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