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 <glusterfs/api/glfs.h> 13 #include "block/block_int.h" 14 #include "qapi/error.h" 15 #include "qapi/qmp/qdict.h" 16 #include "qapi/qmp/qerror.h" 17 #include "qemu/uri.h" 18 #include "qemu/error-report.h" 19 #include "qemu/option.h" 20 #include "qemu/cutils.h" 21 22 #define GLUSTER_OPT_FILENAME "filename" 23 #define GLUSTER_OPT_VOLUME "volume" 24 #define GLUSTER_OPT_PATH "path" 25 #define GLUSTER_OPT_TYPE "type" 26 #define GLUSTER_OPT_SERVER_PATTERN "server." 27 #define GLUSTER_OPT_HOST "host" 28 #define GLUSTER_OPT_PORT "port" 29 #define GLUSTER_OPT_TO "to" 30 #define GLUSTER_OPT_IPV4 "ipv4" 31 #define GLUSTER_OPT_IPV6 "ipv6" 32 #define GLUSTER_OPT_SOCKET "socket" 33 #define GLUSTER_OPT_DEBUG "debug" 34 #define GLUSTER_DEFAULT_PORT 24007 35 #define GLUSTER_DEBUG_DEFAULT 4 36 #define GLUSTER_DEBUG_MAX 9 37 #define GLUSTER_OPT_LOGFILE "logfile" 38 #define GLUSTER_LOGFILE_DEFAULT "-" /* handled in libgfapi as /dev/stderr */ 39 40 #define GERR_INDEX_HINT "hint: check in 'server' array index '%d'\n" 41 42 typedef struct GlusterAIOCB { 43 int64_t size; 44 int ret; 45 Coroutine *coroutine; 46 AioContext *aio_context; 47 } GlusterAIOCB; 48 49 typedef struct BDRVGlusterState { 50 struct glfs *glfs; 51 struct glfs_fd *fd; 52 char *logfile; 53 bool supports_seek_data; 54 int debug; 55 } BDRVGlusterState; 56 57 typedef struct BDRVGlusterReopenState { 58 struct glfs *glfs; 59 struct glfs_fd *fd; 60 } BDRVGlusterReopenState; 61 62 63 typedef struct GlfsPreopened { 64 char *volume; 65 glfs_t *fs; 66 int ref; 67 } GlfsPreopened; 68 69 typedef struct ListElement { 70 QLIST_ENTRY(ListElement) list; 71 GlfsPreopened saved; 72 } ListElement; 73 74 static QLIST_HEAD(glfs_list, ListElement) glfs_list; 75 76 static QemuOptsList qemu_gluster_create_opts = { 77 .name = "qemu-gluster-create-opts", 78 .head = QTAILQ_HEAD_INITIALIZER(qemu_gluster_create_opts.head), 79 .desc = { 80 { 81 .name = BLOCK_OPT_SIZE, 82 .type = QEMU_OPT_SIZE, 83 .help = "Virtual disk size" 84 }, 85 { 86 .name = BLOCK_OPT_PREALLOC, 87 .type = QEMU_OPT_STRING, 88 .help = "Preallocation mode (allowed values: off, full)" 89 }, 90 { 91 .name = GLUSTER_OPT_DEBUG, 92 .type = QEMU_OPT_NUMBER, 93 .help = "Gluster log level, valid range is 0-9", 94 }, 95 { 96 .name = GLUSTER_OPT_LOGFILE, 97 .type = QEMU_OPT_STRING, 98 .help = "Logfile path of libgfapi", 99 }, 100 { /* end of list */ } 101 } 102 }; 103 104 static QemuOptsList runtime_opts = { 105 .name = "gluster", 106 .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head), 107 .desc = { 108 { 109 .name = GLUSTER_OPT_FILENAME, 110 .type = QEMU_OPT_STRING, 111 .help = "URL to the gluster image", 112 }, 113 { 114 .name = GLUSTER_OPT_DEBUG, 115 .type = QEMU_OPT_NUMBER, 116 .help = "Gluster log level, valid range is 0-9", 117 }, 118 { 119 .name = GLUSTER_OPT_LOGFILE, 120 .type = QEMU_OPT_STRING, 121 .help = "Logfile path of libgfapi", 122 }, 123 { /* end of list */ } 124 }, 125 }; 126 127 static QemuOptsList runtime_json_opts = { 128 .name = "gluster_json", 129 .head = QTAILQ_HEAD_INITIALIZER(runtime_json_opts.head), 130 .desc = { 131 { 132 .name = GLUSTER_OPT_VOLUME, 133 .type = QEMU_OPT_STRING, 134 .help = "name of gluster volume where VM image resides", 135 }, 136 { 137 .name = GLUSTER_OPT_PATH, 138 .type = QEMU_OPT_STRING, 139 .help = "absolute path to image file in gluster volume", 140 }, 141 { 142 .name = GLUSTER_OPT_DEBUG, 143 .type = QEMU_OPT_NUMBER, 144 .help = "Gluster log level, valid range is 0-9", 145 }, 146 { /* end of list */ } 147 }, 148 }; 149 150 static QemuOptsList runtime_type_opts = { 151 .name = "gluster_type", 152 .head = QTAILQ_HEAD_INITIALIZER(runtime_type_opts.head), 153 .desc = { 154 { 155 .name = GLUSTER_OPT_TYPE, 156 .type = QEMU_OPT_STRING, 157 .help = "inet|unix", 158 }, 159 { /* end of list */ } 160 }, 161 }; 162 163 static QemuOptsList runtime_unix_opts = { 164 .name = "gluster_unix", 165 .head = QTAILQ_HEAD_INITIALIZER(runtime_unix_opts.head), 166 .desc = { 167 { 168 .name = GLUSTER_OPT_SOCKET, 169 .type = QEMU_OPT_STRING, 170 .help = "socket file path)", 171 }, 172 { /* end of list */ } 173 }, 174 }; 175 176 static QemuOptsList runtime_inet_opts = { 177 .name = "gluster_inet", 178 .head = QTAILQ_HEAD_INITIALIZER(runtime_inet_opts.head), 179 .desc = { 180 { 181 .name = GLUSTER_OPT_TYPE, 182 .type = QEMU_OPT_STRING, 183 .help = "inet|unix", 184 }, 185 { 186 .name = GLUSTER_OPT_HOST, 187 .type = QEMU_OPT_STRING, 188 .help = "host address (hostname/ipv4/ipv6 addresses)", 189 }, 190 { 191 .name = GLUSTER_OPT_PORT, 192 .type = QEMU_OPT_STRING, 193 .help = "port number on which glusterd is listening (default 24007)", 194 }, 195 { 196 .name = "to", 197 .type = QEMU_OPT_NUMBER, 198 .help = "max port number, not supported by gluster", 199 }, 200 { 201 .name = "ipv4", 202 .type = QEMU_OPT_BOOL, 203 .help = "ipv4 bool value, not supported by gluster", 204 }, 205 { 206 .name = "ipv6", 207 .type = QEMU_OPT_BOOL, 208 .help = "ipv6 bool value, not supported by gluster", 209 }, 210 { /* end of list */ } 211 }, 212 }; 213 214 static void glfs_set_preopened(const char *volume, glfs_t *fs) 215 { 216 ListElement *entry = NULL; 217 218 entry = g_new(ListElement, 1); 219 220 entry->saved.volume = g_strdup(volume); 221 222 entry->saved.fs = fs; 223 entry->saved.ref = 1; 224 225 QLIST_INSERT_HEAD(&glfs_list, entry, list); 226 } 227 228 static glfs_t *glfs_find_preopened(const char *volume) 229 { 230 ListElement *entry = NULL; 231 232 QLIST_FOREACH(entry, &glfs_list, list) { 233 if (strcmp(entry->saved.volume, volume) == 0) { 234 entry->saved.ref++; 235 return entry->saved.fs; 236 } 237 } 238 239 return NULL; 240 } 241 242 static void glfs_clear_preopened(glfs_t *fs) 243 { 244 ListElement *entry = NULL; 245 ListElement *next; 246 247 if (fs == NULL) { 248 return; 249 } 250 251 QLIST_FOREACH_SAFE(entry, &glfs_list, list, next) { 252 if (entry->saved.fs == fs) { 253 if (--entry->saved.ref) { 254 return; 255 } 256 257 QLIST_REMOVE(entry, list); 258 259 glfs_fini(entry->saved.fs); 260 g_free(entry->saved.volume); 261 g_free(entry); 262 } 263 } 264 } 265 266 static int parse_volume_options(BlockdevOptionsGluster *gconf, char *path) 267 { 268 char *p, *q; 269 270 if (!path) { 271 return -EINVAL; 272 } 273 274 /* volume */ 275 p = q = path + strspn(path, "/"); 276 p += strcspn(p, "/"); 277 if (*p == '\0') { 278 return -EINVAL; 279 } 280 gconf->volume = g_strndup(q, p - q); 281 282 /* path */ 283 p += strspn(p, "/"); 284 if (*p == '\0') { 285 return -EINVAL; 286 } 287 gconf->path = g_strdup(p); 288 return 0; 289 } 290 291 /* 292 * file=gluster[+transport]://[host[:port]]/volume/path[?socket=...] 293 * 294 * 'gluster' is the protocol. 295 * 296 * 'transport' specifies the transport type used to connect to gluster 297 * management daemon (glusterd). Valid transport types are 298 * tcp or unix. If a transport type isn't specified, then tcp type is assumed. 299 * 300 * 'host' specifies the host where the volume file specification for 301 * the given volume resides. This can be either hostname or ipv4 address. 302 * If transport type is 'unix', then 'host' field should not be specified. 303 * The 'socket' field needs to be populated with the path to unix domain 304 * socket. 305 * 306 * 'port' is the port number on which glusterd is listening. This is optional 307 * and if not specified, QEMU will send 0 which will make gluster to use the 308 * default port. If the transport type is unix, then 'port' should not be 309 * specified. 310 * 311 * 'volume' is the name of the gluster volume which contains the VM image. 312 * 313 * 'path' is the path to the actual VM image that resides on gluster volume. 314 * 315 * Examples: 316 * 317 * file=gluster://1.2.3.4/testvol/a.img 318 * file=gluster+tcp://1.2.3.4/testvol/a.img 319 * file=gluster+tcp://1.2.3.4:24007/testvol/dir/a.img 320 * file=gluster+tcp://host.domain.com:24007/testvol/dir/a.img 321 * file=gluster+unix:///testvol/dir/a.img?socket=/tmp/glusterd.socket 322 */ 323 static int qemu_gluster_parse_uri(BlockdevOptionsGluster *gconf, 324 const char *filename) 325 { 326 SocketAddress *gsconf; 327 URI *uri; 328 QueryParams *qp = NULL; 329 bool is_unix = false; 330 int ret = 0; 331 332 uri = uri_parse(filename); 333 if (!uri) { 334 return -EINVAL; 335 } 336 337 gconf->server = g_new0(SocketAddressList, 1); 338 gconf->server->value = gsconf = g_new0(SocketAddress, 1); 339 340 /* transport */ 341 if (!uri->scheme || !strcmp(uri->scheme, "gluster")) { 342 gsconf->type = SOCKET_ADDRESS_TYPE_INET; 343 } else if (!strcmp(uri->scheme, "gluster+tcp")) { 344 gsconf->type = SOCKET_ADDRESS_TYPE_INET; 345 } else if (!strcmp(uri->scheme, "gluster+unix")) { 346 gsconf->type = SOCKET_ADDRESS_TYPE_UNIX; 347 is_unix = true; 348 } else if (!strcmp(uri->scheme, "gluster+rdma")) { 349 gsconf->type = SOCKET_ADDRESS_TYPE_INET; 350 warn_report("rdma feature is not supported, falling back to tcp"); 351 } else { 352 ret = -EINVAL; 353 goto out; 354 } 355 356 ret = parse_volume_options(gconf, uri->path); 357 if (ret < 0) { 358 goto out; 359 } 360 361 qp = query_params_parse(uri->query); 362 if (qp->n > 1 || (is_unix && !qp->n) || (!is_unix && qp->n)) { 363 ret = -EINVAL; 364 goto out; 365 } 366 367 if (is_unix) { 368 if (uri->server || uri->port) { 369 ret = -EINVAL; 370 goto out; 371 } 372 if (strcmp(qp->p[0].name, "socket")) { 373 ret = -EINVAL; 374 goto out; 375 } 376 gsconf->u.q_unix.path = g_strdup(qp->p[0].value); 377 } else { 378 gsconf->u.inet.host = g_strdup(uri->server ? uri->server : "localhost"); 379 if (uri->port) { 380 gsconf->u.inet.port = g_strdup_printf("%d", uri->port); 381 } else { 382 gsconf->u.inet.port = g_strdup_printf("%d", GLUSTER_DEFAULT_PORT); 383 } 384 } 385 386 out: 387 if (qp) { 388 query_params_free(qp); 389 } 390 uri_free(uri); 391 return ret; 392 } 393 394 static struct glfs *qemu_gluster_glfs_init(BlockdevOptionsGluster *gconf, 395 Error **errp) 396 { 397 struct glfs *glfs; 398 int ret; 399 int old_errno; 400 SocketAddressList *server; 401 unsigned long long port; 402 403 glfs = glfs_find_preopened(gconf->volume); 404 if (glfs) { 405 return glfs; 406 } 407 408 glfs = glfs_new(gconf->volume); 409 if (!glfs) { 410 goto out; 411 } 412 413 glfs_set_preopened(gconf->volume, glfs); 414 415 for (server = gconf->server; server; server = server->next) { 416 switch (server->value->type) { 417 case SOCKET_ADDRESS_TYPE_UNIX: 418 ret = glfs_set_volfile_server(glfs, "unix", 419 server->value->u.q_unix.path, 0); 420 break; 421 case SOCKET_ADDRESS_TYPE_INET: 422 if (parse_uint_full(server->value->u.inet.port, &port, 10) < 0 || 423 port > 65535) { 424 error_setg(errp, "'%s' is not a valid port number", 425 server->value->u.inet.port); 426 errno = EINVAL; 427 goto out; 428 } 429 ret = glfs_set_volfile_server(glfs, "tcp", 430 server->value->u.inet.host, 431 (int)port); 432 break; 433 case SOCKET_ADDRESS_TYPE_VSOCK: 434 case SOCKET_ADDRESS_TYPE_FD: 435 default: 436 abort(); 437 } 438 439 if (ret < 0) { 440 goto out; 441 } 442 } 443 444 ret = glfs_set_logging(glfs, gconf->logfile, gconf->debug); 445 if (ret < 0) { 446 goto out; 447 } 448 449 ret = glfs_init(glfs); 450 if (ret) { 451 error_setg(errp, "Gluster connection for volume %s, path %s failed" 452 " to connect", gconf->volume, gconf->path); 453 for (server = gconf->server; server; server = server->next) { 454 if (server->value->type == SOCKET_ADDRESS_TYPE_UNIX) { 455 error_append_hint(errp, "hint: failed on socket %s ", 456 server->value->u.q_unix.path); 457 } else { 458 error_append_hint(errp, "hint: failed on host %s and port %s ", 459 server->value->u.inet.host, 460 server->value->u.inet.port); 461 } 462 } 463 464 error_append_hint(errp, "Please refer to gluster logs for more info\n"); 465 466 /* glfs_init sometimes doesn't set errno although docs suggest that */ 467 if (errno == 0) { 468 errno = EINVAL; 469 } 470 471 goto out; 472 } 473 return glfs; 474 475 out: 476 if (glfs) { 477 old_errno = errno; 478 glfs_clear_preopened(glfs); 479 errno = old_errno; 480 } 481 return NULL; 482 } 483 484 /* 485 * Convert the json formatted command line into qapi. 486 */ 487 static int qemu_gluster_parse_json(BlockdevOptionsGluster *gconf, 488 QDict *options, Error **errp) 489 { 490 QemuOpts *opts; 491 SocketAddress *gsconf = NULL; 492 SocketAddressList *curr = NULL; 493 QDict *backing_options = NULL; 494 Error *local_err = NULL; 495 char *str = NULL; 496 const char *ptr; 497 int i, type, num_servers; 498 499 /* create opts info from runtime_json_opts list */ 500 opts = qemu_opts_create(&runtime_json_opts, NULL, 0, &error_abort); 501 qemu_opts_absorb_qdict(opts, options, &local_err); 502 if (local_err) { 503 goto out; 504 } 505 506 num_servers = qdict_array_entries(options, GLUSTER_OPT_SERVER_PATTERN); 507 if (num_servers < 1) { 508 error_setg(&local_err, QERR_MISSING_PARAMETER, "server"); 509 goto out; 510 } 511 512 ptr = qemu_opt_get(opts, GLUSTER_OPT_VOLUME); 513 if (!ptr) { 514 error_setg(&local_err, QERR_MISSING_PARAMETER, GLUSTER_OPT_VOLUME); 515 goto out; 516 } 517 gconf->volume = g_strdup(ptr); 518 519 ptr = qemu_opt_get(opts, GLUSTER_OPT_PATH); 520 if (!ptr) { 521 error_setg(&local_err, QERR_MISSING_PARAMETER, GLUSTER_OPT_PATH); 522 goto out; 523 } 524 gconf->path = g_strdup(ptr); 525 qemu_opts_del(opts); 526 527 for (i = 0; i < num_servers; i++) { 528 str = g_strdup_printf(GLUSTER_OPT_SERVER_PATTERN"%d.", i); 529 qdict_extract_subqdict(options, &backing_options, str); 530 531 /* create opts info from runtime_type_opts list */ 532 opts = qemu_opts_create(&runtime_type_opts, NULL, 0, &error_abort); 533 qemu_opts_absorb_qdict(opts, backing_options, &local_err); 534 if (local_err) { 535 goto out; 536 } 537 538 ptr = qemu_opt_get(opts, GLUSTER_OPT_TYPE); 539 if (!ptr) { 540 error_setg(&local_err, QERR_MISSING_PARAMETER, GLUSTER_OPT_TYPE); 541 error_append_hint(&local_err, GERR_INDEX_HINT, i); 542 goto out; 543 544 } 545 gsconf = g_new0(SocketAddress, 1); 546 if (!strcmp(ptr, "tcp")) { 547 ptr = "inet"; /* accept legacy "tcp" */ 548 } 549 type = qapi_enum_parse(&SocketAddressType_lookup, ptr, -1, NULL); 550 if (type != SOCKET_ADDRESS_TYPE_INET 551 && type != SOCKET_ADDRESS_TYPE_UNIX) { 552 error_setg(&local_err, 553 "Parameter '%s' may be 'inet' or 'unix'", 554 GLUSTER_OPT_TYPE); 555 error_append_hint(&local_err, GERR_INDEX_HINT, i); 556 goto out; 557 } 558 gsconf->type = type; 559 qemu_opts_del(opts); 560 561 if (gsconf->type == SOCKET_ADDRESS_TYPE_INET) { 562 /* create opts info from runtime_inet_opts list */ 563 opts = qemu_opts_create(&runtime_inet_opts, NULL, 0, &error_abort); 564 qemu_opts_absorb_qdict(opts, backing_options, &local_err); 565 if (local_err) { 566 goto out; 567 } 568 569 ptr = qemu_opt_get(opts, GLUSTER_OPT_HOST); 570 if (!ptr) { 571 error_setg(&local_err, QERR_MISSING_PARAMETER, 572 GLUSTER_OPT_HOST); 573 error_append_hint(&local_err, GERR_INDEX_HINT, i); 574 goto out; 575 } 576 gsconf->u.inet.host = g_strdup(ptr); 577 ptr = qemu_opt_get(opts, GLUSTER_OPT_PORT); 578 if (!ptr) { 579 error_setg(&local_err, QERR_MISSING_PARAMETER, 580 GLUSTER_OPT_PORT); 581 error_append_hint(&local_err, GERR_INDEX_HINT, i); 582 goto out; 583 } 584 gsconf->u.inet.port = g_strdup(ptr); 585 586 /* defend for unsupported fields in InetSocketAddress, 587 * i.e. @ipv4, @ipv6 and @to 588 */ 589 ptr = qemu_opt_get(opts, GLUSTER_OPT_TO); 590 if (ptr) { 591 gsconf->u.inet.has_to = true; 592 } 593 ptr = qemu_opt_get(opts, GLUSTER_OPT_IPV4); 594 if (ptr) { 595 gsconf->u.inet.has_ipv4 = true; 596 } 597 ptr = qemu_opt_get(opts, GLUSTER_OPT_IPV6); 598 if (ptr) { 599 gsconf->u.inet.has_ipv6 = true; 600 } 601 if (gsconf->u.inet.has_to) { 602 error_setg(&local_err, "Parameter 'to' not supported"); 603 goto out; 604 } 605 if (gsconf->u.inet.has_ipv4 || gsconf->u.inet.has_ipv6) { 606 error_setg(&local_err, "Parameters 'ipv4/ipv6' not supported"); 607 goto out; 608 } 609 qemu_opts_del(opts); 610 } else { 611 /* create opts info from runtime_unix_opts list */ 612 opts = qemu_opts_create(&runtime_unix_opts, NULL, 0, &error_abort); 613 qemu_opts_absorb_qdict(opts, backing_options, &local_err); 614 if (local_err) { 615 goto out; 616 } 617 618 ptr = qemu_opt_get(opts, GLUSTER_OPT_SOCKET); 619 if (!ptr) { 620 error_setg(&local_err, QERR_MISSING_PARAMETER, 621 GLUSTER_OPT_SOCKET); 622 error_append_hint(&local_err, GERR_INDEX_HINT, i); 623 goto out; 624 } 625 gsconf->u.q_unix.path = g_strdup(ptr); 626 qemu_opts_del(opts); 627 } 628 629 if (gconf->server == NULL) { 630 gconf->server = g_new0(SocketAddressList, 1); 631 gconf->server->value = gsconf; 632 curr = gconf->server; 633 } else { 634 curr->next = g_new0(SocketAddressList, 1); 635 curr->next->value = gsconf; 636 curr = curr->next; 637 } 638 gsconf = NULL; 639 640 QDECREF(backing_options); 641 backing_options = NULL; 642 g_free(str); 643 str = NULL; 644 } 645 646 return 0; 647 648 out: 649 error_propagate(errp, local_err); 650 qapi_free_SocketAddress(gsconf); 651 qemu_opts_del(opts); 652 g_free(str); 653 QDECREF(backing_options); 654 errno = EINVAL; 655 return -errno; 656 } 657 658 /* Converts options given in @filename and the @options QDict into the QAPI 659 * object @gconf. */ 660 static int qemu_gluster_parse(BlockdevOptionsGluster *gconf, 661 const char *filename, 662 QDict *options, Error **errp) 663 { 664 int ret; 665 if (filename) { 666 ret = qemu_gluster_parse_uri(gconf, filename); 667 if (ret < 0) { 668 error_setg(errp, "invalid URI %s", filename); 669 error_append_hint(errp, "Usage: file=gluster[+transport]://" 670 "[host[:port]]volume/path[?socket=...]" 671 "[,file.debug=N]" 672 "[,file.logfile=/path/filename.log]\n"); 673 return ret; 674 } 675 } else { 676 ret = qemu_gluster_parse_json(gconf, options, errp); 677 if (ret < 0) { 678 error_append_hint(errp, "Usage: " 679 "-drive driver=qcow2,file.driver=gluster," 680 "file.volume=testvol,file.path=/path/a.qcow2" 681 "[,file.debug=9]" 682 "[,file.logfile=/path/filename.log]," 683 "file.server.0.type=inet," 684 "file.server.0.host=1.2.3.4," 685 "file.server.0.port=24007," 686 "file.server.1.transport=unix," 687 "file.server.1.socket=/var/run/glusterd.socket ..." 688 "\n"); 689 return ret; 690 } 691 } 692 693 return 0; 694 } 695 696 static struct glfs *qemu_gluster_init(BlockdevOptionsGluster *gconf, 697 const char *filename, 698 QDict *options, Error **errp) 699 { 700 int ret; 701 702 ret = qemu_gluster_parse(gconf, filename, options, errp); 703 if (ret < 0) { 704 errno = -ret; 705 return NULL; 706 } 707 708 return qemu_gluster_glfs_init(gconf, errp); 709 } 710 711 /* 712 * AIO callback routine called from GlusterFS thread. 713 */ 714 static void gluster_finish_aiocb(struct glfs_fd *fd, ssize_t ret, void *arg) 715 { 716 GlusterAIOCB *acb = (GlusterAIOCB *)arg; 717 718 if (!ret || ret == acb->size) { 719 acb->ret = 0; /* Success */ 720 } else if (ret < 0) { 721 acb->ret = -errno; /* Read/Write failed */ 722 } else { 723 acb->ret = -EIO; /* Partial read/write - fail it */ 724 } 725 726 aio_co_schedule(acb->aio_context, acb->coroutine); 727 } 728 729 static void qemu_gluster_parse_flags(int bdrv_flags, int *open_flags) 730 { 731 assert(open_flags != NULL); 732 733 *open_flags |= O_BINARY; 734 735 if (bdrv_flags & BDRV_O_RDWR) { 736 *open_flags |= O_RDWR; 737 } else { 738 *open_flags |= O_RDONLY; 739 } 740 741 if ((bdrv_flags & BDRV_O_NOCACHE)) { 742 *open_flags |= O_DIRECT; 743 } 744 } 745 746 /* 747 * Do SEEK_DATA/HOLE to detect if it is functional. Older broken versions of 748 * gfapi incorrectly return the current offset when SEEK_DATA/HOLE is used. 749 * - Corrected versions return -1 and set errno to EINVAL. 750 * - Versions that support SEEK_DATA/HOLE correctly, will return -1 and set 751 * errno to ENXIO when SEEK_DATA is called with a position of EOF. 752 */ 753 static bool qemu_gluster_test_seek(struct glfs_fd *fd) 754 { 755 off_t ret = 0; 756 757 #if defined SEEK_HOLE && defined SEEK_DATA 758 off_t eof; 759 760 eof = glfs_lseek(fd, 0, SEEK_END); 761 if (eof < 0) { 762 /* this should never occur */ 763 return false; 764 } 765 766 /* this should always fail with ENXIO if SEEK_DATA is supported */ 767 ret = glfs_lseek(fd, eof, SEEK_DATA); 768 #endif 769 770 return (ret < 0) && (errno == ENXIO); 771 } 772 773 static int qemu_gluster_open(BlockDriverState *bs, QDict *options, 774 int bdrv_flags, Error **errp) 775 { 776 BDRVGlusterState *s = bs->opaque; 777 int open_flags = 0; 778 int ret = 0; 779 BlockdevOptionsGluster *gconf = NULL; 780 QemuOpts *opts; 781 Error *local_err = NULL; 782 const char *filename, *logfile; 783 784 opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort); 785 qemu_opts_absorb_qdict(opts, options, &local_err); 786 if (local_err) { 787 error_propagate(errp, local_err); 788 ret = -EINVAL; 789 goto out; 790 } 791 792 filename = qemu_opt_get(opts, GLUSTER_OPT_FILENAME); 793 794 s->debug = qemu_opt_get_number(opts, GLUSTER_OPT_DEBUG, 795 GLUSTER_DEBUG_DEFAULT); 796 if (s->debug < 0) { 797 s->debug = 0; 798 } else if (s->debug > GLUSTER_DEBUG_MAX) { 799 s->debug = GLUSTER_DEBUG_MAX; 800 } 801 802 gconf = g_new0(BlockdevOptionsGluster, 1); 803 gconf->debug = s->debug; 804 gconf->has_debug = true; 805 806 logfile = qemu_opt_get(opts, GLUSTER_OPT_LOGFILE); 807 s->logfile = g_strdup(logfile ? logfile : GLUSTER_LOGFILE_DEFAULT); 808 809 gconf->logfile = g_strdup(s->logfile); 810 gconf->has_logfile = true; 811 812 s->glfs = qemu_gluster_init(gconf, filename, options, errp); 813 if (!s->glfs) { 814 ret = -errno; 815 goto out; 816 } 817 818 #ifdef CONFIG_GLUSTERFS_XLATOR_OPT 819 /* Without this, if fsync fails for a recoverable reason (for instance, 820 * ENOSPC), gluster will dump its cache, preventing retries. This means 821 * almost certain data loss. Not all gluster versions support the 822 * 'resync-failed-syncs-after-fsync' key value, but there is no way to 823 * discover during runtime if it is supported (this api returns success for 824 * unknown key/value pairs) */ 825 ret = glfs_set_xlator_option(s->glfs, "*-write-behind", 826 "resync-failed-syncs-after-fsync", 827 "on"); 828 if (ret < 0) { 829 error_setg_errno(errp, errno, "Unable to set xlator key/value pair"); 830 ret = -errno; 831 goto out; 832 } 833 #endif 834 835 qemu_gluster_parse_flags(bdrv_flags, &open_flags); 836 837 s->fd = glfs_open(s->glfs, gconf->path, open_flags); 838 if (!s->fd) { 839 ret = -errno; 840 } 841 842 s->supports_seek_data = qemu_gluster_test_seek(s->fd); 843 844 out: 845 qemu_opts_del(opts); 846 qapi_free_BlockdevOptionsGluster(gconf); 847 if (!ret) { 848 return ret; 849 } 850 g_free(s->logfile); 851 if (s->fd) { 852 glfs_close(s->fd); 853 } 854 855 glfs_clear_preopened(s->glfs); 856 857 return ret; 858 } 859 860 static int qemu_gluster_reopen_prepare(BDRVReopenState *state, 861 BlockReopenQueue *queue, Error **errp) 862 { 863 int ret = 0; 864 BDRVGlusterState *s; 865 BDRVGlusterReopenState *reop_s; 866 BlockdevOptionsGluster *gconf; 867 int open_flags = 0; 868 869 assert(state != NULL); 870 assert(state->bs != NULL); 871 872 s = state->bs->opaque; 873 874 state->opaque = g_new0(BDRVGlusterReopenState, 1); 875 reop_s = state->opaque; 876 877 qemu_gluster_parse_flags(state->flags, &open_flags); 878 879 gconf = g_new0(BlockdevOptionsGluster, 1); 880 gconf->debug = s->debug; 881 gconf->has_debug = true; 882 gconf->logfile = g_strdup(s->logfile); 883 gconf->has_logfile = true; 884 reop_s->glfs = qemu_gluster_init(gconf, state->bs->filename, NULL, errp); 885 if (reop_s->glfs == NULL) { 886 ret = -errno; 887 goto exit; 888 } 889 890 #ifdef CONFIG_GLUSTERFS_XLATOR_OPT 891 ret = glfs_set_xlator_option(reop_s->glfs, "*-write-behind", 892 "resync-failed-syncs-after-fsync", "on"); 893 if (ret < 0) { 894 error_setg_errno(errp, errno, "Unable to set xlator key/value pair"); 895 ret = -errno; 896 goto exit; 897 } 898 #endif 899 900 reop_s->fd = glfs_open(reop_s->glfs, gconf->path, open_flags); 901 if (reop_s->fd == NULL) { 902 /* reops->glfs will be cleaned up in _abort */ 903 ret = -errno; 904 goto exit; 905 } 906 907 exit: 908 /* state->opaque will be freed in either the _abort or _commit */ 909 qapi_free_BlockdevOptionsGluster(gconf); 910 return ret; 911 } 912 913 static void qemu_gluster_reopen_commit(BDRVReopenState *state) 914 { 915 BDRVGlusterReopenState *reop_s = state->opaque; 916 BDRVGlusterState *s = state->bs->opaque; 917 918 919 /* close the old */ 920 if (s->fd) { 921 glfs_close(s->fd); 922 } 923 924 glfs_clear_preopened(s->glfs); 925 926 /* use the newly opened image / connection */ 927 s->fd = reop_s->fd; 928 s->glfs = reop_s->glfs; 929 930 g_free(state->opaque); 931 state->opaque = NULL; 932 933 return; 934 } 935 936 937 static void qemu_gluster_reopen_abort(BDRVReopenState *state) 938 { 939 BDRVGlusterReopenState *reop_s = state->opaque; 940 941 if (reop_s == NULL) { 942 return; 943 } 944 945 if (reop_s->fd) { 946 glfs_close(reop_s->fd); 947 } 948 949 glfs_clear_preopened(reop_s->glfs); 950 951 g_free(state->opaque); 952 state->opaque = NULL; 953 954 return; 955 } 956 957 #ifdef CONFIG_GLUSTERFS_ZEROFILL 958 static coroutine_fn int qemu_gluster_co_pwrite_zeroes(BlockDriverState *bs, 959 int64_t offset, 960 int size, 961 BdrvRequestFlags flags) 962 { 963 int ret; 964 GlusterAIOCB acb; 965 BDRVGlusterState *s = bs->opaque; 966 967 acb.size = size; 968 acb.ret = 0; 969 acb.coroutine = qemu_coroutine_self(); 970 acb.aio_context = bdrv_get_aio_context(bs); 971 972 ret = glfs_zerofill_async(s->fd, offset, size, gluster_finish_aiocb, &acb); 973 if (ret < 0) { 974 return -errno; 975 } 976 977 qemu_coroutine_yield(); 978 return acb.ret; 979 } 980 #endif 981 982 static int qemu_gluster_do_truncate(struct glfs_fd *fd, int64_t offset, 983 PreallocMode prealloc, Error **errp) 984 { 985 int64_t current_length; 986 987 current_length = glfs_lseek(fd, 0, SEEK_END); 988 if (current_length < 0) { 989 error_setg_errno(errp, errno, "Failed to determine current size"); 990 return -errno; 991 } 992 993 if (current_length > offset && prealloc != PREALLOC_MODE_OFF) { 994 error_setg(errp, "Cannot use preallocation for shrinking files"); 995 return -ENOTSUP; 996 } 997 998 if (current_length == offset) { 999 return 0; 1000 } 1001 1002 switch (prealloc) { 1003 #ifdef CONFIG_GLUSTERFS_FALLOCATE 1004 case PREALLOC_MODE_FALLOC: 1005 if (glfs_fallocate(fd, 0, current_length, offset - current_length)) { 1006 error_setg_errno(errp, errno, "Could not preallocate data"); 1007 return -errno; 1008 } 1009 break; 1010 #endif /* CONFIG_GLUSTERFS_FALLOCATE */ 1011 #ifdef CONFIG_GLUSTERFS_ZEROFILL 1012 case PREALLOC_MODE_FULL: 1013 if (glfs_ftruncate(fd, offset)) { 1014 error_setg_errno(errp, errno, "Could not resize file"); 1015 return -errno; 1016 } 1017 if (glfs_zerofill(fd, current_length, offset - current_length)) { 1018 error_setg_errno(errp, errno, "Could not zerofill the new area"); 1019 return -errno; 1020 } 1021 break; 1022 #endif /* CONFIG_GLUSTERFS_ZEROFILL */ 1023 case PREALLOC_MODE_OFF: 1024 if (glfs_ftruncate(fd, offset)) { 1025 error_setg_errno(errp, errno, "Could not resize file"); 1026 return -errno; 1027 } 1028 break; 1029 default: 1030 error_setg(errp, "Unsupported preallocation mode: %s", 1031 PreallocMode_str(prealloc)); 1032 return -EINVAL; 1033 } 1034 1035 return 0; 1036 } 1037 1038 static int qemu_gluster_co_create(BlockdevCreateOptions *options, 1039 Error **errp) 1040 { 1041 BlockdevCreateOptionsGluster *opts = &options->u.gluster; 1042 struct glfs *glfs; 1043 struct glfs_fd *fd = NULL; 1044 int ret = 0; 1045 1046 assert(options->driver == BLOCKDEV_DRIVER_GLUSTER); 1047 1048 glfs = qemu_gluster_glfs_init(opts->location, errp); 1049 if (!glfs) { 1050 ret = -errno; 1051 goto out; 1052 } 1053 1054 fd = glfs_creat(glfs, opts->location->path, 1055 O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, S_IRUSR | S_IWUSR); 1056 if (!fd) { 1057 ret = -errno; 1058 goto out; 1059 } 1060 1061 ret = qemu_gluster_do_truncate(fd, opts->size, opts->preallocation, errp); 1062 1063 out: 1064 if (fd) { 1065 if (glfs_close(fd) != 0 && ret == 0) { 1066 ret = -errno; 1067 } 1068 } 1069 glfs_clear_preopened(glfs); 1070 return ret; 1071 } 1072 1073 static int coroutine_fn qemu_gluster_co_create_opts(const char *filename, 1074 QemuOpts *opts, 1075 Error **errp) 1076 { 1077 BlockdevCreateOptions *options; 1078 BlockdevCreateOptionsGluster *gopts; 1079 BlockdevOptionsGluster *gconf; 1080 char *tmp = NULL; 1081 Error *local_err = NULL; 1082 int ret; 1083 1084 options = g_new0(BlockdevCreateOptions, 1); 1085 options->driver = BLOCKDEV_DRIVER_GLUSTER; 1086 gopts = &options->u.gluster; 1087 1088 gconf = g_new0(BlockdevOptionsGluster, 1); 1089 gopts->location = gconf; 1090 1091 gopts->size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0), 1092 BDRV_SECTOR_SIZE); 1093 1094 tmp = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC); 1095 gopts->preallocation = qapi_enum_parse(&PreallocMode_lookup, tmp, 1096 PREALLOC_MODE_OFF, &local_err); 1097 g_free(tmp); 1098 if (local_err) { 1099 error_propagate(errp, local_err); 1100 ret = -EINVAL; 1101 goto fail; 1102 } 1103 1104 gconf->debug = qemu_opt_get_number_del(opts, GLUSTER_OPT_DEBUG, 1105 GLUSTER_DEBUG_DEFAULT); 1106 if (gconf->debug < 0) { 1107 gconf->debug = 0; 1108 } else if (gconf->debug > GLUSTER_DEBUG_MAX) { 1109 gconf->debug = GLUSTER_DEBUG_MAX; 1110 } 1111 gconf->has_debug = true; 1112 1113 gconf->logfile = qemu_opt_get_del(opts, GLUSTER_OPT_LOGFILE); 1114 if (!gconf->logfile) { 1115 gconf->logfile = g_strdup(GLUSTER_LOGFILE_DEFAULT); 1116 } 1117 gconf->has_logfile = true; 1118 1119 ret = qemu_gluster_parse(gconf, filename, NULL, errp); 1120 if (ret < 0) { 1121 goto fail; 1122 } 1123 1124 ret = qemu_gluster_co_create(options, errp); 1125 if (ret < 0) { 1126 goto fail; 1127 } 1128 1129 ret = 0; 1130 fail: 1131 qapi_free_BlockdevCreateOptions(options); 1132 return ret; 1133 } 1134 1135 static coroutine_fn int qemu_gluster_co_rw(BlockDriverState *bs, 1136 int64_t sector_num, int nb_sectors, 1137 QEMUIOVector *qiov, int write) 1138 { 1139 int ret; 1140 GlusterAIOCB acb; 1141 BDRVGlusterState *s = bs->opaque; 1142 size_t size = nb_sectors * BDRV_SECTOR_SIZE; 1143 off_t offset = sector_num * BDRV_SECTOR_SIZE; 1144 1145 acb.size = size; 1146 acb.ret = 0; 1147 acb.coroutine = qemu_coroutine_self(); 1148 acb.aio_context = bdrv_get_aio_context(bs); 1149 1150 if (write) { 1151 ret = glfs_pwritev_async(s->fd, qiov->iov, qiov->niov, offset, 0, 1152 gluster_finish_aiocb, &acb); 1153 } else { 1154 ret = glfs_preadv_async(s->fd, qiov->iov, qiov->niov, offset, 0, 1155 gluster_finish_aiocb, &acb); 1156 } 1157 1158 if (ret < 0) { 1159 return -errno; 1160 } 1161 1162 qemu_coroutine_yield(); 1163 return acb.ret; 1164 } 1165 1166 static int qemu_gluster_truncate(BlockDriverState *bs, int64_t offset, 1167 PreallocMode prealloc, Error **errp) 1168 { 1169 BDRVGlusterState *s = bs->opaque; 1170 return qemu_gluster_do_truncate(s->fd, offset, prealloc, errp); 1171 } 1172 1173 static coroutine_fn int qemu_gluster_co_readv(BlockDriverState *bs, 1174 int64_t sector_num, 1175 int nb_sectors, 1176 QEMUIOVector *qiov) 1177 { 1178 return qemu_gluster_co_rw(bs, sector_num, nb_sectors, qiov, 0); 1179 } 1180 1181 static coroutine_fn int qemu_gluster_co_writev(BlockDriverState *bs, 1182 int64_t sector_num, 1183 int nb_sectors, 1184 QEMUIOVector *qiov) 1185 { 1186 return qemu_gluster_co_rw(bs, sector_num, nb_sectors, qiov, 1); 1187 } 1188 1189 static void qemu_gluster_close(BlockDriverState *bs) 1190 { 1191 BDRVGlusterState *s = bs->opaque; 1192 1193 g_free(s->logfile); 1194 if (s->fd) { 1195 glfs_close(s->fd); 1196 s->fd = NULL; 1197 } 1198 glfs_clear_preopened(s->glfs); 1199 } 1200 1201 static coroutine_fn int qemu_gluster_co_flush_to_disk(BlockDriverState *bs) 1202 { 1203 int ret; 1204 GlusterAIOCB acb; 1205 BDRVGlusterState *s = bs->opaque; 1206 1207 acb.size = 0; 1208 acb.ret = 0; 1209 acb.coroutine = qemu_coroutine_self(); 1210 acb.aio_context = bdrv_get_aio_context(bs); 1211 1212 ret = glfs_fsync_async(s->fd, gluster_finish_aiocb, &acb); 1213 if (ret < 0) { 1214 ret = -errno; 1215 goto error; 1216 } 1217 1218 qemu_coroutine_yield(); 1219 if (acb.ret < 0) { 1220 ret = acb.ret; 1221 goto error; 1222 } 1223 1224 return acb.ret; 1225 1226 error: 1227 /* Some versions of Gluster (3.5.6 -> 3.5.8?) will not retain its cache 1228 * after a fsync failure, so we have no way of allowing the guest to safely 1229 * continue. Gluster versions prior to 3.5.6 don't retain the cache 1230 * either, but will invalidate the fd on error, so this is again our only 1231 * option. 1232 * 1233 * The 'resync-failed-syncs-after-fsync' xlator option for the 1234 * write-behind cache will cause later gluster versions to retain its 1235 * cache after error, so long as the fd remains open. However, we 1236 * currently have no way of knowing if this option is supported. 1237 * 1238 * TODO: Once gluster provides a way for us to determine if the option 1239 * is supported, bypass the closure and setting drv to NULL. */ 1240 qemu_gluster_close(bs); 1241 bs->drv = NULL; 1242 return ret; 1243 } 1244 1245 #ifdef CONFIG_GLUSTERFS_DISCARD 1246 static coroutine_fn int qemu_gluster_co_pdiscard(BlockDriverState *bs, 1247 int64_t offset, int size) 1248 { 1249 int ret; 1250 GlusterAIOCB acb; 1251 BDRVGlusterState *s = bs->opaque; 1252 1253 acb.size = 0; 1254 acb.ret = 0; 1255 acb.coroutine = qemu_coroutine_self(); 1256 acb.aio_context = bdrv_get_aio_context(bs); 1257 1258 ret = glfs_discard_async(s->fd, offset, size, gluster_finish_aiocb, &acb); 1259 if (ret < 0) { 1260 return -errno; 1261 } 1262 1263 qemu_coroutine_yield(); 1264 return acb.ret; 1265 } 1266 #endif 1267 1268 static int64_t qemu_gluster_getlength(BlockDriverState *bs) 1269 { 1270 BDRVGlusterState *s = bs->opaque; 1271 int64_t ret; 1272 1273 ret = glfs_lseek(s->fd, 0, SEEK_END); 1274 if (ret < 0) { 1275 return -errno; 1276 } else { 1277 return ret; 1278 } 1279 } 1280 1281 static int64_t qemu_gluster_allocated_file_size(BlockDriverState *bs) 1282 { 1283 BDRVGlusterState *s = bs->opaque; 1284 struct stat st; 1285 int ret; 1286 1287 ret = glfs_fstat(s->fd, &st); 1288 if (ret < 0) { 1289 return -errno; 1290 } else { 1291 return st.st_blocks * 512; 1292 } 1293 } 1294 1295 static int qemu_gluster_has_zero_init(BlockDriverState *bs) 1296 { 1297 /* GlusterFS volume could be backed by a block device */ 1298 return 0; 1299 } 1300 1301 /* 1302 * Find allocation range in @bs around offset @start. 1303 * May change underlying file descriptor's file offset. 1304 * If @start is not in a hole, store @start in @data, and the 1305 * beginning of the next hole in @hole, and return 0. 1306 * If @start is in a non-trailing hole, store @start in @hole and the 1307 * beginning of the next non-hole in @data, and return 0. 1308 * If @start is in a trailing hole or beyond EOF, return -ENXIO. 1309 * If we can't find out, return a negative errno other than -ENXIO. 1310 * 1311 * (Shamefully copied from file-posix.c, only miniscule adaptions.) 1312 */ 1313 static int find_allocation(BlockDriverState *bs, off_t start, 1314 off_t *data, off_t *hole) 1315 { 1316 BDRVGlusterState *s = bs->opaque; 1317 1318 if (!s->supports_seek_data) { 1319 goto exit; 1320 } 1321 1322 #if defined SEEK_HOLE && defined SEEK_DATA 1323 off_t offs; 1324 1325 /* 1326 * SEEK_DATA cases: 1327 * D1. offs == start: start is in data 1328 * D2. offs > start: start is in a hole, next data at offs 1329 * D3. offs < 0, errno = ENXIO: either start is in a trailing hole 1330 * or start is beyond EOF 1331 * If the latter happens, the file has been truncated behind 1332 * our back since we opened it. All bets are off then. 1333 * Treating like a trailing hole is simplest. 1334 * D4. offs < 0, errno != ENXIO: we learned nothing 1335 */ 1336 offs = glfs_lseek(s->fd, start, SEEK_DATA); 1337 if (offs < 0) { 1338 return -errno; /* D3 or D4 */ 1339 } 1340 1341 if (offs < start) { 1342 /* This is not a valid return by lseek(). We are safe to just return 1343 * -EIO in this case, and we'll treat it like D4. Unfortunately some 1344 * versions of gluster server will return offs < start, so an assert 1345 * here will unnecessarily abort QEMU. */ 1346 return -EIO; 1347 } 1348 1349 if (offs > start) { 1350 /* D2: in hole, next data at offs */ 1351 *hole = start; 1352 *data = offs; 1353 return 0; 1354 } 1355 1356 /* D1: in data, end not yet known */ 1357 1358 /* 1359 * SEEK_HOLE cases: 1360 * H1. offs == start: start is in a hole 1361 * If this happens here, a hole has been dug behind our back 1362 * since the previous lseek(). 1363 * H2. offs > start: either start is in data, next hole at offs, 1364 * or start is in trailing hole, EOF at offs 1365 * Linux treats trailing holes like any other hole: offs == 1366 * start. Solaris seeks to EOF instead: offs > start (blech). 1367 * If that happens here, a hole has been dug behind our back 1368 * since the previous lseek(). 1369 * H3. offs < 0, errno = ENXIO: start is beyond EOF 1370 * If this happens, the file has been truncated behind our 1371 * back since we opened it. Treat it like a trailing hole. 1372 * H4. offs < 0, errno != ENXIO: we learned nothing 1373 * Pretend we know nothing at all, i.e. "forget" about D1. 1374 */ 1375 offs = glfs_lseek(s->fd, start, SEEK_HOLE); 1376 if (offs < 0) { 1377 return -errno; /* D1 and (H3 or H4) */ 1378 } 1379 1380 if (offs < start) { 1381 /* This is not a valid return by lseek(). We are safe to just return 1382 * -EIO in this case, and we'll treat it like H4. Unfortunately some 1383 * versions of gluster server will return offs < start, so an assert 1384 * here will unnecessarily abort QEMU. */ 1385 return -EIO; 1386 } 1387 1388 if (offs > start) { 1389 /* 1390 * D1 and H2: either in data, next hole at offs, or it was in 1391 * data but is now in a trailing hole. In the latter case, 1392 * all bets are off. Treating it as if it there was data all 1393 * the way to EOF is safe, so simply do that. 1394 */ 1395 *data = start; 1396 *hole = offs; 1397 return 0; 1398 } 1399 1400 /* D1 and H1 */ 1401 return -EBUSY; 1402 #endif 1403 1404 exit: 1405 return -ENOTSUP; 1406 } 1407 1408 /* 1409 * Returns the allocation status of the specified offset. 1410 * 1411 * The block layer guarantees 'offset' and 'bytes' are within bounds. 1412 * 1413 * 'pnum' is set to the number of bytes (including and immediately following 1414 * the specified offset) that are known to be in the same 1415 * allocated/unallocated state. 1416 * 1417 * 'bytes' is the max value 'pnum' should be set to. 1418 * 1419 * (Based on raw_co_block_status() from file-posix.c.) 1420 */ 1421 static int coroutine_fn qemu_gluster_co_block_status(BlockDriverState *bs, 1422 bool want_zero, 1423 int64_t offset, 1424 int64_t bytes, 1425 int64_t *pnum, 1426 int64_t *map, 1427 BlockDriverState **file) 1428 { 1429 BDRVGlusterState *s = bs->opaque; 1430 off_t data = 0, hole = 0; 1431 int ret = -EINVAL; 1432 1433 if (!s->fd) { 1434 return ret; 1435 } 1436 1437 if (!want_zero) { 1438 *pnum = bytes; 1439 *map = offset; 1440 *file = bs; 1441 return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID; 1442 } 1443 1444 ret = find_allocation(bs, offset, &data, &hole); 1445 if (ret == -ENXIO) { 1446 /* Trailing hole */ 1447 *pnum = bytes; 1448 ret = BDRV_BLOCK_ZERO; 1449 } else if (ret < 0) { 1450 /* No info available, so pretend there are no holes */ 1451 *pnum = bytes; 1452 ret = BDRV_BLOCK_DATA; 1453 } else if (data == offset) { 1454 /* On a data extent, compute bytes to the end of the extent, 1455 * possibly including a partial sector at EOF. */ 1456 *pnum = MIN(bytes, hole - offset); 1457 ret = BDRV_BLOCK_DATA; 1458 } else { 1459 /* On a hole, compute bytes to the beginning of the next extent. */ 1460 assert(hole == offset); 1461 *pnum = MIN(bytes, data - offset); 1462 ret = BDRV_BLOCK_ZERO; 1463 } 1464 1465 *map = offset; 1466 *file = bs; 1467 1468 return ret | BDRV_BLOCK_OFFSET_VALID; 1469 } 1470 1471 1472 static BlockDriver bdrv_gluster = { 1473 .format_name = "gluster", 1474 .protocol_name = "gluster", 1475 .instance_size = sizeof(BDRVGlusterState), 1476 .bdrv_needs_filename = false, 1477 .bdrv_file_open = qemu_gluster_open, 1478 .bdrv_reopen_prepare = qemu_gluster_reopen_prepare, 1479 .bdrv_reopen_commit = qemu_gluster_reopen_commit, 1480 .bdrv_reopen_abort = qemu_gluster_reopen_abort, 1481 .bdrv_close = qemu_gluster_close, 1482 .bdrv_co_create = qemu_gluster_co_create, 1483 .bdrv_co_create_opts = qemu_gluster_co_create_opts, 1484 .bdrv_getlength = qemu_gluster_getlength, 1485 .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size, 1486 .bdrv_truncate = qemu_gluster_truncate, 1487 .bdrv_co_readv = qemu_gluster_co_readv, 1488 .bdrv_co_writev = qemu_gluster_co_writev, 1489 .bdrv_co_flush_to_disk = qemu_gluster_co_flush_to_disk, 1490 .bdrv_has_zero_init = qemu_gluster_has_zero_init, 1491 #ifdef CONFIG_GLUSTERFS_DISCARD 1492 .bdrv_co_pdiscard = qemu_gluster_co_pdiscard, 1493 #endif 1494 #ifdef CONFIG_GLUSTERFS_ZEROFILL 1495 .bdrv_co_pwrite_zeroes = qemu_gluster_co_pwrite_zeroes, 1496 #endif 1497 .bdrv_co_block_status = qemu_gluster_co_block_status, 1498 .create_opts = &qemu_gluster_create_opts, 1499 }; 1500 1501 static BlockDriver bdrv_gluster_tcp = { 1502 .format_name = "gluster", 1503 .protocol_name = "gluster+tcp", 1504 .instance_size = sizeof(BDRVGlusterState), 1505 .bdrv_needs_filename = false, 1506 .bdrv_file_open = qemu_gluster_open, 1507 .bdrv_reopen_prepare = qemu_gluster_reopen_prepare, 1508 .bdrv_reopen_commit = qemu_gluster_reopen_commit, 1509 .bdrv_reopen_abort = qemu_gluster_reopen_abort, 1510 .bdrv_close = qemu_gluster_close, 1511 .bdrv_co_create = qemu_gluster_co_create, 1512 .bdrv_co_create_opts = qemu_gluster_co_create_opts, 1513 .bdrv_getlength = qemu_gluster_getlength, 1514 .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size, 1515 .bdrv_truncate = qemu_gluster_truncate, 1516 .bdrv_co_readv = qemu_gluster_co_readv, 1517 .bdrv_co_writev = qemu_gluster_co_writev, 1518 .bdrv_co_flush_to_disk = qemu_gluster_co_flush_to_disk, 1519 .bdrv_has_zero_init = qemu_gluster_has_zero_init, 1520 #ifdef CONFIG_GLUSTERFS_DISCARD 1521 .bdrv_co_pdiscard = qemu_gluster_co_pdiscard, 1522 #endif 1523 #ifdef CONFIG_GLUSTERFS_ZEROFILL 1524 .bdrv_co_pwrite_zeroes = qemu_gluster_co_pwrite_zeroes, 1525 #endif 1526 .bdrv_co_block_status = qemu_gluster_co_block_status, 1527 .create_opts = &qemu_gluster_create_opts, 1528 }; 1529 1530 static BlockDriver bdrv_gluster_unix = { 1531 .format_name = "gluster", 1532 .protocol_name = "gluster+unix", 1533 .instance_size = sizeof(BDRVGlusterState), 1534 .bdrv_needs_filename = true, 1535 .bdrv_file_open = qemu_gluster_open, 1536 .bdrv_reopen_prepare = qemu_gluster_reopen_prepare, 1537 .bdrv_reopen_commit = qemu_gluster_reopen_commit, 1538 .bdrv_reopen_abort = qemu_gluster_reopen_abort, 1539 .bdrv_close = qemu_gluster_close, 1540 .bdrv_co_create = qemu_gluster_co_create, 1541 .bdrv_co_create_opts = qemu_gluster_co_create_opts, 1542 .bdrv_getlength = qemu_gluster_getlength, 1543 .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size, 1544 .bdrv_truncate = qemu_gluster_truncate, 1545 .bdrv_co_readv = qemu_gluster_co_readv, 1546 .bdrv_co_writev = qemu_gluster_co_writev, 1547 .bdrv_co_flush_to_disk = qemu_gluster_co_flush_to_disk, 1548 .bdrv_has_zero_init = qemu_gluster_has_zero_init, 1549 #ifdef CONFIG_GLUSTERFS_DISCARD 1550 .bdrv_co_pdiscard = qemu_gluster_co_pdiscard, 1551 #endif 1552 #ifdef CONFIG_GLUSTERFS_ZEROFILL 1553 .bdrv_co_pwrite_zeroes = qemu_gluster_co_pwrite_zeroes, 1554 #endif 1555 .bdrv_co_block_status = qemu_gluster_co_block_status, 1556 .create_opts = &qemu_gluster_create_opts, 1557 }; 1558 1559 /* rdma is deprecated (actually never supported for volfile fetch). 1560 * Let's maintain it for the protocol compatibility, to make sure things 1561 * won't break immediately. For now, gluster+rdma will fall back to gluster+tcp 1562 * protocol with a warning. 1563 * TODO: remove gluster+rdma interface support 1564 */ 1565 static BlockDriver bdrv_gluster_rdma = { 1566 .format_name = "gluster", 1567 .protocol_name = "gluster+rdma", 1568 .instance_size = sizeof(BDRVGlusterState), 1569 .bdrv_needs_filename = true, 1570 .bdrv_file_open = qemu_gluster_open, 1571 .bdrv_reopen_prepare = qemu_gluster_reopen_prepare, 1572 .bdrv_reopen_commit = qemu_gluster_reopen_commit, 1573 .bdrv_reopen_abort = qemu_gluster_reopen_abort, 1574 .bdrv_close = qemu_gluster_close, 1575 .bdrv_co_create = qemu_gluster_co_create, 1576 .bdrv_co_create_opts = qemu_gluster_co_create_opts, 1577 .bdrv_getlength = qemu_gluster_getlength, 1578 .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size, 1579 .bdrv_truncate = qemu_gluster_truncate, 1580 .bdrv_co_readv = qemu_gluster_co_readv, 1581 .bdrv_co_writev = qemu_gluster_co_writev, 1582 .bdrv_co_flush_to_disk = qemu_gluster_co_flush_to_disk, 1583 .bdrv_has_zero_init = qemu_gluster_has_zero_init, 1584 #ifdef CONFIG_GLUSTERFS_DISCARD 1585 .bdrv_co_pdiscard = qemu_gluster_co_pdiscard, 1586 #endif 1587 #ifdef CONFIG_GLUSTERFS_ZEROFILL 1588 .bdrv_co_pwrite_zeroes = qemu_gluster_co_pwrite_zeroes, 1589 #endif 1590 .bdrv_co_block_status = qemu_gluster_co_block_status, 1591 .create_opts = &qemu_gluster_create_opts, 1592 }; 1593 1594 static void bdrv_gluster_init(void) 1595 { 1596 bdrv_register(&bdrv_gluster_rdma); 1597 bdrv_register(&bdrv_gluster_unix); 1598 bdrv_register(&bdrv_gluster_tcp); 1599 bdrv_register(&bdrv_gluster); 1600 } 1601 1602 block_init(bdrv_gluster_init); 1603