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_int.h" 15 #include "block/qdict.h" 16 #include "qapi/error.h" 17 #include "qapi/qmp/qdict.h" 18 #include "qapi/qmp/qerror.h" 19 #include "qemu/uri.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, char *path) 292 { 293 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 SocketAddress *gsconf; 352 URI *uri; 353 QueryParams *qp = NULL; 354 bool is_unix = false; 355 int ret = 0; 356 357 uri = uri_parse(filename); 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 if (!uri->scheme || !strcmp(uri->scheme, "gluster")) { 367 gsconf->type = SOCKET_ADDRESS_TYPE_INET; 368 } else if (!strcmp(uri->scheme, "gluster+tcp")) { 369 gsconf->type = SOCKET_ADDRESS_TYPE_INET; 370 } else if (!strcmp(uri->scheme, "gluster+unix")) { 371 gsconf->type = SOCKET_ADDRESS_TYPE_UNIX; 372 is_unix = true; 373 } else if (!strcmp(uri->scheme, "gluster+rdma")) { 374 gsconf->type = SOCKET_ADDRESS_TYPE_INET; 375 warn_report("rdma feature is not supported, falling back to tcp"); 376 } else { 377 ret = -EINVAL; 378 goto out; 379 } 380 381 ret = parse_volume_options(gconf, uri->path); 382 if (ret < 0) { 383 goto out; 384 } 385 386 qp = query_params_parse(uri->query); 387 if (qp->n > 1 || (is_unix && !qp->n) || (!is_unix && qp->n)) { 388 ret = -EINVAL; 389 goto out; 390 } 391 392 if (is_unix) { 393 if (uri->server || uri->port) { 394 ret = -EINVAL; 395 goto out; 396 } 397 if (strcmp(qp->p[0].name, "socket")) { 398 ret = -EINVAL; 399 goto out; 400 } 401 gsconf->u.q_unix.path = g_strdup(qp->p[0].value); 402 } else { 403 gsconf->u.inet.host = g_strdup(uri->server ? uri->server : "localhost"); 404 if (uri->port) { 405 gsconf->u.inet.port = g_strdup_printf("%d", uri->port); 406 } else { 407 gsconf->u.inet.port = g_strdup_printf("%d", GLUSTER_DEFAULT_PORT); 408 } 409 } 410 411 out: 412 if (qp) { 413 query_params_free(qp); 414 } 415 uri_free(uri); 416 return ret; 417 } 418 419 static struct glfs *qemu_gluster_glfs_init(BlockdevOptionsGluster *gconf, 420 Error **errp) 421 { 422 struct glfs *glfs; 423 int ret; 424 int old_errno; 425 SocketAddressList *server; 426 unsigned long long port; 427 428 glfs = glfs_find_preopened(gconf->volume); 429 if (glfs) { 430 return glfs; 431 } 432 433 glfs = glfs_new(gconf->volume); 434 if (!glfs) { 435 goto out; 436 } 437 438 glfs_set_preopened(gconf->volume, glfs); 439 440 for (server = gconf->server; server; server = server->next) { 441 switch (server->value->type) { 442 case SOCKET_ADDRESS_TYPE_UNIX: 443 ret = glfs_set_volfile_server(glfs, "unix", 444 server->value->u.q_unix.path, 0); 445 break; 446 case SOCKET_ADDRESS_TYPE_INET: 447 if (parse_uint_full(server->value->u.inet.port, &port, 10) < 0 || 448 port > 65535) { 449 error_setg(errp, "'%s' is not a valid port number", 450 server->value->u.inet.port); 451 errno = EINVAL; 452 goto out; 453 } 454 ret = glfs_set_volfile_server(glfs, "tcp", 455 server->value->u.inet.host, 456 (int)port); 457 break; 458 case SOCKET_ADDRESS_TYPE_VSOCK: 459 case SOCKET_ADDRESS_TYPE_FD: 460 default: 461 abort(); 462 } 463 464 if (ret < 0) { 465 goto out; 466 } 467 } 468 469 ret = glfs_set_logging(glfs, gconf->logfile, gconf->debug); 470 if (ret < 0) { 471 goto out; 472 } 473 474 ret = glfs_init(glfs); 475 if (ret) { 476 error_setg(errp, "Gluster connection for volume %s, path %s failed" 477 " to connect", gconf->volume, gconf->path); 478 for (server = gconf->server; server; server = server->next) { 479 if (server->value->type == SOCKET_ADDRESS_TYPE_UNIX) { 480 error_append_hint(errp, "hint: failed on socket %s ", 481 server->value->u.q_unix.path); 482 } else { 483 error_append_hint(errp, "hint: failed on host %s and port %s ", 484 server->value->u.inet.host, 485 server->value->u.inet.port); 486 } 487 } 488 489 error_append_hint(errp, "Please refer to gluster logs for more info\n"); 490 491 /* glfs_init sometimes doesn't set errno although docs suggest that */ 492 if (errno == 0) { 493 errno = EINVAL; 494 } 495 496 goto out; 497 } 498 return glfs; 499 500 out: 501 if (glfs) { 502 old_errno = errno; 503 glfs_clear_preopened(glfs); 504 errno = old_errno; 505 } 506 return NULL; 507 } 508 509 /* 510 * Convert the json formatted command line into qapi. 511 */ 512 static int qemu_gluster_parse_json(BlockdevOptionsGluster *gconf, 513 QDict *options, Error **errp) 514 { 515 QemuOpts *opts; 516 SocketAddress *gsconf = NULL; 517 SocketAddressList **tail; 518 QDict *backing_options = NULL; 519 Error *local_err = NULL; 520 char *str = NULL; 521 const char *ptr; 522 int i, type, num_servers; 523 524 /* create opts info from runtime_json_opts list */ 525 opts = qemu_opts_create(&runtime_json_opts, NULL, 0, &error_abort); 526 if (!qemu_opts_absorb_qdict(opts, options, errp)) { 527 goto out; 528 } 529 530 num_servers = qdict_array_entries(options, GLUSTER_OPT_SERVER_PATTERN); 531 if (num_servers < 1) { 532 error_setg(&local_err, QERR_MISSING_PARAMETER, "server"); 533 goto out; 534 } 535 536 ptr = qemu_opt_get(opts, GLUSTER_OPT_VOLUME); 537 if (!ptr) { 538 error_setg(&local_err, QERR_MISSING_PARAMETER, GLUSTER_OPT_VOLUME); 539 goto out; 540 } 541 gconf->volume = g_strdup(ptr); 542 543 ptr = qemu_opt_get(opts, GLUSTER_OPT_PATH); 544 if (!ptr) { 545 error_setg(&local_err, QERR_MISSING_PARAMETER, GLUSTER_OPT_PATH); 546 goto out; 547 } 548 gconf->path = g_strdup(ptr); 549 qemu_opts_del(opts); 550 tail = &gconf->server; 551 552 for (i = 0; i < num_servers; i++) { 553 str = g_strdup_printf(GLUSTER_OPT_SERVER_PATTERN"%d.", i); 554 qdict_extract_subqdict(options, &backing_options, str); 555 556 /* create opts info from runtime_type_opts list */ 557 opts = qemu_opts_create(&runtime_type_opts, NULL, 0, &error_abort); 558 if (!qemu_opts_absorb_qdict(opts, backing_options, errp)) { 559 goto out; 560 } 561 562 ptr = qemu_opt_get(opts, GLUSTER_OPT_TYPE); 563 if (!ptr) { 564 error_setg(&local_err, QERR_MISSING_PARAMETER, GLUSTER_OPT_TYPE); 565 error_append_hint(&local_err, GERR_INDEX_HINT, i); 566 goto out; 567 568 } 569 gsconf = g_new0(SocketAddress, 1); 570 if (!strcmp(ptr, "tcp")) { 571 ptr = "inet"; /* accept legacy "tcp" */ 572 } 573 type = qapi_enum_parse(&SocketAddressType_lookup, ptr, -1, NULL); 574 if (type != SOCKET_ADDRESS_TYPE_INET 575 && type != SOCKET_ADDRESS_TYPE_UNIX) { 576 error_setg(&local_err, 577 "Parameter '%s' may be 'inet' or 'unix'", 578 GLUSTER_OPT_TYPE); 579 error_append_hint(&local_err, GERR_INDEX_HINT, i); 580 goto out; 581 } 582 gsconf->type = type; 583 qemu_opts_del(opts); 584 585 if (gsconf->type == SOCKET_ADDRESS_TYPE_INET) { 586 /* create opts info from runtime_inet_opts list */ 587 opts = qemu_opts_create(&runtime_inet_opts, NULL, 0, &error_abort); 588 if (!qemu_opts_absorb_qdict(opts, backing_options, errp)) { 589 goto out; 590 } 591 592 ptr = qemu_opt_get(opts, GLUSTER_OPT_HOST); 593 if (!ptr) { 594 error_setg(&local_err, QERR_MISSING_PARAMETER, 595 GLUSTER_OPT_HOST); 596 error_append_hint(&local_err, GERR_INDEX_HINT, i); 597 goto out; 598 } 599 gsconf->u.inet.host = g_strdup(ptr); 600 ptr = qemu_opt_get(opts, GLUSTER_OPT_PORT); 601 if (!ptr) { 602 error_setg(&local_err, QERR_MISSING_PARAMETER, 603 GLUSTER_OPT_PORT); 604 error_append_hint(&local_err, GERR_INDEX_HINT, i); 605 goto out; 606 } 607 gsconf->u.inet.port = g_strdup(ptr); 608 609 /* defend for unsupported fields in InetSocketAddress, 610 * i.e. @ipv4, @ipv6 and @to 611 */ 612 ptr = qemu_opt_get(opts, GLUSTER_OPT_TO); 613 if (ptr) { 614 gsconf->u.inet.has_to = true; 615 } 616 ptr = qemu_opt_get(opts, GLUSTER_OPT_IPV4); 617 if (ptr) { 618 gsconf->u.inet.has_ipv4 = true; 619 } 620 ptr = qemu_opt_get(opts, GLUSTER_OPT_IPV6); 621 if (ptr) { 622 gsconf->u.inet.has_ipv6 = true; 623 } 624 if (gsconf->u.inet.has_to) { 625 error_setg(&local_err, "Parameter 'to' not supported"); 626 goto out; 627 } 628 if (gsconf->u.inet.has_ipv4 || gsconf->u.inet.has_ipv6) { 629 error_setg(&local_err, "Parameters 'ipv4/ipv6' not supported"); 630 goto out; 631 } 632 qemu_opts_del(opts); 633 } else { 634 /* create opts info from runtime_unix_opts list */ 635 opts = qemu_opts_create(&runtime_unix_opts, NULL, 0, &error_abort); 636 if (!qemu_opts_absorb_qdict(opts, backing_options, errp)) { 637 goto out; 638 } 639 640 ptr = qemu_opt_get(opts, GLUSTER_OPT_PATH); 641 if (!ptr) { 642 ptr = qemu_opt_get(opts, GLUSTER_OPT_SOCKET); 643 } else if (qemu_opt_get(opts, GLUSTER_OPT_SOCKET)) { 644 error_setg(&local_err, 645 "Conflicting parameters 'path' and 'socket'"); 646 error_append_hint(&local_err, GERR_INDEX_HINT, i); 647 goto out; 648 } 649 if (!ptr) { 650 error_setg(&local_err, QERR_MISSING_PARAMETER, 651 GLUSTER_OPT_PATH); 652 error_append_hint(&local_err, GERR_INDEX_HINT, i); 653 goto out; 654 } 655 gsconf->u.q_unix.path = g_strdup(ptr); 656 qemu_opts_del(opts); 657 } 658 659 QAPI_LIST_APPEND(tail, gsconf); 660 gsconf = NULL; 661 662 qobject_unref(backing_options); 663 backing_options = NULL; 664 g_free(str); 665 str = NULL; 666 } 667 668 return 0; 669 670 out: 671 error_propagate(errp, local_err); 672 qapi_free_SocketAddress(gsconf); 673 qemu_opts_del(opts); 674 g_free(str); 675 qobject_unref(backing_options); 676 errno = EINVAL; 677 return -errno; 678 } 679 680 /* Converts options given in @filename and the @options QDict into the QAPI 681 * object @gconf. */ 682 static int qemu_gluster_parse(BlockdevOptionsGluster *gconf, 683 const char *filename, 684 QDict *options, Error **errp) 685 { 686 int ret; 687 if (filename) { 688 ret = qemu_gluster_parse_uri(gconf, filename); 689 if (ret < 0) { 690 error_setg(errp, "invalid URI %s", filename); 691 error_append_hint(errp, "Usage: file=gluster[+transport]://" 692 "[host[:port]]volume/path[?socket=...]" 693 "[,file.debug=N]" 694 "[,file.logfile=/path/filename.log]\n"); 695 return ret; 696 } 697 } else { 698 ret = qemu_gluster_parse_json(gconf, options, errp); 699 if (ret < 0) { 700 error_append_hint(errp, "Usage: " 701 "-drive driver=qcow2,file.driver=gluster," 702 "file.volume=testvol,file.path=/path/a.qcow2" 703 "[,file.debug=9]" 704 "[,file.logfile=/path/filename.log]," 705 "file.server.0.type=inet," 706 "file.server.0.host=1.2.3.4," 707 "file.server.0.port=24007," 708 "file.server.1.transport=unix," 709 "file.server.1.path=/var/run/glusterd.socket ..." 710 "\n"); 711 return ret; 712 } 713 } 714 715 return 0; 716 } 717 718 static struct glfs *qemu_gluster_init(BlockdevOptionsGluster *gconf, 719 const char *filename, 720 QDict *options, Error **errp) 721 { 722 int ret; 723 724 ret = qemu_gluster_parse(gconf, filename, options, errp); 725 if (ret < 0) { 726 errno = -ret; 727 return NULL; 728 } 729 730 return qemu_gluster_glfs_init(gconf, errp); 731 } 732 733 /* 734 * AIO callback routine called from GlusterFS thread. 735 */ 736 static void gluster_finish_aiocb(struct glfs_fd *fd, ssize_t ret, 737 #ifdef CONFIG_GLUSTERFS_IOCB_HAS_STAT 738 struct glfs_stat *pre, struct glfs_stat *post, 739 #endif 740 void *arg) 741 { 742 GlusterAIOCB *acb = (GlusterAIOCB *)arg; 743 744 if (!ret || ret == acb->size) { 745 acb->ret = 0; /* Success */ 746 } else if (ret < 0) { 747 acb->ret = -errno; /* Read/Write failed */ 748 } else { 749 acb->ret = -EIO; /* Partial read/write - fail it */ 750 } 751 752 aio_co_schedule(acb->aio_context, acb->coroutine); 753 } 754 755 static void qemu_gluster_parse_flags(int bdrv_flags, int *open_flags) 756 { 757 assert(open_flags != NULL); 758 759 *open_flags |= O_BINARY; 760 761 if (bdrv_flags & BDRV_O_RDWR) { 762 *open_flags |= O_RDWR; 763 } else { 764 *open_flags |= O_RDONLY; 765 } 766 767 if ((bdrv_flags & BDRV_O_NOCACHE)) { 768 *open_flags |= O_DIRECT; 769 } 770 } 771 772 /* 773 * Do SEEK_DATA/HOLE to detect if it is functional. Older broken versions of 774 * gfapi incorrectly return the current offset when SEEK_DATA/HOLE is used. 775 * - Corrected versions return -1 and set errno to EINVAL. 776 * - Versions that support SEEK_DATA/HOLE correctly, will return -1 and set 777 * errno to ENXIO when SEEK_DATA is called with a position of EOF. 778 */ 779 static bool qemu_gluster_test_seek(struct glfs_fd *fd) 780 { 781 off_t ret = 0; 782 783 #if defined SEEK_HOLE && defined SEEK_DATA 784 off_t eof; 785 786 eof = glfs_lseek(fd, 0, SEEK_END); 787 if (eof < 0) { 788 /* this should never occur */ 789 return false; 790 } 791 792 /* this should always fail with ENXIO if SEEK_DATA is supported */ 793 ret = glfs_lseek(fd, eof, SEEK_DATA); 794 #endif 795 796 return (ret < 0) && (errno == ENXIO); 797 } 798 799 static int qemu_gluster_open(BlockDriverState *bs, QDict *options, 800 int bdrv_flags, Error **errp) 801 { 802 BDRVGlusterState *s = bs->opaque; 803 int open_flags = 0; 804 int ret = 0; 805 BlockdevOptionsGluster *gconf = NULL; 806 QemuOpts *opts; 807 const char *filename, *logfile; 808 809 opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort); 810 if (!qemu_opts_absorb_qdict(opts, options, errp)) { 811 ret = -EINVAL; 812 goto out; 813 } 814 815 filename = qemu_opt_get(opts, GLUSTER_OPT_FILENAME); 816 817 s->debug = qemu_opt_get_number(opts, GLUSTER_OPT_DEBUG, 818 GLUSTER_DEBUG_DEFAULT); 819 if (s->debug < 0) { 820 s->debug = 0; 821 } else if (s->debug > GLUSTER_DEBUG_MAX) { 822 s->debug = GLUSTER_DEBUG_MAX; 823 } 824 825 gconf = g_new0(BlockdevOptionsGluster, 1); 826 gconf->debug = s->debug; 827 gconf->has_debug = true; 828 829 logfile = qemu_opt_get(opts, GLUSTER_OPT_LOGFILE); 830 s->logfile = g_strdup(logfile ? logfile : GLUSTER_LOGFILE_DEFAULT); 831 832 gconf->logfile = g_strdup(s->logfile); 833 834 s->glfs = qemu_gluster_init(gconf, filename, options, errp); 835 if (!s->glfs) { 836 ret = -errno; 837 goto out; 838 } 839 840 #ifdef CONFIG_GLUSTERFS_XLATOR_OPT 841 /* Without this, if fsync fails for a recoverable reason (for instance, 842 * ENOSPC), gluster will dump its cache, preventing retries. This means 843 * almost certain data loss. Not all gluster versions support the 844 * 'resync-failed-syncs-after-fsync' key value, but there is no way to 845 * discover during runtime if it is supported (this api returns success for 846 * unknown key/value pairs) */ 847 ret = glfs_set_xlator_option(s->glfs, "*-write-behind", 848 "resync-failed-syncs-after-fsync", 849 "on"); 850 if (ret < 0) { 851 error_setg_errno(errp, errno, "Unable to set xlator key/value pair"); 852 ret = -errno; 853 goto out; 854 } 855 #endif 856 857 qemu_gluster_parse_flags(bdrv_flags, &open_flags); 858 859 s->fd = glfs_open(s->glfs, gconf->path, open_flags); 860 ret = s->fd ? 0 : -errno; 861 862 if (ret == -EACCES || ret == -EROFS) { 863 /* Try to degrade to read-only, but if it doesn't work, still use the 864 * normal error message. */ 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 } 871 872 s->supports_seek_data = qemu_gluster_test_seek(s->fd); 873 874 out: 875 qemu_opts_del(opts); 876 qapi_free_BlockdevOptionsGluster(gconf); 877 if (!ret) { 878 return ret; 879 } 880 g_free(s->logfile); 881 if (s->fd) { 882 glfs_close(s->fd); 883 } 884 885 glfs_clear_preopened(s->glfs); 886 887 return ret; 888 } 889 890 static void qemu_gluster_refresh_limits(BlockDriverState *bs, Error **errp) 891 { 892 bs->bl.max_transfer = GLUSTER_MAX_TRANSFER; 893 bs->bl.max_pdiscard = MIN(SIZE_MAX, INT64_MAX); 894 } 895 896 static int qemu_gluster_reopen_prepare(BDRVReopenState *state, 897 BlockReopenQueue *queue, Error **errp) 898 { 899 int ret = 0; 900 BDRVGlusterState *s; 901 BDRVGlusterReopenState *reop_s; 902 BlockdevOptionsGluster *gconf; 903 int open_flags = 0; 904 905 assert(state != NULL); 906 assert(state->bs != NULL); 907 908 s = state->bs->opaque; 909 910 state->opaque = g_new0(BDRVGlusterReopenState, 1); 911 reop_s = state->opaque; 912 913 qemu_gluster_parse_flags(state->flags, &open_flags); 914 915 gconf = g_new0(BlockdevOptionsGluster, 1); 916 gconf->debug = s->debug; 917 gconf->has_debug = true; 918 gconf->logfile = g_strdup(s->logfile); 919 920 /* 921 * If 'state->bs->exact_filename' is empty, 'state->options' should contain 922 * the JSON parameters already parsed. 923 */ 924 if (state->bs->exact_filename[0] != '\0') { 925 reop_s->glfs = qemu_gluster_init(gconf, state->bs->exact_filename, NULL, 926 errp); 927 } else { 928 reop_s->glfs = qemu_gluster_init(gconf, NULL, state->options, errp); 929 } 930 if (reop_s->glfs == NULL) { 931 ret = -errno; 932 goto exit; 933 } 934 935 #ifdef CONFIG_GLUSTERFS_XLATOR_OPT 936 ret = glfs_set_xlator_option(reop_s->glfs, "*-write-behind", 937 "resync-failed-syncs-after-fsync", "on"); 938 if (ret < 0) { 939 error_setg_errno(errp, errno, "Unable to set xlator key/value pair"); 940 ret = -errno; 941 goto exit; 942 } 943 #endif 944 945 reop_s->fd = glfs_open(reop_s->glfs, gconf->path, open_flags); 946 if (reop_s->fd == NULL) { 947 /* reops->glfs will be cleaned up in _abort */ 948 ret = -errno; 949 goto exit; 950 } 951 952 exit: 953 /* state->opaque will be freed in either the _abort or _commit */ 954 qapi_free_BlockdevOptionsGluster(gconf); 955 return ret; 956 } 957 958 static void qemu_gluster_reopen_commit(BDRVReopenState *state) 959 { 960 BDRVGlusterReopenState *reop_s = state->opaque; 961 BDRVGlusterState *s = state->bs->opaque; 962 963 964 /* close the old */ 965 if (s->fd) { 966 glfs_close(s->fd); 967 } 968 969 glfs_clear_preopened(s->glfs); 970 971 /* use the newly opened image / connection */ 972 s->fd = reop_s->fd; 973 s->glfs = reop_s->glfs; 974 975 g_free(state->opaque); 976 state->opaque = NULL; 977 978 return; 979 } 980 981 982 static void qemu_gluster_reopen_abort(BDRVReopenState *state) 983 { 984 BDRVGlusterReopenState *reop_s = state->opaque; 985 986 if (reop_s == NULL) { 987 return; 988 } 989 990 if (reop_s->fd) { 991 glfs_close(reop_s->fd); 992 } 993 994 glfs_clear_preopened(reop_s->glfs); 995 996 g_free(state->opaque); 997 state->opaque = NULL; 998 999 return; 1000 } 1001 1002 #ifdef CONFIG_GLUSTERFS_ZEROFILL 1003 static coroutine_fn int qemu_gluster_co_pwrite_zeroes(BlockDriverState *bs, 1004 int64_t offset, 1005 int64_t bytes, 1006 BdrvRequestFlags flags) 1007 { 1008 int ret; 1009 GlusterAIOCB acb; 1010 BDRVGlusterState *s = bs->opaque; 1011 1012 acb.size = bytes; 1013 acb.ret = 0; 1014 acb.coroutine = qemu_coroutine_self(); 1015 acb.aio_context = bdrv_get_aio_context(bs); 1016 1017 ret = glfs_zerofill_async(s->fd, offset, bytes, gluster_finish_aiocb, &acb); 1018 if (ret < 0) { 1019 return -errno; 1020 } 1021 1022 qemu_coroutine_yield(); 1023 return acb.ret; 1024 } 1025 #endif 1026 1027 static int qemu_gluster_do_truncate(struct glfs_fd *fd, int64_t offset, 1028 PreallocMode prealloc, Error **errp) 1029 { 1030 int64_t current_length; 1031 1032 current_length = glfs_lseek(fd, 0, SEEK_END); 1033 if (current_length < 0) { 1034 error_setg_errno(errp, errno, "Failed to determine current size"); 1035 return -errno; 1036 } 1037 1038 if (current_length > offset && prealloc != PREALLOC_MODE_OFF) { 1039 error_setg(errp, "Cannot use preallocation for shrinking files"); 1040 return -ENOTSUP; 1041 } 1042 1043 if (current_length == offset) { 1044 return 0; 1045 } 1046 1047 switch (prealloc) { 1048 #ifdef CONFIG_GLUSTERFS_FALLOCATE 1049 case PREALLOC_MODE_FALLOC: 1050 if (glfs_fallocate(fd, 0, current_length, offset - current_length)) { 1051 error_setg_errno(errp, errno, "Could not preallocate data"); 1052 return -errno; 1053 } 1054 break; 1055 #endif /* CONFIG_GLUSTERFS_FALLOCATE */ 1056 #ifdef CONFIG_GLUSTERFS_ZEROFILL 1057 case PREALLOC_MODE_FULL: 1058 if (glfs_ftruncate(fd, offset)) { 1059 error_setg_errno(errp, errno, "Could not resize file"); 1060 return -errno; 1061 } 1062 if (glfs_zerofill(fd, current_length, offset - current_length)) { 1063 error_setg_errno(errp, errno, "Could not zerofill the new area"); 1064 return -errno; 1065 } 1066 break; 1067 #endif /* CONFIG_GLUSTERFS_ZEROFILL */ 1068 case PREALLOC_MODE_OFF: 1069 if (glfs_ftruncate(fd, offset)) { 1070 error_setg_errno(errp, errno, "Could not resize file"); 1071 return -errno; 1072 } 1073 break; 1074 default: 1075 error_setg(errp, "Unsupported preallocation mode: %s", 1076 PreallocMode_str(prealloc)); 1077 return -EINVAL; 1078 } 1079 1080 return 0; 1081 } 1082 1083 static int qemu_gluster_co_create(BlockdevCreateOptions *options, 1084 Error **errp) 1085 { 1086 BlockdevCreateOptionsGluster *opts = &options->u.gluster; 1087 struct glfs *glfs; 1088 struct glfs_fd *fd = NULL; 1089 int ret = 0; 1090 1091 assert(options->driver == BLOCKDEV_DRIVER_GLUSTER); 1092 1093 glfs = qemu_gluster_glfs_init(opts->location, errp); 1094 if (!glfs) { 1095 ret = -errno; 1096 goto out; 1097 } 1098 1099 fd = glfs_creat(glfs, opts->location->path, 1100 O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, S_IRUSR | S_IWUSR); 1101 if (!fd) { 1102 ret = -errno; 1103 goto out; 1104 } 1105 1106 ret = qemu_gluster_do_truncate(fd, opts->size, opts->preallocation, errp); 1107 1108 out: 1109 if (fd) { 1110 if (glfs_close(fd) != 0 && ret == 0) { 1111 ret = -errno; 1112 } 1113 } 1114 glfs_clear_preopened(glfs); 1115 return ret; 1116 } 1117 1118 static int coroutine_fn qemu_gluster_co_create_opts(BlockDriver *drv, 1119 const char *filename, 1120 QemuOpts *opts, 1121 Error **errp) 1122 { 1123 BlockdevCreateOptions *options; 1124 BlockdevCreateOptionsGluster *gopts; 1125 BlockdevOptionsGluster *gconf; 1126 char *tmp = NULL; 1127 Error *local_err = NULL; 1128 int ret; 1129 1130 options = g_new0(BlockdevCreateOptions, 1); 1131 options->driver = BLOCKDEV_DRIVER_GLUSTER; 1132 gopts = &options->u.gluster; 1133 1134 gconf = g_new0(BlockdevOptionsGluster, 1); 1135 gopts->location = gconf; 1136 1137 gopts->size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0), 1138 BDRV_SECTOR_SIZE); 1139 1140 tmp = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC); 1141 gopts->preallocation = qapi_enum_parse(&PreallocMode_lookup, tmp, 1142 PREALLOC_MODE_OFF, &local_err); 1143 g_free(tmp); 1144 if (local_err) { 1145 error_propagate(errp, local_err); 1146 ret = -EINVAL; 1147 goto fail; 1148 } 1149 1150 gconf->debug = qemu_opt_get_number_del(opts, GLUSTER_OPT_DEBUG, 1151 GLUSTER_DEBUG_DEFAULT); 1152 if (gconf->debug < 0) { 1153 gconf->debug = 0; 1154 } else if (gconf->debug > GLUSTER_DEBUG_MAX) { 1155 gconf->debug = GLUSTER_DEBUG_MAX; 1156 } 1157 gconf->has_debug = true; 1158 1159 gconf->logfile = qemu_opt_get_del(opts, GLUSTER_OPT_LOGFILE); 1160 if (!gconf->logfile) { 1161 gconf->logfile = g_strdup(GLUSTER_LOGFILE_DEFAULT); 1162 } 1163 1164 ret = qemu_gluster_parse(gconf, filename, NULL, errp); 1165 if (ret < 0) { 1166 goto fail; 1167 } 1168 1169 ret = qemu_gluster_co_create(options, errp); 1170 if (ret < 0) { 1171 goto fail; 1172 } 1173 1174 ret = 0; 1175 fail: 1176 qapi_free_BlockdevCreateOptions(options); 1177 return ret; 1178 } 1179 1180 static coroutine_fn int qemu_gluster_co_rw(BlockDriverState *bs, 1181 int64_t sector_num, int nb_sectors, 1182 QEMUIOVector *qiov, int write) 1183 { 1184 int ret; 1185 GlusterAIOCB acb; 1186 BDRVGlusterState *s = bs->opaque; 1187 size_t size = nb_sectors * BDRV_SECTOR_SIZE; 1188 off_t offset = sector_num * BDRV_SECTOR_SIZE; 1189 1190 acb.size = size; 1191 acb.ret = 0; 1192 acb.coroutine = qemu_coroutine_self(); 1193 acb.aio_context = bdrv_get_aio_context(bs); 1194 1195 if (write) { 1196 ret = glfs_pwritev_async(s->fd, qiov->iov, qiov->niov, offset, 0, 1197 gluster_finish_aiocb, &acb); 1198 } else { 1199 ret = glfs_preadv_async(s->fd, qiov->iov, qiov->niov, offset, 0, 1200 gluster_finish_aiocb, &acb); 1201 } 1202 1203 if (ret < 0) { 1204 return -errno; 1205 } 1206 1207 qemu_coroutine_yield(); 1208 return acb.ret; 1209 } 1210 1211 static coroutine_fn int qemu_gluster_co_truncate(BlockDriverState *bs, 1212 int64_t offset, 1213 bool exact, 1214 PreallocMode prealloc, 1215 BdrvRequestFlags flags, 1216 Error **errp) 1217 { 1218 BDRVGlusterState *s = bs->opaque; 1219 return qemu_gluster_do_truncate(s->fd, offset, prealloc, errp); 1220 } 1221 1222 static coroutine_fn int qemu_gluster_co_readv(BlockDriverState *bs, 1223 int64_t sector_num, 1224 int nb_sectors, 1225 QEMUIOVector *qiov) 1226 { 1227 return qemu_gluster_co_rw(bs, sector_num, nb_sectors, qiov, 0); 1228 } 1229 1230 static coroutine_fn int qemu_gluster_co_writev(BlockDriverState *bs, 1231 int64_t sector_num, 1232 int nb_sectors, 1233 QEMUIOVector *qiov, 1234 int flags) 1235 { 1236 return qemu_gluster_co_rw(bs, sector_num, nb_sectors, qiov, 1); 1237 } 1238 1239 static void qemu_gluster_close(BlockDriverState *bs) 1240 { 1241 BDRVGlusterState *s = bs->opaque; 1242 1243 g_free(s->logfile); 1244 if (s->fd) { 1245 glfs_close(s->fd); 1246 s->fd = NULL; 1247 } 1248 glfs_clear_preopened(s->glfs); 1249 } 1250 1251 static coroutine_fn int qemu_gluster_co_flush_to_disk(BlockDriverState *bs) 1252 { 1253 int ret; 1254 GlusterAIOCB acb; 1255 BDRVGlusterState *s = bs->opaque; 1256 1257 acb.size = 0; 1258 acb.ret = 0; 1259 acb.coroutine = qemu_coroutine_self(); 1260 acb.aio_context = bdrv_get_aio_context(bs); 1261 1262 ret = glfs_fsync_async(s->fd, gluster_finish_aiocb, &acb); 1263 if (ret < 0) { 1264 ret = -errno; 1265 goto error; 1266 } 1267 1268 qemu_coroutine_yield(); 1269 if (acb.ret < 0) { 1270 ret = acb.ret; 1271 goto error; 1272 } 1273 1274 return acb.ret; 1275 1276 error: 1277 /* Some versions of Gluster (3.5.6 -> 3.5.8?) will not retain its cache 1278 * after a fsync failure, so we have no way of allowing the guest to safely 1279 * continue. Gluster versions prior to 3.5.6 don't retain the cache 1280 * either, but will invalidate the fd on error, so this is again our only 1281 * option. 1282 * 1283 * The 'resync-failed-syncs-after-fsync' xlator option for the 1284 * write-behind cache will cause later gluster versions to retain its 1285 * cache after error, so long as the fd remains open. However, we 1286 * currently have no way of knowing if this option is supported. 1287 * 1288 * TODO: Once gluster provides a way for us to determine if the option 1289 * is supported, bypass the closure and setting drv to NULL. */ 1290 qemu_gluster_close(bs); 1291 bs->drv = NULL; 1292 return ret; 1293 } 1294 1295 #ifdef CONFIG_GLUSTERFS_DISCARD 1296 static coroutine_fn int qemu_gluster_co_pdiscard(BlockDriverState *bs, 1297 int64_t offset, int64_t bytes) 1298 { 1299 int ret; 1300 GlusterAIOCB acb; 1301 BDRVGlusterState *s = bs->opaque; 1302 1303 assert(bytes <= SIZE_MAX); /* rely on max_pdiscard */ 1304 1305 acb.size = 0; 1306 acb.ret = 0; 1307 acb.coroutine = qemu_coroutine_self(); 1308 acb.aio_context = bdrv_get_aio_context(bs); 1309 1310 ret = glfs_discard_async(s->fd, offset, bytes, gluster_finish_aiocb, &acb); 1311 if (ret < 0) { 1312 return -errno; 1313 } 1314 1315 qemu_coroutine_yield(); 1316 return acb.ret; 1317 } 1318 #endif 1319 1320 static int64_t qemu_gluster_getlength(BlockDriverState *bs) 1321 { 1322 BDRVGlusterState *s = bs->opaque; 1323 int64_t ret; 1324 1325 ret = glfs_lseek(s->fd, 0, SEEK_END); 1326 if (ret < 0) { 1327 return -errno; 1328 } else { 1329 return ret; 1330 } 1331 } 1332 1333 static int64_t qemu_gluster_allocated_file_size(BlockDriverState *bs) 1334 { 1335 BDRVGlusterState *s = bs->opaque; 1336 struct stat st; 1337 int ret; 1338 1339 ret = glfs_fstat(s->fd, &st); 1340 if (ret < 0) { 1341 return -errno; 1342 } else { 1343 return st.st_blocks * 512; 1344 } 1345 } 1346 1347 /* 1348 * Find allocation range in @bs around offset @start. 1349 * May change underlying file descriptor's file offset. 1350 * If @start is not in a hole, store @start in @data, and the 1351 * beginning of the next hole in @hole, and return 0. 1352 * If @start is in a non-trailing hole, store @start in @hole and the 1353 * beginning of the next non-hole in @data, and return 0. 1354 * If @start is in a trailing hole or beyond EOF, return -ENXIO. 1355 * If we can't find out, return a negative errno other than -ENXIO. 1356 * 1357 * (Shamefully copied from file-posix.c, only minuscule adaptions.) 1358 */ 1359 static int find_allocation(BlockDriverState *bs, off_t start, 1360 off_t *data, off_t *hole) 1361 { 1362 BDRVGlusterState *s = bs->opaque; 1363 1364 if (!s->supports_seek_data) { 1365 goto exit; 1366 } 1367 1368 #if defined SEEK_HOLE && defined SEEK_DATA 1369 off_t offs; 1370 1371 /* 1372 * SEEK_DATA cases: 1373 * D1. offs == start: start is in data 1374 * D2. offs > start: start is in a hole, next data at offs 1375 * D3. offs < 0, errno = ENXIO: either start is in a trailing hole 1376 * or start is beyond EOF 1377 * If the latter happens, the file has been truncated behind 1378 * our back since we opened it. All bets are off then. 1379 * Treating like a trailing hole is simplest. 1380 * D4. offs < 0, errno != ENXIO: we learned nothing 1381 */ 1382 offs = glfs_lseek(s->fd, start, SEEK_DATA); 1383 if (offs < 0) { 1384 return -errno; /* D3 or D4 */ 1385 } 1386 1387 if (offs < start) { 1388 /* This is not a valid return by lseek(). We are safe to just return 1389 * -EIO in this case, and we'll treat it like D4. Unfortunately some 1390 * versions of gluster server will return offs < start, so an assert 1391 * here will unnecessarily abort QEMU. */ 1392 return -EIO; 1393 } 1394 1395 if (offs > start) { 1396 /* D2: in hole, next data at offs */ 1397 *hole = start; 1398 *data = offs; 1399 return 0; 1400 } 1401 1402 /* D1: in data, end not yet known */ 1403 1404 /* 1405 * SEEK_HOLE cases: 1406 * H1. offs == start: start is in a hole 1407 * If this happens here, a hole has been dug behind our back 1408 * since the previous lseek(). 1409 * H2. offs > start: either start is in data, next hole at offs, 1410 * or start is in trailing hole, EOF at offs 1411 * Linux treats trailing holes like any other hole: offs == 1412 * start. Solaris seeks to EOF instead: offs > start (blech). 1413 * If that happens here, a hole has been dug behind our back 1414 * since the previous lseek(). 1415 * H3. offs < 0, errno = ENXIO: start is beyond EOF 1416 * If this happens, the file has been truncated behind our 1417 * back since we opened it. Treat it like a trailing hole. 1418 * H4. offs < 0, errno != ENXIO: we learned nothing 1419 * Pretend we know nothing at all, i.e. "forget" about D1. 1420 */ 1421 offs = glfs_lseek(s->fd, start, SEEK_HOLE); 1422 if (offs < 0) { 1423 return -errno; /* D1 and (H3 or H4) */ 1424 } 1425 1426 if (offs < start) { 1427 /* This is not a valid return by lseek(). We are safe to just return 1428 * -EIO in this case, and we'll treat it like H4. Unfortunately some 1429 * versions of gluster server will return offs < start, so an assert 1430 * here will unnecessarily abort QEMU. */ 1431 return -EIO; 1432 } 1433 1434 if (offs > start) { 1435 /* 1436 * D1 and H2: either in data, next hole at offs, or it was in 1437 * data but is now in a trailing hole. In the latter case, 1438 * all bets are off. Treating it as if it there was data all 1439 * the way to EOF is safe, so simply do that. 1440 */ 1441 *data = start; 1442 *hole = offs; 1443 return 0; 1444 } 1445 1446 /* D1 and H1 */ 1447 return -EBUSY; 1448 #endif 1449 1450 exit: 1451 return -ENOTSUP; 1452 } 1453 1454 /* 1455 * Returns the allocation status of the specified offset. 1456 * 1457 * The block layer guarantees 'offset' and 'bytes' are within bounds. 1458 * 1459 * 'pnum' is set to the number of bytes (including and immediately following 1460 * the specified offset) that are known to be in the same 1461 * allocated/unallocated state. 1462 * 1463 * 'bytes' is a soft cap for 'pnum'. If the information is free, 'pnum' may 1464 * well exceed it. 1465 * 1466 * (Based on raw_co_block_status() from file-posix.c.) 1467 */ 1468 static int coroutine_fn qemu_gluster_co_block_status(BlockDriverState *bs, 1469 bool want_zero, 1470 int64_t offset, 1471 int64_t bytes, 1472 int64_t *pnum, 1473 int64_t *map, 1474 BlockDriverState **file) 1475 { 1476 BDRVGlusterState *s = bs->opaque; 1477 off_t data = 0, hole = 0; 1478 int ret = -EINVAL; 1479 1480 assert(QEMU_IS_ALIGNED(offset | bytes, bs->bl.request_alignment)); 1481 1482 if (!s->fd) { 1483 return ret; 1484 } 1485 1486 if (!want_zero) { 1487 *pnum = bytes; 1488 *map = offset; 1489 *file = bs; 1490 return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID; 1491 } 1492 1493 ret = find_allocation(bs, offset, &data, &hole); 1494 if (ret == -ENXIO) { 1495 /* Trailing hole */ 1496 *pnum = bytes; 1497 ret = BDRV_BLOCK_ZERO; 1498 } else if (ret < 0) { 1499 /* No info available, so pretend there are no holes */ 1500 *pnum = bytes; 1501 ret = BDRV_BLOCK_DATA; 1502 } else if (data == offset) { 1503 /* On a data extent, compute bytes to the end of the extent, 1504 * possibly including a partial sector at EOF. */ 1505 *pnum = hole - offset; 1506 1507 /* 1508 * We are not allowed to return partial sectors, though, so 1509 * round up if necessary. 1510 */ 1511 if (!QEMU_IS_ALIGNED(*pnum, bs->bl.request_alignment)) { 1512 int64_t file_length = qemu_gluster_getlength(bs); 1513 if (file_length > 0) { 1514 /* Ignore errors, this is just a safeguard */ 1515 assert(hole == file_length); 1516 } 1517 *pnum = ROUND_UP(*pnum, bs->bl.request_alignment); 1518 } 1519 1520 ret = BDRV_BLOCK_DATA; 1521 } else { 1522 /* On a hole, compute bytes to the beginning of the next extent. */ 1523 assert(hole == offset); 1524 *pnum = data - offset; 1525 ret = BDRV_BLOCK_ZERO; 1526 } 1527 1528 *map = offset; 1529 *file = bs; 1530 1531 return ret | BDRV_BLOCK_OFFSET_VALID; 1532 } 1533 1534 1535 static const char *const gluster_strong_open_opts[] = { 1536 GLUSTER_OPT_VOLUME, 1537 GLUSTER_OPT_PATH, 1538 GLUSTER_OPT_TYPE, 1539 GLUSTER_OPT_SERVER_PATTERN, 1540 GLUSTER_OPT_HOST, 1541 GLUSTER_OPT_PORT, 1542 GLUSTER_OPT_TO, 1543 GLUSTER_OPT_IPV4, 1544 GLUSTER_OPT_IPV6, 1545 GLUSTER_OPT_SOCKET, 1546 1547 NULL 1548 }; 1549 1550 static BlockDriver bdrv_gluster = { 1551 .format_name = "gluster", 1552 .protocol_name = "gluster", 1553 .instance_size = sizeof(BDRVGlusterState), 1554 .bdrv_file_open = qemu_gluster_open, 1555 .bdrv_reopen_prepare = qemu_gluster_reopen_prepare, 1556 .bdrv_reopen_commit = qemu_gluster_reopen_commit, 1557 .bdrv_reopen_abort = qemu_gluster_reopen_abort, 1558 .bdrv_close = qemu_gluster_close, 1559 .bdrv_co_create = qemu_gluster_co_create, 1560 .bdrv_co_create_opts = qemu_gluster_co_create_opts, 1561 .bdrv_getlength = qemu_gluster_getlength, 1562 .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size, 1563 .bdrv_co_truncate = qemu_gluster_co_truncate, 1564 .bdrv_co_readv = qemu_gluster_co_readv, 1565 .bdrv_co_writev = qemu_gluster_co_writev, 1566 .bdrv_co_flush_to_disk = qemu_gluster_co_flush_to_disk, 1567 #ifdef CONFIG_GLUSTERFS_DISCARD 1568 .bdrv_co_pdiscard = qemu_gluster_co_pdiscard, 1569 #endif 1570 #ifdef CONFIG_GLUSTERFS_ZEROFILL 1571 .bdrv_co_pwrite_zeroes = qemu_gluster_co_pwrite_zeroes, 1572 #endif 1573 .bdrv_co_block_status = qemu_gluster_co_block_status, 1574 .bdrv_refresh_limits = qemu_gluster_refresh_limits, 1575 .create_opts = &qemu_gluster_create_opts, 1576 .strong_runtime_opts = gluster_strong_open_opts, 1577 }; 1578 1579 static BlockDriver bdrv_gluster_tcp = { 1580 .format_name = "gluster", 1581 .protocol_name = "gluster+tcp", 1582 .instance_size = sizeof(BDRVGlusterState), 1583 .bdrv_file_open = qemu_gluster_open, 1584 .bdrv_reopen_prepare = qemu_gluster_reopen_prepare, 1585 .bdrv_reopen_commit = qemu_gluster_reopen_commit, 1586 .bdrv_reopen_abort = qemu_gluster_reopen_abort, 1587 .bdrv_close = qemu_gluster_close, 1588 .bdrv_co_create = qemu_gluster_co_create, 1589 .bdrv_co_create_opts = qemu_gluster_co_create_opts, 1590 .bdrv_getlength = qemu_gluster_getlength, 1591 .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size, 1592 .bdrv_co_truncate = qemu_gluster_co_truncate, 1593 .bdrv_co_readv = qemu_gluster_co_readv, 1594 .bdrv_co_writev = qemu_gluster_co_writev, 1595 .bdrv_co_flush_to_disk = qemu_gluster_co_flush_to_disk, 1596 #ifdef CONFIG_GLUSTERFS_DISCARD 1597 .bdrv_co_pdiscard = qemu_gluster_co_pdiscard, 1598 #endif 1599 #ifdef CONFIG_GLUSTERFS_ZEROFILL 1600 .bdrv_co_pwrite_zeroes = qemu_gluster_co_pwrite_zeroes, 1601 #endif 1602 .bdrv_co_block_status = qemu_gluster_co_block_status, 1603 .bdrv_refresh_limits = qemu_gluster_refresh_limits, 1604 .create_opts = &qemu_gluster_create_opts, 1605 .strong_runtime_opts = gluster_strong_open_opts, 1606 }; 1607 1608 static BlockDriver bdrv_gluster_unix = { 1609 .format_name = "gluster", 1610 .protocol_name = "gluster+unix", 1611 .instance_size = sizeof(BDRVGlusterState), 1612 .bdrv_file_open = qemu_gluster_open, 1613 .bdrv_reopen_prepare = qemu_gluster_reopen_prepare, 1614 .bdrv_reopen_commit = qemu_gluster_reopen_commit, 1615 .bdrv_reopen_abort = qemu_gluster_reopen_abort, 1616 .bdrv_close = qemu_gluster_close, 1617 .bdrv_co_create = qemu_gluster_co_create, 1618 .bdrv_co_create_opts = qemu_gluster_co_create_opts, 1619 .bdrv_getlength = qemu_gluster_getlength, 1620 .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size, 1621 .bdrv_co_truncate = qemu_gluster_co_truncate, 1622 .bdrv_co_readv = qemu_gluster_co_readv, 1623 .bdrv_co_writev = qemu_gluster_co_writev, 1624 .bdrv_co_flush_to_disk = qemu_gluster_co_flush_to_disk, 1625 #ifdef CONFIG_GLUSTERFS_DISCARD 1626 .bdrv_co_pdiscard = qemu_gluster_co_pdiscard, 1627 #endif 1628 #ifdef CONFIG_GLUSTERFS_ZEROFILL 1629 .bdrv_co_pwrite_zeroes = qemu_gluster_co_pwrite_zeroes, 1630 #endif 1631 .bdrv_co_block_status = qemu_gluster_co_block_status, 1632 .bdrv_refresh_limits = qemu_gluster_refresh_limits, 1633 .create_opts = &qemu_gluster_create_opts, 1634 .strong_runtime_opts = gluster_strong_open_opts, 1635 }; 1636 1637 /* rdma is deprecated (actually never supported for volfile fetch). 1638 * Let's maintain it for the protocol compatibility, to make sure things 1639 * won't break immediately. For now, gluster+rdma will fall back to gluster+tcp 1640 * protocol with a warning. 1641 * TODO: remove gluster+rdma interface support 1642 */ 1643 static BlockDriver bdrv_gluster_rdma = { 1644 .format_name = "gluster", 1645 .protocol_name = "gluster+rdma", 1646 .instance_size = sizeof(BDRVGlusterState), 1647 .bdrv_file_open = qemu_gluster_open, 1648 .bdrv_reopen_prepare = qemu_gluster_reopen_prepare, 1649 .bdrv_reopen_commit = qemu_gluster_reopen_commit, 1650 .bdrv_reopen_abort = qemu_gluster_reopen_abort, 1651 .bdrv_close = qemu_gluster_close, 1652 .bdrv_co_create = qemu_gluster_co_create, 1653 .bdrv_co_create_opts = qemu_gluster_co_create_opts, 1654 .bdrv_getlength = qemu_gluster_getlength, 1655 .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size, 1656 .bdrv_co_truncate = qemu_gluster_co_truncate, 1657 .bdrv_co_readv = qemu_gluster_co_readv, 1658 .bdrv_co_writev = qemu_gluster_co_writev, 1659 .bdrv_co_flush_to_disk = qemu_gluster_co_flush_to_disk, 1660 #ifdef CONFIG_GLUSTERFS_DISCARD 1661 .bdrv_co_pdiscard = qemu_gluster_co_pdiscard, 1662 #endif 1663 #ifdef CONFIG_GLUSTERFS_ZEROFILL 1664 .bdrv_co_pwrite_zeroes = qemu_gluster_co_pwrite_zeroes, 1665 #endif 1666 .bdrv_co_block_status = qemu_gluster_co_block_status, 1667 .bdrv_refresh_limits = qemu_gluster_refresh_limits, 1668 .create_opts = &qemu_gluster_create_opts, 1669 .strong_runtime_opts = gluster_strong_open_opts, 1670 }; 1671 1672 static void bdrv_gluster_init(void) 1673 { 1674 bdrv_register(&bdrv_gluster_rdma); 1675 bdrv_register(&bdrv_gluster_unix); 1676 bdrv_register(&bdrv_gluster_tcp); 1677 bdrv_register(&bdrv_gluster); 1678 } 1679 1680 block_init(bdrv_gluster_init); 1681