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/aio-wait.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/main-loop.h" 37 #include "qemu/timer.h" 38 39 static bool is_block_job(Job *job) 40 { 41 return job_type(job) == JOB_TYPE_BACKUP || 42 job_type(job) == JOB_TYPE_COMMIT || 43 job_type(job) == JOB_TYPE_MIRROR || 44 job_type(job) == JOB_TYPE_STREAM; 45 } 46 47 BlockJob *block_job_next_locked(BlockJob *bjob) 48 { 49 Job *job = bjob ? &bjob->job : NULL; 50 GLOBAL_STATE_CODE(); 51 52 do { 53 job = job_next_locked(job); 54 } while (job && !is_block_job(job)); 55 56 return job ? container_of(job, BlockJob, job) : NULL; 57 } 58 59 BlockJob *block_job_get_locked(const char *id) 60 { 61 Job *job = job_get_locked(id); 62 GLOBAL_STATE_CODE(); 63 64 if (job && is_block_job(job)) { 65 return container_of(job, BlockJob, job); 66 } else { 67 return NULL; 68 } 69 } 70 71 BlockJob *block_job_get(const char *id) 72 { 73 JOB_LOCK_GUARD(); 74 return block_job_get_locked(id); 75 } 76 77 void block_job_free(Job *job) 78 { 79 BlockJob *bjob = container_of(job, BlockJob, job); 80 GLOBAL_STATE_CODE(); 81 82 block_job_remove_all_bdrv(bjob); 83 ratelimit_destroy(&bjob->limit); 84 error_free(bjob->blocker); 85 } 86 87 static char *child_job_get_parent_desc(BdrvChild *c) 88 { 89 BlockJob *job = c->opaque; 90 return g_strdup_printf("%s job '%s'", job_type_str(&job->job), job->job.id); 91 } 92 93 static void child_job_drained_begin(BdrvChild *c) 94 { 95 BlockJob *job = c->opaque; 96 job_pause(&job->job); 97 } 98 99 static bool child_job_drained_poll(BdrvChild *c) 100 { 101 BlockJob *bjob = c->opaque; 102 Job *job = &bjob->job; 103 const BlockJobDriver *drv = block_job_driver(bjob); 104 105 /* An inactive or completed job doesn't have any pending requests. Jobs 106 * with !job->busy are either already paused or have a pause point after 107 * being reentered, so no job driver code will run before they pause. */ 108 WITH_JOB_LOCK_GUARD() { 109 if (!job->busy || job_is_completed_locked(job)) { 110 return false; 111 } 112 } 113 114 /* Otherwise, assume that it isn't fully stopped yet, but allow the job to 115 * override this assumption. */ 116 if (drv->drained_poll) { 117 return drv->drained_poll(bjob); 118 } else { 119 return true; 120 } 121 } 122 123 static void child_job_drained_end(BdrvChild *c) 124 { 125 BlockJob *job = c->opaque; 126 job_resume(&job->job); 127 } 128 129 typedef struct BdrvStateChildJobContext { 130 AioContext *new_ctx; 131 BlockJob *job; 132 } BdrvStateChildJobContext; 133 134 static void child_job_set_aio_ctx_commit(void *opaque) 135 { 136 BdrvStateChildJobContext *s = opaque; 137 BlockJob *job = s->job; 138 139 job_set_aio_context(&job->job, s->new_ctx); 140 } 141 142 static TransactionActionDrv change_child_job_context = { 143 .commit = child_job_set_aio_ctx_commit, 144 .clean = g_free, 145 }; 146 147 static bool child_job_change_aio_ctx(BdrvChild *c, AioContext *ctx, 148 GHashTable *visited, Transaction *tran, 149 Error **errp) 150 { 151 BlockJob *job = c->opaque; 152 BdrvStateChildJobContext *s; 153 GSList *l; 154 155 for (l = job->nodes; l; l = l->next) { 156 BdrvChild *sibling = l->data; 157 if (!bdrv_child_change_aio_context(sibling, ctx, visited, 158 tran, errp)) { 159 return false; 160 } 161 } 162 163 s = g_new(BdrvStateChildJobContext, 1); 164 *s = (BdrvStateChildJobContext) { 165 .new_ctx = ctx, 166 .job = job, 167 }; 168 169 tran_add(tran, &change_child_job_context, s); 170 return true; 171 } 172 173 static AioContext *child_job_get_parent_aio_context(BdrvChild *c) 174 { 175 BlockJob *job = c->opaque; 176 IO_CODE(); 177 JOB_LOCK_GUARD(); 178 179 return job->job.aio_context; 180 } 181 182 static const BdrvChildClass child_job = { 183 .get_parent_desc = child_job_get_parent_desc, 184 .drained_begin = child_job_drained_begin, 185 .drained_poll = child_job_drained_poll, 186 .drained_end = child_job_drained_end, 187 .change_aio_ctx = child_job_change_aio_ctx, 188 .stay_at_node = true, 189 .get_parent_aio_context = child_job_get_parent_aio_context, 190 }; 191 192 void block_job_remove_all_bdrv(BlockJob *job) 193 { 194 GLOBAL_STATE_CODE(); 195 /* 196 * bdrv_root_unref_child() may reach child_job_[can_]set_aio_ctx(), 197 * which will also traverse job->nodes, so consume the list one by 198 * one to make sure that such a concurrent access does not attempt 199 * to process an already freed BdrvChild. 200 */ 201 aio_context_release(job->job.aio_context); 202 bdrv_graph_wrlock(NULL); 203 aio_context_acquire(job->job.aio_context); 204 while (job->nodes) { 205 GSList *l = job->nodes; 206 BdrvChild *c = l->data; 207 208 job->nodes = l->next; 209 210 bdrv_op_unblock_all(c->bs, job->blocker); 211 bdrv_root_unref_child(c); 212 213 g_slist_free_1(l); 214 } 215 bdrv_graph_wrunlock(); 216 } 217 218 bool block_job_has_bdrv(BlockJob *job, BlockDriverState *bs) 219 { 220 GSList *el; 221 GLOBAL_STATE_CODE(); 222 223 for (el = job->nodes; el; el = el->next) { 224 BdrvChild *c = el->data; 225 if (c->bs == bs) { 226 return true; 227 } 228 } 229 230 return false; 231 } 232 233 int block_job_add_bdrv(BlockJob *job, const char *name, BlockDriverState *bs, 234 uint64_t perm, uint64_t shared_perm, Error **errp) 235 { 236 BdrvChild *c; 237 AioContext *ctx = bdrv_get_aio_context(bs); 238 bool need_context_ops; 239 GLOBAL_STATE_CODE(); 240 241 bdrv_ref(bs); 242 243 need_context_ops = ctx != job->job.aio_context; 244 245 if (need_context_ops) { 246 if (job->job.aio_context != qemu_get_aio_context()) { 247 aio_context_release(job->job.aio_context); 248 } 249 aio_context_acquire(ctx); 250 } 251 c = bdrv_root_attach_child(bs, name, &child_job, 0, perm, shared_perm, job, 252 errp); 253 if (need_context_ops) { 254 aio_context_release(ctx); 255 if (job->job.aio_context != qemu_get_aio_context()) { 256 aio_context_acquire(job->job.aio_context); 257 } 258 } 259 if (c == NULL) { 260 return -EPERM; 261 } 262 263 job->nodes = g_slist_prepend(job->nodes, c); 264 bdrv_op_block_all(bs, job->blocker); 265 266 return 0; 267 } 268 269 /* Called with job_mutex lock held. */ 270 static void block_job_on_idle_locked(Notifier *n, void *opaque) 271 { 272 aio_wait_kick(); 273 } 274 275 bool block_job_is_internal(BlockJob *job) 276 { 277 return (job->job.id == NULL); 278 } 279 280 const BlockJobDriver *block_job_driver(BlockJob *job) 281 { 282 return container_of(job->job.driver, BlockJobDriver, job_driver); 283 } 284 285 /* Assumes the job_mutex is held */ 286 static bool job_timer_pending(Job *job) 287 { 288 return timer_pending(&job->sleep_timer); 289 } 290 291 bool block_job_set_speed_locked(BlockJob *job, int64_t speed, Error **errp) 292 { 293 const BlockJobDriver *drv = block_job_driver(job); 294 int64_t old_speed = job->speed; 295 296 GLOBAL_STATE_CODE(); 297 298 if (job_apply_verb_locked(&job->job, JOB_VERB_SET_SPEED, errp) < 0) { 299 return false; 300 } 301 if (speed < 0) { 302 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "speed", 303 "a non-negative value"); 304 return false; 305 } 306 307 ratelimit_set_speed(&job->limit, speed, BLOCK_JOB_SLICE_TIME); 308 309 job->speed = speed; 310 311 if (drv->set_speed) { 312 job_unlock(); 313 drv->set_speed(job, speed); 314 job_lock(); 315 } 316 317 if (speed && speed <= old_speed) { 318 return true; 319 } 320 321 /* kick only if a timer is pending */ 322 job_enter_cond_locked(&job->job, job_timer_pending); 323 324 return true; 325 } 326 327 static bool block_job_set_speed(BlockJob *job, int64_t speed, Error **errp) 328 { 329 JOB_LOCK_GUARD(); 330 return block_job_set_speed_locked(job, speed, errp); 331 } 332 333 void block_job_ratelimit_processed_bytes(BlockJob *job, uint64_t n) 334 { 335 IO_CODE(); 336 ratelimit_calculate_delay(&job->limit, n); 337 } 338 339 void block_job_ratelimit_sleep(BlockJob *job) 340 { 341 uint64_t delay_ns; 342 343 /* 344 * Sleep at least once. If the job is reentered early, keep waiting until 345 * we've waited for the full time that is necessary to keep the job at the 346 * right speed. 347 * 348 * Make sure to recalculate the delay after each (possibly interrupted) 349 * sleep because the speed can change while the job has yielded. 350 */ 351 do { 352 delay_ns = ratelimit_calculate_delay(&job->limit, 0); 353 job_sleep_ns(&job->job, delay_ns); 354 } while (delay_ns && !job_is_cancelled(&job->job)); 355 } 356 357 BlockJobInfo *block_job_query_locked(BlockJob *job, Error **errp) 358 { 359 BlockJobInfo *info; 360 uint64_t progress_current, progress_total; 361 362 GLOBAL_STATE_CODE(); 363 364 if (block_job_is_internal(job)) { 365 error_setg(errp, "Cannot query QEMU internal jobs"); 366 return NULL; 367 } 368 369 progress_get_snapshot(&job->job.progress, &progress_current, 370 &progress_total); 371 372 info = g_new0(BlockJobInfo, 1); 373 info->type = g_strdup(job_type_str(&job->job)); 374 info->device = g_strdup(job->job.id); 375 info->busy = job->job.busy; 376 info->paused = job->job.pause_count > 0; 377 info->offset = progress_current; 378 info->len = progress_total; 379 info->speed = job->speed; 380 info->io_status = job->iostatus; 381 info->ready = job_is_ready_locked(&job->job), 382 info->status = job->job.status; 383 info->auto_finalize = job->job.auto_finalize; 384 info->auto_dismiss = job->job.auto_dismiss; 385 if (job->job.ret) { 386 info->error = job->job.err ? 387 g_strdup(error_get_pretty(job->job.err)) : 388 g_strdup(strerror(-job->job.ret)); 389 } 390 return info; 391 } 392 393 /* Called with job lock held */ 394 static void block_job_iostatus_set_err_locked(BlockJob *job, int error) 395 { 396 if (job->iostatus == BLOCK_DEVICE_IO_STATUS_OK) { 397 job->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE : 398 BLOCK_DEVICE_IO_STATUS_FAILED; 399 } 400 } 401 402 /* Called with job_mutex lock held. */ 403 static void block_job_event_cancelled_locked(Notifier *n, void *opaque) 404 { 405 BlockJob *job = opaque; 406 uint64_t progress_current, progress_total; 407 408 if (block_job_is_internal(job)) { 409 return; 410 } 411 412 progress_get_snapshot(&job->job.progress, &progress_current, 413 &progress_total); 414 415 qapi_event_send_block_job_cancelled(job_type(&job->job), 416 job->job.id, 417 progress_total, 418 progress_current, 419 job->speed); 420 } 421 422 /* Called with job_mutex lock held. */ 423 static void block_job_event_completed_locked(Notifier *n, void *opaque) 424 { 425 BlockJob *job = opaque; 426 const char *msg = NULL; 427 uint64_t progress_current, progress_total; 428 429 if (block_job_is_internal(job)) { 430 return; 431 } 432 433 if (job->job.ret < 0) { 434 msg = error_get_pretty(job->job.err); 435 } 436 437 progress_get_snapshot(&job->job.progress, &progress_current, 438 &progress_total); 439 440 qapi_event_send_block_job_completed(job_type(&job->job), 441 job->job.id, 442 progress_total, 443 progress_current, 444 job->speed, 445 msg); 446 } 447 448 /* Called with job_mutex lock held. */ 449 static void block_job_event_pending_locked(Notifier *n, void *opaque) 450 { 451 BlockJob *job = opaque; 452 453 if (block_job_is_internal(job)) { 454 return; 455 } 456 457 qapi_event_send_block_job_pending(job_type(&job->job), 458 job->job.id); 459 } 460 461 /* Called with job_mutex lock held. */ 462 static void block_job_event_ready_locked(Notifier *n, void *opaque) 463 { 464 BlockJob *job = opaque; 465 uint64_t progress_current, progress_total; 466 467 if (block_job_is_internal(job)) { 468 return; 469 } 470 471 progress_get_snapshot(&job->job.progress, &progress_current, 472 &progress_total); 473 474 qapi_event_send_block_job_ready(job_type(&job->job), 475 job->job.id, 476 progress_total, 477 progress_current, 478 job->speed); 479 } 480 481 482 void *block_job_create(const char *job_id, const BlockJobDriver *driver, 483 JobTxn *txn, BlockDriverState *bs, uint64_t perm, 484 uint64_t shared_perm, int64_t speed, int flags, 485 BlockCompletionFunc *cb, void *opaque, Error **errp) 486 { 487 BlockJob *job; 488 int ret; 489 GLOBAL_STATE_CODE(); 490 GRAPH_RDLOCK_GUARD_MAINLOOP(); 491 492 if (job_id == NULL && !(flags & JOB_INTERNAL)) { 493 job_id = bdrv_get_device_name(bs); 494 } 495 496 job = job_create(job_id, &driver->job_driver, txn, bdrv_get_aio_context(bs), 497 flags, cb, opaque, errp); 498 if (job == NULL) { 499 return NULL; 500 } 501 502 assert(is_block_job(&job->job)); 503 assert(job->job.driver->free == &block_job_free); 504 assert(job->job.driver->user_resume == &block_job_user_resume); 505 506 ratelimit_init(&job->limit); 507 508 job->finalize_cancelled_notifier.notify = block_job_event_cancelled_locked; 509 job->finalize_completed_notifier.notify = block_job_event_completed_locked; 510 job->pending_notifier.notify = block_job_event_pending_locked; 511 job->ready_notifier.notify = block_job_event_ready_locked; 512 job->idle_notifier.notify = block_job_on_idle_locked; 513 514 WITH_JOB_LOCK_GUARD() { 515 notifier_list_add(&job->job.on_finalize_cancelled, 516 &job->finalize_cancelled_notifier); 517 notifier_list_add(&job->job.on_finalize_completed, 518 &job->finalize_completed_notifier); 519 notifier_list_add(&job->job.on_pending, &job->pending_notifier); 520 notifier_list_add(&job->job.on_ready, &job->ready_notifier); 521 notifier_list_add(&job->job.on_idle, &job->idle_notifier); 522 } 523 524 error_setg(&job->blocker, "block device is in use by block job: %s", 525 job_type_str(&job->job)); 526 527 ret = block_job_add_bdrv(job, "main node", bs, perm, shared_perm, errp); 528 if (ret < 0) { 529 goto fail; 530 } 531 532 bdrv_op_unblock(bs, BLOCK_OP_TYPE_DATAPLANE, job->blocker); 533 534 if (!block_job_set_speed(job, speed, errp)) { 535 goto fail; 536 } 537 538 return job; 539 540 fail: 541 job_early_fail(&job->job); 542 return NULL; 543 } 544 545 void block_job_iostatus_reset_locked(BlockJob *job) 546 { 547 GLOBAL_STATE_CODE(); 548 if (job->iostatus == BLOCK_DEVICE_IO_STATUS_OK) { 549 return; 550 } 551 assert(job->job.user_paused && job->job.pause_count > 0); 552 job->iostatus = BLOCK_DEVICE_IO_STATUS_OK; 553 } 554 555 static void block_job_iostatus_reset(BlockJob *job) 556 { 557 JOB_LOCK_GUARD(); 558 block_job_iostatus_reset_locked(job); 559 } 560 561 void block_job_user_resume(Job *job) 562 { 563 BlockJob *bjob = container_of(job, BlockJob, job); 564 GLOBAL_STATE_CODE(); 565 block_job_iostatus_reset(bjob); 566 } 567 568 BlockErrorAction block_job_error_action(BlockJob *job, BlockdevOnError on_err, 569 int is_read, int error) 570 { 571 BlockErrorAction action; 572 IO_CODE(); 573 574 switch (on_err) { 575 case BLOCKDEV_ON_ERROR_ENOSPC: 576 case BLOCKDEV_ON_ERROR_AUTO: 577 action = (error == ENOSPC) ? 578 BLOCK_ERROR_ACTION_STOP : BLOCK_ERROR_ACTION_REPORT; 579 break; 580 case BLOCKDEV_ON_ERROR_STOP: 581 action = BLOCK_ERROR_ACTION_STOP; 582 break; 583 case BLOCKDEV_ON_ERROR_REPORT: 584 action = BLOCK_ERROR_ACTION_REPORT; 585 break; 586 case BLOCKDEV_ON_ERROR_IGNORE: 587 action = BLOCK_ERROR_ACTION_IGNORE; 588 break; 589 default: 590 abort(); 591 } 592 if (!block_job_is_internal(job)) { 593 qapi_event_send_block_job_error(job->job.id, 594 is_read ? IO_OPERATION_TYPE_READ : 595 IO_OPERATION_TYPE_WRITE, 596 action); 597 } 598 if (action == BLOCK_ERROR_ACTION_STOP) { 599 WITH_JOB_LOCK_GUARD() { 600 if (!job->job.user_paused) { 601 job_pause_locked(&job->job); 602 /* 603 * make the pause user visible, which will be 604 * resumed from QMP. 605 */ 606 job->job.user_paused = true; 607 } 608 block_job_iostatus_set_err_locked(job, error); 609 } 610 } 611 return action; 612 } 613 614 AioContext *block_job_get_aio_context(BlockJob *job) 615 { 616 GLOBAL_STATE_CODE(); 617 return job->job.aio_context; 618 } 619