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"
12 #include <sys/filio.h>
15 static void sofcantrcvmore(struct socket
*so
);
16 static void sofcantsendmore(struct socket
*so
);
19 solookup(struct socket
*head
, struct in_addr laddr
, u_int lport
,
20 struct in_addr faddr
, u_int fport
)
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
)
33 return (struct socket
*)NULL
;
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
44 socreate(Slirp
*slirp
)
48 so
= (struct socket
*)malloc(sizeof(struct socket
));
50 memset(so
, 0, sizeof(struct socket
));
51 so
->so_state
= SS_NOFDREF
;
59 * remque and free a socket, clobber cache
62 sofree(struct socket
*so
)
64 Slirp
*slirp
= so
->slirp
;
66 if (so
->so_emu
==EMU_RSH
&& so
->extra
) {
70 if (so
== slirp
->tcp_last_so
) {
71 slirp
->tcp_last_so
= &slirp
->tcb
;
72 } else if (so
== slirp
->udp_last_so
) {
73 slirp
->udp_last_so
= &slirp
->udb
;
77 if(so
->so_next
&& so
->so_prev
)
78 remque(so
); /* crashes if so is not in a queue */
83 size_t sopreprbuf(struct socket
*so
, struct iovec
*iov
, int *np
)
86 struct sbuf
*sb
= &so
->so_snd
;
87 int len
= sb
->sb_datalen
- sb
->sb_cc
;
88 int mss
= so
->so_tcpcb
->t_maxseg
;
90 DEBUG_CALL("sopreprbuf");
91 DEBUG_ARG("so = %lx", (long )so
);
96 iov
[0].iov_base
= sb
->sb_wptr
;
97 iov
[1].iov_base
= NULL
;
99 if (sb
->sb_wptr
< sb
->sb_rptr
) {
100 iov
[0].iov_len
= sb
->sb_rptr
- sb
->sb_wptr
;
101 /* Should never succeed, but... */
102 if (iov
[0].iov_len
> len
)
103 iov
[0].iov_len
= len
;
104 if (iov
[0].iov_len
> mss
)
105 iov
[0].iov_len
-= iov
[0].iov_len
%mss
;
108 iov
[0].iov_len
= (sb
->sb_data
+ sb
->sb_datalen
) - sb
->sb_wptr
;
109 /* Should never succeed, but... */
110 if (iov
[0].iov_len
> len
) iov
[0].iov_len
= len
;
111 len
-= iov
[0].iov_len
;
113 iov
[1].iov_base
= sb
->sb_data
;
114 iov
[1].iov_len
= sb
->sb_rptr
- sb
->sb_data
;
115 if(iov
[1].iov_len
> len
)
116 iov
[1].iov_len
= len
;
117 total
= iov
[0].iov_len
+ iov
[1].iov_len
;
120 if (iov
[1].iov_len
> lss
) {
121 iov
[1].iov_len
-= lss
;
124 lss
-= iov
[1].iov_len
;
125 iov
[0].iov_len
-= lss
;
131 if (iov
[0].iov_len
> mss
)
132 iov
[0].iov_len
-= iov
[0].iov_len
%mss
;
139 return iov
[0].iov_len
+ (n
- 1) * iov
[1].iov_len
;
143 * Read from so's socket into sb_snd, updating all relevant sbuf fields
144 * NOTE: This will only be called if it is select()ed for reading, so
145 * a read() of 0 (or less) means it's disconnected
148 soread(struct socket
*so
)
151 struct sbuf
*sb
= &so
->so_snd
;
154 DEBUG_CALL("soread");
155 DEBUG_ARG("so = %lx", (long )so
);
158 * No need to check if there's enough room to read.
159 * soread wouldn't have been called if there weren't
161 sopreprbuf(so
, iov
, &n
);
164 nn
= readv(so
->s
, (struct iovec
*)iov
, n
);
165 DEBUG_MISC((dfd
, " ... read nn = %d bytes\n", nn
));
167 nn
= recv(so
->s
, iov
[0].iov_base
, iov
[0].iov_len
,0);
170 if (nn
< 0 && (errno
== EINTR
|| errno
== EAGAIN
))
173 DEBUG_MISC((dfd
, " --- soread() disconnected, nn = %d, errno = %d-%s\n", nn
, errno
,strerror(errno
)));
175 tcp_sockclosed(sototcpcb(so
));
182 * If there was no error, try and read the second time round
183 * We read again if n = 2 (ie, there's another part of the buffer)
184 * and we read as much as we could in the first read
185 * We don't test for <= 0 this time, because there legitimately
186 * might not be any more data (since the socket is non-blocking),
187 * a close will be detected on next iteration.
188 * A return of -1 wont (shouldn't) happen, since it didn't happen above
190 if (n
== 2 && nn
== iov
[0].iov_len
) {
192 ret
= recv(so
->s
, iov
[1].iov_base
, iov
[1].iov_len
,0);
197 DEBUG_MISC((dfd
, " ... read nn = %d bytes\n", nn
));
203 if (sb
->sb_wptr
>= (sb
->sb_data
+ sb
->sb_datalen
))
204 sb
->sb_wptr
-= sb
->sb_datalen
;
208 int soreadbuf(struct socket
*so
, const char *buf
, int size
)
210 int n
, nn
, copy
= size
;
211 struct sbuf
*sb
= &so
->so_snd
;
214 DEBUG_CALL("soreadbuf");
215 DEBUG_ARG("so = %lx", (long )so
);
218 * No need to check if there's enough room to read.
219 * soread wouldn't have been called if there weren't
221 if (sopreprbuf(so
, iov
, &n
) < size
)
224 nn
= MIN(iov
[0].iov_len
, copy
);
225 memcpy(iov
[0].iov_base
, buf
, nn
);
233 memcpy(iov
[1].iov_base
, buf
, copy
);
239 if (sb
->sb_wptr
>= (sb
->sb_data
+ sb
->sb_datalen
))
240 sb
->sb_wptr
-= sb
->sb_datalen
;
245 tcp_sockclosed(sototcpcb(so
));
246 fprintf(stderr
, "soreadbuf buffer to small");
253 * When the socket is created, we set it SO_OOBINLINE,
254 * so when OOB data arrives, we soread() it and everything
255 * in the send buffer is sent as urgent data
258 sorecvoob(struct socket
*so
)
260 struct tcpcb
*tp
= sototcpcb(so
);
262 DEBUG_CALL("sorecvoob");
263 DEBUG_ARG("so = %lx", (long)so
);
266 * We take a guess at how much urgent data has arrived.
267 * In most situations, when urgent data arrives, the next
268 * read() should get all the urgent data. This guess will
269 * be wrong however if more data arrives just after the
270 * urgent data, or the read() doesn't return all the
274 tp
->snd_up
= tp
->snd_una
+ so
->so_snd
.sb_cc
;
282 * There's a lot duplicated code here, but...
285 sosendoob(struct socket
*so
)
287 struct sbuf
*sb
= &so
->so_rcv
;
288 char buff
[2048]; /* XXX Shouldn't be sending more oob data than this */
292 DEBUG_CALL("sosendoob");
293 DEBUG_ARG("so = %lx", (long)so
);
294 DEBUG_ARG("sb->sb_cc = %d", sb
->sb_cc
);
296 if (so
->so_urgc
> 2048)
297 so
->so_urgc
= 2048; /* XXXX */
299 if (sb
->sb_rptr
< sb
->sb_wptr
) {
300 /* We can send it directly */
301 n
= slirp_send(so
, sb
->sb_rptr
, so
->so_urgc
, (MSG_OOB
)); /* |MSG_DONTWAIT)); */
304 DEBUG_MISC((dfd
, " --- sent %d bytes urgent data, %d urgent bytes left\n", n
, so
->so_urgc
));
307 * Since there's no sendv or sendtov like writev,
308 * we must copy all data to a linear buffer then
311 len
= (sb
->sb_data
+ sb
->sb_datalen
) - sb
->sb_rptr
;
312 if (len
> so
->so_urgc
) len
= so
->so_urgc
;
313 memcpy(buff
, sb
->sb_rptr
, len
);
316 n
= sb
->sb_wptr
- sb
->sb_data
;
317 if (n
> so
->so_urgc
) n
= so
->so_urgc
;
318 memcpy((buff
+ len
), sb
->sb_data
, n
);
322 n
= slirp_send(so
, buff
, len
, (MSG_OOB
)); /* |MSG_DONTWAIT)); */
325 DEBUG_ERROR((dfd
, "Didn't send all data urgently XXXXX\n"));
327 DEBUG_MISC((dfd
, " ---2 sent %d bytes urgent data, %d urgent bytes left\n", n
, so
->so_urgc
));
332 if (sb
->sb_rptr
>= (sb
->sb_data
+ sb
->sb_datalen
))
333 sb
->sb_rptr
-= sb
->sb_datalen
;
339 * Write data from so_rcv to so's socket,
340 * updating all sbuf field as necessary
343 sowrite(struct socket
*so
)
346 struct sbuf
*sb
= &so
->so_rcv
;
350 DEBUG_CALL("sowrite");
351 DEBUG_ARG("so = %lx", (long)so
);
360 * No need to check if there's something to write,
361 * sowrite wouldn't have been called otherwise
364 iov
[0].iov_base
= sb
->sb_rptr
;
365 iov
[1].iov_base
= NULL
;
367 if (sb
->sb_rptr
< sb
->sb_wptr
) {
368 iov
[0].iov_len
= sb
->sb_wptr
- sb
->sb_rptr
;
369 /* Should never succeed, but... */
370 if (iov
[0].iov_len
> len
) iov
[0].iov_len
= len
;
373 iov
[0].iov_len
= (sb
->sb_data
+ sb
->sb_datalen
) - sb
->sb_rptr
;
374 if (iov
[0].iov_len
> len
) iov
[0].iov_len
= len
;
375 len
-= iov
[0].iov_len
;
377 iov
[1].iov_base
= sb
->sb_data
;
378 iov
[1].iov_len
= sb
->sb_wptr
- sb
->sb_data
;
379 if (iov
[1].iov_len
> len
) iov
[1].iov_len
= len
;
384 /* Check if there's urgent data to send, and if so, send it */
387 nn
= writev(so
->s
, (const struct iovec
*)iov
, n
);
389 DEBUG_MISC((dfd
, " ... wrote nn = %d bytes\n", nn
));
391 nn
= slirp_send(so
, iov
[0].iov_base
, iov
[0].iov_len
,0);
393 /* This should never happen, but people tell me it does *shrug* */
394 if (nn
< 0 && (errno
== EAGAIN
|| errno
== EINTR
))
398 DEBUG_MISC((dfd
, " --- sowrite disconnected, so->so_state = %x, errno = %d\n",
399 so
->so_state
, errno
));
401 tcp_sockclosed(sototcpcb(so
));
406 if (n
== 2 && nn
== iov
[0].iov_len
) {
408 ret
= slirp_send(so
, iov
[1].iov_base
, iov
[1].iov_len
,0);
412 DEBUG_MISC((dfd
, " ... wrote nn = %d bytes\n", nn
));
418 if (sb
->sb_rptr
>= (sb
->sb_data
+ sb
->sb_datalen
))
419 sb
->sb_rptr
-= sb
->sb_datalen
;
422 * If in DRAIN mode, and there's no more data, set
425 if ((so
->so_state
& SS_FWDRAIN
) && sb
->sb_cc
== 0)
432 * recvfrom() a UDP socket
435 sorecvfrom(struct socket
*so
)
437 struct sockaddr_in addr
;
438 socklen_t addrlen
= sizeof(struct sockaddr_in
);
440 DEBUG_CALL("sorecvfrom");
441 DEBUG_ARG("so = %lx", (long)so
);
443 if (so
->so_type
== IPPROTO_ICMP
) { /* This is a "ping" reply */
447 len
= recvfrom(so
->s
, buff
, 256, 0,
448 (struct sockaddr
*)&addr
, &addrlen
);
449 /* XXX Check if reply is "correct"? */
451 if(len
== -1 || len
== 0) {
452 u_char code
=ICMP_UNREACH_PORT
;
454 if(errno
== EHOSTUNREACH
) code
=ICMP_UNREACH_HOST
;
455 else if(errno
== ENETUNREACH
) code
=ICMP_UNREACH_NET
;
457 DEBUG_MISC((dfd
," udp icmp rx errno = %d-%s\n",
458 errno
,strerror(errno
)));
459 icmp_error(so
->so_m
, ICMP_UNREACH
,code
, 0,strerror(errno
));
461 icmp_reflect(so
->so_m
);
462 so
->so_m
= NULL
; /* Don't m_free() it again! */
464 /* No need for this socket anymore, udp_detach it */
466 } else { /* A "normal" UDP packet */
475 m
= m_get(so
->slirp
);
479 m
->m_data
+= IF_MAXLINKHDR
;
482 * XXX Shouldn't FIONREAD packets destined for port 53,
483 * but I don't know the max packet size for DNS lookups
486 /* if (so->so_fport != htons(53)) { */
487 ioctlsocket(so
->s
, FIONREAD
, &n
);
490 n
= (m
->m_data
- m
->m_dat
) + m
->m_len
+ n
+ 1;
496 m
->m_len
= recvfrom(so
->s
, m
->m_data
, len
, 0,
497 (struct sockaddr
*)&addr
, &addrlen
);
498 DEBUG_MISC((dfd
, " did recvfrom %d, errno = %d-%s\n",
499 m
->m_len
, errno
,strerror(errno
)));
501 u_char code
=ICMP_UNREACH_PORT
;
503 if(errno
== EHOSTUNREACH
) code
=ICMP_UNREACH_HOST
;
504 else if(errno
== ENETUNREACH
) code
=ICMP_UNREACH_NET
;
506 DEBUG_MISC((dfd
," rx error, tx icmp ICMP_UNREACH:%i\n", code
));
507 icmp_error(so
->so_m
, ICMP_UNREACH
,code
, 0,strerror(errno
));
511 * Hack: domain name lookup will be used the most for UDP,
512 * and since they'll only be used once there's no need
513 * for the 4 minute (or whatever) timeout... So we time them
514 * out much quicker (10 seconds for now...)
517 if (so
->so_fport
== htons(53))
518 so
->so_expire
= curtime
+ SO_EXPIREFAST
;
520 so
->so_expire
= curtime
+ SO_EXPIRE
;
524 * If this packet was destined for CTL_ADDR,
525 * make it look like that's where it came from, done by udp_output
527 udp_output(so
, m
, &addr
);
529 } /* if ping packet */
536 sosendto(struct socket
*so
, struct mbuf
*m
)
538 Slirp
*slirp
= so
->slirp
;
540 struct sockaddr_in addr
;
542 DEBUG_CALL("sosendto");
543 DEBUG_ARG("so = %lx", (long)so
);
544 DEBUG_ARG("m = %lx", (long)m
);
546 addr
.sin_family
= AF_INET
;
547 if ((so
->so_faddr
.s_addr
& slirp
->vnetwork_mask
.s_addr
) ==
548 slirp
->vnetwork_addr
.s_addr
) {
550 if (so
->so_faddr
.s_addr
== slirp
->vnameserver_addr
.s_addr
) {
551 if (get_dns_addr(&addr
.sin_addr
) < 0)
552 addr
.sin_addr
= loopback_addr
;
554 addr
.sin_addr
= loopback_addr
;
557 addr
.sin_addr
= so
->so_faddr
;
558 addr
.sin_port
= so
->so_fport
;
560 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
)));
562 /* Don't care what port we get */
563 ret
= sendto(so
->s
, m
->m_data
, m
->m_len
, 0,
564 (struct sockaddr
*)&addr
, sizeof (struct sockaddr
));
569 * Kill the socket if there's no reply in 4 minutes,
570 * but only if it's an expirable socket
573 so
->so_expire
= curtime
+ SO_EXPIRE
;
574 so
->so_state
&= SS_PERSISTENT_MASK
;
575 so
->so_state
|= SS_ISFCONNECTED
; /* So that it gets select()ed */
580 * Listen for incoming TCP connections
583 tcp_listen(Slirp
*slirp
, uint32_t haddr
, u_int hport
, uint32_t laddr
,
584 u_int lport
, int flags
)
586 struct sockaddr_in addr
;
589 socklen_t addrlen
= sizeof(addr
);
590 memset(&addr
, 0, addrlen
);
592 DEBUG_CALL("tcp_listen");
593 DEBUG_ARG("haddr = %x", haddr
);
594 DEBUG_ARG("hport = %d", hport
);
595 DEBUG_ARG("laddr = %x", laddr
);
596 DEBUG_ARG("lport = %d", lport
);
597 DEBUG_ARG("flags = %x", flags
);
599 so
= socreate(slirp
);
604 /* Don't tcp_attach... we don't need so_snd nor so_rcv */
605 if ((so
->so_tcpcb
= tcp_newtcpcb(so
)) == NULL
) {
609 insque(so
, &slirp
->tcb
);
612 * SS_FACCEPTONCE sockets must time out.
614 if (flags
& SS_FACCEPTONCE
)
615 so
->so_tcpcb
->t_timer
[TCPT_KEEP
] = TCPTV_KEEP_INIT
*2;
617 so
->so_state
&= SS_PERSISTENT_MASK
;
618 so
->so_state
|= (SS_FACCEPTCONN
| flags
);
619 so
->so_lport
= lport
; /* Kept in network format */
620 so
->so_laddr
.s_addr
= laddr
; /* Ditto */
622 addr
.sin_family
= AF_INET
;
623 addr
.sin_addr
.s_addr
= haddr
;
624 addr
.sin_port
= hport
;
626 if (((s
= qemu_socket(AF_INET
,SOCK_STREAM
,0)) < 0) ||
627 (setsockopt(s
,SOL_SOCKET
,SO_REUSEADDR
,(char *)&opt
,sizeof(int)) < 0) ||
628 (bind(s
,(struct sockaddr
*)&addr
, sizeof(addr
)) < 0) ||
630 int tmperrno
= errno
; /* Don't clobber the real reason we failed */
634 /* Restore the real errno */
636 WSASetLastError(tmperrno
);
642 setsockopt(s
,SOL_SOCKET
,SO_OOBINLINE
,(char *)&opt
,sizeof(int));
644 getsockname(s
,(struct sockaddr
*)&addr
,&addrlen
);
645 so
->so_fport
= addr
.sin_port
;
646 if (addr
.sin_addr
.s_addr
== 0 || addr
.sin_addr
.s_addr
== loopback_addr
.s_addr
)
647 so
->so_faddr
= slirp
->vhost_addr
;
649 so
->so_faddr
= addr
.sin_addr
;
656 * Various session state calls
657 * XXX Should be #define's
658 * The socket state stuff needs work, these often get call 2 or 3
659 * times each when only 1 was needed
662 soisfconnecting(struct socket
*so
)
664 so
->so_state
&= ~(SS_NOFDREF
|SS_ISFCONNECTED
|SS_FCANTRCVMORE
|
665 SS_FCANTSENDMORE
|SS_FWDRAIN
);
666 so
->so_state
|= SS_ISFCONNECTING
; /* Clobber other states */
670 soisfconnected(struct socket
*so
)
672 so
->so_state
&= ~(SS_ISFCONNECTING
|SS_FWDRAIN
|SS_NOFDREF
);
673 so
->so_state
|= SS_ISFCONNECTED
; /* Clobber other states */
677 sofcantrcvmore(struct socket
*so
)
679 if ((so
->so_state
& SS_NOFDREF
) == 0) {
681 if(global_writefds
) {
682 FD_CLR(so
->s
,global_writefds
);
685 so
->so_state
&= ~(SS_ISFCONNECTING
);
686 if (so
->so_state
& SS_FCANTSENDMORE
) {
687 so
->so_state
&= SS_PERSISTENT_MASK
;
688 so
->so_state
|= SS_NOFDREF
; /* Don't select it */
690 so
->so_state
|= SS_FCANTRCVMORE
;
695 sofcantsendmore(struct socket
*so
)
697 if ((so
->so_state
& SS_NOFDREF
) == 0) {
698 shutdown(so
->s
,1); /* send FIN to fhost */
699 if (global_readfds
) {
700 FD_CLR(so
->s
,global_readfds
);
703 FD_CLR(so
->s
,global_xfds
);
706 so
->so_state
&= ~(SS_ISFCONNECTING
);
707 if (so
->so_state
& SS_FCANTRCVMORE
) {
708 so
->so_state
&= SS_PERSISTENT_MASK
;
709 so
->so_state
|= SS_NOFDREF
; /* as above */
711 so
->so_state
|= SS_FCANTSENDMORE
;
716 * Set write drain mode
717 * Set CANTSENDMORE once all data has been write()n
720 sofwdrain(struct socket
*so
)
722 if (so
->so_rcv
.sb_cc
)
723 so
->so_state
|= SS_FWDRAIN
;