1 /* 2 * QEMU Block driver for NBD 3 * 4 * Copyright (c) 2019 Virtuozzo International GmbH. 5 * Copyright Red Hat 6 * Copyright (C) 2008 Bull S.A.S. 7 * Author: Laurent Vivier <Laurent.Vivier@bull.net> 8 * 9 * Some parts: 10 * Copyright (C) 2007 Anthony Liguori <anthony@codemonkey.ws> 11 * 12 * Permission is hereby granted, free of charge, to any person obtaining a copy 13 * of this software and associated documentation files (the "Software"), to deal 14 * in the Software without restriction, including without limitation the rights 15 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell 16 * copies of the Software, and to permit persons to whom the Software is 17 * furnished to do so, subject to the following conditions: 18 * 19 * The above copyright notice and this permission notice shall be included in 20 * all copies or substantial portions of the Software. 21 * 22 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 23 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 24 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 25 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 26 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, 27 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN 28 * THE SOFTWARE. 29 */ 30 31 #include "qemu/osdep.h" 32 33 #include "trace.h" 34 #include "qemu/uri.h" 35 #include "qemu/option.h" 36 #include "qemu/cutils.h" 37 #include "qemu/main-loop.h" 38 39 #include "qapi/qapi-visit-sockets.h" 40 #include "qapi/qmp/qstring.h" 41 #include "qapi/clone-visitor.h" 42 43 #include "block/qdict.h" 44 #include "block/nbd.h" 45 #include "block/block_int.h" 46 #include "block/coroutines.h" 47 48 #include "qemu/yank.h" 49 50 #define EN_OPTSTR ":exportname=" 51 #define MAX_NBD_REQUESTS 16 52 53 #define COOKIE_TO_INDEX(cookie) ((cookie) - 1) 54 #define INDEX_TO_COOKIE(index) ((index) + 1) 55 56 typedef struct { 57 Coroutine *coroutine; 58 uint64_t offset; /* original offset of the request */ 59 bool receiving; /* sleeping in the yield in nbd_receive_replies */ 60 } NBDClientRequest; 61 62 typedef enum NBDClientState { 63 NBD_CLIENT_CONNECTING_WAIT, 64 NBD_CLIENT_CONNECTING_NOWAIT, 65 NBD_CLIENT_CONNECTED, 66 NBD_CLIENT_QUIT 67 } NBDClientState; 68 69 typedef struct BDRVNBDState { 70 QIOChannel *ioc; /* The current I/O channel */ 71 NBDExportInfo info; 72 73 /* 74 * Protects state, free_sema, in_flight, requests[].coroutine, 75 * reconnect_delay_timer. 76 */ 77 QemuMutex requests_lock; 78 NBDClientState state; 79 CoQueue free_sema; 80 unsigned in_flight; 81 NBDClientRequest requests[MAX_NBD_REQUESTS]; 82 QEMUTimer *reconnect_delay_timer; 83 84 /* Protects sending data on the socket. */ 85 CoMutex send_mutex; 86 87 /* 88 * Protects receiving reply headers from the socket, as well as the 89 * fields reply and requests[].receiving 90 */ 91 CoMutex receive_mutex; 92 NBDReply reply; 93 94 QEMUTimer *open_timer; 95 96 BlockDriverState *bs; 97 98 /* Connection parameters */ 99 uint32_t reconnect_delay; 100 uint32_t open_timeout; 101 SocketAddress *saddr; 102 char *export; 103 char *tlscredsid; 104 QCryptoTLSCreds *tlscreds; 105 char *tlshostname; 106 char *x_dirty_bitmap; 107 bool alloc_depth; 108 109 NBDClientConnection *conn; 110 } BDRVNBDState; 111 112 static void nbd_yank(void *opaque); 113 114 static void nbd_clear_bdrvstate(BlockDriverState *bs) 115 { 116 BDRVNBDState *s = (BDRVNBDState *)bs->opaque; 117 118 nbd_client_connection_release(s->conn); 119 s->conn = NULL; 120 121 yank_unregister_instance(BLOCKDEV_YANK_INSTANCE(bs->node_name)); 122 123 /* Must not leave timers behind that would access freed data */ 124 assert(!s->reconnect_delay_timer); 125 assert(!s->open_timer); 126 127 object_unref(OBJECT(s->tlscreds)); 128 qapi_free_SocketAddress(s->saddr); 129 s->saddr = NULL; 130 g_free(s->export); 131 s->export = NULL; 132 g_free(s->tlscredsid); 133 s->tlscredsid = NULL; 134 g_free(s->tlshostname); 135 s->tlshostname = NULL; 136 g_free(s->x_dirty_bitmap); 137 s->x_dirty_bitmap = NULL; 138 } 139 140 /* Called with s->receive_mutex taken. */ 141 static bool coroutine_fn nbd_recv_coroutine_wake_one(NBDClientRequest *req) 142 { 143 if (req->receiving) { 144 req->receiving = false; 145 aio_co_wake(req->coroutine); 146 return true; 147 } 148 149 return false; 150 } 151 152 static void coroutine_fn nbd_recv_coroutines_wake(BDRVNBDState *s) 153 { 154 int i; 155 156 QEMU_LOCK_GUARD(&s->receive_mutex); 157 for (i = 0; i < MAX_NBD_REQUESTS; i++) { 158 if (nbd_recv_coroutine_wake_one(&s->requests[i])) { 159 return; 160 } 161 } 162 } 163 164 /* Called with s->requests_lock held. */ 165 static void coroutine_fn nbd_channel_error_locked(BDRVNBDState *s, int ret) 166 { 167 if (s->state == NBD_CLIENT_CONNECTED) { 168 qio_channel_shutdown(s->ioc, QIO_CHANNEL_SHUTDOWN_BOTH, NULL); 169 } 170 171 if (ret == -EIO) { 172 if (s->state == NBD_CLIENT_CONNECTED) { 173 s->state = s->reconnect_delay ? NBD_CLIENT_CONNECTING_WAIT : 174 NBD_CLIENT_CONNECTING_NOWAIT; 175 } 176 } else { 177 s->state = NBD_CLIENT_QUIT; 178 } 179 } 180 181 static void coroutine_fn nbd_channel_error(BDRVNBDState *s, int ret) 182 { 183 QEMU_LOCK_GUARD(&s->requests_lock); 184 nbd_channel_error_locked(s, ret); 185 } 186 187 static void reconnect_delay_timer_del(BDRVNBDState *s) 188 { 189 if (s->reconnect_delay_timer) { 190 timer_free(s->reconnect_delay_timer); 191 s->reconnect_delay_timer = NULL; 192 } 193 } 194 195 static void reconnect_delay_timer_cb(void *opaque) 196 { 197 BDRVNBDState *s = opaque; 198 199 reconnect_delay_timer_del(s); 200 WITH_QEMU_LOCK_GUARD(&s->requests_lock) { 201 if (s->state != NBD_CLIENT_CONNECTING_WAIT) { 202 return; 203 } 204 s->state = NBD_CLIENT_CONNECTING_NOWAIT; 205 } 206 nbd_co_establish_connection_cancel(s->conn); 207 } 208 209 static void reconnect_delay_timer_init(BDRVNBDState *s, uint64_t expire_time_ns) 210 { 211 assert(!s->reconnect_delay_timer); 212 s->reconnect_delay_timer = aio_timer_new(bdrv_get_aio_context(s->bs), 213 QEMU_CLOCK_REALTIME, 214 SCALE_NS, 215 reconnect_delay_timer_cb, s); 216 timer_mod(s->reconnect_delay_timer, expire_time_ns); 217 } 218 219 static void nbd_teardown_connection(BlockDriverState *bs) 220 { 221 BDRVNBDState *s = (BDRVNBDState *)bs->opaque; 222 223 assert(!s->in_flight); 224 225 if (s->ioc) { 226 qio_channel_shutdown(s->ioc, QIO_CHANNEL_SHUTDOWN_BOTH, NULL); 227 yank_unregister_function(BLOCKDEV_YANK_INSTANCE(s->bs->node_name), 228 nbd_yank, s->bs); 229 object_unref(OBJECT(s->ioc)); 230 s->ioc = NULL; 231 } 232 233 WITH_QEMU_LOCK_GUARD(&s->requests_lock) { 234 s->state = NBD_CLIENT_QUIT; 235 } 236 } 237 238 static void open_timer_del(BDRVNBDState *s) 239 { 240 if (s->open_timer) { 241 timer_free(s->open_timer); 242 s->open_timer = NULL; 243 } 244 } 245 246 static void open_timer_cb(void *opaque) 247 { 248 BDRVNBDState *s = opaque; 249 250 nbd_co_establish_connection_cancel(s->conn); 251 open_timer_del(s); 252 } 253 254 static void open_timer_init(BDRVNBDState *s, uint64_t expire_time_ns) 255 { 256 assert(!s->open_timer); 257 s->open_timer = aio_timer_new(bdrv_get_aio_context(s->bs), 258 QEMU_CLOCK_REALTIME, 259 SCALE_NS, 260 open_timer_cb, s); 261 timer_mod(s->open_timer, expire_time_ns); 262 } 263 264 static bool nbd_client_will_reconnect(BDRVNBDState *s) 265 { 266 /* 267 * Called only after a socket error, so this is not performance sensitive. 268 */ 269 QEMU_LOCK_GUARD(&s->requests_lock); 270 return s->state == NBD_CLIENT_CONNECTING_WAIT; 271 } 272 273 /* 274 * Update @bs with information learned during a completed negotiation process. 275 * Return failure if the server's advertised options are incompatible with the 276 * client's needs. 277 */ 278 static int coroutine_fn GRAPH_RDLOCK 279 nbd_handle_updated_info(BlockDriverState *bs, Error **errp) 280 { 281 BDRVNBDState *s = (BDRVNBDState *)bs->opaque; 282 int ret; 283 284 if (s->x_dirty_bitmap) { 285 if (!s->info.base_allocation) { 286 error_setg(errp, "requested x-dirty-bitmap %s not found", 287 s->x_dirty_bitmap); 288 return -EINVAL; 289 } 290 if (strcmp(s->x_dirty_bitmap, "qemu:allocation-depth") == 0) { 291 s->alloc_depth = true; 292 } 293 } 294 295 if (s->info.flags & NBD_FLAG_READ_ONLY) { 296 ret = bdrv_apply_auto_read_only(bs, "NBD export is read-only", errp); 297 if (ret < 0) { 298 return ret; 299 } 300 } 301 302 if (s->info.flags & NBD_FLAG_SEND_FUA) { 303 bs->supported_write_flags = BDRV_REQ_FUA; 304 bs->supported_zero_flags |= BDRV_REQ_FUA; 305 } 306 307 if (s->info.flags & NBD_FLAG_SEND_WRITE_ZEROES) { 308 bs->supported_zero_flags |= BDRV_REQ_MAY_UNMAP; 309 if (s->info.flags & NBD_FLAG_SEND_FAST_ZERO) { 310 bs->supported_zero_flags |= BDRV_REQ_NO_FALLBACK; 311 } 312 } 313 314 trace_nbd_client_handshake_success(s->export); 315 316 return 0; 317 } 318 319 int coroutine_fn nbd_co_do_establish_connection(BlockDriverState *bs, 320 bool blocking, Error **errp) 321 { 322 BDRVNBDState *s = (BDRVNBDState *)bs->opaque; 323 int ret; 324 IO_CODE(); 325 326 assert_bdrv_graph_readable(); 327 assert(!s->ioc); 328 329 s->ioc = nbd_co_establish_connection(s->conn, &s->info, blocking, errp); 330 if (!s->ioc) { 331 return -ECONNREFUSED; 332 } 333 334 yank_register_function(BLOCKDEV_YANK_INSTANCE(s->bs->node_name), nbd_yank, 335 bs); 336 337 ret = nbd_handle_updated_info(s->bs, NULL); 338 if (ret < 0) { 339 /* 340 * We have connected, but must fail for other reasons. 341 * Send NBD_CMD_DISC as a courtesy to the server. 342 */ 343 NBDRequest request = { .type = NBD_CMD_DISC, .mode = s->info.mode }; 344 345 nbd_send_request(s->ioc, &request); 346 347 yank_unregister_function(BLOCKDEV_YANK_INSTANCE(s->bs->node_name), 348 nbd_yank, bs); 349 object_unref(OBJECT(s->ioc)); 350 s->ioc = NULL; 351 352 return ret; 353 } 354 355 qio_channel_set_blocking(s->ioc, false, NULL); 356 qio_channel_set_follow_coroutine_ctx(s->ioc, true); 357 358 /* successfully connected */ 359 WITH_QEMU_LOCK_GUARD(&s->requests_lock) { 360 s->state = NBD_CLIENT_CONNECTED; 361 } 362 363 return 0; 364 } 365 366 /* Called with s->requests_lock held. */ 367 static bool nbd_client_connecting(BDRVNBDState *s) 368 { 369 return s->state == NBD_CLIENT_CONNECTING_WAIT || 370 s->state == NBD_CLIENT_CONNECTING_NOWAIT; 371 } 372 373 /* Called with s->requests_lock taken. */ 374 static void coroutine_fn GRAPH_RDLOCK nbd_reconnect_attempt(BDRVNBDState *s) 375 { 376 int ret; 377 bool blocking = s->state == NBD_CLIENT_CONNECTING_WAIT; 378 379 /* 380 * Now we are sure that nobody is accessing the channel, and no one will 381 * try until we set the state to CONNECTED. 382 */ 383 assert(nbd_client_connecting(s)); 384 assert(s->in_flight == 1); 385 386 trace_nbd_reconnect_attempt(s->bs->in_flight); 387 388 if (blocking && !s->reconnect_delay_timer) { 389 /* 390 * It's the first reconnect attempt after switching to 391 * NBD_CLIENT_CONNECTING_WAIT 392 */ 393 g_assert(s->reconnect_delay); 394 reconnect_delay_timer_init(s, 395 qemu_clock_get_ns(QEMU_CLOCK_REALTIME) + 396 s->reconnect_delay * NANOSECONDS_PER_SECOND); 397 } 398 399 /* Finalize previous connection if any */ 400 if (s->ioc) { 401 yank_unregister_function(BLOCKDEV_YANK_INSTANCE(s->bs->node_name), 402 nbd_yank, s->bs); 403 object_unref(OBJECT(s->ioc)); 404 s->ioc = NULL; 405 } 406 407 qemu_mutex_unlock(&s->requests_lock); 408 ret = nbd_co_do_establish_connection(s->bs, blocking, NULL); 409 trace_nbd_reconnect_attempt_result(ret, s->bs->in_flight); 410 qemu_mutex_lock(&s->requests_lock); 411 412 /* 413 * The reconnect attempt is done (maybe successfully, maybe not), so 414 * we no longer need this timer. Delete it so it will not outlive 415 * this I/O request (so draining removes all timers). 416 */ 417 reconnect_delay_timer_del(s); 418 } 419 420 static coroutine_fn int nbd_receive_replies(BDRVNBDState *s, uint64_t cookie, 421 Error **errp) 422 { 423 int ret; 424 uint64_t ind = COOKIE_TO_INDEX(cookie), ind2; 425 QEMU_LOCK_GUARD(&s->receive_mutex); 426 427 while (true) { 428 if (s->reply.cookie == cookie) { 429 /* We are done */ 430 return 0; 431 } 432 433 if (s->reply.cookie != 0) { 434 /* 435 * Some other request is being handled now. It should already be 436 * woken by whoever set s->reply.cookie (or never wait in this 437 * yield). So, we should not wake it here. 438 */ 439 ind2 = COOKIE_TO_INDEX(s->reply.cookie); 440 assert(!s->requests[ind2].receiving); 441 442 s->requests[ind].receiving = true; 443 qemu_co_mutex_unlock(&s->receive_mutex); 444 445 qemu_coroutine_yield(); 446 /* 447 * We may be woken for 2 reasons: 448 * 1. From this function, executing in parallel coroutine, when our 449 * cookie is received. 450 * 2. From nbd_co_receive_one_chunk(), when previous request is 451 * finished and s->reply.cookie set to 0. 452 * Anyway, it's OK to lock the mutex and go to the next iteration. 453 */ 454 455 qemu_co_mutex_lock(&s->receive_mutex); 456 assert(!s->requests[ind].receiving); 457 continue; 458 } 459 460 /* We are under mutex and cookie is 0. We have to do the dirty work. */ 461 assert(s->reply.cookie == 0); 462 ret = nbd_receive_reply(s->bs, s->ioc, &s->reply, s->info.mode, errp); 463 if (ret == 0) { 464 ret = -EIO; 465 error_setg(errp, "server dropped connection"); 466 } 467 if (ret < 0) { 468 nbd_channel_error(s, ret); 469 return ret; 470 } 471 if (nbd_reply_is_structured(&s->reply) && 472 s->info.mode < NBD_MODE_STRUCTURED) { 473 nbd_channel_error(s, -EINVAL); 474 error_setg(errp, "unexpected structured reply"); 475 return -EINVAL; 476 } 477 ind2 = COOKIE_TO_INDEX(s->reply.cookie); 478 if (ind2 >= MAX_NBD_REQUESTS || !s->requests[ind2].coroutine) { 479 nbd_channel_error(s, -EINVAL); 480 error_setg(errp, "unexpected cookie value"); 481 return -EINVAL; 482 } 483 if (s->reply.cookie == cookie) { 484 /* We are done */ 485 return 0; 486 } 487 nbd_recv_coroutine_wake_one(&s->requests[ind2]); 488 } 489 } 490 491 static int coroutine_fn GRAPH_RDLOCK 492 nbd_co_send_request(BlockDriverState *bs, NBDRequest *request, 493 QEMUIOVector *qiov) 494 { 495 BDRVNBDState *s = (BDRVNBDState *)bs->opaque; 496 int rc, i = -1; 497 498 qemu_mutex_lock(&s->requests_lock); 499 while (s->in_flight == MAX_NBD_REQUESTS || 500 (s->state != NBD_CLIENT_CONNECTED && s->in_flight > 0)) { 501 qemu_co_queue_wait(&s->free_sema, &s->requests_lock); 502 } 503 504 s->in_flight++; 505 if (s->state != NBD_CLIENT_CONNECTED) { 506 if (nbd_client_connecting(s)) { 507 nbd_reconnect_attempt(s); 508 qemu_co_queue_restart_all(&s->free_sema); 509 } 510 if (s->state != NBD_CLIENT_CONNECTED) { 511 rc = -EIO; 512 goto err; 513 } 514 } 515 516 for (i = 0; i < MAX_NBD_REQUESTS; i++) { 517 if (s->requests[i].coroutine == NULL) { 518 break; 519 } 520 } 521 522 assert(i < MAX_NBD_REQUESTS); 523 s->requests[i].coroutine = qemu_coroutine_self(); 524 s->requests[i].offset = request->from; 525 s->requests[i].receiving = false; 526 qemu_mutex_unlock(&s->requests_lock); 527 528 qemu_co_mutex_lock(&s->send_mutex); 529 request->cookie = INDEX_TO_COOKIE(i); 530 request->mode = s->info.mode; 531 532 assert(s->ioc); 533 534 if (qiov) { 535 qio_channel_set_cork(s->ioc, true); 536 rc = nbd_send_request(s->ioc, request); 537 if (rc >= 0 && qio_channel_writev_all(s->ioc, qiov->iov, qiov->niov, 538 NULL) < 0) { 539 rc = -EIO; 540 } 541 qio_channel_set_cork(s->ioc, false); 542 } else { 543 rc = nbd_send_request(s->ioc, request); 544 } 545 qemu_co_mutex_unlock(&s->send_mutex); 546 547 if (rc < 0) { 548 qemu_mutex_lock(&s->requests_lock); 549 err: 550 nbd_channel_error_locked(s, rc); 551 if (i != -1) { 552 s->requests[i].coroutine = NULL; 553 } 554 s->in_flight--; 555 qemu_co_queue_next(&s->free_sema); 556 qemu_mutex_unlock(&s->requests_lock); 557 } 558 return rc; 559 } 560 561 static inline uint16_t payload_advance16(uint8_t **payload) 562 { 563 *payload += 2; 564 return lduw_be_p(*payload - 2); 565 } 566 567 static inline uint32_t payload_advance32(uint8_t **payload) 568 { 569 *payload += 4; 570 return ldl_be_p(*payload - 4); 571 } 572 573 static inline uint64_t payload_advance64(uint8_t **payload) 574 { 575 *payload += 8; 576 return ldq_be_p(*payload - 8); 577 } 578 579 static int nbd_parse_offset_hole_payload(BDRVNBDState *s, 580 NBDStructuredReplyChunk *chunk, 581 uint8_t *payload, uint64_t orig_offset, 582 QEMUIOVector *qiov, Error **errp) 583 { 584 uint64_t offset; 585 uint32_t hole_size; 586 587 if (chunk->length != sizeof(offset) + sizeof(hole_size)) { 588 error_setg(errp, "Protocol error: invalid payload for " 589 "NBD_REPLY_TYPE_OFFSET_HOLE"); 590 return -EINVAL; 591 } 592 593 offset = payload_advance64(&payload); 594 hole_size = payload_advance32(&payload); 595 596 if (!hole_size || offset < orig_offset || hole_size > qiov->size || 597 offset > orig_offset + qiov->size - hole_size) { 598 error_setg(errp, "Protocol error: server sent chunk exceeding requested" 599 " region"); 600 return -EINVAL; 601 } 602 if (s->info.min_block && 603 !QEMU_IS_ALIGNED(hole_size, s->info.min_block)) { 604 trace_nbd_structured_read_compliance("hole"); 605 } 606 607 qemu_iovec_memset(qiov, offset - orig_offset, 0, hole_size); 608 609 return 0; 610 } 611 612 /* 613 * nbd_parse_blockstatus_payload 614 * Based on our request, we expect only one extent in reply, for the 615 * base:allocation context. 616 */ 617 static int nbd_parse_blockstatus_payload(BDRVNBDState *s, 618 NBDStructuredReplyChunk *chunk, 619 uint8_t *payload, bool wide, 620 uint64_t orig_length, 621 NBDExtent64 *extent, Error **errp) 622 { 623 uint32_t context_id; 624 uint32_t count; 625 size_t ext_len = wide ? sizeof(*extent) : sizeof(NBDExtent32); 626 size_t pay_len = sizeof(context_id) + wide * sizeof(count) + ext_len; 627 628 /* The server succeeded, so it must have sent [at least] one extent */ 629 if (chunk->length < pay_len) { 630 error_setg(errp, "Protocol error: invalid payload for " 631 "NBD_REPLY_TYPE_BLOCK_STATUS"); 632 return -EINVAL; 633 } 634 635 context_id = payload_advance32(&payload); 636 if (s->info.context_id != context_id) { 637 error_setg(errp, "Protocol error: unexpected context id %d for " 638 "NBD_REPLY_TYPE_BLOCK_STATUS, when negotiated context " 639 "id is %d", context_id, 640 s->info.context_id); 641 return -EINVAL; 642 } 643 644 if (wide) { 645 count = payload_advance32(&payload); 646 extent->length = payload_advance64(&payload); 647 extent->flags = payload_advance64(&payload); 648 } else { 649 count = 0; 650 extent->length = payload_advance32(&payload); 651 extent->flags = payload_advance32(&payload); 652 } 653 654 if (extent->length == 0) { 655 error_setg(errp, "Protocol error: server sent status chunk with " 656 "zero length"); 657 return -EINVAL; 658 } 659 660 /* 661 * A server sending unaligned block status is in violation of the 662 * protocol, but as qemu-nbd 3.1 is such a server (at least for 663 * POSIX files that are not a multiple of 512 bytes, since qemu 664 * rounds files up to 512-byte multiples but lseek(SEEK_HOLE) 665 * still sees an implicit hole beyond the real EOF), it's nicer to 666 * work around the misbehaving server. If the request included 667 * more than the final unaligned block, truncate it back to an 668 * aligned result; if the request was only the final block, round 669 * up to the full block and change the status to fully-allocated 670 * (always a safe status, even if it loses information). 671 */ 672 if (s->info.min_block && !QEMU_IS_ALIGNED(extent->length, 673 s->info.min_block)) { 674 trace_nbd_parse_blockstatus_compliance("extent length is unaligned"); 675 if (extent->length > s->info.min_block) { 676 extent->length = QEMU_ALIGN_DOWN(extent->length, 677 s->info.min_block); 678 } else { 679 extent->length = s->info.min_block; 680 extent->flags = 0; 681 } 682 } 683 684 /* 685 * We used NBD_CMD_FLAG_REQ_ONE, so the server should not have 686 * sent us any more than one extent, nor should it have included 687 * status beyond our request in that extent. Furthermore, a wide 688 * server should have replied with an accurate count (we left 689 * count at 0 for a narrow server). However, it's easy enough to 690 * ignore the server's noncompliance without killing the 691 * connection; just ignore trailing extents, and clamp things to 692 * the length of our request. 693 */ 694 if (count != wide || chunk->length > pay_len) { 695 trace_nbd_parse_blockstatus_compliance("unexpected extent count"); 696 } 697 if (extent->length > orig_length) { 698 extent->length = orig_length; 699 trace_nbd_parse_blockstatus_compliance("extent length too large"); 700 } 701 702 /* 703 * HACK: if we are using x-dirty-bitmaps to access 704 * qemu:allocation-depth, treat all depths > 2 the same as 2, 705 * since nbd_client_co_block_status is only expecting the low two 706 * bits to be set. 707 */ 708 if (s->alloc_depth && extent->flags > 2) { 709 extent->flags = 2; 710 } 711 712 return 0; 713 } 714 715 /* 716 * nbd_parse_error_payload 717 * on success @errp contains message describing nbd error reply 718 */ 719 static int nbd_parse_error_payload(NBDStructuredReplyChunk *chunk, 720 uint8_t *payload, int *request_ret, 721 Error **errp) 722 { 723 uint32_t error; 724 uint16_t message_size; 725 726 assert(chunk->type & (1 << 15)); 727 728 if (chunk->length < sizeof(error) + sizeof(message_size)) { 729 error_setg(errp, 730 "Protocol error: invalid payload for structured error"); 731 return -EINVAL; 732 } 733 734 error = nbd_errno_to_system_errno(payload_advance32(&payload)); 735 if (error == 0) { 736 error_setg(errp, "Protocol error: server sent structured error chunk " 737 "with error = 0"); 738 return -EINVAL; 739 } 740 741 *request_ret = -error; 742 message_size = payload_advance16(&payload); 743 744 if (message_size > chunk->length - sizeof(error) - sizeof(message_size)) { 745 error_setg(errp, "Protocol error: server sent structured error chunk " 746 "with incorrect message size"); 747 return -EINVAL; 748 } 749 750 /* TODO: Add a trace point to mention the server complaint */ 751 752 /* TODO handle ERROR_OFFSET */ 753 754 return 0; 755 } 756 757 static int coroutine_fn 758 nbd_co_receive_offset_data_payload(BDRVNBDState *s, uint64_t orig_offset, 759 QEMUIOVector *qiov, Error **errp) 760 { 761 QEMUIOVector sub_qiov; 762 uint64_t offset; 763 size_t data_size; 764 int ret; 765 NBDStructuredReplyChunk *chunk = &s->reply.structured; 766 767 assert(nbd_reply_is_structured(&s->reply)); 768 769 /* The NBD spec requires at least one byte of payload */ 770 if (chunk->length <= sizeof(offset)) { 771 error_setg(errp, "Protocol error: invalid payload for " 772 "NBD_REPLY_TYPE_OFFSET_DATA"); 773 return -EINVAL; 774 } 775 776 if (nbd_read64(s->ioc, &offset, "OFFSET_DATA offset", errp) < 0) { 777 return -EIO; 778 } 779 780 data_size = chunk->length - sizeof(offset); 781 assert(data_size); 782 if (offset < orig_offset || data_size > qiov->size || 783 offset > orig_offset + qiov->size - data_size) { 784 error_setg(errp, "Protocol error: server sent chunk exceeding requested" 785 " region"); 786 return -EINVAL; 787 } 788 if (s->info.min_block && !QEMU_IS_ALIGNED(data_size, s->info.min_block)) { 789 trace_nbd_structured_read_compliance("data"); 790 } 791 792 qemu_iovec_init(&sub_qiov, qiov->niov); 793 qemu_iovec_concat(&sub_qiov, qiov, offset - orig_offset, data_size); 794 ret = qio_channel_readv_all(s->ioc, sub_qiov.iov, sub_qiov.niov, errp); 795 qemu_iovec_destroy(&sub_qiov); 796 797 return ret < 0 ? -EIO : 0; 798 } 799 800 #define NBD_MAX_MALLOC_PAYLOAD 1000 801 static coroutine_fn int nbd_co_receive_structured_payload( 802 BDRVNBDState *s, void **payload, Error **errp) 803 { 804 int ret; 805 uint32_t len; 806 807 assert(nbd_reply_is_structured(&s->reply)); 808 809 len = s->reply.structured.length; 810 811 if (len == 0) { 812 return 0; 813 } 814 815 if (payload == NULL) { 816 error_setg(errp, "Unexpected structured payload"); 817 return -EINVAL; 818 } 819 820 if (len > NBD_MAX_MALLOC_PAYLOAD) { 821 error_setg(errp, "Payload too large"); 822 return -EINVAL; 823 } 824 825 *payload = g_new(char, len); 826 ret = nbd_read(s->ioc, *payload, len, "structured payload", errp); 827 if (ret < 0) { 828 g_free(*payload); 829 *payload = NULL; 830 return ret; 831 } 832 833 return 0; 834 } 835 836 /* 837 * nbd_co_do_receive_one_chunk 838 * for simple reply: 839 * set request_ret to received reply error 840 * if qiov is not NULL: read payload to @qiov 841 * for structured reply chunk: 842 * if error chunk: read payload, set @request_ret, do not set @payload 843 * else if offset_data chunk: read payload data to @qiov, do not set @payload 844 * else: read payload to @payload 845 * 846 * If function fails, @errp contains corresponding error message, and the 847 * connection with the server is suspect. If it returns 0, then the 848 * transaction succeeded (although @request_ret may be a negative errno 849 * corresponding to the server's error reply), and errp is unchanged. 850 */ 851 static coroutine_fn int nbd_co_do_receive_one_chunk( 852 BDRVNBDState *s, uint64_t cookie, bool only_structured, 853 int *request_ret, QEMUIOVector *qiov, void **payload, Error **errp) 854 { 855 ERRP_GUARD(); 856 int ret; 857 int i = COOKIE_TO_INDEX(cookie); 858 void *local_payload = NULL; 859 NBDStructuredReplyChunk *chunk; 860 861 if (payload) { 862 *payload = NULL; 863 } 864 *request_ret = 0; 865 866 ret = nbd_receive_replies(s, cookie, errp); 867 if (ret < 0) { 868 error_prepend(errp, "Connection closed: "); 869 return -EIO; 870 } 871 assert(s->ioc); 872 873 assert(s->reply.cookie == cookie); 874 875 if (nbd_reply_is_simple(&s->reply)) { 876 if (only_structured) { 877 error_setg(errp, "Protocol error: simple reply when structured " 878 "reply chunk was expected"); 879 return -EINVAL; 880 } 881 882 *request_ret = -nbd_errno_to_system_errno(s->reply.simple.error); 883 if (*request_ret < 0 || !qiov) { 884 return 0; 885 } 886 887 return qio_channel_readv_all(s->ioc, qiov->iov, qiov->niov, 888 errp) < 0 ? -EIO : 0; 889 } 890 891 /* handle structured reply chunk */ 892 assert(s->info.mode >= NBD_MODE_STRUCTURED); 893 chunk = &s->reply.structured; 894 895 if (chunk->type == NBD_REPLY_TYPE_NONE) { 896 if (!(chunk->flags & NBD_REPLY_FLAG_DONE)) { 897 error_setg(errp, "Protocol error: NBD_REPLY_TYPE_NONE chunk without" 898 " NBD_REPLY_FLAG_DONE flag set"); 899 return -EINVAL; 900 } 901 if (chunk->length) { 902 error_setg(errp, "Protocol error: NBD_REPLY_TYPE_NONE chunk with" 903 " nonzero length"); 904 return -EINVAL; 905 } 906 return 0; 907 } 908 909 if (chunk->type == NBD_REPLY_TYPE_OFFSET_DATA) { 910 if (!qiov) { 911 error_setg(errp, "Unexpected NBD_REPLY_TYPE_OFFSET_DATA chunk"); 912 return -EINVAL; 913 } 914 915 return nbd_co_receive_offset_data_payload(s, s->requests[i].offset, 916 qiov, errp); 917 } 918 919 if (nbd_reply_type_is_error(chunk->type)) { 920 payload = &local_payload; 921 } 922 923 ret = nbd_co_receive_structured_payload(s, payload, errp); 924 if (ret < 0) { 925 return ret; 926 } 927 928 if (nbd_reply_type_is_error(chunk->type)) { 929 ret = nbd_parse_error_payload(chunk, local_payload, request_ret, errp); 930 g_free(local_payload); 931 return ret; 932 } 933 934 return 0; 935 } 936 937 /* 938 * nbd_co_receive_one_chunk 939 * Read reply, wake up connection_co and set s->quit if needed. 940 * Return value is a fatal error code or normal nbd reply error code 941 */ 942 static coroutine_fn int nbd_co_receive_one_chunk( 943 BDRVNBDState *s, uint64_t cookie, bool only_structured, 944 int *request_ret, QEMUIOVector *qiov, NBDReply *reply, void **payload, 945 Error **errp) 946 { 947 int ret = nbd_co_do_receive_one_chunk(s, cookie, only_structured, 948 request_ret, qiov, payload, errp); 949 950 if (ret < 0) { 951 memset(reply, 0, sizeof(*reply)); 952 nbd_channel_error(s, ret); 953 } else { 954 /* For assert at loop start in nbd_connection_entry */ 955 *reply = s->reply; 956 } 957 s->reply.cookie = 0; 958 959 nbd_recv_coroutines_wake(s); 960 961 return ret; 962 } 963 964 typedef struct NBDReplyChunkIter { 965 int ret; 966 int request_ret; 967 Error *err; 968 bool done, only_structured; 969 } NBDReplyChunkIter; 970 971 static void nbd_iter_channel_error(NBDReplyChunkIter *iter, 972 int ret, Error **local_err) 973 { 974 assert(local_err && *local_err); 975 assert(ret < 0); 976 977 if (!iter->ret) { 978 iter->ret = ret; 979 error_propagate(&iter->err, *local_err); 980 } else { 981 error_free(*local_err); 982 } 983 984 *local_err = NULL; 985 } 986 987 static void nbd_iter_request_error(NBDReplyChunkIter *iter, int ret) 988 { 989 assert(ret < 0); 990 991 if (!iter->request_ret) { 992 iter->request_ret = ret; 993 } 994 } 995 996 /* 997 * NBD_FOREACH_REPLY_CHUNK 998 * The pointer stored in @payload requires g_free() to free it. 999 */ 1000 #define NBD_FOREACH_REPLY_CHUNK(s, iter, cookie, structured, \ 1001 qiov, reply, payload) \ 1002 for (iter = (NBDReplyChunkIter) { .only_structured = structured }; \ 1003 nbd_reply_chunk_iter_receive(s, &iter, cookie, qiov, reply, payload);) 1004 1005 /* 1006 * nbd_reply_chunk_iter_receive 1007 * The pointer stored in @payload requires g_free() to free it. 1008 */ 1009 static bool coroutine_fn nbd_reply_chunk_iter_receive(BDRVNBDState *s, 1010 NBDReplyChunkIter *iter, 1011 uint64_t cookie, 1012 QEMUIOVector *qiov, 1013 NBDReply *reply, 1014 void **payload) 1015 { 1016 int ret, request_ret; 1017 NBDReply local_reply; 1018 NBDStructuredReplyChunk *chunk; 1019 Error *local_err = NULL; 1020 1021 if (iter->done) { 1022 /* Previous iteration was last. */ 1023 goto break_loop; 1024 } 1025 1026 if (reply == NULL) { 1027 reply = &local_reply; 1028 } 1029 1030 ret = nbd_co_receive_one_chunk(s, cookie, iter->only_structured, 1031 &request_ret, qiov, reply, payload, 1032 &local_err); 1033 if (ret < 0) { 1034 nbd_iter_channel_error(iter, ret, &local_err); 1035 } else if (request_ret < 0) { 1036 nbd_iter_request_error(iter, request_ret); 1037 } 1038 1039 /* Do not execute the body of NBD_FOREACH_REPLY_CHUNK for simple reply. */ 1040 if (nbd_reply_is_simple(reply) || iter->ret < 0) { 1041 goto break_loop; 1042 } 1043 1044 chunk = &reply->structured; 1045 iter->only_structured = true; 1046 1047 if (chunk->type == NBD_REPLY_TYPE_NONE) { 1048 /* NBD_REPLY_FLAG_DONE is already checked in nbd_co_receive_one_chunk */ 1049 assert(chunk->flags & NBD_REPLY_FLAG_DONE); 1050 goto break_loop; 1051 } 1052 1053 if (chunk->flags & NBD_REPLY_FLAG_DONE) { 1054 /* This iteration is last. */ 1055 iter->done = true; 1056 } 1057 1058 /* Execute the loop body */ 1059 return true; 1060 1061 break_loop: 1062 qemu_mutex_lock(&s->requests_lock); 1063 s->requests[COOKIE_TO_INDEX(cookie)].coroutine = NULL; 1064 s->in_flight--; 1065 qemu_co_queue_next(&s->free_sema); 1066 qemu_mutex_unlock(&s->requests_lock); 1067 1068 return false; 1069 } 1070 1071 static int coroutine_fn 1072 nbd_co_receive_return_code(BDRVNBDState *s, uint64_t cookie, 1073 int *request_ret, Error **errp) 1074 { 1075 NBDReplyChunkIter iter; 1076 1077 NBD_FOREACH_REPLY_CHUNK(s, iter, cookie, false, NULL, NULL, NULL) { 1078 /* nbd_reply_chunk_iter_receive does all the work */ 1079 } 1080 1081 error_propagate(errp, iter.err); 1082 *request_ret = iter.request_ret; 1083 return iter.ret; 1084 } 1085 1086 static int coroutine_fn 1087 nbd_co_receive_cmdread_reply(BDRVNBDState *s, uint64_t cookie, 1088 uint64_t offset, QEMUIOVector *qiov, 1089 int *request_ret, Error **errp) 1090 { 1091 NBDReplyChunkIter iter; 1092 NBDReply reply; 1093 void *payload = NULL; 1094 Error *local_err = NULL; 1095 1096 NBD_FOREACH_REPLY_CHUNK(s, iter, cookie, 1097 s->info.mode >= NBD_MODE_STRUCTURED, 1098 qiov, &reply, &payload) 1099 { 1100 int ret; 1101 NBDStructuredReplyChunk *chunk = &reply.structured; 1102 1103 assert(nbd_reply_is_structured(&reply)); 1104 1105 switch (chunk->type) { 1106 case NBD_REPLY_TYPE_OFFSET_DATA: 1107 /* 1108 * special cased in nbd_co_receive_one_chunk, data is already 1109 * in qiov 1110 */ 1111 break; 1112 case NBD_REPLY_TYPE_OFFSET_HOLE: 1113 ret = nbd_parse_offset_hole_payload(s, &reply.structured, payload, 1114 offset, qiov, &local_err); 1115 if (ret < 0) { 1116 nbd_channel_error(s, ret); 1117 nbd_iter_channel_error(&iter, ret, &local_err); 1118 } 1119 break; 1120 default: 1121 if (!nbd_reply_type_is_error(chunk->type)) { 1122 /* not allowed reply type */ 1123 nbd_channel_error(s, -EINVAL); 1124 error_setg(&local_err, 1125 "Unexpected reply type: %d (%s) for CMD_READ", 1126 chunk->type, nbd_reply_type_lookup(chunk->type)); 1127 nbd_iter_channel_error(&iter, -EINVAL, &local_err); 1128 } 1129 } 1130 1131 g_free(payload); 1132 payload = NULL; 1133 } 1134 1135 error_propagate(errp, iter.err); 1136 *request_ret = iter.request_ret; 1137 return iter.ret; 1138 } 1139 1140 static int coroutine_fn 1141 nbd_co_receive_blockstatus_reply(BDRVNBDState *s, uint64_t cookie, 1142 uint64_t length, NBDExtent64 *extent, 1143 int *request_ret, Error **errp) 1144 { 1145 NBDReplyChunkIter iter; 1146 NBDReply reply; 1147 void *payload = NULL; 1148 Error *local_err = NULL; 1149 bool received = false; 1150 1151 assert(!extent->length); 1152 NBD_FOREACH_REPLY_CHUNK(s, iter, cookie, false, NULL, &reply, &payload) { 1153 int ret; 1154 NBDStructuredReplyChunk *chunk = &reply.structured; 1155 bool wide; 1156 1157 assert(nbd_reply_is_structured(&reply)); 1158 1159 switch (chunk->type) { 1160 case NBD_REPLY_TYPE_BLOCK_STATUS_EXT: 1161 case NBD_REPLY_TYPE_BLOCK_STATUS: 1162 wide = chunk->type == NBD_REPLY_TYPE_BLOCK_STATUS_EXT; 1163 if ((s->info.mode >= NBD_MODE_EXTENDED) != wide) { 1164 trace_nbd_extended_headers_compliance("block_status"); 1165 } 1166 if (received) { 1167 nbd_channel_error(s, -EINVAL); 1168 error_setg(&local_err, "Several BLOCK_STATUS chunks in reply"); 1169 nbd_iter_channel_error(&iter, -EINVAL, &local_err); 1170 } 1171 received = true; 1172 1173 ret = nbd_parse_blockstatus_payload( 1174 s, &reply.structured, payload, wide, 1175 length, extent, &local_err); 1176 if (ret < 0) { 1177 nbd_channel_error(s, ret); 1178 nbd_iter_channel_error(&iter, ret, &local_err); 1179 } 1180 break; 1181 default: 1182 if (!nbd_reply_type_is_error(chunk->type)) { 1183 nbd_channel_error(s, -EINVAL); 1184 error_setg(&local_err, 1185 "Unexpected reply type: %d (%s) " 1186 "for CMD_BLOCK_STATUS", 1187 chunk->type, nbd_reply_type_lookup(chunk->type)); 1188 nbd_iter_channel_error(&iter, -EINVAL, &local_err); 1189 } 1190 } 1191 1192 g_free(payload); 1193 payload = NULL; 1194 } 1195 1196 if (!extent->length && !iter.request_ret) { 1197 error_setg(&local_err, "Server did not reply with any status extents"); 1198 nbd_iter_channel_error(&iter, -EIO, &local_err); 1199 } 1200 1201 error_propagate(errp, iter.err); 1202 *request_ret = iter.request_ret; 1203 return iter.ret; 1204 } 1205 1206 static int coroutine_fn GRAPH_RDLOCK 1207 nbd_co_request(BlockDriverState *bs, NBDRequest *request, 1208 QEMUIOVector *write_qiov) 1209 { 1210 int ret, request_ret; 1211 Error *local_err = NULL; 1212 BDRVNBDState *s = (BDRVNBDState *)bs->opaque; 1213 1214 assert(request->type != NBD_CMD_READ); 1215 if (write_qiov) { 1216 assert(request->type == NBD_CMD_WRITE); 1217 assert(request->len == iov_size(write_qiov->iov, write_qiov->niov)); 1218 } else { 1219 assert(request->type != NBD_CMD_WRITE); 1220 } 1221 1222 do { 1223 ret = nbd_co_send_request(bs, request, write_qiov); 1224 if (ret < 0) { 1225 continue; 1226 } 1227 1228 ret = nbd_co_receive_return_code(s, request->cookie, 1229 &request_ret, &local_err); 1230 if (local_err) { 1231 trace_nbd_co_request_fail(request->from, request->len, 1232 request->cookie, request->flags, 1233 request->type, 1234 nbd_cmd_lookup(request->type), 1235 ret, error_get_pretty(local_err)); 1236 error_free(local_err); 1237 local_err = NULL; 1238 } 1239 } while (ret < 0 && nbd_client_will_reconnect(s)); 1240 1241 return ret ? ret : request_ret; 1242 } 1243 1244 static int coroutine_fn GRAPH_RDLOCK 1245 nbd_client_co_preadv(BlockDriverState *bs, int64_t offset, int64_t bytes, 1246 QEMUIOVector *qiov, BdrvRequestFlags flags) 1247 { 1248 int ret, request_ret; 1249 Error *local_err = NULL; 1250 BDRVNBDState *s = (BDRVNBDState *)bs->opaque; 1251 NBDRequest request = { 1252 .type = NBD_CMD_READ, 1253 .from = offset, 1254 .len = bytes, 1255 }; 1256 1257 assert(bytes <= NBD_MAX_BUFFER_SIZE); 1258 1259 if (!bytes) { 1260 return 0; 1261 } 1262 /* 1263 * Work around the fact that the block layer doesn't do 1264 * byte-accurate sizing yet - if the read exceeds the server's 1265 * advertised size because the block layer rounded size up, then 1266 * truncate the request to the server and tail-pad with zero. 1267 */ 1268 if (offset >= s->info.size) { 1269 assert(bytes < BDRV_SECTOR_SIZE); 1270 qemu_iovec_memset(qiov, 0, 0, bytes); 1271 return 0; 1272 } 1273 if (offset + bytes > s->info.size) { 1274 uint64_t slop = offset + bytes - s->info.size; 1275 1276 assert(slop < BDRV_SECTOR_SIZE); 1277 qemu_iovec_memset(qiov, bytes - slop, 0, slop); 1278 request.len -= slop; 1279 } 1280 1281 do { 1282 ret = nbd_co_send_request(bs, &request, NULL); 1283 if (ret < 0) { 1284 continue; 1285 } 1286 1287 ret = nbd_co_receive_cmdread_reply(s, request.cookie, offset, qiov, 1288 &request_ret, &local_err); 1289 if (local_err) { 1290 trace_nbd_co_request_fail(request.from, request.len, request.cookie, 1291 request.flags, request.type, 1292 nbd_cmd_lookup(request.type), 1293 ret, error_get_pretty(local_err)); 1294 error_free(local_err); 1295 local_err = NULL; 1296 } 1297 } while (ret < 0 && nbd_client_will_reconnect(s)); 1298 1299 return ret ? ret : request_ret; 1300 } 1301 1302 static int coroutine_fn GRAPH_RDLOCK 1303 nbd_client_co_pwritev(BlockDriverState *bs, int64_t offset, int64_t bytes, 1304 QEMUIOVector *qiov, BdrvRequestFlags flags) 1305 { 1306 BDRVNBDState *s = (BDRVNBDState *)bs->opaque; 1307 NBDRequest request = { 1308 .type = NBD_CMD_WRITE, 1309 .from = offset, 1310 .len = bytes, 1311 }; 1312 1313 assert(!(s->info.flags & NBD_FLAG_READ_ONLY)); 1314 if (flags & BDRV_REQ_FUA) { 1315 assert(s->info.flags & NBD_FLAG_SEND_FUA); 1316 request.flags |= NBD_CMD_FLAG_FUA; 1317 } 1318 1319 assert(bytes <= NBD_MAX_BUFFER_SIZE); 1320 1321 if (!bytes) { 1322 return 0; 1323 } 1324 return nbd_co_request(bs, &request, qiov); 1325 } 1326 1327 static int coroutine_fn GRAPH_RDLOCK 1328 nbd_client_co_pwrite_zeroes(BlockDriverState *bs, int64_t offset, int64_t bytes, 1329 BdrvRequestFlags flags) 1330 { 1331 BDRVNBDState *s = (BDRVNBDState *)bs->opaque; 1332 NBDRequest request = { 1333 .type = NBD_CMD_WRITE_ZEROES, 1334 .from = offset, 1335 .len = bytes, 1336 }; 1337 1338 /* rely on max_pwrite_zeroes */ 1339 assert(bytes <= UINT32_MAX || s->info.mode >= NBD_MODE_EXTENDED); 1340 1341 assert(!(s->info.flags & NBD_FLAG_READ_ONLY)); 1342 if (!(s->info.flags & NBD_FLAG_SEND_WRITE_ZEROES)) { 1343 return -ENOTSUP; 1344 } 1345 1346 if (flags & BDRV_REQ_FUA) { 1347 assert(s->info.flags & NBD_FLAG_SEND_FUA); 1348 request.flags |= NBD_CMD_FLAG_FUA; 1349 } 1350 if (!(flags & BDRV_REQ_MAY_UNMAP)) { 1351 request.flags |= NBD_CMD_FLAG_NO_HOLE; 1352 } 1353 if (flags & BDRV_REQ_NO_FALLBACK) { 1354 assert(s->info.flags & NBD_FLAG_SEND_FAST_ZERO); 1355 request.flags |= NBD_CMD_FLAG_FAST_ZERO; 1356 } 1357 1358 if (!bytes) { 1359 return 0; 1360 } 1361 return nbd_co_request(bs, &request, NULL); 1362 } 1363 1364 static int coroutine_fn GRAPH_RDLOCK nbd_client_co_flush(BlockDriverState *bs) 1365 { 1366 BDRVNBDState *s = (BDRVNBDState *)bs->opaque; 1367 NBDRequest request = { .type = NBD_CMD_FLUSH }; 1368 1369 if (!(s->info.flags & NBD_FLAG_SEND_FLUSH)) { 1370 return 0; 1371 } 1372 1373 request.from = 0; 1374 request.len = 0; 1375 1376 return nbd_co_request(bs, &request, NULL); 1377 } 1378 1379 static int coroutine_fn GRAPH_RDLOCK 1380 nbd_client_co_pdiscard(BlockDriverState *bs, int64_t offset, int64_t bytes) 1381 { 1382 BDRVNBDState *s = (BDRVNBDState *)bs->opaque; 1383 NBDRequest request = { 1384 .type = NBD_CMD_TRIM, 1385 .from = offset, 1386 .len = bytes, 1387 }; 1388 1389 /* rely on max_pdiscard */ 1390 assert(bytes <= UINT32_MAX || s->info.mode >= NBD_MODE_EXTENDED); 1391 1392 assert(!(s->info.flags & NBD_FLAG_READ_ONLY)); 1393 if (!(s->info.flags & NBD_FLAG_SEND_TRIM) || !bytes) { 1394 return 0; 1395 } 1396 1397 return nbd_co_request(bs, &request, NULL); 1398 } 1399 1400 static int coroutine_fn GRAPH_RDLOCK nbd_client_co_block_status( 1401 BlockDriverState *bs, bool want_zero, int64_t offset, int64_t bytes, 1402 int64_t *pnum, int64_t *map, BlockDriverState **file) 1403 { 1404 int ret, request_ret; 1405 NBDExtent64 extent = { 0 }; 1406 BDRVNBDState *s = (BDRVNBDState *)bs->opaque; 1407 Error *local_err = NULL; 1408 1409 NBDRequest request = { 1410 .type = NBD_CMD_BLOCK_STATUS, 1411 .from = offset, 1412 .len = MIN(bytes, s->info.size - offset), 1413 .flags = NBD_CMD_FLAG_REQ_ONE, 1414 }; 1415 1416 if (!s->info.base_allocation) { 1417 *pnum = bytes; 1418 *map = offset; 1419 *file = bs; 1420 return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID; 1421 } 1422 if (s->info.mode < NBD_MODE_EXTENDED) { 1423 request.len = MIN(QEMU_ALIGN_DOWN(INT_MAX, bs->bl.request_alignment), 1424 request.len); 1425 } 1426 1427 /* 1428 * Work around the fact that the block layer doesn't do 1429 * byte-accurate sizing yet - if the status request exceeds the 1430 * server's advertised size because the block layer rounded size 1431 * up, we truncated the request to the server (above), or are 1432 * called on just the hole. 1433 */ 1434 if (offset >= s->info.size) { 1435 *pnum = bytes; 1436 assert(bytes < BDRV_SECTOR_SIZE); 1437 /* Intentionally don't report offset_valid for the hole */ 1438 return BDRV_BLOCK_ZERO; 1439 } 1440 1441 if (s->info.min_block) { 1442 assert(QEMU_IS_ALIGNED(request.len, s->info.min_block)); 1443 } 1444 do { 1445 ret = nbd_co_send_request(bs, &request, NULL); 1446 if (ret < 0) { 1447 continue; 1448 } 1449 1450 ret = nbd_co_receive_blockstatus_reply(s, request.cookie, bytes, 1451 &extent, &request_ret, 1452 &local_err); 1453 if (local_err) { 1454 trace_nbd_co_request_fail(request.from, request.len, request.cookie, 1455 request.flags, request.type, 1456 nbd_cmd_lookup(request.type), 1457 ret, error_get_pretty(local_err)); 1458 error_free(local_err); 1459 local_err = NULL; 1460 } 1461 } while (ret < 0 && nbd_client_will_reconnect(s)); 1462 1463 if (ret < 0 || request_ret < 0) { 1464 return ret ? ret : request_ret; 1465 } 1466 1467 assert(extent.length); 1468 *pnum = extent.length; 1469 *map = offset; 1470 *file = bs; 1471 return (extent.flags & NBD_STATE_HOLE ? 0 : BDRV_BLOCK_DATA) | 1472 (extent.flags & NBD_STATE_ZERO ? BDRV_BLOCK_ZERO : 0) | 1473 BDRV_BLOCK_OFFSET_VALID; 1474 } 1475 1476 static int nbd_client_reopen_prepare(BDRVReopenState *state, 1477 BlockReopenQueue *queue, Error **errp) 1478 { 1479 BDRVNBDState *s = (BDRVNBDState *)state->bs->opaque; 1480 1481 if ((state->flags & BDRV_O_RDWR) && (s->info.flags & NBD_FLAG_READ_ONLY)) { 1482 error_setg(errp, "Can't reopen read-only NBD mount as read/write"); 1483 return -EACCES; 1484 } 1485 return 0; 1486 } 1487 1488 static void nbd_yank(void *opaque) 1489 { 1490 BlockDriverState *bs = opaque; 1491 BDRVNBDState *s = (BDRVNBDState *)bs->opaque; 1492 1493 QEMU_LOCK_GUARD(&s->requests_lock); 1494 qio_channel_shutdown(s->ioc, QIO_CHANNEL_SHUTDOWN_BOTH, NULL); 1495 s->state = NBD_CLIENT_QUIT; 1496 } 1497 1498 static void nbd_client_close(BlockDriverState *bs) 1499 { 1500 BDRVNBDState *s = (BDRVNBDState *)bs->opaque; 1501 NBDRequest request = { .type = NBD_CMD_DISC, .mode = s->info.mode }; 1502 1503 if (s->ioc) { 1504 nbd_send_request(s->ioc, &request); 1505 } 1506 1507 nbd_teardown_connection(bs); 1508 } 1509 1510 1511 /* 1512 * Parse nbd_open options 1513 */ 1514 1515 static int nbd_parse_uri(const char *filename, QDict *options) 1516 { 1517 URI *uri; 1518 const char *p; 1519 QueryParams *qp = NULL; 1520 int ret = 0; 1521 bool is_unix; 1522 1523 uri = uri_parse(filename); 1524 if (!uri) { 1525 return -EINVAL; 1526 } 1527 1528 /* transport */ 1529 if (!g_strcmp0(uri->scheme, "nbd")) { 1530 is_unix = false; 1531 } else if (!g_strcmp0(uri->scheme, "nbd+tcp")) { 1532 is_unix = false; 1533 } else if (!g_strcmp0(uri->scheme, "nbd+unix")) { 1534 is_unix = true; 1535 } else { 1536 ret = -EINVAL; 1537 goto out; 1538 } 1539 1540 p = uri->path ? uri->path : ""; 1541 if (p[0] == '/') { 1542 p++; 1543 } 1544 if (p[0]) { 1545 qdict_put_str(options, "export", p); 1546 } 1547 1548 qp = query_params_parse(uri->query); 1549 if (qp->n > 1 || (is_unix && !qp->n) || (!is_unix && qp->n)) { 1550 ret = -EINVAL; 1551 goto out; 1552 } 1553 1554 if (is_unix) { 1555 /* nbd+unix:///export?socket=path */ 1556 if (uri->server || uri->port || strcmp(qp->p[0].name, "socket")) { 1557 ret = -EINVAL; 1558 goto out; 1559 } 1560 qdict_put_str(options, "server.type", "unix"); 1561 qdict_put_str(options, "server.path", qp->p[0].value); 1562 } else { 1563 QString *host; 1564 char *port_str; 1565 1566 /* nbd[+tcp]://host[:port]/export */ 1567 if (!uri->server) { 1568 ret = -EINVAL; 1569 goto out; 1570 } 1571 1572 /* strip braces from literal IPv6 address */ 1573 if (uri->server[0] == '[') { 1574 host = qstring_from_substr(uri->server, 1, 1575 strlen(uri->server) - 1); 1576 } else { 1577 host = qstring_from_str(uri->server); 1578 } 1579 1580 qdict_put_str(options, "server.type", "inet"); 1581 qdict_put(options, "server.host", host); 1582 1583 port_str = g_strdup_printf("%d", uri->port ?: NBD_DEFAULT_PORT); 1584 qdict_put_str(options, "server.port", port_str); 1585 g_free(port_str); 1586 } 1587 1588 out: 1589 if (qp) { 1590 query_params_free(qp); 1591 } 1592 uri_free(uri); 1593 return ret; 1594 } 1595 1596 static bool nbd_has_filename_options_conflict(QDict *options, Error **errp) 1597 { 1598 const QDictEntry *e; 1599 1600 for (e = qdict_first(options); e; e = qdict_next(options, e)) { 1601 if (!strcmp(e->key, "host") || 1602 !strcmp(e->key, "port") || 1603 !strcmp(e->key, "path") || 1604 !strcmp(e->key, "export") || 1605 strstart(e->key, "server.", NULL)) 1606 { 1607 error_setg(errp, "Option '%s' cannot be used with a file name", 1608 e->key); 1609 return true; 1610 } 1611 } 1612 1613 return false; 1614 } 1615 1616 static void nbd_parse_filename(const char *filename, QDict *options, 1617 Error **errp) 1618 { 1619 g_autofree char *file = NULL; 1620 char *export_name; 1621 const char *host_spec; 1622 const char *unixpath; 1623 1624 if (nbd_has_filename_options_conflict(options, errp)) { 1625 return; 1626 } 1627 1628 if (strstr(filename, "://")) { 1629 int ret = nbd_parse_uri(filename, options); 1630 if (ret < 0) { 1631 error_setg(errp, "No valid URL specified"); 1632 } 1633 return; 1634 } 1635 1636 file = g_strdup(filename); 1637 1638 export_name = strstr(file, EN_OPTSTR); 1639 if (export_name) { 1640 if (export_name[strlen(EN_OPTSTR)] == 0) { 1641 return; 1642 } 1643 export_name[0] = 0; /* truncate 'file' */ 1644 export_name += strlen(EN_OPTSTR); 1645 1646 qdict_put_str(options, "export", export_name); 1647 } 1648 1649 /* extract the host_spec - fail if it's not nbd:... */ 1650 if (!strstart(file, "nbd:", &host_spec)) { 1651 error_setg(errp, "File name string for NBD must start with 'nbd:'"); 1652 return; 1653 } 1654 1655 if (!*host_spec) { 1656 return; 1657 } 1658 1659 /* are we a UNIX or TCP socket? */ 1660 if (strstart(host_spec, "unix:", &unixpath)) { 1661 qdict_put_str(options, "server.type", "unix"); 1662 qdict_put_str(options, "server.path", unixpath); 1663 } else { 1664 InetSocketAddress *addr = g_new(InetSocketAddress, 1); 1665 1666 if (inet_parse(addr, host_spec, errp)) { 1667 goto out_inet; 1668 } 1669 1670 qdict_put_str(options, "server.type", "inet"); 1671 qdict_put_str(options, "server.host", addr->host); 1672 qdict_put_str(options, "server.port", addr->port); 1673 out_inet: 1674 qapi_free_InetSocketAddress(addr); 1675 } 1676 } 1677 1678 static bool nbd_process_legacy_socket_options(QDict *output_options, 1679 QemuOpts *legacy_opts, 1680 Error **errp) 1681 { 1682 const char *path = qemu_opt_get(legacy_opts, "path"); 1683 const char *host = qemu_opt_get(legacy_opts, "host"); 1684 const char *port = qemu_opt_get(legacy_opts, "port"); 1685 const QDictEntry *e; 1686 1687 if (!path && !host && !port) { 1688 return true; 1689 } 1690 1691 for (e = qdict_first(output_options); e; e = qdict_next(output_options, e)) 1692 { 1693 if (strstart(e->key, "server.", NULL)) { 1694 error_setg(errp, "Cannot use 'server' and path/host/port at the " 1695 "same time"); 1696 return false; 1697 } 1698 } 1699 1700 if (path && host) { 1701 error_setg(errp, "path and host may not be used at the same time"); 1702 return false; 1703 } else if (path) { 1704 if (port) { 1705 error_setg(errp, "port may not be used without host"); 1706 return false; 1707 } 1708 1709 qdict_put_str(output_options, "server.type", "unix"); 1710 qdict_put_str(output_options, "server.path", path); 1711 } else if (host) { 1712 qdict_put_str(output_options, "server.type", "inet"); 1713 qdict_put_str(output_options, "server.host", host); 1714 qdict_put_str(output_options, "server.port", 1715 port ?: stringify(NBD_DEFAULT_PORT)); 1716 } 1717 1718 return true; 1719 } 1720 1721 static SocketAddress *nbd_config(BDRVNBDState *s, QDict *options, 1722 Error **errp) 1723 { 1724 SocketAddress *saddr = NULL; 1725 QDict *addr = NULL; 1726 Visitor *iv = NULL; 1727 1728 qdict_extract_subqdict(options, &addr, "server."); 1729 if (!qdict_size(addr)) { 1730 error_setg(errp, "NBD server address missing"); 1731 goto done; 1732 } 1733 1734 iv = qobject_input_visitor_new_flat_confused(addr, errp); 1735 if (!iv) { 1736 goto done; 1737 } 1738 1739 if (!visit_type_SocketAddress(iv, NULL, &saddr, errp)) { 1740 goto done; 1741 } 1742 1743 if (socket_address_parse_named_fd(saddr, errp) < 0) { 1744 qapi_free_SocketAddress(saddr); 1745 saddr = NULL; 1746 goto done; 1747 } 1748 1749 done: 1750 qobject_unref(addr); 1751 visit_free(iv); 1752 return saddr; 1753 } 1754 1755 static QCryptoTLSCreds *nbd_get_tls_creds(const char *id, Error **errp) 1756 { 1757 Object *obj; 1758 QCryptoTLSCreds *creds; 1759 1760 obj = object_resolve_path_component( 1761 object_get_objects_root(), id); 1762 if (!obj) { 1763 error_setg(errp, "No TLS credentials with id '%s'", 1764 id); 1765 return NULL; 1766 } 1767 creds = (QCryptoTLSCreds *) 1768 object_dynamic_cast(obj, TYPE_QCRYPTO_TLS_CREDS); 1769 if (!creds) { 1770 error_setg(errp, "Object with id '%s' is not TLS credentials", 1771 id); 1772 return NULL; 1773 } 1774 1775 if (!qcrypto_tls_creds_check_endpoint(creds, 1776 QCRYPTO_TLS_CREDS_ENDPOINT_CLIENT, 1777 errp)) { 1778 return NULL; 1779 } 1780 object_ref(obj); 1781 return creds; 1782 } 1783 1784 1785 static QemuOptsList nbd_runtime_opts = { 1786 .name = "nbd", 1787 .head = QTAILQ_HEAD_INITIALIZER(nbd_runtime_opts.head), 1788 .desc = { 1789 { 1790 .name = "host", 1791 .type = QEMU_OPT_STRING, 1792 .help = "TCP host to connect to", 1793 }, 1794 { 1795 .name = "port", 1796 .type = QEMU_OPT_STRING, 1797 .help = "TCP port to connect to", 1798 }, 1799 { 1800 .name = "path", 1801 .type = QEMU_OPT_STRING, 1802 .help = "Unix socket path to connect to", 1803 }, 1804 { 1805 .name = "export", 1806 .type = QEMU_OPT_STRING, 1807 .help = "Name of the NBD export to open", 1808 }, 1809 { 1810 .name = "tls-creds", 1811 .type = QEMU_OPT_STRING, 1812 .help = "ID of the TLS credentials to use", 1813 }, 1814 { 1815 .name = "tls-hostname", 1816 .type = QEMU_OPT_STRING, 1817 .help = "Override hostname for validating TLS x509 certificate", 1818 }, 1819 { 1820 .name = "x-dirty-bitmap", 1821 .type = QEMU_OPT_STRING, 1822 .help = "experimental: expose named dirty bitmap in place of " 1823 "block status", 1824 }, 1825 { 1826 .name = "reconnect-delay", 1827 .type = QEMU_OPT_NUMBER, 1828 .help = "On an unexpected disconnect, the nbd client tries to " 1829 "connect again until succeeding or encountering a serious " 1830 "error. During the first @reconnect-delay seconds, all " 1831 "requests are paused and will be rerun on a successful " 1832 "reconnect. After that time, any delayed requests and all " 1833 "future requests before a successful reconnect will " 1834 "immediately fail. Default 0", 1835 }, 1836 { 1837 .name = "open-timeout", 1838 .type = QEMU_OPT_NUMBER, 1839 .help = "In seconds. If zero, the nbd driver tries the connection " 1840 "only once, and fails to open if the connection fails. " 1841 "If non-zero, the nbd driver will repeat connection " 1842 "attempts until successful or until @open-timeout seconds " 1843 "have elapsed. Default 0", 1844 }, 1845 { /* end of list */ } 1846 }, 1847 }; 1848 1849 static int nbd_process_options(BlockDriverState *bs, QDict *options, 1850 Error **errp) 1851 { 1852 BDRVNBDState *s = bs->opaque; 1853 QemuOpts *opts; 1854 int ret = -EINVAL; 1855 1856 opts = qemu_opts_create(&nbd_runtime_opts, NULL, 0, &error_abort); 1857 if (!qemu_opts_absorb_qdict(opts, options, errp)) { 1858 goto error; 1859 } 1860 1861 /* Translate @host, @port, and @path to a SocketAddress */ 1862 if (!nbd_process_legacy_socket_options(options, opts, errp)) { 1863 goto error; 1864 } 1865 1866 /* Pop the config into our state object. Exit if invalid. */ 1867 s->saddr = nbd_config(s, options, errp); 1868 if (!s->saddr) { 1869 goto error; 1870 } 1871 1872 s->export = g_strdup(qemu_opt_get(opts, "export")); 1873 if (s->export && strlen(s->export) > NBD_MAX_STRING_SIZE) { 1874 error_setg(errp, "export name too long to send to server"); 1875 goto error; 1876 } 1877 1878 s->tlscredsid = g_strdup(qemu_opt_get(opts, "tls-creds")); 1879 if (s->tlscredsid) { 1880 s->tlscreds = nbd_get_tls_creds(s->tlscredsid, errp); 1881 if (!s->tlscreds) { 1882 goto error; 1883 } 1884 1885 s->tlshostname = g_strdup(qemu_opt_get(opts, "tls-hostname")); 1886 if (!s->tlshostname && 1887 s->saddr->type == SOCKET_ADDRESS_TYPE_INET) { 1888 s->tlshostname = g_strdup(s->saddr->u.inet.host); 1889 } 1890 } 1891 1892 s->x_dirty_bitmap = g_strdup(qemu_opt_get(opts, "x-dirty-bitmap")); 1893 if (s->x_dirty_bitmap && strlen(s->x_dirty_bitmap) > NBD_MAX_STRING_SIZE) { 1894 error_setg(errp, "x-dirty-bitmap query too long to send to server"); 1895 goto error; 1896 } 1897 1898 s->reconnect_delay = qemu_opt_get_number(opts, "reconnect-delay", 0); 1899 s->open_timeout = qemu_opt_get_number(opts, "open-timeout", 0); 1900 1901 ret = 0; 1902 1903 error: 1904 qemu_opts_del(opts); 1905 return ret; 1906 } 1907 1908 static int nbd_open(BlockDriverState *bs, QDict *options, int flags, 1909 Error **errp) 1910 { 1911 int ret; 1912 BDRVNBDState *s = (BDRVNBDState *)bs->opaque; 1913 1914 s->bs = bs; 1915 qemu_mutex_init(&s->requests_lock); 1916 qemu_co_queue_init(&s->free_sema); 1917 qemu_co_mutex_init(&s->send_mutex); 1918 qemu_co_mutex_init(&s->receive_mutex); 1919 1920 if (!yank_register_instance(BLOCKDEV_YANK_INSTANCE(bs->node_name), errp)) { 1921 return -EEXIST; 1922 } 1923 1924 ret = nbd_process_options(bs, options, errp); 1925 if (ret < 0) { 1926 goto fail; 1927 } 1928 1929 s->conn = nbd_client_connection_new(s->saddr, true, s->export, 1930 s->x_dirty_bitmap, s->tlscreds, 1931 s->tlshostname); 1932 1933 if (s->open_timeout) { 1934 nbd_client_connection_enable_retry(s->conn); 1935 open_timer_init(s, qemu_clock_get_ns(QEMU_CLOCK_REALTIME) + 1936 s->open_timeout * NANOSECONDS_PER_SECOND); 1937 } 1938 1939 s->state = NBD_CLIENT_CONNECTING_WAIT; 1940 ret = nbd_do_establish_connection(bs, true, errp); 1941 if (ret < 0) { 1942 goto fail; 1943 } 1944 1945 /* 1946 * The connect attempt is done, so we no longer need this timer. 1947 * Delete it, because we do not want it to be around when this node 1948 * is drained or closed. 1949 */ 1950 open_timer_del(s); 1951 1952 nbd_client_connection_enable_retry(s->conn); 1953 1954 return 0; 1955 1956 fail: 1957 open_timer_del(s); 1958 nbd_clear_bdrvstate(bs); 1959 return ret; 1960 } 1961 1962 static void nbd_refresh_limits(BlockDriverState *bs, Error **errp) 1963 { 1964 BDRVNBDState *s = (BDRVNBDState *)bs->opaque; 1965 uint32_t min = s->info.min_block; 1966 uint32_t max = MIN_NON_ZERO(NBD_MAX_BUFFER_SIZE, s->info.max_block); 1967 1968 /* 1969 * If the server did not advertise an alignment: 1970 * - a size that is not sector-aligned implies that an alignment 1971 * of 1 can be used to access those tail bytes 1972 * - advertisement of block status requires an alignment of 1, so 1973 * that we don't violate block layer constraints that block 1974 * status is always aligned (as we can't control whether the 1975 * server will report sub-sector extents, such as a hole at EOF 1976 * on an unaligned POSIX file) 1977 * - otherwise, assume the server is so old that we are safer avoiding 1978 * sub-sector requests 1979 */ 1980 if (!min) { 1981 min = (!QEMU_IS_ALIGNED(s->info.size, BDRV_SECTOR_SIZE) || 1982 s->info.base_allocation) ? 1 : BDRV_SECTOR_SIZE; 1983 } 1984 1985 bs->bl.request_alignment = min; 1986 bs->bl.max_pdiscard = QEMU_ALIGN_DOWN(INT_MAX, min); 1987 bs->bl.max_pwrite_zeroes = max; 1988 bs->bl.max_transfer = max; 1989 1990 /* 1991 * Assume that if the server supports extended headers, it also 1992 * supports unlimited size zero and trim commands. 1993 */ 1994 if (s->info.mode >= NBD_MODE_EXTENDED) { 1995 bs->bl.max_pdiscard = bs->bl.max_pwrite_zeroes = 0; 1996 } 1997 1998 if (s->info.opt_block && 1999 s->info.opt_block > bs->bl.opt_transfer) { 2000 bs->bl.opt_transfer = s->info.opt_block; 2001 } 2002 } 2003 2004 static void nbd_close(BlockDriverState *bs) 2005 { 2006 nbd_client_close(bs); 2007 nbd_clear_bdrvstate(bs); 2008 } 2009 2010 /* 2011 * NBD cannot truncate, but if the caller asks to truncate to the same size, or 2012 * to a smaller size with exact=false, there is no reason to fail the 2013 * operation. 2014 * 2015 * Preallocation mode is ignored since it does not seems useful to fail when 2016 * we never change anything. 2017 */ 2018 static int coroutine_fn nbd_co_truncate(BlockDriverState *bs, int64_t offset, 2019 bool exact, PreallocMode prealloc, 2020 BdrvRequestFlags flags, Error **errp) 2021 { 2022 BDRVNBDState *s = bs->opaque; 2023 2024 if (offset != s->info.size && exact) { 2025 error_setg(errp, "Cannot resize NBD nodes"); 2026 return -ENOTSUP; 2027 } 2028 2029 if (offset > s->info.size) { 2030 error_setg(errp, "Cannot grow NBD nodes"); 2031 return -EINVAL; 2032 } 2033 2034 return 0; 2035 } 2036 2037 static int64_t coroutine_fn nbd_co_getlength(BlockDriverState *bs) 2038 { 2039 BDRVNBDState *s = bs->opaque; 2040 2041 return s->info.size; 2042 } 2043 2044 static void nbd_refresh_filename(BlockDriverState *bs) 2045 { 2046 BDRVNBDState *s = bs->opaque; 2047 const char *host = NULL, *port = NULL, *path = NULL; 2048 size_t len = 0; 2049 2050 if (s->saddr->type == SOCKET_ADDRESS_TYPE_INET) { 2051 const InetSocketAddress *inet = &s->saddr->u.inet; 2052 if (!inet->has_ipv4 && !inet->has_ipv6 && !inet->has_to) { 2053 host = inet->host; 2054 port = inet->port; 2055 } 2056 } else if (s->saddr->type == SOCKET_ADDRESS_TYPE_UNIX) { 2057 path = s->saddr->u.q_unix.path; 2058 } /* else can't represent as pseudo-filename */ 2059 2060 if (path && s->export) { 2061 len = snprintf(bs->exact_filename, sizeof(bs->exact_filename), 2062 "nbd+unix:///%s?socket=%s", s->export, path); 2063 } else if (path && !s->export) { 2064 len = snprintf(bs->exact_filename, sizeof(bs->exact_filename), 2065 "nbd+unix://?socket=%s", path); 2066 } else if (host && s->export) { 2067 len = snprintf(bs->exact_filename, sizeof(bs->exact_filename), 2068 "nbd://%s:%s/%s", host, port, s->export); 2069 } else if (host && !s->export) { 2070 len = snprintf(bs->exact_filename, sizeof(bs->exact_filename), 2071 "nbd://%s:%s", host, port); 2072 } 2073 if (len >= sizeof(bs->exact_filename)) { 2074 /* Name is too long to represent exactly, so leave it empty. */ 2075 bs->exact_filename[0] = '\0'; 2076 } 2077 } 2078 2079 static char *nbd_dirname(BlockDriverState *bs, Error **errp) 2080 { 2081 /* The generic bdrv_dirname() implementation is able to work out some 2082 * directory name for NBD nodes, but that would be wrong. So far there is no 2083 * specification for how "export paths" would work, so NBD does not have 2084 * directory names. */ 2085 error_setg(errp, "Cannot generate a base directory for NBD nodes"); 2086 return NULL; 2087 } 2088 2089 static const char *const nbd_strong_runtime_opts[] = { 2090 "path", 2091 "host", 2092 "port", 2093 "export", 2094 "tls-creds", 2095 "tls-hostname", 2096 "server.", 2097 2098 NULL 2099 }; 2100 2101 static void nbd_cancel_in_flight(BlockDriverState *bs) 2102 { 2103 BDRVNBDState *s = (BDRVNBDState *)bs->opaque; 2104 2105 reconnect_delay_timer_del(s); 2106 2107 qemu_mutex_lock(&s->requests_lock); 2108 if (s->state == NBD_CLIENT_CONNECTING_WAIT) { 2109 s->state = NBD_CLIENT_CONNECTING_NOWAIT; 2110 } 2111 qemu_mutex_unlock(&s->requests_lock); 2112 2113 nbd_co_establish_connection_cancel(s->conn); 2114 } 2115 2116 static void nbd_attach_aio_context(BlockDriverState *bs, 2117 AioContext *new_context) 2118 { 2119 BDRVNBDState *s = bs->opaque; 2120 2121 /* The open_timer is used only during nbd_open() */ 2122 assert(!s->open_timer); 2123 2124 /* 2125 * The reconnect_delay_timer is scheduled in I/O paths when the 2126 * connection is lost, to cancel the reconnection attempt after a 2127 * given time. Once this attempt is done (successfully or not), 2128 * nbd_reconnect_attempt() ensures the timer is deleted before the 2129 * respective I/O request is resumed. 2130 * Since the AioContext can only be changed when a node is drained, 2131 * the reconnect_delay_timer cannot be active here. 2132 */ 2133 assert(!s->reconnect_delay_timer); 2134 } 2135 2136 static void nbd_detach_aio_context(BlockDriverState *bs) 2137 { 2138 BDRVNBDState *s = bs->opaque; 2139 2140 assert(!s->open_timer); 2141 assert(!s->reconnect_delay_timer); 2142 } 2143 2144 static BlockDriver bdrv_nbd = { 2145 .format_name = "nbd", 2146 .protocol_name = "nbd", 2147 .instance_size = sizeof(BDRVNBDState), 2148 .bdrv_parse_filename = nbd_parse_filename, 2149 .bdrv_co_create_opts = bdrv_co_create_opts_simple, 2150 .create_opts = &bdrv_create_opts_simple, 2151 .bdrv_file_open = nbd_open, 2152 .bdrv_reopen_prepare = nbd_client_reopen_prepare, 2153 .bdrv_co_preadv = nbd_client_co_preadv, 2154 .bdrv_co_pwritev = nbd_client_co_pwritev, 2155 .bdrv_co_pwrite_zeroes = nbd_client_co_pwrite_zeroes, 2156 .bdrv_close = nbd_close, 2157 .bdrv_co_flush_to_os = nbd_client_co_flush, 2158 .bdrv_co_pdiscard = nbd_client_co_pdiscard, 2159 .bdrv_refresh_limits = nbd_refresh_limits, 2160 .bdrv_co_truncate = nbd_co_truncate, 2161 .bdrv_co_getlength = nbd_co_getlength, 2162 .bdrv_refresh_filename = nbd_refresh_filename, 2163 .bdrv_co_block_status = nbd_client_co_block_status, 2164 .bdrv_dirname = nbd_dirname, 2165 .strong_runtime_opts = nbd_strong_runtime_opts, 2166 .bdrv_cancel_in_flight = nbd_cancel_in_flight, 2167 2168 .bdrv_attach_aio_context = nbd_attach_aio_context, 2169 .bdrv_detach_aio_context = nbd_detach_aio_context, 2170 }; 2171 2172 static BlockDriver bdrv_nbd_tcp = { 2173 .format_name = "nbd", 2174 .protocol_name = "nbd+tcp", 2175 .instance_size = sizeof(BDRVNBDState), 2176 .bdrv_parse_filename = nbd_parse_filename, 2177 .bdrv_co_create_opts = bdrv_co_create_opts_simple, 2178 .create_opts = &bdrv_create_opts_simple, 2179 .bdrv_file_open = nbd_open, 2180 .bdrv_reopen_prepare = nbd_client_reopen_prepare, 2181 .bdrv_co_preadv = nbd_client_co_preadv, 2182 .bdrv_co_pwritev = nbd_client_co_pwritev, 2183 .bdrv_co_pwrite_zeroes = nbd_client_co_pwrite_zeroes, 2184 .bdrv_close = nbd_close, 2185 .bdrv_co_flush_to_os = nbd_client_co_flush, 2186 .bdrv_co_pdiscard = nbd_client_co_pdiscard, 2187 .bdrv_refresh_limits = nbd_refresh_limits, 2188 .bdrv_co_truncate = nbd_co_truncate, 2189 .bdrv_co_getlength = nbd_co_getlength, 2190 .bdrv_refresh_filename = nbd_refresh_filename, 2191 .bdrv_co_block_status = nbd_client_co_block_status, 2192 .bdrv_dirname = nbd_dirname, 2193 .strong_runtime_opts = nbd_strong_runtime_opts, 2194 .bdrv_cancel_in_flight = nbd_cancel_in_flight, 2195 2196 .bdrv_attach_aio_context = nbd_attach_aio_context, 2197 .bdrv_detach_aio_context = nbd_detach_aio_context, 2198 }; 2199 2200 static BlockDriver bdrv_nbd_unix = { 2201 .format_name = "nbd", 2202 .protocol_name = "nbd+unix", 2203 .instance_size = sizeof(BDRVNBDState), 2204 .bdrv_parse_filename = nbd_parse_filename, 2205 .bdrv_co_create_opts = bdrv_co_create_opts_simple, 2206 .create_opts = &bdrv_create_opts_simple, 2207 .bdrv_file_open = nbd_open, 2208 .bdrv_reopen_prepare = nbd_client_reopen_prepare, 2209 .bdrv_co_preadv = nbd_client_co_preadv, 2210 .bdrv_co_pwritev = nbd_client_co_pwritev, 2211 .bdrv_co_pwrite_zeroes = nbd_client_co_pwrite_zeroes, 2212 .bdrv_close = nbd_close, 2213 .bdrv_co_flush_to_os = nbd_client_co_flush, 2214 .bdrv_co_pdiscard = nbd_client_co_pdiscard, 2215 .bdrv_refresh_limits = nbd_refresh_limits, 2216 .bdrv_co_truncate = nbd_co_truncate, 2217 .bdrv_co_getlength = nbd_co_getlength, 2218 .bdrv_refresh_filename = nbd_refresh_filename, 2219 .bdrv_co_block_status = nbd_client_co_block_status, 2220 .bdrv_dirname = nbd_dirname, 2221 .strong_runtime_opts = nbd_strong_runtime_opts, 2222 .bdrv_cancel_in_flight = nbd_cancel_in_flight, 2223 2224 .bdrv_attach_aio_context = nbd_attach_aio_context, 2225 .bdrv_detach_aio_context = nbd_detach_aio_context, 2226 }; 2227 2228 static void bdrv_nbd_init(void) 2229 { 2230 bdrv_register(&bdrv_nbd); 2231 bdrv_register(&bdrv_nbd_tcp); 2232 bdrv_register(&bdrv_nbd_unix); 2233 } 2234 2235 block_init(bdrv_nbd_init); 2236