slirp/socket.c

   1 /*
   2  * Copyright (c) 1995 Danny Gasparovski.
   3  *
   4  * Please read the file COPYRIGHT for the
   5  * terms and conditions of the copyright.
   6  */
   7
   8 #include "qemu/osdep.h"
   9 #include "qemu-common.h"
  10 #include "slirp.h"
  11 #include "ip_icmp.h"
  12 #ifdef __sun__
  13 #include <sys/filio.h>
  14 #endif
  15
  16 static void sofcantrcvmore(struct socket *so);
  17 static void sofcantsendmore(struct socket *so);
  18
  19 struct socket *solookup(struct socket **last, struct socket *head,
  20         struct sockaddr_storage *lhost, struct sockaddr_storage *fhost)
  21 {
  22     struct socket *so = *last;
  23
  24     /* Optimisation */
  25     if (so != head && sockaddr_equal(&(so->lhost.ss), lhost)
  26             && (!fhost || sockaddr_equal(&so->fhost.ss, fhost))) {
  27         return so;
  28     }
  29
  30     for (so = head->so_next; so != head; so = so->so_next) {
  31         if (sockaddr_equal(&(so->lhost.ss), lhost)
  32                 && (!fhost || sockaddr_equal(&so->fhost.ss, fhost))) {
  33             *last = so;
  34             return so;
  35         }
  36     }
  37
  38     return (struct socket *)NULL;
  39 }
  40
  41 /*
  42  * Create a new socket, initialise the fields
  43  * It is the responsibility of the caller to
  44  * insque() it into the correct linked-list
  45  */
  46 struct socket *
  47 socreate(Slirp *slirp)
  48 {
  49     struct socket *so = g_new(struct socket, 1);
  50
  51     memset(so, 0, sizeof(struct socket));
  52     so->so_state = SS_NOFDREF;
  53     so->s = -1;
  54     so->slirp = slirp;
  55     so->pollfds_idx = -1;
  56
  57     return so;
  58 }
  59
  60 /*
  61  * Remove references to so from the given message queue.
  62  */
  63 static void
  64 soqfree(struct socket *so, struct quehead *qh)
  65 {
  66     struct mbuf *ifq;
  67
  68     for (ifq = (struct mbuf *) qh->qh_link;
  69              (struct quehead *) ifq != qh;
  70              ifq = ifq->ifq_next) {
  71         if (ifq->ifq_so == so) {
  72             struct mbuf *ifm;
  73             ifq->ifq_so = NULL;
  74             for (ifm = ifq->ifs_next; ifm != ifq; ifm = ifm->ifs_next) {
  75                 ifm->ifq_so = NULL;
  76             }
  77         }
  78     }
  79 }
  80
  81 /*
  82  * remque and free a socket, clobber cache
  83  */
  84 void
  85 sofree(struct socket *so)
  86 {
  87   Slirp *slirp = so->slirp;
  88
  89   soqfree(so, &slirp->if_fastq);
  90   soqfree(so, &slirp->if_batchq);
  91
  92   if (so == slirp->tcp_last_so) {
  93       slirp->tcp_last_so = &slirp->tcb;
  94   } else if (so == slirp->udp_last_so) {
  95       slirp->udp_last_so = &slirp->udb;
  96   } else if (so == slirp->icmp_last_so) {
  97       slirp->icmp_last_so = &slirp->icmp;
  98   }
  99   m_free(so->so_m);
 100
 101   if(so->so_next && so->so_prev)
 102     remque(so);  /* crashes if so is not in a queue */
 103
 104   if (so->so_tcpcb) {
 105       free(so->so_tcpcb);
 106   }
 107   g_free(so);
 108 }
 109
 110 size_t sopreprbuf(struct socket *so, struct iovec *iov, int *np)
 111 {
 112         int n, lss, total;
 113         struct sbuf *sb = &so->so_snd;
 114         int len = sb->sb_datalen - sb->sb_cc;
 115         int mss = so->so_tcpcb->t_maxseg;
 116
 117         DEBUG_CALL("sopreprbuf");
 118         DEBUG_ARG("so = %p", so);
 119
 120         if (len <= 0)
 121                 return 0;
 122
 123         iov[0].iov_base = sb->sb_wptr;
 124         iov[1].iov_base = NULL;
 125         iov[1].iov_len = 0;
 126         if (sb->sb_wptr < sb->sb_rptr) {
 127                 iov[0].iov_len = sb->sb_rptr - sb->sb_wptr;
 128                 /* Should never succeed, but... */
 129                 if (iov[0].iov_len > len)
 130                    iov[0].iov_len = len;
 131                 if (iov[0].iov_len > mss)
 132                    iov[0].iov_len -= iov[0].iov_len%mss;
 133                 n = 1;
 134         } else {
 135                 iov[0].iov_len = (sb->sb_data + sb->sb_datalen) - sb->sb_wptr;
 136                 /* Should never succeed, but... */
 137                 if (iov[0].iov_len > len) iov[0].iov_len = len;
 138                 len -= iov[0].iov_len;
 139                 if (len) {
 140                         iov[1].iov_base = sb->sb_data;
 141                         iov[1].iov_len = sb->sb_rptr - sb->sb_data;
 142                         if(iov[1].iov_len > len)
 143                            iov[1].iov_len = len;
 144                         total = iov[0].iov_len + iov[1].iov_len;
 145                         if (total > mss) {
 146                                 lss = total%mss;
 147                                 if (iov[1].iov_len > lss) {
 148                                         iov[1].iov_len -= lss;
 149                                         n = 2;
 150                                 } else {
 151                                         lss -= iov[1].iov_len;
 152                                         iov[0].iov_len -= lss;
 153                                         n = 1;
 154                                 }
 155                         } else
 156                                 n = 2;
 157                 } else {
 158                         if (iov[0].iov_len > mss)
 159                            iov[0].iov_len -= iov[0].iov_len%mss;
 160                         n = 1;
 161                 }
 162         }
 163         if (np)
 164                 *np = n;
 165
 166         return iov[0].iov_len + (n - 1) * iov[1].iov_len;
 167 }
 168
 169 /*
 170  * Read from so's socket into sb_snd, updating all relevant sbuf fields
 171  * NOTE: This will only be called if it is select()ed for reading, so
 172  * a read() of 0 (or less) means it's disconnected
 173  */
 174 int
 175 soread(struct socket *so)
 176 {
 177         int n, nn;
 178         struct sbuf *sb = &so->so_snd;
 179         struct iovec iov[2];
 180
 181         DEBUG_CALL("soread");
 182         DEBUG_ARG("so = %p", so);
 183
 184         /*
 185          * No need to check if there's enough room to read.
 186          * soread wouldn't have been called if there weren't
 187          */
 188         sopreprbuf(so, iov, &n);
 189
 190         nn = qemu_recv(so->s, iov[0].iov_base, iov[0].iov_len,0);
 191         if (nn <= 0) {
 192                 if (nn < 0 && (errno == EINTR || errno == EAGAIN))
 193                         return 0;
 194                 else {
 195                         int err;
 196                         socklen_t elen = sizeof err;
 197                         struct sockaddr_storage addr;
 198                         struct sockaddr *paddr = (struct sockaddr *) &addr;
 199                         socklen_t alen = sizeof addr;
 200
 201                         err = errno;
 202                         if (nn == 0) {
 203                                 if (getpeername(so->s, paddr, &alen) < 0) {
 204                                         err = errno;
 205                                 } else {
 206                                         getsockopt(so->s, SOL_SOCKET, SO_ERROR,
 207                                                 &err, &elen);
 208                                 }
 209                         }
 210
 211                         DEBUG_MISC(" --- soread() disconnected, nn = %d, errno = %d-%s",
 212                        nn, errno,strerror(errno));
 213                         sofcantrcvmore(so);
 214
 215                         if (err == ECONNRESET || err == ECONNREFUSED
 216                             || err == ENOTCONN || err == EPIPE) {
 217                                 tcp_drop(sototcpcb(so), err);
 218                         } else {
 219                                 tcp_sockclosed(sototcpcb(so));
 220                         }
 221                         return -1;
 222                 }
 223         }
 224
 225         /*
 226          * If there was no error, try and read the second time round
 227          * We read again if n = 2 (ie, there's another part of the buffer)
 228          * and we read as much as we could in the first read
 229          * We don't test for <= 0 this time, because there legitimately
 230          * might not be any more data (since the socket is non-blocking),
 231          * a close will be detected on next iteration.
 232          * A return of -1 won't (shouldn't) happen, since it didn't happen above
 233          */
 234         if (n == 2 && nn == iov[0].iov_len) {
 235             int ret;
 236             ret = qemu_recv(so->s, iov[1].iov_base, iov[1].iov_len,0);
 237             if (ret > 0)
 238                 nn += ret;
 239         }
 240
 241         DEBUG_MISC(" ... read nn = %d bytes", nn);
 242
 243         /* Update fields */
 244         sb->sb_cc += nn;
 245         sb->sb_wptr += nn;
 246         if (sb->sb_wptr >= (sb->sb_data + sb->sb_datalen))
 247                 sb->sb_wptr -= sb->sb_datalen;
 248         return nn;
 249 }
 250
 251 int soreadbuf(struct socket *so, const char *buf, int size)
 252 {
 253     int n, nn, copy = size;
 254         struct sbuf *sb = &so->so_snd;
 255         struct iovec iov[2];
 256
 257         DEBUG_CALL("soreadbuf");
 258         DEBUG_ARG("so = %p", so);
 259
 260         /*
 261          * No need to check if there's enough room to read.
 262          * soread wouldn't have been called if there weren't
 263          */
 264         if (sopreprbuf(so, iov, &n) < size)
 265         goto err;
 266
 267     nn = MIN(iov[0].iov_len, copy);
 268     memcpy(iov[0].iov_base, buf, nn);
 269
 270     copy -= nn;
 271     buf += nn;
 272
 273     if (copy == 0)
 274         goto done;
 275
 276     memcpy(iov[1].iov_base, buf, copy);
 277
 278 done:
 279     /* Update fields */
 280         sb->sb_cc += size;
 281         sb->sb_wptr += size;
 282         if (sb->sb_wptr >= (sb->sb_data + sb->sb_datalen))
 283                 sb->sb_wptr -= sb->sb_datalen;
 284     return size;
 285 err:
 286
 287     sofcantrcvmore(so);
 288     tcp_sockclosed(sototcpcb(so));
 289     g_critical("soreadbuf buffer too small");
 290     return -1;
 291 }
 292
 293 /*
 294  * Get urgent data
 295  *
 296  * When the socket is created, we set it SO_OOBINLINE,
 297  * so when OOB data arrives, we soread() it and everything
 298  * in the send buffer is sent as urgent data
 299  */
 300 int
 301 sorecvoob(struct socket *so)
 302 {
 303         struct tcpcb *tp = sototcpcb(so);
 304         int ret;
 305
 306         DEBUG_CALL("sorecvoob");
 307         DEBUG_ARG("so = %p", so);
 308
 309         /*
 310          * We take a guess at how much urgent data has arrived.
 311          * In most situations, when urgent data arrives, the next
 312          * read() should get all the urgent data.  This guess will
 313          * be wrong however if more data arrives just after the
 314          * urgent data, or the read() doesn't return all the
 315          * urgent data.
 316          */
 317         ret = soread(so);
 318         if (ret > 0) {
 319             tp->snd_up = tp->snd_una + so->so_snd.sb_cc;
 320             tp->t_force = 1;
 321             tcp_output(tp);
 322             tp->t_force = 0;
 323         }
 324
 325         return ret;
 326 }
 327
 328 /*
 329  * Send urgent data
 330  * There's a lot duplicated code here, but...
 331  */
 332 int
 333 sosendoob(struct socket *so)
 334 {
 335         struct sbuf *sb = &so->so_rcv;
 336         char buff[2048]; /* XXX Shouldn't be sending more oob data than this */
 337
 338         int n;
 339
 340         DEBUG_CALL("sosendoob");
 341         DEBUG_ARG("so = %p", so);
 342         DEBUG_ARG("sb->sb_cc = %d", sb->sb_cc);
 343
 344         if (so->so_urgc > 2048)
 345            so->so_urgc = 2048; /* XXXX */
 346
 347         if (sb->sb_rptr < sb->sb_wptr) {
 348                 /* We can send it directly */
 349                 n = slirp_send(so, sb->sb_rptr, so->so_urgc, (MSG_OOB)); /* |MSG_DONTWAIT)); */
 350         } else {
 351                 /*
 352                  * Since there's no sendv or sendtov like writev,
 353                  * we must copy all data to a linear buffer then
 354                  * send it all
 355                  */
 356                 uint32_t urgc = so->so_urgc;
 357                 int len = (sb->sb_data + sb->sb_datalen) - sb->sb_rptr;
 358                 if (len > urgc) {
 359                         len = urgc;
 360                 }
 361                 memcpy(buff, sb->sb_rptr, len);
 362                 urgc -= len;
 363                 if (urgc) {
 364                         n = sb->sb_wptr - sb->sb_data;
 365                         if (n > urgc) {
 366                                 n = urgc;
 367                         }
 368                         memcpy((buff + len), sb->sb_data, n);
 369                         len += n;
 370                 }
 371                 n = slirp_send(so, buff, len, (MSG_OOB)); /* |MSG_DONTWAIT)); */
 372 #ifdef DEBUG
 373                 if (n != len) {
 374                         DEBUG_ERROR("Didn't send all data urgently XXXXX");
 375                 }
 376 #endif
 377         }
 378
 379         if (n < 0) {
 380                 return n;
 381         }
 382         so->so_urgc -= n;
 383         DEBUG_MISC(" ---2 sent %d bytes urgent data, %d urgent bytes left", n, so->so_urgc);
 384
 385         sb->sb_cc -= n;
 386         sb->sb_rptr += n;
 387         if (sb->sb_rptr >= (sb->sb_data + sb->sb_datalen))
 388                 sb->sb_rptr -= sb->sb_datalen;
 389
 390         return n;
 391 }
 392
 393 /*
 394  * Write data from so_rcv to so's socket,
 395  * updating all sbuf field as necessary
 396  */
 397 int
 398 sowrite(struct socket *so)
 399 {
 400         int  n,nn;
 401         struct sbuf *sb = &so->so_rcv;
 402         int len = sb->sb_cc;
 403         struct iovec iov[2];
 404
 405         DEBUG_CALL("sowrite");
 406         DEBUG_ARG("so = %p", so);
 407
 408         if (so->so_urgc) {
 409                 uint32_t expected = so->so_urgc;
 410                 if (sosendoob(so) < expected) {
 411                         /* Treat a short write as a fatal error too,
 412                          * rather than continuing on and sending the urgent
 413                          * data as if it were non-urgent and leaving the
 414                          * so_urgc count wrong.
 415                          */
 416                         goto err_disconnected;
 417                 }
 418                 if (sb->sb_cc == 0)
 419                         return 0;
 420         }
 421
 422         /*
 423          * No need to check if there's something to write,
 424          * sowrite wouldn't have been called otherwise
 425          */
 426
 427         iov[0].iov_base = sb->sb_rptr;
 428         iov[1].iov_base = NULL;
 429         iov[1].iov_len = 0;
 430         if (sb->sb_rptr < sb->sb_wptr) {
 431                 iov[0].iov_len = sb->sb_wptr - sb->sb_rptr;
 432                 /* Should never succeed, but... */
 433                 if (iov[0].iov_len > len) iov[0].iov_len = len;
 434                 n = 1;
 435         } else {
 436                 iov[0].iov_len = (sb->sb_data + sb->sb_datalen) - sb->sb_rptr;
 437                 if (iov[0].iov_len > len) iov[0].iov_len = len;
 438                 len -= iov[0].iov_len;
 439                 if (len) {
 440                         iov[1].iov_base = sb->sb_data;
 441                         iov[1].iov_len = sb->sb_wptr - sb->sb_data;
 442                         if (iov[1].iov_len > len) iov[1].iov_len = len;
 443                         n = 2;
 444                 } else
 445                         n = 1;
 446         }
 447         /* Check if there's urgent data to send, and if so, send it */
 448
 449         nn = slirp_send(so, iov[0].iov_base, iov[0].iov_len,0);
 450         /* This should never happen, but people tell me it does *shrug* */
 451         if (nn < 0 && (errno == EAGAIN || errno == EINTR))
 452                 return 0;
 453
 454         if (nn <= 0) {
 455                 goto err_disconnected;
 456         }
 457
 458         if (n == 2 && nn == iov[0].iov_len) {
 459             int ret;
 460             ret = slirp_send(so, iov[1].iov_base, iov[1].iov_len,0);
 461             if (ret > 0)
 462                 nn += ret;
 463         }
 464         DEBUG_MISC("  ... wrote nn = %d bytes", nn);
 465
 466         /* Update sbuf */
 467         sb->sb_cc -= nn;
 468         sb->sb_rptr += nn;
 469         if (sb->sb_rptr >= (sb->sb_data + sb->sb_datalen))
 470                 sb->sb_rptr -= sb->sb_datalen;
 471
 472         /*
 473          * If in DRAIN mode, and there's no more data, set
 474          * it CANTSENDMORE
 475          */
 476         if ((so->so_state & SS_FWDRAIN) && sb->sb_cc == 0)
 477                 sofcantsendmore(so);
 478
 479         return nn;
 480
 481 err_disconnected:
 482         DEBUG_MISC(" --- sowrite disconnected, so->so_state = %x, errno = %d",
 483                so->so_state, errno);
 484         sofcantsendmore(so);
 485         tcp_sockclosed(sototcpcb(so));
 486         return -1;
 487 }
 488
 489 /*
 490  * recvfrom() a UDP socket
 491  */
 492 void
 493 sorecvfrom(struct socket *so)
 494 {
 495         struct sockaddr_storage addr;
 496         struct sockaddr_storage saddr, daddr;
 497         socklen_t addrlen = sizeof(struct sockaddr_storage);
 498
 499         DEBUG_CALL("sorecvfrom");
 500         DEBUG_ARG("so = %p", so);
 501
 502         if (so->so_type == IPPROTO_ICMP) {   /* This is a "ping" reply */
 503           char buff[256];
 504           int len;
 505
 506           len = recvfrom(so->s, buff, 256, 0,
 507                          (struct sockaddr *)&addr, &addrlen);
 508           /* XXX Check if reply is "correct"? */
 509
 510           if(len == -1 || len == 0) {
 511             u_char code=ICMP_UNREACH_PORT;
 512
 513             if(errno == EHOSTUNREACH) code=ICMP_UNREACH_HOST;
 514             else if(errno == ENETUNREACH) code=ICMP_UNREACH_NET;
 515
 516             DEBUG_MISC(" udp icmp rx errno = %d-%s",
 517                    errno,strerror(errno));
 518             icmp_send_error(so->so_m, ICMP_UNREACH, code, 0, strerror(errno));
 519           } else {
 520             icmp_reflect(so->so_m);
 521             so->so_m = NULL; /* Don't m_free() it again! */
 522           }
 523           /* No need for this socket anymore, udp_detach it */
 524           udp_detach(so);
 525         } else {                                /* A "normal" UDP packet */
 526           struct mbuf *m;
 527           int len;
 528 #ifdef _WIN32
 529           unsigned long n;
 530 #else
 531           int n;
 532 #endif
 533
 534           m = m_get(so->slirp);
 535           if (!m) {
 536               return;
 537           }
 538           switch (so->so_ffamily) {
 539           case AF_INET:
 540               m->m_data += IF_MAXLINKHDR + sizeof(struct udpiphdr);
 541               break;
 542           case AF_INET6:
 543               m->m_data += IF_MAXLINKHDR + sizeof(struct ip6)
 544                                          + sizeof(struct udphdr);
 545               break;
 546           default:
 547               g_assert_not_reached();
 548               break;
 549           }
 550
 551           /*
 552            * XXX Shouldn't FIONREAD packets destined for port 53,
 553            * but I don't know the max packet size for DNS lookups
 554            */
 555           len = M_FREEROOM(m);
 556           /* if (so->so_fport != htons(53)) { */
 557           ioctlsocket(so->s, FIONREAD, &n);
 558
 559           if (n > len) {
 560             n = (m->m_data - m->m_dat) + m->m_len + n + 1;
 561             m_inc(m, n);
 562             len = M_FREEROOM(m);
 563           }
 564           /* } */
 565
 566           m->m_len = recvfrom(so->s, m->m_data, len, 0,
 567                               (struct sockaddr *)&addr, &addrlen);
 568           DEBUG_MISC(" did recvfrom %d, errno = %d-%s",
 569                  m->m_len, errno,strerror(errno));
 570           if(m->m_len<0) {
 571             /* Report error as ICMP */
 572             switch (so->so_lfamily) {
 573             uint8_t code;
 574             case AF_INET:
 575               code = ICMP_UNREACH_PORT;
 576
 577               if (errno == EHOSTUNREACH) {
 578                 code = ICMP_UNREACH_HOST;
 579               } else if (errno == ENETUNREACH) {
 580                 code = ICMP_UNREACH_NET;
 581               }
 582
 583               DEBUG_MISC(" rx error, tx icmp ICMP_UNREACH:%i", code);
 584               icmp_send_error(so->so_m, ICMP_UNREACH, code, 0, strerror(errno));
 585               break;
 586             case AF_INET6:
 587               code = ICMP6_UNREACH_PORT;
 588
 589               if (errno == EHOSTUNREACH) {
 590                 code = ICMP6_UNREACH_ADDRESS;
 591               } else if (errno == ENETUNREACH) {
 592                 code = ICMP6_UNREACH_NO_ROUTE;
 593               }
 594
 595               DEBUG_MISC(" rx error, tx icmp6 ICMP_UNREACH:%i", code);
 596               icmp6_send_error(so->so_m, ICMP6_UNREACH, code);
 597               break;
 598             default:
 599               g_assert_not_reached();
 600               break;
 601             }
 602             m_free(m);
 603           } else {
 604           /*
 605            * Hack: domain name lookup will be used the most for UDP,
 606            * and since they'll only be used once there's no need
 607            * for the 4 minute (or whatever) timeout... So we time them
 608            * out much quicker (10 seconds  for now...)
 609            */
 610             if (so->so_expire) {
 611               if (so->so_fport == htons(53))
 612                 so->so_expire = curtime + SO_EXPIREFAST;
 613               else
 614                 so->so_expire = curtime + SO_EXPIRE;
 615             }
 616
 617             /*
 618              * If this packet was destined for CTL_ADDR,
 619              * make it look like that's where it came from
 620              */
 621             saddr = addr;
 622             sotranslate_in(so, &saddr);
 623             daddr = so->lhost.ss;
 624
 625             switch (so->so_ffamily) {
 626             case AF_INET:
 627                 udp_output(so, m, (struct sockaddr_in *) &saddr,
 628                            (struct sockaddr_in *) &daddr,
 629                            so->so_iptos);
 630                 break;
 631             case AF_INET6:
 632                 udp6_output(so, m, (struct sockaddr_in6 *) &saddr,
 633                             (struct sockaddr_in6 *) &daddr);
 634                 break;
 635             default:
 636                 g_assert_not_reached();
 637                 break;
 638             }
 639           } /* rx error */
 640         } /* if ping packet */
 641 }
 642
 643 /*
 644  * sendto() a socket
 645  */
 646 int
 647 sosendto(struct socket *so, struct mbuf *m)
 648 {
 649         int ret;
 650         struct sockaddr_storage addr;
 651
 652         DEBUG_CALL("sosendto");
 653         DEBUG_ARG("so = %p", so);
 654         DEBUG_ARG("m = %p", m);
 655
 656         addr = so->fhost.ss;
 657         DEBUG_CALL(" sendto()ing)");
 658         sotranslate_out(so, &addr);
 659
 660         /* Don't care what port we get */
 661         ret = sendto(so->s, m->m_data, m->m_len, 0,
 662                      (struct sockaddr *)&addr, sockaddr_size(&addr));
 663         if (ret < 0)
 664                 return -1;
 665
 666         /*
 667          * Kill the socket if there's no reply in 4 minutes,
 668          * but only if it's an expirable socket
 669          */
 670         if (so->so_expire)
 671                 so->so_expire = curtime + SO_EXPIRE;
 672         so->so_state &= SS_PERSISTENT_MASK;
 673         so->so_state |= SS_ISFCONNECTED; /* So that it gets select()ed */
 674         return 0;
 675 }
 676
 677 /*
 678  * Listen for incoming TCP connections
 679  */
 680 struct socket *
 681 tcp_listen(Slirp *slirp, uint32_t haddr, u_int hport, uint32_t laddr,
 682            u_int lport, int flags)
 683 {
 684         struct sockaddr_in addr;
 685         struct socket *so;
 686         int s, opt = 1;
 687         socklen_t addrlen = sizeof(addr);
 688         memset(&addr, 0, addrlen);
 689
 690         DEBUG_CALL("tcp_listen");
 691         DEBUG_ARG("haddr = %s", inet_ntoa((struct in_addr){.s_addr = haddr}));
 692         DEBUG_ARG("hport = %d", ntohs(hport));
 693         DEBUG_ARG("laddr = %s", inet_ntoa((struct in_addr){.s_addr = laddr}));
 694         DEBUG_ARG("lport = %d", ntohs(lport));
 695         DEBUG_ARG("flags = %x", flags);
 696
 697         so = socreate(slirp);
 698
 699         /* Don't tcp_attach... we don't need so_snd nor so_rcv */
 700         if ((so->so_tcpcb = tcp_newtcpcb(so)) == NULL) {
 701             g_free(so);
 702             return NULL;
 703         }
 704         insque(so, &slirp->tcb);
 705
 706         /*
 707          * SS_FACCEPTONCE sockets must time out.
 708          */
 709         if (flags & SS_FACCEPTONCE)
 710            so->so_tcpcb->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT*2;
 711
 712         so->so_state &= SS_PERSISTENT_MASK;
 713         so->so_state |= (SS_FACCEPTCONN | flags);
 714         so->so_lfamily = AF_INET;
 715         so->so_lport = lport; /* Kept in network format */
 716         so->so_laddr.s_addr = laddr; /* Ditto */
 717
 718         addr.sin_family = AF_INET;
 719         addr.sin_addr.s_addr = haddr;
 720         addr.sin_port = hport;
 721
 722         if (((s = qemu_socket(AF_INET,SOCK_STREAM,0)) < 0) ||
 723             (socket_set_fast_reuse(s) < 0) ||
 724             (bind(s,(struct sockaddr *)&addr, sizeof(addr)) < 0) ||
 725             (listen(s,1) < 0)) {
 726                 int tmperrno = errno; /* Don't clobber the real reason we failed */
 727
 728                 if (s >= 0) {
 729                     closesocket(s);
 730                 }
 731                 sofree(so);
 732                 /* Restore the real errno */
 733 #ifdef _WIN32
 734                 WSASetLastError(tmperrno);
 735 #else
 736                 errno = tmperrno;
 737 #endif
 738                 return NULL;
 739         }
 740         qemu_setsockopt(s, SOL_SOCKET, SO_OOBINLINE, &opt, sizeof(int));
 741         opt = 1;
 742         qemu_setsockopt(s, IPPROTO_TCP, TCP_NODELAY, &opt, sizeof(int));
 743
 744         getsockname(s,(struct sockaddr *)&addr,&addrlen);
 745         so->so_ffamily = AF_INET;
 746         so->so_fport = addr.sin_port;
 747         if (addr.sin_addr.s_addr == 0 || addr.sin_addr.s_addr == loopback_addr.s_addr)
 748            so->so_faddr = slirp->vhost_addr;
 749         else
 750            so->so_faddr = addr.sin_addr;
 751
 752         so->s = s;
 753         return so;
 754 }
 755
 756 /*
 757  * Various session state calls
 758  * XXX Should be #define's
 759  * The socket state stuff needs work, these often get call 2 or 3
 760  * times each when only 1 was needed
 761  */
 762 void
 763 soisfconnecting(struct socket *so)
 764 {
 765         so->so_state &= ~(SS_NOFDREF|SS_ISFCONNECTED|SS_FCANTRCVMORE|
 766                           SS_FCANTSENDMORE|SS_FWDRAIN);
 767         so->so_state |= SS_ISFCONNECTING; /* Clobber other states */
 768 }
 769
 770 void
 771 soisfconnected(struct socket *so)
 772 {
 773         so->so_state &= ~(SS_ISFCONNECTING|SS_FWDRAIN|SS_NOFDREF);
 774         so->so_state |= SS_ISFCONNECTED; /* Clobber other states */
 775 }
 776
 777 static void
 778 sofcantrcvmore(struct socket *so)
 779 {
 780         if ((so->so_state & SS_NOFDREF) == 0) {
 781                 shutdown(so->s,0);
 782         }
 783         so->so_state &= ~(SS_ISFCONNECTING);
 784         if (so->so_state & SS_FCANTSENDMORE) {
 785            so->so_state &= SS_PERSISTENT_MASK;
 786            so->so_state |= SS_NOFDREF; /* Don't select it */
 787         } else {
 788            so->so_state |= SS_FCANTRCVMORE;
 789         }
 790 }
 791
 792 static void
 793 sofcantsendmore(struct socket *so)
 794 {
 795         if ((so->so_state & SS_NOFDREF) == 0) {
 796             shutdown(so->s,1);           /* send FIN to fhost */
 797         }
 798         so->so_state &= ~(SS_ISFCONNECTING);
 799         if (so->so_state & SS_FCANTRCVMORE) {
 800            so->so_state &= SS_PERSISTENT_MASK;
 801            so->so_state |= SS_NOFDREF; /* as above */
 802         } else {
 803            so->so_state |= SS_FCANTSENDMORE;
 804         }
 805 }
 806
 807 /*
 808  * Set write drain mode
 809  * Set CANTSENDMORE once all data has been write()n
 810  */
 811 void
 812 sofwdrain(struct socket *so)
 813 {
 814         if (so->so_rcv.sb_cc)
 815                 so->so_state |= SS_FWDRAIN;
 816         else
 817                 sofcantsendmore(so);
 818 }
 819
 820 /*
 821  * Translate addr in host addr when it is a virtual address
 822  */
 823 void sotranslate_out(struct socket *so, struct sockaddr_storage *addr)
 824 {
 825     Slirp *slirp = so->slirp;
 826     struct sockaddr_in *sin = (struct sockaddr_in *)addr;
 827     struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)addr;
 828
 829     switch (addr->ss_family) {
 830     case AF_INET:
 831         if ((so->so_faddr.s_addr & slirp->vnetwork_mask.s_addr) ==
 832                 slirp->vnetwork_addr.s_addr) {
 833             /* It's an alias */
 834             if (so->so_faddr.s_addr == slirp->vnameserver_addr.s_addr) {
 835                 if (get_dns_addr(&sin->sin_addr) < 0) {
 836                     sin->sin_addr = loopback_addr;
 837                 }
 838             } else {
 839                 sin->sin_addr = loopback_addr;
 840             }
 841         }
 842
 843         DEBUG_MISC(" addr.sin_port=%d, addr.sin_addr.s_addr=%.16s",
 844                    ntohs(sin->sin_port), inet_ntoa(sin->sin_addr));
 845         break;
 846
 847     case AF_INET6:
 848         if (in6_equal_net(&so->so_faddr6, &slirp->vprefix_addr6,
 849                     slirp->vprefix_len)) {
 850             if (in6_equal(&so->so_faddr6, &slirp->vnameserver_addr6)) {
 851                 uint32_t scope_id;
 852                 if (get_dns6_addr(&sin6->sin6_addr, &scope_id) >= 0) {
 853                     sin6->sin6_scope_id = scope_id;
 854                 } else {
 855                     sin6->sin6_addr = in6addr_loopback;
 856                 }
 857             } else {
 858                 sin6->sin6_addr = in6addr_loopback;
 859             }
 860         }
 861         break;
 862
 863     default:
 864         break;
 865     }
 866 }
 867
 868 void sotranslate_in(struct socket *so, struct sockaddr_storage *addr)
 869 {
 870     Slirp *slirp = so->slirp;
 871     struct sockaddr_in *sin = (struct sockaddr_in *)addr;
 872     struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)addr;
 873
 874     switch (addr->ss_family) {
 875     case AF_INET:
 876         if ((so->so_faddr.s_addr & slirp->vnetwork_mask.s_addr) ==
 877             slirp->vnetwork_addr.s_addr) {
 878             uint32_t inv_mask = ~slirp->vnetwork_mask.s_addr;
 879
 880             if ((so->so_faddr.s_addr & inv_mask) == inv_mask) {
 881                 sin->sin_addr = slirp->vhost_addr;
 882             } else if (sin->sin_addr.s_addr == loopback_addr.s_addr ||
 883                        so->so_faddr.s_addr != slirp->vhost_addr.s_addr) {
 884                 sin->sin_addr = so->so_faddr;
 885             }
 886         }
 887         break;
 888
 889     case AF_INET6:
 890         if (in6_equal_net(&so->so_faddr6, &slirp->vprefix_addr6,
 891                     slirp->vprefix_len)) {
 892             if (in6_equal(&sin6->sin6_addr, &in6addr_loopback)
 893                     || !in6_equal(&so->so_faddr6, &slirp->vhost_addr6)) {
 894                 sin6->sin6_addr = so->so_faddr6;
 895             }
 896         }
 897         break;
 898
 899     default:
 900         break;
 901     }
 902 }
 903
 904 /*
 905  * Translate connections from localhost to the real hostname
 906  */
 907 void sotranslate_accept(struct socket *so)
 908 {
 909     Slirp *slirp = so->slirp;
 910
 911     switch (so->so_ffamily) {
 912     case AF_INET:
 913         if (so->so_faddr.s_addr == INADDR_ANY ||
 914             (so->so_faddr.s_addr & loopback_mask) ==
 915             (loopback_addr.s_addr & loopback_mask)) {
 916            so->so_faddr = slirp->vhost_addr;
 917         }
 918         break;
 919
 920    case AF_INET6:
 921         if (in6_equal(&so->so_faddr6, &in6addr_any) ||
 922                 in6_equal(&so->so_faddr6, &in6addr_loopback)) {
 923            so->so_faddr6 = slirp->vhost_addr6;
 924         }
 925         break;
 926
 927     default:
 928         break;
 929     }
 930 }