xref: /openbmc/qemu/block/gluster.c (revision 795c40b8)
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 "qapi/util.h"
16 #include "qemu/uri.h"
17 #include "qemu/error-report.h"
18 #include "qemu/cutils.h"
19 
20 #define GLUSTER_OPT_FILENAME        "filename"
21 #define GLUSTER_OPT_VOLUME          "volume"
22 #define GLUSTER_OPT_PATH            "path"
23 #define GLUSTER_OPT_TYPE            "type"
24 #define GLUSTER_OPT_SERVER_PATTERN  "server."
25 #define GLUSTER_OPT_HOST            "host"
26 #define GLUSTER_OPT_PORT            "port"
27 #define GLUSTER_OPT_TO              "to"
28 #define GLUSTER_OPT_IPV4            "ipv4"
29 #define GLUSTER_OPT_IPV6            "ipv6"
30 #define GLUSTER_OPT_SOCKET          "socket"
31 #define GLUSTER_OPT_DEBUG           "debug"
32 #define GLUSTER_DEFAULT_PORT        24007
33 #define GLUSTER_DEBUG_DEFAULT       4
34 #define GLUSTER_DEBUG_MAX           9
35 #define GLUSTER_OPT_LOGFILE         "logfile"
36 #define GLUSTER_LOGFILE_DEFAULT     "-" /* handled in libgfapi as /dev/stderr */
37 
38 #define GERR_INDEX_HINT "hint: check in 'server' array index '%d'\n"
39 
40 typedef struct GlusterAIOCB {
41     int64_t size;
42     int ret;
43     Coroutine *coroutine;
44     AioContext *aio_context;
45 } GlusterAIOCB;
46 
47 typedef struct BDRVGlusterState {
48     struct glfs *glfs;
49     struct glfs_fd *fd;
50     char *logfile;
51     bool supports_seek_data;
52     int debug;
53 } BDRVGlusterState;
54 
55 typedef struct BDRVGlusterReopenState {
56     struct glfs *glfs;
57     struct glfs_fd *fd;
58 } BDRVGlusterReopenState;
59 
60 
61 typedef struct GlfsPreopened {
62     char *volume;
63     glfs_t *fs;
64     int ref;
65 } GlfsPreopened;
66 
67 typedef struct ListElement {
68     QLIST_ENTRY(ListElement) list;
69     GlfsPreopened saved;
70 } ListElement;
71 
72 static QLIST_HEAD(glfs_list, ListElement) glfs_list;
73 
74 static QemuOptsList qemu_gluster_create_opts = {
75     .name = "qemu-gluster-create-opts",
76     .head = QTAILQ_HEAD_INITIALIZER(qemu_gluster_create_opts.head),
77     .desc = {
78         {
79             .name = BLOCK_OPT_SIZE,
80             .type = QEMU_OPT_SIZE,
81             .help = "Virtual disk size"
82         },
83         {
84             .name = BLOCK_OPT_PREALLOC,
85             .type = QEMU_OPT_STRING,
86             .help = "Preallocation mode (allowed values: off, full)"
87         },
88         {
89             .name = GLUSTER_OPT_DEBUG,
90             .type = QEMU_OPT_NUMBER,
91             .help = "Gluster log level, valid range is 0-9",
92         },
93         {
94             .name = GLUSTER_OPT_LOGFILE,
95             .type = QEMU_OPT_STRING,
96             .help = "Logfile path of libgfapi",
97         },
98         { /* end of list */ }
99     }
100 };
101 
102 static QemuOptsList runtime_opts = {
103     .name = "gluster",
104     .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
105     .desc = {
106         {
107             .name = GLUSTER_OPT_FILENAME,
108             .type = QEMU_OPT_STRING,
109             .help = "URL to the gluster image",
110         },
111         {
112             .name = GLUSTER_OPT_DEBUG,
113             .type = QEMU_OPT_NUMBER,
114             .help = "Gluster log level, valid range is 0-9",
115         },
116         {
117             .name = GLUSTER_OPT_LOGFILE,
118             .type = QEMU_OPT_STRING,
119             .help = "Logfile path of libgfapi",
120         },
121         { /* end of list */ }
122     },
123 };
124 
125 static QemuOptsList runtime_json_opts = {
126     .name = "gluster_json",
127     .head = QTAILQ_HEAD_INITIALIZER(runtime_json_opts.head),
128     .desc = {
129         {
130             .name = GLUSTER_OPT_VOLUME,
131             .type = QEMU_OPT_STRING,
132             .help = "name of gluster volume where VM image resides",
133         },
134         {
135             .name = GLUSTER_OPT_PATH,
136             .type = QEMU_OPT_STRING,
137             .help = "absolute path to image file in gluster volume",
138         },
139         {
140             .name = GLUSTER_OPT_DEBUG,
141             .type = QEMU_OPT_NUMBER,
142             .help = "Gluster log level, valid range is 0-9",
143         },
144         { /* end of list */ }
145     },
146 };
147 
148 static QemuOptsList runtime_type_opts = {
149     .name = "gluster_type",
150     .head = QTAILQ_HEAD_INITIALIZER(runtime_type_opts.head),
151     .desc = {
152         {
153             .name = GLUSTER_OPT_TYPE,
154             .type = QEMU_OPT_STRING,
155             .help = "inet|unix",
156         },
157         { /* end of list */ }
158     },
159 };
160 
161 static QemuOptsList runtime_unix_opts = {
162     .name = "gluster_unix",
163     .head = QTAILQ_HEAD_INITIALIZER(runtime_unix_opts.head),
164     .desc = {
165         {
166             .name = GLUSTER_OPT_SOCKET,
167             .type = QEMU_OPT_STRING,
168             .help = "socket file path)",
169         },
170         { /* end of list */ }
171     },
172 };
173 
174 static QemuOptsList runtime_inet_opts = {
175     .name = "gluster_inet",
176     .head = QTAILQ_HEAD_INITIALIZER(runtime_inet_opts.head),
177     .desc = {
178         {
179             .name = GLUSTER_OPT_TYPE,
180             .type = QEMU_OPT_STRING,
181             .help = "inet|unix",
182         },
183         {
184             .name = GLUSTER_OPT_HOST,
185             .type = QEMU_OPT_STRING,
186             .help = "host address (hostname/ipv4/ipv6 addresses)",
187         },
188         {
189             .name = GLUSTER_OPT_PORT,
190             .type = QEMU_OPT_STRING,
191             .help = "port number on which glusterd is listening (default 24007)",
192         },
193         {
194             .name = "to",
195             .type = QEMU_OPT_NUMBER,
196             .help = "max port number, not supported by gluster",
197         },
198         {
199             .name = "ipv4",
200             .type = QEMU_OPT_BOOL,
201             .help = "ipv4 bool value, not supported by gluster",
202         },
203         {
204             .name = "ipv6",
205             .type = QEMU_OPT_BOOL,
206             .help = "ipv6 bool value, not supported by gluster",
207         },
208         { /* end of list */ }
209     },
210 };
211 
212 static void glfs_set_preopened(const char *volume, glfs_t *fs)
213 {
214     ListElement *entry = NULL;
215 
216     entry = g_new(ListElement, 1);
217 
218     entry->saved.volume = g_strdup(volume);
219 
220     entry->saved.fs = fs;
221     entry->saved.ref = 1;
222 
223     QLIST_INSERT_HEAD(&glfs_list, entry, list);
224 }
225 
226 static glfs_t *glfs_find_preopened(const char *volume)
227 {
228     ListElement *entry = NULL;
229 
230      QLIST_FOREACH(entry, &glfs_list, list) {
231         if (strcmp(entry->saved.volume, volume) == 0) {
232             entry->saved.ref++;
233             return entry->saved.fs;
234         }
235      }
236 
237     return NULL;
238 }
239 
240 static void glfs_clear_preopened(glfs_t *fs)
241 {
242     ListElement *entry = NULL;
243     ListElement *next;
244 
245     if (fs == NULL) {
246         return;
247     }
248 
249     QLIST_FOREACH_SAFE(entry, &glfs_list, list, next) {
250         if (entry->saved.fs == fs) {
251             if (--entry->saved.ref) {
252                 return;
253             }
254 
255             QLIST_REMOVE(entry, list);
256 
257             glfs_fini(entry->saved.fs);
258             g_free(entry->saved.volume);
259             g_free(entry);
260         }
261     }
262 }
263 
264 static int parse_volume_options(BlockdevOptionsGluster *gconf, char *path)
265 {
266     char *p, *q;
267 
268     if (!path) {
269         return -EINVAL;
270     }
271 
272     /* volume */
273     p = q = path + strspn(path, "/");
274     p += strcspn(p, "/");
275     if (*p == '\0') {
276         return -EINVAL;
277     }
278     gconf->volume = g_strndup(q, p - q);
279 
280     /* path */
281     p += strspn(p, "/");
282     if (*p == '\0') {
283         return -EINVAL;
284     }
285     gconf->path = g_strdup(p);
286     return 0;
287 }
288 
289 /*
290  * file=gluster[+transport]://[host[:port]]/volume/path[?socket=...]
291  *
292  * 'gluster' is the protocol.
293  *
294  * 'transport' specifies the transport type used to connect to gluster
295  * management daemon (glusterd). Valid transport types are
296  * tcp or unix. If a transport type isn't specified, then tcp type is assumed.
297  *
298  * 'host' specifies the host where the volume file specification for
299  * the given volume resides. This can be either hostname or ipv4 address.
300  * If transport type is 'unix', then 'host' field should not be specified.
301  * The 'socket' field needs to be populated with the path to unix domain
302  * socket.
303  *
304  * 'port' is the port number on which glusterd is listening. This is optional
305  * and if not specified, QEMU will send 0 which will make gluster to use the
306  * default port. If the transport type is unix, then 'port' should not be
307  * specified.
308  *
309  * 'volume' is the name of the gluster volume which contains the VM image.
310  *
311  * 'path' is the path to the actual VM image that resides on gluster volume.
312  *
313  * Examples:
314  *
315  * file=gluster://1.2.3.4/testvol/a.img
316  * file=gluster+tcp://1.2.3.4/testvol/a.img
317  * file=gluster+tcp://1.2.3.4:24007/testvol/dir/a.img
318  * file=gluster+tcp://host.domain.com:24007/testvol/dir/a.img
319  * file=gluster+unix:///testvol/dir/a.img?socket=/tmp/glusterd.socket
320  */
321 static int qemu_gluster_parse_uri(BlockdevOptionsGluster *gconf,
322                                   const char *filename)
323 {
324     SocketAddress *gsconf;
325     URI *uri;
326     QueryParams *qp = NULL;
327     bool is_unix = false;
328     int ret = 0;
329 
330     uri = uri_parse(filename);
331     if (!uri) {
332         return -EINVAL;
333     }
334 
335     gconf->server = g_new0(SocketAddressList, 1);
336     gconf->server->value = gsconf = g_new0(SocketAddress, 1);
337 
338     /* transport */
339     if (!uri->scheme || !strcmp(uri->scheme, "gluster")) {
340         gsconf->type = SOCKET_ADDRESS_TYPE_INET;
341     } else if (!strcmp(uri->scheme, "gluster+tcp")) {
342         gsconf->type = SOCKET_ADDRESS_TYPE_INET;
343     } else if (!strcmp(uri->scheme, "gluster+unix")) {
344         gsconf->type = SOCKET_ADDRESS_TYPE_UNIX;
345         is_unix = true;
346     } else if (!strcmp(uri->scheme, "gluster+rdma")) {
347         gsconf->type = SOCKET_ADDRESS_TYPE_INET;
348         error_report("Warning: rdma feature is not supported, falling "
349                      "back to tcp");
350     } else {
351         ret = -EINVAL;
352         goto out;
353     }
354 
355     ret = parse_volume_options(gconf, uri->path);
356     if (ret < 0) {
357         goto out;
358     }
359 
360     qp = query_params_parse(uri->query);
361     if (qp->n > 1 || (is_unix && !qp->n) || (!is_unix && qp->n)) {
362         ret = -EINVAL;
363         goto out;
364     }
365 
366     if (is_unix) {
367         if (uri->server || uri->port) {
368             ret = -EINVAL;
369             goto out;
370         }
371         if (strcmp(qp->p[0].name, "socket")) {
372             ret = -EINVAL;
373             goto out;
374         }
375         gsconf->u.q_unix.path = g_strdup(qp->p[0].value);
376     } else {
377         gsconf->u.inet.host = g_strdup(uri->server ? uri->server : "localhost");
378         if (uri->port) {
379             gsconf->u.inet.port = g_strdup_printf("%d", uri->port);
380         } else {
381             gsconf->u.inet.port = g_strdup_printf("%d", GLUSTER_DEFAULT_PORT);
382         }
383     }
384 
385 out:
386     if (qp) {
387         query_params_free(qp);
388     }
389     uri_free(uri);
390     return ret;
391 }
392 
393 static struct glfs *qemu_gluster_glfs_init(BlockdevOptionsGluster *gconf,
394                                            Error **errp)
395 {
396     struct glfs *glfs;
397     int ret;
398     int old_errno;
399     SocketAddressList *server;
400     unsigned long long port;
401 
402     glfs = glfs_find_preopened(gconf->volume);
403     if (glfs) {
404         return glfs;
405     }
406 
407     glfs = glfs_new(gconf->volume);
408     if (!glfs) {
409         goto out;
410     }
411 
412     glfs_set_preopened(gconf->volume, glfs);
413 
414     for (server = gconf->server; server; server = server->next) {
415         switch (server->value->type) {
416         case SOCKET_ADDRESS_TYPE_UNIX:
417             ret = glfs_set_volfile_server(glfs, "unix",
418                                    server->value->u.q_unix.path, 0);
419             break;
420         case SOCKET_ADDRESS_TYPE_INET:
421             if (parse_uint_full(server->value->u.inet.port, &port, 10) < 0 ||
422                 port > 65535) {
423                 error_setg(errp, "'%s' is not a valid port number",
424                            server->value->u.inet.port);
425                 errno = EINVAL;
426                 goto out;
427             }
428             ret = glfs_set_volfile_server(glfs, "tcp",
429                                    server->value->u.inet.host,
430                                    (int)port);
431             break;
432         case SOCKET_ADDRESS_TYPE_VSOCK:
433         case SOCKET_ADDRESS_TYPE_FD:
434         default:
435             abort();
436         }
437 
438         if (ret < 0) {
439             goto out;
440         }
441     }
442 
443     ret = glfs_set_logging(glfs, gconf->logfile, gconf->debug);
444     if (ret < 0) {
445         goto out;
446     }
447 
448     ret = glfs_init(glfs);
449     if (ret) {
450         error_setg(errp, "Gluster connection for volume %s, path %s failed"
451                          " to connect", gconf->volume, gconf->path);
452         for (server = gconf->server; server; server = server->next) {
453             if (server->value->type  == SOCKET_ADDRESS_TYPE_UNIX) {
454                 error_append_hint(errp, "hint: failed on socket %s ",
455                                   server->value->u.q_unix.path);
456             } else {
457                 error_append_hint(errp, "hint: failed on host %s and port %s ",
458                                   server->value->u.inet.host,
459                                   server->value->u.inet.port);
460             }
461         }
462 
463         error_append_hint(errp, "Please refer to gluster logs for more info\n");
464 
465         /* glfs_init sometimes doesn't set errno although docs suggest that */
466         if (errno == 0) {
467             errno = EINVAL;
468         }
469 
470         goto out;
471     }
472     return glfs;
473 
474 out:
475     if (glfs) {
476         old_errno = errno;
477         glfs_clear_preopened(glfs);
478         errno = old_errno;
479     }
480     return NULL;
481 }
482 
483 /*
484  * Convert the json formatted command line into qapi.
485 */
486 static int qemu_gluster_parse_json(BlockdevOptionsGluster *gconf,
487                                   QDict *options, Error **errp)
488 {
489     QemuOpts *opts;
490     SocketAddress *gsconf = NULL;
491     SocketAddressList *curr = NULL;
492     QDict *backing_options = NULL;
493     Error *local_err = NULL;
494     char *str = NULL;
495     const char *ptr;
496     size_t num_servers;
497     int i, type;
498 
499     /* create opts info from runtime_json_opts list */
500     opts = qemu_opts_create(&runtime_json_opts, NULL, 0, &error_abort);
501     qemu_opts_absorb_qdict(opts, options, &local_err);
502     if (local_err) {
503         goto out;
504     }
505 
506     num_servers = qdict_array_entries(options, GLUSTER_OPT_SERVER_PATTERN);
507     if (num_servers < 1) {
508         error_setg(&local_err, QERR_MISSING_PARAMETER, "server");
509         goto out;
510     }
511 
512     ptr = qemu_opt_get(opts, GLUSTER_OPT_VOLUME);
513     if (!ptr) {
514         error_setg(&local_err, QERR_MISSING_PARAMETER, GLUSTER_OPT_VOLUME);
515         goto out;
516     }
517     gconf->volume = g_strdup(ptr);
518 
519     ptr = qemu_opt_get(opts, GLUSTER_OPT_PATH);
520     if (!ptr) {
521         error_setg(&local_err, QERR_MISSING_PARAMETER, GLUSTER_OPT_PATH);
522         goto out;
523     }
524     gconf->path = g_strdup(ptr);
525     qemu_opts_del(opts);
526 
527     for (i = 0; i < num_servers; i++) {
528         str = g_strdup_printf(GLUSTER_OPT_SERVER_PATTERN"%d.", i);
529         qdict_extract_subqdict(options, &backing_options, str);
530 
531         /* create opts info from runtime_type_opts list */
532         opts = qemu_opts_create(&runtime_type_opts, NULL, 0, &error_abort);
533         qemu_opts_absorb_qdict(opts, backing_options, &local_err);
534         if (local_err) {
535             goto out;
536         }
537 
538         ptr = qemu_opt_get(opts, GLUSTER_OPT_TYPE);
539         if (!ptr) {
540             error_setg(&local_err, QERR_MISSING_PARAMETER, GLUSTER_OPT_TYPE);
541             error_append_hint(&local_err, GERR_INDEX_HINT, i);
542             goto out;
543 
544         }
545         gsconf = g_new0(SocketAddress, 1);
546         if (!strcmp(ptr, "tcp")) {
547             ptr = "inet";       /* accept legacy "tcp" */
548         }
549         type = qapi_enum_parse(SocketAddressType_lookup, ptr,
550                                SOCKET_ADDRESS_TYPE__MAX, -1, NULL);
551         if (type != SOCKET_ADDRESS_TYPE_INET
552             && type != SOCKET_ADDRESS_TYPE_UNIX) {
553             error_setg(&local_err,
554                        "Parameter '%s' may be 'inet' or 'unix'",
555                        GLUSTER_OPT_TYPE);
556             error_append_hint(&local_err, GERR_INDEX_HINT, i);
557             goto out;
558         }
559         gsconf->type = type;
560         qemu_opts_del(opts);
561 
562         if (gsconf->type == SOCKET_ADDRESS_TYPE_INET) {
563             /* create opts info from runtime_inet_opts list */
564             opts = qemu_opts_create(&runtime_inet_opts, NULL, 0, &error_abort);
565             qemu_opts_absorb_qdict(opts, backing_options, &local_err);
566             if (local_err) {
567                 goto out;
568             }
569 
570             ptr = qemu_opt_get(opts, GLUSTER_OPT_HOST);
571             if (!ptr) {
572                 error_setg(&local_err, QERR_MISSING_PARAMETER,
573                            GLUSTER_OPT_HOST);
574                 error_append_hint(&local_err, GERR_INDEX_HINT, i);
575                 goto out;
576             }
577             gsconf->u.inet.host = g_strdup(ptr);
578             ptr = qemu_opt_get(opts, GLUSTER_OPT_PORT);
579             if (!ptr) {
580                 error_setg(&local_err, QERR_MISSING_PARAMETER,
581                            GLUSTER_OPT_PORT);
582                 error_append_hint(&local_err, GERR_INDEX_HINT, i);
583                 goto out;
584             }
585             gsconf->u.inet.port = g_strdup(ptr);
586 
587             /* defend for unsupported fields in InetSocketAddress,
588              * i.e. @ipv4, @ipv6  and @to
589              */
590             ptr = qemu_opt_get(opts, GLUSTER_OPT_TO);
591             if (ptr) {
592                 gsconf->u.inet.has_to = true;
593             }
594             ptr = qemu_opt_get(opts, GLUSTER_OPT_IPV4);
595             if (ptr) {
596                 gsconf->u.inet.has_ipv4 = true;
597             }
598             ptr = qemu_opt_get(opts, GLUSTER_OPT_IPV6);
599             if (ptr) {
600                 gsconf->u.inet.has_ipv6 = true;
601             }
602             if (gsconf->u.inet.has_to) {
603                 error_setg(&local_err, "Parameter 'to' not supported");
604                 goto out;
605             }
606             if (gsconf->u.inet.has_ipv4 || gsconf->u.inet.has_ipv6) {
607                 error_setg(&local_err, "Parameters 'ipv4/ipv6' not supported");
608                 goto out;
609             }
610             qemu_opts_del(opts);
611         } else {
612             /* create opts info from runtime_unix_opts list */
613             opts = qemu_opts_create(&runtime_unix_opts, NULL, 0, &error_abort);
614             qemu_opts_absorb_qdict(opts, backing_options, &local_err);
615             if (local_err) {
616                 goto out;
617             }
618 
619             ptr = qemu_opt_get(opts, GLUSTER_OPT_SOCKET);
620             if (!ptr) {
621                 error_setg(&local_err, QERR_MISSING_PARAMETER,
622                            GLUSTER_OPT_SOCKET);
623                 error_append_hint(&local_err, GERR_INDEX_HINT, i);
624                 goto out;
625             }
626             gsconf->u.q_unix.path = g_strdup(ptr);
627             qemu_opts_del(opts);
628         }
629 
630         if (gconf->server == NULL) {
631             gconf->server = g_new0(SocketAddressList, 1);
632             gconf->server->value = gsconf;
633             curr = gconf->server;
634         } else {
635             curr->next = g_new0(SocketAddressList, 1);
636             curr->next->value = gsconf;
637             curr = curr->next;
638         }
639         gsconf = NULL;
640 
641         QDECREF(backing_options);
642         backing_options = NULL;
643         g_free(str);
644         str = NULL;
645     }
646 
647     return 0;
648 
649 out:
650     error_propagate(errp, local_err);
651     qapi_free_SocketAddress(gsconf);
652     qemu_opts_del(opts);
653     g_free(str);
654     QDECREF(backing_options);
655     errno = EINVAL;
656     return -errno;
657 }
658 
659 static struct glfs *qemu_gluster_init(BlockdevOptionsGluster *gconf,
660                                       const char *filename,
661                                       QDict *options, Error **errp)
662 {
663     int ret;
664     if (filename) {
665         ret = qemu_gluster_parse_uri(gconf, filename);
666         if (ret < 0) {
667             error_setg(errp, "invalid URI");
668             error_append_hint(errp, "Usage: file=gluster[+transport]://"
669                                     "[host[:port]]volume/path[?socket=...]"
670                                     "[,file.debug=N]"
671                                     "[,file.logfile=/path/filename.log]\n");
672             errno = -ret;
673             return NULL;
674         }
675     } else {
676         ret = qemu_gluster_parse_json(gconf, options, errp);
677         if (ret < 0) {
678             error_append_hint(errp, "Usage: "
679                              "-drive driver=qcow2,file.driver=gluster,"
680                              "file.volume=testvol,file.path=/path/a.qcow2"
681                              "[,file.debug=9]"
682                              "[,file.logfile=/path/filename.log],"
683                              "file.server.0.type=inet,"
684                              "file.server.0.host=1.2.3.4,"
685                              "file.server.0.port=24007,"
686                              "file.server.1.transport=unix,"
687                              "file.server.1.socket=/var/run/glusterd.socket ..."
688                              "\n");
689             errno = -ret;
690             return NULL;
691         }
692 
693     }
694 
695     return qemu_gluster_glfs_init(gconf, errp);
696 }
697 
698 /*
699  * AIO callback routine called from GlusterFS thread.
700  */
701 static void gluster_finish_aiocb(struct glfs_fd *fd, ssize_t ret, void *arg)
702 {
703     GlusterAIOCB *acb = (GlusterAIOCB *)arg;
704 
705     if (!ret || ret == acb->size) {
706         acb->ret = 0; /* Success */
707     } else if (ret < 0) {
708         acb->ret = -errno; /* Read/Write failed */
709     } else {
710         acb->ret = -EIO; /* Partial read/write - fail it */
711     }
712 
713     aio_co_schedule(acb->aio_context, acb->coroutine);
714 }
715 
716 static void qemu_gluster_parse_flags(int bdrv_flags, int *open_flags)
717 {
718     assert(open_flags != NULL);
719 
720     *open_flags |= O_BINARY;
721 
722     if (bdrv_flags & BDRV_O_RDWR) {
723         *open_flags |= O_RDWR;
724     } else {
725         *open_flags |= O_RDONLY;
726     }
727 
728     if ((bdrv_flags & BDRV_O_NOCACHE)) {
729         *open_flags |= O_DIRECT;
730     }
731 }
732 
733 /*
734  * Do SEEK_DATA/HOLE to detect if it is functional. Older broken versions of
735  * gfapi incorrectly return the current offset when SEEK_DATA/HOLE is used.
736  * - Corrected versions return -1 and set errno to EINVAL.
737  * - Versions that support SEEK_DATA/HOLE correctly, will return -1 and set
738  *   errno to ENXIO when SEEK_DATA is called with a position of EOF.
739  */
740 static bool qemu_gluster_test_seek(struct glfs_fd *fd)
741 {
742     off_t ret = 0;
743 
744 #if defined SEEK_HOLE && defined SEEK_DATA
745     off_t eof;
746 
747     eof = glfs_lseek(fd, 0, SEEK_END);
748     if (eof < 0) {
749         /* this should never occur */
750         return false;
751     }
752 
753     /* this should always fail with ENXIO if SEEK_DATA is supported */
754     ret = glfs_lseek(fd, eof, SEEK_DATA);
755 #endif
756 
757     return (ret < 0) && (errno == ENXIO);
758 }
759 
760 static int qemu_gluster_open(BlockDriverState *bs,  QDict *options,
761                              int bdrv_flags, Error **errp)
762 {
763     BDRVGlusterState *s = bs->opaque;
764     int open_flags = 0;
765     int ret = 0;
766     BlockdevOptionsGluster *gconf = NULL;
767     QemuOpts *opts;
768     Error *local_err = NULL;
769     const char *filename, *logfile;
770 
771     opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
772     qemu_opts_absorb_qdict(opts, options, &local_err);
773     if (local_err) {
774         error_propagate(errp, local_err);
775         ret = -EINVAL;
776         goto out;
777     }
778 
779     filename = qemu_opt_get(opts, GLUSTER_OPT_FILENAME);
780 
781     s->debug = qemu_opt_get_number(opts, GLUSTER_OPT_DEBUG,
782                                    GLUSTER_DEBUG_DEFAULT);
783     if (s->debug < 0) {
784         s->debug = 0;
785     } else if (s->debug > GLUSTER_DEBUG_MAX) {
786         s->debug = GLUSTER_DEBUG_MAX;
787     }
788 
789     gconf = g_new0(BlockdevOptionsGluster, 1);
790     gconf->debug = s->debug;
791     gconf->has_debug = true;
792 
793     logfile = qemu_opt_get(opts, GLUSTER_OPT_LOGFILE);
794     s->logfile = g_strdup(logfile ? logfile : GLUSTER_LOGFILE_DEFAULT);
795 
796     gconf->logfile = g_strdup(s->logfile);
797     gconf->has_logfile = true;
798 
799     s->glfs = qemu_gluster_init(gconf, filename, options, errp);
800     if (!s->glfs) {
801         ret = -errno;
802         goto out;
803     }
804 
805 #ifdef CONFIG_GLUSTERFS_XLATOR_OPT
806     /* Without this, if fsync fails for a recoverable reason (for instance,
807      * ENOSPC), gluster will dump its cache, preventing retries.  This means
808      * almost certain data loss.  Not all gluster versions support the
809      * 'resync-failed-syncs-after-fsync' key value, but there is no way to
810      * discover during runtime if it is supported (this api returns success for
811      * unknown key/value pairs) */
812     ret = glfs_set_xlator_option(s->glfs, "*-write-behind",
813                                           "resync-failed-syncs-after-fsync",
814                                           "on");
815     if (ret < 0) {
816         error_setg_errno(errp, errno, "Unable to set xlator key/value pair");
817         ret = -errno;
818         goto out;
819     }
820 #endif
821 
822     qemu_gluster_parse_flags(bdrv_flags, &open_flags);
823 
824     s->fd = glfs_open(s->glfs, gconf->path, open_flags);
825     if (!s->fd) {
826         ret = -errno;
827     }
828 
829     s->supports_seek_data = qemu_gluster_test_seek(s->fd);
830 
831 out:
832     qemu_opts_del(opts);
833     qapi_free_BlockdevOptionsGluster(gconf);
834     if (!ret) {
835         return ret;
836     }
837     g_free(s->logfile);
838     if (s->fd) {
839         glfs_close(s->fd);
840     }
841 
842     glfs_clear_preopened(s->glfs);
843 
844     return ret;
845 }
846 
847 static int qemu_gluster_reopen_prepare(BDRVReopenState *state,
848                                        BlockReopenQueue *queue, Error **errp)
849 {
850     int ret = 0;
851     BDRVGlusterState *s;
852     BDRVGlusterReopenState *reop_s;
853     BlockdevOptionsGluster *gconf;
854     int open_flags = 0;
855 
856     assert(state != NULL);
857     assert(state->bs != NULL);
858 
859     s = state->bs->opaque;
860 
861     state->opaque = g_new0(BDRVGlusterReopenState, 1);
862     reop_s = state->opaque;
863 
864     qemu_gluster_parse_flags(state->flags, &open_flags);
865 
866     gconf = g_new0(BlockdevOptionsGluster, 1);
867     gconf->debug = s->debug;
868     gconf->has_debug = true;
869     gconf->logfile = g_strdup(s->logfile);
870     gconf->has_logfile = true;
871     reop_s->glfs = qemu_gluster_init(gconf, state->bs->filename, NULL, errp);
872     if (reop_s->glfs == NULL) {
873         ret = -errno;
874         goto exit;
875     }
876 
877 #ifdef CONFIG_GLUSTERFS_XLATOR_OPT
878     ret = glfs_set_xlator_option(reop_s->glfs, "*-write-behind",
879                                  "resync-failed-syncs-after-fsync", "on");
880     if (ret < 0) {
881         error_setg_errno(errp, errno, "Unable to set xlator key/value pair");
882         ret = -errno;
883         goto exit;
884     }
885 #endif
886 
887     reop_s->fd = glfs_open(reop_s->glfs, gconf->path, open_flags);
888     if (reop_s->fd == NULL) {
889         /* reops->glfs will be cleaned up in _abort */
890         ret = -errno;
891         goto exit;
892     }
893 
894 exit:
895     /* state->opaque will be freed in either the _abort or _commit */
896     qapi_free_BlockdevOptionsGluster(gconf);
897     return ret;
898 }
899 
900 static void qemu_gluster_reopen_commit(BDRVReopenState *state)
901 {
902     BDRVGlusterReopenState *reop_s = state->opaque;
903     BDRVGlusterState *s = state->bs->opaque;
904 
905 
906     /* close the old */
907     if (s->fd) {
908         glfs_close(s->fd);
909     }
910 
911     glfs_clear_preopened(s->glfs);
912 
913     /* use the newly opened image / connection */
914     s->fd         = reop_s->fd;
915     s->glfs       = reop_s->glfs;
916 
917     g_free(state->opaque);
918     state->opaque = NULL;
919 
920     return;
921 }
922 
923 
924 static void qemu_gluster_reopen_abort(BDRVReopenState *state)
925 {
926     BDRVGlusterReopenState *reop_s = state->opaque;
927 
928     if (reop_s == NULL) {
929         return;
930     }
931 
932     if (reop_s->fd) {
933         glfs_close(reop_s->fd);
934     }
935 
936     glfs_clear_preopened(reop_s->glfs);
937 
938     g_free(state->opaque);
939     state->opaque = NULL;
940 
941     return;
942 }
943 
944 #ifdef CONFIG_GLUSTERFS_ZEROFILL
945 static coroutine_fn int qemu_gluster_co_pwrite_zeroes(BlockDriverState *bs,
946                                                       int64_t offset,
947                                                       int size,
948                                                       BdrvRequestFlags flags)
949 {
950     int ret;
951     GlusterAIOCB acb;
952     BDRVGlusterState *s = bs->opaque;
953 
954     acb.size = size;
955     acb.ret = 0;
956     acb.coroutine = qemu_coroutine_self();
957     acb.aio_context = bdrv_get_aio_context(bs);
958 
959     ret = glfs_zerofill_async(s->fd, offset, size, gluster_finish_aiocb, &acb);
960     if (ret < 0) {
961         return -errno;
962     }
963 
964     qemu_coroutine_yield();
965     return acb.ret;
966 }
967 
968 static inline bool gluster_supports_zerofill(void)
969 {
970     return 1;
971 }
972 
973 static inline int qemu_gluster_zerofill(struct glfs_fd *fd, int64_t offset,
974                                         int64_t size)
975 {
976     return glfs_zerofill(fd, offset, size);
977 }
978 
979 #else
980 static inline bool gluster_supports_zerofill(void)
981 {
982     return 0;
983 }
984 
985 static inline int qemu_gluster_zerofill(struct glfs_fd *fd, int64_t offset,
986                                         int64_t size)
987 {
988     return 0;
989 }
990 #endif
991 
992 static int qemu_gluster_create(const char *filename,
993                                QemuOpts *opts, Error **errp)
994 {
995     BlockdevOptionsGluster *gconf;
996     struct glfs *glfs;
997     struct glfs_fd *fd;
998     int ret = 0;
999     int prealloc = 0;
1000     int64_t total_size = 0;
1001     char *tmp = NULL;
1002 
1003     gconf = g_new0(BlockdevOptionsGluster, 1);
1004     gconf->debug = qemu_opt_get_number_del(opts, GLUSTER_OPT_DEBUG,
1005                                            GLUSTER_DEBUG_DEFAULT);
1006     if (gconf->debug < 0) {
1007         gconf->debug = 0;
1008     } else if (gconf->debug > GLUSTER_DEBUG_MAX) {
1009         gconf->debug = GLUSTER_DEBUG_MAX;
1010     }
1011     gconf->has_debug = true;
1012 
1013     gconf->logfile = qemu_opt_get_del(opts, GLUSTER_OPT_LOGFILE);
1014     if (!gconf->logfile) {
1015         gconf->logfile = g_strdup(GLUSTER_LOGFILE_DEFAULT);
1016     }
1017     gconf->has_logfile = true;
1018 
1019     glfs = qemu_gluster_init(gconf, filename, NULL, errp);
1020     if (!glfs) {
1021         ret = -errno;
1022         goto out;
1023     }
1024 
1025     total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
1026                           BDRV_SECTOR_SIZE);
1027 
1028     tmp = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
1029     if (!tmp || !strcmp(tmp, "off")) {
1030         prealloc = 0;
1031     } else if (!strcmp(tmp, "full") && gluster_supports_zerofill()) {
1032         prealloc = 1;
1033     } else {
1034         error_setg(errp, "Invalid preallocation mode: '%s'"
1035                          " or GlusterFS doesn't support zerofill API", tmp);
1036         ret = -EINVAL;
1037         goto out;
1038     }
1039 
1040     fd = glfs_creat(glfs, gconf->path,
1041                     O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, S_IRUSR | S_IWUSR);
1042     if (!fd) {
1043         ret = -errno;
1044     } else {
1045         if (!glfs_ftruncate(fd, total_size)) {
1046             if (prealloc && qemu_gluster_zerofill(fd, 0, total_size)) {
1047                 ret = -errno;
1048             }
1049         } else {
1050             ret = -errno;
1051         }
1052 
1053         if (glfs_close(fd) != 0) {
1054             ret = -errno;
1055         }
1056     }
1057 out:
1058     g_free(tmp);
1059     qapi_free_BlockdevOptionsGluster(gconf);
1060     glfs_clear_preopened(glfs);
1061     return ret;
1062 }
1063 
1064 static coroutine_fn int qemu_gluster_co_rw(BlockDriverState *bs,
1065                                            int64_t sector_num, int nb_sectors,
1066                                            QEMUIOVector *qiov, int write)
1067 {
1068     int ret;
1069     GlusterAIOCB acb;
1070     BDRVGlusterState *s = bs->opaque;
1071     size_t size = nb_sectors * BDRV_SECTOR_SIZE;
1072     off_t offset = sector_num * BDRV_SECTOR_SIZE;
1073 
1074     acb.size = size;
1075     acb.ret = 0;
1076     acb.coroutine = qemu_coroutine_self();
1077     acb.aio_context = bdrv_get_aio_context(bs);
1078 
1079     if (write) {
1080         ret = glfs_pwritev_async(s->fd, qiov->iov, qiov->niov, offset, 0,
1081                                  gluster_finish_aiocb, &acb);
1082     } else {
1083         ret = glfs_preadv_async(s->fd, qiov->iov, qiov->niov, offset, 0,
1084                                 gluster_finish_aiocb, &acb);
1085     }
1086 
1087     if (ret < 0) {
1088         return -errno;
1089     }
1090 
1091     qemu_coroutine_yield();
1092     return acb.ret;
1093 }
1094 
1095 static int qemu_gluster_truncate(BlockDriverState *bs, int64_t offset,
1096                                  Error **errp)
1097 {
1098     int ret;
1099     BDRVGlusterState *s = bs->opaque;
1100 
1101     ret = glfs_ftruncate(s->fd, offset);
1102     if (ret < 0) {
1103         ret = -errno;
1104         error_setg_errno(errp, -ret, "Failed to truncate file");
1105         return ret;
1106     }
1107 
1108     return 0;
1109 }
1110 
1111 static coroutine_fn int qemu_gluster_co_readv(BlockDriverState *bs,
1112                                               int64_t sector_num,
1113                                               int nb_sectors,
1114                                               QEMUIOVector *qiov)
1115 {
1116     return qemu_gluster_co_rw(bs, sector_num, nb_sectors, qiov, 0);
1117 }
1118 
1119 static coroutine_fn int qemu_gluster_co_writev(BlockDriverState *bs,
1120                                                int64_t sector_num,
1121                                                int nb_sectors,
1122                                                QEMUIOVector *qiov)
1123 {
1124     return qemu_gluster_co_rw(bs, sector_num, nb_sectors, qiov, 1);
1125 }
1126 
1127 static void qemu_gluster_close(BlockDriverState *bs)
1128 {
1129     BDRVGlusterState *s = bs->opaque;
1130 
1131     g_free(s->logfile);
1132     if (s->fd) {
1133         glfs_close(s->fd);
1134         s->fd = NULL;
1135     }
1136     glfs_clear_preopened(s->glfs);
1137 }
1138 
1139 static coroutine_fn int qemu_gluster_co_flush_to_disk(BlockDriverState *bs)
1140 {
1141     int ret;
1142     GlusterAIOCB acb;
1143     BDRVGlusterState *s = bs->opaque;
1144 
1145     acb.size = 0;
1146     acb.ret = 0;
1147     acb.coroutine = qemu_coroutine_self();
1148     acb.aio_context = bdrv_get_aio_context(bs);
1149 
1150     ret = glfs_fsync_async(s->fd, gluster_finish_aiocb, &acb);
1151     if (ret < 0) {
1152         ret = -errno;
1153         goto error;
1154     }
1155 
1156     qemu_coroutine_yield();
1157     if (acb.ret < 0) {
1158         ret = acb.ret;
1159         goto error;
1160     }
1161 
1162     return acb.ret;
1163 
1164 error:
1165     /* Some versions of Gluster (3.5.6 -> 3.5.8?) will not retain its cache
1166      * after a fsync failure, so we have no way of allowing the guest to safely
1167      * continue.  Gluster versions prior to 3.5.6 don't retain the cache
1168      * either, but will invalidate the fd on error, so this is again our only
1169      * option.
1170      *
1171      * The 'resync-failed-syncs-after-fsync' xlator option for the
1172      * write-behind cache will cause later gluster versions to retain its
1173      * cache after error, so long as the fd remains open.  However, we
1174      * currently have no way of knowing if this option is supported.
1175      *
1176      * TODO: Once gluster provides a way for us to determine if the option
1177      * is supported, bypass the closure and setting drv to NULL.  */
1178     qemu_gluster_close(bs);
1179     bs->drv = NULL;
1180     return ret;
1181 }
1182 
1183 #ifdef CONFIG_GLUSTERFS_DISCARD
1184 static coroutine_fn int qemu_gluster_co_pdiscard(BlockDriverState *bs,
1185                                                  int64_t offset, int size)
1186 {
1187     int ret;
1188     GlusterAIOCB acb;
1189     BDRVGlusterState *s = bs->opaque;
1190 
1191     acb.size = 0;
1192     acb.ret = 0;
1193     acb.coroutine = qemu_coroutine_self();
1194     acb.aio_context = bdrv_get_aio_context(bs);
1195 
1196     ret = glfs_discard_async(s->fd, offset, size, gluster_finish_aiocb, &acb);
1197     if (ret < 0) {
1198         return -errno;
1199     }
1200 
1201     qemu_coroutine_yield();
1202     return acb.ret;
1203 }
1204 #endif
1205 
1206 static int64_t qemu_gluster_getlength(BlockDriverState *bs)
1207 {
1208     BDRVGlusterState *s = bs->opaque;
1209     int64_t ret;
1210 
1211     ret = glfs_lseek(s->fd, 0, SEEK_END);
1212     if (ret < 0) {
1213         return -errno;
1214     } else {
1215         return ret;
1216     }
1217 }
1218 
1219 static int64_t qemu_gluster_allocated_file_size(BlockDriverState *bs)
1220 {
1221     BDRVGlusterState *s = bs->opaque;
1222     struct stat st;
1223     int ret;
1224 
1225     ret = glfs_fstat(s->fd, &st);
1226     if (ret < 0) {
1227         return -errno;
1228     } else {
1229         return st.st_blocks * 512;
1230     }
1231 }
1232 
1233 static int qemu_gluster_has_zero_init(BlockDriverState *bs)
1234 {
1235     /* GlusterFS volume could be backed by a block device */
1236     return 0;
1237 }
1238 
1239 /*
1240  * Find allocation range in @bs around offset @start.
1241  * May change underlying file descriptor's file offset.
1242  * If @start is not in a hole, store @start in @data, and the
1243  * beginning of the next hole in @hole, and return 0.
1244  * If @start is in a non-trailing hole, store @start in @hole and the
1245  * beginning of the next non-hole in @data, and return 0.
1246  * If @start is in a trailing hole or beyond EOF, return -ENXIO.
1247  * If we can't find out, return a negative errno other than -ENXIO.
1248  *
1249  * (Shamefully copied from file-posix.c, only miniscule adaptions.)
1250  */
1251 static int find_allocation(BlockDriverState *bs, off_t start,
1252                            off_t *data, off_t *hole)
1253 {
1254     BDRVGlusterState *s = bs->opaque;
1255 
1256     if (!s->supports_seek_data) {
1257         goto exit;
1258     }
1259 
1260 #if defined SEEK_HOLE && defined SEEK_DATA
1261     off_t offs;
1262 
1263     /*
1264      * SEEK_DATA cases:
1265      * D1. offs == start: start is in data
1266      * D2. offs > start: start is in a hole, next data at offs
1267      * D3. offs < 0, errno = ENXIO: either start is in a trailing hole
1268      *                              or start is beyond EOF
1269      *     If the latter happens, the file has been truncated behind
1270      *     our back since we opened it.  All bets are off then.
1271      *     Treating like a trailing hole is simplest.
1272      * D4. offs < 0, errno != ENXIO: we learned nothing
1273      */
1274     offs = glfs_lseek(s->fd, start, SEEK_DATA);
1275     if (offs < 0) {
1276         return -errno;          /* D3 or D4 */
1277     }
1278     assert(offs >= start);
1279 
1280     if (offs > start) {
1281         /* D2: in hole, next data at offs */
1282         *hole = start;
1283         *data = offs;
1284         return 0;
1285     }
1286 
1287     /* D1: in data, end not yet known */
1288 
1289     /*
1290      * SEEK_HOLE cases:
1291      * H1. offs == start: start is in a hole
1292      *     If this happens here, a hole has been dug behind our back
1293      *     since the previous lseek().
1294      * H2. offs > start: either start is in data, next hole at offs,
1295      *                   or start is in trailing hole, EOF at offs
1296      *     Linux treats trailing holes like any other hole: offs ==
1297      *     start.  Solaris seeks to EOF instead: offs > start (blech).
1298      *     If that happens here, a hole has been dug behind our back
1299      *     since the previous lseek().
1300      * H3. offs < 0, errno = ENXIO: start is beyond EOF
1301      *     If this happens, the file has been truncated behind our
1302      *     back since we opened it.  Treat it like a trailing hole.
1303      * H4. offs < 0, errno != ENXIO: we learned nothing
1304      *     Pretend we know nothing at all, i.e. "forget" about D1.
1305      */
1306     offs = glfs_lseek(s->fd, start, SEEK_HOLE);
1307     if (offs < 0) {
1308         return -errno;          /* D1 and (H3 or H4) */
1309     }
1310     assert(offs >= start);
1311 
1312     if (offs > start) {
1313         /*
1314          * D1 and H2: either in data, next hole at offs, or it was in
1315          * data but is now in a trailing hole.  In the latter case,
1316          * all bets are off.  Treating it as if it there was data all
1317          * the way to EOF is safe, so simply do that.
1318          */
1319         *data = start;
1320         *hole = offs;
1321         return 0;
1322     }
1323 
1324     /* D1 and H1 */
1325     return -EBUSY;
1326 #endif
1327 
1328 exit:
1329     return -ENOTSUP;
1330 }
1331 
1332 /*
1333  * Returns the allocation status of the specified sectors.
1334  *
1335  * If 'sector_num' is beyond the end of the disk image the return value is 0
1336  * and 'pnum' is set to 0.
1337  *
1338  * 'pnum' is set to the number of sectors (including and immediately following
1339  * the specified sector) that are known to be in the same
1340  * allocated/unallocated state.
1341  *
1342  * 'nb_sectors' is the max value 'pnum' should be set to.  If nb_sectors goes
1343  * beyond the end of the disk image it will be clamped.
1344  *
1345  * (Based on raw_co_get_block_status() from file-posix.c.)
1346  */
1347 static int64_t coroutine_fn qemu_gluster_co_get_block_status(
1348         BlockDriverState *bs, int64_t sector_num, int nb_sectors, int *pnum,
1349         BlockDriverState **file)
1350 {
1351     BDRVGlusterState *s = bs->opaque;
1352     off_t start, data = 0, hole = 0;
1353     int64_t total_size;
1354     int ret = -EINVAL;
1355 
1356     if (!s->fd) {
1357         return ret;
1358     }
1359 
1360     start = sector_num * BDRV_SECTOR_SIZE;
1361     total_size = bdrv_getlength(bs);
1362     if (total_size < 0) {
1363         return total_size;
1364     } else if (start >= total_size) {
1365         *pnum = 0;
1366         return 0;
1367     } else if (start + nb_sectors * BDRV_SECTOR_SIZE > total_size) {
1368         nb_sectors = DIV_ROUND_UP(total_size - start, BDRV_SECTOR_SIZE);
1369     }
1370 
1371     ret = find_allocation(bs, start, &data, &hole);
1372     if (ret == -ENXIO) {
1373         /* Trailing hole */
1374         *pnum = nb_sectors;
1375         ret = BDRV_BLOCK_ZERO;
1376     } else if (ret < 0) {
1377         /* No info available, so pretend there are no holes */
1378         *pnum = nb_sectors;
1379         ret = BDRV_BLOCK_DATA;
1380     } else if (data == start) {
1381         /* On a data extent, compute sectors to the end of the extent,
1382          * possibly including a partial sector at EOF. */
1383         *pnum = MIN(nb_sectors, DIV_ROUND_UP(hole - start, BDRV_SECTOR_SIZE));
1384         ret = BDRV_BLOCK_DATA;
1385     } else {
1386         /* On a hole, compute sectors to the beginning of the next extent.  */
1387         assert(hole == start);
1388         *pnum = MIN(nb_sectors, (data - start) / BDRV_SECTOR_SIZE);
1389         ret = BDRV_BLOCK_ZERO;
1390     }
1391 
1392     *file = bs;
1393 
1394     return ret | BDRV_BLOCK_OFFSET_VALID | start;
1395 }
1396 
1397 
1398 static BlockDriver bdrv_gluster = {
1399     .format_name                  = "gluster",
1400     .protocol_name                = "gluster",
1401     .instance_size                = sizeof(BDRVGlusterState),
1402     .bdrv_needs_filename          = false,
1403     .bdrv_file_open               = qemu_gluster_open,
1404     .bdrv_reopen_prepare          = qemu_gluster_reopen_prepare,
1405     .bdrv_reopen_commit           = qemu_gluster_reopen_commit,
1406     .bdrv_reopen_abort            = qemu_gluster_reopen_abort,
1407     .bdrv_close                   = qemu_gluster_close,
1408     .bdrv_create                  = qemu_gluster_create,
1409     .bdrv_getlength               = qemu_gluster_getlength,
1410     .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size,
1411     .bdrv_truncate                = qemu_gluster_truncate,
1412     .bdrv_co_readv                = qemu_gluster_co_readv,
1413     .bdrv_co_writev               = qemu_gluster_co_writev,
1414     .bdrv_co_flush_to_disk        = qemu_gluster_co_flush_to_disk,
1415     .bdrv_has_zero_init           = qemu_gluster_has_zero_init,
1416 #ifdef CONFIG_GLUSTERFS_DISCARD
1417     .bdrv_co_pdiscard             = qemu_gluster_co_pdiscard,
1418 #endif
1419 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1420     .bdrv_co_pwrite_zeroes        = qemu_gluster_co_pwrite_zeroes,
1421 #endif
1422     .bdrv_co_get_block_status     = qemu_gluster_co_get_block_status,
1423     .create_opts                  = &qemu_gluster_create_opts,
1424 };
1425 
1426 static BlockDriver bdrv_gluster_tcp = {
1427     .format_name                  = "gluster",
1428     .protocol_name                = "gluster+tcp",
1429     .instance_size                = sizeof(BDRVGlusterState),
1430     .bdrv_needs_filename          = false,
1431     .bdrv_file_open               = qemu_gluster_open,
1432     .bdrv_reopen_prepare          = qemu_gluster_reopen_prepare,
1433     .bdrv_reopen_commit           = qemu_gluster_reopen_commit,
1434     .bdrv_reopen_abort            = qemu_gluster_reopen_abort,
1435     .bdrv_close                   = qemu_gluster_close,
1436     .bdrv_create                  = qemu_gluster_create,
1437     .bdrv_getlength               = qemu_gluster_getlength,
1438     .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size,
1439     .bdrv_truncate                = qemu_gluster_truncate,
1440     .bdrv_co_readv                = qemu_gluster_co_readv,
1441     .bdrv_co_writev               = qemu_gluster_co_writev,
1442     .bdrv_co_flush_to_disk        = qemu_gluster_co_flush_to_disk,
1443     .bdrv_has_zero_init           = qemu_gluster_has_zero_init,
1444 #ifdef CONFIG_GLUSTERFS_DISCARD
1445     .bdrv_co_pdiscard             = qemu_gluster_co_pdiscard,
1446 #endif
1447 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1448     .bdrv_co_pwrite_zeroes        = qemu_gluster_co_pwrite_zeroes,
1449 #endif
1450     .bdrv_co_get_block_status     = qemu_gluster_co_get_block_status,
1451     .create_opts                  = &qemu_gluster_create_opts,
1452 };
1453 
1454 static BlockDriver bdrv_gluster_unix = {
1455     .format_name                  = "gluster",
1456     .protocol_name                = "gluster+unix",
1457     .instance_size                = sizeof(BDRVGlusterState),
1458     .bdrv_needs_filename          = true,
1459     .bdrv_file_open               = qemu_gluster_open,
1460     .bdrv_reopen_prepare          = qemu_gluster_reopen_prepare,
1461     .bdrv_reopen_commit           = qemu_gluster_reopen_commit,
1462     .bdrv_reopen_abort            = qemu_gluster_reopen_abort,
1463     .bdrv_close                   = qemu_gluster_close,
1464     .bdrv_create                  = qemu_gluster_create,
1465     .bdrv_getlength               = qemu_gluster_getlength,
1466     .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size,
1467     .bdrv_truncate                = qemu_gluster_truncate,
1468     .bdrv_co_readv                = qemu_gluster_co_readv,
1469     .bdrv_co_writev               = qemu_gluster_co_writev,
1470     .bdrv_co_flush_to_disk        = qemu_gluster_co_flush_to_disk,
1471     .bdrv_has_zero_init           = qemu_gluster_has_zero_init,
1472 #ifdef CONFIG_GLUSTERFS_DISCARD
1473     .bdrv_co_pdiscard             = qemu_gluster_co_pdiscard,
1474 #endif
1475 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1476     .bdrv_co_pwrite_zeroes        = qemu_gluster_co_pwrite_zeroes,
1477 #endif
1478     .bdrv_co_get_block_status     = qemu_gluster_co_get_block_status,
1479     .create_opts                  = &qemu_gluster_create_opts,
1480 };
1481 
1482 /* rdma is deprecated (actually never supported for volfile fetch).
1483  * Let's maintain it for the protocol compatibility, to make sure things
1484  * won't break immediately. For now, gluster+rdma will fall back to gluster+tcp
1485  * protocol with a warning.
1486  * TODO: remove gluster+rdma interface support
1487  */
1488 static BlockDriver bdrv_gluster_rdma = {
1489     .format_name                  = "gluster",
1490     .protocol_name                = "gluster+rdma",
1491     .instance_size                = sizeof(BDRVGlusterState),
1492     .bdrv_needs_filename          = true,
1493     .bdrv_file_open               = qemu_gluster_open,
1494     .bdrv_reopen_prepare          = qemu_gluster_reopen_prepare,
1495     .bdrv_reopen_commit           = qemu_gluster_reopen_commit,
1496     .bdrv_reopen_abort            = qemu_gluster_reopen_abort,
1497     .bdrv_close                   = qemu_gluster_close,
1498     .bdrv_create                  = qemu_gluster_create,
1499     .bdrv_getlength               = qemu_gluster_getlength,
1500     .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size,
1501     .bdrv_truncate                = qemu_gluster_truncate,
1502     .bdrv_co_readv                = qemu_gluster_co_readv,
1503     .bdrv_co_writev               = qemu_gluster_co_writev,
1504     .bdrv_co_flush_to_disk        = qemu_gluster_co_flush_to_disk,
1505     .bdrv_has_zero_init           = qemu_gluster_has_zero_init,
1506 #ifdef CONFIG_GLUSTERFS_DISCARD
1507     .bdrv_co_pdiscard             = qemu_gluster_co_pdiscard,
1508 #endif
1509 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1510     .bdrv_co_pwrite_zeroes        = qemu_gluster_co_pwrite_zeroes,
1511 #endif
1512     .bdrv_co_get_block_status     = qemu_gluster_co_get_block_status,
1513     .create_opts                  = &qemu_gluster_create_opts,
1514 };
1515 
1516 static void bdrv_gluster_init(void)
1517 {
1518     bdrv_register(&bdrv_gluster_rdma);
1519     bdrv_register(&bdrv_gluster_unix);
1520     bdrv_register(&bdrv_gluster_tcp);
1521     bdrv_register(&bdrv_gluster);
1522 }
1523 
1524 block_init(bdrv_gluster_init);
1525