traceroute.pmtu.c

TCP／IP详解

	ident = (getpid() & 0xffff) | 0x8000;

	if ((pe = getprotobyname("icmp")) == NULL) {
		Fprintf(stderr, "icmp: unknown protocol\n");
		exit(10);
	}
	if ((s = socket(AF_INET, SOCK_RAW, pe->p_proto)) < 0) {
		perror("traceroute: icmp socket");
		exit(5);
	}
	if (options & SO_DEBUG)
		(void) setsockopt(s, SOL_SOCKET, SO_DEBUG,
				  (char *)&on, sizeof(on));
	if (options & SO_DONTROUTE)
		(void) setsockopt(s, SOL_SOCKET, SO_DONTROUTE,
				  (char *)&on, sizeof(on));

	if ((sndsock = socket(AF_INET, SOCK_RAW, IPPROTO_RAW)) < 0) {
		perror("traceroute: raw socket");
		exit(5);
	}

	if (lsrr > 0) {		/* loose source routing was specified */
		lsrr++;		/* now fill in total length */
		optlist[IPOPT_OLEN] = IPOPT_MINOFF - 1 + (lsrr*sizeof(u_long));
			/* and place destination addr as final entry */
		bcopy((caddr_t)&to->sin_addr, oix, sizeof(u_long));
Fprintf(stderr, "LSRR to %8x\n", ntohl(to->sin_addr.s_addr));
Fprintf(stderr, "LSRR size %d bytes\n", optlist[IPOPT_OLEN]);
		oix += sizeof(u_long);
		while ((oix - optlist) & 3)
			oix++;		/* pad to an even boundary */

		if ( (pe = getprotobyname("ip")) == NULL) {
			perror("traceroute: unknown protocol ip\n");
			exit(10);
		}
		if ((setsockopt(sndsock, pe->p_proto, IP_OPTIONS,
						optlist, oix-optlist)) < 0) {
			perror("traceroute: lsrr options");
			exit(5);
		}
	}

#ifdef SO_SNDBUF
	if (setsockopt(sndsock, SOL_SOCKET, SO_SNDBUF, (char *)&datalen,
		       sizeof(datalen)) < 0) {
		perror("traceroute: SO_SNDBUF");
		exit(6);
	}
#endif SO_SNDBUF
#ifdef IP_HDRINCL
	if (setsockopt(sndsock, IPPROTO_IP, IP_HDRINCL, (char *)&on,
		       sizeof(on)) < 0) {
		perror("traceroute: IP_HDRINCL");
		exit(6);
	}
#endif IP_HDRINCL
	if (options & SO_DEBUG)
		(void) setsockopt(sndsock, SOL_SOCKET, SO_DEBUG,
				  (char *)&on, sizeof(on));
	if (options & SO_DONTROUTE)
		(void) setsockopt(sndsock, SOL_SOCKET, SO_DONTROUTE,
				  (char *)&on, sizeof(on));

	if (source) {
		(void) bzero((char *)&from, sizeof(struct sockaddr));
		from.sin_family = AF_INET;
		from.sin_addr.s_addr = inet_addr(source);
		if (from.sin_addr.s_addr == -1) {
			Printf("traceroute: unknown host %s\n", source);
			exit(1);
		}
		outpacket->ip.ip_src = from.sin_addr;
#ifndef IP_HDRINCL
		if (bind(sndsock, (struct sockaddr *)&from, sizeof(from)) < 0) {
			perror ("traceroute: bind:");
			exit (1);
		}
#endif IP_HDRINCL
	}

	Fprintf(stderr, "traceroute to %s (%s)", hostname,
		inet_ntoa(to->sin_addr));
	if (source)
		Fprintf(stderr, " from %s", source);
	Fprintf(stderr, ", %d hops max\n", max_ttl);
	(void) fflush(stderr);

	for (ttl = 1; ttl <= max_ttl; ++ttl) {
		u_long lastaddr = 0;
		int got_there = 0;
		int unreachable = 0;

	again:
		Printf("%2d ", ttl);
		for (probe = 0; probe < nprobes; ++probe) {
			int cc;
			struct timeval tv;
			struct ip *ip;

			(void) gettimeofday(&tv, &tz);
			send_probe(++seq, ttl);
			while (cc = wait_for_reply(s, &from)) {
				if ((i = packet_ok(packet, cc, &from, seq))) {
					int dt = deltaT(&tv);
					if (from.sin_addr.s_addr != lastaddr) {
						print(packet, cc, &from);
						lastaddr = from.sin_addr.s_addr;
					}
					Printf("  %d ms", dt);
					switch(i - 1) {
					case ICMP_UNREACH_PORT:
#ifndef ARCHAIC
						ip = (struct ip *)packet;
						if (ip->ip_ttl <= 1)
							Printf(" !");
#endif ARCHAIC
						++got_there;
						break;
					case ICMP_UNREACH_NET:
						++unreachable;
						Printf(" !N");
						break;
					case ICMP_UNREACH_HOST:
						++unreachable;
						Printf(" !H");
						break;
					case ICMP_UNREACH_PROTOCOL:
						++got_there;
						Printf(" !P");
						break;
					case ICMP_UNREACH_NEEDFRAG:
						frag_err(&from);
						goto again;
					case ICMP_UNREACH_SRCFAIL:
						++unreachable;
						Printf(" !S");
						break;
					}
					break;
				}
			}
			if (cc == 0)
				Printf(" *");
			(void) fflush(stdout);
		}
		putchar('\n');
		if (got_there || unreachable >= nprobes-1)
			exit(0);
	}
}

wait_for_reply(sock, from)
	int sock;
	struct sockaddr_in *from;
{
	fd_set fds;
	struct timeval wait;
	int cc = 0;
	int fromlen = sizeof (*from);

	FD_ZERO(&fds);
	FD_SET(sock, &fds);
	wait.tv_sec = waittime; wait.tv_usec = 0;

	if (select(sock+1, &fds, (fd_set *)0, (fd_set *)0, &wait) > 0)
		cc=recvfrom(s, (char *)packet, sizeof(packet), 0,
			    (struct sockaddr *)from, &fromlen);

	return(cc);
}

/*
 * Received ICMP unreachable (fragmentation required and DF set).
 * If the ICMP error was from a "new" router, it'll contain the next-hop
 * MTU that we should use next.  Otherwise we'll just keep going in the
 * mtus[] table, trying until we hit a valid MTU.
 */

long nextmtu;	/* from ICMP error, set by packet_ok(), might be 0 */

frag_err(from)
struct sockaddr_in *from;
{
	int i;

	if (nextmtu > 0) {
		Printf("\nfragmentation required and DF set, next hop MTU = %d\n",
			nextmtu);
		datalen = nextmtu;
		for (i = 0; mtus[i] > 0; i++) {
			if (mtus[i] == nextmtu) {
				mtuptr = &mtus[i+1];	/* next one to try */
				return;
			}
		}
		Printf("(MTU not in table!)\n");
		
	} else {
		datalen = *mtuptr++;
		Printf("\nfragmentation required and DF set, trying new MTU = %d\n",
			datalen);
	}
}

send_probe(seq, ttl)
{
	struct opacket *op = outpacket;
	struct ip *ip = &op->ip;
	struct udphdr *up = &op->udp;
	int i;
	static int firsttime = 1;

again:
	ip->ip_off = 0 | IP_DF;		/* turn on don't-fragment flag */
	ip->ip_p = IPPROTO_UDP;
	ip->ip_len = datalen;
	ip->ip_ttl = ttl;

	up->uh_sport = htons(ident);
	up->uh_dport = htons(port+seq);
	up->uh_ulen = htons((u_short)(datalen - sizeof(struct ip)));
	up->uh_sum = 0;

	op->seq = seq;
	op->ttl = ttl;
	(void) gettimeofday(&op->tv, &tz);

	i = sendto(sndsock, (char *)outpacket, datalen, 0, &whereto,
		   sizeof(struct sockaddr));
	if (i < 0 || i != datalen)  {
		if (i < 0) {
			/*
			 * An errno of EMSGSIZE means we're writing too big a
			 * datagram for the interface.  We have to just decrease
			 * the packet size until we find one that works.
			 */
			if (errno == EMSGSIZE) {
				if (verbose)
				   Fprintf(stderr, "EMSGSIZE for %d\n", datalen);
				datalen = *mtuptr++;
				goto again;
			}
			perror("sendto");
		}
		Printf("traceroute: wrote %s %d chars, ret=%d\n", hostname,
			datalen, i);
		(void) fflush(stdout);
	}
	if (firsttime) {
		Fprintf(stderr, "outgoing MTU = %d\n", datalen);
		firsttime = 0;
	}
}


deltaT(tp)
	struct timeval *tp;
{
	struct timeval tv;

	(void) gettimeofday(&tv, &tz);
	tvsub(&tv, tp);
	return (tv.tv_sec*1000 + (tv.tv_usec + 500)/1000);
}


/*
 * Convert an ICMP "type" field to a printable string.
 */
char *
pr_type(t)
	u_char t;
{
	static char *ttab[] = {
	"Echo Reply",	"ICMP 1",	"ICMP 2",	"Dest Unreachable",
	"Source Quench", "Redirect",	"ICMP 6",	"ICMP 7",
	"Echo",		"ICMP 9",	"ICMP 10",	"Time Exceeded",
	"Param Problem", "Timestamp",	"Timestamp Reply", "Info Request",
	"Info Reply"
	};

	if(t > 16)
		return("OUT-OF-RANGE");

	return(ttab[t]);
}


packet_ok(buf, cc, from, seq)
	u_char *buf;
	int cc;
	struct sockaddr_in *from;
	int seq;
{
	register struct icmp *icp;
	u_char type, code;
	int hlen;
#ifndef ARCHAIC
	struct ip *ip;

	ip = (struct ip *) buf;
	hlen = ip->ip_hl << 2;
	if (cc < hlen + ICMP_MINLEN) {
		if (verbose)
			Printf("packet too short (%d bytes) from %s\n", cc,
				inet_ntoa(from->sin_addr));
		return (0);
	}
	cc -= hlen;
	icp = (struct icmp *)(buf + hlen);
#else
	icp = (struct icmp *)buf;
#endif ARCHAIC
	type = icp->icmp_type; code = icp->icmp_code;
	if ((type == ICMP_TIMXCEED && code == ICMP_TIMXCEED_INTRANS) ||
	    type == ICMP_UNREACH) {
		struct ip *hip;
		struct udphdr *up;

		nextmtu = htonl(icp->icmp_void); /* for frag_err() */
		hip = &icp->icmp_ip;
		hlen = hip->ip_hl << 2;
		up = (struct udphdr *)((u_char *)hip + hlen);
		if (hlen + 12 <= cc && hip->ip_p == IPPROTO_UDP &&
		    up->uh_sport == htons(ident) &&
		    up->uh_dport == htons(port+seq))
			return (type == ICMP_TIMXCEED? -1 : code+1);
	}
#ifndef ARCHAIC
	if (verbose) {
		int i;
		u_long *lp = (u_long *)&icp->icmp_ip;

		Printf("\n%d bytes from %s to %s", cc,
			inet_ntoa(from->sin_addr), inet_ntoa(ip->ip_dst));
		Printf(": icmp type %d (%s) code %d\n", type, pr_type(type),
		       icp->icmp_code);
		for (i = 4; i < cc ; i += sizeof(long))
			Printf("%2d: x%8.8lx\n", i, *lp++);
	}
#endif ARCHAIC
	return(0);
}


print(buf, cc, from)
	u_char *buf;
	int cc;
	struct sockaddr_in *from;
{
	struct ip *ip;
	int hlen;

	ip = (struct ip *) buf;
	hlen = ip->ip_hl << 2;
	cc -= hlen;

	if (nflag)
		Printf(" %s", inet_ntoa(from->sin_addr));
	else
		Printf(" %s (%s)", inetname(from->sin_addr),
		       inet_ntoa(from->sin_addr));

	if (verbose)
		Printf (" %d bytes to %s", cc, inet_ntoa (ip->ip_dst));
}


#ifdef notyet
/*
 * Checksum routine for Internet Protocol family headers (C Version)
 */
in_cksum(addr, len)
u_short *addr;
int len;
{
	register int nleft = len;
	register u_short *w = addr;
	register u_short answer;
	register int sum = 0;

	/*
	 *  Our algorithm is simple, using a 32 bit accumulator (sum),
	 *  we add sequential 16 bit words to it, and at the end, fold
	 *  back all the carry bits from the top 16 bits into the lower
	 *  16 bits.
	 */
	while (nleft > 1)  {
		sum += *w++;
		nleft -= 2;
	}

	/* mop up an odd byte, if necessary */
	if (nleft == 1)
		sum += *(u_char *)w;

	/*
	 * add back carry outs from top 16 bits to low 16 bits
	 */
	sum = (sum >> 16) + (sum & 0xffff);	/* add hi 16 to low 16 */
	sum += (sum >> 16);			/* add carry */
	answer = ~sum;				/* truncate to 16 bits */
	return (answer);
}
#endif notyet

/*
 * Subtract 2 timeval structs:  out = out - in.
 * Out is assumed to be >= in.
 */
tvsub(out, in)
register struct timeval *out, *in;
{
	if ((out->tv_usec -= in->tv_usec) < 0)   {
		out->tv_sec--;
		out->tv_usec += 1000000;
	}
	out->tv_sec -= in->tv_sec;
}


/*
 * Construct an Internet address representation.
 * If the nflag has been supplied, give 
 * numeric value, otherwise try for symbolic name.
 */
char *
inetname(in)
	struct in_addr in;
{
	register char *cp;
	static char line[50];
	struct hostent *hp;
	static char domain[MAXHOSTNAMELEN + 1];
	static int first = 1;
#ifndef	index	/* in case it's #defined as a macro */
	char *index();
#endif

	if (first && !nflag) {
		first = 0;
		if (gethostname(domain, MAXHOSTNAMELEN) == 0 &&
		    (cp = index(domain, '.')))
			(void) strcpy(domain, cp + 1);
		else
			domain[0] = 0;
	}
	cp = 0;
	if (!nflag && in.s_addr != INADDR_ANY) {
		hp = gethostbyaddr((char *)&in, sizeof (in), AF_INET);
		if (hp) {
			if ((cp = index(hp->h_name, '.')) &&
			    !strcmp(cp + 1, domain))
				*cp = 0;
			cp = hp->h_name;
		}
	}
	if (cp)
		(void) strcpy(line, cp);
	else {
		in.s_addr = ntohl(in.s_addr);
#define C(x)	((x) & 0xff)
		Sprintf(line, "%lu.%lu.%lu.%lu", C(in.s_addr >> 24),
			C(in.s_addr >> 16), C(in.s_addr >> 8), C(in.s_addr));
	}
	return (line);
}

usage()
{
	(void)fprintf(stderr, 
"usage: traceroute [-dnrv] [-m max_ttl] [-p port#] [-q nqueries]\n\t\
[-g gateway ] [-G gateway ] [-s src_addr] [-t tos] [-w wait] host [data size]\n");
	exit(1);
}