xref: /openbmc/qemu/block/gluster.c (revision 2f44bea9)
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_int.h"
15 #include "block/qdict.h"
16 #include "qapi/error.h"
17 #include "qapi/qmp/qdict.h"
18 #include "qapi/qmp/qerror.h"
19 #include "qemu/uri.h"
20 #include "qemu/error-report.h"
21 #include "qemu/module.h"
22 #include "qemu/option.h"
23 #include "qemu/cutils.h"
24 
25 #ifdef CONFIG_GLUSTERFS_FTRUNCATE_HAS_STAT
26 # define glfs_ftruncate(fd, offset) glfs_ftruncate(fd, offset, NULL, NULL)
27 #endif
28 
29 #define GLUSTER_OPT_FILENAME        "filename"
30 #define GLUSTER_OPT_VOLUME          "volume"
31 #define GLUSTER_OPT_PATH            "path"
32 #define GLUSTER_OPT_TYPE            "type"
33 #define GLUSTER_OPT_SERVER_PATTERN  "server."
34 #define GLUSTER_OPT_HOST            "host"
35 #define GLUSTER_OPT_PORT            "port"
36 #define GLUSTER_OPT_TO              "to"
37 #define GLUSTER_OPT_IPV4            "ipv4"
38 #define GLUSTER_OPT_IPV6            "ipv6"
39 #define GLUSTER_OPT_SOCKET          "socket"
40 #define GLUSTER_OPT_DEBUG           "debug"
41 #define GLUSTER_DEFAULT_PORT        24007
42 #define GLUSTER_DEBUG_DEFAULT       4
43 #define GLUSTER_DEBUG_MAX           9
44 #define GLUSTER_OPT_LOGFILE         "logfile"
45 #define GLUSTER_LOGFILE_DEFAULT     "-" /* handled in libgfapi as /dev/stderr */
46 /*
47  * Several versions of GlusterFS (3.12? -> 6.0.1) fail when the transfer size
48  * is greater or equal to 1024 MiB, so we are limiting the transfer size to 512
49  * MiB to avoid this rare issue.
50  */
51 #define GLUSTER_MAX_TRANSFER        (512 * MiB)
52 
53 #define GERR_INDEX_HINT "hint: check in 'server' array index '%d'\n"
54 
55 typedef struct GlusterAIOCB {
56     int64_t size;
57     int ret;
58     Coroutine *coroutine;
59     AioContext *aio_context;
60 } GlusterAIOCB;
61 
62 typedef struct BDRVGlusterState {
63     struct glfs *glfs;
64     struct glfs_fd *fd;
65     char *logfile;
66     bool supports_seek_data;
67     int debug;
68 } BDRVGlusterState;
69 
70 typedef struct BDRVGlusterReopenState {
71     struct glfs *glfs;
72     struct glfs_fd *fd;
73 } BDRVGlusterReopenState;
74 
75 
76 typedef struct GlfsPreopened {
77     char *volume;
78     glfs_t *fs;
79     int ref;
80 } GlfsPreopened;
81 
82 typedef struct ListElement {
83     QLIST_ENTRY(ListElement) list;
84     GlfsPreopened saved;
85 } ListElement;
86 
87 static QLIST_HEAD(, ListElement) glfs_list;
88 
89 static QemuOptsList qemu_gluster_create_opts = {
90     .name = "qemu-gluster-create-opts",
91     .head = QTAILQ_HEAD_INITIALIZER(qemu_gluster_create_opts.head),
92     .desc = {
93         {
94             .name = BLOCK_OPT_SIZE,
95             .type = QEMU_OPT_SIZE,
96             .help = "Virtual disk size"
97         },
98         {
99             .name = BLOCK_OPT_PREALLOC,
100             .type = QEMU_OPT_STRING,
101             .help = "Preallocation mode (allowed values: off"
102 #ifdef CONFIG_GLUSTERFS_FALLOCATE
103                     ", falloc"
104 #endif
105 #ifdef CONFIG_GLUSTERFS_ZEROFILL
106                     ", full"
107 #endif
108                     ")"
109         },
110         {
111             .name = GLUSTER_OPT_DEBUG,
112             .type = QEMU_OPT_NUMBER,
113             .help = "Gluster log level, valid range is 0-9",
114         },
115         {
116             .name = GLUSTER_OPT_LOGFILE,
117             .type = QEMU_OPT_STRING,
118             .help = "Logfile path of libgfapi",
119         },
120         { /* end of list */ }
121     }
122 };
123 
124 static QemuOptsList runtime_opts = {
125     .name = "gluster",
126     .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
127     .desc = {
128         {
129             .name = GLUSTER_OPT_FILENAME,
130             .type = QEMU_OPT_STRING,
131             .help = "URL to the gluster image",
132         },
133         {
134             .name = GLUSTER_OPT_DEBUG,
135             .type = QEMU_OPT_NUMBER,
136             .help = "Gluster log level, valid range is 0-9",
137         },
138         {
139             .name = GLUSTER_OPT_LOGFILE,
140             .type = QEMU_OPT_STRING,
141             .help = "Logfile path of libgfapi",
142         },
143         { /* end of list */ }
144     },
145 };
146 
147 static QemuOptsList runtime_json_opts = {
148     .name = "gluster_json",
149     .head = QTAILQ_HEAD_INITIALIZER(runtime_json_opts.head),
150     .desc = {
151         {
152             .name = GLUSTER_OPT_VOLUME,
153             .type = QEMU_OPT_STRING,
154             .help = "name of gluster volume where VM image resides",
155         },
156         {
157             .name = GLUSTER_OPT_PATH,
158             .type = QEMU_OPT_STRING,
159             .help = "absolute path to image file in gluster volume",
160         },
161         {
162             .name = GLUSTER_OPT_DEBUG,
163             .type = QEMU_OPT_NUMBER,
164             .help = "Gluster log level, valid range is 0-9",
165         },
166         { /* end of list */ }
167     },
168 };
169 
170 static QemuOptsList runtime_type_opts = {
171     .name = "gluster_type",
172     .head = QTAILQ_HEAD_INITIALIZER(runtime_type_opts.head),
173     .desc = {
174         {
175             .name = GLUSTER_OPT_TYPE,
176             .type = QEMU_OPT_STRING,
177             .help = "inet|unix",
178         },
179         { /* end of list */ }
180     },
181 };
182 
183 static QemuOptsList runtime_unix_opts = {
184     .name = "gluster_unix",
185     .head = QTAILQ_HEAD_INITIALIZER(runtime_unix_opts.head),
186     .desc = {
187         {
188             .name = GLUSTER_OPT_SOCKET,
189             .type = QEMU_OPT_STRING,
190             .help = "socket file path (legacy)",
191         },
192         {
193             .name = GLUSTER_OPT_PATH,
194             .type = QEMU_OPT_STRING,
195             .help = "socket file path (QAPI)",
196         },
197         { /* end of list */ }
198     },
199 };
200 
201 static QemuOptsList runtime_inet_opts = {
202     .name = "gluster_inet",
203     .head = QTAILQ_HEAD_INITIALIZER(runtime_inet_opts.head),
204     .desc = {
205         {
206             .name = GLUSTER_OPT_TYPE,
207             .type = QEMU_OPT_STRING,
208             .help = "inet|unix",
209         },
210         {
211             .name = GLUSTER_OPT_HOST,
212             .type = QEMU_OPT_STRING,
213             .help = "host address (hostname/ipv4/ipv6 addresses)",
214         },
215         {
216             .name = GLUSTER_OPT_PORT,
217             .type = QEMU_OPT_STRING,
218             .help = "port number on which glusterd is listening (default 24007)",
219         },
220         {
221             .name = "to",
222             .type = QEMU_OPT_NUMBER,
223             .help = "max port number, not supported by gluster",
224         },
225         {
226             .name = "ipv4",
227             .type = QEMU_OPT_BOOL,
228             .help = "ipv4 bool value, not supported by gluster",
229         },
230         {
231             .name = "ipv6",
232             .type = QEMU_OPT_BOOL,
233             .help = "ipv6 bool value, not supported by gluster",
234         },
235         { /* end of list */ }
236     },
237 };
238 
239 static void glfs_set_preopened(const char *volume, glfs_t *fs)
240 {
241     ListElement *entry = NULL;
242 
243     entry = g_new(ListElement, 1);
244 
245     entry->saved.volume = g_strdup(volume);
246 
247     entry->saved.fs = fs;
248     entry->saved.ref = 1;
249 
250     QLIST_INSERT_HEAD(&glfs_list, entry, list);
251 }
252 
253 static glfs_t *glfs_find_preopened(const char *volume)
254 {
255     ListElement *entry = NULL;
256 
257      QLIST_FOREACH(entry, &glfs_list, list) {
258         if (strcmp(entry->saved.volume, volume) == 0) {
259             entry->saved.ref++;
260             return entry->saved.fs;
261         }
262      }
263 
264     return NULL;
265 }
266 
267 static void glfs_clear_preopened(glfs_t *fs)
268 {
269     ListElement *entry = NULL;
270     ListElement *next;
271 
272     if (fs == NULL) {
273         return;
274     }
275 
276     QLIST_FOREACH_SAFE(entry, &glfs_list, list, next) {
277         if (entry->saved.fs == fs) {
278             if (--entry->saved.ref) {
279                 return;
280             }
281 
282             QLIST_REMOVE(entry, list);
283 
284             glfs_fini(entry->saved.fs);
285             g_free(entry->saved.volume);
286             g_free(entry);
287         }
288     }
289 }
290 
291 static int parse_volume_options(BlockdevOptionsGluster *gconf, char *path)
292 {
293     char *p, *q;
294 
295     if (!path) {
296         return -EINVAL;
297     }
298 
299     /* volume */
300     p = q = path + strspn(path, "/");
301     p += strcspn(p, "/");
302     if (*p == '\0') {
303         return -EINVAL;
304     }
305     gconf->volume = g_strndup(q, p - q);
306 
307     /* path */
308     p += strspn(p, "/");
309     if (*p == '\0') {
310         return -EINVAL;
311     }
312     gconf->path = g_strdup(p);
313     return 0;
314 }
315 
316 /*
317  * file=gluster[+transport]://[host[:port]]/volume/path[?socket=...]
318  *
319  * 'gluster' is the protocol.
320  *
321  * 'transport' specifies the transport type used to connect to gluster
322  * management daemon (glusterd). Valid transport types are
323  * tcp or unix. If a transport type isn't specified, then tcp type is assumed.
324  *
325  * 'host' specifies the host where the volume file specification for
326  * the given volume resides. This can be either hostname or ipv4 address.
327  * If transport type is 'unix', then 'host' field should not be specified.
328  * The 'socket' field needs to be populated with the path to unix domain
329  * socket.
330  *
331  * 'port' is the port number on which glusterd is listening. This is optional
332  * and if not specified, QEMU will send 0 which will make gluster to use the
333  * default port. If the transport type is unix, then 'port' should not be
334  * specified.
335  *
336  * 'volume' is the name of the gluster volume which contains the VM image.
337  *
338  * 'path' is the path to the actual VM image that resides on gluster volume.
339  *
340  * Examples:
341  *
342  * file=gluster://1.2.3.4/testvol/a.img
343  * file=gluster+tcp://1.2.3.4/testvol/a.img
344  * file=gluster+tcp://1.2.3.4:24007/testvol/dir/a.img
345  * file=gluster+tcp://host.domain.com:24007/testvol/dir/a.img
346  * file=gluster+unix:///testvol/dir/a.img?socket=/tmp/glusterd.socket
347  */
348 static int qemu_gluster_parse_uri(BlockdevOptionsGluster *gconf,
349                                   const char *filename)
350 {
351     SocketAddress *gsconf;
352     URI *uri;
353     QueryParams *qp = NULL;
354     bool is_unix = false;
355     int ret = 0;
356 
357     uri = uri_parse(filename);
358     if (!uri) {
359         return -EINVAL;
360     }
361 
362     gsconf = g_new0(SocketAddress, 1);
363     QAPI_LIST_PREPEND(gconf->server, gsconf);
364 
365     /* transport */
366     if (!uri->scheme || !strcmp(uri->scheme, "gluster")) {
367         gsconf->type = SOCKET_ADDRESS_TYPE_INET;
368     } else if (!strcmp(uri->scheme, "gluster+tcp")) {
369         gsconf->type = SOCKET_ADDRESS_TYPE_INET;
370     } else if (!strcmp(uri->scheme, "gluster+unix")) {
371         gsconf->type = SOCKET_ADDRESS_TYPE_UNIX;
372         is_unix = true;
373     } else if (!strcmp(uri->scheme, "gluster+rdma")) {
374         gsconf->type = SOCKET_ADDRESS_TYPE_INET;
375         warn_report("rdma feature is not supported, falling back to tcp");
376     } else {
377         ret = -EINVAL;
378         goto out;
379     }
380 
381     ret = parse_volume_options(gconf, uri->path);
382     if (ret < 0) {
383         goto out;
384     }
385 
386     qp = query_params_parse(uri->query);
387     if (qp->n > 1 || (is_unix && !qp->n) || (!is_unix && qp->n)) {
388         ret = -EINVAL;
389         goto out;
390     }
391 
392     if (is_unix) {
393         if (uri->server || uri->port) {
394             ret = -EINVAL;
395             goto out;
396         }
397         if (strcmp(qp->p[0].name, "socket")) {
398             ret = -EINVAL;
399             goto out;
400         }
401         gsconf->u.q_unix.path = g_strdup(qp->p[0].value);
402     } else {
403         gsconf->u.inet.host = g_strdup(uri->server ? uri->server : "localhost");
404         if (uri->port) {
405             gsconf->u.inet.port = g_strdup_printf("%d", uri->port);
406         } else {
407             gsconf->u.inet.port = g_strdup_printf("%d", GLUSTER_DEFAULT_PORT);
408         }
409     }
410 
411 out:
412     if (qp) {
413         query_params_free(qp);
414     }
415     uri_free(uri);
416     return ret;
417 }
418 
419 static struct glfs *qemu_gluster_glfs_init(BlockdevOptionsGluster *gconf,
420                                            Error **errp)
421 {
422     struct glfs *glfs;
423     int ret;
424     int old_errno;
425     SocketAddressList *server;
426     unsigned long long port;
427 
428     glfs = glfs_find_preopened(gconf->volume);
429     if (glfs) {
430         return glfs;
431     }
432 
433     glfs = glfs_new(gconf->volume);
434     if (!glfs) {
435         goto out;
436     }
437 
438     glfs_set_preopened(gconf->volume, glfs);
439 
440     for (server = gconf->server; server; server = server->next) {
441         switch (server->value->type) {
442         case SOCKET_ADDRESS_TYPE_UNIX:
443             ret = glfs_set_volfile_server(glfs, "unix",
444                                    server->value->u.q_unix.path, 0);
445             break;
446         case SOCKET_ADDRESS_TYPE_INET:
447             if (parse_uint_full(server->value->u.inet.port, &port, 10) < 0 ||
448                 port > 65535) {
449                 error_setg(errp, "'%s' is not a valid port number",
450                            server->value->u.inet.port);
451                 errno = EINVAL;
452                 goto out;
453             }
454             ret = glfs_set_volfile_server(glfs, "tcp",
455                                    server->value->u.inet.host,
456                                    (int)port);
457             break;
458         case SOCKET_ADDRESS_TYPE_VSOCK:
459         case SOCKET_ADDRESS_TYPE_FD:
460         default:
461             abort();
462         }
463 
464         if (ret < 0) {
465             goto out;
466         }
467     }
468 
469     ret = glfs_set_logging(glfs, gconf->logfile, gconf->debug);
470     if (ret < 0) {
471         goto out;
472     }
473 
474     ret = glfs_init(glfs);
475     if (ret) {
476         error_setg(errp, "Gluster connection for volume %s, path %s failed"
477                          " to connect", gconf->volume, gconf->path);
478         for (server = gconf->server; server; server = server->next) {
479             if (server->value->type  == SOCKET_ADDRESS_TYPE_UNIX) {
480                 error_append_hint(errp, "hint: failed on socket %s ",
481                                   server->value->u.q_unix.path);
482             } else {
483                 error_append_hint(errp, "hint: failed on host %s and port %s ",
484                                   server->value->u.inet.host,
485                                   server->value->u.inet.port);
486             }
487         }
488 
489         error_append_hint(errp, "Please refer to gluster logs for more info\n");
490 
491         /* glfs_init sometimes doesn't set errno although docs suggest that */
492         if (errno == 0) {
493             errno = EINVAL;
494         }
495 
496         goto out;
497     }
498     return glfs;
499 
500 out:
501     if (glfs) {
502         old_errno = errno;
503         glfs_clear_preopened(glfs);
504         errno = old_errno;
505     }
506     return NULL;
507 }
508 
509 /*
510  * Convert the json formatted command line into qapi.
511 */
512 static int qemu_gluster_parse_json(BlockdevOptionsGluster *gconf,
513                                   QDict *options, Error **errp)
514 {
515     QemuOpts *opts;
516     SocketAddress *gsconf = NULL;
517     SocketAddressList **tail;
518     QDict *backing_options = NULL;
519     Error *local_err = NULL;
520     char *str = NULL;
521     const char *ptr;
522     int i, type, num_servers;
523 
524     /* create opts info from runtime_json_opts list */
525     opts = qemu_opts_create(&runtime_json_opts, NULL, 0, &error_abort);
526     if (!qemu_opts_absorb_qdict(opts, options, errp)) {
527         goto out;
528     }
529 
530     num_servers = qdict_array_entries(options, GLUSTER_OPT_SERVER_PATTERN);
531     if (num_servers < 1) {
532         error_setg(&local_err, QERR_MISSING_PARAMETER, "server");
533         goto out;
534     }
535 
536     ptr = qemu_opt_get(opts, GLUSTER_OPT_VOLUME);
537     if (!ptr) {
538         error_setg(&local_err, QERR_MISSING_PARAMETER, GLUSTER_OPT_VOLUME);
539         goto out;
540     }
541     gconf->volume = g_strdup(ptr);
542 
543     ptr = qemu_opt_get(opts, GLUSTER_OPT_PATH);
544     if (!ptr) {
545         error_setg(&local_err, QERR_MISSING_PARAMETER, GLUSTER_OPT_PATH);
546         goto out;
547     }
548     gconf->path = g_strdup(ptr);
549     qemu_opts_del(opts);
550     tail = &gconf->server;
551 
552     for (i = 0; i < num_servers; i++) {
553         str = g_strdup_printf(GLUSTER_OPT_SERVER_PATTERN"%d.", i);
554         qdict_extract_subqdict(options, &backing_options, str);
555 
556         /* create opts info from runtime_type_opts list */
557         opts = qemu_opts_create(&runtime_type_opts, NULL, 0, &error_abort);
558         if (!qemu_opts_absorb_qdict(opts, backing_options, errp)) {
559             goto out;
560         }
561 
562         ptr = qemu_opt_get(opts, GLUSTER_OPT_TYPE);
563         if (!ptr) {
564             error_setg(&local_err, QERR_MISSING_PARAMETER, GLUSTER_OPT_TYPE);
565             error_append_hint(&local_err, GERR_INDEX_HINT, i);
566             goto out;
567 
568         }
569         gsconf = g_new0(SocketAddress, 1);
570         if (!strcmp(ptr, "tcp")) {
571             ptr = "inet";       /* accept legacy "tcp" */
572         }
573         type = qapi_enum_parse(&SocketAddressType_lookup, ptr, -1, NULL);
574         if (type != SOCKET_ADDRESS_TYPE_INET
575             && type != SOCKET_ADDRESS_TYPE_UNIX) {
576             error_setg(&local_err,
577                        "Parameter '%s' may be 'inet' or 'unix'",
578                        GLUSTER_OPT_TYPE);
579             error_append_hint(&local_err, GERR_INDEX_HINT, i);
580             goto out;
581         }
582         gsconf->type = type;
583         qemu_opts_del(opts);
584 
585         if (gsconf->type == SOCKET_ADDRESS_TYPE_INET) {
586             /* create opts info from runtime_inet_opts list */
587             opts = qemu_opts_create(&runtime_inet_opts, NULL, 0, &error_abort);
588             if (!qemu_opts_absorb_qdict(opts, backing_options, errp)) {
589                 goto out;
590             }
591 
592             ptr = qemu_opt_get(opts, GLUSTER_OPT_HOST);
593             if (!ptr) {
594                 error_setg(&local_err, QERR_MISSING_PARAMETER,
595                            GLUSTER_OPT_HOST);
596                 error_append_hint(&local_err, GERR_INDEX_HINT, i);
597                 goto out;
598             }
599             gsconf->u.inet.host = g_strdup(ptr);
600             ptr = qemu_opt_get(opts, GLUSTER_OPT_PORT);
601             if (!ptr) {
602                 error_setg(&local_err, QERR_MISSING_PARAMETER,
603                            GLUSTER_OPT_PORT);
604                 error_append_hint(&local_err, GERR_INDEX_HINT, i);
605                 goto out;
606             }
607             gsconf->u.inet.port = g_strdup(ptr);
608 
609             /* defend for unsupported fields in InetSocketAddress,
610              * i.e. @ipv4, @ipv6  and @to
611              */
612             ptr = qemu_opt_get(opts, GLUSTER_OPT_TO);
613             if (ptr) {
614                 gsconf->u.inet.has_to = true;
615             }
616             ptr = qemu_opt_get(opts, GLUSTER_OPT_IPV4);
617             if (ptr) {
618                 gsconf->u.inet.has_ipv4 = true;
619             }
620             ptr = qemu_opt_get(opts, GLUSTER_OPT_IPV6);
621             if (ptr) {
622                 gsconf->u.inet.has_ipv6 = true;
623             }
624             if (gsconf->u.inet.has_to) {
625                 error_setg(&local_err, "Parameter 'to' not supported");
626                 goto out;
627             }
628             if (gsconf->u.inet.has_ipv4 || gsconf->u.inet.has_ipv6) {
629                 error_setg(&local_err, "Parameters 'ipv4/ipv6' not supported");
630                 goto out;
631             }
632             qemu_opts_del(opts);
633         } else {
634             /* create opts info from runtime_unix_opts list */
635             opts = qemu_opts_create(&runtime_unix_opts, NULL, 0, &error_abort);
636             if (!qemu_opts_absorb_qdict(opts, backing_options, errp)) {
637                 goto out;
638             }
639 
640             ptr = qemu_opt_get(opts, GLUSTER_OPT_PATH);
641             if (!ptr) {
642                 ptr = qemu_opt_get(opts, GLUSTER_OPT_SOCKET);
643             } else if (qemu_opt_get(opts, GLUSTER_OPT_SOCKET)) {
644                 error_setg(&local_err,
645                            "Conflicting parameters 'path' and 'socket'");
646                 error_append_hint(&local_err, GERR_INDEX_HINT, i);
647                 goto out;
648             }
649             if (!ptr) {
650                 error_setg(&local_err, QERR_MISSING_PARAMETER,
651                            GLUSTER_OPT_PATH);
652                 error_append_hint(&local_err, GERR_INDEX_HINT, i);
653                 goto out;
654             }
655             gsconf->u.q_unix.path = g_strdup(ptr);
656             qemu_opts_del(opts);
657         }
658 
659         QAPI_LIST_APPEND(tail, gsconf);
660         gsconf = NULL;
661 
662         qobject_unref(backing_options);
663         backing_options = NULL;
664         g_free(str);
665         str = NULL;
666     }
667 
668     return 0;
669 
670 out:
671     error_propagate(errp, local_err);
672     qapi_free_SocketAddress(gsconf);
673     qemu_opts_del(opts);
674     g_free(str);
675     qobject_unref(backing_options);
676     errno = EINVAL;
677     return -errno;
678 }
679 
680 /* Converts options given in @filename and the @options QDict into the QAPI
681  * object @gconf. */
682 static int qemu_gluster_parse(BlockdevOptionsGluster *gconf,
683                               const char *filename,
684                               QDict *options, Error **errp)
685 {
686     int ret;
687     if (filename) {
688         ret = qemu_gluster_parse_uri(gconf, filename);
689         if (ret < 0) {
690             error_setg(errp, "invalid URI %s", filename);
691             error_append_hint(errp, "Usage: file=gluster[+transport]://"
692                                     "[host[:port]]volume/path[?socket=...]"
693                                     "[,file.debug=N]"
694                                     "[,file.logfile=/path/filename.log]\n");
695             return ret;
696         }
697     } else {
698         ret = qemu_gluster_parse_json(gconf, options, errp);
699         if (ret < 0) {
700             error_append_hint(errp, "Usage: "
701                              "-drive driver=qcow2,file.driver=gluster,"
702                              "file.volume=testvol,file.path=/path/a.qcow2"
703                              "[,file.debug=9]"
704                              "[,file.logfile=/path/filename.log],"
705                              "file.server.0.type=inet,"
706                              "file.server.0.host=1.2.3.4,"
707                              "file.server.0.port=24007,"
708                              "file.server.1.transport=unix,"
709                              "file.server.1.path=/var/run/glusterd.socket ..."
710                              "\n");
711             return ret;
712         }
713     }
714 
715     return 0;
716 }
717 
718 static struct glfs *qemu_gluster_init(BlockdevOptionsGluster *gconf,
719                                       const char *filename,
720                                       QDict *options, Error **errp)
721 {
722     int ret;
723 
724     ret = qemu_gluster_parse(gconf, filename, options, errp);
725     if (ret < 0) {
726         errno = -ret;
727         return NULL;
728     }
729 
730     return qemu_gluster_glfs_init(gconf, errp);
731 }
732 
733 /*
734  * AIO callback routine called from GlusterFS thread.
735  */
736 static void gluster_finish_aiocb(struct glfs_fd *fd, ssize_t ret,
737 #ifdef CONFIG_GLUSTERFS_IOCB_HAS_STAT
738                                  struct glfs_stat *pre, struct glfs_stat *post,
739 #endif
740                                  void *arg)
741 {
742     GlusterAIOCB *acb = (GlusterAIOCB *)arg;
743 
744     if (!ret || ret == acb->size) {
745         acb->ret = 0; /* Success */
746     } else if (ret < 0) {
747         acb->ret = -errno; /* Read/Write failed */
748     } else {
749         acb->ret = -EIO; /* Partial read/write - fail it */
750     }
751 
752     aio_co_schedule(acb->aio_context, acb->coroutine);
753 }
754 
755 static void qemu_gluster_parse_flags(int bdrv_flags, int *open_flags)
756 {
757     assert(open_flags != NULL);
758 
759     *open_flags |= O_BINARY;
760 
761     if (bdrv_flags & BDRV_O_RDWR) {
762         *open_flags |= O_RDWR;
763     } else {
764         *open_flags |= O_RDONLY;
765     }
766 
767     if ((bdrv_flags & BDRV_O_NOCACHE)) {
768         *open_flags |= O_DIRECT;
769     }
770 }
771 
772 /*
773  * Do SEEK_DATA/HOLE to detect if it is functional. Older broken versions of
774  * gfapi incorrectly return the current offset when SEEK_DATA/HOLE is used.
775  * - Corrected versions return -1 and set errno to EINVAL.
776  * - Versions that support SEEK_DATA/HOLE correctly, will return -1 and set
777  *   errno to ENXIO when SEEK_DATA is called with a position of EOF.
778  */
779 static bool qemu_gluster_test_seek(struct glfs_fd *fd)
780 {
781     off_t ret = 0;
782 
783 #if defined SEEK_HOLE && defined SEEK_DATA
784     off_t eof;
785 
786     eof = glfs_lseek(fd, 0, SEEK_END);
787     if (eof < 0) {
788         /* this should never occur */
789         return false;
790     }
791 
792     /* this should always fail with ENXIO if SEEK_DATA is supported */
793     ret = glfs_lseek(fd, eof, SEEK_DATA);
794 #endif
795 
796     return (ret < 0) && (errno == ENXIO);
797 }
798 
799 static int qemu_gluster_open(BlockDriverState *bs,  QDict *options,
800                              int bdrv_flags, Error **errp)
801 {
802     BDRVGlusterState *s = bs->opaque;
803     int open_flags = 0;
804     int ret = 0;
805     BlockdevOptionsGluster *gconf = NULL;
806     QemuOpts *opts;
807     const char *filename, *logfile;
808 
809     opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
810     if (!qemu_opts_absorb_qdict(opts, options, errp)) {
811         ret = -EINVAL;
812         goto out;
813     }
814 
815     filename = qemu_opt_get(opts, GLUSTER_OPT_FILENAME);
816 
817     s->debug = qemu_opt_get_number(opts, GLUSTER_OPT_DEBUG,
818                                    GLUSTER_DEBUG_DEFAULT);
819     if (s->debug < 0) {
820         s->debug = 0;
821     } else if (s->debug > GLUSTER_DEBUG_MAX) {
822         s->debug = GLUSTER_DEBUG_MAX;
823     }
824 
825     gconf = g_new0(BlockdevOptionsGluster, 1);
826     gconf->debug = s->debug;
827     gconf->has_debug = true;
828 
829     logfile = qemu_opt_get(opts, GLUSTER_OPT_LOGFILE);
830     s->logfile = g_strdup(logfile ? logfile : GLUSTER_LOGFILE_DEFAULT);
831 
832     gconf->logfile = g_strdup(s->logfile);
833     gconf->has_logfile = true;
834 
835     s->glfs = qemu_gluster_init(gconf, filename, options, errp);
836     if (!s->glfs) {
837         ret = -errno;
838         goto out;
839     }
840 
841 #ifdef CONFIG_GLUSTERFS_XLATOR_OPT
842     /* Without this, if fsync fails for a recoverable reason (for instance,
843      * ENOSPC), gluster will dump its cache, preventing retries.  This means
844      * almost certain data loss.  Not all gluster versions support the
845      * 'resync-failed-syncs-after-fsync' key value, but there is no way to
846      * discover during runtime if it is supported (this api returns success for
847      * unknown key/value pairs) */
848     ret = glfs_set_xlator_option(s->glfs, "*-write-behind",
849                                           "resync-failed-syncs-after-fsync",
850                                           "on");
851     if (ret < 0) {
852         error_setg_errno(errp, errno, "Unable to set xlator key/value pair");
853         ret = -errno;
854         goto out;
855     }
856 #endif
857 
858     qemu_gluster_parse_flags(bdrv_flags, &open_flags);
859 
860     s->fd = glfs_open(s->glfs, gconf->path, open_flags);
861     ret = s->fd ? 0 : -errno;
862 
863     if (ret == -EACCES || ret == -EROFS) {
864         /* Try to degrade to read-only, but if it doesn't work, still use the
865          * normal error message. */
866         if (bdrv_apply_auto_read_only(bs, NULL, NULL) == 0) {
867             open_flags = (open_flags & ~O_RDWR) | O_RDONLY;
868             s->fd = glfs_open(s->glfs, gconf->path, open_flags);
869             ret = s->fd ? 0 : -errno;
870         }
871     }
872 
873     s->supports_seek_data = qemu_gluster_test_seek(s->fd);
874 
875 out:
876     qemu_opts_del(opts);
877     qapi_free_BlockdevOptionsGluster(gconf);
878     if (!ret) {
879         return ret;
880     }
881     g_free(s->logfile);
882     if (s->fd) {
883         glfs_close(s->fd);
884     }
885 
886     glfs_clear_preopened(s->glfs);
887 
888     return ret;
889 }
890 
891 static void qemu_gluster_refresh_limits(BlockDriverState *bs, Error **errp)
892 {
893     bs->bl.max_transfer = GLUSTER_MAX_TRANSFER;
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     gconf->has_logfile = true;
920 
921     /*
922      * If 'state->bs->exact_filename' is empty, 'state->options' should contain
923      * the JSON parameters already parsed.
924      */
925     if (state->bs->exact_filename[0] != '\0') {
926         reop_s->glfs = qemu_gluster_init(gconf, state->bs->exact_filename, NULL,
927                                          errp);
928     } else {
929         reop_s->glfs = qemu_gluster_init(gconf, NULL, state->options, errp);
930     }
931     if (reop_s->glfs == NULL) {
932         ret = -errno;
933         goto exit;
934     }
935 
936 #ifdef CONFIG_GLUSTERFS_XLATOR_OPT
937     ret = glfs_set_xlator_option(reop_s->glfs, "*-write-behind",
938                                  "resync-failed-syncs-after-fsync", "on");
939     if (ret < 0) {
940         error_setg_errno(errp, errno, "Unable to set xlator key/value pair");
941         ret = -errno;
942         goto exit;
943     }
944 #endif
945 
946     reop_s->fd = glfs_open(reop_s->glfs, gconf->path, open_flags);
947     if (reop_s->fd == NULL) {
948         /* reops->glfs will be cleaned up in _abort */
949         ret = -errno;
950         goto exit;
951     }
952 
953 exit:
954     /* state->opaque will be freed in either the _abort or _commit */
955     qapi_free_BlockdevOptionsGluster(gconf);
956     return ret;
957 }
958 
959 static void qemu_gluster_reopen_commit(BDRVReopenState *state)
960 {
961     BDRVGlusterReopenState *reop_s = state->opaque;
962     BDRVGlusterState *s = state->bs->opaque;
963 
964 
965     /* close the old */
966     if (s->fd) {
967         glfs_close(s->fd);
968     }
969 
970     glfs_clear_preopened(s->glfs);
971 
972     /* use the newly opened image / connection */
973     s->fd         = reop_s->fd;
974     s->glfs       = reop_s->glfs;
975 
976     g_free(state->opaque);
977     state->opaque = NULL;
978 
979     return;
980 }
981 
982 
983 static void qemu_gluster_reopen_abort(BDRVReopenState *state)
984 {
985     BDRVGlusterReopenState *reop_s = state->opaque;
986 
987     if (reop_s == NULL) {
988         return;
989     }
990 
991     if (reop_s->fd) {
992         glfs_close(reop_s->fd);
993     }
994 
995     glfs_clear_preopened(reop_s->glfs);
996 
997     g_free(state->opaque);
998     state->opaque = NULL;
999 
1000     return;
1001 }
1002 
1003 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1004 static coroutine_fn int qemu_gluster_co_pwrite_zeroes(BlockDriverState *bs,
1005                                                       int64_t offset,
1006                                                       int size,
1007                                                       BdrvRequestFlags flags)
1008 {
1009     int ret;
1010     GlusterAIOCB acb;
1011     BDRVGlusterState *s = bs->opaque;
1012 
1013     acb.size = size;
1014     acb.ret = 0;
1015     acb.coroutine = qemu_coroutine_self();
1016     acb.aio_context = bdrv_get_aio_context(bs);
1017 
1018     ret = glfs_zerofill_async(s->fd, offset, size, gluster_finish_aiocb, &acb);
1019     if (ret < 0) {
1020         return -errno;
1021     }
1022 
1023     qemu_coroutine_yield();
1024     return acb.ret;
1025 }
1026 #endif
1027 
1028 static int qemu_gluster_do_truncate(struct glfs_fd *fd, int64_t offset,
1029                                     PreallocMode prealloc, Error **errp)
1030 {
1031     int64_t current_length;
1032 
1033     current_length = glfs_lseek(fd, 0, SEEK_END);
1034     if (current_length < 0) {
1035         error_setg_errno(errp, errno, "Failed to determine current size");
1036         return -errno;
1037     }
1038 
1039     if (current_length > offset && prealloc != PREALLOC_MODE_OFF) {
1040         error_setg(errp, "Cannot use preallocation for shrinking files");
1041         return -ENOTSUP;
1042     }
1043 
1044     if (current_length == offset) {
1045         return 0;
1046     }
1047 
1048     switch (prealloc) {
1049 #ifdef CONFIG_GLUSTERFS_FALLOCATE
1050     case PREALLOC_MODE_FALLOC:
1051         if (glfs_fallocate(fd, 0, current_length, offset - current_length)) {
1052             error_setg_errno(errp, errno, "Could not preallocate data");
1053             return -errno;
1054         }
1055         break;
1056 #endif /* CONFIG_GLUSTERFS_FALLOCATE */
1057 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1058     case PREALLOC_MODE_FULL:
1059         if (glfs_ftruncate(fd, offset)) {
1060             error_setg_errno(errp, errno, "Could not resize file");
1061             return -errno;
1062         }
1063         if (glfs_zerofill(fd, current_length, offset - current_length)) {
1064             error_setg_errno(errp, errno, "Could not zerofill the new area");
1065             return -errno;
1066         }
1067         break;
1068 #endif /* CONFIG_GLUSTERFS_ZEROFILL */
1069     case PREALLOC_MODE_OFF:
1070         if (glfs_ftruncate(fd, offset)) {
1071             error_setg_errno(errp, errno, "Could not resize file");
1072             return -errno;
1073         }
1074         break;
1075     default:
1076         error_setg(errp, "Unsupported preallocation mode: %s",
1077                    PreallocMode_str(prealloc));
1078         return -EINVAL;
1079     }
1080 
1081     return 0;
1082 }
1083 
1084 static int qemu_gluster_co_create(BlockdevCreateOptions *options,
1085                                   Error **errp)
1086 {
1087     BlockdevCreateOptionsGluster *opts = &options->u.gluster;
1088     struct glfs *glfs;
1089     struct glfs_fd *fd = NULL;
1090     int ret = 0;
1091 
1092     assert(options->driver == BLOCKDEV_DRIVER_GLUSTER);
1093 
1094     glfs = qemu_gluster_glfs_init(opts->location, errp);
1095     if (!glfs) {
1096         ret = -errno;
1097         goto out;
1098     }
1099 
1100     fd = glfs_creat(glfs, opts->location->path,
1101                     O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, S_IRUSR | S_IWUSR);
1102     if (!fd) {
1103         ret = -errno;
1104         goto out;
1105     }
1106 
1107     ret = qemu_gluster_do_truncate(fd, opts->size, opts->preallocation, errp);
1108 
1109 out:
1110     if (fd) {
1111         if (glfs_close(fd) != 0 && ret == 0) {
1112             ret = -errno;
1113         }
1114     }
1115     glfs_clear_preopened(glfs);
1116     return ret;
1117 }
1118 
1119 static int coroutine_fn qemu_gluster_co_create_opts(BlockDriver *drv,
1120                                                     const char *filename,
1121                                                     QemuOpts *opts,
1122                                                     Error **errp)
1123 {
1124     BlockdevCreateOptions *options;
1125     BlockdevCreateOptionsGluster *gopts;
1126     BlockdevOptionsGluster *gconf;
1127     char *tmp = NULL;
1128     Error *local_err = NULL;
1129     int ret;
1130 
1131     options = g_new0(BlockdevCreateOptions, 1);
1132     options->driver = BLOCKDEV_DRIVER_GLUSTER;
1133     gopts = &options->u.gluster;
1134 
1135     gconf = g_new0(BlockdevOptionsGluster, 1);
1136     gopts->location = gconf;
1137 
1138     gopts->size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
1139                            BDRV_SECTOR_SIZE);
1140 
1141     tmp = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
1142     gopts->preallocation = qapi_enum_parse(&PreallocMode_lookup, tmp,
1143                                            PREALLOC_MODE_OFF, &local_err);
1144     g_free(tmp);
1145     if (local_err) {
1146         error_propagate(errp, local_err);
1147         ret = -EINVAL;
1148         goto fail;
1149     }
1150 
1151     gconf->debug = qemu_opt_get_number_del(opts, GLUSTER_OPT_DEBUG,
1152                                            GLUSTER_DEBUG_DEFAULT);
1153     if (gconf->debug < 0) {
1154         gconf->debug = 0;
1155     } else if (gconf->debug > GLUSTER_DEBUG_MAX) {
1156         gconf->debug = GLUSTER_DEBUG_MAX;
1157     }
1158     gconf->has_debug = true;
1159 
1160     gconf->logfile = qemu_opt_get_del(opts, GLUSTER_OPT_LOGFILE);
1161     if (!gconf->logfile) {
1162         gconf->logfile = g_strdup(GLUSTER_LOGFILE_DEFAULT);
1163     }
1164     gconf->has_logfile = true;
1165 
1166     ret = qemu_gluster_parse(gconf, filename, NULL, errp);
1167     if (ret < 0) {
1168         goto fail;
1169     }
1170 
1171     ret = qemu_gluster_co_create(options, errp);
1172     if (ret < 0) {
1173         goto fail;
1174     }
1175 
1176     ret = 0;
1177 fail:
1178     qapi_free_BlockdevCreateOptions(options);
1179     return ret;
1180 }
1181 
1182 static coroutine_fn int qemu_gluster_co_rw(BlockDriverState *bs,
1183                                            int64_t sector_num, int nb_sectors,
1184                                            QEMUIOVector *qiov, int write)
1185 {
1186     int ret;
1187     GlusterAIOCB acb;
1188     BDRVGlusterState *s = bs->opaque;
1189     size_t size = nb_sectors * BDRV_SECTOR_SIZE;
1190     off_t offset = sector_num * BDRV_SECTOR_SIZE;
1191 
1192     acb.size = size;
1193     acb.ret = 0;
1194     acb.coroutine = qemu_coroutine_self();
1195     acb.aio_context = bdrv_get_aio_context(bs);
1196 
1197     if (write) {
1198         ret = glfs_pwritev_async(s->fd, qiov->iov, qiov->niov, offset, 0,
1199                                  gluster_finish_aiocb, &acb);
1200     } else {
1201         ret = glfs_preadv_async(s->fd, qiov->iov, qiov->niov, offset, 0,
1202                                 gluster_finish_aiocb, &acb);
1203     }
1204 
1205     if (ret < 0) {
1206         return -errno;
1207     }
1208 
1209     qemu_coroutine_yield();
1210     return acb.ret;
1211 }
1212 
1213 static coroutine_fn int qemu_gluster_co_truncate(BlockDriverState *bs,
1214                                                  int64_t offset,
1215                                                  bool exact,
1216                                                  PreallocMode prealloc,
1217                                                  BdrvRequestFlags flags,
1218                                                  Error **errp)
1219 {
1220     BDRVGlusterState *s = bs->opaque;
1221     return qemu_gluster_do_truncate(s->fd, offset, prealloc, errp);
1222 }
1223 
1224 static coroutine_fn int qemu_gluster_co_readv(BlockDriverState *bs,
1225                                               int64_t sector_num,
1226                                               int nb_sectors,
1227                                               QEMUIOVector *qiov)
1228 {
1229     return qemu_gluster_co_rw(bs, sector_num, nb_sectors, qiov, 0);
1230 }
1231 
1232 static coroutine_fn int qemu_gluster_co_writev(BlockDriverState *bs,
1233                                                int64_t sector_num,
1234                                                int nb_sectors,
1235                                                QEMUIOVector *qiov,
1236                                                int flags)
1237 {
1238     assert(!flags);
1239     return qemu_gluster_co_rw(bs, sector_num, nb_sectors, qiov, 1);
1240 }
1241 
1242 static void qemu_gluster_close(BlockDriverState *bs)
1243 {
1244     BDRVGlusterState *s = bs->opaque;
1245 
1246     g_free(s->logfile);
1247     if (s->fd) {
1248         glfs_close(s->fd);
1249         s->fd = NULL;
1250     }
1251     glfs_clear_preopened(s->glfs);
1252 }
1253 
1254 static coroutine_fn int qemu_gluster_co_flush_to_disk(BlockDriverState *bs)
1255 {
1256     int ret;
1257     GlusterAIOCB acb;
1258     BDRVGlusterState *s = bs->opaque;
1259 
1260     acb.size = 0;
1261     acb.ret = 0;
1262     acb.coroutine = qemu_coroutine_self();
1263     acb.aio_context = bdrv_get_aio_context(bs);
1264 
1265     ret = glfs_fsync_async(s->fd, gluster_finish_aiocb, &acb);
1266     if (ret < 0) {
1267         ret = -errno;
1268         goto error;
1269     }
1270 
1271     qemu_coroutine_yield();
1272     if (acb.ret < 0) {
1273         ret = acb.ret;
1274         goto error;
1275     }
1276 
1277     return acb.ret;
1278 
1279 error:
1280     /* Some versions of Gluster (3.5.6 -> 3.5.8?) will not retain its cache
1281      * after a fsync failure, so we have no way of allowing the guest to safely
1282      * continue.  Gluster versions prior to 3.5.6 don't retain the cache
1283      * either, but will invalidate the fd on error, so this is again our only
1284      * option.
1285      *
1286      * The 'resync-failed-syncs-after-fsync' xlator option for the
1287      * write-behind cache will cause later gluster versions to retain its
1288      * cache after error, so long as the fd remains open.  However, we
1289      * currently have no way of knowing if this option is supported.
1290      *
1291      * TODO: Once gluster provides a way for us to determine if the option
1292      * is supported, bypass the closure and setting drv to NULL.  */
1293     qemu_gluster_close(bs);
1294     bs->drv = NULL;
1295     return ret;
1296 }
1297 
1298 #ifdef CONFIG_GLUSTERFS_DISCARD
1299 static coroutine_fn int qemu_gluster_co_pdiscard(BlockDriverState *bs,
1300                                                  int64_t offset, int size)
1301 {
1302     int ret;
1303     GlusterAIOCB acb;
1304     BDRVGlusterState *s = bs->opaque;
1305 
1306     acb.size = 0;
1307     acb.ret = 0;
1308     acb.coroutine = qemu_coroutine_self();
1309     acb.aio_context = bdrv_get_aio_context(bs);
1310 
1311     ret = glfs_discard_async(s->fd, offset, size, gluster_finish_aiocb, &acb);
1312     if (ret < 0) {
1313         return -errno;
1314     }
1315 
1316     qemu_coroutine_yield();
1317     return acb.ret;
1318 }
1319 #endif
1320 
1321 static int64_t qemu_gluster_getlength(BlockDriverState *bs)
1322 {
1323     BDRVGlusterState *s = bs->opaque;
1324     int64_t ret;
1325 
1326     ret = glfs_lseek(s->fd, 0, SEEK_END);
1327     if (ret < 0) {
1328         return -errno;
1329     } else {
1330         return ret;
1331     }
1332 }
1333 
1334 static int64_t qemu_gluster_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_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_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 #ifdef CONFIG_GLUSTERFS_DISCARD
1570     .bdrv_co_pdiscard             = qemu_gluster_co_pdiscard,
1571 #endif
1572 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1573     .bdrv_co_pwrite_zeroes        = qemu_gluster_co_pwrite_zeroes,
1574 #endif
1575     .bdrv_co_block_status         = qemu_gluster_co_block_status,
1576     .bdrv_refresh_limits          = qemu_gluster_refresh_limits,
1577     .create_opts                  = &qemu_gluster_create_opts,
1578     .strong_runtime_opts          = gluster_strong_open_opts,
1579 };
1580 
1581 static BlockDriver bdrv_gluster_tcp = {
1582     .format_name                  = "gluster",
1583     .protocol_name                = "gluster+tcp",
1584     .instance_size                = sizeof(BDRVGlusterState),
1585     .bdrv_needs_filename          = false,
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 #ifdef CONFIG_GLUSTERFS_DISCARD
1600     .bdrv_co_pdiscard             = qemu_gluster_co_pdiscard,
1601 #endif
1602 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1603     .bdrv_co_pwrite_zeroes        = qemu_gluster_co_pwrite_zeroes,
1604 #endif
1605     .bdrv_co_block_status         = qemu_gluster_co_block_status,
1606     .bdrv_refresh_limits          = qemu_gluster_refresh_limits,
1607     .create_opts                  = &qemu_gluster_create_opts,
1608     .strong_runtime_opts          = gluster_strong_open_opts,
1609 };
1610 
1611 static BlockDriver bdrv_gluster_unix = {
1612     .format_name                  = "gluster",
1613     .protocol_name                = "gluster+unix",
1614     .instance_size                = sizeof(BDRVGlusterState),
1615     .bdrv_needs_filename          = true,
1616     .bdrv_file_open               = qemu_gluster_open,
1617     .bdrv_reopen_prepare          = qemu_gluster_reopen_prepare,
1618     .bdrv_reopen_commit           = qemu_gluster_reopen_commit,
1619     .bdrv_reopen_abort            = qemu_gluster_reopen_abort,
1620     .bdrv_close                   = qemu_gluster_close,
1621     .bdrv_co_create               = qemu_gluster_co_create,
1622     .bdrv_co_create_opts          = qemu_gluster_co_create_opts,
1623     .bdrv_getlength               = qemu_gluster_getlength,
1624     .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size,
1625     .bdrv_co_truncate             = qemu_gluster_co_truncate,
1626     .bdrv_co_readv                = qemu_gluster_co_readv,
1627     .bdrv_co_writev               = qemu_gluster_co_writev,
1628     .bdrv_co_flush_to_disk        = qemu_gluster_co_flush_to_disk,
1629 #ifdef CONFIG_GLUSTERFS_DISCARD
1630     .bdrv_co_pdiscard             = qemu_gluster_co_pdiscard,
1631 #endif
1632 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1633     .bdrv_co_pwrite_zeroes        = qemu_gluster_co_pwrite_zeroes,
1634 #endif
1635     .bdrv_co_block_status         = qemu_gluster_co_block_status,
1636     .bdrv_refresh_limits          = qemu_gluster_refresh_limits,
1637     .create_opts                  = &qemu_gluster_create_opts,
1638     .strong_runtime_opts          = gluster_strong_open_opts,
1639 };
1640 
1641 /* rdma is deprecated (actually never supported for volfile fetch).
1642  * Let's maintain it for the protocol compatibility, to make sure things
1643  * won't break immediately. For now, gluster+rdma will fall back to gluster+tcp
1644  * protocol with a warning.
1645  * TODO: remove gluster+rdma interface support
1646  */
1647 static BlockDriver bdrv_gluster_rdma = {
1648     .format_name                  = "gluster",
1649     .protocol_name                = "gluster+rdma",
1650     .instance_size                = sizeof(BDRVGlusterState),
1651     .bdrv_needs_filename          = true,
1652     .bdrv_file_open               = qemu_gluster_open,
1653     .bdrv_reopen_prepare          = qemu_gluster_reopen_prepare,
1654     .bdrv_reopen_commit           = qemu_gluster_reopen_commit,
1655     .bdrv_reopen_abort            = qemu_gluster_reopen_abort,
1656     .bdrv_close                   = qemu_gluster_close,
1657     .bdrv_co_create               = qemu_gluster_co_create,
1658     .bdrv_co_create_opts          = qemu_gluster_co_create_opts,
1659     .bdrv_getlength               = qemu_gluster_getlength,
1660     .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size,
1661     .bdrv_co_truncate             = qemu_gluster_co_truncate,
1662     .bdrv_co_readv                = qemu_gluster_co_readv,
1663     .bdrv_co_writev               = qemu_gluster_co_writev,
1664     .bdrv_co_flush_to_disk        = qemu_gluster_co_flush_to_disk,
1665 #ifdef CONFIG_GLUSTERFS_DISCARD
1666     .bdrv_co_pdiscard             = qemu_gluster_co_pdiscard,
1667 #endif
1668 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1669     .bdrv_co_pwrite_zeroes        = qemu_gluster_co_pwrite_zeroes,
1670 #endif
1671     .bdrv_co_block_status         = qemu_gluster_co_block_status,
1672     .bdrv_refresh_limits          = qemu_gluster_refresh_limits,
1673     .create_opts                  = &qemu_gluster_create_opts,
1674     .strong_runtime_opts          = gluster_strong_open_opts,
1675 };
1676 
1677 static void bdrv_gluster_init(void)
1678 {
1679     bdrv_register(&bdrv_gluster_rdma);
1680     bdrv_register(&bdrv_gluster_unix);
1681     bdrv_register(&bdrv_gluster_tcp);
1682     bdrv_register(&bdrv_gluster);
1683 }
1684 
1685 block_init(bdrv_gluster_init);
1686