2 * GlusterFS backend for QEMU
4 * Copyright (C) 2012 Bharata B Rao <bharata@linux.vnet.ibm.com>
6 * This work is licensed under the terms of the GNU GPL, version 2 or later.
7 * See the COPYING file in the top-level directory.
10 #include "qemu/osdep.h"
11 #include <glusterfs/api/glfs.h>
12 #include "block/block_int.h"
13 #include "qapi/error.h"
14 #include "qapi/qmp/qerror.h"
15 #include "qapi/util.h"
17 #include "qemu/error-report.h"
18 #include "qemu/cutils.h"
20 #define GLUSTER_OPT_FILENAME "filename"
21 #define GLUSTER_OPT_VOLUME "volume"
22 #define GLUSTER_OPT_PATH "path"
23 #define GLUSTER_OPT_TYPE "type"
24 #define GLUSTER_OPT_SERVER_PATTERN "server."
25 #define GLUSTER_OPT_HOST "host"
26 #define GLUSTER_OPT_PORT "port"
27 #define GLUSTER_OPT_TO "to"
28 #define GLUSTER_OPT_IPV4 "ipv4"
29 #define GLUSTER_OPT_IPV6 "ipv6"
30 #define GLUSTER_OPT_SOCKET "socket"
31 #define GLUSTER_OPT_DEBUG "debug"
32 #define GLUSTER_DEFAULT_PORT 24007
33 #define GLUSTER_DEBUG_DEFAULT 4
34 #define GLUSTER_DEBUG_MAX 9
35 #define GLUSTER_OPT_LOGFILE "logfile"
36 #define GLUSTER_LOGFILE_DEFAULT "-" /* handled in libgfapi as /dev/stderr */
38 #define GERR_INDEX_HINT "hint: check in 'server' array index '%d'\n"
40 typedef struct GlusterAIOCB
{
44 AioContext
*aio_context
;
47 typedef struct BDRVGlusterState
{
51 bool supports_seek_data
;
55 typedef struct BDRVGlusterReopenState
{
58 } BDRVGlusterReopenState
;
61 typedef struct GlfsPreopened
{
67 typedef struct ListElement
{
68 QLIST_ENTRY(ListElement
) list
;
72 static QLIST_HEAD(glfs_list
, ListElement
) glfs_list
;
74 static QemuOptsList qemu_gluster_create_opts
= {
75 .name
= "qemu-gluster-create-opts",
76 .head
= QTAILQ_HEAD_INITIALIZER(qemu_gluster_create_opts
.head
),
79 .name
= BLOCK_OPT_SIZE
,
80 .type
= QEMU_OPT_SIZE
,
81 .help
= "Virtual disk size"
84 .name
= BLOCK_OPT_PREALLOC
,
85 .type
= QEMU_OPT_STRING
,
86 .help
= "Preallocation mode (allowed values: off, full)"
89 .name
= GLUSTER_OPT_DEBUG
,
90 .type
= QEMU_OPT_NUMBER
,
91 .help
= "Gluster log level, valid range is 0-9",
94 .name
= GLUSTER_OPT_LOGFILE
,
95 .type
= QEMU_OPT_STRING
,
96 .help
= "Logfile path of libgfapi",
102 static QemuOptsList runtime_opts
= {
104 .head
= QTAILQ_HEAD_INITIALIZER(runtime_opts
.head
),
107 .name
= GLUSTER_OPT_FILENAME
,
108 .type
= QEMU_OPT_STRING
,
109 .help
= "URL to the gluster image",
112 .name
= GLUSTER_OPT_DEBUG
,
113 .type
= QEMU_OPT_NUMBER
,
114 .help
= "Gluster log level, valid range is 0-9",
117 .name
= GLUSTER_OPT_LOGFILE
,
118 .type
= QEMU_OPT_STRING
,
119 .help
= "Logfile path of libgfapi",
121 { /* end of list */ }
125 static QemuOptsList runtime_json_opts
= {
126 .name
= "gluster_json",
127 .head
= QTAILQ_HEAD_INITIALIZER(runtime_json_opts
.head
),
130 .name
= GLUSTER_OPT_VOLUME
,
131 .type
= QEMU_OPT_STRING
,
132 .help
= "name of gluster volume where VM image resides",
135 .name
= GLUSTER_OPT_PATH
,
136 .type
= QEMU_OPT_STRING
,
137 .help
= "absolute path to image file in gluster volume",
140 .name
= GLUSTER_OPT_DEBUG
,
141 .type
= QEMU_OPT_NUMBER
,
142 .help
= "Gluster log level, valid range is 0-9",
144 { /* end of list */ }
148 static QemuOptsList runtime_type_opts
= {
149 .name
= "gluster_type",
150 .head
= QTAILQ_HEAD_INITIALIZER(runtime_type_opts
.head
),
153 .name
= GLUSTER_OPT_TYPE
,
154 .type
= QEMU_OPT_STRING
,
157 { /* end of list */ }
161 static QemuOptsList runtime_unix_opts
= {
162 .name
= "gluster_unix",
163 .head
= QTAILQ_HEAD_INITIALIZER(runtime_unix_opts
.head
),
166 .name
= GLUSTER_OPT_SOCKET
,
167 .type
= QEMU_OPT_STRING
,
168 .help
= "socket file path)",
170 { /* end of list */ }
174 static QemuOptsList runtime_inet_opts
= {
175 .name
= "gluster_inet",
176 .head
= QTAILQ_HEAD_INITIALIZER(runtime_inet_opts
.head
),
179 .name
= GLUSTER_OPT_TYPE
,
180 .type
= QEMU_OPT_STRING
,
184 .name
= GLUSTER_OPT_HOST
,
185 .type
= QEMU_OPT_STRING
,
186 .help
= "host address (hostname/ipv4/ipv6 addresses)",
189 .name
= GLUSTER_OPT_PORT
,
190 .type
= QEMU_OPT_STRING
,
191 .help
= "port number on which glusterd is listening (default 24007)",
195 .type
= QEMU_OPT_NUMBER
,
196 .help
= "max port number, not supported by gluster",
200 .type
= QEMU_OPT_BOOL
,
201 .help
= "ipv4 bool value, not supported by gluster",
205 .type
= QEMU_OPT_BOOL
,
206 .help
= "ipv6 bool value, not supported by gluster",
208 { /* end of list */ }
212 static void glfs_set_preopened(const char *volume
, glfs_t
*fs
)
214 ListElement
*entry
= NULL
;
216 entry
= g_new(ListElement
, 1);
218 entry
->saved
.volume
= g_strdup(volume
);
220 entry
->saved
.fs
= fs
;
221 entry
->saved
.ref
= 1;
223 QLIST_INSERT_HEAD(&glfs_list
, entry
, list
);
226 static glfs_t
*glfs_find_preopened(const char *volume
)
228 ListElement
*entry
= NULL
;
230 QLIST_FOREACH(entry
, &glfs_list
, list
) {
231 if (strcmp(entry
->saved
.volume
, volume
) == 0) {
233 return entry
->saved
.fs
;
240 static void glfs_clear_preopened(glfs_t
*fs
)
242 ListElement
*entry
= NULL
;
249 QLIST_FOREACH_SAFE(entry
, &glfs_list
, list
, next
) {
250 if (entry
->saved
.fs
== fs
) {
251 if (--entry
->saved
.ref
) {
255 QLIST_REMOVE(entry
, list
);
257 glfs_fini(entry
->saved
.fs
);
258 g_free(entry
->saved
.volume
);
264 static int parse_volume_options(BlockdevOptionsGluster
*gconf
, char *path
)
273 p
= q
= path
+ strspn(path
, "/");
274 p
+= strcspn(p
, "/");
278 gconf
->volume
= g_strndup(q
, p
- q
);
285 gconf
->path
= g_strdup(p
);
290 * file=gluster[+transport]://[host[:port]]/volume/path[?socket=...]
292 * 'gluster' is the protocol.
294 * 'transport' specifies the transport type used to connect to gluster
295 * management daemon (glusterd). Valid transport types are
296 * tcp or unix. If a transport type isn't specified, then tcp type is assumed.
298 * 'host' specifies the host where the volume file specification for
299 * the given volume resides. This can be either hostname or ipv4 address.
300 * If transport type is 'unix', then 'host' field should not be specified.
301 * The 'socket' field needs to be populated with the path to unix domain
304 * 'port' is the port number on which glusterd is listening. This is optional
305 * and if not specified, QEMU will send 0 which will make gluster to use the
306 * default port. If the transport type is unix, then 'port' should not be
309 * 'volume' is the name of the gluster volume which contains the VM image.
311 * 'path' is the path to the actual VM image that resides on gluster volume.
315 * file=gluster://1.2.3.4/testvol/a.img
316 * file=gluster+tcp://1.2.3.4/testvol/a.img
317 * file=gluster+tcp://1.2.3.4:24007/testvol/dir/a.img
318 * file=gluster+tcp://host.domain.com:24007/testvol/dir/a.img
319 * file=gluster+unix:///testvol/dir/a.img?socket=/tmp/glusterd.socket
321 static int qemu_gluster_parse_uri(BlockdevOptionsGluster
*gconf
,
322 const char *filename
)
324 SocketAddressFlat
*gsconf
;
326 QueryParams
*qp
= NULL
;
327 bool is_unix
= false;
330 uri
= uri_parse(filename
);
335 gconf
->server
= g_new0(SocketAddressFlatList
, 1);
336 gconf
->server
->value
= gsconf
= g_new0(SocketAddressFlat
, 1);
339 if (!uri
->scheme
|| !strcmp(uri
->scheme
, "gluster")) {
340 gsconf
->type
= SOCKET_ADDRESS_FLAT_TYPE_INET
;
341 } else if (!strcmp(uri
->scheme
, "gluster+tcp")) {
342 gsconf
->type
= SOCKET_ADDRESS_FLAT_TYPE_INET
;
343 } else if (!strcmp(uri
->scheme
, "gluster+unix")) {
344 gsconf
->type
= SOCKET_ADDRESS_FLAT_TYPE_UNIX
;
346 } else if (!strcmp(uri
->scheme
, "gluster+rdma")) {
347 gsconf
->type
= SOCKET_ADDRESS_FLAT_TYPE_INET
;
348 error_report("Warning: rdma feature is not supported, falling "
355 ret
= parse_volume_options(gconf
, uri
->path
);
360 qp
= query_params_parse(uri
->query
);
361 if (qp
->n
> 1 || (is_unix
&& !qp
->n
) || (!is_unix
&& qp
->n
)) {
367 if (uri
->server
|| uri
->port
) {
371 if (strcmp(qp
->p
[0].name
, "socket")) {
375 gsconf
->u
.q_unix
.path
= g_strdup(qp
->p
[0].value
);
377 gsconf
->u
.inet
.host
= g_strdup(uri
->server
? uri
->server
: "localhost");
379 gsconf
->u
.inet
.port
= g_strdup_printf("%d", uri
->port
);
381 gsconf
->u
.inet
.port
= g_strdup_printf("%d", GLUSTER_DEFAULT_PORT
);
387 query_params_free(qp
);
393 static struct glfs
*qemu_gluster_glfs_init(BlockdevOptionsGluster
*gconf
,
399 SocketAddressFlatList
*server
;
400 unsigned long long port
;
402 glfs
= glfs_find_preopened(gconf
->volume
);
407 glfs
= glfs_new(gconf
->volume
);
412 glfs_set_preopened(gconf
->volume
, glfs
);
414 for (server
= gconf
->server
; server
; server
= server
->next
) {
415 if (server
->value
->type
== SOCKET_ADDRESS_FLAT_TYPE_UNIX
) {
416 ret
= glfs_set_volfile_server(glfs
, "unix",
417 server
->value
->u
.q_unix
.path
, 0);
419 if (parse_uint_full(server
->value
->u
.inet
.port
, &port
, 10) < 0 ||
421 error_setg(errp
, "'%s' is not a valid port number",
422 server
->value
->u
.inet
.port
);
426 ret
= glfs_set_volfile_server(glfs
, "tcp",
427 server
->value
->u
.inet
.host
,
436 ret
= glfs_set_logging(glfs
, gconf
->logfile
, gconf
->debug
);
441 ret
= glfs_init(glfs
);
443 error_setg(errp
, "Gluster connection for volume %s, path %s failed"
444 " to connect", gconf
->volume
, gconf
->path
);
445 for (server
= gconf
->server
; server
; server
= server
->next
) {
446 if (server
->value
->type
== SOCKET_ADDRESS_FLAT_TYPE_UNIX
) {
447 error_append_hint(errp
, "hint: failed on socket %s ",
448 server
->value
->u
.q_unix
.path
);
450 error_append_hint(errp
, "hint: failed on host %s and port %s ",
451 server
->value
->u
.inet
.host
,
452 server
->value
->u
.inet
.port
);
456 error_append_hint(errp
, "Please refer to gluster logs for more info\n");
458 /* glfs_init sometimes doesn't set errno although docs suggest that */
470 glfs_clear_preopened(glfs
);
477 * Convert the json formatted command line into qapi.
479 static int qemu_gluster_parse_json(BlockdevOptionsGluster
*gconf
,
480 QDict
*options
, Error
**errp
)
483 SocketAddressFlat
*gsconf
= NULL
;
484 SocketAddressFlatList
*curr
= NULL
;
485 QDict
*backing_options
= NULL
;
486 Error
*local_err
= NULL
;
492 /* create opts info from runtime_json_opts list */
493 opts
= qemu_opts_create(&runtime_json_opts
, NULL
, 0, &error_abort
);
494 qemu_opts_absorb_qdict(opts
, options
, &local_err
);
499 num_servers
= qdict_array_entries(options
, GLUSTER_OPT_SERVER_PATTERN
);
500 if (num_servers
< 1) {
501 error_setg(&local_err
, QERR_MISSING_PARAMETER
, "server");
505 ptr
= qemu_opt_get(opts
, GLUSTER_OPT_VOLUME
);
507 error_setg(&local_err
, QERR_MISSING_PARAMETER
, GLUSTER_OPT_VOLUME
);
510 gconf
->volume
= g_strdup(ptr
);
512 ptr
= qemu_opt_get(opts
, GLUSTER_OPT_PATH
);
514 error_setg(&local_err
, QERR_MISSING_PARAMETER
, GLUSTER_OPT_PATH
);
517 gconf
->path
= g_strdup(ptr
);
520 for (i
= 0; i
< num_servers
; i
++) {
521 str
= g_strdup_printf(GLUSTER_OPT_SERVER_PATTERN
"%d.", i
);
522 qdict_extract_subqdict(options
, &backing_options
, str
);
524 /* create opts info from runtime_type_opts list */
525 opts
= qemu_opts_create(&runtime_type_opts
, NULL
, 0, &error_abort
);
526 qemu_opts_absorb_qdict(opts
, backing_options
, &local_err
);
531 ptr
= qemu_opt_get(opts
, GLUSTER_OPT_TYPE
);
533 error_setg(&local_err
, QERR_MISSING_PARAMETER
, GLUSTER_OPT_TYPE
);
534 error_append_hint(&local_err
, GERR_INDEX_HINT
, i
);
538 gsconf
= g_new0(SocketAddressFlat
, 1);
539 if (!strcmp(ptr
, "tcp")) {
540 ptr
= "inet"; /* accept legacy "tcp" */
542 gsconf
->type
= qapi_enum_parse(SocketAddressFlatType_lookup
, ptr
,
543 SOCKET_ADDRESS_FLAT_TYPE__MAX
, -1,
546 error_append_hint(&local_err
,
547 "Parameter '%s' may be 'inet' or 'unix'\n",
549 error_append_hint(&local_err
, GERR_INDEX_HINT
, i
);
554 if (gsconf
->type
== SOCKET_ADDRESS_FLAT_TYPE_INET
) {
555 /* create opts info from runtime_inet_opts list */
556 opts
= qemu_opts_create(&runtime_inet_opts
, NULL
, 0, &error_abort
);
557 qemu_opts_absorb_qdict(opts
, backing_options
, &local_err
);
562 ptr
= qemu_opt_get(opts
, GLUSTER_OPT_HOST
);
564 error_setg(&local_err
, QERR_MISSING_PARAMETER
,
566 error_append_hint(&local_err
, GERR_INDEX_HINT
, i
);
569 gsconf
->u
.inet
.host
= g_strdup(ptr
);
570 ptr
= qemu_opt_get(opts
, GLUSTER_OPT_PORT
);
572 error_setg(&local_err
, QERR_MISSING_PARAMETER
,
574 error_append_hint(&local_err
, GERR_INDEX_HINT
, i
);
577 gsconf
->u
.inet
.port
= g_strdup(ptr
);
579 /* defend for unsupported fields in InetSocketAddress,
580 * i.e. @ipv4, @ipv6 and @to
582 ptr
= qemu_opt_get(opts
, GLUSTER_OPT_TO
);
584 gsconf
->u
.inet
.has_to
= true;
586 ptr
= qemu_opt_get(opts
, GLUSTER_OPT_IPV4
);
588 gsconf
->u
.inet
.has_ipv4
= true;
590 ptr
= qemu_opt_get(opts
, GLUSTER_OPT_IPV6
);
592 gsconf
->u
.inet
.has_ipv6
= true;
594 if (gsconf
->u
.inet
.has_to
) {
595 error_setg(&local_err
, "Parameter 'to' not supported");
598 if (gsconf
->u
.inet
.has_ipv4
|| gsconf
->u
.inet
.has_ipv6
) {
599 error_setg(&local_err
, "Parameters 'ipv4/ipv6' not supported");
604 /* create opts info from runtime_unix_opts list */
605 opts
= qemu_opts_create(&runtime_unix_opts
, NULL
, 0, &error_abort
);
606 qemu_opts_absorb_qdict(opts
, backing_options
, &local_err
);
611 ptr
= qemu_opt_get(opts
, GLUSTER_OPT_SOCKET
);
613 error_setg(&local_err
, QERR_MISSING_PARAMETER
,
615 error_append_hint(&local_err
, GERR_INDEX_HINT
, i
);
618 gsconf
->u
.q_unix
.path
= g_strdup(ptr
);
622 if (gconf
->server
== NULL
) {
623 gconf
->server
= g_new0(SocketAddressFlatList
, 1);
624 gconf
->server
->value
= gsconf
;
625 curr
= gconf
->server
;
627 curr
->next
= g_new0(SocketAddressFlatList
, 1);
628 curr
->next
->value
= gsconf
;
633 QDECREF(backing_options
);
634 backing_options
= NULL
;
642 error_propagate(errp
, local_err
);
643 qapi_free_SocketAddressFlat(gsconf
);
646 QDECREF(backing_options
);
651 static struct glfs
*qemu_gluster_init(BlockdevOptionsGluster
*gconf
,
652 const char *filename
,
653 QDict
*options
, Error
**errp
)
657 ret
= qemu_gluster_parse_uri(gconf
, filename
);
659 error_setg(errp
, "invalid URI");
660 error_append_hint(errp
, "Usage: file=gluster[+transport]://"
661 "[host[:port]]volume/path[?socket=...]"
663 "[,file.logfile=/path/filename.log]\n");
668 ret
= qemu_gluster_parse_json(gconf
, options
, errp
);
670 error_append_hint(errp
, "Usage: "
671 "-drive driver=qcow2,file.driver=gluster,"
672 "file.volume=testvol,file.path=/path/a.qcow2"
674 "[,file.logfile=/path/filename.log],"
675 "file.server.0.type=inet,"
676 "file.server.0.host=1.2.3.4,"
677 "file.server.0.port=24007,"
678 "file.server.1.transport=unix,"
679 "file.server.1.socket=/var/run/glusterd.socket ..."
687 return qemu_gluster_glfs_init(gconf
, errp
);
691 * AIO callback routine called from GlusterFS thread.
693 static void gluster_finish_aiocb(struct glfs_fd
*fd
, ssize_t ret
, void *arg
)
695 GlusterAIOCB
*acb
= (GlusterAIOCB
*)arg
;
697 if (!ret
|| ret
== acb
->size
) {
698 acb
->ret
= 0; /* Success */
699 } else if (ret
< 0) {
700 acb
->ret
= -errno
; /* Read/Write failed */
702 acb
->ret
= -EIO
; /* Partial read/write - fail it */
705 aio_co_schedule(acb
->aio_context
, acb
->coroutine
);
708 static void qemu_gluster_parse_flags(int bdrv_flags
, int *open_flags
)
710 assert(open_flags
!= NULL
);
712 *open_flags
|= O_BINARY
;
714 if (bdrv_flags
& BDRV_O_RDWR
) {
715 *open_flags
|= O_RDWR
;
717 *open_flags
|= O_RDONLY
;
720 if ((bdrv_flags
& BDRV_O_NOCACHE
)) {
721 *open_flags
|= O_DIRECT
;
726 * Do SEEK_DATA/HOLE to detect if it is functional. Older broken versions of
727 * gfapi incorrectly return the current offset when SEEK_DATA/HOLE is used.
728 * - Corrected versions return -1 and set errno to EINVAL.
729 * - Versions that support SEEK_DATA/HOLE correctly, will return -1 and set
730 * errno to ENXIO when SEEK_DATA is called with a position of EOF.
732 static bool qemu_gluster_test_seek(struct glfs_fd
*fd
)
736 #if defined SEEK_HOLE && defined SEEK_DATA
739 eof
= glfs_lseek(fd
, 0, SEEK_END
);
741 /* this should never occur */
745 /* this should always fail with ENXIO if SEEK_DATA is supported */
746 ret
= glfs_lseek(fd
, eof
, SEEK_DATA
);
749 return (ret
< 0) && (errno
== ENXIO
);
752 static int qemu_gluster_open(BlockDriverState
*bs
, QDict
*options
,
753 int bdrv_flags
, Error
**errp
)
755 BDRVGlusterState
*s
= bs
->opaque
;
758 BlockdevOptionsGluster
*gconf
= NULL
;
760 Error
*local_err
= NULL
;
761 const char *filename
, *logfile
;
763 opts
= qemu_opts_create(&runtime_opts
, NULL
, 0, &error_abort
);
764 qemu_opts_absorb_qdict(opts
, options
, &local_err
);
766 error_propagate(errp
, local_err
);
771 filename
= qemu_opt_get(opts
, GLUSTER_OPT_FILENAME
);
773 s
->debug
= qemu_opt_get_number(opts
, GLUSTER_OPT_DEBUG
,
774 GLUSTER_DEBUG_DEFAULT
);
777 } else if (s
->debug
> GLUSTER_DEBUG_MAX
) {
778 s
->debug
= GLUSTER_DEBUG_MAX
;
781 gconf
= g_new0(BlockdevOptionsGluster
, 1);
782 gconf
->debug
= s
->debug
;
783 gconf
->has_debug
= true;
785 logfile
= qemu_opt_get(opts
, GLUSTER_OPT_LOGFILE
);
786 s
->logfile
= g_strdup(logfile
? logfile
: GLUSTER_LOGFILE_DEFAULT
);
788 gconf
->logfile
= g_strdup(s
->logfile
);
789 gconf
->has_logfile
= true;
791 s
->glfs
= qemu_gluster_init(gconf
, filename
, options
, errp
);
797 #ifdef CONFIG_GLUSTERFS_XLATOR_OPT
798 /* Without this, if fsync fails for a recoverable reason (for instance,
799 * ENOSPC), gluster will dump its cache, preventing retries. This means
800 * almost certain data loss. Not all gluster versions support the
801 * 'resync-failed-syncs-after-fsync' key value, but there is no way to
802 * discover during runtime if it is supported (this api returns success for
803 * unknown key/value pairs) */
804 ret
= glfs_set_xlator_option(s
->glfs
, "*-write-behind",
805 "resync-failed-syncs-after-fsync",
808 error_setg_errno(errp
, errno
, "Unable to set xlator key/value pair");
814 qemu_gluster_parse_flags(bdrv_flags
, &open_flags
);
816 s
->fd
= glfs_open(s
->glfs
, gconf
->path
, open_flags
);
821 s
->supports_seek_data
= qemu_gluster_test_seek(s
->fd
);
825 qapi_free_BlockdevOptionsGluster(gconf
);
834 glfs_clear_preopened(s
->glfs
);
839 static int qemu_gluster_reopen_prepare(BDRVReopenState
*state
,
840 BlockReopenQueue
*queue
, Error
**errp
)
844 BDRVGlusterReopenState
*reop_s
;
845 BlockdevOptionsGluster
*gconf
;
848 assert(state
!= NULL
);
849 assert(state
->bs
!= NULL
);
851 s
= state
->bs
->opaque
;
853 state
->opaque
= g_new0(BDRVGlusterReopenState
, 1);
854 reop_s
= state
->opaque
;
856 qemu_gluster_parse_flags(state
->flags
, &open_flags
);
858 gconf
= g_new0(BlockdevOptionsGluster
, 1);
859 gconf
->debug
= s
->debug
;
860 gconf
->has_debug
= true;
861 gconf
->logfile
= g_strdup(s
->logfile
);
862 gconf
->has_logfile
= true;
863 reop_s
->glfs
= qemu_gluster_init(gconf
, state
->bs
->filename
, NULL
, errp
);
864 if (reop_s
->glfs
== NULL
) {
869 #ifdef CONFIG_GLUSTERFS_XLATOR_OPT
870 ret
= glfs_set_xlator_option(reop_s
->glfs
, "*-write-behind",
871 "resync-failed-syncs-after-fsync", "on");
873 error_setg_errno(errp
, errno
, "Unable to set xlator key/value pair");
879 reop_s
->fd
= glfs_open(reop_s
->glfs
, gconf
->path
, open_flags
);
880 if (reop_s
->fd
== NULL
) {
881 /* reops->glfs will be cleaned up in _abort */
887 /* state->opaque will be freed in either the _abort or _commit */
888 qapi_free_BlockdevOptionsGluster(gconf
);
892 static void qemu_gluster_reopen_commit(BDRVReopenState
*state
)
894 BDRVGlusterReopenState
*reop_s
= state
->opaque
;
895 BDRVGlusterState
*s
= state
->bs
->opaque
;
903 glfs_clear_preopened(s
->glfs
);
905 /* use the newly opened image / connection */
907 s
->glfs
= reop_s
->glfs
;
909 g_free(state
->opaque
);
910 state
->opaque
= NULL
;
916 static void qemu_gluster_reopen_abort(BDRVReopenState
*state
)
918 BDRVGlusterReopenState
*reop_s
= state
->opaque
;
920 if (reop_s
== NULL
) {
925 glfs_close(reop_s
->fd
);
928 glfs_clear_preopened(reop_s
->glfs
);
930 g_free(state
->opaque
);
931 state
->opaque
= NULL
;
936 #ifdef CONFIG_GLUSTERFS_ZEROFILL
937 static coroutine_fn
int qemu_gluster_co_pwrite_zeroes(BlockDriverState
*bs
,
940 BdrvRequestFlags flags
)
944 BDRVGlusterState
*s
= bs
->opaque
;
948 acb
.coroutine
= qemu_coroutine_self();
949 acb
.aio_context
= bdrv_get_aio_context(bs
);
951 ret
= glfs_zerofill_async(s
->fd
, offset
, size
, gluster_finish_aiocb
, &acb
);
956 qemu_coroutine_yield();
960 static inline bool gluster_supports_zerofill(void)
965 static inline int qemu_gluster_zerofill(struct glfs_fd
*fd
, int64_t offset
,
968 return glfs_zerofill(fd
, offset
, size
);
972 static inline bool gluster_supports_zerofill(void)
977 static inline int qemu_gluster_zerofill(struct glfs_fd
*fd
, int64_t offset
,
984 static int qemu_gluster_create(const char *filename
,
985 QemuOpts
*opts
, Error
**errp
)
987 BlockdevOptionsGluster
*gconf
;
992 int64_t total_size
= 0;
995 gconf
= g_new0(BlockdevOptionsGluster
, 1);
996 gconf
->debug
= qemu_opt_get_number_del(opts
, GLUSTER_OPT_DEBUG
,
997 GLUSTER_DEBUG_DEFAULT
);
998 if (gconf
->debug
< 0) {
1000 } else if (gconf
->debug
> GLUSTER_DEBUG_MAX
) {
1001 gconf
->debug
= GLUSTER_DEBUG_MAX
;
1003 gconf
->has_debug
= true;
1005 gconf
->logfile
= qemu_opt_get_del(opts
, GLUSTER_OPT_LOGFILE
);
1006 if (!gconf
->logfile
) {
1007 gconf
->logfile
= g_strdup(GLUSTER_LOGFILE_DEFAULT
);
1009 gconf
->has_logfile
= true;
1011 glfs
= qemu_gluster_init(gconf
, filename
, NULL
, errp
);
1017 total_size
= ROUND_UP(qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0),
1020 tmp
= qemu_opt_get_del(opts
, BLOCK_OPT_PREALLOC
);
1021 if (!tmp
|| !strcmp(tmp
, "off")) {
1023 } else if (!strcmp(tmp
, "full") && gluster_supports_zerofill()) {
1026 error_setg(errp
, "Invalid preallocation mode: '%s'"
1027 " or GlusterFS doesn't support zerofill API", tmp
);
1032 fd
= glfs_creat(glfs
, gconf
->path
,
1033 O_WRONLY
| O_CREAT
| O_TRUNC
| O_BINARY
, S_IRUSR
| S_IWUSR
);
1037 if (!glfs_ftruncate(fd
, total_size
)) {
1038 if (prealloc
&& qemu_gluster_zerofill(fd
, 0, total_size
)) {
1045 if (glfs_close(fd
) != 0) {
1051 qapi_free_BlockdevOptionsGluster(gconf
);
1052 glfs_clear_preopened(glfs
);
1056 static coroutine_fn
int qemu_gluster_co_rw(BlockDriverState
*bs
,
1057 int64_t sector_num
, int nb_sectors
,
1058 QEMUIOVector
*qiov
, int write
)
1062 BDRVGlusterState
*s
= bs
->opaque
;
1063 size_t size
= nb_sectors
* BDRV_SECTOR_SIZE
;
1064 off_t offset
= sector_num
* BDRV_SECTOR_SIZE
;
1068 acb
.coroutine
= qemu_coroutine_self();
1069 acb
.aio_context
= bdrv_get_aio_context(bs
);
1072 ret
= glfs_pwritev_async(s
->fd
, qiov
->iov
, qiov
->niov
, offset
, 0,
1073 gluster_finish_aiocb
, &acb
);
1075 ret
= glfs_preadv_async(s
->fd
, qiov
->iov
, qiov
->niov
, offset
, 0,
1076 gluster_finish_aiocb
, &acb
);
1083 qemu_coroutine_yield();
1087 static int qemu_gluster_truncate(BlockDriverState
*bs
, int64_t offset
)
1090 BDRVGlusterState
*s
= bs
->opaque
;
1092 ret
= glfs_ftruncate(s
->fd
, offset
);
1100 static coroutine_fn
int qemu_gluster_co_readv(BlockDriverState
*bs
,
1105 return qemu_gluster_co_rw(bs
, sector_num
, nb_sectors
, qiov
, 0);
1108 static coroutine_fn
int qemu_gluster_co_writev(BlockDriverState
*bs
,
1113 return qemu_gluster_co_rw(bs
, sector_num
, nb_sectors
, qiov
, 1);
1116 static void qemu_gluster_close(BlockDriverState
*bs
)
1118 BDRVGlusterState
*s
= bs
->opaque
;
1125 glfs_clear_preopened(s
->glfs
);
1128 static coroutine_fn
int qemu_gluster_co_flush_to_disk(BlockDriverState
*bs
)
1132 BDRVGlusterState
*s
= bs
->opaque
;
1136 acb
.coroutine
= qemu_coroutine_self();
1137 acb
.aio_context
= bdrv_get_aio_context(bs
);
1139 ret
= glfs_fsync_async(s
->fd
, gluster_finish_aiocb
, &acb
);
1145 qemu_coroutine_yield();
1154 /* Some versions of Gluster (3.5.6 -> 3.5.8?) will not retain its cache
1155 * after a fsync failure, so we have no way of allowing the guest to safely
1156 * continue. Gluster versions prior to 3.5.6 don't retain the cache
1157 * either, but will invalidate the fd on error, so this is again our only
1160 * The 'resync-failed-syncs-after-fsync' xlator option for the
1161 * write-behind cache will cause later gluster versions to retain its
1162 * cache after error, so long as the fd remains open. However, we
1163 * currently have no way of knowing if this option is supported.
1165 * TODO: Once gluster provides a way for us to determine if the option
1166 * is supported, bypass the closure and setting drv to NULL. */
1167 qemu_gluster_close(bs
);
1172 #ifdef CONFIG_GLUSTERFS_DISCARD
1173 static coroutine_fn
int qemu_gluster_co_pdiscard(BlockDriverState
*bs
,
1174 int64_t offset
, int size
)
1178 BDRVGlusterState
*s
= bs
->opaque
;
1182 acb
.coroutine
= qemu_coroutine_self();
1183 acb
.aio_context
= bdrv_get_aio_context(bs
);
1185 ret
= glfs_discard_async(s
->fd
, offset
, size
, gluster_finish_aiocb
, &acb
);
1190 qemu_coroutine_yield();
1195 static int64_t qemu_gluster_getlength(BlockDriverState
*bs
)
1197 BDRVGlusterState
*s
= bs
->opaque
;
1200 ret
= glfs_lseek(s
->fd
, 0, SEEK_END
);
1208 static int64_t qemu_gluster_allocated_file_size(BlockDriverState
*bs
)
1210 BDRVGlusterState
*s
= bs
->opaque
;
1214 ret
= glfs_fstat(s
->fd
, &st
);
1218 return st
.st_blocks
* 512;
1222 static int qemu_gluster_has_zero_init(BlockDriverState
*bs
)
1224 /* GlusterFS volume could be backed by a block device */
1229 * Find allocation range in @bs around offset @start.
1230 * May change underlying file descriptor's file offset.
1231 * If @start is not in a hole, store @start in @data, and the
1232 * beginning of the next hole in @hole, and return 0.
1233 * If @start is in a non-trailing hole, store @start in @hole and the
1234 * beginning of the next non-hole in @data, and return 0.
1235 * If @start is in a trailing hole or beyond EOF, return -ENXIO.
1236 * If we can't find out, return a negative errno other than -ENXIO.
1238 * (Shamefully copied from file-posix.c, only miniscule adaptions.)
1240 static int find_allocation(BlockDriverState
*bs
, off_t start
,
1241 off_t
*data
, off_t
*hole
)
1243 BDRVGlusterState
*s
= bs
->opaque
;
1245 if (!s
->supports_seek_data
) {
1249 #if defined SEEK_HOLE && defined SEEK_DATA
1254 * D1. offs == start: start is in data
1255 * D2. offs > start: start is in a hole, next data at offs
1256 * D3. offs < 0, errno = ENXIO: either start is in a trailing hole
1257 * or start is beyond EOF
1258 * If the latter happens, the file has been truncated behind
1259 * our back since we opened it. All bets are off then.
1260 * Treating like a trailing hole is simplest.
1261 * D4. offs < 0, errno != ENXIO: we learned nothing
1263 offs
= glfs_lseek(s
->fd
, start
, SEEK_DATA
);
1265 return -errno
; /* D3 or D4 */
1267 assert(offs
>= start
);
1270 /* D2: in hole, next data at offs */
1276 /* D1: in data, end not yet known */
1280 * H1. offs == start: start is in a hole
1281 * If this happens here, a hole has been dug behind our back
1282 * since the previous lseek().
1283 * H2. offs > start: either start is in data, next hole at offs,
1284 * or start is in trailing hole, EOF at offs
1285 * Linux treats trailing holes like any other hole: offs ==
1286 * start. Solaris seeks to EOF instead: offs > start (blech).
1287 * If that happens here, a hole has been dug behind our back
1288 * since the previous lseek().
1289 * H3. offs < 0, errno = ENXIO: start is beyond EOF
1290 * If this happens, the file has been truncated behind our
1291 * back since we opened it. Treat it like a trailing hole.
1292 * H4. offs < 0, errno != ENXIO: we learned nothing
1293 * Pretend we know nothing at all, i.e. "forget" about D1.
1295 offs
= glfs_lseek(s
->fd
, start
, SEEK_HOLE
);
1297 return -errno
; /* D1 and (H3 or H4) */
1299 assert(offs
>= start
);
1303 * D1 and H2: either in data, next hole at offs, or it was in
1304 * data but is now in a trailing hole. In the latter case,
1305 * all bets are off. Treating it as if it there was data all
1306 * the way to EOF is safe, so simply do that.
1322 * Returns the allocation status of the specified sectors.
1324 * If 'sector_num' is beyond the end of the disk image the return value is 0
1325 * and 'pnum' is set to 0.
1327 * 'pnum' is set to the number of sectors (including and immediately following
1328 * the specified sector) that are known to be in the same
1329 * allocated/unallocated state.
1331 * 'nb_sectors' is the max value 'pnum' should be set to. If nb_sectors goes
1332 * beyond the end of the disk image it will be clamped.
1334 * (Based on raw_co_get_block_status() from file-posix.c.)
1336 static int64_t coroutine_fn
qemu_gluster_co_get_block_status(
1337 BlockDriverState
*bs
, int64_t sector_num
, int nb_sectors
, int *pnum
,
1338 BlockDriverState
**file
)
1340 BDRVGlusterState
*s
= bs
->opaque
;
1341 off_t start
, data
= 0, hole
= 0;
1349 start
= sector_num
* BDRV_SECTOR_SIZE
;
1350 total_size
= bdrv_getlength(bs
);
1351 if (total_size
< 0) {
1353 } else if (start
>= total_size
) {
1356 } else if (start
+ nb_sectors
* BDRV_SECTOR_SIZE
> total_size
) {
1357 nb_sectors
= DIV_ROUND_UP(total_size
- start
, BDRV_SECTOR_SIZE
);
1360 ret
= find_allocation(bs
, start
, &data
, &hole
);
1361 if (ret
== -ENXIO
) {
1364 ret
= BDRV_BLOCK_ZERO
;
1365 } else if (ret
< 0) {
1366 /* No info available, so pretend there are no holes */
1368 ret
= BDRV_BLOCK_DATA
;
1369 } else if (data
== start
) {
1370 /* On a data extent, compute sectors to the end of the extent,
1371 * possibly including a partial sector at EOF. */
1372 *pnum
= MIN(nb_sectors
, DIV_ROUND_UP(hole
- start
, BDRV_SECTOR_SIZE
));
1373 ret
= BDRV_BLOCK_DATA
;
1375 /* On a hole, compute sectors to the beginning of the next extent. */
1376 assert(hole
== start
);
1377 *pnum
= MIN(nb_sectors
, (data
- start
) / BDRV_SECTOR_SIZE
);
1378 ret
= BDRV_BLOCK_ZERO
;
1383 return ret
| BDRV_BLOCK_OFFSET_VALID
| start
;
1387 static BlockDriver bdrv_gluster
= {
1388 .format_name
= "gluster",
1389 .protocol_name
= "gluster",
1390 .instance_size
= sizeof(BDRVGlusterState
),
1391 .bdrv_needs_filename
= false,
1392 .bdrv_file_open
= qemu_gluster_open
,
1393 .bdrv_reopen_prepare
= qemu_gluster_reopen_prepare
,
1394 .bdrv_reopen_commit
= qemu_gluster_reopen_commit
,
1395 .bdrv_reopen_abort
= qemu_gluster_reopen_abort
,
1396 .bdrv_close
= qemu_gluster_close
,
1397 .bdrv_create
= qemu_gluster_create
,
1398 .bdrv_getlength
= qemu_gluster_getlength
,
1399 .bdrv_get_allocated_file_size
= qemu_gluster_allocated_file_size
,
1400 .bdrv_truncate
= qemu_gluster_truncate
,
1401 .bdrv_co_readv
= qemu_gluster_co_readv
,
1402 .bdrv_co_writev
= qemu_gluster_co_writev
,
1403 .bdrv_co_flush_to_disk
= qemu_gluster_co_flush_to_disk
,
1404 .bdrv_has_zero_init
= qemu_gluster_has_zero_init
,
1405 #ifdef CONFIG_GLUSTERFS_DISCARD
1406 .bdrv_co_pdiscard
= qemu_gluster_co_pdiscard
,
1408 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1409 .bdrv_co_pwrite_zeroes
= qemu_gluster_co_pwrite_zeroes
,
1411 .bdrv_co_get_block_status
= qemu_gluster_co_get_block_status
,
1412 .create_opts
= &qemu_gluster_create_opts
,
1415 static BlockDriver bdrv_gluster_tcp
= {
1416 .format_name
= "gluster",
1417 .protocol_name
= "gluster+tcp",
1418 .instance_size
= sizeof(BDRVGlusterState
),
1419 .bdrv_needs_filename
= false,
1420 .bdrv_file_open
= qemu_gluster_open
,
1421 .bdrv_reopen_prepare
= qemu_gluster_reopen_prepare
,
1422 .bdrv_reopen_commit
= qemu_gluster_reopen_commit
,
1423 .bdrv_reopen_abort
= qemu_gluster_reopen_abort
,
1424 .bdrv_close
= qemu_gluster_close
,
1425 .bdrv_create
= qemu_gluster_create
,
1426 .bdrv_getlength
= qemu_gluster_getlength
,
1427 .bdrv_get_allocated_file_size
= qemu_gluster_allocated_file_size
,
1428 .bdrv_truncate
= qemu_gluster_truncate
,
1429 .bdrv_co_readv
= qemu_gluster_co_readv
,
1430 .bdrv_co_writev
= qemu_gluster_co_writev
,
1431 .bdrv_co_flush_to_disk
= qemu_gluster_co_flush_to_disk
,
1432 .bdrv_has_zero_init
= qemu_gluster_has_zero_init
,
1433 #ifdef CONFIG_GLUSTERFS_DISCARD
1434 .bdrv_co_pdiscard
= qemu_gluster_co_pdiscard
,
1436 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1437 .bdrv_co_pwrite_zeroes
= qemu_gluster_co_pwrite_zeroes
,
1439 .bdrv_co_get_block_status
= qemu_gluster_co_get_block_status
,
1440 .create_opts
= &qemu_gluster_create_opts
,
1443 static BlockDriver bdrv_gluster_unix
= {
1444 .format_name
= "gluster",
1445 .protocol_name
= "gluster+unix",
1446 .instance_size
= sizeof(BDRVGlusterState
),
1447 .bdrv_needs_filename
= true,
1448 .bdrv_file_open
= qemu_gluster_open
,
1449 .bdrv_reopen_prepare
= qemu_gluster_reopen_prepare
,
1450 .bdrv_reopen_commit
= qemu_gluster_reopen_commit
,
1451 .bdrv_reopen_abort
= qemu_gluster_reopen_abort
,
1452 .bdrv_close
= qemu_gluster_close
,
1453 .bdrv_create
= qemu_gluster_create
,
1454 .bdrv_getlength
= qemu_gluster_getlength
,
1455 .bdrv_get_allocated_file_size
= qemu_gluster_allocated_file_size
,
1456 .bdrv_truncate
= qemu_gluster_truncate
,
1457 .bdrv_co_readv
= qemu_gluster_co_readv
,
1458 .bdrv_co_writev
= qemu_gluster_co_writev
,
1459 .bdrv_co_flush_to_disk
= qemu_gluster_co_flush_to_disk
,
1460 .bdrv_has_zero_init
= qemu_gluster_has_zero_init
,
1461 #ifdef CONFIG_GLUSTERFS_DISCARD
1462 .bdrv_co_pdiscard
= qemu_gluster_co_pdiscard
,
1464 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1465 .bdrv_co_pwrite_zeroes
= qemu_gluster_co_pwrite_zeroes
,
1467 .bdrv_co_get_block_status
= qemu_gluster_co_get_block_status
,
1468 .create_opts
= &qemu_gluster_create_opts
,
1471 /* rdma is deprecated (actually never supported for volfile fetch).
1472 * Let's maintain it for the protocol compatibility, to make sure things
1473 * won't break immediately. For now, gluster+rdma will fall back to gluster+tcp
1474 * protocol with a warning.
1475 * TODO: remove gluster+rdma interface support
1477 static BlockDriver bdrv_gluster_rdma
= {
1478 .format_name
= "gluster",
1479 .protocol_name
= "gluster+rdma",
1480 .instance_size
= sizeof(BDRVGlusterState
),
1481 .bdrv_needs_filename
= true,
1482 .bdrv_file_open
= qemu_gluster_open
,
1483 .bdrv_reopen_prepare
= qemu_gluster_reopen_prepare
,
1484 .bdrv_reopen_commit
= qemu_gluster_reopen_commit
,
1485 .bdrv_reopen_abort
= qemu_gluster_reopen_abort
,
1486 .bdrv_close
= qemu_gluster_close
,
1487 .bdrv_create
= qemu_gluster_create
,
1488 .bdrv_getlength
= qemu_gluster_getlength
,
1489 .bdrv_get_allocated_file_size
= qemu_gluster_allocated_file_size
,
1490 .bdrv_truncate
= qemu_gluster_truncate
,
1491 .bdrv_co_readv
= qemu_gluster_co_readv
,
1492 .bdrv_co_writev
= qemu_gluster_co_writev
,
1493 .bdrv_co_flush_to_disk
= qemu_gluster_co_flush_to_disk
,
1494 .bdrv_has_zero_init
= qemu_gluster_has_zero_init
,
1495 #ifdef CONFIG_GLUSTERFS_DISCARD
1496 .bdrv_co_pdiscard
= qemu_gluster_co_pdiscard
,
1498 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1499 .bdrv_co_pwrite_zeroes
= qemu_gluster_co_pwrite_zeroes
,
1501 .bdrv_co_get_block_status
= qemu_gluster_co_get_block_status
,
1502 .create_opts
= &qemu_gluster_create_opts
,
1505 static void bdrv_gluster_init(void)
1507 bdrv_register(&bdrv_gluster_rdma
);
1508 bdrv_register(&bdrv_gluster_unix
);
1509 bdrv_register(&bdrv_gluster_tcp
);
1510 bdrv_register(&bdrv_gluster
);
1513 block_init(bdrv_gluster_init
);