2 * QEMU Block driver for RADOS (Ceph)
7 * This work is licensed under the terms of the GNU GPL, version 2. See
8 * the COPYING file in the top-level directory.
10 * Contributions after 2012-01-13 are licensed under the terms of the
11 * GNU GPL, version 2 or (at your option) any later version.
14 #include "qemu/osdep.h"
16 #include <rbd/librbd.h>
17 #include "qapi/error.h"
18 #include "qemu/error-report.h"
19 #include "qemu/module.h"
20 #include "qemu/option.h"
21 #include "block/block_int.h"
22 #include "block/qdict.h"
23 #include "crypto/secret.h"
24 #include "qemu/cutils.h"
25 #include "sysemu/replay.h"
26 #include "qapi/qmp/qstring.h"
27 #include "qapi/qmp/qdict.h"
28 #include "qapi/qmp/qjson.h"
29 #include "qapi/qmp/qlist.h"
30 #include "qapi/qobject-input-visitor.h"
31 #include "qapi/qapi-visit-block-core.h"
34 * When specifying the image filename use:
36 * rbd:poolname/devicename[@snapshotname][:option1=value1[:option2=value2...]]
38 * poolname must be the name of an existing rados pool.
40 * devicename is the name of the rbd image.
42 * Each option given is used to configure rados, and may be any valid
43 * Ceph option, "id", or "conf".
45 * The "id" option indicates what user we should authenticate as to
46 * the Ceph cluster. If it is excluded we will use the Ceph default
49 * The "conf" option specifies a Ceph configuration file to read. If
50 * it is not specified, we will read from the default Ceph locations
51 * (e.g., /etc/ceph/ceph.conf). To avoid reading _any_ configuration
52 * file, specify conf=/dev/null.
54 * Configuration values containing :, @, or = can be escaped with a
58 /* rbd_aio_discard added in 0.1.2 */
59 #if LIBRBD_VERSION_CODE >= LIBRBD_VERSION(0, 1, 2)
60 #define LIBRBD_SUPPORTS_DISCARD
62 #undef LIBRBD_SUPPORTS_DISCARD
65 #define OBJ_MAX_SIZE (1UL << OBJ_DEFAULT_OBJ_ORDER)
67 #define RBD_MAX_SNAPS 100
69 /* The LIBRBD_SUPPORTS_IOVEC is defined in librbd.h */
70 #ifdef LIBRBD_SUPPORTS_IOVEC
71 #define LIBRBD_USE_IOVEC 1
73 #define LIBRBD_USE_IOVEC 0
83 typedef struct RBDAIOCB {
90 struct BDRVRBDState *s;
93 typedef struct RADOSCB {
95 struct BDRVRBDState *s;
101 typedef struct BDRVRBDState {
103 rados_ioctx_t io_ctx;
110 static int qemu_rbd_connect(rados_t *cluster, rados_ioctx_t *io_ctx,
111 BlockdevOptionsRbd *opts, bool cache,
112 const char *keypairs, const char *secretid,
115 static char *qemu_rbd_next_tok(char *src, char delim, char **p)
121 for (end = src; *end; ++end) {
125 if (*end == '\\' && end[1] != '\0') {
136 static void qemu_rbd_unescape(char *src)
140 for (p = src; *src; ++src, ++p) {
141 if (*src == '\\' && src[1] != '\0') {
149 static void qemu_rbd_parse_filename(const char *filename, QDict *options,
154 QList *keypairs = NULL;
157 if (!strstart(filename, "rbd:", &start)) {
158 error_setg(errp, "File name must start with 'rbd:'");
162 buf = g_strdup(start);
165 found_str = qemu_rbd_next_tok(p, '/', &p);
167 error_setg(errp, "Pool name is required");
170 qemu_rbd_unescape(found_str);
171 qdict_put_str(options, "pool", found_str);
173 if (strchr(p, '@')) {
174 found_str = qemu_rbd_next_tok(p, '@', &p);
175 qemu_rbd_unescape(found_str);
176 qdict_put_str(options, "image", found_str);
178 found_str = qemu_rbd_next_tok(p, ':', &p);
179 qemu_rbd_unescape(found_str);
180 qdict_put_str(options, "snapshot", found_str);
182 found_str = qemu_rbd_next_tok(p, ':', &p);
183 qemu_rbd_unescape(found_str);
184 qdict_put_str(options, "image", found_str);
190 /* The following are essentially all key/value pairs, and we treat
191 * 'id' and 'conf' a bit special. Key/value pairs may be in any order. */
194 name = qemu_rbd_next_tok(p, '=', &p);
196 error_setg(errp, "conf option %s has no value", name);
200 qemu_rbd_unescape(name);
202 value = qemu_rbd_next_tok(p, ':', &p);
203 qemu_rbd_unescape(value);
205 if (!strcmp(name, "conf")) {
206 qdict_put_str(options, "conf", value);
207 } else if (!strcmp(name, "id")) {
208 qdict_put_str(options, "user", value);
211 * We pass these internally to qemu_rbd_set_keypairs(), so
212 * we can get away with the simpler list of [ "key1",
213 * "value1", "key2", "value2" ] rather than a raw dict
214 * { "key1": "value1", "key2": "value2" } where we can't
215 * guarantee order, or even a more correct but complex
216 * [ { "key1": "value1" }, { "key2": "value2" } ]
219 keypairs = qlist_new();
221 qlist_append_str(keypairs, name);
222 qlist_append_str(keypairs, value);
227 qdict_put(options, "=keyvalue-pairs",
228 qobject_to_json(QOBJECT(keypairs)));
233 qobject_unref(keypairs);
238 static void qemu_rbd_refresh_limits(BlockDriverState *bs, Error **errp)
240 /* XXX Does RBD support AIO on less than 512-byte alignment? */
241 bs->bl.request_alignment = 512;
245 static int qemu_rbd_set_auth(rados_t cluster, BlockdevOptionsRbd *opts,
251 RbdAuthModeList *auth;
253 if (opts->key_secret) {
254 key = qcrypto_secret_lookup_as_base64(opts->key_secret, errp);
258 r = rados_conf_set(cluster, "key", key);
261 error_setg_errno(errp, -r, "Could not set 'key'");
266 if (opts->has_auth_client_required) {
267 accu = g_string_new("");
268 for (auth = opts->auth_client_required; auth; auth = auth->next) {
270 g_string_append_c(accu, ';');
272 g_string_append(accu, RbdAuthMode_str(auth->value));
274 acr = g_string_free(accu, FALSE);
275 r = rados_conf_set(cluster, "auth_client_required", acr);
278 error_setg_errno(errp, -r,
279 "Could not set 'auth_client_required'");
287 static int qemu_rbd_set_keypairs(rados_t cluster, const char *keypairs_json,
297 if (!keypairs_json) {
300 keypairs = qobject_to(QList,
301 qobject_from_json(keypairs_json, &error_abort));
302 remaining = qlist_size(keypairs) / 2;
305 while (remaining--) {
306 name = qobject_to(QString, qlist_pop(keypairs));
307 value = qobject_to(QString, qlist_pop(keypairs));
308 assert(name && value);
309 key = qstring_get_str(name);
311 ret = rados_conf_set(cluster, key, qstring_get_str(value));
312 qobject_unref(value);
314 error_setg_errno(errp, -ret, "invalid conf option %s", key);
322 qobject_unref(keypairs);
326 static void qemu_rbd_memset(RADOSCB *rcb, int64_t offs)
328 if (LIBRBD_USE_IOVEC) {
329 RBDAIOCB *acb = rcb->acb;
330 iov_memset(acb->qiov->iov, acb->qiov->niov, offs, 0,
331 acb->qiov->size - offs);
333 memset(rcb->buf + offs, 0, rcb->size - offs);
337 static QemuOptsList runtime_opts = {
339 .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
343 .type = QEMU_OPT_STRING,
344 .help = "Rados pool name",
348 .type = QEMU_OPT_STRING,
349 .help = "Image name in the pool",
353 .type = QEMU_OPT_STRING,
354 .help = "Rados config file location",
358 .type = QEMU_OPT_STRING,
359 .help = "Ceph snapshot name",
362 /* maps to 'id' in rados_create() */
364 .type = QEMU_OPT_STRING,
365 .help = "Rados id name",
368 * server.* extracted manually, see qemu_rbd_mon_host()
370 { /* end of list */ }
374 /* FIXME Deprecate and remove keypairs or make it available in QMP. */
375 static int qemu_rbd_do_create(BlockdevCreateOptions *options,
376 const char *keypairs, const char *password_secret,
379 BlockdevCreateOptionsRbd *opts = &options->u.rbd;
381 rados_ioctx_t io_ctx;
385 assert(options->driver == BLOCKDEV_DRIVER_RBD);
386 if (opts->location->has_snapshot) {
387 error_setg(errp, "Can't use snapshot name for image creation");
391 if (opts->has_cluster_size) {
392 int64_t objsize = opts->cluster_size;
393 if ((objsize - 1) & objsize) { /* not a power of 2? */
394 error_setg(errp, "obj size needs to be power of 2");
397 if (objsize < 4096) {
398 error_setg(errp, "obj size too small");
401 obj_order = ctz32(objsize);
404 ret = qemu_rbd_connect(&cluster, &io_ctx, opts->location, false, keypairs,
405 password_secret, errp);
410 ret = rbd_create(io_ctx, opts->location->image, opts->size, &obj_order);
412 error_setg_errno(errp, -ret, "error rbd create");
418 rados_ioctx_destroy(io_ctx);
419 rados_shutdown(cluster);
423 static int qemu_rbd_co_create(BlockdevCreateOptions *options, Error **errp)
425 return qemu_rbd_do_create(options, NULL, NULL, errp);
428 static int coroutine_fn qemu_rbd_co_create_opts(const char *filename,
432 BlockdevCreateOptions *create_options;
433 BlockdevCreateOptionsRbd *rbd_opts;
434 BlockdevOptionsRbd *loc;
435 Error *local_err = NULL;
436 const char *keypairs, *password_secret;
437 QDict *options = NULL;
440 create_options = g_new0(BlockdevCreateOptions, 1);
441 create_options->driver = BLOCKDEV_DRIVER_RBD;
442 rbd_opts = &create_options->u.rbd;
444 rbd_opts->location = g_new0(BlockdevOptionsRbd, 1);
446 password_secret = qemu_opt_get(opts, "password-secret");
448 /* Read out options */
449 rbd_opts->size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
451 rbd_opts->cluster_size = qemu_opt_get_size_del(opts,
452 BLOCK_OPT_CLUSTER_SIZE, 0);
453 rbd_opts->has_cluster_size = (rbd_opts->cluster_size != 0);
455 options = qdict_new();
456 qemu_rbd_parse_filename(filename, options, &local_err);
459 error_propagate(errp, local_err);
464 * Caution: while qdict_get_try_str() is fine, getting non-string
465 * types would require more care. When @options come from -blockdev
466 * or blockdev_add, its members are typed according to the QAPI
467 * schema, but when they come from -drive, they're all QString.
469 loc = rbd_opts->location;
470 loc->pool = g_strdup(qdict_get_try_str(options, "pool"));
471 loc->conf = g_strdup(qdict_get_try_str(options, "conf"));
472 loc->has_conf = !!loc->conf;
473 loc->user = g_strdup(qdict_get_try_str(options, "user"));
474 loc->has_user = !!loc->user;
475 loc->image = g_strdup(qdict_get_try_str(options, "image"));
476 keypairs = qdict_get_try_str(options, "=keyvalue-pairs");
478 ret = qemu_rbd_do_create(create_options, keypairs, password_secret, errp);
484 qobject_unref(options);
485 qapi_free_BlockdevCreateOptions(create_options);
490 * This aio completion is being called from rbd_finish_bh() and runs in qemu
493 static void qemu_rbd_complete_aio(RADOSCB *rcb)
495 RBDAIOCB *acb = rcb->acb;
500 if (acb->cmd != RBD_AIO_READ) {
504 } else if (!acb->error) {
505 acb->ret = rcb->size;
509 qemu_rbd_memset(rcb, 0);
512 } else if (r < rcb->size) {
513 qemu_rbd_memset(rcb, r);
515 acb->ret = rcb->size;
517 } else if (!acb->error) {
524 if (!LIBRBD_USE_IOVEC) {
525 if (acb->cmd == RBD_AIO_READ) {
526 qemu_iovec_from_buf(acb->qiov, 0, acb->bounce, acb->qiov->size);
528 qemu_vfree(acb->bounce);
531 acb->common.cb(acb->common.opaque, (acb->ret > 0 ? 0 : acb->ret));
536 static char *qemu_rbd_mon_host(BlockdevOptionsRbd *opts, Error **errp)
539 const char *host, *port;
541 InetSocketAddressBaseList *p;
544 if (!opts->has_server) {
548 for (cnt = 0, p = opts->server; p; p = p->next) {
552 vals = g_new(const char *, cnt + 1);
554 for (i = 0, p = opts->server; p; p = p->next, i++) {
555 host = p->value->host;
556 port = p->value->port;
558 if (strchr(host, ':')) {
559 vals[i] = g_strdup_printf("[%s]:%s", host, port);
561 vals[i] = g_strdup_printf("%s:%s", host, port);
566 rados_str = i ? g_strjoinv(";", (char **)vals) : NULL;
567 g_strfreev((char **)vals);
571 static int qemu_rbd_connect(rados_t *cluster, rados_ioctx_t *io_ctx,
572 BlockdevOptionsRbd *opts, bool cache,
573 const char *keypairs, const char *secretid,
576 char *mon_host = NULL;
577 Error *local_err = NULL;
581 if (opts->key_secret) {
583 "Legacy 'password-secret' clashes with 'key-secret'");
586 opts->key_secret = g_strdup(secretid);
587 opts->has_key_secret = true;
590 mon_host = qemu_rbd_mon_host(opts, &local_err);
592 error_propagate(errp, local_err);
597 r = rados_create(cluster, opts->user);
599 error_setg_errno(errp, -r, "error initializing");
603 /* try default location when conf=NULL, but ignore failure */
604 r = rados_conf_read_file(*cluster, opts->conf);
605 if (opts->has_conf && r < 0) {
606 error_setg_errno(errp, -r, "error reading conf file %s", opts->conf);
607 goto failed_shutdown;
610 r = qemu_rbd_set_keypairs(*cluster, keypairs, errp);
612 goto failed_shutdown;
616 r = rados_conf_set(*cluster, "mon_host", mon_host);
618 goto failed_shutdown;
622 r = qemu_rbd_set_auth(*cluster, opts, errp);
624 goto failed_shutdown;
628 * Fallback to more conservative semantics if setting cache
629 * options fails. Ignore errors from setting rbd_cache because the
630 * only possible error is that the option does not exist, and
631 * librbd defaults to no caching. If write through caching cannot
632 * be set up, fall back to no caching.
635 rados_conf_set(*cluster, "rbd_cache", "true");
637 rados_conf_set(*cluster, "rbd_cache", "false");
640 r = rados_connect(*cluster);
642 error_setg_errno(errp, -r, "error connecting");
643 goto failed_shutdown;
646 r = rados_ioctx_create(*cluster, opts->pool, io_ctx);
648 error_setg_errno(errp, -r, "error opening pool %s", opts->pool);
649 goto failed_shutdown;
655 rados_shutdown(*cluster);
661 static int qemu_rbd_convert_options(QDict *options, BlockdevOptionsRbd **opts,
665 Error *local_err = NULL;
667 /* Convert the remaining options into a QAPI object */
668 v = qobject_input_visitor_new_flat_confused(options, errp);
673 visit_type_BlockdevOptionsRbd(v, NULL, opts, &local_err);
677 error_propagate(errp, local_err);
684 static int qemu_rbd_attempt_legacy_options(QDict *options,
685 BlockdevOptionsRbd **opts,
691 filename = g_strdup(qdict_get_try_str(options, "filename"));
695 qdict_del(options, "filename");
697 qemu_rbd_parse_filename(filename, options, NULL);
699 /* keypairs freed by caller */
700 *keypairs = g_strdup(qdict_get_try_str(options, "=keyvalue-pairs"));
702 qdict_del(options, "=keyvalue-pairs");
705 r = qemu_rbd_convert_options(options, opts, NULL);
711 static int qemu_rbd_open(BlockDriverState *bs, QDict *options, int flags,
714 BDRVRBDState *s = bs->opaque;
715 BlockdevOptionsRbd *opts = NULL;
717 Error *local_err = NULL;
718 char *keypairs, *secretid;
721 keypairs = g_strdup(qdict_get_try_str(options, "=keyvalue-pairs"));
723 qdict_del(options, "=keyvalue-pairs");
726 secretid = g_strdup(qdict_get_try_str(options, "password-secret"));
728 qdict_del(options, "password-secret");
731 r = qemu_rbd_convert_options(options, &opts, &local_err);
733 /* If keypairs are present, that means some options are present in
734 * the modern option format. Don't attempt to parse legacy option
735 * formats, as we won't support mixed usage. */
737 error_propagate(errp, local_err);
741 /* If the initial attempt to convert and process the options failed,
742 * we may be attempting to open an image file that has the rbd options
743 * specified in the older format consisting of all key/value pairs
744 * encoded in the filename. Go ahead and attempt to parse the
745 * filename, and see if we can pull out the required options. */
746 r = qemu_rbd_attempt_legacy_options(options, &opts, &keypairs);
748 /* Propagate the original error, not the legacy parsing fallback
749 * error, as the latter was just a best-effort attempt. */
750 error_propagate(errp, local_err);
753 /* Take care whenever deciding to actually deprecate; once this ability
754 * is removed, we will not be able to open any images with legacy-styled
755 * backing image strings. */
756 warn_report("RBD options encoded in the filename as keyvalue pairs "
760 /* Remove the processed options from the QDict (the visitor processes
761 * _all_ options in the QDict) */
762 while ((e = qdict_first(options))) {
763 qdict_del(options, e->key);
766 r = qemu_rbd_connect(&s->cluster, &s->io_ctx, opts,
767 !(flags & BDRV_O_NOCACHE), keypairs, secretid, errp);
772 s->snap = g_strdup(opts->snapshot);
773 s->image_name = g_strdup(opts->image);
775 /* rbd_open is always r/w */
776 r = rbd_open(s->io_ctx, s->image_name, &s->image, s->snap);
778 error_setg_errno(errp, -r, "error reading header from %s",
783 r = rbd_get_size(s->image, &s->image_size);
785 error_setg_errno(errp, -r, "error getting image size from %s",
791 /* If we are using an rbd snapshot, we must be r/o, otherwise
793 if (s->snap != NULL) {
794 r = bdrv_apply_auto_read_only(bs, "rbd snapshots are read-only", errp);
805 rados_ioctx_destroy(s->io_ctx);
807 g_free(s->image_name);
808 rados_shutdown(s->cluster);
810 qapi_free_BlockdevOptionsRbd(opts);
817 /* Since RBD is currently always opened R/W via the API,
818 * we just need to check if we are using a snapshot or not, in
819 * order to determine if we will allow it to be R/W */
820 static int qemu_rbd_reopen_prepare(BDRVReopenState *state,
821 BlockReopenQueue *queue, Error **errp)
823 BDRVRBDState *s = state->bs->opaque;
826 if (s->snap && state->flags & BDRV_O_RDWR) {
828 "Cannot change node '%s' to r/w when using RBD snapshot",
829 bdrv_get_device_or_node_name(state->bs));
836 static void qemu_rbd_close(BlockDriverState *bs)
838 BDRVRBDState *s = bs->opaque;
841 rados_ioctx_destroy(s->io_ctx);
843 g_free(s->image_name);
844 rados_shutdown(s->cluster);
847 /* Resize the RBD image and update the 'image_size' with the current size */
848 static int qemu_rbd_resize(BlockDriverState *bs, uint64_t size)
850 BDRVRBDState *s = bs->opaque;
853 r = rbd_resize(s->image, size);
858 s->image_size = size;
863 static const AIOCBInfo rbd_aiocb_info = {
864 .aiocb_size = sizeof(RBDAIOCB),
867 static void rbd_finish_bh(void *opaque)
869 RADOSCB *rcb = opaque;
870 qemu_rbd_complete_aio(rcb);
874 * This is the callback function for rbd_aio_read and _write
876 * Note: this function is being called from a non qemu thread so
877 * we need to be careful about what we do here. Generally we only
878 * schedule a BH, and do the rest of the io completion handling
879 * from rbd_finish_bh() which runs in a qemu context.
881 static void rbd_finish_aiocb(rbd_completion_t c, RADOSCB *rcb)
883 RBDAIOCB *acb = rcb->acb;
885 rcb->ret = rbd_aio_get_return_value(c);
888 replay_bh_schedule_oneshot_event(bdrv_get_aio_context(acb->common.bs),
892 static int rbd_aio_discard_wrapper(rbd_image_t image,
895 rbd_completion_t comp)
897 #ifdef LIBRBD_SUPPORTS_DISCARD
898 return rbd_aio_discard(image, off, len, comp);
904 static int rbd_aio_flush_wrapper(rbd_image_t image,
905 rbd_completion_t comp)
907 #ifdef LIBRBD_SUPPORTS_AIO_FLUSH
908 return rbd_aio_flush(image, comp);
914 static BlockAIOCB *rbd_start_aio(BlockDriverState *bs,
918 BlockCompletionFunc *cb,
927 BDRVRBDState *s = bs->opaque;
929 acb = qemu_aio_get(&rbd_aiocb_info, bs, cb, opaque);
932 assert(!qiov || qiov->size == size);
934 rcb = g_new(RADOSCB, 1);
936 if (!LIBRBD_USE_IOVEC) {
937 if (cmd == RBD_AIO_DISCARD || cmd == RBD_AIO_FLUSH) {
940 acb->bounce = qemu_try_blockalign(bs, qiov->size);
941 if (acb->bounce == NULL) {
945 if (cmd == RBD_AIO_WRITE) {
946 qemu_iovec_to_buf(acb->qiov, 0, acb->bounce, qiov->size);
948 rcb->buf = acb->bounce;
958 r = rbd_aio_create_completion(rcb, (rbd_callback_t) rbd_finish_aiocb, &c);
964 case RBD_AIO_WRITE: {
966 * RBD APIs don't allow us to write more than actual size, so in order
967 * to support growing images, we resize the image before write
968 * operations that exceed the current size.
970 if (off + size > s->image_size) {
971 r = qemu_rbd_resize(bs, off + size);
973 goto failed_completion;
976 #ifdef LIBRBD_SUPPORTS_IOVEC
977 r = rbd_aio_writev(s->image, qiov->iov, qiov->niov, off, c);
979 r = rbd_aio_write(s->image, off, size, rcb->buf, c);
984 #ifdef LIBRBD_SUPPORTS_IOVEC
985 r = rbd_aio_readv(s->image, qiov->iov, qiov->niov, off, c);
987 r = rbd_aio_read(s->image, off, size, rcb->buf, c);
990 case RBD_AIO_DISCARD:
991 r = rbd_aio_discard_wrapper(s->image, off, size, c);
994 r = rbd_aio_flush_wrapper(s->image, c);
1001 goto failed_completion;
1003 return &acb->common;
1009 if (!LIBRBD_USE_IOVEC) {
1010 qemu_vfree(acb->bounce);
1013 qemu_aio_unref(acb);
1017 static BlockAIOCB *qemu_rbd_aio_preadv(BlockDriverState *bs,
1018 uint64_t offset, uint64_t bytes,
1019 QEMUIOVector *qiov, int flags,
1020 BlockCompletionFunc *cb,
1023 return rbd_start_aio(bs, offset, qiov, bytes, cb, opaque,
1027 static BlockAIOCB *qemu_rbd_aio_pwritev(BlockDriverState *bs,
1028 uint64_t offset, uint64_t bytes,
1029 QEMUIOVector *qiov, int flags,
1030 BlockCompletionFunc *cb,
1033 return rbd_start_aio(bs, offset, qiov, bytes, cb, opaque,
1037 #ifdef LIBRBD_SUPPORTS_AIO_FLUSH
1038 static BlockAIOCB *qemu_rbd_aio_flush(BlockDriverState *bs,
1039 BlockCompletionFunc *cb,
1042 return rbd_start_aio(bs, 0, NULL, 0, cb, opaque, RBD_AIO_FLUSH);
1047 static int qemu_rbd_co_flush(BlockDriverState *bs)
1049 #if LIBRBD_VERSION_CODE >= LIBRBD_VERSION(0, 1, 1)
1050 /* rbd_flush added in 0.1.1 */
1051 BDRVRBDState *s = bs->opaque;
1052 return rbd_flush(s->image);
1059 static int qemu_rbd_getinfo(BlockDriverState *bs, BlockDriverInfo *bdi)
1061 BDRVRBDState *s = bs->opaque;
1062 rbd_image_info_t info;
1065 r = rbd_stat(s->image, &info, sizeof(info));
1070 bdi->cluster_size = info.obj_size;
1074 static int64_t qemu_rbd_getlength(BlockDriverState *bs)
1076 BDRVRBDState *s = bs->opaque;
1077 rbd_image_info_t info;
1080 r = rbd_stat(s->image, &info, sizeof(info));
1088 static int coroutine_fn qemu_rbd_co_truncate(BlockDriverState *bs,
1091 PreallocMode prealloc,
1096 if (prealloc != PREALLOC_MODE_OFF) {
1097 error_setg(errp, "Unsupported preallocation mode '%s'",
1098 PreallocMode_str(prealloc));
1102 r = qemu_rbd_resize(bs, offset);
1104 error_setg_errno(errp, -r, "Failed to resize file");
1111 static int qemu_rbd_snap_create(BlockDriverState *bs,
1112 QEMUSnapshotInfo *sn_info)
1114 BDRVRBDState *s = bs->opaque;
1117 if (sn_info->name[0] == '\0') {
1118 return -EINVAL; /* we need a name for rbd snapshots */
1122 * rbd snapshots are using the name as the user controlled unique identifier
1123 * we can't use the rbd snapid for that purpose, as it can't be set
1125 if (sn_info->id_str[0] != '\0' &&
1126 strcmp(sn_info->id_str, sn_info->name) != 0) {
1130 if (strlen(sn_info->name) >= sizeof(sn_info->id_str)) {
1134 r = rbd_snap_create(s->image, sn_info->name);
1136 error_report("failed to create snap: %s", strerror(-r));
1143 static int qemu_rbd_snap_remove(BlockDriverState *bs,
1144 const char *snapshot_id,
1145 const char *snapshot_name,
1148 BDRVRBDState *s = bs->opaque;
1151 if (!snapshot_name) {
1152 error_setg(errp, "rbd need a valid snapshot name");
1156 /* If snapshot_id is specified, it must be equal to name, see
1157 qemu_rbd_snap_list() */
1158 if (snapshot_id && strcmp(snapshot_id, snapshot_name)) {
1160 "rbd do not support snapshot id, it should be NULL or "
1161 "equal to snapshot name");
1165 r = rbd_snap_remove(s->image, snapshot_name);
1167 error_setg_errno(errp, -r, "Failed to remove the snapshot");
1172 static int qemu_rbd_snap_rollback(BlockDriverState *bs,
1173 const char *snapshot_name)
1175 BDRVRBDState *s = bs->opaque;
1177 return rbd_snap_rollback(s->image, snapshot_name);
1180 static int qemu_rbd_snap_list(BlockDriverState *bs,
1181 QEMUSnapshotInfo **psn_tab)
1183 BDRVRBDState *s = bs->opaque;
1184 QEMUSnapshotInfo *sn_info, *sn_tab = NULL;
1186 rbd_snap_info_t *snaps;
1187 int max_snaps = RBD_MAX_SNAPS;
1190 snaps = g_new(rbd_snap_info_t, max_snaps);
1191 snap_count = rbd_snap_list(s->image, snaps, &max_snaps);
1192 if (snap_count <= 0) {
1195 } while (snap_count == -ERANGE);
1197 if (snap_count <= 0) {
1201 sn_tab = g_new0(QEMUSnapshotInfo, snap_count);
1203 for (i = 0; i < snap_count; i++) {
1204 const char *snap_name = snaps[i].name;
1206 sn_info = sn_tab + i;
1207 pstrcpy(sn_info->id_str, sizeof(sn_info->id_str), snap_name);
1208 pstrcpy(sn_info->name, sizeof(sn_info->name), snap_name);
1210 sn_info->vm_state_size = snaps[i].size;
1211 sn_info->date_sec = 0;
1212 sn_info->date_nsec = 0;
1213 sn_info->vm_clock_nsec = 0;
1215 rbd_snap_list_end(snaps);
1223 #ifdef LIBRBD_SUPPORTS_DISCARD
1224 static BlockAIOCB *qemu_rbd_aio_pdiscard(BlockDriverState *bs,
1227 BlockCompletionFunc *cb,
1230 return rbd_start_aio(bs, offset, NULL, bytes, cb, opaque,
1235 #ifdef LIBRBD_SUPPORTS_INVALIDATE
1236 static void coroutine_fn qemu_rbd_co_invalidate_cache(BlockDriverState *bs,
1239 BDRVRBDState *s = bs->opaque;
1240 int r = rbd_invalidate_cache(s->image);
1242 error_setg_errno(errp, -r, "Failed to invalidate the cache");
1247 static QemuOptsList qemu_rbd_create_opts = {
1248 .name = "rbd-create-opts",
1249 .head = QTAILQ_HEAD_INITIALIZER(qemu_rbd_create_opts.head),
1252 .name = BLOCK_OPT_SIZE,
1253 .type = QEMU_OPT_SIZE,
1254 .help = "Virtual disk size"
1257 .name = BLOCK_OPT_CLUSTER_SIZE,
1258 .type = QEMU_OPT_SIZE,
1259 .help = "RBD object size"
1262 .name = "password-secret",
1263 .type = QEMU_OPT_STRING,
1264 .help = "ID of secret providing the password",
1266 { /* end of list */ }
1270 static const char *const qemu_rbd_strong_runtime_opts[] = {
1282 static BlockDriver bdrv_rbd = {
1283 .format_name = "rbd",
1284 .instance_size = sizeof(BDRVRBDState),
1285 .bdrv_parse_filename = qemu_rbd_parse_filename,
1286 .bdrv_refresh_limits = qemu_rbd_refresh_limits,
1287 .bdrv_file_open = qemu_rbd_open,
1288 .bdrv_close = qemu_rbd_close,
1289 .bdrv_reopen_prepare = qemu_rbd_reopen_prepare,
1290 .bdrv_co_create = qemu_rbd_co_create,
1291 .bdrv_co_create_opts = qemu_rbd_co_create_opts,
1292 .bdrv_has_zero_init = bdrv_has_zero_init_1,
1293 .bdrv_has_zero_init_truncate = bdrv_has_zero_init_1,
1294 .bdrv_get_info = qemu_rbd_getinfo,
1295 .create_opts = &qemu_rbd_create_opts,
1296 .bdrv_getlength = qemu_rbd_getlength,
1297 .bdrv_co_truncate = qemu_rbd_co_truncate,
1298 .protocol_name = "rbd",
1300 .bdrv_aio_preadv = qemu_rbd_aio_preadv,
1301 .bdrv_aio_pwritev = qemu_rbd_aio_pwritev,
1303 #ifdef LIBRBD_SUPPORTS_AIO_FLUSH
1304 .bdrv_aio_flush = qemu_rbd_aio_flush,
1306 .bdrv_co_flush_to_disk = qemu_rbd_co_flush,
1309 #ifdef LIBRBD_SUPPORTS_DISCARD
1310 .bdrv_aio_pdiscard = qemu_rbd_aio_pdiscard,
1313 .bdrv_snapshot_create = qemu_rbd_snap_create,
1314 .bdrv_snapshot_delete = qemu_rbd_snap_remove,
1315 .bdrv_snapshot_list = qemu_rbd_snap_list,
1316 .bdrv_snapshot_goto = qemu_rbd_snap_rollback,
1317 #ifdef LIBRBD_SUPPORTS_INVALIDATE
1318 .bdrv_co_invalidate_cache = qemu_rbd_co_invalidate_cache,
1321 .strong_runtime_opts = qemu_rbd_strong_runtime_opts,
1324 static void bdrv_rbd_init(void)
1326 bdrv_register(&bdrv_rbd);
1329 block_init(bdrv_rbd_init);