qemu/blockjob.c
<<
>>
Prefs
   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
  54static 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
  62BlockJob *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
  73BlockJob *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
  84static void block_job_attached_aio_context(AioContext *new_context,
  85                                           void *opaque);
  86static void block_job_detach_aio_context(void *opaque);
  87
  88void 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
 102static 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
 117void 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
 129static 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
 146static 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
 152static void child_job_drained_begin(BdrvChild *c)
 153{
 154    BlockJob *job = c->opaque;
 155    job_pause(&job->job);
 156}
 157
 158static 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) || job->deferred_to_main_loop) {
 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
 180static void child_job_drained_end(BdrvChild *c)
 181{
 182    BlockJob *job = c->opaque;
 183    job_resume(&job->job);
 184}
 185
 186static 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
 194void 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
 206int 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
 224bool block_job_is_internal(BlockJob *job)
 225{
 226    return (job->job.id == NULL);
 227}
 228
 229const BlockJobDriver *block_job_driver(BlockJob *job)
 230{
 231    return container_of(job->job.driver, BlockJobDriver, job_driver);
 232}
 233
 234/* Assumes the job_mutex is held */
 235static bool job_timer_pending(Job *job)
 236{
 237    return timer_pending(&job->sleep_timer);
 238}
 239
 240void block_job_set_speed(BlockJob *job, int64_t speed, Error **errp)
 241{
 242    int64_t old_speed = job->speed;
 243
 244    if (job_apply_verb(&job->job, JOB_VERB_SET_SPEED, errp)) {
 245        return;
 246    }
 247    if (speed < 0) {
 248        error_setg(errp, QERR_INVALID_PARAMETER, "speed");
 249        return;
 250    }
 251
 252    ratelimit_set_speed(&job->limit, speed, BLOCK_JOB_SLICE_TIME);
 253
 254    job->speed = speed;
 255    if (speed && speed <= old_speed) {
 256        return;
 257    }
 258
 259    /* kick only if a timer is pending */
 260    job_enter_cond(&job->job, job_timer_pending);
 261}
 262
 263int64_t block_job_ratelimit_get_delay(BlockJob *job, uint64_t n)
 264{
 265    if (!job->speed) {
 266        return 0;
 267    }
 268
 269    return ratelimit_calculate_delay(&job->limit, n);
 270}
 271
 272BlockJobInfo *block_job_query(BlockJob *job, Error **errp)
 273{
 274    BlockJobInfo *info;
 275
 276    if (block_job_is_internal(job)) {
 277        error_setg(errp, "Cannot query QEMU internal jobs");
 278        return NULL;
 279    }
 280    info = g_new0(BlockJobInfo, 1);
 281    info->type      = g_strdup(job_type_str(&job->job));
 282    info->device    = g_strdup(job->job.id);
 283    info->busy      = atomic_read(&job->job.busy);
 284    info->paused    = job->job.pause_count > 0;
 285    info->offset    = job->job.progress_current;
 286    info->len       = job->job.progress_total;
 287    info->speed     = job->speed;
 288    info->io_status = job->iostatus;
 289    info->ready     = job_is_ready(&job->job),
 290    info->status    = job->job.status;
 291    info->auto_finalize = job->job.auto_finalize;
 292    info->auto_dismiss  = job->job.auto_dismiss;
 293    info->has_error = job->job.ret != 0;
 294    info->error     = job->job.ret ? g_strdup(strerror(-job->job.ret)) : NULL;
 295    return info;
 296}
 297
 298static void block_job_iostatus_set_err(BlockJob *job, int error)
 299{
 300    if (job->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
 301        job->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE :
 302                                          BLOCK_DEVICE_IO_STATUS_FAILED;
 303    }
 304}
 305
 306static void block_job_event_cancelled(Notifier *n, void *opaque)
 307{
 308    BlockJob *job = opaque;
 309
 310    if (block_job_is_internal(job)) {
 311        return;
 312    }
 313
 314    qapi_event_send_block_job_cancelled(job_type(&job->job),
 315                                        job->job.id,
 316                                        job->job.progress_total,
 317                                        job->job.progress_current,
 318                                        job->speed,
 319                                        &error_abort);
 320}
 321
 322static 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                                        &error_abort);
 343}
 344
 345static void block_job_event_pending(Notifier *n, void *opaque)
 346{
 347    BlockJob *job = opaque;
 348
 349    if (block_job_is_internal(job)) {
 350        return;
 351    }
 352
 353    qapi_event_send_block_job_pending(job_type(&job->job),
 354                                      job->job.id,
 355                                      &error_abort);
 356}
 357
 358static void block_job_event_ready(Notifier *n, void *opaque)
 359{
 360    BlockJob *job = opaque;
 361
 362    if (block_job_is_internal(job)) {
 363        return;
 364    }
 365
 366    qapi_event_send_block_job_ready(job_type(&job->job),
 367                                    job->job.id,
 368                                    job->job.progress_total,
 369                                    job->job.progress_current,
 370                                    job->speed, &error_abort);
 371}
 372
 373
 374/*
 375 * API for block job drivers and the block layer.  These functions are
 376 * declared in blockjob_int.h.
 377 */
 378
 379void *block_job_create(const char *job_id, const BlockJobDriver *driver,
 380                       JobTxn *txn, BlockDriverState *bs, uint64_t perm,
 381                       uint64_t shared_perm, int64_t speed, int flags,
 382                       BlockCompletionFunc *cb, void *opaque, Error **errp)
 383{
 384    BlockBackend *blk;
 385    BlockJob *job;
 386    int ret;
 387
 388    if (bs->job) {
 389        error_setg(errp, QERR_DEVICE_IN_USE, bdrv_get_device_name(bs));
 390        return NULL;
 391    }
 392
 393    if (job_id == NULL && !(flags & JOB_INTERNAL)) {
 394        job_id = bdrv_get_device_name(bs);
 395    }
 396
 397    blk = blk_new(perm, shared_perm);
 398    ret = blk_insert_bs(blk, bs, errp);
 399    if (ret < 0) {
 400        blk_unref(blk);
 401        return NULL;
 402    }
 403
 404    job = job_create(job_id, &driver->job_driver, txn, blk_get_aio_context(blk),
 405                     flags, cb, opaque, errp);
 406    if (job == NULL) {
 407        blk_unref(blk);
 408        return NULL;
 409    }
 410
 411    assert(is_block_job(&job->job));
 412    assert(job->job.driver->free == &block_job_free);
 413    assert(job->job.driver->user_resume == &block_job_user_resume);
 414    assert(job->job.driver->drain == &block_job_drain);
 415
 416    job->blk = blk;
 417
 418    job->finalize_cancelled_notifier.notify = block_job_event_cancelled;
 419    job->finalize_completed_notifier.notify = block_job_event_completed;
 420    job->pending_notifier.notify = block_job_event_pending;
 421    job->ready_notifier.notify = block_job_event_ready;
 422
 423    notifier_list_add(&job->job.on_finalize_cancelled,
 424                      &job->finalize_cancelled_notifier);
 425    notifier_list_add(&job->job.on_finalize_completed,
 426                      &job->finalize_completed_notifier);
 427    notifier_list_add(&job->job.on_pending, &job->pending_notifier);
 428    notifier_list_add(&job->job.on_ready, &job->ready_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_add_aio_context_notifier(blk, block_job_attached_aio_context,
 438                                 block_job_detach_aio_context, job);
 439
 440    /* Only set speed when necessary to avoid NotSupported error */
 441    if (speed != 0) {
 442        Error *local_err = NULL;
 443
 444        block_job_set_speed(job, speed, &local_err);
 445        if (local_err) {
 446            job_early_fail(&job->job);
 447            error_propagate(errp, local_err);
 448            return NULL;
 449        }
 450    }
 451
 452    return job;
 453}
 454
 455void block_job_iostatus_reset(BlockJob *job)
 456{
 457    if (job->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
 458        return;
 459    }
 460    assert(job->job.user_paused && job->job.pause_count > 0);
 461    job->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
 462}
 463
 464void block_job_user_resume(Job *job)
 465{
 466    BlockJob *bjob = container_of(job, BlockJob, job);
 467    block_job_iostatus_reset(bjob);
 468}
 469
 470BlockErrorAction block_job_error_action(BlockJob *job, BlockdevOnError on_err,
 471                                        int is_read, int error)
 472{
 473    BlockErrorAction action;
 474
 475    switch (on_err) {
 476    case BLOCKDEV_ON_ERROR_ENOSPC:
 477    case BLOCKDEV_ON_ERROR_AUTO:
 478        action = (error == ENOSPC) ?
 479                 BLOCK_ERROR_ACTION_STOP : BLOCK_ERROR_ACTION_REPORT;
 480        break;
 481    case BLOCKDEV_ON_ERROR_STOP:
 482        action = BLOCK_ERROR_ACTION_STOP;
 483        break;
 484    case BLOCKDEV_ON_ERROR_REPORT:
 485        action = BLOCK_ERROR_ACTION_REPORT;
 486        break;
 487    case BLOCKDEV_ON_ERROR_IGNORE:
 488        action = BLOCK_ERROR_ACTION_IGNORE;
 489        break;
 490    default:
 491        abort();
 492    }
 493    if (!block_job_is_internal(job)) {
 494        qapi_event_send_block_job_error(job->job.id,
 495                                        is_read ? IO_OPERATION_TYPE_READ :
 496                                        IO_OPERATION_TYPE_WRITE,
 497                                        action, &error_abort);
 498    }
 499    if (action == BLOCK_ERROR_ACTION_STOP) {
 500        job_pause(&job->job);
 501        /* make the pause user visible, which will be resumed from QMP. */
 502        job->job.user_paused = true;
 503        block_job_iostatus_set_err(job, error);
 504    }
 505    return action;
 506}
 507