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); 127 128 QemuMutex lock; 129 130 NBDExport *exp; 131 QCryptoTLSCreds *tlscreds; 132 char *tlsauthz; 133 QIOChannelSocket *sioc; /* The underlying data channel */ 134 QIOChannel *ioc; /* The current I/O channel which may differ (eg TLS) */ 135 136 Coroutine *recv_coroutine; /* protected by lock */ 137 138 CoMutex send_lock; 139 Coroutine *send_coroutine; 140 141 bool read_yielding; /* protected by lock */ 142 bool quiescing; /* protected by lock */ 143 144 QTAILQ_ENTRY(NBDClient) next; 145 int nb_requests; /* protected by lock */ 146 bool closing; /* protected by lock */ 147 148 uint32_t check_align; /* If non-zero, check for aligned client requests */ 149 150 NBDMode mode; 151 NBDMetaContexts contexts; /* Negotiated meta contexts */ 152 153 uint32_t opt; /* Current option being negotiated */ 154 uint32_t optlen; /* remaining length of data in ioc for the option being 155 negotiated now */ 156 }; 157 158 static void nbd_client_receive_next_request(NBDClient *client); 159 160 /* Basic flow for negotiation 161 162 Server Client 163 Negotiate 164 165 or 166 167 Server Client 168 Negotiate #1 169 Option 170 Negotiate #2 171 172 ---- 173 174 followed by 175 176 Server Client 177 Request 178 Response 179 Request 180 Response 181 ... 182 ... 183 Request (type == 2) 184 185 */ 186 187 static inline void set_be_option_rep(NBDOptionReply *rep, uint32_t option, 188 uint32_t type, uint32_t length) 189 { 190 stq_be_p(&rep->magic, NBD_REP_MAGIC); 191 stl_be_p(&rep->option, option); 192 stl_be_p(&rep->type, type); 193 stl_be_p(&rep->length, length); 194 } 195 196 /* Send a reply header, including length, but no payload. 197 * Return -errno on error, 0 on success. */ 198 static coroutine_fn int 199 nbd_negotiate_send_rep_len(NBDClient *client, uint32_t type, 200 uint32_t len, Error **errp) 201 { 202 NBDOptionReply rep; 203 204 trace_nbd_negotiate_send_rep_len(client->opt, nbd_opt_lookup(client->opt), 205 type, nbd_rep_lookup(type), len); 206 207 assert(len < NBD_MAX_BUFFER_SIZE); 208 209 set_be_option_rep(&rep, client->opt, type, len); 210 return nbd_write(client->ioc, &rep, sizeof(rep), errp); 211 } 212 213 /* Send a reply header with default 0 length. 214 * Return -errno on error, 0 on success. */ 215 static coroutine_fn int 216 nbd_negotiate_send_rep(NBDClient *client, uint32_t type, Error **errp) 217 { 218 return nbd_negotiate_send_rep_len(client, type, 0, errp); 219 } 220 221 /* Send an error reply. 222 * Return -errno on error, 0 on success. */ 223 static coroutine_fn int G_GNUC_PRINTF(4, 0) 224 nbd_negotiate_send_rep_verr(NBDClient *client, uint32_t type, 225 Error **errp, const char *fmt, va_list va) 226 { 227 ERRP_GUARD(); 228 g_autofree char *msg = NULL; 229 int ret; 230 size_t len; 231 232 msg = g_strdup_vprintf(fmt, va); 233 len = strlen(msg); 234 assert(len < NBD_MAX_STRING_SIZE); 235 trace_nbd_negotiate_send_rep_err(msg); 236 ret = nbd_negotiate_send_rep_len(client, type, len, errp); 237 if (ret < 0) { 238 return ret; 239 } 240 if (nbd_write(client->ioc, msg, len, errp) < 0) { 241 error_prepend(errp, "write failed (error message): "); 242 return -EIO; 243 } 244 245 return 0; 246 } 247 248 /* 249 * Return a malloc'd copy of @name suitable for use in an error reply. 250 */ 251 static char * 252 nbd_sanitize_name(const char *name) 253 { 254 if (strnlen(name, 80) < 80) { 255 return g_strdup(name); 256 } 257 /* XXX Should we also try to sanitize any control characters? */ 258 return g_strdup_printf("%.80s...", name); 259 } 260 261 /* Send an error reply. 262 * Return -errno on error, 0 on success. */ 263 static coroutine_fn int G_GNUC_PRINTF(4, 5) 264 nbd_negotiate_send_rep_err(NBDClient *client, uint32_t type, 265 Error **errp, const char *fmt, ...) 266 { 267 va_list va; 268 int ret; 269 270 va_start(va, fmt); 271 ret = nbd_negotiate_send_rep_verr(client, type, errp, fmt, va); 272 va_end(va); 273 return ret; 274 } 275 276 /* Drop remainder of the current option, and send a reply with the 277 * given error type and message. Return -errno on read or write 278 * failure; or 0 if connection is still live. */ 279 static coroutine_fn int G_GNUC_PRINTF(4, 0) 280 nbd_opt_vdrop(NBDClient *client, uint32_t type, Error **errp, 281 const char *fmt, va_list va) 282 { 283 int ret = nbd_drop(client->ioc, client->optlen, errp); 284 285 client->optlen = 0; 286 if (!ret) { 287 ret = nbd_negotiate_send_rep_verr(client, type, errp, fmt, va); 288 } 289 return ret; 290 } 291 292 static coroutine_fn int G_GNUC_PRINTF(4, 5) 293 nbd_opt_drop(NBDClient *client, uint32_t type, Error **errp, 294 const char *fmt, ...) 295 { 296 int ret; 297 va_list va; 298 299 va_start(va, fmt); 300 ret = nbd_opt_vdrop(client, type, errp, fmt, va); 301 va_end(va); 302 303 return ret; 304 } 305 306 static coroutine_fn int G_GNUC_PRINTF(3, 4) 307 nbd_opt_invalid(NBDClient *client, Error **errp, const char *fmt, ...) 308 { 309 int ret; 310 va_list va; 311 312 va_start(va, fmt); 313 ret = nbd_opt_vdrop(client, NBD_REP_ERR_INVALID, errp, fmt, va); 314 va_end(va); 315 316 return ret; 317 } 318 319 /* Read size bytes from the unparsed payload of the current option. 320 * If @check_nul, require that no NUL bytes appear in buffer. 321 * Return -errno on I/O error, 0 if option was completely handled by 322 * sending a reply about inconsistent lengths, or 1 on success. */ 323 static coroutine_fn int 324 nbd_opt_read(NBDClient *client, void *buffer, size_t size, 325 bool check_nul, Error **errp) 326 { 327 if (size > client->optlen) { 328 return nbd_opt_invalid(client, errp, 329 "Inconsistent lengths in option %s", 330 nbd_opt_lookup(client->opt)); 331 } 332 client->optlen -= size; 333 if (qio_channel_read_all(client->ioc, buffer, size, errp) < 0) { 334 return -EIO; 335 } 336 337 if (check_nul && strnlen(buffer, size) != size) { 338 return nbd_opt_invalid(client, errp, 339 "Unexpected embedded NUL in option %s", 340 nbd_opt_lookup(client->opt)); 341 } 342 return 1; 343 } 344 345 /* Drop size bytes from the unparsed payload of the current option. 346 * Return -errno on I/O error, 0 if option was completely handled by 347 * sending a reply about inconsistent lengths, or 1 on success. */ 348 static coroutine_fn int 349 nbd_opt_skip(NBDClient *client, size_t size, Error **errp) 350 { 351 if (size > client->optlen) { 352 return nbd_opt_invalid(client, errp, 353 "Inconsistent lengths in option %s", 354 nbd_opt_lookup(client->opt)); 355 } 356 client->optlen -= size; 357 return nbd_drop(client->ioc, size, errp) < 0 ? -EIO : 1; 358 } 359 360 /* nbd_opt_read_name 361 * 362 * Read a string with the format: 363 * uint32_t len (<= NBD_MAX_STRING_SIZE) 364 * len bytes string (not 0-terminated) 365 * 366 * On success, @name will be allocated. 367 * If @length is non-null, it will be set to the actual string length. 368 * 369 * Return -errno on I/O error, 0 if option was completely handled by 370 * sending a reply about inconsistent lengths, or 1 on success. 371 */ 372 static coroutine_fn int 373 nbd_opt_read_name(NBDClient *client, char **name, uint32_t *length, 374 Error **errp) 375 { 376 int ret; 377 uint32_t len; 378 g_autofree char *local_name = NULL; 379 380 *name = NULL; 381 ret = nbd_opt_read(client, &len, sizeof(len), false, errp); 382 if (ret <= 0) { 383 return ret; 384 } 385 len = cpu_to_be32(len); 386 387 if (len > NBD_MAX_STRING_SIZE) { 388 return nbd_opt_invalid(client, errp, 389 "Invalid name length: %" PRIu32, len); 390 } 391 392 local_name = g_malloc(len + 1); 393 ret = nbd_opt_read(client, local_name, len, true, errp); 394 if (ret <= 0) { 395 return ret; 396 } 397 local_name[len] = '\0'; 398 399 if (length) { 400 *length = len; 401 } 402 *name = g_steal_pointer(&local_name); 403 404 return 1; 405 } 406 407 /* Send a single NBD_REP_SERVER reply to NBD_OPT_LIST, including payload. 408 * Return -errno on error, 0 on success. */ 409 static coroutine_fn int 410 nbd_negotiate_send_rep_list(NBDClient *client, NBDExport *exp, Error **errp) 411 { 412 ERRP_GUARD(); 413 size_t name_len, desc_len; 414 uint32_t len; 415 const char *name = exp->name ? exp->name : ""; 416 const char *desc = exp->description ? exp->description : ""; 417 QIOChannel *ioc = client->ioc; 418 int ret; 419 420 trace_nbd_negotiate_send_rep_list(name, desc); 421 name_len = strlen(name); 422 desc_len = strlen(desc); 423 assert(name_len <= NBD_MAX_STRING_SIZE && desc_len <= NBD_MAX_STRING_SIZE); 424 len = name_len + desc_len + sizeof(len); 425 ret = nbd_negotiate_send_rep_len(client, NBD_REP_SERVER, len, errp); 426 if (ret < 0) { 427 return ret; 428 } 429 430 len = cpu_to_be32(name_len); 431 if (nbd_write(ioc, &len, sizeof(len), errp) < 0) { 432 error_prepend(errp, "write failed (name length): "); 433 return -EINVAL; 434 } 435 436 if (nbd_write(ioc, name, name_len, errp) < 0) { 437 error_prepend(errp, "write failed (name buffer): "); 438 return -EINVAL; 439 } 440 441 if (nbd_write(ioc, desc, desc_len, errp) < 0) { 442 error_prepend(errp, "write failed (description buffer): "); 443 return -EINVAL; 444 } 445 446 return 0; 447 } 448 449 /* Process the NBD_OPT_LIST command, with a potential series of replies. 450 * Return -errno on error, 0 on success. */ 451 static coroutine_fn int 452 nbd_negotiate_handle_list(NBDClient *client, Error **errp) 453 { 454 NBDExport *exp; 455 assert(client->opt == NBD_OPT_LIST); 456 457 /* For each export, send a NBD_REP_SERVER reply. */ 458 QTAILQ_FOREACH(exp, &exports, next) { 459 if (nbd_negotiate_send_rep_list(client, exp, errp)) { 460 return -EINVAL; 461 } 462 } 463 /* Finish with a NBD_REP_ACK. */ 464 return nbd_negotiate_send_rep(client, NBD_REP_ACK, errp); 465 } 466 467 static coroutine_fn void 468 nbd_check_meta_export(NBDClient *client, NBDExport *exp) 469 { 470 if (exp != client->contexts.exp) { 471 client->contexts.count = 0; 472 } 473 } 474 475 /* Send a reply to NBD_OPT_EXPORT_NAME. 476 * Return -errno on error, 0 on success. */ 477 static coroutine_fn int 478 nbd_negotiate_handle_export_name(NBDClient *client, bool no_zeroes, 479 Error **errp) 480 { 481 ERRP_GUARD(); 482 g_autofree char *name = NULL; 483 char buf[NBD_REPLY_EXPORT_NAME_SIZE] = ""; 484 size_t len; 485 int ret; 486 uint16_t myflags; 487 488 /* Client sends: 489 [20 .. xx] export name (length bytes) 490 Server replies: 491 [ 0 .. 7] size 492 [ 8 .. 9] export flags 493 [10 .. 133] reserved (0) [unless no_zeroes] 494 */ 495 trace_nbd_negotiate_handle_export_name(); 496 if (client->mode >= NBD_MODE_EXTENDED) { 497 error_setg(errp, "Extended headers already negotiated"); 498 return -EINVAL; 499 } 500 if (client->optlen > NBD_MAX_STRING_SIZE) { 501 error_setg(errp, "Bad length received"); 502 return -EINVAL; 503 } 504 name = g_malloc(client->optlen + 1); 505 if (nbd_read(client->ioc, name, client->optlen, "export name", errp) < 0) { 506 return -EIO; 507 } 508 name[client->optlen] = '\0'; 509 client->optlen = 0; 510 511 trace_nbd_negotiate_handle_export_name_request(name); 512 513 client->exp = nbd_export_find(name); 514 if (!client->exp) { 515 error_setg(errp, "export not found"); 516 return -EINVAL; 517 } 518 nbd_check_meta_export(client, client->exp); 519 520 myflags = client->exp->nbdflags; 521 if (client->mode >= NBD_MODE_STRUCTURED) { 522 myflags |= NBD_FLAG_SEND_DF; 523 } 524 if (client->mode >= NBD_MODE_EXTENDED && client->contexts.count) { 525 myflags |= NBD_FLAG_BLOCK_STAT_PAYLOAD; 526 } 527 trace_nbd_negotiate_new_style_size_flags(client->exp->size, myflags); 528 stq_be_p(buf, client->exp->size); 529 stw_be_p(buf + 8, myflags); 530 len = no_zeroes ? 10 : sizeof(buf); 531 ret = nbd_write(client->ioc, buf, len, errp); 532 if (ret < 0) { 533 error_prepend(errp, "write failed: "); 534 return ret; 535 } 536 537 QTAILQ_INSERT_TAIL(&client->exp->clients, client, next); 538 blk_exp_ref(&client->exp->common); 539 540 return 0; 541 } 542 543 /* Send a single NBD_REP_INFO, with a buffer @buf of @length bytes. 544 * The buffer does NOT include the info type prefix. 545 * Return -errno on error, 0 if ready to send more. */ 546 static coroutine_fn int 547 nbd_negotiate_send_info(NBDClient *client, uint16_t info, uint32_t length, 548 void *buf, Error **errp) 549 { 550 int rc; 551 552 trace_nbd_negotiate_send_info(info, nbd_info_lookup(info), length); 553 rc = nbd_negotiate_send_rep_len(client, NBD_REP_INFO, 554 sizeof(info) + length, errp); 555 if (rc < 0) { 556 return rc; 557 } 558 info = cpu_to_be16(info); 559 if (nbd_write(client->ioc, &info, sizeof(info), errp) < 0) { 560 return -EIO; 561 } 562 if (nbd_write(client->ioc, buf, length, errp) < 0) { 563 return -EIO; 564 } 565 return 0; 566 } 567 568 /* nbd_reject_length: Handle any unexpected payload. 569 * @fatal requests that we quit talking to the client, even if we are able 570 * to successfully send an error reply. 571 * Return: 572 * -errno transmission error occurred or @fatal was requested, errp is set 573 * 0 error message successfully sent to client, errp is not set 574 */ 575 static coroutine_fn int 576 nbd_reject_length(NBDClient *client, bool fatal, Error **errp) 577 { 578 int ret; 579 580 assert(client->optlen); 581 ret = nbd_opt_invalid(client, errp, "option '%s' has unexpected length", 582 nbd_opt_lookup(client->opt)); 583 if (fatal && !ret) { 584 error_setg(errp, "option '%s' has unexpected length", 585 nbd_opt_lookup(client->opt)); 586 return -EINVAL; 587 } 588 return ret; 589 } 590 591 /* Handle NBD_OPT_INFO and NBD_OPT_GO. 592 * Return -errno on error, 0 if ready for next option, and 1 to move 593 * into transmission phase. */ 594 static coroutine_fn int 595 nbd_negotiate_handle_info(NBDClient *client, Error **errp) 596 { 597 int rc; 598 g_autofree char *name = NULL; 599 NBDExport *exp; 600 uint16_t requests; 601 uint16_t request; 602 uint32_t namelen = 0; 603 bool sendname = false; 604 bool blocksize = false; 605 uint32_t sizes[3]; 606 char buf[sizeof(uint64_t) + sizeof(uint16_t)]; 607 uint32_t check_align = 0; 608 uint16_t myflags; 609 610 /* Client sends: 611 4 bytes: L, name length (can be 0) 612 L bytes: export name 613 2 bytes: N, number of requests (can be 0) 614 N * 2 bytes: N requests 615 */ 616 rc = nbd_opt_read_name(client, &name, &namelen, errp); 617 if (rc <= 0) { 618 return rc; 619 } 620 trace_nbd_negotiate_handle_export_name_request(name); 621 622 rc = nbd_opt_read(client, &requests, sizeof(requests), false, errp); 623 if (rc <= 0) { 624 return rc; 625 } 626 requests = be16_to_cpu(requests); 627 trace_nbd_negotiate_handle_info_requests(requests); 628 while (requests--) { 629 rc = nbd_opt_read(client, &request, sizeof(request), false, errp); 630 if (rc <= 0) { 631 return rc; 632 } 633 request = be16_to_cpu(request); 634 trace_nbd_negotiate_handle_info_request(request, 635 nbd_info_lookup(request)); 636 /* We care about NBD_INFO_NAME and NBD_INFO_BLOCK_SIZE; 637 * everything else is either a request we don't know or 638 * something we send regardless of request */ 639 switch (request) { 640 case NBD_INFO_NAME: 641 sendname = true; 642 break; 643 case NBD_INFO_BLOCK_SIZE: 644 blocksize = true; 645 break; 646 } 647 } 648 if (client->optlen) { 649 return nbd_reject_length(client, false, errp); 650 } 651 652 exp = nbd_export_find(name); 653 if (!exp) { 654 g_autofree char *sane_name = nbd_sanitize_name(name); 655 656 return nbd_negotiate_send_rep_err(client, NBD_REP_ERR_UNKNOWN, 657 errp, "export '%s' not present", 658 sane_name); 659 } 660 if (client->opt == NBD_OPT_GO) { 661 nbd_check_meta_export(client, exp); 662 } 663 664 /* Don't bother sending NBD_INFO_NAME unless client requested it */ 665 if (sendname) { 666 rc = nbd_negotiate_send_info(client, NBD_INFO_NAME, namelen, name, 667 errp); 668 if (rc < 0) { 669 return rc; 670 } 671 } 672 673 /* Send NBD_INFO_DESCRIPTION only if available, regardless of 674 * client request */ 675 if (exp->description) { 676 size_t len = strlen(exp->description); 677 678 assert(len <= NBD_MAX_STRING_SIZE); 679 rc = nbd_negotiate_send_info(client, NBD_INFO_DESCRIPTION, 680 len, exp->description, errp); 681 if (rc < 0) { 682 return rc; 683 } 684 } 685 686 /* Send NBD_INFO_BLOCK_SIZE always, but tweak the minimum size 687 * according to whether the client requested it, and according to 688 * whether this is OPT_INFO or OPT_GO. */ 689 /* minimum - 1 for back-compat, or actual if client will obey it. */ 690 if (client->opt == NBD_OPT_INFO || blocksize) { 691 check_align = sizes[0] = blk_get_request_alignment(exp->common.blk); 692 } else { 693 sizes[0] = 1; 694 } 695 assert(sizes[0] <= NBD_MAX_BUFFER_SIZE); 696 /* preferred - Hard-code to 4096 for now. 697 * TODO: is blk_bs(blk)->bl.opt_transfer appropriate? */ 698 sizes[1] = MAX(4096, sizes[0]); 699 /* maximum - At most 32M, but smaller as appropriate. */ 700 sizes[2] = MIN(blk_get_max_transfer(exp->common.blk), NBD_MAX_BUFFER_SIZE); 701 trace_nbd_negotiate_handle_info_block_size(sizes[0], sizes[1], sizes[2]); 702 sizes[0] = cpu_to_be32(sizes[0]); 703 sizes[1] = cpu_to_be32(sizes[1]); 704 sizes[2] = cpu_to_be32(sizes[2]); 705 rc = nbd_negotiate_send_info(client, NBD_INFO_BLOCK_SIZE, 706 sizeof(sizes), sizes, errp); 707 if (rc < 0) { 708 return rc; 709 } 710 711 /* Send NBD_INFO_EXPORT always */ 712 myflags = exp->nbdflags; 713 if (client->mode >= NBD_MODE_STRUCTURED) { 714 myflags |= NBD_FLAG_SEND_DF; 715 } 716 if (client->mode >= NBD_MODE_EXTENDED && 717 (client->contexts.count || client->opt == NBD_OPT_INFO)) { 718 myflags |= NBD_FLAG_BLOCK_STAT_PAYLOAD; 719 } 720 trace_nbd_negotiate_new_style_size_flags(exp->size, myflags); 721 stq_be_p(buf, exp->size); 722 stw_be_p(buf + 8, myflags); 723 rc = nbd_negotiate_send_info(client, NBD_INFO_EXPORT, 724 sizeof(buf), buf, errp); 725 if (rc < 0) { 726 return rc; 727 } 728 729 /* 730 * If the client is just asking for NBD_OPT_INFO, but forgot to 731 * request block sizes in a situation that would impact 732 * performance, then return an error. But for NBD_OPT_GO, we 733 * tolerate all clients, regardless of alignments. 734 */ 735 if (client->opt == NBD_OPT_INFO && !blocksize && 736 blk_get_request_alignment(exp->common.blk) > 1) { 737 return nbd_negotiate_send_rep_err(client, 738 NBD_REP_ERR_BLOCK_SIZE_REQD, 739 errp, 740 "request NBD_INFO_BLOCK_SIZE to " 741 "use this export"); 742 } 743 744 /* Final reply */ 745 rc = nbd_negotiate_send_rep(client, NBD_REP_ACK, errp); 746 if (rc < 0) { 747 return rc; 748 } 749 750 if (client->opt == NBD_OPT_GO) { 751 client->exp = exp; 752 client->check_align = check_align; 753 QTAILQ_INSERT_TAIL(&client->exp->clients, client, next); 754 blk_exp_ref(&client->exp->common); 755 rc = 1; 756 } 757 return rc; 758 } 759 760 /* Callback to learn when QIO TLS upgrade is complete */ 761 struct NBDTLSServerHandshakeData { 762 bool complete; 763 Error *error; 764 Coroutine *co; 765 }; 766 767 static void 768 nbd_server_tls_handshake(QIOTask *task, void *opaque) 769 { 770 struct NBDTLSServerHandshakeData *data = opaque; 771 772 qio_task_propagate_error(task, &data->error); 773 data->complete = true; 774 if (!qemu_coroutine_entered(data->co)) { 775 aio_co_wake(data->co); 776 } 777 } 778 779 /* Handle NBD_OPT_STARTTLS. Return NULL to drop connection, or else the 780 * new channel for all further (now-encrypted) communication. */ 781 static coroutine_fn QIOChannel * 782 nbd_negotiate_handle_starttls(NBDClient *client, Error **errp) 783 { 784 QIOChannel *ioc; 785 QIOChannelTLS *tioc; 786 struct NBDTLSServerHandshakeData data = { 0 }; 787 788 assert(client->opt == NBD_OPT_STARTTLS); 789 790 trace_nbd_negotiate_handle_starttls(); 791 ioc = client->ioc; 792 793 if (nbd_negotiate_send_rep(client, NBD_REP_ACK, errp) < 0) { 794 return NULL; 795 } 796 797 tioc = qio_channel_tls_new_server(ioc, 798 client->tlscreds, 799 client->tlsauthz, 800 errp); 801 if (!tioc) { 802 return NULL; 803 } 804 805 qio_channel_set_name(QIO_CHANNEL(tioc), "nbd-server-tls"); 806 trace_nbd_negotiate_handle_starttls_handshake(); 807 data.co = qemu_coroutine_self(); 808 qio_channel_tls_handshake(tioc, 809 nbd_server_tls_handshake, 810 &data, 811 NULL, 812 NULL); 813 814 if (!data.complete) { 815 qemu_coroutine_yield(); 816 assert(data.complete); 817 } 818 819 if (data.error) { 820 object_unref(OBJECT(tioc)); 821 error_propagate(errp, data.error); 822 return NULL; 823 } 824 825 return QIO_CHANNEL(tioc); 826 } 827 828 /* nbd_negotiate_send_meta_context 829 * 830 * Send one chunk of reply to NBD_OPT_{LIST,SET}_META_CONTEXT 831 * 832 * For NBD_OPT_LIST_META_CONTEXT @context_id is ignored, 0 is used instead. 833 */ 834 static coroutine_fn int 835 nbd_negotiate_send_meta_context(NBDClient *client, const char *context, 836 uint32_t context_id, Error **errp) 837 { 838 NBDOptionReplyMetaContext opt; 839 struct iovec iov[] = { 840 {.iov_base = &opt, .iov_len = sizeof(opt)}, 841 {.iov_base = (void *)context, .iov_len = strlen(context)} 842 }; 843 844 assert(iov[1].iov_len <= NBD_MAX_STRING_SIZE); 845 if (client->opt == NBD_OPT_LIST_META_CONTEXT) { 846 context_id = 0; 847 } 848 849 trace_nbd_negotiate_meta_query_reply(context, context_id); 850 set_be_option_rep(&opt.h, client->opt, NBD_REP_META_CONTEXT, 851 sizeof(opt) - sizeof(opt.h) + iov[1].iov_len); 852 stl_be_p(&opt.context_id, context_id); 853 854 return qio_channel_writev_all(client->ioc, iov, 2, errp) < 0 ? -EIO : 0; 855 } 856 857 /* 858 * Return true if @query matches @pattern, or if @query is empty when 859 * the @client is performing _LIST_. 860 */ 861 static coroutine_fn bool 862 nbd_meta_empty_or_pattern(NBDClient *client, const char *pattern, 863 const char *query) 864 { 865 if (!*query) { 866 trace_nbd_negotiate_meta_query_parse("empty"); 867 return client->opt == NBD_OPT_LIST_META_CONTEXT; 868 } 869 if (strcmp(query, pattern) == 0) { 870 trace_nbd_negotiate_meta_query_parse(pattern); 871 return true; 872 } 873 trace_nbd_negotiate_meta_query_skip("pattern not matched"); 874 return false; 875 } 876 877 /* 878 * Return true and adjust @str in place if it begins with @prefix. 879 */ 880 static coroutine_fn bool 881 nbd_strshift(const char **str, const char *prefix) 882 { 883 size_t len = strlen(prefix); 884 885 if (strncmp(*str, prefix, len) == 0) { 886 *str += len; 887 return true; 888 } 889 return false; 890 } 891 892 /* nbd_meta_base_query 893 * 894 * Handle queries to 'base' namespace. For now, only the base:allocation 895 * context is available. Return true if @query has been handled. 896 */ 897 static coroutine_fn bool 898 nbd_meta_base_query(NBDClient *client, NBDMetaContexts *meta, 899 const char *query) 900 { 901 if (!nbd_strshift(&query, "base:")) { 902 return false; 903 } 904 trace_nbd_negotiate_meta_query_parse("base:"); 905 906 if (nbd_meta_empty_or_pattern(client, "allocation", query)) { 907 meta->base_allocation = true; 908 } 909 return true; 910 } 911 912 /* nbd_meta_qemu_query 913 * 914 * Handle queries to 'qemu' namespace. For now, only the qemu:dirty-bitmap: 915 * and qemu:allocation-depth contexts are available. Return true if @query 916 * has been handled. 917 */ 918 static coroutine_fn bool 919 nbd_meta_qemu_query(NBDClient *client, NBDMetaContexts *meta, 920 const char *query) 921 { 922 size_t i; 923 924 if (!nbd_strshift(&query, "qemu:")) { 925 return false; 926 } 927 trace_nbd_negotiate_meta_query_parse("qemu:"); 928 929 if (!*query) { 930 if (client->opt == NBD_OPT_LIST_META_CONTEXT) { 931 meta->allocation_depth = meta->exp->allocation_depth; 932 if (meta->exp->nr_export_bitmaps) { 933 memset(meta->bitmaps, 1, meta->exp->nr_export_bitmaps); 934 } 935 } 936 trace_nbd_negotiate_meta_query_parse("empty"); 937 return true; 938 } 939 940 if (strcmp(query, "allocation-depth") == 0) { 941 trace_nbd_negotiate_meta_query_parse("allocation-depth"); 942 meta->allocation_depth = meta->exp->allocation_depth; 943 return true; 944 } 945 946 if (nbd_strshift(&query, "dirty-bitmap:")) { 947 trace_nbd_negotiate_meta_query_parse("dirty-bitmap:"); 948 if (!*query) { 949 if (client->opt == NBD_OPT_LIST_META_CONTEXT && 950 meta->exp->nr_export_bitmaps) { 951 memset(meta->bitmaps, 1, meta->exp->nr_export_bitmaps); 952 } 953 trace_nbd_negotiate_meta_query_parse("empty"); 954 return true; 955 } 956 957 for (i = 0; i < meta->exp->nr_export_bitmaps; i++) { 958 const char *bm_name; 959 960 bm_name = bdrv_dirty_bitmap_name(meta->exp->export_bitmaps[i]); 961 if (strcmp(bm_name, query) == 0) { 962 meta->bitmaps[i] = true; 963 trace_nbd_negotiate_meta_query_parse(query); 964 return true; 965 } 966 } 967 trace_nbd_negotiate_meta_query_skip("no dirty-bitmap match"); 968 return true; 969 } 970 971 trace_nbd_negotiate_meta_query_skip("unknown qemu context"); 972 return true; 973 } 974 975 /* nbd_negotiate_meta_query 976 * 977 * Parse namespace name and call corresponding function to parse body of the 978 * query. 979 * 980 * The only supported namespaces are 'base' and 'qemu'. 981 * 982 * Return -errno on I/O error, 0 if option was completely handled by 983 * sending a reply about inconsistent lengths, or 1 on success. */ 984 static coroutine_fn int 985 nbd_negotiate_meta_query(NBDClient *client, 986 NBDMetaContexts *meta, Error **errp) 987 { 988 int ret; 989 g_autofree char *query = NULL; 990 uint32_t len; 991 992 ret = nbd_opt_read(client, &len, sizeof(len), false, errp); 993 if (ret <= 0) { 994 return ret; 995 } 996 len = cpu_to_be32(len); 997 998 if (len > NBD_MAX_STRING_SIZE) { 999 trace_nbd_negotiate_meta_query_skip("length too long"); 1000 return nbd_opt_skip(client, len, errp); 1001 } 1002 1003 query = g_malloc(len + 1); 1004 ret = nbd_opt_read(client, query, len, true, errp); 1005 if (ret <= 0) { 1006 return ret; 1007 } 1008 query[len] = '\0'; 1009 1010 if (nbd_meta_base_query(client, meta, query)) { 1011 return 1; 1012 } 1013 if (nbd_meta_qemu_query(client, meta, query)) { 1014 return 1; 1015 } 1016 1017 trace_nbd_negotiate_meta_query_skip("unknown namespace"); 1018 return 1; 1019 } 1020 1021 /* nbd_negotiate_meta_queries 1022 * Handle NBD_OPT_LIST_META_CONTEXT and NBD_OPT_SET_META_CONTEXT 1023 * 1024 * Return -errno on I/O error, or 0 if option was completely handled. */ 1025 static coroutine_fn int 1026 nbd_negotiate_meta_queries(NBDClient *client, Error **errp) 1027 { 1028 int ret; 1029 g_autofree char *export_name = NULL; 1030 /* Mark unused to work around https://bugs.llvm.org/show_bug.cgi?id=3888 */ 1031 g_autofree G_GNUC_UNUSED bool *bitmaps = NULL; 1032 NBDMetaContexts local_meta = {0}; 1033 NBDMetaContexts *meta; 1034 uint32_t nb_queries; 1035 size_t i; 1036 size_t count = 0; 1037 1038 if (client->opt == NBD_OPT_SET_META_CONTEXT && 1039 client->mode < NBD_MODE_STRUCTURED) { 1040 return nbd_opt_invalid(client, errp, 1041 "request option '%s' when structured reply " 1042 "is not negotiated", 1043 nbd_opt_lookup(client->opt)); 1044 } 1045 1046 if (client->opt == NBD_OPT_LIST_META_CONTEXT) { 1047 /* Only change the caller's meta on SET. */ 1048 meta = &local_meta; 1049 } else { 1050 meta = &client->contexts; 1051 } 1052 1053 g_free(meta->bitmaps); 1054 memset(meta, 0, sizeof(*meta)); 1055 1056 ret = nbd_opt_read_name(client, &export_name, NULL, errp); 1057 if (ret <= 0) { 1058 return ret; 1059 } 1060 1061 meta->exp = nbd_export_find(export_name); 1062 if (meta->exp == NULL) { 1063 g_autofree char *sane_name = nbd_sanitize_name(export_name); 1064 1065 return nbd_opt_drop(client, NBD_REP_ERR_UNKNOWN, errp, 1066 "export '%s' not present", sane_name); 1067 } 1068 meta->bitmaps = g_new0(bool, meta->exp->nr_export_bitmaps); 1069 if (client->opt == NBD_OPT_LIST_META_CONTEXT) { 1070 bitmaps = meta->bitmaps; 1071 } 1072 1073 ret = nbd_opt_read(client, &nb_queries, sizeof(nb_queries), false, errp); 1074 if (ret <= 0) { 1075 return ret; 1076 } 1077 nb_queries = cpu_to_be32(nb_queries); 1078 trace_nbd_negotiate_meta_context(nbd_opt_lookup(client->opt), 1079 export_name, nb_queries); 1080 1081 if (client->opt == NBD_OPT_LIST_META_CONTEXT && !nb_queries) { 1082 /* enable all known contexts */ 1083 meta->base_allocation = true; 1084 meta->allocation_depth = meta->exp->allocation_depth; 1085 if (meta->exp->nr_export_bitmaps) { 1086 memset(meta->bitmaps, 1, meta->exp->nr_export_bitmaps); 1087 } 1088 } else { 1089 for (i = 0; i < nb_queries; ++i) { 1090 ret = nbd_negotiate_meta_query(client, meta, errp); 1091 if (ret <= 0) { 1092 return ret; 1093 } 1094 } 1095 } 1096 1097 if (meta->base_allocation) { 1098 ret = nbd_negotiate_send_meta_context(client, "base:allocation", 1099 NBD_META_ID_BASE_ALLOCATION, 1100 errp); 1101 if (ret < 0) { 1102 return ret; 1103 } 1104 count++; 1105 } 1106 1107 if (meta->allocation_depth) { 1108 ret = nbd_negotiate_send_meta_context(client, "qemu:allocation-depth", 1109 NBD_META_ID_ALLOCATION_DEPTH, 1110 errp); 1111 if (ret < 0) { 1112 return ret; 1113 } 1114 count++; 1115 } 1116 1117 for (i = 0; i < meta->exp->nr_export_bitmaps; i++) { 1118 const char *bm_name; 1119 g_autofree char *context = NULL; 1120 1121 if (!meta->bitmaps[i]) { 1122 continue; 1123 } 1124 1125 bm_name = bdrv_dirty_bitmap_name(meta->exp->export_bitmaps[i]); 1126 context = g_strdup_printf("qemu:dirty-bitmap:%s", bm_name); 1127 1128 ret = nbd_negotiate_send_meta_context(client, context, 1129 NBD_META_ID_DIRTY_BITMAP + i, 1130 errp); 1131 if (ret < 0) { 1132 return ret; 1133 } 1134 count++; 1135 } 1136 1137 ret = nbd_negotiate_send_rep(client, NBD_REP_ACK, errp); 1138 if (ret == 0) { 1139 meta->count = count; 1140 } 1141 1142 return ret; 1143 } 1144 1145 /* nbd_negotiate_options 1146 * Process all NBD_OPT_* client option commands, during fixed newstyle 1147 * negotiation. 1148 * Return: 1149 * -errno on error, errp is set 1150 * 0 on successful negotiation, errp is not set 1151 * 1 if client sent NBD_OPT_ABORT, i.e. on valid disconnect, 1152 * errp is not set 1153 */ 1154 static coroutine_fn int 1155 nbd_negotiate_options(NBDClient *client, Error **errp) 1156 { 1157 uint32_t flags; 1158 bool fixedNewstyle = false; 1159 bool no_zeroes = false; 1160 1161 /* Client sends: 1162 [ 0 .. 3] client flags 1163 1164 Then we loop until NBD_OPT_EXPORT_NAME or NBD_OPT_GO: 1165 [ 0 .. 7] NBD_OPTS_MAGIC 1166 [ 8 .. 11] NBD option 1167 [12 .. 15] Data length 1168 ... Rest of request 1169 1170 [ 0 .. 7] NBD_OPTS_MAGIC 1171 [ 8 .. 11] Second NBD option 1172 [12 .. 15] Data length 1173 ... Rest of request 1174 */ 1175 1176 if (nbd_read32(client->ioc, &flags, "flags", errp) < 0) { 1177 return -EIO; 1178 } 1179 client->mode = NBD_MODE_EXPORT_NAME; 1180 trace_nbd_negotiate_options_flags(flags); 1181 if (flags & NBD_FLAG_C_FIXED_NEWSTYLE) { 1182 fixedNewstyle = true; 1183 flags &= ~NBD_FLAG_C_FIXED_NEWSTYLE; 1184 client->mode = NBD_MODE_SIMPLE; 1185 } 1186 if (flags & NBD_FLAG_C_NO_ZEROES) { 1187 no_zeroes = true; 1188 flags &= ~NBD_FLAG_C_NO_ZEROES; 1189 } 1190 if (flags != 0) { 1191 error_setg(errp, "Unknown client flags 0x%" PRIx32 " received", flags); 1192 return -EINVAL; 1193 } 1194 1195 while (1) { 1196 int ret; 1197 uint32_t option, length; 1198 uint64_t magic; 1199 1200 if (nbd_read64(client->ioc, &magic, "opts magic", errp) < 0) { 1201 return -EINVAL; 1202 } 1203 trace_nbd_negotiate_options_check_magic(magic); 1204 if (magic != NBD_OPTS_MAGIC) { 1205 error_setg(errp, "Bad magic received"); 1206 return -EINVAL; 1207 } 1208 1209 if (nbd_read32(client->ioc, &option, "option", errp) < 0) { 1210 return -EINVAL; 1211 } 1212 client->opt = option; 1213 1214 if (nbd_read32(client->ioc, &length, "option length", errp) < 0) { 1215 return -EINVAL; 1216 } 1217 assert(!client->optlen); 1218 client->optlen = length; 1219 1220 if (length > NBD_MAX_BUFFER_SIZE) { 1221 error_setg(errp, "len (%" PRIu32 ") is larger than max len (%u)", 1222 length, NBD_MAX_BUFFER_SIZE); 1223 return -EINVAL; 1224 } 1225 1226 trace_nbd_negotiate_options_check_option(option, 1227 nbd_opt_lookup(option)); 1228 if (client->tlscreds && 1229 client->ioc == (QIOChannel *)client->sioc) { 1230 QIOChannel *tioc; 1231 if (!fixedNewstyle) { 1232 error_setg(errp, "Unsupported option 0x%" PRIx32, option); 1233 return -EINVAL; 1234 } 1235 switch (option) { 1236 case NBD_OPT_STARTTLS: 1237 if (length) { 1238 /* Unconditionally drop the connection if the client 1239 * can't start a TLS negotiation correctly */ 1240 return nbd_reject_length(client, true, errp); 1241 } 1242 tioc = nbd_negotiate_handle_starttls(client, errp); 1243 if (!tioc) { 1244 return -EIO; 1245 } 1246 ret = 0; 1247 object_unref(OBJECT(client->ioc)); 1248 client->ioc = tioc; 1249 break; 1250 1251 case NBD_OPT_EXPORT_NAME: 1252 /* No way to return an error to client, so drop connection */ 1253 error_setg(errp, "Option 0x%x not permitted before TLS", 1254 option); 1255 return -EINVAL; 1256 1257 default: 1258 /* Let the client keep trying, unless they asked to 1259 * quit. Always try to give an error back to the 1260 * client; but when replying to OPT_ABORT, be aware 1261 * that the client may hang up before receiving the 1262 * error, in which case we are fine ignoring the 1263 * resulting EPIPE. */ 1264 ret = nbd_opt_drop(client, NBD_REP_ERR_TLS_REQD, 1265 option == NBD_OPT_ABORT ? NULL : errp, 1266 "Option 0x%" PRIx32 1267 " not permitted before TLS", option); 1268 if (option == NBD_OPT_ABORT) { 1269 return 1; 1270 } 1271 break; 1272 } 1273 } else if (fixedNewstyle) { 1274 switch (option) { 1275 case NBD_OPT_LIST: 1276 if (length) { 1277 ret = nbd_reject_length(client, false, errp); 1278 } else { 1279 ret = nbd_negotiate_handle_list(client, errp); 1280 } 1281 break; 1282 1283 case NBD_OPT_ABORT: 1284 /* NBD spec says we must try to reply before 1285 * disconnecting, but that we must also tolerate 1286 * guests that don't wait for our reply. */ 1287 nbd_negotiate_send_rep(client, NBD_REP_ACK, NULL); 1288 return 1; 1289 1290 case NBD_OPT_EXPORT_NAME: 1291 return nbd_negotiate_handle_export_name(client, no_zeroes, 1292 errp); 1293 1294 case NBD_OPT_INFO: 1295 case NBD_OPT_GO: 1296 ret = nbd_negotiate_handle_info(client, errp); 1297 if (ret == 1) { 1298 assert(option == NBD_OPT_GO); 1299 return 0; 1300 } 1301 break; 1302 1303 case NBD_OPT_STARTTLS: 1304 if (length) { 1305 ret = nbd_reject_length(client, false, errp); 1306 } else if (client->tlscreds) { 1307 ret = nbd_negotiate_send_rep_err(client, 1308 NBD_REP_ERR_INVALID, errp, 1309 "TLS already enabled"); 1310 } else { 1311 ret = nbd_negotiate_send_rep_err(client, 1312 NBD_REP_ERR_POLICY, errp, 1313 "TLS not configured"); 1314 } 1315 break; 1316 1317 case NBD_OPT_STRUCTURED_REPLY: 1318 if (length) { 1319 ret = nbd_reject_length(client, false, errp); 1320 } else if (client->mode >= NBD_MODE_EXTENDED) { 1321 ret = nbd_negotiate_send_rep_err( 1322 client, NBD_REP_ERR_EXT_HEADER_REQD, errp, 1323 "extended headers already negotiated"); 1324 } else if (client->mode >= NBD_MODE_STRUCTURED) { 1325 ret = nbd_negotiate_send_rep_err( 1326 client, NBD_REP_ERR_INVALID, errp, 1327 "structured reply already negotiated"); 1328 } else { 1329 ret = nbd_negotiate_send_rep(client, NBD_REP_ACK, errp); 1330 client->mode = NBD_MODE_STRUCTURED; 1331 } 1332 break; 1333 1334 case NBD_OPT_LIST_META_CONTEXT: 1335 case NBD_OPT_SET_META_CONTEXT: 1336 ret = nbd_negotiate_meta_queries(client, errp); 1337 break; 1338 1339 case NBD_OPT_EXTENDED_HEADERS: 1340 if (length) { 1341 ret = nbd_reject_length(client, false, errp); 1342 } else if (client->mode >= NBD_MODE_EXTENDED) { 1343 ret = nbd_negotiate_send_rep_err( 1344 client, NBD_REP_ERR_INVALID, errp, 1345 "extended headers already negotiated"); 1346 } else { 1347 ret = nbd_negotiate_send_rep(client, NBD_REP_ACK, errp); 1348 client->mode = NBD_MODE_EXTENDED; 1349 } 1350 break; 1351 1352 default: 1353 ret = nbd_opt_drop(client, NBD_REP_ERR_UNSUP, errp, 1354 "Unsupported option %" PRIu32 " (%s)", 1355 option, nbd_opt_lookup(option)); 1356 break; 1357 } 1358 } else { 1359 /* 1360 * If broken new-style we should drop the connection 1361 * for anything except NBD_OPT_EXPORT_NAME 1362 */ 1363 switch (option) { 1364 case NBD_OPT_EXPORT_NAME: 1365 return nbd_negotiate_handle_export_name(client, no_zeroes, 1366 errp); 1367 1368 default: 1369 error_setg(errp, "Unsupported option %" PRIu32 " (%s)", 1370 option, nbd_opt_lookup(option)); 1371 return -EINVAL; 1372 } 1373 } 1374 if (ret < 0) { 1375 return ret; 1376 } 1377 } 1378 } 1379 1380 /* nbd_negotiate 1381 * Return: 1382 * -errno on error, errp is set 1383 * 0 on successful negotiation, errp is not set 1384 * 1 if client sent NBD_OPT_ABORT, i.e. on valid disconnect, 1385 * errp is not set 1386 */ 1387 static coroutine_fn int nbd_negotiate(NBDClient *client, Error **errp) 1388 { 1389 ERRP_GUARD(); 1390 char buf[NBD_OLDSTYLE_NEGOTIATE_SIZE] = ""; 1391 int ret; 1392 1393 /* Old style negotiation header, no room for options 1394 [ 0 .. 7] passwd ("NBDMAGIC") 1395 [ 8 .. 15] magic (NBD_CLIENT_MAGIC) 1396 [16 .. 23] size 1397 [24 .. 27] export flags (zero-extended) 1398 [28 .. 151] reserved (0) 1399 1400 New style negotiation header, client can send options 1401 [ 0 .. 7] passwd ("NBDMAGIC") 1402 [ 8 .. 15] magic (NBD_OPTS_MAGIC) 1403 [16 .. 17] server flags (0) 1404 ....options sent, ending in NBD_OPT_EXPORT_NAME or NBD_OPT_GO.... 1405 */ 1406 1407 qio_channel_set_blocking(client->ioc, false, NULL); 1408 qio_channel_set_follow_coroutine_ctx(client->ioc, true); 1409 1410 trace_nbd_negotiate_begin(); 1411 memcpy(buf, "NBDMAGIC", 8); 1412 1413 stq_be_p(buf + 8, NBD_OPTS_MAGIC); 1414 stw_be_p(buf + 16, NBD_FLAG_FIXED_NEWSTYLE | NBD_FLAG_NO_ZEROES); 1415 1416 if (nbd_write(client->ioc, buf, 18, errp) < 0) { 1417 error_prepend(errp, "write failed: "); 1418 return -EINVAL; 1419 } 1420 ret = nbd_negotiate_options(client, errp); 1421 if (ret != 0) { 1422 if (ret < 0) { 1423 error_prepend(errp, "option negotiation failed: "); 1424 } 1425 return ret; 1426 } 1427 1428 assert(!client->optlen); 1429 trace_nbd_negotiate_success(); 1430 1431 return 0; 1432 } 1433 1434 /* nbd_read_eof 1435 * Tries to read @size bytes from @ioc. This is a local implementation of 1436 * qio_channel_readv_all_eof. We have it here because we need it to be 1437 * interruptible and to know when the coroutine is yielding. 1438 * Returns 1 on success 1439 * 0 on eof, when no data was read (errp is not set) 1440 * negative errno on failure (errp is set) 1441 */ 1442 static inline int coroutine_fn 1443 nbd_read_eof(NBDClient *client, void *buffer, size_t size, Error **errp) 1444 { 1445 bool partial = false; 1446 1447 assert(size); 1448 while (size > 0) { 1449 struct iovec iov = { .iov_base = buffer, .iov_len = size }; 1450 ssize_t len; 1451 1452 len = qio_channel_readv(client->ioc, &iov, 1, errp); 1453 if (len == QIO_CHANNEL_ERR_BLOCK) { 1454 WITH_QEMU_LOCK_GUARD(&client->lock) { 1455 client->read_yielding = true; 1456 1457 /* Prompt main loop thread to re-run nbd_drained_poll() */ 1458 aio_wait_kick(); 1459 } 1460 qio_channel_yield(client->ioc, G_IO_IN); 1461 WITH_QEMU_LOCK_GUARD(&client->lock) { 1462 client->read_yielding = false; 1463 if (client->quiescing) { 1464 return -EAGAIN; 1465 } 1466 } 1467 continue; 1468 } else if (len < 0) { 1469 return -EIO; 1470 } else if (len == 0) { 1471 if (partial) { 1472 error_setg(errp, 1473 "Unexpected end-of-file before all bytes were read"); 1474 return -EIO; 1475 } else { 1476 return 0; 1477 } 1478 } 1479 1480 partial = true; 1481 size -= len; 1482 buffer = (uint8_t *) buffer + len; 1483 } 1484 return 1; 1485 } 1486 1487 static int coroutine_fn nbd_receive_request(NBDClient *client, NBDRequest *request, 1488 Error **errp) 1489 { 1490 uint8_t buf[NBD_EXTENDED_REQUEST_SIZE]; 1491 uint32_t magic, expect; 1492 int ret; 1493 size_t size = client->mode >= NBD_MODE_EXTENDED ? 1494 NBD_EXTENDED_REQUEST_SIZE : NBD_REQUEST_SIZE; 1495 1496 ret = nbd_read_eof(client, buf, size, errp); 1497 if (ret < 0) { 1498 return ret; 1499 } 1500 if (ret == 0) { 1501 return -EIO; 1502 } 1503 1504 /* 1505 * Compact request 1506 * [ 0 .. 3] magic (NBD_REQUEST_MAGIC) 1507 * [ 4 .. 5] flags (NBD_CMD_FLAG_FUA, ...) 1508 * [ 6 .. 7] type (NBD_CMD_READ, ...) 1509 * [ 8 .. 15] cookie 1510 * [16 .. 23] from 1511 * [24 .. 27] len 1512 * Extended request 1513 * [ 0 .. 3] magic (NBD_EXTENDED_REQUEST_MAGIC) 1514 * [ 4 .. 5] flags (NBD_CMD_FLAG_FUA, NBD_CMD_FLAG_PAYLOAD_LEN, ...) 1515 * [ 6 .. 7] type (NBD_CMD_READ, ...) 1516 * [ 8 .. 15] cookie 1517 * [16 .. 23] from 1518 * [24 .. 31] len 1519 */ 1520 1521 magic = ldl_be_p(buf); 1522 request->flags = lduw_be_p(buf + 4); 1523 request->type = lduw_be_p(buf + 6); 1524 request->cookie = ldq_be_p(buf + 8); 1525 request->from = ldq_be_p(buf + 16); 1526 if (client->mode >= NBD_MODE_EXTENDED) { 1527 request->len = ldq_be_p(buf + 24); 1528 expect = NBD_EXTENDED_REQUEST_MAGIC; 1529 } else { 1530 request->len = (uint32_t)ldl_be_p(buf + 24); /* widen 32 to 64 bits */ 1531 expect = NBD_REQUEST_MAGIC; 1532 } 1533 1534 trace_nbd_receive_request(magic, request->flags, request->type, 1535 request->from, request->len); 1536 1537 if (magic != expect) { 1538 error_setg(errp, "invalid magic (got 0x%" PRIx32 ", expected 0x%" 1539 PRIx32 ")", magic, expect); 1540 return -EINVAL; 1541 } 1542 return 0; 1543 } 1544 1545 #define MAX_NBD_REQUESTS 16 1546 1547 /* Runs in export AioContext and main loop thread */ 1548 void nbd_client_get(NBDClient *client) 1549 { 1550 qatomic_inc(&client->refcount); 1551 } 1552 1553 void nbd_client_put(NBDClient *client) 1554 { 1555 assert(qemu_in_main_thread()); 1556 1557 if (qatomic_fetch_dec(&client->refcount) == 1) { 1558 /* The last reference should be dropped by client->close, 1559 * which is called by client_close. 1560 */ 1561 assert(client->closing); 1562 1563 object_unref(OBJECT(client->sioc)); 1564 object_unref(OBJECT(client->ioc)); 1565 if (client->tlscreds) { 1566 object_unref(OBJECT(client->tlscreds)); 1567 } 1568 g_free(client->tlsauthz); 1569 if (client->exp) { 1570 QTAILQ_REMOVE(&client->exp->clients, client, next); 1571 blk_exp_unref(&client->exp->common); 1572 } 1573 g_free(client->contexts.bitmaps); 1574 qemu_mutex_destroy(&client->lock); 1575 g_free(client); 1576 } 1577 } 1578 1579 /* 1580 * Tries to release the reference to @client, but only if other references 1581 * remain. This is an optimization for the common case where we want to avoid 1582 * the expense of scheduling nbd_client_put() in the main loop thread. 1583 * 1584 * Returns true upon success or false if the reference was not released because 1585 * it is the last reference. 1586 */ 1587 static bool nbd_client_put_nonzero(NBDClient *client) 1588 { 1589 int old = qatomic_read(&client->refcount); 1590 int expected; 1591 1592 do { 1593 if (old == 1) { 1594 return false; 1595 } 1596 1597 expected = old; 1598 old = qatomic_cmpxchg(&client->refcount, expected, expected - 1); 1599 } while (old != expected); 1600 1601 return true; 1602 } 1603 1604 static void client_close(NBDClient *client, bool negotiated) 1605 { 1606 assert(qemu_in_main_thread()); 1607 1608 WITH_QEMU_LOCK_GUARD(&client->lock) { 1609 if (client->closing) { 1610 return; 1611 } 1612 1613 client->closing = true; 1614 } 1615 1616 /* Force requests to finish. They will drop their own references, 1617 * then we'll close the socket and free the NBDClient. 1618 */ 1619 qio_channel_shutdown(client->ioc, QIO_CHANNEL_SHUTDOWN_BOTH, 1620 NULL); 1621 1622 /* Also tell the client, so that they release their reference. */ 1623 if (client->close_fn) { 1624 client->close_fn(client, negotiated); 1625 } 1626 } 1627 1628 /* Runs in export AioContext with client->lock held */ 1629 static NBDRequestData *nbd_request_get(NBDClient *client) 1630 { 1631 NBDRequestData *req; 1632 1633 assert(client->nb_requests <= MAX_NBD_REQUESTS - 1); 1634 client->nb_requests++; 1635 1636 req = g_new0(NBDRequestData, 1); 1637 req->client = client; 1638 return req; 1639 } 1640 1641 /* Runs in export AioContext with client->lock held */ 1642 static void nbd_request_put(NBDRequestData *req) 1643 { 1644 NBDClient *client = req->client; 1645 1646 if (req->data) { 1647 qemu_vfree(req->data); 1648 } 1649 g_free(req); 1650 1651 client->nb_requests--; 1652 1653 if (client->quiescing && client->nb_requests == 0) { 1654 aio_wait_kick(); 1655 } 1656 1657 nbd_client_receive_next_request(client); 1658 } 1659 1660 static void blk_aio_attached(AioContext *ctx, void *opaque) 1661 { 1662 NBDExport *exp = opaque; 1663 NBDClient *client; 1664 1665 assert(qemu_in_main_thread()); 1666 1667 trace_nbd_blk_aio_attached(exp->name, ctx); 1668 1669 exp->common.ctx = ctx; 1670 1671 QTAILQ_FOREACH(client, &exp->clients, next) { 1672 WITH_QEMU_LOCK_GUARD(&client->lock) { 1673 assert(client->nb_requests == 0); 1674 assert(client->recv_coroutine == NULL); 1675 assert(client->send_coroutine == NULL); 1676 } 1677 } 1678 } 1679 1680 static void blk_aio_detach(void *opaque) 1681 { 1682 NBDExport *exp = opaque; 1683 1684 assert(qemu_in_main_thread()); 1685 1686 trace_nbd_blk_aio_detach(exp->name, exp->common.ctx); 1687 1688 exp->common.ctx = NULL; 1689 } 1690 1691 static void nbd_drained_begin(void *opaque) 1692 { 1693 NBDExport *exp = opaque; 1694 NBDClient *client; 1695 1696 assert(qemu_in_main_thread()); 1697 1698 QTAILQ_FOREACH(client, &exp->clients, next) { 1699 WITH_QEMU_LOCK_GUARD(&client->lock) { 1700 client->quiescing = true; 1701 } 1702 } 1703 } 1704 1705 static void nbd_drained_end(void *opaque) 1706 { 1707 NBDExport *exp = opaque; 1708 NBDClient *client; 1709 1710 assert(qemu_in_main_thread()); 1711 1712 QTAILQ_FOREACH(client, &exp->clients, next) { 1713 WITH_QEMU_LOCK_GUARD(&client->lock) { 1714 client->quiescing = false; 1715 nbd_client_receive_next_request(client); 1716 } 1717 } 1718 } 1719 1720 /* Runs in export AioContext */ 1721 static void nbd_wake_read_bh(void *opaque) 1722 { 1723 NBDClient *client = opaque; 1724 qio_channel_wake_read(client->ioc); 1725 } 1726 1727 static bool nbd_drained_poll(void *opaque) 1728 { 1729 NBDExport *exp = opaque; 1730 NBDClient *client; 1731 1732 assert(qemu_in_main_thread()); 1733 1734 QTAILQ_FOREACH(client, &exp->clients, next) { 1735 WITH_QEMU_LOCK_GUARD(&client->lock) { 1736 if (client->nb_requests != 0) { 1737 /* 1738 * If there's a coroutine waiting for a request on nbd_read_eof() 1739 * enter it here so we don't depend on the client to wake it up. 1740 * 1741 * Schedule a BH in the export AioContext to avoid missing the 1742 * wake up due to the race between qio_channel_wake_read() and 1743 * qio_channel_yield(). 1744 */ 1745 if (client->recv_coroutine != NULL && client->read_yielding) { 1746 aio_bh_schedule_oneshot(nbd_export_aio_context(client->exp), 1747 nbd_wake_read_bh, client); 1748 } 1749 1750 return true; 1751 } 1752 } 1753 } 1754 1755 return false; 1756 } 1757 1758 static void nbd_eject_notifier(Notifier *n, void *data) 1759 { 1760 NBDExport *exp = container_of(n, NBDExport, eject_notifier); 1761 1762 assert(qemu_in_main_thread()); 1763 1764 blk_exp_request_shutdown(&exp->common); 1765 } 1766 1767 void nbd_export_set_on_eject_blk(BlockExport *exp, BlockBackend *blk) 1768 { 1769 NBDExport *nbd_exp = container_of(exp, NBDExport, common); 1770 assert(exp->drv == &blk_exp_nbd); 1771 assert(nbd_exp->eject_notifier_blk == NULL); 1772 1773 blk_ref(blk); 1774 nbd_exp->eject_notifier_blk = blk; 1775 nbd_exp->eject_notifier.notify = nbd_eject_notifier; 1776 blk_add_remove_bs_notifier(blk, &nbd_exp->eject_notifier); 1777 } 1778 1779 static const BlockDevOps nbd_block_ops = { 1780 .drained_begin = nbd_drained_begin, 1781 .drained_end = nbd_drained_end, 1782 .drained_poll = nbd_drained_poll, 1783 }; 1784 1785 static int nbd_export_create(BlockExport *blk_exp, BlockExportOptions *exp_args, 1786 Error **errp) 1787 { 1788 NBDExport *exp = container_of(blk_exp, NBDExport, common); 1789 BlockExportOptionsNbd *arg = &exp_args->u.nbd; 1790 const char *name = arg->name ?: exp_args->node_name; 1791 BlockBackend *blk = blk_exp->blk; 1792 int64_t size; 1793 uint64_t perm, shared_perm; 1794 bool readonly = !exp_args->writable; 1795 BlockDirtyBitmapOrStrList *bitmaps; 1796 size_t i; 1797 int ret; 1798 1799 GLOBAL_STATE_CODE(); 1800 assert(exp_args->type == BLOCK_EXPORT_TYPE_NBD); 1801 1802 if (!nbd_server_is_running()) { 1803 error_setg(errp, "NBD server not running"); 1804 return -EINVAL; 1805 } 1806 1807 if (strlen(name) > NBD_MAX_STRING_SIZE) { 1808 error_setg(errp, "export name '%s' too long", name); 1809 return -EINVAL; 1810 } 1811 1812 if (arg->description && strlen(arg->description) > NBD_MAX_STRING_SIZE) { 1813 error_setg(errp, "description '%s' too long", arg->description); 1814 return -EINVAL; 1815 } 1816 1817 if (nbd_export_find(name)) { 1818 error_setg(errp, "NBD server already has export named '%s'", name); 1819 return -EEXIST; 1820 } 1821 1822 size = blk_getlength(blk); 1823 if (size < 0) { 1824 error_setg_errno(errp, -size, 1825 "Failed to determine the NBD export's length"); 1826 return size; 1827 } 1828 1829 /* Don't allow resize while the NBD server is running, otherwise we don't 1830 * care what happens with the node. */ 1831 blk_get_perm(blk, &perm, &shared_perm); 1832 ret = blk_set_perm(blk, perm, shared_perm & ~BLK_PERM_RESIZE, errp); 1833 if (ret < 0) { 1834 return ret; 1835 } 1836 1837 QTAILQ_INIT(&exp->clients); 1838 exp->name = g_strdup(name); 1839 exp->description = g_strdup(arg->description); 1840 exp->nbdflags = (NBD_FLAG_HAS_FLAGS | NBD_FLAG_SEND_FLUSH | 1841 NBD_FLAG_SEND_FUA | NBD_FLAG_SEND_CACHE); 1842 1843 if (nbd_server_max_connections() != 1) { 1844 exp->nbdflags |= NBD_FLAG_CAN_MULTI_CONN; 1845 } 1846 if (readonly) { 1847 exp->nbdflags |= NBD_FLAG_READ_ONLY; 1848 } else { 1849 exp->nbdflags |= (NBD_FLAG_SEND_TRIM | NBD_FLAG_SEND_WRITE_ZEROES | 1850 NBD_FLAG_SEND_FAST_ZERO); 1851 } 1852 exp->size = QEMU_ALIGN_DOWN(size, BDRV_SECTOR_SIZE); 1853 1854 bdrv_graph_rdlock_main_loop(); 1855 1856 for (bitmaps = arg->bitmaps; bitmaps; bitmaps = bitmaps->next) { 1857 exp->nr_export_bitmaps++; 1858 } 1859 exp->export_bitmaps = g_new0(BdrvDirtyBitmap *, exp->nr_export_bitmaps); 1860 for (i = 0, bitmaps = arg->bitmaps; bitmaps; 1861 i++, bitmaps = bitmaps->next) 1862 { 1863 const char *bitmap; 1864 BlockDriverState *bs = blk_bs(blk); 1865 BdrvDirtyBitmap *bm = NULL; 1866 1867 switch (bitmaps->value->type) { 1868 case QTYPE_QSTRING: 1869 bitmap = bitmaps->value->u.local; 1870 while (bs) { 1871 bm = bdrv_find_dirty_bitmap(bs, bitmap); 1872 if (bm != NULL) { 1873 break; 1874 } 1875 1876 bs = bdrv_filter_or_cow_bs(bs); 1877 } 1878 1879 if (bm == NULL) { 1880 ret = -ENOENT; 1881 error_setg(errp, "Bitmap '%s' is not found", 1882 bitmaps->value->u.local); 1883 goto fail; 1884 } 1885 1886 if (readonly && bdrv_is_writable(bs) && 1887 bdrv_dirty_bitmap_enabled(bm)) { 1888 ret = -EINVAL; 1889 error_setg(errp, "Enabled bitmap '%s' incompatible with " 1890 "readonly export", bitmap); 1891 goto fail; 1892 } 1893 break; 1894 case QTYPE_QDICT: 1895 bitmap = bitmaps->value->u.external.name; 1896 bm = block_dirty_bitmap_lookup(bitmaps->value->u.external.node, 1897 bitmap, NULL, errp); 1898 if (!bm) { 1899 ret = -ENOENT; 1900 goto fail; 1901 } 1902 break; 1903 default: 1904 abort(); 1905 } 1906 1907 assert(bm); 1908 1909 if (bdrv_dirty_bitmap_check(bm, BDRV_BITMAP_ALLOW_RO, errp)) { 1910 ret = -EINVAL; 1911 goto fail; 1912 } 1913 1914 exp->export_bitmaps[i] = bm; 1915 assert(strlen(bitmap) <= BDRV_BITMAP_MAX_NAME_SIZE); 1916 } 1917 1918 /* Mark bitmaps busy in a separate loop, to simplify roll-back concerns. */ 1919 for (i = 0; i < exp->nr_export_bitmaps; i++) { 1920 bdrv_dirty_bitmap_set_busy(exp->export_bitmaps[i], true); 1921 } 1922 1923 exp->allocation_depth = arg->allocation_depth; 1924 1925 /* 1926 * We need to inhibit request queuing in the block layer to ensure we can 1927 * be properly quiesced when entering a drained section, as our coroutines 1928 * servicing pending requests might enter blk_pread(). 1929 */ 1930 blk_set_disable_request_queuing(blk, true); 1931 1932 blk_add_aio_context_notifier(blk, blk_aio_attached, blk_aio_detach, exp); 1933 1934 blk_set_dev_ops(blk, &nbd_block_ops, exp); 1935 1936 QTAILQ_INSERT_TAIL(&exports, exp, next); 1937 1938 bdrv_graph_rdunlock_main_loop(); 1939 1940 return 0; 1941 1942 fail: 1943 bdrv_graph_rdunlock_main_loop(); 1944 g_free(exp->export_bitmaps); 1945 g_free(exp->name); 1946 g_free(exp->description); 1947 return ret; 1948 } 1949 1950 NBDExport *nbd_export_find(const char *name) 1951 { 1952 NBDExport *exp; 1953 QTAILQ_FOREACH(exp, &exports, next) { 1954 if (strcmp(name, exp->name) == 0) { 1955 return exp; 1956 } 1957 } 1958 1959 return NULL; 1960 } 1961 1962 AioContext * 1963 nbd_export_aio_context(NBDExport *exp) 1964 { 1965 return exp->common.ctx; 1966 } 1967 1968 static void nbd_export_request_shutdown(BlockExport *blk_exp) 1969 { 1970 NBDExport *exp = container_of(blk_exp, NBDExport, common); 1971 NBDClient *client, *next; 1972 1973 blk_exp_ref(&exp->common); 1974 /* 1975 * TODO: Should we expand QMP NbdServerRemoveNode enum to allow a 1976 * close mode that stops advertising the export to new clients but 1977 * still permits existing clients to run to completion? Because of 1978 * that possibility, nbd_export_close() can be called more than 1979 * once on an export. 1980 */ 1981 QTAILQ_FOREACH_SAFE(client, &exp->clients, next, next) { 1982 client_close(client, true); 1983 } 1984 if (exp->name) { 1985 g_free(exp->name); 1986 exp->name = NULL; 1987 QTAILQ_REMOVE(&exports, exp, next); 1988 } 1989 blk_exp_unref(&exp->common); 1990 } 1991 1992 static void nbd_export_delete(BlockExport *blk_exp) 1993 { 1994 size_t i; 1995 NBDExport *exp = container_of(blk_exp, NBDExport, common); 1996 1997 assert(exp->name == NULL); 1998 assert(QTAILQ_EMPTY(&exp->clients)); 1999 2000 g_free(exp->description); 2001 exp->description = NULL; 2002 2003 if (exp->eject_notifier_blk) { 2004 notifier_remove(&exp->eject_notifier); 2005 blk_unref(exp->eject_notifier_blk); 2006 } 2007 blk_remove_aio_context_notifier(exp->common.blk, blk_aio_attached, 2008 blk_aio_detach, exp); 2009 blk_set_disable_request_queuing(exp->common.blk, false); 2010 2011 for (i = 0; i < exp->nr_export_bitmaps; i++) { 2012 bdrv_dirty_bitmap_set_busy(exp->export_bitmaps[i], false); 2013 } 2014 } 2015 2016 const BlockExportDriver blk_exp_nbd = { 2017 .type = BLOCK_EXPORT_TYPE_NBD, 2018 .instance_size = sizeof(NBDExport), 2019 .create = nbd_export_create, 2020 .delete = nbd_export_delete, 2021 .request_shutdown = nbd_export_request_shutdown, 2022 }; 2023 2024 static int coroutine_fn nbd_co_send_iov(NBDClient *client, struct iovec *iov, 2025 unsigned niov, Error **errp) 2026 { 2027 int ret; 2028 2029 g_assert(qemu_in_coroutine()); 2030 qemu_co_mutex_lock(&client->send_lock); 2031 client->send_coroutine = qemu_coroutine_self(); 2032 2033 ret = qio_channel_writev_all(client->ioc, iov, niov, errp) < 0 ? -EIO : 0; 2034 2035 client->send_coroutine = NULL; 2036 qemu_co_mutex_unlock(&client->send_lock); 2037 2038 return ret; 2039 } 2040 2041 static inline void set_be_simple_reply(NBDSimpleReply *reply, uint64_t error, 2042 uint64_t cookie) 2043 { 2044 stl_be_p(&reply->magic, NBD_SIMPLE_REPLY_MAGIC); 2045 stl_be_p(&reply->error, error); 2046 stq_be_p(&reply->cookie, cookie); 2047 } 2048 2049 static int coroutine_fn nbd_co_send_simple_reply(NBDClient *client, 2050 NBDRequest *request, 2051 uint32_t error, 2052 void *data, 2053 uint64_t len, 2054 Error **errp) 2055 { 2056 NBDSimpleReply reply; 2057 int nbd_err = system_errno_to_nbd_errno(error); 2058 struct iovec iov[] = { 2059 {.iov_base = &reply, .iov_len = sizeof(reply)}, 2060 {.iov_base = data, .iov_len = len} 2061 }; 2062 2063 assert(!len || !nbd_err); 2064 assert(len <= NBD_MAX_BUFFER_SIZE); 2065 assert(client->mode < NBD_MODE_STRUCTURED || 2066 (client->mode == NBD_MODE_STRUCTURED && 2067 request->type != NBD_CMD_READ)); 2068 trace_nbd_co_send_simple_reply(request->cookie, nbd_err, 2069 nbd_err_lookup(nbd_err), len); 2070 set_be_simple_reply(&reply, nbd_err, request->cookie); 2071 2072 return nbd_co_send_iov(client, iov, 2, errp); 2073 } 2074 2075 /* 2076 * Prepare the header of a reply chunk for network transmission. 2077 * 2078 * On input, @iov is partially initialized: iov[0].iov_base must point 2079 * to an uninitialized NBDReply, while the remaining @niov elements 2080 * (if any) must be ready for transmission. This function then 2081 * populates iov[0] for transmission. 2082 */ 2083 static inline void set_be_chunk(NBDClient *client, struct iovec *iov, 2084 size_t niov, uint16_t flags, uint16_t type, 2085 NBDRequest *request) 2086 { 2087 size_t i, length = 0; 2088 2089 for (i = 1; i < niov; i++) { 2090 length += iov[i].iov_len; 2091 } 2092 assert(length <= NBD_MAX_BUFFER_SIZE + sizeof(NBDStructuredReadData)); 2093 2094 if (client->mode >= NBD_MODE_EXTENDED) { 2095 NBDExtendedReplyChunk *chunk = iov->iov_base; 2096 2097 iov[0].iov_len = sizeof(*chunk); 2098 stl_be_p(&chunk->magic, NBD_EXTENDED_REPLY_MAGIC); 2099 stw_be_p(&chunk->flags, flags); 2100 stw_be_p(&chunk->type, type); 2101 stq_be_p(&chunk->cookie, request->cookie); 2102 stq_be_p(&chunk->offset, request->from); 2103 stq_be_p(&chunk->length, length); 2104 } else { 2105 NBDStructuredReplyChunk *chunk = iov->iov_base; 2106 2107 iov[0].iov_len = sizeof(*chunk); 2108 stl_be_p(&chunk->magic, NBD_STRUCTURED_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 stl_be_p(&chunk->length, length); 2113 } 2114 } 2115 2116 static int coroutine_fn nbd_co_send_chunk_done(NBDClient *client, 2117 NBDRequest *request, 2118 Error **errp) 2119 { 2120 NBDReply hdr; 2121 struct iovec iov[] = { 2122 {.iov_base = &hdr}, 2123 }; 2124 2125 trace_nbd_co_send_chunk_done(request->cookie); 2126 set_be_chunk(client, iov, 1, NBD_REPLY_FLAG_DONE, 2127 NBD_REPLY_TYPE_NONE, request); 2128 return nbd_co_send_iov(client, iov, 1, errp); 2129 } 2130 2131 static int coroutine_fn nbd_co_send_chunk_read(NBDClient *client, 2132 NBDRequest *request, 2133 uint64_t offset, 2134 void *data, 2135 uint64_t size, 2136 bool final, 2137 Error **errp) 2138 { 2139 NBDReply hdr; 2140 NBDStructuredReadData chunk; 2141 struct iovec iov[] = { 2142 {.iov_base = &hdr}, 2143 {.iov_base = &chunk, .iov_len = sizeof(chunk)}, 2144 {.iov_base = data, .iov_len = size} 2145 }; 2146 2147 assert(size && size <= NBD_MAX_BUFFER_SIZE); 2148 trace_nbd_co_send_chunk_read(request->cookie, offset, data, size); 2149 set_be_chunk(client, iov, 3, final ? NBD_REPLY_FLAG_DONE : 0, 2150 NBD_REPLY_TYPE_OFFSET_DATA, request); 2151 stq_be_p(&chunk.offset, offset); 2152 2153 return nbd_co_send_iov(client, iov, 3, errp); 2154 } 2155 2156 static int coroutine_fn nbd_co_send_chunk_error(NBDClient *client, 2157 NBDRequest *request, 2158 uint32_t error, 2159 const char *msg, 2160 Error **errp) 2161 { 2162 NBDReply hdr; 2163 NBDStructuredError chunk; 2164 int nbd_err = system_errno_to_nbd_errno(error); 2165 struct iovec iov[] = { 2166 {.iov_base = &hdr}, 2167 {.iov_base = &chunk, .iov_len = sizeof(chunk)}, 2168 {.iov_base = (char *)msg, .iov_len = msg ? strlen(msg) : 0}, 2169 }; 2170 2171 assert(nbd_err); 2172 trace_nbd_co_send_chunk_error(request->cookie, nbd_err, 2173 nbd_err_lookup(nbd_err), msg ? msg : ""); 2174 set_be_chunk(client, iov, 3, NBD_REPLY_FLAG_DONE, 2175 NBD_REPLY_TYPE_ERROR, request); 2176 stl_be_p(&chunk.error, nbd_err); 2177 stw_be_p(&chunk.message_length, iov[2].iov_len); 2178 2179 return nbd_co_send_iov(client, iov, 3, errp); 2180 } 2181 2182 /* Do a sparse read and send the structured reply to the client. 2183 * Returns -errno if sending fails. blk_co_block_status_above() failure is 2184 * reported to the client, at which point this function succeeds. 2185 */ 2186 static int coroutine_fn nbd_co_send_sparse_read(NBDClient *client, 2187 NBDRequest *request, 2188 uint64_t offset, 2189 uint8_t *data, 2190 uint64_t size, 2191 Error **errp) 2192 { 2193 int ret = 0; 2194 NBDExport *exp = client->exp; 2195 size_t progress = 0; 2196 2197 assert(size <= NBD_MAX_BUFFER_SIZE); 2198 while (progress < size) { 2199 int64_t pnum; 2200 int status = blk_co_block_status_above(exp->common.blk, NULL, 2201 offset + progress, 2202 size - progress, &pnum, NULL, 2203 NULL); 2204 bool final; 2205 2206 if (status < 0) { 2207 char *msg = g_strdup_printf("unable to check for holes: %s", 2208 strerror(-status)); 2209 2210 ret = nbd_co_send_chunk_error(client, request, -status, msg, errp); 2211 g_free(msg); 2212 return ret; 2213 } 2214 assert(pnum && pnum <= size - progress); 2215 final = progress + pnum == size; 2216 if (status & BDRV_BLOCK_ZERO) { 2217 NBDReply hdr; 2218 NBDStructuredReadHole chunk; 2219 struct iovec iov[] = { 2220 {.iov_base = &hdr}, 2221 {.iov_base = &chunk, .iov_len = sizeof(chunk)}, 2222 }; 2223 2224 trace_nbd_co_send_chunk_read_hole(request->cookie, 2225 offset + progress, pnum); 2226 set_be_chunk(client, iov, 2, 2227 final ? NBD_REPLY_FLAG_DONE : 0, 2228 NBD_REPLY_TYPE_OFFSET_HOLE, request); 2229 stq_be_p(&chunk.offset, offset + progress); 2230 stl_be_p(&chunk.length, pnum); 2231 ret = nbd_co_send_iov(client, iov, 2, errp); 2232 } else { 2233 ret = blk_co_pread(exp->common.blk, offset + progress, pnum, 2234 data + progress, 0); 2235 if (ret < 0) { 2236 error_setg_errno(errp, -ret, "reading from file failed"); 2237 break; 2238 } 2239 ret = nbd_co_send_chunk_read(client, request, offset + progress, 2240 data + progress, pnum, final, errp); 2241 } 2242 2243 if (ret < 0) { 2244 break; 2245 } 2246 progress += pnum; 2247 } 2248 return ret; 2249 } 2250 2251 typedef struct NBDExtentArray { 2252 NBDExtent64 *extents; 2253 unsigned int nb_alloc; 2254 unsigned int count; 2255 uint64_t total_length; 2256 bool extended; 2257 bool can_add; 2258 bool converted_to_be; 2259 } NBDExtentArray; 2260 2261 static NBDExtentArray *nbd_extent_array_new(unsigned int nb_alloc, 2262 NBDMode mode) 2263 { 2264 NBDExtentArray *ea = g_new0(NBDExtentArray, 1); 2265 2266 assert(mode >= NBD_MODE_STRUCTURED); 2267 ea->nb_alloc = nb_alloc; 2268 ea->extents = g_new(NBDExtent64, nb_alloc); 2269 ea->extended = mode >= NBD_MODE_EXTENDED; 2270 ea->can_add = true; 2271 2272 return ea; 2273 } 2274 2275 static void nbd_extent_array_free(NBDExtentArray *ea) 2276 { 2277 g_free(ea->extents); 2278 g_free(ea); 2279 } 2280 G_DEFINE_AUTOPTR_CLEANUP_FUNC(NBDExtentArray, nbd_extent_array_free) 2281 2282 /* Further modifications of the array after conversion are abandoned */ 2283 static void nbd_extent_array_convert_to_be(NBDExtentArray *ea) 2284 { 2285 int i; 2286 2287 assert(!ea->converted_to_be); 2288 assert(ea->extended); 2289 ea->can_add = false; 2290 ea->converted_to_be = true; 2291 2292 for (i = 0; i < ea->count; i++) { 2293 ea->extents[i].length = cpu_to_be64(ea->extents[i].length); 2294 ea->extents[i].flags = cpu_to_be64(ea->extents[i].flags); 2295 } 2296 } 2297 2298 /* Further modifications of the array after conversion are abandoned */ 2299 static NBDExtent32 *nbd_extent_array_convert_to_narrow(NBDExtentArray *ea) 2300 { 2301 int i; 2302 NBDExtent32 *extents = g_new(NBDExtent32, ea->count); 2303 2304 assert(!ea->converted_to_be); 2305 assert(!ea->extended); 2306 ea->can_add = false; 2307 ea->converted_to_be = true; 2308 2309 for (i = 0; i < ea->count; i++) { 2310 assert((ea->extents[i].length | ea->extents[i].flags) <= UINT32_MAX); 2311 extents[i].length = cpu_to_be32(ea->extents[i].length); 2312 extents[i].flags = cpu_to_be32(ea->extents[i].flags); 2313 } 2314 2315 return extents; 2316 } 2317 2318 /* 2319 * Add extent to NBDExtentArray. If extent can't be added (no available space), 2320 * return -1. 2321 * For safety, when returning -1 for the first time, .can_add is set to false, 2322 * and further calls to nbd_extent_array_add() will crash. 2323 * (this avoids the situation where a caller ignores failure to add one extent, 2324 * where adding another extent that would squash into the last array entry 2325 * would result in an incorrect range reported to the client) 2326 */ 2327 static int nbd_extent_array_add(NBDExtentArray *ea, 2328 uint64_t length, uint32_t flags) 2329 { 2330 assert(ea->can_add); 2331 2332 if (!length) { 2333 return 0; 2334 } 2335 if (!ea->extended) { 2336 assert(length <= UINT32_MAX); 2337 } 2338 2339 /* Extend previous extent if flags are the same */ 2340 if (ea->count > 0 && flags == ea->extents[ea->count - 1].flags) { 2341 uint64_t sum = length + ea->extents[ea->count - 1].length; 2342 2343 /* 2344 * sum cannot overflow: the block layer bounds image size at 2345 * 2^63, and ea->extents[].length comes from the block layer. 2346 */ 2347 assert(sum >= length); 2348 if (sum <= UINT32_MAX || ea->extended) { 2349 ea->extents[ea->count - 1].length = sum; 2350 ea->total_length += length; 2351 return 0; 2352 } 2353 } 2354 2355 if (ea->count >= ea->nb_alloc) { 2356 ea->can_add = false; 2357 return -1; 2358 } 2359 2360 ea->total_length += length; 2361 ea->extents[ea->count] = (NBDExtent64) {.length = length, .flags = flags}; 2362 ea->count++; 2363 2364 return 0; 2365 } 2366 2367 static int coroutine_fn blockstatus_to_extents(BlockBackend *blk, 2368 uint64_t offset, uint64_t bytes, 2369 NBDExtentArray *ea) 2370 { 2371 while (bytes) { 2372 uint32_t flags; 2373 int64_t num; 2374 int ret = blk_co_block_status_above(blk, NULL, offset, bytes, &num, 2375 NULL, NULL); 2376 2377 if (ret < 0) { 2378 return ret; 2379 } 2380 2381 flags = (ret & BDRV_BLOCK_DATA ? 0 : NBD_STATE_HOLE) | 2382 (ret & BDRV_BLOCK_ZERO ? NBD_STATE_ZERO : 0); 2383 2384 if (nbd_extent_array_add(ea, num, flags) < 0) { 2385 return 0; 2386 } 2387 2388 offset += num; 2389 bytes -= num; 2390 } 2391 2392 return 0; 2393 } 2394 2395 static int coroutine_fn blockalloc_to_extents(BlockBackend *blk, 2396 uint64_t offset, uint64_t bytes, 2397 NBDExtentArray *ea) 2398 { 2399 while (bytes) { 2400 int64_t num; 2401 int ret = blk_co_is_allocated_above(blk, NULL, false, offset, bytes, 2402 &num); 2403 2404 if (ret < 0) { 2405 return ret; 2406 } 2407 2408 if (nbd_extent_array_add(ea, num, ret) < 0) { 2409 return 0; 2410 } 2411 2412 offset += num; 2413 bytes -= num; 2414 } 2415 2416 return 0; 2417 } 2418 2419 /* 2420 * nbd_co_send_extents 2421 * 2422 * @ea is converted to BE by the function 2423 * @last controls whether NBD_REPLY_FLAG_DONE is sent. 2424 */ 2425 static int coroutine_fn 2426 nbd_co_send_extents(NBDClient *client, NBDRequest *request, NBDExtentArray *ea, 2427 bool last, uint32_t context_id, Error **errp) 2428 { 2429 NBDReply hdr; 2430 NBDStructuredMeta meta; 2431 NBDExtendedMeta meta_ext; 2432 g_autofree NBDExtent32 *extents = NULL; 2433 uint16_t type; 2434 struct iovec iov[] = { {.iov_base = &hdr}, {0}, {0} }; 2435 2436 if (client->mode >= NBD_MODE_EXTENDED) { 2437 type = NBD_REPLY_TYPE_BLOCK_STATUS_EXT; 2438 2439 iov[1].iov_base = &meta_ext; 2440 iov[1].iov_len = sizeof(meta_ext); 2441 stl_be_p(&meta_ext.context_id, context_id); 2442 stl_be_p(&meta_ext.count, ea->count); 2443 2444 nbd_extent_array_convert_to_be(ea); 2445 iov[2].iov_base = ea->extents; 2446 iov[2].iov_len = ea->count * sizeof(ea->extents[0]); 2447 } else { 2448 type = NBD_REPLY_TYPE_BLOCK_STATUS; 2449 2450 iov[1].iov_base = &meta; 2451 iov[1].iov_len = sizeof(meta); 2452 stl_be_p(&meta.context_id, context_id); 2453 2454 extents = nbd_extent_array_convert_to_narrow(ea); 2455 iov[2].iov_base = extents; 2456 iov[2].iov_len = ea->count * sizeof(extents[0]); 2457 } 2458 2459 trace_nbd_co_send_extents(request->cookie, ea->count, context_id, 2460 ea->total_length, last); 2461 set_be_chunk(client, iov, 3, last ? NBD_REPLY_FLAG_DONE : 0, type, 2462 request); 2463 2464 return nbd_co_send_iov(client, iov, 3, errp); 2465 } 2466 2467 /* Get block status from the exported device and send it to the client */ 2468 static int 2469 coroutine_fn nbd_co_send_block_status(NBDClient *client, NBDRequest *request, 2470 BlockBackend *blk, uint64_t offset, 2471 uint64_t length, bool dont_fragment, 2472 bool last, uint32_t context_id, 2473 Error **errp) 2474 { 2475 int ret; 2476 unsigned int nb_extents = dont_fragment ? 1 : NBD_MAX_BLOCK_STATUS_EXTENTS; 2477 g_autoptr(NBDExtentArray) ea = 2478 nbd_extent_array_new(nb_extents, client->mode); 2479 2480 if (context_id == NBD_META_ID_BASE_ALLOCATION) { 2481 ret = blockstatus_to_extents(blk, offset, length, ea); 2482 } else { 2483 ret = blockalloc_to_extents(blk, offset, length, ea); 2484 } 2485 if (ret < 0) { 2486 return nbd_co_send_chunk_error(client, request, -ret, 2487 "can't get block status", errp); 2488 } 2489 2490 return nbd_co_send_extents(client, request, ea, last, context_id, errp); 2491 } 2492 2493 /* Populate @ea from a dirty bitmap. */ 2494 static void bitmap_to_extents(BdrvDirtyBitmap *bitmap, 2495 uint64_t offset, uint64_t length, 2496 NBDExtentArray *es) 2497 { 2498 int64_t start, dirty_start, dirty_count; 2499 int64_t end = offset + length; 2500 bool full = false; 2501 int64_t bound = es->extended ? INT64_MAX : INT32_MAX; 2502 2503 bdrv_dirty_bitmap_lock(bitmap); 2504 2505 for (start = offset; 2506 bdrv_dirty_bitmap_next_dirty_area(bitmap, start, end, bound, 2507 &dirty_start, &dirty_count); 2508 start = dirty_start + dirty_count) 2509 { 2510 if ((nbd_extent_array_add(es, dirty_start - start, 0) < 0) || 2511 (nbd_extent_array_add(es, dirty_count, NBD_STATE_DIRTY) < 0)) 2512 { 2513 full = true; 2514 break; 2515 } 2516 } 2517 2518 if (!full) { 2519 /* last non dirty extent, nothing to do if array is now full */ 2520 (void) nbd_extent_array_add(es, end - start, 0); 2521 } 2522 2523 bdrv_dirty_bitmap_unlock(bitmap); 2524 } 2525 2526 static int coroutine_fn nbd_co_send_bitmap(NBDClient *client, 2527 NBDRequest *request, 2528 BdrvDirtyBitmap *bitmap, 2529 uint64_t offset, 2530 uint64_t length, bool dont_fragment, 2531 bool last, uint32_t context_id, 2532 Error **errp) 2533 { 2534 unsigned int nb_extents = dont_fragment ? 1 : NBD_MAX_BLOCK_STATUS_EXTENTS; 2535 g_autoptr(NBDExtentArray) ea = 2536 nbd_extent_array_new(nb_extents, client->mode); 2537 2538 bitmap_to_extents(bitmap, offset, length, ea); 2539 2540 return nbd_co_send_extents(client, request, ea, last, context_id, errp); 2541 } 2542 2543 /* 2544 * nbd_co_block_status_payload_read 2545 * Called when a client wants a subset of negotiated contexts via a 2546 * BLOCK_STATUS payload. Check the payload for valid length and 2547 * contents. On success, return 0 with request updated to effective 2548 * length. If request was invalid but all payload consumed, return 0 2549 * with request->len and request->contexts->count set to 0 (which will 2550 * trigger an appropriate NBD_EINVAL response later on). Return 2551 * negative errno if the payload was not fully consumed. 2552 */ 2553 static int 2554 nbd_co_block_status_payload_read(NBDClient *client, NBDRequest *request, 2555 Error **errp) 2556 { 2557 uint64_t payload_len = request->len; 2558 g_autofree char *buf = NULL; 2559 size_t count, i, nr_bitmaps; 2560 uint32_t id; 2561 2562 if (payload_len > NBD_MAX_BUFFER_SIZE) { 2563 error_setg(errp, "len (%" PRIu64 ") is larger than max len (%u)", 2564 request->len, NBD_MAX_BUFFER_SIZE); 2565 return -EINVAL; 2566 } 2567 2568 assert(client->contexts.exp == client->exp); 2569 nr_bitmaps = client->exp->nr_export_bitmaps; 2570 request->contexts = g_new0(NBDMetaContexts, 1); 2571 request->contexts->exp = client->exp; 2572 2573 if (payload_len % sizeof(uint32_t) || 2574 payload_len < sizeof(NBDBlockStatusPayload) || 2575 payload_len > (sizeof(NBDBlockStatusPayload) + 2576 sizeof(id) * client->contexts.count)) { 2577 goto skip; 2578 } 2579 2580 buf = g_malloc(payload_len); 2581 if (nbd_read(client->ioc, buf, payload_len, 2582 "CMD_BLOCK_STATUS data", errp) < 0) { 2583 return -EIO; 2584 } 2585 trace_nbd_co_receive_request_payload_received(request->cookie, 2586 payload_len); 2587 request->contexts->bitmaps = g_new0(bool, nr_bitmaps); 2588 count = (payload_len - sizeof(NBDBlockStatusPayload)) / sizeof(id); 2589 payload_len = 0; 2590 2591 for (i = 0; i < count; i++) { 2592 id = ldl_be_p(buf + sizeof(NBDBlockStatusPayload) + sizeof(id) * i); 2593 if (id == NBD_META_ID_BASE_ALLOCATION) { 2594 if (!client->contexts.base_allocation || 2595 request->contexts->base_allocation) { 2596 goto skip; 2597 } 2598 request->contexts->base_allocation = true; 2599 } else if (id == NBD_META_ID_ALLOCATION_DEPTH) { 2600 if (!client->contexts.allocation_depth || 2601 request->contexts->allocation_depth) { 2602 goto skip; 2603 } 2604 request->contexts->allocation_depth = true; 2605 } else { 2606 unsigned idx = id - NBD_META_ID_DIRTY_BITMAP; 2607 2608 if (idx >= nr_bitmaps || !client->contexts.bitmaps[idx] || 2609 request->contexts->bitmaps[idx]) { 2610 goto skip; 2611 } 2612 request->contexts->bitmaps[idx] = true; 2613 } 2614 } 2615 2616 request->len = ldq_be_p(buf); 2617 request->contexts->count = count; 2618 return 0; 2619 2620 skip: 2621 trace_nbd_co_receive_block_status_payload_compliance(request->from, 2622 request->len); 2623 request->len = request->contexts->count = 0; 2624 return nbd_drop(client->ioc, payload_len, errp); 2625 } 2626 2627 /* nbd_co_receive_request 2628 * Collect a client request. Return 0 if request looks valid, -EIO to drop 2629 * connection right away, -EAGAIN to indicate we were interrupted and the 2630 * channel should be quiesced, and any other negative value to report an error 2631 * to the client (although the caller may still need to disconnect after 2632 * reporting the error). 2633 */ 2634 static int coroutine_fn nbd_co_receive_request(NBDRequestData *req, 2635 NBDRequest *request, 2636 Error **errp) 2637 { 2638 NBDClient *client = req->client; 2639 bool extended_with_payload; 2640 bool check_length = false; 2641 bool check_rofs = false; 2642 bool allocate_buffer = false; 2643 bool payload_okay = false; 2644 uint64_t payload_len = 0; 2645 int valid_flags = NBD_CMD_FLAG_FUA; 2646 int ret; 2647 2648 g_assert(qemu_in_coroutine()); 2649 ret = nbd_receive_request(client, request, errp); 2650 if (ret < 0) { 2651 return ret; 2652 } 2653 2654 trace_nbd_co_receive_request_decode_type(request->cookie, request->type, 2655 nbd_cmd_lookup(request->type)); 2656 extended_with_payload = client->mode >= NBD_MODE_EXTENDED && 2657 request->flags & NBD_CMD_FLAG_PAYLOAD_LEN; 2658 if (extended_with_payload) { 2659 payload_len = request->len; 2660 check_length = true; 2661 } 2662 2663 switch (request->type) { 2664 case NBD_CMD_DISC: 2665 /* Special case: we're going to disconnect without a reply, 2666 * whether or not flags, from, or len are bogus */ 2667 req->complete = true; 2668 return -EIO; 2669 2670 case NBD_CMD_READ: 2671 if (client->mode >= NBD_MODE_STRUCTURED) { 2672 valid_flags |= NBD_CMD_FLAG_DF; 2673 } 2674 check_length = true; 2675 allocate_buffer = true; 2676 break; 2677 2678 case NBD_CMD_WRITE: 2679 if (client->mode >= NBD_MODE_EXTENDED) { 2680 if (!extended_with_payload) { 2681 /* The client is noncompliant. Trace it, but proceed. */ 2682 trace_nbd_co_receive_ext_payload_compliance(request->from, 2683 request->len); 2684 } 2685 valid_flags |= NBD_CMD_FLAG_PAYLOAD_LEN; 2686 } 2687 payload_okay = true; 2688 payload_len = request->len; 2689 check_length = true; 2690 allocate_buffer = true; 2691 check_rofs = true; 2692 break; 2693 2694 case NBD_CMD_FLUSH: 2695 break; 2696 2697 case NBD_CMD_TRIM: 2698 check_rofs = true; 2699 break; 2700 2701 case NBD_CMD_CACHE: 2702 check_length = true; 2703 break; 2704 2705 case NBD_CMD_WRITE_ZEROES: 2706 valid_flags |= NBD_CMD_FLAG_NO_HOLE | NBD_CMD_FLAG_FAST_ZERO; 2707 check_rofs = true; 2708 break; 2709 2710 case NBD_CMD_BLOCK_STATUS: 2711 if (extended_with_payload) { 2712 ret = nbd_co_block_status_payload_read(client, request, errp); 2713 if (ret < 0) { 2714 return ret; 2715 } 2716 /* payload now consumed */ 2717 check_length = false; 2718 payload_len = 0; 2719 valid_flags |= NBD_CMD_FLAG_PAYLOAD_LEN; 2720 } else { 2721 request->contexts = &client->contexts; 2722 } 2723 valid_flags |= NBD_CMD_FLAG_REQ_ONE; 2724 break; 2725 2726 default: 2727 /* Unrecognized, will fail later */ 2728 ; 2729 } 2730 2731 /* Payload and buffer handling. */ 2732 if (!payload_len) { 2733 req->complete = true; 2734 } 2735 if (check_length && request->len > NBD_MAX_BUFFER_SIZE) { 2736 /* READ, WRITE, CACHE */ 2737 error_setg(errp, "len (%" PRIu64 ") is larger than max len (%u)", 2738 request->len, NBD_MAX_BUFFER_SIZE); 2739 return -EINVAL; 2740 } 2741 if (payload_len && !payload_okay) { 2742 /* 2743 * For now, we don't support payloads on other commands; but 2744 * we can keep the connection alive by ignoring the payload. 2745 * We will fail the command later with NBD_EINVAL for the use 2746 * of an unsupported flag (and not for access beyond bounds). 2747 */ 2748 assert(request->type != NBD_CMD_WRITE); 2749 request->len = 0; 2750 } 2751 if (allocate_buffer) { 2752 /* READ, WRITE */ 2753 req->data = blk_try_blockalign(client->exp->common.blk, 2754 request->len); 2755 if (req->data == NULL) { 2756 error_setg(errp, "No memory"); 2757 return -ENOMEM; 2758 } 2759 } 2760 if (payload_len) { 2761 if (payload_okay) { 2762 /* WRITE */ 2763 assert(req->data); 2764 ret = nbd_read(client->ioc, req->data, payload_len, 2765 "CMD_WRITE data", errp); 2766 } else { 2767 ret = nbd_drop(client->ioc, payload_len, errp); 2768 } 2769 if (ret < 0) { 2770 return -EIO; 2771 } 2772 req->complete = true; 2773 trace_nbd_co_receive_request_payload_received(request->cookie, 2774 payload_len); 2775 } 2776 2777 /* Sanity checks. */ 2778 if (client->exp->nbdflags & NBD_FLAG_READ_ONLY && check_rofs) { 2779 /* WRITE, TRIM, WRITE_ZEROES */ 2780 error_setg(errp, "Export is read-only"); 2781 return -EROFS; 2782 } 2783 if (request->from > client->exp->size || 2784 request->len > client->exp->size - request->from) { 2785 error_setg(errp, "operation past EOF; From: %" PRIu64 ", Len: %" PRIu64 2786 ", Size: %" PRIu64, request->from, request->len, 2787 client->exp->size); 2788 return (request->type == NBD_CMD_WRITE || 2789 request->type == NBD_CMD_WRITE_ZEROES) ? -ENOSPC : -EINVAL; 2790 } 2791 if (client->check_align && !QEMU_IS_ALIGNED(request->from | request->len, 2792 client->check_align)) { 2793 /* 2794 * The block layer gracefully handles unaligned requests, but 2795 * it's still worth tracing client non-compliance 2796 */ 2797 trace_nbd_co_receive_align_compliance(nbd_cmd_lookup(request->type), 2798 request->from, 2799 request->len, 2800 client->check_align); 2801 } 2802 if (request->flags & ~valid_flags) { 2803 error_setg(errp, "unsupported flags for command %s (got 0x%x)", 2804 nbd_cmd_lookup(request->type), request->flags); 2805 return -EINVAL; 2806 } 2807 2808 return 0; 2809 } 2810 2811 /* Send simple reply without a payload, or a structured error 2812 * @error_msg is ignored if @ret >= 0 2813 * Returns 0 if connection is still live, -errno on failure to talk to client 2814 */ 2815 static coroutine_fn int nbd_send_generic_reply(NBDClient *client, 2816 NBDRequest *request, 2817 int ret, 2818 const char *error_msg, 2819 Error **errp) 2820 { 2821 if (client->mode >= NBD_MODE_STRUCTURED && ret < 0) { 2822 return nbd_co_send_chunk_error(client, request, -ret, error_msg, errp); 2823 } else if (client->mode >= NBD_MODE_EXTENDED) { 2824 return nbd_co_send_chunk_done(client, request, errp); 2825 } else { 2826 return nbd_co_send_simple_reply(client, request, ret < 0 ? -ret : 0, 2827 NULL, 0, errp); 2828 } 2829 } 2830 2831 /* Handle NBD_CMD_READ request. 2832 * Return -errno if sending fails. Other errors are reported directly to the 2833 * client as an error reply. */ 2834 static coroutine_fn int nbd_do_cmd_read(NBDClient *client, NBDRequest *request, 2835 uint8_t *data, Error **errp) 2836 { 2837 int ret; 2838 NBDExport *exp = client->exp; 2839 2840 assert(request->type == NBD_CMD_READ); 2841 assert(request->len <= NBD_MAX_BUFFER_SIZE); 2842 2843 /* XXX: NBD Protocol only documents use of FUA with WRITE */ 2844 if (request->flags & NBD_CMD_FLAG_FUA) { 2845 ret = blk_co_flush(exp->common.blk); 2846 if (ret < 0) { 2847 return nbd_send_generic_reply(client, request, ret, 2848 "flush failed", errp); 2849 } 2850 } 2851 2852 if (client->mode >= NBD_MODE_STRUCTURED && 2853 !(request->flags & NBD_CMD_FLAG_DF) && request->len) 2854 { 2855 return nbd_co_send_sparse_read(client, request, request->from, 2856 data, request->len, errp); 2857 } 2858 2859 ret = blk_co_pread(exp->common.blk, request->from, request->len, data, 0); 2860 if (ret < 0) { 2861 return nbd_send_generic_reply(client, request, ret, 2862 "reading from file failed", errp); 2863 } 2864 2865 if (client->mode >= NBD_MODE_STRUCTURED) { 2866 if (request->len) { 2867 return nbd_co_send_chunk_read(client, request, request->from, data, 2868 request->len, true, errp); 2869 } else { 2870 return nbd_co_send_chunk_done(client, request, errp); 2871 } 2872 } else { 2873 return nbd_co_send_simple_reply(client, request, 0, 2874 data, request->len, errp); 2875 } 2876 } 2877 2878 /* 2879 * nbd_do_cmd_cache 2880 * 2881 * Handle NBD_CMD_CACHE request. 2882 * Return -errno if sending fails. Other errors are reported directly to the 2883 * client as an error reply. 2884 */ 2885 static coroutine_fn int nbd_do_cmd_cache(NBDClient *client, NBDRequest *request, 2886 Error **errp) 2887 { 2888 int ret; 2889 NBDExport *exp = client->exp; 2890 2891 assert(request->type == NBD_CMD_CACHE); 2892 assert(request->len <= NBD_MAX_BUFFER_SIZE); 2893 2894 ret = blk_co_preadv(exp->common.blk, request->from, request->len, 2895 NULL, BDRV_REQ_COPY_ON_READ | BDRV_REQ_PREFETCH); 2896 2897 return nbd_send_generic_reply(client, request, ret, 2898 "caching data failed", errp); 2899 } 2900 2901 /* Handle NBD request. 2902 * Return -errno if sending fails. Other errors are reported directly to the 2903 * client as an error reply. */ 2904 static coroutine_fn int nbd_handle_request(NBDClient *client, 2905 NBDRequest *request, 2906 uint8_t *data, Error **errp) 2907 { 2908 int ret; 2909 int flags; 2910 NBDExport *exp = client->exp; 2911 char *msg; 2912 size_t i; 2913 2914 switch (request->type) { 2915 case NBD_CMD_CACHE: 2916 return nbd_do_cmd_cache(client, request, errp); 2917 2918 case NBD_CMD_READ: 2919 return nbd_do_cmd_read(client, request, data, errp); 2920 2921 case NBD_CMD_WRITE: 2922 flags = 0; 2923 if (request->flags & NBD_CMD_FLAG_FUA) { 2924 flags |= BDRV_REQ_FUA; 2925 } 2926 assert(request->len <= NBD_MAX_BUFFER_SIZE); 2927 ret = blk_co_pwrite(exp->common.blk, request->from, request->len, data, 2928 flags); 2929 return nbd_send_generic_reply(client, request, ret, 2930 "writing to file failed", errp); 2931 2932 case NBD_CMD_WRITE_ZEROES: 2933 flags = 0; 2934 if (request->flags & NBD_CMD_FLAG_FUA) { 2935 flags |= BDRV_REQ_FUA; 2936 } 2937 if (!(request->flags & NBD_CMD_FLAG_NO_HOLE)) { 2938 flags |= BDRV_REQ_MAY_UNMAP; 2939 } 2940 if (request->flags & NBD_CMD_FLAG_FAST_ZERO) { 2941 flags |= BDRV_REQ_NO_FALLBACK; 2942 } 2943 ret = blk_co_pwrite_zeroes(exp->common.blk, request->from, request->len, 2944 flags); 2945 return nbd_send_generic_reply(client, request, ret, 2946 "writing to file failed", errp); 2947 2948 case NBD_CMD_DISC: 2949 /* unreachable, thanks to special case in nbd_co_receive_request() */ 2950 abort(); 2951 2952 case NBD_CMD_FLUSH: 2953 ret = blk_co_flush(exp->common.blk); 2954 return nbd_send_generic_reply(client, request, ret, 2955 "flush failed", errp); 2956 2957 case NBD_CMD_TRIM: 2958 ret = blk_co_pdiscard(exp->common.blk, request->from, request->len); 2959 if (ret >= 0 && request->flags & NBD_CMD_FLAG_FUA) { 2960 ret = blk_co_flush(exp->common.blk); 2961 } 2962 return nbd_send_generic_reply(client, request, ret, 2963 "discard failed", errp); 2964 2965 case NBD_CMD_BLOCK_STATUS: 2966 assert(request->contexts); 2967 assert(client->mode >= NBD_MODE_EXTENDED || 2968 request->len <= UINT32_MAX); 2969 if (request->contexts->count) { 2970 bool dont_fragment = request->flags & NBD_CMD_FLAG_REQ_ONE; 2971 int contexts_remaining = request->contexts->count; 2972 2973 if (!request->len) { 2974 return nbd_send_generic_reply(client, request, -EINVAL, 2975 "need non-zero length", errp); 2976 } 2977 if (request->contexts->base_allocation) { 2978 ret = nbd_co_send_block_status(client, request, 2979 exp->common.blk, 2980 request->from, 2981 request->len, dont_fragment, 2982 !--contexts_remaining, 2983 NBD_META_ID_BASE_ALLOCATION, 2984 errp); 2985 if (ret < 0) { 2986 return ret; 2987 } 2988 } 2989 2990 if (request->contexts->allocation_depth) { 2991 ret = nbd_co_send_block_status(client, request, 2992 exp->common.blk, 2993 request->from, request->len, 2994 dont_fragment, 2995 !--contexts_remaining, 2996 NBD_META_ID_ALLOCATION_DEPTH, 2997 errp); 2998 if (ret < 0) { 2999 return ret; 3000 } 3001 } 3002 3003 assert(request->contexts->exp == client->exp); 3004 for (i = 0; i < client->exp->nr_export_bitmaps; i++) { 3005 if (!request->contexts->bitmaps[i]) { 3006 continue; 3007 } 3008 ret = nbd_co_send_bitmap(client, request, 3009 client->exp->export_bitmaps[i], 3010 request->from, request->len, 3011 dont_fragment, !--contexts_remaining, 3012 NBD_META_ID_DIRTY_BITMAP + i, errp); 3013 if (ret < 0) { 3014 return ret; 3015 } 3016 } 3017 3018 assert(!contexts_remaining); 3019 3020 return 0; 3021 } else if (client->contexts.count) { 3022 return nbd_send_generic_reply(client, request, -EINVAL, 3023 "CMD_BLOCK_STATUS payload not valid", 3024 errp); 3025 } else { 3026 return nbd_send_generic_reply(client, request, -EINVAL, 3027 "CMD_BLOCK_STATUS not negotiated", 3028 errp); 3029 } 3030 3031 default: 3032 msg = g_strdup_printf("invalid request type (%" PRIu32 ") received", 3033 request->type); 3034 ret = nbd_send_generic_reply(client, request, -EINVAL, msg, 3035 errp); 3036 g_free(msg); 3037 return ret; 3038 } 3039 } 3040 3041 /* Owns a reference to the NBDClient passed as opaque. */ 3042 static coroutine_fn void nbd_trip(void *opaque) 3043 { 3044 NBDRequestData *req = opaque; 3045 NBDClient *client = req->client; 3046 NBDRequest request = { 0 }; /* GCC thinks it can be used uninitialized */ 3047 int ret; 3048 Error *local_err = NULL; 3049 3050 /* 3051 * Note that nbd_client_put() and client_close() must be called from the 3052 * main loop thread. Use aio_co_reschedule_self() to switch AioContext 3053 * before calling these functions. 3054 */ 3055 3056 trace_nbd_trip(); 3057 3058 qemu_mutex_lock(&client->lock); 3059 3060 if (client->closing) { 3061 goto done; 3062 } 3063 3064 if (client->quiescing) { 3065 /* 3066 * We're switching between AIO contexts. Don't attempt to receive a new 3067 * request and kick the main context which may be waiting for us. 3068 */ 3069 client->recv_coroutine = NULL; 3070 aio_wait_kick(); 3071 goto done; 3072 } 3073 3074 /* 3075 * nbd_co_receive_request() returns -EAGAIN when nbd_drained_begin() has 3076 * set client->quiescing but by the time we get back nbd_drained_end() may 3077 * have already cleared client->quiescing. In that case we try again 3078 * because nothing else will spawn an nbd_trip() coroutine until we set 3079 * client->recv_coroutine = NULL further down. 3080 */ 3081 do { 3082 assert(client->recv_coroutine == qemu_coroutine_self()); 3083 qemu_mutex_unlock(&client->lock); 3084 ret = nbd_co_receive_request(req, &request, &local_err); 3085 qemu_mutex_lock(&client->lock); 3086 } while (ret == -EAGAIN && !client->quiescing); 3087 3088 client->recv_coroutine = NULL; 3089 3090 if (client->closing) { 3091 /* 3092 * The client may be closed when we are blocked in 3093 * nbd_co_receive_request() 3094 */ 3095 goto done; 3096 } 3097 3098 if (ret == -EAGAIN) { 3099 goto done; 3100 } 3101 3102 nbd_client_receive_next_request(client); 3103 3104 if (ret == -EIO) { 3105 goto disconnect; 3106 } 3107 3108 qemu_mutex_unlock(&client->lock); 3109 qio_channel_set_cork(client->ioc, true); 3110 3111 if (ret < 0) { 3112 /* It wasn't -EIO, so, according to nbd_co_receive_request() 3113 * semantics, we should return the error to the client. */ 3114 Error *export_err = local_err; 3115 3116 local_err = NULL; 3117 ret = nbd_send_generic_reply(client, &request, -EINVAL, 3118 error_get_pretty(export_err), &local_err); 3119 error_free(export_err); 3120 } else { 3121 ret = nbd_handle_request(client, &request, req->data, &local_err); 3122 } 3123 if (request.contexts && request.contexts != &client->contexts) { 3124 assert(request.type == NBD_CMD_BLOCK_STATUS); 3125 g_free(request.contexts->bitmaps); 3126 g_free(request.contexts); 3127 } 3128 3129 qio_channel_set_cork(client->ioc, false); 3130 qemu_mutex_lock(&client->lock); 3131 3132 if (ret < 0) { 3133 error_prepend(&local_err, "Failed to send reply: "); 3134 goto disconnect; 3135 } 3136 3137 /* 3138 * We must disconnect after NBD_CMD_WRITE or BLOCK_STATUS with 3139 * payload if we did not read the payload. 3140 */ 3141 if (!req->complete) { 3142 error_setg(&local_err, "Request handling failed in intermediate state"); 3143 goto disconnect; 3144 } 3145 3146 done: 3147 nbd_request_put(req); 3148 3149 qemu_mutex_unlock(&client->lock); 3150 3151 if (!nbd_client_put_nonzero(client)) { 3152 aio_co_reschedule_self(qemu_get_aio_context()); 3153 nbd_client_put(client); 3154 } 3155 return; 3156 3157 disconnect: 3158 if (local_err) { 3159 error_reportf_err(local_err, "Disconnect client, due to: "); 3160 } 3161 3162 nbd_request_put(req); 3163 qemu_mutex_unlock(&client->lock); 3164 3165 aio_co_reschedule_self(qemu_get_aio_context()); 3166 client_close(client, true); 3167 nbd_client_put(client); 3168 } 3169 3170 /* 3171 * Runs in export AioContext and main loop thread. Caller must hold 3172 * client->lock. 3173 */ 3174 static void nbd_client_receive_next_request(NBDClient *client) 3175 { 3176 NBDRequestData *req; 3177 3178 if (!client->recv_coroutine && client->nb_requests < MAX_NBD_REQUESTS && 3179 !client->quiescing) { 3180 nbd_client_get(client); 3181 req = nbd_request_get(client); 3182 client->recv_coroutine = qemu_coroutine_create(nbd_trip, req); 3183 aio_co_schedule(client->exp->common.ctx, client->recv_coroutine); 3184 } 3185 } 3186 3187 static coroutine_fn void nbd_co_client_start(void *opaque) 3188 { 3189 NBDClient *client = opaque; 3190 Error *local_err = NULL; 3191 3192 qemu_co_mutex_init(&client->send_lock); 3193 3194 if (nbd_negotiate(client, &local_err)) { 3195 if (local_err) { 3196 error_report_err(local_err); 3197 } 3198 client_close(client, false); 3199 return; 3200 } 3201 3202 WITH_QEMU_LOCK_GUARD(&client->lock) { 3203 nbd_client_receive_next_request(client); 3204 } 3205 } 3206 3207 /* 3208 * Create a new client listener using the given channel @sioc. 3209 * Begin servicing it in a coroutine. When the connection closes, call 3210 * @close_fn with an indication of whether the client completed negotiation. 3211 */ 3212 void nbd_client_new(QIOChannelSocket *sioc, 3213 QCryptoTLSCreds *tlscreds, 3214 const char *tlsauthz, 3215 void (*close_fn)(NBDClient *, bool)) 3216 { 3217 NBDClient *client; 3218 Coroutine *co; 3219 3220 client = g_new0(NBDClient, 1); 3221 qemu_mutex_init(&client->lock); 3222 client->refcount = 1; 3223 client->tlscreds = tlscreds; 3224 if (tlscreds) { 3225 object_ref(OBJECT(client->tlscreds)); 3226 } 3227 client->tlsauthz = g_strdup(tlsauthz); 3228 client->sioc = sioc; 3229 qio_channel_set_delay(QIO_CHANNEL(sioc), false); 3230 object_ref(OBJECT(client->sioc)); 3231 client->ioc = QIO_CHANNEL(sioc); 3232 object_ref(OBJECT(client->ioc)); 3233 client->close_fn = close_fn; 3234 3235 co = qemu_coroutine_create(nbd_co_client_start, client); 3236 qemu_coroutine_enter(co); 3237 } 3238