1 /* 2 * Copyright (C) 2016-2018 Red Hat, Inc. 3 * Copyright (C) 2005 Anthony Liguori <anthony@codemonkey.ws> 4 * 5 * Network Block Device Server Side 6 * 7 * This program is free software; you can redistribute it and/or modify 8 * it under the terms of the GNU General Public License as published by 9 * the Free Software Foundation; under version 2 of the License. 10 * 11 * This program is distributed in the hope that it will be useful, 12 * but WITHOUT ANY WARRANTY; without even the implied warranty of 13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 14 * GNU General Public License for more details. 15 * 16 * You should have received a copy of the GNU General Public License 17 * along with this program; if not, see <http://www.gnu.org/licenses/>. 18 */ 19 20 #include "qemu/osdep.h" 21 #include "qapi/error.h" 22 #include "trace.h" 23 #include "nbd-internal.h" 24 25 static int system_errno_to_nbd_errno(int err) 26 { 27 switch (err) { 28 case 0: 29 return NBD_SUCCESS; 30 case EPERM: 31 case EROFS: 32 return NBD_EPERM; 33 case EIO: 34 return NBD_EIO; 35 case ENOMEM: 36 return NBD_ENOMEM; 37 #ifdef EDQUOT 38 case EDQUOT: 39 #endif 40 case EFBIG: 41 case ENOSPC: 42 return NBD_ENOSPC; 43 case EOVERFLOW: 44 return NBD_EOVERFLOW; 45 case ESHUTDOWN: 46 return NBD_ESHUTDOWN; 47 case EINVAL: 48 default: 49 return NBD_EINVAL; 50 } 51 } 52 53 /* Definitions for opaque data types */ 54 55 typedef struct NBDRequestData NBDRequestData; 56 57 struct NBDRequestData { 58 QSIMPLEQ_ENTRY(NBDRequestData) entry; 59 NBDClient *client; 60 uint8_t *data; 61 bool complete; 62 }; 63 64 struct NBDExport { 65 int refcount; 66 void (*close)(NBDExport *exp); 67 68 BlockBackend *blk; 69 char *name; 70 char *description; 71 off_t dev_offset; 72 off_t size; 73 uint16_t nbdflags; 74 QTAILQ_HEAD(, NBDClient) clients; 75 QTAILQ_ENTRY(NBDExport) next; 76 77 AioContext *ctx; 78 79 BlockBackend *eject_notifier_blk; 80 Notifier eject_notifier; 81 }; 82 83 static QTAILQ_HEAD(, NBDExport) exports = QTAILQ_HEAD_INITIALIZER(exports); 84 85 struct NBDClient { 86 int refcount; 87 void (*close_fn)(NBDClient *client, bool negotiated); 88 89 NBDExport *exp; 90 QCryptoTLSCreds *tlscreds; 91 char *tlsaclname; 92 QIOChannelSocket *sioc; /* The underlying data channel */ 93 QIOChannel *ioc; /* The current I/O channel which may differ (eg TLS) */ 94 95 Coroutine *recv_coroutine; 96 97 CoMutex send_lock; 98 Coroutine *send_coroutine; 99 100 QTAILQ_ENTRY(NBDClient) next; 101 int nb_requests; 102 bool closing; 103 104 bool structured_reply; 105 106 uint32_t opt; /* Current option being negotiated */ 107 uint32_t optlen; /* remaining length of data in ioc for the option being 108 negotiated now */ 109 }; 110 111 static void nbd_client_receive_next_request(NBDClient *client); 112 113 /* Basic flow for negotiation 114 115 Server Client 116 Negotiate 117 118 or 119 120 Server Client 121 Negotiate #1 122 Option 123 Negotiate #2 124 125 ---- 126 127 followed by 128 129 Server Client 130 Request 131 Response 132 Request 133 Response 134 ... 135 ... 136 Request (type == 2) 137 138 */ 139 140 static inline void set_be_option_rep(NBDOptionReply *rep, uint32_t option, 141 uint32_t type, uint32_t length) 142 { 143 stq_be_p(&rep->magic, NBD_REP_MAGIC); 144 stl_be_p(&rep->option, option); 145 stl_be_p(&rep->type, type); 146 stl_be_p(&rep->length, length); 147 } 148 149 /* Send a reply header, including length, but no payload. 150 * Return -errno on error, 0 on success. */ 151 static int nbd_negotiate_send_rep_len(NBDClient *client, uint32_t type, 152 uint32_t len, Error **errp) 153 { 154 NBDOptionReply rep; 155 156 trace_nbd_negotiate_send_rep_len(client->opt, nbd_opt_lookup(client->opt), 157 type, nbd_rep_lookup(type), len); 158 159 assert(len < NBD_MAX_BUFFER_SIZE); 160 161 set_be_option_rep(&rep, client->opt, type, len); 162 return nbd_write(client->ioc, &rep, sizeof(rep), errp); 163 } 164 165 /* Send a reply header with default 0 length. 166 * Return -errno on error, 0 on success. */ 167 static int nbd_negotiate_send_rep(NBDClient *client, uint32_t type, 168 Error **errp) 169 { 170 return nbd_negotiate_send_rep_len(client, type, 0, errp); 171 } 172 173 /* Send an error reply. 174 * Return -errno on error, 0 on success. */ 175 static int GCC_FMT_ATTR(4, 0) 176 nbd_negotiate_send_rep_verr(NBDClient *client, uint32_t type, 177 Error **errp, const char *fmt, va_list va) 178 { 179 char *msg; 180 int ret; 181 size_t len; 182 183 msg = g_strdup_vprintf(fmt, va); 184 len = strlen(msg); 185 assert(len < 4096); 186 trace_nbd_negotiate_send_rep_err(msg); 187 ret = nbd_negotiate_send_rep_len(client, type, len, errp); 188 if (ret < 0) { 189 goto out; 190 } 191 if (nbd_write(client->ioc, msg, len, errp) < 0) { 192 error_prepend(errp, "write failed (error message): "); 193 ret = -EIO; 194 } else { 195 ret = 0; 196 } 197 198 out: 199 g_free(msg); 200 return ret; 201 } 202 203 /* Send an error reply. 204 * Return -errno on error, 0 on success. */ 205 static int GCC_FMT_ATTR(4, 5) 206 nbd_negotiate_send_rep_err(NBDClient *client, uint32_t type, 207 Error **errp, const char *fmt, ...) 208 { 209 va_list va; 210 int ret; 211 212 va_start(va, fmt); 213 ret = nbd_negotiate_send_rep_verr(client, type, errp, fmt, va); 214 va_end(va); 215 return ret; 216 } 217 218 /* Drop remainder of the current option, and send a reply with the 219 * given error type and message. Return -errno on read or write 220 * failure; or 0 if connection is still live. */ 221 static int GCC_FMT_ATTR(4, 5) 222 nbd_opt_drop(NBDClient *client, uint32_t type, Error **errp, 223 const char *fmt, ...) 224 { 225 int ret = nbd_drop(client->ioc, client->optlen, errp); 226 va_list va; 227 228 client->optlen = 0; 229 if (!ret) { 230 va_start(va, fmt); 231 ret = nbd_negotiate_send_rep_verr(client, type, errp, fmt, va); 232 va_end(va); 233 } 234 return ret; 235 } 236 237 /* Read size bytes from the unparsed payload of the current option. 238 * Return -errno on I/O error, 0 if option was completely handled by 239 * sending a reply about inconsistent lengths, or 1 on success. */ 240 static int nbd_opt_read(NBDClient *client, void *buffer, size_t size, 241 Error **errp) 242 { 243 if (size > client->optlen) { 244 return nbd_opt_drop(client, NBD_REP_ERR_INVALID, errp, 245 "Inconsistent lengths in option %s", 246 nbd_opt_lookup(client->opt)); 247 } 248 client->optlen -= size; 249 return qio_channel_read_all(client->ioc, buffer, size, errp) < 0 ? -EIO : 1; 250 } 251 252 /* Send a single NBD_REP_SERVER reply to NBD_OPT_LIST, including payload. 253 * Return -errno on error, 0 on success. */ 254 static int nbd_negotiate_send_rep_list(NBDClient *client, NBDExport *exp, 255 Error **errp) 256 { 257 size_t name_len, desc_len; 258 uint32_t len; 259 const char *name = exp->name ? exp->name : ""; 260 const char *desc = exp->description ? exp->description : ""; 261 QIOChannel *ioc = client->ioc; 262 int ret; 263 264 trace_nbd_negotiate_send_rep_list(name, desc); 265 name_len = strlen(name); 266 desc_len = strlen(desc); 267 len = name_len + desc_len + sizeof(len); 268 ret = nbd_negotiate_send_rep_len(client, NBD_REP_SERVER, len, errp); 269 if (ret < 0) { 270 return ret; 271 } 272 273 len = cpu_to_be32(name_len); 274 if (nbd_write(ioc, &len, sizeof(len), errp) < 0) { 275 error_prepend(errp, "write failed (name length): "); 276 return -EINVAL; 277 } 278 279 if (nbd_write(ioc, name, name_len, errp) < 0) { 280 error_prepend(errp, "write failed (name buffer): "); 281 return -EINVAL; 282 } 283 284 if (nbd_write(ioc, desc, desc_len, errp) < 0) { 285 error_prepend(errp, "write failed (description buffer): "); 286 return -EINVAL; 287 } 288 289 return 0; 290 } 291 292 /* Process the NBD_OPT_LIST command, with a potential series of replies. 293 * Return -errno on error, 0 on success. */ 294 static int nbd_negotiate_handle_list(NBDClient *client, Error **errp) 295 { 296 NBDExport *exp; 297 assert(client->opt == NBD_OPT_LIST); 298 299 /* For each export, send a NBD_REP_SERVER reply. */ 300 QTAILQ_FOREACH(exp, &exports, next) { 301 if (nbd_negotiate_send_rep_list(client, exp, errp)) { 302 return -EINVAL; 303 } 304 } 305 /* Finish with a NBD_REP_ACK. */ 306 return nbd_negotiate_send_rep(client, NBD_REP_ACK, errp); 307 } 308 309 /* Send a reply to NBD_OPT_EXPORT_NAME. 310 * Return -errno on error, 0 on success. */ 311 static int nbd_negotiate_handle_export_name(NBDClient *client, 312 uint16_t myflags, bool no_zeroes, 313 Error **errp) 314 { 315 char name[NBD_MAX_NAME_SIZE + 1]; 316 char buf[NBD_REPLY_EXPORT_NAME_SIZE] = ""; 317 size_t len; 318 int ret; 319 320 /* Client sends: 321 [20 .. xx] export name (length bytes) 322 Server replies: 323 [ 0 .. 7] size 324 [ 8 .. 9] export flags 325 [10 .. 133] reserved (0) [unless no_zeroes] 326 */ 327 trace_nbd_negotiate_handle_export_name(); 328 if (client->optlen >= sizeof(name)) { 329 error_setg(errp, "Bad length received"); 330 return -EINVAL; 331 } 332 if (nbd_read(client->ioc, name, client->optlen, errp) < 0) { 333 error_prepend(errp, "read failed: "); 334 return -EIO; 335 } 336 name[client->optlen] = '\0'; 337 client->optlen = 0; 338 339 trace_nbd_negotiate_handle_export_name_request(name); 340 341 client->exp = nbd_export_find(name); 342 if (!client->exp) { 343 error_setg(errp, "export not found"); 344 return -EINVAL; 345 } 346 347 trace_nbd_negotiate_new_style_size_flags(client->exp->size, 348 client->exp->nbdflags | myflags); 349 stq_be_p(buf, client->exp->size); 350 stw_be_p(buf + 8, client->exp->nbdflags | myflags); 351 len = no_zeroes ? 10 : sizeof(buf); 352 ret = nbd_write(client->ioc, buf, len, errp); 353 if (ret < 0) { 354 error_prepend(errp, "write failed: "); 355 return ret; 356 } 357 358 QTAILQ_INSERT_TAIL(&client->exp->clients, client, next); 359 nbd_export_get(client->exp); 360 361 return 0; 362 } 363 364 /* Send a single NBD_REP_INFO, with a buffer @buf of @length bytes. 365 * The buffer does NOT include the info type prefix. 366 * Return -errno on error, 0 if ready to send more. */ 367 static int nbd_negotiate_send_info(NBDClient *client, 368 uint16_t info, uint32_t length, void *buf, 369 Error **errp) 370 { 371 int rc; 372 373 trace_nbd_negotiate_send_info(info, nbd_info_lookup(info), length); 374 rc = nbd_negotiate_send_rep_len(client, NBD_REP_INFO, 375 sizeof(info) + length, errp); 376 if (rc < 0) { 377 return rc; 378 } 379 cpu_to_be16s(&info); 380 if (nbd_write(client->ioc, &info, sizeof(info), errp) < 0) { 381 return -EIO; 382 } 383 if (nbd_write(client->ioc, buf, length, errp) < 0) { 384 return -EIO; 385 } 386 return 0; 387 } 388 389 /* nbd_reject_length: Handle any unexpected payload. 390 * @fatal requests that we quit talking to the client, even if we are able 391 * to successfully send an error reply. 392 * Return: 393 * -errno transmission error occurred or @fatal was requested, errp is set 394 * 0 error message successfully sent to client, errp is not set 395 */ 396 static int nbd_reject_length(NBDClient *client, bool fatal, Error **errp) 397 { 398 int ret; 399 400 assert(client->optlen); 401 ret = nbd_opt_drop(client, NBD_REP_ERR_INVALID, errp, 402 "option '%s' has unexpected length", 403 nbd_opt_lookup(client->opt)); 404 if (fatal && !ret) { 405 error_setg(errp, "option '%s' has unexpected length", 406 nbd_opt_lookup(client->opt)); 407 return -EINVAL; 408 } 409 return ret; 410 } 411 412 /* Handle NBD_OPT_INFO and NBD_OPT_GO. 413 * Return -errno on error, 0 if ready for next option, and 1 to move 414 * into transmission phase. */ 415 static int nbd_negotiate_handle_info(NBDClient *client, uint16_t myflags, 416 Error **errp) 417 { 418 int rc; 419 char name[NBD_MAX_NAME_SIZE + 1]; 420 NBDExport *exp; 421 uint16_t requests; 422 uint16_t request; 423 uint32_t namelen; 424 bool sendname = false; 425 bool blocksize = false; 426 uint32_t sizes[3]; 427 char buf[sizeof(uint64_t) + sizeof(uint16_t)]; 428 429 /* Client sends: 430 4 bytes: L, name length (can be 0) 431 L bytes: export name 432 2 bytes: N, number of requests (can be 0) 433 N * 2 bytes: N requests 434 */ 435 rc = nbd_opt_read(client, &namelen, sizeof(namelen), errp); 436 if (rc <= 0) { 437 return rc; 438 } 439 be32_to_cpus(&namelen); 440 if (namelen >= sizeof(name)) { 441 return nbd_opt_drop(client, NBD_REP_ERR_INVALID, errp, 442 "name too long for qemu"); 443 } 444 rc = nbd_opt_read(client, name, namelen, errp); 445 if (rc <= 0) { 446 return rc; 447 } 448 name[namelen] = '\0'; 449 trace_nbd_negotiate_handle_export_name_request(name); 450 451 rc = nbd_opt_read(client, &requests, sizeof(requests), errp); 452 if (rc <= 0) { 453 return rc; 454 } 455 be16_to_cpus(&requests); 456 trace_nbd_negotiate_handle_info_requests(requests); 457 while (requests--) { 458 rc = nbd_opt_read(client, &request, sizeof(request), errp); 459 if (rc <= 0) { 460 return rc; 461 } 462 be16_to_cpus(&request); 463 trace_nbd_negotiate_handle_info_request(request, 464 nbd_info_lookup(request)); 465 /* We care about NBD_INFO_NAME and NBD_INFO_BLOCK_SIZE; 466 * everything else is either a request we don't know or 467 * something we send regardless of request */ 468 switch (request) { 469 case NBD_INFO_NAME: 470 sendname = true; 471 break; 472 case NBD_INFO_BLOCK_SIZE: 473 blocksize = true; 474 break; 475 } 476 } 477 if (client->optlen) { 478 return nbd_reject_length(client, false, errp); 479 } 480 481 exp = nbd_export_find(name); 482 if (!exp) { 483 return nbd_negotiate_send_rep_err(client, NBD_REP_ERR_UNKNOWN, 484 errp, "export '%s' not present", 485 name); 486 } 487 488 /* Don't bother sending NBD_INFO_NAME unless client requested it */ 489 if (sendname) { 490 rc = nbd_negotiate_send_info(client, NBD_INFO_NAME, namelen, name, 491 errp); 492 if (rc < 0) { 493 return rc; 494 } 495 } 496 497 /* Send NBD_INFO_DESCRIPTION only if available, regardless of 498 * client request */ 499 if (exp->description) { 500 size_t len = strlen(exp->description); 501 502 rc = nbd_negotiate_send_info(client, NBD_INFO_DESCRIPTION, 503 len, exp->description, errp); 504 if (rc < 0) { 505 return rc; 506 } 507 } 508 509 /* Send NBD_INFO_BLOCK_SIZE always, but tweak the minimum size 510 * according to whether the client requested it, and according to 511 * whether this is OPT_INFO or OPT_GO. */ 512 /* minimum - 1 for back-compat, or 512 if client is new enough. 513 * TODO: consult blk_bs(blk)->bl.request_alignment? */ 514 sizes[0] = 515 (client->opt == NBD_OPT_INFO || blocksize) ? BDRV_SECTOR_SIZE : 1; 516 /* preferred - Hard-code to 4096 for now. 517 * TODO: is blk_bs(blk)->bl.opt_transfer appropriate? */ 518 sizes[1] = 4096; 519 /* maximum - At most 32M, but smaller as appropriate. */ 520 sizes[2] = MIN(blk_get_max_transfer(exp->blk), NBD_MAX_BUFFER_SIZE); 521 trace_nbd_negotiate_handle_info_block_size(sizes[0], sizes[1], sizes[2]); 522 cpu_to_be32s(&sizes[0]); 523 cpu_to_be32s(&sizes[1]); 524 cpu_to_be32s(&sizes[2]); 525 rc = nbd_negotiate_send_info(client, NBD_INFO_BLOCK_SIZE, 526 sizeof(sizes), sizes, errp); 527 if (rc < 0) { 528 return rc; 529 } 530 531 /* Send NBD_INFO_EXPORT always */ 532 trace_nbd_negotiate_new_style_size_flags(exp->size, 533 exp->nbdflags | myflags); 534 stq_be_p(buf, exp->size); 535 stw_be_p(buf + 8, exp->nbdflags | myflags); 536 rc = nbd_negotiate_send_info(client, NBD_INFO_EXPORT, 537 sizeof(buf), buf, errp); 538 if (rc < 0) { 539 return rc; 540 } 541 542 /* If the client is just asking for NBD_OPT_INFO, but forgot to 543 * request block sizes, return an error. 544 * TODO: consult blk_bs(blk)->request_align, and only error if it 545 * is not 1? */ 546 if (client->opt == NBD_OPT_INFO && !blocksize) { 547 return nbd_negotiate_send_rep_err(client, 548 NBD_REP_ERR_BLOCK_SIZE_REQD, 549 errp, 550 "request NBD_INFO_BLOCK_SIZE to " 551 "use this export"); 552 } 553 554 /* Final reply */ 555 rc = nbd_negotiate_send_rep(client, NBD_REP_ACK, errp); 556 if (rc < 0) { 557 return rc; 558 } 559 560 if (client->opt == NBD_OPT_GO) { 561 client->exp = exp; 562 QTAILQ_INSERT_TAIL(&client->exp->clients, client, next); 563 nbd_export_get(client->exp); 564 rc = 1; 565 } 566 return rc; 567 } 568 569 570 /* Handle NBD_OPT_STARTTLS. Return NULL to drop connection, or else the 571 * new channel for all further (now-encrypted) communication. */ 572 static QIOChannel *nbd_negotiate_handle_starttls(NBDClient *client, 573 Error **errp) 574 { 575 QIOChannel *ioc; 576 QIOChannelTLS *tioc; 577 struct NBDTLSHandshakeData data = { 0 }; 578 579 assert(client->opt == NBD_OPT_STARTTLS); 580 581 trace_nbd_negotiate_handle_starttls(); 582 ioc = client->ioc; 583 584 if (nbd_negotiate_send_rep(client, NBD_REP_ACK, errp) < 0) { 585 return NULL; 586 } 587 588 tioc = qio_channel_tls_new_server(ioc, 589 client->tlscreds, 590 client->tlsaclname, 591 errp); 592 if (!tioc) { 593 return NULL; 594 } 595 596 qio_channel_set_name(QIO_CHANNEL(tioc), "nbd-server-tls"); 597 trace_nbd_negotiate_handle_starttls_handshake(); 598 data.loop = g_main_loop_new(g_main_context_default(), FALSE); 599 qio_channel_tls_handshake(tioc, 600 nbd_tls_handshake, 601 &data, 602 NULL, 603 NULL); 604 605 if (!data.complete) { 606 g_main_loop_run(data.loop); 607 } 608 g_main_loop_unref(data.loop); 609 if (data.error) { 610 object_unref(OBJECT(tioc)); 611 error_propagate(errp, data.error); 612 return NULL; 613 } 614 615 return QIO_CHANNEL(tioc); 616 } 617 618 /* nbd_negotiate_options 619 * Process all NBD_OPT_* client option commands, during fixed newstyle 620 * negotiation. 621 * Return: 622 * -errno on error, errp is set 623 * 0 on successful negotiation, errp is not set 624 * 1 if client sent NBD_OPT_ABORT, i.e. on valid disconnect, 625 * errp is not set 626 */ 627 static int nbd_negotiate_options(NBDClient *client, uint16_t myflags, 628 Error **errp) 629 { 630 uint32_t flags; 631 bool fixedNewstyle = false; 632 bool no_zeroes = false; 633 634 /* Client sends: 635 [ 0 .. 3] client flags 636 637 Then we loop until NBD_OPT_EXPORT_NAME or NBD_OPT_GO: 638 [ 0 .. 7] NBD_OPTS_MAGIC 639 [ 8 .. 11] NBD option 640 [12 .. 15] Data length 641 ... Rest of request 642 643 [ 0 .. 7] NBD_OPTS_MAGIC 644 [ 8 .. 11] Second NBD option 645 [12 .. 15] Data length 646 ... Rest of request 647 */ 648 649 if (nbd_read(client->ioc, &flags, sizeof(flags), errp) < 0) { 650 error_prepend(errp, "read failed: "); 651 return -EIO; 652 } 653 be32_to_cpus(&flags); 654 trace_nbd_negotiate_options_flags(flags); 655 if (flags & NBD_FLAG_C_FIXED_NEWSTYLE) { 656 fixedNewstyle = true; 657 flags &= ~NBD_FLAG_C_FIXED_NEWSTYLE; 658 } 659 if (flags & NBD_FLAG_C_NO_ZEROES) { 660 no_zeroes = true; 661 flags &= ~NBD_FLAG_C_NO_ZEROES; 662 } 663 if (flags != 0) { 664 error_setg(errp, "Unknown client flags 0x%" PRIx32 " received", flags); 665 return -EINVAL; 666 } 667 668 while (1) { 669 int ret; 670 uint32_t option, length; 671 uint64_t magic; 672 673 if (nbd_read(client->ioc, &magic, sizeof(magic), errp) < 0) { 674 error_prepend(errp, "read failed: "); 675 return -EINVAL; 676 } 677 magic = be64_to_cpu(magic); 678 trace_nbd_negotiate_options_check_magic(magic); 679 if (magic != NBD_OPTS_MAGIC) { 680 error_setg(errp, "Bad magic received"); 681 return -EINVAL; 682 } 683 684 if (nbd_read(client->ioc, &option, 685 sizeof(option), errp) < 0) { 686 error_prepend(errp, "read failed: "); 687 return -EINVAL; 688 } 689 option = be32_to_cpu(option); 690 client->opt = option; 691 692 if (nbd_read(client->ioc, &length, sizeof(length), errp) < 0) { 693 error_prepend(errp, "read failed: "); 694 return -EINVAL; 695 } 696 length = be32_to_cpu(length); 697 assert(!client->optlen); 698 client->optlen = length; 699 700 if (length > NBD_MAX_BUFFER_SIZE) { 701 error_setg(errp, "len (%" PRIu32" ) is larger than max len (%u)", 702 length, NBD_MAX_BUFFER_SIZE); 703 return -EINVAL; 704 } 705 706 trace_nbd_negotiate_options_check_option(option, 707 nbd_opt_lookup(option)); 708 if (client->tlscreds && 709 client->ioc == (QIOChannel *)client->sioc) { 710 QIOChannel *tioc; 711 if (!fixedNewstyle) { 712 error_setg(errp, "Unsupported option 0x%" PRIx32, option); 713 return -EINVAL; 714 } 715 switch (option) { 716 case NBD_OPT_STARTTLS: 717 if (length) { 718 /* Unconditionally drop the connection if the client 719 * can't start a TLS negotiation correctly */ 720 return nbd_reject_length(client, true, errp); 721 } 722 tioc = nbd_negotiate_handle_starttls(client, errp); 723 if (!tioc) { 724 return -EIO; 725 } 726 ret = 0; 727 object_unref(OBJECT(client->ioc)); 728 client->ioc = QIO_CHANNEL(tioc); 729 break; 730 731 case NBD_OPT_EXPORT_NAME: 732 /* No way to return an error to client, so drop connection */ 733 error_setg(errp, "Option 0x%x not permitted before TLS", 734 option); 735 return -EINVAL; 736 737 default: 738 ret = nbd_opt_drop(client, NBD_REP_ERR_TLS_REQD, errp, 739 "Option 0x%" PRIx32 740 "not permitted before TLS", option); 741 /* Let the client keep trying, unless they asked to 742 * quit. In this mode, we've already sent an error, so 743 * we can't ack the abort. */ 744 if (option == NBD_OPT_ABORT) { 745 return 1; 746 } 747 break; 748 } 749 } else if (fixedNewstyle) { 750 switch (option) { 751 case NBD_OPT_LIST: 752 if (length) { 753 ret = nbd_reject_length(client, false, errp); 754 } else { 755 ret = nbd_negotiate_handle_list(client, errp); 756 } 757 break; 758 759 case NBD_OPT_ABORT: 760 /* NBD spec says we must try to reply before 761 * disconnecting, but that we must also tolerate 762 * guests that don't wait for our reply. */ 763 nbd_negotiate_send_rep(client, NBD_REP_ACK, NULL); 764 return 1; 765 766 case NBD_OPT_EXPORT_NAME: 767 return nbd_negotiate_handle_export_name(client, 768 myflags, no_zeroes, 769 errp); 770 771 case NBD_OPT_INFO: 772 case NBD_OPT_GO: 773 ret = nbd_negotiate_handle_info(client, myflags, errp); 774 if (ret == 1) { 775 assert(option == NBD_OPT_GO); 776 return 0; 777 } 778 break; 779 780 case NBD_OPT_STARTTLS: 781 if (length) { 782 ret = nbd_reject_length(client, false, errp); 783 } else if (client->tlscreds) { 784 ret = nbd_negotiate_send_rep_err(client, 785 NBD_REP_ERR_INVALID, errp, 786 "TLS already enabled"); 787 } else { 788 ret = nbd_negotiate_send_rep_err(client, 789 NBD_REP_ERR_POLICY, errp, 790 "TLS not configured"); 791 } 792 break; 793 794 case NBD_OPT_STRUCTURED_REPLY: 795 if (length) { 796 ret = nbd_reject_length(client, false, errp); 797 } else if (client->structured_reply) { 798 ret = nbd_negotiate_send_rep_err( 799 client, NBD_REP_ERR_INVALID, errp, 800 "structured reply already negotiated"); 801 } else { 802 ret = nbd_negotiate_send_rep(client, NBD_REP_ACK, errp); 803 client->structured_reply = true; 804 myflags |= NBD_FLAG_SEND_DF; 805 } 806 break; 807 808 default: 809 ret = nbd_opt_drop(client, NBD_REP_ERR_UNSUP, errp, 810 "Unsupported option %" PRIu32 " (%s)", 811 option, nbd_opt_lookup(option)); 812 break; 813 } 814 } else { 815 /* 816 * If broken new-style we should drop the connection 817 * for anything except NBD_OPT_EXPORT_NAME 818 */ 819 switch (option) { 820 case NBD_OPT_EXPORT_NAME: 821 return nbd_negotiate_handle_export_name(client, 822 myflags, no_zeroes, 823 errp); 824 825 default: 826 error_setg(errp, "Unsupported option %" PRIu32 " (%s)", 827 option, nbd_opt_lookup(option)); 828 return -EINVAL; 829 } 830 } 831 if (ret < 0) { 832 return ret; 833 } 834 } 835 } 836 837 /* nbd_negotiate 838 * Return: 839 * -errno on error, errp is set 840 * 0 on successful negotiation, errp is not set 841 * 1 if client sent NBD_OPT_ABORT, i.e. on valid disconnect, 842 * errp is not set 843 */ 844 static coroutine_fn int nbd_negotiate(NBDClient *client, Error **errp) 845 { 846 char buf[NBD_OLDSTYLE_NEGOTIATE_SIZE] = ""; 847 int ret; 848 const uint16_t myflags = (NBD_FLAG_HAS_FLAGS | NBD_FLAG_SEND_TRIM | 849 NBD_FLAG_SEND_FLUSH | NBD_FLAG_SEND_FUA | 850 NBD_FLAG_SEND_WRITE_ZEROES); 851 bool oldStyle; 852 853 /* Old style negotiation header, no room for options 854 [ 0 .. 7] passwd ("NBDMAGIC") 855 [ 8 .. 15] magic (NBD_CLIENT_MAGIC) 856 [16 .. 23] size 857 [24 .. 27] export flags (zero-extended) 858 [28 .. 151] reserved (0) 859 860 New style negotiation header, client can send options 861 [ 0 .. 7] passwd ("NBDMAGIC") 862 [ 8 .. 15] magic (NBD_OPTS_MAGIC) 863 [16 .. 17] server flags (0) 864 ....options sent, ending in NBD_OPT_EXPORT_NAME or NBD_OPT_GO.... 865 */ 866 867 qio_channel_set_blocking(client->ioc, false, NULL); 868 869 trace_nbd_negotiate_begin(); 870 memcpy(buf, "NBDMAGIC", 8); 871 872 oldStyle = client->exp != NULL && !client->tlscreds; 873 if (oldStyle) { 874 trace_nbd_negotiate_old_style(client->exp->size, 875 client->exp->nbdflags | myflags); 876 stq_be_p(buf + 8, NBD_CLIENT_MAGIC); 877 stq_be_p(buf + 16, client->exp->size); 878 stl_be_p(buf + 24, client->exp->nbdflags | myflags); 879 880 if (nbd_write(client->ioc, buf, sizeof(buf), errp) < 0) { 881 error_prepend(errp, "write failed: "); 882 return -EINVAL; 883 } 884 } else { 885 stq_be_p(buf + 8, NBD_OPTS_MAGIC); 886 stw_be_p(buf + 16, NBD_FLAG_FIXED_NEWSTYLE | NBD_FLAG_NO_ZEROES); 887 888 if (nbd_write(client->ioc, buf, 18, errp) < 0) { 889 error_prepend(errp, "write failed: "); 890 return -EINVAL; 891 } 892 ret = nbd_negotiate_options(client, myflags, errp); 893 if (ret != 0) { 894 if (ret < 0) { 895 error_prepend(errp, "option negotiation failed: "); 896 } 897 return ret; 898 } 899 } 900 901 assert(!client->optlen); 902 trace_nbd_negotiate_success(); 903 904 return 0; 905 } 906 907 static int nbd_receive_request(QIOChannel *ioc, NBDRequest *request, 908 Error **errp) 909 { 910 uint8_t buf[NBD_REQUEST_SIZE]; 911 uint32_t magic; 912 int ret; 913 914 ret = nbd_read(ioc, buf, sizeof(buf), errp); 915 if (ret < 0) { 916 return ret; 917 } 918 919 /* Request 920 [ 0 .. 3] magic (NBD_REQUEST_MAGIC) 921 [ 4 .. 5] flags (NBD_CMD_FLAG_FUA, ...) 922 [ 6 .. 7] type (NBD_CMD_READ, ...) 923 [ 8 .. 15] handle 924 [16 .. 23] from 925 [24 .. 27] len 926 */ 927 928 magic = ldl_be_p(buf); 929 request->flags = lduw_be_p(buf + 4); 930 request->type = lduw_be_p(buf + 6); 931 request->handle = ldq_be_p(buf + 8); 932 request->from = ldq_be_p(buf + 16); 933 request->len = ldl_be_p(buf + 24); 934 935 trace_nbd_receive_request(magic, request->flags, request->type, 936 request->from, request->len); 937 938 if (magic != NBD_REQUEST_MAGIC) { 939 error_setg(errp, "invalid magic (got 0x%" PRIx32 ")", magic); 940 return -EINVAL; 941 } 942 return 0; 943 } 944 945 #define MAX_NBD_REQUESTS 16 946 947 void nbd_client_get(NBDClient *client) 948 { 949 client->refcount++; 950 } 951 952 void nbd_client_put(NBDClient *client) 953 { 954 if (--client->refcount == 0) { 955 /* The last reference should be dropped by client->close, 956 * which is called by client_close. 957 */ 958 assert(client->closing); 959 960 qio_channel_detach_aio_context(client->ioc); 961 object_unref(OBJECT(client->sioc)); 962 object_unref(OBJECT(client->ioc)); 963 if (client->tlscreds) { 964 object_unref(OBJECT(client->tlscreds)); 965 } 966 g_free(client->tlsaclname); 967 if (client->exp) { 968 QTAILQ_REMOVE(&client->exp->clients, client, next); 969 nbd_export_put(client->exp); 970 } 971 g_free(client); 972 } 973 } 974 975 static void client_close(NBDClient *client, bool negotiated) 976 { 977 if (client->closing) { 978 return; 979 } 980 981 client->closing = true; 982 983 /* Force requests to finish. They will drop their own references, 984 * then we'll close the socket and free the NBDClient. 985 */ 986 qio_channel_shutdown(client->ioc, QIO_CHANNEL_SHUTDOWN_BOTH, 987 NULL); 988 989 /* Also tell the client, so that they release their reference. */ 990 if (client->close_fn) { 991 client->close_fn(client, negotiated); 992 } 993 } 994 995 static NBDRequestData *nbd_request_get(NBDClient *client) 996 { 997 NBDRequestData *req; 998 999 assert(client->nb_requests <= MAX_NBD_REQUESTS - 1); 1000 client->nb_requests++; 1001 1002 req = g_new0(NBDRequestData, 1); 1003 nbd_client_get(client); 1004 req->client = client; 1005 return req; 1006 } 1007 1008 static void nbd_request_put(NBDRequestData *req) 1009 { 1010 NBDClient *client = req->client; 1011 1012 if (req->data) { 1013 qemu_vfree(req->data); 1014 } 1015 g_free(req); 1016 1017 client->nb_requests--; 1018 nbd_client_receive_next_request(client); 1019 1020 nbd_client_put(client); 1021 } 1022 1023 static void blk_aio_attached(AioContext *ctx, void *opaque) 1024 { 1025 NBDExport *exp = opaque; 1026 NBDClient *client; 1027 1028 trace_nbd_blk_aio_attached(exp->name, ctx); 1029 1030 exp->ctx = ctx; 1031 1032 QTAILQ_FOREACH(client, &exp->clients, next) { 1033 qio_channel_attach_aio_context(client->ioc, ctx); 1034 if (client->recv_coroutine) { 1035 aio_co_schedule(ctx, client->recv_coroutine); 1036 } 1037 if (client->send_coroutine) { 1038 aio_co_schedule(ctx, client->send_coroutine); 1039 } 1040 } 1041 } 1042 1043 static void blk_aio_detach(void *opaque) 1044 { 1045 NBDExport *exp = opaque; 1046 NBDClient *client; 1047 1048 trace_nbd_blk_aio_detach(exp->name, exp->ctx); 1049 1050 QTAILQ_FOREACH(client, &exp->clients, next) { 1051 qio_channel_detach_aio_context(client->ioc); 1052 } 1053 1054 exp->ctx = NULL; 1055 } 1056 1057 static void nbd_eject_notifier(Notifier *n, void *data) 1058 { 1059 NBDExport *exp = container_of(n, NBDExport, eject_notifier); 1060 nbd_export_close(exp); 1061 } 1062 1063 NBDExport *nbd_export_new(BlockDriverState *bs, off_t dev_offset, off_t size, 1064 uint16_t nbdflags, void (*close)(NBDExport *), 1065 bool writethrough, BlockBackend *on_eject_blk, 1066 Error **errp) 1067 { 1068 AioContext *ctx; 1069 BlockBackend *blk; 1070 NBDExport *exp = g_new0(NBDExport, 1); 1071 uint64_t perm; 1072 int ret; 1073 1074 /* 1075 * NBD exports are used for non-shared storage migration. Make sure 1076 * that BDRV_O_INACTIVE is cleared and the image is ready for write 1077 * access since the export could be available before migration handover. 1078 */ 1079 ctx = bdrv_get_aio_context(bs); 1080 aio_context_acquire(ctx); 1081 bdrv_invalidate_cache(bs, NULL); 1082 aio_context_release(ctx); 1083 1084 /* Don't allow resize while the NBD server is running, otherwise we don't 1085 * care what happens with the node. */ 1086 perm = BLK_PERM_CONSISTENT_READ; 1087 if ((nbdflags & NBD_FLAG_READ_ONLY) == 0) { 1088 perm |= BLK_PERM_WRITE; 1089 } 1090 blk = blk_new(perm, BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED | 1091 BLK_PERM_WRITE | BLK_PERM_GRAPH_MOD); 1092 ret = blk_insert_bs(blk, bs, errp); 1093 if (ret < 0) { 1094 goto fail; 1095 } 1096 blk_set_enable_write_cache(blk, !writethrough); 1097 1098 exp->refcount = 1; 1099 QTAILQ_INIT(&exp->clients); 1100 exp->blk = blk; 1101 exp->dev_offset = dev_offset; 1102 exp->nbdflags = nbdflags; 1103 exp->size = size < 0 ? blk_getlength(blk) : size; 1104 if (exp->size < 0) { 1105 error_setg_errno(errp, -exp->size, 1106 "Failed to determine the NBD export's length"); 1107 goto fail; 1108 } 1109 exp->size -= exp->size % BDRV_SECTOR_SIZE; 1110 1111 exp->close = close; 1112 exp->ctx = blk_get_aio_context(blk); 1113 blk_add_aio_context_notifier(blk, blk_aio_attached, blk_aio_detach, exp); 1114 1115 if (on_eject_blk) { 1116 blk_ref(on_eject_blk); 1117 exp->eject_notifier_blk = on_eject_blk; 1118 exp->eject_notifier.notify = nbd_eject_notifier; 1119 blk_add_remove_bs_notifier(on_eject_blk, &exp->eject_notifier); 1120 } 1121 return exp; 1122 1123 fail: 1124 blk_unref(blk); 1125 g_free(exp); 1126 return NULL; 1127 } 1128 1129 NBDExport *nbd_export_find(const char *name) 1130 { 1131 NBDExport *exp; 1132 QTAILQ_FOREACH(exp, &exports, next) { 1133 if (strcmp(name, exp->name) == 0) { 1134 return exp; 1135 } 1136 } 1137 1138 return NULL; 1139 } 1140 1141 void nbd_export_set_name(NBDExport *exp, const char *name) 1142 { 1143 if (exp->name == name) { 1144 return; 1145 } 1146 1147 nbd_export_get(exp); 1148 if (exp->name != NULL) { 1149 g_free(exp->name); 1150 exp->name = NULL; 1151 QTAILQ_REMOVE(&exports, exp, next); 1152 nbd_export_put(exp); 1153 } 1154 if (name != NULL) { 1155 nbd_export_get(exp); 1156 exp->name = g_strdup(name); 1157 QTAILQ_INSERT_TAIL(&exports, exp, next); 1158 } 1159 nbd_export_put(exp); 1160 } 1161 1162 void nbd_export_set_description(NBDExport *exp, const char *description) 1163 { 1164 g_free(exp->description); 1165 exp->description = g_strdup(description); 1166 } 1167 1168 void nbd_export_close(NBDExport *exp) 1169 { 1170 NBDClient *client, *next; 1171 1172 nbd_export_get(exp); 1173 QTAILQ_FOREACH_SAFE(client, &exp->clients, next, next) { 1174 client_close(client, true); 1175 } 1176 nbd_export_set_name(exp, NULL); 1177 nbd_export_set_description(exp, NULL); 1178 nbd_export_put(exp); 1179 } 1180 1181 void nbd_export_remove(NBDExport *exp, NbdServerRemoveMode mode, Error **errp) 1182 { 1183 if (mode == NBD_SERVER_REMOVE_MODE_HARD || QTAILQ_EMPTY(&exp->clients)) { 1184 nbd_export_close(exp); 1185 return; 1186 } 1187 1188 assert(mode == NBD_SERVER_REMOVE_MODE_SAFE); 1189 1190 error_setg(errp, "export '%s' still in use", exp->name); 1191 error_append_hint(errp, "Use mode='hard' to force client disconnect\n"); 1192 } 1193 1194 void nbd_export_get(NBDExport *exp) 1195 { 1196 assert(exp->refcount > 0); 1197 exp->refcount++; 1198 } 1199 1200 void nbd_export_put(NBDExport *exp) 1201 { 1202 assert(exp->refcount > 0); 1203 if (exp->refcount == 1) { 1204 nbd_export_close(exp); 1205 } 1206 1207 /* nbd_export_close() may theoretically reduce refcount to 0. It may happen 1208 * if someone calls nbd_export_put() on named export not through 1209 * nbd_export_set_name() when refcount is 1. So, let's assert that 1210 * it is > 0. 1211 */ 1212 assert(exp->refcount > 0); 1213 if (--exp->refcount == 0) { 1214 assert(exp->name == NULL); 1215 assert(exp->description == NULL); 1216 1217 if (exp->close) { 1218 exp->close(exp); 1219 } 1220 1221 if (exp->blk) { 1222 if (exp->eject_notifier_blk) { 1223 notifier_remove(&exp->eject_notifier); 1224 blk_unref(exp->eject_notifier_blk); 1225 } 1226 blk_remove_aio_context_notifier(exp->blk, blk_aio_attached, 1227 blk_aio_detach, exp); 1228 blk_unref(exp->blk); 1229 exp->blk = NULL; 1230 } 1231 1232 g_free(exp); 1233 } 1234 } 1235 1236 BlockBackend *nbd_export_get_blockdev(NBDExport *exp) 1237 { 1238 return exp->blk; 1239 } 1240 1241 void nbd_export_close_all(void) 1242 { 1243 NBDExport *exp, *next; 1244 1245 QTAILQ_FOREACH_SAFE(exp, &exports, next, next) { 1246 nbd_export_close(exp); 1247 } 1248 } 1249 1250 static int coroutine_fn nbd_co_send_iov(NBDClient *client, struct iovec *iov, 1251 unsigned niov, Error **errp) 1252 { 1253 int ret; 1254 1255 g_assert(qemu_in_coroutine()); 1256 qemu_co_mutex_lock(&client->send_lock); 1257 client->send_coroutine = qemu_coroutine_self(); 1258 1259 ret = qio_channel_writev_all(client->ioc, iov, niov, errp) < 0 ? -EIO : 0; 1260 1261 client->send_coroutine = NULL; 1262 qemu_co_mutex_unlock(&client->send_lock); 1263 1264 return ret; 1265 } 1266 1267 static inline void set_be_simple_reply(NBDSimpleReply *reply, uint64_t error, 1268 uint64_t handle) 1269 { 1270 stl_be_p(&reply->magic, NBD_SIMPLE_REPLY_MAGIC); 1271 stl_be_p(&reply->error, error); 1272 stq_be_p(&reply->handle, handle); 1273 } 1274 1275 static int nbd_co_send_simple_reply(NBDClient *client, 1276 uint64_t handle, 1277 uint32_t error, 1278 void *data, 1279 size_t len, 1280 Error **errp) 1281 { 1282 NBDSimpleReply reply; 1283 int nbd_err = system_errno_to_nbd_errno(error); 1284 struct iovec iov[] = { 1285 {.iov_base = &reply, .iov_len = sizeof(reply)}, 1286 {.iov_base = data, .iov_len = len} 1287 }; 1288 1289 trace_nbd_co_send_simple_reply(handle, nbd_err, nbd_err_lookup(nbd_err), 1290 len); 1291 set_be_simple_reply(&reply, nbd_err, handle); 1292 1293 return nbd_co_send_iov(client, iov, len ? 2 : 1, errp); 1294 } 1295 1296 static inline void set_be_chunk(NBDStructuredReplyChunk *chunk, uint16_t flags, 1297 uint16_t type, uint64_t handle, uint32_t length) 1298 { 1299 stl_be_p(&chunk->magic, NBD_STRUCTURED_REPLY_MAGIC); 1300 stw_be_p(&chunk->flags, flags); 1301 stw_be_p(&chunk->type, type); 1302 stq_be_p(&chunk->handle, handle); 1303 stl_be_p(&chunk->length, length); 1304 } 1305 1306 static int coroutine_fn nbd_co_send_structured_done(NBDClient *client, 1307 uint64_t handle, 1308 Error **errp) 1309 { 1310 NBDStructuredReplyChunk chunk; 1311 struct iovec iov[] = { 1312 {.iov_base = &chunk, .iov_len = sizeof(chunk)}, 1313 }; 1314 1315 trace_nbd_co_send_structured_done(handle); 1316 set_be_chunk(&chunk, NBD_REPLY_FLAG_DONE, NBD_REPLY_TYPE_NONE, handle, 0); 1317 1318 return nbd_co_send_iov(client, iov, 1, errp); 1319 } 1320 1321 static int coroutine_fn nbd_co_send_structured_read(NBDClient *client, 1322 uint64_t handle, 1323 uint64_t offset, 1324 void *data, 1325 size_t size, 1326 bool final, 1327 Error **errp) 1328 { 1329 NBDStructuredReadData chunk; 1330 struct iovec iov[] = { 1331 {.iov_base = &chunk, .iov_len = sizeof(chunk)}, 1332 {.iov_base = data, .iov_len = size} 1333 }; 1334 1335 assert(size); 1336 trace_nbd_co_send_structured_read(handle, offset, data, size); 1337 set_be_chunk(&chunk.h, final ? NBD_REPLY_FLAG_DONE : 0, 1338 NBD_REPLY_TYPE_OFFSET_DATA, handle, 1339 sizeof(chunk) - sizeof(chunk.h) + size); 1340 stq_be_p(&chunk.offset, offset); 1341 1342 return nbd_co_send_iov(client, iov, 2, errp); 1343 } 1344 1345 static int coroutine_fn nbd_co_send_sparse_read(NBDClient *client, 1346 uint64_t handle, 1347 uint64_t offset, 1348 uint8_t *data, 1349 size_t size, 1350 Error **errp) 1351 { 1352 int ret = 0; 1353 NBDExport *exp = client->exp; 1354 size_t progress = 0; 1355 1356 while (progress < size) { 1357 int64_t pnum; 1358 int status = bdrv_block_status_above(blk_bs(exp->blk), NULL, 1359 offset + progress, 1360 size - progress, &pnum, NULL, 1361 NULL); 1362 bool final; 1363 1364 if (status < 0) { 1365 error_setg_errno(errp, -status, "unable to check for holes"); 1366 return status; 1367 } 1368 assert(pnum && pnum <= size - progress); 1369 final = progress + pnum == size; 1370 if (status & BDRV_BLOCK_ZERO) { 1371 NBDStructuredReadHole chunk; 1372 struct iovec iov[] = { 1373 {.iov_base = &chunk, .iov_len = sizeof(chunk)}, 1374 }; 1375 1376 trace_nbd_co_send_structured_read_hole(handle, offset + progress, 1377 pnum); 1378 set_be_chunk(&chunk.h, final ? NBD_REPLY_FLAG_DONE : 0, 1379 NBD_REPLY_TYPE_OFFSET_HOLE, 1380 handle, sizeof(chunk) - sizeof(chunk.h)); 1381 stq_be_p(&chunk.offset, offset + progress); 1382 stl_be_p(&chunk.length, pnum); 1383 ret = nbd_co_send_iov(client, iov, 1, errp); 1384 } else { 1385 ret = blk_pread(exp->blk, offset + progress + exp->dev_offset, 1386 data + progress, pnum); 1387 if (ret < 0) { 1388 error_setg_errno(errp, -ret, "reading from file failed"); 1389 break; 1390 } 1391 ret = nbd_co_send_structured_read(client, handle, offset + progress, 1392 data + progress, pnum, final, 1393 errp); 1394 } 1395 1396 if (ret < 0) { 1397 break; 1398 } 1399 progress += pnum; 1400 } 1401 return ret; 1402 } 1403 1404 static int coroutine_fn nbd_co_send_structured_error(NBDClient *client, 1405 uint64_t handle, 1406 uint32_t error, 1407 const char *msg, 1408 Error **errp) 1409 { 1410 NBDStructuredError chunk; 1411 int nbd_err = system_errno_to_nbd_errno(error); 1412 struct iovec iov[] = { 1413 {.iov_base = &chunk, .iov_len = sizeof(chunk)}, 1414 {.iov_base = (char *)msg, .iov_len = msg ? strlen(msg) : 0}, 1415 }; 1416 1417 assert(nbd_err); 1418 trace_nbd_co_send_structured_error(handle, nbd_err, 1419 nbd_err_lookup(nbd_err), msg ? msg : ""); 1420 set_be_chunk(&chunk.h, NBD_REPLY_FLAG_DONE, NBD_REPLY_TYPE_ERROR, handle, 1421 sizeof(chunk) - sizeof(chunk.h) + iov[1].iov_len); 1422 stl_be_p(&chunk.error, nbd_err); 1423 stw_be_p(&chunk.message_length, iov[1].iov_len); 1424 1425 return nbd_co_send_iov(client, iov, 1 + !!iov[1].iov_len, errp); 1426 } 1427 1428 /* nbd_co_receive_request 1429 * Collect a client request. Return 0 if request looks valid, -EIO to drop 1430 * connection right away, and any other negative value to report an error to 1431 * the client (although the caller may still need to disconnect after reporting 1432 * the error). 1433 */ 1434 static int nbd_co_receive_request(NBDRequestData *req, NBDRequest *request, 1435 Error **errp) 1436 { 1437 NBDClient *client = req->client; 1438 int valid_flags; 1439 1440 g_assert(qemu_in_coroutine()); 1441 assert(client->recv_coroutine == qemu_coroutine_self()); 1442 if (nbd_receive_request(client->ioc, request, errp) < 0) { 1443 return -EIO; 1444 } 1445 1446 trace_nbd_co_receive_request_decode_type(request->handle, request->type, 1447 nbd_cmd_lookup(request->type)); 1448 1449 if (request->type != NBD_CMD_WRITE) { 1450 /* No payload, we are ready to read the next request. */ 1451 req->complete = true; 1452 } 1453 1454 if (request->type == NBD_CMD_DISC) { 1455 /* Special case: we're going to disconnect without a reply, 1456 * whether or not flags, from, or len are bogus */ 1457 return -EIO; 1458 } 1459 1460 if (request->type == NBD_CMD_READ || request->type == NBD_CMD_WRITE) { 1461 if (request->len > NBD_MAX_BUFFER_SIZE) { 1462 error_setg(errp, "len (%" PRIu32" ) is larger than max len (%u)", 1463 request->len, NBD_MAX_BUFFER_SIZE); 1464 return -EINVAL; 1465 } 1466 1467 req->data = blk_try_blockalign(client->exp->blk, request->len); 1468 if (req->data == NULL) { 1469 error_setg(errp, "No memory"); 1470 return -ENOMEM; 1471 } 1472 } 1473 if (request->type == NBD_CMD_WRITE) { 1474 if (nbd_read(client->ioc, req->data, request->len, errp) < 0) { 1475 error_prepend(errp, "reading from socket failed: "); 1476 return -EIO; 1477 } 1478 req->complete = true; 1479 1480 trace_nbd_co_receive_request_payload_received(request->handle, 1481 request->len); 1482 } 1483 1484 /* Sanity checks. */ 1485 if (client->exp->nbdflags & NBD_FLAG_READ_ONLY && 1486 (request->type == NBD_CMD_WRITE || 1487 request->type == NBD_CMD_WRITE_ZEROES || 1488 request->type == NBD_CMD_TRIM)) { 1489 error_setg(errp, "Export is read-only"); 1490 return -EROFS; 1491 } 1492 if (request->from > client->exp->size || 1493 request->from + request->len > client->exp->size) { 1494 error_setg(errp, "operation past EOF; From: %" PRIu64 ", Len: %" PRIu32 1495 ", Size: %" PRIu64, request->from, request->len, 1496 (uint64_t)client->exp->size); 1497 return (request->type == NBD_CMD_WRITE || 1498 request->type == NBD_CMD_WRITE_ZEROES) ? -ENOSPC : -EINVAL; 1499 } 1500 valid_flags = NBD_CMD_FLAG_FUA; 1501 if (request->type == NBD_CMD_READ && client->structured_reply) { 1502 valid_flags |= NBD_CMD_FLAG_DF; 1503 } else if (request->type == NBD_CMD_WRITE_ZEROES) { 1504 valid_flags |= NBD_CMD_FLAG_NO_HOLE; 1505 } 1506 if (request->flags & ~valid_flags) { 1507 error_setg(errp, "unsupported flags for command %s (got 0x%x)", 1508 nbd_cmd_lookup(request->type), request->flags); 1509 return -EINVAL; 1510 } 1511 1512 return 0; 1513 } 1514 1515 /* Owns a reference to the NBDClient passed as opaque. */ 1516 static coroutine_fn void nbd_trip(void *opaque) 1517 { 1518 NBDClient *client = opaque; 1519 NBDExport *exp = client->exp; 1520 NBDRequestData *req; 1521 NBDRequest request = { 0 }; /* GCC thinks it can be used uninitialized */ 1522 int ret; 1523 int flags; 1524 int reply_data_len = 0; 1525 Error *local_err = NULL; 1526 char *msg = NULL; 1527 1528 trace_nbd_trip(); 1529 if (client->closing) { 1530 nbd_client_put(client); 1531 return; 1532 } 1533 1534 req = nbd_request_get(client); 1535 ret = nbd_co_receive_request(req, &request, &local_err); 1536 client->recv_coroutine = NULL; 1537 nbd_client_receive_next_request(client); 1538 if (ret == -EIO) { 1539 goto disconnect; 1540 } 1541 1542 if (ret < 0) { 1543 goto reply; 1544 } 1545 1546 if (client->closing) { 1547 /* 1548 * The client may be closed when we are blocked in 1549 * nbd_co_receive_request() 1550 */ 1551 goto done; 1552 } 1553 1554 switch (request.type) { 1555 case NBD_CMD_READ: 1556 /* XXX: NBD Protocol only documents use of FUA with WRITE */ 1557 if (request.flags & NBD_CMD_FLAG_FUA) { 1558 ret = blk_co_flush(exp->blk); 1559 if (ret < 0) { 1560 error_setg_errno(&local_err, -ret, "flush failed"); 1561 break; 1562 } 1563 } 1564 1565 if (client->structured_reply && !(request.flags & NBD_CMD_FLAG_DF) && 1566 request.len) { 1567 ret = nbd_co_send_sparse_read(req->client, request.handle, 1568 request.from, req->data, request.len, 1569 &local_err); 1570 if (ret < 0) { 1571 goto reply; 1572 } 1573 goto done; 1574 } 1575 1576 ret = blk_pread(exp->blk, request.from + exp->dev_offset, 1577 req->data, request.len); 1578 if (ret < 0) { 1579 error_setg_errno(&local_err, -ret, "reading from file failed"); 1580 break; 1581 } 1582 1583 reply_data_len = request.len; 1584 1585 break; 1586 case NBD_CMD_WRITE: 1587 flags = 0; 1588 if (request.flags & NBD_CMD_FLAG_FUA) { 1589 flags |= BDRV_REQ_FUA; 1590 } 1591 ret = blk_pwrite(exp->blk, request.from + exp->dev_offset, 1592 req->data, request.len, flags); 1593 if (ret < 0) { 1594 error_setg_errno(&local_err, -ret, "writing to file failed"); 1595 } 1596 1597 break; 1598 case NBD_CMD_WRITE_ZEROES: 1599 flags = 0; 1600 if (request.flags & NBD_CMD_FLAG_FUA) { 1601 flags |= BDRV_REQ_FUA; 1602 } 1603 if (!(request.flags & NBD_CMD_FLAG_NO_HOLE)) { 1604 flags |= BDRV_REQ_MAY_UNMAP; 1605 } 1606 ret = blk_pwrite_zeroes(exp->blk, request.from + exp->dev_offset, 1607 request.len, flags); 1608 if (ret < 0) { 1609 error_setg_errno(&local_err, -ret, "writing to file failed"); 1610 } 1611 1612 break; 1613 case NBD_CMD_DISC: 1614 /* unreachable, thanks to special case in nbd_co_receive_request() */ 1615 abort(); 1616 1617 case NBD_CMD_FLUSH: 1618 ret = blk_co_flush(exp->blk); 1619 if (ret < 0) { 1620 error_setg_errno(&local_err, -ret, "flush failed"); 1621 } 1622 1623 break; 1624 case NBD_CMD_TRIM: 1625 ret = blk_co_pdiscard(exp->blk, request.from + exp->dev_offset, 1626 request.len); 1627 if (ret < 0) { 1628 error_setg_errno(&local_err, -ret, "discard failed"); 1629 } 1630 1631 break; 1632 default: 1633 error_setg(&local_err, "invalid request type (%" PRIu32 ") received", 1634 request.type); 1635 ret = -EINVAL; 1636 } 1637 1638 reply: 1639 if (local_err) { 1640 /* If we get here, local_err was not a fatal error, and should be sent 1641 * to the client. */ 1642 assert(ret < 0); 1643 msg = g_strdup(error_get_pretty(local_err)); 1644 error_report_err(local_err); 1645 local_err = NULL; 1646 } 1647 1648 if (client->structured_reply && 1649 (ret < 0 || request.type == NBD_CMD_READ)) { 1650 if (ret < 0) { 1651 ret = nbd_co_send_structured_error(req->client, request.handle, 1652 -ret, msg, &local_err); 1653 } else if (reply_data_len) { 1654 ret = nbd_co_send_structured_read(req->client, request.handle, 1655 request.from, req->data, 1656 reply_data_len, true, 1657 &local_err); 1658 } else { 1659 ret = nbd_co_send_structured_done(req->client, request.handle, 1660 &local_err); 1661 } 1662 } else { 1663 ret = nbd_co_send_simple_reply(req->client, request.handle, 1664 ret < 0 ? -ret : 0, 1665 req->data, reply_data_len, &local_err); 1666 } 1667 g_free(msg); 1668 if (ret < 0) { 1669 error_prepend(&local_err, "Failed to send reply: "); 1670 goto disconnect; 1671 } 1672 1673 /* We must disconnect after NBD_CMD_WRITE if we did not 1674 * read the payload. 1675 */ 1676 if (!req->complete) { 1677 error_setg(&local_err, "Request handling failed in intermediate state"); 1678 goto disconnect; 1679 } 1680 1681 done: 1682 nbd_request_put(req); 1683 nbd_client_put(client); 1684 return; 1685 1686 disconnect: 1687 if (local_err) { 1688 error_reportf_err(local_err, "Disconnect client, due to: "); 1689 } 1690 nbd_request_put(req); 1691 client_close(client, true); 1692 nbd_client_put(client); 1693 } 1694 1695 static void nbd_client_receive_next_request(NBDClient *client) 1696 { 1697 if (!client->recv_coroutine && client->nb_requests < MAX_NBD_REQUESTS) { 1698 nbd_client_get(client); 1699 client->recv_coroutine = qemu_coroutine_create(nbd_trip, client); 1700 aio_co_schedule(client->exp->ctx, client->recv_coroutine); 1701 } 1702 } 1703 1704 static coroutine_fn void nbd_co_client_start(void *opaque) 1705 { 1706 NBDClient *client = opaque; 1707 NBDExport *exp = client->exp; 1708 Error *local_err = NULL; 1709 1710 if (exp) { 1711 nbd_export_get(exp); 1712 QTAILQ_INSERT_TAIL(&exp->clients, client, next); 1713 } 1714 qemu_co_mutex_init(&client->send_lock); 1715 1716 if (nbd_negotiate(client, &local_err)) { 1717 if (local_err) { 1718 error_report_err(local_err); 1719 } 1720 client_close(client, false); 1721 return; 1722 } 1723 1724 nbd_client_receive_next_request(client); 1725 } 1726 1727 /* 1728 * Create a new client listener on the given export @exp, using the 1729 * given channel @sioc. Begin servicing it in a coroutine. When the 1730 * connection closes, call @close_fn with an indication of whether the 1731 * client completed negotiation. 1732 */ 1733 void nbd_client_new(NBDExport *exp, 1734 QIOChannelSocket *sioc, 1735 QCryptoTLSCreds *tlscreds, 1736 const char *tlsaclname, 1737 void (*close_fn)(NBDClient *, bool)) 1738 { 1739 NBDClient *client; 1740 Coroutine *co; 1741 1742 client = g_new0(NBDClient, 1); 1743 client->refcount = 1; 1744 client->exp = exp; 1745 client->tlscreds = tlscreds; 1746 if (tlscreds) { 1747 object_ref(OBJECT(client->tlscreds)); 1748 } 1749 client->tlsaclname = g_strdup(tlsaclname); 1750 client->sioc = sioc; 1751 object_ref(OBJECT(client->sioc)); 1752 client->ioc = QIO_CHANNEL(sioc); 1753 object_ref(OBJECT(client->ioc)); 1754 client->close_fn = close_fn; 1755 1756 co = qemu_coroutine_create(nbd_co_client_start, client); 1757 qemu_coroutine_enter(co); 1758 } 1759