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 #define NBD_META_ID_BASE_ALLOCATION 0 26 #define NBD_META_ID_DIRTY_BITMAP 1 27 28 /* NBD_MAX_BITMAP_EXTENTS: 1 mb of extents data. An empirical 29 * constant. If an increase is needed, note that the NBD protocol 30 * recommends no larger than 32 mb, so that the client won't consider 31 * the reply as a denial of service attack. */ 32 #define NBD_MAX_BITMAP_EXTENTS (0x100000 / 8) 33 34 static int system_errno_to_nbd_errno(int err) 35 { 36 switch (err) { 37 case 0: 38 return NBD_SUCCESS; 39 case EPERM: 40 case EROFS: 41 return NBD_EPERM; 42 case EIO: 43 return NBD_EIO; 44 case ENOMEM: 45 return NBD_ENOMEM; 46 #ifdef EDQUOT 47 case EDQUOT: 48 #endif 49 case EFBIG: 50 case ENOSPC: 51 return NBD_ENOSPC; 52 case EOVERFLOW: 53 return NBD_EOVERFLOW; 54 case ESHUTDOWN: 55 return NBD_ESHUTDOWN; 56 case EINVAL: 57 default: 58 return NBD_EINVAL; 59 } 60 } 61 62 /* Definitions for opaque data types */ 63 64 typedef struct NBDRequestData NBDRequestData; 65 66 struct NBDRequestData { 67 QSIMPLEQ_ENTRY(NBDRequestData) entry; 68 NBDClient *client; 69 uint8_t *data; 70 bool complete; 71 }; 72 73 struct NBDExport { 74 int refcount; 75 void (*close)(NBDExport *exp); 76 77 BlockBackend *blk; 78 char *name; 79 char *description; 80 off_t dev_offset; 81 off_t size; 82 uint16_t nbdflags; 83 QTAILQ_HEAD(, NBDClient) clients; 84 QTAILQ_ENTRY(NBDExport) next; 85 86 AioContext *ctx; 87 88 BlockBackend *eject_notifier_blk; 89 Notifier eject_notifier; 90 91 BdrvDirtyBitmap *export_bitmap; 92 char *export_bitmap_context; 93 }; 94 95 static QTAILQ_HEAD(, NBDExport) exports = QTAILQ_HEAD_INITIALIZER(exports); 96 97 /* NBDExportMetaContexts represents a list of contexts to be exported, 98 * as selected by NBD_OPT_SET_META_CONTEXT. Also used for 99 * NBD_OPT_LIST_META_CONTEXT. */ 100 typedef struct NBDExportMetaContexts { 101 NBDExport *exp; 102 bool valid; /* means that negotiation of the option finished without 103 errors */ 104 bool base_allocation; /* export base:allocation context (block status) */ 105 bool bitmap; /* export qemu:dirty-bitmap:<export bitmap name> */ 106 } NBDExportMetaContexts; 107 108 struct NBDClient { 109 int refcount; 110 void (*close_fn)(NBDClient *client, bool negotiated); 111 112 NBDExport *exp; 113 QCryptoTLSCreds *tlscreds; 114 char *tlsaclname; 115 QIOChannelSocket *sioc; /* The underlying data channel */ 116 QIOChannel *ioc; /* The current I/O channel which may differ (eg TLS) */ 117 118 Coroutine *recv_coroutine; 119 120 CoMutex send_lock; 121 Coroutine *send_coroutine; 122 123 QTAILQ_ENTRY(NBDClient) next; 124 int nb_requests; 125 bool closing; 126 127 bool structured_reply; 128 NBDExportMetaContexts export_meta; 129 130 uint32_t opt; /* Current option being negotiated */ 131 uint32_t optlen; /* remaining length of data in ioc for the option being 132 negotiated now */ 133 }; 134 135 static void nbd_client_receive_next_request(NBDClient *client); 136 137 /* Basic flow for negotiation 138 139 Server Client 140 Negotiate 141 142 or 143 144 Server Client 145 Negotiate #1 146 Option 147 Negotiate #2 148 149 ---- 150 151 followed by 152 153 Server Client 154 Request 155 Response 156 Request 157 Response 158 ... 159 ... 160 Request (type == 2) 161 162 */ 163 164 static inline void set_be_option_rep(NBDOptionReply *rep, uint32_t option, 165 uint32_t type, uint32_t length) 166 { 167 stq_be_p(&rep->magic, NBD_REP_MAGIC); 168 stl_be_p(&rep->option, option); 169 stl_be_p(&rep->type, type); 170 stl_be_p(&rep->length, length); 171 } 172 173 /* Send a reply header, including length, but no payload. 174 * Return -errno on error, 0 on success. */ 175 static int nbd_negotiate_send_rep_len(NBDClient *client, uint32_t type, 176 uint32_t len, Error **errp) 177 { 178 NBDOptionReply rep; 179 180 trace_nbd_negotiate_send_rep_len(client->opt, nbd_opt_lookup(client->opt), 181 type, nbd_rep_lookup(type), len); 182 183 assert(len < NBD_MAX_BUFFER_SIZE); 184 185 set_be_option_rep(&rep, client->opt, type, len); 186 return nbd_write(client->ioc, &rep, sizeof(rep), errp); 187 } 188 189 /* Send a reply header with default 0 length. 190 * Return -errno on error, 0 on success. */ 191 static int nbd_negotiate_send_rep(NBDClient *client, uint32_t type, 192 Error **errp) 193 { 194 return nbd_negotiate_send_rep_len(client, type, 0, errp); 195 } 196 197 /* Send an error reply. 198 * Return -errno on error, 0 on success. */ 199 static int GCC_FMT_ATTR(4, 0) 200 nbd_negotiate_send_rep_verr(NBDClient *client, uint32_t type, 201 Error **errp, const char *fmt, va_list va) 202 { 203 char *msg; 204 int ret; 205 size_t len; 206 207 msg = g_strdup_vprintf(fmt, va); 208 len = strlen(msg); 209 assert(len < 4096); 210 trace_nbd_negotiate_send_rep_err(msg); 211 ret = nbd_negotiate_send_rep_len(client, type, len, errp); 212 if (ret < 0) { 213 goto out; 214 } 215 if (nbd_write(client->ioc, msg, len, errp) < 0) { 216 error_prepend(errp, "write failed (error message): "); 217 ret = -EIO; 218 } else { 219 ret = 0; 220 } 221 222 out: 223 g_free(msg); 224 return ret; 225 } 226 227 /* Send an error reply. 228 * Return -errno on error, 0 on success. */ 229 static int GCC_FMT_ATTR(4, 5) 230 nbd_negotiate_send_rep_err(NBDClient *client, uint32_t type, 231 Error **errp, const char *fmt, ...) 232 { 233 va_list va; 234 int ret; 235 236 va_start(va, fmt); 237 ret = nbd_negotiate_send_rep_verr(client, type, errp, fmt, va); 238 va_end(va); 239 return ret; 240 } 241 242 /* Drop remainder of the current option, and send a reply with the 243 * given error type and message. Return -errno on read or write 244 * failure; or 0 if connection is still live. */ 245 static int GCC_FMT_ATTR(4, 0) 246 nbd_opt_vdrop(NBDClient *client, uint32_t type, Error **errp, 247 const char *fmt, va_list va) 248 { 249 int ret = nbd_drop(client->ioc, client->optlen, errp); 250 251 client->optlen = 0; 252 if (!ret) { 253 ret = nbd_negotiate_send_rep_verr(client, type, errp, fmt, va); 254 } 255 return ret; 256 } 257 258 static int GCC_FMT_ATTR(4, 5) 259 nbd_opt_drop(NBDClient *client, uint32_t type, Error **errp, 260 const char *fmt, ...) 261 { 262 int ret; 263 va_list va; 264 265 va_start(va, fmt); 266 ret = nbd_opt_vdrop(client, type, errp, fmt, va); 267 va_end(va); 268 269 return ret; 270 } 271 272 static int GCC_FMT_ATTR(3, 4) 273 nbd_opt_invalid(NBDClient *client, Error **errp, const char *fmt, ...) 274 { 275 int ret; 276 va_list va; 277 278 va_start(va, fmt); 279 ret = nbd_opt_vdrop(client, NBD_REP_ERR_INVALID, errp, fmt, va); 280 va_end(va); 281 282 return ret; 283 } 284 285 /* Read size bytes from the unparsed payload of the current option. 286 * Return -errno on I/O error, 0 if option was completely handled by 287 * sending a reply about inconsistent lengths, or 1 on success. */ 288 static int nbd_opt_read(NBDClient *client, void *buffer, size_t size, 289 Error **errp) 290 { 291 if (size > client->optlen) { 292 return nbd_opt_invalid(client, errp, 293 "Inconsistent lengths in option %s", 294 nbd_opt_lookup(client->opt)); 295 } 296 client->optlen -= size; 297 return qio_channel_read_all(client->ioc, buffer, size, errp) < 0 ? -EIO : 1; 298 } 299 300 /* Drop size bytes from the unparsed payload of the current option. 301 * Return -errno on I/O error, 0 if option was completely handled by 302 * sending a reply about inconsistent lengths, or 1 on success. */ 303 static int nbd_opt_skip(NBDClient *client, size_t size, Error **errp) 304 { 305 if (size > client->optlen) { 306 return nbd_opt_invalid(client, errp, 307 "Inconsistent lengths in option %s", 308 nbd_opt_lookup(client->opt)); 309 } 310 client->optlen -= size; 311 return nbd_drop(client->ioc, size, errp) < 0 ? -EIO : 1; 312 } 313 314 /* nbd_opt_read_name 315 * 316 * Read a string with the format: 317 * uint32_t len (<= NBD_MAX_NAME_SIZE) 318 * len bytes string (not 0-terminated) 319 * 320 * @name should be enough to store NBD_MAX_NAME_SIZE+1. 321 * If @length is non-null, it will be set to the actual string length. 322 * 323 * Return -errno on I/O error, 0 if option was completely handled by 324 * sending a reply about inconsistent lengths, or 1 on success. 325 */ 326 static int nbd_opt_read_name(NBDClient *client, char *name, uint32_t *length, 327 Error **errp) 328 { 329 int ret; 330 uint32_t len; 331 332 ret = nbd_opt_read(client, &len, sizeof(len), errp); 333 if (ret <= 0) { 334 return ret; 335 } 336 cpu_to_be32s(&len); 337 338 if (len > NBD_MAX_NAME_SIZE) { 339 return nbd_opt_invalid(client, errp, 340 "Invalid name length: %" PRIu32, len); 341 } 342 343 ret = nbd_opt_read(client, name, len, errp); 344 if (ret <= 0) { 345 return ret; 346 } 347 name[len] = '\0'; 348 349 if (length) { 350 *length = len; 351 } 352 353 return 1; 354 } 355 356 /* Send a single NBD_REP_SERVER reply to NBD_OPT_LIST, including payload. 357 * Return -errno on error, 0 on success. */ 358 static int nbd_negotiate_send_rep_list(NBDClient *client, NBDExport *exp, 359 Error **errp) 360 { 361 size_t name_len, desc_len; 362 uint32_t len; 363 const char *name = exp->name ? exp->name : ""; 364 const char *desc = exp->description ? exp->description : ""; 365 QIOChannel *ioc = client->ioc; 366 int ret; 367 368 trace_nbd_negotiate_send_rep_list(name, desc); 369 name_len = strlen(name); 370 desc_len = strlen(desc); 371 len = name_len + desc_len + sizeof(len); 372 ret = nbd_negotiate_send_rep_len(client, NBD_REP_SERVER, len, errp); 373 if (ret < 0) { 374 return ret; 375 } 376 377 len = cpu_to_be32(name_len); 378 if (nbd_write(ioc, &len, sizeof(len), errp) < 0) { 379 error_prepend(errp, "write failed (name length): "); 380 return -EINVAL; 381 } 382 383 if (nbd_write(ioc, name, name_len, errp) < 0) { 384 error_prepend(errp, "write failed (name buffer): "); 385 return -EINVAL; 386 } 387 388 if (nbd_write(ioc, desc, desc_len, errp) < 0) { 389 error_prepend(errp, "write failed (description buffer): "); 390 return -EINVAL; 391 } 392 393 return 0; 394 } 395 396 /* Process the NBD_OPT_LIST command, with a potential series of replies. 397 * Return -errno on error, 0 on success. */ 398 static int nbd_negotiate_handle_list(NBDClient *client, Error **errp) 399 { 400 NBDExport *exp; 401 assert(client->opt == NBD_OPT_LIST); 402 403 /* For each export, send a NBD_REP_SERVER reply. */ 404 QTAILQ_FOREACH(exp, &exports, next) { 405 if (nbd_negotiate_send_rep_list(client, exp, errp)) { 406 return -EINVAL; 407 } 408 } 409 /* Finish with a NBD_REP_ACK. */ 410 return nbd_negotiate_send_rep(client, NBD_REP_ACK, errp); 411 } 412 413 static void nbd_check_meta_export(NBDClient *client) 414 { 415 client->export_meta.valid &= client->exp == client->export_meta.exp; 416 } 417 418 /* Send a reply to NBD_OPT_EXPORT_NAME. 419 * Return -errno on error, 0 on success. */ 420 static int nbd_negotiate_handle_export_name(NBDClient *client, 421 uint16_t myflags, bool no_zeroes, 422 Error **errp) 423 { 424 char name[NBD_MAX_NAME_SIZE + 1]; 425 char buf[NBD_REPLY_EXPORT_NAME_SIZE] = ""; 426 size_t len; 427 int ret; 428 429 /* Client sends: 430 [20 .. xx] export name (length bytes) 431 Server replies: 432 [ 0 .. 7] size 433 [ 8 .. 9] export flags 434 [10 .. 133] reserved (0) [unless no_zeroes] 435 */ 436 trace_nbd_negotiate_handle_export_name(); 437 if (client->optlen >= sizeof(name)) { 438 error_setg(errp, "Bad length received"); 439 return -EINVAL; 440 } 441 if (nbd_read(client->ioc, name, client->optlen, errp) < 0) { 442 error_prepend(errp, "read failed: "); 443 return -EIO; 444 } 445 name[client->optlen] = '\0'; 446 client->optlen = 0; 447 448 trace_nbd_negotiate_handle_export_name_request(name); 449 450 client->exp = nbd_export_find(name); 451 if (!client->exp) { 452 error_setg(errp, "export not found"); 453 return -EINVAL; 454 } 455 456 trace_nbd_negotiate_new_style_size_flags(client->exp->size, 457 client->exp->nbdflags | myflags); 458 stq_be_p(buf, client->exp->size); 459 stw_be_p(buf + 8, client->exp->nbdflags | myflags); 460 len = no_zeroes ? 10 : sizeof(buf); 461 ret = nbd_write(client->ioc, buf, len, errp); 462 if (ret < 0) { 463 error_prepend(errp, "write failed: "); 464 return ret; 465 } 466 467 QTAILQ_INSERT_TAIL(&client->exp->clients, client, next); 468 nbd_export_get(client->exp); 469 nbd_check_meta_export(client); 470 471 return 0; 472 } 473 474 /* Send a single NBD_REP_INFO, with a buffer @buf of @length bytes. 475 * The buffer does NOT include the info type prefix. 476 * Return -errno on error, 0 if ready to send more. */ 477 static int nbd_negotiate_send_info(NBDClient *client, 478 uint16_t info, uint32_t length, void *buf, 479 Error **errp) 480 { 481 int rc; 482 483 trace_nbd_negotiate_send_info(info, nbd_info_lookup(info), length); 484 rc = nbd_negotiate_send_rep_len(client, NBD_REP_INFO, 485 sizeof(info) + length, errp); 486 if (rc < 0) { 487 return rc; 488 } 489 cpu_to_be16s(&info); 490 if (nbd_write(client->ioc, &info, sizeof(info), errp) < 0) { 491 return -EIO; 492 } 493 if (nbd_write(client->ioc, buf, length, errp) < 0) { 494 return -EIO; 495 } 496 return 0; 497 } 498 499 /* nbd_reject_length: Handle any unexpected payload. 500 * @fatal requests that we quit talking to the client, even if we are able 501 * to successfully send an error reply. 502 * Return: 503 * -errno transmission error occurred or @fatal was requested, errp is set 504 * 0 error message successfully sent to client, errp is not set 505 */ 506 static int nbd_reject_length(NBDClient *client, bool fatal, Error **errp) 507 { 508 int ret; 509 510 assert(client->optlen); 511 ret = nbd_opt_invalid(client, errp, "option '%s' has unexpected length", 512 nbd_opt_lookup(client->opt)); 513 if (fatal && !ret) { 514 error_setg(errp, "option '%s' has unexpected length", 515 nbd_opt_lookup(client->opt)); 516 return -EINVAL; 517 } 518 return ret; 519 } 520 521 /* Handle NBD_OPT_INFO and NBD_OPT_GO. 522 * Return -errno on error, 0 if ready for next option, and 1 to move 523 * into transmission phase. */ 524 static int nbd_negotiate_handle_info(NBDClient *client, uint16_t myflags, 525 Error **errp) 526 { 527 int rc; 528 char name[NBD_MAX_NAME_SIZE + 1]; 529 NBDExport *exp; 530 uint16_t requests; 531 uint16_t request; 532 uint32_t namelen; 533 bool sendname = false; 534 bool blocksize = false; 535 uint32_t sizes[3]; 536 char buf[sizeof(uint64_t) + sizeof(uint16_t)]; 537 538 /* Client sends: 539 4 bytes: L, name length (can be 0) 540 L bytes: export name 541 2 bytes: N, number of requests (can be 0) 542 N * 2 bytes: N requests 543 */ 544 rc = nbd_opt_read_name(client, name, &namelen, errp); 545 if (rc <= 0) { 546 return rc; 547 } 548 trace_nbd_negotiate_handle_export_name_request(name); 549 550 rc = nbd_opt_read(client, &requests, sizeof(requests), errp); 551 if (rc <= 0) { 552 return rc; 553 } 554 be16_to_cpus(&requests); 555 trace_nbd_negotiate_handle_info_requests(requests); 556 while (requests--) { 557 rc = nbd_opt_read(client, &request, sizeof(request), errp); 558 if (rc <= 0) { 559 return rc; 560 } 561 be16_to_cpus(&request); 562 trace_nbd_negotiate_handle_info_request(request, 563 nbd_info_lookup(request)); 564 /* We care about NBD_INFO_NAME and NBD_INFO_BLOCK_SIZE; 565 * everything else is either a request we don't know or 566 * something we send regardless of request */ 567 switch (request) { 568 case NBD_INFO_NAME: 569 sendname = true; 570 break; 571 case NBD_INFO_BLOCK_SIZE: 572 blocksize = true; 573 break; 574 } 575 } 576 if (client->optlen) { 577 return nbd_reject_length(client, false, errp); 578 } 579 580 exp = nbd_export_find(name); 581 if (!exp) { 582 return nbd_negotiate_send_rep_err(client, NBD_REP_ERR_UNKNOWN, 583 errp, "export '%s' not present", 584 name); 585 } 586 587 /* Don't bother sending NBD_INFO_NAME unless client requested it */ 588 if (sendname) { 589 rc = nbd_negotiate_send_info(client, NBD_INFO_NAME, namelen, name, 590 errp); 591 if (rc < 0) { 592 return rc; 593 } 594 } 595 596 /* Send NBD_INFO_DESCRIPTION only if available, regardless of 597 * client request */ 598 if (exp->description) { 599 size_t len = strlen(exp->description); 600 601 rc = nbd_negotiate_send_info(client, NBD_INFO_DESCRIPTION, 602 len, exp->description, errp); 603 if (rc < 0) { 604 return rc; 605 } 606 } 607 608 /* Send NBD_INFO_BLOCK_SIZE always, but tweak the minimum size 609 * according to whether the client requested it, and according to 610 * whether this is OPT_INFO or OPT_GO. */ 611 /* minimum - 1 for back-compat, or 512 if client is new enough. 612 * TODO: consult blk_bs(blk)->bl.request_alignment? */ 613 sizes[0] = 614 (client->opt == NBD_OPT_INFO || blocksize) ? BDRV_SECTOR_SIZE : 1; 615 /* preferred - Hard-code to 4096 for now. 616 * TODO: is blk_bs(blk)->bl.opt_transfer appropriate? */ 617 sizes[1] = 4096; 618 /* maximum - At most 32M, but smaller as appropriate. */ 619 sizes[2] = MIN(blk_get_max_transfer(exp->blk), NBD_MAX_BUFFER_SIZE); 620 trace_nbd_negotiate_handle_info_block_size(sizes[0], sizes[1], sizes[2]); 621 cpu_to_be32s(&sizes[0]); 622 cpu_to_be32s(&sizes[1]); 623 cpu_to_be32s(&sizes[2]); 624 rc = nbd_negotiate_send_info(client, NBD_INFO_BLOCK_SIZE, 625 sizeof(sizes), sizes, errp); 626 if (rc < 0) { 627 return rc; 628 } 629 630 /* Send NBD_INFO_EXPORT always */ 631 trace_nbd_negotiate_new_style_size_flags(exp->size, 632 exp->nbdflags | myflags); 633 stq_be_p(buf, exp->size); 634 stw_be_p(buf + 8, exp->nbdflags | myflags); 635 rc = nbd_negotiate_send_info(client, NBD_INFO_EXPORT, 636 sizeof(buf), buf, errp); 637 if (rc < 0) { 638 return rc; 639 } 640 641 /* If the client is just asking for NBD_OPT_INFO, but forgot to 642 * request block sizes, return an error. 643 * TODO: consult blk_bs(blk)->request_align, and only error if it 644 * is not 1? */ 645 if (client->opt == NBD_OPT_INFO && !blocksize) { 646 return nbd_negotiate_send_rep_err(client, 647 NBD_REP_ERR_BLOCK_SIZE_REQD, 648 errp, 649 "request NBD_INFO_BLOCK_SIZE to " 650 "use this export"); 651 } 652 653 /* Final reply */ 654 rc = nbd_negotiate_send_rep(client, NBD_REP_ACK, errp); 655 if (rc < 0) { 656 return rc; 657 } 658 659 if (client->opt == NBD_OPT_GO) { 660 client->exp = exp; 661 QTAILQ_INSERT_TAIL(&client->exp->clients, client, next); 662 nbd_export_get(client->exp); 663 nbd_check_meta_export(client); 664 rc = 1; 665 } 666 return rc; 667 } 668 669 670 /* Handle NBD_OPT_STARTTLS. Return NULL to drop connection, or else the 671 * new channel for all further (now-encrypted) communication. */ 672 static QIOChannel *nbd_negotiate_handle_starttls(NBDClient *client, 673 Error **errp) 674 { 675 QIOChannel *ioc; 676 QIOChannelTLS *tioc; 677 struct NBDTLSHandshakeData data = { 0 }; 678 679 assert(client->opt == NBD_OPT_STARTTLS); 680 681 trace_nbd_negotiate_handle_starttls(); 682 ioc = client->ioc; 683 684 if (nbd_negotiate_send_rep(client, NBD_REP_ACK, errp) < 0) { 685 return NULL; 686 } 687 688 tioc = qio_channel_tls_new_server(ioc, 689 client->tlscreds, 690 client->tlsaclname, 691 errp); 692 if (!tioc) { 693 return NULL; 694 } 695 696 qio_channel_set_name(QIO_CHANNEL(tioc), "nbd-server-tls"); 697 trace_nbd_negotiate_handle_starttls_handshake(); 698 data.loop = g_main_loop_new(g_main_context_default(), FALSE); 699 qio_channel_tls_handshake(tioc, 700 nbd_tls_handshake, 701 &data, 702 NULL, 703 NULL); 704 705 if (!data.complete) { 706 g_main_loop_run(data.loop); 707 } 708 g_main_loop_unref(data.loop); 709 if (data.error) { 710 object_unref(OBJECT(tioc)); 711 error_propagate(errp, data.error); 712 return NULL; 713 } 714 715 return QIO_CHANNEL(tioc); 716 } 717 718 /* nbd_negotiate_send_meta_context 719 * 720 * Send one chunk of reply to NBD_OPT_{LIST,SET}_META_CONTEXT 721 * 722 * For NBD_OPT_LIST_META_CONTEXT @context_id is ignored, 0 is used instead. 723 */ 724 static int nbd_negotiate_send_meta_context(NBDClient *client, 725 const char *context, 726 uint32_t context_id, 727 Error **errp) 728 { 729 NBDOptionReplyMetaContext opt; 730 struct iovec iov[] = { 731 {.iov_base = &opt, .iov_len = sizeof(opt)}, 732 {.iov_base = (void *)context, .iov_len = strlen(context)} 733 }; 734 735 if (client->opt == NBD_OPT_LIST_META_CONTEXT) { 736 context_id = 0; 737 } 738 739 trace_nbd_negotiate_meta_query_reply(context, context_id); 740 set_be_option_rep(&opt.h, client->opt, NBD_REP_META_CONTEXT, 741 sizeof(opt) - sizeof(opt.h) + iov[1].iov_len); 742 stl_be_p(&opt.context_id, context_id); 743 744 return qio_channel_writev_all(client->ioc, iov, 2, errp) < 0 ? -EIO : 0; 745 } 746 747 /* Read strlen(@pattern) bytes, and set @match to true if they match @pattern. 748 * @match is never set to false. 749 * 750 * Return -errno on I/O error, 0 if option was completely handled by 751 * sending a reply about inconsistent lengths, or 1 on success. 752 * 753 * Note: return code = 1 doesn't mean that we've read exactly @pattern. 754 * It only means that there are no errors. 755 */ 756 static int nbd_meta_pattern(NBDClient *client, const char *pattern, bool *match, 757 Error **errp) 758 { 759 int ret; 760 char *query; 761 size_t len = strlen(pattern); 762 763 assert(len); 764 765 query = g_malloc(len); 766 ret = nbd_opt_read(client, query, len, errp); 767 if (ret <= 0) { 768 g_free(query); 769 return ret; 770 } 771 772 if (strncmp(query, pattern, len) == 0) { 773 trace_nbd_negotiate_meta_query_parse(pattern); 774 *match = true; 775 } else { 776 trace_nbd_negotiate_meta_query_skip("pattern not matched"); 777 } 778 g_free(query); 779 780 return 1; 781 } 782 783 /* 784 * Read @len bytes, and set @match to true if they match @pattern, or if @len 785 * is 0 and the client is performing _LIST_. @match is never set to false. 786 * 787 * Return -errno on I/O error, 0 if option was completely handled by 788 * sending a reply about inconsistent lengths, or 1 on success. 789 * 790 * Note: return code = 1 doesn't mean that we've read exactly @pattern. 791 * It only means that there are no errors. 792 */ 793 static int nbd_meta_empty_or_pattern(NBDClient *client, const char *pattern, 794 uint32_t len, bool *match, Error **errp) 795 { 796 if (len == 0) { 797 if (client->opt == NBD_OPT_LIST_META_CONTEXT) { 798 *match = true; 799 } 800 trace_nbd_negotiate_meta_query_parse("empty"); 801 return 1; 802 } 803 804 if (len != strlen(pattern)) { 805 trace_nbd_negotiate_meta_query_skip("different lengths"); 806 return nbd_opt_skip(client, len, errp); 807 } 808 809 return nbd_meta_pattern(client, pattern, match, errp); 810 } 811 812 /* nbd_meta_base_query 813 * 814 * Handle queries to 'base' namespace. For now, only the base:allocation 815 * context is available. 'len' is the amount of text remaining to be read from 816 * the current name, after the 'base:' portion has been stripped. 817 * 818 * Return -errno on I/O error, 0 if option was completely handled by 819 * sending a reply about inconsistent lengths, or 1 on success. 820 */ 821 static int nbd_meta_base_query(NBDClient *client, NBDExportMetaContexts *meta, 822 uint32_t len, Error **errp) 823 { 824 return nbd_meta_empty_or_pattern(client, "allocation", len, 825 &meta->base_allocation, errp); 826 } 827 828 /* nbd_meta_bitmap_query 829 * 830 * Handle query to 'qemu:' namespace. 831 * @len is the amount of text remaining to be read from the current name, after 832 * the 'qemu:' portion has been stripped. 833 * 834 * Return -errno on I/O error, 0 if option was completely handled by 835 * sending a reply about inconsistent lengths, or 1 on success. */ 836 static int nbd_meta_qemu_query(NBDClient *client, NBDExportMetaContexts *meta, 837 uint32_t len, Error **errp) 838 { 839 bool dirty_bitmap = false; 840 size_t dirty_bitmap_len = strlen("dirty-bitmap:"); 841 int ret; 842 843 if (!meta->exp->export_bitmap) { 844 trace_nbd_negotiate_meta_query_skip("no dirty-bitmap exported"); 845 return nbd_opt_skip(client, len, errp); 846 } 847 848 if (len == 0) { 849 if (client->opt == NBD_OPT_LIST_META_CONTEXT) { 850 meta->bitmap = true; 851 } 852 trace_nbd_negotiate_meta_query_parse("empty"); 853 return 1; 854 } 855 856 if (len < dirty_bitmap_len) { 857 trace_nbd_negotiate_meta_query_skip("not dirty-bitmap:"); 858 return nbd_opt_skip(client, len, errp); 859 } 860 861 len -= dirty_bitmap_len; 862 ret = nbd_meta_pattern(client, "dirty-bitmap:", &dirty_bitmap, errp); 863 if (ret <= 0) { 864 return ret; 865 } 866 if (!dirty_bitmap) { 867 trace_nbd_negotiate_meta_query_skip("not dirty-bitmap:"); 868 return nbd_opt_skip(client, len, errp); 869 } 870 871 trace_nbd_negotiate_meta_query_parse("dirty-bitmap:"); 872 873 return nbd_meta_empty_or_pattern( 874 client, meta->exp->export_bitmap_context + 875 strlen("qemu:dirty_bitmap:"), len, &meta->bitmap, errp); 876 } 877 878 /* nbd_negotiate_meta_query 879 * 880 * Parse namespace name and call corresponding function to parse body of the 881 * query. 882 * 883 * The only supported namespace now is 'base'. 884 * 885 * The function aims not wasting time and memory to read long unknown namespace 886 * names. 887 * 888 * Return -errno on I/O error, 0 if option was completely handled by 889 * sending a reply about inconsistent lengths, or 1 on success. */ 890 static int nbd_negotiate_meta_query(NBDClient *client, 891 NBDExportMetaContexts *meta, Error **errp) 892 { 893 /* 894 * Both 'qemu' and 'base' namespaces have length = 5 including a 895 * colon. If another length namespace is later introduced, this 896 * should certainly be refactored. 897 */ 898 int ret; 899 size_t ns_len = 5; 900 char ns[5]; 901 uint32_t len; 902 903 ret = nbd_opt_read(client, &len, sizeof(len), errp); 904 if (ret <= 0) { 905 return ret; 906 } 907 cpu_to_be32s(&len); 908 909 if (len < ns_len) { 910 trace_nbd_negotiate_meta_query_skip("length too short"); 911 return nbd_opt_skip(client, len, errp); 912 } 913 914 len -= ns_len; 915 ret = nbd_opt_read(client, ns, ns_len, errp); 916 if (ret <= 0) { 917 return ret; 918 } 919 920 if (!strncmp(ns, "base:", ns_len)) { 921 trace_nbd_negotiate_meta_query_parse("base:"); 922 return nbd_meta_base_query(client, meta, len, errp); 923 } else if (!strncmp(ns, "qemu:", ns_len)) { 924 trace_nbd_negotiate_meta_query_parse("qemu:"); 925 return nbd_meta_qemu_query(client, meta, len, errp); 926 } 927 928 trace_nbd_negotiate_meta_query_skip("unknown namespace"); 929 return nbd_opt_skip(client, len, errp); 930 } 931 932 /* nbd_negotiate_meta_queries 933 * Handle NBD_OPT_LIST_META_CONTEXT and NBD_OPT_SET_META_CONTEXT 934 * 935 * Return -errno on I/O error, or 0 if option was completely handled. */ 936 static int nbd_negotiate_meta_queries(NBDClient *client, 937 NBDExportMetaContexts *meta, Error **errp) 938 { 939 int ret; 940 char export_name[NBD_MAX_NAME_SIZE + 1]; 941 NBDExportMetaContexts local_meta; 942 uint32_t nb_queries; 943 int i; 944 945 if (!client->structured_reply) { 946 return nbd_opt_invalid(client, errp, 947 "request option '%s' when structured reply " 948 "is not negotiated", 949 nbd_opt_lookup(client->opt)); 950 } 951 952 if (client->opt == NBD_OPT_LIST_META_CONTEXT) { 953 /* Only change the caller's meta on SET. */ 954 meta = &local_meta; 955 } 956 957 memset(meta, 0, sizeof(*meta)); 958 959 ret = nbd_opt_read_name(client, export_name, NULL, errp); 960 if (ret <= 0) { 961 return ret; 962 } 963 964 meta->exp = nbd_export_find(export_name); 965 if (meta->exp == NULL) { 966 return nbd_opt_drop(client, NBD_REP_ERR_UNKNOWN, errp, 967 "export '%s' not present", export_name); 968 } 969 970 ret = nbd_opt_read(client, &nb_queries, sizeof(nb_queries), errp); 971 if (ret <= 0) { 972 return ret; 973 } 974 cpu_to_be32s(&nb_queries); 975 trace_nbd_negotiate_meta_context(nbd_opt_lookup(client->opt), 976 export_name, nb_queries); 977 978 if (client->opt == NBD_OPT_LIST_META_CONTEXT && !nb_queries) { 979 /* enable all known contexts */ 980 meta->base_allocation = true; 981 } else { 982 for (i = 0; i < nb_queries; ++i) { 983 ret = nbd_negotiate_meta_query(client, meta, errp); 984 if (ret <= 0) { 985 return ret; 986 } 987 } 988 } 989 990 if (meta->base_allocation) { 991 ret = nbd_negotiate_send_meta_context(client, "base:allocation", 992 NBD_META_ID_BASE_ALLOCATION, 993 errp); 994 if (ret < 0) { 995 return ret; 996 } 997 } 998 999 if (meta->bitmap) { 1000 ret = nbd_negotiate_send_meta_context(client, 1001 meta->exp->export_bitmap_context, 1002 NBD_META_ID_DIRTY_BITMAP, 1003 errp); 1004 if (ret < 0) { 1005 return ret; 1006 } 1007 } 1008 1009 ret = nbd_negotiate_send_rep(client, NBD_REP_ACK, errp); 1010 if (ret == 0) { 1011 meta->valid = true; 1012 } 1013 1014 return ret; 1015 } 1016 1017 /* nbd_negotiate_options 1018 * Process all NBD_OPT_* client option commands, during fixed newstyle 1019 * negotiation. 1020 * Return: 1021 * -errno on error, errp is set 1022 * 0 on successful negotiation, errp is not set 1023 * 1 if client sent NBD_OPT_ABORT, i.e. on valid disconnect, 1024 * errp is not set 1025 */ 1026 static int nbd_negotiate_options(NBDClient *client, uint16_t myflags, 1027 Error **errp) 1028 { 1029 uint32_t flags; 1030 bool fixedNewstyle = false; 1031 bool no_zeroes = false; 1032 1033 /* Client sends: 1034 [ 0 .. 3] client flags 1035 1036 Then we loop until NBD_OPT_EXPORT_NAME or NBD_OPT_GO: 1037 [ 0 .. 7] NBD_OPTS_MAGIC 1038 [ 8 .. 11] NBD option 1039 [12 .. 15] Data length 1040 ... Rest of request 1041 1042 [ 0 .. 7] NBD_OPTS_MAGIC 1043 [ 8 .. 11] Second NBD option 1044 [12 .. 15] Data length 1045 ... Rest of request 1046 */ 1047 1048 if (nbd_read(client->ioc, &flags, sizeof(flags), errp) < 0) { 1049 error_prepend(errp, "read failed: "); 1050 return -EIO; 1051 } 1052 be32_to_cpus(&flags); 1053 trace_nbd_negotiate_options_flags(flags); 1054 if (flags & NBD_FLAG_C_FIXED_NEWSTYLE) { 1055 fixedNewstyle = true; 1056 flags &= ~NBD_FLAG_C_FIXED_NEWSTYLE; 1057 } 1058 if (flags & NBD_FLAG_C_NO_ZEROES) { 1059 no_zeroes = true; 1060 flags &= ~NBD_FLAG_C_NO_ZEROES; 1061 } 1062 if (flags != 0) { 1063 error_setg(errp, "Unknown client flags 0x%" PRIx32 " received", flags); 1064 return -EINVAL; 1065 } 1066 1067 while (1) { 1068 int ret; 1069 uint32_t option, length; 1070 uint64_t magic; 1071 1072 if (nbd_read(client->ioc, &magic, sizeof(magic), errp) < 0) { 1073 error_prepend(errp, "read failed: "); 1074 return -EINVAL; 1075 } 1076 magic = be64_to_cpu(magic); 1077 trace_nbd_negotiate_options_check_magic(magic); 1078 if (magic != NBD_OPTS_MAGIC) { 1079 error_setg(errp, "Bad magic received"); 1080 return -EINVAL; 1081 } 1082 1083 if (nbd_read(client->ioc, &option, 1084 sizeof(option), errp) < 0) { 1085 error_prepend(errp, "read failed: "); 1086 return -EINVAL; 1087 } 1088 option = be32_to_cpu(option); 1089 client->opt = option; 1090 1091 if (nbd_read(client->ioc, &length, sizeof(length), errp) < 0) { 1092 error_prepend(errp, "read failed: "); 1093 return -EINVAL; 1094 } 1095 length = be32_to_cpu(length); 1096 assert(!client->optlen); 1097 client->optlen = length; 1098 1099 if (length > NBD_MAX_BUFFER_SIZE) { 1100 error_setg(errp, "len (%" PRIu32" ) is larger than max len (%u)", 1101 length, NBD_MAX_BUFFER_SIZE); 1102 return -EINVAL; 1103 } 1104 1105 trace_nbd_negotiate_options_check_option(option, 1106 nbd_opt_lookup(option)); 1107 if (client->tlscreds && 1108 client->ioc == (QIOChannel *)client->sioc) { 1109 QIOChannel *tioc; 1110 if (!fixedNewstyle) { 1111 error_setg(errp, "Unsupported option 0x%" PRIx32, option); 1112 return -EINVAL; 1113 } 1114 switch (option) { 1115 case NBD_OPT_STARTTLS: 1116 if (length) { 1117 /* Unconditionally drop the connection if the client 1118 * can't start a TLS negotiation correctly */ 1119 return nbd_reject_length(client, true, errp); 1120 } 1121 tioc = nbd_negotiate_handle_starttls(client, errp); 1122 if (!tioc) { 1123 return -EIO; 1124 } 1125 ret = 0; 1126 object_unref(OBJECT(client->ioc)); 1127 client->ioc = QIO_CHANNEL(tioc); 1128 break; 1129 1130 case NBD_OPT_EXPORT_NAME: 1131 /* No way to return an error to client, so drop connection */ 1132 error_setg(errp, "Option 0x%x not permitted before TLS", 1133 option); 1134 return -EINVAL; 1135 1136 default: 1137 ret = nbd_opt_drop(client, NBD_REP_ERR_TLS_REQD, errp, 1138 "Option 0x%" PRIx32 1139 "not permitted before TLS", option); 1140 /* Let the client keep trying, unless they asked to 1141 * quit. In this mode, we've already sent an error, so 1142 * we can't ack the abort. */ 1143 if (option == NBD_OPT_ABORT) { 1144 return 1; 1145 } 1146 break; 1147 } 1148 } else if (fixedNewstyle) { 1149 switch (option) { 1150 case NBD_OPT_LIST: 1151 if (length) { 1152 ret = nbd_reject_length(client, false, errp); 1153 } else { 1154 ret = nbd_negotiate_handle_list(client, errp); 1155 } 1156 break; 1157 1158 case NBD_OPT_ABORT: 1159 /* NBD spec says we must try to reply before 1160 * disconnecting, but that we must also tolerate 1161 * guests that don't wait for our reply. */ 1162 nbd_negotiate_send_rep(client, NBD_REP_ACK, NULL); 1163 return 1; 1164 1165 case NBD_OPT_EXPORT_NAME: 1166 return nbd_negotiate_handle_export_name(client, 1167 myflags, no_zeroes, 1168 errp); 1169 1170 case NBD_OPT_INFO: 1171 case NBD_OPT_GO: 1172 ret = nbd_negotiate_handle_info(client, myflags, errp); 1173 if (ret == 1) { 1174 assert(option == NBD_OPT_GO); 1175 return 0; 1176 } 1177 break; 1178 1179 case NBD_OPT_STARTTLS: 1180 if (length) { 1181 ret = nbd_reject_length(client, false, errp); 1182 } else if (client->tlscreds) { 1183 ret = nbd_negotiate_send_rep_err(client, 1184 NBD_REP_ERR_INVALID, errp, 1185 "TLS already enabled"); 1186 } else { 1187 ret = nbd_negotiate_send_rep_err(client, 1188 NBD_REP_ERR_POLICY, errp, 1189 "TLS not configured"); 1190 } 1191 break; 1192 1193 case NBD_OPT_STRUCTURED_REPLY: 1194 if (length) { 1195 ret = nbd_reject_length(client, false, errp); 1196 } else if (client->structured_reply) { 1197 ret = nbd_negotiate_send_rep_err( 1198 client, NBD_REP_ERR_INVALID, errp, 1199 "structured reply already negotiated"); 1200 } else { 1201 ret = nbd_negotiate_send_rep(client, NBD_REP_ACK, errp); 1202 client->structured_reply = true; 1203 myflags |= NBD_FLAG_SEND_DF; 1204 } 1205 break; 1206 1207 case NBD_OPT_LIST_META_CONTEXT: 1208 case NBD_OPT_SET_META_CONTEXT: 1209 ret = nbd_negotiate_meta_queries(client, &client->export_meta, 1210 errp); 1211 break; 1212 1213 default: 1214 ret = nbd_opt_drop(client, NBD_REP_ERR_UNSUP, errp, 1215 "Unsupported option %" PRIu32 " (%s)", 1216 option, nbd_opt_lookup(option)); 1217 break; 1218 } 1219 } else { 1220 /* 1221 * If broken new-style we should drop the connection 1222 * for anything except NBD_OPT_EXPORT_NAME 1223 */ 1224 switch (option) { 1225 case NBD_OPT_EXPORT_NAME: 1226 return nbd_negotiate_handle_export_name(client, 1227 myflags, no_zeroes, 1228 errp); 1229 1230 default: 1231 error_setg(errp, "Unsupported option %" PRIu32 " (%s)", 1232 option, nbd_opt_lookup(option)); 1233 return -EINVAL; 1234 } 1235 } 1236 if (ret < 0) { 1237 return ret; 1238 } 1239 } 1240 } 1241 1242 /* nbd_negotiate 1243 * Return: 1244 * -errno on error, errp is set 1245 * 0 on successful negotiation, errp is not set 1246 * 1 if client sent NBD_OPT_ABORT, i.e. on valid disconnect, 1247 * errp is not set 1248 */ 1249 static coroutine_fn int nbd_negotiate(NBDClient *client, Error **errp) 1250 { 1251 char buf[NBD_OLDSTYLE_NEGOTIATE_SIZE] = ""; 1252 int ret; 1253 const uint16_t myflags = (NBD_FLAG_HAS_FLAGS | NBD_FLAG_SEND_TRIM | 1254 NBD_FLAG_SEND_FLUSH | NBD_FLAG_SEND_FUA | 1255 NBD_FLAG_SEND_WRITE_ZEROES | NBD_FLAG_SEND_CACHE); 1256 bool oldStyle; 1257 1258 /* Old style negotiation header, no room for options 1259 [ 0 .. 7] passwd ("NBDMAGIC") 1260 [ 8 .. 15] magic (NBD_CLIENT_MAGIC) 1261 [16 .. 23] size 1262 [24 .. 27] export flags (zero-extended) 1263 [28 .. 151] reserved (0) 1264 1265 New style negotiation header, client can send options 1266 [ 0 .. 7] passwd ("NBDMAGIC") 1267 [ 8 .. 15] magic (NBD_OPTS_MAGIC) 1268 [16 .. 17] server flags (0) 1269 ....options sent, ending in NBD_OPT_EXPORT_NAME or NBD_OPT_GO.... 1270 */ 1271 1272 qio_channel_set_blocking(client->ioc, false, NULL); 1273 1274 trace_nbd_negotiate_begin(); 1275 memcpy(buf, "NBDMAGIC", 8); 1276 1277 oldStyle = client->exp != NULL && !client->tlscreds; 1278 if (oldStyle) { 1279 trace_nbd_negotiate_old_style(client->exp->size, 1280 client->exp->nbdflags | myflags); 1281 stq_be_p(buf + 8, NBD_CLIENT_MAGIC); 1282 stq_be_p(buf + 16, client->exp->size); 1283 stl_be_p(buf + 24, client->exp->nbdflags | myflags); 1284 1285 if (nbd_write(client->ioc, buf, sizeof(buf), errp) < 0) { 1286 error_prepend(errp, "write failed: "); 1287 return -EINVAL; 1288 } 1289 } else { 1290 stq_be_p(buf + 8, NBD_OPTS_MAGIC); 1291 stw_be_p(buf + 16, NBD_FLAG_FIXED_NEWSTYLE | NBD_FLAG_NO_ZEROES); 1292 1293 if (nbd_write(client->ioc, buf, 18, errp) < 0) { 1294 error_prepend(errp, "write failed: "); 1295 return -EINVAL; 1296 } 1297 ret = nbd_negotiate_options(client, myflags, errp); 1298 if (ret != 0) { 1299 if (ret < 0) { 1300 error_prepend(errp, "option negotiation failed: "); 1301 } 1302 return ret; 1303 } 1304 } 1305 1306 assert(!client->optlen); 1307 trace_nbd_negotiate_success(); 1308 1309 return 0; 1310 } 1311 1312 static int nbd_receive_request(QIOChannel *ioc, NBDRequest *request, 1313 Error **errp) 1314 { 1315 uint8_t buf[NBD_REQUEST_SIZE]; 1316 uint32_t magic; 1317 int ret; 1318 1319 ret = nbd_read(ioc, buf, sizeof(buf), errp); 1320 if (ret < 0) { 1321 return ret; 1322 } 1323 1324 /* Request 1325 [ 0 .. 3] magic (NBD_REQUEST_MAGIC) 1326 [ 4 .. 5] flags (NBD_CMD_FLAG_FUA, ...) 1327 [ 6 .. 7] type (NBD_CMD_READ, ...) 1328 [ 8 .. 15] handle 1329 [16 .. 23] from 1330 [24 .. 27] len 1331 */ 1332 1333 magic = ldl_be_p(buf); 1334 request->flags = lduw_be_p(buf + 4); 1335 request->type = lduw_be_p(buf + 6); 1336 request->handle = ldq_be_p(buf + 8); 1337 request->from = ldq_be_p(buf + 16); 1338 request->len = ldl_be_p(buf + 24); 1339 1340 trace_nbd_receive_request(magic, request->flags, request->type, 1341 request->from, request->len); 1342 1343 if (magic != NBD_REQUEST_MAGIC) { 1344 error_setg(errp, "invalid magic (got 0x%" PRIx32 ")", magic); 1345 return -EINVAL; 1346 } 1347 return 0; 1348 } 1349 1350 #define MAX_NBD_REQUESTS 16 1351 1352 void nbd_client_get(NBDClient *client) 1353 { 1354 client->refcount++; 1355 } 1356 1357 void nbd_client_put(NBDClient *client) 1358 { 1359 if (--client->refcount == 0) { 1360 /* The last reference should be dropped by client->close, 1361 * which is called by client_close. 1362 */ 1363 assert(client->closing); 1364 1365 qio_channel_detach_aio_context(client->ioc); 1366 object_unref(OBJECT(client->sioc)); 1367 object_unref(OBJECT(client->ioc)); 1368 if (client->tlscreds) { 1369 object_unref(OBJECT(client->tlscreds)); 1370 } 1371 g_free(client->tlsaclname); 1372 if (client->exp) { 1373 QTAILQ_REMOVE(&client->exp->clients, client, next); 1374 nbd_export_put(client->exp); 1375 } 1376 g_free(client); 1377 } 1378 } 1379 1380 static void client_close(NBDClient *client, bool negotiated) 1381 { 1382 if (client->closing) { 1383 return; 1384 } 1385 1386 client->closing = true; 1387 1388 /* Force requests to finish. They will drop their own references, 1389 * then we'll close the socket and free the NBDClient. 1390 */ 1391 qio_channel_shutdown(client->ioc, QIO_CHANNEL_SHUTDOWN_BOTH, 1392 NULL); 1393 1394 /* Also tell the client, so that they release their reference. */ 1395 if (client->close_fn) { 1396 client->close_fn(client, negotiated); 1397 } 1398 } 1399 1400 static NBDRequestData *nbd_request_get(NBDClient *client) 1401 { 1402 NBDRequestData *req; 1403 1404 assert(client->nb_requests <= MAX_NBD_REQUESTS - 1); 1405 client->nb_requests++; 1406 1407 req = g_new0(NBDRequestData, 1); 1408 nbd_client_get(client); 1409 req->client = client; 1410 return req; 1411 } 1412 1413 static void nbd_request_put(NBDRequestData *req) 1414 { 1415 NBDClient *client = req->client; 1416 1417 if (req->data) { 1418 qemu_vfree(req->data); 1419 } 1420 g_free(req); 1421 1422 client->nb_requests--; 1423 nbd_client_receive_next_request(client); 1424 1425 nbd_client_put(client); 1426 } 1427 1428 static void blk_aio_attached(AioContext *ctx, void *opaque) 1429 { 1430 NBDExport *exp = opaque; 1431 NBDClient *client; 1432 1433 trace_nbd_blk_aio_attached(exp->name, ctx); 1434 1435 exp->ctx = ctx; 1436 1437 QTAILQ_FOREACH(client, &exp->clients, next) { 1438 qio_channel_attach_aio_context(client->ioc, ctx); 1439 if (client->recv_coroutine) { 1440 aio_co_schedule(ctx, client->recv_coroutine); 1441 } 1442 if (client->send_coroutine) { 1443 aio_co_schedule(ctx, client->send_coroutine); 1444 } 1445 } 1446 } 1447 1448 static void blk_aio_detach(void *opaque) 1449 { 1450 NBDExport *exp = opaque; 1451 NBDClient *client; 1452 1453 trace_nbd_blk_aio_detach(exp->name, exp->ctx); 1454 1455 QTAILQ_FOREACH(client, &exp->clients, next) { 1456 qio_channel_detach_aio_context(client->ioc); 1457 } 1458 1459 exp->ctx = NULL; 1460 } 1461 1462 static void nbd_eject_notifier(Notifier *n, void *data) 1463 { 1464 NBDExport *exp = container_of(n, NBDExport, eject_notifier); 1465 nbd_export_close(exp); 1466 } 1467 1468 NBDExport *nbd_export_new(BlockDriverState *bs, off_t dev_offset, off_t size, 1469 uint16_t nbdflags, void (*close)(NBDExport *), 1470 bool writethrough, BlockBackend *on_eject_blk, 1471 Error **errp) 1472 { 1473 AioContext *ctx; 1474 BlockBackend *blk; 1475 NBDExport *exp = g_new0(NBDExport, 1); 1476 uint64_t perm; 1477 int ret; 1478 1479 /* 1480 * NBD exports are used for non-shared storage migration. Make sure 1481 * that BDRV_O_INACTIVE is cleared and the image is ready for write 1482 * access since the export could be available before migration handover. 1483 */ 1484 ctx = bdrv_get_aio_context(bs); 1485 aio_context_acquire(ctx); 1486 bdrv_invalidate_cache(bs, NULL); 1487 aio_context_release(ctx); 1488 1489 /* Don't allow resize while the NBD server is running, otherwise we don't 1490 * care what happens with the node. */ 1491 perm = BLK_PERM_CONSISTENT_READ; 1492 if ((nbdflags & NBD_FLAG_READ_ONLY) == 0) { 1493 perm |= BLK_PERM_WRITE; 1494 } 1495 blk = blk_new(perm, BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED | 1496 BLK_PERM_WRITE | BLK_PERM_GRAPH_MOD); 1497 ret = blk_insert_bs(blk, bs, errp); 1498 if (ret < 0) { 1499 goto fail; 1500 } 1501 blk_set_enable_write_cache(blk, !writethrough); 1502 1503 exp->refcount = 1; 1504 QTAILQ_INIT(&exp->clients); 1505 exp->blk = blk; 1506 exp->dev_offset = dev_offset; 1507 exp->nbdflags = nbdflags; 1508 exp->size = size < 0 ? blk_getlength(blk) : size; 1509 if (exp->size < 0) { 1510 error_setg_errno(errp, -exp->size, 1511 "Failed to determine the NBD export's length"); 1512 goto fail; 1513 } 1514 exp->size -= exp->size % BDRV_SECTOR_SIZE; 1515 1516 exp->close = close; 1517 exp->ctx = blk_get_aio_context(blk); 1518 blk_add_aio_context_notifier(blk, blk_aio_attached, blk_aio_detach, exp); 1519 1520 if (on_eject_blk) { 1521 blk_ref(on_eject_blk); 1522 exp->eject_notifier_blk = on_eject_blk; 1523 exp->eject_notifier.notify = nbd_eject_notifier; 1524 blk_add_remove_bs_notifier(on_eject_blk, &exp->eject_notifier); 1525 } 1526 return exp; 1527 1528 fail: 1529 blk_unref(blk); 1530 g_free(exp); 1531 return NULL; 1532 } 1533 1534 NBDExport *nbd_export_find(const char *name) 1535 { 1536 NBDExport *exp; 1537 QTAILQ_FOREACH(exp, &exports, next) { 1538 if (strcmp(name, exp->name) == 0) { 1539 return exp; 1540 } 1541 } 1542 1543 return NULL; 1544 } 1545 1546 void nbd_export_set_name(NBDExport *exp, const char *name) 1547 { 1548 if (exp->name == name) { 1549 return; 1550 } 1551 1552 nbd_export_get(exp); 1553 if (exp->name != NULL) { 1554 g_free(exp->name); 1555 exp->name = NULL; 1556 QTAILQ_REMOVE(&exports, exp, next); 1557 nbd_export_put(exp); 1558 } 1559 if (name != NULL) { 1560 nbd_export_get(exp); 1561 exp->name = g_strdup(name); 1562 QTAILQ_INSERT_TAIL(&exports, exp, next); 1563 } 1564 nbd_export_put(exp); 1565 } 1566 1567 void nbd_export_set_description(NBDExport *exp, const char *description) 1568 { 1569 g_free(exp->description); 1570 exp->description = g_strdup(description); 1571 } 1572 1573 void nbd_export_close(NBDExport *exp) 1574 { 1575 NBDClient *client, *next; 1576 1577 nbd_export_get(exp); 1578 QTAILQ_FOREACH_SAFE(client, &exp->clients, next, next) { 1579 client_close(client, true); 1580 } 1581 nbd_export_set_name(exp, NULL); 1582 nbd_export_set_description(exp, NULL); 1583 nbd_export_put(exp); 1584 } 1585 1586 void nbd_export_remove(NBDExport *exp, NbdServerRemoveMode mode, Error **errp) 1587 { 1588 if (mode == NBD_SERVER_REMOVE_MODE_HARD || QTAILQ_EMPTY(&exp->clients)) { 1589 nbd_export_close(exp); 1590 return; 1591 } 1592 1593 assert(mode == NBD_SERVER_REMOVE_MODE_SAFE); 1594 1595 error_setg(errp, "export '%s' still in use", exp->name); 1596 error_append_hint(errp, "Use mode='hard' to force client disconnect\n"); 1597 } 1598 1599 void nbd_export_get(NBDExport *exp) 1600 { 1601 assert(exp->refcount > 0); 1602 exp->refcount++; 1603 } 1604 1605 void nbd_export_put(NBDExport *exp) 1606 { 1607 assert(exp->refcount > 0); 1608 if (exp->refcount == 1) { 1609 nbd_export_close(exp); 1610 } 1611 1612 /* nbd_export_close() may theoretically reduce refcount to 0. It may happen 1613 * if someone calls nbd_export_put() on named export not through 1614 * nbd_export_set_name() when refcount is 1. So, let's assert that 1615 * it is > 0. 1616 */ 1617 assert(exp->refcount > 0); 1618 if (--exp->refcount == 0) { 1619 assert(exp->name == NULL); 1620 assert(exp->description == NULL); 1621 1622 if (exp->close) { 1623 exp->close(exp); 1624 } 1625 1626 if (exp->blk) { 1627 if (exp->eject_notifier_blk) { 1628 notifier_remove(&exp->eject_notifier); 1629 blk_unref(exp->eject_notifier_blk); 1630 } 1631 blk_remove_aio_context_notifier(exp->blk, blk_aio_attached, 1632 blk_aio_detach, exp); 1633 blk_unref(exp->blk); 1634 exp->blk = NULL; 1635 } 1636 1637 if (exp->export_bitmap) { 1638 bdrv_dirty_bitmap_set_qmp_locked(exp->export_bitmap, false); 1639 g_free(exp->export_bitmap_context); 1640 } 1641 1642 g_free(exp); 1643 } 1644 } 1645 1646 BlockBackend *nbd_export_get_blockdev(NBDExport *exp) 1647 { 1648 return exp->blk; 1649 } 1650 1651 void nbd_export_close_all(void) 1652 { 1653 NBDExport *exp, *next; 1654 1655 QTAILQ_FOREACH_SAFE(exp, &exports, next, next) { 1656 nbd_export_close(exp); 1657 } 1658 } 1659 1660 static int coroutine_fn nbd_co_send_iov(NBDClient *client, struct iovec *iov, 1661 unsigned niov, Error **errp) 1662 { 1663 int ret; 1664 1665 g_assert(qemu_in_coroutine()); 1666 qemu_co_mutex_lock(&client->send_lock); 1667 client->send_coroutine = qemu_coroutine_self(); 1668 1669 ret = qio_channel_writev_all(client->ioc, iov, niov, errp) < 0 ? -EIO : 0; 1670 1671 client->send_coroutine = NULL; 1672 qemu_co_mutex_unlock(&client->send_lock); 1673 1674 return ret; 1675 } 1676 1677 static inline void set_be_simple_reply(NBDSimpleReply *reply, uint64_t error, 1678 uint64_t handle) 1679 { 1680 stl_be_p(&reply->magic, NBD_SIMPLE_REPLY_MAGIC); 1681 stl_be_p(&reply->error, error); 1682 stq_be_p(&reply->handle, handle); 1683 } 1684 1685 static int nbd_co_send_simple_reply(NBDClient *client, 1686 uint64_t handle, 1687 uint32_t error, 1688 void *data, 1689 size_t len, 1690 Error **errp) 1691 { 1692 NBDSimpleReply reply; 1693 int nbd_err = system_errno_to_nbd_errno(error); 1694 struct iovec iov[] = { 1695 {.iov_base = &reply, .iov_len = sizeof(reply)}, 1696 {.iov_base = data, .iov_len = len} 1697 }; 1698 1699 trace_nbd_co_send_simple_reply(handle, nbd_err, nbd_err_lookup(nbd_err), 1700 len); 1701 set_be_simple_reply(&reply, nbd_err, handle); 1702 1703 return nbd_co_send_iov(client, iov, len ? 2 : 1, errp); 1704 } 1705 1706 static inline void set_be_chunk(NBDStructuredReplyChunk *chunk, uint16_t flags, 1707 uint16_t type, uint64_t handle, uint32_t length) 1708 { 1709 stl_be_p(&chunk->magic, NBD_STRUCTURED_REPLY_MAGIC); 1710 stw_be_p(&chunk->flags, flags); 1711 stw_be_p(&chunk->type, type); 1712 stq_be_p(&chunk->handle, handle); 1713 stl_be_p(&chunk->length, length); 1714 } 1715 1716 static int coroutine_fn nbd_co_send_structured_done(NBDClient *client, 1717 uint64_t handle, 1718 Error **errp) 1719 { 1720 NBDStructuredReplyChunk chunk; 1721 struct iovec iov[] = { 1722 {.iov_base = &chunk, .iov_len = sizeof(chunk)}, 1723 }; 1724 1725 trace_nbd_co_send_structured_done(handle); 1726 set_be_chunk(&chunk, NBD_REPLY_FLAG_DONE, NBD_REPLY_TYPE_NONE, handle, 0); 1727 1728 return nbd_co_send_iov(client, iov, 1, errp); 1729 } 1730 1731 static int coroutine_fn nbd_co_send_structured_read(NBDClient *client, 1732 uint64_t handle, 1733 uint64_t offset, 1734 void *data, 1735 size_t size, 1736 bool final, 1737 Error **errp) 1738 { 1739 NBDStructuredReadData chunk; 1740 struct iovec iov[] = { 1741 {.iov_base = &chunk, .iov_len = sizeof(chunk)}, 1742 {.iov_base = data, .iov_len = size} 1743 }; 1744 1745 assert(size); 1746 trace_nbd_co_send_structured_read(handle, offset, data, size); 1747 set_be_chunk(&chunk.h, final ? NBD_REPLY_FLAG_DONE : 0, 1748 NBD_REPLY_TYPE_OFFSET_DATA, handle, 1749 sizeof(chunk) - sizeof(chunk.h) + size); 1750 stq_be_p(&chunk.offset, offset); 1751 1752 return nbd_co_send_iov(client, iov, 2, errp); 1753 } 1754 1755 static int coroutine_fn nbd_co_send_structured_error(NBDClient *client, 1756 uint64_t handle, 1757 uint32_t error, 1758 const char *msg, 1759 Error **errp) 1760 { 1761 NBDStructuredError chunk; 1762 int nbd_err = system_errno_to_nbd_errno(error); 1763 struct iovec iov[] = { 1764 {.iov_base = &chunk, .iov_len = sizeof(chunk)}, 1765 {.iov_base = (char *)msg, .iov_len = msg ? strlen(msg) : 0}, 1766 }; 1767 1768 assert(nbd_err); 1769 trace_nbd_co_send_structured_error(handle, nbd_err, 1770 nbd_err_lookup(nbd_err), msg ? msg : ""); 1771 set_be_chunk(&chunk.h, NBD_REPLY_FLAG_DONE, NBD_REPLY_TYPE_ERROR, handle, 1772 sizeof(chunk) - sizeof(chunk.h) + iov[1].iov_len); 1773 stl_be_p(&chunk.error, nbd_err); 1774 stw_be_p(&chunk.message_length, iov[1].iov_len); 1775 1776 return nbd_co_send_iov(client, iov, 1 + !!iov[1].iov_len, errp); 1777 } 1778 1779 /* Do a sparse read and send the structured reply to the client. 1780 * Returns -errno if sending fails. bdrv_block_status_above() failure is 1781 * reported to the client, at which point this function succeeds. 1782 */ 1783 static int coroutine_fn nbd_co_send_sparse_read(NBDClient *client, 1784 uint64_t handle, 1785 uint64_t offset, 1786 uint8_t *data, 1787 size_t size, 1788 Error **errp) 1789 { 1790 int ret = 0; 1791 NBDExport *exp = client->exp; 1792 size_t progress = 0; 1793 1794 while (progress < size) { 1795 int64_t pnum; 1796 int status = bdrv_block_status_above(blk_bs(exp->blk), NULL, 1797 offset + progress, 1798 size - progress, &pnum, NULL, 1799 NULL); 1800 bool final; 1801 1802 if (status < 0) { 1803 char *msg = g_strdup_printf("unable to check for holes: %s", 1804 strerror(-status)); 1805 1806 ret = nbd_co_send_structured_error(client, handle, -status, msg, 1807 errp); 1808 g_free(msg); 1809 return ret; 1810 } 1811 assert(pnum && pnum <= size - progress); 1812 final = progress + pnum == size; 1813 if (status & BDRV_BLOCK_ZERO) { 1814 NBDStructuredReadHole chunk; 1815 struct iovec iov[] = { 1816 {.iov_base = &chunk, .iov_len = sizeof(chunk)}, 1817 }; 1818 1819 trace_nbd_co_send_structured_read_hole(handle, offset + progress, 1820 pnum); 1821 set_be_chunk(&chunk.h, final ? NBD_REPLY_FLAG_DONE : 0, 1822 NBD_REPLY_TYPE_OFFSET_HOLE, 1823 handle, sizeof(chunk) - sizeof(chunk.h)); 1824 stq_be_p(&chunk.offset, offset + progress); 1825 stl_be_p(&chunk.length, pnum); 1826 ret = nbd_co_send_iov(client, iov, 1, errp); 1827 } else { 1828 ret = blk_pread(exp->blk, offset + progress + exp->dev_offset, 1829 data + progress, pnum); 1830 if (ret < 0) { 1831 error_setg_errno(errp, -ret, "reading from file failed"); 1832 break; 1833 } 1834 ret = nbd_co_send_structured_read(client, handle, offset + progress, 1835 data + progress, pnum, final, 1836 errp); 1837 } 1838 1839 if (ret < 0) { 1840 break; 1841 } 1842 progress += pnum; 1843 } 1844 return ret; 1845 } 1846 1847 static int blockstatus_to_extent_be(BlockDriverState *bs, uint64_t offset, 1848 uint64_t bytes, NBDExtent *extent) 1849 { 1850 uint64_t remaining_bytes = bytes; 1851 1852 while (remaining_bytes) { 1853 uint32_t flags; 1854 int64_t num; 1855 int ret = bdrv_block_status_above(bs, NULL, offset, remaining_bytes, 1856 &num, NULL, NULL); 1857 if (ret < 0) { 1858 return ret; 1859 } 1860 1861 flags = (ret & BDRV_BLOCK_ALLOCATED ? 0 : NBD_STATE_HOLE) | 1862 (ret & BDRV_BLOCK_ZERO ? NBD_STATE_ZERO : 0); 1863 1864 if (remaining_bytes == bytes) { 1865 extent->flags = flags; 1866 } 1867 1868 if (flags != extent->flags) { 1869 break; 1870 } 1871 1872 offset += num; 1873 remaining_bytes -= num; 1874 } 1875 1876 cpu_to_be32s(&extent->flags); 1877 extent->length = cpu_to_be32(bytes - remaining_bytes); 1878 1879 return 0; 1880 } 1881 1882 /* nbd_co_send_extents 1883 * 1884 * @length is only for tracing purposes (and may be smaller or larger 1885 * than the client's original request). @last controls whether 1886 * NBD_REPLY_FLAG_DONE is sent. @extents should already be in 1887 * big-endian format. 1888 */ 1889 static int nbd_co_send_extents(NBDClient *client, uint64_t handle, 1890 NBDExtent *extents, unsigned int nb_extents, 1891 uint64_t length, bool last, 1892 uint32_t context_id, Error **errp) 1893 { 1894 NBDStructuredMeta chunk; 1895 1896 struct iovec iov[] = { 1897 {.iov_base = &chunk, .iov_len = sizeof(chunk)}, 1898 {.iov_base = extents, .iov_len = nb_extents * sizeof(extents[0])} 1899 }; 1900 1901 trace_nbd_co_send_extents(handle, nb_extents, context_id, length, last); 1902 set_be_chunk(&chunk.h, last ? NBD_REPLY_FLAG_DONE : 0, 1903 NBD_REPLY_TYPE_BLOCK_STATUS, 1904 handle, sizeof(chunk) - sizeof(chunk.h) + iov[1].iov_len); 1905 stl_be_p(&chunk.context_id, context_id); 1906 1907 return nbd_co_send_iov(client, iov, 2, errp); 1908 } 1909 1910 /* Get block status from the exported device and send it to the client */ 1911 static int nbd_co_send_block_status(NBDClient *client, uint64_t handle, 1912 BlockDriverState *bs, uint64_t offset, 1913 uint32_t length, bool last, 1914 uint32_t context_id, Error **errp) 1915 { 1916 int ret; 1917 NBDExtent extent; 1918 1919 ret = blockstatus_to_extent_be(bs, offset, length, &extent); 1920 if (ret < 0) { 1921 return nbd_co_send_structured_error( 1922 client, handle, -ret, "can't get block status", errp); 1923 } 1924 1925 return nbd_co_send_extents(client, handle, &extent, 1, 1926 be32_to_cpu(extent.length), last, 1927 context_id, errp); 1928 } 1929 1930 /* 1931 * Populate @extents from a dirty bitmap. Unless @dont_fragment, the 1932 * final extent may exceed the original @length. Store in @length the 1933 * byte length encoded (which may be smaller or larger than the 1934 * original), and return the number of extents used. 1935 */ 1936 static unsigned int bitmap_to_extents(BdrvDirtyBitmap *bitmap, uint64_t offset, 1937 uint64_t *length, NBDExtent *extents, 1938 unsigned int nb_extents, 1939 bool dont_fragment) 1940 { 1941 uint64_t begin = offset, end = offset; 1942 uint64_t overall_end = offset + *length; 1943 unsigned int i = 0; 1944 BdrvDirtyBitmapIter *it; 1945 bool dirty; 1946 1947 bdrv_dirty_bitmap_lock(bitmap); 1948 1949 it = bdrv_dirty_iter_new(bitmap); 1950 dirty = bdrv_get_dirty_locked(NULL, bitmap, offset); 1951 1952 assert(begin < overall_end && nb_extents); 1953 while (begin < overall_end && i < nb_extents) { 1954 if (dirty) { 1955 end = bdrv_dirty_bitmap_next_zero(bitmap, begin); 1956 } else { 1957 bdrv_set_dirty_iter(it, begin); 1958 end = bdrv_dirty_iter_next(it); 1959 } 1960 if (end == -1 || end - begin > UINT32_MAX) { 1961 /* Cap to an aligned value < 4G beyond begin. */ 1962 end = MIN(bdrv_dirty_bitmap_size(bitmap), 1963 begin + UINT32_MAX + 1 - 1964 bdrv_dirty_bitmap_granularity(bitmap)); 1965 } 1966 if (dont_fragment && end > overall_end) { 1967 end = overall_end; 1968 } 1969 1970 extents[i].length = cpu_to_be32(end - begin); 1971 extents[i].flags = cpu_to_be32(dirty ? NBD_STATE_DIRTY : 0); 1972 i++; 1973 begin = end; 1974 dirty = !dirty; 1975 } 1976 1977 bdrv_dirty_iter_free(it); 1978 1979 bdrv_dirty_bitmap_unlock(bitmap); 1980 1981 assert(offset < end); 1982 *length = end - offset; 1983 return i; 1984 } 1985 1986 static int nbd_co_send_bitmap(NBDClient *client, uint64_t handle, 1987 BdrvDirtyBitmap *bitmap, uint64_t offset, 1988 uint32_t length, bool dont_fragment, bool last, 1989 uint32_t context_id, Error **errp) 1990 { 1991 int ret; 1992 unsigned int nb_extents = dont_fragment ? 1 : NBD_MAX_BITMAP_EXTENTS; 1993 NBDExtent *extents = g_new(NBDExtent, nb_extents); 1994 uint64_t final_length = length; 1995 1996 nb_extents = bitmap_to_extents(bitmap, offset, &final_length, extents, 1997 nb_extents, dont_fragment); 1998 1999 ret = nbd_co_send_extents(client, handle, extents, nb_extents, 2000 final_length, last, context_id, errp); 2001 2002 g_free(extents); 2003 2004 return ret; 2005 } 2006 2007 /* nbd_co_receive_request 2008 * Collect a client request. Return 0 if request looks valid, -EIO to drop 2009 * connection right away, and any other negative value to report an error to 2010 * the client (although the caller may still need to disconnect after reporting 2011 * the error). 2012 */ 2013 static int nbd_co_receive_request(NBDRequestData *req, NBDRequest *request, 2014 Error **errp) 2015 { 2016 NBDClient *client = req->client; 2017 int valid_flags; 2018 2019 g_assert(qemu_in_coroutine()); 2020 assert(client->recv_coroutine == qemu_coroutine_self()); 2021 if (nbd_receive_request(client->ioc, request, errp) < 0) { 2022 return -EIO; 2023 } 2024 2025 trace_nbd_co_receive_request_decode_type(request->handle, request->type, 2026 nbd_cmd_lookup(request->type)); 2027 2028 if (request->type != NBD_CMD_WRITE) { 2029 /* No payload, we are ready to read the next request. */ 2030 req->complete = true; 2031 } 2032 2033 if (request->type == NBD_CMD_DISC) { 2034 /* Special case: we're going to disconnect without a reply, 2035 * whether or not flags, from, or len are bogus */ 2036 return -EIO; 2037 } 2038 2039 if (request->type == NBD_CMD_READ || request->type == NBD_CMD_WRITE || 2040 request->type == NBD_CMD_CACHE) 2041 { 2042 if (request->len > NBD_MAX_BUFFER_SIZE) { 2043 error_setg(errp, "len (%" PRIu32" ) is larger than max len (%u)", 2044 request->len, NBD_MAX_BUFFER_SIZE); 2045 return -EINVAL; 2046 } 2047 2048 req->data = blk_try_blockalign(client->exp->blk, request->len); 2049 if (req->data == NULL) { 2050 error_setg(errp, "No memory"); 2051 return -ENOMEM; 2052 } 2053 } 2054 if (request->type == NBD_CMD_WRITE) { 2055 if (nbd_read(client->ioc, req->data, request->len, errp) < 0) { 2056 error_prepend(errp, "reading from socket failed: "); 2057 return -EIO; 2058 } 2059 req->complete = true; 2060 2061 trace_nbd_co_receive_request_payload_received(request->handle, 2062 request->len); 2063 } 2064 2065 /* Sanity checks. */ 2066 if (client->exp->nbdflags & NBD_FLAG_READ_ONLY && 2067 (request->type == NBD_CMD_WRITE || 2068 request->type == NBD_CMD_WRITE_ZEROES || 2069 request->type == NBD_CMD_TRIM)) { 2070 error_setg(errp, "Export is read-only"); 2071 return -EROFS; 2072 } 2073 if (request->from > client->exp->size || 2074 request->from + request->len > client->exp->size) { 2075 error_setg(errp, "operation past EOF; From: %" PRIu64 ", Len: %" PRIu32 2076 ", Size: %" PRIu64, request->from, request->len, 2077 (uint64_t)client->exp->size); 2078 return (request->type == NBD_CMD_WRITE || 2079 request->type == NBD_CMD_WRITE_ZEROES) ? -ENOSPC : -EINVAL; 2080 } 2081 valid_flags = NBD_CMD_FLAG_FUA; 2082 if (request->type == NBD_CMD_READ && client->structured_reply) { 2083 valid_flags |= NBD_CMD_FLAG_DF; 2084 } else if (request->type == NBD_CMD_WRITE_ZEROES) { 2085 valid_flags |= NBD_CMD_FLAG_NO_HOLE; 2086 } else if (request->type == NBD_CMD_BLOCK_STATUS) { 2087 valid_flags |= NBD_CMD_FLAG_REQ_ONE; 2088 } 2089 if (request->flags & ~valid_flags) { 2090 error_setg(errp, "unsupported flags for command %s (got 0x%x)", 2091 nbd_cmd_lookup(request->type), request->flags); 2092 return -EINVAL; 2093 } 2094 2095 return 0; 2096 } 2097 2098 /* Send simple reply without a payload, or a structured error 2099 * @error_msg is ignored if @ret >= 0 2100 * Returns 0 if connection is still live, -errno on failure to talk to client 2101 */ 2102 static coroutine_fn int nbd_send_generic_reply(NBDClient *client, 2103 uint64_t handle, 2104 int ret, 2105 const char *error_msg, 2106 Error **errp) 2107 { 2108 if (client->structured_reply && ret < 0) { 2109 return nbd_co_send_structured_error(client, handle, -ret, error_msg, 2110 errp); 2111 } else { 2112 return nbd_co_send_simple_reply(client, handle, ret < 0 ? -ret : 0, 2113 NULL, 0, errp); 2114 } 2115 } 2116 2117 /* Handle NBD_CMD_READ request. 2118 * Return -errno if sending fails. Other errors are reported directly to the 2119 * client as an error reply. */ 2120 static coroutine_fn int nbd_do_cmd_read(NBDClient *client, NBDRequest *request, 2121 uint8_t *data, Error **errp) 2122 { 2123 int ret; 2124 NBDExport *exp = client->exp; 2125 2126 assert(request->type == NBD_CMD_READ || request->type == NBD_CMD_CACHE); 2127 2128 /* XXX: NBD Protocol only documents use of FUA with WRITE */ 2129 if (request->flags & NBD_CMD_FLAG_FUA) { 2130 ret = blk_co_flush(exp->blk); 2131 if (ret < 0) { 2132 return nbd_send_generic_reply(client, request->handle, ret, 2133 "flush failed", errp); 2134 } 2135 } 2136 2137 if (client->structured_reply && !(request->flags & NBD_CMD_FLAG_DF) && 2138 request->len) { 2139 return nbd_co_send_sparse_read(client, request->handle, request->from, 2140 data, request->len, errp); 2141 } 2142 2143 ret = blk_pread(exp->blk, request->from + exp->dev_offset, data, 2144 request->len); 2145 if (ret < 0 || request->type == NBD_CMD_CACHE) { 2146 return nbd_send_generic_reply(client, request->handle, ret, 2147 "reading from file failed", errp); 2148 } 2149 2150 if (client->structured_reply) { 2151 if (request->len) { 2152 return nbd_co_send_structured_read(client, request->handle, 2153 request->from, data, 2154 request->len, true, errp); 2155 } else { 2156 return nbd_co_send_structured_done(client, request->handle, errp); 2157 } 2158 } else { 2159 return nbd_co_send_simple_reply(client, request->handle, 0, 2160 data, request->len, errp); 2161 } 2162 } 2163 2164 /* Handle NBD request. 2165 * Return -errno if sending fails. Other errors are reported directly to the 2166 * client as an error reply. */ 2167 static coroutine_fn int nbd_handle_request(NBDClient *client, 2168 NBDRequest *request, 2169 uint8_t *data, Error **errp) 2170 { 2171 int ret; 2172 int flags; 2173 NBDExport *exp = client->exp; 2174 char *msg; 2175 2176 switch (request->type) { 2177 case NBD_CMD_READ: 2178 case NBD_CMD_CACHE: 2179 return nbd_do_cmd_read(client, request, data, errp); 2180 2181 case NBD_CMD_WRITE: 2182 flags = 0; 2183 if (request->flags & NBD_CMD_FLAG_FUA) { 2184 flags |= BDRV_REQ_FUA; 2185 } 2186 ret = blk_pwrite(exp->blk, request->from + exp->dev_offset, 2187 data, request->len, flags); 2188 return nbd_send_generic_reply(client, request->handle, ret, 2189 "writing to file failed", errp); 2190 2191 case NBD_CMD_WRITE_ZEROES: 2192 flags = 0; 2193 if (request->flags & NBD_CMD_FLAG_FUA) { 2194 flags |= BDRV_REQ_FUA; 2195 } 2196 if (!(request->flags & NBD_CMD_FLAG_NO_HOLE)) { 2197 flags |= BDRV_REQ_MAY_UNMAP; 2198 } 2199 ret = blk_pwrite_zeroes(exp->blk, request->from + exp->dev_offset, 2200 request->len, flags); 2201 return nbd_send_generic_reply(client, request->handle, ret, 2202 "writing to file failed", errp); 2203 2204 case NBD_CMD_DISC: 2205 /* unreachable, thanks to special case in nbd_co_receive_request() */ 2206 abort(); 2207 2208 case NBD_CMD_FLUSH: 2209 ret = blk_co_flush(exp->blk); 2210 return nbd_send_generic_reply(client, request->handle, ret, 2211 "flush failed", errp); 2212 2213 case NBD_CMD_TRIM: 2214 ret = blk_co_pdiscard(exp->blk, request->from + exp->dev_offset, 2215 request->len); 2216 if (ret == 0 && request->flags & NBD_CMD_FLAG_FUA) { 2217 ret = blk_co_flush(exp->blk); 2218 } 2219 return nbd_send_generic_reply(client, request->handle, ret, 2220 "discard failed", errp); 2221 2222 case NBD_CMD_BLOCK_STATUS: 2223 if (!request->len) { 2224 return nbd_send_generic_reply(client, request->handle, -EINVAL, 2225 "need non-zero length", errp); 2226 } 2227 if (client->export_meta.valid && 2228 (client->export_meta.base_allocation || 2229 client->export_meta.bitmap)) 2230 { 2231 if (client->export_meta.base_allocation) { 2232 ret = nbd_co_send_block_status(client, request->handle, 2233 blk_bs(exp->blk), request->from, 2234 request->len, 2235 !client->export_meta.bitmap, 2236 NBD_META_ID_BASE_ALLOCATION, 2237 errp); 2238 if (ret < 0) { 2239 return ret; 2240 } 2241 } 2242 2243 if (client->export_meta.bitmap) { 2244 ret = nbd_co_send_bitmap(client, request->handle, 2245 client->exp->export_bitmap, 2246 request->from, request->len, 2247 request->flags & NBD_CMD_FLAG_REQ_ONE, 2248 true, NBD_META_ID_DIRTY_BITMAP, errp); 2249 if (ret < 0) { 2250 return ret; 2251 } 2252 } 2253 2254 return ret; 2255 } else { 2256 return nbd_send_generic_reply(client, request->handle, -EINVAL, 2257 "CMD_BLOCK_STATUS not negotiated", 2258 errp); 2259 } 2260 2261 default: 2262 msg = g_strdup_printf("invalid request type (%" PRIu32 ") received", 2263 request->type); 2264 ret = nbd_send_generic_reply(client, request->handle, -EINVAL, msg, 2265 errp); 2266 g_free(msg); 2267 return ret; 2268 } 2269 } 2270 2271 /* Owns a reference to the NBDClient passed as opaque. */ 2272 static coroutine_fn void nbd_trip(void *opaque) 2273 { 2274 NBDClient *client = opaque; 2275 NBDRequestData *req; 2276 NBDRequest request = { 0 }; /* GCC thinks it can be used uninitialized */ 2277 int ret; 2278 Error *local_err = NULL; 2279 2280 trace_nbd_trip(); 2281 if (client->closing) { 2282 nbd_client_put(client); 2283 return; 2284 } 2285 2286 req = nbd_request_get(client); 2287 ret = nbd_co_receive_request(req, &request, &local_err); 2288 client->recv_coroutine = NULL; 2289 2290 if (client->closing) { 2291 /* 2292 * The client may be closed when we are blocked in 2293 * nbd_co_receive_request() 2294 */ 2295 goto done; 2296 } 2297 2298 nbd_client_receive_next_request(client); 2299 if (ret == -EIO) { 2300 goto disconnect; 2301 } 2302 2303 if (ret < 0) { 2304 /* It wans't -EIO, so, according to nbd_co_receive_request() 2305 * semantics, we should return the error to the client. */ 2306 Error *export_err = local_err; 2307 2308 local_err = NULL; 2309 ret = nbd_send_generic_reply(client, request.handle, -EINVAL, 2310 error_get_pretty(export_err), &local_err); 2311 error_free(export_err); 2312 } else { 2313 ret = nbd_handle_request(client, &request, req->data, &local_err); 2314 } 2315 if (ret < 0) { 2316 error_prepend(&local_err, "Failed to send reply: "); 2317 goto disconnect; 2318 } 2319 2320 /* We must disconnect after NBD_CMD_WRITE if we did not 2321 * read the payload. 2322 */ 2323 if (!req->complete) { 2324 error_setg(&local_err, "Request handling failed in intermediate state"); 2325 goto disconnect; 2326 } 2327 2328 done: 2329 nbd_request_put(req); 2330 nbd_client_put(client); 2331 return; 2332 2333 disconnect: 2334 if (local_err) { 2335 error_reportf_err(local_err, "Disconnect client, due to: "); 2336 } 2337 nbd_request_put(req); 2338 client_close(client, true); 2339 nbd_client_put(client); 2340 } 2341 2342 static void nbd_client_receive_next_request(NBDClient *client) 2343 { 2344 if (!client->recv_coroutine && client->nb_requests < MAX_NBD_REQUESTS) { 2345 nbd_client_get(client); 2346 client->recv_coroutine = qemu_coroutine_create(nbd_trip, client); 2347 aio_co_schedule(client->exp->ctx, client->recv_coroutine); 2348 } 2349 } 2350 2351 static coroutine_fn void nbd_co_client_start(void *opaque) 2352 { 2353 NBDClient *client = opaque; 2354 NBDExport *exp = client->exp; 2355 Error *local_err = NULL; 2356 2357 if (exp) { 2358 nbd_export_get(exp); 2359 QTAILQ_INSERT_TAIL(&exp->clients, client, next); 2360 } 2361 qemu_co_mutex_init(&client->send_lock); 2362 2363 if (nbd_negotiate(client, &local_err)) { 2364 if (local_err) { 2365 error_report_err(local_err); 2366 } 2367 client_close(client, false); 2368 return; 2369 } 2370 2371 nbd_client_receive_next_request(client); 2372 } 2373 2374 /* 2375 * Create a new client listener on the given export @exp, using the 2376 * given channel @sioc. Begin servicing it in a coroutine. When the 2377 * connection closes, call @close_fn with an indication of whether the 2378 * client completed negotiation. 2379 */ 2380 void nbd_client_new(NBDExport *exp, 2381 QIOChannelSocket *sioc, 2382 QCryptoTLSCreds *tlscreds, 2383 const char *tlsaclname, 2384 void (*close_fn)(NBDClient *, bool)) 2385 { 2386 NBDClient *client; 2387 Coroutine *co; 2388 2389 client = g_new0(NBDClient, 1); 2390 client->refcount = 1; 2391 client->exp = exp; 2392 client->tlscreds = tlscreds; 2393 if (tlscreds) { 2394 object_ref(OBJECT(client->tlscreds)); 2395 } 2396 client->tlsaclname = g_strdup(tlsaclname); 2397 client->sioc = sioc; 2398 object_ref(OBJECT(client->sioc)); 2399 client->ioc = QIO_CHANNEL(sioc); 2400 object_ref(OBJECT(client->ioc)); 2401 client->close_fn = close_fn; 2402 2403 co = qemu_coroutine_create(nbd_co_client_start, client); 2404 qemu_coroutine_enter(co); 2405 } 2406 2407 void nbd_export_bitmap(NBDExport *exp, const char *bitmap, 2408 const char *bitmap_export_name, Error **errp) 2409 { 2410 BdrvDirtyBitmap *bm = NULL; 2411 BlockDriverState *bs = blk_bs(exp->blk); 2412 2413 if (exp->export_bitmap) { 2414 error_setg(errp, "Export bitmap is already set"); 2415 return; 2416 } 2417 2418 while (true) { 2419 bm = bdrv_find_dirty_bitmap(bs, bitmap); 2420 if (bm != NULL || bs->backing == NULL) { 2421 break; 2422 } 2423 2424 bs = bs->backing->bs; 2425 } 2426 2427 if (bm == NULL) { 2428 error_setg(errp, "Bitmap '%s' is not found", bitmap); 2429 return; 2430 } 2431 2432 if (bdrv_dirty_bitmap_enabled(bm)) { 2433 error_setg(errp, "Bitmap '%s' is enabled", bitmap); 2434 return; 2435 } 2436 2437 if (bdrv_dirty_bitmap_qmp_locked(bm)) { 2438 error_setg(errp, "Bitmap '%s' is locked", bitmap); 2439 return; 2440 } 2441 2442 bdrv_dirty_bitmap_set_qmp_locked(bm, true); 2443 exp->export_bitmap = bm; 2444 exp->export_bitmap_context = 2445 g_strdup_printf("qemu:dirty-bitmap:%s", bitmap_export_name); 2446 } 2447