2 * QEMU Block driver for RADOS (Ceph)
4 * Copyright (C) 2010-2011 Christian Brunner <chb@muc.de>,
5 * Josh Durgin <josh.durgin@dreamhost.com>
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/option.h"
20 #include "block/block_int.h"
21 #include "block/qdict.h"
22 #include "crypto/secret.h"
23 #include "qemu/cutils.h"
24 #include "qapi/qmp/qstring.h"
25 #include "qapi/qmp/qdict.h"
26 #include "qapi/qmp/qjson.h"
27 #include "qapi/qmp/qlist.h"
28 #include "qapi/qobject-input-visitor.h"
29 #include "qapi/qapi-visit-block-core.h"
32 * When specifying the image filename use:
34 * rbd:poolname/devicename[@snapshotname][:option1=value1[:option2=value2...]]
36 * poolname must be the name of an existing rados pool.
38 * devicename is the name of the rbd image.
40 * Each option given is used to configure rados, and may be any valid
41 * Ceph option, "id", or "conf".
43 * The "id" option indicates what user we should authenticate as to
44 * the Ceph cluster. If it is excluded we will use the Ceph default
47 * The "conf" option specifies a Ceph configuration file to read. If
48 * it is not specified, we will read from the default Ceph locations
49 * (e.g., /etc/ceph/ceph.conf). To avoid reading _any_ configuration
50 * file, specify conf=/dev/null.
52 * Configuration values containing :, @, or = can be escaped with a
56 /* rbd_aio_discard added in 0.1.2 */
57 #if LIBRBD_VERSION_CODE >= LIBRBD_VERSION(0, 1, 2)
58 #define LIBRBD_SUPPORTS_DISCARD
60 #undef LIBRBD_SUPPORTS_DISCARD
63 #define OBJ_MAX_SIZE (1UL << OBJ_DEFAULT_OBJ_ORDER)
65 #define RBD_MAX_SNAPS 100
67 /* The LIBRBD_SUPPORTS_IOVEC is defined in librbd.h */
68 #ifdef LIBRBD_SUPPORTS_IOVEC
69 #define LIBRBD_USE_IOVEC 1
71 #define LIBRBD_USE_IOVEC 0
81 typedef struct RBDAIOCB
{
88 struct BDRVRBDState
*s
;
91 typedef struct RADOSCB
{
93 struct BDRVRBDState
*s
;
99 typedef struct BDRVRBDState
{
101 rados_ioctx_t io_ctx
;
107 static int qemu_rbd_connect(rados_t
*cluster
, rados_ioctx_t
*io_ctx
,
108 BlockdevOptionsRbd
*opts
, bool cache
,
109 const char *keypairs
, const char *secretid
,
112 static char *qemu_rbd_next_tok(char *src
, char delim
, char **p
)
118 for (end
= src
; *end
; ++end
) {
122 if (*end
== '\\' && end
[1] != '\0') {
133 static void qemu_rbd_unescape(char *src
)
137 for (p
= src
; *src
; ++src
, ++p
) {
138 if (*src
== '\\' && src
[1] != '\0') {
146 static void qemu_rbd_parse_filename(const char *filename
, QDict
*options
,
151 QList
*keypairs
= NULL
;
154 if (!strstart(filename
, "rbd:", &start
)) {
155 error_setg(errp
, "File name must start with 'rbd:'");
159 buf
= g_strdup(start
);
162 found_str
= qemu_rbd_next_tok(p
, '/', &p
);
164 error_setg(errp
, "Pool name is required");
167 qemu_rbd_unescape(found_str
);
168 qdict_put_str(options
, "pool", found_str
);
170 if (strchr(p
, '@')) {
171 found_str
= qemu_rbd_next_tok(p
, '@', &p
);
172 qemu_rbd_unescape(found_str
);
173 qdict_put_str(options
, "image", found_str
);
175 found_str
= qemu_rbd_next_tok(p
, ':', &p
);
176 qemu_rbd_unescape(found_str
);
177 qdict_put_str(options
, "snapshot", found_str
);
179 found_str
= qemu_rbd_next_tok(p
, ':', &p
);
180 qemu_rbd_unescape(found_str
);
181 qdict_put_str(options
, "image", found_str
);
187 /* The following are essentially all key/value pairs, and we treat
188 * 'id' and 'conf' a bit special. Key/value pairs may be in any order. */
191 name
= qemu_rbd_next_tok(p
, '=', &p
);
193 error_setg(errp
, "conf option %s has no value", name
);
197 qemu_rbd_unescape(name
);
199 value
= qemu_rbd_next_tok(p
, ':', &p
);
200 qemu_rbd_unescape(value
);
202 if (!strcmp(name
, "conf")) {
203 qdict_put_str(options
, "conf", value
);
204 } else if (!strcmp(name
, "id")) {
205 qdict_put_str(options
, "user", value
);
208 * We pass these internally to qemu_rbd_set_keypairs(), so
209 * we can get away with the simpler list of [ "key1",
210 * "value1", "key2", "value2" ] rather than a raw dict
211 * { "key1": "value1", "key2": "value2" } where we can't
212 * guarantee order, or even a more correct but complex
213 * [ { "key1": "value1" }, { "key2": "value2" } ]
216 keypairs
= qlist_new();
218 qlist_append_str(keypairs
, name
);
219 qlist_append_str(keypairs
, value
);
224 qdict_put(options
, "=keyvalue-pairs",
225 qobject_to_json(QOBJECT(keypairs
)));
230 qobject_unref(keypairs
);
235 static void qemu_rbd_refresh_limits(BlockDriverState
*bs
, Error
**errp
)
237 /* XXX Does RBD support AIO on less than 512-byte alignment? */
238 bs
->bl
.request_alignment
= 512;
242 static int qemu_rbd_set_auth(rados_t cluster
, BlockdevOptionsRbd
*opts
,
248 RbdAuthModeList
*auth
;
250 if (opts
->key_secret
) {
251 key
= qcrypto_secret_lookup_as_base64(opts
->key_secret
, errp
);
255 r
= rados_conf_set(cluster
, "key", key
);
258 error_setg_errno(errp
, -r
, "Could not set 'key'");
263 if (opts
->has_auth_client_required
) {
264 accu
= g_string_new("");
265 for (auth
= opts
->auth_client_required
; auth
; auth
= auth
->next
) {
267 g_string_append_c(accu
, ';');
269 g_string_append(accu
, RbdAuthMode_str(auth
->value
));
271 acr
= g_string_free(accu
, FALSE
);
272 r
= rados_conf_set(cluster
, "auth_client_required", acr
);
275 error_setg_errno(errp
, -r
,
276 "Could not set 'auth_client_required'");
284 #pragma GCC diagnostic ignored "-Wsuggest-attribute=format"
285 static int qemu_rbd_set_keypairs(rados_t cluster
, const char *keypairs_json
,
295 if (!keypairs_json
) {
298 keypairs
= qobject_to(QList
,
299 qobject_from_json(keypairs_json
, &error_abort
));
300 remaining
= qlist_size(keypairs
) / 2;
303 while (remaining
--) {
304 name
= qobject_to(QString
, qlist_pop(keypairs
));
305 value
= qobject_to(QString
, qlist_pop(keypairs
));
306 assert(name
&& value
);
307 key
= qstring_get_str(name
);
309 ret
= rados_conf_set(cluster
, key
, qstring_get_str(value
));
310 qobject_unref(value
);
312 error_setg_errno(errp
, -ret
, "invalid conf option %s", key
);
320 qobject_unref(keypairs
);
324 static void qemu_rbd_memset(RADOSCB
*rcb
, int64_t offs
)
326 if (LIBRBD_USE_IOVEC
) {
327 RBDAIOCB
*acb
= rcb
->acb
;
328 iov_memset(acb
->qiov
->iov
, acb
->qiov
->niov
, offs
, 0,
329 acb
->qiov
->size
- offs
);
331 memset(rcb
->buf
+ offs
, 0, rcb
->size
- offs
);
335 static QemuOptsList runtime_opts
= {
337 .head
= QTAILQ_HEAD_INITIALIZER(runtime_opts
.head
),
341 .type
= QEMU_OPT_STRING
,
342 .help
= "Rados pool name",
346 .type
= QEMU_OPT_STRING
,
347 .help
= "Image name in the pool",
351 .type
= QEMU_OPT_STRING
,
352 .help
= "Rados config file location",
356 .type
= QEMU_OPT_STRING
,
357 .help
= "Ceph snapshot name",
360 /* maps to 'id' in rados_create() */
362 .type
= QEMU_OPT_STRING
,
363 .help
= "Rados id name",
366 * server.* extracted manually, see qemu_rbd_mon_host()
368 { /* end of list */ }
372 /* FIXME Deprecate and remove keypairs or make it available in QMP. */
373 static int qemu_rbd_do_create(BlockdevCreateOptions
*options
,
374 const char *keypairs
, const char *password_secret
,
377 BlockdevCreateOptionsRbd
*opts
= &options
->u
.rbd
;
379 rados_ioctx_t io_ctx
;
383 assert(options
->driver
== BLOCKDEV_DRIVER_RBD
);
384 if (opts
->location
->has_snapshot
) {
385 error_setg(errp
, "Can't use snapshot name for image creation");
389 if (opts
->has_cluster_size
) {
390 int64_t objsize
= opts
->cluster_size
;
391 if ((objsize
- 1) & objsize
) { /* not a power of 2? */
392 error_setg(errp
, "obj size needs to be power of 2");
395 if (objsize
< 4096) {
396 error_setg(errp
, "obj size too small");
399 obj_order
= ctz32(objsize
);
402 ret
= qemu_rbd_connect(&cluster
, &io_ctx
, opts
->location
, false, keypairs
,
403 password_secret
, errp
);
408 ret
= rbd_create(io_ctx
, opts
->location
->image
, opts
->size
, &obj_order
);
410 error_setg_errno(errp
, -ret
, "error rbd create");
416 rados_ioctx_destroy(io_ctx
);
417 rados_shutdown(cluster
);
421 static int qemu_rbd_co_create(BlockdevCreateOptions
*options
, Error
**errp
)
423 return qemu_rbd_do_create(options
, NULL
, NULL
, errp
);
426 static int coroutine_fn
qemu_rbd_co_create_opts(const char *filename
,
430 BlockdevCreateOptions
*create_options
;
431 BlockdevCreateOptionsRbd
*rbd_opts
;
432 BlockdevOptionsRbd
*loc
;
433 Error
*local_err
= NULL
;
434 const char *keypairs
, *password_secret
;
435 QDict
*options
= NULL
;
438 create_options
= g_new0(BlockdevCreateOptions
, 1);
439 create_options
->driver
= BLOCKDEV_DRIVER_RBD
;
440 rbd_opts
= &create_options
->u
.rbd
;
442 rbd_opts
->location
= g_new0(BlockdevOptionsRbd
, 1);
444 password_secret
= qemu_opt_get(opts
, "password-secret");
446 /* Read out options */
447 rbd_opts
->size
= ROUND_UP(qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0),
449 rbd_opts
->cluster_size
= qemu_opt_get_size_del(opts
,
450 BLOCK_OPT_CLUSTER_SIZE
, 0);
451 rbd_opts
->has_cluster_size
= (rbd_opts
->cluster_size
!= 0);
453 options
= qdict_new();
454 qemu_rbd_parse_filename(filename
, options
, &local_err
);
457 error_propagate(errp
, local_err
);
462 * Caution: while qdict_get_try_str() is fine, getting non-string
463 * types would require more care. When @options come from -blockdev
464 * or blockdev_add, its members are typed according to the QAPI
465 * schema, but when they come from -drive, they're all QString.
467 loc
= rbd_opts
->location
;
468 loc
->pool
= g_strdup(qdict_get_try_str(options
, "pool"));
469 loc
->conf
= g_strdup(qdict_get_try_str(options
, "conf"));
470 loc
->has_conf
= !!loc
->conf
;
471 loc
->user
= g_strdup(qdict_get_try_str(options
, "user"));
472 loc
->has_user
= !!loc
->user
;
473 loc
->image
= g_strdup(qdict_get_try_str(options
, "image"));
474 keypairs
= qdict_get_try_str(options
, "=keyvalue-pairs");
476 ret
= qemu_rbd_do_create(create_options
, keypairs
, password_secret
, errp
);
482 qobject_unref(options
);
483 qapi_free_BlockdevCreateOptions(create_options
);
488 * This aio completion is being called from rbd_finish_bh() and runs in qemu
491 static void qemu_rbd_complete_aio(RADOSCB
*rcb
)
493 RBDAIOCB
*acb
= rcb
->acb
;
498 if (acb
->cmd
!= RBD_AIO_READ
) {
502 } else if (!acb
->error
) {
503 acb
->ret
= rcb
->size
;
507 qemu_rbd_memset(rcb
, 0);
510 } else if (r
< rcb
->size
) {
511 qemu_rbd_memset(rcb
, r
);
513 acb
->ret
= rcb
->size
;
515 } else if (!acb
->error
) {
522 if (!LIBRBD_USE_IOVEC
) {
523 if (acb
->cmd
== RBD_AIO_READ
) {
524 qemu_iovec_from_buf(acb
->qiov
, 0, acb
->bounce
, acb
->qiov
->size
);
526 qemu_vfree(acb
->bounce
);
529 acb
->common
.cb(acb
->common
.opaque
, (acb
->ret
> 0 ? 0 : acb
->ret
));
534 static char *qemu_rbd_mon_host(BlockdevOptionsRbd
*opts
, Error
**errp
)
537 const char *host
, *port
;
539 InetSocketAddressBaseList
*p
;
542 if (!opts
->has_server
) {
546 for (cnt
= 0, p
= opts
->server
; p
; p
= p
->next
) {
550 vals
= g_new(const char *, cnt
+ 1);
552 for (i
= 0, p
= opts
->server
; p
; p
= p
->next
, i
++) {
553 host
= p
->value
->host
;
554 port
= p
->value
->port
;
556 if (strchr(host
, ':')) {
557 vals
[i
] = g_strdup_printf("[%s]:%s", host
, port
);
559 vals
[i
] = g_strdup_printf("%s:%s", host
, port
);
564 rados_str
= i
? g_strjoinv(";", (char **)vals
) : NULL
;
565 g_strfreev((char **)vals
);
569 static int qemu_rbd_connect(rados_t
*cluster
, rados_ioctx_t
*io_ctx
,
570 BlockdevOptionsRbd
*opts
, bool cache
,
571 const char *keypairs
, const char *secretid
,
574 char *mon_host
= NULL
;
575 Error
*local_err
= NULL
;
579 if (opts
->key_secret
) {
581 "Legacy 'password-secret' clashes with 'key-secret'");
584 opts
->key_secret
= g_strdup(secretid
);
585 opts
->has_key_secret
= true;
588 mon_host
= qemu_rbd_mon_host(opts
, &local_err
);
590 error_propagate(errp
, local_err
);
595 r
= rados_create(cluster
, opts
->user
);
597 error_setg_errno(errp
, -r
, "error initializing");
601 /* try default location when conf=NULL, but ignore failure */
602 r
= rados_conf_read_file(*cluster
, opts
->conf
);
603 if (opts
->has_conf
&& r
< 0) {
604 error_setg_errno(errp
, -r
, "error reading conf file %s", opts
->conf
);
605 goto failed_shutdown
;
608 r
= qemu_rbd_set_keypairs(*cluster
, keypairs
, errp
);
610 goto failed_shutdown
;
614 r
= rados_conf_set(*cluster
, "mon_host", mon_host
);
616 goto failed_shutdown
;
620 r
= qemu_rbd_set_auth(*cluster
, opts
, errp
);
622 goto failed_shutdown
;
626 * Fallback to more conservative semantics if setting cache
627 * options fails. Ignore errors from setting rbd_cache because the
628 * only possible error is that the option does not exist, and
629 * librbd defaults to no caching. If write through caching cannot
630 * be set up, fall back to no caching.
633 rados_conf_set(*cluster
, "rbd_cache", "true");
635 rados_conf_set(*cluster
, "rbd_cache", "false");
638 r
= rados_connect(*cluster
);
640 error_setg_errno(errp
, -r
, "error connecting");
641 goto failed_shutdown
;
644 r
= rados_ioctx_create(*cluster
, opts
->pool
, io_ctx
);
646 error_setg_errno(errp
, -r
, "error opening pool %s", opts
->pool
);
647 goto failed_shutdown
;
653 rados_shutdown(*cluster
);
659 static int qemu_rbd_convert_options(QDict
*options
, BlockdevOptionsRbd
**opts
,
663 Error
*local_err
= NULL
;
665 /* Convert the remaining options into a QAPI object */
666 v
= qobject_input_visitor_new_flat_confused(options
, errp
);
671 visit_type_BlockdevOptionsRbd(v
, NULL
, opts
, &local_err
);
675 error_propagate(errp
, local_err
);
682 static int qemu_rbd_attempt_legacy_options(QDict
*options
,
683 BlockdevOptionsRbd
**opts
,
689 filename
= g_strdup(qdict_get_try_str(options
, "filename"));
693 qdict_del(options
, "filename");
695 qemu_rbd_parse_filename(filename
, options
, NULL
);
697 /* keypairs freed by caller */
698 *keypairs
= g_strdup(qdict_get_try_str(options
, "=keyvalue-pairs"));
700 qdict_del(options
, "=keyvalue-pairs");
703 r
= qemu_rbd_convert_options(options
, opts
, NULL
);
709 static int qemu_rbd_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
712 BDRVRBDState
*s
= bs
->opaque
;
713 BlockdevOptionsRbd
*opts
= NULL
;
715 Error
*local_err
= NULL
;
716 char *keypairs
, *secretid
;
719 keypairs
= g_strdup(qdict_get_try_str(options
, "=keyvalue-pairs"));
721 qdict_del(options
, "=keyvalue-pairs");
724 secretid
= g_strdup(qdict_get_try_str(options
, "password-secret"));
726 qdict_del(options
, "password-secret");
729 r
= qemu_rbd_convert_options(options
, &opts
, &local_err
);
731 /* If keypairs are present, that means some options are present in
732 * the modern option format. Don't attempt to parse legacy option
733 * formats, as we won't support mixed usage. */
735 error_propagate(errp
, local_err
);
739 /* If the initial attempt to convert and process the options failed,
740 * we may be attempting to open an image file that has the rbd options
741 * specified in the older format consisting of all key/value pairs
742 * encoded in the filename. Go ahead and attempt to parse the
743 * filename, and see if we can pull out the required options. */
744 r
= qemu_rbd_attempt_legacy_options(options
, &opts
, &keypairs
);
746 /* Propagate the original error, not the legacy parsing fallback
747 * error, as the latter was just a best-effort attempt. */
748 error_propagate(errp
, local_err
);
751 /* Take care whenever deciding to actually deprecate; once this ability
752 * is removed, we will not be able to open any images with legacy-styled
753 * backing image strings. */
754 warn_report("RBD options encoded in the filename as keyvalue pairs "
758 /* Remove the processed options from the QDict (the visitor processes
759 * _all_ options in the QDict) */
760 while ((e
= qdict_first(options
))) {
761 qdict_del(options
, e
->key
);
764 r
= qemu_rbd_connect(&s
->cluster
, &s
->io_ctx
, opts
,
765 !(flags
& BDRV_O_NOCACHE
), keypairs
, secretid
, errp
);
770 s
->snap
= g_strdup(opts
->snapshot
);
771 s
->image_name
= g_strdup(opts
->image
);
773 /* rbd_open is always r/w */
774 r
= rbd_open(s
->io_ctx
, s
->image_name
, &s
->image
, s
->snap
);
776 error_setg_errno(errp
, -r
, "error reading header from %s",
781 /* If we are using an rbd snapshot, we must be r/o, otherwise
783 if (s
->snap
!= NULL
) {
784 r
= bdrv_apply_auto_read_only(bs
, "rbd snapshots are read-only", errp
);
795 rados_ioctx_destroy(s
->io_ctx
);
797 g_free(s
->image_name
);
798 rados_shutdown(s
->cluster
);
800 qapi_free_BlockdevOptionsRbd(opts
);
807 /* Since RBD is currently always opened R/W via the API,
808 * we just need to check if we are using a snapshot or not, in
809 * order to determine if we will allow it to be R/W */
810 static int qemu_rbd_reopen_prepare(BDRVReopenState
*state
,
811 BlockReopenQueue
*queue
, Error
**errp
)
813 BDRVRBDState
*s
= state
->bs
->opaque
;
816 if (s
->snap
&& state
->flags
& BDRV_O_RDWR
) {
818 "Cannot change node '%s' to r/w when using RBD snapshot",
819 bdrv_get_device_or_node_name(state
->bs
));
826 static void qemu_rbd_close(BlockDriverState
*bs
)
828 BDRVRBDState
*s
= bs
->opaque
;
831 rados_ioctx_destroy(s
->io_ctx
);
833 g_free(s
->image_name
);
834 rados_shutdown(s
->cluster
);
837 static const AIOCBInfo rbd_aiocb_info
= {
838 .aiocb_size
= sizeof(RBDAIOCB
),
841 static void rbd_finish_bh(void *opaque
)
843 RADOSCB
*rcb
= opaque
;
844 qemu_rbd_complete_aio(rcb
);
848 * This is the callback function for rbd_aio_read and _write
850 * Note: this function is being called from a non qemu thread so
851 * we need to be careful about what we do here. Generally we only
852 * schedule a BH, and do the rest of the io completion handling
853 * from rbd_finish_bh() which runs in a qemu context.
855 static void rbd_finish_aiocb(rbd_completion_t c
, RADOSCB
*rcb
)
857 RBDAIOCB
*acb
= rcb
->acb
;
859 rcb
->ret
= rbd_aio_get_return_value(c
);
862 aio_bh_schedule_oneshot(bdrv_get_aio_context(acb
->common
.bs
),
866 static int rbd_aio_discard_wrapper(rbd_image_t image
,
869 rbd_completion_t comp
)
871 #ifdef LIBRBD_SUPPORTS_DISCARD
872 return rbd_aio_discard(image
, off
, len
, comp
);
878 static int rbd_aio_flush_wrapper(rbd_image_t image
,
879 rbd_completion_t comp
)
881 #ifdef LIBRBD_SUPPORTS_AIO_FLUSH
882 return rbd_aio_flush(image
, comp
);
888 static BlockAIOCB
*rbd_start_aio(BlockDriverState
*bs
,
892 BlockCompletionFunc
*cb
,
901 BDRVRBDState
*s
= bs
->opaque
;
903 acb
= qemu_aio_get(&rbd_aiocb_info
, bs
, cb
, opaque
);
906 assert(!qiov
|| qiov
->size
== size
);
908 rcb
= g_new(RADOSCB
, 1);
910 if (!LIBRBD_USE_IOVEC
) {
911 if (cmd
== RBD_AIO_DISCARD
|| cmd
== RBD_AIO_FLUSH
) {
914 acb
->bounce
= qemu_try_blockalign(bs
, qiov
->size
);
915 if (acb
->bounce
== NULL
) {
919 if (cmd
== RBD_AIO_WRITE
) {
920 qemu_iovec_to_buf(acb
->qiov
, 0, acb
->bounce
, qiov
->size
);
922 rcb
->buf
= acb
->bounce
;
932 r
= rbd_aio_create_completion(rcb
, (rbd_callback_t
) rbd_finish_aiocb
, &c
);
939 #ifdef LIBRBD_SUPPORTS_IOVEC
940 r
= rbd_aio_writev(s
->image
, qiov
->iov
, qiov
->niov
, off
, c
);
942 r
= rbd_aio_write(s
->image
, off
, size
, rcb
->buf
, c
);
946 #ifdef LIBRBD_SUPPORTS_IOVEC
947 r
= rbd_aio_readv(s
->image
, qiov
->iov
, qiov
->niov
, off
, c
);
949 r
= rbd_aio_read(s
->image
, off
, size
, rcb
->buf
, c
);
952 case RBD_AIO_DISCARD
:
953 r
= rbd_aio_discard_wrapper(s
->image
, off
, size
, c
);
956 r
= rbd_aio_flush_wrapper(s
->image
, c
);
963 goto failed_completion
;
971 if (!LIBRBD_USE_IOVEC
) {
972 qemu_vfree(acb
->bounce
);
979 static BlockAIOCB
*qemu_rbd_aio_preadv(BlockDriverState
*bs
,
980 uint64_t offset
, uint64_t bytes
,
981 QEMUIOVector
*qiov
, int flags
,
982 BlockCompletionFunc
*cb
,
985 return rbd_start_aio(bs
, offset
, qiov
, bytes
, cb
, opaque
,
989 static BlockAIOCB
*qemu_rbd_aio_pwritev(BlockDriverState
*bs
,
990 uint64_t offset
, uint64_t bytes
,
991 QEMUIOVector
*qiov
, int flags
,
992 BlockCompletionFunc
*cb
,
995 return rbd_start_aio(bs
, offset
, qiov
, bytes
, cb
, opaque
,
999 #ifdef LIBRBD_SUPPORTS_AIO_FLUSH
1000 static BlockAIOCB
*qemu_rbd_aio_flush(BlockDriverState
*bs
,
1001 BlockCompletionFunc
*cb
,
1004 return rbd_start_aio(bs
, 0, NULL
, 0, cb
, opaque
, RBD_AIO_FLUSH
);
1009 static int qemu_rbd_co_flush(BlockDriverState
*bs
)
1011 #if LIBRBD_VERSION_CODE >= LIBRBD_VERSION(0, 1, 1)
1012 /* rbd_flush added in 0.1.1 */
1013 BDRVRBDState
*s
= bs
->opaque
;
1014 return rbd_flush(s
->image
);
1021 static int qemu_rbd_getinfo(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
1023 BDRVRBDState
*s
= bs
->opaque
;
1024 rbd_image_info_t info
;
1027 r
= rbd_stat(s
->image
, &info
, sizeof(info
));
1032 bdi
->cluster_size
= info
.obj_size
;
1036 static int64_t qemu_rbd_getlength(BlockDriverState
*bs
)
1038 BDRVRBDState
*s
= bs
->opaque
;
1039 rbd_image_info_t info
;
1042 r
= rbd_stat(s
->image
, &info
, sizeof(info
));
1050 static int coroutine_fn
qemu_rbd_co_truncate(BlockDriverState
*bs
,
1052 PreallocMode prealloc
,
1055 BDRVRBDState
*s
= bs
->opaque
;
1058 if (prealloc
!= PREALLOC_MODE_OFF
) {
1059 error_setg(errp
, "Unsupported preallocation mode '%s'",
1060 PreallocMode_str(prealloc
));
1064 r
= rbd_resize(s
->image
, offset
);
1066 error_setg_errno(errp
, -r
, "Failed to resize file");
1073 static int qemu_rbd_snap_create(BlockDriverState
*bs
,
1074 QEMUSnapshotInfo
*sn_info
)
1076 BDRVRBDState
*s
= bs
->opaque
;
1079 if (sn_info
->name
[0] == '\0') {
1080 return -EINVAL
; /* we need a name for rbd snapshots */
1084 * rbd snapshots are using the name as the user controlled unique identifier
1085 * we can't use the rbd snapid for that purpose, as it can't be set
1087 if (sn_info
->id_str
[0] != '\0' &&
1088 strcmp(sn_info
->id_str
, sn_info
->name
) != 0) {
1092 if (strlen(sn_info
->name
) >= sizeof(sn_info
->id_str
)) {
1096 r
= rbd_snap_create(s
->image
, sn_info
->name
);
1098 error_report("failed to create snap: %s", strerror(-r
));
1105 static int qemu_rbd_snap_remove(BlockDriverState
*bs
,
1106 const char *snapshot_id
,
1107 const char *snapshot_name
,
1110 BDRVRBDState
*s
= bs
->opaque
;
1113 if (!snapshot_name
) {
1114 error_setg(errp
, "rbd need a valid snapshot name");
1118 /* If snapshot_id is specified, it must be equal to name, see
1119 qemu_rbd_snap_list() */
1120 if (snapshot_id
&& strcmp(snapshot_id
, snapshot_name
)) {
1122 "rbd do not support snapshot id, it should be NULL or "
1123 "equal to snapshot name");
1127 r
= rbd_snap_remove(s
->image
, snapshot_name
);
1129 error_setg_errno(errp
, -r
, "Failed to remove the snapshot");
1134 static int qemu_rbd_snap_rollback(BlockDriverState
*bs
,
1135 const char *snapshot_name
)
1137 BDRVRBDState
*s
= bs
->opaque
;
1139 return rbd_snap_rollback(s
->image
, snapshot_name
);
1142 static int qemu_rbd_snap_list(BlockDriverState
*bs
,
1143 QEMUSnapshotInfo
**psn_tab
)
1145 BDRVRBDState
*s
= bs
->opaque
;
1146 QEMUSnapshotInfo
*sn_info
, *sn_tab
= NULL
;
1148 rbd_snap_info_t
*snaps
;
1149 int max_snaps
= RBD_MAX_SNAPS
;
1152 snaps
= g_new(rbd_snap_info_t
, max_snaps
);
1153 snap_count
= rbd_snap_list(s
->image
, snaps
, &max_snaps
);
1154 if (snap_count
<= 0) {
1157 } while (snap_count
== -ERANGE
);
1159 if (snap_count
<= 0) {
1163 sn_tab
= g_new0(QEMUSnapshotInfo
, snap_count
);
1165 for (i
= 0; i
< snap_count
; i
++) {
1166 const char *snap_name
= snaps
[i
].name
;
1168 sn_info
= sn_tab
+ i
;
1169 pstrcpy(sn_info
->id_str
, sizeof(sn_info
->id_str
), snap_name
);
1170 pstrcpy(sn_info
->name
, sizeof(sn_info
->name
), snap_name
);
1172 sn_info
->vm_state_size
= snaps
[i
].size
;
1173 sn_info
->date_sec
= 0;
1174 sn_info
->date_nsec
= 0;
1175 sn_info
->vm_clock_nsec
= 0;
1177 rbd_snap_list_end(snaps
);
1185 #ifdef LIBRBD_SUPPORTS_DISCARD
1186 static BlockAIOCB
*qemu_rbd_aio_pdiscard(BlockDriverState
*bs
,
1189 BlockCompletionFunc
*cb
,
1192 return rbd_start_aio(bs
, offset
, NULL
, bytes
, cb
, opaque
,
1197 #ifdef LIBRBD_SUPPORTS_INVALIDATE
1198 static void coroutine_fn
qemu_rbd_co_invalidate_cache(BlockDriverState
*bs
,
1201 BDRVRBDState
*s
= bs
->opaque
;
1202 int r
= rbd_invalidate_cache(s
->image
);
1204 error_setg_errno(errp
, -r
, "Failed to invalidate the cache");
1209 static QemuOptsList qemu_rbd_create_opts
= {
1210 .name
= "rbd-create-opts",
1211 .head
= QTAILQ_HEAD_INITIALIZER(qemu_rbd_create_opts
.head
),
1214 .name
= BLOCK_OPT_SIZE
,
1215 .type
= QEMU_OPT_SIZE
,
1216 .help
= "Virtual disk size"
1219 .name
= BLOCK_OPT_CLUSTER_SIZE
,
1220 .type
= QEMU_OPT_SIZE
,
1221 .help
= "RBD object size"
1224 .name
= "password-secret",
1225 .type
= QEMU_OPT_STRING
,
1226 .help
= "ID of secret providing the password",
1228 { /* end of list */ }
1232 static const char *const qemu_rbd_strong_runtime_opts
[] = {
1244 static BlockDriver bdrv_rbd
= {
1245 .format_name
= "rbd",
1246 .instance_size
= sizeof(BDRVRBDState
),
1247 .bdrv_parse_filename
= qemu_rbd_parse_filename
,
1248 .bdrv_refresh_limits
= qemu_rbd_refresh_limits
,
1249 .bdrv_file_open
= qemu_rbd_open
,
1250 .bdrv_close
= qemu_rbd_close
,
1251 .bdrv_reopen_prepare
= qemu_rbd_reopen_prepare
,
1252 .bdrv_co_create
= qemu_rbd_co_create
,
1253 .bdrv_co_create_opts
= qemu_rbd_co_create_opts
,
1254 .bdrv_has_zero_init
= bdrv_has_zero_init_1
,
1255 .bdrv_get_info
= qemu_rbd_getinfo
,
1256 .create_opts
= &qemu_rbd_create_opts
,
1257 .bdrv_getlength
= qemu_rbd_getlength
,
1258 .bdrv_co_truncate
= qemu_rbd_co_truncate
,
1259 .protocol_name
= "rbd",
1261 .bdrv_aio_preadv
= qemu_rbd_aio_preadv
,
1262 .bdrv_aio_pwritev
= qemu_rbd_aio_pwritev
,
1264 #ifdef LIBRBD_SUPPORTS_AIO_FLUSH
1265 .bdrv_aio_flush
= qemu_rbd_aio_flush
,
1267 .bdrv_co_flush_to_disk
= qemu_rbd_co_flush
,
1270 #ifdef LIBRBD_SUPPORTS_DISCARD
1271 .bdrv_aio_pdiscard
= qemu_rbd_aio_pdiscard
,
1274 .bdrv_snapshot_create
= qemu_rbd_snap_create
,
1275 .bdrv_snapshot_delete
= qemu_rbd_snap_remove
,
1276 .bdrv_snapshot_list
= qemu_rbd_snap_list
,
1277 .bdrv_snapshot_goto
= qemu_rbd_snap_rollback
,
1278 #ifdef LIBRBD_SUPPORTS_INVALIDATE
1279 .bdrv_co_invalidate_cache
= qemu_rbd_co_invalidate_cache
,
1282 .strong_runtime_opts
= qemu_rbd_strong_runtime_opts
,
1285 static void bdrv_rbd_init(void)
1287 bdrv_register(&bdrv_rbd
);
1290 block_init(bdrv_rbd_init
);