block/nbd-client.c

   1 /*
   2  * QEMU Block driver for  NBD
   3  *
   4  * Copyright (C) 2016 Red Hat, Inc.
   5  * Copyright (C) 2008 Bull S.A.S.
   6  *     Author: Laurent Vivier <Laurent.Vivier@bull.net>
   7  *
   8  * Some parts:
   9  *    Copyright (C) 2007 Anthony Liguori <anthony@codemonkey.ws>
  10  *
  11  * Permission is hereby granted, free of charge, to any person obtaining a copy
  12  * of this software and associated documentation files (the "Software"), to deal
  13  * in the Software without restriction, including without limitation the rights
  14  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
  15  * copies of the Software, and to permit persons to whom the Software is
  16  * furnished to do so, subject to the following conditions:
  17  *
  18  * The above copyright notice and this permission notice shall be included in
  19  * all copies or substantial portions of the Software.
  20  *
  21  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  22  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  23  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
  24  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  25  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  26  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
  27  * THE SOFTWARE.
  28  */
  29
  30 #include "qemu/osdep.h"
  31 #include "qapi/error.h"
  32 #include "nbd-client.h"
  33
  34 #define HANDLE_TO_INDEX(bs, handle) ((handle) ^ (uint64_t)(intptr_t)(bs))
  35 #define INDEX_TO_HANDLE(bs, index)  ((index)  ^ (uint64_t)(intptr_t)(bs))
  36
  37 static void nbd_recv_coroutines_wake_all(NBDClientSession *s)
  38 {
  39     int i;
  40
  41     for (i = 0; i < MAX_NBD_REQUESTS; i++) {
  42         NBDClientRequest *req = &s->requests[i];
  43
  44         if (req->coroutine && req->receiving) {
  45             aio_co_wake(req->coroutine);
  46         }
  47     }
  48 }
  49
  50 static void nbd_teardown_connection(BlockDriverState *bs)
  51 {
  52     NBDClientSession *client = nbd_get_client_session(bs);
  53
  54     if (!client->ioc) { /* Already closed */
  55         return;
  56     }
  57
  58     /* finish any pending coroutines */
  59     qio_channel_shutdown(client->ioc,
  60                          QIO_CHANNEL_SHUTDOWN_BOTH,
  61                          NULL);
  62     BDRV_POLL_WHILE(bs, client->read_reply_co);
  63
  64     nbd_client_detach_aio_context(bs);
  65     object_unref(OBJECT(client->sioc));
  66     client->sioc = NULL;
  67     object_unref(OBJECT(client->ioc));
  68     client->ioc = NULL;
  69 }
  70
  71 static coroutine_fn void nbd_read_reply_entry(void *opaque)
  72 {
  73     NBDClientSession *s = opaque;
  74     uint64_t i;
  75     int ret = 0;
  76     Error *local_err = NULL;
  77
  78     while (!s->quit) {
  79         assert(s->reply.handle == 0);
  80         ret = nbd_receive_reply(s->ioc, &s->reply, &local_err);
  81         if (ret < 0) {
  82             error_report_err(local_err);
  83         }
  84         if (ret <= 0) {
  85             break;
  86         }
  87
  88         /* There's no need for a mutex on the receive side, because the
  89          * handler acts as a synchronization point and ensures that only
  90          * one coroutine is called until the reply finishes.
  91          */
  92         i = HANDLE_TO_INDEX(s, s->reply.handle);
  93         if (i >= MAX_NBD_REQUESTS ||
  94             !s->requests[i].coroutine ||
  95             !s->requests[i].receiving ||
  96             (nbd_reply_is_structured(&s->reply) && !s->info.structured_reply))
  97         {
  98             break;
  99         }
 100
 101         /* We're woken up again by the request itself.  Note that there
 102          * is no race between yielding and reentering read_reply_co.  This
 103          * is because:
 104          *
 105          * - if the request runs on the same AioContext, it is only
 106          *   entered after we yield
 107          *
 108          * - if the request runs on a different AioContext, reentering
 109          *   read_reply_co happens through a bottom half, which can only
 110          *   run after we yield.
 111          */
 112         aio_co_wake(s->requests[i].coroutine);
 113         qemu_coroutine_yield();
 114     }
 115
 116     s->quit = true;
 117     nbd_recv_coroutines_wake_all(s);
 118     s->read_reply_co = NULL;
 119 }
 120
 121 static int nbd_co_send_request(BlockDriverState *bs,
 122                                NBDRequest *request,
 123                                QEMUIOVector *qiov)
 124 {
 125     NBDClientSession *s = nbd_get_client_session(bs);
 126     int rc, i;
 127
 128     qemu_co_mutex_lock(&s->send_mutex);
 129     while (s->in_flight == MAX_NBD_REQUESTS) {
 130         qemu_co_queue_wait(&s->free_sema, &s->send_mutex);
 131     }
 132     s->in_flight++;
 133
 134     for (i = 0; i < MAX_NBD_REQUESTS; i++) {
 135         if (s->requests[i].coroutine == NULL) {
 136             break;
 137         }
 138     }
 139
 140     g_assert(qemu_in_coroutine());
 141     assert(i < MAX_NBD_REQUESTS);
 142
 143     s->requests[i].coroutine = qemu_coroutine_self();
 144     s->requests[i].offset = request->from;
 145     s->requests[i].receiving = false;
 146
 147     request->handle = INDEX_TO_HANDLE(s, i);
 148
 149     if (s->quit) {
 150         rc = -EIO;
 151         goto err;
 152     }
 153     if (!s->ioc) {
 154         rc = -EPIPE;
 155         goto err;
 156     }
 157
 158     if (qiov) {
 159         qio_channel_set_cork(s->ioc, true);
 160         rc = nbd_send_request(s->ioc, request);
 161         if (rc >= 0 && !s->quit) {
 162             if (qio_channel_writev_all(s->ioc, qiov->iov, qiov->niov,
 163                                        NULL) < 0) {
 164                 rc = -EIO;
 165             }
 166         } else if (rc >= 0) {
 167             rc = -EIO;
 168         }
 169         qio_channel_set_cork(s->ioc, false);
 170     } else {
 171         rc = nbd_send_request(s->ioc, request);
 172     }
 173
 174 err:
 175     if (rc < 0) {
 176         s->quit = true;
 177         s->requests[i].coroutine = NULL;
 178         s->in_flight--;
 179         qemu_co_queue_next(&s->free_sema);
 180     }
 181     qemu_co_mutex_unlock(&s->send_mutex);
 182     return rc;
 183 }
 184
 185 static inline uint16_t payload_advance16(uint8_t **payload)
 186 {
 187     *payload += 2;
 188     return lduw_be_p(*payload - 2);
 189 }
 190
 191 static inline uint32_t payload_advance32(uint8_t **payload)
 192 {
 193     *payload += 4;
 194     return ldl_be_p(*payload - 4);
 195 }
 196
 197 static inline uint64_t payload_advance64(uint8_t **payload)
 198 {
 199     *payload += 8;
 200     return ldq_be_p(*payload - 8);
 201 }
 202
 203 static int nbd_parse_offset_hole_payload(NBDStructuredReplyChunk *chunk,
 204                                          uint8_t *payload, uint64_t orig_offset,
 205                                          QEMUIOVector *qiov, Error **errp)
 206 {
 207     uint64_t offset;
 208     uint32_t hole_size;
 209
 210     if (chunk->length != sizeof(offset) + sizeof(hole_size)) {
 211         error_setg(errp, "Protocol error: invalid payload for "
 212                          "NBD_REPLY_TYPE_OFFSET_HOLE");
 213         return -EINVAL;
 214     }
 215
 216     offset = payload_advance64(&payload);
 217     hole_size = payload_advance32(&payload);
 218
 219     if (offset < orig_offset || hole_size > qiov->size ||
 220         offset > orig_offset + qiov->size - hole_size) {
 221         error_setg(errp, "Protocol error: server sent chunk exceeding requested"
 222                          " region");
 223         return -EINVAL;
 224     }
 225
 226     qemu_iovec_memset(qiov, offset - orig_offset, 0, hole_size);
 227
 228     return 0;
 229 }
 230
 231 /* nbd_parse_error_payload
 232  * on success @errp contains message describing nbd error reply
 233  */
 234 static int nbd_parse_error_payload(NBDStructuredReplyChunk *chunk,
 235                                    uint8_t *payload, int *request_ret,
 236                                    Error **errp)
 237 {
 238     uint32_t error;
 239     uint16_t message_size;
 240
 241     assert(chunk->type & (1 << 15));
 242
 243     if (chunk->length < sizeof(error) + sizeof(message_size)) {
 244         error_setg(errp,
 245                    "Protocol error: invalid payload for structured error");
 246         return -EINVAL;
 247     }
 248
 249     error = nbd_errno_to_system_errno(payload_advance32(&payload));
 250     if (error == 0) {
 251         error_setg(errp, "Protocol error: server sent structured error chunk"
 252                          "with error = 0");
 253         return -EINVAL;
 254     }
 255
 256     *request_ret = -error;
 257     message_size = payload_advance16(&payload);
 258
 259     if (message_size > chunk->length - sizeof(error) - sizeof(message_size)) {
 260         error_setg(errp, "Protocol error: server sent structured error chunk"
 261                          "with incorrect message size");
 262         return -EINVAL;
 263     }
 264
 265     /* TODO: Add a trace point to mention the server complaint */
 266
 267     /* TODO handle ERROR_OFFSET */
 268
 269     return 0;
 270 }
 271
 272 static int nbd_co_receive_offset_data_payload(NBDClientSession *s,
 273                                               uint64_t orig_offset,
 274                                               QEMUIOVector *qiov, Error **errp)
 275 {
 276     QEMUIOVector sub_qiov;
 277     uint64_t offset;
 278     size_t data_size;
 279     int ret;
 280     NBDStructuredReplyChunk *chunk = &s->reply.structured;
 281
 282     assert(nbd_reply_is_structured(&s->reply));
 283
 284     if (chunk->length < sizeof(offset)) {
 285         error_setg(errp, "Protocol error: invalid payload for "
 286                          "NBD_REPLY_TYPE_OFFSET_DATA");
 287         return -EINVAL;
 288     }
 289
 290     if (nbd_read(s->ioc, &offset, sizeof(offset), errp) < 0) {
 291         return -EIO;
 292     }
 293     be64_to_cpus(&offset);
 294
 295     data_size = chunk->length - sizeof(offset);
 296     if (offset < orig_offset || data_size > qiov->size ||
 297         offset > orig_offset + qiov->size - data_size) {
 298         error_setg(errp, "Protocol error: server sent chunk exceeding requested"
 299                          " region");
 300         return -EINVAL;
 301     }
 302
 303     qemu_iovec_init(&sub_qiov, qiov->niov);
 304     qemu_iovec_concat(&sub_qiov, qiov, offset - orig_offset, data_size);
 305     ret = qio_channel_readv_all(s->ioc, sub_qiov.iov, sub_qiov.niov, errp);
 306     qemu_iovec_destroy(&sub_qiov);
 307
 308     return ret < 0 ? -EIO : 0;
 309 }
 310
 311 #define NBD_MAX_MALLOC_PAYLOAD 1000
 312 /* nbd_co_receive_structured_payload
 313  */
 314 static coroutine_fn int nbd_co_receive_structured_payload(
 315         NBDClientSession *s, void **payload, Error **errp)
 316 {
 317     int ret;
 318     uint32_t len;
 319
 320     assert(nbd_reply_is_structured(&s->reply));
 321
 322     len = s->reply.structured.length;
 323
 324     if (len == 0) {
 325         return 0;
 326     }
 327
 328     if (payload == NULL) {
 329         error_setg(errp, "Unexpected structured payload");
 330         return -EINVAL;
 331     }
 332
 333     if (len > NBD_MAX_MALLOC_PAYLOAD) {
 334         error_setg(errp, "Payload too large");
 335         return -EINVAL;
 336     }
 337
 338     *payload = g_new(char, len);
 339     ret = nbd_read(s->ioc, *payload, len, errp);
 340     if (ret < 0) {
 341         g_free(*payload);
 342         *payload = NULL;
 343         return ret;
 344     }
 345
 346     return 0;
 347 }
 348
 349 /* nbd_co_do_receive_one_chunk
 350  * for simple reply:
 351  *   set request_ret to received reply error
 352  *   if qiov is not NULL: read payload to @qiov
 353  * for structured reply chunk:
 354  *   if error chunk: read payload, set @request_ret, do not set @payload
 355  *   else if offset_data chunk: read payload data to @qiov, do not set @payload
 356  *   else: read payload to @payload
 357  *
 358  * If function fails, @errp contains corresponding error message, and the
 359  * connection with the server is suspect.  If it returns 0, then the
 360  * transaction succeeded (although @request_ret may be a negative errno
 361  * corresponding to the server's error reply), and errp is unchanged.
 362  */
 363 static coroutine_fn int nbd_co_do_receive_one_chunk(
 364         NBDClientSession *s, uint64_t handle, bool only_structured,
 365         int *request_ret, QEMUIOVector *qiov, void **payload, Error **errp)
 366 {
 367     int ret;
 368     int i = HANDLE_TO_INDEX(s, handle);
 369     void *local_payload = NULL;
 370     NBDStructuredReplyChunk *chunk;
 371
 372     if (payload) {
 373         *payload = NULL;
 374     }
 375     *request_ret = 0;
 376
 377     /* Wait until we're woken up by nbd_read_reply_entry.  */
 378     s->requests[i].receiving = true;
 379     qemu_coroutine_yield();
 380     s->requests[i].receiving = false;
 381     if (!s->ioc || s->quit) {
 382         error_setg(errp, "Connection closed");
 383         return -EIO;
 384     }
 385
 386     assert(s->reply.handle == handle);
 387
 388     if (nbd_reply_is_simple(&s->reply)) {
 389         if (only_structured) {
 390             error_setg(errp, "Protocol error: simple reply when structured "
 391                              "reply chunk was expected");
 392             return -EINVAL;
 393         }
 394
 395         *request_ret = -nbd_errno_to_system_errno(s->reply.simple.error);
 396         if (*request_ret < 0 || !qiov) {
 397             return 0;
 398         }
 399
 400         return qio_channel_readv_all(s->ioc, qiov->iov, qiov->niov,
 401                                      errp) < 0 ? -EIO : 0;
 402     }
 403
 404     /* handle structured reply chunk */
 405     assert(s->info.structured_reply);
 406     chunk = &s->reply.structured;
 407
 408     if (chunk->type == NBD_REPLY_TYPE_NONE) {
 409         if (!(chunk->flags & NBD_REPLY_FLAG_DONE)) {
 410             error_setg(errp, "Protocol error: NBD_REPLY_TYPE_NONE chunk without"
 411                              "NBD_REPLY_FLAG_DONE flag set");
 412             return -EINVAL;
 413         }
 414         return 0;
 415     }
 416
 417     if (chunk->type == NBD_REPLY_TYPE_OFFSET_DATA) {
 418         if (!qiov) {
 419             error_setg(errp, "Unexpected NBD_REPLY_TYPE_OFFSET_DATA chunk");
 420             return -EINVAL;
 421         }
 422
 423         return nbd_co_receive_offset_data_payload(s, s->requests[i].offset,
 424                                                   qiov, errp);
 425     }
 426
 427     if (nbd_reply_type_is_error(chunk->type)) {
 428         payload = &local_payload;
 429     }
 430
 431     ret = nbd_co_receive_structured_payload(s, payload, errp);
 432     if (ret < 0) {
 433         return ret;
 434     }
 435
 436     if (nbd_reply_type_is_error(chunk->type)) {
 437         ret = nbd_parse_error_payload(chunk, local_payload, request_ret, errp);
 438         g_free(local_payload);
 439         return ret;
 440     }
 441
 442     return 0;
 443 }
 444
 445 /* nbd_co_receive_one_chunk
 446  * Read reply, wake up read_reply_co and set s->quit if needed.
 447  * Return value is a fatal error code or normal nbd reply error code
 448  */
 449 static coroutine_fn int nbd_co_receive_one_chunk(
 450         NBDClientSession *s, uint64_t handle, bool only_structured,
 451         QEMUIOVector *qiov, NBDReply *reply, void **payload, Error **errp)
 452 {
 453     int request_ret;
 454     int ret = nbd_co_do_receive_one_chunk(s, handle, only_structured,
 455                                           &request_ret, qiov, payload, errp);
 456
 457     if (ret < 0) {
 458         s->quit = true;
 459     } else {
 460         /* For assert at loop start in nbd_read_reply_entry */
 461         if (reply) {
 462             *reply = s->reply;
 463         }
 464         s->reply.handle = 0;
 465         ret = request_ret;
 466     }
 467
 468     if (s->read_reply_co) {
 469         aio_co_wake(s->read_reply_co);
 470     }
 471
 472     return ret;
 473 }
 474
 475 typedef struct NBDReplyChunkIter {
 476     int ret;
 477     Error *err;
 478     bool done, only_structured;
 479 } NBDReplyChunkIter;
 480
 481 static void nbd_iter_error(NBDReplyChunkIter *iter, bool fatal,
 482                            int ret, Error **local_err)
 483 {
 484     assert(ret < 0);
 485
 486     if (fatal || iter->ret == 0) {
 487         if (iter->ret != 0) {
 488             error_free(iter->err);
 489             iter->err = NULL;
 490         }
 491         iter->ret = ret;
 492         error_propagate(&iter->err, *local_err);
 493     } else {
 494         error_free(*local_err);
 495     }
 496
 497     *local_err = NULL;
 498 }
 499
 500 /* NBD_FOREACH_REPLY_CHUNK
 501  */
 502 #define NBD_FOREACH_REPLY_CHUNK(s, iter, handle, structured, \
 503                                 qiov, reply, payload) \
 504     for (iter = (NBDReplyChunkIter) { .only_structured = structured }; \
 505          nbd_reply_chunk_iter_receive(s, &iter, handle, qiov, reply, payload);)
 506
 507 /* nbd_reply_chunk_iter_receive
 508  */
 509 static bool nbd_reply_chunk_iter_receive(NBDClientSession *s,
 510                                          NBDReplyChunkIter *iter,
 511                                          uint64_t handle,
 512                                          QEMUIOVector *qiov, NBDReply *reply,
 513                                          void **payload)
 514 {
 515     int ret;
 516     NBDReply local_reply;
 517     NBDStructuredReplyChunk *chunk;
 518     Error *local_err = NULL;
 519     if (s->quit) {
 520         error_setg(&local_err, "Connection closed");
 521         nbd_iter_error(iter, true, -EIO, &local_err);
 522         goto break_loop;
 523     }
 524
 525     if (iter->done) {
 526         /* Previous iteration was last. */
 527         goto break_loop;
 528     }
 529
 530     if (reply == NULL) {
 531         reply = &local_reply;
 532     }
 533
 534     ret = nbd_co_receive_one_chunk(s, handle, iter->only_structured,
 535                                    qiov, reply, payload, &local_err);
 536     if (ret < 0) {
 537         /* If it is a fatal error s->quit is set by nbd_co_receive_one_chunk */
 538         nbd_iter_error(iter, s->quit, ret, &local_err);
 539     }
 540
 541     /* Do not execute the body of NBD_FOREACH_REPLY_CHUNK for simple reply. */
 542     if (nbd_reply_is_simple(&s->reply) || s->quit) {
 543         goto break_loop;
 544     }
 545
 546     chunk = &reply->structured;
 547     iter->only_structured = true;
 548
 549     if (chunk->type == NBD_REPLY_TYPE_NONE) {
 550         /* NBD_REPLY_FLAG_DONE is already checked in nbd_co_receive_one_chunk */
 551         assert(chunk->flags & NBD_REPLY_FLAG_DONE);
 552         goto break_loop;
 553     }
 554
 555     if (chunk->flags & NBD_REPLY_FLAG_DONE) {
 556         /* This iteration is last. */
 557         iter->done = true;
 558     }
 559
 560     /* Execute the loop body */
 561     return true;
 562
 563 break_loop:
 564     s->requests[HANDLE_TO_INDEX(s, handle)].coroutine = NULL;
 565
 566     qemu_co_mutex_lock(&s->send_mutex);
 567     s->in_flight--;
 568     qemu_co_queue_next(&s->free_sema);
 569     qemu_co_mutex_unlock(&s->send_mutex);
 570
 571     return false;
 572 }
 573
 574 static int nbd_co_receive_return_code(NBDClientSession *s, uint64_t handle,
 575                                       Error **errp)
 576 {
 577     NBDReplyChunkIter iter;
 578
 579     NBD_FOREACH_REPLY_CHUNK(s, iter, handle, false, NULL, NULL, NULL) {
 580         /* nbd_reply_chunk_iter_receive does all the work */
 581     }
 582
 583     error_propagate(errp, iter.err);
 584     return iter.ret;
 585 }
 586
 587 static int nbd_co_receive_cmdread_reply(NBDClientSession *s, uint64_t handle,
 588                                         uint64_t offset, QEMUIOVector *qiov,
 589                                         Error **errp)
 590 {
 591     NBDReplyChunkIter iter;
 592     NBDReply reply;
 593     void *payload = NULL;
 594     Error *local_err = NULL;
 595
 596     NBD_FOREACH_REPLY_CHUNK(s, iter, handle, s->info.structured_reply,
 597                             qiov, &reply, &payload)
 598     {
 599         int ret;
 600         NBDStructuredReplyChunk *chunk = &reply.structured;
 601
 602         assert(nbd_reply_is_structured(&reply));
 603
 604         switch (chunk->type) {
 605         case NBD_REPLY_TYPE_OFFSET_DATA:
 606             /* special cased in nbd_co_receive_one_chunk, data is already
 607              * in qiov */
 608             break;
 609         case NBD_REPLY_TYPE_OFFSET_HOLE:
 610             ret = nbd_parse_offset_hole_payload(&reply.structured, payload,
 611                                                 offset, qiov, &local_err);
 612             if (ret < 0) {
 613                 s->quit = true;
 614                 nbd_iter_error(&iter, true, ret, &local_err);
 615             }
 616             break;
 617         default:
 618             if (!nbd_reply_type_is_error(chunk->type)) {
 619                 /* not allowed reply type */
 620                 s->quit = true;
 621                 error_setg(&local_err,
 622                            "Unexpected reply type: %d (%s) for CMD_READ",
 623                            chunk->type, nbd_reply_type_lookup(chunk->type));
 624                 nbd_iter_error(&iter, true, -EINVAL, &local_err);
 625             }
 626         }
 627
 628         g_free(payload);
 629         payload = NULL;
 630     }
 631
 632     error_propagate(errp, iter.err);
 633     return iter.ret;
 634 }
 635
 636 static int nbd_co_request(BlockDriverState *bs, NBDRequest *request,
 637                           QEMUIOVector *write_qiov)
 638 {
 639     int ret;
 640     Error *local_err = NULL;
 641     NBDClientSession *client = nbd_get_client_session(bs);
 642
 643     assert(request->type != NBD_CMD_READ);
 644     if (write_qiov) {
 645         assert(request->type == NBD_CMD_WRITE);
 646         assert(request->len == iov_size(write_qiov->iov, write_qiov->niov));
 647     } else {
 648         assert(request->type != NBD_CMD_WRITE);
 649     }
 650     ret = nbd_co_send_request(bs, request, write_qiov);
 651     if (ret < 0) {
 652         return ret;
 653     }
 654
 655     ret = nbd_co_receive_return_code(client, request->handle, &local_err);
 656     if (local_err) {
 657         error_report_err(local_err);
 658     }
 659     return ret;
 660 }
 661
 662 int nbd_client_co_preadv(BlockDriverState *bs, uint64_t offset,
 663                          uint64_t bytes, QEMUIOVector *qiov, int flags)
 664 {
 665     int ret;
 666     Error *local_err = NULL;
 667     NBDClientSession *client = nbd_get_client_session(bs);
 668     NBDRequest request = {
 669         .type = NBD_CMD_READ,
 670         .from = offset,
 671         .len = bytes,
 672     };
 673
 674     assert(bytes <= NBD_MAX_BUFFER_SIZE);
 675     assert(!flags);
 676
 677     ret = nbd_co_send_request(bs, &request, NULL);
 678     if (ret < 0) {
 679         return ret;
 680     }
 681
 682     ret = nbd_co_receive_cmdread_reply(client, request.handle, offset, qiov,
 683                                        &local_err);
 684     if (ret < 0) {
 685         error_report_err(local_err);
 686     }
 687     return ret;
 688 }
 689
 690 int nbd_client_co_pwritev(BlockDriverState *bs, uint64_t offset,
 691                           uint64_t bytes, QEMUIOVector *qiov, int flags)
 692 {
 693     NBDClientSession *client = nbd_get_client_session(bs);
 694     NBDRequest request = {
 695         .type = NBD_CMD_WRITE,
 696         .from = offset,
 697         .len = bytes,
 698     };
 699
 700     if (flags & BDRV_REQ_FUA) {
 701         assert(client->info.flags & NBD_FLAG_SEND_FUA);
 702         request.flags |= NBD_CMD_FLAG_FUA;
 703     }
 704
 705     assert(bytes <= NBD_MAX_BUFFER_SIZE);
 706
 707     return nbd_co_request(bs, &request, qiov);
 708 }
 709
 710 int nbd_client_co_pwrite_zeroes(BlockDriverState *bs, int64_t offset,
 711                                 int bytes, BdrvRequestFlags flags)
 712 {
 713     NBDClientSession *client = nbd_get_client_session(bs);
 714     NBDRequest request = {
 715         .type = NBD_CMD_WRITE_ZEROES,
 716         .from = offset,
 717         .len = bytes,
 718     };
 719
 720     if (!(client->info.flags & NBD_FLAG_SEND_WRITE_ZEROES)) {
 721         return -ENOTSUP;
 722     }
 723
 724     if (flags & BDRV_REQ_FUA) {
 725         assert(client->info.flags & NBD_FLAG_SEND_FUA);
 726         request.flags |= NBD_CMD_FLAG_FUA;
 727     }
 728     if (!(flags & BDRV_REQ_MAY_UNMAP)) {
 729         request.flags |= NBD_CMD_FLAG_NO_HOLE;
 730     }
 731
 732     return nbd_co_request(bs, &request, NULL);
 733 }
 734
 735 int nbd_client_co_flush(BlockDriverState *bs)
 736 {
 737     NBDClientSession *client = nbd_get_client_session(bs);
 738     NBDRequest request = { .type = NBD_CMD_FLUSH };
 739
 740     if (!(client->info.flags & NBD_FLAG_SEND_FLUSH)) {
 741         return 0;
 742     }
 743
 744     request.from = 0;
 745     request.len = 0;
 746
 747     return nbd_co_request(bs, &request, NULL);
 748 }
 749
 750 int nbd_client_co_pdiscard(BlockDriverState *bs, int64_t offset, int bytes)
 751 {
 752     NBDClientSession *client = nbd_get_client_session(bs);
 753     NBDRequest request = {
 754         .type = NBD_CMD_TRIM,
 755         .from = offset,
 756         .len = bytes,
 757     };
 758
 759     if (!(client->info.flags & NBD_FLAG_SEND_TRIM)) {
 760         return 0;
 761     }
 762
 763     return nbd_co_request(bs, &request, NULL);
 764 }
 765
 766 void nbd_client_detach_aio_context(BlockDriverState *bs)
 767 {
 768     NBDClientSession *client = nbd_get_client_session(bs);
 769     qio_channel_detach_aio_context(QIO_CHANNEL(client->ioc));
 770 }
 771
 772 void nbd_client_attach_aio_context(BlockDriverState *bs,
 773                                    AioContext *new_context)
 774 {
 775     NBDClientSession *client = nbd_get_client_session(bs);
 776     qio_channel_attach_aio_context(QIO_CHANNEL(client->ioc), new_context);
 777     aio_co_schedule(new_context, client->read_reply_co);
 778 }
 779
 780 void nbd_client_close(BlockDriverState *bs)
 781 {
 782     NBDClientSession *client = nbd_get_client_session(bs);
 783     NBDRequest request = { .type = NBD_CMD_DISC };
 784
 785     if (client->ioc == NULL) {
 786         return;
 787     }
 788
 789     nbd_send_request(client->ioc, &request);
 790
 791     nbd_teardown_connection(bs);
 792 }
 793
 794 int nbd_client_init(BlockDriverState *bs,
 795                     QIOChannelSocket *sioc,
 796                     const char *export,
 797                     QCryptoTLSCreds *tlscreds,
 798                     const char *hostname,
 799                     Error **errp)
 800 {
 801     NBDClientSession *client = nbd_get_client_session(bs);
 802     int ret;
 803
 804     /* NBD handshake */
 805     logout("session init %s\n", export);
 806     qio_channel_set_blocking(QIO_CHANNEL(sioc), true, NULL);
 807
 808     client->info.request_sizes = true;
 809     client->info.structured_reply = true;
 810     ret = nbd_receive_negotiate(QIO_CHANNEL(sioc), export,
 811                                 tlscreds, hostname,
 812                                 &client->ioc, &client->info, errp);
 813     if (ret < 0) {
 814         logout("Failed to negotiate with the NBD server\n");
 815         return ret;
 816     }
 817     if (client->info.flags & NBD_FLAG_SEND_FUA) {
 818         bs->supported_write_flags = BDRV_REQ_FUA;
 819         bs->supported_zero_flags |= BDRV_REQ_FUA;
 820     }
 821     if (client->info.flags & NBD_FLAG_SEND_WRITE_ZEROES) {
 822         bs->supported_zero_flags |= BDRV_REQ_MAY_UNMAP;
 823     }
 824     if (client->info.min_block > bs->bl.request_alignment) {
 825         bs->bl.request_alignment = client->info.min_block;
 826     }
 827
 828     qemu_co_mutex_init(&client->send_mutex);
 829     qemu_co_queue_init(&client->free_sema);
 830     client->sioc = sioc;
 831     object_ref(OBJECT(client->sioc));
 832
 833     if (!client->ioc) {
 834         client->ioc = QIO_CHANNEL(sioc);
 835         object_ref(OBJECT(client->ioc));
 836     }
 837
 838     /* Now that we're connected, set the socket to be non-blocking and
 839      * kick the reply mechanism.  */
 840     qio_channel_set_blocking(QIO_CHANNEL(sioc), false, NULL);
 841     client->read_reply_co = qemu_coroutine_create(nbd_read_reply_entry, client);
 842     nbd_client_attach_aio_context(bs, bdrv_get_aio_context(bs));
 843
 844     logout("Established connection with NBD server\n");
 845     return 0;
 846 }