc *cb, void *opaque, Error **errp)
 387 {
 388     BlockBackend *blk;
 389     BlockJob *job;
 390     int ret;
 391
 392     if (job_id == NULL && !(flags & JOB_INTERNAL)) {
 393         job_id = bdrv_get_device_name(bs);
 394     }
 395
 396     blk = blk_new(bdrv_get_aio_context(bs), perm, shared_perm);
 397     ret = blk_insert_bs(blk, bs, errp);
 398     if (ret < 0) {
 399         blk_unref(blk);
 400         return NULL;
 401     }
 402
 403     job = job_create(job_id, &driver->job_driver, txn, blk_get_aio_context(blk),
 404                      flags, cb, opaque, errp);
 405     if (job == NULL) {
 406         blk_unref(blk);
 407         return NULL;
 408     }
 409
 410     assert(is_block_job(&job->job));
 411     assert(job->job.driver->free == &block_job_free);
 412     assert(job->job.driver->user_resume == &block_job_user_resume);
 413
 414     job->blk = blk;
 415
 416     job->finalize_cancelled_notifier.notify = block_job_event_cancelled;
 417     job->finalize_completed_notifier.notify = block_job_event_completed;
 418     job->pending_notifier.notify = block_job_event_pending;
 419     job->ready_notifier.notify = block_job_event_ready;
 420     job->idle_notifier.notify = block_job_on_idle;
 421
 422     notifier_list_add(&job->job.on_finalize_cancelled,
 423                       &job->finalize_cancelled_notifier);
 424     notifier_list_add(&job->job.on_finalize_completed,
 425                       &job->finalize_completed_notifier);
 426     notifier_list_add(&job->job.on_pending, &job->pending_notifier);
 427     notifier_list_add(&job->job.on_ready, &job->ready_notifier);
 428     notifier_list_add(&job->job.on_idle, &job->idle_notifier);
 429
 430     error_setg(&job->blocker, "block device is in use by block job: %s",
 431                job_type_str(&job->job));
 432     block_job_add_bdrv(job, "main node", bs, 0, BLK_PERM_ALL, &error_abort);
 433
 434     bdrv_op_unblock(bs, BLOCK_OP_TYPE_DATAPLANE, job->blocker);
 435
 436     /* Disable request queuing in the BlockBackend to avoid deadlocks on drain:
 437      * The job reports that it's busy until it reaches a pause point. */
 438     blk_set_disable_request_queuing(blk, true);
 439     blk_set_allow_aio_context_change(blk, true);
 440
 441     /* Only set speed when necessary to avoid NotSupported error */
 442     if (speed != 0) {
 443         Error *local_err = NULL;
 444
 445         block_job_set_speed(job, speed, &local_err);
 446         if (local_err) {
 447             job_early_fail(&job->job);
 448             error_propagate(errp, local_err);
 449             return NULL;
 450         }
 451     }
 452
 453     return job;
 454 }
 455
 456 void block_job_iostatus_reset(BlockJob *job)
 457 {
 458     if (job->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
 459         return;
 460     }
 461     assert(job->job.user_paused && job->job.pause_count > 0);
 462     job->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
 463 }
 464
 465 void block_job_user_resume(Job *job)
 466 {
 467     BlockJob *bjob = container_of(job, BlockJob, job);
 468     block_job_iostatus_reset(bjob);
 469 }
 470
 471 BlockErrorAction block_job_error_action(BlockJob *job, BlockdevOnError on_err,
 472                                         int is_read, int error)
 473 {
 474     BlockErrorAction action;
 475
 476     switch (on_err) {
 477     case BLOCKDEV_ON_ERROR_ENOSPC:
 478     case BLOCKDEV_ON_ERROR_AUTO:
 479         action = (error == ENOSPC) ?
 480                  BLOCK_ERROR_ACTION_STOP : BLOCK_ERROR_ACTION_REPORT;
 481         break;
 482     case BLOCKDEV_ON_ERROR_STOP:
 483         action = BLOCK_ERROR_ACTION_STOP;
 484         break;
 485     case BLOCKDEV_ON_ERROR_REPORT:
 486         action = BLOCK_ERROR_ACTION_REPORT;
 487         break;
 488     case BLOCKDEV_ON_ERROR_IGNORE:
 489         action = BLOCK_ERROR_ACTION_IGNORE;
 490         break;
 491     default:
 492         abort();
 493     }
 494     if (!block_job_is_internal(job)) {
 495         qapi_event_send_block_job_error(job->job.id,
 496                                         is_read ? IO_OPERATION_TYPE_READ :
 497                                         IO_OPERATION_TYPE_WRITE,
 498                                         action);
 499     }
 500     if (action == BLOCK_ERROR_ACTION_STOP) {
 501         if (!job->job.user_paused) {
 502             job_pause(&job->job);
 503             /* make the pause user visible, which will be resumed from QMP. */
 504             job->job.user_paused = true;
 505         }
 506         block_job_iostatus_set_err(job, error);
 507     }
 508     return action;
 509 }