1 /* 2 * Copyright Red Hat 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 22 #include "block/block_int.h" 23 #include "block/export.h" 24 #include "block/dirty-bitmap.h" 25 #include "qapi/error.h" 26 #include "qemu/queue.h" 27 #include "trace.h" 28 #include "nbd-internal.h" 29 #include "qemu/units.h" 30 #include "qemu/memalign.h" 31 32 #define NBD_META_ID_BASE_ALLOCATION 0 33 #define NBD_META_ID_ALLOCATION_DEPTH 1 34 /* Dirty bitmaps use 'NBD_META_ID_DIRTY_BITMAP + i', so keep this id last. */ 35 #define NBD_META_ID_DIRTY_BITMAP 2 36 37 /* 38 * NBD_MAX_BLOCK_STATUS_EXTENTS: 1 MiB of extents data. An empirical 39 * constant. If an increase is needed, note that the NBD protocol 40 * recommends no larger than 32 mb, so that the client won't consider 41 * the reply as a denial of service attack. 42 */ 43 #define NBD_MAX_BLOCK_STATUS_EXTENTS (1 * MiB / 8) 44 45 static int system_errno_to_nbd_errno(int err) 46 { 47 switch (err) { 48 case 0: 49 return NBD_SUCCESS; 50 case EPERM: 51 case EROFS: 52 return NBD_EPERM; 53 case EIO: 54 return NBD_EIO; 55 case ENOMEM: 56 return NBD_ENOMEM; 57 #ifdef EDQUOT 58 case EDQUOT: 59 #endif 60 case EFBIG: 61 case ENOSPC: 62 return NBD_ENOSPC; 63 case EOVERFLOW: 64 return NBD_EOVERFLOW; 65 case ENOTSUP: 66 #if ENOTSUP != EOPNOTSUPP 67 case EOPNOTSUPP: 68 #endif 69 return NBD_ENOTSUP; 70 case ESHUTDOWN: 71 return NBD_ESHUTDOWN; 72 case EINVAL: 73 default: 74 return NBD_EINVAL; 75 } 76 } 77 78 /* Definitions for opaque data types */ 79 80 typedef struct NBDRequestData NBDRequestData; 81 82 struct NBDRequestData { 83 NBDClient *client; 84 uint8_t *data; 85 bool complete; 86 }; 87 88 struct NBDExport { 89 BlockExport common; 90 91 char *name; 92 char *description; 93 uint64_t size; 94 uint16_t nbdflags; 95 QTAILQ_HEAD(, NBDClient) clients; 96 QTAILQ_ENTRY(NBDExport) next; 97 98 BlockBackend *eject_notifier_blk; 99 Notifier eject_notifier; 100 101 bool allocation_depth; 102 BdrvDirtyBitmap **export_bitmaps; 103 size_t nr_export_bitmaps; 104 }; 105 106 static QTAILQ_HEAD(, NBDExport) exports = QTAILQ_HEAD_INITIALIZER(exports); 107 108 /* 109 * NBDMetaContexts represents a list of meta contexts in use, 110 * as selected by NBD_OPT_SET_META_CONTEXT. Also used for 111 * NBD_OPT_LIST_META_CONTEXT. 112 */ 113 struct NBDMetaContexts { 114 const NBDExport *exp; /* associated export */ 115 size_t count; /* number of negotiated contexts */ 116 bool base_allocation; /* export base:allocation context (block status) */ 117 bool allocation_depth; /* export qemu:allocation-depth */ 118 bool *bitmaps; /* 119 * export qemu:dirty-bitmap:<export bitmap name>, 120 * sized by exp->nr_export_bitmaps 121 */ 122 }; 123 124 struct NBDClient { 125 int refcount; /* atomic */ 126 void (*close_fn)(NBDClient *client, bool negotiated); 127 void *owner; 128 129 QemuMutex lock; 130 131 NBDExport *exp; 132 QCryptoTLSCreds *tlscreds; 133 char *tlsauthz; 134 uint32_t handshake_max_secs; 135 QIOChannelSocket *sioc; /* The underlying data channel */ 136 QIOChannel *ioc; /* The current I/O channel which may differ (eg TLS) */ 137 138 Coroutine *recv_coroutine; /* protected by lock */ 139 140 CoMutex send_lock; 141 Coroutine *send_coroutine; 142 143 bool read_yielding; /* protected by lock */ 144 bool quiescing; /* protected by lock */ 145 146 QTAILQ_ENTRY(NBDClient) next; 147 int nb_requests; /* protected by lock */ 148 bool closing; /* protected by lock */ 149 150 uint32_t check_align; /* If non-zero, check for aligned client requests */ 151 152 NBDMode mode; 153 NBDMetaContexts contexts; /* Negotiated meta contexts */ 154 155 uint32_t opt; /* Current option being negotiated */ 156 uint32_t optlen; /* remaining length of data in ioc for the option being 157 negotiated now */ 158 }; 159 160 static void nbd_client_receive_next_request(NBDClient *client); 161 162 /* Basic flow for negotiation 163 164 Server Client 165 Negotiate 166 167 or 168 169 Server Client 170 Negotiate #1 171 Option 172 Negotiate #2 173 174 ---- 175 176 followed by 177 178 Server Client 179 Request 180 Response 181 Request 182 Response 183 ... 184 ... 185 Request (type == 2) 186 187 */ 188 189 static inline void set_be_option_rep(NBDOptionReply *rep, uint32_t option, 190 uint32_t type, uint32_t length) 191 { 192 stq_be_p(&rep->magic, NBD_REP_MAGIC); 193 stl_be_p(&rep->option, option); 194 stl_be_p(&rep->type, type); 195 stl_be_p(&rep->length, length); 196 } 197 198 /* Send a reply header, including length, but no payload. 199 * Return -errno on error, 0 on success. */ 200 static coroutine_fn int 201 nbd_negotiate_send_rep_len(NBDClient *client, uint32_t type, 202 uint32_t len, Error **errp) 203 { 204 NBDOptionReply rep; 205 206 trace_nbd_negotiate_send_rep_len(client->opt, nbd_opt_lookup(client->opt), 207 type, nbd_rep_lookup(type), len); 208 209 assert(len < NBD_MAX_BUFFER_SIZE); 210 211 set_be_option_rep(&rep, client->opt, type, len); 212 return nbd_write(client->ioc, &rep, sizeof(rep), errp); 213 } 214 215 /* Send a reply header with default 0 length. 216 * Return -errno on error, 0 on success. */ 217 static coroutine_fn int 218 nbd_negotiate_send_rep(NBDClient *client, uint32_t type, Error **errp) 219 { 220 return nbd_negotiate_send_rep_len(client, type, 0, errp); 221 } 222 223 /* Send an error reply. 224 * Return -errno on error, 0 on success. */ 225 static coroutine_fn int G_GNUC_PRINTF(4, 0) 226 nbd_negotiate_send_rep_verr(NBDClient *client, uint32_t type, 227 Error **errp, const char *fmt, va_list va) 228 { 229 ERRP_GUARD(); 230 g_autofree char *msg = NULL; 231 int ret; 232 size_t len; 233 234 msg = g_strdup_vprintf(fmt, va); 235 len = strlen(msg); 236 assert(len < NBD_MAX_STRING_SIZE); 237 trace_nbd_negotiate_send_rep_err(msg); 238 ret = nbd_negotiate_send_rep_len(client, type, len, errp); 239 if (ret < 0) { 240 return ret; 241 } 242 if (nbd_write(client->ioc, msg, len, errp) < 0) { 243 error_prepend(errp, "write failed (error message): "); 244 return -EIO; 245 } 246 247 return 0; 248 } 249 250 /* 251 * Return a malloc'd copy of @name suitable for use in an error reply. 252 */ 253 static char * 254 nbd_sanitize_name(const char *name) 255 { 256 if (strnlen(name, 80) < 80) { 257 return g_strdup(name); 258 } 259 /* XXX Should we also try to sanitize any control characters? */ 260 return g_strdup_printf("%.80s...", name); 261 } 262 263 /* Send an error reply. 264 * Return -errno on error, 0 on success. */ 265 static coroutine_fn int G_GNUC_PRINTF(4, 5) 266 nbd_negotiate_send_rep_err(NBDClient *client, uint32_t type, 267 Error **errp, const char *fmt, ...) 268 { 269 va_list va; 270 int ret; 271 272 va_start(va, fmt); 273 ret = nbd_negotiate_send_rep_verr(client, type, errp, fmt, va); 274 va_end(va); 275 return ret; 276 } 277 278 /* Drop remainder of the current option, and send a reply with the 279 * given error type and message. Return -errno on read or write 280 * failure; or 0 if connection is still live. */ 281 static coroutine_fn int G_GNUC_PRINTF(4, 0) 282 nbd_opt_vdrop(NBDClient *client, uint32_t type, Error **errp, 283 const char *fmt, va_list va) 284 { 285 int ret = nbd_drop(client->ioc, client->optlen, errp); 286 287 client->optlen = 0; 288 if (!ret) { 289 ret = nbd_negotiate_send_rep_verr(client, type, errp, fmt, va); 290 } 291 return ret; 292 } 293 294 static coroutine_fn int G_GNUC_PRINTF(4, 5) 295 nbd_opt_drop(NBDClient *client, uint32_t type, Error **errp, 296 const char *fmt, ...) 297 { 298 int ret; 299 va_list va; 300 301 va_start(va, fmt); 302 ret = nbd_opt_vdrop(client, type, errp, fmt, va); 303 va_end(va); 304 305 return ret; 306 } 307 308 static coroutine_fn int G_GNUC_PRINTF(3, 4) 309 nbd_opt_invalid(NBDClient *client, Error **errp, const char *fmt, ...) 310 { 311 int ret; 312 va_list va; 313 314 va_start(va, fmt); 315 ret = nbd_opt_vdrop(client, NBD_REP_ERR_INVALID, errp, fmt, va); 316 va_end(va); 317 318 return ret; 319 } 320 321 /* Read size bytes from the unparsed payload of the current option. 322 * If @check_nul, require that no NUL bytes appear in buffer. 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 static coroutine_fn int 326 nbd_opt_read(NBDClient *client, void *buffer, size_t size, 327 bool check_nul, Error **errp) 328 { 329 if (size > client->optlen) { 330 return nbd_opt_invalid(client, errp, 331 "Inconsistent lengths in option %s", 332 nbd_opt_lookup(client->opt)); 333 } 334 client->optlen -= size; 335 if (qio_channel_read_all(client->ioc, buffer, size, errp) < 0) { 336 return -EIO; 337 } 338 339 if (check_nul && strnlen(buffer, size) != size) { 340 return nbd_opt_invalid(client, errp, 341 "Unexpected embedded NUL in option %s", 342 nbd_opt_lookup(client->opt)); 343 } 344 return 1; 345 } 346 347 /* Drop size bytes from the unparsed payload of the current option. 348 * Return -errno on I/O error, 0 if option was completely handled by 349 * sending a reply about inconsistent lengths, or 1 on success. */ 350 static coroutine_fn int 351 nbd_opt_skip(NBDClient *client, size_t size, Error **errp) 352 { 353 if (size > client->optlen) { 354 return nbd_opt_invalid(client, errp, 355 "Inconsistent lengths in option %s", 356 nbd_opt_lookup(client->opt)); 357 } 358 client->optlen -= size; 359 return nbd_drop(client->ioc, size, errp) < 0 ? -EIO : 1; 360 } 361 362 /* nbd_opt_read_name 363 * 364 * Read a string with the format: 365 * uint32_t len (<= NBD_MAX_STRING_SIZE) 366 * len bytes string (not 0-terminated) 367 * 368 * On success, @name will be allocated. 369 * If @length is non-null, it will be set to the actual string length. 370 * 371 * Return -errno on I/O error, 0 if option was completely handled by 372 * sending a reply about inconsistent lengths, or 1 on success. 373 */ 374 static coroutine_fn int 375 nbd_opt_read_name(NBDClient *client, char **name, uint32_t *length, 376 Error **errp) 377 { 378 int ret; 379 uint32_t len; 380 g_autofree char *local_name = NULL; 381 382 *name = NULL; 383 ret = nbd_opt_read(client, &len, sizeof(len), false, errp); 384 if (ret <= 0) { 385 return ret; 386 } 387 len = cpu_to_be32(len); 388 389 if (len > NBD_MAX_STRING_SIZE) { 390 return nbd_opt_invalid(client, errp, 391 "Invalid name length: %" PRIu32, len); 392 } 393 394 local_name = g_malloc(len + 1); 395 ret = nbd_opt_read(client, local_name, len, true, errp); 396 if (ret <= 0) { 397 return ret; 398 } 399 local_name[len] = '\0'; 400 401 if (length) { 402 *length = len; 403 } 404 *name = g_steal_pointer(&local_name); 405 406 return 1; 407 } 408 409 /* Send a single NBD_REP_SERVER reply to NBD_OPT_LIST, including payload. 410 * Return -errno on error, 0 on success. */ 411 static coroutine_fn int 412 nbd_negotiate_send_rep_list(NBDClient *client, NBDExport *exp, Error **errp) 413 { 414 ERRP_GUARD(); 415 size_t name_len, desc_len; 416 uint32_t len; 417 const char *name = exp->name ? exp->name : ""; 418 const char *desc = exp->description ? exp->description : ""; 419 QIOChannel *ioc = client->ioc; 420 int ret; 421 422 trace_nbd_negotiate_send_rep_list(name, desc); 423 name_len = strlen(name); 424 desc_len = strlen(desc); 425 assert(name_len <= NBD_MAX_STRING_SIZE && desc_len <= NBD_MAX_STRING_SIZE); 426 len = name_len + desc_len + sizeof(len); 427 ret = nbd_negotiate_send_rep_len(client, NBD_REP_SERVER, len, errp); 428 if (ret < 0) { 429 return ret; 430 } 431 432 len = cpu_to_be32(name_len); 433 if (nbd_write(ioc, &len, sizeof(len), errp) < 0) { 434 error_prepend(errp, "write failed (name length): "); 435 return -EINVAL; 436 } 437 438 if (nbd_write(ioc, name, name_len, errp) < 0) { 439 error_prepend(errp, "write failed (name buffer): "); 440 return -EINVAL; 441 } 442 443 if (nbd_write(ioc, desc, desc_len, errp) < 0) { 444 error_prepend(errp, "write failed (description buffer): "); 445 return -EINVAL; 446 } 447 448 return 0; 449 } 450 451 /* Process the NBD_OPT_LIST command, with a potential series of replies. 452 * Return -errno on error, 0 on success. */ 453 static coroutine_fn int 454 nbd_negotiate_handle_list(NBDClient *client, Error **errp) 455 { 456 NBDExport *exp; 457 assert(client->opt == NBD_OPT_LIST); 458 459 /* For each export, send a NBD_REP_SERVER reply. */ 460 QTAILQ_FOREACH(exp, &exports, next) { 461 if (nbd_negotiate_send_rep_list(client, exp, errp)) { 462 return -EINVAL; 463 } 464 } 465 /* Finish with a NBD_REP_ACK. */ 466 return nbd_negotiate_send_rep(client, NBD_REP_ACK, errp); 467 } 468 469 static coroutine_fn void 470 nbd_check_meta_export(NBDClient *client, NBDExport *exp) 471 { 472 if (exp != client->contexts.exp) { 473 client->contexts.count = 0; 474 } 475 } 476 477 /* Send a reply to NBD_OPT_EXPORT_NAME. 478 * Return -errno on error, 0 on success. */ 479 static coroutine_fn int 480 nbd_negotiate_handle_export_name(NBDClient *client, bool no_zeroes, 481 Error **errp) 482 { 483 ERRP_GUARD(); 484 g_autofree char *name = NULL; 485 char buf[NBD_REPLY_EXPORT_NAME_SIZE] = ""; 486 size_t len; 487 int ret; 488 uint16_t myflags; 489 490 /* Client sends: 491 [20 .. xx] export name (length bytes) 492 Server replies: 493 [ 0 .. 7] size 494 [ 8 .. 9] export flags 495 [10 .. 133] reserved (0) [unless no_zeroes] 496 */ 497 trace_nbd_negotiate_handle_export_name(); 498 if (client->mode >= NBD_MODE_EXTENDED) { 499 error_setg(errp, "Extended headers already negotiated"); 500 return -EINVAL; 501 } 502 if (client->optlen > NBD_MAX_STRING_SIZE) { 503 error_setg(errp, "Bad length received"); 504 return -EINVAL; 505 } 506 name = g_malloc(client->optlen + 1); 507 if (nbd_read(client->ioc, name, client->optlen, "export name", errp) < 0) { 508 return -EIO; 509 } 510 name[client->optlen] = '\0'; 511 client->optlen = 0; 512 513 trace_nbd_negotiate_handle_export_name_request(name); 514 515 client->exp = nbd_export_find(name); 516 if (!client->exp) { 517 error_setg(errp, "export not found"); 518 return -EINVAL; 519 } 520 nbd_check_meta_export(client, client->exp); 521 522 myflags = client->exp->nbdflags; 523 if (client->mode >= NBD_MODE_STRUCTURED) { 524 myflags |= NBD_FLAG_SEND_DF; 525 } 526 if (client->mode >= NBD_MODE_EXTENDED && client->contexts.count) { 527 myflags |= NBD_FLAG_BLOCK_STAT_PAYLOAD; 528 } 529 trace_nbd_negotiate_new_style_size_flags(client->exp->size, myflags); 530 stq_be_p(buf, client->exp->size); 531 stw_be_p(buf + 8, myflags); 532 len = no_zeroes ? 10 : sizeof(buf); 533 ret = nbd_write(client->ioc, buf, len, errp); 534 if (ret < 0) { 535 error_prepend(errp, "write failed: "); 536 return ret; 537 } 538 539 QTAILQ_INSERT_TAIL(&client->exp->clients, client, next); 540 blk_exp_ref(&client->exp->common); 541 542 return 0; 543 } 544 545 /* Send a single NBD_REP_INFO, with a buffer @buf of @length bytes. 546 * The buffer does NOT include the info type prefix. 547 * Return -errno on error, 0 if ready to send more. */ 548 static coroutine_fn int 549 nbd_negotiate_send_info(NBDClient *client, uint16_t info, uint32_t length, 550 void *buf, Error **errp) 551 { 552 int rc; 553 554 trace_nbd_negotiate_send_info(info, nbd_info_lookup(info), length); 555 rc = nbd_negotiate_send_rep_len(client, NBD_REP_INFO, 556 sizeof(info) + length, errp); 557 if (rc < 0) { 558 return rc; 559 } 560 info = cpu_to_be16(info); 561 if (nbd_write(client->ioc, &info, sizeof(info), errp) < 0) { 562 return -EIO; 563 } 564 if (nbd_write(client->ioc, buf, length, errp) < 0) { 565 return -EIO; 566 } 567 return 0; 568 } 569 570 /* nbd_reject_length: Handle any unexpected payload. 571 * @fatal requests that we quit talking to the client, even if we are able 572 * to successfully send an error reply. 573 * Return: 574 * -errno transmission error occurred or @fatal was requested, errp is set 575 * 0 error message successfully sent to client, errp is not set 576 */ 577 static coroutine_fn int 578 nbd_reject_length(NBDClient *client, bool fatal, Error **errp) 579 { 580 int ret; 581 582 assert(client->optlen); 583 ret = nbd_opt_invalid(client, errp, "option '%s' has unexpected length", 584 nbd_opt_lookup(client->opt)); 585 if (fatal && !ret) { 586 error_setg(errp, "option '%s' has unexpected length", 587 nbd_opt_lookup(client->opt)); 588 return -EINVAL; 589 } 590 return ret; 591 } 592 593 /* Handle NBD_OPT_INFO and NBD_OPT_GO. 594 * Return -errno on error, 0 if ready for next option, and 1 to move 595 * into transmission phase. */ 596 static coroutine_fn int 597 nbd_negotiate_handle_info(NBDClient *client, Error **errp) 598 { 599 int rc; 600 g_autofree char *name = NULL; 601 NBDExport *exp; 602 uint16_t requests; 603 uint16_t request; 604 uint32_t namelen = 0; 605 bool sendname = false; 606 bool blocksize = false; 607 uint32_t sizes[3]; 608 char buf[sizeof(uint64_t) + sizeof(uint16_t)]; 609 uint32_t check_align = 0; 610 uint16_t myflags; 611 612 /* Client sends: 613 4 bytes: L, name length (can be 0) 614 L bytes: export name 615 2 bytes: N, number of requests (can be 0) 616 N * 2 bytes: N requests 617 */ 618 rc = nbd_opt_read_name(client, &name, &namelen, errp); 619 if (rc <= 0) { 620 return rc; 621 } 622 trace_nbd_negotiate_handle_export_name_request(name); 623 624 rc = nbd_opt_read(client, &requests, sizeof(requests), false, errp); 625 if (rc <= 0) { 626 return rc; 627 } 628 requests = be16_to_cpu(requests); 629 trace_nbd_negotiate_handle_info_requests(requests); 630 while (requests--) { 631 rc = nbd_opt_read(client, &request, sizeof(request), false, errp); 632 if (rc <= 0) { 633 return rc; 634 } 635 request = be16_to_cpu(request); 636 trace_nbd_negotiate_handle_info_request(request, 637 nbd_info_lookup(request)); 638 /* We care about NBD_INFO_NAME and NBD_INFO_BLOCK_SIZE; 639 * everything else is either a request we don't know or 640 * something we send regardless of request */ 641 switch (request) { 642 case NBD_INFO_NAME: 643 sendname = true; 644 break; 645 case NBD_INFO_BLOCK_SIZE: 646 blocksize = true; 647 break; 648 } 649 } 650 if (client->optlen) { 651 return nbd_reject_length(client, false, errp); 652 } 653 654 exp = nbd_export_find(name); 655 if (!exp) { 656 g_autofree char *sane_name = nbd_sanitize_name(name); 657 658 return nbd_negotiate_send_rep_err(client, NBD_REP_ERR_UNKNOWN, 659 errp, "export '%s' not present", 660 sane_name); 661 } 662 if (client->opt == NBD_OPT_GO) { 663 nbd_check_meta_export(client, exp); 664 } 665 666 /* Don't bother sending NBD_INFO_NAME unless client requested it */ 667 if (sendname) { 668 rc = nbd_negotiate_send_info(client, NBD_INFO_NAME, namelen, name, 669 errp); 670 if (rc < 0) { 671 return rc; 672 } 673 } 674 675 /* Send NBD_INFO_DESCRIPTION only if available, regardless of 676 * client request */ 677 if (exp->description) { 678 size_t len = strlen(exp->description); 679 680 assert(len <= NBD_MAX_STRING_SIZE); 681 rc = nbd_negotiate_send_info(client, NBD_INFO_DESCRIPTION, 682 len, exp->description, errp); 683 if (rc < 0) { 684 return rc; 685 } 686 } 687 688 /* Send NBD_INFO_BLOCK_SIZE always, but tweak the minimum size 689 * according to whether the client requested it, and according to 690 * whether this is OPT_INFO or OPT_GO. */ 691 /* minimum - 1 for back-compat, or actual if client will obey it. */ 692 if (client->opt == NBD_OPT_INFO || blocksize) { 693 check_align = sizes[0] = blk_get_request_alignment(exp->common.blk); 694 } else { 695 sizes[0] = 1; 696 } 697 assert(sizes[0] <= NBD_MAX_BUFFER_SIZE); 698 /* preferred - Hard-code to 4096 for now. 699 * TODO: is blk_bs(blk)->bl.opt_transfer appropriate? */ 700 sizes[1] = MAX(4096, sizes[0]); 701 /* maximum - At most 32M, but smaller as appropriate. */ 702 sizes[2] = MIN(blk_get_max_transfer(exp->common.blk), NBD_MAX_BUFFER_SIZE); 703 trace_nbd_negotiate_handle_info_block_size(sizes[0], sizes[1], sizes[2]); 704 sizes[0] = cpu_to_be32(sizes[0]); 705 sizes[1] = cpu_to_be32(sizes[1]); 706 sizes[2] = cpu_to_be32(sizes[2]); 707 rc = nbd_negotiate_send_info(client, NBD_INFO_BLOCK_SIZE, 708 sizeof(sizes), sizes, errp); 709 if (rc < 0) { 710 return rc; 711 } 712 713 /* Send NBD_INFO_EXPORT always */ 714 myflags = exp->nbdflags; 715 if (client->mode >= NBD_MODE_STRUCTURED) { 716 myflags |= NBD_FLAG_SEND_DF; 717 } 718 if (client->mode >= NBD_MODE_EXTENDED && 719 (client->contexts.count || client->opt == NBD_OPT_INFO)) { 720 myflags |= NBD_FLAG_BLOCK_STAT_PAYLOAD; 721 } 722 trace_nbd_negotiate_new_style_size_flags(exp->size, myflags); 723 stq_be_p(buf, exp->size); 724 stw_be_p(buf + 8, myflags); 725 rc = nbd_negotiate_send_info(client, NBD_INFO_EXPORT, 726 sizeof(buf), buf, errp); 727 if (rc < 0) { 728 return rc; 729 } 730 731 /* 732 * If the client is just asking for NBD_OPT_INFO, but forgot to 733 * request block sizes in a situation that would impact 734 * performance, then return an error. But for NBD_OPT_GO, we 735 * tolerate all clients, regardless of alignments. 736 */ 737 if (client->opt == NBD_OPT_INFO && !blocksize && 738 blk_get_request_alignment(exp->common.blk) > 1) { 739 return nbd_negotiate_send_rep_err(client, 740 NBD_REP_ERR_BLOCK_SIZE_REQD, 741 errp, 742 "request NBD_INFO_BLOCK_SIZE to " 743 "use this export"); 744 } 745 746 /* Final reply */ 747 rc = nbd_negotiate_send_rep(client, NBD_REP_ACK, errp); 748 if (rc < 0) { 749 return rc; 750 } 751 752 if (client->opt == NBD_OPT_GO) { 753 client->exp = exp; 754 client->check_align = check_align; 755 QTAILQ_INSERT_TAIL(&client->exp->clients, client, next); 756 blk_exp_ref(&client->exp->common); 757 rc = 1; 758 } 759 return rc; 760 } 761 762 /* Callback to learn when QIO TLS upgrade is complete */ 763 struct NBDTLSServerHandshakeData { 764 bool complete; 765 Error *error; 766 Coroutine *co; 767 }; 768 769 static void 770 nbd_server_tls_handshake(QIOTask *task, void *opaque) 771 { 772 struct NBDTLSServerHandshakeData *data = opaque; 773 774 qio_task_propagate_error(task, &data->error); 775 data->complete = true; 776 if (!qemu_coroutine_entered(data->co)) { 777 aio_co_wake(data->co); 778 } 779 } 780 781 /* Handle NBD_OPT_STARTTLS. Return NULL to drop connection, or else the 782 * new channel for all further (now-encrypted) communication. */ 783 static coroutine_fn QIOChannel * 784 nbd_negotiate_handle_starttls(NBDClient *client, Error **errp) 785 { 786 QIOChannel *ioc; 787 QIOChannelTLS *tioc; 788 struct NBDTLSServerHandshakeData data = { 0 }; 789 790 assert(client->opt == NBD_OPT_STARTTLS); 791 792 trace_nbd_negotiate_handle_starttls(); 793 ioc = client->ioc; 794 795 if (nbd_negotiate_send_rep(client, NBD_REP_ACK, errp) < 0) { 796 return NULL; 797 } 798 799 tioc = qio_channel_tls_new_server(ioc, 800 client->tlscreds, 801 client->tlsauthz, 802 errp); 803 if (!tioc) { 804 return NULL; 805 } 806 807 qio_channel_set_name(QIO_CHANNEL(tioc), "nbd-server-tls"); 808 trace_nbd_negotiate_handle_starttls_handshake(); 809 data.co = qemu_coroutine_self(); 810 qio_channel_tls_handshake(tioc, 811 nbd_server_tls_handshake, 812 &data, 813 NULL, 814 NULL); 815 816 if (!data.complete) { 817 qemu_coroutine_yield(); 818 assert(data.complete); 819 } 820 821 if (data.error) { 822 object_unref(OBJECT(tioc)); 823 error_propagate(errp, data.error); 824 return NULL; 825 } 826 827 return QIO_CHANNEL(tioc); 828 } 829 830 /* nbd_negotiate_send_meta_context 831 * 832 * Send one chunk of reply to NBD_OPT_{LIST,SET}_META_CONTEXT 833 * 834 * For NBD_OPT_LIST_META_CONTEXT @context_id is ignored, 0 is used instead. 835 */ 836 static coroutine_fn int 837 nbd_negotiate_send_meta_context(NBDClient *client, const char *context, 838 uint32_t context_id, Error **errp) 839 { 840 NBDOptionReplyMetaContext opt; 841 struct iovec iov[] = { 842 {.iov_base = &opt, .iov_len = sizeof(opt)}, 843 {.iov_base = (void *)context, .iov_len = strlen(context)} 844 }; 845 846 assert(iov[1].iov_len <= NBD_MAX_STRING_SIZE); 847 if (client->opt == NBD_OPT_LIST_META_CONTEXT) { 848 context_id = 0; 849 } 850 851 trace_nbd_negotiate_meta_query_reply(context, context_id); 852 set_be_option_rep(&opt.h, client->opt, NBD_REP_META_CONTEXT, 853 sizeof(opt) - sizeof(opt.h) + iov[1].iov_len); 854 stl_be_p(&opt.context_id, context_id); 855 856 return qio_channel_writev_all(client->ioc, iov, 2, errp) < 0 ? -EIO : 0; 857 } 858 859 /* 860 * Return true if @query matches @pattern, or if @query is empty when 861 * the @client is performing _LIST_. 862 */ 863 static coroutine_fn bool 864 nbd_meta_empty_or_pattern(NBDClient *client, const char *pattern, 865 const char *query) 866 { 867 if (!*query) { 868 trace_nbd_negotiate_meta_query_parse("empty"); 869 return client->opt == NBD_OPT_LIST_META_CONTEXT; 870 } 871 if (strcmp(query, pattern) == 0) { 872 trace_nbd_negotiate_meta_query_parse(pattern); 873 return true; 874 } 875 trace_nbd_negotiate_meta_query_skip("pattern not matched"); 876 return false; 877 } 878 879 /* 880 * Return true and adjust @str in place if it begins with @prefix. 881 */ 882 static coroutine_fn bool 883 nbd_strshift(const char **str, const char *prefix) 884 { 885 size_t len = strlen(prefix); 886 887 if (strncmp(*str, prefix, len) == 0) { 888 *str += len; 889 return true; 890 } 891 return false; 892 } 893 894 /* nbd_meta_base_query 895 * 896 * Handle queries to 'base' namespace. For now, only the base:allocation 897 * context is available. Return true if @query has been handled. 898 */ 899 static coroutine_fn bool 900 nbd_meta_base_query(NBDClient *client, NBDMetaContexts *meta, 901 const char *query) 902 { 903 if (!nbd_strshift(&query, "base:")) { 904 return false; 905 } 906 trace_nbd_negotiate_meta_query_parse("base:"); 907 908 if (nbd_meta_empty_or_pattern(client, "allocation", query)) { 909 meta->base_allocation = true; 910 } 911 return true; 912 } 913 914 /* nbd_meta_qemu_query 915 * 916 * Handle queries to 'qemu' namespace. For now, only the qemu:dirty-bitmap: 917 * and qemu:allocation-depth contexts are available. Return true if @query 918 * has been handled. 919 */ 920 static coroutine_fn bool 921 nbd_meta_qemu_query(NBDClient *client, NBDMetaContexts *meta, 922 const char *query) 923 { 924 size_t i; 925 926 if (!nbd_strshift(&query, "qemu:")) { 927 return false; 928 } 929 trace_nbd_negotiate_meta_query_parse("qemu:"); 930 931 if (!*query) { 932 if (client->opt == NBD_OPT_LIST_META_CONTEXT) { 933 meta->allocation_depth = meta->exp->allocation_depth; 934 if (meta->exp->nr_export_bitmaps) { 935 memset(meta->bitmaps, 1, meta->exp->nr_export_bitmaps); 936 } 937 } 938 trace_nbd_negotiate_meta_query_parse("empty"); 939 return true; 940 } 941 942 if (strcmp(query, "allocation-depth") == 0) { 943 trace_nbd_negotiate_meta_query_parse("allocation-depth"); 944 meta->allocation_depth = meta->exp->allocation_depth; 945 return true; 946 } 947 948 if (nbd_strshift(&query, "dirty-bitmap:")) { 949 trace_nbd_negotiate_meta_query_parse("dirty-bitmap:"); 950 if (!*query) { 951 if (client->opt == NBD_OPT_LIST_META_CONTEXT && 952 meta->exp->nr_export_bitmaps) { 953 memset(meta->bitmaps, 1, meta->exp->nr_export_bitmaps); 954 } 955 trace_nbd_negotiate_meta_query_parse("empty"); 956 return true; 957 } 958 959 for (i = 0; i < meta->exp->nr_export_bitmaps; i++) { 960 const char *bm_name; 961 962 bm_name = bdrv_dirty_bitmap_name(meta->exp->export_bitmaps[i]); 963 if (strcmp(bm_name, query) == 0) { 964 meta->bitmaps[i] = true; 965 trace_nbd_negotiate_meta_query_parse(query); 966 return true; 967 } 968 } 969 trace_nbd_negotiate_meta_query_skip("no dirty-bitmap match"); 970 return true; 971 } 972 973 trace_nbd_negotiate_meta_query_skip("unknown qemu context"); 974 return true; 975 } 976 977 /* nbd_negotiate_meta_query 978 * 979 * Parse namespace name and call corresponding function to parse body of the 980 * query. 981 * 982 * The only supported namespaces are 'base' and 'qemu'. 983 * 984 * Return -errno on I/O error, 0 if option was completely handled by 985 * sending a reply about inconsistent lengths, or 1 on success. */ 986 static coroutine_fn int 987 nbd_negotiate_meta_query(NBDClient *client, 988 NBDMetaContexts *meta, Error **errp) 989 { 990 int ret; 991 g_autofree char *query = NULL; 992 uint32_t len; 993 994 ret = nbd_opt_read(client, &len, sizeof(len), false, errp); 995 if (ret <= 0) { 996 return ret; 997 } 998 len = cpu_to_be32(len); 999 1000 if (len > NBD_MAX_STRING_SIZE) { 1001 trace_nbd_negotiate_meta_query_skip("length too long"); 1002 return nbd_opt_skip(client, len, errp); 1003 } 1004 1005 query = g_malloc(len + 1); 1006 ret = nbd_opt_read(client, query, len, true, errp); 1007 if (ret <= 0) { 1008 return ret; 1009 } 1010 query[len] = '\0'; 1011 1012 if (nbd_meta_base_query(client, meta, query)) { 1013 return 1; 1014 } 1015 if (nbd_meta_qemu_query(client, meta, query)) { 1016 return 1; 1017 } 1018 1019 trace_nbd_negotiate_meta_query_skip("unknown namespace"); 1020 return 1; 1021 } 1022 1023 /* nbd_negotiate_meta_queries 1024 * Handle NBD_OPT_LIST_META_CONTEXT and NBD_OPT_SET_META_CONTEXT 1025 * 1026 * Return -errno on I/O error, or 0 if option was completely handled. */ 1027 static coroutine_fn int 1028 nbd_negotiate_meta_queries(NBDClient *client, Error **errp) 1029 { 1030 int ret; 1031 g_autofree char *export_name = NULL; 1032 /* Mark unused to work around https://bugs.llvm.org/show_bug.cgi?id=3888 */ 1033 g_autofree G_GNUC_UNUSED bool *bitmaps = NULL; 1034 NBDMetaContexts local_meta = {0}; 1035 NBDMetaContexts *meta; 1036 uint32_t nb_queries; 1037 size_t i; 1038 size_t count = 0; 1039 1040 if (client->opt == NBD_OPT_SET_META_CONTEXT && 1041 client->mode < NBD_MODE_STRUCTURED) { 1042 return nbd_opt_invalid(client, errp, 1043 "request option '%s' when structured reply " 1044 "is not negotiated", 1045 nbd_opt_lookup(client->opt)); 1046 } 1047 1048 if (client->opt == NBD_OPT_LIST_META_CONTEXT) { 1049 /* Only change the caller's meta on SET. */ 1050 meta = &local_meta; 1051 } else { 1052 meta = &client->contexts; 1053 } 1054 1055 g_free(meta->bitmaps); 1056 memset(meta, 0, sizeof(*meta)); 1057 1058 ret = nbd_opt_read_name(client, &export_name, NULL, errp); 1059 if (ret <= 0) { 1060 return ret; 1061 } 1062 1063 meta->exp = nbd_export_find(export_name); 1064 if (meta->exp == NULL) { 1065 g_autofree char *sane_name = nbd_sanitize_name(export_name); 1066 1067 return nbd_opt_drop(client, NBD_REP_ERR_UNKNOWN, errp, 1068 "export '%s' not present", sane_name); 1069 } 1070 meta->bitmaps = g_new0(bool, meta->exp->nr_export_bitmaps); 1071 if (client->opt == NBD_OPT_LIST_META_CONTEXT) { 1072 bitmaps = meta->bitmaps; 1073 } 1074 1075 ret = nbd_opt_read(client, &nb_queries, sizeof(nb_queries), false, errp); 1076 if (ret <= 0) { 1077 return ret; 1078 } 1079 nb_queries = cpu_to_be32(nb_queries); 1080 trace_nbd_negotiate_meta_context(nbd_opt_lookup(client->opt), 1081 export_name, nb_queries); 1082 1083 if (client->opt == NBD_OPT_LIST_META_CONTEXT && !nb_queries) { 1084 /* enable all known contexts */ 1085 meta->base_allocation = true; 1086 meta->allocation_depth = meta->exp->allocation_depth; 1087 if (meta->exp->nr_export_bitmaps) { 1088 memset(meta->bitmaps, 1, meta->exp->nr_export_bitmaps); 1089 } 1090 } else { 1091 for (i = 0; i < nb_queries; ++i) { 1092 ret = nbd_negotiate_meta_query(client, meta, errp); 1093 if (ret <= 0) { 1094 return ret; 1095 } 1096 } 1097 } 1098 1099 if (meta->base_allocation) { 1100 ret = nbd_negotiate_send_meta_context(client, "base:allocation", 1101 NBD_META_ID_BASE_ALLOCATION, 1102 errp); 1103 if (ret < 0) { 1104 return ret; 1105 } 1106 count++; 1107 } 1108 1109 if (meta->allocation_depth) { 1110 ret = nbd_negotiate_send_meta_context(client, "qemu:allocation-depth", 1111 NBD_META_ID_ALLOCATION_DEPTH, 1112 errp); 1113 if (ret < 0) { 1114 return ret; 1115 } 1116 count++; 1117 } 1118 1119 for (i = 0; i < meta->exp->nr_export_bitmaps; i++) { 1120 const char *bm_name; 1121 g_autofree char *context = NULL; 1122 1123 if (!meta->bitmaps[i]) { 1124 continue; 1125 } 1126 1127 bm_name = bdrv_dirty_bitmap_name(meta->exp->export_bitmaps[i]); 1128 context = g_strdup_printf("qemu:dirty-bitmap:%s", bm_name); 1129 1130 ret = nbd_negotiate_send_meta_context(client, context, 1131 NBD_META_ID_DIRTY_BITMAP + i, 1132 errp); 1133 if (ret < 0) { 1134 return ret; 1135 } 1136 count++; 1137 } 1138 1139 ret = nbd_negotiate_send_rep(client, NBD_REP_ACK, errp); 1140 if (ret == 0) { 1141 meta->count = count; 1142 } 1143 1144 return ret; 1145 } 1146 1147 /* nbd_negotiate_options 1148 * Process all NBD_OPT_* client option commands, during fixed newstyle 1149 * negotiation. 1150 * Return: 1151 * -errno on error, errp is set 1152 * 0 on successful negotiation, errp is not set 1153 * 1 if client sent NBD_OPT_ABORT (i.e. on valid disconnect) or never 1154 * wrote anything (i.e. port probe); errp is not set 1155 */ 1156 static coroutine_fn int 1157 nbd_negotiate_options(NBDClient *client, Error **errp) 1158 { 1159 uint32_t flags; 1160 bool fixedNewstyle = false; 1161 bool no_zeroes = false; 1162 1163 /* Client sends: 1164 [ 0 .. 3] client flags 1165 1166 Then we loop until NBD_OPT_EXPORT_NAME or NBD_OPT_GO: 1167 [ 0 .. 7] NBD_OPTS_MAGIC 1168 [ 8 .. 11] NBD option 1169 [12 .. 15] Data length 1170 ... Rest of request 1171 1172 [ 0 .. 7] NBD_OPTS_MAGIC 1173 [ 8 .. 11] Second NBD option 1174 [12 .. 15] Data length 1175 ... Rest of request 1176 */ 1177 1178 /* 1179 * Intentionally ignore errors on this first read - we do not want 1180 * to be noisy about a mere port probe, but only for clients that 1181 * start talking the protocol and then quit abruptly. 1182 */ 1183 if (nbd_read32(client->ioc, &flags, "flags", NULL) < 0) { 1184 return 1; 1185 } 1186 client->mode = NBD_MODE_EXPORT_NAME; 1187 trace_nbd_negotiate_options_flags(flags); 1188 if (flags & NBD_FLAG_C_FIXED_NEWSTYLE) { 1189 fixedNewstyle = true; 1190 flags &= ~NBD_FLAG_C_FIXED_NEWSTYLE; 1191 client->mode = NBD_MODE_SIMPLE; 1192 } 1193 if (flags & NBD_FLAG_C_NO_ZEROES) { 1194 no_zeroes = true; 1195 flags &= ~NBD_FLAG_C_NO_ZEROES; 1196 } 1197 if (flags != 0) { 1198 error_setg(errp, "Unknown client flags 0x%" PRIx32 " received", flags); 1199 return -EINVAL; 1200 } 1201 1202 while (1) { 1203 int ret; 1204 uint32_t option, length; 1205 uint64_t magic; 1206 1207 if (nbd_read64(client->ioc, &magic, "opts magic", errp) < 0) { 1208 return -EINVAL; 1209 } 1210 trace_nbd_negotiate_options_check_magic(magic); 1211 if (magic != NBD_OPTS_MAGIC) { 1212 error_setg(errp, "Bad magic received"); 1213 return -EINVAL; 1214 } 1215 1216 if (nbd_read32(client->ioc, &option, "option", errp) < 0) { 1217 return -EINVAL; 1218 } 1219 client->opt = option; 1220 1221 if (nbd_read32(client->ioc, &length, "option length", errp) < 0) { 1222 return -EINVAL; 1223 } 1224 assert(!client->optlen); 1225 client->optlen = length; 1226 1227 if (length > NBD_MAX_BUFFER_SIZE) { 1228 error_setg(errp, "len (%" PRIu32 ") is larger than max len (%u)", 1229 length, NBD_MAX_BUFFER_SIZE); 1230 return -EINVAL; 1231 } 1232 1233 trace_nbd_negotiate_options_check_option(option, 1234 nbd_opt_lookup(option)); 1235 if (client->tlscreds && 1236 client->ioc == (QIOChannel *)client->sioc) { 1237 QIOChannel *tioc; 1238 if (!fixedNewstyle) { 1239 error_setg(errp, "Unsupported option 0x%" PRIx32, option); 1240 return -EINVAL; 1241 } 1242 switch (option) { 1243 case NBD_OPT_STARTTLS: 1244 if (length) { 1245 /* Unconditionally drop the connection if the client 1246 * can't start a TLS negotiation correctly */ 1247 return nbd_reject_length(client, true, errp); 1248 } 1249 tioc = nbd_negotiate_handle_starttls(client, errp); 1250 if (!tioc) { 1251 return -EIO; 1252 } 1253 ret = 0; 1254 object_unref(OBJECT(client->ioc)); 1255 client->ioc = tioc; 1256 break; 1257 1258 case NBD_OPT_EXPORT_NAME: 1259 /* No way to return an error to client, so drop connection */ 1260 error_setg(errp, "Option 0x%x not permitted before TLS", 1261 option); 1262 return -EINVAL; 1263 1264 default: 1265 /* Let the client keep trying, unless they asked to 1266 * quit. Always try to give an error back to the 1267 * client; but when replying to OPT_ABORT, be aware 1268 * that the client may hang up before receiving the 1269 * error, in which case we are fine ignoring the 1270 * resulting EPIPE. */ 1271 ret = nbd_opt_drop(client, NBD_REP_ERR_TLS_REQD, 1272 option == NBD_OPT_ABORT ? NULL : errp, 1273 "Option 0x%" PRIx32 1274 " not permitted before TLS", option); 1275 if (option == NBD_OPT_ABORT) { 1276 return 1; 1277 } 1278 break; 1279 } 1280 } else if (fixedNewstyle) { 1281 switch (option) { 1282 case NBD_OPT_LIST: 1283 if (length) { 1284 ret = nbd_reject_length(client, false, errp); 1285 } else { 1286 ret = nbd_negotiate_handle_list(client, errp); 1287 } 1288 break; 1289 1290 case NBD_OPT_ABORT: 1291 /* NBD spec says we must try to reply before 1292 * disconnecting, but that we must also tolerate 1293 * guests that don't wait for our reply. */ 1294 nbd_negotiate_send_rep(client, NBD_REP_ACK, NULL); 1295 return 1; 1296 1297 case NBD_OPT_EXPORT_NAME: 1298 return nbd_negotiate_handle_export_name(client, no_zeroes, 1299 errp); 1300 1301 case NBD_OPT_INFO: 1302 case NBD_OPT_GO: 1303 ret = nbd_negotiate_handle_info(client, errp); 1304 if (ret == 1) { 1305 assert(option == NBD_OPT_GO); 1306 return 0; 1307 } 1308 break; 1309 1310 case NBD_OPT_STARTTLS: 1311 if (length) { 1312 ret = nbd_reject_length(client, false, errp); 1313 } else if (client->tlscreds) { 1314 ret = nbd_negotiate_send_rep_err(client, 1315 NBD_REP_ERR_INVALID, errp, 1316 "TLS already enabled"); 1317 } else { 1318 ret = nbd_negotiate_send_rep_err(client, 1319 NBD_REP_ERR_POLICY, errp, 1320 "TLS not configured"); 1321 } 1322 break; 1323 1324 case NBD_OPT_STRUCTURED_REPLY: 1325 if (length) { 1326 ret = nbd_reject_length(client, false, errp); 1327 } else if (client->mode >= NBD_MODE_EXTENDED) { 1328 ret = nbd_negotiate_send_rep_err( 1329 client, NBD_REP_ERR_EXT_HEADER_REQD, errp, 1330 "extended headers already negotiated"); 1331 } else if (client->mode >= NBD_MODE_STRUCTURED) { 1332 ret = nbd_negotiate_send_rep_err( 1333 client, NBD_REP_ERR_INVALID, errp, 1334 "structured reply already negotiated"); 1335 } else { 1336 ret = nbd_negotiate_send_rep(client, NBD_REP_ACK, errp); 1337 client->mode = NBD_MODE_STRUCTURED; 1338 } 1339 break; 1340 1341 case NBD_OPT_LIST_META_CONTEXT: 1342 case NBD_OPT_SET_META_CONTEXT: 1343 ret = nbd_negotiate_meta_queries(client, errp); 1344 break; 1345 1346 case NBD_OPT_EXTENDED_HEADERS: 1347 if (length) { 1348 ret = nbd_reject_length(client, false, errp); 1349 } else if (client->mode >= NBD_MODE_EXTENDED) { 1350 ret = nbd_negotiate_send_rep_err( 1351 client, NBD_REP_ERR_INVALID, errp, 1352 "extended headers already negotiated"); 1353 } else { 1354 ret = nbd_negotiate_send_rep(client, NBD_REP_ACK, errp); 1355 client->mode = NBD_MODE_EXTENDED; 1356 } 1357 break; 1358 1359 default: 1360 ret = nbd_opt_drop(client, NBD_REP_ERR_UNSUP, errp, 1361 "Unsupported option %" PRIu32 " (%s)", 1362 option, nbd_opt_lookup(option)); 1363 break; 1364 } 1365 } else { 1366 /* 1367 * If broken new-style we should drop the connection 1368 * for anything except NBD_OPT_EXPORT_NAME 1369 */ 1370 switch (option) { 1371 case NBD_OPT_EXPORT_NAME: 1372 return nbd_negotiate_handle_export_name(client, no_zeroes, 1373 errp); 1374 1375 default: 1376 error_setg(errp, "Unsupported option %" PRIu32 " (%s)", 1377 option, nbd_opt_lookup(option)); 1378 return -EINVAL; 1379 } 1380 } 1381 if (ret < 0) { 1382 return ret; 1383 } 1384 } 1385 } 1386 1387 /* nbd_negotiate 1388 * Return: 1389 * -errno on error, errp is set 1390 * 0 on successful negotiation, errp is not set 1391 * 1 if client sent NBD_OPT_ABORT (i.e. on valid disconnect) or never 1392 * wrote anything (i.e. port probe); errp is not set 1393 */ 1394 static coroutine_fn int nbd_negotiate(NBDClient *client, Error **errp) 1395 { 1396 ERRP_GUARD(); 1397 char buf[NBD_OLDSTYLE_NEGOTIATE_SIZE] = ""; 1398 int ret; 1399 1400 /* Old style negotiation header, no room for options 1401 [ 0 .. 7] passwd ("NBDMAGIC") 1402 [ 8 .. 15] magic (NBD_CLIENT_MAGIC) 1403 [16 .. 23] size 1404 [24 .. 27] export flags (zero-extended) 1405 [28 .. 151] reserved (0) 1406 1407 New style negotiation header, client can send options 1408 [ 0 .. 7] passwd ("NBDMAGIC") 1409 [ 8 .. 15] magic (NBD_OPTS_MAGIC) 1410 [16 .. 17] server flags (0) 1411 ....options sent, ending in NBD_OPT_EXPORT_NAME or NBD_OPT_GO.... 1412 */ 1413 1414 qio_channel_set_blocking(client->ioc, false, NULL); 1415 qio_channel_set_follow_coroutine_ctx(client->ioc, true); 1416 1417 trace_nbd_negotiate_begin(); 1418 memcpy(buf, "NBDMAGIC", 8); 1419 1420 stq_be_p(buf + 8, NBD_OPTS_MAGIC); 1421 stw_be_p(buf + 16, NBD_FLAG_FIXED_NEWSTYLE | NBD_FLAG_NO_ZEROES); 1422 1423 /* 1424 * Be silent about failure to write our greeting: there is nothing 1425 * wrong with a client testing if our port is alive. 1426 */ 1427 if (nbd_write(client->ioc, buf, 18, NULL) < 0) { 1428 return 1; 1429 } 1430 ret = nbd_negotiate_options(client, errp); 1431 if (ret != 0) { 1432 if (ret < 0) { 1433 error_prepend(errp, "option negotiation failed: "); 1434 } 1435 return ret; 1436 } 1437 1438 assert(!client->optlen); 1439 trace_nbd_negotiate_success(); 1440 1441 return 0; 1442 } 1443 1444 /* nbd_read_eof 1445 * Tries to read @size bytes from @ioc. This is a local implementation of 1446 * qio_channel_readv_all_eof. We have it here because we need it to be 1447 * interruptible and to know when the coroutine is yielding. 1448 * Returns 1 on success 1449 * 0 on eof, when no data was read (errp is not set) 1450 * negative errno on failure (errp is set) 1451 */ 1452 static inline int coroutine_fn 1453 nbd_read_eof(NBDClient *client, void *buffer, size_t size, Error **errp) 1454 { 1455 bool partial = false; 1456 1457 assert(size); 1458 while (size > 0) { 1459 struct iovec iov = { .iov_base = buffer, .iov_len = size }; 1460 ssize_t len; 1461 1462 len = qio_channel_readv(client->ioc, &iov, 1, errp); 1463 if (len == QIO_CHANNEL_ERR_BLOCK) { 1464 WITH_QEMU_LOCK_GUARD(&client->lock) { 1465 client->read_yielding = true; 1466 1467 /* Prompt main loop thread to re-run nbd_drained_poll() */ 1468 aio_wait_kick(); 1469 } 1470 qio_channel_yield(client->ioc, G_IO_IN); 1471 WITH_QEMU_LOCK_GUARD(&client->lock) { 1472 client->read_yielding = false; 1473 if (client->quiescing) { 1474 return -EAGAIN; 1475 } 1476 } 1477 continue; 1478 } else if (len < 0) { 1479 return -EIO; 1480 } else if (len == 0) { 1481 if (partial) { 1482 error_setg(errp, 1483 "Unexpected end-of-file before all bytes were read"); 1484 return -EIO; 1485 } else { 1486 return 0; 1487 } 1488 } 1489 1490 partial = true; 1491 size -= len; 1492 buffer = (uint8_t *) buffer + len; 1493 } 1494 return 1; 1495 } 1496 1497 static int coroutine_fn nbd_receive_request(NBDClient *client, NBDRequest *request, 1498 Error **errp) 1499 { 1500 uint8_t buf[NBD_EXTENDED_REQUEST_SIZE]; 1501 uint32_t magic, expect; 1502 int ret; 1503 size_t size = client->mode >= NBD_MODE_EXTENDED ? 1504 NBD_EXTENDED_REQUEST_SIZE : NBD_REQUEST_SIZE; 1505 1506 ret = nbd_read_eof(client, buf, size, errp); 1507 if (ret < 0) { 1508 return ret; 1509 } 1510 if (ret == 0) { 1511 return -EIO; 1512 } 1513 1514 /* 1515 * Compact request 1516 * [ 0 .. 3] magic (NBD_REQUEST_MAGIC) 1517 * [ 4 .. 5] flags (NBD_CMD_FLAG_FUA, ...) 1518 * [ 6 .. 7] type (NBD_CMD_READ, ...) 1519 * [ 8 .. 15] cookie 1520 * [16 .. 23] from 1521 * [24 .. 27] len 1522 * Extended request 1523 * [ 0 .. 3] magic (NBD_EXTENDED_REQUEST_MAGIC) 1524 * [ 4 .. 5] flags (NBD_CMD_FLAG_FUA, NBD_CMD_FLAG_PAYLOAD_LEN, ...) 1525 * [ 6 .. 7] type (NBD_CMD_READ, ...) 1526 * [ 8 .. 15] cookie 1527 * [16 .. 23] from 1528 * [24 .. 31] len 1529 */ 1530 1531 magic = ldl_be_p(buf); 1532 request->flags = lduw_be_p(buf + 4); 1533 request->type = lduw_be_p(buf + 6); 1534 request->cookie = ldq_be_p(buf + 8); 1535 request->from = ldq_be_p(buf + 16); 1536 if (client->mode >= NBD_MODE_EXTENDED) { 1537 request->len = ldq_be_p(buf + 24); 1538 expect = NBD_EXTENDED_REQUEST_MAGIC; 1539 } else { 1540 request->len = (uint32_t)ldl_be_p(buf + 24); /* widen 32 to 64 bits */ 1541 expect = NBD_REQUEST_MAGIC; 1542 } 1543 1544 trace_nbd_receive_request(magic, request->flags, request->type, 1545 request->from, request->len); 1546 1547 if (magic != expect) { 1548 error_setg(errp, "invalid magic (got 0x%" PRIx32 ", expected 0x%" 1549 PRIx32 ")", magic, expect); 1550 return -EINVAL; 1551 } 1552 return 0; 1553 } 1554 1555 #define MAX_NBD_REQUESTS 16 1556 1557 /* Runs in export AioContext and main loop thread */ 1558 void nbd_client_get(NBDClient *client) 1559 { 1560 qatomic_inc(&client->refcount); 1561 } 1562 1563 void nbd_client_put(NBDClient *client) 1564 { 1565 assert(qemu_in_main_thread()); 1566 1567 if (qatomic_fetch_dec(&client->refcount) == 1) { 1568 /* The last reference should be dropped by client->close, 1569 * which is called by client_close. 1570 */ 1571 assert(client->closing); 1572 1573 object_unref(OBJECT(client->sioc)); 1574 object_unref(OBJECT(client->ioc)); 1575 if (client->tlscreds) { 1576 object_unref(OBJECT(client->tlscreds)); 1577 } 1578 g_free(client->tlsauthz); 1579 if (client->exp) { 1580 QTAILQ_REMOVE(&client->exp->clients, client, next); 1581 blk_exp_unref(&client->exp->common); 1582 } 1583 g_free(client->contexts.bitmaps); 1584 qemu_mutex_destroy(&client->lock); 1585 g_free(client); 1586 } 1587 } 1588 1589 /* 1590 * Tries to release the reference to @client, but only if other references 1591 * remain. This is an optimization for the common case where we want to avoid 1592 * the expense of scheduling nbd_client_put() in the main loop thread. 1593 * 1594 * Returns true upon success or false if the reference was not released because 1595 * it is the last reference. 1596 */ 1597 static bool nbd_client_put_nonzero(NBDClient *client) 1598 { 1599 int old = qatomic_read(&client->refcount); 1600 int expected; 1601 1602 do { 1603 if (old == 1) { 1604 return false; 1605 } 1606 1607 expected = old; 1608 old = qatomic_cmpxchg(&client->refcount, expected, expected - 1); 1609 } while (old != expected); 1610 1611 return true; 1612 } 1613 1614 static void client_close(NBDClient *client, bool negotiated) 1615 { 1616 assert(qemu_in_main_thread()); 1617 1618 WITH_QEMU_LOCK_GUARD(&client->lock) { 1619 if (client->closing) { 1620 return; 1621 } 1622 1623 client->closing = true; 1624 } 1625 1626 /* Force requests to finish. They will drop their own references, 1627 * then we'll close the socket and free the NBDClient. 1628 */ 1629 qio_channel_shutdown(client->ioc, QIO_CHANNEL_SHUTDOWN_BOTH, 1630 NULL); 1631 1632 /* Also tell the client, so that they release their reference. */ 1633 if (client->close_fn) { 1634 client->close_fn(client, negotiated); 1635 } 1636 } 1637 1638 /* Runs in export AioContext with client->lock held */ 1639 static NBDRequestData *nbd_request_get(NBDClient *client) 1640 { 1641 NBDRequestData *req; 1642 1643 assert(client->nb_requests <= MAX_NBD_REQUESTS - 1); 1644 client->nb_requests++; 1645 1646 req = g_new0(NBDRequestData, 1); 1647 req->client = client; 1648 return req; 1649 } 1650 1651 /* Runs in export AioContext with client->lock held */ 1652 static void nbd_request_put(NBDRequestData *req) 1653 { 1654 NBDClient *client = req->client; 1655 1656 if (req->data) { 1657 qemu_vfree(req->data); 1658 } 1659 g_free(req); 1660 1661 client->nb_requests--; 1662 1663 if (client->quiescing && client->nb_requests == 0) { 1664 aio_wait_kick(); 1665 } 1666 1667 nbd_client_receive_next_request(client); 1668 } 1669 1670 static void blk_aio_attached(AioContext *ctx, void *opaque) 1671 { 1672 NBDExport *exp = opaque; 1673 NBDClient *client; 1674 1675 assert(qemu_in_main_thread()); 1676 1677 trace_nbd_blk_aio_attached(exp->name, ctx); 1678 1679 exp->common.ctx = ctx; 1680 1681 QTAILQ_FOREACH(client, &exp->clients, next) { 1682 WITH_QEMU_LOCK_GUARD(&client->lock) { 1683 assert(client->nb_requests == 0); 1684 assert(client->recv_coroutine == NULL); 1685 assert(client->send_coroutine == NULL); 1686 } 1687 } 1688 } 1689 1690 static void blk_aio_detach(void *opaque) 1691 { 1692 NBDExport *exp = opaque; 1693 1694 assert(qemu_in_main_thread()); 1695 1696 trace_nbd_blk_aio_detach(exp->name, exp->common.ctx); 1697 1698 exp->common.ctx = NULL; 1699 } 1700 1701 static void nbd_drained_begin(void *opaque) 1702 { 1703 NBDExport *exp = opaque; 1704 NBDClient *client; 1705 1706 assert(qemu_in_main_thread()); 1707 1708 QTAILQ_FOREACH(client, &exp->clients, next) { 1709 WITH_QEMU_LOCK_GUARD(&client->lock) { 1710 client->quiescing = true; 1711 } 1712 } 1713 } 1714 1715 static void nbd_drained_end(void *opaque) 1716 { 1717 NBDExport *exp = opaque; 1718 NBDClient *client; 1719 1720 assert(qemu_in_main_thread()); 1721 1722 QTAILQ_FOREACH(client, &exp->clients, next) { 1723 WITH_QEMU_LOCK_GUARD(&client->lock) { 1724 client->quiescing = false; 1725 nbd_client_receive_next_request(client); 1726 } 1727 } 1728 } 1729 1730 /* Runs in export AioContext */ 1731 static void nbd_wake_read_bh(void *opaque) 1732 { 1733 NBDClient *client = opaque; 1734 qio_channel_wake_read(client->ioc); 1735 } 1736 1737 static bool nbd_drained_poll(void *opaque) 1738 { 1739 NBDExport *exp = opaque; 1740 NBDClient *client; 1741 1742 assert(qemu_in_main_thread()); 1743 1744 QTAILQ_FOREACH(client, &exp->clients, next) { 1745 WITH_QEMU_LOCK_GUARD(&client->lock) { 1746 if (client->nb_requests != 0) { 1747 /* 1748 * If there's a coroutine waiting for a request on nbd_read_eof() 1749 * enter it here so we don't depend on the client to wake it up. 1750 * 1751 * Schedule a BH in the export AioContext to avoid missing the 1752 * wake up due to the race between qio_channel_wake_read() and 1753 * qio_channel_yield(). 1754 */ 1755 if (client->recv_coroutine != NULL && client->read_yielding) { 1756 aio_bh_schedule_oneshot(nbd_export_aio_context(client->exp), 1757 nbd_wake_read_bh, client); 1758 } 1759 1760 return true; 1761 } 1762 } 1763 } 1764 1765 return false; 1766 } 1767 1768 static void nbd_eject_notifier(Notifier *n, void *data) 1769 { 1770 NBDExport *exp = container_of(n, NBDExport, eject_notifier); 1771 1772 assert(qemu_in_main_thread()); 1773 1774 blk_exp_request_shutdown(&exp->common); 1775 } 1776 1777 void nbd_export_set_on_eject_blk(BlockExport *exp, BlockBackend *blk) 1778 { 1779 NBDExport *nbd_exp = container_of(exp, NBDExport, common); 1780 assert(exp->drv == &blk_exp_nbd); 1781 assert(nbd_exp->eject_notifier_blk == NULL); 1782 1783 blk_ref(blk); 1784 nbd_exp->eject_notifier_blk = blk; 1785 nbd_exp->eject_notifier.notify = nbd_eject_notifier; 1786 blk_add_remove_bs_notifier(blk, &nbd_exp->eject_notifier); 1787 } 1788 1789 static const BlockDevOps nbd_block_ops = { 1790 .drained_begin = nbd_drained_begin, 1791 .drained_end = nbd_drained_end, 1792 .drained_poll = nbd_drained_poll, 1793 }; 1794 1795 static int nbd_export_create(BlockExport *blk_exp, BlockExportOptions *exp_args, 1796 Error **errp) 1797 { 1798 NBDExport *exp = container_of(blk_exp, NBDExport, common); 1799 BlockExportOptionsNbd *arg = &exp_args->u.nbd; 1800 const char *name = arg->name ?: exp_args->node_name; 1801 BlockBackend *blk = blk_exp->blk; 1802 int64_t size; 1803 uint64_t perm, shared_perm; 1804 bool readonly = !exp_args->writable; 1805 BlockDirtyBitmapOrStrList *bitmaps; 1806 size_t i; 1807 int ret; 1808 1809 GLOBAL_STATE_CODE(); 1810 assert(exp_args->type == BLOCK_EXPORT_TYPE_NBD); 1811 1812 if (!nbd_server_is_running()) { 1813 error_setg(errp, "NBD server not running"); 1814 return -EINVAL; 1815 } 1816 1817 if (strlen(name) > NBD_MAX_STRING_SIZE) { 1818 error_setg(errp, "export name '%s' too long", name); 1819 return -EINVAL; 1820 } 1821 1822 if (arg->description && strlen(arg->description) > NBD_MAX_STRING_SIZE) { 1823 error_setg(errp, "description '%s' too long", arg->description); 1824 return -EINVAL; 1825 } 1826 1827 if (nbd_export_find(name)) { 1828 error_setg(errp, "NBD server already has export named '%s'", name); 1829 return -EEXIST; 1830 } 1831 1832 size = blk_getlength(blk); 1833 if (size < 0) { 1834 error_setg_errno(errp, -size, 1835 "Failed to determine the NBD export's length"); 1836 return size; 1837 } 1838 1839 /* Don't allow resize while the NBD server is running, otherwise we don't 1840 * care what happens with the node. */ 1841 blk_get_perm(blk, &perm, &shared_perm); 1842 ret = blk_set_perm(blk, perm, shared_perm & ~BLK_PERM_RESIZE, errp); 1843 if (ret < 0) { 1844 return ret; 1845 } 1846 1847 QTAILQ_INIT(&exp->clients); 1848 exp->name = g_strdup(name); 1849 exp->description = g_strdup(arg->description); 1850 exp->nbdflags = (NBD_FLAG_HAS_FLAGS | NBD_FLAG_SEND_FLUSH | 1851 NBD_FLAG_SEND_FUA | NBD_FLAG_SEND_CACHE); 1852 1853 if (nbd_server_max_connections() != 1) { 1854 exp->nbdflags |= NBD_FLAG_CAN_MULTI_CONN; 1855 } 1856 if (readonly) { 1857 exp->nbdflags |= NBD_FLAG_READ_ONLY; 1858 } else { 1859 exp->nbdflags |= (NBD_FLAG_SEND_TRIM | NBD_FLAG_SEND_WRITE_ZEROES | 1860 NBD_FLAG_SEND_FAST_ZERO); 1861 } 1862 exp->size = QEMU_ALIGN_DOWN(size, BDRV_SECTOR_SIZE); 1863 1864 bdrv_graph_rdlock_main_loop(); 1865 1866 for (bitmaps = arg->bitmaps; bitmaps; bitmaps = bitmaps->next) { 1867 exp->nr_export_bitmaps++; 1868 } 1869 exp->export_bitmaps = g_new0(BdrvDirtyBitmap *, exp->nr_export_bitmaps); 1870 for (i = 0, bitmaps = arg->bitmaps; bitmaps; 1871 i++, bitmaps = bitmaps->next) 1872 { 1873 const char *bitmap; 1874 BlockDriverState *bs = blk_bs(blk); 1875 BdrvDirtyBitmap *bm = NULL; 1876 1877 switch (bitmaps->value->type) { 1878 case QTYPE_QSTRING: 1879 bitmap = bitmaps->value->u.local; 1880 while (bs) { 1881 bm = bdrv_find_dirty_bitmap(bs, bitmap); 1882 if (bm != NULL) { 1883 break; 1884 } 1885 1886 bs = bdrv_filter_or_cow_bs(bs); 1887 } 1888 1889 if (bm == NULL) { 1890 ret = -ENOENT; 1891 error_setg(errp, "Bitmap '%s' is not found", 1892 bitmaps->value->u.local); 1893 goto fail; 1894 } 1895 1896 if (readonly && bdrv_is_writable(bs) && 1897 bdrv_dirty_bitmap_enabled(bm)) { 1898 ret = -EINVAL; 1899 error_setg(errp, "Enabled bitmap '%s' incompatible with " 1900 "readonly export", bitmap); 1901 goto fail; 1902 } 1903 break; 1904 case QTYPE_QDICT: 1905 bitmap = bitmaps->value->u.external.name; 1906 bm = block_dirty_bitmap_lookup(bitmaps->value->u.external.node, 1907 bitmap, NULL, errp); 1908 if (!bm) { 1909 ret = -ENOENT; 1910 goto fail; 1911 } 1912 break; 1913 default: 1914 abort(); 1915 } 1916 1917 assert(bm); 1918 1919 if (bdrv_dirty_bitmap_check(bm, BDRV_BITMAP_ALLOW_RO, errp)) { 1920 ret = -EINVAL; 1921 goto fail; 1922 } 1923 1924 exp->export_bitmaps[i] = bm; 1925 assert(strlen(bitmap) <= BDRV_BITMAP_MAX_NAME_SIZE); 1926 } 1927 1928 /* Mark bitmaps busy in a separate loop, to simplify roll-back concerns. */ 1929 for (i = 0; i < exp->nr_export_bitmaps; i++) { 1930 bdrv_dirty_bitmap_set_busy(exp->export_bitmaps[i], true); 1931 } 1932 1933 exp->allocation_depth = arg->allocation_depth; 1934 1935 /* 1936 * We need to inhibit request queuing in the block layer to ensure we can 1937 * be properly quiesced when entering a drained section, as our coroutines 1938 * servicing pending requests might enter blk_pread(). 1939 */ 1940 blk_set_disable_request_queuing(blk, true); 1941 1942 blk_add_aio_context_notifier(blk, blk_aio_attached, blk_aio_detach, exp); 1943 1944 blk_set_dev_ops(blk, &nbd_block_ops, exp); 1945 1946 QTAILQ_INSERT_TAIL(&exports, exp, next); 1947 1948 bdrv_graph_rdunlock_main_loop(); 1949 1950 return 0; 1951 1952 fail: 1953 bdrv_graph_rdunlock_main_loop(); 1954 g_free(exp->export_bitmaps); 1955 g_free(exp->name); 1956 g_free(exp->description); 1957 return ret; 1958 } 1959 1960 NBDExport *nbd_export_find(const char *name) 1961 { 1962 NBDExport *exp; 1963 QTAILQ_FOREACH(exp, &exports, next) { 1964 if (strcmp(name, exp->name) == 0) { 1965 return exp; 1966 } 1967 } 1968 1969 return NULL; 1970 } 1971 1972 AioContext * 1973 nbd_export_aio_context(NBDExport *exp) 1974 { 1975 return exp->common.ctx; 1976 } 1977 1978 static void nbd_export_request_shutdown(BlockExport *blk_exp) 1979 { 1980 NBDExport *exp = container_of(blk_exp, NBDExport, common); 1981 NBDClient *client, *next; 1982 1983 blk_exp_ref(&exp->common); 1984 /* 1985 * TODO: Should we expand QMP BlockExportRemoveMode enum to allow a 1986 * close mode that stops advertising the export to new clients but 1987 * still permits existing clients to run to completion? Because of 1988 * that possibility, nbd_export_close() can be called more than 1989 * once on an export. 1990 */ 1991 QTAILQ_FOREACH_SAFE(client, &exp->clients, next, next) { 1992 client_close(client, true); 1993 } 1994 if (exp->name) { 1995 g_free(exp->name); 1996 exp->name = NULL; 1997 QTAILQ_REMOVE(&exports, exp, next); 1998 } 1999 blk_exp_unref(&exp->common); 2000 } 2001 2002 static void nbd_export_delete(BlockExport *blk_exp) 2003 { 2004 size_t i; 2005 NBDExport *exp = container_of(blk_exp, NBDExport, common); 2006 2007 assert(exp->name == NULL); 2008 assert(QTAILQ_EMPTY(&exp->clients)); 2009 2010 g_free(exp->description); 2011 exp->description = NULL; 2012 2013 if (exp->eject_notifier_blk) { 2014 notifier_remove(&exp->eject_notifier); 2015 blk_unref(exp->eject_notifier_blk); 2016 } 2017 blk_remove_aio_context_notifier(exp->common.blk, blk_aio_attached, 2018 blk_aio_detach, exp); 2019 blk_set_disable_request_queuing(exp->common.blk, false); 2020 2021 for (i = 0; i < exp->nr_export_bitmaps; i++) { 2022 bdrv_dirty_bitmap_set_busy(exp->export_bitmaps[i], false); 2023 } 2024 } 2025 2026 const BlockExportDriver blk_exp_nbd = { 2027 .type = BLOCK_EXPORT_TYPE_NBD, 2028 .instance_size = sizeof(NBDExport), 2029 .create = nbd_export_create, 2030 .delete = nbd_export_delete, 2031 .request_shutdown = nbd_export_request_shutdown, 2032 }; 2033 2034 static int coroutine_fn nbd_co_send_iov(NBDClient *client, struct iovec *iov, 2035 unsigned niov, Error **errp) 2036 { 2037 int ret; 2038 2039 g_assert(qemu_in_coroutine()); 2040 qemu_co_mutex_lock(&client->send_lock); 2041 client->send_coroutine = qemu_coroutine_self(); 2042 2043 ret = qio_channel_writev_all(client->ioc, iov, niov, errp) < 0 ? -EIO : 0; 2044 2045 client->send_coroutine = NULL; 2046 qemu_co_mutex_unlock(&client->send_lock); 2047 2048 return ret; 2049 } 2050 2051 static inline void set_be_simple_reply(NBDSimpleReply *reply, uint64_t error, 2052 uint64_t cookie) 2053 { 2054 stl_be_p(&reply->magic, NBD_SIMPLE_REPLY_MAGIC); 2055 stl_be_p(&reply->error, error); 2056 stq_be_p(&reply->cookie, cookie); 2057 } 2058 2059 static int coroutine_fn nbd_co_send_simple_reply(NBDClient *client, 2060 NBDRequest *request, 2061 uint32_t error, 2062 void *data, 2063 uint64_t len, 2064 Error **errp) 2065 { 2066 NBDSimpleReply reply; 2067 int nbd_err = system_errno_to_nbd_errno(error); 2068 struct iovec iov[] = { 2069 {.iov_base = &reply, .iov_len = sizeof(reply)}, 2070 {.iov_base = data, .iov_len = len} 2071 }; 2072 2073 assert(!len || !nbd_err); 2074 assert(len <= NBD_MAX_BUFFER_SIZE); 2075 assert(client->mode < NBD_MODE_STRUCTURED || 2076 (client->mode == NBD_MODE_STRUCTURED && 2077 request->type != NBD_CMD_READ)); 2078 trace_nbd_co_send_simple_reply(request->cookie, nbd_err, 2079 nbd_err_lookup(nbd_err), len); 2080 set_be_simple_reply(&reply, nbd_err, request->cookie); 2081 2082 return nbd_co_send_iov(client, iov, 2, errp); 2083 } 2084 2085 /* 2086 * Prepare the header of a reply chunk for network transmission. 2087 * 2088 * On input, @iov is partially initialized: iov[0].iov_base must point 2089 * to an uninitialized NBDReply, while the remaining @niov elements 2090 * (if any) must be ready for transmission. This function then 2091 * populates iov[0] for transmission. 2092 */ 2093 static inline void set_be_chunk(NBDClient *client, struct iovec *iov, 2094 size_t niov, uint16_t flags, uint16_t type, 2095 NBDRequest *request) 2096 { 2097 size_t i, length = 0; 2098 2099 for (i = 1; i < niov; i++) { 2100 length += iov[i].iov_len; 2101 } 2102 assert(length <= NBD_MAX_BUFFER_SIZE + sizeof(NBDStructuredReadData)); 2103 2104 if (client->mode >= NBD_MODE_EXTENDED) { 2105 NBDExtendedReplyChunk *chunk = iov->iov_base; 2106 2107 iov[0].iov_len = sizeof(*chunk); 2108 stl_be_p(&chunk->magic, NBD_EXTENDED_REPLY_MAGIC); 2109 stw_be_p(&chunk->flags, flags); 2110 stw_be_p(&chunk->type, type); 2111 stq_be_p(&chunk->cookie, request->cookie); 2112 stq_be_p(&chunk->offset, request->from); 2113 stq_be_p(&chunk->length, length); 2114 } else { 2115 NBDStructuredReplyChunk *chunk = iov->iov_base; 2116 2117 iov[0].iov_len = sizeof(*chunk); 2118 stl_be_p(&chunk->magic, NBD_STRUCTURED_REPLY_MAGIC); 2119 stw_be_p(&chunk->flags, flags); 2120 stw_be_p(&chunk->type, type); 2121 stq_be_p(&chunk->cookie, request->cookie); 2122 stl_be_p(&chunk->length, length); 2123 } 2124 } 2125 2126 static int coroutine_fn nbd_co_send_chunk_done(NBDClient *client, 2127 NBDRequest *request, 2128 Error **errp) 2129 { 2130 NBDReply hdr; 2131 struct iovec iov[] = { 2132 {.iov_base = &hdr}, 2133 }; 2134 2135 trace_nbd_co_send_chunk_done(request->cookie); 2136 set_be_chunk(client, iov, 1, NBD_REPLY_FLAG_DONE, 2137 NBD_REPLY_TYPE_NONE, request); 2138 return nbd_co_send_iov(client, iov, 1, errp); 2139 } 2140 2141 static int coroutine_fn nbd_co_send_chunk_read(NBDClient *client, 2142 NBDRequest *request, 2143 uint64_t offset, 2144 void *data, 2145 uint64_t size, 2146 bool final, 2147 Error **errp) 2148 { 2149 NBDReply hdr; 2150 NBDStructuredReadData chunk; 2151 struct iovec iov[] = { 2152 {.iov_base = &hdr}, 2153 {.iov_base = &chunk, .iov_len = sizeof(chunk)}, 2154 {.iov_base = data, .iov_len = size} 2155 }; 2156 2157 assert(size && size <= NBD_MAX_BUFFER_SIZE); 2158 trace_nbd_co_send_chunk_read(request->cookie, offset, data, size); 2159 set_be_chunk(client, iov, 3, final ? NBD_REPLY_FLAG_DONE : 0, 2160 NBD_REPLY_TYPE_OFFSET_DATA, request); 2161 stq_be_p(&chunk.offset, offset); 2162 2163 return nbd_co_send_iov(client, iov, 3, errp); 2164 } 2165 2166 static int coroutine_fn nbd_co_send_chunk_error(NBDClient *client, 2167 NBDRequest *request, 2168 uint32_t error, 2169 const char *msg, 2170 Error **errp) 2171 { 2172 NBDReply hdr; 2173 NBDStructuredError chunk; 2174 int nbd_err = system_errno_to_nbd_errno(error); 2175 struct iovec iov[] = { 2176 {.iov_base = &hdr}, 2177 {.iov_base = &chunk, .iov_len = sizeof(chunk)}, 2178 {.iov_base = (char *)msg, .iov_len = msg ? strlen(msg) : 0}, 2179 }; 2180 2181 assert(nbd_err); 2182 trace_nbd_co_send_chunk_error(request->cookie, nbd_err, 2183 nbd_err_lookup(nbd_err), msg ? msg : ""); 2184 set_be_chunk(client, iov, 3, NBD_REPLY_FLAG_DONE, 2185 NBD_REPLY_TYPE_ERROR, request); 2186 stl_be_p(&chunk.error, nbd_err); 2187 stw_be_p(&chunk.message_length, iov[2].iov_len); 2188 2189 return nbd_co_send_iov(client, iov, 3, errp); 2190 } 2191 2192 /* Do a sparse read and send the structured reply to the client. 2193 * Returns -errno if sending fails. blk_co_block_status_above() failure is 2194 * reported to the client, at which point this function succeeds. 2195 */ 2196 static int coroutine_fn nbd_co_send_sparse_read(NBDClient *client, 2197 NBDRequest *request, 2198 uint64_t offset, 2199 uint8_t *data, 2200 uint64_t size, 2201 Error **errp) 2202 { 2203 int ret = 0; 2204 NBDExport *exp = client->exp; 2205 size_t progress = 0; 2206 2207 assert(size <= NBD_MAX_BUFFER_SIZE); 2208 while (progress < size) { 2209 int64_t pnum; 2210 int status = blk_co_block_status_above(exp->common.blk, NULL, 2211 offset + progress, 2212 size - progress, &pnum, NULL, 2213 NULL); 2214 bool final; 2215 2216 if (status < 0) { 2217 char *msg = g_strdup_printf("unable to check for holes: %s", 2218 strerror(-status)); 2219 2220 ret = nbd_co_send_chunk_error(client, request, -status, msg, errp); 2221 g_free(msg); 2222 return ret; 2223 } 2224 assert(pnum && pnum <= size - progress); 2225 final = progress + pnum == size; 2226 if (status & BDRV_BLOCK_ZERO) { 2227 NBDReply hdr; 2228 NBDStructuredReadHole chunk; 2229 struct iovec iov[] = { 2230 {.iov_base = &hdr}, 2231 {.iov_base = &chunk, .iov_len = sizeof(chunk)}, 2232 }; 2233 2234 trace_nbd_co_send_chunk_read_hole(request->cookie, 2235 offset + progress, pnum); 2236 set_be_chunk(client, iov, 2, 2237 final ? NBD_REPLY_FLAG_DONE : 0, 2238 NBD_REPLY_TYPE_OFFSET_HOLE, request); 2239 stq_be_p(&chunk.offset, offset + progress); 2240 stl_be_p(&chunk.length, pnum); 2241 ret = nbd_co_send_iov(client, iov, 2, errp); 2242 } else { 2243 ret = blk_co_pread(exp->common.blk, offset + progress, pnum, 2244 data + progress, 0); 2245 if (ret < 0) { 2246 error_setg_errno(errp, -ret, "reading from file failed"); 2247 break; 2248 } 2249 ret = nbd_co_send_chunk_read(client, request, offset + progress, 2250 data + progress, pnum, final, errp); 2251 } 2252 2253 if (ret < 0) { 2254 break; 2255 } 2256 progress += pnum; 2257 } 2258 return ret; 2259 } 2260 2261 typedef struct NBDExtentArray { 2262 NBDExtent64 *extents; 2263 unsigned int nb_alloc; 2264 unsigned int count; 2265 uint64_t total_length; 2266 bool extended; 2267 bool can_add; 2268 bool converted_to_be; 2269 } NBDExtentArray; 2270 2271 static NBDExtentArray *nbd_extent_array_new(unsigned int nb_alloc, 2272 NBDMode mode) 2273 { 2274 NBDExtentArray *ea = g_new0(NBDExtentArray, 1); 2275 2276 assert(mode >= NBD_MODE_STRUCTURED); 2277 ea->nb_alloc = nb_alloc; 2278 ea->extents = g_new(NBDExtent64, nb_alloc); 2279 ea->extended = mode >= NBD_MODE_EXTENDED; 2280 ea->can_add = true; 2281 2282 return ea; 2283 } 2284 2285 static void nbd_extent_array_free(NBDExtentArray *ea) 2286 { 2287 g_free(ea->extents); 2288 g_free(ea); 2289 } 2290 G_DEFINE_AUTOPTR_CLEANUP_FUNC(NBDExtentArray, nbd_extent_array_free) 2291 2292 /* Further modifications of the array after conversion are abandoned */ 2293 static void nbd_extent_array_convert_to_be(NBDExtentArray *ea) 2294 { 2295 int i; 2296 2297 assert(!ea->converted_to_be); 2298 assert(ea->extended); 2299 ea->can_add = false; 2300 ea->converted_to_be = true; 2301 2302 for (i = 0; i < ea->count; i++) { 2303 ea->extents[i].length = cpu_to_be64(ea->extents[i].length); 2304 ea->extents[i].flags = cpu_to_be64(ea->extents[i].flags); 2305 } 2306 } 2307 2308 /* Further modifications of the array after conversion are abandoned */ 2309 static NBDExtent32 *nbd_extent_array_convert_to_narrow(NBDExtentArray *ea) 2310 { 2311 int i; 2312 NBDExtent32 *extents = g_new(NBDExtent32, ea->count); 2313 2314 assert(!ea->converted_to_be); 2315 assert(!ea->extended); 2316 ea->can_add = false; 2317 ea->converted_to_be = true; 2318 2319 for (i = 0; i < ea->count; i++) { 2320 assert((ea->extents[i].length | ea->extents[i].flags) <= UINT32_MAX); 2321 extents[i].length = cpu_to_be32(ea->extents[i].length); 2322 extents[i].flags = cpu_to_be32(ea->extents[i].flags); 2323 } 2324 2325 return extents; 2326 } 2327 2328 /* 2329 * Add extent to NBDExtentArray. If extent can't be added (no available space), 2330 * return -1. 2331 * For safety, when returning -1 for the first time, .can_add is set to false, 2332 * and further calls to nbd_extent_array_add() will crash. 2333 * (this avoids the situation where a caller ignores failure to add one extent, 2334 * where adding another extent that would squash into the last array entry 2335 * would result in an incorrect range reported to the client) 2336 */ 2337 static int nbd_extent_array_add(NBDExtentArray *ea, 2338 uint64_t length, uint32_t flags) 2339 { 2340 assert(ea->can_add); 2341 2342 if (!length) { 2343 return 0; 2344 } 2345 if (!ea->extended) { 2346 assert(length <= UINT32_MAX); 2347 } 2348 2349 /* Extend previous extent if flags are the same */ 2350 if (ea->count > 0 && flags == ea->extents[ea->count - 1].flags) { 2351 uint64_t sum = length + ea->extents[ea->count - 1].length; 2352 2353 /* 2354 * sum cannot overflow: the block layer bounds image size at 2355 * 2^63, and ea->extents[].length comes from the block layer. 2356 */ 2357 assert(sum >= length); 2358 if (sum <= UINT32_MAX || ea->extended) { 2359 ea->extents[ea->count - 1].length = sum; 2360 ea->total_length += length; 2361 return 0; 2362 } 2363 } 2364 2365 if (ea->count >= ea->nb_alloc) { 2366 ea->can_add = false; 2367 return -1; 2368 } 2369 2370 ea->total_length += length; 2371 ea->extents[ea->count] = (NBDExtent64) {.length = length, .flags = flags}; 2372 ea->count++; 2373 2374 return 0; 2375 } 2376 2377 static int coroutine_fn blockstatus_to_extents(BlockBackend *blk, 2378 uint64_t offset, uint64_t bytes, 2379 NBDExtentArray *ea) 2380 { 2381 while (bytes) { 2382 uint32_t flags; 2383 int64_t num; 2384 int ret = blk_co_block_status_above(blk, NULL, offset, bytes, &num, 2385 NULL, NULL); 2386 2387 if (ret < 0) { 2388 return ret; 2389 } 2390 2391 flags = (ret & BDRV_BLOCK_DATA ? 0 : NBD_STATE_HOLE) | 2392 (ret & BDRV_BLOCK_ZERO ? NBD_STATE_ZERO : 0); 2393 2394 if (nbd_extent_array_add(ea, num, flags) < 0) { 2395 return 0; 2396 } 2397 2398 offset += num; 2399 bytes -= num; 2400 } 2401 2402 return 0; 2403 } 2404 2405 static int coroutine_fn blockalloc_to_extents(BlockBackend *blk, 2406 uint64_t offset, uint64_t bytes, 2407 NBDExtentArray *ea) 2408 { 2409 while (bytes) { 2410 int64_t num; 2411 int ret = blk_co_is_allocated_above(blk, NULL, false, offset, bytes, 2412 &num); 2413 2414 if (ret < 0) { 2415 return ret; 2416 } 2417 2418 if (nbd_extent_array_add(ea, num, ret) < 0) { 2419 return 0; 2420 } 2421 2422 offset += num; 2423 bytes -= num; 2424 } 2425 2426 return 0; 2427 } 2428 2429 /* 2430 * nbd_co_send_extents 2431 * 2432 * @ea is converted to BE by the function 2433 * @last controls whether NBD_REPLY_FLAG_DONE is sent. 2434 */ 2435 static int coroutine_fn 2436 nbd_co_send_extents(NBDClient *client, NBDRequest *request, NBDExtentArray *ea, 2437 bool last, uint32_t context_id, Error **errp) 2438 { 2439 NBDReply hdr; 2440 NBDStructuredMeta meta; 2441 NBDExtendedMeta meta_ext; 2442 g_autofree NBDExtent32 *extents = NULL; 2443 uint16_t type; 2444 struct iovec iov[] = { {.iov_base = &hdr}, {0}, {0} }; 2445 2446 if (client->mode >= NBD_MODE_EXTENDED) { 2447 type = NBD_REPLY_TYPE_BLOCK_STATUS_EXT; 2448 2449 iov[1].iov_base = &meta_ext; 2450 iov[1].iov_len = sizeof(meta_ext); 2451 stl_be_p(&meta_ext.context_id, context_id); 2452 stl_be_p(&meta_ext.count, ea->count); 2453 2454 nbd_extent_array_convert_to_be(ea); 2455 iov[2].iov_base = ea->extents; 2456 iov[2].iov_len = ea->count * sizeof(ea->extents[0]); 2457 } else { 2458 type = NBD_REPLY_TYPE_BLOCK_STATUS; 2459 2460 iov[1].iov_base = &meta; 2461 iov[1].iov_len = sizeof(meta); 2462 stl_be_p(&meta.context_id, context_id); 2463 2464 extents = nbd_extent_array_convert_to_narrow(ea); 2465 iov[2].iov_base = extents; 2466 iov[2].iov_len = ea->count * sizeof(extents[0]); 2467 } 2468 2469 trace_nbd_co_send_extents(request->cookie, ea->count, context_id, 2470 ea->total_length, last); 2471 set_be_chunk(client, iov, 3, last ? NBD_REPLY_FLAG_DONE : 0, type, 2472 request); 2473 2474 return nbd_co_send_iov(client, iov, 3, errp); 2475 } 2476 2477 /* Get block status from the exported device and send it to the client */ 2478 static int 2479 coroutine_fn nbd_co_send_block_status(NBDClient *client, NBDRequest *request, 2480 BlockBackend *blk, uint64_t offset, 2481 uint64_t length, bool dont_fragment, 2482 bool last, uint32_t context_id, 2483 Error **errp) 2484 { 2485 int ret; 2486 unsigned int nb_extents = dont_fragment ? 1 : NBD_MAX_BLOCK_STATUS_EXTENTS; 2487 g_autoptr(NBDExtentArray) ea = 2488 nbd_extent_array_new(nb_extents, client->mode); 2489 2490 if (context_id == NBD_META_ID_BASE_ALLOCATION) { 2491 ret = blockstatus_to_extents(blk, offset, length, ea); 2492 } else { 2493 ret = blockalloc_to_extents(blk, offset, length, ea); 2494 } 2495 if (ret < 0) { 2496 return nbd_co_send_chunk_error(client, request, -ret, 2497 "can't get block status", errp); 2498 } 2499 2500 return nbd_co_send_extents(client, request, ea, last, context_id, errp); 2501 } 2502 2503 /* Populate @ea from a dirty bitmap. */ 2504 static void bitmap_to_extents(BdrvDirtyBitmap *bitmap, 2505 uint64_t offset, uint64_t length, 2506 NBDExtentArray *es) 2507 { 2508 int64_t start, dirty_start, dirty_count; 2509 int64_t end = offset + length; 2510 bool full = false; 2511 int64_t bound = es->extended ? INT64_MAX : INT32_MAX; 2512 2513 bdrv_dirty_bitmap_lock(bitmap); 2514 2515 for (start = offset; 2516 bdrv_dirty_bitmap_next_dirty_area(bitmap, start, end, bound, 2517 &dirty_start, &dirty_count); 2518 start = dirty_start + dirty_count) 2519 { 2520 if ((nbd_extent_array_add(es, dirty_start - start, 0) < 0) || 2521 (nbd_extent_array_add(es, dirty_count, NBD_STATE_DIRTY) < 0)) 2522 { 2523 full = true; 2524 break; 2525 } 2526 } 2527 2528 if (!full) { 2529 /* last non dirty extent, nothing to do if array is now full */ 2530 (void) nbd_extent_array_add(es, end - start, 0); 2531 } 2532 2533 bdrv_dirty_bitmap_unlock(bitmap); 2534 } 2535 2536 static int coroutine_fn nbd_co_send_bitmap(NBDClient *client, 2537 NBDRequest *request, 2538 BdrvDirtyBitmap *bitmap, 2539 uint64_t offset, 2540 uint64_t length, bool dont_fragment, 2541 bool last, uint32_t context_id, 2542 Error **errp) 2543 { 2544 unsigned int nb_extents = dont_fragment ? 1 : NBD_MAX_BLOCK_STATUS_EXTENTS; 2545 g_autoptr(NBDExtentArray) ea = 2546 nbd_extent_array_new(nb_extents, client->mode); 2547 2548 bitmap_to_extents(bitmap, offset, length, ea); 2549 2550 return nbd_co_send_extents(client, request, ea, last, context_id, errp); 2551 } 2552 2553 /* 2554 * nbd_co_block_status_payload_read 2555 * Called when a client wants a subset of negotiated contexts via a 2556 * BLOCK_STATUS payload. Check the payload for valid length and 2557 * contents. On success, return 0 with request updated to effective 2558 * length. If request was invalid but all payload consumed, return 0 2559 * with request->len and request->contexts->count set to 0 (which will 2560 * trigger an appropriate NBD_EINVAL response later on). Return 2561 * negative errno if the payload was not fully consumed. 2562 */ 2563 static int 2564 nbd_co_block_status_payload_read(NBDClient *client, NBDRequest *request, 2565 Error **errp) 2566 { 2567 uint64_t payload_len = request->len; 2568 g_autofree char *buf = NULL; 2569 size_t count, i, nr_bitmaps; 2570 uint32_t id; 2571 2572 if (payload_len > NBD_MAX_BUFFER_SIZE) { 2573 error_setg(errp, "len (%" PRIu64 ") is larger than max len (%u)", 2574 request->len, NBD_MAX_BUFFER_SIZE); 2575 return -EINVAL; 2576 } 2577 2578 assert(client->contexts.exp == client->exp); 2579 nr_bitmaps = client->exp->nr_export_bitmaps; 2580 request->contexts = g_new0(NBDMetaContexts, 1); 2581 request->contexts->exp = client->exp; 2582 2583 if (payload_len % sizeof(uint32_t) || 2584 payload_len < sizeof(NBDBlockStatusPayload) || 2585 payload_len > (sizeof(NBDBlockStatusPayload) + 2586 sizeof(id) * client->contexts.count)) { 2587 goto skip; 2588 } 2589 2590 buf = g_malloc(payload_len); 2591 if (nbd_read(client->ioc, buf, payload_len, 2592 "CMD_BLOCK_STATUS data", errp) < 0) { 2593 return -EIO; 2594 } 2595 trace_nbd_co_receive_request_payload_received(request->cookie, 2596 payload_len); 2597 request->contexts->bitmaps = g_new0(bool, nr_bitmaps); 2598 count = (payload_len - sizeof(NBDBlockStatusPayload)) / sizeof(id); 2599 payload_len = 0; 2600 2601 for (i = 0; i < count; i++) { 2602 id = ldl_be_p(buf + sizeof(NBDBlockStatusPayload) + sizeof(id) * i); 2603 if (id == NBD_META_ID_BASE_ALLOCATION) { 2604 if (!client->contexts.base_allocation || 2605 request->contexts->base_allocation) { 2606 goto skip; 2607 } 2608 request->contexts->base_allocation = true; 2609 } else if (id == NBD_META_ID_ALLOCATION_DEPTH) { 2610 if (!client->contexts.allocation_depth || 2611 request->contexts->allocation_depth) { 2612 goto skip; 2613 } 2614 request->contexts->allocation_depth = true; 2615 } else { 2616 unsigned idx = id - NBD_META_ID_DIRTY_BITMAP; 2617 2618 if (idx >= nr_bitmaps || !client->contexts.bitmaps[idx] || 2619 request->contexts->bitmaps[idx]) { 2620 goto skip; 2621 } 2622 request->contexts->bitmaps[idx] = true; 2623 } 2624 } 2625 2626 request->len = ldq_be_p(buf); 2627 request->contexts->count = count; 2628 return 0; 2629 2630 skip: 2631 trace_nbd_co_receive_block_status_payload_compliance(request->from, 2632 request->len); 2633 request->len = request->contexts->count = 0; 2634 return nbd_drop(client->ioc, payload_len, errp); 2635 } 2636 2637 /* nbd_co_receive_request 2638 * Collect a client request. Return 0 if request looks valid, -EIO to drop 2639 * connection right away, -EAGAIN to indicate we were interrupted and the 2640 * channel should be quiesced, and any other negative value to report an error 2641 * to the client (although the caller may still need to disconnect after 2642 * reporting the error). 2643 */ 2644 static int coroutine_fn nbd_co_receive_request(NBDRequestData *req, 2645 NBDRequest *request, 2646 Error **errp) 2647 { 2648 NBDClient *client = req->client; 2649 bool extended_with_payload; 2650 bool check_length = false; 2651 bool check_rofs = false; 2652 bool allocate_buffer = false; 2653 bool payload_okay = false; 2654 uint64_t payload_len = 0; 2655 int valid_flags = NBD_CMD_FLAG_FUA; 2656 int ret; 2657 2658 g_assert(qemu_in_coroutine()); 2659 ret = nbd_receive_request(client, request, errp); 2660 if (ret < 0) { 2661 return ret; 2662 } 2663 2664 trace_nbd_co_receive_request_decode_type(request->cookie, request->type, 2665 nbd_cmd_lookup(request->type)); 2666 extended_with_payload = client->mode >= NBD_MODE_EXTENDED && 2667 request->flags & NBD_CMD_FLAG_PAYLOAD_LEN; 2668 if (extended_with_payload) { 2669 payload_len = request->len; 2670 check_length = true; 2671 } 2672 2673 switch (request->type) { 2674 case NBD_CMD_DISC: 2675 /* Special case: we're going to disconnect without a reply, 2676 * whether or not flags, from, or len are bogus */ 2677 req->complete = true; 2678 return -EIO; 2679 2680 case NBD_CMD_READ: 2681 if (client->mode >= NBD_MODE_STRUCTURED) { 2682 valid_flags |= NBD_CMD_FLAG_DF; 2683 } 2684 check_length = true; 2685 allocate_buffer = true; 2686 break; 2687 2688 case NBD_CMD_WRITE: 2689 if (client->mode >= NBD_MODE_EXTENDED) { 2690 if (!extended_with_payload) { 2691 /* The client is noncompliant. Trace it, but proceed. */ 2692 trace_nbd_co_receive_ext_payload_compliance(request->from, 2693 request->len); 2694 } 2695 valid_flags |= NBD_CMD_FLAG_PAYLOAD_LEN; 2696 } 2697 payload_okay = true; 2698 payload_len = request->len; 2699 check_length = true; 2700 allocate_buffer = true; 2701 check_rofs = true; 2702 break; 2703 2704 case NBD_CMD_FLUSH: 2705 break; 2706 2707 case NBD_CMD_TRIM: 2708 check_rofs = true; 2709 break; 2710 2711 case NBD_CMD_CACHE: 2712 check_length = true; 2713 break; 2714 2715 case NBD_CMD_WRITE_ZEROES: 2716 valid_flags |= NBD_CMD_FLAG_NO_HOLE | NBD_CMD_FLAG_FAST_ZERO; 2717 check_rofs = true; 2718 break; 2719 2720 case NBD_CMD_BLOCK_STATUS: 2721 if (extended_with_payload) { 2722 ret = nbd_co_block_status_payload_read(client, request, errp); 2723 if (ret < 0) { 2724 return ret; 2725 } 2726 /* payload now consumed */ 2727 check_length = false; 2728 payload_len = 0; 2729 valid_flags |= NBD_CMD_FLAG_PAYLOAD_LEN; 2730 } else { 2731 request->contexts = &client->contexts; 2732 } 2733 valid_flags |= NBD_CMD_FLAG_REQ_ONE; 2734 break; 2735 2736 default: 2737 /* Unrecognized, will fail later */ 2738 ; 2739 } 2740 2741 /* Payload and buffer handling. */ 2742 if (!payload_len) { 2743 req->complete = true; 2744 } 2745 if (check_length && request->len > NBD_MAX_BUFFER_SIZE) { 2746 /* READ, WRITE, CACHE */ 2747 error_setg(errp, "len (%" PRIu64 ") is larger than max len (%u)", 2748 request->len, NBD_MAX_BUFFER_SIZE); 2749 return -EINVAL; 2750 } 2751 if (payload_len && !payload_okay) { 2752 /* 2753 * For now, we don't support payloads on other commands; but 2754 * we can keep the connection alive by ignoring the payload. 2755 * We will fail the command later with NBD_EINVAL for the use 2756 * of an unsupported flag (and not for access beyond bounds). 2757 */ 2758 assert(request->type != NBD_CMD_WRITE); 2759 request->len = 0; 2760 } 2761 if (allocate_buffer) { 2762 /* READ, WRITE */ 2763 req->data = blk_try_blockalign(client->exp->common.blk, 2764 request->len); 2765 if (req->data == NULL) { 2766 error_setg(errp, "No memory"); 2767 return -ENOMEM; 2768 } 2769 } 2770 if (payload_len) { 2771 if (payload_okay) { 2772 /* WRITE */ 2773 assert(req->data); 2774 ret = nbd_read(client->ioc, req->data, payload_len, 2775 "CMD_WRITE data", errp); 2776 } else { 2777 ret = nbd_drop(client->ioc, payload_len, errp); 2778 } 2779 if (ret < 0) { 2780 return -EIO; 2781 } 2782 req->complete = true; 2783 trace_nbd_co_receive_request_payload_received(request->cookie, 2784 payload_len); 2785 } 2786 2787 /* Sanity checks. */ 2788 if (client->exp->nbdflags & NBD_FLAG_READ_ONLY && check_rofs) { 2789 /* WRITE, TRIM, WRITE_ZEROES */ 2790 error_setg(errp, "Export is read-only"); 2791 return -EROFS; 2792 } 2793 if (request->from > client->exp->size || 2794 request->len > client->exp->size - request->from) { 2795 error_setg(errp, "operation past EOF; From: %" PRIu64 ", Len: %" PRIu64 2796 ", Size: %" PRIu64, request->from, request->len, 2797 client->exp->size); 2798 return (request->type == NBD_CMD_WRITE || 2799 request->type == NBD_CMD_WRITE_ZEROES) ? -ENOSPC : -EINVAL; 2800 } 2801 if (client->check_align && !QEMU_IS_ALIGNED(request->from | request->len, 2802 client->check_align)) { 2803 /* 2804 * The block layer gracefully handles unaligned requests, but 2805 * it's still worth tracing client non-compliance 2806 */ 2807 trace_nbd_co_receive_align_compliance(nbd_cmd_lookup(request->type), 2808 request->from, 2809 request->len, 2810 client->check_align); 2811 } 2812 if (request->flags & ~valid_flags) { 2813 error_setg(errp, "unsupported flags for command %s (got 0x%x)", 2814 nbd_cmd_lookup(request->type), request->flags); 2815 return -EINVAL; 2816 } 2817 2818 return 0; 2819 } 2820 2821 /* Send simple reply without a payload, or a structured error 2822 * @error_msg is ignored if @ret >= 0 2823 * Returns 0 if connection is still live, -errno on failure to talk to client 2824 */ 2825 static coroutine_fn int nbd_send_generic_reply(NBDClient *client, 2826 NBDRequest *request, 2827 int ret, 2828 const char *error_msg, 2829 Error **errp) 2830 { 2831 if (client->mode >= NBD_MODE_STRUCTURED && ret < 0) { 2832 return nbd_co_send_chunk_error(client, request, -ret, error_msg, errp); 2833 } else if (client->mode >= NBD_MODE_EXTENDED) { 2834 return nbd_co_send_chunk_done(client, request, errp); 2835 } else { 2836 return nbd_co_send_simple_reply(client, request, ret < 0 ? -ret : 0, 2837 NULL, 0, errp); 2838 } 2839 } 2840 2841 /* Handle NBD_CMD_READ request. 2842 * Return -errno if sending fails. Other errors are reported directly to the 2843 * client as an error reply. */ 2844 static coroutine_fn int nbd_do_cmd_read(NBDClient *client, NBDRequest *request, 2845 uint8_t *data, Error **errp) 2846 { 2847 int ret; 2848 NBDExport *exp = client->exp; 2849 2850 assert(request->type == NBD_CMD_READ); 2851 assert(request->len <= NBD_MAX_BUFFER_SIZE); 2852 2853 /* XXX: NBD Protocol only documents use of FUA with WRITE */ 2854 if (request->flags & NBD_CMD_FLAG_FUA) { 2855 ret = blk_co_flush(exp->common.blk); 2856 if (ret < 0) { 2857 return nbd_send_generic_reply(client, request, ret, 2858 "flush failed", errp); 2859 } 2860 } 2861 2862 if (client->mode >= NBD_MODE_STRUCTURED && 2863 !(request->flags & NBD_CMD_FLAG_DF) && request->len) 2864 { 2865 return nbd_co_send_sparse_read(client, request, request->from, 2866 data, request->len, errp); 2867 } 2868 2869 ret = blk_co_pread(exp->common.blk, request->from, request->len, data, 0); 2870 if (ret < 0) { 2871 return nbd_send_generic_reply(client, request, ret, 2872 "reading from file failed", errp); 2873 } 2874 2875 if (client->mode >= NBD_MODE_STRUCTURED) { 2876 if (request->len) { 2877 return nbd_co_send_chunk_read(client, request, request->from, data, 2878 request->len, true, errp); 2879 } else { 2880 return nbd_co_send_chunk_done(client, request, errp); 2881 } 2882 } else { 2883 return nbd_co_send_simple_reply(client, request, 0, 2884 data, request->len, errp); 2885 } 2886 } 2887 2888 /* 2889 * nbd_do_cmd_cache 2890 * 2891 * Handle NBD_CMD_CACHE request. 2892 * Return -errno if sending fails. Other errors are reported directly to the 2893 * client as an error reply. 2894 */ 2895 static coroutine_fn int nbd_do_cmd_cache(NBDClient *client, NBDRequest *request, 2896 Error **errp) 2897 { 2898 int ret; 2899 NBDExport *exp = client->exp; 2900 2901 assert(request->type == NBD_CMD_CACHE); 2902 assert(request->len <= NBD_MAX_BUFFER_SIZE); 2903 2904 ret = blk_co_preadv(exp->common.blk, request->from, request->len, 2905 NULL, BDRV_REQ_COPY_ON_READ | BDRV_REQ_PREFETCH); 2906 2907 return nbd_send_generic_reply(client, request, ret, 2908 "caching data failed", errp); 2909 } 2910 2911 /* Handle NBD request. 2912 * Return -errno if sending fails. Other errors are reported directly to the 2913 * client as an error reply. */ 2914 static coroutine_fn int nbd_handle_request(NBDClient *client, 2915 NBDRequest *request, 2916 uint8_t *data, Error **errp) 2917 { 2918 int ret; 2919 int flags; 2920 NBDExport *exp = client->exp; 2921 char *msg; 2922 size_t i; 2923 2924 switch (request->type) { 2925 case NBD_CMD_CACHE: 2926 return nbd_do_cmd_cache(client, request, errp); 2927 2928 case NBD_CMD_READ: 2929 return nbd_do_cmd_read(client, request, data, errp); 2930 2931 case NBD_CMD_WRITE: 2932 flags = 0; 2933 if (request->flags & NBD_CMD_FLAG_FUA) { 2934 flags |= BDRV_REQ_FUA; 2935 } 2936 assert(request->len <= NBD_MAX_BUFFER_SIZE); 2937 ret = blk_co_pwrite(exp->common.blk, request->from, request->len, data, 2938 flags); 2939 return nbd_send_generic_reply(client, request, ret, 2940 "writing to file failed", errp); 2941 2942 case NBD_CMD_WRITE_ZEROES: 2943 flags = 0; 2944 if (request->flags & NBD_CMD_FLAG_FUA) { 2945 flags |= BDRV_REQ_FUA; 2946 } 2947 if (!(request->flags & NBD_CMD_FLAG_NO_HOLE)) { 2948 flags |= BDRV_REQ_MAY_UNMAP; 2949 } 2950 if (request->flags & NBD_CMD_FLAG_FAST_ZERO) { 2951 flags |= BDRV_REQ_NO_FALLBACK; 2952 } 2953 ret = blk_co_pwrite_zeroes(exp->common.blk, request->from, request->len, 2954 flags); 2955 return nbd_send_generic_reply(client, request, ret, 2956 "writing to file failed", errp); 2957 2958 case NBD_CMD_DISC: 2959 /* unreachable, thanks to special case in nbd_co_receive_request() */ 2960 abort(); 2961 2962 case NBD_CMD_FLUSH: 2963 ret = blk_co_flush(exp->common.blk); 2964 return nbd_send_generic_reply(client, request, ret, 2965 "flush failed", errp); 2966 2967 case NBD_CMD_TRIM: 2968 ret = blk_co_pdiscard(exp->common.blk, request->from, request->len); 2969 if (ret >= 0 && request->flags & NBD_CMD_FLAG_FUA) { 2970 ret = blk_co_flush(exp->common.blk); 2971 } 2972 return nbd_send_generic_reply(client, request, ret, 2973 "discard failed", errp); 2974 2975 case NBD_CMD_BLOCK_STATUS: 2976 assert(request->contexts); 2977 assert(client->mode >= NBD_MODE_EXTENDED || 2978 request->len <= UINT32_MAX); 2979 if (request->contexts->count) { 2980 bool dont_fragment = request->flags & NBD_CMD_FLAG_REQ_ONE; 2981 int contexts_remaining = request->contexts->count; 2982 2983 if (!request->len) { 2984 return nbd_send_generic_reply(client, request, -EINVAL, 2985 "need non-zero length", errp); 2986 } 2987 if (request->contexts->base_allocation) { 2988 ret = nbd_co_send_block_status(client, request, 2989 exp->common.blk, 2990 request->from, 2991 request->len, dont_fragment, 2992 !--contexts_remaining, 2993 NBD_META_ID_BASE_ALLOCATION, 2994 errp); 2995 if (ret < 0) { 2996 return ret; 2997 } 2998 } 2999 3000 if (request->contexts->allocation_depth) { 3001 ret = nbd_co_send_block_status(client, request, 3002 exp->common.blk, 3003 request->from, request->len, 3004 dont_fragment, 3005 !--contexts_remaining, 3006 NBD_META_ID_ALLOCATION_DEPTH, 3007 errp); 3008 if (ret < 0) { 3009 return ret; 3010 } 3011 } 3012 3013 assert(request->contexts->exp == client->exp); 3014 for (i = 0; i < client->exp->nr_export_bitmaps; i++) { 3015 if (!request->contexts->bitmaps[i]) { 3016 continue; 3017 } 3018 ret = nbd_co_send_bitmap(client, request, 3019 client->exp->export_bitmaps[i], 3020 request->from, request->len, 3021 dont_fragment, !--contexts_remaining, 3022 NBD_META_ID_DIRTY_BITMAP + i, errp); 3023 if (ret < 0) { 3024 return ret; 3025 } 3026 } 3027 3028 assert(!contexts_remaining); 3029 3030 return 0; 3031 } else if (client->contexts.count) { 3032 return nbd_send_generic_reply(client, request, -EINVAL, 3033 "CMD_BLOCK_STATUS payload not valid", 3034 errp); 3035 } else { 3036 return nbd_send_generic_reply(client, request, -EINVAL, 3037 "CMD_BLOCK_STATUS not negotiated", 3038 errp); 3039 } 3040 3041 default: 3042 msg = g_strdup_printf("invalid request type (%" PRIu32 ") received", 3043 request->type); 3044 ret = nbd_send_generic_reply(client, request, -EINVAL, msg, 3045 errp); 3046 g_free(msg); 3047 return ret; 3048 } 3049 } 3050 3051 /* Owns a reference to the NBDClient passed as opaque. */ 3052 static coroutine_fn void nbd_trip(void *opaque) 3053 { 3054 NBDRequestData *req = opaque; 3055 NBDClient *client = req->client; 3056 NBDRequest request = { 0 }; /* GCC thinks it can be used uninitialized */ 3057 int ret; 3058 Error *local_err = NULL; 3059 3060 /* 3061 * Note that nbd_client_put() and client_close() must be called from the 3062 * main loop thread. Use aio_co_reschedule_self() to switch AioContext 3063 * before calling these functions. 3064 */ 3065 3066 trace_nbd_trip(); 3067 3068 qemu_mutex_lock(&client->lock); 3069 3070 if (client->closing) { 3071 goto done; 3072 } 3073 3074 if (client->quiescing) { 3075 /* 3076 * We're switching between AIO contexts. Don't attempt to receive a new 3077 * request and kick the main context which may be waiting for us. 3078 */ 3079 client->recv_coroutine = NULL; 3080 aio_wait_kick(); 3081 goto done; 3082 } 3083 3084 /* 3085 * nbd_co_receive_request() returns -EAGAIN when nbd_drained_begin() has 3086 * set client->quiescing but by the time we get back nbd_drained_end() may 3087 * have already cleared client->quiescing. In that case we try again 3088 * because nothing else will spawn an nbd_trip() coroutine until we set 3089 * client->recv_coroutine = NULL further down. 3090 */ 3091 do { 3092 assert(client->recv_coroutine == qemu_coroutine_self()); 3093 qemu_mutex_unlock(&client->lock); 3094 ret = nbd_co_receive_request(req, &request, &local_err); 3095 qemu_mutex_lock(&client->lock); 3096 } while (ret == -EAGAIN && !client->quiescing); 3097 3098 client->recv_coroutine = NULL; 3099 3100 if (client->closing) { 3101 /* 3102 * The client may be closed when we are blocked in 3103 * nbd_co_receive_request() 3104 */ 3105 goto done; 3106 } 3107 3108 if (ret == -EAGAIN) { 3109 goto done; 3110 } 3111 3112 nbd_client_receive_next_request(client); 3113 3114 if (ret == -EIO) { 3115 goto disconnect; 3116 } 3117 3118 qemu_mutex_unlock(&client->lock); 3119 qio_channel_set_cork(client->ioc, true); 3120 3121 if (ret < 0) { 3122 /* It wasn't -EIO, so, according to nbd_co_receive_request() 3123 * semantics, we should return the error to the client. */ 3124 Error *export_err = local_err; 3125 3126 local_err = NULL; 3127 ret = nbd_send_generic_reply(client, &request, -EINVAL, 3128 error_get_pretty(export_err), &local_err); 3129 error_free(export_err); 3130 } else { 3131 ret = nbd_handle_request(client, &request, req->data, &local_err); 3132 } 3133 if (request.contexts && request.contexts != &client->contexts) { 3134 assert(request.type == NBD_CMD_BLOCK_STATUS); 3135 g_free(request.contexts->bitmaps); 3136 g_free(request.contexts); 3137 } 3138 3139 qio_channel_set_cork(client->ioc, false); 3140 qemu_mutex_lock(&client->lock); 3141 3142 if (ret < 0) { 3143 error_prepend(&local_err, "Failed to send reply: "); 3144 goto disconnect; 3145 } 3146 3147 /* 3148 * We must disconnect after NBD_CMD_WRITE or BLOCK_STATUS with 3149 * payload if we did not read the payload. 3150 */ 3151 if (!req->complete) { 3152 error_setg(&local_err, "Request handling failed in intermediate state"); 3153 goto disconnect; 3154 } 3155 3156 done: 3157 nbd_request_put(req); 3158 3159 qemu_mutex_unlock(&client->lock); 3160 3161 if (!nbd_client_put_nonzero(client)) { 3162 aio_co_reschedule_self(qemu_get_aio_context()); 3163 nbd_client_put(client); 3164 } 3165 return; 3166 3167 disconnect: 3168 if (local_err) { 3169 error_reportf_err(local_err, "Disconnect client, due to: "); 3170 } 3171 3172 nbd_request_put(req); 3173 qemu_mutex_unlock(&client->lock); 3174 3175 aio_co_reschedule_self(qemu_get_aio_context()); 3176 client_close(client, true); 3177 nbd_client_put(client); 3178 } 3179 3180 /* 3181 * Runs in export AioContext and main loop thread. Caller must hold 3182 * client->lock. 3183 */ 3184 static void nbd_client_receive_next_request(NBDClient *client) 3185 { 3186 NBDRequestData *req; 3187 3188 if (!client->recv_coroutine && client->nb_requests < MAX_NBD_REQUESTS && 3189 !client->quiescing) { 3190 nbd_client_get(client); 3191 req = nbd_request_get(client); 3192 client->recv_coroutine = qemu_coroutine_create(nbd_trip, req); 3193 aio_co_schedule(client->exp->common.ctx, client->recv_coroutine); 3194 } 3195 } 3196 3197 static void nbd_handshake_timer_cb(void *opaque) 3198 { 3199 QIOChannel *ioc = opaque; 3200 3201 trace_nbd_handshake_timer_cb(); 3202 qio_channel_shutdown(ioc, QIO_CHANNEL_SHUTDOWN_BOTH, NULL); 3203 } 3204 3205 static coroutine_fn void nbd_co_client_start(void *opaque) 3206 { 3207 NBDClient *client = opaque; 3208 Error *local_err = NULL; 3209 QEMUTimer *handshake_timer = NULL; 3210 3211 qemu_co_mutex_init(&client->send_lock); 3212 3213 /* 3214 * Create a timer to bound the time spent in negotiation. If the 3215 * timer expires, it is likely nbd_negotiate will fail because the 3216 * socket was shutdown. 3217 */ 3218 if (client->handshake_max_secs > 0) { 3219 handshake_timer = aio_timer_new(qemu_get_aio_context(), 3220 QEMU_CLOCK_REALTIME, 3221 SCALE_NS, 3222 nbd_handshake_timer_cb, 3223 client->sioc); 3224 timer_mod(handshake_timer, 3225 qemu_clock_get_ns(QEMU_CLOCK_REALTIME) + 3226 client->handshake_max_secs * NANOSECONDS_PER_SECOND); 3227 } 3228 3229 if (nbd_negotiate(client, &local_err)) { 3230 if (local_err) { 3231 error_report_err(local_err); 3232 } 3233 timer_free(handshake_timer); 3234 client_close(client, false); 3235 return; 3236 } 3237 3238 timer_free(handshake_timer); 3239 WITH_QEMU_LOCK_GUARD(&client->lock) { 3240 nbd_client_receive_next_request(client); 3241 } 3242 } 3243 3244 /* 3245 * Create a new client listener using the given channel @sioc and @owner. 3246 * Begin servicing it in a coroutine. When the connection closes, call 3247 * @close_fn with an indication of whether the client completed negotiation 3248 * within @handshake_max_secs seconds (0 for unbounded). 3249 */ 3250 void nbd_client_new(QIOChannelSocket *sioc, 3251 uint32_t handshake_max_secs, 3252 QCryptoTLSCreds *tlscreds, 3253 const char *tlsauthz, 3254 void (*close_fn)(NBDClient *, bool), 3255 void *owner) 3256 { 3257 NBDClient *client; 3258 Coroutine *co; 3259 3260 client = g_new0(NBDClient, 1); 3261 qemu_mutex_init(&client->lock); 3262 client->refcount = 1; 3263 client->tlscreds = tlscreds; 3264 if (tlscreds) { 3265 object_ref(OBJECT(client->tlscreds)); 3266 } 3267 client->tlsauthz = g_strdup(tlsauthz); 3268 client->handshake_max_secs = handshake_max_secs; 3269 client->sioc = sioc; 3270 qio_channel_set_delay(QIO_CHANNEL(sioc), false); 3271 object_ref(OBJECT(client->sioc)); 3272 client->ioc = QIO_CHANNEL(sioc); 3273 object_ref(OBJECT(client->ioc)); 3274 client->close_fn = close_fn; 3275 client->owner = owner; 3276 3277 co = qemu_coroutine_create(nbd_co_client_start, client); 3278 qemu_coroutine_enter(co); 3279 } 3280 3281 void * 3282 nbd_client_owner(NBDClient *client) 3283 { 3284 return client->owner; 3285 } 3286