2 * Replication Block filter
4 * Copyright (c) 2016 HUAWEI TECHNOLOGIES CO., LTD.
5 * Copyright (c) 2016 Intel Corporation
6 * Copyright (c) 2016 FUJITSU LIMITED
11 * This work is licensed under the terms of the GNU GPL, version 2 or later.
12 * See the COPYING file in the top-level directory.
15 #include "qemu/osdep.h"
16 #include "qemu-common.h"
17 #include "block/nbd.h"
18 #include "block/blockjob.h"
19 #include "block/block_int.h"
20 #include "block/block_backup.h"
21 #include "sysemu/block-backend.h"
22 #include "qapi/error.h"
23 #include "replication.h"
26 BLOCK_REPLICATION_NONE, /* block replication is not started */
27 BLOCK_REPLICATION_RUNNING, /* block replication is running */
28 BLOCK_REPLICATION_FAILOVER, /* failover is running in background */
29 BLOCK_REPLICATION_FAILOVER_FAILED, /* failover failed */
30 BLOCK_REPLICATION_DONE, /* block replication is done */
33 typedef struct BDRVReplicationState {
35 ReplicationStage stage;
36 BdrvChild *active_disk;
37 BdrvChild *hidden_disk;
38 BdrvChild *secondary_disk;
42 int orig_hidden_flags;
43 int orig_secondary_flags;
45 } BDRVReplicationState;
47 static void replication_start(ReplicationState *rs, ReplicationMode mode,
49 static void replication_do_checkpoint(ReplicationState *rs, Error **errp);
50 static void replication_get_error(ReplicationState *rs, Error **errp);
51 static void replication_stop(ReplicationState *rs, bool failover,
54 #define REPLICATION_MODE "mode"
55 #define REPLICATION_TOP_ID "top-id"
56 static QemuOptsList replication_runtime_opts = {
57 .name = "replication",
58 .head = QTAILQ_HEAD_INITIALIZER(replication_runtime_opts.head),
61 .name = REPLICATION_MODE,
62 .type = QEMU_OPT_STRING,
65 .name = REPLICATION_TOP_ID,
66 .type = QEMU_OPT_STRING,
72 static ReplicationOps replication_ops = {
73 .start = replication_start,
74 .checkpoint = replication_do_checkpoint,
75 .get_error = replication_get_error,
76 .stop = replication_stop,
79 static int replication_open(BlockDriverState *bs, QDict *options,
80 int flags, Error **errp)
83 BDRVReplicationState *s = bs->opaque;
84 Error *local_err = NULL;
85 QemuOpts *opts = NULL;
89 bs->file = bdrv_open_child(NULL, options, "file", bs, &child_file,
96 opts = qemu_opts_create(&replication_runtime_opts, NULL, 0, &error_abort);
97 qemu_opts_absorb_qdict(opts, options, &local_err);
102 mode = qemu_opt_get(opts, REPLICATION_MODE);
104 error_setg(&local_err, "Missing the option mode");
108 if (!strcmp(mode, "primary")) {
109 s->mode = REPLICATION_MODE_PRIMARY;
110 top_id = qemu_opt_get(opts, REPLICATION_TOP_ID);
112 error_setg(&local_err, "The primary side does not support option top-id");
115 } else if (!strcmp(mode, "secondary")) {
116 s->mode = REPLICATION_MODE_SECONDARY;
117 top_id = qemu_opt_get(opts, REPLICATION_TOP_ID);
118 s->top_id = g_strdup(top_id);
120 error_setg(&local_err, "Missing the option top-id");
124 error_setg(&local_err,
125 "The option mode's value should be primary or secondary");
129 s->rs = replication_new(bs, &replication_ops);
135 error_propagate(errp, local_err);
140 static void replication_close(BlockDriverState *bs)
142 BDRVReplicationState *s = bs->opaque;
144 if (s->stage == BLOCK_REPLICATION_RUNNING) {
145 replication_stop(s->rs, false, NULL);
147 if (s->stage == BLOCK_REPLICATION_FAILOVER) {
148 block_job_cancel_sync(s->active_disk->bs->job);
151 if (s->mode == REPLICATION_MODE_SECONDARY) {
155 replication_remove(s->rs);
158 static void replication_child_perm(BlockDriverState *bs, BdrvChild *c,
159 const BdrvChildRole *role,
160 uint64_t perm, uint64_t shared,
161 uint64_t *nperm, uint64_t *nshared)
163 *nperm = *nshared = BLK_PERM_CONSISTENT_READ \
165 | BLK_PERM_WRITE_UNCHANGED;
170 static int64_t replication_getlength(BlockDriverState *bs)
172 return bdrv_getlength(bs->file->bs);
175 static int replication_get_io_status(BDRVReplicationState *s)
178 case BLOCK_REPLICATION_NONE:
180 case BLOCK_REPLICATION_RUNNING:
182 case BLOCK_REPLICATION_FAILOVER:
183 return s->mode == REPLICATION_MODE_PRIMARY ? -EIO : 0;
184 case BLOCK_REPLICATION_FAILOVER_FAILED:
185 return s->mode == REPLICATION_MODE_PRIMARY ? -EIO : 1;
186 case BLOCK_REPLICATION_DONE:
188 * active commit job completes, and active disk and secondary_disk
189 * is swapped, so we can operate bs->file directly
191 return s->mode == REPLICATION_MODE_PRIMARY ? -EIO : 0;
197 static int replication_return_value(BDRVReplicationState *s, int ret)
199 if (s->mode == REPLICATION_MODE_SECONDARY) {
211 static coroutine_fn int replication_co_readv(BlockDriverState *bs,
213 int remaining_sectors,
216 BDRVReplicationState *s = bs->opaque;
217 BdrvChild *child = s->secondary_disk;
218 BlockJob *job = NULL;
222 if (s->mode == REPLICATION_MODE_PRIMARY) {
223 /* We only use it to forward primary write requests */
227 ret = replication_get_io_status(s);
232 if (child && child->bs) {
233 job = child->bs->job;
237 backup_wait_for_overlapping_requests(child->bs->job, sector_num,
239 backup_cow_request_begin(&req, child->bs->job, sector_num,
241 ret = bdrv_co_readv(bs->file, sector_num, remaining_sectors,
243 backup_cow_request_end(&req);
247 ret = bdrv_co_readv(bs->file, sector_num, remaining_sectors, qiov);
249 return replication_return_value(s, ret);
252 static coroutine_fn int replication_co_writev(BlockDriverState *bs,
254 int remaining_sectors,
257 BDRVReplicationState *s = bs->opaque;
258 QEMUIOVector hd_qiov;
259 uint64_t bytes_done = 0;
260 BdrvChild *top = bs->file;
261 BdrvChild *base = s->secondary_disk;
265 ret = replication_get_io_status(s);
271 ret = bdrv_co_writev(top, sector_num,
272 remaining_sectors, qiov);
273 return replication_return_value(s, ret);
277 * Failover failed, only write to active disk if the sectors
278 * have already been allocated in active disk/hidden disk.
280 qemu_iovec_init(&hd_qiov, qiov->niov);
281 while (remaining_sectors > 0) {
282 ret = bdrv_is_allocated_above(top->bs, base->bs, sector_num,
283 remaining_sectors, &n);
288 qemu_iovec_reset(&hd_qiov);
289 qemu_iovec_concat(&hd_qiov, qiov, bytes_done, n * BDRV_SECTOR_SIZE);
291 target = ret ? top : base;
292 ret = bdrv_co_writev(target, sector_num, n, &hd_qiov);
297 remaining_sectors -= n;
299 bytes_done += n * BDRV_SECTOR_SIZE;
303 qemu_iovec_destroy(&hd_qiov);
308 static bool replication_recurse_is_first_non_filter(BlockDriverState *bs,
309 BlockDriverState *candidate)
311 return bdrv_recurse_is_first_non_filter(bs->file->bs, candidate);
314 static void secondary_do_checkpoint(BDRVReplicationState *s, Error **errp)
316 Error *local_err = NULL;
319 if (!s->secondary_disk->bs->job) {
320 error_setg(errp, "Backup job was cancelled unexpectedly");
324 backup_do_checkpoint(s->secondary_disk->bs->job, &local_err);
326 error_propagate(errp, local_err);
330 ret = s->active_disk->bs->drv->bdrv_make_empty(s->active_disk->bs);
332 error_setg(errp, "Cannot make active disk empty");
336 ret = s->hidden_disk->bs->drv->bdrv_make_empty(s->hidden_disk->bs);
338 error_setg(errp, "Cannot make hidden disk empty");
343 static void reopen_backing_file(BlockDriverState *bs, bool writable,
346 BDRVReplicationState *s = bs->opaque;
347 BlockReopenQueue *reopen_queue = NULL;
348 int orig_hidden_flags, orig_secondary_flags;
349 int new_hidden_flags, new_secondary_flags;
350 Error *local_err = NULL;
353 orig_hidden_flags = s->orig_hidden_flags =
354 bdrv_get_flags(s->hidden_disk->bs);
355 new_hidden_flags = (orig_hidden_flags | BDRV_O_RDWR) &
357 orig_secondary_flags = s->orig_secondary_flags =
358 bdrv_get_flags(s->secondary_disk->bs);
359 new_secondary_flags = (orig_secondary_flags | BDRV_O_RDWR) &
362 orig_hidden_flags = (s->orig_hidden_flags | BDRV_O_RDWR) &
364 new_hidden_flags = s->orig_hidden_flags;
365 orig_secondary_flags = (s->orig_secondary_flags | BDRV_O_RDWR) &
367 new_secondary_flags = s->orig_secondary_flags;
370 if (orig_hidden_flags != new_hidden_flags) {
371 reopen_queue = bdrv_reopen_queue(reopen_queue, s->hidden_disk->bs, NULL,
375 if (!(orig_secondary_flags & BDRV_O_RDWR)) {
376 reopen_queue = bdrv_reopen_queue(reopen_queue, s->secondary_disk->bs,
377 NULL, new_secondary_flags);
381 bdrv_reopen_multiple(bdrv_get_aio_context(bs),
382 reopen_queue, &local_err);
383 error_propagate(errp, local_err);
387 static void backup_job_cleanup(BlockDriverState *bs)
389 BDRVReplicationState *s = bs->opaque;
390 BlockDriverState *top_bs;
392 top_bs = bdrv_lookup_bs(s->top_id, s->top_id, NULL);
396 bdrv_op_unblock_all(top_bs, s->blocker);
397 error_free(s->blocker);
398 reopen_backing_file(bs, false, NULL);
401 static void backup_job_completed(void *opaque, int ret)
403 BlockDriverState *bs = opaque;
404 BDRVReplicationState *s = bs->opaque;
406 if (s->stage != BLOCK_REPLICATION_FAILOVER) {
407 /* The backup job is cancelled unexpectedly */
411 backup_job_cleanup(bs);
414 static bool check_top_bs(BlockDriverState *top_bs, BlockDriverState *bs)
418 /* The bs itself is the top_bs */
423 /* Iterate over top_bs's children */
424 QLIST_FOREACH(child, &top_bs->children, next) {
425 if (child->bs == bs || check_top_bs(child->bs, bs)) {
433 static void replication_start(ReplicationState *rs, ReplicationMode mode,
436 BlockDriverState *bs = rs->opaque;
437 BDRVReplicationState *s;
438 BlockDriverState *top_bs;
439 int64_t active_length, hidden_length, disk_length;
440 AioContext *aio_context;
441 Error *local_err = NULL;
444 aio_context = bdrv_get_aio_context(bs);
445 aio_context_acquire(aio_context);
448 if (s->stage != BLOCK_REPLICATION_NONE) {
449 error_setg(errp, "Block replication is running or done");
450 aio_context_release(aio_context);
454 if (s->mode != mode) {
455 error_setg(errp, "The parameter mode's value is invalid, needs %d,"
456 " but got %d", s->mode, mode);
457 aio_context_release(aio_context);
462 case REPLICATION_MODE_PRIMARY:
464 case REPLICATION_MODE_SECONDARY:
465 s->active_disk = bs->file;
466 if (!s->active_disk || !s->active_disk->bs ||
467 !s->active_disk->bs->backing) {
468 error_setg(errp, "Active disk doesn't have backing file");
469 aio_context_release(aio_context);
473 s->hidden_disk = s->active_disk->bs->backing;
474 if (!s->hidden_disk->bs || !s->hidden_disk->bs->backing) {
475 error_setg(errp, "Hidden disk doesn't have backing file");
476 aio_context_release(aio_context);
480 s->secondary_disk = s->hidden_disk->bs->backing;
481 if (!s->secondary_disk->bs || !bdrv_has_blk(s->secondary_disk->bs)) {
482 error_setg(errp, "The secondary disk doesn't have block backend");
483 aio_context_release(aio_context);
487 /* verify the length */
488 active_length = bdrv_getlength(s->active_disk->bs);
489 hidden_length = bdrv_getlength(s->hidden_disk->bs);
490 disk_length = bdrv_getlength(s->secondary_disk->bs);
491 if (active_length < 0 || hidden_length < 0 || disk_length < 0 ||
492 active_length != hidden_length || hidden_length != disk_length) {
493 error_setg(errp, "Active disk, hidden disk, secondary disk's length"
494 " are not the same");
495 aio_context_release(aio_context);
499 if (!s->active_disk->bs->drv->bdrv_make_empty ||
500 !s->hidden_disk->bs->drv->bdrv_make_empty) {
502 "Active disk or hidden disk doesn't support make_empty");
503 aio_context_release(aio_context);
507 /* reopen the backing file in r/w mode */
508 reopen_backing_file(bs, true, &local_err);
510 error_propagate(errp, local_err);
511 aio_context_release(aio_context);
515 /* start backup job now */
516 error_setg(&s->blocker,
517 "Block device is in use by internal backup job");
519 top_bs = bdrv_lookup_bs(s->top_id, s->top_id, NULL);
520 if (!top_bs || !bdrv_is_root_node(top_bs) ||
521 !check_top_bs(top_bs, bs)) {
522 error_setg(errp, "No top_bs or it is invalid");
523 reopen_backing_file(bs, false, NULL);
524 aio_context_release(aio_context);
527 bdrv_op_block_all(top_bs, s->blocker);
528 bdrv_op_unblock(top_bs, BLOCK_OP_TYPE_DATAPLANE, s->blocker);
530 job = backup_job_create(NULL, s->secondary_disk->bs, s->hidden_disk->bs,
531 0, MIRROR_SYNC_MODE_NONE, NULL, false,
532 BLOCKDEV_ON_ERROR_REPORT,
533 BLOCKDEV_ON_ERROR_REPORT, BLOCK_JOB_INTERNAL,
534 backup_job_completed, bs, NULL, &local_err);
536 error_propagate(errp, local_err);
537 backup_job_cleanup(bs);
538 aio_context_release(aio_context);
541 block_job_start(job);
544 aio_context_release(aio_context);
548 s->stage = BLOCK_REPLICATION_RUNNING;
550 if (s->mode == REPLICATION_MODE_SECONDARY) {
551 secondary_do_checkpoint(s, errp);
555 aio_context_release(aio_context);
558 static void replication_do_checkpoint(ReplicationState *rs, Error **errp)
560 BlockDriverState *bs = rs->opaque;
561 BDRVReplicationState *s;
562 AioContext *aio_context;
564 aio_context = bdrv_get_aio_context(bs);
565 aio_context_acquire(aio_context);
568 if (s->mode == REPLICATION_MODE_SECONDARY) {
569 secondary_do_checkpoint(s, errp);
571 aio_context_release(aio_context);
574 static void replication_get_error(ReplicationState *rs, Error **errp)
576 BlockDriverState *bs = rs->opaque;
577 BDRVReplicationState *s;
578 AioContext *aio_context;
580 aio_context = bdrv_get_aio_context(bs);
581 aio_context_acquire(aio_context);
584 if (s->stage != BLOCK_REPLICATION_RUNNING) {
585 error_setg(errp, "Block replication is not running");
586 aio_context_release(aio_context);
591 error_setg(errp, "I/O error occurred");
592 aio_context_release(aio_context);
595 aio_context_release(aio_context);
598 static void replication_done(void *opaque, int ret)
600 BlockDriverState *bs = opaque;
601 BDRVReplicationState *s = bs->opaque;
604 s->stage = BLOCK_REPLICATION_DONE;
606 /* refresh top bs's filename */
607 bdrv_refresh_filename(bs);
608 s->active_disk = NULL;
609 s->secondary_disk = NULL;
610 s->hidden_disk = NULL;
613 s->stage = BLOCK_REPLICATION_FAILOVER_FAILED;
618 static void replication_stop(ReplicationState *rs, bool failover, Error **errp)
620 BlockDriverState *bs = rs->opaque;
621 BDRVReplicationState *s;
622 AioContext *aio_context;
624 aio_context = bdrv_get_aio_context(bs);
625 aio_context_acquire(aio_context);
628 if (s->stage != BLOCK_REPLICATION_RUNNING) {
629 error_setg(errp, "Block replication is not running");
630 aio_context_release(aio_context);
635 case REPLICATION_MODE_PRIMARY:
636 s->stage = BLOCK_REPLICATION_DONE;
639 case REPLICATION_MODE_SECONDARY:
641 * This BDS will be closed, and the job should be completed
642 * before the BDS is closed, because we will access hidden
643 * disk, secondary disk in backup_job_completed().
645 if (s->secondary_disk->bs->job) {
646 block_job_cancel_sync(s->secondary_disk->bs->job);
650 secondary_do_checkpoint(s, errp);
651 s->stage = BLOCK_REPLICATION_DONE;
652 aio_context_release(aio_context);
656 s->stage = BLOCK_REPLICATION_FAILOVER;
657 commit_active_start(NULL, s->active_disk->bs, s->secondary_disk->bs,
658 BLOCK_JOB_INTERNAL, 0, BLOCKDEV_ON_ERROR_REPORT,
659 NULL, replication_done, bs, true, errp);
662 aio_context_release(aio_context);
665 aio_context_release(aio_context);
668 BlockDriver bdrv_replication = {
669 .format_name = "replication",
670 .protocol_name = "replication",
671 .instance_size = sizeof(BDRVReplicationState),
673 .bdrv_open = replication_open,
674 .bdrv_close = replication_close,
675 .bdrv_child_perm = replication_child_perm,
677 .bdrv_getlength = replication_getlength,
678 .bdrv_co_readv = replication_co_readv,
679 .bdrv_co_writev = replication_co_writev,
682 .bdrv_recurse_is_first_non_filter = replication_recurse_is_first_non_filter,
684 .has_variable_length = true,
687 static void bdrv_replication_init(void)
689 bdrv_register(&bdrv_replication);
692 block_init(bdrv_replication_init);