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 static void block_job_attached_aio_context(AioContext *new_context,
  85                                            void *opaque);
  86 static void block_job_detach_aio_context(void *opaque);
  87
  88 void block_job_free(Job *job)
  89 {
  90     BlockJob *bjob = container_of(job, BlockJob, job);
  91     BlockDriverState *bs = blk_bs(bjob->blk);
  92
  93     bs->job = NULL;
  94     block_job_remove_all_bdrv(bjob);
  95     blk_remove_aio_context_notifier(bjob->blk,
  96                                     block_job_attached_aio_context,
  97                                     block_job_detach_aio_context, bjob);
  98     blk_unref(bjob->blk);
  99     error_free(bjob->blocker);
 100 }
 101
 102 static void block_job_attached_aio_context(AioContext *new_context,
 103                                            void *opaque)
 104 {
 105     BlockJob *job = opaque;
 106     const JobDriver *drv = job->job.driver;
 107     BlockJobDriver *bjdrv = container_of(drv, BlockJobDriver, job_driver);
 108
 109     job->job.aio_context = new_context;
 110     if (bjdrv->attached_aio_context) {
 111         bjdrv->attached_aio_context(job, new_context);
 112     }
 113
 114     job_resume(&job->job);
 115 }
 116
 117 void block_job_drain(Job *job)
 118 {
 119     BlockJob *bjob = container_of(job, BlockJob, job);
 120     const JobDriver *drv = job->driver;
 121     BlockJobDriver *bjdrv = container_of(drv, BlockJobDriver, job_driver);
 122
 123     blk_drain(bjob->blk);
 124     if (bjdrv->drain) {
 125         bjdrv->drain(bjob);
 126     }
 127 }
 128
 129 static void block_job_detach_aio_context(void *opaque)
 130 {
 131     BlockJob *job = opaque;
 132
 133     /* In case the job terminates during aio_poll()... */
 134     job_ref(&job->job);
 135
 136     job_pause(&job->job);
 137
 138     while (!job->job.paused && !job_is_completed(&job->job)) {
 139         job_drain(&job->job);
 140     }
 141
 142     job->job.aio_context = NULL;
 143     job_unref(&job->job);
 144 }
 145
 146 static char *child_job_get_parent_desc(BdrvChild *c)
 147 {
 148     BlockJob *job = c->opaque;
 149     return g_strdup_printf("%s job '%s'", job_type_str(&job->job), job->job.id);
 150 }
 151
 152 static void child_job_drained_begin(BdrvChild *c)
 153 {
 154     BlockJob *job = c->opaque;
 155     job_pause(&job->job);
 156 }
 157
 158 static bool child_job_drained_poll(BdrvChild *c)
 159 {
 160     BlockJob *bjob = c->opaque;
 161     Job *job = &bjob->job;
 162     const BlockJobDriver *drv = block_job_driver(bjob);
 163
 164     /* An inactive or completed job doesn't have any pending requests. Jobs
 165      * with !job->busy are either already paused or have a pause point after
 166      * being reentered, so no job driver code will run before they pause. */
 167     if (!job->busy || job_is_completed(job)) {
 168         return false;
 169     }
 170
 171     /* Otherwise, assume that it isn't fully stopped yet, but allow the job to
 172      * override this assumption. */
 173     if (drv->drained_poll) {
 174         return drv->drained_poll(bjob);
 175     } else {
 176         return true;
 177     }
 178 }
 179
 180 static void child_job_drained_end(BdrvChild *c)
 181 {
 182     BlockJob *job = c->opaque;
 183     job_resume(&job->job);
 184 }
 185
 186 static const BdrvChildRole child_job = {
 187     .get_parent_desc    = child_job_get_parent_desc,
 188     .drained_begin      = child_job_drained_begin,
 189     .drained_poll       = child_job_drained_poll,
 190     .drained_end        = child_job_drained_end,
 191     .stay_at_node       = true,
 192 };
 193
 194 void block_job_remove_all_bdrv(BlockJob *job)
 195 {
 196     GSList *l;
 197     for (l = job->nodes; l; l = l->next) {
 198         BdrvChild *c = l->data;
 199         bdrv_op_unblock_all(c->bs, job->blocker);
 200         bdrv_root_unref_child(c);
 201     }
 202     g_slist_free(job->nodes);
 203     job->nodes = NULL;
 204 }
 205
 206 int block_job_add_bdrv(BlockJob *job, const char *name, BlockDriverState *bs,
 207                        uint64_t perm, uint64_t shared_perm, Error **errp)
 208 {
 209     BdrvChild *c;
 210
 211     c = bdrv_root_attach_child(bs, name, &child_job, perm, shared_perm,
 212                                job, errp);
 213     if (c == NULL) {
 214         return -EPERM;
 215     }
 216
 217     job->nodes = g_slist_prepend(job->nodes, c);
 218     bdrv_ref(bs);
 219     bdrv_op_block_all(bs, job->blocker);
 220
 221     return 0;
 222 }
 223
 224 static void block_job_on_idle(Notifier *n, void *opaque)
 225 {
 226     aio_wait_kick();
 227 }
 228
 229 bool block_job_is_internal(BlockJob *job)
 230 {
 231     return (job->job.id == NULL);
 232 }
 233
 234 const BlockJobDriver *block_job_driver(BlockJob *job)
 235 {
 236     return container_of(job->job.driver, BlockJobDriver, job_driver);
 237 }
 238
 239 /* Assumes the job_mutex is held */
 240 static bool job_timer_pending(Job *job)
 241 {
 242     return timer_pending(&job->sleep_timer);
 243 }
 244
 245 void block_job_set_speed(BlockJob *job, int64_t speed, Error **errp)
 246 {
 247     int64_t old_speed = job->speed;
 248
 249     if (job_apply_verb(&job->job, JOB_VERB_SET_SPEED, errp)) {
 250         return;
 251     }
 252     if (speed < 0) {
 253         error_setg(errp, QERR_INVALID_PARAMETER, "speed");
 254         return;
 255     }
 256
 257     ratelimit_set_speed(&job->limit, speed, BLOCK_JOB_SLICE_TIME);
 258
 259     job->speed = speed;
 260     if (speed && speed <= old_speed) {
 261         return;
 262     }
 263
 264     /* kick only if a timer is pending */
 265     job_enter_cond(&job->job, job_timer_pending);
 266 }
 267
 268 int64_t block_job_ratelimit_get_delay(BlockJob *job, uint64_t n)
 269 {
 270     if (!job->speed) {
 271         return 0;
 272     }
 273
 274     return ratelimit_calculate_delay(&job->limit, n);
 275 }
 276
 277 BlockJobInfo *block_job_query(BlockJob *job, Error **errp)
 278 {
 279     BlockJobInfo *info;
 280
 281     if (block_job_is_internal(job)) {
 282         error_setg(errp, "Cannot query QEMU internal jobs");
 283         return NULL;
 284     }
 285     info = g_new0(BlockJobInfo, 1);
 286     info->type      = g_strdup(job_type_str(&job->job));
 287     info->device    = g_strdup(job->job.id);
 288     info->busy      = atomic_read(&job->job.busy);
 289     info->paused    = job->job.pause_count > 0;
 290     info->offset    = job->job.progress_current;
 291     info->len       = job->job.progress_total;
 292     info->speed     = job->speed;
 293     info->io_status = job->iostatus;
 294     info->ready     = job_is_ready(&job->job),
 295     info->status    = job->job.status;
 296     info->auto_finalize = job->job.auto_finalize;
 297     info->auto_dismiss  = job->job.auto_dismiss;
 298     info->has_error = job->job.ret != 0;
 299     info->error     = job->job.ret ? g_strdup(strerror(-job->job.ret)) : NULL;
 300     return info;
 301 }
 302
 303 static void block_job_iostatus_set_err(BlockJob *job, int error)
 304 {
 305     if (job->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
 306         job->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE :
 307                                           BLOCK_DEVICE_IO_STATUS_FAILED;
 308     }
 309 }
 310
 311 static void block_job_event_cancelled(Notifier *n, void *opaque)
 312 {
 313     BlockJob *job = opaque;
 314
 315     if (block_job_is_internal(job)) {
 316         return;
 317     }
 318
 319     qapi_event_send_block_job_cancelled(job_type(&job->job),
 320                                         job->job.id,
 321                                         job->job.progress_total,
 322                                         job->job.progress_current,
 323                                         job->speed);
 324 }
 325
 326 static void block_job_event_completed(Notifier *n, void *opaque)
 327 {
 328     BlockJob *job = opaque;
 329     const char *msg = NULL;
 330
 331     if (block_job_is_internal(job)) {
 332         return;
 333     }
 334
 335     if (job->job.ret < 0) {
 336         msg = strerror(-job->job.ret);
 337     }
 338
 339     qapi_event_send_block_job_completed(job_type(&job->job),
 340                                         job->job.id,
 341                                         job->job.progress_total,
 342                                         job->job.progress_current,
 343                                         job->speed,
 344                                         !!msg,
 345                                         msg);
 346 }
 347
 348 static void block_job_event_pending(Notifier *n, void *opaque)
 349 {
 350     BlockJob *job = opaque;
 351
 352     if (block_job_is_internal(job)) {
 353         return;
 354     }
 355
 356     qapi_event_send_block_job_pending(job_type(&job->job),
 357                                       job->job.id);
 358 }
 359
 360 static void block_job_event_ready(Notifier *n, void *opaque)
 361 {
 362     BlockJob *job = opaque;
 363
 364     if (block_job_is_internal(job)) {
 365         return;
 366     }
 367
 368     qapi_event_send_block_job_ready(job_type(&job->job),
 369                                     job->job.id,
 370                                     job->job.progress_total,
 371                                     job->job.progress_current,
 372                                     job->speed);
 373 }
 374
 375
 376 /*
 377  * API for block job drivers and the block layer.  These functions are
 378  * declared in blockjob_int.h.
 379  */
 380
 381 void *block_job_create(const char *job_id, const BlockJobDriver *driver,
 382                        JobTxn *txn, BlockDriverState *bs, uint64_t perm,
 383                        uint64_t shared_perm, int64_t speed, int flags,
 384                        BlockCompletionFunc *cb, void *opaque, Error **errp)
 385 {
 386     BlockBackend *blk;
 387     BlockJob *job;
 388     int ret;
 389
 390     if (bs->job) {
 391         error_setg(errp, QERR_DEVICE_IN_USE, bdrv_get_device_name(bs));
 392         return NULL;
 393     }
 394
 395     if (job_id == NULL && !(flags & JOB_INTERNAL)) {
 396         job_id = bdrv_get_device_name(bs);
 397     }
 398
 399     blk = blk_new(perm, shared_perm);
 400     ret = blk_insert_bs(blk, bs, errp);
 401     if (ret < 0) {
 402         blk_unref(blk);
 403         return NULL;
 404     }
 405
 406     job = job_create(job_id, &driver->job_driver, txn, blk_get_aio_context(blk),
 407                      flags, cb, opaque, errp);
 408     if (job == NULL) {
 409         blk_unref(blk);
 410         return NULL;
 411     }
 412
 413     assert(is_block_job(&job->job));
 414     assert(job->job.driver->free == &block_job_free);
 415     assert(job->job.driver->user_resume == &block_job_user_resume);
 416     assert(job->job.driver->drain == &block_job_drain);
 417
 418     job->blk = blk;
 419
 420     job->finalize_cancelled_notifier.notify = block_job_event_cancelled;
 421     job->finalize_completed_notifier.notify = block_job_event_completed;
 422     job->pending_notifier.notify = block_job_event_pending;
 423     job->ready_notifier.notify = block_job_event_ready;
 424     job->idle_notifier.notify = block_job_on_idle;
 425
 426     notifier_list_add(&job->job.on_finalize_cancelled,
 427                       &job->finalize_cancelled_notifier);
 428     notifier_list_add(&job->job.on_finalize_completed,
 429                       &job->finalize_completed_notifier);
 430     notifier_list_add(&job->job.on_pending, &job->pending_notifier);
 431     notifier_list_add(&job->job.on_ready, &job->ready_notifier);
 432     notifier_list_add(&job->job.on_idle, &job->idle_notifier);
 433
 434     error_setg(&job->blocker, "block device is in use by block job: %s",
 435                job_type_str(&job->job));
 436     block_job_add_bdrv(job, "main node", bs, 0, BLK_PERM_ALL, &error_abort);
 437     bs->job = job;
 438
 439     bdrv_op_unblock(bs, BLOCK_OP_TYPE_DATAPLANE, job->blocker);
 440
 441     blk_add_aio_context_notifier(blk, block_job_attached_aio_context,
 442                                  block_job_detach_aio_context, job);
 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         job_pause(&job->job);
 505         /* make the pause user visible, which will be resumed from QMP. */
 506         job->job.user_paused = true;
 507         block_job_iostatus_set_err(job, error);
 508     }
 509     return action;
 510 }