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/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 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 static QemuOptsList runtime_opts
= {
346 .head
= QTAILQ_HEAD_INITIALIZER(runtime_opts
.head
),
350 .type
= QEMU_OPT_STRING
,
351 .help
= "Rados pool name",
355 .type
= QEMU_OPT_STRING
,
356 .help
= "Rados namespace name in the pool",
360 .type
= QEMU_OPT_STRING
,
361 .help
= "Image name in the pool",
365 .type
= QEMU_OPT_STRING
,
366 .help
= "Rados config file location",
370 .type
= QEMU_OPT_STRING
,
371 .help
= "Ceph snapshot name",
374 /* maps to 'id' in rados_create() */
376 .type
= QEMU_OPT_STRING
,
377 .help
= "Rados id name",
380 * server.* extracted manually, see qemu_rbd_mon_host()
382 { /* end of list */ }
386 /* FIXME Deprecate and remove keypairs or make it available in QMP. */
387 static int qemu_rbd_do_create(BlockdevCreateOptions
*options
,
388 const char *keypairs
, const char *password_secret
,
391 BlockdevCreateOptionsRbd
*opts
= &options
->u
.rbd
;
393 rados_ioctx_t io_ctx
;
397 assert(options
->driver
== BLOCKDEV_DRIVER_RBD
);
398 if (opts
->location
->has_snapshot
) {
399 error_setg(errp
, "Can't use snapshot name for image creation");
403 if (opts
->has_cluster_size
) {
404 int64_t objsize
= opts
->cluster_size
;
405 if ((objsize
- 1) & objsize
) { /* not a power of 2? */
406 error_setg(errp
, "obj size needs to be power of 2");
409 if (objsize
< 4096) {
410 error_setg(errp
, "obj size too small");
413 obj_order
= ctz32(objsize
);
416 ret
= qemu_rbd_connect(&cluster
, &io_ctx
, opts
->location
, false, keypairs
,
417 password_secret
, errp
);
422 ret
= rbd_create(io_ctx
, opts
->location
->image
, opts
->size
, &obj_order
);
424 error_setg_errno(errp
, -ret
, "error rbd create");
430 rados_ioctx_destroy(io_ctx
);
431 rados_shutdown(cluster
);
435 static int qemu_rbd_co_create(BlockdevCreateOptions
*options
, Error
**errp
)
437 return qemu_rbd_do_create(options
, NULL
, NULL
, errp
);
440 static int coroutine_fn
qemu_rbd_co_create_opts(const char *filename
,
444 BlockdevCreateOptions
*create_options
;
445 BlockdevCreateOptionsRbd
*rbd_opts
;
446 BlockdevOptionsRbd
*loc
;
447 Error
*local_err
= NULL
;
448 const char *keypairs
, *password_secret
;
449 QDict
*options
= NULL
;
452 create_options
= g_new0(BlockdevCreateOptions
, 1);
453 create_options
->driver
= BLOCKDEV_DRIVER_RBD
;
454 rbd_opts
= &create_options
->u
.rbd
;
456 rbd_opts
->location
= g_new0(BlockdevOptionsRbd
, 1);
458 password_secret
= qemu_opt_get(opts
, "password-secret");
460 /* Read out options */
461 rbd_opts
->size
= ROUND_UP(qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0),
463 rbd_opts
->cluster_size
= qemu_opt_get_size_del(opts
,
464 BLOCK_OPT_CLUSTER_SIZE
, 0);
465 rbd_opts
->has_cluster_size
= (rbd_opts
->cluster_size
!= 0);
467 options
= qdict_new();
468 qemu_rbd_parse_filename(filename
, options
, &local_err
);
471 error_propagate(errp
, local_err
);
476 * Caution: while qdict_get_try_str() is fine, getting non-string
477 * types would require more care. When @options come from -blockdev
478 * or blockdev_add, its members are typed according to the QAPI
479 * schema, but when they come from -drive, they're all QString.
481 loc
= rbd_opts
->location
;
482 loc
->pool
= g_strdup(qdict_get_try_str(options
, "pool"));
483 loc
->conf
= g_strdup(qdict_get_try_str(options
, "conf"));
484 loc
->has_conf
= !!loc
->conf
;
485 loc
->user
= g_strdup(qdict_get_try_str(options
, "user"));
486 loc
->has_user
= !!loc
->user
;
487 loc
->q_namespace
= g_strdup(qdict_get_try_str(options
, "namespace"));
488 loc
->image
= g_strdup(qdict_get_try_str(options
, "image"));
489 keypairs
= qdict_get_try_str(options
, "=keyvalue-pairs");
491 ret
= qemu_rbd_do_create(create_options
, keypairs
, password_secret
, errp
);
497 qobject_unref(options
);
498 qapi_free_BlockdevCreateOptions(create_options
);
503 * This aio completion is being called from rbd_finish_bh() and runs in qemu
506 static void qemu_rbd_complete_aio(RADOSCB
*rcb
)
508 RBDAIOCB
*acb
= rcb
->acb
;
513 if (acb
->cmd
!= RBD_AIO_READ
) {
517 } else if (!acb
->error
) {
518 acb
->ret
= rcb
->size
;
522 qemu_rbd_memset(rcb
, 0);
525 } else if (r
< rcb
->size
) {
526 qemu_rbd_memset(rcb
, r
);
528 acb
->ret
= rcb
->size
;
530 } else if (!acb
->error
) {
537 if (!LIBRBD_USE_IOVEC
) {
538 if (acb
->cmd
== RBD_AIO_READ
) {
539 qemu_iovec_from_buf(acb
->qiov
, 0, acb
->bounce
, acb
->qiov
->size
);
541 qemu_vfree(acb
->bounce
);
544 acb
->common
.cb(acb
->common
.opaque
, (acb
->ret
> 0 ? 0 : acb
->ret
));
549 static char *qemu_rbd_mon_host(BlockdevOptionsRbd
*opts
, Error
**errp
)
552 const char *host
, *port
;
554 InetSocketAddressBaseList
*p
;
557 if (!opts
->has_server
) {
561 for (cnt
= 0, p
= opts
->server
; p
; p
= p
->next
) {
565 vals
= g_new(const char *, cnt
+ 1);
567 for (i
= 0, p
= opts
->server
; p
; p
= p
->next
, i
++) {
568 host
= p
->value
->host
;
569 port
= p
->value
->port
;
571 if (strchr(host
, ':')) {
572 vals
[i
] = g_strdup_printf("[%s]:%s", host
, port
);
574 vals
[i
] = g_strdup_printf("%s:%s", host
, port
);
579 rados_str
= i
? g_strjoinv(";", (char **)vals
) : NULL
;
580 g_strfreev((char **)vals
);
584 static int qemu_rbd_connect(rados_t
*cluster
, rados_ioctx_t
*io_ctx
,
585 BlockdevOptionsRbd
*opts
, bool cache
,
586 const char *keypairs
, const char *secretid
,
589 char *mon_host
= NULL
;
590 Error
*local_err
= NULL
;
594 if (opts
->key_secret
) {
596 "Legacy 'password-secret' clashes with 'key-secret'");
599 opts
->key_secret
= g_strdup(secretid
);
600 opts
->has_key_secret
= true;
603 mon_host
= qemu_rbd_mon_host(opts
, &local_err
);
605 error_propagate(errp
, local_err
);
610 r
= rados_create(cluster
, opts
->user
);
612 error_setg_errno(errp
, -r
, "error initializing");
616 /* try default location when conf=NULL, but ignore failure */
617 r
= rados_conf_read_file(*cluster
, opts
->conf
);
618 if (opts
->has_conf
&& r
< 0) {
619 error_setg_errno(errp
, -r
, "error reading conf file %s", opts
->conf
);
620 goto failed_shutdown
;
623 r
= qemu_rbd_set_keypairs(*cluster
, keypairs
, errp
);
625 goto failed_shutdown
;
629 r
= rados_conf_set(*cluster
, "mon_host", mon_host
);
631 goto failed_shutdown
;
635 r
= qemu_rbd_set_auth(*cluster
, opts
, errp
);
637 goto failed_shutdown
;
641 * Fallback to more conservative semantics if setting cache
642 * options fails. Ignore errors from setting rbd_cache because the
643 * only possible error is that the option does not exist, and
644 * librbd defaults to no caching. If write through caching cannot
645 * be set up, fall back to no caching.
648 rados_conf_set(*cluster
, "rbd_cache", "true");
650 rados_conf_set(*cluster
, "rbd_cache", "false");
653 r
= rados_connect(*cluster
);
655 error_setg_errno(errp
, -r
, "error connecting");
656 goto failed_shutdown
;
659 r
= rados_ioctx_create(*cluster
, opts
->pool
, io_ctx
);
661 error_setg_errno(errp
, -r
, "error opening pool %s", opts
->pool
);
662 goto failed_shutdown
;
665 * Set the namespace after opening the io context on the pool,
666 * if nspace == NULL or if nspace == "", it is just as we did nothing
668 rados_ioctx_set_namespace(*io_ctx
, opts
->q_namespace
);
673 rados_shutdown(*cluster
);
679 static int qemu_rbd_convert_options(QDict
*options
, BlockdevOptionsRbd
**opts
,
683 Error
*local_err
= NULL
;
685 /* Convert the remaining options into a QAPI object */
686 v
= qobject_input_visitor_new_flat_confused(options
, errp
);
691 visit_type_BlockdevOptionsRbd(v
, NULL
, opts
, &local_err
);
695 error_propagate(errp
, local_err
);
702 static int qemu_rbd_attempt_legacy_options(QDict
*options
,
703 BlockdevOptionsRbd
**opts
,
709 filename
= g_strdup(qdict_get_try_str(options
, "filename"));
713 qdict_del(options
, "filename");
715 qemu_rbd_parse_filename(filename
, options
, NULL
);
717 /* keypairs freed by caller */
718 *keypairs
= g_strdup(qdict_get_try_str(options
, "=keyvalue-pairs"));
720 qdict_del(options
, "=keyvalue-pairs");
723 r
= qemu_rbd_convert_options(options
, opts
, NULL
);
729 static int qemu_rbd_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
732 BDRVRBDState
*s
= bs
->opaque
;
733 BlockdevOptionsRbd
*opts
= NULL
;
735 Error
*local_err
= NULL
;
736 char *keypairs
, *secretid
;
739 keypairs
= g_strdup(qdict_get_try_str(options
, "=keyvalue-pairs"));
741 qdict_del(options
, "=keyvalue-pairs");
744 secretid
= g_strdup(qdict_get_try_str(options
, "password-secret"));
746 qdict_del(options
, "password-secret");
749 r
= qemu_rbd_convert_options(options
, &opts
, &local_err
);
751 /* If keypairs are present, that means some options are present in
752 * the modern option format. Don't attempt to parse legacy option
753 * formats, as we won't support mixed usage. */
755 error_propagate(errp
, local_err
);
759 /* If the initial attempt to convert and process the options failed,
760 * we may be attempting to open an image file that has the rbd options
761 * specified in the older format consisting of all key/value pairs
762 * encoded in the filename. Go ahead and attempt to parse the
763 * filename, and see if we can pull out the required options. */
764 r
= qemu_rbd_attempt_legacy_options(options
, &opts
, &keypairs
);
766 /* Propagate the original error, not the legacy parsing fallback
767 * error, as the latter was just a best-effort attempt. */
768 error_propagate(errp
, local_err
);
771 /* Take care whenever deciding to actually deprecate; once this ability
772 * is removed, we will not be able to open any images with legacy-styled
773 * backing image strings. */
774 warn_report("RBD options encoded in the filename as keyvalue pairs "
778 /* Remove the processed options from the QDict (the visitor processes
779 * _all_ options in the QDict) */
780 while ((e
= qdict_first(options
))) {
781 qdict_del(options
, e
->key
);
784 r
= qemu_rbd_connect(&s
->cluster
, &s
->io_ctx
, opts
,
785 !(flags
& BDRV_O_NOCACHE
), keypairs
, secretid
, errp
);
790 s
->snap
= g_strdup(opts
->snapshot
);
791 s
->image_name
= g_strdup(opts
->image
);
793 /* rbd_open is always r/w */
794 r
= rbd_open(s
->io_ctx
, s
->image_name
, &s
->image
, s
->snap
);
796 error_setg_errno(errp
, -r
, "error reading header from %s",
801 r
= rbd_get_size(s
->image
, &s
->image_size
);
803 error_setg_errno(errp
, -r
, "error getting image size from %s",
809 /* If we are using an rbd snapshot, we must be r/o, otherwise
811 if (s
->snap
!= NULL
) {
812 r
= bdrv_apply_auto_read_only(bs
, "rbd snapshots are read-only", errp
);
823 rados_ioctx_destroy(s
->io_ctx
);
825 g_free(s
->image_name
);
826 rados_shutdown(s
->cluster
);
828 qapi_free_BlockdevOptionsRbd(opts
);
835 /* Since RBD is currently always opened R/W via the API,
836 * we just need to check if we are using a snapshot or not, in
837 * order to determine if we will allow it to be R/W */
838 static int qemu_rbd_reopen_prepare(BDRVReopenState
*state
,
839 BlockReopenQueue
*queue
, Error
**errp
)
841 BDRVRBDState
*s
= state
->bs
->opaque
;
844 if (s
->snap
&& state
->flags
& BDRV_O_RDWR
) {
846 "Cannot change node '%s' to r/w when using RBD snapshot",
847 bdrv_get_device_or_node_name(state
->bs
));
854 static void qemu_rbd_close(BlockDriverState
*bs
)
856 BDRVRBDState
*s
= bs
->opaque
;
859 rados_ioctx_destroy(s
->io_ctx
);
861 g_free(s
->image_name
);
862 rados_shutdown(s
->cluster
);
865 /* Resize the RBD image and update the 'image_size' with the current size */
866 static int qemu_rbd_resize(BlockDriverState
*bs
, uint64_t size
)
868 BDRVRBDState
*s
= bs
->opaque
;
871 r
= rbd_resize(s
->image
, size
);
876 s
->image_size
= size
;
881 static const AIOCBInfo rbd_aiocb_info
= {
882 .aiocb_size
= sizeof(RBDAIOCB
),
885 static void rbd_finish_bh(void *opaque
)
887 RADOSCB
*rcb
= opaque
;
888 qemu_rbd_complete_aio(rcb
);
892 * This is the callback function for rbd_aio_read and _write
894 * Note: this function is being called from a non qemu thread so
895 * we need to be careful about what we do here. Generally we only
896 * schedule a BH, and do the rest of the io completion handling
897 * from rbd_finish_bh() which runs in a qemu context.
899 static void rbd_finish_aiocb(rbd_completion_t c
, RADOSCB
*rcb
)
901 RBDAIOCB
*acb
= rcb
->acb
;
903 rcb
->ret
= rbd_aio_get_return_value(c
);
906 replay_bh_schedule_oneshot_event(bdrv_get_aio_context(acb
->common
.bs
),
910 static int rbd_aio_discard_wrapper(rbd_image_t image
,
913 rbd_completion_t comp
)
915 #ifdef LIBRBD_SUPPORTS_DISCARD
916 return rbd_aio_discard(image
, off
, len
, comp
);
922 static int rbd_aio_flush_wrapper(rbd_image_t image
,
923 rbd_completion_t comp
)
925 #ifdef LIBRBD_SUPPORTS_AIO_FLUSH
926 return rbd_aio_flush(image
, comp
);
932 static BlockAIOCB
*rbd_start_aio(BlockDriverState
*bs
,
936 BlockCompletionFunc
*cb
,
945 BDRVRBDState
*s
= bs
->opaque
;
947 acb
= qemu_aio_get(&rbd_aiocb_info
, bs
, cb
, opaque
);
950 assert(!qiov
|| qiov
->size
== size
);
952 rcb
= g_new(RADOSCB
, 1);
954 if (!LIBRBD_USE_IOVEC
) {
955 if (cmd
== RBD_AIO_DISCARD
|| cmd
== RBD_AIO_FLUSH
) {
958 acb
->bounce
= qemu_try_blockalign(bs
, qiov
->size
);
959 if (acb
->bounce
== NULL
) {
963 if (cmd
== RBD_AIO_WRITE
) {
964 qemu_iovec_to_buf(acb
->qiov
, 0, acb
->bounce
, qiov
->size
);
966 rcb
->buf
= acb
->bounce
;
976 r
= rbd_aio_create_completion(rcb
, (rbd_callback_t
) rbd_finish_aiocb
, &c
);
982 case RBD_AIO_WRITE
: {
984 * RBD APIs don't allow us to write more than actual size, so in order
985 * to support growing images, we resize the image before write
986 * operations that exceed the current size.
988 if (off
+ size
> s
->image_size
) {
989 r
= qemu_rbd_resize(bs
, off
+ size
);
991 goto failed_completion
;
994 #ifdef LIBRBD_SUPPORTS_IOVEC
995 r
= rbd_aio_writev(s
->image
, qiov
->iov
, qiov
->niov
, off
, c
);
997 r
= rbd_aio_write(s
->image
, off
, size
, rcb
->buf
, c
);
1002 #ifdef LIBRBD_SUPPORTS_IOVEC
1003 r
= rbd_aio_readv(s
->image
, qiov
->iov
, qiov
->niov
, off
, c
);
1005 r
= rbd_aio_read(s
->image
, off
, size
, rcb
->buf
, c
);
1008 case RBD_AIO_DISCARD
:
1009 r
= rbd_aio_discard_wrapper(s
->image
, off
, size
, c
);
1012 r
= rbd_aio_flush_wrapper(s
->image
, c
);
1019 goto failed_completion
;
1021 return &acb
->common
;
1027 if (!LIBRBD_USE_IOVEC
) {
1028 qemu_vfree(acb
->bounce
);
1031 qemu_aio_unref(acb
);
1035 static BlockAIOCB
*qemu_rbd_aio_preadv(BlockDriverState
*bs
,
1036 uint64_t offset
, uint64_t bytes
,
1037 QEMUIOVector
*qiov
, int flags
,
1038 BlockCompletionFunc
*cb
,
1041 return rbd_start_aio(bs
, offset
, qiov
, bytes
, cb
, opaque
,
1045 static BlockAIOCB
*qemu_rbd_aio_pwritev(BlockDriverState
*bs
,
1046 uint64_t offset
, uint64_t bytes
,
1047 QEMUIOVector
*qiov
, int flags
,
1048 BlockCompletionFunc
*cb
,
1051 return rbd_start_aio(bs
, offset
, qiov
, bytes
, cb
, opaque
,
1055 #ifdef LIBRBD_SUPPORTS_AIO_FLUSH
1056 static BlockAIOCB
*qemu_rbd_aio_flush(BlockDriverState
*bs
,
1057 BlockCompletionFunc
*cb
,
1060 return rbd_start_aio(bs
, 0, NULL
, 0, cb
, opaque
, RBD_AIO_FLUSH
);
1065 static int qemu_rbd_co_flush(BlockDriverState
*bs
)
1067 #if LIBRBD_VERSION_CODE >= LIBRBD_VERSION(0, 1, 1)
1068 /* rbd_flush added in 0.1.1 */
1069 BDRVRBDState
*s
= bs
->opaque
;
1070 return rbd_flush(s
->image
);
1077 static int qemu_rbd_getinfo(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
1079 BDRVRBDState
*s
= bs
->opaque
;
1080 rbd_image_info_t info
;
1083 r
= rbd_stat(s
->image
, &info
, sizeof(info
));
1088 bdi
->cluster_size
= info
.obj_size
;
1092 static int64_t qemu_rbd_getlength(BlockDriverState
*bs
)
1094 BDRVRBDState
*s
= bs
->opaque
;
1095 rbd_image_info_t info
;
1098 r
= rbd_stat(s
->image
, &info
, sizeof(info
));
1106 static int coroutine_fn
qemu_rbd_co_truncate(BlockDriverState
*bs
,
1109 PreallocMode prealloc
,
1114 if (prealloc
!= PREALLOC_MODE_OFF
) {
1115 error_setg(errp
, "Unsupported preallocation mode '%s'",
1116 PreallocMode_str(prealloc
));
1120 r
= qemu_rbd_resize(bs
, offset
);
1122 error_setg_errno(errp
, -r
, "Failed to resize file");
1129 static int qemu_rbd_snap_create(BlockDriverState
*bs
,
1130 QEMUSnapshotInfo
*sn_info
)
1132 BDRVRBDState
*s
= bs
->opaque
;
1135 if (sn_info
->name
[0] == '\0') {
1136 return -EINVAL
; /* we need a name for rbd snapshots */
1140 * rbd snapshots are using the name as the user controlled unique identifier
1141 * we can't use the rbd snapid for that purpose, as it can't be set
1143 if (sn_info
->id_str
[0] != '\0' &&
1144 strcmp(sn_info
->id_str
, sn_info
->name
) != 0) {
1148 if (strlen(sn_info
->name
) >= sizeof(sn_info
->id_str
)) {
1152 r
= rbd_snap_create(s
->image
, sn_info
->name
);
1154 error_report("failed to create snap: %s", strerror(-r
));
1161 static int qemu_rbd_snap_remove(BlockDriverState
*bs
,
1162 const char *snapshot_id
,
1163 const char *snapshot_name
,
1166 BDRVRBDState
*s
= bs
->opaque
;
1169 if (!snapshot_name
) {
1170 error_setg(errp
, "rbd need a valid snapshot name");
1174 /* If snapshot_id is specified, it must be equal to name, see
1175 qemu_rbd_snap_list() */
1176 if (snapshot_id
&& strcmp(snapshot_id
, snapshot_name
)) {
1178 "rbd do not support snapshot id, it should be NULL or "
1179 "equal to snapshot name");
1183 r
= rbd_snap_remove(s
->image
, snapshot_name
);
1185 error_setg_errno(errp
, -r
, "Failed to remove the snapshot");
1190 static int qemu_rbd_snap_rollback(BlockDriverState
*bs
,
1191 const char *snapshot_name
)
1193 BDRVRBDState
*s
= bs
->opaque
;
1195 return rbd_snap_rollback(s
->image
, snapshot_name
);
1198 static int qemu_rbd_snap_list(BlockDriverState
*bs
,
1199 QEMUSnapshotInfo
**psn_tab
)
1201 BDRVRBDState
*s
= bs
->opaque
;
1202 QEMUSnapshotInfo
*sn_info
, *sn_tab
= NULL
;
1204 rbd_snap_info_t
*snaps
;
1205 int max_snaps
= RBD_MAX_SNAPS
;
1208 snaps
= g_new(rbd_snap_info_t
, max_snaps
);
1209 snap_count
= rbd_snap_list(s
->image
, snaps
, &max_snaps
);
1210 if (snap_count
<= 0) {
1213 } while (snap_count
== -ERANGE
);
1215 if (snap_count
<= 0) {
1219 sn_tab
= g_new0(QEMUSnapshotInfo
, snap_count
);
1221 for (i
= 0; i
< snap_count
; i
++) {
1222 const char *snap_name
= snaps
[i
].name
;
1224 sn_info
= sn_tab
+ i
;
1225 pstrcpy(sn_info
->id_str
, sizeof(sn_info
->id_str
), snap_name
);
1226 pstrcpy(sn_info
->name
, sizeof(sn_info
->name
), snap_name
);
1228 sn_info
->vm_state_size
= snaps
[i
].size
;
1229 sn_info
->date_sec
= 0;
1230 sn_info
->date_nsec
= 0;
1231 sn_info
->vm_clock_nsec
= 0;
1233 rbd_snap_list_end(snaps
);
1241 #ifdef LIBRBD_SUPPORTS_DISCARD
1242 static BlockAIOCB
*qemu_rbd_aio_pdiscard(BlockDriverState
*bs
,
1245 BlockCompletionFunc
*cb
,
1248 return rbd_start_aio(bs
, offset
, NULL
, bytes
, cb
, opaque
,
1253 #ifdef LIBRBD_SUPPORTS_INVALIDATE
1254 static void coroutine_fn
qemu_rbd_co_invalidate_cache(BlockDriverState
*bs
,
1257 BDRVRBDState
*s
= bs
->opaque
;
1258 int r
= rbd_invalidate_cache(s
->image
);
1260 error_setg_errno(errp
, -r
, "Failed to invalidate the cache");
1265 static QemuOptsList qemu_rbd_create_opts
= {
1266 .name
= "rbd-create-opts",
1267 .head
= QTAILQ_HEAD_INITIALIZER(qemu_rbd_create_opts
.head
),
1270 .name
= BLOCK_OPT_SIZE
,
1271 .type
= QEMU_OPT_SIZE
,
1272 .help
= "Virtual disk size"
1275 .name
= BLOCK_OPT_CLUSTER_SIZE
,
1276 .type
= QEMU_OPT_SIZE
,
1277 .help
= "RBD object size"
1280 .name
= "password-secret",
1281 .type
= QEMU_OPT_STRING
,
1282 .help
= "ID of secret providing the password",
1284 { /* end of list */ }
1288 static const char *const qemu_rbd_strong_runtime_opts
[] = {
1300 static BlockDriver bdrv_rbd
= {
1301 .format_name
= "rbd",
1302 .instance_size
= sizeof(BDRVRBDState
),
1303 .bdrv_parse_filename
= qemu_rbd_parse_filename
,
1304 .bdrv_refresh_limits
= qemu_rbd_refresh_limits
,
1305 .bdrv_file_open
= qemu_rbd_open
,
1306 .bdrv_close
= qemu_rbd_close
,
1307 .bdrv_reopen_prepare
= qemu_rbd_reopen_prepare
,
1308 .bdrv_co_create
= qemu_rbd_co_create
,
1309 .bdrv_co_create_opts
= qemu_rbd_co_create_opts
,
1310 .bdrv_has_zero_init
= bdrv_has_zero_init_1
,
1311 .bdrv_has_zero_init_truncate
= bdrv_has_zero_init_1
,
1312 .bdrv_get_info
= qemu_rbd_getinfo
,
1313 .create_opts
= &qemu_rbd_create_opts
,
1314 .bdrv_getlength
= qemu_rbd_getlength
,
1315 .bdrv_co_truncate
= qemu_rbd_co_truncate
,
1316 .protocol_name
= "rbd",
1318 .bdrv_aio_preadv
= qemu_rbd_aio_preadv
,
1319 .bdrv_aio_pwritev
= qemu_rbd_aio_pwritev
,
1321 #ifdef LIBRBD_SUPPORTS_AIO_FLUSH
1322 .bdrv_aio_flush
= qemu_rbd_aio_flush
,
1324 .bdrv_co_flush_to_disk
= qemu_rbd_co_flush
,
1327 #ifdef LIBRBD_SUPPORTS_DISCARD
1328 .bdrv_aio_pdiscard
= qemu_rbd_aio_pdiscard
,
1331 .bdrv_snapshot_create
= qemu_rbd_snap_create
,
1332 .bdrv_snapshot_delete
= qemu_rbd_snap_remove
,
1333 .bdrv_snapshot_list
= qemu_rbd_snap_list
,
1334 .bdrv_snapshot_goto
= qemu_rbd_snap_rollback
,
1335 #ifdef LIBRBD_SUPPORTS_INVALIDATE
1336 .bdrv_co_invalidate_cache
= qemu_rbd_co_invalidate_cache
,
1339 .strong_runtime_opts
= qemu_rbd_strong_runtime_opts
,
1342 static void bdrv_rbd_init(void)
1344 bdrv_register(&bdrv_rbd
);
1347 block_init(bdrv_rbd_init
);