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