]> git.proxmox.com Git - mirror_qemu.git/blame - block/gluster.c
Merge remote-tracking branch 'remotes/gkurz/tags/for-upstream' into staging
[mirror_qemu.git] / block / gluster.c
CommitLineData
8d6d89cb
BR
1/*
2 * GlusterFS backend for QEMU
3 *
4 * Copyright (C) 2012 Bharata B Rao <bharata@linux.vnet.ibm.com>
5 *
85c09bc0
BR
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.
8d6d89cb 8 *
8d6d89cb 9 */
80c71a24 10#include "qemu/osdep.h"
8d6d89cb 11#include <glusterfs/api/glfs.h>
737e150e 12#include "block/block_int.h"
da34e65c 13#include "qapi/error.h"
6c7189bb 14#include "qapi/qmp/qerror.h"
1de7afc9 15#include "qemu/uri.h"
0552ff24 16#include "qemu/error-report.h"
c56ac33b 17#include "qemu/cutils.h"
8d6d89cb 18
f70c50c8 19#define GLUSTER_OPT_FILENAME "filename"
6c7189bb
PKK
20#define GLUSTER_OPT_VOLUME "volume"
21#define GLUSTER_OPT_PATH "path"
22#define GLUSTER_OPT_TYPE "type"
23#define GLUSTER_OPT_SERVER_PATTERN "server."
24#define GLUSTER_OPT_HOST "host"
25#define GLUSTER_OPT_PORT "port"
26#define GLUSTER_OPT_TO "to"
27#define GLUSTER_OPT_IPV4 "ipv4"
28#define GLUSTER_OPT_IPV6 "ipv6"
29#define GLUSTER_OPT_SOCKET "socket"
f70c50c8 30#define GLUSTER_OPT_DEBUG "debug"
7edac2dd 31#define GLUSTER_DEFAULT_PORT 24007
f70c50c8
PKK
32#define GLUSTER_DEBUG_DEFAULT 4
33#define GLUSTER_DEBUG_MAX 9
e9db8ff3
PKK
34#define GLUSTER_OPT_LOGFILE "logfile"
35#define GLUSTER_LOGFILE_DEFAULT "-" /* handled in libgfapi as /dev/stderr */
f70c50c8 36
6c7189bb 37#define GERR_INDEX_HINT "hint: check in 'server' array index '%d'\n"
f70c50c8 38
8d6d89cb 39typedef struct GlusterAIOCB {
8d6d89cb
BR
40 int64_t size;
41 int ret;
15744b0b 42 Coroutine *coroutine;
6ee50af2 43 AioContext *aio_context;
8d6d89cb
BR
44} GlusterAIOCB;
45
46typedef struct BDRVGlusterState {
47 struct glfs *glfs;
8d6d89cb 48 struct glfs_fd *fd;
e9db8ff3 49 char *logfile;
947eb203 50 bool supports_seek_data;
1a417e46 51 int debug;
8d6d89cb
BR
52} BDRVGlusterState;
53
f70c50c8
PKK
54typedef struct BDRVGlusterReopenState {
55 struct glfs *glfs;
56 struct glfs_fd *fd;
57} BDRVGlusterReopenState;
58
f70c50c8 59
6349c154
PKK
60typedef struct GlfsPreopened {
61 char *volume;
62 glfs_t *fs;
63 int ref;
64} GlfsPreopened;
65
66typedef struct ListElement {
67 QLIST_ENTRY(ListElement) list;
68 GlfsPreopened saved;
69} ListElement;
70
71static QLIST_HEAD(glfs_list, ListElement) glfs_list;
72
f70c50c8
PKK
73static QemuOptsList qemu_gluster_create_opts = {
74 .name = "qemu-gluster-create-opts",
75 .head = QTAILQ_HEAD_INITIALIZER(qemu_gluster_create_opts.head),
76 .desc = {
77 {
78 .name = BLOCK_OPT_SIZE,
79 .type = QEMU_OPT_SIZE,
80 .help = "Virtual disk size"
81 },
82 {
83 .name = BLOCK_OPT_PREALLOC,
84 .type = QEMU_OPT_STRING,
85 .help = "Preallocation mode (allowed values: off, full)"
86 },
87 {
88 .name = GLUSTER_OPT_DEBUG,
89 .type = QEMU_OPT_NUMBER,
90 .help = "Gluster log level, valid range is 0-9",
91 },
e9db8ff3
PKK
92 {
93 .name = GLUSTER_OPT_LOGFILE,
94 .type = QEMU_OPT_STRING,
95 .help = "Logfile path of libgfapi",
96 },
f70c50c8
PKK
97 { /* end of list */ }
98 }
99};
100
101static QemuOptsList runtime_opts = {
102 .name = "gluster",
103 .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
104 .desc = {
105 {
106 .name = GLUSTER_OPT_FILENAME,
107 .type = QEMU_OPT_STRING,
108 .help = "URL to the gluster image",
109 },
110 {
111 .name = GLUSTER_OPT_DEBUG,
112 .type = QEMU_OPT_NUMBER,
113 .help = "Gluster log level, valid range is 0-9",
114 },
e9db8ff3
PKK
115 {
116 .name = GLUSTER_OPT_LOGFILE,
117 .type = QEMU_OPT_STRING,
118 .help = "Logfile path of libgfapi",
119 },
f70c50c8
PKK
120 { /* end of list */ }
121 },
122};
123
6c7189bb
PKK
124static QemuOptsList runtime_json_opts = {
125 .name = "gluster_json",
126 .head = QTAILQ_HEAD_INITIALIZER(runtime_json_opts.head),
127 .desc = {
128 {
129 .name = GLUSTER_OPT_VOLUME,
130 .type = QEMU_OPT_STRING,
131 .help = "name of gluster volume where VM image resides",
132 },
133 {
134 .name = GLUSTER_OPT_PATH,
135 .type = QEMU_OPT_STRING,
136 .help = "absolute path to image file in gluster volume",
137 },
138 {
139 .name = GLUSTER_OPT_DEBUG,
140 .type = QEMU_OPT_NUMBER,
141 .help = "Gluster log level, valid range is 0-9",
142 },
143 { /* end of list */ }
144 },
145};
146
147static QemuOptsList runtime_type_opts = {
148 .name = "gluster_type",
149 .head = QTAILQ_HEAD_INITIALIZER(runtime_type_opts.head),
150 .desc = {
151 {
152 .name = GLUSTER_OPT_TYPE,
153 .type = QEMU_OPT_STRING,
c5f1ae3a 154 .help = "inet|unix",
6c7189bb
PKK
155 },
156 { /* end of list */ }
157 },
158};
159
160static QemuOptsList runtime_unix_opts = {
161 .name = "gluster_unix",
162 .head = QTAILQ_HEAD_INITIALIZER(runtime_unix_opts.head),
163 .desc = {
164 {
165 .name = GLUSTER_OPT_SOCKET,
166 .type = QEMU_OPT_STRING,
167 .help = "socket file path)",
168 },
169 { /* end of list */ }
170 },
171};
172
c5f1ae3a
MA
173static QemuOptsList runtime_inet_opts = {
174 .name = "gluster_inet",
175 .head = QTAILQ_HEAD_INITIALIZER(runtime_inet_opts.head),
6c7189bb
PKK
176 .desc = {
177 {
178 .name = GLUSTER_OPT_TYPE,
179 .type = QEMU_OPT_STRING,
c5f1ae3a 180 .help = "inet|unix",
6c7189bb
PKK
181 },
182 {
183 .name = GLUSTER_OPT_HOST,
184 .type = QEMU_OPT_STRING,
185 .help = "host address (hostname/ipv4/ipv6 addresses)",
186 },
187 {
188 .name = GLUSTER_OPT_PORT,
53d9837f 189 .type = QEMU_OPT_STRING,
6c7189bb
PKK
190 .help = "port number on which glusterd is listening (default 24007)",
191 },
192 {
193 .name = "to",
194 .type = QEMU_OPT_NUMBER,
195 .help = "max port number, not supported by gluster",
196 },
197 {
198 .name = "ipv4",
199 .type = QEMU_OPT_BOOL,
200 .help = "ipv4 bool value, not supported by gluster",
201 },
202 {
203 .name = "ipv6",
204 .type = QEMU_OPT_BOOL,
205 .help = "ipv6 bool value, not supported by gluster",
206 },
207 { /* end of list */ }
208 },
209};
f70c50c8 210
6349c154
PKK
211static void glfs_set_preopened(const char *volume, glfs_t *fs)
212{
213 ListElement *entry = NULL;
214
215 entry = g_new(ListElement, 1);
216
217 entry->saved.volume = g_strdup(volume);
218
219 entry->saved.fs = fs;
220 entry->saved.ref = 1;
221
222 QLIST_INSERT_HEAD(&glfs_list, entry, list);
223}
224
225static glfs_t *glfs_find_preopened(const char *volume)
226{
227 ListElement *entry = NULL;
228
229 QLIST_FOREACH(entry, &glfs_list, list) {
230 if (strcmp(entry->saved.volume, volume) == 0) {
231 entry->saved.ref++;
232 return entry->saved.fs;
233 }
234 }
235
236 return NULL;
237}
238
239static void glfs_clear_preopened(glfs_t *fs)
240{
241 ListElement *entry = NULL;
668c0e44 242 ListElement *next;
6349c154
PKK
243
244 if (fs == NULL) {
245 return;
246 }
247
668c0e44 248 QLIST_FOREACH_SAFE(entry, &glfs_list, list, next) {
6349c154
PKK
249 if (entry->saved.fs == fs) {
250 if (--entry->saved.ref) {
251 return;
252 }
253
254 QLIST_REMOVE(entry, list);
255
256 glfs_fini(entry->saved.fs);
257 g_free(entry->saved.volume);
258 g_free(entry);
259 }
260 }
261}
262
7edac2dd 263static int parse_volume_options(BlockdevOptionsGluster *gconf, char *path)
8d6d89cb
BR
264{
265 char *p, *q;
266
267 if (!path) {
268 return -EINVAL;
269 }
270
271 /* volume */
272 p = q = path + strspn(path, "/");
273 p += strcspn(p, "/");
274 if (*p == '\0') {
275 return -EINVAL;
276 }
d5cf4079 277 gconf->volume = g_strndup(q, p - q);
8d6d89cb 278
d5cf4079 279 /* path */
8d6d89cb
BR
280 p += strspn(p, "/");
281 if (*p == '\0') {
282 return -EINVAL;
283 }
d5cf4079 284 gconf->path = g_strdup(p);
8d6d89cb
BR
285 return 0;
286}
287
288/*
d5cf4079 289 * file=gluster[+transport]://[host[:port]]/volume/path[?socket=...]
8d6d89cb
BR
290 *
291 * 'gluster' is the protocol.
292 *
293 * 'transport' specifies the transport type used to connect to gluster
294 * management daemon (glusterd). Valid transport types are
0552ff24 295 * tcp or unix. If a transport type isn't specified, then tcp type is assumed.
8d6d89cb 296 *
d5cf4079 297 * 'host' specifies the host where the volume file specification for
0552ff24 298 * the given volume resides. This can be either hostname or ipv4 address.
d5cf4079 299 * If transport type is 'unix', then 'host' field should not be specified.
8d6d89cb
BR
300 * The 'socket' field needs to be populated with the path to unix domain
301 * socket.
302 *
303 * 'port' is the port number on which glusterd is listening. This is optional
304 * and if not specified, QEMU will send 0 which will make gluster to use the
305 * default port. If the transport type is unix, then 'port' should not be
306 * specified.
307 *
d5cf4079 308 * 'volume' is the name of the gluster volume which contains the VM image.
8d6d89cb 309 *
d5cf4079 310 * 'path' is the path to the actual VM image that resides on gluster volume.
8d6d89cb
BR
311 *
312 * Examples:
313 *
314 * file=gluster://1.2.3.4/testvol/a.img
315 * file=gluster+tcp://1.2.3.4/testvol/a.img
316 * file=gluster+tcp://1.2.3.4:24007/testvol/dir/a.img
d5cf4079 317 * file=gluster+tcp://host.domain.com:24007/testvol/dir/a.img
8d6d89cb 318 * file=gluster+unix:///testvol/dir/a.img?socket=/tmp/glusterd.socket
8d6d89cb 319 */
7edac2dd
PKK
320static int qemu_gluster_parse_uri(BlockdevOptionsGluster *gconf,
321 const char *filename)
8d6d89cb 322{
62cf396b 323 SocketAddress *gsconf;
8d6d89cb
BR
324 URI *uri;
325 QueryParams *qp = NULL;
326 bool is_unix = false;
327 int ret = 0;
328
329 uri = uri_parse(filename);
330 if (!uri) {
331 return -EINVAL;
332 }
333
62cf396b
MA
334 gconf->server = g_new0(SocketAddressList, 1);
335 gconf->server->value = gsconf = g_new0(SocketAddress, 1);
7edac2dd 336
8d6d89cb 337 /* transport */
24897a76 338 if (!uri->scheme || !strcmp(uri->scheme, "gluster")) {
62cf396b 339 gsconf->type = SOCKET_ADDRESS_TYPE_INET;
8d6d89cb 340 } else if (!strcmp(uri->scheme, "gluster+tcp")) {
62cf396b 341 gsconf->type = SOCKET_ADDRESS_TYPE_INET;
8d6d89cb 342 } else if (!strcmp(uri->scheme, "gluster+unix")) {
62cf396b 343 gsconf->type = SOCKET_ADDRESS_TYPE_UNIX;
8d6d89cb
BR
344 is_unix = true;
345 } else if (!strcmp(uri->scheme, "gluster+rdma")) {
62cf396b 346 gsconf->type = SOCKET_ADDRESS_TYPE_INET;
3dc6f869 347 warn_report("rdma feature is not supported, falling back to tcp");
8d6d89cb
BR
348 } else {
349 ret = -EINVAL;
350 goto out;
351 }
352
353 ret = parse_volume_options(gconf, uri->path);
354 if (ret < 0) {
355 goto out;
356 }
357
358 qp = query_params_parse(uri->query);
359 if (qp->n > 1 || (is_unix && !qp->n) || (!is_unix && qp->n)) {
360 ret = -EINVAL;
361 goto out;
362 }
363
364 if (is_unix) {
365 if (uri->server || uri->port) {
366 ret = -EINVAL;
367 goto out;
368 }
369 if (strcmp(qp->p[0].name, "socket")) {
370 ret = -EINVAL;
371 goto out;
372 }
7edac2dd 373 gsconf->u.q_unix.path = g_strdup(qp->p[0].value);
8d6d89cb 374 } else {
c5f1ae3a 375 gsconf->u.inet.host = g_strdup(uri->server ? uri->server : "localhost");
7edac2dd 376 if (uri->port) {
c5f1ae3a 377 gsconf->u.inet.port = g_strdup_printf("%d", uri->port);
7edac2dd 378 } else {
c5f1ae3a 379 gsconf->u.inet.port = g_strdup_printf("%d", GLUSTER_DEFAULT_PORT);
7edac2dd 380 }
8d6d89cb
BR
381 }
382
383out:
384 if (qp) {
385 query_params_free(qp);
386 }
387 uri_free(uri);
388 return ret;
389}
390
6c7189bb
PKK
391static struct glfs *qemu_gluster_glfs_init(BlockdevOptionsGluster *gconf,
392 Error **errp)
8d6d89cb 393{
6c7189bb 394 struct glfs *glfs;
8d6d89cb
BR
395 int ret;
396 int old_errno;
62cf396b 397 SocketAddressList *server;
c56ac33b 398 unsigned long long port;
8d6d89cb 399
6349c154
PKK
400 glfs = glfs_find_preopened(gconf->volume);
401 if (glfs) {
402 return glfs;
403 }
404
d5cf4079 405 glfs = glfs_new(gconf->volume);
8d6d89cb
BR
406 if (!glfs) {
407 goto out;
408 }
409
6349c154
PKK
410 glfs_set_preopened(gconf->volume, glfs);
411
6c7189bb 412 for (server = gconf->server; server; server = server->next) {
fce5d538 413 switch (server->value->type) {
62cf396b 414 case SOCKET_ADDRESS_TYPE_UNIX:
fc29458d 415 ret = glfs_set_volfile_server(glfs, "unix",
6c7189bb 416 server->value->u.q_unix.path, 0);
fce5d538 417 break;
62cf396b 418 case SOCKET_ADDRESS_TYPE_INET:
c5f1ae3a 419 if (parse_uint_full(server->value->u.inet.port, &port, 10) < 0 ||
c56ac33b
PKK
420 port > 65535) {
421 error_setg(errp, "'%s' is not a valid port number",
c5f1ae3a 422 server->value->u.inet.port);
c56ac33b
PKK
423 errno = EINVAL;
424 goto out;
425 }
fc29458d 426 ret = glfs_set_volfile_server(glfs, "tcp",
c5f1ae3a 427 server->value->u.inet.host,
c56ac33b 428 (int)port);
fce5d538 429 break;
62cf396b
MA
430 case SOCKET_ADDRESS_TYPE_VSOCK:
431 case SOCKET_ADDRESS_TYPE_FD:
fce5d538
MA
432 default:
433 abort();
6c7189bb
PKK
434 }
435
436 if (ret < 0) {
437 goto out;
438 }
8d6d89cb
BR
439 }
440
1a417e46 441 ret = glfs_set_logging(glfs, gconf->logfile, gconf->debug);
8d6d89cb
BR
442 if (ret < 0) {
443 goto out;
444 }
445
446 ret = glfs_init(glfs);
447 if (ret) {
6c7189bb
PKK
448 error_setg(errp, "Gluster connection for volume %s, path %s failed"
449 " to connect", gconf->volume, gconf->path);
450 for (server = gconf->server; server; server = server->next) {
62cf396b 451 if (server->value->type == SOCKET_ADDRESS_TYPE_UNIX) {
6c7189bb
PKK
452 error_append_hint(errp, "hint: failed on socket %s ",
453 server->value->u.q_unix.path);
454 } else {
455 error_append_hint(errp, "hint: failed on host %s and port %s ",
c5f1ae3a
MA
456 server->value->u.inet.host,
457 server->value->u.inet.port);
6c7189bb 458 }
7edac2dd 459 }
4557117d 460
6c7189bb
PKK
461 error_append_hint(errp, "Please refer to gluster logs for more info\n");
462
4557117d 463 /* glfs_init sometimes doesn't set errno although docs suggest that */
7edac2dd 464 if (errno == 0) {
4557117d 465 errno = EINVAL;
7edac2dd 466 }
4557117d 467
8d6d89cb
BR
468 goto out;
469 }
470 return glfs;
471
472out:
473 if (glfs) {
474 old_errno = errno;
6349c154 475 glfs_clear_preopened(glfs);
8d6d89cb
BR
476 errno = old_errno;
477 }
478 return NULL;
479}
480
6c7189bb
PKK
481/*
482 * Convert the json formatted command line into qapi.
483*/
484static int qemu_gluster_parse_json(BlockdevOptionsGluster *gconf,
485 QDict *options, Error **errp)
486{
487 QemuOpts *opts;
62cf396b
MA
488 SocketAddress *gsconf = NULL;
489 SocketAddressList *curr = NULL;
6c7189bb
PKK
490 QDict *backing_options = NULL;
491 Error *local_err = NULL;
492 char *str = NULL;
493 const char *ptr;
56faeb9b 494 int i, type, num_servers;
6c7189bb
PKK
495
496 /* create opts info from runtime_json_opts list */
497 opts = qemu_opts_create(&runtime_json_opts, NULL, 0, &error_abort);
498 qemu_opts_absorb_qdict(opts, options, &local_err);
499 if (local_err) {
500 goto out;
501 }
502
503 num_servers = qdict_array_entries(options, GLUSTER_OPT_SERVER_PATTERN);
504 if (num_servers < 1) {
505 error_setg(&local_err, QERR_MISSING_PARAMETER, "server");
506 goto out;
507 }
508
509 ptr = qemu_opt_get(opts, GLUSTER_OPT_VOLUME);
510 if (!ptr) {
511 error_setg(&local_err, QERR_MISSING_PARAMETER, GLUSTER_OPT_VOLUME);
512 goto out;
513 }
514 gconf->volume = g_strdup(ptr);
515
516 ptr = qemu_opt_get(opts, GLUSTER_OPT_PATH);
517 if (!ptr) {
518 error_setg(&local_err, QERR_MISSING_PARAMETER, GLUSTER_OPT_PATH);
519 goto out;
520 }
521 gconf->path = g_strdup(ptr);
522 qemu_opts_del(opts);
523
524 for (i = 0; i < num_servers; i++) {
525 str = g_strdup_printf(GLUSTER_OPT_SERVER_PATTERN"%d.", i);
526 qdict_extract_subqdict(options, &backing_options, str);
527
528 /* create opts info from runtime_type_opts list */
529 opts = qemu_opts_create(&runtime_type_opts, NULL, 0, &error_abort);
530 qemu_opts_absorb_qdict(opts, backing_options, &local_err);
531 if (local_err) {
532 goto out;
533 }
534
535 ptr = qemu_opt_get(opts, GLUSTER_OPT_TYPE);
6c7189bb
PKK
536 if (!ptr) {
537 error_setg(&local_err, QERR_MISSING_PARAMETER, GLUSTER_OPT_TYPE);
538 error_append_hint(&local_err, GERR_INDEX_HINT, i);
539 goto out;
540
541 }
62cf396b 542 gsconf = g_new0(SocketAddress, 1);
c5f1ae3a
MA
543 if (!strcmp(ptr, "tcp")) {
544 ptr = "inet"; /* accept legacy "tcp" */
545 }
f7abe0ec 546 type = qapi_enum_parse(&SocketAddressType_lookup, ptr, -1, NULL);
62cf396b
MA
547 if (type != SOCKET_ADDRESS_TYPE_INET
548 && type != SOCKET_ADDRESS_TYPE_UNIX) {
fce5d538
MA
549 error_setg(&local_err,
550 "Parameter '%s' may be 'inet' or 'unix'",
551 GLUSTER_OPT_TYPE);
6c7189bb
PKK
552 error_append_hint(&local_err, GERR_INDEX_HINT, i);
553 goto out;
554 }
fce5d538 555 gsconf->type = type;
6c7189bb
PKK
556 qemu_opts_del(opts);
557
62cf396b 558 if (gsconf->type == SOCKET_ADDRESS_TYPE_INET) {
c5f1ae3a
MA
559 /* create opts info from runtime_inet_opts list */
560 opts = qemu_opts_create(&runtime_inet_opts, NULL, 0, &error_abort);
6c7189bb
PKK
561 qemu_opts_absorb_qdict(opts, backing_options, &local_err);
562 if (local_err) {
563 goto out;
564 }
565
566 ptr = qemu_opt_get(opts, GLUSTER_OPT_HOST);
567 if (!ptr) {
568 error_setg(&local_err, QERR_MISSING_PARAMETER,
569 GLUSTER_OPT_HOST);
570 error_append_hint(&local_err, GERR_INDEX_HINT, i);
571 goto out;
572 }
c5f1ae3a 573 gsconf->u.inet.host = g_strdup(ptr);
6c7189bb
PKK
574 ptr = qemu_opt_get(opts, GLUSTER_OPT_PORT);
575 if (!ptr) {
576 error_setg(&local_err, QERR_MISSING_PARAMETER,
577 GLUSTER_OPT_PORT);
578 error_append_hint(&local_err, GERR_INDEX_HINT, i);
579 goto out;
580 }
c5f1ae3a 581 gsconf->u.inet.port = g_strdup(ptr);
6c7189bb
PKK
582
583 /* defend for unsupported fields in InetSocketAddress,
584 * i.e. @ipv4, @ipv6 and @to
585 */
586 ptr = qemu_opt_get(opts, GLUSTER_OPT_TO);
587 if (ptr) {
c5f1ae3a 588 gsconf->u.inet.has_to = true;
6c7189bb
PKK
589 }
590 ptr = qemu_opt_get(opts, GLUSTER_OPT_IPV4);
591 if (ptr) {
c5f1ae3a 592 gsconf->u.inet.has_ipv4 = true;
6c7189bb
PKK
593 }
594 ptr = qemu_opt_get(opts, GLUSTER_OPT_IPV6);
595 if (ptr) {
c5f1ae3a 596 gsconf->u.inet.has_ipv6 = true;
6c7189bb 597 }
c5f1ae3a 598 if (gsconf->u.inet.has_to) {
6c7189bb
PKK
599 error_setg(&local_err, "Parameter 'to' not supported");
600 goto out;
601 }
c5f1ae3a 602 if (gsconf->u.inet.has_ipv4 || gsconf->u.inet.has_ipv6) {
6c7189bb
PKK
603 error_setg(&local_err, "Parameters 'ipv4/ipv6' not supported");
604 goto out;
605 }
606 qemu_opts_del(opts);
607 } else {
608 /* create opts info from runtime_unix_opts list */
609 opts = qemu_opts_create(&runtime_unix_opts, NULL, 0, &error_abort);
610 qemu_opts_absorb_qdict(opts, backing_options, &local_err);
611 if (local_err) {
612 goto out;
613 }
614
615 ptr = qemu_opt_get(opts, GLUSTER_OPT_SOCKET);
616 if (!ptr) {
617 error_setg(&local_err, QERR_MISSING_PARAMETER,
618 GLUSTER_OPT_SOCKET);
619 error_append_hint(&local_err, GERR_INDEX_HINT, i);
620 goto out;
621 }
622 gsconf->u.q_unix.path = g_strdup(ptr);
623 qemu_opts_del(opts);
624 }
625
626 if (gconf->server == NULL) {
62cf396b 627 gconf->server = g_new0(SocketAddressList, 1);
6c7189bb
PKK
628 gconf->server->value = gsconf;
629 curr = gconf->server;
630 } else {
62cf396b 631 curr->next = g_new0(SocketAddressList, 1);
6c7189bb
PKK
632 curr->next->value = gsconf;
633 curr = curr->next;
634 }
85a82e85 635 gsconf = NULL;
6c7189bb 636
85a82e85
MA
637 QDECREF(backing_options);
638 backing_options = NULL;
6c7189bb
PKK
639 g_free(str);
640 str = NULL;
641 }
642
643 return 0;
644
645out:
646 error_propagate(errp, local_err);
62cf396b 647 qapi_free_SocketAddress(gsconf);
6c7189bb 648 qemu_opts_del(opts);
85a82e85
MA
649 g_free(str);
650 QDECREF(backing_options);
6c7189bb
PKK
651 errno = EINVAL;
652 return -errno;
653}
654
655static struct glfs *qemu_gluster_init(BlockdevOptionsGluster *gconf,
656 const char *filename,
657 QDict *options, Error **errp)
658{
659 int ret;
660 if (filename) {
661 ret = qemu_gluster_parse_uri(gconf, filename);
662 if (ret < 0) {
663 error_setg(errp, "invalid URI");
664 error_append_hint(errp, "Usage: file=gluster[+transport]://"
e9db8ff3
PKK
665 "[host[:port]]volume/path[?socket=...]"
666 "[,file.debug=N]"
667 "[,file.logfile=/path/filename.log]\n");
6c7189bb
PKK
668 errno = -ret;
669 return NULL;
670 }
671 } else {
672 ret = qemu_gluster_parse_json(gconf, options, errp);
673 if (ret < 0) {
674 error_append_hint(errp, "Usage: "
675 "-drive driver=qcow2,file.driver=gluster,"
676 "file.volume=testvol,file.path=/path/a.qcow2"
e9db8ff3
PKK
677 "[,file.debug=9]"
678 "[,file.logfile=/path/filename.log],"
c5f1ae3a 679 "file.server.0.type=inet,"
6c7189bb
PKK
680 "file.server.0.host=1.2.3.4,"
681 "file.server.0.port=24007,"
682 "file.server.1.transport=unix,"
683 "file.server.1.socket=/var/run/glusterd.socket ..."
684 "\n");
685 errno = -ret;
686 return NULL;
687 }
688
689 }
690
691 return qemu_gluster_glfs_init(gconf, errp);
692}
693
7c815372
BR
694/*
695 * AIO callback routine called from GlusterFS thread.
696 */
697static void gluster_finish_aiocb(struct glfs_fd *fd, ssize_t ret, void *arg)
698{
699 GlusterAIOCB *acb = (GlusterAIOCB *)arg;
700
701 if (!ret || ret == acb->size) {
702 acb->ret = 0; /* Success */
703 } else if (ret < 0) {
a8827453 704 acb->ret = -errno; /* Read/Write failed */
7c815372
BR
705 } else {
706 acb->ret = -EIO; /* Partial read/write - fail it */
707 }
708
1919631e 709 aio_co_schedule(acb->aio_context, acb->coroutine);
7c815372
BR
710}
711
1b37b344
JC
712static void qemu_gluster_parse_flags(int bdrv_flags, int *open_flags)
713{
714 assert(open_flags != NULL);
715
716 *open_flags |= O_BINARY;
717
718 if (bdrv_flags & BDRV_O_RDWR) {
719 *open_flags |= O_RDWR;
720 } else {
721 *open_flags |= O_RDONLY;
722 }
723
724 if ((bdrv_flags & BDRV_O_NOCACHE)) {
725 *open_flags |= O_DIRECT;
726 }
727}
728
947eb203
NV
729/*
730 * Do SEEK_DATA/HOLE to detect if it is functional. Older broken versions of
731 * gfapi incorrectly return the current offset when SEEK_DATA/HOLE is used.
732 * - Corrected versions return -1 and set errno to EINVAL.
733 * - Versions that support SEEK_DATA/HOLE correctly, will return -1 and set
734 * errno to ENXIO when SEEK_DATA is called with a position of EOF.
735 */
736static bool qemu_gluster_test_seek(struct glfs_fd *fd)
737{
d9b78974
JC
738 off_t ret = 0;
739
740#if defined SEEK_HOLE && defined SEEK_DATA
741 off_t eof;
947eb203
NV
742
743 eof = glfs_lseek(fd, 0, SEEK_END);
744 if (eof < 0) {
745 /* this should never occur */
746 return false;
747 }
748
749 /* this should always fail with ENXIO if SEEK_DATA is supported */
750 ret = glfs_lseek(fd, eof, SEEK_DATA);
d9b78974
JC
751#endif
752
947eb203
NV
753 return (ret < 0) && (errno == ENXIO);
754}
755
56d1b4d2 756static int qemu_gluster_open(BlockDriverState *bs, QDict *options,
015a1036 757 int bdrv_flags, Error **errp)
8d6d89cb
BR
758{
759 BDRVGlusterState *s = bs->opaque;
1b37b344 760 int open_flags = 0;
8d6d89cb 761 int ret = 0;
7edac2dd 762 BlockdevOptionsGluster *gconf = NULL;
b4894776
KW
763 QemuOpts *opts;
764 Error *local_err = NULL;
e9db8ff3 765 const char *filename, *logfile;
b4894776 766
87ea75d5 767 opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
b4894776 768 qemu_opts_absorb_qdict(opts, options, &local_err);
84d18f06 769 if (local_err) {
a7451cb8 770 error_propagate(errp, local_err);
b4894776
KW
771 ret = -EINVAL;
772 goto out;
773 }
774
7eac868a
JC
775 filename = qemu_opt_get(opts, GLUSTER_OPT_FILENAME);
776
1a417e46
PKK
777 s->debug = qemu_opt_get_number(opts, GLUSTER_OPT_DEBUG,
778 GLUSTER_DEBUG_DEFAULT);
779 if (s->debug < 0) {
780 s->debug = 0;
781 } else if (s->debug > GLUSTER_DEBUG_MAX) {
782 s->debug = GLUSTER_DEBUG_MAX;
7eac868a 783 }
b4894776 784
7edac2dd 785 gconf = g_new0(BlockdevOptionsGluster, 1);
1a417e46
PKK
786 gconf->debug = s->debug;
787 gconf->has_debug = true;
e9db8ff3
PKK
788
789 logfile = qemu_opt_get(opts, GLUSTER_OPT_LOGFILE);
790 s->logfile = g_strdup(logfile ? logfile : GLUSTER_LOGFILE_DEFAULT);
791
792 gconf->logfile = g_strdup(s->logfile);
793 gconf->has_logfile = true;
794
6c7189bb 795 s->glfs = qemu_gluster_init(gconf, filename, options, errp);
8d6d89cb
BR
796 if (!s->glfs) {
797 ret = -errno;
798 goto out;
799 }
800
d85fa9eb
JC
801#ifdef CONFIG_GLUSTERFS_XLATOR_OPT
802 /* Without this, if fsync fails for a recoverable reason (for instance,
803 * ENOSPC), gluster will dump its cache, preventing retries. This means
804 * almost certain data loss. Not all gluster versions support the
805 * 'resync-failed-syncs-after-fsync' key value, but there is no way to
806 * discover during runtime if it is supported (this api returns success for
807 * unknown key/value pairs) */
808 ret = glfs_set_xlator_option(s->glfs, "*-write-behind",
809 "resync-failed-syncs-after-fsync",
810 "on");
811 if (ret < 0) {
812 error_setg_errno(errp, errno, "Unable to set xlator key/value pair");
813 ret = -errno;
814 goto out;
815 }
816#endif
817
1b37b344 818 qemu_gluster_parse_flags(bdrv_flags, &open_flags);
8d6d89cb 819
d5cf4079 820 s->fd = glfs_open(s->glfs, gconf->path, open_flags);
8d6d89cb
BR
821 if (!s->fd) {
822 ret = -errno;
8d6d89cb 823 }
8d6d89cb 824
947eb203
NV
825 s->supports_seek_data = qemu_gluster_test_seek(s->fd);
826
8d6d89cb 827out:
b4894776 828 qemu_opts_del(opts);
7edac2dd 829 qapi_free_BlockdevOptionsGluster(gconf);
8d6d89cb
BR
830 if (!ret) {
831 return ret;
832 }
e9db8ff3 833 g_free(s->logfile);
8d6d89cb
BR
834 if (s->fd) {
835 glfs_close(s->fd);
836 }
6349c154
PKK
837
838 glfs_clear_preopened(s->glfs);
839
8d6d89cb
BR
840 return ret;
841}
842
adccfbcd
JC
843static int qemu_gluster_reopen_prepare(BDRVReopenState *state,
844 BlockReopenQueue *queue, Error **errp)
845{
846 int ret = 0;
7eac868a 847 BDRVGlusterState *s;
adccfbcd 848 BDRVGlusterReopenState *reop_s;
7edac2dd 849 BlockdevOptionsGluster *gconf;
adccfbcd
JC
850 int open_flags = 0;
851
852 assert(state != NULL);
853 assert(state->bs != NULL);
854
7eac868a
JC
855 s = state->bs->opaque;
856
5839e53b 857 state->opaque = g_new0(BDRVGlusterReopenState, 1);
adccfbcd
JC
858 reop_s = state->opaque;
859
860 qemu_gluster_parse_flags(state->flags, &open_flags);
861
7edac2dd 862 gconf = g_new0(BlockdevOptionsGluster, 1);
1a417e46
PKK
863 gconf->debug = s->debug;
864 gconf->has_debug = true;
e9db8ff3
PKK
865 gconf->logfile = g_strdup(s->logfile);
866 gconf->has_logfile = true;
6c7189bb 867 reop_s->glfs = qemu_gluster_init(gconf, state->bs->filename, NULL, errp);
adccfbcd
JC
868 if (reop_s->glfs == NULL) {
869 ret = -errno;
870 goto exit;
871 }
872
d85fa9eb
JC
873#ifdef CONFIG_GLUSTERFS_XLATOR_OPT
874 ret = glfs_set_xlator_option(reop_s->glfs, "*-write-behind",
875 "resync-failed-syncs-after-fsync", "on");
876 if (ret < 0) {
877 error_setg_errno(errp, errno, "Unable to set xlator key/value pair");
878 ret = -errno;
879 goto exit;
880 }
881#endif
882
d5cf4079 883 reop_s->fd = glfs_open(reop_s->glfs, gconf->path, open_flags);
adccfbcd
JC
884 if (reop_s->fd == NULL) {
885 /* reops->glfs will be cleaned up in _abort */
886 ret = -errno;
887 goto exit;
888 }
889
890exit:
891 /* state->opaque will be freed in either the _abort or _commit */
7edac2dd 892 qapi_free_BlockdevOptionsGluster(gconf);
adccfbcd
JC
893 return ret;
894}
895
896static void qemu_gluster_reopen_commit(BDRVReopenState *state)
897{
898 BDRVGlusterReopenState *reop_s = state->opaque;
899 BDRVGlusterState *s = state->bs->opaque;
900
901
902 /* close the old */
903 if (s->fd) {
904 glfs_close(s->fd);
905 }
6349c154
PKK
906
907 glfs_clear_preopened(s->glfs);
adccfbcd
JC
908
909 /* use the newly opened image / connection */
910 s->fd = reop_s->fd;
911 s->glfs = reop_s->glfs;
912
913 g_free(state->opaque);
914 state->opaque = NULL;
915
916 return;
917}
918
919
920static void qemu_gluster_reopen_abort(BDRVReopenState *state)
921{
922 BDRVGlusterReopenState *reop_s = state->opaque;
923
924 if (reop_s == NULL) {
925 return;
926 }
927
928 if (reop_s->fd) {
929 glfs_close(reop_s->fd);
930 }
931
6349c154 932 glfs_clear_preopened(reop_s->glfs);
adccfbcd
JC
933
934 g_free(state->opaque);
935 state->opaque = NULL;
936
937 return;
938}
939
7c815372 940#ifdef CONFIG_GLUSTERFS_ZEROFILL
e88a36eb 941static coroutine_fn int qemu_gluster_co_pwrite_zeroes(BlockDriverState *bs,
f70c50c8
PKK
942 int64_t offset,
943 int size,
944 BdrvRequestFlags flags)
7c815372
BR
945{
946 int ret;
c833d1e8 947 GlusterAIOCB acb;
7c815372 948 BDRVGlusterState *s = bs->opaque;
7c815372 949
c833d1e8
PB
950 acb.size = size;
951 acb.ret = 0;
952 acb.coroutine = qemu_coroutine_self();
953 acb.aio_context = bdrv_get_aio_context(bs);
7c815372 954
c833d1e8 955 ret = glfs_zerofill_async(s->fd, offset, size, gluster_finish_aiocb, &acb);
7c815372 956 if (ret < 0) {
c833d1e8 957 return -errno;
7c815372
BR
958 }
959
960 qemu_coroutine_yield();
c833d1e8 961 return acb.ret;
7c815372
BR
962}
963#endif
964
8d6d89cb 965static int qemu_gluster_create(const char *filename,
90c772de 966 QemuOpts *opts, Error **errp)
8d6d89cb 967{
7edac2dd 968 BlockdevOptionsGluster *gconf;
8d6d89cb
BR
969 struct glfs *glfs;
970 struct glfs_fd *fd;
971 int ret = 0;
df3a429a 972 PreallocMode prealloc;
8d6d89cb 973 int64_t total_size = 0;
90c772de 974 char *tmp = NULL;
df3a429a 975 Error *local_err = NULL;
8d6d89cb 976
7edac2dd 977 gconf = g_new0(BlockdevOptionsGluster, 1);
1a417e46
PKK
978 gconf->debug = qemu_opt_get_number_del(opts, GLUSTER_OPT_DEBUG,
979 GLUSTER_DEBUG_DEFAULT);
980 if (gconf->debug < 0) {
981 gconf->debug = 0;
982 } else if (gconf->debug > GLUSTER_DEBUG_MAX) {
983 gconf->debug = GLUSTER_DEBUG_MAX;
984 }
985 gconf->has_debug = true;
7eac868a 986
e9db8ff3
PKK
987 gconf->logfile = qemu_opt_get_del(opts, GLUSTER_OPT_LOGFILE);
988 if (!gconf->logfile) {
989 gconf->logfile = g_strdup(GLUSTER_LOGFILE_DEFAULT);
990 }
991 gconf->has_logfile = true;
992
6c7189bb 993 glfs = qemu_gluster_init(gconf, filename, NULL, errp);
8d6d89cb 994 if (!glfs) {
4557117d 995 ret = -errno;
8d6d89cb
BR
996 goto out;
997 }
998
180e9526
HT
999 total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
1000 BDRV_SECTOR_SIZE);
90c772de
CL
1001
1002 tmp = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
f7abe0ec 1003 prealloc = qapi_enum_parse(&PreallocMode_lookup, tmp, PREALLOC_MODE_OFF,
df3a429a
NV
1004 &local_err);
1005 g_free(tmp);
1006 if (local_err) {
1007 error_propagate(errp, local_err);
90c772de
CL
1008 ret = -EINVAL;
1009 goto out;
8d6d89cb
BR
1010 }
1011
d5cf4079 1012 fd = glfs_creat(glfs, gconf->path,
f70c50c8 1013 O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, S_IRUSR | S_IWUSR);
8d6d89cb
BR
1014 if (!fd) {
1015 ret = -errno;
df3a429a
NV
1016 goto out;
1017 }
1018
1019 switch (prealloc) {
1020#ifdef CONFIG_GLUSTERFS_FALLOCATE
1021 case PREALLOC_MODE_FALLOC:
1022 if (glfs_fallocate(fd, 0, 0, total_size)) {
1023 error_setg(errp, "Could not preallocate data for the new file");
1024 ret = -errno;
1025 }
1026 break;
1027#endif /* CONFIG_GLUSTERFS_FALLOCATE */
1028#ifdef CONFIG_GLUSTERFS_ZEROFILL
1029 case PREALLOC_MODE_FULL:
180e9526 1030 if (!glfs_ftruncate(fd, total_size)) {
df3a429a
NV
1031 if (glfs_zerofill(fd, 0, total_size)) {
1032 error_setg(errp, "Could not zerofill the new file");
cf7f616b
BR
1033 ret = -errno;
1034 }
1035 } else {
df3a429a 1036 error_setg(errp, "Could not resize file");
8d6d89cb
BR
1037 ret = -errno;
1038 }
df3a429a
NV
1039 break;
1040#endif /* CONFIG_GLUSTERFS_ZEROFILL */
1041 case PREALLOC_MODE_OFF:
1042 if (glfs_ftruncate(fd, total_size) != 0) {
8d6d89cb 1043 ret = -errno;
df3a429a 1044 error_setg(errp, "Could not resize file");
8d6d89cb 1045 }
df3a429a
NV
1046 break;
1047 default:
1048 ret = -EINVAL;
1049 error_setg(errp, "Unsupported preallocation mode: %s",
977c736f 1050 PreallocMode_str(prealloc));
df3a429a
NV
1051 break;
1052 }
1053
1054 if (glfs_close(fd) != 0) {
1055 ret = -errno;
8d6d89cb
BR
1056 }
1057out:
7edac2dd 1058 qapi_free_BlockdevOptionsGluster(gconf);
6349c154 1059 glfs_clear_preopened(glfs);
8d6d89cb
BR
1060 return ret;
1061}
1062
15744b0b 1063static coroutine_fn int qemu_gluster_co_rw(BlockDriverState *bs,
f70c50c8
PKK
1064 int64_t sector_num, int nb_sectors,
1065 QEMUIOVector *qiov, int write)
8d6d89cb
BR
1066{
1067 int ret;
c833d1e8 1068 GlusterAIOCB acb;
8d6d89cb 1069 BDRVGlusterState *s = bs->opaque;
15744b0b
BR
1070 size_t size = nb_sectors * BDRV_SECTOR_SIZE;
1071 off_t offset = sector_num * BDRV_SECTOR_SIZE;
8d6d89cb 1072
c833d1e8
PB
1073 acb.size = size;
1074 acb.ret = 0;
1075 acb.coroutine = qemu_coroutine_self();
1076 acb.aio_context = bdrv_get_aio_context(bs);
8d6d89cb
BR
1077
1078 if (write) {
1079 ret = glfs_pwritev_async(s->fd, qiov->iov, qiov->niov, offset, 0,
f70c50c8 1080 gluster_finish_aiocb, &acb);
8d6d89cb
BR
1081 } else {
1082 ret = glfs_preadv_async(s->fd, qiov->iov, qiov->niov, offset, 0,
f70c50c8 1083 gluster_finish_aiocb, &acb);
8d6d89cb
BR
1084 }
1085
1086 if (ret < 0) {
c833d1e8 1087 return -errno;
8d6d89cb 1088 }
15744b0b
BR
1089
1090 qemu_coroutine_yield();
c833d1e8 1091 return acb.ret;
8d6d89cb
BR
1092}
1093
4bff28b8 1094static int qemu_gluster_truncate(BlockDriverState *bs, int64_t offset,
8243ccb7 1095 PreallocMode prealloc, Error **errp)
42ec24e2
PB
1096{
1097 int ret;
1098 BDRVGlusterState *s = bs->opaque;
1099
8243ccb7
HR
1100 if (prealloc != PREALLOC_MODE_OFF) {
1101 error_setg(errp, "Unsupported preallocation mode '%s'",
977c736f 1102 PreallocMode_str(prealloc));
8243ccb7
HR
1103 return -ENOTSUP;
1104 }
1105
42ec24e2
PB
1106 ret = glfs_ftruncate(s->fd, offset);
1107 if (ret < 0) {
f59adb32
HR
1108 ret = -errno;
1109 error_setg_errno(errp, -ret, "Failed to truncate file");
1110 return ret;
42ec24e2
PB
1111 }
1112
1113 return 0;
1114}
1115
15744b0b 1116static coroutine_fn int qemu_gluster_co_readv(BlockDriverState *bs,
f70c50c8
PKK
1117 int64_t sector_num,
1118 int nb_sectors,
1119 QEMUIOVector *qiov)
8d6d89cb 1120{
15744b0b 1121 return qemu_gluster_co_rw(bs, sector_num, nb_sectors, qiov, 0);
8d6d89cb
BR
1122}
1123
15744b0b 1124static coroutine_fn int qemu_gluster_co_writev(BlockDriverState *bs,
f70c50c8
PKK
1125 int64_t sector_num,
1126 int nb_sectors,
1127 QEMUIOVector *qiov)
8d6d89cb 1128{
15744b0b 1129 return qemu_gluster_co_rw(bs, sector_num, nb_sectors, qiov, 1);
8d6d89cb
BR
1130}
1131
5d4343e6
JC
1132static void qemu_gluster_close(BlockDriverState *bs)
1133{
1134 BDRVGlusterState *s = bs->opaque;
1135
e9db8ff3 1136 g_free(s->logfile);
5d4343e6
JC
1137 if (s->fd) {
1138 glfs_close(s->fd);
1139 s->fd = NULL;
1140 }
6349c154 1141 glfs_clear_preopened(s->glfs);
5d4343e6
JC
1142}
1143
15744b0b 1144static coroutine_fn int qemu_gluster_co_flush_to_disk(BlockDriverState *bs)
8d6d89cb
BR
1145{
1146 int ret;
c833d1e8 1147 GlusterAIOCB acb;
8d6d89cb
BR
1148 BDRVGlusterState *s = bs->opaque;
1149
c833d1e8
PB
1150 acb.size = 0;
1151 acb.ret = 0;
1152 acb.coroutine = qemu_coroutine_self();
1153 acb.aio_context = bdrv_get_aio_context(bs);
8d6d89cb 1154
c833d1e8 1155 ret = glfs_fsync_async(s->fd, gluster_finish_aiocb, &acb);
8d6d89cb 1156 if (ret < 0) {
d85fa9eb
JC
1157 ret = -errno;
1158 goto error;
8d6d89cb 1159 }
15744b0b
BR
1160
1161 qemu_coroutine_yield();
d85fa9eb
JC
1162 if (acb.ret < 0) {
1163 ret = acb.ret;
1164 goto error;
1165 }
1166
c833d1e8 1167 return acb.ret;
d85fa9eb
JC
1168
1169error:
1170 /* Some versions of Gluster (3.5.6 -> 3.5.8?) will not retain its cache
1171 * after a fsync failure, so we have no way of allowing the guest to safely
1172 * continue. Gluster versions prior to 3.5.6 don't retain the cache
1173 * either, but will invalidate the fd on error, so this is again our only
1174 * option.
1175 *
1176 * The 'resync-failed-syncs-after-fsync' xlator option for the
1177 * write-behind cache will cause later gluster versions to retain its
1178 * cache after error, so long as the fd remains open. However, we
1179 * currently have no way of knowing if this option is supported.
1180 *
1181 * TODO: Once gluster provides a way for us to determine if the option
1182 * is supported, bypass the closure and setting drv to NULL. */
1183 qemu_gluster_close(bs);
1184 bs->drv = NULL;
1185 return ret;
8d6d89cb
BR
1186}
1187
0c14fb47 1188#ifdef CONFIG_GLUSTERFS_DISCARD
1014170b
EB
1189static coroutine_fn int qemu_gluster_co_pdiscard(BlockDriverState *bs,
1190 int64_t offset, int size)
0c14fb47
BR
1191{
1192 int ret;
c833d1e8 1193 GlusterAIOCB acb;
0c14fb47 1194 BDRVGlusterState *s = bs->opaque;
0c14fb47 1195
c833d1e8
PB
1196 acb.size = 0;
1197 acb.ret = 0;
1198 acb.coroutine = qemu_coroutine_self();
1199 acb.aio_context = bdrv_get_aio_context(bs);
0c14fb47 1200
c833d1e8 1201 ret = glfs_discard_async(s->fd, offset, size, gluster_finish_aiocb, &acb);
0c14fb47 1202 if (ret < 0) {
c833d1e8 1203 return -errno;
0c14fb47 1204 }
15744b0b
BR
1205
1206 qemu_coroutine_yield();
c833d1e8 1207 return acb.ret;
0c14fb47
BR
1208}
1209#endif
1210
8d6d89cb
BR
1211static int64_t qemu_gluster_getlength(BlockDriverState *bs)
1212{
1213 BDRVGlusterState *s = bs->opaque;
1214 int64_t ret;
1215
1216 ret = glfs_lseek(s->fd, 0, SEEK_END);
1217 if (ret < 0) {
1218 return -errno;
1219 } else {
1220 return ret;
1221 }
1222}
1223
1224static int64_t qemu_gluster_allocated_file_size(BlockDriverState *bs)
1225{
1226 BDRVGlusterState *s = bs->opaque;
1227 struct stat st;
1228 int ret;
1229
1230 ret = glfs_fstat(s->fd, &st);
1231 if (ret < 0) {
1232 return -errno;
1233 } else {
1234 return st.st_blocks * 512;
1235 }
1236}
1237
8ab6feec
KW
1238static int qemu_gluster_has_zero_init(BlockDriverState *bs)
1239{
1240 /* GlusterFS volume could be backed by a block device */
1241 return 0;
1242}
1243
947eb203
NV
1244/*
1245 * Find allocation range in @bs around offset @start.
1246 * May change underlying file descriptor's file offset.
1247 * If @start is not in a hole, store @start in @data, and the
1248 * beginning of the next hole in @hole, and return 0.
1249 * If @start is in a non-trailing hole, store @start in @hole and the
1250 * beginning of the next non-hole in @data, and return 0.
1251 * If @start is in a trailing hole or beyond EOF, return -ENXIO.
1252 * If we can't find out, return a negative errno other than -ENXIO.
1253 *
c1bb86cd 1254 * (Shamefully copied from file-posix.c, only miniscule adaptions.)
947eb203
NV
1255 */
1256static int find_allocation(BlockDriverState *bs, off_t start,
1257 off_t *data, off_t *hole)
1258{
1259 BDRVGlusterState *s = bs->opaque;
947eb203
NV
1260
1261 if (!s->supports_seek_data) {
d9b78974 1262 goto exit;
947eb203
NV
1263 }
1264
d9b78974
JC
1265#if defined SEEK_HOLE && defined SEEK_DATA
1266 off_t offs;
1267
947eb203
NV
1268 /*
1269 * SEEK_DATA cases:
1270 * D1. offs == start: start is in data
1271 * D2. offs > start: start is in a hole, next data at offs
1272 * D3. offs < 0, errno = ENXIO: either start is in a trailing hole
1273 * or start is beyond EOF
1274 * If the latter happens, the file has been truncated behind
1275 * our back since we opened it. All bets are off then.
1276 * Treating like a trailing hole is simplest.
1277 * D4. offs < 0, errno != ENXIO: we learned nothing
1278 */
1279 offs = glfs_lseek(s->fd, start, SEEK_DATA);
1280 if (offs < 0) {
1281 return -errno; /* D3 or D4 */
1282 }
223a23c1
JC
1283
1284 if (offs < start) {
1285 /* This is not a valid return by lseek(). We are safe to just return
1286 * -EIO in this case, and we'll treat it like D4. Unfortunately some
1287 * versions of gluster server will return offs < start, so an assert
1288 * here will unnecessarily abort QEMU. */
1289 return -EIO;
1290 }
947eb203
NV
1291
1292 if (offs > start) {
1293 /* D2: in hole, next data at offs */
1294 *hole = start;
1295 *data = offs;
1296 return 0;
1297 }
1298
1299 /* D1: in data, end not yet known */
1300
1301 /*
1302 * SEEK_HOLE cases:
1303 * H1. offs == start: start is in a hole
1304 * If this happens here, a hole has been dug behind our back
1305 * since the previous lseek().
1306 * H2. offs > start: either start is in data, next hole at offs,
1307 * or start is in trailing hole, EOF at offs
1308 * Linux treats trailing holes like any other hole: offs ==
1309 * start. Solaris seeks to EOF instead: offs > start (blech).
1310 * If that happens here, a hole has been dug behind our back
1311 * since the previous lseek().
1312 * H3. offs < 0, errno = ENXIO: start is beyond EOF
1313 * If this happens, the file has been truncated behind our
1314 * back since we opened it. Treat it like a trailing hole.
1315 * H4. offs < 0, errno != ENXIO: we learned nothing
1316 * Pretend we know nothing at all, i.e. "forget" about D1.
1317 */
1318 offs = glfs_lseek(s->fd, start, SEEK_HOLE);
1319 if (offs < 0) {
1320 return -errno; /* D1 and (H3 or H4) */
1321 }
223a23c1
JC
1322
1323 if (offs < start) {
1324 /* This is not a valid return by lseek(). We are safe to just return
1325 * -EIO in this case, and we'll treat it like H4. Unfortunately some
1326 * versions of gluster server will return offs < start, so an assert
1327 * here will unnecessarily abort QEMU. */
1328 return -EIO;
1329 }
947eb203
NV
1330
1331 if (offs > start) {
1332 /*
1333 * D1 and H2: either in data, next hole at offs, or it was in
1334 * data but is now in a trailing hole. In the latter case,
1335 * all bets are off. Treating it as if it there was data all
1336 * the way to EOF is safe, so simply do that.
1337 */
1338 *data = start;
1339 *hole = offs;
1340 return 0;
1341 }
1342
1343 /* D1 and H1 */
1344 return -EBUSY;
d9b78974
JC
1345#endif
1346
1347exit:
1348 return -ENOTSUP;
947eb203
NV
1349}
1350
1351/*
1352 * Returns the allocation status of the specified sectors.
1353 *
1354 * If 'sector_num' is beyond the end of the disk image the return value is 0
1355 * and 'pnum' is set to 0.
1356 *
1357 * 'pnum' is set to the number of sectors (including and immediately following
1358 * the specified sector) that are known to be in the same
1359 * allocated/unallocated state.
1360 *
1361 * 'nb_sectors' is the max value 'pnum' should be set to. If nb_sectors goes
1362 * beyond the end of the disk image it will be clamped.
1363 *
c1bb86cd 1364 * (Based on raw_co_get_block_status() from file-posix.c.)
947eb203
NV
1365 */
1366static int64_t coroutine_fn qemu_gluster_co_get_block_status(
1367 BlockDriverState *bs, int64_t sector_num, int nb_sectors, int *pnum,
1368 BlockDriverState **file)
1369{
1370 BDRVGlusterState *s = bs->opaque;
1371 off_t start, data = 0, hole = 0;
1372 int64_t total_size;
1373 int ret = -EINVAL;
1374
1375 if (!s->fd) {
1376 return ret;
1377 }
1378
1379 start = sector_num * BDRV_SECTOR_SIZE;
1380 total_size = bdrv_getlength(bs);
1381 if (total_size < 0) {
1382 return total_size;
1383 } else if (start >= total_size) {
1384 *pnum = 0;
1385 return 0;
1386 } else if (start + nb_sectors * BDRV_SECTOR_SIZE > total_size) {
1387 nb_sectors = DIV_ROUND_UP(total_size - start, BDRV_SECTOR_SIZE);
1388 }
1389
1390 ret = find_allocation(bs, start, &data, &hole);
1391 if (ret == -ENXIO) {
1392 /* Trailing hole */
1393 *pnum = nb_sectors;
1394 ret = BDRV_BLOCK_ZERO;
1395 } else if (ret < 0) {
1396 /* No info available, so pretend there are no holes */
1397 *pnum = nb_sectors;
1398 ret = BDRV_BLOCK_DATA;
1399 } else if (data == start) {
1400 /* On a data extent, compute sectors to the end of the extent,
1401 * possibly including a partial sector at EOF. */
1402 *pnum = MIN(nb_sectors, DIV_ROUND_UP(hole - start, BDRV_SECTOR_SIZE));
1403 ret = BDRV_BLOCK_DATA;
1404 } else {
1405 /* On a hole, compute sectors to the beginning of the next extent. */
1406 assert(hole == start);
1407 *pnum = MIN(nb_sectors, (data - start) / BDRV_SECTOR_SIZE);
1408 ret = BDRV_BLOCK_ZERO;
1409 }
1410
1411 *file = bs;
1412
1413 return ret | BDRV_BLOCK_OFFSET_VALID | start;
1414}
1415
1416
8d6d89cb
BR
1417static BlockDriver bdrv_gluster = {
1418 .format_name = "gluster",
1419 .protocol_name = "gluster",
1420 .instance_size = sizeof(BDRVGlusterState),
6c7189bb 1421 .bdrv_needs_filename = false,
8d6d89cb 1422 .bdrv_file_open = qemu_gluster_open,
adccfbcd
JC
1423 .bdrv_reopen_prepare = qemu_gluster_reopen_prepare,
1424 .bdrv_reopen_commit = qemu_gluster_reopen_commit,
1425 .bdrv_reopen_abort = qemu_gluster_reopen_abort,
8d6d89cb 1426 .bdrv_close = qemu_gluster_close,
c282e1fd 1427 .bdrv_create = qemu_gluster_create,
8d6d89cb
BR
1428 .bdrv_getlength = qemu_gluster_getlength,
1429 .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size,
42ec24e2 1430 .bdrv_truncate = qemu_gluster_truncate,
15744b0b
BR
1431 .bdrv_co_readv = qemu_gluster_co_readv,
1432 .bdrv_co_writev = qemu_gluster_co_writev,
1433 .bdrv_co_flush_to_disk = qemu_gluster_co_flush_to_disk,
8ab6feec 1434 .bdrv_has_zero_init = qemu_gluster_has_zero_init,
0c14fb47 1435#ifdef CONFIG_GLUSTERFS_DISCARD
1014170b 1436 .bdrv_co_pdiscard = qemu_gluster_co_pdiscard,
7c815372
BR
1437#endif
1438#ifdef CONFIG_GLUSTERFS_ZEROFILL
e88a36eb 1439 .bdrv_co_pwrite_zeroes = qemu_gluster_co_pwrite_zeroes,
0c14fb47 1440#endif
947eb203 1441 .bdrv_co_get_block_status = qemu_gluster_co_get_block_status,
90c772de 1442 .create_opts = &qemu_gluster_create_opts,
8d6d89cb
BR
1443};
1444
1445static BlockDriver bdrv_gluster_tcp = {
1446 .format_name = "gluster",
1447 .protocol_name = "gluster+tcp",
1448 .instance_size = sizeof(BDRVGlusterState),
6c7189bb 1449 .bdrv_needs_filename = false,
8d6d89cb 1450 .bdrv_file_open = qemu_gluster_open,
adccfbcd
JC
1451 .bdrv_reopen_prepare = qemu_gluster_reopen_prepare,
1452 .bdrv_reopen_commit = qemu_gluster_reopen_commit,
1453 .bdrv_reopen_abort = qemu_gluster_reopen_abort,
8d6d89cb 1454 .bdrv_close = qemu_gluster_close,
c282e1fd 1455 .bdrv_create = qemu_gluster_create,
8d6d89cb
BR
1456 .bdrv_getlength = qemu_gluster_getlength,
1457 .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size,
42ec24e2 1458 .bdrv_truncate = qemu_gluster_truncate,
15744b0b
BR
1459 .bdrv_co_readv = qemu_gluster_co_readv,
1460 .bdrv_co_writev = qemu_gluster_co_writev,
1461 .bdrv_co_flush_to_disk = qemu_gluster_co_flush_to_disk,
8ab6feec 1462 .bdrv_has_zero_init = qemu_gluster_has_zero_init,
0c14fb47 1463#ifdef CONFIG_GLUSTERFS_DISCARD
1014170b 1464 .bdrv_co_pdiscard = qemu_gluster_co_pdiscard,
7c815372
BR
1465#endif
1466#ifdef CONFIG_GLUSTERFS_ZEROFILL
e88a36eb 1467 .bdrv_co_pwrite_zeroes = qemu_gluster_co_pwrite_zeroes,
0c14fb47 1468#endif
947eb203 1469 .bdrv_co_get_block_status = qemu_gluster_co_get_block_status,
90c772de 1470 .create_opts = &qemu_gluster_create_opts,
8d6d89cb
BR
1471};
1472
1473static BlockDriver bdrv_gluster_unix = {
1474 .format_name = "gluster",
1475 .protocol_name = "gluster+unix",
1476 .instance_size = sizeof(BDRVGlusterState),
030be321 1477 .bdrv_needs_filename = true,
8d6d89cb 1478 .bdrv_file_open = qemu_gluster_open,
adccfbcd
JC
1479 .bdrv_reopen_prepare = qemu_gluster_reopen_prepare,
1480 .bdrv_reopen_commit = qemu_gluster_reopen_commit,
1481 .bdrv_reopen_abort = qemu_gluster_reopen_abort,
8d6d89cb 1482 .bdrv_close = qemu_gluster_close,
c282e1fd 1483 .bdrv_create = qemu_gluster_create,
8d6d89cb
BR
1484 .bdrv_getlength = qemu_gluster_getlength,
1485 .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size,
42ec24e2 1486 .bdrv_truncate = qemu_gluster_truncate,
15744b0b
BR
1487 .bdrv_co_readv = qemu_gluster_co_readv,
1488 .bdrv_co_writev = qemu_gluster_co_writev,
1489 .bdrv_co_flush_to_disk = qemu_gluster_co_flush_to_disk,
8ab6feec 1490 .bdrv_has_zero_init = qemu_gluster_has_zero_init,
0c14fb47 1491#ifdef CONFIG_GLUSTERFS_DISCARD
1014170b 1492 .bdrv_co_pdiscard = qemu_gluster_co_pdiscard,
7c815372
BR
1493#endif
1494#ifdef CONFIG_GLUSTERFS_ZEROFILL
e88a36eb 1495 .bdrv_co_pwrite_zeroes = qemu_gluster_co_pwrite_zeroes,
0c14fb47 1496#endif
947eb203 1497 .bdrv_co_get_block_status = qemu_gluster_co_get_block_status,
90c772de 1498 .create_opts = &qemu_gluster_create_opts,
8d6d89cb
BR
1499};
1500
0552ff24
PKK
1501/* rdma is deprecated (actually never supported for volfile fetch).
1502 * Let's maintain it for the protocol compatibility, to make sure things
1503 * won't break immediately. For now, gluster+rdma will fall back to gluster+tcp
1504 * protocol with a warning.
1505 * TODO: remove gluster+rdma interface support
1506 */
8d6d89cb
BR
1507static BlockDriver bdrv_gluster_rdma = {
1508 .format_name = "gluster",
1509 .protocol_name = "gluster+rdma",
1510 .instance_size = sizeof(BDRVGlusterState),
030be321 1511 .bdrv_needs_filename = true,
8d6d89cb 1512 .bdrv_file_open = qemu_gluster_open,
adccfbcd
JC
1513 .bdrv_reopen_prepare = qemu_gluster_reopen_prepare,
1514 .bdrv_reopen_commit = qemu_gluster_reopen_commit,
1515 .bdrv_reopen_abort = qemu_gluster_reopen_abort,
8d6d89cb 1516 .bdrv_close = qemu_gluster_close,
c282e1fd 1517 .bdrv_create = qemu_gluster_create,
8d6d89cb
BR
1518 .bdrv_getlength = qemu_gluster_getlength,
1519 .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size,
42ec24e2 1520 .bdrv_truncate = qemu_gluster_truncate,
15744b0b
BR
1521 .bdrv_co_readv = qemu_gluster_co_readv,
1522 .bdrv_co_writev = qemu_gluster_co_writev,
1523 .bdrv_co_flush_to_disk = qemu_gluster_co_flush_to_disk,
8ab6feec 1524 .bdrv_has_zero_init = qemu_gluster_has_zero_init,
0c14fb47 1525#ifdef CONFIG_GLUSTERFS_DISCARD
1014170b 1526 .bdrv_co_pdiscard = qemu_gluster_co_pdiscard,
7c815372
BR
1527#endif
1528#ifdef CONFIG_GLUSTERFS_ZEROFILL
e88a36eb 1529 .bdrv_co_pwrite_zeroes = qemu_gluster_co_pwrite_zeroes,
0c14fb47 1530#endif
947eb203 1531 .bdrv_co_get_block_status = qemu_gluster_co_get_block_status,
90c772de 1532 .create_opts = &qemu_gluster_create_opts,
8d6d89cb
BR
1533};
1534
1535static void bdrv_gluster_init(void)
1536{
1537 bdrv_register(&bdrv_gluster_rdma);
1538 bdrv_register(&bdrv_gluster_unix);
1539 bdrv_register(&bdrv_gluster_tcp);
1540 bdrv_register(&bdrv_gluster);
1541}
1542
1543block_init(bdrv_gluster_init);