xref: /openbmc/qemu/block/gluster.c (revision 500016e5)
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 "block/qdict.h"
15 #include "qapi/error.h"
16 #include "qapi/qmp/qdict.h"
17 #include "qapi/qmp/qerror.h"
18 #include "qemu/uri.h"
19 #include "qemu/error-report.h"
20 #include "qemu/option.h"
21 #include "qemu/cutils.h"
22 
23 #ifdef CONFIG_GLUSTERFS_FTRUNCATE_HAS_STAT
24 # define glfs_ftruncate(fd, offset) glfs_ftruncate(fd, offset, NULL, NULL)
25 #endif
26 
27 #define GLUSTER_OPT_FILENAME        "filename"
28 #define GLUSTER_OPT_VOLUME          "volume"
29 #define GLUSTER_OPT_PATH            "path"
30 #define GLUSTER_OPT_TYPE            "type"
31 #define GLUSTER_OPT_SERVER_PATTERN  "server."
32 #define GLUSTER_OPT_HOST            "host"
33 #define GLUSTER_OPT_PORT            "port"
34 #define GLUSTER_OPT_TO              "to"
35 #define GLUSTER_OPT_IPV4            "ipv4"
36 #define GLUSTER_OPT_IPV6            "ipv6"
37 #define GLUSTER_OPT_SOCKET          "socket"
38 #define GLUSTER_OPT_DEBUG           "debug"
39 #define GLUSTER_DEFAULT_PORT        24007
40 #define GLUSTER_DEBUG_DEFAULT       4
41 #define GLUSTER_DEBUG_MAX           9
42 #define GLUSTER_OPT_LOGFILE         "logfile"
43 #define GLUSTER_LOGFILE_DEFAULT     "-" /* handled in libgfapi as /dev/stderr */
44 
45 #define GERR_INDEX_HINT "hint: check in 'server' array index '%d'\n"
46 
47 typedef struct GlusterAIOCB {
48     int64_t size;
49     int ret;
50     Coroutine *coroutine;
51     AioContext *aio_context;
52 } GlusterAIOCB;
53 
54 typedef struct BDRVGlusterState {
55     struct glfs *glfs;
56     struct glfs_fd *fd;
57     char *logfile;
58     bool supports_seek_data;
59     int debug;
60 } BDRVGlusterState;
61 
62 typedef struct BDRVGlusterReopenState {
63     struct glfs *glfs;
64     struct glfs_fd *fd;
65 } BDRVGlusterReopenState;
66 
67 
68 typedef struct GlfsPreopened {
69     char *volume;
70     glfs_t *fs;
71     int ref;
72 } GlfsPreopened;
73 
74 typedef struct ListElement {
75     QLIST_ENTRY(ListElement) list;
76     GlfsPreopened saved;
77 } ListElement;
78 
79 static QLIST_HEAD(, ListElement) glfs_list;
80 
81 static QemuOptsList qemu_gluster_create_opts = {
82     .name = "qemu-gluster-create-opts",
83     .head = QTAILQ_HEAD_INITIALIZER(qemu_gluster_create_opts.head),
84     .desc = {
85         {
86             .name = BLOCK_OPT_SIZE,
87             .type = QEMU_OPT_SIZE,
88             .help = "Virtual disk size"
89         },
90         {
91             .name = BLOCK_OPT_PREALLOC,
92             .type = QEMU_OPT_STRING,
93             .help = "Preallocation mode (allowed values: off, full)"
94         },
95         {
96             .name = GLUSTER_OPT_DEBUG,
97             .type = QEMU_OPT_NUMBER,
98             .help = "Gluster log level, valid range is 0-9",
99         },
100         {
101             .name = GLUSTER_OPT_LOGFILE,
102             .type = QEMU_OPT_STRING,
103             .help = "Logfile path of libgfapi",
104         },
105         { /* end of list */ }
106     }
107 };
108 
109 static QemuOptsList runtime_opts = {
110     .name = "gluster",
111     .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
112     .desc = {
113         {
114             .name = GLUSTER_OPT_FILENAME,
115             .type = QEMU_OPT_STRING,
116             .help = "URL to the gluster image",
117         },
118         {
119             .name = GLUSTER_OPT_DEBUG,
120             .type = QEMU_OPT_NUMBER,
121             .help = "Gluster log level, valid range is 0-9",
122         },
123         {
124             .name = GLUSTER_OPT_LOGFILE,
125             .type = QEMU_OPT_STRING,
126             .help = "Logfile path of libgfapi",
127         },
128         { /* end of list */ }
129     },
130 };
131 
132 static QemuOptsList runtime_json_opts = {
133     .name = "gluster_json",
134     .head = QTAILQ_HEAD_INITIALIZER(runtime_json_opts.head),
135     .desc = {
136         {
137             .name = GLUSTER_OPT_VOLUME,
138             .type = QEMU_OPT_STRING,
139             .help = "name of gluster volume where VM image resides",
140         },
141         {
142             .name = GLUSTER_OPT_PATH,
143             .type = QEMU_OPT_STRING,
144             .help = "absolute path to image file in gluster volume",
145         },
146         {
147             .name = GLUSTER_OPT_DEBUG,
148             .type = QEMU_OPT_NUMBER,
149             .help = "Gluster log level, valid range is 0-9",
150         },
151         { /* end of list */ }
152     },
153 };
154 
155 static QemuOptsList runtime_type_opts = {
156     .name = "gluster_type",
157     .head = QTAILQ_HEAD_INITIALIZER(runtime_type_opts.head),
158     .desc = {
159         {
160             .name = GLUSTER_OPT_TYPE,
161             .type = QEMU_OPT_STRING,
162             .help = "inet|unix",
163         },
164         { /* end of list */ }
165     },
166 };
167 
168 static QemuOptsList runtime_unix_opts = {
169     .name = "gluster_unix",
170     .head = QTAILQ_HEAD_INITIALIZER(runtime_unix_opts.head),
171     .desc = {
172         {
173             .name = GLUSTER_OPT_SOCKET,
174             .type = QEMU_OPT_STRING,
175             .help = "socket file path (legacy)",
176         },
177         {
178             .name = GLUSTER_OPT_PATH,
179             .type = QEMU_OPT_STRING,
180             .help = "socket file path (QAPI)",
181         },
182         { /* end of list */ }
183     },
184 };
185 
186 static QemuOptsList runtime_inet_opts = {
187     .name = "gluster_inet",
188     .head = QTAILQ_HEAD_INITIALIZER(runtime_inet_opts.head),
189     .desc = {
190         {
191             .name = GLUSTER_OPT_TYPE,
192             .type = QEMU_OPT_STRING,
193             .help = "inet|unix",
194         },
195         {
196             .name = GLUSTER_OPT_HOST,
197             .type = QEMU_OPT_STRING,
198             .help = "host address (hostname/ipv4/ipv6 addresses)",
199         },
200         {
201             .name = GLUSTER_OPT_PORT,
202             .type = QEMU_OPT_STRING,
203             .help = "port number on which glusterd is listening (default 24007)",
204         },
205         {
206             .name = "to",
207             .type = QEMU_OPT_NUMBER,
208             .help = "max port number, not supported by gluster",
209         },
210         {
211             .name = "ipv4",
212             .type = QEMU_OPT_BOOL,
213             .help = "ipv4 bool value, not supported by gluster",
214         },
215         {
216             .name = "ipv6",
217             .type = QEMU_OPT_BOOL,
218             .help = "ipv6 bool value, not supported by gluster",
219         },
220         { /* end of list */ }
221     },
222 };
223 
224 static void glfs_set_preopened(const char *volume, glfs_t *fs)
225 {
226     ListElement *entry = NULL;
227 
228     entry = g_new(ListElement, 1);
229 
230     entry->saved.volume = g_strdup(volume);
231 
232     entry->saved.fs = fs;
233     entry->saved.ref = 1;
234 
235     QLIST_INSERT_HEAD(&glfs_list, entry, list);
236 }
237 
238 static glfs_t *glfs_find_preopened(const char *volume)
239 {
240     ListElement *entry = NULL;
241 
242      QLIST_FOREACH(entry, &glfs_list, list) {
243         if (strcmp(entry->saved.volume, volume) == 0) {
244             entry->saved.ref++;
245             return entry->saved.fs;
246         }
247      }
248 
249     return NULL;
250 }
251 
252 static void glfs_clear_preopened(glfs_t *fs)
253 {
254     ListElement *entry = NULL;
255     ListElement *next;
256 
257     if (fs == NULL) {
258         return;
259     }
260 
261     QLIST_FOREACH_SAFE(entry, &glfs_list, list, next) {
262         if (entry->saved.fs == fs) {
263             if (--entry->saved.ref) {
264                 return;
265             }
266 
267             QLIST_REMOVE(entry, list);
268 
269             glfs_fini(entry->saved.fs);
270             g_free(entry->saved.volume);
271             g_free(entry);
272         }
273     }
274 }
275 
276 static int parse_volume_options(BlockdevOptionsGluster *gconf, char *path)
277 {
278     char *p, *q;
279 
280     if (!path) {
281         return -EINVAL;
282     }
283 
284     /* volume */
285     p = q = path + strspn(path, "/");
286     p += strcspn(p, "/");
287     if (*p == '\0') {
288         return -EINVAL;
289     }
290     gconf->volume = g_strndup(q, p - q);
291 
292     /* path */
293     p += strspn(p, "/");
294     if (*p == '\0') {
295         return -EINVAL;
296     }
297     gconf->path = g_strdup(p);
298     return 0;
299 }
300 
301 /*
302  * file=gluster[+transport]://[host[:port]]/volume/path[?socket=...]
303  *
304  * 'gluster' is the protocol.
305  *
306  * 'transport' specifies the transport type used to connect to gluster
307  * management daemon (glusterd). Valid transport types are
308  * tcp or unix. If a transport type isn't specified, then tcp type is assumed.
309  *
310  * 'host' specifies the host where the volume file specification for
311  * the given volume resides. This can be either hostname or ipv4 address.
312  * If transport type is 'unix', then 'host' field should not be specified.
313  * The 'socket' field needs to be populated with the path to unix domain
314  * socket.
315  *
316  * 'port' is the port number on which glusterd is listening. This is optional
317  * and if not specified, QEMU will send 0 which will make gluster to use the
318  * default port. If the transport type is unix, then 'port' should not be
319  * specified.
320  *
321  * 'volume' is the name of the gluster volume which contains the VM image.
322  *
323  * 'path' is the path to the actual VM image that resides on gluster volume.
324  *
325  * Examples:
326  *
327  * file=gluster://1.2.3.4/testvol/a.img
328  * file=gluster+tcp://1.2.3.4/testvol/a.img
329  * file=gluster+tcp://1.2.3.4:24007/testvol/dir/a.img
330  * file=gluster+tcp://host.domain.com:24007/testvol/dir/a.img
331  * file=gluster+unix:///testvol/dir/a.img?socket=/tmp/glusterd.socket
332  */
333 static int qemu_gluster_parse_uri(BlockdevOptionsGluster *gconf,
334                                   const char *filename)
335 {
336     SocketAddress *gsconf;
337     URI *uri;
338     QueryParams *qp = NULL;
339     bool is_unix = false;
340     int ret = 0;
341 
342     uri = uri_parse(filename);
343     if (!uri) {
344         return -EINVAL;
345     }
346 
347     gconf->server = g_new0(SocketAddressList, 1);
348     gconf->server->value = gsconf = g_new0(SocketAddress, 1);
349 
350     /* transport */
351     if (!uri->scheme || !strcmp(uri->scheme, "gluster")) {
352         gsconf->type = SOCKET_ADDRESS_TYPE_INET;
353     } else if (!strcmp(uri->scheme, "gluster+tcp")) {
354         gsconf->type = SOCKET_ADDRESS_TYPE_INET;
355     } else if (!strcmp(uri->scheme, "gluster+unix")) {
356         gsconf->type = SOCKET_ADDRESS_TYPE_UNIX;
357         is_unix = true;
358     } else if (!strcmp(uri->scheme, "gluster+rdma")) {
359         gsconf->type = SOCKET_ADDRESS_TYPE_INET;
360         warn_report("rdma feature is not supported, falling back to tcp");
361     } else {
362         ret = -EINVAL;
363         goto out;
364     }
365 
366     ret = parse_volume_options(gconf, uri->path);
367     if (ret < 0) {
368         goto out;
369     }
370 
371     qp = query_params_parse(uri->query);
372     if (qp->n > 1 || (is_unix && !qp->n) || (!is_unix && qp->n)) {
373         ret = -EINVAL;
374         goto out;
375     }
376 
377     if (is_unix) {
378         if (uri->server || uri->port) {
379             ret = -EINVAL;
380             goto out;
381         }
382         if (strcmp(qp->p[0].name, "socket")) {
383             ret = -EINVAL;
384             goto out;
385         }
386         gsconf->u.q_unix.path = g_strdup(qp->p[0].value);
387     } else {
388         gsconf->u.inet.host = g_strdup(uri->server ? uri->server : "localhost");
389         if (uri->port) {
390             gsconf->u.inet.port = g_strdup_printf("%d", uri->port);
391         } else {
392             gsconf->u.inet.port = g_strdup_printf("%d", GLUSTER_DEFAULT_PORT);
393         }
394     }
395 
396 out:
397     if (qp) {
398         query_params_free(qp);
399     }
400     uri_free(uri);
401     return ret;
402 }
403 
404 static struct glfs *qemu_gluster_glfs_init(BlockdevOptionsGluster *gconf,
405                                            Error **errp)
406 {
407     struct glfs *glfs;
408     int ret;
409     int old_errno;
410     SocketAddressList *server;
411     unsigned long long port;
412 
413     glfs = glfs_find_preopened(gconf->volume);
414     if (glfs) {
415         return glfs;
416     }
417 
418     glfs = glfs_new(gconf->volume);
419     if (!glfs) {
420         goto out;
421     }
422 
423     glfs_set_preopened(gconf->volume, glfs);
424 
425     for (server = gconf->server; server; server = server->next) {
426         switch (server->value->type) {
427         case SOCKET_ADDRESS_TYPE_UNIX:
428             ret = glfs_set_volfile_server(glfs, "unix",
429                                    server->value->u.q_unix.path, 0);
430             break;
431         case SOCKET_ADDRESS_TYPE_INET:
432             if (parse_uint_full(server->value->u.inet.port, &port, 10) < 0 ||
433                 port > 65535) {
434                 error_setg(errp, "'%s' is not a valid port number",
435                            server->value->u.inet.port);
436                 errno = EINVAL;
437                 goto out;
438             }
439             ret = glfs_set_volfile_server(glfs, "tcp",
440                                    server->value->u.inet.host,
441                                    (int)port);
442             break;
443         case SOCKET_ADDRESS_TYPE_VSOCK:
444         case SOCKET_ADDRESS_TYPE_FD:
445         default:
446             abort();
447         }
448 
449         if (ret < 0) {
450             goto out;
451         }
452     }
453 
454     ret = glfs_set_logging(glfs, gconf->logfile, gconf->debug);
455     if (ret < 0) {
456         goto out;
457     }
458 
459     ret = glfs_init(glfs);
460     if (ret) {
461         error_setg(errp, "Gluster connection for volume %s, path %s failed"
462                          " to connect", gconf->volume, gconf->path);
463         for (server = gconf->server; server; server = server->next) {
464             if (server->value->type  == SOCKET_ADDRESS_TYPE_UNIX) {
465                 error_append_hint(errp, "hint: failed on socket %s ",
466                                   server->value->u.q_unix.path);
467             } else {
468                 error_append_hint(errp, "hint: failed on host %s and port %s ",
469                                   server->value->u.inet.host,
470                                   server->value->u.inet.port);
471             }
472         }
473 
474         error_append_hint(errp, "Please refer to gluster logs for more info\n");
475 
476         /* glfs_init sometimes doesn't set errno although docs suggest that */
477         if (errno == 0) {
478             errno = EINVAL;
479         }
480 
481         goto out;
482     }
483     return glfs;
484 
485 out:
486     if (glfs) {
487         old_errno = errno;
488         glfs_clear_preopened(glfs);
489         errno = old_errno;
490     }
491     return NULL;
492 }
493 
494 /*
495  * Convert the json formatted command line into qapi.
496 */
497 static int qemu_gluster_parse_json(BlockdevOptionsGluster *gconf,
498                                   QDict *options, Error **errp)
499 {
500     QemuOpts *opts;
501     SocketAddress *gsconf = NULL;
502     SocketAddressList *curr = NULL;
503     QDict *backing_options = NULL;
504     Error *local_err = NULL;
505     char *str = NULL;
506     const char *ptr;
507     int i, type, num_servers;
508 
509     /* create opts info from runtime_json_opts list */
510     opts = qemu_opts_create(&runtime_json_opts, NULL, 0, &error_abort);
511     qemu_opts_absorb_qdict(opts, options, &local_err);
512     if (local_err) {
513         goto out;
514     }
515 
516     num_servers = qdict_array_entries(options, GLUSTER_OPT_SERVER_PATTERN);
517     if (num_servers < 1) {
518         error_setg(&local_err, QERR_MISSING_PARAMETER, "server");
519         goto out;
520     }
521 
522     ptr = qemu_opt_get(opts, GLUSTER_OPT_VOLUME);
523     if (!ptr) {
524         error_setg(&local_err, QERR_MISSING_PARAMETER, GLUSTER_OPT_VOLUME);
525         goto out;
526     }
527     gconf->volume = g_strdup(ptr);
528 
529     ptr = qemu_opt_get(opts, GLUSTER_OPT_PATH);
530     if (!ptr) {
531         error_setg(&local_err, QERR_MISSING_PARAMETER, GLUSTER_OPT_PATH);
532         goto out;
533     }
534     gconf->path = g_strdup(ptr);
535     qemu_opts_del(opts);
536 
537     for (i = 0; i < num_servers; i++) {
538         str = g_strdup_printf(GLUSTER_OPT_SERVER_PATTERN"%d.", i);
539         qdict_extract_subqdict(options, &backing_options, str);
540 
541         /* create opts info from runtime_type_opts list */
542         opts = qemu_opts_create(&runtime_type_opts, NULL, 0, &error_abort);
543         qemu_opts_absorb_qdict(opts, backing_options, &local_err);
544         if (local_err) {
545             goto out;
546         }
547 
548         ptr = qemu_opt_get(opts, GLUSTER_OPT_TYPE);
549         if (!ptr) {
550             error_setg(&local_err, QERR_MISSING_PARAMETER, GLUSTER_OPT_TYPE);
551             error_append_hint(&local_err, GERR_INDEX_HINT, i);
552             goto out;
553 
554         }
555         gsconf = g_new0(SocketAddress, 1);
556         if (!strcmp(ptr, "tcp")) {
557             ptr = "inet";       /* accept legacy "tcp" */
558         }
559         type = qapi_enum_parse(&SocketAddressType_lookup, ptr, -1, NULL);
560         if (type != SOCKET_ADDRESS_TYPE_INET
561             && type != SOCKET_ADDRESS_TYPE_UNIX) {
562             error_setg(&local_err,
563                        "Parameter '%s' may be 'inet' or 'unix'",
564                        GLUSTER_OPT_TYPE);
565             error_append_hint(&local_err, GERR_INDEX_HINT, i);
566             goto out;
567         }
568         gsconf->type = type;
569         qemu_opts_del(opts);
570 
571         if (gsconf->type == SOCKET_ADDRESS_TYPE_INET) {
572             /* create opts info from runtime_inet_opts list */
573             opts = qemu_opts_create(&runtime_inet_opts, NULL, 0, &error_abort);
574             qemu_opts_absorb_qdict(opts, backing_options, &local_err);
575             if (local_err) {
576                 goto out;
577             }
578 
579             ptr = qemu_opt_get(opts, GLUSTER_OPT_HOST);
580             if (!ptr) {
581                 error_setg(&local_err, QERR_MISSING_PARAMETER,
582                            GLUSTER_OPT_HOST);
583                 error_append_hint(&local_err, GERR_INDEX_HINT, i);
584                 goto out;
585             }
586             gsconf->u.inet.host = g_strdup(ptr);
587             ptr = qemu_opt_get(opts, GLUSTER_OPT_PORT);
588             if (!ptr) {
589                 error_setg(&local_err, QERR_MISSING_PARAMETER,
590                            GLUSTER_OPT_PORT);
591                 error_append_hint(&local_err, GERR_INDEX_HINT, i);
592                 goto out;
593             }
594             gsconf->u.inet.port = g_strdup(ptr);
595 
596             /* defend for unsupported fields in InetSocketAddress,
597              * i.e. @ipv4, @ipv6  and @to
598              */
599             ptr = qemu_opt_get(opts, GLUSTER_OPT_TO);
600             if (ptr) {
601                 gsconf->u.inet.has_to = true;
602             }
603             ptr = qemu_opt_get(opts, GLUSTER_OPT_IPV4);
604             if (ptr) {
605                 gsconf->u.inet.has_ipv4 = true;
606             }
607             ptr = qemu_opt_get(opts, GLUSTER_OPT_IPV6);
608             if (ptr) {
609                 gsconf->u.inet.has_ipv6 = true;
610             }
611             if (gsconf->u.inet.has_to) {
612                 error_setg(&local_err, "Parameter 'to' not supported");
613                 goto out;
614             }
615             if (gsconf->u.inet.has_ipv4 || gsconf->u.inet.has_ipv6) {
616                 error_setg(&local_err, "Parameters 'ipv4/ipv6' not supported");
617                 goto out;
618             }
619             qemu_opts_del(opts);
620         } else {
621             /* create opts info from runtime_unix_opts list */
622             opts = qemu_opts_create(&runtime_unix_opts, NULL, 0, &error_abort);
623             qemu_opts_absorb_qdict(opts, backing_options, &local_err);
624             if (local_err) {
625                 goto out;
626             }
627 
628             ptr = qemu_opt_get(opts, GLUSTER_OPT_PATH);
629             if (!ptr) {
630                 ptr = qemu_opt_get(opts, GLUSTER_OPT_SOCKET);
631             } else if (qemu_opt_get(opts, GLUSTER_OPT_SOCKET)) {
632                 error_setg(&local_err,
633                            "Conflicting parameters 'path' and 'socket'");
634                 error_append_hint(&local_err, GERR_INDEX_HINT, i);
635                 goto out;
636             }
637             if (!ptr) {
638                 error_setg(&local_err, QERR_MISSING_PARAMETER,
639                            GLUSTER_OPT_PATH);
640                 error_append_hint(&local_err, GERR_INDEX_HINT, i);
641                 goto out;
642             }
643             gsconf->u.q_unix.path = g_strdup(ptr);
644             qemu_opts_del(opts);
645         }
646 
647         if (gconf->server == NULL) {
648             gconf->server = g_new0(SocketAddressList, 1);
649             gconf->server->value = gsconf;
650             curr = gconf->server;
651         } else {
652             curr->next = g_new0(SocketAddressList, 1);
653             curr->next->value = gsconf;
654             curr = curr->next;
655         }
656         gsconf = NULL;
657 
658         qobject_unref(backing_options);
659         backing_options = NULL;
660         g_free(str);
661         str = NULL;
662     }
663 
664     return 0;
665 
666 out:
667     error_propagate(errp, local_err);
668     qapi_free_SocketAddress(gsconf);
669     qemu_opts_del(opts);
670     g_free(str);
671     qobject_unref(backing_options);
672     errno = EINVAL;
673     return -errno;
674 }
675 
676 /* Converts options given in @filename and the @options QDict into the QAPI
677  * object @gconf. */
678 static int qemu_gluster_parse(BlockdevOptionsGluster *gconf,
679                               const char *filename,
680                               QDict *options, Error **errp)
681 {
682     int ret;
683     if (filename) {
684         ret = qemu_gluster_parse_uri(gconf, filename);
685         if (ret < 0) {
686             error_setg(errp, "invalid URI %s", filename);
687             error_append_hint(errp, "Usage: file=gluster[+transport]://"
688                                     "[host[:port]]volume/path[?socket=...]"
689                                     "[,file.debug=N]"
690                                     "[,file.logfile=/path/filename.log]\n");
691             return ret;
692         }
693     } else {
694         ret = qemu_gluster_parse_json(gconf, options, errp);
695         if (ret < 0) {
696             error_append_hint(errp, "Usage: "
697                              "-drive driver=qcow2,file.driver=gluster,"
698                              "file.volume=testvol,file.path=/path/a.qcow2"
699                              "[,file.debug=9]"
700                              "[,file.logfile=/path/filename.log],"
701                              "file.server.0.type=inet,"
702                              "file.server.0.host=1.2.3.4,"
703                              "file.server.0.port=24007,"
704                              "file.server.1.transport=unix,"
705                              "file.server.1.path=/var/run/glusterd.socket ..."
706                              "\n");
707             return ret;
708         }
709     }
710 
711     return 0;
712 }
713 
714 static struct glfs *qemu_gluster_init(BlockdevOptionsGluster *gconf,
715                                       const char *filename,
716                                       QDict *options, Error **errp)
717 {
718     int ret;
719 
720     ret = qemu_gluster_parse(gconf, filename, options, errp);
721     if (ret < 0) {
722         errno = -ret;
723         return NULL;
724     }
725 
726     return qemu_gluster_glfs_init(gconf, errp);
727 }
728 
729 /*
730  * AIO callback routine called from GlusterFS thread.
731  */
732 static void gluster_finish_aiocb(struct glfs_fd *fd, ssize_t ret,
733 #ifdef CONFIG_GLUSTERFS_IOCB_HAS_STAT
734                                  struct glfs_stat *pre, struct glfs_stat *post,
735 #endif
736                                  void *arg)
737 {
738     GlusterAIOCB *acb = (GlusterAIOCB *)arg;
739 
740     if (!ret || ret == acb->size) {
741         acb->ret = 0; /* Success */
742     } else if (ret < 0) {
743         acb->ret = -errno; /* Read/Write failed */
744     } else {
745         acb->ret = -EIO; /* Partial read/write - fail it */
746     }
747 
748     aio_co_schedule(acb->aio_context, acb->coroutine);
749 }
750 
751 static void qemu_gluster_parse_flags(int bdrv_flags, int *open_flags)
752 {
753     assert(open_flags != NULL);
754 
755     *open_flags |= O_BINARY;
756 
757     if (bdrv_flags & BDRV_O_RDWR) {
758         *open_flags |= O_RDWR;
759     } else {
760         *open_flags |= O_RDONLY;
761     }
762 
763     if ((bdrv_flags & BDRV_O_NOCACHE)) {
764         *open_flags |= O_DIRECT;
765     }
766 }
767 
768 /*
769  * Do SEEK_DATA/HOLE to detect if it is functional. Older broken versions of
770  * gfapi incorrectly return the current offset when SEEK_DATA/HOLE is used.
771  * - Corrected versions return -1 and set errno to EINVAL.
772  * - Versions that support SEEK_DATA/HOLE correctly, will return -1 and set
773  *   errno to ENXIO when SEEK_DATA is called with a position of EOF.
774  */
775 static bool qemu_gluster_test_seek(struct glfs_fd *fd)
776 {
777     off_t ret = 0;
778 
779 #if defined SEEK_HOLE && defined SEEK_DATA
780     off_t eof;
781 
782     eof = glfs_lseek(fd, 0, SEEK_END);
783     if (eof < 0) {
784         /* this should never occur */
785         return false;
786     }
787 
788     /* this should always fail with ENXIO if SEEK_DATA is supported */
789     ret = glfs_lseek(fd, eof, SEEK_DATA);
790 #endif
791 
792     return (ret < 0) && (errno == ENXIO);
793 }
794 
795 static int qemu_gluster_open(BlockDriverState *bs,  QDict *options,
796                              int bdrv_flags, Error **errp)
797 {
798     BDRVGlusterState *s = bs->opaque;
799     int open_flags = 0;
800     int ret = 0;
801     BlockdevOptionsGluster *gconf = NULL;
802     QemuOpts *opts;
803     Error *local_err = NULL;
804     const char *filename, *logfile;
805 
806     opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
807     qemu_opts_absorb_qdict(opts, options, &local_err);
808     if (local_err) {
809         error_propagate(errp, local_err);
810         ret = -EINVAL;
811         goto out;
812     }
813 
814     filename = qemu_opt_get(opts, GLUSTER_OPT_FILENAME);
815 
816     s->debug = qemu_opt_get_number(opts, GLUSTER_OPT_DEBUG,
817                                    GLUSTER_DEBUG_DEFAULT);
818     if (s->debug < 0) {
819         s->debug = 0;
820     } else if (s->debug > GLUSTER_DEBUG_MAX) {
821         s->debug = GLUSTER_DEBUG_MAX;
822     }
823 
824     gconf = g_new0(BlockdevOptionsGluster, 1);
825     gconf->debug = s->debug;
826     gconf->has_debug = true;
827 
828     logfile = qemu_opt_get(opts, GLUSTER_OPT_LOGFILE);
829     s->logfile = g_strdup(logfile ? logfile : GLUSTER_LOGFILE_DEFAULT);
830 
831     gconf->logfile = g_strdup(s->logfile);
832     gconf->has_logfile = true;
833 
834     s->glfs = qemu_gluster_init(gconf, filename, options, errp);
835     if (!s->glfs) {
836         ret = -errno;
837         goto out;
838     }
839 
840 #ifdef CONFIG_GLUSTERFS_XLATOR_OPT
841     /* Without this, if fsync fails for a recoverable reason (for instance,
842      * ENOSPC), gluster will dump its cache, preventing retries.  This means
843      * almost certain data loss.  Not all gluster versions support the
844      * 'resync-failed-syncs-after-fsync' key value, but there is no way to
845      * discover during runtime if it is supported (this api returns success for
846      * unknown key/value pairs) */
847     ret = glfs_set_xlator_option(s->glfs, "*-write-behind",
848                                           "resync-failed-syncs-after-fsync",
849                                           "on");
850     if (ret < 0) {
851         error_setg_errno(errp, errno, "Unable to set xlator key/value pair");
852         ret = -errno;
853         goto out;
854     }
855 #endif
856 
857     qemu_gluster_parse_flags(bdrv_flags, &open_flags);
858 
859     s->fd = glfs_open(s->glfs, gconf->path, open_flags);
860     ret = s->fd ? 0 : -errno;
861 
862     if (ret == -EACCES || ret == -EROFS) {
863         /* Try to degrade to read-only, but if it doesn't work, still use the
864          * normal error message. */
865         if (bdrv_apply_auto_read_only(bs, NULL, NULL) == 0) {
866             open_flags = (open_flags & ~O_RDWR) | O_RDONLY;
867             s->fd = glfs_open(s->glfs, gconf->path, open_flags);
868             ret = s->fd ? 0 : -errno;
869         }
870     }
871 
872     s->supports_seek_data = qemu_gluster_test_seek(s->fd);
873 
874 out:
875     qemu_opts_del(opts);
876     qapi_free_BlockdevOptionsGluster(gconf);
877     if (!ret) {
878         return ret;
879     }
880     g_free(s->logfile);
881     if (s->fd) {
882         glfs_close(s->fd);
883     }
884 
885     glfs_clear_preopened(s->glfs);
886 
887     return ret;
888 }
889 
890 static int qemu_gluster_reopen_prepare(BDRVReopenState *state,
891                                        BlockReopenQueue *queue, Error **errp)
892 {
893     int ret = 0;
894     BDRVGlusterState *s;
895     BDRVGlusterReopenState *reop_s;
896     BlockdevOptionsGluster *gconf;
897     int open_flags = 0;
898 
899     assert(state != NULL);
900     assert(state->bs != NULL);
901 
902     s = state->bs->opaque;
903 
904     state->opaque = g_new0(BDRVGlusterReopenState, 1);
905     reop_s = state->opaque;
906 
907     qemu_gluster_parse_flags(state->flags, &open_flags);
908 
909     gconf = g_new0(BlockdevOptionsGluster, 1);
910     gconf->debug = s->debug;
911     gconf->has_debug = true;
912     gconf->logfile = g_strdup(s->logfile);
913     gconf->has_logfile = true;
914     reop_s->glfs = qemu_gluster_init(gconf, state->bs->filename, NULL, errp);
915     if (reop_s->glfs == NULL) {
916         ret = -errno;
917         goto exit;
918     }
919 
920 #ifdef CONFIG_GLUSTERFS_XLATOR_OPT
921     ret = glfs_set_xlator_option(reop_s->glfs, "*-write-behind",
922                                  "resync-failed-syncs-after-fsync", "on");
923     if (ret < 0) {
924         error_setg_errno(errp, errno, "Unable to set xlator key/value pair");
925         ret = -errno;
926         goto exit;
927     }
928 #endif
929 
930     reop_s->fd = glfs_open(reop_s->glfs, gconf->path, open_flags);
931     if (reop_s->fd == NULL) {
932         /* reops->glfs will be cleaned up in _abort */
933         ret = -errno;
934         goto exit;
935     }
936 
937 exit:
938     /* state->opaque will be freed in either the _abort or _commit */
939     qapi_free_BlockdevOptionsGluster(gconf);
940     return ret;
941 }
942 
943 static void qemu_gluster_reopen_commit(BDRVReopenState *state)
944 {
945     BDRVGlusterReopenState *reop_s = state->opaque;
946     BDRVGlusterState *s = state->bs->opaque;
947 
948 
949     /* close the old */
950     if (s->fd) {
951         glfs_close(s->fd);
952     }
953 
954     glfs_clear_preopened(s->glfs);
955 
956     /* use the newly opened image / connection */
957     s->fd         = reop_s->fd;
958     s->glfs       = reop_s->glfs;
959 
960     g_free(state->opaque);
961     state->opaque = NULL;
962 
963     return;
964 }
965 
966 
967 static void qemu_gluster_reopen_abort(BDRVReopenState *state)
968 {
969     BDRVGlusterReopenState *reop_s = state->opaque;
970 
971     if (reop_s == NULL) {
972         return;
973     }
974 
975     if (reop_s->fd) {
976         glfs_close(reop_s->fd);
977     }
978 
979     glfs_clear_preopened(reop_s->glfs);
980 
981     g_free(state->opaque);
982     state->opaque = NULL;
983 
984     return;
985 }
986 
987 #ifdef CONFIG_GLUSTERFS_ZEROFILL
988 static coroutine_fn int qemu_gluster_co_pwrite_zeroes(BlockDriverState *bs,
989                                                       int64_t offset,
990                                                       int size,
991                                                       BdrvRequestFlags flags)
992 {
993     int ret;
994     GlusterAIOCB acb;
995     BDRVGlusterState *s = bs->opaque;
996 
997     acb.size = size;
998     acb.ret = 0;
999     acb.coroutine = qemu_coroutine_self();
1000     acb.aio_context = bdrv_get_aio_context(bs);
1001 
1002     ret = glfs_zerofill_async(s->fd, offset, size, gluster_finish_aiocb, &acb);
1003     if (ret < 0) {
1004         return -errno;
1005     }
1006 
1007     qemu_coroutine_yield();
1008     return acb.ret;
1009 }
1010 #endif
1011 
1012 static int qemu_gluster_do_truncate(struct glfs_fd *fd, int64_t offset,
1013                                     PreallocMode prealloc, Error **errp)
1014 {
1015     int64_t current_length;
1016 
1017     current_length = glfs_lseek(fd, 0, SEEK_END);
1018     if (current_length < 0) {
1019         error_setg_errno(errp, errno, "Failed to determine current size");
1020         return -errno;
1021     }
1022 
1023     if (current_length > offset && prealloc != PREALLOC_MODE_OFF) {
1024         error_setg(errp, "Cannot use preallocation for shrinking files");
1025         return -ENOTSUP;
1026     }
1027 
1028     if (current_length == offset) {
1029         return 0;
1030     }
1031 
1032     switch (prealloc) {
1033 #ifdef CONFIG_GLUSTERFS_FALLOCATE
1034     case PREALLOC_MODE_FALLOC:
1035         if (glfs_fallocate(fd, 0, current_length, offset - current_length)) {
1036             error_setg_errno(errp, errno, "Could not preallocate data");
1037             return -errno;
1038         }
1039         break;
1040 #endif /* CONFIG_GLUSTERFS_FALLOCATE */
1041 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1042     case PREALLOC_MODE_FULL:
1043         if (glfs_ftruncate(fd, offset)) {
1044             error_setg_errno(errp, errno, "Could not resize file");
1045             return -errno;
1046         }
1047         if (glfs_zerofill(fd, current_length, offset - current_length)) {
1048             error_setg_errno(errp, errno, "Could not zerofill the new area");
1049             return -errno;
1050         }
1051         break;
1052 #endif /* CONFIG_GLUSTERFS_ZEROFILL */
1053     case PREALLOC_MODE_OFF:
1054         if (glfs_ftruncate(fd, offset)) {
1055             error_setg_errno(errp, errno, "Could not resize file");
1056             return -errno;
1057         }
1058         break;
1059     default:
1060         error_setg(errp, "Unsupported preallocation mode: %s",
1061                    PreallocMode_str(prealloc));
1062         return -EINVAL;
1063     }
1064 
1065     return 0;
1066 }
1067 
1068 static int qemu_gluster_co_create(BlockdevCreateOptions *options,
1069                                   Error **errp)
1070 {
1071     BlockdevCreateOptionsGluster *opts = &options->u.gluster;
1072     struct glfs *glfs;
1073     struct glfs_fd *fd = NULL;
1074     int ret = 0;
1075 
1076     assert(options->driver == BLOCKDEV_DRIVER_GLUSTER);
1077 
1078     glfs = qemu_gluster_glfs_init(opts->location, errp);
1079     if (!glfs) {
1080         ret = -errno;
1081         goto out;
1082     }
1083 
1084     fd = glfs_creat(glfs, opts->location->path,
1085                     O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, S_IRUSR | S_IWUSR);
1086     if (!fd) {
1087         ret = -errno;
1088         goto out;
1089     }
1090 
1091     ret = qemu_gluster_do_truncate(fd, opts->size, opts->preallocation, errp);
1092 
1093 out:
1094     if (fd) {
1095         if (glfs_close(fd) != 0 && ret == 0) {
1096             ret = -errno;
1097         }
1098     }
1099     glfs_clear_preopened(glfs);
1100     return ret;
1101 }
1102 
1103 static int coroutine_fn qemu_gluster_co_create_opts(const char *filename,
1104                                                     QemuOpts *opts,
1105                                                     Error **errp)
1106 {
1107     BlockdevCreateOptions *options;
1108     BlockdevCreateOptionsGluster *gopts;
1109     BlockdevOptionsGluster *gconf;
1110     char *tmp = NULL;
1111     Error *local_err = NULL;
1112     int ret;
1113 
1114     options = g_new0(BlockdevCreateOptions, 1);
1115     options->driver = BLOCKDEV_DRIVER_GLUSTER;
1116     gopts = &options->u.gluster;
1117 
1118     gconf = g_new0(BlockdevOptionsGluster, 1);
1119     gopts->location = gconf;
1120 
1121     gopts->size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
1122                            BDRV_SECTOR_SIZE);
1123 
1124     tmp = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
1125     gopts->preallocation = qapi_enum_parse(&PreallocMode_lookup, tmp,
1126                                            PREALLOC_MODE_OFF, &local_err);
1127     g_free(tmp);
1128     if (local_err) {
1129         error_propagate(errp, local_err);
1130         ret = -EINVAL;
1131         goto fail;
1132     }
1133 
1134     gconf->debug = qemu_opt_get_number_del(opts, GLUSTER_OPT_DEBUG,
1135                                            GLUSTER_DEBUG_DEFAULT);
1136     if (gconf->debug < 0) {
1137         gconf->debug = 0;
1138     } else if (gconf->debug > GLUSTER_DEBUG_MAX) {
1139         gconf->debug = GLUSTER_DEBUG_MAX;
1140     }
1141     gconf->has_debug = true;
1142 
1143     gconf->logfile = qemu_opt_get_del(opts, GLUSTER_OPT_LOGFILE);
1144     if (!gconf->logfile) {
1145         gconf->logfile = g_strdup(GLUSTER_LOGFILE_DEFAULT);
1146     }
1147     gconf->has_logfile = true;
1148 
1149     ret = qemu_gluster_parse(gconf, filename, NULL, errp);
1150     if (ret < 0) {
1151         goto fail;
1152     }
1153 
1154     ret = qemu_gluster_co_create(options, errp);
1155     if (ret < 0) {
1156         goto fail;
1157     }
1158 
1159     ret = 0;
1160 fail:
1161     qapi_free_BlockdevCreateOptions(options);
1162     return ret;
1163 }
1164 
1165 static coroutine_fn int qemu_gluster_co_rw(BlockDriverState *bs,
1166                                            int64_t sector_num, int nb_sectors,
1167                                            QEMUIOVector *qiov, int write)
1168 {
1169     int ret;
1170     GlusterAIOCB acb;
1171     BDRVGlusterState *s = bs->opaque;
1172     size_t size = nb_sectors * BDRV_SECTOR_SIZE;
1173     off_t offset = sector_num * BDRV_SECTOR_SIZE;
1174 
1175     acb.size = size;
1176     acb.ret = 0;
1177     acb.coroutine = qemu_coroutine_self();
1178     acb.aio_context = bdrv_get_aio_context(bs);
1179 
1180     if (write) {
1181         ret = glfs_pwritev_async(s->fd, qiov->iov, qiov->niov, offset, 0,
1182                                  gluster_finish_aiocb, &acb);
1183     } else {
1184         ret = glfs_preadv_async(s->fd, qiov->iov, qiov->niov, offset, 0,
1185                                 gluster_finish_aiocb, &acb);
1186     }
1187 
1188     if (ret < 0) {
1189         return -errno;
1190     }
1191 
1192     qemu_coroutine_yield();
1193     return acb.ret;
1194 }
1195 
1196 static coroutine_fn int qemu_gluster_co_truncate(BlockDriverState *bs,
1197                                                  int64_t offset,
1198                                                  PreallocMode prealloc,
1199                                                  Error **errp)
1200 {
1201     BDRVGlusterState *s = bs->opaque;
1202     return qemu_gluster_do_truncate(s->fd, offset, prealloc, errp);
1203 }
1204 
1205 static coroutine_fn int qemu_gluster_co_readv(BlockDriverState *bs,
1206                                               int64_t sector_num,
1207                                               int nb_sectors,
1208                                               QEMUIOVector *qiov)
1209 {
1210     return qemu_gluster_co_rw(bs, sector_num, nb_sectors, qiov, 0);
1211 }
1212 
1213 static coroutine_fn int qemu_gluster_co_writev(BlockDriverState *bs,
1214                                                int64_t sector_num,
1215                                                int nb_sectors,
1216                                                QEMUIOVector *qiov,
1217                                                int flags)
1218 {
1219     assert(!flags);
1220     return qemu_gluster_co_rw(bs, sector_num, nb_sectors, qiov, 1);
1221 }
1222 
1223 static void qemu_gluster_close(BlockDriverState *bs)
1224 {
1225     BDRVGlusterState *s = bs->opaque;
1226 
1227     g_free(s->logfile);
1228     if (s->fd) {
1229         glfs_close(s->fd);
1230         s->fd = NULL;
1231     }
1232     glfs_clear_preopened(s->glfs);
1233 }
1234 
1235 static coroutine_fn int qemu_gluster_co_flush_to_disk(BlockDriverState *bs)
1236 {
1237     int ret;
1238     GlusterAIOCB acb;
1239     BDRVGlusterState *s = bs->opaque;
1240 
1241     acb.size = 0;
1242     acb.ret = 0;
1243     acb.coroutine = qemu_coroutine_self();
1244     acb.aio_context = bdrv_get_aio_context(bs);
1245 
1246     ret = glfs_fsync_async(s->fd, gluster_finish_aiocb, &acb);
1247     if (ret < 0) {
1248         ret = -errno;
1249         goto error;
1250     }
1251 
1252     qemu_coroutine_yield();
1253     if (acb.ret < 0) {
1254         ret = acb.ret;
1255         goto error;
1256     }
1257 
1258     return acb.ret;
1259 
1260 error:
1261     /* Some versions of Gluster (3.5.6 -> 3.5.8?) will not retain its cache
1262      * after a fsync failure, so we have no way of allowing the guest to safely
1263      * continue.  Gluster versions prior to 3.5.6 don't retain the cache
1264      * either, but will invalidate the fd on error, so this is again our only
1265      * option.
1266      *
1267      * The 'resync-failed-syncs-after-fsync' xlator option for the
1268      * write-behind cache will cause later gluster versions to retain its
1269      * cache after error, so long as the fd remains open.  However, we
1270      * currently have no way of knowing if this option is supported.
1271      *
1272      * TODO: Once gluster provides a way for us to determine if the option
1273      * is supported, bypass the closure and setting drv to NULL.  */
1274     qemu_gluster_close(bs);
1275     bs->drv = NULL;
1276     return ret;
1277 }
1278 
1279 #ifdef CONFIG_GLUSTERFS_DISCARD
1280 static coroutine_fn int qemu_gluster_co_pdiscard(BlockDriverState *bs,
1281                                                  int64_t offset, int size)
1282 {
1283     int ret;
1284     GlusterAIOCB acb;
1285     BDRVGlusterState *s = bs->opaque;
1286 
1287     acb.size = 0;
1288     acb.ret = 0;
1289     acb.coroutine = qemu_coroutine_self();
1290     acb.aio_context = bdrv_get_aio_context(bs);
1291 
1292     ret = glfs_discard_async(s->fd, offset, size, gluster_finish_aiocb, &acb);
1293     if (ret < 0) {
1294         return -errno;
1295     }
1296 
1297     qemu_coroutine_yield();
1298     return acb.ret;
1299 }
1300 #endif
1301 
1302 static int64_t qemu_gluster_getlength(BlockDriverState *bs)
1303 {
1304     BDRVGlusterState *s = bs->opaque;
1305     int64_t ret;
1306 
1307     ret = glfs_lseek(s->fd, 0, SEEK_END);
1308     if (ret < 0) {
1309         return -errno;
1310     } else {
1311         return ret;
1312     }
1313 }
1314 
1315 static int64_t qemu_gluster_allocated_file_size(BlockDriverState *bs)
1316 {
1317     BDRVGlusterState *s = bs->opaque;
1318     struct stat st;
1319     int ret;
1320 
1321     ret = glfs_fstat(s->fd, &st);
1322     if (ret < 0) {
1323         return -errno;
1324     } else {
1325         return st.st_blocks * 512;
1326     }
1327 }
1328 
1329 static int qemu_gluster_has_zero_init(BlockDriverState *bs)
1330 {
1331     /* GlusterFS volume could be backed by a block device */
1332     return 0;
1333 }
1334 
1335 /*
1336  * Find allocation range in @bs around offset @start.
1337  * May change underlying file descriptor's file offset.
1338  * If @start is not in a hole, store @start in @data, and the
1339  * beginning of the next hole in @hole, and return 0.
1340  * If @start is in a non-trailing hole, store @start in @hole and the
1341  * beginning of the next non-hole in @data, and return 0.
1342  * If @start is in a trailing hole or beyond EOF, return -ENXIO.
1343  * If we can't find out, return a negative errno other than -ENXIO.
1344  *
1345  * (Shamefully copied from file-posix.c, only minuscule adaptions.)
1346  */
1347 static int find_allocation(BlockDriverState *bs, off_t start,
1348                            off_t *data, off_t *hole)
1349 {
1350     BDRVGlusterState *s = bs->opaque;
1351 
1352     if (!s->supports_seek_data) {
1353         goto exit;
1354     }
1355 
1356 #if defined SEEK_HOLE && defined SEEK_DATA
1357     off_t offs;
1358 
1359     /*
1360      * SEEK_DATA cases:
1361      * D1. offs == start: start is in data
1362      * D2. offs > start: start is in a hole, next data at offs
1363      * D3. offs < 0, errno = ENXIO: either start is in a trailing hole
1364      *                              or start is beyond EOF
1365      *     If the latter happens, the file has been truncated behind
1366      *     our back since we opened it.  All bets are off then.
1367      *     Treating like a trailing hole is simplest.
1368      * D4. offs < 0, errno != ENXIO: we learned nothing
1369      */
1370     offs = glfs_lseek(s->fd, start, SEEK_DATA);
1371     if (offs < 0) {
1372         return -errno;          /* D3 or D4 */
1373     }
1374 
1375     if (offs < start) {
1376         /* This is not a valid return by lseek().  We are safe to just return
1377          * -EIO in this case, and we'll treat it like D4. Unfortunately some
1378          *  versions of gluster server will return offs < start, so an assert
1379          *  here will unnecessarily abort QEMU. */
1380         return -EIO;
1381     }
1382 
1383     if (offs > start) {
1384         /* D2: in hole, next data at offs */
1385         *hole = start;
1386         *data = offs;
1387         return 0;
1388     }
1389 
1390     /* D1: in data, end not yet known */
1391 
1392     /*
1393      * SEEK_HOLE cases:
1394      * H1. offs == start: start is in a hole
1395      *     If this happens here, a hole has been dug behind our back
1396      *     since the previous lseek().
1397      * H2. offs > start: either start is in data, next hole at offs,
1398      *                   or start is in trailing hole, EOF at offs
1399      *     Linux treats trailing holes like any other hole: offs ==
1400      *     start.  Solaris seeks to EOF instead: offs > start (blech).
1401      *     If that happens here, a hole has been dug behind our back
1402      *     since the previous lseek().
1403      * H3. offs < 0, errno = ENXIO: start is beyond EOF
1404      *     If this happens, the file has been truncated behind our
1405      *     back since we opened it.  Treat it like a trailing hole.
1406      * H4. offs < 0, errno != ENXIO: we learned nothing
1407      *     Pretend we know nothing at all, i.e. "forget" about D1.
1408      */
1409     offs = glfs_lseek(s->fd, start, SEEK_HOLE);
1410     if (offs < 0) {
1411         return -errno;          /* D1 and (H3 or H4) */
1412     }
1413 
1414     if (offs < start) {
1415         /* This is not a valid return by lseek().  We are safe to just return
1416          * -EIO in this case, and we'll treat it like H4. Unfortunately some
1417          *  versions of gluster server will return offs < start, so an assert
1418          *  here will unnecessarily abort QEMU. */
1419         return -EIO;
1420     }
1421 
1422     if (offs > start) {
1423         /*
1424          * D1 and H2: either in data, next hole at offs, or it was in
1425          * data but is now in a trailing hole.  In the latter case,
1426          * all bets are off.  Treating it as if it there was data all
1427          * the way to EOF is safe, so simply do that.
1428          */
1429         *data = start;
1430         *hole = offs;
1431         return 0;
1432     }
1433 
1434     /* D1 and H1 */
1435     return -EBUSY;
1436 #endif
1437 
1438 exit:
1439     return -ENOTSUP;
1440 }
1441 
1442 /*
1443  * Returns the allocation status of the specified offset.
1444  *
1445  * The block layer guarantees 'offset' and 'bytes' are within bounds.
1446  *
1447  * 'pnum' is set to the number of bytes (including and immediately following
1448  * the specified offset) that are known to be in the same
1449  * allocated/unallocated state.
1450  *
1451  * 'bytes' is the max value 'pnum' should be set to.
1452  *
1453  * (Based on raw_co_block_status() from file-posix.c.)
1454  */
1455 static int coroutine_fn qemu_gluster_co_block_status(BlockDriverState *bs,
1456                                                      bool want_zero,
1457                                                      int64_t offset,
1458                                                      int64_t bytes,
1459                                                      int64_t *pnum,
1460                                                      int64_t *map,
1461                                                      BlockDriverState **file)
1462 {
1463     BDRVGlusterState *s = bs->opaque;
1464     off_t data = 0, hole = 0;
1465     int ret = -EINVAL;
1466 
1467     if (!s->fd) {
1468         return ret;
1469     }
1470 
1471     if (!want_zero) {
1472         *pnum = bytes;
1473         *map = offset;
1474         *file = bs;
1475         return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID;
1476     }
1477 
1478     ret = find_allocation(bs, offset, &data, &hole);
1479     if (ret == -ENXIO) {
1480         /* Trailing hole */
1481         *pnum = bytes;
1482         ret = BDRV_BLOCK_ZERO;
1483     } else if (ret < 0) {
1484         /* No info available, so pretend there are no holes */
1485         *pnum = bytes;
1486         ret = BDRV_BLOCK_DATA;
1487     } else if (data == offset) {
1488         /* On a data extent, compute bytes to the end of the extent,
1489          * possibly including a partial sector at EOF. */
1490         *pnum = MIN(bytes, hole - offset);
1491         ret = BDRV_BLOCK_DATA;
1492     } else {
1493         /* On a hole, compute bytes to the beginning of the next extent.  */
1494         assert(hole == offset);
1495         *pnum = MIN(bytes, data - offset);
1496         ret = BDRV_BLOCK_ZERO;
1497     }
1498 
1499     *map = offset;
1500     *file = bs;
1501 
1502     return ret | BDRV_BLOCK_OFFSET_VALID;
1503 }
1504 
1505 
1506 static const char *const gluster_strong_open_opts[] = {
1507     GLUSTER_OPT_VOLUME,
1508     GLUSTER_OPT_PATH,
1509     GLUSTER_OPT_TYPE,
1510     GLUSTER_OPT_SERVER_PATTERN,
1511     GLUSTER_OPT_HOST,
1512     GLUSTER_OPT_PORT,
1513     GLUSTER_OPT_TO,
1514     GLUSTER_OPT_IPV4,
1515     GLUSTER_OPT_IPV6,
1516     GLUSTER_OPT_SOCKET,
1517 
1518     NULL
1519 };
1520 
1521 static BlockDriver bdrv_gluster = {
1522     .format_name                  = "gluster",
1523     .protocol_name                = "gluster",
1524     .instance_size                = sizeof(BDRVGlusterState),
1525     .bdrv_needs_filename          = false,
1526     .bdrv_file_open               = qemu_gluster_open,
1527     .bdrv_reopen_prepare          = qemu_gluster_reopen_prepare,
1528     .bdrv_reopen_commit           = qemu_gluster_reopen_commit,
1529     .bdrv_reopen_abort            = qemu_gluster_reopen_abort,
1530     .bdrv_close                   = qemu_gluster_close,
1531     .bdrv_co_create               = qemu_gluster_co_create,
1532     .bdrv_co_create_opts          = qemu_gluster_co_create_opts,
1533     .bdrv_getlength               = qemu_gluster_getlength,
1534     .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size,
1535     .bdrv_co_truncate             = qemu_gluster_co_truncate,
1536     .bdrv_co_readv                = qemu_gluster_co_readv,
1537     .bdrv_co_writev               = qemu_gluster_co_writev,
1538     .bdrv_co_flush_to_disk        = qemu_gluster_co_flush_to_disk,
1539     .bdrv_has_zero_init           = qemu_gluster_has_zero_init,
1540 #ifdef CONFIG_GLUSTERFS_DISCARD
1541     .bdrv_co_pdiscard             = qemu_gluster_co_pdiscard,
1542 #endif
1543 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1544     .bdrv_co_pwrite_zeroes        = qemu_gluster_co_pwrite_zeroes,
1545 #endif
1546     .bdrv_co_block_status         = qemu_gluster_co_block_status,
1547     .create_opts                  = &qemu_gluster_create_opts,
1548     .strong_runtime_opts          = gluster_strong_open_opts,
1549 };
1550 
1551 static BlockDriver bdrv_gluster_tcp = {
1552     .format_name                  = "gluster",
1553     .protocol_name                = "gluster+tcp",
1554     .instance_size                = sizeof(BDRVGlusterState),
1555     .bdrv_needs_filename          = false,
1556     .bdrv_file_open               = qemu_gluster_open,
1557     .bdrv_reopen_prepare          = qemu_gluster_reopen_prepare,
1558     .bdrv_reopen_commit           = qemu_gluster_reopen_commit,
1559     .bdrv_reopen_abort            = qemu_gluster_reopen_abort,
1560     .bdrv_close                   = qemu_gluster_close,
1561     .bdrv_co_create               = qemu_gluster_co_create,
1562     .bdrv_co_create_opts          = qemu_gluster_co_create_opts,
1563     .bdrv_getlength               = qemu_gluster_getlength,
1564     .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size,
1565     .bdrv_co_truncate             = qemu_gluster_co_truncate,
1566     .bdrv_co_readv                = qemu_gluster_co_readv,
1567     .bdrv_co_writev               = qemu_gluster_co_writev,
1568     .bdrv_co_flush_to_disk        = qemu_gluster_co_flush_to_disk,
1569     .bdrv_has_zero_init           = qemu_gluster_has_zero_init,
1570 #ifdef CONFIG_GLUSTERFS_DISCARD
1571     .bdrv_co_pdiscard             = qemu_gluster_co_pdiscard,
1572 #endif
1573 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1574     .bdrv_co_pwrite_zeroes        = qemu_gluster_co_pwrite_zeroes,
1575 #endif
1576     .bdrv_co_block_status         = qemu_gluster_co_block_status,
1577     .create_opts                  = &qemu_gluster_create_opts,
1578     .strong_runtime_opts          = gluster_strong_open_opts,
1579 };
1580 
1581 static BlockDriver bdrv_gluster_unix = {
1582     .format_name                  = "gluster",
1583     .protocol_name                = "gluster+unix",
1584     .instance_size                = sizeof(BDRVGlusterState),
1585     .bdrv_needs_filename          = true,
1586     .bdrv_file_open               = qemu_gluster_open,
1587     .bdrv_reopen_prepare          = qemu_gluster_reopen_prepare,
1588     .bdrv_reopen_commit           = qemu_gluster_reopen_commit,
1589     .bdrv_reopen_abort            = qemu_gluster_reopen_abort,
1590     .bdrv_close                   = qemu_gluster_close,
1591     .bdrv_co_create               = qemu_gluster_co_create,
1592     .bdrv_co_create_opts          = qemu_gluster_co_create_opts,
1593     .bdrv_getlength               = qemu_gluster_getlength,
1594     .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size,
1595     .bdrv_co_truncate             = qemu_gluster_co_truncate,
1596     .bdrv_co_readv                = qemu_gluster_co_readv,
1597     .bdrv_co_writev               = qemu_gluster_co_writev,
1598     .bdrv_co_flush_to_disk        = qemu_gluster_co_flush_to_disk,
1599     .bdrv_has_zero_init           = qemu_gluster_has_zero_init,
1600 #ifdef CONFIG_GLUSTERFS_DISCARD
1601     .bdrv_co_pdiscard             = qemu_gluster_co_pdiscard,
1602 #endif
1603 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1604     .bdrv_co_pwrite_zeroes        = qemu_gluster_co_pwrite_zeroes,
1605 #endif
1606     .bdrv_co_block_status         = qemu_gluster_co_block_status,
1607     .create_opts                  = &qemu_gluster_create_opts,
1608     .strong_runtime_opts          = gluster_strong_open_opts,
1609 };
1610 
1611 /* rdma is deprecated (actually never supported for volfile fetch).
1612  * Let's maintain it for the protocol compatibility, to make sure things
1613  * won't break immediately. For now, gluster+rdma will fall back to gluster+tcp
1614  * protocol with a warning.
1615  * TODO: remove gluster+rdma interface support
1616  */
1617 static BlockDriver bdrv_gluster_rdma = {
1618     .format_name                  = "gluster",
1619     .protocol_name                = "gluster+rdma",
1620     .instance_size                = sizeof(BDRVGlusterState),
1621     .bdrv_needs_filename          = true,
1622     .bdrv_file_open               = qemu_gluster_open,
1623     .bdrv_reopen_prepare          = qemu_gluster_reopen_prepare,
1624     .bdrv_reopen_commit           = qemu_gluster_reopen_commit,
1625     .bdrv_reopen_abort            = qemu_gluster_reopen_abort,
1626     .bdrv_close                   = qemu_gluster_close,
1627     .bdrv_co_create               = qemu_gluster_co_create,
1628     .bdrv_co_create_opts          = qemu_gluster_co_create_opts,
1629     .bdrv_getlength               = qemu_gluster_getlength,
1630     .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size,
1631     .bdrv_co_truncate             = qemu_gluster_co_truncate,
1632     .bdrv_co_readv                = qemu_gluster_co_readv,
1633     .bdrv_co_writev               = qemu_gluster_co_writev,
1634     .bdrv_co_flush_to_disk        = qemu_gluster_co_flush_to_disk,
1635     .bdrv_has_zero_init           = qemu_gluster_has_zero_init,
1636 #ifdef CONFIG_GLUSTERFS_DISCARD
1637     .bdrv_co_pdiscard             = qemu_gluster_co_pdiscard,
1638 #endif
1639 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1640     .bdrv_co_pwrite_zeroes        = qemu_gluster_co_pwrite_zeroes,
1641 #endif
1642     .bdrv_co_block_status         = qemu_gluster_co_block_status,
1643     .create_opts                  = &qemu_gluster_create_opts,
1644     .strong_runtime_opts          = gluster_strong_open_opts,
1645 };
1646 
1647 static void bdrv_gluster_init(void)
1648 {
1649     bdrv_register(&bdrv_gluster_rdma);
1650     bdrv_register(&bdrv_gluster_unix);
1651     bdrv_register(&bdrv_gluster_tcp);
1652     bdrv_register(&bdrv_gluster);
1653 }
1654 
1655 block_init(bdrv_gluster_init);
1656