]> git.proxmox.com Git - qemu.git/blob - slirp/socket.c
initial user mode network support
[qemu.git] / slirp / socket.c
1 /*
2 * Copyright (c) 1995 Danny Gasparovski.
3 *
4 * Please read the file COPYRIGHT for the
5 * terms and conditions of the copyright.
6 */
7
8 #define WANT_SYS_IOCTL_H
9 #include <slirp.h>
10 #include "ip_icmp.h"
11 #include "main.h"
12
13 void
14 so_init()
15 {
16 /* Nothing yet */
17 }
18
19
20 struct socket *
21 solookup(head, laddr, lport, faddr, fport)
22 struct socket *head;
23 struct in_addr laddr;
24 u_int lport;
25 struct in_addr faddr;
26 u_int fport;
27 {
28 struct socket *so;
29
30 for (so = head->so_next; so != head; so = so->so_next) {
31 if (so->so_lport == lport &&
32 so->so_laddr.s_addr == laddr.s_addr &&
33 so->so_faddr.s_addr == faddr.s_addr &&
34 so->so_fport == fport)
35 break;
36 }
37
38 if (so == head)
39 return (struct socket *)NULL;
40 return so;
41
42 }
43
44 /*
45 * Create a new socket, initialise the fields
46 * It is the responsibility of the caller to
47 * insque() it into the correct linked-list
48 */
49 struct socket *
50 socreate()
51 {
52 struct socket *so;
53
54 so = (struct socket *)malloc(sizeof(struct socket));
55 if(so) {
56 memset(so, 0, sizeof(struct socket));
57 so->so_state = SS_NOFDREF;
58 so->s = -1;
59 }
60 return(so);
61 }
62
63 /*
64 * remque and free a socket, clobber cache
65 */
66 void
67 sofree(so)
68 struct socket *so;
69 {
70 if (so->so_emu==EMU_RSH && so->extra) {
71 sofree(so->extra);
72 so->extra=NULL;
73 }
74 if (so == tcp_last_so)
75 tcp_last_so = &tcb;
76 else if (so == udp_last_so)
77 udp_last_so = &udb;
78
79 m_free(so->so_m);
80
81 if(so->so_next && so->so_prev)
82 remque(so); /* crashes if so is not in a queue */
83
84 free(so);
85 }
86
87 /*
88 * Read from so's socket into sb_snd, updating all relevant sbuf fields
89 * NOTE: This will only be called if it is select()ed for reading, so
90 * a read() of 0 (or less) means it's disconnected
91 */
92 int
93 soread(so)
94 struct socket *so;
95 {
96 int n, nn, lss, total;
97 struct sbuf *sb = &so->so_snd;
98 int len = sb->sb_datalen - sb->sb_cc;
99 struct iovec iov[2];
100 int mss = so->so_tcpcb->t_maxseg;
101
102 DEBUG_CALL("soread");
103 DEBUG_ARG("so = %lx", (long )so);
104
105 /*
106 * No need to check if there's enough room to read.
107 * soread wouldn't have been called if there weren't
108 */
109
110 len = sb->sb_datalen - sb->sb_cc;
111
112 iov[0].iov_base = sb->sb_wptr;
113 if (sb->sb_wptr < sb->sb_rptr) {
114 iov[0].iov_len = sb->sb_rptr - sb->sb_wptr;
115 /* Should never succeed, but... */
116 if (iov[0].iov_len > len)
117 iov[0].iov_len = len;
118 if (iov[0].iov_len > mss)
119 iov[0].iov_len -= iov[0].iov_len%mss;
120 n = 1;
121 } else {
122 iov[0].iov_len = (sb->sb_data + sb->sb_datalen) - sb->sb_wptr;
123 /* Should never succeed, but... */
124 if (iov[0].iov_len > len) iov[0].iov_len = len;
125 len -= iov[0].iov_len;
126 if (len) {
127 iov[1].iov_base = sb->sb_data;
128 iov[1].iov_len = sb->sb_rptr - sb->sb_data;
129 if(iov[1].iov_len > len)
130 iov[1].iov_len = len;
131 total = iov[0].iov_len + iov[1].iov_len;
132 if (total > mss) {
133 lss = total%mss;
134 if (iov[1].iov_len > lss) {
135 iov[1].iov_len -= lss;
136 n = 2;
137 } else {
138 lss -= iov[1].iov_len;
139 iov[0].iov_len -= lss;
140 n = 1;
141 }
142 } else
143 n = 2;
144 } else {
145 if (iov[0].iov_len > mss)
146 iov[0].iov_len -= iov[0].iov_len%mss;
147 n = 1;
148 }
149 }
150
151 #ifdef HAVE_READV
152 nn = readv(so->s, (struct iovec *)iov, n);
153 DEBUG_MISC((dfd, " ... read nn = %d bytes\n", nn));
154 #else
155 nn = read(so->s, iov[0].iov_base, iov[0].iov_len);
156 #endif
157 if (nn <= 0) {
158 if (nn < 0 && (errno == EINTR || errno == EAGAIN))
159 return 0;
160 else {
161 DEBUG_MISC((dfd, " --- soread() disconnected, nn = %d, errno = %d-%s\n", nn, errno,strerror(errno)));
162 sofcantrcvmore(so);
163 tcp_sockclosed(sototcpcb(so));
164 return -1;
165 }
166 }
167
168 #ifndef HAVE_READV
169 /*
170 * If there was no error, try and read the second time round
171 * We read again if n = 2 (ie, there's another part of the buffer)
172 * and we read as much as we could in the first read
173 * We don't test for <= 0 this time, because there legitimately
174 * might not be any more data (since the socket is non-blocking),
175 * a close will be detected on next iteration.
176 * A return of -1 wont (shouldn't) happen, since it didn't happen above
177 */
178 if (n == 2 && nn == iov[0].iov_len)
179 nn += read(so->s, iov[1].iov_base, iov[1].iov_len);
180
181 DEBUG_MISC((dfd, " ... read nn = %d bytes\n", nn));
182 #endif
183
184 /* Update fields */
185 sb->sb_cc += nn;
186 sb->sb_wptr += nn;
187 if (sb->sb_wptr >= (sb->sb_data + sb->sb_datalen))
188 sb->sb_wptr -= sb->sb_datalen;
189 return nn;
190 }
191
192 /*
193 * Get urgent data
194 *
195 * When the socket is created, we set it SO_OOBINLINE,
196 * so when OOB data arrives, we soread() it and everything
197 * in the send buffer is sent as urgent data
198 */
199 void
200 sorecvoob(so)
201 struct socket *so;
202 {
203 struct tcpcb *tp = sototcpcb(so);
204
205 DEBUG_CALL("sorecvoob");
206 DEBUG_ARG("so = %lx", (long)so);
207
208 /*
209 * We take a guess at how much urgent data has arrived.
210 * In most situations, when urgent data arrives, the next
211 * read() should get all the urgent data. This guess will
212 * be wrong however if more data arrives just after the
213 * urgent data, or the read() doesn't return all the
214 * urgent data.
215 */
216 soread(so);
217 tp->snd_up = tp->snd_una + so->so_snd.sb_cc;
218 tp->t_force = 1;
219 tcp_output(tp);
220 tp->t_force = 0;
221 }
222
223 /*
224 * Send urgent data
225 * There's a lot duplicated code here, but...
226 */
227 int
228 sosendoob(so)
229 struct socket *so;
230 {
231 struct sbuf *sb = &so->so_rcv;
232 char buff[2048]; /* XXX Shouldn't be sending more oob data than this */
233
234 int n, len;
235
236 DEBUG_CALL("sosendoob");
237 DEBUG_ARG("so = %lx", (long)so);
238 DEBUG_ARG("sb->sb_cc = %d", sb->sb_cc);
239
240 if (so->so_urgc > 2048)
241 so->so_urgc = 2048; /* XXXX */
242
243 if (sb->sb_rptr < sb->sb_wptr) {
244 /* We can send it directly */
245 n = send(so->s, sb->sb_rptr, so->so_urgc, (MSG_OOB)); /* |MSG_DONTWAIT)); */
246 so->so_urgc -= n;
247
248 DEBUG_MISC((dfd, " --- sent %d bytes urgent data, %d urgent bytes left\n", n, so->so_urgc));
249 } else {
250 /*
251 * Since there's no sendv or sendtov like writev,
252 * we must copy all data to a linear buffer then
253 * send it all
254 */
255 len = (sb->sb_data + sb->sb_datalen) - sb->sb_rptr;
256 if (len > so->so_urgc) len = so->so_urgc;
257 memcpy(buff, sb->sb_rptr, len);
258 so->so_urgc -= len;
259 if (so->so_urgc) {
260 n = sb->sb_wptr - sb->sb_data;
261 if (n > so->so_urgc) n = so->so_urgc;
262 memcpy((buff + len), sb->sb_data, n);
263 so->so_urgc -= n;
264 len += n;
265 }
266 n = send(so->s, buff, len, (MSG_OOB)); /* |MSG_DONTWAIT)); */
267 #ifdef DEBUG
268 if (n != len)
269 DEBUG_ERROR((dfd, "Didn't send all data urgently XXXXX\n"));
270 #endif
271 DEBUG_MISC((dfd, " ---2 sent %d bytes urgent data, %d urgent bytes left\n", n, so->so_urgc));
272 }
273
274 sb->sb_cc -= n;
275 sb->sb_rptr += n;
276 if (sb->sb_rptr >= (sb->sb_data + sb->sb_datalen))
277 sb->sb_rptr -= sb->sb_datalen;
278
279 return n;
280 }
281
282 /*
283 * Write data from so_rcv to so's socket,
284 * updating all sbuf field as necessary
285 */
286 int
287 sowrite(so)
288 struct socket *so;
289 {
290 int n,nn;
291 struct sbuf *sb = &so->so_rcv;
292 int len = sb->sb_cc;
293 struct iovec iov[2];
294
295 DEBUG_CALL("sowrite");
296 DEBUG_ARG("so = %lx", (long)so);
297
298 if (so->so_urgc) {
299 sosendoob(so);
300 if (sb->sb_cc == 0)
301 return 0;
302 }
303
304 /*
305 * No need to check if there's something to write,
306 * sowrite wouldn't have been called otherwise
307 */
308
309 len = sb->sb_cc;
310
311 iov[0].iov_base = sb->sb_rptr;
312 if (sb->sb_rptr < sb->sb_wptr) {
313 iov[0].iov_len = sb->sb_wptr - sb->sb_rptr;
314 /* Should never succeed, but... */
315 if (iov[0].iov_len > len) iov[0].iov_len = len;
316 n = 1;
317 } else {
318 iov[0].iov_len = (sb->sb_data + sb->sb_datalen) - sb->sb_rptr;
319 if (iov[0].iov_len > len) iov[0].iov_len = len;
320 len -= iov[0].iov_len;
321 if (len) {
322 iov[1].iov_base = sb->sb_data;
323 iov[1].iov_len = sb->sb_wptr - sb->sb_data;
324 if (iov[1].iov_len > len) iov[1].iov_len = len;
325 n = 2;
326 } else
327 n = 1;
328 }
329 /* Check if there's urgent data to send, and if so, send it */
330
331 #ifdef HAVE_READV
332 nn = writev(so->s, (const struct iovec *)iov, n);
333
334 DEBUG_MISC((dfd, " ... wrote nn = %d bytes\n", nn));
335 #else
336 nn = write(so->s, iov[0].iov_base, iov[0].iov_len);
337 #endif
338 /* This should never happen, but people tell me it does *shrug* */
339 if (nn < 0 && (errno == EAGAIN || errno == EINTR))
340 return 0;
341
342 if (nn <= 0) {
343 DEBUG_MISC((dfd, " --- sowrite disconnected, so->so_state = %x, errno = %d\n",
344 so->so_state, errno));
345 sofcantsendmore(so);
346 tcp_sockclosed(sototcpcb(so));
347 return -1;
348 }
349
350 #ifndef HAVE_READV
351 if (n == 2 && nn == iov[0].iov_len)
352 nn += write(so->s, iov[1].iov_base, iov[1].iov_len);
353 DEBUG_MISC((dfd, " ... wrote nn = %d bytes\n", nn));
354 #endif
355
356 /* Update sbuf */
357 sb->sb_cc -= nn;
358 sb->sb_rptr += nn;
359 if (sb->sb_rptr >= (sb->sb_data + sb->sb_datalen))
360 sb->sb_rptr -= sb->sb_datalen;
361
362 /*
363 * If in DRAIN mode, and there's no more data, set
364 * it CANTSENDMORE
365 */
366 if ((so->so_state & SS_FWDRAIN) && sb->sb_cc == 0)
367 sofcantsendmore(so);
368
369 return nn;
370 }
371
372 /*
373 * recvfrom() a UDP socket
374 */
375 void
376 sorecvfrom(so)
377 struct socket *so;
378 {
379 struct sockaddr_in addr;
380 int addrlen = sizeof(struct sockaddr_in);
381
382 DEBUG_CALL("sorecvfrom");
383 DEBUG_ARG("so = %lx", (long)so);
384
385 if (so->so_type == IPPROTO_ICMP) { /* This is a "ping" reply */
386 char buff[256];
387 int len;
388
389 len = recvfrom(so->s, buff, 256, 0,
390 (struct sockaddr *)&addr, &addrlen);
391 /* XXX Check if reply is "correct"? */
392
393 if(len == -1 || len == 0) {
394 u_char code=ICMP_UNREACH_PORT;
395
396 if(errno == EHOSTUNREACH) code=ICMP_UNREACH_HOST;
397 else if(errno == ENETUNREACH) code=ICMP_UNREACH_NET;
398
399 DEBUG_MISC((dfd," udp icmp rx errno = %d-%s\n",
400 errno,strerror(errno)));
401 icmp_error(so->so_m, ICMP_UNREACH,code, 0,strerror(errno));
402 } else {
403 icmp_reflect(so->so_m);
404 so->so_m = 0; /* Don't m_free() it again! */
405 }
406 /* No need for this socket anymore, udp_detach it */
407 udp_detach(so);
408 } else { /* A "normal" UDP packet */
409 struct mbuf *m;
410 int len, n;
411
412 if (!(m = m_get())) return;
413 m->m_data += if_maxlinkhdr;
414
415 /*
416 * XXX Shouldn't FIONREAD packets destined for port 53,
417 * but I don't know the max packet size for DNS lookups
418 */
419 len = M_FREEROOM(m);
420 /* if (so->so_fport != htons(53)) { */
421 ioctl(so->s, FIONREAD, &n);
422
423 if (n > len) {
424 n = (m->m_data - m->m_dat) + m->m_len + n + 1;
425 m_inc(m, n);
426 len = M_FREEROOM(m);
427 }
428 /* } */
429
430 m->m_len = recvfrom(so->s, m->m_data, len, 0,
431 (struct sockaddr *)&addr, &addrlen);
432 DEBUG_MISC((dfd, " did recvfrom %d, errno = %d-%s\n",
433 m->m_len, errno,strerror(errno)));
434 if(m->m_len<0) {
435 u_char code=ICMP_UNREACH_PORT;
436
437 if(errno == EHOSTUNREACH) code=ICMP_UNREACH_HOST;
438 else if(errno == ENETUNREACH) code=ICMP_UNREACH_NET;
439
440 DEBUG_MISC((dfd," rx error, tx icmp ICMP_UNREACH:%i\n", code));
441 icmp_error(so->so_m, ICMP_UNREACH,code, 0,strerror(errno));
442 m_free(m);
443 } else {
444 /*
445 * Hack: domain name lookup will be used the most for UDP,
446 * and since they'll only be used once there's no need
447 * for the 4 minute (or whatever) timeout... So we time them
448 * out much quicker (10 seconds for now...)
449 */
450 if (so->so_expire) {
451 if (so->so_fport == htons(53))
452 so->so_expire = curtime + SO_EXPIREFAST;
453 else
454 so->so_expire = curtime + SO_EXPIRE;
455 }
456
457 /* if (m->m_len == len) {
458 * m_inc(m, MINCSIZE);
459 * m->m_len = 0;
460 * }
461 */
462
463 /*
464 * If this packet was destined for CTL_ADDR,
465 * make it look like that's where it came from, done by udp_output
466 */
467 udp_output(so, m, &addr);
468 } /* rx error */
469 } /* if ping packet */
470 }
471
472 /*
473 * sendto() a socket
474 */
475 int
476 sosendto(so, m)
477 struct socket *so;
478 struct mbuf *m;
479 {
480 int ret;
481 struct sockaddr_in addr;
482
483 DEBUG_CALL("sosendto");
484 DEBUG_ARG("so = %lx", (long)so);
485 DEBUG_ARG("m = %lx", (long)m);
486
487 addr.sin_family = AF_INET;
488 if ((so->so_faddr.s_addr & htonl(0xffffff00)) == special_addr.s_addr) {
489 /* It's an alias */
490 switch(ntohl(so->so_faddr.s_addr) & 0xff) {
491 case CTL_DNS:
492 addr.sin_addr = dns_addr;
493 break;
494 case CTL_ALIAS:
495 default:
496 addr.sin_addr = loopback_addr;
497 break;
498 }
499 } else
500 addr.sin_addr = so->so_faddr;
501 addr.sin_port = so->so_fport;
502
503 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)));
504
505 /* Don't care what port we get */
506 ret = sendto(so->s, m->m_data, m->m_len, 0,
507 (struct sockaddr *)&addr, sizeof (struct sockaddr));
508 if (ret < 0)
509 return -1;
510
511 /*
512 * Kill the socket if there's no reply in 4 minutes,
513 * but only if it's an expirable socket
514 */
515 if (so->so_expire)
516 so->so_expire = curtime + SO_EXPIRE;
517 so->so_state = SS_ISFCONNECTED; /* So that it gets select()ed */
518 return 0;
519 }
520
521 /*
522 * XXX This should really be tcp_listen
523 */
524 struct socket *
525 solisten(port, laddr, lport, flags)
526 u_int port;
527 u_int32_t laddr;
528 u_int lport;
529 int flags;
530 {
531 struct sockaddr_in addr;
532 struct socket *so;
533 int s, addrlen = sizeof(addr), opt = 1;
534
535 DEBUG_CALL("solisten");
536 DEBUG_ARG("port = %d", port);
537 DEBUG_ARG("laddr = %x", laddr);
538 DEBUG_ARG("lport = %d", lport);
539 DEBUG_ARG("flags = %x", flags);
540
541 if ((so = socreate()) == NULL) {
542 /* free(so); Not sofree() ??? free(NULL) == NOP */
543 return NULL;
544 }
545
546 /* Don't tcp_attach... we don't need so_snd nor so_rcv */
547 if ((so->so_tcpcb = tcp_newtcpcb(so)) == NULL) {
548 free(so);
549 return NULL;
550 }
551 insque(so,&tcb);
552
553 /*
554 * SS_FACCEPTONCE sockets must time out.
555 */
556 if (flags & SS_FACCEPTONCE)
557 so->so_tcpcb->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT*2;
558
559 so->so_state = (SS_FACCEPTCONN|flags);
560 so->so_lport = lport; /* Kept in network format */
561 so->so_laddr.s_addr = laddr; /* Ditto */
562
563 addr.sin_family = AF_INET;
564 addr.sin_addr.s_addr = INADDR_ANY;
565 addr.sin_port = port;
566
567 if (((s = socket(AF_INET,SOCK_STREAM,0)) < 0) ||
568 (bind(s,(struct sockaddr *)&addr, sizeof(addr)) < 0) ||
569 (listen(s,1) < 0)) {
570 int tmperrno = errno; /* Don't clobber the real reason we failed */
571
572 close(s);
573 sofree(so);
574 /* Restore the real errno */
575 errno = tmperrno;
576 return NULL;
577 }
578 setsockopt(s,SOL_SOCKET,SO_REUSEADDR,(char *)&opt,sizeof(int));
579 setsockopt(s,SOL_SOCKET,SO_OOBINLINE,(char *)&opt,sizeof(int));
580
581 getsockname(s,(struct sockaddr *)&addr,&addrlen);
582 so->so_fport = addr.sin_port;
583 if (addr.sin_addr.s_addr == 0 || addr.sin_addr.s_addr == loopback_addr.s_addr)
584 so->so_faddr = our_addr;
585 else
586 so->so_faddr = addr.sin_addr;
587
588 so->s = s;
589 return so;
590 }
591
592 /*
593 * Data is available in so_rcv
594 * Just write() the data to the socket
595 * XXX not yet...
596 */
597 void
598 sorwakeup(so)
599 struct socket *so;
600 {
601 /* sowrite(so); */
602 /* FD_CLR(so->s,&writefds); */
603 }
604
605 /*
606 * Data has been freed in so_snd
607 * We have room for a read() if we want to
608 * For now, don't read, it'll be done in the main loop
609 */
610 void
611 sowwakeup(so)
612 struct socket *so;
613 {
614 /* Nothing, yet */
615 }
616
617 /*
618 * Various session state calls
619 * XXX Should be #define's
620 * The socket state stuff needs work, these often get call 2 or 3
621 * times each when only 1 was needed
622 */
623 void
624 soisfconnecting(so)
625 register struct socket *so;
626 {
627 so->so_state &= ~(SS_NOFDREF|SS_ISFCONNECTED|SS_FCANTRCVMORE|
628 SS_FCANTSENDMORE|SS_FWDRAIN);
629 so->so_state |= SS_ISFCONNECTING; /* Clobber other states */
630 }
631
632 void
633 soisfconnected(so)
634 register struct socket *so;
635 {
636 so->so_state &= ~(SS_ISFCONNECTING|SS_FWDRAIN|SS_NOFDREF);
637 so->so_state |= SS_ISFCONNECTED; /* Clobber other states */
638 }
639
640 void
641 sofcantrcvmore(so)
642 struct socket *so;
643 {
644 if ((so->so_state & SS_NOFDREF) == 0) {
645 shutdown(so->s,0);
646 FD_CLR(so->s, global_writefds);
647 }
648 so->so_state &= ~(SS_ISFCONNECTING);
649 if (so->so_state & SS_FCANTSENDMORE)
650 so->so_state = SS_NOFDREF; /* Don't select it */ /* XXX close() here as well? */
651 else
652 so->so_state |= SS_FCANTRCVMORE;
653 }
654
655 void
656 sofcantsendmore(so)
657 struct socket *so;
658 {
659 if ((so->so_state & SS_NOFDREF) == 0) {
660 shutdown(so->s,1); /* send FIN to fhost */
661 FD_CLR(so->s, global_readfds);
662 FD_CLR(so->s, global_xfds);
663 }
664 so->so_state &= ~(SS_ISFCONNECTING);
665 if (so->so_state & SS_FCANTRCVMORE)
666 so->so_state = SS_NOFDREF; /* as above */
667 else
668 so->so_state |= SS_FCANTSENDMORE;
669 }
670
671 void
672 soisfdisconnected(so)
673 struct socket *so;
674 {
675 /* so->so_state &= ~(SS_ISFCONNECTING|SS_ISFCONNECTED); */
676 /* close(so->s); */
677 /* so->so_state = SS_ISFDISCONNECTED; */
678 /*
679 * XXX Do nothing ... ?
680 */
681 }
682
683 /*
684 * Set write drain mode
685 * Set CANTSENDMORE once all data has been write()n
686 */
687 void
688 sofwdrain(so)
689 struct socket *so;
690 {
691 if (so->so_rcv.sb_cc)
692 so->so_state |= SS_FWDRAIN;
693 else
694 sofcantsendmore(so);
695 }
696