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->so_emu==EMU_RSH && so->extra) {
  93         sofree(so->extra);
  94         so->extra=NULL;
  95   }
  96   if (so == slirp->tcp_last_so) {
  97       slirp->tcp_last_so = &slirp->tcb;
  98   } else if (so == slirp->udp_last_so) {
  99       slirp->udp_last_so = &slirp->udb;
 100   } else if (so == slirp->icmp_last_so) {
 101       slirp->icmp_last_so = &slirp->icmp;
 102   }
 103   m_free(so->so_m);
 104
 105   if(so->so_next && so->so_prev)
 106     remque(so);  /* crashes if so is not in a queue */
 107
 108   if (so->so_tcpcb) {
 109       free(so->so_tcpcb);
 110   }
 111   g_free(so);
 112 }
 113
 114 size_t sopreprbuf(struct socket *so, struct iovec *iov, int *np)
 115 {
 116         int n, lss, total;
 117         struct sbuf *sb = &so->so_snd;
 118         int len = sb->sb_datalen - sb->sb_cc;
 119         int mss = so->so_tcpcb->t_maxseg;
 120
 121         DEBUG_CALL("sopreprbuf");
 122         DEBUG_ARG("so = %p", so);
 123
 124         if (len <= 0)
 125                 return 0;
 126
 127         iov[0].iov_base = sb->sb_wptr;
 128         iov[1].iov_base = NULL;
 129         iov[1].iov_len = 0;
 130         if (sb->sb_wptr < sb->sb_rptr) {
 131                 iov[0].iov_len = sb->sb_rptr - sb->sb_wptr;
 132                 /* Should never succeed, but... */
 133                 if (iov[0].iov_len > len)
 134                    iov[0].iov_len = len;
 135                 if (iov[0].iov_len > mss)
 136                    iov[0].iov_len -= iov[0].iov_len%mss;
 137                 n = 1;
 138         } else {
 139                 iov[0].iov_len = (sb->sb_data + sb->sb_datalen) - sb->sb_wptr;
 140                 /* Should never succeed, but... */
 141                 if (iov[0].iov_len > len) iov[0].iov_len = len;
 142                 len -= iov[0].iov_len;
 143                 if (len) {
 144                         iov[1].iov_base = sb->sb_data;
 145                         iov[1].iov_len = sb->sb_rptr - sb->sb_data;
 146                         if(iov[1].iov_len > len)
 147                            iov[1].iov_len = len;
 148                         total = iov[0].iov_len + iov[1].iov_len;
 149                         if (total > mss) {
 150                                 lss = total%mss;
 151                                 if (iov[1].iov_len > lss) {
 152                                         iov[1].iov_len -= lss;
 153                                         n = 2;
 154                                 } else {
 155                                         lss -= iov[1].iov_len;
 156                                         iov[0].iov_len -= lss;
 157                                         n = 1;
 158                                 }
 159                         } else
 160                                 n = 2;
 161                 } else {
 162                         if (iov[0].iov_len > mss)
 163                            iov[0].iov_len -= iov[0].iov_len%mss;
 164                         n = 1;
 165                 }
 166         }
 167         if (np)
 168                 *np = n;
 169
 170         return iov[0].iov_len + (n - 1) * iov[1].iov_len;
 171 }
 172
 173 /*
 174  * Read from so's socket into sb_snd, updating all relevant sbuf fields
 175  * NOTE: This will only be called if it is select()ed for reading, so
 176  * a read() of 0 (or less) means it's disconnected
 177  */
 178 int
 179 soread(struct socket *so)
 180 {
 181         int n, nn;
 182         struct sbuf *sb = &so->so_snd;
 183         struct iovec iov[2];
 184
 185         DEBUG_CALL("soread");
 186         DEBUG_ARG("so = %p", so);
 187
 188         /*
 189          * No need to check if there's enough room to read.
 190          * soread wouldn't have been called if there weren't
 191          */
 192         sopreprbuf(so, iov, &n);
 193
 194 #ifdef HAVE_READV
 195         nn = readv(so->s, (struct iovec *)iov, n);
 196         DEBUG_MISC((dfd, " ... read nn = %d bytes\n", nn));
 197 #else
 198         nn = qemu_recv(so->s, iov[0].iov_base, iov[0].iov_len,0);
 199 #endif
 200         if (nn <= 0) {
 201                 if (nn < 0 && (errno == EINTR || errno == EAGAIN))
 202                         return 0;
 203                 else {
 204                         int err;
 205                         socklen_t elen = sizeof err;
 206                         struct sockaddr_storage addr;
 207                         struct sockaddr *paddr = (struct sockaddr *) &addr;
 208                         socklen_t alen = sizeof addr;
 209
 210                         err = errno;
 211                         if (nn == 0) {
 212                                 if (getpeername(so->s, paddr, &alen) < 0) {
 213                                         err = errno;
 214                                 } else {
 215                                         getsockopt(so->s, SOL_SOCKET, SO_ERROR,
 216                                                 &err, &elen);
 217                                 }
 218                         }
 219
 220                         DEBUG_MISC((dfd, " --- soread() disconnected, nn = %d, errno = %d-%s\n", nn, errno,strerror(errno)));
 221                         sofcantrcvmore(so);
 222
 223                         if (err == ECONNRESET || err == ECONNREFUSED
 224                             || err == ENOTCONN || err == EPIPE) {
 225                                 tcp_drop(sototcpcb(so), err);
 226                         } else {
 227                                 tcp_sockclosed(sototcpcb(so));
 228                         }
 229                         return -1;
 230                 }
 231         }
 232
 233 #ifndef HAVE_READV
 234         /*
 235          * If there was no error, try and read the second time round
 236          * We read again if n = 2 (ie, there's another part of the buffer)
 237          * and we read as much as we could in the first read
 238          * We don't test for <= 0 this time, because there legitimately
 239          * might not be any more data (since the socket is non-blocking),
 240          * a close will be detected on next iteration.
 241          * A return of -1 won't (shouldn't) happen, since it didn't happen above
 242          */
 243         if (n == 2 && nn == iov[0].iov_len) {
 244             int ret;
 245             ret = qemu_recv(so->s, iov[1].iov_base, iov[1].iov_len,0);
 246             if (ret > 0)
 247                 nn += ret;
 248         }
 249
 250         DEBUG_MISC((dfd, " ... read nn = %d bytes\n", nn));
 251 #endif
 252
 253         /* Update fields */
 254         sb->sb_cc += nn;
 255         sb->sb_wptr += nn;
 256         if (sb->sb_wptr >= (sb->sb_data + sb->sb_datalen))
 257                 sb->sb_wptr -= sb->sb_datalen;
 258         return nn;
 259 }
 260
 261 int soreadbuf(struct socket *so, const char *buf, int size)
 262 {
 263     int n, nn, copy = size;
 264         struct sbuf *sb = &so->so_snd;
 265         struct iovec iov[2];
 266
 267         DEBUG_CALL("soreadbuf");
 268         DEBUG_ARG("so = %p", so);
 269
 270         /*
 271          * No need to check if there's enough room to read.
 272          * soread wouldn't have been called if there weren't
 273          */
 274         if (sopreprbuf(so, iov, &n) < size)
 275         goto err;
 276
 277     nn = MIN(iov[0].iov_len, copy);
 278     memcpy(iov[0].iov_base, buf, nn);
 279
 280     copy -= nn;
 281     buf += nn;
 282
 283     if (copy == 0)
 284         goto done;
 285
 286     memcpy(iov[1].iov_base, buf, copy);
 287
 288 done:
 289     /* Update fields */
 290         sb->sb_cc += size;
 291         sb->sb_wptr += size;
 292         if (sb->sb_wptr >= (sb->sb_data + sb->sb_datalen))
 293                 sb->sb_wptr -= sb->sb_datalen;
 294     return size;
 295 err:
 296
 297     sofcantrcvmore(so);
 298     tcp_sockclosed(sototcpcb(so));
 299     fprintf(stderr, "soreadbuf buffer to small");
 300     return -1;
 301 }
 302
 303 /*
 304  * Get urgent data
 305  *
 306  * When the socket is created, we set it SO_OOBINLINE,
 307  * so when OOB data arrives, we soread() it and everything
 308  * in the send buffer is sent as urgent data
 309  */
 310 int
 311 sorecvoob(struct socket *so)
 312 {
 313         struct tcpcb *tp = sototcpcb(so);
 314         int ret;
 315
 316         DEBUG_CALL("sorecvoob");
 317         DEBUG_ARG("so = %p", so);
 318
 319         /*
 320          * We take a guess at how much urgent data has arrived.
 321          * In most situations, when urgent data arrives, the next
 322          * read() should get all the urgent data.  This guess will
 323          * be wrong however if more data arrives just after the
 324          * urgent data, or the read() doesn't return all the
 325          * urgent data.
 326          */
 327         ret = soread(so);
 328         if (ret > 0) {
 329             tp->snd_up = tp->snd_una + so->so_snd.sb_cc;
 330             tp->t_force = 1;
 331             tcp_output(tp);
 332             tp->t_force = 0;
 333         }
 334
 335         return ret;
 336 }
 337
 338 /*
 339  * Send urgent data
 340  * There's a lot duplicated code here, but...
 341  */
 342 int
 343 sosendoob(struct socket *so)
 344 {
 345         struct sbuf *sb = &so->so_rcv;
 346         char buff[2048]; /* XXX Shouldn't be sending more oob data than this */
 347
 348         int n;
 349
 350         DEBUG_CALL("sosendoob");
 351         DEBUG_ARG("so = %p", so);
 352         DEBUG_ARG("sb->sb_cc = %d", sb->sb_cc);
 353
 354         if (so->so_urgc > 2048)
 355            so->so_urgc = 2048; /* XXXX */
 356
 357         if (sb->sb_rptr < sb->sb_wptr) {
 358                 /* We can send it directly */
 359                 n = slirp_send(so, sb->sb_rptr, so->so_urgc, (MSG_OOB)); /* |MSG_DONTWAIT)); */
 360         } else {
 361                 /*
 362                  * Since there's no sendv or sendtov like writev,
 363                  * we must copy all data to a linear buffer then
 364                  * send it all
 365                  */
 366                 uint32_t urgc = so->so_urgc;
 367                 int len = (sb->sb_data + sb->sb_datalen) - sb->sb_rptr;
 368                 if (len > urgc) {
 369                         len = urgc;
 370                 }
 371                 memcpy(buff, sb->sb_rptr, len);
 372                 urgc -= len;
 373                 if (urgc) {
 374                         n = sb->sb_wptr - sb->sb_data;
 375                         if (n > urgc) {
 376                                 n = urgc;
 377                         }
 378                         memcpy((buff + len), sb->sb_data, n);
 379                         len += n;
 380                 }
 381                 n = slirp_send(so, buff, len, (MSG_OOB)); /* |MSG_DONTWAIT)); */
 382 #ifdef DEBUG
 383                 if (n != len) {
 384                         DEBUG_ERROR((dfd, "Didn't send all data urgently XXXXX\n"));
 385                 }
 386 #endif
 387         }
 388
 389         if (n < 0) {
 390                 return n;
 391         }
 392         so->so_urgc -= n;
 393         DEBUG_MISC((dfd, " ---2 sent %d bytes urgent data, %d urgent bytes left\n", n, so->so_urgc));
 394
 395         sb->sb_cc -= n;
 396         sb->sb_rptr += n;
 397         if (sb->sb_rptr >= (sb->sb_data + sb->sb_datalen))
 398                 sb->sb_rptr -= sb->sb_datalen;
 399
 400         return n;
 401 }
 402
 403 /*
 404  * Write data from so_rcv to so's socket,
 405  * updating all sbuf field as necessary
 406  */
 407 int
 408 sowrite(struct socket *so)
 409 {
 410         int  n,nn;
 411         struct sbuf *sb = &so->so_rcv;
 412         int len = sb->sb_cc;
 413         struct iovec iov[2];
 414
 415         DEBUG_CALL("sowrite");
 416         DEBUG_ARG("so = %p", so);
 417
 418         if (so->so_urgc) {
 419                 uint32_t expected = so->so_urgc;
 420                 if (sosendoob(so) < expected) {
 421                         /* Treat a short write as a fatal error too,
 422                          * rather than continuing on and sending the urgent
 423                          * data as if it were non-urgent and leaving the
 424                          * so_urgc count wrong.
 425                          */
 426                         goto err_disconnected;
 427                 }
 428                 if (sb->sb_cc == 0)
 429                         return 0;
 430         }
 431
 432         /*
 433          * No need to check if there's something to write,
 434          * sowrite wouldn't have been called otherwise
 435          */
 436
 437         iov[0].iov_base = sb->sb_rptr;
 438         iov[1].iov_base = NULL;
 439         iov[1].iov_len = 0;
 440         if (sb->sb_rptr < sb->sb_wptr) {
 441                 iov[0].iov_len = sb->sb_wptr - sb->sb_rptr;
 442                 /* Should never succeed, but... */
 443                 if (iov[0].iov_len > len) iov[0].iov_len = len;
 444                 n = 1;
 445         } else {
 446                 iov[0].iov_len = (sb->sb_data + sb->sb_datalen) - sb->sb_rptr;
 447                 if (iov[0].iov_len > len) iov[0].iov_len = len;
 448                 len -= iov[0].iov_len;
 449                 if (len) {
 450                         iov[1].iov_base = sb->sb_data;
 451                         iov[1].iov_len = sb->sb_wptr - sb->sb_data;
 452                         if (iov[1].iov_len > len) iov[1].iov_len = len;
 453                         n = 2;
 454                 } else
 455                         n = 1;
 456         }
 457         /* Check if there's urgent data to send, and if so, send it */
 458
 459 #ifdef HAVE_READV
 460         nn = writev(so->s, (const struct iovec *)iov, n);
 461
 462         DEBUG_MISC((dfd, "  ... wrote nn = %d bytes\n", nn));
 463 #else
 464         nn = slirp_send(so, iov[0].iov_base, iov[0].iov_len,0);
 465 #endif
 466         /* This should never happen, but people tell me it does *shrug* */
 467         if (nn < 0 && (errno == EAGAIN || errno == EINTR))
 468                 return 0;
 469
 470         if (nn <= 0) {
 471                 goto err_disconnected;
 472         }
 473
 474 #ifndef HAVE_READV
 475         if (n == 2 && nn == iov[0].iov_len) {
 476             int ret;
 477             ret = slirp_send(so, iov[1].iov_base, iov[1].iov_len,0);
 478             if (ret > 0)
 479                 nn += ret;
 480         }
 481         DEBUG_MISC((dfd, "  ... wrote nn = %d bytes\n", nn));
 482 #endif
 483
 484         /* Update sbuf */
 485         sb->sb_cc -= nn;
 486         sb->sb_rptr += nn;
 487         if (sb->sb_rptr >= (sb->sb_data + sb->sb_datalen))
 488                 sb->sb_rptr -= sb->sb_datalen;
 489
 490         /*
 491          * If in DRAIN mode, and there's no more data, set
 492          * it CANTSENDMORE
 493          */
 494         if ((so->so_state & SS_FWDRAIN) && sb->sb_cc == 0)
 495                 sofcantsendmore(so);
 496
 497         return nn;
 498
 499 err_disconnected:
 500         DEBUG_MISC((dfd, " --- sowrite disconnected, so->so_state = %x, errno = %d\n",
 501                     so->so_state, errno));
 502         sofcantsendmore(so);
 503         tcp_sockclosed(sototcpcb(so));
 504         return -1;
 505 }
 506
 507 /*
 508  * recvfrom() a UDP socket
 509  */
 510 void
 511 sorecvfrom(struct socket *so)
 512 {
 513         struct sockaddr_storage addr;
 514         struct sockaddr_storage saddr, daddr;
 515         socklen_t addrlen = sizeof(struct sockaddr_storage);
 516
 517         DEBUG_CALL("sorecvfrom");
 518         DEBUG_ARG("so = %p", so);
 519
 520         if (so->so_type == IPPROTO_ICMP) {   /* This is a "ping" reply */
 521           char buff[256];
 522           int len;
 523
 524           len = recvfrom(so->s, buff, 256, 0,
 525                          (struct sockaddr *)&addr, &addrlen);
 526           /* XXX Check if reply is "correct"? */
 527
 528           if(len == -1 || len == 0) {
 529             u_char code=ICMP_UNREACH_PORT;
 530
 531             if(errno == EHOSTUNREACH) code=ICMP_UNREACH_HOST;
 532             else if(errno == ENETUNREACH) code=ICMP_UNREACH_NET;
 533
 534             DEBUG_MISC((dfd," udp icmp rx errno = %d-%s\n",
 535                         errno,strerror(errno)));
 536             icmp_send_error(so->so_m, ICMP_UNREACH, code, 0, strerror(errno));
 537           } else {
 538             icmp_reflect(so->so_m);
 539             so->so_m = NULL; /* Don't m_free() it again! */
 540           }
 541           /* No need for this socket anymore, udp_detach it */
 542           udp_detach(so);
 543         } else {                                /* A "normal" UDP packet */
 544           struct mbuf *m;
 545           int len;
 546 #ifdef _WIN32
 547           unsigned long n;
 548 #else
 549           int n;
 550 #endif
 551
 552           m = m_get(so->slirp);
 553           if (!m) {
 554               return;
 555           }
 556           switch (so->so_ffamily) {
 557           case AF_INET:
 558               m->m_data += IF_MAXLINKHDR + sizeof(struct udpiphdr);
 559               break;
 560           case AF_INET6:
 561               m->m_data += IF_MAXLINKHDR + sizeof(struct ip6)
 562                                          + sizeof(struct udphdr);
 563               break;
 564           default:
 565               g_assert_not_reached();
 566               break;
 567           }
 568
 569           /*
 570            * XXX Shouldn't FIONREAD packets destined for port 53,
 571            * but I don't know the max packet size for DNS lookups
 572            */
 573           len = M_FREEROOM(m);
 574           /* if (so->so_fport != htons(53)) { */
 575           ioctlsocket(so->s, FIONREAD, &n);
 576
 577           if (n > len) {
 578             n = (m->m_data - m->m_dat) + m->m_len + n + 1;
 579             m_inc(m, n);
 580             len = M_FREEROOM(m);
 581           }
 582           /* } */
 583
 584           m->m_len = recvfrom(so->s, m->m_data, len, 0,
 585                               (struct sockaddr *)&addr, &addrlen);
 586           DEBUG_MISC((dfd, " did recvfrom %d, errno = %d-%s\n",
 587                       m->m_len, errno,strerror(errno)));
 588           if(m->m_len<0) {
 589             /* Report error as ICMP */
 590             switch (so->so_lfamily) {
 591             uint8_t code;
 592             case AF_INET:
 593               code = ICMP_UNREACH_PORT;
 594
 595               if (errno == EHOSTUNREACH) {
 596                 code = ICMP_UNREACH_HOST;
 597               } else if (errno == ENETUNREACH) {
 598                 code = ICMP_UNREACH_NET;
 599               }
 600
 601               DEBUG_MISC((dfd, " rx error, tx icmp ICMP_UNREACH:%i\n", code));
 602               icmp_send_error(so->so_m, ICMP_UNREACH, code, 0, strerror(errno));
 603               break;
 604             case AF_INET6:
 605               code = ICMP6_UNREACH_PORT;
 606
 607               if (errno == EHOSTUNREACH) {
 608                 code = ICMP6_UNREACH_ADDRESS;
 609               } else if (errno == ENETUNREACH) {
 610                 code = ICMP6_UNREACH_NO_ROUTE;
 611               }
 612
 613               DEBUG_MISC((dfd, " rx error, tx icmp6 ICMP_UNREACH:%i\n", code));
 614               icmp6_send_error(so->so_m, ICMP6_UNREACH, code);
 615               break;
 616             default:
 617               g_assert_not_reached();
 618               break;
 619             }
 620             m_free(m);
 621           } else {
 622           /*
 623            * Hack: domain name lookup will be used the most for UDP,
 624            * and since they'll only be used once there's no need
 625            * for the 4 minute (or whatever) timeout... So we time them
 626            * out much quicker (10 seconds  for now...)
 627            */
 628             if (so->so_expire) {
 629               if (so->so_fport == htons(53))
 630                 so->so_expire = curtime + SO_EXPIREFAST;
 631               else
 632                 so->so_expire = curtime + SO_EXPIRE;
 633             }
 634
 635             /*
 636              * If this packet was destined for CTL_ADDR,
 637              * make it look like that's where it came from
 638              */
 639             saddr = addr;
 640             sotranslate_in(so, &saddr);
 641             daddr = so->lhost.ss;
 642
 643             switch (so->so_ffamily) {
 644             case AF_INET:
 645                 udp_output(so, m, (struct sockaddr_in *) &saddr,
 646                            (struct sockaddr_in *) &daddr,
 647                            so->so_iptos);
 648                 break;
 649             case AF_INET6:
 650                 udp6_output(so, m, (struct sockaddr_in6 *) &saddr,
 651                             (struct sockaddr_in6 *) &daddr);
 652                 break;
 653             default:
 654                 g_assert_not_reached();
 655                 break;
 656             }
 657           } /* rx error */
 658         } /* if ping packet */
 659 }
 660
 661 /*
 662  * sendto() a socket
 663  */
 664 int
 665 sosendto(struct socket *so, struct mbuf *m)
 666 {
 667         int ret;
 668         struct sockaddr_storage addr;
 669
 670         DEBUG_CALL("sosendto");
 671         DEBUG_ARG("so = %p", so);
 672         DEBUG_ARG("m = %p", m);
 673
 674         addr = so->fhost.ss;
 675         DEBUG_CALL(" sendto()ing)");
 676         sotranslate_out(so, &addr);
 677
 678         /* Don't care what port we get */
 679         ret = sendto(so->s, m->m_data, m->m_len, 0,
 680                      (struct sockaddr *)&addr, sockaddr_size(&addr));
 681         if (ret < 0)
 682                 return -1;
 683
 684         /*
 685          * Kill the socket if there's no reply in 4 minutes,
 686          * but only if it's an expirable socket
 687          */
 688         if (so->so_expire)
 689                 so->so_expire = curtime + SO_EXPIRE;
 690         so->so_state &= SS_PERSISTENT_MASK;
 691         so->so_state |= SS_ISFCONNECTED; /* So that it gets select()ed */
 692         return 0;
 693 }
 694
 695 /*
 696  * Listen for incoming TCP connections
 697  */
 698 struct socket *
 699 tcp_listen(Slirp *slirp, uint32_t haddr, u_int hport, uint32_t laddr,
 700            u_int lport, int flags)
 701 {
 702         struct sockaddr_in addr;
 703         struct socket *so;
 704         int s, opt = 1;
 705         socklen_t addrlen = sizeof(addr);
 706         memset(&addr, 0, addrlen);
 707
 708         DEBUG_CALL("tcp_listen");
 709         DEBUG_ARG("haddr = %s", inet_ntoa((struct in_addr){.s_addr = haddr}));
 710         DEBUG_ARG("hport = %d", ntohs(hport));
 711         DEBUG_ARG("laddr = %s", inet_ntoa((struct in_addr){.s_addr = laddr}));
 712         DEBUG_ARG("lport = %d", ntohs(lport));
 713         DEBUG_ARG("flags = %x", flags);
 714
 715         so = socreate(slirp);
 716
 717         /* Don't tcp_attach... we don't need so_snd nor so_rcv */
 718         if ((so->so_tcpcb = tcp_newtcpcb(so)) == NULL) {
 719             g_free(so);
 720             return NULL;
 721         }
 722         insque(so, &slirp->tcb);
 723
 724         /*
 725          * SS_FACCEPTONCE sockets must time out.
 726          */
 727         if (flags & SS_FACCEPTONCE)
 728            so->so_tcpcb->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT*2;
 729
 730         so->so_state &= SS_PERSISTENT_MASK;
 731         so->so_state |= (SS_FACCEPTCONN | flags);
 732         so->so_lfamily = AF_INET;
 733         so->so_lport = lport; /* Kept in network format */
 734         so->so_laddr.s_addr = laddr; /* Ditto */
 735
 736         addr.sin_family = AF_INET;
 737         addr.sin_addr.s_addr = haddr;
 738         addr.sin_port = hport;
 739
 740         if (((s = qemu_socket(AF_INET,SOCK_STREAM,0)) < 0) ||
 741             (socket_set_fast_reuse(s) < 0) ||
 742             (bind(s,(struct sockaddr *)&addr, sizeof(addr)) < 0) ||
 743             (listen(s,1) < 0)) {
 744                 int tmperrno = errno; /* Don't clobber the real reason we failed */
 745
 746                 if (s >= 0) {
 747                     closesocket(s);
 748                 }
 749                 sofree(so);
 750                 /* Restore the real errno */
 751 #ifdef _WIN32
 752                 WSASetLastError(tmperrno);
 753 #else
 754                 errno = tmperrno;
 755 #endif
 756                 return NULL;
 757         }
 758         qemu_setsockopt(s, SOL_SOCKET, SO_OOBINLINE, &opt, sizeof(int));
 759         opt = 1;
 760         qemu_setsockopt(s, IPPROTO_TCP, TCP_NODELAY, &opt, sizeof(int));
 761
 762         getsockname(s,(struct sockaddr *)&addr,&addrlen);
 763         so->so_ffamily = AF_INET;
 764         so->so_fport = addr.sin_port;
 765         if (addr.sin_addr.s_addr == 0 || addr.sin_addr.s_addr == loopback_addr.s_addr)
 766            so->so_faddr = slirp->vhost_addr;
 767         else
 768            so->so_faddr = addr.sin_addr;
 769
 770         so->s = s;
 771         return so;
 772 }
 773
 774 /*
 775  * Various session state calls
 776  * XXX Should be #define's
 777  * The socket state stuff needs work, these often get call 2 or 3
 778  * times each when only 1 was needed
 779  */
 780 void
 781 soisfconnecting(struct socket *so)
 782 {
 783         so->so_state &= ~(SS_NOFDREF|SS_ISFCONNECTED|SS_FCANTRCVMORE|
 784                           SS_FCANTSENDMORE|SS_FWDRAIN);
 785         so->so_state |= SS_ISFCONNECTING; /* Clobber other states */
 786 }
 787
 788 void
 789 soisfconnected(struct socket *so)
 790 {
 791         so->so_state &= ~(SS_ISFCONNECTING|SS_FWDRAIN|SS_NOFDREF);
 792         so->so_state |= SS_ISFCONNECTED; /* Clobber other states */
 793 }
 794
 795 static void
 796 sofcantrcvmore(struct socket *so)
 797 {
 798         if ((so->so_state & SS_NOFDREF) == 0) {
 799                 shutdown(so->s,0);
 800         }
 801         so->so_state &= ~(SS_ISFCONNECTING);
 802         if (so->so_state & SS_FCANTSENDMORE) {
 803            so->so_state &= SS_PERSISTENT_MASK;
 804            so->so_state |= SS_NOFDREF; /* Don't select it */
 805         } else {
 806            so->so_state |= SS_FCANTRCVMORE;
 807         }
 808 }
 809
 810 static void
 811 sofcantsendmore(struct socket *so)
 812 {
 813         if ((so->so_state & SS_NOFDREF) == 0) {
 814             shutdown(so->s,1);           /* send FIN to fhost */
 815         }
 816         so->so_state &= ~(SS_ISFCONNECTING);
 817         if (so->so_state & SS_FCANTRCVMORE) {
 818            so->so_state &= SS_PERSISTENT_MASK;
 819            so->so_state |= SS_NOFDREF; /* as above */
 820         } else {
 821            so->so_state |= SS_FCANTSENDMORE;
 822         }
 823 }
 824
 825 /*
 826  * Set write drain mode
 827  * Set CANTSENDMORE once all data has been write()n
 828  */
 829 void
 830 sofwdrain(struct socket *so)
 831 {
 832         if (so->so_rcv.sb_cc)
 833                 so->so_state |= SS_FWDRAIN;
 834         else
 835                 sofcantsendmore(so);
 836 }
 837
 838 /*
 839  * Translate addr in host addr when it is a virtual address
 840  */
 841 void sotranslate_out(struct socket *so, struct sockaddr_storage *addr)
 842 {
 843     Slirp *slirp = so->slirp;
 844     struct sockaddr_in *sin = (struct sockaddr_in *)addr;
 845     struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)addr;
 846
 847     switch (addr->ss_family) {
 848     case AF_INET:
 849         if ((so->so_faddr.s_addr & slirp->vnetwork_mask.s_addr) ==
 850                 slirp->vnetwork_addr.s_addr) {
 851             /* It's an alias */
 852             if (so->so_faddr.s_addr == slirp->vnameserver_addr.s_addr) {
 853                 if (get_dns_addr(&sin->sin_addr) < 0) {
 854                     sin->sin_addr = loopback_addr;
 855                 }
 856             } else {
 857                 sin->sin_addr = loopback_addr;
 858             }
 859         }
 860
 861         DEBUG_MISC((dfd, " addr.sin_port=%d, "
 862             "addr.sin_addr.s_addr=%.16s\n",
 863             ntohs(sin->sin_port), inet_ntoa(sin->sin_addr)));
 864         break;
 865
 866     case AF_INET6:
 867         if (in6_equal_net(&so->so_faddr6, &slirp->vprefix_addr6,
 868                     slirp->vprefix_len)) {
 869             if (in6_equal(&so->so_faddr6, &slirp->vnameserver_addr6)) {
 870                 uint32_t scope_id;
 871                 if (get_dns6_addr(&sin6->sin6_addr, &scope_id) >= 0) {
 872                     sin6->sin6_scope_id = scope_id;
 873                 } else {
 874                     sin6->sin6_addr = in6addr_loopback;
 875                 }
 876             } else {
 877                 sin6->sin6_addr = in6addr_loopback;
 878             }
 879         }
 880         break;
 881
 882     default:
 883         break;
 884     }
 885 }
 886
 887 void sotranslate_in(struct socket *so, struct sockaddr_storage *addr)
 888 {
 889     Slirp *slirp = so->slirp;
 890     struct sockaddr_in *sin = (struct sockaddr_in *)addr;
 891     struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)addr;
 892
 893     switch (addr->ss_family) {
 894     case AF_INET:
 895         if ((so->so_faddr.s_addr & slirp->vnetwork_mask.s_addr) ==
 896             slirp->vnetwork_addr.s_addr) {
 897             uint32_t inv_mask = ~slirp->vnetwork_mask.s_addr;
 898
 899             if ((so->so_faddr.s_addr & inv_mask) == inv_mask) {
 900                 sin->sin_addr = slirp->vhost_addr;
 901             } else if (sin->sin_addr.s_addr == loopback_addr.s_addr ||
 902                        so->so_faddr.s_addr != slirp->vhost_addr.s_addr) {
 903                 sin->sin_addr = so->so_faddr;
 904             }
 905         }
 906         break;
 907
 908     case AF_INET6:
 909         if (in6_equal_net(&so->so_faddr6, &slirp->vprefix_addr6,
 910                     slirp->vprefix_len)) {
 911             if (in6_equal(&sin6->sin6_addr, &in6addr_loopback)
 912                     || !in6_equal(&so->so_faddr6, &slirp->vhost_addr6)) {
 913                 sin6->sin6_addr = so->so_faddr6;
 914             }
 915         }
 916         break;
 917
 918     default:
 919         break;
 920     }
 921 }
 922
 923 /*
 924  * Translate connections from localhost to the real hostname
 925  */
 926 void sotranslate_accept(struct socket *so)
 927 {
 928     Slirp *slirp = so->slirp;
 929
 930     switch (so->so_ffamily) {
 931     case AF_INET:
 932         if (so->so_faddr.s_addr == INADDR_ANY ||
 933             (so->so_faddr.s_addr & loopback_mask) ==
 934             (loopback_addr.s_addr & loopback_mask)) {
 935            so->so_faddr = slirp->vhost_addr;
 936         }
 937         break;
 938
 939    case AF_INET6:
 940         if (in6_equal(&so->so_faddr6, &in6addr_any) ||
 941                 in6_equal(&so->so_faddr6, &in6addr_loopback)) {
 942            so->so_faddr6 = slirp->vhost_addr6;
 943         }
 944         break;
 945
 946     default:
 947         break;
 948     }
 949 }