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 #pragma GCC diagnostic ignored "-Wsuggest-attribute=format"
295 static int qemu_rbd_set_keypairs(rados_t cluster
, const char *keypairs_json
,
305 if (!keypairs_json
) {
308 keypairs
= qobject_to(QList
,
309 qobject_from_json(keypairs_json
, &error_abort
));
310 remaining
= qlist_size(keypairs
) / 2;
313 while (remaining
--) {
314 name
= qobject_to(QString
, qlist_pop(keypairs
));
315 value
= qobject_to(QString
, qlist_pop(keypairs
));
316 assert(name
&& value
);
317 key
= qstring_get_str(name
);
319 ret
= rados_conf_set(cluster
, key
, qstring_get_str(value
));
320 qobject_unref(value
);
322 error_setg_errno(errp
, -ret
, "invalid conf option %s", key
);
330 qobject_unref(keypairs
);
334 static void qemu_rbd_memset(RADOSCB
*rcb
, int64_t offs
)
336 if (LIBRBD_USE_IOVEC
) {
337 RBDAIOCB
*acb
= rcb
->acb
;
338 iov_memset(acb
->qiov
->iov
, acb
->qiov
->niov
, offs
, 0,
339 acb
->qiov
->size
- offs
);
341 memset(rcb
->buf
+ offs
, 0, rcb
->size
- offs
);
345 /* FIXME Deprecate and remove keypairs or make it available in QMP. */
346 static int qemu_rbd_do_create(BlockdevCreateOptions
*options
,
347 const char *keypairs
, const char *password_secret
,
350 BlockdevCreateOptionsRbd
*opts
= &options
->u
.rbd
;
352 rados_ioctx_t io_ctx
;
356 assert(options
->driver
== BLOCKDEV_DRIVER_RBD
);
357 if (opts
->location
->has_snapshot
) {
358 error_setg(errp
, "Can't use snapshot name for image creation");
362 if (opts
->has_cluster_size
) {
363 int64_t objsize
= opts
->cluster_size
;
364 if ((objsize
- 1) & objsize
) { /* not a power of 2? */
365 error_setg(errp
, "obj size needs to be power of 2");
368 if (objsize
< 4096) {
369 error_setg(errp
, "obj size too small");
372 obj_order
= ctz32(objsize
);
375 ret
= qemu_rbd_connect(&cluster
, &io_ctx
, opts
->location
, false, keypairs
,
376 password_secret
, errp
);
381 ret
= rbd_create(io_ctx
, opts
->location
->image
, opts
->size
, &obj_order
);
383 error_setg_errno(errp
, -ret
, "error rbd create");
389 rados_ioctx_destroy(io_ctx
);
390 rados_shutdown(cluster
);
394 static int qemu_rbd_co_create(BlockdevCreateOptions
*options
, Error
**errp
)
396 return qemu_rbd_do_create(options
, NULL
, NULL
, errp
);
399 static int coroutine_fn
qemu_rbd_co_create_opts(BlockDriver
*drv
,
400 const char *filename
,
404 BlockdevCreateOptions
*create_options
;
405 BlockdevCreateOptionsRbd
*rbd_opts
;
406 BlockdevOptionsRbd
*loc
;
407 Error
*local_err
= NULL
;
408 const char *keypairs
, *password_secret
;
409 QDict
*options
= NULL
;
412 create_options
= g_new0(BlockdevCreateOptions
, 1);
413 create_options
->driver
= BLOCKDEV_DRIVER_RBD
;
414 rbd_opts
= &create_options
->u
.rbd
;
416 rbd_opts
->location
= g_new0(BlockdevOptionsRbd
, 1);
418 password_secret
= qemu_opt_get(opts
, "password-secret");
420 /* Read out options */
421 rbd_opts
->size
= ROUND_UP(qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0),
423 rbd_opts
->cluster_size
= qemu_opt_get_size_del(opts
,
424 BLOCK_OPT_CLUSTER_SIZE
, 0);
425 rbd_opts
->has_cluster_size
= (rbd_opts
->cluster_size
!= 0);
427 options
= qdict_new();
428 qemu_rbd_parse_filename(filename
, options
, &local_err
);
431 error_propagate(errp
, local_err
);
436 * Caution: while qdict_get_try_str() is fine, getting non-string
437 * types would require more care. When @options come from -blockdev
438 * or blockdev_add, its members are typed according to the QAPI
439 * schema, but when they come from -drive, they're all QString.
441 loc
= rbd_opts
->location
;
442 loc
->pool
= g_strdup(qdict_get_try_str(options
, "pool"));
443 loc
->conf
= g_strdup(qdict_get_try_str(options
, "conf"));
444 loc
->has_conf
= !!loc
->conf
;
445 loc
->user
= g_strdup(qdict_get_try_str(options
, "user"));
446 loc
->has_user
= !!loc
->user
;
447 loc
->q_namespace
= g_strdup(qdict_get_try_str(options
, "namespace"));
448 loc
->image
= g_strdup(qdict_get_try_str(options
, "image"));
449 keypairs
= qdict_get_try_str(options
, "=keyvalue-pairs");
451 ret
= qemu_rbd_do_create(create_options
, keypairs
, password_secret
, errp
);
457 qobject_unref(options
);
458 qapi_free_BlockdevCreateOptions(create_options
);
463 * This aio completion is being called from rbd_finish_bh() and runs in qemu
466 static void qemu_rbd_complete_aio(RADOSCB
*rcb
)
468 RBDAIOCB
*acb
= rcb
->acb
;
473 if (acb
->cmd
!= RBD_AIO_READ
) {
477 } else if (!acb
->error
) {
478 acb
->ret
= rcb
->size
;
482 qemu_rbd_memset(rcb
, 0);
485 } else if (r
< rcb
->size
) {
486 qemu_rbd_memset(rcb
, r
);
488 acb
->ret
= rcb
->size
;
490 } else if (!acb
->error
) {
497 if (!LIBRBD_USE_IOVEC
) {
498 if (acb
->cmd
== RBD_AIO_READ
) {
499 qemu_iovec_from_buf(acb
->qiov
, 0, acb
->bounce
, acb
->qiov
->size
);
501 qemu_vfree(acb
->bounce
);
504 acb
->common
.cb(acb
->common
.opaque
, (acb
->ret
> 0 ? 0 : acb
->ret
));
509 static char *qemu_rbd_mon_host(BlockdevOptionsRbd
*opts
, Error
**errp
)
512 const char *host
, *port
;
514 InetSocketAddressBaseList
*p
;
517 if (!opts
->has_server
) {
521 for (cnt
= 0, p
= opts
->server
; p
; p
= p
->next
) {
525 vals
= g_new(const char *, cnt
+ 1);
527 for (i
= 0, p
= opts
->server
; p
; p
= p
->next
, i
++) {
528 host
= p
->value
->host
;
529 port
= p
->value
->port
;
531 if (strchr(host
, ':')) {
532 vals
[i
] = g_strdup_printf("[%s]:%s", host
, port
);
534 vals
[i
] = g_strdup_printf("%s:%s", host
, port
);
539 rados_str
= i
? g_strjoinv(";", (char **)vals
) : NULL
;
540 g_strfreev((char **)vals
);
544 static int qemu_rbd_connect(rados_t
*cluster
, rados_ioctx_t
*io_ctx
,
545 BlockdevOptionsRbd
*opts
, bool cache
,
546 const char *keypairs
, const char *secretid
,
549 char *mon_host
= NULL
;
550 Error
*local_err
= NULL
;
554 if (opts
->key_secret
) {
556 "Legacy 'password-secret' clashes with 'key-secret'");
559 opts
->key_secret
= g_strdup(secretid
);
560 opts
->has_key_secret
= true;
563 mon_host
= qemu_rbd_mon_host(opts
, &local_err
);
565 error_propagate(errp
, local_err
);
570 r
= rados_create(cluster
, opts
->user
);
572 error_setg_errno(errp
, -r
, "error initializing");
576 /* try default location when conf=NULL, but ignore failure */
577 r
= rados_conf_read_file(*cluster
, opts
->conf
);
578 if (opts
->has_conf
&& r
< 0) {
579 error_setg_errno(errp
, -r
, "error reading conf file %s", opts
->conf
);
580 goto failed_shutdown
;
583 r
= qemu_rbd_set_keypairs(*cluster
, keypairs
, errp
);
585 goto failed_shutdown
;
589 r
= rados_conf_set(*cluster
, "mon_host", mon_host
);
591 goto failed_shutdown
;
595 r
= qemu_rbd_set_auth(*cluster
, opts
, errp
);
597 goto failed_shutdown
;
601 * Fallback to more conservative semantics if setting cache
602 * options fails. Ignore errors from setting rbd_cache because the
603 * only possible error is that the option does not exist, and
604 * librbd defaults to no caching. If write through caching cannot
605 * be set up, fall back to no caching.
608 rados_conf_set(*cluster
, "rbd_cache", "true");
610 rados_conf_set(*cluster
, "rbd_cache", "false");
613 r
= rados_connect(*cluster
);
615 error_setg_errno(errp
, -r
, "error connecting");
616 goto failed_shutdown
;
619 r
= rados_ioctx_create(*cluster
, opts
->pool
, io_ctx
);
621 error_setg_errno(errp
, -r
, "error opening pool %s", opts
->pool
);
622 goto failed_shutdown
;
625 * Set the namespace after opening the io context on the pool,
626 * if nspace == NULL or if nspace == "", it is just as we did nothing
628 rados_ioctx_set_namespace(*io_ctx
, opts
->q_namespace
);
633 rados_shutdown(*cluster
);
639 static int qemu_rbd_convert_options(QDict
*options
, BlockdevOptionsRbd
**opts
,
644 /* Convert the remaining options into a QAPI object */
645 v
= qobject_input_visitor_new_flat_confused(options
, errp
);
650 visit_type_BlockdevOptionsRbd(v
, NULL
, opts
, errp
);
659 static int qemu_rbd_attempt_legacy_options(QDict
*options
,
660 BlockdevOptionsRbd
**opts
,
666 filename
= g_strdup(qdict_get_try_str(options
, "filename"));
670 qdict_del(options
, "filename");
672 qemu_rbd_parse_filename(filename
, options
, NULL
);
674 /* keypairs freed by caller */
675 *keypairs
= g_strdup(qdict_get_try_str(options
, "=keyvalue-pairs"));
677 qdict_del(options
, "=keyvalue-pairs");
680 r
= qemu_rbd_convert_options(options
, opts
, NULL
);
686 static int qemu_rbd_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
689 BDRVRBDState
*s
= bs
->opaque
;
690 BlockdevOptionsRbd
*opts
= NULL
;
692 Error
*local_err
= NULL
;
693 char *keypairs
, *secretid
;
696 keypairs
= g_strdup(qdict_get_try_str(options
, "=keyvalue-pairs"));
698 qdict_del(options
, "=keyvalue-pairs");
701 secretid
= g_strdup(qdict_get_try_str(options
, "password-secret"));
703 qdict_del(options
, "password-secret");
706 r
= qemu_rbd_convert_options(options
, &opts
, &local_err
);
708 /* If keypairs are present, that means some options are present in
709 * the modern option format. Don't attempt to parse legacy option
710 * formats, as we won't support mixed usage. */
712 error_propagate(errp
, local_err
);
716 /* If the initial attempt to convert and process the options failed,
717 * we may be attempting to open an image file that has the rbd options
718 * specified in the older format consisting of all key/value pairs
719 * encoded in the filename. Go ahead and attempt to parse the
720 * filename, and see if we can pull out the required options. */
721 r
= qemu_rbd_attempt_legacy_options(options
, &opts
, &keypairs
);
723 /* Propagate the original error, not the legacy parsing fallback
724 * error, as the latter was just a best-effort attempt. */
725 error_propagate(errp
, local_err
);
728 /* Take care whenever deciding to actually deprecate; once this ability
729 * is removed, we will not be able to open any images with legacy-styled
730 * backing image strings. */
731 warn_report("RBD options encoded in the filename as keyvalue pairs "
735 /* Remove the processed options from the QDict (the visitor processes
736 * _all_ options in the QDict) */
737 while ((e
= qdict_first(options
))) {
738 qdict_del(options
, e
->key
);
741 r
= qemu_rbd_connect(&s
->cluster
, &s
->io_ctx
, opts
,
742 !(flags
& BDRV_O_NOCACHE
), keypairs
, secretid
, errp
);
747 s
->snap
= g_strdup(opts
->snapshot
);
748 s
->image_name
= g_strdup(opts
->image
);
750 /* rbd_open is always r/w */
751 r
= rbd_open(s
->io_ctx
, s
->image_name
, &s
->image
, s
->snap
);
753 error_setg_errno(errp
, -r
, "error reading header from %s",
758 r
= rbd_get_size(s
->image
, &s
->image_size
);
760 error_setg_errno(errp
, -r
, "error getting image size from %s",
766 /* If we are using an rbd snapshot, we must be r/o, otherwise
768 if (s
->snap
!= NULL
) {
769 r
= bdrv_apply_auto_read_only(bs
, "rbd snapshots are read-only", errp
);
776 /* When extending regular files, we get zeros from the OS */
777 bs
->supported_truncate_flags
= BDRV_REQ_ZERO_WRITE
;
783 rados_ioctx_destroy(s
->io_ctx
);
785 g_free(s
->image_name
);
786 rados_shutdown(s
->cluster
);
788 qapi_free_BlockdevOptionsRbd(opts
);
795 /* Since RBD is currently always opened R/W via the API,
796 * we just need to check if we are using a snapshot or not, in
797 * order to determine if we will allow it to be R/W */
798 static int qemu_rbd_reopen_prepare(BDRVReopenState
*state
,
799 BlockReopenQueue
*queue
, Error
**errp
)
801 BDRVRBDState
*s
= state
->bs
->opaque
;
804 if (s
->snap
&& state
->flags
& BDRV_O_RDWR
) {
806 "Cannot change node '%s' to r/w when using RBD snapshot",
807 bdrv_get_device_or_node_name(state
->bs
));
814 static void qemu_rbd_close(BlockDriverState
*bs
)
816 BDRVRBDState
*s
= bs
->opaque
;
819 rados_ioctx_destroy(s
->io_ctx
);
821 g_free(s
->image_name
);
822 rados_shutdown(s
->cluster
);
825 /* Resize the RBD image and update the 'image_size' with the current size */
826 static int qemu_rbd_resize(BlockDriverState
*bs
, uint64_t size
)
828 BDRVRBDState
*s
= bs
->opaque
;
831 r
= rbd_resize(s
->image
, size
);
836 s
->image_size
= size
;
841 static const AIOCBInfo rbd_aiocb_info
= {
842 .aiocb_size
= sizeof(RBDAIOCB
),
845 static void rbd_finish_bh(void *opaque
)
847 RADOSCB
*rcb
= opaque
;
848 qemu_rbd_complete_aio(rcb
);
852 * This is the callback function for rbd_aio_read and _write
854 * Note: this function is being called from a non qemu thread so
855 * we need to be careful about what we do here. Generally we only
856 * schedule a BH, and do the rest of the io completion handling
857 * from rbd_finish_bh() which runs in a qemu context.
859 static void rbd_finish_aiocb(rbd_completion_t c
, RADOSCB
*rcb
)
861 RBDAIOCB
*acb
= rcb
->acb
;
863 rcb
->ret
= rbd_aio_get_return_value(c
);
866 replay_bh_schedule_oneshot_event(bdrv_get_aio_context(acb
->common
.bs
),
870 static int rbd_aio_discard_wrapper(rbd_image_t image
,
873 rbd_completion_t comp
)
875 #ifdef LIBRBD_SUPPORTS_DISCARD
876 return rbd_aio_discard(image
, off
, len
, comp
);
882 static int rbd_aio_flush_wrapper(rbd_image_t image
,
883 rbd_completion_t comp
)
885 #ifdef LIBRBD_SUPPORTS_AIO_FLUSH
886 return rbd_aio_flush(image
, comp
);
892 static BlockAIOCB
*rbd_start_aio(BlockDriverState
*bs
,
896 BlockCompletionFunc
*cb
,
905 BDRVRBDState
*s
= bs
->opaque
;
907 acb
= qemu_aio_get(&rbd_aiocb_info
, bs
, cb
, opaque
);
910 assert(!qiov
|| qiov
->size
== size
);
912 rcb
= g_new(RADOSCB
, 1);
914 if (!LIBRBD_USE_IOVEC
) {
915 if (cmd
== RBD_AIO_DISCARD
|| cmd
== RBD_AIO_FLUSH
) {
918 acb
->bounce
= qemu_try_blockalign(bs
, qiov
->size
);
919 if (acb
->bounce
== NULL
) {
923 if (cmd
== RBD_AIO_WRITE
) {
924 qemu_iovec_to_buf(acb
->qiov
, 0, acb
->bounce
, qiov
->size
);
926 rcb
->buf
= acb
->bounce
;
936 r
= rbd_aio_create_completion(rcb
, (rbd_callback_t
) rbd_finish_aiocb
, &c
);
942 case RBD_AIO_WRITE
: {
944 * RBD APIs don't allow us to write more than actual size, so in order
945 * to support growing images, we resize the image before write
946 * operations that exceed the current size.
948 if (off
+ size
> s
->image_size
) {
949 r
= qemu_rbd_resize(bs
, off
+ size
);
951 goto failed_completion
;
954 #ifdef LIBRBD_SUPPORTS_IOVEC
955 r
= rbd_aio_writev(s
->image
, qiov
->iov
, qiov
->niov
, off
, c
);
957 r
= rbd_aio_write(s
->image
, off
, size
, rcb
->buf
, c
);
962 #ifdef LIBRBD_SUPPORTS_IOVEC
963 r
= rbd_aio_readv(s
->image
, qiov
->iov
, qiov
->niov
, off
, c
);
965 r
= rbd_aio_read(s
->image
, off
, size
, rcb
->buf
, c
);
968 case RBD_AIO_DISCARD
:
969 r
= rbd_aio_discard_wrapper(s
->image
, off
, size
, c
);
972 r
= rbd_aio_flush_wrapper(s
->image
, c
);
979 goto failed_completion
;
987 if (!LIBRBD_USE_IOVEC
) {
988 qemu_vfree(acb
->bounce
);
995 static BlockAIOCB
*qemu_rbd_aio_preadv(BlockDriverState
*bs
,
996 uint64_t offset
, uint64_t bytes
,
997 QEMUIOVector
*qiov
, int flags
,
998 BlockCompletionFunc
*cb
,
1001 return rbd_start_aio(bs
, offset
, qiov
, bytes
, cb
, opaque
,
1005 static BlockAIOCB
*qemu_rbd_aio_pwritev(BlockDriverState
*bs
,
1006 uint64_t offset
, uint64_t bytes
,
1007 QEMUIOVector
*qiov
, int flags
,
1008 BlockCompletionFunc
*cb
,
1011 return rbd_start_aio(bs
, offset
, qiov
, bytes
, cb
, opaque
,
1015 #ifdef LIBRBD_SUPPORTS_AIO_FLUSH
1016 static BlockAIOCB
*qemu_rbd_aio_flush(BlockDriverState
*bs
,
1017 BlockCompletionFunc
*cb
,
1020 return rbd_start_aio(bs
, 0, NULL
, 0, cb
, opaque
, RBD_AIO_FLUSH
);
1025 static int qemu_rbd_co_flush(BlockDriverState
*bs
)
1027 #if LIBRBD_VERSION_CODE >= LIBRBD_VERSION(0, 1, 1)
1028 /* rbd_flush added in 0.1.1 */
1029 BDRVRBDState
*s
= bs
->opaque
;
1030 return rbd_flush(s
->image
);
1037 static int qemu_rbd_getinfo(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
1039 BDRVRBDState
*s
= bs
->opaque
;
1040 rbd_image_info_t info
;
1043 r
= rbd_stat(s
->image
, &info
, sizeof(info
));
1048 bdi
->cluster_size
= info
.obj_size
;
1052 static int64_t qemu_rbd_getlength(BlockDriverState
*bs
)
1054 BDRVRBDState
*s
= bs
->opaque
;
1055 rbd_image_info_t info
;
1058 r
= rbd_stat(s
->image
, &info
, sizeof(info
));
1066 static int coroutine_fn
qemu_rbd_co_truncate(BlockDriverState
*bs
,
1069 PreallocMode prealloc
,
1070 BdrvRequestFlags flags
,
1075 if (prealloc
!= PREALLOC_MODE_OFF
) {
1076 error_setg(errp
, "Unsupported preallocation mode '%s'",
1077 PreallocMode_str(prealloc
));
1081 r
= qemu_rbd_resize(bs
, offset
);
1083 error_setg_errno(errp
, -r
, "Failed to resize file");
1090 static int qemu_rbd_snap_create(BlockDriverState
*bs
,
1091 QEMUSnapshotInfo
*sn_info
)
1093 BDRVRBDState
*s
= bs
->opaque
;
1096 if (sn_info
->name
[0] == '\0') {
1097 return -EINVAL
; /* we need a name for rbd snapshots */
1101 * rbd snapshots are using the name as the user controlled unique identifier
1102 * we can't use the rbd snapid for that purpose, as it can't be set
1104 if (sn_info
->id_str
[0] != '\0' &&
1105 strcmp(sn_info
->id_str
, sn_info
->name
) != 0) {
1109 if (strlen(sn_info
->name
) >= sizeof(sn_info
->id_str
)) {
1113 r
= rbd_snap_create(s
->image
, sn_info
->name
);
1115 error_report("failed to create snap: %s", strerror(-r
));
1122 static int qemu_rbd_snap_remove(BlockDriverState
*bs
,
1123 const char *snapshot_id
,
1124 const char *snapshot_name
,
1127 BDRVRBDState
*s
= bs
->opaque
;
1130 if (!snapshot_name
) {
1131 error_setg(errp
, "rbd need a valid snapshot name");
1135 /* If snapshot_id is specified, it must be equal to name, see
1136 qemu_rbd_snap_list() */
1137 if (snapshot_id
&& strcmp(snapshot_id
, snapshot_name
)) {
1139 "rbd do not support snapshot id, it should be NULL or "
1140 "equal to snapshot name");
1144 r
= rbd_snap_remove(s
->image
, snapshot_name
);
1146 error_setg_errno(errp
, -r
, "Failed to remove the snapshot");
1151 static int qemu_rbd_snap_rollback(BlockDriverState
*bs
,
1152 const char *snapshot_name
)
1154 BDRVRBDState
*s
= bs
->opaque
;
1156 return rbd_snap_rollback(s
->image
, snapshot_name
);
1159 static int qemu_rbd_snap_list(BlockDriverState
*bs
,
1160 QEMUSnapshotInfo
**psn_tab
)
1162 BDRVRBDState
*s
= bs
->opaque
;
1163 QEMUSnapshotInfo
*sn_info
, *sn_tab
= NULL
;
1165 rbd_snap_info_t
*snaps
;
1166 int max_snaps
= RBD_MAX_SNAPS
;
1169 snaps
= g_new(rbd_snap_info_t
, max_snaps
);
1170 snap_count
= rbd_snap_list(s
->image
, snaps
, &max_snaps
);
1171 if (snap_count
<= 0) {
1174 } while (snap_count
== -ERANGE
);
1176 if (snap_count
<= 0) {
1180 sn_tab
= g_new0(QEMUSnapshotInfo
, snap_count
);
1182 for (i
= 0; i
< snap_count
; i
++) {
1183 const char *snap_name
= snaps
[i
].name
;
1185 sn_info
= sn_tab
+ i
;
1186 pstrcpy(sn_info
->id_str
, sizeof(sn_info
->id_str
), snap_name
);
1187 pstrcpy(sn_info
->name
, sizeof(sn_info
->name
), snap_name
);
1189 sn_info
->vm_state_size
= snaps
[i
].size
;
1190 sn_info
->date_sec
= 0;
1191 sn_info
->date_nsec
= 0;
1192 sn_info
->vm_clock_nsec
= 0;
1194 rbd_snap_list_end(snaps
);
1202 #ifdef LIBRBD_SUPPORTS_DISCARD
1203 static BlockAIOCB
*qemu_rbd_aio_pdiscard(BlockDriverState
*bs
,
1206 BlockCompletionFunc
*cb
,
1209 return rbd_start_aio(bs
, offset
, NULL
, bytes
, cb
, opaque
,
1214 #ifdef LIBRBD_SUPPORTS_INVALIDATE
1215 static void coroutine_fn
qemu_rbd_co_invalidate_cache(BlockDriverState
*bs
,
1218 BDRVRBDState
*s
= bs
->opaque
;
1219 int r
= rbd_invalidate_cache(s
->image
);
1221 error_setg_errno(errp
, -r
, "Failed to invalidate the cache");
1226 static QemuOptsList qemu_rbd_create_opts
= {
1227 .name
= "rbd-create-opts",
1228 .head
= QTAILQ_HEAD_INITIALIZER(qemu_rbd_create_opts
.head
),
1231 .name
= BLOCK_OPT_SIZE
,
1232 .type
= QEMU_OPT_SIZE
,
1233 .help
= "Virtual disk size"
1236 .name
= BLOCK_OPT_CLUSTER_SIZE
,
1237 .type
= QEMU_OPT_SIZE
,
1238 .help
= "RBD object size"
1241 .name
= "password-secret",
1242 .type
= QEMU_OPT_STRING
,
1243 .help
= "ID of secret providing the password",
1245 { /* end of list */ }
1249 static const char *const qemu_rbd_strong_runtime_opts
[] = {
1262 static BlockDriver bdrv_rbd
= {
1263 .format_name
= "rbd",
1264 .instance_size
= sizeof(BDRVRBDState
),
1265 .bdrv_parse_filename
= qemu_rbd_parse_filename
,
1266 .bdrv_refresh_limits
= qemu_rbd_refresh_limits
,
1267 .bdrv_file_open
= qemu_rbd_open
,
1268 .bdrv_close
= qemu_rbd_close
,
1269 .bdrv_reopen_prepare
= qemu_rbd_reopen_prepare
,
1270 .bdrv_co_create
= qemu_rbd_co_create
,
1271 .bdrv_co_create_opts
= qemu_rbd_co_create_opts
,
1272 .bdrv_has_zero_init
= bdrv_has_zero_init_1
,
1273 .bdrv_get_info
= qemu_rbd_getinfo
,
1274 .create_opts
= &qemu_rbd_create_opts
,
1275 .bdrv_getlength
= qemu_rbd_getlength
,
1276 .bdrv_co_truncate
= qemu_rbd_co_truncate
,
1277 .protocol_name
= "rbd",
1279 .bdrv_aio_preadv
= qemu_rbd_aio_preadv
,
1280 .bdrv_aio_pwritev
= qemu_rbd_aio_pwritev
,
1282 #ifdef LIBRBD_SUPPORTS_AIO_FLUSH
1283 .bdrv_aio_flush
= qemu_rbd_aio_flush
,
1285 .bdrv_co_flush_to_disk
= qemu_rbd_co_flush
,
1288 #ifdef LIBRBD_SUPPORTS_DISCARD
1289 .bdrv_aio_pdiscard
= qemu_rbd_aio_pdiscard
,
1292 .bdrv_snapshot_create
= qemu_rbd_snap_create
,
1293 .bdrv_snapshot_delete
= qemu_rbd_snap_remove
,
1294 .bdrv_snapshot_list
= qemu_rbd_snap_list
,
1295 .bdrv_snapshot_goto
= qemu_rbd_snap_rollback
,
1296 #ifdef LIBRBD_SUPPORTS_INVALIDATE
1297 .bdrv_co_invalidate_cache
= qemu_rbd_co_invalidate_cache
,
1300 .strong_runtime_opts
= qemu_rbd_strong_runtime_opts
,
1303 static void bdrv_rbd_init(void)
1305 bdrv_register(&bdrv_rbd
);
1308 block_init(bdrv_rbd_init
);