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 /* Definitions for opaque data types */
  26
  27 static QTAILQ_HEAD(, NBDExport) exports = QTAILQ_HEAD_INITIALIZER(exports);
  28
  29 /* That's all folks */
  30
  31 /* Basic flow for negotiation
  32
  33    Server         Client
  34    Negotiate
  35
  36    or
  37
  38    Server         Client
  39    Negotiate #1
  40                   Option
  41    Negotiate #2
  42
  43    ----
  44
  45    followed by
  46
  47    Server         Client
  48                   Request
  49    Response
  50                   Request
  51    Response
  52                   ...
  53    ...
  54                   Request (type == 2)
  55
  56 */
  57
  58 /* Send an option request.
  59  *
  60  * The request is for option @opt, with @data containing @len bytes of
  61  * additional payload for the request (@len may be -1 to treat @data as
  62  * a C string; and @data may be NULL if @len is 0).
  63  * Return 0 if successful, -1 with errp set if it is impossible to
  64  * continue. */
  65 static int nbd_send_option_request(QIOChannel *ioc, uint32_t opt,
  66                                    uint32_t len, const char *data,
  67                                    Error **errp)
  68 {
  69     nbd_option req;
  70     QEMU_BUILD_BUG_ON(sizeof(req) != 16);
  71
  72     if (len == -1) {
  73         req.length = len = strlen(data);
  74     }
  75     trace_nbd_send_option_request(opt, nbd_opt_lookup(opt), len);
  76
  77     stq_be_p(&req.magic, NBD_OPTS_MAGIC);
  78     stl_be_p(&req.option, opt);
  79     stl_be_p(&req.length, len);
  80
  81     if (nbd_write(ioc, &req, sizeof(req), errp) < 0) {
  82         error_prepend(errp, "Failed to send option request header: ");
  83         return -1;
  84     }
  85
  86     if (len && nbd_write(ioc, (char *) data, len, errp) < 0) {
  87         error_prepend(errp, "Failed to send option request data: ");
  88         return -1;
  89     }
  90
  91     return 0;
  92 }
  93
  94 /* Send NBD_OPT_ABORT as a courtesy to let the server know that we are
  95  * not going to attempt further negotiation. */
  96 static void nbd_send_opt_abort(QIOChannel *ioc)
  97 {
  98     /* Technically, a compliant server is supposed to reply to us; but
  99      * older servers disconnected instead. At any rate, we're allowed
 100      * to disconnect without waiting for the server reply, so we don't
 101      * even care if the request makes it to the server, let alone
 102      * waiting around for whether the server replies. */
 103     nbd_send_option_request(ioc, NBD_OPT_ABORT, 0, NULL, NULL);
 104 }
 105
 106
 107 /* Receive the header of an option reply, which should match the given
 108  * opt.  Read through the length field, but NOT the length bytes of
 109  * payload. Return 0 if successful, -1 with errp set if it is
 110  * impossible to continue. */
 111 static int nbd_receive_option_reply(QIOChannel *ioc, uint32_t opt,
 112                                     nbd_opt_reply *reply, Error **errp)
 113 {
 114     QEMU_BUILD_BUG_ON(sizeof(*reply) != 20);
 115     if (nbd_read(ioc, reply, sizeof(*reply), errp) < 0) {
 116         error_prepend(errp, "failed to read option reply: ");
 117         nbd_send_opt_abort(ioc);
 118         return -1;
 119     }
 120     be64_to_cpus(&reply->magic);
 121     be32_to_cpus(&reply->option);
 122     be32_to_cpus(&reply->type);
 123     be32_to_cpus(&reply->length);
 124
 125     trace_nbd_receive_option_reply(reply->option, nbd_opt_lookup(reply->option),
 126                                    reply->type, nbd_rep_lookup(reply->type),
 127                                    reply->length);
 128
 129     if (reply->magic != NBD_REP_MAGIC) {
 130         error_setg(errp, "Unexpected option reply magic");
 131         nbd_send_opt_abort(ioc);
 132         return -1;
 133     }
 134     if (reply->option != opt) {
 135         error_setg(errp, "Unexpected option type %x expected %x",
 136                    reply->option, opt);
 137         nbd_send_opt_abort(ioc);
 138         return -1;
 139     }
 140     return 0;
 141 }
 142
 143 /* If reply represents success, return 1 without further action.
 144  * If reply represents an error, consume the optional payload of
 145  * the packet on ioc.  Then return 0 for unsupported (so the client
 146  * can fall back to other approaches), or -1 with errp set for other
 147  * errors.
 148  */
 149 static int nbd_handle_reply_err(QIOChannel *ioc, nbd_opt_reply *reply,
 150                                 Error **errp)
 151 {
 152     char *msg = NULL;
 153     int result = -1;
 154
 155     if (!(reply->type & (1 << 31))) {
 156         return 1;
 157     }
 158
 159     if (reply->length) {
 160         if (reply->length > NBD_MAX_BUFFER_SIZE) {
 161             error_setg(errp, "server error 0x%" PRIx32
 162                        " (%s) message is too long",
 163                        reply->type, nbd_rep_lookup(reply->type));
 164             goto cleanup;
 165         }
 166         msg = g_malloc(reply->length + 1);
 167         if (nbd_read(ioc, msg, reply->length, errp) < 0) {
 168             error_prepend(errp, "failed to read option error 0x%" PRIx32
 169                           " (%s) message: ",
 170                           reply->type, nbd_rep_lookup(reply->type));
 171             goto cleanup;
 172         }
 173         msg[reply->length] = '\0';
 174     }
 175
 176     switch (reply->type) {
 177     case NBD_REP_ERR_UNSUP:
 178         trace_nbd_reply_err_unsup(reply->option, nbd_opt_lookup(reply->option));
 179         result = 0;
 180         goto cleanup;
 181
 182     case NBD_REP_ERR_POLICY:
 183         error_setg(errp, "Denied by server for option %" PRIx32 " (%s)",
 184                    reply->option, nbd_opt_lookup(reply->option));
 185         break;
 186
 187     case NBD_REP_ERR_INVALID:
 188         error_setg(errp, "Invalid data length for option %" PRIx32 " (%s)",
 189                    reply->option, nbd_opt_lookup(reply->option));
 190         break;
 191
 192     case NBD_REP_ERR_PLATFORM:
 193         error_setg(errp, "Server lacks support for option %" PRIx32 " (%s)",
 194                    reply->option, nbd_opt_lookup(reply->option));
 195         break;
 196
 197     case NBD_REP_ERR_TLS_REQD:
 198         error_setg(errp, "TLS negotiation required before option %" PRIx32
 199                    " (%s)", reply->option, nbd_opt_lookup(reply->option));
 200         break;
 201
 202     case NBD_REP_ERR_UNKNOWN:
 203         error_setg(errp, "Requested export not available");
 204         break;
 205
 206     case NBD_REP_ERR_SHUTDOWN:
 207         error_setg(errp, "Server shutting down before option %" PRIx32 " (%s)",
 208                    reply->option, nbd_opt_lookup(reply->option));
 209         break;
 210
 211     case NBD_REP_ERR_BLOCK_SIZE_REQD:
 212         error_setg(errp, "Server requires INFO_BLOCK_SIZE for option %" PRIx32
 213                    " (%s)", reply->option, nbd_opt_lookup(reply->option));
 214         break;
 215
 216     default:
 217         error_setg(errp, "Unknown error code when asking for option %" PRIx32
 218                    " (%s)", reply->option, nbd_opt_lookup(reply->option));
 219         break;
 220     }
 221
 222     if (msg) {
 223         error_append_hint(errp, "server reported: %s\n", msg);
 224     }
 225
 226  cleanup:
 227     g_free(msg);
 228     if (result < 0) {
 229         nbd_send_opt_abort(ioc);
 230     }
 231     return result;
 232 }
 233
 234 /* Process another portion of the NBD_OPT_LIST reply.  Set *@match if
 235  * the current reply matches @want or if the server does not support
 236  * NBD_OPT_LIST, otherwise leave @match alone.  Return 0 if iteration
 237  * is complete, positive if more replies are expected, or negative
 238  * with @errp set if an unrecoverable error occurred. */
 239 static int nbd_receive_list(QIOChannel *ioc, const char *want, bool *match,
 240                             Error **errp)
 241 {
 242     nbd_opt_reply reply;
 243     uint32_t len;
 244     uint32_t namelen;
 245     char name[NBD_MAX_NAME_SIZE + 1];
 246     int error;
 247
 248     if (nbd_receive_option_reply(ioc, NBD_OPT_LIST, &reply, errp) < 0) {
 249         return -1;
 250     }
 251     error = nbd_handle_reply_err(ioc, &reply, errp);
 252     if (error <= 0) {
 253         /* The server did not support NBD_OPT_LIST, so set *match on
 254          * the assumption that any name will be accepted.  */
 255         *match = true;
 256         return error;
 257     }
 258     len = reply.length;
 259
 260     if (reply.type == NBD_REP_ACK) {
 261         if (len != 0) {
 262             error_setg(errp, "length too long for option end");
 263             nbd_send_opt_abort(ioc);
 264             return -1;
 265         }
 266         return 0;
 267     } else if (reply.type != NBD_REP_SERVER) {
 268         error_setg(errp, "Unexpected reply type %" PRIx32 " expected %x",
 269                    reply.type, NBD_REP_SERVER);
 270         nbd_send_opt_abort(ioc);
 271         return -1;
 272     }
 273
 274     if (len < sizeof(namelen) || len > NBD_MAX_BUFFER_SIZE) {
 275         error_setg(errp, "incorrect option length %" PRIu32, len);
 276         nbd_send_opt_abort(ioc);
 277         return -1;
 278     }
 279     if (nbd_read(ioc, &namelen, sizeof(namelen), errp) < 0) {
 280         error_prepend(errp, "failed to read option name length: ");
 281         nbd_send_opt_abort(ioc);
 282         return -1;
 283     }
 284     namelen = be32_to_cpu(namelen);
 285     len -= sizeof(namelen);
 286     if (len < namelen) {
 287         error_setg(errp, "incorrect option name length");
 288         nbd_send_opt_abort(ioc);
 289         return -1;
 290     }
 291     if (namelen != strlen(want)) {
 292         if (nbd_drop(ioc, len, errp) < 0) {
 293             error_prepend(errp,
 294                           "failed to skip export name with wrong length: ");
 295             nbd_send_opt_abort(ioc);
 296             return -1;
 297         }
 298         return 1;
 299     }
 300
 301     assert(namelen < sizeof(name));
 302     if (nbd_read(ioc, name, namelen, errp) < 0) {
 303         error_prepend(errp, "failed to read export name: ");
 304         nbd_send_opt_abort(ioc);
 305         return -1;
 306     }
 307     name[namelen] = '\0';
 308     len -= namelen;
 309     if (nbd_drop(ioc, len, errp) < 0) {
 310         error_prepend(errp, "failed to read export description: ");
 311         nbd_send_opt_abort(ioc);
 312         return -1;
 313     }
 314     if (!strcmp(name, want)) {
 315         *match = true;
 316     }
 317     return 1;
 318 }
 319
 320
 321 /* Returns -1 if NBD_OPT_GO proves the export @wantname cannot be
 322  * used, 0 if NBD_OPT_GO is unsupported (fall back to NBD_OPT_LIST and
 323  * NBD_OPT_EXPORT_NAME in that case), and > 0 if the export is good to
 324  * go (with @info populated). */
 325 static int nbd_opt_go(QIOChannel *ioc, const char *wantname,
 326                       NBDExportInfo *info, Error **errp)
 327 {
 328     nbd_opt_reply reply;
 329     uint32_t len = strlen(wantname);
 330     uint16_t type;
 331     int error;
 332     char *buf;
 333
 334     /* The protocol requires that the server send NBD_INFO_EXPORT with
 335      * a non-zero flags (at least NBD_FLAG_HAS_FLAGS must be set); so
 336      * flags still 0 is a witness of a broken server. */
 337     info->flags = 0;
 338
 339     trace_nbd_opt_go_start(wantname);
 340     buf = g_malloc(4 + len + 2 + 2 * info->request_sizes + 1);
 341     stl_be_p(buf, len);
 342     memcpy(buf + 4, wantname, len);
 343     /* At most one request, everything else up to server */
 344     stw_be_p(buf + 4 + len, info->request_sizes);
 345     if (info->request_sizes) {
 346         stw_be_p(buf + 4 + len + 2, NBD_INFO_BLOCK_SIZE);
 347     }
 348     error = nbd_send_option_request(ioc, NBD_OPT_GO,
 349                                     4 + len + 2 + 2 * info->request_sizes,
 350                                     buf, errp);
 351     g_free(buf);
 352     if (error < 0) {
 353         return -1;
 354     }
 355
 356     while (1) {
 357         if (nbd_receive_option_reply(ioc, NBD_OPT_GO, &reply, errp) < 0) {
 358             return -1;
 359         }
 360         error = nbd_handle_reply_err(ioc, &reply, errp);
 361         if (error <= 0) {
 362             return error;
 363         }
 364         len = reply.length;
 365
 366         if (reply.type == NBD_REP_ACK) {
 367             /* Server is done sending info and moved into transmission
 368                phase, but make sure it sent flags */
 369             if (len) {
 370                 error_setg(errp, "server sent invalid NBD_REP_ACK");
 371                 return -1;
 372             }
 373             if (!info->flags) {
 374                 error_setg(errp, "broken server omitted NBD_INFO_EXPORT");
 375                 return -1;
 376             }
 377             trace_nbd_opt_go_success();
 378             return 1;
 379         }
 380         if (reply.type != NBD_REP_INFO) {
 381             error_setg(errp, "unexpected reply type %" PRIx32
 382                        " (%s), expected %x",
 383                        reply.type, nbd_rep_lookup(reply.type), NBD_REP_INFO);
 384             nbd_send_opt_abort(ioc);
 385             return -1;
 386         }
 387         if (len < sizeof(type)) {
 388             error_setg(errp, "NBD_REP_INFO length %" PRIu32 " is too short",
 389                        len);
 390             nbd_send_opt_abort(ioc);
 391             return -1;
 392         }
 393         if (nbd_read(ioc, &type, sizeof(type), errp) < 0) {
 394             error_prepend(errp, "failed to read info type: ");
 395             nbd_send_opt_abort(ioc);
 396             return -1;
 397         }
 398         len -= sizeof(type);
 399         be16_to_cpus(&type);
 400         switch (type) {
 401         case NBD_INFO_EXPORT:
 402             if (len != sizeof(info->size) + sizeof(info->flags)) {
 403                 error_setg(errp, "remaining export info len %" PRIu32
 404                            " is unexpected size", len);
 405                 nbd_send_opt_abort(ioc);
 406                 return -1;
 407             }
 408             if (nbd_read(ioc, &info->size, sizeof(info->size), errp) < 0) {
 409                 error_prepend(errp, "failed to read info size: ");
 410                 nbd_send_opt_abort(ioc);
 411                 return -1;
 412             }
 413             be64_to_cpus(&info->size);
 414             if (nbd_read(ioc, &info->flags, sizeof(info->flags), errp) < 0) {
 415                 error_prepend(errp, "failed to read info flags: ");
 416                 nbd_send_opt_abort(ioc);
 417                 return -1;
 418             }
 419             be16_to_cpus(&info->flags);
 420             trace_nbd_receive_negotiate_size_flags(info->size, info->flags);
 421             break;
 422
 423         case NBD_INFO_BLOCK_SIZE:
 424             if (len != sizeof(info->min_block) * 3) {
 425                 error_setg(errp, "remaining export info len %" PRIu32
 426                            " is unexpected size", len);
 427                 nbd_send_opt_abort(ioc);
 428                 return -1;
 429             }
 430             if (nbd_read(ioc, &info->min_block, sizeof(info->min_block),
 431                          errp) < 0) {
 432                 error_prepend(errp, "failed to read info minimum block size: ");
 433                 nbd_send_opt_abort(ioc);
 434                 return -1;
 435             }
 436             be32_to_cpus(&info->min_block);
 437             if (!is_power_of_2(info->min_block)) {
 438                 error_setg(errp, "server minimum block size %" PRId32
 439                            "is not a power of two", info->min_block);
 440                 nbd_send_opt_abort(ioc);
 441                 return -1;
 442             }
 443             if (nbd_read(ioc, &info->opt_block, sizeof(info->opt_block),
 444                          errp) < 0) {
 445                 error_prepend(errp,
 446                               "failed to read info preferred block size: ");
 447                 nbd_send_opt_abort(ioc);
 448                 return -1;
 449             }
 450             be32_to_cpus(&info->opt_block);
 451             if (!is_power_of_2(info->opt_block) ||
 452                 info->opt_block < info->min_block) {
 453                 error_setg(errp, "server preferred block size %" PRId32
 454                            "is not valid", info->opt_block);
 455                 nbd_send_opt_abort(ioc);
 456                 return -1;
 457             }
 458             if (nbd_read(ioc, &info->max_block, sizeof(info->max_block),
 459                          errp) < 0) {
 460                 error_prepend(errp, "failed to read info maximum block size: ");
 461                 nbd_send_opt_abort(ioc);
 462                 return -1;
 463             }
 464             be32_to_cpus(&info->max_block);
 465             trace_nbd_opt_go_info_block_size(info->min_block, info->opt_block,
 466                                              info->max_block);
 467             break;
 468
 469         default:
 470             trace_nbd_opt_go_info_unknown(type, nbd_info_lookup(type));
 471             if (nbd_drop(ioc, len, errp) < 0) {
 472                 error_prepend(errp, "Failed to read info payload: ");
 473                 nbd_send_opt_abort(ioc);
 474                 return -1;
 475             }
 476             break;
 477         }
 478     }
 479 }
 480
 481 /* Return -1 on failure, 0 if wantname is an available export. */
 482 static int nbd_receive_query_exports(QIOChannel *ioc,
 483                                      const char *wantname,
 484                                      Error **errp)
 485 {
 486     bool foundExport = false;
 487
 488     trace_nbd_receive_query_exports_start(wantname);
 489     if (nbd_send_option_request(ioc, NBD_OPT_LIST, 0, NULL, errp) < 0) {
 490         return -1;
 491     }
 492
 493     while (1) {
 494         int ret = nbd_receive_list(ioc, wantname, &foundExport, errp);
 495
 496         if (ret < 0) {
 497             /* Server gave unexpected reply */
 498             return -1;
 499         } else if (ret == 0) {
 500             /* Done iterating. */
 501             if (!foundExport) {
 502                 error_setg(errp, "No export with name '%s' available",
 503                            wantname);
 504                 nbd_send_opt_abort(ioc);
 505                 return -1;
 506             }
 507             trace_nbd_receive_query_exports_success(wantname);
 508             return 0;
 509         }
 510     }
 511 }
 512
 513 /* nbd_request_simple_option: Send an option request, and parse the reply
 514  * return 1 for successful negotiation,
 515  *        0 if operation is unsupported,
 516  *        -1 with errp set for any other error
 517  */
 518 static int nbd_request_simple_option(QIOChannel *ioc, int opt, Error **errp)
 519 {
 520     nbd_opt_reply reply;
 521     int error;
 522
 523     if (nbd_send_option_request(ioc, opt, 0, NULL, errp) < 0) {
 524         return -1;
 525     }
 526
 527     if (nbd_receive_option_reply(ioc, opt, &reply, errp) < 0) {
 528         return -1;
 529     }
 530     error = nbd_handle_reply_err(ioc, &reply, errp);
 531     if (error <= 0) {
 532         return error;
 533     }
 534
 535     if (reply.type != NBD_REP_ACK) {
 536         error_setg(errp, "Server answered option %d (%s) with unexpected "
 537                    "reply %" PRIx32 " (%s)", opt, nbd_opt_lookup(opt),
 538                    reply.type, nbd_rep_lookup(reply.type));
 539         nbd_send_opt_abort(ioc);
 540         return -1;
 541     }
 542
 543     if (reply.length != 0) {
 544         error_setg(errp, "Option %d ('%s') response length is %" PRIu32
 545                    " (it should be zero)", opt, nbd_opt_lookup(opt),
 546                    reply.length);
 547         nbd_send_opt_abort(ioc);
 548         return -1;
 549     }
 550
 551     return 1;
 552 }
 553
 554 static QIOChannel *nbd_receive_starttls(QIOChannel *ioc,
 555                                         QCryptoTLSCreds *tlscreds,
 556                                         const char *hostname, Error **errp)
 557 {
 558     int ret;
 559     QIOChannelTLS *tioc;
 560     struct NBDTLSHandshakeData data = { 0 };
 561
 562     ret = nbd_request_simple_option(ioc, NBD_OPT_STARTTLS, errp);
 563     if (ret <= 0) {
 564         if (ret == 0) {
 565             error_setg(errp, "Server don't support STARTTLS option");
 566             nbd_send_opt_abort(ioc);
 567         }
 568         return NULL;
 569     }
 570
 571     trace_nbd_receive_starttls_new_client();
 572     tioc = qio_channel_tls_new_client(ioc, tlscreds, hostname, errp);
 573     if (!tioc) {
 574         return NULL;
 575     }
 576     qio_channel_set_name(QIO_CHANNEL(tioc), "nbd-client-tls");
 577     data.loop = g_main_loop_new(g_main_context_default(), FALSE);
 578     trace_nbd_receive_starttls_tls_handshake();
 579     qio_channel_tls_handshake(tioc,
 580                               nbd_tls_handshake,
 581                               &data,
 582                               NULL);
 583
 584     if (!data.complete) {
 585         g_main_loop_run(data.loop);
 586     }
 587     g_main_loop_unref(data.loop);
 588     if (data.error) {
 589         error_propagate(errp, data.error);
 590         object_unref(OBJECT(tioc));
 591         return NULL;
 592     }
 593
 594     return QIO_CHANNEL(tioc);
 595 }
 596
 597
 598 int nbd_receive_negotiate(QIOChannel *ioc, const char *name,
 599                           QCryptoTLSCreds *tlscreds, const char *hostname,
 600                           QIOChannel **outioc, NBDExportInfo *info,
 601                           Error **errp)
 602 {
 603     char buf[256];
 604     uint64_t magic;
 605     int rc;
 606     bool zeroes = true;
 607     bool structured_reply = info->structured_reply;
 608
 609     trace_nbd_receive_negotiate(tlscreds, hostname ? hostname : "<null>");
 610
 611     info->structured_reply = false;
 612     rc = -EINVAL;
 613
 614     if (outioc) {
 615         *outioc = NULL;
 616     }
 617     if (tlscreds && !outioc) {
 618         error_setg(errp, "Output I/O channel required for TLS");
 619         goto fail;
 620     }
 621
 622     if (nbd_read(ioc, buf, 8, errp) < 0) {
 623         error_prepend(errp, "Failed to read data: ");
 624         goto fail;
 625     }
 626
 627     buf[8] = '\0';
 628     if (strlen(buf) == 0) {
 629         error_setg(errp, "Server connection closed unexpectedly");
 630         goto fail;
 631     }
 632
 633     magic = ldq_be_p(buf);
 634     trace_nbd_receive_negotiate_magic(magic);
 635
 636     if (memcmp(buf, "NBDMAGIC", 8) != 0) {
 637         error_setg(errp, "Invalid magic received");
 638         goto fail;
 639     }
 640
 641     if (nbd_read(ioc, &magic, sizeof(magic), errp) < 0) {
 642         error_prepend(errp, "Failed to read magic: ");
 643         goto fail;
 644     }
 645     magic = be64_to_cpu(magic);
 646     trace_nbd_receive_negotiate_magic(magic);
 647
 648     if (magic == NBD_OPTS_MAGIC) {
 649         uint32_t clientflags = 0;
 650         uint16_t globalflags;
 651         bool fixedNewStyle = false;
 652
 653         if (nbd_read(ioc, &globalflags, sizeof(globalflags), errp) < 0) {
 654             error_prepend(errp, "Failed to read server flags: ");
 655             goto fail;
 656         }
 657         globalflags = be16_to_cpu(globalflags);
 658         trace_nbd_receive_negotiate_server_flags(globalflags);
 659         if (globalflags & NBD_FLAG_FIXED_NEWSTYLE) {
 660             fixedNewStyle = true;
 661             clientflags |= NBD_FLAG_C_FIXED_NEWSTYLE;
 662         }
 663         if (globalflags & NBD_FLAG_NO_ZEROES) {
 664             zeroes = false;
 665             clientflags |= NBD_FLAG_C_NO_ZEROES;
 666         }
 667         /* client requested flags */
 668         clientflags = cpu_to_be32(clientflags);
 669         if (nbd_write(ioc, &clientflags, sizeof(clientflags), errp) < 0) {
 670             error_prepend(errp, "Failed to send clientflags field: ");
 671             goto fail;
 672         }
 673         if (tlscreds) {
 674             if (fixedNewStyle) {
 675                 *outioc = nbd_receive_starttls(ioc, tlscreds, hostname, errp);
 676                 if (!*outioc) {
 677                     goto fail;
 678                 }
 679                 ioc = *outioc;
 680             } else {
 681                 error_setg(errp, "Server does not support STARTTLS");
 682                 goto fail;
 683             }
 684         }
 685         if (!name) {
 686             trace_nbd_receive_negotiate_default_name();
 687             name = "";
 688         }
 689         if (fixedNewStyle) {
 690             int result;
 691
 692             if (structured_reply) {
 693                 result = nbd_request_simple_option(ioc,
 694                                                    NBD_OPT_STRUCTURED_REPLY,
 695                                                    errp);
 696                 if (result < 0) {
 697                     goto fail;
 698                 }
 699                 info->structured_reply = result == 1;
 700             }
 701
 702             /* Try NBD_OPT_GO first - if it works, we are done (it
 703              * also gives us a good message if the server requires
 704              * TLS).  If it is not available, fall back to
 705              * NBD_OPT_LIST for nicer error messages about a missing
 706              * export, then use NBD_OPT_EXPORT_NAME.  */
 707             result = nbd_opt_go(ioc, name, info, errp);
 708             if (result < 0) {
 709                 goto fail;
 710             }
 711             if (result > 0) {
 712                 return 0;
 713             }
 714             /* Check our desired export is present in the
 715              * server export list. Since NBD_OPT_EXPORT_NAME
 716              * cannot return an error message, running this
 717              * query gives us better error reporting if the
 718              * export name is not available.
 719              */
 720             if (nbd_receive_query_exports(ioc, name, errp) < 0) {
 721                 goto fail;
 722             }
 723         }
 724         /* write the export name request */
 725         if (nbd_send_option_request(ioc, NBD_OPT_EXPORT_NAME, -1, name,
 726                                     errp) < 0) {
 727             goto fail;
 728         }
 729
 730         /* Read the response */
 731         if (nbd_read(ioc, &info->size, sizeof(info->size), errp) < 0) {
 732             error_prepend(errp, "Failed to read export length: ");
 733             goto fail;
 734         }
 735         be64_to_cpus(&info->size);
 736
 737         if (nbd_read(ioc, &info->flags, sizeof(info->flags), errp) < 0) {
 738             error_prepend(errp, "Failed to read export flags: ");
 739             goto fail;
 740         }
 741         be16_to_cpus(&info->flags);
 742     } else if (magic == NBD_CLIENT_MAGIC) {
 743         uint32_t oldflags;
 744
 745         if (name) {
 746             error_setg(errp, "Server does not support export names");
 747             goto fail;
 748         }
 749         if (tlscreds) {
 750             error_setg(errp, "Server does not support STARTTLS");
 751             goto fail;
 752         }
 753
 754         if (nbd_read(ioc, &info->size, sizeof(info->size), errp) < 0) {
 755             error_prepend(errp, "Failed to read export length: ");
 756             goto fail;
 757         }
 758         be64_to_cpus(&info->size);
 759
 760         if (nbd_read(ioc, &oldflags, sizeof(oldflags), errp) < 0) {
 761             error_prepend(errp, "Failed to read export flags: ");
 762             goto fail;
 763         }
 764         be32_to_cpus(&oldflags);
 765         if (oldflags & ~0xffff) {
 766             error_setg(errp, "Unexpected export flags %0x" PRIx32, oldflags);
 767             goto fail;
 768         }
 769         info->flags = oldflags;
 770     } else {
 771         error_setg(errp, "Bad magic received");
 772         goto fail;
 773     }
 774
 775     trace_nbd_receive_negotiate_size_flags(info->size, info->flags);
 776     if (zeroes && nbd_drop(ioc, 124, errp) < 0) {
 777         error_prepend(errp, "Failed to read reserved block: ");
 778         goto fail;
 779     }
 780     rc = 0;
 781
 782 fail:
 783     return rc;
 784 }
 785
 786 #ifdef __linux__
 787 int nbd_init(int fd, QIOChannelSocket *sioc, NBDExportInfo *info,
 788              Error **errp)
 789 {
 790     unsigned long sector_size = MAX(BDRV_SECTOR_SIZE, info->min_block);
 791     unsigned long sectors = info->size / sector_size;
 792
 793     /* FIXME: Once the kernel module is patched to honor block sizes,
 794      * and to advertise that fact to user space, we should update the
 795      * hand-off to the kernel to use any block sizes we learned. */
 796     assert(!info->request_sizes);
 797     if (info->size / sector_size != sectors) {
 798         error_setg(errp, "Export size %" PRIu64 " too large for 32-bit kernel",
 799                    info->size);
 800         return -E2BIG;
 801     }
 802
 803     trace_nbd_init_set_socket();
 804
 805     if (ioctl(fd, NBD_SET_SOCK, (unsigned long) sioc->fd) < 0) {
 806         int serrno = errno;
 807         error_setg(errp, "Failed to set NBD socket");
 808         return -serrno;
 809     }
 810
 811     trace_nbd_init_set_block_size(sector_size);
 812
 813     if (ioctl(fd, NBD_SET_BLKSIZE, sector_size) < 0) {
 814         int serrno = errno;
 815         error_setg(errp, "Failed setting NBD block size");
 816         return -serrno;
 817     }
 818
 819     trace_nbd_init_set_size(sectors);
 820     if (info->size % sector_size) {
 821         trace_nbd_init_trailing_bytes(info->size % sector_size);
 822     }
 823
 824     if (ioctl(fd, NBD_SET_SIZE_BLOCKS, sectors) < 0) {
 825         int serrno = errno;
 826         error_setg(errp, "Failed setting size (in blocks)");
 827         return -serrno;
 828     }
 829
 830     if (ioctl(fd, NBD_SET_FLAGS, (unsigned long) info->flags) < 0) {
 831         if (errno == ENOTTY) {
 832             int read_only = (info->flags & NBD_FLAG_READ_ONLY) != 0;
 833             trace_nbd_init_set_readonly();
 834
 835             if (ioctl(fd, BLKROSET, (unsigned long) &read_only) < 0) {
 836                 int serrno = errno;
 837                 error_setg(errp, "Failed setting read-only attribute");
 838                 return -serrno;
 839             }
 840         } else {
 841             int serrno = errno;
 842             error_setg(errp, "Failed setting flags");
 843             return -serrno;
 844         }
 845     }
 846
 847     trace_nbd_init_finish();
 848
 849     return 0;
 850 }
 851
 852 int nbd_client(int fd)
 853 {
 854     int ret;
 855     int serrno;
 856
 857     trace_nbd_client_loop();
 858
 859     ret = ioctl(fd, NBD_DO_IT);
 860     if (ret < 0 && errno == EPIPE) {
 861         /* NBD_DO_IT normally returns EPIPE when someone has disconnected
 862          * the socket via NBD_DISCONNECT.  We do not want to return 1 in
 863          * that case.
 864          */
 865         ret = 0;
 866     }
 867     serrno = errno;
 868
 869     trace_nbd_client_loop_ret(ret, strerror(serrno));
 870
 871     trace_nbd_client_clear_queue();
 872     ioctl(fd, NBD_CLEAR_QUE);
 873
 874     trace_nbd_client_clear_socket();
 875     ioctl(fd, NBD_CLEAR_SOCK);
 876
 877     errno = serrno;
 878     return ret;
 879 }
 880
 881 int nbd_disconnect(int fd)
 882 {
 883     ioctl(fd, NBD_CLEAR_QUE);
 884     ioctl(fd, NBD_DISCONNECT);
 885     ioctl(fd, NBD_CLEAR_SOCK);
 886     return 0;
 887 }
 888
 889 #else
 890 int nbd_init(int fd, QIOChannelSocket *ioc, NBDExportInfo *info,
 891              Error **errp)
 892 {
 893     error_setg(errp, "nbd_init is only supported on Linux");
 894     return -ENOTSUP;
 895 }
 896
 897 int nbd_client(int fd)
 898 {
 899     return -ENOTSUP;
 900 }
 901 int nbd_disconnect(int fd)
 902 {
 903     return -ENOTSUP;
 904 }
 905 #endif
 906
 907 int nbd_send_request(QIOChannel *ioc, NBDRequest *request)
 908 {
 909     uint8_t buf[NBD_REQUEST_SIZE];
 910
 911     trace_nbd_send_request(request->from, request->len, request->handle,
 912                            request->flags, request->type,
 913                            nbd_cmd_lookup(request->type));
 914
 915     stl_be_p(buf, NBD_REQUEST_MAGIC);
 916     stw_be_p(buf + 4, request->flags);
 917     stw_be_p(buf + 6, request->type);
 918     stq_be_p(buf + 8, request->handle);
 919     stq_be_p(buf + 16, request->from);
 920     stl_be_p(buf + 24, request->len);
 921
 922     return nbd_write(ioc, buf, sizeof(buf), NULL);
 923 }
 924
 925 /* nbd_receive_simple_reply
 926  * Read simple reply except magic field (which should be already read).
 927  * Payload is not read (payload is possible for CMD_READ, but here we even
 928  * don't know whether it take place or not).
 929  */
 930 static int nbd_receive_simple_reply(QIOChannel *ioc, NBDSimpleReply *reply,
 931                                     Error **errp)
 932 {
 933     int ret;
 934
 935     assert(reply->magic == NBD_SIMPLE_REPLY_MAGIC);
 936
 937     ret = nbd_read(ioc, (uint8_t *)reply + sizeof(reply->magic),
 938                    sizeof(*reply) - sizeof(reply->magic), errp);
 939     if (ret < 0) {
 940         return ret;
 941     }
 942
 943     be32_to_cpus(&reply->error);
 944     be64_to_cpus(&reply->handle);
 945
 946     return 0;
 947 }
 948
 949 /* nbd_receive_structured_reply_chunk
 950  * Read structured reply chunk except magic field (which should be already
 951  * read).
 952  * Payload is not read.
 953  */
 954 static int nbd_receive_structured_reply_chunk(QIOChannel *ioc,
 955                                               NBDStructuredReplyChunk *chunk,
 956                                               Error **errp)
 957 {
 958     int ret;
 959
 960     assert(chunk->magic == NBD_STRUCTURED_REPLY_MAGIC);
 961
 962     ret = nbd_read(ioc, (uint8_t *)chunk + sizeof(chunk->magic),
 963                    sizeof(*chunk) - sizeof(chunk->magic), errp);
 964     if (ret < 0) {
 965         return ret;
 966     }
 967
 968     be16_to_cpus(&chunk->flags);
 969     be16_to_cpus(&chunk->type);
 970     be64_to_cpus(&chunk->handle);
 971     be32_to_cpus(&chunk->length);
 972
 973     return 0;
 974 }
 975
 976 /* nbd_receive_reply
 977  * Returns 1 on success
 978  *         0 on eof, when no data was read (errp is not set)
 979  *         negative errno on failure (errp is set)
 980  */
 981 int nbd_receive_reply(QIOChannel *ioc, NBDReply *reply, Error **errp)
 982 {
 983     int ret;
 984     const char *type;
 985
 986     ret = nbd_read_eof(ioc, &reply->magic, sizeof(reply->magic), errp);
 987     if (ret <= 0) {
 988         return ret;
 989     }
 990
 991     be32_to_cpus(&reply->magic);
 992
 993     switch (reply->magic) {
 994     case NBD_SIMPLE_REPLY_MAGIC:
 995         ret = nbd_receive_simple_reply(ioc, &reply->simple, errp);
 996         if (ret < 0) {
 997             break;
 998         }
 999         trace_nbd_receive_simple_reply(reply->simple.error,
1000                                        nbd_err_lookup(reply->simple.error),
1001                                        reply->handle);
1002         break;
1003     case NBD_STRUCTURED_REPLY_MAGIC:
1004         ret = nbd_receive_structured_reply_chunk(ioc, &reply->structured, errp);
1005         if (ret < 0) {
1006             break;
1007         }
1008         type = nbd_reply_type_lookup(reply->structured.type);
1009         trace_nbd_receive_structured_reply_chunk(reply->structured.flags,
1010                                                  reply->structured.type, type,
1011                                                  reply->structured.handle,
1012                                                  reply->structured.length);
1013         break;
1014     default:
1015         error_setg(errp, "invalid magic (got 0x%" PRIx32 ")", reply->magic);
1016         return -EINVAL;
1017     }
1018     if (ret < 0) {
1019         return ret;
1020     }
1021
1022     return 1;
1023 }
1024