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"
17 #include "qemu/cutils.h"
19 #define GLUSTER_OPT_FILENAME "filename"
20 #define GLUSTER_OPT_VOLUME "volume"
21 #define GLUSTER_OPT_PATH "path"
22 #define GLUSTER_OPT_TYPE "type"
23 #define GLUSTER_OPT_SERVER_PATTERN "server."
24 #define GLUSTER_OPT_HOST "host"
25 #define GLUSTER_OPT_PORT "port"
26 #define GLUSTER_OPT_TO "to"
27 #define GLUSTER_OPT_IPV4 "ipv4"
28 #define GLUSTER_OPT_IPV6 "ipv6"
29 #define GLUSTER_OPT_SOCKET "socket"
30 #define GLUSTER_OPT_DEBUG "debug"
31 #define GLUSTER_DEFAULT_PORT 24007
32 #define GLUSTER_DEBUG_DEFAULT 4
33 #define GLUSTER_DEBUG_MAX 9
34 #define GLUSTER_OPT_LOGFILE "logfile"
35 #define GLUSTER_LOGFILE_DEFAULT "-" /* handled in libgfapi as /dev/stderr */
37 #define GERR_INDEX_HINT "hint: check in 'server' array index '%d'\n"
39 typedef struct GlusterAIOCB {
43 AioContext *aio_context;
46 typedef struct BDRVGlusterState {
50 bool supports_seek_data;
54 typedef struct BDRVGlusterReopenState {
57 } BDRVGlusterReopenState;
60 typedef struct GlfsPreopened {
66 typedef struct ListElement {
67 QLIST_ENTRY(ListElement) list;
71 static QLIST_HEAD(glfs_list, ListElement) glfs_list;
73 static QemuOptsList qemu_gluster_create_opts = {
74 .name = "qemu-gluster-create-opts",
75 .head = QTAILQ_HEAD_INITIALIZER(qemu_gluster_create_opts.head),
78 .name = BLOCK_OPT_SIZE,
79 .type = QEMU_OPT_SIZE,
80 .help = "Virtual disk size"
83 .name = BLOCK_OPT_PREALLOC,
84 .type = QEMU_OPT_STRING,
85 .help = "Preallocation mode (allowed values: off, full)"
88 .name = GLUSTER_OPT_DEBUG,
89 .type = QEMU_OPT_NUMBER,
90 .help = "Gluster log level, valid range is 0-9",
93 .name = GLUSTER_OPT_LOGFILE,
94 .type = QEMU_OPT_STRING,
95 .help = "Logfile path of libgfapi",
101 static QemuOptsList runtime_opts = {
103 .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
106 .name = GLUSTER_OPT_FILENAME,
107 .type = QEMU_OPT_STRING,
108 .help = "URL to the gluster image",
111 .name = GLUSTER_OPT_DEBUG,
112 .type = QEMU_OPT_NUMBER,
113 .help = "Gluster log level, valid range is 0-9",
116 .name = GLUSTER_OPT_LOGFILE,
117 .type = QEMU_OPT_STRING,
118 .help = "Logfile path of libgfapi",
120 { /* end of list */ }
124 static QemuOptsList runtime_json_opts = {
125 .name = "gluster_json",
126 .head = QTAILQ_HEAD_INITIALIZER(runtime_json_opts.head),
129 .name = GLUSTER_OPT_VOLUME,
130 .type = QEMU_OPT_STRING,
131 .help = "name of gluster volume where VM image resides",
134 .name = GLUSTER_OPT_PATH,
135 .type = QEMU_OPT_STRING,
136 .help = "absolute path to image file in gluster volume",
139 .name = GLUSTER_OPT_DEBUG,
140 .type = QEMU_OPT_NUMBER,
141 .help = "Gluster log level, valid range is 0-9",
143 { /* end of list */ }
147 static QemuOptsList runtime_type_opts = {
148 .name = "gluster_type",
149 .head = QTAILQ_HEAD_INITIALIZER(runtime_type_opts.head),
152 .name = GLUSTER_OPT_TYPE,
153 .type = QEMU_OPT_STRING,
156 { /* end of list */ }
160 static QemuOptsList runtime_unix_opts = {
161 .name = "gluster_unix",
162 .head = QTAILQ_HEAD_INITIALIZER(runtime_unix_opts.head),
165 .name = GLUSTER_OPT_SOCKET,
166 .type = QEMU_OPT_STRING,
167 .help = "socket file path)",
169 { /* end of list */ }
173 static QemuOptsList runtime_inet_opts = {
174 .name = "gluster_inet",
175 .head = QTAILQ_HEAD_INITIALIZER(runtime_inet_opts.head),
178 .name = GLUSTER_OPT_TYPE,
179 .type = QEMU_OPT_STRING,
183 .name = GLUSTER_OPT_HOST,
184 .type = QEMU_OPT_STRING,
185 .help = "host address (hostname/ipv4/ipv6 addresses)",
188 .name = GLUSTER_OPT_PORT,
189 .type = QEMU_OPT_STRING,
190 .help = "port number on which glusterd is listening (default 24007)",
194 .type = QEMU_OPT_NUMBER,
195 .help = "max port number, not supported by gluster",
199 .type = QEMU_OPT_BOOL,
200 .help = "ipv4 bool value, not supported by gluster",
204 .type = QEMU_OPT_BOOL,
205 .help = "ipv6 bool value, not supported by gluster",
207 { /* end of list */ }
211 static void glfs_set_preopened(const char *volume, glfs_t *fs)
213 ListElement *entry = NULL;
215 entry = g_new(ListElement, 1);
217 entry->saved.volume = g_strdup(volume);
219 entry->saved.fs = fs;
220 entry->saved.ref = 1;
222 QLIST_INSERT_HEAD(&glfs_list, entry, list);
225 static glfs_t *glfs_find_preopened(const char *volume)
227 ListElement *entry = NULL;
229 QLIST_FOREACH(entry, &glfs_list, list) {
230 if (strcmp(entry->saved.volume, volume) == 0) {
232 return entry->saved.fs;
239 static void glfs_clear_preopened(glfs_t *fs)
241 ListElement *entry = NULL;
248 QLIST_FOREACH_SAFE(entry, &glfs_list, list, next) {
249 if (entry->saved.fs == fs) {
250 if (--entry->saved.ref) {
254 QLIST_REMOVE(entry, list);
256 glfs_fini(entry->saved.fs);
257 g_free(entry->saved.volume);
263 static int parse_volume_options(BlockdevOptionsGluster *gconf, char *path)
272 p = q = path + strspn(path, "/");
273 p += strcspn(p, "/");
277 gconf->volume = g_strndup(q, p - q);
284 gconf->path = g_strdup(p);
289 * file=gluster[+transport]://[host[:port]]/volume/path[?socket=...]
291 * 'gluster' is the protocol.
293 * 'transport' specifies the transport type used to connect to gluster
294 * management daemon (glusterd). Valid transport types are
295 * tcp or unix. If a transport type isn't specified, then tcp type is assumed.
297 * 'host' specifies the host where the volume file specification for
298 * the given volume resides. This can be either hostname or ipv4 address.
299 * If transport type is 'unix', then 'host' field should not be specified.
300 * The 'socket' field needs to be populated with the path to unix domain
303 * 'port' is the port number on which glusterd is listening. This is optional
304 * and if not specified, QEMU will send 0 which will make gluster to use the
305 * default port. If the transport type is unix, then 'port' should not be
308 * 'volume' is the name of the gluster volume which contains the VM image.
310 * 'path' is the path to the actual VM image that resides on gluster volume.
314 * file=gluster://1.2.3.4/testvol/a.img
315 * file=gluster+tcp://1.2.3.4/testvol/a.img
316 * file=gluster+tcp://1.2.3.4:24007/testvol/dir/a.img
317 * file=gluster+tcp://host.domain.com:24007/testvol/dir/a.img
318 * file=gluster+unix:///testvol/dir/a.img?socket=/tmp/glusterd.socket
320 static int qemu_gluster_parse_uri(BlockdevOptionsGluster *gconf,
321 const char *filename)
323 SocketAddress *gsconf;
325 QueryParams *qp = NULL;
326 bool is_unix = false;
329 uri = uri_parse(filename);
334 gconf->server = g_new0(SocketAddressList, 1);
335 gconf->server->value = gsconf = g_new0(SocketAddress, 1);
338 if (!uri->scheme || !strcmp(uri->scheme, "gluster")) {
339 gsconf->type = SOCKET_ADDRESS_TYPE_INET;
340 } else if (!strcmp(uri->scheme, "gluster+tcp")) {
341 gsconf->type = SOCKET_ADDRESS_TYPE_INET;
342 } else if (!strcmp(uri->scheme, "gluster+unix")) {
343 gsconf->type = SOCKET_ADDRESS_TYPE_UNIX;
345 } else if (!strcmp(uri->scheme, "gluster+rdma")) {
346 gsconf->type = SOCKET_ADDRESS_TYPE_INET;
347 warn_report("rdma feature is not supported, falling back to tcp");
353 ret = parse_volume_options(gconf, uri->path);
358 qp = query_params_parse(uri->query);
359 if (qp->n > 1 || (is_unix && !qp->n) || (!is_unix && qp->n)) {
365 if (uri->server || uri->port) {
369 if (strcmp(qp->p[0].name, "socket")) {
373 gsconf->u.q_unix.path = g_strdup(qp->p[0].value);
375 gsconf->u.inet.host = g_strdup(uri->server ? uri->server : "localhost");
377 gsconf->u.inet.port = g_strdup_printf("%d", uri->port);
379 gsconf->u.inet.port = g_strdup_printf("%d", GLUSTER_DEFAULT_PORT);
385 query_params_free(qp);
391 static struct glfs *qemu_gluster_glfs_init(BlockdevOptionsGluster *gconf,
397 SocketAddressList *server;
398 unsigned long long port;
400 glfs = glfs_find_preopened(gconf->volume);
405 glfs = glfs_new(gconf->volume);
410 glfs_set_preopened(gconf->volume, glfs);
412 for (server = gconf->server; server; server = server->next) {
413 switch (server->value->type) {
414 case SOCKET_ADDRESS_TYPE_UNIX:
415 ret = glfs_set_volfile_server(glfs, "unix",
416 server->value->u.q_unix.path, 0);
418 case SOCKET_ADDRESS_TYPE_INET:
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,
430 case SOCKET_ADDRESS_TYPE_VSOCK:
431 case SOCKET_ADDRESS_TYPE_FD:
441 ret = glfs_set_logging(glfs, gconf->logfile, gconf->debug);
446 ret = glfs_init(glfs);
448 error_setg(errp, "Gluster connection for volume %s, path %s failed"
449 " to connect", gconf->volume, gconf->path);
450 for (server = gconf->server; server; server = server->next) {
451 if (server->value->type == SOCKET_ADDRESS_TYPE_UNIX) {
452 error_append_hint(errp, "hint: failed on socket %s ",
453 server->value->u.q_unix.path);
455 error_append_hint(errp, "hint: failed on host %s and port %s ",
456 server->value->u.inet.host,
457 server->value->u.inet.port);
461 error_append_hint(errp, "Please refer to gluster logs for more info\n");
463 /* glfs_init sometimes doesn't set errno although docs suggest that */
475 glfs_clear_preopened(glfs);
482 * Convert the json formatted command line into qapi.
484 static int qemu_gluster_parse_json(BlockdevOptionsGluster *gconf,
485 QDict *options, Error **errp)
488 SocketAddress *gsconf = NULL;
489 SocketAddressList *curr = NULL;
490 QDict *backing_options = NULL;
491 Error *local_err = NULL;
494 int i, type, num_servers;
496 /* create opts info from runtime_json_opts list */
497 opts = qemu_opts_create(&runtime_json_opts, NULL, 0, &error_abort);
498 qemu_opts_absorb_qdict(opts, options, &local_err);
503 num_servers = qdict_array_entries(options, GLUSTER_OPT_SERVER_PATTERN);
504 if (num_servers < 1) {
505 error_setg(&local_err, QERR_MISSING_PARAMETER, "server");
509 ptr = qemu_opt_get(opts, GLUSTER_OPT_VOLUME);
511 error_setg(&local_err, QERR_MISSING_PARAMETER, GLUSTER_OPT_VOLUME);
514 gconf->volume = g_strdup(ptr);
516 ptr = qemu_opt_get(opts, GLUSTER_OPT_PATH);
518 error_setg(&local_err, QERR_MISSING_PARAMETER, GLUSTER_OPT_PATH);
521 gconf->path = g_strdup(ptr);
524 for (i = 0; i < num_servers; i++) {
525 str = g_strdup_printf(GLUSTER_OPT_SERVER_PATTERN"%d.", i);
526 qdict_extract_subqdict(options, &backing_options, str);
528 /* create opts info from runtime_type_opts list */
529 opts = qemu_opts_create(&runtime_type_opts, NULL, 0, &error_abort);
530 qemu_opts_absorb_qdict(opts, backing_options, &local_err);
535 ptr = qemu_opt_get(opts, GLUSTER_OPT_TYPE);
537 error_setg(&local_err, QERR_MISSING_PARAMETER, GLUSTER_OPT_TYPE);
538 error_append_hint(&local_err, GERR_INDEX_HINT, i);
542 gsconf = g_new0(SocketAddress, 1);
543 if (!strcmp(ptr, "tcp")) {
544 ptr = "inet"; /* accept legacy "tcp" */
546 type = qapi_enum_parse(&SocketAddressType_lookup, ptr, -1, NULL);
547 if (type != SOCKET_ADDRESS_TYPE_INET
548 && type != SOCKET_ADDRESS_TYPE_UNIX) {
549 error_setg(&local_err,
550 "Parameter '%s' may be 'inet' or 'unix'",
552 error_append_hint(&local_err, GERR_INDEX_HINT, i);
558 if (gsconf->type == SOCKET_ADDRESS_TYPE_INET) {
559 /* create opts info from runtime_inet_opts list */
560 opts = qemu_opts_create(&runtime_inet_opts, NULL, 0, &error_abort);
561 qemu_opts_absorb_qdict(opts, backing_options, &local_err);
566 ptr = qemu_opt_get(opts, GLUSTER_OPT_HOST);
568 error_setg(&local_err, QERR_MISSING_PARAMETER,
570 error_append_hint(&local_err, GERR_INDEX_HINT, i);
573 gsconf->u.inet.host = g_strdup(ptr);
574 ptr = qemu_opt_get(opts, GLUSTER_OPT_PORT);
576 error_setg(&local_err, QERR_MISSING_PARAMETER,
578 error_append_hint(&local_err, GERR_INDEX_HINT, i);
581 gsconf->u.inet.port = g_strdup(ptr);
583 /* defend for unsupported fields in InetSocketAddress,
584 * i.e. @ipv4, @ipv6 and @to
586 ptr = qemu_opt_get(opts, GLUSTER_OPT_TO);
588 gsconf->u.inet.has_to = true;
590 ptr = qemu_opt_get(opts, GLUSTER_OPT_IPV4);
592 gsconf->u.inet.has_ipv4 = true;
594 ptr = qemu_opt_get(opts, GLUSTER_OPT_IPV6);
596 gsconf->u.inet.has_ipv6 = true;
598 if (gsconf->u.inet.has_to) {
599 error_setg(&local_err, "Parameter 'to' not supported");
602 if (gsconf->u.inet.has_ipv4 || gsconf->u.inet.has_ipv6) {
603 error_setg(&local_err, "Parameters 'ipv4/ipv6' not supported");
608 /* create opts info from runtime_unix_opts list */
609 opts = qemu_opts_create(&runtime_unix_opts, NULL, 0, &error_abort);
610 qemu_opts_absorb_qdict(opts, backing_options, &local_err);
615 ptr = qemu_opt_get(opts, GLUSTER_OPT_SOCKET);
617 error_setg(&local_err, QERR_MISSING_PARAMETER,
619 error_append_hint(&local_err, GERR_INDEX_HINT, i);
622 gsconf->u.q_unix.path = g_strdup(ptr);
626 if (gconf->server == NULL) {
627 gconf->server = g_new0(SocketAddressList, 1);
628 gconf->server->value = gsconf;
629 curr = gconf->server;
631 curr->next = g_new0(SocketAddressList, 1);
632 curr->next->value = gsconf;
637 QDECREF(backing_options);
638 backing_options = NULL;
646 error_propagate(errp, local_err);
647 qapi_free_SocketAddress(gsconf);
650 QDECREF(backing_options);
655 static struct glfs *qemu_gluster_init(BlockdevOptionsGluster *gconf,
656 const char *filename,
657 QDict *options, Error **errp)
661 ret = qemu_gluster_parse_uri(gconf, filename);
663 error_setg(errp, "invalid URI");
664 error_append_hint(errp, "Usage: file=gluster[+transport]://"
665 "[host[:port]]volume/path[?socket=...]"
667 "[,file.logfile=/path/filename.log]\n");
672 ret = qemu_gluster_parse_json(gconf, options, errp);
674 error_append_hint(errp, "Usage: "
675 "-drive driver=qcow2,file.driver=gluster,"
676 "file.volume=testvol,file.path=/path/a.qcow2"
678 "[,file.logfile=/path/filename.log],"
679 "file.server.0.type=inet,"
680 "file.server.0.host=1.2.3.4,"
681 "file.server.0.port=24007,"
682 "file.server.1.transport=unix,"
683 "file.server.1.socket=/var/run/glusterd.socket ..."
691 return qemu_gluster_glfs_init(gconf, errp);
695 * AIO callback routine called from GlusterFS thread.
697 static void gluster_finish_aiocb(struct glfs_fd *fd, ssize_t ret, void *arg)
699 GlusterAIOCB *acb = (GlusterAIOCB *)arg;
701 if (!ret || ret == acb->size) {
702 acb->ret = 0; /* Success */
703 } else if (ret < 0) {
704 acb->ret = -errno; /* Read/Write failed */
706 acb->ret = -EIO; /* Partial read/write - fail it */
709 aio_co_schedule(acb->aio_context, acb->coroutine);
712 static void qemu_gluster_parse_flags(int bdrv_flags, int *open_flags)
714 assert(open_flags != NULL);
716 *open_flags |= O_BINARY;
718 if (bdrv_flags & BDRV_O_RDWR) {
719 *open_flags |= O_RDWR;
721 *open_flags |= O_RDONLY;
724 if ((bdrv_flags & BDRV_O_NOCACHE)) {
725 *open_flags |= O_DIRECT;
730 * Do SEEK_DATA/HOLE to detect if it is functional. Older broken versions of
731 * gfapi incorrectly return the current offset when SEEK_DATA/HOLE is used.
732 * - Corrected versions return -1 and set errno to EINVAL.
733 * - Versions that support SEEK_DATA/HOLE correctly, will return -1 and set
734 * errno to ENXIO when SEEK_DATA is called with a position of EOF.
736 static bool qemu_gluster_test_seek(struct glfs_fd *fd)
740 #if defined SEEK_HOLE && defined SEEK_DATA
743 eof = glfs_lseek(fd, 0, SEEK_END);
745 /* this should never occur */
749 /* this should always fail with ENXIO if SEEK_DATA is supported */
750 ret = glfs_lseek(fd, eof, SEEK_DATA);
753 return (ret < 0) && (errno == ENXIO);
756 static int qemu_gluster_open(BlockDriverState *bs, QDict *options,
757 int bdrv_flags, Error **errp)
759 BDRVGlusterState *s = bs->opaque;
762 BlockdevOptionsGluster *gconf = NULL;
764 Error *local_err = NULL;
765 const char *filename, *logfile;
767 opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
768 qemu_opts_absorb_qdict(opts, options, &local_err);
770 error_propagate(errp, local_err);
775 filename = qemu_opt_get(opts, GLUSTER_OPT_FILENAME);
777 s->debug = qemu_opt_get_number(opts, GLUSTER_OPT_DEBUG,
778 GLUSTER_DEBUG_DEFAULT);
781 } else if (s->debug > GLUSTER_DEBUG_MAX) {
782 s->debug = GLUSTER_DEBUG_MAX;
785 gconf = g_new0(BlockdevOptionsGluster, 1);
786 gconf->debug = s->debug;
787 gconf->has_debug = true;
789 logfile = qemu_opt_get(opts, GLUSTER_OPT_LOGFILE);
790 s->logfile = g_strdup(logfile ? logfile : GLUSTER_LOGFILE_DEFAULT);
792 gconf->logfile = g_strdup(s->logfile);
793 gconf->has_logfile = true;
795 s->glfs = qemu_gluster_init(gconf, filename, options, errp);
801 #ifdef CONFIG_GLUSTERFS_XLATOR_OPT
802 /* Without this, if fsync fails for a recoverable reason (for instance,
803 * ENOSPC), gluster will dump its cache, preventing retries. This means
804 * almost certain data loss. Not all gluster versions support the
805 * 'resync-failed-syncs-after-fsync' key value, but there is no way to
806 * discover during runtime if it is supported (this api returns success for
807 * unknown key/value pairs) */
808 ret = glfs_set_xlator_option(s->glfs, "*-write-behind",
809 "resync-failed-syncs-after-fsync",
812 error_setg_errno(errp, errno, "Unable to set xlator key/value pair");
818 qemu_gluster_parse_flags(bdrv_flags, &open_flags);
820 s->fd = glfs_open(s->glfs, gconf->path, open_flags);
825 s->supports_seek_data = qemu_gluster_test_seek(s->fd);
829 qapi_free_BlockdevOptionsGluster(gconf);
838 glfs_clear_preopened(s->glfs);
843 static int qemu_gluster_reopen_prepare(BDRVReopenState *state,
844 BlockReopenQueue *queue, Error **errp)
848 BDRVGlusterReopenState *reop_s;
849 BlockdevOptionsGluster *gconf;
852 assert(state != NULL);
853 assert(state->bs != NULL);
855 s = state->bs->opaque;
857 state->opaque = g_new0(BDRVGlusterReopenState, 1);
858 reop_s = state->opaque;
860 qemu_gluster_parse_flags(state->flags, &open_flags);
862 gconf = g_new0(BlockdevOptionsGluster, 1);
863 gconf->debug = s->debug;
864 gconf->has_debug = true;
865 gconf->logfile = g_strdup(s->logfile);
866 gconf->has_logfile = true;
867 reop_s->glfs = qemu_gluster_init(gconf, state->bs->filename, NULL, errp);
868 if (reop_s->glfs == NULL) {
873 #ifdef CONFIG_GLUSTERFS_XLATOR_OPT
874 ret = glfs_set_xlator_option(reop_s->glfs, "*-write-behind",
875 "resync-failed-syncs-after-fsync", "on");
877 error_setg_errno(errp, errno, "Unable to set xlator key/value pair");
883 reop_s->fd = glfs_open(reop_s->glfs, gconf->path, open_flags);
884 if (reop_s->fd == NULL) {
885 /* reops->glfs will be cleaned up in _abort */
891 /* state->opaque will be freed in either the _abort or _commit */
892 qapi_free_BlockdevOptionsGluster(gconf);
896 static void qemu_gluster_reopen_commit(BDRVReopenState *state)
898 BDRVGlusterReopenState *reop_s = state->opaque;
899 BDRVGlusterState *s = state->bs->opaque;
907 glfs_clear_preopened(s->glfs);
909 /* use the newly opened image / connection */
911 s->glfs = reop_s->glfs;
913 g_free(state->opaque);
914 state->opaque = NULL;
920 static void qemu_gluster_reopen_abort(BDRVReopenState *state)
922 BDRVGlusterReopenState *reop_s = state->opaque;
924 if (reop_s == NULL) {
929 glfs_close(reop_s->fd);
932 glfs_clear_preopened(reop_s->glfs);
934 g_free(state->opaque);
935 state->opaque = NULL;
940 #ifdef CONFIG_GLUSTERFS_ZEROFILL
941 static coroutine_fn int qemu_gluster_co_pwrite_zeroes(BlockDriverState *bs,
944 BdrvRequestFlags flags)
948 BDRVGlusterState *s = bs->opaque;
952 acb.coroutine = qemu_coroutine_self();
953 acb.aio_context = bdrv_get_aio_context(bs);
955 ret = glfs_zerofill_async(s->fd, offset, size, gluster_finish_aiocb, &acb);
960 qemu_coroutine_yield();
965 static int qemu_gluster_create(const char *filename,
966 QemuOpts *opts, Error **errp)
968 BlockdevOptionsGluster *gconf;
972 PreallocMode prealloc;
973 int64_t total_size = 0;
975 Error *local_err = NULL;
977 gconf = g_new0(BlockdevOptionsGluster, 1);
978 gconf->debug = qemu_opt_get_number_del(opts, GLUSTER_OPT_DEBUG,
979 GLUSTER_DEBUG_DEFAULT);
980 if (gconf->debug < 0) {
982 } else if (gconf->debug > GLUSTER_DEBUG_MAX) {
983 gconf->debug = GLUSTER_DEBUG_MAX;
985 gconf->has_debug = true;
987 gconf->logfile = qemu_opt_get_del(opts, GLUSTER_OPT_LOGFILE);
988 if (!gconf->logfile) {
989 gconf->logfile = g_strdup(GLUSTER_LOGFILE_DEFAULT);
991 gconf->has_logfile = true;
993 glfs = qemu_gluster_init(gconf, filename, NULL, errp);
999 total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
1002 tmp = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
1003 prealloc = qapi_enum_parse(&PreallocMode_lookup, tmp, PREALLOC_MODE_OFF,
1007 error_propagate(errp, local_err);
1012 fd = glfs_creat(glfs, gconf->path,
1013 O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, S_IRUSR | S_IWUSR);
1020 #ifdef CONFIG_GLUSTERFS_FALLOCATE
1021 case PREALLOC_MODE_FALLOC:
1022 if (glfs_fallocate(fd, 0, 0, total_size)) {
1023 error_setg(errp, "Could not preallocate data for the new file");
1027 #endif /* CONFIG_GLUSTERFS_FALLOCATE */
1028 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1029 case PREALLOC_MODE_FULL:
1030 if (!glfs_ftruncate(fd, total_size)) {
1031 if (glfs_zerofill(fd, 0, total_size)) {
1032 error_setg(errp, "Could not zerofill the new file");
1036 error_setg(errp, "Could not resize file");
1040 #endif /* CONFIG_GLUSTERFS_ZEROFILL */
1041 case PREALLOC_MODE_OFF:
1042 if (glfs_ftruncate(fd, total_size) != 0) {
1044 error_setg(errp, "Could not resize file");
1049 error_setg(errp, "Unsupported preallocation mode: %s",
1050 PreallocMode_str(prealloc));
1054 if (glfs_close(fd) != 0) {
1058 qapi_free_BlockdevOptionsGluster(gconf);
1059 glfs_clear_preopened(glfs);
1063 static coroutine_fn int qemu_gluster_co_rw(BlockDriverState *bs,
1064 int64_t sector_num, int nb_sectors,
1065 QEMUIOVector *qiov, int write)
1069 BDRVGlusterState *s = bs->opaque;
1070 size_t size = nb_sectors * BDRV_SECTOR_SIZE;
1071 off_t offset = sector_num * BDRV_SECTOR_SIZE;
1075 acb.coroutine = qemu_coroutine_self();
1076 acb.aio_context = bdrv_get_aio_context(bs);
1079 ret = glfs_pwritev_async(s->fd, qiov->iov, qiov->niov, offset, 0,
1080 gluster_finish_aiocb, &acb);
1082 ret = glfs_preadv_async(s->fd, qiov->iov, qiov->niov, offset, 0,
1083 gluster_finish_aiocb, &acb);
1090 qemu_coroutine_yield();
1094 static int qemu_gluster_truncate(BlockDriverState *bs, int64_t offset,
1095 PreallocMode prealloc, Error **errp)
1098 BDRVGlusterState *s = bs->opaque;
1100 if (prealloc != PREALLOC_MODE_OFF) {
1101 error_setg(errp, "Unsupported preallocation mode '%s'",
1102 PreallocMode_str(prealloc));
1106 ret = glfs_ftruncate(s->fd, offset);
1109 error_setg_errno(errp, -ret, "Failed to truncate file");
1116 static coroutine_fn int qemu_gluster_co_readv(BlockDriverState *bs,
1121 return qemu_gluster_co_rw(bs, sector_num, nb_sectors, qiov, 0);
1124 static coroutine_fn int qemu_gluster_co_writev(BlockDriverState *bs,
1129 return qemu_gluster_co_rw(bs, sector_num, nb_sectors, qiov, 1);
1132 static void qemu_gluster_close(BlockDriverState *bs)
1134 BDRVGlusterState *s = bs->opaque;
1141 glfs_clear_preopened(s->glfs);
1144 static coroutine_fn int qemu_gluster_co_flush_to_disk(BlockDriverState *bs)
1148 BDRVGlusterState *s = bs->opaque;
1152 acb.coroutine = qemu_coroutine_self();
1153 acb.aio_context = bdrv_get_aio_context(bs);
1155 ret = glfs_fsync_async(s->fd, gluster_finish_aiocb, &acb);
1161 qemu_coroutine_yield();
1170 /* Some versions of Gluster (3.5.6 -> 3.5.8?) will not retain its cache
1171 * after a fsync failure, so we have no way of allowing the guest to safely
1172 * continue. Gluster versions prior to 3.5.6 don't retain the cache
1173 * either, but will invalidate the fd on error, so this is again our only
1176 * The 'resync-failed-syncs-after-fsync' xlator option for the
1177 * write-behind cache will cause later gluster versions to retain its
1178 * cache after error, so long as the fd remains open. However, we
1179 * currently have no way of knowing if this option is supported.
1181 * TODO: Once gluster provides a way for us to determine if the option
1182 * is supported, bypass the closure and setting drv to NULL. */
1183 qemu_gluster_close(bs);
1188 #ifdef CONFIG_GLUSTERFS_DISCARD
1189 static coroutine_fn int qemu_gluster_co_pdiscard(BlockDriverState *bs,
1190 int64_t offset, int size)
1194 BDRVGlusterState *s = bs->opaque;
1198 acb.coroutine = qemu_coroutine_self();
1199 acb.aio_context = bdrv_get_aio_context(bs);
1201 ret = glfs_discard_async(s->fd, offset, size, gluster_finish_aiocb, &acb);
1206 qemu_coroutine_yield();
1211 static int64_t qemu_gluster_getlength(BlockDriverState *bs)
1213 BDRVGlusterState *s = bs->opaque;
1216 ret = glfs_lseek(s->fd, 0, SEEK_END);
1224 static int64_t qemu_gluster_allocated_file_size(BlockDriverState *bs)
1226 BDRVGlusterState *s = bs->opaque;
1230 ret = glfs_fstat(s->fd, &st);
1234 return st.st_blocks * 512;
1238 static int qemu_gluster_has_zero_init(BlockDriverState *bs)
1240 /* GlusterFS volume could be backed by a block device */
1245 * Find allocation range in @bs around offset @start.
1246 * May change underlying file descriptor's file offset.
1247 * If @start is not in a hole, store @start in @data, and the
1248 * beginning of the next hole in @hole, and return 0.
1249 * If @start is in a non-trailing hole, store @start in @hole and the
1250 * beginning of the next non-hole in @data, and return 0.
1251 * If @start is in a trailing hole or beyond EOF, return -ENXIO.
1252 * If we can't find out, return a negative errno other than -ENXIO.
1254 * (Shamefully copied from file-posix.c, only miniscule adaptions.)
1256 static int find_allocation(BlockDriverState *bs, off_t start,
1257 off_t *data, off_t *hole)
1259 BDRVGlusterState *s = bs->opaque;
1261 if (!s->supports_seek_data) {
1265 #if defined SEEK_HOLE && defined SEEK_DATA
1270 * D1. offs == start: start is in data
1271 * D2. offs > start: start is in a hole, next data at offs
1272 * D3. offs < 0, errno = ENXIO: either start is in a trailing hole
1273 * or start is beyond EOF
1274 * If the latter happens, the file has been truncated behind
1275 * our back since we opened it. All bets are off then.
1276 * Treating like a trailing hole is simplest.
1277 * D4. offs < 0, errno != ENXIO: we learned nothing
1279 offs = glfs_lseek(s->fd, start, SEEK_DATA);
1281 return -errno; /* D3 or D4 */
1285 /* This is not a valid return by lseek(). We are safe to just return
1286 * -EIO in this case, and we'll treat it like D4. Unfortunately some
1287 * versions of gluster server will return offs < start, so an assert
1288 * here will unnecessarily abort QEMU. */
1293 /* D2: in hole, next data at offs */
1299 /* D1: in data, end not yet known */
1303 * H1. offs == start: start is in a hole
1304 * If this happens here, a hole has been dug behind our back
1305 * since the previous lseek().
1306 * H2. offs > start: either start is in data, next hole at offs,
1307 * or start is in trailing hole, EOF at offs
1308 * Linux treats trailing holes like any other hole: offs ==
1309 * start. Solaris seeks to EOF instead: offs > start (blech).
1310 * If that happens here, a hole has been dug behind our back
1311 * since the previous lseek().
1312 * H3. offs < 0, errno = ENXIO: start is beyond EOF
1313 * If this happens, the file has been truncated behind our
1314 * back since we opened it. Treat it like a trailing hole.
1315 * H4. offs < 0, errno != ENXIO: we learned nothing
1316 * Pretend we know nothing at all, i.e. "forget" about D1.
1318 offs = glfs_lseek(s->fd, start, SEEK_HOLE);
1320 return -errno; /* D1 and (H3 or H4) */
1324 /* This is not a valid return by lseek(). We are safe to just return
1325 * -EIO in this case, and we'll treat it like H4. Unfortunately some
1326 * versions of gluster server will return offs < start, so an assert
1327 * here will unnecessarily abort QEMU. */
1333 * D1 and H2: either in data, next hole at offs, or it was in
1334 * data but is now in a trailing hole. In the latter case,
1335 * all bets are off. Treating it as if it there was data all
1336 * the way to EOF is safe, so simply do that.
1352 * Returns the allocation status of the specified sectors.
1354 * If 'sector_num' is beyond the end of the disk image the return value is 0
1355 * and 'pnum' is set to 0.
1357 * 'pnum' is set to the number of sectors (including and immediately following
1358 * the specified sector) that are known to be in the same
1359 * allocated/unallocated state.
1361 * 'nb_sectors' is the max value 'pnum' should be set to. If nb_sectors goes
1362 * beyond the end of the disk image it will be clamped.
1364 * (Based on raw_co_get_block_status() from file-posix.c.)
1366 static int64_t coroutine_fn qemu_gluster_co_get_block_status(
1367 BlockDriverState *bs, int64_t sector_num, int nb_sectors, int *pnum,
1368 BlockDriverState **file)
1370 BDRVGlusterState *s = bs->opaque;
1371 off_t start, data = 0, hole = 0;
1379 start = sector_num * BDRV_SECTOR_SIZE;
1380 total_size = bdrv_getlength(bs);
1381 if (total_size < 0) {
1383 } else if (start >= total_size) {
1386 } else if (start + nb_sectors * BDRV_SECTOR_SIZE > total_size) {
1387 nb_sectors = DIV_ROUND_UP(total_size - start, BDRV_SECTOR_SIZE);
1390 ret = find_allocation(bs, start, &data, &hole);
1391 if (ret == -ENXIO) {
1394 ret = BDRV_BLOCK_ZERO;
1395 } else if (ret < 0) {
1396 /* No info available, so pretend there are no holes */
1398 ret = BDRV_BLOCK_DATA;
1399 } else if (data == start) {
1400 /* On a data extent, compute sectors to the end of the extent,
1401 * possibly including a partial sector at EOF. */
1402 *pnum = MIN(nb_sectors, DIV_ROUND_UP(hole - start, BDRV_SECTOR_SIZE));
1403 ret = BDRV_BLOCK_DATA;
1405 /* On a hole, compute sectors to the beginning of the next extent. */
1406 assert(hole == start);
1407 *pnum = MIN(nb_sectors, (data - start) / BDRV_SECTOR_SIZE);
1408 ret = BDRV_BLOCK_ZERO;
1413 return ret | BDRV_BLOCK_OFFSET_VALID | start;
1417 static BlockDriver bdrv_gluster = {
1418 .format_name = "gluster",
1419 .protocol_name = "gluster",
1420 .instance_size = sizeof(BDRVGlusterState),
1421 .bdrv_needs_filename = false,
1422 .bdrv_file_open = qemu_gluster_open,
1423 .bdrv_reopen_prepare = qemu_gluster_reopen_prepare,
1424 .bdrv_reopen_commit = qemu_gluster_reopen_commit,
1425 .bdrv_reopen_abort = qemu_gluster_reopen_abort,
1426 .bdrv_close = qemu_gluster_close,
1427 .bdrv_create = qemu_gluster_create,
1428 .bdrv_getlength = qemu_gluster_getlength,
1429 .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size,
1430 .bdrv_truncate = qemu_gluster_truncate,
1431 .bdrv_co_readv = qemu_gluster_co_readv,
1432 .bdrv_co_writev = qemu_gluster_co_writev,
1433 .bdrv_co_flush_to_disk = qemu_gluster_co_flush_to_disk,
1434 .bdrv_has_zero_init = qemu_gluster_has_zero_init,
1435 #ifdef CONFIG_GLUSTERFS_DISCARD
1436 .bdrv_co_pdiscard = qemu_gluster_co_pdiscard,
1438 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1439 .bdrv_co_pwrite_zeroes = qemu_gluster_co_pwrite_zeroes,
1441 .bdrv_co_get_block_status = qemu_gluster_co_get_block_status,
1442 .create_opts = &qemu_gluster_create_opts,
1445 static BlockDriver bdrv_gluster_tcp = {
1446 .format_name = "gluster",
1447 .protocol_name = "gluster+tcp",
1448 .instance_size = sizeof(BDRVGlusterState),
1449 .bdrv_needs_filename = false,
1450 .bdrv_file_open = qemu_gluster_open,
1451 .bdrv_reopen_prepare = qemu_gluster_reopen_prepare,
1452 .bdrv_reopen_commit = qemu_gluster_reopen_commit,
1453 .bdrv_reopen_abort = qemu_gluster_reopen_abort,
1454 .bdrv_close = qemu_gluster_close,
1455 .bdrv_create = qemu_gluster_create,
1456 .bdrv_getlength = qemu_gluster_getlength,
1457 .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size,
1458 .bdrv_truncate = qemu_gluster_truncate,
1459 .bdrv_co_readv = qemu_gluster_co_readv,
1460 .bdrv_co_writev = qemu_gluster_co_writev,
1461 .bdrv_co_flush_to_disk = qemu_gluster_co_flush_to_disk,
1462 .bdrv_has_zero_init = qemu_gluster_has_zero_init,
1463 #ifdef CONFIG_GLUSTERFS_DISCARD
1464 .bdrv_co_pdiscard = qemu_gluster_co_pdiscard,
1466 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1467 .bdrv_co_pwrite_zeroes = qemu_gluster_co_pwrite_zeroes,
1469 .bdrv_co_get_block_status = qemu_gluster_co_get_block_status,
1470 .create_opts = &qemu_gluster_create_opts,
1473 static BlockDriver bdrv_gluster_unix = {
1474 .format_name = "gluster",
1475 .protocol_name = "gluster+unix",
1476 .instance_size = sizeof(BDRVGlusterState),
1477 .bdrv_needs_filename = true,
1478 .bdrv_file_open = qemu_gluster_open,
1479 .bdrv_reopen_prepare = qemu_gluster_reopen_prepare,
1480 .bdrv_reopen_commit = qemu_gluster_reopen_commit,
1481 .bdrv_reopen_abort = qemu_gluster_reopen_abort,
1482 .bdrv_close = qemu_gluster_close,
1483 .bdrv_create = qemu_gluster_create,
1484 .bdrv_getlength = qemu_gluster_getlength,
1485 .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size,
1486 .bdrv_truncate = qemu_gluster_truncate,
1487 .bdrv_co_readv = qemu_gluster_co_readv,
1488 .bdrv_co_writev = qemu_gluster_co_writev,
1489 .bdrv_co_flush_to_disk = qemu_gluster_co_flush_to_disk,
1490 .bdrv_has_zero_init = qemu_gluster_has_zero_init,
1491 #ifdef CONFIG_GLUSTERFS_DISCARD
1492 .bdrv_co_pdiscard = qemu_gluster_co_pdiscard,
1494 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1495 .bdrv_co_pwrite_zeroes = qemu_gluster_co_pwrite_zeroes,
1497 .bdrv_co_get_block_status = qemu_gluster_co_get_block_status,
1498 .create_opts = &qemu_gluster_create_opts,
1501 /* rdma is deprecated (actually never supported for volfile fetch).
1502 * Let's maintain it for the protocol compatibility, to make sure things
1503 * won't break immediately. For now, gluster+rdma will fall back to gluster+tcp
1504 * protocol with a warning.
1505 * TODO: remove gluster+rdma interface support
1507 static BlockDriver bdrv_gluster_rdma = {
1508 .format_name = "gluster",
1509 .protocol_name = "gluster+rdma",
1510 .instance_size = sizeof(BDRVGlusterState),
1511 .bdrv_needs_filename = true,
1512 .bdrv_file_open = qemu_gluster_open,
1513 .bdrv_reopen_prepare = qemu_gluster_reopen_prepare,
1514 .bdrv_reopen_commit = qemu_gluster_reopen_commit,
1515 .bdrv_reopen_abort = qemu_gluster_reopen_abort,
1516 .bdrv_close = qemu_gluster_close,
1517 .bdrv_create = qemu_gluster_create,
1518 .bdrv_getlength = qemu_gluster_getlength,
1519 .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size,
1520 .bdrv_truncate = qemu_gluster_truncate,
1521 .bdrv_co_readv = qemu_gluster_co_readv,
1522 .bdrv_co_writev = qemu_gluster_co_writev,
1523 .bdrv_co_flush_to_disk = qemu_gluster_co_flush_to_disk,
1524 .bdrv_has_zero_init = qemu_gluster_has_zero_init,
1525 #ifdef CONFIG_GLUSTERFS_DISCARD
1526 .bdrv_co_pdiscard = qemu_gluster_co_pdiscard,
1528 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1529 .bdrv_co_pwrite_zeroes = qemu_gluster_co_pwrite_zeroes,
1531 .bdrv_co_get_block_status = qemu_gluster_co_get_block_status,
1532 .create_opts = &qemu_gluster_create_opts,
1535 static void bdrv_gluster_init(void)
1537 bdrv_register(&bdrv_gluster_rdma);
1538 bdrv_register(&bdrv_gluster_unix);
1539 bdrv_register(&bdrv_gluster_tcp);
1540 bdrv_register(&bdrv_gluster);
1543 block_init(bdrv_gluster_init);