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_level; 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; 242 243 if (fs == NULL) { 244 return; 245 } 246 247 QLIST_FOREACH(entry, &glfs_list, list) { 248 if (entry->saved.fs == fs) { 249 if (--entry->saved.ref) { 250 return; 251 } 252 253 QLIST_REMOVE(entry, list); 254 255 glfs_fini(entry->saved.fs); 256 g_free(entry->saved.volume); 257 g_free(entry); 258 } 259 } 260 } 261 262 static int parse_volume_options(BlockdevOptionsGluster *gconf, char *path) 263 { 264 char *p, *q; 265 266 if (!path) { 267 return -EINVAL; 268 } 269 270 /* volume */ 271 p = q = path + strspn(path, "/"); 272 p += strcspn(p, "/"); 273 if (*p == '\0') { 274 return -EINVAL; 275 } 276 gconf->volume = g_strndup(q, p - q); 277 278 /* path */ 279 p += strspn(p, "/"); 280 if (*p == '\0') { 281 return -EINVAL; 282 } 283 gconf->path = g_strdup(p); 284 return 0; 285 } 286 287 /* 288 * file=gluster[+transport]://[host[:port]]/volume/path[?socket=...] 289 * 290 * 'gluster' is the protocol. 291 * 292 * 'transport' specifies the transport type used to connect to gluster 293 * management daemon (glusterd). Valid transport types are 294 * tcp or unix. If a transport type isn't specified, then tcp type is assumed. 295 * 296 * 'host' specifies the host where the volume file specification for 297 * the given volume resides. This can be either hostname or ipv4 address. 298 * If transport type is 'unix', then 'host' field should not be specified. 299 * The 'socket' field needs to be populated with the path to unix domain 300 * socket. 301 * 302 * 'port' is the port number on which glusterd is listening. This is optional 303 * and if not specified, QEMU will send 0 which will make gluster to use the 304 * default port. If the transport type is unix, then 'port' should not be 305 * specified. 306 * 307 * 'volume' is the name of the gluster volume which contains the VM image. 308 * 309 * 'path' is the path to the actual VM image that resides on gluster volume. 310 * 311 * Examples: 312 * 313 * file=gluster://1.2.3.4/testvol/a.img 314 * file=gluster+tcp://1.2.3.4/testvol/a.img 315 * file=gluster+tcp://1.2.3.4:24007/testvol/dir/a.img 316 * file=gluster+tcp://host.domain.com:24007/testvol/dir/a.img 317 * file=gluster+unix:///testvol/dir/a.img?socket=/tmp/glusterd.socket 318 */ 319 static int qemu_gluster_parse_uri(BlockdevOptionsGluster *gconf, 320 const char *filename) 321 { 322 GlusterServer *gsconf; 323 URI *uri; 324 QueryParams *qp = NULL; 325 bool is_unix = false; 326 int ret = 0; 327 328 uri = uri_parse(filename); 329 if (!uri) { 330 return -EINVAL; 331 } 332 333 gconf->server = g_new0(GlusterServerList, 1); 334 gconf->server->value = gsconf = g_new0(GlusterServer, 1); 335 336 /* transport */ 337 if (!uri->scheme || !strcmp(uri->scheme, "gluster")) { 338 gsconf->type = GLUSTER_TRANSPORT_TCP; 339 } else if (!strcmp(uri->scheme, "gluster+tcp")) { 340 gsconf->type = GLUSTER_TRANSPORT_TCP; 341 } else if (!strcmp(uri->scheme, "gluster+unix")) { 342 gsconf->type = GLUSTER_TRANSPORT_UNIX; 343 is_unix = true; 344 } else if (!strcmp(uri->scheme, "gluster+rdma")) { 345 gsconf->type = GLUSTER_TRANSPORT_TCP; 346 error_report("Warning: rdma feature is not supported, falling " 347 "back to tcp"); 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 } 373 gsconf->u.q_unix.path = g_strdup(qp->p[0].value); 374 } else { 375 gsconf->u.tcp.host = g_strdup(uri->server ? uri->server : "localhost"); 376 if (uri->port) { 377 gsconf->u.tcp.port = g_strdup_printf("%d", uri->port); 378 } else { 379 gsconf->u.tcp.port = g_strdup_printf("%d", GLUSTER_DEFAULT_PORT); 380 } 381 } 382 383 out: 384 if (qp) { 385 query_params_free(qp); 386 } 387 uri_free(uri); 388 return ret; 389 } 390 391 static struct glfs *qemu_gluster_glfs_init(BlockdevOptionsGluster *gconf, 392 Error **errp) 393 { 394 struct glfs *glfs; 395 int ret; 396 int old_errno; 397 GlusterServerList *server; 398 unsigned long long port; 399 400 glfs = glfs_find_preopened(gconf->volume); 401 if (glfs) { 402 return glfs; 403 } 404 405 glfs = glfs_new(gconf->volume); 406 if (!glfs) { 407 goto out; 408 } 409 410 glfs_set_preopened(gconf->volume, glfs); 411 412 for (server = gconf->server; server; server = server->next) { 413 if (server->value->type == GLUSTER_TRANSPORT_UNIX) { 414 ret = glfs_set_volfile_server(glfs, 415 GlusterTransport_lookup[server->value->type], 416 server->value->u.q_unix.path, 0); 417 } else { 418 if (parse_uint_full(server->value->u.tcp.port, &port, 10) < 0 || 419 port > 65535) { 420 error_setg(errp, "'%s' is not a valid port number", 421 server->value->u.tcp.port); 422 errno = EINVAL; 423 goto out; 424 } 425 ret = glfs_set_volfile_server(glfs, 426 GlusterTransport_lookup[server->value->type], 427 server->value->u.tcp.host, 428 (int)port); 429 } 430 431 if (ret < 0) { 432 goto out; 433 } 434 } 435 436 ret = glfs_set_logging(glfs, gconf->logfile, gconf->debug_level); 437 if (ret < 0) { 438 goto out; 439 } 440 441 ret = glfs_init(glfs); 442 if (ret) { 443 error_setg(errp, "Gluster connection for volume %s, path %s failed" 444 " to connect", gconf->volume, gconf->path); 445 for (server = gconf->server; server; server = server->next) { 446 if (server->value->type == GLUSTER_TRANSPORT_UNIX) { 447 error_append_hint(errp, "hint: failed on socket %s ", 448 server->value->u.q_unix.path); 449 } else { 450 error_append_hint(errp, "hint: failed on host %s and port %s ", 451 server->value->u.tcp.host, 452 server->value->u.tcp.port); 453 } 454 } 455 456 error_append_hint(errp, "Please refer to gluster logs for more info\n"); 457 458 /* glfs_init sometimes doesn't set errno although docs suggest that */ 459 if (errno == 0) { 460 errno = EINVAL; 461 } 462 463 goto out; 464 } 465 return glfs; 466 467 out: 468 if (glfs) { 469 old_errno = errno; 470 glfs_clear_preopened(glfs); 471 errno = old_errno; 472 } 473 return NULL; 474 } 475 476 static int qapi_enum_parse(const char *opt) 477 { 478 int i; 479 480 if (!opt) { 481 return GLUSTER_TRANSPORT__MAX; 482 } 483 484 for (i = 0; i < GLUSTER_TRANSPORT__MAX; i++) { 485 if (!strcmp(opt, GlusterTransport_lookup[i])) { 486 return i; 487 } 488 } 489 490 return i; 491 } 492 493 /* 494 * Convert the json formatted command line into qapi. 495 */ 496 static int qemu_gluster_parse_json(BlockdevOptionsGluster *gconf, 497 QDict *options, Error **errp) 498 { 499 QemuOpts *opts; 500 GlusterServer *gsconf; 501 GlusterServerList *curr = NULL; 502 QDict *backing_options = NULL; 503 Error *local_err = NULL; 504 char *str = NULL; 505 const char *ptr; 506 size_t num_servers; 507 int i; 508 509 /* create opts info from runtime_json_opts list */ 510 opts = qemu_opts_create(&runtime_json_opts, NULL, 0, &error_abort); 511 qemu_opts_absorb_qdict(opts, options, &local_err); 512 if (local_err) { 513 goto out; 514 } 515 516 num_servers = qdict_array_entries(options, GLUSTER_OPT_SERVER_PATTERN); 517 if (num_servers < 1) { 518 error_setg(&local_err, QERR_MISSING_PARAMETER, "server"); 519 goto out; 520 } 521 522 ptr = qemu_opt_get(opts, GLUSTER_OPT_VOLUME); 523 if (!ptr) { 524 error_setg(&local_err, QERR_MISSING_PARAMETER, GLUSTER_OPT_VOLUME); 525 goto out; 526 } 527 gconf->volume = g_strdup(ptr); 528 529 ptr = qemu_opt_get(opts, GLUSTER_OPT_PATH); 530 if (!ptr) { 531 error_setg(&local_err, QERR_MISSING_PARAMETER, GLUSTER_OPT_PATH); 532 goto out; 533 } 534 gconf->path = g_strdup(ptr); 535 qemu_opts_del(opts); 536 537 for (i = 0; i < num_servers; i++) { 538 str = g_strdup_printf(GLUSTER_OPT_SERVER_PATTERN"%d.", i); 539 qdict_extract_subqdict(options, &backing_options, str); 540 541 /* create opts info from runtime_type_opts list */ 542 opts = qemu_opts_create(&runtime_type_opts, NULL, 0, &error_abort); 543 qemu_opts_absorb_qdict(opts, backing_options, &local_err); 544 if (local_err) { 545 goto out; 546 } 547 548 ptr = qemu_opt_get(opts, GLUSTER_OPT_TYPE); 549 gsconf = g_new0(GlusterServer, 1); 550 gsconf->type = qapi_enum_parse(ptr); 551 if (!ptr) { 552 error_setg(&local_err, QERR_MISSING_PARAMETER, GLUSTER_OPT_TYPE); 553 error_append_hint(&local_err, GERR_INDEX_HINT, i); 554 goto out; 555 556 } 557 if (gsconf->type == GLUSTER_TRANSPORT__MAX) { 558 error_setg(&local_err, QERR_INVALID_PARAMETER_VALUE, 559 GLUSTER_OPT_TYPE, "tcp or unix"); 560 error_append_hint(&local_err, GERR_INDEX_HINT, i); 561 goto out; 562 } 563 qemu_opts_del(opts); 564 565 if (gsconf->type == GLUSTER_TRANSPORT_TCP) { 566 /* create opts info from runtime_tcp_opts list */ 567 opts = qemu_opts_create(&runtime_tcp_opts, NULL, 0, &error_abort); 568 qemu_opts_absorb_qdict(opts, backing_options, &local_err); 569 if (local_err) { 570 goto out; 571 } 572 573 ptr = qemu_opt_get(opts, GLUSTER_OPT_HOST); 574 if (!ptr) { 575 error_setg(&local_err, QERR_MISSING_PARAMETER, 576 GLUSTER_OPT_HOST); 577 error_append_hint(&local_err, GERR_INDEX_HINT, i); 578 goto out; 579 } 580 gsconf->u.tcp.host = g_strdup(ptr); 581 ptr = qemu_opt_get(opts, GLUSTER_OPT_PORT); 582 if (!ptr) { 583 error_setg(&local_err, QERR_MISSING_PARAMETER, 584 GLUSTER_OPT_PORT); 585 error_append_hint(&local_err, GERR_INDEX_HINT, i); 586 goto out; 587 } 588 gsconf->u.tcp.port = g_strdup(ptr); 589 590 /* defend for unsupported fields in InetSocketAddress, 591 * i.e. @ipv4, @ipv6 and @to 592 */ 593 ptr = qemu_opt_get(opts, GLUSTER_OPT_TO); 594 if (ptr) { 595 gsconf->u.tcp.has_to = true; 596 } 597 ptr = qemu_opt_get(opts, GLUSTER_OPT_IPV4); 598 if (ptr) { 599 gsconf->u.tcp.has_ipv4 = true; 600 } 601 ptr = qemu_opt_get(opts, GLUSTER_OPT_IPV6); 602 if (ptr) { 603 gsconf->u.tcp.has_ipv6 = true; 604 } 605 if (gsconf->u.tcp.has_to) { 606 error_setg(&local_err, "Parameter 'to' not supported"); 607 goto out; 608 } 609 if (gsconf->u.tcp.has_ipv4 || gsconf->u.tcp.has_ipv6) { 610 error_setg(&local_err, "Parameters 'ipv4/ipv6' not supported"); 611 goto out; 612 } 613 qemu_opts_del(opts); 614 } else { 615 /* create opts info from runtime_unix_opts list */ 616 opts = qemu_opts_create(&runtime_unix_opts, NULL, 0, &error_abort); 617 qemu_opts_absorb_qdict(opts, backing_options, &local_err); 618 if (local_err) { 619 goto out; 620 } 621 622 ptr = qemu_opt_get(opts, GLUSTER_OPT_SOCKET); 623 if (!ptr) { 624 error_setg(&local_err, QERR_MISSING_PARAMETER, 625 GLUSTER_OPT_SOCKET); 626 error_append_hint(&local_err, GERR_INDEX_HINT, i); 627 goto out; 628 } 629 gsconf->u.q_unix.path = g_strdup(ptr); 630 qemu_opts_del(opts); 631 } 632 633 if (gconf->server == NULL) { 634 gconf->server = g_new0(GlusterServerList, 1); 635 gconf->server->value = gsconf; 636 curr = gconf->server; 637 } else { 638 curr->next = g_new0(GlusterServerList, 1); 639 curr->next->value = gsconf; 640 curr = curr->next; 641 } 642 643 qdict_del(backing_options, str); 644 g_free(str); 645 str = NULL; 646 } 647 648 return 0; 649 650 out: 651 error_propagate(errp, local_err); 652 qemu_opts_del(opts); 653 if (str) { 654 qdict_del(backing_options, str); 655 g_free(str); 656 } 657 errno = EINVAL; 658 return -errno; 659 } 660 661 static struct glfs *qemu_gluster_init(BlockdevOptionsGluster *gconf, 662 const char *filename, 663 QDict *options, Error **errp) 664 { 665 int ret; 666 if (filename) { 667 ret = qemu_gluster_parse_uri(gconf, filename); 668 if (ret < 0) { 669 error_setg(errp, "invalid URI"); 670 error_append_hint(errp, "Usage: file=gluster[+transport]://" 671 "[host[:port]]volume/path[?socket=...]" 672 "[,file.debug=N]" 673 "[,file.logfile=/path/filename.log]\n"); 674 errno = -ret; 675 return NULL; 676 } 677 } else { 678 ret = qemu_gluster_parse_json(gconf, options, errp); 679 if (ret < 0) { 680 error_append_hint(errp, "Usage: " 681 "-drive driver=qcow2,file.driver=gluster," 682 "file.volume=testvol,file.path=/path/a.qcow2" 683 "[,file.debug=9]" 684 "[,file.logfile=/path/filename.log]," 685 "file.server.0.type=tcp," 686 "file.server.0.host=1.2.3.4," 687 "file.server.0.port=24007," 688 "file.server.1.transport=unix," 689 "file.server.1.socket=/var/run/glusterd.socket ..." 690 "\n"); 691 errno = -ret; 692 return NULL; 693 } 694 695 } 696 697 return qemu_gluster_glfs_init(gconf, errp); 698 } 699 700 static void qemu_gluster_complete_aio(void *opaque) 701 { 702 GlusterAIOCB *acb = (GlusterAIOCB *)opaque; 703 704 qemu_coroutine_enter(acb->coroutine); 705 } 706 707 /* 708 * AIO callback routine called from GlusterFS thread. 709 */ 710 static void gluster_finish_aiocb(struct glfs_fd *fd, ssize_t ret, void *arg) 711 { 712 GlusterAIOCB *acb = (GlusterAIOCB *)arg; 713 714 if (!ret || ret == acb->size) { 715 acb->ret = 0; /* Success */ 716 } else if (ret < 0) { 717 acb->ret = -errno; /* Read/Write failed */ 718 } else { 719 acb->ret = -EIO; /* Partial read/write - fail it */ 720 } 721 722 aio_bh_schedule_oneshot(acb->aio_context, qemu_gluster_complete_aio, acb); 723 } 724 725 static void qemu_gluster_parse_flags(int bdrv_flags, int *open_flags) 726 { 727 assert(open_flags != NULL); 728 729 *open_flags |= O_BINARY; 730 731 if (bdrv_flags & BDRV_O_RDWR) { 732 *open_flags |= O_RDWR; 733 } else { 734 *open_flags |= O_RDONLY; 735 } 736 737 if ((bdrv_flags & BDRV_O_NOCACHE)) { 738 *open_flags |= O_DIRECT; 739 } 740 } 741 742 /* 743 * Do SEEK_DATA/HOLE to detect if it is functional. Older broken versions of 744 * gfapi incorrectly return the current offset when SEEK_DATA/HOLE is used. 745 * - Corrected versions return -1 and set errno to EINVAL. 746 * - Versions that support SEEK_DATA/HOLE correctly, will return -1 and set 747 * errno to ENXIO when SEEK_DATA is called with a position of EOF. 748 */ 749 static bool qemu_gluster_test_seek(struct glfs_fd *fd) 750 { 751 off_t ret = 0; 752 753 #if defined SEEK_HOLE && defined SEEK_DATA 754 off_t eof; 755 756 eof = glfs_lseek(fd, 0, SEEK_END); 757 if (eof < 0) { 758 /* this should never occur */ 759 return false; 760 } 761 762 /* this should always fail with ENXIO if SEEK_DATA is supported */ 763 ret = glfs_lseek(fd, eof, SEEK_DATA); 764 #endif 765 766 return (ret < 0) && (errno == ENXIO); 767 } 768 769 static int qemu_gluster_open(BlockDriverState *bs, QDict *options, 770 int bdrv_flags, Error **errp) 771 { 772 BDRVGlusterState *s = bs->opaque; 773 int open_flags = 0; 774 int ret = 0; 775 BlockdevOptionsGluster *gconf = NULL; 776 QemuOpts *opts; 777 Error *local_err = NULL; 778 const char *filename, *logfile; 779 780 opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort); 781 qemu_opts_absorb_qdict(opts, options, &local_err); 782 if (local_err) { 783 error_propagate(errp, local_err); 784 ret = -EINVAL; 785 goto out; 786 } 787 788 filename = qemu_opt_get(opts, GLUSTER_OPT_FILENAME); 789 790 s->debug_level = qemu_opt_get_number(opts, GLUSTER_OPT_DEBUG, 791 GLUSTER_DEBUG_DEFAULT); 792 if (s->debug_level < 0) { 793 s->debug_level = 0; 794 } else if (s->debug_level > GLUSTER_DEBUG_MAX) { 795 s->debug_level = GLUSTER_DEBUG_MAX; 796 } 797 798 gconf = g_new0(BlockdevOptionsGluster, 1); 799 gconf->debug_level = s->debug_level; 800 gconf->has_debug_level = true; 801 802 logfile = qemu_opt_get(opts, GLUSTER_OPT_LOGFILE); 803 s->logfile = g_strdup(logfile ? logfile : GLUSTER_LOGFILE_DEFAULT); 804 805 gconf->logfile = g_strdup(s->logfile); 806 gconf->has_logfile = true; 807 808 s->glfs = qemu_gluster_init(gconf, filename, options, errp); 809 if (!s->glfs) { 810 ret = -errno; 811 goto out; 812 } 813 814 #ifdef CONFIG_GLUSTERFS_XLATOR_OPT 815 /* Without this, if fsync fails for a recoverable reason (for instance, 816 * ENOSPC), gluster will dump its cache, preventing retries. This means 817 * almost certain data loss. Not all gluster versions support the 818 * 'resync-failed-syncs-after-fsync' key value, but there is no way to 819 * discover during runtime if it is supported (this api returns success for 820 * unknown key/value pairs) */ 821 ret = glfs_set_xlator_option(s->glfs, "*-write-behind", 822 "resync-failed-syncs-after-fsync", 823 "on"); 824 if (ret < 0) { 825 error_setg_errno(errp, errno, "Unable to set xlator key/value pair"); 826 ret = -errno; 827 goto out; 828 } 829 #endif 830 831 qemu_gluster_parse_flags(bdrv_flags, &open_flags); 832 833 s->fd = glfs_open(s->glfs, gconf->path, open_flags); 834 if (!s->fd) { 835 ret = -errno; 836 } 837 838 s->supports_seek_data = qemu_gluster_test_seek(s->fd); 839 840 out: 841 qemu_opts_del(opts); 842 qapi_free_BlockdevOptionsGluster(gconf); 843 if (!ret) { 844 return ret; 845 } 846 g_free(s->logfile); 847 if (s->fd) { 848 glfs_close(s->fd); 849 } 850 851 glfs_clear_preopened(s->glfs); 852 853 return ret; 854 } 855 856 static int qemu_gluster_reopen_prepare(BDRVReopenState *state, 857 BlockReopenQueue *queue, Error **errp) 858 { 859 int ret = 0; 860 BDRVGlusterState *s; 861 BDRVGlusterReopenState *reop_s; 862 BlockdevOptionsGluster *gconf; 863 int open_flags = 0; 864 865 assert(state != NULL); 866 assert(state->bs != NULL); 867 868 s = state->bs->opaque; 869 870 state->opaque = g_new0(BDRVGlusterReopenState, 1); 871 reop_s = state->opaque; 872 873 qemu_gluster_parse_flags(state->flags, &open_flags); 874 875 gconf = g_new0(BlockdevOptionsGluster, 1); 876 gconf->debug_level = s->debug_level; 877 gconf->has_debug_level = true; 878 gconf->logfile = g_strdup(s->logfile); 879 gconf->has_logfile = true; 880 reop_s->glfs = qemu_gluster_init(gconf, state->bs->filename, NULL, errp); 881 if (reop_s->glfs == NULL) { 882 ret = -errno; 883 goto exit; 884 } 885 886 #ifdef CONFIG_GLUSTERFS_XLATOR_OPT 887 ret = glfs_set_xlator_option(reop_s->glfs, "*-write-behind", 888 "resync-failed-syncs-after-fsync", "on"); 889 if (ret < 0) { 890 error_setg_errno(errp, errno, "Unable to set xlator key/value pair"); 891 ret = -errno; 892 goto exit; 893 } 894 #endif 895 896 reop_s->fd = glfs_open(reop_s->glfs, gconf->path, open_flags); 897 if (reop_s->fd == NULL) { 898 /* reops->glfs will be cleaned up in _abort */ 899 ret = -errno; 900 goto exit; 901 } 902 903 exit: 904 /* state->opaque will be freed in either the _abort or _commit */ 905 qapi_free_BlockdevOptionsGluster(gconf); 906 return ret; 907 } 908 909 static void qemu_gluster_reopen_commit(BDRVReopenState *state) 910 { 911 BDRVGlusterReopenState *reop_s = state->opaque; 912 BDRVGlusterState *s = state->bs->opaque; 913 914 915 /* close the old */ 916 if (s->fd) { 917 glfs_close(s->fd); 918 } 919 920 glfs_clear_preopened(s->glfs); 921 922 /* use the newly opened image / connection */ 923 s->fd = reop_s->fd; 924 s->glfs = reop_s->glfs; 925 926 g_free(state->opaque); 927 state->opaque = NULL; 928 929 return; 930 } 931 932 933 static void qemu_gluster_reopen_abort(BDRVReopenState *state) 934 { 935 BDRVGlusterReopenState *reop_s = state->opaque; 936 937 if (reop_s == NULL) { 938 return; 939 } 940 941 if (reop_s->fd) { 942 glfs_close(reop_s->fd); 943 } 944 945 glfs_clear_preopened(reop_s->glfs); 946 947 g_free(state->opaque); 948 state->opaque = NULL; 949 950 return; 951 } 952 953 #ifdef CONFIG_GLUSTERFS_ZEROFILL 954 static coroutine_fn int qemu_gluster_co_pwrite_zeroes(BlockDriverState *bs, 955 int64_t offset, 956 int size, 957 BdrvRequestFlags flags) 958 { 959 int ret; 960 GlusterAIOCB acb; 961 BDRVGlusterState *s = bs->opaque; 962 963 acb.size = size; 964 acb.ret = 0; 965 acb.coroutine = qemu_coroutine_self(); 966 acb.aio_context = bdrv_get_aio_context(bs); 967 968 ret = glfs_zerofill_async(s->fd, offset, size, gluster_finish_aiocb, &acb); 969 if (ret < 0) { 970 return -errno; 971 } 972 973 qemu_coroutine_yield(); 974 return acb.ret; 975 } 976 977 static inline bool gluster_supports_zerofill(void) 978 { 979 return 1; 980 } 981 982 static inline int qemu_gluster_zerofill(struct glfs_fd *fd, int64_t offset, 983 int64_t size) 984 { 985 return glfs_zerofill(fd, offset, size); 986 } 987 988 #else 989 static inline bool gluster_supports_zerofill(void) 990 { 991 return 0; 992 } 993 994 static inline int qemu_gluster_zerofill(struct glfs_fd *fd, int64_t offset, 995 int64_t size) 996 { 997 return 0; 998 } 999 #endif 1000 1001 static int qemu_gluster_create(const char *filename, 1002 QemuOpts *opts, Error **errp) 1003 { 1004 BlockdevOptionsGluster *gconf; 1005 struct glfs *glfs; 1006 struct glfs_fd *fd; 1007 int ret = 0; 1008 int prealloc = 0; 1009 int64_t total_size = 0; 1010 char *tmp = NULL; 1011 1012 gconf = g_new0(BlockdevOptionsGluster, 1); 1013 gconf->debug_level = qemu_opt_get_number_del(opts, GLUSTER_OPT_DEBUG, 1014 GLUSTER_DEBUG_DEFAULT); 1015 if (gconf->debug_level < 0) { 1016 gconf->debug_level = 0; 1017 } else if (gconf->debug_level > GLUSTER_DEBUG_MAX) { 1018 gconf->debug_level = GLUSTER_DEBUG_MAX; 1019 } 1020 gconf->has_debug_level = true; 1021 1022 gconf->logfile = qemu_opt_get_del(opts, GLUSTER_OPT_LOGFILE); 1023 if (!gconf->logfile) { 1024 gconf->logfile = g_strdup(GLUSTER_LOGFILE_DEFAULT); 1025 } 1026 gconf->has_logfile = true; 1027 1028 glfs = qemu_gluster_init(gconf, filename, NULL, errp); 1029 if (!glfs) { 1030 ret = -errno; 1031 goto out; 1032 } 1033 1034 total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0), 1035 BDRV_SECTOR_SIZE); 1036 1037 tmp = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC); 1038 if (!tmp || !strcmp(tmp, "off")) { 1039 prealloc = 0; 1040 } else if (!strcmp(tmp, "full") && gluster_supports_zerofill()) { 1041 prealloc = 1; 1042 } else { 1043 error_setg(errp, "Invalid preallocation mode: '%s'" 1044 " or GlusterFS doesn't support zerofill API", tmp); 1045 ret = -EINVAL; 1046 goto out; 1047 } 1048 1049 fd = glfs_creat(glfs, gconf->path, 1050 O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, S_IRUSR | S_IWUSR); 1051 if (!fd) { 1052 ret = -errno; 1053 } else { 1054 if (!glfs_ftruncate(fd, total_size)) { 1055 if (prealloc && qemu_gluster_zerofill(fd, 0, total_size)) { 1056 ret = -errno; 1057 } 1058 } else { 1059 ret = -errno; 1060 } 1061 1062 if (glfs_close(fd) != 0) { 1063 ret = -errno; 1064 } 1065 } 1066 out: 1067 g_free(tmp); 1068 qapi_free_BlockdevOptionsGluster(gconf); 1069 glfs_clear_preopened(glfs); 1070 return ret; 1071 } 1072 1073 static coroutine_fn int qemu_gluster_co_rw(BlockDriverState *bs, 1074 int64_t sector_num, int nb_sectors, 1075 QEMUIOVector *qiov, int write) 1076 { 1077 int ret; 1078 GlusterAIOCB acb; 1079 BDRVGlusterState *s = bs->opaque; 1080 size_t size = nb_sectors * BDRV_SECTOR_SIZE; 1081 off_t offset = sector_num * BDRV_SECTOR_SIZE; 1082 1083 acb.size = size; 1084 acb.ret = 0; 1085 acb.coroutine = qemu_coroutine_self(); 1086 acb.aio_context = bdrv_get_aio_context(bs); 1087 1088 if (write) { 1089 ret = glfs_pwritev_async(s->fd, qiov->iov, qiov->niov, offset, 0, 1090 gluster_finish_aiocb, &acb); 1091 } else { 1092 ret = glfs_preadv_async(s->fd, qiov->iov, qiov->niov, offset, 0, 1093 gluster_finish_aiocb, &acb); 1094 } 1095 1096 if (ret < 0) { 1097 return -errno; 1098 } 1099 1100 qemu_coroutine_yield(); 1101 return acb.ret; 1102 } 1103 1104 static int qemu_gluster_truncate(BlockDriverState *bs, int64_t offset) 1105 { 1106 int ret; 1107 BDRVGlusterState *s = bs->opaque; 1108 1109 ret = glfs_ftruncate(s->fd, offset); 1110 if (ret < 0) { 1111 return -errno; 1112 } 1113 1114 return 0; 1115 } 1116 1117 static coroutine_fn int qemu_gluster_co_readv(BlockDriverState *bs, 1118 int64_t sector_num, 1119 int nb_sectors, 1120 QEMUIOVector *qiov) 1121 { 1122 return qemu_gluster_co_rw(bs, sector_num, nb_sectors, qiov, 0); 1123 } 1124 1125 static coroutine_fn int qemu_gluster_co_writev(BlockDriverState *bs, 1126 int64_t sector_num, 1127 int nb_sectors, 1128 QEMUIOVector *qiov) 1129 { 1130 return qemu_gluster_co_rw(bs, sector_num, nb_sectors, qiov, 1); 1131 } 1132 1133 static void qemu_gluster_close(BlockDriverState *bs) 1134 { 1135 BDRVGlusterState *s = bs->opaque; 1136 1137 g_free(s->logfile); 1138 if (s->fd) { 1139 glfs_close(s->fd); 1140 s->fd = NULL; 1141 } 1142 glfs_clear_preopened(s->glfs); 1143 } 1144 1145 static coroutine_fn int qemu_gluster_co_flush_to_disk(BlockDriverState *bs) 1146 { 1147 int ret; 1148 GlusterAIOCB acb; 1149 BDRVGlusterState *s = bs->opaque; 1150 1151 acb.size = 0; 1152 acb.ret = 0; 1153 acb.coroutine = qemu_coroutine_self(); 1154 acb.aio_context = bdrv_get_aio_context(bs); 1155 1156 ret = glfs_fsync_async(s->fd, gluster_finish_aiocb, &acb); 1157 if (ret < 0) { 1158 ret = -errno; 1159 goto error; 1160 } 1161 1162 qemu_coroutine_yield(); 1163 if (acb.ret < 0) { 1164 ret = acb.ret; 1165 goto error; 1166 } 1167 1168 return acb.ret; 1169 1170 error: 1171 /* Some versions of Gluster (3.5.6 -> 3.5.8?) will not retain its cache 1172 * after a fsync failure, so we have no way of allowing the guest to safely 1173 * continue. Gluster versions prior to 3.5.6 don't retain the cache 1174 * either, but will invalidate the fd on error, so this is again our only 1175 * option. 1176 * 1177 * The 'resync-failed-syncs-after-fsync' xlator option for the 1178 * write-behind cache will cause later gluster versions to retain its 1179 * cache after error, so long as the fd remains open. However, we 1180 * currently have no way of knowing if this option is supported. 1181 * 1182 * TODO: Once gluster provides a way for us to determine if the option 1183 * is supported, bypass the closure and setting drv to NULL. */ 1184 qemu_gluster_close(bs); 1185 bs->drv = NULL; 1186 return ret; 1187 } 1188 1189 #ifdef CONFIG_GLUSTERFS_DISCARD 1190 static coroutine_fn int qemu_gluster_co_pdiscard(BlockDriverState *bs, 1191 int64_t offset, int size) 1192 { 1193 int ret; 1194 GlusterAIOCB acb; 1195 BDRVGlusterState *s = bs->opaque; 1196 1197 acb.size = 0; 1198 acb.ret = 0; 1199 acb.coroutine = qemu_coroutine_self(); 1200 acb.aio_context = bdrv_get_aio_context(bs); 1201 1202 ret = glfs_discard_async(s->fd, offset, size, gluster_finish_aiocb, &acb); 1203 if (ret < 0) { 1204 return -errno; 1205 } 1206 1207 qemu_coroutine_yield(); 1208 return acb.ret; 1209 } 1210 #endif 1211 1212 static int64_t qemu_gluster_getlength(BlockDriverState *bs) 1213 { 1214 BDRVGlusterState *s = bs->opaque; 1215 int64_t ret; 1216 1217 ret = glfs_lseek(s->fd, 0, SEEK_END); 1218 if (ret < 0) { 1219 return -errno; 1220 } else { 1221 return ret; 1222 } 1223 } 1224 1225 static int64_t qemu_gluster_allocated_file_size(BlockDriverState *bs) 1226 { 1227 BDRVGlusterState *s = bs->opaque; 1228 struct stat st; 1229 int ret; 1230 1231 ret = glfs_fstat(s->fd, &st); 1232 if (ret < 0) { 1233 return -errno; 1234 } else { 1235 return st.st_blocks * 512; 1236 } 1237 } 1238 1239 static int qemu_gluster_has_zero_init(BlockDriverState *bs) 1240 { 1241 /* GlusterFS volume could be backed by a block device */ 1242 return 0; 1243 } 1244 1245 /* 1246 * Find allocation range in @bs around offset @start. 1247 * May change underlying file descriptor's file offset. 1248 * If @start is not in a hole, store @start in @data, and the 1249 * beginning of the next hole in @hole, and return 0. 1250 * If @start is in a non-trailing hole, store @start in @hole and the 1251 * beginning of the next non-hole in @data, and return 0. 1252 * If @start is in a trailing hole or beyond EOF, return -ENXIO. 1253 * If we can't find out, return a negative errno other than -ENXIO. 1254 * 1255 * (Shamefully copied from raw-posix.c, only miniscule adaptions.) 1256 */ 1257 static int find_allocation(BlockDriverState *bs, off_t start, 1258 off_t *data, off_t *hole) 1259 { 1260 BDRVGlusterState *s = bs->opaque; 1261 1262 if (!s->supports_seek_data) { 1263 goto exit; 1264 } 1265 1266 #if defined SEEK_HOLE && defined SEEK_DATA 1267 off_t offs; 1268 1269 /* 1270 * SEEK_DATA cases: 1271 * D1. offs == start: start is in data 1272 * D2. offs > start: start is in a hole, next data at offs 1273 * D3. offs < 0, errno = ENXIO: either start is in a trailing hole 1274 * or start is beyond EOF 1275 * If the latter happens, the file has been truncated behind 1276 * our back since we opened it. All bets are off then. 1277 * Treating like a trailing hole is simplest. 1278 * D4. offs < 0, errno != ENXIO: we learned nothing 1279 */ 1280 offs = glfs_lseek(s->fd, start, SEEK_DATA); 1281 if (offs < 0) { 1282 return -errno; /* D3 or D4 */ 1283 } 1284 assert(offs >= start); 1285 1286 if (offs > start) { 1287 /* D2: in hole, next data at offs */ 1288 *hole = start; 1289 *data = offs; 1290 return 0; 1291 } 1292 1293 /* D1: in data, end not yet known */ 1294 1295 /* 1296 * SEEK_HOLE cases: 1297 * H1. offs == start: start is in a hole 1298 * If this happens here, a hole has been dug behind our back 1299 * since the previous lseek(). 1300 * H2. offs > start: either start is in data, next hole at offs, 1301 * or start is in trailing hole, EOF at offs 1302 * Linux treats trailing holes like any other hole: offs == 1303 * start. Solaris seeks to EOF instead: offs > start (blech). 1304 * If that happens here, a hole has been dug behind our back 1305 * since the previous lseek(). 1306 * H3. offs < 0, errno = ENXIO: start is beyond EOF 1307 * If this happens, the file has been truncated behind our 1308 * back since we opened it. Treat it like a trailing hole. 1309 * H4. offs < 0, errno != ENXIO: we learned nothing 1310 * Pretend we know nothing at all, i.e. "forget" about D1. 1311 */ 1312 offs = glfs_lseek(s->fd, start, SEEK_HOLE); 1313 if (offs < 0) { 1314 return -errno; /* D1 and (H3 or H4) */ 1315 } 1316 assert(offs >= start); 1317 1318 if (offs > start) { 1319 /* 1320 * D1 and H2: either in data, next hole at offs, or it was in 1321 * data but is now in a trailing hole. In the latter case, 1322 * all bets are off. Treating it as if it there was data all 1323 * the way to EOF is safe, so simply do that. 1324 */ 1325 *data = start; 1326 *hole = offs; 1327 return 0; 1328 } 1329 1330 /* D1 and H1 */ 1331 return -EBUSY; 1332 #endif 1333 1334 exit: 1335 return -ENOTSUP; 1336 } 1337 1338 /* 1339 * Returns the allocation status of the specified sectors. 1340 * 1341 * If 'sector_num' is beyond the end of the disk image the return value is 0 1342 * and 'pnum' is set to 0. 1343 * 1344 * 'pnum' is set to the number of sectors (including and immediately following 1345 * the specified sector) that are known to be in the same 1346 * allocated/unallocated state. 1347 * 1348 * 'nb_sectors' is the max value 'pnum' should be set to. If nb_sectors goes 1349 * beyond the end of the disk image it will be clamped. 1350 * 1351 * (Based on raw_co_get_block_status() from raw-posix.c.) 1352 */ 1353 static int64_t coroutine_fn qemu_gluster_co_get_block_status( 1354 BlockDriverState *bs, int64_t sector_num, int nb_sectors, int *pnum, 1355 BlockDriverState **file) 1356 { 1357 BDRVGlusterState *s = bs->opaque; 1358 off_t start, data = 0, hole = 0; 1359 int64_t total_size; 1360 int ret = -EINVAL; 1361 1362 if (!s->fd) { 1363 return ret; 1364 } 1365 1366 start = sector_num * BDRV_SECTOR_SIZE; 1367 total_size = bdrv_getlength(bs); 1368 if (total_size < 0) { 1369 return total_size; 1370 } else if (start >= total_size) { 1371 *pnum = 0; 1372 return 0; 1373 } else if (start + nb_sectors * BDRV_SECTOR_SIZE > total_size) { 1374 nb_sectors = DIV_ROUND_UP(total_size - start, BDRV_SECTOR_SIZE); 1375 } 1376 1377 ret = find_allocation(bs, start, &data, &hole); 1378 if (ret == -ENXIO) { 1379 /* Trailing hole */ 1380 *pnum = nb_sectors; 1381 ret = BDRV_BLOCK_ZERO; 1382 } else if (ret < 0) { 1383 /* No info available, so pretend there are no holes */ 1384 *pnum = nb_sectors; 1385 ret = BDRV_BLOCK_DATA; 1386 } else if (data == start) { 1387 /* On a data extent, compute sectors to the end of the extent, 1388 * possibly including a partial sector at EOF. */ 1389 *pnum = MIN(nb_sectors, DIV_ROUND_UP(hole - start, BDRV_SECTOR_SIZE)); 1390 ret = BDRV_BLOCK_DATA; 1391 } else { 1392 /* On a hole, compute sectors to the beginning of the next extent. */ 1393 assert(hole == start); 1394 *pnum = MIN(nb_sectors, (data - start) / BDRV_SECTOR_SIZE); 1395 ret = BDRV_BLOCK_ZERO; 1396 } 1397 1398 *file = bs; 1399 1400 return ret | BDRV_BLOCK_OFFSET_VALID | start; 1401 } 1402 1403 1404 static BlockDriver bdrv_gluster = { 1405 .format_name = "gluster", 1406 .protocol_name = "gluster", 1407 .instance_size = sizeof(BDRVGlusterState), 1408 .bdrv_needs_filename = false, 1409 .bdrv_file_open = qemu_gluster_open, 1410 .bdrv_reopen_prepare = qemu_gluster_reopen_prepare, 1411 .bdrv_reopen_commit = qemu_gluster_reopen_commit, 1412 .bdrv_reopen_abort = qemu_gluster_reopen_abort, 1413 .bdrv_close = qemu_gluster_close, 1414 .bdrv_create = qemu_gluster_create, 1415 .bdrv_getlength = qemu_gluster_getlength, 1416 .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size, 1417 .bdrv_truncate = qemu_gluster_truncate, 1418 .bdrv_co_readv = qemu_gluster_co_readv, 1419 .bdrv_co_writev = qemu_gluster_co_writev, 1420 .bdrv_co_flush_to_disk = qemu_gluster_co_flush_to_disk, 1421 .bdrv_has_zero_init = qemu_gluster_has_zero_init, 1422 #ifdef CONFIG_GLUSTERFS_DISCARD 1423 .bdrv_co_pdiscard = qemu_gluster_co_pdiscard, 1424 #endif 1425 #ifdef CONFIG_GLUSTERFS_ZEROFILL 1426 .bdrv_co_pwrite_zeroes = qemu_gluster_co_pwrite_zeroes, 1427 #endif 1428 .bdrv_co_get_block_status = qemu_gluster_co_get_block_status, 1429 .create_opts = &qemu_gluster_create_opts, 1430 }; 1431 1432 static BlockDriver bdrv_gluster_tcp = { 1433 .format_name = "gluster", 1434 .protocol_name = "gluster+tcp", 1435 .instance_size = sizeof(BDRVGlusterState), 1436 .bdrv_needs_filename = false, 1437 .bdrv_file_open = qemu_gluster_open, 1438 .bdrv_reopen_prepare = qemu_gluster_reopen_prepare, 1439 .bdrv_reopen_commit = qemu_gluster_reopen_commit, 1440 .bdrv_reopen_abort = qemu_gluster_reopen_abort, 1441 .bdrv_close = qemu_gluster_close, 1442 .bdrv_create = qemu_gluster_create, 1443 .bdrv_getlength = qemu_gluster_getlength, 1444 .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size, 1445 .bdrv_truncate = qemu_gluster_truncate, 1446 .bdrv_co_readv = qemu_gluster_co_readv, 1447 .bdrv_co_writev = qemu_gluster_co_writev, 1448 .bdrv_co_flush_to_disk = qemu_gluster_co_flush_to_disk, 1449 .bdrv_has_zero_init = qemu_gluster_has_zero_init, 1450 #ifdef CONFIG_GLUSTERFS_DISCARD 1451 .bdrv_co_pdiscard = qemu_gluster_co_pdiscard, 1452 #endif 1453 #ifdef CONFIG_GLUSTERFS_ZEROFILL 1454 .bdrv_co_pwrite_zeroes = qemu_gluster_co_pwrite_zeroes, 1455 #endif 1456 .bdrv_co_get_block_status = qemu_gluster_co_get_block_status, 1457 .create_opts = &qemu_gluster_create_opts, 1458 }; 1459 1460 static BlockDriver bdrv_gluster_unix = { 1461 .format_name = "gluster", 1462 .protocol_name = "gluster+unix", 1463 .instance_size = sizeof(BDRVGlusterState), 1464 .bdrv_needs_filename = true, 1465 .bdrv_file_open = qemu_gluster_open, 1466 .bdrv_reopen_prepare = qemu_gluster_reopen_prepare, 1467 .bdrv_reopen_commit = qemu_gluster_reopen_commit, 1468 .bdrv_reopen_abort = qemu_gluster_reopen_abort, 1469 .bdrv_close = qemu_gluster_close, 1470 .bdrv_create = qemu_gluster_create, 1471 .bdrv_getlength = qemu_gluster_getlength, 1472 .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size, 1473 .bdrv_truncate = qemu_gluster_truncate, 1474 .bdrv_co_readv = qemu_gluster_co_readv, 1475 .bdrv_co_writev = qemu_gluster_co_writev, 1476 .bdrv_co_flush_to_disk = qemu_gluster_co_flush_to_disk, 1477 .bdrv_has_zero_init = qemu_gluster_has_zero_init, 1478 #ifdef CONFIG_GLUSTERFS_DISCARD 1479 .bdrv_co_pdiscard = qemu_gluster_co_pdiscard, 1480 #endif 1481 #ifdef CONFIG_GLUSTERFS_ZEROFILL 1482 .bdrv_co_pwrite_zeroes = qemu_gluster_co_pwrite_zeroes, 1483 #endif 1484 .bdrv_co_get_block_status = qemu_gluster_co_get_block_status, 1485 .create_opts = &qemu_gluster_create_opts, 1486 }; 1487 1488 /* rdma is deprecated (actually never supported for volfile fetch). 1489 * Let's maintain it for the protocol compatibility, to make sure things 1490 * won't break immediately. For now, gluster+rdma will fall back to gluster+tcp 1491 * protocol with a warning. 1492 * TODO: remove gluster+rdma interface support 1493 */ 1494 static BlockDriver bdrv_gluster_rdma = { 1495 .format_name = "gluster", 1496 .protocol_name = "gluster+rdma", 1497 .instance_size = sizeof(BDRVGlusterState), 1498 .bdrv_needs_filename = true, 1499 .bdrv_file_open = qemu_gluster_open, 1500 .bdrv_reopen_prepare = qemu_gluster_reopen_prepare, 1501 .bdrv_reopen_commit = qemu_gluster_reopen_commit, 1502 .bdrv_reopen_abort = qemu_gluster_reopen_abort, 1503 .bdrv_close = qemu_gluster_close, 1504 .bdrv_create = qemu_gluster_create, 1505 .bdrv_getlength = qemu_gluster_getlength, 1506 .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size, 1507 .bdrv_truncate = qemu_gluster_truncate, 1508 .bdrv_co_readv = qemu_gluster_co_readv, 1509 .bdrv_co_writev = qemu_gluster_co_writev, 1510 .bdrv_co_flush_to_disk = qemu_gluster_co_flush_to_disk, 1511 .bdrv_has_zero_init = qemu_gluster_has_zero_init, 1512 #ifdef CONFIG_GLUSTERFS_DISCARD 1513 .bdrv_co_pdiscard = qemu_gluster_co_pdiscard, 1514 #endif 1515 #ifdef CONFIG_GLUSTERFS_ZEROFILL 1516 .bdrv_co_pwrite_zeroes = qemu_gluster_co_pwrite_zeroes, 1517 #endif 1518 .bdrv_co_get_block_status = qemu_gluster_co_get_block_status, 1519 .create_opts = &qemu_gluster_create_opts, 1520 }; 1521 1522 static void bdrv_gluster_init(void) 1523 { 1524 bdrv_register(&bdrv_gluster_rdma); 1525 bdrv_register(&bdrv_gluster_unix); 1526 bdrv_register(&bdrv_gluster_tcp); 1527 bdrv_register(&bdrv_gluster); 1528 } 1529 1530 block_init(bdrv_gluster_init); 1531