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 + 2 * info->request_sizes + 1);
 373     stl_be_p(buf, len);
 374     memcpy(buf + 4, wantname, len);
 375     /* At most one request, everything else up to server */
 376     stw_be_p(buf + 4 + len, info->request_sizes);
 377     if (info->request_sizes) {
 378         stw_be_p(buf + 4 + len + 2, NBD_INFO_BLOCK_SIZE);
 379     }
 380     if (nbd_send_option_request(ioc, NBD_OPT_GO,
 381                                 4 + len + 2 + 2 * info->request_sizes, buf,
 382                                 errp) < 0) {
 383         return -1;
 384     }
 385
 386     while (1) {
 387         if (nbd_receive_option_reply(ioc, NBD_OPT_GO, &reply, errp) < 0) {
 388             return -1;
 389         }
 390         error = nbd_handle_reply_err(ioc, &reply, errp);
 391         if (error <= 0) {
 392             return error;
 393         }
 394         len = reply.length;
 395
 396         if (reply.type == NBD_REP_ACK) {
 397             /* Server is done sending info and moved into transmission
 398                phase, but make sure it sent flags */
 399             if (len) {
 400                 error_setg(errp, "server sent invalid NBD_REP_ACK");
 401                 nbd_send_opt_abort(ioc);
 402                 return -1;
 403             }
 404             if (!info->flags) {
 405                 error_setg(errp, "broken server omitted NBD_INFO_EXPORT");
 406                 nbd_send_opt_abort(ioc);
 407                 return -1;
 408             }
 409             trace_nbd_opt_go_success();
 410             return 1;
 411         }
 412         if (reply.type != NBD_REP_INFO) {
 413             error_setg(errp, "unexpected reply type %" PRIx32
 414                        " (%s), expected %x",
 415                        reply.type, nbd_rep_lookup(reply.type), NBD_REP_INFO);
 416             nbd_send_opt_abort(ioc);
 417             return -1;
 418         }
 419         if (len < sizeof(type)) {
 420             error_setg(errp, "NBD_REP_INFO length %" PRIu32 " is too short",
 421                        len);
 422             nbd_send_opt_abort(ioc);
 423             return -1;
 424         }
 425         if (nbd_read(ioc, &type, sizeof(type), errp) < 0) {
 426             error_prepend(errp, "failed to read info type");
 427             nbd_send_opt_abort(ioc);
 428             return -1;
 429         }
 430         len -= sizeof(type);
 431         be16_to_cpus(&type);
 432         switch (type) {
 433         case NBD_INFO_EXPORT:
 434             if (len != sizeof(info->size) + sizeof(info->flags)) {
 435                 error_setg(errp, "remaining export info len %" PRIu32
 436                            " is unexpected size", len);
 437                 nbd_send_opt_abort(ioc);
 438                 return -1;
 439             }
 440             if (nbd_read(ioc, &info->size, sizeof(info->size), errp) < 0) {
 441                 error_prepend(errp, "failed to read info size");
 442                 nbd_send_opt_abort(ioc);
 443                 return -1;
 444             }
 445             be64_to_cpus(&info->size);
 446             if (nbd_read(ioc, &info->flags, sizeof(info->flags), errp) < 0) {
 447                 error_prepend(errp, "failed to read info flags");
 448                 nbd_send_opt_abort(ioc);
 449                 return -1;
 450             }
 451             be16_to_cpus(&info->flags);
 452             trace_nbd_receive_negotiate_size_flags(info->size, info->flags);
 453             break;
 454
 455         case NBD_INFO_BLOCK_SIZE:
 456             if (len != sizeof(info->min_block) * 3) {
 457                 error_setg(errp, "remaining export info len %" PRIu32
 458                            " is unexpected size", len);
 459                 nbd_send_opt_abort(ioc);
 460                 return -1;
 461             }
 462             if (nbd_read(ioc, &info->min_block, sizeof(info->min_block),
 463                          errp) < 0) {
 464                 error_prepend(errp, "failed to read info minimum block size");
 465                 nbd_send_opt_abort(ioc);
 466                 return -1;
 467             }
 468             be32_to_cpus(&info->min_block);
 469             if (!is_power_of_2(info->min_block)) {
 470                 error_setg(errp, "server minimum block size %" PRId32
 471                            "is not a power of two", info->min_block);
 472                 nbd_send_opt_abort(ioc);
 473                 return -1;
 474             }
 475             if (nbd_read(ioc, &info->opt_block, sizeof(info->opt_block),
 476                          errp) < 0) {
 477                 error_prepend(errp, "failed to read info preferred block size");
 478                 nbd_send_opt_abort(ioc);
 479                 return -1;
 480             }
 481             be32_to_cpus(&info->opt_block);
 482             if (!is_power_of_2(info->opt_block) ||
 483                 info->opt_block < info->min_block) {
 484                 error_setg(errp, "server preferred block size %" PRId32
 485                            "is not valid", info->opt_block);
 486                 nbd_send_opt_abort(ioc);
 487                 return -1;
 488             }
 489             if (nbd_read(ioc, &info->max_block, sizeof(info->max_block),
 490                          errp) < 0) {
 491                 error_prepend(errp, "failed to read info maximum block size");
 492                 nbd_send_opt_abort(ioc);
 493                 return -1;
 494             }
 495             be32_to_cpus(&info->max_block);
 496             trace_nbd_opt_go_info_block_size(info->min_block, info->opt_block,
 497                                              info->max_block);
 498             break;
 499
 500         default:
 501             trace_nbd_opt_go_info_unknown(type, nbd_info_lookup(type));
 502             if (nbd_drop(ioc, len, errp) < 0) {
 503                 error_prepend(errp, "Failed to read info payload");
 504                 nbd_send_opt_abort(ioc);
 505                 return -1;
 506             }
 507             break;
 508         }
 509     }
 510 }
 511
 512 /* Return -1 on failure, 0 if wantname is an available export. */
 513 static int nbd_receive_query_exports(QIOChannel *ioc,
 514                                      const char *wantname,
 515                                      Error **errp)
 516 {
 517     bool foundExport = false;
 518
 519     trace_nbd_receive_query_exports_start(wantname);
 520     if (nbd_send_option_request(ioc, NBD_OPT_LIST, 0, NULL, errp) < 0) {
 521         return -1;
 522     }
 523
 524     while (1) {
 525         int ret = nbd_receive_list(ioc, wantname, &foundExport, errp);
 526
 527         if (ret < 0) {
 528             /* Server gave unexpected reply */
 529             return -1;
 530         } else if (ret == 0) {
 531             /* Done iterating. */
 532             if (!foundExport) {
 533                 error_setg(errp, "No export with name '%s' available",
 534                            wantname);
 535                 nbd_send_opt_abort(ioc);
 536                 return -1;
 537             }
 538             trace_nbd_receive_query_exports_success(wantname);
 539             return 0;
 540         }
 541     }
 542 }
 543
 544 static QIOChannel *nbd_receive_starttls(QIOChannel *ioc,
 545                                         QCryptoTLSCreds *tlscreds,
 546                                         const char *hostname, Error **errp)
 547 {
 548     nbd_opt_reply reply;
 549     QIOChannelTLS *tioc;
 550     struct NBDTLSHandshakeData data = { 0 };
 551
 552     trace_nbd_receive_starttls_request();
 553     if (nbd_send_option_request(ioc, NBD_OPT_STARTTLS, 0, NULL, errp) < 0) {
 554         return NULL;
 555     }
 556
 557     trace_nbd_receive_starttls_reply();
 558     if (nbd_receive_option_reply(ioc, NBD_OPT_STARTTLS, &reply, errp) < 0) {
 559         return NULL;
 560     }
 561
 562     if (reply.type != NBD_REP_ACK) {
 563         error_setg(errp, "Server rejected request to start TLS %" PRIx32,
 564                    reply.type);
 565         nbd_send_opt_abort(ioc);
 566         return NULL;
 567     }
 568
 569     if (reply.length != 0) {
 570         error_setg(errp, "Start TLS response was not zero %" PRIu32,
 571                    reply.length);
 572         nbd_send_opt_abort(ioc);
 573         return NULL;
 574     }
 575
 576     trace_nbd_receive_starttls_new_client();
 577     tioc = qio_channel_tls_new_client(ioc, tlscreds, hostname, errp);
 578     if (!tioc) {
 579         return NULL;
 580     }
 581     qio_channel_set_name(QIO_CHANNEL(tioc), "nbd-client-tls");
 582     data.loop = g_main_loop_new(g_main_context_default(), FALSE);
 583     trace_nbd_receive_starttls_tls_handshake();
 584     qio_channel_tls_handshake(tioc,
 585                               nbd_tls_handshake,
 586                               &data,
 587                               NULL);
 588
 589     if (!data.complete) {
 590         g_main_loop_run(data.loop);
 591     }
 592     g_main_loop_unref(data.loop);
 593     if (data.error) {
 594         error_propagate(errp, data.error);
 595         object_unref(OBJECT(tioc));
 596         return NULL;
 597     }
 598
 599     return QIO_CHANNEL(tioc);
 600 }
 601
 602
 603 int nbd_receive_negotiate(QIOChannel *ioc, const char *name,
 604                           QCryptoTLSCreds *tlscreds, const char *hostname,
 605                           QIOChannel **outioc, NBDExportInfo *info,
 606                           Error **errp)
 607 {
 608     char buf[256];
 609     uint64_t magic;
 610     int rc;
 611     bool zeroes = true;
 612
 613     trace_nbd_receive_negotiate(tlscreds, hostname ? hostname : "<null>");
 614
 615     rc = -EINVAL;
 616
 617     if (outioc) {
 618         *outioc = NULL;
 619     }
 620     if (tlscreds && !outioc) {
 621         error_setg(errp, "Output I/O channel required for TLS");
 622         goto fail;
 623     }
 624
 625     if (nbd_read(ioc, buf, 8, errp) < 0) {
 626         error_prepend(errp, "Failed to read data");
 627         goto fail;
 628     }
 629
 630     buf[8] = '\0';
 631     if (strlen(buf) == 0) {
 632         error_setg(errp, "Server connection closed unexpectedly");
 633         goto fail;
 634     }
 635
 636     magic = ldq_be_p(buf);
 637     trace_nbd_receive_negotiate_magic(magic);
 638
 639     if (memcmp(buf, "NBDMAGIC", 8) != 0) {
 640         error_setg(errp, "Invalid magic received");
 641         goto fail;
 642     }
 643
 644     if (nbd_read(ioc, &magic, sizeof(magic), errp) < 0) {
 645         error_prepend(errp, "Failed to read magic");
 646         goto fail;
 647     }
 648     magic = be64_to_cpu(magic);
 649     trace_nbd_receive_negotiate_magic(magic);
 650
 651     if (magic == NBD_OPTS_MAGIC) {
 652         uint32_t clientflags = 0;
 653         uint16_t globalflags;
 654         bool fixedNewStyle = false;
 655
 656         if (nbd_read(ioc, &globalflags, sizeof(globalflags), errp) < 0) {
 657             error_prepend(errp, "Failed to read server flags");
 658             goto fail;
 659         }
 660         globalflags = be16_to_cpu(globalflags);
 661         trace_nbd_receive_negotiate_server_flags(globalflags);
 662         if (globalflags & NBD_FLAG_FIXED_NEWSTYLE) {
 663             fixedNewStyle = true;
 664             clientflags |= NBD_FLAG_C_FIXED_NEWSTYLE;
 665         }
 666         if (globalflags & NBD_FLAG_NO_ZEROES) {
 667             zeroes = false;
 668             clientflags |= NBD_FLAG_C_NO_ZEROES;
 669         }
 670         /* client requested flags */
 671         clientflags = cpu_to_be32(clientflags);
 672         if (nbd_write(ioc, &clientflags, sizeof(clientflags), errp) < 0) {
 673             error_prepend(errp, "Failed to send clientflags field");
 674             goto fail;
 675         }
 676         if (tlscreds) {
 677             if (fixedNewStyle) {
 678                 *outioc = nbd_receive_starttls(ioc, tlscreds, hostname, errp);
 679                 if (!*outioc) {
 680                     goto fail;
 681                 }
 682                 ioc = *outioc;
 683             } else {
 684                 error_setg(errp, "Server does not support STARTTLS");
 685                 goto fail;
 686             }
 687         }
 688         if (!name) {
 689             trace_nbd_receive_negotiate_default_name();
 690             name = "";
 691         }
 692         if (fixedNewStyle) {
 693             int result;
 694
 695             /* Try NBD_OPT_GO first - if it works, we are done (it
 696              * also gives us a good message if the server requires
 697              * TLS).  If it is not available, fall back to
 698              * NBD_OPT_LIST for nicer error messages about a missing
 699              * export, then use NBD_OPT_EXPORT_NAME.  */
 700             result = nbd_opt_go(ioc, name, info, errp);
 701             if (result < 0) {
 702                 goto fail;
 703             }
 704             if (result > 0) {
 705                 return 0;
 706             }
 707             /* Check our desired export is present in the
 708              * server export list. Since NBD_OPT_EXPORT_NAME
 709              * cannot return an error message, running this
 710              * query gives us better error reporting if the
 711              * export name is not available.
 712              */
 713             if (nbd_receive_query_exports(ioc, name, errp) < 0) {
 714                 goto fail;
 715             }
 716         }
 717         /* write the export name request */
 718         if (nbd_send_option_request(ioc, NBD_OPT_EXPORT_NAME, -1, name,
 719                                     errp) < 0) {
 720             goto fail;
 721         }
 722
 723         /* Read the response */
 724         if (nbd_read(ioc, &info->size, sizeof(info->size), errp) < 0) {
 725             error_prepend(errp, "Failed to read export length");
 726             goto fail;
 727         }
 728         be64_to_cpus(&info->size);
 729
 730         if (nbd_read(ioc, &info->flags, sizeof(info->flags), errp) < 0) {
 731             error_prepend(errp, "Failed to read export flags");
 732             goto fail;
 733         }
 734         be16_to_cpus(&info->flags);
 735     } else if (magic == NBD_CLIENT_MAGIC) {
 736         uint32_t oldflags;
 737
 738         if (name) {
 739             error_setg(errp, "Server does not support export names");
 740             goto fail;
 741         }
 742         if (tlscreds) {
 743             error_setg(errp, "Server does not support STARTTLS");
 744             goto fail;
 745         }
 746
 747         if (nbd_read(ioc, &info->size, sizeof(info->size), errp) < 0) {
 748             error_prepend(errp, "Failed to read export length");
 749             goto fail;
 750         }
 751         be64_to_cpus(&info->size);
 752
 753         if (nbd_read(ioc, &oldflags, sizeof(oldflags), errp) < 0) {
 754             error_prepend(errp, "Failed to read export flags");
 755             goto fail;
 756         }
 757         be32_to_cpus(&oldflags);
 758         if (oldflags & ~0xffff) {
 759             error_setg(errp, "Unexpected export flags %0x" PRIx32, oldflags);
 760             goto fail;
 761         }
 762         info->flags = oldflags;
 763     } else {
 764         error_setg(errp, "Bad magic received");
 765         goto fail;
 766     }
 767
 768     trace_nbd_receive_negotiate_size_flags(info->size, info->flags);
 769     if (zeroes && nbd_drop(ioc, 124, errp) < 0) {
 770         error_prepend(errp, "Failed to read reserved block");
 771         goto fail;
 772     }
 773     rc = 0;
 774
 775 fail:
 776     return rc;
 777 }
 778
 779 #ifdef __linux__
 780 int nbd_init(int fd, QIOChannelSocket *sioc, NBDExportInfo *info,
 781              Error **errp)
 782 {
 783     unsigned long sector_size = MAX(BDRV_SECTOR_SIZE, info->min_block);
 784     unsigned long sectors = info->size / sector_size;
 785
 786     /* FIXME: Once the kernel module is patched to honor block sizes,
 787      * and to advertise that fact to user space, we should update the
 788      * hand-off to the kernel to use any block sizes we learned. */
 789     assert(!info->request_sizes);
 790     if (info->size / sector_size != sectors) {
 791         error_setg(errp, "Export size %" PRIu64 " too large for 32-bit kernel",
 792                    info->size);
 793         return -E2BIG;
 794     }
 795
 796     trace_nbd_init_set_socket();
 797
 798     if (ioctl(fd, NBD_SET_SOCK, (unsigned long) sioc->fd) < 0) {
 799         int serrno = errno;
 800         error_setg(errp, "Failed to set NBD socket");
 801         return -serrno;
 802     }
 803
 804     trace_nbd_init_set_block_size(sector_size);
 805
 806     if (ioctl(fd, NBD_SET_BLKSIZE, sector_size) < 0) {
 807         int serrno = errno;
 808         error_setg(errp, "Failed setting NBD block size");
 809         return -serrno;
 810     }
 811
 812     trace_nbd_init_set_size(sectors);
 813     if (info->size % sector_size) {
 814         trace_nbd_init_trailing_bytes(info->size % sector_size);
 815     }
 816
 817     if (ioctl(fd, NBD_SET_SIZE_BLOCKS, sectors) < 0) {
 818         int serrno = errno;
 819         error_setg(errp, "Failed setting size (in blocks)");
 820         return -serrno;
 821     }
 822
 823     if (ioctl(fd, NBD_SET_FLAGS, (unsigned long) info->flags) < 0) {
 824         if (errno == ENOTTY) {
 825             int read_only = (info->flags & NBD_FLAG_READ_ONLY) != 0;
 826             trace_nbd_init_set_readonly();
 827
 828             if (ioctl(fd, BLKROSET, (unsigned long) &read_only) < 0) {
 829                 int serrno = errno;
 830                 error_setg(errp, "Failed setting read-only attribute");
 831                 return -serrno;
 832             }
 833         } else {
 834             int serrno = errno;
 835             error_setg(errp, "Failed setting flags");
 836             return -serrno;
 837         }
 838     }
 839
 840     trace_nbd_init_finish();
 841
 842     return 0;
 843 }
 844
 845 int nbd_client(int fd)
 846 {
 847     int ret;
 848     int serrno;
 849
 850     trace_nbd_client_loop();
 851
 852     ret = ioctl(fd, NBD_DO_IT);
 853     if (ret < 0 && errno == EPIPE) {
 854         /* NBD_DO_IT normally returns EPIPE when someone has disconnected
 855          * the socket via NBD_DISCONNECT.  We do not want to return 1 in
 856          * that case.
 857          */
 858         ret = 0;
 859     }
 860     serrno = errno;
 861
 862     trace_nbd_client_loop_ret(ret, strerror(serrno));
 863
 864     trace_nbd_client_clear_queue();
 865     ioctl(fd, NBD_CLEAR_QUE);
 866
 867     trace_nbd_client_clear_socket();
 868     ioctl(fd, NBD_CLEAR_SOCK);
 869
 870     errno = serrno;
 871     return ret;
 872 }
 873
 874 int nbd_disconnect(int fd)
 875 {
 876     ioctl(fd, NBD_CLEAR_QUE);
 877     ioctl(fd, NBD_DISCONNECT);
 878     ioctl(fd, NBD_CLEAR_SOCK);
 879     return 0;
 880 }
 881
 882 #else
 883 int nbd_init(int fd, QIOChannelSocket *ioc, NBDExportInfo *info,
 884              Error **errp)
 885 {
 886     error_setg(errp, "nbd_init is only supported on Linux");
 887     return -ENOTSUP;
 888 }
 889
 890 int nbd_client(int fd)
 891 {
 892     return -ENOTSUP;
 893 }
 894 int nbd_disconnect(int fd)
 895 {
 896     return -ENOTSUP;
 897 }
 898 #endif
 899
 900 ssize_t nbd_send_request(QIOChannel *ioc, NBDRequest *request)
 901 {
 902     uint8_t buf[NBD_REQUEST_SIZE];
 903
 904     trace_nbd_send_request(request->from, request->len, request->handle,
 905                            request->flags, 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