1 /* 2 * Image mirroring 3 * 4 * Copyright Red Hat, Inc. 2012 5 * 6 * Authors: 7 * Paolo Bonzini <pbonzini@redhat.com> 8 * 9 * This work is licensed under the terms of the GNU LGPL, version 2 or later. 10 * See the COPYING.LIB file in the top-level directory. 11 * 12 */ 13 14 #include "trace.h" 15 #include "block/blockjob.h" 16 #include "block/block_int.h" 17 #include "sysemu/block-backend.h" 18 #include "qapi/qmp/qerror.h" 19 #include "qemu/ratelimit.h" 20 #include "qemu/bitmap.h" 21 22 #define SLICE_TIME 100000000ULL /* ns */ 23 #define MAX_IN_FLIGHT 16 24 #define DEFAULT_MIRROR_BUF_SIZE (10 << 20) 25 26 /* The mirroring buffer is a list of granularity-sized chunks. 27 * Free chunks are organized in a list. 28 */ 29 typedef struct MirrorBuffer { 30 QSIMPLEQ_ENTRY(MirrorBuffer) next; 31 } MirrorBuffer; 32 33 typedef struct MirrorBlockJob { 34 BlockJob common; 35 RateLimit limit; 36 BlockDriverState *target; 37 BlockDriverState *base; 38 /* The name of the graph node to replace */ 39 char *replaces; 40 /* The BDS to replace */ 41 BlockDriverState *to_replace; 42 /* Used to block operations on the drive-mirror-replace target */ 43 Error *replace_blocker; 44 bool is_none_mode; 45 BlockdevOnError on_source_error, on_target_error; 46 bool synced; 47 bool should_complete; 48 int64_t sector_num; 49 int64_t granularity; 50 size_t buf_size; 51 int64_t bdev_length; 52 unsigned long *cow_bitmap; 53 BdrvDirtyBitmap *dirty_bitmap; 54 HBitmapIter hbi; 55 uint8_t *buf; 56 QSIMPLEQ_HEAD(, MirrorBuffer) buf_free; 57 int buf_free_count; 58 59 unsigned long *in_flight_bitmap; 60 int in_flight; 61 int sectors_in_flight; 62 int ret; 63 bool unmap; 64 bool waiting_for_io; 65 } MirrorBlockJob; 66 67 typedef struct MirrorOp { 68 MirrorBlockJob *s; 69 QEMUIOVector qiov; 70 int64_t sector_num; 71 int nb_sectors; 72 } MirrorOp; 73 74 static BlockErrorAction mirror_error_action(MirrorBlockJob *s, bool read, 75 int error) 76 { 77 s->synced = false; 78 if (read) { 79 return block_job_error_action(&s->common, s->common.bs, 80 s->on_source_error, true, error); 81 } else { 82 return block_job_error_action(&s->common, s->target, 83 s->on_target_error, false, error); 84 } 85 } 86 87 static void mirror_iteration_done(MirrorOp *op, int ret) 88 { 89 MirrorBlockJob *s = op->s; 90 struct iovec *iov; 91 int64_t chunk_num; 92 int i, nb_chunks, sectors_per_chunk; 93 94 trace_mirror_iteration_done(s, op->sector_num, op->nb_sectors, ret); 95 96 s->in_flight--; 97 s->sectors_in_flight -= op->nb_sectors; 98 iov = op->qiov.iov; 99 for (i = 0; i < op->qiov.niov; i++) { 100 MirrorBuffer *buf = (MirrorBuffer *) iov[i].iov_base; 101 QSIMPLEQ_INSERT_TAIL(&s->buf_free, buf, next); 102 s->buf_free_count++; 103 } 104 105 sectors_per_chunk = s->granularity >> BDRV_SECTOR_BITS; 106 chunk_num = op->sector_num / sectors_per_chunk; 107 nb_chunks = op->nb_sectors / sectors_per_chunk; 108 bitmap_clear(s->in_flight_bitmap, chunk_num, nb_chunks); 109 if (ret >= 0) { 110 if (s->cow_bitmap) { 111 bitmap_set(s->cow_bitmap, chunk_num, nb_chunks); 112 } 113 s->common.offset += (uint64_t)op->nb_sectors * BDRV_SECTOR_SIZE; 114 } 115 116 qemu_iovec_destroy(&op->qiov); 117 g_free(op); 118 119 if (s->waiting_for_io) { 120 qemu_coroutine_enter(s->common.co, NULL); 121 } 122 } 123 124 static void mirror_write_complete(void *opaque, int ret) 125 { 126 MirrorOp *op = opaque; 127 MirrorBlockJob *s = op->s; 128 if (ret < 0) { 129 BlockErrorAction action; 130 131 bdrv_set_dirty_bitmap(s->dirty_bitmap, op->sector_num, op->nb_sectors); 132 action = mirror_error_action(s, false, -ret); 133 if (action == BLOCK_ERROR_ACTION_REPORT && s->ret >= 0) { 134 s->ret = ret; 135 } 136 } 137 mirror_iteration_done(op, ret); 138 } 139 140 static void mirror_read_complete(void *opaque, int ret) 141 { 142 MirrorOp *op = opaque; 143 MirrorBlockJob *s = op->s; 144 if (ret < 0) { 145 BlockErrorAction action; 146 147 bdrv_set_dirty_bitmap(s->dirty_bitmap, op->sector_num, op->nb_sectors); 148 action = mirror_error_action(s, true, -ret); 149 if (action == BLOCK_ERROR_ACTION_REPORT && s->ret >= 0) { 150 s->ret = ret; 151 } 152 153 mirror_iteration_done(op, ret); 154 return; 155 } 156 bdrv_aio_writev(s->target, op->sector_num, &op->qiov, op->nb_sectors, 157 mirror_write_complete, op); 158 } 159 160 static uint64_t coroutine_fn mirror_iteration(MirrorBlockJob *s) 161 { 162 BlockDriverState *source = s->common.bs; 163 int nb_sectors, sectors_per_chunk, nb_chunks; 164 int64_t end, sector_num, next_chunk, next_sector, hbitmap_next_sector; 165 uint64_t delay_ns = 0; 166 MirrorOp *op; 167 int pnum; 168 int64_t ret; 169 170 s->sector_num = hbitmap_iter_next(&s->hbi); 171 if (s->sector_num < 0) { 172 bdrv_dirty_iter_init(s->dirty_bitmap, &s->hbi); 173 s->sector_num = hbitmap_iter_next(&s->hbi); 174 trace_mirror_restart_iter(s, bdrv_get_dirty_count(s->dirty_bitmap)); 175 assert(s->sector_num >= 0); 176 } 177 178 hbitmap_next_sector = s->sector_num; 179 sector_num = s->sector_num; 180 sectors_per_chunk = s->granularity >> BDRV_SECTOR_BITS; 181 end = s->bdev_length / BDRV_SECTOR_SIZE; 182 183 /* Extend the QEMUIOVector to include all adjacent blocks that will 184 * be copied in this operation. 185 * 186 * We have to do this if we have no backing file yet in the destination, 187 * and the cluster size is very large. Then we need to do COW ourselves. 188 * The first time a cluster is copied, copy it entirely. Note that, 189 * because both the granularity and the cluster size are powers of two, 190 * the number of sectors to copy cannot exceed one cluster. 191 * 192 * We also want to extend the QEMUIOVector to include more adjacent 193 * dirty blocks if possible, to limit the number of I/O operations and 194 * run efficiently even with a small granularity. 195 */ 196 nb_chunks = 0; 197 nb_sectors = 0; 198 next_sector = sector_num; 199 next_chunk = sector_num / sectors_per_chunk; 200 201 /* Wait for I/O to this cluster (from a previous iteration) to be done. */ 202 while (test_bit(next_chunk, s->in_flight_bitmap)) { 203 trace_mirror_yield_in_flight(s, sector_num, s->in_flight); 204 s->waiting_for_io = true; 205 qemu_coroutine_yield(); 206 s->waiting_for_io = false; 207 } 208 209 do { 210 int added_sectors, added_chunks; 211 212 if (!bdrv_get_dirty(source, s->dirty_bitmap, next_sector) || 213 test_bit(next_chunk, s->in_flight_bitmap)) { 214 assert(nb_sectors > 0); 215 break; 216 } 217 218 added_sectors = sectors_per_chunk; 219 if (s->cow_bitmap && !test_bit(next_chunk, s->cow_bitmap)) { 220 bdrv_round_to_clusters(s->target, 221 next_sector, added_sectors, 222 &next_sector, &added_sectors); 223 224 /* On the first iteration, the rounding may make us copy 225 * sectors before the first dirty one. 226 */ 227 if (next_sector < sector_num) { 228 assert(nb_sectors == 0); 229 sector_num = next_sector; 230 next_chunk = next_sector / sectors_per_chunk; 231 } 232 } 233 234 added_sectors = MIN(added_sectors, end - (sector_num + nb_sectors)); 235 added_chunks = (added_sectors + sectors_per_chunk - 1) / sectors_per_chunk; 236 237 /* When doing COW, it may happen that there is not enough space for 238 * a full cluster. Wait if that is the case. 239 */ 240 while (nb_chunks == 0 && s->buf_free_count < added_chunks) { 241 trace_mirror_yield_buf_busy(s, nb_chunks, s->in_flight); 242 s->waiting_for_io = true; 243 qemu_coroutine_yield(); 244 s->waiting_for_io = false; 245 } 246 if (s->buf_free_count < nb_chunks + added_chunks) { 247 trace_mirror_break_buf_busy(s, nb_chunks, s->in_flight); 248 break; 249 } 250 if (IOV_MAX < nb_chunks + added_chunks) { 251 trace_mirror_break_iov_max(s, nb_chunks, added_chunks); 252 break; 253 } 254 255 /* We have enough free space to copy these sectors. */ 256 bitmap_set(s->in_flight_bitmap, next_chunk, added_chunks); 257 258 nb_sectors += added_sectors; 259 nb_chunks += added_chunks; 260 next_sector += added_sectors; 261 next_chunk += added_chunks; 262 if (!s->synced && s->common.speed) { 263 delay_ns = ratelimit_calculate_delay(&s->limit, added_sectors); 264 } 265 } while (delay_ns == 0 && next_sector < end); 266 267 /* Allocate a MirrorOp that is used as an AIO callback. */ 268 op = g_new(MirrorOp, 1); 269 op->s = s; 270 op->sector_num = sector_num; 271 op->nb_sectors = nb_sectors; 272 273 /* Now make a QEMUIOVector taking enough granularity-sized chunks 274 * from s->buf_free. 275 */ 276 qemu_iovec_init(&op->qiov, nb_chunks); 277 next_sector = sector_num; 278 while (nb_chunks-- > 0) { 279 MirrorBuffer *buf = QSIMPLEQ_FIRST(&s->buf_free); 280 size_t remaining = (nb_sectors * BDRV_SECTOR_SIZE) - op->qiov.size; 281 282 QSIMPLEQ_REMOVE_HEAD(&s->buf_free, next); 283 s->buf_free_count--; 284 qemu_iovec_add(&op->qiov, buf, MIN(s->granularity, remaining)); 285 286 /* Advance the HBitmapIter in parallel, so that we do not examine 287 * the same sector twice. 288 */ 289 if (next_sector > hbitmap_next_sector 290 && bdrv_get_dirty(source, s->dirty_bitmap, next_sector)) { 291 hbitmap_next_sector = hbitmap_iter_next(&s->hbi); 292 } 293 294 next_sector += sectors_per_chunk; 295 } 296 297 bdrv_reset_dirty_bitmap(s->dirty_bitmap, sector_num, nb_sectors); 298 299 /* Copy the dirty cluster. */ 300 s->in_flight++; 301 s->sectors_in_flight += nb_sectors; 302 trace_mirror_one_iteration(s, sector_num, nb_sectors); 303 304 ret = bdrv_get_block_status_above(source, NULL, sector_num, 305 nb_sectors, &pnum); 306 if (ret < 0 || pnum < nb_sectors || 307 (ret & BDRV_BLOCK_DATA && !(ret & BDRV_BLOCK_ZERO))) { 308 bdrv_aio_readv(source, sector_num, &op->qiov, nb_sectors, 309 mirror_read_complete, op); 310 } else if (ret & BDRV_BLOCK_ZERO) { 311 bdrv_aio_write_zeroes(s->target, sector_num, op->nb_sectors, 312 s->unmap ? BDRV_REQ_MAY_UNMAP : 0, 313 mirror_write_complete, op); 314 } else { 315 assert(!(ret & BDRV_BLOCK_DATA)); 316 bdrv_aio_discard(s->target, sector_num, op->nb_sectors, 317 mirror_write_complete, op); 318 } 319 return delay_ns; 320 } 321 322 static void mirror_free_init(MirrorBlockJob *s) 323 { 324 int granularity = s->granularity; 325 size_t buf_size = s->buf_size; 326 uint8_t *buf = s->buf; 327 328 assert(s->buf_free_count == 0); 329 QSIMPLEQ_INIT(&s->buf_free); 330 while (buf_size != 0) { 331 MirrorBuffer *cur = (MirrorBuffer *)buf; 332 QSIMPLEQ_INSERT_TAIL(&s->buf_free, cur, next); 333 s->buf_free_count++; 334 buf_size -= granularity; 335 buf += granularity; 336 } 337 } 338 339 static void mirror_drain(MirrorBlockJob *s) 340 { 341 while (s->in_flight > 0) { 342 s->waiting_for_io = true; 343 qemu_coroutine_yield(); 344 s->waiting_for_io = false; 345 } 346 } 347 348 typedef struct { 349 int ret; 350 } MirrorExitData; 351 352 static void mirror_exit(BlockJob *job, void *opaque) 353 { 354 MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); 355 MirrorExitData *data = opaque; 356 AioContext *replace_aio_context = NULL; 357 BlockDriverState *src = s->common.bs; 358 359 /* Make sure that the source BDS doesn't go away before we called 360 * block_job_completed(). */ 361 bdrv_ref(src); 362 363 if (s->to_replace) { 364 replace_aio_context = bdrv_get_aio_context(s->to_replace); 365 aio_context_acquire(replace_aio_context); 366 } 367 368 if (s->should_complete && data->ret == 0) { 369 BlockDriverState *to_replace = s->common.bs; 370 if (s->to_replace) { 371 to_replace = s->to_replace; 372 } 373 if (bdrv_get_flags(s->target) != bdrv_get_flags(to_replace)) { 374 bdrv_reopen(s->target, bdrv_get_flags(to_replace), NULL); 375 } 376 bdrv_replace_in_backing_chain(to_replace, s->target); 377 } 378 if (s->to_replace) { 379 bdrv_op_unblock_all(s->to_replace, s->replace_blocker); 380 error_free(s->replace_blocker); 381 bdrv_unref(s->to_replace); 382 } 383 if (replace_aio_context) { 384 aio_context_release(replace_aio_context); 385 } 386 g_free(s->replaces); 387 bdrv_unref(s->target); 388 block_job_completed(&s->common, data->ret); 389 g_free(data); 390 bdrv_unref(src); 391 } 392 393 static void coroutine_fn mirror_run(void *opaque) 394 { 395 MirrorBlockJob *s = opaque; 396 MirrorExitData *data; 397 BlockDriverState *bs = s->common.bs; 398 int64_t sector_num, end, length; 399 uint64_t last_pause_ns; 400 BlockDriverInfo bdi; 401 char backing_filename[2]; /* we only need 2 characters because we are only 402 checking for a NULL string */ 403 int ret = 0; 404 int n; 405 406 if (block_job_is_cancelled(&s->common)) { 407 goto immediate_exit; 408 } 409 410 s->bdev_length = bdrv_getlength(bs); 411 if (s->bdev_length < 0) { 412 ret = s->bdev_length; 413 goto immediate_exit; 414 } else if (s->bdev_length == 0) { 415 /* Report BLOCK_JOB_READY and wait for complete. */ 416 block_job_event_ready(&s->common); 417 s->synced = true; 418 while (!block_job_is_cancelled(&s->common) && !s->should_complete) { 419 block_job_yield(&s->common); 420 } 421 s->common.cancelled = false; 422 goto immediate_exit; 423 } 424 425 length = DIV_ROUND_UP(s->bdev_length, s->granularity); 426 s->in_flight_bitmap = bitmap_new(length); 427 428 /* If we have no backing file yet in the destination, we cannot let 429 * the destination do COW. Instead, we copy sectors around the 430 * dirty data if needed. We need a bitmap to do that. 431 */ 432 bdrv_get_backing_filename(s->target, backing_filename, 433 sizeof(backing_filename)); 434 if (backing_filename[0] && !s->target->backing) { 435 ret = bdrv_get_info(s->target, &bdi); 436 if (ret < 0) { 437 goto immediate_exit; 438 } 439 if (s->granularity < bdi.cluster_size) { 440 s->buf_size = MAX(s->buf_size, bdi.cluster_size); 441 s->cow_bitmap = bitmap_new(length); 442 } 443 } 444 445 end = s->bdev_length / BDRV_SECTOR_SIZE; 446 s->buf = qemu_try_blockalign(bs, s->buf_size); 447 if (s->buf == NULL) { 448 ret = -ENOMEM; 449 goto immediate_exit; 450 } 451 452 mirror_free_init(s); 453 454 last_pause_ns = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); 455 if (!s->is_none_mode) { 456 /* First part, loop on the sectors and initialize the dirty bitmap. */ 457 BlockDriverState *base = s->base; 458 bool mark_all_dirty = s->base == NULL && !bdrv_has_zero_init(s->target); 459 460 for (sector_num = 0; sector_num < end; ) { 461 /* Just to make sure we are not exceeding int limit. */ 462 int nb_sectors = MIN(INT_MAX >> BDRV_SECTOR_BITS, 463 end - sector_num); 464 int64_t now = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); 465 466 if (now - last_pause_ns > SLICE_TIME) { 467 last_pause_ns = now; 468 block_job_sleep_ns(&s->common, QEMU_CLOCK_REALTIME, 0); 469 } 470 471 if (block_job_is_cancelled(&s->common)) { 472 goto immediate_exit; 473 } 474 475 ret = bdrv_is_allocated_above(bs, base, sector_num, nb_sectors, &n); 476 477 if (ret < 0) { 478 goto immediate_exit; 479 } 480 481 assert(n > 0); 482 if (ret == 1 || mark_all_dirty) { 483 bdrv_set_dirty_bitmap(s->dirty_bitmap, sector_num, n); 484 } 485 sector_num += n; 486 } 487 } 488 489 bdrv_dirty_iter_init(s->dirty_bitmap, &s->hbi); 490 for (;;) { 491 uint64_t delay_ns = 0; 492 int64_t cnt; 493 bool should_complete; 494 495 if (s->ret < 0) { 496 ret = s->ret; 497 goto immediate_exit; 498 } 499 500 cnt = bdrv_get_dirty_count(s->dirty_bitmap); 501 /* s->common.offset contains the number of bytes already processed so 502 * far, cnt is the number of dirty sectors remaining and 503 * s->sectors_in_flight is the number of sectors currently being 504 * processed; together those are the current total operation length */ 505 s->common.len = s->common.offset + 506 (cnt + s->sectors_in_flight) * BDRV_SECTOR_SIZE; 507 508 /* Note that even when no rate limit is applied we need to yield 509 * periodically with no pending I/O so that bdrv_drain_all() returns. 510 * We do so every SLICE_TIME nanoseconds, or when there is an error, 511 * or when the source is clean, whichever comes first. 512 */ 513 if (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - last_pause_ns < SLICE_TIME && 514 s->common.iostatus == BLOCK_DEVICE_IO_STATUS_OK) { 515 if (s->in_flight == MAX_IN_FLIGHT || s->buf_free_count == 0 || 516 (cnt == 0 && s->in_flight > 0)) { 517 trace_mirror_yield(s, s->in_flight, s->buf_free_count, cnt); 518 s->waiting_for_io = true; 519 qemu_coroutine_yield(); 520 s->waiting_for_io = false; 521 continue; 522 } else if (cnt != 0) { 523 delay_ns = mirror_iteration(s); 524 } 525 } 526 527 should_complete = false; 528 if (s->in_flight == 0 && cnt == 0) { 529 trace_mirror_before_flush(s); 530 ret = bdrv_flush(s->target); 531 if (ret < 0) { 532 if (mirror_error_action(s, false, -ret) == 533 BLOCK_ERROR_ACTION_REPORT) { 534 goto immediate_exit; 535 } 536 } else { 537 /* We're out of the streaming phase. From now on, if the job 538 * is cancelled we will actually complete all pending I/O and 539 * report completion. This way, block-job-cancel will leave 540 * the target in a consistent state. 541 */ 542 if (!s->synced) { 543 block_job_event_ready(&s->common); 544 s->synced = true; 545 } 546 547 should_complete = s->should_complete || 548 block_job_is_cancelled(&s->common); 549 cnt = bdrv_get_dirty_count(s->dirty_bitmap); 550 } 551 } 552 553 if (cnt == 0 && should_complete) { 554 /* The dirty bitmap is not updated while operations are pending. 555 * If we're about to exit, wait for pending operations before 556 * calling bdrv_get_dirty_count(bs), or we may exit while the 557 * source has dirty data to copy! 558 * 559 * Note that I/O can be submitted by the guest while 560 * mirror_populate runs. 561 */ 562 trace_mirror_before_drain(s, cnt); 563 bdrv_drain(bs); 564 cnt = bdrv_get_dirty_count(s->dirty_bitmap); 565 } 566 567 ret = 0; 568 trace_mirror_before_sleep(s, cnt, s->synced, delay_ns); 569 if (!s->synced) { 570 block_job_sleep_ns(&s->common, QEMU_CLOCK_REALTIME, delay_ns); 571 if (block_job_is_cancelled(&s->common)) { 572 break; 573 } 574 } else if (!should_complete) { 575 delay_ns = (s->in_flight == 0 && cnt == 0 ? SLICE_TIME : 0); 576 block_job_sleep_ns(&s->common, QEMU_CLOCK_REALTIME, delay_ns); 577 } else if (cnt == 0) { 578 /* The two disks are in sync. Exit and report successful 579 * completion. 580 */ 581 assert(QLIST_EMPTY(&bs->tracked_requests)); 582 s->common.cancelled = false; 583 break; 584 } 585 last_pause_ns = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); 586 } 587 588 immediate_exit: 589 if (s->in_flight > 0) { 590 /* We get here only if something went wrong. Either the job failed, 591 * or it was cancelled prematurely so that we do not guarantee that 592 * the target is a copy of the source. 593 */ 594 assert(ret < 0 || (!s->synced && block_job_is_cancelled(&s->common))); 595 mirror_drain(s); 596 } 597 598 assert(s->in_flight == 0); 599 qemu_vfree(s->buf); 600 g_free(s->cow_bitmap); 601 g_free(s->in_flight_bitmap); 602 bdrv_release_dirty_bitmap(bs, s->dirty_bitmap); 603 if (s->target->blk) { 604 blk_iostatus_disable(s->target->blk); 605 } 606 607 data = g_malloc(sizeof(*data)); 608 data->ret = ret; 609 block_job_defer_to_main_loop(&s->common, mirror_exit, data); 610 } 611 612 static void mirror_set_speed(BlockJob *job, int64_t speed, Error **errp) 613 { 614 MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); 615 616 if (speed < 0) { 617 error_setg(errp, QERR_INVALID_PARAMETER, "speed"); 618 return; 619 } 620 ratelimit_set_speed(&s->limit, speed / BDRV_SECTOR_SIZE, SLICE_TIME); 621 } 622 623 static void mirror_iostatus_reset(BlockJob *job) 624 { 625 MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); 626 627 if (s->target->blk) { 628 blk_iostatus_reset(s->target->blk); 629 } 630 } 631 632 static void mirror_complete(BlockJob *job, Error **errp) 633 { 634 MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); 635 Error *local_err = NULL; 636 int ret; 637 638 ret = bdrv_open_backing_file(s->target, NULL, &local_err); 639 if (ret < 0) { 640 error_propagate(errp, local_err); 641 return; 642 } 643 if (!s->synced) { 644 error_setg(errp, QERR_BLOCK_JOB_NOT_READY, job->id); 645 return; 646 } 647 648 /* check the target bs is not blocked and block all operations on it */ 649 if (s->replaces) { 650 AioContext *replace_aio_context; 651 652 s->to_replace = bdrv_find_node(s->replaces); 653 if (!s->to_replace) { 654 error_setg(errp, "Node name '%s' not found", s->replaces); 655 return; 656 } 657 658 replace_aio_context = bdrv_get_aio_context(s->to_replace); 659 aio_context_acquire(replace_aio_context); 660 661 error_setg(&s->replace_blocker, 662 "block device is in use by block-job-complete"); 663 bdrv_op_block_all(s->to_replace, s->replace_blocker); 664 bdrv_ref(s->to_replace); 665 666 aio_context_release(replace_aio_context); 667 } 668 669 s->should_complete = true; 670 block_job_enter(&s->common); 671 } 672 673 static const BlockJobDriver mirror_job_driver = { 674 .instance_size = sizeof(MirrorBlockJob), 675 .job_type = BLOCK_JOB_TYPE_MIRROR, 676 .set_speed = mirror_set_speed, 677 .iostatus_reset= mirror_iostatus_reset, 678 .complete = mirror_complete, 679 }; 680 681 static const BlockJobDriver commit_active_job_driver = { 682 .instance_size = sizeof(MirrorBlockJob), 683 .job_type = BLOCK_JOB_TYPE_COMMIT, 684 .set_speed = mirror_set_speed, 685 .iostatus_reset 686 = mirror_iostatus_reset, 687 .complete = mirror_complete, 688 }; 689 690 static void mirror_start_job(BlockDriverState *bs, BlockDriverState *target, 691 const char *replaces, 692 int64_t speed, uint32_t granularity, 693 int64_t buf_size, 694 BlockdevOnError on_source_error, 695 BlockdevOnError on_target_error, 696 bool unmap, 697 BlockCompletionFunc *cb, 698 void *opaque, Error **errp, 699 const BlockJobDriver *driver, 700 bool is_none_mode, BlockDriverState *base) 701 { 702 MirrorBlockJob *s; 703 704 if (granularity == 0) { 705 granularity = bdrv_get_default_bitmap_granularity(target); 706 } 707 708 assert ((granularity & (granularity - 1)) == 0); 709 710 if ((on_source_error == BLOCKDEV_ON_ERROR_STOP || 711 on_source_error == BLOCKDEV_ON_ERROR_ENOSPC) && 712 (!bs->blk || !blk_iostatus_is_enabled(bs->blk))) { 713 error_setg(errp, QERR_INVALID_PARAMETER, "on-source-error"); 714 return; 715 } 716 717 if (buf_size < 0) { 718 error_setg(errp, "Invalid parameter 'buf-size'"); 719 return; 720 } 721 722 if (buf_size == 0) { 723 buf_size = DEFAULT_MIRROR_BUF_SIZE; 724 } 725 726 s = block_job_create(driver, bs, speed, cb, opaque, errp); 727 if (!s) { 728 return; 729 } 730 731 s->replaces = g_strdup(replaces); 732 s->on_source_error = on_source_error; 733 s->on_target_error = on_target_error; 734 s->target = target; 735 s->is_none_mode = is_none_mode; 736 s->base = base; 737 s->granularity = granularity; 738 s->buf_size = ROUND_UP(buf_size, granularity); 739 s->unmap = unmap; 740 741 s->dirty_bitmap = bdrv_create_dirty_bitmap(bs, granularity, NULL, errp); 742 if (!s->dirty_bitmap) { 743 g_free(s->replaces); 744 block_job_release(bs); 745 return; 746 } 747 bdrv_set_enable_write_cache(s->target, true); 748 if (s->target->blk) { 749 blk_set_on_error(s->target->blk, on_target_error, on_target_error); 750 blk_iostatus_enable(s->target->blk); 751 } 752 s->common.co = qemu_coroutine_create(mirror_run); 753 trace_mirror_start(bs, s, s->common.co, opaque); 754 qemu_coroutine_enter(s->common.co, s); 755 } 756 757 void mirror_start(BlockDriverState *bs, BlockDriverState *target, 758 const char *replaces, 759 int64_t speed, uint32_t granularity, int64_t buf_size, 760 MirrorSyncMode mode, BlockdevOnError on_source_error, 761 BlockdevOnError on_target_error, 762 bool unmap, 763 BlockCompletionFunc *cb, 764 void *opaque, Error **errp) 765 { 766 bool is_none_mode; 767 BlockDriverState *base; 768 769 if (mode == MIRROR_SYNC_MODE_INCREMENTAL) { 770 error_setg(errp, "Sync mode 'incremental' not supported"); 771 return; 772 } 773 is_none_mode = mode == MIRROR_SYNC_MODE_NONE; 774 base = mode == MIRROR_SYNC_MODE_TOP ? backing_bs(bs) : NULL; 775 mirror_start_job(bs, target, replaces, 776 speed, granularity, buf_size, 777 on_source_error, on_target_error, unmap, cb, opaque, errp, 778 &mirror_job_driver, is_none_mode, base); 779 } 780 781 void commit_active_start(BlockDriverState *bs, BlockDriverState *base, 782 int64_t speed, 783 BlockdevOnError on_error, 784 BlockCompletionFunc *cb, 785 void *opaque, Error **errp) 786 { 787 int64_t length, base_length; 788 int orig_base_flags; 789 int ret; 790 Error *local_err = NULL; 791 792 orig_base_flags = bdrv_get_flags(base); 793 794 if (bdrv_reopen(base, bs->open_flags, errp)) { 795 return; 796 } 797 798 length = bdrv_getlength(bs); 799 if (length < 0) { 800 error_setg_errno(errp, -length, 801 "Unable to determine length of %s", bs->filename); 802 goto error_restore_flags; 803 } 804 805 base_length = bdrv_getlength(base); 806 if (base_length < 0) { 807 error_setg_errno(errp, -base_length, 808 "Unable to determine length of %s", base->filename); 809 goto error_restore_flags; 810 } 811 812 if (length > base_length) { 813 ret = bdrv_truncate(base, length); 814 if (ret < 0) { 815 error_setg_errno(errp, -ret, 816 "Top image %s is larger than base image %s, and " 817 "resize of base image failed", 818 bs->filename, base->filename); 819 goto error_restore_flags; 820 } 821 } 822 823 bdrv_ref(base); 824 mirror_start_job(bs, base, NULL, speed, 0, 0, 825 on_error, on_error, false, cb, opaque, &local_err, 826 &commit_active_job_driver, false, base); 827 if (local_err) { 828 error_propagate(errp, local_err); 829 goto error_restore_flags; 830 } 831 832 return; 833 834 error_restore_flags: 835 /* ignore error and errp for bdrv_reopen, because we want to propagate 836 * the original error */ 837 bdrv_reopen(base, orig_base_flags, NULL); 838 return; 839 } 840