2 * GlusterFS backend for QEMU
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.
11 #include "qemu/osdep.h"
12 #include <glusterfs/api/glfs.h>
13 #include "block/block_int.h"
14 #include "block/qdict.h"
15 #include "qapi/error.h"
16 #include "qapi/qmp/qdict.h"
17 #include "qapi/qmp/qerror.h"
19 #include "qemu/error-report.h"
20 #include "qemu/option.h"
21 #include "qemu/cutils.h"
23 #define GLUSTER_OPT_FILENAME "filename"
24 #define GLUSTER_OPT_VOLUME "volume"
25 #define GLUSTER_OPT_PATH "path"
26 #define GLUSTER_OPT_TYPE "type"
27 #define GLUSTER_OPT_SERVER_PATTERN "server."
28 #define GLUSTER_OPT_HOST "host"
29 #define GLUSTER_OPT_PORT "port"
30 #define GLUSTER_OPT_TO "to"
31 #define GLUSTER_OPT_IPV4 "ipv4"
32 #define GLUSTER_OPT_IPV6 "ipv6"
33 #define GLUSTER_OPT_SOCKET "socket"
34 #define GLUSTER_OPT_DEBUG "debug"
35 #define GLUSTER_DEFAULT_PORT 24007
36 #define GLUSTER_DEBUG_DEFAULT 4
37 #define GLUSTER_DEBUG_MAX 9
38 #define GLUSTER_OPT_LOGFILE "logfile"
39 #define GLUSTER_LOGFILE_DEFAULT "-" /* handled in libgfapi as /dev/stderr */
41 #define GERR_INDEX_HINT "hint: check in 'server' array index '%d'\n"
43 typedef struct GlusterAIOCB {
47 AioContext *aio_context;
50 typedef struct BDRVGlusterState {
54 bool supports_seek_data;
58 typedef struct BDRVGlusterReopenState {
61 } BDRVGlusterReopenState;
64 typedef struct GlfsPreopened {
70 typedef struct ListElement {
71 QLIST_ENTRY(ListElement) list;
75 static QLIST_HEAD(, ListElement) glfs_list;
77 static QemuOptsList qemu_gluster_create_opts = {
78 .name = "qemu-gluster-create-opts",
79 .head = QTAILQ_HEAD_INITIALIZER(qemu_gluster_create_opts.head),
82 .name = BLOCK_OPT_SIZE,
83 .type = QEMU_OPT_SIZE,
84 .help = "Virtual disk size"
87 .name = BLOCK_OPT_PREALLOC,
88 .type = QEMU_OPT_STRING,
89 .help = "Preallocation mode (allowed values: off, full)"
92 .name = GLUSTER_OPT_DEBUG,
93 .type = QEMU_OPT_NUMBER,
94 .help = "Gluster log level, valid range is 0-9",
97 .name = GLUSTER_OPT_LOGFILE,
98 .type = QEMU_OPT_STRING,
99 .help = "Logfile path of libgfapi",
101 { /* end of list */ }
105 static QemuOptsList runtime_opts = {
107 .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
110 .name = GLUSTER_OPT_FILENAME,
111 .type = QEMU_OPT_STRING,
112 .help = "URL to the gluster image",
115 .name = GLUSTER_OPT_DEBUG,
116 .type = QEMU_OPT_NUMBER,
117 .help = "Gluster log level, valid range is 0-9",
120 .name = GLUSTER_OPT_LOGFILE,
121 .type = QEMU_OPT_STRING,
122 .help = "Logfile path of libgfapi",
124 { /* end of list */ }
128 static QemuOptsList runtime_json_opts = {
129 .name = "gluster_json",
130 .head = QTAILQ_HEAD_INITIALIZER(runtime_json_opts.head),
133 .name = GLUSTER_OPT_VOLUME,
134 .type = QEMU_OPT_STRING,
135 .help = "name of gluster volume where VM image resides",
138 .name = GLUSTER_OPT_PATH,
139 .type = QEMU_OPT_STRING,
140 .help = "absolute path to image file in gluster volume",
143 .name = GLUSTER_OPT_DEBUG,
144 .type = QEMU_OPT_NUMBER,
145 .help = "Gluster log level, valid range is 0-9",
147 { /* end of list */ }
151 static QemuOptsList runtime_type_opts = {
152 .name = "gluster_type",
153 .head = QTAILQ_HEAD_INITIALIZER(runtime_type_opts.head),
156 .name = GLUSTER_OPT_TYPE,
157 .type = QEMU_OPT_STRING,
160 { /* end of list */ }
164 static QemuOptsList runtime_unix_opts = {
165 .name = "gluster_unix",
166 .head = QTAILQ_HEAD_INITIALIZER(runtime_unix_opts.head),
169 .name = GLUSTER_OPT_SOCKET,
170 .type = QEMU_OPT_STRING,
171 .help = "socket file path (legacy)",
174 .name = GLUSTER_OPT_PATH,
175 .type = QEMU_OPT_STRING,
176 .help = "socket file path (QAPI)",
178 { /* end of list */ }
182 static QemuOptsList runtime_inet_opts = {
183 .name = "gluster_inet",
184 .head = QTAILQ_HEAD_INITIALIZER(runtime_inet_opts.head),
187 .name = GLUSTER_OPT_TYPE,
188 .type = QEMU_OPT_STRING,
192 .name = GLUSTER_OPT_HOST,
193 .type = QEMU_OPT_STRING,
194 .help = "host address (hostname/ipv4/ipv6 addresses)",
197 .name = GLUSTER_OPT_PORT,
198 .type = QEMU_OPT_STRING,
199 .help = "port number on which glusterd is listening (default 24007)",
203 .type = QEMU_OPT_NUMBER,
204 .help = "max port number, not supported by gluster",
208 .type = QEMU_OPT_BOOL,
209 .help = "ipv4 bool value, not supported by gluster",
213 .type = QEMU_OPT_BOOL,
214 .help = "ipv6 bool value, not supported by gluster",
216 { /* end of list */ }
220 static void glfs_set_preopened(const char *volume, glfs_t *fs)
222 ListElement *entry = NULL;
224 entry = g_new(ListElement, 1);
226 entry->saved.volume = g_strdup(volume);
228 entry->saved.fs = fs;
229 entry->saved.ref = 1;
231 QLIST_INSERT_HEAD(&glfs_list, entry, list);
234 static glfs_t *glfs_find_preopened(const char *volume)
236 ListElement *entry = NULL;
238 QLIST_FOREACH(entry, &glfs_list, list) {
239 if (strcmp(entry->saved.volume, volume) == 0) {
241 return entry->saved.fs;
248 static void glfs_clear_preopened(glfs_t *fs)
250 ListElement *entry = NULL;
257 QLIST_FOREACH_SAFE(entry, &glfs_list, list, next) {
258 if (entry->saved.fs == fs) {
259 if (--entry->saved.ref) {
263 QLIST_REMOVE(entry, list);
265 glfs_fini(entry->saved.fs);
266 g_free(entry->saved.volume);
272 static int parse_volume_options(BlockdevOptionsGluster *gconf, char *path)
281 p = q = path + strspn(path, "/");
282 p += strcspn(p, "/");
286 gconf->volume = g_strndup(q, p - q);
293 gconf->path = g_strdup(p);
298 * file=gluster[+transport]://[host[:port]]/volume/path[?socket=...]
300 * 'gluster' is the protocol.
302 * 'transport' specifies the transport type used to connect to gluster
303 * management daemon (glusterd). Valid transport types are
304 * tcp or unix. If a transport type isn't specified, then tcp type is assumed.
306 * 'host' specifies the host where the volume file specification for
307 * the given volume resides. This can be either hostname or ipv4 address.
308 * If transport type is 'unix', then 'host' field should not be specified.
309 * The 'socket' field needs to be populated with the path to unix domain
312 * 'port' is the port number on which glusterd is listening. This is optional
313 * and if not specified, QEMU will send 0 which will make gluster to use the
314 * default port. If the transport type is unix, then 'port' should not be
317 * 'volume' is the name of the gluster volume which contains the VM image.
319 * 'path' is the path to the actual VM image that resides on gluster volume.
323 * file=gluster://1.2.3.4/testvol/a.img
324 * file=gluster+tcp://1.2.3.4/testvol/a.img
325 * file=gluster+tcp://1.2.3.4:24007/testvol/dir/a.img
326 * file=gluster+tcp://host.domain.com:24007/testvol/dir/a.img
327 * file=gluster+unix:///testvol/dir/a.img?socket=/tmp/glusterd.socket
329 static int qemu_gluster_parse_uri(BlockdevOptionsGluster *gconf,
330 const char *filename)
332 SocketAddress *gsconf;
334 QueryParams *qp = NULL;
335 bool is_unix = false;
338 uri = uri_parse(filename);
343 gconf->server = g_new0(SocketAddressList, 1);
344 gconf->server->value = gsconf = g_new0(SocketAddress, 1);
347 if (!uri->scheme || !strcmp(uri->scheme, "gluster")) {
348 gsconf->type = SOCKET_ADDRESS_TYPE_INET;
349 } else if (!strcmp(uri->scheme, "gluster+tcp")) {
350 gsconf->type = SOCKET_ADDRESS_TYPE_INET;
351 } else if (!strcmp(uri->scheme, "gluster+unix")) {
352 gsconf->type = SOCKET_ADDRESS_TYPE_UNIX;
354 } else if (!strcmp(uri->scheme, "gluster+rdma")) {
355 gsconf->type = SOCKET_ADDRESS_TYPE_INET;
356 warn_report("rdma feature is not supported, falling back to tcp");
362 ret = parse_volume_options(gconf, uri->path);
367 qp = query_params_parse(uri->query);
368 if (qp->n > 1 || (is_unix && !qp->n) || (!is_unix && qp->n)) {
374 if (uri->server || uri->port) {
378 if (strcmp(qp->p[0].name, "socket")) {
382 gsconf->u.q_unix.path = g_strdup(qp->p[0].value);
384 gsconf->u.inet.host = g_strdup(uri->server ? uri->server : "localhost");
386 gsconf->u.inet.port = g_strdup_printf("%d", uri->port);
388 gsconf->u.inet.port = g_strdup_printf("%d", GLUSTER_DEFAULT_PORT);
394 query_params_free(qp);
400 static struct glfs *qemu_gluster_glfs_init(BlockdevOptionsGluster *gconf,
406 SocketAddressList *server;
407 unsigned long long port;
409 glfs = glfs_find_preopened(gconf->volume);
414 glfs = glfs_new(gconf->volume);
419 glfs_set_preopened(gconf->volume, glfs);
421 for (server = gconf->server; server; server = server->next) {
422 switch (server->value->type) {
423 case SOCKET_ADDRESS_TYPE_UNIX:
424 ret = glfs_set_volfile_server(glfs, "unix",
425 server->value->u.q_unix.path, 0);
427 case SOCKET_ADDRESS_TYPE_INET:
428 if (parse_uint_full(server->value->u.inet.port, &port, 10) < 0 ||
430 error_setg(errp, "'%s' is not a valid port number",
431 server->value->u.inet.port);
435 ret = glfs_set_volfile_server(glfs, "tcp",
436 server->value->u.inet.host,
439 case SOCKET_ADDRESS_TYPE_VSOCK:
440 case SOCKET_ADDRESS_TYPE_FD:
450 ret = glfs_set_logging(glfs, gconf->logfile, gconf->debug);
455 ret = glfs_init(glfs);
457 error_setg(errp, "Gluster connection for volume %s, path %s failed"
458 " to connect", gconf->volume, gconf->path);
459 for (server = gconf->server; server; server = server->next) {
460 if (server->value->type == SOCKET_ADDRESS_TYPE_UNIX) {
461 error_append_hint(errp, "hint: failed on socket %s ",
462 server->value->u.q_unix.path);
464 error_append_hint(errp, "hint: failed on host %s and port %s ",
465 server->value->u.inet.host,
466 server->value->u.inet.port);
470 error_append_hint(errp, "Please refer to gluster logs for more info\n");
472 /* glfs_init sometimes doesn't set errno although docs suggest that */
484 glfs_clear_preopened(glfs);
491 * Convert the json formatted command line into qapi.
493 static int qemu_gluster_parse_json(BlockdevOptionsGluster *gconf,
494 QDict *options, Error **errp)
497 SocketAddress *gsconf = NULL;
498 SocketAddressList *curr = NULL;
499 QDict *backing_options = NULL;
500 Error *local_err = NULL;
503 int i, type, num_servers;
505 /* create opts info from runtime_json_opts list */
506 opts = qemu_opts_create(&runtime_json_opts, NULL, 0, &error_abort);
507 qemu_opts_absorb_qdict(opts, options, &local_err);
512 num_servers = qdict_array_entries(options, GLUSTER_OPT_SERVER_PATTERN);
513 if (num_servers < 1) {
514 error_setg(&local_err, QERR_MISSING_PARAMETER, "server");
518 ptr = qemu_opt_get(opts, GLUSTER_OPT_VOLUME);
520 error_setg(&local_err, QERR_MISSING_PARAMETER, GLUSTER_OPT_VOLUME);
523 gconf->volume = g_strdup(ptr);
525 ptr = qemu_opt_get(opts, GLUSTER_OPT_PATH);
527 error_setg(&local_err, QERR_MISSING_PARAMETER, GLUSTER_OPT_PATH);
530 gconf->path = g_strdup(ptr);
533 for (i = 0; i < num_servers; i++) {
534 str = g_strdup_printf(GLUSTER_OPT_SERVER_PATTERN"%d.", i);
535 qdict_extract_subqdict(options, &backing_options, str);
537 /* create opts info from runtime_type_opts list */
538 opts = qemu_opts_create(&runtime_type_opts, NULL, 0, &error_abort);
539 qemu_opts_absorb_qdict(opts, backing_options, &local_err);
544 ptr = qemu_opt_get(opts, GLUSTER_OPT_TYPE);
546 error_setg(&local_err, QERR_MISSING_PARAMETER, GLUSTER_OPT_TYPE);
547 error_append_hint(&local_err, GERR_INDEX_HINT, i);
551 gsconf = g_new0(SocketAddress, 1);
552 if (!strcmp(ptr, "tcp")) {
553 ptr = "inet"; /* accept legacy "tcp" */
555 type = qapi_enum_parse(&SocketAddressType_lookup, ptr, -1, NULL);
556 if (type != SOCKET_ADDRESS_TYPE_INET
557 && type != SOCKET_ADDRESS_TYPE_UNIX) {
558 error_setg(&local_err,
559 "Parameter '%s' may be 'inet' or 'unix'",
561 error_append_hint(&local_err, GERR_INDEX_HINT, i);
567 if (gsconf->type == SOCKET_ADDRESS_TYPE_INET) {
568 /* create opts info from runtime_inet_opts list */
569 opts = qemu_opts_create(&runtime_inet_opts, NULL, 0, &error_abort);
570 qemu_opts_absorb_qdict(opts, backing_options, &local_err);
575 ptr = qemu_opt_get(opts, GLUSTER_OPT_HOST);
577 error_setg(&local_err, QERR_MISSING_PARAMETER,
579 error_append_hint(&local_err, GERR_INDEX_HINT, i);
582 gsconf->u.inet.host = g_strdup(ptr);
583 ptr = qemu_opt_get(opts, GLUSTER_OPT_PORT);
585 error_setg(&local_err, QERR_MISSING_PARAMETER,
587 error_append_hint(&local_err, GERR_INDEX_HINT, i);
590 gsconf->u.inet.port = g_strdup(ptr);
592 /* defend for unsupported fields in InetSocketAddress,
593 * i.e. @ipv4, @ipv6 and @to
595 ptr = qemu_opt_get(opts, GLUSTER_OPT_TO);
597 gsconf->u.inet.has_to = true;
599 ptr = qemu_opt_get(opts, GLUSTER_OPT_IPV4);
601 gsconf->u.inet.has_ipv4 = true;
603 ptr = qemu_opt_get(opts, GLUSTER_OPT_IPV6);
605 gsconf->u.inet.has_ipv6 = true;
607 if (gsconf->u.inet.has_to) {
608 error_setg(&local_err, "Parameter 'to' not supported");
611 if (gsconf->u.inet.has_ipv4 || gsconf->u.inet.has_ipv6) {
612 error_setg(&local_err, "Parameters 'ipv4/ipv6' not supported");
617 /* create opts info from runtime_unix_opts list */
618 opts = qemu_opts_create(&runtime_unix_opts, NULL, 0, &error_abort);
619 qemu_opts_absorb_qdict(opts, backing_options, &local_err);
624 ptr = qemu_opt_get(opts, GLUSTER_OPT_PATH);
626 ptr = qemu_opt_get(opts, GLUSTER_OPT_SOCKET);
627 } else if (qemu_opt_get(opts, GLUSTER_OPT_SOCKET)) {
628 error_setg(&local_err,
629 "Conflicting parameters 'path' and 'socket'");
630 error_append_hint(&local_err, GERR_INDEX_HINT, i);
634 error_setg(&local_err, QERR_MISSING_PARAMETER,
636 error_append_hint(&local_err, GERR_INDEX_HINT, i);
639 gsconf->u.q_unix.path = g_strdup(ptr);
643 if (gconf->server == NULL) {
644 gconf->server = g_new0(SocketAddressList, 1);
645 gconf->server->value = gsconf;
646 curr = gconf->server;
648 curr->next = g_new0(SocketAddressList, 1);
649 curr->next->value = gsconf;
654 qobject_unref(backing_options);
655 backing_options = NULL;
663 error_propagate(errp, local_err);
664 qapi_free_SocketAddress(gsconf);
667 qobject_unref(backing_options);
672 /* Converts options given in @filename and the @options QDict into the QAPI
674 static int qemu_gluster_parse(BlockdevOptionsGluster *gconf,
675 const char *filename,
676 QDict *options, Error **errp)
680 ret = qemu_gluster_parse_uri(gconf, filename);
682 error_setg(errp, "invalid URI %s", filename);
683 error_append_hint(errp, "Usage: file=gluster[+transport]://"
684 "[host[:port]]volume/path[?socket=...]"
686 "[,file.logfile=/path/filename.log]\n");
690 ret = qemu_gluster_parse_json(gconf, options, errp);
692 error_append_hint(errp, "Usage: "
693 "-drive driver=qcow2,file.driver=gluster,"
694 "file.volume=testvol,file.path=/path/a.qcow2"
696 "[,file.logfile=/path/filename.log],"
697 "file.server.0.type=inet,"
698 "file.server.0.host=1.2.3.4,"
699 "file.server.0.port=24007,"
700 "file.server.1.transport=unix,"
701 "file.server.1.path=/var/run/glusterd.socket ..."
710 static struct glfs *qemu_gluster_init(BlockdevOptionsGluster *gconf,
711 const char *filename,
712 QDict *options, Error **errp)
716 ret = qemu_gluster_parse(gconf, filename, options, errp);
722 return qemu_gluster_glfs_init(gconf, errp);
726 * AIO callback routine called from GlusterFS thread.
728 static void gluster_finish_aiocb(struct glfs_fd *fd, ssize_t ret, void *arg)
730 GlusterAIOCB *acb = (GlusterAIOCB *)arg;
732 if (!ret || ret == acb->size) {
733 acb->ret = 0; /* Success */
734 } else if (ret < 0) {
735 acb->ret = -errno; /* Read/Write failed */
737 acb->ret = -EIO; /* Partial read/write - fail it */
740 aio_co_schedule(acb->aio_context, acb->coroutine);
743 static void qemu_gluster_parse_flags(int bdrv_flags, int *open_flags)
745 assert(open_flags != NULL);
747 *open_flags |= O_BINARY;
749 if (bdrv_flags & BDRV_O_RDWR) {
750 *open_flags |= O_RDWR;
752 *open_flags |= O_RDONLY;
755 if ((bdrv_flags & BDRV_O_NOCACHE)) {
756 *open_flags |= O_DIRECT;
761 * Do SEEK_DATA/HOLE to detect if it is functional. Older broken versions of
762 * gfapi incorrectly return the current offset when SEEK_DATA/HOLE is used.
763 * - Corrected versions return -1 and set errno to EINVAL.
764 * - Versions that support SEEK_DATA/HOLE correctly, will return -1 and set
765 * errno to ENXIO when SEEK_DATA is called with a position of EOF.
767 static bool qemu_gluster_test_seek(struct glfs_fd *fd)
771 #if defined SEEK_HOLE && defined SEEK_DATA
774 eof = glfs_lseek(fd, 0, SEEK_END);
776 /* this should never occur */
780 /* this should always fail with ENXIO if SEEK_DATA is supported */
781 ret = glfs_lseek(fd, eof, SEEK_DATA);
784 return (ret < 0) && (errno == ENXIO);
787 static int qemu_gluster_open(BlockDriverState *bs, QDict *options,
788 int bdrv_flags, Error **errp)
790 BDRVGlusterState *s = bs->opaque;
793 BlockdevOptionsGluster *gconf = NULL;
795 Error *local_err = NULL;
796 const char *filename, *logfile;
798 opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
799 qemu_opts_absorb_qdict(opts, options, &local_err);
801 error_propagate(errp, local_err);
806 filename = qemu_opt_get(opts, GLUSTER_OPT_FILENAME);
808 s->debug = qemu_opt_get_number(opts, GLUSTER_OPT_DEBUG,
809 GLUSTER_DEBUG_DEFAULT);
812 } else if (s->debug > GLUSTER_DEBUG_MAX) {
813 s->debug = GLUSTER_DEBUG_MAX;
816 gconf = g_new0(BlockdevOptionsGluster, 1);
817 gconf->debug = s->debug;
818 gconf->has_debug = true;
820 logfile = qemu_opt_get(opts, GLUSTER_OPT_LOGFILE);
821 s->logfile = g_strdup(logfile ? logfile : GLUSTER_LOGFILE_DEFAULT);
823 gconf->logfile = g_strdup(s->logfile);
824 gconf->has_logfile = true;
826 s->glfs = qemu_gluster_init(gconf, filename, options, errp);
832 #ifdef CONFIG_GLUSTERFS_XLATOR_OPT
833 /* Without this, if fsync fails for a recoverable reason (for instance,
834 * ENOSPC), gluster will dump its cache, preventing retries. This means
835 * almost certain data loss. Not all gluster versions support the
836 * 'resync-failed-syncs-after-fsync' key value, but there is no way to
837 * discover during runtime if it is supported (this api returns success for
838 * unknown key/value pairs) */
839 ret = glfs_set_xlator_option(s->glfs, "*-write-behind",
840 "resync-failed-syncs-after-fsync",
843 error_setg_errno(errp, errno, "Unable to set xlator key/value pair");
849 qemu_gluster_parse_flags(bdrv_flags, &open_flags);
851 s->fd = glfs_open(s->glfs, gconf->path, open_flags);
852 ret = s->fd ? 0 : -errno;
854 if (ret == -EACCES || ret == -EROFS) {
855 /* Try to degrade to read-only, but if it doesn't work, still use the
856 * normal error message. */
857 if (bdrv_apply_auto_read_only(bs, NULL, NULL) == 0) {
858 open_flags = (open_flags & ~O_RDWR) | O_RDONLY;
859 s->fd = glfs_open(s->glfs, gconf->path, open_flags);
860 ret = s->fd ? 0 : -errno;
864 s->supports_seek_data = qemu_gluster_test_seek(s->fd);
868 qapi_free_BlockdevOptionsGluster(gconf);
877 glfs_clear_preopened(s->glfs);
882 static int qemu_gluster_reopen_prepare(BDRVReopenState *state,
883 BlockReopenQueue *queue, Error **errp)
887 BDRVGlusterReopenState *reop_s;
888 BlockdevOptionsGluster *gconf;
891 assert(state != NULL);
892 assert(state->bs != NULL);
894 s = state->bs->opaque;
896 state->opaque = g_new0(BDRVGlusterReopenState, 1);
897 reop_s = state->opaque;
899 qemu_gluster_parse_flags(state->flags, &open_flags);
901 gconf = g_new0(BlockdevOptionsGluster, 1);
902 gconf->debug = s->debug;
903 gconf->has_debug = true;
904 gconf->logfile = g_strdup(s->logfile);
905 gconf->has_logfile = true;
906 reop_s->glfs = qemu_gluster_init(gconf, state->bs->filename, NULL, errp);
907 if (reop_s->glfs == NULL) {
912 #ifdef CONFIG_GLUSTERFS_XLATOR_OPT
913 ret = glfs_set_xlator_option(reop_s->glfs, "*-write-behind",
914 "resync-failed-syncs-after-fsync", "on");
916 error_setg_errno(errp, errno, "Unable to set xlator key/value pair");
922 reop_s->fd = glfs_open(reop_s->glfs, gconf->path, open_flags);
923 if (reop_s->fd == NULL) {
924 /* reops->glfs will be cleaned up in _abort */
930 /* state->opaque will be freed in either the _abort or _commit */
931 qapi_free_BlockdevOptionsGluster(gconf);
935 static void qemu_gluster_reopen_commit(BDRVReopenState *state)
937 BDRVGlusterReopenState *reop_s = state->opaque;
938 BDRVGlusterState *s = state->bs->opaque;
946 glfs_clear_preopened(s->glfs);
948 /* use the newly opened image / connection */
950 s->glfs = reop_s->glfs;
952 g_free(state->opaque);
953 state->opaque = NULL;
959 static void qemu_gluster_reopen_abort(BDRVReopenState *state)
961 BDRVGlusterReopenState *reop_s = state->opaque;
963 if (reop_s == NULL) {
968 glfs_close(reop_s->fd);
971 glfs_clear_preopened(reop_s->glfs);
973 g_free(state->opaque);
974 state->opaque = NULL;
979 #ifdef CONFIG_GLUSTERFS_ZEROFILL
980 static coroutine_fn int qemu_gluster_co_pwrite_zeroes(BlockDriverState *bs,
983 BdrvRequestFlags flags)
987 BDRVGlusterState *s = bs->opaque;
991 acb.coroutine = qemu_coroutine_self();
992 acb.aio_context = bdrv_get_aio_context(bs);
994 ret = glfs_zerofill_async(s->fd, offset, size, gluster_finish_aiocb, &acb);
999 qemu_coroutine_yield();
1004 static int qemu_gluster_do_truncate(struct glfs_fd *fd, int64_t offset,
1005 PreallocMode prealloc, Error **errp)
1007 int64_t current_length;
1009 current_length = glfs_lseek(fd, 0, SEEK_END);
1010 if (current_length < 0) {
1011 error_setg_errno(errp, errno, "Failed to determine current size");
1015 if (current_length > offset && prealloc != PREALLOC_MODE_OFF) {
1016 error_setg(errp, "Cannot use preallocation for shrinking files");
1020 if (current_length == offset) {
1025 #ifdef CONFIG_GLUSTERFS_FALLOCATE
1026 case PREALLOC_MODE_FALLOC:
1027 if (glfs_fallocate(fd, 0, current_length, offset - current_length)) {
1028 error_setg_errno(errp, errno, "Could not preallocate data");
1032 #endif /* CONFIG_GLUSTERFS_FALLOCATE */
1033 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1034 case PREALLOC_MODE_FULL:
1035 if (glfs_ftruncate(fd, offset)) {
1036 error_setg_errno(errp, errno, "Could not resize file");
1039 if (glfs_zerofill(fd, current_length, offset - current_length)) {
1040 error_setg_errno(errp, errno, "Could not zerofill the new area");
1044 #endif /* CONFIG_GLUSTERFS_ZEROFILL */
1045 case PREALLOC_MODE_OFF:
1046 if (glfs_ftruncate(fd, offset)) {
1047 error_setg_errno(errp, errno, "Could not resize file");
1052 error_setg(errp, "Unsupported preallocation mode: %s",
1053 PreallocMode_str(prealloc));
1060 static int qemu_gluster_co_create(BlockdevCreateOptions *options,
1063 BlockdevCreateOptionsGluster *opts = &options->u.gluster;
1065 struct glfs_fd *fd = NULL;
1068 assert(options->driver == BLOCKDEV_DRIVER_GLUSTER);
1070 glfs = qemu_gluster_glfs_init(opts->location, errp);
1076 fd = glfs_creat(glfs, opts->location->path,
1077 O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, S_IRUSR | S_IWUSR);
1083 ret = qemu_gluster_do_truncate(fd, opts->size, opts->preallocation, errp);
1087 if (glfs_close(fd) != 0 && ret == 0) {
1091 glfs_clear_preopened(glfs);
1095 static int coroutine_fn qemu_gluster_co_create_opts(const char *filename,
1099 BlockdevCreateOptions *options;
1100 BlockdevCreateOptionsGluster *gopts;
1101 BlockdevOptionsGluster *gconf;
1103 Error *local_err = NULL;
1106 options = g_new0(BlockdevCreateOptions, 1);
1107 options->driver = BLOCKDEV_DRIVER_GLUSTER;
1108 gopts = &options->u.gluster;
1110 gconf = g_new0(BlockdevOptionsGluster, 1);
1111 gopts->location = gconf;
1113 gopts->size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
1116 tmp = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
1117 gopts->preallocation = qapi_enum_parse(&PreallocMode_lookup, tmp,
1118 PREALLOC_MODE_OFF, &local_err);
1121 error_propagate(errp, local_err);
1126 gconf->debug = qemu_opt_get_number_del(opts, GLUSTER_OPT_DEBUG,
1127 GLUSTER_DEBUG_DEFAULT);
1128 if (gconf->debug < 0) {
1130 } else if (gconf->debug > GLUSTER_DEBUG_MAX) {
1131 gconf->debug = GLUSTER_DEBUG_MAX;
1133 gconf->has_debug = true;
1135 gconf->logfile = qemu_opt_get_del(opts, GLUSTER_OPT_LOGFILE);
1136 if (!gconf->logfile) {
1137 gconf->logfile = g_strdup(GLUSTER_LOGFILE_DEFAULT);
1139 gconf->has_logfile = true;
1141 ret = qemu_gluster_parse(gconf, filename, NULL, errp);
1146 ret = qemu_gluster_co_create(options, errp);
1153 qapi_free_BlockdevCreateOptions(options);
1157 static coroutine_fn int qemu_gluster_co_rw(BlockDriverState *bs,
1158 int64_t sector_num, int nb_sectors,
1159 QEMUIOVector *qiov, int write)
1163 BDRVGlusterState *s = bs->opaque;
1164 size_t size = nb_sectors * BDRV_SECTOR_SIZE;
1165 off_t offset = sector_num * BDRV_SECTOR_SIZE;
1169 acb.coroutine = qemu_coroutine_self();
1170 acb.aio_context = bdrv_get_aio_context(bs);
1173 ret = glfs_pwritev_async(s->fd, qiov->iov, qiov->niov, offset, 0,
1174 gluster_finish_aiocb, &acb);
1176 ret = glfs_preadv_async(s->fd, qiov->iov, qiov->niov, offset, 0,
1177 gluster_finish_aiocb, &acb);
1184 qemu_coroutine_yield();
1188 static coroutine_fn int qemu_gluster_co_truncate(BlockDriverState *bs,
1190 PreallocMode prealloc,
1193 BDRVGlusterState *s = bs->opaque;
1194 return qemu_gluster_do_truncate(s->fd, offset, prealloc, errp);
1197 static coroutine_fn int qemu_gluster_co_readv(BlockDriverState *bs,
1202 return qemu_gluster_co_rw(bs, sector_num, nb_sectors, qiov, 0);
1205 static coroutine_fn int qemu_gluster_co_writev(BlockDriverState *bs,
1212 return qemu_gluster_co_rw(bs, sector_num, nb_sectors, qiov, 1);
1215 static void qemu_gluster_close(BlockDriverState *bs)
1217 BDRVGlusterState *s = bs->opaque;
1224 glfs_clear_preopened(s->glfs);
1227 static coroutine_fn int qemu_gluster_co_flush_to_disk(BlockDriverState *bs)
1231 BDRVGlusterState *s = bs->opaque;
1235 acb.coroutine = qemu_coroutine_self();
1236 acb.aio_context = bdrv_get_aio_context(bs);
1238 ret = glfs_fsync_async(s->fd, gluster_finish_aiocb, &acb);
1244 qemu_coroutine_yield();
1253 /* Some versions of Gluster (3.5.6 -> 3.5.8?) will not retain its cache
1254 * after a fsync failure, so we have no way of allowing the guest to safely
1255 * continue. Gluster versions prior to 3.5.6 don't retain the cache
1256 * either, but will invalidate the fd on error, so this is again our only
1259 * The 'resync-failed-syncs-after-fsync' xlator option for the
1260 * write-behind cache will cause later gluster versions to retain its
1261 * cache after error, so long as the fd remains open. However, we
1262 * currently have no way of knowing if this option is supported.
1264 * TODO: Once gluster provides a way for us to determine if the option
1265 * is supported, bypass the closure and setting drv to NULL. */
1266 qemu_gluster_close(bs);
1271 #ifdef CONFIG_GLUSTERFS_DISCARD
1272 static coroutine_fn int qemu_gluster_co_pdiscard(BlockDriverState *bs,
1273 int64_t offset, int size)
1277 BDRVGlusterState *s = bs->opaque;
1281 acb.coroutine = qemu_coroutine_self();
1282 acb.aio_context = bdrv_get_aio_context(bs);
1284 ret = glfs_discard_async(s->fd, offset, size, gluster_finish_aiocb, &acb);
1289 qemu_coroutine_yield();
1294 static int64_t qemu_gluster_getlength(BlockDriverState *bs)
1296 BDRVGlusterState *s = bs->opaque;
1299 ret = glfs_lseek(s->fd, 0, SEEK_END);
1307 static int64_t qemu_gluster_allocated_file_size(BlockDriverState *bs)
1309 BDRVGlusterState *s = bs->opaque;
1313 ret = glfs_fstat(s->fd, &st);
1317 return st.st_blocks * 512;
1321 static int qemu_gluster_has_zero_init(BlockDriverState *bs)
1323 /* GlusterFS volume could be backed by a block device */
1328 * Find allocation range in @bs around offset @start.
1329 * May change underlying file descriptor's file offset.
1330 * If @start is not in a hole, store @start in @data, and the
1331 * beginning of the next hole in @hole, and return 0.
1332 * If @start is in a non-trailing hole, store @start in @hole and the
1333 * beginning of the next non-hole in @data, and return 0.
1334 * If @start is in a trailing hole or beyond EOF, return -ENXIO.
1335 * If we can't find out, return a negative errno other than -ENXIO.
1337 * (Shamefully copied from file-posix.c, only minuscule adaptions.)
1339 static int find_allocation(BlockDriverState *bs, off_t start,
1340 off_t *data, off_t *hole)
1342 BDRVGlusterState *s = bs->opaque;
1344 if (!s->supports_seek_data) {
1348 #if defined SEEK_HOLE && defined SEEK_DATA
1353 * D1. offs == start: start is in data
1354 * D2. offs > start: start is in a hole, next data at offs
1355 * D3. offs < 0, errno = ENXIO: either start is in a trailing hole
1356 * or start is beyond EOF
1357 * If the latter happens, the file has been truncated behind
1358 * our back since we opened it. All bets are off then.
1359 * Treating like a trailing hole is simplest.
1360 * D4. offs < 0, errno != ENXIO: we learned nothing
1362 offs = glfs_lseek(s->fd, start, SEEK_DATA);
1364 return -errno; /* D3 or D4 */
1368 /* This is not a valid return by lseek(). We are safe to just return
1369 * -EIO in this case, and we'll treat it like D4. Unfortunately some
1370 * versions of gluster server will return offs < start, so an assert
1371 * here will unnecessarily abort QEMU. */
1376 /* D2: in hole, next data at offs */
1382 /* D1: in data, end not yet known */
1386 * H1. offs == start: start is in a hole
1387 * If this happens here, a hole has been dug behind our back
1388 * since the previous lseek().
1389 * H2. offs > start: either start is in data, next hole at offs,
1390 * or start is in trailing hole, EOF at offs
1391 * Linux treats trailing holes like any other hole: offs ==
1392 * start. Solaris seeks to EOF instead: offs > start (blech).
1393 * If that happens here, a hole has been dug behind our back
1394 * since the previous lseek().
1395 * H3. offs < 0, errno = ENXIO: start is beyond EOF
1396 * If this happens, the file has been truncated behind our
1397 * back since we opened it. Treat it like a trailing hole.
1398 * H4. offs < 0, errno != ENXIO: we learned nothing
1399 * Pretend we know nothing at all, i.e. "forget" about D1.
1401 offs = glfs_lseek(s->fd, start, SEEK_HOLE);
1403 return -errno; /* D1 and (H3 or H4) */
1407 /* This is not a valid return by lseek(). We are safe to just return
1408 * -EIO in this case, and we'll treat it like H4. Unfortunately some
1409 * versions of gluster server will return offs < start, so an assert
1410 * here will unnecessarily abort QEMU. */
1416 * D1 and H2: either in data, next hole at offs, or it was in
1417 * data but is now in a trailing hole. In the latter case,
1418 * all bets are off. Treating it as if it there was data all
1419 * the way to EOF is safe, so simply do that.
1435 * Returns the allocation status of the specified offset.
1437 * The block layer guarantees 'offset' and 'bytes' are within bounds.
1439 * 'pnum' is set to the number of bytes (including and immediately following
1440 * the specified offset) that are known to be in the same
1441 * allocated/unallocated state.
1443 * 'bytes' is the max value 'pnum' should be set to.
1445 * (Based on raw_co_block_status() from file-posix.c.)
1447 static int coroutine_fn qemu_gluster_co_block_status(BlockDriverState *bs,
1453 BlockDriverState **file)
1455 BDRVGlusterState *s = bs->opaque;
1456 off_t data = 0, hole = 0;
1467 return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID;
1470 ret = find_allocation(bs, offset, &data, &hole);
1471 if (ret == -ENXIO) {
1474 ret = BDRV_BLOCK_ZERO;
1475 } else if (ret < 0) {
1476 /* No info available, so pretend there are no holes */
1478 ret = BDRV_BLOCK_DATA;
1479 } else if (data == offset) {
1480 /* On a data extent, compute bytes to the end of the extent,
1481 * possibly including a partial sector at EOF. */
1482 *pnum = MIN(bytes, hole - offset);
1483 ret = BDRV_BLOCK_DATA;
1485 /* On a hole, compute bytes to the beginning of the next extent. */
1486 assert(hole == offset);
1487 *pnum = MIN(bytes, data - offset);
1488 ret = BDRV_BLOCK_ZERO;
1494 return ret | BDRV_BLOCK_OFFSET_VALID;
1498 static BlockDriver bdrv_gluster = {
1499 .format_name = "gluster",
1500 .protocol_name = "gluster",
1501 .instance_size = sizeof(BDRVGlusterState),
1502 .bdrv_needs_filename = false,
1503 .bdrv_file_open = qemu_gluster_open,
1504 .bdrv_reopen_prepare = qemu_gluster_reopen_prepare,
1505 .bdrv_reopen_commit = qemu_gluster_reopen_commit,
1506 .bdrv_reopen_abort = qemu_gluster_reopen_abort,
1507 .bdrv_close = qemu_gluster_close,
1508 .bdrv_co_create = qemu_gluster_co_create,
1509 .bdrv_co_create_opts = qemu_gluster_co_create_opts,
1510 .bdrv_getlength = qemu_gluster_getlength,
1511 .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size,
1512 .bdrv_co_truncate = qemu_gluster_co_truncate,
1513 .bdrv_co_readv = qemu_gluster_co_readv,
1514 .bdrv_co_writev = qemu_gluster_co_writev,
1515 .bdrv_co_flush_to_disk = qemu_gluster_co_flush_to_disk,
1516 .bdrv_has_zero_init = qemu_gluster_has_zero_init,
1517 #ifdef CONFIG_GLUSTERFS_DISCARD
1518 .bdrv_co_pdiscard = qemu_gluster_co_pdiscard,
1520 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1521 .bdrv_co_pwrite_zeroes = qemu_gluster_co_pwrite_zeroes,
1523 .bdrv_co_block_status = qemu_gluster_co_block_status,
1524 .create_opts = &qemu_gluster_create_opts,
1527 static BlockDriver bdrv_gluster_tcp = {
1528 .format_name = "gluster",
1529 .protocol_name = "gluster+tcp",
1530 .instance_size = sizeof(BDRVGlusterState),
1531 .bdrv_needs_filename = false,
1532 .bdrv_file_open = qemu_gluster_open,
1533 .bdrv_reopen_prepare = qemu_gluster_reopen_prepare,
1534 .bdrv_reopen_commit = qemu_gluster_reopen_commit,
1535 .bdrv_reopen_abort = qemu_gluster_reopen_abort,
1536 .bdrv_close = qemu_gluster_close,
1537 .bdrv_co_create = qemu_gluster_co_create,
1538 .bdrv_co_create_opts = qemu_gluster_co_create_opts,
1539 .bdrv_getlength = qemu_gluster_getlength,
1540 .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size,
1541 .bdrv_co_truncate = qemu_gluster_co_truncate,
1542 .bdrv_co_readv = qemu_gluster_co_readv,
1543 .bdrv_co_writev = qemu_gluster_co_writev,
1544 .bdrv_co_flush_to_disk = qemu_gluster_co_flush_to_disk,
1545 .bdrv_has_zero_init = qemu_gluster_has_zero_init,
1546 #ifdef CONFIG_GLUSTERFS_DISCARD
1547 .bdrv_co_pdiscard = qemu_gluster_co_pdiscard,
1549 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1550 .bdrv_co_pwrite_zeroes = qemu_gluster_co_pwrite_zeroes,
1552 .bdrv_co_block_status = qemu_gluster_co_block_status,
1553 .create_opts = &qemu_gluster_create_opts,
1556 static BlockDriver bdrv_gluster_unix = {
1557 .format_name = "gluster",
1558 .protocol_name = "gluster+unix",
1559 .instance_size = sizeof(BDRVGlusterState),
1560 .bdrv_needs_filename = true,
1561 .bdrv_file_open = qemu_gluster_open,
1562 .bdrv_reopen_prepare = qemu_gluster_reopen_prepare,
1563 .bdrv_reopen_commit = qemu_gluster_reopen_commit,
1564 .bdrv_reopen_abort = qemu_gluster_reopen_abort,
1565 .bdrv_close = qemu_gluster_close,
1566 .bdrv_co_create = qemu_gluster_co_create,
1567 .bdrv_co_create_opts = qemu_gluster_co_create_opts,
1568 .bdrv_getlength = qemu_gluster_getlength,
1569 .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size,
1570 .bdrv_co_truncate = qemu_gluster_co_truncate,
1571 .bdrv_co_readv = qemu_gluster_co_readv,
1572 .bdrv_co_writev = qemu_gluster_co_writev,
1573 .bdrv_co_flush_to_disk = qemu_gluster_co_flush_to_disk,
1574 .bdrv_has_zero_init = qemu_gluster_has_zero_init,
1575 #ifdef CONFIG_GLUSTERFS_DISCARD
1576 .bdrv_co_pdiscard = qemu_gluster_co_pdiscard,
1578 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1579 .bdrv_co_pwrite_zeroes = qemu_gluster_co_pwrite_zeroes,
1581 .bdrv_co_block_status = qemu_gluster_co_block_status,
1582 .create_opts = &qemu_gluster_create_opts,
1585 /* rdma is deprecated (actually never supported for volfile fetch).
1586 * Let's maintain it for the protocol compatibility, to make sure things
1587 * won't break immediately. For now, gluster+rdma will fall back to gluster+tcp
1588 * protocol with a warning.
1589 * TODO: remove gluster+rdma interface support
1591 static BlockDriver bdrv_gluster_rdma = {
1592 .format_name = "gluster",
1593 .protocol_name = "gluster+rdma",
1594 .instance_size = sizeof(BDRVGlusterState),
1595 .bdrv_needs_filename = true,
1596 .bdrv_file_open = qemu_gluster_open,
1597 .bdrv_reopen_prepare = qemu_gluster_reopen_prepare,
1598 .bdrv_reopen_commit = qemu_gluster_reopen_commit,
1599 .bdrv_reopen_abort = qemu_gluster_reopen_abort,
1600 .bdrv_close = qemu_gluster_close,
1601 .bdrv_co_create = qemu_gluster_co_create,
1602 .bdrv_co_create_opts = qemu_gluster_co_create_opts,
1603 .bdrv_getlength = qemu_gluster_getlength,
1604 .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size,
1605 .bdrv_co_truncate = qemu_gluster_co_truncate,
1606 .bdrv_co_readv = qemu_gluster_co_readv,
1607 .bdrv_co_writev = qemu_gluster_co_writev,
1608 .bdrv_co_flush_to_disk = qemu_gluster_co_flush_to_disk,
1609 .bdrv_has_zero_init = qemu_gluster_has_zero_init,
1610 #ifdef CONFIG_GLUSTERFS_DISCARD
1611 .bdrv_co_pdiscard = qemu_gluster_co_pdiscard,
1613 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1614 .bdrv_co_pwrite_zeroes = qemu_gluster_co_pwrite_zeroes,
1616 .bdrv_co_block_status = qemu_gluster_co_block_status,
1617 .create_opts = &qemu_gluster_create_opts,
1620 static void bdrv_gluster_init(void)
1622 bdrv_register(&bdrv_gluster_rdma);
1623 bdrv_register(&bdrv_gluster_unix);
1624 bdrv_register(&bdrv_gluster_tcp);
1625 bdrv_register(&bdrv_gluster);
1628 block_init(bdrv_gluster_init);