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 11 #include "qemu/osdep.h" 12 #include "qemu/units.h" 13 #include <glusterfs/api/glfs.h> 14 #include "block/block-io.h" 15 #include "block/block_int.h" 16 #include "block/qdict.h" 17 #include "qapi/error.h" 18 #include "qapi/qmp/qdict.h" 19 #include "qapi/qmp/qerror.h" 20 #include "qemu/error-report.h" 21 #include "qemu/module.h" 22 #include "qemu/option.h" 23 #include "qemu/cutils.h" 24 25 #ifdef CONFIG_GLUSTERFS_FTRUNCATE_HAS_STAT 26 # define glfs_ftruncate(fd, offset) glfs_ftruncate(fd, offset, NULL, NULL) 27 #endif 28 29 #define GLUSTER_OPT_FILENAME "filename" 30 #define GLUSTER_OPT_VOLUME "volume" 31 #define GLUSTER_OPT_PATH "path" 32 #define GLUSTER_OPT_TYPE "type" 33 #define GLUSTER_OPT_SERVER_PATTERN "server." 34 #define GLUSTER_OPT_HOST "host" 35 #define GLUSTER_OPT_PORT "port" 36 #define GLUSTER_OPT_TO "to" 37 #define GLUSTER_OPT_IPV4 "ipv4" 38 #define GLUSTER_OPT_IPV6 "ipv6" 39 #define GLUSTER_OPT_SOCKET "socket" 40 #define GLUSTER_OPT_DEBUG "debug" 41 #define GLUSTER_DEFAULT_PORT 24007 42 #define GLUSTER_DEBUG_DEFAULT 4 43 #define GLUSTER_DEBUG_MAX 9 44 #define GLUSTER_OPT_LOGFILE "logfile" 45 #define GLUSTER_LOGFILE_DEFAULT "-" /* handled in libgfapi as /dev/stderr */ 46 /* 47 * Several versions of GlusterFS (3.12? -> 6.0.1) fail when the transfer size 48 * is greater or equal to 1024 MiB, so we are limiting the transfer size to 512 49 * MiB to avoid this rare issue. 50 */ 51 #define GLUSTER_MAX_TRANSFER (512 * MiB) 52 53 #define GERR_INDEX_HINT "hint: check in 'server' array index '%d'\n" 54 55 typedef struct GlusterAIOCB { 56 int64_t size; 57 int ret; 58 Coroutine *coroutine; 59 AioContext *aio_context; 60 } GlusterAIOCB; 61 62 typedef struct BDRVGlusterState { 63 struct glfs *glfs; 64 struct glfs_fd *fd; 65 char *logfile; 66 bool supports_seek_data; 67 int debug; 68 } BDRVGlusterState; 69 70 typedef struct BDRVGlusterReopenState { 71 struct glfs *glfs; 72 struct glfs_fd *fd; 73 } BDRVGlusterReopenState; 74 75 76 typedef struct GlfsPreopened { 77 char *volume; 78 glfs_t *fs; 79 int ref; 80 } GlfsPreopened; 81 82 typedef struct ListElement { 83 QLIST_ENTRY(ListElement) list; 84 GlfsPreopened saved; 85 } ListElement; 86 87 static QLIST_HEAD(, ListElement) glfs_list; 88 89 static QemuOptsList qemu_gluster_create_opts = { 90 .name = "qemu-gluster-create-opts", 91 .head = QTAILQ_HEAD_INITIALIZER(qemu_gluster_create_opts.head), 92 .desc = { 93 { 94 .name = BLOCK_OPT_SIZE, 95 .type = QEMU_OPT_SIZE, 96 .help = "Virtual disk size" 97 }, 98 { 99 .name = BLOCK_OPT_PREALLOC, 100 .type = QEMU_OPT_STRING, 101 .help = "Preallocation mode (allowed values: off" 102 #ifdef CONFIG_GLUSTERFS_FALLOCATE 103 ", falloc" 104 #endif 105 #ifdef CONFIG_GLUSTERFS_ZEROFILL 106 ", full" 107 #endif 108 ")" 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_opts = { 125 .name = "gluster", 126 .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head), 127 .desc = { 128 { 129 .name = GLUSTER_OPT_FILENAME, 130 .type = QEMU_OPT_STRING, 131 .help = "URL to the gluster image", 132 }, 133 { 134 .name = GLUSTER_OPT_DEBUG, 135 .type = QEMU_OPT_NUMBER, 136 .help = "Gluster log level, valid range is 0-9", 137 }, 138 { 139 .name = GLUSTER_OPT_LOGFILE, 140 .type = QEMU_OPT_STRING, 141 .help = "Logfile path of libgfapi", 142 }, 143 { /* end of list */ } 144 }, 145 }; 146 147 static QemuOptsList runtime_json_opts = { 148 .name = "gluster_json", 149 .head = QTAILQ_HEAD_INITIALIZER(runtime_json_opts.head), 150 .desc = { 151 { 152 .name = GLUSTER_OPT_VOLUME, 153 .type = QEMU_OPT_STRING, 154 .help = "name of gluster volume where VM image resides", 155 }, 156 { 157 .name = GLUSTER_OPT_PATH, 158 .type = QEMU_OPT_STRING, 159 .help = "absolute path to image file in gluster volume", 160 }, 161 { 162 .name = GLUSTER_OPT_DEBUG, 163 .type = QEMU_OPT_NUMBER, 164 .help = "Gluster log level, valid range is 0-9", 165 }, 166 { /* end of list */ } 167 }, 168 }; 169 170 static QemuOptsList runtime_type_opts = { 171 .name = "gluster_type", 172 .head = QTAILQ_HEAD_INITIALIZER(runtime_type_opts.head), 173 .desc = { 174 { 175 .name = GLUSTER_OPT_TYPE, 176 .type = QEMU_OPT_STRING, 177 .help = "inet|unix", 178 }, 179 { /* end of list */ } 180 }, 181 }; 182 183 static QemuOptsList runtime_unix_opts = { 184 .name = "gluster_unix", 185 .head = QTAILQ_HEAD_INITIALIZER(runtime_unix_opts.head), 186 .desc = { 187 { 188 .name = GLUSTER_OPT_SOCKET, 189 .type = QEMU_OPT_STRING, 190 .help = "socket file path (legacy)", 191 }, 192 { 193 .name = GLUSTER_OPT_PATH, 194 .type = QEMU_OPT_STRING, 195 .help = "socket file path (QAPI)", 196 }, 197 { /* end of list */ } 198 }, 199 }; 200 201 static QemuOptsList runtime_inet_opts = { 202 .name = "gluster_inet", 203 .head = QTAILQ_HEAD_INITIALIZER(runtime_inet_opts.head), 204 .desc = { 205 { 206 .name = GLUSTER_OPT_TYPE, 207 .type = QEMU_OPT_STRING, 208 .help = "inet|unix", 209 }, 210 { 211 .name = GLUSTER_OPT_HOST, 212 .type = QEMU_OPT_STRING, 213 .help = "host address (hostname/ipv4/ipv6 addresses)", 214 }, 215 { 216 .name = GLUSTER_OPT_PORT, 217 .type = QEMU_OPT_STRING, 218 .help = "port number on which glusterd is listening (default 24007)", 219 }, 220 { 221 .name = "to", 222 .type = QEMU_OPT_NUMBER, 223 .help = "max port number, not supported by gluster", 224 }, 225 { 226 .name = "ipv4", 227 .type = QEMU_OPT_BOOL, 228 .help = "ipv4 bool value, not supported by gluster", 229 }, 230 { 231 .name = "ipv6", 232 .type = QEMU_OPT_BOOL, 233 .help = "ipv6 bool value, not supported by gluster", 234 }, 235 { /* end of list */ } 236 }, 237 }; 238 239 static void glfs_set_preopened(const char *volume, glfs_t *fs) 240 { 241 ListElement *entry = NULL; 242 243 entry = g_new(ListElement, 1); 244 245 entry->saved.volume = g_strdup(volume); 246 247 entry->saved.fs = fs; 248 entry->saved.ref = 1; 249 250 QLIST_INSERT_HEAD(&glfs_list, entry, list); 251 } 252 253 static glfs_t *glfs_find_preopened(const char *volume) 254 { 255 ListElement *entry = NULL; 256 257 QLIST_FOREACH(entry, &glfs_list, list) { 258 if (strcmp(entry->saved.volume, volume) == 0) { 259 entry->saved.ref++; 260 return entry->saved.fs; 261 } 262 } 263 264 return NULL; 265 } 266 267 static void glfs_clear_preopened(glfs_t *fs) 268 { 269 ListElement *entry = NULL; 270 ListElement *next; 271 272 if (fs == NULL) { 273 return; 274 } 275 276 QLIST_FOREACH_SAFE(entry, &glfs_list, list, next) { 277 if (entry->saved.fs == fs) { 278 if (--entry->saved.ref) { 279 return; 280 } 281 282 QLIST_REMOVE(entry, list); 283 284 glfs_fini(entry->saved.fs); 285 g_free(entry->saved.volume); 286 g_free(entry); 287 } 288 } 289 } 290 291 static int parse_volume_options(BlockdevOptionsGluster *gconf, const char *path) 292 { 293 const char *p, *q; 294 295 if (!path) { 296 return -EINVAL; 297 } 298 299 /* volume */ 300 p = q = path + strspn(path, "/"); 301 p += strcspn(p, "/"); 302 if (*p == '\0') { 303 return -EINVAL; 304 } 305 gconf->volume = g_strndup(q, p - q); 306 307 /* path */ 308 p += strspn(p, "/"); 309 if (*p == '\0') { 310 return -EINVAL; 311 } 312 gconf->path = g_strdup(p); 313 return 0; 314 } 315 316 /* 317 * file=gluster[+transport]://[host[:port]]/volume/path[?socket=...] 318 * 319 * 'gluster' is the protocol. 320 * 321 * 'transport' specifies the transport type used to connect to gluster 322 * management daemon (glusterd). Valid transport types are 323 * tcp or unix. If a transport type isn't specified, then tcp type is assumed. 324 * 325 * 'host' specifies the host where the volume file specification for 326 * the given volume resides. This can be either hostname or ipv4 address. 327 * If transport type is 'unix', then 'host' field should not be specified. 328 * The 'socket' field needs to be populated with the path to unix domain 329 * socket. 330 * 331 * 'port' is the port number on which glusterd is listening. This is optional 332 * and if not specified, QEMU will send 0 which will make gluster to use the 333 * default port. If the transport type is unix, then 'port' should not be 334 * specified. 335 * 336 * 'volume' is the name of the gluster volume which contains the VM image. 337 * 338 * 'path' is the path to the actual VM image that resides on gluster volume. 339 * 340 * Examples: 341 * 342 * file=gluster://1.2.3.4/testvol/a.img 343 * file=gluster+tcp://1.2.3.4/testvol/a.img 344 * file=gluster+tcp://1.2.3.4:24007/testvol/dir/a.img 345 * file=gluster+tcp://host.domain.com:24007/testvol/dir/a.img 346 * file=gluster+unix:///testvol/dir/a.img?socket=/tmp/glusterd.socket 347 */ 348 static int qemu_gluster_parse_uri(BlockdevOptionsGluster *gconf, 349 const char *filename) 350 { 351 g_autoptr(GUri) uri = g_uri_parse(filename, G_URI_FLAGS_NONE, NULL); 352 g_autoptr(GHashTable) qp = NULL; 353 SocketAddress *gsconf; 354 bool is_unix = false; 355 const char *uri_scheme, *uri_query, *uri_server; 356 int uri_port, ret; 357 358 if (!uri) { 359 return -EINVAL; 360 } 361 362 gsconf = g_new0(SocketAddress, 1); 363 QAPI_LIST_PREPEND(gconf->server, gsconf); 364 365 /* transport */ 366 uri_scheme = g_uri_get_scheme(uri); 367 if (!uri_scheme || !strcmp(uri_scheme, "gluster")) { 368 gsconf->type = SOCKET_ADDRESS_TYPE_INET; 369 } else if (!strcmp(uri_scheme, "gluster+tcp")) { 370 gsconf->type = SOCKET_ADDRESS_TYPE_INET; 371 } else if (!strcmp(uri_scheme, "gluster+unix")) { 372 gsconf->type = SOCKET_ADDRESS_TYPE_UNIX; 373 is_unix = true; 374 } else { 375 return -EINVAL; 376 } 377 378 ret = parse_volume_options(gconf, g_uri_get_path(uri)); 379 if (ret < 0) { 380 return ret; 381 } 382 383 uri_query = g_uri_get_query(uri); 384 if (uri_query) { 385 qp = g_uri_parse_params(uri_query, -1, "&", G_URI_PARAMS_NONE, NULL); 386 if (!qp) { 387 return -EINVAL; 388 } 389 ret = g_hash_table_size(qp); 390 if (ret > 1 || (is_unix && !ret) || (!is_unix && ret)) { 391 return -EINVAL; 392 } 393 } 394 395 uri_server = g_uri_get_host(uri); 396 uri_port = g_uri_get_port(uri); 397 398 if (is_unix) { 399 char *uri_socket = g_hash_table_lookup(qp, "socket"); 400 if (uri_server || uri_port != -1 || !uri_socket) { 401 return -EINVAL; 402 } 403 gsconf->u.q_unix.path = g_strdup(uri_socket); 404 } else { 405 gsconf->u.inet.host = g_strdup(uri_server ? uri_server : "localhost"); 406 if (uri_port > 0) { 407 gsconf->u.inet.port = g_strdup_printf("%d", uri_port); 408 } else { 409 gsconf->u.inet.port = g_strdup_printf("%d", GLUSTER_DEFAULT_PORT); 410 } 411 } 412 413 return 0; 414 } 415 416 static struct glfs *qemu_gluster_glfs_init(BlockdevOptionsGluster *gconf, 417 Error **errp) 418 { 419 struct glfs *glfs; 420 int ret; 421 int old_errno; 422 SocketAddressList *server; 423 uint64_t port; 424 425 glfs = glfs_find_preopened(gconf->volume); 426 if (glfs) { 427 return glfs; 428 } 429 430 glfs = glfs_new(gconf->volume); 431 if (!glfs) { 432 goto out; 433 } 434 435 glfs_set_preopened(gconf->volume, glfs); 436 437 for (server = gconf->server; server; server = server->next) { 438 switch (server->value->type) { 439 case SOCKET_ADDRESS_TYPE_UNIX: 440 ret = glfs_set_volfile_server(glfs, "unix", 441 server->value->u.q_unix.path, 0); 442 break; 443 case SOCKET_ADDRESS_TYPE_INET: 444 if (parse_uint_full(server->value->u.inet.port, 10, &port) < 0 || 445 port > 65535) { 446 error_setg(errp, "'%s' is not a valid port number", 447 server->value->u.inet.port); 448 errno = EINVAL; 449 goto out; 450 } 451 ret = glfs_set_volfile_server(glfs, "tcp", 452 server->value->u.inet.host, 453 (int)port); 454 break; 455 case SOCKET_ADDRESS_TYPE_VSOCK: 456 case SOCKET_ADDRESS_TYPE_FD: 457 default: 458 abort(); 459 } 460 461 if (ret < 0) { 462 goto out; 463 } 464 } 465 466 ret = glfs_set_logging(glfs, gconf->logfile, gconf->debug); 467 if (ret < 0) { 468 goto out; 469 } 470 471 ret = glfs_init(glfs); 472 if (ret) { 473 error_setg(errp, "Gluster connection for volume %s, path %s failed" 474 " to connect", gconf->volume, gconf->path); 475 for (server = gconf->server; server; server = server->next) { 476 if (server->value->type == SOCKET_ADDRESS_TYPE_UNIX) { 477 error_append_hint(errp, "hint: failed on socket %s ", 478 server->value->u.q_unix.path); 479 } else { 480 error_append_hint(errp, "hint: failed on host %s and port %s ", 481 server->value->u.inet.host, 482 server->value->u.inet.port); 483 } 484 } 485 486 error_append_hint(errp, "Please refer to gluster logs for more info\n"); 487 488 /* glfs_init sometimes doesn't set errno although docs suggest that */ 489 if (errno == 0) { 490 errno = EINVAL; 491 } 492 493 goto out; 494 } 495 return glfs; 496 497 out: 498 if (glfs) { 499 old_errno = errno; 500 glfs_clear_preopened(glfs); 501 errno = old_errno; 502 } 503 return NULL; 504 } 505 506 /* 507 * Convert the json formatted command line into qapi. 508 */ 509 static int qemu_gluster_parse_json(BlockdevOptionsGluster *gconf, 510 QDict *options, Error **errp) 511 { 512 QemuOpts *opts; 513 SocketAddress *gsconf = NULL; 514 SocketAddressList **tail; 515 QDict *backing_options = NULL; 516 Error *local_err = NULL; 517 char *str = NULL; 518 const char *ptr; 519 int i, type, num_servers; 520 521 /* create opts info from runtime_json_opts list */ 522 opts = qemu_opts_create(&runtime_json_opts, NULL, 0, &error_abort); 523 if (!qemu_opts_absorb_qdict(opts, options, errp)) { 524 goto out; 525 } 526 527 num_servers = qdict_array_entries(options, GLUSTER_OPT_SERVER_PATTERN); 528 if (num_servers < 1) { 529 error_setg(&local_err, QERR_MISSING_PARAMETER, "server"); 530 goto out; 531 } 532 533 ptr = qemu_opt_get(opts, GLUSTER_OPT_VOLUME); 534 if (!ptr) { 535 error_setg(&local_err, QERR_MISSING_PARAMETER, GLUSTER_OPT_VOLUME); 536 goto out; 537 } 538 gconf->volume = g_strdup(ptr); 539 540 ptr = qemu_opt_get(opts, GLUSTER_OPT_PATH); 541 if (!ptr) { 542 error_setg(&local_err, QERR_MISSING_PARAMETER, GLUSTER_OPT_PATH); 543 goto out; 544 } 545 gconf->path = g_strdup(ptr); 546 qemu_opts_del(opts); 547 tail = &gconf->server; 548 549 for (i = 0; i < num_servers; i++) { 550 str = g_strdup_printf(GLUSTER_OPT_SERVER_PATTERN"%d.", i); 551 qdict_extract_subqdict(options, &backing_options, str); 552 553 /* create opts info from runtime_type_opts list */ 554 opts = qemu_opts_create(&runtime_type_opts, NULL, 0, &error_abort); 555 if (!qemu_opts_absorb_qdict(opts, backing_options, errp)) { 556 goto out; 557 } 558 559 ptr = qemu_opt_get(opts, GLUSTER_OPT_TYPE); 560 if (!ptr) { 561 error_setg(&local_err, QERR_MISSING_PARAMETER, GLUSTER_OPT_TYPE); 562 error_append_hint(&local_err, GERR_INDEX_HINT, i); 563 goto out; 564 565 } 566 gsconf = g_new0(SocketAddress, 1); 567 if (!strcmp(ptr, "tcp")) { 568 ptr = "inet"; /* accept legacy "tcp" */ 569 } 570 type = qapi_enum_parse(&SocketAddressType_lookup, ptr, -1, NULL); 571 if (type != SOCKET_ADDRESS_TYPE_INET 572 && type != SOCKET_ADDRESS_TYPE_UNIX) { 573 error_setg(&local_err, 574 "Parameter '%s' may be 'inet' or 'unix'", 575 GLUSTER_OPT_TYPE); 576 error_append_hint(&local_err, GERR_INDEX_HINT, i); 577 goto out; 578 } 579 gsconf->type = type; 580 qemu_opts_del(opts); 581 582 if (gsconf->type == SOCKET_ADDRESS_TYPE_INET) { 583 /* create opts info from runtime_inet_opts list */ 584 opts = qemu_opts_create(&runtime_inet_opts, NULL, 0, &error_abort); 585 if (!qemu_opts_absorb_qdict(opts, backing_options, errp)) { 586 goto out; 587 } 588 589 ptr = qemu_opt_get(opts, GLUSTER_OPT_HOST); 590 if (!ptr) { 591 error_setg(&local_err, QERR_MISSING_PARAMETER, 592 GLUSTER_OPT_HOST); 593 error_append_hint(&local_err, GERR_INDEX_HINT, i); 594 goto out; 595 } 596 gsconf->u.inet.host = g_strdup(ptr); 597 ptr = qemu_opt_get(opts, GLUSTER_OPT_PORT); 598 if (!ptr) { 599 error_setg(&local_err, QERR_MISSING_PARAMETER, 600 GLUSTER_OPT_PORT); 601 error_append_hint(&local_err, GERR_INDEX_HINT, i); 602 goto out; 603 } 604 gsconf->u.inet.port = g_strdup(ptr); 605 606 /* defend for unsupported fields in InetSocketAddress, 607 * i.e. @ipv4, @ipv6 and @to 608 */ 609 ptr = qemu_opt_get(opts, GLUSTER_OPT_TO); 610 if (ptr) { 611 gsconf->u.inet.has_to = true; 612 } 613 ptr = qemu_opt_get(opts, GLUSTER_OPT_IPV4); 614 if (ptr) { 615 gsconf->u.inet.has_ipv4 = true; 616 } 617 ptr = qemu_opt_get(opts, GLUSTER_OPT_IPV6); 618 if (ptr) { 619 gsconf->u.inet.has_ipv6 = true; 620 } 621 if (gsconf->u.inet.has_to) { 622 error_setg(&local_err, "Parameter 'to' not supported"); 623 goto out; 624 } 625 if (gsconf->u.inet.has_ipv4 || gsconf->u.inet.has_ipv6) { 626 error_setg(&local_err, "Parameters 'ipv4/ipv6' not supported"); 627 goto out; 628 } 629 qemu_opts_del(opts); 630 } else { 631 /* create opts info from runtime_unix_opts list */ 632 opts = qemu_opts_create(&runtime_unix_opts, NULL, 0, &error_abort); 633 if (!qemu_opts_absorb_qdict(opts, backing_options, errp)) { 634 goto out; 635 } 636 637 ptr = qemu_opt_get(opts, GLUSTER_OPT_PATH); 638 if (!ptr) { 639 ptr = qemu_opt_get(opts, GLUSTER_OPT_SOCKET); 640 } else if (qemu_opt_get(opts, GLUSTER_OPT_SOCKET)) { 641 error_setg(&local_err, 642 "Conflicting parameters 'path' and 'socket'"); 643 error_append_hint(&local_err, GERR_INDEX_HINT, i); 644 goto out; 645 } 646 if (!ptr) { 647 error_setg(&local_err, QERR_MISSING_PARAMETER, 648 GLUSTER_OPT_PATH); 649 error_append_hint(&local_err, GERR_INDEX_HINT, i); 650 goto out; 651 } 652 gsconf->u.q_unix.path = g_strdup(ptr); 653 qemu_opts_del(opts); 654 } 655 656 QAPI_LIST_APPEND(tail, gsconf); 657 gsconf = NULL; 658 659 qobject_unref(backing_options); 660 backing_options = NULL; 661 g_free(str); 662 str = NULL; 663 } 664 665 return 0; 666 667 out: 668 error_propagate(errp, local_err); 669 qapi_free_SocketAddress(gsconf); 670 qemu_opts_del(opts); 671 g_free(str); 672 qobject_unref(backing_options); 673 errno = EINVAL; 674 return -errno; 675 } 676 677 /* Converts options given in @filename and the @options QDict into the QAPI 678 * object @gconf. */ 679 static int qemu_gluster_parse(BlockdevOptionsGluster *gconf, 680 const char *filename, 681 QDict *options, Error **errp) 682 { 683 int ret; 684 if (filename) { 685 ret = qemu_gluster_parse_uri(gconf, filename); 686 if (ret < 0) { 687 error_setg(errp, "invalid URI %s", filename); 688 error_append_hint(errp, "Usage: file=gluster[+transport]://" 689 "[host[:port]]volume/path[?socket=...]" 690 "[,file.debug=N]" 691 "[,file.logfile=/path/filename.log]\n"); 692 return ret; 693 } 694 } else { 695 ret = qemu_gluster_parse_json(gconf, options, errp); 696 if (ret < 0) { 697 error_append_hint(errp, "Usage: " 698 "-drive driver=qcow2,file.driver=gluster," 699 "file.volume=testvol,file.path=/path/a.qcow2" 700 "[,file.debug=9]" 701 "[,file.logfile=/path/filename.log]," 702 "file.server.0.type=inet," 703 "file.server.0.host=1.2.3.4," 704 "file.server.0.port=24007," 705 "file.server.1.transport=unix," 706 "file.server.1.path=/var/run/glusterd.socket ..." 707 "\n"); 708 return ret; 709 } 710 } 711 712 return 0; 713 } 714 715 static struct glfs *qemu_gluster_init(BlockdevOptionsGluster *gconf, 716 const char *filename, 717 QDict *options, Error **errp) 718 { 719 int ret; 720 721 ret = qemu_gluster_parse(gconf, filename, options, errp); 722 if (ret < 0) { 723 errno = -ret; 724 return NULL; 725 } 726 727 return qemu_gluster_glfs_init(gconf, errp); 728 } 729 730 /* 731 * AIO callback routine called from GlusterFS thread. 732 */ 733 static void gluster_finish_aiocb(struct glfs_fd *fd, ssize_t ret, 734 #ifdef CONFIG_GLUSTERFS_IOCB_HAS_STAT 735 struct glfs_stat *pre, struct glfs_stat *post, 736 #endif 737 void *arg) 738 { 739 GlusterAIOCB *acb = (GlusterAIOCB *)arg; 740 741 if (!ret || ret == acb->size) { 742 acb->ret = 0; /* Success */ 743 } else if (ret < 0) { 744 acb->ret = -errno; /* Read/Write failed */ 745 } else { 746 acb->ret = -EIO; /* Partial read/write - fail it */ 747 } 748 749 aio_co_schedule(acb->aio_context, acb->coroutine); 750 } 751 752 static void qemu_gluster_parse_flags(int bdrv_flags, int *open_flags) 753 { 754 assert(open_flags != NULL); 755 756 *open_flags |= O_BINARY; 757 758 if (bdrv_flags & BDRV_O_RDWR) { 759 *open_flags |= O_RDWR; 760 } else { 761 *open_flags |= O_RDONLY; 762 } 763 764 if ((bdrv_flags & BDRV_O_NOCACHE)) { 765 *open_flags |= O_DIRECT; 766 } 767 } 768 769 /* 770 * Do SEEK_DATA/HOLE to detect if it is functional. Older broken versions of 771 * gfapi incorrectly return the current offset when SEEK_DATA/HOLE is used. 772 * - Corrected versions return -1 and set errno to EINVAL. 773 * - Versions that support SEEK_DATA/HOLE correctly, will return -1 and set 774 * errno to ENXIO when SEEK_DATA is called with a position of EOF. 775 */ 776 static bool qemu_gluster_test_seek(struct glfs_fd *fd) 777 { 778 off_t ret = 0; 779 780 #if defined SEEK_HOLE && defined SEEK_DATA 781 off_t eof; 782 783 eof = glfs_lseek(fd, 0, SEEK_END); 784 if (eof < 0) { 785 /* this should never occur */ 786 return false; 787 } 788 789 /* this should always fail with ENXIO if SEEK_DATA is supported */ 790 ret = glfs_lseek(fd, eof, SEEK_DATA); 791 #endif 792 793 return (ret < 0) && (errno == ENXIO); 794 } 795 796 static int qemu_gluster_open(BlockDriverState *bs, QDict *options, 797 int bdrv_flags, Error **errp) 798 { 799 BDRVGlusterState *s = bs->opaque; 800 int open_flags = 0; 801 int ret = 0; 802 BlockdevOptionsGluster *gconf = NULL; 803 QemuOpts *opts; 804 const char *filename, *logfile; 805 806 opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort); 807 if (!qemu_opts_absorb_qdict(opts, options, errp)) { 808 ret = -EINVAL; 809 goto out; 810 } 811 812 warn_report_once("'gluster' is deprecated"); 813 814 filename = qemu_opt_get(opts, GLUSTER_OPT_FILENAME); 815 816 s->debug = qemu_opt_get_number(opts, GLUSTER_OPT_DEBUG, 817 GLUSTER_DEBUG_DEFAULT); 818 if (s->debug < 0) { 819 s->debug = 0; 820 } else if (s->debug > GLUSTER_DEBUG_MAX) { 821 s->debug = GLUSTER_DEBUG_MAX; 822 } 823 824 gconf = g_new0(BlockdevOptionsGluster, 1); 825 gconf->debug = s->debug; 826 gconf->has_debug = true; 827 828 logfile = qemu_opt_get(opts, GLUSTER_OPT_LOGFILE); 829 s->logfile = g_strdup(logfile ? logfile : GLUSTER_LOGFILE_DEFAULT); 830 831 gconf->logfile = g_strdup(s->logfile); 832 833 s->glfs = qemu_gluster_init(gconf, filename, options, errp); 834 if (!s->glfs) { 835 ret = -errno; 836 goto out; 837 } 838 839 #ifdef CONFIG_GLUSTERFS_XLATOR_OPT 840 /* Without this, if fsync fails for a recoverable reason (for instance, 841 * ENOSPC), gluster will dump its cache, preventing retries. This means 842 * almost certain data loss. Not all gluster versions support the 843 * 'resync-failed-syncs-after-fsync' key value, but there is no way to 844 * discover during runtime if it is supported (this api returns success for 845 * unknown key/value pairs) */ 846 ret = glfs_set_xlator_option(s->glfs, "*-write-behind", 847 "resync-failed-syncs-after-fsync", 848 "on"); 849 if (ret < 0) { 850 error_setg_errno(errp, errno, "Unable to set xlator key/value pair"); 851 ret = -errno; 852 goto out; 853 } 854 #endif 855 856 qemu_gluster_parse_flags(bdrv_flags, &open_flags); 857 858 s->fd = glfs_open(s->glfs, gconf->path, open_flags); 859 ret = s->fd ? 0 : -errno; 860 861 if (ret == -EACCES || ret == -EROFS) { 862 /* Try to degrade to read-only, but if it doesn't work, still use the 863 * normal error message. */ 864 bdrv_graph_rdlock_main_loop(); 865 if (bdrv_apply_auto_read_only(bs, NULL, NULL) == 0) { 866 open_flags = (open_flags & ~O_RDWR) | O_RDONLY; 867 s->fd = glfs_open(s->glfs, gconf->path, open_flags); 868 ret = s->fd ? 0 : -errno; 869 } 870 bdrv_graph_rdunlock_main_loop(); 871 } 872 873 s->supports_seek_data = qemu_gluster_test_seek(s->fd); 874 875 out: 876 qemu_opts_del(opts); 877 qapi_free_BlockdevOptionsGluster(gconf); 878 if (!ret) { 879 return ret; 880 } 881 g_free(s->logfile); 882 if (s->fd) { 883 glfs_close(s->fd); 884 } 885 886 glfs_clear_preopened(s->glfs); 887 888 return ret; 889 } 890 891 static void qemu_gluster_refresh_limits(BlockDriverState *bs, Error **errp) 892 { 893 bs->bl.max_transfer = GLUSTER_MAX_TRANSFER; 894 bs->bl.max_pdiscard = MIN(SIZE_MAX, INT64_MAX); 895 } 896 897 static int qemu_gluster_reopen_prepare(BDRVReopenState *state, 898 BlockReopenQueue *queue, Error **errp) 899 { 900 int ret = 0; 901 BDRVGlusterState *s; 902 BDRVGlusterReopenState *reop_s; 903 BlockdevOptionsGluster *gconf; 904 int open_flags = 0; 905 906 assert(state != NULL); 907 assert(state->bs != NULL); 908 909 s = state->bs->opaque; 910 911 state->opaque = g_new0(BDRVGlusterReopenState, 1); 912 reop_s = state->opaque; 913 914 qemu_gluster_parse_flags(state->flags, &open_flags); 915 916 gconf = g_new0(BlockdevOptionsGluster, 1); 917 gconf->debug = s->debug; 918 gconf->has_debug = true; 919 gconf->logfile = g_strdup(s->logfile); 920 921 /* 922 * If 'state->bs->exact_filename' is empty, 'state->options' should contain 923 * the JSON parameters already parsed. 924 */ 925 if (state->bs->exact_filename[0] != '\0') { 926 reop_s->glfs = qemu_gluster_init(gconf, state->bs->exact_filename, NULL, 927 errp); 928 } else { 929 reop_s->glfs = qemu_gluster_init(gconf, NULL, state->options, errp); 930 } 931 if (reop_s->glfs == NULL) { 932 ret = -errno; 933 goto exit; 934 } 935 936 #ifdef CONFIG_GLUSTERFS_XLATOR_OPT 937 ret = glfs_set_xlator_option(reop_s->glfs, "*-write-behind", 938 "resync-failed-syncs-after-fsync", "on"); 939 if (ret < 0) { 940 error_setg_errno(errp, errno, "Unable to set xlator key/value pair"); 941 ret = -errno; 942 goto exit; 943 } 944 #endif 945 946 reop_s->fd = glfs_open(reop_s->glfs, gconf->path, open_flags); 947 if (reop_s->fd == NULL) { 948 /* reops->glfs will be cleaned up in _abort */ 949 ret = -errno; 950 goto exit; 951 } 952 953 exit: 954 /* state->opaque will be freed in either the _abort or _commit */ 955 qapi_free_BlockdevOptionsGluster(gconf); 956 return ret; 957 } 958 959 static void qemu_gluster_reopen_commit(BDRVReopenState *state) 960 { 961 BDRVGlusterReopenState *reop_s = state->opaque; 962 BDRVGlusterState *s = state->bs->opaque; 963 964 965 /* close the old */ 966 if (s->fd) { 967 glfs_close(s->fd); 968 } 969 970 glfs_clear_preopened(s->glfs); 971 972 /* use the newly opened image / connection */ 973 s->fd = reop_s->fd; 974 s->glfs = reop_s->glfs; 975 976 g_free(state->opaque); 977 state->opaque = NULL; 978 979 return; 980 } 981 982 983 static void qemu_gluster_reopen_abort(BDRVReopenState *state) 984 { 985 BDRVGlusterReopenState *reop_s = state->opaque; 986 987 if (reop_s == NULL) { 988 return; 989 } 990 991 if (reop_s->fd) { 992 glfs_close(reop_s->fd); 993 } 994 995 glfs_clear_preopened(reop_s->glfs); 996 997 g_free(state->opaque); 998 state->opaque = NULL; 999 1000 return; 1001 } 1002 1003 #ifdef CONFIG_GLUSTERFS_ZEROFILL 1004 static coroutine_fn int qemu_gluster_co_pwrite_zeroes(BlockDriverState *bs, 1005 int64_t offset, 1006 int64_t bytes, 1007 BdrvRequestFlags flags) 1008 { 1009 int ret; 1010 GlusterAIOCB acb; 1011 BDRVGlusterState *s = bs->opaque; 1012 1013 acb.size = bytes; 1014 acb.ret = 0; 1015 acb.coroutine = qemu_coroutine_self(); 1016 acb.aio_context = bdrv_get_aio_context(bs); 1017 1018 ret = glfs_zerofill_async(s->fd, offset, bytes, gluster_finish_aiocb, &acb); 1019 if (ret < 0) { 1020 return -errno; 1021 } 1022 1023 qemu_coroutine_yield(); 1024 return acb.ret; 1025 } 1026 #endif 1027 1028 static int qemu_gluster_do_truncate(struct glfs_fd *fd, int64_t offset, 1029 PreallocMode prealloc, Error **errp) 1030 { 1031 int64_t current_length; 1032 1033 current_length = glfs_lseek(fd, 0, SEEK_END); 1034 if (current_length < 0) { 1035 error_setg_errno(errp, errno, "Failed to determine current size"); 1036 return -errno; 1037 } 1038 1039 if (current_length > offset && prealloc != PREALLOC_MODE_OFF) { 1040 error_setg(errp, "Cannot use preallocation for shrinking files"); 1041 return -ENOTSUP; 1042 } 1043 1044 if (current_length == offset) { 1045 return 0; 1046 } 1047 1048 switch (prealloc) { 1049 #ifdef CONFIG_GLUSTERFS_FALLOCATE 1050 case PREALLOC_MODE_FALLOC: 1051 if (glfs_fallocate(fd, 0, current_length, offset - current_length)) { 1052 error_setg_errno(errp, errno, "Could not preallocate data"); 1053 return -errno; 1054 } 1055 break; 1056 #endif /* CONFIG_GLUSTERFS_FALLOCATE */ 1057 #ifdef CONFIG_GLUSTERFS_ZEROFILL 1058 case PREALLOC_MODE_FULL: 1059 if (glfs_ftruncate(fd, offset)) { 1060 error_setg_errno(errp, errno, "Could not resize file"); 1061 return -errno; 1062 } 1063 if (glfs_zerofill(fd, current_length, offset - current_length)) { 1064 error_setg_errno(errp, errno, "Could not zerofill the new area"); 1065 return -errno; 1066 } 1067 break; 1068 #endif /* CONFIG_GLUSTERFS_ZEROFILL */ 1069 case PREALLOC_MODE_OFF: 1070 if (glfs_ftruncate(fd, offset)) { 1071 error_setg_errno(errp, errno, "Could not resize file"); 1072 return -errno; 1073 } 1074 break; 1075 default: 1076 error_setg(errp, "Unsupported preallocation mode: %s", 1077 PreallocMode_str(prealloc)); 1078 return -EINVAL; 1079 } 1080 1081 return 0; 1082 } 1083 1084 static int qemu_gluster_co_create(BlockdevCreateOptions *options, 1085 Error **errp) 1086 { 1087 BlockdevCreateOptionsGluster *opts = &options->u.gluster; 1088 struct glfs *glfs; 1089 struct glfs_fd *fd = NULL; 1090 int ret = 0; 1091 1092 assert(options->driver == BLOCKDEV_DRIVER_GLUSTER); 1093 1094 glfs = qemu_gluster_glfs_init(opts->location, errp); 1095 if (!glfs) { 1096 ret = -errno; 1097 goto out; 1098 } 1099 1100 fd = glfs_creat(glfs, opts->location->path, 1101 O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, S_IRUSR | S_IWUSR); 1102 if (!fd) { 1103 ret = -errno; 1104 goto out; 1105 } 1106 1107 ret = qemu_gluster_do_truncate(fd, opts->size, opts->preallocation, errp); 1108 1109 out: 1110 if (fd) { 1111 if (glfs_close(fd) != 0 && ret == 0) { 1112 ret = -errno; 1113 } 1114 } 1115 glfs_clear_preopened(glfs); 1116 return ret; 1117 } 1118 1119 static int coroutine_fn qemu_gluster_co_create_opts(BlockDriver *drv, 1120 const char *filename, 1121 QemuOpts *opts, 1122 Error **errp) 1123 { 1124 BlockdevCreateOptions *options; 1125 BlockdevCreateOptionsGluster *gopts; 1126 BlockdevOptionsGluster *gconf; 1127 char *tmp = NULL; 1128 Error *local_err = NULL; 1129 int ret; 1130 1131 options = g_new0(BlockdevCreateOptions, 1); 1132 options->driver = BLOCKDEV_DRIVER_GLUSTER; 1133 gopts = &options->u.gluster; 1134 1135 gconf = g_new0(BlockdevOptionsGluster, 1); 1136 gopts->location = gconf; 1137 1138 gopts->size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0), 1139 BDRV_SECTOR_SIZE); 1140 1141 tmp = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC); 1142 gopts->preallocation = qapi_enum_parse(&PreallocMode_lookup, tmp, 1143 PREALLOC_MODE_OFF, &local_err); 1144 g_free(tmp); 1145 if (local_err) { 1146 error_propagate(errp, local_err); 1147 ret = -EINVAL; 1148 goto fail; 1149 } 1150 1151 gconf->debug = qemu_opt_get_number_del(opts, GLUSTER_OPT_DEBUG, 1152 GLUSTER_DEBUG_DEFAULT); 1153 if (gconf->debug < 0) { 1154 gconf->debug = 0; 1155 } else if (gconf->debug > GLUSTER_DEBUG_MAX) { 1156 gconf->debug = GLUSTER_DEBUG_MAX; 1157 } 1158 gconf->has_debug = true; 1159 1160 gconf->logfile = qemu_opt_get_del(opts, GLUSTER_OPT_LOGFILE); 1161 if (!gconf->logfile) { 1162 gconf->logfile = g_strdup(GLUSTER_LOGFILE_DEFAULT); 1163 } 1164 1165 ret = qemu_gluster_parse(gconf, filename, NULL, errp); 1166 if (ret < 0) { 1167 goto fail; 1168 } 1169 1170 ret = qemu_gluster_co_create(options, errp); 1171 if (ret < 0) { 1172 goto fail; 1173 } 1174 1175 ret = 0; 1176 fail: 1177 qapi_free_BlockdevCreateOptions(options); 1178 return ret; 1179 } 1180 1181 static coroutine_fn int qemu_gluster_co_rw(BlockDriverState *bs, 1182 int64_t sector_num, int nb_sectors, 1183 QEMUIOVector *qiov, int write) 1184 { 1185 int ret; 1186 GlusterAIOCB acb; 1187 BDRVGlusterState *s = bs->opaque; 1188 size_t size = nb_sectors * BDRV_SECTOR_SIZE; 1189 off_t offset = sector_num * BDRV_SECTOR_SIZE; 1190 1191 acb.size = size; 1192 acb.ret = 0; 1193 acb.coroutine = qemu_coroutine_self(); 1194 acb.aio_context = bdrv_get_aio_context(bs); 1195 1196 if (write) { 1197 ret = glfs_pwritev_async(s->fd, qiov->iov, qiov->niov, offset, 0, 1198 gluster_finish_aiocb, &acb); 1199 } else { 1200 ret = glfs_preadv_async(s->fd, qiov->iov, qiov->niov, offset, 0, 1201 gluster_finish_aiocb, &acb); 1202 } 1203 1204 if (ret < 0) { 1205 return -errno; 1206 } 1207 1208 qemu_coroutine_yield(); 1209 return acb.ret; 1210 } 1211 1212 static coroutine_fn int qemu_gluster_co_truncate(BlockDriverState *bs, 1213 int64_t offset, 1214 bool exact, 1215 PreallocMode prealloc, 1216 BdrvRequestFlags flags, 1217 Error **errp) 1218 { 1219 BDRVGlusterState *s = bs->opaque; 1220 return qemu_gluster_do_truncate(s->fd, offset, prealloc, errp); 1221 } 1222 1223 static coroutine_fn int qemu_gluster_co_readv(BlockDriverState *bs, 1224 int64_t sector_num, 1225 int nb_sectors, 1226 QEMUIOVector *qiov) 1227 { 1228 return qemu_gluster_co_rw(bs, sector_num, nb_sectors, qiov, 0); 1229 } 1230 1231 static coroutine_fn int qemu_gluster_co_writev(BlockDriverState *bs, 1232 int64_t sector_num, 1233 int nb_sectors, 1234 QEMUIOVector *qiov, 1235 int flags) 1236 { 1237 return qemu_gluster_co_rw(bs, sector_num, nb_sectors, qiov, 1); 1238 } 1239 1240 static void qemu_gluster_close(BlockDriverState *bs) 1241 { 1242 BDRVGlusterState *s = bs->opaque; 1243 1244 g_free(s->logfile); 1245 if (s->fd) { 1246 glfs_close(s->fd); 1247 s->fd = NULL; 1248 } 1249 glfs_clear_preopened(s->glfs); 1250 } 1251 1252 static coroutine_fn int qemu_gluster_co_flush_to_disk(BlockDriverState *bs) 1253 { 1254 int ret; 1255 GlusterAIOCB acb; 1256 BDRVGlusterState *s = bs->opaque; 1257 1258 acb.size = 0; 1259 acb.ret = 0; 1260 acb.coroutine = qemu_coroutine_self(); 1261 acb.aio_context = bdrv_get_aio_context(bs); 1262 1263 ret = glfs_fsync_async(s->fd, gluster_finish_aiocb, &acb); 1264 if (ret < 0) { 1265 ret = -errno; 1266 goto error; 1267 } 1268 1269 qemu_coroutine_yield(); 1270 if (acb.ret < 0) { 1271 ret = acb.ret; 1272 goto error; 1273 } 1274 1275 return acb.ret; 1276 1277 error: 1278 /* Some versions of Gluster (3.5.6 -> 3.5.8?) will not retain its cache 1279 * after a fsync failure, so we have no way of allowing the guest to safely 1280 * continue. Gluster versions prior to 3.5.6 don't retain the cache 1281 * either, but will invalidate the fd on error, so this is again our only 1282 * option. 1283 * 1284 * The 'resync-failed-syncs-after-fsync' xlator option for the 1285 * write-behind cache will cause later gluster versions to retain its 1286 * cache after error, so long as the fd remains open. However, we 1287 * currently have no way of knowing if this option is supported. 1288 * 1289 * TODO: Once gluster provides a way for us to determine if the option 1290 * is supported, bypass the closure and setting drv to NULL. */ 1291 qemu_gluster_close(bs); 1292 bs->drv = NULL; 1293 return ret; 1294 } 1295 1296 #ifdef CONFIG_GLUSTERFS_DISCARD 1297 static coroutine_fn int qemu_gluster_co_pdiscard(BlockDriverState *bs, 1298 int64_t offset, int64_t bytes) 1299 { 1300 int ret; 1301 GlusterAIOCB acb; 1302 BDRVGlusterState *s = bs->opaque; 1303 1304 assert(bytes <= SIZE_MAX); /* rely on max_pdiscard */ 1305 1306 acb.size = 0; 1307 acb.ret = 0; 1308 acb.coroutine = qemu_coroutine_self(); 1309 acb.aio_context = bdrv_get_aio_context(bs); 1310 1311 ret = glfs_discard_async(s->fd, offset, bytes, gluster_finish_aiocb, &acb); 1312 if (ret < 0) { 1313 return -errno; 1314 } 1315 1316 qemu_coroutine_yield(); 1317 return acb.ret; 1318 } 1319 #endif 1320 1321 static int64_t coroutine_fn qemu_gluster_co_getlength(BlockDriverState *bs) 1322 { 1323 BDRVGlusterState *s = bs->opaque; 1324 int64_t ret; 1325 1326 ret = glfs_lseek(s->fd, 0, SEEK_END); 1327 if (ret < 0) { 1328 return -errno; 1329 } else { 1330 return ret; 1331 } 1332 } 1333 1334 static int64_t coroutine_fn 1335 qemu_gluster_co_get_allocated_file_size(BlockDriverState *bs) 1336 { 1337 BDRVGlusterState *s = bs->opaque; 1338 struct stat st; 1339 int ret; 1340 1341 ret = glfs_fstat(s->fd, &st); 1342 if (ret < 0) { 1343 return -errno; 1344 } else { 1345 return st.st_blocks * 512; 1346 } 1347 } 1348 1349 /* 1350 * Find allocation range in @bs around offset @start. 1351 * May change underlying file descriptor's file offset. 1352 * If @start is not in a hole, store @start in @data, and the 1353 * beginning of the next hole in @hole, and return 0. 1354 * If @start is in a non-trailing hole, store @start in @hole and the 1355 * beginning of the next non-hole in @data, and return 0. 1356 * If @start is in a trailing hole or beyond EOF, return -ENXIO. 1357 * If we can't find out, return a negative errno other than -ENXIO. 1358 * 1359 * (Shamefully copied from file-posix.c, only minuscule adaptions.) 1360 */ 1361 static int find_allocation(BlockDriverState *bs, off_t start, 1362 off_t *data, off_t *hole) 1363 { 1364 BDRVGlusterState *s = bs->opaque; 1365 1366 if (!s->supports_seek_data) { 1367 goto exit; 1368 } 1369 1370 #if defined SEEK_HOLE && defined SEEK_DATA 1371 off_t offs; 1372 1373 /* 1374 * SEEK_DATA cases: 1375 * D1. offs == start: start is in data 1376 * D2. offs > start: start is in a hole, next data at offs 1377 * D3. offs < 0, errno = ENXIO: either start is in a trailing hole 1378 * or start is beyond EOF 1379 * If the latter happens, the file has been truncated behind 1380 * our back since we opened it. All bets are off then. 1381 * Treating like a trailing hole is simplest. 1382 * D4. offs < 0, errno != ENXIO: we learned nothing 1383 */ 1384 offs = glfs_lseek(s->fd, start, SEEK_DATA); 1385 if (offs < 0) { 1386 return -errno; /* D3 or D4 */ 1387 } 1388 1389 if (offs < start) { 1390 /* This is not a valid return by lseek(). We are safe to just return 1391 * -EIO in this case, and we'll treat it like D4. Unfortunately some 1392 * versions of gluster server will return offs < start, so an assert 1393 * here will unnecessarily abort QEMU. */ 1394 return -EIO; 1395 } 1396 1397 if (offs > start) { 1398 /* D2: in hole, next data at offs */ 1399 *hole = start; 1400 *data = offs; 1401 return 0; 1402 } 1403 1404 /* D1: in data, end not yet known */ 1405 1406 /* 1407 * SEEK_HOLE cases: 1408 * H1. offs == start: start is in a hole 1409 * If this happens here, a hole has been dug behind our back 1410 * since the previous lseek(). 1411 * H2. offs > start: either start is in data, next hole at offs, 1412 * or start is in trailing hole, EOF at offs 1413 * Linux treats trailing holes like any other hole: offs == 1414 * start. Solaris seeks to EOF instead: offs > start (blech). 1415 * If that happens here, a hole has been dug behind our back 1416 * since the previous lseek(). 1417 * H3. offs < 0, errno = ENXIO: start is beyond EOF 1418 * If this happens, the file has been truncated behind our 1419 * back since we opened it. Treat it like a trailing hole. 1420 * H4. offs < 0, errno != ENXIO: we learned nothing 1421 * Pretend we know nothing at all, i.e. "forget" about D1. 1422 */ 1423 offs = glfs_lseek(s->fd, start, SEEK_HOLE); 1424 if (offs < 0) { 1425 return -errno; /* D1 and (H3 or H4) */ 1426 } 1427 1428 if (offs < start) { 1429 /* This is not a valid return by lseek(). We are safe to just return 1430 * -EIO in this case, and we'll treat it like H4. Unfortunately some 1431 * versions of gluster server will return offs < start, so an assert 1432 * here will unnecessarily abort QEMU. */ 1433 return -EIO; 1434 } 1435 1436 if (offs > start) { 1437 /* 1438 * D1 and H2: either in data, next hole at offs, or it was in 1439 * data but is now in a trailing hole. In the latter case, 1440 * all bets are off. Treating it as if it there was data all 1441 * the way to EOF is safe, so simply do that. 1442 */ 1443 *data = start; 1444 *hole = offs; 1445 return 0; 1446 } 1447 1448 /* D1 and H1 */ 1449 return -EBUSY; 1450 #endif 1451 1452 exit: 1453 return -ENOTSUP; 1454 } 1455 1456 /* 1457 * Returns the allocation status of the specified offset. 1458 * 1459 * The block layer guarantees 'offset' and 'bytes' are within bounds. 1460 * 1461 * 'pnum' is set to the number of bytes (including and immediately following 1462 * the specified offset) that are known to be in the same 1463 * allocated/unallocated state. 1464 * 1465 * 'bytes' is a soft cap for 'pnum'. If the information is free, 'pnum' may 1466 * well exceed it. 1467 * 1468 * (Based on raw_co_block_status() from file-posix.c.) 1469 */ 1470 static int coroutine_fn qemu_gluster_co_block_status(BlockDriverState *bs, 1471 bool want_zero, 1472 int64_t offset, 1473 int64_t bytes, 1474 int64_t *pnum, 1475 int64_t *map, 1476 BlockDriverState **file) 1477 { 1478 BDRVGlusterState *s = bs->opaque; 1479 off_t data = 0, hole = 0; 1480 int ret = -EINVAL; 1481 1482 assert(QEMU_IS_ALIGNED(offset | bytes, bs->bl.request_alignment)); 1483 1484 if (!s->fd) { 1485 return ret; 1486 } 1487 1488 if (!want_zero) { 1489 *pnum = bytes; 1490 *map = offset; 1491 *file = bs; 1492 return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID; 1493 } 1494 1495 ret = find_allocation(bs, offset, &data, &hole); 1496 if (ret == -ENXIO) { 1497 /* Trailing hole */ 1498 *pnum = bytes; 1499 ret = BDRV_BLOCK_ZERO; 1500 } else if (ret < 0) { 1501 /* No info available, so pretend there are no holes */ 1502 *pnum = bytes; 1503 ret = BDRV_BLOCK_DATA; 1504 } else if (data == offset) { 1505 /* On a data extent, compute bytes to the end of the extent, 1506 * possibly including a partial sector at EOF. */ 1507 *pnum = hole - offset; 1508 1509 /* 1510 * We are not allowed to return partial sectors, though, so 1511 * round up if necessary. 1512 */ 1513 if (!QEMU_IS_ALIGNED(*pnum, bs->bl.request_alignment)) { 1514 int64_t file_length = qemu_gluster_co_getlength(bs); 1515 if (file_length > 0) { 1516 /* Ignore errors, this is just a safeguard */ 1517 assert(hole == file_length); 1518 } 1519 *pnum = ROUND_UP(*pnum, bs->bl.request_alignment); 1520 } 1521 1522 ret = BDRV_BLOCK_DATA; 1523 } else { 1524 /* On a hole, compute bytes to the beginning of the next extent. */ 1525 assert(hole == offset); 1526 *pnum = data - offset; 1527 ret = BDRV_BLOCK_ZERO; 1528 } 1529 1530 *map = offset; 1531 *file = bs; 1532 1533 return ret | BDRV_BLOCK_OFFSET_VALID; 1534 } 1535 1536 1537 static const char *const gluster_strong_open_opts[] = { 1538 GLUSTER_OPT_VOLUME, 1539 GLUSTER_OPT_PATH, 1540 GLUSTER_OPT_TYPE, 1541 GLUSTER_OPT_SERVER_PATTERN, 1542 GLUSTER_OPT_HOST, 1543 GLUSTER_OPT_PORT, 1544 GLUSTER_OPT_TO, 1545 GLUSTER_OPT_IPV4, 1546 GLUSTER_OPT_IPV6, 1547 GLUSTER_OPT_SOCKET, 1548 1549 NULL 1550 }; 1551 1552 static BlockDriver bdrv_gluster = { 1553 .format_name = "gluster", 1554 .protocol_name = "gluster", 1555 .instance_size = sizeof(BDRVGlusterState), 1556 .bdrv_open = qemu_gluster_open, 1557 .bdrv_reopen_prepare = qemu_gluster_reopen_prepare, 1558 .bdrv_reopen_commit = qemu_gluster_reopen_commit, 1559 .bdrv_reopen_abort = qemu_gluster_reopen_abort, 1560 .bdrv_close = qemu_gluster_close, 1561 .bdrv_co_create = qemu_gluster_co_create, 1562 .bdrv_co_create_opts = qemu_gluster_co_create_opts, 1563 .bdrv_co_getlength = qemu_gluster_co_getlength, 1564 .bdrv_co_get_allocated_file_size = qemu_gluster_co_get_allocated_file_size, 1565 .bdrv_co_truncate = qemu_gluster_co_truncate, 1566 .bdrv_co_readv = qemu_gluster_co_readv, 1567 .bdrv_co_writev = qemu_gluster_co_writev, 1568 .bdrv_co_flush_to_disk = qemu_gluster_co_flush_to_disk, 1569 #ifdef CONFIG_GLUSTERFS_DISCARD 1570 .bdrv_co_pdiscard = qemu_gluster_co_pdiscard, 1571 #endif 1572 #ifdef CONFIG_GLUSTERFS_ZEROFILL 1573 .bdrv_co_pwrite_zeroes = qemu_gluster_co_pwrite_zeroes, 1574 #endif 1575 .bdrv_co_block_status = qemu_gluster_co_block_status, 1576 .bdrv_refresh_limits = qemu_gluster_refresh_limits, 1577 .create_opts = &qemu_gluster_create_opts, 1578 .strong_runtime_opts = gluster_strong_open_opts, 1579 }; 1580 1581 static BlockDriver bdrv_gluster_tcp = { 1582 .format_name = "gluster", 1583 .protocol_name = "gluster+tcp", 1584 .instance_size = sizeof(BDRVGlusterState), 1585 .bdrv_open = qemu_gluster_open, 1586 .bdrv_reopen_prepare = qemu_gluster_reopen_prepare, 1587 .bdrv_reopen_commit = qemu_gluster_reopen_commit, 1588 .bdrv_reopen_abort = qemu_gluster_reopen_abort, 1589 .bdrv_close = qemu_gluster_close, 1590 .bdrv_co_create = qemu_gluster_co_create, 1591 .bdrv_co_create_opts = qemu_gluster_co_create_opts, 1592 .bdrv_co_getlength = qemu_gluster_co_getlength, 1593 .bdrv_co_get_allocated_file_size = qemu_gluster_co_get_allocated_file_size, 1594 .bdrv_co_truncate = qemu_gluster_co_truncate, 1595 .bdrv_co_readv = qemu_gluster_co_readv, 1596 .bdrv_co_writev = qemu_gluster_co_writev, 1597 .bdrv_co_flush_to_disk = qemu_gluster_co_flush_to_disk, 1598 #ifdef CONFIG_GLUSTERFS_DISCARD 1599 .bdrv_co_pdiscard = qemu_gluster_co_pdiscard, 1600 #endif 1601 #ifdef CONFIG_GLUSTERFS_ZEROFILL 1602 .bdrv_co_pwrite_zeroes = qemu_gluster_co_pwrite_zeroes, 1603 #endif 1604 .bdrv_co_block_status = qemu_gluster_co_block_status, 1605 .bdrv_refresh_limits = qemu_gluster_refresh_limits, 1606 .create_opts = &qemu_gluster_create_opts, 1607 .strong_runtime_opts = gluster_strong_open_opts, 1608 }; 1609 1610 static BlockDriver bdrv_gluster_unix = { 1611 .format_name = "gluster", 1612 .protocol_name = "gluster+unix", 1613 .instance_size = sizeof(BDRVGlusterState), 1614 .bdrv_open = qemu_gluster_open, 1615 .bdrv_reopen_prepare = qemu_gluster_reopen_prepare, 1616 .bdrv_reopen_commit = qemu_gluster_reopen_commit, 1617 .bdrv_reopen_abort = qemu_gluster_reopen_abort, 1618 .bdrv_close = qemu_gluster_close, 1619 .bdrv_co_create = qemu_gluster_co_create, 1620 .bdrv_co_create_opts = qemu_gluster_co_create_opts, 1621 .bdrv_co_getlength = qemu_gluster_co_getlength, 1622 .bdrv_co_get_allocated_file_size = qemu_gluster_co_get_allocated_file_size, 1623 .bdrv_co_truncate = qemu_gluster_co_truncate, 1624 .bdrv_co_readv = qemu_gluster_co_readv, 1625 .bdrv_co_writev = qemu_gluster_co_writev, 1626 .bdrv_co_flush_to_disk = qemu_gluster_co_flush_to_disk, 1627 #ifdef CONFIG_GLUSTERFS_DISCARD 1628 .bdrv_co_pdiscard = qemu_gluster_co_pdiscard, 1629 #endif 1630 #ifdef CONFIG_GLUSTERFS_ZEROFILL 1631 .bdrv_co_pwrite_zeroes = qemu_gluster_co_pwrite_zeroes, 1632 #endif 1633 .bdrv_co_block_status = qemu_gluster_co_block_status, 1634 .bdrv_refresh_limits = qemu_gluster_refresh_limits, 1635 .create_opts = &qemu_gluster_create_opts, 1636 .strong_runtime_opts = gluster_strong_open_opts, 1637 }; 1638 1639 static void bdrv_gluster_init(void) 1640 { 1641 bdrv_register(&bdrv_gluster_unix); 1642 bdrv_register(&bdrv_gluster_tcp); 1643 bdrv_register(&bdrv_gluster); 1644 } 1645 1646 block_init(bdrv_gluster_init); 1647