xref: /openbmc/qemu/block/gluster.c (revision f0984d40)
1 /*
2  * GlusterFS backend for QEMU
3  *
4  * Copyright (C) 2012 Bharata B Rao <bharata@linux.vnet.ibm.com>
5  *
6  * This work is licensed under the terms of the GNU GPL, version 2 or later.
7  * See the COPYING file in the top-level directory.
8  *
9  */
10 
11 #include "qemu/osdep.h"
12 #include "qemu/units.h"
13 #include <glusterfs/api/glfs.h>
14 #include "block/block-io.h"
15 #include "block/block_int.h"
16 #include "block/qdict.h"
17 #include "qapi/error.h"
18 #include "qapi/qmp/qdict.h"
19 #include "qapi/qmp/qerror.h"
20 #include "qemu/uri.h"
21 #include "qemu/error-report.h"
22 #include "qemu/module.h"
23 #include "qemu/option.h"
24 #include "qemu/cutils.h"
25 
26 #ifdef CONFIG_GLUSTERFS_FTRUNCATE_HAS_STAT
27 # define glfs_ftruncate(fd, offset) glfs_ftruncate(fd, offset, NULL, NULL)
28 #endif
29 
30 #define GLUSTER_OPT_FILENAME        "filename"
31 #define GLUSTER_OPT_VOLUME          "volume"
32 #define GLUSTER_OPT_PATH            "path"
33 #define GLUSTER_OPT_TYPE            "type"
34 #define GLUSTER_OPT_SERVER_PATTERN  "server."
35 #define GLUSTER_OPT_HOST            "host"
36 #define GLUSTER_OPT_PORT            "port"
37 #define GLUSTER_OPT_TO              "to"
38 #define GLUSTER_OPT_IPV4            "ipv4"
39 #define GLUSTER_OPT_IPV6            "ipv6"
40 #define GLUSTER_OPT_SOCKET          "socket"
41 #define GLUSTER_OPT_DEBUG           "debug"
42 #define GLUSTER_DEFAULT_PORT        24007
43 #define GLUSTER_DEBUG_DEFAULT       4
44 #define GLUSTER_DEBUG_MAX           9
45 #define GLUSTER_OPT_LOGFILE         "logfile"
46 #define GLUSTER_LOGFILE_DEFAULT     "-" /* handled in libgfapi as /dev/stderr */
47 /*
48  * Several versions of GlusterFS (3.12? -> 6.0.1) fail when the transfer size
49  * is greater or equal to 1024 MiB, so we are limiting the transfer size to 512
50  * MiB to avoid this rare issue.
51  */
52 #define GLUSTER_MAX_TRANSFER        (512 * MiB)
53 
54 #define GERR_INDEX_HINT "hint: check in 'server' array index '%d'\n"
55 
56 typedef struct GlusterAIOCB {
57     int64_t size;
58     int ret;
59     Coroutine *coroutine;
60     AioContext *aio_context;
61 } GlusterAIOCB;
62 
63 typedef struct BDRVGlusterState {
64     struct glfs *glfs;
65     struct glfs_fd *fd;
66     char *logfile;
67     bool supports_seek_data;
68     int debug;
69 } BDRVGlusterState;
70 
71 typedef struct BDRVGlusterReopenState {
72     struct glfs *glfs;
73     struct glfs_fd *fd;
74 } BDRVGlusterReopenState;
75 
76 
77 typedef struct GlfsPreopened {
78     char *volume;
79     glfs_t *fs;
80     int ref;
81 } GlfsPreopened;
82 
83 typedef struct ListElement {
84     QLIST_ENTRY(ListElement) list;
85     GlfsPreopened saved;
86 } ListElement;
87 
88 static QLIST_HEAD(, ListElement) glfs_list;
89 
90 static QemuOptsList qemu_gluster_create_opts = {
91     .name = "qemu-gluster-create-opts",
92     .head = QTAILQ_HEAD_INITIALIZER(qemu_gluster_create_opts.head),
93     .desc = {
94         {
95             .name = BLOCK_OPT_SIZE,
96             .type = QEMU_OPT_SIZE,
97             .help = "Virtual disk size"
98         },
99         {
100             .name = BLOCK_OPT_PREALLOC,
101             .type = QEMU_OPT_STRING,
102             .help = "Preallocation mode (allowed values: off"
103 #ifdef CONFIG_GLUSTERFS_FALLOCATE
104                     ", falloc"
105 #endif
106 #ifdef CONFIG_GLUSTERFS_ZEROFILL
107                     ", full"
108 #endif
109                     ")"
110         },
111         {
112             .name = GLUSTER_OPT_DEBUG,
113             .type = QEMU_OPT_NUMBER,
114             .help = "Gluster log level, valid range is 0-9",
115         },
116         {
117             .name = GLUSTER_OPT_LOGFILE,
118             .type = QEMU_OPT_STRING,
119             .help = "Logfile path of libgfapi",
120         },
121         { /* end of list */ }
122     }
123 };
124 
125 static QemuOptsList runtime_opts = {
126     .name = "gluster",
127     .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
128     .desc = {
129         {
130             .name = GLUSTER_OPT_FILENAME,
131             .type = QEMU_OPT_STRING,
132             .help = "URL to the gluster image",
133         },
134         {
135             .name = GLUSTER_OPT_DEBUG,
136             .type = QEMU_OPT_NUMBER,
137             .help = "Gluster log level, valid range is 0-9",
138         },
139         {
140             .name = GLUSTER_OPT_LOGFILE,
141             .type = QEMU_OPT_STRING,
142             .help = "Logfile path of libgfapi",
143         },
144         { /* end of list */ }
145     },
146 };
147 
148 static QemuOptsList runtime_json_opts = {
149     .name = "gluster_json",
150     .head = QTAILQ_HEAD_INITIALIZER(runtime_json_opts.head),
151     .desc = {
152         {
153             .name = GLUSTER_OPT_VOLUME,
154             .type = QEMU_OPT_STRING,
155             .help = "name of gluster volume where VM image resides",
156         },
157         {
158             .name = GLUSTER_OPT_PATH,
159             .type = QEMU_OPT_STRING,
160             .help = "absolute path to image file in gluster volume",
161         },
162         {
163             .name = GLUSTER_OPT_DEBUG,
164             .type = QEMU_OPT_NUMBER,
165             .help = "Gluster log level, valid range is 0-9",
166         },
167         { /* end of list */ }
168     },
169 };
170 
171 static QemuOptsList runtime_type_opts = {
172     .name = "gluster_type",
173     .head = QTAILQ_HEAD_INITIALIZER(runtime_type_opts.head),
174     .desc = {
175         {
176             .name = GLUSTER_OPT_TYPE,
177             .type = QEMU_OPT_STRING,
178             .help = "inet|unix",
179         },
180         { /* end of list */ }
181     },
182 };
183 
184 static QemuOptsList runtime_unix_opts = {
185     .name = "gluster_unix",
186     .head = QTAILQ_HEAD_INITIALIZER(runtime_unix_opts.head),
187     .desc = {
188         {
189             .name = GLUSTER_OPT_SOCKET,
190             .type = QEMU_OPT_STRING,
191             .help = "socket file path (legacy)",
192         },
193         {
194             .name = GLUSTER_OPT_PATH,
195             .type = QEMU_OPT_STRING,
196             .help = "socket file path (QAPI)",
197         },
198         { /* end of list */ }
199     },
200 };
201 
202 static QemuOptsList runtime_inet_opts = {
203     .name = "gluster_inet",
204     .head = QTAILQ_HEAD_INITIALIZER(runtime_inet_opts.head),
205     .desc = {
206         {
207             .name = GLUSTER_OPT_TYPE,
208             .type = QEMU_OPT_STRING,
209             .help = "inet|unix",
210         },
211         {
212             .name = GLUSTER_OPT_HOST,
213             .type = QEMU_OPT_STRING,
214             .help = "host address (hostname/ipv4/ipv6 addresses)",
215         },
216         {
217             .name = GLUSTER_OPT_PORT,
218             .type = QEMU_OPT_STRING,
219             .help = "port number on which glusterd is listening (default 24007)",
220         },
221         {
222             .name = "to",
223             .type = QEMU_OPT_NUMBER,
224             .help = "max port number, not supported by gluster",
225         },
226         {
227             .name = "ipv4",
228             .type = QEMU_OPT_BOOL,
229             .help = "ipv4 bool value, not supported by gluster",
230         },
231         {
232             .name = "ipv6",
233             .type = QEMU_OPT_BOOL,
234             .help = "ipv6 bool value, not supported by gluster",
235         },
236         { /* end of list */ }
237     },
238 };
239 
240 static void glfs_set_preopened(const char *volume, glfs_t *fs)
241 {
242     ListElement *entry = NULL;
243 
244     entry = g_new(ListElement, 1);
245 
246     entry->saved.volume = g_strdup(volume);
247 
248     entry->saved.fs = fs;
249     entry->saved.ref = 1;
250 
251     QLIST_INSERT_HEAD(&glfs_list, entry, list);
252 }
253 
254 static glfs_t *glfs_find_preopened(const char *volume)
255 {
256     ListElement *entry = NULL;
257 
258      QLIST_FOREACH(entry, &glfs_list, list) {
259         if (strcmp(entry->saved.volume, volume) == 0) {
260             entry->saved.ref++;
261             return entry->saved.fs;
262         }
263      }
264 
265     return NULL;
266 }
267 
268 static void glfs_clear_preopened(glfs_t *fs)
269 {
270     ListElement *entry = NULL;
271     ListElement *next;
272 
273     if (fs == NULL) {
274         return;
275     }
276 
277     QLIST_FOREACH_SAFE(entry, &glfs_list, list, next) {
278         if (entry->saved.fs == fs) {
279             if (--entry->saved.ref) {
280                 return;
281             }
282 
283             QLIST_REMOVE(entry, list);
284 
285             glfs_fini(entry->saved.fs);
286             g_free(entry->saved.volume);
287             g_free(entry);
288         }
289     }
290 }
291 
292 static int parse_volume_options(BlockdevOptionsGluster *gconf, char *path)
293 {
294     char *p, *q;
295 
296     if (!path) {
297         return -EINVAL;
298     }
299 
300     /* volume */
301     p = q = path + strspn(path, "/");
302     p += strcspn(p, "/");
303     if (*p == '\0') {
304         return -EINVAL;
305     }
306     gconf->volume = g_strndup(q, p - q);
307 
308     /* path */
309     p += strspn(p, "/");
310     if (*p == '\0') {
311         return -EINVAL;
312     }
313     gconf->path = g_strdup(p);
314     return 0;
315 }
316 
317 /*
318  * file=gluster[+transport]://[host[:port]]/volume/path[?socket=...]
319  *
320  * 'gluster' is the protocol.
321  *
322  * 'transport' specifies the transport type used to connect to gluster
323  * management daemon (glusterd). Valid transport types are
324  * tcp or unix. If a transport type isn't specified, then tcp type is assumed.
325  *
326  * 'host' specifies the host where the volume file specification for
327  * the given volume resides. This can be either hostname or ipv4 address.
328  * If transport type is 'unix', then 'host' field should not be specified.
329  * The 'socket' field needs to be populated with the path to unix domain
330  * socket.
331  *
332  * 'port' is the port number on which glusterd is listening. This is optional
333  * and if not specified, QEMU will send 0 which will make gluster to use the
334  * default port. If the transport type is unix, then 'port' should not be
335  * specified.
336  *
337  * 'volume' is the name of the gluster volume which contains the VM image.
338  *
339  * 'path' is the path to the actual VM image that resides on gluster volume.
340  *
341  * Examples:
342  *
343  * file=gluster://1.2.3.4/testvol/a.img
344  * file=gluster+tcp://1.2.3.4/testvol/a.img
345  * file=gluster+tcp://1.2.3.4:24007/testvol/dir/a.img
346  * file=gluster+tcp://host.domain.com:24007/testvol/dir/a.img
347  * file=gluster+unix:///testvol/dir/a.img?socket=/tmp/glusterd.socket
348  */
349 static int qemu_gluster_parse_uri(BlockdevOptionsGluster *gconf,
350                                   const char *filename)
351 {
352     SocketAddress *gsconf;
353     URI *uri;
354     QueryParams *qp = NULL;
355     bool is_unix = false;
356     int ret = 0;
357 
358     uri = uri_parse(filename);
359     if (!uri) {
360         return -EINVAL;
361     }
362 
363     gsconf = g_new0(SocketAddress, 1);
364     QAPI_LIST_PREPEND(gconf->server, gsconf);
365 
366     /* transport */
367     if (!uri->scheme || !strcmp(uri->scheme, "gluster")) {
368         gsconf->type = SOCKET_ADDRESS_TYPE_INET;
369     } else if (!strcmp(uri->scheme, "gluster+tcp")) {
370         gsconf->type = SOCKET_ADDRESS_TYPE_INET;
371     } else if (!strcmp(uri->scheme, "gluster+unix")) {
372         gsconf->type = SOCKET_ADDRESS_TYPE_UNIX;
373         is_unix = true;
374     } else if (!strcmp(uri->scheme, "gluster+rdma")) {
375         gsconf->type = SOCKET_ADDRESS_TYPE_INET;
376         warn_report("rdma feature is not supported, falling back to tcp");
377     } else {
378         ret = -EINVAL;
379         goto out;
380     }
381 
382     ret = parse_volume_options(gconf, uri->path);
383     if (ret < 0) {
384         goto out;
385     }
386 
387     qp = query_params_parse(uri->query);
388     if (qp->n > 1 || (is_unix && !qp->n) || (!is_unix && qp->n)) {
389         ret = -EINVAL;
390         goto out;
391     }
392 
393     if (is_unix) {
394         if (uri->server || uri->port) {
395             ret = -EINVAL;
396             goto out;
397         }
398         if (strcmp(qp->p[0].name, "socket")) {
399             ret = -EINVAL;
400             goto out;
401         }
402         gsconf->u.q_unix.path = g_strdup(qp->p[0].value);
403     } else {
404         gsconf->u.inet.host = g_strdup(uri->server ? uri->server : "localhost");
405         if (uri->port) {
406             gsconf->u.inet.port = g_strdup_printf("%d", uri->port);
407         } else {
408             gsconf->u.inet.port = g_strdup_printf("%d", GLUSTER_DEFAULT_PORT);
409         }
410     }
411 
412 out:
413     if (qp) {
414         query_params_free(qp);
415     }
416     uri_free(uri);
417     return ret;
418 }
419 
420 static struct glfs *qemu_gluster_glfs_init(BlockdevOptionsGluster *gconf,
421                                            Error **errp)
422 {
423     struct glfs *glfs;
424     int ret;
425     int old_errno;
426     SocketAddressList *server;
427     unsigned long long port;
428 
429     glfs = glfs_find_preopened(gconf->volume);
430     if (glfs) {
431         return glfs;
432     }
433 
434     glfs = glfs_new(gconf->volume);
435     if (!glfs) {
436         goto out;
437     }
438 
439     glfs_set_preopened(gconf->volume, glfs);
440 
441     for (server = gconf->server; server; server = server->next) {
442         switch (server->value->type) {
443         case SOCKET_ADDRESS_TYPE_UNIX:
444             ret = glfs_set_volfile_server(glfs, "unix",
445                                    server->value->u.q_unix.path, 0);
446             break;
447         case SOCKET_ADDRESS_TYPE_INET:
448             if (parse_uint_full(server->value->u.inet.port, &port, 10) < 0 ||
449                 port > 65535) {
450                 error_setg(errp, "'%s' is not a valid port number",
451                            server->value->u.inet.port);
452                 errno = EINVAL;
453                 goto out;
454             }
455             ret = glfs_set_volfile_server(glfs, "tcp",
456                                    server->value->u.inet.host,
457                                    (int)port);
458             break;
459         case SOCKET_ADDRESS_TYPE_VSOCK:
460         case SOCKET_ADDRESS_TYPE_FD:
461         default:
462             abort();
463         }
464 
465         if (ret < 0) {
466             goto out;
467         }
468     }
469 
470     ret = glfs_set_logging(glfs, gconf->logfile, gconf->debug);
471     if (ret < 0) {
472         goto out;
473     }
474 
475     ret = glfs_init(glfs);
476     if (ret) {
477         error_setg(errp, "Gluster connection for volume %s, path %s failed"
478                          " to connect", gconf->volume, gconf->path);
479         for (server = gconf->server; server; server = server->next) {
480             if (server->value->type  == SOCKET_ADDRESS_TYPE_UNIX) {
481                 error_append_hint(errp, "hint: failed on socket %s ",
482                                   server->value->u.q_unix.path);
483             } else {
484                 error_append_hint(errp, "hint: failed on host %s and port %s ",
485                                   server->value->u.inet.host,
486                                   server->value->u.inet.port);
487             }
488         }
489 
490         error_append_hint(errp, "Please refer to gluster logs for more info\n");
491 
492         /* glfs_init sometimes doesn't set errno although docs suggest that */
493         if (errno == 0) {
494             errno = EINVAL;
495         }
496 
497         goto out;
498     }
499     return glfs;
500 
501 out:
502     if (glfs) {
503         old_errno = errno;
504         glfs_clear_preopened(glfs);
505         errno = old_errno;
506     }
507     return NULL;
508 }
509 
510 /*
511  * Convert the json formatted command line into qapi.
512 */
513 static int qemu_gluster_parse_json(BlockdevOptionsGluster *gconf,
514                                   QDict *options, Error **errp)
515 {
516     QemuOpts *opts;
517     SocketAddress *gsconf = NULL;
518     SocketAddressList **tail;
519     QDict *backing_options = NULL;
520     Error *local_err = NULL;
521     char *str = NULL;
522     const char *ptr;
523     int i, type, num_servers;
524 
525     /* create opts info from runtime_json_opts list */
526     opts = qemu_opts_create(&runtime_json_opts, NULL, 0, &error_abort);
527     if (!qemu_opts_absorb_qdict(opts, options, errp)) {
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     tail = &gconf->server;
552 
553     for (i = 0; i < num_servers; i++) {
554         str = g_strdup_printf(GLUSTER_OPT_SERVER_PATTERN"%d.", i);
555         qdict_extract_subqdict(options, &backing_options, str);
556 
557         /* create opts info from runtime_type_opts list */
558         opts = qemu_opts_create(&runtime_type_opts, NULL, 0, &error_abort);
559         if (!qemu_opts_absorb_qdict(opts, backing_options, errp)) {
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             if (!qemu_opts_absorb_qdict(opts, backing_options, errp)) {
590                 goto out;
591             }
592 
593             ptr = qemu_opt_get(opts, GLUSTER_OPT_HOST);
594             if (!ptr) {
595                 error_setg(&local_err, QERR_MISSING_PARAMETER,
596                            GLUSTER_OPT_HOST);
597                 error_append_hint(&local_err, GERR_INDEX_HINT, i);
598                 goto out;
599             }
600             gsconf->u.inet.host = g_strdup(ptr);
601             ptr = qemu_opt_get(opts, GLUSTER_OPT_PORT);
602             if (!ptr) {
603                 error_setg(&local_err, QERR_MISSING_PARAMETER,
604                            GLUSTER_OPT_PORT);
605                 error_append_hint(&local_err, GERR_INDEX_HINT, i);
606                 goto out;
607             }
608             gsconf->u.inet.port = g_strdup(ptr);
609 
610             /* defend for unsupported fields in InetSocketAddress,
611              * i.e. @ipv4, @ipv6  and @to
612              */
613             ptr = qemu_opt_get(opts, GLUSTER_OPT_TO);
614             if (ptr) {
615                 gsconf->u.inet.has_to = true;
616             }
617             ptr = qemu_opt_get(opts, GLUSTER_OPT_IPV4);
618             if (ptr) {
619                 gsconf->u.inet.has_ipv4 = true;
620             }
621             ptr = qemu_opt_get(opts, GLUSTER_OPT_IPV6);
622             if (ptr) {
623                 gsconf->u.inet.has_ipv6 = true;
624             }
625             if (gsconf->u.inet.has_to) {
626                 error_setg(&local_err, "Parameter 'to' not supported");
627                 goto out;
628             }
629             if (gsconf->u.inet.has_ipv4 || gsconf->u.inet.has_ipv6) {
630                 error_setg(&local_err, "Parameters 'ipv4/ipv6' not supported");
631                 goto out;
632             }
633             qemu_opts_del(opts);
634         } else {
635             /* create opts info from runtime_unix_opts list */
636             opts = qemu_opts_create(&runtime_unix_opts, NULL, 0, &error_abort);
637             if (!qemu_opts_absorb_qdict(opts, backing_options, errp)) {
638                 goto out;
639             }
640 
641             ptr = qemu_opt_get(opts, GLUSTER_OPT_PATH);
642             if (!ptr) {
643                 ptr = qemu_opt_get(opts, GLUSTER_OPT_SOCKET);
644             } else if (qemu_opt_get(opts, GLUSTER_OPT_SOCKET)) {
645                 error_setg(&local_err,
646                            "Conflicting parameters 'path' and 'socket'");
647                 error_append_hint(&local_err, GERR_INDEX_HINT, i);
648                 goto out;
649             }
650             if (!ptr) {
651                 error_setg(&local_err, QERR_MISSING_PARAMETER,
652                            GLUSTER_OPT_PATH);
653                 error_append_hint(&local_err, GERR_INDEX_HINT, i);
654                 goto out;
655             }
656             gsconf->u.q_unix.path = g_strdup(ptr);
657             qemu_opts_del(opts);
658         }
659 
660         QAPI_LIST_APPEND(tail, gsconf);
661         gsconf = NULL;
662 
663         qobject_unref(backing_options);
664         backing_options = NULL;
665         g_free(str);
666         str = NULL;
667     }
668 
669     return 0;
670 
671 out:
672     error_propagate(errp, local_err);
673     qapi_free_SocketAddress(gsconf);
674     qemu_opts_del(opts);
675     g_free(str);
676     qobject_unref(backing_options);
677     errno = EINVAL;
678     return -errno;
679 }
680 
681 /* Converts options given in @filename and the @options QDict into the QAPI
682  * object @gconf. */
683 static int qemu_gluster_parse(BlockdevOptionsGluster *gconf,
684                               const char *filename,
685                               QDict *options, Error **errp)
686 {
687     int ret;
688     if (filename) {
689         ret = qemu_gluster_parse_uri(gconf, filename);
690         if (ret < 0) {
691             error_setg(errp, "invalid URI %s", filename);
692             error_append_hint(errp, "Usage: file=gluster[+transport]://"
693                                     "[host[:port]]volume/path[?socket=...]"
694                                     "[,file.debug=N]"
695                                     "[,file.logfile=/path/filename.log]\n");
696             return ret;
697         }
698     } else {
699         ret = qemu_gluster_parse_json(gconf, options, errp);
700         if (ret < 0) {
701             error_append_hint(errp, "Usage: "
702                              "-drive driver=qcow2,file.driver=gluster,"
703                              "file.volume=testvol,file.path=/path/a.qcow2"
704                              "[,file.debug=9]"
705                              "[,file.logfile=/path/filename.log],"
706                              "file.server.0.type=inet,"
707                              "file.server.0.host=1.2.3.4,"
708                              "file.server.0.port=24007,"
709                              "file.server.1.transport=unix,"
710                              "file.server.1.path=/var/run/glusterd.socket ..."
711                              "\n");
712             return ret;
713         }
714     }
715 
716     return 0;
717 }
718 
719 static struct glfs *qemu_gluster_init(BlockdevOptionsGluster *gconf,
720                                       const char *filename,
721                                       QDict *options, Error **errp)
722 {
723     int ret;
724 
725     ret = qemu_gluster_parse(gconf, filename, options, errp);
726     if (ret < 0) {
727         errno = -ret;
728         return NULL;
729     }
730 
731     return qemu_gluster_glfs_init(gconf, errp);
732 }
733 
734 /*
735  * AIO callback routine called from GlusterFS thread.
736  */
737 static void gluster_finish_aiocb(struct glfs_fd *fd, ssize_t ret,
738 #ifdef CONFIG_GLUSTERFS_IOCB_HAS_STAT
739                                  struct glfs_stat *pre, struct glfs_stat *post,
740 #endif
741                                  void *arg)
742 {
743     GlusterAIOCB *acb = (GlusterAIOCB *)arg;
744 
745     if (!ret || ret == acb->size) {
746         acb->ret = 0; /* Success */
747     } else if (ret < 0) {
748         acb->ret = -errno; /* Read/Write failed */
749     } else {
750         acb->ret = -EIO; /* Partial read/write - fail it */
751     }
752 
753     aio_co_schedule(acb->aio_context, acb->coroutine);
754 }
755 
756 static void qemu_gluster_parse_flags(int bdrv_flags, int *open_flags)
757 {
758     assert(open_flags != NULL);
759 
760     *open_flags |= O_BINARY;
761 
762     if (bdrv_flags & BDRV_O_RDWR) {
763         *open_flags |= O_RDWR;
764     } else {
765         *open_flags |= O_RDONLY;
766     }
767 
768     if ((bdrv_flags & BDRV_O_NOCACHE)) {
769         *open_flags |= O_DIRECT;
770     }
771 }
772 
773 /*
774  * Do SEEK_DATA/HOLE to detect if it is functional. Older broken versions of
775  * gfapi incorrectly return the current offset when SEEK_DATA/HOLE is used.
776  * - Corrected versions return -1 and set errno to EINVAL.
777  * - Versions that support SEEK_DATA/HOLE correctly, will return -1 and set
778  *   errno to ENXIO when SEEK_DATA is called with a position of EOF.
779  */
780 static bool qemu_gluster_test_seek(struct glfs_fd *fd)
781 {
782     off_t ret = 0;
783 
784 #if defined SEEK_HOLE && defined SEEK_DATA
785     off_t eof;
786 
787     eof = glfs_lseek(fd, 0, SEEK_END);
788     if (eof < 0) {
789         /* this should never occur */
790         return false;
791     }
792 
793     /* this should always fail with ENXIO if SEEK_DATA is supported */
794     ret = glfs_lseek(fd, eof, SEEK_DATA);
795 #endif
796 
797     return (ret < 0) && (errno == ENXIO);
798 }
799 
800 static int qemu_gluster_open(BlockDriverState *bs,  QDict *options,
801                              int bdrv_flags, Error **errp)
802 {
803     BDRVGlusterState *s = bs->opaque;
804     int open_flags = 0;
805     int ret = 0;
806     BlockdevOptionsGluster *gconf = NULL;
807     QemuOpts *opts;
808     const char *filename, *logfile;
809 
810     opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
811     if (!qemu_opts_absorb_qdict(opts, options, errp)) {
812         ret = -EINVAL;
813         goto out;
814     }
815 
816     filename = qemu_opt_get(opts, GLUSTER_OPT_FILENAME);
817 
818     s->debug = qemu_opt_get_number(opts, GLUSTER_OPT_DEBUG,
819                                    GLUSTER_DEBUG_DEFAULT);
820     if (s->debug < 0) {
821         s->debug = 0;
822     } else if (s->debug > GLUSTER_DEBUG_MAX) {
823         s->debug = GLUSTER_DEBUG_MAX;
824     }
825 
826     gconf = g_new0(BlockdevOptionsGluster, 1);
827     gconf->debug = s->debug;
828     gconf->has_debug = true;
829 
830     logfile = qemu_opt_get(opts, GLUSTER_OPT_LOGFILE);
831     s->logfile = g_strdup(logfile ? logfile : GLUSTER_LOGFILE_DEFAULT);
832 
833     gconf->logfile = g_strdup(s->logfile);
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     bs->bl.max_pdiscard = MIN(SIZE_MAX, INT64_MAX);
895 }
896 
897 static int qemu_gluster_reopen_prepare(BDRVReopenState *state,
898                                        BlockReopenQueue *queue, Error **errp)
899 {
900     int ret = 0;
901     BDRVGlusterState *s;
902     BDRVGlusterReopenState *reop_s;
903     BlockdevOptionsGluster *gconf;
904     int open_flags = 0;
905 
906     assert(state != NULL);
907     assert(state->bs != NULL);
908 
909     s = state->bs->opaque;
910 
911     state->opaque = g_new0(BDRVGlusterReopenState, 1);
912     reop_s = state->opaque;
913 
914     qemu_gluster_parse_flags(state->flags, &open_flags);
915 
916     gconf = g_new0(BlockdevOptionsGluster, 1);
917     gconf->debug = s->debug;
918     gconf->has_debug = true;
919     gconf->logfile = g_strdup(s->logfile);
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                                                       int64_t bytes,
1007                                                       BdrvRequestFlags flags)
1008 {
1009     int ret;
1010     GlusterAIOCB acb;
1011     BDRVGlusterState *s = bs->opaque;
1012 
1013     acb.size = bytes;
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, bytes, 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 
1165     ret = qemu_gluster_parse(gconf, filename, NULL, errp);
1166     if (ret < 0) {
1167         goto fail;
1168     }
1169 
1170     ret = qemu_gluster_co_create(options, errp);
1171     if (ret < 0) {
1172         goto fail;
1173     }
1174 
1175     ret = 0;
1176 fail:
1177     qapi_free_BlockdevCreateOptions(options);
1178     return ret;
1179 }
1180 
1181 static coroutine_fn int qemu_gluster_co_rw(BlockDriverState *bs,
1182                                            int64_t sector_num, int nb_sectors,
1183                                            QEMUIOVector *qiov, int write)
1184 {
1185     int ret;
1186     GlusterAIOCB acb;
1187     BDRVGlusterState *s = bs->opaque;
1188     size_t size = nb_sectors * BDRV_SECTOR_SIZE;
1189     off_t offset = sector_num * BDRV_SECTOR_SIZE;
1190 
1191     acb.size = size;
1192     acb.ret = 0;
1193     acb.coroutine = qemu_coroutine_self();
1194     acb.aio_context = bdrv_get_aio_context(bs);
1195 
1196     if (write) {
1197         ret = glfs_pwritev_async(s->fd, qiov->iov, qiov->niov, offset, 0,
1198                                  gluster_finish_aiocb, &acb);
1199     } else {
1200         ret = glfs_preadv_async(s->fd, qiov->iov, qiov->niov, offset, 0,
1201                                 gluster_finish_aiocb, &acb);
1202     }
1203 
1204     if (ret < 0) {
1205         return -errno;
1206     }
1207 
1208     qemu_coroutine_yield();
1209     return acb.ret;
1210 }
1211 
1212 static coroutine_fn int qemu_gluster_co_truncate(BlockDriverState *bs,
1213                                                  int64_t offset,
1214                                                  bool exact,
1215                                                  PreallocMode prealloc,
1216                                                  BdrvRequestFlags flags,
1217                                                  Error **errp)
1218 {
1219     BDRVGlusterState *s = bs->opaque;
1220     return qemu_gluster_do_truncate(s->fd, offset, prealloc, errp);
1221 }
1222 
1223 static coroutine_fn int qemu_gluster_co_readv(BlockDriverState *bs,
1224                                               int64_t sector_num,
1225                                               int nb_sectors,
1226                                               QEMUIOVector *qiov)
1227 {
1228     return qemu_gluster_co_rw(bs, sector_num, nb_sectors, qiov, 0);
1229 }
1230 
1231 static coroutine_fn int qemu_gluster_co_writev(BlockDriverState *bs,
1232                                                int64_t sector_num,
1233                                                int nb_sectors,
1234                                                QEMUIOVector *qiov,
1235                                                int flags)
1236 {
1237     return qemu_gluster_co_rw(bs, sector_num, nb_sectors, qiov, 1);
1238 }
1239 
1240 static void qemu_gluster_close(BlockDriverState *bs)
1241 {
1242     BDRVGlusterState *s = bs->opaque;
1243 
1244     g_free(s->logfile);
1245     if (s->fd) {
1246         glfs_close(s->fd);
1247         s->fd = NULL;
1248     }
1249     glfs_clear_preopened(s->glfs);
1250 }
1251 
1252 static coroutine_fn int qemu_gluster_co_flush_to_disk(BlockDriverState *bs)
1253 {
1254     int ret;
1255     GlusterAIOCB acb;
1256     BDRVGlusterState *s = bs->opaque;
1257 
1258     acb.size = 0;
1259     acb.ret = 0;
1260     acb.coroutine = qemu_coroutine_self();
1261     acb.aio_context = bdrv_get_aio_context(bs);
1262 
1263     ret = glfs_fsync_async(s->fd, gluster_finish_aiocb, &acb);
1264     if (ret < 0) {
1265         ret = -errno;
1266         goto error;
1267     }
1268 
1269     qemu_coroutine_yield();
1270     if (acb.ret < 0) {
1271         ret = acb.ret;
1272         goto error;
1273     }
1274 
1275     return acb.ret;
1276 
1277 error:
1278     /* Some versions of Gluster (3.5.6 -> 3.5.8?) will not retain its cache
1279      * after a fsync failure, so we have no way of allowing the guest to safely
1280      * continue.  Gluster versions prior to 3.5.6 don't retain the cache
1281      * either, but will invalidate the fd on error, so this is again our only
1282      * option.
1283      *
1284      * The 'resync-failed-syncs-after-fsync' xlator option for the
1285      * write-behind cache will cause later gluster versions to retain its
1286      * cache after error, so long as the fd remains open.  However, we
1287      * currently have no way of knowing if this option is supported.
1288      *
1289      * TODO: Once gluster provides a way for us to determine if the option
1290      * is supported, bypass the closure and setting drv to NULL.  */
1291     qemu_gluster_close(bs);
1292     bs->drv = NULL;
1293     return ret;
1294 }
1295 
1296 #ifdef CONFIG_GLUSTERFS_DISCARD
1297 static coroutine_fn int qemu_gluster_co_pdiscard(BlockDriverState *bs,
1298                                                  int64_t offset, int64_t bytes)
1299 {
1300     int ret;
1301     GlusterAIOCB acb;
1302     BDRVGlusterState *s = bs->opaque;
1303 
1304     assert(bytes <= SIZE_MAX); /* rely on max_pdiscard */
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, bytes, 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 coroutine_fn qemu_gluster_co_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 coroutine_fn
1335 qemu_gluster_co_get_allocated_file_size(BlockDriverState *bs)
1336 {
1337     BDRVGlusterState *s = bs->opaque;
1338     struct stat st;
1339     int ret;
1340 
1341     ret = glfs_fstat(s->fd, &st);
1342     if (ret < 0) {
1343         return -errno;
1344     } else {
1345         return st.st_blocks * 512;
1346     }
1347 }
1348 
1349 /*
1350  * Find allocation range in @bs around offset @start.
1351  * May change underlying file descriptor's file offset.
1352  * If @start is not in a hole, store @start in @data, and the
1353  * beginning of the next hole in @hole, and return 0.
1354  * If @start is in a non-trailing hole, store @start in @hole and the
1355  * beginning of the next non-hole in @data, and return 0.
1356  * If @start is in a trailing hole or beyond EOF, return -ENXIO.
1357  * If we can't find out, return a negative errno other than -ENXIO.
1358  *
1359  * (Shamefully copied from file-posix.c, only minuscule adaptions.)
1360  */
1361 static int find_allocation(BlockDriverState *bs, off_t start,
1362                            off_t *data, off_t *hole)
1363 {
1364     BDRVGlusterState *s = bs->opaque;
1365 
1366     if (!s->supports_seek_data) {
1367         goto exit;
1368     }
1369 
1370 #if defined SEEK_HOLE && defined SEEK_DATA
1371     off_t offs;
1372 
1373     /*
1374      * SEEK_DATA cases:
1375      * D1. offs == start: start is in data
1376      * D2. offs > start: start is in a hole, next data at offs
1377      * D3. offs < 0, errno = ENXIO: either start is in a trailing hole
1378      *                              or start is beyond EOF
1379      *     If the latter happens, the file has been truncated behind
1380      *     our back since we opened it.  All bets are off then.
1381      *     Treating like a trailing hole is simplest.
1382      * D4. offs < 0, errno != ENXIO: we learned nothing
1383      */
1384     offs = glfs_lseek(s->fd, start, SEEK_DATA);
1385     if (offs < 0) {
1386         return -errno;          /* D3 or D4 */
1387     }
1388 
1389     if (offs < start) {
1390         /* This is not a valid return by lseek().  We are safe to just return
1391          * -EIO in this case, and we'll treat it like D4. Unfortunately some
1392          *  versions of gluster server will return offs < start, so an assert
1393          *  here will unnecessarily abort QEMU. */
1394         return -EIO;
1395     }
1396 
1397     if (offs > start) {
1398         /* D2: in hole, next data at offs */
1399         *hole = start;
1400         *data = offs;
1401         return 0;
1402     }
1403 
1404     /* D1: in data, end not yet known */
1405 
1406     /*
1407      * SEEK_HOLE cases:
1408      * H1. offs == start: start is in a hole
1409      *     If this happens here, a hole has been dug behind our back
1410      *     since the previous lseek().
1411      * H2. offs > start: either start is in data, next hole at offs,
1412      *                   or start is in trailing hole, EOF at offs
1413      *     Linux treats trailing holes like any other hole: offs ==
1414      *     start.  Solaris seeks to EOF instead: offs > start (blech).
1415      *     If that happens here, a hole has been dug behind our back
1416      *     since the previous lseek().
1417      * H3. offs < 0, errno = ENXIO: start is beyond EOF
1418      *     If this happens, the file has been truncated behind our
1419      *     back since we opened it.  Treat it like a trailing hole.
1420      * H4. offs < 0, errno != ENXIO: we learned nothing
1421      *     Pretend we know nothing at all, i.e. "forget" about D1.
1422      */
1423     offs = glfs_lseek(s->fd, start, SEEK_HOLE);
1424     if (offs < 0) {
1425         return -errno;          /* D1 and (H3 or H4) */
1426     }
1427 
1428     if (offs < start) {
1429         /* This is not a valid return by lseek().  We are safe to just return
1430          * -EIO in this case, and we'll treat it like H4. Unfortunately some
1431          *  versions of gluster server will return offs < start, so an assert
1432          *  here will unnecessarily abort QEMU. */
1433         return -EIO;
1434     }
1435 
1436     if (offs > start) {
1437         /*
1438          * D1 and H2: either in data, next hole at offs, or it was in
1439          * data but is now in a trailing hole.  In the latter case,
1440          * all bets are off.  Treating it as if it there was data all
1441          * the way to EOF is safe, so simply do that.
1442          */
1443         *data = start;
1444         *hole = offs;
1445         return 0;
1446     }
1447 
1448     /* D1 and H1 */
1449     return -EBUSY;
1450 #endif
1451 
1452 exit:
1453     return -ENOTSUP;
1454 }
1455 
1456 /*
1457  * Returns the allocation status of the specified offset.
1458  *
1459  * The block layer guarantees 'offset' and 'bytes' are within bounds.
1460  *
1461  * 'pnum' is set to the number of bytes (including and immediately following
1462  * the specified offset) that are known to be in the same
1463  * allocated/unallocated state.
1464  *
1465  * 'bytes' is a soft cap for 'pnum'.  If the information is free, 'pnum' may
1466  * well exceed it.
1467  *
1468  * (Based on raw_co_block_status() from file-posix.c.)
1469  */
1470 static int coroutine_fn qemu_gluster_co_block_status(BlockDriverState *bs,
1471                                                      bool want_zero,
1472                                                      int64_t offset,
1473                                                      int64_t bytes,
1474                                                      int64_t *pnum,
1475                                                      int64_t *map,
1476                                                      BlockDriverState **file)
1477 {
1478     BDRVGlusterState *s = bs->opaque;
1479     off_t data = 0, hole = 0;
1480     int ret = -EINVAL;
1481 
1482     assert(QEMU_IS_ALIGNED(offset | bytes, bs->bl.request_alignment));
1483 
1484     if (!s->fd) {
1485         return ret;
1486     }
1487 
1488     if (!want_zero) {
1489         *pnum = bytes;
1490         *map = offset;
1491         *file = bs;
1492         return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID;
1493     }
1494 
1495     ret = find_allocation(bs, offset, &data, &hole);
1496     if (ret == -ENXIO) {
1497         /* Trailing hole */
1498         *pnum = bytes;
1499         ret = BDRV_BLOCK_ZERO;
1500     } else if (ret < 0) {
1501         /* No info available, so pretend there are no holes */
1502         *pnum = bytes;
1503         ret = BDRV_BLOCK_DATA;
1504     } else if (data == offset) {
1505         /* On a data extent, compute bytes to the end of the extent,
1506          * possibly including a partial sector at EOF. */
1507         *pnum = hole - offset;
1508 
1509         /*
1510          * We are not allowed to return partial sectors, though, so
1511          * round up if necessary.
1512          */
1513         if (!QEMU_IS_ALIGNED(*pnum, bs->bl.request_alignment)) {
1514             int64_t file_length = qemu_gluster_co_getlength(bs);
1515             if (file_length > 0) {
1516                 /* Ignore errors, this is just a safeguard */
1517                 assert(hole == file_length);
1518             }
1519             *pnum = ROUND_UP(*pnum, bs->bl.request_alignment);
1520         }
1521 
1522         ret = BDRV_BLOCK_DATA;
1523     } else {
1524         /* On a hole, compute bytes to the beginning of the next extent.  */
1525         assert(hole == offset);
1526         *pnum = data - offset;
1527         ret = BDRV_BLOCK_ZERO;
1528     }
1529 
1530     *map = offset;
1531     *file = bs;
1532 
1533     return ret | BDRV_BLOCK_OFFSET_VALID;
1534 }
1535 
1536 
1537 static const char *const gluster_strong_open_opts[] = {
1538     GLUSTER_OPT_VOLUME,
1539     GLUSTER_OPT_PATH,
1540     GLUSTER_OPT_TYPE,
1541     GLUSTER_OPT_SERVER_PATTERN,
1542     GLUSTER_OPT_HOST,
1543     GLUSTER_OPT_PORT,
1544     GLUSTER_OPT_TO,
1545     GLUSTER_OPT_IPV4,
1546     GLUSTER_OPT_IPV6,
1547     GLUSTER_OPT_SOCKET,
1548 
1549     NULL
1550 };
1551 
1552 static BlockDriver bdrv_gluster = {
1553     .format_name                  = "gluster",
1554     .protocol_name                = "gluster",
1555     .instance_size                = sizeof(BDRVGlusterState),
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_co_getlength            = qemu_gluster_co_getlength,
1564     .bdrv_co_get_allocated_file_size = qemu_gluster_co_get_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_file_open               = qemu_gluster_open,
1586     .bdrv_reopen_prepare          = qemu_gluster_reopen_prepare,
1587     .bdrv_reopen_commit           = qemu_gluster_reopen_commit,
1588     .bdrv_reopen_abort            = qemu_gluster_reopen_abort,
1589     .bdrv_close                   = qemu_gluster_close,
1590     .bdrv_co_create               = qemu_gluster_co_create,
1591     .bdrv_co_create_opts          = qemu_gluster_co_create_opts,
1592     .bdrv_co_getlength            = qemu_gluster_co_getlength,
1593     .bdrv_co_get_allocated_file_size = qemu_gluster_co_get_allocated_file_size,
1594     .bdrv_co_truncate             = qemu_gluster_co_truncate,
1595     .bdrv_co_readv                = qemu_gluster_co_readv,
1596     .bdrv_co_writev               = qemu_gluster_co_writev,
1597     .bdrv_co_flush_to_disk        = qemu_gluster_co_flush_to_disk,
1598 #ifdef CONFIG_GLUSTERFS_DISCARD
1599     .bdrv_co_pdiscard             = qemu_gluster_co_pdiscard,
1600 #endif
1601 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1602     .bdrv_co_pwrite_zeroes        = qemu_gluster_co_pwrite_zeroes,
1603 #endif
1604     .bdrv_co_block_status         = qemu_gluster_co_block_status,
1605     .bdrv_refresh_limits          = qemu_gluster_refresh_limits,
1606     .create_opts                  = &qemu_gluster_create_opts,
1607     .strong_runtime_opts          = gluster_strong_open_opts,
1608 };
1609 
1610 static BlockDriver bdrv_gluster_unix = {
1611     .format_name                  = "gluster",
1612     .protocol_name                = "gluster+unix",
1613     .instance_size                = sizeof(BDRVGlusterState),
1614     .bdrv_file_open               = qemu_gluster_open,
1615     .bdrv_reopen_prepare          = qemu_gluster_reopen_prepare,
1616     .bdrv_reopen_commit           = qemu_gluster_reopen_commit,
1617     .bdrv_reopen_abort            = qemu_gluster_reopen_abort,
1618     .bdrv_close                   = qemu_gluster_close,
1619     .bdrv_co_create               = qemu_gluster_co_create,
1620     .bdrv_co_create_opts          = qemu_gluster_co_create_opts,
1621     .bdrv_co_getlength            = qemu_gluster_co_getlength,
1622     .bdrv_co_get_allocated_file_size = qemu_gluster_co_get_allocated_file_size,
1623     .bdrv_co_truncate             = qemu_gluster_co_truncate,
1624     .bdrv_co_readv                = qemu_gluster_co_readv,
1625     .bdrv_co_writev               = qemu_gluster_co_writev,
1626     .bdrv_co_flush_to_disk        = qemu_gluster_co_flush_to_disk,
1627 #ifdef CONFIG_GLUSTERFS_DISCARD
1628     .bdrv_co_pdiscard             = qemu_gluster_co_pdiscard,
1629 #endif
1630 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1631     .bdrv_co_pwrite_zeroes        = qemu_gluster_co_pwrite_zeroes,
1632 #endif
1633     .bdrv_co_block_status         = qemu_gluster_co_block_status,
1634     .bdrv_refresh_limits          = qemu_gluster_refresh_limits,
1635     .create_opts                  = &qemu_gluster_create_opts,
1636     .strong_runtime_opts          = gluster_strong_open_opts,
1637 };
1638 
1639 /* rdma is deprecated (actually never supported for volfile fetch).
1640  * Let's maintain it for the protocol compatibility, to make sure things
1641  * won't break immediately. For now, gluster+rdma will fall back to gluster+tcp
1642  * protocol with a warning.
1643  * TODO: remove gluster+rdma interface support
1644  */
1645 static BlockDriver bdrv_gluster_rdma = {
1646     .format_name                  = "gluster",
1647     .protocol_name                = "gluster+rdma",
1648     .instance_size                = sizeof(BDRVGlusterState),
1649     .bdrv_file_open               = qemu_gluster_open,
1650     .bdrv_reopen_prepare          = qemu_gluster_reopen_prepare,
1651     .bdrv_reopen_commit           = qemu_gluster_reopen_commit,
1652     .bdrv_reopen_abort            = qemu_gluster_reopen_abort,
1653     .bdrv_close                   = qemu_gluster_close,
1654     .bdrv_co_create               = qemu_gluster_co_create,
1655     .bdrv_co_create_opts          = qemu_gluster_co_create_opts,
1656     .bdrv_co_getlength            = qemu_gluster_co_getlength,
1657     .bdrv_co_get_allocated_file_size = qemu_gluster_co_get_allocated_file_size,
1658     .bdrv_co_truncate             = qemu_gluster_co_truncate,
1659     .bdrv_co_readv                = qemu_gluster_co_readv,
1660     .bdrv_co_writev               = qemu_gluster_co_writev,
1661     .bdrv_co_flush_to_disk        = qemu_gluster_co_flush_to_disk,
1662 #ifdef CONFIG_GLUSTERFS_DISCARD
1663     .bdrv_co_pdiscard             = qemu_gluster_co_pdiscard,
1664 #endif
1665 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1666     .bdrv_co_pwrite_zeroes        = qemu_gluster_co_pwrite_zeroes,
1667 #endif
1668     .bdrv_co_block_status         = qemu_gluster_co_block_status,
1669     .bdrv_refresh_limits          = qemu_gluster_refresh_limits,
1670     .create_opts                  = &qemu_gluster_create_opts,
1671     .strong_runtime_opts          = gluster_strong_open_opts,
1672 };
1673 
1674 static void bdrv_gluster_init(void)
1675 {
1676     bdrv_register(&bdrv_gluster_rdma);
1677     bdrv_register(&bdrv_gluster_unix);
1678     bdrv_register(&bdrv_gluster_tcp);
1679     bdrv_register(&bdrv_gluster);
1680 }
1681 
1682 block_init(bdrv_gluster_init);
1683