xref: /openbmc/qemu/block/gluster.c (revision 922a01a013d2270682a188258cbccacfecf8129c)
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 
11 #include "qemu/osdep.h"
12 #include <glusterfs/api/glfs.h>
13 #include "block/block_int.h"
14 #include "qapi/error.h"
15 #include "qapi/qmp/qdict.h"
16 #include "qapi/qmp/qerror.h"
17 #include "qemu/uri.h"
18 #include "qemu/error-report.h"
19 #include "qemu/option.h"
20 #include "qemu/cutils.h"
21 
22 #define GLUSTER_OPT_FILENAME        "filename"
23 #define GLUSTER_OPT_VOLUME          "volume"
24 #define GLUSTER_OPT_PATH            "path"
25 #define GLUSTER_OPT_TYPE            "type"
26 #define GLUSTER_OPT_SERVER_PATTERN  "server."
27 #define GLUSTER_OPT_HOST            "host"
28 #define GLUSTER_OPT_PORT            "port"
29 #define GLUSTER_OPT_TO              "to"
30 #define GLUSTER_OPT_IPV4            "ipv4"
31 #define GLUSTER_OPT_IPV6            "ipv6"
32 #define GLUSTER_OPT_SOCKET          "socket"
33 #define GLUSTER_OPT_DEBUG           "debug"
34 #define GLUSTER_DEFAULT_PORT        24007
35 #define GLUSTER_DEBUG_DEFAULT       4
36 #define GLUSTER_DEBUG_MAX           9
37 #define GLUSTER_OPT_LOGFILE         "logfile"
38 #define GLUSTER_LOGFILE_DEFAULT     "-" /* handled in libgfapi as /dev/stderr */
39 
40 #define GERR_INDEX_HINT "hint: check in 'server' array index '%d'\n"
41 
42 typedef struct GlusterAIOCB {
43     int64_t size;
44     int ret;
45     Coroutine *coroutine;
46     AioContext *aio_context;
47 } GlusterAIOCB;
48 
49 typedef struct BDRVGlusterState {
50     struct glfs *glfs;
51     struct glfs_fd *fd;
52     char *logfile;
53     bool supports_seek_data;
54     int debug;
55 } BDRVGlusterState;
56 
57 typedef struct BDRVGlusterReopenState {
58     struct glfs *glfs;
59     struct glfs_fd *fd;
60 } BDRVGlusterReopenState;
61 
62 
63 typedef struct GlfsPreopened {
64     char *volume;
65     glfs_t *fs;
66     int ref;
67 } GlfsPreopened;
68 
69 typedef struct ListElement {
70     QLIST_ENTRY(ListElement) list;
71     GlfsPreopened saved;
72 } ListElement;
73 
74 static QLIST_HEAD(glfs_list, ListElement) glfs_list;
75 
76 static QemuOptsList qemu_gluster_create_opts = {
77     .name = "qemu-gluster-create-opts",
78     .head = QTAILQ_HEAD_INITIALIZER(qemu_gluster_create_opts.head),
79     .desc = {
80         {
81             .name = BLOCK_OPT_SIZE,
82             .type = QEMU_OPT_SIZE,
83             .help = "Virtual disk size"
84         },
85         {
86             .name = BLOCK_OPT_PREALLOC,
87             .type = QEMU_OPT_STRING,
88             .help = "Preallocation mode (allowed values: off, full)"
89         },
90         {
91             .name = GLUSTER_OPT_DEBUG,
92             .type = QEMU_OPT_NUMBER,
93             .help = "Gluster log level, valid range is 0-9",
94         },
95         {
96             .name = GLUSTER_OPT_LOGFILE,
97             .type = QEMU_OPT_STRING,
98             .help = "Logfile path of libgfapi",
99         },
100         { /* end of list */ }
101     }
102 };
103 
104 static QemuOptsList runtime_opts = {
105     .name = "gluster",
106     .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
107     .desc = {
108         {
109             .name = GLUSTER_OPT_FILENAME,
110             .type = QEMU_OPT_STRING,
111             .help = "URL to the gluster image",
112         },
113         {
114             .name = GLUSTER_OPT_DEBUG,
115             .type = QEMU_OPT_NUMBER,
116             .help = "Gluster log level, valid range is 0-9",
117         },
118         {
119             .name = GLUSTER_OPT_LOGFILE,
120             .type = QEMU_OPT_STRING,
121             .help = "Logfile path of libgfapi",
122         },
123         { /* end of list */ }
124     },
125 };
126 
127 static QemuOptsList runtime_json_opts = {
128     .name = "gluster_json",
129     .head = QTAILQ_HEAD_INITIALIZER(runtime_json_opts.head),
130     .desc = {
131         {
132             .name = GLUSTER_OPT_VOLUME,
133             .type = QEMU_OPT_STRING,
134             .help = "name of gluster volume where VM image resides",
135         },
136         {
137             .name = GLUSTER_OPT_PATH,
138             .type = QEMU_OPT_STRING,
139             .help = "absolute path to image file in gluster volume",
140         },
141         {
142             .name = GLUSTER_OPT_DEBUG,
143             .type = QEMU_OPT_NUMBER,
144             .help = "Gluster log level, valid range is 0-9",
145         },
146         { /* end of list */ }
147     },
148 };
149 
150 static QemuOptsList runtime_type_opts = {
151     .name = "gluster_type",
152     .head = QTAILQ_HEAD_INITIALIZER(runtime_type_opts.head),
153     .desc = {
154         {
155             .name = GLUSTER_OPT_TYPE,
156             .type = QEMU_OPT_STRING,
157             .help = "inet|unix",
158         },
159         { /* end of list */ }
160     },
161 };
162 
163 static QemuOptsList runtime_unix_opts = {
164     .name = "gluster_unix",
165     .head = QTAILQ_HEAD_INITIALIZER(runtime_unix_opts.head),
166     .desc = {
167         {
168             .name = GLUSTER_OPT_SOCKET,
169             .type = QEMU_OPT_STRING,
170             .help = "socket file path)",
171         },
172         { /* end of list */ }
173     },
174 };
175 
176 static QemuOptsList runtime_inet_opts = {
177     .name = "gluster_inet",
178     .head = QTAILQ_HEAD_INITIALIZER(runtime_inet_opts.head),
179     .desc = {
180         {
181             .name = GLUSTER_OPT_TYPE,
182             .type = QEMU_OPT_STRING,
183             .help = "inet|unix",
184         },
185         {
186             .name = GLUSTER_OPT_HOST,
187             .type = QEMU_OPT_STRING,
188             .help = "host address (hostname/ipv4/ipv6 addresses)",
189         },
190         {
191             .name = GLUSTER_OPT_PORT,
192             .type = QEMU_OPT_STRING,
193             .help = "port number on which glusterd is listening (default 24007)",
194         },
195         {
196             .name = "to",
197             .type = QEMU_OPT_NUMBER,
198             .help = "max port number, not supported by gluster",
199         },
200         {
201             .name = "ipv4",
202             .type = QEMU_OPT_BOOL,
203             .help = "ipv4 bool value, not supported by gluster",
204         },
205         {
206             .name = "ipv6",
207             .type = QEMU_OPT_BOOL,
208             .help = "ipv6 bool value, not supported by gluster",
209         },
210         { /* end of list */ }
211     },
212 };
213 
214 static void glfs_set_preopened(const char *volume, glfs_t *fs)
215 {
216     ListElement *entry = NULL;
217 
218     entry = g_new(ListElement, 1);
219 
220     entry->saved.volume = g_strdup(volume);
221 
222     entry->saved.fs = fs;
223     entry->saved.ref = 1;
224 
225     QLIST_INSERT_HEAD(&glfs_list, entry, list);
226 }
227 
228 static glfs_t *glfs_find_preopened(const char *volume)
229 {
230     ListElement *entry = NULL;
231 
232      QLIST_FOREACH(entry, &glfs_list, list) {
233         if (strcmp(entry->saved.volume, volume) == 0) {
234             entry->saved.ref++;
235             return entry->saved.fs;
236         }
237      }
238 
239     return NULL;
240 }
241 
242 static void glfs_clear_preopened(glfs_t *fs)
243 {
244     ListElement *entry = NULL;
245     ListElement *next;
246 
247     if (fs == NULL) {
248         return;
249     }
250 
251     QLIST_FOREACH_SAFE(entry, &glfs_list, list, next) {
252         if (entry->saved.fs == fs) {
253             if (--entry->saved.ref) {
254                 return;
255             }
256 
257             QLIST_REMOVE(entry, list);
258 
259             glfs_fini(entry->saved.fs);
260             g_free(entry->saved.volume);
261             g_free(entry);
262         }
263     }
264 }
265 
266 static int parse_volume_options(BlockdevOptionsGluster *gconf, char *path)
267 {
268     char *p, *q;
269 
270     if (!path) {
271         return -EINVAL;
272     }
273 
274     /* volume */
275     p = q = path + strspn(path, "/");
276     p += strcspn(p, "/");
277     if (*p == '\0') {
278         return -EINVAL;
279     }
280     gconf->volume = g_strndup(q, p - q);
281 
282     /* path */
283     p += strspn(p, "/");
284     if (*p == '\0') {
285         return -EINVAL;
286     }
287     gconf->path = g_strdup(p);
288     return 0;
289 }
290 
291 /*
292  * file=gluster[+transport]://[host[:port]]/volume/path[?socket=...]
293  *
294  * 'gluster' is the protocol.
295  *
296  * 'transport' specifies the transport type used to connect to gluster
297  * management daemon (glusterd). Valid transport types are
298  * tcp or unix. If a transport type isn't specified, then tcp type is assumed.
299  *
300  * 'host' specifies the host where the volume file specification for
301  * the given volume resides. This can be either hostname or ipv4 address.
302  * If transport type is 'unix', then 'host' field should not be specified.
303  * The 'socket' field needs to be populated with the path to unix domain
304  * socket.
305  *
306  * 'port' is the port number on which glusterd is listening. This is optional
307  * and if not specified, QEMU will send 0 which will make gluster to use the
308  * default port. If the transport type is unix, then 'port' should not be
309  * specified.
310  *
311  * 'volume' is the name of the gluster volume which contains the VM image.
312  *
313  * 'path' is the path to the actual VM image that resides on gluster volume.
314  *
315  * Examples:
316  *
317  * file=gluster://1.2.3.4/testvol/a.img
318  * file=gluster+tcp://1.2.3.4/testvol/a.img
319  * file=gluster+tcp://1.2.3.4:24007/testvol/dir/a.img
320  * file=gluster+tcp://host.domain.com:24007/testvol/dir/a.img
321  * file=gluster+unix:///testvol/dir/a.img?socket=/tmp/glusterd.socket
322  */
323 static int qemu_gluster_parse_uri(BlockdevOptionsGluster *gconf,
324                                   const char *filename)
325 {
326     SocketAddress *gsconf;
327     URI *uri;
328     QueryParams *qp = NULL;
329     bool is_unix = false;
330     int ret = 0;
331 
332     uri = uri_parse(filename);
333     if (!uri) {
334         return -EINVAL;
335     }
336 
337     gconf->server = g_new0(SocketAddressList, 1);
338     gconf->server->value = gsconf = g_new0(SocketAddress, 1);
339 
340     /* transport */
341     if (!uri->scheme || !strcmp(uri->scheme, "gluster")) {
342         gsconf->type = SOCKET_ADDRESS_TYPE_INET;
343     } else if (!strcmp(uri->scheme, "gluster+tcp")) {
344         gsconf->type = SOCKET_ADDRESS_TYPE_INET;
345     } else if (!strcmp(uri->scheme, "gluster+unix")) {
346         gsconf->type = SOCKET_ADDRESS_TYPE_UNIX;
347         is_unix = true;
348     } else if (!strcmp(uri->scheme, "gluster+rdma")) {
349         gsconf->type = SOCKET_ADDRESS_TYPE_INET;
350         warn_report("rdma feature is not supported, falling back to tcp");
351     } else {
352         ret = -EINVAL;
353         goto out;
354     }
355 
356     ret = parse_volume_options(gconf, uri->path);
357     if (ret < 0) {
358         goto out;
359     }
360 
361     qp = query_params_parse(uri->query);
362     if (qp->n > 1 || (is_unix && !qp->n) || (!is_unix && qp->n)) {
363         ret = -EINVAL;
364         goto out;
365     }
366 
367     if (is_unix) {
368         if (uri->server || uri->port) {
369             ret = -EINVAL;
370             goto out;
371         }
372         if (strcmp(qp->p[0].name, "socket")) {
373             ret = -EINVAL;
374             goto out;
375         }
376         gsconf->u.q_unix.path = g_strdup(qp->p[0].value);
377     } else {
378         gsconf->u.inet.host = g_strdup(uri->server ? uri->server : "localhost");
379         if (uri->port) {
380             gsconf->u.inet.port = g_strdup_printf("%d", uri->port);
381         } else {
382             gsconf->u.inet.port = g_strdup_printf("%d", GLUSTER_DEFAULT_PORT);
383         }
384     }
385 
386 out:
387     if (qp) {
388         query_params_free(qp);
389     }
390     uri_free(uri);
391     return ret;
392 }
393 
394 static struct glfs *qemu_gluster_glfs_init(BlockdevOptionsGluster *gconf,
395                                            Error **errp)
396 {
397     struct glfs *glfs;
398     int ret;
399     int old_errno;
400     SocketAddressList *server;
401     unsigned long long port;
402 
403     glfs = glfs_find_preopened(gconf->volume);
404     if (glfs) {
405         return glfs;
406     }
407 
408     glfs = glfs_new(gconf->volume);
409     if (!glfs) {
410         goto out;
411     }
412 
413     glfs_set_preopened(gconf->volume, glfs);
414 
415     for (server = gconf->server; server; server = server->next) {
416         switch (server->value->type) {
417         case SOCKET_ADDRESS_TYPE_UNIX:
418             ret = glfs_set_volfile_server(glfs, "unix",
419                                    server->value->u.q_unix.path, 0);
420             break;
421         case SOCKET_ADDRESS_TYPE_INET:
422             if (parse_uint_full(server->value->u.inet.port, &port, 10) < 0 ||
423                 port > 65535) {
424                 error_setg(errp, "'%s' is not a valid port number",
425                            server->value->u.inet.port);
426                 errno = EINVAL;
427                 goto out;
428             }
429             ret = glfs_set_volfile_server(glfs, "tcp",
430                                    server->value->u.inet.host,
431                                    (int)port);
432             break;
433         case SOCKET_ADDRESS_TYPE_VSOCK:
434         case SOCKET_ADDRESS_TYPE_FD:
435         default:
436             abort();
437         }
438 
439         if (ret < 0) {
440             goto out;
441         }
442     }
443 
444     ret = glfs_set_logging(glfs, gconf->logfile, gconf->debug);
445     if (ret < 0) {
446         goto out;
447     }
448 
449     ret = glfs_init(glfs);
450     if (ret) {
451         error_setg(errp, "Gluster connection for volume %s, path %s failed"
452                          " to connect", gconf->volume, gconf->path);
453         for (server = gconf->server; server; server = server->next) {
454             if (server->value->type  == SOCKET_ADDRESS_TYPE_UNIX) {
455                 error_append_hint(errp, "hint: failed on socket %s ",
456                                   server->value->u.q_unix.path);
457             } else {
458                 error_append_hint(errp, "hint: failed on host %s and port %s ",
459                                   server->value->u.inet.host,
460                                   server->value->u.inet.port);
461             }
462         }
463 
464         error_append_hint(errp, "Please refer to gluster logs for more info\n");
465 
466         /* glfs_init sometimes doesn't set errno although docs suggest that */
467         if (errno == 0) {
468             errno = EINVAL;
469         }
470 
471         goto out;
472     }
473     return glfs;
474 
475 out:
476     if (glfs) {
477         old_errno = errno;
478         glfs_clear_preopened(glfs);
479         errno = old_errno;
480     }
481     return NULL;
482 }
483 
484 /*
485  * Convert the json formatted command line into qapi.
486 */
487 static int qemu_gluster_parse_json(BlockdevOptionsGluster *gconf,
488                                   QDict *options, Error **errp)
489 {
490     QemuOpts *opts;
491     SocketAddress *gsconf = NULL;
492     SocketAddressList *curr = NULL;
493     QDict *backing_options = NULL;
494     Error *local_err = NULL;
495     char *str = NULL;
496     const char *ptr;
497     int i, type, num_servers;
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, -1, NULL);
550         if (type != SOCKET_ADDRESS_TYPE_INET
551             && type != SOCKET_ADDRESS_TYPE_UNIX) {
552             error_setg(&local_err,
553                        "Parameter '%s' may be 'inet' or 'unix'",
554                        GLUSTER_OPT_TYPE);
555             error_append_hint(&local_err, GERR_INDEX_HINT, i);
556             goto out;
557         }
558         gsconf->type = type;
559         qemu_opts_del(opts);
560 
561         if (gsconf->type == SOCKET_ADDRESS_TYPE_INET) {
562             /* create opts info from runtime_inet_opts list */
563             opts = qemu_opts_create(&runtime_inet_opts, NULL, 0, &error_abort);
564             qemu_opts_absorb_qdict(opts, backing_options, &local_err);
565             if (local_err) {
566                 goto out;
567             }
568 
569             ptr = qemu_opt_get(opts, GLUSTER_OPT_HOST);
570             if (!ptr) {
571                 error_setg(&local_err, QERR_MISSING_PARAMETER,
572                            GLUSTER_OPT_HOST);
573                 error_append_hint(&local_err, GERR_INDEX_HINT, i);
574                 goto out;
575             }
576             gsconf->u.inet.host = g_strdup(ptr);
577             ptr = qemu_opt_get(opts, GLUSTER_OPT_PORT);
578             if (!ptr) {
579                 error_setg(&local_err, QERR_MISSING_PARAMETER,
580                            GLUSTER_OPT_PORT);
581                 error_append_hint(&local_err, GERR_INDEX_HINT, i);
582                 goto out;
583             }
584             gsconf->u.inet.port = g_strdup(ptr);
585 
586             /* defend for unsupported fields in InetSocketAddress,
587              * i.e. @ipv4, @ipv6  and @to
588              */
589             ptr = qemu_opt_get(opts, GLUSTER_OPT_TO);
590             if (ptr) {
591                 gsconf->u.inet.has_to = true;
592             }
593             ptr = qemu_opt_get(opts, GLUSTER_OPT_IPV4);
594             if (ptr) {
595                 gsconf->u.inet.has_ipv4 = true;
596             }
597             ptr = qemu_opt_get(opts, GLUSTER_OPT_IPV6);
598             if (ptr) {
599                 gsconf->u.inet.has_ipv6 = true;
600             }
601             if (gsconf->u.inet.has_to) {
602                 error_setg(&local_err, "Parameter 'to' not supported");
603                 goto out;
604             }
605             if (gsconf->u.inet.has_ipv4 || gsconf->u.inet.has_ipv6) {
606                 error_setg(&local_err, "Parameters 'ipv4/ipv6' not supported");
607                 goto out;
608             }
609             qemu_opts_del(opts);
610         } else {
611             /* create opts info from runtime_unix_opts list */
612             opts = qemu_opts_create(&runtime_unix_opts, NULL, 0, &error_abort);
613             qemu_opts_absorb_qdict(opts, backing_options, &local_err);
614             if (local_err) {
615                 goto out;
616             }
617 
618             ptr = qemu_opt_get(opts, GLUSTER_OPT_SOCKET);
619             if (!ptr) {
620                 error_setg(&local_err, QERR_MISSING_PARAMETER,
621                            GLUSTER_OPT_SOCKET);
622                 error_append_hint(&local_err, GERR_INDEX_HINT, i);
623                 goto out;
624             }
625             gsconf->u.q_unix.path = g_strdup(ptr);
626             qemu_opts_del(opts);
627         }
628 
629         if (gconf->server == NULL) {
630             gconf->server = g_new0(SocketAddressList, 1);
631             gconf->server->value = gsconf;
632             curr = gconf->server;
633         } else {
634             curr->next = g_new0(SocketAddressList, 1);
635             curr->next->value = gsconf;
636             curr = curr->next;
637         }
638         gsconf = NULL;
639 
640         QDECREF(backing_options);
641         backing_options = NULL;
642         g_free(str);
643         str = NULL;
644     }
645 
646     return 0;
647 
648 out:
649     error_propagate(errp, local_err);
650     qapi_free_SocketAddress(gsconf);
651     qemu_opts_del(opts);
652     g_free(str);
653     QDECREF(backing_options);
654     errno = EINVAL;
655     return -errno;
656 }
657 
658 static struct glfs *qemu_gluster_init(BlockdevOptionsGluster *gconf,
659                                       const char *filename,
660                                       QDict *options, Error **errp)
661 {
662     int ret;
663     if (filename) {
664         ret = qemu_gluster_parse_uri(gconf, filename);
665         if (ret < 0) {
666             error_setg(errp, "invalid URI");
667             error_append_hint(errp, "Usage: file=gluster[+transport]://"
668                                     "[host[:port]]volume/path[?socket=...]"
669                                     "[,file.debug=N]"
670                                     "[,file.logfile=/path/filename.log]\n");
671             errno = -ret;
672             return NULL;
673         }
674     } else {
675         ret = qemu_gluster_parse_json(gconf, options, errp);
676         if (ret < 0) {
677             error_append_hint(errp, "Usage: "
678                              "-drive driver=qcow2,file.driver=gluster,"
679                              "file.volume=testvol,file.path=/path/a.qcow2"
680                              "[,file.debug=9]"
681                              "[,file.logfile=/path/filename.log],"
682                              "file.server.0.type=inet,"
683                              "file.server.0.host=1.2.3.4,"
684                              "file.server.0.port=24007,"
685                              "file.server.1.transport=unix,"
686                              "file.server.1.socket=/var/run/glusterd.socket ..."
687                              "\n");
688             errno = -ret;
689             return NULL;
690         }
691 
692     }
693 
694     return qemu_gluster_glfs_init(gconf, errp);
695 }
696 
697 /*
698  * AIO callback routine called from GlusterFS thread.
699  */
700 static void gluster_finish_aiocb(struct glfs_fd *fd, ssize_t ret, void *arg)
701 {
702     GlusterAIOCB *acb = (GlusterAIOCB *)arg;
703 
704     if (!ret || ret == acb->size) {
705         acb->ret = 0; /* Success */
706     } else if (ret < 0) {
707         acb->ret = -errno; /* Read/Write failed */
708     } else {
709         acb->ret = -EIO; /* Partial read/write - fail it */
710     }
711 
712     aio_co_schedule(acb->aio_context, acb->coroutine);
713 }
714 
715 static void qemu_gluster_parse_flags(int bdrv_flags, int *open_flags)
716 {
717     assert(open_flags != NULL);
718 
719     *open_flags |= O_BINARY;
720 
721     if (bdrv_flags & BDRV_O_RDWR) {
722         *open_flags |= O_RDWR;
723     } else {
724         *open_flags |= O_RDONLY;
725     }
726 
727     if ((bdrv_flags & BDRV_O_NOCACHE)) {
728         *open_flags |= O_DIRECT;
729     }
730 }
731 
732 /*
733  * Do SEEK_DATA/HOLE to detect if it is functional. Older broken versions of
734  * gfapi incorrectly return the current offset when SEEK_DATA/HOLE is used.
735  * - Corrected versions return -1 and set errno to EINVAL.
736  * - Versions that support SEEK_DATA/HOLE correctly, will return -1 and set
737  *   errno to ENXIO when SEEK_DATA is called with a position of EOF.
738  */
739 static bool qemu_gluster_test_seek(struct glfs_fd *fd)
740 {
741     off_t ret = 0;
742 
743 #if defined SEEK_HOLE && defined SEEK_DATA
744     off_t eof;
745 
746     eof = glfs_lseek(fd, 0, SEEK_END);
747     if (eof < 0) {
748         /* this should never occur */
749         return false;
750     }
751 
752     /* this should always fail with ENXIO if SEEK_DATA is supported */
753     ret = glfs_lseek(fd, eof, SEEK_DATA);
754 #endif
755 
756     return (ret < 0) && (errno == ENXIO);
757 }
758 
759 static int qemu_gluster_open(BlockDriverState *bs,  QDict *options,
760                              int bdrv_flags, Error **errp)
761 {
762     BDRVGlusterState *s = bs->opaque;
763     int open_flags = 0;
764     int ret = 0;
765     BlockdevOptionsGluster *gconf = NULL;
766     QemuOpts *opts;
767     Error *local_err = NULL;
768     const char *filename, *logfile;
769 
770     opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
771     qemu_opts_absorb_qdict(opts, options, &local_err);
772     if (local_err) {
773         error_propagate(errp, local_err);
774         ret = -EINVAL;
775         goto out;
776     }
777 
778     filename = qemu_opt_get(opts, GLUSTER_OPT_FILENAME);
779 
780     s->debug = qemu_opt_get_number(opts, GLUSTER_OPT_DEBUG,
781                                    GLUSTER_DEBUG_DEFAULT);
782     if (s->debug < 0) {
783         s->debug = 0;
784     } else if (s->debug > GLUSTER_DEBUG_MAX) {
785         s->debug = GLUSTER_DEBUG_MAX;
786     }
787 
788     gconf = g_new0(BlockdevOptionsGluster, 1);
789     gconf->debug = s->debug;
790     gconf->has_debug = true;
791 
792     logfile = qemu_opt_get(opts, GLUSTER_OPT_LOGFILE);
793     s->logfile = g_strdup(logfile ? logfile : GLUSTER_LOGFILE_DEFAULT);
794 
795     gconf->logfile = g_strdup(s->logfile);
796     gconf->has_logfile = true;
797 
798     s->glfs = qemu_gluster_init(gconf, filename, options, errp);
799     if (!s->glfs) {
800         ret = -errno;
801         goto out;
802     }
803 
804 #ifdef CONFIG_GLUSTERFS_XLATOR_OPT
805     /* Without this, if fsync fails for a recoverable reason (for instance,
806      * ENOSPC), gluster will dump its cache, preventing retries.  This means
807      * almost certain data loss.  Not all gluster versions support the
808      * 'resync-failed-syncs-after-fsync' key value, but there is no way to
809      * discover during runtime if it is supported (this api returns success for
810      * unknown key/value pairs) */
811     ret = glfs_set_xlator_option(s->glfs, "*-write-behind",
812                                           "resync-failed-syncs-after-fsync",
813                                           "on");
814     if (ret < 0) {
815         error_setg_errno(errp, errno, "Unable to set xlator key/value pair");
816         ret = -errno;
817         goto out;
818     }
819 #endif
820 
821     qemu_gluster_parse_flags(bdrv_flags, &open_flags);
822 
823     s->fd = glfs_open(s->glfs, gconf->path, open_flags);
824     if (!s->fd) {
825         ret = -errno;
826     }
827 
828     s->supports_seek_data = qemu_gluster_test_seek(s->fd);
829 
830 out:
831     qemu_opts_del(opts);
832     qapi_free_BlockdevOptionsGluster(gconf);
833     if (!ret) {
834         return ret;
835     }
836     g_free(s->logfile);
837     if (s->fd) {
838         glfs_close(s->fd);
839     }
840 
841     glfs_clear_preopened(s->glfs);
842 
843     return ret;
844 }
845 
846 static int qemu_gluster_reopen_prepare(BDRVReopenState *state,
847                                        BlockReopenQueue *queue, Error **errp)
848 {
849     int ret = 0;
850     BDRVGlusterState *s;
851     BDRVGlusterReopenState *reop_s;
852     BlockdevOptionsGluster *gconf;
853     int open_flags = 0;
854 
855     assert(state != NULL);
856     assert(state->bs != NULL);
857 
858     s = state->bs->opaque;
859 
860     state->opaque = g_new0(BDRVGlusterReopenState, 1);
861     reop_s = state->opaque;
862 
863     qemu_gluster_parse_flags(state->flags, &open_flags);
864 
865     gconf = g_new0(BlockdevOptionsGluster, 1);
866     gconf->debug = s->debug;
867     gconf->has_debug = true;
868     gconf->logfile = g_strdup(s->logfile);
869     gconf->has_logfile = true;
870     reop_s->glfs = qemu_gluster_init(gconf, state->bs->filename, NULL, errp);
871     if (reop_s->glfs == NULL) {
872         ret = -errno;
873         goto exit;
874     }
875 
876 #ifdef CONFIG_GLUSTERFS_XLATOR_OPT
877     ret = glfs_set_xlator_option(reop_s->glfs, "*-write-behind",
878                                  "resync-failed-syncs-after-fsync", "on");
879     if (ret < 0) {
880         error_setg_errno(errp, errno, "Unable to set xlator key/value pair");
881         ret = -errno;
882         goto exit;
883     }
884 #endif
885 
886     reop_s->fd = glfs_open(reop_s->glfs, gconf->path, open_flags);
887     if (reop_s->fd == NULL) {
888         /* reops->glfs will be cleaned up in _abort */
889         ret = -errno;
890         goto exit;
891     }
892 
893 exit:
894     /* state->opaque will be freed in either the _abort or _commit */
895     qapi_free_BlockdevOptionsGluster(gconf);
896     return ret;
897 }
898 
899 static void qemu_gluster_reopen_commit(BDRVReopenState *state)
900 {
901     BDRVGlusterReopenState *reop_s = state->opaque;
902     BDRVGlusterState *s = state->bs->opaque;
903 
904 
905     /* close the old */
906     if (s->fd) {
907         glfs_close(s->fd);
908     }
909 
910     glfs_clear_preopened(s->glfs);
911 
912     /* use the newly opened image / connection */
913     s->fd         = reop_s->fd;
914     s->glfs       = reop_s->glfs;
915 
916     g_free(state->opaque);
917     state->opaque = NULL;
918 
919     return;
920 }
921 
922 
923 static void qemu_gluster_reopen_abort(BDRVReopenState *state)
924 {
925     BDRVGlusterReopenState *reop_s = state->opaque;
926 
927     if (reop_s == NULL) {
928         return;
929     }
930 
931     if (reop_s->fd) {
932         glfs_close(reop_s->fd);
933     }
934 
935     glfs_clear_preopened(reop_s->glfs);
936 
937     g_free(state->opaque);
938     state->opaque = NULL;
939 
940     return;
941 }
942 
943 #ifdef CONFIG_GLUSTERFS_ZEROFILL
944 static coroutine_fn int qemu_gluster_co_pwrite_zeroes(BlockDriverState *bs,
945                                                       int64_t offset,
946                                                       int size,
947                                                       BdrvRequestFlags flags)
948 {
949     int ret;
950     GlusterAIOCB acb;
951     BDRVGlusterState *s = bs->opaque;
952 
953     acb.size = size;
954     acb.ret = 0;
955     acb.coroutine = qemu_coroutine_self();
956     acb.aio_context = bdrv_get_aio_context(bs);
957 
958     ret = glfs_zerofill_async(s->fd, offset, size, gluster_finish_aiocb, &acb);
959     if (ret < 0) {
960         return -errno;
961     }
962 
963     qemu_coroutine_yield();
964     return acb.ret;
965 }
966 #endif
967 
968 static int qemu_gluster_create(const char *filename,
969                                QemuOpts *opts, Error **errp)
970 {
971     BlockdevOptionsGluster *gconf;
972     struct glfs *glfs;
973     struct glfs_fd *fd;
974     int ret = 0;
975     PreallocMode prealloc;
976     int64_t total_size = 0;
977     char *tmp = NULL;
978     Error *local_err = NULL;
979 
980     gconf = g_new0(BlockdevOptionsGluster, 1);
981     gconf->debug = qemu_opt_get_number_del(opts, GLUSTER_OPT_DEBUG,
982                                            GLUSTER_DEBUG_DEFAULT);
983     if (gconf->debug < 0) {
984         gconf->debug = 0;
985     } else if (gconf->debug > GLUSTER_DEBUG_MAX) {
986         gconf->debug = GLUSTER_DEBUG_MAX;
987     }
988     gconf->has_debug = true;
989 
990     gconf->logfile = qemu_opt_get_del(opts, GLUSTER_OPT_LOGFILE);
991     if (!gconf->logfile) {
992         gconf->logfile = g_strdup(GLUSTER_LOGFILE_DEFAULT);
993     }
994     gconf->has_logfile = true;
995 
996     glfs = qemu_gluster_init(gconf, filename, NULL, errp);
997     if (!glfs) {
998         ret = -errno;
999         goto out;
1000     }
1001 
1002     total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
1003                           BDRV_SECTOR_SIZE);
1004 
1005     tmp = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
1006     prealloc = qapi_enum_parse(&PreallocMode_lookup, tmp, PREALLOC_MODE_OFF,
1007                                &local_err);
1008     g_free(tmp);
1009     if (local_err) {
1010         error_propagate(errp, local_err);
1011         ret = -EINVAL;
1012         goto out;
1013     }
1014 
1015     fd = glfs_creat(glfs, gconf->path,
1016                     O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, S_IRUSR | S_IWUSR);
1017     if (!fd) {
1018         ret = -errno;
1019         goto out;
1020     }
1021 
1022     switch (prealloc) {
1023 #ifdef CONFIG_GLUSTERFS_FALLOCATE
1024     case PREALLOC_MODE_FALLOC:
1025         if (glfs_fallocate(fd, 0, 0, total_size)) {
1026             error_setg(errp, "Could not preallocate data for the new file");
1027             ret = -errno;
1028         }
1029         break;
1030 #endif /* CONFIG_GLUSTERFS_FALLOCATE */
1031 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1032     case PREALLOC_MODE_FULL:
1033         if (!glfs_ftruncate(fd, total_size)) {
1034             if (glfs_zerofill(fd, 0, total_size)) {
1035                 error_setg(errp, "Could not zerofill the new file");
1036                 ret = -errno;
1037             }
1038         } else {
1039             error_setg(errp, "Could not resize file");
1040             ret = -errno;
1041         }
1042         break;
1043 #endif /* CONFIG_GLUSTERFS_ZEROFILL */
1044     case PREALLOC_MODE_OFF:
1045         if (glfs_ftruncate(fd, total_size) != 0) {
1046             ret = -errno;
1047             error_setg(errp, "Could not resize file");
1048         }
1049         break;
1050     default:
1051         ret = -EINVAL;
1052         error_setg(errp, "Unsupported preallocation mode: %s",
1053                    PreallocMode_str(prealloc));
1054         break;
1055     }
1056 
1057     if (glfs_close(fd) != 0) {
1058         ret = -errno;
1059     }
1060 out:
1061     qapi_free_BlockdevOptionsGluster(gconf);
1062     glfs_clear_preopened(glfs);
1063     return ret;
1064 }
1065 
1066 static coroutine_fn int qemu_gluster_co_rw(BlockDriverState *bs,
1067                                            int64_t sector_num, int nb_sectors,
1068                                            QEMUIOVector *qiov, int write)
1069 {
1070     int ret;
1071     GlusterAIOCB acb;
1072     BDRVGlusterState *s = bs->opaque;
1073     size_t size = nb_sectors * BDRV_SECTOR_SIZE;
1074     off_t offset = sector_num * BDRV_SECTOR_SIZE;
1075 
1076     acb.size = size;
1077     acb.ret = 0;
1078     acb.coroutine = qemu_coroutine_self();
1079     acb.aio_context = bdrv_get_aio_context(bs);
1080 
1081     if (write) {
1082         ret = glfs_pwritev_async(s->fd, qiov->iov, qiov->niov, offset, 0,
1083                                  gluster_finish_aiocb, &acb);
1084     } else {
1085         ret = glfs_preadv_async(s->fd, qiov->iov, qiov->niov, offset, 0,
1086                                 gluster_finish_aiocb, &acb);
1087     }
1088 
1089     if (ret < 0) {
1090         return -errno;
1091     }
1092 
1093     qemu_coroutine_yield();
1094     return acb.ret;
1095 }
1096 
1097 static int qemu_gluster_truncate(BlockDriverState *bs, int64_t offset,
1098                                  PreallocMode prealloc, Error **errp)
1099 {
1100     int ret;
1101     BDRVGlusterState *s = bs->opaque;
1102 
1103     if (prealloc != PREALLOC_MODE_OFF) {
1104         error_setg(errp, "Unsupported preallocation mode '%s'",
1105                    PreallocMode_str(prealloc));
1106         return -ENOTSUP;
1107     }
1108 
1109     ret = glfs_ftruncate(s->fd, offset);
1110     if (ret < 0) {
1111         ret = -errno;
1112         error_setg_errno(errp, -ret, "Failed to truncate file");
1113         return ret;
1114     }
1115 
1116     return 0;
1117 }
1118 
1119 static coroutine_fn int qemu_gluster_co_readv(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, 0);
1125 }
1126 
1127 static coroutine_fn int qemu_gluster_co_writev(BlockDriverState *bs,
1128                                                int64_t sector_num,
1129                                                int nb_sectors,
1130                                                QEMUIOVector *qiov)
1131 {
1132     return qemu_gluster_co_rw(bs, sector_num, nb_sectors, qiov, 1);
1133 }
1134 
1135 static void qemu_gluster_close(BlockDriverState *bs)
1136 {
1137     BDRVGlusterState *s = bs->opaque;
1138 
1139     g_free(s->logfile);
1140     if (s->fd) {
1141         glfs_close(s->fd);
1142         s->fd = NULL;
1143     }
1144     glfs_clear_preopened(s->glfs);
1145 }
1146 
1147 static coroutine_fn int qemu_gluster_co_flush_to_disk(BlockDriverState *bs)
1148 {
1149     int ret;
1150     GlusterAIOCB acb;
1151     BDRVGlusterState *s = bs->opaque;
1152 
1153     acb.size = 0;
1154     acb.ret = 0;
1155     acb.coroutine = qemu_coroutine_self();
1156     acb.aio_context = bdrv_get_aio_context(bs);
1157 
1158     ret = glfs_fsync_async(s->fd, gluster_finish_aiocb, &acb);
1159     if (ret < 0) {
1160         ret = -errno;
1161         goto error;
1162     }
1163 
1164     qemu_coroutine_yield();
1165     if (acb.ret < 0) {
1166         ret = acb.ret;
1167         goto error;
1168     }
1169 
1170     return acb.ret;
1171 
1172 error:
1173     /* Some versions of Gluster (3.5.6 -> 3.5.8?) will not retain its cache
1174      * after a fsync failure, so we have no way of allowing the guest to safely
1175      * continue.  Gluster versions prior to 3.5.6 don't retain the cache
1176      * either, but will invalidate the fd on error, so this is again our only
1177      * option.
1178      *
1179      * The 'resync-failed-syncs-after-fsync' xlator option for the
1180      * write-behind cache will cause later gluster versions to retain its
1181      * cache after error, so long as the fd remains open.  However, we
1182      * currently have no way of knowing if this option is supported.
1183      *
1184      * TODO: Once gluster provides a way for us to determine if the option
1185      * is supported, bypass the closure and setting drv to NULL.  */
1186     qemu_gluster_close(bs);
1187     bs->drv = NULL;
1188     return ret;
1189 }
1190 
1191 #ifdef CONFIG_GLUSTERFS_DISCARD
1192 static coroutine_fn int qemu_gluster_co_pdiscard(BlockDriverState *bs,
1193                                                  int64_t offset, int size)
1194 {
1195     int ret;
1196     GlusterAIOCB acb;
1197     BDRVGlusterState *s = bs->opaque;
1198 
1199     acb.size = 0;
1200     acb.ret = 0;
1201     acb.coroutine = qemu_coroutine_self();
1202     acb.aio_context = bdrv_get_aio_context(bs);
1203 
1204     ret = glfs_discard_async(s->fd, offset, size, gluster_finish_aiocb, &acb);
1205     if (ret < 0) {
1206         return -errno;
1207     }
1208 
1209     qemu_coroutine_yield();
1210     return acb.ret;
1211 }
1212 #endif
1213 
1214 static int64_t qemu_gluster_getlength(BlockDriverState *bs)
1215 {
1216     BDRVGlusterState *s = bs->opaque;
1217     int64_t ret;
1218 
1219     ret = glfs_lseek(s->fd, 0, SEEK_END);
1220     if (ret < 0) {
1221         return -errno;
1222     } else {
1223         return ret;
1224     }
1225 }
1226 
1227 static int64_t qemu_gluster_allocated_file_size(BlockDriverState *bs)
1228 {
1229     BDRVGlusterState *s = bs->opaque;
1230     struct stat st;
1231     int ret;
1232 
1233     ret = glfs_fstat(s->fd, &st);
1234     if (ret < 0) {
1235         return -errno;
1236     } else {
1237         return st.st_blocks * 512;
1238     }
1239 }
1240 
1241 static int qemu_gluster_has_zero_init(BlockDriverState *bs)
1242 {
1243     /* GlusterFS volume could be backed by a block device */
1244     return 0;
1245 }
1246 
1247 /*
1248  * Find allocation range in @bs around offset @start.
1249  * May change underlying file descriptor's file offset.
1250  * If @start is not in a hole, store @start in @data, and the
1251  * beginning of the next hole in @hole, and return 0.
1252  * If @start is in a non-trailing hole, store @start in @hole and the
1253  * beginning of the next non-hole in @data, and return 0.
1254  * If @start is in a trailing hole or beyond EOF, return -ENXIO.
1255  * If we can't find out, return a negative errno other than -ENXIO.
1256  *
1257  * (Shamefully copied from file-posix.c, only miniscule adaptions.)
1258  */
1259 static int find_allocation(BlockDriverState *bs, off_t start,
1260                            off_t *data, off_t *hole)
1261 {
1262     BDRVGlusterState *s = bs->opaque;
1263 
1264     if (!s->supports_seek_data) {
1265         goto exit;
1266     }
1267 
1268 #if defined SEEK_HOLE && defined SEEK_DATA
1269     off_t offs;
1270 
1271     /*
1272      * SEEK_DATA cases:
1273      * D1. offs == start: start is in data
1274      * D2. offs > start: start is in a hole, next data at offs
1275      * D3. offs < 0, errno = ENXIO: either start is in a trailing hole
1276      *                              or start is beyond EOF
1277      *     If the latter happens, the file has been truncated behind
1278      *     our back since we opened it.  All bets are off then.
1279      *     Treating like a trailing hole is simplest.
1280      * D4. offs < 0, errno != ENXIO: we learned nothing
1281      */
1282     offs = glfs_lseek(s->fd, start, SEEK_DATA);
1283     if (offs < 0) {
1284         return -errno;          /* D3 or D4 */
1285     }
1286 
1287     if (offs < start) {
1288         /* This is not a valid return by lseek().  We are safe to just return
1289          * -EIO in this case, and we'll treat it like D4. Unfortunately some
1290          *  versions of gluster server will return offs < start, so an assert
1291          *  here will unnecessarily abort QEMU. */
1292         return -EIO;
1293     }
1294 
1295     if (offs > start) {
1296         /* D2: in hole, next data at offs */
1297         *hole = start;
1298         *data = offs;
1299         return 0;
1300     }
1301 
1302     /* D1: in data, end not yet known */
1303 
1304     /*
1305      * SEEK_HOLE cases:
1306      * H1. offs == start: start is in a hole
1307      *     If this happens here, a hole has been dug behind our back
1308      *     since the previous lseek().
1309      * H2. offs > start: either start is in data, next hole at offs,
1310      *                   or start is in trailing hole, EOF at offs
1311      *     Linux treats trailing holes like any other hole: offs ==
1312      *     start.  Solaris seeks to EOF instead: offs > start (blech).
1313      *     If that happens here, a hole has been dug behind our back
1314      *     since the previous lseek().
1315      * H3. offs < 0, errno = ENXIO: start is beyond EOF
1316      *     If this happens, the file has been truncated behind our
1317      *     back since we opened it.  Treat it like a trailing hole.
1318      * H4. offs < 0, errno != ENXIO: we learned nothing
1319      *     Pretend we know nothing at all, i.e. "forget" about D1.
1320      */
1321     offs = glfs_lseek(s->fd, start, SEEK_HOLE);
1322     if (offs < 0) {
1323         return -errno;          /* D1 and (H3 or H4) */
1324     }
1325 
1326     if (offs < start) {
1327         /* This is not a valid return by lseek().  We are safe to just return
1328          * -EIO in this case, and we'll treat it like H4. Unfortunately some
1329          *  versions of gluster server will return offs < start, so an assert
1330          *  here will unnecessarily abort QEMU. */
1331         return -EIO;
1332     }
1333 
1334     if (offs > start) {
1335         /*
1336          * D1 and H2: either in data, next hole at offs, or it was in
1337          * data but is now in a trailing hole.  In the latter case,
1338          * all bets are off.  Treating it as if it there was data all
1339          * the way to EOF is safe, so simply do that.
1340          */
1341         *data = start;
1342         *hole = offs;
1343         return 0;
1344     }
1345 
1346     /* D1 and H1 */
1347     return -EBUSY;
1348 #endif
1349 
1350 exit:
1351     return -ENOTSUP;
1352 }
1353 
1354 /*
1355  * Returns the allocation status of the specified sectors.
1356  *
1357  * If 'sector_num' is beyond the end of the disk image the return value is 0
1358  * and 'pnum' is set to 0.
1359  *
1360  * 'pnum' is set to the number of sectors (including and immediately following
1361  * the specified sector) that are known to be in the same
1362  * allocated/unallocated state.
1363  *
1364  * 'nb_sectors' is the max value 'pnum' should be set to.  If nb_sectors goes
1365  * beyond the end of the disk image it will be clamped.
1366  *
1367  * (Based on raw_co_get_block_status() from file-posix.c.)
1368  */
1369 static int64_t coroutine_fn qemu_gluster_co_get_block_status(
1370         BlockDriverState *bs, int64_t sector_num, int nb_sectors, int *pnum,
1371         BlockDriverState **file)
1372 {
1373     BDRVGlusterState *s = bs->opaque;
1374     off_t start, data = 0, hole = 0;
1375     int64_t total_size;
1376     int ret = -EINVAL;
1377 
1378     if (!s->fd) {
1379         return ret;
1380     }
1381 
1382     start = sector_num * BDRV_SECTOR_SIZE;
1383     total_size = bdrv_getlength(bs);
1384     if (total_size < 0) {
1385         return total_size;
1386     } else if (start >= total_size) {
1387         *pnum = 0;
1388         return 0;
1389     } else if (start + nb_sectors * BDRV_SECTOR_SIZE > total_size) {
1390         nb_sectors = DIV_ROUND_UP(total_size - start, BDRV_SECTOR_SIZE);
1391     }
1392 
1393     ret = find_allocation(bs, start, &data, &hole);
1394     if (ret == -ENXIO) {
1395         /* Trailing hole */
1396         *pnum = nb_sectors;
1397         ret = BDRV_BLOCK_ZERO;
1398     } else if (ret < 0) {
1399         /* No info available, so pretend there are no holes */
1400         *pnum = nb_sectors;
1401         ret = BDRV_BLOCK_DATA;
1402     } else if (data == start) {
1403         /* On a data extent, compute sectors to the end of the extent,
1404          * possibly including a partial sector at EOF. */
1405         *pnum = MIN(nb_sectors, DIV_ROUND_UP(hole - start, BDRV_SECTOR_SIZE));
1406         ret = BDRV_BLOCK_DATA;
1407     } else {
1408         /* On a hole, compute sectors to the beginning of the next extent.  */
1409         assert(hole == start);
1410         *pnum = MIN(nb_sectors, (data - start) / BDRV_SECTOR_SIZE);
1411         ret = BDRV_BLOCK_ZERO;
1412     }
1413 
1414     *file = bs;
1415 
1416     return ret | BDRV_BLOCK_OFFSET_VALID | start;
1417 }
1418 
1419 
1420 static BlockDriver bdrv_gluster = {
1421     .format_name                  = "gluster",
1422     .protocol_name                = "gluster",
1423     .instance_size                = sizeof(BDRVGlusterState),
1424     .bdrv_needs_filename          = false,
1425     .bdrv_file_open               = qemu_gluster_open,
1426     .bdrv_reopen_prepare          = qemu_gluster_reopen_prepare,
1427     .bdrv_reopen_commit           = qemu_gluster_reopen_commit,
1428     .bdrv_reopen_abort            = qemu_gluster_reopen_abort,
1429     .bdrv_close                   = qemu_gluster_close,
1430     .bdrv_create                  = qemu_gluster_create,
1431     .bdrv_getlength               = qemu_gluster_getlength,
1432     .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size,
1433     .bdrv_truncate                = qemu_gluster_truncate,
1434     .bdrv_co_readv                = qemu_gluster_co_readv,
1435     .bdrv_co_writev               = qemu_gluster_co_writev,
1436     .bdrv_co_flush_to_disk        = qemu_gluster_co_flush_to_disk,
1437     .bdrv_has_zero_init           = qemu_gluster_has_zero_init,
1438 #ifdef CONFIG_GLUSTERFS_DISCARD
1439     .bdrv_co_pdiscard             = qemu_gluster_co_pdiscard,
1440 #endif
1441 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1442     .bdrv_co_pwrite_zeroes        = qemu_gluster_co_pwrite_zeroes,
1443 #endif
1444     .bdrv_co_get_block_status     = qemu_gluster_co_get_block_status,
1445     .create_opts                  = &qemu_gluster_create_opts,
1446 };
1447 
1448 static BlockDriver bdrv_gluster_tcp = {
1449     .format_name                  = "gluster",
1450     .protocol_name                = "gluster+tcp",
1451     .instance_size                = sizeof(BDRVGlusterState),
1452     .bdrv_needs_filename          = false,
1453     .bdrv_file_open               = qemu_gluster_open,
1454     .bdrv_reopen_prepare          = qemu_gluster_reopen_prepare,
1455     .bdrv_reopen_commit           = qemu_gluster_reopen_commit,
1456     .bdrv_reopen_abort            = qemu_gluster_reopen_abort,
1457     .bdrv_close                   = qemu_gluster_close,
1458     .bdrv_create                  = qemu_gluster_create,
1459     .bdrv_getlength               = qemu_gluster_getlength,
1460     .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size,
1461     .bdrv_truncate                = qemu_gluster_truncate,
1462     .bdrv_co_readv                = qemu_gluster_co_readv,
1463     .bdrv_co_writev               = qemu_gluster_co_writev,
1464     .bdrv_co_flush_to_disk        = qemu_gluster_co_flush_to_disk,
1465     .bdrv_has_zero_init           = qemu_gluster_has_zero_init,
1466 #ifdef CONFIG_GLUSTERFS_DISCARD
1467     .bdrv_co_pdiscard             = qemu_gluster_co_pdiscard,
1468 #endif
1469 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1470     .bdrv_co_pwrite_zeroes        = qemu_gluster_co_pwrite_zeroes,
1471 #endif
1472     .bdrv_co_get_block_status     = qemu_gluster_co_get_block_status,
1473     .create_opts                  = &qemu_gluster_create_opts,
1474 };
1475 
1476 static BlockDriver bdrv_gluster_unix = {
1477     .format_name                  = "gluster",
1478     .protocol_name                = "gluster+unix",
1479     .instance_size                = sizeof(BDRVGlusterState),
1480     .bdrv_needs_filename          = true,
1481     .bdrv_file_open               = qemu_gluster_open,
1482     .bdrv_reopen_prepare          = qemu_gluster_reopen_prepare,
1483     .bdrv_reopen_commit           = qemu_gluster_reopen_commit,
1484     .bdrv_reopen_abort            = qemu_gluster_reopen_abort,
1485     .bdrv_close                   = qemu_gluster_close,
1486     .bdrv_create                  = qemu_gluster_create,
1487     .bdrv_getlength               = qemu_gluster_getlength,
1488     .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size,
1489     .bdrv_truncate                = qemu_gluster_truncate,
1490     .bdrv_co_readv                = qemu_gluster_co_readv,
1491     .bdrv_co_writev               = qemu_gluster_co_writev,
1492     .bdrv_co_flush_to_disk        = qemu_gluster_co_flush_to_disk,
1493     .bdrv_has_zero_init           = qemu_gluster_has_zero_init,
1494 #ifdef CONFIG_GLUSTERFS_DISCARD
1495     .bdrv_co_pdiscard             = qemu_gluster_co_pdiscard,
1496 #endif
1497 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1498     .bdrv_co_pwrite_zeroes        = qemu_gluster_co_pwrite_zeroes,
1499 #endif
1500     .bdrv_co_get_block_status     = qemu_gluster_co_get_block_status,
1501     .create_opts                  = &qemu_gluster_create_opts,
1502 };
1503 
1504 /* rdma is deprecated (actually never supported for volfile fetch).
1505  * Let's maintain it for the protocol compatibility, to make sure things
1506  * won't break immediately. For now, gluster+rdma will fall back to gluster+tcp
1507  * protocol with a warning.
1508  * TODO: remove gluster+rdma interface support
1509  */
1510 static BlockDriver bdrv_gluster_rdma = {
1511     .format_name                  = "gluster",
1512     .protocol_name                = "gluster+rdma",
1513     .instance_size                = sizeof(BDRVGlusterState),
1514     .bdrv_needs_filename          = true,
1515     .bdrv_file_open               = qemu_gluster_open,
1516     .bdrv_reopen_prepare          = qemu_gluster_reopen_prepare,
1517     .bdrv_reopen_commit           = qemu_gluster_reopen_commit,
1518     .bdrv_reopen_abort            = qemu_gluster_reopen_abort,
1519     .bdrv_close                   = qemu_gluster_close,
1520     .bdrv_create                  = qemu_gluster_create,
1521     .bdrv_getlength               = qemu_gluster_getlength,
1522     .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size,
1523     .bdrv_truncate                = qemu_gluster_truncate,
1524     .bdrv_co_readv                = qemu_gluster_co_readv,
1525     .bdrv_co_writev               = qemu_gluster_co_writev,
1526     .bdrv_co_flush_to_disk        = qemu_gluster_co_flush_to_disk,
1527     .bdrv_has_zero_init           = qemu_gluster_has_zero_init,
1528 #ifdef CONFIG_GLUSTERFS_DISCARD
1529     .bdrv_co_pdiscard             = qemu_gluster_co_pdiscard,
1530 #endif
1531 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1532     .bdrv_co_pwrite_zeroes        = qemu_gluster_co_pwrite_zeroes,
1533 #endif
1534     .bdrv_co_get_block_status     = qemu_gluster_co_get_block_status,
1535     .create_opts                  = &qemu_gluster_create_opts,
1536 };
1537 
1538 static void bdrv_gluster_init(void)
1539 {
1540     bdrv_register(&bdrv_gluster_rdma);
1541     bdrv_register(&bdrv_gluster_unix);
1542     bdrv_register(&bdrv_gluster_tcp);
1543     bdrv_register(&bdrv_gluster);
1544 }
1545 
1546 block_init(bdrv_gluster_init);
1547