vhost.c: include <linux/vhost.h> last
[qemu.git] / slirp / socket.c
blob8a257acc83f82fa6ac87df98060712ba58c09bdb
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;
55 return(so);
59 * remque and free a socket, clobber cache
61 void
62 sofree(struct socket *so)
64 Slirp *slirp = so->slirp;
66 if (so->so_emu==EMU_RSH && so->extra) {
67 sofree(so->extra);
68 so->extra=NULL;
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;
75 m_free(so->so_m);
77 if(so->so_next && so->so_prev)
78 remque(so); /* crashes if so is not in a queue */
80 free(so);
83 size_t sopreprbuf(struct socket *so, struct iovec *iov, int *np)
85 int n, lss, total;
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);
93 if (len <= 0)
94 return 0;
96 iov[0].iov_base = sb->sb_wptr;
97 iov[1].iov_base = NULL;
98 iov[1].iov_len = 0;
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;
106 n = 1;
107 } else {
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;
112 if (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;
118 if (total > mss) {
119 lss = total%mss;
120 if (iov[1].iov_len > lss) {
121 iov[1].iov_len -= lss;
122 n = 2;
123 } else {
124 lss -= iov[1].iov_len;
125 iov[0].iov_len -= lss;
126 n = 1;
128 } else
129 n = 2;
130 } else {
131 if (iov[0].iov_len > mss)
132 iov[0].iov_len -= iov[0].iov_len%mss;
133 n = 1;
136 if (np)
137 *np = n;
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)
150 int n, nn;
151 struct sbuf *sb = &so->so_snd;
152 struct iovec iov[2];
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);
163 #ifdef HAVE_READV
164 nn = readv(so->s, (struct iovec *)iov, n);
165 DEBUG_MISC((dfd, " ... read nn = %d bytes\n", nn));
166 #else
167 nn = recv(so->s, iov[0].iov_base, iov[0].iov_len,0);
168 #endif
169 if (nn <= 0) {
170 if (nn < 0 && (errno == EINTR || errno == EAGAIN))
171 return 0;
172 else {
173 DEBUG_MISC((dfd, " --- soread() disconnected, nn = %d, errno = %d-%s\n", nn, errno,strerror(errno)));
174 sofcantrcvmore(so);
175 tcp_sockclosed(sototcpcb(so));
176 return -1;
180 #ifndef HAVE_READV
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) {
191 int ret;
192 ret = recv(so->s, iov[1].iov_base, iov[1].iov_len,0);
193 if (ret > 0)
194 nn += ret;
197 DEBUG_MISC((dfd, " ... read nn = %d bytes\n", nn));
198 #endif
200 /* Update fields */
201 sb->sb_cc += nn;
202 sb->sb_wptr += nn;
203 if (sb->sb_wptr >= (sb->sb_data + sb->sb_datalen))
204 sb->sb_wptr -= sb->sb_datalen;
205 return nn;
208 int soreadbuf(struct socket *so, const char *buf, int size)
210 int n, nn, copy = size;
211 struct sbuf *sb = &so->so_snd;
212 struct iovec iov[2];
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)
222 goto err;
224 nn = MIN(iov[0].iov_len, copy);
225 memcpy(iov[0].iov_base, buf, nn);
227 copy -= nn;
228 buf += nn;
230 if (copy == 0)
231 goto done;
233 memcpy(iov[1].iov_base, buf, copy);
235 done:
236 /* Update fields */
237 sb->sb_cc += size;
238 sb->sb_wptr += size;
239 if (sb->sb_wptr >= (sb->sb_data + sb->sb_datalen))
240 sb->sb_wptr -= sb->sb_datalen;
241 return size;
242 err:
244 sofcantrcvmore(so);
245 tcp_sockclosed(sototcpcb(so));
246 fprintf(stderr, "soreadbuf buffer to small");
247 return -1;
251 * Get urgent data
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
257 void
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
271 * urgent data.
273 soread(so);
274 tp->snd_up = tp->snd_una + so->so_snd.sb_cc;
275 tp->t_force = 1;
276 tcp_output(tp);
277 tp->t_force = 0;
281 * Send urgent data
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 */
290 int n, len;
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)); */
302 so->so_urgc -= n;
304 DEBUG_MISC((dfd, " --- sent %d bytes urgent data, %d urgent bytes left\n", n, so->so_urgc));
305 } else {
307 * Since there's no sendv or sendtov like writev,
308 * we must copy all data to a linear buffer then
309 * send it all
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);
314 so->so_urgc -= len;
315 if (so->so_urgc) {
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);
319 so->so_urgc -= n;
320 len += n;
322 n = slirp_send(so, buff, len, (MSG_OOB)); /* |MSG_DONTWAIT)); */
323 #ifdef DEBUG
324 if (n != len)
325 DEBUG_ERROR((dfd, "Didn't send all data urgently XXXXX\n"));
326 #endif
327 DEBUG_MISC((dfd, " ---2 sent %d bytes urgent data, %d urgent bytes left\n", n, so->so_urgc));
330 sb->sb_cc -= n;
331 sb->sb_rptr += n;
332 if (sb->sb_rptr >= (sb->sb_data + sb->sb_datalen))
333 sb->sb_rptr -= sb->sb_datalen;
335 return n;
339 * Write data from so_rcv to so's socket,
340 * updating all sbuf field as necessary
343 sowrite(struct socket *so)
345 int n,nn;
346 struct sbuf *sb = &so->so_rcv;
347 int len = sb->sb_cc;
348 struct iovec iov[2];
350 DEBUG_CALL("sowrite");
351 DEBUG_ARG("so = %lx", (long)so);
353 if (so->so_urgc) {
354 sosendoob(so);
355 if (sb->sb_cc == 0)
356 return 0;
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;
366 iov[1].iov_len = 0;
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;
371 n = 1;
372 } else {
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;
376 if (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;
380 n = 2;
381 } else
382 n = 1;
384 /* Check if there's urgent data to send, and if so, send it */
386 #ifdef HAVE_READV
387 nn = writev(so->s, (const struct iovec *)iov, n);
389 DEBUG_MISC((dfd, " ... wrote nn = %d bytes\n", nn));
390 #else
391 nn = slirp_send(so, iov[0].iov_base, iov[0].iov_len,0);
392 #endif
393 /* This should never happen, but people tell me it does *shrug* */
394 if (nn < 0 && (errno == EAGAIN || errno == EINTR))
395 return 0;
397 if (nn <= 0) {
398 DEBUG_MISC((dfd, " --- sowrite disconnected, so->so_state = %x, errno = %d\n",
399 so->so_state, errno));
400 sofcantsendmore(so);
401 tcp_sockclosed(sototcpcb(so));
402 return -1;
405 #ifndef HAVE_READV
406 if (n == 2 && nn == iov[0].iov_len) {
407 int ret;
408 ret = slirp_send(so, iov[1].iov_base, iov[1].iov_len,0);
409 if (ret > 0)
410 nn += ret;
412 DEBUG_MISC((dfd, " ... wrote nn = %d bytes\n", nn));
413 #endif
415 /* Update sbuf */
416 sb->sb_cc -= nn;
417 sb->sb_rptr += 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
423 * it CANTSENDMORE
425 if ((so->so_state & SS_FWDRAIN) && sb->sb_cc == 0)
426 sofcantsendmore(so);
428 return nn;
432 * recvfrom() a UDP socket
434 void
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 */
444 char buff[256];
445 int len;
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));
460 } else {
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 */
465 udp_detach(so);
466 } else { /* A "normal" UDP packet */
467 struct mbuf *m;
468 int len;
469 #ifdef _WIN32
470 unsigned long n;
471 #else
472 int n;
473 #endif
475 m = m_get(so->slirp);
476 if (!m) {
477 return;
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
485 len = M_FREEROOM(m);
486 /* if (so->so_fport != htons(53)) { */
487 ioctlsocket(so->s, FIONREAD, &n);
489 if (n > len) {
490 n = (m->m_data - m->m_dat) + m->m_len + n + 1;
491 m_inc(m, n);
492 len = M_FREEROOM(m);
494 /* } */
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)));
500 if(m->m_len<0) {
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));
508 m_free(m);
509 } else {
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...)
516 if (so->so_expire) {
517 if (so->so_fport == htons(53))
518 so->so_expire = curtime + SO_EXPIREFAST;
519 else
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);
528 } /* rx error */
529 } /* if ping packet */
533 * sendto() a socket
536 sosendto(struct socket *so, struct mbuf *m)
538 Slirp *slirp = so->slirp;
539 int ret;
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) {
549 /* It's an alias */
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;
553 } else {
554 addr.sin_addr = loopback_addr;
556 } else
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));
565 if (ret < 0)
566 return -1;
569 * Kill the socket if there's no reply in 4 minutes,
570 * but only if it's an expirable socket
572 if (so->so_expire)
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 */
576 return 0;
580 * Listen for incoming TCP connections
582 struct socket *
583 tcp_listen(Slirp *slirp, u_int32_t haddr, u_int hport, u_int32_t laddr,
584 u_int lport, int flags)
586 struct sockaddr_in addr;
587 struct socket *so;
588 int s, opt = 1;
589 socklen_t addrlen = sizeof(addr);
591 DEBUG_CALL("tcp_listen");
592 DEBUG_ARG("haddr = %x", haddr);
593 DEBUG_ARG("hport = %d", hport);
594 DEBUG_ARG("laddr = %x", laddr);
595 DEBUG_ARG("lport = %d", lport);
596 DEBUG_ARG("flags = %x", flags);
598 so = socreate(slirp);
599 if (!so) {
600 return NULL;
603 /* Don't tcp_attach... we don't need so_snd nor so_rcv */
604 if ((so->so_tcpcb = tcp_newtcpcb(so)) == NULL) {
605 free(so);
606 return NULL;
608 insque(so, &slirp->tcb);
611 * SS_FACCEPTONCE sockets must time out.
613 if (flags & SS_FACCEPTONCE)
614 so->so_tcpcb->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT*2;
616 so->so_state &= SS_PERSISTENT_MASK;
617 so->so_state |= (SS_FACCEPTCONN | flags);
618 so->so_lport = lport; /* Kept in network format */
619 so->so_laddr.s_addr = laddr; /* Ditto */
621 addr.sin_family = AF_INET;
622 addr.sin_addr.s_addr = haddr;
623 addr.sin_port = hport;
625 if (((s = qemu_socket(AF_INET,SOCK_STREAM,0)) < 0) ||
626 (setsockopt(s,SOL_SOCKET,SO_REUSEADDR,(char *)&opt,sizeof(int)) < 0) ||
627 (bind(s,(struct sockaddr *)&addr, sizeof(addr)) < 0) ||
628 (listen(s,1) < 0)) {
629 int tmperrno = errno; /* Don't clobber the real reason we failed */
631 close(s);
632 sofree(so);
633 /* Restore the real errno */
634 #ifdef _WIN32
635 WSASetLastError(tmperrno);
636 #else
637 errno = tmperrno;
638 #endif
639 return NULL;
641 setsockopt(s,SOL_SOCKET,SO_OOBINLINE,(char *)&opt,sizeof(int));
643 getsockname(s,(struct sockaddr *)&addr,&addrlen);
644 so->so_fport = addr.sin_port;
645 if (addr.sin_addr.s_addr == 0 || addr.sin_addr.s_addr == loopback_addr.s_addr)
646 so->so_faddr = slirp->vhost_addr;
647 else
648 so->so_faddr = addr.sin_addr;
650 so->s = s;
651 return so;
655 * Various session state calls
656 * XXX Should be #define's
657 * The socket state stuff needs work, these often get call 2 or 3
658 * times each when only 1 was needed
660 void
661 soisfconnecting(struct socket *so)
663 so->so_state &= ~(SS_NOFDREF|SS_ISFCONNECTED|SS_FCANTRCVMORE|
664 SS_FCANTSENDMORE|SS_FWDRAIN);
665 so->so_state |= SS_ISFCONNECTING; /* Clobber other states */
668 void
669 soisfconnected(struct socket *so)
671 so->so_state &= ~(SS_ISFCONNECTING|SS_FWDRAIN|SS_NOFDREF);
672 so->so_state |= SS_ISFCONNECTED; /* Clobber other states */
675 static void
676 sofcantrcvmore(struct socket *so)
678 if ((so->so_state & SS_NOFDREF) == 0) {
679 shutdown(so->s,0);
680 if(global_writefds) {
681 FD_CLR(so->s,global_writefds);
684 so->so_state &= ~(SS_ISFCONNECTING);
685 if (so->so_state & SS_FCANTSENDMORE) {
686 so->so_state &= SS_PERSISTENT_MASK;
687 so->so_state |= SS_NOFDREF; /* Don't select it */
688 } else {
689 so->so_state |= SS_FCANTRCVMORE;
693 static void
694 sofcantsendmore(struct socket *so)
696 if ((so->so_state & SS_NOFDREF) == 0) {
697 shutdown(so->s,1); /* send FIN to fhost */
698 if (global_readfds) {
699 FD_CLR(so->s,global_readfds);
701 if (global_xfds) {
702 FD_CLR(so->s,global_xfds);
705 so->so_state &= ~(SS_ISFCONNECTING);
706 if (so->so_state & SS_FCANTRCVMORE) {
707 so->so_state &= SS_PERSISTENT_MASK;
708 so->so_state |= SS_NOFDREF; /* as above */
709 } else {
710 so->so_state |= SS_FCANTSENDMORE;
715 * Set write drain mode
716 * Set CANTSENDMORE once all data has been write()n
718 void
719 sofwdrain(struct socket *so)
721 if (so->so_rcv.sb_cc)
722 so->so_state |= SS_FWDRAIN;
723 else
724 sofcantsendmore(so);