xref: /original-bsd/sys/netinet/if_ether.c (revision feb5f8e2)
1 /*
2  * Copyright (c) 1982, 1986, 1988 Regents of the University of California.
3  * All rights reserved.
4  *
5  * %sccs.include.redist.c%
6  *
7  *	@(#)if_ether.c	7.13 (Berkeley) 10/31/90
8  */
9 
10 /*
11  * Ethernet address resolution protocol.
12  * TODO:
13  *	run at splnet (add ARP protocol intr.)
14  *	link entries onto hash chains, keep free list
15  *	add "inuse/lock" bit (or ref. count) along with valid bit
16  */
17 
18 #include "param.h"
19 #include "systm.h"
20 #include "malloc.h"
21 #include "mbuf.h"
22 #include "socket.h"
23 #include "time.h"
24 #include "kernel.h"
25 #include "errno.h"
26 #include "ioctl.h"
27 #include "syslog.h"
28 
29 #include "../net/if.h"
30 #include "in.h"
31 #include "in_systm.h"
32 #include "in_var.h"
33 #include "ip.h"
34 #include "if_ether.h"
35 
36 #ifdef GATEWAY
37 #define	ARPTAB_BSIZ	16		/* bucket size */
38 #define	ARPTAB_NB	37		/* number of buckets */
39 #else
40 #define	ARPTAB_BSIZ	9		/* bucket size */
41 #define	ARPTAB_NB	19		/* number of buckets */
42 #endif
43 #define	ARPTAB_SIZE	(ARPTAB_BSIZ * ARPTAB_NB)
44 struct	arptab arptab[ARPTAB_SIZE];
45 int	arptab_size = ARPTAB_SIZE;	/* for arp command */
46 
47 /*
48  * ARP trailer negotiation.  Trailer protocol is not IP specific,
49  * but ARP request/response use IP addresses.
50  */
51 #define ETHERTYPE_IPTRAILERS ETHERTYPE_TRAIL
52 
53 #define	ARPTAB_HASH(a) \
54 	((u_long)(a) % ARPTAB_NB)
55 
56 #define	ARPTAB_LOOK(at,addr) { \
57 	register n; \
58 	at = &arptab[ARPTAB_HASH(addr) * ARPTAB_BSIZ]; \
59 	for (n = 0 ; n < ARPTAB_BSIZ ; n++,at++) \
60 		if (at->at_iaddr.s_addr == addr) \
61 			break; \
62 	if (n >= ARPTAB_BSIZ) \
63 		at = 0; \
64 }
65 
66 /* timer values */
67 #define	ARPT_AGE	(60*1)	/* aging timer, 1 min. */
68 #define	ARPT_KILLC	20	/* kill completed entry in 20 mins. */
69 #define	ARPT_KILLI	3	/* kill incomplete entry in 3 minutes */
70 
71 extern struct ifnet loif;
72 
73 /*
74  * Timeout routine.  Age arp_tab entries once a minute.
75  */
76 arptimer()
77 {
78 	register struct arptab *at;
79 	register i;
80 
81 	timeout(arptimer, (caddr_t)0, ARPT_AGE * hz);
82 	at = &arptab[0];
83 	for (i = 0; i < ARPTAB_SIZE; i++, at++) {
84 		if (at->at_flags == 0 || (at->at_flags & ATF_PERM))
85 			continue;
86 		if (++at->at_timer < ((at->at_flags&ATF_COM) ?
87 		    ARPT_KILLC : ARPT_KILLI))
88 			continue;
89 		/* timer has expired, clear entry */
90 		arptfree(at);
91 	}
92 }
93 
94 /*
95  * Broadcast an ARP packet, asking who has addr on interface ac.
96  */
97 arpwhohas(ac, addr)
98 	register struct arpcom *ac;
99 	struct in_addr *addr;
100 {
101 	register struct mbuf *m;
102 	register struct ether_header *eh;
103 	register struct ether_arp *ea;
104 	struct sockaddr sa;
105 
106 	if ((m = m_gethdr(M_DONTWAIT, MT_DATA)) == NULL)
107 		return;
108 	m->m_len = sizeof(*ea);
109 	m->m_pkthdr.len = sizeof(*ea);
110 	MH_ALIGN(m, sizeof(*ea));
111 	ea = mtod(m, struct ether_arp *);
112 	eh = (struct ether_header *)sa.sa_data;
113 	bzero((caddr_t)ea, sizeof (*ea));
114 	bcopy((caddr_t)etherbroadcastaddr, (caddr_t)eh->ether_dhost,
115 	    sizeof(eh->ether_dhost));
116 	eh->ether_type = ETHERTYPE_ARP;		/* if_output will swap */
117 	ea->arp_hrd = htons(ARPHRD_ETHER);
118 	ea->arp_pro = htons(ETHERTYPE_IP);
119 	ea->arp_hln = sizeof(ea->arp_sha);	/* hardware address length */
120 	ea->arp_pln = sizeof(ea->arp_spa);	/* protocol address length */
121 	ea->arp_op = htons(ARPOP_REQUEST);
122 	bcopy((caddr_t)ac->ac_enaddr, (caddr_t)ea->arp_sha,
123 	   sizeof(ea->arp_sha));
124 	bcopy((caddr_t)&ac->ac_ipaddr, (caddr_t)ea->arp_spa,
125 	   sizeof(ea->arp_spa));
126 	bcopy((caddr_t)addr, (caddr_t)ea->arp_tpa, sizeof(ea->arp_tpa));
127 	sa.sa_family = AF_UNSPEC;
128 	sa.sa_len = sizeof(sa);
129 	(*ac->ac_if.if_output)(&ac->ac_if, m, &sa, (struct rtentry *)0);
130 }
131 
132 int	useloopback = 1;	/* use loopback interface for local traffic */
133 
134 /*
135  * Resolve an IP address into an ethernet address.  If success,
136  * desten is filled in.  If there is no entry in arptab,
137  * set one up and broadcast a request for the IP address.
138  * Hold onto this mbuf and resend it once the address
139  * is finally resolved.  A return value of 1 indicates
140  * that desten has been filled in and the packet should be sent
141  * normally; a 0 return indicates that the packet has been
142  * taken over here, either now or for later transmission.
143  *
144  * We do some (conservative) locking here at splimp, since
145  * arptab is also altered from input interrupt service (ecintr/ilintr
146  * calls arpinput when ETHERTYPE_ARP packets come in).
147  */
148 arpresolve(ac, m, destip, desten, usetrailers)
149 	register struct arpcom *ac;
150 	struct mbuf *m;
151 	register struct in_addr *destip;
152 	register u_char *desten;
153 	int *usetrailers;
154 {
155 	register struct arptab *at;
156 	struct sockaddr_in sin;
157 	register struct in_ifaddr *ia;
158 	u_long lna;
159 	int s;
160 
161 	*usetrailers = 0;
162 	if (m->m_flags & M_BCAST) {	/* broadcast */
163 		bcopy((caddr_t)etherbroadcastaddr, (caddr_t)desten,
164 		    sizeof(etherbroadcastaddr));
165 		return (1);
166 	}
167 	lna = in_lnaof(*destip);
168 	/* if for us, use software loopback driver if up */
169 	for (ia = in_ifaddr; ia; ia = ia->ia_next)
170 	    if ((ia->ia_ifp == &ac->ac_if) &&
171 		(destip->s_addr == ia->ia_addr.sin_addr.s_addr)) {
172 		/*
173 		 * This test used to be
174 		 *	if (loif.if_flags & IFF_UP)
175 		 * It allowed local traffic to be forced
176 		 * through the hardware by configuring the loopback down.
177 		 * However, it causes problems during network configuration
178 		 * for boards that can't receive packets they send.
179 		 * It is now necessary to clear "useloopback"
180 		 * to force traffic out to the hardware.
181 		 */
182 		if (useloopback) {
183 			sin.sin_family = AF_INET;
184 			sin.sin_addr = *destip;
185 			(void) looutput(&loif, m, (struct sockaddr *)&sin, 0);
186 			/*
187 			 * The packet has already been sent and freed.
188 			 */
189 			return (0);
190 		} else {
191 			bcopy((caddr_t)ac->ac_enaddr, (caddr_t)desten,
192 			    sizeof(ac->ac_enaddr));
193 			return (1);
194 		}
195 	}
196 	s = splimp();
197 	ARPTAB_LOOK(at, destip->s_addr);
198 	if (at == 0) {			/* not found */
199 		if (ac->ac_if.if_flags & IFF_NOARP) {
200 			bcopy((caddr_t)ac->ac_enaddr, (caddr_t)desten, 3);
201 			desten[3] = (lna >> 16) & 0x7f;
202 			desten[4] = (lna >> 8) & 0xff;
203 			desten[5] = lna & 0xff;
204 			splx(s);
205 			return (1);
206 		} else {
207 			at = arptnew(destip);
208 			if (at == 0)
209 				panic("arpresolve: no free entry");
210 			at->at_hold = m;
211 			arpwhohas(ac, destip);
212 			splx(s);
213 			return (0);
214 		}
215 	}
216 	at->at_timer = 0;		/* restart the timer */
217 	if (at->at_flags & ATF_COM) {	/* entry IS complete */
218 		bcopy((caddr_t)at->at_enaddr, (caddr_t)desten,
219 		    sizeof(at->at_enaddr));
220 		if (at->at_flags & ATF_USETRAILERS)
221 			*usetrailers = 1;
222 		splx(s);
223 		return (1);
224 	}
225 	/*
226 	 * There is an arptab entry, but no ethernet address
227 	 * response yet.  Replace the held mbuf with this
228 	 * latest one.
229 	 */
230 	if (at->at_hold)
231 		m_freem(at->at_hold);
232 	at->at_hold = m;
233 	arpwhohas(ac, destip);		/* ask again */
234 	splx(s);
235 	return (0);
236 }
237 
238 /*
239  * Called from 10 Mb/s Ethernet interrupt handlers
240  * when ether packet type ETHERTYPE_ARP
241  * is received.  Common length and type checks are done here,
242  * then the protocol-specific routine is called.
243  */
244 arpinput(ac, m)
245 	struct arpcom *ac;
246 	struct mbuf *m;
247 {
248 	register struct arphdr *ar;
249 
250 	if (ac->ac_if.if_flags & IFF_NOARP)
251 		goto out;
252 	if (m->m_len < sizeof(struct arphdr))
253 		goto out;
254 	ar = mtod(m, struct arphdr *);
255 	if (ntohs(ar->ar_hrd) != ARPHRD_ETHER)
256 		goto out;
257 	if (m->m_len < sizeof(struct arphdr) + 2 * ar->ar_hln + 2 * ar->ar_pln)
258 		goto out;
259 
260 	switch (ntohs(ar->ar_pro)) {
261 
262 	case ETHERTYPE_IP:
263 	case ETHERTYPE_IPTRAILERS:
264 		in_arpinput(ac, m);
265 		return;
266 
267 	default:
268 		break;
269 	}
270 out:
271 	m_freem(m);
272 }
273 
274 /*
275  * ARP for Internet protocols on 10 Mb/s Ethernet.
276  * Algorithm is that given in RFC 826.
277  * In addition, a sanity check is performed on the sender
278  * protocol address, to catch impersonators.
279  * We also handle negotiations for use of trailer protocol:
280  * ARP replies for protocol type ETHERTYPE_TRAIL are sent
281  * along with IP replies if we want trailers sent to us,
282  * and also send them in response to IP replies.
283  * This allows either end to announce the desire to receive
284  * trailer packets.
285  * We reply to requests for ETHERTYPE_TRAIL protocol as well,
286  * but don't normally send requests.
287  */
288 in_arpinput(ac, m)
289 	register struct arpcom *ac;
290 	struct mbuf *m;
291 {
292 	register struct ether_arp *ea;
293 	struct ether_header *eh;
294 	register struct arptab *at;  /* same as "merge" flag */
295 	register struct in_ifaddr *ia;
296 	struct in_ifaddr *maybe_ia = 0;
297 	struct mbuf *mcopy = 0;
298 	struct sockaddr_in sin;
299 	struct sockaddr sa;
300 	struct in_addr isaddr, itaddr, myaddr;
301 	int proto, op, s, completed = 0;
302 
303 	ea = mtod(m, struct ether_arp *);
304 	proto = ntohs(ea->arp_pro);
305 	op = ntohs(ea->arp_op);
306 	bcopy((caddr_t)ea->arp_spa, (caddr_t)&isaddr, sizeof (isaddr));
307 	bcopy((caddr_t)ea->arp_tpa, (caddr_t)&itaddr, sizeof (itaddr));
308 	for (ia = in_ifaddr; ia; ia = ia->ia_next)
309 		if (ia->ia_ifp == &ac->ac_if) {
310 			maybe_ia = ia;
311 			if ((itaddr.s_addr == ia->ia_addr.sin_addr.s_addr) ||
312 			     (isaddr.s_addr == ia->ia_addr.sin_addr.s_addr))
313 				break;
314 		}
315 	if (maybe_ia == 0)
316 		goto out;
317 	myaddr = ia ? ia->ia_addr.sin_addr : maybe_ia->ia_addr.sin_addr;
318 	if (!bcmp((caddr_t)ea->arp_sha, (caddr_t)ac->ac_enaddr,
319 	    sizeof (ea->arp_sha)))
320 		goto out;	/* it's from me, ignore it. */
321 	if (!bcmp((caddr_t)ea->arp_sha, (caddr_t)etherbroadcastaddr,
322 	    sizeof (ea->arp_sha))) {
323 		log(LOG_ERR,
324 		    "arp: ether address is broadcast for IP address %x!\n",
325 		    ntohl(isaddr.s_addr));
326 		goto out;
327 	}
328 	if (isaddr.s_addr == myaddr.s_addr) {
329 		log(LOG_ERR,
330 		   "duplicate IP address %x!! sent from ethernet address: %s\n",
331 		   ntohl(isaddr.s_addr), ether_sprintf(ea->arp_sha));
332 		itaddr = myaddr;
333 		if (op == ARPOP_REQUEST)
334 			goto reply;
335 		goto out;
336 	}
337 	s = splimp();
338 	ARPTAB_LOOK(at, isaddr.s_addr);
339 	if (at) {
340 		bcopy((caddr_t)ea->arp_sha, (caddr_t)at->at_enaddr,
341 		    sizeof(ea->arp_sha));
342 		if ((at->at_flags & ATF_COM) == 0)
343 			completed = 1;
344 		at->at_flags |= ATF_COM;
345 		if (at->at_hold) {
346 			sin.sin_family = AF_INET;
347 			sin.sin_addr = isaddr;
348 			(*ac->ac_if.if_output)(&ac->ac_if, at->at_hold,
349 				(struct sockaddr *)&sin, (struct rtentry *)0);
350 			at->at_hold = 0;
351 		}
352 	}
353 	if (at == 0 && itaddr.s_addr == myaddr.s_addr) {
354 		/* ensure we have a table entry */
355 		if (at = arptnew(&isaddr)) {
356 			bcopy((caddr_t)ea->arp_sha, (caddr_t)at->at_enaddr,
357 			    sizeof(ea->arp_sha));
358 			completed = 1;
359 			at->at_flags |= ATF_COM;
360 		}
361 	}
362 	splx(s);
363 reply:
364 	switch (proto) {
365 
366 	case ETHERTYPE_IPTRAILERS:
367 		/* partner says trailers are OK */
368 		if (at)
369 			at->at_flags |= ATF_USETRAILERS;
370 		/*
371 		 * Reply to request iff we want trailers.
372 		 */
373 		if (op != ARPOP_REQUEST || ac->ac_if.if_flags & IFF_NOTRAILERS)
374 			goto out;
375 		break;
376 
377 	case ETHERTYPE_IP:
378 		/*
379 		 * Reply if this is an IP request,
380 		 * or if we want to send a trailer response.
381 		 * Send the latter only to the IP response
382 		 * that completes the current ARP entry.
383 		 */
384 		if (op != ARPOP_REQUEST &&
385 		    (completed == 0 || ac->ac_if.if_flags & IFF_NOTRAILERS))
386 			goto out;
387 	}
388 	if (itaddr.s_addr == myaddr.s_addr) {
389 		/* I am the target */
390 		bcopy((caddr_t)ea->arp_sha, (caddr_t)ea->arp_tha,
391 		    sizeof(ea->arp_sha));
392 		bcopy((caddr_t)ac->ac_enaddr, (caddr_t)ea->arp_sha,
393 		    sizeof(ea->arp_sha));
394 	} else {
395 		ARPTAB_LOOK(at, itaddr.s_addr);
396 		if (at == NULL || (at->at_flags & ATF_PUBL) == 0)
397 			goto out;
398 		bcopy((caddr_t)ea->arp_sha, (caddr_t)ea->arp_tha,
399 		    sizeof(ea->arp_sha));
400 		bcopy((caddr_t)at->at_enaddr, (caddr_t)ea->arp_sha,
401 		    sizeof(ea->arp_sha));
402 	}
403 
404 	bcopy((caddr_t)ea->arp_spa, (caddr_t)ea->arp_tpa,
405 	    sizeof(ea->arp_spa));
406 	bcopy((caddr_t)&itaddr, (caddr_t)ea->arp_spa,
407 	    sizeof(ea->arp_spa));
408 	ea->arp_op = htons(ARPOP_REPLY);
409 	/*
410 	 * If incoming packet was an IP reply,
411 	 * we are sending a reply for type IPTRAILERS.
412 	 * If we are sending a reply for type IP
413 	 * and we want to receive trailers,
414 	 * send a trailer reply as well.
415 	 */
416 	if (op == ARPOP_REPLY)
417 		ea->arp_pro = htons(ETHERTYPE_IPTRAILERS);
418 	else if (proto == ETHERTYPE_IP &&
419 	    (ac->ac_if.if_flags & IFF_NOTRAILERS) == 0)
420 		mcopy = m_copy(m, 0, (int)M_COPYALL);
421 	eh = (struct ether_header *)sa.sa_data;
422 	bcopy((caddr_t)ea->arp_tha, (caddr_t)eh->ether_dhost,
423 	    sizeof(eh->ether_dhost));
424 	eh->ether_type = ETHERTYPE_ARP;
425 	sa.sa_family = AF_UNSPEC;
426 	sa.sa_len = sizeof(sa);
427 	(*ac->ac_if.if_output)(&ac->ac_if, m, &sa, (struct rtentry *)0);
428 	if (mcopy) {
429 		ea = mtod(mcopy, struct ether_arp *);
430 		ea->arp_pro = htons(ETHERTYPE_IPTRAILERS);
431 		(*ac->ac_if.if_output)(&ac->ac_if,
432 					mcopy, &sa, (struct rtentry *)0);
433 	}
434 	return;
435 out:
436 	m_freem(m);
437 	return;
438 }
439 
440 /*
441  * Free an arptab entry.
442  */
443 arptfree(at)
444 	register struct arptab *at;
445 {
446 	int s = splimp();
447 
448 	if (at->at_hold)
449 		m_freem(at->at_hold);
450 	at->at_hold = 0;
451 	at->at_timer = at->at_flags = 0;
452 	at->at_iaddr.s_addr = 0;
453 	splx(s);
454 }
455 
456 /*
457  * Enter a new address in arptab, pushing out the oldest entry
458  * from the bucket if there is no room.
459  * This always succeeds since no bucket can be completely filled
460  * with permanent entries (except from arpioctl when testing whether
461  * another permanent entry will fit).
462  * MUST BE CALLED AT SPLIMP.
463  */
464 struct arptab *
465 arptnew(addr)
466 	struct in_addr *addr;
467 {
468 	register n;
469 	int oldest = -1;
470 	register struct arptab *at, *ato = NULL;
471 	static int first = 1;
472 
473 	if (first) {
474 		first = 0;
475 		timeout(arptimer, (caddr_t)0, hz);
476 	}
477 	at = &arptab[ARPTAB_HASH(addr->s_addr) * ARPTAB_BSIZ];
478 	for (n = 0; n < ARPTAB_BSIZ; n++,at++) {
479 		if (at->at_flags == 0)
480 			goto out;	 /* found an empty entry */
481 		if (at->at_flags & ATF_PERM)
482 			continue;
483 		if ((int) at->at_timer > oldest) {
484 			oldest = at->at_timer;
485 			ato = at;
486 		}
487 	}
488 	if (ato == NULL)
489 		return (NULL);
490 	at = ato;
491 	arptfree(at);
492 out:
493 	at->at_iaddr = *addr;
494 	at->at_flags = ATF_INUSE;
495 	return (at);
496 }
497 
498 arpioctl(cmd, data)
499 	int cmd;
500 	caddr_t data;
501 {
502 	register struct arpreq *ar = (struct arpreq *)data;
503 	register struct arptab *at;
504 	register struct sockaddr_in *sin;
505 	int s;
506 
507 	sin = (struct sockaddr_in *)&ar->arp_ha;
508 #if defined(COMPAT_43) && BYTE_ORDER != BIG_ENDIAN
509 	if (sin->sin_family == 0 && sin->sin_len < 16)
510 		sin->sin_family = sin->sin_len;
511 #endif
512 	sin->sin_len = sizeof(ar->arp_ha);
513 	sin = (struct sockaddr_in *)&ar->arp_pa;
514 #if defined(COMPAT_43) && BYTE_ORDER != BIG_ENDIAN
515 	if (sin->sin_family == 0 && sin->sin_len < 16)
516 		sin->sin_family = sin->sin_len;
517 #endif
518 	sin->sin_len = sizeof(ar->arp_pa);
519 	if (ar->arp_pa.sa_family != AF_INET ||
520 	    ar->arp_ha.sa_family != AF_UNSPEC)
521 		return (EAFNOSUPPORT);
522 	s = splimp();
523 	ARPTAB_LOOK(at, sin->sin_addr.s_addr);
524 	if (at == NULL) {		/* not found */
525 		if (cmd != SIOCSARP) {
526 			splx(s);
527 			return (ENXIO);
528 		}
529 		if (ifa_ifwithnet(&ar->arp_pa) == NULL) {
530 			splx(s);
531 			return (ENETUNREACH);
532 		}
533 	}
534 	switch (cmd) {
535 
536 	case SIOCSARP:		/* set entry */
537 		if (at == NULL) {
538 			at = arptnew(&sin->sin_addr);
539 			if (at == NULL) {
540 				splx(s);
541 				return (EADDRNOTAVAIL);
542 			}
543 			if (ar->arp_flags & ATF_PERM) {
544 			/* never make all entries in a bucket permanent */
545 				register struct arptab *tat;
546 
547 				/* try to re-allocate */
548 				tat = arptnew(&sin->sin_addr);
549 				if (tat == NULL) {
550 					arptfree(at);
551 					splx(s);
552 					return (EADDRNOTAVAIL);
553 				}
554 				arptfree(tat);
555 			}
556 		}
557 		bcopy((caddr_t)ar->arp_ha.sa_data, (caddr_t)at->at_enaddr,
558 		    sizeof(at->at_enaddr));
559 		at->at_flags = ATF_COM | ATF_INUSE |
560 			(ar->arp_flags & (ATF_PERM|ATF_PUBL|ATF_USETRAILERS));
561 		at->at_timer = 0;
562 		break;
563 
564 	case SIOCDARP:		/* delete entry */
565 		arptfree(at);
566 		break;
567 
568 	case SIOCGARP:		/* get entry */
569 	case OSIOCGARP:
570 		bcopy((caddr_t)at->at_enaddr, (caddr_t)ar->arp_ha.sa_data,
571 		    sizeof(at->at_enaddr));
572 #ifdef COMPAT_43
573 		if (cmd == OSIOCGARP)
574 			*(u_short *)&ar->arp_ha = ar->arp_ha.sa_family;
575 #endif
576 		ar->arp_flags = at->at_flags;
577 		break;
578 	}
579 	splx(s);
580 	return (0);
581 }
582