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