tests/acceptance/boot_linux: Accept SSH pubkey
[qemu.git] / block / replication.c
blob0c70215784da9cd4516919f25928f55a06679c41
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 "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 BdrvChild *active_disk;
39 BlockJob *commit_job;
40 BdrvChild *hidden_disk;
41 BdrvChild *secondary_disk;
42 BlockJob *backup_job;
43 char *top_id;
44 ReplicationState *rs;
45 Error *blocker;
46 bool orig_hidden_read_only;
47 bool orig_secondary_read_only;
48 int error;
49 } BDRVReplicationState;
51 static void replication_start(ReplicationState *rs, ReplicationMode mode,
52 Error **errp);
53 static void replication_do_checkpoint(ReplicationState *rs, Error **errp);
54 static void replication_get_error(ReplicationState *rs, Error **errp);
55 static void replication_stop(ReplicationState *rs, bool failover,
56 Error **errp);
58 #define REPLICATION_MODE "mode"
59 #define REPLICATION_TOP_ID "top-id"
60 static QemuOptsList replication_runtime_opts = {
61 .name = "replication",
62 .head = QTAILQ_HEAD_INITIALIZER(replication_runtime_opts.head),
63 .desc = {
65 .name = REPLICATION_MODE,
66 .type = QEMU_OPT_STRING,
69 .name = REPLICATION_TOP_ID,
70 .type = QEMU_OPT_STRING,
72 { /* end of list */ }
76 static ReplicationOps replication_ops = {
77 .start = replication_start,
78 .checkpoint = replication_do_checkpoint,
79 .get_error = replication_get_error,
80 .stop = replication_stop,
83 static int replication_open(BlockDriverState *bs, QDict *options,
84 int flags, Error **errp)
86 int ret;
87 BDRVReplicationState *s = bs->opaque;
88 QemuOpts *opts = NULL;
89 const char *mode;
90 const char *top_id;
92 bs->file = bdrv_open_child(NULL, options, "file", bs, &child_of_bds,
93 BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY,
94 false, errp);
95 if (!bs->file) {
96 return -EINVAL;
99 ret = -EINVAL;
100 opts = qemu_opts_create(&replication_runtime_opts, NULL, 0, &error_abort);
101 if (!qemu_opts_absorb_qdict(opts, options, errp)) {
102 goto fail;
105 mode = qemu_opt_get(opts, REPLICATION_MODE);
106 if (!mode) {
107 error_setg(errp, "Missing the option mode");
108 goto fail;
111 if (!strcmp(mode, "primary")) {
112 s->mode = REPLICATION_MODE_PRIMARY;
113 top_id = qemu_opt_get(opts, REPLICATION_TOP_ID);
114 if (top_id) {
115 error_setg(errp,
116 "The primary side does not support option top-id");
117 goto fail;
119 } else if (!strcmp(mode, "secondary")) {
120 s->mode = REPLICATION_MODE_SECONDARY;
121 top_id = qemu_opt_get(opts, REPLICATION_TOP_ID);
122 s->top_id = g_strdup(top_id);
123 if (!s->top_id) {
124 error_setg(errp, "Missing the option top-id");
125 goto fail;
127 } else {
128 error_setg(errp,
129 "The option mode's value should be primary or secondary");
130 goto fail;
133 s->rs = replication_new(bs, &replication_ops);
135 ret = 0;
137 fail:
138 qemu_opts_del(opts);
139 return ret;
142 static void replication_close(BlockDriverState *bs)
144 BDRVReplicationState *s = bs->opaque;
145 Job *commit_job;
147 if (s->stage == BLOCK_REPLICATION_RUNNING) {
148 replication_stop(s->rs, false, NULL);
150 if (s->stage == BLOCK_REPLICATION_FAILOVER) {
151 commit_job = &s->commit_job->job;
152 assert(commit_job->aio_context == qemu_get_current_aio_context());
153 job_cancel_sync(commit_job);
156 if (s->mode == REPLICATION_MODE_SECONDARY) {
157 g_free(s->top_id);
160 replication_remove(s->rs);
163 static void replication_child_perm(BlockDriverState *bs, BdrvChild *c,
164 BdrvChildRole role,
165 BlockReopenQueue *reopen_queue,
166 uint64_t perm, uint64_t shared,
167 uint64_t *nperm, uint64_t *nshared)
169 *nperm = BLK_PERM_CONSISTENT_READ;
170 if ((bs->open_flags & (BDRV_O_INACTIVE | BDRV_O_RDWR)) == BDRV_O_RDWR) {
171 *nperm |= BLK_PERM_WRITE;
173 *nshared = BLK_PERM_CONSISTENT_READ
174 | BLK_PERM_WRITE
175 | BLK_PERM_WRITE_UNCHANGED;
176 return;
179 static int64_t replication_getlength(BlockDriverState *bs)
181 return bdrv_getlength(bs->file->bs);
184 static int replication_get_io_status(BDRVReplicationState *s)
186 switch (s->stage) {
187 case BLOCK_REPLICATION_NONE:
188 return -EIO;
189 case BLOCK_REPLICATION_RUNNING:
190 return 0;
191 case BLOCK_REPLICATION_FAILOVER:
192 return s->mode == REPLICATION_MODE_PRIMARY ? -EIO : 0;
193 case BLOCK_REPLICATION_FAILOVER_FAILED:
194 return s->mode == REPLICATION_MODE_PRIMARY ? -EIO : 1;
195 case BLOCK_REPLICATION_DONE:
197 * active commit job completes, and active disk and secondary_disk
198 * is swapped, so we can operate bs->file directly
200 return s->mode == REPLICATION_MODE_PRIMARY ? -EIO : 0;
201 default:
202 abort();
206 static int replication_return_value(BDRVReplicationState *s, int ret)
208 if (s->mode == REPLICATION_MODE_SECONDARY) {
209 return ret;
212 if (ret < 0) {
213 s->error = ret;
214 ret = 0;
217 return ret;
220 static coroutine_fn int replication_co_readv(BlockDriverState *bs,
221 int64_t sector_num,
222 int remaining_sectors,
223 QEMUIOVector *qiov)
225 BDRVReplicationState *s = bs->opaque;
226 int ret;
228 if (s->mode == REPLICATION_MODE_PRIMARY) {
229 /* We only use it to forward primary write requests */
230 return -EIO;
233 ret = replication_get_io_status(s);
234 if (ret < 0) {
235 return ret;
238 ret = bdrv_co_preadv(bs->file, sector_num * BDRV_SECTOR_SIZE,
239 remaining_sectors * BDRV_SECTOR_SIZE, qiov, 0);
241 return replication_return_value(s, ret);
244 static coroutine_fn int replication_co_writev(BlockDriverState *bs,
245 int64_t sector_num,
246 int remaining_sectors,
247 QEMUIOVector *qiov,
248 int flags)
250 BDRVReplicationState *s = bs->opaque;
251 QEMUIOVector hd_qiov;
252 uint64_t bytes_done = 0;
253 BdrvChild *top = bs->file;
254 BdrvChild *base = s->secondary_disk;
255 BdrvChild *target;
256 int ret;
257 int64_t n;
259 assert(!flags);
260 ret = replication_get_io_status(s);
261 if (ret < 0) {
262 goto out;
265 if (ret == 0) {
266 ret = bdrv_co_pwritev(top, sector_num * BDRV_SECTOR_SIZE,
267 remaining_sectors * BDRV_SECTOR_SIZE, qiov, 0);
268 return replication_return_value(s, ret);
272 * Failover failed, only write to active disk if the sectors
273 * have already been allocated in active disk/hidden disk.
275 qemu_iovec_init(&hd_qiov, qiov->niov);
276 while (remaining_sectors > 0) {
277 int64_t count;
279 ret = bdrv_is_allocated_above(top->bs, base->bs, false,
280 sector_num * BDRV_SECTOR_SIZE,
281 remaining_sectors * BDRV_SECTOR_SIZE,
282 &count);
283 if (ret < 0) {
284 goto out1;
287 assert(QEMU_IS_ALIGNED(count, BDRV_SECTOR_SIZE));
288 n = count >> BDRV_SECTOR_BITS;
289 qemu_iovec_reset(&hd_qiov);
290 qemu_iovec_concat(&hd_qiov, qiov, bytes_done, count);
292 target = ret ? top : base;
293 ret = bdrv_co_pwritev(target, sector_num * BDRV_SECTOR_SIZE,
294 n * BDRV_SECTOR_SIZE, &hd_qiov, 0);
295 if (ret < 0) {
296 goto out1;
299 remaining_sectors -= n;
300 sector_num += n;
301 bytes_done += count;
304 out1:
305 qemu_iovec_destroy(&hd_qiov);
306 out:
307 return ret;
310 static void secondary_do_checkpoint(BDRVReplicationState *s, Error **errp)
312 Error *local_err = NULL;
313 int ret;
315 if (!s->backup_job) {
316 error_setg(errp, "Backup job was cancelled unexpectedly");
317 return;
320 backup_do_checkpoint(s->backup_job, &local_err);
321 if (local_err) {
322 error_propagate(errp, local_err);
323 return;
326 if (!s->active_disk->bs->drv) {
327 error_setg(errp, "Active disk %s is ejected",
328 s->active_disk->bs->node_name);
329 return;
332 ret = bdrv_make_empty(s->active_disk, errp);
333 if (ret < 0) {
334 return;
337 if (!s->hidden_disk->bs->drv) {
338 error_setg(errp, "Hidden disk %s is ejected",
339 s->hidden_disk->bs->node_name);
340 return;
343 BlockBackend *blk = blk_new(qemu_get_current_aio_context(),
344 BLK_PERM_WRITE, BLK_PERM_ALL);
345 blk_insert_bs(blk, s->hidden_disk->bs, &local_err);
346 if (local_err) {
347 error_propagate(errp, local_err);
348 blk_unref(blk);
349 return;
352 ret = blk_make_empty(blk, errp);
353 blk_unref(blk);
354 if (ret < 0) {
355 return;
359 /* This function is supposed to be called twice:
360 * first with writable = true, then with writable = false.
361 * The first call puts s->hidden_disk and s->secondary_disk in
362 * r/w mode, and the second puts them back in their original state.
364 static void reopen_backing_file(BlockDriverState *bs, bool writable,
365 Error **errp)
367 BDRVReplicationState *s = bs->opaque;
368 BlockReopenQueue *reopen_queue = NULL;
370 if (writable) {
371 s->orig_hidden_read_only = bdrv_is_read_only(s->hidden_disk->bs);
372 s->orig_secondary_read_only = bdrv_is_read_only(s->secondary_disk->bs);
375 bdrv_subtree_drained_begin(s->hidden_disk->bs);
376 bdrv_subtree_drained_begin(s->secondary_disk->bs);
378 if (s->orig_hidden_read_only) {
379 QDict *opts = qdict_new();
380 qdict_put_bool(opts, BDRV_OPT_READ_ONLY, !writable);
381 reopen_queue = bdrv_reopen_queue(reopen_queue, s->hidden_disk->bs,
382 opts, true);
385 if (s->orig_secondary_read_only) {
386 QDict *opts = qdict_new();
387 qdict_put_bool(opts, BDRV_OPT_READ_ONLY, !writable);
388 reopen_queue = bdrv_reopen_queue(reopen_queue, s->secondary_disk->bs,
389 opts, true);
392 if (reopen_queue) {
393 bdrv_reopen_multiple(reopen_queue, errp);
396 bdrv_subtree_drained_end(s->hidden_disk->bs);
397 bdrv_subtree_drained_end(s->secondary_disk->bs);
400 static void backup_job_cleanup(BlockDriverState *bs)
402 BDRVReplicationState *s = bs->opaque;
403 BlockDriverState *top_bs;
405 s->backup_job = NULL;
407 top_bs = bdrv_lookup_bs(s->top_id, s->top_id, NULL);
408 if (!top_bs) {
409 return;
411 bdrv_op_unblock_all(top_bs, s->blocker);
412 error_free(s->blocker);
413 reopen_backing_file(bs, false, NULL);
416 static void backup_job_completed(void *opaque, int ret)
418 BlockDriverState *bs = opaque;
419 BDRVReplicationState *s = bs->opaque;
421 if (s->stage != BLOCK_REPLICATION_FAILOVER) {
422 /* The backup job is cancelled unexpectedly */
423 s->error = -EIO;
426 backup_job_cleanup(bs);
429 static bool check_top_bs(BlockDriverState *top_bs, BlockDriverState *bs)
431 BdrvChild *child;
433 /* The bs itself is the top_bs */
434 if (top_bs == bs) {
435 return true;
438 /* Iterate over top_bs's children */
439 QLIST_FOREACH(child, &top_bs->children, next) {
440 if (child->bs == bs || check_top_bs(child->bs, bs)) {
441 return true;
445 return false;
448 static void replication_start(ReplicationState *rs, ReplicationMode mode,
449 Error **errp)
451 BlockDriverState *bs = rs->opaque;
452 BDRVReplicationState *s;
453 BlockDriverState *top_bs;
454 int64_t active_length, hidden_length, disk_length;
455 AioContext *aio_context;
456 Error *local_err = NULL;
458 aio_context = bdrv_get_aio_context(bs);
459 aio_context_acquire(aio_context);
460 s = bs->opaque;
462 if (s->stage == BLOCK_REPLICATION_DONE ||
463 s->stage == BLOCK_REPLICATION_FAILOVER) {
465 * This case happens when a secondary is promoted to primary.
466 * Ignore the request because the secondary side of replication
467 * doesn't have to do anything anymore.
469 aio_context_release(aio_context);
470 return;
473 if (s->stage != BLOCK_REPLICATION_NONE) {
474 error_setg(errp, "Block replication is running or done");
475 aio_context_release(aio_context);
476 return;
479 if (s->mode != mode) {
480 error_setg(errp, "The parameter mode's value is invalid, needs %d,"
481 " but got %d", s->mode, mode);
482 aio_context_release(aio_context);
483 return;
486 switch (s->mode) {
487 case REPLICATION_MODE_PRIMARY:
488 break;
489 case REPLICATION_MODE_SECONDARY:
490 s->active_disk = bs->file;
491 if (!s->active_disk || !s->active_disk->bs ||
492 !s->active_disk->bs->backing) {
493 error_setg(errp, "Active disk doesn't have backing file");
494 aio_context_release(aio_context);
495 return;
498 s->hidden_disk = s->active_disk->bs->backing;
499 if (!s->hidden_disk->bs || !s->hidden_disk->bs->backing) {
500 error_setg(errp, "Hidden disk doesn't have backing file");
501 aio_context_release(aio_context);
502 return;
505 s->secondary_disk = s->hidden_disk->bs->backing;
506 if (!s->secondary_disk->bs || !bdrv_has_blk(s->secondary_disk->bs)) {
507 error_setg(errp, "The secondary disk doesn't have block backend");
508 aio_context_release(aio_context);
509 return;
512 /* verify the length */
513 active_length = bdrv_getlength(s->active_disk->bs);
514 hidden_length = bdrv_getlength(s->hidden_disk->bs);
515 disk_length = bdrv_getlength(s->secondary_disk->bs);
516 if (active_length < 0 || hidden_length < 0 || disk_length < 0 ||
517 active_length != hidden_length || hidden_length != disk_length) {
518 error_setg(errp, "Active disk, hidden disk, secondary disk's length"
519 " are not the same");
520 aio_context_release(aio_context);
521 return;
524 /* Must be true, or the bdrv_getlength() calls would have failed */
525 assert(s->active_disk->bs->drv && s->hidden_disk->bs->drv);
527 if (!s->active_disk->bs->drv->bdrv_make_empty ||
528 !s->hidden_disk->bs->drv->bdrv_make_empty) {
529 error_setg(errp,
530 "Active disk or hidden disk doesn't support make_empty");
531 aio_context_release(aio_context);
532 return;
535 /* reopen the backing file in r/w mode */
536 reopen_backing_file(bs, true, &local_err);
537 if (local_err) {
538 error_propagate(errp, local_err);
539 aio_context_release(aio_context);
540 return;
543 /* start backup job now */
544 error_setg(&s->blocker,
545 "Block device is in use by internal backup job");
547 top_bs = bdrv_lookup_bs(s->top_id, s->top_id, NULL);
548 if (!top_bs || !bdrv_is_root_node(top_bs) ||
549 !check_top_bs(top_bs, bs)) {
550 error_setg(errp, "No top_bs or it is invalid");
551 reopen_backing_file(bs, false, NULL);
552 aio_context_release(aio_context);
553 return;
555 bdrv_op_block_all(top_bs, s->blocker);
556 bdrv_op_unblock(top_bs, BLOCK_OP_TYPE_DATAPLANE, s->blocker);
558 s->backup_job = backup_job_create(
559 NULL, s->secondary_disk->bs, s->hidden_disk->bs,
560 0, MIRROR_SYNC_MODE_NONE, NULL, 0, false, NULL,
561 BLOCKDEV_ON_ERROR_REPORT,
562 BLOCKDEV_ON_ERROR_REPORT, JOB_INTERNAL,
563 backup_job_completed, bs, NULL, &local_err);
564 if (local_err) {
565 error_propagate(errp, local_err);
566 backup_job_cleanup(bs);
567 aio_context_release(aio_context);
568 return;
570 job_start(&s->backup_job->job);
571 break;
572 default:
573 aio_context_release(aio_context);
574 abort();
577 s->stage = BLOCK_REPLICATION_RUNNING;
579 if (s->mode == REPLICATION_MODE_SECONDARY) {
580 secondary_do_checkpoint(s, errp);
583 s->error = 0;
584 aio_context_release(aio_context);
587 static void replication_do_checkpoint(ReplicationState *rs, Error **errp)
589 BlockDriverState *bs = rs->opaque;
590 BDRVReplicationState *s;
591 AioContext *aio_context;
593 aio_context = bdrv_get_aio_context(bs);
594 aio_context_acquire(aio_context);
595 s = bs->opaque;
597 if (s->stage == BLOCK_REPLICATION_DONE ||
598 s->stage == BLOCK_REPLICATION_FAILOVER) {
600 * This case happens when a secondary was promoted to primary.
601 * Ignore the request because the secondary side of replication
602 * doesn't have to do anything anymore.
604 aio_context_release(aio_context);
605 return;
608 if (s->mode == REPLICATION_MODE_SECONDARY) {
609 secondary_do_checkpoint(s, errp);
611 aio_context_release(aio_context);
614 static void replication_get_error(ReplicationState *rs, Error **errp)
616 BlockDriverState *bs = rs->opaque;
617 BDRVReplicationState *s;
618 AioContext *aio_context;
620 aio_context = bdrv_get_aio_context(bs);
621 aio_context_acquire(aio_context);
622 s = bs->opaque;
624 if (s->stage == BLOCK_REPLICATION_NONE) {
625 error_setg(errp, "Block replication is not running");
626 aio_context_release(aio_context);
627 return;
630 if (s->error) {
631 error_setg(errp, "I/O error occurred");
632 aio_context_release(aio_context);
633 return;
635 aio_context_release(aio_context);
638 static void replication_done(void *opaque, int ret)
640 BlockDriverState *bs = opaque;
641 BDRVReplicationState *s = bs->opaque;
643 if (ret == 0) {
644 s->stage = BLOCK_REPLICATION_DONE;
646 s->active_disk = NULL;
647 s->secondary_disk = NULL;
648 s->hidden_disk = NULL;
649 s->error = 0;
650 } else {
651 s->stage = BLOCK_REPLICATION_FAILOVER_FAILED;
652 s->error = -EIO;
656 static void replication_stop(ReplicationState *rs, bool failover, Error **errp)
658 BlockDriverState *bs = rs->opaque;
659 BDRVReplicationState *s;
660 AioContext *aio_context;
662 aio_context = bdrv_get_aio_context(bs);
663 aio_context_acquire(aio_context);
664 s = bs->opaque;
666 if (s->stage == BLOCK_REPLICATION_DONE ||
667 s->stage == BLOCK_REPLICATION_FAILOVER) {
669 * This case happens when a secondary was promoted to primary.
670 * Ignore the request because the secondary side of replication
671 * doesn't have to do anything anymore.
673 aio_context_release(aio_context);
674 return;
677 if (s->stage != BLOCK_REPLICATION_RUNNING) {
678 error_setg(errp, "Block replication is not running");
679 aio_context_release(aio_context);
680 return;
683 switch (s->mode) {
684 case REPLICATION_MODE_PRIMARY:
685 s->stage = BLOCK_REPLICATION_DONE;
686 s->error = 0;
687 break;
688 case REPLICATION_MODE_SECONDARY:
690 * This BDS will be closed, and the job should be completed
691 * before the BDS is closed, because we will access hidden
692 * disk, secondary disk in backup_job_completed().
694 if (s->backup_job) {
695 job_cancel_sync(&s->backup_job->job);
698 if (!failover) {
699 secondary_do_checkpoint(s, errp);
700 s->stage = BLOCK_REPLICATION_DONE;
701 aio_context_release(aio_context);
702 return;
705 s->stage = BLOCK_REPLICATION_FAILOVER;
706 s->commit_job = commit_active_start(
707 NULL, s->active_disk->bs, s->secondary_disk->bs,
708 JOB_INTERNAL, 0, BLOCKDEV_ON_ERROR_REPORT,
709 NULL, replication_done, bs, true, errp);
710 break;
711 default:
712 aio_context_release(aio_context);
713 abort();
715 aio_context_release(aio_context);
718 static const char *const replication_strong_runtime_opts[] = {
719 REPLICATION_MODE,
720 REPLICATION_TOP_ID,
722 NULL
725 static BlockDriver bdrv_replication = {
726 .format_name = "replication",
727 .instance_size = sizeof(BDRVReplicationState),
729 .bdrv_open = replication_open,
730 .bdrv_close = replication_close,
731 .bdrv_child_perm = replication_child_perm,
733 .bdrv_getlength = replication_getlength,
734 .bdrv_co_readv = replication_co_readv,
735 .bdrv_co_writev = replication_co_writev,
737 .is_filter = true,
739 .has_variable_length = true,
740 .strong_runtime_opts = replication_strong_runtime_opts,
743 static void bdrv_replication_init(void)
745 bdrv_register(&bdrv_replication);
748 block_init(bdrv_replication_init);