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 "qemu/osdep.h" 15 #include "qemu/cutils.h" 16 #include "qemu/coroutine.h" 17 #include "qemu/range.h" 18 #include "trace.h" 19 #include "block/blockjob_int.h" 20 #include "block/block_int.h" 21 #include "block/dirty-bitmap.h" 22 #include "sysemu/block-backend.h" 23 #include "qapi/error.h" 24 #include "qemu/ratelimit.h" 25 #include "qemu/bitmap.h" 26 #include "qemu/memalign.h" 27 28 #define MAX_IN_FLIGHT 16 29 #define MAX_IO_BYTES (1 << 20) /* 1 Mb */ 30 #define DEFAULT_MIRROR_BUF_SIZE (MAX_IN_FLIGHT * MAX_IO_BYTES) 31 32 /* The mirroring buffer is a list of granularity-sized chunks. 33 * Free chunks are organized in a list. 34 */ 35 typedef struct MirrorBuffer { 36 QSIMPLEQ_ENTRY(MirrorBuffer) next; 37 } MirrorBuffer; 38 39 typedef struct MirrorOp MirrorOp; 40 41 typedef struct MirrorBlockJob { 42 BlockJob common; 43 BlockBackend *target; 44 BlockDriverState *mirror_top_bs; 45 BlockDriverState *base; 46 BlockDriverState *base_overlay; 47 48 /* The name of the graph node to replace */ 49 char *replaces; 50 /* The BDS to replace */ 51 BlockDriverState *to_replace; 52 /* Used to block operations on the drive-mirror-replace target */ 53 Error *replace_blocker; 54 bool is_none_mode; 55 BlockMirrorBackingMode backing_mode; 56 /* Whether the target image requires explicit zero-initialization */ 57 bool zero_target; 58 /* 59 * To be accesssed with atomics. Written only under the BQL (required by the 60 * current implementation of mirror_change()). 61 */ 62 MirrorCopyMode copy_mode; 63 BlockdevOnError on_source_error, on_target_error; 64 /* 65 * To be accessed with atomics. 66 * 67 * Set when the target is synced (dirty bitmap is clean, nothing in flight) 68 * and the job is running in active mode. 69 */ 70 bool actively_synced; 71 bool should_complete; 72 int64_t granularity; 73 size_t buf_size; 74 int64_t bdev_length; 75 unsigned long *cow_bitmap; 76 BdrvDirtyBitmap *dirty_bitmap; 77 BdrvDirtyBitmapIter *dbi; 78 uint8_t *buf; 79 QSIMPLEQ_HEAD(, MirrorBuffer) buf_free; 80 int buf_free_count; 81 82 uint64_t last_pause_ns; 83 unsigned long *in_flight_bitmap; 84 unsigned in_flight; 85 int64_t bytes_in_flight; 86 QTAILQ_HEAD(, MirrorOp) ops_in_flight; 87 int ret; 88 bool unmap; 89 int target_cluster_size; 90 int max_iov; 91 bool initial_zeroing_ongoing; 92 int in_active_write_counter; 93 int64_t active_write_bytes_in_flight; 94 bool prepared; 95 bool in_drain; 96 bool base_ro; 97 } MirrorBlockJob; 98 99 typedef struct MirrorBDSOpaque { 100 MirrorBlockJob *job; 101 bool stop; 102 bool is_commit; 103 } MirrorBDSOpaque; 104 105 struct MirrorOp { 106 MirrorBlockJob *s; 107 QEMUIOVector qiov; 108 int64_t offset; 109 uint64_t bytes; 110 111 /* The pointee is set by mirror_co_read(), mirror_co_zero(), and 112 * mirror_co_discard() before yielding for the first time */ 113 int64_t *bytes_handled; 114 115 bool is_pseudo_op; 116 bool is_active_write; 117 bool is_in_flight; 118 CoQueue waiting_requests; 119 Coroutine *co; 120 MirrorOp *waiting_for_op; 121 122 QTAILQ_ENTRY(MirrorOp) next; 123 }; 124 125 typedef enum MirrorMethod { 126 MIRROR_METHOD_COPY, 127 MIRROR_METHOD_ZERO, 128 MIRROR_METHOD_DISCARD, 129 } MirrorMethod; 130 131 static BlockErrorAction mirror_error_action(MirrorBlockJob *s, bool read, 132 int error) 133 { 134 qatomic_set(&s->actively_synced, false); 135 if (read) { 136 return block_job_error_action(&s->common, s->on_source_error, 137 true, error); 138 } else { 139 return block_job_error_action(&s->common, s->on_target_error, 140 false, error); 141 } 142 } 143 144 static void coroutine_fn mirror_wait_on_conflicts(MirrorOp *self, 145 MirrorBlockJob *s, 146 uint64_t offset, 147 uint64_t bytes) 148 { 149 uint64_t self_start_chunk = offset / s->granularity; 150 uint64_t self_end_chunk = DIV_ROUND_UP(offset + bytes, s->granularity); 151 uint64_t self_nb_chunks = self_end_chunk - self_start_chunk; 152 153 while (find_next_bit(s->in_flight_bitmap, self_end_chunk, 154 self_start_chunk) < self_end_chunk && 155 s->ret >= 0) 156 { 157 MirrorOp *op; 158 159 QTAILQ_FOREACH(op, &s->ops_in_flight, next) { 160 uint64_t op_start_chunk = op->offset / s->granularity; 161 uint64_t op_nb_chunks = DIV_ROUND_UP(op->offset + op->bytes, 162 s->granularity) - 163 op_start_chunk; 164 165 if (op == self) { 166 continue; 167 } 168 169 if (ranges_overlap(self_start_chunk, self_nb_chunks, 170 op_start_chunk, op_nb_chunks)) 171 { 172 if (self) { 173 /* 174 * If the operation is already (indirectly) waiting for us, 175 * or will wait for us as soon as it wakes up, then just go 176 * on (instead of producing a deadlock in the former case). 177 */ 178 if (op->waiting_for_op) { 179 continue; 180 } 181 182 self->waiting_for_op = op; 183 } 184 185 qemu_co_queue_wait(&op->waiting_requests, NULL); 186 187 if (self) { 188 self->waiting_for_op = NULL; 189 } 190 191 break; 192 } 193 } 194 } 195 } 196 197 static void coroutine_fn mirror_iteration_done(MirrorOp *op, int ret) 198 { 199 MirrorBlockJob *s = op->s; 200 struct iovec *iov; 201 int64_t chunk_num; 202 int i, nb_chunks; 203 204 trace_mirror_iteration_done(s, op->offset, op->bytes, ret); 205 206 s->in_flight--; 207 s->bytes_in_flight -= op->bytes; 208 iov = op->qiov.iov; 209 for (i = 0; i < op->qiov.niov; i++) { 210 MirrorBuffer *buf = (MirrorBuffer *) iov[i].iov_base; 211 QSIMPLEQ_INSERT_TAIL(&s->buf_free, buf, next); 212 s->buf_free_count++; 213 } 214 215 chunk_num = op->offset / s->granularity; 216 nb_chunks = DIV_ROUND_UP(op->bytes, s->granularity); 217 218 bitmap_clear(s->in_flight_bitmap, chunk_num, nb_chunks); 219 QTAILQ_REMOVE(&s->ops_in_flight, op, next); 220 if (ret >= 0) { 221 if (s->cow_bitmap) { 222 bitmap_set(s->cow_bitmap, chunk_num, nb_chunks); 223 } 224 if (!s->initial_zeroing_ongoing) { 225 job_progress_update(&s->common.job, op->bytes); 226 } 227 } 228 qemu_iovec_destroy(&op->qiov); 229 230 qemu_co_queue_restart_all(&op->waiting_requests); 231 g_free(op); 232 } 233 234 static void coroutine_fn mirror_write_complete(MirrorOp *op, int ret) 235 { 236 MirrorBlockJob *s = op->s; 237 238 if (ret < 0) { 239 BlockErrorAction action; 240 241 bdrv_set_dirty_bitmap(s->dirty_bitmap, op->offset, op->bytes); 242 action = mirror_error_action(s, false, -ret); 243 if (action == BLOCK_ERROR_ACTION_REPORT && s->ret >= 0) { 244 s->ret = ret; 245 } 246 } 247 248 mirror_iteration_done(op, ret); 249 } 250 251 static void coroutine_fn mirror_read_complete(MirrorOp *op, int ret) 252 { 253 MirrorBlockJob *s = op->s; 254 255 if (ret < 0) { 256 BlockErrorAction action; 257 258 bdrv_set_dirty_bitmap(s->dirty_bitmap, op->offset, op->bytes); 259 action = mirror_error_action(s, true, -ret); 260 if (action == BLOCK_ERROR_ACTION_REPORT && s->ret >= 0) { 261 s->ret = ret; 262 } 263 264 mirror_iteration_done(op, ret); 265 return; 266 } 267 268 ret = blk_co_pwritev(s->target, op->offset, op->qiov.size, &op->qiov, 0); 269 mirror_write_complete(op, ret); 270 } 271 272 /* Clip bytes relative to offset to not exceed end-of-file */ 273 static inline int64_t mirror_clip_bytes(MirrorBlockJob *s, 274 int64_t offset, 275 int64_t bytes) 276 { 277 return MIN(bytes, s->bdev_length - offset); 278 } 279 280 /* Round offset and/or bytes to target cluster if COW is needed, and 281 * return the offset of the adjusted tail against original. */ 282 static int coroutine_fn mirror_cow_align(MirrorBlockJob *s, int64_t *offset, 283 uint64_t *bytes) 284 { 285 bool need_cow; 286 int ret = 0; 287 int64_t align_offset = *offset; 288 int64_t align_bytes = *bytes; 289 int max_bytes = s->granularity * s->max_iov; 290 291 need_cow = !test_bit(*offset / s->granularity, s->cow_bitmap); 292 need_cow |= !test_bit((*offset + *bytes - 1) / s->granularity, 293 s->cow_bitmap); 294 if (need_cow) { 295 bdrv_round_to_subclusters(blk_bs(s->target), *offset, *bytes, 296 &align_offset, &align_bytes); 297 } 298 299 if (align_bytes > max_bytes) { 300 align_bytes = max_bytes; 301 if (need_cow) { 302 align_bytes = QEMU_ALIGN_DOWN(align_bytes, s->target_cluster_size); 303 } 304 } 305 /* Clipping may result in align_bytes unaligned to chunk boundary, but 306 * that doesn't matter because it's already the end of source image. */ 307 align_bytes = mirror_clip_bytes(s, align_offset, align_bytes); 308 309 ret = align_offset + align_bytes - (*offset + *bytes); 310 *offset = align_offset; 311 *bytes = align_bytes; 312 assert(ret >= 0); 313 return ret; 314 } 315 316 static inline void coroutine_fn 317 mirror_wait_for_free_in_flight_slot(MirrorBlockJob *s) 318 { 319 MirrorOp *op; 320 321 QTAILQ_FOREACH(op, &s->ops_in_flight, next) { 322 /* 323 * Do not wait on pseudo ops, because it may in turn wait on 324 * some other operation to start, which may in fact be the 325 * caller of this function. Since there is only one pseudo op 326 * at any given time, we will always find some real operation 327 * to wait on. 328 * Also, do not wait on active operations, because they do not 329 * use up in-flight slots. 330 */ 331 if (!op->is_pseudo_op && op->is_in_flight && !op->is_active_write) { 332 qemu_co_queue_wait(&op->waiting_requests, NULL); 333 return; 334 } 335 } 336 abort(); 337 } 338 339 /* Perform a mirror copy operation. 340 * 341 * *op->bytes_handled is set to the number of bytes copied after and 342 * including offset, excluding any bytes copied prior to offset due 343 * to alignment. This will be op->bytes if no alignment is necessary, 344 * or (new_end - op->offset) if the tail is rounded up or down due to 345 * alignment or buffer limit. 346 */ 347 static void coroutine_fn mirror_co_read(void *opaque) 348 { 349 MirrorOp *op = opaque; 350 MirrorBlockJob *s = op->s; 351 int nb_chunks; 352 uint64_t ret; 353 uint64_t max_bytes; 354 355 max_bytes = s->granularity * s->max_iov; 356 357 /* We can only handle as much as buf_size at a time. */ 358 op->bytes = MIN(s->buf_size, MIN(max_bytes, op->bytes)); 359 assert(op->bytes); 360 assert(op->bytes < BDRV_REQUEST_MAX_BYTES); 361 *op->bytes_handled = op->bytes; 362 363 if (s->cow_bitmap) { 364 *op->bytes_handled += mirror_cow_align(s, &op->offset, &op->bytes); 365 } 366 /* Cannot exceed BDRV_REQUEST_MAX_BYTES + INT_MAX */ 367 assert(*op->bytes_handled <= UINT_MAX); 368 assert(op->bytes <= s->buf_size); 369 /* The offset is granularity-aligned because: 370 * 1) Caller passes in aligned values; 371 * 2) mirror_cow_align is used only when target cluster is larger. */ 372 assert(QEMU_IS_ALIGNED(op->offset, s->granularity)); 373 /* The range is sector-aligned, since bdrv_getlength() rounds up. */ 374 assert(QEMU_IS_ALIGNED(op->bytes, BDRV_SECTOR_SIZE)); 375 nb_chunks = DIV_ROUND_UP(op->bytes, s->granularity); 376 377 while (s->buf_free_count < nb_chunks) { 378 trace_mirror_yield_in_flight(s, op->offset, s->in_flight); 379 mirror_wait_for_free_in_flight_slot(s); 380 } 381 382 /* Now make a QEMUIOVector taking enough granularity-sized chunks 383 * from s->buf_free. 384 */ 385 qemu_iovec_init(&op->qiov, nb_chunks); 386 while (nb_chunks-- > 0) { 387 MirrorBuffer *buf = QSIMPLEQ_FIRST(&s->buf_free); 388 size_t remaining = op->bytes - op->qiov.size; 389 390 QSIMPLEQ_REMOVE_HEAD(&s->buf_free, next); 391 s->buf_free_count--; 392 qemu_iovec_add(&op->qiov, buf, MIN(s->granularity, remaining)); 393 } 394 395 /* Copy the dirty cluster. */ 396 s->in_flight++; 397 s->bytes_in_flight += op->bytes; 398 op->is_in_flight = true; 399 trace_mirror_one_iteration(s, op->offset, op->bytes); 400 401 WITH_GRAPH_RDLOCK_GUARD() { 402 ret = bdrv_co_preadv(s->mirror_top_bs->backing, op->offset, op->bytes, 403 &op->qiov, 0); 404 } 405 mirror_read_complete(op, ret); 406 } 407 408 static void coroutine_fn mirror_co_zero(void *opaque) 409 { 410 MirrorOp *op = opaque; 411 int ret; 412 413 op->s->in_flight++; 414 op->s->bytes_in_flight += op->bytes; 415 *op->bytes_handled = op->bytes; 416 op->is_in_flight = true; 417 418 ret = blk_co_pwrite_zeroes(op->s->target, op->offset, op->bytes, 419 op->s->unmap ? BDRV_REQ_MAY_UNMAP : 0); 420 mirror_write_complete(op, ret); 421 } 422 423 static void coroutine_fn mirror_co_discard(void *opaque) 424 { 425 MirrorOp *op = opaque; 426 int ret; 427 428 op->s->in_flight++; 429 op->s->bytes_in_flight += op->bytes; 430 *op->bytes_handled = op->bytes; 431 op->is_in_flight = true; 432 433 ret = blk_co_pdiscard(op->s->target, op->offset, op->bytes); 434 mirror_write_complete(op, ret); 435 } 436 437 static unsigned mirror_perform(MirrorBlockJob *s, int64_t offset, 438 unsigned bytes, MirrorMethod mirror_method) 439 { 440 MirrorOp *op; 441 Coroutine *co; 442 int64_t bytes_handled = -1; 443 444 op = g_new(MirrorOp, 1); 445 *op = (MirrorOp){ 446 .s = s, 447 .offset = offset, 448 .bytes = bytes, 449 .bytes_handled = &bytes_handled, 450 }; 451 qemu_co_queue_init(&op->waiting_requests); 452 453 switch (mirror_method) { 454 case MIRROR_METHOD_COPY: 455 co = qemu_coroutine_create(mirror_co_read, op); 456 break; 457 case MIRROR_METHOD_ZERO: 458 co = qemu_coroutine_create(mirror_co_zero, op); 459 break; 460 case MIRROR_METHOD_DISCARD: 461 co = qemu_coroutine_create(mirror_co_discard, op); 462 break; 463 default: 464 abort(); 465 } 466 op->co = co; 467 468 QTAILQ_INSERT_TAIL(&s->ops_in_flight, op, next); 469 qemu_coroutine_enter(co); 470 /* At this point, ownership of op has been moved to the coroutine 471 * and the object may already be freed */ 472 473 /* Assert that this value has been set */ 474 assert(bytes_handled >= 0); 475 476 /* Same assertion as in mirror_co_read() (and for mirror_co_read() 477 * and mirror_co_discard(), bytes_handled == op->bytes, which 478 * is the @bytes parameter given to this function) */ 479 assert(bytes_handled <= UINT_MAX); 480 return bytes_handled; 481 } 482 483 static void coroutine_fn GRAPH_UNLOCKED mirror_iteration(MirrorBlockJob *s) 484 { 485 BlockDriverState *source; 486 MirrorOp *pseudo_op; 487 int64_t offset; 488 /* At least the first dirty chunk is mirrored in one iteration. */ 489 int nb_chunks = 1; 490 bool write_zeroes_ok = bdrv_can_write_zeroes_with_unmap(blk_bs(s->target)); 491 int max_io_bytes = MAX(s->buf_size / MAX_IN_FLIGHT, MAX_IO_BYTES); 492 493 bdrv_graph_co_rdlock(); 494 source = s->mirror_top_bs->backing->bs; 495 bdrv_graph_co_rdunlock(); 496 497 bdrv_dirty_bitmap_lock(s->dirty_bitmap); 498 offset = bdrv_dirty_iter_next(s->dbi); 499 if (offset < 0) { 500 bdrv_set_dirty_iter(s->dbi, 0); 501 offset = bdrv_dirty_iter_next(s->dbi); 502 trace_mirror_restart_iter(s, bdrv_get_dirty_count(s->dirty_bitmap)); 503 assert(offset >= 0); 504 } 505 bdrv_dirty_bitmap_unlock(s->dirty_bitmap); 506 507 /* 508 * Wait for concurrent requests to @offset. The next loop will limit the 509 * copied area based on in_flight_bitmap so we only copy an area that does 510 * not overlap with concurrent in-flight requests. Still, we would like to 511 * copy something, so wait until there are at least no more requests to the 512 * very beginning of the area. 513 */ 514 mirror_wait_on_conflicts(NULL, s, offset, 1); 515 516 job_pause_point(&s->common.job); 517 518 /* Find the number of consecutive dirty chunks following the first dirty 519 * one, and wait for in flight requests in them. */ 520 bdrv_dirty_bitmap_lock(s->dirty_bitmap); 521 while (nb_chunks * s->granularity < s->buf_size) { 522 int64_t next_dirty; 523 int64_t next_offset = offset + nb_chunks * s->granularity; 524 int64_t next_chunk = next_offset / s->granularity; 525 if (next_offset >= s->bdev_length || 526 !bdrv_dirty_bitmap_get_locked(s->dirty_bitmap, next_offset)) { 527 break; 528 } 529 if (test_bit(next_chunk, s->in_flight_bitmap)) { 530 break; 531 } 532 533 next_dirty = bdrv_dirty_iter_next(s->dbi); 534 if (next_dirty > next_offset || next_dirty < 0) { 535 /* The bitmap iterator's cache is stale, refresh it */ 536 bdrv_set_dirty_iter(s->dbi, next_offset); 537 next_dirty = bdrv_dirty_iter_next(s->dbi); 538 } 539 assert(next_dirty == next_offset); 540 nb_chunks++; 541 } 542 543 /* Clear dirty bits before querying the block status, because 544 * calling bdrv_block_status_above could yield - if some blocks are 545 * marked dirty in this window, we need to know. 546 */ 547 bdrv_reset_dirty_bitmap_locked(s->dirty_bitmap, offset, 548 nb_chunks * s->granularity); 549 bdrv_dirty_bitmap_unlock(s->dirty_bitmap); 550 551 /* Before claiming an area in the in-flight bitmap, we have to 552 * create a MirrorOp for it so that conflicting requests can wait 553 * for it. mirror_perform() will create the real MirrorOps later, 554 * for now we just create a pseudo operation that will wake up all 555 * conflicting requests once all real operations have been 556 * launched. */ 557 pseudo_op = g_new(MirrorOp, 1); 558 *pseudo_op = (MirrorOp){ 559 .offset = offset, 560 .bytes = nb_chunks * s->granularity, 561 .is_pseudo_op = true, 562 }; 563 qemu_co_queue_init(&pseudo_op->waiting_requests); 564 QTAILQ_INSERT_TAIL(&s->ops_in_flight, pseudo_op, next); 565 566 bitmap_set(s->in_flight_bitmap, offset / s->granularity, nb_chunks); 567 while (nb_chunks > 0 && offset < s->bdev_length) { 568 int ret; 569 int64_t io_bytes; 570 int64_t io_bytes_acct; 571 MirrorMethod mirror_method = MIRROR_METHOD_COPY; 572 573 assert(!(offset % s->granularity)); 574 WITH_GRAPH_RDLOCK_GUARD() { 575 ret = bdrv_co_block_status_above(source, NULL, offset, 576 nb_chunks * s->granularity, 577 &io_bytes, NULL, NULL); 578 } 579 if (ret < 0) { 580 io_bytes = MIN(nb_chunks * s->granularity, max_io_bytes); 581 } else if (ret & BDRV_BLOCK_DATA) { 582 io_bytes = MIN(io_bytes, max_io_bytes); 583 } 584 585 io_bytes -= io_bytes % s->granularity; 586 if (io_bytes < s->granularity) { 587 io_bytes = s->granularity; 588 } else if (ret >= 0 && !(ret & BDRV_BLOCK_DATA)) { 589 int64_t target_offset; 590 int64_t target_bytes; 591 WITH_GRAPH_RDLOCK_GUARD() { 592 bdrv_round_to_subclusters(blk_bs(s->target), offset, io_bytes, 593 &target_offset, &target_bytes); 594 } 595 if (target_offset == offset && 596 target_bytes == io_bytes) { 597 mirror_method = ret & BDRV_BLOCK_ZERO ? 598 MIRROR_METHOD_ZERO : 599 MIRROR_METHOD_DISCARD; 600 } 601 } 602 603 while (s->in_flight >= MAX_IN_FLIGHT) { 604 trace_mirror_yield_in_flight(s, offset, s->in_flight); 605 mirror_wait_for_free_in_flight_slot(s); 606 } 607 608 if (s->ret < 0) { 609 ret = 0; 610 goto fail; 611 } 612 613 io_bytes = mirror_clip_bytes(s, offset, io_bytes); 614 io_bytes = mirror_perform(s, offset, io_bytes, mirror_method); 615 if (mirror_method != MIRROR_METHOD_COPY && write_zeroes_ok) { 616 io_bytes_acct = 0; 617 } else { 618 io_bytes_acct = io_bytes; 619 } 620 assert(io_bytes); 621 offset += io_bytes; 622 nb_chunks -= DIV_ROUND_UP(io_bytes, s->granularity); 623 block_job_ratelimit_processed_bytes(&s->common, io_bytes_acct); 624 } 625 626 fail: 627 QTAILQ_REMOVE(&s->ops_in_flight, pseudo_op, next); 628 qemu_co_queue_restart_all(&pseudo_op->waiting_requests); 629 g_free(pseudo_op); 630 } 631 632 static void mirror_free_init(MirrorBlockJob *s) 633 { 634 int granularity = s->granularity; 635 size_t buf_size = s->buf_size; 636 uint8_t *buf = s->buf; 637 638 assert(s->buf_free_count == 0); 639 QSIMPLEQ_INIT(&s->buf_free); 640 while (buf_size != 0) { 641 MirrorBuffer *cur = (MirrorBuffer *)buf; 642 QSIMPLEQ_INSERT_TAIL(&s->buf_free, cur, next); 643 s->buf_free_count++; 644 buf_size -= granularity; 645 buf += granularity; 646 } 647 } 648 649 /* This is also used for the .pause callback. There is no matching 650 * mirror_resume() because mirror_run() will begin iterating again 651 * when the job is resumed. 652 */ 653 static void coroutine_fn mirror_wait_for_all_io(MirrorBlockJob *s) 654 { 655 while (s->in_flight > 0) { 656 mirror_wait_for_free_in_flight_slot(s); 657 } 658 } 659 660 /** 661 * mirror_exit_common: handle both abort() and prepare() cases. 662 * for .prepare, returns 0 on success and -errno on failure. 663 * for .abort cases, denoted by abort = true, MUST return 0. 664 */ 665 static int mirror_exit_common(Job *job) 666 { 667 MirrorBlockJob *s = container_of(job, MirrorBlockJob, common.job); 668 BlockJob *bjob = &s->common; 669 MirrorBDSOpaque *bs_opaque; 670 BlockDriverState *src; 671 BlockDriverState *target_bs; 672 BlockDriverState *mirror_top_bs; 673 Error *local_err = NULL; 674 bool abort = job->ret < 0; 675 int ret = 0; 676 677 GLOBAL_STATE_CODE(); 678 679 if (s->prepared) { 680 return 0; 681 } 682 s->prepared = true; 683 684 bdrv_graph_rdlock_main_loop(); 685 686 mirror_top_bs = s->mirror_top_bs; 687 bs_opaque = mirror_top_bs->opaque; 688 src = mirror_top_bs->backing->bs; 689 target_bs = blk_bs(s->target); 690 691 if (bdrv_chain_contains(src, target_bs)) { 692 bdrv_unfreeze_backing_chain(mirror_top_bs, target_bs); 693 } 694 695 bdrv_release_dirty_bitmap(s->dirty_bitmap); 696 697 /* Make sure that the source BDS doesn't go away during bdrv_replace_node, 698 * before we can call bdrv_drained_end */ 699 bdrv_ref(src); 700 bdrv_ref(mirror_top_bs); 701 bdrv_ref(target_bs); 702 703 bdrv_graph_rdunlock_main_loop(); 704 705 /* 706 * Remove target parent that still uses BLK_PERM_WRITE/RESIZE before 707 * inserting target_bs at s->to_replace, where we might not be able to get 708 * these permissions. 709 */ 710 blk_unref(s->target); 711 s->target = NULL; 712 713 /* We don't access the source any more. Dropping any WRITE/RESIZE is 714 * required before it could become a backing file of target_bs. Not having 715 * these permissions any more means that we can't allow any new requests on 716 * mirror_top_bs from now on, so keep it drained. */ 717 bdrv_drained_begin(mirror_top_bs); 718 bdrv_drained_begin(target_bs); 719 bs_opaque->stop = true; 720 721 bdrv_graph_rdlock_main_loop(); 722 bdrv_child_refresh_perms(mirror_top_bs, mirror_top_bs->backing, 723 &error_abort); 724 725 if (!abort && s->backing_mode == MIRROR_SOURCE_BACKING_CHAIN) { 726 BlockDriverState *backing = s->is_none_mode ? src : s->base; 727 BlockDriverState *unfiltered_target = bdrv_skip_filters(target_bs); 728 729 if (bdrv_cow_bs(unfiltered_target) != backing) { 730 bdrv_set_backing_hd(unfiltered_target, backing, &local_err); 731 if (local_err) { 732 error_report_err(local_err); 733 local_err = NULL; 734 ret = -EPERM; 735 } 736 } 737 } else if (!abort && s->backing_mode == MIRROR_OPEN_BACKING_CHAIN) { 738 assert(!bdrv_backing_chain_next(target_bs)); 739 ret = bdrv_open_backing_file(bdrv_skip_filters(target_bs), NULL, 740 "backing", &local_err); 741 if (ret < 0) { 742 error_report_err(local_err); 743 local_err = NULL; 744 } 745 } 746 bdrv_graph_rdunlock_main_loop(); 747 748 if (s->should_complete && !abort) { 749 BlockDriverState *to_replace = s->to_replace ?: src; 750 bool ro = bdrv_is_read_only(to_replace); 751 752 if (ro != bdrv_is_read_only(target_bs)) { 753 bdrv_reopen_set_read_only(target_bs, ro, NULL); 754 } 755 756 /* The mirror job has no requests in flight any more, but we need to 757 * drain potential other users of the BDS before changing the graph. */ 758 assert(s->in_drain); 759 bdrv_drained_begin(to_replace); 760 /* 761 * Cannot use check_to_replace_node() here, because that would 762 * check for an op blocker on @to_replace, and we have our own 763 * there. 764 */ 765 bdrv_graph_wrlock(); 766 if (bdrv_recurse_can_replace(src, to_replace)) { 767 bdrv_replace_node(to_replace, target_bs, &local_err); 768 } else { 769 error_setg(&local_err, "Can no longer replace '%s' by '%s', " 770 "because it can no longer be guaranteed that doing so " 771 "would not lead to an abrupt change of visible data", 772 to_replace->node_name, target_bs->node_name); 773 } 774 bdrv_graph_wrunlock(); 775 bdrv_drained_end(to_replace); 776 if (local_err) { 777 error_report_err(local_err); 778 ret = -EPERM; 779 } 780 } 781 if (s->to_replace) { 782 bdrv_op_unblock_all(s->to_replace, s->replace_blocker); 783 error_free(s->replace_blocker); 784 bdrv_unref(s->to_replace); 785 } 786 g_free(s->replaces); 787 788 /* 789 * Remove the mirror filter driver from the graph. Before this, get rid of 790 * the blockers on the intermediate nodes so that the resulting state is 791 * valid. 792 */ 793 block_job_remove_all_bdrv(bjob); 794 bdrv_graph_wrlock(); 795 bdrv_replace_node(mirror_top_bs, mirror_top_bs->backing->bs, &error_abort); 796 bdrv_graph_wrunlock(); 797 798 if (abort && s->base_ro && !bdrv_is_read_only(target_bs)) { 799 bdrv_reopen_set_read_only(target_bs, true, NULL); 800 } 801 802 bdrv_drained_end(target_bs); 803 bdrv_unref(target_bs); 804 805 bs_opaque->job = NULL; 806 807 bdrv_drained_end(src); 808 bdrv_drained_end(mirror_top_bs); 809 s->in_drain = false; 810 bdrv_unref(mirror_top_bs); 811 bdrv_unref(src); 812 813 return ret; 814 } 815 816 static int mirror_prepare(Job *job) 817 { 818 return mirror_exit_common(job); 819 } 820 821 static void mirror_abort(Job *job) 822 { 823 int ret = mirror_exit_common(job); 824 assert(ret == 0); 825 } 826 827 static void coroutine_fn mirror_throttle(MirrorBlockJob *s) 828 { 829 int64_t now = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); 830 831 if (now - s->last_pause_ns > BLOCK_JOB_SLICE_TIME) { 832 s->last_pause_ns = now; 833 job_sleep_ns(&s->common.job, 0); 834 } else { 835 job_pause_point(&s->common.job); 836 } 837 } 838 839 static int coroutine_fn GRAPH_UNLOCKED mirror_dirty_init(MirrorBlockJob *s) 840 { 841 int64_t offset; 842 BlockDriverState *bs; 843 BlockDriverState *target_bs = blk_bs(s->target); 844 int ret; 845 int64_t count; 846 847 bdrv_graph_co_rdlock(); 848 bs = s->mirror_top_bs->backing->bs; 849 bdrv_graph_co_rdunlock(); 850 851 if (s->zero_target) { 852 if (!bdrv_can_write_zeroes_with_unmap(target_bs)) { 853 bdrv_set_dirty_bitmap(s->dirty_bitmap, 0, s->bdev_length); 854 return 0; 855 } 856 857 s->initial_zeroing_ongoing = true; 858 for (offset = 0; offset < s->bdev_length; ) { 859 int bytes = MIN(s->bdev_length - offset, 860 QEMU_ALIGN_DOWN(INT_MAX, s->granularity)); 861 862 mirror_throttle(s); 863 864 if (job_is_cancelled(&s->common.job)) { 865 s->initial_zeroing_ongoing = false; 866 return 0; 867 } 868 869 if (s->in_flight >= MAX_IN_FLIGHT) { 870 trace_mirror_yield(s, UINT64_MAX, s->buf_free_count, 871 s->in_flight); 872 mirror_wait_for_free_in_flight_slot(s); 873 continue; 874 } 875 876 mirror_perform(s, offset, bytes, MIRROR_METHOD_ZERO); 877 offset += bytes; 878 } 879 880 mirror_wait_for_all_io(s); 881 s->initial_zeroing_ongoing = false; 882 } 883 884 /* First part, loop on the sectors and initialize the dirty bitmap. */ 885 for (offset = 0; offset < s->bdev_length; ) { 886 /* Just to make sure we are not exceeding int limit. */ 887 int bytes = MIN(s->bdev_length - offset, 888 QEMU_ALIGN_DOWN(INT_MAX, s->granularity)); 889 890 mirror_throttle(s); 891 892 if (job_is_cancelled(&s->common.job)) { 893 return 0; 894 } 895 896 WITH_GRAPH_RDLOCK_GUARD() { 897 ret = bdrv_co_is_allocated_above(bs, s->base_overlay, true, offset, 898 bytes, &count); 899 } 900 if (ret < 0) { 901 return ret; 902 } 903 904 assert(count); 905 if (ret > 0) { 906 bdrv_set_dirty_bitmap(s->dirty_bitmap, offset, count); 907 } 908 offset += count; 909 } 910 return 0; 911 } 912 913 /* Called when going out of the streaming phase to flush the bulk of the 914 * data to the medium, or just before completing. 915 */ 916 static int coroutine_fn mirror_flush(MirrorBlockJob *s) 917 { 918 int ret = blk_co_flush(s->target); 919 if (ret < 0) { 920 if (mirror_error_action(s, false, -ret) == BLOCK_ERROR_ACTION_REPORT) { 921 s->ret = ret; 922 } 923 } 924 return ret; 925 } 926 927 static int coroutine_fn mirror_run(Job *job, Error **errp) 928 { 929 MirrorBlockJob *s = container_of(job, MirrorBlockJob, common.job); 930 BlockDriverState *bs; 931 MirrorBDSOpaque *mirror_top_opaque = s->mirror_top_bs->opaque; 932 BlockDriverState *target_bs = blk_bs(s->target); 933 bool need_drain = true; 934 BlockDeviceIoStatus iostatus; 935 int64_t length; 936 int64_t target_length; 937 BlockDriverInfo bdi; 938 char backing_filename[2]; /* we only need 2 characters because we are only 939 checking for a NULL string */ 940 int ret = 0; 941 942 bdrv_graph_co_rdlock(); 943 bs = bdrv_filter_bs(s->mirror_top_bs); 944 bdrv_graph_co_rdunlock(); 945 946 if (job_is_cancelled(&s->common.job)) { 947 goto immediate_exit; 948 } 949 950 bdrv_graph_co_rdlock(); 951 s->bdev_length = bdrv_co_getlength(bs); 952 bdrv_graph_co_rdunlock(); 953 954 if (s->bdev_length < 0) { 955 ret = s->bdev_length; 956 goto immediate_exit; 957 } 958 959 target_length = blk_co_getlength(s->target); 960 if (target_length < 0) { 961 ret = target_length; 962 goto immediate_exit; 963 } 964 965 /* Active commit must resize the base image if its size differs from the 966 * active layer. */ 967 if (s->base == blk_bs(s->target)) { 968 if (s->bdev_length > target_length) { 969 ret = blk_co_truncate(s->target, s->bdev_length, false, 970 PREALLOC_MODE_OFF, 0, NULL); 971 if (ret < 0) { 972 goto immediate_exit; 973 } 974 } 975 } else if (s->bdev_length != target_length) { 976 error_setg(errp, "Source and target image have different sizes"); 977 ret = -EINVAL; 978 goto immediate_exit; 979 } 980 981 if (s->bdev_length == 0) { 982 /* Transition to the READY state and wait for complete. */ 983 job_transition_to_ready(&s->common.job); 984 qatomic_set(&s->actively_synced, true); 985 while (!job_cancel_requested(&s->common.job) && !s->should_complete) { 986 job_yield(&s->common.job); 987 } 988 goto immediate_exit; 989 } 990 991 length = DIV_ROUND_UP(s->bdev_length, s->granularity); 992 s->in_flight_bitmap = bitmap_new(length); 993 994 /* If we have no backing file yet in the destination, we cannot let 995 * the destination do COW. Instead, we copy sectors around the 996 * dirty data if needed. We need a bitmap to do that. 997 */ 998 bdrv_get_backing_filename(target_bs, backing_filename, 999 sizeof(backing_filename)); 1000 bdrv_graph_co_rdlock(); 1001 if (!bdrv_co_get_info(target_bs, &bdi) && bdi.cluster_size) { 1002 s->target_cluster_size = bdi.cluster_size; 1003 } else { 1004 s->target_cluster_size = BDRV_SECTOR_SIZE; 1005 } 1006 if (backing_filename[0] && !bdrv_backing_chain_next(target_bs) && 1007 s->granularity < s->target_cluster_size) { 1008 s->buf_size = MAX(s->buf_size, s->target_cluster_size); 1009 s->cow_bitmap = bitmap_new(length); 1010 } 1011 s->max_iov = MIN(bs->bl.max_iov, target_bs->bl.max_iov); 1012 bdrv_graph_co_rdunlock(); 1013 1014 s->buf = qemu_try_blockalign(bs, s->buf_size); 1015 if (s->buf == NULL) { 1016 ret = -ENOMEM; 1017 goto immediate_exit; 1018 } 1019 1020 mirror_free_init(s); 1021 1022 s->last_pause_ns = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); 1023 if (!s->is_none_mode) { 1024 ret = mirror_dirty_init(s); 1025 if (ret < 0 || job_is_cancelled(&s->common.job)) { 1026 goto immediate_exit; 1027 } 1028 } 1029 1030 /* 1031 * Only now the job is fully initialised and mirror_top_bs should start 1032 * accessing it. 1033 */ 1034 mirror_top_opaque->job = s; 1035 1036 assert(!s->dbi); 1037 s->dbi = bdrv_dirty_iter_new(s->dirty_bitmap); 1038 for (;;) { 1039 int64_t cnt, delta; 1040 bool should_complete; 1041 1042 if (s->ret < 0) { 1043 ret = s->ret; 1044 goto immediate_exit; 1045 } 1046 1047 job_pause_point(&s->common.job); 1048 1049 if (job_is_cancelled(&s->common.job)) { 1050 ret = 0; 1051 goto immediate_exit; 1052 } 1053 1054 cnt = bdrv_get_dirty_count(s->dirty_bitmap); 1055 /* cnt is the number of dirty bytes remaining and s->bytes_in_flight is 1056 * the number of bytes currently being processed; together those are 1057 * the current remaining operation length */ 1058 job_progress_set_remaining(&s->common.job, 1059 s->bytes_in_flight + cnt + 1060 s->active_write_bytes_in_flight); 1061 1062 /* Note that even when no rate limit is applied we need to yield 1063 * periodically with no pending I/O so that bdrv_drain_all() returns. 1064 * We do so every BLKOCK_JOB_SLICE_TIME nanoseconds, or when there is 1065 * an error, or when the source is clean, whichever comes first. */ 1066 delta = qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - s->last_pause_ns; 1067 WITH_JOB_LOCK_GUARD() { 1068 iostatus = s->common.iostatus; 1069 } 1070 if (delta < BLOCK_JOB_SLICE_TIME && 1071 iostatus == BLOCK_DEVICE_IO_STATUS_OK) { 1072 if (s->in_flight >= MAX_IN_FLIGHT || s->buf_free_count == 0 || 1073 (cnt == 0 && s->in_flight > 0)) { 1074 trace_mirror_yield(s, cnt, s->buf_free_count, s->in_flight); 1075 mirror_wait_for_free_in_flight_slot(s); 1076 continue; 1077 } else if (cnt != 0) { 1078 mirror_iteration(s); 1079 } 1080 } 1081 1082 should_complete = false; 1083 if (s->in_flight == 0 && cnt == 0) { 1084 trace_mirror_before_flush(s); 1085 if (!job_is_ready(&s->common.job)) { 1086 if (mirror_flush(s) < 0) { 1087 /* Go check s->ret. */ 1088 continue; 1089 } 1090 /* We're out of the streaming phase. From now on, if the job 1091 * is cancelled we will actually complete all pending I/O and 1092 * report completion. This way, block-job-cancel will leave 1093 * the target in a consistent state. 1094 */ 1095 job_transition_to_ready(&s->common.job); 1096 } 1097 if (qatomic_read(&s->copy_mode) != MIRROR_COPY_MODE_BACKGROUND) { 1098 qatomic_set(&s->actively_synced, true); 1099 } 1100 1101 should_complete = s->should_complete || 1102 job_cancel_requested(&s->common.job); 1103 cnt = bdrv_get_dirty_count(s->dirty_bitmap); 1104 } 1105 1106 if (cnt == 0 && should_complete) { 1107 /* The dirty bitmap is not updated while operations are pending. 1108 * If we're about to exit, wait for pending operations before 1109 * calling bdrv_get_dirty_count(bs), or we may exit while the 1110 * source has dirty data to copy! 1111 * 1112 * Note that I/O can be submitted by the guest while 1113 * mirror_populate runs, so pause it now. Before deciding 1114 * whether to switch to target check one last time if I/O has 1115 * come in the meanwhile, and if not flush the data to disk. 1116 */ 1117 trace_mirror_before_drain(s, cnt); 1118 1119 s->in_drain = true; 1120 bdrv_drained_begin(bs); 1121 1122 /* Must be zero because we are drained */ 1123 assert(s->in_active_write_counter == 0); 1124 1125 cnt = bdrv_get_dirty_count(s->dirty_bitmap); 1126 if (cnt > 0 || mirror_flush(s) < 0) { 1127 bdrv_drained_end(bs); 1128 s->in_drain = false; 1129 continue; 1130 } 1131 1132 /* The two disks are in sync. Exit and report successful 1133 * completion. 1134 */ 1135 assert(QLIST_EMPTY(&bs->tracked_requests)); 1136 need_drain = false; 1137 break; 1138 } 1139 1140 if (job_is_ready(&s->common.job) && !should_complete) { 1141 if (s->in_flight == 0 && cnt == 0) { 1142 trace_mirror_before_sleep(s, cnt, job_is_ready(&s->common.job), 1143 BLOCK_JOB_SLICE_TIME); 1144 job_sleep_ns(&s->common.job, BLOCK_JOB_SLICE_TIME); 1145 } 1146 } else { 1147 block_job_ratelimit_sleep(&s->common); 1148 } 1149 s->last_pause_ns = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); 1150 } 1151 1152 immediate_exit: 1153 if (s->in_flight > 0) { 1154 /* We get here only if something went wrong. Either the job failed, 1155 * or it was cancelled prematurely so that we do not guarantee that 1156 * the target is a copy of the source. 1157 */ 1158 assert(ret < 0 || job_is_cancelled(&s->common.job)); 1159 assert(need_drain); 1160 mirror_wait_for_all_io(s); 1161 } 1162 1163 assert(s->in_flight == 0); 1164 qemu_vfree(s->buf); 1165 g_free(s->cow_bitmap); 1166 g_free(s->in_flight_bitmap); 1167 bdrv_dirty_iter_free(s->dbi); 1168 1169 if (need_drain) { 1170 s->in_drain = true; 1171 bdrv_drained_begin(bs); 1172 } 1173 1174 return ret; 1175 } 1176 1177 static void mirror_complete(Job *job, Error **errp) 1178 { 1179 MirrorBlockJob *s = container_of(job, MirrorBlockJob, common.job); 1180 1181 if (!job_is_ready(job)) { 1182 error_setg(errp, "The active block job '%s' cannot be completed", 1183 job->id); 1184 return; 1185 } 1186 1187 /* block all operations on to_replace bs */ 1188 if (s->replaces) { 1189 s->to_replace = bdrv_find_node(s->replaces); 1190 if (!s->to_replace) { 1191 error_setg(errp, "Node name '%s' not found", s->replaces); 1192 return; 1193 } 1194 1195 /* TODO Translate this into child freeze system. */ 1196 error_setg(&s->replace_blocker, 1197 "block device is in use by block-job-complete"); 1198 bdrv_op_block_all(s->to_replace, s->replace_blocker); 1199 bdrv_ref(s->to_replace); 1200 } 1201 1202 s->should_complete = true; 1203 1204 /* If the job is paused, it will be re-entered when it is resumed */ 1205 WITH_JOB_LOCK_GUARD() { 1206 if (!job->paused) { 1207 job_enter_cond_locked(job, NULL); 1208 } 1209 } 1210 } 1211 1212 static void coroutine_fn mirror_pause(Job *job) 1213 { 1214 MirrorBlockJob *s = container_of(job, MirrorBlockJob, common.job); 1215 1216 mirror_wait_for_all_io(s); 1217 } 1218 1219 static bool mirror_drained_poll(BlockJob *job) 1220 { 1221 MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); 1222 1223 /* If the job isn't paused nor cancelled, we can't be sure that it won't 1224 * issue more requests. We make an exception if we've reached this point 1225 * from one of our own drain sections, to avoid a deadlock waiting for 1226 * ourselves. 1227 */ 1228 WITH_JOB_LOCK_GUARD() { 1229 if (!s->common.job.paused && !job_is_cancelled_locked(&job->job) 1230 && !s->in_drain) { 1231 return true; 1232 } 1233 } 1234 1235 return !!s->in_flight; 1236 } 1237 1238 static bool mirror_cancel(Job *job, bool force) 1239 { 1240 MirrorBlockJob *s = container_of(job, MirrorBlockJob, common.job); 1241 BlockDriverState *target = blk_bs(s->target); 1242 1243 /* 1244 * Before the job is READY, we treat any cancellation like a 1245 * force-cancellation. 1246 */ 1247 force = force || !job_is_ready(job); 1248 1249 if (force) { 1250 bdrv_cancel_in_flight(target); 1251 } 1252 return force; 1253 } 1254 1255 static bool commit_active_cancel(Job *job, bool force) 1256 { 1257 /* Same as above in mirror_cancel() */ 1258 return force || !job_is_ready(job); 1259 } 1260 1261 static void mirror_change(BlockJob *job, BlockJobChangeOptions *opts, 1262 Error **errp) 1263 { 1264 MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); 1265 BlockJobChangeOptionsMirror *change_opts = &opts->u.mirror; 1266 MirrorCopyMode current; 1267 1268 /* 1269 * The implementation relies on the fact that copy_mode is only written 1270 * under the BQL. Otherwise, further synchronization would be required. 1271 */ 1272 1273 GLOBAL_STATE_CODE(); 1274 1275 if (qatomic_read(&s->copy_mode) == change_opts->copy_mode) { 1276 return; 1277 } 1278 1279 if (change_opts->copy_mode != MIRROR_COPY_MODE_WRITE_BLOCKING) { 1280 error_setg(errp, "Change to copy mode '%s' is not implemented", 1281 MirrorCopyMode_str(change_opts->copy_mode)); 1282 return; 1283 } 1284 1285 current = qatomic_cmpxchg(&s->copy_mode, MIRROR_COPY_MODE_BACKGROUND, 1286 change_opts->copy_mode); 1287 if (current != MIRROR_COPY_MODE_BACKGROUND) { 1288 error_setg(errp, "Expected current copy mode '%s', got '%s'", 1289 MirrorCopyMode_str(MIRROR_COPY_MODE_BACKGROUND), 1290 MirrorCopyMode_str(current)); 1291 } 1292 } 1293 1294 static void mirror_query(BlockJob *job, BlockJobInfo *info) 1295 { 1296 MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); 1297 1298 info->u.mirror = (BlockJobInfoMirror) { 1299 .actively_synced = qatomic_read(&s->actively_synced), 1300 }; 1301 } 1302 1303 static const BlockJobDriver mirror_job_driver = { 1304 .job_driver = { 1305 .instance_size = sizeof(MirrorBlockJob), 1306 .job_type = JOB_TYPE_MIRROR, 1307 .free = block_job_free, 1308 .user_resume = block_job_user_resume, 1309 .run = mirror_run, 1310 .prepare = mirror_prepare, 1311 .abort = mirror_abort, 1312 .pause = mirror_pause, 1313 .complete = mirror_complete, 1314 .cancel = mirror_cancel, 1315 }, 1316 .drained_poll = mirror_drained_poll, 1317 .change = mirror_change, 1318 .query = mirror_query, 1319 }; 1320 1321 static const BlockJobDriver commit_active_job_driver = { 1322 .job_driver = { 1323 .instance_size = sizeof(MirrorBlockJob), 1324 .job_type = JOB_TYPE_COMMIT, 1325 .free = block_job_free, 1326 .user_resume = block_job_user_resume, 1327 .run = mirror_run, 1328 .prepare = mirror_prepare, 1329 .abort = mirror_abort, 1330 .pause = mirror_pause, 1331 .complete = mirror_complete, 1332 .cancel = commit_active_cancel, 1333 }, 1334 .drained_poll = mirror_drained_poll, 1335 }; 1336 1337 static void coroutine_fn 1338 do_sync_target_write(MirrorBlockJob *job, MirrorMethod method, 1339 uint64_t offset, uint64_t bytes, 1340 QEMUIOVector *qiov, int flags) 1341 { 1342 int ret; 1343 size_t qiov_offset = 0; 1344 int64_t bitmap_offset, bitmap_end; 1345 1346 if (!QEMU_IS_ALIGNED(offset, job->granularity) && 1347 bdrv_dirty_bitmap_get(job->dirty_bitmap, offset)) 1348 { 1349 /* 1350 * Dirty unaligned padding: ignore it. 1351 * 1352 * Reasoning: 1353 * 1. If we copy it, we can't reset corresponding bit in 1354 * dirty_bitmap as there may be some "dirty" bytes still not 1355 * copied. 1356 * 2. It's already dirty, so skipping it we don't diverge mirror 1357 * progress. 1358 * 1359 * Note, that because of this, guest write may have no contribution 1360 * into mirror converge, but that's not bad, as we have background 1361 * process of mirroring. If under some bad circumstances (high guest 1362 * IO load) background process starve, we will not converge anyway, 1363 * even if each write will contribute, as guest is not guaranteed to 1364 * rewrite the whole disk. 1365 */ 1366 qiov_offset = QEMU_ALIGN_UP(offset, job->granularity) - offset; 1367 if (bytes <= qiov_offset) { 1368 /* nothing to do after shrink */ 1369 return; 1370 } 1371 offset += qiov_offset; 1372 bytes -= qiov_offset; 1373 } 1374 1375 if (!QEMU_IS_ALIGNED(offset + bytes, job->granularity) && 1376 bdrv_dirty_bitmap_get(job->dirty_bitmap, offset + bytes - 1)) 1377 { 1378 uint64_t tail = (offset + bytes) % job->granularity; 1379 1380 if (bytes <= tail) { 1381 /* nothing to do after shrink */ 1382 return; 1383 } 1384 bytes -= tail; 1385 } 1386 1387 /* 1388 * Tails are either clean or shrunk, so for bitmap resetting 1389 * we safely align the range down. 1390 */ 1391 bitmap_offset = QEMU_ALIGN_UP(offset, job->granularity); 1392 bitmap_end = QEMU_ALIGN_DOWN(offset + bytes, job->granularity); 1393 if (bitmap_offset < bitmap_end) { 1394 bdrv_reset_dirty_bitmap(job->dirty_bitmap, bitmap_offset, 1395 bitmap_end - bitmap_offset); 1396 } 1397 1398 job_progress_increase_remaining(&job->common.job, bytes); 1399 job->active_write_bytes_in_flight += bytes; 1400 1401 switch (method) { 1402 case MIRROR_METHOD_COPY: 1403 ret = blk_co_pwritev_part(job->target, offset, bytes, 1404 qiov, qiov_offset, flags); 1405 break; 1406 1407 case MIRROR_METHOD_ZERO: 1408 assert(!qiov); 1409 ret = blk_co_pwrite_zeroes(job->target, offset, bytes, flags); 1410 break; 1411 1412 case MIRROR_METHOD_DISCARD: 1413 assert(!qiov); 1414 ret = blk_co_pdiscard(job->target, offset, bytes); 1415 break; 1416 1417 default: 1418 abort(); 1419 } 1420 1421 job->active_write_bytes_in_flight -= bytes; 1422 if (ret >= 0) { 1423 job_progress_update(&job->common.job, bytes); 1424 } else { 1425 BlockErrorAction action; 1426 1427 /* 1428 * We failed, so we should mark dirty the whole area, aligned up. 1429 * Note that we don't care about shrunk tails if any: they were dirty 1430 * at function start, and they must be still dirty, as we've locked 1431 * the region for in-flight op. 1432 */ 1433 bitmap_offset = QEMU_ALIGN_DOWN(offset, job->granularity); 1434 bitmap_end = QEMU_ALIGN_UP(offset + bytes, job->granularity); 1435 bdrv_set_dirty_bitmap(job->dirty_bitmap, bitmap_offset, 1436 bitmap_end - bitmap_offset); 1437 qatomic_set(&job->actively_synced, false); 1438 1439 action = mirror_error_action(job, false, -ret); 1440 if (action == BLOCK_ERROR_ACTION_REPORT) { 1441 if (!job->ret) { 1442 job->ret = ret; 1443 } 1444 } 1445 } 1446 } 1447 1448 static MirrorOp *coroutine_fn active_write_prepare(MirrorBlockJob *s, 1449 uint64_t offset, 1450 uint64_t bytes) 1451 { 1452 MirrorOp *op; 1453 uint64_t start_chunk = offset / s->granularity; 1454 uint64_t end_chunk = DIV_ROUND_UP(offset + bytes, s->granularity); 1455 1456 op = g_new(MirrorOp, 1); 1457 *op = (MirrorOp){ 1458 .s = s, 1459 .offset = offset, 1460 .bytes = bytes, 1461 .is_active_write = true, 1462 .is_in_flight = true, 1463 .co = qemu_coroutine_self(), 1464 }; 1465 qemu_co_queue_init(&op->waiting_requests); 1466 QTAILQ_INSERT_TAIL(&s->ops_in_flight, op, next); 1467 1468 s->in_active_write_counter++; 1469 1470 /* 1471 * Wait for concurrent requests affecting the area. If there are already 1472 * running requests that are copying off now-to-be stale data in the area, 1473 * we must wait for them to finish before we begin writing fresh data to the 1474 * target so that the write operations appear in the correct order. 1475 * Note that background requests (see mirror_iteration()) in contrast only 1476 * wait for conflicting requests at the start of the dirty area, and then 1477 * (based on the in_flight_bitmap) truncate the area to copy so it will not 1478 * conflict with any requests beyond that. For active writes, however, we 1479 * cannot truncate that area. The request from our parent must be blocked 1480 * until the area is copied in full. Therefore, we must wait for the whole 1481 * area to become free of concurrent requests. 1482 */ 1483 mirror_wait_on_conflicts(op, s, offset, bytes); 1484 1485 bitmap_set(s->in_flight_bitmap, start_chunk, end_chunk - start_chunk); 1486 1487 return op; 1488 } 1489 1490 static void coroutine_fn GRAPH_RDLOCK active_write_settle(MirrorOp *op) 1491 { 1492 uint64_t start_chunk = op->offset / op->s->granularity; 1493 uint64_t end_chunk = DIV_ROUND_UP(op->offset + op->bytes, 1494 op->s->granularity); 1495 1496 if (!--op->s->in_active_write_counter && 1497 qatomic_read(&op->s->actively_synced)) { 1498 BdrvChild *source = op->s->mirror_top_bs->backing; 1499 1500 if (QLIST_FIRST(&source->bs->parents) == source && 1501 QLIST_NEXT(source, next_parent) == NULL) 1502 { 1503 /* Assert that we are back in sync once all active write 1504 * operations are settled. 1505 * Note that we can only assert this if the mirror node 1506 * is the source node's only parent. */ 1507 assert(!bdrv_get_dirty_count(op->s->dirty_bitmap)); 1508 } 1509 } 1510 bitmap_clear(op->s->in_flight_bitmap, start_chunk, end_chunk - start_chunk); 1511 QTAILQ_REMOVE(&op->s->ops_in_flight, op, next); 1512 qemu_co_queue_restart_all(&op->waiting_requests); 1513 g_free(op); 1514 } 1515 1516 static int coroutine_fn GRAPH_RDLOCK 1517 bdrv_mirror_top_preadv(BlockDriverState *bs, int64_t offset, int64_t bytes, 1518 QEMUIOVector *qiov, BdrvRequestFlags flags) 1519 { 1520 return bdrv_co_preadv(bs->backing, offset, bytes, qiov, flags); 1521 } 1522 1523 static bool should_copy_to_target(MirrorBDSOpaque *s) 1524 { 1525 return s->job && s->job->ret >= 0 && 1526 !job_is_cancelled(&s->job->common.job) && 1527 qatomic_read(&s->job->copy_mode) == MIRROR_COPY_MODE_WRITE_BLOCKING; 1528 } 1529 1530 static int coroutine_fn GRAPH_RDLOCK 1531 bdrv_mirror_top_do_write(BlockDriverState *bs, MirrorMethod method, 1532 bool copy_to_target, uint64_t offset, uint64_t bytes, 1533 QEMUIOVector *qiov, int flags) 1534 { 1535 MirrorOp *op = NULL; 1536 MirrorBDSOpaque *s = bs->opaque; 1537 int ret = 0; 1538 1539 if (copy_to_target) { 1540 op = active_write_prepare(s->job, offset, bytes); 1541 } 1542 1543 switch (method) { 1544 case MIRROR_METHOD_COPY: 1545 ret = bdrv_co_pwritev(bs->backing, offset, bytes, qiov, flags); 1546 break; 1547 1548 case MIRROR_METHOD_ZERO: 1549 ret = bdrv_co_pwrite_zeroes(bs->backing, offset, bytes, flags); 1550 break; 1551 1552 case MIRROR_METHOD_DISCARD: 1553 ret = bdrv_co_pdiscard(bs->backing, offset, bytes); 1554 break; 1555 1556 default: 1557 abort(); 1558 } 1559 1560 if (!copy_to_target && s->job && s->job->dirty_bitmap) { 1561 qatomic_set(&s->job->actively_synced, false); 1562 bdrv_set_dirty_bitmap(s->job->dirty_bitmap, offset, bytes); 1563 } 1564 1565 if (ret < 0) { 1566 goto out; 1567 } 1568 1569 if (copy_to_target) { 1570 do_sync_target_write(s->job, method, offset, bytes, qiov, flags); 1571 } 1572 1573 out: 1574 if (copy_to_target) { 1575 active_write_settle(op); 1576 } 1577 return ret; 1578 } 1579 1580 static int coroutine_fn GRAPH_RDLOCK 1581 bdrv_mirror_top_pwritev(BlockDriverState *bs, int64_t offset, int64_t bytes, 1582 QEMUIOVector *qiov, BdrvRequestFlags flags) 1583 { 1584 QEMUIOVector bounce_qiov; 1585 void *bounce_buf; 1586 int ret = 0; 1587 bool copy_to_target = should_copy_to_target(bs->opaque); 1588 1589 if (copy_to_target) { 1590 /* The guest might concurrently modify the data to write; but 1591 * the data on source and destination must match, so we have 1592 * to use a bounce buffer if we are going to write to the 1593 * target now. */ 1594 bounce_buf = qemu_blockalign(bs, bytes); 1595 iov_to_buf_full(qiov->iov, qiov->niov, 0, bounce_buf, bytes); 1596 1597 qemu_iovec_init(&bounce_qiov, 1); 1598 qemu_iovec_add(&bounce_qiov, bounce_buf, bytes); 1599 qiov = &bounce_qiov; 1600 1601 flags &= ~BDRV_REQ_REGISTERED_BUF; 1602 } 1603 1604 ret = bdrv_mirror_top_do_write(bs, MIRROR_METHOD_COPY, copy_to_target, 1605 offset, bytes, qiov, flags); 1606 1607 if (copy_to_target) { 1608 qemu_iovec_destroy(&bounce_qiov); 1609 qemu_vfree(bounce_buf); 1610 } 1611 1612 return ret; 1613 } 1614 1615 static int coroutine_fn GRAPH_RDLOCK bdrv_mirror_top_flush(BlockDriverState *bs) 1616 { 1617 if (bs->backing == NULL) { 1618 /* we can be here after failed bdrv_append in mirror_start_job */ 1619 return 0; 1620 } 1621 return bdrv_co_flush(bs->backing->bs); 1622 } 1623 1624 static int coroutine_fn GRAPH_RDLOCK 1625 bdrv_mirror_top_pwrite_zeroes(BlockDriverState *bs, int64_t offset, 1626 int64_t bytes, BdrvRequestFlags flags) 1627 { 1628 bool copy_to_target = should_copy_to_target(bs->opaque); 1629 return bdrv_mirror_top_do_write(bs, MIRROR_METHOD_ZERO, copy_to_target, 1630 offset, bytes, NULL, flags); 1631 } 1632 1633 static int coroutine_fn GRAPH_RDLOCK 1634 bdrv_mirror_top_pdiscard(BlockDriverState *bs, int64_t offset, int64_t bytes) 1635 { 1636 bool copy_to_target = should_copy_to_target(bs->opaque); 1637 return bdrv_mirror_top_do_write(bs, MIRROR_METHOD_DISCARD, copy_to_target, 1638 offset, bytes, NULL, 0); 1639 } 1640 1641 static void GRAPH_RDLOCK bdrv_mirror_top_refresh_filename(BlockDriverState *bs) 1642 { 1643 if (bs->backing == NULL) { 1644 /* we can be here after failed bdrv_attach_child in 1645 * bdrv_set_backing_hd */ 1646 return; 1647 } 1648 pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), 1649 bs->backing->bs->filename); 1650 } 1651 1652 static void bdrv_mirror_top_child_perm(BlockDriverState *bs, BdrvChild *c, 1653 BdrvChildRole role, 1654 BlockReopenQueue *reopen_queue, 1655 uint64_t perm, uint64_t shared, 1656 uint64_t *nperm, uint64_t *nshared) 1657 { 1658 MirrorBDSOpaque *s = bs->opaque; 1659 1660 if (s->stop) { 1661 /* 1662 * If the job is to be stopped, we do not need to forward 1663 * anything to the real image. 1664 */ 1665 *nperm = 0; 1666 *nshared = BLK_PERM_ALL; 1667 return; 1668 } 1669 1670 bdrv_default_perms(bs, c, role, reopen_queue, 1671 perm, shared, nperm, nshared); 1672 1673 if (s->is_commit) { 1674 /* 1675 * For commit jobs, we cannot take CONSISTENT_READ, because 1676 * that permission is unshared for everything above the base 1677 * node (except for filters on the base node). 1678 * We also have to force-share the WRITE permission, or 1679 * otherwise we would block ourselves at the base node (if 1680 * writes are blocked for a node, they are also blocked for 1681 * its backing file). 1682 * (We could also share RESIZE, because it may be needed for 1683 * the target if its size is less than the top node's; but 1684 * bdrv_default_perms_for_cow() automatically shares RESIZE 1685 * for backing nodes if WRITE is shared, so there is no need 1686 * to do it here.) 1687 */ 1688 *nperm &= ~BLK_PERM_CONSISTENT_READ; 1689 *nshared |= BLK_PERM_WRITE; 1690 } 1691 } 1692 1693 /* Dummy node that provides consistent read to its users without requiring it 1694 * from its backing file and that allows writes on the backing file chain. */ 1695 static BlockDriver bdrv_mirror_top = { 1696 .format_name = "mirror_top", 1697 .bdrv_co_preadv = bdrv_mirror_top_preadv, 1698 .bdrv_co_pwritev = bdrv_mirror_top_pwritev, 1699 .bdrv_co_pwrite_zeroes = bdrv_mirror_top_pwrite_zeroes, 1700 .bdrv_co_pdiscard = bdrv_mirror_top_pdiscard, 1701 .bdrv_co_flush = bdrv_mirror_top_flush, 1702 .bdrv_refresh_filename = bdrv_mirror_top_refresh_filename, 1703 .bdrv_child_perm = bdrv_mirror_top_child_perm, 1704 1705 .is_filter = true, 1706 .filtered_child_is_backing = true, 1707 }; 1708 1709 static BlockJob *mirror_start_job( 1710 const char *job_id, BlockDriverState *bs, 1711 int creation_flags, BlockDriverState *target, 1712 const char *replaces, int64_t speed, 1713 uint32_t granularity, int64_t buf_size, 1714 BlockMirrorBackingMode backing_mode, 1715 bool zero_target, 1716 BlockdevOnError on_source_error, 1717 BlockdevOnError on_target_error, 1718 bool unmap, 1719 BlockCompletionFunc *cb, 1720 void *opaque, 1721 const BlockJobDriver *driver, 1722 bool is_none_mode, BlockDriverState *base, 1723 bool auto_complete, const char *filter_node_name, 1724 bool is_mirror, MirrorCopyMode copy_mode, 1725 bool base_ro, 1726 Error **errp) 1727 { 1728 MirrorBlockJob *s; 1729 MirrorBDSOpaque *bs_opaque; 1730 BlockDriverState *mirror_top_bs; 1731 bool target_is_backing; 1732 uint64_t target_perms, target_shared_perms; 1733 int ret; 1734 1735 GLOBAL_STATE_CODE(); 1736 1737 if (granularity == 0) { 1738 granularity = bdrv_get_default_bitmap_granularity(target); 1739 } 1740 1741 assert(is_power_of_2(granularity)); 1742 1743 if (buf_size < 0) { 1744 error_setg(errp, "Invalid parameter 'buf-size'"); 1745 return NULL; 1746 } 1747 1748 if (buf_size == 0) { 1749 buf_size = DEFAULT_MIRROR_BUF_SIZE; 1750 } 1751 1752 bdrv_graph_rdlock_main_loop(); 1753 if (bdrv_skip_filters(bs) == bdrv_skip_filters(target)) { 1754 error_setg(errp, "Can't mirror node into itself"); 1755 bdrv_graph_rdunlock_main_loop(); 1756 return NULL; 1757 } 1758 1759 target_is_backing = bdrv_chain_contains(bs, target); 1760 bdrv_graph_rdunlock_main_loop(); 1761 1762 /* In the case of active commit, add dummy driver to provide consistent 1763 * reads on the top, while disabling it in the intermediate nodes, and make 1764 * the backing chain writable. */ 1765 mirror_top_bs = bdrv_new_open_driver(&bdrv_mirror_top, filter_node_name, 1766 BDRV_O_RDWR, errp); 1767 if (mirror_top_bs == NULL) { 1768 return NULL; 1769 } 1770 if (!filter_node_name) { 1771 mirror_top_bs->implicit = true; 1772 } 1773 1774 /* So that we can always drop this node */ 1775 mirror_top_bs->never_freeze = true; 1776 1777 mirror_top_bs->total_sectors = bs->total_sectors; 1778 mirror_top_bs->supported_write_flags = BDRV_REQ_WRITE_UNCHANGED; 1779 mirror_top_bs->supported_zero_flags = BDRV_REQ_WRITE_UNCHANGED | 1780 BDRV_REQ_NO_FALLBACK; 1781 bs_opaque = g_new0(MirrorBDSOpaque, 1); 1782 mirror_top_bs->opaque = bs_opaque; 1783 1784 bs_opaque->is_commit = target_is_backing; 1785 1786 bdrv_drained_begin(bs); 1787 ret = bdrv_append(mirror_top_bs, bs, errp); 1788 bdrv_drained_end(bs); 1789 1790 if (ret < 0) { 1791 bdrv_unref(mirror_top_bs); 1792 return NULL; 1793 } 1794 1795 /* Make sure that the source is not resized while the job is running */ 1796 s = block_job_create(job_id, driver, NULL, mirror_top_bs, 1797 BLK_PERM_CONSISTENT_READ, 1798 BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED | 1799 BLK_PERM_WRITE, speed, 1800 creation_flags, cb, opaque, errp); 1801 if (!s) { 1802 goto fail; 1803 } 1804 1805 /* The block job now has a reference to this node */ 1806 bdrv_unref(mirror_top_bs); 1807 1808 s->mirror_top_bs = mirror_top_bs; 1809 s->base_ro = base_ro; 1810 1811 /* No resize for the target either; while the mirror is still running, a 1812 * consistent read isn't necessarily possible. We could possibly allow 1813 * writes and graph modifications, though it would likely defeat the 1814 * purpose of a mirror, so leave them blocked for now. 1815 * 1816 * In the case of active commit, things look a bit different, though, 1817 * because the target is an already populated backing file in active use. 1818 * We can allow anything except resize there.*/ 1819 1820 target_perms = BLK_PERM_WRITE; 1821 target_shared_perms = BLK_PERM_WRITE_UNCHANGED; 1822 1823 if (target_is_backing) { 1824 int64_t bs_size, target_size; 1825 bs_size = bdrv_getlength(bs); 1826 if (bs_size < 0) { 1827 error_setg_errno(errp, -bs_size, 1828 "Could not inquire top image size"); 1829 goto fail; 1830 } 1831 1832 target_size = bdrv_getlength(target); 1833 if (target_size < 0) { 1834 error_setg_errno(errp, -target_size, 1835 "Could not inquire base image size"); 1836 goto fail; 1837 } 1838 1839 if (target_size < bs_size) { 1840 target_perms |= BLK_PERM_RESIZE; 1841 } 1842 1843 target_shared_perms |= BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE; 1844 } else { 1845 bdrv_graph_rdlock_main_loop(); 1846 if (bdrv_chain_contains(bs, bdrv_skip_filters(target))) { 1847 /* 1848 * We may want to allow this in the future, but it would 1849 * require taking some extra care. 1850 */ 1851 error_setg(errp, "Cannot mirror to a filter on top of a node in " 1852 "the source's backing chain"); 1853 bdrv_graph_rdunlock_main_loop(); 1854 goto fail; 1855 } 1856 bdrv_graph_rdunlock_main_loop(); 1857 } 1858 1859 s->target = blk_new(s->common.job.aio_context, 1860 target_perms, target_shared_perms); 1861 ret = blk_insert_bs(s->target, target, errp); 1862 if (ret < 0) { 1863 goto fail; 1864 } 1865 if (is_mirror) { 1866 /* XXX: Mirror target could be a NBD server of target QEMU in the case 1867 * of non-shared block migration. To allow migration completion, we 1868 * have to allow "inactivate" of the target BB. When that happens, we 1869 * know the job is drained, and the vcpus are stopped, so no write 1870 * operation will be performed. Block layer already has assertions to 1871 * ensure that. */ 1872 blk_set_force_allow_inactivate(s->target); 1873 } 1874 blk_set_allow_aio_context_change(s->target, true); 1875 blk_set_disable_request_queuing(s->target, true); 1876 1877 bdrv_graph_rdlock_main_loop(); 1878 s->replaces = g_strdup(replaces); 1879 s->on_source_error = on_source_error; 1880 s->on_target_error = on_target_error; 1881 s->is_none_mode = is_none_mode; 1882 s->backing_mode = backing_mode; 1883 s->zero_target = zero_target; 1884 qatomic_set(&s->copy_mode, copy_mode); 1885 s->base = base; 1886 s->base_overlay = bdrv_find_overlay(bs, base); 1887 s->granularity = granularity; 1888 s->buf_size = ROUND_UP(buf_size, granularity); 1889 s->unmap = unmap; 1890 if (auto_complete) { 1891 s->should_complete = true; 1892 } 1893 bdrv_graph_rdunlock_main_loop(); 1894 1895 s->dirty_bitmap = bdrv_create_dirty_bitmap(s->mirror_top_bs, granularity, 1896 NULL, errp); 1897 if (!s->dirty_bitmap) { 1898 goto fail; 1899 } 1900 1901 /* 1902 * The dirty bitmap is set by bdrv_mirror_top_do_write() when not in active 1903 * mode. 1904 */ 1905 bdrv_disable_dirty_bitmap(s->dirty_bitmap); 1906 1907 bdrv_graph_wrlock(); 1908 ret = block_job_add_bdrv(&s->common, "source", bs, 0, 1909 BLK_PERM_WRITE_UNCHANGED | BLK_PERM_WRITE | 1910 BLK_PERM_CONSISTENT_READ, 1911 errp); 1912 if (ret < 0) { 1913 bdrv_graph_wrunlock(); 1914 goto fail; 1915 } 1916 1917 /* Required permissions are already taken with blk_new() */ 1918 block_job_add_bdrv(&s->common, "target", target, 0, BLK_PERM_ALL, 1919 &error_abort); 1920 1921 /* In commit_active_start() all intermediate nodes disappear, so 1922 * any jobs in them must be blocked */ 1923 if (target_is_backing) { 1924 BlockDriverState *iter, *filtered_target; 1925 uint64_t iter_shared_perms; 1926 1927 /* 1928 * The topmost node with 1929 * bdrv_skip_filters(filtered_target) == bdrv_skip_filters(target) 1930 */ 1931 filtered_target = bdrv_cow_bs(bdrv_find_overlay(bs, target)); 1932 1933 assert(bdrv_skip_filters(filtered_target) == 1934 bdrv_skip_filters(target)); 1935 1936 /* 1937 * XXX BLK_PERM_WRITE needs to be allowed so we don't block 1938 * ourselves at s->base (if writes are blocked for a node, they are 1939 * also blocked for its backing file). The other options would be a 1940 * second filter driver above s->base (== target). 1941 */ 1942 iter_shared_perms = BLK_PERM_WRITE_UNCHANGED | BLK_PERM_WRITE; 1943 1944 for (iter = bdrv_filter_or_cow_bs(bs); iter != target; 1945 iter = bdrv_filter_or_cow_bs(iter)) 1946 { 1947 if (iter == filtered_target) { 1948 /* 1949 * From here on, all nodes are filters on the base. 1950 * This allows us to share BLK_PERM_CONSISTENT_READ. 1951 */ 1952 iter_shared_perms |= BLK_PERM_CONSISTENT_READ; 1953 } 1954 1955 ret = block_job_add_bdrv(&s->common, "intermediate node", iter, 0, 1956 iter_shared_perms, errp); 1957 if (ret < 0) { 1958 bdrv_graph_wrunlock(); 1959 goto fail; 1960 } 1961 } 1962 1963 if (bdrv_freeze_backing_chain(mirror_top_bs, target, errp) < 0) { 1964 bdrv_graph_wrunlock(); 1965 goto fail; 1966 } 1967 } 1968 bdrv_graph_wrunlock(); 1969 1970 QTAILQ_INIT(&s->ops_in_flight); 1971 1972 trace_mirror_start(bs, s, opaque); 1973 job_start(&s->common.job); 1974 1975 return &s->common; 1976 1977 fail: 1978 if (s) { 1979 /* Make sure this BDS does not go away until we have completed the graph 1980 * changes below */ 1981 bdrv_ref(mirror_top_bs); 1982 1983 g_free(s->replaces); 1984 blk_unref(s->target); 1985 bs_opaque->job = NULL; 1986 if (s->dirty_bitmap) { 1987 bdrv_release_dirty_bitmap(s->dirty_bitmap); 1988 } 1989 job_early_fail(&s->common.job); 1990 } 1991 1992 bs_opaque->stop = true; 1993 bdrv_drained_begin(bs); 1994 bdrv_graph_wrlock(); 1995 assert(mirror_top_bs->backing->bs == bs); 1996 bdrv_child_refresh_perms(mirror_top_bs, mirror_top_bs->backing, 1997 &error_abort); 1998 bdrv_replace_node(mirror_top_bs, bs, &error_abort); 1999 bdrv_graph_wrunlock(); 2000 bdrv_drained_end(bs); 2001 2002 bdrv_unref(mirror_top_bs); 2003 2004 return NULL; 2005 } 2006 2007 void mirror_start(const char *job_id, BlockDriverState *bs, 2008 BlockDriverState *target, const char *replaces, 2009 int creation_flags, int64_t speed, 2010 uint32_t granularity, int64_t buf_size, 2011 MirrorSyncMode mode, BlockMirrorBackingMode backing_mode, 2012 bool zero_target, 2013 BlockdevOnError on_source_error, 2014 BlockdevOnError on_target_error, 2015 bool unmap, const char *filter_node_name, 2016 MirrorCopyMode copy_mode, Error **errp) 2017 { 2018 bool is_none_mode; 2019 BlockDriverState *base; 2020 2021 GLOBAL_STATE_CODE(); 2022 2023 if ((mode == MIRROR_SYNC_MODE_INCREMENTAL) || 2024 (mode == MIRROR_SYNC_MODE_BITMAP)) { 2025 error_setg(errp, "Sync mode '%s' not supported", 2026 MirrorSyncMode_str(mode)); 2027 return; 2028 } 2029 2030 bdrv_graph_rdlock_main_loop(); 2031 is_none_mode = mode == MIRROR_SYNC_MODE_NONE; 2032 base = mode == MIRROR_SYNC_MODE_TOP ? bdrv_backing_chain_next(bs) : NULL; 2033 bdrv_graph_rdunlock_main_loop(); 2034 2035 mirror_start_job(job_id, bs, creation_flags, target, replaces, 2036 speed, granularity, buf_size, backing_mode, zero_target, 2037 on_source_error, on_target_error, unmap, NULL, NULL, 2038 &mirror_job_driver, is_none_mode, base, false, 2039 filter_node_name, true, copy_mode, false, errp); 2040 } 2041 2042 BlockJob *commit_active_start(const char *job_id, BlockDriverState *bs, 2043 BlockDriverState *base, int creation_flags, 2044 int64_t speed, BlockdevOnError on_error, 2045 const char *filter_node_name, 2046 BlockCompletionFunc *cb, void *opaque, 2047 bool auto_complete, Error **errp) 2048 { 2049 bool base_read_only; 2050 BlockJob *job; 2051 2052 GLOBAL_STATE_CODE(); 2053 2054 base_read_only = bdrv_is_read_only(base); 2055 2056 if (base_read_only) { 2057 if (bdrv_reopen_set_read_only(base, false, errp) < 0) { 2058 return NULL; 2059 } 2060 } 2061 2062 job = mirror_start_job( 2063 job_id, bs, creation_flags, base, NULL, speed, 0, 0, 2064 MIRROR_LEAVE_BACKING_CHAIN, false, 2065 on_error, on_error, true, cb, opaque, 2066 &commit_active_job_driver, false, base, auto_complete, 2067 filter_node_name, false, MIRROR_COPY_MODE_BACKGROUND, 2068 base_read_only, errp); 2069 if (!job) { 2070 goto error_restore_flags; 2071 } 2072 2073 return job; 2074 2075 error_restore_flags: 2076 /* ignore error and errp for bdrv_reopen, because we want to propagate 2077 * the original error */ 2078 if (base_read_only) { 2079 bdrv_reopen_set_read_only(base, true, NULL); 2080 } 2081 return NULL; 2082 } 2083