2 * Copyright (c) 1995 Danny Gasparovski.
4 * Please read the file COPYRIGHT for the
5 * terms and conditions of the copyright.
8 #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 */
95 * Read from so's socket into sb_snd, updating all relevant sbuf fields
96 * NOTE: This will only be called if it is select()ed for reading, so
97 * a read() of 0 (or less) means it's disconnected
103 int n
, nn
, lss
, total
;
104 struct sbuf
*sb
= &so
->so_snd
;
105 int len
= sb
->sb_datalen
- sb
->sb_cc
;
107 int mss
= so
->so_tcpcb
->t_maxseg
;
109 DEBUG_CALL("soread");
110 DEBUG_ARG("so = %lx", (long )so
);
113 * No need to check if there's enough room to read.
114 * soread wouldn't have been called if there weren't
117 len
= sb
->sb_datalen
- sb
->sb_cc
;
119 iov
[0].iov_base
= sb
->sb_wptr
;
120 if (sb
->sb_wptr
< sb
->sb_rptr
) {
121 iov
[0].iov_len
= sb
->sb_rptr
- sb
->sb_wptr
;
122 /* Should never succeed, but... */
123 if (iov
[0].iov_len
> len
)
124 iov
[0].iov_len
= len
;
125 if (iov
[0].iov_len
> mss
)
126 iov
[0].iov_len
-= iov
[0].iov_len
%mss
;
129 iov
[0].iov_len
= (sb
->sb_data
+ sb
->sb_datalen
) - sb
->sb_wptr
;
130 /* Should never succeed, but... */
131 if (iov
[0].iov_len
> len
) iov
[0].iov_len
= len
;
132 len
-= iov
[0].iov_len
;
134 iov
[1].iov_base
= sb
->sb_data
;
135 iov
[1].iov_len
= sb
->sb_rptr
- sb
->sb_data
;
136 if(iov
[1].iov_len
> len
)
137 iov
[1].iov_len
= len
;
138 total
= iov
[0].iov_len
+ iov
[1].iov_len
;
141 if (iov
[1].iov_len
> lss
) {
142 iov
[1].iov_len
-= lss
;
145 lss
-= iov
[1].iov_len
;
146 iov
[0].iov_len
-= lss
;
152 if (iov
[0].iov_len
> mss
)
153 iov
[0].iov_len
-= iov
[0].iov_len
%mss
;
159 nn
= readv(so
->s
, (struct iovec
*)iov
, n
);
160 DEBUG_MISC((dfd
, " ... read nn = %d bytes\n", nn
));
162 nn
= recv(so
->s
, iov
[0].iov_base
, iov
[0].iov_len
,0);
165 if (nn
< 0 && (errno
== EINTR
|| errno
== EAGAIN
))
168 DEBUG_MISC((dfd
, " --- soread() disconnected, nn = %d, errno = %d-%s\n", nn
, errno
,strerror(errno
)));
170 tcp_sockclosed(sototcpcb(so
));
177 * If there was no error, try and read the second time round
178 * We read again if n = 2 (ie, there's another part of the buffer)
179 * and we read as much as we could in the first read
180 * We don't test for <= 0 this time, because there legitimately
181 * might not be any more data (since the socket is non-blocking),
182 * a close will be detected on next iteration.
183 * A return of -1 wont (shouldn't) happen, since it didn't happen above
185 if (n
== 2 && nn
== iov
[0].iov_len
) {
187 ret
= recv(so
->s
, iov
[1].iov_base
, iov
[1].iov_len
,0);
192 DEBUG_MISC((dfd
, " ... read nn = %d bytes\n", nn
));
198 if (sb
->sb_wptr
>= (sb
->sb_data
+ sb
->sb_datalen
))
199 sb
->sb_wptr
-= sb
->sb_datalen
;
206 * When the socket is created, we set it SO_OOBINLINE,
207 * so when OOB data arrives, we soread() it and everything
208 * in the send buffer is sent as urgent data
214 struct tcpcb
*tp
= sototcpcb(so
);
216 DEBUG_CALL("sorecvoob");
217 DEBUG_ARG("so = %lx", (long)so
);
220 * We take a guess at how much urgent data has arrived.
221 * In most situations, when urgent data arrives, the next
222 * read() should get all the urgent data. This guess will
223 * be wrong however if more data arrives just after the
224 * urgent data, or the read() doesn't return all the
228 tp
->snd_up
= tp
->snd_una
+ so
->so_snd
.sb_cc
;
236 * There's a lot duplicated code here, but...
242 struct sbuf
*sb
= &so
->so_rcv
;
243 char buff
[2048]; /* XXX Shouldn't be sending more oob data than this */
247 DEBUG_CALL("sosendoob");
248 DEBUG_ARG("so = %lx", (long)so
);
249 DEBUG_ARG("sb->sb_cc = %d", sb
->sb_cc
);
251 if (so
->so_urgc
> 2048)
252 so
->so_urgc
= 2048; /* XXXX */
254 if (sb
->sb_rptr
< sb
->sb_wptr
) {
255 /* We can send it directly */
256 n
= send(so
->s
, sb
->sb_rptr
, so
->so_urgc
, (MSG_OOB
)); /* |MSG_DONTWAIT)); */
259 DEBUG_MISC((dfd
, " --- sent %d bytes urgent data, %d urgent bytes left\n", n
, so
->so_urgc
));
262 * Since there's no sendv or sendtov like writev,
263 * we must copy all data to a linear buffer then
266 len
= (sb
->sb_data
+ sb
->sb_datalen
) - sb
->sb_rptr
;
267 if (len
> so
->so_urgc
) len
= so
->so_urgc
;
268 memcpy(buff
, sb
->sb_rptr
, len
);
271 n
= sb
->sb_wptr
- sb
->sb_data
;
272 if (n
> so
->so_urgc
) n
= so
->so_urgc
;
273 memcpy((buff
+ len
), sb
->sb_data
, n
);
277 n
= send(so
->s
, buff
, len
, (MSG_OOB
)); /* |MSG_DONTWAIT)); */
280 DEBUG_ERROR((dfd
, "Didn't send all data urgently XXXXX\n"));
282 DEBUG_MISC((dfd
, " ---2 sent %d bytes urgent data, %d urgent bytes left\n", n
, so
->so_urgc
));
287 if (sb
->sb_rptr
>= (sb
->sb_data
+ sb
->sb_datalen
))
288 sb
->sb_rptr
-= sb
->sb_datalen
;
294 * Write data from so_rcv to so's socket,
295 * updating all sbuf field as necessary
302 struct sbuf
*sb
= &so
->so_rcv
;
306 DEBUG_CALL("sowrite");
307 DEBUG_ARG("so = %lx", (long)so
);
316 * No need to check if there's something to write,
317 * sowrite wouldn't have been called otherwise
322 iov
[0].iov_base
= sb
->sb_rptr
;
323 if (sb
->sb_rptr
< sb
->sb_wptr
) {
324 iov
[0].iov_len
= sb
->sb_wptr
- sb
->sb_rptr
;
325 /* Should never succeed, but... */
326 if (iov
[0].iov_len
> len
) iov
[0].iov_len
= len
;
329 iov
[0].iov_len
= (sb
->sb_data
+ sb
->sb_datalen
) - sb
->sb_rptr
;
330 if (iov
[0].iov_len
> len
) iov
[0].iov_len
= len
;
331 len
-= iov
[0].iov_len
;
333 iov
[1].iov_base
= sb
->sb_data
;
334 iov
[1].iov_len
= sb
->sb_wptr
- sb
->sb_data
;
335 if (iov
[1].iov_len
> len
) iov
[1].iov_len
= len
;
340 /* Check if there's urgent data to send, and if so, send it */
343 nn
= writev(so
->s
, (const struct iovec
*)iov
, n
);
345 DEBUG_MISC((dfd
, " ... wrote nn = %d bytes\n", nn
));
347 nn
= send(so
->s
, iov
[0].iov_base
, iov
[0].iov_len
,0);
349 /* This should never happen, but people tell me it does *shrug* */
350 if (nn
< 0 && (errno
== EAGAIN
|| errno
== EINTR
))
354 DEBUG_MISC((dfd
, " --- sowrite disconnected, so->so_state = %x, errno = %d\n",
355 so
->so_state
, errno
));
357 tcp_sockclosed(sototcpcb(so
));
362 if (n
== 2 && nn
== iov
[0].iov_len
) {
364 ret
= send(so
->s
, iov
[1].iov_base
, iov
[1].iov_len
,0);
368 DEBUG_MISC((dfd
, " ... wrote nn = %d bytes\n", nn
));
374 if (sb
->sb_rptr
>= (sb
->sb_data
+ sb
->sb_datalen
))
375 sb
->sb_rptr
-= sb
->sb_datalen
;
378 * If in DRAIN mode, and there's no more data, set
381 if ((so
->so_state
& SS_FWDRAIN
) && sb
->sb_cc
== 0)
388 * recvfrom() a UDP socket
394 struct sockaddr_in addr
;
395 int addrlen
= sizeof(struct sockaddr_in
);
397 DEBUG_CALL("sorecvfrom");
398 DEBUG_ARG("so = %lx", (long)so
);
400 if (so
->so_type
== IPPROTO_ICMP
) { /* This is a "ping" reply */
404 len
= recvfrom(so
->s
, buff
, 256, 0,
405 (struct sockaddr
*)&addr
, &addrlen
);
406 /* XXX Check if reply is "correct"? */
408 if(len
== -1 || len
== 0) {
409 u_char code
=ICMP_UNREACH_PORT
;
411 if(errno
== EHOSTUNREACH
) code
=ICMP_UNREACH_HOST
;
412 else if(errno
== ENETUNREACH
) code
=ICMP_UNREACH_NET
;
414 DEBUG_MISC((dfd
," udp icmp rx errno = %d-%s\n",
415 errno
,strerror(errno
)));
416 icmp_error(so
->so_m
, ICMP_UNREACH
,code
, 0,strerror(errno
));
418 icmp_reflect(so
->so_m
);
419 so
->so_m
= 0; /* Don't m_free() it again! */
421 /* No need for this socket anymore, udp_detach it */
423 } else { /* A "normal" UDP packet */
427 if (!(m
= m_get())) return;
428 m
->m_data
+= IF_MAXLINKHDR
;
431 * XXX Shouldn't FIONREAD packets destined for port 53,
432 * but I don't know the max packet size for DNS lookups
435 /* if (so->so_fport != htons(53)) { */
436 ioctlsocket(so
->s
, FIONREAD
, &n
);
439 n
= (m
->m_data
- m
->m_dat
) + m
->m_len
+ n
+ 1;
445 m
->m_len
= recvfrom(so
->s
, m
->m_data
, len
, 0,
446 (struct sockaddr
*)&addr
, &addrlen
);
447 DEBUG_MISC((dfd
, " did recvfrom %d, errno = %d-%s\n",
448 m
->m_len
, errno
,strerror(errno
)));
450 u_char code
=ICMP_UNREACH_PORT
;
452 if(errno
== EHOSTUNREACH
) code
=ICMP_UNREACH_HOST
;
453 else if(errno
== ENETUNREACH
) code
=ICMP_UNREACH_NET
;
455 DEBUG_MISC((dfd
," rx error, tx icmp ICMP_UNREACH:%i\n", code
));
456 icmp_error(so
->so_m
, ICMP_UNREACH
,code
, 0,strerror(errno
));
460 * Hack: domain name lookup will be used the most for UDP,
461 * and since they'll only be used once there's no need
462 * for the 4 minute (or whatever) timeout... So we time them
463 * out much quicker (10 seconds for now...)
466 if (so
->so_fport
== htons(53))
467 so
->so_expire
= curtime
+ SO_EXPIREFAST
;
469 so
->so_expire
= curtime
+ SO_EXPIRE
;
472 /* if (m->m_len == len) {
473 * m_inc(m, MINCSIZE);
479 * If this packet was destined for CTL_ADDR,
480 * make it look like that's where it came from, done by udp_output
482 udp_output(so
, m
, &addr
);
484 } /* if ping packet */
496 struct sockaddr_in addr
;
498 DEBUG_CALL("sosendto");
499 DEBUG_ARG("so = %lx", (long)so
);
500 DEBUG_ARG("m = %lx", (long)m
);
502 addr
.sin_family
= AF_INET
;
503 if ((so
->so_faddr
.s_addr
& htonl(0xffffff00)) == special_addr
.s_addr
) {
505 switch(ntohl(so
->so_faddr
.s_addr
) & 0xff) {
507 addr
.sin_addr
= dns_addr
;
511 addr
.sin_addr
= loopback_addr
;
515 addr
.sin_addr
= so
->so_faddr
;
516 addr
.sin_port
= so
->so_fport
;
518 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
)));
520 /* Don't care what port we get */
521 ret
= sendto(so
->s
, m
->m_data
, m
->m_len
, 0,
522 (struct sockaddr
*)&addr
, sizeof (struct sockaddr
));
527 * Kill the socket if there's no reply in 4 minutes,
528 * but only if it's an expirable socket
531 so
->so_expire
= curtime
+ SO_EXPIRE
;
532 so
->so_state
= SS_ISFCONNECTED
; /* So that it gets select()ed */
537 * XXX This should really be tcp_listen
540 solisten(port
, laddr
, lport
, flags
)
546 struct sockaddr_in addr
;
548 int s
, addrlen
= sizeof(addr
), opt
= 1;
550 DEBUG_CALL("solisten");
551 DEBUG_ARG("port = %d", port
);
552 DEBUG_ARG("laddr = %x", laddr
);
553 DEBUG_ARG("lport = %d", lport
);
554 DEBUG_ARG("flags = %x", flags
);
556 if ((so
= socreate()) == NULL
) {
557 /* free(so); Not sofree() ??? free(NULL) == NOP */
561 /* Don't tcp_attach... we don't need so_snd nor so_rcv */
562 if ((so
->so_tcpcb
= tcp_newtcpcb(so
)) == NULL
) {
569 * SS_FACCEPTONCE sockets must time out.
571 if (flags
& SS_FACCEPTONCE
)
572 so
->so_tcpcb
->t_timer
[TCPT_KEEP
] = TCPTV_KEEP_INIT
*2;
574 so
->so_state
= (SS_FACCEPTCONN
|flags
);
575 so
->so_lport
= lport
; /* Kept in network format */
576 so
->so_laddr
.s_addr
= laddr
; /* Ditto */
578 addr
.sin_family
= AF_INET
;
579 addr
.sin_addr
.s_addr
= INADDR_ANY
;
580 addr
.sin_port
= port
;
582 if (((s
= socket(AF_INET
,SOCK_STREAM
,0)) < 0) ||
583 (setsockopt(s
,SOL_SOCKET
,SO_REUSEADDR
,(char *)&opt
,sizeof(int)) < 0) ||
584 (bind(s
,(struct sockaddr
*)&addr
, sizeof(addr
)) < 0) ||
586 int tmperrno
= errno
; /* Don't clobber the real reason we failed */
590 /* Restore the real errno */
592 WSASetLastError(tmperrno
);
598 setsockopt(s
,SOL_SOCKET
,SO_OOBINLINE
,(char *)&opt
,sizeof(int));
600 getsockname(s
,(struct sockaddr
*)&addr
,&addrlen
);
601 so
->so_fport
= addr
.sin_port
;
602 if (addr
.sin_addr
.s_addr
== 0 || addr
.sin_addr
.s_addr
== loopback_addr
.s_addr
)
603 so
->so_faddr
= alias_addr
;
605 so
->so_faddr
= addr
.sin_addr
;
613 * Data is available in so_rcv
614 * Just write() the data to the socket
622 /* FD_CLR(so->s,&writefds); */
626 * Data has been freed in so_snd
627 * We have room for a read() if we want to
628 * For now, don't read, it'll be done in the main loop
639 * Various session state calls
640 * XXX Should be #define's
641 * The socket state stuff needs work, these often get call 2 or 3
642 * times each when only 1 was needed
646 register struct socket
*so
;
648 so
->so_state
&= ~(SS_NOFDREF
|SS_ISFCONNECTED
|SS_FCANTRCVMORE
|
649 SS_FCANTSENDMORE
|SS_FWDRAIN
);
650 so
->so_state
|= SS_ISFCONNECTING
; /* Clobber other states */
655 register struct socket
*so
;
657 so
->so_state
&= ~(SS_ISFCONNECTING
|SS_FWDRAIN
|SS_NOFDREF
);
658 so
->so_state
|= SS_ISFCONNECTED
; /* Clobber other states */
662 sofcantrcvmore(struct socket
*so
)
664 if ((so
->so_state
& SS_NOFDREF
) == 0) {
666 if(global_writefds
) {
667 FD_CLR(so
->s
,global_writefds
);
670 so
->so_state
&= ~(SS_ISFCONNECTING
);
671 if (so
->so_state
& SS_FCANTSENDMORE
)
672 so
->so_state
= SS_NOFDREF
; /* Don't select it */ /* XXX close() here as well? */
674 so
->so_state
|= SS_FCANTRCVMORE
;
678 sofcantsendmore(struct socket
*so
)
680 if ((so
->so_state
& SS_NOFDREF
) == 0) {
681 shutdown(so
->s
,1); /* send FIN to fhost */
682 if (global_readfds
) {
683 FD_CLR(so
->s
,global_readfds
);
686 FD_CLR(so
->s
,global_xfds
);
689 so
->so_state
&= ~(SS_ISFCONNECTING
);
690 if (so
->so_state
& SS_FCANTRCVMORE
)
691 so
->so_state
= SS_NOFDREF
; /* as above */
693 so
->so_state
|= SS_FCANTSENDMORE
;
697 soisfdisconnected(so
)
700 /* so->so_state &= ~(SS_ISFCONNECTING|SS_ISFCONNECTED); */
702 /* so->so_state = SS_ISFDISCONNECTED; */
704 * XXX Do nothing ... ?
709 * Set write drain mode
710 * Set CANTSENDMORE once all data has been write()n
716 if (so
->so_rcv
.sb_cc
)
717 so
->so_state
|= SS_FWDRAIN
;