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 = 0;
  94     char small[1024];
  95     char *buffer;
  96
  97     buffer = sizeof(small) < size ? small : g_malloc(MIN(65536, size));
  98     while (size > 0) {
  99         ssize_t count = read_sync(ioc, buffer, MIN(65536, size));
 100
 101         if (count <= 0) {
 102             goto cleanup;
 103         }
 104         assert(count <= size);
 105         size -= count;
 106         ret += count;
 107     }
 108
 109  cleanup:
 110     if (buffer != small) {
 111         g_free(buffer);
 112     }
 113     return ret;
 114 }
 115
 116 /* Send an option request.
 117  *
 118  * The request is for option @opt, with @data containing @len bytes of
 119  * additional payload for the request (@len may be -1 to treat @data as
 120  * a C string; and @data may be NULL if @len is 0).
 121  * Return 0 if successful, -1 with errp set if it is impossible to
 122  * continue. */
 123 static int nbd_send_option_request(QIOChannel *ioc, uint32_t opt,
 124                                    uint32_t len, const char *data,
 125                                    Error **errp)
 126 {
 127     nbd_option req;
 128     QEMU_BUILD_BUG_ON(sizeof(req) != 16);
 129
 130     if (len == -1) {
 131         req.length = len = strlen(data);
 132     }
 133     TRACE("Sending option request %" PRIu32", len %" PRIu32, opt, len);
 134
 135     stq_be_p(&req.magic, NBD_OPTS_MAGIC);
 136     stl_be_p(&req.option, opt);
 137     stl_be_p(&req.length, len);
 138
 139     if (write_sync(ioc, &req, sizeof(req)) != sizeof(req)) {
 140         error_setg(errp, "Failed to send option request header");
 141         return -1;
 142     }
 143
 144     if (len && write_sync(ioc, (char *) data, len) != len) {
 145         error_setg(errp, "Failed to send option request data");
 146         return -1;
 147     }
 148
 149     return 0;
 150 }
 151
 152 /* Send NBD_OPT_ABORT as a courtesy to let the server know that we are
 153  * not going to attempt further negotiation. */
 154 static void nbd_send_opt_abort(QIOChannel *ioc)
 155 {
 156     /* Technically, a compliant server is supposed to reply to us; but
 157      * older servers disconnected instead. At any rate, we're allowed
 158      * to disconnect without waiting for the server reply, so we don't
 159      * even care if the request makes it to the server, let alone
 160      * waiting around for whether the server replies. */
 161     nbd_send_option_request(ioc, NBD_OPT_ABORT, 0, NULL, NULL);
 162 }
 163
 164
 165 /* Receive the header of an option reply, which should match the given
 166  * opt.  Read through the length field, but NOT the length bytes of
 167  * payload. Return 0 if successful, -1 with errp set if it is
 168  * impossible to continue. */
 169 static int nbd_receive_option_reply(QIOChannel *ioc, uint32_t opt,
 170                                     nbd_opt_reply *reply, Error **errp)
 171 {
 172     QEMU_BUILD_BUG_ON(sizeof(*reply) != 20);
 173     if (read_sync(ioc, reply, sizeof(*reply)) != sizeof(*reply)) {
 174         error_setg(errp, "failed to read option reply");
 175         nbd_send_opt_abort(ioc);
 176         return -1;
 177     }
 178     be64_to_cpus(&reply->magic);
 179     be32_to_cpus(&reply->option);
 180     be32_to_cpus(&reply->type);
 181     be32_to_cpus(&reply->length);
 182
 183     TRACE("Received option reply %" PRIx32", type %" PRIx32", len %" PRIu32,
 184           reply->option, reply->type, reply->length);
 185
 186     if (reply->magic != NBD_REP_MAGIC) {
 187         error_setg(errp, "Unexpected option reply magic");
 188         nbd_send_opt_abort(ioc);
 189         return -1;
 190     }
 191     if (reply->option != opt) {
 192         error_setg(errp, "Unexpected option type %x expected %x",
 193                    reply->option, opt);
 194         nbd_send_opt_abort(ioc);
 195         return -1;
 196     }
 197     return 0;
 198 }
 199
 200 /* If reply represents success, return 1 without further action.
 201  * If reply represents an error, consume the optional payload of
 202  * the packet on ioc.  Then return 0 for unsupported (so the client
 203  * can fall back to other approaches), or -1 with errp set for other
 204  * errors.
 205  */
 206 static int nbd_handle_reply_err(QIOChannel *ioc, nbd_opt_reply *reply,
 207                                 Error **errp)
 208 {
 209     char *msg = NULL;
 210     int result = -1;
 211
 212     if (!(reply->type & (1 << 31))) {
 213         return 1;
 214     }
 215
 216     if (reply->length) {
 217         if (reply->length > NBD_MAX_BUFFER_SIZE) {
 218             error_setg(errp, "server's error message is too long");
 219             goto cleanup;
 220         }
 221         msg = g_malloc(reply->length + 1);
 222         if (read_sync(ioc, msg, reply->length) != reply->length) {
 223             error_setg(errp, "failed to read option error message");
 224             goto cleanup;
 225         }
 226         msg[reply->length] = '\0';
 227     }
 228
 229     switch (reply->type) {
 230     case NBD_REP_ERR_UNSUP:
 231         TRACE("server doesn't understand request %" PRIx32
 232               ", attempting fallback", reply->option);
 233         result = 0;
 234         goto cleanup;
 235
 236     case NBD_REP_ERR_POLICY:
 237         error_setg(errp, "Denied by server for option %" PRIx32,
 238                    reply->option);
 239         break;
 240
 241     case NBD_REP_ERR_INVALID:
 242         error_setg(errp, "Invalid data length for option %" PRIx32,
 243                    reply->option);
 244         break;
 245
 246     case NBD_REP_ERR_PLATFORM:
 247         error_setg(errp, "Server lacks support for option %" PRIx32,
 248                    reply->option);
 249         break;
 250
 251     case NBD_REP_ERR_TLS_REQD:
 252         error_setg(errp, "TLS negotiation required before option %" PRIx32,
 253                    reply->option);
 254         break;
 255
 256     case NBD_REP_ERR_SHUTDOWN:
 257         error_setg(errp, "Server shutting down before option %" PRIx32,
 258                    reply->option);
 259         break;
 260
 261     default:
 262         error_setg(errp, "Unknown error code when asking for option %" PRIx32,
 263                    reply->option);
 264         break;
 265     }
 266
 267     if (msg) {
 268         error_append_hint(errp, "%s\n", msg);
 269     }
 270
 271  cleanup:
 272     g_free(msg);
 273     if (result < 0) {
 274         nbd_send_opt_abort(ioc);
 275     }
 276     return result;
 277 }
 278
 279 /* Process another portion of the NBD_OPT_LIST reply.  Set *@match if
 280  * the current reply matches @want or if the server does not support
 281  * NBD_OPT_LIST, otherwise leave @match alone.  Return 0 if iteration
 282  * is complete, positive if more replies are expected, or negative
 283  * with @errp set if an unrecoverable error occurred. */
 284 static int nbd_receive_list(QIOChannel *ioc, const char *want, bool *match,
 285                             Error **errp)
 286 {
 287     nbd_opt_reply reply;
 288     uint32_t len;
 289     uint32_t namelen;
 290     char name[NBD_MAX_NAME_SIZE + 1];
 291     int error;
 292
 293     if (nbd_receive_option_reply(ioc, NBD_OPT_LIST, &reply, errp) < 0) {
 294         return -1;
 295     }
 296     error = nbd_handle_reply_err(ioc, &reply, errp);
 297     if (error <= 0) {
 298         /* The server did not support NBD_OPT_LIST, so set *match on
 299          * the assumption that any name will be accepted.  */
 300         *match = true;
 301         return error;
 302     }
 303     len = reply.length;
 304
 305     if (reply.type == NBD_REP_ACK) {
 306         if (len != 0) {
 307             error_setg(errp, "length too long for option end");
 308             nbd_send_opt_abort(ioc);
 309             return -1;
 310         }
 311         return 0;
 312     } else if (reply.type != NBD_REP_SERVER) {
 313         error_setg(errp, "Unexpected reply type %" PRIx32 " expected %x",
 314                    reply.type, NBD_REP_SERVER);
 315         nbd_send_opt_abort(ioc);
 316         return -1;
 317     }
 318
 319     if (len < sizeof(namelen) || len > NBD_MAX_BUFFER_SIZE) {
 320         error_setg(errp, "incorrect option length %" PRIu32, len);
 321         nbd_send_opt_abort(ioc);
 322         return -1;
 323     }
 324     if (read_sync(ioc, &namelen, sizeof(namelen)) != sizeof(namelen)) {
 325         error_setg(errp, "failed to read option name length");
 326         nbd_send_opt_abort(ioc);
 327         return -1;
 328     }
 329     namelen = be32_to_cpu(namelen);
 330     len -= sizeof(namelen);
 331     if (len < namelen) {
 332         error_setg(errp, "incorrect option name length");
 333         nbd_send_opt_abort(ioc);
 334         return -1;
 335     }
 336     if (namelen != strlen(want)) {
 337         if (drop_sync(ioc, len) != len) {
 338             error_setg(errp, "failed to skip export name with wrong length");
 339             nbd_send_opt_abort(ioc);
 340             return -1;
 341         }
 342         return 1;
 343     }
 344
 345     assert(namelen < sizeof(name));
 346     if (read_sync(ioc, name, namelen) != namelen) {
 347         error_setg(errp, "failed to read export name");
 348         nbd_send_opt_abort(ioc);
 349         return -1;
 350     }
 351     name[namelen] = '\0';
 352     len -= namelen;
 353     if (drop_sync(ioc, len) != len) {
 354         error_setg(errp, "failed to read export description");
 355         nbd_send_opt_abort(ioc);
 356         return -1;
 357     }
 358     if (!strcmp(name, want)) {
 359         *match = true;
 360     }
 361     return 1;
 362 }
 363
 364
 365 /* Return -1 on failure, 0 if wantname is an available export. */
 366 static int nbd_receive_query_exports(QIOChannel *ioc,
 367                                      const char *wantname,
 368                                      Error **errp)
 369 {
 370     bool foundExport = false;
 371
 372     TRACE("Querying export list for '%s'", wantname);
 373     if (nbd_send_option_request(ioc, NBD_OPT_LIST, 0, NULL, errp) < 0) {
 374         return -1;
 375     }
 376
 377     TRACE("Reading available export names");
 378     while (1) {
 379         int ret = nbd_receive_list(ioc, wantname, &foundExport, errp);
 380
 381         if (ret < 0) {
 382             /* Server gave unexpected reply */
 383             return -1;
 384         } else if (ret == 0) {
 385             /* Done iterating. */
 386             if (!foundExport) {
 387                 error_setg(errp, "No export with name '%s' available",
 388                            wantname);
 389                 nbd_send_opt_abort(ioc);
 390                 return -1;
 391             }
 392             TRACE("Found desired export name '%s'", wantname);
 393             return 0;
 394         }
 395     }
 396 }
 397
 398 static QIOChannel *nbd_receive_starttls(QIOChannel *ioc,
 399                                         QCryptoTLSCreds *tlscreds,
 400                                         const char *hostname, Error **errp)
 401 {
 402     nbd_opt_reply reply;
 403     QIOChannelTLS *tioc;
 404     struct NBDTLSHandshakeData data = { 0 };
 405
 406     TRACE("Requesting TLS from server");
 407     if (nbd_send_option_request(ioc, NBD_OPT_STARTTLS, 0, NULL, errp) < 0) {
 408         return NULL;
 409     }
 410
 411     TRACE("Getting TLS reply from server");
 412     if (nbd_receive_option_reply(ioc, NBD_OPT_STARTTLS, &reply, errp) < 0) {
 413         return NULL;
 414     }
 415
 416     if (reply.type != NBD_REP_ACK) {
 417         error_setg(errp, "Server rejected request to start TLS %" PRIx32,
 418                    reply.type);
 419         nbd_send_opt_abort(ioc);
 420         return NULL;
 421     }
 422
 423     if (reply.length != 0) {
 424         error_setg(errp, "Start TLS response was not zero %" PRIu32,
 425                    reply.length);
 426         nbd_send_opt_abort(ioc);
 427         return NULL;
 428     }
 429
 430     TRACE("TLS request approved, setting up TLS");
 431     tioc = qio_channel_tls_new_client(ioc, tlscreds, hostname, errp);
 432     if (!tioc) {
 433         return NULL;
 434     }
 435     qio_channel_set_name(QIO_CHANNEL(tioc), "nbd-client-tls");
 436     data.loop = g_main_loop_new(g_main_context_default(), FALSE);
 437     TRACE("Starting TLS handshake");
 438     qio_channel_tls_handshake(tioc,
 439                               nbd_tls_handshake,
 440                               &data,
 441                               NULL);
 442
 443     if (!data.complete) {
 444         g_main_loop_run(data.loop);
 445     }
 446     g_main_loop_unref(data.loop);
 447     if (data.error) {
 448         error_propagate(errp, data.error);
 449         object_unref(OBJECT(tioc));
 450         return NULL;
 451     }
 452
 453     return QIO_CHANNEL(tioc);
 454 }
 455
 456
 457 int nbd_receive_negotiate(QIOChannel *ioc, const char *name, uint16_t *flags,
 458                           QCryptoTLSCreds *tlscreds, const char *hostname,
 459                           QIOChannel **outioc,
 460                           off_t *size, Error **errp)
 461 {
 462     char buf[256];
 463     uint64_t magic, s;
 464     int rc;
 465     bool zeroes = true;
 466
 467     TRACE("Receiving negotiation tlscreds=%p hostname=%s.",
 468           tlscreds, hostname ? hostname : "<null>");
 469
 470     rc = -EINVAL;
 471
 472     if (outioc) {
 473         *outioc = NULL;
 474     }
 475     if (tlscreds && !outioc) {
 476         error_setg(errp, "Output I/O channel required for TLS");
 477         goto fail;
 478     }
 479
 480     if (read_sync(ioc, buf, 8) != 8) {
 481         error_setg(errp, "Failed to read data");
 482         goto fail;
 483     }
 484
 485     buf[8] = '\0';
 486     if (strlen(buf) == 0) {
 487         error_setg(errp, "Server connection closed unexpectedly");
 488         goto fail;
 489     }
 490
 491     TRACE("Magic is %c%c%c%c%c%c%c%c",
 492           qemu_isprint(buf[0]) ? buf[0] : '.',
 493           qemu_isprint(buf[1]) ? buf[1] : '.',
 494           qemu_isprint(buf[2]) ? buf[2] : '.',
 495           qemu_isprint(buf[3]) ? buf[3] : '.',
 496           qemu_isprint(buf[4]) ? buf[4] : '.',
 497           qemu_isprint(buf[5]) ? buf[5] : '.',
 498           qemu_isprint(buf[6]) ? buf[6] : '.',
 499           qemu_isprint(buf[7]) ? buf[7] : '.');
 500
 501     if (memcmp(buf, "NBDMAGIC", 8) != 0) {
 502         error_setg(errp, "Invalid magic received");
 503         goto fail;
 504     }
 505
 506     if (read_sync(ioc, &magic, sizeof(magic)) != sizeof(magic)) {
 507         error_setg(errp, "Failed to read magic");
 508         goto fail;
 509     }
 510     magic = be64_to_cpu(magic);
 511     TRACE("Magic is 0x%" PRIx64, magic);
 512
 513     if (magic == NBD_OPTS_MAGIC) {
 514         uint32_t clientflags = 0;
 515         uint16_t globalflags;
 516         bool fixedNewStyle = false;
 517
 518         if (read_sync(ioc, &globalflags, sizeof(globalflags)) !=
 519             sizeof(globalflags)) {
 520             error_setg(errp, "Failed to read server flags");
 521             goto fail;
 522         }
 523         globalflags = be16_to_cpu(globalflags);
 524         TRACE("Global flags are %" PRIx32, globalflags);
 525         if (globalflags & NBD_FLAG_FIXED_NEWSTYLE) {
 526             fixedNewStyle = true;
 527             TRACE("Server supports fixed new style");
 528             clientflags |= NBD_FLAG_C_FIXED_NEWSTYLE;
 529         }
 530         if (globalflags & NBD_FLAG_NO_ZEROES) {
 531             zeroes = false;
 532             TRACE("Server supports no zeroes");
 533             clientflags |= NBD_FLAG_C_NO_ZEROES;
 534         }
 535         /* client requested flags */
 536         clientflags = cpu_to_be32(clientflags);
 537         if (write_sync(ioc, &clientflags, sizeof(clientflags)) !=
 538             sizeof(clientflags)) {
 539             error_setg(errp, "Failed to send clientflags field");
 540             goto fail;
 541         }
 542         if (tlscreds) {
 543             if (fixedNewStyle) {
 544                 *outioc = nbd_receive_starttls(ioc, tlscreds, hostname, errp);
 545                 if (!*outioc) {
 546                     goto fail;
 547                 }
 548                 ioc = *outioc;
 549             } else {
 550                 error_setg(errp, "Server does not support STARTTLS");
 551                 goto fail;
 552             }
 553         }
 554         if (!name) {
 555             TRACE("Using default NBD export name \"\"");
 556             name = "";
 557         }
 558         if (fixedNewStyle) {
 559             /* Check our desired export is present in the
 560              * server export list. Since NBD_OPT_EXPORT_NAME
 561              * cannot return an error message, running this
 562              * query gives us good error reporting if the
 563              * server required TLS
 564              */
 565             if (nbd_receive_query_exports(ioc, name, errp) < 0) {
 566                 goto fail;
 567             }
 568         }
 569         /* write the export name request */
 570         if (nbd_send_option_request(ioc, NBD_OPT_EXPORT_NAME, -1, name,
 571                                     errp) < 0) {
 572             goto fail;
 573         }
 574
 575         /* Read the response */
 576         if (read_sync(ioc, &s, sizeof(s)) != sizeof(s)) {
 577             error_setg(errp, "Failed to read export length");
 578             goto fail;
 579         }
 580         *size = be64_to_cpu(s);
 581
 582         if (read_sync(ioc, flags, sizeof(*flags)) != sizeof(*flags)) {
 583             error_setg(errp, "Failed to read export flags");
 584             goto fail;
 585         }
 586         be16_to_cpus(flags);
 587     } else if (magic == NBD_CLIENT_MAGIC) {
 588         uint32_t oldflags;
 589
 590         if (name) {
 591             error_setg(errp, "Server does not support export names");
 592             goto fail;
 593         }
 594         if (tlscreds) {
 595             error_setg(errp, "Server does not support STARTTLS");
 596             goto fail;
 597         }
 598
 599         if (read_sync(ioc, &s, sizeof(s)) != sizeof(s)) {
 600             error_setg(errp, "Failed to read export length");
 601             goto fail;
 602         }
 603         *size = be64_to_cpu(s);
 604         TRACE("Size is %" PRIu64, *size);
 605
 606         if (read_sync(ioc, &oldflags, sizeof(oldflags)) != sizeof(oldflags)) {
 607             error_setg(errp, "Failed to read export flags");
 608             goto fail;
 609         }
 610         be32_to_cpus(&oldflags);
 611         if (oldflags & ~0xffff) {
 612             error_setg(errp, "Unexpected export flags %0x" PRIx32, oldflags);
 613             goto fail;
 614         }
 615         *flags = oldflags;
 616     } else {
 617         error_setg(errp, "Bad magic received");
 618         goto fail;
 619     }
 620
 621     TRACE("Size is %" PRIu64 ", export flags %" PRIx16, *size, *flags);
 622     if (zeroes && drop_sync(ioc, 124) != 124) {
 623         error_setg(errp, "Failed to read reserved block");
 624         goto fail;
 625     }
 626     rc = 0;
 627
 628 fail:
 629     return rc;
 630 }
 631
 632 #ifdef __linux__
 633 int nbd_init(int fd, QIOChannelSocket *sioc, uint16_t flags, off_t size)
 634 {
 635     unsigned long sectors = size / BDRV_SECTOR_SIZE;
 636     if (size / BDRV_SECTOR_SIZE != sectors) {
 637         LOG("Export size %lld too large for 32-bit kernel", (long long) size);
 638         return -E2BIG;
 639     }
 640
 641     TRACE("Setting NBD socket");
 642
 643     if (ioctl(fd, NBD_SET_SOCK, (unsigned long) sioc->fd) < 0) {
 644         int serrno = errno;
 645         LOG("Failed to set NBD socket");
 646         return -serrno;
 647     }
 648
 649     TRACE("Setting block size to %lu", (unsigned long)BDRV_SECTOR_SIZE);
 650
 651     if (ioctl(fd, NBD_SET_BLKSIZE, (unsigned long)BDRV_SECTOR_SIZE) < 0) {
 652         int serrno = errno;
 653         LOG("Failed setting NBD block size");
 654         return -serrno;
 655     }
 656
 657     TRACE("Setting size to %lu block(s)", sectors);
 658     if (size % BDRV_SECTOR_SIZE) {
 659         TRACE("Ignoring trailing %d bytes of export",
 660               (int) (size % BDRV_SECTOR_SIZE));
 661     }
 662
 663     if (ioctl(fd, NBD_SET_SIZE_BLOCKS, sectors) < 0) {
 664         int serrno = errno;
 665         LOG("Failed setting size (in blocks)");
 666         return -serrno;
 667     }
 668
 669     if (ioctl(fd, NBD_SET_FLAGS, (unsigned long) flags) < 0) {
 670         if (errno == ENOTTY) {
 671             int read_only = (flags & NBD_FLAG_READ_ONLY) != 0;
 672             TRACE("Setting readonly attribute");
 673
 674             if (ioctl(fd, BLKROSET, (unsigned long) &read_only) < 0) {
 675                 int serrno = errno;
 676                 LOG("Failed setting read-only attribute");
 677                 return -serrno;
 678             }
 679         } else {
 680             int serrno = errno;
 681             LOG("Failed setting flags");
 682             return -serrno;
 683         }
 684     }
 685
 686     TRACE("Negotiation ended");
 687
 688     return 0;
 689 }
 690
 691 int nbd_client(int fd)
 692 {
 693     int ret;
 694     int serrno;
 695
 696     TRACE("Doing NBD loop");
 697
 698     ret = ioctl(fd, NBD_DO_IT);
 699     if (ret < 0 && errno == EPIPE) {
 700         /* NBD_DO_IT normally returns EPIPE when someone has disconnected
 701          * the socket via NBD_DISCONNECT.  We do not want to return 1 in
 702          * that case.
 703          */
 704         ret = 0;
 705     }
 706     serrno = errno;
 707
 708     TRACE("NBD loop returned %d: %s", ret, strerror(serrno));
 709
 710     TRACE("Clearing NBD queue");
 711     ioctl(fd, NBD_CLEAR_QUE);
 712
 713     TRACE("Clearing NBD socket");
 714     ioctl(fd, NBD_CLEAR_SOCK);
 715
 716     errno = serrno;
 717     return ret;
 718 }
 719
 720 int nbd_disconnect(int fd)
 721 {
 722     ioctl(fd, NBD_CLEAR_QUE);
 723     ioctl(fd, NBD_DISCONNECT);
 724     ioctl(fd, NBD_CLEAR_SOCK);
 725     return 0;
 726 }
 727
 728 #else
 729 int nbd_init(int fd, QIOChannelSocket *ioc, uint16_t flags, off_t size)
 730 {
 731     return -ENOTSUP;
 732 }
 733
 734 int nbd_client(int fd)
 735 {
 736     return -ENOTSUP;
 737 }
 738 int nbd_disconnect(int fd)
 739 {
 740     return -ENOTSUP;
 741 }
 742 #endif
 743
 744 ssize_t nbd_send_request(QIOChannel *ioc, NBDRequest *request)
 745 {
 746     uint8_t buf[NBD_REQUEST_SIZE];
 747     ssize_t ret;
 748
 749     TRACE("Sending request to server: "
 750           "{ .from = %" PRIu64", .len = %" PRIu32 ", .handle = %" PRIu64
 751           ", .flags = %" PRIx16 ", .type = %" PRIu16 " }",
 752           request->from, request->len, request->handle,
 753           request->flags, request->type);
 754
 755     stl_be_p(buf, NBD_REQUEST_MAGIC);
 756     stw_be_p(buf + 4, request->flags);
 757     stw_be_p(buf + 6, request->type);
 758     stq_be_p(buf + 8, request->handle);
 759     stq_be_p(buf + 16, request->from);
 760     stl_be_p(buf + 24, request->len);
 761
 762     ret = write_sync(ioc, buf, sizeof(buf));
 763     if (ret < 0) {
 764         return ret;
 765     }
 766
 767     if (ret != sizeof(buf)) {
 768         LOG("writing to socket failed");
 769         return -EINVAL;
 770     }
 771     return 0;
 772 }
 773
 774 ssize_t nbd_receive_reply(QIOChannel *ioc, NBDReply *reply)
 775 {
 776     uint8_t buf[NBD_REPLY_SIZE];
 777     uint32_t magic;
 778     ssize_t ret;
 779
 780     ret = read_sync(ioc, buf, sizeof(buf));
 781     if (ret < 0) {
 782         return ret;
 783     }
 784
 785     if (ret != sizeof(buf)) {
 786         LOG("read failed");
 787         return -EINVAL;
 788     }
 789
 790     /* Reply
 791        [ 0 ..  3]    magic   (NBD_REPLY_MAGIC)
 792        [ 4 ..  7]    error   (0 == no error)
 793        [ 7 .. 15]    handle
 794      */
 795
 796     magic = ldl_be_p(buf);
 797     reply->error  = ldl_be_p(buf + 4);
 798     reply->handle = ldq_be_p(buf + 8);
 799
 800     reply->error = nbd_errno_to_system_errno(reply->error);
 801
 802     if (reply->error == ESHUTDOWN) {
 803         /* This works even on mingw which lacks a native ESHUTDOWN */
 804         LOG("server shutting down");
 805         return -EINVAL;
 806     }
 807     TRACE("Got reply: { magic = 0x%" PRIx32 ", .error = % " PRId32
 808           ", handle = %" PRIu64" }",
 809           magic, reply->error, reply->handle);
 810
 811     if (magic != NBD_REPLY_MAGIC) {
 812         LOG("invalid magic (got 0x%" PRIx32 ")", magic);
 813         return -EINVAL;
 814     }
 815     return 0;
 816 }
 817