xref: /openbmc/qemu/block/gluster.c (revision fd38b162)
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     gconf->server = g_new0(SocketAddressList, 1);
363     gconf->server->value = gsconf = g_new0(SocketAddress, 1);
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 *curr = NULL;
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     qemu_opts_absorb_qdict(opts, options, &local_err);
527     if (local_err) {
528         goto out;
529     }
530 
531     num_servers = qdict_array_entries(options, GLUSTER_OPT_SERVER_PATTERN);
532     if (num_servers < 1) {
533         error_setg(&local_err, QERR_MISSING_PARAMETER, "server");
534         goto out;
535     }
536 
537     ptr = qemu_opt_get(opts, GLUSTER_OPT_VOLUME);
538     if (!ptr) {
539         error_setg(&local_err, QERR_MISSING_PARAMETER, GLUSTER_OPT_VOLUME);
540         goto out;
541     }
542     gconf->volume = g_strdup(ptr);
543 
544     ptr = qemu_opt_get(opts, GLUSTER_OPT_PATH);
545     if (!ptr) {
546         error_setg(&local_err, QERR_MISSING_PARAMETER, GLUSTER_OPT_PATH);
547         goto out;
548     }
549     gconf->path = g_strdup(ptr);
550     qemu_opts_del(opts);
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         qemu_opts_absorb_qdict(opts, backing_options, &local_err);
559         if (local_err) {
560             goto out;
561         }
562 
563         ptr = qemu_opt_get(opts, GLUSTER_OPT_TYPE);
564         if (!ptr) {
565             error_setg(&local_err, QERR_MISSING_PARAMETER, GLUSTER_OPT_TYPE);
566             error_append_hint(&local_err, GERR_INDEX_HINT, i);
567             goto out;
568 
569         }
570         gsconf = g_new0(SocketAddress, 1);
571         if (!strcmp(ptr, "tcp")) {
572             ptr = "inet";       /* accept legacy "tcp" */
573         }
574         type = qapi_enum_parse(&SocketAddressType_lookup, ptr, -1, NULL);
575         if (type != SOCKET_ADDRESS_TYPE_INET
576             && type != SOCKET_ADDRESS_TYPE_UNIX) {
577             error_setg(&local_err,
578                        "Parameter '%s' may be 'inet' or 'unix'",
579                        GLUSTER_OPT_TYPE);
580             error_append_hint(&local_err, GERR_INDEX_HINT, i);
581             goto out;
582         }
583         gsconf->type = type;
584         qemu_opts_del(opts);
585 
586         if (gsconf->type == SOCKET_ADDRESS_TYPE_INET) {
587             /* create opts info from runtime_inet_opts list */
588             opts = qemu_opts_create(&runtime_inet_opts, NULL, 0, &error_abort);
589             qemu_opts_absorb_qdict(opts, backing_options, &local_err);
590             if (local_err) {
591                 goto out;
592             }
593 
594             ptr = qemu_opt_get(opts, GLUSTER_OPT_HOST);
595             if (!ptr) {
596                 error_setg(&local_err, QERR_MISSING_PARAMETER,
597                            GLUSTER_OPT_HOST);
598                 error_append_hint(&local_err, GERR_INDEX_HINT, i);
599                 goto out;
600             }
601             gsconf->u.inet.host = g_strdup(ptr);
602             ptr = qemu_opt_get(opts, GLUSTER_OPT_PORT);
603             if (!ptr) {
604                 error_setg(&local_err, QERR_MISSING_PARAMETER,
605                            GLUSTER_OPT_PORT);
606                 error_append_hint(&local_err, GERR_INDEX_HINT, i);
607                 goto out;
608             }
609             gsconf->u.inet.port = g_strdup(ptr);
610 
611             /* defend for unsupported fields in InetSocketAddress,
612              * i.e. @ipv4, @ipv6  and @to
613              */
614             ptr = qemu_opt_get(opts, GLUSTER_OPT_TO);
615             if (ptr) {
616                 gsconf->u.inet.has_to = true;
617             }
618             ptr = qemu_opt_get(opts, GLUSTER_OPT_IPV4);
619             if (ptr) {
620                 gsconf->u.inet.has_ipv4 = true;
621             }
622             ptr = qemu_opt_get(opts, GLUSTER_OPT_IPV6);
623             if (ptr) {
624                 gsconf->u.inet.has_ipv6 = true;
625             }
626             if (gsconf->u.inet.has_to) {
627                 error_setg(&local_err, "Parameter 'to' not supported");
628                 goto out;
629             }
630             if (gsconf->u.inet.has_ipv4 || gsconf->u.inet.has_ipv6) {
631                 error_setg(&local_err, "Parameters 'ipv4/ipv6' not supported");
632                 goto out;
633             }
634             qemu_opts_del(opts);
635         } else {
636             /* create opts info from runtime_unix_opts list */
637             opts = qemu_opts_create(&runtime_unix_opts, NULL, 0, &error_abort);
638             qemu_opts_absorb_qdict(opts, backing_options, &local_err);
639             if (local_err) {
640                 goto out;
641             }
642 
643             ptr = qemu_opt_get(opts, GLUSTER_OPT_PATH);
644             if (!ptr) {
645                 ptr = qemu_opt_get(opts, GLUSTER_OPT_SOCKET);
646             } else if (qemu_opt_get(opts, GLUSTER_OPT_SOCKET)) {
647                 error_setg(&local_err,
648                            "Conflicting parameters 'path' and 'socket'");
649                 error_append_hint(&local_err, GERR_INDEX_HINT, i);
650                 goto out;
651             }
652             if (!ptr) {
653                 error_setg(&local_err, QERR_MISSING_PARAMETER,
654                            GLUSTER_OPT_PATH);
655                 error_append_hint(&local_err, GERR_INDEX_HINT, i);
656                 goto out;
657             }
658             gsconf->u.q_unix.path = g_strdup(ptr);
659             qemu_opts_del(opts);
660         }
661 
662         if (gconf->server == NULL) {
663             gconf->server = g_new0(SocketAddressList, 1);
664             gconf->server->value = gsconf;
665             curr = gconf->server;
666         } else {
667             curr->next = g_new0(SocketAddressList, 1);
668             curr->next->value = gsconf;
669             curr = curr->next;
670         }
671         gsconf = NULL;
672 
673         qobject_unref(backing_options);
674         backing_options = NULL;
675         g_free(str);
676         str = NULL;
677     }
678 
679     return 0;
680 
681 out:
682     error_propagate(errp, local_err);
683     qapi_free_SocketAddress(gsconf);
684     qemu_opts_del(opts);
685     g_free(str);
686     qobject_unref(backing_options);
687     errno = EINVAL;
688     return -errno;
689 }
690 
691 /* Converts options given in @filename and the @options QDict into the QAPI
692  * object @gconf. */
693 static int qemu_gluster_parse(BlockdevOptionsGluster *gconf,
694                               const char *filename,
695                               QDict *options, Error **errp)
696 {
697     int ret;
698     if (filename) {
699         ret = qemu_gluster_parse_uri(gconf, filename);
700         if (ret < 0) {
701             error_setg(errp, "invalid URI %s", filename);
702             error_append_hint(errp, "Usage: file=gluster[+transport]://"
703                                     "[host[:port]]volume/path[?socket=...]"
704                                     "[,file.debug=N]"
705                                     "[,file.logfile=/path/filename.log]\n");
706             return ret;
707         }
708     } else {
709         ret = qemu_gluster_parse_json(gconf, options, errp);
710         if (ret < 0) {
711             error_append_hint(errp, "Usage: "
712                              "-drive driver=qcow2,file.driver=gluster,"
713                              "file.volume=testvol,file.path=/path/a.qcow2"
714                              "[,file.debug=9]"
715                              "[,file.logfile=/path/filename.log],"
716                              "file.server.0.type=inet,"
717                              "file.server.0.host=1.2.3.4,"
718                              "file.server.0.port=24007,"
719                              "file.server.1.transport=unix,"
720                              "file.server.1.path=/var/run/glusterd.socket ..."
721                              "\n");
722             return ret;
723         }
724     }
725 
726     return 0;
727 }
728 
729 static struct glfs *qemu_gluster_init(BlockdevOptionsGluster *gconf,
730                                       const char *filename,
731                                       QDict *options, Error **errp)
732 {
733     int ret;
734 
735     ret = qemu_gluster_parse(gconf, filename, options, errp);
736     if (ret < 0) {
737         errno = -ret;
738         return NULL;
739     }
740 
741     return qemu_gluster_glfs_init(gconf, errp);
742 }
743 
744 /*
745  * AIO callback routine called from GlusterFS thread.
746  */
747 static void gluster_finish_aiocb(struct glfs_fd *fd, ssize_t ret,
748 #ifdef CONFIG_GLUSTERFS_IOCB_HAS_STAT
749                                  struct glfs_stat *pre, struct glfs_stat *post,
750 #endif
751                                  void *arg)
752 {
753     GlusterAIOCB *acb = (GlusterAIOCB *)arg;
754 
755     if (!ret || ret == acb->size) {
756         acb->ret = 0; /* Success */
757     } else if (ret < 0) {
758         acb->ret = -errno; /* Read/Write failed */
759     } else {
760         acb->ret = -EIO; /* Partial read/write - fail it */
761     }
762 
763     aio_co_schedule(acb->aio_context, acb->coroutine);
764 }
765 
766 static void qemu_gluster_parse_flags(int bdrv_flags, int *open_flags)
767 {
768     assert(open_flags != NULL);
769 
770     *open_flags |= O_BINARY;
771 
772     if (bdrv_flags & BDRV_O_RDWR) {
773         *open_flags |= O_RDWR;
774     } else {
775         *open_flags |= O_RDONLY;
776     }
777 
778     if ((bdrv_flags & BDRV_O_NOCACHE)) {
779         *open_flags |= O_DIRECT;
780     }
781 }
782 
783 /*
784  * Do SEEK_DATA/HOLE to detect if it is functional. Older broken versions of
785  * gfapi incorrectly return the current offset when SEEK_DATA/HOLE is used.
786  * - Corrected versions return -1 and set errno to EINVAL.
787  * - Versions that support SEEK_DATA/HOLE correctly, will return -1 and set
788  *   errno to ENXIO when SEEK_DATA is called with a position of EOF.
789  */
790 static bool qemu_gluster_test_seek(struct glfs_fd *fd)
791 {
792     off_t ret = 0;
793 
794 #if defined SEEK_HOLE && defined SEEK_DATA
795     off_t eof;
796 
797     eof = glfs_lseek(fd, 0, SEEK_END);
798     if (eof < 0) {
799         /* this should never occur */
800         return false;
801     }
802 
803     /* this should always fail with ENXIO if SEEK_DATA is supported */
804     ret = glfs_lseek(fd, eof, SEEK_DATA);
805 #endif
806 
807     return (ret < 0) && (errno == ENXIO);
808 }
809 
810 static int qemu_gluster_open(BlockDriverState *bs,  QDict *options,
811                              int bdrv_flags, Error **errp)
812 {
813     BDRVGlusterState *s = bs->opaque;
814     int open_flags = 0;
815     int ret = 0;
816     BlockdevOptionsGluster *gconf = NULL;
817     QemuOpts *opts;
818     Error *local_err = NULL;
819     const char *filename, *logfile;
820 
821     opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
822     qemu_opts_absorb_qdict(opts, options, &local_err);
823     if (local_err) {
824         error_propagate(errp, local_err);
825         ret = -EINVAL;
826         goto out;
827     }
828 
829     filename = qemu_opt_get(opts, GLUSTER_OPT_FILENAME);
830 
831     s->debug = qemu_opt_get_number(opts, GLUSTER_OPT_DEBUG,
832                                    GLUSTER_DEBUG_DEFAULT);
833     if (s->debug < 0) {
834         s->debug = 0;
835     } else if (s->debug > GLUSTER_DEBUG_MAX) {
836         s->debug = GLUSTER_DEBUG_MAX;
837     }
838 
839     gconf = g_new0(BlockdevOptionsGluster, 1);
840     gconf->debug = s->debug;
841     gconf->has_debug = true;
842 
843     logfile = qemu_opt_get(opts, GLUSTER_OPT_LOGFILE);
844     s->logfile = g_strdup(logfile ? logfile : GLUSTER_LOGFILE_DEFAULT);
845 
846     gconf->logfile = g_strdup(s->logfile);
847     gconf->has_logfile = true;
848 
849     s->glfs = qemu_gluster_init(gconf, filename, options, errp);
850     if (!s->glfs) {
851         ret = -errno;
852         goto out;
853     }
854 
855 #ifdef CONFIG_GLUSTERFS_XLATOR_OPT
856     /* Without this, if fsync fails for a recoverable reason (for instance,
857      * ENOSPC), gluster will dump its cache, preventing retries.  This means
858      * almost certain data loss.  Not all gluster versions support the
859      * 'resync-failed-syncs-after-fsync' key value, but there is no way to
860      * discover during runtime if it is supported (this api returns success for
861      * unknown key/value pairs) */
862     ret = glfs_set_xlator_option(s->glfs, "*-write-behind",
863                                           "resync-failed-syncs-after-fsync",
864                                           "on");
865     if (ret < 0) {
866         error_setg_errno(errp, errno, "Unable to set xlator key/value pair");
867         ret = -errno;
868         goto out;
869     }
870 #endif
871 
872     qemu_gluster_parse_flags(bdrv_flags, &open_flags);
873 
874     s->fd = glfs_open(s->glfs, gconf->path, open_flags);
875     ret = s->fd ? 0 : -errno;
876 
877     if (ret == -EACCES || ret == -EROFS) {
878         /* Try to degrade to read-only, but if it doesn't work, still use the
879          * normal error message. */
880         if (bdrv_apply_auto_read_only(bs, NULL, NULL) == 0) {
881             open_flags = (open_flags & ~O_RDWR) | O_RDONLY;
882             s->fd = glfs_open(s->glfs, gconf->path, open_flags);
883             ret = s->fd ? 0 : -errno;
884         }
885     }
886 
887     s->supports_seek_data = qemu_gluster_test_seek(s->fd);
888 
889 out:
890     qemu_opts_del(opts);
891     qapi_free_BlockdevOptionsGluster(gconf);
892     if (!ret) {
893         return ret;
894     }
895     g_free(s->logfile);
896     if (s->fd) {
897         glfs_close(s->fd);
898     }
899 
900     glfs_clear_preopened(s->glfs);
901 
902     return ret;
903 }
904 
905 static void qemu_gluster_refresh_limits(BlockDriverState *bs, Error **errp)
906 {
907     bs->bl.max_transfer = GLUSTER_MAX_TRANSFER;
908 }
909 
910 static int qemu_gluster_reopen_prepare(BDRVReopenState *state,
911                                        BlockReopenQueue *queue, Error **errp)
912 {
913     int ret = 0;
914     BDRVGlusterState *s;
915     BDRVGlusterReopenState *reop_s;
916     BlockdevOptionsGluster *gconf;
917     int open_flags = 0;
918 
919     assert(state != NULL);
920     assert(state->bs != NULL);
921 
922     s = state->bs->opaque;
923 
924     state->opaque = g_new0(BDRVGlusterReopenState, 1);
925     reop_s = state->opaque;
926 
927     qemu_gluster_parse_flags(state->flags, &open_flags);
928 
929     gconf = g_new0(BlockdevOptionsGluster, 1);
930     gconf->debug = s->debug;
931     gconf->has_debug = true;
932     gconf->logfile = g_strdup(s->logfile);
933     gconf->has_logfile = true;
934 
935     /*
936      * If 'state->bs->exact_filename' is empty, 'state->options' should contain
937      * the JSON parameters already parsed.
938      */
939     if (state->bs->exact_filename[0] != '\0') {
940         reop_s->glfs = qemu_gluster_init(gconf, state->bs->exact_filename, NULL,
941                                          errp);
942     } else {
943         reop_s->glfs = qemu_gluster_init(gconf, NULL, state->options, errp);
944     }
945     if (reop_s->glfs == NULL) {
946         ret = -errno;
947         goto exit;
948     }
949 
950 #ifdef CONFIG_GLUSTERFS_XLATOR_OPT
951     ret = glfs_set_xlator_option(reop_s->glfs, "*-write-behind",
952                                  "resync-failed-syncs-after-fsync", "on");
953     if (ret < 0) {
954         error_setg_errno(errp, errno, "Unable to set xlator key/value pair");
955         ret = -errno;
956         goto exit;
957     }
958 #endif
959 
960     reop_s->fd = glfs_open(reop_s->glfs, gconf->path, open_flags);
961     if (reop_s->fd == NULL) {
962         /* reops->glfs will be cleaned up in _abort */
963         ret = -errno;
964         goto exit;
965     }
966 
967 exit:
968     /* state->opaque will be freed in either the _abort or _commit */
969     qapi_free_BlockdevOptionsGluster(gconf);
970     return ret;
971 }
972 
973 static void qemu_gluster_reopen_commit(BDRVReopenState *state)
974 {
975     BDRVGlusterReopenState *reop_s = state->opaque;
976     BDRVGlusterState *s = state->bs->opaque;
977 
978 
979     /* close the old */
980     if (s->fd) {
981         glfs_close(s->fd);
982     }
983 
984     glfs_clear_preopened(s->glfs);
985 
986     /* use the newly opened image / connection */
987     s->fd         = reop_s->fd;
988     s->glfs       = reop_s->glfs;
989 
990     g_free(state->opaque);
991     state->opaque = NULL;
992 
993     return;
994 }
995 
996 
997 static void qemu_gluster_reopen_abort(BDRVReopenState *state)
998 {
999     BDRVGlusterReopenState *reop_s = state->opaque;
1000 
1001     if (reop_s == NULL) {
1002         return;
1003     }
1004 
1005     if (reop_s->fd) {
1006         glfs_close(reop_s->fd);
1007     }
1008 
1009     glfs_clear_preopened(reop_s->glfs);
1010 
1011     g_free(state->opaque);
1012     state->opaque = NULL;
1013 
1014     return;
1015 }
1016 
1017 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1018 static coroutine_fn int qemu_gluster_co_pwrite_zeroes(BlockDriverState *bs,
1019                                                       int64_t offset,
1020                                                       int size,
1021                                                       BdrvRequestFlags flags)
1022 {
1023     int ret;
1024     GlusterAIOCB acb;
1025     BDRVGlusterState *s = bs->opaque;
1026 
1027     acb.size = size;
1028     acb.ret = 0;
1029     acb.coroutine = qemu_coroutine_self();
1030     acb.aio_context = bdrv_get_aio_context(bs);
1031 
1032     ret = glfs_zerofill_async(s->fd, offset, size, gluster_finish_aiocb, &acb);
1033     if (ret < 0) {
1034         return -errno;
1035     }
1036 
1037     qemu_coroutine_yield();
1038     return acb.ret;
1039 }
1040 #endif
1041 
1042 static int qemu_gluster_do_truncate(struct glfs_fd *fd, int64_t offset,
1043                                     PreallocMode prealloc, Error **errp)
1044 {
1045     int64_t current_length;
1046 
1047     current_length = glfs_lseek(fd, 0, SEEK_END);
1048     if (current_length < 0) {
1049         error_setg_errno(errp, errno, "Failed to determine current size");
1050         return -errno;
1051     }
1052 
1053     if (current_length > offset && prealloc != PREALLOC_MODE_OFF) {
1054         error_setg(errp, "Cannot use preallocation for shrinking files");
1055         return -ENOTSUP;
1056     }
1057 
1058     if (current_length == offset) {
1059         return 0;
1060     }
1061 
1062     switch (prealloc) {
1063 #ifdef CONFIG_GLUSTERFS_FALLOCATE
1064     case PREALLOC_MODE_FALLOC:
1065         if (glfs_fallocate(fd, 0, current_length, offset - current_length)) {
1066             error_setg_errno(errp, errno, "Could not preallocate data");
1067             return -errno;
1068         }
1069         break;
1070 #endif /* CONFIG_GLUSTERFS_FALLOCATE */
1071 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1072     case PREALLOC_MODE_FULL:
1073         if (glfs_ftruncate(fd, offset)) {
1074             error_setg_errno(errp, errno, "Could not resize file");
1075             return -errno;
1076         }
1077         if (glfs_zerofill(fd, current_length, offset - current_length)) {
1078             error_setg_errno(errp, errno, "Could not zerofill the new area");
1079             return -errno;
1080         }
1081         break;
1082 #endif /* CONFIG_GLUSTERFS_ZEROFILL */
1083     case PREALLOC_MODE_OFF:
1084         if (glfs_ftruncate(fd, offset)) {
1085             error_setg_errno(errp, errno, "Could not resize file");
1086             return -errno;
1087         }
1088         break;
1089     default:
1090         error_setg(errp, "Unsupported preallocation mode: %s",
1091                    PreallocMode_str(prealloc));
1092         return -EINVAL;
1093     }
1094 
1095     return 0;
1096 }
1097 
1098 static int qemu_gluster_co_create(BlockdevCreateOptions *options,
1099                                   Error **errp)
1100 {
1101     BlockdevCreateOptionsGluster *opts = &options->u.gluster;
1102     struct glfs *glfs;
1103     struct glfs_fd *fd = NULL;
1104     int ret = 0;
1105 
1106     assert(options->driver == BLOCKDEV_DRIVER_GLUSTER);
1107 
1108     glfs = qemu_gluster_glfs_init(opts->location, errp);
1109     if (!glfs) {
1110         ret = -errno;
1111         goto out;
1112     }
1113 
1114     fd = glfs_creat(glfs, opts->location->path,
1115                     O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, S_IRUSR | S_IWUSR);
1116     if (!fd) {
1117         ret = -errno;
1118         goto out;
1119     }
1120 
1121     ret = qemu_gluster_do_truncate(fd, opts->size, opts->preallocation, errp);
1122 
1123 out:
1124     if (fd) {
1125         if (glfs_close(fd) != 0 && ret == 0) {
1126             ret = -errno;
1127         }
1128     }
1129     glfs_clear_preopened(glfs);
1130     return ret;
1131 }
1132 
1133 static int coroutine_fn qemu_gluster_co_create_opts(const char *filename,
1134                                                     QemuOpts *opts,
1135                                                     Error **errp)
1136 {
1137     BlockdevCreateOptions *options;
1138     BlockdevCreateOptionsGluster *gopts;
1139     BlockdevOptionsGluster *gconf;
1140     char *tmp = NULL;
1141     Error *local_err = NULL;
1142     int ret;
1143 
1144     options = g_new0(BlockdevCreateOptions, 1);
1145     options->driver = BLOCKDEV_DRIVER_GLUSTER;
1146     gopts = &options->u.gluster;
1147 
1148     gconf = g_new0(BlockdevOptionsGluster, 1);
1149     gopts->location = gconf;
1150 
1151     gopts->size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
1152                            BDRV_SECTOR_SIZE);
1153 
1154     tmp = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
1155     gopts->preallocation = qapi_enum_parse(&PreallocMode_lookup, tmp,
1156                                            PREALLOC_MODE_OFF, &local_err);
1157     g_free(tmp);
1158     if (local_err) {
1159         error_propagate(errp, local_err);
1160         ret = -EINVAL;
1161         goto fail;
1162     }
1163 
1164     gconf->debug = qemu_opt_get_number_del(opts, GLUSTER_OPT_DEBUG,
1165                                            GLUSTER_DEBUG_DEFAULT);
1166     if (gconf->debug < 0) {
1167         gconf->debug = 0;
1168     } else if (gconf->debug > GLUSTER_DEBUG_MAX) {
1169         gconf->debug = GLUSTER_DEBUG_MAX;
1170     }
1171     gconf->has_debug = true;
1172 
1173     gconf->logfile = qemu_opt_get_del(opts, GLUSTER_OPT_LOGFILE);
1174     if (!gconf->logfile) {
1175         gconf->logfile = g_strdup(GLUSTER_LOGFILE_DEFAULT);
1176     }
1177     gconf->has_logfile = true;
1178 
1179     ret = qemu_gluster_parse(gconf, filename, NULL, errp);
1180     if (ret < 0) {
1181         goto fail;
1182     }
1183 
1184     ret = qemu_gluster_co_create(options, errp);
1185     if (ret < 0) {
1186         goto fail;
1187     }
1188 
1189     ret = 0;
1190 fail:
1191     qapi_free_BlockdevCreateOptions(options);
1192     return ret;
1193 }
1194 
1195 static coroutine_fn int qemu_gluster_co_rw(BlockDriverState *bs,
1196                                            int64_t sector_num, int nb_sectors,
1197                                            QEMUIOVector *qiov, int write)
1198 {
1199     int ret;
1200     GlusterAIOCB acb;
1201     BDRVGlusterState *s = bs->opaque;
1202     size_t size = nb_sectors * BDRV_SECTOR_SIZE;
1203     off_t offset = sector_num * BDRV_SECTOR_SIZE;
1204 
1205     acb.size = size;
1206     acb.ret = 0;
1207     acb.coroutine = qemu_coroutine_self();
1208     acb.aio_context = bdrv_get_aio_context(bs);
1209 
1210     if (write) {
1211         ret = glfs_pwritev_async(s->fd, qiov->iov, qiov->niov, offset, 0,
1212                                  gluster_finish_aiocb, &acb);
1213     } else {
1214         ret = glfs_preadv_async(s->fd, qiov->iov, qiov->niov, offset, 0,
1215                                 gluster_finish_aiocb, &acb);
1216     }
1217 
1218     if (ret < 0) {
1219         return -errno;
1220     }
1221 
1222     qemu_coroutine_yield();
1223     return acb.ret;
1224 }
1225 
1226 static coroutine_fn int qemu_gluster_co_truncate(BlockDriverState *bs,
1227                                                  int64_t offset,
1228                                                  PreallocMode prealloc,
1229                                                  Error **errp)
1230 {
1231     BDRVGlusterState *s = bs->opaque;
1232     return qemu_gluster_do_truncate(s->fd, offset, prealloc, errp);
1233 }
1234 
1235 static coroutine_fn int qemu_gluster_co_readv(BlockDriverState *bs,
1236                                               int64_t sector_num,
1237                                               int nb_sectors,
1238                                               QEMUIOVector *qiov)
1239 {
1240     return qemu_gluster_co_rw(bs, sector_num, nb_sectors, qiov, 0);
1241 }
1242 
1243 static coroutine_fn int qemu_gluster_co_writev(BlockDriverState *bs,
1244                                                int64_t sector_num,
1245                                                int nb_sectors,
1246                                                QEMUIOVector *qiov,
1247                                                int flags)
1248 {
1249     assert(!flags);
1250     return qemu_gluster_co_rw(bs, sector_num, nb_sectors, qiov, 1);
1251 }
1252 
1253 static void qemu_gluster_close(BlockDriverState *bs)
1254 {
1255     BDRVGlusterState *s = bs->opaque;
1256 
1257     g_free(s->logfile);
1258     if (s->fd) {
1259         glfs_close(s->fd);
1260         s->fd = NULL;
1261     }
1262     glfs_clear_preopened(s->glfs);
1263 }
1264 
1265 static coroutine_fn int qemu_gluster_co_flush_to_disk(BlockDriverState *bs)
1266 {
1267     int ret;
1268     GlusterAIOCB acb;
1269     BDRVGlusterState *s = bs->opaque;
1270 
1271     acb.size = 0;
1272     acb.ret = 0;
1273     acb.coroutine = qemu_coroutine_self();
1274     acb.aio_context = bdrv_get_aio_context(bs);
1275 
1276     ret = glfs_fsync_async(s->fd, gluster_finish_aiocb, &acb);
1277     if (ret < 0) {
1278         ret = -errno;
1279         goto error;
1280     }
1281 
1282     qemu_coroutine_yield();
1283     if (acb.ret < 0) {
1284         ret = acb.ret;
1285         goto error;
1286     }
1287 
1288     return acb.ret;
1289 
1290 error:
1291     /* Some versions of Gluster (3.5.6 -> 3.5.8?) will not retain its cache
1292      * after a fsync failure, so we have no way of allowing the guest to safely
1293      * continue.  Gluster versions prior to 3.5.6 don't retain the cache
1294      * either, but will invalidate the fd on error, so this is again our only
1295      * option.
1296      *
1297      * The 'resync-failed-syncs-after-fsync' xlator option for the
1298      * write-behind cache will cause later gluster versions to retain its
1299      * cache after error, so long as the fd remains open.  However, we
1300      * currently have no way of knowing if this option is supported.
1301      *
1302      * TODO: Once gluster provides a way for us to determine if the option
1303      * is supported, bypass the closure and setting drv to NULL.  */
1304     qemu_gluster_close(bs);
1305     bs->drv = NULL;
1306     return ret;
1307 }
1308 
1309 #ifdef CONFIG_GLUSTERFS_DISCARD
1310 static coroutine_fn int qemu_gluster_co_pdiscard(BlockDriverState *bs,
1311                                                  int64_t offset, int size)
1312 {
1313     int ret;
1314     GlusterAIOCB acb;
1315     BDRVGlusterState *s = bs->opaque;
1316 
1317     acb.size = 0;
1318     acb.ret = 0;
1319     acb.coroutine = qemu_coroutine_self();
1320     acb.aio_context = bdrv_get_aio_context(bs);
1321 
1322     ret = glfs_discard_async(s->fd, offset, size, gluster_finish_aiocb, &acb);
1323     if (ret < 0) {
1324         return -errno;
1325     }
1326 
1327     qemu_coroutine_yield();
1328     return acb.ret;
1329 }
1330 #endif
1331 
1332 static int64_t qemu_gluster_getlength(BlockDriverState *bs)
1333 {
1334     BDRVGlusterState *s = bs->opaque;
1335     int64_t ret;
1336 
1337     ret = glfs_lseek(s->fd, 0, SEEK_END);
1338     if (ret < 0) {
1339         return -errno;
1340     } else {
1341         return ret;
1342     }
1343 }
1344 
1345 static int64_t qemu_gluster_allocated_file_size(BlockDriverState *bs)
1346 {
1347     BDRVGlusterState *s = bs->opaque;
1348     struct stat st;
1349     int ret;
1350 
1351     ret = glfs_fstat(s->fd, &st);
1352     if (ret < 0) {
1353         return -errno;
1354     } else {
1355         return st.st_blocks * 512;
1356     }
1357 }
1358 
1359 static int qemu_gluster_has_zero_init(BlockDriverState *bs)
1360 {
1361     /* GlusterFS volume could be backed by a block device */
1362     return 0;
1363 }
1364 
1365 /*
1366  * Find allocation range in @bs around offset @start.
1367  * May change underlying file descriptor's file offset.
1368  * If @start is not in a hole, store @start in @data, and the
1369  * beginning of the next hole in @hole, and return 0.
1370  * If @start is in a non-trailing hole, store @start in @hole and the
1371  * beginning of the next non-hole in @data, and return 0.
1372  * If @start is in a trailing hole or beyond EOF, return -ENXIO.
1373  * If we can't find out, return a negative errno other than -ENXIO.
1374  *
1375  * (Shamefully copied from file-posix.c, only minuscule adaptions.)
1376  */
1377 static int find_allocation(BlockDriverState *bs, off_t start,
1378                            off_t *data, off_t *hole)
1379 {
1380     BDRVGlusterState *s = bs->opaque;
1381 
1382     if (!s->supports_seek_data) {
1383         goto exit;
1384     }
1385 
1386 #if defined SEEK_HOLE && defined SEEK_DATA
1387     off_t offs;
1388 
1389     /*
1390      * SEEK_DATA cases:
1391      * D1. offs == start: start is in data
1392      * D2. offs > start: start is in a hole, next data at offs
1393      * D3. offs < 0, errno = ENXIO: either start is in a trailing hole
1394      *                              or start is beyond EOF
1395      *     If the latter happens, the file has been truncated behind
1396      *     our back since we opened it.  All bets are off then.
1397      *     Treating like a trailing hole is simplest.
1398      * D4. offs < 0, errno != ENXIO: we learned nothing
1399      */
1400     offs = glfs_lseek(s->fd, start, SEEK_DATA);
1401     if (offs < 0) {
1402         return -errno;          /* D3 or D4 */
1403     }
1404 
1405     if (offs < start) {
1406         /* This is not a valid return by lseek().  We are safe to just return
1407          * -EIO in this case, and we'll treat it like D4. Unfortunately some
1408          *  versions of gluster server will return offs < start, so an assert
1409          *  here will unnecessarily abort QEMU. */
1410         return -EIO;
1411     }
1412 
1413     if (offs > start) {
1414         /* D2: in hole, next data at offs */
1415         *hole = start;
1416         *data = offs;
1417         return 0;
1418     }
1419 
1420     /* D1: in data, end not yet known */
1421 
1422     /*
1423      * SEEK_HOLE cases:
1424      * H1. offs == start: start is in a hole
1425      *     If this happens here, a hole has been dug behind our back
1426      *     since the previous lseek().
1427      * H2. offs > start: either start is in data, next hole at offs,
1428      *                   or start is in trailing hole, EOF at offs
1429      *     Linux treats trailing holes like any other hole: offs ==
1430      *     start.  Solaris seeks to EOF instead: offs > start (blech).
1431      *     If that happens here, a hole has been dug behind our back
1432      *     since the previous lseek().
1433      * H3. offs < 0, errno = ENXIO: start is beyond EOF
1434      *     If this happens, the file has been truncated behind our
1435      *     back since we opened it.  Treat it like a trailing hole.
1436      * H4. offs < 0, errno != ENXIO: we learned nothing
1437      *     Pretend we know nothing at all, i.e. "forget" about D1.
1438      */
1439     offs = glfs_lseek(s->fd, start, SEEK_HOLE);
1440     if (offs < 0) {
1441         return -errno;          /* D1 and (H3 or H4) */
1442     }
1443 
1444     if (offs < start) {
1445         /* This is not a valid return by lseek().  We are safe to just return
1446          * -EIO in this case, and we'll treat it like H4. Unfortunately some
1447          *  versions of gluster server will return offs < start, so an assert
1448          *  here will unnecessarily abort QEMU. */
1449         return -EIO;
1450     }
1451 
1452     if (offs > start) {
1453         /*
1454          * D1 and H2: either in data, next hole at offs, or it was in
1455          * data but is now in a trailing hole.  In the latter case,
1456          * all bets are off.  Treating it as if it there was data all
1457          * the way to EOF is safe, so simply do that.
1458          */
1459         *data = start;
1460         *hole = offs;
1461         return 0;
1462     }
1463 
1464     /* D1 and H1 */
1465     return -EBUSY;
1466 #endif
1467 
1468 exit:
1469     return -ENOTSUP;
1470 }
1471 
1472 /*
1473  * Returns the allocation status of the specified offset.
1474  *
1475  * The block layer guarantees 'offset' and 'bytes' are within bounds.
1476  *
1477  * 'pnum' is set to the number of bytes (including and immediately following
1478  * the specified offset) that are known to be in the same
1479  * allocated/unallocated state.
1480  *
1481  * 'bytes' is the max value 'pnum' should be set to.
1482  *
1483  * (Based on raw_co_block_status() from file-posix.c.)
1484  */
1485 static int coroutine_fn qemu_gluster_co_block_status(BlockDriverState *bs,
1486                                                      bool want_zero,
1487                                                      int64_t offset,
1488                                                      int64_t bytes,
1489                                                      int64_t *pnum,
1490                                                      int64_t *map,
1491                                                      BlockDriverState **file)
1492 {
1493     BDRVGlusterState *s = bs->opaque;
1494     off_t data = 0, hole = 0;
1495     int ret = -EINVAL;
1496 
1497     if (!s->fd) {
1498         return ret;
1499     }
1500 
1501     if (!want_zero) {
1502         *pnum = bytes;
1503         *map = offset;
1504         *file = bs;
1505         return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID;
1506     }
1507 
1508     ret = find_allocation(bs, offset, &data, &hole);
1509     if (ret == -ENXIO) {
1510         /* Trailing hole */
1511         *pnum = bytes;
1512         ret = BDRV_BLOCK_ZERO;
1513     } else if (ret < 0) {
1514         /* No info available, so pretend there are no holes */
1515         *pnum = bytes;
1516         ret = BDRV_BLOCK_DATA;
1517     } else if (data == offset) {
1518         /* On a data extent, compute bytes to the end of the extent,
1519          * possibly including a partial sector at EOF. */
1520         *pnum = MIN(bytes, hole - offset);
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 = MIN(bytes, 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     .bdrv_has_zero_init           = qemu_gluster_has_zero_init,
1570     .bdrv_has_zero_init_truncate  = qemu_gluster_has_zero_init,
1571 #ifdef CONFIG_GLUSTERFS_DISCARD
1572     .bdrv_co_pdiscard             = qemu_gluster_co_pdiscard,
1573 #endif
1574 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1575     .bdrv_co_pwrite_zeroes        = qemu_gluster_co_pwrite_zeroes,
1576 #endif
1577     .bdrv_co_block_status         = qemu_gluster_co_block_status,
1578     .bdrv_refresh_limits          = qemu_gluster_refresh_limits,
1579     .create_opts                  = &qemu_gluster_create_opts,
1580     .strong_runtime_opts          = gluster_strong_open_opts,
1581 };
1582 
1583 static BlockDriver bdrv_gluster_tcp = {
1584     .format_name                  = "gluster",
1585     .protocol_name                = "gluster+tcp",
1586     .instance_size                = sizeof(BDRVGlusterState),
1587     .bdrv_needs_filename          = false,
1588     .bdrv_file_open               = qemu_gluster_open,
1589     .bdrv_reopen_prepare          = qemu_gluster_reopen_prepare,
1590     .bdrv_reopen_commit           = qemu_gluster_reopen_commit,
1591     .bdrv_reopen_abort            = qemu_gluster_reopen_abort,
1592     .bdrv_close                   = qemu_gluster_close,
1593     .bdrv_co_create               = qemu_gluster_co_create,
1594     .bdrv_co_create_opts          = qemu_gluster_co_create_opts,
1595     .bdrv_getlength               = qemu_gluster_getlength,
1596     .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size,
1597     .bdrv_co_truncate             = qemu_gluster_co_truncate,
1598     .bdrv_co_readv                = qemu_gluster_co_readv,
1599     .bdrv_co_writev               = qemu_gluster_co_writev,
1600     .bdrv_co_flush_to_disk        = qemu_gluster_co_flush_to_disk,
1601     .bdrv_has_zero_init           = qemu_gluster_has_zero_init,
1602     .bdrv_has_zero_init_truncate  = qemu_gluster_has_zero_init,
1603 #ifdef CONFIG_GLUSTERFS_DISCARD
1604     .bdrv_co_pdiscard             = qemu_gluster_co_pdiscard,
1605 #endif
1606 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1607     .bdrv_co_pwrite_zeroes        = qemu_gluster_co_pwrite_zeroes,
1608 #endif
1609     .bdrv_co_block_status         = qemu_gluster_co_block_status,
1610     .bdrv_refresh_limits          = qemu_gluster_refresh_limits,
1611     .create_opts                  = &qemu_gluster_create_opts,
1612     .strong_runtime_opts          = gluster_strong_open_opts,
1613 };
1614 
1615 static BlockDriver bdrv_gluster_unix = {
1616     .format_name                  = "gluster",
1617     .protocol_name                = "gluster+unix",
1618     .instance_size                = sizeof(BDRVGlusterState),
1619     .bdrv_needs_filename          = true,
1620     .bdrv_file_open               = qemu_gluster_open,
1621     .bdrv_reopen_prepare          = qemu_gluster_reopen_prepare,
1622     .bdrv_reopen_commit           = qemu_gluster_reopen_commit,
1623     .bdrv_reopen_abort            = qemu_gluster_reopen_abort,
1624     .bdrv_close                   = qemu_gluster_close,
1625     .bdrv_co_create               = qemu_gluster_co_create,
1626     .bdrv_co_create_opts          = qemu_gluster_co_create_opts,
1627     .bdrv_getlength               = qemu_gluster_getlength,
1628     .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size,
1629     .bdrv_co_truncate             = qemu_gluster_co_truncate,
1630     .bdrv_co_readv                = qemu_gluster_co_readv,
1631     .bdrv_co_writev               = qemu_gluster_co_writev,
1632     .bdrv_co_flush_to_disk        = qemu_gluster_co_flush_to_disk,
1633     .bdrv_has_zero_init           = qemu_gluster_has_zero_init,
1634     .bdrv_has_zero_init_truncate  = qemu_gluster_has_zero_init,
1635 #ifdef CONFIG_GLUSTERFS_DISCARD
1636     .bdrv_co_pdiscard             = qemu_gluster_co_pdiscard,
1637 #endif
1638 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1639     .bdrv_co_pwrite_zeroes        = qemu_gluster_co_pwrite_zeroes,
1640 #endif
1641     .bdrv_co_block_status         = qemu_gluster_co_block_status,
1642     .bdrv_refresh_limits          = qemu_gluster_refresh_limits,
1643     .create_opts                  = &qemu_gluster_create_opts,
1644     .strong_runtime_opts          = gluster_strong_open_opts,
1645 };
1646 
1647 /* rdma is deprecated (actually never supported for volfile fetch).
1648  * Let's maintain it for the protocol compatibility, to make sure things
1649  * won't break immediately. For now, gluster+rdma will fall back to gluster+tcp
1650  * protocol with a warning.
1651  * TODO: remove gluster+rdma interface support
1652  */
1653 static BlockDriver bdrv_gluster_rdma = {
1654     .format_name                  = "gluster",
1655     .protocol_name                = "gluster+rdma",
1656     .instance_size                = sizeof(BDRVGlusterState),
1657     .bdrv_needs_filename          = true,
1658     .bdrv_file_open               = qemu_gluster_open,
1659     .bdrv_reopen_prepare          = qemu_gluster_reopen_prepare,
1660     .bdrv_reopen_commit           = qemu_gluster_reopen_commit,
1661     .bdrv_reopen_abort            = qemu_gluster_reopen_abort,
1662     .bdrv_close                   = qemu_gluster_close,
1663     .bdrv_co_create               = qemu_gluster_co_create,
1664     .bdrv_co_create_opts          = qemu_gluster_co_create_opts,
1665     .bdrv_getlength               = qemu_gluster_getlength,
1666     .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size,
1667     .bdrv_co_truncate             = qemu_gluster_co_truncate,
1668     .bdrv_co_readv                = qemu_gluster_co_readv,
1669     .bdrv_co_writev               = qemu_gluster_co_writev,
1670     .bdrv_co_flush_to_disk        = qemu_gluster_co_flush_to_disk,
1671     .bdrv_has_zero_init           = qemu_gluster_has_zero_init,
1672     .bdrv_has_zero_init_truncate  = qemu_gluster_has_zero_init,
1673 #ifdef CONFIG_GLUSTERFS_DISCARD
1674     .bdrv_co_pdiscard             = qemu_gluster_co_pdiscard,
1675 #endif
1676 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1677     .bdrv_co_pwrite_zeroes        = qemu_gluster_co_pwrite_zeroes,
1678 #endif
1679     .bdrv_co_block_status         = qemu_gluster_co_block_status,
1680     .bdrv_refresh_limits          = qemu_gluster_refresh_limits,
1681     .create_opts                  = &qemu_gluster_create_opts,
1682     .strong_runtime_opts          = gluster_strong_open_opts,
1683 };
1684 
1685 static void bdrv_gluster_init(void)
1686 {
1687     bdrv_register(&bdrv_gluster_rdma);
1688     bdrv_register(&bdrv_gluster_unix);
1689     bdrv_register(&bdrv_gluster_tcp);
1690     bdrv_register(&bdrv_gluster);
1691 }
1692 
1693 block_init(bdrv_gluster_init);
1694