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;
111 static int qemu_rbd_connect(rados_t *cluster, rados_ioctx_t *io_ctx,
112 BlockdevOptionsRbd *opts, bool cache,
113 const char *keypairs, const char *secretid,
116 static char *qemu_rbd_next_tok(char *src, char delim, char **p)
122 for (end = src; *end; ++end) {
126 if (*end == '\\' && end[1] != '\0') {
137 static void qemu_rbd_unescape(char *src)
141 for (p = src; *src; ++src, ++p) {
142 if (*src == '\\' && src[1] != '\0') {
150 static void qemu_rbd_parse_filename(const char *filename, QDict *options,
155 QList *keypairs = NULL;
156 char *found_str, *image_name;
158 if (!strstart(filename, "rbd:", &start)) {
159 error_setg(errp, "File name must start with 'rbd:'");
163 buf = g_strdup(start);
166 found_str = qemu_rbd_next_tok(p, '/', &p);
168 error_setg(errp, "Pool name is required");
171 qemu_rbd_unescape(found_str);
172 qdict_put_str(options, "pool", found_str);
174 if (strchr(p, '@')) {
175 image_name = qemu_rbd_next_tok(p, '@', &p);
177 found_str = qemu_rbd_next_tok(p, ':', &p);
178 qemu_rbd_unescape(found_str);
179 qdict_put_str(options, "snapshot", found_str);
181 image_name = qemu_rbd_next_tok(p, ':', &p);
183 /* Check for namespace in the image_name */
184 if (strchr(image_name, '/')) {
185 found_str = qemu_rbd_next_tok(image_name, '/', &image_name);
186 qemu_rbd_unescape(found_str);
187 qdict_put_str(options, "namespace", found_str);
189 qdict_put_str(options, "namespace", "");
191 qemu_rbd_unescape(image_name);
192 qdict_put_str(options, "image", image_name);
197 /* The following are essentially all key/value pairs, and we treat
198 * 'id' and 'conf' a bit special. Key/value pairs may be in any order. */
201 name = qemu_rbd_next_tok(p, '=', &p);
203 error_setg(errp, "conf option %s has no value", name);
207 qemu_rbd_unescape(name);
209 value = qemu_rbd_next_tok(p, ':', &p);
210 qemu_rbd_unescape(value);
212 if (!strcmp(name, "conf")) {
213 qdict_put_str(options, "conf", value);
214 } else if (!strcmp(name, "id")) {
215 qdict_put_str(options, "user", value);
218 * We pass these internally to qemu_rbd_set_keypairs(), so
219 * we can get away with the simpler list of [ "key1",
220 * "value1", "key2", "value2" ] rather than a raw dict
221 * { "key1": "value1", "key2": "value2" } where we can't
222 * guarantee order, or even a more correct but complex
223 * [ { "key1": "value1" }, { "key2": "value2" } ]
226 keypairs = qlist_new();
228 qlist_append_str(keypairs, name);
229 qlist_append_str(keypairs, value);
234 qdict_put(options, "=keyvalue-pairs",
235 qstring_from_gstring(qobject_to_json(QOBJECT(keypairs))));
240 qobject_unref(keypairs);
245 static void qemu_rbd_refresh_limits(BlockDriverState *bs, Error **errp)
247 /* XXX Does RBD support AIO on less than 512-byte alignment? */
248 bs->bl.request_alignment = 512;
252 static int qemu_rbd_set_auth(rados_t cluster, BlockdevOptionsRbd *opts,
258 RbdAuthModeList *auth;
260 if (opts->key_secret) {
261 key = qcrypto_secret_lookup_as_base64(opts->key_secret, errp);
265 r = rados_conf_set(cluster, "key", key);
268 error_setg_errno(errp, -r, "Could not set 'key'");
273 if (opts->has_auth_client_required) {
274 accu = g_string_new("");
275 for (auth = opts->auth_client_required; auth; auth = auth->next) {
277 g_string_append_c(accu, ';');
279 g_string_append(accu, RbdAuthMode_str(auth->value));
281 acr = g_string_free(accu, FALSE);
282 r = rados_conf_set(cluster, "auth_client_required", acr);
285 error_setg_errno(errp, -r,
286 "Could not set 'auth_client_required'");
294 static int qemu_rbd_set_keypairs(rados_t cluster, const char *keypairs_json,
304 if (!keypairs_json) {
307 keypairs = qobject_to(QList,
308 qobject_from_json(keypairs_json, &error_abort));
309 remaining = qlist_size(keypairs) / 2;
312 while (remaining--) {
313 name = qobject_to(QString, qlist_pop(keypairs));
314 value = qobject_to(QString, qlist_pop(keypairs));
315 assert(name && value);
316 key = qstring_get_str(name);
318 ret = rados_conf_set(cluster, key, qstring_get_str(value));
319 qobject_unref(value);
321 error_setg_errno(errp, -ret, "invalid conf option %s", key);
329 qobject_unref(keypairs);
333 static void qemu_rbd_memset(RADOSCB *rcb, int64_t offs)
335 if (LIBRBD_USE_IOVEC) {
336 RBDAIOCB *acb = rcb->acb;
337 iov_memset(acb->qiov->iov, acb->qiov->niov, offs, 0,
338 acb->qiov->size - offs);
340 memset(rcb->buf + offs, 0, rcb->size - offs);
344 /* FIXME Deprecate and remove keypairs or make it available in QMP. */
345 static int qemu_rbd_do_create(BlockdevCreateOptions *options,
346 const char *keypairs, const char *password_secret,
349 BlockdevCreateOptionsRbd *opts = &options->u.rbd;
351 rados_ioctx_t io_ctx;
355 assert(options->driver == BLOCKDEV_DRIVER_RBD);
356 if (opts->location->has_snapshot) {
357 error_setg(errp, "Can't use snapshot name for image creation");
361 if (opts->has_cluster_size) {
362 int64_t objsize = opts->cluster_size;
363 if ((objsize - 1) & objsize) { /* not a power of 2? */
364 error_setg(errp, "obj size needs to be power of 2");
367 if (objsize < 4096) {
368 error_setg(errp, "obj size too small");
371 obj_order = ctz32(objsize);
374 ret = qemu_rbd_connect(&cluster, &io_ctx, opts->location, false, keypairs,
375 password_secret, errp);
380 ret = rbd_create(io_ctx, opts->location->image, opts->size, &obj_order);
382 error_setg_errno(errp, -ret, "error rbd create");
388 rados_ioctx_destroy(io_ctx);
389 rados_shutdown(cluster);
393 static int qemu_rbd_co_create(BlockdevCreateOptions *options, Error **errp)
395 return qemu_rbd_do_create(options, NULL, NULL, errp);
398 static int coroutine_fn qemu_rbd_co_create_opts(BlockDriver *drv,
399 const char *filename,
403 BlockdevCreateOptions *create_options;
404 BlockdevCreateOptionsRbd *rbd_opts;
405 BlockdevOptionsRbd *loc;
406 Error *local_err = NULL;
407 const char *keypairs, *password_secret;
408 QDict *options = NULL;
411 create_options = g_new0(BlockdevCreateOptions, 1);
412 create_options->driver = BLOCKDEV_DRIVER_RBD;
413 rbd_opts = &create_options->u.rbd;
415 rbd_opts->location = g_new0(BlockdevOptionsRbd, 1);
417 password_secret = qemu_opt_get(opts, "password-secret");
419 /* Read out options */
420 rbd_opts->size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
422 rbd_opts->cluster_size = qemu_opt_get_size_del(opts,
423 BLOCK_OPT_CLUSTER_SIZE, 0);
424 rbd_opts->has_cluster_size = (rbd_opts->cluster_size != 0);
426 options = qdict_new();
427 qemu_rbd_parse_filename(filename, options, &local_err);
430 error_propagate(errp, local_err);
435 * Caution: while qdict_get_try_str() is fine, getting non-string
436 * types would require more care. When @options come from -blockdev
437 * or blockdev_add, its members are typed according to the QAPI
438 * schema, but when they come from -drive, they're all QString.
440 loc = rbd_opts->location;
441 loc->pool = g_strdup(qdict_get_try_str(options, "pool"));
442 loc->conf = g_strdup(qdict_get_try_str(options, "conf"));
443 loc->has_conf = !!loc->conf;
444 loc->user = g_strdup(qdict_get_try_str(options, "user"));
445 loc->has_user = !!loc->user;
446 loc->q_namespace = g_strdup(qdict_get_try_str(options, "namespace"));
447 loc->image = g_strdup(qdict_get_try_str(options, "image"));
448 keypairs = qdict_get_try_str(options, "=keyvalue-pairs");
450 ret = qemu_rbd_do_create(create_options, keypairs, password_secret, errp);
456 qobject_unref(options);
457 qapi_free_BlockdevCreateOptions(create_options);
462 * This aio completion is being called from rbd_finish_bh() and runs in qemu
465 static void qemu_rbd_complete_aio(RADOSCB *rcb)
467 RBDAIOCB *acb = rcb->acb;
472 if (acb->cmd != RBD_AIO_READ) {
476 } else if (!acb->error) {
477 acb->ret = rcb->size;
481 qemu_rbd_memset(rcb, 0);
484 } else if (r < rcb->size) {
485 qemu_rbd_memset(rcb, r);
487 acb->ret = rcb->size;
489 } else if (!acb->error) {
496 if (!LIBRBD_USE_IOVEC) {
497 if (acb->cmd == RBD_AIO_READ) {
498 qemu_iovec_from_buf(acb->qiov, 0, acb->bounce, acb->qiov->size);
500 qemu_vfree(acb->bounce);
503 acb->common.cb(acb->common.opaque, (acb->ret > 0 ? 0 : acb->ret));
508 static char *qemu_rbd_mon_host(BlockdevOptionsRbd *opts, Error **errp)
511 const char *host, *port;
513 InetSocketAddressBaseList *p;
516 if (!opts->has_server) {
520 for (cnt = 0, p = opts->server; p; p = p->next) {
524 vals = g_new(const char *, cnt + 1);
526 for (i = 0, p = opts->server; p; p = p->next, i++) {
527 host = p->value->host;
528 port = p->value->port;
530 if (strchr(host, ':')) {
531 vals[i] = g_strdup_printf("[%s]:%s", host, port);
533 vals[i] = g_strdup_printf("%s:%s", host, port);
538 rados_str = i ? g_strjoinv(";", (char **)vals) : NULL;
539 g_strfreev((char **)vals);
543 static int qemu_rbd_connect(rados_t *cluster, rados_ioctx_t *io_ctx,
544 BlockdevOptionsRbd *opts, bool cache,
545 const char *keypairs, const char *secretid,
548 char *mon_host = NULL;
549 Error *local_err = NULL;
553 if (opts->key_secret) {
555 "Legacy 'password-secret' clashes with 'key-secret'");
558 opts->key_secret = g_strdup(secretid);
559 opts->has_key_secret = true;
562 mon_host = qemu_rbd_mon_host(opts, &local_err);
564 error_propagate(errp, local_err);
569 r = rados_create(cluster, opts->user);
571 error_setg_errno(errp, -r, "error initializing");
575 /* try default location when conf=NULL, but ignore failure */
576 r = rados_conf_read_file(*cluster, opts->conf);
577 if (opts->has_conf && r < 0) {
578 error_setg_errno(errp, -r, "error reading conf file %s", opts->conf);
579 goto failed_shutdown;
582 r = qemu_rbd_set_keypairs(*cluster, keypairs, errp);
584 goto failed_shutdown;
588 r = rados_conf_set(*cluster, "mon_host", mon_host);
590 goto failed_shutdown;
594 r = qemu_rbd_set_auth(*cluster, opts, errp);
596 goto failed_shutdown;
600 * Fallback to more conservative semantics if setting cache
601 * options fails. Ignore errors from setting rbd_cache because the
602 * only possible error is that the option does not exist, and
603 * librbd defaults to no caching. If write through caching cannot
604 * be set up, fall back to no caching.
607 rados_conf_set(*cluster, "rbd_cache", "true");
609 rados_conf_set(*cluster, "rbd_cache", "false");
612 r = rados_connect(*cluster);
614 error_setg_errno(errp, -r, "error connecting");
615 goto failed_shutdown;
618 r = rados_ioctx_create(*cluster, opts->pool, io_ctx);
620 error_setg_errno(errp, -r, "error opening pool %s", opts->pool);
621 goto failed_shutdown;
624 * Set the namespace after opening the io context on the pool,
625 * if nspace == NULL or if nspace == "", it is just as we did nothing
627 rados_ioctx_set_namespace(*io_ctx, opts->q_namespace);
632 rados_shutdown(*cluster);
638 static int qemu_rbd_convert_options(QDict *options, BlockdevOptionsRbd **opts,
643 /* Convert the remaining options into a QAPI object */
644 v = qobject_input_visitor_new_flat_confused(options, errp);
649 visit_type_BlockdevOptionsRbd(v, NULL, opts, errp);
658 static int qemu_rbd_attempt_legacy_options(QDict *options,
659 BlockdevOptionsRbd **opts,
665 filename = g_strdup(qdict_get_try_str(options, "filename"));
669 qdict_del(options, "filename");
671 qemu_rbd_parse_filename(filename, options, NULL);
673 /* keypairs freed by caller */
674 *keypairs = g_strdup(qdict_get_try_str(options, "=keyvalue-pairs"));
676 qdict_del(options, "=keyvalue-pairs");
679 r = qemu_rbd_convert_options(options, opts, NULL);
685 static int qemu_rbd_open(BlockDriverState *bs, QDict *options, int flags,
688 BDRVRBDState *s = bs->opaque;
689 BlockdevOptionsRbd *opts = NULL;
691 Error *local_err = NULL;
692 char *keypairs, *secretid;
695 keypairs = g_strdup(qdict_get_try_str(options, "=keyvalue-pairs"));
697 qdict_del(options, "=keyvalue-pairs");
700 secretid = g_strdup(qdict_get_try_str(options, "password-secret"));
702 qdict_del(options, "password-secret");
705 r = qemu_rbd_convert_options(options, &opts, &local_err);
707 /* If keypairs are present, that means some options are present in
708 * the modern option format. Don't attempt to parse legacy option
709 * formats, as we won't support mixed usage. */
711 error_propagate(errp, local_err);
715 /* If the initial attempt to convert and process the options failed,
716 * we may be attempting to open an image file that has the rbd options
717 * specified in the older format consisting of all key/value pairs
718 * encoded in the filename. Go ahead and attempt to parse the
719 * filename, and see if we can pull out the required options. */
720 r = qemu_rbd_attempt_legacy_options(options, &opts, &keypairs);
722 /* Propagate the original error, not the legacy parsing fallback
723 * error, as the latter was just a best-effort attempt. */
724 error_propagate(errp, local_err);
727 /* Take care whenever deciding to actually deprecate; once this ability
728 * is removed, we will not be able to open any images with legacy-styled
729 * backing image strings. */
730 warn_report("RBD options encoded in the filename as keyvalue pairs "
734 /* Remove the processed options from the QDict (the visitor processes
735 * _all_ options in the QDict) */
736 while ((e = qdict_first(options))) {
737 qdict_del(options, e->key);
740 r = qemu_rbd_connect(&s->cluster, &s->io_ctx, opts,
741 !(flags & BDRV_O_NOCACHE), keypairs, secretid, errp);
746 s->snap = g_strdup(opts->snapshot);
747 s->image_name = g_strdup(opts->image);
749 /* rbd_open is always r/w */
750 r = rbd_open(s->io_ctx, s->image_name, &s->image, s->snap);
752 error_setg_errno(errp, -r, "error reading header from %s",
757 r = rbd_get_size(s->image, &s->image_size);
759 error_setg_errno(errp, -r, "error getting image size from %s",
765 /* If we are using an rbd snapshot, we must be r/o, otherwise
767 if (s->snap != NULL) {
768 r = bdrv_apply_auto_read_only(bs, "rbd snapshots are read-only", errp);
775 /* When extending regular files, we get zeros from the OS */
776 bs->supported_truncate_flags = BDRV_REQ_ZERO_WRITE;
782 rados_ioctx_destroy(s->io_ctx);
784 g_free(s->image_name);
785 rados_shutdown(s->cluster);
787 qapi_free_BlockdevOptionsRbd(opts);
794 /* Since RBD is currently always opened R/W via the API,
795 * we just need to check if we are using a snapshot or not, in
796 * order to determine if we will allow it to be R/W */
797 static int qemu_rbd_reopen_prepare(BDRVReopenState *state,
798 BlockReopenQueue *queue, Error **errp)
800 BDRVRBDState *s = state->bs->opaque;
803 if (s->snap && state->flags & BDRV_O_RDWR) {
805 "Cannot change node '%s' to r/w when using RBD snapshot",
806 bdrv_get_device_or_node_name(state->bs));
813 static void qemu_rbd_close(BlockDriverState *bs)
815 BDRVRBDState *s = bs->opaque;
818 rados_ioctx_destroy(s->io_ctx);
820 g_free(s->image_name);
821 rados_shutdown(s->cluster);
824 /* Resize the RBD image and update the 'image_size' with the current size */
825 static int qemu_rbd_resize(BlockDriverState *bs, uint64_t size)
827 BDRVRBDState *s = bs->opaque;
830 r = rbd_resize(s->image, size);
835 s->image_size = size;
840 static const AIOCBInfo rbd_aiocb_info = {
841 .aiocb_size = sizeof(RBDAIOCB),
844 static void rbd_finish_bh(void *opaque)
846 RADOSCB *rcb = opaque;
847 qemu_rbd_complete_aio(rcb);
851 * This is the callback function for rbd_aio_read and _write
853 * Note: this function is being called from a non qemu thread so
854 * we need to be careful about what we do here. Generally we only
855 * schedule a BH, and do the rest of the io completion handling
856 * from rbd_finish_bh() which runs in a qemu context.
858 static void rbd_finish_aiocb(rbd_completion_t c, RADOSCB *rcb)
860 RBDAIOCB *acb = rcb->acb;
862 rcb->ret = rbd_aio_get_return_value(c);
865 replay_bh_schedule_oneshot_event(bdrv_get_aio_context(acb->common.bs),
869 static int rbd_aio_discard_wrapper(rbd_image_t image,
872 rbd_completion_t comp)
874 #ifdef LIBRBD_SUPPORTS_DISCARD
875 return rbd_aio_discard(image, off, len, comp);
881 static int rbd_aio_flush_wrapper(rbd_image_t image,
882 rbd_completion_t comp)
884 #ifdef LIBRBD_SUPPORTS_AIO_FLUSH
885 return rbd_aio_flush(image, comp);
891 static BlockAIOCB *rbd_start_aio(BlockDriverState *bs,
895 BlockCompletionFunc *cb,
904 BDRVRBDState *s = bs->opaque;
906 acb = qemu_aio_get(&rbd_aiocb_info, bs, cb, opaque);
909 assert(!qiov || qiov->size == size);
911 rcb = g_new(RADOSCB, 1);
913 if (!LIBRBD_USE_IOVEC) {
914 if (cmd == RBD_AIO_DISCARD || cmd == RBD_AIO_FLUSH) {
917 acb->bounce = qemu_try_blockalign(bs, qiov->size);
918 if (acb->bounce == NULL) {
922 if (cmd == RBD_AIO_WRITE) {
923 qemu_iovec_to_buf(acb->qiov, 0, acb->bounce, qiov->size);
925 rcb->buf = acb->bounce;
935 r = rbd_aio_create_completion(rcb, (rbd_callback_t) rbd_finish_aiocb, &c);
941 case RBD_AIO_WRITE: {
943 * RBD APIs don't allow us to write more than actual size, so in order
944 * to support growing images, we resize the image before write
945 * operations that exceed the current size.
947 if (off + size > s->image_size) {
948 r = qemu_rbd_resize(bs, off + size);
950 goto failed_completion;
953 #ifdef LIBRBD_SUPPORTS_IOVEC
954 r = rbd_aio_writev(s->image, qiov->iov, qiov->niov, off, c);
956 r = rbd_aio_write(s->image, off, size, rcb->buf, c);
961 #ifdef LIBRBD_SUPPORTS_IOVEC
962 r = rbd_aio_readv(s->image, qiov->iov, qiov->niov, off, c);
964 r = rbd_aio_read(s->image, off, size, rcb->buf, c);
967 case RBD_AIO_DISCARD:
968 r = rbd_aio_discard_wrapper(s->image, off, size, c);
971 r = rbd_aio_flush_wrapper(s->image, c);
978 goto failed_completion;
986 if (!LIBRBD_USE_IOVEC) {
987 qemu_vfree(acb->bounce);
994 static BlockAIOCB *qemu_rbd_aio_preadv(BlockDriverState *bs,
995 uint64_t offset, uint64_t bytes,
996 QEMUIOVector *qiov, int flags,
997 BlockCompletionFunc *cb,
1000 return rbd_start_aio(bs, offset, qiov, bytes, cb, opaque,
1004 static BlockAIOCB *qemu_rbd_aio_pwritev(BlockDriverState *bs,
1005 uint64_t offset, uint64_t bytes,
1006 QEMUIOVector *qiov, int flags,
1007 BlockCompletionFunc *cb,
1010 return rbd_start_aio(bs, offset, qiov, bytes, cb, opaque,
1014 #ifdef LIBRBD_SUPPORTS_AIO_FLUSH
1015 static BlockAIOCB *qemu_rbd_aio_flush(BlockDriverState *bs,
1016 BlockCompletionFunc *cb,
1019 return rbd_start_aio(bs, 0, NULL, 0, cb, opaque, RBD_AIO_FLUSH);
1024 static int qemu_rbd_co_flush(BlockDriverState *bs)
1026 #if LIBRBD_VERSION_CODE >= LIBRBD_VERSION(0, 1, 1)
1027 /* rbd_flush added in 0.1.1 */
1028 BDRVRBDState *s = bs->opaque;
1029 return rbd_flush(s->image);
1036 static int qemu_rbd_getinfo(BlockDriverState *bs, BlockDriverInfo *bdi)
1038 BDRVRBDState *s = bs->opaque;
1039 rbd_image_info_t info;
1042 r = rbd_stat(s->image, &info, sizeof(info));
1047 bdi->cluster_size = info.obj_size;
1051 static int64_t qemu_rbd_getlength(BlockDriverState *bs)
1053 BDRVRBDState *s = bs->opaque;
1054 rbd_image_info_t info;
1057 r = rbd_stat(s->image, &info, sizeof(info));
1065 static int coroutine_fn qemu_rbd_co_truncate(BlockDriverState *bs,
1068 PreallocMode prealloc,
1069 BdrvRequestFlags flags,
1074 if (prealloc != PREALLOC_MODE_OFF) {
1075 error_setg(errp, "Unsupported preallocation mode '%s'",
1076 PreallocMode_str(prealloc));
1080 r = qemu_rbd_resize(bs, offset);
1082 error_setg_errno(errp, -r, "Failed to resize file");
1089 static int qemu_rbd_snap_create(BlockDriverState *bs,
1090 QEMUSnapshotInfo *sn_info)
1092 BDRVRBDState *s = bs->opaque;
1095 if (sn_info->name[0] == '\0') {
1096 return -EINVAL; /* we need a name for rbd snapshots */
1100 * rbd snapshots are using the name as the user controlled unique identifier
1101 * we can't use the rbd snapid for that purpose, as it can't be set
1103 if (sn_info->id_str[0] != '\0' &&
1104 strcmp(sn_info->id_str, sn_info->name) != 0) {
1108 if (strlen(sn_info->name) >= sizeof(sn_info->id_str)) {
1112 r = rbd_snap_create(s->image, sn_info->name);
1114 error_report("failed to create snap: %s", strerror(-r));
1121 static int qemu_rbd_snap_remove(BlockDriverState *bs,
1122 const char *snapshot_id,
1123 const char *snapshot_name,
1126 BDRVRBDState *s = bs->opaque;
1129 if (!snapshot_name) {
1130 error_setg(errp, "rbd need a valid snapshot name");
1134 /* If snapshot_id is specified, it must be equal to name, see
1135 qemu_rbd_snap_list() */
1136 if (snapshot_id && strcmp(snapshot_id, snapshot_name)) {
1138 "rbd do not support snapshot id, it should be NULL or "
1139 "equal to snapshot name");
1143 r = rbd_snap_remove(s->image, snapshot_name);
1145 error_setg_errno(errp, -r, "Failed to remove the snapshot");
1150 static int qemu_rbd_snap_rollback(BlockDriverState *bs,
1151 const char *snapshot_name)
1153 BDRVRBDState *s = bs->opaque;
1155 return rbd_snap_rollback(s->image, snapshot_name);
1158 static int qemu_rbd_snap_list(BlockDriverState *bs,
1159 QEMUSnapshotInfo **psn_tab)
1161 BDRVRBDState *s = bs->opaque;
1162 QEMUSnapshotInfo *sn_info, *sn_tab = NULL;
1164 rbd_snap_info_t *snaps;
1165 int max_snaps = RBD_MAX_SNAPS;
1168 snaps = g_new(rbd_snap_info_t, max_snaps);
1169 snap_count = rbd_snap_list(s->image, snaps, &max_snaps);
1170 if (snap_count <= 0) {
1173 } while (snap_count == -ERANGE);
1175 if (snap_count <= 0) {
1179 sn_tab = g_new0(QEMUSnapshotInfo, snap_count);
1181 for (i = 0; i < snap_count; i++) {
1182 const char *snap_name = snaps[i].name;
1184 sn_info = sn_tab + i;
1185 pstrcpy(sn_info->id_str, sizeof(sn_info->id_str), snap_name);
1186 pstrcpy(sn_info->name, sizeof(sn_info->name), snap_name);
1188 sn_info->vm_state_size = snaps[i].size;
1189 sn_info->date_sec = 0;
1190 sn_info->date_nsec = 0;
1191 sn_info->vm_clock_nsec = 0;
1193 rbd_snap_list_end(snaps);
1201 #ifdef LIBRBD_SUPPORTS_DISCARD
1202 static BlockAIOCB *qemu_rbd_aio_pdiscard(BlockDriverState *bs,
1205 BlockCompletionFunc *cb,
1208 return rbd_start_aio(bs, offset, NULL, bytes, cb, opaque,
1213 #ifdef LIBRBD_SUPPORTS_INVALIDATE
1214 static void coroutine_fn qemu_rbd_co_invalidate_cache(BlockDriverState *bs,
1217 BDRVRBDState *s = bs->opaque;
1218 int r = rbd_invalidate_cache(s->image);
1220 error_setg_errno(errp, -r, "Failed to invalidate the cache");
1225 static QemuOptsList qemu_rbd_create_opts = {
1226 .name = "rbd-create-opts",
1227 .head = QTAILQ_HEAD_INITIALIZER(qemu_rbd_create_opts.head),
1230 .name = BLOCK_OPT_SIZE,
1231 .type = QEMU_OPT_SIZE,
1232 .help = "Virtual disk size"
1235 .name = BLOCK_OPT_CLUSTER_SIZE,
1236 .type = QEMU_OPT_SIZE,
1237 .help = "RBD object size"
1240 .name = "password-secret",
1241 .type = QEMU_OPT_STRING,
1242 .help = "ID of secret providing the password",
1244 { /* end of list */ }
1248 static const char *const qemu_rbd_strong_runtime_opts[] = {
1261 static BlockDriver bdrv_rbd = {
1262 .format_name = "rbd",
1263 .instance_size = sizeof(BDRVRBDState),
1264 .bdrv_parse_filename = qemu_rbd_parse_filename,
1265 .bdrv_refresh_limits = qemu_rbd_refresh_limits,
1266 .bdrv_file_open = qemu_rbd_open,
1267 .bdrv_close = qemu_rbd_close,
1268 .bdrv_reopen_prepare = qemu_rbd_reopen_prepare,
1269 .bdrv_co_create = qemu_rbd_co_create,
1270 .bdrv_co_create_opts = qemu_rbd_co_create_opts,
1271 .bdrv_has_zero_init = bdrv_has_zero_init_1,
1272 .bdrv_get_info = qemu_rbd_getinfo,
1273 .create_opts = &qemu_rbd_create_opts,
1274 .bdrv_getlength = qemu_rbd_getlength,
1275 .bdrv_co_truncate = qemu_rbd_co_truncate,
1276 .protocol_name = "rbd",
1278 .bdrv_aio_preadv = qemu_rbd_aio_preadv,
1279 .bdrv_aio_pwritev = qemu_rbd_aio_pwritev,
1281 #ifdef LIBRBD_SUPPORTS_AIO_FLUSH
1282 .bdrv_aio_flush = qemu_rbd_aio_flush,
1284 .bdrv_co_flush_to_disk = qemu_rbd_co_flush,
1287 #ifdef LIBRBD_SUPPORTS_DISCARD
1288 .bdrv_aio_pdiscard = qemu_rbd_aio_pdiscard,
1291 .bdrv_snapshot_create = qemu_rbd_snap_create,
1292 .bdrv_snapshot_delete = qemu_rbd_snap_remove,
1293 .bdrv_snapshot_list = qemu_rbd_snap_list,
1294 .bdrv_snapshot_goto = qemu_rbd_snap_rollback,
1295 #ifdef LIBRBD_SUPPORTS_INVALIDATE
1296 .bdrv_co_invalidate_cache = qemu_rbd_co_invalidate_cache,
1299 .strong_runtime_opts = qemu_rbd_strong_runtime_opts,
1302 static void bdrv_rbd_init(void)
1304 bdrv_register(&bdrv_rbd);
1307 block_init(bdrv_rbd_init);