xref: /openbmc/qemu/block/quorum.c (revision 10358b6a)
1 /*
2  * Quorum Block filter
3  *
4  * Copyright (C) 2012-2014 Nodalink, EURL.
5  *
6  * Author:
7  *   Benoît Canet <benoit.canet@irqsave.net>
8  *
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).
11  *
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.
14  */
15 
16 #include <gnutls/gnutls.h>
17 #include <gnutls/crypto.h>
18 #include "block/block_int.h"
19 #include "qapi/qmp/qjson.h"
20 #include "qapi-event.h"
21 
22 #define HASH_LENGTH 32
23 
24 #define QUORUM_OPT_VOTE_THRESHOLD "vote-threshold"
25 #define QUORUM_OPT_BLKVERIFY      "blkverify"
26 
27 /* This union holds a vote hash value */
28 typedef union QuorumVoteValue {
29     char h[HASH_LENGTH];       /* SHA-256 hash */
30     int64_t l;                 /* simpler 64 bits hash */
31 } QuorumVoteValue;
32 
33 /* A vote item */
34 typedef struct QuorumVoteItem {
35     int index;
36     QLIST_ENTRY(QuorumVoteItem) next;
37 } QuorumVoteItem;
38 
39 /* this structure is a vote version. A version is the set of votes sharing the
40  * same vote value.
41  * The set of votes will be tracked with the items field and its cardinality is
42  * vote_count.
43  */
44 typedef struct QuorumVoteVersion {
45     QuorumVoteValue value;
46     int index;
47     int vote_count;
48     QLIST_HEAD(, QuorumVoteItem) items;
49     QLIST_ENTRY(QuorumVoteVersion) next;
50 } QuorumVoteVersion;
51 
52 /* this structure holds a group of vote versions together */
53 typedef struct QuorumVotes {
54     QLIST_HEAD(, QuorumVoteVersion) vote_list;
55     bool (*compare)(QuorumVoteValue *a, QuorumVoteValue *b);
56 } QuorumVotes;
57 
58 /* the following structure holds the state of one quorum instance */
59 typedef struct BDRVQuorumState {
60     BlockDriverState **bs; /* children BlockDriverStates */
61     int num_children;      /* children count */
62     int threshold;         /* if less than threshold children reads gave the
63                             * same result a quorum error occurs.
64                             */
65     bool is_blkverify;     /* true if the driver is in blkverify mode
66                             * Writes are mirrored on two children devices.
67                             * On reads the two children devices' contents are
68                             * compared and if a difference is spotted its
69                             * location is printed and the code aborts.
70                             * It is useful to debug other block drivers by
71                             * comparing them with a reference one.
72                             */
73 } BDRVQuorumState;
74 
75 typedef struct QuorumAIOCB QuorumAIOCB;
76 
77 /* Quorum will create one instance of the following structure per operation it
78  * performs on its children.
79  * So for each read/write operation coming from the upper layer there will be
80  * $children_count QuorumChildRequest.
81  */
82 typedef struct QuorumChildRequest {
83     BlockDriverAIOCB *aiocb;
84     QEMUIOVector qiov;
85     uint8_t *buf;
86     int ret;
87     QuorumAIOCB *parent;
88 } QuorumChildRequest;
89 
90 /* Quorum will use the following structure to track progress of each read/write
91  * operation received by the upper layer.
92  * This structure hold pointers to the QuorumChildRequest structures instances
93  * used to do operations on each children and track overall progress.
94  */
95 struct QuorumAIOCB {
96     BlockDriverAIOCB common;
97 
98     /* Request metadata */
99     uint64_t sector_num;
100     int nb_sectors;
101 
102     QEMUIOVector *qiov;         /* calling IOV */
103 
104     QuorumChildRequest *qcrs;   /* individual child requests */
105     int count;                  /* number of completed AIOCB */
106     int success_count;          /* number of successfully completed AIOCB */
107 
108     QuorumVotes votes;
109 
110     bool is_read;
111     int vote_ret;
112 };
113 
114 static void quorum_vote(QuorumAIOCB *acb);
115 
116 static void quorum_aio_cancel(BlockDriverAIOCB *blockacb)
117 {
118     QuorumAIOCB *acb = container_of(blockacb, QuorumAIOCB, common);
119     BDRVQuorumState *s = acb->common.bs->opaque;
120     int i;
121 
122     /* cancel all callbacks */
123     for (i = 0; i < s->num_children; i++) {
124         bdrv_aio_cancel(acb->qcrs[i].aiocb);
125     }
126 
127     g_free(acb->qcrs);
128     qemu_aio_release(acb);
129 }
130 
131 static AIOCBInfo quorum_aiocb_info = {
132     .aiocb_size         = sizeof(QuorumAIOCB),
133     .cancel             = quorum_aio_cancel,
134 };
135 
136 static void quorum_aio_finalize(QuorumAIOCB *acb)
137 {
138     BDRVQuorumState *s = acb->common.bs->opaque;
139     int i, ret = 0;
140 
141     if (acb->vote_ret) {
142         ret = acb->vote_ret;
143     }
144 
145     acb->common.cb(acb->common.opaque, ret);
146 
147     if (acb->is_read) {
148         for (i = 0; i < s->num_children; i++) {
149             qemu_vfree(acb->qcrs[i].buf);
150             qemu_iovec_destroy(&acb->qcrs[i].qiov);
151         }
152     }
153 
154     g_free(acb->qcrs);
155     qemu_aio_release(acb);
156 }
157 
158 static bool quorum_sha256_compare(QuorumVoteValue *a, QuorumVoteValue *b)
159 {
160     return !memcmp(a->h, b->h, HASH_LENGTH);
161 }
162 
163 static bool quorum_64bits_compare(QuorumVoteValue *a, QuorumVoteValue *b)
164 {
165     return a->l == b->l;
166 }
167 
168 static QuorumAIOCB *quorum_aio_get(BDRVQuorumState *s,
169                                    BlockDriverState *bs,
170                                    QEMUIOVector *qiov,
171                                    uint64_t sector_num,
172                                    int nb_sectors,
173                                    BlockDriverCompletionFunc *cb,
174                                    void *opaque)
175 {
176     QuorumAIOCB *acb = qemu_aio_get(&quorum_aiocb_info, bs, cb, opaque);
177     int i;
178 
179     acb->common.bs->opaque = s;
180     acb->sector_num = sector_num;
181     acb->nb_sectors = nb_sectors;
182     acb->qiov = qiov;
183     acb->qcrs = g_new0(QuorumChildRequest, s->num_children);
184     acb->count = 0;
185     acb->success_count = 0;
186     acb->votes.compare = quorum_sha256_compare;
187     QLIST_INIT(&acb->votes.vote_list);
188     acb->is_read = false;
189     acb->vote_ret = 0;
190 
191     for (i = 0; i < s->num_children; i++) {
192         acb->qcrs[i].buf = NULL;
193         acb->qcrs[i].ret = 0;
194         acb->qcrs[i].parent = acb;
195     }
196 
197     return acb;
198 }
199 
200 static void quorum_report_bad(QuorumAIOCB *acb, char *node_name, int ret)
201 {
202     const char *msg = NULL;
203     if (ret < 0) {
204         msg = strerror(-ret);
205     }
206     qapi_event_send_quorum_report_bad(!!msg, msg, node_name,
207                                       acb->sector_num, acb->nb_sectors, &error_abort);
208 }
209 
210 static void quorum_report_failure(QuorumAIOCB *acb)
211 {
212     const char *reference = acb->common.bs->device_name[0] ?
213                             acb->common.bs->device_name :
214                             acb->common.bs->node_name;
215 
216     qapi_event_send_quorum_failure(reference, acb->sector_num,
217                                    acb->nb_sectors, &error_abort);
218 }
219 
220 static int quorum_vote_error(QuorumAIOCB *acb);
221 
222 static bool quorum_has_too_much_io_failed(QuorumAIOCB *acb)
223 {
224     BDRVQuorumState *s = acb->common.bs->opaque;
225 
226     if (acb->success_count < s->threshold) {
227         acb->vote_ret = quorum_vote_error(acb);
228         quorum_report_failure(acb);
229         return true;
230     }
231 
232     return false;
233 }
234 
235 static void quorum_aio_cb(void *opaque, int ret)
236 {
237     QuorumChildRequest *sacb = opaque;
238     QuorumAIOCB *acb = sacb->parent;
239     BDRVQuorumState *s = acb->common.bs->opaque;
240 
241     sacb->ret = ret;
242     acb->count++;
243     if (ret == 0) {
244         acb->success_count++;
245     } else {
246         quorum_report_bad(acb, sacb->aiocb->bs->node_name, ret);
247     }
248     assert(acb->count <= s->num_children);
249     assert(acb->success_count <= s->num_children);
250     if (acb->count < s->num_children) {
251         return;
252     }
253 
254     /* Do the vote on read */
255     if (acb->is_read) {
256         quorum_vote(acb);
257     } else {
258         quorum_has_too_much_io_failed(acb);
259     }
260 
261     quorum_aio_finalize(acb);
262 }
263 
264 static void quorum_report_bad_versions(BDRVQuorumState *s,
265                                        QuorumAIOCB *acb,
266                                        QuorumVoteValue *value)
267 {
268     QuorumVoteVersion *version;
269     QuorumVoteItem *item;
270 
271     QLIST_FOREACH(version, &acb->votes.vote_list, next) {
272         if (acb->votes.compare(&version->value, value)) {
273             continue;
274         }
275         QLIST_FOREACH(item, &version->items, next) {
276             quorum_report_bad(acb, s->bs[item->index]->node_name, 0);
277         }
278     }
279 }
280 
281 static void quorum_copy_qiov(QEMUIOVector *dest, QEMUIOVector *source)
282 {
283     int i;
284     assert(dest->niov == source->niov);
285     assert(dest->size == source->size);
286     for (i = 0; i < source->niov; i++) {
287         assert(dest->iov[i].iov_len == source->iov[i].iov_len);
288         memcpy(dest->iov[i].iov_base,
289                source->iov[i].iov_base,
290                source->iov[i].iov_len);
291     }
292 }
293 
294 static void quorum_count_vote(QuorumVotes *votes,
295                               QuorumVoteValue *value,
296                               int index)
297 {
298     QuorumVoteVersion *v = NULL, *version = NULL;
299     QuorumVoteItem *item;
300 
301     /* look if we have something with this hash */
302     QLIST_FOREACH(v, &votes->vote_list, next) {
303         if (votes->compare(&v->value, value)) {
304             version = v;
305             break;
306         }
307     }
308 
309     /* It's a version not yet in the list add it */
310     if (!version) {
311         version = g_new0(QuorumVoteVersion, 1);
312         QLIST_INIT(&version->items);
313         memcpy(&version->value, value, sizeof(version->value));
314         version->index = index;
315         version->vote_count = 0;
316         QLIST_INSERT_HEAD(&votes->vote_list, version, next);
317     }
318 
319     version->vote_count++;
320 
321     item = g_new0(QuorumVoteItem, 1);
322     item->index = index;
323     QLIST_INSERT_HEAD(&version->items, item, next);
324 }
325 
326 static void quorum_free_vote_list(QuorumVotes *votes)
327 {
328     QuorumVoteVersion *version, *next_version;
329     QuorumVoteItem *item, *next_item;
330 
331     QLIST_FOREACH_SAFE(version, &votes->vote_list, next, next_version) {
332         QLIST_REMOVE(version, next);
333         QLIST_FOREACH_SAFE(item, &version->items, next, next_item) {
334             QLIST_REMOVE(item, next);
335             g_free(item);
336         }
337         g_free(version);
338     }
339 }
340 
341 static int quorum_compute_hash(QuorumAIOCB *acb, int i, QuorumVoteValue *hash)
342 {
343     int j, ret;
344     gnutls_hash_hd_t dig;
345     QEMUIOVector *qiov = &acb->qcrs[i].qiov;
346 
347     ret = gnutls_hash_init(&dig, GNUTLS_DIG_SHA256);
348 
349     if (ret < 0) {
350         return ret;
351     }
352 
353     for (j = 0; j < qiov->niov; j++) {
354         ret = gnutls_hash(dig, qiov->iov[j].iov_base, qiov->iov[j].iov_len);
355         if (ret < 0) {
356             break;
357         }
358     }
359 
360     gnutls_hash_deinit(dig, (void *) hash);
361     return ret;
362 }
363 
364 static QuorumVoteVersion *quorum_get_vote_winner(QuorumVotes *votes)
365 {
366     int max = 0;
367     QuorumVoteVersion *candidate, *winner = NULL;
368 
369     QLIST_FOREACH(candidate, &votes->vote_list, next) {
370         if (candidate->vote_count > max) {
371             max = candidate->vote_count;
372             winner = candidate;
373         }
374     }
375 
376     return winner;
377 }
378 
379 /* qemu_iovec_compare is handy for blkverify mode because it returns the first
380  * differing byte location. Yet it is handcoded to compare vectors one byte
381  * after another so it does not benefit from the libc SIMD optimizations.
382  * quorum_iovec_compare is written for speed and should be used in the non
383  * blkverify mode of quorum.
384  */
385 static bool quorum_iovec_compare(QEMUIOVector *a, QEMUIOVector *b)
386 {
387     int i;
388     int result;
389 
390     assert(a->niov == b->niov);
391     for (i = 0; i < a->niov; i++) {
392         assert(a->iov[i].iov_len == b->iov[i].iov_len);
393         result = memcmp(a->iov[i].iov_base,
394                         b->iov[i].iov_base,
395                         a->iov[i].iov_len);
396         if (result) {
397             return false;
398         }
399     }
400 
401     return true;
402 }
403 
404 static void GCC_FMT_ATTR(2, 3) quorum_err(QuorumAIOCB *acb,
405                                           const char *fmt, ...)
406 {
407     va_list ap;
408 
409     va_start(ap, fmt);
410     fprintf(stderr, "quorum: sector_num=%" PRId64 " nb_sectors=%d ",
411             acb->sector_num, acb->nb_sectors);
412     vfprintf(stderr, fmt, ap);
413     fprintf(stderr, "\n");
414     va_end(ap);
415     exit(1);
416 }
417 
418 static bool quorum_compare(QuorumAIOCB *acb,
419                            QEMUIOVector *a,
420                            QEMUIOVector *b)
421 {
422     BDRVQuorumState *s = acb->common.bs->opaque;
423     ssize_t offset;
424 
425     /* This driver will replace blkverify in this particular case */
426     if (s->is_blkverify) {
427         offset = qemu_iovec_compare(a, b);
428         if (offset != -1) {
429             quorum_err(acb, "contents mismatch in sector %" PRId64,
430                        acb->sector_num +
431                        (uint64_t)(offset / BDRV_SECTOR_SIZE));
432         }
433         return true;
434     }
435 
436     return quorum_iovec_compare(a, b);
437 }
438 
439 /* Do a vote to get the error code */
440 static int quorum_vote_error(QuorumAIOCB *acb)
441 {
442     BDRVQuorumState *s = acb->common.bs->opaque;
443     QuorumVoteVersion *winner = NULL;
444     QuorumVotes error_votes;
445     QuorumVoteValue result_value;
446     int i, ret = 0;
447     bool error = false;
448 
449     QLIST_INIT(&error_votes.vote_list);
450     error_votes.compare = quorum_64bits_compare;
451 
452     for (i = 0; i < s->num_children; i++) {
453         ret = acb->qcrs[i].ret;
454         if (ret) {
455             error = true;
456             result_value.l = ret;
457             quorum_count_vote(&error_votes, &result_value, i);
458         }
459     }
460 
461     if (error) {
462         winner = quorum_get_vote_winner(&error_votes);
463         ret = winner->value.l;
464     }
465 
466     quorum_free_vote_list(&error_votes);
467 
468     return ret;
469 }
470 
471 static void quorum_vote(QuorumAIOCB *acb)
472 {
473     bool quorum = true;
474     int i, j, ret;
475     QuorumVoteValue hash;
476     BDRVQuorumState *s = acb->common.bs->opaque;
477     QuorumVoteVersion *winner;
478 
479     if (quorum_has_too_much_io_failed(acb)) {
480         return;
481     }
482 
483     /* get the index of the first successful read */
484     for (i = 0; i < s->num_children; i++) {
485         if (!acb->qcrs[i].ret) {
486             break;
487         }
488     }
489 
490     assert(i < s->num_children);
491 
492     /* compare this read with all other successful reads stopping at quorum
493      * failure
494      */
495     for (j = i + 1; j < s->num_children; j++) {
496         if (acb->qcrs[j].ret) {
497             continue;
498         }
499         quorum = quorum_compare(acb, &acb->qcrs[i].qiov, &acb->qcrs[j].qiov);
500         if (!quorum) {
501             break;
502        }
503     }
504 
505     /* Every successful read agrees */
506     if (quorum) {
507         quorum_copy_qiov(acb->qiov, &acb->qcrs[i].qiov);
508         return;
509     }
510 
511     /* compute hashes for each successful read, also store indexes */
512     for (i = 0; i < s->num_children; i++) {
513         if (acb->qcrs[i].ret) {
514             continue;
515         }
516         ret = quorum_compute_hash(acb, i, &hash);
517         /* if ever the hash computation failed */
518         if (ret < 0) {
519             acb->vote_ret = ret;
520             goto free_exit;
521         }
522         quorum_count_vote(&acb->votes, &hash, i);
523     }
524 
525     /* vote to select the most represented version */
526     winner = quorum_get_vote_winner(&acb->votes);
527 
528     /* if the winner count is smaller than threshold the read fails */
529     if (winner->vote_count < s->threshold) {
530         quorum_report_failure(acb);
531         acb->vote_ret = -EIO;
532         goto free_exit;
533     }
534 
535     /* we have a winner: copy it */
536     quorum_copy_qiov(acb->qiov, &acb->qcrs[winner->index].qiov);
537 
538     /* some versions are bad print them */
539     quorum_report_bad_versions(s, acb, &winner->value);
540 
541 free_exit:
542     /* free lists */
543     quorum_free_vote_list(&acb->votes);
544 }
545 
546 static BlockDriverAIOCB *quorum_aio_readv(BlockDriverState *bs,
547                                          int64_t sector_num,
548                                          QEMUIOVector *qiov,
549                                          int nb_sectors,
550                                          BlockDriverCompletionFunc *cb,
551                                          void *opaque)
552 {
553     BDRVQuorumState *s = bs->opaque;
554     QuorumAIOCB *acb = quorum_aio_get(s, bs, qiov, sector_num,
555                                       nb_sectors, cb, opaque);
556     int i;
557 
558     acb->is_read = true;
559 
560     for (i = 0; i < s->num_children; i++) {
561         acb->qcrs[i].buf = qemu_blockalign(s->bs[i], qiov->size);
562         qemu_iovec_init(&acb->qcrs[i].qiov, qiov->niov);
563         qemu_iovec_clone(&acb->qcrs[i].qiov, qiov, acb->qcrs[i].buf);
564     }
565 
566     for (i = 0; i < s->num_children; i++) {
567         bdrv_aio_readv(s->bs[i], sector_num, &acb->qcrs[i].qiov, nb_sectors,
568                        quorum_aio_cb, &acb->qcrs[i]);
569     }
570 
571     return &acb->common;
572 }
573 
574 static BlockDriverAIOCB *quorum_aio_writev(BlockDriverState *bs,
575                                           int64_t sector_num,
576                                           QEMUIOVector *qiov,
577                                           int nb_sectors,
578                                           BlockDriverCompletionFunc *cb,
579                                           void *opaque)
580 {
581     BDRVQuorumState *s = bs->opaque;
582     QuorumAIOCB *acb = quorum_aio_get(s, bs, qiov, sector_num, nb_sectors,
583                                       cb, opaque);
584     int i;
585 
586     for (i = 0; i < s->num_children; i++) {
587         acb->qcrs[i].aiocb = bdrv_aio_writev(s->bs[i], sector_num, qiov,
588                                              nb_sectors, &quorum_aio_cb,
589                                              &acb->qcrs[i]);
590     }
591 
592     return &acb->common;
593 }
594 
595 static int64_t quorum_getlength(BlockDriverState *bs)
596 {
597     BDRVQuorumState *s = bs->opaque;
598     int64_t result;
599     int i;
600 
601     /* check that all file have the same length */
602     result = bdrv_getlength(s->bs[0]);
603     if (result < 0) {
604         return result;
605     }
606     for (i = 1; i < s->num_children; i++) {
607         int64_t value = bdrv_getlength(s->bs[i]);
608         if (value < 0) {
609             return value;
610         }
611         if (value != result) {
612             return -EIO;
613         }
614     }
615 
616     return result;
617 }
618 
619 static void quorum_invalidate_cache(BlockDriverState *bs, Error **errp)
620 {
621     BDRVQuorumState *s = bs->opaque;
622     Error *local_err = NULL;
623     int i;
624 
625     for (i = 0; i < s->num_children; i++) {
626         bdrv_invalidate_cache(s->bs[i], &local_err);
627         if (local_err) {
628             error_propagate(errp, local_err);
629             return;
630         }
631     }
632 }
633 
634 static coroutine_fn int quorum_co_flush(BlockDriverState *bs)
635 {
636     BDRVQuorumState *s = bs->opaque;
637     QuorumVoteVersion *winner = NULL;
638     QuorumVotes error_votes;
639     QuorumVoteValue result_value;
640     int i;
641     int result = 0;
642 
643     QLIST_INIT(&error_votes.vote_list);
644     error_votes.compare = quorum_64bits_compare;
645 
646     for (i = 0; i < s->num_children; i++) {
647         result = bdrv_co_flush(s->bs[i]);
648         result_value.l = result;
649         quorum_count_vote(&error_votes, &result_value, i);
650     }
651 
652     winner = quorum_get_vote_winner(&error_votes);
653     result = winner->value.l;
654 
655     quorum_free_vote_list(&error_votes);
656 
657     return result;
658 }
659 
660 static bool quorum_recurse_is_first_non_filter(BlockDriverState *bs,
661                                                BlockDriverState *candidate)
662 {
663     BDRVQuorumState *s = bs->opaque;
664     int i;
665 
666     for (i = 0; i < s->num_children; i++) {
667         bool perm = bdrv_recurse_is_first_non_filter(s->bs[i],
668                                                      candidate);
669         if (perm) {
670             return true;
671         }
672     }
673 
674     return false;
675 }
676 
677 static int quorum_valid_threshold(int threshold, int num_children, Error **errp)
678 {
679 
680     if (threshold < 1) {
681         error_set(errp, QERR_INVALID_PARAMETER_VALUE,
682                   "vote-threshold", "value >= 1");
683         return -ERANGE;
684     }
685 
686     if (threshold > num_children) {
687         error_setg(errp, "threshold may not exceed children count");
688         return -ERANGE;
689     }
690 
691     return 0;
692 }
693 
694 static QemuOptsList quorum_runtime_opts = {
695     .name = "quorum",
696     .head = QTAILQ_HEAD_INITIALIZER(quorum_runtime_opts.head),
697     .desc = {
698         {
699             .name = QUORUM_OPT_VOTE_THRESHOLD,
700             .type = QEMU_OPT_NUMBER,
701             .help = "The number of vote needed for reaching quorum",
702         },
703         {
704             .name = QUORUM_OPT_BLKVERIFY,
705             .type = QEMU_OPT_BOOL,
706             .help = "Trigger block verify mode if set",
707         },
708         { /* end of list */ }
709     },
710 };
711 
712 static int quorum_open(BlockDriverState *bs, QDict *options, int flags,
713                        Error **errp)
714 {
715     BDRVQuorumState *s = bs->opaque;
716     Error *local_err = NULL;
717     QemuOpts *opts;
718     bool *opened;
719     QDict *sub = NULL;
720     QList *list = NULL;
721     const QListEntry *lentry;
722     int i;
723     int ret = 0;
724 
725     qdict_flatten(options);
726     qdict_extract_subqdict(options, &sub, "children.");
727     qdict_array_split(sub, &list);
728 
729     if (qdict_size(sub)) {
730         error_setg(&local_err, "Invalid option children.%s",
731                    qdict_first(sub)->key);
732         ret = -EINVAL;
733         goto exit;
734     }
735 
736     /* count how many different children are present */
737     s->num_children = qlist_size(list);
738     if (s->num_children < 2) {
739         error_setg(&local_err,
740                    "Number of provided children must be greater than 1");
741         ret = -EINVAL;
742         goto exit;
743     }
744 
745     opts = qemu_opts_create(&quorum_runtime_opts, NULL, 0, &error_abort);
746     qemu_opts_absorb_qdict(opts, options, &local_err);
747     if (local_err) {
748         ret = -EINVAL;
749         goto exit;
750     }
751 
752     s->threshold = qemu_opt_get_number(opts, QUORUM_OPT_VOTE_THRESHOLD, 0);
753 
754     /* and validate it against s->num_children */
755     ret = quorum_valid_threshold(s->threshold, s->num_children, &local_err);
756     if (ret < 0) {
757         goto exit;
758     }
759 
760     /* is the driver in blkverify mode */
761     if (qemu_opt_get_bool(opts, QUORUM_OPT_BLKVERIFY, false) &&
762         s->num_children == 2 && s->threshold == 2) {
763         s->is_blkverify = true;
764     } else if (qemu_opt_get_bool(opts, QUORUM_OPT_BLKVERIFY, false)) {
765         fprintf(stderr, "blkverify mode is set by setting blkverify=on "
766                 "and using two files with vote_threshold=2\n");
767     }
768 
769     /* allocate the children BlockDriverState array */
770     s->bs = g_new0(BlockDriverState *, s->num_children);
771     opened = g_new0(bool, s->num_children);
772 
773     for (i = 0, lentry = qlist_first(list); lentry;
774          lentry = qlist_next(lentry), i++) {
775         QDict *d;
776         QString *string;
777 
778         switch (qobject_type(lentry->value))
779         {
780             /* List of options */
781             case QTYPE_QDICT:
782                 d = qobject_to_qdict(lentry->value);
783                 QINCREF(d);
784                 ret = bdrv_open(&s->bs[i], NULL, NULL, d, flags, NULL,
785                                 &local_err);
786                 break;
787 
788             /* QMP reference */
789             case QTYPE_QSTRING:
790                 string = qobject_to_qstring(lentry->value);
791                 ret = bdrv_open(&s->bs[i], NULL, qstring_get_str(string), NULL,
792                                 flags, NULL, &local_err);
793                 break;
794 
795             default:
796                 error_setg(&local_err, "Specification of child block device %i "
797                            "is invalid", i);
798                 ret = -EINVAL;
799         }
800 
801         if (ret < 0) {
802             goto close_exit;
803         }
804         opened[i] = true;
805     }
806 
807     g_free(opened);
808     goto exit;
809 
810 close_exit:
811     /* cleanup on error */
812     for (i = 0; i < s->num_children; i++) {
813         if (!opened[i]) {
814             continue;
815         }
816         bdrv_unref(s->bs[i]);
817     }
818     g_free(s->bs);
819     g_free(opened);
820 exit:
821     /* propagate error */
822     if (local_err) {
823         error_propagate(errp, local_err);
824     }
825     QDECREF(list);
826     QDECREF(sub);
827     return ret;
828 }
829 
830 static void quorum_close(BlockDriverState *bs)
831 {
832     BDRVQuorumState *s = bs->opaque;
833     int i;
834 
835     for (i = 0; i < s->num_children; i++) {
836         bdrv_unref(s->bs[i]);
837     }
838 
839     g_free(s->bs);
840 }
841 
842 static void quorum_detach_aio_context(BlockDriverState *bs)
843 {
844     BDRVQuorumState *s = bs->opaque;
845     int i;
846 
847     for (i = 0; i < s->num_children; i++) {
848         bdrv_detach_aio_context(s->bs[i]);
849     }
850 }
851 
852 static void quorum_attach_aio_context(BlockDriverState *bs,
853                                       AioContext *new_context)
854 {
855     BDRVQuorumState *s = bs->opaque;
856     int i;
857 
858     for (i = 0; i < s->num_children; i++) {
859         bdrv_attach_aio_context(s->bs[i], new_context);
860     }
861 }
862 
863 static BlockDriver bdrv_quorum = {
864     .format_name                        = "quorum",
865     .protocol_name                      = "quorum",
866 
867     .instance_size                      = sizeof(BDRVQuorumState),
868 
869     .bdrv_file_open                     = quorum_open,
870     .bdrv_close                         = quorum_close,
871 
872     .bdrv_co_flush_to_disk              = quorum_co_flush,
873 
874     .bdrv_getlength                     = quorum_getlength,
875 
876     .bdrv_aio_readv                     = quorum_aio_readv,
877     .bdrv_aio_writev                    = quorum_aio_writev,
878     .bdrv_invalidate_cache              = quorum_invalidate_cache,
879 
880     .bdrv_detach_aio_context            = quorum_detach_aio_context,
881     .bdrv_attach_aio_context            = quorum_attach_aio_context,
882 
883     .is_filter                          = true,
884     .bdrv_recurse_is_first_non_filter   = quorum_recurse_is_first_non_filter,
885 };
886 
887 static void bdrv_quorum_init(void)
888 {
889     bdrv_register(&bdrv_quorum);
890 }
891 
892 block_init(bdrv_quorum_init);
893