2 * GlusterFS backend for QEMU
6 * Pipe handling mechanism in AIO implementation is derived from
12 * This work is licensed under the terms of the GNU GPL, version 2. See
13 * the COPYING file in the top-level directory.
15 * Contributions after 2012-01-13 are licensed under the terms of the
16 * GNU GPL, version 2 or (at your option) any later version.
18 #include <glusterfs/api/glfs.h>
19 #include "block/block_int.h"
20 #include "qemu/sockets.h"
23 typedef struct GlusterAIOCB {
30 typedef struct BDRVGlusterState {
35 #define GLUSTER_FD_READ 0
36 #define GLUSTER_FD_WRITE 1
38 typedef struct GlusterConf {
46 static void qemu_gluster_gconf_free(GlusterConf *gconf)
48 g_free(gconf->server);
49 g_free(gconf->volname);
51 g_free(gconf->transport);
55 static int parse_volume_options(GlusterConf *gconf, char *path)
64 p = q = path + strspn(path, "/");
69 gconf->volname = g_strndup(q, p - q);
76 gconf->image = g_strdup(p);
81 * file=gluster[+transport]://[server[:port]]/volname/image[?socket=...]
83 * 'gluster' is the protocol.
85 * 'transport' specifies the transport type used to connect to gluster
86 * management daemon (glusterd). Valid transport types are
87 * tcp, unix and rdma. If a transport type isn't specified, then tcp
90 * 'server' specifies the server where the volume file specification for
91 * the given volume resides. This can be either hostname, ipv4 address
92 * or ipv6 address. ipv6 address needs to be within square brackets [ ].
93 * If transport type is 'unix', then 'server' field should not be specifed.
94 * The 'socket' field needs to be populated with the path to unix domain
97 * 'port' is the port number on which glusterd is listening. This is optional
98 * and if not specified, QEMU will send 0 which will make gluster to use the
99 * default port. If the transport type is unix, then 'port' should not be
102 * 'volname' is the name of the gluster volume which contains the VM image.
104 * 'image' is the path to the actual VM image that resides on gluster volume.
108 * file=gluster://1.2.3.4/testvol/a.img
109 * file=gluster+tcp://1.2.3.4/testvol/a.img
110 * file=gluster+tcp://1.2.3.4:24007/testvol/dir/a.img
111 * file=gluster+tcp://[1:2:3:4:5:6:7:8]/testvol/dir/a.img
112 * file=gluster+tcp://[1:2:3:4:5:6:7:8]:24007/testvol/dir/a.img
113 * file=gluster+tcp://server.domain.com:24007/testvol/dir/a.img
114 * file=gluster+unix:///testvol/dir/a.img?socket=/tmp/glusterd.socket
115 * file=gluster+rdma://1.2.3.4:24007/testvol/a.img
117 static int qemu_gluster_parseuri(GlusterConf *gconf, const char *filename)
120 QueryParams *qp = NULL;
121 bool is_unix = false;
124 uri = uri_parse(filename);
130 if (!strcmp(uri->scheme, "gluster")) {
131 gconf->transport = g_strdup("tcp");
132 } else if (!strcmp(uri->scheme, "gluster+tcp")) {
133 gconf->transport = g_strdup("tcp");
134 } else if (!strcmp(uri->scheme, "gluster+unix")) {
135 gconf->transport = g_strdup("unix");
137 } else if (!strcmp(uri->scheme, "gluster+rdma")) {
138 gconf->transport = g_strdup("rdma");
144 ret = parse_volume_options(gconf, uri->path);
149 qp = query_params_parse(uri->query);
150 if (qp->n > 1 || (is_unix && !qp->n) || (!is_unix && qp->n)) {
156 if (uri->server || uri->port) {
160 if (strcmp(qp->p[0].name, "socket")) {
164 gconf->server = g_strdup(qp->p[0].value);
166 gconf->server = g_strdup(uri->server);
167 gconf->port = uri->port;
172 query_params_free(qp);
178 static struct glfs *qemu_gluster_init(GlusterConf *gconf, const char *filename)
180 struct glfs *glfs = NULL;
184 ret = qemu_gluster_parseuri(gconf, filename);
186 error_report("Usage: file=gluster[+transport]://[server[:port]]/"
187 "volname/image[?socket=...]");
192 glfs = glfs_new(gconf->volname);
197 ret = glfs_set_volfile_server(glfs, gconf->transport, gconf->server,
204 * TODO: Use GF_LOG_ERROR instead of hard code value of 4 here when
205 * GlusterFS makes GF_LOG_* macros available to libgfapi users.
207 ret = glfs_set_logging(glfs, "-", 4);
212 ret = glfs_init(glfs);
214 error_report("Gluster connection failed for server=%s port=%d "
215 "volume=%s image=%s transport=%s", gconf->server, gconf->port,
216 gconf->volname, gconf->image, gconf->transport);
230 static void qemu_gluster_complete_aio(void *opaque)
232 GlusterAIOCB *acb = (GlusterAIOCB *)opaque;
234 qemu_bh_delete(acb->bh);
236 qemu_coroutine_enter(acb->coroutine, NULL);
240 * AIO callback routine called from GlusterFS thread.
242 static void gluster_finish_aiocb(struct glfs_fd *fd, ssize_t ret, void *arg)
244 GlusterAIOCB *acb = (GlusterAIOCB *)arg;
246 if (!ret || ret == acb->size) {
247 acb->ret = 0; /* Success */
248 } else if (ret < 0) {
249 acb->ret = ret; /* Read/Write failed */
251 acb->ret = -EIO; /* Partial read/write - fail it */
254 acb->bh = qemu_bh_new(qemu_gluster_complete_aio, acb);
255 qemu_bh_schedule(acb->bh);
258 /* TODO Convert to fine grained options */
259 static QemuOptsList runtime_opts = {
261 .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
265 .type = QEMU_OPT_STRING,
266 .help = "URL to the gluster image",
268 { /* end of list */ }
272 static int qemu_gluster_open(BlockDriverState *bs, QDict *options,
273 int bdrv_flags, Error **errp)
275 BDRVGlusterState *s = bs->opaque;
276 int open_flags = O_BINARY;
278 GlusterConf *gconf = g_malloc0(sizeof(GlusterConf));
280 Error *local_err = NULL;
281 const char *filename;
283 opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
284 qemu_opts_absorb_qdict(opts, options, &local_err);
286 qerror_report_err(local_err);
287 error_free(local_err);
292 filename = qemu_opt_get(opts, "filename");
294 s->glfs = qemu_gluster_init(gconf, filename);
300 if (bdrv_flags & BDRV_O_RDWR) {
301 open_flags |= O_RDWR;
303 open_flags |= O_RDONLY;
306 if ((bdrv_flags & BDRV_O_NOCACHE)) {
307 open_flags |= O_DIRECT;
310 s->fd = glfs_open(s->glfs, gconf->image, open_flags);
317 qemu_gluster_gconf_free(gconf);
330 #ifdef CONFIG_GLUSTERFS_ZEROFILL
331 static coroutine_fn int qemu_gluster_co_write_zeroes(BlockDriverState *bs,
332 int64_t sector_num, int nb_sectors, BdrvRequestFlags flags)
335 GlusterAIOCB *acb = g_slice_new(GlusterAIOCB);
336 BDRVGlusterState *s = bs->opaque;
337 off_t size = nb_sectors * BDRV_SECTOR_SIZE;
338 off_t offset = sector_num * BDRV_SECTOR_SIZE;
342 acb->coroutine = qemu_coroutine_self();
344 ret = glfs_zerofill_async(s->fd, offset, size, &gluster_finish_aiocb, acb);
350 qemu_coroutine_yield();
354 g_slice_free(GlusterAIOCB, acb);
358 static inline bool gluster_supports_zerofill(void)
363 static inline int qemu_gluster_zerofill(struct glfs_fd *fd, int64_t offset,
366 return glfs_zerofill(fd, offset, size);
370 static inline bool gluster_supports_zerofill(void)
375 static inline int qemu_gluster_zerofill(struct glfs_fd *fd, int64_t offset,
382 static int qemu_gluster_create(const char *filename,
383 QEMUOptionParameter *options, Error **errp)
389 int64_t total_size = 0;
390 GlusterConf *gconf = g_malloc0(sizeof(GlusterConf));
392 glfs = qemu_gluster_init(gconf, filename);
398 while (options && options->name) {
399 if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
400 total_size = options->value.n / BDRV_SECTOR_SIZE;
401 } else if (!strcmp(options->name, BLOCK_OPT_PREALLOC)) {
402 if (!options->value.s || !strcmp(options->value.s, "off")) {
404 } else if (!strcmp(options->value.s, "full") &&
405 gluster_supports_zerofill()) {
408 error_setg(errp, "Invalid preallocation mode: '%s'"
409 " or GlusterFS doesn't support zerofill API",
418 fd = glfs_creat(glfs, gconf->image,
419 O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, S_IRUSR | S_IWUSR);
423 if (!glfs_ftruncate(fd, total_size * BDRV_SECTOR_SIZE)) {
424 if (prealloc && qemu_gluster_zerofill(fd, 0,
425 total_size * BDRV_SECTOR_SIZE)) {
432 if (glfs_close(fd) != 0) {
437 qemu_gluster_gconf_free(gconf);
444 static coroutine_fn int qemu_gluster_co_rw(BlockDriverState *bs,
445 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov, int write)
448 GlusterAIOCB *acb = g_slice_new(GlusterAIOCB);
449 BDRVGlusterState *s = bs->opaque;
450 size_t size = nb_sectors * BDRV_SECTOR_SIZE;
451 off_t offset = sector_num * BDRV_SECTOR_SIZE;
455 acb->coroutine = qemu_coroutine_self();
458 ret = glfs_pwritev_async(s->fd, qiov->iov, qiov->niov, offset, 0,
459 &gluster_finish_aiocb, acb);
461 ret = glfs_preadv_async(s->fd, qiov->iov, qiov->niov, offset, 0,
462 &gluster_finish_aiocb, acb);
470 qemu_coroutine_yield();
474 g_slice_free(GlusterAIOCB, acb);
478 static int qemu_gluster_truncate(BlockDriverState *bs, int64_t offset)
481 BDRVGlusterState *s = bs->opaque;
483 ret = glfs_ftruncate(s->fd, offset);
491 static coroutine_fn int qemu_gluster_co_readv(BlockDriverState *bs,
492 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov)
494 return qemu_gluster_co_rw(bs, sector_num, nb_sectors, qiov, 0);
497 static coroutine_fn int qemu_gluster_co_writev(BlockDriverState *bs,
498 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov)
500 return qemu_gluster_co_rw(bs, sector_num, nb_sectors, qiov, 1);
503 static coroutine_fn int qemu_gluster_co_flush_to_disk(BlockDriverState *bs)
506 GlusterAIOCB *acb = g_slice_new(GlusterAIOCB);
507 BDRVGlusterState *s = bs->opaque;
511 acb->coroutine = qemu_coroutine_self();
513 ret = glfs_fsync_async(s->fd, &gluster_finish_aiocb, acb);
519 qemu_coroutine_yield();
523 g_slice_free(GlusterAIOCB, acb);
527 #ifdef CONFIG_GLUSTERFS_DISCARD
528 static coroutine_fn int qemu_gluster_co_discard(BlockDriverState *bs,
529 int64_t sector_num, int nb_sectors)
532 GlusterAIOCB *acb = g_slice_new(GlusterAIOCB);
533 BDRVGlusterState *s = bs->opaque;
534 size_t size = nb_sectors * BDRV_SECTOR_SIZE;
535 off_t offset = sector_num * BDRV_SECTOR_SIZE;
539 acb->coroutine = qemu_coroutine_self();
541 ret = glfs_discard_async(s->fd, offset, size, &gluster_finish_aiocb, acb);
547 qemu_coroutine_yield();
551 g_slice_free(GlusterAIOCB, acb);
556 static int64_t qemu_gluster_getlength(BlockDriverState *bs)
558 BDRVGlusterState *s = bs->opaque;
561 ret = glfs_lseek(s->fd, 0, SEEK_END);
569 static int64_t qemu_gluster_allocated_file_size(BlockDriverState *bs)
571 BDRVGlusterState *s = bs->opaque;
575 ret = glfs_fstat(s->fd, &st);
579 return st.st_blocks * 512;
583 static void qemu_gluster_close(BlockDriverState *bs)
585 BDRVGlusterState *s = bs->opaque;
594 static int qemu_gluster_has_zero_init(BlockDriverState *bs)
596 /* GlusterFS volume could be backed by a block device */
600 static QEMUOptionParameter qemu_gluster_create_options[] = {
602 .name = BLOCK_OPT_SIZE,
604 .help = "Virtual disk size"
607 .name = BLOCK_OPT_PREALLOC,
609 .help = "Preallocation mode (allowed values: off, full)"
614 static BlockDriver bdrv_gluster = {
615 .format_name = "gluster",
616 .protocol_name = "gluster",
617 .instance_size = sizeof(BDRVGlusterState),
618 .bdrv_needs_filename = true,
619 .bdrv_file_open = qemu_gluster_open,
620 .bdrv_close = qemu_gluster_close,
621 .bdrv_create = qemu_gluster_create,
622 .bdrv_getlength = qemu_gluster_getlength,
623 .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size,
624 .bdrv_truncate = qemu_gluster_truncate,
625 .bdrv_co_readv = qemu_gluster_co_readv,
626 .bdrv_co_writev = qemu_gluster_co_writev,
627 .bdrv_co_flush_to_disk = qemu_gluster_co_flush_to_disk,
628 .bdrv_has_zero_init = qemu_gluster_has_zero_init,
629 #ifdef CONFIG_GLUSTERFS_DISCARD
630 .bdrv_co_discard = qemu_gluster_co_discard,
632 #ifdef CONFIG_GLUSTERFS_ZEROFILL
633 .bdrv_co_write_zeroes = qemu_gluster_co_write_zeroes,
635 .create_options = qemu_gluster_create_options,
638 static BlockDriver bdrv_gluster_tcp = {
639 .format_name = "gluster",
640 .protocol_name = "gluster+tcp",
641 .instance_size = sizeof(BDRVGlusterState),
642 .bdrv_needs_filename = true,
643 .bdrv_file_open = qemu_gluster_open,
644 .bdrv_close = qemu_gluster_close,
645 .bdrv_create = qemu_gluster_create,
646 .bdrv_getlength = qemu_gluster_getlength,
647 .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size,
648 .bdrv_truncate = qemu_gluster_truncate,
649 .bdrv_co_readv = qemu_gluster_co_readv,
650 .bdrv_co_writev = qemu_gluster_co_writev,
651 .bdrv_co_flush_to_disk = qemu_gluster_co_flush_to_disk,
652 .bdrv_has_zero_init = qemu_gluster_has_zero_init,
653 #ifdef CONFIG_GLUSTERFS_DISCARD
654 .bdrv_co_discard = qemu_gluster_co_discard,
656 #ifdef CONFIG_GLUSTERFS_ZEROFILL
657 .bdrv_co_write_zeroes = qemu_gluster_co_write_zeroes,
659 .create_options = qemu_gluster_create_options,
662 static BlockDriver bdrv_gluster_unix = {
663 .format_name = "gluster",
664 .protocol_name = "gluster+unix",
665 .instance_size = sizeof(BDRVGlusterState),
666 .bdrv_needs_filename = true,
667 .bdrv_file_open = qemu_gluster_open,
668 .bdrv_close = qemu_gluster_close,
669 .bdrv_create = qemu_gluster_create,
670 .bdrv_getlength = qemu_gluster_getlength,
671 .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size,
672 .bdrv_truncate = qemu_gluster_truncate,
673 .bdrv_co_readv = qemu_gluster_co_readv,
674 .bdrv_co_writev = qemu_gluster_co_writev,
675 .bdrv_co_flush_to_disk = qemu_gluster_co_flush_to_disk,
676 .bdrv_has_zero_init = qemu_gluster_has_zero_init,
677 #ifdef CONFIG_GLUSTERFS_DISCARD
678 .bdrv_co_discard = qemu_gluster_co_discard,
680 #ifdef CONFIG_GLUSTERFS_ZEROFILL
681 .bdrv_co_write_zeroes = qemu_gluster_co_write_zeroes,
683 .create_options = qemu_gluster_create_options,
686 static BlockDriver bdrv_gluster_rdma = {
687 .format_name = "gluster",
688 .protocol_name = "gluster+rdma",
689 .instance_size = sizeof(BDRVGlusterState),
690 .bdrv_needs_filename = true,
691 .bdrv_file_open = qemu_gluster_open,
692 .bdrv_close = qemu_gluster_close,
693 .bdrv_create = qemu_gluster_create,
694 .bdrv_getlength = qemu_gluster_getlength,
695 .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size,
696 .bdrv_truncate = qemu_gluster_truncate,
697 .bdrv_co_readv = qemu_gluster_co_readv,
698 .bdrv_co_writev = qemu_gluster_co_writev,
699 .bdrv_co_flush_to_disk = qemu_gluster_co_flush_to_disk,
700 .bdrv_has_zero_init = qemu_gluster_has_zero_init,
701 #ifdef CONFIG_GLUSTERFS_DISCARD
702 .bdrv_co_discard = qemu_gluster_co_discard,
704 #ifdef CONFIG_GLUSTERFS_ZEROFILL
705 .bdrv_co_write_zeroes = qemu_gluster_co_write_zeroes,
707 .create_options = qemu_gluster_create_options,
710 static void bdrv_gluster_init(void)
712 bdrv_register(&bdrv_gluster_rdma);
713 bdrv_register(&bdrv_gluster_unix);
714 bdrv_register(&bdrv_gluster_tcp);
715 bdrv_register(&bdrv_gluster);
718 block_init(bdrv_gluster_init);