xref: /openbmc/qemu/block/gluster.c (revision af175e85)
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     const char *filename, *logfile;
815 
816     opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
817     if (!qemu_opts_absorb_qdict(opts, options, errp)) {
818         ret = -EINVAL;
819         goto out;
820     }
821 
822     filename = qemu_opt_get(opts, GLUSTER_OPT_FILENAME);
823 
824     s->debug = qemu_opt_get_number(opts, GLUSTER_OPT_DEBUG,
825                                    GLUSTER_DEBUG_DEFAULT);
826     if (s->debug < 0) {
827         s->debug = 0;
828     } else if (s->debug > GLUSTER_DEBUG_MAX) {
829         s->debug = GLUSTER_DEBUG_MAX;
830     }
831 
832     gconf = g_new0(BlockdevOptionsGluster, 1);
833     gconf->debug = s->debug;
834     gconf->has_debug = true;
835 
836     logfile = qemu_opt_get(opts, GLUSTER_OPT_LOGFILE);
837     s->logfile = g_strdup(logfile ? logfile : GLUSTER_LOGFILE_DEFAULT);
838 
839     gconf->logfile = g_strdup(s->logfile);
840     gconf->has_logfile = true;
841 
842     s->glfs = qemu_gluster_init(gconf, filename, options, errp);
843     if (!s->glfs) {
844         ret = -errno;
845         goto out;
846     }
847 
848 #ifdef CONFIG_GLUSTERFS_XLATOR_OPT
849     /* Without this, if fsync fails for a recoverable reason (for instance,
850      * ENOSPC), gluster will dump its cache, preventing retries.  This means
851      * almost certain data loss.  Not all gluster versions support the
852      * 'resync-failed-syncs-after-fsync' key value, but there is no way to
853      * discover during runtime if it is supported (this api returns success for
854      * unknown key/value pairs) */
855     ret = glfs_set_xlator_option(s->glfs, "*-write-behind",
856                                           "resync-failed-syncs-after-fsync",
857                                           "on");
858     if (ret < 0) {
859         error_setg_errno(errp, errno, "Unable to set xlator key/value pair");
860         ret = -errno;
861         goto out;
862     }
863 #endif
864 
865     qemu_gluster_parse_flags(bdrv_flags, &open_flags);
866 
867     s->fd = glfs_open(s->glfs, gconf->path, open_flags);
868     ret = s->fd ? 0 : -errno;
869 
870     if (ret == -EACCES || ret == -EROFS) {
871         /* Try to degrade to read-only, but if it doesn't work, still use the
872          * normal error message. */
873         if (bdrv_apply_auto_read_only(bs, NULL, NULL) == 0) {
874             open_flags = (open_flags & ~O_RDWR) | O_RDONLY;
875             s->fd = glfs_open(s->glfs, gconf->path, open_flags);
876             ret = s->fd ? 0 : -errno;
877         }
878     }
879 
880     s->supports_seek_data = qemu_gluster_test_seek(s->fd);
881 
882 out:
883     qemu_opts_del(opts);
884     qapi_free_BlockdevOptionsGluster(gconf);
885     if (!ret) {
886         return ret;
887     }
888     g_free(s->logfile);
889     if (s->fd) {
890         glfs_close(s->fd);
891     }
892 
893     glfs_clear_preopened(s->glfs);
894 
895     return ret;
896 }
897 
898 static void qemu_gluster_refresh_limits(BlockDriverState *bs, Error **errp)
899 {
900     bs->bl.max_transfer = GLUSTER_MAX_TRANSFER;
901 }
902 
903 static int qemu_gluster_reopen_prepare(BDRVReopenState *state,
904                                        BlockReopenQueue *queue, Error **errp)
905 {
906     int ret = 0;
907     BDRVGlusterState *s;
908     BDRVGlusterReopenState *reop_s;
909     BlockdevOptionsGluster *gconf;
910     int open_flags = 0;
911 
912     assert(state != NULL);
913     assert(state->bs != NULL);
914 
915     s = state->bs->opaque;
916 
917     state->opaque = g_new0(BDRVGlusterReopenState, 1);
918     reop_s = state->opaque;
919 
920     qemu_gluster_parse_flags(state->flags, &open_flags);
921 
922     gconf = g_new0(BlockdevOptionsGluster, 1);
923     gconf->debug = s->debug;
924     gconf->has_debug = true;
925     gconf->logfile = g_strdup(s->logfile);
926     gconf->has_logfile = true;
927 
928     /*
929      * If 'state->bs->exact_filename' is empty, 'state->options' should contain
930      * the JSON parameters already parsed.
931      */
932     if (state->bs->exact_filename[0] != '\0') {
933         reop_s->glfs = qemu_gluster_init(gconf, state->bs->exact_filename, NULL,
934                                          errp);
935     } else {
936         reop_s->glfs = qemu_gluster_init(gconf, NULL, state->options, errp);
937     }
938     if (reop_s->glfs == NULL) {
939         ret = -errno;
940         goto exit;
941     }
942 
943 #ifdef CONFIG_GLUSTERFS_XLATOR_OPT
944     ret = glfs_set_xlator_option(reop_s->glfs, "*-write-behind",
945                                  "resync-failed-syncs-after-fsync", "on");
946     if (ret < 0) {
947         error_setg_errno(errp, errno, "Unable to set xlator key/value pair");
948         ret = -errno;
949         goto exit;
950     }
951 #endif
952 
953     reop_s->fd = glfs_open(reop_s->glfs, gconf->path, open_flags);
954     if (reop_s->fd == NULL) {
955         /* reops->glfs will be cleaned up in _abort */
956         ret = -errno;
957         goto exit;
958     }
959 
960 exit:
961     /* state->opaque will be freed in either the _abort or _commit */
962     qapi_free_BlockdevOptionsGluster(gconf);
963     return ret;
964 }
965 
966 static void qemu_gluster_reopen_commit(BDRVReopenState *state)
967 {
968     BDRVGlusterReopenState *reop_s = state->opaque;
969     BDRVGlusterState *s = state->bs->opaque;
970 
971 
972     /* close the old */
973     if (s->fd) {
974         glfs_close(s->fd);
975     }
976 
977     glfs_clear_preopened(s->glfs);
978 
979     /* use the newly opened image / connection */
980     s->fd         = reop_s->fd;
981     s->glfs       = reop_s->glfs;
982 
983     g_free(state->opaque);
984     state->opaque = NULL;
985 
986     return;
987 }
988 
989 
990 static void qemu_gluster_reopen_abort(BDRVReopenState *state)
991 {
992     BDRVGlusterReopenState *reop_s = state->opaque;
993 
994     if (reop_s == NULL) {
995         return;
996     }
997 
998     if (reop_s->fd) {
999         glfs_close(reop_s->fd);
1000     }
1001 
1002     glfs_clear_preopened(reop_s->glfs);
1003 
1004     g_free(state->opaque);
1005     state->opaque = NULL;
1006 
1007     return;
1008 }
1009 
1010 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1011 static coroutine_fn int qemu_gluster_co_pwrite_zeroes(BlockDriverState *bs,
1012                                                       int64_t offset,
1013                                                       int size,
1014                                                       BdrvRequestFlags flags)
1015 {
1016     int ret;
1017     GlusterAIOCB acb;
1018     BDRVGlusterState *s = bs->opaque;
1019 
1020     acb.size = size;
1021     acb.ret = 0;
1022     acb.coroutine = qemu_coroutine_self();
1023     acb.aio_context = bdrv_get_aio_context(bs);
1024 
1025     ret = glfs_zerofill_async(s->fd, offset, size, gluster_finish_aiocb, &acb);
1026     if (ret < 0) {
1027         return -errno;
1028     }
1029 
1030     qemu_coroutine_yield();
1031     return acb.ret;
1032 }
1033 #endif
1034 
1035 static int qemu_gluster_do_truncate(struct glfs_fd *fd, int64_t offset,
1036                                     PreallocMode prealloc, Error **errp)
1037 {
1038     int64_t current_length;
1039 
1040     current_length = glfs_lseek(fd, 0, SEEK_END);
1041     if (current_length < 0) {
1042         error_setg_errno(errp, errno, "Failed to determine current size");
1043         return -errno;
1044     }
1045 
1046     if (current_length > offset && prealloc != PREALLOC_MODE_OFF) {
1047         error_setg(errp, "Cannot use preallocation for shrinking files");
1048         return -ENOTSUP;
1049     }
1050 
1051     if (current_length == offset) {
1052         return 0;
1053     }
1054 
1055     switch (prealloc) {
1056 #ifdef CONFIG_GLUSTERFS_FALLOCATE
1057     case PREALLOC_MODE_FALLOC:
1058         if (glfs_fallocate(fd, 0, current_length, offset - current_length)) {
1059             error_setg_errno(errp, errno, "Could not preallocate data");
1060             return -errno;
1061         }
1062         break;
1063 #endif /* CONFIG_GLUSTERFS_FALLOCATE */
1064 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1065     case PREALLOC_MODE_FULL:
1066         if (glfs_ftruncate(fd, offset)) {
1067             error_setg_errno(errp, errno, "Could not resize file");
1068             return -errno;
1069         }
1070         if (glfs_zerofill(fd, current_length, offset - current_length)) {
1071             error_setg_errno(errp, errno, "Could not zerofill the new area");
1072             return -errno;
1073         }
1074         break;
1075 #endif /* CONFIG_GLUSTERFS_ZEROFILL */
1076     case PREALLOC_MODE_OFF:
1077         if (glfs_ftruncate(fd, offset)) {
1078             error_setg_errno(errp, errno, "Could not resize file");
1079             return -errno;
1080         }
1081         break;
1082     default:
1083         error_setg(errp, "Unsupported preallocation mode: %s",
1084                    PreallocMode_str(prealloc));
1085         return -EINVAL;
1086     }
1087 
1088     return 0;
1089 }
1090 
1091 static int qemu_gluster_co_create(BlockdevCreateOptions *options,
1092                                   Error **errp)
1093 {
1094     BlockdevCreateOptionsGluster *opts = &options->u.gluster;
1095     struct glfs *glfs;
1096     struct glfs_fd *fd = NULL;
1097     int ret = 0;
1098 
1099     assert(options->driver == BLOCKDEV_DRIVER_GLUSTER);
1100 
1101     glfs = qemu_gluster_glfs_init(opts->location, errp);
1102     if (!glfs) {
1103         ret = -errno;
1104         goto out;
1105     }
1106 
1107     fd = glfs_creat(glfs, opts->location->path,
1108                     O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, S_IRUSR | S_IWUSR);
1109     if (!fd) {
1110         ret = -errno;
1111         goto out;
1112     }
1113 
1114     ret = qemu_gluster_do_truncate(fd, opts->size, opts->preallocation, errp);
1115 
1116 out:
1117     if (fd) {
1118         if (glfs_close(fd) != 0 && ret == 0) {
1119             ret = -errno;
1120         }
1121     }
1122     glfs_clear_preopened(glfs);
1123     return ret;
1124 }
1125 
1126 static int coroutine_fn qemu_gluster_co_create_opts(BlockDriver *drv,
1127                                                     const char *filename,
1128                                                     QemuOpts *opts,
1129                                                     Error **errp)
1130 {
1131     BlockdevCreateOptions *options;
1132     BlockdevCreateOptionsGluster *gopts;
1133     BlockdevOptionsGluster *gconf;
1134     char *tmp = NULL;
1135     Error *local_err = NULL;
1136     int ret;
1137 
1138     options = g_new0(BlockdevCreateOptions, 1);
1139     options->driver = BLOCKDEV_DRIVER_GLUSTER;
1140     gopts = &options->u.gluster;
1141 
1142     gconf = g_new0(BlockdevOptionsGluster, 1);
1143     gopts->location = gconf;
1144 
1145     gopts->size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
1146                            BDRV_SECTOR_SIZE);
1147 
1148     tmp = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
1149     gopts->preallocation = qapi_enum_parse(&PreallocMode_lookup, tmp,
1150                                            PREALLOC_MODE_OFF, &local_err);
1151     g_free(tmp);
1152     if (local_err) {
1153         error_propagate(errp, local_err);
1154         ret = -EINVAL;
1155         goto fail;
1156     }
1157 
1158     gconf->debug = qemu_opt_get_number_del(opts, GLUSTER_OPT_DEBUG,
1159                                            GLUSTER_DEBUG_DEFAULT);
1160     if (gconf->debug < 0) {
1161         gconf->debug = 0;
1162     } else if (gconf->debug > GLUSTER_DEBUG_MAX) {
1163         gconf->debug = GLUSTER_DEBUG_MAX;
1164     }
1165     gconf->has_debug = true;
1166 
1167     gconf->logfile = qemu_opt_get_del(opts, GLUSTER_OPT_LOGFILE);
1168     if (!gconf->logfile) {
1169         gconf->logfile = g_strdup(GLUSTER_LOGFILE_DEFAULT);
1170     }
1171     gconf->has_logfile = true;
1172 
1173     ret = qemu_gluster_parse(gconf, filename, NULL, errp);
1174     if (ret < 0) {
1175         goto fail;
1176     }
1177 
1178     ret = qemu_gluster_co_create(options, errp);
1179     if (ret < 0) {
1180         goto fail;
1181     }
1182 
1183     ret = 0;
1184 fail:
1185     qapi_free_BlockdevCreateOptions(options);
1186     return ret;
1187 }
1188 
1189 static coroutine_fn int qemu_gluster_co_rw(BlockDriverState *bs,
1190                                            int64_t sector_num, int nb_sectors,
1191                                            QEMUIOVector *qiov, int write)
1192 {
1193     int ret;
1194     GlusterAIOCB acb;
1195     BDRVGlusterState *s = bs->opaque;
1196     size_t size = nb_sectors * BDRV_SECTOR_SIZE;
1197     off_t offset = sector_num * BDRV_SECTOR_SIZE;
1198 
1199     acb.size = size;
1200     acb.ret = 0;
1201     acb.coroutine = qemu_coroutine_self();
1202     acb.aio_context = bdrv_get_aio_context(bs);
1203 
1204     if (write) {
1205         ret = glfs_pwritev_async(s->fd, qiov->iov, qiov->niov, offset, 0,
1206                                  gluster_finish_aiocb, &acb);
1207     } else {
1208         ret = glfs_preadv_async(s->fd, qiov->iov, qiov->niov, offset, 0,
1209                                 gluster_finish_aiocb, &acb);
1210     }
1211 
1212     if (ret < 0) {
1213         return -errno;
1214     }
1215 
1216     qemu_coroutine_yield();
1217     return acb.ret;
1218 }
1219 
1220 static coroutine_fn int qemu_gluster_co_truncate(BlockDriverState *bs,
1221                                                  int64_t offset,
1222                                                  bool exact,
1223                                                  PreallocMode prealloc,
1224                                                  BdrvRequestFlags flags,
1225                                                  Error **errp)
1226 {
1227     BDRVGlusterState *s = bs->opaque;
1228     return qemu_gluster_do_truncate(s->fd, offset, prealloc, errp);
1229 }
1230 
1231 static coroutine_fn int qemu_gluster_co_readv(BlockDriverState *bs,
1232                                               int64_t sector_num,
1233                                               int nb_sectors,
1234                                               QEMUIOVector *qiov)
1235 {
1236     return qemu_gluster_co_rw(bs, sector_num, nb_sectors, qiov, 0);
1237 }
1238 
1239 static coroutine_fn int qemu_gluster_co_writev(BlockDriverState *bs,
1240                                                int64_t sector_num,
1241                                                int nb_sectors,
1242                                                QEMUIOVector *qiov,
1243                                                int flags)
1244 {
1245     assert(!flags);
1246     return qemu_gluster_co_rw(bs, sector_num, nb_sectors, qiov, 1);
1247 }
1248 
1249 static void qemu_gluster_close(BlockDriverState *bs)
1250 {
1251     BDRVGlusterState *s = bs->opaque;
1252 
1253     g_free(s->logfile);
1254     if (s->fd) {
1255         glfs_close(s->fd);
1256         s->fd = NULL;
1257     }
1258     glfs_clear_preopened(s->glfs);
1259 }
1260 
1261 static coroutine_fn int qemu_gluster_co_flush_to_disk(BlockDriverState *bs)
1262 {
1263     int ret;
1264     GlusterAIOCB acb;
1265     BDRVGlusterState *s = bs->opaque;
1266 
1267     acb.size = 0;
1268     acb.ret = 0;
1269     acb.coroutine = qemu_coroutine_self();
1270     acb.aio_context = bdrv_get_aio_context(bs);
1271 
1272     ret = glfs_fsync_async(s->fd, gluster_finish_aiocb, &acb);
1273     if (ret < 0) {
1274         ret = -errno;
1275         goto error;
1276     }
1277 
1278     qemu_coroutine_yield();
1279     if (acb.ret < 0) {
1280         ret = acb.ret;
1281         goto error;
1282     }
1283 
1284     return acb.ret;
1285 
1286 error:
1287     /* Some versions of Gluster (3.5.6 -> 3.5.8?) will not retain its cache
1288      * after a fsync failure, so we have no way of allowing the guest to safely
1289      * continue.  Gluster versions prior to 3.5.6 don't retain the cache
1290      * either, but will invalidate the fd on error, so this is again our only
1291      * option.
1292      *
1293      * The 'resync-failed-syncs-after-fsync' xlator option for the
1294      * write-behind cache will cause later gluster versions to retain its
1295      * cache after error, so long as the fd remains open.  However, we
1296      * currently have no way of knowing if this option is supported.
1297      *
1298      * TODO: Once gluster provides a way for us to determine if the option
1299      * is supported, bypass the closure and setting drv to NULL.  */
1300     qemu_gluster_close(bs);
1301     bs->drv = NULL;
1302     return ret;
1303 }
1304 
1305 #ifdef CONFIG_GLUSTERFS_DISCARD
1306 static coroutine_fn int qemu_gluster_co_pdiscard(BlockDriverState *bs,
1307                                                  int64_t offset, int size)
1308 {
1309     int ret;
1310     GlusterAIOCB acb;
1311     BDRVGlusterState *s = bs->opaque;
1312 
1313     acb.size = 0;
1314     acb.ret = 0;
1315     acb.coroutine = qemu_coroutine_self();
1316     acb.aio_context = bdrv_get_aio_context(bs);
1317 
1318     ret = glfs_discard_async(s->fd, offset, size, gluster_finish_aiocb, &acb);
1319     if (ret < 0) {
1320         return -errno;
1321     }
1322 
1323     qemu_coroutine_yield();
1324     return acb.ret;
1325 }
1326 #endif
1327 
1328 static int64_t qemu_gluster_getlength(BlockDriverState *bs)
1329 {
1330     BDRVGlusterState *s = bs->opaque;
1331     int64_t ret;
1332 
1333     ret = glfs_lseek(s->fd, 0, SEEK_END);
1334     if (ret < 0) {
1335         return -errno;
1336     } else {
1337         return ret;
1338     }
1339 }
1340 
1341 static int64_t qemu_gluster_allocated_file_size(BlockDriverState *bs)
1342 {
1343     BDRVGlusterState *s = bs->opaque;
1344     struct stat st;
1345     int ret;
1346 
1347     ret = glfs_fstat(s->fd, &st);
1348     if (ret < 0) {
1349         return -errno;
1350     } else {
1351         return st.st_blocks * 512;
1352     }
1353 }
1354 
1355 /*
1356  * Find allocation range in @bs around offset @start.
1357  * May change underlying file descriptor's file offset.
1358  * If @start is not in a hole, store @start in @data, and the
1359  * beginning of the next hole in @hole, and return 0.
1360  * If @start is in a non-trailing hole, store @start in @hole and the
1361  * beginning of the next non-hole in @data, and return 0.
1362  * If @start is in a trailing hole or beyond EOF, return -ENXIO.
1363  * If we can't find out, return a negative errno other than -ENXIO.
1364  *
1365  * (Shamefully copied from file-posix.c, only minuscule adaptions.)
1366  */
1367 static int find_allocation(BlockDriverState *bs, off_t start,
1368                            off_t *data, off_t *hole)
1369 {
1370     BDRVGlusterState *s = bs->opaque;
1371 
1372     if (!s->supports_seek_data) {
1373         goto exit;
1374     }
1375 
1376 #if defined SEEK_HOLE && defined SEEK_DATA
1377     off_t offs;
1378 
1379     /*
1380      * SEEK_DATA cases:
1381      * D1. offs == start: start is in data
1382      * D2. offs > start: start is in a hole, next data at offs
1383      * D3. offs < 0, errno = ENXIO: either start is in a trailing hole
1384      *                              or start is beyond EOF
1385      *     If the latter happens, the file has been truncated behind
1386      *     our back since we opened it.  All bets are off then.
1387      *     Treating like a trailing hole is simplest.
1388      * D4. offs < 0, errno != ENXIO: we learned nothing
1389      */
1390     offs = glfs_lseek(s->fd, start, SEEK_DATA);
1391     if (offs < 0) {
1392         return -errno;          /* D3 or D4 */
1393     }
1394 
1395     if (offs < start) {
1396         /* This is not a valid return by lseek().  We are safe to just return
1397          * -EIO in this case, and we'll treat it like D4. Unfortunately some
1398          *  versions of gluster server will return offs < start, so an assert
1399          *  here will unnecessarily abort QEMU. */
1400         return -EIO;
1401     }
1402 
1403     if (offs > start) {
1404         /* D2: in hole, next data at offs */
1405         *hole = start;
1406         *data = offs;
1407         return 0;
1408     }
1409 
1410     /* D1: in data, end not yet known */
1411 
1412     /*
1413      * SEEK_HOLE cases:
1414      * H1. offs == start: start is in a hole
1415      *     If this happens here, a hole has been dug behind our back
1416      *     since the previous lseek().
1417      * H2. offs > start: either start is in data, next hole at offs,
1418      *                   or start is in trailing hole, EOF at offs
1419      *     Linux treats trailing holes like any other hole: offs ==
1420      *     start.  Solaris seeks to EOF instead: offs > start (blech).
1421      *     If that happens here, a hole has been dug behind our back
1422      *     since the previous lseek().
1423      * H3. offs < 0, errno = ENXIO: start is beyond EOF
1424      *     If this happens, the file has been truncated behind our
1425      *     back since we opened it.  Treat it like a trailing hole.
1426      * H4. offs < 0, errno != ENXIO: we learned nothing
1427      *     Pretend we know nothing at all, i.e. "forget" about D1.
1428      */
1429     offs = glfs_lseek(s->fd, start, SEEK_HOLE);
1430     if (offs < 0) {
1431         return -errno;          /* D1 and (H3 or H4) */
1432     }
1433 
1434     if (offs < start) {
1435         /* This is not a valid return by lseek().  We are safe to just return
1436          * -EIO in this case, and we'll treat it like H4. Unfortunately some
1437          *  versions of gluster server will return offs < start, so an assert
1438          *  here will unnecessarily abort QEMU. */
1439         return -EIO;
1440     }
1441 
1442     if (offs > start) {
1443         /*
1444          * D1 and H2: either in data, next hole at offs, or it was in
1445          * data but is now in a trailing hole.  In the latter case,
1446          * all bets are off.  Treating it as if it there was data all
1447          * the way to EOF is safe, so simply do that.
1448          */
1449         *data = start;
1450         *hole = offs;
1451         return 0;
1452     }
1453 
1454     /* D1 and H1 */
1455     return -EBUSY;
1456 #endif
1457 
1458 exit:
1459     return -ENOTSUP;
1460 }
1461 
1462 /*
1463  * Returns the allocation status of the specified offset.
1464  *
1465  * The block layer guarantees 'offset' and 'bytes' are within bounds.
1466  *
1467  * 'pnum' is set to the number of bytes (including and immediately following
1468  * the specified offset) that are known to be in the same
1469  * allocated/unallocated state.
1470  *
1471  * 'bytes' is the max value 'pnum' should be set to.
1472  *
1473  * (Based on raw_co_block_status() from file-posix.c.)
1474  */
1475 static int coroutine_fn qemu_gluster_co_block_status(BlockDriverState *bs,
1476                                                      bool want_zero,
1477                                                      int64_t offset,
1478                                                      int64_t bytes,
1479                                                      int64_t *pnum,
1480                                                      int64_t *map,
1481                                                      BlockDriverState **file)
1482 {
1483     BDRVGlusterState *s = bs->opaque;
1484     off_t data = 0, hole = 0;
1485     int ret = -EINVAL;
1486 
1487     if (!s->fd) {
1488         return ret;
1489     }
1490 
1491     if (!want_zero) {
1492         *pnum = bytes;
1493         *map = offset;
1494         *file = bs;
1495         return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID;
1496     }
1497 
1498     ret = find_allocation(bs, offset, &data, &hole);
1499     if (ret == -ENXIO) {
1500         /* Trailing hole */
1501         *pnum = bytes;
1502         ret = BDRV_BLOCK_ZERO;
1503     } else if (ret < 0) {
1504         /* No info available, so pretend there are no holes */
1505         *pnum = bytes;
1506         ret = BDRV_BLOCK_DATA;
1507     } else if (data == offset) {
1508         /* On a data extent, compute bytes to the end of the extent,
1509          * possibly including a partial sector at EOF. */
1510         *pnum = MIN(bytes, hole - offset);
1511         ret = BDRV_BLOCK_DATA;
1512     } else {
1513         /* On a hole, compute bytes to the beginning of the next extent.  */
1514         assert(hole == offset);
1515         *pnum = MIN(bytes, data - offset);
1516         ret = BDRV_BLOCK_ZERO;
1517     }
1518 
1519     *map = offset;
1520     *file = bs;
1521 
1522     return ret | BDRV_BLOCK_OFFSET_VALID;
1523 }
1524 
1525 
1526 static const char *const gluster_strong_open_opts[] = {
1527     GLUSTER_OPT_VOLUME,
1528     GLUSTER_OPT_PATH,
1529     GLUSTER_OPT_TYPE,
1530     GLUSTER_OPT_SERVER_PATTERN,
1531     GLUSTER_OPT_HOST,
1532     GLUSTER_OPT_PORT,
1533     GLUSTER_OPT_TO,
1534     GLUSTER_OPT_IPV4,
1535     GLUSTER_OPT_IPV6,
1536     GLUSTER_OPT_SOCKET,
1537 
1538     NULL
1539 };
1540 
1541 static BlockDriver bdrv_gluster = {
1542     .format_name                  = "gluster",
1543     .protocol_name                = "gluster",
1544     .instance_size                = sizeof(BDRVGlusterState),
1545     .bdrv_needs_filename          = false,
1546     .bdrv_file_open               = qemu_gluster_open,
1547     .bdrv_reopen_prepare          = qemu_gluster_reopen_prepare,
1548     .bdrv_reopen_commit           = qemu_gluster_reopen_commit,
1549     .bdrv_reopen_abort            = qemu_gluster_reopen_abort,
1550     .bdrv_close                   = qemu_gluster_close,
1551     .bdrv_co_create               = qemu_gluster_co_create,
1552     .bdrv_co_create_opts          = qemu_gluster_co_create_opts,
1553     .bdrv_getlength               = qemu_gluster_getlength,
1554     .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size,
1555     .bdrv_co_truncate             = qemu_gluster_co_truncate,
1556     .bdrv_co_readv                = qemu_gluster_co_readv,
1557     .bdrv_co_writev               = qemu_gluster_co_writev,
1558     .bdrv_co_flush_to_disk        = qemu_gluster_co_flush_to_disk,
1559 #ifdef CONFIG_GLUSTERFS_DISCARD
1560     .bdrv_co_pdiscard             = qemu_gluster_co_pdiscard,
1561 #endif
1562 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1563     .bdrv_co_pwrite_zeroes        = qemu_gluster_co_pwrite_zeroes,
1564 #endif
1565     .bdrv_co_block_status         = qemu_gluster_co_block_status,
1566     .bdrv_refresh_limits          = qemu_gluster_refresh_limits,
1567     .create_opts                  = &qemu_gluster_create_opts,
1568     .strong_runtime_opts          = gluster_strong_open_opts,
1569 };
1570 
1571 static BlockDriver bdrv_gluster_tcp = {
1572     .format_name                  = "gluster",
1573     .protocol_name                = "gluster+tcp",
1574     .instance_size                = sizeof(BDRVGlusterState),
1575     .bdrv_needs_filename          = false,
1576     .bdrv_file_open               = qemu_gluster_open,
1577     .bdrv_reopen_prepare          = qemu_gluster_reopen_prepare,
1578     .bdrv_reopen_commit           = qemu_gluster_reopen_commit,
1579     .bdrv_reopen_abort            = qemu_gluster_reopen_abort,
1580     .bdrv_close                   = qemu_gluster_close,
1581     .bdrv_co_create               = qemu_gluster_co_create,
1582     .bdrv_co_create_opts          = qemu_gluster_co_create_opts,
1583     .bdrv_getlength               = qemu_gluster_getlength,
1584     .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size,
1585     .bdrv_co_truncate             = qemu_gluster_co_truncate,
1586     .bdrv_co_readv                = qemu_gluster_co_readv,
1587     .bdrv_co_writev               = qemu_gluster_co_writev,
1588     .bdrv_co_flush_to_disk        = qemu_gluster_co_flush_to_disk,
1589 #ifdef CONFIG_GLUSTERFS_DISCARD
1590     .bdrv_co_pdiscard             = qemu_gluster_co_pdiscard,
1591 #endif
1592 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1593     .bdrv_co_pwrite_zeroes        = qemu_gluster_co_pwrite_zeroes,
1594 #endif
1595     .bdrv_co_block_status         = qemu_gluster_co_block_status,
1596     .bdrv_refresh_limits          = qemu_gluster_refresh_limits,
1597     .create_opts                  = &qemu_gluster_create_opts,
1598     .strong_runtime_opts          = gluster_strong_open_opts,
1599 };
1600 
1601 static BlockDriver bdrv_gluster_unix = {
1602     .format_name                  = "gluster",
1603     .protocol_name                = "gluster+unix",
1604     .instance_size                = sizeof(BDRVGlusterState),
1605     .bdrv_needs_filename          = true,
1606     .bdrv_file_open               = qemu_gluster_open,
1607     .bdrv_reopen_prepare          = qemu_gluster_reopen_prepare,
1608     .bdrv_reopen_commit           = qemu_gluster_reopen_commit,
1609     .bdrv_reopen_abort            = qemu_gluster_reopen_abort,
1610     .bdrv_close                   = qemu_gluster_close,
1611     .bdrv_co_create               = qemu_gluster_co_create,
1612     .bdrv_co_create_opts          = qemu_gluster_co_create_opts,
1613     .bdrv_getlength               = qemu_gluster_getlength,
1614     .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size,
1615     .bdrv_co_truncate             = qemu_gluster_co_truncate,
1616     .bdrv_co_readv                = qemu_gluster_co_readv,
1617     .bdrv_co_writev               = qemu_gluster_co_writev,
1618     .bdrv_co_flush_to_disk        = qemu_gluster_co_flush_to_disk,
1619 #ifdef CONFIG_GLUSTERFS_DISCARD
1620     .bdrv_co_pdiscard             = qemu_gluster_co_pdiscard,
1621 #endif
1622 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1623     .bdrv_co_pwrite_zeroes        = qemu_gluster_co_pwrite_zeroes,
1624 #endif
1625     .bdrv_co_block_status         = qemu_gluster_co_block_status,
1626     .bdrv_refresh_limits          = qemu_gluster_refresh_limits,
1627     .create_opts                  = &qemu_gluster_create_opts,
1628     .strong_runtime_opts          = gluster_strong_open_opts,
1629 };
1630 
1631 /* rdma is deprecated (actually never supported for volfile fetch).
1632  * Let's maintain it for the protocol compatibility, to make sure things
1633  * won't break immediately. For now, gluster+rdma will fall back to gluster+tcp
1634  * protocol with a warning.
1635  * TODO: remove gluster+rdma interface support
1636  */
1637 static BlockDriver bdrv_gluster_rdma = {
1638     .format_name                  = "gluster",
1639     .protocol_name                = "gluster+rdma",
1640     .instance_size                = sizeof(BDRVGlusterState),
1641     .bdrv_needs_filename          = true,
1642     .bdrv_file_open               = qemu_gluster_open,
1643     .bdrv_reopen_prepare          = qemu_gluster_reopen_prepare,
1644     .bdrv_reopen_commit           = qemu_gluster_reopen_commit,
1645     .bdrv_reopen_abort            = qemu_gluster_reopen_abort,
1646     .bdrv_close                   = qemu_gluster_close,
1647     .bdrv_co_create               = qemu_gluster_co_create,
1648     .bdrv_co_create_opts          = qemu_gluster_co_create_opts,
1649     .bdrv_getlength               = qemu_gluster_getlength,
1650     .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size,
1651     .bdrv_co_truncate             = qemu_gluster_co_truncate,
1652     .bdrv_co_readv                = qemu_gluster_co_readv,
1653     .bdrv_co_writev               = qemu_gluster_co_writev,
1654     .bdrv_co_flush_to_disk        = qemu_gluster_co_flush_to_disk,
1655 #ifdef CONFIG_GLUSTERFS_DISCARD
1656     .bdrv_co_pdiscard             = qemu_gluster_co_pdiscard,
1657 #endif
1658 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1659     .bdrv_co_pwrite_zeroes        = qemu_gluster_co_pwrite_zeroes,
1660 #endif
1661     .bdrv_co_block_status         = qemu_gluster_co_block_status,
1662     .bdrv_refresh_limits          = qemu_gluster_refresh_limits,
1663     .create_opts                  = &qemu_gluster_create_opts,
1664     .strong_runtime_opts          = gluster_strong_open_opts,
1665 };
1666 
1667 static void bdrv_gluster_init(void)
1668 {
1669     bdrv_register(&bdrv_gluster_rdma);
1670     bdrv_register(&bdrv_gluster_unix);
1671     bdrv_register(&bdrv_gluster_tcp);
1672     bdrv_register(&bdrv_gluster);
1673 }
1674 
1675 block_init(bdrv_gluster_init);
1676