Merge remote-tracking branch 'remotes/kraxel/tags/pull-vga-20170103-1' into staging
[qemu.git] / block / gluster.c
1 /*
2 * GlusterFS backend for QEMU
3 *
4 * Copyright (C) 2012 Bharata B Rao <bharata@linux.vnet.ibm.com>
5 *
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.
8 *
9 */
10 #include "qemu/osdep.h"
11 #include <glusterfs/api/glfs.h>
12 #include "block/block_int.h"
13 #include "qapi/error.h"
14 #include "qapi/qmp/qerror.h"
15 #include "qemu/uri.h"
16 #include "qemu/error-report.h"
17 #include "qemu/cutils.h"
18
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 */
36
37 #define GERR_INDEX_HINT "hint: check in 'server' array index '%d'\n"
38
39 typedef struct GlusterAIOCB {
40 int64_t size;
41 int ret;
42 Coroutine *coroutine;
43 AioContext *aio_context;
44 } GlusterAIOCB;
45
46 typedef struct BDRVGlusterState {
47 struct glfs *glfs;
48 struct glfs_fd *fd;
49 char *logfile;
50 bool supports_seek_data;
51 int debug;
52 } BDRVGlusterState;
53
54 typedef struct BDRVGlusterReopenState {
55 struct glfs *glfs;
56 struct glfs_fd *fd;
57 } BDRVGlusterReopenState;
58
59
60 typedef struct GlfsPreopened {
61 char *volume;
62 glfs_t *fs;
63 int ref;
64 } GlfsPreopened;
65
66 typedef struct ListElement {
67 QLIST_ENTRY(ListElement) list;
68 GlfsPreopened saved;
69 } ListElement;
70
71 static QLIST_HEAD(glfs_list, ListElement) glfs_list;
72
73 static QemuOptsList qemu_gluster_create_opts = {
74 .name = "qemu-gluster-create-opts",
75 .head = QTAILQ_HEAD_INITIALIZER(qemu_gluster_create_opts.head),
76 .desc = {
77 {
78 .name = BLOCK_OPT_SIZE,
79 .type = QEMU_OPT_SIZE,
80 .help = "Virtual disk size"
81 },
82 {
83 .name = BLOCK_OPT_PREALLOC,
84 .type = QEMU_OPT_STRING,
85 .help = "Preallocation mode (allowed values: off, full)"
86 },
87 {
88 .name = GLUSTER_OPT_DEBUG,
89 .type = QEMU_OPT_NUMBER,
90 .help = "Gluster log level, valid range is 0-9",
91 },
92 {
93 .name = GLUSTER_OPT_LOGFILE,
94 .type = QEMU_OPT_STRING,
95 .help = "Logfile path of libgfapi",
96 },
97 { /* end of list */ }
98 }
99 };
100
101 static QemuOptsList runtime_opts = {
102 .name = "gluster",
103 .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
104 .desc = {
105 {
106 .name = GLUSTER_OPT_FILENAME,
107 .type = QEMU_OPT_STRING,
108 .help = "URL to the gluster image",
109 },
110 {
111 .name = GLUSTER_OPT_DEBUG,
112 .type = QEMU_OPT_NUMBER,
113 .help = "Gluster log level, valid range is 0-9",
114 },
115 {
116 .name = GLUSTER_OPT_LOGFILE,
117 .type = QEMU_OPT_STRING,
118 .help = "Logfile path of libgfapi",
119 },
120 { /* end of list */ }
121 },
122 };
123
124 static QemuOptsList runtime_json_opts = {
125 .name = "gluster_json",
126 .head = QTAILQ_HEAD_INITIALIZER(runtime_json_opts.head),
127 .desc = {
128 {
129 .name = GLUSTER_OPT_VOLUME,
130 .type = QEMU_OPT_STRING,
131 .help = "name of gluster volume where VM image resides",
132 },
133 {
134 .name = GLUSTER_OPT_PATH,
135 .type = QEMU_OPT_STRING,
136 .help = "absolute path to image file in gluster volume",
137 },
138 {
139 .name = GLUSTER_OPT_DEBUG,
140 .type = QEMU_OPT_NUMBER,
141 .help = "Gluster log level, valid range is 0-9",
142 },
143 { /* end of list */ }
144 },
145 };
146
147 static QemuOptsList runtime_type_opts = {
148 .name = "gluster_type",
149 .head = QTAILQ_HEAD_INITIALIZER(runtime_type_opts.head),
150 .desc = {
151 {
152 .name = GLUSTER_OPT_TYPE,
153 .type = QEMU_OPT_STRING,
154 .help = "tcp|unix",
155 },
156 { /* end of list */ }
157 },
158 };
159
160 static QemuOptsList runtime_unix_opts = {
161 .name = "gluster_unix",
162 .head = QTAILQ_HEAD_INITIALIZER(runtime_unix_opts.head),
163 .desc = {
164 {
165 .name = GLUSTER_OPT_SOCKET,
166 .type = QEMU_OPT_STRING,
167 .help = "socket file path)",
168 },
169 { /* end of list */ }
170 },
171 };
172
173 static QemuOptsList runtime_tcp_opts = {
174 .name = "gluster_tcp",
175 .head = QTAILQ_HEAD_INITIALIZER(runtime_tcp_opts.head),
176 .desc = {
177 {
178 .name = GLUSTER_OPT_TYPE,
179 .type = QEMU_OPT_STRING,
180 .help = "tcp|unix",
181 },
182 {
183 .name = GLUSTER_OPT_HOST,
184 .type = QEMU_OPT_STRING,
185 .help = "host address (hostname/ipv4/ipv6 addresses)",
186 },
187 {
188 .name = GLUSTER_OPT_PORT,
189 .type = QEMU_OPT_STRING,
190 .help = "port number on which glusterd is listening (default 24007)",
191 },
192 {
193 .name = "to",
194 .type = QEMU_OPT_NUMBER,
195 .help = "max port number, not supported by gluster",
196 },
197 {
198 .name = "ipv4",
199 .type = QEMU_OPT_BOOL,
200 .help = "ipv4 bool value, not supported by gluster",
201 },
202 {
203 .name = "ipv6",
204 .type = QEMU_OPT_BOOL,
205 .help = "ipv6 bool value, not supported by gluster",
206 },
207 { /* end of list */ }
208 },
209 };
210
211 static void glfs_set_preopened(const char *volume, glfs_t *fs)
212 {
213 ListElement *entry = NULL;
214
215 entry = g_new(ListElement, 1);
216
217 entry->saved.volume = g_strdup(volume);
218
219 entry->saved.fs = fs;
220 entry->saved.ref = 1;
221
222 QLIST_INSERT_HEAD(&glfs_list, entry, list);
223 }
224
225 static glfs_t *glfs_find_preopened(const char *volume)
226 {
227 ListElement *entry = NULL;
228
229 QLIST_FOREACH(entry, &glfs_list, list) {
230 if (strcmp(entry->saved.volume, volume) == 0) {
231 entry->saved.ref++;
232 return entry->saved.fs;
233 }
234 }
235
236 return NULL;
237 }
238
239 static void glfs_clear_preopened(glfs_t *fs)
240 {
241 ListElement *entry = NULL;
242 ListElement *next;
243
244 if (fs == NULL) {
245 return;
246 }
247
248 QLIST_FOREACH_SAFE(entry, &glfs_list, list, next) {
249 if (entry->saved.fs == fs) {
250 if (--entry->saved.ref) {
251 return;
252 }
253
254 QLIST_REMOVE(entry, list);
255
256 glfs_fini(entry->saved.fs);
257 g_free(entry->saved.volume);
258 g_free(entry);
259 }
260 }
261 }
262
263 static int parse_volume_options(BlockdevOptionsGluster *gconf, char *path)
264 {
265 char *p, *q;
266
267 if (!path) {
268 return -EINVAL;
269 }
270
271 /* volume */
272 p = q = path + strspn(path, "/");
273 p += strcspn(p, "/");
274 if (*p == '\0') {
275 return -EINVAL;
276 }
277 gconf->volume = g_strndup(q, p - q);
278
279 /* path */
280 p += strspn(p, "/");
281 if (*p == '\0') {
282 return -EINVAL;
283 }
284 gconf->path = g_strdup(p);
285 return 0;
286 }
287
288 /*
289 * file=gluster[+transport]://[host[:port]]/volume/path[?socket=...]
290 *
291 * 'gluster' is the protocol.
292 *
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.
296 *
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
301 * socket.
302 *
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
306 * specified.
307 *
308 * 'volume' is the name of the gluster volume which contains the VM image.
309 *
310 * 'path' is the path to the actual VM image that resides on gluster volume.
311 *
312 * Examples:
313 *
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
319 */
320 static int qemu_gluster_parse_uri(BlockdevOptionsGluster *gconf,
321 const char *filename)
322 {
323 GlusterServer *gsconf;
324 URI *uri;
325 QueryParams *qp = NULL;
326 bool is_unix = false;
327 int ret = 0;
328
329 uri = uri_parse(filename);
330 if (!uri) {
331 return -EINVAL;
332 }
333
334 gconf->server = g_new0(GlusterServerList, 1);
335 gconf->server->value = gsconf = g_new0(GlusterServer, 1);
336
337 /* transport */
338 if (!uri->scheme || !strcmp(uri->scheme, "gluster")) {
339 gsconf->type = GLUSTER_TRANSPORT_TCP;
340 } else if (!strcmp(uri->scheme, "gluster+tcp")) {
341 gsconf->type = GLUSTER_TRANSPORT_TCP;
342 } else if (!strcmp(uri->scheme, "gluster+unix")) {
343 gsconf->type = GLUSTER_TRANSPORT_UNIX;
344 is_unix = true;
345 } else if (!strcmp(uri->scheme, "gluster+rdma")) {
346 gsconf->type = GLUSTER_TRANSPORT_TCP;
347 error_report("Warning: rdma feature is not supported, falling "
348 "back to tcp");
349 } else {
350 ret = -EINVAL;
351 goto out;
352 }
353
354 ret = parse_volume_options(gconf, uri->path);
355 if (ret < 0) {
356 goto out;
357 }
358
359 qp = query_params_parse(uri->query);
360 if (qp->n > 1 || (is_unix && !qp->n) || (!is_unix && qp->n)) {
361 ret = -EINVAL;
362 goto out;
363 }
364
365 if (is_unix) {
366 if (uri->server || uri->port) {
367 ret = -EINVAL;
368 goto out;
369 }
370 if (strcmp(qp->p[0].name, "socket")) {
371 ret = -EINVAL;
372 goto out;
373 }
374 gsconf->u.q_unix.path = g_strdup(qp->p[0].value);
375 } else {
376 gsconf->u.tcp.host = g_strdup(uri->server ? uri->server : "localhost");
377 if (uri->port) {
378 gsconf->u.tcp.port = g_strdup_printf("%d", uri->port);
379 } else {
380 gsconf->u.tcp.port = g_strdup_printf("%d", GLUSTER_DEFAULT_PORT);
381 }
382 }
383
384 out:
385 if (qp) {
386 query_params_free(qp);
387 }
388 uri_free(uri);
389 return ret;
390 }
391
392 static struct glfs *qemu_gluster_glfs_init(BlockdevOptionsGluster *gconf,
393 Error **errp)
394 {
395 struct glfs *glfs;
396 int ret;
397 int old_errno;
398 GlusterServerList *server;
399 unsigned long long port;
400
401 glfs = glfs_find_preopened(gconf->volume);
402 if (glfs) {
403 return glfs;
404 }
405
406 glfs = glfs_new(gconf->volume);
407 if (!glfs) {
408 goto out;
409 }
410
411 glfs_set_preopened(gconf->volume, glfs);
412
413 for (server = gconf->server; server; server = server->next) {
414 if (server->value->type == GLUSTER_TRANSPORT_UNIX) {
415 ret = glfs_set_volfile_server(glfs,
416 GlusterTransport_lookup[server->value->type],
417 server->value->u.q_unix.path, 0);
418 } else {
419 if (parse_uint_full(server->value->u.tcp.port, &port, 10) < 0 ||
420 port > 65535) {
421 error_setg(errp, "'%s' is not a valid port number",
422 server->value->u.tcp.port);
423 errno = EINVAL;
424 goto out;
425 }
426 ret = glfs_set_volfile_server(glfs,
427 GlusterTransport_lookup[server->value->type],
428 server->value->u.tcp.host,
429 (int)port);
430 }
431
432 if (ret < 0) {
433 goto out;
434 }
435 }
436
437 ret = glfs_set_logging(glfs, gconf->logfile, gconf->debug);
438 if (ret < 0) {
439 goto out;
440 }
441
442 ret = glfs_init(glfs);
443 if (ret) {
444 error_setg(errp, "Gluster connection for volume %s, path %s failed"
445 " to connect", gconf->volume, gconf->path);
446 for (server = gconf->server; server; server = server->next) {
447 if (server->value->type == GLUSTER_TRANSPORT_UNIX) {
448 error_append_hint(errp, "hint: failed on socket %s ",
449 server->value->u.q_unix.path);
450 } else {
451 error_append_hint(errp, "hint: failed on host %s and port %s ",
452 server->value->u.tcp.host,
453 server->value->u.tcp.port);
454 }
455 }
456
457 error_append_hint(errp, "Please refer to gluster logs for more info\n");
458
459 /* glfs_init sometimes doesn't set errno although docs suggest that */
460 if (errno == 0) {
461 errno = EINVAL;
462 }
463
464 goto out;
465 }
466 return glfs;
467
468 out:
469 if (glfs) {
470 old_errno = errno;
471 glfs_clear_preopened(glfs);
472 errno = old_errno;
473 }
474 return NULL;
475 }
476
477 static int qapi_enum_parse(const char *opt)
478 {
479 int i;
480
481 if (!opt) {
482 return GLUSTER_TRANSPORT__MAX;
483 }
484
485 for (i = 0; i < GLUSTER_TRANSPORT__MAX; i++) {
486 if (!strcmp(opt, GlusterTransport_lookup[i])) {
487 return i;
488 }
489 }
490
491 return i;
492 }
493
494 /*
495 * Convert the json formatted command line into qapi.
496 */
497 static int qemu_gluster_parse_json(BlockdevOptionsGluster *gconf,
498 QDict *options, Error **errp)
499 {
500 QemuOpts *opts;
501 GlusterServer *gsconf;
502 GlusterServerList *curr = NULL;
503 QDict *backing_options = NULL;
504 Error *local_err = NULL;
505 char *str = NULL;
506 const char *ptr;
507 size_t num_servers;
508 int i;
509
510 /* create opts info from runtime_json_opts list */
511 opts = qemu_opts_create(&runtime_json_opts, NULL, 0, &error_abort);
512 qemu_opts_absorb_qdict(opts, options, &local_err);
513 if (local_err) {
514 goto out;
515 }
516
517 num_servers = qdict_array_entries(options, GLUSTER_OPT_SERVER_PATTERN);
518 if (num_servers < 1) {
519 error_setg(&local_err, QERR_MISSING_PARAMETER, "server");
520 goto out;
521 }
522
523 ptr = qemu_opt_get(opts, GLUSTER_OPT_VOLUME);
524 if (!ptr) {
525 error_setg(&local_err, QERR_MISSING_PARAMETER, GLUSTER_OPT_VOLUME);
526 goto out;
527 }
528 gconf->volume = g_strdup(ptr);
529
530 ptr = qemu_opt_get(opts, GLUSTER_OPT_PATH);
531 if (!ptr) {
532 error_setg(&local_err, QERR_MISSING_PARAMETER, GLUSTER_OPT_PATH);
533 goto out;
534 }
535 gconf->path = g_strdup(ptr);
536 qemu_opts_del(opts);
537
538 for (i = 0; i < num_servers; i++) {
539 str = g_strdup_printf(GLUSTER_OPT_SERVER_PATTERN"%d.", i);
540 qdict_extract_subqdict(options, &backing_options, str);
541
542 /* create opts info from runtime_type_opts list */
543 opts = qemu_opts_create(&runtime_type_opts, NULL, 0, &error_abort);
544 qemu_opts_absorb_qdict(opts, backing_options, &local_err);
545 if (local_err) {
546 goto out;
547 }
548
549 ptr = qemu_opt_get(opts, GLUSTER_OPT_TYPE);
550 gsconf = g_new0(GlusterServer, 1);
551 gsconf->type = qapi_enum_parse(ptr);
552 if (!ptr) {
553 error_setg(&local_err, QERR_MISSING_PARAMETER, GLUSTER_OPT_TYPE);
554 error_append_hint(&local_err, GERR_INDEX_HINT, i);
555 goto out;
556
557 }
558 if (gsconf->type == GLUSTER_TRANSPORT__MAX) {
559 error_setg(&local_err, QERR_INVALID_PARAMETER_VALUE,
560 GLUSTER_OPT_TYPE, "tcp or unix");
561 error_append_hint(&local_err, GERR_INDEX_HINT, i);
562 goto out;
563 }
564 qemu_opts_del(opts);
565
566 if (gsconf->type == GLUSTER_TRANSPORT_TCP) {
567 /* create opts info from runtime_tcp_opts list */
568 opts = qemu_opts_create(&runtime_tcp_opts, NULL, 0, &error_abort);
569 qemu_opts_absorb_qdict(opts, backing_options, &local_err);
570 if (local_err) {
571 goto out;
572 }
573
574 ptr = qemu_opt_get(opts, GLUSTER_OPT_HOST);
575 if (!ptr) {
576 error_setg(&local_err, QERR_MISSING_PARAMETER,
577 GLUSTER_OPT_HOST);
578 error_append_hint(&local_err, GERR_INDEX_HINT, i);
579 goto out;
580 }
581 gsconf->u.tcp.host = g_strdup(ptr);
582 ptr = qemu_opt_get(opts, GLUSTER_OPT_PORT);
583 if (!ptr) {
584 error_setg(&local_err, QERR_MISSING_PARAMETER,
585 GLUSTER_OPT_PORT);
586 error_append_hint(&local_err, GERR_INDEX_HINT, i);
587 goto out;
588 }
589 gsconf->u.tcp.port = g_strdup(ptr);
590
591 /* defend for unsupported fields in InetSocketAddress,
592 * i.e. @ipv4, @ipv6 and @to
593 */
594 ptr = qemu_opt_get(opts, GLUSTER_OPT_TO);
595 if (ptr) {
596 gsconf->u.tcp.has_to = true;
597 }
598 ptr = qemu_opt_get(opts, GLUSTER_OPT_IPV4);
599 if (ptr) {
600 gsconf->u.tcp.has_ipv4 = true;
601 }
602 ptr = qemu_opt_get(opts, GLUSTER_OPT_IPV6);
603 if (ptr) {
604 gsconf->u.tcp.has_ipv6 = true;
605 }
606 if (gsconf->u.tcp.has_to) {
607 error_setg(&local_err, "Parameter 'to' not supported");
608 goto out;
609 }
610 if (gsconf->u.tcp.has_ipv4 || gsconf->u.tcp.has_ipv6) {
611 error_setg(&local_err, "Parameters 'ipv4/ipv6' not supported");
612 goto out;
613 }
614 qemu_opts_del(opts);
615 } else {
616 /* create opts info from runtime_unix_opts list */
617 opts = qemu_opts_create(&runtime_unix_opts, NULL, 0, &error_abort);
618 qemu_opts_absorb_qdict(opts, backing_options, &local_err);
619 if (local_err) {
620 goto out;
621 }
622
623 ptr = qemu_opt_get(opts, GLUSTER_OPT_SOCKET);
624 if (!ptr) {
625 error_setg(&local_err, QERR_MISSING_PARAMETER,
626 GLUSTER_OPT_SOCKET);
627 error_append_hint(&local_err, GERR_INDEX_HINT, i);
628 goto out;
629 }
630 gsconf->u.q_unix.path = g_strdup(ptr);
631 qemu_opts_del(opts);
632 }
633
634 if (gconf->server == NULL) {
635 gconf->server = g_new0(GlusterServerList, 1);
636 gconf->server->value = gsconf;
637 curr = gconf->server;
638 } else {
639 curr->next = g_new0(GlusterServerList, 1);
640 curr->next->value = gsconf;
641 curr = curr->next;
642 }
643
644 qdict_del(backing_options, str);
645 g_free(str);
646 str = NULL;
647 }
648
649 return 0;
650
651 out:
652 error_propagate(errp, local_err);
653 qemu_opts_del(opts);
654 if (str) {
655 qdict_del(backing_options, str);
656 g_free(str);
657 }
658 errno = EINVAL;
659 return -errno;
660 }
661
662 static struct glfs *qemu_gluster_init(BlockdevOptionsGluster *gconf,
663 const char *filename,
664 QDict *options, Error **errp)
665 {
666 int ret;
667 if (filename) {
668 ret = qemu_gluster_parse_uri(gconf, filename);
669 if (ret < 0) {
670 error_setg(errp, "invalid URI");
671 error_append_hint(errp, "Usage: file=gluster[+transport]://"
672 "[host[:port]]volume/path[?socket=...]"
673 "[,file.debug=N]"
674 "[,file.logfile=/path/filename.log]\n");
675 errno = -ret;
676 return NULL;
677 }
678 } else {
679 ret = qemu_gluster_parse_json(gconf, options, errp);
680 if (ret < 0) {
681 error_append_hint(errp, "Usage: "
682 "-drive driver=qcow2,file.driver=gluster,"
683 "file.volume=testvol,file.path=/path/a.qcow2"
684 "[,file.debug=9]"
685 "[,file.logfile=/path/filename.log],"
686 "file.server.0.type=tcp,"
687 "file.server.0.host=1.2.3.4,"
688 "file.server.0.port=24007,"
689 "file.server.1.transport=unix,"
690 "file.server.1.socket=/var/run/glusterd.socket ..."
691 "\n");
692 errno = -ret;
693 return NULL;
694 }
695
696 }
697
698 return qemu_gluster_glfs_init(gconf, errp);
699 }
700
701 static void qemu_gluster_complete_aio(void *opaque)
702 {
703 GlusterAIOCB *acb = (GlusterAIOCB *)opaque;
704
705 qemu_coroutine_enter(acb->coroutine);
706 }
707
708 /*
709 * AIO callback routine called from GlusterFS thread.
710 */
711 static void gluster_finish_aiocb(struct glfs_fd *fd, ssize_t ret, void *arg)
712 {
713 GlusterAIOCB *acb = (GlusterAIOCB *)arg;
714
715 if (!ret || ret == acb->size) {
716 acb->ret = 0; /* Success */
717 } else if (ret < 0) {
718 acb->ret = -errno; /* Read/Write failed */
719 } else {
720 acb->ret = -EIO; /* Partial read/write - fail it */
721 }
722
723 aio_bh_schedule_oneshot(acb->aio_context, qemu_gluster_complete_aio, acb);
724 }
725
726 static void qemu_gluster_parse_flags(int bdrv_flags, int *open_flags)
727 {
728 assert(open_flags != NULL);
729
730 *open_flags |= O_BINARY;
731
732 if (bdrv_flags & BDRV_O_RDWR) {
733 *open_flags |= O_RDWR;
734 } else {
735 *open_flags |= O_RDONLY;
736 }
737
738 if ((bdrv_flags & BDRV_O_NOCACHE)) {
739 *open_flags |= O_DIRECT;
740 }
741 }
742
743 /*
744 * Do SEEK_DATA/HOLE to detect if it is functional. Older broken versions of
745 * gfapi incorrectly return the current offset when SEEK_DATA/HOLE is used.
746 * - Corrected versions return -1 and set errno to EINVAL.
747 * - Versions that support SEEK_DATA/HOLE correctly, will return -1 and set
748 * errno to ENXIO when SEEK_DATA is called with a position of EOF.
749 */
750 static bool qemu_gluster_test_seek(struct glfs_fd *fd)
751 {
752 off_t ret = 0;
753
754 #if defined SEEK_HOLE && defined SEEK_DATA
755 off_t eof;
756
757 eof = glfs_lseek(fd, 0, SEEK_END);
758 if (eof < 0) {
759 /* this should never occur */
760 return false;
761 }
762
763 /* this should always fail with ENXIO if SEEK_DATA is supported */
764 ret = glfs_lseek(fd, eof, SEEK_DATA);
765 #endif
766
767 return (ret < 0) && (errno == ENXIO);
768 }
769
770 static int qemu_gluster_open(BlockDriverState *bs, QDict *options,
771 int bdrv_flags, Error **errp)
772 {
773 BDRVGlusterState *s = bs->opaque;
774 int open_flags = 0;
775 int ret = 0;
776 BlockdevOptionsGluster *gconf = NULL;
777 QemuOpts *opts;
778 Error *local_err = NULL;
779 const char *filename, *logfile;
780
781 opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
782 qemu_opts_absorb_qdict(opts, options, &local_err);
783 if (local_err) {
784 error_propagate(errp, local_err);
785 ret = -EINVAL;
786 goto out;
787 }
788
789 filename = qemu_opt_get(opts, GLUSTER_OPT_FILENAME);
790
791 s->debug = qemu_opt_get_number(opts, GLUSTER_OPT_DEBUG,
792 GLUSTER_DEBUG_DEFAULT);
793 if (s->debug < 0) {
794 s->debug = 0;
795 } else if (s->debug > GLUSTER_DEBUG_MAX) {
796 s->debug = GLUSTER_DEBUG_MAX;
797 }
798
799 gconf = g_new0(BlockdevOptionsGluster, 1);
800 gconf->debug = s->debug;
801 gconf->has_debug = true;
802
803 logfile = qemu_opt_get(opts, GLUSTER_OPT_LOGFILE);
804 s->logfile = g_strdup(logfile ? logfile : GLUSTER_LOGFILE_DEFAULT);
805
806 gconf->logfile = g_strdup(s->logfile);
807 gconf->has_logfile = true;
808
809 s->glfs = qemu_gluster_init(gconf, filename, options, errp);
810 if (!s->glfs) {
811 ret = -errno;
812 goto out;
813 }
814
815 #ifdef CONFIG_GLUSTERFS_XLATOR_OPT
816 /* Without this, if fsync fails for a recoverable reason (for instance,
817 * ENOSPC), gluster will dump its cache, preventing retries. This means
818 * almost certain data loss. Not all gluster versions support the
819 * 'resync-failed-syncs-after-fsync' key value, but there is no way to
820 * discover during runtime if it is supported (this api returns success for
821 * unknown key/value pairs) */
822 ret = glfs_set_xlator_option(s->glfs, "*-write-behind",
823 "resync-failed-syncs-after-fsync",
824 "on");
825 if (ret < 0) {
826 error_setg_errno(errp, errno, "Unable to set xlator key/value pair");
827 ret = -errno;
828 goto out;
829 }
830 #endif
831
832 qemu_gluster_parse_flags(bdrv_flags, &open_flags);
833
834 s->fd = glfs_open(s->glfs, gconf->path, open_flags);
835 if (!s->fd) {
836 ret = -errno;
837 }
838
839 s->supports_seek_data = qemu_gluster_test_seek(s->fd);
840
841 out:
842 qemu_opts_del(opts);
843 qapi_free_BlockdevOptionsGluster(gconf);
844 if (!ret) {
845 return ret;
846 }
847 g_free(s->logfile);
848 if (s->fd) {
849 glfs_close(s->fd);
850 }
851
852 glfs_clear_preopened(s->glfs);
853
854 return ret;
855 }
856
857 static int qemu_gluster_reopen_prepare(BDRVReopenState *state,
858 BlockReopenQueue *queue, Error **errp)
859 {
860 int ret = 0;
861 BDRVGlusterState *s;
862 BDRVGlusterReopenState *reop_s;
863 BlockdevOptionsGluster *gconf;
864 int open_flags = 0;
865
866 assert(state != NULL);
867 assert(state->bs != NULL);
868
869 s = state->bs->opaque;
870
871 state->opaque = g_new0(BDRVGlusterReopenState, 1);
872 reop_s = state->opaque;
873
874 qemu_gluster_parse_flags(state->flags, &open_flags);
875
876 gconf = g_new0(BlockdevOptionsGluster, 1);
877 gconf->debug = s->debug;
878 gconf->has_debug = true;
879 gconf->logfile = g_strdup(s->logfile);
880 gconf->has_logfile = true;
881 reop_s->glfs = qemu_gluster_init(gconf, state->bs->filename, NULL, errp);
882 if (reop_s->glfs == NULL) {
883 ret = -errno;
884 goto exit;
885 }
886
887 #ifdef CONFIG_GLUSTERFS_XLATOR_OPT
888 ret = glfs_set_xlator_option(reop_s->glfs, "*-write-behind",
889 "resync-failed-syncs-after-fsync", "on");
890 if (ret < 0) {
891 error_setg_errno(errp, errno, "Unable to set xlator key/value pair");
892 ret = -errno;
893 goto exit;
894 }
895 #endif
896
897 reop_s->fd = glfs_open(reop_s->glfs, gconf->path, open_flags);
898 if (reop_s->fd == NULL) {
899 /* reops->glfs will be cleaned up in _abort */
900 ret = -errno;
901 goto exit;
902 }
903
904 exit:
905 /* state->opaque will be freed in either the _abort or _commit */
906 qapi_free_BlockdevOptionsGluster(gconf);
907 return ret;
908 }
909
910 static void qemu_gluster_reopen_commit(BDRVReopenState *state)
911 {
912 BDRVGlusterReopenState *reop_s = state->opaque;
913 BDRVGlusterState *s = state->bs->opaque;
914
915
916 /* close the old */
917 if (s->fd) {
918 glfs_close(s->fd);
919 }
920
921 glfs_clear_preopened(s->glfs);
922
923 /* use the newly opened image / connection */
924 s->fd = reop_s->fd;
925 s->glfs = reop_s->glfs;
926
927 g_free(state->opaque);
928 state->opaque = NULL;
929
930 return;
931 }
932
933
934 static void qemu_gluster_reopen_abort(BDRVReopenState *state)
935 {
936 BDRVGlusterReopenState *reop_s = state->opaque;
937
938 if (reop_s == NULL) {
939 return;
940 }
941
942 if (reop_s->fd) {
943 glfs_close(reop_s->fd);
944 }
945
946 glfs_clear_preopened(reop_s->glfs);
947
948 g_free(state->opaque);
949 state->opaque = NULL;
950
951 return;
952 }
953
954 #ifdef CONFIG_GLUSTERFS_ZEROFILL
955 static coroutine_fn int qemu_gluster_co_pwrite_zeroes(BlockDriverState *bs,
956 int64_t offset,
957 int size,
958 BdrvRequestFlags flags)
959 {
960 int ret;
961 GlusterAIOCB acb;
962 BDRVGlusterState *s = bs->opaque;
963
964 acb.size = size;
965 acb.ret = 0;
966 acb.coroutine = qemu_coroutine_self();
967 acb.aio_context = bdrv_get_aio_context(bs);
968
969 ret = glfs_zerofill_async(s->fd, offset, size, gluster_finish_aiocb, &acb);
970 if (ret < 0) {
971 return -errno;
972 }
973
974 qemu_coroutine_yield();
975 return acb.ret;
976 }
977
978 static inline bool gluster_supports_zerofill(void)
979 {
980 return 1;
981 }
982
983 static inline int qemu_gluster_zerofill(struct glfs_fd *fd, int64_t offset,
984 int64_t size)
985 {
986 return glfs_zerofill(fd, offset, size);
987 }
988
989 #else
990 static inline bool gluster_supports_zerofill(void)
991 {
992 return 0;
993 }
994
995 static inline int qemu_gluster_zerofill(struct glfs_fd *fd, int64_t offset,
996 int64_t size)
997 {
998 return 0;
999 }
1000 #endif
1001
1002 static int qemu_gluster_create(const char *filename,
1003 QemuOpts *opts, Error **errp)
1004 {
1005 BlockdevOptionsGluster *gconf;
1006 struct glfs *glfs;
1007 struct glfs_fd *fd;
1008 int ret = 0;
1009 int prealloc = 0;
1010 int64_t total_size = 0;
1011 char *tmp = NULL;
1012
1013 gconf = g_new0(BlockdevOptionsGluster, 1);
1014 gconf->debug = qemu_opt_get_number_del(opts, GLUSTER_OPT_DEBUG,
1015 GLUSTER_DEBUG_DEFAULT);
1016 if (gconf->debug < 0) {
1017 gconf->debug = 0;
1018 } else if (gconf->debug > GLUSTER_DEBUG_MAX) {
1019 gconf->debug = GLUSTER_DEBUG_MAX;
1020 }
1021 gconf->has_debug = true;
1022
1023 gconf->logfile = qemu_opt_get_del(opts, GLUSTER_OPT_LOGFILE);
1024 if (!gconf->logfile) {
1025 gconf->logfile = g_strdup(GLUSTER_LOGFILE_DEFAULT);
1026 }
1027 gconf->has_logfile = true;
1028
1029 glfs = qemu_gluster_init(gconf, filename, NULL, errp);
1030 if (!glfs) {
1031 ret = -errno;
1032 goto out;
1033 }
1034
1035 total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
1036 BDRV_SECTOR_SIZE);
1037
1038 tmp = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
1039 if (!tmp || !strcmp(tmp, "off")) {
1040 prealloc = 0;
1041 } else if (!strcmp(tmp, "full") && gluster_supports_zerofill()) {
1042 prealloc = 1;
1043 } else {
1044 error_setg(errp, "Invalid preallocation mode: '%s'"
1045 " or GlusterFS doesn't support zerofill API", tmp);
1046 ret = -EINVAL;
1047 goto out;
1048 }
1049
1050 fd = glfs_creat(glfs, gconf->path,
1051 O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, S_IRUSR | S_IWUSR);
1052 if (!fd) {
1053 ret = -errno;
1054 } else {
1055 if (!glfs_ftruncate(fd, total_size)) {
1056 if (prealloc && qemu_gluster_zerofill(fd, 0, total_size)) {
1057 ret = -errno;
1058 }
1059 } else {
1060 ret = -errno;
1061 }
1062
1063 if (glfs_close(fd) != 0) {
1064 ret = -errno;
1065 }
1066 }
1067 out:
1068 g_free(tmp);
1069 qapi_free_BlockdevOptionsGluster(gconf);
1070 glfs_clear_preopened(glfs);
1071 return ret;
1072 }
1073
1074 static coroutine_fn int qemu_gluster_co_rw(BlockDriverState *bs,
1075 int64_t sector_num, int nb_sectors,
1076 QEMUIOVector *qiov, int write)
1077 {
1078 int ret;
1079 GlusterAIOCB acb;
1080 BDRVGlusterState *s = bs->opaque;
1081 size_t size = nb_sectors * BDRV_SECTOR_SIZE;
1082 off_t offset = sector_num * BDRV_SECTOR_SIZE;
1083
1084 acb.size = size;
1085 acb.ret = 0;
1086 acb.coroutine = qemu_coroutine_self();
1087 acb.aio_context = bdrv_get_aio_context(bs);
1088
1089 if (write) {
1090 ret = glfs_pwritev_async(s->fd, qiov->iov, qiov->niov, offset, 0,
1091 gluster_finish_aiocb, &acb);
1092 } else {
1093 ret = glfs_preadv_async(s->fd, qiov->iov, qiov->niov, offset, 0,
1094 gluster_finish_aiocb, &acb);
1095 }
1096
1097 if (ret < 0) {
1098 return -errno;
1099 }
1100
1101 qemu_coroutine_yield();
1102 return acb.ret;
1103 }
1104
1105 static int qemu_gluster_truncate(BlockDriverState *bs, int64_t offset)
1106 {
1107 int ret;
1108 BDRVGlusterState *s = bs->opaque;
1109
1110 ret = glfs_ftruncate(s->fd, offset);
1111 if (ret < 0) {
1112 return -errno;
1113 }
1114
1115 return 0;
1116 }
1117
1118 static coroutine_fn int qemu_gluster_co_readv(BlockDriverState *bs,
1119 int64_t sector_num,
1120 int nb_sectors,
1121 QEMUIOVector *qiov)
1122 {
1123 return qemu_gluster_co_rw(bs, sector_num, nb_sectors, qiov, 0);
1124 }
1125
1126 static coroutine_fn int qemu_gluster_co_writev(BlockDriverState *bs,
1127 int64_t sector_num,
1128 int nb_sectors,
1129 QEMUIOVector *qiov)
1130 {
1131 return qemu_gluster_co_rw(bs, sector_num, nb_sectors, qiov, 1);
1132 }
1133
1134 static void qemu_gluster_close(BlockDriverState *bs)
1135 {
1136 BDRVGlusterState *s = bs->opaque;
1137
1138 g_free(s->logfile);
1139 if (s->fd) {
1140 glfs_close(s->fd);
1141 s->fd = NULL;
1142 }
1143 glfs_clear_preopened(s->glfs);
1144 }
1145
1146 static coroutine_fn int qemu_gluster_co_flush_to_disk(BlockDriverState *bs)
1147 {
1148 int ret;
1149 GlusterAIOCB acb;
1150 BDRVGlusterState *s = bs->opaque;
1151
1152 acb.size = 0;
1153 acb.ret = 0;
1154 acb.coroutine = qemu_coroutine_self();
1155 acb.aio_context = bdrv_get_aio_context(bs);
1156
1157 ret = glfs_fsync_async(s->fd, gluster_finish_aiocb, &acb);
1158 if (ret < 0) {
1159 ret = -errno;
1160 goto error;
1161 }
1162
1163 qemu_coroutine_yield();
1164 if (acb.ret < 0) {
1165 ret = acb.ret;
1166 goto error;
1167 }
1168
1169 return acb.ret;
1170
1171 error:
1172 /* Some versions of Gluster (3.5.6 -> 3.5.8?) will not retain its cache
1173 * after a fsync failure, so we have no way of allowing the guest to safely
1174 * continue. Gluster versions prior to 3.5.6 don't retain the cache
1175 * either, but will invalidate the fd on error, so this is again our only
1176 * option.
1177 *
1178 * The 'resync-failed-syncs-after-fsync' xlator option for the
1179 * write-behind cache will cause later gluster versions to retain its
1180 * cache after error, so long as the fd remains open. However, we
1181 * currently have no way of knowing if this option is supported.
1182 *
1183 * TODO: Once gluster provides a way for us to determine if the option
1184 * is supported, bypass the closure and setting drv to NULL. */
1185 qemu_gluster_close(bs);
1186 bs->drv = NULL;
1187 return ret;
1188 }
1189
1190 #ifdef CONFIG_GLUSTERFS_DISCARD
1191 static coroutine_fn int qemu_gluster_co_pdiscard(BlockDriverState *bs,
1192 int64_t offset, int size)
1193 {
1194 int ret;
1195 GlusterAIOCB acb;
1196 BDRVGlusterState *s = bs->opaque;
1197
1198 acb.size = 0;
1199 acb.ret = 0;
1200 acb.coroutine = qemu_coroutine_self();
1201 acb.aio_context = bdrv_get_aio_context(bs);
1202
1203 ret = glfs_discard_async(s->fd, offset, size, gluster_finish_aiocb, &acb);
1204 if (ret < 0) {
1205 return -errno;
1206 }
1207
1208 qemu_coroutine_yield();
1209 return acb.ret;
1210 }
1211 #endif
1212
1213 static int64_t qemu_gluster_getlength(BlockDriverState *bs)
1214 {
1215 BDRVGlusterState *s = bs->opaque;
1216 int64_t ret;
1217
1218 ret = glfs_lseek(s->fd, 0, SEEK_END);
1219 if (ret < 0) {
1220 return -errno;
1221 } else {
1222 return ret;
1223 }
1224 }
1225
1226 static int64_t qemu_gluster_allocated_file_size(BlockDriverState *bs)
1227 {
1228 BDRVGlusterState *s = bs->opaque;
1229 struct stat st;
1230 int ret;
1231
1232 ret = glfs_fstat(s->fd, &st);
1233 if (ret < 0) {
1234 return -errno;
1235 } else {
1236 return st.st_blocks * 512;
1237 }
1238 }
1239
1240 static int qemu_gluster_has_zero_init(BlockDriverState *bs)
1241 {
1242 /* GlusterFS volume could be backed by a block device */
1243 return 0;
1244 }
1245
1246 /*
1247 * Find allocation range in @bs around offset @start.
1248 * May change underlying file descriptor's file offset.
1249 * If @start is not in a hole, store @start in @data, and the
1250 * beginning of the next hole in @hole, and return 0.
1251 * If @start is in a non-trailing hole, store @start in @hole and the
1252 * beginning of the next non-hole in @data, and return 0.
1253 * If @start is in a trailing hole or beyond EOF, return -ENXIO.
1254 * If we can't find out, return a negative errno other than -ENXIO.
1255 *
1256 * (Shamefully copied from raw-posix.c, only miniscule adaptions.)
1257 */
1258 static int find_allocation(BlockDriverState *bs, off_t start,
1259 off_t *data, off_t *hole)
1260 {
1261 BDRVGlusterState *s = bs->opaque;
1262
1263 if (!s->supports_seek_data) {
1264 goto exit;
1265 }
1266
1267 #if defined SEEK_HOLE && defined SEEK_DATA
1268 off_t offs;
1269
1270 /*
1271 * SEEK_DATA cases:
1272 * D1. offs == start: start is in data
1273 * D2. offs > start: start is in a hole, next data at offs
1274 * D3. offs < 0, errno = ENXIO: either start is in a trailing hole
1275 * or start is beyond EOF
1276 * If the latter happens, the file has been truncated behind
1277 * our back since we opened it. All bets are off then.
1278 * Treating like a trailing hole is simplest.
1279 * D4. offs < 0, errno != ENXIO: we learned nothing
1280 */
1281 offs = glfs_lseek(s->fd, start, SEEK_DATA);
1282 if (offs < 0) {
1283 return -errno; /* D3 or D4 */
1284 }
1285 assert(offs >= start);
1286
1287 if (offs > start) {
1288 /* D2: in hole, next data at offs */
1289 *hole = start;
1290 *data = offs;
1291 return 0;
1292 }
1293
1294 /* D1: in data, end not yet known */
1295
1296 /*
1297 * SEEK_HOLE cases:
1298 * H1. offs == start: start is in a hole
1299 * If this happens here, a hole has been dug behind our back
1300 * since the previous lseek().
1301 * H2. offs > start: either start is in data, next hole at offs,
1302 * or start is in trailing hole, EOF at offs
1303 * Linux treats trailing holes like any other hole: offs ==
1304 * start. Solaris seeks to EOF instead: offs > start (blech).
1305 * If that happens here, a hole has been dug behind our back
1306 * since the previous lseek().
1307 * H3. offs < 0, errno = ENXIO: start is beyond EOF
1308 * If this happens, the file has been truncated behind our
1309 * back since we opened it. Treat it like a trailing hole.
1310 * H4. offs < 0, errno != ENXIO: we learned nothing
1311 * Pretend we know nothing at all, i.e. "forget" about D1.
1312 */
1313 offs = glfs_lseek(s->fd, start, SEEK_HOLE);
1314 if (offs < 0) {
1315 return -errno; /* D1 and (H3 or H4) */
1316 }
1317 assert(offs >= start);
1318
1319 if (offs > start) {
1320 /*
1321 * D1 and H2: either in data, next hole at offs, or it was in
1322 * data but is now in a trailing hole. In the latter case,
1323 * all bets are off. Treating it as if it there was data all
1324 * the way to EOF is safe, so simply do that.
1325 */
1326 *data = start;
1327 *hole = offs;
1328 return 0;
1329 }
1330
1331 /* D1 and H1 */
1332 return -EBUSY;
1333 #endif
1334
1335 exit:
1336 return -ENOTSUP;
1337 }
1338
1339 /*
1340 * Returns the allocation status of the specified sectors.
1341 *
1342 * If 'sector_num' is beyond the end of the disk image the return value is 0
1343 * and 'pnum' is set to 0.
1344 *
1345 * 'pnum' is set to the number of sectors (including and immediately following
1346 * the specified sector) that are known to be in the same
1347 * allocated/unallocated state.
1348 *
1349 * 'nb_sectors' is the max value 'pnum' should be set to. If nb_sectors goes
1350 * beyond the end of the disk image it will be clamped.
1351 *
1352 * (Based on raw_co_get_block_status() from raw-posix.c.)
1353 */
1354 static int64_t coroutine_fn qemu_gluster_co_get_block_status(
1355 BlockDriverState *bs, int64_t sector_num, int nb_sectors, int *pnum,
1356 BlockDriverState **file)
1357 {
1358 BDRVGlusterState *s = bs->opaque;
1359 off_t start, data = 0, hole = 0;
1360 int64_t total_size;
1361 int ret = -EINVAL;
1362
1363 if (!s->fd) {
1364 return ret;
1365 }
1366
1367 start = sector_num * BDRV_SECTOR_SIZE;
1368 total_size = bdrv_getlength(bs);
1369 if (total_size < 0) {
1370 return total_size;
1371 } else if (start >= total_size) {
1372 *pnum = 0;
1373 return 0;
1374 } else if (start + nb_sectors * BDRV_SECTOR_SIZE > total_size) {
1375 nb_sectors = DIV_ROUND_UP(total_size - start, BDRV_SECTOR_SIZE);
1376 }
1377
1378 ret = find_allocation(bs, start, &data, &hole);
1379 if (ret == -ENXIO) {
1380 /* Trailing hole */
1381 *pnum = nb_sectors;
1382 ret = BDRV_BLOCK_ZERO;
1383 } else if (ret < 0) {
1384 /* No info available, so pretend there are no holes */
1385 *pnum = nb_sectors;
1386 ret = BDRV_BLOCK_DATA;
1387 } else if (data == start) {
1388 /* On a data extent, compute sectors to the end of the extent,
1389 * possibly including a partial sector at EOF. */
1390 *pnum = MIN(nb_sectors, DIV_ROUND_UP(hole - start, BDRV_SECTOR_SIZE));
1391 ret = BDRV_BLOCK_DATA;
1392 } else {
1393 /* On a hole, compute sectors to the beginning of the next extent. */
1394 assert(hole == start);
1395 *pnum = MIN(nb_sectors, (data - start) / BDRV_SECTOR_SIZE);
1396 ret = BDRV_BLOCK_ZERO;
1397 }
1398
1399 *file = bs;
1400
1401 return ret | BDRV_BLOCK_OFFSET_VALID | start;
1402 }
1403
1404
1405 static BlockDriver bdrv_gluster = {
1406 .format_name = "gluster",
1407 .protocol_name = "gluster",
1408 .instance_size = sizeof(BDRVGlusterState),
1409 .bdrv_needs_filename = false,
1410 .bdrv_file_open = qemu_gluster_open,
1411 .bdrv_reopen_prepare = qemu_gluster_reopen_prepare,
1412 .bdrv_reopen_commit = qemu_gluster_reopen_commit,
1413 .bdrv_reopen_abort = qemu_gluster_reopen_abort,
1414 .bdrv_close = qemu_gluster_close,
1415 .bdrv_create = qemu_gluster_create,
1416 .bdrv_getlength = qemu_gluster_getlength,
1417 .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size,
1418 .bdrv_truncate = qemu_gluster_truncate,
1419 .bdrv_co_readv = qemu_gluster_co_readv,
1420 .bdrv_co_writev = qemu_gluster_co_writev,
1421 .bdrv_co_flush_to_disk = qemu_gluster_co_flush_to_disk,
1422 .bdrv_has_zero_init = qemu_gluster_has_zero_init,
1423 #ifdef CONFIG_GLUSTERFS_DISCARD
1424 .bdrv_co_pdiscard = qemu_gluster_co_pdiscard,
1425 #endif
1426 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1427 .bdrv_co_pwrite_zeroes = qemu_gluster_co_pwrite_zeroes,
1428 #endif
1429 .bdrv_co_get_block_status = qemu_gluster_co_get_block_status,
1430 .create_opts = &qemu_gluster_create_opts,
1431 };
1432
1433 static BlockDriver bdrv_gluster_tcp = {
1434 .format_name = "gluster",
1435 .protocol_name = "gluster+tcp",
1436 .instance_size = sizeof(BDRVGlusterState),
1437 .bdrv_needs_filename = false,
1438 .bdrv_file_open = qemu_gluster_open,
1439 .bdrv_reopen_prepare = qemu_gluster_reopen_prepare,
1440 .bdrv_reopen_commit = qemu_gluster_reopen_commit,
1441 .bdrv_reopen_abort = qemu_gluster_reopen_abort,
1442 .bdrv_close = qemu_gluster_close,
1443 .bdrv_create = qemu_gluster_create,
1444 .bdrv_getlength = qemu_gluster_getlength,
1445 .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size,
1446 .bdrv_truncate = qemu_gluster_truncate,
1447 .bdrv_co_readv = qemu_gluster_co_readv,
1448 .bdrv_co_writev = qemu_gluster_co_writev,
1449 .bdrv_co_flush_to_disk = qemu_gluster_co_flush_to_disk,
1450 .bdrv_has_zero_init = qemu_gluster_has_zero_init,
1451 #ifdef CONFIG_GLUSTERFS_DISCARD
1452 .bdrv_co_pdiscard = qemu_gluster_co_pdiscard,
1453 #endif
1454 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1455 .bdrv_co_pwrite_zeroes = qemu_gluster_co_pwrite_zeroes,
1456 #endif
1457 .bdrv_co_get_block_status = qemu_gluster_co_get_block_status,
1458 .create_opts = &qemu_gluster_create_opts,
1459 };
1460
1461 static BlockDriver bdrv_gluster_unix = {
1462 .format_name = "gluster",
1463 .protocol_name = "gluster+unix",
1464 .instance_size = sizeof(BDRVGlusterState),
1465 .bdrv_needs_filename = true,
1466 .bdrv_file_open = qemu_gluster_open,
1467 .bdrv_reopen_prepare = qemu_gluster_reopen_prepare,
1468 .bdrv_reopen_commit = qemu_gluster_reopen_commit,
1469 .bdrv_reopen_abort = qemu_gluster_reopen_abort,
1470 .bdrv_close = qemu_gluster_close,
1471 .bdrv_create = qemu_gluster_create,
1472 .bdrv_getlength = qemu_gluster_getlength,
1473 .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size,
1474 .bdrv_truncate = qemu_gluster_truncate,
1475 .bdrv_co_readv = qemu_gluster_co_readv,
1476 .bdrv_co_writev = qemu_gluster_co_writev,
1477 .bdrv_co_flush_to_disk = qemu_gluster_co_flush_to_disk,
1478 .bdrv_has_zero_init = qemu_gluster_has_zero_init,
1479 #ifdef CONFIG_GLUSTERFS_DISCARD
1480 .bdrv_co_pdiscard = qemu_gluster_co_pdiscard,
1481 #endif
1482 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1483 .bdrv_co_pwrite_zeroes = qemu_gluster_co_pwrite_zeroes,
1484 #endif
1485 .bdrv_co_get_block_status = qemu_gluster_co_get_block_status,
1486 .create_opts = &qemu_gluster_create_opts,
1487 };
1488
1489 /* rdma is deprecated (actually never supported for volfile fetch).
1490 * Let's maintain it for the protocol compatibility, to make sure things
1491 * won't break immediately. For now, gluster+rdma will fall back to gluster+tcp
1492 * protocol with a warning.
1493 * TODO: remove gluster+rdma interface support
1494 */
1495 static BlockDriver bdrv_gluster_rdma = {
1496 .format_name = "gluster",
1497 .protocol_name = "gluster+rdma",
1498 .instance_size = sizeof(BDRVGlusterState),
1499 .bdrv_needs_filename = true,
1500 .bdrv_file_open = qemu_gluster_open,
1501 .bdrv_reopen_prepare = qemu_gluster_reopen_prepare,
1502 .bdrv_reopen_commit = qemu_gluster_reopen_commit,
1503 .bdrv_reopen_abort = qemu_gluster_reopen_abort,
1504 .bdrv_close = qemu_gluster_close,
1505 .bdrv_create = qemu_gluster_create,
1506 .bdrv_getlength = qemu_gluster_getlength,
1507 .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size,
1508 .bdrv_truncate = qemu_gluster_truncate,
1509 .bdrv_co_readv = qemu_gluster_co_readv,
1510 .bdrv_co_writev = qemu_gluster_co_writev,
1511 .bdrv_co_flush_to_disk = qemu_gluster_co_flush_to_disk,
1512 .bdrv_has_zero_init = qemu_gluster_has_zero_init,
1513 #ifdef CONFIG_GLUSTERFS_DISCARD
1514 .bdrv_co_pdiscard = qemu_gluster_co_pdiscard,
1515 #endif
1516 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1517 .bdrv_co_pwrite_zeroes = qemu_gluster_co_pwrite_zeroes,
1518 #endif
1519 .bdrv_co_get_block_status = qemu_gluster_co_get_block_status,
1520 .create_opts = &qemu_gluster_create_opts,
1521 };
1522
1523 static void bdrv_gluster_init(void)
1524 {
1525 bdrv_register(&bdrv_gluster_rdma);
1526 bdrv_register(&bdrv_gluster_unix);
1527 bdrv_register(&bdrv_gluster_tcp);
1528 bdrv_register(&bdrv_gluster);
1529 }
1530
1531 block_init(bdrv_gluster_init);