nbd/client.c

   1 /*
   2  *  Copyright (C) 2016-2017 Red Hat, Inc.
   3  *  Copyright (C) 2005  Anthony Liguori <anthony@codemonkey.ws>
   4  *
   5  *  Network Block Device Client Side
   6  *
   7  *  This program is free software; you can redistribute it and/or modify
   8  *  it under the terms of the GNU General Public License as published by
   9  *  the Free Software Foundation; under version 2 of the License.
  10  *
  11  *  This program is distributed in the hope that it will be useful,
  12  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
  13  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  14  *  GNU General Public License for more details.
  15  *
  16  *  You should have received a copy of the GNU General Public License
  17  *  along with this program; if not, see <http://www.gnu.org/licenses/>.
  18  */
  19
  20 #include "qemu/osdep.h"
  21 #include "qapi/error.h"
  22 #include "trace.h"
  23 #include "nbd-internal.h"
  24
  25 static int nbd_errno_to_system_errno(int err)
  26 {
  27     int ret;
  28     switch (err) {
  29     case NBD_SUCCESS:
  30         ret = 0;
  31         break;
  32     case NBD_EPERM:
  33         ret = EPERM;
  34         break;
  35     case NBD_EIO:
  36         ret = EIO;
  37         break;
  38     case NBD_ENOMEM:
  39         ret = ENOMEM;
  40         break;
  41     case NBD_ENOSPC:
  42         ret = ENOSPC;
  43         break;
  44     case NBD_ESHUTDOWN:
  45         ret = ESHUTDOWN;
  46         break;
  47     default:
  48         trace_nbd_unknown_error(err);
  49         /* fallthrough */
  50     case NBD_EINVAL:
  51         ret = EINVAL;
  52         break;
  53     }
  54     return ret;
  55 }
  56
  57 /* Definitions for opaque data types */
  58
  59 static QTAILQ_HEAD(, NBDExport) exports = QTAILQ_HEAD_INITIALIZER(exports);
  60
  61 /* That's all folks */
  62
  63 /* Basic flow for negotiation
  64
  65    Server         Client
  66    Negotiate
  67
  68    or
  69
  70    Server         Client
  71    Negotiate #1
  72                   Option
  73    Negotiate #2
  74
  75    ----
  76
  77    followed by
  78
  79    Server         Client
  80                   Request
  81    Response
  82                   Request
  83    Response
  84                   ...
  85    ...
  86                   Request (type == 2)
  87
  88 */
  89
  90 /* Send an option request.
  91  *
  92  * The request is for option @opt, with @data containing @len bytes of
  93  * additional payload for the request (@len may be -1 to treat @data as
  94  * a C string; and @data may be NULL if @len is 0).
  95  * Return 0 if successful, -1 with errp set if it is impossible to
  96  * continue. */
  97 static int nbd_send_option_request(QIOChannel *ioc, uint32_t opt,
  98                                    uint32_t len, const char *data,
  99                                    Error **errp)
 100 {
 101     nbd_option req;
 102     QEMU_BUILD_BUG_ON(sizeof(req) != 16);
 103
 104     if (len == -1) {
 105         req.length = len = strlen(data);
 106     }
 107     trace_nbd_send_option_request(opt, nbd_opt_lookup(opt), len);
 108
 109     stq_be_p(&req.magic, NBD_OPTS_MAGIC);
 110     stl_be_p(&req.option, opt);
 111     stl_be_p(&req.length, len);
 112
 113     if (nbd_write(ioc, &req, sizeof(req), errp) < 0) {
 114         error_prepend(errp, "Failed to send option request header");
 115         return -1;
 116     }
 117
 118     if (len && nbd_write(ioc, (char *) data, len, errp) < 0) {
 119         error_prepend(errp, "Failed to send option request data");
 120         return -1;
 121     }
 122
 123     return 0;
 124 }
 125
 126 /* Send NBD_OPT_ABORT as a courtesy to let the server know that we are
 127  * not going to attempt further negotiation. */
 128 static void nbd_send_opt_abort(QIOChannel *ioc)
 129 {
 130     /* Technically, a compliant server is supposed to reply to us; but
 131      * older servers disconnected instead. At any rate, we're allowed
 132      * to disconnect without waiting for the server reply, so we don't
 133      * even care if the request makes it to the server, let alone
 134      * waiting around for whether the server replies. */
 135     nbd_send_option_request(ioc, NBD_OPT_ABORT, 0, NULL, NULL);
 136 }
 137
 138
 139 /* Receive the header of an option reply, which should match the given
 140  * opt.  Read through the length field, but NOT the length bytes of
 141  * payload. Return 0 if successful, -1 with errp set if it is
 142  * impossible to continue. */
 143 static int nbd_receive_option_reply(QIOChannel *ioc, uint32_t opt,
 144                                     nbd_opt_reply *reply, Error **errp)
 145 {
 146     QEMU_BUILD_BUG_ON(sizeof(*reply) != 20);
 147     if (nbd_read(ioc, reply, sizeof(*reply), errp) < 0) {
 148         error_prepend(errp, "failed to read option reply");
 149         nbd_send_opt_abort(ioc);
 150         return -1;
 151     }
 152     be64_to_cpus(&reply->magic);
 153     be32_to_cpus(&reply->option);
 154     be32_to_cpus(&reply->type);
 155     be32_to_cpus(&reply->length);
 156
 157     trace_nbd_receive_option_reply(reply->option, nbd_opt_lookup(reply->option),
 158                                    reply->type, nbd_rep_lookup(reply->type),
 159                                    reply->length);
 160
 161     if (reply->magic != NBD_REP_MAGIC) {
 162         error_setg(errp, "Unexpected option reply magic");
 163         nbd_send_opt_abort(ioc);
 164         return -1;
 165     }
 166     if (reply->option != opt) {
 167         error_setg(errp, "Unexpected option type %x expected %x",
 168                    reply->option, opt);
 169         nbd_send_opt_abort(ioc);
 170         return -1;
 171     }
 172     return 0;
 173 }
 174
 175 /* If reply represents success, return 1 without further action.
 176  * If reply represents an error, consume the optional payload of
 177  * the packet on ioc.  Then return 0 for unsupported (so the client
 178  * can fall back to other approaches), or -1 with errp set for other
 179  * errors.
 180  */
 181 static int nbd_handle_reply_err(QIOChannel *ioc, nbd_opt_reply *reply,
 182                                 Error **errp)
 183 {
 184     char *msg = NULL;
 185     int result = -1;
 186
 187     if (!(reply->type & (1 << 31))) {
 188         return 1;
 189     }
 190
 191     if (reply->length) {
 192         if (reply->length > NBD_MAX_BUFFER_SIZE) {
 193             error_setg(errp, "server error 0x%" PRIx32
 194                        " (%s) message is too long",
 195                        reply->type, nbd_rep_lookup(reply->type));
 196             goto cleanup;
 197         }
 198         msg = g_malloc(reply->length + 1);
 199         if (nbd_read(ioc, msg, reply->length, errp) < 0) {
 200             error_prepend(errp, "failed to read option error 0x%" PRIx32
 201                           " (%s) message",
 202                           reply->type, nbd_rep_lookup(reply->type));
 203             goto cleanup;
 204         }
 205         msg[reply->length] = '\0';
 206     }
 207
 208     switch (reply->type) {
 209     case NBD_REP_ERR_UNSUP:
 210         trace_nbd_reply_err_unsup(reply->option, nbd_opt_lookup(reply->option));
 211         result = 0;
 212         goto cleanup;
 213
 214     case NBD_REP_ERR_POLICY:
 215         error_setg(errp, "Denied by server for option %" PRIx32 " (%s)",
 216                    reply->option, nbd_opt_lookup(reply->option));
 217         break;
 218
 219     case NBD_REP_ERR_INVALID:
 220         error_setg(errp, "Invalid data length for option %" PRIx32 " (%s)",
 221                    reply->option, nbd_opt_lookup(reply->option));
 222         break;
 223
 224     case NBD_REP_ERR_PLATFORM:
 225         error_setg(errp, "Server lacks support for option %" PRIx32 " (%s)",
 226                    reply->option, nbd_opt_lookup(reply->option));
 227         break;
 228
 229     case NBD_REP_ERR_TLS_REQD:
 230         error_setg(errp, "TLS negotiation required before option %" PRIx32
 231                    " (%s)", reply->option, nbd_opt_lookup(reply->option));
 232         break;
 233
 234     case NBD_REP_ERR_UNKNOWN:
 235         error_setg(errp, "Requested export not available");
 236         break;
 237
 238     case NBD_REP_ERR_SHUTDOWN:
 239         error_setg(errp, "Server shutting down before option %" PRIx32 " (%s)",
 240                    reply->option, nbd_opt_lookup(reply->option));
 241         break;
 242
 243     case NBD_REP_ERR_BLOCK_SIZE_REQD:
 244         error_setg(errp, "Server requires INFO_BLOCK_SIZE for option %" PRIx32
 245                    " (%s)", reply->option, nbd_opt_lookup(reply->option));
 246         break;
 247
 248     default:
 249         error_setg(errp, "Unknown error code when asking for option %" PRIx32
 250                    " (%s)", reply->option, nbd_opt_lookup(reply->option));
 251         break;
 252     }
 253
 254     if (msg) {
 255         error_append_hint(errp, "server reported: %s\n", msg);
 256     }
 257
 258  cleanup:
 259     g_free(msg);
 260     if (result < 0) {
 261         nbd_send_opt_abort(ioc);
 262     }
 263     return result;
 264 }
 265
 266 /* Process another portion of the NBD_OPT_LIST reply.  Set *@match if
 267  * the current reply matches @want or if the server does not support
 268  * NBD_OPT_LIST, otherwise leave @match alone.  Return 0 if iteration
 269  * is complete, positive if more replies are expected, or negative
 270  * with @errp set if an unrecoverable error occurred. */
 271 static int nbd_receive_list(QIOChannel *ioc, const char *want, bool *match,
 272                             Error **errp)
 273 {
 274     nbd_opt_reply reply;
 275     uint32_t len;
 276     uint32_t namelen;
 277     char name[NBD_MAX_NAME_SIZE + 1];
 278     int error;
 279
 280     if (nbd_receive_option_reply(ioc, NBD_OPT_LIST, &reply, errp) < 0) {
 281         return -1;
 282     }
 283     error = nbd_handle_reply_err(ioc, &reply, errp);
 284     if (error <= 0) {
 285         /* The server did not support NBD_OPT_LIST, so set *match on
 286          * the assumption that any name will be accepted.  */
 287         *match = true;
 288         return error;
 289     }
 290     len = reply.length;
 291
 292     if (reply.type == NBD_REP_ACK) {
 293         if (len != 0) {
 294             error_setg(errp, "length too long for option end");
 295             nbd_send_opt_abort(ioc);
 296             return -1;
 297         }
 298         return 0;
 299     } else if (reply.type != NBD_REP_SERVER) {
 300         error_setg(errp, "Unexpected reply type %" PRIx32 " expected %x",
 301                    reply.type, NBD_REP_SERVER);
 302         nbd_send_opt_abort(ioc);
 303         return -1;
 304     }
 305
 306     if (len < sizeof(namelen) || len > NBD_MAX_BUFFER_SIZE) {
 307         error_setg(errp, "incorrect option length %" PRIu32, len);
 308         nbd_send_opt_abort(ioc);
 309         return -1;
 310     }
 311     if (nbd_read(ioc, &namelen, sizeof(namelen), errp) < 0) {
 312         error_prepend(errp, "failed to read option name length");
 313         nbd_send_opt_abort(ioc);
 314         return -1;
 315     }
 316     namelen = be32_to_cpu(namelen);
 317     len -= sizeof(namelen);
 318     if (len < namelen) {
 319         error_setg(errp, "incorrect option name length");
 320         nbd_send_opt_abort(ioc);
 321         return -1;
 322     }
 323     if (namelen != strlen(want)) {
 324         if (nbd_drop(ioc, len, errp) < 0) {
 325             error_prepend(errp, "failed to skip export name with wrong length");
 326             nbd_send_opt_abort(ioc);
 327             return -1;
 328         }
 329         return 1;
 330     }
 331
 332     assert(namelen < sizeof(name));
 333     if (nbd_read(ioc, name, namelen, errp) < 0) {
 334         error_prepend(errp, "failed to read export name");
 335         nbd_send_opt_abort(ioc);
 336         return -1;
 337     }
 338     name[namelen] = '\0';
 339     len -= namelen;
 340     if (nbd_drop(ioc, len, errp) < 0) {
 341         error_prepend(errp, "failed to read export description");
 342         nbd_send_opt_abort(ioc);
 343         return -1;
 344     }
 345     if (!strcmp(name, want)) {
 346         *match = true;
 347     }
 348     return 1;
 349 }
 350
 351
 352 /* Returns -1 if NBD_OPT_GO proves the export @wantname cannot be
 353  * used, 0 if NBD_OPT_GO is unsupported (fall back to NBD_OPT_LIST and
 354  * NBD_OPT_EXPORT_NAME in that case), and > 0 if the export is good to
 355  * go (with @info populated). */
 356 static int nbd_opt_go(QIOChannel *ioc, const char *wantname,
 357                       NBDExportInfo *info, Error **errp)
 358 {
 359     nbd_opt_reply reply;
 360     uint32_t len = strlen(wantname);
 361     uint16_t type;
 362     int error;
 363     char *buf;
 364
 365     /* The protocol requires that the server send NBD_INFO_EXPORT with
 366      * a non-zero flags (at least NBD_FLAG_HAS_FLAGS must be set); so
 367      * flags still 0 is a witness of a broken server. */
 368     info->flags = 0;
 369
 370     trace_nbd_opt_go_start(wantname);
 371     buf = g_malloc(4 + len + 2 + 2 * info->request_sizes + 1);
 372     stl_be_p(buf, len);
 373     memcpy(buf + 4, wantname, len);
 374     /* At most one request, everything else up to server */
 375     stw_be_p(buf + 4 + len, info->request_sizes);
 376     if (info->request_sizes) {
 377         stw_be_p(buf + 4 + len + 2, NBD_INFO_BLOCK_SIZE);
 378     }
 379     if (nbd_send_option_request(ioc, NBD_OPT_GO,
 380                                 4 + len + 2 + 2 * info->request_sizes, buf,
 381                                 errp) < 0) {
 382         return -1;
 383     }
 384
 385     while (1) {
 386         if (nbd_receive_option_reply(ioc, NBD_OPT_GO, &reply, errp) < 0) {
 387             return -1;
 388         }
 389         error = nbd_handle_reply_err(ioc, &reply, errp);
 390         if (error <= 0) {
 391             return error;
 392         }
 393         len = reply.length;
 394
 395         if (reply.type == NBD_REP_ACK) {
 396             /* Server is done sending info and moved into transmission
 397                phase, but make sure it sent flags */
 398             if (len) {
 399                 error_setg(errp, "server sent invalid NBD_REP_ACK");
 400                 nbd_send_opt_abort(ioc);
 401                 return -1;
 402             }
 403             if (!info->flags) {
 404                 error_setg(errp, "broken server omitted NBD_INFO_EXPORT");
 405                 nbd_send_opt_abort(ioc);
 406                 return -1;
 407             }
 408             trace_nbd_opt_go_success();
 409             return 1;
 410         }
 411         if (reply.type != NBD_REP_INFO) {
 412             error_setg(errp, "unexpected reply type %" PRIx32
 413                        " (%s), expected %x",
 414                        reply.type, nbd_rep_lookup(reply.type), NBD_REP_INFO);
 415             nbd_send_opt_abort(ioc);
 416             return -1;
 417         }
 418         if (len < sizeof(type)) {
 419             error_setg(errp, "NBD_REP_INFO length %" PRIu32 " is too short",
 420                        len);
 421             nbd_send_opt_abort(ioc);
 422             return -1;
 423         }
 424         if (nbd_read(ioc, &type, sizeof(type), errp) < 0) {
 425             error_prepend(errp, "failed to read info type");
 426             nbd_send_opt_abort(ioc);
 427             return -1;
 428         }
 429         len -= sizeof(type);
 430         be16_to_cpus(&type);
 431         switch (type) {
 432         case NBD_INFO_EXPORT:
 433             if (len != sizeof(info->size) + sizeof(info->flags)) {
 434                 error_setg(errp, "remaining export info len %" PRIu32
 435                            " is unexpected size", len);
 436                 nbd_send_opt_abort(ioc);
 437                 return -1;
 438             }
 439             if (nbd_read(ioc, &info->size, sizeof(info->size), errp) < 0) {
 440                 error_prepend(errp, "failed to read info size");
 441                 nbd_send_opt_abort(ioc);
 442                 return -1;
 443             }
 444             be64_to_cpus(&info->size);
 445             if (nbd_read(ioc, &info->flags, sizeof(info->flags), errp) < 0) {
 446                 error_prepend(errp, "failed to read info flags");
 447                 nbd_send_opt_abort(ioc);
 448                 return -1;
 449             }
 450             be16_to_cpus(&info->flags);
 451             trace_nbd_receive_negotiate_size_flags(info->size, info->flags);
 452             break;
 453
 454         case NBD_INFO_BLOCK_SIZE:
 455             if (len != sizeof(info->min_block) * 3) {
 456                 error_setg(errp, "remaining export info len %" PRIu32
 457                            " is unexpected size", len);
 458                 nbd_send_opt_abort(ioc);
 459                 return -1;
 460             }
 461             if (nbd_read(ioc, &info->min_block, sizeof(info->min_block),
 462                          errp) < 0) {
 463                 error_prepend(errp, "failed to read info minimum block size");
 464                 nbd_send_opt_abort(ioc);
 465                 return -1;
 466             }
 467             be32_to_cpus(&info->min_block);
 468             if (!is_power_of_2(info->min_block)) {
 469                 error_setg(errp, "server minimum block size %" PRId32
 470                            "is not a power of two", info->min_block);
 471                 nbd_send_opt_abort(ioc);
 472                 return -1;
 473             }
 474             if (nbd_read(ioc, &info->opt_block, sizeof(info->opt_block),
 475                          errp) < 0) {
 476                 error_prepend(errp, "failed to read info preferred block size");
 477                 nbd_send_opt_abort(ioc);
 478                 return -1;
 479             }
 480             be32_to_cpus(&info->opt_block);
 481             if (!is_power_of_2(info->opt_block) ||
 482                 info->opt_block < info->min_block) {
 483                 error_setg(errp, "server preferred block size %" PRId32
 484                            "is not valid", info->opt_block);
 485                 nbd_send_opt_abort(ioc);
 486                 return -1;
 487             }
 488             if (nbd_read(ioc, &info->max_block, sizeof(info->max_block),
 489                          errp) < 0) {
 490                 error_prepend(errp, "failed to read info maximum block size");
 491                 nbd_send_opt_abort(ioc);
 492                 return -1;
 493             }
 494             be32_to_cpus(&info->max_block);
 495             trace_nbd_opt_go_info_block_size(info->min_block, info->opt_block,
 496                                              info->max_block);
 497             break;
 498
 499         default:
 500             trace_nbd_opt_go_info_unknown(type, nbd_info_lookup(type));
 501             if (nbd_drop(ioc, len, errp) < 0) {
 502                 error_prepend(errp, "Failed to read info payload");
 503                 nbd_send_opt_abort(ioc);
 504                 return -1;
 505             }
 506             break;
 507         }
 508     }
 509 }
 510
 511 /* Return -1 on failure, 0 if wantname is an available export. */
 512 static int nbd_receive_query_exports(QIOChannel *ioc,
 513                                      const char *wantname,
 514                                      Error **errp)
 515 {
 516     bool foundExport = false;
 517
 518     trace_nbd_receive_query_exports_start(wantname);
 519     if (nbd_send_option_request(ioc, NBD_OPT_LIST, 0, NULL, errp) < 0) {
 520         return -1;
 521     }
 522
 523     while (1) {
 524         int ret = nbd_receive_list(ioc, wantname, &foundExport, errp);
 525
 526         if (ret < 0) {
 527             /* Server gave unexpected reply */
 528             return -1;
 529         } else if (ret == 0) {
 530             /* Done iterating. */
 531             if (!foundExport) {
 532                 error_setg(errp, "No export with name '%s' available",
 533                            wantname);
 534                 nbd_send_opt_abort(ioc);
 535                 return -1;
 536             }
 537             trace_nbd_receive_query_exports_success(wantname);
 538             return 0;
 539         }
 540     }
 541 }
 542
 543 static QIOChannel *nbd_receive_starttls(QIOChannel *ioc,
 544                                         QCryptoTLSCreds *tlscreds,
 545                                         const char *hostname, Error **errp)
 546 {
 547     nbd_opt_reply reply;
 548     QIOChannelTLS *tioc;
 549     struct NBDTLSHandshakeData data = { 0 };
 550
 551     trace_nbd_receive_starttls_request();
 552     if (nbd_send_option_request(ioc, NBD_OPT_STARTTLS, 0, NULL, errp) < 0) {
 553         return NULL;
 554     }
 555
 556     trace_nbd_receive_starttls_reply();
 557     if (nbd_receive_option_reply(ioc, NBD_OPT_STARTTLS, &reply, errp) < 0) {
 558         return NULL;
 559     }
 560
 561     if (reply.type != NBD_REP_ACK) {
 562         error_setg(errp, "Server rejected request to start TLS %" PRIx32,
 563                    reply.type);
 564         nbd_send_opt_abort(ioc);
 565         return NULL;
 566     }
 567
 568     if (reply.length != 0) {
 569         error_setg(errp, "Start TLS response was not zero %" PRIu32,
 570                    reply.length);
 571         nbd_send_opt_abort(ioc);
 572         return NULL;
 573     }
 574
 575     trace_nbd_receive_starttls_new_client();
 576     tioc = qio_channel_tls_new_client(ioc, tlscreds, hostname, errp);
 577     if (!tioc) {
 578         return NULL;
 579     }
 580     qio_channel_set_name(QIO_CHANNEL(tioc), "nbd-client-tls");
 581     data.loop = g_main_loop_new(g_main_context_default(), FALSE);
 582     trace_nbd_receive_starttls_tls_handshake();
 583     qio_channel_tls_handshake(tioc,
 584                               nbd_tls_handshake,
 585                               &data,
 586                               NULL);
 587
 588     if (!data.complete) {
 589         g_main_loop_run(data.loop);
 590     }
 591     g_main_loop_unref(data.loop);
 592     if (data.error) {
 593         error_propagate(errp, data.error);
 594         object_unref(OBJECT(tioc));
 595         return NULL;
 596     }
 597
 598     return QIO_CHANNEL(tioc);
 599 }
 600
 601
 602 int nbd_receive_negotiate(QIOChannel *ioc, const char *name,
 603                           QCryptoTLSCreds *tlscreds, const char *hostname,
 604                           QIOChannel **outioc, NBDExportInfo *info,
 605                           Error **errp)
 606 {
 607     char buf[256];
 608     uint64_t magic;
 609     int rc;
 610     bool zeroes = true;
 611
 612     trace_nbd_receive_negotiate(tlscreds, hostname ? hostname : "<null>");
 613
 614     rc = -EINVAL;
 615
 616     if (outioc) {
 617         *outioc = NULL;
 618     }
 619     if (tlscreds && !outioc) {
 620         error_setg(errp, "Output I/O channel required for TLS");
 621         goto fail;
 622     }
 623
 624     if (nbd_read(ioc, buf, 8, errp) < 0) {
 625         error_prepend(errp, "Failed to read data");
 626         goto fail;
 627     }
 628
 629     buf[8] = '\0';
 630     if (strlen(buf) == 0) {
 631         error_setg(errp, "Server connection closed unexpectedly");
 632         goto fail;
 633     }
 634
 635     magic = ldq_be_p(buf);
 636     trace_nbd_receive_negotiate_magic(magic);
 637
 638     if (memcmp(buf, "NBDMAGIC", 8) != 0) {
 639         error_setg(errp, "Invalid magic received");
 640         goto fail;
 641     }
 642
 643     if (nbd_read(ioc, &magic, sizeof(magic), errp) < 0) {
 644         error_prepend(errp, "Failed to read magic");
 645         goto fail;
 646     }
 647     magic = be64_to_cpu(magic);
 648     trace_nbd_receive_negotiate_magic(magic);
 649
 650     if (magic == NBD_OPTS_MAGIC) {
 651         uint32_t clientflags = 0;
 652         uint16_t globalflags;
 653         bool fixedNewStyle = false;
 654
 655         if (nbd_read(ioc, &globalflags, sizeof(globalflags), errp) < 0) {
 656             error_prepend(errp, "Failed to read server flags");
 657             goto fail;
 658         }
 659         globalflags = be16_to_cpu(globalflags);
 660         trace_nbd_receive_negotiate_server_flags(globalflags);
 661         if (globalflags & NBD_FLAG_FIXED_NEWSTYLE) {
 662             fixedNewStyle = true;
 663             clientflags |= NBD_FLAG_C_FIXED_NEWSTYLE;
 664         }
 665         if (globalflags & NBD_FLAG_NO_ZEROES) {
 666             zeroes = false;
 667             clientflags |= NBD_FLAG_C_NO_ZEROES;
 668         }
 669         /* client requested flags */
 670         clientflags = cpu_to_be32(clientflags);
 671         if (nbd_write(ioc, &clientflags, sizeof(clientflags), errp) < 0) {
 672             error_prepend(errp, "Failed to send clientflags field");
 673             goto fail;
 674         }
 675         if (tlscreds) {
 676             if (fixedNewStyle) {
 677                 *outioc = nbd_receive_starttls(ioc, tlscreds, hostname, errp);
 678                 if (!*outioc) {
 679                     goto fail;
 680                 }
 681                 ioc = *outioc;
 682             } else {
 683                 error_setg(errp, "Server does not support STARTTLS");
 684                 goto fail;
 685             }
 686         }
 687         if (!name) {
 688             trace_nbd_receive_negotiate_default_name();
 689             name = "";
 690         }
 691         if (fixedNewStyle) {
 692             int result;
 693
 694             /* Try NBD_OPT_GO first - if it works, we are done (it
 695              * also gives us a good message if the server requires
 696              * TLS).  If it is not available, fall back to
 697              * NBD_OPT_LIST for nicer error messages about a missing
 698              * export, then use NBD_OPT_EXPORT_NAME.  */
 699             result = nbd_opt_go(ioc, name, info, errp);
 700             if (result < 0) {
 701                 goto fail;
 702             }
 703             if (result > 0) {
 704                 return 0;
 705             }
 706             /* Check our desired export is present in the
 707              * server export list. Since NBD_OPT_EXPORT_NAME
 708              * cannot return an error message, running this
 709              * query gives us better error reporting if the
 710              * export name is not available.
 711              */
 712             if (nbd_receive_query_exports(ioc, name, errp) < 0) {
 713                 goto fail;
 714             }
 715         }
 716         /* write the export name request */
 717         if (nbd_send_option_request(ioc, NBD_OPT_EXPORT_NAME, -1, name,
 718                                     errp) < 0) {
 719             goto fail;
 720         }
 721
 722         /* Read the response */
 723         if (nbd_read(ioc, &info->size, sizeof(info->size), errp) < 0) {
 724             error_prepend(errp, "Failed to read export length");
 725             goto fail;
 726         }
 727         be64_to_cpus(&info->size);
 728
 729         if (nbd_read(ioc, &info->flags, sizeof(info->flags), errp) < 0) {
 730             error_prepend(errp, "Failed to read export flags");
 731             goto fail;
 732         }
 733         be16_to_cpus(&info->flags);
 734     } else if (magic == NBD_CLIENT_MAGIC) {
 735         uint32_t oldflags;
 736
 737         if (name) {
 738             error_setg(errp, "Server does not support export names");
 739             goto fail;
 740         }
 741         if (tlscreds) {
 742             error_setg(errp, "Server does not support STARTTLS");
 743             goto fail;
 744         }
 745
 746         if (nbd_read(ioc, &info->size, sizeof(info->size), errp) < 0) {
 747             error_prepend(errp, "Failed to read export length");
 748             goto fail;
 749         }
 750         be64_to_cpus(&info->size);
 751
 752         if (nbd_read(ioc, &oldflags, sizeof(oldflags), errp) < 0) {
 753             error_prepend(errp, "Failed to read export flags");
 754             goto fail;
 755         }
 756         be32_to_cpus(&oldflags);
 757         if (oldflags & ~0xffff) {
 758             error_setg(errp, "Unexpected export flags %0x" PRIx32, oldflags);
 759             goto fail;
 760         }
 761         info->flags = oldflags;
 762     } else {
 763         error_setg(errp, "Bad magic received");
 764         goto fail;
 765     }
 766
 767     trace_nbd_receive_negotiate_size_flags(info->size, info->flags);
 768     if (zeroes && nbd_drop(ioc, 124, errp) < 0) {
 769         error_prepend(errp, "Failed to read reserved block");
 770         goto fail;
 771     }
 772     rc = 0;
 773
 774 fail:
 775     return rc;
 776 }
 777
 778 #ifdef __linux__
 779 int nbd_init(int fd, QIOChannelSocket *sioc, NBDExportInfo *info,
 780              Error **errp)
 781 {
 782     unsigned long sector_size = MAX(BDRV_SECTOR_SIZE, info->min_block);
 783     unsigned long sectors = info->size / sector_size;
 784
 785     /* FIXME: Once the kernel module is patched to honor block sizes,
 786      * and to advertise that fact to user space, we should update the
 787      * hand-off to the kernel to use any block sizes we learned. */
 788     assert(!info->request_sizes);
 789     if (info->size / sector_size != sectors) {
 790         error_setg(errp, "Export size %" PRIu64 " too large for 32-bit kernel",
 791                    info->size);
 792         return -E2BIG;
 793     }
 794
 795     trace_nbd_init_set_socket();
 796
 797     if (ioctl(fd, NBD_SET_SOCK, (unsigned long) sioc->fd) < 0) {
 798         int serrno = errno;
 799         error_setg(errp, "Failed to set NBD socket");
 800         return -serrno;
 801     }
 802
 803     trace_nbd_init_set_block_size(sector_size);
 804
 805     if (ioctl(fd, NBD_SET_BLKSIZE, sector_size) < 0) {
 806         int serrno = errno;
 807         error_setg(errp, "Failed setting NBD block size");
 808         return -serrno;
 809     }
 810
 811     trace_nbd_init_set_size(sectors);
 812     if (info->size % sector_size) {
 813         trace_nbd_init_trailing_bytes(info->size % sector_size);
 814     }
 815
 816     if (ioctl(fd, NBD_SET_SIZE_BLOCKS, sectors) < 0) {
 817         int serrno = errno;
 818         error_setg(errp, "Failed setting size (in blocks)");
 819         return -serrno;
 820     }
 821
 822     if (ioctl(fd, NBD_SET_FLAGS, (unsigned long) info->flags) < 0) {
 823         if (errno == ENOTTY) {
 824             int read_only = (info->flags & NBD_FLAG_READ_ONLY) != 0;
 825             trace_nbd_init_set_readonly();
 826
 827             if (ioctl(fd, BLKROSET, (unsigned long) &read_only) < 0) {
 828                 int serrno = errno;
 829                 error_setg(errp, "Failed setting read-only attribute");
 830                 return -serrno;
 831             }
 832         } else {
 833             int serrno = errno;
 834             error_setg(errp, "Failed setting flags");
 835             return -serrno;
 836         }
 837     }
 838
 839     trace_nbd_init_finish();
 840
 841     return 0;
 842 }
 843
 844 int nbd_client(int fd)
 845 {
 846     int ret;
 847     int serrno;
 848
 849     trace_nbd_client_loop();
 850
 851     ret = ioctl(fd, NBD_DO_IT);
 852     if (ret < 0 && errno == EPIPE) {
 853         /* NBD_DO_IT normally returns EPIPE when someone has disconnected
 854          * the socket via NBD_DISCONNECT.  We do not want to return 1 in
 855          * that case.
 856          */
 857         ret = 0;
 858     }
 859     serrno = errno;
 860
 861     trace_nbd_client_loop_ret(ret, strerror(serrno));
 862
 863     trace_nbd_client_clear_queue();
 864     ioctl(fd, NBD_CLEAR_QUE);
 865
 866     trace_nbd_client_clear_socket();
 867     ioctl(fd, NBD_CLEAR_SOCK);
 868
 869     errno = serrno;
 870     return ret;
 871 }
 872
 873 int nbd_disconnect(int fd)
 874 {
 875     ioctl(fd, NBD_CLEAR_QUE);
 876     ioctl(fd, NBD_DISCONNECT);
 877     ioctl(fd, NBD_CLEAR_SOCK);
 878     return 0;
 879 }
 880
 881 #else
 882 int nbd_init(int fd, QIOChannelSocket *ioc, NBDExportInfo *info,
 883              Error **errp)
 884 {
 885     error_setg(errp, "nbd_init is only supported on Linux");
 886     return -ENOTSUP;
 887 }
 888
 889 int nbd_client(int fd)
 890 {
 891     return -ENOTSUP;
 892 }
 893 int nbd_disconnect(int fd)
 894 {
 895     return -ENOTSUP;
 896 }
 897 #endif
 898
 899 ssize_t nbd_send_request(QIOChannel *ioc, NBDRequest *request)
 900 {
 901     uint8_t buf[NBD_REQUEST_SIZE];
 902
 903     trace_nbd_send_request(request->from, request->len, request->handle,
 904                            request->flags, request->type,
 905                            nbd_cmd_lookup(request->type));
 906
 907     stl_be_p(buf, NBD_REQUEST_MAGIC);
 908     stw_be_p(buf + 4, request->flags);
 909     stw_be_p(buf + 6, request->type);
 910     stq_be_p(buf + 8, request->handle);
 911     stq_be_p(buf + 16, request->from);
 912     stl_be_p(buf + 24, request->len);
 913
 914     return nbd_write(ioc, buf, sizeof(buf), NULL);
 915 }
 916
 917 ssize_t nbd_receive_reply(QIOChannel *ioc, NBDReply *reply, Error **errp)
 918 {
 919     uint8_t buf[NBD_REPLY_SIZE];
 920     uint32_t magic;
 921     ssize_t ret;
 922
 923     ret = nbd_read_eof(ioc, buf, sizeof(buf), errp);
 924     if (ret <= 0) {
 925         return ret;
 926     }
 927
 928     if (ret != sizeof(buf)) {
 929         error_setg(errp, "read failed");
 930         return -EINVAL;
 931     }
 932
 933     /* Reply
 934        [ 0 ..  3]    magic   (NBD_REPLY_MAGIC)
 935        [ 4 ..  7]    error   (0 == no error)
 936        [ 7 .. 15]    handle
 937      */
 938
 939     magic = ldl_be_p(buf);
 940     reply->error  = ldl_be_p(buf + 4);
 941     reply->handle = ldq_be_p(buf + 8);
 942
 943     reply->error = nbd_errno_to_system_errno(reply->error);
 944
 945     if (reply->error == ESHUTDOWN) {
 946         /* This works even on mingw which lacks a native ESHUTDOWN */
 947         error_setg(errp, "server shutting down");
 948         return -EINVAL;
 949     }
 950     trace_nbd_receive_reply(magic, reply->error, reply->handle);
 951
 952     if (magic != NBD_REPLY_MAGIC) {
 953         error_setg(errp, "invalid magic (got 0x%" PRIx32 ")", magic);
 954         return -EINVAL;
 955     }
 956     return sizeof(buf);
 957 }
 958