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/timer.h"
  37
  38 /*
  39  * The block job API is composed of two categories of functions.
  40  *
  41  * The first includes functions used by the monitor.  The monitor is
  42  * peculiar in that it accesses the block job list with block_job_get, and
  43  * therefore needs consistency across block_job_get and the actual operation
  44  * (e.g. block_job_set_speed).  The consistency is achieved with
  45  * aio_context_acquire/release.  These functions are declared in blockjob.h.
  46  *
  47  * The second includes functions used by the block job drivers and sometimes
  48  * by the core block layer.  These do not care about locking, because the
  49  * whole coroutine runs under the AioContext lock, and are declared in
  50  * blockjob_int.h.
  51  */
  52
  53 static bool is_block_job(Job *job)
  54 {
  55     return job_type(job) == JOB_TYPE_BACKUP ||
  56            job_type(job) == JOB_TYPE_COMMIT ||
  57            job_type(job) == JOB_TYPE_MIRROR ||
  58            job_type(job) == JOB_TYPE_STREAM;
  59 }
  60
  61 BlockJob *block_job_next(BlockJob *bjob)
  62 {
  63     Job *job = bjob ? &bjob->job : NULL;
  64
  65     do {
  66         job = job_next(job);
  67     } while (job && !is_block_job(job));
  68
  69     return job ? container_of(job, BlockJob, job) : NULL;
  70 }
  71
  72 BlockJob *block_job_get(const char *id)
  73 {
  74     Job *job = job_get(id);
  75
  76     if (job && is_block_job(job)) {
  77         return container_of(job, BlockJob, job);
  78     } else {
  79         return NULL;
  80     }
  81 }
  82
  83 void block_job_free(Job *job)
  84 {
  85     BlockJob *bjob = container_of(job, BlockJob, job);
  86     BlockDriverState *bs = blk_bs(bjob->blk);
  87
  88     bs->job = NULL;
  89     block_job_remove_all_bdrv(bjob);
  90     blk_unref(bjob->blk);
  91     error_free(bjob->blocker);
  92 }
  93
  94 void block_job_drain(Job *job)
  95 {
  96     BlockJob *bjob = container_of(job, BlockJob, job);
  97     const JobDriver *drv = job->driver;
  98     BlockJobDriver *bjdrv = container_of(drv, BlockJobDriver, job_driver);
  99
 100     blk_drain(bjob->blk);
 101     if (bjdrv->drain) {
 102         bjdrv->drain(bjob);
 103     }
 104 }
 105
 106 static char *child_job_get_parent_desc(BdrvChild *c)
 107 {
 108     BlockJob *job = c->opaque;
 109     return g_strdup_printf("%s job '%s'", job_type_str(&job->job), job->job.id);
 110 }
 111
 112 static void child_job_drained_begin(BdrvChild *c)
 113 {
 114     BlockJob *job = c->opaque;
 115     job_pause(&job->job);
 116 }
 117
 118 static bool child_job_drained_poll(BdrvChild *c)
 119 {
 120     BlockJob *bjob = c->opaque;
 121     Job *job = &bjob->job;
 122     const BlockJobDriver *drv = block_job_driver(bjob);
 123
 124     /* An inactive or completed job doesn't have any pending requests. Jobs
 125      * with !job->busy are either already paused or have a pause point after
 126      * being reentered, so no job driver code will run before they pause. */
 127     if (!job->busy || job_is_completed(job)) {
 128         return false;
 129     }
 130
 131     /* Otherwise, assume that it isn't fully stopped yet, but allow the job to
 132      * override this assumption. */
 133     if (drv->drained_poll) {
 134         return drv->drained_poll(bjob);
 135     } else {
 136         return true;
 137     }
 138 }
 139
 140 static void child_job_drained_end(BdrvChild *c)
 141 {
 142     BlockJob *job = c->opaque;
 143     job_resume(&job->job);
 144 }
 145
 146 static bool child_job_can_set_aio_ctx(BdrvChild *c, AioContext *ctx,
 147                                       GSList **ignore, Error **errp)
 148 {
 149     BlockJob *job = c->opaque;
 150     GSList *l;
 151
 152     for (l = job->nodes; l; l = l->next) {
 153         BdrvChild *sibling = l->data;
 154         if (!bdrv_child_can_set_aio_context(sibling, ctx, ignore, errp)) {
 155             return false;
 156         }
 157     }
 158     return true;
 159 }
 160
 161 static void child_job_set_aio_ctx(BdrvChild *c, AioContext *ctx,
 162                                   GSList **ignore)
 163 {
 164     BlockJob *job = c->opaque;
 165     GSList *l;
 166
 167     for (l = job->nodes; l; l = l->next) {
 168         BdrvChild *sibling = l->data;
 169         if (g_slist_find(*ignore, sibling)) {
 170             continue;
 171         }
 172         *ignore = g_slist_prepend(*ignore, sibling);
 173         bdrv_set_aio_context_ignore(sibling->bs, ctx, ignore);
 174     }
 175
 176     job->job.aio_context = ctx;
 177 }
 178
 179 static const BdrvChildRole child_job = {
 180     .get_parent_desc    = child_job_get_parent_desc,
 181     .drained_begin      = child_job_drained_begin,
 182     .drained_poll       = child_job_drained_poll,
 183     .drained_end        = child_job_drained_end,
 184     .can_set_aio_ctx    = child_job_can_set_aio_ctx,
 185     .set_aio_ctx        = child_job_set_aio_ctx,
 186     .stay_at_node       = true,
 187 };
 188
 189 void block_job_remove_all_bdrv(BlockJob *job)
 190 {
 191     GSList *l;
 192     for (l = job->nodes; l; l = l->next) {
 193         BdrvChild *c = l->data;
 194         bdrv_op_unblock_all(c->bs, job->blocker);
 195         bdrv_root_unref_child(c);
 196     }
 197     g_slist_free(job->nodes);
 198     job->nodes = NULL;
 199 }
 200
 201 int block_job_add_bdrv(BlockJob *job, const char *name, BlockDriverState *bs,
 202                        uint64_t perm, uint64_t shared_perm, Error **errp)
 203 {
 204     BdrvChild *c;
 205
 206     bdrv_ref(bs);
 207     if (job->job.aio_context != qemu_get_aio_context()) {
 208         aio_context_release(job->job.aio_context);
 209     }
 210     c = bdrv_root_attach_child(bs, name, &child_job, job->job.aio_context,
 211                                perm, shared_perm, job, errp);
 212     if (job->job.aio_context != qemu_get_aio_context()) {
 213         aio_context_acquire(job->job.aio_context);
 214     }
 215     if (c == NULL) {
 216         return -EPERM;
 217     }
 218
 219     job->nodes = g_slist_prepend(job->nodes, c);
 220     bdrv_op_block_all(bs, job->blocker);
 221
 222     return 0;
 223 }
 224
 225 static void block_job_on_idle(Notifier *n, void *opaque)
 226 {
 227     aio_wait_kick();
 228 }
 229
 230 bool block_job_is_internal(BlockJob *job)
 231 {
 232     return (job->job.id == NULL);
 233 }
 234
 235 const BlockJobDriver *block_job_driver(BlockJob *job)
 236 {
 237     return container_of(job->job.driver, BlockJobDriver, job_driver);
 238 }
 239
 240 /* Assumes the job_mutex is held */
 241 static bool job_timer_pending(Job *job)
 242 {
 243     return timer_pending(&job->sleep_timer);
 244 }
 245
 246 void block_job_set_speed(BlockJob *job, int64_t speed, Error **errp)
 247 {
 248     int64_t old_speed = job->speed;
 249
 250     if (job_apply_verb(&job->job, JOB_VERB_SET_SPEED, errp)) {
 251         return;
 252     }
 253     if (speed < 0) {
 254         error_setg(errp, QERR_INVALID_PARAMETER, "speed");
 255         return;
 256     }
 257
 258     ratelimit_set_speed(&job->limit, speed, BLOCK_JOB_SLICE_TIME);
 259
 260     job->speed = speed;
 261     if (speed && speed <= old_speed) {
 262         return;
 263     }
 264
 265     /* kick only if a timer is pending */
 266     job_enter_cond(&job->job, job_timer_pending);
 267 }
 268
 269 int64_t block_job_ratelimit_get_delay(BlockJob *job, uint64_t n)
 270 {
 271     if (!job->speed) {
 272         return 0;
 273     }
 274
 275     return ratelimit_calculate_delay(&job->limit, n);
 276 }
 277
 278 BlockJobInfo *block_job_query(BlockJob *job, Error **errp)
 279 {
 280     BlockJobInfo *info;
 281
 282     if (block_job_is_internal(job)) {
 283         error_setg(errp, "Cannot query QEMU internal jobs");
 284         return NULL;
 285     }
 286     info = g_new0(BlockJobInfo, 1);
 287     info->type      = g_strdup(job_type_str(&job->job));
 288     info->device    = g_strdup(job->job.id);
 289     info->busy      = atomic_read(&job->job.busy);
 290     info->paused    = job->job.pause_count > 0;
 291     info->offset    = job->job.progress_current;
 292     info->len       = job->job.progress_total;
 293     info->speed     = job->speed;
 294     info->io_status = job->iostatus;
 295     info->ready     = job_is_ready(&job->job),
 296     info->status    = job->job.status;
 297     info->auto_finalize = job->job.auto_finalize;
 298     info->auto_dismiss  = job->job.auto_dismiss;
 299     info->has_error = job->job.ret != 0;
 300     info->error     = job->job.ret ? g_strdup(strerror(-job->job.ret)) : NULL;
 301     return info;
 302 }
 303
 304 static void block_job_iostatus_set_err(BlockJob *job, int error)
 305 {
 306     if (job->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
 307         job->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE :
 308                                           BLOCK_DEVICE_IO_STATUS_FAILED;
 309     }
 310 }
 311
 312 static void block_job_event_cancelled(Notifier *n, void *opaque)
 313 {
 314     BlockJob *job = opaque;
 315
 316     if (block_job_is_internal(job)) {
 317         return;
 318     }
 319
 320     qapi_event_send_block_job_cancelled(job_type(&job->job),
 321                                         job->job.id,
 322                                         job->job.progress_total,
 323                                         job->job.progress_current,
 324                                         job->speed);
 325 }
 326
 327 static void block_job_event_completed(Notifier *n, void *opaque)
 328 {
 329     BlockJob *job = opaque;
 330     const char *msg = NULL;
 331
 332     if (block_job_is_internal(job)) {
 333         return;
 334     }
 335
 336     if (job->job.ret < 0) {
 337         msg = strerror(-job->job.ret);
 338     }
 339
 340     qapi_event_send_block_job_completed(job_type(&job->job),
 341                                         job->job.id,
 342                                         job->job.progress_total,
 343                                         job->job.progress_current,
 344                                         job->speed,
 345                                         !!msg,
 346                                         msg);
 347 }
 348
 349 static void block_job_event_pending(Notifier *n, void *opaque)
 350 {
 351     BlockJob *job = opaque;
 352
 353     if (block_job_is_internal(job)) {
 354         return;
 355     }
 356
 357     qapi_event_send_block_job_pending(job_type(&job->job),
 358                                       job->job.id);
 359 }
 360
 361 static void block_job_event_ready(Notifier *n, void *opaque)
 362 {
 363     BlockJob *job = opaque;
 364
 365     if (block_job_is_internal(job)) {
 366         return;
 367     }
 368
 369     qapi_event_send_block_job_ready(job_type(&job->job),
 370                                     job->job.id,
 371                                     job->job.progress_total,
 372                                     job->job.progress_current,
 373                                     job->speed);
 374 }
 375
 376
 377 /*
 378  * API for block job drivers and the block layer.  These functions are
 379  * declared in blockjob_int.h.
 380  */
 381
 382 void *block_job_create(const char *job_id, const BlockJobDriver *driver,
 383                        JobTxn *txn, BlockDriverState *bs, uint64_t perm,
 384                        uint64_t shared_perm, int64_t speed, int flags,
 385                        BlockCompletionFunc *cb, void *opaque, Error **errp)
 386 {
 387     BlockBackend *blk;
 388     BlockJob *job;
 389     int ret;
 390
 391     if (bs->job) {
 392         error_setg(errp, QERR_DEVICE_IN_USE, bdrv_get_device_name(bs));
 393         return NULL;
 394     }
 395
 396     if (job_id == NULL && !(flags & JOB_INTERNAL)) {
 397         job_id = bdrv_get_device_name(bs);
 398     }
 399
 400     blk = blk_new(bdrv_get_aio_context(bs), perm, shared_perm);
 401     ret = blk_insert_bs(blk, bs, errp);
 402     if (ret < 0) {
 403         blk_unref(blk);
 404         return NULL;
 405     }
 406
 407     job = job_create(job_id, &driver->job_driver, txn, blk_get_aio_context(blk),
 408                      flags, cb, opaque, errp);
 409     if (job == NULL) {
 410         blk_unref(blk);
 411         return NULL;
 412     }
 413
 414     assert(is_block_job(&job->job));
 415     assert(job->job.driver->free == &block_job_free);
 416     assert(job->job.driver->user_resume == &block_job_user_resume);
 417     assert(job->job.driver->drain == &block_job_drain);
 418
 419     job->blk = blk;
 420
 421     job->finalize_cancelled_notifier.notify = block_job_event_cancelled;
 422     job->finalize_completed_notifier.notify = block_job_event_completed;
 423     job->pending_notifier.notify = block_job_event_pending;
 424     job->ready_notifier.notify = block_job_event_ready;
 425     job->idle_notifier.notify = block_job_on_idle;
 426
 427     notifier_list_add(&job->job.on_finalize_cancelled,
 428                       &job->finalize_cancelled_notifier);
 429     notifier_list_add(&job->job.on_finalize_completed,
 430                       &job->finalize_completed_notifier);
 431     notifier_list_add(&job->job.on_pending, &job->pending_notifier);
 432     notifier_list_add(&job->job.on_ready, &job->ready_notifier);
 433     notifier_list_add(&job->job.on_idle, &job->idle_notifier);
 434
 435     error_setg(&job->blocker, "block device is in use by block job: %s",
 436                job_type_str(&job->job));
 437     block_job_add_bdrv(job, "main node", bs, 0, BLK_PERM_ALL, &error_abort);
 438     bs->job = job;
 439
 440     bdrv_op_unblock(bs, BLOCK_OP_TYPE_DATAPLANE, job->blocker);
 441
 442     blk_set_allow_aio_context_change(blk, true);
 443
 444     /* Only set speed when necessary to avoid NotSupported error */
 445     if (speed != 0) {
 446         Error *local_err = NULL;
 447
 448         block_job_set_speed(job, speed, &local_err);
 449         if (local_err) {
 450             job_early_fail(&job->job);
 451             error_propagate(errp, local_err);
 452             return NULL;
 453         }
 454     }
 455
 456     return job;
 457 }
 458
 459 void block_job_iostatus_reset(BlockJob *job)
 460 {
 461     if (job->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
 462         return;
 463     }
 464     assert(job->job.user_paused && job->job.pause_count > 0);
 465     job->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
 466 }
 467
 468 void block_job_user_resume(Job *job)
 469 {
 470     BlockJob *bjob = container_of(job, BlockJob, job);
 471     block_job_iostatus_reset(bjob);
 472 }
 473
 474 BlockErrorAction block_job_error_action(BlockJob *job, BlockdevOnError on_err,
 475                                         int is_read, int error)
 476 {
 477     BlockErrorAction action;
 478
 479     switch (on_err) {
 480     case BLOCKDEV_ON_ERROR_ENOSPC:
 481     case BLOCKDEV_ON_ERROR_AUTO:
 482         action = (error == ENOSPC) ?
 483                  BLOCK_ERROR_ACTION_STOP : BLOCK_ERROR_ACTION_REPORT;
 484         break;
 485     case BLOCKDEV_ON_ERROR_STOP:
 486         action = BLOCK_ERROR_ACTION_STOP;
 487         break;
 488     case BLOCKDEV_ON_ERROR_REPORT:
 489         action = BLOCK_ERROR_ACTION_REPORT;
 490         break;
 491     case BLOCKDEV_ON_ERROR_IGNORE:
 492         action = BLOCK_ERROR_ACTION_IGNORE;
 493         break;
 494     default:
 495         abort();
 496     }
 497     if (!block_job_is_internal(job)) {
 498         qapi_event_send_block_job_error(job->job.id,
 499                                         is_read ? IO_OPERATION_TYPE_READ :
 500                                         IO_OPERATION_TYPE_WRITE,
 501                                         action);
 502     }
 503     if (action == BLOCK_ERROR_ACTION_STOP) {
 504         if (!job->job.user_paused) {
 505             job_pause(&job->job);
 506             /* make the pause user visible, which will be resumed from QMP. */
 507             job->job.user_paused = true;
 508         }
 509         block_job_iostatus_set_err(job, error);
 510     }
 511     return action;
 512 }