2 * Copyright (c) 1995 Danny Gasparovski.
4 * Please read the file COPYRIGHT for the
5 * terms and conditions of the copyright.
11 #include <sys/filio.h>
14 static void sofcantrcvmore(struct socket *so);
15 static void sofcantsendmore(struct socket *so);
17 struct socket *solookup(struct socket **last, struct socket *head,
18 struct sockaddr_storage *lhost, struct sockaddr_storage *fhost)
20 struct socket *so = *last;
23 if (so != head && sockaddr_equal(&(so->lhost.ss), lhost)
24 && (!fhost || sockaddr_equal(&so->fhost.ss, fhost))) {
28 for (so = head->so_next; so != head; so = so->so_next) {
29 if (sockaddr_equal(&(so->lhost.ss), lhost)
30 && (!fhost || sockaddr_equal(&so->fhost.ss, fhost))) {
36 return (struct socket *)NULL;
40 * Create a new socket, initialise the fields
41 * It is the responsibility of the caller to
42 * insque() it into the correct linked-list
45 socreate(Slirp *slirp)
47 struct socket *so = g_new(struct socket, 1);
49 memset(so, 0, sizeof(struct socket));
50 so->so_state = SS_NOFDREF;
59 * Remove references to so from the given message queue.
62 soqfree(struct socket *so, struct quehead *qh)
66 for (ifq = (struct mbuf *) qh->qh_link;
67 (struct quehead *) ifq != qh;
68 ifq = ifq->ifq_next) {
69 if (ifq->ifq_so == so) {
72 for (ifm = ifq->ifs_next; ifm != ifq; ifm = ifm->ifs_next) {
80 * remque and free a socket, clobber cache
83 sofree(struct socket *so)
85 Slirp *slirp = so->slirp;
87 soqfree(so, &slirp->if_fastq);
88 soqfree(so, &slirp->if_batchq);
90 if (so == slirp->tcp_last_so) {
91 slirp->tcp_last_so = &slirp->tcb;
92 } else if (so == slirp->udp_last_so) {
93 slirp->udp_last_so = &slirp->udb;
94 } else if (so == slirp->icmp_last_so) {
95 slirp->icmp_last_so = &slirp->icmp;
99 if(so->so_next && so->so_prev)
100 remque(so); /* crashes if so is not in a queue */
108 size_t sopreprbuf(struct socket *so, struct iovec *iov, int *np)
111 struct sbuf *sb = &so->so_snd;
112 int len = sb->sb_datalen - sb->sb_cc;
113 int mss = so->so_tcpcb->t_maxseg;
115 DEBUG_CALL("sopreprbuf");
116 DEBUG_ARG("so = %p", so);
121 iov[0].iov_base = sb->sb_wptr;
122 iov[1].iov_base = NULL;
124 if (sb->sb_wptr < sb->sb_rptr) {
125 iov[0].iov_len = sb->sb_rptr - sb->sb_wptr;
126 /* Should never succeed, but... */
127 if (iov[0].iov_len > len)
128 iov[0].iov_len = len;
129 if (iov[0].iov_len > mss)
130 iov[0].iov_len -= iov[0].iov_len%mss;
133 iov[0].iov_len = (sb->sb_data + sb->sb_datalen) - sb->sb_wptr;
134 /* Should never succeed, but... */
135 if (iov[0].iov_len > len) iov[0].iov_len = len;
136 len -= iov[0].iov_len;
138 iov[1].iov_base = sb->sb_data;
139 iov[1].iov_len = sb->sb_rptr - sb->sb_data;
140 if(iov[1].iov_len > len)
141 iov[1].iov_len = len;
142 total = iov[0].iov_len + iov[1].iov_len;
145 if (iov[1].iov_len > lss) {
146 iov[1].iov_len -= lss;
149 lss -= iov[1].iov_len;
150 iov[0].iov_len -= lss;
156 if (iov[0].iov_len > mss)
157 iov[0].iov_len -= iov[0].iov_len%mss;
164 return iov[0].iov_len + (n - 1) * iov[1].iov_len;
168 * Read from so's socket into sb_snd, updating all relevant sbuf fields
169 * NOTE: This will only be called if it is select()ed for reading, so
170 * a read() of 0 (or less) means it's disconnected
173 soread(struct socket *so)
176 struct sbuf *sb = &so->so_snd;
179 DEBUG_CALL("soread");
180 DEBUG_ARG("so = %p", so);
183 * No need to check if there's enough room to read.
184 * soread wouldn't have been called if there weren't
186 sopreprbuf(so, iov, &n);
188 nn = recv(so->s, iov[0].iov_base, iov[0].iov_len,0);
190 if (nn < 0 && (errno == EINTR || errno == EAGAIN))
194 socklen_t elen = sizeof err;
195 struct sockaddr_storage addr;
196 struct sockaddr *paddr = (struct sockaddr *) &addr;
197 socklen_t alen = sizeof addr;
201 if (getpeername(so->s, paddr, &alen) < 0) {
204 getsockopt(so->s, SOL_SOCKET, SO_ERROR,
209 DEBUG_MISC(" --- soread() disconnected, nn = %d, errno = %d-%s",
210 nn, errno,strerror(errno));
213 if (err == ECONNRESET || err == ECONNREFUSED
214 || err == ENOTCONN || err == EPIPE) {
215 tcp_drop(sototcpcb(so), err);
217 tcp_sockclosed(sototcpcb(so));
224 * If there was no error, try and read the second time round
225 * We read again if n = 2 (ie, there's another part of the buffer)
226 * and we read as much as we could in the first read
227 * We don't test for <= 0 this time, because there legitimately
228 * might not be any more data (since the socket is non-blocking),
229 * a close will be detected on next iteration.
230 * A return of -1 won't (shouldn't) happen, since it didn't happen above
232 if (n == 2 && nn == iov[0].iov_len) {
234 ret = recv(so->s, iov[1].iov_base, iov[1].iov_len,0);
239 DEBUG_MISC(" ... read nn = %d bytes", nn);
244 if (sb->sb_wptr >= (sb->sb_data + sb->sb_datalen))
245 sb->sb_wptr -= sb->sb_datalen;
249 int soreadbuf(struct socket *so, const char *buf, int size)
251 int n, nn, copy = size;
252 struct sbuf *sb = &so->so_snd;
255 DEBUG_CALL("soreadbuf");
256 DEBUG_ARG("so = %p", so);
259 * No need to check if there's enough room to read.
260 * soread wouldn't have been called if there weren't
262 if (sopreprbuf(so, iov, &n) < size)
265 nn = MIN(iov[0].iov_len, copy);
266 memcpy(iov[0].iov_base, buf, nn);
274 memcpy(iov[1].iov_base, buf, copy);
280 if (sb->sb_wptr >= (sb->sb_data + sb->sb_datalen))
281 sb->sb_wptr -= sb->sb_datalen;
286 tcp_sockclosed(sototcpcb(so));
287 g_critical("soreadbuf buffer too small");
294 * When the socket is created, we set it SO_OOBINLINE,
295 * so when OOB data arrives, we soread() it and everything
296 * in the send buffer is sent as urgent data
299 sorecvoob(struct socket *so)
301 struct tcpcb *tp = sototcpcb(so);
304 DEBUG_CALL("sorecvoob");
305 DEBUG_ARG("so = %p", so);
308 * We take a guess at how much urgent data has arrived.
309 * In most situations, when urgent data arrives, the next
310 * read() should get all the urgent data. This guess will
311 * be wrong however if more data arrives just after the
312 * urgent data, or the read() doesn't return all the
317 tp->snd_up = tp->snd_una + so->so_snd.sb_cc;
328 * There's a lot duplicated code here, but...
331 sosendoob(struct socket *so)
333 struct sbuf *sb = &so->so_rcv;
334 char buff[2048]; /* XXX Shouldn't be sending more oob data than this */
338 DEBUG_CALL("sosendoob");
339 DEBUG_ARG("so = %p", so);
340 DEBUG_ARG("sb->sb_cc = %d", sb->sb_cc);
342 if (so->so_urgc > 2048)
343 so->so_urgc = 2048; /* XXXX */
345 if (sb->sb_rptr < sb->sb_wptr) {
346 /* We can send it directly */
347 n = slirp_send(so, sb->sb_rptr, so->so_urgc, (MSG_OOB)); /* |MSG_DONTWAIT)); */
350 * Since there's no sendv or sendtov like writev,
351 * we must copy all data to a linear buffer then
354 uint32_t urgc = so->so_urgc;
355 int len = (sb->sb_data + sb->sb_datalen) - sb->sb_rptr;
359 memcpy(buff, sb->sb_rptr, len);
362 n = sb->sb_wptr - sb->sb_data;
366 memcpy((buff + len), sb->sb_data, n);
369 n = slirp_send(so, buff, len, (MSG_OOB)); /* |MSG_DONTWAIT)); */
372 DEBUG_ERROR("Didn't send all data urgently XXXXX");
381 DEBUG_MISC(" ---2 sent %d bytes urgent data, %d urgent bytes left", n, so->so_urgc);
385 if (sb->sb_rptr >= (sb->sb_data + sb->sb_datalen))
386 sb->sb_rptr -= sb->sb_datalen;
392 * Write data from so_rcv to so's socket,
393 * updating all sbuf field as necessary
396 sowrite(struct socket *so)
399 struct sbuf *sb = &so->so_rcv;
403 DEBUG_CALL("sowrite");
404 DEBUG_ARG("so = %p", so);
407 uint32_t expected = so->so_urgc;
408 if (sosendoob(so) < expected) {
409 /* Treat a short write as a fatal error too,
410 * rather than continuing on and sending the urgent
411 * data as if it were non-urgent and leaving the
412 * so_urgc count wrong.
414 goto err_disconnected;
421 * No need to check if there's something to write,
422 * sowrite wouldn't have been called otherwise
425 iov[0].iov_base = sb->sb_rptr;
426 iov[1].iov_base = NULL;
428 if (sb->sb_rptr < sb->sb_wptr) {
429 iov[0].iov_len = sb->sb_wptr - sb->sb_rptr;
430 /* Should never succeed, but... */
431 if (iov[0].iov_len > len) iov[0].iov_len = len;
434 iov[0].iov_len = (sb->sb_data + sb->sb_datalen) - sb->sb_rptr;
435 if (iov[0].iov_len > len) iov[0].iov_len = len;
436 len -= iov[0].iov_len;
438 iov[1].iov_base = sb->sb_data;
439 iov[1].iov_len = sb->sb_wptr - sb->sb_data;
440 if (iov[1].iov_len > len) iov[1].iov_len = len;
445 /* Check if there's urgent data to send, and if so, send it */
447 nn = slirp_send(so, iov[0].iov_base, iov[0].iov_len,0);
448 /* This should never happen, but people tell me it does *shrug* */
449 if (nn < 0 && (errno == EAGAIN || errno == EINTR))
453 goto err_disconnected;
456 if (n == 2 && nn == iov[0].iov_len) {
458 ret = slirp_send(so, iov[1].iov_base, iov[1].iov_len,0);
462 DEBUG_MISC(" ... wrote nn = %d bytes", nn);
467 if (sb->sb_rptr >= (sb->sb_data + sb->sb_datalen))
468 sb->sb_rptr -= sb->sb_datalen;
471 * If in DRAIN mode, and there's no more data, set
474 if ((so->so_state & SS_FWDRAIN) && sb->sb_cc == 0)
480 DEBUG_MISC(" --- sowrite disconnected, so->so_state = %x, errno = %d",
481 so->so_state, errno);
483 tcp_sockclosed(sototcpcb(so));
488 * recvfrom() a UDP socket
491 sorecvfrom(struct socket *so)
493 struct sockaddr_storage addr;
494 struct sockaddr_storage saddr, daddr;
495 socklen_t addrlen = sizeof(struct sockaddr_storage);
497 DEBUG_CALL("sorecvfrom");
498 DEBUG_ARG("so = %p", so);
500 if (so->so_type == IPPROTO_ICMP) { /* This is a "ping" reply */
504 len = recvfrom(so->s, buff, 256, 0,
505 (struct sockaddr *)&addr, &addrlen);
506 /* XXX Check if reply is "correct"? */
508 if(len == -1 || len == 0) {
509 uint8_t code=ICMP_UNREACH_PORT;
511 if(errno == EHOSTUNREACH) code=ICMP_UNREACH_HOST;
512 else if(errno == ENETUNREACH) code=ICMP_UNREACH_NET;
514 DEBUG_MISC(" udp icmp rx errno = %d-%s",
515 errno,strerror(errno));
516 icmp_send_error(so->so_m, ICMP_UNREACH, code, 0, strerror(errno));
518 icmp_reflect(so->so_m);
519 so->so_m = NULL; /* Don't m_free() it again! */
521 /* No need for this socket anymore, udp_detach it */
523 } else { /* A "normal" UDP packet */
532 m = m_get(so->slirp);
536 switch (so->so_ffamily) {
538 m->m_data += IF_MAXLINKHDR + sizeof(struct udpiphdr);
541 m->m_data += IF_MAXLINKHDR + sizeof(struct ip6)
542 + sizeof(struct udphdr);
545 g_assert_not_reached();
550 * XXX Shouldn't FIONREAD packets destined for port 53,
551 * but I don't know the max packet size for DNS lookups
554 /* if (so->so_fport != htons(53)) { */
555 ioctlsocket(so->s, FIONREAD, &n);
558 n = (m->m_data - m->m_dat) + m->m_len + n + 1;
564 m->m_len = recvfrom(so->s, m->m_data, len, 0,
565 (struct sockaddr *)&addr, &addrlen);
566 DEBUG_MISC(" did recvfrom %d, errno = %d-%s",
567 m->m_len, errno,strerror(errno));
569 /* Report error as ICMP */
570 switch (so->so_lfamily) {
573 code = ICMP_UNREACH_PORT;
575 if (errno == EHOSTUNREACH) {
576 code = ICMP_UNREACH_HOST;
577 } else if (errno == ENETUNREACH) {
578 code = ICMP_UNREACH_NET;
581 DEBUG_MISC(" rx error, tx icmp ICMP_UNREACH:%i", code);
582 icmp_send_error(so->so_m, ICMP_UNREACH, code, 0, strerror(errno));
585 code = ICMP6_UNREACH_PORT;
587 if (errno == EHOSTUNREACH) {
588 code = ICMP6_UNREACH_ADDRESS;
589 } else if (errno == ENETUNREACH) {
590 code = ICMP6_UNREACH_NO_ROUTE;
593 DEBUG_MISC(" rx error, tx icmp6 ICMP_UNREACH:%i", code);
594 icmp6_send_error(so->so_m, ICMP6_UNREACH, code);
597 g_assert_not_reached();
603 * Hack: domain name lookup will be used the most for UDP,
604 * and since they'll only be used once there's no need
605 * for the 4 minute (or whatever) timeout... So we time them
606 * out much quicker (10 seconds for now...)
609 if (so->so_fport == htons(53))
610 so->so_expire = curtime + SO_EXPIREFAST;
612 so->so_expire = curtime + SO_EXPIRE;
616 * If this packet was destined for CTL_ADDR,
617 * make it look like that's where it came from
620 sotranslate_in(so, &saddr);
621 daddr = so->lhost.ss;
623 switch (so->so_ffamily) {
625 udp_output(so, m, (struct sockaddr_in *) &saddr,
626 (struct sockaddr_in *) &daddr,
630 udp6_output(so, m, (struct sockaddr_in6 *) &saddr,
631 (struct sockaddr_in6 *) &daddr);
634 g_assert_not_reached();
638 } /* if ping packet */
645 sosendto(struct socket *so, struct mbuf *m)
648 struct sockaddr_storage addr;
650 DEBUG_CALL("sosendto");
651 DEBUG_ARG("so = %p", so);
652 DEBUG_ARG("m = %p", m);
655 DEBUG_CALL(" sendto()ing)");
656 sotranslate_out(so, &addr);
658 /* Don't care what port we get */
659 ret = sendto(so->s, m->m_data, m->m_len, 0,
660 (struct sockaddr *)&addr, sockaddr_size(&addr));
665 * Kill the socket if there's no reply in 4 minutes,
666 * but only if it's an expirable socket
669 so->so_expire = curtime + SO_EXPIRE;
670 so->so_state &= SS_PERSISTENT_MASK;
671 so->so_state |= SS_ISFCONNECTED; /* So that it gets select()ed */
676 * Listen for incoming TCP connections
679 tcp_listen(Slirp *slirp, uint32_t haddr, unsigned hport, uint32_t laddr,
680 unsigned lport, int flags)
682 struct sockaddr_in addr;
685 socklen_t addrlen = sizeof(addr);
686 memset(&addr, 0, addrlen);
688 DEBUG_CALL("tcp_listen");
689 DEBUG_ARG("haddr = %s", inet_ntoa((struct in_addr){.s_addr = haddr}));
690 DEBUG_ARG("hport = %d", ntohs(hport));
691 DEBUG_ARG("laddr = %s", inet_ntoa((struct in_addr){.s_addr = laddr}));
692 DEBUG_ARG("lport = %d", ntohs(lport));
693 DEBUG_ARG("flags = %x", flags);
695 so = socreate(slirp);
697 /* Don't tcp_attach... we don't need so_snd nor so_rcv */
698 if ((so->so_tcpcb = tcp_newtcpcb(so)) == NULL) {
702 insque(so, &slirp->tcb);
705 * SS_FACCEPTONCE sockets must time out.
707 if (flags & SS_FACCEPTONCE)
708 so->so_tcpcb->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT*2;
710 so->so_state &= SS_PERSISTENT_MASK;
711 so->so_state |= (SS_FACCEPTCONN | flags);
712 so->so_lfamily = AF_INET;
713 so->so_lport = lport; /* Kept in network format */
714 so->so_laddr.s_addr = laddr; /* Ditto */
716 addr.sin_family = AF_INET;
717 addr.sin_addr.s_addr = haddr;
718 addr.sin_port = hport;
720 if (((s = slirp_socket(AF_INET,SOCK_STREAM,0)) < 0) ||
721 (slirp_socket_set_fast_reuse(s) < 0) ||
722 (bind(s,(struct sockaddr *)&addr, sizeof(addr)) < 0) ||
724 int tmperrno = errno; /* Don't clobber the real reason we failed */
730 /* Restore the real errno */
732 WSASetLastError(tmperrno);
738 setsockopt(s, SOL_SOCKET, SO_OOBINLINE, &opt, sizeof(int));
740 setsockopt(s, IPPROTO_TCP, TCP_NODELAY, &opt, sizeof(int));
742 getsockname(s,(struct sockaddr *)&addr,&addrlen);
743 so->so_ffamily = AF_INET;
744 so->so_fport = addr.sin_port;
745 if (addr.sin_addr.s_addr == 0 || addr.sin_addr.s_addr == loopback_addr.s_addr)
746 so->so_faddr = slirp->vhost_addr;
748 so->so_faddr = addr.sin_addr;
755 * Various session state calls
756 * XXX Should be #define's
757 * The socket state stuff needs work, these often get call 2 or 3
758 * times each when only 1 was needed
761 soisfconnecting(struct socket *so)
763 so->so_state &= ~(SS_NOFDREF|SS_ISFCONNECTED|SS_FCANTRCVMORE|
764 SS_FCANTSENDMORE|SS_FWDRAIN);
765 so->so_state |= SS_ISFCONNECTING; /* Clobber other states */
769 soisfconnected(struct socket *so)
771 so->so_state &= ~(SS_ISFCONNECTING|SS_FWDRAIN|SS_NOFDREF);
772 so->so_state |= SS_ISFCONNECTED; /* Clobber other states */
776 sofcantrcvmore(struct socket *so)
778 if ((so->so_state & SS_NOFDREF) == 0) {
781 so->so_state &= ~(SS_ISFCONNECTING);
782 if (so->so_state & SS_FCANTSENDMORE) {
783 so->so_state &= SS_PERSISTENT_MASK;
784 so->so_state |= SS_NOFDREF; /* Don't select it */
786 so->so_state |= SS_FCANTRCVMORE;
791 sofcantsendmore(struct socket *so)
793 if ((so->so_state & SS_NOFDREF) == 0) {
794 shutdown(so->s,1); /* send FIN to fhost */
796 so->so_state &= ~(SS_ISFCONNECTING);
797 if (so->so_state & SS_FCANTRCVMORE) {
798 so->so_state &= SS_PERSISTENT_MASK;
799 so->so_state |= SS_NOFDREF; /* as above */
801 so->so_state |= SS_FCANTSENDMORE;
806 * Set write drain mode
807 * Set CANTSENDMORE once all data has been write()n
810 sofwdrain(struct socket *so)
812 if (so->so_rcv.sb_cc)
813 so->so_state |= SS_FWDRAIN;
819 * Translate addr in host addr when it is a virtual address
821 void sotranslate_out(struct socket *so, struct sockaddr_storage *addr)
823 Slirp *slirp = so->slirp;
824 struct sockaddr_in *sin = (struct sockaddr_in *)addr;
825 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)addr;
827 switch (addr->ss_family) {
829 if ((so->so_faddr.s_addr & slirp->vnetwork_mask.s_addr) ==
830 slirp->vnetwork_addr.s_addr) {
832 if (so->so_faddr.s_addr == slirp->vnameserver_addr.s_addr) {
833 if (get_dns_addr(&sin->sin_addr) < 0) {
834 sin->sin_addr = loopback_addr;
837 sin->sin_addr = loopback_addr;
841 DEBUG_MISC(" addr.sin_port=%d, addr.sin_addr.s_addr=%.16s",
842 ntohs(sin->sin_port), inet_ntoa(sin->sin_addr));
846 if (in6_equal_net(&so->so_faddr6, &slirp->vprefix_addr6,
847 slirp->vprefix_len)) {
848 if (in6_equal(&so->so_faddr6, &slirp->vnameserver_addr6)) {
850 if (get_dns6_addr(&sin6->sin6_addr, &scope_id) >= 0) {
851 sin6->sin6_scope_id = scope_id;
853 sin6->sin6_addr = in6addr_loopback;
856 sin6->sin6_addr = in6addr_loopback;
866 void sotranslate_in(struct socket *so, struct sockaddr_storage *addr)
868 Slirp *slirp = so->slirp;
869 struct sockaddr_in *sin = (struct sockaddr_in *)addr;
870 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)addr;
872 switch (addr->ss_family) {
874 if ((so->so_faddr.s_addr & slirp->vnetwork_mask.s_addr) ==
875 slirp->vnetwork_addr.s_addr) {
876 uint32_t inv_mask = ~slirp->vnetwork_mask.s_addr;
878 if ((so->so_faddr.s_addr & inv_mask) == inv_mask) {
879 sin->sin_addr = slirp->vhost_addr;
880 } else if (sin->sin_addr.s_addr == loopback_addr.s_addr ||
881 so->so_faddr.s_addr != slirp->vhost_addr.s_addr) {
882 sin->sin_addr = so->so_faddr;
888 if (in6_equal_net(&so->so_faddr6, &slirp->vprefix_addr6,
889 slirp->vprefix_len)) {
890 if (in6_equal(&sin6->sin6_addr, &in6addr_loopback)
891 || !in6_equal(&so->so_faddr6, &slirp->vhost_addr6)) {
892 sin6->sin6_addr = so->so_faddr6;
903 * Translate connections from localhost to the real hostname
905 void sotranslate_accept(struct socket *so)
907 Slirp *slirp = so->slirp;
909 switch (so->so_ffamily) {
911 if (so->so_faddr.s_addr == INADDR_ANY ||
912 (so->so_faddr.s_addr & loopback_mask) ==
913 (loopback_addr.s_addr & loopback_mask)) {
914 so->so_faddr = slirp->vhost_addr;
919 if (in6_equal(&so->so_faddr6, &in6addr_any) ||
920 in6_equal(&so->so_faddr6, &in6addr_loopback)) {
921 so->so_faddr6 = slirp->vhost_addr6;
930 void sodrop(struct socket *s, int num)
932 if (sbdrop(&s->so_snd, num)) {
933 s->slirp->cb->notify(s->slirp->opaque);