2 * Copyright (c) 1995 Danny Gasparovski.
4 * Please read the file COPYRIGHT for the
5 * terms and conditions of the copyright.
8 #include "qemu-common.h"
9 #define WANT_SYS_IOCTL_H
13 #include <sys/filio.h>
16 static void sofcantrcvmore(struct socket *so);
17 static void sofcantsendmore(struct socket *so);
28 solookup(head, laddr, lport, faddr, fport)
37 for (so = head->so_next; so != head; so = so->so_next) {
38 if (so->so_lport == lport &&
39 so->so_laddr.s_addr == laddr.s_addr &&
40 so->so_faddr.s_addr == faddr.s_addr &&
41 so->so_fport == fport)
46 return (struct socket *)NULL;
52 * Create a new socket, initialise the fields
53 * It is the responsibility of the caller to
54 * insque() it into the correct linked-list
61 so = (struct socket *)malloc(sizeof(struct socket));
63 memset(so, 0, sizeof(struct socket));
64 so->so_state = SS_NOFDREF;
71 * remque and free a socket, clobber cache
77 if (so->so_emu==EMU_RSH && so->extra) {
81 if (so == tcp_last_so)
83 else if (so == udp_last_so)
88 if(so->so_next && so->so_prev)
89 remque(so); /* crashes if so is not in a queue */
94 size_t sopreprbuf(struct socket *so, struct iovec *iov, int *np)
97 struct sbuf *sb = &so->so_snd;
98 int len = sb->sb_datalen - sb->sb_cc;
99 int mss = so->so_tcpcb->t_maxseg;
101 DEBUG_CALL("sopreprbuf");
102 DEBUG_ARG("so = %lx", (long )so);
104 len = sb->sb_datalen - sb->sb_cc;
109 iov[0].iov_base = sb->sb_wptr;
110 iov[1].iov_base = NULL;
112 if (sb->sb_wptr < sb->sb_rptr) {
113 iov[0].iov_len = sb->sb_rptr - sb->sb_wptr;
114 /* Should never succeed, but... */
115 if (iov[0].iov_len > len)
116 iov[0].iov_len = len;
117 if (iov[0].iov_len > mss)
118 iov[0].iov_len -= iov[0].iov_len%mss;
121 iov[0].iov_len = (sb->sb_data + sb->sb_datalen) - sb->sb_wptr;
122 /* Should never succeed, but... */
123 if (iov[0].iov_len > len) iov[0].iov_len = len;
124 len -= iov[0].iov_len;
126 iov[1].iov_base = sb->sb_data;
127 iov[1].iov_len = sb->sb_rptr - sb->sb_data;
128 if(iov[1].iov_len > len)
129 iov[1].iov_len = len;
130 total = iov[0].iov_len + iov[1].iov_len;
133 if (iov[1].iov_len > lss) {
134 iov[1].iov_len -= lss;
137 lss -= iov[1].iov_len;
138 iov[0].iov_len -= lss;
144 if (iov[0].iov_len > mss)
145 iov[0].iov_len -= iov[0].iov_len%mss;
152 return iov[0].iov_len + (n - 1) * iov[1].iov_len;
156 * Read from so's socket into sb_snd, updating all relevant sbuf fields
157 * NOTE: This will only be called if it is select()ed for reading, so
158 * a read() of 0 (or less) means it's disconnected
165 struct sbuf *sb = &so->so_snd;
168 DEBUG_CALL("soread");
169 DEBUG_ARG("so = %lx", (long )so);
172 * No need to check if there's enough room to read.
173 * soread wouldn't have been called if there weren't
175 sopreprbuf(so, iov, &n);
178 nn = readv(so->s, (struct iovec *)iov, n);
179 DEBUG_MISC((dfd, " ... read nn = %d bytes\n", nn));
181 nn = recv(so->s, iov[0].iov_base, iov[0].iov_len,0);
184 if (nn < 0 && (errno == EINTR || errno == EAGAIN))
187 DEBUG_MISC((dfd, " --- soread() disconnected, nn = %d, errno = %d-%s\n", nn, errno,strerror(errno)));
189 tcp_sockclosed(sototcpcb(so));
196 * If there was no error, try and read the second time round
197 * We read again if n = 2 (ie, there's another part of the buffer)
198 * and we read as much as we could in the first read
199 * We don't test for <= 0 this time, because there legitimately
200 * might not be any more data (since the socket is non-blocking),
201 * a close will be detected on next iteration.
202 * A return of -1 wont (shouldn't) happen, since it didn't happen above
204 if (n == 2 && nn == iov[0].iov_len) {
206 ret = recv(so->s, iov[1].iov_base, iov[1].iov_len,0);
211 DEBUG_MISC((dfd, " ... read nn = %d bytes\n", nn));
217 if (sb->sb_wptr >= (sb->sb_data + sb->sb_datalen))
218 sb->sb_wptr -= sb->sb_datalen;
222 int soreadbuf(struct socket *so, const char *buf, int size)
224 int n, nn, copy = size;
225 struct sbuf *sb = &so->so_snd;
228 DEBUG_CALL("soreadbuf");
229 DEBUG_ARG("so = %lx", (long )so);
232 * No need to check if there's enough room to read.
233 * soread wouldn't have been called if there weren't
235 if (sopreprbuf(so, iov, &n) < size)
238 nn = MIN(iov[0].iov_len, copy);
239 memcpy(iov[0].iov_base, buf, nn);
247 memcpy(iov[1].iov_base, buf, copy);
253 if (sb->sb_wptr >= (sb->sb_data + sb->sb_datalen))
254 sb->sb_wptr -= sb->sb_datalen;
259 tcp_sockclosed(sototcpcb(so));
260 fprintf(stderr, "soreadbuf buffer to small");
267 * When the socket is created, we set it SO_OOBINLINE,
268 * so when OOB data arrives, we soread() it and everything
269 * in the send buffer is sent as urgent data
275 struct tcpcb *tp = sototcpcb(so);
277 DEBUG_CALL("sorecvoob");
278 DEBUG_ARG("so = %lx", (long)so);
281 * We take a guess at how much urgent data has arrived.
282 * In most situations, when urgent data arrives, the next
283 * read() should get all the urgent data. This guess will
284 * be wrong however if more data arrives just after the
285 * urgent data, or the read() doesn't return all the
289 tp->snd_up = tp->snd_una + so->so_snd.sb_cc;
297 * There's a lot duplicated code here, but...
303 struct sbuf *sb = &so->so_rcv;
304 char buff[2048]; /* XXX Shouldn't be sending more oob data than this */
308 DEBUG_CALL("sosendoob");
309 DEBUG_ARG("so = %lx", (long)so);
310 DEBUG_ARG("sb->sb_cc = %d", sb->sb_cc);
312 if (so->so_urgc > 2048)
313 so->so_urgc = 2048; /* XXXX */
315 if (sb->sb_rptr < sb->sb_wptr) {
316 /* We can send it directly */
317 n = slirp_send(so, sb->sb_rptr, so->so_urgc, (MSG_OOB)); /* |MSG_DONTWAIT)); */
320 DEBUG_MISC((dfd, " --- sent %d bytes urgent data, %d urgent bytes left\n", n, so->so_urgc));
323 * Since there's no sendv or sendtov like writev,
324 * we must copy all data to a linear buffer then
327 len = (sb->sb_data + sb->sb_datalen) - sb->sb_rptr;
328 if (len > so->so_urgc) len = so->so_urgc;
329 memcpy(buff, sb->sb_rptr, len);
332 n = sb->sb_wptr - sb->sb_data;
333 if (n > so->so_urgc) n = so->so_urgc;
334 memcpy((buff + len), sb->sb_data, n);
338 n = slirp_send(so, buff, len, (MSG_OOB)); /* |MSG_DONTWAIT)); */
341 DEBUG_ERROR((dfd, "Didn't send all data urgently XXXXX\n"));
343 DEBUG_MISC((dfd, " ---2 sent %d bytes urgent data, %d urgent bytes left\n", n, so->so_urgc));
348 if (sb->sb_rptr >= (sb->sb_data + sb->sb_datalen))
349 sb->sb_rptr -= sb->sb_datalen;
355 * Write data from so_rcv to so's socket,
356 * updating all sbuf field as necessary
363 struct sbuf *sb = &so->so_rcv;
367 DEBUG_CALL("sowrite");
368 DEBUG_ARG("so = %lx", (long)so);
377 * No need to check if there's something to write,
378 * sowrite wouldn't have been called otherwise
383 iov[0].iov_base = sb->sb_rptr;
384 iov[1].iov_base = NULL;
386 if (sb->sb_rptr < sb->sb_wptr) {
387 iov[0].iov_len = sb->sb_wptr - sb->sb_rptr;
388 /* Should never succeed, but... */
389 if (iov[0].iov_len > len) iov[0].iov_len = len;
392 iov[0].iov_len = (sb->sb_data + sb->sb_datalen) - sb->sb_rptr;
393 if (iov[0].iov_len > len) iov[0].iov_len = len;
394 len -= iov[0].iov_len;
396 iov[1].iov_base = sb->sb_data;
397 iov[1].iov_len = sb->sb_wptr - sb->sb_data;
398 if (iov[1].iov_len > len) iov[1].iov_len = len;
403 /* Check if there's urgent data to send, and if so, send it */
406 nn = writev(so->s, (const struct iovec *)iov, n);
408 DEBUG_MISC((dfd, " ... wrote nn = %d bytes\n", nn));
410 nn = slirp_send(so, iov[0].iov_base, iov[0].iov_len,0);
412 /* This should never happen, but people tell me it does *shrug* */
413 if (nn < 0 && (errno == EAGAIN || errno == EINTR))
417 DEBUG_MISC((dfd, " --- sowrite disconnected, so->so_state = %x, errno = %d\n",
418 so->so_state, errno));
420 tcp_sockclosed(sototcpcb(so));
425 if (n == 2 && nn == iov[0].iov_len) {
427 ret = slirp_send(so, iov[1].iov_base, iov[1].iov_len,0);
431 DEBUG_MISC((dfd, " ... wrote nn = %d bytes\n", nn));
437 if (sb->sb_rptr >= (sb->sb_data + sb->sb_datalen))
438 sb->sb_rptr -= sb->sb_datalen;
441 * If in DRAIN mode, and there's no more data, set
444 if ((so->so_state & SS_FWDRAIN) && sb->sb_cc == 0)
451 * recvfrom() a UDP socket
457 struct sockaddr_in addr;
458 socklen_t addrlen = sizeof(struct sockaddr_in);
460 DEBUG_CALL("sorecvfrom");
461 DEBUG_ARG("so = %lx", (long)so);
463 if (so->so_type == IPPROTO_ICMP) { /* This is a "ping" reply */
467 len = recvfrom(so->s, buff, 256, 0,
468 (struct sockaddr *)&addr, &addrlen);
469 /* XXX Check if reply is "correct"? */
471 if(len == -1 || len == 0) {
472 u_char code=ICMP_UNREACH_PORT;
474 if(errno == EHOSTUNREACH) code=ICMP_UNREACH_HOST;
475 else if(errno == ENETUNREACH) code=ICMP_UNREACH_NET;
477 DEBUG_MISC((dfd," udp icmp rx errno = %d-%s\n",
478 errno,strerror(errno)));
479 icmp_error(so->so_m, ICMP_UNREACH,code, 0,strerror(errno));
481 icmp_reflect(so->so_m);
482 so->so_m = 0; /* Don't m_free() it again! */
484 /* No need for this socket anymore, udp_detach it */
486 } else { /* A "normal" UDP packet */
490 if (!(m = m_get())) return;
491 m->m_data += IF_MAXLINKHDR;
494 * XXX Shouldn't FIONREAD packets destined for port 53,
495 * but I don't know the max packet size for DNS lookups
498 /* if (so->so_fport != htons(53)) { */
499 ioctlsocket(so->s, FIONREAD, &n);
502 n = (m->m_data - m->m_dat) + m->m_len + n + 1;
508 m->m_len = recvfrom(so->s, m->m_data, len, 0,
509 (struct sockaddr *)&addr, &addrlen);
510 DEBUG_MISC((dfd, " did recvfrom %d, errno = %d-%s\n",
511 m->m_len, errno,strerror(errno)));
513 u_char code=ICMP_UNREACH_PORT;
515 if(errno == EHOSTUNREACH) code=ICMP_UNREACH_HOST;
516 else if(errno == ENETUNREACH) code=ICMP_UNREACH_NET;
518 DEBUG_MISC((dfd," rx error, tx icmp ICMP_UNREACH:%i\n", code));
519 icmp_error(so->so_m, ICMP_UNREACH,code, 0,strerror(errno));
523 * Hack: domain name lookup will be used the most for UDP,
524 * and since they'll only be used once there's no need
525 * for the 4 minute (or whatever) timeout... So we time them
526 * out much quicker (10 seconds for now...)
529 if (so->so_fport == htons(53))
530 so->so_expire = curtime + SO_EXPIREFAST;
532 so->so_expire = curtime + SO_EXPIRE;
535 /* if (m->m_len == len) {
536 * m_inc(m, MINCSIZE);
542 * If this packet was destined for CTL_ADDR,
543 * make it look like that's where it came from, done by udp_output
545 udp_output(so, m, &addr);
547 } /* if ping packet */
559 struct sockaddr_in addr;
561 DEBUG_CALL("sosendto");
562 DEBUG_ARG("so = %lx", (long)so);
563 DEBUG_ARG("m = %lx", (long)m);
565 addr.sin_family = AF_INET;
566 if ((so->so_faddr.s_addr & htonl(0xffffff00)) == special_addr.s_addr) {
568 switch(ntohl(so->so_faddr.s_addr) & 0xff) {
570 addr.sin_addr = dns_addr;
574 addr.sin_addr = loopback_addr;
578 addr.sin_addr = so->so_faddr;
579 addr.sin_port = so->so_fport;
581 DEBUG_MISC((dfd, " sendto()ing, addr.sin_port=%d, addr.sin_addr.s_addr=%.16s\n", ntohs(addr.sin_port), inet_ntoa(addr.sin_addr)));
583 /* Don't care what port we get */
584 ret = sendto(so->s, m->m_data, m->m_len, 0,
585 (struct sockaddr *)&addr, sizeof (struct sockaddr));
590 * Kill the socket if there's no reply in 4 minutes,
591 * but only if it's an expirable socket
594 so->so_expire = curtime + SO_EXPIRE;
595 so->so_state = SS_ISFCONNECTED; /* So that it gets select()ed */
600 * XXX This should really be tcp_listen
603 solisten(port, laddr, lport, flags)
609 struct sockaddr_in addr;
612 socklen_t addrlen = sizeof(addr);
614 DEBUG_CALL("solisten");
615 DEBUG_ARG("port = %d", port);
616 DEBUG_ARG("laddr = %x", laddr);
617 DEBUG_ARG("lport = %d", lport);
618 DEBUG_ARG("flags = %x", flags);
620 if ((so = socreate()) == NULL) {
621 /* free(so); Not sofree() ??? free(NULL) == NOP */
625 /* Don't tcp_attach... we don't need so_snd nor so_rcv */
626 if ((so->so_tcpcb = tcp_newtcpcb(so)) == NULL) {
633 * SS_FACCEPTONCE sockets must time out.
635 if (flags & SS_FACCEPTONCE)
636 so->so_tcpcb->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT*2;
638 so->so_state = (SS_FACCEPTCONN|flags);
639 so->so_lport = lport; /* Kept in network format */
640 so->so_laddr.s_addr = laddr; /* Ditto */
642 addr.sin_family = AF_INET;
643 addr.sin_addr.s_addr = INADDR_ANY;
644 addr.sin_port = port;
646 if (((s = socket(AF_INET,SOCK_STREAM,0)) < 0) ||
647 (setsockopt(s,SOL_SOCKET,SO_REUSEADDR,(char *)&opt,sizeof(int)) < 0) ||
648 (bind(s,(struct sockaddr *)&addr, sizeof(addr)) < 0) ||
650 int tmperrno = errno; /* Don't clobber the real reason we failed */
654 /* Restore the real errno */
656 WSASetLastError(tmperrno);
662 setsockopt(s,SOL_SOCKET,SO_OOBINLINE,(char *)&opt,sizeof(int));
664 getsockname(s,(struct sockaddr *)&addr,&addrlen);
665 so->so_fport = addr.sin_port;
666 if (addr.sin_addr.s_addr == 0 || addr.sin_addr.s_addr == loopback_addr.s_addr)
667 so->so_faddr = alias_addr;
669 so->so_faddr = addr.sin_addr;
677 * Data is available in so_rcv
678 * Just write() the data to the socket
686 /* FD_CLR(so->s,&writefds); */
690 * Data has been freed in so_snd
691 * We have room for a read() if we want to
692 * For now, don't read, it'll be done in the main loop
703 * Various session state calls
704 * XXX Should be #define's
705 * The socket state stuff needs work, these often get call 2 or 3
706 * times each when only 1 was needed
710 register struct socket *so;
712 so->so_state &= ~(SS_NOFDREF|SS_ISFCONNECTED|SS_FCANTRCVMORE|
713 SS_FCANTSENDMORE|SS_FWDRAIN);
714 so->so_state |= SS_ISFCONNECTING; /* Clobber other states */
719 register struct socket *so;
721 so->so_state &= ~(SS_ISFCONNECTING|SS_FWDRAIN|SS_NOFDREF);
722 so->so_state |= SS_ISFCONNECTED; /* Clobber other states */
726 sofcantrcvmore(struct socket *so)
728 if ((so->so_state & SS_NOFDREF) == 0) {
730 if(global_writefds) {
731 FD_CLR(so->s,global_writefds);
734 so->so_state &= ~(SS_ISFCONNECTING);
735 if (so->so_state & SS_FCANTSENDMORE)
736 so->so_state = SS_NOFDREF; /* Don't select it */ /* XXX close() here as well? */
738 so->so_state |= SS_FCANTRCVMORE;
742 sofcantsendmore(struct socket *so)
744 if ((so->so_state & SS_NOFDREF) == 0) {
745 shutdown(so->s,1); /* send FIN to fhost */
746 if (global_readfds) {
747 FD_CLR(so->s,global_readfds);
750 FD_CLR(so->s,global_xfds);
753 so->so_state &= ~(SS_ISFCONNECTING);
754 if (so->so_state & SS_FCANTRCVMORE)
755 so->so_state = SS_NOFDREF; /* as above */
757 so->so_state |= SS_FCANTSENDMORE;
761 soisfdisconnected(so)
764 /* so->so_state &= ~(SS_ISFCONNECTING|SS_ISFCONNECTED); */
766 /* so->so_state = SS_ISFDISCONNECTED; */
768 * XXX Do nothing ... ?
773 * Set write drain mode
774 * Set CANTSENDMORE once all data has been write()n
780 if (so->so_rcv.sb_cc)
781 so->so_state |= SS_FWDRAIN;