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 (local_err) {
  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 (!hole_size || 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     /* The NBD spec requires at least one byte of payload */
 285     if (chunk->length <= sizeof(offset)) {
 286         error_setg(errp, "Protocol error: invalid payload for "
 287                          "NBD_REPLY_TYPE_OFFSET_DATA");
 288         return -EINVAL;
 289     }
 290
 291     if (nbd_read(s->ioc, &offset, sizeof(offset), errp) < 0) {
 292         return -EIO;
 293     }
 294     be64_to_cpus(&offset);
 295
 296     data_size = chunk->length - sizeof(offset);
 297     assert(data_size);
 298     if (offset < orig_offset || data_size > qiov->size ||
 299         offset > orig_offset + qiov->size - data_size) {
 300         error_setg(errp, "Protocol error: server sent chunk exceeding requested"
 301                          " region");
 302         return -EINVAL;
 303     }
 304
 305     qemu_iovec_init(&sub_qiov, qiov->niov);
 306     qemu_iovec_concat(&sub_qiov, qiov, offset - orig_offset, data_size);
 307     ret = qio_channel_readv_all(s->ioc, sub_qiov.iov, sub_qiov.niov, errp);
 308     qemu_iovec_destroy(&sub_qiov);
 309
 310     return ret < 0 ? -EIO : 0;
 311 }
 312
 313 #define NBD_MAX_MALLOC_PAYLOAD 1000
 314 /* nbd_co_receive_structured_payload
 315  */
 316 static coroutine_fn int nbd_co_receive_structured_payload(
 317         NBDClientSession *s, void **payload, Error **errp)
 318 {
 319     int ret;
 320     uint32_t len;
 321
 322     assert(nbd_reply_is_structured(&s->reply));
 323
 324     len = s->reply.structured.length;
 325
 326     if (len == 0) {
 327         return 0;
 328     }
 329
 330     if (payload == NULL) {
 331         error_setg(errp, "Unexpected structured payload");
 332         return -EINVAL;
 333     }
 334
 335     if (len > NBD_MAX_MALLOC_PAYLOAD) {
 336         error_setg(errp, "Payload too large");
 337         return -EINVAL;
 338     }
 339
 340     *payload = g_new(char, len);
 341     ret = nbd_read(s->ioc, *payload, len, errp);
 342     if (ret < 0) {
 343         g_free(*payload);
 344         *payload = NULL;
 345         return ret;
 346     }
 347
 348     return 0;
 349 }
 350
 351 /* nbd_co_do_receive_one_chunk
 352  * for simple reply:
 353  *   set request_ret to received reply error
 354  *   if qiov is not NULL: read payload to @qiov
 355  * for structured reply chunk:
 356  *   if error chunk: read payload, set @request_ret, do not set @payload
 357  *   else if offset_data chunk: read payload data to @qiov, do not set @payload
 358  *   else: read payload to @payload
 359  *
 360  * If function fails, @errp contains corresponding error message, and the
 361  * connection with the server is suspect.  If it returns 0, then the
 362  * transaction succeeded (although @request_ret may be a negative errno
 363  * corresponding to the server's error reply), and errp is unchanged.
 364  */
 365 static coroutine_fn int nbd_co_do_receive_one_chunk(
 366         NBDClientSession *s, uint64_t handle, bool only_structured,
 367         int *request_ret, QEMUIOVector *qiov, void **payload, Error **errp)
 368 {
 369     int ret;
 370     int i = HANDLE_TO_INDEX(s, handle);
 371     void *local_payload = NULL;
 372     NBDStructuredReplyChunk *chunk;
 373
 374     if (payload) {
 375         *payload = NULL;
 376     }
 377     *request_ret = 0;
 378
 379     /* Wait until we're woken up by nbd_read_reply_entry.  */
 380     s->requests[i].receiving = true;
 381     qemu_coroutine_yield();
 382     s->requests[i].receiving = false;
 383     if (!s->ioc || s->quit) {
 384         error_setg(errp, "Connection closed");
 385         return -EIO;
 386     }
 387
 388     assert(s->reply.handle == handle);
 389
 390     if (nbd_reply_is_simple(&s->reply)) {
 391         if (only_structured) {
 392             error_setg(errp, "Protocol error: simple reply when structured "
 393                              "reply chunk was expected");
 394             return -EINVAL;
 395         }
 396
 397         *request_ret = -nbd_errno_to_system_errno(s->reply.simple.error);
 398         if (*request_ret < 0 || !qiov) {
 399             return 0;
 400         }
 401
 402         return qio_channel_readv_all(s->ioc, qiov->iov, qiov->niov,
 403                                      errp) < 0 ? -EIO : 0;
 404     }
 405
 406     /* handle structured reply chunk */
 407     assert(s->info.structured_reply);
 408     chunk = &s->reply.structured;
 409
 410     if (chunk->type == NBD_REPLY_TYPE_NONE) {
 411         if (!(chunk->flags & NBD_REPLY_FLAG_DONE)) {
 412             error_setg(errp, "Protocol error: NBD_REPLY_TYPE_NONE chunk without"
 413                        " NBD_REPLY_FLAG_DONE flag set");
 414             return -EINVAL;
 415         }
 416         if (chunk->length) {
 417             error_setg(errp, "Protocol error: NBD_REPLY_TYPE_NONE chunk with"
 418                        " nonzero length");
 419             return -EINVAL;
 420         }
 421         return 0;
 422     }
 423
 424     if (chunk->type == NBD_REPLY_TYPE_OFFSET_DATA) {
 425         if (!qiov) {
 426             error_setg(errp, "Unexpected NBD_REPLY_TYPE_OFFSET_DATA chunk");
 427             return -EINVAL;
 428         }
 429
 430         return nbd_co_receive_offset_data_payload(s, s->requests[i].offset,
 431                                                   qiov, errp);
 432     }
 433
 434     if (nbd_reply_type_is_error(chunk->type)) {
 435         payload = &local_payload;
 436     }
 437
 438     ret = nbd_co_receive_structured_payload(s, payload, errp);
 439     if (ret < 0) {
 440         return ret;
 441     }
 442
 443     if (nbd_reply_type_is_error(chunk->type)) {
 444         ret = nbd_parse_error_payload(chunk, local_payload, request_ret, errp);
 445         g_free(local_payload);
 446         return ret;
 447     }
 448
 449     return 0;
 450 }
 451
 452 /* nbd_co_receive_one_chunk
 453  * Read reply, wake up read_reply_co and set s->quit if needed.
 454  * Return value is a fatal error code or normal nbd reply error code
 455  */
 456 static coroutine_fn int nbd_co_receive_one_chunk(
 457         NBDClientSession *s, uint64_t handle, bool only_structured,
 458         QEMUIOVector *qiov, NBDReply *reply, void **payload, Error **errp)
 459 {
 460     int request_ret;
 461     int ret = nbd_co_do_receive_one_chunk(s, handle, only_structured,
 462                                           &request_ret, qiov, payload, errp);
 463
 464     if (ret < 0) {
 465         s->quit = true;
 466     } else {
 467         /* For assert at loop start in nbd_read_reply_entry */
 468         if (reply) {
 469             *reply = s->reply;
 470         }
 471         s->reply.handle = 0;
 472         ret = request_ret;
 473     }
 474
 475     if (s->read_reply_co) {
 476         aio_co_wake(s->read_reply_co);
 477     }
 478
 479     return ret;
 480 }
 481
 482 typedef struct NBDReplyChunkIter {
 483     int ret;
 484     Error *err;
 485     bool done, only_structured;
 486 } NBDReplyChunkIter;
 487
 488 static void nbd_iter_error(NBDReplyChunkIter *iter, bool fatal,
 489                            int ret, Error **local_err)
 490 {
 491     assert(ret < 0);
 492
 493     if (fatal || iter->ret == 0) {
 494         if (iter->ret != 0) {
 495             error_free(iter->err);
 496             iter->err = NULL;
 497         }
 498         iter->ret = ret;
 499         error_propagate(&iter->err, *local_err);
 500     } else {
 501         error_free(*local_err);
 502     }
 503
 504     *local_err = NULL;
 505 }
 506
 507 /* NBD_FOREACH_REPLY_CHUNK
 508  */
 509 #define NBD_FOREACH_REPLY_CHUNK(s, iter, handle, structured, \
 510                                 qiov, reply, payload) \
 511     for (iter = (NBDReplyChunkIter) { .only_structured = structured }; \
 512          nbd_reply_chunk_iter_receive(s, &iter, handle, qiov, reply, payload);)
 513
 514 /* nbd_reply_chunk_iter_receive
 515  */
 516 static bool nbd_reply_chunk_iter_receive(NBDClientSession *s,
 517                                          NBDReplyChunkIter *iter,
 518                                          uint64_t handle,
 519                                          QEMUIOVector *qiov, NBDReply *reply,
 520                                          void **payload)
 521 {
 522     int ret;
 523     NBDReply local_reply;
 524     NBDStructuredReplyChunk *chunk;
 525     Error *local_err = NULL;
 526     if (s->quit) {
 527         error_setg(&local_err, "Connection closed");
 528         nbd_iter_error(iter, true, -EIO, &local_err);
 529         goto break_loop;
 530     }
 531
 532     if (iter->done) {
 533         /* Previous iteration was last. */
 534         goto break_loop;
 535     }
 536
 537     if (reply == NULL) {
 538         reply = &local_reply;
 539     }
 540
 541     ret = nbd_co_receive_one_chunk(s, handle, iter->only_structured,
 542                                    qiov, reply, payload, &local_err);
 543     if (ret < 0) {
 544         /* If it is a fatal error s->quit is set by nbd_co_receive_one_chunk */
 545         nbd_iter_error(iter, s->quit, ret, &local_err);
 546     }
 547
 548     /* Do not execute the body of NBD_FOREACH_REPLY_CHUNK for simple reply. */
 549     if (nbd_reply_is_simple(&s->reply) || s->quit) {
 550         goto break_loop;
 551     }
 552
 553     chunk = &reply->structured;
 554     iter->only_structured = true;
 555
 556     if (chunk->type == NBD_REPLY_TYPE_NONE) {
 557         /* NBD_REPLY_FLAG_DONE is already checked in nbd_co_receive_one_chunk */
 558         assert(chunk->flags & NBD_REPLY_FLAG_DONE);
 559         goto break_loop;
 560     }
 561
 562     if (chunk->flags & NBD_REPLY_FLAG_DONE) {
 563         /* This iteration is last. */
 564         iter->done = true;
 565     }
 566
 567     /* Execute the loop body */
 568     return true;
 569
 570 break_loop:
 571     s->requests[HANDLE_TO_INDEX(s, handle)].coroutine = NULL;
 572
 573     qemu_co_mutex_lock(&s->send_mutex);
 574     s->in_flight--;
 575     qemu_co_queue_next(&s->free_sema);
 576     qemu_co_mutex_unlock(&s->send_mutex);
 577
 578     return false;
 579 }
 580
 581 static int nbd_co_receive_return_code(NBDClientSession *s, uint64_t handle,
 582                                       Error **errp)
 583 {
 584     NBDReplyChunkIter iter;
 585
 586     NBD_FOREACH_REPLY_CHUNK(s, iter, handle, false, NULL, NULL, NULL) {
 587         /* nbd_reply_chunk_iter_receive does all the work */
 588     }
 589
 590     error_propagate(errp, iter.err);
 591     return iter.ret;
 592 }
 593
 594 static int nbd_co_receive_cmdread_reply(NBDClientSession *s, uint64_t handle,
 595                                         uint64_t offset, QEMUIOVector *qiov,
 596                                         Error **errp)
 597 {
 598     NBDReplyChunkIter iter;
 599     NBDReply reply;
 600     void *payload = NULL;
 601     Error *local_err = NULL;
 602
 603     NBD_FOREACH_REPLY_CHUNK(s, iter, handle, s->info.structured_reply,
 604                             qiov, &reply, &payload)
 605     {
 606         int ret;
 607         NBDStructuredReplyChunk *chunk = &reply.structured;
 608
 609         assert(nbd_reply_is_structured(&reply));
 610
 611         switch (chunk->type) {
 612         case NBD_REPLY_TYPE_OFFSET_DATA:
 613             /* special cased in nbd_co_receive_one_chunk, data is already
 614              * in qiov */
 615             break;
 616         case NBD_REPLY_TYPE_OFFSET_HOLE:
 617             ret = nbd_parse_offset_hole_payload(&reply.structured, payload,
 618                                                 offset, qiov, &local_err);
 619             if (ret < 0) {
 620                 s->quit = true;
 621                 nbd_iter_error(&iter, true, ret, &local_err);
 622             }
 623             break;
 624         default:
 625             if (!nbd_reply_type_is_error(chunk->type)) {
 626                 /* not allowed reply type */
 627                 s->quit = true;
 628                 error_setg(&local_err,
 629                            "Unexpected reply type: %d (%s) for CMD_READ",
 630                            chunk->type, nbd_reply_type_lookup(chunk->type));
 631                 nbd_iter_error(&iter, true, -EINVAL, &local_err);
 632             }
 633         }
 634
 635         g_free(payload);
 636         payload = NULL;
 637     }
 638
 639     error_propagate(errp, iter.err);
 640     return iter.ret;
 641 }
 642
 643 static int nbd_co_request(BlockDriverState *bs, NBDRequest *request,
 644                           QEMUIOVector *write_qiov)
 645 {
 646     int ret;
 647     Error *local_err = NULL;
 648     NBDClientSession *client = nbd_get_client_session(bs);
 649
 650     assert(request->type != NBD_CMD_READ);
 651     if (write_qiov) {
 652         assert(request->type == NBD_CMD_WRITE);
 653         assert(request->len == iov_size(write_qiov->iov, write_qiov->niov));
 654     } else {
 655         assert(request->type != NBD_CMD_WRITE);
 656     }
 657     ret = nbd_co_send_request(bs, request, write_qiov);
 658     if (ret < 0) {
 659         return ret;
 660     }
 661
 662     ret = nbd_co_receive_return_code(client, request->handle, &local_err);
 663     if (local_err) {
 664         error_report_err(local_err);
 665     }
 666     return ret;
 667 }
 668
 669 int nbd_client_co_preadv(BlockDriverState *bs, uint64_t offset,
 670                          uint64_t bytes, QEMUIOVector *qiov, int flags)
 671 {
 672     int ret;
 673     Error *local_err = NULL;
 674     NBDClientSession *client = nbd_get_client_session(bs);
 675     NBDRequest request = {
 676         .type = NBD_CMD_READ,
 677         .from = offset,
 678         .len = bytes,
 679     };
 680
 681     assert(bytes <= NBD_MAX_BUFFER_SIZE);
 682     assert(!flags);
 683
 684     if (!bytes) {
 685         return 0;
 686     }
 687     ret = nbd_co_send_request(bs, &request, NULL);
 688     if (ret < 0) {
 689         return ret;
 690     }
 691
 692     ret = nbd_co_receive_cmdread_reply(client, request.handle, offset, qiov,
 693                                        &local_err);
 694     if (local_err) {
 695         error_report_err(local_err);
 696     }
 697     return ret;
 698 }
 699
 700 int nbd_client_co_pwritev(BlockDriverState *bs, uint64_t offset,
 701                           uint64_t bytes, QEMUIOVector *qiov, int flags)
 702 {
 703     NBDClientSession *client = nbd_get_client_session(bs);
 704     NBDRequest request = {
 705         .type = NBD_CMD_WRITE,
 706         .from = offset,
 707         .len = bytes,
 708     };
 709
 710     assert(!(client->info.flags & NBD_FLAG_READ_ONLY));
 711     if (flags & BDRV_REQ_FUA) {
 712         assert(client->info.flags & NBD_FLAG_SEND_FUA);
 713         request.flags |= NBD_CMD_FLAG_FUA;
 714     }
 715
 716     assert(bytes <= NBD_MAX_BUFFER_SIZE);
 717
 718     if (!bytes) {
 719         return 0;
 720     }
 721     return nbd_co_request(bs, &request, qiov);
 722 }
 723
 724 int nbd_client_co_pwrite_zeroes(BlockDriverState *bs, int64_t offset,
 725                                 int bytes, BdrvRequestFlags flags)
 726 {
 727     NBDClientSession *client = nbd_get_client_session(bs);
 728     NBDRequest request = {
 729         .type = NBD_CMD_WRITE_ZEROES,
 730         .from = offset,
 731         .len = bytes,
 732     };
 733
 734     assert(!(client->info.flags & NBD_FLAG_READ_ONLY));
 735     if (!(client->info.flags & NBD_FLAG_SEND_WRITE_ZEROES)) {
 736         return -ENOTSUP;
 737     }
 738
 739     if (flags & BDRV_REQ_FUA) {
 740         assert(client->info.flags & NBD_FLAG_SEND_FUA);
 741         request.flags |= NBD_CMD_FLAG_FUA;
 742     }
 743     if (!(flags & BDRV_REQ_MAY_UNMAP)) {
 744         request.flags |= NBD_CMD_FLAG_NO_HOLE;
 745     }
 746
 747     if (!bytes) {
 748         return 0;
 749     }
 750     return nbd_co_request(bs, &request, NULL);
 751 }
 752
 753 int nbd_client_co_flush(BlockDriverState *bs)
 754 {
 755     NBDClientSession *client = nbd_get_client_session(bs);
 756     NBDRequest request = { .type = NBD_CMD_FLUSH };
 757
 758     if (!(client->info.flags & NBD_FLAG_SEND_FLUSH)) {
 759         return 0;
 760     }
 761
 762     request.from = 0;
 763     request.len = 0;
 764
 765     return nbd_co_request(bs, &request, NULL);
 766 }
 767
 768 int nbd_client_co_pdiscard(BlockDriverState *bs, int64_t offset, int bytes)
 769 {
 770     NBDClientSession *client = nbd_get_client_session(bs);
 771     NBDRequest request = {
 772         .type = NBD_CMD_TRIM,
 773         .from = offset,
 774         .len = bytes,
 775     };
 776
 777     assert(!(client->info.flags & NBD_FLAG_READ_ONLY));
 778     if (!(client->info.flags & NBD_FLAG_SEND_TRIM) || !bytes) {
 779         return 0;
 780     }
 781
 782     return nbd_co_request(bs, &request, NULL);
 783 }
 784
 785 void nbd_client_detach_aio_context(BlockDriverState *bs)
 786 {
 787     NBDClientSession *client = nbd_get_client_session(bs);
 788     qio_channel_detach_aio_context(QIO_CHANNEL(client->ioc));
 789 }
 790
 791 void nbd_client_attach_aio_context(BlockDriverState *bs,
 792                                    AioContext *new_context)
 793 {
 794     NBDClientSession *client = nbd_get_client_session(bs);
 795     qio_channel_attach_aio_context(QIO_CHANNEL(client->ioc), new_context);
 796     aio_co_schedule(new_context, client->read_reply_co);
 797 }
 798
 799 void nbd_client_close(BlockDriverState *bs)
 800 {
 801     NBDClientSession *client = nbd_get_client_session(bs);
 802     NBDRequest request = { .type = NBD_CMD_DISC };
 803
 804     if (client->ioc == NULL) {
 805         return;
 806     }
 807
 808     nbd_send_request(client->ioc, &request);
 809
 810     nbd_teardown_connection(bs);
 811 }
 812
 813 int nbd_client_init(BlockDriverState *bs,
 814                     QIOChannelSocket *sioc,
 815                     const char *export,
 816                     QCryptoTLSCreds *tlscreds,
 817                     const char *hostname,
 818                     Error **errp)
 819 {
 820     NBDClientSession *client = nbd_get_client_session(bs);
 821     int ret;
 822
 823     /* NBD handshake */
 824     logout("session init %s\n", export);
 825     qio_channel_set_blocking(QIO_CHANNEL(sioc), true, NULL);
 826
 827     client->info.request_sizes = true;
 828     client->info.structured_reply = true;
 829     ret = nbd_receive_negotiate(QIO_CHANNEL(sioc), export,
 830                                 tlscreds, hostname,
 831                                 &client->ioc, &client->info, errp);
 832     if (ret < 0) {
 833         logout("Failed to negotiate with the NBD server\n");
 834         return ret;
 835     }
 836     if (client->info.flags & NBD_FLAG_READ_ONLY &&
 837         !bdrv_is_read_only(bs)) {
 838         error_setg(errp,
 839                    "request for write access conflicts with read-only export");
 840         return -EACCES;
 841     }
 842     if (client->info.flags & NBD_FLAG_SEND_FUA) {
 843         bs->supported_write_flags = BDRV_REQ_FUA;
 844         bs->supported_zero_flags |= BDRV_REQ_FUA;
 845     }
 846     if (client->info.flags & NBD_FLAG_SEND_WRITE_ZEROES) {
 847         bs->supported_zero_flags |= BDRV_REQ_MAY_UNMAP;
 848     }
 849
 850     qemu_co_mutex_init(&client->send_mutex);
 851     qemu_co_queue_init(&client->free_sema);
 852     client->sioc = sioc;
 853     object_ref(OBJECT(client->sioc));
 854
 855     if (!client->ioc) {
 856         client->ioc = QIO_CHANNEL(sioc);
 857         object_ref(OBJECT(client->ioc));
 858     }
 859
 860     /* Now that we're connected, set the socket to be non-blocking and
 861      * kick the reply mechanism.  */
 862     qio_channel_set_blocking(QIO_CHANNEL(sioc), false, NULL);
 863     client->read_reply_co = qemu_coroutine_create(nbd_read_reply_entry, client);
 864     nbd_client_attach_aio_context(bs, bdrv_get_aio_context(bs));
 865
 866     logout("Established connection with NBD server\n");
 867     return 0;
 868 }