Merge tag 'qemu-macppc-20230206' of https://github.com/mcayland/qemu into staging
[qemu.git] / block / replication.c
bloba27417d310802d09c25ef6e9902daa594f02efed
1 /*
2 * Replication Block filter
4 * Copyright (c) 2016 HUAWEI TECHNOLOGIES CO., LTD.
5 * Copyright (c) 2016 Intel Corporation
6 * Copyright (c) 2016 FUJITSU LIMITED
8 * Author:
9 * Wen Congyang <wency@cn.fujitsu.com>
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/module.h"
17 #include "qemu/option.h"
18 #include "block/nbd.h"
19 #include "block/blockjob.h"
20 #include "block/block_int.h"
21 #include "block/block_backup.h"
22 #include "sysemu/block-backend.h"
23 #include "qapi/error.h"
24 #include "qapi/qmp/qdict.h"
25 #include "block/replication.h"
27 typedef enum {
28 BLOCK_REPLICATION_NONE, /* block replication is not started */
29 BLOCK_REPLICATION_RUNNING, /* block replication is running */
30 BLOCK_REPLICATION_FAILOVER, /* failover is running in background */
31 BLOCK_REPLICATION_FAILOVER_FAILED, /* failover failed */
32 BLOCK_REPLICATION_DONE, /* block replication is done */
33 } ReplicationStage;
35 typedef struct BDRVReplicationState {
36 ReplicationMode mode;
37 ReplicationStage stage;
38 BlockJob *commit_job;
39 BdrvChild *hidden_disk;
40 BdrvChild *secondary_disk;
41 BlockJob *backup_job;
42 char *top_id;
43 ReplicationState *rs;
44 Error *blocker;
45 bool orig_hidden_read_only;
46 bool orig_secondary_read_only;
47 int error;
48 } BDRVReplicationState;
50 static void replication_start(ReplicationState *rs, ReplicationMode mode,
51 Error **errp);
52 static void replication_do_checkpoint(ReplicationState *rs, Error **errp);
53 static void replication_get_error(ReplicationState *rs, Error **errp);
54 static void replication_stop(ReplicationState *rs, bool failover,
55 Error **errp);
57 #define REPLICATION_MODE "mode"
58 #define REPLICATION_TOP_ID "top-id"
59 static QemuOptsList replication_runtime_opts = {
60 .name = "replication",
61 .head = QTAILQ_HEAD_INITIALIZER(replication_runtime_opts.head),
62 .desc = {
64 .name = REPLICATION_MODE,
65 .type = QEMU_OPT_STRING,
68 .name = REPLICATION_TOP_ID,
69 .type = QEMU_OPT_STRING,
71 { /* end of list */ }
75 static ReplicationOps replication_ops = {
76 .start = replication_start,
77 .checkpoint = replication_do_checkpoint,
78 .get_error = replication_get_error,
79 .stop = replication_stop,
82 static int replication_open(BlockDriverState *bs, QDict *options,
83 int flags, Error **errp)
85 int ret;
86 BDRVReplicationState *s = bs->opaque;
87 QemuOpts *opts = NULL;
88 const char *mode;
89 const char *top_id;
91 ret = bdrv_open_file_child(NULL, options, "file", bs, errp);
92 if (ret < 0) {
93 return ret;
96 ret = -EINVAL;
97 opts = qemu_opts_create(&replication_runtime_opts, NULL, 0, &error_abort);
98 if (!qemu_opts_absorb_qdict(opts, options, errp)) {
99 goto fail;
102 mode = qemu_opt_get(opts, REPLICATION_MODE);
103 if (!mode) {
104 error_setg(errp, "Missing the option mode");
105 goto fail;
108 if (!strcmp(mode, "primary")) {
109 s->mode = REPLICATION_MODE_PRIMARY;
110 top_id = qemu_opt_get(opts, REPLICATION_TOP_ID);
111 if (top_id) {
112 error_setg(errp,
113 "The primary side does not support option top-id");
114 goto fail;
116 } else if (!strcmp(mode, "secondary")) {
117 s->mode = REPLICATION_MODE_SECONDARY;
118 top_id = qemu_opt_get(opts, REPLICATION_TOP_ID);
119 s->top_id = g_strdup(top_id);
120 if (!s->top_id) {
121 error_setg(errp, "Missing the option top-id");
122 goto fail;
124 } else {
125 error_setg(errp,
126 "The option mode's value should be primary or secondary");
127 goto fail;
130 s->rs = replication_new(bs, &replication_ops);
132 ret = 0;
134 fail:
135 qemu_opts_del(opts);
136 return ret;
139 static void replication_close(BlockDriverState *bs)
141 BDRVReplicationState *s = bs->opaque;
142 Job *commit_job;
143 GLOBAL_STATE_CODE();
145 if (s->stage == BLOCK_REPLICATION_RUNNING) {
146 replication_stop(s->rs, false, NULL);
148 if (s->stage == BLOCK_REPLICATION_FAILOVER) {
149 commit_job = &s->commit_job->job;
150 assert(commit_job->aio_context == qemu_get_current_aio_context());
151 job_cancel_sync(commit_job, false);
154 if (s->mode == REPLICATION_MODE_SECONDARY) {
155 g_free(s->top_id);
158 replication_remove(s->rs);
161 static void replication_child_perm(BlockDriverState *bs, BdrvChild *c,
162 BdrvChildRole role,
163 BlockReopenQueue *reopen_queue,
164 uint64_t perm, uint64_t shared,
165 uint64_t *nperm, uint64_t *nshared)
167 if (role & BDRV_CHILD_PRIMARY) {
168 *nperm = BLK_PERM_CONSISTENT_READ;
169 } else {
170 *nperm = 0;
173 if ((bs->open_flags & (BDRV_O_INACTIVE | BDRV_O_RDWR)) == BDRV_O_RDWR) {
174 *nperm |= BLK_PERM_WRITE;
176 *nshared = BLK_PERM_CONSISTENT_READ
177 | BLK_PERM_WRITE
178 | BLK_PERM_WRITE_UNCHANGED;
179 return;
182 static int64_t coroutine_fn replication_co_getlength(BlockDriverState *bs)
184 return bdrv_co_getlength(bs->file->bs);
187 static int replication_get_io_status(BDRVReplicationState *s)
189 switch (s->stage) {
190 case BLOCK_REPLICATION_NONE:
191 return -EIO;
192 case BLOCK_REPLICATION_RUNNING:
193 return 0;
194 case BLOCK_REPLICATION_FAILOVER:
195 return s->mode == REPLICATION_MODE_PRIMARY ? -EIO : 0;
196 case BLOCK_REPLICATION_FAILOVER_FAILED:
197 return s->mode == REPLICATION_MODE_PRIMARY ? -EIO : 1;
198 case BLOCK_REPLICATION_DONE:
200 * active commit job completes, and active disk and secondary_disk
201 * is swapped, so we can operate bs->file directly
203 return s->mode == REPLICATION_MODE_PRIMARY ? -EIO : 0;
204 default:
205 abort();
209 static int replication_return_value(BDRVReplicationState *s, int ret)
211 if (s->mode == REPLICATION_MODE_SECONDARY) {
212 return ret;
215 if (ret < 0) {
216 s->error = ret;
217 ret = 0;
220 return ret;
223 static coroutine_fn int replication_co_readv(BlockDriverState *bs,
224 int64_t sector_num,
225 int remaining_sectors,
226 QEMUIOVector *qiov)
228 BDRVReplicationState *s = bs->opaque;
229 int ret;
231 if (s->mode == REPLICATION_MODE_PRIMARY) {
232 /* We only use it to forward primary write requests */
233 return -EIO;
236 ret = replication_get_io_status(s);
237 if (ret < 0) {
238 return ret;
241 ret = bdrv_co_preadv(bs->file, sector_num * BDRV_SECTOR_SIZE,
242 remaining_sectors * BDRV_SECTOR_SIZE, qiov, 0);
244 return replication_return_value(s, ret);
247 static coroutine_fn int replication_co_writev(BlockDriverState *bs,
248 int64_t sector_num,
249 int remaining_sectors,
250 QEMUIOVector *qiov,
251 int flags)
253 BDRVReplicationState *s = bs->opaque;
254 QEMUIOVector hd_qiov;
255 uint64_t bytes_done = 0;
256 BdrvChild *top = bs->file;
257 BdrvChild *base = s->secondary_disk;
258 BdrvChild *target;
259 int ret;
260 int64_t n;
262 ret = replication_get_io_status(s);
263 if (ret < 0) {
264 goto out;
267 if (ret == 0) {
268 ret = bdrv_co_pwritev(top, sector_num * BDRV_SECTOR_SIZE,
269 remaining_sectors * BDRV_SECTOR_SIZE, qiov, 0);
270 return replication_return_value(s, ret);
274 * Failover failed, only write to active disk if the sectors
275 * have already been allocated in active disk/hidden disk.
277 qemu_iovec_init(&hd_qiov, qiov->niov);
278 while (remaining_sectors > 0) {
279 int64_t count;
281 ret = bdrv_is_allocated_above(top->bs, base->bs, false,
282 sector_num * BDRV_SECTOR_SIZE,
283 remaining_sectors * BDRV_SECTOR_SIZE,
284 &count);
285 if (ret < 0) {
286 goto out1;
289 assert(QEMU_IS_ALIGNED(count, BDRV_SECTOR_SIZE));
290 n = count >> BDRV_SECTOR_BITS;
291 qemu_iovec_reset(&hd_qiov);
292 qemu_iovec_concat(&hd_qiov, qiov, bytes_done, count);
294 target = ret ? top : base;
295 ret = bdrv_co_pwritev(target, sector_num * BDRV_SECTOR_SIZE,
296 n * BDRV_SECTOR_SIZE, &hd_qiov, 0);
297 if (ret < 0) {
298 goto out1;
301 remaining_sectors -= n;
302 sector_num += n;
303 bytes_done += count;
306 out1:
307 qemu_iovec_destroy(&hd_qiov);
308 out:
309 return ret;
312 static void secondary_do_checkpoint(BlockDriverState *bs, Error **errp)
314 BDRVReplicationState *s = bs->opaque;
315 BdrvChild *active_disk = bs->file;
316 Error *local_err = NULL;
317 int ret;
319 if (!s->backup_job) {
320 error_setg(errp, "Backup job was cancelled unexpectedly");
321 return;
324 backup_do_checkpoint(s->backup_job, &local_err);
325 if (local_err) {
326 error_propagate(errp, local_err);
327 return;
330 if (!active_disk->bs->drv) {
331 error_setg(errp, "Active disk %s is ejected",
332 active_disk->bs->node_name);
333 return;
336 ret = bdrv_make_empty(active_disk, errp);
337 if (ret < 0) {
338 return;
341 if (!s->hidden_disk->bs->drv) {
342 error_setg(errp, "Hidden disk %s is ejected",
343 s->hidden_disk->bs->node_name);
344 return;
347 ret = bdrv_make_empty(s->hidden_disk, errp);
348 if (ret < 0) {
349 return;
353 /* This function is supposed to be called twice:
354 * first with writable = true, then with writable = false.
355 * The first call puts s->hidden_disk and s->secondary_disk in
356 * r/w mode, and the second puts them back in their original state.
358 static void reopen_backing_file(BlockDriverState *bs, bool writable,
359 Error **errp)
361 BDRVReplicationState *s = bs->opaque;
362 BdrvChild *hidden_disk, *secondary_disk;
363 BlockReopenQueue *reopen_queue = NULL;
366 * s->hidden_disk and s->secondary_disk may not be set yet, as they will
367 * only be set after the children are writable.
369 hidden_disk = bs->file->bs->backing;
370 secondary_disk = hidden_disk->bs->backing;
372 if (writable) {
373 s->orig_hidden_read_only = bdrv_is_read_only(hidden_disk->bs);
374 s->orig_secondary_read_only = bdrv_is_read_only(secondary_disk->bs);
377 if (s->orig_hidden_read_only) {
378 QDict *opts = qdict_new();
379 qdict_put_bool(opts, BDRV_OPT_READ_ONLY, !writable);
380 reopen_queue = bdrv_reopen_queue(reopen_queue, hidden_disk->bs,
381 opts, true);
384 if (s->orig_secondary_read_only) {
385 QDict *opts = qdict_new();
386 qdict_put_bool(opts, BDRV_OPT_READ_ONLY, !writable);
387 reopen_queue = bdrv_reopen_queue(reopen_queue, secondary_disk->bs,
388 opts, true);
391 if (reopen_queue) {
392 AioContext *ctx = bdrv_get_aio_context(bs);
393 if (ctx != qemu_get_aio_context()) {
394 aio_context_release(ctx);
396 bdrv_reopen_multiple(reopen_queue, errp);
397 if (ctx != qemu_get_aio_context()) {
398 aio_context_acquire(ctx);
403 static void backup_job_cleanup(BlockDriverState *bs)
405 BDRVReplicationState *s = bs->opaque;
406 BlockDriverState *top_bs;
408 s->backup_job = NULL;
410 top_bs = bdrv_lookup_bs(s->top_id, s->top_id, NULL);
411 if (!top_bs) {
412 return;
414 bdrv_op_unblock_all(top_bs, s->blocker);
415 error_free(s->blocker);
416 reopen_backing_file(bs, false, NULL);
419 static void backup_job_completed(void *opaque, int ret)
421 BlockDriverState *bs = opaque;
422 BDRVReplicationState *s = bs->opaque;
424 if (s->stage != BLOCK_REPLICATION_FAILOVER) {
425 /* The backup job is cancelled unexpectedly */
426 s->error = -EIO;
429 backup_job_cleanup(bs);
432 static bool check_top_bs(BlockDriverState *top_bs, BlockDriverState *bs)
434 BdrvChild *child;
436 /* The bs itself is the top_bs */
437 if (top_bs == bs) {
438 return true;
441 /* Iterate over top_bs's children */
442 QLIST_FOREACH(child, &top_bs->children, next) {
443 if (child->bs == bs || check_top_bs(child->bs, bs)) {
444 return true;
448 return false;
451 static void replication_start(ReplicationState *rs, ReplicationMode mode,
452 Error **errp)
454 BlockDriverState *bs = rs->opaque;
455 BDRVReplicationState *s;
456 BlockDriverState *top_bs;
457 BdrvChild *active_disk, *hidden_disk, *secondary_disk;
458 int64_t active_length, hidden_length, disk_length;
459 AioContext *aio_context;
460 Error *local_err = NULL;
461 BackupPerf perf = { .use_copy_range = true, .max_workers = 1 };
463 aio_context = bdrv_get_aio_context(bs);
464 aio_context_acquire(aio_context);
465 s = bs->opaque;
467 if (s->stage == BLOCK_REPLICATION_DONE ||
468 s->stage == BLOCK_REPLICATION_FAILOVER) {
470 * This case happens when a secondary is promoted to primary.
471 * Ignore the request because the secondary side of replication
472 * doesn't have to do anything anymore.
474 aio_context_release(aio_context);
475 return;
478 if (s->stage != BLOCK_REPLICATION_NONE) {
479 error_setg(errp, "Block replication is running or done");
480 aio_context_release(aio_context);
481 return;
484 if (s->mode != mode) {
485 error_setg(errp, "The parameter mode's value is invalid, needs %d,"
486 " but got %d", s->mode, mode);
487 aio_context_release(aio_context);
488 return;
491 switch (s->mode) {
492 case REPLICATION_MODE_PRIMARY:
493 break;
494 case REPLICATION_MODE_SECONDARY:
495 active_disk = bs->file;
496 if (!active_disk || !active_disk->bs || !active_disk->bs->backing) {
497 error_setg(errp, "Active disk doesn't have backing file");
498 aio_context_release(aio_context);
499 return;
502 hidden_disk = active_disk->bs->backing;
503 if (!hidden_disk->bs || !hidden_disk->bs->backing) {
504 error_setg(errp, "Hidden disk doesn't have backing file");
505 aio_context_release(aio_context);
506 return;
509 secondary_disk = hidden_disk->bs->backing;
510 if (!secondary_disk->bs || !bdrv_has_blk(secondary_disk->bs)) {
511 error_setg(errp, "The secondary disk doesn't have block backend");
512 aio_context_release(aio_context);
513 return;
516 /* verify the length */
517 active_length = bdrv_getlength(active_disk->bs);
518 hidden_length = bdrv_getlength(hidden_disk->bs);
519 disk_length = bdrv_getlength(secondary_disk->bs);
520 if (active_length < 0 || hidden_length < 0 || disk_length < 0 ||
521 active_length != hidden_length || hidden_length != disk_length) {
522 error_setg(errp, "Active disk, hidden disk, secondary disk's length"
523 " are not the same");
524 aio_context_release(aio_context);
525 return;
528 /* Must be true, or the bdrv_getlength() calls would have failed */
529 assert(active_disk->bs->drv && hidden_disk->bs->drv);
531 if (!active_disk->bs->drv->bdrv_make_empty ||
532 !hidden_disk->bs->drv->bdrv_make_empty) {
533 error_setg(errp,
534 "Active disk or hidden disk doesn't support make_empty");
535 aio_context_release(aio_context);
536 return;
539 /* reopen the backing file in r/w mode */
540 reopen_backing_file(bs, true, &local_err);
541 if (local_err) {
542 error_propagate(errp, local_err);
543 aio_context_release(aio_context);
544 return;
547 bdrv_ref(hidden_disk->bs);
548 s->hidden_disk = bdrv_attach_child(bs, hidden_disk->bs, "hidden disk",
549 &child_of_bds, BDRV_CHILD_DATA,
550 &local_err);
551 if (local_err) {
552 error_propagate(errp, local_err);
553 aio_context_release(aio_context);
554 return;
557 bdrv_ref(secondary_disk->bs);
558 s->secondary_disk = bdrv_attach_child(bs, secondary_disk->bs,
559 "secondary disk", &child_of_bds,
560 BDRV_CHILD_DATA, &local_err);
561 if (local_err) {
562 error_propagate(errp, local_err);
563 aio_context_release(aio_context);
564 return;
567 /* start backup job now */
568 error_setg(&s->blocker,
569 "Block device is in use by internal backup job");
571 top_bs = bdrv_lookup_bs(s->top_id, s->top_id, NULL);
572 if (!top_bs || !bdrv_is_root_node(top_bs) ||
573 !check_top_bs(top_bs, bs)) {
574 error_setg(errp, "No top_bs or it is invalid");
575 reopen_backing_file(bs, false, NULL);
576 aio_context_release(aio_context);
577 return;
579 bdrv_op_block_all(top_bs, s->blocker);
580 bdrv_op_unblock(top_bs, BLOCK_OP_TYPE_DATAPLANE, s->blocker);
582 s->backup_job = backup_job_create(
583 NULL, s->secondary_disk->bs, s->hidden_disk->bs,
584 0, MIRROR_SYNC_MODE_NONE, NULL, 0, false, NULL,
585 &perf,
586 BLOCKDEV_ON_ERROR_REPORT,
587 BLOCKDEV_ON_ERROR_REPORT, JOB_INTERNAL,
588 backup_job_completed, bs, NULL, &local_err);
589 if (local_err) {
590 error_propagate(errp, local_err);
591 backup_job_cleanup(bs);
592 aio_context_release(aio_context);
593 return;
595 job_start(&s->backup_job->job);
596 break;
597 default:
598 aio_context_release(aio_context);
599 abort();
602 s->stage = BLOCK_REPLICATION_RUNNING;
604 if (s->mode == REPLICATION_MODE_SECONDARY) {
605 secondary_do_checkpoint(bs, errp);
608 s->error = 0;
609 aio_context_release(aio_context);
612 static void replication_do_checkpoint(ReplicationState *rs, Error **errp)
614 BlockDriverState *bs = rs->opaque;
615 BDRVReplicationState *s;
616 AioContext *aio_context;
618 aio_context = bdrv_get_aio_context(bs);
619 aio_context_acquire(aio_context);
620 s = bs->opaque;
622 if (s->stage == BLOCK_REPLICATION_DONE ||
623 s->stage == BLOCK_REPLICATION_FAILOVER) {
625 * This case happens when a secondary was promoted to primary.
626 * Ignore the request because the secondary side of replication
627 * doesn't have to do anything anymore.
629 aio_context_release(aio_context);
630 return;
633 if (s->mode == REPLICATION_MODE_SECONDARY) {
634 secondary_do_checkpoint(bs, errp);
636 aio_context_release(aio_context);
639 static void replication_get_error(ReplicationState *rs, Error **errp)
641 BlockDriverState *bs = rs->opaque;
642 BDRVReplicationState *s;
643 AioContext *aio_context;
645 aio_context = bdrv_get_aio_context(bs);
646 aio_context_acquire(aio_context);
647 s = bs->opaque;
649 if (s->stage == BLOCK_REPLICATION_NONE) {
650 error_setg(errp, "Block replication is not running");
651 aio_context_release(aio_context);
652 return;
655 if (s->error) {
656 error_setg(errp, "I/O error occurred");
657 aio_context_release(aio_context);
658 return;
660 aio_context_release(aio_context);
663 static void replication_done(void *opaque, int ret)
665 BlockDriverState *bs = opaque;
666 BDRVReplicationState *s = bs->opaque;
668 if (ret == 0) {
669 s->stage = BLOCK_REPLICATION_DONE;
671 bdrv_unref_child(bs, s->secondary_disk);
672 s->secondary_disk = NULL;
673 bdrv_unref_child(bs, s->hidden_disk);
674 s->hidden_disk = NULL;
675 s->error = 0;
676 } else {
677 s->stage = BLOCK_REPLICATION_FAILOVER_FAILED;
678 s->error = -EIO;
682 static void replication_stop(ReplicationState *rs, bool failover, Error **errp)
684 BlockDriverState *bs = rs->opaque;
685 BDRVReplicationState *s;
686 AioContext *aio_context;
688 aio_context = bdrv_get_aio_context(bs);
689 aio_context_acquire(aio_context);
690 s = bs->opaque;
692 if (s->stage == BLOCK_REPLICATION_DONE ||
693 s->stage == BLOCK_REPLICATION_FAILOVER) {
695 * This case happens when a secondary was promoted to primary.
696 * Ignore the request because the secondary side of replication
697 * doesn't have to do anything anymore.
699 aio_context_release(aio_context);
700 return;
703 if (s->stage != BLOCK_REPLICATION_RUNNING) {
704 error_setg(errp, "Block replication is not running");
705 aio_context_release(aio_context);
706 return;
709 switch (s->mode) {
710 case REPLICATION_MODE_PRIMARY:
711 s->stage = BLOCK_REPLICATION_DONE;
712 s->error = 0;
713 break;
714 case REPLICATION_MODE_SECONDARY:
716 * This BDS will be closed, and the job should be completed
717 * before the BDS is closed, because we will access hidden
718 * disk, secondary disk in backup_job_completed().
720 if (s->backup_job) {
721 aio_context_release(aio_context);
722 job_cancel_sync(&s->backup_job->job, true);
723 aio_context_acquire(aio_context);
726 if (!failover) {
727 secondary_do_checkpoint(bs, errp);
728 s->stage = BLOCK_REPLICATION_DONE;
729 aio_context_release(aio_context);
730 return;
733 s->stage = BLOCK_REPLICATION_FAILOVER;
734 s->commit_job = commit_active_start(
735 NULL, bs->file->bs, s->secondary_disk->bs,
736 JOB_INTERNAL, 0, BLOCKDEV_ON_ERROR_REPORT,
737 NULL, replication_done, bs, true, errp);
738 break;
739 default:
740 aio_context_release(aio_context);
741 abort();
743 aio_context_release(aio_context);
746 static const char *const replication_strong_runtime_opts[] = {
747 REPLICATION_MODE,
748 REPLICATION_TOP_ID,
750 NULL
753 static BlockDriver bdrv_replication = {
754 .format_name = "replication",
755 .instance_size = sizeof(BDRVReplicationState),
757 .bdrv_open = replication_open,
758 .bdrv_close = replication_close,
759 .bdrv_child_perm = replication_child_perm,
761 .bdrv_co_getlength = replication_co_getlength,
762 .bdrv_co_readv = replication_co_readv,
763 .bdrv_co_writev = replication_co_writev,
765 .is_filter = true,
767 .has_variable_length = true,
768 .strong_runtime_opts = replication_strong_runtime_opts,
771 static void bdrv_replication_init(void)
773 bdrv_register(&bdrv_replication);
776 block_init(bdrv_replication_init);