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 /* NBDExportMetaContexts represents a list of contexts to be exported, 109 * as selected by NBD_OPT_SET_META_CONTEXT. Also used for 110 * NBD_OPT_LIST_META_CONTEXT. */ 111 typedef struct NBDExportMetaContexts { 112 NBDExport *exp; 113 size_t count; /* number of negotiated contexts */ 114 bool base_allocation; /* export base:allocation context (block status) */ 115 bool allocation_depth; /* export qemu:allocation-depth */ 116 bool *bitmaps; /* 117 * export qemu:dirty-bitmap:<export bitmap name>, 118 * sized by exp->nr_export_bitmaps 119 */ 120 } NBDExportMetaContexts; 121 122 struct NBDClient { 123 int refcount; 124 void (*close_fn)(NBDClient *client, bool negotiated); 125 126 NBDExport *exp; 127 QCryptoTLSCreds *tlscreds; 128 char *tlsauthz; 129 QIOChannelSocket *sioc; /* The underlying data channel */ 130 QIOChannel *ioc; /* The current I/O channel which may differ (eg TLS) */ 131 132 Coroutine *recv_coroutine; 133 134 CoMutex send_lock; 135 Coroutine *send_coroutine; 136 137 bool read_yielding; 138 bool quiescing; 139 140 QTAILQ_ENTRY(NBDClient) next; 141 int nb_requests; 142 bool closing; 143 144 uint32_t check_align; /* If non-zero, check for aligned client requests */ 145 146 NBDMode mode; 147 NBDExportMetaContexts export_meta; 148 149 uint32_t opt; /* Current option being negotiated */ 150 uint32_t optlen; /* remaining length of data in ioc for the option being 151 negotiated now */ 152 }; 153 154 static void nbd_client_receive_next_request(NBDClient *client); 155 156 /* Basic flow for negotiation 157 158 Server Client 159 Negotiate 160 161 or 162 163 Server Client 164 Negotiate #1 165 Option 166 Negotiate #2 167 168 ---- 169 170 followed by 171 172 Server Client 173 Request 174 Response 175 Request 176 Response 177 ... 178 ... 179 Request (type == 2) 180 181 */ 182 183 static inline void set_be_option_rep(NBDOptionReply *rep, uint32_t option, 184 uint32_t type, uint32_t length) 185 { 186 stq_be_p(&rep->magic, NBD_REP_MAGIC); 187 stl_be_p(&rep->option, option); 188 stl_be_p(&rep->type, type); 189 stl_be_p(&rep->length, length); 190 } 191 192 /* Send a reply header, including length, but no payload. 193 * Return -errno on error, 0 on success. */ 194 static int nbd_negotiate_send_rep_len(NBDClient *client, uint32_t type, 195 uint32_t len, Error **errp) 196 { 197 NBDOptionReply rep; 198 199 trace_nbd_negotiate_send_rep_len(client->opt, nbd_opt_lookup(client->opt), 200 type, nbd_rep_lookup(type), len); 201 202 assert(len < NBD_MAX_BUFFER_SIZE); 203 204 set_be_option_rep(&rep, client->opt, type, len); 205 return nbd_write(client->ioc, &rep, sizeof(rep), errp); 206 } 207 208 /* Send a reply header with default 0 length. 209 * Return -errno on error, 0 on success. */ 210 static int nbd_negotiate_send_rep(NBDClient *client, uint32_t type, 211 Error **errp) 212 { 213 return nbd_negotiate_send_rep_len(client, type, 0, errp); 214 } 215 216 /* Send an error reply. 217 * Return -errno on error, 0 on success. */ 218 static int G_GNUC_PRINTF(4, 0) 219 nbd_negotiate_send_rep_verr(NBDClient *client, uint32_t type, 220 Error **errp, const char *fmt, va_list va) 221 { 222 ERRP_GUARD(); 223 g_autofree char *msg = NULL; 224 int ret; 225 size_t len; 226 227 msg = g_strdup_vprintf(fmt, va); 228 len = strlen(msg); 229 assert(len < NBD_MAX_STRING_SIZE); 230 trace_nbd_negotiate_send_rep_err(msg); 231 ret = nbd_negotiate_send_rep_len(client, type, len, errp); 232 if (ret < 0) { 233 return ret; 234 } 235 if (nbd_write(client->ioc, msg, len, errp) < 0) { 236 error_prepend(errp, "write failed (error message): "); 237 return -EIO; 238 } 239 240 return 0; 241 } 242 243 /* 244 * Return a malloc'd copy of @name suitable for use in an error reply. 245 */ 246 static char * 247 nbd_sanitize_name(const char *name) 248 { 249 if (strnlen(name, 80) < 80) { 250 return g_strdup(name); 251 } 252 /* XXX Should we also try to sanitize any control characters? */ 253 return g_strdup_printf("%.80s...", name); 254 } 255 256 /* Send an error reply. 257 * Return -errno on error, 0 on success. */ 258 static int G_GNUC_PRINTF(4, 5) 259 nbd_negotiate_send_rep_err(NBDClient *client, uint32_t type, 260 Error **errp, const char *fmt, ...) 261 { 262 va_list va; 263 int ret; 264 265 va_start(va, fmt); 266 ret = nbd_negotiate_send_rep_verr(client, type, errp, fmt, va); 267 va_end(va); 268 return ret; 269 } 270 271 /* Drop remainder of the current option, and send a reply with the 272 * given error type and message. Return -errno on read or write 273 * failure; or 0 if connection is still live. */ 274 static int G_GNUC_PRINTF(4, 0) 275 nbd_opt_vdrop(NBDClient *client, uint32_t type, Error **errp, 276 const char *fmt, va_list va) 277 { 278 int ret = nbd_drop(client->ioc, client->optlen, errp); 279 280 client->optlen = 0; 281 if (!ret) { 282 ret = nbd_negotiate_send_rep_verr(client, type, errp, fmt, va); 283 } 284 return ret; 285 } 286 287 static int G_GNUC_PRINTF(4, 5) 288 nbd_opt_drop(NBDClient *client, uint32_t type, Error **errp, 289 const char *fmt, ...) 290 { 291 int ret; 292 va_list va; 293 294 va_start(va, fmt); 295 ret = nbd_opt_vdrop(client, type, errp, fmt, va); 296 va_end(va); 297 298 return ret; 299 } 300 301 static int G_GNUC_PRINTF(3, 4) 302 nbd_opt_invalid(NBDClient *client, Error **errp, const char *fmt, ...) 303 { 304 int ret; 305 va_list va; 306 307 va_start(va, fmt); 308 ret = nbd_opt_vdrop(client, NBD_REP_ERR_INVALID, errp, fmt, va); 309 va_end(va); 310 311 return ret; 312 } 313 314 /* Read size bytes from the unparsed payload of the current option. 315 * If @check_nul, require that no NUL bytes appear in buffer. 316 * Return -errno on I/O error, 0 if option was completely handled by 317 * sending a reply about inconsistent lengths, or 1 on success. */ 318 static int nbd_opt_read(NBDClient *client, void *buffer, size_t size, 319 bool check_nul, Error **errp) 320 { 321 if (size > client->optlen) { 322 return nbd_opt_invalid(client, errp, 323 "Inconsistent lengths in option %s", 324 nbd_opt_lookup(client->opt)); 325 } 326 client->optlen -= size; 327 if (qio_channel_read_all(client->ioc, buffer, size, errp) < 0) { 328 return -EIO; 329 } 330 331 if (check_nul && strnlen(buffer, size) != size) { 332 return nbd_opt_invalid(client, errp, 333 "Unexpected embedded NUL in option %s", 334 nbd_opt_lookup(client->opt)); 335 } 336 return 1; 337 } 338 339 /* Drop size bytes from the unparsed payload of the current option. 340 * Return -errno on I/O error, 0 if option was completely handled by 341 * sending a reply about inconsistent lengths, or 1 on success. */ 342 static int nbd_opt_skip(NBDClient *client, size_t size, Error **errp) 343 { 344 if (size > client->optlen) { 345 return nbd_opt_invalid(client, errp, 346 "Inconsistent lengths in option %s", 347 nbd_opt_lookup(client->opt)); 348 } 349 client->optlen -= size; 350 return nbd_drop(client->ioc, size, errp) < 0 ? -EIO : 1; 351 } 352 353 /* nbd_opt_read_name 354 * 355 * Read a string with the format: 356 * uint32_t len (<= NBD_MAX_STRING_SIZE) 357 * len bytes string (not 0-terminated) 358 * 359 * On success, @name will be allocated. 360 * If @length is non-null, it will be set to the actual string length. 361 * 362 * Return -errno on I/O error, 0 if option was completely handled by 363 * sending a reply about inconsistent lengths, or 1 on success. 364 */ 365 static int nbd_opt_read_name(NBDClient *client, char **name, uint32_t *length, 366 Error **errp) 367 { 368 int ret; 369 uint32_t len; 370 g_autofree char *local_name = NULL; 371 372 *name = NULL; 373 ret = nbd_opt_read(client, &len, sizeof(len), false, errp); 374 if (ret <= 0) { 375 return ret; 376 } 377 len = cpu_to_be32(len); 378 379 if (len > NBD_MAX_STRING_SIZE) { 380 return nbd_opt_invalid(client, errp, 381 "Invalid name length: %" PRIu32, len); 382 } 383 384 local_name = g_malloc(len + 1); 385 ret = nbd_opt_read(client, local_name, len, true, errp); 386 if (ret <= 0) { 387 return ret; 388 } 389 local_name[len] = '\0'; 390 391 if (length) { 392 *length = len; 393 } 394 *name = g_steal_pointer(&local_name); 395 396 return 1; 397 } 398 399 /* Send a single NBD_REP_SERVER reply to NBD_OPT_LIST, including payload. 400 * Return -errno on error, 0 on success. */ 401 static int nbd_negotiate_send_rep_list(NBDClient *client, NBDExport *exp, 402 Error **errp) 403 { 404 ERRP_GUARD(); 405 size_t name_len, desc_len; 406 uint32_t len; 407 const char *name = exp->name ? exp->name : ""; 408 const char *desc = exp->description ? exp->description : ""; 409 QIOChannel *ioc = client->ioc; 410 int ret; 411 412 trace_nbd_negotiate_send_rep_list(name, desc); 413 name_len = strlen(name); 414 desc_len = strlen(desc); 415 assert(name_len <= NBD_MAX_STRING_SIZE && desc_len <= NBD_MAX_STRING_SIZE); 416 len = name_len + desc_len + sizeof(len); 417 ret = nbd_negotiate_send_rep_len(client, NBD_REP_SERVER, len, errp); 418 if (ret < 0) { 419 return ret; 420 } 421 422 len = cpu_to_be32(name_len); 423 if (nbd_write(ioc, &len, sizeof(len), errp) < 0) { 424 error_prepend(errp, "write failed (name length): "); 425 return -EINVAL; 426 } 427 428 if (nbd_write(ioc, name, name_len, errp) < 0) { 429 error_prepend(errp, "write failed (name buffer): "); 430 return -EINVAL; 431 } 432 433 if (nbd_write(ioc, desc, desc_len, errp) < 0) { 434 error_prepend(errp, "write failed (description buffer): "); 435 return -EINVAL; 436 } 437 438 return 0; 439 } 440 441 /* Process the NBD_OPT_LIST command, with a potential series of replies. 442 * Return -errno on error, 0 on success. */ 443 static int nbd_negotiate_handle_list(NBDClient *client, Error **errp) 444 { 445 NBDExport *exp; 446 assert(client->opt == NBD_OPT_LIST); 447 448 /* For each export, send a NBD_REP_SERVER reply. */ 449 QTAILQ_FOREACH(exp, &exports, next) { 450 if (nbd_negotiate_send_rep_list(client, exp, errp)) { 451 return -EINVAL; 452 } 453 } 454 /* Finish with a NBD_REP_ACK. */ 455 return nbd_negotiate_send_rep(client, NBD_REP_ACK, errp); 456 } 457 458 static void nbd_check_meta_export(NBDClient *client) 459 { 460 if (client->exp != client->export_meta.exp) { 461 client->export_meta.count = 0; 462 } 463 } 464 465 /* Send a reply to NBD_OPT_EXPORT_NAME. 466 * Return -errno on error, 0 on success. */ 467 static int nbd_negotiate_handle_export_name(NBDClient *client, bool no_zeroes, 468 Error **errp) 469 { 470 ERRP_GUARD(); 471 g_autofree char *name = NULL; 472 char buf[NBD_REPLY_EXPORT_NAME_SIZE] = ""; 473 size_t len; 474 int ret; 475 uint16_t myflags; 476 477 /* Client sends: 478 [20 .. xx] export name (length bytes) 479 Server replies: 480 [ 0 .. 7] size 481 [ 8 .. 9] export flags 482 [10 .. 133] reserved (0) [unless no_zeroes] 483 */ 484 trace_nbd_negotiate_handle_export_name(); 485 if (client->mode >= NBD_MODE_EXTENDED) { 486 error_setg(errp, "Extended headers already negotiated"); 487 return -EINVAL; 488 } 489 if (client->optlen > NBD_MAX_STRING_SIZE) { 490 error_setg(errp, "Bad length received"); 491 return -EINVAL; 492 } 493 name = g_malloc(client->optlen + 1); 494 if (nbd_read(client->ioc, name, client->optlen, "export name", errp) < 0) { 495 return -EIO; 496 } 497 name[client->optlen] = '\0'; 498 client->optlen = 0; 499 500 trace_nbd_negotiate_handle_export_name_request(name); 501 502 client->exp = nbd_export_find(name); 503 if (!client->exp) { 504 error_setg(errp, "export not found"); 505 return -EINVAL; 506 } 507 508 myflags = client->exp->nbdflags; 509 if (client->mode >= NBD_MODE_STRUCTURED) { 510 myflags |= NBD_FLAG_SEND_DF; 511 } 512 trace_nbd_negotiate_new_style_size_flags(client->exp->size, myflags); 513 stq_be_p(buf, client->exp->size); 514 stw_be_p(buf + 8, myflags); 515 len = no_zeroes ? 10 : sizeof(buf); 516 ret = nbd_write(client->ioc, buf, len, errp); 517 if (ret < 0) { 518 error_prepend(errp, "write failed: "); 519 return ret; 520 } 521 522 QTAILQ_INSERT_TAIL(&client->exp->clients, client, next); 523 blk_exp_ref(&client->exp->common); 524 nbd_check_meta_export(client); 525 526 return 0; 527 } 528 529 /* Send a single NBD_REP_INFO, with a buffer @buf of @length bytes. 530 * The buffer does NOT include the info type prefix. 531 * Return -errno on error, 0 if ready to send more. */ 532 static int nbd_negotiate_send_info(NBDClient *client, 533 uint16_t info, uint32_t length, void *buf, 534 Error **errp) 535 { 536 int rc; 537 538 trace_nbd_negotiate_send_info(info, nbd_info_lookup(info), length); 539 rc = nbd_negotiate_send_rep_len(client, NBD_REP_INFO, 540 sizeof(info) + length, errp); 541 if (rc < 0) { 542 return rc; 543 } 544 info = cpu_to_be16(info); 545 if (nbd_write(client->ioc, &info, sizeof(info), errp) < 0) { 546 return -EIO; 547 } 548 if (nbd_write(client->ioc, buf, length, errp) < 0) { 549 return -EIO; 550 } 551 return 0; 552 } 553 554 /* nbd_reject_length: Handle any unexpected payload. 555 * @fatal requests that we quit talking to the client, even if we are able 556 * to successfully send an error reply. 557 * Return: 558 * -errno transmission error occurred or @fatal was requested, errp is set 559 * 0 error message successfully sent to client, errp is not set 560 */ 561 static int nbd_reject_length(NBDClient *client, bool fatal, Error **errp) 562 { 563 int ret; 564 565 assert(client->optlen); 566 ret = nbd_opt_invalid(client, errp, "option '%s' has unexpected length", 567 nbd_opt_lookup(client->opt)); 568 if (fatal && !ret) { 569 error_setg(errp, "option '%s' has unexpected length", 570 nbd_opt_lookup(client->opt)); 571 return -EINVAL; 572 } 573 return ret; 574 } 575 576 /* Handle NBD_OPT_INFO and NBD_OPT_GO. 577 * Return -errno on error, 0 if ready for next option, and 1 to move 578 * into transmission phase. */ 579 static int nbd_negotiate_handle_info(NBDClient *client, Error **errp) 580 { 581 int rc; 582 g_autofree char *name = NULL; 583 NBDExport *exp; 584 uint16_t requests; 585 uint16_t request; 586 uint32_t namelen = 0; 587 bool sendname = false; 588 bool blocksize = false; 589 uint32_t sizes[3]; 590 char buf[sizeof(uint64_t) + sizeof(uint16_t)]; 591 uint32_t check_align = 0; 592 uint16_t myflags; 593 594 /* Client sends: 595 4 bytes: L, name length (can be 0) 596 L bytes: export name 597 2 bytes: N, number of requests (can be 0) 598 N * 2 bytes: N requests 599 */ 600 rc = nbd_opt_read_name(client, &name, &namelen, errp); 601 if (rc <= 0) { 602 return rc; 603 } 604 trace_nbd_negotiate_handle_export_name_request(name); 605 606 rc = nbd_opt_read(client, &requests, sizeof(requests), false, errp); 607 if (rc <= 0) { 608 return rc; 609 } 610 requests = be16_to_cpu(requests); 611 trace_nbd_negotiate_handle_info_requests(requests); 612 while (requests--) { 613 rc = nbd_opt_read(client, &request, sizeof(request), false, errp); 614 if (rc <= 0) { 615 return rc; 616 } 617 request = be16_to_cpu(request); 618 trace_nbd_negotiate_handle_info_request(request, 619 nbd_info_lookup(request)); 620 /* We care about NBD_INFO_NAME and NBD_INFO_BLOCK_SIZE; 621 * everything else is either a request we don't know or 622 * something we send regardless of request */ 623 switch (request) { 624 case NBD_INFO_NAME: 625 sendname = true; 626 break; 627 case NBD_INFO_BLOCK_SIZE: 628 blocksize = true; 629 break; 630 } 631 } 632 if (client->optlen) { 633 return nbd_reject_length(client, false, errp); 634 } 635 636 exp = nbd_export_find(name); 637 if (!exp) { 638 g_autofree char *sane_name = nbd_sanitize_name(name); 639 640 return nbd_negotiate_send_rep_err(client, NBD_REP_ERR_UNKNOWN, 641 errp, "export '%s' not present", 642 sane_name); 643 } 644 645 /* Don't bother sending NBD_INFO_NAME unless client requested it */ 646 if (sendname) { 647 rc = nbd_negotiate_send_info(client, NBD_INFO_NAME, namelen, name, 648 errp); 649 if (rc < 0) { 650 return rc; 651 } 652 } 653 654 /* Send NBD_INFO_DESCRIPTION only if available, regardless of 655 * client request */ 656 if (exp->description) { 657 size_t len = strlen(exp->description); 658 659 assert(len <= NBD_MAX_STRING_SIZE); 660 rc = nbd_negotiate_send_info(client, NBD_INFO_DESCRIPTION, 661 len, exp->description, errp); 662 if (rc < 0) { 663 return rc; 664 } 665 } 666 667 /* Send NBD_INFO_BLOCK_SIZE always, but tweak the minimum size 668 * according to whether the client requested it, and according to 669 * whether this is OPT_INFO or OPT_GO. */ 670 /* minimum - 1 for back-compat, or actual if client will obey it. */ 671 if (client->opt == NBD_OPT_INFO || blocksize) { 672 check_align = sizes[0] = blk_get_request_alignment(exp->common.blk); 673 } else { 674 sizes[0] = 1; 675 } 676 assert(sizes[0] <= NBD_MAX_BUFFER_SIZE); 677 /* preferred - Hard-code to 4096 for now. 678 * TODO: is blk_bs(blk)->bl.opt_transfer appropriate? */ 679 sizes[1] = MAX(4096, sizes[0]); 680 /* maximum - At most 32M, but smaller as appropriate. */ 681 sizes[2] = MIN(blk_get_max_transfer(exp->common.blk), NBD_MAX_BUFFER_SIZE); 682 trace_nbd_negotiate_handle_info_block_size(sizes[0], sizes[1], sizes[2]); 683 sizes[0] = cpu_to_be32(sizes[0]); 684 sizes[1] = cpu_to_be32(sizes[1]); 685 sizes[2] = cpu_to_be32(sizes[2]); 686 rc = nbd_negotiate_send_info(client, NBD_INFO_BLOCK_SIZE, 687 sizeof(sizes), sizes, errp); 688 if (rc < 0) { 689 return rc; 690 } 691 692 /* Send NBD_INFO_EXPORT always */ 693 myflags = exp->nbdflags; 694 if (client->mode >= NBD_MODE_STRUCTURED) { 695 myflags |= NBD_FLAG_SEND_DF; 696 } 697 trace_nbd_negotiate_new_style_size_flags(exp->size, myflags); 698 stq_be_p(buf, exp->size); 699 stw_be_p(buf + 8, myflags); 700 rc = nbd_negotiate_send_info(client, NBD_INFO_EXPORT, 701 sizeof(buf), buf, errp); 702 if (rc < 0) { 703 return rc; 704 } 705 706 /* 707 * If the client is just asking for NBD_OPT_INFO, but forgot to 708 * request block sizes in a situation that would impact 709 * performance, then return an error. But for NBD_OPT_GO, we 710 * tolerate all clients, regardless of alignments. 711 */ 712 if (client->opt == NBD_OPT_INFO && !blocksize && 713 blk_get_request_alignment(exp->common.blk) > 1) { 714 return nbd_negotiate_send_rep_err(client, 715 NBD_REP_ERR_BLOCK_SIZE_REQD, 716 errp, 717 "request NBD_INFO_BLOCK_SIZE to " 718 "use this export"); 719 } 720 721 /* Final reply */ 722 rc = nbd_negotiate_send_rep(client, NBD_REP_ACK, errp); 723 if (rc < 0) { 724 return rc; 725 } 726 727 if (client->opt == NBD_OPT_GO) { 728 client->exp = exp; 729 client->check_align = check_align; 730 QTAILQ_INSERT_TAIL(&client->exp->clients, client, next); 731 blk_exp_ref(&client->exp->common); 732 nbd_check_meta_export(client); 733 rc = 1; 734 } 735 return rc; 736 } 737 738 739 /* Handle NBD_OPT_STARTTLS. Return NULL to drop connection, or else the 740 * new channel for all further (now-encrypted) communication. */ 741 static QIOChannel *nbd_negotiate_handle_starttls(NBDClient *client, 742 Error **errp) 743 { 744 QIOChannel *ioc; 745 QIOChannelTLS *tioc; 746 struct NBDTLSHandshakeData data = { 0 }; 747 748 assert(client->opt == NBD_OPT_STARTTLS); 749 750 trace_nbd_negotiate_handle_starttls(); 751 ioc = client->ioc; 752 753 if (nbd_negotiate_send_rep(client, NBD_REP_ACK, errp) < 0) { 754 return NULL; 755 } 756 757 tioc = qio_channel_tls_new_server(ioc, 758 client->tlscreds, 759 client->tlsauthz, 760 errp); 761 if (!tioc) { 762 return NULL; 763 } 764 765 qio_channel_set_name(QIO_CHANNEL(tioc), "nbd-server-tls"); 766 trace_nbd_negotiate_handle_starttls_handshake(); 767 data.loop = g_main_loop_new(g_main_context_default(), FALSE); 768 qio_channel_tls_handshake(tioc, 769 nbd_tls_handshake, 770 &data, 771 NULL, 772 NULL); 773 774 if (!data.complete) { 775 g_main_loop_run(data.loop); 776 } 777 g_main_loop_unref(data.loop); 778 if (data.error) { 779 object_unref(OBJECT(tioc)); 780 error_propagate(errp, data.error); 781 return NULL; 782 } 783 784 return QIO_CHANNEL(tioc); 785 } 786 787 /* nbd_negotiate_send_meta_context 788 * 789 * Send one chunk of reply to NBD_OPT_{LIST,SET}_META_CONTEXT 790 * 791 * For NBD_OPT_LIST_META_CONTEXT @context_id is ignored, 0 is used instead. 792 */ 793 static int nbd_negotiate_send_meta_context(NBDClient *client, 794 const char *context, 795 uint32_t context_id, 796 Error **errp) 797 { 798 NBDOptionReplyMetaContext opt; 799 struct iovec iov[] = { 800 {.iov_base = &opt, .iov_len = sizeof(opt)}, 801 {.iov_base = (void *)context, .iov_len = strlen(context)} 802 }; 803 804 assert(iov[1].iov_len <= NBD_MAX_STRING_SIZE); 805 if (client->opt == NBD_OPT_LIST_META_CONTEXT) { 806 context_id = 0; 807 } 808 809 trace_nbd_negotiate_meta_query_reply(context, context_id); 810 set_be_option_rep(&opt.h, client->opt, NBD_REP_META_CONTEXT, 811 sizeof(opt) - sizeof(opt.h) + iov[1].iov_len); 812 stl_be_p(&opt.context_id, context_id); 813 814 return qio_channel_writev_all(client->ioc, iov, 2, errp) < 0 ? -EIO : 0; 815 } 816 817 /* 818 * Return true if @query matches @pattern, or if @query is empty when 819 * the @client is performing _LIST_. 820 */ 821 static bool nbd_meta_empty_or_pattern(NBDClient *client, const char *pattern, 822 const char *query) 823 { 824 if (!*query) { 825 trace_nbd_negotiate_meta_query_parse("empty"); 826 return client->opt == NBD_OPT_LIST_META_CONTEXT; 827 } 828 if (strcmp(query, pattern) == 0) { 829 trace_nbd_negotiate_meta_query_parse(pattern); 830 return true; 831 } 832 trace_nbd_negotiate_meta_query_skip("pattern not matched"); 833 return false; 834 } 835 836 /* 837 * Return true and adjust @str in place if it begins with @prefix. 838 */ 839 static bool nbd_strshift(const char **str, const char *prefix) 840 { 841 size_t len = strlen(prefix); 842 843 if (strncmp(*str, prefix, len) == 0) { 844 *str += len; 845 return true; 846 } 847 return false; 848 } 849 850 /* nbd_meta_base_query 851 * 852 * Handle queries to 'base' namespace. For now, only the base:allocation 853 * context is available. Return true if @query has been handled. 854 */ 855 static bool nbd_meta_base_query(NBDClient *client, NBDExportMetaContexts *meta, 856 const char *query) 857 { 858 if (!nbd_strshift(&query, "base:")) { 859 return false; 860 } 861 trace_nbd_negotiate_meta_query_parse("base:"); 862 863 if (nbd_meta_empty_or_pattern(client, "allocation", query)) { 864 meta->base_allocation = true; 865 } 866 return true; 867 } 868 869 /* nbd_meta_qemu_query 870 * 871 * Handle queries to 'qemu' namespace. For now, only the qemu:dirty-bitmap: 872 * and qemu:allocation-depth contexts are available. Return true if @query 873 * has been handled. 874 */ 875 static bool nbd_meta_qemu_query(NBDClient *client, NBDExportMetaContexts *meta, 876 const char *query) 877 { 878 size_t i; 879 880 if (!nbd_strshift(&query, "qemu:")) { 881 return false; 882 } 883 trace_nbd_negotiate_meta_query_parse("qemu:"); 884 885 if (!*query) { 886 if (client->opt == NBD_OPT_LIST_META_CONTEXT) { 887 meta->allocation_depth = meta->exp->allocation_depth; 888 if (meta->exp->nr_export_bitmaps) { 889 memset(meta->bitmaps, 1, meta->exp->nr_export_bitmaps); 890 } 891 } 892 trace_nbd_negotiate_meta_query_parse("empty"); 893 return true; 894 } 895 896 if (strcmp(query, "allocation-depth") == 0) { 897 trace_nbd_negotiate_meta_query_parse("allocation-depth"); 898 meta->allocation_depth = meta->exp->allocation_depth; 899 return true; 900 } 901 902 if (nbd_strshift(&query, "dirty-bitmap:")) { 903 trace_nbd_negotiate_meta_query_parse("dirty-bitmap:"); 904 if (!*query) { 905 if (client->opt == NBD_OPT_LIST_META_CONTEXT && 906 meta->exp->nr_export_bitmaps) { 907 memset(meta->bitmaps, 1, meta->exp->nr_export_bitmaps); 908 } 909 trace_nbd_negotiate_meta_query_parse("empty"); 910 return true; 911 } 912 913 for (i = 0; i < meta->exp->nr_export_bitmaps; i++) { 914 const char *bm_name; 915 916 bm_name = bdrv_dirty_bitmap_name(meta->exp->export_bitmaps[i]); 917 if (strcmp(bm_name, query) == 0) { 918 meta->bitmaps[i] = true; 919 trace_nbd_negotiate_meta_query_parse(query); 920 return true; 921 } 922 } 923 trace_nbd_negotiate_meta_query_skip("no dirty-bitmap match"); 924 return true; 925 } 926 927 trace_nbd_negotiate_meta_query_skip("unknown qemu context"); 928 return true; 929 } 930 931 /* nbd_negotiate_meta_query 932 * 933 * Parse namespace name and call corresponding function to parse body of the 934 * query. 935 * 936 * The only supported namespaces are 'base' and 'qemu'. 937 * 938 * Return -errno on I/O error, 0 if option was completely handled by 939 * sending a reply about inconsistent lengths, or 1 on success. */ 940 static int nbd_negotiate_meta_query(NBDClient *client, 941 NBDExportMetaContexts *meta, Error **errp) 942 { 943 int ret; 944 g_autofree char *query = NULL; 945 uint32_t len; 946 947 ret = nbd_opt_read(client, &len, sizeof(len), false, errp); 948 if (ret <= 0) { 949 return ret; 950 } 951 len = cpu_to_be32(len); 952 953 if (len > NBD_MAX_STRING_SIZE) { 954 trace_nbd_negotiate_meta_query_skip("length too long"); 955 return nbd_opt_skip(client, len, errp); 956 } 957 958 query = g_malloc(len + 1); 959 ret = nbd_opt_read(client, query, len, true, errp); 960 if (ret <= 0) { 961 return ret; 962 } 963 query[len] = '\0'; 964 965 if (nbd_meta_base_query(client, meta, query)) { 966 return 1; 967 } 968 if (nbd_meta_qemu_query(client, meta, query)) { 969 return 1; 970 } 971 972 trace_nbd_negotiate_meta_query_skip("unknown namespace"); 973 return 1; 974 } 975 976 /* nbd_negotiate_meta_queries 977 * Handle NBD_OPT_LIST_META_CONTEXT and NBD_OPT_SET_META_CONTEXT 978 * 979 * Return -errno on I/O error, or 0 if option was completely handled. */ 980 static int nbd_negotiate_meta_queries(NBDClient *client, 981 NBDExportMetaContexts *meta, Error **errp) 982 { 983 int ret; 984 g_autofree char *export_name = NULL; 985 /* Mark unused to work around https://bugs.llvm.org/show_bug.cgi?id=3888 */ 986 g_autofree G_GNUC_UNUSED bool *bitmaps = NULL; 987 NBDExportMetaContexts local_meta = {0}; 988 uint32_t nb_queries; 989 size_t i; 990 size_t count = 0; 991 992 if (client->opt == NBD_OPT_SET_META_CONTEXT && 993 client->mode < NBD_MODE_STRUCTURED) { 994 return nbd_opt_invalid(client, errp, 995 "request option '%s' when structured reply " 996 "is not negotiated", 997 nbd_opt_lookup(client->opt)); 998 } 999 1000 if (client->opt == NBD_OPT_LIST_META_CONTEXT) { 1001 /* Only change the caller's meta on SET. */ 1002 meta = &local_meta; 1003 } 1004 1005 g_free(meta->bitmaps); 1006 memset(meta, 0, sizeof(*meta)); 1007 1008 ret = nbd_opt_read_name(client, &export_name, NULL, errp); 1009 if (ret <= 0) { 1010 return ret; 1011 } 1012 1013 meta->exp = nbd_export_find(export_name); 1014 if (meta->exp == NULL) { 1015 g_autofree char *sane_name = nbd_sanitize_name(export_name); 1016 1017 return nbd_opt_drop(client, NBD_REP_ERR_UNKNOWN, errp, 1018 "export '%s' not present", sane_name); 1019 } 1020 meta->bitmaps = g_new0(bool, meta->exp->nr_export_bitmaps); 1021 if (client->opt == NBD_OPT_LIST_META_CONTEXT) { 1022 bitmaps = meta->bitmaps; 1023 } 1024 1025 ret = nbd_opt_read(client, &nb_queries, sizeof(nb_queries), false, errp); 1026 if (ret <= 0) { 1027 return ret; 1028 } 1029 nb_queries = cpu_to_be32(nb_queries); 1030 trace_nbd_negotiate_meta_context(nbd_opt_lookup(client->opt), 1031 export_name, nb_queries); 1032 1033 if (client->opt == NBD_OPT_LIST_META_CONTEXT && !nb_queries) { 1034 /* enable all known contexts */ 1035 meta->base_allocation = true; 1036 meta->allocation_depth = meta->exp->allocation_depth; 1037 if (meta->exp->nr_export_bitmaps) { 1038 memset(meta->bitmaps, 1, meta->exp->nr_export_bitmaps); 1039 } 1040 } else { 1041 for (i = 0; i < nb_queries; ++i) { 1042 ret = nbd_negotiate_meta_query(client, meta, errp); 1043 if (ret <= 0) { 1044 return ret; 1045 } 1046 } 1047 } 1048 1049 if (meta->base_allocation) { 1050 ret = nbd_negotiate_send_meta_context(client, "base:allocation", 1051 NBD_META_ID_BASE_ALLOCATION, 1052 errp); 1053 if (ret < 0) { 1054 return ret; 1055 } 1056 count++; 1057 } 1058 1059 if (meta->allocation_depth) { 1060 ret = nbd_negotiate_send_meta_context(client, "qemu:allocation-depth", 1061 NBD_META_ID_ALLOCATION_DEPTH, 1062 errp); 1063 if (ret < 0) { 1064 return ret; 1065 } 1066 count++; 1067 } 1068 1069 for (i = 0; i < meta->exp->nr_export_bitmaps; i++) { 1070 const char *bm_name; 1071 g_autofree char *context = NULL; 1072 1073 if (!meta->bitmaps[i]) { 1074 continue; 1075 } 1076 1077 bm_name = bdrv_dirty_bitmap_name(meta->exp->export_bitmaps[i]); 1078 context = g_strdup_printf("qemu:dirty-bitmap:%s", bm_name); 1079 1080 ret = nbd_negotiate_send_meta_context(client, context, 1081 NBD_META_ID_DIRTY_BITMAP + i, 1082 errp); 1083 if (ret < 0) { 1084 return ret; 1085 } 1086 count++; 1087 } 1088 1089 ret = nbd_negotiate_send_rep(client, NBD_REP_ACK, errp); 1090 if (ret == 0) { 1091 meta->count = count; 1092 } 1093 1094 return ret; 1095 } 1096 1097 /* nbd_negotiate_options 1098 * Process all NBD_OPT_* client option commands, during fixed newstyle 1099 * negotiation. 1100 * Return: 1101 * -errno on error, errp is set 1102 * 0 on successful negotiation, errp is not set 1103 * 1 if client sent NBD_OPT_ABORT, i.e. on valid disconnect, 1104 * errp is not set 1105 */ 1106 static int nbd_negotiate_options(NBDClient *client, Error **errp) 1107 { 1108 uint32_t flags; 1109 bool fixedNewstyle = false; 1110 bool no_zeroes = false; 1111 1112 /* Client sends: 1113 [ 0 .. 3] client flags 1114 1115 Then we loop until NBD_OPT_EXPORT_NAME or NBD_OPT_GO: 1116 [ 0 .. 7] NBD_OPTS_MAGIC 1117 [ 8 .. 11] NBD option 1118 [12 .. 15] Data length 1119 ... Rest of request 1120 1121 [ 0 .. 7] NBD_OPTS_MAGIC 1122 [ 8 .. 11] Second NBD option 1123 [12 .. 15] Data length 1124 ... Rest of request 1125 */ 1126 1127 if (nbd_read32(client->ioc, &flags, "flags", errp) < 0) { 1128 return -EIO; 1129 } 1130 client->mode = NBD_MODE_EXPORT_NAME; 1131 trace_nbd_negotiate_options_flags(flags); 1132 if (flags & NBD_FLAG_C_FIXED_NEWSTYLE) { 1133 fixedNewstyle = true; 1134 flags &= ~NBD_FLAG_C_FIXED_NEWSTYLE; 1135 client->mode = NBD_MODE_SIMPLE; 1136 } 1137 if (flags & NBD_FLAG_C_NO_ZEROES) { 1138 no_zeroes = true; 1139 flags &= ~NBD_FLAG_C_NO_ZEROES; 1140 } 1141 if (flags != 0) { 1142 error_setg(errp, "Unknown client flags 0x%" PRIx32 " received", flags); 1143 return -EINVAL; 1144 } 1145 1146 while (1) { 1147 int ret; 1148 uint32_t option, length; 1149 uint64_t magic; 1150 1151 if (nbd_read64(client->ioc, &magic, "opts magic", errp) < 0) { 1152 return -EINVAL; 1153 } 1154 trace_nbd_negotiate_options_check_magic(magic); 1155 if (magic != NBD_OPTS_MAGIC) { 1156 error_setg(errp, "Bad magic received"); 1157 return -EINVAL; 1158 } 1159 1160 if (nbd_read32(client->ioc, &option, "option", errp) < 0) { 1161 return -EINVAL; 1162 } 1163 client->opt = option; 1164 1165 if (nbd_read32(client->ioc, &length, "option length", errp) < 0) { 1166 return -EINVAL; 1167 } 1168 assert(!client->optlen); 1169 client->optlen = length; 1170 1171 if (length > NBD_MAX_BUFFER_SIZE) { 1172 error_setg(errp, "len (%" PRIu32 ") is larger than max len (%u)", 1173 length, NBD_MAX_BUFFER_SIZE); 1174 return -EINVAL; 1175 } 1176 1177 trace_nbd_negotiate_options_check_option(option, 1178 nbd_opt_lookup(option)); 1179 if (client->tlscreds && 1180 client->ioc == (QIOChannel *)client->sioc) { 1181 QIOChannel *tioc; 1182 if (!fixedNewstyle) { 1183 error_setg(errp, "Unsupported option 0x%" PRIx32, option); 1184 return -EINVAL; 1185 } 1186 switch (option) { 1187 case NBD_OPT_STARTTLS: 1188 if (length) { 1189 /* Unconditionally drop the connection if the client 1190 * can't start a TLS negotiation correctly */ 1191 return nbd_reject_length(client, true, errp); 1192 } 1193 tioc = nbd_negotiate_handle_starttls(client, errp); 1194 if (!tioc) { 1195 return -EIO; 1196 } 1197 ret = 0; 1198 object_unref(OBJECT(client->ioc)); 1199 client->ioc = tioc; 1200 break; 1201 1202 case NBD_OPT_EXPORT_NAME: 1203 /* No way to return an error to client, so drop connection */ 1204 error_setg(errp, "Option 0x%x not permitted before TLS", 1205 option); 1206 return -EINVAL; 1207 1208 default: 1209 /* Let the client keep trying, unless they asked to 1210 * quit. Always try to give an error back to the 1211 * client; but when replying to OPT_ABORT, be aware 1212 * that the client may hang up before receiving the 1213 * error, in which case we are fine ignoring the 1214 * resulting EPIPE. */ 1215 ret = nbd_opt_drop(client, NBD_REP_ERR_TLS_REQD, 1216 option == NBD_OPT_ABORT ? NULL : errp, 1217 "Option 0x%" PRIx32 1218 " not permitted before TLS", option); 1219 if (option == NBD_OPT_ABORT) { 1220 return 1; 1221 } 1222 break; 1223 } 1224 } else if (fixedNewstyle) { 1225 switch (option) { 1226 case NBD_OPT_LIST: 1227 if (length) { 1228 ret = nbd_reject_length(client, false, errp); 1229 } else { 1230 ret = nbd_negotiate_handle_list(client, errp); 1231 } 1232 break; 1233 1234 case NBD_OPT_ABORT: 1235 /* NBD spec says we must try to reply before 1236 * disconnecting, but that we must also tolerate 1237 * guests that don't wait for our reply. */ 1238 nbd_negotiate_send_rep(client, NBD_REP_ACK, NULL); 1239 return 1; 1240 1241 case NBD_OPT_EXPORT_NAME: 1242 return nbd_negotiate_handle_export_name(client, no_zeroes, 1243 errp); 1244 1245 case NBD_OPT_INFO: 1246 case NBD_OPT_GO: 1247 ret = nbd_negotiate_handle_info(client, errp); 1248 if (ret == 1) { 1249 assert(option == NBD_OPT_GO); 1250 return 0; 1251 } 1252 break; 1253 1254 case NBD_OPT_STARTTLS: 1255 if (length) { 1256 ret = nbd_reject_length(client, false, errp); 1257 } else if (client->tlscreds) { 1258 ret = nbd_negotiate_send_rep_err(client, 1259 NBD_REP_ERR_INVALID, errp, 1260 "TLS already enabled"); 1261 } else { 1262 ret = nbd_negotiate_send_rep_err(client, 1263 NBD_REP_ERR_POLICY, errp, 1264 "TLS not configured"); 1265 } 1266 break; 1267 1268 case NBD_OPT_STRUCTURED_REPLY: 1269 if (length) { 1270 ret = nbd_reject_length(client, false, errp); 1271 } else if (client->mode >= NBD_MODE_EXTENDED) { 1272 ret = nbd_negotiate_send_rep_err( 1273 client, NBD_REP_ERR_EXT_HEADER_REQD, errp, 1274 "extended headers already negotiated"); 1275 } else if (client->mode >= NBD_MODE_STRUCTURED) { 1276 ret = nbd_negotiate_send_rep_err( 1277 client, NBD_REP_ERR_INVALID, errp, 1278 "structured reply already negotiated"); 1279 } else { 1280 ret = nbd_negotiate_send_rep(client, NBD_REP_ACK, errp); 1281 client->mode = NBD_MODE_STRUCTURED; 1282 } 1283 break; 1284 1285 case NBD_OPT_LIST_META_CONTEXT: 1286 case NBD_OPT_SET_META_CONTEXT: 1287 ret = nbd_negotiate_meta_queries(client, &client->export_meta, 1288 errp); 1289 break; 1290 1291 case NBD_OPT_EXTENDED_HEADERS: 1292 if (length) { 1293 ret = nbd_reject_length(client, false, errp); 1294 } else if (client->mode >= NBD_MODE_EXTENDED) { 1295 ret = nbd_negotiate_send_rep_err( 1296 client, NBD_REP_ERR_INVALID, errp, 1297 "extended headers already negotiated"); 1298 } else { 1299 ret = nbd_negotiate_send_rep(client, NBD_REP_ACK, errp); 1300 client->mode = NBD_MODE_EXTENDED; 1301 } 1302 break; 1303 1304 default: 1305 ret = nbd_opt_drop(client, NBD_REP_ERR_UNSUP, errp, 1306 "Unsupported option %" PRIu32 " (%s)", 1307 option, nbd_opt_lookup(option)); 1308 break; 1309 } 1310 } else { 1311 /* 1312 * If broken new-style we should drop the connection 1313 * for anything except NBD_OPT_EXPORT_NAME 1314 */ 1315 switch (option) { 1316 case NBD_OPT_EXPORT_NAME: 1317 return nbd_negotiate_handle_export_name(client, no_zeroes, 1318 errp); 1319 1320 default: 1321 error_setg(errp, "Unsupported option %" PRIu32 " (%s)", 1322 option, nbd_opt_lookup(option)); 1323 return -EINVAL; 1324 } 1325 } 1326 if (ret < 0) { 1327 return ret; 1328 } 1329 } 1330 } 1331 1332 /* nbd_negotiate 1333 * Return: 1334 * -errno on error, errp is set 1335 * 0 on successful negotiation, errp is not set 1336 * 1 if client sent NBD_OPT_ABORT, i.e. on valid disconnect, 1337 * errp is not set 1338 */ 1339 static coroutine_fn int nbd_negotiate(NBDClient *client, Error **errp) 1340 { 1341 ERRP_GUARD(); 1342 char buf[NBD_OLDSTYLE_NEGOTIATE_SIZE] = ""; 1343 int ret; 1344 1345 /* Old style negotiation header, no room for options 1346 [ 0 .. 7] passwd ("NBDMAGIC") 1347 [ 8 .. 15] magic (NBD_CLIENT_MAGIC) 1348 [16 .. 23] size 1349 [24 .. 27] export flags (zero-extended) 1350 [28 .. 151] reserved (0) 1351 1352 New style negotiation header, client can send options 1353 [ 0 .. 7] passwd ("NBDMAGIC") 1354 [ 8 .. 15] magic (NBD_OPTS_MAGIC) 1355 [16 .. 17] server flags (0) 1356 ....options sent, ending in NBD_OPT_EXPORT_NAME or NBD_OPT_GO.... 1357 */ 1358 1359 qio_channel_set_blocking(client->ioc, false, NULL); 1360 qio_channel_set_follow_coroutine_ctx(client->ioc, true); 1361 1362 trace_nbd_negotiate_begin(); 1363 memcpy(buf, "NBDMAGIC", 8); 1364 1365 stq_be_p(buf + 8, NBD_OPTS_MAGIC); 1366 stw_be_p(buf + 16, NBD_FLAG_FIXED_NEWSTYLE | NBD_FLAG_NO_ZEROES); 1367 1368 if (nbd_write(client->ioc, buf, 18, errp) < 0) { 1369 error_prepend(errp, "write failed: "); 1370 return -EINVAL; 1371 } 1372 ret = nbd_negotiate_options(client, errp); 1373 if (ret != 0) { 1374 if (ret < 0) { 1375 error_prepend(errp, "option negotiation failed: "); 1376 } 1377 return ret; 1378 } 1379 1380 assert(!client->optlen); 1381 trace_nbd_negotiate_success(); 1382 1383 return 0; 1384 } 1385 1386 /* nbd_read_eof 1387 * Tries to read @size bytes from @ioc. This is a local implementation of 1388 * qio_channel_readv_all_eof. We have it here because we need it to be 1389 * interruptible and to know when the coroutine is yielding. 1390 * Returns 1 on success 1391 * 0 on eof, when no data was read (errp is not set) 1392 * negative errno on failure (errp is set) 1393 */ 1394 static inline int coroutine_fn 1395 nbd_read_eof(NBDClient *client, void *buffer, size_t size, Error **errp) 1396 { 1397 bool partial = false; 1398 1399 assert(size); 1400 while (size > 0) { 1401 struct iovec iov = { .iov_base = buffer, .iov_len = size }; 1402 ssize_t len; 1403 1404 len = qio_channel_readv(client->ioc, &iov, 1, errp); 1405 if (len == QIO_CHANNEL_ERR_BLOCK) { 1406 client->read_yielding = true; 1407 qio_channel_yield(client->ioc, G_IO_IN); 1408 client->read_yielding = false; 1409 if (client->quiescing) { 1410 return -EAGAIN; 1411 } 1412 continue; 1413 } else if (len < 0) { 1414 return -EIO; 1415 } else if (len == 0) { 1416 if (partial) { 1417 error_setg(errp, 1418 "Unexpected end-of-file before all bytes were read"); 1419 return -EIO; 1420 } else { 1421 return 0; 1422 } 1423 } 1424 1425 partial = true; 1426 size -= len; 1427 buffer = (uint8_t *) buffer + len; 1428 } 1429 return 1; 1430 } 1431 1432 static int coroutine_fn nbd_receive_request(NBDClient *client, NBDRequest *request, 1433 Error **errp) 1434 { 1435 uint8_t buf[NBD_EXTENDED_REQUEST_SIZE]; 1436 uint32_t magic, expect; 1437 int ret; 1438 size_t size = client->mode >= NBD_MODE_EXTENDED ? 1439 NBD_EXTENDED_REQUEST_SIZE : NBD_REQUEST_SIZE; 1440 1441 ret = nbd_read_eof(client, buf, size, errp); 1442 if (ret < 0) { 1443 return ret; 1444 } 1445 if (ret == 0) { 1446 return -EIO; 1447 } 1448 1449 /* 1450 * Compact request 1451 * [ 0 .. 3] magic (NBD_REQUEST_MAGIC) 1452 * [ 4 .. 5] flags (NBD_CMD_FLAG_FUA, ...) 1453 * [ 6 .. 7] type (NBD_CMD_READ, ...) 1454 * [ 8 .. 15] cookie 1455 * [16 .. 23] from 1456 * [24 .. 27] len 1457 * Extended request 1458 * [ 0 .. 3] magic (NBD_EXTENDED_REQUEST_MAGIC) 1459 * [ 4 .. 5] flags (NBD_CMD_FLAG_FUA, NBD_CMD_FLAG_PAYLOAD_LEN, ...) 1460 * [ 6 .. 7] type (NBD_CMD_READ, ...) 1461 * [ 8 .. 15] cookie 1462 * [16 .. 23] from 1463 * [24 .. 31] len 1464 */ 1465 1466 magic = ldl_be_p(buf); 1467 request->flags = lduw_be_p(buf + 4); 1468 request->type = lduw_be_p(buf + 6); 1469 request->cookie = ldq_be_p(buf + 8); 1470 request->from = ldq_be_p(buf + 16); 1471 if (client->mode >= NBD_MODE_EXTENDED) { 1472 request->len = ldq_be_p(buf + 24); 1473 expect = NBD_EXTENDED_REQUEST_MAGIC; 1474 } else { 1475 request->len = (uint32_t)ldl_be_p(buf + 24); /* widen 32 to 64 bits */ 1476 expect = NBD_REQUEST_MAGIC; 1477 } 1478 1479 trace_nbd_receive_request(magic, request->flags, request->type, 1480 request->from, request->len); 1481 1482 if (magic != expect) { 1483 error_setg(errp, "invalid magic (got 0x%" PRIx32 ", expected 0x%" 1484 PRIx32 ")", magic, expect); 1485 return -EINVAL; 1486 } 1487 return 0; 1488 } 1489 1490 #define MAX_NBD_REQUESTS 16 1491 1492 void nbd_client_get(NBDClient *client) 1493 { 1494 client->refcount++; 1495 } 1496 1497 void nbd_client_put(NBDClient *client) 1498 { 1499 if (--client->refcount == 0) { 1500 /* The last reference should be dropped by client->close, 1501 * which is called by client_close. 1502 */ 1503 assert(client->closing); 1504 1505 object_unref(OBJECT(client->sioc)); 1506 object_unref(OBJECT(client->ioc)); 1507 if (client->tlscreds) { 1508 object_unref(OBJECT(client->tlscreds)); 1509 } 1510 g_free(client->tlsauthz); 1511 if (client->exp) { 1512 QTAILQ_REMOVE(&client->exp->clients, client, next); 1513 blk_exp_unref(&client->exp->common); 1514 } 1515 g_free(client->export_meta.bitmaps); 1516 g_free(client); 1517 } 1518 } 1519 1520 static void client_close(NBDClient *client, bool negotiated) 1521 { 1522 if (client->closing) { 1523 return; 1524 } 1525 1526 client->closing = true; 1527 1528 /* Force requests to finish. They will drop their own references, 1529 * then we'll close the socket and free the NBDClient. 1530 */ 1531 qio_channel_shutdown(client->ioc, QIO_CHANNEL_SHUTDOWN_BOTH, 1532 NULL); 1533 1534 /* Also tell the client, so that they release their reference. */ 1535 if (client->close_fn) { 1536 client->close_fn(client, negotiated); 1537 } 1538 } 1539 1540 static NBDRequestData *nbd_request_get(NBDClient *client) 1541 { 1542 NBDRequestData *req; 1543 1544 assert(client->nb_requests <= MAX_NBD_REQUESTS - 1); 1545 client->nb_requests++; 1546 1547 req = g_new0(NBDRequestData, 1); 1548 nbd_client_get(client); 1549 req->client = client; 1550 return req; 1551 } 1552 1553 static void nbd_request_put(NBDRequestData *req) 1554 { 1555 NBDClient *client = req->client; 1556 1557 if (req->data) { 1558 qemu_vfree(req->data); 1559 } 1560 g_free(req); 1561 1562 client->nb_requests--; 1563 1564 if (client->quiescing && client->nb_requests == 0) { 1565 aio_wait_kick(); 1566 } 1567 1568 nbd_client_receive_next_request(client); 1569 1570 nbd_client_put(client); 1571 } 1572 1573 static void blk_aio_attached(AioContext *ctx, void *opaque) 1574 { 1575 NBDExport *exp = opaque; 1576 NBDClient *client; 1577 1578 trace_nbd_blk_aio_attached(exp->name, ctx); 1579 1580 exp->common.ctx = ctx; 1581 1582 QTAILQ_FOREACH(client, &exp->clients, next) { 1583 assert(client->nb_requests == 0); 1584 assert(client->recv_coroutine == NULL); 1585 assert(client->send_coroutine == NULL); 1586 } 1587 } 1588 1589 static void blk_aio_detach(void *opaque) 1590 { 1591 NBDExport *exp = opaque; 1592 1593 trace_nbd_blk_aio_detach(exp->name, exp->common.ctx); 1594 1595 exp->common.ctx = NULL; 1596 } 1597 1598 static void nbd_drained_begin(void *opaque) 1599 { 1600 NBDExport *exp = opaque; 1601 NBDClient *client; 1602 1603 QTAILQ_FOREACH(client, &exp->clients, next) { 1604 client->quiescing = true; 1605 } 1606 } 1607 1608 static void nbd_drained_end(void *opaque) 1609 { 1610 NBDExport *exp = opaque; 1611 NBDClient *client; 1612 1613 QTAILQ_FOREACH(client, &exp->clients, next) { 1614 client->quiescing = false; 1615 nbd_client_receive_next_request(client); 1616 } 1617 } 1618 1619 static bool nbd_drained_poll(void *opaque) 1620 { 1621 NBDExport *exp = opaque; 1622 NBDClient *client; 1623 1624 QTAILQ_FOREACH(client, &exp->clients, next) { 1625 if (client->nb_requests != 0) { 1626 /* 1627 * If there's a coroutine waiting for a request on nbd_read_eof() 1628 * enter it here so we don't depend on the client to wake it up. 1629 */ 1630 if (client->recv_coroutine != NULL && client->read_yielding) { 1631 qio_channel_wake_read(client->ioc); 1632 } 1633 1634 return true; 1635 } 1636 } 1637 1638 return false; 1639 } 1640 1641 static void nbd_eject_notifier(Notifier *n, void *data) 1642 { 1643 NBDExport *exp = container_of(n, NBDExport, eject_notifier); 1644 1645 blk_exp_request_shutdown(&exp->common); 1646 } 1647 1648 void nbd_export_set_on_eject_blk(BlockExport *exp, BlockBackend *blk) 1649 { 1650 NBDExport *nbd_exp = container_of(exp, NBDExport, common); 1651 assert(exp->drv == &blk_exp_nbd); 1652 assert(nbd_exp->eject_notifier_blk == NULL); 1653 1654 blk_ref(blk); 1655 nbd_exp->eject_notifier_blk = blk; 1656 nbd_exp->eject_notifier.notify = nbd_eject_notifier; 1657 blk_add_remove_bs_notifier(blk, &nbd_exp->eject_notifier); 1658 } 1659 1660 static const BlockDevOps nbd_block_ops = { 1661 .drained_begin = nbd_drained_begin, 1662 .drained_end = nbd_drained_end, 1663 .drained_poll = nbd_drained_poll, 1664 }; 1665 1666 static int nbd_export_create(BlockExport *blk_exp, BlockExportOptions *exp_args, 1667 Error **errp) 1668 { 1669 NBDExport *exp = container_of(blk_exp, NBDExport, common); 1670 BlockExportOptionsNbd *arg = &exp_args->u.nbd; 1671 const char *name = arg->name ?: exp_args->node_name; 1672 BlockBackend *blk = blk_exp->blk; 1673 int64_t size; 1674 uint64_t perm, shared_perm; 1675 bool readonly = !exp_args->writable; 1676 BlockDirtyBitmapOrStrList *bitmaps; 1677 size_t i; 1678 int ret; 1679 1680 assert(exp_args->type == BLOCK_EXPORT_TYPE_NBD); 1681 1682 if (!nbd_server_is_running()) { 1683 error_setg(errp, "NBD server not running"); 1684 return -EINVAL; 1685 } 1686 1687 if (strlen(name) > NBD_MAX_STRING_SIZE) { 1688 error_setg(errp, "export name '%s' too long", name); 1689 return -EINVAL; 1690 } 1691 1692 if (arg->description && strlen(arg->description) > NBD_MAX_STRING_SIZE) { 1693 error_setg(errp, "description '%s' too long", arg->description); 1694 return -EINVAL; 1695 } 1696 1697 if (nbd_export_find(name)) { 1698 error_setg(errp, "NBD server already has export named '%s'", name); 1699 return -EEXIST; 1700 } 1701 1702 size = blk_getlength(blk); 1703 if (size < 0) { 1704 error_setg_errno(errp, -size, 1705 "Failed to determine the NBD export's length"); 1706 return size; 1707 } 1708 1709 /* Don't allow resize while the NBD server is running, otherwise we don't 1710 * care what happens with the node. */ 1711 blk_get_perm(blk, &perm, &shared_perm); 1712 ret = blk_set_perm(blk, perm, shared_perm & ~BLK_PERM_RESIZE, errp); 1713 if (ret < 0) { 1714 return ret; 1715 } 1716 1717 QTAILQ_INIT(&exp->clients); 1718 exp->name = g_strdup(name); 1719 exp->description = g_strdup(arg->description); 1720 exp->nbdflags = (NBD_FLAG_HAS_FLAGS | NBD_FLAG_SEND_FLUSH | 1721 NBD_FLAG_SEND_FUA | NBD_FLAG_SEND_CACHE); 1722 1723 if (nbd_server_max_connections() != 1) { 1724 exp->nbdflags |= NBD_FLAG_CAN_MULTI_CONN; 1725 } 1726 if (readonly) { 1727 exp->nbdflags |= NBD_FLAG_READ_ONLY; 1728 } else { 1729 exp->nbdflags |= (NBD_FLAG_SEND_TRIM | NBD_FLAG_SEND_WRITE_ZEROES | 1730 NBD_FLAG_SEND_FAST_ZERO); 1731 } 1732 exp->size = QEMU_ALIGN_DOWN(size, BDRV_SECTOR_SIZE); 1733 1734 for (bitmaps = arg->bitmaps; bitmaps; bitmaps = bitmaps->next) { 1735 exp->nr_export_bitmaps++; 1736 } 1737 exp->export_bitmaps = g_new0(BdrvDirtyBitmap *, exp->nr_export_bitmaps); 1738 for (i = 0, bitmaps = arg->bitmaps; bitmaps; 1739 i++, bitmaps = bitmaps->next) 1740 { 1741 const char *bitmap; 1742 BlockDriverState *bs = blk_bs(blk); 1743 BdrvDirtyBitmap *bm = NULL; 1744 1745 switch (bitmaps->value->type) { 1746 case QTYPE_QSTRING: 1747 bitmap = bitmaps->value->u.local; 1748 while (bs) { 1749 bm = bdrv_find_dirty_bitmap(bs, bitmap); 1750 if (bm != NULL) { 1751 break; 1752 } 1753 1754 bs = bdrv_filter_or_cow_bs(bs); 1755 } 1756 1757 if (bm == NULL) { 1758 ret = -ENOENT; 1759 error_setg(errp, "Bitmap '%s' is not found", 1760 bitmaps->value->u.local); 1761 goto fail; 1762 } 1763 1764 if (readonly && bdrv_is_writable(bs) && 1765 bdrv_dirty_bitmap_enabled(bm)) { 1766 ret = -EINVAL; 1767 error_setg(errp, "Enabled bitmap '%s' incompatible with " 1768 "readonly export", bitmap); 1769 goto fail; 1770 } 1771 break; 1772 case QTYPE_QDICT: 1773 bitmap = bitmaps->value->u.external.name; 1774 bm = block_dirty_bitmap_lookup(bitmaps->value->u.external.node, 1775 bitmap, NULL, errp); 1776 if (!bm) { 1777 ret = -ENOENT; 1778 goto fail; 1779 } 1780 break; 1781 default: 1782 abort(); 1783 } 1784 1785 assert(bm); 1786 1787 if (bdrv_dirty_bitmap_check(bm, BDRV_BITMAP_ALLOW_RO, errp)) { 1788 ret = -EINVAL; 1789 goto fail; 1790 } 1791 1792 exp->export_bitmaps[i] = bm; 1793 assert(strlen(bitmap) <= BDRV_BITMAP_MAX_NAME_SIZE); 1794 } 1795 1796 /* Mark bitmaps busy in a separate loop, to simplify roll-back concerns. */ 1797 for (i = 0; i < exp->nr_export_bitmaps; i++) { 1798 bdrv_dirty_bitmap_set_busy(exp->export_bitmaps[i], true); 1799 } 1800 1801 exp->allocation_depth = arg->allocation_depth; 1802 1803 /* 1804 * We need to inhibit request queuing in the block layer to ensure we can 1805 * be properly quiesced when entering a drained section, as our coroutines 1806 * servicing pending requests might enter blk_pread(). 1807 */ 1808 blk_set_disable_request_queuing(blk, true); 1809 1810 blk_add_aio_context_notifier(blk, blk_aio_attached, blk_aio_detach, exp); 1811 1812 blk_set_dev_ops(blk, &nbd_block_ops, exp); 1813 1814 QTAILQ_INSERT_TAIL(&exports, exp, next); 1815 1816 return 0; 1817 1818 fail: 1819 g_free(exp->export_bitmaps); 1820 g_free(exp->name); 1821 g_free(exp->description); 1822 return ret; 1823 } 1824 1825 NBDExport *nbd_export_find(const char *name) 1826 { 1827 NBDExport *exp; 1828 QTAILQ_FOREACH(exp, &exports, next) { 1829 if (strcmp(name, exp->name) == 0) { 1830 return exp; 1831 } 1832 } 1833 1834 return NULL; 1835 } 1836 1837 AioContext * 1838 nbd_export_aio_context(NBDExport *exp) 1839 { 1840 return exp->common.ctx; 1841 } 1842 1843 static void nbd_export_request_shutdown(BlockExport *blk_exp) 1844 { 1845 NBDExport *exp = container_of(blk_exp, NBDExport, common); 1846 NBDClient *client, *next; 1847 1848 blk_exp_ref(&exp->common); 1849 /* 1850 * TODO: Should we expand QMP NbdServerRemoveNode enum to allow a 1851 * close mode that stops advertising the export to new clients but 1852 * still permits existing clients to run to completion? Because of 1853 * that possibility, nbd_export_close() can be called more than 1854 * once on an export. 1855 */ 1856 QTAILQ_FOREACH_SAFE(client, &exp->clients, next, next) { 1857 client_close(client, true); 1858 } 1859 if (exp->name) { 1860 g_free(exp->name); 1861 exp->name = NULL; 1862 QTAILQ_REMOVE(&exports, exp, next); 1863 } 1864 blk_exp_unref(&exp->common); 1865 } 1866 1867 static void nbd_export_delete(BlockExport *blk_exp) 1868 { 1869 size_t i; 1870 NBDExport *exp = container_of(blk_exp, NBDExport, common); 1871 1872 assert(exp->name == NULL); 1873 assert(QTAILQ_EMPTY(&exp->clients)); 1874 1875 g_free(exp->description); 1876 exp->description = NULL; 1877 1878 if (exp->eject_notifier_blk) { 1879 notifier_remove(&exp->eject_notifier); 1880 blk_unref(exp->eject_notifier_blk); 1881 } 1882 blk_remove_aio_context_notifier(exp->common.blk, blk_aio_attached, 1883 blk_aio_detach, exp); 1884 blk_set_disable_request_queuing(exp->common.blk, false); 1885 1886 for (i = 0; i < exp->nr_export_bitmaps; i++) { 1887 bdrv_dirty_bitmap_set_busy(exp->export_bitmaps[i], false); 1888 } 1889 } 1890 1891 const BlockExportDriver blk_exp_nbd = { 1892 .type = BLOCK_EXPORT_TYPE_NBD, 1893 .instance_size = sizeof(NBDExport), 1894 .create = nbd_export_create, 1895 .delete = nbd_export_delete, 1896 .request_shutdown = nbd_export_request_shutdown, 1897 }; 1898 1899 static int coroutine_fn nbd_co_send_iov(NBDClient *client, struct iovec *iov, 1900 unsigned niov, Error **errp) 1901 { 1902 int ret; 1903 1904 g_assert(qemu_in_coroutine()); 1905 qemu_co_mutex_lock(&client->send_lock); 1906 client->send_coroutine = qemu_coroutine_self(); 1907 1908 ret = qio_channel_writev_all(client->ioc, iov, niov, errp) < 0 ? -EIO : 0; 1909 1910 client->send_coroutine = NULL; 1911 qemu_co_mutex_unlock(&client->send_lock); 1912 1913 return ret; 1914 } 1915 1916 static inline void set_be_simple_reply(NBDSimpleReply *reply, uint64_t error, 1917 uint64_t cookie) 1918 { 1919 stl_be_p(&reply->magic, NBD_SIMPLE_REPLY_MAGIC); 1920 stl_be_p(&reply->error, error); 1921 stq_be_p(&reply->cookie, cookie); 1922 } 1923 1924 static int coroutine_fn nbd_co_send_simple_reply(NBDClient *client, 1925 NBDRequest *request, 1926 uint32_t error, 1927 void *data, 1928 uint64_t len, 1929 Error **errp) 1930 { 1931 NBDSimpleReply reply; 1932 int nbd_err = system_errno_to_nbd_errno(error); 1933 struct iovec iov[] = { 1934 {.iov_base = &reply, .iov_len = sizeof(reply)}, 1935 {.iov_base = data, .iov_len = len} 1936 }; 1937 1938 assert(!len || !nbd_err); 1939 assert(len <= NBD_MAX_BUFFER_SIZE); 1940 assert(client->mode < NBD_MODE_STRUCTURED || 1941 (client->mode == NBD_MODE_STRUCTURED && 1942 request->type != NBD_CMD_READ)); 1943 trace_nbd_co_send_simple_reply(request->cookie, nbd_err, 1944 nbd_err_lookup(nbd_err), len); 1945 set_be_simple_reply(&reply, nbd_err, request->cookie); 1946 1947 return nbd_co_send_iov(client, iov, 2, errp); 1948 } 1949 1950 /* 1951 * Prepare the header of a reply chunk for network transmission. 1952 * 1953 * On input, @iov is partially initialized: iov[0].iov_base must point 1954 * to an uninitialized NBDReply, while the remaining @niov elements 1955 * (if any) must be ready for transmission. This function then 1956 * populates iov[0] for transmission. 1957 */ 1958 static inline void set_be_chunk(NBDClient *client, struct iovec *iov, 1959 size_t niov, uint16_t flags, uint16_t type, 1960 NBDRequest *request) 1961 { 1962 size_t i, length = 0; 1963 1964 for (i = 1; i < niov; i++) { 1965 length += iov[i].iov_len; 1966 } 1967 assert(length <= NBD_MAX_BUFFER_SIZE + sizeof(NBDStructuredReadData)); 1968 1969 if (client->mode >= NBD_MODE_EXTENDED) { 1970 NBDExtendedReplyChunk *chunk = iov->iov_base; 1971 1972 iov[0].iov_len = sizeof(*chunk); 1973 stl_be_p(&chunk->magic, NBD_EXTENDED_REPLY_MAGIC); 1974 stw_be_p(&chunk->flags, flags); 1975 stw_be_p(&chunk->type, type); 1976 stq_be_p(&chunk->cookie, request->cookie); 1977 stq_be_p(&chunk->offset, request->from); 1978 stq_be_p(&chunk->length, length); 1979 } else { 1980 NBDStructuredReplyChunk *chunk = iov->iov_base; 1981 1982 iov[0].iov_len = sizeof(*chunk); 1983 stl_be_p(&chunk->magic, NBD_STRUCTURED_REPLY_MAGIC); 1984 stw_be_p(&chunk->flags, flags); 1985 stw_be_p(&chunk->type, type); 1986 stq_be_p(&chunk->cookie, request->cookie); 1987 stl_be_p(&chunk->length, length); 1988 } 1989 } 1990 1991 static int coroutine_fn nbd_co_send_chunk_done(NBDClient *client, 1992 NBDRequest *request, 1993 Error **errp) 1994 { 1995 NBDReply hdr; 1996 struct iovec iov[] = { 1997 {.iov_base = &hdr}, 1998 }; 1999 2000 trace_nbd_co_send_chunk_done(request->cookie); 2001 set_be_chunk(client, iov, 1, NBD_REPLY_FLAG_DONE, 2002 NBD_REPLY_TYPE_NONE, request); 2003 return nbd_co_send_iov(client, iov, 1, errp); 2004 } 2005 2006 static int coroutine_fn nbd_co_send_chunk_read(NBDClient *client, 2007 NBDRequest *request, 2008 uint64_t offset, 2009 void *data, 2010 uint64_t size, 2011 bool final, 2012 Error **errp) 2013 { 2014 NBDReply hdr; 2015 NBDStructuredReadData chunk; 2016 struct iovec iov[] = { 2017 {.iov_base = &hdr}, 2018 {.iov_base = &chunk, .iov_len = sizeof(chunk)}, 2019 {.iov_base = data, .iov_len = size} 2020 }; 2021 2022 assert(size && size <= NBD_MAX_BUFFER_SIZE); 2023 trace_nbd_co_send_chunk_read(request->cookie, offset, data, size); 2024 set_be_chunk(client, iov, 3, final ? NBD_REPLY_FLAG_DONE : 0, 2025 NBD_REPLY_TYPE_OFFSET_DATA, request); 2026 stq_be_p(&chunk.offset, offset); 2027 2028 return nbd_co_send_iov(client, iov, 3, errp); 2029 } 2030 2031 static int coroutine_fn nbd_co_send_chunk_error(NBDClient *client, 2032 NBDRequest *request, 2033 uint32_t error, 2034 const char *msg, 2035 Error **errp) 2036 { 2037 NBDReply hdr; 2038 NBDStructuredError chunk; 2039 int nbd_err = system_errno_to_nbd_errno(error); 2040 struct iovec iov[] = { 2041 {.iov_base = &hdr}, 2042 {.iov_base = &chunk, .iov_len = sizeof(chunk)}, 2043 {.iov_base = (char *)msg, .iov_len = msg ? strlen(msg) : 0}, 2044 }; 2045 2046 assert(nbd_err); 2047 trace_nbd_co_send_chunk_error(request->cookie, nbd_err, 2048 nbd_err_lookup(nbd_err), msg ? msg : ""); 2049 set_be_chunk(client, iov, 3, NBD_REPLY_FLAG_DONE, 2050 NBD_REPLY_TYPE_ERROR, request); 2051 stl_be_p(&chunk.error, nbd_err); 2052 stw_be_p(&chunk.message_length, iov[2].iov_len); 2053 2054 return nbd_co_send_iov(client, iov, 3, errp); 2055 } 2056 2057 /* Do a sparse read and send the structured reply to the client. 2058 * Returns -errno if sending fails. blk_co_block_status_above() failure is 2059 * reported to the client, at which point this function succeeds. 2060 */ 2061 static int coroutine_fn nbd_co_send_sparse_read(NBDClient *client, 2062 NBDRequest *request, 2063 uint64_t offset, 2064 uint8_t *data, 2065 uint64_t size, 2066 Error **errp) 2067 { 2068 int ret = 0; 2069 NBDExport *exp = client->exp; 2070 size_t progress = 0; 2071 2072 assert(size <= NBD_MAX_BUFFER_SIZE); 2073 while (progress < size) { 2074 int64_t pnum; 2075 int status = blk_co_block_status_above(exp->common.blk, NULL, 2076 offset + progress, 2077 size - progress, &pnum, NULL, 2078 NULL); 2079 bool final; 2080 2081 if (status < 0) { 2082 char *msg = g_strdup_printf("unable to check for holes: %s", 2083 strerror(-status)); 2084 2085 ret = nbd_co_send_chunk_error(client, request, -status, msg, errp); 2086 g_free(msg); 2087 return ret; 2088 } 2089 assert(pnum && pnum <= size - progress); 2090 final = progress + pnum == size; 2091 if (status & BDRV_BLOCK_ZERO) { 2092 NBDReply hdr; 2093 NBDStructuredReadHole chunk; 2094 struct iovec iov[] = { 2095 {.iov_base = &hdr}, 2096 {.iov_base = &chunk, .iov_len = sizeof(chunk)}, 2097 }; 2098 2099 trace_nbd_co_send_chunk_read_hole(request->cookie, 2100 offset + progress, pnum); 2101 set_be_chunk(client, iov, 2, 2102 final ? NBD_REPLY_FLAG_DONE : 0, 2103 NBD_REPLY_TYPE_OFFSET_HOLE, request); 2104 stq_be_p(&chunk.offset, offset + progress); 2105 stl_be_p(&chunk.length, pnum); 2106 ret = nbd_co_send_iov(client, iov, 2, errp); 2107 } else { 2108 ret = blk_co_pread(exp->common.blk, offset + progress, pnum, 2109 data + progress, 0); 2110 if (ret < 0) { 2111 error_setg_errno(errp, -ret, "reading from file failed"); 2112 break; 2113 } 2114 ret = nbd_co_send_chunk_read(client, request, offset + progress, 2115 data + progress, pnum, final, errp); 2116 } 2117 2118 if (ret < 0) { 2119 break; 2120 } 2121 progress += pnum; 2122 } 2123 return ret; 2124 } 2125 2126 typedef struct NBDExtentArray { 2127 NBDExtent64 *extents; 2128 unsigned int nb_alloc; 2129 unsigned int count; 2130 uint64_t total_length; 2131 bool extended; 2132 bool can_add; 2133 bool converted_to_be; 2134 } NBDExtentArray; 2135 2136 static NBDExtentArray *nbd_extent_array_new(unsigned int nb_alloc, 2137 NBDMode mode) 2138 { 2139 NBDExtentArray *ea = g_new0(NBDExtentArray, 1); 2140 2141 assert(mode >= NBD_MODE_STRUCTURED); 2142 ea->nb_alloc = nb_alloc; 2143 ea->extents = g_new(NBDExtent64, nb_alloc); 2144 ea->extended = mode >= NBD_MODE_EXTENDED; 2145 ea->can_add = true; 2146 2147 return ea; 2148 } 2149 2150 static void nbd_extent_array_free(NBDExtentArray *ea) 2151 { 2152 g_free(ea->extents); 2153 g_free(ea); 2154 } 2155 G_DEFINE_AUTOPTR_CLEANUP_FUNC(NBDExtentArray, nbd_extent_array_free) 2156 2157 /* Further modifications of the array after conversion are abandoned */ 2158 static void nbd_extent_array_convert_to_be(NBDExtentArray *ea) 2159 { 2160 int i; 2161 2162 assert(!ea->converted_to_be); 2163 assert(ea->extended); 2164 ea->can_add = false; 2165 ea->converted_to_be = true; 2166 2167 for (i = 0; i < ea->count; i++) { 2168 ea->extents[i].length = cpu_to_be64(ea->extents[i].length); 2169 ea->extents[i].flags = cpu_to_be64(ea->extents[i].flags); 2170 } 2171 } 2172 2173 /* Further modifications of the array after conversion are abandoned */ 2174 static NBDExtent32 *nbd_extent_array_convert_to_narrow(NBDExtentArray *ea) 2175 { 2176 int i; 2177 NBDExtent32 *extents = g_new(NBDExtent32, ea->count); 2178 2179 assert(!ea->converted_to_be); 2180 assert(!ea->extended); 2181 ea->can_add = false; 2182 ea->converted_to_be = true; 2183 2184 for (i = 0; i < ea->count; i++) { 2185 assert((ea->extents[i].length | ea->extents[i].flags) <= UINT32_MAX); 2186 extents[i].length = cpu_to_be32(ea->extents[i].length); 2187 extents[i].flags = cpu_to_be32(ea->extents[i].flags); 2188 } 2189 2190 return extents; 2191 } 2192 2193 /* 2194 * Add extent to NBDExtentArray. If extent can't be added (no available space), 2195 * return -1. 2196 * For safety, when returning -1 for the first time, .can_add is set to false, 2197 * and further calls to nbd_extent_array_add() will crash. 2198 * (this avoids the situation where a caller ignores failure to add one extent, 2199 * where adding another extent that would squash into the last array entry 2200 * would result in an incorrect range reported to the client) 2201 */ 2202 static int nbd_extent_array_add(NBDExtentArray *ea, 2203 uint64_t length, uint32_t flags) 2204 { 2205 assert(ea->can_add); 2206 2207 if (!length) { 2208 return 0; 2209 } 2210 if (!ea->extended) { 2211 assert(length <= UINT32_MAX); 2212 } 2213 2214 /* Extend previous extent if flags are the same */ 2215 if (ea->count > 0 && flags == ea->extents[ea->count - 1].flags) { 2216 uint64_t sum = length + ea->extents[ea->count - 1].length; 2217 2218 /* 2219 * sum cannot overflow: the block layer bounds image size at 2220 * 2^63, and ea->extents[].length comes from the block layer. 2221 */ 2222 assert(sum >= length); 2223 if (sum <= UINT32_MAX || ea->extended) { 2224 ea->extents[ea->count - 1].length = sum; 2225 ea->total_length += length; 2226 return 0; 2227 } 2228 } 2229 2230 if (ea->count >= ea->nb_alloc) { 2231 ea->can_add = false; 2232 return -1; 2233 } 2234 2235 ea->total_length += length; 2236 ea->extents[ea->count] = (NBDExtent64) {.length = length, .flags = flags}; 2237 ea->count++; 2238 2239 return 0; 2240 } 2241 2242 static int coroutine_fn blockstatus_to_extents(BlockBackend *blk, 2243 uint64_t offset, uint64_t bytes, 2244 NBDExtentArray *ea) 2245 { 2246 while (bytes) { 2247 uint32_t flags; 2248 int64_t num; 2249 int ret = blk_co_block_status_above(blk, NULL, offset, bytes, &num, 2250 NULL, NULL); 2251 2252 if (ret < 0) { 2253 return ret; 2254 } 2255 2256 flags = (ret & BDRV_BLOCK_DATA ? 0 : NBD_STATE_HOLE) | 2257 (ret & BDRV_BLOCK_ZERO ? NBD_STATE_ZERO : 0); 2258 2259 if (nbd_extent_array_add(ea, num, flags) < 0) { 2260 return 0; 2261 } 2262 2263 offset += num; 2264 bytes -= num; 2265 } 2266 2267 return 0; 2268 } 2269 2270 static int coroutine_fn blockalloc_to_extents(BlockBackend *blk, 2271 uint64_t offset, uint64_t bytes, 2272 NBDExtentArray *ea) 2273 { 2274 while (bytes) { 2275 int64_t num; 2276 int ret = blk_co_is_allocated_above(blk, NULL, false, offset, bytes, 2277 &num); 2278 2279 if (ret < 0) { 2280 return ret; 2281 } 2282 2283 if (nbd_extent_array_add(ea, num, ret) < 0) { 2284 return 0; 2285 } 2286 2287 offset += num; 2288 bytes -= num; 2289 } 2290 2291 return 0; 2292 } 2293 2294 /* 2295 * nbd_co_send_extents 2296 * 2297 * @ea is converted to BE by the function 2298 * @last controls whether NBD_REPLY_FLAG_DONE is sent. 2299 */ 2300 static int coroutine_fn 2301 nbd_co_send_extents(NBDClient *client, NBDRequest *request, NBDExtentArray *ea, 2302 bool last, uint32_t context_id, Error **errp) 2303 { 2304 NBDReply hdr; 2305 NBDStructuredMeta meta; 2306 NBDExtendedMeta meta_ext; 2307 g_autofree NBDExtent32 *extents = NULL; 2308 uint16_t type; 2309 struct iovec iov[] = { {.iov_base = &hdr}, {0}, {0} }; 2310 2311 if (client->mode >= NBD_MODE_EXTENDED) { 2312 type = NBD_REPLY_TYPE_BLOCK_STATUS_EXT; 2313 2314 iov[1].iov_base = &meta_ext; 2315 iov[1].iov_len = sizeof(meta_ext); 2316 stl_be_p(&meta_ext.context_id, context_id); 2317 stl_be_p(&meta_ext.count, ea->count); 2318 2319 nbd_extent_array_convert_to_be(ea); 2320 iov[2].iov_base = ea->extents; 2321 iov[2].iov_len = ea->count * sizeof(ea->extents[0]); 2322 } else { 2323 type = NBD_REPLY_TYPE_BLOCK_STATUS; 2324 2325 iov[1].iov_base = &meta; 2326 iov[1].iov_len = sizeof(meta); 2327 stl_be_p(&meta.context_id, context_id); 2328 2329 extents = nbd_extent_array_convert_to_narrow(ea); 2330 iov[2].iov_base = extents; 2331 iov[2].iov_len = ea->count * sizeof(extents[0]); 2332 } 2333 2334 trace_nbd_co_send_extents(request->cookie, ea->count, context_id, 2335 ea->total_length, last); 2336 set_be_chunk(client, iov, 3, last ? NBD_REPLY_FLAG_DONE : 0, type, 2337 request); 2338 2339 return nbd_co_send_iov(client, iov, 3, errp); 2340 } 2341 2342 /* Get block status from the exported device and send it to the client */ 2343 static int 2344 coroutine_fn nbd_co_send_block_status(NBDClient *client, NBDRequest *request, 2345 BlockBackend *blk, uint64_t offset, 2346 uint64_t length, bool dont_fragment, 2347 bool last, uint32_t context_id, 2348 Error **errp) 2349 { 2350 int ret; 2351 unsigned int nb_extents = dont_fragment ? 1 : NBD_MAX_BLOCK_STATUS_EXTENTS; 2352 g_autoptr(NBDExtentArray) ea = 2353 nbd_extent_array_new(nb_extents, client->mode); 2354 2355 if (context_id == NBD_META_ID_BASE_ALLOCATION) { 2356 ret = blockstatus_to_extents(blk, offset, length, ea); 2357 } else { 2358 ret = blockalloc_to_extents(blk, offset, length, ea); 2359 } 2360 if (ret < 0) { 2361 return nbd_co_send_chunk_error(client, request, -ret, 2362 "can't get block status", errp); 2363 } 2364 2365 return nbd_co_send_extents(client, request, ea, last, context_id, errp); 2366 } 2367 2368 /* Populate @ea from a dirty bitmap. */ 2369 static void bitmap_to_extents(BdrvDirtyBitmap *bitmap, 2370 uint64_t offset, uint64_t length, 2371 NBDExtentArray *es) 2372 { 2373 int64_t start, dirty_start, dirty_count; 2374 int64_t end = offset + length; 2375 bool full = false; 2376 int64_t bound = es->extended ? INT64_MAX : INT32_MAX; 2377 2378 bdrv_dirty_bitmap_lock(bitmap); 2379 2380 for (start = offset; 2381 bdrv_dirty_bitmap_next_dirty_area(bitmap, start, end, bound, 2382 &dirty_start, &dirty_count); 2383 start = dirty_start + dirty_count) 2384 { 2385 if ((nbd_extent_array_add(es, dirty_start - start, 0) < 0) || 2386 (nbd_extent_array_add(es, dirty_count, NBD_STATE_DIRTY) < 0)) 2387 { 2388 full = true; 2389 break; 2390 } 2391 } 2392 2393 if (!full) { 2394 /* last non dirty extent, nothing to do if array is now full */ 2395 (void) nbd_extent_array_add(es, end - start, 0); 2396 } 2397 2398 bdrv_dirty_bitmap_unlock(bitmap); 2399 } 2400 2401 static int coroutine_fn nbd_co_send_bitmap(NBDClient *client, 2402 NBDRequest *request, 2403 BdrvDirtyBitmap *bitmap, 2404 uint64_t offset, 2405 uint64_t length, bool dont_fragment, 2406 bool last, uint32_t context_id, 2407 Error **errp) 2408 { 2409 unsigned int nb_extents = dont_fragment ? 1 : NBD_MAX_BLOCK_STATUS_EXTENTS; 2410 g_autoptr(NBDExtentArray) ea = 2411 nbd_extent_array_new(nb_extents, client->mode); 2412 2413 bitmap_to_extents(bitmap, offset, length, ea); 2414 2415 return nbd_co_send_extents(client, request, ea, last, context_id, errp); 2416 } 2417 2418 /* nbd_co_receive_request 2419 * Collect a client request. Return 0 if request looks valid, -EIO to drop 2420 * connection right away, -EAGAIN to indicate we were interrupted and the 2421 * channel should be quiesced, and any other negative value to report an error 2422 * to the client (although the caller may still need to disconnect after 2423 * reporting the error). 2424 */ 2425 static int coroutine_fn nbd_co_receive_request(NBDRequestData *req, 2426 NBDRequest *request, 2427 Error **errp) 2428 { 2429 NBDClient *client = req->client; 2430 bool extended_with_payload; 2431 bool check_length = false; 2432 bool check_rofs = false; 2433 bool allocate_buffer = false; 2434 bool payload_okay = false; 2435 uint64_t payload_len = 0; 2436 int valid_flags = NBD_CMD_FLAG_FUA; 2437 int ret; 2438 2439 g_assert(qemu_in_coroutine()); 2440 assert(client->recv_coroutine == qemu_coroutine_self()); 2441 ret = nbd_receive_request(client, request, errp); 2442 if (ret < 0) { 2443 return ret; 2444 } 2445 2446 trace_nbd_co_receive_request_decode_type(request->cookie, request->type, 2447 nbd_cmd_lookup(request->type)); 2448 extended_with_payload = client->mode >= NBD_MODE_EXTENDED && 2449 request->flags & NBD_CMD_FLAG_PAYLOAD_LEN; 2450 if (extended_with_payload) { 2451 payload_len = request->len; 2452 check_length = true; 2453 } 2454 2455 switch (request->type) { 2456 case NBD_CMD_DISC: 2457 /* Special case: we're going to disconnect without a reply, 2458 * whether or not flags, from, or len are bogus */ 2459 req->complete = true; 2460 return -EIO; 2461 2462 case NBD_CMD_READ: 2463 if (client->mode >= NBD_MODE_STRUCTURED) { 2464 valid_flags |= NBD_CMD_FLAG_DF; 2465 } 2466 check_length = true; 2467 allocate_buffer = true; 2468 break; 2469 2470 case NBD_CMD_WRITE: 2471 if (client->mode >= NBD_MODE_EXTENDED) { 2472 if (!extended_with_payload) { 2473 /* The client is noncompliant. Trace it, but proceed. */ 2474 trace_nbd_co_receive_ext_payload_compliance(request->from, 2475 request->len); 2476 } 2477 valid_flags |= NBD_CMD_FLAG_PAYLOAD_LEN; 2478 } 2479 payload_okay = true; 2480 payload_len = request->len; 2481 check_length = true; 2482 allocate_buffer = true; 2483 check_rofs = true; 2484 break; 2485 2486 case NBD_CMD_FLUSH: 2487 break; 2488 2489 case NBD_CMD_TRIM: 2490 check_rofs = true; 2491 break; 2492 2493 case NBD_CMD_CACHE: 2494 check_length = true; 2495 break; 2496 2497 case NBD_CMD_WRITE_ZEROES: 2498 valid_flags |= NBD_CMD_FLAG_NO_HOLE | NBD_CMD_FLAG_FAST_ZERO; 2499 check_rofs = true; 2500 break; 2501 2502 case NBD_CMD_BLOCK_STATUS: 2503 valid_flags |= NBD_CMD_FLAG_REQ_ONE; 2504 break; 2505 2506 default: 2507 /* Unrecognized, will fail later */ 2508 ; 2509 } 2510 2511 /* Payload and buffer handling. */ 2512 if (!payload_len) { 2513 req->complete = true; 2514 } 2515 if (check_length && request->len > NBD_MAX_BUFFER_SIZE) { 2516 /* READ, WRITE, CACHE */ 2517 error_setg(errp, "len (%" PRIu64 ") is larger than max len (%u)", 2518 request->len, NBD_MAX_BUFFER_SIZE); 2519 return -EINVAL; 2520 } 2521 if (payload_len && !payload_okay) { 2522 /* 2523 * For now, we don't support payloads on other commands; but 2524 * we can keep the connection alive by ignoring the payload. 2525 * We will fail the command later with NBD_EINVAL for the use 2526 * of an unsupported flag (and not for access beyond bounds). 2527 */ 2528 assert(request->type != NBD_CMD_WRITE); 2529 request->len = 0; 2530 } 2531 if (allocate_buffer) { 2532 /* READ, WRITE */ 2533 req->data = blk_try_blockalign(client->exp->common.blk, 2534 request->len); 2535 if (req->data == NULL) { 2536 error_setg(errp, "No memory"); 2537 return -ENOMEM; 2538 } 2539 } 2540 if (payload_len) { 2541 if (payload_okay) { 2542 /* WRITE */ 2543 assert(req->data); 2544 ret = nbd_read(client->ioc, req->data, payload_len, 2545 "CMD_WRITE data", errp); 2546 } else { 2547 ret = nbd_drop(client->ioc, payload_len, errp); 2548 } 2549 if (ret < 0) { 2550 return -EIO; 2551 } 2552 req->complete = true; 2553 trace_nbd_co_receive_request_payload_received(request->cookie, 2554 payload_len); 2555 } 2556 2557 /* Sanity checks. */ 2558 if (client->exp->nbdflags & NBD_FLAG_READ_ONLY && check_rofs) { 2559 /* WRITE, TRIM, WRITE_ZEROES */ 2560 error_setg(errp, "Export is read-only"); 2561 return -EROFS; 2562 } 2563 if (request->from > client->exp->size || 2564 request->len > client->exp->size - request->from) { 2565 error_setg(errp, "operation past EOF; From: %" PRIu64 ", Len: %" PRIu64 2566 ", Size: %" PRIu64, request->from, request->len, 2567 client->exp->size); 2568 return (request->type == NBD_CMD_WRITE || 2569 request->type == NBD_CMD_WRITE_ZEROES) ? -ENOSPC : -EINVAL; 2570 } 2571 if (client->check_align && !QEMU_IS_ALIGNED(request->from | request->len, 2572 client->check_align)) { 2573 /* 2574 * The block layer gracefully handles unaligned requests, but 2575 * it's still worth tracing client non-compliance 2576 */ 2577 trace_nbd_co_receive_align_compliance(nbd_cmd_lookup(request->type), 2578 request->from, 2579 request->len, 2580 client->check_align); 2581 } 2582 if (request->flags & ~valid_flags) { 2583 error_setg(errp, "unsupported flags for command %s (got 0x%x)", 2584 nbd_cmd_lookup(request->type), request->flags); 2585 return -EINVAL; 2586 } 2587 2588 return 0; 2589 } 2590 2591 /* Send simple reply without a payload, or a structured error 2592 * @error_msg is ignored if @ret >= 0 2593 * Returns 0 if connection is still live, -errno on failure to talk to client 2594 */ 2595 static coroutine_fn int nbd_send_generic_reply(NBDClient *client, 2596 NBDRequest *request, 2597 int ret, 2598 const char *error_msg, 2599 Error **errp) 2600 { 2601 if (client->mode >= NBD_MODE_STRUCTURED && ret < 0) { 2602 return nbd_co_send_chunk_error(client, request, -ret, error_msg, errp); 2603 } else if (client->mode >= NBD_MODE_EXTENDED) { 2604 return nbd_co_send_chunk_done(client, request, errp); 2605 } else { 2606 return nbd_co_send_simple_reply(client, request, ret < 0 ? -ret : 0, 2607 NULL, 0, errp); 2608 } 2609 } 2610 2611 /* Handle NBD_CMD_READ request. 2612 * Return -errno if sending fails. Other errors are reported directly to the 2613 * client as an error reply. */ 2614 static coroutine_fn int nbd_do_cmd_read(NBDClient *client, NBDRequest *request, 2615 uint8_t *data, Error **errp) 2616 { 2617 int ret; 2618 NBDExport *exp = client->exp; 2619 2620 assert(request->type == NBD_CMD_READ); 2621 assert(request->len <= NBD_MAX_BUFFER_SIZE); 2622 2623 /* XXX: NBD Protocol only documents use of FUA with WRITE */ 2624 if (request->flags & NBD_CMD_FLAG_FUA) { 2625 ret = blk_co_flush(exp->common.blk); 2626 if (ret < 0) { 2627 return nbd_send_generic_reply(client, request, ret, 2628 "flush failed", errp); 2629 } 2630 } 2631 2632 if (client->mode >= NBD_MODE_STRUCTURED && 2633 !(request->flags & NBD_CMD_FLAG_DF) && request->len) 2634 { 2635 return nbd_co_send_sparse_read(client, request, request->from, 2636 data, request->len, errp); 2637 } 2638 2639 ret = blk_co_pread(exp->common.blk, request->from, request->len, data, 0); 2640 if (ret < 0) { 2641 return nbd_send_generic_reply(client, request, ret, 2642 "reading from file failed", errp); 2643 } 2644 2645 if (client->mode >= NBD_MODE_STRUCTURED) { 2646 if (request->len) { 2647 return nbd_co_send_chunk_read(client, request, request->from, data, 2648 request->len, true, errp); 2649 } else { 2650 return nbd_co_send_chunk_done(client, request, errp); 2651 } 2652 } else { 2653 return nbd_co_send_simple_reply(client, request, 0, 2654 data, request->len, errp); 2655 } 2656 } 2657 2658 /* 2659 * nbd_do_cmd_cache 2660 * 2661 * Handle NBD_CMD_CACHE request. 2662 * Return -errno if sending fails. Other errors are reported directly to the 2663 * client as an error reply. 2664 */ 2665 static coroutine_fn int nbd_do_cmd_cache(NBDClient *client, NBDRequest *request, 2666 Error **errp) 2667 { 2668 int ret; 2669 NBDExport *exp = client->exp; 2670 2671 assert(request->type == NBD_CMD_CACHE); 2672 assert(request->len <= NBD_MAX_BUFFER_SIZE); 2673 2674 ret = blk_co_preadv(exp->common.blk, request->from, request->len, 2675 NULL, BDRV_REQ_COPY_ON_READ | BDRV_REQ_PREFETCH); 2676 2677 return nbd_send_generic_reply(client, request, ret, 2678 "caching data failed", errp); 2679 } 2680 2681 /* Handle NBD request. 2682 * Return -errno if sending fails. Other errors are reported directly to the 2683 * client as an error reply. */ 2684 static coroutine_fn int nbd_handle_request(NBDClient *client, 2685 NBDRequest *request, 2686 uint8_t *data, Error **errp) 2687 { 2688 int ret; 2689 int flags; 2690 NBDExport *exp = client->exp; 2691 char *msg; 2692 size_t i; 2693 2694 switch (request->type) { 2695 case NBD_CMD_CACHE: 2696 return nbd_do_cmd_cache(client, request, errp); 2697 2698 case NBD_CMD_READ: 2699 return nbd_do_cmd_read(client, request, data, errp); 2700 2701 case NBD_CMD_WRITE: 2702 flags = 0; 2703 if (request->flags & NBD_CMD_FLAG_FUA) { 2704 flags |= BDRV_REQ_FUA; 2705 } 2706 assert(request->len <= NBD_MAX_BUFFER_SIZE); 2707 ret = blk_co_pwrite(exp->common.blk, request->from, request->len, data, 2708 flags); 2709 return nbd_send_generic_reply(client, request, ret, 2710 "writing to file failed", errp); 2711 2712 case NBD_CMD_WRITE_ZEROES: 2713 flags = 0; 2714 if (request->flags & NBD_CMD_FLAG_FUA) { 2715 flags |= BDRV_REQ_FUA; 2716 } 2717 if (!(request->flags & NBD_CMD_FLAG_NO_HOLE)) { 2718 flags |= BDRV_REQ_MAY_UNMAP; 2719 } 2720 if (request->flags & NBD_CMD_FLAG_FAST_ZERO) { 2721 flags |= BDRV_REQ_NO_FALLBACK; 2722 } 2723 ret = blk_co_pwrite_zeroes(exp->common.blk, request->from, request->len, 2724 flags); 2725 return nbd_send_generic_reply(client, request, ret, 2726 "writing to file failed", errp); 2727 2728 case NBD_CMD_DISC: 2729 /* unreachable, thanks to special case in nbd_co_receive_request() */ 2730 abort(); 2731 2732 case NBD_CMD_FLUSH: 2733 ret = blk_co_flush(exp->common.blk); 2734 return nbd_send_generic_reply(client, request, ret, 2735 "flush failed", errp); 2736 2737 case NBD_CMD_TRIM: 2738 ret = blk_co_pdiscard(exp->common.blk, request->from, request->len); 2739 if (ret >= 0 && request->flags & NBD_CMD_FLAG_FUA) { 2740 ret = blk_co_flush(exp->common.blk); 2741 } 2742 return nbd_send_generic_reply(client, request, ret, 2743 "discard failed", errp); 2744 2745 case NBD_CMD_BLOCK_STATUS: 2746 if (!request->len) { 2747 return nbd_send_generic_reply(client, request, -EINVAL, 2748 "need non-zero length", errp); 2749 } 2750 assert(client->mode >= NBD_MODE_EXTENDED || 2751 request->len <= UINT32_MAX); 2752 if (client->export_meta.count) { 2753 bool dont_fragment = request->flags & NBD_CMD_FLAG_REQ_ONE; 2754 int contexts_remaining = client->export_meta.count; 2755 2756 if (client->export_meta.base_allocation) { 2757 ret = nbd_co_send_block_status(client, request, 2758 exp->common.blk, 2759 request->from, 2760 request->len, dont_fragment, 2761 !--contexts_remaining, 2762 NBD_META_ID_BASE_ALLOCATION, 2763 errp); 2764 if (ret < 0) { 2765 return ret; 2766 } 2767 } 2768 2769 if (client->export_meta.allocation_depth) { 2770 ret = nbd_co_send_block_status(client, request, 2771 exp->common.blk, 2772 request->from, request->len, 2773 dont_fragment, 2774 !--contexts_remaining, 2775 NBD_META_ID_ALLOCATION_DEPTH, 2776 errp); 2777 if (ret < 0) { 2778 return ret; 2779 } 2780 } 2781 2782 for (i = 0; i < client->exp->nr_export_bitmaps; i++) { 2783 if (!client->export_meta.bitmaps[i]) { 2784 continue; 2785 } 2786 ret = nbd_co_send_bitmap(client, request, 2787 client->exp->export_bitmaps[i], 2788 request->from, request->len, 2789 dont_fragment, !--contexts_remaining, 2790 NBD_META_ID_DIRTY_BITMAP + i, errp); 2791 if (ret < 0) { 2792 return ret; 2793 } 2794 } 2795 2796 assert(!contexts_remaining); 2797 2798 return 0; 2799 } else { 2800 return nbd_send_generic_reply(client, request, -EINVAL, 2801 "CMD_BLOCK_STATUS not negotiated", 2802 errp); 2803 } 2804 2805 default: 2806 msg = g_strdup_printf("invalid request type (%" PRIu32 ") received", 2807 request->type); 2808 ret = nbd_send_generic_reply(client, request, -EINVAL, msg, 2809 errp); 2810 g_free(msg); 2811 return ret; 2812 } 2813 } 2814 2815 /* Owns a reference to the NBDClient passed as opaque. */ 2816 static coroutine_fn void nbd_trip(void *opaque) 2817 { 2818 NBDClient *client = opaque; 2819 NBDRequestData *req; 2820 NBDRequest request = { 0 }; /* GCC thinks it can be used uninitialized */ 2821 int ret; 2822 Error *local_err = NULL; 2823 2824 trace_nbd_trip(); 2825 if (client->closing) { 2826 nbd_client_put(client); 2827 return; 2828 } 2829 2830 if (client->quiescing) { 2831 /* 2832 * We're switching between AIO contexts. Don't attempt to receive a new 2833 * request and kick the main context which may be waiting for us. 2834 */ 2835 nbd_client_put(client); 2836 client->recv_coroutine = NULL; 2837 aio_wait_kick(); 2838 return; 2839 } 2840 2841 req = nbd_request_get(client); 2842 ret = nbd_co_receive_request(req, &request, &local_err); 2843 client->recv_coroutine = NULL; 2844 2845 if (client->closing) { 2846 /* 2847 * The client may be closed when we are blocked in 2848 * nbd_co_receive_request() 2849 */ 2850 goto done; 2851 } 2852 2853 if (ret == -EAGAIN) { 2854 assert(client->quiescing); 2855 goto done; 2856 } 2857 2858 nbd_client_receive_next_request(client); 2859 if (ret == -EIO) { 2860 goto disconnect; 2861 } 2862 2863 qio_channel_set_cork(client->ioc, true); 2864 2865 if (ret < 0) { 2866 /* It wasn't -EIO, so, according to nbd_co_receive_request() 2867 * semantics, we should return the error to the client. */ 2868 Error *export_err = local_err; 2869 2870 local_err = NULL; 2871 ret = nbd_send_generic_reply(client, &request, -EINVAL, 2872 error_get_pretty(export_err), &local_err); 2873 error_free(export_err); 2874 } else { 2875 ret = nbd_handle_request(client, &request, req->data, &local_err); 2876 } 2877 if (ret < 0) { 2878 error_prepend(&local_err, "Failed to send reply: "); 2879 goto disconnect; 2880 } 2881 2882 /* We must disconnect after NBD_CMD_WRITE if we did not 2883 * read the payload. 2884 */ 2885 if (!req->complete) { 2886 error_setg(&local_err, "Request handling failed in intermediate state"); 2887 goto disconnect; 2888 } 2889 2890 qio_channel_set_cork(client->ioc, false); 2891 done: 2892 nbd_request_put(req); 2893 nbd_client_put(client); 2894 return; 2895 2896 disconnect: 2897 if (local_err) { 2898 error_reportf_err(local_err, "Disconnect client, due to: "); 2899 } 2900 nbd_request_put(req); 2901 client_close(client, true); 2902 nbd_client_put(client); 2903 } 2904 2905 static void nbd_client_receive_next_request(NBDClient *client) 2906 { 2907 if (!client->recv_coroutine && client->nb_requests < MAX_NBD_REQUESTS && 2908 !client->quiescing) { 2909 nbd_client_get(client); 2910 client->recv_coroutine = qemu_coroutine_create(nbd_trip, client); 2911 aio_co_schedule(client->exp->common.ctx, client->recv_coroutine); 2912 } 2913 } 2914 2915 static coroutine_fn void nbd_co_client_start(void *opaque) 2916 { 2917 NBDClient *client = opaque; 2918 Error *local_err = NULL; 2919 2920 qemu_co_mutex_init(&client->send_lock); 2921 2922 if (nbd_negotiate(client, &local_err)) { 2923 if (local_err) { 2924 error_report_err(local_err); 2925 } 2926 client_close(client, false); 2927 return; 2928 } 2929 2930 nbd_client_receive_next_request(client); 2931 } 2932 2933 /* 2934 * Create a new client listener using the given channel @sioc. 2935 * Begin servicing it in a coroutine. When the connection closes, call 2936 * @close_fn with an indication of whether the client completed negotiation. 2937 */ 2938 void nbd_client_new(QIOChannelSocket *sioc, 2939 QCryptoTLSCreds *tlscreds, 2940 const char *tlsauthz, 2941 void (*close_fn)(NBDClient *, bool)) 2942 { 2943 NBDClient *client; 2944 Coroutine *co; 2945 2946 client = g_new0(NBDClient, 1); 2947 client->refcount = 1; 2948 client->tlscreds = tlscreds; 2949 if (tlscreds) { 2950 object_ref(OBJECT(client->tlscreds)); 2951 } 2952 client->tlsauthz = g_strdup(tlsauthz); 2953 client->sioc = sioc; 2954 qio_channel_set_delay(QIO_CHANNEL(sioc), false); 2955 object_ref(OBJECT(client->sioc)); 2956 client->ioc = QIO_CHANNEL(sioc); 2957 object_ref(OBJECT(client->ioc)); 2958 client->close_fn = close_fn; 2959 2960 co = qemu_coroutine_create(nbd_co_client_start, client); 2961 qemu_coroutine_enter(co); 2962 } 2963