xref: /openbmc/qemu/block/gluster.c (revision 109f1a43)
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 "qemu/units.h"
13 #include <glusterfs/api/glfs.h>
14 #include "block/block-io.h"
15 #include "block/block_int.h"
16 #include "block/qdict.h"
17 #include "qapi/error.h"
18 #include "qapi/qmp/qdict.h"
19 #include "qapi/qmp/qerror.h"
20 #include "qemu/uri.h"
21 #include "qemu/error-report.h"
22 #include "qemu/module.h"
23 #include "qemu/option.h"
24 #include "qemu/cutils.h"
25 
26 #ifdef CONFIG_GLUSTERFS_FTRUNCATE_HAS_STAT
27 # define glfs_ftruncate(fd, offset) glfs_ftruncate(fd, offset, NULL, NULL)
28 #endif
29 
30 #define GLUSTER_OPT_FILENAME        "filename"
31 #define GLUSTER_OPT_VOLUME          "volume"
32 #define GLUSTER_OPT_PATH            "path"
33 #define GLUSTER_OPT_TYPE            "type"
34 #define GLUSTER_OPT_SERVER_PATTERN  "server."
35 #define GLUSTER_OPT_HOST            "host"
36 #define GLUSTER_OPT_PORT            "port"
37 #define GLUSTER_OPT_TO              "to"
38 #define GLUSTER_OPT_IPV4            "ipv4"
39 #define GLUSTER_OPT_IPV6            "ipv6"
40 #define GLUSTER_OPT_SOCKET          "socket"
41 #define GLUSTER_OPT_DEBUG           "debug"
42 #define GLUSTER_DEFAULT_PORT        24007
43 #define GLUSTER_DEBUG_DEFAULT       4
44 #define GLUSTER_DEBUG_MAX           9
45 #define GLUSTER_OPT_LOGFILE         "logfile"
46 #define GLUSTER_LOGFILE_DEFAULT     "-" /* handled in libgfapi as /dev/stderr */
47 /*
48  * Several versions of GlusterFS (3.12? -> 6.0.1) fail when the transfer size
49  * is greater or equal to 1024 MiB, so we are limiting the transfer size to 512
50  * MiB to avoid this rare issue.
51  */
52 #define GLUSTER_MAX_TRANSFER        (512 * MiB)
53 
54 #define GERR_INDEX_HINT "hint: check in 'server' array index '%d'\n"
55 
56 typedef struct GlusterAIOCB {
57     int64_t size;
58     int ret;
59     Coroutine *coroutine;
60     AioContext *aio_context;
61 } GlusterAIOCB;
62 
63 typedef struct BDRVGlusterState {
64     struct glfs *glfs;
65     struct glfs_fd *fd;
66     char *logfile;
67     bool supports_seek_data;
68     int debug;
69 } BDRVGlusterState;
70 
71 typedef struct BDRVGlusterReopenState {
72     struct glfs *glfs;
73     struct glfs_fd *fd;
74 } BDRVGlusterReopenState;
75 
76 
77 typedef struct GlfsPreopened {
78     char *volume;
79     glfs_t *fs;
80     int ref;
81 } GlfsPreopened;
82 
83 typedef struct ListElement {
84     QLIST_ENTRY(ListElement) list;
85     GlfsPreopened saved;
86 } ListElement;
87 
88 static QLIST_HEAD(, ListElement) glfs_list;
89 
90 static QemuOptsList qemu_gluster_create_opts = {
91     .name = "qemu-gluster-create-opts",
92     .head = QTAILQ_HEAD_INITIALIZER(qemu_gluster_create_opts.head),
93     .desc = {
94         {
95             .name = BLOCK_OPT_SIZE,
96             .type = QEMU_OPT_SIZE,
97             .help = "Virtual disk size"
98         },
99         {
100             .name = BLOCK_OPT_PREALLOC,
101             .type = QEMU_OPT_STRING,
102             .help = "Preallocation mode (allowed values: off"
103 #ifdef CONFIG_GLUSTERFS_FALLOCATE
104                     ", falloc"
105 #endif
106 #ifdef CONFIG_GLUSTERFS_ZEROFILL
107                     ", full"
108 #endif
109                     ")"
110         },
111         {
112             .name = GLUSTER_OPT_DEBUG,
113             .type = QEMU_OPT_NUMBER,
114             .help = "Gluster log level, valid range is 0-9",
115         },
116         {
117             .name = GLUSTER_OPT_LOGFILE,
118             .type = QEMU_OPT_STRING,
119             .help = "Logfile path of libgfapi",
120         },
121         { /* end of list */ }
122     }
123 };
124 
125 static QemuOptsList runtime_opts = {
126     .name = "gluster",
127     .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
128     .desc = {
129         {
130             .name = GLUSTER_OPT_FILENAME,
131             .type = QEMU_OPT_STRING,
132             .help = "URL to the gluster image",
133         },
134         {
135             .name = GLUSTER_OPT_DEBUG,
136             .type = QEMU_OPT_NUMBER,
137             .help = "Gluster log level, valid range is 0-9",
138         },
139         {
140             .name = GLUSTER_OPT_LOGFILE,
141             .type = QEMU_OPT_STRING,
142             .help = "Logfile path of libgfapi",
143         },
144         { /* end of list */ }
145     },
146 };
147 
148 static QemuOptsList runtime_json_opts = {
149     .name = "gluster_json",
150     .head = QTAILQ_HEAD_INITIALIZER(runtime_json_opts.head),
151     .desc = {
152         {
153             .name = GLUSTER_OPT_VOLUME,
154             .type = QEMU_OPT_STRING,
155             .help = "name of gluster volume where VM image resides",
156         },
157         {
158             .name = GLUSTER_OPT_PATH,
159             .type = QEMU_OPT_STRING,
160             .help = "absolute path to image file in gluster volume",
161         },
162         {
163             .name = GLUSTER_OPT_DEBUG,
164             .type = QEMU_OPT_NUMBER,
165             .help = "Gluster log level, valid range is 0-9",
166         },
167         { /* end of list */ }
168     },
169 };
170 
171 static QemuOptsList runtime_type_opts = {
172     .name = "gluster_type",
173     .head = QTAILQ_HEAD_INITIALIZER(runtime_type_opts.head),
174     .desc = {
175         {
176             .name = GLUSTER_OPT_TYPE,
177             .type = QEMU_OPT_STRING,
178             .help = "inet|unix",
179         },
180         { /* end of list */ }
181     },
182 };
183 
184 static QemuOptsList runtime_unix_opts = {
185     .name = "gluster_unix",
186     .head = QTAILQ_HEAD_INITIALIZER(runtime_unix_opts.head),
187     .desc = {
188         {
189             .name = GLUSTER_OPT_SOCKET,
190             .type = QEMU_OPT_STRING,
191             .help = "socket file path (legacy)",
192         },
193         {
194             .name = GLUSTER_OPT_PATH,
195             .type = QEMU_OPT_STRING,
196             .help = "socket file path (QAPI)",
197         },
198         { /* end of list */ }
199     },
200 };
201 
202 static QemuOptsList runtime_inet_opts = {
203     .name = "gluster_inet",
204     .head = QTAILQ_HEAD_INITIALIZER(runtime_inet_opts.head),
205     .desc = {
206         {
207             .name = GLUSTER_OPT_TYPE,
208             .type = QEMU_OPT_STRING,
209             .help = "inet|unix",
210         },
211         {
212             .name = GLUSTER_OPT_HOST,
213             .type = QEMU_OPT_STRING,
214             .help = "host address (hostname/ipv4/ipv6 addresses)",
215         },
216         {
217             .name = GLUSTER_OPT_PORT,
218             .type = QEMU_OPT_STRING,
219             .help = "port number on which glusterd is listening (default 24007)",
220         },
221         {
222             .name = "to",
223             .type = QEMU_OPT_NUMBER,
224             .help = "max port number, not supported by gluster",
225         },
226         {
227             .name = "ipv4",
228             .type = QEMU_OPT_BOOL,
229             .help = "ipv4 bool value, not supported by gluster",
230         },
231         {
232             .name = "ipv6",
233             .type = QEMU_OPT_BOOL,
234             .help = "ipv6 bool value, not supported by gluster",
235         },
236         { /* end of list */ }
237     },
238 };
239 
240 static void glfs_set_preopened(const char *volume, glfs_t *fs)
241 {
242     ListElement *entry = NULL;
243 
244     entry = g_new(ListElement, 1);
245 
246     entry->saved.volume = g_strdup(volume);
247 
248     entry->saved.fs = fs;
249     entry->saved.ref = 1;
250 
251     QLIST_INSERT_HEAD(&glfs_list, entry, list);
252 }
253 
254 static glfs_t *glfs_find_preopened(const char *volume)
255 {
256     ListElement *entry = NULL;
257 
258      QLIST_FOREACH(entry, &glfs_list, list) {
259         if (strcmp(entry->saved.volume, volume) == 0) {
260             entry->saved.ref++;
261             return entry->saved.fs;
262         }
263      }
264 
265     return NULL;
266 }
267 
268 static void glfs_clear_preopened(glfs_t *fs)
269 {
270     ListElement *entry = NULL;
271     ListElement *next;
272 
273     if (fs == NULL) {
274         return;
275     }
276 
277     QLIST_FOREACH_SAFE(entry, &glfs_list, list, next) {
278         if (entry->saved.fs == fs) {
279             if (--entry->saved.ref) {
280                 return;
281             }
282 
283             QLIST_REMOVE(entry, list);
284 
285             glfs_fini(entry->saved.fs);
286             g_free(entry->saved.volume);
287             g_free(entry);
288         }
289     }
290 }
291 
292 static int parse_volume_options(BlockdevOptionsGluster *gconf, char *path)
293 {
294     char *p, *q;
295 
296     if (!path) {
297         return -EINVAL;
298     }
299 
300     /* volume */
301     p = q = path + strspn(path, "/");
302     p += strcspn(p, "/");
303     if (*p == '\0') {
304         return -EINVAL;
305     }
306     gconf->volume = g_strndup(q, p - q);
307 
308     /* path */
309     p += strspn(p, "/");
310     if (*p == '\0') {
311         return -EINVAL;
312     }
313     gconf->path = g_strdup(p);
314     return 0;
315 }
316 
317 /*
318  * file=gluster[+transport]://[host[:port]]/volume/path[?socket=...]
319  *
320  * 'gluster' is the protocol.
321  *
322  * 'transport' specifies the transport type used to connect to gluster
323  * management daemon (glusterd). Valid transport types are
324  * tcp or unix. If a transport type isn't specified, then tcp type is assumed.
325  *
326  * 'host' specifies the host where the volume file specification for
327  * the given volume resides. This can be either hostname or ipv4 address.
328  * If transport type is 'unix', then 'host' field should not be specified.
329  * The 'socket' field needs to be populated with the path to unix domain
330  * socket.
331  *
332  * 'port' is the port number on which glusterd is listening. This is optional
333  * and if not specified, QEMU will send 0 which will make gluster to use the
334  * default port. If the transport type is unix, then 'port' should not be
335  * specified.
336  *
337  * 'volume' is the name of the gluster volume which contains the VM image.
338  *
339  * 'path' is the path to the actual VM image that resides on gluster volume.
340  *
341  * Examples:
342  *
343  * file=gluster://1.2.3.4/testvol/a.img
344  * file=gluster+tcp://1.2.3.4/testvol/a.img
345  * file=gluster+tcp://1.2.3.4:24007/testvol/dir/a.img
346  * file=gluster+tcp://host.domain.com:24007/testvol/dir/a.img
347  * file=gluster+unix:///testvol/dir/a.img?socket=/tmp/glusterd.socket
348  */
349 static int qemu_gluster_parse_uri(BlockdevOptionsGluster *gconf,
350                                   const char *filename)
351 {
352     SocketAddress *gsconf;
353     URI *uri;
354     QueryParams *qp = NULL;
355     bool is_unix = false;
356     int ret = 0;
357 
358     uri = uri_parse(filename);
359     if (!uri) {
360         return -EINVAL;
361     }
362 
363     gsconf = g_new0(SocketAddress, 1);
364     QAPI_LIST_PREPEND(gconf->server, gsconf);
365 
366     /* transport */
367     if (!uri->scheme || !strcmp(uri->scheme, "gluster")) {
368         gsconf->type = SOCKET_ADDRESS_TYPE_INET;
369     } else if (!strcmp(uri->scheme, "gluster+tcp")) {
370         gsconf->type = SOCKET_ADDRESS_TYPE_INET;
371     } else if (!strcmp(uri->scheme, "gluster+unix")) {
372         gsconf->type = SOCKET_ADDRESS_TYPE_UNIX;
373         is_unix = true;
374     } else {
375         ret = -EINVAL;
376         goto out;
377     }
378 
379     ret = parse_volume_options(gconf, uri->path);
380     if (ret < 0) {
381         goto out;
382     }
383 
384     qp = query_params_parse(uri->query);
385     if (qp->n > 1 || (is_unix && !qp->n) || (!is_unix && qp->n)) {
386         ret = -EINVAL;
387         goto out;
388     }
389 
390     if (is_unix) {
391         if (uri->server || uri->port) {
392             ret = -EINVAL;
393             goto out;
394         }
395         if (strcmp(qp->p[0].name, "socket")) {
396             ret = -EINVAL;
397             goto out;
398         }
399         gsconf->u.q_unix.path = g_strdup(qp->p[0].value);
400     } else {
401         gsconf->u.inet.host = g_strdup(uri->server ? uri->server : "localhost");
402         if (uri->port) {
403             gsconf->u.inet.port = g_strdup_printf("%d", uri->port);
404         } else {
405             gsconf->u.inet.port = g_strdup_printf("%d", GLUSTER_DEFAULT_PORT);
406         }
407     }
408 
409 out:
410     if (qp) {
411         query_params_free(qp);
412     }
413     uri_free(uri);
414     return ret;
415 }
416 
417 static struct glfs *qemu_gluster_glfs_init(BlockdevOptionsGluster *gconf,
418                                            Error **errp)
419 {
420     struct glfs *glfs;
421     int ret;
422     int old_errno;
423     SocketAddressList *server;
424     uint64_t port;
425 
426     glfs = glfs_find_preopened(gconf->volume);
427     if (glfs) {
428         return glfs;
429     }
430 
431     glfs = glfs_new(gconf->volume);
432     if (!glfs) {
433         goto out;
434     }
435 
436     glfs_set_preopened(gconf->volume, glfs);
437 
438     for (server = gconf->server; server; server = server->next) {
439         switch (server->value->type) {
440         case SOCKET_ADDRESS_TYPE_UNIX:
441             ret = glfs_set_volfile_server(glfs, "unix",
442                                    server->value->u.q_unix.path, 0);
443             break;
444         case SOCKET_ADDRESS_TYPE_INET:
445             if (parse_uint_full(server->value->u.inet.port, 10, &port) < 0 ||
446                 port > 65535) {
447                 error_setg(errp, "'%s' is not a valid port number",
448                            server->value->u.inet.port);
449                 errno = EINVAL;
450                 goto out;
451             }
452             ret = glfs_set_volfile_server(glfs, "tcp",
453                                    server->value->u.inet.host,
454                                    (int)port);
455             break;
456         case SOCKET_ADDRESS_TYPE_VSOCK:
457         case SOCKET_ADDRESS_TYPE_FD:
458         default:
459             abort();
460         }
461 
462         if (ret < 0) {
463             goto out;
464         }
465     }
466 
467     ret = glfs_set_logging(glfs, gconf->logfile, gconf->debug);
468     if (ret < 0) {
469         goto out;
470     }
471 
472     ret = glfs_init(glfs);
473     if (ret) {
474         error_setg(errp, "Gluster connection for volume %s, path %s failed"
475                          " to connect", gconf->volume, gconf->path);
476         for (server = gconf->server; server; server = server->next) {
477             if (server->value->type  == SOCKET_ADDRESS_TYPE_UNIX) {
478                 error_append_hint(errp, "hint: failed on socket %s ",
479                                   server->value->u.q_unix.path);
480             } else {
481                 error_append_hint(errp, "hint: failed on host %s and port %s ",
482                                   server->value->u.inet.host,
483                                   server->value->u.inet.port);
484             }
485         }
486 
487         error_append_hint(errp, "Please refer to gluster logs for more info\n");
488 
489         /* glfs_init sometimes doesn't set errno although docs suggest that */
490         if (errno == 0) {
491             errno = EINVAL;
492         }
493 
494         goto out;
495     }
496     return glfs;
497 
498 out:
499     if (glfs) {
500         old_errno = errno;
501         glfs_clear_preopened(glfs);
502         errno = old_errno;
503     }
504     return NULL;
505 }
506 
507 /*
508  * Convert the json formatted command line into qapi.
509 */
510 static int qemu_gluster_parse_json(BlockdevOptionsGluster *gconf,
511                                   QDict *options, Error **errp)
512 {
513     QemuOpts *opts;
514     SocketAddress *gsconf = NULL;
515     SocketAddressList **tail;
516     QDict *backing_options = NULL;
517     Error *local_err = NULL;
518     char *str = NULL;
519     const char *ptr;
520     int i, type, num_servers;
521 
522     /* create opts info from runtime_json_opts list */
523     opts = qemu_opts_create(&runtime_json_opts, NULL, 0, &error_abort);
524     if (!qemu_opts_absorb_qdict(opts, options, errp)) {
525         goto out;
526     }
527 
528     num_servers = qdict_array_entries(options, GLUSTER_OPT_SERVER_PATTERN);
529     if (num_servers < 1) {
530         error_setg(&local_err, QERR_MISSING_PARAMETER, "server");
531         goto out;
532     }
533 
534     ptr = qemu_opt_get(opts, GLUSTER_OPT_VOLUME);
535     if (!ptr) {
536         error_setg(&local_err, QERR_MISSING_PARAMETER, GLUSTER_OPT_VOLUME);
537         goto out;
538     }
539     gconf->volume = g_strdup(ptr);
540 
541     ptr = qemu_opt_get(opts, GLUSTER_OPT_PATH);
542     if (!ptr) {
543         error_setg(&local_err, QERR_MISSING_PARAMETER, GLUSTER_OPT_PATH);
544         goto out;
545     }
546     gconf->path = g_strdup(ptr);
547     qemu_opts_del(opts);
548     tail = &gconf->server;
549 
550     for (i = 0; i < num_servers; i++) {
551         str = g_strdup_printf(GLUSTER_OPT_SERVER_PATTERN"%d.", i);
552         qdict_extract_subqdict(options, &backing_options, str);
553 
554         /* create opts info from runtime_type_opts list */
555         opts = qemu_opts_create(&runtime_type_opts, NULL, 0, &error_abort);
556         if (!qemu_opts_absorb_qdict(opts, backing_options, errp)) {
557             goto out;
558         }
559 
560         ptr = qemu_opt_get(opts, GLUSTER_OPT_TYPE);
561         if (!ptr) {
562             error_setg(&local_err, QERR_MISSING_PARAMETER, GLUSTER_OPT_TYPE);
563             error_append_hint(&local_err, GERR_INDEX_HINT, i);
564             goto out;
565 
566         }
567         gsconf = g_new0(SocketAddress, 1);
568         if (!strcmp(ptr, "tcp")) {
569             ptr = "inet";       /* accept legacy "tcp" */
570         }
571         type = qapi_enum_parse(&SocketAddressType_lookup, ptr, -1, NULL);
572         if (type != SOCKET_ADDRESS_TYPE_INET
573             && type != SOCKET_ADDRESS_TYPE_UNIX) {
574             error_setg(&local_err,
575                        "Parameter '%s' may be 'inet' or 'unix'",
576                        GLUSTER_OPT_TYPE);
577             error_append_hint(&local_err, GERR_INDEX_HINT, i);
578             goto out;
579         }
580         gsconf->type = type;
581         qemu_opts_del(opts);
582 
583         if (gsconf->type == SOCKET_ADDRESS_TYPE_INET) {
584             /* create opts info from runtime_inet_opts list */
585             opts = qemu_opts_create(&runtime_inet_opts, NULL, 0, &error_abort);
586             if (!qemu_opts_absorb_qdict(opts, backing_options, errp)) {
587                 goto out;
588             }
589 
590             ptr = qemu_opt_get(opts, GLUSTER_OPT_HOST);
591             if (!ptr) {
592                 error_setg(&local_err, QERR_MISSING_PARAMETER,
593                            GLUSTER_OPT_HOST);
594                 error_append_hint(&local_err, GERR_INDEX_HINT, i);
595                 goto out;
596             }
597             gsconf->u.inet.host = g_strdup(ptr);
598             ptr = qemu_opt_get(opts, GLUSTER_OPT_PORT);
599             if (!ptr) {
600                 error_setg(&local_err, QERR_MISSING_PARAMETER,
601                            GLUSTER_OPT_PORT);
602                 error_append_hint(&local_err, GERR_INDEX_HINT, i);
603                 goto out;
604             }
605             gsconf->u.inet.port = g_strdup(ptr);
606 
607             /* defend for unsupported fields in InetSocketAddress,
608              * i.e. @ipv4, @ipv6  and @to
609              */
610             ptr = qemu_opt_get(opts, GLUSTER_OPT_TO);
611             if (ptr) {
612                 gsconf->u.inet.has_to = true;
613             }
614             ptr = qemu_opt_get(opts, GLUSTER_OPT_IPV4);
615             if (ptr) {
616                 gsconf->u.inet.has_ipv4 = true;
617             }
618             ptr = qemu_opt_get(opts, GLUSTER_OPT_IPV6);
619             if (ptr) {
620                 gsconf->u.inet.has_ipv6 = true;
621             }
622             if (gsconf->u.inet.has_to) {
623                 error_setg(&local_err, "Parameter 'to' not supported");
624                 goto out;
625             }
626             if (gsconf->u.inet.has_ipv4 || gsconf->u.inet.has_ipv6) {
627                 error_setg(&local_err, "Parameters 'ipv4/ipv6' not supported");
628                 goto out;
629             }
630             qemu_opts_del(opts);
631         } else {
632             /* create opts info from runtime_unix_opts list */
633             opts = qemu_opts_create(&runtime_unix_opts, NULL, 0, &error_abort);
634             if (!qemu_opts_absorb_qdict(opts, backing_options, errp)) {
635                 goto out;
636             }
637 
638             ptr = qemu_opt_get(opts, GLUSTER_OPT_PATH);
639             if (!ptr) {
640                 ptr = qemu_opt_get(opts, GLUSTER_OPT_SOCKET);
641             } else if (qemu_opt_get(opts, GLUSTER_OPT_SOCKET)) {
642                 error_setg(&local_err,
643                            "Conflicting parameters 'path' and 'socket'");
644                 error_append_hint(&local_err, GERR_INDEX_HINT, i);
645                 goto out;
646             }
647             if (!ptr) {
648                 error_setg(&local_err, QERR_MISSING_PARAMETER,
649                            GLUSTER_OPT_PATH);
650                 error_append_hint(&local_err, GERR_INDEX_HINT, i);
651                 goto out;
652             }
653             gsconf->u.q_unix.path = g_strdup(ptr);
654             qemu_opts_del(opts);
655         }
656 
657         QAPI_LIST_APPEND(tail, gsconf);
658         gsconf = NULL;
659 
660         qobject_unref(backing_options);
661         backing_options = NULL;
662         g_free(str);
663         str = NULL;
664     }
665 
666     return 0;
667 
668 out:
669     error_propagate(errp, local_err);
670     qapi_free_SocketAddress(gsconf);
671     qemu_opts_del(opts);
672     g_free(str);
673     qobject_unref(backing_options);
674     errno = EINVAL;
675     return -errno;
676 }
677 
678 /* Converts options given in @filename and the @options QDict into the QAPI
679  * object @gconf. */
680 static int qemu_gluster_parse(BlockdevOptionsGluster *gconf,
681                               const char *filename,
682                               QDict *options, Error **errp)
683 {
684     int ret;
685     if (filename) {
686         ret = qemu_gluster_parse_uri(gconf, filename);
687         if (ret < 0) {
688             error_setg(errp, "invalid URI %s", filename);
689             error_append_hint(errp, "Usage: file=gluster[+transport]://"
690                                     "[host[:port]]volume/path[?socket=...]"
691                                     "[,file.debug=N]"
692                                     "[,file.logfile=/path/filename.log]\n");
693             return ret;
694         }
695     } else {
696         ret = qemu_gluster_parse_json(gconf, options, errp);
697         if (ret < 0) {
698             error_append_hint(errp, "Usage: "
699                              "-drive driver=qcow2,file.driver=gluster,"
700                              "file.volume=testvol,file.path=/path/a.qcow2"
701                              "[,file.debug=9]"
702                              "[,file.logfile=/path/filename.log],"
703                              "file.server.0.type=inet,"
704                              "file.server.0.host=1.2.3.4,"
705                              "file.server.0.port=24007,"
706                              "file.server.1.transport=unix,"
707                              "file.server.1.path=/var/run/glusterd.socket ..."
708                              "\n");
709             return ret;
710         }
711     }
712 
713     return 0;
714 }
715 
716 static struct glfs *qemu_gluster_init(BlockdevOptionsGluster *gconf,
717                                       const char *filename,
718                                       QDict *options, Error **errp)
719 {
720     int ret;
721 
722     ret = qemu_gluster_parse(gconf, filename, options, errp);
723     if (ret < 0) {
724         errno = -ret;
725         return NULL;
726     }
727 
728     return qemu_gluster_glfs_init(gconf, errp);
729 }
730 
731 /*
732  * AIO callback routine called from GlusterFS thread.
733  */
734 static void gluster_finish_aiocb(struct glfs_fd *fd, ssize_t ret,
735 #ifdef CONFIG_GLUSTERFS_IOCB_HAS_STAT
736                                  struct glfs_stat *pre, struct glfs_stat *post,
737 #endif
738                                  void *arg)
739 {
740     GlusterAIOCB *acb = (GlusterAIOCB *)arg;
741 
742     if (!ret || ret == acb->size) {
743         acb->ret = 0; /* Success */
744     } else if (ret < 0) {
745         acb->ret = -errno; /* Read/Write failed */
746     } else {
747         acb->ret = -EIO; /* Partial read/write - fail it */
748     }
749 
750     aio_co_schedule(acb->aio_context, acb->coroutine);
751 }
752 
753 static void qemu_gluster_parse_flags(int bdrv_flags, int *open_flags)
754 {
755     assert(open_flags != NULL);
756 
757     *open_flags |= O_BINARY;
758 
759     if (bdrv_flags & BDRV_O_RDWR) {
760         *open_flags |= O_RDWR;
761     } else {
762         *open_flags |= O_RDONLY;
763     }
764 
765     if ((bdrv_flags & BDRV_O_NOCACHE)) {
766         *open_flags |= O_DIRECT;
767     }
768 }
769 
770 /*
771  * Do SEEK_DATA/HOLE to detect if it is functional. Older broken versions of
772  * gfapi incorrectly return the current offset when SEEK_DATA/HOLE is used.
773  * - Corrected versions return -1 and set errno to EINVAL.
774  * - Versions that support SEEK_DATA/HOLE correctly, will return -1 and set
775  *   errno to ENXIO when SEEK_DATA is called with a position of EOF.
776  */
777 static bool qemu_gluster_test_seek(struct glfs_fd *fd)
778 {
779     off_t ret = 0;
780 
781 #if defined SEEK_HOLE && defined SEEK_DATA
782     off_t eof;
783 
784     eof = glfs_lseek(fd, 0, SEEK_END);
785     if (eof < 0) {
786         /* this should never occur */
787         return false;
788     }
789 
790     /* this should always fail with ENXIO if SEEK_DATA is supported */
791     ret = glfs_lseek(fd, eof, SEEK_DATA);
792 #endif
793 
794     return (ret < 0) && (errno == ENXIO);
795 }
796 
797 static int qemu_gluster_open(BlockDriverState *bs,  QDict *options,
798                              int bdrv_flags, Error **errp)
799 {
800     BDRVGlusterState *s = bs->opaque;
801     int open_flags = 0;
802     int ret = 0;
803     BlockdevOptionsGluster *gconf = NULL;
804     QemuOpts *opts;
805     const char *filename, *logfile;
806 
807     opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
808     if (!qemu_opts_absorb_qdict(opts, options, errp)) {
809         ret = -EINVAL;
810         goto out;
811     }
812 
813     filename = qemu_opt_get(opts, GLUSTER_OPT_FILENAME);
814 
815     s->debug = qemu_opt_get_number(opts, GLUSTER_OPT_DEBUG,
816                                    GLUSTER_DEBUG_DEFAULT);
817     if (s->debug < 0) {
818         s->debug = 0;
819     } else if (s->debug > GLUSTER_DEBUG_MAX) {
820         s->debug = GLUSTER_DEBUG_MAX;
821     }
822 
823     gconf = g_new0(BlockdevOptionsGluster, 1);
824     gconf->debug = s->debug;
825     gconf->has_debug = true;
826 
827     logfile = qemu_opt_get(opts, GLUSTER_OPT_LOGFILE);
828     s->logfile = g_strdup(logfile ? logfile : GLUSTER_LOGFILE_DEFAULT);
829 
830     gconf->logfile = g_strdup(s->logfile);
831 
832     s->glfs = qemu_gluster_init(gconf, filename, options, errp);
833     if (!s->glfs) {
834         ret = -errno;
835         goto out;
836     }
837 
838 #ifdef CONFIG_GLUSTERFS_XLATOR_OPT
839     /* Without this, if fsync fails for a recoverable reason (for instance,
840      * ENOSPC), gluster will dump its cache, preventing retries.  This means
841      * almost certain data loss.  Not all gluster versions support the
842      * 'resync-failed-syncs-after-fsync' key value, but there is no way to
843      * discover during runtime if it is supported (this api returns success for
844      * unknown key/value pairs) */
845     ret = glfs_set_xlator_option(s->glfs, "*-write-behind",
846                                           "resync-failed-syncs-after-fsync",
847                                           "on");
848     if (ret < 0) {
849         error_setg_errno(errp, errno, "Unable to set xlator key/value pair");
850         ret = -errno;
851         goto out;
852     }
853 #endif
854 
855     qemu_gluster_parse_flags(bdrv_flags, &open_flags);
856 
857     s->fd = glfs_open(s->glfs, gconf->path, open_flags);
858     ret = s->fd ? 0 : -errno;
859 
860     if (ret == -EACCES || ret == -EROFS) {
861         /* Try to degrade to read-only, but if it doesn't work, still use the
862          * normal error message. */
863         bdrv_graph_rdlock_main_loop();
864         if (bdrv_apply_auto_read_only(bs, NULL, NULL) == 0) {
865             open_flags = (open_flags & ~O_RDWR) | O_RDONLY;
866             s->fd = glfs_open(s->glfs, gconf->path, open_flags);
867             ret = s->fd ? 0 : -errno;
868         }
869         bdrv_graph_rdunlock_main_loop();
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 void qemu_gluster_refresh_limits(BlockDriverState *bs, Error **errp)
891 {
892     bs->bl.max_transfer = GLUSTER_MAX_TRANSFER;
893     bs->bl.max_pdiscard = MIN(SIZE_MAX, INT64_MAX);
894 }
895 
896 static int qemu_gluster_reopen_prepare(BDRVReopenState *state,
897                                        BlockReopenQueue *queue, Error **errp)
898 {
899     int ret = 0;
900     BDRVGlusterState *s;
901     BDRVGlusterReopenState *reop_s;
902     BlockdevOptionsGluster *gconf;
903     int open_flags = 0;
904 
905     assert(state != NULL);
906     assert(state->bs != NULL);
907 
908     s = state->bs->opaque;
909 
910     state->opaque = g_new0(BDRVGlusterReopenState, 1);
911     reop_s = state->opaque;
912 
913     qemu_gluster_parse_flags(state->flags, &open_flags);
914 
915     gconf = g_new0(BlockdevOptionsGluster, 1);
916     gconf->debug = s->debug;
917     gconf->has_debug = true;
918     gconf->logfile = g_strdup(s->logfile);
919 
920     /*
921      * If 'state->bs->exact_filename' is empty, 'state->options' should contain
922      * the JSON parameters already parsed.
923      */
924     if (state->bs->exact_filename[0] != '\0') {
925         reop_s->glfs = qemu_gluster_init(gconf, state->bs->exact_filename, NULL,
926                                          errp);
927     } else {
928         reop_s->glfs = qemu_gluster_init(gconf, NULL, state->options, errp);
929     }
930     if (reop_s->glfs == NULL) {
931         ret = -errno;
932         goto exit;
933     }
934 
935 #ifdef CONFIG_GLUSTERFS_XLATOR_OPT
936     ret = glfs_set_xlator_option(reop_s->glfs, "*-write-behind",
937                                  "resync-failed-syncs-after-fsync", "on");
938     if (ret < 0) {
939         error_setg_errno(errp, errno, "Unable to set xlator key/value pair");
940         ret = -errno;
941         goto exit;
942     }
943 #endif
944 
945     reop_s->fd = glfs_open(reop_s->glfs, gconf->path, open_flags);
946     if (reop_s->fd == NULL) {
947         /* reops->glfs will be cleaned up in _abort */
948         ret = -errno;
949         goto exit;
950     }
951 
952 exit:
953     /* state->opaque will be freed in either the _abort or _commit */
954     qapi_free_BlockdevOptionsGluster(gconf);
955     return ret;
956 }
957 
958 static void qemu_gluster_reopen_commit(BDRVReopenState *state)
959 {
960     BDRVGlusterReopenState *reop_s = state->opaque;
961     BDRVGlusterState *s = state->bs->opaque;
962 
963 
964     /* close the old */
965     if (s->fd) {
966         glfs_close(s->fd);
967     }
968 
969     glfs_clear_preopened(s->glfs);
970 
971     /* use the newly opened image / connection */
972     s->fd         = reop_s->fd;
973     s->glfs       = reop_s->glfs;
974 
975     g_free(state->opaque);
976     state->opaque = NULL;
977 
978     return;
979 }
980 
981 
982 static void qemu_gluster_reopen_abort(BDRVReopenState *state)
983 {
984     BDRVGlusterReopenState *reop_s = state->opaque;
985 
986     if (reop_s == NULL) {
987         return;
988     }
989 
990     if (reop_s->fd) {
991         glfs_close(reop_s->fd);
992     }
993 
994     glfs_clear_preopened(reop_s->glfs);
995 
996     g_free(state->opaque);
997     state->opaque = NULL;
998 
999     return;
1000 }
1001 
1002 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1003 static coroutine_fn int qemu_gluster_co_pwrite_zeroes(BlockDriverState *bs,
1004                                                       int64_t offset,
1005                                                       int64_t bytes,
1006                                                       BdrvRequestFlags flags)
1007 {
1008     int ret;
1009     GlusterAIOCB acb;
1010     BDRVGlusterState *s = bs->opaque;
1011 
1012     acb.size = bytes;
1013     acb.ret = 0;
1014     acb.coroutine = qemu_coroutine_self();
1015     acb.aio_context = bdrv_get_aio_context(bs);
1016 
1017     ret = glfs_zerofill_async(s->fd, offset, bytes, gluster_finish_aiocb, &acb);
1018     if (ret < 0) {
1019         return -errno;
1020     }
1021 
1022     qemu_coroutine_yield();
1023     return acb.ret;
1024 }
1025 #endif
1026 
1027 static int qemu_gluster_do_truncate(struct glfs_fd *fd, int64_t offset,
1028                                     PreallocMode prealloc, Error **errp)
1029 {
1030     int64_t current_length;
1031 
1032     current_length = glfs_lseek(fd, 0, SEEK_END);
1033     if (current_length < 0) {
1034         error_setg_errno(errp, errno, "Failed to determine current size");
1035         return -errno;
1036     }
1037 
1038     if (current_length > offset && prealloc != PREALLOC_MODE_OFF) {
1039         error_setg(errp, "Cannot use preallocation for shrinking files");
1040         return -ENOTSUP;
1041     }
1042 
1043     if (current_length == offset) {
1044         return 0;
1045     }
1046 
1047     switch (prealloc) {
1048 #ifdef CONFIG_GLUSTERFS_FALLOCATE
1049     case PREALLOC_MODE_FALLOC:
1050         if (glfs_fallocate(fd, 0, current_length, offset - current_length)) {
1051             error_setg_errno(errp, errno, "Could not preallocate data");
1052             return -errno;
1053         }
1054         break;
1055 #endif /* CONFIG_GLUSTERFS_FALLOCATE */
1056 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1057     case PREALLOC_MODE_FULL:
1058         if (glfs_ftruncate(fd, offset)) {
1059             error_setg_errno(errp, errno, "Could not resize file");
1060             return -errno;
1061         }
1062         if (glfs_zerofill(fd, current_length, offset - current_length)) {
1063             error_setg_errno(errp, errno, "Could not zerofill the new area");
1064             return -errno;
1065         }
1066         break;
1067 #endif /* CONFIG_GLUSTERFS_ZEROFILL */
1068     case PREALLOC_MODE_OFF:
1069         if (glfs_ftruncate(fd, offset)) {
1070             error_setg_errno(errp, errno, "Could not resize file");
1071             return -errno;
1072         }
1073         break;
1074     default:
1075         error_setg(errp, "Unsupported preallocation mode: %s",
1076                    PreallocMode_str(prealloc));
1077         return -EINVAL;
1078     }
1079 
1080     return 0;
1081 }
1082 
1083 static int qemu_gluster_co_create(BlockdevCreateOptions *options,
1084                                   Error **errp)
1085 {
1086     BlockdevCreateOptionsGluster *opts = &options->u.gluster;
1087     struct glfs *glfs;
1088     struct glfs_fd *fd = NULL;
1089     int ret = 0;
1090 
1091     assert(options->driver == BLOCKDEV_DRIVER_GLUSTER);
1092 
1093     glfs = qemu_gluster_glfs_init(opts->location, errp);
1094     if (!glfs) {
1095         ret = -errno;
1096         goto out;
1097     }
1098 
1099     fd = glfs_creat(glfs, opts->location->path,
1100                     O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, S_IRUSR | S_IWUSR);
1101     if (!fd) {
1102         ret = -errno;
1103         goto out;
1104     }
1105 
1106     ret = qemu_gluster_do_truncate(fd, opts->size, opts->preallocation, errp);
1107 
1108 out:
1109     if (fd) {
1110         if (glfs_close(fd) != 0 && ret == 0) {
1111             ret = -errno;
1112         }
1113     }
1114     glfs_clear_preopened(glfs);
1115     return ret;
1116 }
1117 
1118 static int coroutine_fn qemu_gluster_co_create_opts(BlockDriver *drv,
1119                                                     const char *filename,
1120                                                     QemuOpts *opts,
1121                                                     Error **errp)
1122 {
1123     BlockdevCreateOptions *options;
1124     BlockdevCreateOptionsGluster *gopts;
1125     BlockdevOptionsGluster *gconf;
1126     char *tmp = NULL;
1127     Error *local_err = NULL;
1128     int ret;
1129 
1130     options = g_new0(BlockdevCreateOptions, 1);
1131     options->driver = BLOCKDEV_DRIVER_GLUSTER;
1132     gopts = &options->u.gluster;
1133 
1134     gconf = g_new0(BlockdevOptionsGluster, 1);
1135     gopts->location = gconf;
1136 
1137     gopts->size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
1138                            BDRV_SECTOR_SIZE);
1139 
1140     tmp = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
1141     gopts->preallocation = qapi_enum_parse(&PreallocMode_lookup, tmp,
1142                                            PREALLOC_MODE_OFF, &local_err);
1143     g_free(tmp);
1144     if (local_err) {
1145         error_propagate(errp, local_err);
1146         ret = -EINVAL;
1147         goto fail;
1148     }
1149 
1150     gconf->debug = qemu_opt_get_number_del(opts, GLUSTER_OPT_DEBUG,
1151                                            GLUSTER_DEBUG_DEFAULT);
1152     if (gconf->debug < 0) {
1153         gconf->debug = 0;
1154     } else if (gconf->debug > GLUSTER_DEBUG_MAX) {
1155         gconf->debug = GLUSTER_DEBUG_MAX;
1156     }
1157     gconf->has_debug = true;
1158 
1159     gconf->logfile = qemu_opt_get_del(opts, GLUSTER_OPT_LOGFILE);
1160     if (!gconf->logfile) {
1161         gconf->logfile = g_strdup(GLUSTER_LOGFILE_DEFAULT);
1162     }
1163 
1164     ret = qemu_gluster_parse(gconf, filename, NULL, errp);
1165     if (ret < 0) {
1166         goto fail;
1167     }
1168 
1169     ret = qemu_gluster_co_create(options, errp);
1170     if (ret < 0) {
1171         goto fail;
1172     }
1173 
1174     ret = 0;
1175 fail:
1176     qapi_free_BlockdevCreateOptions(options);
1177     return ret;
1178 }
1179 
1180 static coroutine_fn int qemu_gluster_co_rw(BlockDriverState *bs,
1181                                            int64_t sector_num, int nb_sectors,
1182                                            QEMUIOVector *qiov, int write)
1183 {
1184     int ret;
1185     GlusterAIOCB acb;
1186     BDRVGlusterState *s = bs->opaque;
1187     size_t size = nb_sectors * BDRV_SECTOR_SIZE;
1188     off_t offset = sector_num * BDRV_SECTOR_SIZE;
1189 
1190     acb.size = size;
1191     acb.ret = 0;
1192     acb.coroutine = qemu_coroutine_self();
1193     acb.aio_context = bdrv_get_aio_context(bs);
1194 
1195     if (write) {
1196         ret = glfs_pwritev_async(s->fd, qiov->iov, qiov->niov, offset, 0,
1197                                  gluster_finish_aiocb, &acb);
1198     } else {
1199         ret = glfs_preadv_async(s->fd, qiov->iov, qiov->niov, offset, 0,
1200                                 gluster_finish_aiocb, &acb);
1201     }
1202 
1203     if (ret < 0) {
1204         return -errno;
1205     }
1206 
1207     qemu_coroutine_yield();
1208     return acb.ret;
1209 }
1210 
1211 static coroutine_fn int qemu_gluster_co_truncate(BlockDriverState *bs,
1212                                                  int64_t offset,
1213                                                  bool exact,
1214                                                  PreallocMode prealloc,
1215                                                  BdrvRequestFlags flags,
1216                                                  Error **errp)
1217 {
1218     BDRVGlusterState *s = bs->opaque;
1219     return qemu_gluster_do_truncate(s->fd, offset, prealloc, errp);
1220 }
1221 
1222 static coroutine_fn int qemu_gluster_co_readv(BlockDriverState *bs,
1223                                               int64_t sector_num,
1224                                               int nb_sectors,
1225                                               QEMUIOVector *qiov)
1226 {
1227     return qemu_gluster_co_rw(bs, sector_num, nb_sectors, qiov, 0);
1228 }
1229 
1230 static coroutine_fn int qemu_gluster_co_writev(BlockDriverState *bs,
1231                                                int64_t sector_num,
1232                                                int nb_sectors,
1233                                                QEMUIOVector *qiov,
1234                                                int flags)
1235 {
1236     return qemu_gluster_co_rw(bs, sector_num, nb_sectors, qiov, 1);
1237 }
1238 
1239 static void qemu_gluster_close(BlockDriverState *bs)
1240 {
1241     BDRVGlusterState *s = bs->opaque;
1242 
1243     g_free(s->logfile);
1244     if (s->fd) {
1245         glfs_close(s->fd);
1246         s->fd = NULL;
1247     }
1248     glfs_clear_preopened(s->glfs);
1249 }
1250 
1251 static coroutine_fn int qemu_gluster_co_flush_to_disk(BlockDriverState *bs)
1252 {
1253     int ret;
1254     GlusterAIOCB acb;
1255     BDRVGlusterState *s = bs->opaque;
1256 
1257     acb.size = 0;
1258     acb.ret = 0;
1259     acb.coroutine = qemu_coroutine_self();
1260     acb.aio_context = bdrv_get_aio_context(bs);
1261 
1262     ret = glfs_fsync_async(s->fd, gluster_finish_aiocb, &acb);
1263     if (ret < 0) {
1264         ret = -errno;
1265         goto error;
1266     }
1267 
1268     qemu_coroutine_yield();
1269     if (acb.ret < 0) {
1270         ret = acb.ret;
1271         goto error;
1272     }
1273 
1274     return acb.ret;
1275 
1276 error:
1277     /* Some versions of Gluster (3.5.6 -> 3.5.8?) will not retain its cache
1278      * after a fsync failure, so we have no way of allowing the guest to safely
1279      * continue.  Gluster versions prior to 3.5.6 don't retain the cache
1280      * either, but will invalidate the fd on error, so this is again our only
1281      * option.
1282      *
1283      * The 'resync-failed-syncs-after-fsync' xlator option for the
1284      * write-behind cache will cause later gluster versions to retain its
1285      * cache after error, so long as the fd remains open.  However, we
1286      * currently have no way of knowing if this option is supported.
1287      *
1288      * TODO: Once gluster provides a way for us to determine if the option
1289      * is supported, bypass the closure and setting drv to NULL.  */
1290     qemu_gluster_close(bs);
1291     bs->drv = NULL;
1292     return ret;
1293 }
1294 
1295 #ifdef CONFIG_GLUSTERFS_DISCARD
1296 static coroutine_fn int qemu_gluster_co_pdiscard(BlockDriverState *bs,
1297                                                  int64_t offset, int64_t bytes)
1298 {
1299     int ret;
1300     GlusterAIOCB acb;
1301     BDRVGlusterState *s = bs->opaque;
1302 
1303     assert(bytes <= SIZE_MAX); /* rely on max_pdiscard */
1304 
1305     acb.size = 0;
1306     acb.ret = 0;
1307     acb.coroutine = qemu_coroutine_self();
1308     acb.aio_context = bdrv_get_aio_context(bs);
1309 
1310     ret = glfs_discard_async(s->fd, offset, bytes, gluster_finish_aiocb, &acb);
1311     if (ret < 0) {
1312         return -errno;
1313     }
1314 
1315     qemu_coroutine_yield();
1316     return acb.ret;
1317 }
1318 #endif
1319 
1320 static int64_t coroutine_fn qemu_gluster_co_getlength(BlockDriverState *bs)
1321 {
1322     BDRVGlusterState *s = bs->opaque;
1323     int64_t ret;
1324 
1325     ret = glfs_lseek(s->fd, 0, SEEK_END);
1326     if (ret < 0) {
1327         return -errno;
1328     } else {
1329         return ret;
1330     }
1331 }
1332 
1333 static int64_t coroutine_fn
1334 qemu_gluster_co_get_allocated_file_size(BlockDriverState *bs)
1335 {
1336     BDRVGlusterState *s = bs->opaque;
1337     struct stat st;
1338     int ret;
1339 
1340     ret = glfs_fstat(s->fd, &st);
1341     if (ret < 0) {
1342         return -errno;
1343     } else {
1344         return st.st_blocks * 512;
1345     }
1346 }
1347 
1348 /*
1349  * Find allocation range in @bs around offset @start.
1350  * May change underlying file descriptor's file offset.
1351  * If @start is not in a hole, store @start in @data, and the
1352  * beginning of the next hole in @hole, and return 0.
1353  * If @start is in a non-trailing hole, store @start in @hole and the
1354  * beginning of the next non-hole in @data, and return 0.
1355  * If @start is in a trailing hole or beyond EOF, return -ENXIO.
1356  * If we can't find out, return a negative errno other than -ENXIO.
1357  *
1358  * (Shamefully copied from file-posix.c, only minuscule adaptions.)
1359  */
1360 static int find_allocation(BlockDriverState *bs, off_t start,
1361                            off_t *data, off_t *hole)
1362 {
1363     BDRVGlusterState *s = bs->opaque;
1364 
1365     if (!s->supports_seek_data) {
1366         goto exit;
1367     }
1368 
1369 #if defined SEEK_HOLE && defined SEEK_DATA
1370     off_t offs;
1371 
1372     /*
1373      * SEEK_DATA cases:
1374      * D1. offs == start: start is in data
1375      * D2. offs > start: start is in a hole, next data at offs
1376      * D3. offs < 0, errno = ENXIO: either start is in a trailing hole
1377      *                              or start is beyond EOF
1378      *     If the latter happens, the file has been truncated behind
1379      *     our back since we opened it.  All bets are off then.
1380      *     Treating like a trailing hole is simplest.
1381      * D4. offs < 0, errno != ENXIO: we learned nothing
1382      */
1383     offs = glfs_lseek(s->fd, start, SEEK_DATA);
1384     if (offs < 0) {
1385         return -errno;          /* D3 or D4 */
1386     }
1387 
1388     if (offs < start) {
1389         /* This is not a valid return by lseek().  We are safe to just return
1390          * -EIO in this case, and we'll treat it like D4. Unfortunately some
1391          *  versions of gluster server will return offs < start, so an assert
1392          *  here will unnecessarily abort QEMU. */
1393         return -EIO;
1394     }
1395 
1396     if (offs > start) {
1397         /* D2: in hole, next data at offs */
1398         *hole = start;
1399         *data = offs;
1400         return 0;
1401     }
1402 
1403     /* D1: in data, end not yet known */
1404 
1405     /*
1406      * SEEK_HOLE cases:
1407      * H1. offs == start: start is in a hole
1408      *     If this happens here, a hole has been dug behind our back
1409      *     since the previous lseek().
1410      * H2. offs > start: either start is in data, next hole at offs,
1411      *                   or start is in trailing hole, EOF at offs
1412      *     Linux treats trailing holes like any other hole: offs ==
1413      *     start.  Solaris seeks to EOF instead: offs > start (blech).
1414      *     If that happens here, a hole has been dug behind our back
1415      *     since the previous lseek().
1416      * H3. offs < 0, errno = ENXIO: start is beyond EOF
1417      *     If this happens, the file has been truncated behind our
1418      *     back since we opened it.  Treat it like a trailing hole.
1419      * H4. offs < 0, errno != ENXIO: we learned nothing
1420      *     Pretend we know nothing at all, i.e. "forget" about D1.
1421      */
1422     offs = glfs_lseek(s->fd, start, SEEK_HOLE);
1423     if (offs < 0) {
1424         return -errno;          /* D1 and (H3 or H4) */
1425     }
1426 
1427     if (offs < start) {
1428         /* This is not a valid return by lseek().  We are safe to just return
1429          * -EIO in this case, and we'll treat it like H4. Unfortunately some
1430          *  versions of gluster server will return offs < start, so an assert
1431          *  here will unnecessarily abort QEMU. */
1432         return -EIO;
1433     }
1434 
1435     if (offs > start) {
1436         /*
1437          * D1 and H2: either in data, next hole at offs, or it was in
1438          * data but is now in a trailing hole.  In the latter case,
1439          * all bets are off.  Treating it as if it there was data all
1440          * the way to EOF is safe, so simply do that.
1441          */
1442         *data = start;
1443         *hole = offs;
1444         return 0;
1445     }
1446 
1447     /* D1 and H1 */
1448     return -EBUSY;
1449 #endif
1450 
1451 exit:
1452     return -ENOTSUP;
1453 }
1454 
1455 /*
1456  * Returns the allocation status of the specified offset.
1457  *
1458  * The block layer guarantees 'offset' and 'bytes' are within bounds.
1459  *
1460  * 'pnum' is set to the number of bytes (including and immediately following
1461  * the specified offset) that are known to be in the same
1462  * allocated/unallocated state.
1463  *
1464  * 'bytes' is a soft cap for 'pnum'.  If the information is free, 'pnum' may
1465  * well exceed it.
1466  *
1467  * (Based on raw_co_block_status() from file-posix.c.)
1468  */
1469 static int coroutine_fn qemu_gluster_co_block_status(BlockDriverState *bs,
1470                                                      bool want_zero,
1471                                                      int64_t offset,
1472                                                      int64_t bytes,
1473                                                      int64_t *pnum,
1474                                                      int64_t *map,
1475                                                      BlockDriverState **file)
1476 {
1477     BDRVGlusterState *s = bs->opaque;
1478     off_t data = 0, hole = 0;
1479     int ret = -EINVAL;
1480 
1481     assert(QEMU_IS_ALIGNED(offset | bytes, bs->bl.request_alignment));
1482 
1483     if (!s->fd) {
1484         return ret;
1485     }
1486 
1487     if (!want_zero) {
1488         *pnum = bytes;
1489         *map = offset;
1490         *file = bs;
1491         return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID;
1492     }
1493 
1494     ret = find_allocation(bs, offset, &data, &hole);
1495     if (ret == -ENXIO) {
1496         /* Trailing hole */
1497         *pnum = bytes;
1498         ret = BDRV_BLOCK_ZERO;
1499     } else if (ret < 0) {
1500         /* No info available, so pretend there are no holes */
1501         *pnum = bytes;
1502         ret = BDRV_BLOCK_DATA;
1503     } else if (data == offset) {
1504         /* On a data extent, compute bytes to the end of the extent,
1505          * possibly including a partial sector at EOF. */
1506         *pnum = hole - offset;
1507 
1508         /*
1509          * We are not allowed to return partial sectors, though, so
1510          * round up if necessary.
1511          */
1512         if (!QEMU_IS_ALIGNED(*pnum, bs->bl.request_alignment)) {
1513             int64_t file_length = qemu_gluster_co_getlength(bs);
1514             if (file_length > 0) {
1515                 /* Ignore errors, this is just a safeguard */
1516                 assert(hole == file_length);
1517             }
1518             *pnum = ROUND_UP(*pnum, bs->bl.request_alignment);
1519         }
1520 
1521         ret = BDRV_BLOCK_DATA;
1522     } else {
1523         /* On a hole, compute bytes to the beginning of the next extent.  */
1524         assert(hole == offset);
1525         *pnum = data - offset;
1526         ret = BDRV_BLOCK_ZERO;
1527     }
1528 
1529     *map = offset;
1530     *file = bs;
1531 
1532     return ret | BDRV_BLOCK_OFFSET_VALID;
1533 }
1534 
1535 
1536 static const char *const gluster_strong_open_opts[] = {
1537     GLUSTER_OPT_VOLUME,
1538     GLUSTER_OPT_PATH,
1539     GLUSTER_OPT_TYPE,
1540     GLUSTER_OPT_SERVER_PATTERN,
1541     GLUSTER_OPT_HOST,
1542     GLUSTER_OPT_PORT,
1543     GLUSTER_OPT_TO,
1544     GLUSTER_OPT_IPV4,
1545     GLUSTER_OPT_IPV6,
1546     GLUSTER_OPT_SOCKET,
1547 
1548     NULL
1549 };
1550 
1551 static BlockDriver bdrv_gluster = {
1552     .format_name                  = "gluster",
1553     .protocol_name                = "gluster",
1554     .instance_size                = sizeof(BDRVGlusterState),
1555     .bdrv_file_open               = qemu_gluster_open,
1556     .bdrv_reopen_prepare          = qemu_gluster_reopen_prepare,
1557     .bdrv_reopen_commit           = qemu_gluster_reopen_commit,
1558     .bdrv_reopen_abort            = qemu_gluster_reopen_abort,
1559     .bdrv_close                   = qemu_gluster_close,
1560     .bdrv_co_create               = qemu_gluster_co_create,
1561     .bdrv_co_create_opts          = qemu_gluster_co_create_opts,
1562     .bdrv_co_getlength            = qemu_gluster_co_getlength,
1563     .bdrv_co_get_allocated_file_size = qemu_gluster_co_get_allocated_file_size,
1564     .bdrv_co_truncate             = qemu_gluster_co_truncate,
1565     .bdrv_co_readv                = qemu_gluster_co_readv,
1566     .bdrv_co_writev               = qemu_gluster_co_writev,
1567     .bdrv_co_flush_to_disk        = qemu_gluster_co_flush_to_disk,
1568 #ifdef CONFIG_GLUSTERFS_DISCARD
1569     .bdrv_co_pdiscard             = qemu_gluster_co_pdiscard,
1570 #endif
1571 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1572     .bdrv_co_pwrite_zeroes        = qemu_gluster_co_pwrite_zeroes,
1573 #endif
1574     .bdrv_co_block_status         = qemu_gluster_co_block_status,
1575     .bdrv_refresh_limits          = qemu_gluster_refresh_limits,
1576     .create_opts                  = &qemu_gluster_create_opts,
1577     .strong_runtime_opts          = gluster_strong_open_opts,
1578 };
1579 
1580 static BlockDriver bdrv_gluster_tcp = {
1581     .format_name                  = "gluster",
1582     .protocol_name                = "gluster+tcp",
1583     .instance_size                = sizeof(BDRVGlusterState),
1584     .bdrv_file_open               = qemu_gluster_open,
1585     .bdrv_reopen_prepare          = qemu_gluster_reopen_prepare,
1586     .bdrv_reopen_commit           = qemu_gluster_reopen_commit,
1587     .bdrv_reopen_abort            = qemu_gluster_reopen_abort,
1588     .bdrv_close                   = qemu_gluster_close,
1589     .bdrv_co_create               = qemu_gluster_co_create,
1590     .bdrv_co_create_opts          = qemu_gluster_co_create_opts,
1591     .bdrv_co_getlength            = qemu_gluster_co_getlength,
1592     .bdrv_co_get_allocated_file_size = qemu_gluster_co_get_allocated_file_size,
1593     .bdrv_co_truncate             = qemu_gluster_co_truncate,
1594     .bdrv_co_readv                = qemu_gluster_co_readv,
1595     .bdrv_co_writev               = qemu_gluster_co_writev,
1596     .bdrv_co_flush_to_disk        = qemu_gluster_co_flush_to_disk,
1597 #ifdef CONFIG_GLUSTERFS_DISCARD
1598     .bdrv_co_pdiscard             = qemu_gluster_co_pdiscard,
1599 #endif
1600 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1601     .bdrv_co_pwrite_zeroes        = qemu_gluster_co_pwrite_zeroes,
1602 #endif
1603     .bdrv_co_block_status         = qemu_gluster_co_block_status,
1604     .bdrv_refresh_limits          = qemu_gluster_refresh_limits,
1605     .create_opts                  = &qemu_gluster_create_opts,
1606     .strong_runtime_opts          = gluster_strong_open_opts,
1607 };
1608 
1609 static BlockDriver bdrv_gluster_unix = {
1610     .format_name                  = "gluster",
1611     .protocol_name                = "gluster+unix",
1612     .instance_size                = sizeof(BDRVGlusterState),
1613     .bdrv_file_open               = qemu_gluster_open,
1614     .bdrv_reopen_prepare          = qemu_gluster_reopen_prepare,
1615     .bdrv_reopen_commit           = qemu_gluster_reopen_commit,
1616     .bdrv_reopen_abort            = qemu_gluster_reopen_abort,
1617     .bdrv_close                   = qemu_gluster_close,
1618     .bdrv_co_create               = qemu_gluster_co_create,
1619     .bdrv_co_create_opts          = qemu_gluster_co_create_opts,
1620     .bdrv_co_getlength            = qemu_gluster_co_getlength,
1621     .bdrv_co_get_allocated_file_size = qemu_gluster_co_get_allocated_file_size,
1622     .bdrv_co_truncate             = qemu_gluster_co_truncate,
1623     .bdrv_co_readv                = qemu_gluster_co_readv,
1624     .bdrv_co_writev               = qemu_gluster_co_writev,
1625     .bdrv_co_flush_to_disk        = qemu_gluster_co_flush_to_disk,
1626 #ifdef CONFIG_GLUSTERFS_DISCARD
1627     .bdrv_co_pdiscard             = qemu_gluster_co_pdiscard,
1628 #endif
1629 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1630     .bdrv_co_pwrite_zeroes        = qemu_gluster_co_pwrite_zeroes,
1631 #endif
1632     .bdrv_co_block_status         = qemu_gluster_co_block_status,
1633     .bdrv_refresh_limits          = qemu_gluster_refresh_limits,
1634     .create_opts                  = &qemu_gluster_create_opts,
1635     .strong_runtime_opts          = gluster_strong_open_opts,
1636 };
1637 
1638 static void bdrv_gluster_init(void)
1639 {
1640     bdrv_register(&bdrv_gluster_unix);
1641     bdrv_register(&bdrv_gluster_tcp);
1642     bdrv_register(&bdrv_gluster);
1643 }
1644 
1645 block_init(bdrv_gluster_init);
1646