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