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/uri.h" 21 #include "qemu/error-report.h" 22 #include "qemu/module.h" 23 #include "qemu/option.h" 24 #include "qemu/cutils.h" 25 26 #ifdef CONFIG_GLUSTERFS_FTRUNCATE_HAS_STAT 27 # define glfs_ftruncate(fd, offset) glfs_ftruncate(fd, offset, NULL, NULL) 28 #endif 29 30 #define GLUSTER_OPT_FILENAME "filename" 31 #define GLUSTER_OPT_VOLUME "volume" 32 #define GLUSTER_OPT_PATH "path" 33 #define GLUSTER_OPT_TYPE "type" 34 #define GLUSTER_OPT_SERVER_PATTERN "server." 35 #define GLUSTER_OPT_HOST "host" 36 #define GLUSTER_OPT_PORT "port" 37 #define GLUSTER_OPT_TO "to" 38 #define GLUSTER_OPT_IPV4 "ipv4" 39 #define GLUSTER_OPT_IPV6 "ipv6" 40 #define GLUSTER_OPT_SOCKET "socket" 41 #define GLUSTER_OPT_DEBUG "debug" 42 #define GLUSTER_DEFAULT_PORT 24007 43 #define GLUSTER_DEBUG_DEFAULT 4 44 #define GLUSTER_DEBUG_MAX 9 45 #define GLUSTER_OPT_LOGFILE "logfile" 46 #define GLUSTER_LOGFILE_DEFAULT "-" /* handled in libgfapi as /dev/stderr */ 47 /* 48 * Several versions of GlusterFS (3.12? -> 6.0.1) fail when the transfer size 49 * is greater or equal to 1024 MiB, so we are limiting the transfer size to 512 50 * MiB to avoid this rare issue. 51 */ 52 #define GLUSTER_MAX_TRANSFER (512 * MiB) 53 54 #define GERR_INDEX_HINT "hint: check in 'server' array index '%d'\n" 55 56 typedef struct GlusterAIOCB { 57 int64_t size; 58 int ret; 59 Coroutine *coroutine; 60 AioContext *aio_context; 61 } GlusterAIOCB; 62 63 typedef struct BDRVGlusterState { 64 struct glfs *glfs; 65 struct glfs_fd *fd; 66 char *logfile; 67 bool supports_seek_data; 68 int debug; 69 } BDRVGlusterState; 70 71 typedef struct BDRVGlusterReopenState { 72 struct glfs *glfs; 73 struct glfs_fd *fd; 74 } BDRVGlusterReopenState; 75 76 77 typedef struct GlfsPreopened { 78 char *volume; 79 glfs_t *fs; 80 int ref; 81 } GlfsPreopened; 82 83 typedef struct ListElement { 84 QLIST_ENTRY(ListElement) list; 85 GlfsPreopened saved; 86 } ListElement; 87 88 static QLIST_HEAD(, ListElement) glfs_list; 89 90 static QemuOptsList qemu_gluster_create_opts = { 91 .name = "qemu-gluster-create-opts", 92 .head = QTAILQ_HEAD_INITIALIZER(qemu_gluster_create_opts.head), 93 .desc = { 94 { 95 .name = BLOCK_OPT_SIZE, 96 .type = QEMU_OPT_SIZE, 97 .help = "Virtual disk size" 98 }, 99 { 100 .name = BLOCK_OPT_PREALLOC, 101 .type = QEMU_OPT_STRING, 102 .help = "Preallocation mode (allowed values: off" 103 #ifdef CONFIG_GLUSTERFS_FALLOCATE 104 ", falloc" 105 #endif 106 #ifdef CONFIG_GLUSTERFS_ZEROFILL 107 ", full" 108 #endif 109 ")" 110 }, 111 { 112 .name = GLUSTER_OPT_DEBUG, 113 .type = QEMU_OPT_NUMBER, 114 .help = "Gluster log level, valid range is 0-9", 115 }, 116 { 117 .name = GLUSTER_OPT_LOGFILE, 118 .type = QEMU_OPT_STRING, 119 .help = "Logfile path of libgfapi", 120 }, 121 { /* end of list */ } 122 } 123 }; 124 125 static QemuOptsList runtime_opts = { 126 .name = "gluster", 127 .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head), 128 .desc = { 129 { 130 .name = GLUSTER_OPT_FILENAME, 131 .type = QEMU_OPT_STRING, 132 .help = "URL to the gluster image", 133 }, 134 { 135 .name = GLUSTER_OPT_DEBUG, 136 .type = QEMU_OPT_NUMBER, 137 .help = "Gluster log level, valid range is 0-9", 138 }, 139 { 140 .name = GLUSTER_OPT_LOGFILE, 141 .type = QEMU_OPT_STRING, 142 .help = "Logfile path of libgfapi", 143 }, 144 { /* end of list */ } 145 }, 146 }; 147 148 static QemuOptsList runtime_json_opts = { 149 .name = "gluster_json", 150 .head = QTAILQ_HEAD_INITIALIZER(runtime_json_opts.head), 151 .desc = { 152 { 153 .name = GLUSTER_OPT_VOLUME, 154 .type = QEMU_OPT_STRING, 155 .help = "name of gluster volume where VM image resides", 156 }, 157 { 158 .name = GLUSTER_OPT_PATH, 159 .type = QEMU_OPT_STRING, 160 .help = "absolute path to image file in gluster volume", 161 }, 162 { 163 .name = GLUSTER_OPT_DEBUG, 164 .type = QEMU_OPT_NUMBER, 165 .help = "Gluster log level, valid range is 0-9", 166 }, 167 { /* end of list */ } 168 }, 169 }; 170 171 static QemuOptsList runtime_type_opts = { 172 .name = "gluster_type", 173 .head = QTAILQ_HEAD_INITIALIZER(runtime_type_opts.head), 174 .desc = { 175 { 176 .name = GLUSTER_OPT_TYPE, 177 .type = QEMU_OPT_STRING, 178 .help = "inet|unix", 179 }, 180 { /* end of list */ } 181 }, 182 }; 183 184 static QemuOptsList runtime_unix_opts = { 185 .name = "gluster_unix", 186 .head = QTAILQ_HEAD_INITIALIZER(runtime_unix_opts.head), 187 .desc = { 188 { 189 .name = GLUSTER_OPT_SOCKET, 190 .type = QEMU_OPT_STRING, 191 .help = "socket file path (legacy)", 192 }, 193 { 194 .name = GLUSTER_OPT_PATH, 195 .type = QEMU_OPT_STRING, 196 .help = "socket file path (QAPI)", 197 }, 198 { /* end of list */ } 199 }, 200 }; 201 202 static QemuOptsList runtime_inet_opts = { 203 .name = "gluster_inet", 204 .head = QTAILQ_HEAD_INITIALIZER(runtime_inet_opts.head), 205 .desc = { 206 { 207 .name = GLUSTER_OPT_TYPE, 208 .type = QEMU_OPT_STRING, 209 .help = "inet|unix", 210 }, 211 { 212 .name = GLUSTER_OPT_HOST, 213 .type = QEMU_OPT_STRING, 214 .help = "host address (hostname/ipv4/ipv6 addresses)", 215 }, 216 { 217 .name = GLUSTER_OPT_PORT, 218 .type = QEMU_OPT_STRING, 219 .help = "port number on which glusterd is listening (default 24007)", 220 }, 221 { 222 .name = "to", 223 .type = QEMU_OPT_NUMBER, 224 .help = "max port number, not supported by gluster", 225 }, 226 { 227 .name = "ipv4", 228 .type = QEMU_OPT_BOOL, 229 .help = "ipv4 bool value, not supported by gluster", 230 }, 231 { 232 .name = "ipv6", 233 .type = QEMU_OPT_BOOL, 234 .help = "ipv6 bool value, not supported by gluster", 235 }, 236 { /* end of list */ } 237 }, 238 }; 239 240 static void glfs_set_preopened(const char *volume, glfs_t *fs) 241 { 242 ListElement *entry = NULL; 243 244 entry = g_new(ListElement, 1); 245 246 entry->saved.volume = g_strdup(volume); 247 248 entry->saved.fs = fs; 249 entry->saved.ref = 1; 250 251 QLIST_INSERT_HEAD(&glfs_list, entry, list); 252 } 253 254 static glfs_t *glfs_find_preopened(const char *volume) 255 { 256 ListElement *entry = NULL; 257 258 QLIST_FOREACH(entry, &glfs_list, list) { 259 if (strcmp(entry->saved.volume, volume) == 0) { 260 entry->saved.ref++; 261 return entry->saved.fs; 262 } 263 } 264 265 return NULL; 266 } 267 268 static void glfs_clear_preopened(glfs_t *fs) 269 { 270 ListElement *entry = NULL; 271 ListElement *next; 272 273 if (fs == NULL) { 274 return; 275 } 276 277 QLIST_FOREACH_SAFE(entry, &glfs_list, list, next) { 278 if (entry->saved.fs == fs) { 279 if (--entry->saved.ref) { 280 return; 281 } 282 283 QLIST_REMOVE(entry, list); 284 285 glfs_fini(entry->saved.fs); 286 g_free(entry->saved.volume); 287 g_free(entry); 288 } 289 } 290 } 291 292 static int parse_volume_options(BlockdevOptionsGluster *gconf, char *path) 293 { 294 char *p, *q; 295 296 if (!path) { 297 return -EINVAL; 298 } 299 300 /* volume */ 301 p = q = path + strspn(path, "/"); 302 p += strcspn(p, "/"); 303 if (*p == '\0') { 304 return -EINVAL; 305 } 306 gconf->volume = g_strndup(q, p - q); 307 308 /* path */ 309 p += strspn(p, "/"); 310 if (*p == '\0') { 311 return -EINVAL; 312 } 313 gconf->path = g_strdup(p); 314 return 0; 315 } 316 317 /* 318 * file=gluster[+transport]://[host[:port]]/volume/path[?socket=...] 319 * 320 * 'gluster' is the protocol. 321 * 322 * 'transport' specifies the transport type used to connect to gluster 323 * management daemon (glusterd). Valid transport types are 324 * tcp or unix. If a transport type isn't specified, then tcp type is assumed. 325 * 326 * 'host' specifies the host where the volume file specification for 327 * the given volume resides. This can be either hostname or ipv4 address. 328 * If transport type is 'unix', then 'host' field should not be specified. 329 * The 'socket' field needs to be populated with the path to unix domain 330 * socket. 331 * 332 * 'port' is the port number on which glusterd is listening. This is optional 333 * and if not specified, QEMU will send 0 which will make gluster to use the 334 * default port. If the transport type is unix, then 'port' should not be 335 * specified. 336 * 337 * 'volume' is the name of the gluster volume which contains the VM image. 338 * 339 * 'path' is the path to the actual VM image that resides on gluster volume. 340 * 341 * Examples: 342 * 343 * file=gluster://1.2.3.4/testvol/a.img 344 * file=gluster+tcp://1.2.3.4/testvol/a.img 345 * file=gluster+tcp://1.2.3.4:24007/testvol/dir/a.img 346 * file=gluster+tcp://host.domain.com:24007/testvol/dir/a.img 347 * file=gluster+unix:///testvol/dir/a.img?socket=/tmp/glusterd.socket 348 */ 349 static int qemu_gluster_parse_uri(BlockdevOptionsGluster *gconf, 350 const char *filename) 351 { 352 SocketAddress *gsconf; 353 URI *uri; 354 QueryParams *qp = NULL; 355 bool is_unix = false; 356 int ret = 0; 357 358 uri = uri_parse(filename); 359 if (!uri) { 360 return -EINVAL; 361 } 362 363 gsconf = g_new0(SocketAddress, 1); 364 QAPI_LIST_PREPEND(gconf->server, gsconf); 365 366 /* transport */ 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 if (!strcmp(uri->scheme, "gluster+rdma")) { 375 gsconf->type = SOCKET_ADDRESS_TYPE_INET; 376 warn_report("rdma feature is not supported, falling back to tcp"); 377 } else { 378 ret = -EINVAL; 379 goto out; 380 } 381 382 ret = parse_volume_options(gconf, uri->path); 383 if (ret < 0) { 384 goto out; 385 } 386 387 qp = query_params_parse(uri->query); 388 if (qp->n > 1 || (is_unix && !qp->n) || (!is_unix && qp->n)) { 389 ret = -EINVAL; 390 goto out; 391 } 392 393 if (is_unix) { 394 if (uri->server || uri->port) { 395 ret = -EINVAL; 396 goto out; 397 } 398 if (strcmp(qp->p[0].name, "socket")) { 399 ret = -EINVAL; 400 goto out; 401 } 402 gsconf->u.q_unix.path = g_strdup(qp->p[0].value); 403 } else { 404 gsconf->u.inet.host = g_strdup(uri->server ? uri->server : "localhost"); 405 if (uri->port) { 406 gsconf->u.inet.port = g_strdup_printf("%d", uri->port); 407 } else { 408 gsconf->u.inet.port = g_strdup_printf("%d", GLUSTER_DEFAULT_PORT); 409 } 410 } 411 412 out: 413 if (qp) { 414 query_params_free(qp); 415 } 416 uri_free(uri); 417 return ret; 418 } 419 420 static struct glfs *qemu_gluster_glfs_init(BlockdevOptionsGluster *gconf, 421 Error **errp) 422 { 423 struct glfs *glfs; 424 int ret; 425 int old_errno; 426 SocketAddressList *server; 427 uint64_t port; 428 429 glfs = glfs_find_preopened(gconf->volume); 430 if (glfs) { 431 return glfs; 432 } 433 434 glfs = glfs_new(gconf->volume); 435 if (!glfs) { 436 goto out; 437 } 438 439 glfs_set_preopened(gconf->volume, glfs); 440 441 for (server = gconf->server; server; server = server->next) { 442 switch (server->value->type) { 443 case SOCKET_ADDRESS_TYPE_UNIX: 444 ret = glfs_set_volfile_server(glfs, "unix", 445 server->value->u.q_unix.path, 0); 446 break; 447 case SOCKET_ADDRESS_TYPE_INET: 448 if (parse_uint_full(server->value->u.inet.port, 10, &port) < 0 || 449 port > 65535) { 450 error_setg(errp, "'%s' is not a valid port number", 451 server->value->u.inet.port); 452 errno = EINVAL; 453 goto out; 454 } 455 ret = glfs_set_volfile_server(glfs, "tcp", 456 server->value->u.inet.host, 457 (int)port); 458 break; 459 case SOCKET_ADDRESS_TYPE_VSOCK: 460 case SOCKET_ADDRESS_TYPE_FD: 461 default: 462 abort(); 463 } 464 465 if (ret < 0) { 466 goto out; 467 } 468 } 469 470 ret = glfs_set_logging(glfs, gconf->logfile, gconf->debug); 471 if (ret < 0) { 472 goto out; 473 } 474 475 ret = glfs_init(glfs); 476 if (ret) { 477 error_setg(errp, "Gluster connection for volume %s, path %s failed" 478 " to connect", gconf->volume, gconf->path); 479 for (server = gconf->server; server; server = server->next) { 480 if (server->value->type == SOCKET_ADDRESS_TYPE_UNIX) { 481 error_append_hint(errp, "hint: failed on socket %s ", 482 server->value->u.q_unix.path); 483 } else { 484 error_append_hint(errp, "hint: failed on host %s and port %s ", 485 server->value->u.inet.host, 486 server->value->u.inet.port); 487 } 488 } 489 490 error_append_hint(errp, "Please refer to gluster logs for more info\n"); 491 492 /* glfs_init sometimes doesn't set errno although docs suggest that */ 493 if (errno == 0) { 494 errno = EINVAL; 495 } 496 497 goto out; 498 } 499 return glfs; 500 501 out: 502 if (glfs) { 503 old_errno = errno; 504 glfs_clear_preopened(glfs); 505 errno = old_errno; 506 } 507 return NULL; 508 } 509 510 /* 511 * Convert the json formatted command line into qapi. 512 */ 513 static int qemu_gluster_parse_json(BlockdevOptionsGluster *gconf, 514 QDict *options, Error **errp) 515 { 516 QemuOpts *opts; 517 SocketAddress *gsconf = NULL; 518 SocketAddressList **tail; 519 QDict *backing_options = NULL; 520 Error *local_err = NULL; 521 char *str = NULL; 522 const char *ptr; 523 int i, type, num_servers; 524 525 /* create opts info from runtime_json_opts list */ 526 opts = qemu_opts_create(&runtime_json_opts, NULL, 0, &error_abort); 527 if (!qemu_opts_absorb_qdict(opts, options, errp)) { 528 goto out; 529 } 530 531 num_servers = qdict_array_entries(options, GLUSTER_OPT_SERVER_PATTERN); 532 if (num_servers < 1) { 533 error_setg(&local_err, QERR_MISSING_PARAMETER, "server"); 534 goto out; 535 } 536 537 ptr = qemu_opt_get(opts, GLUSTER_OPT_VOLUME); 538 if (!ptr) { 539 error_setg(&local_err, QERR_MISSING_PARAMETER, GLUSTER_OPT_VOLUME); 540 goto out; 541 } 542 gconf->volume = g_strdup(ptr); 543 544 ptr = qemu_opt_get(opts, GLUSTER_OPT_PATH); 545 if (!ptr) { 546 error_setg(&local_err, QERR_MISSING_PARAMETER, GLUSTER_OPT_PATH); 547 goto out; 548 } 549 gconf->path = g_strdup(ptr); 550 qemu_opts_del(opts); 551 tail = &gconf->server; 552 553 for (i = 0; i < num_servers; i++) { 554 str = g_strdup_printf(GLUSTER_OPT_SERVER_PATTERN"%d.", i); 555 qdict_extract_subqdict(options, &backing_options, str); 556 557 /* create opts info from runtime_type_opts list */ 558 opts = qemu_opts_create(&runtime_type_opts, NULL, 0, &error_abort); 559 if (!qemu_opts_absorb_qdict(opts, backing_options, errp)) { 560 goto out; 561 } 562 563 ptr = qemu_opt_get(opts, GLUSTER_OPT_TYPE); 564 if (!ptr) { 565 error_setg(&local_err, QERR_MISSING_PARAMETER, GLUSTER_OPT_TYPE); 566 error_append_hint(&local_err, GERR_INDEX_HINT, i); 567 goto out; 568 569 } 570 gsconf = g_new0(SocketAddress, 1); 571 if (!strcmp(ptr, "tcp")) { 572 ptr = "inet"; /* accept legacy "tcp" */ 573 } 574 type = qapi_enum_parse(&SocketAddressType_lookup, ptr, -1, NULL); 575 if (type != SOCKET_ADDRESS_TYPE_INET 576 && type != SOCKET_ADDRESS_TYPE_UNIX) { 577 error_setg(&local_err, 578 "Parameter '%s' may be 'inet' or 'unix'", 579 GLUSTER_OPT_TYPE); 580 error_append_hint(&local_err, GERR_INDEX_HINT, i); 581 goto out; 582 } 583 gsconf->type = type; 584 qemu_opts_del(opts); 585 586 if (gsconf->type == SOCKET_ADDRESS_TYPE_INET) { 587 /* create opts info from runtime_inet_opts list */ 588 opts = qemu_opts_create(&runtime_inet_opts, NULL, 0, &error_abort); 589 if (!qemu_opts_absorb_qdict(opts, backing_options, errp)) { 590 goto out; 591 } 592 593 ptr = qemu_opt_get(opts, GLUSTER_OPT_HOST); 594 if (!ptr) { 595 error_setg(&local_err, QERR_MISSING_PARAMETER, 596 GLUSTER_OPT_HOST); 597 error_append_hint(&local_err, GERR_INDEX_HINT, i); 598 goto out; 599 } 600 gsconf->u.inet.host = g_strdup(ptr); 601 ptr = qemu_opt_get(opts, GLUSTER_OPT_PORT); 602 if (!ptr) { 603 error_setg(&local_err, QERR_MISSING_PARAMETER, 604 GLUSTER_OPT_PORT); 605 error_append_hint(&local_err, GERR_INDEX_HINT, i); 606 goto out; 607 } 608 gsconf->u.inet.port = g_strdup(ptr); 609 610 /* defend for unsupported fields in InetSocketAddress, 611 * i.e. @ipv4, @ipv6 and @to 612 */ 613 ptr = qemu_opt_get(opts, GLUSTER_OPT_TO); 614 if (ptr) { 615 gsconf->u.inet.has_to = true; 616 } 617 ptr = qemu_opt_get(opts, GLUSTER_OPT_IPV4); 618 if (ptr) { 619 gsconf->u.inet.has_ipv4 = true; 620 } 621 ptr = qemu_opt_get(opts, GLUSTER_OPT_IPV6); 622 if (ptr) { 623 gsconf->u.inet.has_ipv6 = true; 624 } 625 if (gsconf->u.inet.has_to) { 626 error_setg(&local_err, "Parameter 'to' not supported"); 627 goto out; 628 } 629 if (gsconf->u.inet.has_ipv4 || gsconf->u.inet.has_ipv6) { 630 error_setg(&local_err, "Parameters 'ipv4/ipv6' not supported"); 631 goto out; 632 } 633 qemu_opts_del(opts); 634 } else { 635 /* create opts info from runtime_unix_opts list */ 636 opts = qemu_opts_create(&runtime_unix_opts, NULL, 0, &error_abort); 637 if (!qemu_opts_absorb_qdict(opts, backing_options, errp)) { 638 goto out; 639 } 640 641 ptr = qemu_opt_get(opts, GLUSTER_OPT_PATH); 642 if (!ptr) { 643 ptr = qemu_opt_get(opts, GLUSTER_OPT_SOCKET); 644 } else if (qemu_opt_get(opts, GLUSTER_OPT_SOCKET)) { 645 error_setg(&local_err, 646 "Conflicting parameters 'path' and 'socket'"); 647 error_append_hint(&local_err, GERR_INDEX_HINT, i); 648 goto out; 649 } 650 if (!ptr) { 651 error_setg(&local_err, QERR_MISSING_PARAMETER, 652 GLUSTER_OPT_PATH); 653 error_append_hint(&local_err, GERR_INDEX_HINT, i); 654 goto out; 655 } 656 gsconf->u.q_unix.path = g_strdup(ptr); 657 qemu_opts_del(opts); 658 } 659 660 QAPI_LIST_APPEND(tail, gsconf); 661 gsconf = NULL; 662 663 qobject_unref(backing_options); 664 backing_options = NULL; 665 g_free(str); 666 str = NULL; 667 } 668 669 return 0; 670 671 out: 672 error_propagate(errp, local_err); 673 qapi_free_SocketAddress(gsconf); 674 qemu_opts_del(opts); 675 g_free(str); 676 qobject_unref(backing_options); 677 errno = EINVAL; 678 return -errno; 679 } 680 681 /* Converts options given in @filename and the @options QDict into the QAPI 682 * object @gconf. */ 683 static int qemu_gluster_parse(BlockdevOptionsGluster *gconf, 684 const char *filename, 685 QDict *options, Error **errp) 686 { 687 int ret; 688 if (filename) { 689 ret = qemu_gluster_parse_uri(gconf, filename); 690 if (ret < 0) { 691 error_setg(errp, "invalid URI %s", filename); 692 error_append_hint(errp, "Usage: file=gluster[+transport]://" 693 "[host[:port]]volume/path[?socket=...]" 694 "[,file.debug=N]" 695 "[,file.logfile=/path/filename.log]\n"); 696 return ret; 697 } 698 } else { 699 ret = qemu_gluster_parse_json(gconf, options, errp); 700 if (ret < 0) { 701 error_append_hint(errp, "Usage: " 702 "-drive driver=qcow2,file.driver=gluster," 703 "file.volume=testvol,file.path=/path/a.qcow2" 704 "[,file.debug=9]" 705 "[,file.logfile=/path/filename.log]," 706 "file.server.0.type=inet," 707 "file.server.0.host=1.2.3.4," 708 "file.server.0.port=24007," 709 "file.server.1.transport=unix," 710 "file.server.1.path=/var/run/glusterd.socket ..." 711 "\n"); 712 return ret; 713 } 714 } 715 716 return 0; 717 } 718 719 static struct glfs *qemu_gluster_init(BlockdevOptionsGluster *gconf, 720 const char *filename, 721 QDict *options, Error **errp) 722 { 723 int ret; 724 725 ret = qemu_gluster_parse(gconf, filename, options, errp); 726 if (ret < 0) { 727 errno = -ret; 728 return NULL; 729 } 730 731 return qemu_gluster_glfs_init(gconf, errp); 732 } 733 734 /* 735 * AIO callback routine called from GlusterFS thread. 736 */ 737 static void gluster_finish_aiocb(struct glfs_fd *fd, ssize_t ret, 738 #ifdef CONFIG_GLUSTERFS_IOCB_HAS_STAT 739 struct glfs_stat *pre, struct glfs_stat *post, 740 #endif 741 void *arg) 742 { 743 GlusterAIOCB *acb = (GlusterAIOCB *)arg; 744 745 if (!ret || ret == acb->size) { 746 acb->ret = 0; /* Success */ 747 } else if (ret < 0) { 748 acb->ret = -errno; /* Read/Write failed */ 749 } else { 750 acb->ret = -EIO; /* Partial read/write - fail it */ 751 } 752 753 aio_co_schedule(acb->aio_context, acb->coroutine); 754 } 755 756 static void qemu_gluster_parse_flags(int bdrv_flags, int *open_flags) 757 { 758 assert(open_flags != NULL); 759 760 *open_flags |= O_BINARY; 761 762 if (bdrv_flags & BDRV_O_RDWR) { 763 *open_flags |= O_RDWR; 764 } else { 765 *open_flags |= O_RDONLY; 766 } 767 768 if ((bdrv_flags & BDRV_O_NOCACHE)) { 769 *open_flags |= O_DIRECT; 770 } 771 } 772 773 /* 774 * Do SEEK_DATA/HOLE to detect if it is functional. Older broken versions of 775 * gfapi incorrectly return the current offset when SEEK_DATA/HOLE is used. 776 * - Corrected versions return -1 and set errno to EINVAL. 777 * - Versions that support SEEK_DATA/HOLE correctly, will return -1 and set 778 * errno to ENXIO when SEEK_DATA is called with a position of EOF. 779 */ 780 static bool qemu_gluster_test_seek(struct glfs_fd *fd) 781 { 782 off_t ret = 0; 783 784 #if defined SEEK_HOLE && defined SEEK_DATA 785 off_t eof; 786 787 eof = glfs_lseek(fd, 0, SEEK_END); 788 if (eof < 0) { 789 /* this should never occur */ 790 return false; 791 } 792 793 /* this should always fail with ENXIO if SEEK_DATA is supported */ 794 ret = glfs_lseek(fd, eof, SEEK_DATA); 795 #endif 796 797 return (ret < 0) && (errno == ENXIO); 798 } 799 800 static int qemu_gluster_open(BlockDriverState *bs, QDict *options, 801 int bdrv_flags, Error **errp) 802 { 803 BDRVGlusterState *s = bs->opaque; 804 int open_flags = 0; 805 int ret = 0; 806 BlockdevOptionsGluster *gconf = NULL; 807 QemuOpts *opts; 808 const char *filename, *logfile; 809 810 opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort); 811 if (!qemu_opts_absorb_qdict(opts, options, errp)) { 812 ret = -EINVAL; 813 goto out; 814 } 815 816 filename = qemu_opt_get(opts, GLUSTER_OPT_FILENAME); 817 818 s->debug = qemu_opt_get_number(opts, GLUSTER_OPT_DEBUG, 819 GLUSTER_DEBUG_DEFAULT); 820 if (s->debug < 0) { 821 s->debug = 0; 822 } else if (s->debug > GLUSTER_DEBUG_MAX) { 823 s->debug = GLUSTER_DEBUG_MAX; 824 } 825 826 gconf = g_new0(BlockdevOptionsGluster, 1); 827 gconf->debug = s->debug; 828 gconf->has_debug = true; 829 830 logfile = qemu_opt_get(opts, GLUSTER_OPT_LOGFILE); 831 s->logfile = g_strdup(logfile ? logfile : GLUSTER_LOGFILE_DEFAULT); 832 833 gconf->logfile = g_strdup(s->logfile); 834 835 s->glfs = qemu_gluster_init(gconf, filename, options, errp); 836 if (!s->glfs) { 837 ret = -errno; 838 goto out; 839 } 840 841 #ifdef CONFIG_GLUSTERFS_XLATOR_OPT 842 /* Without this, if fsync fails for a recoverable reason (for instance, 843 * ENOSPC), gluster will dump its cache, preventing retries. This means 844 * almost certain data loss. Not all gluster versions support the 845 * 'resync-failed-syncs-after-fsync' key value, but there is no way to 846 * discover during runtime if it is supported (this api returns success for 847 * unknown key/value pairs) */ 848 ret = glfs_set_xlator_option(s->glfs, "*-write-behind", 849 "resync-failed-syncs-after-fsync", 850 "on"); 851 if (ret < 0) { 852 error_setg_errno(errp, errno, "Unable to set xlator key/value pair"); 853 ret = -errno; 854 goto out; 855 } 856 #endif 857 858 qemu_gluster_parse_flags(bdrv_flags, &open_flags); 859 860 s->fd = glfs_open(s->glfs, gconf->path, open_flags); 861 ret = s->fd ? 0 : -errno; 862 863 if (ret == -EACCES || ret == -EROFS) { 864 /* Try to degrade to read-only, but if it doesn't work, still use the 865 * normal error message. */ 866 bdrv_graph_rdlock_main_loop(); 867 if (bdrv_apply_auto_read_only(bs, NULL, NULL) == 0) { 868 open_flags = (open_flags & ~O_RDWR) | O_RDONLY; 869 s->fd = glfs_open(s->glfs, gconf->path, open_flags); 870 ret = s->fd ? 0 : -errno; 871 } 872 bdrv_graph_rdunlock_main_loop(); 873 } 874 875 s->supports_seek_data = qemu_gluster_test_seek(s->fd); 876 877 out: 878 qemu_opts_del(opts); 879 qapi_free_BlockdevOptionsGluster(gconf); 880 if (!ret) { 881 return ret; 882 } 883 g_free(s->logfile); 884 if (s->fd) { 885 glfs_close(s->fd); 886 } 887 888 glfs_clear_preopened(s->glfs); 889 890 return ret; 891 } 892 893 static void qemu_gluster_refresh_limits(BlockDriverState *bs, Error **errp) 894 { 895 bs->bl.max_transfer = GLUSTER_MAX_TRANSFER; 896 bs->bl.max_pdiscard = MIN(SIZE_MAX, INT64_MAX); 897 } 898 899 static int qemu_gluster_reopen_prepare(BDRVReopenState *state, 900 BlockReopenQueue *queue, Error **errp) 901 { 902 int ret = 0; 903 BDRVGlusterState *s; 904 BDRVGlusterReopenState *reop_s; 905 BlockdevOptionsGluster *gconf; 906 int open_flags = 0; 907 908 assert(state != NULL); 909 assert(state->bs != NULL); 910 911 s = state->bs->opaque; 912 913 state->opaque = g_new0(BDRVGlusterReopenState, 1); 914 reop_s = state->opaque; 915 916 qemu_gluster_parse_flags(state->flags, &open_flags); 917 918 gconf = g_new0(BlockdevOptionsGluster, 1); 919 gconf->debug = s->debug; 920 gconf->has_debug = true; 921 gconf->logfile = g_strdup(s->logfile); 922 923 /* 924 * If 'state->bs->exact_filename' is empty, 'state->options' should contain 925 * the JSON parameters already parsed. 926 */ 927 if (state->bs->exact_filename[0] != '\0') { 928 reop_s->glfs = qemu_gluster_init(gconf, state->bs->exact_filename, NULL, 929 errp); 930 } else { 931 reop_s->glfs = qemu_gluster_init(gconf, NULL, state->options, errp); 932 } 933 if (reop_s->glfs == NULL) { 934 ret = -errno; 935 goto exit; 936 } 937 938 #ifdef CONFIG_GLUSTERFS_XLATOR_OPT 939 ret = glfs_set_xlator_option(reop_s->glfs, "*-write-behind", 940 "resync-failed-syncs-after-fsync", "on"); 941 if (ret < 0) { 942 error_setg_errno(errp, errno, "Unable to set xlator key/value pair"); 943 ret = -errno; 944 goto exit; 945 } 946 #endif 947 948 reop_s->fd = glfs_open(reop_s->glfs, gconf->path, open_flags); 949 if (reop_s->fd == NULL) { 950 /* reops->glfs will be cleaned up in _abort */ 951 ret = -errno; 952 goto exit; 953 } 954 955 exit: 956 /* state->opaque will be freed in either the _abort or _commit */ 957 qapi_free_BlockdevOptionsGluster(gconf); 958 return ret; 959 } 960 961 static void qemu_gluster_reopen_commit(BDRVReopenState *state) 962 { 963 BDRVGlusterReopenState *reop_s = state->opaque; 964 BDRVGlusterState *s = state->bs->opaque; 965 966 967 /* close the old */ 968 if (s->fd) { 969 glfs_close(s->fd); 970 } 971 972 glfs_clear_preopened(s->glfs); 973 974 /* use the newly opened image / connection */ 975 s->fd = reop_s->fd; 976 s->glfs = reop_s->glfs; 977 978 g_free(state->opaque); 979 state->opaque = NULL; 980 981 return; 982 } 983 984 985 static void qemu_gluster_reopen_abort(BDRVReopenState *state) 986 { 987 BDRVGlusterReopenState *reop_s = state->opaque; 988 989 if (reop_s == NULL) { 990 return; 991 } 992 993 if (reop_s->fd) { 994 glfs_close(reop_s->fd); 995 } 996 997 glfs_clear_preopened(reop_s->glfs); 998 999 g_free(state->opaque); 1000 state->opaque = NULL; 1001 1002 return; 1003 } 1004 1005 #ifdef CONFIG_GLUSTERFS_ZEROFILL 1006 static coroutine_fn int qemu_gluster_co_pwrite_zeroes(BlockDriverState *bs, 1007 int64_t offset, 1008 int64_t bytes, 1009 BdrvRequestFlags flags) 1010 { 1011 int ret; 1012 GlusterAIOCB acb; 1013 BDRVGlusterState *s = bs->opaque; 1014 1015 acb.size = bytes; 1016 acb.ret = 0; 1017 acb.coroutine = qemu_coroutine_self(); 1018 acb.aio_context = bdrv_get_aio_context(bs); 1019 1020 ret = glfs_zerofill_async(s->fd, offset, bytes, gluster_finish_aiocb, &acb); 1021 if (ret < 0) { 1022 return -errno; 1023 } 1024 1025 qemu_coroutine_yield(); 1026 return acb.ret; 1027 } 1028 #endif 1029 1030 static int qemu_gluster_do_truncate(struct glfs_fd *fd, int64_t offset, 1031 PreallocMode prealloc, Error **errp) 1032 { 1033 int64_t current_length; 1034 1035 current_length = glfs_lseek(fd, 0, SEEK_END); 1036 if (current_length < 0) { 1037 error_setg_errno(errp, errno, "Failed to determine current size"); 1038 return -errno; 1039 } 1040 1041 if (current_length > offset && prealloc != PREALLOC_MODE_OFF) { 1042 error_setg(errp, "Cannot use preallocation for shrinking files"); 1043 return -ENOTSUP; 1044 } 1045 1046 if (current_length == offset) { 1047 return 0; 1048 } 1049 1050 switch (prealloc) { 1051 #ifdef CONFIG_GLUSTERFS_FALLOCATE 1052 case PREALLOC_MODE_FALLOC: 1053 if (glfs_fallocate(fd, 0, current_length, offset - current_length)) { 1054 error_setg_errno(errp, errno, "Could not preallocate data"); 1055 return -errno; 1056 } 1057 break; 1058 #endif /* CONFIG_GLUSTERFS_FALLOCATE */ 1059 #ifdef CONFIG_GLUSTERFS_ZEROFILL 1060 case PREALLOC_MODE_FULL: 1061 if (glfs_ftruncate(fd, offset)) { 1062 error_setg_errno(errp, errno, "Could not resize file"); 1063 return -errno; 1064 } 1065 if (glfs_zerofill(fd, current_length, offset - current_length)) { 1066 error_setg_errno(errp, errno, "Could not zerofill the new area"); 1067 return -errno; 1068 } 1069 break; 1070 #endif /* CONFIG_GLUSTERFS_ZEROFILL */ 1071 case PREALLOC_MODE_OFF: 1072 if (glfs_ftruncate(fd, offset)) { 1073 error_setg_errno(errp, errno, "Could not resize file"); 1074 return -errno; 1075 } 1076 break; 1077 default: 1078 error_setg(errp, "Unsupported preallocation mode: %s", 1079 PreallocMode_str(prealloc)); 1080 return -EINVAL; 1081 } 1082 1083 return 0; 1084 } 1085 1086 static int qemu_gluster_co_create(BlockdevCreateOptions *options, 1087 Error **errp) 1088 { 1089 BlockdevCreateOptionsGluster *opts = &options->u.gluster; 1090 struct glfs *glfs; 1091 struct glfs_fd *fd = NULL; 1092 int ret = 0; 1093 1094 assert(options->driver == BLOCKDEV_DRIVER_GLUSTER); 1095 1096 glfs = qemu_gluster_glfs_init(opts->location, errp); 1097 if (!glfs) { 1098 ret = -errno; 1099 goto out; 1100 } 1101 1102 fd = glfs_creat(glfs, opts->location->path, 1103 O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, S_IRUSR | S_IWUSR); 1104 if (!fd) { 1105 ret = -errno; 1106 goto out; 1107 } 1108 1109 ret = qemu_gluster_do_truncate(fd, opts->size, opts->preallocation, errp); 1110 1111 out: 1112 if (fd) { 1113 if (glfs_close(fd) != 0 && ret == 0) { 1114 ret = -errno; 1115 } 1116 } 1117 glfs_clear_preopened(glfs); 1118 return ret; 1119 } 1120 1121 static int coroutine_fn qemu_gluster_co_create_opts(BlockDriver *drv, 1122 const char *filename, 1123 QemuOpts *opts, 1124 Error **errp) 1125 { 1126 BlockdevCreateOptions *options; 1127 BlockdevCreateOptionsGluster *gopts; 1128 BlockdevOptionsGluster *gconf; 1129 char *tmp = NULL; 1130 Error *local_err = NULL; 1131 int ret; 1132 1133 options = g_new0(BlockdevCreateOptions, 1); 1134 options->driver = BLOCKDEV_DRIVER_GLUSTER; 1135 gopts = &options->u.gluster; 1136 1137 gconf = g_new0(BlockdevOptionsGluster, 1); 1138 gopts->location = gconf; 1139 1140 gopts->size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0), 1141 BDRV_SECTOR_SIZE); 1142 1143 tmp = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC); 1144 gopts->preallocation = qapi_enum_parse(&PreallocMode_lookup, tmp, 1145 PREALLOC_MODE_OFF, &local_err); 1146 g_free(tmp); 1147 if (local_err) { 1148 error_propagate(errp, local_err); 1149 ret = -EINVAL; 1150 goto fail; 1151 } 1152 1153 gconf->debug = qemu_opt_get_number_del(opts, GLUSTER_OPT_DEBUG, 1154 GLUSTER_DEBUG_DEFAULT); 1155 if (gconf->debug < 0) { 1156 gconf->debug = 0; 1157 } else if (gconf->debug > GLUSTER_DEBUG_MAX) { 1158 gconf->debug = GLUSTER_DEBUG_MAX; 1159 } 1160 gconf->has_debug = true; 1161 1162 gconf->logfile = qemu_opt_get_del(opts, GLUSTER_OPT_LOGFILE); 1163 if (!gconf->logfile) { 1164 gconf->logfile = g_strdup(GLUSTER_LOGFILE_DEFAULT); 1165 } 1166 1167 ret = qemu_gluster_parse(gconf, filename, NULL, errp); 1168 if (ret < 0) { 1169 goto fail; 1170 } 1171 1172 ret = qemu_gluster_co_create(options, errp); 1173 if (ret < 0) { 1174 goto fail; 1175 } 1176 1177 ret = 0; 1178 fail: 1179 qapi_free_BlockdevCreateOptions(options); 1180 return ret; 1181 } 1182 1183 static coroutine_fn int qemu_gluster_co_rw(BlockDriverState *bs, 1184 int64_t sector_num, int nb_sectors, 1185 QEMUIOVector *qiov, int write) 1186 { 1187 int ret; 1188 GlusterAIOCB acb; 1189 BDRVGlusterState *s = bs->opaque; 1190 size_t size = nb_sectors * BDRV_SECTOR_SIZE; 1191 off_t offset = sector_num * BDRV_SECTOR_SIZE; 1192 1193 acb.size = size; 1194 acb.ret = 0; 1195 acb.coroutine = qemu_coroutine_self(); 1196 acb.aio_context = bdrv_get_aio_context(bs); 1197 1198 if (write) { 1199 ret = glfs_pwritev_async(s->fd, qiov->iov, qiov->niov, offset, 0, 1200 gluster_finish_aiocb, &acb); 1201 } else { 1202 ret = glfs_preadv_async(s->fd, qiov->iov, qiov->niov, offset, 0, 1203 gluster_finish_aiocb, &acb); 1204 } 1205 1206 if (ret < 0) { 1207 return -errno; 1208 } 1209 1210 qemu_coroutine_yield(); 1211 return acb.ret; 1212 } 1213 1214 static coroutine_fn int qemu_gluster_co_truncate(BlockDriverState *bs, 1215 int64_t offset, 1216 bool exact, 1217 PreallocMode prealloc, 1218 BdrvRequestFlags flags, 1219 Error **errp) 1220 { 1221 BDRVGlusterState *s = bs->opaque; 1222 return qemu_gluster_do_truncate(s->fd, offset, prealloc, errp); 1223 } 1224 1225 static coroutine_fn int qemu_gluster_co_readv(BlockDriverState *bs, 1226 int64_t sector_num, 1227 int nb_sectors, 1228 QEMUIOVector *qiov) 1229 { 1230 return qemu_gluster_co_rw(bs, sector_num, nb_sectors, qiov, 0); 1231 } 1232 1233 static coroutine_fn int qemu_gluster_co_writev(BlockDriverState *bs, 1234 int64_t sector_num, 1235 int nb_sectors, 1236 QEMUIOVector *qiov, 1237 int flags) 1238 { 1239 return qemu_gluster_co_rw(bs, sector_num, nb_sectors, qiov, 1); 1240 } 1241 1242 static void qemu_gluster_close(BlockDriverState *bs) 1243 { 1244 BDRVGlusterState *s = bs->opaque; 1245 1246 g_free(s->logfile); 1247 if (s->fd) { 1248 glfs_close(s->fd); 1249 s->fd = NULL; 1250 } 1251 glfs_clear_preopened(s->glfs); 1252 } 1253 1254 static coroutine_fn int qemu_gluster_co_flush_to_disk(BlockDriverState *bs) 1255 { 1256 int ret; 1257 GlusterAIOCB acb; 1258 BDRVGlusterState *s = bs->opaque; 1259 1260 acb.size = 0; 1261 acb.ret = 0; 1262 acb.coroutine = qemu_coroutine_self(); 1263 acb.aio_context = bdrv_get_aio_context(bs); 1264 1265 ret = glfs_fsync_async(s->fd, gluster_finish_aiocb, &acb); 1266 if (ret < 0) { 1267 ret = -errno; 1268 goto error; 1269 } 1270 1271 qemu_coroutine_yield(); 1272 if (acb.ret < 0) { 1273 ret = acb.ret; 1274 goto error; 1275 } 1276 1277 return acb.ret; 1278 1279 error: 1280 /* Some versions of Gluster (3.5.6 -> 3.5.8?) will not retain its cache 1281 * after a fsync failure, so we have no way of allowing the guest to safely 1282 * continue. Gluster versions prior to 3.5.6 don't retain the cache 1283 * either, but will invalidate the fd on error, so this is again our only 1284 * option. 1285 * 1286 * The 'resync-failed-syncs-after-fsync' xlator option for the 1287 * write-behind cache will cause later gluster versions to retain its 1288 * cache after error, so long as the fd remains open. However, we 1289 * currently have no way of knowing if this option is supported. 1290 * 1291 * TODO: Once gluster provides a way for us to determine if the option 1292 * is supported, bypass the closure and setting drv to NULL. */ 1293 qemu_gluster_close(bs); 1294 bs->drv = NULL; 1295 return ret; 1296 } 1297 1298 #ifdef CONFIG_GLUSTERFS_DISCARD 1299 static coroutine_fn int qemu_gluster_co_pdiscard(BlockDriverState *bs, 1300 int64_t offset, int64_t bytes) 1301 { 1302 int ret; 1303 GlusterAIOCB acb; 1304 BDRVGlusterState *s = bs->opaque; 1305 1306 assert(bytes <= SIZE_MAX); /* rely on max_pdiscard */ 1307 1308 acb.size = 0; 1309 acb.ret = 0; 1310 acb.coroutine = qemu_coroutine_self(); 1311 acb.aio_context = bdrv_get_aio_context(bs); 1312 1313 ret = glfs_discard_async(s->fd, offset, bytes, gluster_finish_aiocb, &acb); 1314 if (ret < 0) { 1315 return -errno; 1316 } 1317 1318 qemu_coroutine_yield(); 1319 return acb.ret; 1320 } 1321 #endif 1322 1323 static int64_t coroutine_fn qemu_gluster_co_getlength(BlockDriverState *bs) 1324 { 1325 BDRVGlusterState *s = bs->opaque; 1326 int64_t ret; 1327 1328 ret = glfs_lseek(s->fd, 0, SEEK_END); 1329 if (ret < 0) { 1330 return -errno; 1331 } else { 1332 return ret; 1333 } 1334 } 1335 1336 static int64_t coroutine_fn 1337 qemu_gluster_co_get_allocated_file_size(BlockDriverState *bs) 1338 { 1339 BDRVGlusterState *s = bs->opaque; 1340 struct stat st; 1341 int ret; 1342 1343 ret = glfs_fstat(s->fd, &st); 1344 if (ret < 0) { 1345 return -errno; 1346 } else { 1347 return st.st_blocks * 512; 1348 } 1349 } 1350 1351 /* 1352 * Find allocation range in @bs around offset @start. 1353 * May change underlying file descriptor's file offset. 1354 * If @start is not in a hole, store @start in @data, and the 1355 * beginning of the next hole in @hole, and return 0. 1356 * If @start is in a non-trailing hole, store @start in @hole and the 1357 * beginning of the next non-hole in @data, and return 0. 1358 * If @start is in a trailing hole or beyond EOF, return -ENXIO. 1359 * If we can't find out, return a negative errno other than -ENXIO. 1360 * 1361 * (Shamefully copied from file-posix.c, only minuscule adaptions.) 1362 */ 1363 static int find_allocation(BlockDriverState *bs, off_t start, 1364 off_t *data, off_t *hole) 1365 { 1366 BDRVGlusterState *s = bs->opaque; 1367 1368 if (!s->supports_seek_data) { 1369 goto exit; 1370 } 1371 1372 #if defined SEEK_HOLE && defined SEEK_DATA 1373 off_t offs; 1374 1375 /* 1376 * SEEK_DATA cases: 1377 * D1. offs == start: start is in data 1378 * D2. offs > start: start is in a hole, next data at offs 1379 * D3. offs < 0, errno = ENXIO: either start is in a trailing hole 1380 * or start is beyond EOF 1381 * If the latter happens, the file has been truncated behind 1382 * our back since we opened it. All bets are off then. 1383 * Treating like a trailing hole is simplest. 1384 * D4. offs < 0, errno != ENXIO: we learned nothing 1385 */ 1386 offs = glfs_lseek(s->fd, start, SEEK_DATA); 1387 if (offs < 0) { 1388 return -errno; /* D3 or D4 */ 1389 } 1390 1391 if (offs < start) { 1392 /* This is not a valid return by lseek(). We are safe to just return 1393 * -EIO in this case, and we'll treat it like D4. Unfortunately some 1394 * versions of gluster server will return offs < start, so an assert 1395 * here will unnecessarily abort QEMU. */ 1396 return -EIO; 1397 } 1398 1399 if (offs > start) { 1400 /* D2: in hole, next data at offs */ 1401 *hole = start; 1402 *data = offs; 1403 return 0; 1404 } 1405 1406 /* D1: in data, end not yet known */ 1407 1408 /* 1409 * SEEK_HOLE cases: 1410 * H1. offs == start: start is in a hole 1411 * If this happens here, a hole has been dug behind our back 1412 * since the previous lseek(). 1413 * H2. offs > start: either start is in data, next hole at offs, 1414 * or start is in trailing hole, EOF at offs 1415 * Linux treats trailing holes like any other hole: offs == 1416 * start. Solaris seeks to EOF instead: offs > start (blech). 1417 * If that happens here, a hole has been dug behind our back 1418 * since the previous lseek(). 1419 * H3. offs < 0, errno = ENXIO: start is beyond EOF 1420 * If this happens, the file has been truncated behind our 1421 * back since we opened it. Treat it like a trailing hole. 1422 * H4. offs < 0, errno != ENXIO: we learned nothing 1423 * Pretend we know nothing at all, i.e. "forget" about D1. 1424 */ 1425 offs = glfs_lseek(s->fd, start, SEEK_HOLE); 1426 if (offs < 0) { 1427 return -errno; /* D1 and (H3 or H4) */ 1428 } 1429 1430 if (offs < start) { 1431 /* This is not a valid return by lseek(). We are safe to just return 1432 * -EIO in this case, and we'll treat it like H4. Unfortunately some 1433 * versions of gluster server will return offs < start, so an assert 1434 * here will unnecessarily abort QEMU. */ 1435 return -EIO; 1436 } 1437 1438 if (offs > start) { 1439 /* 1440 * D1 and H2: either in data, next hole at offs, or it was in 1441 * data but is now in a trailing hole. In the latter case, 1442 * all bets are off. Treating it as if it there was data all 1443 * the way to EOF is safe, so simply do that. 1444 */ 1445 *data = start; 1446 *hole = offs; 1447 return 0; 1448 } 1449 1450 /* D1 and H1 */ 1451 return -EBUSY; 1452 #endif 1453 1454 exit: 1455 return -ENOTSUP; 1456 } 1457 1458 /* 1459 * Returns the allocation status of the specified offset. 1460 * 1461 * The block layer guarantees 'offset' and 'bytes' are within bounds. 1462 * 1463 * 'pnum' is set to the number of bytes (including and immediately following 1464 * the specified offset) that are known to be in the same 1465 * allocated/unallocated state. 1466 * 1467 * 'bytes' is a soft cap for 'pnum'. If the information is free, 'pnum' may 1468 * well exceed it. 1469 * 1470 * (Based on raw_co_block_status() from file-posix.c.) 1471 */ 1472 static int coroutine_fn qemu_gluster_co_block_status(BlockDriverState *bs, 1473 bool want_zero, 1474 int64_t offset, 1475 int64_t bytes, 1476 int64_t *pnum, 1477 int64_t *map, 1478 BlockDriverState **file) 1479 { 1480 BDRVGlusterState *s = bs->opaque; 1481 off_t data = 0, hole = 0; 1482 int ret = -EINVAL; 1483 1484 assert(QEMU_IS_ALIGNED(offset | bytes, bs->bl.request_alignment)); 1485 1486 if (!s->fd) { 1487 return ret; 1488 } 1489 1490 if (!want_zero) { 1491 *pnum = bytes; 1492 *map = offset; 1493 *file = bs; 1494 return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID; 1495 } 1496 1497 ret = find_allocation(bs, offset, &data, &hole); 1498 if (ret == -ENXIO) { 1499 /* Trailing hole */ 1500 *pnum = bytes; 1501 ret = BDRV_BLOCK_ZERO; 1502 } else if (ret < 0) { 1503 /* No info available, so pretend there are no holes */ 1504 *pnum = bytes; 1505 ret = BDRV_BLOCK_DATA; 1506 } else if (data == offset) { 1507 /* On a data extent, compute bytes to the end of the extent, 1508 * possibly including a partial sector at EOF. */ 1509 *pnum = hole - offset; 1510 1511 /* 1512 * We are not allowed to return partial sectors, though, so 1513 * round up if necessary. 1514 */ 1515 if (!QEMU_IS_ALIGNED(*pnum, bs->bl.request_alignment)) { 1516 int64_t file_length = qemu_gluster_co_getlength(bs); 1517 if (file_length > 0) { 1518 /* Ignore errors, this is just a safeguard */ 1519 assert(hole == file_length); 1520 } 1521 *pnum = ROUND_UP(*pnum, bs->bl.request_alignment); 1522 } 1523 1524 ret = BDRV_BLOCK_DATA; 1525 } else { 1526 /* On a hole, compute bytes to the beginning of the next extent. */ 1527 assert(hole == offset); 1528 *pnum = data - offset; 1529 ret = BDRV_BLOCK_ZERO; 1530 } 1531 1532 *map = offset; 1533 *file = bs; 1534 1535 return ret | BDRV_BLOCK_OFFSET_VALID; 1536 } 1537 1538 1539 static const char *const gluster_strong_open_opts[] = { 1540 GLUSTER_OPT_VOLUME, 1541 GLUSTER_OPT_PATH, 1542 GLUSTER_OPT_TYPE, 1543 GLUSTER_OPT_SERVER_PATTERN, 1544 GLUSTER_OPT_HOST, 1545 GLUSTER_OPT_PORT, 1546 GLUSTER_OPT_TO, 1547 GLUSTER_OPT_IPV4, 1548 GLUSTER_OPT_IPV6, 1549 GLUSTER_OPT_SOCKET, 1550 1551 NULL 1552 }; 1553 1554 static BlockDriver bdrv_gluster = { 1555 .format_name = "gluster", 1556 .protocol_name = "gluster", 1557 .instance_size = sizeof(BDRVGlusterState), 1558 .bdrv_file_open = qemu_gluster_open, 1559 .bdrv_reopen_prepare = qemu_gluster_reopen_prepare, 1560 .bdrv_reopen_commit = qemu_gluster_reopen_commit, 1561 .bdrv_reopen_abort = qemu_gluster_reopen_abort, 1562 .bdrv_close = qemu_gluster_close, 1563 .bdrv_co_create = qemu_gluster_co_create, 1564 .bdrv_co_create_opts = qemu_gluster_co_create_opts, 1565 .bdrv_co_getlength = qemu_gluster_co_getlength, 1566 .bdrv_co_get_allocated_file_size = qemu_gluster_co_get_allocated_file_size, 1567 .bdrv_co_truncate = qemu_gluster_co_truncate, 1568 .bdrv_co_readv = qemu_gluster_co_readv, 1569 .bdrv_co_writev = qemu_gluster_co_writev, 1570 .bdrv_co_flush_to_disk = qemu_gluster_co_flush_to_disk, 1571 #ifdef CONFIG_GLUSTERFS_DISCARD 1572 .bdrv_co_pdiscard = qemu_gluster_co_pdiscard, 1573 #endif 1574 #ifdef CONFIG_GLUSTERFS_ZEROFILL 1575 .bdrv_co_pwrite_zeroes = qemu_gluster_co_pwrite_zeroes, 1576 #endif 1577 .bdrv_co_block_status = qemu_gluster_co_block_status, 1578 .bdrv_refresh_limits = qemu_gluster_refresh_limits, 1579 .create_opts = &qemu_gluster_create_opts, 1580 .strong_runtime_opts = gluster_strong_open_opts, 1581 }; 1582 1583 static BlockDriver bdrv_gluster_tcp = { 1584 .format_name = "gluster", 1585 .protocol_name = "gluster+tcp", 1586 .instance_size = sizeof(BDRVGlusterState), 1587 .bdrv_file_open = qemu_gluster_open, 1588 .bdrv_reopen_prepare = qemu_gluster_reopen_prepare, 1589 .bdrv_reopen_commit = qemu_gluster_reopen_commit, 1590 .bdrv_reopen_abort = qemu_gluster_reopen_abort, 1591 .bdrv_close = qemu_gluster_close, 1592 .bdrv_co_create = qemu_gluster_co_create, 1593 .bdrv_co_create_opts = qemu_gluster_co_create_opts, 1594 .bdrv_co_getlength = qemu_gluster_co_getlength, 1595 .bdrv_co_get_allocated_file_size = qemu_gluster_co_get_allocated_file_size, 1596 .bdrv_co_truncate = qemu_gluster_co_truncate, 1597 .bdrv_co_readv = qemu_gluster_co_readv, 1598 .bdrv_co_writev = qemu_gluster_co_writev, 1599 .bdrv_co_flush_to_disk = qemu_gluster_co_flush_to_disk, 1600 #ifdef CONFIG_GLUSTERFS_DISCARD 1601 .bdrv_co_pdiscard = qemu_gluster_co_pdiscard, 1602 #endif 1603 #ifdef CONFIG_GLUSTERFS_ZEROFILL 1604 .bdrv_co_pwrite_zeroes = qemu_gluster_co_pwrite_zeroes, 1605 #endif 1606 .bdrv_co_block_status = qemu_gluster_co_block_status, 1607 .bdrv_refresh_limits = qemu_gluster_refresh_limits, 1608 .create_opts = &qemu_gluster_create_opts, 1609 .strong_runtime_opts = gluster_strong_open_opts, 1610 }; 1611 1612 static BlockDriver bdrv_gluster_unix = { 1613 .format_name = "gluster", 1614 .protocol_name = "gluster+unix", 1615 .instance_size = sizeof(BDRVGlusterState), 1616 .bdrv_file_open = qemu_gluster_open, 1617 .bdrv_reopen_prepare = qemu_gluster_reopen_prepare, 1618 .bdrv_reopen_commit = qemu_gluster_reopen_commit, 1619 .bdrv_reopen_abort = qemu_gluster_reopen_abort, 1620 .bdrv_close = qemu_gluster_close, 1621 .bdrv_co_create = qemu_gluster_co_create, 1622 .bdrv_co_create_opts = qemu_gluster_co_create_opts, 1623 .bdrv_co_getlength = qemu_gluster_co_getlength, 1624 .bdrv_co_get_allocated_file_size = qemu_gluster_co_get_allocated_file_size, 1625 .bdrv_co_truncate = qemu_gluster_co_truncate, 1626 .bdrv_co_readv = qemu_gluster_co_readv, 1627 .bdrv_co_writev = qemu_gluster_co_writev, 1628 .bdrv_co_flush_to_disk = qemu_gluster_co_flush_to_disk, 1629 #ifdef CONFIG_GLUSTERFS_DISCARD 1630 .bdrv_co_pdiscard = qemu_gluster_co_pdiscard, 1631 #endif 1632 #ifdef CONFIG_GLUSTERFS_ZEROFILL 1633 .bdrv_co_pwrite_zeroes = qemu_gluster_co_pwrite_zeroes, 1634 #endif 1635 .bdrv_co_block_status = qemu_gluster_co_block_status, 1636 .bdrv_refresh_limits = qemu_gluster_refresh_limits, 1637 .create_opts = &qemu_gluster_create_opts, 1638 .strong_runtime_opts = gluster_strong_open_opts, 1639 }; 1640 1641 /* rdma is deprecated (actually never supported for volfile fetch). 1642 * Let's maintain it for the protocol compatibility, to make sure things 1643 * won't break immediately. For now, gluster+rdma will fall back to gluster+tcp 1644 * protocol with a warning. 1645 * TODO: remove gluster+rdma interface support 1646 */ 1647 static BlockDriver bdrv_gluster_rdma = { 1648 .format_name = "gluster", 1649 .protocol_name = "gluster+rdma", 1650 .instance_size = sizeof(BDRVGlusterState), 1651 .bdrv_file_open = qemu_gluster_open, 1652 .bdrv_reopen_prepare = qemu_gluster_reopen_prepare, 1653 .bdrv_reopen_commit = qemu_gluster_reopen_commit, 1654 .bdrv_reopen_abort = qemu_gluster_reopen_abort, 1655 .bdrv_close = qemu_gluster_close, 1656 .bdrv_co_create = qemu_gluster_co_create, 1657 .bdrv_co_create_opts = qemu_gluster_co_create_opts, 1658 .bdrv_co_getlength = qemu_gluster_co_getlength, 1659 .bdrv_co_get_allocated_file_size = qemu_gluster_co_get_allocated_file_size, 1660 .bdrv_co_truncate = qemu_gluster_co_truncate, 1661 .bdrv_co_readv = qemu_gluster_co_readv, 1662 .bdrv_co_writev = qemu_gluster_co_writev, 1663 .bdrv_co_flush_to_disk = qemu_gluster_co_flush_to_disk, 1664 #ifdef CONFIG_GLUSTERFS_DISCARD 1665 .bdrv_co_pdiscard = qemu_gluster_co_pdiscard, 1666 #endif 1667 #ifdef CONFIG_GLUSTERFS_ZEROFILL 1668 .bdrv_co_pwrite_zeroes = qemu_gluster_co_pwrite_zeroes, 1669 #endif 1670 .bdrv_co_block_status = qemu_gluster_co_block_status, 1671 .bdrv_refresh_limits = qemu_gluster_refresh_limits, 1672 .create_opts = &qemu_gluster_create_opts, 1673 .strong_runtime_opts = gluster_strong_open_opts, 1674 }; 1675 1676 static void bdrv_gluster_init(void) 1677 { 1678 bdrv_register(&bdrv_gluster_rdma); 1679 bdrv_register(&bdrv_gluster_unix); 1680 bdrv_register(&bdrv_gluster_tcp); 1681 bdrv_register(&bdrv_gluster); 1682 } 1683 1684 block_init(bdrv_gluster_init); 1685