xref: /openbmc/qemu/block/gluster.c (revision 3f53bc61)
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     SocketAddressFlat *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(SocketAddressFlatList, 1);
336     gconf->server->value = gsconf = g_new0(SocketAddressFlat, 1);
337 
338     /* transport */
339     if (!uri->scheme || !strcmp(uri->scheme, "gluster")) {
340         gsconf->type = SOCKET_ADDRESS_FLAT_TYPE_INET;
341     } else if (!strcmp(uri->scheme, "gluster+tcp")) {
342         gsconf->type = SOCKET_ADDRESS_FLAT_TYPE_INET;
343     } else if (!strcmp(uri->scheme, "gluster+unix")) {
344         gsconf->type = SOCKET_ADDRESS_FLAT_TYPE_UNIX;
345         is_unix = true;
346     } else if (!strcmp(uri->scheme, "gluster+rdma")) {
347         gsconf->type = SOCKET_ADDRESS_FLAT_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     SocketAddressFlatList *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         if (server->value->type  == SOCKET_ADDRESS_FLAT_TYPE_UNIX) {
416             ret = glfs_set_volfile_server(glfs, "unix",
417                                    server->value->u.q_unix.path, 0);
418         } else {
419             if (parse_uint_full(server->value->u.inet.port, &port, 10) < 0 ||
420                 port > 65535) {
421                 error_setg(errp, "'%s' is not a valid port number",
422                            server->value->u.inet.port);
423                 errno = EINVAL;
424                 goto out;
425             }
426             ret = glfs_set_volfile_server(glfs, "tcp",
427                                    server->value->u.inet.host,
428                                    (int)port);
429         }
430 
431         if (ret < 0) {
432             goto out;
433         }
434     }
435 
436     ret = glfs_set_logging(glfs, gconf->logfile, gconf->debug);
437     if (ret < 0) {
438         goto out;
439     }
440 
441     ret = glfs_init(glfs);
442     if (ret) {
443         error_setg(errp, "Gluster connection for volume %s, path %s failed"
444                          " to connect", gconf->volume, gconf->path);
445         for (server = gconf->server; server; server = server->next) {
446             if (server->value->type  == SOCKET_ADDRESS_FLAT_TYPE_UNIX) {
447                 error_append_hint(errp, "hint: failed on socket %s ",
448                                   server->value->u.q_unix.path);
449             } else {
450                 error_append_hint(errp, "hint: failed on host %s and port %s ",
451                                   server->value->u.inet.host,
452                                   server->value->u.inet.port);
453             }
454         }
455 
456         error_append_hint(errp, "Please refer to gluster logs for more info\n");
457 
458         /* glfs_init sometimes doesn't set errno although docs suggest that */
459         if (errno == 0) {
460             errno = EINVAL;
461         }
462 
463         goto out;
464     }
465     return glfs;
466 
467 out:
468     if (glfs) {
469         old_errno = errno;
470         glfs_clear_preopened(glfs);
471         errno = old_errno;
472     }
473     return NULL;
474 }
475 
476 /*
477  * Convert the json formatted command line into qapi.
478 */
479 static int qemu_gluster_parse_json(BlockdevOptionsGluster *gconf,
480                                   QDict *options, Error **errp)
481 {
482     QemuOpts *opts;
483     SocketAddressFlat *gsconf = NULL;
484     SocketAddressFlatList *curr = NULL;
485     QDict *backing_options = NULL;
486     Error *local_err = NULL;
487     char *str = NULL;
488     const char *ptr;
489     size_t num_servers;
490     int i;
491 
492     /* create opts info from runtime_json_opts list */
493     opts = qemu_opts_create(&runtime_json_opts, NULL, 0, &error_abort);
494     qemu_opts_absorb_qdict(opts, options, &local_err);
495     if (local_err) {
496         goto out;
497     }
498 
499     num_servers = qdict_array_entries(options, GLUSTER_OPT_SERVER_PATTERN);
500     if (num_servers < 1) {
501         error_setg(&local_err, QERR_MISSING_PARAMETER, "server");
502         goto out;
503     }
504 
505     ptr = qemu_opt_get(opts, GLUSTER_OPT_VOLUME);
506     if (!ptr) {
507         error_setg(&local_err, QERR_MISSING_PARAMETER, GLUSTER_OPT_VOLUME);
508         goto out;
509     }
510     gconf->volume = g_strdup(ptr);
511 
512     ptr = qemu_opt_get(opts, GLUSTER_OPT_PATH);
513     if (!ptr) {
514         error_setg(&local_err, QERR_MISSING_PARAMETER, GLUSTER_OPT_PATH);
515         goto out;
516     }
517     gconf->path = g_strdup(ptr);
518     qemu_opts_del(opts);
519 
520     for (i = 0; i < num_servers; i++) {
521         str = g_strdup_printf(GLUSTER_OPT_SERVER_PATTERN"%d.", i);
522         qdict_extract_subqdict(options, &backing_options, str);
523 
524         /* create opts info from runtime_type_opts list */
525         opts = qemu_opts_create(&runtime_type_opts, NULL, 0, &error_abort);
526         qemu_opts_absorb_qdict(opts, backing_options, &local_err);
527         if (local_err) {
528             goto out;
529         }
530 
531         ptr = qemu_opt_get(opts, GLUSTER_OPT_TYPE);
532         if (!ptr) {
533             error_setg(&local_err, QERR_MISSING_PARAMETER, GLUSTER_OPT_TYPE);
534             error_append_hint(&local_err, GERR_INDEX_HINT, i);
535             goto out;
536 
537         }
538         gsconf = g_new0(SocketAddressFlat, 1);
539         if (!strcmp(ptr, "tcp")) {
540             ptr = "inet";       /* accept legacy "tcp" */
541         }
542         gsconf->type = qapi_enum_parse(SocketAddressFlatType_lookup, ptr,
543                                        SOCKET_ADDRESS_FLAT_TYPE__MAX, -1,
544                                        &local_err);
545         if (local_err) {
546             error_append_hint(&local_err,
547                               "Parameter '%s' may be 'inet' or 'unix'\n",
548                               GLUSTER_OPT_TYPE);
549             error_append_hint(&local_err, GERR_INDEX_HINT, i);
550             goto out;
551         }
552         qemu_opts_del(opts);
553 
554         if (gsconf->type == SOCKET_ADDRESS_FLAT_TYPE_INET) {
555             /* create opts info from runtime_inet_opts list */
556             opts = qemu_opts_create(&runtime_inet_opts, NULL, 0, &error_abort);
557             qemu_opts_absorb_qdict(opts, backing_options, &local_err);
558             if (local_err) {
559                 goto out;
560             }
561 
562             ptr = qemu_opt_get(opts, GLUSTER_OPT_HOST);
563             if (!ptr) {
564                 error_setg(&local_err, QERR_MISSING_PARAMETER,
565                            GLUSTER_OPT_HOST);
566                 error_append_hint(&local_err, GERR_INDEX_HINT, i);
567                 goto out;
568             }
569             gsconf->u.inet.host = g_strdup(ptr);
570             ptr = qemu_opt_get(opts, GLUSTER_OPT_PORT);
571             if (!ptr) {
572                 error_setg(&local_err, QERR_MISSING_PARAMETER,
573                            GLUSTER_OPT_PORT);
574                 error_append_hint(&local_err, GERR_INDEX_HINT, i);
575                 goto out;
576             }
577             gsconf->u.inet.port = g_strdup(ptr);
578 
579             /* defend for unsupported fields in InetSocketAddress,
580              * i.e. @ipv4, @ipv6  and @to
581              */
582             ptr = qemu_opt_get(opts, GLUSTER_OPT_TO);
583             if (ptr) {
584                 gsconf->u.inet.has_to = true;
585             }
586             ptr = qemu_opt_get(opts, GLUSTER_OPT_IPV4);
587             if (ptr) {
588                 gsconf->u.inet.has_ipv4 = true;
589             }
590             ptr = qemu_opt_get(opts, GLUSTER_OPT_IPV6);
591             if (ptr) {
592                 gsconf->u.inet.has_ipv6 = true;
593             }
594             if (gsconf->u.inet.has_to) {
595                 error_setg(&local_err, "Parameter 'to' not supported");
596                 goto out;
597             }
598             if (gsconf->u.inet.has_ipv4 || gsconf->u.inet.has_ipv6) {
599                 error_setg(&local_err, "Parameters 'ipv4/ipv6' not supported");
600                 goto out;
601             }
602             qemu_opts_del(opts);
603         } else {
604             /* create opts info from runtime_unix_opts list */
605             opts = qemu_opts_create(&runtime_unix_opts, NULL, 0, &error_abort);
606             qemu_opts_absorb_qdict(opts, backing_options, &local_err);
607             if (local_err) {
608                 goto out;
609             }
610 
611             ptr = qemu_opt_get(opts, GLUSTER_OPT_SOCKET);
612             if (!ptr) {
613                 error_setg(&local_err, QERR_MISSING_PARAMETER,
614                            GLUSTER_OPT_SOCKET);
615                 error_append_hint(&local_err, GERR_INDEX_HINT, i);
616                 goto out;
617             }
618             gsconf->u.q_unix.path = g_strdup(ptr);
619             qemu_opts_del(opts);
620         }
621 
622         if (gconf->server == NULL) {
623             gconf->server = g_new0(SocketAddressFlatList, 1);
624             gconf->server->value = gsconf;
625             curr = gconf->server;
626         } else {
627             curr->next = g_new0(SocketAddressFlatList, 1);
628             curr->next->value = gsconf;
629             curr = curr->next;
630         }
631         gsconf = NULL;
632 
633         QDECREF(backing_options);
634         backing_options = NULL;
635         g_free(str);
636         str = NULL;
637     }
638 
639     return 0;
640 
641 out:
642     error_propagate(errp, local_err);
643     qapi_free_SocketAddressFlat(gsconf);
644     qemu_opts_del(opts);
645     g_free(str);
646     QDECREF(backing_options);
647     errno = EINVAL;
648     return -errno;
649 }
650 
651 static struct glfs *qemu_gluster_init(BlockdevOptionsGluster *gconf,
652                                       const char *filename,
653                                       QDict *options, Error **errp)
654 {
655     int ret;
656     if (filename) {
657         ret = qemu_gluster_parse_uri(gconf, filename);
658         if (ret < 0) {
659             error_setg(errp, "invalid URI");
660             error_append_hint(errp, "Usage: file=gluster[+transport]://"
661                                     "[host[:port]]volume/path[?socket=...]"
662                                     "[,file.debug=N]"
663                                     "[,file.logfile=/path/filename.log]\n");
664             errno = -ret;
665             return NULL;
666         }
667     } else {
668         ret = qemu_gluster_parse_json(gconf, options, errp);
669         if (ret < 0) {
670             error_append_hint(errp, "Usage: "
671                              "-drive driver=qcow2,file.driver=gluster,"
672                              "file.volume=testvol,file.path=/path/a.qcow2"
673                              "[,file.debug=9]"
674                              "[,file.logfile=/path/filename.log],"
675                              "file.server.0.type=inet,"
676                              "file.server.0.host=1.2.3.4,"
677                              "file.server.0.port=24007,"
678                              "file.server.1.transport=unix,"
679                              "file.server.1.socket=/var/run/glusterd.socket ..."
680                              "\n");
681             errno = -ret;
682             return NULL;
683         }
684 
685     }
686 
687     return qemu_gluster_glfs_init(gconf, errp);
688 }
689 
690 /*
691  * AIO callback routine called from GlusterFS thread.
692  */
693 static void gluster_finish_aiocb(struct glfs_fd *fd, ssize_t ret, void *arg)
694 {
695     GlusterAIOCB *acb = (GlusterAIOCB *)arg;
696 
697     if (!ret || ret == acb->size) {
698         acb->ret = 0; /* Success */
699     } else if (ret < 0) {
700         acb->ret = -errno; /* Read/Write failed */
701     } else {
702         acb->ret = -EIO; /* Partial read/write - fail it */
703     }
704 
705     aio_co_schedule(acb->aio_context, acb->coroutine);
706 }
707 
708 static void qemu_gluster_parse_flags(int bdrv_flags, int *open_flags)
709 {
710     assert(open_flags != NULL);
711 
712     *open_flags |= O_BINARY;
713 
714     if (bdrv_flags & BDRV_O_RDWR) {
715         *open_flags |= O_RDWR;
716     } else {
717         *open_flags |= O_RDONLY;
718     }
719 
720     if ((bdrv_flags & BDRV_O_NOCACHE)) {
721         *open_flags |= O_DIRECT;
722     }
723 }
724 
725 /*
726  * Do SEEK_DATA/HOLE to detect if it is functional. Older broken versions of
727  * gfapi incorrectly return the current offset when SEEK_DATA/HOLE is used.
728  * - Corrected versions return -1 and set errno to EINVAL.
729  * - Versions that support SEEK_DATA/HOLE correctly, will return -1 and set
730  *   errno to ENXIO when SEEK_DATA is called with a position of EOF.
731  */
732 static bool qemu_gluster_test_seek(struct glfs_fd *fd)
733 {
734     off_t ret = 0;
735 
736 #if defined SEEK_HOLE && defined SEEK_DATA
737     off_t eof;
738 
739     eof = glfs_lseek(fd, 0, SEEK_END);
740     if (eof < 0) {
741         /* this should never occur */
742         return false;
743     }
744 
745     /* this should always fail with ENXIO if SEEK_DATA is supported */
746     ret = glfs_lseek(fd, eof, SEEK_DATA);
747 #endif
748 
749     return (ret < 0) && (errno == ENXIO);
750 }
751 
752 static int qemu_gluster_open(BlockDriverState *bs,  QDict *options,
753                              int bdrv_flags, Error **errp)
754 {
755     BDRVGlusterState *s = bs->opaque;
756     int open_flags = 0;
757     int ret = 0;
758     BlockdevOptionsGluster *gconf = NULL;
759     QemuOpts *opts;
760     Error *local_err = NULL;
761     const char *filename, *logfile;
762 
763     opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
764     qemu_opts_absorb_qdict(opts, options, &local_err);
765     if (local_err) {
766         error_propagate(errp, local_err);
767         ret = -EINVAL;
768         goto out;
769     }
770 
771     filename = qemu_opt_get(opts, GLUSTER_OPT_FILENAME);
772 
773     s->debug = qemu_opt_get_number(opts, GLUSTER_OPT_DEBUG,
774                                    GLUSTER_DEBUG_DEFAULT);
775     if (s->debug < 0) {
776         s->debug = 0;
777     } else if (s->debug > GLUSTER_DEBUG_MAX) {
778         s->debug = GLUSTER_DEBUG_MAX;
779     }
780 
781     gconf = g_new0(BlockdevOptionsGluster, 1);
782     gconf->debug = s->debug;
783     gconf->has_debug = true;
784 
785     logfile = qemu_opt_get(opts, GLUSTER_OPT_LOGFILE);
786     s->logfile = g_strdup(logfile ? logfile : GLUSTER_LOGFILE_DEFAULT);
787 
788     gconf->logfile = g_strdup(s->logfile);
789     gconf->has_logfile = true;
790 
791     s->glfs = qemu_gluster_init(gconf, filename, options, errp);
792     if (!s->glfs) {
793         ret = -errno;
794         goto out;
795     }
796 
797 #ifdef CONFIG_GLUSTERFS_XLATOR_OPT
798     /* Without this, if fsync fails for a recoverable reason (for instance,
799      * ENOSPC), gluster will dump its cache, preventing retries.  This means
800      * almost certain data loss.  Not all gluster versions support the
801      * 'resync-failed-syncs-after-fsync' key value, but there is no way to
802      * discover during runtime if it is supported (this api returns success for
803      * unknown key/value pairs) */
804     ret = glfs_set_xlator_option(s->glfs, "*-write-behind",
805                                           "resync-failed-syncs-after-fsync",
806                                           "on");
807     if (ret < 0) {
808         error_setg_errno(errp, errno, "Unable to set xlator key/value pair");
809         ret = -errno;
810         goto out;
811     }
812 #endif
813 
814     qemu_gluster_parse_flags(bdrv_flags, &open_flags);
815 
816     s->fd = glfs_open(s->glfs, gconf->path, open_flags);
817     if (!s->fd) {
818         ret = -errno;
819     }
820 
821     s->supports_seek_data = qemu_gluster_test_seek(s->fd);
822 
823 out:
824     qemu_opts_del(opts);
825     qapi_free_BlockdevOptionsGluster(gconf);
826     if (!ret) {
827         return ret;
828     }
829     g_free(s->logfile);
830     if (s->fd) {
831         glfs_close(s->fd);
832     }
833 
834     glfs_clear_preopened(s->glfs);
835 
836     return ret;
837 }
838 
839 static int qemu_gluster_reopen_prepare(BDRVReopenState *state,
840                                        BlockReopenQueue *queue, Error **errp)
841 {
842     int ret = 0;
843     BDRVGlusterState *s;
844     BDRVGlusterReopenState *reop_s;
845     BlockdevOptionsGluster *gconf;
846     int open_flags = 0;
847 
848     assert(state != NULL);
849     assert(state->bs != NULL);
850 
851     s = state->bs->opaque;
852 
853     state->opaque = g_new0(BDRVGlusterReopenState, 1);
854     reop_s = state->opaque;
855 
856     qemu_gluster_parse_flags(state->flags, &open_flags);
857 
858     gconf = g_new0(BlockdevOptionsGluster, 1);
859     gconf->debug = s->debug;
860     gconf->has_debug = true;
861     gconf->logfile = g_strdup(s->logfile);
862     gconf->has_logfile = true;
863     reop_s->glfs = qemu_gluster_init(gconf, state->bs->filename, NULL, errp);
864     if (reop_s->glfs == NULL) {
865         ret = -errno;
866         goto exit;
867     }
868 
869 #ifdef CONFIG_GLUSTERFS_XLATOR_OPT
870     ret = glfs_set_xlator_option(reop_s->glfs, "*-write-behind",
871                                  "resync-failed-syncs-after-fsync", "on");
872     if (ret < 0) {
873         error_setg_errno(errp, errno, "Unable to set xlator key/value pair");
874         ret = -errno;
875         goto exit;
876     }
877 #endif
878 
879     reop_s->fd = glfs_open(reop_s->glfs, gconf->path, open_flags);
880     if (reop_s->fd == NULL) {
881         /* reops->glfs will be cleaned up in _abort */
882         ret = -errno;
883         goto exit;
884     }
885 
886 exit:
887     /* state->opaque will be freed in either the _abort or _commit */
888     qapi_free_BlockdevOptionsGluster(gconf);
889     return ret;
890 }
891 
892 static void qemu_gluster_reopen_commit(BDRVReopenState *state)
893 {
894     BDRVGlusterReopenState *reop_s = state->opaque;
895     BDRVGlusterState *s = state->bs->opaque;
896 
897 
898     /* close the old */
899     if (s->fd) {
900         glfs_close(s->fd);
901     }
902 
903     glfs_clear_preopened(s->glfs);
904 
905     /* use the newly opened image / connection */
906     s->fd         = reop_s->fd;
907     s->glfs       = reop_s->glfs;
908 
909     g_free(state->opaque);
910     state->opaque = NULL;
911 
912     return;
913 }
914 
915 
916 static void qemu_gluster_reopen_abort(BDRVReopenState *state)
917 {
918     BDRVGlusterReopenState *reop_s = state->opaque;
919 
920     if (reop_s == NULL) {
921         return;
922     }
923 
924     if (reop_s->fd) {
925         glfs_close(reop_s->fd);
926     }
927 
928     glfs_clear_preopened(reop_s->glfs);
929 
930     g_free(state->opaque);
931     state->opaque = NULL;
932 
933     return;
934 }
935 
936 #ifdef CONFIG_GLUSTERFS_ZEROFILL
937 static coroutine_fn int qemu_gluster_co_pwrite_zeroes(BlockDriverState *bs,
938                                                       int64_t offset,
939                                                       int size,
940                                                       BdrvRequestFlags flags)
941 {
942     int ret;
943     GlusterAIOCB acb;
944     BDRVGlusterState *s = bs->opaque;
945 
946     acb.size = size;
947     acb.ret = 0;
948     acb.coroutine = qemu_coroutine_self();
949     acb.aio_context = bdrv_get_aio_context(bs);
950 
951     ret = glfs_zerofill_async(s->fd, offset, size, gluster_finish_aiocb, &acb);
952     if (ret < 0) {
953         return -errno;
954     }
955 
956     qemu_coroutine_yield();
957     return acb.ret;
958 }
959 
960 static inline bool gluster_supports_zerofill(void)
961 {
962     return 1;
963 }
964 
965 static inline int qemu_gluster_zerofill(struct glfs_fd *fd, int64_t offset,
966                                         int64_t size)
967 {
968     return glfs_zerofill(fd, offset, size);
969 }
970 
971 #else
972 static inline bool gluster_supports_zerofill(void)
973 {
974     return 0;
975 }
976 
977 static inline int qemu_gluster_zerofill(struct glfs_fd *fd, int64_t offset,
978                                         int64_t size)
979 {
980     return 0;
981 }
982 #endif
983 
984 static int qemu_gluster_create(const char *filename,
985                                QemuOpts *opts, Error **errp)
986 {
987     BlockdevOptionsGluster *gconf;
988     struct glfs *glfs;
989     struct glfs_fd *fd;
990     int ret = 0;
991     int prealloc = 0;
992     int64_t total_size = 0;
993     char *tmp = NULL;
994 
995     gconf = g_new0(BlockdevOptionsGluster, 1);
996     gconf->debug = qemu_opt_get_number_del(opts, GLUSTER_OPT_DEBUG,
997                                            GLUSTER_DEBUG_DEFAULT);
998     if (gconf->debug < 0) {
999         gconf->debug = 0;
1000     } else if (gconf->debug > GLUSTER_DEBUG_MAX) {
1001         gconf->debug = GLUSTER_DEBUG_MAX;
1002     }
1003     gconf->has_debug = true;
1004 
1005     gconf->logfile = qemu_opt_get_del(opts, GLUSTER_OPT_LOGFILE);
1006     if (!gconf->logfile) {
1007         gconf->logfile = g_strdup(GLUSTER_LOGFILE_DEFAULT);
1008     }
1009     gconf->has_logfile = true;
1010 
1011     glfs = qemu_gluster_init(gconf, filename, NULL, errp);
1012     if (!glfs) {
1013         ret = -errno;
1014         goto out;
1015     }
1016 
1017     total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
1018                           BDRV_SECTOR_SIZE);
1019 
1020     tmp = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
1021     if (!tmp || !strcmp(tmp, "off")) {
1022         prealloc = 0;
1023     } else if (!strcmp(tmp, "full") && gluster_supports_zerofill()) {
1024         prealloc = 1;
1025     } else {
1026         error_setg(errp, "Invalid preallocation mode: '%s'"
1027                          " or GlusterFS doesn't support zerofill API", tmp);
1028         ret = -EINVAL;
1029         goto out;
1030     }
1031 
1032     fd = glfs_creat(glfs, gconf->path,
1033                     O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, S_IRUSR | S_IWUSR);
1034     if (!fd) {
1035         ret = -errno;
1036     } else {
1037         if (!glfs_ftruncate(fd, total_size)) {
1038             if (prealloc && qemu_gluster_zerofill(fd, 0, total_size)) {
1039                 ret = -errno;
1040             }
1041         } else {
1042             ret = -errno;
1043         }
1044 
1045         if (glfs_close(fd) != 0) {
1046             ret = -errno;
1047         }
1048     }
1049 out:
1050     g_free(tmp);
1051     qapi_free_BlockdevOptionsGluster(gconf);
1052     glfs_clear_preopened(glfs);
1053     return ret;
1054 }
1055 
1056 static coroutine_fn int qemu_gluster_co_rw(BlockDriverState *bs,
1057                                            int64_t sector_num, int nb_sectors,
1058                                            QEMUIOVector *qiov, int write)
1059 {
1060     int ret;
1061     GlusterAIOCB acb;
1062     BDRVGlusterState *s = bs->opaque;
1063     size_t size = nb_sectors * BDRV_SECTOR_SIZE;
1064     off_t offset = sector_num * BDRV_SECTOR_SIZE;
1065 
1066     acb.size = size;
1067     acb.ret = 0;
1068     acb.coroutine = qemu_coroutine_self();
1069     acb.aio_context = bdrv_get_aio_context(bs);
1070 
1071     if (write) {
1072         ret = glfs_pwritev_async(s->fd, qiov->iov, qiov->niov, offset, 0,
1073                                  gluster_finish_aiocb, &acb);
1074     } else {
1075         ret = glfs_preadv_async(s->fd, qiov->iov, qiov->niov, offset, 0,
1076                                 gluster_finish_aiocb, &acb);
1077     }
1078 
1079     if (ret < 0) {
1080         return -errno;
1081     }
1082 
1083     qemu_coroutine_yield();
1084     return acb.ret;
1085 }
1086 
1087 static int qemu_gluster_truncate(BlockDriverState *bs, int64_t offset)
1088 {
1089     int ret;
1090     BDRVGlusterState *s = bs->opaque;
1091 
1092     ret = glfs_ftruncate(s->fd, offset);
1093     if (ret < 0) {
1094         return -errno;
1095     }
1096 
1097     return 0;
1098 }
1099 
1100 static coroutine_fn int qemu_gluster_co_readv(BlockDriverState *bs,
1101                                               int64_t sector_num,
1102                                               int nb_sectors,
1103                                               QEMUIOVector *qiov)
1104 {
1105     return qemu_gluster_co_rw(bs, sector_num, nb_sectors, qiov, 0);
1106 }
1107 
1108 static coroutine_fn int qemu_gluster_co_writev(BlockDriverState *bs,
1109                                                int64_t sector_num,
1110                                                int nb_sectors,
1111                                                QEMUIOVector *qiov)
1112 {
1113     return qemu_gluster_co_rw(bs, sector_num, nb_sectors, qiov, 1);
1114 }
1115 
1116 static void qemu_gluster_close(BlockDriverState *bs)
1117 {
1118     BDRVGlusterState *s = bs->opaque;
1119 
1120     g_free(s->logfile);
1121     if (s->fd) {
1122         glfs_close(s->fd);
1123         s->fd = NULL;
1124     }
1125     glfs_clear_preopened(s->glfs);
1126 }
1127 
1128 static coroutine_fn int qemu_gluster_co_flush_to_disk(BlockDriverState *bs)
1129 {
1130     int ret;
1131     GlusterAIOCB acb;
1132     BDRVGlusterState *s = bs->opaque;
1133 
1134     acb.size = 0;
1135     acb.ret = 0;
1136     acb.coroutine = qemu_coroutine_self();
1137     acb.aio_context = bdrv_get_aio_context(bs);
1138 
1139     ret = glfs_fsync_async(s->fd, gluster_finish_aiocb, &acb);
1140     if (ret < 0) {
1141         ret = -errno;
1142         goto error;
1143     }
1144 
1145     qemu_coroutine_yield();
1146     if (acb.ret < 0) {
1147         ret = acb.ret;
1148         goto error;
1149     }
1150 
1151     return acb.ret;
1152 
1153 error:
1154     /* Some versions of Gluster (3.5.6 -> 3.5.8?) will not retain its cache
1155      * after a fsync failure, so we have no way of allowing the guest to safely
1156      * continue.  Gluster versions prior to 3.5.6 don't retain the cache
1157      * either, but will invalidate the fd on error, so this is again our only
1158      * option.
1159      *
1160      * The 'resync-failed-syncs-after-fsync' xlator option for the
1161      * write-behind cache will cause later gluster versions to retain its
1162      * cache after error, so long as the fd remains open.  However, we
1163      * currently have no way of knowing if this option is supported.
1164      *
1165      * TODO: Once gluster provides a way for us to determine if the option
1166      * is supported, bypass the closure and setting drv to NULL.  */
1167     qemu_gluster_close(bs);
1168     bs->drv = NULL;
1169     return ret;
1170 }
1171 
1172 #ifdef CONFIG_GLUSTERFS_DISCARD
1173 static coroutine_fn int qemu_gluster_co_pdiscard(BlockDriverState *bs,
1174                                                  int64_t offset, int size)
1175 {
1176     int ret;
1177     GlusterAIOCB acb;
1178     BDRVGlusterState *s = bs->opaque;
1179 
1180     acb.size = 0;
1181     acb.ret = 0;
1182     acb.coroutine = qemu_coroutine_self();
1183     acb.aio_context = bdrv_get_aio_context(bs);
1184 
1185     ret = glfs_discard_async(s->fd, offset, size, gluster_finish_aiocb, &acb);
1186     if (ret < 0) {
1187         return -errno;
1188     }
1189 
1190     qemu_coroutine_yield();
1191     return acb.ret;
1192 }
1193 #endif
1194 
1195 static int64_t qemu_gluster_getlength(BlockDriverState *bs)
1196 {
1197     BDRVGlusterState *s = bs->opaque;
1198     int64_t ret;
1199 
1200     ret = glfs_lseek(s->fd, 0, SEEK_END);
1201     if (ret < 0) {
1202         return -errno;
1203     } else {
1204         return ret;
1205     }
1206 }
1207 
1208 static int64_t qemu_gluster_allocated_file_size(BlockDriverState *bs)
1209 {
1210     BDRVGlusterState *s = bs->opaque;
1211     struct stat st;
1212     int ret;
1213 
1214     ret = glfs_fstat(s->fd, &st);
1215     if (ret < 0) {
1216         return -errno;
1217     } else {
1218         return st.st_blocks * 512;
1219     }
1220 }
1221 
1222 static int qemu_gluster_has_zero_init(BlockDriverState *bs)
1223 {
1224     /* GlusterFS volume could be backed by a block device */
1225     return 0;
1226 }
1227 
1228 /*
1229  * Find allocation range in @bs around offset @start.
1230  * May change underlying file descriptor's file offset.
1231  * If @start is not in a hole, store @start in @data, and the
1232  * beginning of the next hole in @hole, and return 0.
1233  * If @start is in a non-trailing hole, store @start in @hole and the
1234  * beginning of the next non-hole in @data, and return 0.
1235  * If @start is in a trailing hole or beyond EOF, return -ENXIO.
1236  * If we can't find out, return a negative errno other than -ENXIO.
1237  *
1238  * (Shamefully copied from file-posix.c, only miniscule adaptions.)
1239  */
1240 static int find_allocation(BlockDriverState *bs, off_t start,
1241                            off_t *data, off_t *hole)
1242 {
1243     BDRVGlusterState *s = bs->opaque;
1244 
1245     if (!s->supports_seek_data) {
1246         goto exit;
1247     }
1248 
1249 #if defined SEEK_HOLE && defined SEEK_DATA
1250     off_t offs;
1251 
1252     /*
1253      * SEEK_DATA cases:
1254      * D1. offs == start: start is in data
1255      * D2. offs > start: start is in a hole, next data at offs
1256      * D3. offs < 0, errno = ENXIO: either start is in a trailing hole
1257      *                              or start is beyond EOF
1258      *     If the latter happens, the file has been truncated behind
1259      *     our back since we opened it.  All bets are off then.
1260      *     Treating like a trailing hole is simplest.
1261      * D4. offs < 0, errno != ENXIO: we learned nothing
1262      */
1263     offs = glfs_lseek(s->fd, start, SEEK_DATA);
1264     if (offs < 0) {
1265         return -errno;          /* D3 or D4 */
1266     }
1267     assert(offs >= start);
1268 
1269     if (offs > start) {
1270         /* D2: in hole, next data at offs */
1271         *hole = start;
1272         *data = offs;
1273         return 0;
1274     }
1275 
1276     /* D1: in data, end not yet known */
1277 
1278     /*
1279      * SEEK_HOLE cases:
1280      * H1. offs == start: start is in a hole
1281      *     If this happens here, a hole has been dug behind our back
1282      *     since the previous lseek().
1283      * H2. offs > start: either start is in data, next hole at offs,
1284      *                   or start is in trailing hole, EOF at offs
1285      *     Linux treats trailing holes like any other hole: offs ==
1286      *     start.  Solaris seeks to EOF instead: offs > start (blech).
1287      *     If that happens here, a hole has been dug behind our back
1288      *     since the previous lseek().
1289      * H3. offs < 0, errno = ENXIO: start is beyond EOF
1290      *     If this happens, the file has been truncated behind our
1291      *     back since we opened it.  Treat it like a trailing hole.
1292      * H4. offs < 0, errno != ENXIO: we learned nothing
1293      *     Pretend we know nothing at all, i.e. "forget" about D1.
1294      */
1295     offs = glfs_lseek(s->fd, start, SEEK_HOLE);
1296     if (offs < 0) {
1297         return -errno;          /* D1 and (H3 or H4) */
1298     }
1299     assert(offs >= start);
1300 
1301     if (offs > start) {
1302         /*
1303          * D1 and H2: either in data, next hole at offs, or it was in
1304          * data but is now in a trailing hole.  In the latter case,
1305          * all bets are off.  Treating it as if it there was data all
1306          * the way to EOF is safe, so simply do that.
1307          */
1308         *data = start;
1309         *hole = offs;
1310         return 0;
1311     }
1312 
1313     /* D1 and H1 */
1314     return -EBUSY;
1315 #endif
1316 
1317 exit:
1318     return -ENOTSUP;
1319 }
1320 
1321 /*
1322  * Returns the allocation status of the specified sectors.
1323  *
1324  * If 'sector_num' is beyond the end of the disk image the return value is 0
1325  * and 'pnum' is set to 0.
1326  *
1327  * 'pnum' is set to the number of sectors (including and immediately following
1328  * the specified sector) that are known to be in the same
1329  * allocated/unallocated state.
1330  *
1331  * 'nb_sectors' is the max value 'pnum' should be set to.  If nb_sectors goes
1332  * beyond the end of the disk image it will be clamped.
1333  *
1334  * (Based on raw_co_get_block_status() from file-posix.c.)
1335  */
1336 static int64_t coroutine_fn qemu_gluster_co_get_block_status(
1337         BlockDriverState *bs, int64_t sector_num, int nb_sectors, int *pnum,
1338         BlockDriverState **file)
1339 {
1340     BDRVGlusterState *s = bs->opaque;
1341     off_t start, data = 0, hole = 0;
1342     int64_t total_size;
1343     int ret = -EINVAL;
1344 
1345     if (!s->fd) {
1346         return ret;
1347     }
1348 
1349     start = sector_num * BDRV_SECTOR_SIZE;
1350     total_size = bdrv_getlength(bs);
1351     if (total_size < 0) {
1352         return total_size;
1353     } else if (start >= total_size) {
1354         *pnum = 0;
1355         return 0;
1356     } else if (start + nb_sectors * BDRV_SECTOR_SIZE > total_size) {
1357         nb_sectors = DIV_ROUND_UP(total_size - start, BDRV_SECTOR_SIZE);
1358     }
1359 
1360     ret = find_allocation(bs, start, &data, &hole);
1361     if (ret == -ENXIO) {
1362         /* Trailing hole */
1363         *pnum = nb_sectors;
1364         ret = BDRV_BLOCK_ZERO;
1365     } else if (ret < 0) {
1366         /* No info available, so pretend there are no holes */
1367         *pnum = nb_sectors;
1368         ret = BDRV_BLOCK_DATA;
1369     } else if (data == start) {
1370         /* On a data extent, compute sectors to the end of the extent,
1371          * possibly including a partial sector at EOF. */
1372         *pnum = MIN(nb_sectors, DIV_ROUND_UP(hole - start, BDRV_SECTOR_SIZE));
1373         ret = BDRV_BLOCK_DATA;
1374     } else {
1375         /* On a hole, compute sectors to the beginning of the next extent.  */
1376         assert(hole == start);
1377         *pnum = MIN(nb_sectors, (data - start) / BDRV_SECTOR_SIZE);
1378         ret = BDRV_BLOCK_ZERO;
1379     }
1380 
1381     *file = bs;
1382 
1383     return ret | BDRV_BLOCK_OFFSET_VALID | start;
1384 }
1385 
1386 
1387 static BlockDriver bdrv_gluster = {
1388     .format_name                  = "gluster",
1389     .protocol_name                = "gluster",
1390     .instance_size                = sizeof(BDRVGlusterState),
1391     .bdrv_needs_filename          = false,
1392     .bdrv_file_open               = qemu_gluster_open,
1393     .bdrv_reopen_prepare          = qemu_gluster_reopen_prepare,
1394     .bdrv_reopen_commit           = qemu_gluster_reopen_commit,
1395     .bdrv_reopen_abort            = qemu_gluster_reopen_abort,
1396     .bdrv_close                   = qemu_gluster_close,
1397     .bdrv_create                  = qemu_gluster_create,
1398     .bdrv_getlength               = qemu_gluster_getlength,
1399     .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size,
1400     .bdrv_truncate                = qemu_gluster_truncate,
1401     .bdrv_co_readv                = qemu_gluster_co_readv,
1402     .bdrv_co_writev               = qemu_gluster_co_writev,
1403     .bdrv_co_flush_to_disk        = qemu_gluster_co_flush_to_disk,
1404     .bdrv_has_zero_init           = qemu_gluster_has_zero_init,
1405 #ifdef CONFIG_GLUSTERFS_DISCARD
1406     .bdrv_co_pdiscard             = qemu_gluster_co_pdiscard,
1407 #endif
1408 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1409     .bdrv_co_pwrite_zeroes        = qemu_gluster_co_pwrite_zeroes,
1410 #endif
1411     .bdrv_co_get_block_status     = qemu_gluster_co_get_block_status,
1412     .create_opts                  = &qemu_gluster_create_opts,
1413 };
1414 
1415 static BlockDriver bdrv_gluster_tcp = {
1416     .format_name                  = "gluster",
1417     .protocol_name                = "gluster+tcp",
1418     .instance_size                = sizeof(BDRVGlusterState),
1419     .bdrv_needs_filename          = false,
1420     .bdrv_file_open               = qemu_gluster_open,
1421     .bdrv_reopen_prepare          = qemu_gluster_reopen_prepare,
1422     .bdrv_reopen_commit           = qemu_gluster_reopen_commit,
1423     .bdrv_reopen_abort            = qemu_gluster_reopen_abort,
1424     .bdrv_close                   = qemu_gluster_close,
1425     .bdrv_create                  = qemu_gluster_create,
1426     .bdrv_getlength               = qemu_gluster_getlength,
1427     .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size,
1428     .bdrv_truncate                = qemu_gluster_truncate,
1429     .bdrv_co_readv                = qemu_gluster_co_readv,
1430     .bdrv_co_writev               = qemu_gluster_co_writev,
1431     .bdrv_co_flush_to_disk        = qemu_gluster_co_flush_to_disk,
1432     .bdrv_has_zero_init           = qemu_gluster_has_zero_init,
1433 #ifdef CONFIG_GLUSTERFS_DISCARD
1434     .bdrv_co_pdiscard             = qemu_gluster_co_pdiscard,
1435 #endif
1436 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1437     .bdrv_co_pwrite_zeroes        = qemu_gluster_co_pwrite_zeroes,
1438 #endif
1439     .bdrv_co_get_block_status     = qemu_gluster_co_get_block_status,
1440     .create_opts                  = &qemu_gluster_create_opts,
1441 };
1442 
1443 static BlockDriver bdrv_gluster_unix = {
1444     .format_name                  = "gluster",
1445     .protocol_name                = "gluster+unix",
1446     .instance_size                = sizeof(BDRVGlusterState),
1447     .bdrv_needs_filename          = true,
1448     .bdrv_file_open               = qemu_gluster_open,
1449     .bdrv_reopen_prepare          = qemu_gluster_reopen_prepare,
1450     .bdrv_reopen_commit           = qemu_gluster_reopen_commit,
1451     .bdrv_reopen_abort            = qemu_gluster_reopen_abort,
1452     .bdrv_close                   = qemu_gluster_close,
1453     .bdrv_create                  = qemu_gluster_create,
1454     .bdrv_getlength               = qemu_gluster_getlength,
1455     .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size,
1456     .bdrv_truncate                = qemu_gluster_truncate,
1457     .bdrv_co_readv                = qemu_gluster_co_readv,
1458     .bdrv_co_writev               = qemu_gluster_co_writev,
1459     .bdrv_co_flush_to_disk        = qemu_gluster_co_flush_to_disk,
1460     .bdrv_has_zero_init           = qemu_gluster_has_zero_init,
1461 #ifdef CONFIG_GLUSTERFS_DISCARD
1462     .bdrv_co_pdiscard             = qemu_gluster_co_pdiscard,
1463 #endif
1464 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1465     .bdrv_co_pwrite_zeroes        = qemu_gluster_co_pwrite_zeroes,
1466 #endif
1467     .bdrv_co_get_block_status     = qemu_gluster_co_get_block_status,
1468     .create_opts                  = &qemu_gluster_create_opts,
1469 };
1470 
1471 /* rdma is deprecated (actually never supported for volfile fetch).
1472  * Let's maintain it for the protocol compatibility, to make sure things
1473  * won't break immediately. For now, gluster+rdma will fall back to gluster+tcp
1474  * protocol with a warning.
1475  * TODO: remove gluster+rdma interface support
1476  */
1477 static BlockDriver bdrv_gluster_rdma = {
1478     .format_name                  = "gluster",
1479     .protocol_name                = "gluster+rdma",
1480     .instance_size                = sizeof(BDRVGlusterState),
1481     .bdrv_needs_filename          = true,
1482     .bdrv_file_open               = qemu_gluster_open,
1483     .bdrv_reopen_prepare          = qemu_gluster_reopen_prepare,
1484     .bdrv_reopen_commit           = qemu_gluster_reopen_commit,
1485     .bdrv_reopen_abort            = qemu_gluster_reopen_abort,
1486     .bdrv_close                   = qemu_gluster_close,
1487     .bdrv_create                  = qemu_gluster_create,
1488     .bdrv_getlength               = qemu_gluster_getlength,
1489     .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size,
1490     .bdrv_truncate                = qemu_gluster_truncate,
1491     .bdrv_co_readv                = qemu_gluster_co_readv,
1492     .bdrv_co_writev               = qemu_gluster_co_writev,
1493     .bdrv_co_flush_to_disk        = qemu_gluster_co_flush_to_disk,
1494     .bdrv_has_zero_init           = qemu_gluster_has_zero_init,
1495 #ifdef CONFIG_GLUSTERFS_DISCARD
1496     .bdrv_co_pdiscard             = qemu_gluster_co_pdiscard,
1497 #endif
1498 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1499     .bdrv_co_pwrite_zeroes        = qemu_gluster_co_pwrite_zeroes,
1500 #endif
1501     .bdrv_co_get_block_status     = qemu_gluster_co_get_block_status,
1502     .create_opts                  = &qemu_gluster_create_opts,
1503 };
1504 
1505 static void bdrv_gluster_init(void)
1506 {
1507     bdrv_register(&bdrv_gluster_rdma);
1508     bdrv_register(&bdrv_gluster_unix);
1509     bdrv_register(&bdrv_gluster_tcp);
1510     bdrv_register(&bdrv_gluster);
1511 }
1512 
1513 block_init(bdrv_gluster_init);
1514