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 void child_job_drained_end(BdrvChild *c)
 159 {
 160     BlockJob *job = c->opaque;
 161     job_resume(&job->job);
 162 }
 163
 164 static const BdrvChildRole child_job = {
 165     .get_parent_desc    = child_job_get_parent_desc,
 166     .drained_begin      = child_job_drained_begin,
 167     .drained_end        = child_job_drained_end,
 168     .stay_at_node       = true,
 169 };
 170
 171 void block_job_remove_all_bdrv(BlockJob *job)
 172 {
 173     GSList *l;
 174     for (l = job->nodes; l; l = l->next) {
 175         BdrvChild *c = l->data;
 176         bdrv_op_unblock_all(c->bs, job->blocker);
 177         bdrv_root_unref_child(c);
 178     }
 179     g_slist_free(job->nodes);
 180     job->nodes = NULL;
 181 }
 182
 183 int block_job_add_bdrv(BlockJob *job, const char *name, BlockDriverState *bs,
 184                        uint64_t perm, uint64_t shared_perm, Error **errp)
 185 {
 186     BdrvChild *c;
 187
 188     c = bdrv_root_attach_child(bs, name, &child_job, perm, shared_perm,
 189                                job, errp);
 190     if (c == NULL) {
 191         return -EPERM;
 192     }
 193
 194     job->nodes = g_slist_prepend(job->nodes, c);
 195     bdrv_ref(bs);
 196     bdrv_op_block_all(bs, job->blocker);
 197
 198     return 0;
 199 }
 200
 201 bool block_job_is_internal(BlockJob *job)
 202 {
 203     return (job->job.id == NULL);
 204 }
 205
 206 const BlockJobDriver *block_job_driver(BlockJob *job)
 207 {
 208     return container_of(job->job.driver, BlockJobDriver, job_driver);
 209 }
 210
 211 /* Assumes the job_mutex is held */
 212 static bool job_timer_pending(Job *job)
 213 {
 214     return timer_pending(&job->sleep_timer);
 215 }
 216
 217 void block_job_set_speed(BlockJob *job, int64_t speed, Error **errp)
 218 {
 219     int64_t old_speed = job->speed;
 220
 221     if (job_apply_verb(&job->job, JOB_VERB_SET_SPEED, errp)) {
 222         return;
 223     }
 224     if (speed < 0) {
 225         error_setg(errp, QERR_INVALID_PARAMETER, "speed");
 226         return;
 227     }
 228
 229     ratelimit_set_speed(&job->limit, speed, BLOCK_JOB_SLICE_TIME);
 230
 231     job->speed = speed;
 232     if (speed && speed <= old_speed) {
 233         return;
 234     }
 235
 236     /* kick only if a timer is pending */
 237     job_enter_cond(&job->job, job_timer_pending);
 238 }
 239
 240 int64_t block_job_ratelimit_get_delay(BlockJob *job, uint64_t n)
 241 {
 242     if (!job->speed) {
 243         return 0;
 244     }
 245
 246     return ratelimit_calculate_delay(&job->limit, n);
 247 }
 248
 249 BlockJobInfo *block_job_query(BlockJob *job, Error **errp)
 250 {
 251     BlockJobInfo *info;
 252
 253     if (block_job_is_internal(job)) {
 254         error_setg(errp, "Cannot query QEMU internal jobs");
 255         return NULL;
 256     }
 257     info = g_new0(BlockJobInfo, 1);
 258     info->type      = g_strdup(job_type_str(&job->job));
 259     info->device    = g_strdup(job->job.id);
 260     info->busy      = atomic_read(&job->job.busy);
 261     info->paused    = job->job.pause_count > 0;
 262     info->offset    = job->job.progress_current;
 263     info->len       = job->job.progress_total;
 264     info->speed     = job->speed;
 265     info->io_status = job->iostatus;
 266     info->ready     = job_is_ready(&job->job),
 267     info->status    = job->job.status;
 268     info->auto_finalize = job->job.auto_finalize;
 269     info->auto_dismiss  = job->job.auto_dismiss;
 270     info->has_error = job->job.ret != 0;
 271     info->error     = job->job.ret ? g_strdup(strerror(-job->job.ret)) : NULL;
 272     return info;
 273 }
 274
 275 static void block_job_iostatus_set_err(BlockJob *job, int error)
 276 {
 277     if (job->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
 278         job->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE :
 279                                           BLOCK_DEVICE_IO_STATUS_FAILED;
 280     }
 281 }
 282
 283 static void block_job_event_cancelled(Notifier *n, void *opaque)
 284 {
 285     BlockJob *job = opaque;
 286
 287     if (block_job_is_internal(job)) {
 288         return;
 289     }
 290
 291     qapi_event_send_block_job_cancelled(job_type(&job->job),
 292                                         job->job.id,
 293                                         job->job.progress_total,
 294                                         job->job.progress_current,
 295                                         job->speed,
 296                                         &error_abort);
 297 }
 298
 299 static void block_job_event_completed(Notifier *n, void *opaque)
 300 {
 301     BlockJob *job = opaque;
 302     const char *msg = NULL;
 303
 304     if (block_job_is_internal(job)) {
 305         return;
 306     }
 307
 308     if (job->job.ret < 0) {
 309         msg = strerror(-job->job.ret);
 310     }
 311
 312     qapi_event_send_block_job_completed(job_type(&job->job),
 313                                         job->job.id,
 314                                         job->job.progress_total,
 315                                         job->job.progress_current,
 316                                         job->speed,
 317                                         !!msg,
 318                                         msg,
 319                                         &error_abort);
 320 }
 321
 322 static void block_job_event_pending(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_pending(job_type(&job->job),
 331                                       job->job.id,
 332                                       &error_abort);
 333 }
 334
 335 static void block_job_event_ready(Notifier *n, void *opaque)
 336 {
 337     BlockJob *job = opaque;
 338
 339     if (block_job_is_internal(job)) {
 340         return;
 341     }
 342
 343     qapi_event_send_block_job_ready(job_type(&job->job),
 344                                     job->job.id,
 345                                     job->job.progress_total,
 346                                     job->job.progress_current,
 347                                     job->speed, &error_abort);
 348 }
 349
 350
 351 /*
 352  * API for block job drivers and the block layer.  These functions are
 353  * declared in blockjob_int.h.
 354  */
 355
 356 void *block_job_create(const char *job_id, const BlockJobDriver *driver,
 357                        JobTxn *txn, BlockDriverState *bs, uint64_t perm,
 358                        uint64_t shared_perm, int64_t speed, int flags,
 359                        BlockCompletionFunc *cb, void *opaque, Error **errp)
 360 {
 361     BlockBackend *blk;
 362     BlockJob *job;
 363     int ret;
 364
 365     if (bs->job) {
 366         error_setg(errp, QERR_DEVICE_IN_USE, bdrv_get_device_name(bs));
 367         return NULL;
 368     }
 369
 370     if (job_id == NULL && !(flags & JOB_INTERNAL)) {
 371         job_id = bdrv_get_device_name(bs);
 372     }
 373
 374     blk = blk_new(perm, shared_perm);
 375     ret = blk_insert_bs(blk, bs, errp);
 376     if (ret < 0) {
 377         blk_unref(blk);
 378         return NULL;
 379     }
 380
 381     job = job_create(job_id, &driver->job_driver, txn, blk_get_aio_context(blk),
 382                      flags, cb, opaque, errp);
 383     if (job == NULL) {
 384         blk_unref(blk);
 385         return NULL;
 386     }
 387
 388     assert(is_block_job(&job->job));
 389     assert(job->job.driver->free == &block_job_free);
 390     assert(job->job.driver->user_resume == &block_job_user_resume);
 391     assert(job->job.driver->drain == &block_job_drain);
 392
 393     job->blk = blk;
 394
 395     job->finalize_cancelled_notifier.notify = block_job_event_cancelled;
 396     job->finalize_completed_notifier.notify = block_job_event_completed;
 397     job->pending_notifier.notify = block_job_event_pending;
 398     job->ready_notifier.notify = block_job_event_ready;
 399
 400     notifier_list_add(&job->job.on_finalize_cancelled,
 401                       &job->finalize_cancelled_notifier);
 402     notifier_list_add(&job->job.on_finalize_completed,
 403                       &job->finalize_completed_notifier);
 404     notifier_list_add(&job->job.on_pending, &job->pending_notifier);
 405     notifier_list_add(&job->job.on_ready, &job->ready_notifier);
 406
 407     error_setg(&job->blocker, "block device is in use by block job: %s",
 408                job_type_str(&job->job));
 409     block_job_add_bdrv(job, "main node", bs, 0, BLK_PERM_ALL, &error_abort);
 410     bs->job = job;
 411
 412     bdrv_op_unblock(bs, BLOCK_OP_TYPE_DATAPLANE, job->blocker);
 413
 414     blk_add_aio_context_notifier(blk, block_job_attached_aio_context,
 415                                  block_job_detach_aio_context, job);
 416
 417     /* Only set speed when necessary to avoid NotSupported error */
 418     if (speed != 0) {
 419         Error *local_err = NULL;
 420
 421         block_job_set_speed(job, speed, &local_err);
 422         if (local_err) {
 423             job_early_fail(&job->job);
 424             error_propagate(errp, local_err);
 425             return NULL;
 426         }
 427     }
 428
 429     return job;
 430 }
 431
 432 void block_job_iostatus_reset(BlockJob *job)
 433 {
 434     if (job->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
 435         return;
 436     }
 437     assert(job->job.user_paused && job->job.pause_count > 0);
 438     job->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
 439 }
 440
 441 void block_job_user_resume(Job *job)
 442 {
 443     BlockJob *bjob = container_of(job, BlockJob, job);
 444     block_job_iostatus_reset(bjob);
 445 }
 446
 447 BlockErrorAction block_job_error_action(BlockJob *job, BlockdevOnError on_err,
 448                                         int is_read, int error)
 449 {
 450     BlockErrorAction action;
 451
 452     switch (on_err) {
 453     case BLOCKDEV_ON_ERROR_ENOSPC:
 454     case BLOCKDEV_ON_ERROR_AUTO:
 455         action = (error == ENOSPC) ?
 456                  BLOCK_ERROR_ACTION_STOP : BLOCK_ERROR_ACTION_REPORT;
 457         break;
 458     case BLOCKDEV_ON_ERROR_STOP:
 459         action = BLOCK_ERROR_ACTION_STOP;
 460         break;
 461     case BLOCKDEV_ON_ERROR_REPORT:
 462         action = BLOCK_ERROR_ACTION_REPORT;
 463         break;
 464     case BLOCKDEV_ON_ERROR_IGNORE:
 465         action = BLOCK_ERROR_ACTION_IGNORE;
 466         break;
 467     default:
 468         abort();
 469     }
 470     if (!block_job_is_internal(job)) {
 471         qapi_event_send_block_job_error(job->job.id,
 472                                         is_read ? IO_OPERATION_TYPE_READ :
 473                                         IO_OPERATION_TYPE_WRITE,
 474                                         action, &error_abort);
 475     }
 476     if (action == BLOCK_ERROR_ACTION_STOP) {
 477         job_pause(&job->job);
 478         /* make the pause user visible, which will be resumed from QMP. */
 479         job->job.user_paused = true;
 480         block_job_iostatus_set_err(job, error);
 481     }
 482     return action;
 483 }