xref: /openbmc/qemu/block/gluster.c (revision 3c4b89c3)
1 /*
2  * GlusterFS backend for QEMU
3  *
4  * Copyright (C) 2012 Bharata B Rao <bharata@linux.vnet.ibm.com>
5  *
6  * This work is licensed under the terms of the GNU GPL, version 2 or later.
7  * See the COPYING file in the top-level directory.
8  *
9  */
10 
11 #include "qemu/osdep.h"
12 #include "qemu/units.h"
13 #include <glusterfs/api/glfs.h>
14 #include "block/block_int.h"
15 #include "block/qdict.h"
16 #include "qapi/error.h"
17 #include "qapi/qmp/qdict.h"
18 #include "qapi/qmp/qerror.h"
19 #include "qemu/uri.h"
20 #include "qemu/error-report.h"
21 #include "qemu/module.h"
22 #include "qemu/option.h"
23 #include "qemu/cutils.h"
24 
25 #ifdef CONFIG_GLUSTERFS_FTRUNCATE_HAS_STAT
26 # define glfs_ftruncate(fd, offset) glfs_ftruncate(fd, offset, NULL, NULL)
27 #endif
28 
29 #define GLUSTER_OPT_FILENAME        "filename"
30 #define GLUSTER_OPT_VOLUME          "volume"
31 #define GLUSTER_OPT_PATH            "path"
32 #define GLUSTER_OPT_TYPE            "type"
33 #define GLUSTER_OPT_SERVER_PATTERN  "server."
34 #define GLUSTER_OPT_HOST            "host"
35 #define GLUSTER_OPT_PORT            "port"
36 #define GLUSTER_OPT_TO              "to"
37 #define GLUSTER_OPT_IPV4            "ipv4"
38 #define GLUSTER_OPT_IPV6            "ipv6"
39 #define GLUSTER_OPT_SOCKET          "socket"
40 #define GLUSTER_OPT_DEBUG           "debug"
41 #define GLUSTER_DEFAULT_PORT        24007
42 #define GLUSTER_DEBUG_DEFAULT       4
43 #define GLUSTER_DEBUG_MAX           9
44 #define GLUSTER_OPT_LOGFILE         "logfile"
45 #define GLUSTER_LOGFILE_DEFAULT     "-" /* handled in libgfapi as /dev/stderr */
46 /*
47  * Several versions of GlusterFS (3.12? -> 6.0.1) fail when the transfer size
48  * is greater or equal to 1024 MiB, so we are limiting the transfer size to 512
49  * MiB to avoid this rare issue.
50  */
51 #define GLUSTER_MAX_TRANSFER        (512 * MiB)
52 
53 #define GERR_INDEX_HINT "hint: check in 'server' array index '%d'\n"
54 
55 typedef struct GlusterAIOCB {
56     int64_t size;
57     int ret;
58     Coroutine *coroutine;
59     AioContext *aio_context;
60 } GlusterAIOCB;
61 
62 typedef struct BDRVGlusterState {
63     struct glfs *glfs;
64     struct glfs_fd *fd;
65     char *logfile;
66     bool supports_seek_data;
67     int debug;
68 } BDRVGlusterState;
69 
70 typedef struct BDRVGlusterReopenState {
71     struct glfs *glfs;
72     struct glfs_fd *fd;
73 } BDRVGlusterReopenState;
74 
75 
76 typedef struct GlfsPreopened {
77     char *volume;
78     glfs_t *fs;
79     int ref;
80 } GlfsPreopened;
81 
82 typedef struct ListElement {
83     QLIST_ENTRY(ListElement) list;
84     GlfsPreopened saved;
85 } ListElement;
86 
87 static QLIST_HEAD(, ListElement) glfs_list;
88 
89 static QemuOptsList qemu_gluster_create_opts = {
90     .name = "qemu-gluster-create-opts",
91     .head = QTAILQ_HEAD_INITIALIZER(qemu_gluster_create_opts.head),
92     .desc = {
93         {
94             .name = BLOCK_OPT_SIZE,
95             .type = QEMU_OPT_SIZE,
96             .help = "Virtual disk size"
97         },
98         {
99             .name = BLOCK_OPT_PREALLOC,
100             .type = QEMU_OPT_STRING,
101             .help = "Preallocation mode (allowed values: off"
102 #ifdef CONFIG_GLUSTERFS_FALLOCATE
103                     ", falloc"
104 #endif
105 #ifdef CONFIG_GLUSTERFS_ZEROFILL
106                     ", full"
107 #endif
108                     ")"
109         },
110         {
111             .name = GLUSTER_OPT_DEBUG,
112             .type = QEMU_OPT_NUMBER,
113             .help = "Gluster log level, valid range is 0-9",
114         },
115         {
116             .name = GLUSTER_OPT_LOGFILE,
117             .type = QEMU_OPT_STRING,
118             .help = "Logfile path of libgfapi",
119         },
120         { /* end of list */ }
121     }
122 };
123 
124 static QemuOptsList runtime_opts = {
125     .name = "gluster",
126     .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
127     .desc = {
128         {
129             .name = GLUSTER_OPT_FILENAME,
130             .type = QEMU_OPT_STRING,
131             .help = "URL to the gluster image",
132         },
133         {
134             .name = GLUSTER_OPT_DEBUG,
135             .type = QEMU_OPT_NUMBER,
136             .help = "Gluster log level, valid range is 0-9",
137         },
138         {
139             .name = GLUSTER_OPT_LOGFILE,
140             .type = QEMU_OPT_STRING,
141             .help = "Logfile path of libgfapi",
142         },
143         { /* end of list */ }
144     },
145 };
146 
147 static QemuOptsList runtime_json_opts = {
148     .name = "gluster_json",
149     .head = QTAILQ_HEAD_INITIALIZER(runtime_json_opts.head),
150     .desc = {
151         {
152             .name = GLUSTER_OPT_VOLUME,
153             .type = QEMU_OPT_STRING,
154             .help = "name of gluster volume where VM image resides",
155         },
156         {
157             .name = GLUSTER_OPT_PATH,
158             .type = QEMU_OPT_STRING,
159             .help = "absolute path to image file in gluster volume",
160         },
161         {
162             .name = GLUSTER_OPT_DEBUG,
163             .type = QEMU_OPT_NUMBER,
164             .help = "Gluster log level, valid range is 0-9",
165         },
166         { /* end of list */ }
167     },
168 };
169 
170 static QemuOptsList runtime_type_opts = {
171     .name = "gluster_type",
172     .head = QTAILQ_HEAD_INITIALIZER(runtime_type_opts.head),
173     .desc = {
174         {
175             .name = GLUSTER_OPT_TYPE,
176             .type = QEMU_OPT_STRING,
177             .help = "inet|unix",
178         },
179         { /* end of list */ }
180     },
181 };
182 
183 static QemuOptsList runtime_unix_opts = {
184     .name = "gluster_unix",
185     .head = QTAILQ_HEAD_INITIALIZER(runtime_unix_opts.head),
186     .desc = {
187         {
188             .name = GLUSTER_OPT_SOCKET,
189             .type = QEMU_OPT_STRING,
190             .help = "socket file path (legacy)",
191         },
192         {
193             .name = GLUSTER_OPT_PATH,
194             .type = QEMU_OPT_STRING,
195             .help = "socket file path (QAPI)",
196         },
197         { /* end of list */ }
198     },
199 };
200 
201 static QemuOptsList runtime_inet_opts = {
202     .name = "gluster_inet",
203     .head = QTAILQ_HEAD_INITIALIZER(runtime_inet_opts.head),
204     .desc = {
205         {
206             .name = GLUSTER_OPT_TYPE,
207             .type = QEMU_OPT_STRING,
208             .help = "inet|unix",
209         },
210         {
211             .name = GLUSTER_OPT_HOST,
212             .type = QEMU_OPT_STRING,
213             .help = "host address (hostname/ipv4/ipv6 addresses)",
214         },
215         {
216             .name = GLUSTER_OPT_PORT,
217             .type = QEMU_OPT_STRING,
218             .help = "port number on which glusterd is listening (default 24007)",
219         },
220         {
221             .name = "to",
222             .type = QEMU_OPT_NUMBER,
223             .help = "max port number, not supported by gluster",
224         },
225         {
226             .name = "ipv4",
227             .type = QEMU_OPT_BOOL,
228             .help = "ipv4 bool value, not supported by gluster",
229         },
230         {
231             .name = "ipv6",
232             .type = QEMU_OPT_BOOL,
233             .help = "ipv6 bool value, not supported by gluster",
234         },
235         { /* end of list */ }
236     },
237 };
238 
239 static void glfs_set_preopened(const char *volume, glfs_t *fs)
240 {
241     ListElement *entry = NULL;
242 
243     entry = g_new(ListElement, 1);
244 
245     entry->saved.volume = g_strdup(volume);
246 
247     entry->saved.fs = fs;
248     entry->saved.ref = 1;
249 
250     QLIST_INSERT_HEAD(&glfs_list, entry, list);
251 }
252 
253 static glfs_t *glfs_find_preopened(const char *volume)
254 {
255     ListElement *entry = NULL;
256 
257      QLIST_FOREACH(entry, &glfs_list, list) {
258         if (strcmp(entry->saved.volume, volume) == 0) {
259             entry->saved.ref++;
260             return entry->saved.fs;
261         }
262      }
263 
264     return NULL;
265 }
266 
267 static void glfs_clear_preopened(glfs_t *fs)
268 {
269     ListElement *entry = NULL;
270     ListElement *next;
271 
272     if (fs == NULL) {
273         return;
274     }
275 
276     QLIST_FOREACH_SAFE(entry, &glfs_list, list, next) {
277         if (entry->saved.fs == fs) {
278             if (--entry->saved.ref) {
279                 return;
280             }
281 
282             QLIST_REMOVE(entry, list);
283 
284             glfs_fini(entry->saved.fs);
285             g_free(entry->saved.volume);
286             g_free(entry);
287         }
288     }
289 }
290 
291 static int parse_volume_options(BlockdevOptionsGluster *gconf, char *path)
292 {
293     char *p, *q;
294 
295     if (!path) {
296         return -EINVAL;
297     }
298 
299     /* volume */
300     p = q = path + strspn(path, "/");
301     p += strcspn(p, "/");
302     if (*p == '\0') {
303         return -EINVAL;
304     }
305     gconf->volume = g_strndup(q, p - q);
306 
307     /* path */
308     p += strspn(p, "/");
309     if (*p == '\0') {
310         return -EINVAL;
311     }
312     gconf->path = g_strdup(p);
313     return 0;
314 }
315 
316 /*
317  * file=gluster[+transport]://[host[:port]]/volume/path[?socket=...]
318  *
319  * 'gluster' is the protocol.
320  *
321  * 'transport' specifies the transport type used to connect to gluster
322  * management daemon (glusterd). Valid transport types are
323  * tcp or unix. If a transport type isn't specified, then tcp type is assumed.
324  *
325  * 'host' specifies the host where the volume file specification for
326  * the given volume resides. This can be either hostname or ipv4 address.
327  * If transport type is 'unix', then 'host' field should not be specified.
328  * The 'socket' field needs to be populated with the path to unix domain
329  * socket.
330  *
331  * 'port' is the port number on which glusterd is listening. This is optional
332  * and if not specified, QEMU will send 0 which will make gluster to use the
333  * default port. If the transport type is unix, then 'port' should not be
334  * specified.
335  *
336  * 'volume' is the name of the gluster volume which contains the VM image.
337  *
338  * 'path' is the path to the actual VM image that resides on gluster volume.
339  *
340  * Examples:
341  *
342  * file=gluster://1.2.3.4/testvol/a.img
343  * file=gluster+tcp://1.2.3.4/testvol/a.img
344  * file=gluster+tcp://1.2.3.4:24007/testvol/dir/a.img
345  * file=gluster+tcp://host.domain.com:24007/testvol/dir/a.img
346  * file=gluster+unix:///testvol/dir/a.img?socket=/tmp/glusterd.socket
347  */
348 static int qemu_gluster_parse_uri(BlockdevOptionsGluster *gconf,
349                                   const char *filename)
350 {
351     SocketAddress *gsconf;
352     URI *uri;
353     QueryParams *qp = NULL;
354     bool is_unix = false;
355     int ret = 0;
356 
357     uri = uri_parse(filename);
358     if (!uri) {
359         return -EINVAL;
360     }
361 
362     gconf->server = g_new0(SocketAddressList, 1);
363     gconf->server->value = gsconf = g_new0(SocketAddress, 1);
364 
365     /* transport */
366     if (!uri->scheme || !strcmp(uri->scheme, "gluster")) {
367         gsconf->type = SOCKET_ADDRESS_TYPE_INET;
368     } else if (!strcmp(uri->scheme, "gluster+tcp")) {
369         gsconf->type = SOCKET_ADDRESS_TYPE_INET;
370     } else if (!strcmp(uri->scheme, "gluster+unix")) {
371         gsconf->type = SOCKET_ADDRESS_TYPE_UNIX;
372         is_unix = true;
373     } else if (!strcmp(uri->scheme, "gluster+rdma")) {
374         gsconf->type = SOCKET_ADDRESS_TYPE_INET;
375         warn_report("rdma feature is not supported, falling back to tcp");
376     } else {
377         ret = -EINVAL;
378         goto out;
379     }
380 
381     ret = parse_volume_options(gconf, uri->path);
382     if (ret < 0) {
383         goto out;
384     }
385 
386     qp = query_params_parse(uri->query);
387     if (qp->n > 1 || (is_unix && !qp->n) || (!is_unix && qp->n)) {
388         ret = -EINVAL;
389         goto out;
390     }
391 
392     if (is_unix) {
393         if (uri->server || uri->port) {
394             ret = -EINVAL;
395             goto out;
396         }
397         if (strcmp(qp->p[0].name, "socket")) {
398             ret = -EINVAL;
399             goto out;
400         }
401         gsconf->u.q_unix.path = g_strdup(qp->p[0].value);
402     } else {
403         gsconf->u.inet.host = g_strdup(uri->server ? uri->server : "localhost");
404         if (uri->port) {
405             gsconf->u.inet.port = g_strdup_printf("%d", uri->port);
406         } else {
407             gsconf->u.inet.port = g_strdup_printf("%d", GLUSTER_DEFAULT_PORT);
408         }
409     }
410 
411 out:
412     if (qp) {
413         query_params_free(qp);
414     }
415     uri_free(uri);
416     return ret;
417 }
418 
419 static struct glfs *qemu_gluster_glfs_init(BlockdevOptionsGluster *gconf,
420                                            Error **errp)
421 {
422     struct glfs *glfs;
423     int ret;
424     int old_errno;
425     SocketAddressList *server;
426     unsigned long long port;
427 
428     glfs = glfs_find_preopened(gconf->volume);
429     if (glfs) {
430         return glfs;
431     }
432 
433     glfs = glfs_new(gconf->volume);
434     if (!glfs) {
435         goto out;
436     }
437 
438     glfs_set_preopened(gconf->volume, glfs);
439 
440     for (server = gconf->server; server; server = server->next) {
441         switch (server->value->type) {
442         case SOCKET_ADDRESS_TYPE_UNIX:
443             ret = glfs_set_volfile_server(glfs, "unix",
444                                    server->value->u.q_unix.path, 0);
445             break;
446         case SOCKET_ADDRESS_TYPE_INET:
447             if (parse_uint_full(server->value->u.inet.port, &port, 10) < 0 ||
448                 port > 65535) {
449                 error_setg(errp, "'%s' is not a valid port number",
450                            server->value->u.inet.port);
451                 errno = EINVAL;
452                 goto out;
453             }
454             ret = glfs_set_volfile_server(glfs, "tcp",
455                                    server->value->u.inet.host,
456                                    (int)port);
457             break;
458         case SOCKET_ADDRESS_TYPE_VSOCK:
459         case SOCKET_ADDRESS_TYPE_FD:
460         default:
461             abort();
462         }
463 
464         if (ret < 0) {
465             goto out;
466         }
467     }
468 
469     ret = glfs_set_logging(glfs, gconf->logfile, gconf->debug);
470     if (ret < 0) {
471         goto out;
472     }
473 
474     ret = glfs_init(glfs);
475     if (ret) {
476         error_setg(errp, "Gluster connection for volume %s, path %s failed"
477                          " to connect", gconf->volume, gconf->path);
478         for (server = gconf->server; server; server = server->next) {
479             if (server->value->type  == SOCKET_ADDRESS_TYPE_UNIX) {
480                 error_append_hint(errp, "hint: failed on socket %s ",
481                                   server->value->u.q_unix.path);
482             } else {
483                 error_append_hint(errp, "hint: failed on host %s and port %s ",
484                                   server->value->u.inet.host,
485                                   server->value->u.inet.port);
486             }
487         }
488 
489         error_append_hint(errp, "Please refer to gluster logs for more info\n");
490 
491         /* glfs_init sometimes doesn't set errno although docs suggest that */
492         if (errno == 0) {
493             errno = EINVAL;
494         }
495 
496         goto out;
497     }
498     return glfs;
499 
500 out:
501     if (glfs) {
502         old_errno = errno;
503         glfs_clear_preopened(glfs);
504         errno = old_errno;
505     }
506     return NULL;
507 }
508 
509 /*
510  * Convert the json formatted command line into qapi.
511 */
512 static int qemu_gluster_parse_json(BlockdevOptionsGluster *gconf,
513                                   QDict *options, Error **errp)
514 {
515     QemuOpts *opts;
516     SocketAddress *gsconf = NULL;
517     SocketAddressList *curr = NULL;
518     QDict *backing_options = NULL;
519     Error *local_err = NULL;
520     char *str = NULL;
521     const char *ptr;
522     int i, type, num_servers;
523 
524     /* create opts info from runtime_json_opts list */
525     opts = qemu_opts_create(&runtime_json_opts, NULL, 0, &error_abort);
526     if (!qemu_opts_absorb_qdict(opts, options, &local_err)) {
527         goto out;
528     }
529 
530     num_servers = qdict_array_entries(options, GLUSTER_OPT_SERVER_PATTERN);
531     if (num_servers < 1) {
532         error_setg(&local_err, QERR_MISSING_PARAMETER, "server");
533         goto out;
534     }
535 
536     ptr = qemu_opt_get(opts, GLUSTER_OPT_VOLUME);
537     if (!ptr) {
538         error_setg(&local_err, QERR_MISSING_PARAMETER, GLUSTER_OPT_VOLUME);
539         goto out;
540     }
541     gconf->volume = g_strdup(ptr);
542 
543     ptr = qemu_opt_get(opts, GLUSTER_OPT_PATH);
544     if (!ptr) {
545         error_setg(&local_err, QERR_MISSING_PARAMETER, GLUSTER_OPT_PATH);
546         goto out;
547     }
548     gconf->path = g_strdup(ptr);
549     qemu_opts_del(opts);
550 
551     for (i = 0; i < num_servers; i++) {
552         str = g_strdup_printf(GLUSTER_OPT_SERVER_PATTERN"%d.", i);
553         qdict_extract_subqdict(options, &backing_options, str);
554 
555         /* create opts info from runtime_type_opts list */
556         opts = qemu_opts_create(&runtime_type_opts, NULL, 0, &error_abort);
557         if (!qemu_opts_absorb_qdict(opts, backing_options, &local_err)) {
558             goto out;
559         }
560 
561         ptr = qemu_opt_get(opts, GLUSTER_OPT_TYPE);
562         if (!ptr) {
563             error_setg(&local_err, QERR_MISSING_PARAMETER, GLUSTER_OPT_TYPE);
564             error_append_hint(&local_err, GERR_INDEX_HINT, i);
565             goto out;
566 
567         }
568         gsconf = g_new0(SocketAddress, 1);
569         if (!strcmp(ptr, "tcp")) {
570             ptr = "inet";       /* accept legacy "tcp" */
571         }
572         type = qapi_enum_parse(&SocketAddressType_lookup, ptr, -1, NULL);
573         if (type != SOCKET_ADDRESS_TYPE_INET
574             && type != SOCKET_ADDRESS_TYPE_UNIX) {
575             error_setg(&local_err,
576                        "Parameter '%s' may be 'inet' or 'unix'",
577                        GLUSTER_OPT_TYPE);
578             error_append_hint(&local_err, GERR_INDEX_HINT, i);
579             goto out;
580         }
581         gsconf->type = type;
582         qemu_opts_del(opts);
583 
584         if (gsconf->type == SOCKET_ADDRESS_TYPE_INET) {
585             /* create opts info from runtime_inet_opts list */
586             opts = qemu_opts_create(&runtime_inet_opts, NULL, 0, &error_abort);
587             if (!qemu_opts_absorb_qdict(opts, backing_options, &local_err)) {
588                 goto out;
589             }
590 
591             ptr = qemu_opt_get(opts, GLUSTER_OPT_HOST);
592             if (!ptr) {
593                 error_setg(&local_err, QERR_MISSING_PARAMETER,
594                            GLUSTER_OPT_HOST);
595                 error_append_hint(&local_err, GERR_INDEX_HINT, i);
596                 goto out;
597             }
598             gsconf->u.inet.host = g_strdup(ptr);
599             ptr = qemu_opt_get(opts, GLUSTER_OPT_PORT);
600             if (!ptr) {
601                 error_setg(&local_err, QERR_MISSING_PARAMETER,
602                            GLUSTER_OPT_PORT);
603                 error_append_hint(&local_err, GERR_INDEX_HINT, i);
604                 goto out;
605             }
606             gsconf->u.inet.port = g_strdup(ptr);
607 
608             /* defend for unsupported fields in InetSocketAddress,
609              * i.e. @ipv4, @ipv6  and @to
610              */
611             ptr = qemu_opt_get(opts, GLUSTER_OPT_TO);
612             if (ptr) {
613                 gsconf->u.inet.has_to = true;
614             }
615             ptr = qemu_opt_get(opts, GLUSTER_OPT_IPV4);
616             if (ptr) {
617                 gsconf->u.inet.has_ipv4 = true;
618             }
619             ptr = qemu_opt_get(opts, GLUSTER_OPT_IPV6);
620             if (ptr) {
621                 gsconf->u.inet.has_ipv6 = true;
622             }
623             if (gsconf->u.inet.has_to) {
624                 error_setg(&local_err, "Parameter 'to' not supported");
625                 goto out;
626             }
627             if (gsconf->u.inet.has_ipv4 || gsconf->u.inet.has_ipv6) {
628                 error_setg(&local_err, "Parameters 'ipv4/ipv6' not supported");
629                 goto out;
630             }
631             qemu_opts_del(opts);
632         } else {
633             /* create opts info from runtime_unix_opts list */
634             opts = qemu_opts_create(&runtime_unix_opts, NULL, 0, &error_abort);
635             if (!qemu_opts_absorb_qdict(opts, backing_options, &local_err)) {
636                 goto out;
637             }
638 
639             ptr = qemu_opt_get(opts, GLUSTER_OPT_PATH);
640             if (!ptr) {
641                 ptr = qemu_opt_get(opts, GLUSTER_OPT_SOCKET);
642             } else if (qemu_opt_get(opts, GLUSTER_OPT_SOCKET)) {
643                 error_setg(&local_err,
644                            "Conflicting parameters 'path' and 'socket'");
645                 error_append_hint(&local_err, GERR_INDEX_HINT, i);
646                 goto out;
647             }
648             if (!ptr) {
649                 error_setg(&local_err, QERR_MISSING_PARAMETER,
650                            GLUSTER_OPT_PATH);
651                 error_append_hint(&local_err, GERR_INDEX_HINT, i);
652                 goto out;
653             }
654             gsconf->u.q_unix.path = g_strdup(ptr);
655             qemu_opts_del(opts);
656         }
657 
658         if (gconf->server == NULL) {
659             gconf->server = g_new0(SocketAddressList, 1);
660             gconf->server->value = gsconf;
661             curr = gconf->server;
662         } else {
663             curr->next = g_new0(SocketAddressList, 1);
664             curr->next->value = gsconf;
665             curr = curr->next;
666         }
667         gsconf = NULL;
668 
669         qobject_unref(backing_options);
670         backing_options = NULL;
671         g_free(str);
672         str = NULL;
673     }
674 
675     return 0;
676 
677 out:
678     error_propagate(errp, local_err);
679     qapi_free_SocketAddress(gsconf);
680     qemu_opts_del(opts);
681     g_free(str);
682     qobject_unref(backing_options);
683     errno = EINVAL;
684     return -errno;
685 }
686 
687 /* Converts options given in @filename and the @options QDict into the QAPI
688  * object @gconf. */
689 static int qemu_gluster_parse(BlockdevOptionsGluster *gconf,
690                               const char *filename,
691                               QDict *options, Error **errp)
692 {
693     int ret;
694     if (filename) {
695         ret = qemu_gluster_parse_uri(gconf, filename);
696         if (ret < 0) {
697             error_setg(errp, "invalid URI %s", filename);
698             error_append_hint(errp, "Usage: file=gluster[+transport]://"
699                                     "[host[:port]]volume/path[?socket=...]"
700                                     "[,file.debug=N]"
701                                     "[,file.logfile=/path/filename.log]\n");
702             return ret;
703         }
704     } else {
705         ret = qemu_gluster_parse_json(gconf, options, errp);
706         if (ret < 0) {
707             error_append_hint(errp, "Usage: "
708                              "-drive driver=qcow2,file.driver=gluster,"
709                              "file.volume=testvol,file.path=/path/a.qcow2"
710                              "[,file.debug=9]"
711                              "[,file.logfile=/path/filename.log],"
712                              "file.server.0.type=inet,"
713                              "file.server.0.host=1.2.3.4,"
714                              "file.server.0.port=24007,"
715                              "file.server.1.transport=unix,"
716                              "file.server.1.path=/var/run/glusterd.socket ..."
717                              "\n");
718             return ret;
719         }
720     }
721 
722     return 0;
723 }
724 
725 static struct glfs *qemu_gluster_init(BlockdevOptionsGluster *gconf,
726                                       const char *filename,
727                                       QDict *options, Error **errp)
728 {
729     int ret;
730 
731     ret = qemu_gluster_parse(gconf, filename, options, errp);
732     if (ret < 0) {
733         errno = -ret;
734         return NULL;
735     }
736 
737     return qemu_gluster_glfs_init(gconf, errp);
738 }
739 
740 /*
741  * AIO callback routine called from GlusterFS thread.
742  */
743 static void gluster_finish_aiocb(struct glfs_fd *fd, ssize_t ret,
744 #ifdef CONFIG_GLUSTERFS_IOCB_HAS_STAT
745                                  struct glfs_stat *pre, struct glfs_stat *post,
746 #endif
747                                  void *arg)
748 {
749     GlusterAIOCB *acb = (GlusterAIOCB *)arg;
750 
751     if (!ret || ret == acb->size) {
752         acb->ret = 0; /* Success */
753     } else if (ret < 0) {
754         acb->ret = -errno; /* Read/Write failed */
755     } else {
756         acb->ret = -EIO; /* Partial read/write - fail it */
757     }
758 
759     aio_co_schedule(acb->aio_context, acb->coroutine);
760 }
761 
762 static void qemu_gluster_parse_flags(int bdrv_flags, int *open_flags)
763 {
764     assert(open_flags != NULL);
765 
766     *open_flags |= O_BINARY;
767 
768     if (bdrv_flags & BDRV_O_RDWR) {
769         *open_flags |= O_RDWR;
770     } else {
771         *open_flags |= O_RDONLY;
772     }
773 
774     if ((bdrv_flags & BDRV_O_NOCACHE)) {
775         *open_flags |= O_DIRECT;
776     }
777 }
778 
779 /*
780  * Do SEEK_DATA/HOLE to detect if it is functional. Older broken versions of
781  * gfapi incorrectly return the current offset when SEEK_DATA/HOLE is used.
782  * - Corrected versions return -1 and set errno to EINVAL.
783  * - Versions that support SEEK_DATA/HOLE correctly, will return -1 and set
784  *   errno to ENXIO when SEEK_DATA is called with a position of EOF.
785  */
786 static bool qemu_gluster_test_seek(struct glfs_fd *fd)
787 {
788     off_t ret = 0;
789 
790 #if defined SEEK_HOLE && defined SEEK_DATA
791     off_t eof;
792 
793     eof = glfs_lseek(fd, 0, SEEK_END);
794     if (eof < 0) {
795         /* this should never occur */
796         return false;
797     }
798 
799     /* this should always fail with ENXIO if SEEK_DATA is supported */
800     ret = glfs_lseek(fd, eof, SEEK_DATA);
801 #endif
802 
803     return (ret < 0) && (errno == ENXIO);
804 }
805 
806 static int qemu_gluster_open(BlockDriverState *bs,  QDict *options,
807                              int bdrv_flags, Error **errp)
808 {
809     BDRVGlusterState *s = bs->opaque;
810     int open_flags = 0;
811     int ret = 0;
812     BlockdevOptionsGluster *gconf = NULL;
813     QemuOpts *opts;
814     Error *local_err = NULL;
815     const char *filename, *logfile;
816 
817     opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
818     if (!qemu_opts_absorb_qdict(opts, options, &local_err)) {
819         error_propagate(errp, local_err);
820         ret = -EINVAL;
821         goto out;
822     }
823 
824     filename = qemu_opt_get(opts, GLUSTER_OPT_FILENAME);
825 
826     s->debug = qemu_opt_get_number(opts, GLUSTER_OPT_DEBUG,
827                                    GLUSTER_DEBUG_DEFAULT);
828     if (s->debug < 0) {
829         s->debug = 0;
830     } else if (s->debug > GLUSTER_DEBUG_MAX) {
831         s->debug = GLUSTER_DEBUG_MAX;
832     }
833 
834     gconf = g_new0(BlockdevOptionsGluster, 1);
835     gconf->debug = s->debug;
836     gconf->has_debug = true;
837 
838     logfile = qemu_opt_get(opts, GLUSTER_OPT_LOGFILE);
839     s->logfile = g_strdup(logfile ? logfile : GLUSTER_LOGFILE_DEFAULT);
840 
841     gconf->logfile = g_strdup(s->logfile);
842     gconf->has_logfile = true;
843 
844     s->glfs = qemu_gluster_init(gconf, filename, options, errp);
845     if (!s->glfs) {
846         ret = -errno;
847         goto out;
848     }
849 
850 #ifdef CONFIG_GLUSTERFS_XLATOR_OPT
851     /* Without this, if fsync fails for a recoverable reason (for instance,
852      * ENOSPC), gluster will dump its cache, preventing retries.  This means
853      * almost certain data loss.  Not all gluster versions support the
854      * 'resync-failed-syncs-after-fsync' key value, but there is no way to
855      * discover during runtime if it is supported (this api returns success for
856      * unknown key/value pairs) */
857     ret = glfs_set_xlator_option(s->glfs, "*-write-behind",
858                                           "resync-failed-syncs-after-fsync",
859                                           "on");
860     if (ret < 0) {
861         error_setg_errno(errp, errno, "Unable to set xlator key/value pair");
862         ret = -errno;
863         goto out;
864     }
865 #endif
866 
867     qemu_gluster_parse_flags(bdrv_flags, &open_flags);
868 
869     s->fd = glfs_open(s->glfs, gconf->path, open_flags);
870     ret = s->fd ? 0 : -errno;
871 
872     if (ret == -EACCES || ret == -EROFS) {
873         /* Try to degrade to read-only, but if it doesn't work, still use the
874          * normal error message. */
875         if (bdrv_apply_auto_read_only(bs, NULL, NULL) == 0) {
876             open_flags = (open_flags & ~O_RDWR) | O_RDONLY;
877             s->fd = glfs_open(s->glfs, gconf->path, open_flags);
878             ret = s->fd ? 0 : -errno;
879         }
880     }
881 
882     s->supports_seek_data = qemu_gluster_test_seek(s->fd);
883 
884 out:
885     qemu_opts_del(opts);
886     qapi_free_BlockdevOptionsGluster(gconf);
887     if (!ret) {
888         return ret;
889     }
890     g_free(s->logfile);
891     if (s->fd) {
892         glfs_close(s->fd);
893     }
894 
895     glfs_clear_preopened(s->glfs);
896 
897     return ret;
898 }
899 
900 static void qemu_gluster_refresh_limits(BlockDriverState *bs, Error **errp)
901 {
902     bs->bl.max_transfer = GLUSTER_MAX_TRANSFER;
903 }
904 
905 static int qemu_gluster_reopen_prepare(BDRVReopenState *state,
906                                        BlockReopenQueue *queue, Error **errp)
907 {
908     int ret = 0;
909     BDRVGlusterState *s;
910     BDRVGlusterReopenState *reop_s;
911     BlockdevOptionsGluster *gconf;
912     int open_flags = 0;
913 
914     assert(state != NULL);
915     assert(state->bs != NULL);
916 
917     s = state->bs->opaque;
918 
919     state->opaque = g_new0(BDRVGlusterReopenState, 1);
920     reop_s = state->opaque;
921 
922     qemu_gluster_parse_flags(state->flags, &open_flags);
923 
924     gconf = g_new0(BlockdevOptionsGluster, 1);
925     gconf->debug = s->debug;
926     gconf->has_debug = true;
927     gconf->logfile = g_strdup(s->logfile);
928     gconf->has_logfile = true;
929 
930     /*
931      * If 'state->bs->exact_filename' is empty, 'state->options' should contain
932      * the JSON parameters already parsed.
933      */
934     if (state->bs->exact_filename[0] != '\0') {
935         reop_s->glfs = qemu_gluster_init(gconf, state->bs->exact_filename, NULL,
936                                          errp);
937     } else {
938         reop_s->glfs = qemu_gluster_init(gconf, NULL, state->options, errp);
939     }
940     if (reop_s->glfs == NULL) {
941         ret = -errno;
942         goto exit;
943     }
944 
945 #ifdef CONFIG_GLUSTERFS_XLATOR_OPT
946     ret = glfs_set_xlator_option(reop_s->glfs, "*-write-behind",
947                                  "resync-failed-syncs-after-fsync", "on");
948     if (ret < 0) {
949         error_setg_errno(errp, errno, "Unable to set xlator key/value pair");
950         ret = -errno;
951         goto exit;
952     }
953 #endif
954 
955     reop_s->fd = glfs_open(reop_s->glfs, gconf->path, open_flags);
956     if (reop_s->fd == NULL) {
957         /* reops->glfs will be cleaned up in _abort */
958         ret = -errno;
959         goto exit;
960     }
961 
962 exit:
963     /* state->opaque will be freed in either the _abort or _commit */
964     qapi_free_BlockdevOptionsGluster(gconf);
965     return ret;
966 }
967 
968 static void qemu_gluster_reopen_commit(BDRVReopenState *state)
969 {
970     BDRVGlusterReopenState *reop_s = state->opaque;
971     BDRVGlusterState *s = state->bs->opaque;
972 
973 
974     /* close the old */
975     if (s->fd) {
976         glfs_close(s->fd);
977     }
978 
979     glfs_clear_preopened(s->glfs);
980 
981     /* use the newly opened image / connection */
982     s->fd         = reop_s->fd;
983     s->glfs       = reop_s->glfs;
984 
985     g_free(state->opaque);
986     state->opaque = NULL;
987 
988     return;
989 }
990 
991 
992 static void qemu_gluster_reopen_abort(BDRVReopenState *state)
993 {
994     BDRVGlusterReopenState *reop_s = state->opaque;
995 
996     if (reop_s == NULL) {
997         return;
998     }
999 
1000     if (reop_s->fd) {
1001         glfs_close(reop_s->fd);
1002     }
1003 
1004     glfs_clear_preopened(reop_s->glfs);
1005 
1006     g_free(state->opaque);
1007     state->opaque = NULL;
1008 
1009     return;
1010 }
1011 
1012 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1013 static coroutine_fn int qemu_gluster_co_pwrite_zeroes(BlockDriverState *bs,
1014                                                       int64_t offset,
1015                                                       int size,
1016                                                       BdrvRequestFlags flags)
1017 {
1018     int ret;
1019     GlusterAIOCB acb;
1020     BDRVGlusterState *s = bs->opaque;
1021 
1022     acb.size = size;
1023     acb.ret = 0;
1024     acb.coroutine = qemu_coroutine_self();
1025     acb.aio_context = bdrv_get_aio_context(bs);
1026 
1027     ret = glfs_zerofill_async(s->fd, offset, size, gluster_finish_aiocb, &acb);
1028     if (ret < 0) {
1029         return -errno;
1030     }
1031 
1032     qemu_coroutine_yield();
1033     return acb.ret;
1034 }
1035 #endif
1036 
1037 static int qemu_gluster_do_truncate(struct glfs_fd *fd, int64_t offset,
1038                                     PreallocMode prealloc, Error **errp)
1039 {
1040     int64_t current_length;
1041 
1042     current_length = glfs_lseek(fd, 0, SEEK_END);
1043     if (current_length < 0) {
1044         error_setg_errno(errp, errno, "Failed to determine current size");
1045         return -errno;
1046     }
1047 
1048     if (current_length > offset && prealloc != PREALLOC_MODE_OFF) {
1049         error_setg(errp, "Cannot use preallocation for shrinking files");
1050         return -ENOTSUP;
1051     }
1052 
1053     if (current_length == offset) {
1054         return 0;
1055     }
1056 
1057     switch (prealloc) {
1058 #ifdef CONFIG_GLUSTERFS_FALLOCATE
1059     case PREALLOC_MODE_FALLOC:
1060         if (glfs_fallocate(fd, 0, current_length, offset - current_length)) {
1061             error_setg_errno(errp, errno, "Could not preallocate data");
1062             return -errno;
1063         }
1064         break;
1065 #endif /* CONFIG_GLUSTERFS_FALLOCATE */
1066 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1067     case PREALLOC_MODE_FULL:
1068         if (glfs_ftruncate(fd, offset)) {
1069             error_setg_errno(errp, errno, "Could not resize file");
1070             return -errno;
1071         }
1072         if (glfs_zerofill(fd, current_length, offset - current_length)) {
1073             error_setg_errno(errp, errno, "Could not zerofill the new area");
1074             return -errno;
1075         }
1076         break;
1077 #endif /* CONFIG_GLUSTERFS_ZEROFILL */
1078     case PREALLOC_MODE_OFF:
1079         if (glfs_ftruncate(fd, offset)) {
1080             error_setg_errno(errp, errno, "Could not resize file");
1081             return -errno;
1082         }
1083         break;
1084     default:
1085         error_setg(errp, "Unsupported preallocation mode: %s",
1086                    PreallocMode_str(prealloc));
1087         return -EINVAL;
1088     }
1089 
1090     return 0;
1091 }
1092 
1093 static int qemu_gluster_co_create(BlockdevCreateOptions *options,
1094                                   Error **errp)
1095 {
1096     BlockdevCreateOptionsGluster *opts = &options->u.gluster;
1097     struct glfs *glfs;
1098     struct glfs_fd *fd = NULL;
1099     int ret = 0;
1100 
1101     assert(options->driver == BLOCKDEV_DRIVER_GLUSTER);
1102 
1103     glfs = qemu_gluster_glfs_init(opts->location, errp);
1104     if (!glfs) {
1105         ret = -errno;
1106         goto out;
1107     }
1108 
1109     fd = glfs_creat(glfs, opts->location->path,
1110                     O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, S_IRUSR | S_IWUSR);
1111     if (!fd) {
1112         ret = -errno;
1113         goto out;
1114     }
1115 
1116     ret = qemu_gluster_do_truncate(fd, opts->size, opts->preallocation, errp);
1117 
1118 out:
1119     if (fd) {
1120         if (glfs_close(fd) != 0 && ret == 0) {
1121             ret = -errno;
1122         }
1123     }
1124     glfs_clear_preopened(glfs);
1125     return ret;
1126 }
1127 
1128 static int coroutine_fn qemu_gluster_co_create_opts(BlockDriver *drv,
1129                                                     const char *filename,
1130                                                     QemuOpts *opts,
1131                                                     Error **errp)
1132 {
1133     BlockdevCreateOptions *options;
1134     BlockdevCreateOptionsGluster *gopts;
1135     BlockdevOptionsGluster *gconf;
1136     char *tmp = NULL;
1137     Error *local_err = NULL;
1138     int ret;
1139 
1140     options = g_new0(BlockdevCreateOptions, 1);
1141     options->driver = BLOCKDEV_DRIVER_GLUSTER;
1142     gopts = &options->u.gluster;
1143 
1144     gconf = g_new0(BlockdevOptionsGluster, 1);
1145     gopts->location = gconf;
1146 
1147     gopts->size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
1148                            BDRV_SECTOR_SIZE);
1149 
1150     tmp = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
1151     gopts->preallocation = qapi_enum_parse(&PreallocMode_lookup, tmp,
1152                                            PREALLOC_MODE_OFF, &local_err);
1153     g_free(tmp);
1154     if (local_err) {
1155         error_propagate(errp, local_err);
1156         ret = -EINVAL;
1157         goto fail;
1158     }
1159 
1160     gconf->debug = qemu_opt_get_number_del(opts, GLUSTER_OPT_DEBUG,
1161                                            GLUSTER_DEBUG_DEFAULT);
1162     if (gconf->debug < 0) {
1163         gconf->debug = 0;
1164     } else if (gconf->debug > GLUSTER_DEBUG_MAX) {
1165         gconf->debug = GLUSTER_DEBUG_MAX;
1166     }
1167     gconf->has_debug = true;
1168 
1169     gconf->logfile = qemu_opt_get_del(opts, GLUSTER_OPT_LOGFILE);
1170     if (!gconf->logfile) {
1171         gconf->logfile = g_strdup(GLUSTER_LOGFILE_DEFAULT);
1172     }
1173     gconf->has_logfile = true;
1174 
1175     ret = qemu_gluster_parse(gconf, filename, NULL, errp);
1176     if (ret < 0) {
1177         goto fail;
1178     }
1179 
1180     ret = qemu_gluster_co_create(options, errp);
1181     if (ret < 0) {
1182         goto fail;
1183     }
1184 
1185     ret = 0;
1186 fail:
1187     qapi_free_BlockdevCreateOptions(options);
1188     return ret;
1189 }
1190 
1191 static coroutine_fn int qemu_gluster_co_rw(BlockDriverState *bs,
1192                                            int64_t sector_num, int nb_sectors,
1193                                            QEMUIOVector *qiov, int write)
1194 {
1195     int ret;
1196     GlusterAIOCB acb;
1197     BDRVGlusterState *s = bs->opaque;
1198     size_t size = nb_sectors * BDRV_SECTOR_SIZE;
1199     off_t offset = sector_num * BDRV_SECTOR_SIZE;
1200 
1201     acb.size = size;
1202     acb.ret = 0;
1203     acb.coroutine = qemu_coroutine_self();
1204     acb.aio_context = bdrv_get_aio_context(bs);
1205 
1206     if (write) {
1207         ret = glfs_pwritev_async(s->fd, qiov->iov, qiov->niov, offset, 0,
1208                                  gluster_finish_aiocb, &acb);
1209     } else {
1210         ret = glfs_preadv_async(s->fd, qiov->iov, qiov->niov, offset, 0,
1211                                 gluster_finish_aiocb, &acb);
1212     }
1213 
1214     if (ret < 0) {
1215         return -errno;
1216     }
1217 
1218     qemu_coroutine_yield();
1219     return acb.ret;
1220 }
1221 
1222 static coroutine_fn int qemu_gluster_co_truncate(BlockDriverState *bs,
1223                                                  int64_t offset,
1224                                                  bool exact,
1225                                                  PreallocMode prealloc,
1226                                                  BdrvRequestFlags flags,
1227                                                  Error **errp)
1228 {
1229     BDRVGlusterState *s = bs->opaque;
1230     return qemu_gluster_do_truncate(s->fd, offset, prealloc, errp);
1231 }
1232 
1233 static coroutine_fn int qemu_gluster_co_readv(BlockDriverState *bs,
1234                                               int64_t sector_num,
1235                                               int nb_sectors,
1236                                               QEMUIOVector *qiov)
1237 {
1238     return qemu_gluster_co_rw(bs, sector_num, nb_sectors, qiov, 0);
1239 }
1240 
1241 static coroutine_fn int qemu_gluster_co_writev(BlockDriverState *bs,
1242                                                int64_t sector_num,
1243                                                int nb_sectors,
1244                                                QEMUIOVector *qiov,
1245                                                int flags)
1246 {
1247     assert(!flags);
1248     return qemu_gluster_co_rw(bs, sector_num, nb_sectors, qiov, 1);
1249 }
1250 
1251 static void qemu_gluster_close(BlockDriverState *bs)
1252 {
1253     BDRVGlusterState *s = bs->opaque;
1254 
1255     g_free(s->logfile);
1256     if (s->fd) {
1257         glfs_close(s->fd);
1258         s->fd = NULL;
1259     }
1260     glfs_clear_preopened(s->glfs);
1261 }
1262 
1263 static coroutine_fn int qemu_gluster_co_flush_to_disk(BlockDriverState *bs)
1264 {
1265     int ret;
1266     GlusterAIOCB acb;
1267     BDRVGlusterState *s = bs->opaque;
1268 
1269     acb.size = 0;
1270     acb.ret = 0;
1271     acb.coroutine = qemu_coroutine_self();
1272     acb.aio_context = bdrv_get_aio_context(bs);
1273 
1274     ret = glfs_fsync_async(s->fd, gluster_finish_aiocb, &acb);
1275     if (ret < 0) {
1276         ret = -errno;
1277         goto error;
1278     }
1279 
1280     qemu_coroutine_yield();
1281     if (acb.ret < 0) {
1282         ret = acb.ret;
1283         goto error;
1284     }
1285 
1286     return acb.ret;
1287 
1288 error:
1289     /* Some versions of Gluster (3.5.6 -> 3.5.8?) will not retain its cache
1290      * after a fsync failure, so we have no way of allowing the guest to safely
1291      * continue.  Gluster versions prior to 3.5.6 don't retain the cache
1292      * either, but will invalidate the fd on error, so this is again our only
1293      * option.
1294      *
1295      * The 'resync-failed-syncs-after-fsync' xlator option for the
1296      * write-behind cache will cause later gluster versions to retain its
1297      * cache after error, so long as the fd remains open.  However, we
1298      * currently have no way of knowing if this option is supported.
1299      *
1300      * TODO: Once gluster provides a way for us to determine if the option
1301      * is supported, bypass the closure and setting drv to NULL.  */
1302     qemu_gluster_close(bs);
1303     bs->drv = NULL;
1304     return ret;
1305 }
1306 
1307 #ifdef CONFIG_GLUSTERFS_DISCARD
1308 static coroutine_fn int qemu_gluster_co_pdiscard(BlockDriverState *bs,
1309                                                  int64_t offset, int size)
1310 {
1311     int ret;
1312     GlusterAIOCB acb;
1313     BDRVGlusterState *s = bs->opaque;
1314 
1315     acb.size = 0;
1316     acb.ret = 0;
1317     acb.coroutine = qemu_coroutine_self();
1318     acb.aio_context = bdrv_get_aio_context(bs);
1319 
1320     ret = glfs_discard_async(s->fd, offset, size, gluster_finish_aiocb, &acb);
1321     if (ret < 0) {
1322         return -errno;
1323     }
1324 
1325     qemu_coroutine_yield();
1326     return acb.ret;
1327 }
1328 #endif
1329 
1330 static int64_t qemu_gluster_getlength(BlockDriverState *bs)
1331 {
1332     BDRVGlusterState *s = bs->opaque;
1333     int64_t ret;
1334 
1335     ret = glfs_lseek(s->fd, 0, SEEK_END);
1336     if (ret < 0) {
1337         return -errno;
1338     } else {
1339         return ret;
1340     }
1341 }
1342 
1343 static int64_t qemu_gluster_allocated_file_size(BlockDriverState *bs)
1344 {
1345     BDRVGlusterState *s = bs->opaque;
1346     struct stat st;
1347     int ret;
1348 
1349     ret = glfs_fstat(s->fd, &st);
1350     if (ret < 0) {
1351         return -errno;
1352     } else {
1353         return st.st_blocks * 512;
1354     }
1355 }
1356 
1357 /*
1358  * Find allocation range in @bs around offset @start.
1359  * May change underlying file descriptor's file offset.
1360  * If @start is not in a hole, store @start in @data, and the
1361  * beginning of the next hole in @hole, and return 0.
1362  * If @start is in a non-trailing hole, store @start in @hole and the
1363  * beginning of the next non-hole in @data, and return 0.
1364  * If @start is in a trailing hole or beyond EOF, return -ENXIO.
1365  * If we can't find out, return a negative errno other than -ENXIO.
1366  *
1367  * (Shamefully copied from file-posix.c, only minuscule adaptions.)
1368  */
1369 static int find_allocation(BlockDriverState *bs, off_t start,
1370                            off_t *data, off_t *hole)
1371 {
1372     BDRVGlusterState *s = bs->opaque;
1373 
1374     if (!s->supports_seek_data) {
1375         goto exit;
1376     }
1377 
1378 #if defined SEEK_HOLE && defined SEEK_DATA
1379     off_t offs;
1380 
1381     /*
1382      * SEEK_DATA cases:
1383      * D1. offs == start: start is in data
1384      * D2. offs > start: start is in a hole, next data at offs
1385      * D3. offs < 0, errno = ENXIO: either start is in a trailing hole
1386      *                              or start is beyond EOF
1387      *     If the latter happens, the file has been truncated behind
1388      *     our back since we opened it.  All bets are off then.
1389      *     Treating like a trailing hole is simplest.
1390      * D4. offs < 0, errno != ENXIO: we learned nothing
1391      */
1392     offs = glfs_lseek(s->fd, start, SEEK_DATA);
1393     if (offs < 0) {
1394         return -errno;          /* D3 or D4 */
1395     }
1396 
1397     if (offs < start) {
1398         /* This is not a valid return by lseek().  We are safe to just return
1399          * -EIO in this case, and we'll treat it like D4. Unfortunately some
1400          *  versions of gluster server will return offs < start, so an assert
1401          *  here will unnecessarily abort QEMU. */
1402         return -EIO;
1403     }
1404 
1405     if (offs > start) {
1406         /* D2: in hole, next data at offs */
1407         *hole = start;
1408         *data = offs;
1409         return 0;
1410     }
1411 
1412     /* D1: in data, end not yet known */
1413 
1414     /*
1415      * SEEK_HOLE cases:
1416      * H1. offs == start: start is in a hole
1417      *     If this happens here, a hole has been dug behind our back
1418      *     since the previous lseek().
1419      * H2. offs > start: either start is in data, next hole at offs,
1420      *                   or start is in trailing hole, EOF at offs
1421      *     Linux treats trailing holes like any other hole: offs ==
1422      *     start.  Solaris seeks to EOF instead: offs > start (blech).
1423      *     If that happens here, a hole has been dug behind our back
1424      *     since the previous lseek().
1425      * H3. offs < 0, errno = ENXIO: start is beyond EOF
1426      *     If this happens, the file has been truncated behind our
1427      *     back since we opened it.  Treat it like a trailing hole.
1428      * H4. offs < 0, errno != ENXIO: we learned nothing
1429      *     Pretend we know nothing at all, i.e. "forget" about D1.
1430      */
1431     offs = glfs_lseek(s->fd, start, SEEK_HOLE);
1432     if (offs < 0) {
1433         return -errno;          /* D1 and (H3 or H4) */
1434     }
1435 
1436     if (offs < start) {
1437         /* This is not a valid return by lseek().  We are safe to just return
1438          * -EIO in this case, and we'll treat it like H4. Unfortunately some
1439          *  versions of gluster server will return offs < start, so an assert
1440          *  here will unnecessarily abort QEMU. */
1441         return -EIO;
1442     }
1443 
1444     if (offs > start) {
1445         /*
1446          * D1 and H2: either in data, next hole at offs, or it was in
1447          * data but is now in a trailing hole.  In the latter case,
1448          * all bets are off.  Treating it as if it there was data all
1449          * the way to EOF is safe, so simply do that.
1450          */
1451         *data = start;
1452         *hole = offs;
1453         return 0;
1454     }
1455 
1456     /* D1 and H1 */
1457     return -EBUSY;
1458 #endif
1459 
1460 exit:
1461     return -ENOTSUP;
1462 }
1463 
1464 /*
1465  * Returns the allocation status of the specified offset.
1466  *
1467  * The block layer guarantees 'offset' and 'bytes' are within bounds.
1468  *
1469  * 'pnum' is set to the number of bytes (including and immediately following
1470  * the specified offset) that are known to be in the same
1471  * allocated/unallocated state.
1472  *
1473  * 'bytes' is the max value 'pnum' should be set to.
1474  *
1475  * (Based on raw_co_block_status() from file-posix.c.)
1476  */
1477 static int coroutine_fn qemu_gluster_co_block_status(BlockDriverState *bs,
1478                                                      bool want_zero,
1479                                                      int64_t offset,
1480                                                      int64_t bytes,
1481                                                      int64_t *pnum,
1482                                                      int64_t *map,
1483                                                      BlockDriverState **file)
1484 {
1485     BDRVGlusterState *s = bs->opaque;
1486     off_t data = 0, hole = 0;
1487     int ret = -EINVAL;
1488 
1489     if (!s->fd) {
1490         return ret;
1491     }
1492 
1493     if (!want_zero) {
1494         *pnum = bytes;
1495         *map = offset;
1496         *file = bs;
1497         return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID;
1498     }
1499 
1500     ret = find_allocation(bs, offset, &data, &hole);
1501     if (ret == -ENXIO) {
1502         /* Trailing hole */
1503         *pnum = bytes;
1504         ret = BDRV_BLOCK_ZERO;
1505     } else if (ret < 0) {
1506         /* No info available, so pretend there are no holes */
1507         *pnum = bytes;
1508         ret = BDRV_BLOCK_DATA;
1509     } else if (data == offset) {
1510         /* On a data extent, compute bytes to the end of the extent,
1511          * possibly including a partial sector at EOF. */
1512         *pnum = MIN(bytes, hole - offset);
1513         ret = BDRV_BLOCK_DATA;
1514     } else {
1515         /* On a hole, compute bytes to the beginning of the next extent.  */
1516         assert(hole == offset);
1517         *pnum = MIN(bytes, data - offset);
1518         ret = BDRV_BLOCK_ZERO;
1519     }
1520 
1521     *map = offset;
1522     *file = bs;
1523 
1524     return ret | BDRV_BLOCK_OFFSET_VALID;
1525 }
1526 
1527 
1528 static const char *const gluster_strong_open_opts[] = {
1529     GLUSTER_OPT_VOLUME,
1530     GLUSTER_OPT_PATH,
1531     GLUSTER_OPT_TYPE,
1532     GLUSTER_OPT_SERVER_PATTERN,
1533     GLUSTER_OPT_HOST,
1534     GLUSTER_OPT_PORT,
1535     GLUSTER_OPT_TO,
1536     GLUSTER_OPT_IPV4,
1537     GLUSTER_OPT_IPV6,
1538     GLUSTER_OPT_SOCKET,
1539 
1540     NULL
1541 };
1542 
1543 static BlockDriver bdrv_gluster = {
1544     .format_name                  = "gluster",
1545     .protocol_name                = "gluster",
1546     .instance_size                = sizeof(BDRVGlusterState),
1547     .bdrv_needs_filename          = false,
1548     .bdrv_file_open               = qemu_gluster_open,
1549     .bdrv_reopen_prepare          = qemu_gluster_reopen_prepare,
1550     .bdrv_reopen_commit           = qemu_gluster_reopen_commit,
1551     .bdrv_reopen_abort            = qemu_gluster_reopen_abort,
1552     .bdrv_close                   = qemu_gluster_close,
1553     .bdrv_co_create               = qemu_gluster_co_create,
1554     .bdrv_co_create_opts          = qemu_gluster_co_create_opts,
1555     .bdrv_getlength               = qemu_gluster_getlength,
1556     .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size,
1557     .bdrv_co_truncate             = qemu_gluster_co_truncate,
1558     .bdrv_co_readv                = qemu_gluster_co_readv,
1559     .bdrv_co_writev               = qemu_gluster_co_writev,
1560     .bdrv_co_flush_to_disk        = qemu_gluster_co_flush_to_disk,
1561 #ifdef CONFIG_GLUSTERFS_DISCARD
1562     .bdrv_co_pdiscard             = qemu_gluster_co_pdiscard,
1563 #endif
1564 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1565     .bdrv_co_pwrite_zeroes        = qemu_gluster_co_pwrite_zeroes,
1566 #endif
1567     .bdrv_co_block_status         = qemu_gluster_co_block_status,
1568     .bdrv_refresh_limits          = qemu_gluster_refresh_limits,
1569     .create_opts                  = &qemu_gluster_create_opts,
1570     .strong_runtime_opts          = gluster_strong_open_opts,
1571 };
1572 
1573 static BlockDriver bdrv_gluster_tcp = {
1574     .format_name                  = "gluster",
1575     .protocol_name                = "gluster+tcp",
1576     .instance_size                = sizeof(BDRVGlusterState),
1577     .bdrv_needs_filename          = false,
1578     .bdrv_file_open               = qemu_gluster_open,
1579     .bdrv_reopen_prepare          = qemu_gluster_reopen_prepare,
1580     .bdrv_reopen_commit           = qemu_gluster_reopen_commit,
1581     .bdrv_reopen_abort            = qemu_gluster_reopen_abort,
1582     .bdrv_close                   = qemu_gluster_close,
1583     .bdrv_co_create               = qemu_gluster_co_create,
1584     .bdrv_co_create_opts          = qemu_gluster_co_create_opts,
1585     .bdrv_getlength               = qemu_gluster_getlength,
1586     .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size,
1587     .bdrv_co_truncate             = qemu_gluster_co_truncate,
1588     .bdrv_co_readv                = qemu_gluster_co_readv,
1589     .bdrv_co_writev               = qemu_gluster_co_writev,
1590     .bdrv_co_flush_to_disk        = qemu_gluster_co_flush_to_disk,
1591 #ifdef CONFIG_GLUSTERFS_DISCARD
1592     .bdrv_co_pdiscard             = qemu_gluster_co_pdiscard,
1593 #endif
1594 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1595     .bdrv_co_pwrite_zeroes        = qemu_gluster_co_pwrite_zeroes,
1596 #endif
1597     .bdrv_co_block_status         = qemu_gluster_co_block_status,
1598     .bdrv_refresh_limits          = qemu_gluster_refresh_limits,
1599     .create_opts                  = &qemu_gluster_create_opts,
1600     .strong_runtime_opts          = gluster_strong_open_opts,
1601 };
1602 
1603 static BlockDriver bdrv_gluster_unix = {
1604     .format_name                  = "gluster",
1605     .protocol_name                = "gluster+unix",
1606     .instance_size                = sizeof(BDRVGlusterState),
1607     .bdrv_needs_filename          = true,
1608     .bdrv_file_open               = qemu_gluster_open,
1609     .bdrv_reopen_prepare          = qemu_gluster_reopen_prepare,
1610     .bdrv_reopen_commit           = qemu_gluster_reopen_commit,
1611     .bdrv_reopen_abort            = qemu_gluster_reopen_abort,
1612     .bdrv_close                   = qemu_gluster_close,
1613     .bdrv_co_create               = qemu_gluster_co_create,
1614     .bdrv_co_create_opts          = qemu_gluster_co_create_opts,
1615     .bdrv_getlength               = qemu_gluster_getlength,
1616     .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size,
1617     .bdrv_co_truncate             = qemu_gluster_co_truncate,
1618     .bdrv_co_readv                = qemu_gluster_co_readv,
1619     .bdrv_co_writev               = qemu_gluster_co_writev,
1620     .bdrv_co_flush_to_disk        = qemu_gluster_co_flush_to_disk,
1621 #ifdef CONFIG_GLUSTERFS_DISCARD
1622     .bdrv_co_pdiscard             = qemu_gluster_co_pdiscard,
1623 #endif
1624 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1625     .bdrv_co_pwrite_zeroes        = qemu_gluster_co_pwrite_zeroes,
1626 #endif
1627     .bdrv_co_block_status         = qemu_gluster_co_block_status,
1628     .bdrv_refresh_limits          = qemu_gluster_refresh_limits,
1629     .create_opts                  = &qemu_gluster_create_opts,
1630     .strong_runtime_opts          = gluster_strong_open_opts,
1631 };
1632 
1633 /* rdma is deprecated (actually never supported for volfile fetch).
1634  * Let's maintain it for the protocol compatibility, to make sure things
1635  * won't break immediately. For now, gluster+rdma will fall back to gluster+tcp
1636  * protocol with a warning.
1637  * TODO: remove gluster+rdma interface support
1638  */
1639 static BlockDriver bdrv_gluster_rdma = {
1640     .format_name                  = "gluster",
1641     .protocol_name                = "gluster+rdma",
1642     .instance_size                = sizeof(BDRVGlusterState),
1643     .bdrv_needs_filename          = true,
1644     .bdrv_file_open               = qemu_gluster_open,
1645     .bdrv_reopen_prepare          = qemu_gluster_reopen_prepare,
1646     .bdrv_reopen_commit           = qemu_gluster_reopen_commit,
1647     .bdrv_reopen_abort            = qemu_gluster_reopen_abort,
1648     .bdrv_close                   = qemu_gluster_close,
1649     .bdrv_co_create               = qemu_gluster_co_create,
1650     .bdrv_co_create_opts          = qemu_gluster_co_create_opts,
1651     .bdrv_getlength               = qemu_gluster_getlength,
1652     .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size,
1653     .bdrv_co_truncate             = qemu_gluster_co_truncate,
1654     .bdrv_co_readv                = qemu_gluster_co_readv,
1655     .bdrv_co_writev               = qemu_gluster_co_writev,
1656     .bdrv_co_flush_to_disk        = qemu_gluster_co_flush_to_disk,
1657 #ifdef CONFIG_GLUSTERFS_DISCARD
1658     .bdrv_co_pdiscard             = qemu_gluster_co_pdiscard,
1659 #endif
1660 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1661     .bdrv_co_pwrite_zeroes        = qemu_gluster_co_pwrite_zeroes,
1662 #endif
1663     .bdrv_co_block_status         = qemu_gluster_co_block_status,
1664     .bdrv_refresh_limits          = qemu_gluster_refresh_limits,
1665     .create_opts                  = &qemu_gluster_create_opts,
1666     .strong_runtime_opts          = gluster_strong_open_opts,
1667 };
1668 
1669 static void bdrv_gluster_init(void)
1670 {
1671     bdrv_register(&bdrv_gluster_rdma);
1672     bdrv_register(&bdrv_gluster_unix);
1673     bdrv_register(&bdrv_gluster_tcp);
1674     bdrv_register(&bdrv_gluster);
1675 }
1676 
1677 block_init(bdrv_gluster_init);
1678