xref: /openbmc/qemu/block/gluster.c (revision ca693d1c)
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/option.h"
22 #include "qemu/cutils.h"
23 
24 #ifdef CONFIG_GLUSTERFS_FTRUNCATE_HAS_STAT
25 # define glfs_ftruncate(fd, offset) glfs_ftruncate(fd, offset, NULL, NULL)
26 #endif
27 
28 #define GLUSTER_OPT_FILENAME        "filename"
29 #define GLUSTER_OPT_VOLUME          "volume"
30 #define GLUSTER_OPT_PATH            "path"
31 #define GLUSTER_OPT_TYPE            "type"
32 #define GLUSTER_OPT_SERVER_PATTERN  "server."
33 #define GLUSTER_OPT_HOST            "host"
34 #define GLUSTER_OPT_PORT            "port"
35 #define GLUSTER_OPT_TO              "to"
36 #define GLUSTER_OPT_IPV4            "ipv4"
37 #define GLUSTER_OPT_IPV6            "ipv6"
38 #define GLUSTER_OPT_SOCKET          "socket"
39 #define GLUSTER_OPT_DEBUG           "debug"
40 #define GLUSTER_DEFAULT_PORT        24007
41 #define GLUSTER_DEBUG_DEFAULT       4
42 #define GLUSTER_DEBUG_MAX           9
43 #define GLUSTER_OPT_LOGFILE         "logfile"
44 #define GLUSTER_LOGFILE_DEFAULT     "-" /* handled in libgfapi as /dev/stderr */
45 /*
46  * Several versions of GlusterFS (3.12? -> 6.0.1) fail when the transfer size
47  * is greater or equal to 1024 MiB, so we are limiting the transfer size to 512
48  * MiB to avoid this rare issue.
49  */
50 #define GLUSTER_MAX_TRANSFER        (512 * MiB)
51 
52 #define GERR_INDEX_HINT "hint: check in 'server' array index '%d'\n"
53 
54 typedef struct GlusterAIOCB {
55     int64_t size;
56     int ret;
57     Coroutine *coroutine;
58     AioContext *aio_context;
59 } GlusterAIOCB;
60 
61 typedef struct BDRVGlusterState {
62     struct glfs *glfs;
63     struct glfs_fd *fd;
64     char *logfile;
65     bool supports_seek_data;
66     int debug;
67 } BDRVGlusterState;
68 
69 typedef struct BDRVGlusterReopenState {
70     struct glfs *glfs;
71     struct glfs_fd *fd;
72 } BDRVGlusterReopenState;
73 
74 
75 typedef struct GlfsPreopened {
76     char *volume;
77     glfs_t *fs;
78     int ref;
79 } GlfsPreopened;
80 
81 typedef struct ListElement {
82     QLIST_ENTRY(ListElement) list;
83     GlfsPreopened saved;
84 } ListElement;
85 
86 static QLIST_HEAD(, ListElement) glfs_list;
87 
88 static QemuOptsList qemu_gluster_create_opts = {
89     .name = "qemu-gluster-create-opts",
90     .head = QTAILQ_HEAD_INITIALIZER(qemu_gluster_create_opts.head),
91     .desc = {
92         {
93             .name = BLOCK_OPT_SIZE,
94             .type = QEMU_OPT_SIZE,
95             .help = "Virtual disk size"
96         },
97         {
98             .name = BLOCK_OPT_PREALLOC,
99             .type = QEMU_OPT_STRING,
100             .help = "Preallocation mode (allowed values: off, full)"
101         },
102         {
103             .name = GLUSTER_OPT_DEBUG,
104             .type = QEMU_OPT_NUMBER,
105             .help = "Gluster log level, valid range is 0-9",
106         },
107         {
108             .name = GLUSTER_OPT_LOGFILE,
109             .type = QEMU_OPT_STRING,
110             .help = "Logfile path of libgfapi",
111         },
112         { /* end of list */ }
113     }
114 };
115 
116 static QemuOptsList runtime_opts = {
117     .name = "gluster",
118     .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
119     .desc = {
120         {
121             .name = GLUSTER_OPT_FILENAME,
122             .type = QEMU_OPT_STRING,
123             .help = "URL to the gluster image",
124         },
125         {
126             .name = GLUSTER_OPT_DEBUG,
127             .type = QEMU_OPT_NUMBER,
128             .help = "Gluster log level, valid range is 0-9",
129         },
130         {
131             .name = GLUSTER_OPT_LOGFILE,
132             .type = QEMU_OPT_STRING,
133             .help = "Logfile path of libgfapi",
134         },
135         { /* end of list */ }
136     },
137 };
138 
139 static QemuOptsList runtime_json_opts = {
140     .name = "gluster_json",
141     .head = QTAILQ_HEAD_INITIALIZER(runtime_json_opts.head),
142     .desc = {
143         {
144             .name = GLUSTER_OPT_VOLUME,
145             .type = QEMU_OPT_STRING,
146             .help = "name of gluster volume where VM image resides",
147         },
148         {
149             .name = GLUSTER_OPT_PATH,
150             .type = QEMU_OPT_STRING,
151             .help = "absolute path to image file in gluster volume",
152         },
153         {
154             .name = GLUSTER_OPT_DEBUG,
155             .type = QEMU_OPT_NUMBER,
156             .help = "Gluster log level, valid range is 0-9",
157         },
158         { /* end of list */ }
159     },
160 };
161 
162 static QemuOptsList runtime_type_opts = {
163     .name = "gluster_type",
164     .head = QTAILQ_HEAD_INITIALIZER(runtime_type_opts.head),
165     .desc = {
166         {
167             .name = GLUSTER_OPT_TYPE,
168             .type = QEMU_OPT_STRING,
169             .help = "inet|unix",
170         },
171         { /* end of list */ }
172     },
173 };
174 
175 static QemuOptsList runtime_unix_opts = {
176     .name = "gluster_unix",
177     .head = QTAILQ_HEAD_INITIALIZER(runtime_unix_opts.head),
178     .desc = {
179         {
180             .name = GLUSTER_OPT_SOCKET,
181             .type = QEMU_OPT_STRING,
182             .help = "socket file path (legacy)",
183         },
184         {
185             .name = GLUSTER_OPT_PATH,
186             .type = QEMU_OPT_STRING,
187             .help = "socket file path (QAPI)",
188         },
189         { /* end of list */ }
190     },
191 };
192 
193 static QemuOptsList runtime_inet_opts = {
194     .name = "gluster_inet",
195     .head = QTAILQ_HEAD_INITIALIZER(runtime_inet_opts.head),
196     .desc = {
197         {
198             .name = GLUSTER_OPT_TYPE,
199             .type = QEMU_OPT_STRING,
200             .help = "inet|unix",
201         },
202         {
203             .name = GLUSTER_OPT_HOST,
204             .type = QEMU_OPT_STRING,
205             .help = "host address (hostname/ipv4/ipv6 addresses)",
206         },
207         {
208             .name = GLUSTER_OPT_PORT,
209             .type = QEMU_OPT_STRING,
210             .help = "port number on which glusterd is listening (default 24007)",
211         },
212         {
213             .name = "to",
214             .type = QEMU_OPT_NUMBER,
215             .help = "max port number, not supported by gluster",
216         },
217         {
218             .name = "ipv4",
219             .type = QEMU_OPT_BOOL,
220             .help = "ipv4 bool value, not supported by gluster",
221         },
222         {
223             .name = "ipv6",
224             .type = QEMU_OPT_BOOL,
225             .help = "ipv6 bool value, not supported by gluster",
226         },
227         { /* end of list */ }
228     },
229 };
230 
231 static void glfs_set_preopened(const char *volume, glfs_t *fs)
232 {
233     ListElement *entry = NULL;
234 
235     entry = g_new(ListElement, 1);
236 
237     entry->saved.volume = g_strdup(volume);
238 
239     entry->saved.fs = fs;
240     entry->saved.ref = 1;
241 
242     QLIST_INSERT_HEAD(&glfs_list, entry, list);
243 }
244 
245 static glfs_t *glfs_find_preopened(const char *volume)
246 {
247     ListElement *entry = NULL;
248 
249      QLIST_FOREACH(entry, &glfs_list, list) {
250         if (strcmp(entry->saved.volume, volume) == 0) {
251             entry->saved.ref++;
252             return entry->saved.fs;
253         }
254      }
255 
256     return NULL;
257 }
258 
259 static void glfs_clear_preopened(glfs_t *fs)
260 {
261     ListElement *entry = NULL;
262     ListElement *next;
263 
264     if (fs == NULL) {
265         return;
266     }
267 
268     QLIST_FOREACH_SAFE(entry, &glfs_list, list, next) {
269         if (entry->saved.fs == fs) {
270             if (--entry->saved.ref) {
271                 return;
272             }
273 
274             QLIST_REMOVE(entry, list);
275 
276             glfs_fini(entry->saved.fs);
277             g_free(entry->saved.volume);
278             g_free(entry);
279         }
280     }
281 }
282 
283 static int parse_volume_options(BlockdevOptionsGluster *gconf, char *path)
284 {
285     char *p, *q;
286 
287     if (!path) {
288         return -EINVAL;
289     }
290 
291     /* volume */
292     p = q = path + strspn(path, "/");
293     p += strcspn(p, "/");
294     if (*p == '\0') {
295         return -EINVAL;
296     }
297     gconf->volume = g_strndup(q, p - q);
298 
299     /* path */
300     p += strspn(p, "/");
301     if (*p == '\0') {
302         return -EINVAL;
303     }
304     gconf->path = g_strdup(p);
305     return 0;
306 }
307 
308 /*
309  * file=gluster[+transport]://[host[:port]]/volume/path[?socket=...]
310  *
311  * 'gluster' is the protocol.
312  *
313  * 'transport' specifies the transport type used to connect to gluster
314  * management daemon (glusterd). Valid transport types are
315  * tcp or unix. If a transport type isn't specified, then tcp type is assumed.
316  *
317  * 'host' specifies the host where the volume file specification for
318  * the given volume resides. This can be either hostname or ipv4 address.
319  * If transport type is 'unix', then 'host' field should not be specified.
320  * The 'socket' field needs to be populated with the path to unix domain
321  * socket.
322  *
323  * 'port' is the port number on which glusterd is listening. This is optional
324  * and if not specified, QEMU will send 0 which will make gluster to use the
325  * default port. If the transport type is unix, then 'port' should not be
326  * specified.
327  *
328  * 'volume' is the name of the gluster volume which contains the VM image.
329  *
330  * 'path' is the path to the actual VM image that resides on gluster volume.
331  *
332  * Examples:
333  *
334  * file=gluster://1.2.3.4/testvol/a.img
335  * file=gluster+tcp://1.2.3.4/testvol/a.img
336  * file=gluster+tcp://1.2.3.4:24007/testvol/dir/a.img
337  * file=gluster+tcp://host.domain.com:24007/testvol/dir/a.img
338  * file=gluster+unix:///testvol/dir/a.img?socket=/tmp/glusterd.socket
339  */
340 static int qemu_gluster_parse_uri(BlockdevOptionsGluster *gconf,
341                                   const char *filename)
342 {
343     SocketAddress *gsconf;
344     URI *uri;
345     QueryParams *qp = NULL;
346     bool is_unix = false;
347     int ret = 0;
348 
349     uri = uri_parse(filename);
350     if (!uri) {
351         return -EINVAL;
352     }
353 
354     gconf->server = g_new0(SocketAddressList, 1);
355     gconf->server->value = gsconf = g_new0(SocketAddress, 1);
356 
357     /* transport */
358     if (!uri->scheme || !strcmp(uri->scheme, "gluster")) {
359         gsconf->type = SOCKET_ADDRESS_TYPE_INET;
360     } else if (!strcmp(uri->scheme, "gluster+tcp")) {
361         gsconf->type = SOCKET_ADDRESS_TYPE_INET;
362     } else if (!strcmp(uri->scheme, "gluster+unix")) {
363         gsconf->type = SOCKET_ADDRESS_TYPE_UNIX;
364         is_unix = true;
365     } else if (!strcmp(uri->scheme, "gluster+rdma")) {
366         gsconf->type = SOCKET_ADDRESS_TYPE_INET;
367         warn_report("rdma feature is not supported, falling back to tcp");
368     } else {
369         ret = -EINVAL;
370         goto out;
371     }
372 
373     ret = parse_volume_options(gconf, uri->path);
374     if (ret < 0) {
375         goto out;
376     }
377 
378     qp = query_params_parse(uri->query);
379     if (qp->n > 1 || (is_unix && !qp->n) || (!is_unix && qp->n)) {
380         ret = -EINVAL;
381         goto out;
382     }
383 
384     if (is_unix) {
385         if (uri->server || uri->port) {
386             ret = -EINVAL;
387             goto out;
388         }
389         if (strcmp(qp->p[0].name, "socket")) {
390             ret = -EINVAL;
391             goto out;
392         }
393         gsconf->u.q_unix.path = g_strdup(qp->p[0].value);
394     } else {
395         gsconf->u.inet.host = g_strdup(uri->server ? uri->server : "localhost");
396         if (uri->port) {
397             gsconf->u.inet.port = g_strdup_printf("%d", uri->port);
398         } else {
399             gsconf->u.inet.port = g_strdup_printf("%d", GLUSTER_DEFAULT_PORT);
400         }
401     }
402 
403 out:
404     if (qp) {
405         query_params_free(qp);
406     }
407     uri_free(uri);
408     return ret;
409 }
410 
411 static struct glfs *qemu_gluster_glfs_init(BlockdevOptionsGluster *gconf,
412                                            Error **errp)
413 {
414     struct glfs *glfs;
415     int ret;
416     int old_errno;
417     SocketAddressList *server;
418     unsigned long long port;
419 
420     glfs = glfs_find_preopened(gconf->volume);
421     if (glfs) {
422         return glfs;
423     }
424 
425     glfs = glfs_new(gconf->volume);
426     if (!glfs) {
427         goto out;
428     }
429 
430     glfs_set_preopened(gconf->volume, glfs);
431 
432     for (server = gconf->server; server; server = server->next) {
433         switch (server->value->type) {
434         case SOCKET_ADDRESS_TYPE_UNIX:
435             ret = glfs_set_volfile_server(glfs, "unix",
436                                    server->value->u.q_unix.path, 0);
437             break;
438         case SOCKET_ADDRESS_TYPE_INET:
439             if (parse_uint_full(server->value->u.inet.port, &port, 10) < 0 ||
440                 port > 65535) {
441                 error_setg(errp, "'%s' is not a valid port number",
442                            server->value->u.inet.port);
443                 errno = EINVAL;
444                 goto out;
445             }
446             ret = glfs_set_volfile_server(glfs, "tcp",
447                                    server->value->u.inet.host,
448                                    (int)port);
449             break;
450         case SOCKET_ADDRESS_TYPE_VSOCK:
451         case SOCKET_ADDRESS_TYPE_FD:
452         default:
453             abort();
454         }
455 
456         if (ret < 0) {
457             goto out;
458         }
459     }
460 
461     ret = glfs_set_logging(glfs, gconf->logfile, gconf->debug);
462     if (ret < 0) {
463         goto out;
464     }
465 
466     ret = glfs_init(glfs);
467     if (ret) {
468         error_setg(errp, "Gluster connection for volume %s, path %s failed"
469                          " to connect", gconf->volume, gconf->path);
470         for (server = gconf->server; server; server = server->next) {
471             if (server->value->type  == SOCKET_ADDRESS_TYPE_UNIX) {
472                 error_append_hint(errp, "hint: failed on socket %s ",
473                                   server->value->u.q_unix.path);
474             } else {
475                 error_append_hint(errp, "hint: failed on host %s and port %s ",
476                                   server->value->u.inet.host,
477                                   server->value->u.inet.port);
478             }
479         }
480 
481         error_append_hint(errp, "Please refer to gluster logs for more info\n");
482 
483         /* glfs_init sometimes doesn't set errno although docs suggest that */
484         if (errno == 0) {
485             errno = EINVAL;
486         }
487 
488         goto out;
489     }
490     return glfs;
491 
492 out:
493     if (glfs) {
494         old_errno = errno;
495         glfs_clear_preopened(glfs);
496         errno = old_errno;
497     }
498     return NULL;
499 }
500 
501 /*
502  * Convert the json formatted command line into qapi.
503 */
504 static int qemu_gluster_parse_json(BlockdevOptionsGluster *gconf,
505                                   QDict *options, Error **errp)
506 {
507     QemuOpts *opts;
508     SocketAddress *gsconf = NULL;
509     SocketAddressList *curr = NULL;
510     QDict *backing_options = NULL;
511     Error *local_err = NULL;
512     char *str = NULL;
513     const char *ptr;
514     int i, type, num_servers;
515 
516     /* create opts info from runtime_json_opts list */
517     opts = qemu_opts_create(&runtime_json_opts, NULL, 0, &error_abort);
518     qemu_opts_absorb_qdict(opts, options, &local_err);
519     if (local_err) {
520         goto out;
521     }
522 
523     num_servers = qdict_array_entries(options, GLUSTER_OPT_SERVER_PATTERN);
524     if (num_servers < 1) {
525         error_setg(&local_err, QERR_MISSING_PARAMETER, "server");
526         goto out;
527     }
528 
529     ptr = qemu_opt_get(opts, GLUSTER_OPT_VOLUME);
530     if (!ptr) {
531         error_setg(&local_err, QERR_MISSING_PARAMETER, GLUSTER_OPT_VOLUME);
532         goto out;
533     }
534     gconf->volume = g_strdup(ptr);
535 
536     ptr = qemu_opt_get(opts, GLUSTER_OPT_PATH);
537     if (!ptr) {
538         error_setg(&local_err, QERR_MISSING_PARAMETER, GLUSTER_OPT_PATH);
539         goto out;
540     }
541     gconf->path = g_strdup(ptr);
542     qemu_opts_del(opts);
543 
544     for (i = 0; i < num_servers; i++) {
545         str = g_strdup_printf(GLUSTER_OPT_SERVER_PATTERN"%d.", i);
546         qdict_extract_subqdict(options, &backing_options, str);
547 
548         /* create opts info from runtime_type_opts list */
549         opts = qemu_opts_create(&runtime_type_opts, NULL, 0, &error_abort);
550         qemu_opts_absorb_qdict(opts, backing_options, &local_err);
551         if (local_err) {
552             goto out;
553         }
554 
555         ptr = qemu_opt_get(opts, GLUSTER_OPT_TYPE);
556         if (!ptr) {
557             error_setg(&local_err, QERR_MISSING_PARAMETER, GLUSTER_OPT_TYPE);
558             error_append_hint(&local_err, GERR_INDEX_HINT, i);
559             goto out;
560 
561         }
562         gsconf = g_new0(SocketAddress, 1);
563         if (!strcmp(ptr, "tcp")) {
564             ptr = "inet";       /* accept legacy "tcp" */
565         }
566         type = qapi_enum_parse(&SocketAddressType_lookup, ptr, -1, NULL);
567         if (type != SOCKET_ADDRESS_TYPE_INET
568             && type != SOCKET_ADDRESS_TYPE_UNIX) {
569             error_setg(&local_err,
570                        "Parameter '%s' may be 'inet' or 'unix'",
571                        GLUSTER_OPT_TYPE);
572             error_append_hint(&local_err, GERR_INDEX_HINT, i);
573             goto out;
574         }
575         gsconf->type = type;
576         qemu_opts_del(opts);
577 
578         if (gsconf->type == SOCKET_ADDRESS_TYPE_INET) {
579             /* create opts info from runtime_inet_opts list */
580             opts = qemu_opts_create(&runtime_inet_opts, NULL, 0, &error_abort);
581             qemu_opts_absorb_qdict(opts, backing_options, &local_err);
582             if (local_err) {
583                 goto out;
584             }
585 
586             ptr = qemu_opt_get(opts, GLUSTER_OPT_HOST);
587             if (!ptr) {
588                 error_setg(&local_err, QERR_MISSING_PARAMETER,
589                            GLUSTER_OPT_HOST);
590                 error_append_hint(&local_err, GERR_INDEX_HINT, i);
591                 goto out;
592             }
593             gsconf->u.inet.host = g_strdup(ptr);
594             ptr = qemu_opt_get(opts, GLUSTER_OPT_PORT);
595             if (!ptr) {
596                 error_setg(&local_err, QERR_MISSING_PARAMETER,
597                            GLUSTER_OPT_PORT);
598                 error_append_hint(&local_err, GERR_INDEX_HINT, i);
599                 goto out;
600             }
601             gsconf->u.inet.port = g_strdup(ptr);
602 
603             /* defend for unsupported fields in InetSocketAddress,
604              * i.e. @ipv4, @ipv6  and @to
605              */
606             ptr = qemu_opt_get(opts, GLUSTER_OPT_TO);
607             if (ptr) {
608                 gsconf->u.inet.has_to = true;
609             }
610             ptr = qemu_opt_get(opts, GLUSTER_OPT_IPV4);
611             if (ptr) {
612                 gsconf->u.inet.has_ipv4 = true;
613             }
614             ptr = qemu_opt_get(opts, GLUSTER_OPT_IPV6);
615             if (ptr) {
616                 gsconf->u.inet.has_ipv6 = true;
617             }
618             if (gsconf->u.inet.has_to) {
619                 error_setg(&local_err, "Parameter 'to' not supported");
620                 goto out;
621             }
622             if (gsconf->u.inet.has_ipv4 || gsconf->u.inet.has_ipv6) {
623                 error_setg(&local_err, "Parameters 'ipv4/ipv6' not supported");
624                 goto out;
625             }
626             qemu_opts_del(opts);
627         } else {
628             /* create opts info from runtime_unix_opts list */
629             opts = qemu_opts_create(&runtime_unix_opts, NULL, 0, &error_abort);
630             qemu_opts_absorb_qdict(opts, backing_options, &local_err);
631             if (local_err) {
632                 goto out;
633             }
634 
635             ptr = qemu_opt_get(opts, GLUSTER_OPT_PATH);
636             if (!ptr) {
637                 ptr = qemu_opt_get(opts, GLUSTER_OPT_SOCKET);
638             } else if (qemu_opt_get(opts, GLUSTER_OPT_SOCKET)) {
639                 error_setg(&local_err,
640                            "Conflicting parameters 'path' and 'socket'");
641                 error_append_hint(&local_err, GERR_INDEX_HINT, i);
642                 goto out;
643             }
644             if (!ptr) {
645                 error_setg(&local_err, QERR_MISSING_PARAMETER,
646                            GLUSTER_OPT_PATH);
647                 error_append_hint(&local_err, GERR_INDEX_HINT, i);
648                 goto out;
649             }
650             gsconf->u.q_unix.path = g_strdup(ptr);
651             qemu_opts_del(opts);
652         }
653 
654         if (gconf->server == NULL) {
655             gconf->server = g_new0(SocketAddressList, 1);
656             gconf->server->value = gsconf;
657             curr = gconf->server;
658         } else {
659             curr->next = g_new0(SocketAddressList, 1);
660             curr->next->value = gsconf;
661             curr = curr->next;
662         }
663         gsconf = NULL;
664 
665         qobject_unref(backing_options);
666         backing_options = NULL;
667         g_free(str);
668         str = NULL;
669     }
670 
671     return 0;
672 
673 out:
674     error_propagate(errp, local_err);
675     qapi_free_SocketAddress(gsconf);
676     qemu_opts_del(opts);
677     g_free(str);
678     qobject_unref(backing_options);
679     errno = EINVAL;
680     return -errno;
681 }
682 
683 /* Converts options given in @filename and the @options QDict into the QAPI
684  * object @gconf. */
685 static int qemu_gluster_parse(BlockdevOptionsGluster *gconf,
686                               const char *filename,
687                               QDict *options, Error **errp)
688 {
689     int ret;
690     if (filename) {
691         ret = qemu_gluster_parse_uri(gconf, filename);
692         if (ret < 0) {
693             error_setg(errp, "invalid URI %s", filename);
694             error_append_hint(errp, "Usage: file=gluster[+transport]://"
695                                     "[host[:port]]volume/path[?socket=...]"
696                                     "[,file.debug=N]"
697                                     "[,file.logfile=/path/filename.log]\n");
698             return ret;
699         }
700     } else {
701         ret = qemu_gluster_parse_json(gconf, options, errp);
702         if (ret < 0) {
703             error_append_hint(errp, "Usage: "
704                              "-drive driver=qcow2,file.driver=gluster,"
705                              "file.volume=testvol,file.path=/path/a.qcow2"
706                              "[,file.debug=9]"
707                              "[,file.logfile=/path/filename.log],"
708                              "file.server.0.type=inet,"
709                              "file.server.0.host=1.2.3.4,"
710                              "file.server.0.port=24007,"
711                              "file.server.1.transport=unix,"
712                              "file.server.1.path=/var/run/glusterd.socket ..."
713                              "\n");
714             return ret;
715         }
716     }
717 
718     return 0;
719 }
720 
721 static struct glfs *qemu_gluster_init(BlockdevOptionsGluster *gconf,
722                                       const char *filename,
723                                       QDict *options, Error **errp)
724 {
725     int ret;
726 
727     ret = qemu_gluster_parse(gconf, filename, options, errp);
728     if (ret < 0) {
729         errno = -ret;
730         return NULL;
731     }
732 
733     return qemu_gluster_glfs_init(gconf, errp);
734 }
735 
736 /*
737  * AIO callback routine called from GlusterFS thread.
738  */
739 static void gluster_finish_aiocb(struct glfs_fd *fd, ssize_t ret,
740 #ifdef CONFIG_GLUSTERFS_IOCB_HAS_STAT
741                                  struct glfs_stat *pre, struct glfs_stat *post,
742 #endif
743                                  void *arg)
744 {
745     GlusterAIOCB *acb = (GlusterAIOCB *)arg;
746 
747     if (!ret || ret == acb->size) {
748         acb->ret = 0; /* Success */
749     } else if (ret < 0) {
750         acb->ret = -errno; /* Read/Write failed */
751     } else {
752         acb->ret = -EIO; /* Partial read/write - fail it */
753     }
754 
755     aio_co_schedule(acb->aio_context, acb->coroutine);
756 }
757 
758 static void qemu_gluster_parse_flags(int bdrv_flags, int *open_flags)
759 {
760     assert(open_flags != NULL);
761 
762     *open_flags |= O_BINARY;
763 
764     if (bdrv_flags & BDRV_O_RDWR) {
765         *open_flags |= O_RDWR;
766     } else {
767         *open_flags |= O_RDONLY;
768     }
769 
770     if ((bdrv_flags & BDRV_O_NOCACHE)) {
771         *open_flags |= O_DIRECT;
772     }
773 }
774 
775 /*
776  * Do SEEK_DATA/HOLE to detect if it is functional. Older broken versions of
777  * gfapi incorrectly return the current offset when SEEK_DATA/HOLE is used.
778  * - Corrected versions return -1 and set errno to EINVAL.
779  * - Versions that support SEEK_DATA/HOLE correctly, will return -1 and set
780  *   errno to ENXIO when SEEK_DATA is called with a position of EOF.
781  */
782 static bool qemu_gluster_test_seek(struct glfs_fd *fd)
783 {
784     off_t ret = 0;
785 
786 #if defined SEEK_HOLE && defined SEEK_DATA
787     off_t eof;
788 
789     eof = glfs_lseek(fd, 0, SEEK_END);
790     if (eof < 0) {
791         /* this should never occur */
792         return false;
793     }
794 
795     /* this should always fail with ENXIO if SEEK_DATA is supported */
796     ret = glfs_lseek(fd, eof, SEEK_DATA);
797 #endif
798 
799     return (ret < 0) && (errno == ENXIO);
800 }
801 
802 static int qemu_gluster_open(BlockDriverState *bs,  QDict *options,
803                              int bdrv_flags, Error **errp)
804 {
805     BDRVGlusterState *s = bs->opaque;
806     int open_flags = 0;
807     int ret = 0;
808     BlockdevOptionsGluster *gconf = NULL;
809     QemuOpts *opts;
810     Error *local_err = NULL;
811     const char *filename, *logfile;
812 
813     opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
814     qemu_opts_absorb_qdict(opts, options, &local_err);
815     if (local_err) {
816         error_propagate(errp, local_err);
817         ret = -EINVAL;
818         goto out;
819     }
820 
821     filename = qemu_opt_get(opts, GLUSTER_OPT_FILENAME);
822 
823     s->debug = qemu_opt_get_number(opts, GLUSTER_OPT_DEBUG,
824                                    GLUSTER_DEBUG_DEFAULT);
825     if (s->debug < 0) {
826         s->debug = 0;
827     } else if (s->debug > GLUSTER_DEBUG_MAX) {
828         s->debug = GLUSTER_DEBUG_MAX;
829     }
830 
831     gconf = g_new0(BlockdevOptionsGluster, 1);
832     gconf->debug = s->debug;
833     gconf->has_debug = true;
834 
835     logfile = qemu_opt_get(opts, GLUSTER_OPT_LOGFILE);
836     s->logfile = g_strdup(logfile ? logfile : GLUSTER_LOGFILE_DEFAULT);
837 
838     gconf->logfile = g_strdup(s->logfile);
839     gconf->has_logfile = true;
840 
841     s->glfs = qemu_gluster_init(gconf, filename, options, errp);
842     if (!s->glfs) {
843         ret = -errno;
844         goto out;
845     }
846 
847 #ifdef CONFIG_GLUSTERFS_XLATOR_OPT
848     /* Without this, if fsync fails for a recoverable reason (for instance,
849      * ENOSPC), gluster will dump its cache, preventing retries.  This means
850      * almost certain data loss.  Not all gluster versions support the
851      * 'resync-failed-syncs-after-fsync' key value, but there is no way to
852      * discover during runtime if it is supported (this api returns success for
853      * unknown key/value pairs) */
854     ret = glfs_set_xlator_option(s->glfs, "*-write-behind",
855                                           "resync-failed-syncs-after-fsync",
856                                           "on");
857     if (ret < 0) {
858         error_setg_errno(errp, errno, "Unable to set xlator key/value pair");
859         ret = -errno;
860         goto out;
861     }
862 #endif
863 
864     qemu_gluster_parse_flags(bdrv_flags, &open_flags);
865 
866     s->fd = glfs_open(s->glfs, gconf->path, open_flags);
867     ret = s->fd ? 0 : -errno;
868 
869     if (ret == -EACCES || ret == -EROFS) {
870         /* Try to degrade to read-only, but if it doesn't work, still use the
871          * normal error message. */
872         if (bdrv_apply_auto_read_only(bs, NULL, NULL) == 0) {
873             open_flags = (open_flags & ~O_RDWR) | O_RDONLY;
874             s->fd = glfs_open(s->glfs, gconf->path, open_flags);
875             ret = s->fd ? 0 : -errno;
876         }
877     }
878 
879     s->supports_seek_data = qemu_gluster_test_seek(s->fd);
880 
881 out:
882     qemu_opts_del(opts);
883     qapi_free_BlockdevOptionsGluster(gconf);
884     if (!ret) {
885         return ret;
886     }
887     g_free(s->logfile);
888     if (s->fd) {
889         glfs_close(s->fd);
890     }
891 
892     glfs_clear_preopened(s->glfs);
893 
894     return ret;
895 }
896 
897 static void qemu_gluster_refresh_limits(BlockDriverState *bs, Error **errp)
898 {
899     bs->bl.max_transfer = GLUSTER_MAX_TRANSFER;
900 }
901 
902 static int qemu_gluster_reopen_prepare(BDRVReopenState *state,
903                                        BlockReopenQueue *queue, Error **errp)
904 {
905     int ret = 0;
906     BDRVGlusterState *s;
907     BDRVGlusterReopenState *reop_s;
908     BlockdevOptionsGluster *gconf;
909     int open_flags = 0;
910 
911     assert(state != NULL);
912     assert(state->bs != NULL);
913 
914     s = state->bs->opaque;
915 
916     state->opaque = g_new0(BDRVGlusterReopenState, 1);
917     reop_s = state->opaque;
918 
919     qemu_gluster_parse_flags(state->flags, &open_flags);
920 
921     gconf = g_new0(BlockdevOptionsGluster, 1);
922     gconf->debug = s->debug;
923     gconf->has_debug = true;
924     gconf->logfile = g_strdup(s->logfile);
925     gconf->has_logfile = true;
926     reop_s->glfs = qemu_gluster_init(gconf, state->bs->filename, NULL, errp);
927     if (reop_s->glfs == NULL) {
928         ret = -errno;
929         goto exit;
930     }
931 
932 #ifdef CONFIG_GLUSTERFS_XLATOR_OPT
933     ret = glfs_set_xlator_option(reop_s->glfs, "*-write-behind",
934                                  "resync-failed-syncs-after-fsync", "on");
935     if (ret < 0) {
936         error_setg_errno(errp, errno, "Unable to set xlator key/value pair");
937         ret = -errno;
938         goto exit;
939     }
940 #endif
941 
942     reop_s->fd = glfs_open(reop_s->glfs, gconf->path, open_flags);
943     if (reop_s->fd == NULL) {
944         /* reops->glfs will be cleaned up in _abort */
945         ret = -errno;
946         goto exit;
947     }
948 
949 exit:
950     /* state->opaque will be freed in either the _abort or _commit */
951     qapi_free_BlockdevOptionsGluster(gconf);
952     return ret;
953 }
954 
955 static void qemu_gluster_reopen_commit(BDRVReopenState *state)
956 {
957     BDRVGlusterReopenState *reop_s = state->opaque;
958     BDRVGlusterState *s = state->bs->opaque;
959 
960 
961     /* close the old */
962     if (s->fd) {
963         glfs_close(s->fd);
964     }
965 
966     glfs_clear_preopened(s->glfs);
967 
968     /* use the newly opened image / connection */
969     s->fd         = reop_s->fd;
970     s->glfs       = reop_s->glfs;
971 
972     g_free(state->opaque);
973     state->opaque = NULL;
974 
975     return;
976 }
977 
978 
979 static void qemu_gluster_reopen_abort(BDRVReopenState *state)
980 {
981     BDRVGlusterReopenState *reop_s = state->opaque;
982 
983     if (reop_s == NULL) {
984         return;
985     }
986 
987     if (reop_s->fd) {
988         glfs_close(reop_s->fd);
989     }
990 
991     glfs_clear_preopened(reop_s->glfs);
992 
993     g_free(state->opaque);
994     state->opaque = NULL;
995 
996     return;
997 }
998 
999 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1000 static coroutine_fn int qemu_gluster_co_pwrite_zeroes(BlockDriverState *bs,
1001                                                       int64_t offset,
1002                                                       int size,
1003                                                       BdrvRequestFlags flags)
1004 {
1005     int ret;
1006     GlusterAIOCB acb;
1007     BDRVGlusterState *s = bs->opaque;
1008 
1009     acb.size = size;
1010     acb.ret = 0;
1011     acb.coroutine = qemu_coroutine_self();
1012     acb.aio_context = bdrv_get_aio_context(bs);
1013 
1014     ret = glfs_zerofill_async(s->fd, offset, size, gluster_finish_aiocb, &acb);
1015     if (ret < 0) {
1016         return -errno;
1017     }
1018 
1019     qemu_coroutine_yield();
1020     return acb.ret;
1021 }
1022 #endif
1023 
1024 static int qemu_gluster_do_truncate(struct glfs_fd *fd, int64_t offset,
1025                                     PreallocMode prealloc, Error **errp)
1026 {
1027     int64_t current_length;
1028 
1029     current_length = glfs_lseek(fd, 0, SEEK_END);
1030     if (current_length < 0) {
1031         error_setg_errno(errp, errno, "Failed to determine current size");
1032         return -errno;
1033     }
1034 
1035     if (current_length > offset && prealloc != PREALLOC_MODE_OFF) {
1036         error_setg(errp, "Cannot use preallocation for shrinking files");
1037         return -ENOTSUP;
1038     }
1039 
1040     if (current_length == offset) {
1041         return 0;
1042     }
1043 
1044     switch (prealloc) {
1045 #ifdef CONFIG_GLUSTERFS_FALLOCATE
1046     case PREALLOC_MODE_FALLOC:
1047         if (glfs_fallocate(fd, 0, current_length, offset - current_length)) {
1048             error_setg_errno(errp, errno, "Could not preallocate data");
1049             return -errno;
1050         }
1051         break;
1052 #endif /* CONFIG_GLUSTERFS_FALLOCATE */
1053 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1054     case PREALLOC_MODE_FULL:
1055         if (glfs_ftruncate(fd, offset)) {
1056             error_setg_errno(errp, errno, "Could not resize file");
1057             return -errno;
1058         }
1059         if (glfs_zerofill(fd, current_length, offset - current_length)) {
1060             error_setg_errno(errp, errno, "Could not zerofill the new area");
1061             return -errno;
1062         }
1063         break;
1064 #endif /* CONFIG_GLUSTERFS_ZEROFILL */
1065     case PREALLOC_MODE_OFF:
1066         if (glfs_ftruncate(fd, offset)) {
1067             error_setg_errno(errp, errno, "Could not resize file");
1068             return -errno;
1069         }
1070         break;
1071     default:
1072         error_setg(errp, "Unsupported preallocation mode: %s",
1073                    PreallocMode_str(prealloc));
1074         return -EINVAL;
1075     }
1076 
1077     return 0;
1078 }
1079 
1080 static int qemu_gluster_co_create(BlockdevCreateOptions *options,
1081                                   Error **errp)
1082 {
1083     BlockdevCreateOptionsGluster *opts = &options->u.gluster;
1084     struct glfs *glfs;
1085     struct glfs_fd *fd = NULL;
1086     int ret = 0;
1087 
1088     assert(options->driver == BLOCKDEV_DRIVER_GLUSTER);
1089 
1090     glfs = qemu_gluster_glfs_init(opts->location, errp);
1091     if (!glfs) {
1092         ret = -errno;
1093         goto out;
1094     }
1095 
1096     fd = glfs_creat(glfs, opts->location->path,
1097                     O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, S_IRUSR | S_IWUSR);
1098     if (!fd) {
1099         ret = -errno;
1100         goto out;
1101     }
1102 
1103     ret = qemu_gluster_do_truncate(fd, opts->size, opts->preallocation, errp);
1104 
1105 out:
1106     if (fd) {
1107         if (glfs_close(fd) != 0 && ret == 0) {
1108             ret = -errno;
1109         }
1110     }
1111     glfs_clear_preopened(glfs);
1112     return ret;
1113 }
1114 
1115 static int coroutine_fn qemu_gluster_co_create_opts(const char *filename,
1116                                                     QemuOpts *opts,
1117                                                     Error **errp)
1118 {
1119     BlockdevCreateOptions *options;
1120     BlockdevCreateOptionsGluster *gopts;
1121     BlockdevOptionsGluster *gconf;
1122     char *tmp = NULL;
1123     Error *local_err = NULL;
1124     int ret;
1125 
1126     options = g_new0(BlockdevCreateOptions, 1);
1127     options->driver = BLOCKDEV_DRIVER_GLUSTER;
1128     gopts = &options->u.gluster;
1129 
1130     gconf = g_new0(BlockdevOptionsGluster, 1);
1131     gopts->location = gconf;
1132 
1133     gopts->size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
1134                            BDRV_SECTOR_SIZE);
1135 
1136     tmp = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
1137     gopts->preallocation = qapi_enum_parse(&PreallocMode_lookup, tmp,
1138                                            PREALLOC_MODE_OFF, &local_err);
1139     g_free(tmp);
1140     if (local_err) {
1141         error_propagate(errp, local_err);
1142         ret = -EINVAL;
1143         goto fail;
1144     }
1145 
1146     gconf->debug = qemu_opt_get_number_del(opts, GLUSTER_OPT_DEBUG,
1147                                            GLUSTER_DEBUG_DEFAULT);
1148     if (gconf->debug < 0) {
1149         gconf->debug = 0;
1150     } else if (gconf->debug > GLUSTER_DEBUG_MAX) {
1151         gconf->debug = GLUSTER_DEBUG_MAX;
1152     }
1153     gconf->has_debug = true;
1154 
1155     gconf->logfile = qemu_opt_get_del(opts, GLUSTER_OPT_LOGFILE);
1156     if (!gconf->logfile) {
1157         gconf->logfile = g_strdup(GLUSTER_LOGFILE_DEFAULT);
1158     }
1159     gconf->has_logfile = true;
1160 
1161     ret = qemu_gluster_parse(gconf, filename, NULL, errp);
1162     if (ret < 0) {
1163         goto fail;
1164     }
1165 
1166     ret = qemu_gluster_co_create(options, errp);
1167     if (ret < 0) {
1168         goto fail;
1169     }
1170 
1171     ret = 0;
1172 fail:
1173     qapi_free_BlockdevCreateOptions(options);
1174     return ret;
1175 }
1176 
1177 static coroutine_fn int qemu_gluster_co_rw(BlockDriverState *bs,
1178                                            int64_t sector_num, int nb_sectors,
1179                                            QEMUIOVector *qiov, int write)
1180 {
1181     int ret;
1182     GlusterAIOCB acb;
1183     BDRVGlusterState *s = bs->opaque;
1184     size_t size = nb_sectors * BDRV_SECTOR_SIZE;
1185     off_t offset = sector_num * BDRV_SECTOR_SIZE;
1186 
1187     acb.size = size;
1188     acb.ret = 0;
1189     acb.coroutine = qemu_coroutine_self();
1190     acb.aio_context = bdrv_get_aio_context(bs);
1191 
1192     if (write) {
1193         ret = glfs_pwritev_async(s->fd, qiov->iov, qiov->niov, offset, 0,
1194                                  gluster_finish_aiocb, &acb);
1195     } else {
1196         ret = glfs_preadv_async(s->fd, qiov->iov, qiov->niov, offset, 0,
1197                                 gluster_finish_aiocb, &acb);
1198     }
1199 
1200     if (ret < 0) {
1201         return -errno;
1202     }
1203 
1204     qemu_coroutine_yield();
1205     return acb.ret;
1206 }
1207 
1208 static coroutine_fn int qemu_gluster_co_truncate(BlockDriverState *bs,
1209                                                  int64_t offset,
1210                                                  PreallocMode prealloc,
1211                                                  Error **errp)
1212 {
1213     BDRVGlusterState *s = bs->opaque;
1214     return qemu_gluster_do_truncate(s->fd, offset, prealloc, errp);
1215 }
1216 
1217 static coroutine_fn int qemu_gluster_co_readv(BlockDriverState *bs,
1218                                               int64_t sector_num,
1219                                               int nb_sectors,
1220                                               QEMUIOVector *qiov)
1221 {
1222     return qemu_gluster_co_rw(bs, sector_num, nb_sectors, qiov, 0);
1223 }
1224 
1225 static coroutine_fn int qemu_gluster_co_writev(BlockDriverState *bs,
1226                                                int64_t sector_num,
1227                                                int nb_sectors,
1228                                                QEMUIOVector *qiov,
1229                                                int flags)
1230 {
1231     assert(!flags);
1232     return qemu_gluster_co_rw(bs, sector_num, nb_sectors, qiov, 1);
1233 }
1234 
1235 static void qemu_gluster_close(BlockDriverState *bs)
1236 {
1237     BDRVGlusterState *s = bs->opaque;
1238 
1239     g_free(s->logfile);
1240     if (s->fd) {
1241         glfs_close(s->fd);
1242         s->fd = NULL;
1243     }
1244     glfs_clear_preopened(s->glfs);
1245 }
1246 
1247 static coroutine_fn int qemu_gluster_co_flush_to_disk(BlockDriverState *bs)
1248 {
1249     int ret;
1250     GlusterAIOCB acb;
1251     BDRVGlusterState *s = bs->opaque;
1252 
1253     acb.size = 0;
1254     acb.ret = 0;
1255     acb.coroutine = qemu_coroutine_self();
1256     acb.aio_context = bdrv_get_aio_context(bs);
1257 
1258     ret = glfs_fsync_async(s->fd, gluster_finish_aiocb, &acb);
1259     if (ret < 0) {
1260         ret = -errno;
1261         goto error;
1262     }
1263 
1264     qemu_coroutine_yield();
1265     if (acb.ret < 0) {
1266         ret = acb.ret;
1267         goto error;
1268     }
1269 
1270     return acb.ret;
1271 
1272 error:
1273     /* Some versions of Gluster (3.5.6 -> 3.5.8?) will not retain its cache
1274      * after a fsync failure, so we have no way of allowing the guest to safely
1275      * continue.  Gluster versions prior to 3.5.6 don't retain the cache
1276      * either, but will invalidate the fd on error, so this is again our only
1277      * option.
1278      *
1279      * The 'resync-failed-syncs-after-fsync' xlator option for the
1280      * write-behind cache will cause later gluster versions to retain its
1281      * cache after error, so long as the fd remains open.  However, we
1282      * currently have no way of knowing if this option is supported.
1283      *
1284      * TODO: Once gluster provides a way for us to determine if the option
1285      * is supported, bypass the closure and setting drv to NULL.  */
1286     qemu_gluster_close(bs);
1287     bs->drv = NULL;
1288     return ret;
1289 }
1290 
1291 #ifdef CONFIG_GLUSTERFS_DISCARD
1292 static coroutine_fn int qemu_gluster_co_pdiscard(BlockDriverState *bs,
1293                                                  int64_t offset, int size)
1294 {
1295     int ret;
1296     GlusterAIOCB acb;
1297     BDRVGlusterState *s = bs->opaque;
1298 
1299     acb.size = 0;
1300     acb.ret = 0;
1301     acb.coroutine = qemu_coroutine_self();
1302     acb.aio_context = bdrv_get_aio_context(bs);
1303 
1304     ret = glfs_discard_async(s->fd, offset, size, gluster_finish_aiocb, &acb);
1305     if (ret < 0) {
1306         return -errno;
1307     }
1308 
1309     qemu_coroutine_yield();
1310     return acb.ret;
1311 }
1312 #endif
1313 
1314 static int64_t qemu_gluster_getlength(BlockDriverState *bs)
1315 {
1316     BDRVGlusterState *s = bs->opaque;
1317     int64_t ret;
1318 
1319     ret = glfs_lseek(s->fd, 0, SEEK_END);
1320     if (ret < 0) {
1321         return -errno;
1322     } else {
1323         return ret;
1324     }
1325 }
1326 
1327 static int64_t qemu_gluster_allocated_file_size(BlockDriverState *bs)
1328 {
1329     BDRVGlusterState *s = bs->opaque;
1330     struct stat st;
1331     int ret;
1332 
1333     ret = glfs_fstat(s->fd, &st);
1334     if (ret < 0) {
1335         return -errno;
1336     } else {
1337         return st.st_blocks * 512;
1338     }
1339 }
1340 
1341 static int qemu_gluster_has_zero_init(BlockDriverState *bs)
1342 {
1343     /* GlusterFS volume could be backed by a block device */
1344     return 0;
1345 }
1346 
1347 /*
1348  * Find allocation range in @bs around offset @start.
1349  * May change underlying file descriptor's file offset.
1350  * If @start is not in a hole, store @start in @data, and the
1351  * beginning of the next hole in @hole, and return 0.
1352  * If @start is in a non-trailing hole, store @start in @hole and the
1353  * beginning of the next non-hole in @data, and return 0.
1354  * If @start is in a trailing hole or beyond EOF, return -ENXIO.
1355  * If we can't find out, return a negative errno other than -ENXIO.
1356  *
1357  * (Shamefully copied from file-posix.c, only minuscule adaptions.)
1358  */
1359 static int find_allocation(BlockDriverState *bs, off_t start,
1360                            off_t *data, off_t *hole)
1361 {
1362     BDRVGlusterState *s = bs->opaque;
1363 
1364     if (!s->supports_seek_data) {
1365         goto exit;
1366     }
1367 
1368 #if defined SEEK_HOLE && defined SEEK_DATA
1369     off_t offs;
1370 
1371     /*
1372      * SEEK_DATA cases:
1373      * D1. offs == start: start is in data
1374      * D2. offs > start: start is in a hole, next data at offs
1375      * D3. offs < 0, errno = ENXIO: either start is in a trailing hole
1376      *                              or start is beyond EOF
1377      *     If the latter happens, the file has been truncated behind
1378      *     our back since we opened it.  All bets are off then.
1379      *     Treating like a trailing hole is simplest.
1380      * D4. offs < 0, errno != ENXIO: we learned nothing
1381      */
1382     offs = glfs_lseek(s->fd, start, SEEK_DATA);
1383     if (offs < 0) {
1384         return -errno;          /* D3 or D4 */
1385     }
1386 
1387     if (offs < start) {
1388         /* This is not a valid return by lseek().  We are safe to just return
1389          * -EIO in this case, and we'll treat it like D4. Unfortunately some
1390          *  versions of gluster server will return offs < start, so an assert
1391          *  here will unnecessarily abort QEMU. */
1392         return -EIO;
1393     }
1394 
1395     if (offs > start) {
1396         /* D2: in hole, next data at offs */
1397         *hole = start;
1398         *data = offs;
1399         return 0;
1400     }
1401 
1402     /* D1: in data, end not yet known */
1403 
1404     /*
1405      * SEEK_HOLE cases:
1406      * H1. offs == start: start is in a hole
1407      *     If this happens here, a hole has been dug behind our back
1408      *     since the previous lseek().
1409      * H2. offs > start: either start is in data, next hole at offs,
1410      *                   or start is in trailing hole, EOF at offs
1411      *     Linux treats trailing holes like any other hole: offs ==
1412      *     start.  Solaris seeks to EOF instead: offs > start (blech).
1413      *     If that happens here, a hole has been dug behind our back
1414      *     since the previous lseek().
1415      * H3. offs < 0, errno = ENXIO: start is beyond EOF
1416      *     If this happens, the file has been truncated behind our
1417      *     back since we opened it.  Treat it like a trailing hole.
1418      * H4. offs < 0, errno != ENXIO: we learned nothing
1419      *     Pretend we know nothing at all, i.e. "forget" about D1.
1420      */
1421     offs = glfs_lseek(s->fd, start, SEEK_HOLE);
1422     if (offs < 0) {
1423         return -errno;          /* D1 and (H3 or H4) */
1424     }
1425 
1426     if (offs < start) {
1427         /* This is not a valid return by lseek().  We are safe to just return
1428          * -EIO in this case, and we'll treat it like H4. Unfortunately some
1429          *  versions of gluster server will return offs < start, so an assert
1430          *  here will unnecessarily abort QEMU. */
1431         return -EIO;
1432     }
1433 
1434     if (offs > start) {
1435         /*
1436          * D1 and H2: either in data, next hole at offs, or it was in
1437          * data but is now in a trailing hole.  In the latter case,
1438          * all bets are off.  Treating it as if it there was data all
1439          * the way to EOF is safe, so simply do that.
1440          */
1441         *data = start;
1442         *hole = offs;
1443         return 0;
1444     }
1445 
1446     /* D1 and H1 */
1447     return -EBUSY;
1448 #endif
1449 
1450 exit:
1451     return -ENOTSUP;
1452 }
1453 
1454 /*
1455  * Returns the allocation status of the specified offset.
1456  *
1457  * The block layer guarantees 'offset' and 'bytes' are within bounds.
1458  *
1459  * 'pnum' is set to the number of bytes (including and immediately following
1460  * the specified offset) that are known to be in the same
1461  * allocated/unallocated state.
1462  *
1463  * 'bytes' is the max value 'pnum' should be set to.
1464  *
1465  * (Based on raw_co_block_status() from file-posix.c.)
1466  */
1467 static int coroutine_fn qemu_gluster_co_block_status(BlockDriverState *bs,
1468                                                      bool want_zero,
1469                                                      int64_t offset,
1470                                                      int64_t bytes,
1471                                                      int64_t *pnum,
1472                                                      int64_t *map,
1473                                                      BlockDriverState **file)
1474 {
1475     BDRVGlusterState *s = bs->opaque;
1476     off_t data = 0, hole = 0;
1477     int ret = -EINVAL;
1478 
1479     if (!s->fd) {
1480         return ret;
1481     }
1482 
1483     if (!want_zero) {
1484         *pnum = bytes;
1485         *map = offset;
1486         *file = bs;
1487         return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID;
1488     }
1489 
1490     ret = find_allocation(bs, offset, &data, &hole);
1491     if (ret == -ENXIO) {
1492         /* Trailing hole */
1493         *pnum = bytes;
1494         ret = BDRV_BLOCK_ZERO;
1495     } else if (ret < 0) {
1496         /* No info available, so pretend there are no holes */
1497         *pnum = bytes;
1498         ret = BDRV_BLOCK_DATA;
1499     } else if (data == offset) {
1500         /* On a data extent, compute bytes to the end of the extent,
1501          * possibly including a partial sector at EOF. */
1502         *pnum = MIN(bytes, hole - offset);
1503         ret = BDRV_BLOCK_DATA;
1504     } else {
1505         /* On a hole, compute bytes to the beginning of the next extent.  */
1506         assert(hole == offset);
1507         *pnum = MIN(bytes, data - offset);
1508         ret = BDRV_BLOCK_ZERO;
1509     }
1510 
1511     *map = offset;
1512     *file = bs;
1513 
1514     return ret | BDRV_BLOCK_OFFSET_VALID;
1515 }
1516 
1517 
1518 static const char *const gluster_strong_open_opts[] = {
1519     GLUSTER_OPT_VOLUME,
1520     GLUSTER_OPT_PATH,
1521     GLUSTER_OPT_TYPE,
1522     GLUSTER_OPT_SERVER_PATTERN,
1523     GLUSTER_OPT_HOST,
1524     GLUSTER_OPT_PORT,
1525     GLUSTER_OPT_TO,
1526     GLUSTER_OPT_IPV4,
1527     GLUSTER_OPT_IPV6,
1528     GLUSTER_OPT_SOCKET,
1529 
1530     NULL
1531 };
1532 
1533 static BlockDriver bdrv_gluster = {
1534     .format_name                  = "gluster",
1535     .protocol_name                = "gluster",
1536     .instance_size                = sizeof(BDRVGlusterState),
1537     .bdrv_needs_filename          = false,
1538     .bdrv_file_open               = qemu_gluster_open,
1539     .bdrv_reopen_prepare          = qemu_gluster_reopen_prepare,
1540     .bdrv_reopen_commit           = qemu_gluster_reopen_commit,
1541     .bdrv_reopen_abort            = qemu_gluster_reopen_abort,
1542     .bdrv_close                   = qemu_gluster_close,
1543     .bdrv_co_create               = qemu_gluster_co_create,
1544     .bdrv_co_create_opts          = qemu_gluster_co_create_opts,
1545     .bdrv_getlength               = qemu_gluster_getlength,
1546     .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size,
1547     .bdrv_co_truncate             = qemu_gluster_co_truncate,
1548     .bdrv_co_readv                = qemu_gluster_co_readv,
1549     .bdrv_co_writev               = qemu_gluster_co_writev,
1550     .bdrv_co_flush_to_disk        = qemu_gluster_co_flush_to_disk,
1551     .bdrv_has_zero_init           = qemu_gluster_has_zero_init,
1552 #ifdef CONFIG_GLUSTERFS_DISCARD
1553     .bdrv_co_pdiscard             = qemu_gluster_co_pdiscard,
1554 #endif
1555 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1556     .bdrv_co_pwrite_zeroes        = qemu_gluster_co_pwrite_zeroes,
1557 #endif
1558     .bdrv_co_block_status         = qemu_gluster_co_block_status,
1559     .bdrv_refresh_limits          = qemu_gluster_refresh_limits,
1560     .create_opts                  = &qemu_gluster_create_opts,
1561     .strong_runtime_opts          = gluster_strong_open_opts,
1562 };
1563 
1564 static BlockDriver bdrv_gluster_tcp = {
1565     .format_name                  = "gluster",
1566     .protocol_name                = "gluster+tcp",
1567     .instance_size                = sizeof(BDRVGlusterState),
1568     .bdrv_needs_filename          = false,
1569     .bdrv_file_open               = qemu_gluster_open,
1570     .bdrv_reopen_prepare          = qemu_gluster_reopen_prepare,
1571     .bdrv_reopen_commit           = qemu_gluster_reopen_commit,
1572     .bdrv_reopen_abort            = qemu_gluster_reopen_abort,
1573     .bdrv_close                   = qemu_gluster_close,
1574     .bdrv_co_create               = qemu_gluster_co_create,
1575     .bdrv_co_create_opts          = qemu_gluster_co_create_opts,
1576     .bdrv_getlength               = qemu_gluster_getlength,
1577     .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size,
1578     .bdrv_co_truncate             = qemu_gluster_co_truncate,
1579     .bdrv_co_readv                = qemu_gluster_co_readv,
1580     .bdrv_co_writev               = qemu_gluster_co_writev,
1581     .bdrv_co_flush_to_disk        = qemu_gluster_co_flush_to_disk,
1582     .bdrv_has_zero_init           = qemu_gluster_has_zero_init,
1583 #ifdef CONFIG_GLUSTERFS_DISCARD
1584     .bdrv_co_pdiscard             = qemu_gluster_co_pdiscard,
1585 #endif
1586 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1587     .bdrv_co_pwrite_zeroes        = qemu_gluster_co_pwrite_zeroes,
1588 #endif
1589     .bdrv_co_block_status         = qemu_gluster_co_block_status,
1590     .bdrv_refresh_limits          = qemu_gluster_refresh_limits,
1591     .create_opts                  = &qemu_gluster_create_opts,
1592     .strong_runtime_opts          = gluster_strong_open_opts,
1593 };
1594 
1595 static BlockDriver bdrv_gluster_unix = {
1596     .format_name                  = "gluster",
1597     .protocol_name                = "gluster+unix",
1598     .instance_size                = sizeof(BDRVGlusterState),
1599     .bdrv_needs_filename          = true,
1600     .bdrv_file_open               = qemu_gluster_open,
1601     .bdrv_reopen_prepare          = qemu_gluster_reopen_prepare,
1602     .bdrv_reopen_commit           = qemu_gluster_reopen_commit,
1603     .bdrv_reopen_abort            = qemu_gluster_reopen_abort,
1604     .bdrv_close                   = qemu_gluster_close,
1605     .bdrv_co_create               = qemu_gluster_co_create,
1606     .bdrv_co_create_opts          = qemu_gluster_co_create_opts,
1607     .bdrv_getlength               = qemu_gluster_getlength,
1608     .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size,
1609     .bdrv_co_truncate             = qemu_gluster_co_truncate,
1610     .bdrv_co_readv                = qemu_gluster_co_readv,
1611     .bdrv_co_writev               = qemu_gluster_co_writev,
1612     .bdrv_co_flush_to_disk        = qemu_gluster_co_flush_to_disk,
1613     .bdrv_has_zero_init           = qemu_gluster_has_zero_init,
1614 #ifdef CONFIG_GLUSTERFS_DISCARD
1615     .bdrv_co_pdiscard             = qemu_gluster_co_pdiscard,
1616 #endif
1617 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1618     .bdrv_co_pwrite_zeroes        = qemu_gluster_co_pwrite_zeroes,
1619 #endif
1620     .bdrv_co_block_status         = qemu_gluster_co_block_status,
1621     .bdrv_refresh_limits          = qemu_gluster_refresh_limits,
1622     .create_opts                  = &qemu_gluster_create_opts,
1623     .strong_runtime_opts          = gluster_strong_open_opts,
1624 };
1625 
1626 /* rdma is deprecated (actually never supported for volfile fetch).
1627  * Let's maintain it for the protocol compatibility, to make sure things
1628  * won't break immediately. For now, gluster+rdma will fall back to gluster+tcp
1629  * protocol with a warning.
1630  * TODO: remove gluster+rdma interface support
1631  */
1632 static BlockDriver bdrv_gluster_rdma = {
1633     .format_name                  = "gluster",
1634     .protocol_name                = "gluster+rdma",
1635     .instance_size                = sizeof(BDRVGlusterState),
1636     .bdrv_needs_filename          = true,
1637     .bdrv_file_open               = qemu_gluster_open,
1638     .bdrv_reopen_prepare          = qemu_gluster_reopen_prepare,
1639     .bdrv_reopen_commit           = qemu_gluster_reopen_commit,
1640     .bdrv_reopen_abort            = qemu_gluster_reopen_abort,
1641     .bdrv_close                   = qemu_gluster_close,
1642     .bdrv_co_create               = qemu_gluster_co_create,
1643     .bdrv_co_create_opts          = qemu_gluster_co_create_opts,
1644     .bdrv_getlength               = qemu_gluster_getlength,
1645     .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size,
1646     .bdrv_co_truncate             = qemu_gluster_co_truncate,
1647     .bdrv_co_readv                = qemu_gluster_co_readv,
1648     .bdrv_co_writev               = qemu_gluster_co_writev,
1649     .bdrv_co_flush_to_disk        = qemu_gluster_co_flush_to_disk,
1650     .bdrv_has_zero_init           = qemu_gluster_has_zero_init,
1651 #ifdef CONFIG_GLUSTERFS_DISCARD
1652     .bdrv_co_pdiscard             = qemu_gluster_co_pdiscard,
1653 #endif
1654 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1655     .bdrv_co_pwrite_zeroes        = qemu_gluster_co_pwrite_zeroes,
1656 #endif
1657     .bdrv_co_block_status         = qemu_gluster_co_block_status,
1658     .bdrv_refresh_limits          = qemu_gluster_refresh_limits,
1659     .create_opts                  = &qemu_gluster_create_opts,
1660     .strong_runtime_opts          = gluster_strong_open_opts,
1661 };
1662 
1663 static void bdrv_gluster_init(void)
1664 {
1665     bdrv_register(&bdrv_gluster_rdma);
1666     bdrv_register(&bdrv_gluster_unix);
1667     bdrv_register(&bdrv_gluster_tcp);
1668     bdrv_register(&bdrv_gluster);
1669 }
1670 
1671 block_init(bdrv_gluster_init);
1672