1 /*
2 * GlusterFS backend for QEMU
3 *
4 * Copyright (C) 2012 Bharata B Rao <bharata@linux.vnet.ibm.com>
5 *
6 * This work is licensed under the terms of the GNU GPL, version 2 or later.
7 * See the COPYING file in the top-level directory.
8 *
9 */
10
11 #include "qemu/osdep.h"
12 #include "qemu/units.h"
13 #include <glusterfs/api/glfs.h>
14 #include "block/block-io.h"
15 #include "block/block_int.h"
16 #include "block/qdict.h"
17 #include "qapi/error.h"
18 #include "qapi/qmp/qdict.h"
19 #include "qapi/qmp/qerror.h"
20 #include "qemu/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
glfs_set_preopened(const char * volume,glfs_t * fs)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
glfs_find_preopened(const char * volume)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
glfs_clear_preopened(glfs_t * fs)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
parse_volume_options(BlockdevOptionsGluster * gconf,const char * path)291 static int parse_volume_options(BlockdevOptionsGluster *gconf, const char *path)
292 {
293 const 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 */
qemu_gluster_parse_uri(BlockdevOptionsGluster * gconf,const char * filename)348 static int qemu_gluster_parse_uri(BlockdevOptionsGluster *gconf,
349 const char *filename)
350 {
351 g_autoptr(GUri) uri = g_uri_parse(filename, G_URI_FLAGS_NONE, NULL);
352 g_autoptr(GHashTable) qp = NULL;
353 SocketAddress *gsconf;
354 bool is_unix = false;
355 const char *uri_scheme, *uri_query, *uri_server;
356 int uri_port, ret;
357
358 if (!uri) {
359 return -EINVAL;
360 }
361
362 gsconf = g_new0(SocketAddress, 1);
363 QAPI_LIST_PREPEND(gconf->server, gsconf);
364
365 /* transport */
366 uri_scheme = g_uri_get_scheme(uri);
367 if (!uri_scheme || !strcmp(uri_scheme, "gluster")) {
368 gsconf->type = SOCKET_ADDRESS_TYPE_INET;
369 } else if (!strcmp(uri_scheme, "gluster+tcp")) {
370 gsconf->type = SOCKET_ADDRESS_TYPE_INET;
371 } else if (!strcmp(uri_scheme, "gluster+unix")) {
372 gsconf->type = SOCKET_ADDRESS_TYPE_UNIX;
373 is_unix = true;
374 } else {
375 return -EINVAL;
376 }
377
378 ret = parse_volume_options(gconf, g_uri_get_path(uri));
379 if (ret < 0) {
380 return ret;
381 }
382
383 uri_query = g_uri_get_query(uri);
384 if (uri_query) {
385 qp = g_uri_parse_params(uri_query, -1, "&", G_URI_PARAMS_NONE, NULL);
386 if (!qp) {
387 return -EINVAL;
388 }
389 ret = g_hash_table_size(qp);
390 if (ret > 1 || (is_unix && !ret) || (!is_unix && ret)) {
391 return -EINVAL;
392 }
393 }
394
395 uri_server = g_uri_get_host(uri);
396 uri_port = g_uri_get_port(uri);
397
398 if (is_unix) {
399 char *uri_socket = g_hash_table_lookup(qp, "socket");
400 if (uri_server || uri_port != -1 || !uri_socket) {
401 return -EINVAL;
402 }
403 gsconf->u.q_unix.path = g_strdup(uri_socket);
404 } else {
405 gsconf->u.inet.host = g_strdup(uri_server ? uri_server : "localhost");
406 if (uri_port > 0) {
407 gsconf->u.inet.port = g_strdup_printf("%d", uri_port);
408 } else {
409 gsconf->u.inet.port = g_strdup_printf("%d", GLUSTER_DEFAULT_PORT);
410 }
411 }
412
413 return 0;
414 }
415
qemu_gluster_glfs_init(BlockdevOptionsGluster * gconf,Error ** errp)416 static struct glfs *qemu_gluster_glfs_init(BlockdevOptionsGluster *gconf,
417 Error **errp)
418 {
419 struct glfs *glfs;
420 int ret;
421 int old_errno;
422 SocketAddressList *server;
423 uint64_t port;
424
425 glfs = glfs_find_preopened(gconf->volume);
426 if (glfs) {
427 return glfs;
428 }
429
430 glfs = glfs_new(gconf->volume);
431 if (!glfs) {
432 goto out;
433 }
434
435 glfs_set_preopened(gconf->volume, glfs);
436
437 for (server = gconf->server; server; server = server->next) {
438 switch (server->value->type) {
439 case SOCKET_ADDRESS_TYPE_UNIX:
440 ret = glfs_set_volfile_server(glfs, "unix",
441 server->value->u.q_unix.path, 0);
442 break;
443 case SOCKET_ADDRESS_TYPE_INET:
444 if (parse_uint_full(server->value->u.inet.port, 10, &port) < 0 ||
445 port > 65535) {
446 error_setg(errp, "'%s' is not a valid port number",
447 server->value->u.inet.port);
448 errno = EINVAL;
449 goto out;
450 }
451 ret = glfs_set_volfile_server(glfs, "tcp",
452 server->value->u.inet.host,
453 (int)port);
454 break;
455 case SOCKET_ADDRESS_TYPE_VSOCK:
456 case SOCKET_ADDRESS_TYPE_FD:
457 default:
458 abort();
459 }
460
461 if (ret < 0) {
462 goto out;
463 }
464 }
465
466 ret = glfs_set_logging(glfs, gconf->logfile, gconf->debug);
467 if (ret < 0) {
468 goto out;
469 }
470
471 ret = glfs_init(glfs);
472 if (ret) {
473 error_setg(errp, "Gluster connection for volume %s, path %s failed"
474 " to connect", gconf->volume, gconf->path);
475 for (server = gconf->server; server; server = server->next) {
476 if (server->value->type == SOCKET_ADDRESS_TYPE_UNIX) {
477 error_append_hint(errp, "hint: failed on socket %s ",
478 server->value->u.q_unix.path);
479 } else {
480 error_append_hint(errp, "hint: failed on host %s and port %s ",
481 server->value->u.inet.host,
482 server->value->u.inet.port);
483 }
484 }
485
486 error_append_hint(errp, "Please refer to gluster logs for more info\n");
487
488 /* glfs_init sometimes doesn't set errno although docs suggest that */
489 if (errno == 0) {
490 errno = EINVAL;
491 }
492
493 goto out;
494 }
495 return glfs;
496
497 out:
498 if (glfs) {
499 old_errno = errno;
500 glfs_clear_preopened(glfs);
501 errno = old_errno;
502 }
503 return NULL;
504 }
505
506 /*
507 * Convert the json formatted command line into qapi.
508 */
qemu_gluster_parse_json(BlockdevOptionsGluster * gconf,QDict * options,Error ** errp)509 static int qemu_gluster_parse_json(BlockdevOptionsGluster *gconf,
510 QDict *options, Error **errp)
511 {
512 QemuOpts *opts;
513 SocketAddress *gsconf = NULL;
514 SocketAddressList **tail;
515 QDict *backing_options = NULL;
516 Error *local_err = NULL;
517 const char *ptr;
518 int i, type, num_servers;
519
520 /* create opts info from runtime_json_opts list */
521 opts = qemu_opts_create(&runtime_json_opts, NULL, 0, &error_abort);
522 if (!qemu_opts_absorb_qdict(opts, options, errp)) {
523 goto out;
524 }
525
526 num_servers = qdict_array_entries(options, GLUSTER_OPT_SERVER_PATTERN);
527 if (num_servers < 1) {
528 error_setg(&local_err, QERR_MISSING_PARAMETER, "server");
529 goto out;
530 }
531
532 ptr = qemu_opt_get(opts, GLUSTER_OPT_VOLUME);
533 if (!ptr) {
534 error_setg(&local_err, QERR_MISSING_PARAMETER, GLUSTER_OPT_VOLUME);
535 goto out;
536 }
537 gconf->volume = g_strdup(ptr);
538
539 ptr = qemu_opt_get(opts, GLUSTER_OPT_PATH);
540 if (!ptr) {
541 error_setg(&local_err, QERR_MISSING_PARAMETER, GLUSTER_OPT_PATH);
542 goto out;
543 }
544 gconf->path = g_strdup(ptr);
545 qemu_opts_del(opts);
546 tail = &gconf->server;
547
548 for (i = 0; i < num_servers; i++) {
549 g_autofree char *str = g_strdup_printf(GLUSTER_OPT_SERVER_PATTERN"%d.",
550 i);
551 qdict_extract_subqdict(options, &backing_options, str);
552
553 /* create opts info from runtime_type_opts list */
554 opts = qemu_opts_create(&runtime_type_opts, NULL, 0, &error_abort);
555 if (!qemu_opts_absorb_qdict(opts, backing_options, errp)) {
556 goto out;
557 }
558
559 ptr = qemu_opt_get(opts, GLUSTER_OPT_TYPE);
560 if (!ptr) {
561 error_setg(&local_err, QERR_MISSING_PARAMETER, GLUSTER_OPT_TYPE);
562 error_append_hint(&local_err, GERR_INDEX_HINT, i);
563 goto out;
564
565 }
566 gsconf = g_new0(SocketAddress, 1);
567 if (!strcmp(ptr, "tcp")) {
568 ptr = "inet"; /* accept legacy "tcp" */
569 }
570 type = qapi_enum_parse(&SocketAddressType_lookup, ptr, -1, NULL);
571 if (type != SOCKET_ADDRESS_TYPE_INET
572 && type != SOCKET_ADDRESS_TYPE_UNIX) {
573 error_setg(&local_err,
574 "Parameter '%s' may be 'inet' or 'unix'",
575 GLUSTER_OPT_TYPE);
576 error_append_hint(&local_err, GERR_INDEX_HINT, i);
577 goto out;
578 }
579 gsconf->type = type;
580 qemu_opts_del(opts);
581
582 if (gsconf->type == SOCKET_ADDRESS_TYPE_INET) {
583 /* create opts info from runtime_inet_opts list */
584 opts = qemu_opts_create(&runtime_inet_opts, NULL, 0, &error_abort);
585 if (!qemu_opts_absorb_qdict(opts, backing_options, errp)) {
586 goto out;
587 }
588
589 ptr = qemu_opt_get(opts, GLUSTER_OPT_HOST);
590 if (!ptr) {
591 error_setg(&local_err, QERR_MISSING_PARAMETER,
592 GLUSTER_OPT_HOST);
593 error_append_hint(&local_err, GERR_INDEX_HINT, i);
594 goto out;
595 }
596 gsconf->u.inet.host = g_strdup(ptr);
597 ptr = qemu_opt_get(opts, GLUSTER_OPT_PORT);
598 if (!ptr) {
599 error_setg(&local_err, QERR_MISSING_PARAMETER,
600 GLUSTER_OPT_PORT);
601 error_append_hint(&local_err, GERR_INDEX_HINT, i);
602 goto out;
603 }
604 gsconf->u.inet.port = g_strdup(ptr);
605
606 /* defend for unsupported fields in InetSocketAddress,
607 * i.e. @ipv4, @ipv6 and @to
608 */
609 ptr = qemu_opt_get(opts, GLUSTER_OPT_TO);
610 if (ptr) {
611 gsconf->u.inet.has_to = true;
612 }
613 ptr = qemu_opt_get(opts, GLUSTER_OPT_IPV4);
614 if (ptr) {
615 gsconf->u.inet.has_ipv4 = true;
616 }
617 ptr = qemu_opt_get(opts, GLUSTER_OPT_IPV6);
618 if (ptr) {
619 gsconf->u.inet.has_ipv6 = true;
620 }
621 if (gsconf->u.inet.has_to) {
622 error_setg(&local_err, "Parameter 'to' not supported");
623 goto out;
624 }
625 if (gsconf->u.inet.has_ipv4 || gsconf->u.inet.has_ipv6) {
626 error_setg(&local_err, "Parameters 'ipv4/ipv6' not supported");
627 goto out;
628 }
629 qemu_opts_del(opts);
630 } else {
631 /* create opts info from runtime_unix_opts list */
632 opts = qemu_opts_create(&runtime_unix_opts, NULL, 0, &error_abort);
633 if (!qemu_opts_absorb_qdict(opts, backing_options, errp)) {
634 goto out;
635 }
636
637 ptr = qemu_opt_get(opts, GLUSTER_OPT_PATH);
638 if (!ptr) {
639 ptr = qemu_opt_get(opts, GLUSTER_OPT_SOCKET);
640 } else if (qemu_opt_get(opts, GLUSTER_OPT_SOCKET)) {
641 error_setg(&local_err,
642 "Conflicting parameters 'path' and 'socket'");
643 error_append_hint(&local_err, GERR_INDEX_HINT, i);
644 goto out;
645 }
646 if (!ptr) {
647 error_setg(&local_err, QERR_MISSING_PARAMETER,
648 GLUSTER_OPT_PATH);
649 error_append_hint(&local_err, GERR_INDEX_HINT, i);
650 goto out;
651 }
652 gsconf->u.q_unix.path = g_strdup(ptr);
653 qemu_opts_del(opts);
654 }
655
656 QAPI_LIST_APPEND(tail, gsconf);
657 gsconf = NULL;
658
659 qobject_unref(backing_options);
660 backing_options = NULL;
661 }
662
663 return 0;
664
665 out:
666 error_propagate(errp, local_err);
667 qapi_free_SocketAddress(gsconf);
668 qemu_opts_del(opts);
669 qobject_unref(backing_options);
670 errno = EINVAL;
671 return -errno;
672 }
673
674 /* Converts options given in @filename and the @options QDict into the QAPI
675 * object @gconf. */
qemu_gluster_parse(BlockdevOptionsGluster * gconf,const char * filename,QDict * options,Error ** errp)676 static int qemu_gluster_parse(BlockdevOptionsGluster *gconf,
677 const char *filename,
678 QDict *options, Error **errp)
679 {
680 int ret;
681 if (filename) {
682 ret = qemu_gluster_parse_uri(gconf, filename);
683 if (ret < 0) {
684 error_setg(errp, "invalid URI %s", filename);
685 error_append_hint(errp, "Usage: file=gluster[+transport]://"
686 "[host[:port]]volume/path[?socket=...]"
687 "[,file.debug=N]"
688 "[,file.logfile=/path/filename.log]\n");
689 return ret;
690 }
691 } else {
692 ret = qemu_gluster_parse_json(gconf, options, errp);
693 if (ret < 0) {
694 error_append_hint(errp, "Usage: "
695 "-drive driver=qcow2,file.driver=gluster,"
696 "file.volume=testvol,file.path=/path/a.qcow2"
697 "[,file.debug=9]"
698 "[,file.logfile=/path/filename.log],"
699 "file.server.0.type=inet,"
700 "file.server.0.host=1.2.3.4,"
701 "file.server.0.port=24007,"
702 "file.server.1.transport=unix,"
703 "file.server.1.path=/var/run/glusterd.socket ..."
704 "\n");
705 return ret;
706 }
707 }
708
709 return 0;
710 }
711
qemu_gluster_init(BlockdevOptionsGluster * gconf,const char * filename,QDict * options,Error ** errp)712 static struct glfs *qemu_gluster_init(BlockdevOptionsGluster *gconf,
713 const char *filename,
714 QDict *options, Error **errp)
715 {
716 int ret;
717
718 ret = qemu_gluster_parse(gconf, filename, options, errp);
719 if (ret < 0) {
720 errno = -ret;
721 return NULL;
722 }
723
724 return qemu_gluster_glfs_init(gconf, errp);
725 }
726
727 /*
728 * AIO callback routine called from GlusterFS thread.
729 */
gluster_finish_aiocb(struct glfs_fd * fd,ssize_t ret,struct glfs_stat * pre,struct glfs_stat * post,void * arg)730 static void gluster_finish_aiocb(struct glfs_fd *fd, ssize_t ret,
731 #ifdef CONFIG_GLUSTERFS_IOCB_HAS_STAT
732 struct glfs_stat *pre, struct glfs_stat *post,
733 #endif
734 void *arg)
735 {
736 GlusterAIOCB *acb = (GlusterAIOCB *)arg;
737
738 if (!ret || ret == acb->size) {
739 acb->ret = 0; /* Success */
740 } else if (ret < 0) {
741 acb->ret = -errno; /* Read/Write failed */
742 } else {
743 acb->ret = -EIO; /* Partial read/write - fail it */
744 }
745
746 aio_co_schedule(acb->aio_context, acb->coroutine);
747 }
748
qemu_gluster_parse_flags(int bdrv_flags,int * open_flags)749 static void qemu_gluster_parse_flags(int bdrv_flags, int *open_flags)
750 {
751 assert(open_flags != NULL);
752
753 *open_flags |= O_BINARY;
754
755 if (bdrv_flags & BDRV_O_RDWR) {
756 *open_flags |= O_RDWR;
757 } else {
758 *open_flags |= O_RDONLY;
759 }
760
761 if ((bdrv_flags & BDRV_O_NOCACHE)) {
762 *open_flags |= O_DIRECT;
763 }
764 }
765
766 /*
767 * Do SEEK_DATA/HOLE to detect if it is functional. Older broken versions of
768 * gfapi incorrectly return the current offset when SEEK_DATA/HOLE is used.
769 * - Corrected versions return -1 and set errno to EINVAL.
770 * - Versions that support SEEK_DATA/HOLE correctly, will return -1 and set
771 * errno to ENXIO when SEEK_DATA is called with a position of EOF.
772 */
qemu_gluster_test_seek(struct glfs_fd * fd)773 static bool qemu_gluster_test_seek(struct glfs_fd *fd)
774 {
775 off_t ret = 0;
776
777 #if defined SEEK_HOLE && defined SEEK_DATA
778 off_t eof;
779
780 eof = glfs_lseek(fd, 0, SEEK_END);
781 if (eof < 0) {
782 /* this should never occur */
783 return false;
784 }
785
786 /* this should always fail with ENXIO if SEEK_DATA is supported */
787 ret = glfs_lseek(fd, eof, SEEK_DATA);
788 #endif
789
790 return (ret < 0) && (errno == ENXIO);
791 }
792
qemu_gluster_open(BlockDriverState * bs,QDict * options,int bdrv_flags,Error ** errp)793 static int qemu_gluster_open(BlockDriverState *bs, QDict *options,
794 int bdrv_flags, Error **errp)
795 {
796 BDRVGlusterState *s = bs->opaque;
797 int open_flags = 0;
798 int ret = 0;
799 BlockdevOptionsGluster *gconf = NULL;
800 QemuOpts *opts;
801 const char *filename, *logfile;
802
803 opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
804 if (!qemu_opts_absorb_qdict(opts, options, errp)) {
805 ret = -EINVAL;
806 goto out;
807 }
808
809 warn_report_once("'gluster' is deprecated");
810
811 filename = qemu_opt_get(opts, GLUSTER_OPT_FILENAME);
812
813 s->debug = qemu_opt_get_number(opts, GLUSTER_OPT_DEBUG,
814 GLUSTER_DEBUG_DEFAULT);
815 if (s->debug < 0) {
816 s->debug = 0;
817 } else if (s->debug > GLUSTER_DEBUG_MAX) {
818 s->debug = GLUSTER_DEBUG_MAX;
819 }
820
821 gconf = g_new0(BlockdevOptionsGluster, 1);
822 gconf->debug = s->debug;
823 gconf->has_debug = true;
824
825 logfile = qemu_opt_get(opts, GLUSTER_OPT_LOGFILE);
826 s->logfile = g_strdup(logfile ? logfile : GLUSTER_LOGFILE_DEFAULT);
827
828 gconf->logfile = g_strdup(s->logfile);
829
830 s->glfs = qemu_gluster_init(gconf, filename, options, errp);
831 if (!s->glfs) {
832 ret = -errno;
833 goto out;
834 }
835
836 #ifdef CONFIG_GLUSTERFS_XLATOR_OPT
837 /* Without this, if fsync fails for a recoverable reason (for instance,
838 * ENOSPC), gluster will dump its cache, preventing retries. This means
839 * almost certain data loss. Not all gluster versions support the
840 * 'resync-failed-syncs-after-fsync' key value, but there is no way to
841 * discover during runtime if it is supported (this api returns success for
842 * unknown key/value pairs) */
843 ret = glfs_set_xlator_option(s->glfs, "*-write-behind",
844 "resync-failed-syncs-after-fsync",
845 "on");
846 if (ret < 0) {
847 error_setg_errno(errp, errno, "Unable to set xlator key/value pair");
848 ret = -errno;
849 goto out;
850 }
851 #endif
852
853 qemu_gluster_parse_flags(bdrv_flags, &open_flags);
854
855 s->fd = glfs_open(s->glfs, gconf->path, open_flags);
856 ret = s->fd ? 0 : -errno;
857
858 if (ret == -EACCES || ret == -EROFS) {
859 /* Try to degrade to read-only, but if it doesn't work, still use the
860 * normal error message. */
861 bdrv_graph_rdlock_main_loop();
862 if (bdrv_apply_auto_read_only(bs, NULL, NULL) == 0) {
863 open_flags = (open_flags & ~O_RDWR) | O_RDONLY;
864 s->fd = glfs_open(s->glfs, gconf->path, open_flags);
865 ret = s->fd ? 0 : -errno;
866 }
867 bdrv_graph_rdunlock_main_loop();
868 }
869
870 s->supports_seek_data = qemu_gluster_test_seek(s->fd);
871
872 out:
873 qemu_opts_del(opts);
874 qapi_free_BlockdevOptionsGluster(gconf);
875 if (!ret) {
876 return ret;
877 }
878 g_free(s->logfile);
879 if (s->fd) {
880 glfs_close(s->fd);
881 }
882
883 glfs_clear_preopened(s->glfs);
884
885 return ret;
886 }
887
qemu_gluster_refresh_limits(BlockDriverState * bs,Error ** errp)888 static void qemu_gluster_refresh_limits(BlockDriverState *bs, Error **errp)
889 {
890 bs->bl.max_transfer = GLUSTER_MAX_TRANSFER;
891 bs->bl.max_pdiscard = MIN(SIZE_MAX, INT64_MAX);
892 }
893
qemu_gluster_reopen_prepare(BDRVReopenState * state,BlockReopenQueue * queue,Error ** errp)894 static int qemu_gluster_reopen_prepare(BDRVReopenState *state,
895 BlockReopenQueue *queue, Error **errp)
896 {
897 int ret = 0;
898 BDRVGlusterState *s;
899 BDRVGlusterReopenState *reop_s;
900 BlockdevOptionsGluster *gconf;
901 int open_flags = 0;
902
903 assert(state != NULL);
904 assert(state->bs != NULL);
905
906 s = state->bs->opaque;
907
908 state->opaque = g_new0(BDRVGlusterReopenState, 1);
909 reop_s = state->opaque;
910
911 qemu_gluster_parse_flags(state->flags, &open_flags);
912
913 gconf = g_new0(BlockdevOptionsGluster, 1);
914 gconf->debug = s->debug;
915 gconf->has_debug = true;
916 gconf->logfile = g_strdup(s->logfile);
917
918 /*
919 * If 'state->bs->exact_filename' is empty, 'state->options' should contain
920 * the JSON parameters already parsed.
921 */
922 if (state->bs->exact_filename[0] != '\0') {
923 reop_s->glfs = qemu_gluster_init(gconf, state->bs->exact_filename, NULL,
924 errp);
925 } else {
926 reop_s->glfs = qemu_gluster_init(gconf, NULL, state->options, errp);
927 }
928 if (reop_s->glfs == NULL) {
929 ret = -errno;
930 goto exit;
931 }
932
933 #ifdef CONFIG_GLUSTERFS_XLATOR_OPT
934 ret = glfs_set_xlator_option(reop_s->glfs, "*-write-behind",
935 "resync-failed-syncs-after-fsync", "on");
936 if (ret < 0) {
937 error_setg_errno(errp, errno, "Unable to set xlator key/value pair");
938 ret = -errno;
939 goto exit;
940 }
941 #endif
942
943 reop_s->fd = glfs_open(reop_s->glfs, gconf->path, open_flags);
944 if (reop_s->fd == NULL) {
945 /* reops->glfs will be cleaned up in _abort */
946 ret = -errno;
947 goto exit;
948 }
949
950 exit:
951 /* state->opaque will be freed in either the _abort or _commit */
952 qapi_free_BlockdevOptionsGluster(gconf);
953 return ret;
954 }
955
qemu_gluster_reopen_commit(BDRVReopenState * state)956 static void qemu_gluster_reopen_commit(BDRVReopenState *state)
957 {
958 BDRVGlusterReopenState *reop_s = state->opaque;
959 BDRVGlusterState *s = state->bs->opaque;
960
961
962 /* close the old */
963 if (s->fd) {
964 glfs_close(s->fd);
965 }
966
967 glfs_clear_preopened(s->glfs);
968
969 /* use the newly opened image / connection */
970 s->fd = reop_s->fd;
971 s->glfs = reop_s->glfs;
972
973 g_free(state->opaque);
974 state->opaque = NULL;
975
976 return;
977 }
978
979
qemu_gluster_reopen_abort(BDRVReopenState * state)980 static void qemu_gluster_reopen_abort(BDRVReopenState *state)
981 {
982 BDRVGlusterReopenState *reop_s = state->opaque;
983
984 if (reop_s == NULL) {
985 return;
986 }
987
988 if (reop_s->fd) {
989 glfs_close(reop_s->fd);
990 }
991
992 glfs_clear_preopened(reop_s->glfs);
993
994 g_free(state->opaque);
995 state->opaque = NULL;
996
997 return;
998 }
999
1000 #ifdef CONFIG_GLUSTERFS_ZEROFILL
qemu_gluster_co_pwrite_zeroes(BlockDriverState * bs,int64_t offset,int64_t bytes,BdrvRequestFlags flags)1001 static coroutine_fn int qemu_gluster_co_pwrite_zeroes(BlockDriverState *bs,
1002 int64_t offset,
1003 int64_t bytes,
1004 BdrvRequestFlags flags)
1005 {
1006 int ret;
1007 GlusterAIOCB acb;
1008 BDRVGlusterState *s = bs->opaque;
1009
1010 acb.size = bytes;
1011 acb.ret = 0;
1012 acb.coroutine = qemu_coroutine_self();
1013 acb.aio_context = bdrv_get_aio_context(bs);
1014
1015 ret = glfs_zerofill_async(s->fd, offset, bytes, gluster_finish_aiocb, &acb);
1016 if (ret < 0) {
1017 return -errno;
1018 }
1019
1020 qemu_coroutine_yield();
1021 return acb.ret;
1022 }
1023 #endif
1024
qemu_gluster_do_truncate(struct glfs_fd * fd,int64_t offset,PreallocMode prealloc,Error ** errp)1025 static int qemu_gluster_do_truncate(struct glfs_fd *fd, int64_t offset,
1026 PreallocMode prealloc, Error **errp)
1027 {
1028 int64_t current_length;
1029
1030 current_length = glfs_lseek(fd, 0, SEEK_END);
1031 if (current_length < 0) {
1032 error_setg_errno(errp, errno, "Failed to determine current size");
1033 return -errno;
1034 }
1035
1036 if (current_length > offset && prealloc != PREALLOC_MODE_OFF) {
1037 error_setg(errp, "Cannot use preallocation for shrinking files");
1038 return -ENOTSUP;
1039 }
1040
1041 if (current_length == offset) {
1042 return 0;
1043 }
1044
1045 switch (prealloc) {
1046 #ifdef CONFIG_GLUSTERFS_FALLOCATE
1047 case PREALLOC_MODE_FALLOC:
1048 if (glfs_fallocate(fd, 0, current_length, offset - current_length)) {
1049 error_setg_errno(errp, errno, "Could not preallocate data");
1050 return -errno;
1051 }
1052 break;
1053 #endif /* CONFIG_GLUSTERFS_FALLOCATE */
1054 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1055 case PREALLOC_MODE_FULL:
1056 if (glfs_ftruncate(fd, offset)) {
1057 error_setg_errno(errp, errno, "Could not resize file");
1058 return -errno;
1059 }
1060 if (glfs_zerofill(fd, current_length, offset - current_length)) {
1061 error_setg_errno(errp, errno, "Could not zerofill the new area");
1062 return -errno;
1063 }
1064 break;
1065 #endif /* CONFIG_GLUSTERFS_ZEROFILL */
1066 case PREALLOC_MODE_OFF:
1067 if (glfs_ftruncate(fd, offset)) {
1068 error_setg_errno(errp, errno, "Could not resize file");
1069 return -errno;
1070 }
1071 break;
1072 default:
1073 error_setg(errp, "Unsupported preallocation mode: %s",
1074 PreallocMode_str(prealloc));
1075 return -EINVAL;
1076 }
1077
1078 return 0;
1079 }
1080
qemu_gluster_co_create(BlockdevCreateOptions * options,Error ** errp)1081 static int qemu_gluster_co_create(BlockdevCreateOptions *options,
1082 Error **errp)
1083 {
1084 BlockdevCreateOptionsGluster *opts = &options->u.gluster;
1085 struct glfs *glfs;
1086 struct glfs_fd *fd = NULL;
1087 int ret = 0;
1088
1089 assert(options->driver == BLOCKDEV_DRIVER_GLUSTER);
1090
1091 glfs = qemu_gluster_glfs_init(opts->location, errp);
1092 if (!glfs) {
1093 ret = -errno;
1094 goto out;
1095 }
1096
1097 fd = glfs_creat(glfs, opts->location->path,
1098 O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, S_IRUSR | S_IWUSR);
1099 if (!fd) {
1100 ret = -errno;
1101 goto out;
1102 }
1103
1104 ret = qemu_gluster_do_truncate(fd, opts->size, opts->preallocation, errp);
1105
1106 out:
1107 if (fd) {
1108 if (glfs_close(fd) != 0 && ret == 0) {
1109 ret = -errno;
1110 }
1111 }
1112 glfs_clear_preopened(glfs);
1113 return ret;
1114 }
1115
qemu_gluster_co_create_opts(BlockDriver * drv,const char * filename,QemuOpts * opts,Error ** errp)1116 static int coroutine_fn qemu_gluster_co_create_opts(BlockDriver *drv,
1117 const char *filename,
1118 QemuOpts *opts,
1119 Error **errp)
1120 {
1121 BlockdevCreateOptions *options;
1122 BlockdevCreateOptionsGluster *gopts;
1123 BlockdevOptionsGluster *gconf;
1124 char *tmp = NULL;
1125 Error *local_err = NULL;
1126 int ret;
1127
1128 options = g_new0(BlockdevCreateOptions, 1);
1129 options->driver = BLOCKDEV_DRIVER_GLUSTER;
1130 gopts = &options->u.gluster;
1131
1132 gconf = g_new0(BlockdevOptionsGluster, 1);
1133 gopts->location = gconf;
1134
1135 gopts->size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
1136 BDRV_SECTOR_SIZE);
1137
1138 tmp = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
1139 gopts->preallocation = qapi_enum_parse(&PreallocMode_lookup, tmp,
1140 PREALLOC_MODE_OFF, &local_err);
1141 g_free(tmp);
1142 if (local_err) {
1143 error_propagate(errp, local_err);
1144 ret = -EINVAL;
1145 goto fail;
1146 }
1147
1148 gconf->debug = qemu_opt_get_number_del(opts, GLUSTER_OPT_DEBUG,
1149 GLUSTER_DEBUG_DEFAULT);
1150 if (gconf->debug < 0) {
1151 gconf->debug = 0;
1152 } else if (gconf->debug > GLUSTER_DEBUG_MAX) {
1153 gconf->debug = GLUSTER_DEBUG_MAX;
1154 }
1155 gconf->has_debug = true;
1156
1157 gconf->logfile = qemu_opt_get_del(opts, GLUSTER_OPT_LOGFILE);
1158 if (!gconf->logfile) {
1159 gconf->logfile = g_strdup(GLUSTER_LOGFILE_DEFAULT);
1160 }
1161
1162 ret = qemu_gluster_parse(gconf, filename, NULL, errp);
1163 if (ret < 0) {
1164 goto fail;
1165 }
1166
1167 ret = qemu_gluster_co_create(options, errp);
1168 if (ret < 0) {
1169 goto fail;
1170 }
1171
1172 ret = 0;
1173 fail:
1174 qapi_free_BlockdevCreateOptions(options);
1175 return ret;
1176 }
1177
qemu_gluster_co_rw(BlockDriverState * bs,int64_t sector_num,int nb_sectors,QEMUIOVector * qiov,int write)1178 static coroutine_fn int qemu_gluster_co_rw(BlockDriverState *bs,
1179 int64_t sector_num, int nb_sectors,
1180 QEMUIOVector *qiov, int write)
1181 {
1182 int ret;
1183 GlusterAIOCB acb;
1184 BDRVGlusterState *s = bs->opaque;
1185 size_t size = nb_sectors * BDRV_SECTOR_SIZE;
1186 off_t offset = sector_num * BDRV_SECTOR_SIZE;
1187
1188 acb.size = size;
1189 acb.ret = 0;
1190 acb.coroutine = qemu_coroutine_self();
1191 acb.aio_context = bdrv_get_aio_context(bs);
1192
1193 if (write) {
1194 ret = glfs_pwritev_async(s->fd, qiov->iov, qiov->niov, offset, 0,
1195 gluster_finish_aiocb, &acb);
1196 } else {
1197 ret = glfs_preadv_async(s->fd, qiov->iov, qiov->niov, offset, 0,
1198 gluster_finish_aiocb, &acb);
1199 }
1200
1201 if (ret < 0) {
1202 return -errno;
1203 }
1204
1205 qemu_coroutine_yield();
1206 return acb.ret;
1207 }
1208
qemu_gluster_co_truncate(BlockDriverState * bs,int64_t offset,bool exact,PreallocMode prealloc,BdrvRequestFlags flags,Error ** errp)1209 static coroutine_fn int qemu_gluster_co_truncate(BlockDriverState *bs,
1210 int64_t offset,
1211 bool exact,
1212 PreallocMode prealloc,
1213 BdrvRequestFlags flags,
1214 Error **errp)
1215 {
1216 BDRVGlusterState *s = bs->opaque;
1217 return qemu_gluster_do_truncate(s->fd, offset, prealloc, errp);
1218 }
1219
qemu_gluster_co_readv(BlockDriverState * bs,int64_t sector_num,int nb_sectors,QEMUIOVector * qiov)1220 static coroutine_fn int qemu_gluster_co_readv(BlockDriverState *bs,
1221 int64_t sector_num,
1222 int nb_sectors,
1223 QEMUIOVector *qiov)
1224 {
1225 return qemu_gluster_co_rw(bs, sector_num, nb_sectors, qiov, 0);
1226 }
1227
qemu_gluster_co_writev(BlockDriverState * bs,int64_t sector_num,int nb_sectors,QEMUIOVector * qiov,int flags)1228 static coroutine_fn int qemu_gluster_co_writev(BlockDriverState *bs,
1229 int64_t sector_num,
1230 int nb_sectors,
1231 QEMUIOVector *qiov,
1232 int flags)
1233 {
1234 return qemu_gluster_co_rw(bs, sector_num, nb_sectors, qiov, 1);
1235 }
1236
qemu_gluster_close(BlockDriverState * bs)1237 static void qemu_gluster_close(BlockDriverState *bs)
1238 {
1239 BDRVGlusterState *s = bs->opaque;
1240
1241 g_free(s->logfile);
1242 if (s->fd) {
1243 glfs_close(s->fd);
1244 s->fd = NULL;
1245 }
1246 glfs_clear_preopened(s->glfs);
1247 }
1248
qemu_gluster_co_flush_to_disk(BlockDriverState * bs)1249 static coroutine_fn int qemu_gluster_co_flush_to_disk(BlockDriverState *bs)
1250 {
1251 int ret;
1252 GlusterAIOCB acb;
1253 BDRVGlusterState *s = bs->opaque;
1254
1255 acb.size = 0;
1256 acb.ret = 0;
1257 acb.coroutine = qemu_coroutine_self();
1258 acb.aio_context = bdrv_get_aio_context(bs);
1259
1260 ret = glfs_fsync_async(s->fd, gluster_finish_aiocb, &acb);
1261 if (ret < 0) {
1262 ret = -errno;
1263 goto error;
1264 }
1265
1266 qemu_coroutine_yield();
1267 if (acb.ret < 0) {
1268 ret = acb.ret;
1269 goto error;
1270 }
1271
1272 return acb.ret;
1273
1274 error:
1275 /* Some versions of Gluster (3.5.6 -> 3.5.8?) will not retain its cache
1276 * after a fsync failure, so we have no way of allowing the guest to safely
1277 * continue. Gluster versions prior to 3.5.6 don't retain the cache
1278 * either, but will invalidate the fd on error, so this is again our only
1279 * option.
1280 *
1281 * The 'resync-failed-syncs-after-fsync' xlator option for the
1282 * write-behind cache will cause later gluster versions to retain its
1283 * cache after error, so long as the fd remains open. However, we
1284 * currently have no way of knowing if this option is supported.
1285 *
1286 * TODO: Once gluster provides a way for us to determine if the option
1287 * is supported, bypass the closure and setting drv to NULL. */
1288 qemu_gluster_close(bs);
1289 bs->drv = NULL;
1290 return ret;
1291 }
1292
1293 #ifdef CONFIG_GLUSTERFS_DISCARD
qemu_gluster_co_pdiscard(BlockDriverState * bs,int64_t offset,int64_t bytes)1294 static coroutine_fn int qemu_gluster_co_pdiscard(BlockDriverState *bs,
1295 int64_t offset, int64_t bytes)
1296 {
1297 int ret;
1298 GlusterAIOCB acb;
1299 BDRVGlusterState *s = bs->opaque;
1300
1301 assert(bytes <= SIZE_MAX); /* rely on max_pdiscard */
1302
1303 acb.size = 0;
1304 acb.ret = 0;
1305 acb.coroutine = qemu_coroutine_self();
1306 acb.aio_context = bdrv_get_aio_context(bs);
1307
1308 ret = glfs_discard_async(s->fd, offset, bytes, gluster_finish_aiocb, &acb);
1309 if (ret < 0) {
1310 return -errno;
1311 }
1312
1313 qemu_coroutine_yield();
1314 return acb.ret;
1315 }
1316 #endif
1317
qemu_gluster_co_getlength(BlockDriverState * bs)1318 static int64_t coroutine_fn qemu_gluster_co_getlength(BlockDriverState *bs)
1319 {
1320 BDRVGlusterState *s = bs->opaque;
1321 int64_t ret;
1322
1323 ret = glfs_lseek(s->fd, 0, SEEK_END);
1324 if (ret < 0) {
1325 return -errno;
1326 } else {
1327 return ret;
1328 }
1329 }
1330
1331 static int64_t coroutine_fn
qemu_gluster_co_get_allocated_file_size(BlockDriverState * bs)1332 qemu_gluster_co_get_allocated_file_size(BlockDriverState *bs)
1333 {
1334 BDRVGlusterState *s = bs->opaque;
1335 struct stat st;
1336 int ret;
1337
1338 ret = glfs_fstat(s->fd, &st);
1339 if (ret < 0) {
1340 return -errno;
1341 } else {
1342 return st.st_blocks * 512;
1343 }
1344 }
1345
1346 /*
1347 * Find allocation range in @bs around offset @start.
1348 * May change underlying file descriptor's file offset.
1349 * If @start is not in a hole, store @start in @data, and the
1350 * beginning of the next hole in @hole, and return 0.
1351 * If @start is in a non-trailing hole, store @start in @hole and the
1352 * beginning of the next non-hole in @data, and return 0.
1353 * If @start is in a trailing hole or beyond EOF, return -ENXIO.
1354 * If we can't find out, return a negative errno other than -ENXIO.
1355 *
1356 * (Shamefully copied from file-posix.c, only minuscule adaptions.)
1357 */
find_allocation(BlockDriverState * bs,off_t start,off_t * data,off_t * hole)1358 static int find_allocation(BlockDriverState *bs, off_t start,
1359 off_t *data, off_t *hole)
1360 {
1361 BDRVGlusterState *s = bs->opaque;
1362
1363 if (!s->supports_seek_data) {
1364 goto exit;
1365 }
1366
1367 #if defined SEEK_HOLE && defined SEEK_DATA
1368 off_t offs;
1369
1370 /*
1371 * SEEK_DATA cases:
1372 * D1. offs == start: start is in data
1373 * D2. offs > start: start is in a hole, next data at offs
1374 * D3. offs < 0, errno = ENXIO: either start is in a trailing hole
1375 * or start is beyond EOF
1376 * If the latter happens, the file has been truncated behind
1377 * our back since we opened it. All bets are off then.
1378 * Treating like a trailing hole is simplest.
1379 * D4. offs < 0, errno != ENXIO: we learned nothing
1380 */
1381 offs = glfs_lseek(s->fd, start, SEEK_DATA);
1382 if (offs < 0) {
1383 return -errno; /* D3 or D4 */
1384 }
1385
1386 if (offs < start) {
1387 /* This is not a valid return by lseek(). We are safe to just return
1388 * -EIO in this case, and we'll treat it like D4. Unfortunately some
1389 * versions of gluster server will return offs < start, so an assert
1390 * here will unnecessarily abort QEMU. */
1391 return -EIO;
1392 }
1393
1394 if (offs > start) {
1395 /* D2: in hole, next data at offs */
1396 *hole = start;
1397 *data = offs;
1398 return 0;
1399 }
1400
1401 /* D1: in data, end not yet known */
1402
1403 /*
1404 * SEEK_HOLE cases:
1405 * H1. offs == start: start is in a hole
1406 * If this happens here, a hole has been dug behind our back
1407 * since the previous lseek().
1408 * H2. offs > start: either start is in data, next hole at offs,
1409 * or start is in trailing hole, EOF at offs
1410 * Linux treats trailing holes like any other hole: offs ==
1411 * start. Solaris seeks to EOF instead: offs > start (blech).
1412 * If that happens here, a hole has been dug behind our back
1413 * since the previous lseek().
1414 * H3. offs < 0, errno = ENXIO: start is beyond EOF
1415 * If this happens, the file has been truncated behind our
1416 * back since we opened it. Treat it like a trailing hole.
1417 * H4. offs < 0, errno != ENXIO: we learned nothing
1418 * Pretend we know nothing at all, i.e. "forget" about D1.
1419 */
1420 offs = glfs_lseek(s->fd, start, SEEK_HOLE);
1421 if (offs < 0) {
1422 return -errno; /* D1 and (H3 or H4) */
1423 }
1424
1425 if (offs < start) {
1426 /* This is not a valid return by lseek(). We are safe to just return
1427 * -EIO in this case, and we'll treat it like H4. Unfortunately some
1428 * versions of gluster server will return offs < start, so an assert
1429 * here will unnecessarily abort QEMU. */
1430 return -EIO;
1431 }
1432
1433 if (offs > start) {
1434 /*
1435 * D1 and H2: either in data, next hole at offs, or it was in
1436 * data but is now in a trailing hole. In the latter case,
1437 * all bets are off. Treating it as if it there was data all
1438 * the way to EOF is safe, so simply do that.
1439 */
1440 *data = start;
1441 *hole = offs;
1442 return 0;
1443 }
1444
1445 /* D1 and H1 */
1446 return -EBUSY;
1447 #endif
1448
1449 exit:
1450 return -ENOTSUP;
1451 }
1452
1453 /*
1454 * Returns the allocation status of the specified offset.
1455 *
1456 * The block layer guarantees 'offset' and 'bytes' are within bounds.
1457 *
1458 * 'pnum' is set to the number of bytes (including and immediately following
1459 * the specified offset) that are known to be in the same
1460 * allocated/unallocated state.
1461 *
1462 * 'bytes' is a soft cap for 'pnum'. If the information is free, 'pnum' may
1463 * well exceed it.
1464 *
1465 * (Based on raw_co_block_status() from file-posix.c.)
1466 */
qemu_gluster_co_block_status(BlockDriverState * bs,bool want_zero,int64_t offset,int64_t bytes,int64_t * pnum,int64_t * map,BlockDriverState ** file)1467 static int coroutine_fn qemu_gluster_co_block_status(BlockDriverState *bs,
1468 bool want_zero,
1469 int64_t offset,
1470 int64_t bytes,
1471 int64_t *pnum,
1472 int64_t *map,
1473 BlockDriverState **file)
1474 {
1475 BDRVGlusterState *s = bs->opaque;
1476 off_t data = 0, hole = 0;
1477 int ret = -EINVAL;
1478
1479 assert(QEMU_IS_ALIGNED(offset | bytes, bs->bl.request_alignment));
1480
1481 if (!s->fd) {
1482 return ret;
1483 }
1484
1485 if (!want_zero) {
1486 *pnum = bytes;
1487 *map = offset;
1488 *file = bs;
1489 return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID;
1490 }
1491
1492 ret = find_allocation(bs, offset, &data, &hole);
1493 if (ret == -ENXIO) {
1494 /* Trailing hole */
1495 *pnum = bytes;
1496 ret = BDRV_BLOCK_ZERO;
1497 } else if (ret < 0) {
1498 /* No info available, so pretend there are no holes */
1499 *pnum = bytes;
1500 ret = BDRV_BLOCK_DATA;
1501 } else if (data == offset) {
1502 /* On a data extent, compute bytes to the end of the extent,
1503 * possibly including a partial sector at EOF. */
1504 *pnum = hole - offset;
1505
1506 /*
1507 * We are not allowed to return partial sectors, though, so
1508 * round up if necessary.
1509 */
1510 if (!QEMU_IS_ALIGNED(*pnum, bs->bl.request_alignment)) {
1511 int64_t file_length = qemu_gluster_co_getlength(bs);
1512 if (file_length > 0) {
1513 /* Ignore errors, this is just a safeguard */
1514 assert(hole == file_length);
1515 }
1516 *pnum = ROUND_UP(*pnum, bs->bl.request_alignment);
1517 }
1518
1519 ret = BDRV_BLOCK_DATA;
1520 } else {
1521 /* On a hole, compute bytes to the beginning of the next extent. */
1522 assert(hole == offset);
1523 *pnum = data - offset;
1524 ret = BDRV_BLOCK_ZERO;
1525 }
1526
1527 *map = offset;
1528 *file = bs;
1529
1530 return ret | BDRV_BLOCK_OFFSET_VALID;
1531 }
1532
1533
1534 static const char *const gluster_strong_open_opts[] = {
1535 GLUSTER_OPT_VOLUME,
1536 GLUSTER_OPT_PATH,
1537 GLUSTER_OPT_TYPE,
1538 GLUSTER_OPT_SERVER_PATTERN,
1539 GLUSTER_OPT_HOST,
1540 GLUSTER_OPT_PORT,
1541 GLUSTER_OPT_TO,
1542 GLUSTER_OPT_IPV4,
1543 GLUSTER_OPT_IPV6,
1544 GLUSTER_OPT_SOCKET,
1545
1546 NULL
1547 };
1548
1549 static BlockDriver bdrv_gluster = {
1550 .format_name = "gluster",
1551 .protocol_name = "gluster",
1552 .instance_size = sizeof(BDRVGlusterState),
1553 .bdrv_open = qemu_gluster_open,
1554 .bdrv_reopen_prepare = qemu_gluster_reopen_prepare,
1555 .bdrv_reopen_commit = qemu_gluster_reopen_commit,
1556 .bdrv_reopen_abort = qemu_gluster_reopen_abort,
1557 .bdrv_close = qemu_gluster_close,
1558 .bdrv_co_create = qemu_gluster_co_create,
1559 .bdrv_co_create_opts = qemu_gluster_co_create_opts,
1560 .bdrv_co_getlength = qemu_gluster_co_getlength,
1561 .bdrv_co_get_allocated_file_size = qemu_gluster_co_get_allocated_file_size,
1562 .bdrv_co_truncate = qemu_gluster_co_truncate,
1563 .bdrv_co_readv = qemu_gluster_co_readv,
1564 .bdrv_co_writev = qemu_gluster_co_writev,
1565 .bdrv_co_flush_to_disk = qemu_gluster_co_flush_to_disk,
1566 #ifdef CONFIG_GLUSTERFS_DISCARD
1567 .bdrv_co_pdiscard = qemu_gluster_co_pdiscard,
1568 #endif
1569 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1570 .bdrv_co_pwrite_zeroes = qemu_gluster_co_pwrite_zeroes,
1571 #endif
1572 .bdrv_co_block_status = qemu_gluster_co_block_status,
1573 .bdrv_refresh_limits = qemu_gluster_refresh_limits,
1574 .create_opts = &qemu_gluster_create_opts,
1575 .strong_runtime_opts = gluster_strong_open_opts,
1576 };
1577
1578 static BlockDriver bdrv_gluster_tcp = {
1579 .format_name = "gluster",
1580 .protocol_name = "gluster+tcp",
1581 .instance_size = sizeof(BDRVGlusterState),
1582 .bdrv_open = qemu_gluster_open,
1583 .bdrv_reopen_prepare = qemu_gluster_reopen_prepare,
1584 .bdrv_reopen_commit = qemu_gluster_reopen_commit,
1585 .bdrv_reopen_abort = qemu_gluster_reopen_abort,
1586 .bdrv_close = qemu_gluster_close,
1587 .bdrv_co_create = qemu_gluster_co_create,
1588 .bdrv_co_create_opts = qemu_gluster_co_create_opts,
1589 .bdrv_co_getlength = qemu_gluster_co_getlength,
1590 .bdrv_co_get_allocated_file_size = qemu_gluster_co_get_allocated_file_size,
1591 .bdrv_co_truncate = qemu_gluster_co_truncate,
1592 .bdrv_co_readv = qemu_gluster_co_readv,
1593 .bdrv_co_writev = qemu_gluster_co_writev,
1594 .bdrv_co_flush_to_disk = qemu_gluster_co_flush_to_disk,
1595 #ifdef CONFIG_GLUSTERFS_DISCARD
1596 .bdrv_co_pdiscard = qemu_gluster_co_pdiscard,
1597 #endif
1598 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1599 .bdrv_co_pwrite_zeroes = qemu_gluster_co_pwrite_zeroes,
1600 #endif
1601 .bdrv_co_block_status = qemu_gluster_co_block_status,
1602 .bdrv_refresh_limits = qemu_gluster_refresh_limits,
1603 .create_opts = &qemu_gluster_create_opts,
1604 .strong_runtime_opts = gluster_strong_open_opts,
1605 };
1606
1607 static BlockDriver bdrv_gluster_unix = {
1608 .format_name = "gluster",
1609 .protocol_name = "gluster+unix",
1610 .instance_size = sizeof(BDRVGlusterState),
1611 .bdrv_open = qemu_gluster_open,
1612 .bdrv_reopen_prepare = qemu_gluster_reopen_prepare,
1613 .bdrv_reopen_commit = qemu_gluster_reopen_commit,
1614 .bdrv_reopen_abort = qemu_gluster_reopen_abort,
1615 .bdrv_close = qemu_gluster_close,
1616 .bdrv_co_create = qemu_gluster_co_create,
1617 .bdrv_co_create_opts = qemu_gluster_co_create_opts,
1618 .bdrv_co_getlength = qemu_gluster_co_getlength,
1619 .bdrv_co_get_allocated_file_size = qemu_gluster_co_get_allocated_file_size,
1620 .bdrv_co_truncate = qemu_gluster_co_truncate,
1621 .bdrv_co_readv = qemu_gluster_co_readv,
1622 .bdrv_co_writev = qemu_gluster_co_writev,
1623 .bdrv_co_flush_to_disk = qemu_gluster_co_flush_to_disk,
1624 #ifdef CONFIG_GLUSTERFS_DISCARD
1625 .bdrv_co_pdiscard = qemu_gluster_co_pdiscard,
1626 #endif
1627 #ifdef CONFIG_GLUSTERFS_ZEROFILL
1628 .bdrv_co_pwrite_zeroes = qemu_gluster_co_pwrite_zeroes,
1629 #endif
1630 .bdrv_co_block_status = qemu_gluster_co_block_status,
1631 .bdrv_refresh_limits = qemu_gluster_refresh_limits,
1632 .create_opts = &qemu_gluster_create_opts,
1633 .strong_runtime_opts = gluster_strong_open_opts,
1634 };
1635
bdrv_gluster_init(void)1636 static void bdrv_gluster_init(void)
1637 {
1638 bdrv_register(&bdrv_gluster_unix);
1639 bdrv_register(&bdrv_gluster_tcp);
1640 bdrv_register(&bdrv_gluster);
1641 }
1642
1643 block_init(bdrv_gluster_init);
1644