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 for option %" PRIx32
 236                    " (%s)", reply->option, nbd_opt_lookup(reply->option));
 237         break;
 238
 239     case NBD_REP_ERR_SHUTDOWN:
 240         error_setg(errp, "Server shutting down before option %" PRIx32 " (%s)",
 241                    reply->option, nbd_opt_lookup(reply->option));
 242         break;
 243
 244     case NBD_REP_ERR_BLOCK_SIZE_REQD:
 245         error_setg(errp, "Server requires INFO_BLOCK_SIZE for option %" PRIx32
 246                    " (%s)", reply->option, nbd_opt_lookup(reply->option));
 247         break;
 248
 249     default:
 250         error_setg(errp, "Unknown error code when asking for option %" PRIx32
 251                    " (%s)", reply->option, nbd_opt_lookup(reply->option));
 252         break;
 253     }
 254
 255     if (msg) {
 256         error_append_hint(errp, "%s\n", msg);
 257     }
 258
 259  cleanup:
 260     g_free(msg);
 261     if (result < 0) {
 262         nbd_send_opt_abort(ioc);
 263     }
 264     return result;
 265 }
 266
 267 /* Process another portion of the NBD_OPT_LIST reply.  Set *@match if
 268  * the current reply matches @want or if the server does not support
 269  * NBD_OPT_LIST, otherwise leave @match alone.  Return 0 if iteration
 270  * is complete, positive if more replies are expected, or negative
 271  * with @errp set if an unrecoverable error occurred. */
 272 static int nbd_receive_list(QIOChannel *ioc, const char *want, bool *match,
 273                             Error **errp)
 274 {
 275     nbd_opt_reply reply;
 276     uint32_t len;
 277     uint32_t namelen;
 278     char name[NBD_MAX_NAME_SIZE + 1];
 279     int error;
 280
 281     if (nbd_receive_option_reply(ioc, NBD_OPT_LIST, &reply, errp) < 0) {
 282         return -1;
 283     }
 284     error = nbd_handle_reply_err(ioc, &reply, errp);
 285     if (error <= 0) {
 286         /* The server did not support NBD_OPT_LIST, so set *match on
 287          * the assumption that any name will be accepted.  */
 288         *match = true;
 289         return error;
 290     }
 291     len = reply.length;
 292
 293     if (reply.type == NBD_REP_ACK) {
 294         if (len != 0) {
 295             error_setg(errp, "length too long for option end");
 296             nbd_send_opt_abort(ioc);
 297             return -1;
 298         }
 299         return 0;
 300     } else if (reply.type != NBD_REP_SERVER) {
 301         error_setg(errp, "Unexpected reply type %" PRIx32 " expected %x",
 302                    reply.type, NBD_REP_SERVER);
 303         nbd_send_opt_abort(ioc);
 304         return -1;
 305     }
 306
 307     if (len < sizeof(namelen) || len > NBD_MAX_BUFFER_SIZE) {
 308         error_setg(errp, "incorrect option length %" PRIu32, len);
 309         nbd_send_opt_abort(ioc);
 310         return -1;
 311     }
 312     if (nbd_read(ioc, &namelen, sizeof(namelen), errp) < 0) {
 313         error_prepend(errp, "failed to read option name length");
 314         nbd_send_opt_abort(ioc);
 315         return -1;
 316     }
 317     namelen = be32_to_cpu(namelen);
 318     len -= sizeof(namelen);
 319     if (len < namelen) {
 320         error_setg(errp, "incorrect option name length");
 321         nbd_send_opt_abort(ioc);
 322         return -1;
 323     }
 324     if (namelen != strlen(want)) {
 325         if (nbd_drop(ioc, len, errp) < 0) {
 326             error_prepend(errp, "failed to skip export name with wrong length");
 327             nbd_send_opt_abort(ioc);
 328             return -1;
 329         }
 330         return 1;
 331     }
 332
 333     assert(namelen < sizeof(name));
 334     if (nbd_read(ioc, name, namelen, errp) < 0) {
 335         error_prepend(errp, "failed to read export name");
 336         nbd_send_opt_abort(ioc);
 337         return -1;
 338     }
 339     name[namelen] = '\0';
 340     len -= namelen;
 341     if (nbd_drop(ioc, len, errp) < 0) {
 342         error_prepend(errp, "failed to read export description");
 343         nbd_send_opt_abort(ioc);
 344         return -1;
 345     }
 346     if (!strcmp(name, want)) {
 347         *match = true;
 348     }
 349     return 1;
 350 }
 351
 352
 353 /* Returns -1 if NBD_OPT_GO proves the export @wantname cannot be
 354  * used, 0 if NBD_OPT_GO is unsupported (fall back to NBD_OPT_LIST and
 355  * NBD_OPT_EXPORT_NAME in that case), and > 0 if the export is good to
 356  * go (with @info populated). */
 357 static int nbd_opt_go(QIOChannel *ioc, const char *wantname,
 358                       NBDExportInfo *info, Error **errp)
 359 {
 360     nbd_opt_reply reply;
 361     uint32_t len = strlen(wantname);
 362     uint16_t type;
 363     int error;
 364     char *buf;
 365
 366     /* The protocol requires that the server send NBD_INFO_EXPORT with
 367      * a non-zero flags (at least NBD_FLAG_HAS_FLAGS must be set); so
 368      * flags still 0 is a witness of a broken server. */
 369     info->flags = 0;
 370
 371     trace_nbd_opt_go_start(wantname);
 372     buf = g_malloc(4 + len + 2 + 1);
 373     stl_be_p(buf, len);
 374     memcpy(buf + 4, wantname, len);
 375     /* No requests, live with whatever server sends */
 376     stw_be_p(buf + 4 + len, 0);
 377     if (nbd_send_option_request(ioc, NBD_OPT_GO, len + 6, buf, errp) < 0) {
 378         return -1;
 379     }
 380
 381     while (1) {
 382         if (nbd_receive_option_reply(ioc, NBD_OPT_GO, &reply, errp) < 0) {
 383             return -1;
 384         }
 385         error = nbd_handle_reply_err(ioc, &reply, errp);
 386         if (error <= 0) {
 387             return error;
 388         }
 389         len = reply.length;
 390
 391         if (reply.type == NBD_REP_ACK) {
 392             /* Server is done sending info and moved into transmission
 393                phase, but make sure it sent flags */
 394             if (len) {
 395                 error_setg(errp, "server sent invalid NBD_REP_ACK");
 396                 nbd_send_opt_abort(ioc);
 397                 return -1;
 398             }
 399             if (!info->flags) {
 400                 error_setg(errp, "broken server omitted NBD_INFO_EXPORT");
 401                 nbd_send_opt_abort(ioc);
 402                 return -1;
 403             }
 404             trace_nbd_opt_go_success();
 405             return 1;
 406         }
 407         if (reply.type != NBD_REP_INFO) {
 408             error_setg(errp, "unexpected reply type %" PRIx32 ", expected %x",
 409                        reply.type, NBD_REP_INFO);
 410             nbd_send_opt_abort(ioc);
 411             return -1;
 412         }
 413         if (len < sizeof(type)) {
 414             error_setg(errp, "NBD_REP_INFO length %" PRIu32 " is too short",
 415                        len);
 416             nbd_send_opt_abort(ioc);
 417             return -1;
 418         }
 419         if (nbd_read(ioc, &type, sizeof(type), errp) < 0) {
 420             error_prepend(errp, "failed to read info type");
 421             nbd_send_opt_abort(ioc);
 422             return -1;
 423         }
 424         len -= sizeof(type);
 425         be16_to_cpus(&type);
 426         switch (type) {
 427         case NBD_INFO_EXPORT:
 428             if (len != sizeof(info->size) + sizeof(info->flags)) {
 429                 error_setg(errp, "remaining export info len %" PRIu32
 430                            " is unexpected size", len);
 431                 nbd_send_opt_abort(ioc);
 432                 return -1;
 433             }
 434             if (nbd_read(ioc, &info->size, sizeof(info->size), errp) < 0) {
 435                 error_prepend(errp, "failed to read info size");
 436                 nbd_send_opt_abort(ioc);
 437                 return -1;
 438             }
 439             be64_to_cpus(&info->size);
 440             if (nbd_read(ioc, &info->flags, sizeof(info->flags), errp) < 0) {
 441                 error_prepend(errp, "failed to read info flags");
 442                 nbd_send_opt_abort(ioc);
 443                 return -1;
 444             }
 445             be16_to_cpus(&info->flags);
 446             trace_nbd_receive_negotiate_size_flags(info->size, info->flags);
 447             break;
 448
 449         default:
 450             trace_nbd_opt_go_info_unknown(type, nbd_info_lookup(type));
 451             if (nbd_drop(ioc, len, errp) < 0) {
 452                 error_prepend(errp, "Failed to read info payload");
 453                 nbd_send_opt_abort(ioc);
 454                 return -1;
 455             }
 456             break;
 457         }
 458     }
 459 }
 460
 461 /* Return -1 on failure, 0 if wantname is an available export. */
 462 static int nbd_receive_query_exports(QIOChannel *ioc,
 463                                      const char *wantname,
 464                                      Error **errp)
 465 {
 466     bool foundExport = false;
 467
 468     trace_nbd_receive_query_exports_start(wantname);
 469     if (nbd_send_option_request(ioc, NBD_OPT_LIST, 0, NULL, errp) < 0) {
 470         return -1;
 471     }
 472
 473     while (1) {
 474         int ret = nbd_receive_list(ioc, wantname, &foundExport, errp);
 475
 476         if (ret < 0) {
 477             /* Server gave unexpected reply */
 478             return -1;
 479         } else if (ret == 0) {
 480             /* Done iterating. */
 481             if (!foundExport) {
 482                 error_setg(errp, "No export with name '%s' available",
 483                            wantname);
 484                 nbd_send_opt_abort(ioc);
 485                 return -1;
 486             }
 487             trace_nbd_receive_query_exports_success(wantname);
 488             return 0;
 489         }
 490     }
 491 }
 492
 493 static QIOChannel *nbd_receive_starttls(QIOChannel *ioc,
 494                                         QCryptoTLSCreds *tlscreds,
 495                                         const char *hostname, Error **errp)
 496 {
 497     nbd_opt_reply reply;
 498     QIOChannelTLS *tioc;
 499     struct NBDTLSHandshakeData data = { 0 };
 500
 501     trace_nbd_receive_starttls_request();
 502     if (nbd_send_option_request(ioc, NBD_OPT_STARTTLS, 0, NULL, errp) < 0) {
 503         return NULL;
 504     }
 505
 506     trace_nbd_receive_starttls_reply();
 507     if (nbd_receive_option_reply(ioc, NBD_OPT_STARTTLS, &reply, errp) < 0) {
 508         return NULL;
 509     }
 510
 511     if (reply.type != NBD_REP_ACK) {
 512         error_setg(errp, "Server rejected request to start TLS %" PRIx32,
 513                    reply.type);
 514         nbd_send_opt_abort(ioc);
 515         return NULL;
 516     }
 517
 518     if (reply.length != 0) {
 519         error_setg(errp, "Start TLS response was not zero %" PRIu32,
 520                    reply.length);
 521         nbd_send_opt_abort(ioc);
 522         return NULL;
 523     }
 524
 525     trace_nbd_receive_starttls_new_client();
 526     tioc = qio_channel_tls_new_client(ioc, tlscreds, hostname, errp);
 527     if (!tioc) {
 528         return NULL;
 529     }
 530     qio_channel_set_name(QIO_CHANNEL(tioc), "nbd-client-tls");
 531     data.loop = g_main_loop_new(g_main_context_default(), FALSE);
 532     trace_nbd_receive_starttls_tls_handshake();
 533     qio_channel_tls_handshake(tioc,
 534                               nbd_tls_handshake,
 535                               &data,
 536                               NULL);
 537
 538     if (!data.complete) {
 539         g_main_loop_run(data.loop);
 540     }
 541     g_main_loop_unref(data.loop);
 542     if (data.error) {
 543         error_propagate(errp, data.error);
 544         object_unref(OBJECT(tioc));
 545         return NULL;
 546     }
 547
 548     return QIO_CHANNEL(tioc);
 549 }
 550
 551
 552 int nbd_receive_negotiate(QIOChannel *ioc, const char *name,
 553                           QCryptoTLSCreds *tlscreds, const char *hostname,
 554                           QIOChannel **outioc, NBDExportInfo *info,
 555                           Error **errp)
 556 {
 557     char buf[256];
 558     uint64_t magic;
 559     int rc;
 560     bool zeroes = true;
 561
 562     trace_nbd_receive_negotiate(tlscreds, hostname ? hostname : "<null>");
 563
 564     rc = -EINVAL;
 565
 566     if (outioc) {
 567         *outioc = NULL;
 568     }
 569     if (tlscreds && !outioc) {
 570         error_setg(errp, "Output I/O channel required for TLS");
 571         goto fail;
 572     }
 573
 574     if (nbd_read(ioc, buf, 8, errp) < 0) {
 575         error_prepend(errp, "Failed to read data");
 576         goto fail;
 577     }
 578
 579     buf[8] = '\0';
 580     if (strlen(buf) == 0) {
 581         error_setg(errp, "Server connection closed unexpectedly");
 582         goto fail;
 583     }
 584
 585     magic = ldq_be_p(buf);
 586     trace_nbd_receive_negotiate_magic(magic);
 587
 588     if (memcmp(buf, "NBDMAGIC", 8) != 0) {
 589         error_setg(errp, "Invalid magic received");
 590         goto fail;
 591     }
 592
 593     if (nbd_read(ioc, &magic, sizeof(magic), errp) < 0) {
 594         error_prepend(errp, "Failed to read magic");
 595         goto fail;
 596     }
 597     magic = be64_to_cpu(magic);
 598     trace_nbd_receive_negotiate_magic(magic);
 599
 600     if (magic == NBD_OPTS_MAGIC) {
 601         uint32_t clientflags = 0;
 602         uint16_t globalflags;
 603         bool fixedNewStyle = false;
 604
 605         if (nbd_read(ioc, &globalflags, sizeof(globalflags), errp) < 0) {
 606             error_prepend(errp, "Failed to read server flags");
 607             goto fail;
 608         }
 609         globalflags = be16_to_cpu(globalflags);
 610         trace_nbd_receive_negotiate_server_flags(globalflags);
 611         if (globalflags & NBD_FLAG_FIXED_NEWSTYLE) {
 612             fixedNewStyle = true;
 613             clientflags |= NBD_FLAG_C_FIXED_NEWSTYLE;
 614         }
 615         if (globalflags & NBD_FLAG_NO_ZEROES) {
 616             zeroes = false;
 617             clientflags |= NBD_FLAG_C_NO_ZEROES;
 618         }
 619         /* client requested flags */
 620         clientflags = cpu_to_be32(clientflags);
 621         if (nbd_write(ioc, &clientflags, sizeof(clientflags), errp) < 0) {
 622             error_prepend(errp, "Failed to send clientflags field");
 623             goto fail;
 624         }
 625         if (tlscreds) {
 626             if (fixedNewStyle) {
 627                 *outioc = nbd_receive_starttls(ioc, tlscreds, hostname, errp);
 628                 if (!*outioc) {
 629                     goto fail;
 630                 }
 631                 ioc = *outioc;
 632             } else {
 633                 error_setg(errp, "Server does not support STARTTLS");
 634                 goto fail;
 635             }
 636         }
 637         if (!name) {
 638             trace_nbd_receive_negotiate_default_name();
 639             name = "";
 640         }
 641         if (fixedNewStyle) {
 642             int result;
 643
 644             /* Try NBD_OPT_GO first - if it works, we are done (it
 645              * also gives us a good message if the server requires
 646              * TLS).  If it is not available, fall back to
 647              * NBD_OPT_LIST for nicer error messages about a missing
 648              * export, then use NBD_OPT_EXPORT_NAME.  */
 649             result = nbd_opt_go(ioc, name, info, errp);
 650             if (result < 0) {
 651                 goto fail;
 652             }
 653             if (result > 0) {
 654                 return 0;
 655             }
 656             /* Check our desired export is present in the
 657              * server export list. Since NBD_OPT_EXPORT_NAME
 658              * cannot return an error message, running this
 659              * query gives us better error reporting if the
 660              * export name is not available.
 661              */
 662             if (nbd_receive_query_exports(ioc, name, errp) < 0) {
 663                 goto fail;
 664             }
 665         }
 666         /* write the export name request */
 667         if (nbd_send_option_request(ioc, NBD_OPT_EXPORT_NAME, -1, name,
 668                                     errp) < 0) {
 669             goto fail;
 670         }
 671
 672         /* Read the response */
 673         if (nbd_read(ioc, &info->size, sizeof(info->size), errp) < 0) {
 674             error_prepend(errp, "Failed to read export length");
 675             goto fail;
 676         }
 677         be64_to_cpus(&info->size);
 678
 679         if (nbd_read(ioc, &info->flags, sizeof(info->flags), errp) < 0) {
 680             error_prepend(errp, "Failed to read export flags");
 681             goto fail;
 682         }
 683         be16_to_cpus(&info->flags);
 684     } else if (magic == NBD_CLIENT_MAGIC) {
 685         uint32_t oldflags;
 686
 687         if (name) {
 688             error_setg(errp, "Server does not support export names");
 689             goto fail;
 690         }
 691         if (tlscreds) {
 692             error_setg(errp, "Server does not support STARTTLS");
 693             goto fail;
 694         }
 695
 696         if (nbd_read(ioc, &info->size, sizeof(info->size), errp) < 0) {
 697             error_prepend(errp, "Failed to read export length");
 698             goto fail;
 699         }
 700         be64_to_cpus(&info->size);
 701
 702         if (nbd_read(ioc, &oldflags, sizeof(oldflags), errp) < 0) {
 703             error_prepend(errp, "Failed to read export flags");
 704             goto fail;
 705         }
 706         be32_to_cpus(&oldflags);
 707         if (oldflags & ~0xffff) {
 708             error_setg(errp, "Unexpected export flags %0x" PRIx32, oldflags);
 709             goto fail;
 710         }
 711         info->flags = oldflags;
 712     } else {
 713         error_setg(errp, "Bad magic received");
 714         goto fail;
 715     }
 716
 717     trace_nbd_receive_negotiate_size_flags(info->size, info->flags);
 718     if (zeroes && nbd_drop(ioc, 124, errp) < 0) {
 719         error_prepend(errp, "Failed to read reserved block");
 720         goto fail;
 721     }
 722     rc = 0;
 723
 724 fail:
 725     return rc;
 726 }
 727
 728 #ifdef __linux__
 729 int nbd_init(int fd, QIOChannelSocket *sioc, NBDExportInfo *info,
 730              Error **errp)
 731 {
 732     unsigned long sectors = info->size / BDRV_SECTOR_SIZE;
 733     if (info->size / BDRV_SECTOR_SIZE != sectors) {
 734         error_setg(errp, "Export size %" PRIu64 " too large for 32-bit kernel",
 735                    info->size);
 736         return -E2BIG;
 737     }
 738
 739     trace_nbd_init_set_socket();
 740
 741     if (ioctl(fd, NBD_SET_SOCK, (unsigned long) sioc->fd) < 0) {
 742         int serrno = errno;
 743         error_setg(errp, "Failed to set NBD socket");
 744         return -serrno;
 745     }
 746
 747     trace_nbd_init_set_block_size(BDRV_SECTOR_SIZE);
 748
 749     if (ioctl(fd, NBD_SET_BLKSIZE, (unsigned long)BDRV_SECTOR_SIZE) < 0) {
 750         int serrno = errno;
 751         error_setg(errp, "Failed setting NBD block size");
 752         return -serrno;
 753     }
 754
 755     trace_nbd_init_set_size(sectors);
 756     if (info->size % BDRV_SECTOR_SIZE) {
 757         trace_nbd_init_trailing_bytes(info->size % BDRV_SECTOR_SIZE);
 758     }
 759
 760     if (ioctl(fd, NBD_SET_SIZE_BLOCKS, sectors) < 0) {
 761         int serrno = errno;
 762         error_setg(errp, "Failed setting size (in blocks)");
 763         return -serrno;
 764     }
 765
 766     if (ioctl(fd, NBD_SET_FLAGS, (unsigned long) info->flags) < 0) {
 767         if (errno == ENOTTY) {
 768             int read_only = (info->flags & NBD_FLAG_READ_ONLY) != 0;
 769             trace_nbd_init_set_readonly();
 770
 771             if (ioctl(fd, BLKROSET, (unsigned long) &read_only) < 0) {
 772                 int serrno = errno;
 773                 error_setg(errp, "Failed setting read-only attribute");
 774                 return -serrno;
 775             }
 776         } else {
 777             int serrno = errno;
 778             error_setg(errp, "Failed setting flags");
 779             return -serrno;
 780         }
 781     }
 782
 783     trace_nbd_init_finish();
 784
 785     return 0;
 786 }
 787
 788 int nbd_client(int fd)
 789 {
 790     int ret;
 791     int serrno;
 792
 793     trace_nbd_client_loop();
 794
 795     ret = ioctl(fd, NBD_DO_IT);
 796     if (ret < 0 && errno == EPIPE) {
 797         /* NBD_DO_IT normally returns EPIPE when someone has disconnected
 798          * the socket via NBD_DISCONNECT.  We do not want to return 1 in
 799          * that case.
 800          */
 801         ret = 0;
 802     }
 803     serrno = errno;
 804
 805     trace_nbd_client_loop_ret(ret, strerror(serrno));
 806
 807     trace_nbd_client_clear_queue();
 808     ioctl(fd, NBD_CLEAR_QUE);
 809
 810     trace_nbd_client_clear_socket();
 811     ioctl(fd, NBD_CLEAR_SOCK);
 812
 813     errno = serrno;
 814     return ret;
 815 }
 816
 817 int nbd_disconnect(int fd)
 818 {
 819     ioctl(fd, NBD_CLEAR_QUE);
 820     ioctl(fd, NBD_DISCONNECT);
 821     ioctl(fd, NBD_CLEAR_SOCK);
 822     return 0;
 823 }
 824
 825 #else
 826 int nbd_init(int fd, QIOChannelSocket *ioc, NBDExportInfo *info,
 827              Error **errp)
 828 {
 829     error_setg(errp, "nbd_init is only supported on Linux");
 830     return -ENOTSUP;
 831 }
 832
 833 int nbd_client(int fd)
 834 {
 835     return -ENOTSUP;
 836 }
 837 int nbd_disconnect(int fd)
 838 {
 839     return -ENOTSUP;
 840 }
 841 #endif
 842
 843 ssize_t nbd_send_request(QIOChannel *ioc, NBDRequest *request)
 844 {
 845     uint8_t buf[NBD_REQUEST_SIZE];
 846
 847     trace_nbd_send_request(request->from, request->len, request->handle,
 848                            request->flags, request->type);
 849
 850     stl_be_p(buf, NBD_REQUEST_MAGIC);
 851     stw_be_p(buf + 4, request->flags);
 852     stw_be_p(buf + 6, request->type);
 853     stq_be_p(buf + 8, request->handle);
 854     stq_be_p(buf + 16, request->from);
 855     stl_be_p(buf + 24, request->len);
 856
 857     return nbd_write(ioc, buf, sizeof(buf), NULL);
 858 }
 859
 860 ssize_t nbd_receive_reply(QIOChannel *ioc, NBDReply *reply, Error **errp)
 861 {
 862     uint8_t buf[NBD_REPLY_SIZE];
 863     uint32_t magic;
 864     ssize_t ret;
 865
 866     ret = nbd_read_eof(ioc, buf, sizeof(buf), errp);
 867     if (ret <= 0) {
 868         return ret;
 869     }
 870
 871     if (ret != sizeof(buf)) {
 872         error_setg(errp, "read failed");
 873         return -EINVAL;
 874     }
 875
 876     /* Reply
 877        [ 0 ..  3]    magic   (NBD_REPLY_MAGIC)
 878        [ 4 ..  7]    error   (0 == no error)
 879        [ 7 .. 15]    handle
 880      */
 881
 882     magic = ldl_be_p(buf);
 883     reply->error  = ldl_be_p(buf + 4);
 884     reply->handle = ldq_be_p(buf + 8);
 885
 886     reply->error = nbd_errno_to_system_errno(reply->error);
 887
 888     if (reply->error == ESHUTDOWN) {
 889         /* This works even on mingw which lacks a native ESHUTDOWN */
 890         error_setg(errp, "server shutting down");
 891         return -EINVAL;
 892     }
 893     trace_nbd_receive_reply(magic, reply->error, reply->handle);
 894
 895     if (magic != NBD_REPLY_MAGIC) {
 896         error_setg(errp, "invalid magic (got 0x%" PRIx32 ")", magic);
 897         return -EINVAL;
 898     }
 899     return sizeof(buf);
 900 }
 901