nbd/client.c

   1 /*
   2  *  Copyright (C) 2016 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 "nbd-internal.h"
  23
  24 static int nbd_errno_to_system_errno(int err)
  25 {
  26     int ret;
  27     switch (err) {
  28     case NBD_SUCCESS:
  29         ret = 0;
  30         break;
  31     case NBD_EPERM:
  32         ret = EPERM;
  33         break;
  34     case NBD_EIO:
  35         ret = EIO;
  36         break;
  37     case NBD_ENOMEM:
  38         ret = ENOMEM;
  39         break;
  40     case NBD_ENOSPC:
  41         ret = ENOSPC;
  42         break;
  43     case NBD_ESHUTDOWN:
  44         ret = ESHUTDOWN;
  45         break;
  46     default:
  47         TRACE("Squashing unexpected error %d to EINVAL", err);
  48         /* fallthrough */
  49     case NBD_EINVAL:
  50         ret = EINVAL;
  51         break;
  52     }
  53     return ret;
  54 }
  55
  56 /* Definitions for opaque data types */
  57
  58 static QTAILQ_HEAD(, NBDExport) exports = QTAILQ_HEAD_INITIALIZER(exports);
  59
  60 /* That's all folks */
  61
  62 /* Basic flow for negotiation
  63
  64    Server         Client
  65    Negotiate
  66
  67    or
  68
  69    Server         Client
  70    Negotiate #1
  71                   Option
  72    Negotiate #2
  73
  74    ----
  75
  76    followed by
  77
  78    Server         Client
  79                   Request
  80    Response
  81                   Request
  82    Response
  83                   ...
  84    ...
  85                   Request (type == 2)
  86
  87 */
  88
  89 /* Discard length bytes from channel.  Return -errno on failure, or
  90  * the amount of bytes consumed. */
  91 static ssize_t drop_sync(QIOChannel *ioc, size_t size)
  92 {
  93     ssize_t ret, dropped = size;
  94     char small[1024];
  95     char *buffer;
  96
  97     buffer = sizeof(small) < size ? small : g_malloc(MIN(65536, size));
  98     while (size > 0) {
  99         ret = read_sync(ioc, buffer, MIN(65536, size));
 100         if (ret < 0) {
 101             goto cleanup;
 102         }
 103         assert(ret <= size);
 104         size -= ret;
 105     }
 106     ret = dropped;
 107
 108  cleanup:
 109     if (buffer != small) {
 110         g_free(buffer);
 111     }
 112     return ret;
 113 }
 114
 115 /* Send an option request.
 116  *
 117  * The request is for option @opt, with @data containing @len bytes of
 118  * additional payload for the request (@len may be -1 to treat @data as
 119  * a C string; and @data may be NULL if @len is 0).
 120  * Return 0 if successful, -1 with errp set if it is impossible to
 121  * continue. */
 122 static int nbd_send_option_request(QIOChannel *ioc, uint32_t opt,
 123                                    uint32_t len, const char *data,
 124                                    Error **errp)
 125 {
 126     nbd_option req;
 127     QEMU_BUILD_BUG_ON(sizeof(req) != 16);
 128
 129     if (len == -1) {
 130         req.length = len = strlen(data);
 131     }
 132     TRACE("Sending option request %" PRIu32", len %" PRIu32, opt, len);
 133
 134     stq_be_p(&req.magic, NBD_OPTS_MAGIC);
 135     stl_be_p(&req.option, opt);
 136     stl_be_p(&req.length, len);
 137
 138     if (write_sync(ioc, &req, sizeof(req)) != sizeof(req)) {
 139         error_setg(errp, "Failed to send option request header");
 140         return -1;
 141     }
 142
 143     if (len && write_sync(ioc, (char *) data, len) != len) {
 144         error_setg(errp, "Failed to send option request data");
 145         return -1;
 146     }
 147
 148     return 0;
 149 }
 150
 151 /* Send NBD_OPT_ABORT as a courtesy to let the server know that we are
 152  * not going to attempt further negotiation. */
 153 static void nbd_send_opt_abort(QIOChannel *ioc)
 154 {
 155     /* Technically, a compliant server is supposed to reply to us; but
 156      * older servers disconnected instead. At any rate, we're allowed
 157      * to disconnect without waiting for the server reply, so we don't
 158      * even care if the request makes it to the server, let alone
 159      * waiting around for whether the server replies. */
 160     nbd_send_option_request(ioc, NBD_OPT_ABORT, 0, NULL, NULL);
 161 }
 162
 163
 164 /* Receive the header of an option reply, which should match the given
 165  * opt.  Read through the length field, but NOT the length bytes of
 166  * payload. Return 0 if successful, -1 with errp set if it is
 167  * impossible to continue. */
 168 static int nbd_receive_option_reply(QIOChannel *ioc, uint32_t opt,
 169                                     nbd_opt_reply *reply, Error **errp)
 170 {
 171     QEMU_BUILD_BUG_ON(sizeof(*reply) != 20);
 172     if (read_sync(ioc, reply, sizeof(*reply)) != sizeof(*reply)) {
 173         error_setg(errp, "failed to read option reply");
 174         nbd_send_opt_abort(ioc);
 175         return -1;
 176     }
 177     be64_to_cpus(&reply->magic);
 178     be32_to_cpus(&reply->option);
 179     be32_to_cpus(&reply->type);
 180     be32_to_cpus(&reply->length);
 181
 182     TRACE("Received option reply %" PRIx32", type %" PRIx32", len %" PRIu32,
 183           reply->option, reply->type, reply->length);
 184
 185     if (reply->magic != NBD_REP_MAGIC) {
 186         error_setg(errp, "Unexpected option reply magic");
 187         nbd_send_opt_abort(ioc);
 188         return -1;
 189     }
 190     if (reply->option != opt) {
 191         error_setg(errp, "Unexpected option type %x expected %x",
 192                    reply->option, opt);
 193         nbd_send_opt_abort(ioc);
 194         return -1;
 195     }
 196     return 0;
 197 }
 198
 199 /* If reply represents success, return 1 without further action.
 200  * If reply represents an error, consume the optional payload of
 201  * the packet on ioc.  Then return 0 for unsupported (so the client
 202  * can fall back to other approaches), or -1 with errp set for other
 203  * errors.
 204  */
 205 static int nbd_handle_reply_err(QIOChannel *ioc, nbd_opt_reply *reply,
 206                                 Error **errp)
 207 {
 208     char *msg = NULL;
 209     int result = -1;
 210
 211     if (!(reply->type & (1 << 31))) {
 212         return 1;
 213     }
 214
 215     if (reply->length) {
 216         if (reply->length > NBD_MAX_BUFFER_SIZE) {
 217             error_setg(errp, "server's error message is too long");
 218             goto cleanup;
 219         }
 220         msg = g_malloc(reply->length + 1);
 221         if (read_sync(ioc, msg, reply->length) != reply->length) {
 222             error_setg(errp, "failed to read option error message");
 223             goto cleanup;
 224         }
 225         msg[reply->length] = '\0';
 226     }
 227
 228     switch (reply->type) {
 229     case NBD_REP_ERR_UNSUP:
 230         TRACE("server doesn't understand request %" PRIx32
 231               ", attempting fallback", reply->option);
 232         result = 0;
 233         goto cleanup;
 234
 235     case NBD_REP_ERR_POLICY:
 236         error_setg(errp, "Denied by server for option %" PRIx32,
 237                    reply->option);
 238         break;
 239
 240     case NBD_REP_ERR_INVALID:
 241         error_setg(errp, "Invalid data length for option %" PRIx32,
 242                    reply->option);
 243         break;
 244
 245     case NBD_REP_ERR_PLATFORM:
 246         error_setg(errp, "Server lacks support for option %" PRIx32,
 247                    reply->option);
 248         break;
 249
 250     case NBD_REP_ERR_TLS_REQD:
 251         error_setg(errp, "TLS negotiation required before option %" PRIx32,
 252                    reply->option);
 253         break;
 254
 255     case NBD_REP_ERR_SHUTDOWN:
 256         error_setg(errp, "Server shutting down before option %" PRIx32,
 257                    reply->option);
 258         break;
 259
 260     default:
 261         error_setg(errp, "Unknown error code when asking for option %" PRIx32,
 262                    reply->option);
 263         break;
 264     }
 265
 266     if (msg) {
 267         error_append_hint(errp, "%s\n", msg);
 268     }
 269
 270  cleanup:
 271     g_free(msg);
 272     if (result < 0) {
 273         nbd_send_opt_abort(ioc);
 274     }
 275     return result;
 276 }
 277
 278 /* Process another portion of the NBD_OPT_LIST reply.  Set *@match if
 279  * the current reply matches @want or if the server does not support
 280  * NBD_OPT_LIST, otherwise leave @match alone.  Return 0 if iteration
 281  * is complete, positive if more replies are expected, or negative
 282  * with @errp set if an unrecoverable error occurred. */
 283 static int nbd_receive_list(QIOChannel *ioc, const char *want, bool *match,
 284                             Error **errp)
 285 {
 286     nbd_opt_reply reply;
 287     uint32_t len;
 288     uint32_t namelen;
 289     char name[NBD_MAX_NAME_SIZE + 1];
 290     int error;
 291
 292     if (nbd_receive_option_reply(ioc, NBD_OPT_LIST, &reply, errp) < 0) {
 293         return -1;
 294     }
 295     error = nbd_handle_reply_err(ioc, &reply, errp);
 296     if (error <= 0) {
 297         /* The server did not support NBD_OPT_LIST, so set *match on
 298          * the assumption that any name will be accepted.  */
 299         *match = true;
 300         return error;
 301     }
 302     len = reply.length;
 303
 304     if (reply.type == NBD_REP_ACK) {
 305         if (len != 0) {
 306             error_setg(errp, "length too long for option end");
 307             nbd_send_opt_abort(ioc);
 308             return -1;
 309         }
 310         return 0;
 311     } else if (reply.type != NBD_REP_SERVER) {
 312         error_setg(errp, "Unexpected reply type %" PRIx32 " expected %x",
 313                    reply.type, NBD_REP_SERVER);
 314         nbd_send_opt_abort(ioc);
 315         return -1;
 316     }
 317
 318     if (len < sizeof(namelen) || len > NBD_MAX_BUFFER_SIZE) {
 319         error_setg(errp, "incorrect option length %" PRIu32, len);
 320         nbd_send_opt_abort(ioc);
 321         return -1;
 322     }
 323     if (read_sync(ioc, &namelen, sizeof(namelen)) != sizeof(namelen)) {
 324         error_setg(errp, "failed to read option name length");
 325         nbd_send_opt_abort(ioc);
 326         return -1;
 327     }
 328     namelen = be32_to_cpu(namelen);
 329     len -= sizeof(namelen);
 330     if (len < namelen) {
 331         error_setg(errp, "incorrect option name length");
 332         nbd_send_opt_abort(ioc);
 333         return -1;
 334     }
 335     if (namelen != strlen(want)) {
 336         if (drop_sync(ioc, len) != len) {
 337             error_setg(errp, "failed to skip export name with wrong length");
 338             nbd_send_opt_abort(ioc);
 339             return -1;
 340         }
 341         return 1;
 342     }
 343
 344     assert(namelen < sizeof(name));
 345     if (read_sync(ioc, name, namelen) != namelen) {
 346         error_setg(errp, "failed to read export name");
 347         nbd_send_opt_abort(ioc);
 348         return -1;
 349     }
 350     name[namelen] = '\0';
 351     len -= namelen;
 352     if (drop_sync(ioc, len) != len) {
 353         error_setg(errp, "failed to read export description");
 354         nbd_send_opt_abort(ioc);
 355         return -1;
 356     }
 357     if (!strcmp(name, want)) {
 358         *match = true;
 359     }
 360     return 1;
 361 }
 362
 363
 364 /* Return -1 on failure, 0 if wantname is an available export. */
 365 static int nbd_receive_query_exports(QIOChannel *ioc,
 366                                      const char *wantname,
 367                                      Error **errp)
 368 {
 369     bool foundExport = false;
 370
 371     TRACE("Querying export list for '%s'", wantname);
 372     if (nbd_send_option_request(ioc, NBD_OPT_LIST, 0, NULL, errp) < 0) {
 373         return -1;
 374     }
 375
 376     TRACE("Reading available export names");
 377     while (1) {
 378         int ret = nbd_receive_list(ioc, wantname, &foundExport, errp);
 379
 380         if (ret < 0) {
 381             /* Server gave unexpected reply */
 382             return -1;
 383         } else if (ret == 0) {
 384             /* Done iterating. */
 385             if (!foundExport) {
 386                 error_setg(errp, "No export with name '%s' available",
 387                            wantname);
 388                 nbd_send_opt_abort(ioc);
 389                 return -1;
 390             }
 391             TRACE("Found desired export name '%s'", wantname);
 392             return 0;
 393         }
 394     }
 395 }
 396
 397 static QIOChannel *nbd_receive_starttls(QIOChannel *ioc,
 398                                         QCryptoTLSCreds *tlscreds,
 399                                         const char *hostname, Error **errp)
 400 {
 401     nbd_opt_reply reply;
 402     QIOChannelTLS *tioc;
 403     struct NBDTLSHandshakeData data = { 0 };
 404
 405     TRACE("Requesting TLS from server");
 406     if (nbd_send_option_request(ioc, NBD_OPT_STARTTLS, 0, NULL, errp) < 0) {
 407         return NULL;
 408     }
 409
 410     TRACE("Getting TLS reply from server");
 411     if (nbd_receive_option_reply(ioc, NBD_OPT_STARTTLS, &reply, errp) < 0) {
 412         return NULL;
 413     }
 414
 415     if (reply.type != NBD_REP_ACK) {
 416         error_setg(errp, "Server rejected request to start TLS %" PRIx32,
 417                    reply.type);
 418         nbd_send_opt_abort(ioc);
 419         return NULL;
 420     }
 421
 422     if (reply.length != 0) {
 423         error_setg(errp, "Start TLS response was not zero %" PRIu32,
 424                    reply.length);
 425         nbd_send_opt_abort(ioc);
 426         return NULL;
 427     }
 428
 429     TRACE("TLS request approved, setting up TLS");
 430     tioc = qio_channel_tls_new_client(ioc, tlscreds, hostname, errp);
 431     if (!tioc) {
 432         return NULL;
 433     }
 434     qio_channel_set_name(QIO_CHANNEL(tioc), "nbd-client-tls");
 435     data.loop = g_main_loop_new(g_main_context_default(), FALSE);
 436     TRACE("Starting TLS handshake");
 437     qio_channel_tls_handshake(tioc,
 438                               nbd_tls_handshake,
 439                               &data,
 440                               NULL);
 441
 442     if (!data.complete) {
 443         g_main_loop_run(data.loop);
 444     }
 445     g_main_loop_unref(data.loop);
 446     if (data.error) {
 447         error_propagate(errp, data.error);
 448         object_unref(OBJECT(tioc));
 449         return NULL;
 450     }
 451
 452     return QIO_CHANNEL(tioc);
 453 }
 454
 455
 456 int nbd_receive_negotiate(QIOChannel *ioc, const char *name, uint16_t *flags,
 457                           QCryptoTLSCreds *tlscreds, const char *hostname,
 458                           QIOChannel **outioc,
 459                           off_t *size, Error **errp)
 460 {
 461     char buf[256];
 462     uint64_t magic, s;
 463     int rc;
 464     bool zeroes = true;
 465
 466     TRACE("Receiving negotiation tlscreds=%p hostname=%s.",
 467           tlscreds, hostname ? hostname : "<null>");
 468
 469     rc = -EINVAL;
 470
 471     if (outioc) {
 472         *outioc = NULL;
 473     }
 474     if (tlscreds && !outioc) {
 475         error_setg(errp, "Output I/O channel required for TLS");
 476         goto fail;
 477     }
 478
 479     if (read_sync(ioc, buf, 8) != 8) {
 480         error_setg(errp, "Failed to read data");
 481         goto fail;
 482     }
 483
 484     buf[8] = '\0';
 485     if (strlen(buf) == 0) {
 486         error_setg(errp, "Server connection closed unexpectedly");
 487         goto fail;
 488     }
 489
 490     TRACE("Magic is %c%c%c%c%c%c%c%c",
 491           qemu_isprint(buf[0]) ? buf[0] : '.',
 492           qemu_isprint(buf[1]) ? buf[1] : '.',
 493           qemu_isprint(buf[2]) ? buf[2] : '.',
 494           qemu_isprint(buf[3]) ? buf[3] : '.',
 495           qemu_isprint(buf[4]) ? buf[4] : '.',
 496           qemu_isprint(buf[5]) ? buf[5] : '.',
 497           qemu_isprint(buf[6]) ? buf[6] : '.',
 498           qemu_isprint(buf[7]) ? buf[7] : '.');
 499
 500     if (memcmp(buf, "NBDMAGIC", 8) != 0) {
 501         error_setg(errp, "Invalid magic received");
 502         goto fail;
 503     }
 504
 505     if (read_sync(ioc, &magic, sizeof(magic)) != sizeof(magic)) {
 506         error_setg(errp, "Failed to read magic");
 507         goto fail;
 508     }
 509     magic = be64_to_cpu(magic);
 510     TRACE("Magic is 0x%" PRIx64, magic);
 511
 512     if (magic == NBD_OPTS_MAGIC) {
 513         uint32_t clientflags = 0;
 514         uint16_t globalflags;
 515         bool fixedNewStyle = false;
 516
 517         if (read_sync(ioc, &globalflags, sizeof(globalflags)) !=
 518             sizeof(globalflags)) {
 519             error_setg(errp, "Failed to read server flags");
 520             goto fail;
 521         }
 522         globalflags = be16_to_cpu(globalflags);
 523         TRACE("Global flags are %" PRIx32, globalflags);
 524         if (globalflags & NBD_FLAG_FIXED_NEWSTYLE) {
 525             fixedNewStyle = true;
 526             TRACE("Server supports fixed new style");
 527             clientflags |= NBD_FLAG_C_FIXED_NEWSTYLE;
 528         }
 529         if (globalflags & NBD_FLAG_NO_ZEROES) {
 530             zeroes = false;
 531             TRACE("Server supports no zeroes");
 532             clientflags |= NBD_FLAG_C_NO_ZEROES;
 533         }
 534         /* client requested flags */
 535         clientflags = cpu_to_be32(clientflags);
 536         if (write_sync(ioc, &clientflags, sizeof(clientflags)) !=
 537             sizeof(clientflags)) {
 538             error_setg(errp, "Failed to send clientflags field");
 539             goto fail;
 540         }
 541         if (tlscreds) {
 542             if (fixedNewStyle) {
 543                 *outioc = nbd_receive_starttls(ioc, tlscreds, hostname, errp);
 544                 if (!*outioc) {
 545                     goto fail;
 546                 }
 547                 ioc = *outioc;
 548             } else {
 549                 error_setg(errp, "Server does not support STARTTLS");
 550                 goto fail;
 551             }
 552         }
 553         if (!name) {
 554             TRACE("Using default NBD export name \"\"");
 555             name = "";
 556         }
 557         if (fixedNewStyle) {
 558             /* Check our desired export is present in the
 559              * server export list. Since NBD_OPT_EXPORT_NAME
 560              * cannot return an error message, running this
 561              * query gives us good error reporting if the
 562              * server required TLS
 563              */
 564             if (nbd_receive_query_exports(ioc, name, errp) < 0) {
 565                 goto fail;
 566             }
 567         }
 568         /* write the export name request */
 569         if (nbd_send_option_request(ioc, NBD_OPT_EXPORT_NAME, -1, name,
 570                                     errp) < 0) {
 571             goto fail;
 572         }
 573
 574         /* Read the response */
 575         if (read_sync(ioc, &s, sizeof(s)) != sizeof(s)) {
 576             error_setg(errp, "Failed to read export length");
 577             goto fail;
 578         }
 579         *size = be64_to_cpu(s);
 580
 581         if (read_sync(ioc, flags, sizeof(*flags)) != sizeof(*flags)) {
 582             error_setg(errp, "Failed to read export flags");
 583             goto fail;
 584         }
 585         be16_to_cpus(flags);
 586     } else if (magic == NBD_CLIENT_MAGIC) {
 587         uint32_t oldflags;
 588
 589         if (name) {
 590             error_setg(errp, "Server does not support export names");
 591             goto fail;
 592         }
 593         if (tlscreds) {
 594             error_setg(errp, "Server does not support STARTTLS");
 595             goto fail;
 596         }
 597
 598         if (read_sync(ioc, &s, sizeof(s)) != sizeof(s)) {
 599             error_setg(errp, "Failed to read export length");
 600             goto fail;
 601         }
 602         *size = be64_to_cpu(s);
 603         TRACE("Size is %" PRIu64, *size);
 604
 605         if (read_sync(ioc, &oldflags, sizeof(oldflags)) != sizeof(oldflags)) {
 606             error_setg(errp, "Failed to read export flags");
 607             goto fail;
 608         }
 609         be32_to_cpus(&oldflags);
 610         if (oldflags & ~0xffff) {
 611             error_setg(errp, "Unexpected export flags %0x" PRIx32, oldflags);
 612             goto fail;
 613         }
 614         *flags = oldflags;
 615     } else {
 616         error_setg(errp, "Bad magic received");
 617         goto fail;
 618     }
 619
 620     TRACE("Size is %" PRIu64 ", export flags %" PRIx16, *size, *flags);
 621     if (zeroes && drop_sync(ioc, 124) != 124) {
 622         error_setg(errp, "Failed to read reserved block");
 623         goto fail;
 624     }
 625     rc = 0;
 626
 627 fail:
 628     return rc;
 629 }
 630
 631 #ifdef __linux__
 632 int nbd_init(int fd, QIOChannelSocket *sioc, uint16_t flags, off_t size)
 633 {
 634     unsigned long sectors = size / BDRV_SECTOR_SIZE;
 635     if (size / BDRV_SECTOR_SIZE != sectors) {
 636         LOG("Export size %lld too large for 32-bit kernel", (long long) size);
 637         return -E2BIG;
 638     }
 639
 640     TRACE("Setting NBD socket");
 641
 642     if (ioctl(fd, NBD_SET_SOCK, (unsigned long) sioc->fd) < 0) {
 643         int serrno = errno;
 644         LOG("Failed to set NBD socket");
 645         return -serrno;
 646     }
 647
 648     TRACE("Setting block size to %lu", (unsigned long)BDRV_SECTOR_SIZE);
 649
 650     if (ioctl(fd, NBD_SET_BLKSIZE, (unsigned long)BDRV_SECTOR_SIZE) < 0) {
 651         int serrno = errno;
 652         LOG("Failed setting NBD block size");
 653         return -serrno;
 654     }
 655
 656     TRACE("Setting size to %lu block(s)", sectors);
 657     if (size % BDRV_SECTOR_SIZE) {
 658         TRACE("Ignoring trailing %d bytes of export",
 659               (int) (size % BDRV_SECTOR_SIZE));
 660     }
 661
 662     if (ioctl(fd, NBD_SET_SIZE_BLOCKS, sectors) < 0) {
 663         int serrno = errno;
 664         LOG("Failed setting size (in blocks)");
 665         return -serrno;
 666     }
 667
 668     if (ioctl(fd, NBD_SET_FLAGS, (unsigned long) flags) < 0) {
 669         if (errno == ENOTTY) {
 670             int read_only = (flags & NBD_FLAG_READ_ONLY) != 0;
 671             TRACE("Setting readonly attribute");
 672
 673             if (ioctl(fd, BLKROSET, (unsigned long) &read_only) < 0) {
 674                 int serrno = errno;
 675                 LOG("Failed setting read-only attribute");
 676                 return -serrno;
 677             }
 678         } else {
 679             int serrno = errno;
 680             LOG("Failed setting flags");
 681             return -serrno;
 682         }
 683     }
 684
 685     TRACE("Negotiation ended");
 686
 687     return 0;
 688 }
 689
 690 int nbd_client(int fd)
 691 {
 692     int ret;
 693     int serrno;
 694
 695     TRACE("Doing NBD loop");
 696
 697     ret = ioctl(fd, NBD_DO_IT);
 698     if (ret < 0 && errno == EPIPE) {
 699         /* NBD_DO_IT normally returns EPIPE when someone has disconnected
 700          * the socket via NBD_DISCONNECT.  We do not want to return 1 in
 701          * that case.
 702          */
 703         ret = 0;
 704     }
 705     serrno = errno;
 706
 707     TRACE("NBD loop returned %d: %s", ret, strerror(serrno));
 708
 709     TRACE("Clearing NBD queue");
 710     ioctl(fd, NBD_CLEAR_QUE);
 711
 712     TRACE("Clearing NBD socket");
 713     ioctl(fd, NBD_CLEAR_SOCK);
 714
 715     errno = serrno;
 716     return ret;
 717 }
 718
 719 int nbd_disconnect(int fd)
 720 {
 721     ioctl(fd, NBD_CLEAR_QUE);
 722     ioctl(fd, NBD_DISCONNECT);
 723     ioctl(fd, NBD_CLEAR_SOCK);
 724     return 0;
 725 }
 726
 727 #else
 728 int nbd_init(int fd, QIOChannelSocket *ioc, uint16_t flags, off_t size)
 729 {
 730     return -ENOTSUP;
 731 }
 732
 733 int nbd_client(int fd)
 734 {
 735     return -ENOTSUP;
 736 }
 737 int nbd_disconnect(int fd)
 738 {
 739     return -ENOTSUP;
 740 }
 741 #endif
 742
 743 ssize_t nbd_send_request(QIOChannel *ioc, NBDRequest *request)
 744 {
 745     uint8_t buf[NBD_REQUEST_SIZE];
 746     ssize_t ret;
 747
 748     TRACE("Sending request to server: "
 749           "{ .from = %" PRIu64", .len = %" PRIu32 ", .handle = %" PRIu64
 750           ", .flags = %" PRIx16 ", .type = %" PRIu16 " }",
 751           request->from, request->len, request->handle,
 752           request->flags, request->type);
 753
 754     stl_be_p(buf, NBD_REQUEST_MAGIC);
 755     stw_be_p(buf + 4, request->flags);
 756     stw_be_p(buf + 6, request->type);
 757     stq_be_p(buf + 8, request->handle);
 758     stq_be_p(buf + 16, request->from);
 759     stl_be_p(buf + 24, request->len);
 760
 761     ret = write_sync(ioc, buf, sizeof(buf));
 762     if (ret < 0) {
 763         return ret;
 764     }
 765
 766     if (ret != sizeof(buf)) {
 767         LOG("writing to socket failed");
 768         return -EINVAL;
 769     }
 770     return 0;
 771 }
 772
 773 ssize_t nbd_receive_reply(QIOChannel *ioc, NBDReply *reply)
 774 {
 775     uint8_t buf[NBD_REPLY_SIZE];
 776     uint32_t magic;
 777     ssize_t ret;
 778
 779     ret = read_sync(ioc, buf, sizeof(buf));
 780     if (ret < 0) {
 781         return ret;
 782     }
 783
 784     if (ret != sizeof(buf)) {
 785         LOG("read failed");
 786         return -EINVAL;
 787     }
 788
 789     /* Reply
 790        [ 0 ..  3]    magic   (NBD_REPLY_MAGIC)
 791        [ 4 ..  7]    error   (0 == no error)
 792        [ 7 .. 15]    handle
 793      */
 794
 795     magic = ldl_be_p(buf);
 796     reply->error  = ldl_be_p(buf + 4);
 797     reply->handle = ldq_be_p(buf + 8);
 798
 799     reply->error = nbd_errno_to_system_errno(reply->error);
 800
 801     if (reply->error == ESHUTDOWN) {
 802         /* This works even on mingw which lacks a native ESHUTDOWN */
 803         LOG("server shutting down");
 804         return -EINVAL;
 805     }
 806     TRACE("Got reply: { magic = 0x%" PRIx32 ", .error = % " PRId32
 807           ", handle = %" PRIu64" }",
 808           magic, reply->error, reply->handle);
 809
 810     if (magic != NBD_REPLY_MAGIC) {
 811         LOG("invalid magic (got 0x%" PRIx32 ")", magic);
 812         return -EINVAL;
 813     }
 814     return 0;
 815 }
 816