slirp: Make Socket structure IPv6 compatible
[qemu/ar7.git] / slirp / socket.c
blobbf603c9185bedddf4c6c4d3771e07a553f05ed4a
1 /*
2 * Copyright (c) 1995 Danny Gasparovski.
4 * Please read the file COPYRIGHT for the
5 * terms and conditions of the copyright.
6 */
8 #include "qemu-common.h"
9 #include <slirp.h>
10 #include "ip_icmp.h"
11 #ifdef __sun__
12 #include <sys/filio.h>
13 #endif
15 static void sofcantrcvmore(struct socket *so);
16 static void sofcantsendmore(struct socket *so);
18 struct socket *
19 solookup(struct socket *head, struct in_addr laddr, u_int lport,
20 struct in_addr faddr, u_int fport)
22 struct socket *so;
24 for (so = head->so_next; so != head; so = so->so_next) {
25 if (so->so_lport == lport &&
26 so->so_laddr.s_addr == laddr.s_addr &&
27 so->so_faddr.s_addr == faddr.s_addr &&
28 so->so_fport == fport)
29 break;
32 if (so == head)
33 return (struct socket *)NULL;
34 return so;
39 * Create a new socket, initialise the fields
40 * It is the responsibility of the caller to
41 * insque() it into the correct linked-list
43 struct socket *
44 socreate(Slirp *slirp)
46 struct socket *so;
48 so = (struct socket *)malloc(sizeof(struct socket));
49 if(so) {
50 memset(so, 0, sizeof(struct socket));
51 so->so_state = SS_NOFDREF;
52 so->s = -1;
53 so->slirp = slirp;
54 so->pollfds_idx = -1;
56 return(so);
60 * remque and free a socket, clobber cache
62 void
63 sofree(struct socket *so)
65 Slirp *slirp = so->slirp;
67 if (so->so_emu==EMU_RSH && so->extra) {
68 sofree(so->extra);
69 so->extra=NULL;
71 if (so == slirp->tcp_last_so) {
72 slirp->tcp_last_so = &slirp->tcb;
73 } else if (so == slirp->udp_last_so) {
74 slirp->udp_last_so = &slirp->udb;
75 } else if (so == slirp->icmp_last_so) {
76 slirp->icmp_last_so = &slirp->icmp;
78 m_free(so->so_m);
80 if(so->so_next && so->so_prev)
81 remque(so); /* crashes if so is not in a queue */
83 free(so);
86 size_t sopreprbuf(struct socket *so, struct iovec *iov, int *np)
88 int n, lss, total;
89 struct sbuf *sb = &so->so_snd;
90 int len = sb->sb_datalen - sb->sb_cc;
91 int mss = so->so_tcpcb->t_maxseg;
93 DEBUG_CALL("sopreprbuf");
94 DEBUG_ARG("so = %p", so);
96 if (len <= 0)
97 return 0;
99 iov[0].iov_base = sb->sb_wptr;
100 iov[1].iov_base = NULL;
101 iov[1].iov_len = 0;
102 if (sb->sb_wptr < sb->sb_rptr) {
103 iov[0].iov_len = sb->sb_rptr - sb->sb_wptr;
104 /* Should never succeed, but... */
105 if (iov[0].iov_len > len)
106 iov[0].iov_len = len;
107 if (iov[0].iov_len > mss)
108 iov[0].iov_len -= iov[0].iov_len%mss;
109 n = 1;
110 } else {
111 iov[0].iov_len = (sb->sb_data + sb->sb_datalen) - sb->sb_wptr;
112 /* Should never succeed, but... */
113 if (iov[0].iov_len > len) iov[0].iov_len = len;
114 len -= iov[0].iov_len;
115 if (len) {
116 iov[1].iov_base = sb->sb_data;
117 iov[1].iov_len = sb->sb_rptr - sb->sb_data;
118 if(iov[1].iov_len > len)
119 iov[1].iov_len = len;
120 total = iov[0].iov_len + iov[1].iov_len;
121 if (total > mss) {
122 lss = total%mss;
123 if (iov[1].iov_len > lss) {
124 iov[1].iov_len -= lss;
125 n = 2;
126 } else {
127 lss -= iov[1].iov_len;
128 iov[0].iov_len -= lss;
129 n = 1;
131 } else
132 n = 2;
133 } else {
134 if (iov[0].iov_len > mss)
135 iov[0].iov_len -= iov[0].iov_len%mss;
136 n = 1;
139 if (np)
140 *np = n;
142 return iov[0].iov_len + (n - 1) * iov[1].iov_len;
146 * Read from so's socket into sb_snd, updating all relevant sbuf fields
147 * NOTE: This will only be called if it is select()ed for reading, so
148 * a read() of 0 (or less) means it's disconnected
151 soread(struct socket *so)
153 int n, nn;
154 struct sbuf *sb = &so->so_snd;
155 struct iovec iov[2];
157 DEBUG_CALL("soread");
158 DEBUG_ARG("so = %p", so);
161 * No need to check if there's enough room to read.
162 * soread wouldn't have been called if there weren't
164 sopreprbuf(so, iov, &n);
166 #ifdef HAVE_READV
167 nn = readv(so->s, (struct iovec *)iov, n);
168 DEBUG_MISC((dfd, " ... read nn = %d bytes\n", nn));
169 #else
170 nn = qemu_recv(so->s, iov[0].iov_base, iov[0].iov_len,0);
171 #endif
172 if (nn <= 0) {
173 if (nn < 0 && (errno == EINTR || errno == EAGAIN))
174 return 0;
175 else {
176 DEBUG_MISC((dfd, " --- soread() disconnected, nn = %d, errno = %d-%s\n", nn, errno,strerror(errno)));
177 sofcantrcvmore(so);
178 tcp_sockclosed(sototcpcb(so));
179 return -1;
183 #ifndef HAVE_READV
185 * If there was no error, try and read the second time round
186 * We read again if n = 2 (ie, there's another part of the buffer)
187 * and we read as much as we could in the first read
188 * We don't test for <= 0 this time, because there legitimately
189 * might not be any more data (since the socket is non-blocking),
190 * a close will be detected on next iteration.
191 * A return of -1 wont (shouldn't) happen, since it didn't happen above
193 if (n == 2 && nn == iov[0].iov_len) {
194 int ret;
195 ret = qemu_recv(so->s, iov[1].iov_base, iov[1].iov_len,0);
196 if (ret > 0)
197 nn += ret;
200 DEBUG_MISC((dfd, " ... read nn = %d bytes\n", nn));
201 #endif
203 /* Update fields */
204 sb->sb_cc += nn;
205 sb->sb_wptr += nn;
206 if (sb->sb_wptr >= (sb->sb_data + sb->sb_datalen))
207 sb->sb_wptr -= sb->sb_datalen;
208 return nn;
211 int soreadbuf(struct socket *so, const char *buf, int size)
213 int n, nn, copy = size;
214 struct sbuf *sb = &so->so_snd;
215 struct iovec iov[2];
217 DEBUG_CALL("soreadbuf");
218 DEBUG_ARG("so = %p", so);
221 * No need to check if there's enough room to read.
222 * soread wouldn't have been called if there weren't
224 if (sopreprbuf(so, iov, &n) < size)
225 goto err;
227 nn = MIN(iov[0].iov_len, copy);
228 memcpy(iov[0].iov_base, buf, nn);
230 copy -= nn;
231 buf += nn;
233 if (copy == 0)
234 goto done;
236 memcpy(iov[1].iov_base, buf, copy);
238 done:
239 /* Update fields */
240 sb->sb_cc += size;
241 sb->sb_wptr += size;
242 if (sb->sb_wptr >= (sb->sb_data + sb->sb_datalen))
243 sb->sb_wptr -= sb->sb_datalen;
244 return size;
245 err:
247 sofcantrcvmore(so);
248 tcp_sockclosed(sototcpcb(so));
249 fprintf(stderr, "soreadbuf buffer to small");
250 return -1;
254 * Get urgent data
256 * When the socket is created, we set it SO_OOBINLINE,
257 * so when OOB data arrives, we soread() it and everything
258 * in the send buffer is sent as urgent data
260 void
261 sorecvoob(struct socket *so)
263 struct tcpcb *tp = sototcpcb(so);
265 DEBUG_CALL("sorecvoob");
266 DEBUG_ARG("so = %p", so);
269 * We take a guess at how much urgent data has arrived.
270 * In most situations, when urgent data arrives, the next
271 * read() should get all the urgent data. This guess will
272 * be wrong however if more data arrives just after the
273 * urgent data, or the read() doesn't return all the
274 * urgent data.
276 soread(so);
277 tp->snd_up = tp->snd_una + so->so_snd.sb_cc;
278 tp->t_force = 1;
279 tcp_output(tp);
280 tp->t_force = 0;
284 * Send urgent data
285 * There's a lot duplicated code here, but...
288 sosendoob(struct socket *so)
290 struct sbuf *sb = &so->so_rcv;
291 char buff[2048]; /* XXX Shouldn't be sending more oob data than this */
293 int n, len;
295 DEBUG_CALL("sosendoob");
296 DEBUG_ARG("so = %p", so);
297 DEBUG_ARG("sb->sb_cc = %d", sb->sb_cc);
299 if (so->so_urgc > 2048)
300 so->so_urgc = 2048; /* XXXX */
302 if (sb->sb_rptr < sb->sb_wptr) {
303 /* We can send it directly */
304 n = slirp_send(so, sb->sb_rptr, so->so_urgc, (MSG_OOB)); /* |MSG_DONTWAIT)); */
305 so->so_urgc -= n;
307 DEBUG_MISC((dfd, " --- sent %d bytes urgent data, %d urgent bytes left\n", n, so->so_urgc));
308 } else {
310 * Since there's no sendv or sendtov like writev,
311 * we must copy all data to a linear buffer then
312 * send it all
314 len = (sb->sb_data + sb->sb_datalen) - sb->sb_rptr;
315 if (len > so->so_urgc) len = so->so_urgc;
316 memcpy(buff, sb->sb_rptr, len);
317 so->so_urgc -= len;
318 if (so->so_urgc) {
319 n = sb->sb_wptr - sb->sb_data;
320 if (n > so->so_urgc) n = so->so_urgc;
321 memcpy((buff + len), sb->sb_data, n);
322 so->so_urgc -= n;
323 len += n;
325 n = slirp_send(so, buff, len, (MSG_OOB)); /* |MSG_DONTWAIT)); */
326 #ifdef DEBUG
327 if (n != len)
328 DEBUG_ERROR((dfd, "Didn't send all data urgently XXXXX\n"));
329 #endif
330 DEBUG_MISC((dfd, " ---2 sent %d bytes urgent data, %d urgent bytes left\n", n, so->so_urgc));
333 sb->sb_cc -= n;
334 sb->sb_rptr += n;
335 if (sb->sb_rptr >= (sb->sb_data + sb->sb_datalen))
336 sb->sb_rptr -= sb->sb_datalen;
338 return n;
342 * Write data from so_rcv to so's socket,
343 * updating all sbuf field as necessary
346 sowrite(struct socket *so)
348 int n,nn;
349 struct sbuf *sb = &so->so_rcv;
350 int len = sb->sb_cc;
351 struct iovec iov[2];
353 DEBUG_CALL("sowrite");
354 DEBUG_ARG("so = %p", so);
356 if (so->so_urgc) {
357 sosendoob(so);
358 if (sb->sb_cc == 0)
359 return 0;
363 * No need to check if there's something to write,
364 * sowrite wouldn't have been called otherwise
367 iov[0].iov_base = sb->sb_rptr;
368 iov[1].iov_base = NULL;
369 iov[1].iov_len = 0;
370 if (sb->sb_rptr < sb->sb_wptr) {
371 iov[0].iov_len = sb->sb_wptr - sb->sb_rptr;
372 /* Should never succeed, but... */
373 if (iov[0].iov_len > len) iov[0].iov_len = len;
374 n = 1;
375 } else {
376 iov[0].iov_len = (sb->sb_data + sb->sb_datalen) - sb->sb_rptr;
377 if (iov[0].iov_len > len) iov[0].iov_len = len;
378 len -= iov[0].iov_len;
379 if (len) {
380 iov[1].iov_base = sb->sb_data;
381 iov[1].iov_len = sb->sb_wptr - sb->sb_data;
382 if (iov[1].iov_len > len) iov[1].iov_len = len;
383 n = 2;
384 } else
385 n = 1;
387 /* Check if there's urgent data to send, and if so, send it */
389 #ifdef HAVE_READV
390 nn = writev(so->s, (const struct iovec *)iov, n);
392 DEBUG_MISC((dfd, " ... wrote nn = %d bytes\n", nn));
393 #else
394 nn = slirp_send(so, iov[0].iov_base, iov[0].iov_len,0);
395 #endif
396 /* This should never happen, but people tell me it does *shrug* */
397 if (nn < 0 && (errno == EAGAIN || errno == EINTR))
398 return 0;
400 if (nn <= 0) {
401 DEBUG_MISC((dfd, " --- sowrite disconnected, so->so_state = %x, errno = %d\n",
402 so->so_state, errno));
403 sofcantsendmore(so);
404 tcp_sockclosed(sototcpcb(so));
405 return -1;
408 #ifndef HAVE_READV
409 if (n == 2 && nn == iov[0].iov_len) {
410 int ret;
411 ret = slirp_send(so, iov[1].iov_base, iov[1].iov_len,0);
412 if (ret > 0)
413 nn += ret;
415 DEBUG_MISC((dfd, " ... wrote nn = %d bytes\n", nn));
416 #endif
418 /* Update sbuf */
419 sb->sb_cc -= nn;
420 sb->sb_rptr += nn;
421 if (sb->sb_rptr >= (sb->sb_data + sb->sb_datalen))
422 sb->sb_rptr -= sb->sb_datalen;
425 * If in DRAIN mode, and there's no more data, set
426 * it CANTSENDMORE
428 if ((so->so_state & SS_FWDRAIN) && sb->sb_cc == 0)
429 sofcantsendmore(so);
431 return nn;
435 * recvfrom() a UDP socket
437 void
438 sorecvfrom(struct socket *so)
440 struct sockaddr_storage addr;
441 socklen_t addrlen = sizeof(struct sockaddr_storage);
443 DEBUG_CALL("sorecvfrom");
444 DEBUG_ARG("so = %p", so);
446 if (so->so_type == IPPROTO_ICMP) { /* This is a "ping" reply */
447 char buff[256];
448 int len;
450 len = recvfrom(so->s, buff, 256, 0,
451 (struct sockaddr *)&addr, &addrlen);
452 /* XXX Check if reply is "correct"? */
454 if(len == -1 || len == 0) {
455 u_char code=ICMP_UNREACH_PORT;
457 if(errno == EHOSTUNREACH) code=ICMP_UNREACH_HOST;
458 else if(errno == ENETUNREACH) code=ICMP_UNREACH_NET;
460 DEBUG_MISC((dfd," udp icmp rx errno = %d-%s\n",
461 errno,strerror(errno)));
462 icmp_error(so->so_m, ICMP_UNREACH,code, 0,strerror(errno));
463 } else {
464 icmp_reflect(so->so_m);
465 so->so_m = NULL; /* Don't m_free() it again! */
467 /* No need for this socket anymore, udp_detach it */
468 udp_detach(so);
469 } else { /* A "normal" UDP packet */
470 struct mbuf *m;
471 int len;
472 #ifdef _WIN32
473 unsigned long n;
474 #else
475 int n;
476 #endif
478 m = m_get(so->slirp);
479 if (!m) {
480 return;
482 m->m_data += IF_MAXLINKHDR;
485 * XXX Shouldn't FIONREAD packets destined for port 53,
486 * but I don't know the max packet size for DNS lookups
488 len = M_FREEROOM(m);
489 /* if (so->so_fport != htons(53)) { */
490 ioctlsocket(so->s, FIONREAD, &n);
492 if (n > len) {
493 n = (m->m_data - m->m_dat) + m->m_len + n + 1;
494 m_inc(m, n);
495 len = M_FREEROOM(m);
497 /* } */
499 m->m_len = recvfrom(so->s, m->m_data, len, 0,
500 (struct sockaddr *)&addr, &addrlen);
501 DEBUG_MISC((dfd, " did recvfrom %d, errno = %d-%s\n",
502 m->m_len, errno,strerror(errno)));
503 if(m->m_len<0) {
504 u_char code=ICMP_UNREACH_PORT;
506 if(errno == EHOSTUNREACH) code=ICMP_UNREACH_HOST;
507 else if(errno == ENETUNREACH) code=ICMP_UNREACH_NET;
509 DEBUG_MISC((dfd," rx error, tx icmp ICMP_UNREACH:%i\n", code));
510 icmp_error(so->so_m, ICMP_UNREACH,code, 0,strerror(errno));
511 m_free(m);
512 } else {
514 * Hack: domain name lookup will be used the most for UDP,
515 * and since they'll only be used once there's no need
516 * for the 4 minute (or whatever) timeout... So we time them
517 * out much quicker (10 seconds for now...)
519 if (so->so_expire) {
520 if (so->so_fport == htons(53))
521 so->so_expire = curtime + SO_EXPIREFAST;
522 else
523 so->so_expire = curtime + SO_EXPIRE;
527 * If this packet was destined for CTL_ADDR,
528 * make it look like that's where it came from, done by udp_output
530 switch (so->so_ffamily) {
531 case AF_INET:
532 udp_output(so, m, (struct sockaddr_in *) &addr);
533 break;
534 default:
535 break;
537 } /* rx error */
538 } /* if ping packet */
542 * sendto() a socket
545 sosendto(struct socket *so, struct mbuf *m)
547 Slirp *slirp = so->slirp;
548 int ret;
549 struct sockaddr_in addr;
551 DEBUG_CALL("sosendto");
552 DEBUG_ARG("so = %p", so);
553 DEBUG_ARG("m = %p", m);
555 addr.sin_family = AF_INET;
556 if ((so->so_faddr.s_addr & slirp->vnetwork_mask.s_addr) ==
557 slirp->vnetwork_addr.s_addr) {
558 /* It's an alias */
559 if (so->so_faddr.s_addr == slirp->vnameserver_addr.s_addr) {
560 if (get_dns_addr(&addr.sin_addr) < 0)
561 addr.sin_addr = loopback_addr;
562 } else {
563 addr.sin_addr = loopback_addr;
565 } else
566 addr.sin_addr = so->so_faddr;
567 addr.sin_port = so->so_fport;
569 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)));
571 /* Don't care what port we get */
572 ret = sendto(so->s, m->m_data, m->m_len, 0,
573 (struct sockaddr *)&addr, sizeof (struct sockaddr));
574 if (ret < 0)
575 return -1;
578 * Kill the socket if there's no reply in 4 minutes,
579 * but only if it's an expirable socket
581 if (so->so_expire)
582 so->so_expire = curtime + SO_EXPIRE;
583 so->so_state &= SS_PERSISTENT_MASK;
584 so->so_state |= SS_ISFCONNECTED; /* So that it gets select()ed */
585 return 0;
589 * Listen for incoming TCP connections
591 struct socket *
592 tcp_listen(Slirp *slirp, uint32_t haddr, u_int hport, uint32_t laddr,
593 u_int lport, int flags)
595 struct sockaddr_in addr;
596 struct socket *so;
597 int s, opt = 1;
598 socklen_t addrlen = sizeof(addr);
599 memset(&addr, 0, addrlen);
601 DEBUG_CALL("tcp_listen");
602 DEBUG_ARG("haddr = %x", haddr);
603 DEBUG_ARG("hport = %d", hport);
604 DEBUG_ARG("laddr = %x", laddr);
605 DEBUG_ARG("lport = %d", lport);
606 DEBUG_ARG("flags = %x", flags);
608 so = socreate(slirp);
609 if (!so) {
610 return NULL;
613 /* Don't tcp_attach... we don't need so_snd nor so_rcv */
614 if ((so->so_tcpcb = tcp_newtcpcb(so)) == NULL) {
615 free(so);
616 return NULL;
618 insque(so, &slirp->tcb);
621 * SS_FACCEPTONCE sockets must time out.
623 if (flags & SS_FACCEPTONCE)
624 so->so_tcpcb->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT*2;
626 so->so_state &= SS_PERSISTENT_MASK;
627 so->so_state |= (SS_FACCEPTCONN | flags);
628 so->so_lfamily = AF_INET;
629 so->so_lport = lport; /* Kept in network format */
630 so->so_laddr.s_addr = laddr; /* Ditto */
632 addr.sin_family = AF_INET;
633 addr.sin_addr.s_addr = haddr;
634 addr.sin_port = hport;
636 if (((s = qemu_socket(AF_INET,SOCK_STREAM,0)) < 0) ||
637 (socket_set_fast_reuse(s) < 0) ||
638 (bind(s,(struct sockaddr *)&addr, sizeof(addr)) < 0) ||
639 (listen(s,1) < 0)) {
640 int tmperrno = errno; /* Don't clobber the real reason we failed */
642 close(s);
643 sofree(so);
644 /* Restore the real errno */
645 #ifdef _WIN32
646 WSASetLastError(tmperrno);
647 #else
648 errno = tmperrno;
649 #endif
650 return NULL;
652 qemu_setsockopt(s, SOL_SOCKET, SO_OOBINLINE, &opt, sizeof(int));
654 getsockname(s,(struct sockaddr *)&addr,&addrlen);
655 so->so_ffamily = AF_INET;
656 so->so_fport = addr.sin_port;
657 if (addr.sin_addr.s_addr == 0 || addr.sin_addr.s_addr == loopback_addr.s_addr)
658 so->so_faddr = slirp->vhost_addr;
659 else
660 so->so_faddr = addr.sin_addr;
662 so->s = s;
663 return so;
667 * Various session state calls
668 * XXX Should be #define's
669 * The socket state stuff needs work, these often get call 2 or 3
670 * times each when only 1 was needed
672 void
673 soisfconnecting(struct socket *so)
675 so->so_state &= ~(SS_NOFDREF|SS_ISFCONNECTED|SS_FCANTRCVMORE|
676 SS_FCANTSENDMORE|SS_FWDRAIN);
677 so->so_state |= SS_ISFCONNECTING; /* Clobber other states */
680 void
681 soisfconnected(struct socket *so)
683 so->so_state &= ~(SS_ISFCONNECTING|SS_FWDRAIN|SS_NOFDREF);
684 so->so_state |= SS_ISFCONNECTED; /* Clobber other states */
687 static void
688 sofcantrcvmore(struct socket *so)
690 if ((so->so_state & SS_NOFDREF) == 0) {
691 shutdown(so->s,0);
693 so->so_state &= ~(SS_ISFCONNECTING);
694 if (so->so_state & SS_FCANTSENDMORE) {
695 so->so_state &= SS_PERSISTENT_MASK;
696 so->so_state |= SS_NOFDREF; /* Don't select it */
697 } else {
698 so->so_state |= SS_FCANTRCVMORE;
702 static void
703 sofcantsendmore(struct socket *so)
705 if ((so->so_state & SS_NOFDREF) == 0) {
706 shutdown(so->s,1); /* send FIN to fhost */
708 so->so_state &= ~(SS_ISFCONNECTING);
709 if (so->so_state & SS_FCANTRCVMORE) {
710 so->so_state &= SS_PERSISTENT_MASK;
711 so->so_state |= SS_NOFDREF; /* as above */
712 } else {
713 so->so_state |= SS_FCANTSENDMORE;
718 * Set write drain mode
719 * Set CANTSENDMORE once all data has been write()n
721 void
722 sofwdrain(struct socket *so)
724 if (so->so_rcv.sb_cc)
725 so->so_state |= SS_FWDRAIN;
726 else
727 sofcantsendmore(so);