1 /* 2 * Copyright (C) 2016 Red Hat, Inc. 3 * Copyright (C) 2005 Anthony Liguori <anthony@codemonkey.ws> 4 * 5 * Network Block Device Server Side 6 * 7 * This program is free software; you can redistribute it and/or modify 8 * it under the terms of the GNU General Public License as published by 9 * the Free Software Foundation; under version 2 of the License. 10 * 11 * This program is distributed in the hope that it will be useful, 12 * but WITHOUT ANY WARRANTY; without even the implied warranty of 13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 14 * GNU General Public License for more details. 15 * 16 * You should have received a copy of the GNU General Public License 17 * along with this program; if not, see <http://www.gnu.org/licenses/>. 18 */ 19 20 #include "qemu/osdep.h" 21 #include "qapi/error.h" 22 #include "nbd-internal.h" 23 24 static int system_errno_to_nbd_errno(int err) 25 { 26 switch (err) { 27 case 0: 28 return NBD_SUCCESS; 29 case EPERM: 30 case EROFS: 31 return NBD_EPERM; 32 case EIO: 33 return NBD_EIO; 34 case ENOMEM: 35 return NBD_ENOMEM; 36 #ifdef EDQUOT 37 case EDQUOT: 38 #endif 39 case EFBIG: 40 case ENOSPC: 41 return NBD_ENOSPC; 42 case ESHUTDOWN: 43 return NBD_ESHUTDOWN; 44 case EINVAL: 45 default: 46 return NBD_EINVAL; 47 } 48 } 49 50 /* Definitions for opaque data types */ 51 52 typedef struct NBDRequestData NBDRequestData; 53 54 struct NBDRequestData { 55 QSIMPLEQ_ENTRY(NBDRequestData) entry; 56 NBDClient *client; 57 uint8_t *data; 58 bool complete; 59 }; 60 61 struct NBDExport { 62 int refcount; 63 void (*close)(NBDExport *exp); 64 65 BlockBackend *blk; 66 char *name; 67 char *description; 68 off_t dev_offset; 69 off_t size; 70 uint16_t nbdflags; 71 QTAILQ_HEAD(, NBDClient) clients; 72 QTAILQ_ENTRY(NBDExport) next; 73 74 AioContext *ctx; 75 76 BlockBackend *eject_notifier_blk; 77 Notifier eject_notifier; 78 }; 79 80 static QTAILQ_HEAD(, NBDExport) exports = QTAILQ_HEAD_INITIALIZER(exports); 81 82 struct NBDClient { 83 int refcount; 84 void (*close)(NBDClient *client); 85 86 bool no_zeroes; 87 NBDExport *exp; 88 QCryptoTLSCreds *tlscreds; 89 char *tlsaclname; 90 QIOChannelSocket *sioc; /* The underlying data channel */ 91 QIOChannel *ioc; /* The current I/O channel which may differ (eg TLS) */ 92 93 Coroutine *recv_coroutine; 94 95 CoMutex send_lock; 96 Coroutine *send_coroutine; 97 98 QTAILQ_ENTRY(NBDClient) next; 99 int nb_requests; 100 bool closing; 101 }; 102 103 /* That's all folks */ 104 105 static void nbd_client_receive_next_request(NBDClient *client); 106 107 static gboolean nbd_negotiate_continue(QIOChannel *ioc, 108 GIOCondition condition, 109 void *opaque) 110 { 111 qemu_coroutine_enter(opaque); 112 return TRUE; 113 } 114 115 static ssize_t nbd_negotiate_read(QIOChannel *ioc, void *buffer, size_t size) 116 { 117 ssize_t ret; 118 guint watch; 119 120 assert(qemu_in_coroutine()); 121 /* Negotiation are always in main loop. */ 122 watch = qio_channel_add_watch(ioc, 123 G_IO_IN, 124 nbd_negotiate_continue, 125 qemu_coroutine_self(), 126 NULL); 127 ret = read_sync(ioc, buffer, size); 128 g_source_remove(watch); 129 return ret; 130 131 } 132 133 static ssize_t nbd_negotiate_write(QIOChannel *ioc, const void *buffer, 134 size_t size) 135 { 136 ssize_t ret; 137 guint watch; 138 139 assert(qemu_in_coroutine()); 140 /* Negotiation are always in main loop. */ 141 watch = qio_channel_add_watch(ioc, 142 G_IO_OUT, 143 nbd_negotiate_continue, 144 qemu_coroutine_self(), 145 NULL); 146 ret = write_sync(ioc, buffer, size); 147 g_source_remove(watch); 148 return ret; 149 } 150 151 static ssize_t nbd_negotiate_drop_sync(QIOChannel *ioc, size_t size) 152 { 153 ssize_t ret, dropped = size; 154 uint8_t *buffer = g_malloc(MIN(65536, size)); 155 156 while (size > 0) { 157 ret = nbd_negotiate_read(ioc, buffer, MIN(65536, size)); 158 if (ret < 0) { 159 g_free(buffer); 160 return ret; 161 } 162 163 assert(ret <= size); 164 size -= ret; 165 } 166 167 g_free(buffer); 168 return dropped; 169 } 170 171 /* Basic flow for negotiation 172 173 Server Client 174 Negotiate 175 176 or 177 178 Server Client 179 Negotiate #1 180 Option 181 Negotiate #2 182 183 ---- 184 185 followed by 186 187 Server Client 188 Request 189 Response 190 Request 191 Response 192 ... 193 ... 194 Request (type == 2) 195 196 */ 197 198 /* Send a reply header, including length, but no payload. 199 * Return -errno on error, 0 on success. */ 200 static int nbd_negotiate_send_rep_len(QIOChannel *ioc, uint32_t type, 201 uint32_t opt, uint32_t len) 202 { 203 uint64_t magic; 204 205 TRACE("Reply opt=%" PRIx32 " type=%" PRIx32 " len=%" PRIu32, 206 type, opt, len); 207 208 magic = cpu_to_be64(NBD_REP_MAGIC); 209 if (nbd_negotiate_write(ioc, &magic, sizeof(magic)) != sizeof(magic)) { 210 LOG("write failed (rep magic)"); 211 return -EINVAL; 212 } 213 opt = cpu_to_be32(opt); 214 if (nbd_negotiate_write(ioc, &opt, sizeof(opt)) != sizeof(opt)) { 215 LOG("write failed (rep opt)"); 216 return -EINVAL; 217 } 218 type = cpu_to_be32(type); 219 if (nbd_negotiate_write(ioc, &type, sizeof(type)) != sizeof(type)) { 220 LOG("write failed (rep type)"); 221 return -EINVAL; 222 } 223 len = cpu_to_be32(len); 224 if (nbd_negotiate_write(ioc, &len, sizeof(len)) != sizeof(len)) { 225 LOG("write failed (rep data length)"); 226 return -EINVAL; 227 } 228 return 0; 229 } 230 231 /* Send a reply header with default 0 length. 232 * Return -errno on error, 0 on success. */ 233 static int nbd_negotiate_send_rep(QIOChannel *ioc, uint32_t type, uint32_t opt) 234 { 235 return nbd_negotiate_send_rep_len(ioc, type, opt, 0); 236 } 237 238 /* Send an error reply. 239 * Return -errno on error, 0 on success. */ 240 static int GCC_FMT_ATTR(4, 5) 241 nbd_negotiate_send_rep_err(QIOChannel *ioc, uint32_t type, 242 uint32_t opt, const char *fmt, ...) 243 { 244 va_list va; 245 char *msg; 246 int ret; 247 size_t len; 248 249 va_start(va, fmt); 250 msg = g_strdup_vprintf(fmt, va); 251 va_end(va); 252 len = strlen(msg); 253 assert(len < 4096); 254 TRACE("sending error message \"%s\"", msg); 255 ret = nbd_negotiate_send_rep_len(ioc, type, opt, len); 256 if (ret < 0) { 257 goto out; 258 } 259 if (nbd_negotiate_write(ioc, msg, len) != len) { 260 LOG("write failed (error message)"); 261 ret = -EIO; 262 } else { 263 ret = 0; 264 } 265 out: 266 g_free(msg); 267 return ret; 268 } 269 270 /* Send a single NBD_REP_SERVER reply to NBD_OPT_LIST, including payload. 271 * Return -errno on error, 0 on success. */ 272 static int nbd_negotiate_send_rep_list(QIOChannel *ioc, NBDExport *exp) 273 { 274 size_t name_len, desc_len; 275 uint32_t len; 276 const char *name = exp->name ? exp->name : ""; 277 const char *desc = exp->description ? exp->description : ""; 278 int rc; 279 280 TRACE("Advertising export name '%s' description '%s'", name, desc); 281 name_len = strlen(name); 282 desc_len = strlen(desc); 283 len = name_len + desc_len + sizeof(len); 284 rc = nbd_negotiate_send_rep_len(ioc, NBD_REP_SERVER, NBD_OPT_LIST, len); 285 if (rc < 0) { 286 return rc; 287 } 288 289 len = cpu_to_be32(name_len); 290 if (nbd_negotiate_write(ioc, &len, sizeof(len)) != sizeof(len)) { 291 LOG("write failed (name length)"); 292 return -EINVAL; 293 } 294 if (nbd_negotiate_write(ioc, name, name_len) != name_len) { 295 LOG("write failed (name buffer)"); 296 return -EINVAL; 297 } 298 if (nbd_negotiate_write(ioc, desc, desc_len) != desc_len) { 299 LOG("write failed (description buffer)"); 300 return -EINVAL; 301 } 302 return 0; 303 } 304 305 /* Process the NBD_OPT_LIST command, with a potential series of replies. 306 * Return -errno on error, 0 on success. */ 307 static int nbd_negotiate_handle_list(NBDClient *client, uint32_t length) 308 { 309 NBDExport *exp; 310 311 if (length) { 312 if (nbd_negotiate_drop_sync(client->ioc, length) != length) { 313 return -EIO; 314 } 315 return nbd_negotiate_send_rep_err(client->ioc, 316 NBD_REP_ERR_INVALID, NBD_OPT_LIST, 317 "OPT_LIST should not have length"); 318 } 319 320 /* For each export, send a NBD_REP_SERVER reply. */ 321 QTAILQ_FOREACH(exp, &exports, next) { 322 if (nbd_negotiate_send_rep_list(client->ioc, exp)) { 323 return -EINVAL; 324 } 325 } 326 /* Finish with a NBD_REP_ACK. */ 327 return nbd_negotiate_send_rep(client->ioc, NBD_REP_ACK, NBD_OPT_LIST); 328 } 329 330 static int nbd_negotiate_handle_export_name(NBDClient *client, uint32_t length) 331 { 332 int rc = -EINVAL; 333 char name[NBD_MAX_NAME_SIZE + 1]; 334 335 /* Client sends: 336 [20 .. xx] export name (length bytes) 337 */ 338 TRACE("Checking length"); 339 if (length >= sizeof(name)) { 340 LOG("Bad length received"); 341 goto fail; 342 } 343 if (nbd_negotiate_read(client->ioc, name, length) != length) { 344 LOG("read failed"); 345 goto fail; 346 } 347 name[length] = '\0'; 348 349 TRACE("Client requested export '%s'", name); 350 351 client->exp = nbd_export_find(name); 352 if (!client->exp) { 353 LOG("export not found"); 354 goto fail; 355 } 356 357 QTAILQ_INSERT_TAIL(&client->exp->clients, client, next); 358 nbd_export_get(client->exp); 359 rc = 0; 360 fail: 361 return rc; 362 } 363 364 /* Handle NBD_OPT_STARTTLS. Return NULL to drop connection, or else the 365 * new channel for all further (now-encrypted) communication. */ 366 static QIOChannel *nbd_negotiate_handle_starttls(NBDClient *client, 367 uint32_t length) 368 { 369 QIOChannel *ioc; 370 QIOChannelTLS *tioc; 371 struct NBDTLSHandshakeData data = { 0 }; 372 373 TRACE("Setting up TLS"); 374 ioc = client->ioc; 375 if (length) { 376 if (nbd_negotiate_drop_sync(ioc, length) != length) { 377 return NULL; 378 } 379 nbd_negotiate_send_rep_err(ioc, NBD_REP_ERR_INVALID, NBD_OPT_STARTTLS, 380 "OPT_STARTTLS should not have length"); 381 return NULL; 382 } 383 384 if (nbd_negotiate_send_rep(client->ioc, NBD_REP_ACK, 385 NBD_OPT_STARTTLS) < 0) { 386 return NULL; 387 } 388 389 tioc = qio_channel_tls_new_server(ioc, 390 client->tlscreds, 391 client->tlsaclname, 392 NULL); 393 if (!tioc) { 394 return NULL; 395 } 396 397 qio_channel_set_name(QIO_CHANNEL(tioc), "nbd-server-tls"); 398 TRACE("Starting TLS handshake"); 399 data.loop = g_main_loop_new(g_main_context_default(), FALSE); 400 qio_channel_tls_handshake(tioc, 401 nbd_tls_handshake, 402 &data, 403 NULL); 404 405 if (!data.complete) { 406 g_main_loop_run(data.loop); 407 } 408 g_main_loop_unref(data.loop); 409 if (data.error) { 410 object_unref(OBJECT(tioc)); 411 error_free(data.error); 412 return NULL; 413 } 414 415 return QIO_CHANNEL(tioc); 416 } 417 418 419 /* Process all NBD_OPT_* client option commands. 420 * Return -errno on error, 0 on success. */ 421 static int nbd_negotiate_options(NBDClient *client) 422 { 423 uint32_t flags; 424 bool fixedNewstyle = false; 425 426 /* Client sends: 427 [ 0 .. 3] client flags 428 429 [ 0 .. 7] NBD_OPTS_MAGIC 430 [ 8 .. 11] NBD option 431 [12 .. 15] Data length 432 ... Rest of request 433 434 [ 0 .. 7] NBD_OPTS_MAGIC 435 [ 8 .. 11] Second NBD option 436 [12 .. 15] Data length 437 ... Rest of request 438 */ 439 440 if (nbd_negotiate_read(client->ioc, &flags, sizeof(flags)) != 441 sizeof(flags)) { 442 LOG("read failed"); 443 return -EIO; 444 } 445 TRACE("Checking client flags"); 446 be32_to_cpus(&flags); 447 if (flags & NBD_FLAG_C_FIXED_NEWSTYLE) { 448 TRACE("Client supports fixed newstyle handshake"); 449 fixedNewstyle = true; 450 flags &= ~NBD_FLAG_C_FIXED_NEWSTYLE; 451 } 452 if (flags & NBD_FLAG_C_NO_ZEROES) { 453 TRACE("Client supports no zeroes at handshake end"); 454 client->no_zeroes = true; 455 flags &= ~NBD_FLAG_C_NO_ZEROES; 456 } 457 if (flags != 0) { 458 TRACE("Unknown client flags 0x%" PRIx32 " received", flags); 459 return -EIO; 460 } 461 462 while (1) { 463 int ret; 464 uint32_t clientflags, length; 465 uint64_t magic; 466 467 if (nbd_negotiate_read(client->ioc, &magic, sizeof(magic)) != 468 sizeof(magic)) { 469 LOG("read failed"); 470 return -EINVAL; 471 } 472 TRACE("Checking opts magic"); 473 if (magic != be64_to_cpu(NBD_OPTS_MAGIC)) { 474 LOG("Bad magic received"); 475 return -EINVAL; 476 } 477 478 if (nbd_negotiate_read(client->ioc, &clientflags, 479 sizeof(clientflags)) != sizeof(clientflags)) { 480 LOG("read failed"); 481 return -EINVAL; 482 } 483 clientflags = be32_to_cpu(clientflags); 484 485 if (nbd_negotiate_read(client->ioc, &length, sizeof(length)) != 486 sizeof(length)) { 487 LOG("read failed"); 488 return -EINVAL; 489 } 490 length = be32_to_cpu(length); 491 492 TRACE("Checking option 0x%" PRIx32, clientflags); 493 if (client->tlscreds && 494 client->ioc == (QIOChannel *)client->sioc) { 495 QIOChannel *tioc; 496 if (!fixedNewstyle) { 497 TRACE("Unsupported option 0x%" PRIx32, clientflags); 498 return -EINVAL; 499 } 500 switch (clientflags) { 501 case NBD_OPT_STARTTLS: 502 tioc = nbd_negotiate_handle_starttls(client, length); 503 if (!tioc) { 504 return -EIO; 505 } 506 object_unref(OBJECT(client->ioc)); 507 client->ioc = QIO_CHANNEL(tioc); 508 break; 509 510 case NBD_OPT_EXPORT_NAME: 511 /* No way to return an error to client, so drop connection */ 512 TRACE("Option 0x%x not permitted before TLS", clientflags); 513 return -EINVAL; 514 515 default: 516 if (nbd_negotiate_drop_sync(client->ioc, length) != length) { 517 return -EIO; 518 } 519 ret = nbd_negotiate_send_rep_err(client->ioc, 520 NBD_REP_ERR_TLS_REQD, 521 clientflags, 522 "Option 0x%" PRIx32 523 "not permitted before TLS", 524 clientflags); 525 if (ret < 0) { 526 return ret; 527 } 528 /* Let the client keep trying, unless they asked to quit */ 529 if (clientflags == NBD_OPT_ABORT) { 530 return -EINVAL; 531 } 532 break; 533 } 534 } else if (fixedNewstyle) { 535 switch (clientflags) { 536 case NBD_OPT_LIST: 537 ret = nbd_negotiate_handle_list(client, length); 538 if (ret < 0) { 539 return ret; 540 } 541 break; 542 543 case NBD_OPT_ABORT: 544 /* NBD spec says we must try to reply before 545 * disconnecting, but that we must also tolerate 546 * guests that don't wait for our reply. */ 547 nbd_negotiate_send_rep(client->ioc, NBD_REP_ACK, clientflags); 548 return -EINVAL; 549 550 case NBD_OPT_EXPORT_NAME: 551 return nbd_negotiate_handle_export_name(client, length); 552 553 case NBD_OPT_STARTTLS: 554 if (nbd_negotiate_drop_sync(client->ioc, length) != length) { 555 return -EIO; 556 } 557 if (client->tlscreds) { 558 ret = nbd_negotiate_send_rep_err(client->ioc, 559 NBD_REP_ERR_INVALID, 560 clientflags, 561 "TLS already enabled"); 562 } else { 563 ret = nbd_negotiate_send_rep_err(client->ioc, 564 NBD_REP_ERR_POLICY, 565 clientflags, 566 "TLS not configured"); 567 } 568 if (ret < 0) { 569 return ret; 570 } 571 break; 572 default: 573 if (nbd_negotiate_drop_sync(client->ioc, length) != length) { 574 return -EIO; 575 } 576 ret = nbd_negotiate_send_rep_err(client->ioc, 577 NBD_REP_ERR_UNSUP, 578 clientflags, 579 "Unsupported option 0x%" 580 PRIx32, 581 clientflags); 582 if (ret < 0) { 583 return ret; 584 } 585 break; 586 } 587 } else { 588 /* 589 * If broken new-style we should drop the connection 590 * for anything except NBD_OPT_EXPORT_NAME 591 */ 592 switch (clientflags) { 593 case NBD_OPT_EXPORT_NAME: 594 return nbd_negotiate_handle_export_name(client, length); 595 596 default: 597 TRACE("Unsupported option 0x%" PRIx32, clientflags); 598 return -EINVAL; 599 } 600 } 601 } 602 } 603 604 typedef struct { 605 NBDClient *client; 606 Coroutine *co; 607 } NBDClientNewData; 608 609 static coroutine_fn int nbd_negotiate(NBDClientNewData *data) 610 { 611 NBDClient *client = data->client; 612 char buf[8 + 8 + 8 + 128]; 613 int rc; 614 const uint16_t myflags = (NBD_FLAG_HAS_FLAGS | NBD_FLAG_SEND_TRIM | 615 NBD_FLAG_SEND_FLUSH | NBD_FLAG_SEND_FUA | 616 NBD_FLAG_SEND_WRITE_ZEROES); 617 bool oldStyle; 618 size_t len; 619 620 /* Old style negotiation header without options 621 [ 0 .. 7] passwd ("NBDMAGIC") 622 [ 8 .. 15] magic (NBD_CLIENT_MAGIC) 623 [16 .. 23] size 624 [24 .. 25] server flags (0) 625 [26 .. 27] export flags 626 [28 .. 151] reserved (0) 627 628 New style negotiation header with options 629 [ 0 .. 7] passwd ("NBDMAGIC") 630 [ 8 .. 15] magic (NBD_OPTS_MAGIC) 631 [16 .. 17] server flags (0) 632 ....options sent.... 633 [18 .. 25] size 634 [26 .. 27] export flags 635 [28 .. 151] reserved (0, omit if no_zeroes) 636 */ 637 638 qio_channel_set_blocking(client->ioc, false, NULL); 639 rc = -EINVAL; 640 641 TRACE("Beginning negotiation."); 642 memset(buf, 0, sizeof(buf)); 643 memcpy(buf, "NBDMAGIC", 8); 644 645 oldStyle = client->exp != NULL && !client->tlscreds; 646 if (oldStyle) { 647 TRACE("advertising size %" PRIu64 " and flags %x", 648 client->exp->size, client->exp->nbdflags | myflags); 649 stq_be_p(buf + 8, NBD_CLIENT_MAGIC); 650 stq_be_p(buf + 16, client->exp->size); 651 stw_be_p(buf + 26, client->exp->nbdflags | myflags); 652 } else { 653 stq_be_p(buf + 8, NBD_OPTS_MAGIC); 654 stw_be_p(buf + 16, NBD_FLAG_FIXED_NEWSTYLE | NBD_FLAG_NO_ZEROES); 655 } 656 657 if (oldStyle) { 658 if (client->tlscreds) { 659 TRACE("TLS cannot be enabled with oldstyle protocol"); 660 goto fail; 661 } 662 if (nbd_negotiate_write(client->ioc, buf, sizeof(buf)) != sizeof(buf)) { 663 LOG("write failed"); 664 goto fail; 665 } 666 } else { 667 if (nbd_negotiate_write(client->ioc, buf, 18) != 18) { 668 LOG("write failed"); 669 goto fail; 670 } 671 rc = nbd_negotiate_options(client); 672 if (rc != 0) { 673 LOG("option negotiation failed"); 674 goto fail; 675 } 676 677 TRACE("advertising size %" PRIu64 " and flags %x", 678 client->exp->size, client->exp->nbdflags | myflags); 679 stq_be_p(buf + 18, client->exp->size); 680 stw_be_p(buf + 26, client->exp->nbdflags | myflags); 681 len = client->no_zeroes ? 10 : sizeof(buf) - 18; 682 if (nbd_negotiate_write(client->ioc, buf + 18, len) != len) { 683 LOG("write failed"); 684 goto fail; 685 } 686 } 687 688 TRACE("Negotiation succeeded."); 689 rc = 0; 690 fail: 691 return rc; 692 } 693 694 static ssize_t nbd_receive_request(QIOChannel *ioc, NBDRequest *request) 695 { 696 uint8_t buf[NBD_REQUEST_SIZE]; 697 uint32_t magic; 698 ssize_t ret; 699 700 ret = read_sync(ioc, buf, sizeof(buf)); 701 if (ret < 0) { 702 return ret; 703 } 704 705 if (ret != sizeof(buf)) { 706 LOG("read failed"); 707 return -EINVAL; 708 } 709 710 /* Request 711 [ 0 .. 3] magic (NBD_REQUEST_MAGIC) 712 [ 4 .. 5] flags (NBD_CMD_FLAG_FUA, ...) 713 [ 6 .. 7] type (NBD_CMD_READ, ...) 714 [ 8 .. 15] handle 715 [16 .. 23] from 716 [24 .. 27] len 717 */ 718 719 magic = ldl_be_p(buf); 720 request->flags = lduw_be_p(buf + 4); 721 request->type = lduw_be_p(buf + 6); 722 request->handle = ldq_be_p(buf + 8); 723 request->from = ldq_be_p(buf + 16); 724 request->len = ldl_be_p(buf + 24); 725 726 TRACE("Got request: { magic = 0x%" PRIx32 ", .flags = %" PRIx16 727 ", .type = %" PRIx16 ", from = %" PRIu64 ", len = %" PRIu32 " }", 728 magic, request->flags, request->type, request->from, request->len); 729 730 if (magic != NBD_REQUEST_MAGIC) { 731 LOG("invalid magic (got 0x%" PRIx32 ")", magic); 732 return -EINVAL; 733 } 734 return 0; 735 } 736 737 static ssize_t nbd_send_reply(QIOChannel *ioc, NBDReply *reply) 738 { 739 uint8_t buf[NBD_REPLY_SIZE]; 740 ssize_t ret; 741 742 reply->error = system_errno_to_nbd_errno(reply->error); 743 744 TRACE("Sending response to client: { .error = %" PRId32 745 ", handle = %" PRIu64 " }", 746 reply->error, reply->handle); 747 748 /* Reply 749 [ 0 .. 3] magic (NBD_REPLY_MAGIC) 750 [ 4 .. 7] error (0 == no error) 751 [ 7 .. 15] handle 752 */ 753 stl_be_p(buf, NBD_REPLY_MAGIC); 754 stl_be_p(buf + 4, reply->error); 755 stq_be_p(buf + 8, reply->handle); 756 757 ret = write_sync(ioc, buf, sizeof(buf)); 758 if (ret < 0) { 759 return ret; 760 } 761 762 if (ret != sizeof(buf)) { 763 LOG("writing to socket failed"); 764 return -EINVAL; 765 } 766 return 0; 767 } 768 769 #define MAX_NBD_REQUESTS 16 770 771 void nbd_client_get(NBDClient *client) 772 { 773 client->refcount++; 774 } 775 776 void nbd_client_put(NBDClient *client) 777 { 778 if (--client->refcount == 0) { 779 /* The last reference should be dropped by client->close, 780 * which is called by client_close. 781 */ 782 assert(client->closing); 783 784 qio_channel_detach_aio_context(client->ioc); 785 object_unref(OBJECT(client->sioc)); 786 object_unref(OBJECT(client->ioc)); 787 if (client->tlscreds) { 788 object_unref(OBJECT(client->tlscreds)); 789 } 790 g_free(client->tlsaclname); 791 if (client->exp) { 792 QTAILQ_REMOVE(&client->exp->clients, client, next); 793 nbd_export_put(client->exp); 794 } 795 g_free(client); 796 } 797 } 798 799 static void client_close(NBDClient *client) 800 { 801 if (client->closing) { 802 return; 803 } 804 805 client->closing = true; 806 807 /* Force requests to finish. They will drop their own references, 808 * then we'll close the socket and free the NBDClient. 809 */ 810 qio_channel_shutdown(client->ioc, QIO_CHANNEL_SHUTDOWN_BOTH, 811 NULL); 812 813 /* Also tell the client, so that they release their reference. */ 814 if (client->close) { 815 client->close(client); 816 } 817 } 818 819 static NBDRequestData *nbd_request_get(NBDClient *client) 820 { 821 NBDRequestData *req; 822 823 assert(client->nb_requests <= MAX_NBD_REQUESTS - 1); 824 client->nb_requests++; 825 826 req = g_new0(NBDRequestData, 1); 827 nbd_client_get(client); 828 req->client = client; 829 return req; 830 } 831 832 static void nbd_request_put(NBDRequestData *req) 833 { 834 NBDClient *client = req->client; 835 836 if (req->data) { 837 qemu_vfree(req->data); 838 } 839 g_free(req); 840 841 client->nb_requests--; 842 nbd_client_receive_next_request(client); 843 844 nbd_client_put(client); 845 } 846 847 static void blk_aio_attached(AioContext *ctx, void *opaque) 848 { 849 NBDExport *exp = opaque; 850 NBDClient *client; 851 852 TRACE("Export %s: Attaching clients to AIO context %p\n", exp->name, ctx); 853 854 exp->ctx = ctx; 855 856 QTAILQ_FOREACH(client, &exp->clients, next) { 857 qio_channel_attach_aio_context(client->ioc, ctx); 858 if (client->recv_coroutine) { 859 aio_co_schedule(ctx, client->recv_coroutine); 860 } 861 if (client->send_coroutine) { 862 aio_co_schedule(ctx, client->send_coroutine); 863 } 864 } 865 } 866 867 static void blk_aio_detach(void *opaque) 868 { 869 NBDExport *exp = opaque; 870 NBDClient *client; 871 872 TRACE("Export %s: Detaching clients from AIO context %p\n", exp->name, exp->ctx); 873 874 QTAILQ_FOREACH(client, &exp->clients, next) { 875 qio_channel_detach_aio_context(client->ioc); 876 } 877 878 exp->ctx = NULL; 879 } 880 881 static void nbd_eject_notifier(Notifier *n, void *data) 882 { 883 NBDExport *exp = container_of(n, NBDExport, eject_notifier); 884 nbd_export_close(exp); 885 } 886 887 NBDExport *nbd_export_new(BlockDriverState *bs, off_t dev_offset, off_t size, 888 uint16_t nbdflags, void (*close)(NBDExport *), 889 bool writethrough, BlockBackend *on_eject_blk, 890 Error **errp) 891 { 892 BlockBackend *blk; 893 NBDExport *exp = g_malloc0(sizeof(NBDExport)); 894 895 blk = blk_new(); 896 blk_insert_bs(blk, bs); 897 blk_set_enable_write_cache(blk, !writethrough); 898 899 exp->refcount = 1; 900 QTAILQ_INIT(&exp->clients); 901 exp->blk = blk; 902 exp->dev_offset = dev_offset; 903 exp->nbdflags = nbdflags; 904 exp->size = size < 0 ? blk_getlength(blk) : size; 905 if (exp->size < 0) { 906 error_setg_errno(errp, -exp->size, 907 "Failed to determine the NBD export's length"); 908 goto fail; 909 } 910 exp->size -= exp->size % BDRV_SECTOR_SIZE; 911 912 exp->close = close; 913 exp->ctx = blk_get_aio_context(blk); 914 blk_add_aio_context_notifier(blk, blk_aio_attached, blk_aio_detach, exp); 915 916 if (on_eject_blk) { 917 blk_ref(on_eject_blk); 918 exp->eject_notifier_blk = on_eject_blk; 919 exp->eject_notifier.notify = nbd_eject_notifier; 920 blk_add_remove_bs_notifier(on_eject_blk, &exp->eject_notifier); 921 } 922 923 /* 924 * NBD exports are used for non-shared storage migration. Make sure 925 * that BDRV_O_INACTIVE is cleared and the image is ready for write 926 * access since the export could be available before migration handover. 927 */ 928 aio_context_acquire(exp->ctx); 929 blk_invalidate_cache(blk, NULL); 930 aio_context_release(exp->ctx); 931 return exp; 932 933 fail: 934 blk_unref(blk); 935 g_free(exp); 936 return NULL; 937 } 938 939 NBDExport *nbd_export_find(const char *name) 940 { 941 NBDExport *exp; 942 QTAILQ_FOREACH(exp, &exports, next) { 943 if (strcmp(name, exp->name) == 0) { 944 return exp; 945 } 946 } 947 948 return NULL; 949 } 950 951 void nbd_export_set_name(NBDExport *exp, const char *name) 952 { 953 if (exp->name == name) { 954 return; 955 } 956 957 nbd_export_get(exp); 958 if (exp->name != NULL) { 959 g_free(exp->name); 960 exp->name = NULL; 961 QTAILQ_REMOVE(&exports, exp, next); 962 nbd_export_put(exp); 963 } 964 if (name != NULL) { 965 nbd_export_get(exp); 966 exp->name = g_strdup(name); 967 QTAILQ_INSERT_TAIL(&exports, exp, next); 968 } 969 nbd_export_put(exp); 970 } 971 972 void nbd_export_set_description(NBDExport *exp, const char *description) 973 { 974 g_free(exp->description); 975 exp->description = g_strdup(description); 976 } 977 978 void nbd_export_close(NBDExport *exp) 979 { 980 NBDClient *client, *next; 981 982 nbd_export_get(exp); 983 QTAILQ_FOREACH_SAFE(client, &exp->clients, next, next) { 984 client_close(client); 985 } 986 nbd_export_set_name(exp, NULL); 987 nbd_export_set_description(exp, NULL); 988 nbd_export_put(exp); 989 } 990 991 void nbd_export_get(NBDExport *exp) 992 { 993 assert(exp->refcount > 0); 994 exp->refcount++; 995 } 996 997 void nbd_export_put(NBDExport *exp) 998 { 999 assert(exp->refcount > 0); 1000 if (exp->refcount == 1) { 1001 nbd_export_close(exp); 1002 } 1003 1004 if (--exp->refcount == 0) { 1005 assert(exp->name == NULL); 1006 assert(exp->description == NULL); 1007 1008 if (exp->close) { 1009 exp->close(exp); 1010 } 1011 1012 if (exp->blk) { 1013 if (exp->eject_notifier_blk) { 1014 notifier_remove(&exp->eject_notifier); 1015 blk_unref(exp->eject_notifier_blk); 1016 } 1017 blk_remove_aio_context_notifier(exp->blk, blk_aio_attached, 1018 blk_aio_detach, exp); 1019 blk_unref(exp->blk); 1020 exp->blk = NULL; 1021 } 1022 1023 g_free(exp); 1024 } 1025 } 1026 1027 BlockBackend *nbd_export_get_blockdev(NBDExport *exp) 1028 { 1029 return exp->blk; 1030 } 1031 1032 void nbd_export_close_all(void) 1033 { 1034 NBDExport *exp, *next; 1035 1036 QTAILQ_FOREACH_SAFE(exp, &exports, next, next) { 1037 nbd_export_close(exp); 1038 } 1039 } 1040 1041 static ssize_t nbd_co_send_reply(NBDRequestData *req, NBDReply *reply, 1042 int len) 1043 { 1044 NBDClient *client = req->client; 1045 ssize_t rc, ret; 1046 1047 g_assert(qemu_in_coroutine()); 1048 qemu_co_mutex_lock(&client->send_lock); 1049 client->send_coroutine = qemu_coroutine_self(); 1050 1051 if (!len) { 1052 rc = nbd_send_reply(client->ioc, reply); 1053 } else { 1054 qio_channel_set_cork(client->ioc, true); 1055 rc = nbd_send_reply(client->ioc, reply); 1056 if (rc >= 0) { 1057 ret = write_sync(client->ioc, req->data, len); 1058 if (ret != len) { 1059 rc = -EIO; 1060 } 1061 } 1062 qio_channel_set_cork(client->ioc, false); 1063 } 1064 1065 client->send_coroutine = NULL; 1066 qemu_co_mutex_unlock(&client->send_lock); 1067 return rc; 1068 } 1069 1070 /* Collect a client request. Return 0 if request looks valid, -EAGAIN 1071 * to keep trying the collection, -EIO to drop connection right away, 1072 * and any other negative value to report an error to the client 1073 * (although the caller may still need to disconnect after reporting 1074 * the error). */ 1075 static ssize_t nbd_co_receive_request(NBDRequestData *req, 1076 NBDRequest *request) 1077 { 1078 NBDClient *client = req->client; 1079 ssize_t rc; 1080 1081 g_assert(qemu_in_coroutine()); 1082 assert(client->recv_coroutine == qemu_coroutine_self()); 1083 rc = nbd_receive_request(client->ioc, request); 1084 if (rc < 0) { 1085 if (rc != -EAGAIN) { 1086 rc = -EIO; 1087 } 1088 goto out; 1089 } 1090 1091 TRACE("Decoding type"); 1092 1093 if (request->type != NBD_CMD_WRITE) { 1094 /* No payload, we are ready to read the next request. */ 1095 req->complete = true; 1096 } 1097 1098 if (request->type == NBD_CMD_DISC) { 1099 /* Special case: we're going to disconnect without a reply, 1100 * whether or not flags, from, or len are bogus */ 1101 TRACE("Request type is DISCONNECT"); 1102 rc = -EIO; 1103 goto out; 1104 } 1105 1106 /* Check for sanity in the parameters, part 1. Defer as many 1107 * checks as possible until after reading any NBD_CMD_WRITE 1108 * payload, so we can try and keep the connection alive. */ 1109 if ((request->from + request->len) < request->from) { 1110 LOG("integer overflow detected, you're probably being attacked"); 1111 rc = -EINVAL; 1112 goto out; 1113 } 1114 1115 if (request->type == NBD_CMD_READ || request->type == NBD_CMD_WRITE) { 1116 if (request->len > NBD_MAX_BUFFER_SIZE) { 1117 LOG("len (%" PRIu32" ) is larger than max len (%u)", 1118 request->len, NBD_MAX_BUFFER_SIZE); 1119 rc = -EINVAL; 1120 goto out; 1121 } 1122 1123 req->data = blk_try_blockalign(client->exp->blk, request->len); 1124 if (req->data == NULL) { 1125 rc = -ENOMEM; 1126 goto out; 1127 } 1128 } 1129 if (request->type == NBD_CMD_WRITE) { 1130 TRACE("Reading %" PRIu32 " byte(s)", request->len); 1131 1132 if (read_sync(client->ioc, req->data, request->len) != request->len) { 1133 LOG("reading from socket failed"); 1134 rc = -EIO; 1135 goto out; 1136 } 1137 req->complete = true; 1138 } 1139 1140 /* Sanity checks, part 2. */ 1141 if (request->from + request->len > client->exp->size) { 1142 LOG("operation past EOF; From: %" PRIu64 ", Len: %" PRIu32 1143 ", Size: %" PRIu64, request->from, request->len, 1144 (uint64_t)client->exp->size); 1145 rc = request->type == NBD_CMD_WRITE ? -ENOSPC : -EINVAL; 1146 goto out; 1147 } 1148 if (request->flags & ~(NBD_CMD_FLAG_FUA | NBD_CMD_FLAG_NO_HOLE)) { 1149 LOG("unsupported flags (got 0x%x)", request->flags); 1150 rc = -EINVAL; 1151 goto out; 1152 } 1153 if (request->type != NBD_CMD_WRITE_ZEROES && 1154 (request->flags & NBD_CMD_FLAG_NO_HOLE)) { 1155 LOG("unexpected flags (got 0x%x)", request->flags); 1156 rc = -EINVAL; 1157 goto out; 1158 } 1159 1160 rc = 0; 1161 1162 out: 1163 client->recv_coroutine = NULL; 1164 nbd_client_receive_next_request(client); 1165 1166 return rc; 1167 } 1168 1169 /* Owns a reference to the NBDClient passed as opaque. */ 1170 static coroutine_fn void nbd_trip(void *opaque) 1171 { 1172 NBDClient *client = opaque; 1173 NBDExport *exp = client->exp; 1174 NBDRequestData *req; 1175 NBDRequest request = { 0 }; /* GCC thinks it can be used uninitialized */ 1176 NBDReply reply; 1177 ssize_t ret; 1178 int flags; 1179 1180 TRACE("Reading request."); 1181 if (client->closing) { 1182 nbd_client_put(client); 1183 return; 1184 } 1185 1186 req = nbd_request_get(client); 1187 ret = nbd_co_receive_request(req, &request); 1188 if (ret == -EAGAIN) { 1189 goto done; 1190 } 1191 if (ret == -EIO) { 1192 goto out; 1193 } 1194 1195 reply.handle = request.handle; 1196 reply.error = 0; 1197 1198 if (ret < 0) { 1199 reply.error = -ret; 1200 goto error_reply; 1201 } 1202 1203 if (client->closing) { 1204 /* 1205 * The client may be closed when we are blocked in 1206 * nbd_co_receive_request() 1207 */ 1208 goto done; 1209 } 1210 1211 switch (request.type) { 1212 case NBD_CMD_READ: 1213 TRACE("Request type is READ"); 1214 1215 /* XXX: NBD Protocol only documents use of FUA with WRITE */ 1216 if (request.flags & NBD_CMD_FLAG_FUA) { 1217 ret = blk_co_flush(exp->blk); 1218 if (ret < 0) { 1219 LOG("flush failed"); 1220 reply.error = -ret; 1221 goto error_reply; 1222 } 1223 } 1224 1225 ret = blk_pread(exp->blk, request.from + exp->dev_offset, 1226 req->data, request.len); 1227 if (ret < 0) { 1228 LOG("reading from file failed"); 1229 reply.error = -ret; 1230 goto error_reply; 1231 } 1232 1233 TRACE("Read %" PRIu32" byte(s)", request.len); 1234 if (nbd_co_send_reply(req, &reply, request.len) < 0) 1235 goto out; 1236 break; 1237 case NBD_CMD_WRITE: 1238 TRACE("Request type is WRITE"); 1239 1240 if (exp->nbdflags & NBD_FLAG_READ_ONLY) { 1241 TRACE("Server is read-only, return error"); 1242 reply.error = EROFS; 1243 goto error_reply; 1244 } 1245 1246 TRACE("Writing to device"); 1247 1248 flags = 0; 1249 if (request.flags & NBD_CMD_FLAG_FUA) { 1250 flags |= BDRV_REQ_FUA; 1251 } 1252 ret = blk_pwrite(exp->blk, request.from + exp->dev_offset, 1253 req->data, request.len, flags); 1254 if (ret < 0) { 1255 LOG("writing to file failed"); 1256 reply.error = -ret; 1257 goto error_reply; 1258 } 1259 1260 if (nbd_co_send_reply(req, &reply, 0) < 0) { 1261 goto out; 1262 } 1263 break; 1264 1265 case NBD_CMD_WRITE_ZEROES: 1266 TRACE("Request type is WRITE_ZEROES"); 1267 1268 if (exp->nbdflags & NBD_FLAG_READ_ONLY) { 1269 TRACE("Server is read-only, return error"); 1270 reply.error = EROFS; 1271 goto error_reply; 1272 } 1273 1274 TRACE("Writing to device"); 1275 1276 flags = 0; 1277 if (request.flags & NBD_CMD_FLAG_FUA) { 1278 flags |= BDRV_REQ_FUA; 1279 } 1280 if (!(request.flags & NBD_CMD_FLAG_NO_HOLE)) { 1281 flags |= BDRV_REQ_MAY_UNMAP; 1282 } 1283 ret = blk_pwrite_zeroes(exp->blk, request.from + exp->dev_offset, 1284 request.len, flags); 1285 if (ret < 0) { 1286 LOG("writing to file failed"); 1287 reply.error = -ret; 1288 goto error_reply; 1289 } 1290 1291 if (nbd_co_send_reply(req, &reply, 0) < 0) { 1292 goto out; 1293 } 1294 break; 1295 1296 case NBD_CMD_DISC: 1297 /* unreachable, thanks to special case in nbd_co_receive_request() */ 1298 abort(); 1299 1300 case NBD_CMD_FLUSH: 1301 TRACE("Request type is FLUSH"); 1302 1303 ret = blk_co_flush(exp->blk); 1304 if (ret < 0) { 1305 LOG("flush failed"); 1306 reply.error = -ret; 1307 } 1308 if (nbd_co_send_reply(req, &reply, 0) < 0) { 1309 goto out; 1310 } 1311 break; 1312 case NBD_CMD_TRIM: 1313 TRACE("Request type is TRIM"); 1314 ret = blk_co_pdiscard(exp->blk, request.from + exp->dev_offset, 1315 request.len); 1316 if (ret < 0) { 1317 LOG("discard failed"); 1318 reply.error = -ret; 1319 } 1320 if (nbd_co_send_reply(req, &reply, 0) < 0) { 1321 goto out; 1322 } 1323 break; 1324 default: 1325 LOG("invalid request type (%" PRIu32 ") received", request.type); 1326 reply.error = EINVAL; 1327 error_reply: 1328 /* We must disconnect after NBD_CMD_WRITE if we did not 1329 * read the payload. 1330 */ 1331 if (nbd_co_send_reply(req, &reply, 0) < 0 || !req->complete) { 1332 goto out; 1333 } 1334 break; 1335 } 1336 1337 TRACE("Request/Reply complete"); 1338 1339 done: 1340 nbd_request_put(req); 1341 nbd_client_put(client); 1342 return; 1343 1344 out: 1345 nbd_request_put(req); 1346 client_close(client); 1347 nbd_client_put(client); 1348 } 1349 1350 static void nbd_client_receive_next_request(NBDClient *client) 1351 { 1352 if (!client->recv_coroutine && client->nb_requests < MAX_NBD_REQUESTS) { 1353 nbd_client_get(client); 1354 client->recv_coroutine = qemu_coroutine_create(nbd_trip, client); 1355 aio_co_schedule(client->exp->ctx, client->recv_coroutine); 1356 } 1357 } 1358 1359 static coroutine_fn void nbd_co_client_start(void *opaque) 1360 { 1361 NBDClientNewData *data = opaque; 1362 NBDClient *client = data->client; 1363 NBDExport *exp = client->exp; 1364 1365 if (exp) { 1366 nbd_export_get(exp); 1367 } 1368 if (nbd_negotiate(data)) { 1369 client_close(client); 1370 goto out; 1371 } 1372 qemu_co_mutex_init(&client->send_lock); 1373 1374 if (exp) { 1375 QTAILQ_INSERT_TAIL(&exp->clients, client, next); 1376 } 1377 1378 nbd_client_receive_next_request(client); 1379 1380 out: 1381 g_free(data); 1382 } 1383 1384 void nbd_client_new(NBDExport *exp, 1385 QIOChannelSocket *sioc, 1386 QCryptoTLSCreds *tlscreds, 1387 const char *tlsaclname, 1388 void (*close_fn)(NBDClient *)) 1389 { 1390 NBDClient *client; 1391 NBDClientNewData *data = g_new(NBDClientNewData, 1); 1392 1393 client = g_malloc0(sizeof(NBDClient)); 1394 client->refcount = 1; 1395 client->exp = exp; 1396 client->tlscreds = tlscreds; 1397 if (tlscreds) { 1398 object_ref(OBJECT(client->tlscreds)); 1399 } 1400 client->tlsaclname = g_strdup(tlsaclname); 1401 client->sioc = sioc; 1402 object_ref(OBJECT(client->sioc)); 1403 client->ioc = QIO_CHANNEL(sioc); 1404 object_ref(OBJECT(client->ioc)); 1405 client->close = close_fn; 1406 1407 data->client = client; 1408 data->co = qemu_coroutine_create(nbd_co_client_start, data); 1409 qemu_coroutine_enter(data->co); 1410 } 1411