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.
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"
16 #include "qemu/error-report.h"
18 #define GLUSTER_OPT_FILENAME "filename"
19 #define GLUSTER_OPT_VOLUME "volume"
20 #define GLUSTER_OPT_PATH "path"
21 #define GLUSTER_OPT_TYPE "type"
22 #define GLUSTER_OPT_SERVER_PATTERN "server."
23 #define GLUSTER_OPT_HOST "host"
24 #define GLUSTER_OPT_PORT "port"
25 #define GLUSTER_OPT_TO "to"
26 #define GLUSTER_OPT_IPV4 "ipv4"
27 #define GLUSTER_OPT_IPV6 "ipv6"
28 #define GLUSTER_OPT_SOCKET "socket"
29 #define GLUSTER_OPT_DEBUG "debug"
30 #define GLUSTER_DEFAULT_PORT 24007
31 #define GLUSTER_DEBUG_DEFAULT 4
32 #define GLUSTER_DEBUG_MAX 9
34 #define GERR_INDEX_HINT "hint: check in 'server' array index '%d'\n"
36 typedef struct GlusterAIOCB {
41 AioContext *aio_context;
44 typedef struct BDRVGlusterState {
47 bool supports_seek_data;
51 typedef struct BDRVGlusterReopenState {
54 } BDRVGlusterReopenState;
57 static QemuOptsList qemu_gluster_create_opts = {
58 .name = "qemu-gluster-create-opts",
59 .head = QTAILQ_HEAD_INITIALIZER(qemu_gluster_create_opts.head),
62 .name = BLOCK_OPT_SIZE,
63 .type = QEMU_OPT_SIZE,
64 .help = "Virtual disk size"
67 .name = BLOCK_OPT_PREALLOC,
68 .type = QEMU_OPT_STRING,
69 .help = "Preallocation mode (allowed values: off, full)"
72 .name = GLUSTER_OPT_DEBUG,
73 .type = QEMU_OPT_NUMBER,
74 .help = "Gluster log level, valid range is 0-9",
80 static QemuOptsList runtime_opts = {
82 .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
85 .name = GLUSTER_OPT_FILENAME,
86 .type = QEMU_OPT_STRING,
87 .help = "URL to the gluster image",
90 .name = GLUSTER_OPT_DEBUG,
91 .type = QEMU_OPT_NUMBER,
92 .help = "Gluster log level, valid range is 0-9",
98 static QemuOptsList runtime_json_opts = {
99 .name = "gluster_json",
100 .head = QTAILQ_HEAD_INITIALIZER(runtime_json_opts.head),
103 .name = GLUSTER_OPT_VOLUME,
104 .type = QEMU_OPT_STRING,
105 .help = "name of gluster volume where VM image resides",
108 .name = GLUSTER_OPT_PATH,
109 .type = QEMU_OPT_STRING,
110 .help = "absolute path to image file in gluster volume",
113 .name = GLUSTER_OPT_DEBUG,
114 .type = QEMU_OPT_NUMBER,
115 .help = "Gluster log level, valid range is 0-9",
117 { /* end of list */ }
121 static QemuOptsList runtime_type_opts = {
122 .name = "gluster_type",
123 .head = QTAILQ_HEAD_INITIALIZER(runtime_type_opts.head),
126 .name = GLUSTER_OPT_TYPE,
127 .type = QEMU_OPT_STRING,
130 { /* end of list */ }
134 static QemuOptsList runtime_unix_opts = {
135 .name = "gluster_unix",
136 .head = QTAILQ_HEAD_INITIALIZER(runtime_unix_opts.head),
139 .name = GLUSTER_OPT_SOCKET,
140 .type = QEMU_OPT_STRING,
141 .help = "socket file path)",
143 { /* end of list */ }
147 static QemuOptsList runtime_tcp_opts = {
148 .name = "gluster_tcp",
149 .head = QTAILQ_HEAD_INITIALIZER(runtime_tcp_opts.head),
152 .name = GLUSTER_OPT_TYPE,
153 .type = QEMU_OPT_STRING,
157 .name = GLUSTER_OPT_HOST,
158 .type = QEMU_OPT_STRING,
159 .help = "host address (hostname/ipv4/ipv6 addresses)",
162 .name = GLUSTER_OPT_PORT,
163 .type = QEMU_OPT_NUMBER,
164 .help = "port number on which glusterd is listening (default 24007)",
168 .type = QEMU_OPT_NUMBER,
169 .help = "max port number, not supported by gluster",
173 .type = QEMU_OPT_BOOL,
174 .help = "ipv4 bool value, not supported by gluster",
178 .type = QEMU_OPT_BOOL,
179 .help = "ipv6 bool value, not supported by gluster",
181 { /* end of list */ }
185 static int parse_volume_options(BlockdevOptionsGluster *gconf, char *path)
194 p = q = path + strspn(path, "/");
195 p += strcspn(p, "/");
199 gconf->volume = g_strndup(q, p - q);
206 gconf->path = g_strdup(p);
211 * file=gluster[+transport]://[host[:port]]/volume/path[?socket=...]
213 * 'gluster' is the protocol.
215 * 'transport' specifies the transport type used to connect to gluster
216 * management daemon (glusterd). Valid transport types are
217 * tcp or unix. If a transport type isn't specified, then tcp type is assumed.
219 * 'host' specifies the host where the volume file specification for
220 * the given volume resides. This can be either hostname or ipv4 address.
221 * If transport type is 'unix', then 'host' field should not be specified.
222 * The 'socket' field needs to be populated with the path to unix domain
225 * 'port' is the port number on which glusterd is listening. This is optional
226 * and if not specified, QEMU will send 0 which will make gluster to use the
227 * default port. If the transport type is unix, then 'port' should not be
230 * 'volume' is the name of the gluster volume which contains the VM image.
232 * 'path' is the path to the actual VM image that resides on gluster volume.
236 * file=gluster://1.2.3.4/testvol/a.img
237 * file=gluster+tcp://1.2.3.4/testvol/a.img
238 * file=gluster+tcp://1.2.3.4:24007/testvol/dir/a.img
239 * file=gluster+tcp://host.domain.com:24007/testvol/dir/a.img
240 * file=gluster+unix:///testvol/dir/a.img?socket=/tmp/glusterd.socket
242 static int qemu_gluster_parse_uri(BlockdevOptionsGluster *gconf,
243 const char *filename)
245 GlusterServer *gsconf;
247 QueryParams *qp = NULL;
248 bool is_unix = false;
251 uri = uri_parse(filename);
256 gconf->server = g_new0(GlusterServerList, 1);
257 gconf->server->value = gsconf = g_new0(GlusterServer, 1);
260 if (!uri->scheme || !strcmp(uri->scheme, "gluster")) {
261 gsconf->type = GLUSTER_TRANSPORT_TCP;
262 } else if (!strcmp(uri->scheme, "gluster+tcp")) {
263 gsconf->type = GLUSTER_TRANSPORT_TCP;
264 } else if (!strcmp(uri->scheme, "gluster+unix")) {
265 gsconf->type = GLUSTER_TRANSPORT_UNIX;
267 } else if (!strcmp(uri->scheme, "gluster+rdma")) {
268 gsconf->type = GLUSTER_TRANSPORT_TCP;
269 error_report("Warning: rdma feature is not supported, falling "
276 ret = parse_volume_options(gconf, uri->path);
281 qp = query_params_parse(uri->query);
282 if (qp->n > 1 || (is_unix && !qp->n) || (!is_unix && qp->n)) {
288 if (uri->server || uri->port) {
292 if (strcmp(qp->p[0].name, "socket")) {
296 gsconf->u.q_unix.path = g_strdup(qp->p[0].value);
298 gsconf->u.tcp.host = g_strdup(uri->server ? uri->server : "localhost");
300 gsconf->u.tcp.port = g_strdup_printf("%d", uri->port);
302 gsconf->u.tcp.port = g_strdup_printf("%d", GLUSTER_DEFAULT_PORT);
308 query_params_free(qp);
314 static struct glfs *qemu_gluster_glfs_init(BlockdevOptionsGluster *gconf,
320 GlusterServerList *server;
322 glfs = glfs_new(gconf->volume);
327 for (server = gconf->server; server; server = server->next) {
328 if (server->value->type == GLUSTER_TRANSPORT_UNIX) {
329 ret = glfs_set_volfile_server(glfs,
330 GlusterTransport_lookup[server->value->type],
331 server->value->u.q_unix.path, 0);
333 ret = glfs_set_volfile_server(glfs,
334 GlusterTransport_lookup[server->value->type],
335 server->value->u.tcp.host,
336 atoi(server->value->u.tcp.port));
344 ret = glfs_set_logging(glfs, "-", gconf->debug_level);
349 ret = glfs_init(glfs);
351 error_setg(errp, "Gluster connection for volume %s, path %s failed"
352 " to connect", gconf->volume, gconf->path);
353 for (server = gconf->server; server; server = server->next) {
354 if (server->value->type == GLUSTER_TRANSPORT_UNIX) {
355 error_append_hint(errp, "hint: failed on socket %s ",
356 server->value->u.q_unix.path);
358 error_append_hint(errp, "hint: failed on host %s and port %s ",
359 server->value->u.tcp.host,
360 server->value->u.tcp.port);
364 error_append_hint(errp, "Please refer to gluster logs for more info\n");
366 /* glfs_init sometimes doesn't set errno although docs suggest that */
384 static int qapi_enum_parse(const char *opt)
389 return GLUSTER_TRANSPORT__MAX;
392 for (i = 0; i < GLUSTER_TRANSPORT__MAX; i++) {
393 if (!strcmp(opt, GlusterTransport_lookup[i])) {
402 * Convert the json formatted command line into qapi.
404 static int qemu_gluster_parse_json(BlockdevOptionsGluster *gconf,
405 QDict *options, Error **errp)
408 GlusterServer *gsconf;
409 GlusterServerList *curr = NULL;
410 QDict *backing_options = NULL;
411 Error *local_err = NULL;
417 /* create opts info from runtime_json_opts list */
418 opts = qemu_opts_create(&runtime_json_opts, NULL, 0, &error_abort);
419 qemu_opts_absorb_qdict(opts, options, &local_err);
424 num_servers = qdict_array_entries(options, GLUSTER_OPT_SERVER_PATTERN);
425 if (num_servers < 1) {
426 error_setg(&local_err, QERR_MISSING_PARAMETER, "server");
430 ptr = qemu_opt_get(opts, GLUSTER_OPT_VOLUME);
432 error_setg(&local_err, QERR_MISSING_PARAMETER, GLUSTER_OPT_VOLUME);
435 gconf->volume = g_strdup(ptr);
437 ptr = qemu_opt_get(opts, GLUSTER_OPT_PATH);
439 error_setg(&local_err, QERR_MISSING_PARAMETER, GLUSTER_OPT_PATH);
442 gconf->path = g_strdup(ptr);
445 for (i = 0; i < num_servers; i++) {
446 str = g_strdup_printf(GLUSTER_OPT_SERVER_PATTERN"%d.", i);
447 qdict_extract_subqdict(options, &backing_options, str);
449 /* create opts info from runtime_type_opts list */
450 opts = qemu_opts_create(&runtime_type_opts, NULL, 0, &error_abort);
451 qemu_opts_absorb_qdict(opts, backing_options, &local_err);
456 ptr = qemu_opt_get(opts, GLUSTER_OPT_TYPE);
457 gsconf = g_new0(GlusterServer, 1);
458 gsconf->type = qapi_enum_parse(ptr);
460 error_setg(&local_err, QERR_MISSING_PARAMETER, GLUSTER_OPT_TYPE);
461 error_append_hint(&local_err, GERR_INDEX_HINT, i);
465 if (gsconf->type == GLUSTER_TRANSPORT__MAX) {
466 error_setg(&local_err, QERR_INVALID_PARAMETER_VALUE,
467 GLUSTER_OPT_TYPE, "tcp or unix");
468 error_append_hint(&local_err, GERR_INDEX_HINT, i);
473 if (gsconf->type == GLUSTER_TRANSPORT_TCP) {
474 /* create opts info from runtime_tcp_opts list */
475 opts = qemu_opts_create(&runtime_tcp_opts, NULL, 0, &error_abort);
476 qemu_opts_absorb_qdict(opts, backing_options, &local_err);
481 ptr = qemu_opt_get(opts, GLUSTER_OPT_HOST);
483 error_setg(&local_err, QERR_MISSING_PARAMETER,
485 error_append_hint(&local_err, GERR_INDEX_HINT, i);
488 gsconf->u.tcp.host = g_strdup(ptr);
489 ptr = qemu_opt_get(opts, GLUSTER_OPT_PORT);
491 error_setg(&local_err, QERR_MISSING_PARAMETER,
493 error_append_hint(&local_err, GERR_INDEX_HINT, i);
496 gsconf->u.tcp.port = g_strdup(ptr);
498 /* defend for unsupported fields in InetSocketAddress,
499 * i.e. @ipv4, @ipv6 and @to
501 ptr = qemu_opt_get(opts, GLUSTER_OPT_TO);
503 gsconf->u.tcp.has_to = true;
505 ptr = qemu_opt_get(opts, GLUSTER_OPT_IPV4);
507 gsconf->u.tcp.has_ipv4 = true;
509 ptr = qemu_opt_get(opts, GLUSTER_OPT_IPV6);
511 gsconf->u.tcp.has_ipv6 = true;
513 if (gsconf->u.tcp.has_to) {
514 error_setg(&local_err, "Parameter 'to' not supported");
517 if (gsconf->u.tcp.has_ipv4 || gsconf->u.tcp.has_ipv6) {
518 error_setg(&local_err, "Parameters 'ipv4/ipv6' not supported");
523 /* create opts info from runtime_unix_opts list */
524 opts = qemu_opts_create(&runtime_unix_opts, NULL, 0, &error_abort);
525 qemu_opts_absorb_qdict(opts, backing_options, &local_err);
530 ptr = qemu_opt_get(opts, GLUSTER_OPT_SOCKET);
532 error_setg(&local_err, QERR_MISSING_PARAMETER,
534 error_append_hint(&local_err, GERR_INDEX_HINT, i);
537 gsconf->u.q_unix.path = g_strdup(ptr);
541 if (gconf->server == NULL) {
542 gconf->server = g_new0(GlusterServerList, 1);
543 gconf->server->value = gsconf;
544 curr = gconf->server;
546 curr->next = g_new0(GlusterServerList, 1);
547 curr->next->value = gsconf;
551 qdict_del(backing_options, str);
559 error_propagate(errp, local_err);
562 qdict_del(backing_options, str);
569 static struct glfs *qemu_gluster_init(BlockdevOptionsGluster *gconf,
570 const char *filename,
571 QDict *options, Error **errp)
575 ret = qemu_gluster_parse_uri(gconf, filename);
577 error_setg(errp, "invalid URI");
578 error_append_hint(errp, "Usage: file=gluster[+transport]://"
579 "[host[:port]]/volume/path[?socket=...]\n");
584 ret = qemu_gluster_parse_json(gconf, options, errp);
586 error_append_hint(errp, "Usage: "
587 "-drive driver=qcow2,file.driver=gluster,"
588 "file.volume=testvol,file.path=/path/a.qcow2"
589 "[,file.debug=9],file.server.0.type=tcp,"
590 "file.server.0.host=1.2.3.4,"
591 "file.server.0.port=24007,"
592 "file.server.1.transport=unix,"
593 "file.server.1.socket=/var/run/glusterd.socket ..."
601 return qemu_gluster_glfs_init(gconf, errp);
604 static void qemu_gluster_complete_aio(void *opaque)
606 GlusterAIOCB *acb = (GlusterAIOCB *)opaque;
608 qemu_bh_delete(acb->bh);
610 qemu_coroutine_enter(acb->coroutine);
614 * AIO callback routine called from GlusterFS thread.
616 static void gluster_finish_aiocb(struct glfs_fd *fd, ssize_t ret, void *arg)
618 GlusterAIOCB *acb = (GlusterAIOCB *)arg;
620 if (!ret || ret == acb->size) {
621 acb->ret = 0; /* Success */
622 } else if (ret < 0) {
623 acb->ret = -errno; /* Read/Write failed */
625 acb->ret = -EIO; /* Partial read/write - fail it */
628 acb->bh = aio_bh_new(acb->aio_context, qemu_gluster_complete_aio, acb);
629 qemu_bh_schedule(acb->bh);
632 static void qemu_gluster_parse_flags(int bdrv_flags, int *open_flags)
634 assert(open_flags != NULL);
636 *open_flags |= O_BINARY;
638 if (bdrv_flags & BDRV_O_RDWR) {
639 *open_flags |= O_RDWR;
641 *open_flags |= O_RDONLY;
644 if ((bdrv_flags & BDRV_O_NOCACHE)) {
645 *open_flags |= O_DIRECT;
650 * Do SEEK_DATA/HOLE to detect if it is functional. Older broken versions of
651 * gfapi incorrectly return the current offset when SEEK_DATA/HOLE is used.
652 * - Corrected versions return -1 and set errno to EINVAL.
653 * - Versions that support SEEK_DATA/HOLE correctly, will return -1 and set
654 * errno to ENXIO when SEEK_DATA is called with a position of EOF.
656 static bool qemu_gluster_test_seek(struct glfs_fd *fd)
660 eof = glfs_lseek(fd, 0, SEEK_END);
662 /* this should never occur */
666 /* this should always fail with ENXIO if SEEK_DATA is supported */
667 ret = glfs_lseek(fd, eof, SEEK_DATA);
668 return (ret < 0) && (errno == ENXIO);
671 static int qemu_gluster_open(BlockDriverState *bs, QDict *options,
672 int bdrv_flags, Error **errp)
674 BDRVGlusterState *s = bs->opaque;
677 BlockdevOptionsGluster *gconf = NULL;
679 Error *local_err = NULL;
680 const char *filename;
682 opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
683 qemu_opts_absorb_qdict(opts, options, &local_err);
685 error_propagate(errp, local_err);
690 filename = qemu_opt_get(opts, GLUSTER_OPT_FILENAME);
692 s->debug_level = qemu_opt_get_number(opts, GLUSTER_OPT_DEBUG,
693 GLUSTER_DEBUG_DEFAULT);
694 if (s->debug_level < 0) {
696 } else if (s->debug_level > GLUSTER_DEBUG_MAX) {
697 s->debug_level = GLUSTER_DEBUG_MAX;
700 gconf = g_new0(BlockdevOptionsGluster, 1);
701 gconf->debug_level = s->debug_level;
702 gconf->has_debug_level = true;
703 s->glfs = qemu_gluster_init(gconf, filename, options, errp);
709 #ifdef CONFIG_GLUSTERFS_XLATOR_OPT
710 /* Without this, if fsync fails for a recoverable reason (for instance,
711 * ENOSPC), gluster will dump its cache, preventing retries. This means
712 * almost certain data loss. Not all gluster versions support the
713 * 'resync-failed-syncs-after-fsync' key value, but there is no way to
714 * discover during runtime if it is supported (this api returns success for
715 * unknown key/value pairs) */
716 ret = glfs_set_xlator_option(s->glfs, "*-write-behind",
717 "resync-failed-syncs-after-fsync",
720 error_setg_errno(errp, errno, "Unable to set xlator key/value pair");
726 qemu_gluster_parse_flags(bdrv_flags, &open_flags);
728 s->fd = glfs_open(s->glfs, gconf->path, open_flags);
733 s->supports_seek_data = qemu_gluster_test_seek(s->fd);
737 qapi_free_BlockdevOptionsGluster(gconf);
750 static int qemu_gluster_reopen_prepare(BDRVReopenState *state,
751 BlockReopenQueue *queue, Error **errp)
755 BDRVGlusterReopenState *reop_s;
756 BlockdevOptionsGluster *gconf;
759 assert(state != NULL);
760 assert(state->bs != NULL);
762 s = state->bs->opaque;
764 state->opaque = g_new0(BDRVGlusterReopenState, 1);
765 reop_s = state->opaque;
767 qemu_gluster_parse_flags(state->flags, &open_flags);
769 gconf = g_new0(BlockdevOptionsGluster, 1);
770 gconf->debug_level = s->debug_level;
771 gconf->has_debug_level = true;
772 reop_s->glfs = qemu_gluster_init(gconf, state->bs->filename, NULL, errp);
773 if (reop_s->glfs == NULL) {
778 #ifdef CONFIG_GLUSTERFS_XLATOR_OPT
779 ret = glfs_set_xlator_option(reop_s->glfs, "*-write-behind",
780 "resync-failed-syncs-after-fsync", "on");
782 error_setg_errno(errp, errno, "Unable to set xlator key/value pair");
788 reop_s->fd = glfs_open(reop_s->glfs, gconf->path, open_flags);
789 if (reop_s->fd == NULL) {
790 /* reops->glfs will be cleaned up in _abort */
796 /* state->opaque will be freed in either the _abort or _commit */
797 qapi_free_BlockdevOptionsGluster(gconf);
801 static void qemu_gluster_reopen_commit(BDRVReopenState *state)
803 BDRVGlusterReopenState *reop_s = state->opaque;
804 BDRVGlusterState *s = state->bs->opaque;
815 /* use the newly opened image / connection */
817 s->glfs = reop_s->glfs;
819 g_free(state->opaque);
820 state->opaque = NULL;
826 static void qemu_gluster_reopen_abort(BDRVReopenState *state)
828 BDRVGlusterReopenState *reop_s = state->opaque;
830 if (reop_s == NULL) {
835 glfs_close(reop_s->fd);
839 glfs_fini(reop_s->glfs);
842 g_free(state->opaque);
843 state->opaque = NULL;
848 #ifdef CONFIG_GLUSTERFS_ZEROFILL
849 static coroutine_fn int qemu_gluster_co_pwrite_zeroes(BlockDriverState *bs,
852 BdrvRequestFlags flags)
856 BDRVGlusterState *s = bs->opaque;
860 acb.coroutine = qemu_coroutine_self();
861 acb.aio_context = bdrv_get_aio_context(bs);
863 ret = glfs_zerofill_async(s->fd, offset, size, gluster_finish_aiocb, &acb);
868 qemu_coroutine_yield();
872 static inline bool gluster_supports_zerofill(void)
877 static inline int qemu_gluster_zerofill(struct glfs_fd *fd, int64_t offset,
880 return glfs_zerofill(fd, offset, size);
884 static inline bool gluster_supports_zerofill(void)
889 static inline int qemu_gluster_zerofill(struct glfs_fd *fd, int64_t offset,
896 static int qemu_gluster_create(const char *filename,
897 QemuOpts *opts, Error **errp)
899 BlockdevOptionsGluster *gconf;
904 int64_t total_size = 0;
907 gconf = g_new0(BlockdevOptionsGluster, 1);
908 gconf->debug_level = qemu_opt_get_number_del(opts, GLUSTER_OPT_DEBUG,
909 GLUSTER_DEBUG_DEFAULT);
910 if (gconf->debug_level < 0) {
911 gconf->debug_level = 0;
912 } else if (gconf->debug_level > GLUSTER_DEBUG_MAX) {
913 gconf->debug_level = GLUSTER_DEBUG_MAX;
915 gconf->has_debug_level = true;
917 glfs = qemu_gluster_init(gconf, filename, NULL, errp);
923 total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
926 tmp = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
927 if (!tmp || !strcmp(tmp, "off")) {
929 } else if (!strcmp(tmp, "full") && gluster_supports_zerofill()) {
932 error_setg(errp, "Invalid preallocation mode: '%s'"
933 " or GlusterFS doesn't support zerofill API", tmp);
938 fd = glfs_creat(glfs, gconf->path,
939 O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, S_IRUSR | S_IWUSR);
943 if (!glfs_ftruncate(fd, total_size)) {
944 if (prealloc && qemu_gluster_zerofill(fd, 0, total_size)) {
951 if (glfs_close(fd) != 0) {
957 qapi_free_BlockdevOptionsGluster(gconf);
964 static coroutine_fn int qemu_gluster_co_rw(BlockDriverState *bs,
965 int64_t sector_num, int nb_sectors,
966 QEMUIOVector *qiov, int write)
970 BDRVGlusterState *s = bs->opaque;
971 size_t size = nb_sectors * BDRV_SECTOR_SIZE;
972 off_t offset = sector_num * BDRV_SECTOR_SIZE;
976 acb.coroutine = qemu_coroutine_self();
977 acb.aio_context = bdrv_get_aio_context(bs);
980 ret = glfs_pwritev_async(s->fd, qiov->iov, qiov->niov, offset, 0,
981 gluster_finish_aiocb, &acb);
983 ret = glfs_preadv_async(s->fd, qiov->iov, qiov->niov, offset, 0,
984 gluster_finish_aiocb, &acb);
991 qemu_coroutine_yield();
995 static int qemu_gluster_truncate(BlockDriverState *bs, int64_t offset)
998 BDRVGlusterState *s = bs->opaque;
1000 ret = glfs_ftruncate(s->fd, offset);
1008 static coroutine_fn int qemu_gluster_co_readv(BlockDriverState *bs,
1013 return qemu_gluster_co_rw(bs, sector_num, nb_sectors, qiov, 0);
1016 static coroutine_fn int qemu_gluster_co_writev(BlockDriverState *bs,
1021 return qemu_gluster_co_rw(bs, sector_num, nb_sectors, qiov, 1);
1024 static void qemu_gluster_close(BlockDriverState *bs)
1026 BDRVGlusterState *s = bs->opaque;
1035 static coroutine_fn int qemu_gluster_co_flush_to_disk(BlockDriverState *bs)
1039 BDRVGlusterState *s = bs->opaque;
1043 acb.coroutine = qemu_coroutine_self();
1044 acb.aio_context = bdrv_get_aio_context(bs);
1046 ret = glfs_fsync_async(s->fd, gluster_finish_aiocb, &acb);
1052 qemu_coroutine_yield();
1061 /* Some versions of Gluster (3.5.6 -> 3.5.8?) will not retain its cache
1062 * after a fsync failure, so we have no way of allowing the guest to safely
1063 * continue. Gluster versions prior to 3.5.6 don't retain the cache
1064 * either, but will invalidate the fd on error, so this is again our only
1067 * The 'resync-failed-syncs-after-fsync' xlator option for the
1068 * write-behind cache will cause later gluster versions to retain its
1069 * cache after error, so long as the fd remains open. However, we
1070 * currently have no way of knowing if this option is supported.
1072 * TODO: Once gluster provides a way for us to determine if the option
1073 * is supported, bypass the closure and setting drv to NULL. */
1074 qemu_gluster_close(bs);
1079 #ifdef CONFIG_GLUSTERFS_DISCARD
1080 static coroutine_fn int qemu_gluster_co_pdiscard(BlockDriverState *bs,
1081 int64_t offset, int size)
1085 BDRVGlusterState *s = bs->opaque;
1089 acb.coroutine = qemu_coroutine_self();
1090 acb.aio_context = bdrv_get_aio_context(bs);
1092 ret = glfs_discard_async(s->fd, offset, size, gluster_finish_aiocb, &acb);
1097 qemu_coroutine_yield();
1102 static int64_t qemu_gluster_getlength(BlockDriverState *bs)
1104 BDRVGlusterState *s = bs->opaque;
1107 ret = glfs_lseek(s->fd, 0, SEEK_END);
1115 static int64_t qemu_gluster_allocated_file_size(BlockDriverState *bs)
1117 BDRVGlusterState *s = bs->opaque;
1121 ret = glfs_fstat(s->fd, &st);
1125 return st.st_blocks * 512;
1129 static int qemu_gluster_has_zero_init(BlockDriverState *bs)
1131 /* GlusterFS volume could be backed by a block device */
1136 * Find allocation range in @bs around offset @start.
1137 * May change underlying file descriptor's file offset.
1138 * If @start is not in a hole, store @start in @data, and the
1139 * beginning of the next hole in @hole, and return 0.
1140 * If @start is in a non-trailing hole, store @start in @hole and the
1141 * beginning of the next non-hole in @data, and return 0.
1142 * If @start is in a trailing hole or beyond EOF, return -ENXIO.
1143 * If we can't find out, return a negative errno other than -ENXIO.
1145 * (Shamefully copied from raw-posix.c, only miniscule adaptions.)
1147 static int find_allocation(BlockDriverState *bs, off_t start,
1148 off_t *data, off_t *hole)
1150 BDRVGlusterState *s = bs->opaque;
1153 if (!s->supports_seek_data) {
1159 * D1. offs == start: start is in data
1160 * D2. offs > start: start is in a hole, next data at offs
1161 * D3. offs < 0, errno = ENXIO: either start is in a trailing hole
1162 * or start is beyond EOF
1163 * If the latter happens, the file has been truncated behind
1164 * our back since we opened it. All bets are off then.
1165 * Treating like a trailing hole is simplest.
1166 * D4. offs < 0, errno != ENXIO: we learned nothing
1168 offs = glfs_lseek(s->fd, start, SEEK_DATA);
1170 return -errno; /* D3 or D4 */
1172 assert(offs >= start);
1175 /* D2: in hole, next data at offs */
1181 /* D1: in data, end not yet known */
1185 * H1. offs == start: start is in a hole
1186 * If this happens here, a hole has been dug behind our back
1187 * since the previous lseek().
1188 * H2. offs > start: either start is in data, next hole at offs,
1189 * or start is in trailing hole, EOF at offs
1190 * Linux treats trailing holes like any other hole: offs ==
1191 * start. Solaris seeks to EOF instead: offs > start (blech).
1192 * If that happens here, a hole has been dug behind our back
1193 * since the previous lseek().
1194 * H3. offs < 0, errno = ENXIO: start is beyond EOF
1195 * If this happens, the file has been truncated behind our
1196 * back since we opened it. Treat it like a trailing hole.
1197 * H4. offs < 0, errno != ENXIO: we learned nothing
1198 * Pretend we know nothing at all, i.e. "forget" about D1.
1200 offs = glfs_lseek(s->fd, start, SEEK_HOLE);
1202 return -errno; /* D1 and (H3 or H4) */
1204 assert(offs >= start);
1208 * D1 and H2: either in data, next hole at offs, or it was in
1209 * data but is now in a trailing hole. In the latter case,
1210 * all bets are off. Treating it as if it there was data all
1211 * the way to EOF is safe, so simply do that.
1223 * Returns the allocation status of the specified sectors.
1225 * If 'sector_num' is beyond the end of the disk image the return value is 0
1226 * and 'pnum' is set to 0.
1228 * 'pnum' is set to the number of sectors (including and immediately following
1229 * the specified sector) that are known to be in the same
1230 * allocated/unallocated state.
1232 * 'nb_sectors' is the max value 'pnum' should be set to. If nb_sectors goes
1233 * beyond the end of the disk image it will be clamped.
1235 * (Based on raw_co_get_block_status() from raw-posix.c.)
1237 static int64_t coroutine_fn qemu_gluster_co_get_block_status(
1238 BlockDriverState *bs, int64_t sector_num, int nb_sectors, int *pnum,
1239 BlockDriverState **file)
1241 BDRVGlusterState *s = bs->opaque;
1242 off_t start, data = 0, hole = 0;
1250 start = sector_num * BDRV_SECTOR_SIZE;
1251 total_size = bdrv_getlength(bs);
1252 if (total_size < 0) {
1254 } else if (start >= total_size) {
1257 } else if (start + nb_sectors * BDRV_SECTOR_SIZE > total_size) {
1258 nb_sectors = DIV_ROUND_UP(total_size - start, BDRV_SECTOR_SIZE);
1261 ret = find_allocation(bs, start, &data, &hole);
1262 if (ret == -ENXIO) {
1265 ret = BDRV_BLOCK_ZERO;
1266 } else if (ret < 0) {
1267 /* No info available, so pretend there are no holes */
1269 ret = BDRV_BLOCK_DATA;
1270 } else if (data == start) {
1271 /* On a data extent, compute sectors to the end of the extent,
1272 * possibly including a partial sector at EOF. */
1273 *pnum = MIN(nb_sectors, DIV_ROUND_UP(hole - start, BDRV_SECTOR_SIZE));
1274 ret = BDRV_BLOCK_DATA;
1276 /* On a hole, compute sectors to the beginning of the next extent. */
1277 assert(hole == start);
1278 *pnum = MIN(nb_sectors, (data - start) / BDRV_SECTOR_SIZE);
1279 ret = BDRV_BLOCK_ZERO;
1284 return ret | BDRV_BLOCK_OFFSET_VALID | start;
1288 static BlockDriver bdrv_gluster = {
1289 .format_name = "gluster",
1290 .protocol_name = "gluster",
1291 .instance_size = sizeof(BDRVGlusterState),
1292 .bdrv_needs_filename = false,
1293 .bdrv_file_open = qemu_gluster_open,
1294 .bdrv_reopen_prepare = qemu_gluster_reopen_prepare,
1295 .bdrv_reopen_commit = qemu_gluster_reopen_commit,
1296 .bdrv_reopen_abort = qemu_gluster_reopen_abort,
1297 .bdrv_close = qemu_gluster_close,
1298 .bdrv_create = qemu_gluster_create,
1299 .bdrv_getlength = qemu_gluster_getlength,
1300 .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size,
1301 .bdrv_truncate = qemu_gluster_truncate,
1302 .bdrv_co_readv = qemu_gluster_co_readv,
1303 .bdrv_co_writev = qemu_gluster_co_writev,
1304 .bdrv_co_flush_to_disk = qemu_gluster_co_flush_to_disk,
1305 .bdrv_has_zero_init = qemu_gluster_has_zero_init,
1306 #ifdef CONFIG_GLUSTERFS_DISCARD
1307 .bdrv_co_pdiscard = qemu_gluster_co_pdiscard,
1309 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1310 .bdrv_co_pwrite_zeroes = qemu_gluster_co_pwrite_zeroes,
1312 .bdrv_co_get_block_status = qemu_gluster_co_get_block_status,
1313 .create_opts = &qemu_gluster_create_opts,
1316 static BlockDriver bdrv_gluster_tcp = {
1317 .format_name = "gluster",
1318 .protocol_name = "gluster+tcp",
1319 .instance_size = sizeof(BDRVGlusterState),
1320 .bdrv_needs_filename = false,
1321 .bdrv_file_open = qemu_gluster_open,
1322 .bdrv_reopen_prepare = qemu_gluster_reopen_prepare,
1323 .bdrv_reopen_commit = qemu_gluster_reopen_commit,
1324 .bdrv_reopen_abort = qemu_gluster_reopen_abort,
1325 .bdrv_close = qemu_gluster_close,
1326 .bdrv_create = qemu_gluster_create,
1327 .bdrv_getlength = qemu_gluster_getlength,
1328 .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size,
1329 .bdrv_truncate = qemu_gluster_truncate,
1330 .bdrv_co_readv = qemu_gluster_co_readv,
1331 .bdrv_co_writev = qemu_gluster_co_writev,
1332 .bdrv_co_flush_to_disk = qemu_gluster_co_flush_to_disk,
1333 .bdrv_has_zero_init = qemu_gluster_has_zero_init,
1334 #ifdef CONFIG_GLUSTERFS_DISCARD
1335 .bdrv_co_pdiscard = qemu_gluster_co_pdiscard,
1337 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1338 .bdrv_co_pwrite_zeroes = qemu_gluster_co_pwrite_zeroes,
1340 .bdrv_co_get_block_status = qemu_gluster_co_get_block_status,
1341 .create_opts = &qemu_gluster_create_opts,
1344 static BlockDriver bdrv_gluster_unix = {
1345 .format_name = "gluster",
1346 .protocol_name = "gluster+unix",
1347 .instance_size = sizeof(BDRVGlusterState),
1348 .bdrv_needs_filename = true,
1349 .bdrv_file_open = qemu_gluster_open,
1350 .bdrv_reopen_prepare = qemu_gluster_reopen_prepare,
1351 .bdrv_reopen_commit = qemu_gluster_reopen_commit,
1352 .bdrv_reopen_abort = qemu_gluster_reopen_abort,
1353 .bdrv_close = qemu_gluster_close,
1354 .bdrv_create = qemu_gluster_create,
1355 .bdrv_getlength = qemu_gluster_getlength,
1356 .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size,
1357 .bdrv_truncate = qemu_gluster_truncate,
1358 .bdrv_co_readv = qemu_gluster_co_readv,
1359 .bdrv_co_writev = qemu_gluster_co_writev,
1360 .bdrv_co_flush_to_disk = qemu_gluster_co_flush_to_disk,
1361 .bdrv_has_zero_init = qemu_gluster_has_zero_init,
1362 #ifdef CONFIG_GLUSTERFS_DISCARD
1363 .bdrv_co_pdiscard = qemu_gluster_co_pdiscard,
1365 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1366 .bdrv_co_pwrite_zeroes = qemu_gluster_co_pwrite_zeroes,
1368 .bdrv_co_get_block_status = qemu_gluster_co_get_block_status,
1369 .create_opts = &qemu_gluster_create_opts,
1372 /* rdma is deprecated (actually never supported for volfile fetch).
1373 * Let's maintain it for the protocol compatibility, to make sure things
1374 * won't break immediately. For now, gluster+rdma will fall back to gluster+tcp
1375 * protocol with a warning.
1376 * TODO: remove gluster+rdma interface support
1378 static BlockDriver bdrv_gluster_rdma = {
1379 .format_name = "gluster",
1380 .protocol_name = "gluster+rdma",
1381 .instance_size = sizeof(BDRVGlusterState),
1382 .bdrv_needs_filename = true,
1383 .bdrv_file_open = qemu_gluster_open,
1384 .bdrv_reopen_prepare = qemu_gluster_reopen_prepare,
1385 .bdrv_reopen_commit = qemu_gluster_reopen_commit,
1386 .bdrv_reopen_abort = qemu_gluster_reopen_abort,
1387 .bdrv_close = qemu_gluster_close,
1388 .bdrv_create = qemu_gluster_create,
1389 .bdrv_getlength = qemu_gluster_getlength,
1390 .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size,
1391 .bdrv_truncate = qemu_gluster_truncate,
1392 .bdrv_co_readv = qemu_gluster_co_readv,
1393 .bdrv_co_writev = qemu_gluster_co_writev,
1394 .bdrv_co_flush_to_disk = qemu_gluster_co_flush_to_disk,
1395 .bdrv_has_zero_init = qemu_gluster_has_zero_init,
1396 #ifdef CONFIG_GLUSTERFS_DISCARD
1397 .bdrv_co_pdiscard = qemu_gluster_co_pdiscard,
1399 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1400 .bdrv_co_pwrite_zeroes = qemu_gluster_co_pwrite_zeroes,
1402 .bdrv_co_get_block_status = qemu_gluster_co_get_block_status,
1403 .create_opts = &qemu_gluster_create_opts,
1406 static void bdrv_gluster_init(void)
1408 bdrv_register(&bdrv_gluster_rdma);
1409 bdrv_register(&bdrv_gluster_unix);
1410 bdrv_register(&bdrv_gluster_tcp);
1411 bdrv_register(&bdrv_gluster);
1414 block_init(bdrv_gluster_init);