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(struct socket
*head
, struct in_addr laddr
, u_int lport
,
29 struct in_addr faddr
, u_int fport
)
33 for (so
= head
->so_next
; so
!= head
; so
= so
->so_next
) {
34 if (so
->so_lport
== lport
&&
35 so
->so_laddr
.s_addr
== laddr
.s_addr
&&
36 so
->so_faddr
.s_addr
== faddr
.s_addr
&&
37 so
->so_fport
== fport
)
42 return (struct socket
*)NULL
;
48 * Create a new socket, initialise the fields
49 * It is the responsibility of the caller to
50 * insque() it into the correct linked-list
57 so
= (struct socket
*)malloc(sizeof(struct socket
));
59 memset(so
, 0, sizeof(struct socket
));
60 so
->so_state
= SS_NOFDREF
;
67 * remque and free a socket, clobber cache
70 sofree(struct socket
*so
)
72 if (so
->so_emu
==EMU_RSH
&& so
->extra
) {
76 if (so
== tcp_last_so
)
78 else if (so
== udp_last_so
)
83 if(so
->so_next
&& so
->so_prev
)
84 remque(so
); /* crashes if so is not in a queue */
89 size_t sopreprbuf(struct socket
*so
, struct iovec
*iov
, int *np
)
92 struct sbuf
*sb
= &so
->so_snd
;
93 int len
= sb
->sb_datalen
- sb
->sb_cc
;
94 int mss
= so
->so_tcpcb
->t_maxseg
;
96 DEBUG_CALL("sopreprbuf");
97 DEBUG_ARG("so = %lx", (long )so
);
99 len
= sb
->sb_datalen
- sb
->sb_cc
;
104 iov
[0].iov_base
= sb
->sb_wptr
;
105 iov
[1].iov_base
= NULL
;
107 if (sb
->sb_wptr
< sb
->sb_rptr
) {
108 iov
[0].iov_len
= sb
->sb_rptr
- sb
->sb_wptr
;
109 /* Should never succeed, but... */
110 if (iov
[0].iov_len
> len
)
111 iov
[0].iov_len
= len
;
112 if (iov
[0].iov_len
> mss
)
113 iov
[0].iov_len
-= iov
[0].iov_len
%mss
;
116 iov
[0].iov_len
= (sb
->sb_data
+ sb
->sb_datalen
) - sb
->sb_wptr
;
117 /* Should never succeed, but... */
118 if (iov
[0].iov_len
> len
) iov
[0].iov_len
= len
;
119 len
-= iov
[0].iov_len
;
121 iov
[1].iov_base
= sb
->sb_data
;
122 iov
[1].iov_len
= sb
->sb_rptr
- sb
->sb_data
;
123 if(iov
[1].iov_len
> len
)
124 iov
[1].iov_len
= len
;
125 total
= iov
[0].iov_len
+ iov
[1].iov_len
;
128 if (iov
[1].iov_len
> lss
) {
129 iov
[1].iov_len
-= lss
;
132 lss
-= iov
[1].iov_len
;
133 iov
[0].iov_len
-= lss
;
139 if (iov
[0].iov_len
> mss
)
140 iov
[0].iov_len
-= iov
[0].iov_len
%mss
;
147 return iov
[0].iov_len
+ (n
- 1) * iov
[1].iov_len
;
151 * Read from so's socket into sb_snd, updating all relevant sbuf fields
152 * NOTE: This will only be called if it is select()ed for reading, so
153 * a read() of 0 (or less) means it's disconnected
156 soread(struct socket
*so
)
159 struct sbuf
*sb
= &so
->so_snd
;
162 DEBUG_CALL("soread");
163 DEBUG_ARG("so = %lx", (long )so
);
166 * No need to check if there's enough room to read.
167 * soread wouldn't have been called if there weren't
169 sopreprbuf(so
, iov
, &n
);
172 nn
= readv(so
->s
, (struct iovec
*)iov
, n
);
173 DEBUG_MISC((dfd
, " ... read nn = %d bytes\n", nn
));
175 nn
= recv(so
->s
, iov
[0].iov_base
, iov
[0].iov_len
,0);
178 if (nn
< 0 && (errno
== EINTR
|| errno
== EAGAIN
))
181 DEBUG_MISC((dfd
, " --- soread() disconnected, nn = %d, errno = %d-%s\n", nn
, errno
,strerror(errno
)));
183 tcp_sockclosed(sototcpcb(so
));
190 * If there was no error, try and read the second time round
191 * We read again if n = 2 (ie, there's another part of the buffer)
192 * and we read as much as we could in the first read
193 * We don't test for <= 0 this time, because there legitimately
194 * might not be any more data (since the socket is non-blocking),
195 * a close will be detected on next iteration.
196 * A return of -1 wont (shouldn't) happen, since it didn't happen above
198 if (n
== 2 && nn
== iov
[0].iov_len
) {
200 ret
= recv(so
->s
, iov
[1].iov_base
, iov
[1].iov_len
,0);
205 DEBUG_MISC((dfd
, " ... read nn = %d bytes\n", nn
));
211 if (sb
->sb_wptr
>= (sb
->sb_data
+ sb
->sb_datalen
))
212 sb
->sb_wptr
-= sb
->sb_datalen
;
216 int soreadbuf(struct socket
*so
, const char *buf
, int size
)
218 int n
, nn
, copy
= size
;
219 struct sbuf
*sb
= &so
->so_snd
;
222 DEBUG_CALL("soreadbuf");
223 DEBUG_ARG("so = %lx", (long )so
);
226 * No need to check if there's enough room to read.
227 * soread wouldn't have been called if there weren't
229 if (sopreprbuf(so
, iov
, &n
) < size
)
232 nn
= MIN(iov
[0].iov_len
, copy
);
233 memcpy(iov
[0].iov_base
, buf
, nn
);
241 memcpy(iov
[1].iov_base
, buf
, copy
);
247 if (sb
->sb_wptr
>= (sb
->sb_data
+ sb
->sb_datalen
))
248 sb
->sb_wptr
-= sb
->sb_datalen
;
253 tcp_sockclosed(sototcpcb(so
));
254 fprintf(stderr
, "soreadbuf buffer to small");
261 * When the socket is created, we set it SO_OOBINLINE,
262 * so when OOB data arrives, we soread() it and everything
263 * in the send buffer is sent as urgent data
266 sorecvoob(struct socket
*so
)
268 struct tcpcb
*tp
= sototcpcb(so
);
270 DEBUG_CALL("sorecvoob");
271 DEBUG_ARG("so = %lx", (long)so
);
274 * We take a guess at how much urgent data has arrived.
275 * In most situations, when urgent data arrives, the next
276 * read() should get all the urgent data. This guess will
277 * be wrong however if more data arrives just after the
278 * urgent data, or the read() doesn't return all the
282 tp
->snd_up
= tp
->snd_una
+ so
->so_snd
.sb_cc
;
290 * There's a lot duplicated code here, but...
293 sosendoob(struct socket
*so
)
295 struct sbuf
*sb
= &so
->so_rcv
;
296 char buff
[2048]; /* XXX Shouldn't be sending more oob data than this */
300 DEBUG_CALL("sosendoob");
301 DEBUG_ARG("so = %lx", (long)so
);
302 DEBUG_ARG("sb->sb_cc = %d", sb
->sb_cc
);
304 if (so
->so_urgc
> 2048)
305 so
->so_urgc
= 2048; /* XXXX */
307 if (sb
->sb_rptr
< sb
->sb_wptr
) {
308 /* We can send it directly */
309 n
= slirp_send(so
, sb
->sb_rptr
, so
->so_urgc
, (MSG_OOB
)); /* |MSG_DONTWAIT)); */
312 DEBUG_MISC((dfd
, " --- sent %d bytes urgent data, %d urgent bytes left\n", n
, so
->so_urgc
));
315 * Since there's no sendv or sendtov like writev,
316 * we must copy all data to a linear buffer then
319 len
= (sb
->sb_data
+ sb
->sb_datalen
) - sb
->sb_rptr
;
320 if (len
> so
->so_urgc
) len
= so
->so_urgc
;
321 memcpy(buff
, sb
->sb_rptr
, len
);
324 n
= sb
->sb_wptr
- sb
->sb_data
;
325 if (n
> so
->so_urgc
) n
= so
->so_urgc
;
326 memcpy((buff
+ len
), sb
->sb_data
, n
);
330 n
= slirp_send(so
, buff
, len
, (MSG_OOB
)); /* |MSG_DONTWAIT)); */
333 DEBUG_ERROR((dfd
, "Didn't send all data urgently XXXXX\n"));
335 DEBUG_MISC((dfd
, " ---2 sent %d bytes urgent data, %d urgent bytes left\n", n
, so
->so_urgc
));
340 if (sb
->sb_rptr
>= (sb
->sb_data
+ sb
->sb_datalen
))
341 sb
->sb_rptr
-= sb
->sb_datalen
;
347 * Write data from so_rcv to so's socket,
348 * updating all sbuf field as necessary
351 sowrite(struct socket
*so
)
354 struct sbuf
*sb
= &so
->so_rcv
;
358 DEBUG_CALL("sowrite");
359 DEBUG_ARG("so = %lx", (long)so
);
368 * No need to check if there's something to write,
369 * sowrite wouldn't have been called otherwise
374 iov
[0].iov_base
= sb
->sb_rptr
;
375 iov
[1].iov_base
= NULL
;
377 if (sb
->sb_rptr
< sb
->sb_wptr
) {
378 iov
[0].iov_len
= sb
->sb_wptr
- sb
->sb_rptr
;
379 /* Should never succeed, but... */
380 if (iov
[0].iov_len
> len
) iov
[0].iov_len
= len
;
383 iov
[0].iov_len
= (sb
->sb_data
+ sb
->sb_datalen
) - sb
->sb_rptr
;
384 if (iov
[0].iov_len
> len
) iov
[0].iov_len
= len
;
385 len
-= iov
[0].iov_len
;
387 iov
[1].iov_base
= sb
->sb_data
;
388 iov
[1].iov_len
= sb
->sb_wptr
- sb
->sb_data
;
389 if (iov
[1].iov_len
> len
) iov
[1].iov_len
= len
;
394 /* Check if there's urgent data to send, and if so, send it */
397 nn
= writev(so
->s
, (const struct iovec
*)iov
, n
);
399 DEBUG_MISC((dfd
, " ... wrote nn = %d bytes\n", nn
));
401 nn
= slirp_send(so
, iov
[0].iov_base
, iov
[0].iov_len
,0);
403 /* This should never happen, but people tell me it does *shrug* */
404 if (nn
< 0 && (errno
== EAGAIN
|| errno
== EINTR
))
408 DEBUG_MISC((dfd
, " --- sowrite disconnected, so->so_state = %x, errno = %d\n",
409 so
->so_state
, errno
));
411 tcp_sockclosed(sototcpcb(so
));
416 if (n
== 2 && nn
== iov
[0].iov_len
) {
418 ret
= slirp_send(so
, iov
[1].iov_base
, iov
[1].iov_len
,0);
422 DEBUG_MISC((dfd
, " ... wrote nn = %d bytes\n", nn
));
428 if (sb
->sb_rptr
>= (sb
->sb_data
+ sb
->sb_datalen
))
429 sb
->sb_rptr
-= sb
->sb_datalen
;
432 * If in DRAIN mode, and there's no more data, set
435 if ((so
->so_state
& SS_FWDRAIN
) && sb
->sb_cc
== 0)
442 * recvfrom() a UDP socket
445 sorecvfrom(struct socket
*so
)
447 struct sockaddr_in addr
;
448 socklen_t addrlen
= sizeof(struct sockaddr_in
);
450 DEBUG_CALL("sorecvfrom");
451 DEBUG_ARG("so = %lx", (long)so
);
453 if (so
->so_type
== IPPROTO_ICMP
) { /* This is a "ping" reply */
457 len
= recvfrom(so
->s
, buff
, 256, 0,
458 (struct sockaddr
*)&addr
, &addrlen
);
459 /* XXX Check if reply is "correct"? */
461 if(len
== -1 || len
== 0) {
462 u_char code
=ICMP_UNREACH_PORT
;
464 if(errno
== EHOSTUNREACH
) code
=ICMP_UNREACH_HOST
;
465 else if(errno
== ENETUNREACH
) code
=ICMP_UNREACH_NET
;
467 DEBUG_MISC((dfd
," udp icmp rx errno = %d-%s\n",
468 errno
,strerror(errno
)));
469 icmp_error(so
->so_m
, ICMP_UNREACH
,code
, 0,strerror(errno
));
471 icmp_reflect(so
->so_m
);
472 so
->so_m
= NULL
; /* Don't m_free() it again! */
474 /* No need for this socket anymore, udp_detach it */
476 } else { /* A "normal" UDP packet */
480 if (!(m
= m_get())) return;
481 m
->m_data
+= IF_MAXLINKHDR
;
484 * XXX Shouldn't FIONREAD packets destined for port 53,
485 * but I don't know the max packet size for DNS lookups
488 /* if (so->so_fport != htons(53)) { */
489 ioctlsocket(so
->s
, FIONREAD
, &n
);
492 n
= (m
->m_data
- m
->m_dat
) + m
->m_len
+ n
+ 1;
498 m
->m_len
= recvfrom(so
->s
, m
->m_data
, len
, 0,
499 (struct sockaddr
*)&addr
, &addrlen
);
500 DEBUG_MISC((dfd
, " did recvfrom %d, errno = %d-%s\n",
501 m
->m_len
, errno
,strerror(errno
)));
503 u_char code
=ICMP_UNREACH_PORT
;
505 if(errno
== EHOSTUNREACH
) code
=ICMP_UNREACH_HOST
;
506 else if(errno
== ENETUNREACH
) code
=ICMP_UNREACH_NET
;
508 DEBUG_MISC((dfd
," rx error, tx icmp ICMP_UNREACH:%i\n", code
));
509 icmp_error(so
->so_m
, ICMP_UNREACH
,code
, 0,strerror(errno
));
513 * Hack: domain name lookup will be used the most for UDP,
514 * and since they'll only be used once there's no need
515 * for the 4 minute (or whatever) timeout... So we time them
516 * out much quicker (10 seconds for now...)
519 if (so
->so_fport
== htons(53))
520 so
->so_expire
= curtime
+ SO_EXPIREFAST
;
522 so
->so_expire
= curtime
+ SO_EXPIRE
;
525 /* if (m->m_len == len) {
526 * m_inc(m, MINCSIZE);
532 * If this packet was destined for CTL_ADDR,
533 * make it look like that's where it came from, done by udp_output
535 udp_output(so
, m
, &addr
);
537 } /* if ping packet */
544 sosendto(struct socket
*so
, struct mbuf
*m
)
547 struct sockaddr_in addr
;
549 DEBUG_CALL("sosendto");
550 DEBUG_ARG("so = %lx", (long)so
);
551 DEBUG_ARG("m = %lx", (long)m
);
553 addr
.sin_family
= AF_INET
;
554 if ((so
->so_faddr
.s_addr
& htonl(0xffffff00)) == special_addr
.s_addr
) {
556 switch(ntohl(so
->so_faddr
.s_addr
) & 0xff) {
558 addr
.sin_addr
= dns_addr
;
562 addr
.sin_addr
= loopback_addr
;
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
));
578 * Kill the socket if there's no reply in 4 minutes,
579 * but only if it's an expirable socket
582 so
->so_expire
= curtime
+ SO_EXPIRE
;
583 so
->so_state
= SS_ISFCONNECTED
; /* So that it gets select()ed */
588 * XXX This should really be tcp_listen
591 solisten(u_int port
, u_int32_t laddr
, u_int lport
, int flags
)
593 struct sockaddr_in addr
;
596 socklen_t addrlen
= sizeof(addr
);
598 DEBUG_CALL("solisten");
599 DEBUG_ARG("port = %d", port
);
600 DEBUG_ARG("laddr = %x", laddr
);
601 DEBUG_ARG("lport = %d", lport
);
602 DEBUG_ARG("flags = %x", flags
);
604 if ((so
= socreate()) == NULL
) {
605 /* free(so); Not sofree() ??? free(NULL) == NOP */
609 /* Don't tcp_attach... we don't need so_snd nor so_rcv */
610 if ((so
->so_tcpcb
= tcp_newtcpcb(so
)) == NULL
) {
617 * SS_FACCEPTONCE sockets must time out.
619 if (flags
& SS_FACCEPTONCE
)
620 so
->so_tcpcb
->t_timer
[TCPT_KEEP
] = TCPTV_KEEP_INIT
*2;
622 so
->so_state
= (SS_FACCEPTCONN
|flags
);
623 so
->so_lport
= lport
; /* Kept in network format */
624 so
->so_laddr
.s_addr
= laddr
; /* Ditto */
626 addr
.sin_family
= AF_INET
;
627 addr
.sin_addr
.s_addr
= INADDR_ANY
;
628 addr
.sin_port
= port
;
630 if (((s
= socket(AF_INET
,SOCK_STREAM
,0)) < 0) ||
631 (setsockopt(s
,SOL_SOCKET
,SO_REUSEADDR
,(char *)&opt
,sizeof(int)) < 0) ||
632 (bind(s
,(struct sockaddr
*)&addr
, sizeof(addr
)) < 0) ||
634 int tmperrno
= errno
; /* Don't clobber the real reason we failed */
638 /* Restore the real errno */
640 WSASetLastError(tmperrno
);
646 setsockopt(s
,SOL_SOCKET
,SO_OOBINLINE
,(char *)&opt
,sizeof(int));
648 getsockname(s
,(struct sockaddr
*)&addr
,&addrlen
);
649 so
->so_fport
= addr
.sin_port
;
650 if (addr
.sin_addr
.s_addr
== 0 || addr
.sin_addr
.s_addr
== loopback_addr
.s_addr
)
651 so
->so_faddr
= alias_addr
;
653 so
->so_faddr
= addr
.sin_addr
;
661 * Data is available in so_rcv
662 * Just write() the data to the socket
670 /* FD_CLR(so->s,&writefds); */
674 * Data has been freed in so_snd
675 * We have room for a read() if we want to
676 * For now, don't read, it'll be done in the main loop
687 * Various session state calls
688 * XXX Should be #define's
689 * The socket state stuff needs work, these often get call 2 or 3
690 * times each when only 1 was needed
693 soisfconnecting(struct socket
*so
)
695 so
->so_state
&= ~(SS_NOFDREF
|SS_ISFCONNECTED
|SS_FCANTRCVMORE
|
696 SS_FCANTSENDMORE
|SS_FWDRAIN
);
697 so
->so_state
|= SS_ISFCONNECTING
; /* Clobber other states */
701 soisfconnected(struct socket
*so
)
703 so
->so_state
&= ~(SS_ISFCONNECTING
|SS_FWDRAIN
|SS_NOFDREF
);
704 so
->so_state
|= SS_ISFCONNECTED
; /* Clobber other states */
708 sofcantrcvmore(struct socket
*so
)
710 if ((so
->so_state
& SS_NOFDREF
) == 0) {
712 if(global_writefds
) {
713 FD_CLR(so
->s
,global_writefds
);
716 so
->so_state
&= ~(SS_ISFCONNECTING
);
717 if (so
->so_state
& SS_FCANTSENDMORE
)
718 so
->so_state
= SS_NOFDREF
; /* Don't select it */ /* XXX close() here as well? */
720 so
->so_state
|= SS_FCANTRCVMORE
;
724 sofcantsendmore(struct socket
*so
)
726 if ((so
->so_state
& SS_NOFDREF
) == 0) {
727 shutdown(so
->s
,1); /* send FIN to fhost */
728 if (global_readfds
) {
729 FD_CLR(so
->s
,global_readfds
);
732 FD_CLR(so
->s
,global_xfds
);
735 so
->so_state
&= ~(SS_ISFCONNECTING
);
736 if (so
->so_state
& SS_FCANTRCVMORE
)
737 so
->so_state
= SS_NOFDREF
; /* as above */
739 so
->so_state
|= SS_FCANTSENDMORE
;
743 soisfdisconnected(struct socket
*so
)
745 /* so->so_state &= ~(SS_ISFCONNECTING|SS_ISFCONNECTED); */
747 /* so->so_state = SS_ISFDISCONNECTED; */
749 * XXX Do nothing ... ?
754 * Set write drain mode
755 * Set CANTSENDMORE once all data has been write()n
758 sofwdrain(struct socket
*so
)
760 if (so
->so_rcv
.sb_cc
)
761 so
->so_state
|= SS_FWDRAIN
;