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