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 "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
  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
  84void 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
  93static 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
  99static void child_job_drained_begin(BdrvChild *c)
 100{
 101    BlockJob *job = c->opaque;
 102    job_pause(&job->job);
 103}
 104
 105static 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
 127static void child_job_drained_end(BdrvChild *c, int *drained_end_counter)
 128{
 129    BlockJob *job = c->opaque;
 130    job_resume(&job->job);
 131}
 132
 133static 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
 148static 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
 166static 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
 176void block_job_remove_all_bdrv(BlockJob *job)
 177{
 178    /*
 179     * bdrv_root_unref_child() may reach child_job_[can_]set_aio_ctx(),
 180     * which will also traverse job->nodes, so consume the list one by
 181     * one to make sure that such a concurrent access does not attempt
 182     * to process an already freed BdrvChild.
 183     */
 184    while (job->nodes) {
 185        GSList *l = job->nodes;
 186        BdrvChild *c = l->data;
 187
 188        job->nodes = l->next;
 189
 190        bdrv_op_unblock_all(c->bs, job->blocker);
 191        bdrv_root_unref_child(c);
 192
 193        g_slist_free_1(l);
 194    }
 195}
 196
 197bool block_job_has_bdrv(BlockJob *job, BlockDriverState *bs)
 198{
 199    GSList *el;
 200
 201    for (el = job->nodes; el; el = el->next) {
 202        BdrvChild *c = el->data;
 203        if (c->bs == bs) {
 204            return true;
 205        }
 206    }
 207
 208    return false;
 209}
 210
 211int block_job_add_bdrv(BlockJob *job, const char *name, BlockDriverState *bs,
 212                       uint64_t perm, uint64_t shared_perm, Error **errp)
 213{
 214    BdrvChild *c;
 215
 216    bdrv_ref(bs);
 217    if (job->job.aio_context != qemu_get_aio_context()) {
 218        aio_context_release(job->job.aio_context);
 219    }
 220    c = bdrv_root_attach_child(bs, name, &child_job, job->job.aio_context,
 221                               perm, shared_perm, job, errp);
 222    if (job->job.aio_context != qemu_get_aio_context()) {
 223        aio_context_acquire(job->job.aio_context);
 224    }
 225    if (c == NULL) {
 226        return -EPERM;
 227    }
 228
 229    job->nodes = g_slist_prepend(job->nodes, c);
 230    bdrv_op_block_all(bs, job->blocker);
 231
 232    return 0;
 233}
 234
 235static void block_job_on_idle(Notifier *n, void *opaque)
 236{
 237    aio_wait_kick();
 238}
 239
 240bool block_job_is_internal(BlockJob *job)
 241{
 242    return (job->job.id == NULL);
 243}
 244
 245const BlockJobDriver *block_job_driver(BlockJob *job)
 246{
 247    return container_of(job->job.driver, BlockJobDriver, job_driver);
 248}
 249
 250/* Assumes the job_mutex is held */
 251static bool job_timer_pending(Job *job)
 252{
 253    return timer_pending(&job->sleep_timer);
 254}
 255
 256void block_job_set_speed(BlockJob *job, int64_t speed, Error **errp)
 257{
 258    int64_t old_speed = job->speed;
 259
 260    if (job_apply_verb(&job->job, JOB_VERB_SET_SPEED, errp)) {
 261        return;
 262    }
 263    if (speed < 0) {
 264        error_setg(errp, QERR_INVALID_PARAMETER, "speed");
 265        return;
 266    }
 267
 268    ratelimit_set_speed(&job->limit, speed, BLOCK_JOB_SLICE_TIME);
 269
 270    job->speed = speed;
 271    if (speed && speed <= old_speed) {
 272        return;
 273    }
 274
 275    /* kick only if a timer is pending */
 276    job_enter_cond(&job->job, job_timer_pending);
 277}
 278
 279int64_t block_job_ratelimit_get_delay(BlockJob *job, uint64_t n)
 280{
 281    if (!job->speed) {
 282        return 0;
 283    }
 284
 285    return ratelimit_calculate_delay(&job->limit, n);
 286}
 287
 288BlockJobInfo *block_job_query(BlockJob *job, Error **errp)
 289{
 290    BlockJobInfo *info;
 291
 292    if (block_job_is_internal(job)) {
 293        error_setg(errp, "Cannot query QEMU internal jobs");
 294        return NULL;
 295    }
 296    info = g_new0(BlockJobInfo, 1);
 297    info->type      = g_strdup(job_type_str(&job->job));
 298    info->device    = g_strdup(job->job.id);
 299    info->busy      = atomic_read(&job->job.busy);
 300    info->paused    = job->job.pause_count > 0;
 301    info->offset    = job->job.progress_current;
 302    info->len       = job->job.progress_total;
 303    info->speed     = job->speed;
 304    info->io_status = job->iostatus;
 305    info->ready     = job_is_ready(&job->job),
 306    info->status    = job->job.status;
 307    info->auto_finalize = job->job.auto_finalize;
 308    info->auto_dismiss  = job->job.auto_dismiss;
 309    info->has_error = job->job.ret != 0;
 310    info->error     = job->job.ret ? g_strdup(strerror(-job->job.ret)) : NULL;
 311    return info;
 312}
 313
 314static void block_job_iostatus_set_err(BlockJob *job, int error)
 315{
 316    if (job->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
 317        job->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE :
 318                                          BLOCK_DEVICE_IO_STATUS_FAILED;
 319    }
 320}
 321
 322static void block_job_event_cancelled(Notifier *n, void *opaque)
 323{
 324    BlockJob *job = opaque;
 325
 326    if (block_job_is_internal(job)) {
 327        return;
 328    }
 329
 330    qapi_event_send_block_job_cancelled(job_type(&job->job),
 331                                        job->job.id,
 332                                        job->job.progress_total,
 333                                        job->job.progress_current,
 334                                        job->speed);
 335}
 336
 337static void block_job_event_completed(Notifier *n, void *opaque)
 338{
 339    BlockJob *job = opaque;
 340    const char *msg = NULL;
 341
 342    if (block_job_is_internal(job)) {
 343        return;
 344    }
 345
 346    if (job->job.ret < 0) {
 347        msg = strerror(-job->job.ret);
 348    }
 349
 350    qapi_event_send_block_job_completed(job_type(&job->job),
 351                                        job->job.id,
 352                                        job->job.progress_total,
 353                                        job->job.progress_current,
 354                                        job->speed,
 355                                        !!msg,
 356                                        msg);
 357}
 358
 359static void block_job_event_pending(Notifier *n, void *opaque)
 360{
 361    BlockJob *job = opaque;
 362
 363    if (block_job_is_internal(job)) {
 364        return;
 365    }
 366
 367    qapi_event_send_block_job_pending(job_type(&job->job),
 368                                      job->job.id);
 369}
 370
 371static void block_job_event_ready(Notifier *n, void *opaque)
 372{
 373    BlockJob *job = opaque;
 374
 375    if (block_job_is_internal(job)) {
 376        return;
 377    }
 378
 379    qapi_event_send_block_job_ready(job_type(&job->job),
 380                                    job->job.id,
 381                                    job->job.progress_total,
 382                                    job->job.progress_current,
 383                                    job->speed);
 384}
 385
 386
 387/*
 388 * API for block job drivers and the block layer.  These functions are
 389 * declared in blockjob_int.h.
 390 */
 391
 392void *block_job_create(const char *job_id, const BlockJobDriver *driver,
 393                       JobTxn *txn, BlockDriverState *bs, uint64_t perm,
 394                       uint64_t shared_perm, int64_t speed, int flags,
 395                       BlockCompletionFunc *cb, void *opaque, Error **errp)
 396{
 397    BlockBackend *blk;
 398    BlockJob *job;
 399    int ret;
 400
 401    if (job_id == NULL && !(flags & JOB_INTERNAL)) {
 402        job_id = bdrv_get_device_name(bs);
 403    }
 404
 405    blk = blk_new(bdrv_get_aio_context(bs), perm, shared_perm);
 406    ret = blk_insert_bs(blk, bs, errp);
 407    if (ret < 0) {
 408        blk_unref(blk);
 409        return NULL;
 410    }
 411
 412    job = job_create(job_id, &driver->job_driver, txn, blk_get_aio_context(blk),
 413                     flags, cb, opaque, errp);
 414    if (job == NULL) {
 415        blk_unref(blk);
 416        return NULL;
 417    }
 418
 419    assert(is_block_job(&job->job));
 420    assert(job->job.driver->free == &block_job_free);
 421    assert(job->job.driver->user_resume == &block_job_user_resume);
 422
 423    job->blk = blk;
 424
 425    job->finalize_cancelled_notifier.notify = block_job_event_cancelled;
 426    job->finalize_completed_notifier.notify = block_job_event_completed;
 427    job->pending_notifier.notify = block_job_event_pending;
 428    job->ready_notifier.notify = block_job_event_ready;
 429    job->idle_notifier.notify = block_job_on_idle;
 430
 431    notifier_list_add(&job->job.on_finalize_cancelled,
 432                      &job->finalize_cancelled_notifier);
 433    notifier_list_add(&job->job.on_finalize_completed,
 434                      &job->finalize_completed_notifier);
 435    notifier_list_add(&job->job.on_pending, &job->pending_notifier);
 436    notifier_list_add(&job->job.on_ready, &job->ready_notifier);
 437    notifier_list_add(&job->job.on_idle, &job->idle_notifier);
 438
 439    error_setg(&job->blocker, "block device is in use by block job: %s",
 440               job_type_str(&job->job));
 441    block_job_add_bdrv(job, "main node", bs, 0, BLK_PERM_ALL, &error_abort);
 442
 443    bdrv_op_unblock(bs, BLOCK_OP_TYPE_DATAPLANE, job->blocker);
 444
 445    /* Disable request queuing in the BlockBackend to avoid deadlocks on drain:
 446     * The job reports that it's busy until it reaches a pause point. */
 447    blk_set_disable_request_queuing(blk, true);
 448    blk_set_allow_aio_context_change(blk, true);
 449
 450    /* Only set speed when necessary to avoid NotSupported error */
 451    if (speed != 0) {
 452        Error *local_err = NULL;
 453
 454        block_job_set_speed(job, speed, &local_err);
 455        if (local_err) {
 456            job_early_fail(&job->job);
 457            error_propagate(errp, local_err);
 458            return NULL;
 459        }
 460    }
 461
 462    return job;
 463}
 464
 465void block_job_iostatus_reset(BlockJob *job)
 466{
 467    if (job->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
 468        return;
 469    }
 470    assert(job->job.user_paused && job->job.pause_count > 0);
 471    job->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
 472}
 473
 474void block_job_user_resume(Job *job)
 475{
 476    BlockJob *bjob = container_of(job, BlockJob, job);
 477    block_job_iostatus_reset(bjob);
 478}
 479
 480BlockErrorAction block_job_error_action(BlockJob *job, BlockdevOnError on_err,
 481                                        int is_read, int error)
 482{
 483    BlockErrorAction action;
 484
 485    switch (on_err) {
 486    case BLOCKDEV_ON_ERROR_ENOSPC:
 487    case BLOCKDEV_ON_ERROR_AUTO:
 488        action = (error == ENOSPC) ?
 489                 BLOCK_ERROR_ACTION_STOP : BLOCK_ERROR_ACTION_REPORT;
 490        break;
 491    case BLOCKDEV_ON_ERROR_STOP:
 492        action = BLOCK_ERROR_ACTION_STOP;
 493        break;
 494    case BLOCKDEV_ON_ERROR_REPORT:
 495        action = BLOCK_ERROR_ACTION_REPORT;
 496        break;
 497    case BLOCKDEV_ON_ERROR_IGNORE:
 498        action = BLOCK_ERROR_ACTION_IGNORE;
 499        break;
 500    default:
 501        abort();
 502    }
 503    if (!block_job_is_internal(job)) {
 504        qapi_event_send_block_job_error(job->job.id,
 505                                        is_read ? IO_OPERATION_TYPE_READ :
 506                                        IO_OPERATION_TYPE_WRITE,
 507                                        action);
 508    }
 509    if (action == BLOCK_ERROR_ACTION_STOP) {
 510        if (!job->job.user_paused) {
 511            job_pause(&job->job);
 512            /* make the pause user visible, which will be resumed from QMP. */
 513            job->job.user_paused = true;
 514        }
 515        block_job_iostatus_set_err(job, error);
 516    }
 517    return action;
 518}
 519