isa: use get_uint() for "io-base"
[qemu/ar7.git] / block / quorum.c
blob55ba91665576f66bcf980c40b5b3ccf57fa68981
1 /*
2 * Quorum Block filter
4 * Copyright (C) 2012-2014 Nodalink, EURL.
6 * Author:
7 * BenoƮt Canet <benoit.canet@irqsave.net>
9 * Based on the design and code of blkverify.c (Copyright (C) 2010 IBM, Corp)
10 * and blkmirror.c (Copyright (C) 2011 Red Hat, Inc).
12 * This work is licensed under the terms of the GNU GPL, version 2 or later.
13 * See the COPYING file in the top-level directory.
16 #include "qemu/osdep.h"
17 #include "qemu/cutils.h"
18 #include "block/block_int.h"
19 #include "qapi/qmp/qbool.h"
20 #include "qapi/qmp/qdict.h"
21 #include "qapi/qmp/qerror.h"
22 #include "qapi/qmp/qjson.h"
23 #include "qapi/qmp/qlist.h"
24 #include "qapi/qmp/qstring.h"
25 #include "qapi-event.h"
26 #include "crypto/hash.h"
28 #define HASH_LENGTH 32
30 #define QUORUM_OPT_VOTE_THRESHOLD "vote-threshold"
31 #define QUORUM_OPT_BLKVERIFY "blkverify"
32 #define QUORUM_OPT_REWRITE "rewrite-corrupted"
33 #define QUORUM_OPT_READ_PATTERN "read-pattern"
35 /* This union holds a vote hash value */
36 typedef union QuorumVoteValue {
37 uint8_t h[HASH_LENGTH]; /* SHA-256 hash */
38 int64_t l; /* simpler 64 bits hash */
39 } QuorumVoteValue;
41 /* A vote item */
42 typedef struct QuorumVoteItem {
43 int index;
44 QLIST_ENTRY(QuorumVoteItem) next;
45 } QuorumVoteItem;
47 /* this structure is a vote version. A version is the set of votes sharing the
48 * same vote value.
49 * The set of votes will be tracked with the items field and its cardinality is
50 * vote_count.
52 typedef struct QuorumVoteVersion {
53 QuorumVoteValue value;
54 int index;
55 int vote_count;
56 QLIST_HEAD(, QuorumVoteItem) items;
57 QLIST_ENTRY(QuorumVoteVersion) next;
58 } QuorumVoteVersion;
60 /* this structure holds a group of vote versions together */
61 typedef struct QuorumVotes {
62 QLIST_HEAD(, QuorumVoteVersion) vote_list;
63 bool (*compare)(QuorumVoteValue *a, QuorumVoteValue *b);
64 } QuorumVotes;
66 /* the following structure holds the state of one quorum instance */
67 typedef struct BDRVQuorumState {
68 BdrvChild **children; /* children BlockDriverStates */
69 int num_children; /* children count */
70 unsigned next_child_index; /* the index of the next child that should
71 * be added
73 int threshold; /* if less than threshold children reads gave the
74 * same result a quorum error occurs.
76 bool is_blkverify; /* true if the driver is in blkverify mode
77 * Writes are mirrored on two children devices.
78 * On reads the two children devices' contents are
79 * compared and if a difference is spotted its
80 * location is printed and the code aborts.
81 * It is useful to debug other block drivers by
82 * comparing them with a reference one.
84 bool rewrite_corrupted;/* true if the driver must rewrite-on-read corrupted
85 * block if Quorum is reached.
88 QuorumReadPattern read_pattern;
89 } BDRVQuorumState;
91 typedef struct QuorumAIOCB QuorumAIOCB;
93 /* Quorum will create one instance of the following structure per operation it
94 * performs on its children.
95 * So for each read/write operation coming from the upper layer there will be
96 * $children_count QuorumChildRequest.
98 typedef struct QuorumChildRequest {
99 BlockDriverState *bs;
100 QEMUIOVector qiov;
101 uint8_t *buf;
102 int ret;
103 QuorumAIOCB *parent;
104 } QuorumChildRequest;
106 /* Quorum will use the following structure to track progress of each read/write
107 * operation received by the upper layer.
108 * This structure hold pointers to the QuorumChildRequest structures instances
109 * used to do operations on each children and track overall progress.
111 struct QuorumAIOCB {
112 BlockDriverState *bs;
113 Coroutine *co;
115 /* Request metadata */
116 uint64_t offset;
117 uint64_t bytes;
119 QEMUIOVector *qiov; /* calling IOV */
121 QuorumChildRequest *qcrs; /* individual child requests */
122 int count; /* number of completed AIOCB */
123 int success_count; /* number of successfully completed AIOCB */
125 int rewrite_count; /* number of replica to rewrite: count down to
126 * zero once writes are fired
129 QuorumVotes votes;
131 bool is_read;
132 int vote_ret;
133 int children_read; /* how many children have been read from */
136 typedef struct QuorumCo {
137 QuorumAIOCB *acb;
138 int idx;
139 } QuorumCo;
141 static void quorum_aio_finalize(QuorumAIOCB *acb)
143 g_free(acb->qcrs);
144 g_free(acb);
147 static bool quorum_sha256_compare(QuorumVoteValue *a, QuorumVoteValue *b)
149 return !memcmp(a->h, b->h, HASH_LENGTH);
152 static bool quorum_64bits_compare(QuorumVoteValue *a, QuorumVoteValue *b)
154 return a->l == b->l;
157 static QuorumAIOCB *quorum_aio_get(BlockDriverState *bs,
158 QEMUIOVector *qiov,
159 uint64_t offset,
160 uint64_t bytes)
162 BDRVQuorumState *s = bs->opaque;
163 QuorumAIOCB *acb = g_new(QuorumAIOCB, 1);
164 int i;
166 *acb = (QuorumAIOCB) {
167 .co = qemu_coroutine_self(),
168 .bs = bs,
169 .offset = offset,
170 .bytes = bytes,
171 .qiov = qiov,
172 .votes.compare = quorum_sha256_compare,
173 .votes.vote_list = QLIST_HEAD_INITIALIZER(acb.votes.vote_list),
176 acb->qcrs = g_new0(QuorumChildRequest, s->num_children);
177 for (i = 0; i < s->num_children; i++) {
178 acb->qcrs[i].buf = NULL;
179 acb->qcrs[i].ret = 0;
180 acb->qcrs[i].parent = acb;
183 return acb;
186 static void quorum_report_bad(QuorumOpType type, uint64_t offset,
187 uint64_t bytes, char *node_name, int ret)
189 const char *msg = NULL;
190 int64_t start_sector = offset / BDRV_SECTOR_SIZE;
191 int64_t end_sector = DIV_ROUND_UP(offset + bytes, BDRV_SECTOR_SIZE);
193 if (ret < 0) {
194 msg = strerror(-ret);
197 qapi_event_send_quorum_report_bad(type, !!msg, msg, node_name, start_sector,
198 end_sector - start_sector, &error_abort);
201 static void quorum_report_failure(QuorumAIOCB *acb)
203 const char *reference = bdrv_get_device_or_node_name(acb->bs);
204 int64_t start_sector = acb->offset / BDRV_SECTOR_SIZE;
205 int64_t end_sector = DIV_ROUND_UP(acb->offset + acb->bytes,
206 BDRV_SECTOR_SIZE);
208 qapi_event_send_quorum_failure(reference, start_sector,
209 end_sector - start_sector, &error_abort);
212 static int quorum_vote_error(QuorumAIOCB *acb);
214 static bool quorum_has_too_much_io_failed(QuorumAIOCB *acb)
216 BDRVQuorumState *s = acb->bs->opaque;
218 if (acb->success_count < s->threshold) {
219 acb->vote_ret = quorum_vote_error(acb);
220 quorum_report_failure(acb);
221 return true;
224 return false;
227 static int read_fifo_child(QuorumAIOCB *acb);
229 static void quorum_copy_qiov(QEMUIOVector *dest, QEMUIOVector *source)
231 int i;
232 assert(dest->niov == source->niov);
233 assert(dest->size == source->size);
234 for (i = 0; i < source->niov; i++) {
235 assert(dest->iov[i].iov_len == source->iov[i].iov_len);
236 memcpy(dest->iov[i].iov_base,
237 source->iov[i].iov_base,
238 source->iov[i].iov_len);
242 static void quorum_report_bad_acb(QuorumChildRequest *sacb, int ret)
244 QuorumAIOCB *acb = sacb->parent;
245 QuorumOpType type = acb->is_read ? QUORUM_OP_TYPE_READ : QUORUM_OP_TYPE_WRITE;
246 quorum_report_bad(type, acb->offset, acb->bytes, sacb->bs->node_name, ret);
249 static void quorum_report_bad_versions(BDRVQuorumState *s,
250 QuorumAIOCB *acb,
251 QuorumVoteValue *value)
253 QuorumVoteVersion *version;
254 QuorumVoteItem *item;
256 QLIST_FOREACH(version, &acb->votes.vote_list, next) {
257 if (acb->votes.compare(&version->value, value)) {
258 continue;
260 QLIST_FOREACH(item, &version->items, next) {
261 quorum_report_bad(QUORUM_OP_TYPE_READ, acb->offset, acb->bytes,
262 s->children[item->index]->bs->node_name, 0);
267 static void quorum_rewrite_entry(void *opaque)
269 QuorumCo *co = opaque;
270 QuorumAIOCB *acb = co->acb;
271 BDRVQuorumState *s = acb->bs->opaque;
273 /* Ignore any errors, it's just a correction attempt for already
274 * corrupted data. */
275 bdrv_co_pwritev(s->children[co->idx], acb->offset, acb->bytes,
276 acb->qiov, 0);
278 /* Wake up the caller after the last rewrite */
279 acb->rewrite_count--;
280 if (!acb->rewrite_count) {
281 qemu_coroutine_enter_if_inactive(acb->co);
285 static bool quorum_rewrite_bad_versions(QuorumAIOCB *acb,
286 QuorumVoteValue *value)
288 QuorumVoteVersion *version;
289 QuorumVoteItem *item;
290 int count = 0;
292 /* first count the number of bad versions: done first to avoid concurrency
293 * issues.
295 QLIST_FOREACH(version, &acb->votes.vote_list, next) {
296 if (acb->votes.compare(&version->value, value)) {
297 continue;
299 QLIST_FOREACH(item, &version->items, next) {
300 count++;
304 /* quorum_rewrite_entry will count down this to zero */
305 acb->rewrite_count = count;
307 /* now fire the correcting rewrites */
308 QLIST_FOREACH(version, &acb->votes.vote_list, next) {
309 if (acb->votes.compare(&version->value, value)) {
310 continue;
312 QLIST_FOREACH(item, &version->items, next) {
313 Coroutine *co;
314 QuorumCo data = {
315 .acb = acb,
316 .idx = item->index,
319 co = qemu_coroutine_create(quorum_rewrite_entry, &data);
320 qemu_coroutine_enter(co);
324 /* return true if any rewrite is done else false */
325 return count;
328 static void quorum_count_vote(QuorumVotes *votes,
329 QuorumVoteValue *value,
330 int index)
332 QuorumVoteVersion *v = NULL, *version = NULL;
333 QuorumVoteItem *item;
335 /* look if we have something with this hash */
336 QLIST_FOREACH(v, &votes->vote_list, next) {
337 if (votes->compare(&v->value, value)) {
338 version = v;
339 break;
343 /* It's a version not yet in the list add it */
344 if (!version) {
345 version = g_new0(QuorumVoteVersion, 1);
346 QLIST_INIT(&version->items);
347 memcpy(&version->value, value, sizeof(version->value));
348 version->index = index;
349 version->vote_count = 0;
350 QLIST_INSERT_HEAD(&votes->vote_list, version, next);
353 version->vote_count++;
355 item = g_new0(QuorumVoteItem, 1);
356 item->index = index;
357 QLIST_INSERT_HEAD(&version->items, item, next);
360 static void quorum_free_vote_list(QuorumVotes *votes)
362 QuorumVoteVersion *version, *next_version;
363 QuorumVoteItem *item, *next_item;
365 QLIST_FOREACH_SAFE(version, &votes->vote_list, next, next_version) {
366 QLIST_REMOVE(version, next);
367 QLIST_FOREACH_SAFE(item, &version->items, next, next_item) {
368 QLIST_REMOVE(item, next);
369 g_free(item);
371 g_free(version);
375 static int quorum_compute_hash(QuorumAIOCB *acb, int i, QuorumVoteValue *hash)
377 QEMUIOVector *qiov = &acb->qcrs[i].qiov;
378 size_t len = sizeof(hash->h);
379 uint8_t *data = hash->h;
381 /* XXX - would be nice if we could pass in the Error **
382 * and propagate that back, but this quorum code is
383 * restricted to just errno values currently */
384 if (qcrypto_hash_bytesv(QCRYPTO_HASH_ALG_SHA256,
385 qiov->iov, qiov->niov,
386 &data, &len,
387 NULL) < 0) {
388 return -EINVAL;
391 return 0;
394 static QuorumVoteVersion *quorum_get_vote_winner(QuorumVotes *votes)
396 int max = 0;
397 QuorumVoteVersion *candidate, *winner = NULL;
399 QLIST_FOREACH(candidate, &votes->vote_list, next) {
400 if (candidate->vote_count > max) {
401 max = candidate->vote_count;
402 winner = candidate;
406 return winner;
409 /* qemu_iovec_compare is handy for blkverify mode because it returns the first
410 * differing byte location. Yet it is handcoded to compare vectors one byte
411 * after another so it does not benefit from the libc SIMD optimizations.
412 * quorum_iovec_compare is written for speed and should be used in the non
413 * blkverify mode of quorum.
415 static bool quorum_iovec_compare(QEMUIOVector *a, QEMUIOVector *b)
417 int i;
418 int result;
420 assert(a->niov == b->niov);
421 for (i = 0; i < a->niov; i++) {
422 assert(a->iov[i].iov_len == b->iov[i].iov_len);
423 result = memcmp(a->iov[i].iov_base,
424 b->iov[i].iov_base,
425 a->iov[i].iov_len);
426 if (result) {
427 return false;
431 return true;
434 static void GCC_FMT_ATTR(2, 3) quorum_err(QuorumAIOCB *acb,
435 const char *fmt, ...)
437 va_list ap;
439 va_start(ap, fmt);
440 fprintf(stderr, "quorum: offset=%" PRIu64 " bytes=%" PRIu64 " ",
441 acb->offset, acb->bytes);
442 vfprintf(stderr, fmt, ap);
443 fprintf(stderr, "\n");
444 va_end(ap);
445 exit(1);
448 static bool quorum_compare(QuorumAIOCB *acb,
449 QEMUIOVector *a,
450 QEMUIOVector *b)
452 BDRVQuorumState *s = acb->bs->opaque;
453 ssize_t offset;
455 /* This driver will replace blkverify in this particular case */
456 if (s->is_blkverify) {
457 offset = qemu_iovec_compare(a, b);
458 if (offset != -1) {
459 quorum_err(acb, "contents mismatch at offset %" PRIu64,
460 acb->offset + offset);
462 return true;
465 return quorum_iovec_compare(a, b);
468 /* Do a vote to get the error code */
469 static int quorum_vote_error(QuorumAIOCB *acb)
471 BDRVQuorumState *s = acb->bs->opaque;
472 QuorumVoteVersion *winner = NULL;
473 QuorumVotes error_votes;
474 QuorumVoteValue result_value;
475 int i, ret = 0;
476 bool error = false;
478 QLIST_INIT(&error_votes.vote_list);
479 error_votes.compare = quorum_64bits_compare;
481 for (i = 0; i < s->num_children; i++) {
482 ret = acb->qcrs[i].ret;
483 if (ret) {
484 error = true;
485 result_value.l = ret;
486 quorum_count_vote(&error_votes, &result_value, i);
490 if (error) {
491 winner = quorum_get_vote_winner(&error_votes);
492 ret = winner->value.l;
495 quorum_free_vote_list(&error_votes);
497 return ret;
500 static void quorum_vote(QuorumAIOCB *acb)
502 bool quorum = true;
503 int i, j, ret;
504 QuorumVoteValue hash;
505 BDRVQuorumState *s = acb->bs->opaque;
506 QuorumVoteVersion *winner;
508 if (quorum_has_too_much_io_failed(acb)) {
509 return;
512 /* get the index of the first successful read */
513 for (i = 0; i < s->num_children; i++) {
514 if (!acb->qcrs[i].ret) {
515 break;
519 assert(i < s->num_children);
521 /* compare this read with all other successful reads stopping at quorum
522 * failure
524 for (j = i + 1; j < s->num_children; j++) {
525 if (acb->qcrs[j].ret) {
526 continue;
528 quorum = quorum_compare(acb, &acb->qcrs[i].qiov, &acb->qcrs[j].qiov);
529 if (!quorum) {
530 break;
534 /* Every successful read agrees */
535 if (quorum) {
536 quorum_copy_qiov(acb->qiov, &acb->qcrs[i].qiov);
537 return;
540 /* compute hashes for each successful read, also store indexes */
541 for (i = 0; i < s->num_children; i++) {
542 if (acb->qcrs[i].ret) {
543 continue;
545 ret = quorum_compute_hash(acb, i, &hash);
546 /* if ever the hash computation failed */
547 if (ret < 0) {
548 acb->vote_ret = ret;
549 goto free_exit;
551 quorum_count_vote(&acb->votes, &hash, i);
554 /* vote to select the most represented version */
555 winner = quorum_get_vote_winner(&acb->votes);
557 /* if the winner count is smaller than threshold the read fails */
558 if (winner->vote_count < s->threshold) {
559 quorum_report_failure(acb);
560 acb->vote_ret = -EIO;
561 goto free_exit;
564 /* we have a winner: copy it */
565 quorum_copy_qiov(acb->qiov, &acb->qcrs[winner->index].qiov);
567 /* some versions are bad print them */
568 quorum_report_bad_versions(s, acb, &winner->value);
570 /* corruption correction is enabled */
571 if (s->rewrite_corrupted) {
572 quorum_rewrite_bad_versions(acb, &winner->value);
575 free_exit:
576 /* free lists */
577 quorum_free_vote_list(&acb->votes);
580 static void read_quorum_children_entry(void *opaque)
582 QuorumCo *co = opaque;
583 QuorumAIOCB *acb = co->acb;
584 BDRVQuorumState *s = acb->bs->opaque;
585 int i = co->idx;
586 QuorumChildRequest *sacb = &acb->qcrs[i];
588 sacb->bs = s->children[i]->bs;
589 sacb->ret = bdrv_co_preadv(s->children[i], acb->offset, acb->bytes,
590 &acb->qcrs[i].qiov, 0);
592 if (sacb->ret == 0) {
593 acb->success_count++;
594 } else {
595 quorum_report_bad_acb(sacb, sacb->ret);
598 acb->count++;
599 assert(acb->count <= s->num_children);
600 assert(acb->success_count <= s->num_children);
602 /* Wake up the caller after the last read */
603 if (acb->count == s->num_children) {
604 qemu_coroutine_enter_if_inactive(acb->co);
608 static int read_quorum_children(QuorumAIOCB *acb)
610 BDRVQuorumState *s = acb->bs->opaque;
611 int i, ret;
613 acb->children_read = s->num_children;
614 for (i = 0; i < s->num_children; i++) {
615 acb->qcrs[i].buf = qemu_blockalign(s->children[i]->bs, acb->qiov->size);
616 qemu_iovec_init(&acb->qcrs[i].qiov, acb->qiov->niov);
617 qemu_iovec_clone(&acb->qcrs[i].qiov, acb->qiov, acb->qcrs[i].buf);
620 for (i = 0; i < s->num_children; i++) {
621 Coroutine *co;
622 QuorumCo data = {
623 .acb = acb,
624 .idx = i,
627 co = qemu_coroutine_create(read_quorum_children_entry, &data);
628 qemu_coroutine_enter(co);
631 while (acb->count < s->num_children) {
632 qemu_coroutine_yield();
635 /* Do the vote on read */
636 quorum_vote(acb);
637 for (i = 0; i < s->num_children; i++) {
638 qemu_vfree(acb->qcrs[i].buf);
639 qemu_iovec_destroy(&acb->qcrs[i].qiov);
642 while (acb->rewrite_count) {
643 qemu_coroutine_yield();
646 ret = acb->vote_ret;
648 return ret;
651 static int read_fifo_child(QuorumAIOCB *acb)
653 BDRVQuorumState *s = acb->bs->opaque;
654 int n, ret;
656 /* We try to read the next child in FIFO order if we failed to read */
657 do {
658 n = acb->children_read++;
659 acb->qcrs[n].bs = s->children[n]->bs;
660 ret = bdrv_co_preadv(s->children[n], acb->offset, acb->bytes,
661 acb->qiov, 0);
662 if (ret < 0) {
663 quorum_report_bad_acb(&acb->qcrs[n], ret);
665 } while (ret < 0 && acb->children_read < s->num_children);
667 /* FIXME: rewrite failed children if acb->children_read > 1? */
669 return ret;
672 static int quorum_co_preadv(BlockDriverState *bs, uint64_t offset,
673 uint64_t bytes, QEMUIOVector *qiov, int flags)
675 BDRVQuorumState *s = bs->opaque;
676 QuorumAIOCB *acb = quorum_aio_get(bs, qiov, offset, bytes);
677 int ret;
679 acb->is_read = true;
680 acb->children_read = 0;
682 if (s->read_pattern == QUORUM_READ_PATTERN_QUORUM) {
683 ret = read_quorum_children(acb);
684 } else {
685 ret = read_fifo_child(acb);
687 quorum_aio_finalize(acb);
689 return ret;
692 static void write_quorum_entry(void *opaque)
694 QuorumCo *co = opaque;
695 QuorumAIOCB *acb = co->acb;
696 BDRVQuorumState *s = acb->bs->opaque;
697 int i = co->idx;
698 QuorumChildRequest *sacb = &acb->qcrs[i];
700 sacb->bs = s->children[i]->bs;
701 sacb->ret = bdrv_co_pwritev(s->children[i], acb->offset, acb->bytes,
702 acb->qiov, 0);
703 if (sacb->ret == 0) {
704 acb->success_count++;
705 } else {
706 quorum_report_bad_acb(sacb, sacb->ret);
708 acb->count++;
709 assert(acb->count <= s->num_children);
710 assert(acb->success_count <= s->num_children);
712 /* Wake up the caller after the last write */
713 if (acb->count == s->num_children) {
714 qemu_coroutine_enter_if_inactive(acb->co);
718 static int quorum_co_pwritev(BlockDriverState *bs, uint64_t offset,
719 uint64_t bytes, QEMUIOVector *qiov, int flags)
721 BDRVQuorumState *s = bs->opaque;
722 QuorumAIOCB *acb = quorum_aio_get(bs, qiov, offset, bytes);
723 int i, ret;
725 for (i = 0; i < s->num_children; i++) {
726 Coroutine *co;
727 QuorumCo data = {
728 .acb = acb,
729 .idx = i,
732 co = qemu_coroutine_create(write_quorum_entry, &data);
733 qemu_coroutine_enter(co);
736 while (acb->count < s->num_children) {
737 qemu_coroutine_yield();
740 quorum_has_too_much_io_failed(acb);
742 ret = acb->vote_ret;
743 quorum_aio_finalize(acb);
745 return ret;
748 static int64_t quorum_getlength(BlockDriverState *bs)
750 BDRVQuorumState *s = bs->opaque;
751 int64_t result;
752 int i;
754 /* check that all file have the same length */
755 result = bdrv_getlength(s->children[0]->bs);
756 if (result < 0) {
757 return result;
759 for (i = 1; i < s->num_children; i++) {
760 int64_t value = bdrv_getlength(s->children[i]->bs);
761 if (value < 0) {
762 return value;
764 if (value != result) {
765 return -EIO;
769 return result;
772 static coroutine_fn int quorum_co_flush(BlockDriverState *bs)
774 BDRVQuorumState *s = bs->opaque;
775 QuorumVoteVersion *winner = NULL;
776 QuorumVotes error_votes;
777 QuorumVoteValue result_value;
778 int i;
779 int result = 0;
780 int success_count = 0;
782 QLIST_INIT(&error_votes.vote_list);
783 error_votes.compare = quorum_64bits_compare;
785 for (i = 0; i < s->num_children; i++) {
786 result = bdrv_co_flush(s->children[i]->bs);
787 if (result) {
788 quorum_report_bad(QUORUM_OP_TYPE_FLUSH, 0,
789 bdrv_getlength(s->children[i]->bs),
790 s->children[i]->bs->node_name, result);
791 result_value.l = result;
792 quorum_count_vote(&error_votes, &result_value, i);
793 } else {
794 success_count++;
798 if (success_count >= s->threshold) {
799 result = 0;
800 } else {
801 winner = quorum_get_vote_winner(&error_votes);
802 result = winner->value.l;
804 quorum_free_vote_list(&error_votes);
806 return result;
809 static bool quorum_recurse_is_first_non_filter(BlockDriverState *bs,
810 BlockDriverState *candidate)
812 BDRVQuorumState *s = bs->opaque;
813 int i;
815 for (i = 0; i < s->num_children; i++) {
816 bool perm = bdrv_recurse_is_first_non_filter(s->children[i]->bs,
817 candidate);
818 if (perm) {
819 return true;
823 return false;
826 static int quorum_valid_threshold(int threshold, int num_children, Error **errp)
829 if (threshold < 1) {
830 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
831 "vote-threshold", "value >= 1");
832 return -ERANGE;
835 if (threshold > num_children) {
836 error_setg(errp, "threshold may not exceed children count");
837 return -ERANGE;
840 return 0;
843 static QemuOptsList quorum_runtime_opts = {
844 .name = "quorum",
845 .head = QTAILQ_HEAD_INITIALIZER(quorum_runtime_opts.head),
846 .desc = {
848 .name = QUORUM_OPT_VOTE_THRESHOLD,
849 .type = QEMU_OPT_NUMBER,
850 .help = "The number of vote needed for reaching quorum",
853 .name = QUORUM_OPT_BLKVERIFY,
854 .type = QEMU_OPT_BOOL,
855 .help = "Trigger block verify mode if set",
858 .name = QUORUM_OPT_REWRITE,
859 .type = QEMU_OPT_BOOL,
860 .help = "Rewrite corrupted block on read quorum",
863 .name = QUORUM_OPT_READ_PATTERN,
864 .type = QEMU_OPT_STRING,
865 .help = "Allowed pattern: quorum, fifo. Quorum is default",
867 { /* end of list */ }
871 static int parse_read_pattern(const char *opt)
873 int i;
875 if (!opt) {
876 /* Set quorum as default */
877 return QUORUM_READ_PATTERN_QUORUM;
880 for (i = 0; i < QUORUM_READ_PATTERN__MAX; i++) {
881 if (!strcmp(opt, QuorumReadPattern_lookup[i])) {
882 return i;
886 return -EINVAL;
889 static int quorum_open(BlockDriverState *bs, QDict *options, int flags,
890 Error **errp)
892 BDRVQuorumState *s = bs->opaque;
893 Error *local_err = NULL;
894 QemuOpts *opts = NULL;
895 bool *opened;
896 int i;
897 int ret = 0;
899 qdict_flatten(options);
901 /* count how many different children are present */
902 s->num_children = qdict_array_entries(options, "children.");
903 if (s->num_children < 0) {
904 error_setg(&local_err, "Option children is not a valid array");
905 ret = -EINVAL;
906 goto exit;
908 if (s->num_children < 1) {
909 error_setg(&local_err,
910 "Number of provided children must be 1 or more");
911 ret = -EINVAL;
912 goto exit;
915 opts = qemu_opts_create(&quorum_runtime_opts, NULL, 0, &error_abort);
916 qemu_opts_absorb_qdict(opts, options, &local_err);
917 if (local_err) {
918 ret = -EINVAL;
919 goto exit;
922 s->threshold = qemu_opt_get_number(opts, QUORUM_OPT_VOTE_THRESHOLD, 0);
923 /* and validate it against s->num_children */
924 ret = quorum_valid_threshold(s->threshold, s->num_children, &local_err);
925 if (ret < 0) {
926 goto exit;
929 ret = parse_read_pattern(qemu_opt_get(opts, QUORUM_OPT_READ_PATTERN));
930 if (ret < 0) {
931 error_setg(&local_err, "Please set read-pattern as fifo or quorum");
932 goto exit;
934 s->read_pattern = ret;
936 if (s->read_pattern == QUORUM_READ_PATTERN_QUORUM) {
937 /* is the driver in blkverify mode */
938 if (qemu_opt_get_bool(opts, QUORUM_OPT_BLKVERIFY, false) &&
939 s->num_children == 2 && s->threshold == 2) {
940 s->is_blkverify = true;
941 } else if (qemu_opt_get_bool(opts, QUORUM_OPT_BLKVERIFY, false)) {
942 fprintf(stderr, "blkverify mode is set by setting blkverify=on "
943 "and using two files with vote_threshold=2\n");
946 s->rewrite_corrupted = qemu_opt_get_bool(opts, QUORUM_OPT_REWRITE,
947 false);
948 if (s->rewrite_corrupted && s->is_blkverify) {
949 error_setg(&local_err,
950 "rewrite-corrupted=on cannot be used with blkverify=on");
951 ret = -EINVAL;
952 goto exit;
956 /* allocate the children array */
957 s->children = g_new0(BdrvChild *, s->num_children);
958 opened = g_new0(bool, s->num_children);
960 for (i = 0; i < s->num_children; i++) {
961 char indexstr[32];
962 ret = snprintf(indexstr, 32, "children.%d", i);
963 assert(ret < 32);
965 s->children[i] = bdrv_open_child(NULL, options, indexstr, bs,
966 &child_format, false, &local_err);
967 if (local_err) {
968 ret = -EINVAL;
969 goto close_exit;
972 opened[i] = true;
974 s->next_child_index = s->num_children;
976 g_free(opened);
977 goto exit;
979 close_exit:
980 /* cleanup on error */
981 for (i = 0; i < s->num_children; i++) {
982 if (!opened[i]) {
983 continue;
985 bdrv_unref_child(bs, s->children[i]);
987 g_free(s->children);
988 g_free(opened);
989 exit:
990 qemu_opts_del(opts);
991 /* propagate error */
992 error_propagate(errp, local_err);
993 return ret;
996 static void quorum_close(BlockDriverState *bs)
998 BDRVQuorumState *s = bs->opaque;
999 int i;
1001 for (i = 0; i < s->num_children; i++) {
1002 bdrv_unref_child(bs, s->children[i]);
1005 g_free(s->children);
1008 static void quorum_add_child(BlockDriverState *bs, BlockDriverState *child_bs,
1009 Error **errp)
1011 BDRVQuorumState *s = bs->opaque;
1012 BdrvChild *child;
1013 char indexstr[32];
1014 int ret;
1016 assert(s->num_children <= INT_MAX / sizeof(BdrvChild *));
1017 if (s->num_children == INT_MAX / sizeof(BdrvChild *) ||
1018 s->next_child_index == UINT_MAX) {
1019 error_setg(errp, "Too many children");
1020 return;
1023 ret = snprintf(indexstr, 32, "children.%u", s->next_child_index);
1024 if (ret < 0 || ret >= 32) {
1025 error_setg(errp, "cannot generate child name");
1026 return;
1028 s->next_child_index++;
1030 bdrv_drained_begin(bs);
1032 /* We can safely add the child now */
1033 bdrv_ref(child_bs);
1035 child = bdrv_attach_child(bs, child_bs, indexstr, &child_format, errp);
1036 if (child == NULL) {
1037 s->next_child_index--;
1038 bdrv_unref(child_bs);
1039 goto out;
1041 s->children = g_renew(BdrvChild *, s->children, s->num_children + 1);
1042 s->children[s->num_children++] = child;
1044 out:
1045 bdrv_drained_end(bs);
1048 static void quorum_del_child(BlockDriverState *bs, BdrvChild *child,
1049 Error **errp)
1051 BDRVQuorumState *s = bs->opaque;
1052 int i;
1054 for (i = 0; i < s->num_children; i++) {
1055 if (s->children[i] == child) {
1056 break;
1060 /* we have checked it in bdrv_del_child() */
1061 assert(i < s->num_children);
1063 if (s->num_children <= s->threshold) {
1064 error_setg(errp,
1065 "The number of children cannot be lower than the vote threshold %d",
1066 s->threshold);
1067 return;
1070 bdrv_drained_begin(bs);
1072 /* We can safely remove this child now */
1073 memmove(&s->children[i], &s->children[i + 1],
1074 (s->num_children - i - 1) * sizeof(BdrvChild *));
1075 s->children = g_renew(BdrvChild *, s->children, --s->num_children);
1076 bdrv_unref_child(bs, child);
1078 bdrv_drained_end(bs);
1081 static void quorum_refresh_filename(BlockDriverState *bs, QDict *options)
1083 BDRVQuorumState *s = bs->opaque;
1084 QDict *opts;
1085 QList *children;
1086 int i;
1088 for (i = 0; i < s->num_children; i++) {
1089 bdrv_refresh_filename(s->children[i]->bs);
1090 if (!s->children[i]->bs->full_open_options) {
1091 return;
1095 children = qlist_new();
1096 for (i = 0; i < s->num_children; i++) {
1097 QINCREF(s->children[i]->bs->full_open_options);
1098 qlist_append(children, s->children[i]->bs->full_open_options);
1101 opts = qdict_new();
1102 qdict_put_str(opts, "driver", "quorum");
1103 qdict_put_int(opts, QUORUM_OPT_VOTE_THRESHOLD, s->threshold);
1104 qdict_put_bool(opts, QUORUM_OPT_BLKVERIFY, s->is_blkverify);
1105 qdict_put_bool(opts, QUORUM_OPT_REWRITE, s->rewrite_corrupted);
1106 qdict_put(opts, "children", children);
1108 bs->full_open_options = opts;
1111 static BlockDriver bdrv_quorum = {
1112 .format_name = "quorum",
1113 .protocol_name = "quorum",
1115 .instance_size = sizeof(BDRVQuorumState),
1117 .bdrv_file_open = quorum_open,
1118 .bdrv_close = quorum_close,
1119 .bdrv_refresh_filename = quorum_refresh_filename,
1121 .bdrv_co_flush_to_disk = quorum_co_flush,
1123 .bdrv_getlength = quorum_getlength,
1125 .bdrv_co_preadv = quorum_co_preadv,
1126 .bdrv_co_pwritev = quorum_co_pwritev,
1128 .bdrv_add_child = quorum_add_child,
1129 .bdrv_del_child = quorum_del_child,
1131 .bdrv_child_perm = bdrv_filter_default_perms,
1133 .is_filter = true,
1134 .bdrv_recurse_is_first_non_filter = quorum_recurse_is_first_non_filter,
1137 static void bdrv_quorum_init(void)
1139 if (!qcrypto_hash_supports(QCRYPTO_HASH_ALG_SHA256)) {
1140 /* SHA256 hash support is required for quorum device */
1141 return;
1143 bdrv_register(&bdrv_quorum);
1146 block_init(bdrv_quorum_init);