745 lines
21 KiB
C
745 lines
21 KiB
C
/* $NetBSD: input.c,v 1.18 1996/09/24 16:24:14 christos Exp $ */
|
|
|
|
/*
|
|
* Copyright (c) 1983, 1988, 1993
|
|
* The Regents of the University of California. All rights reserved.
|
|
*
|
|
* Redistribution and use in source and binary forms, with or without
|
|
* modification, are permitted provided that the following conditions
|
|
* are met:
|
|
* 1. Redistributions of source code must retain the above copyright
|
|
* notice, this list of conditions and the following disclaimer.
|
|
* 2. Redistributions in binary form must reproduce the above copyright
|
|
* notice, this list of conditions and the following disclaimer in the
|
|
* documentation and/or other materials provided with the distribution.
|
|
* 3. All advertising materials mentioning features or use of this software
|
|
* must display the following acknowledgement:
|
|
* This product includes software developed by the University of
|
|
* California, Berkeley and its contributors.
|
|
* 4. Neither the name of the University nor the names of its contributors
|
|
* may be used to endorse or promote products derived from this software
|
|
* without specific prior written permission.
|
|
*
|
|
* THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
|
|
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
|
|
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
|
|
* ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
|
|
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
|
|
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
|
|
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
|
|
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
|
|
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
|
|
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
|
|
* SUCH DAMAGE.
|
|
*/
|
|
|
|
#if !defined(lint) && !defined(sgi) && !defined(__NetBSD__)
|
|
static char sccsid[] = "@(#)input.c 8.1 (Berkeley) 6/5/93";
|
|
#elif defined(__NetBSD__)
|
|
static char rcsid[] = "$NetBSD: input.c,v 1.18 1996/09/24 16:24:14 christos Exp $";
|
|
#endif
|
|
|
|
#include "defs.h"
|
|
|
|
static void input(struct sockaddr_in *, struct interface*, struct rip *, int);
|
|
static void input_route(struct interface *, naddr,
|
|
naddr, naddr, naddr, struct netinfo *);
|
|
|
|
|
|
/* process RIP input
|
|
*/
|
|
void
|
|
read_rip(int sock,
|
|
struct interface *ifp)
|
|
{
|
|
struct sockaddr_in from;
|
|
int fromlen, cc;
|
|
union pkt_buf inbuf;
|
|
|
|
|
|
for (;;) {
|
|
fromlen = sizeof(from);
|
|
cc = recvfrom(sock, &inbuf, sizeof(inbuf), 0,
|
|
(struct sockaddr*)&from, &fromlen);
|
|
if (cc <= 0) {
|
|
if (cc < 0 && errno != EWOULDBLOCK)
|
|
LOGERR("recvfrom(rip)");
|
|
break;
|
|
}
|
|
if (fromlen != sizeof(struct sockaddr_in))
|
|
logbad(1,"impossible recvfrom(rip) fromlen=%d",
|
|
fromlen);
|
|
|
|
input(&from, ifp, &inbuf.rip, cc);
|
|
}
|
|
}
|
|
|
|
|
|
/* Process a RIP packet
|
|
*/
|
|
static void
|
|
input(struct sockaddr_in *from, /* received from this IP address */
|
|
struct interface *sifp, /* interface by which it arrived */
|
|
struct rip *rip,
|
|
int size)
|
|
{
|
|
# define FROM_NADDR from->sin_addr.s_addr
|
|
static naddr use_auth, bad_len, bad_mask;
|
|
static naddr unk_router, bad_router, bad_nhop;
|
|
|
|
struct interface *aifp; /* interface if via 1 hop */
|
|
struct rt_entry *rt;
|
|
struct netinfo *n, *lim;
|
|
struct interface *ifp1;
|
|
naddr gate, mask, v1_mask, dst, ddst_h;
|
|
int i;
|
|
|
|
aifp = iflookup(from->sin_addr.s_addr);
|
|
if (sifp == 0)
|
|
sifp = aifp;
|
|
|
|
if (sifp != 0)
|
|
sifp->int_state |= IS_ACTIVE;
|
|
|
|
trace_rip("Recv", "from", from, sifp, rip, size);
|
|
|
|
if (rip->rip_vers == 0) {
|
|
if (from->sin_addr.s_addr != bad_router)
|
|
msglog("RIP version 0, cmd %d, packet received"
|
|
" from %s",
|
|
rip->rip_cmd, naddr_ntoa(FROM_NADDR));
|
|
bad_router = from->sin_addr.s_addr;
|
|
return;
|
|
} else if (rip->rip_vers > RIPv2) {
|
|
rip->rip_vers = RIPv2;
|
|
}
|
|
if (size > MAXPACKETSIZE) {
|
|
if (from->sin_addr.s_addr != bad_router)
|
|
msglog("packet at least %d bytes too long received"
|
|
" from %s",
|
|
size-MAXPACKETSIZE, naddr_ntoa(FROM_NADDR));
|
|
bad_router = from->sin_addr.s_addr;
|
|
return;
|
|
}
|
|
|
|
n = rip->rip_nets;
|
|
lim = (struct netinfo *)((char*)rip + size);
|
|
|
|
/* Notice authentication.
|
|
* As required by section 4.2 in RFC 1723, discard authenticated
|
|
* RIPv2 messages, but only if configured for that silliness.
|
|
*
|
|
* RIPv2 authentication is lame, since snooping on the wire makes
|
|
* its simple passwords evident. Also, why authenticate queries?
|
|
* Why should a RIPv2 implementation with authentication disabled
|
|
* not be able to listen to RIPv2 packets with authenication, while
|
|
* RIPv1 systems will listen? Crazy!
|
|
*/
|
|
if (!auth_ok
|
|
&& rip->rip_vers == RIPv2
|
|
&& n < lim && n->n_family == RIP_AF_AUTH) {
|
|
if (from->sin_addr.s_addr != use_auth)
|
|
msglog("RIPv2 message with authentication"
|
|
" from %s discarded",
|
|
naddr_ntoa(FROM_NADDR));
|
|
use_auth = from->sin_addr.s_addr;
|
|
trace_pkt("discard authenticated RIPv2 message\n");
|
|
return;
|
|
}
|
|
|
|
switch (rip->rip_cmd) {
|
|
case RIPCMD_REQUEST:
|
|
/* did the request come from a router?
|
|
*/
|
|
if (from->sin_port == htons(RIP_PORT)) {
|
|
/* yes, ignore it if RIP is off so that it does not
|
|
* depend on us.
|
|
*/
|
|
if (rip_sock < 0) {
|
|
trace_pkt("ignore request while RIP off\n");
|
|
return;
|
|
}
|
|
|
|
/* Ignore the request if we talking to ourself
|
|
* (and not a remote gateway).
|
|
*/
|
|
if (ifwithaddr(FROM_NADDR, 0, 0) != 0) {
|
|
trace_pkt("discard our own RIP request\n");
|
|
return;
|
|
}
|
|
}
|
|
|
|
/* According to RFC 1723, we should ignore unathenticated
|
|
* queries. That is too silly to bother with. Sheesh!
|
|
* Are forwarding tables supposed to be secret? When
|
|
* a bad guy can infer them with test traffic?
|
|
* Maybe on firewalls you'd care, but not enough to
|
|
* give up the diagnostic facilities of remote probing.
|
|
*/
|
|
|
|
if (n >= lim
|
|
|| size%sizeof(*n) != sizeof(struct rip)%sizeof(*n)) {
|
|
if (from->sin_addr.s_addr != bad_len)
|
|
msglog("request of bad length (%d) from %s",
|
|
size, naddr_ntoa(FROM_NADDR));
|
|
bad_len = from->sin_addr.s_addr;
|
|
}
|
|
for (; n < lim; n++) {
|
|
n->n_metric = ntohl(n->n_metric);
|
|
|
|
/* A single entry with family RIP_AF_UNSPEC and
|
|
* metric HOPCNT_INFINITY means "all routes".
|
|
* We respond to routers only if we are acting
|
|
* as a supplier, or to anyone other than a router
|
|
* (i.e. a query).
|
|
*/
|
|
if (n->n_family == RIP_AF_UNSPEC
|
|
&& n->n_metric == HOPCNT_INFINITY
|
|
&& n == rip->rip_nets
|
|
&& n+1 == lim) {
|
|
if (from->sin_port != htons(RIP_PORT)) {
|
|
/* Answer a query from a utility
|
|
* program with all we know.
|
|
*/
|
|
supply(from, sifp, OUT_QUERY, 0,
|
|
rip->rip_vers);
|
|
return;
|
|
}
|
|
/* A router trying to prime its tables.
|
|
* Filter the answer in the about same way
|
|
* broadcasts are filtered.
|
|
*
|
|
* Only answer a router if we are a supplier
|
|
* to keep an unwary host that is just starting
|
|
* from picking us as a router. Respond with
|
|
* RIPv1 instead of RIPv2 if that is what we
|
|
* are broadcasting on the interface to keep
|
|
* the remote router from getting the wrong
|
|
* initial idea of the routes we send.
|
|
*/
|
|
if (!supplier
|
|
|| aifp == 0
|
|
|| (aifp->int_state & IS_PASSIVE)
|
|
|| (aifp->int_state & IS_ALIAS)
|
|
|| ((aifp->int_state & IS_NO_RIPV1_OUT)
|
|
&& (aifp->int_state&IS_NO_RIPV2_OUT)))
|
|
return;
|
|
|
|
supply(from, aifp, OUT_UNICAST, 0,
|
|
(aifp->int_state&IS_NO_RIPV1_OUT)
|
|
? RIPv2 : RIPv1);
|
|
return;
|
|
}
|
|
|
|
if (n->n_family != RIP_AF_INET) {
|
|
if (from->sin_addr.s_addr != bad_router)
|
|
msglog("request from %s"
|
|
" for unsupported (af %d) %s",
|
|
naddr_ntoa(FROM_NADDR),
|
|
ntohs(n->n_family),
|
|
naddr_ntoa(n->n_dst));
|
|
bad_router = from->sin_addr.s_addr;
|
|
return;
|
|
}
|
|
|
|
dst = n->n_dst;
|
|
if (!check_dst(dst)) {
|
|
if (from->sin_addr.s_addr != bad_router)
|
|
msglog("bad queried destination"
|
|
" %s from %s",
|
|
naddr_ntoa(dst),
|
|
naddr_ntoa(FROM_NADDR));
|
|
bad_router = from->sin_addr.s_addr;
|
|
return;
|
|
}
|
|
|
|
if (rip->rip_vers == RIPv1
|
|
|| 0 == (mask = ntohl(n->n_mask))
|
|
|| 0 != (ntohl(dst) & ~mask))
|
|
mask = ripv1_mask_host(dst,sifp);
|
|
|
|
rt = rtget(dst, mask);
|
|
if (!rt && dst != RIP_DEFAULT)
|
|
rt = rtfind(n->n_dst);
|
|
|
|
n->n_tag = 0;
|
|
n->n_nhop = 0;
|
|
if (rip->rip_vers == RIPv1) {
|
|
n->n_mask = 0;
|
|
} else {
|
|
n->n_mask = mask;
|
|
}
|
|
if (rt == 0) {
|
|
n->n_metric = HOPCNT_INFINITY;
|
|
} else {
|
|
n->n_metric = rt->rt_metric+1;
|
|
n->n_metric += (sifp!=0)?sifp->int_metric : 1;
|
|
if (n->n_metric > HOPCNT_INFINITY)
|
|
n->n_metric = HOPCNT_INFINITY;
|
|
if (rip->rip_vers != RIPv1) {
|
|
n->n_tag = rt->rt_tag;
|
|
if (sifp != 0
|
|
&& on_net(rt->rt_gate,
|
|
sifp->int_net,
|
|
sifp->int_mask)
|
|
&& rt->rt_gate != sifp->int_addr)
|
|
n->n_nhop = rt->rt_gate;
|
|
}
|
|
}
|
|
HTONL(n->n_metric);
|
|
}
|
|
/* Answer about specific routes.
|
|
* Only answer a router if we are a supplier
|
|
* to keep an unwary host that is just starting
|
|
* from picking us an a router.
|
|
*/
|
|
rip->rip_cmd = RIPCMD_RESPONSE;
|
|
rip->rip_res1 = 0;
|
|
if (rip->rip_vers != RIPv1)
|
|
rip->rip_vers = RIPv2;
|
|
if (from->sin_port != htons(RIP_PORT)) {
|
|
/* query */
|
|
(void)output(OUT_QUERY, from, sifp, rip, size);
|
|
} else if (supplier) {
|
|
(void)output(OUT_UNICAST, from, sifp, rip, size);
|
|
}
|
|
return;
|
|
|
|
case RIPCMD_TRACEON:
|
|
case RIPCMD_TRACEOFF:
|
|
/* verify message came from a privileged port */
|
|
if (ntohs(from->sin_port) > IPPORT_RESERVED) {
|
|
msglog("trace command from untrusted port on %s",
|
|
naddr_ntoa(FROM_NADDR));
|
|
return;
|
|
}
|
|
if (aifp == 0) {
|
|
msglog("trace command from unknown router %s",
|
|
naddr_ntoa(FROM_NADDR));
|
|
return;
|
|
}
|
|
if (rip->rip_cmd == RIPCMD_TRACEON) {
|
|
rip->rip_tracefile[size-4] = '\0';
|
|
trace_on((char*)rip->rip_tracefile, 0);
|
|
} else {
|
|
trace_off("tracing turned off by %s\n",
|
|
naddr_ntoa(FROM_NADDR));
|
|
}
|
|
return;
|
|
|
|
case RIPCMD_RESPONSE:
|
|
if (size%sizeof(*n) != sizeof(struct rip)%sizeof(*n)) {
|
|
if (from->sin_addr.s_addr != bad_len)
|
|
msglog("response of bad length (%d) from %s",
|
|
size, naddr_ntoa(FROM_NADDR));
|
|
bad_len = from->sin_addr.s_addr;
|
|
}
|
|
|
|
/* verify message came from a router */
|
|
if (from->sin_port != ntohs(RIP_PORT)) {
|
|
trace_pkt("discard RIP response from unknown port\n");
|
|
return;
|
|
}
|
|
|
|
if (rip_sock < 0) {
|
|
trace_pkt("discard response while RIP off\n");
|
|
return;
|
|
}
|
|
|
|
/* Are we talking to ourself or a remote gateway?
|
|
*/
|
|
ifp1 = ifwithaddr(FROM_NADDR, 0, 1);
|
|
if (ifp1) {
|
|
if (ifp1->int_state & IS_REMOTE) {
|
|
if (ifp1->int_state & IS_PASSIVE) {
|
|
msglog("bogus input from %s on"
|
|
" supposedly passive %s",
|
|
naddr_ntoa(FROM_NADDR),
|
|
ifp1->int_name);
|
|
} else {
|
|
ifp1->int_act_time = now.tv_sec;
|
|
if (if_ok(ifp1, "remote "))
|
|
addrouteforif(ifp1);
|
|
}
|
|
} else {
|
|
trace_pkt("discard our own RIP response\n");
|
|
}
|
|
return;
|
|
}
|
|
|
|
/* Check the router from which message originated. We accept
|
|
* routing packets from routers directly connected via
|
|
* broadcast or point-to-point networks, and from
|
|
* those listed in /etc/gateways.
|
|
*/
|
|
if (!aifp) {
|
|
if (from->sin_addr.s_addr != unk_router)
|
|
msglog("discard packet from unknown router %s"
|
|
" or via unidentified interface",
|
|
naddr_ntoa(FROM_NADDR));
|
|
unk_router = from->sin_addr.s_addr;
|
|
return;
|
|
}
|
|
if (aifp->int_state & IS_PASSIVE) {
|
|
trace_act("discard packet from %s"
|
|
" via passive interface %s\n",
|
|
naddr_ntoa(FROM_NADDR),
|
|
aifp->int_name);
|
|
return;
|
|
}
|
|
|
|
/* Check required version
|
|
*/
|
|
if (((aifp->int_state & IS_NO_RIPV1_IN)
|
|
&& rip->rip_vers == RIPv1)
|
|
|| ((aifp->int_state & IS_NO_RIPV2_IN)
|
|
&& rip->rip_vers != RIPv1)) {
|
|
trace_pkt("discard RIPv%d response\n",
|
|
rip->rip_vers);
|
|
return;
|
|
}
|
|
|
|
/* Ignore routes via dead interface.
|
|
*/
|
|
if (aifp->int_state & IS_BROKE) {
|
|
trace_pkt("discard response via broken interface %s\n",
|
|
aifp->int_name);
|
|
return;
|
|
}
|
|
|
|
/* Authenticate the packet if we have a secret.
|
|
*/
|
|
if (aifp->int_passwd[0] != '\0') {
|
|
if (n >= lim
|
|
|| n->n_family != RIP_AF_AUTH
|
|
|| ((struct netauth*)n)->a_type != RIP_AUTH_PW) {
|
|
if (from->sin_addr.s_addr != use_auth)
|
|
msglog("missing password from %s",
|
|
naddr_ntoa(FROM_NADDR));
|
|
use_auth = from->sin_addr.s_addr;
|
|
return;
|
|
|
|
} else if (0 != bcmp(((struct netauth*)n)->au.au_pw,
|
|
aifp->int_passwd,
|
|
sizeof(aifp->int_passwd))) {
|
|
if (from->sin_addr.s_addr != use_auth)
|
|
msglog("bad password from %s",
|
|
naddr_ntoa(FROM_NADDR));
|
|
use_auth = from->sin_addr.s_addr;
|
|
return;
|
|
}
|
|
}
|
|
|
|
auth_ok:
|
|
|
|
for (; n < lim; n++) {
|
|
if (n->n_family == RIP_AF_AUTH)
|
|
continue;
|
|
|
|
NTOHL(n->n_metric);
|
|
dst = n->n_dst;
|
|
if (n->n_family != RIP_AF_INET
|
|
&& (n->n_family != RIP_AF_UNSPEC
|
|
|| dst != RIP_DEFAULT)) {
|
|
if (from->sin_addr.s_addr != bad_router)
|
|
msglog("route from %s to unsupported"
|
|
" address family %d,"
|
|
" destination %s",
|
|
naddr_ntoa(FROM_NADDR),
|
|
n->n_family,
|
|
naddr_ntoa(dst));
|
|
bad_router = from->sin_addr.s_addr;
|
|
continue;
|
|
}
|
|
if (!check_dst(dst)) {
|
|
if (from->sin_addr.s_addr != bad_router)
|
|
msglog("bad destination %s from %s",
|
|
naddr_ntoa(dst),
|
|
naddr_ntoa(FROM_NADDR));
|
|
bad_router = from->sin_addr.s_addr;
|
|
return;
|
|
}
|
|
if (n->n_metric == 0
|
|
|| n->n_metric > HOPCNT_INFINITY) {
|
|
if (from->sin_addr.s_addr != bad_router)
|
|
msglog("bad metric %d from %s"
|
|
" for destination %s",
|
|
n->n_metric,
|
|
naddr_ntoa(FROM_NADDR),
|
|
naddr_ntoa(dst));
|
|
bad_router = from->sin_addr.s_addr;
|
|
return;
|
|
}
|
|
|
|
/* Notice the next-hop.
|
|
*/
|
|
gate = from->sin_addr.s_addr;
|
|
if (n->n_nhop != 0) {
|
|
if (rip->rip_vers == RIPv2) {
|
|
n->n_nhop = 0;
|
|
} else {
|
|
/* Use it only if it is valid. */
|
|
if (on_net(n->n_nhop,
|
|
aifp->int_net, aifp->int_mask)
|
|
&& check_dst(n->n_nhop)) {
|
|
gate = n->n_nhop;
|
|
} else {
|
|
if (bad_nhop != from->sin_addr.s_addr)
|
|
msglog("router %s to %s has"
|
|
" bad next hop %s",
|
|
naddr_ntoa(FROM_NADDR),
|
|
naddr_ntoa(dst),
|
|
naddr_ntoa(n->n_nhop));
|
|
bad_nhop = from->sin_addr.s_addr;
|
|
n->n_nhop = 0;
|
|
}
|
|
}
|
|
}
|
|
|
|
if (rip->rip_vers == RIPv1
|
|
|| 0 == (mask = ntohl(n->n_mask))) {
|
|
mask = ripv1_mask_host(dst,aifp);
|
|
} else if ((ntohl(dst) & ~mask) != 0) {
|
|
if (bad_mask != from->sin_addr.s_addr) {
|
|
msglog("router %s sent bad netmask"
|
|
" %#x with %s",
|
|
naddr_ntoa(FROM_NADDR),
|
|
mask,
|
|
naddr_ntoa(dst));
|
|
bad_mask = from->sin_addr.s_addr;
|
|
}
|
|
continue;
|
|
}
|
|
if (rip->rip_vers == RIPv1)
|
|
n->n_tag = 0;
|
|
|
|
/* Adjust metric according to incoming interface..
|
|
*/
|
|
n->n_metric += aifp->int_metric;
|
|
if (n->n_metric > HOPCNT_INFINITY)
|
|
n->n_metric = HOPCNT_INFINITY;
|
|
|
|
/* Recognize and ignore a default route we faked
|
|
* which is being sent back to us by a machine with
|
|
* broken split-horizon.
|
|
* Be a little more paranoid than that, and reject
|
|
* default routes with the same metric we advertised.
|
|
*/
|
|
if (aifp->int_d_metric != 0
|
|
&& dst == RIP_DEFAULT
|
|
&& n->n_metric >= aifp->int_d_metric)
|
|
continue;
|
|
|
|
/* We can receive aggregated RIPv2 routes that must
|
|
* be broken down before they are transmitted by
|
|
* RIPv1 via an interface on a subnet.
|
|
* We might also receive the same routes aggregated
|
|
* via other RIPv2 interfaces.
|
|
* This could cause duplicate routes to be sent on
|
|
* the RIPv1 interfaces. "Longest matching variable
|
|
* length netmasks" lets RIPv2 listeners understand,
|
|
* but breaking down the aggregated routes for RIPv1
|
|
* listeners can produce duplicate routes.
|
|
*
|
|
* Breaking down aggregated routes here bloats
|
|
* the daemon table, but does not hurt the kernel
|
|
* table, since routes are always aggregated for
|
|
* the kernel.
|
|
*
|
|
* Notice that this does not break down network
|
|
* routes corresponding to subnets. This is part
|
|
* of the defense against RS_NET_SYN.
|
|
*/
|
|
if (have_ripv1_out
|
|
&& (v1_mask = ripv1_mask_net(dst,0)) > mask
|
|
&& (((rt = rtget(dst,mask)) == 0
|
|
|| !(rt->rt_state & RS_NET_SYN)))) {
|
|
ddst_h = v1_mask & -v1_mask;
|
|
i = (v1_mask & ~mask)/ddst_h;
|
|
if (i >= 511) {
|
|
/* Punt if we would have to generate
|
|
* an unreasonable number of routes.
|
|
*/
|
|
#ifdef DEBUG
|
|
msglog("accept %s from %s as 1"
|
|
" instead of %d routes",
|
|
addrname(dst,mask,0),
|
|
naddr_ntoa(FROM_NADDR),
|
|
i+1);
|
|
#endif
|
|
i = 0;
|
|
} else {
|
|
mask = v1_mask;
|
|
}
|
|
} else {
|
|
i = 0;
|
|
}
|
|
|
|
for (;;) {
|
|
input_route(aifp, FROM_NADDR,
|
|
dst, mask, gate, n);
|
|
if (i-- == 0)
|
|
break;
|
|
dst = htonl(ntohl(dst) + ddst_h);
|
|
}
|
|
}
|
|
break;
|
|
}
|
|
}
|
|
|
|
|
|
/* Process a single input route.
|
|
*/
|
|
static void
|
|
input_route(struct interface *ifp,
|
|
naddr from,
|
|
naddr dst,
|
|
naddr mask,
|
|
naddr gate,
|
|
struct netinfo *n)
|
|
{
|
|
int i;
|
|
struct rt_entry *rt;
|
|
struct rt_spare *rts, *rts0;
|
|
struct interface *ifp1;
|
|
time_t new_time;
|
|
|
|
|
|
/* See if the other guy is telling us to send our packets to him.
|
|
* Sometimes network routes arrive over a point-to-point link for
|
|
* the network containing the address(es) of the link.
|
|
*
|
|
* If our interface is broken, switch to using the other guy.
|
|
*/
|
|
ifp1 = ifwithaddr(dst, 1, 1);
|
|
if (ifp1 != 0
|
|
&& !(ifp1->int_state & IS_BROKE))
|
|
return;
|
|
|
|
/* Look for the route in our table.
|
|
*/
|
|
rt = rtget(dst, mask);
|
|
|
|
/* Consider adding the route if we do not already have it.
|
|
*/
|
|
if (rt == 0) {
|
|
/* Ignore unknown routes being poisoned.
|
|
*/
|
|
if (n->n_metric == HOPCNT_INFINITY)
|
|
return;
|
|
|
|
/* Ignore the route if it points to us */
|
|
if (n->n_nhop != 0
|
|
&& 0 != ifwithaddr(n->n_nhop, 1, 0))
|
|
return;
|
|
|
|
/* If something has not gone crazy and tried to fill
|
|
* our memory, accept the new route.
|
|
*/
|
|
if (total_routes < MAX_ROUTES)
|
|
rtadd(dst, mask, gate, from, n->n_metric,
|
|
n->n_tag, 0, ifp);
|
|
return;
|
|
}
|
|
|
|
/* We already know about the route. Consider this update.
|
|
*
|
|
* If (rt->rt_state & RS_NET_SYN), then this route
|
|
* is the same as a network route we have inferred
|
|
* for subnets we know, in order to tell RIPv1 routers
|
|
* about the subnets.
|
|
*
|
|
* It is impossible to tell if the route is coming
|
|
* from a distant RIPv2 router with the standard
|
|
* netmask because that router knows about the entire
|
|
* network, or if it is a round-about echo of a
|
|
* synthetic, RIPv1 network route of our own.
|
|
* The worst is that both kinds of routes might be
|
|
* received, and the bad one might have the smaller
|
|
* metric. Partly solve this problem by never
|
|
* aggregating into such a route. Also keep it
|
|
* around as long as the interface exists.
|
|
*/
|
|
|
|
rts0 = rt->rt_spares;
|
|
for (rts = rts0, i = NUM_SPARES; i != 0; i--, rts++) {
|
|
if (rts->rts_router == from)
|
|
break;
|
|
/* Note the worst slot to reuse,
|
|
* other than the current slot.
|
|
*/
|
|
if (rts0 == rt->rt_spares
|
|
|| BETTER_LINK(rt, rts0, rts))
|
|
rts0 = rts;
|
|
}
|
|
if (i != 0) {
|
|
/* Found the router
|
|
*/
|
|
int old_metric = rts->rts_metric;
|
|
|
|
/* Keep poisoned routes around only long enough to pass
|
|
* the poison on. Get a new timestamp for good routes.
|
|
*/
|
|
new_time =((old_metric == HOPCNT_INFINITY)
|
|
? rts->rts_time
|
|
: now.tv_sec);
|
|
|
|
/* If this is an update for the router we currently prefer,
|
|
* then note it.
|
|
*/
|
|
if (i == NUM_SPARES) {
|
|
rtchange(rt,rt->rt_state, gate,rt->rt_router,
|
|
n->n_metric, n->n_tag, ifp, new_time, 0);
|
|
/* If the route got worse, check for something better.
|
|
*/
|
|
if (n->n_metric > old_metric)
|
|
rtswitch(rt, 0);
|
|
return;
|
|
}
|
|
|
|
/* This is an update for a spare route.
|
|
* Finished if the route is unchanged.
|
|
*/
|
|
if (rts->rts_gate == gate
|
|
&& old_metric == n->n_metric
|
|
&& rts->rts_tag == n->n_tag) {
|
|
rts->rts_time = new_time;
|
|
return;
|
|
}
|
|
|
|
} else {
|
|
/* The update is for a route we know about,
|
|
* but not from a familiar router.
|
|
*
|
|
* Ignore the route if it points to us.
|
|
*/
|
|
if (n->n_nhop != 0
|
|
&& 0 != ifwithaddr(n->n_nhop, 1, 0))
|
|
return;
|
|
|
|
rts = rts0;
|
|
|
|
/* Save the route as a spare only if it has
|
|
* a better metric than our worst spare.
|
|
* This also ignores poisoned routes (those
|
|
* received with metric HOPCNT_INFINITY).
|
|
*/
|
|
if (n->n_metric >= rts->rts_metric)
|
|
return;
|
|
|
|
new_time = now.tv_sec;
|
|
}
|
|
|
|
trace_upslot(rt, rts, gate, from, ifp, n->n_metric,n->n_tag, new_time);
|
|
|
|
rts->rts_gate = gate;
|
|
rts->rts_router = from;
|
|
rts->rts_metric = n->n_metric;
|
|
rts->rts_tag = n->n_tag;
|
|
rts->rts_time = new_time;
|
|
rts->rts_ifp = ifp;
|
|
|
|
/* try to switch to a better route */
|
|
rtswitch(rt, rts);
|
|
}
|