freebsd-src/usr.sbin/route6d/route6d.c
Hajimu UMEMOTO bcfb36405d - do not check/advertise cloned routes, nor dynamic routes
- cast size_t on printing (be friendly with alpha)

Obtained from:	KAME
2000-11-27 17:44:53 +00:00

3169 lines
78 KiB
C

/* $FreeBSD$ */
/* $KAME: route6d.c,v 1.37 2000/10/10 13:02:30 itojun Exp $ */
/*
* Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
* 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. Neither the name of the project 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 PROJECT 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 PROJECT 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.
*/
#ifndef lint
static char _rcsid[] = "$KAME: route6d.c,v 1.37 2000/10/10 13:02:30 itojun Exp $";
#endif
#include <stdio.h>
#include <time.h>
#include <unistd.h>
#include <stdlib.h>
#include <string.h>
#include <signal.h>
#ifdef __STDC__
#include <stdarg.h>
#else
#include <varargs.h>
#endif
#include <syslog.h>
#include <stddef.h>
#include <errno.h>
#include <err.h>
#include <sys/types.h>
#include <sys/param.h>
#include <sys/file.h>
#include <sys/socket.h>
#include <sys/ioctl.h>
#include <sys/sysctl.h>
#include <sys/uio.h>
#include <net/if.h>
#if defined(__FreeBSD__) && __FreeBSD__ >= 3
#include <net/if_var.h>
#endif /* __FreeBSD__ >= 3 */
#define KERNEL 1
#define _KERNEL 1
#include <net/route.h>
#undef KERNEL
#undef _KERNEL
#include <netinet/in.h>
#include <netinet/in_var.h>
#include <netinet/ip6.h>
#include <netinet/udp.h>
#include <netdb.h>
#ifdef HAVE_GETIFADDRS
#include <ifaddrs.h>
#endif
#include <arpa/inet.h>
#include "route6d.h"
#define MAXFILTER 40
#ifdef DEBUG
#define INIT_INTERVAL6 6
#else
#define INIT_INTERVAL6 10 /* Wait to submit a initial riprequest */
#endif
/* alignment constraint for routing socket */
#define ROUNDUP(a) \
((a) > 0 ? (1 + (((a) - 1) | (sizeof(long) - 1))) : sizeof(long))
#define ADVANCE(x, n) (x += ROUNDUP((n)->sa_len))
/*
* Following two macros are highly depending on KAME Release
*/
#define IN6_LINKLOCAL_IFINDEX(addr) \
((addr).s6_addr[2] << 8 | (addr).s6_addr[3])
#define SET_IN6_LINKLOCAL_IFINDEX(addr, index) \
do { \
(addr).s6_addr[2] = ((index) >> 8) & 0xff; \
(addr).s6_addr[3] = (index) & 0xff; \
} while (0)
struct ifc { /* Configuration of an interface */
char *ifc_name; /* if name */
struct ifc *ifc_next;
int ifc_index; /* if index */
int ifc_mtu; /* if mtu */
int ifc_metric; /* if metric */
short ifc_flags; /* flags */
struct in6_addr ifc_mylladdr; /* my link-local address */
struct sockaddr_in6 ifc_ripsin; /* rip multicast address */
struct iff *ifc_filter; /* filter structure */
struct ifac *ifc_addr; /* list of AF_INET6 addresses */
int ifc_joined; /* joined to ff02::9 */
};
struct ifac { /* Adddress associated to an interface */
struct ifc *ifa_conf; /* back pointer */
struct ifac *ifa_next;
struct in6_addr ifa_addr; /* address */
struct in6_addr ifa_raddr; /* remote address, valid in p2p */
int ifa_plen; /* prefix length */
};
struct iff {
int iff_type;
struct in6_addr iff_addr;
int iff_plen;
struct iff *iff_next;
};
struct ifc *ifc;
int nifc; /* number of valid ifc's */
struct ifc **index2ifc;
int nindex2ifc;
struct ifc *loopifcp = NULL; /* pointing to loopback */
int loopifindex = 0; /* ditto */
fd_set sockvec; /* vector to select() for receiving */
int rtsock; /* the routing socket */
int ripsock; /* socket to send/receive RIP datagram */
struct rip6 *ripbuf; /* packet buffer for sending */
/*
* Maintain the routes in a linked list. When the number of the routes
* grows, somebody would like to introduce a hash based or a radix tree
* based strucutre. I believe the number of routes handled by RIP is
* limited and I don't have to manage a complex data structure, however.
*
* One of the major drawbacks of the linear linked list is the difficulty
* of representing the relationship between a couple of routes. This may
* be a significant problem when we have to support route aggregation with
* supressing the specifices covered by the aggregate.
*/
struct riprt {
struct riprt *rrt_next; /* next destination */
struct riprt *rrt_same; /* same destination - future use */
struct netinfo6 rrt_info; /* network info */
struct in6_addr rrt_gw; /* gateway */
u_long rrt_flags; /* kernel routing table flags */
u_long rrt_rflags; /* route6d routing table flags */
time_t rrt_t; /* when the route validated */
int rrt_index; /* ifindex from which this route got */
};
struct riprt *riprt = 0;
int dflag = 0; /* debug flag */
int qflag = 0; /* quiet flag */
int nflag = 0; /* don't update kernel routing table */
int aflag = 0; /* age out even the statically defined routes */
int hflag = 0; /* don't split horizon */
int lflag = 0; /* exchange site local routes */
int sflag = 0; /* announce static routes w/ split horizon */
int Sflag = 0; /* announce static routes to every interface */
unsigned long routetag = 0; /* route tag attached on originating case */
char *filter[MAXFILTER];
int filtertype[MAXFILTER];
int nfilter = 0;
pid_t pid;
struct sockaddr_storage ripsin;
struct rtentry rtentry;
int interval = 1;
time_t nextalarm = 0;
time_t sup_trig_update = 0;
FILE *rtlog = NULL;
int logopened = 0;
static u_long seq = 0;
#define RRTF_AGGREGATE 0x08000000
#define RRTF_NOADVERTISE 0x10000000
#define RRTF_NH_NOT_LLADDR 0x20000000
#define RRTF_SENDANYWAY 0x40000000
#define RRTF_CHANGED 0x80000000
int main __P((int, char **));
void ripalarm __P((int));
void riprecv __P((void));
void ripsend __P((struct ifc *, struct sockaddr_in6 *, int));
void init __P((void));
void sockopt __P((struct ifc *));
void ifconfig __P((void));
void ifconfig1 __P((const char *, const struct sockaddr *, struct ifc *, int));
void rtrecv __P((void));
int rt_del __P((const struct sockaddr_in6 *, const struct sockaddr_in6 *,
const struct sockaddr_in6 *));
int rt_deladdr __P((struct ifc *, const struct sockaddr_in6 *,
const struct sockaddr_in6 *));
void filterconfig __P((void));
int getifmtu __P((int));
const char *rttypes __P((struct rt_msghdr *rtm));
const char *rtflags __P((struct rt_msghdr *rtm));
const char *ifflags __P((int flags));
void ifrt __P((struct ifc *, int));
void ifrt_p2p __P((struct ifc *, int));
void applymask __P((struct in6_addr *, struct in6_addr *));
void applyplen __P((struct in6_addr *, int));
void ifrtdump __P((int));
void ifdump __P((int));
void ifdump0 __P((FILE *, const struct ifc *));
void rtdump __P((int));
void rt_entry __P((struct rt_msghdr *, int));
void rtdexit __P((int));
void riprequest __P((struct ifc *, struct netinfo6 *, int, struct sockaddr_in6 *));
void ripflush __P((struct ifc *, struct sockaddr_in6 *));
void sendrequest __P((struct ifc *));
int mask2len __P((const struct in6_addr *, int));
int sendpacket __P((struct sockaddr_in6 *, int));
int addroute __P((struct riprt *, const struct in6_addr *, struct ifc *));
int delroute __P((struct netinfo6 *, struct in6_addr *));
struct in6_addr *getroute __P((struct netinfo6 *, struct in6_addr *));
void krtread __P((int));
int tobeadv __P((struct riprt *, struct ifc *));
char *allocopy __P((char *));
char *hms __P((void));
const char *inet6_n2p __P((const struct in6_addr *));
struct ifac *ifa_match __P((const struct ifc *, const struct in6_addr *, int));
struct in6_addr *plen2mask __P((int));
struct riprt *rtsearch __P((struct netinfo6 *));
int ripinterval __P((int));
time_t ripsuptrig __P((void));
void fatal __P((const char *, ...))
__attribute__((__format__(__printf__, 1, 2)));
void trace __P((int, const char *, ...))
__attribute__((__format__(__printf__, 2, 3)));
void tracet __P((int, const char *, ...))
__attribute__((__format__(__printf__, 2, 3)));
unsigned int if_maxindex __P((void));
struct ifc *ifc_find __P((char *));
struct iff *iff_find __P((struct ifc *, int));
void setindex2ifc __P((int, struct ifc *));
#define MALLOC(type) ((type *)malloc(sizeof(type)))
int
main(argc, argv)
int argc;
char **argv;
{
int ch;
int error = 0;
struct ifc *ifcp;
sigset_t mask, omask;
FILE *pidfile;
char *progname;
char *ep;
progname = strrchr(*argv, '/');
if (progname)
progname++;
else
progname = *argv;
pid = getpid();
while ((ch = getopt(argc, argv, "A:N:O:R:T:L:t:adDhlnqsS")) != EOF) {
switch (ch) {
case 'A':
case 'N':
case 'O':
case 'T':
case 'L':
if (nfilter >= MAXFILTER) {
fatal("Exceeds MAXFILTER");
/*NOTREACHED*/
}
filtertype[nfilter] = ch;
filter[nfilter++] = allocopy(optarg);
break;
case 't':
ep = NULL;
routetag = strtoul(optarg, &ep, 0);
if (!ep || *ep != '\0' || (routetag & ~0xffff) != 0) {
fatal("invalid route tag");
/*NOTREACHED*/
}
break;
case 'R':
if ((rtlog = fopen(optarg, "w")) == NULL) {
fatal("Can not write to routelog");
/*NOTREACHED*/
}
break;
#define FLAG(c, flag, n) case c: do { flag = n; break; } while(0)
FLAG('a', aflag, 1); break;
FLAG('d', dflag, 1); break;
FLAG('D', dflag, 2); break;
FLAG('h', hflag, 1); break;
FLAG('l', lflag, 1); break;
FLAG('n', nflag, 1); break;
FLAG('q', qflag, 1); break;
FLAG('s', sflag, 1); break;
FLAG('S', Sflag, 1); break;
#undef FLAG
default:
fatal("Invalid option specified, terminating");
/*NOTREACHED*/
}
}
argc -= optind;
argv += optind;
if (argc > 0)
fatal("bogus extra arguments");
if (geteuid()) {
nflag = 1;
fprintf(stderr, "No kernel update is allowed\n");
}
openlog(progname, LOG_NDELAY|LOG_PID, LOG_DAEMON);
logopened++;
init();
ifconfig();
for (ifcp = ifc; ifcp; ifcp = ifcp->ifc_next) {
if (ifcp->ifc_index < 0) {
fprintf(stderr,
"No ifindex found at %s (no link-local address?)\n",
ifcp->ifc_name);
error++;
}
}
if (error)
exit(1);
if (loopifcp == NULL)
fatal("No loopback found");
loopifindex = loopifcp->ifc_index;
for (ifcp = ifc; ifcp; ifcp = ifcp->ifc_next)
ifrt(ifcp, 0);
filterconfig();
krtread(0);
if (dflag)
ifrtdump(0);
if (dflag == 0) {
#if 1
if (daemon(0, 0) < 0)
fatal("daemon");
#else
if (fork())
exit(0);
if (setsid() < 0)
fatal("setid");
#endif
}
pid = getpid();
if ((pidfile = fopen(ROUTE6D_PID, "w")) != NULL) {
fprintf(pidfile, "%d\n", pid);
fclose(pidfile);
}
if ((ripbuf = (struct rip6 *)malloc(RIP6_MAXMTU)) == NULL)
fatal("malloc");
memset(ripbuf, 0, RIP6_MAXMTU);
ripbuf->rip6_cmd = RIP6_RESPONSE;
ripbuf->rip6_vers = RIP6_VERSION;
ripbuf->rip6_res1[0] = 0;
ripbuf->rip6_res1[1] = 0;
if (signal(SIGALRM, ripalarm) == SIG_ERR)
fatal("signal: SIGALRM");
if (signal(SIGQUIT, rtdexit) == SIG_ERR)
fatal("signal: SIGQUIT");
if (signal(SIGTERM, rtdexit) == SIG_ERR)
fatal("signal: SIGTERM");
if (signal(SIGUSR1, ifrtdump) == SIG_ERR)
fatal("signal: SIGUSR1");
if (signal(SIGHUP, ifrtdump) == SIG_ERR)
fatal("signal: SIGHUP");
if (signal(SIGINT, ifrtdump) == SIG_ERR)
fatal("signal: SIGINT");
/*
* To avoid rip packet congestion (not on a cable but in this
* process), wait for a moment to send the first RIP6_RESPONSE
* packets.
*/
alarm(ripinterval(INIT_INTERVAL6));
for (ifcp = ifc; ifcp; ifcp = ifcp->ifc_next) {
if (ifcp->ifc_index > 0 && (ifcp->ifc_flags & IFF_UP))
sendrequest(ifcp);
}
syslog(LOG_INFO, "**** Started ****");
sigemptyset(&mask);
sigaddset(&mask, SIGALRM);
while (1) {
fd_set recvec;
FD_COPY(&sockvec, &recvec);
switch (select(FD_SETSIZE, &recvec, 0, 0, 0)) {
case -1:
if (errno == EINTR)
continue;
fatal("select");
case 0:
continue;
default:
if (FD_ISSET(ripsock, &recvec)) {
sigprocmask(SIG_BLOCK, &mask, &omask);
riprecv();
sigprocmask(SIG_SETMASK, &omask, NULL);
}
if (FD_ISSET(rtsock, &recvec)) {
sigprocmask(SIG_BLOCK, &mask, &omask);
rtrecv();
sigprocmask(SIG_SETMASK, &omask, NULL);
}
}
}
}
/*
* gracefully exits after resetting sockopts.
*/
/* ARGSUSED */
void
rtdexit(sig)
int sig;
{
struct riprt *rrt;
alarm(0);
for (rrt = riprt; rrt; rrt = rrt->rrt_next) {
if (rrt->rrt_rflags & RRTF_AGGREGATE) {
delroute(&rrt->rrt_info, &rrt->rrt_gw);
}
}
close(ripsock);
close(rtsock);
syslog(LOG_INFO, "**** Terminated ****");
closelog();
exit(1);
}
/*
* Called periodically:
* 1. age out the learned route. remove it if necessary.
* 2. submit RIP6_RESPONSE packets.
* Invoked in every SUPPLY_INTERVAL6 (30) seconds. I believe we don't have
* to invoke this function in every 1 or 5 or 10 seconds only to age the
* routes more precisely.
*/
/* ARGSUSED */
void
ripalarm(sig)
int sig;
{
struct ifc *ifcp;
struct riprt *rrt, *rrt_prev, *rrt_next;
time_t t_lifetime, t_holddown;
/* age the RIP routes */
rrt_prev = 0;
t_lifetime = time(NULL) - RIP_LIFETIME;
t_holddown = t_lifetime - RIP_HOLDDOWN;
for (rrt = riprt; rrt; rrt = rrt_next) {
rrt_next = rrt->rrt_next;
if (rrt->rrt_t == 0) {
rrt_prev = rrt;
continue;
}
if (rrt->rrt_t < t_holddown) {
if (rrt_prev) {
rrt_prev->rrt_next = rrt->rrt_next;
} else {
riprt = rrt->rrt_next;
}
delroute(&rrt->rrt_info, &rrt->rrt_gw);
free(rrt);
continue;
}
if (rrt->rrt_t < t_lifetime)
rrt->rrt_info.rip6_metric = HOPCNT_INFINITY6;
rrt_prev = rrt;
}
/* Supply updates */
for (ifcp = ifc; ifcp; ifcp = ifcp->ifc_next) {
if (ifcp->ifc_index > 0 && (ifcp->ifc_flags & IFF_UP))
ripsend(ifcp, &ifcp->ifc_ripsin, 0);
}
alarm(ripinterval(SUPPLY_INTERVAL6));
}
void
init()
{
int i, int0, int255, error;
struct addrinfo hints, *res;
char port[10];
ifc = (struct ifc *)NULL;
nifc = 0;
nindex2ifc = 0; /*initial guess*/
index2ifc = NULL;
snprintf(port, sizeof(port), "%d", RIP6_PORT);
memset(&hints, 0, sizeof(hints));
hints.ai_family = PF_INET6;
hints.ai_socktype = SOCK_DGRAM;
hints.ai_flags = AI_PASSIVE;
error = getaddrinfo(NULL, port, &hints, &res);
if (error)
fatal("%s", gai_strerror(error));
if (res->ai_next)
fatal(":: resolved to multiple address");
int0 = 0; int255 = 255;
ripsock = socket(res->ai_family, res->ai_socktype, res->ai_protocol);
if (ripsock < 0)
fatal("rip socket");
if (bind(ripsock, res->ai_addr, res->ai_addrlen) < 0)
fatal("rip bind");
if (setsockopt(ripsock, IPPROTO_IPV6, IPV6_MULTICAST_HOPS,
&int255, sizeof(int255)) < 0)
fatal("rip IPV6_MULTICAST_HOPS");
if (setsockopt(ripsock, IPPROTO_IPV6, IPV6_MULTICAST_LOOP,
&int0, sizeof(int0)) < 0)
fatal("rip IPV6_MULTICAST_LOOP");
i = 1;
#ifdef IPV6_RECVPKTINFO
if (setsockopt(ripsock, IPPROTO_IPV6, IPV6_RECVPKTINFO, &i,
sizeof(i)) < 0)
fatal("rip IPV6_RECVPKTINFO");
#else /* old adv. API */
if (setsockopt(ripsock, IPPROTO_IPV6, IPV6_PKTINFO, &i,
sizeof(i)) < 0)
fatal("rip IPV6_PKTINFO");
#endif
memset(&hints, 0, sizeof(hints));
hints.ai_family = PF_INET6;
hints.ai_socktype = SOCK_DGRAM;
error = getaddrinfo(RIP6_DEST, port, &hints, &res);
if (error)
fatal("%s", gai_strerror(error));
if (res->ai_next)
fatal("%s resolved to multiple address", RIP6_DEST);
memcpy(&ripsin, res->ai_addr, res->ai_addrlen);
#ifdef FD_ZERO
FD_ZERO(&sockvec);
#else
memset(&sockvec, 0, sizeof(sockvec));
#endif
FD_SET(ripsock, &sockvec);
if (nflag == 0) {
if ((rtsock = socket(PF_ROUTE, SOCK_RAW, 0)) < 0)
fatal("route socket");
FD_SET(rtsock, &sockvec);
} else
rtsock = -1; /*just for safety */
}
#define RIPSIZE(n) \
(sizeof(struct rip6) + ((n)-1) * sizeof(struct netinfo6))
/*
* ripflush flushes the rip datagram stored in the rip buffer
*/
static int nrt;
static struct netinfo6 *np;
void
ripflush(ifcp, sin)
struct ifc *ifcp;
struct sockaddr_in6 *sin;
{
int i;
int error;
if (ifcp)
tracet(1, "Send(%s): info(%d) to %s.%d\n",
ifcp->ifc_name, nrt,
inet6_n2p(&sin->sin6_addr), ntohs(sin->sin6_port));
else
tracet(1, "Send: info(%d) to %s.%d\n",
nrt, inet6_n2p(&sin->sin6_addr), ntohs(sin->sin6_port));
if (dflag >= 2) {
np = ripbuf->rip6_nets;
for (i = 0; i < nrt; i++, np++) {
if (np->rip6_metric == NEXTHOP_METRIC) {
if (IN6_IS_ADDR_UNSPECIFIED(&np->rip6_dest))
trace(2, " NextHop reset");
else {
trace(2, " NextHop %s",
inet6_n2p(&np->rip6_dest));
}
} else {
trace(2, " %s/%d[%d]",
inet6_n2p(&np->rip6_dest),
np->rip6_plen, np->rip6_metric);
}
if (np->rip6_tag) {
trace(2, " tag=0x%04x",
ntohs(np->rip6_tag) & 0xffff);
}
trace(2, "\n");
}
}
error = sendpacket(sin, RIPSIZE(nrt));
if (error == EAFNOSUPPORT) {
/* Protocol not supported */
tracet(1, "Could not send info to %s (%s): "
"set IFF_UP to 0\n",
ifcp->ifc_name, inet6_n2p(&ifcp->ifc_ripsin.sin6_addr));
ifcp->ifc_flags &= ~IFF_UP; /* As if down for AF_INET6 */
}
nrt = 0; np = ripbuf->rip6_nets;
}
/*
* Generate RIP6_RESPONSE packets and send them.
*/
void
ripsend(ifcp, sin, flag)
struct ifc *ifcp;
struct sockaddr_in6 *sin;
int flag;
{
struct riprt *rrt;
struct in6_addr *nh; /* next hop */
struct in6_addr ia;
struct iff *iffp;
int maxrte, ok;
if (ifcp == NULL) {
/*
* Request from non-link local address is not
* a regular route6d update.
*/
maxrte = (IFMINMTU - sizeof(struct ip6_hdr) -
sizeof(struct udphdr) -
sizeof(struct rip6) + sizeof(struct netinfo6)) /
sizeof(struct netinfo6);
nrt = 0; np = ripbuf->rip6_nets; nh = NULL;
for (rrt = riprt; rrt; rrt = rrt->rrt_next) {
if (rrt->rrt_rflags & RRTF_NOADVERTISE)
continue;
/* Put the route to the buffer */
*np = rrt->rrt_info;
np++; nrt++;
if (nrt == maxrte) {
ripflush(NULL, sin);
nh = NULL;
}
}
if (nrt) /* Send last packet */
ripflush(NULL, sin);
return;
}
if ((flag & RRTF_SENDANYWAY) == 0 &&
(qflag || (ifcp->ifc_flags & IFF_LOOPBACK)))
return;
if (iff_find(ifcp, 'N') != NULL)
return;
if (iff_find(ifcp, 'T') != NULL) {
struct netinfo6 rrt_info;
memset(&rrt_info, 0, sizeof(struct netinfo6));
rrt_info.rip6_dest = in6addr_any;
rrt_info.rip6_plen = 0;
rrt_info.rip6_metric = 1;
rrt_info.rip6_tag = htons(routetag & 0xffff);
np = ripbuf->rip6_nets;
*np = rrt_info;
nrt = 1;
ripflush(ifcp, sin);
return;
}
maxrte = (ifcp->ifc_mtu - sizeof(struct ip6_hdr) -
sizeof(struct udphdr) -
sizeof(struct rip6) + sizeof(struct netinfo6)) /
sizeof(struct netinfo6);
nrt = 0; np = ripbuf->rip6_nets; nh = NULL;
for (rrt = riprt; rrt; rrt = rrt->rrt_next) {
if (rrt->rrt_rflags & RRTF_NOADVERTISE)
continue;
/* Need to check filer here */
ok = 1;
for (iffp = ifcp->ifc_filter; iffp; iffp = iffp->iff_next) {
if (iffp->iff_type != 'A')
continue;
if (rrt->rrt_info.rip6_plen <= iffp->iff_plen)
continue;
ia = rrt->rrt_info.rip6_dest;
applyplen(&ia, iffp->iff_plen);
if (IN6_ARE_ADDR_EQUAL(&ia, &iffp->iff_addr)) {
ok = 0;
break;
}
}
if (!ok)
continue;
for (iffp = ifcp->ifc_filter; iffp; iffp = iffp->iff_next) {
if (iffp->iff_type != 'O')
continue;
ok = 0;
if (rrt->rrt_info.rip6_plen < iffp->iff_plen)
continue;
ia = rrt->rrt_info.rip6_dest;
applyplen(&ia, iffp->iff_plen);
if (IN6_ARE_ADDR_EQUAL(&ia, &iffp->iff_addr)) {
ok = 1;
break;
}
}
if (!ok)
continue;
/* Check split horizon and other conditions */
if (tobeadv(rrt, ifcp) == 0)
continue;
/* Only considers the routes with flag if specified */
if ((flag & RRTF_CHANGED) &&
(rrt->rrt_rflags & RRTF_CHANGED) == 0)
continue;
/* Check nexthop */
if (rrt->rrt_index == ifcp->ifc_index &&
!IN6_IS_ADDR_UNSPECIFIED(&rrt->rrt_gw) &&
(rrt->rrt_rflags & RRTF_NH_NOT_LLADDR) == 0) {
if (nh == NULL || !IN6_ARE_ADDR_EQUAL(nh, &rrt->rrt_gw)) {
if (nrt == maxrte - 2)
ripflush(ifcp, sin);
np->rip6_dest = rrt->rrt_gw;
if (IN6_IS_ADDR_LINKLOCAL(&np->rip6_dest))
SET_IN6_LINKLOCAL_IFINDEX(np->rip6_dest, 0);
np->rip6_plen = 0;
np->rip6_tag = 0;
np->rip6_metric = NEXTHOP_METRIC;
nh = &rrt->rrt_gw;
np++; nrt++;
}
} else if (nh && (rrt->rrt_index != ifcp->ifc_index ||
!IN6_ARE_ADDR_EQUAL(nh, &rrt->rrt_gw) ||
rrt->rrt_rflags & RRTF_NH_NOT_LLADDR)) {
/* Reset nexthop */
if (nrt == maxrte - 2)
ripflush(ifcp, sin);
memset(np, 0, sizeof(struct netinfo6));
np->rip6_metric = NEXTHOP_METRIC;
nh = NULL;
np++; nrt++;
}
/* Put the route to the buffer */
*np = rrt->rrt_info;
np++; nrt++;
if (nrt == maxrte) {
ripflush(ifcp, sin);
nh = NULL;
}
}
if (nrt) /* Send last packet */
ripflush(ifcp, sin);
}
/*
* Determine if the route is to be advertised on the specified interface.
* It checks options specified in the arguments and the split horizon rule.
*/
int
tobeadv(rrt, ifcp)
struct riprt *rrt;
struct ifc *ifcp;
{
/* Special care for static routes */
if (rrt->rrt_flags & RTF_STATIC) {
/* XXX don't advertise reject/blackhole routes */
if (rrt->rrt_flags & (RTF_REJECT | RTF_BLACKHOLE))
return 0;
if (Sflag) /* Yes, advertise it anyway */
return 1;
if (sflag && rrt->rrt_index != ifcp->ifc_index)
return 1;
return 0;
}
/* Regular split horizon */
if (hflag == 0 && rrt->rrt_index == ifcp->ifc_index)
return 0;
return 1;
}
/*
* Send a rip packet actually.
*/
int
sendpacket(sin, len)
struct sockaddr_in6 *sin;
int len;
{
/*
* MSG_DONTROUTE should not be specified when it responds with a
* RIP6_REQUEST message. SO_DONTROUTE has been specified to
* other sockets.
*/
struct msghdr m;
struct cmsghdr *cm;
struct iovec iov[2];
u_char cmsgbuf[256];
struct in6_pktinfo *pi;
int index;
struct sockaddr_in6 sincopy;
/* do not overwrite the given sin */
sincopy = *sin;
sin = &sincopy;
if (IN6_IS_ADDR_LINKLOCAL(&sin->sin6_addr)
|| IN6_IS_ADDR_MULTICAST(&sin->sin6_addr)) {
index = IN6_LINKLOCAL_IFINDEX(sin->sin6_addr);
SET_IN6_LINKLOCAL_IFINDEX(sin->sin6_addr, 0);
} else
index = 0;
m.msg_name = (caddr_t)sin;
m.msg_namelen = sizeof(*sin);
iov[0].iov_base = (caddr_t)ripbuf;
iov[0].iov_len = len;
m.msg_iov = iov;
m.msg_iovlen = 1;
if (!index) {
m.msg_control = NULL;
m.msg_controllen = 0;
} else {
memset(cmsgbuf, 0, sizeof(cmsgbuf));
cm = (struct cmsghdr *)cmsgbuf;
m.msg_control = (caddr_t)cm;
m.msg_controllen = CMSG_SPACE(sizeof(struct in6_pktinfo));
cm->cmsg_len = CMSG_LEN(sizeof(struct in6_pktinfo));
cm->cmsg_level = IPPROTO_IPV6;
cm->cmsg_type = IPV6_PKTINFO;
pi = (struct in6_pktinfo *)CMSG_DATA(cm);
memset(&pi->ipi6_addr, 0, sizeof(pi->ipi6_addr)); /*::*/
pi->ipi6_ifindex = index;
}
if (sendmsg(ripsock, &m, 0 /*MSG_DONTROUTE*/) < 0) {
trace(1, "sendmsg: %s\n", strerror(errno));
return errno;
}
return 0;
}
/*
* Receive and process RIP packets. Update the routes/kernel forwarding
* table if necessary.
*/
void
riprecv()
{
struct ifc *ifcp, *ic;
struct sockaddr_in6 fsock;
struct in6_addr nh; /* next hop */
struct rip6 *rp;
struct netinfo6 *np, *nq;
struct riprt *rrt;
int len, nn, need_trigger, index;
char buf[4 * RIP6_MAXMTU];
time_t t;
struct msghdr m;
struct cmsghdr *cm;
struct iovec iov[2];
u_char cmsgbuf[256];
struct in6_pktinfo *pi;
struct iff *iffp;
struct in6_addr ia;
int ok;
need_trigger = 0;
m.msg_name = (caddr_t)&fsock;
m.msg_namelen = sizeof(fsock);
iov[0].iov_base = (caddr_t)buf;
iov[0].iov_len = sizeof(buf);
m.msg_iov = iov;
m.msg_iovlen = 1;
cm = (struct cmsghdr *)cmsgbuf;
m.msg_control = (caddr_t)cm;
m.msg_controllen = sizeof(cmsgbuf);
if ((len = recvmsg(ripsock, &m, 0)) < 0)
fatal("recvmsg");
index = 0;
for (cm = (struct cmsghdr *)CMSG_FIRSTHDR(&m);
cm;
cm = (struct cmsghdr *)CMSG_NXTHDR(&m, cm)) {
if (cm->cmsg_level == IPPROTO_IPV6
&& cm->cmsg_type == IPV6_PKTINFO) {
pi = (struct in6_pktinfo *)(CMSG_DATA(cm));
index = pi->ipi6_ifindex;
break;
}
}
if (index && IN6_IS_ADDR_LINKLOCAL(&fsock.sin6_addr))
SET_IN6_LINKLOCAL_IFINDEX(fsock.sin6_addr, index);
nh = fsock.sin6_addr;
nn = (len - sizeof(struct rip6) + sizeof(struct netinfo6)) /
sizeof(struct netinfo6);
rp = (struct rip6 *)buf;
np = rp->rip6_nets;
if (rp->rip6_vers != RIP6_VERSION) {
trace(1, "Incorrect RIP version %d\n", rp->rip6_vers);
return;
}
if (rp->rip6_cmd == RIP6_REQUEST) {
if (index && index < nindex2ifc) {
ifcp = index2ifc[index];
riprequest(ifcp, np, nn, &fsock);
} else {
riprequest(NULL, np, nn, &fsock);
}
return;
}
if (!IN6_IS_ADDR_LINKLOCAL(&fsock.sin6_addr)) {
trace(1, "Packets from non-ll addr: %s\n",
inet6_n2p(&fsock.sin6_addr));
return; /* Ignore packets from non-link-local addr */
}
index = IN6_LINKLOCAL_IFINDEX(fsock.sin6_addr);
ifcp = (index < nindex2ifc) ? index2ifc[index] : NULL;
if (!ifcp) {
trace(1, "Packets to unknown interface index %d\n", index);
return; /* Ignore it */
}
if (IN6_ARE_ADDR_EQUAL(&ifcp->ifc_mylladdr, &fsock.sin6_addr))
return; /* The packet is from me; ignore */
if (rp->rip6_cmd != RIP6_RESPONSE) {
trace(1, "Invalid command %d\n", rp->rip6_cmd);
return;
}
if (iff_find(ifcp, 'N') != NULL)
return;
tracet(1, "Recv(%s): from %s.%d info(%d)\n",
ifcp->ifc_name, inet6_n2p(&nh), ntohs(fsock.sin6_port), nn);
t = time(NULL);
for (; nn; nn--, np++) {
if (np->rip6_metric == NEXTHOP_METRIC) {
/* modify neighbor address */
if (IN6_IS_ADDR_LINKLOCAL(&np->rip6_dest)) {
nh = np->rip6_dest;
SET_IN6_LINKLOCAL_IFINDEX(nh, index);
trace(1, "\tNexthop: %s\n", inet6_n2p(&nh));
} else if (IN6_IS_ADDR_UNSPECIFIED(&np->rip6_dest)) {
nh = fsock.sin6_addr;
trace(1, "\tNexthop: %s\n", inet6_n2p(&nh));
} else {
nh = fsock.sin6_addr;
trace(1, "\tInvalid Nexthop: %s\n",
inet6_n2p(&np->rip6_dest));
}
continue;
}
if (IN6_IS_ADDR_MULTICAST(&np->rip6_dest)) {
trace(1, "\tMulticast netinfo6: %s/%d [%d]\n",
inet6_n2p(&np->rip6_dest),
np->rip6_plen, np->rip6_metric);
continue;
}
if (IN6_IS_ADDR_LOOPBACK(&np->rip6_dest)) {
trace(1, "\tLoopback netinfo6: %s/%d [%d]\n",
inet6_n2p(&np->rip6_dest),
np->rip6_plen, np->rip6_metric);
continue;
}
if (IN6_IS_ADDR_LINKLOCAL(&np->rip6_dest)) {
trace(1, "\tLink Local netinfo6: %s/%d [%d]\n",
inet6_n2p(&np->rip6_dest),
np->rip6_plen, np->rip6_metric);
continue;
}
/* may need to pass sitelocal prefix in some case, however*/
if (IN6_IS_ADDR_SITELOCAL(&np->rip6_dest) && !lflag) {
trace(1, "\tSite Local netinfo6: %s/%d [%d]\n",
inet6_n2p(&np->rip6_dest),
np->rip6_plen, np->rip6_metric);
continue;
}
trace(2, "\tnetinfo6: %s/%d [%d]",
inet6_n2p(&np->rip6_dest),
np->rip6_plen, np->rip6_metric);
if (np->rip6_tag)
trace(2, " tag=0x%04x", ntohs(np->rip6_tag) & 0xffff);
if (dflag >= 2) {
ia = np->rip6_dest;
applyplen(&ia, np->rip6_plen);
if (!IN6_ARE_ADDR_EQUAL(&ia, &np->rip6_dest))
trace(2, " [junk outside prefix]");
}
/* Listen-only filter */
ok = 1; /* if there's no L filter, it is ok */
for (iffp = ifcp->ifc_filter; iffp; iffp = iffp->iff_next) {
if (iffp->iff_type != 'L')
continue;
ok = 0;
if (np->rip6_plen < iffp->iff_plen)
continue;
/* special rule: ::/0 means default, not "in /0" */
if (iffp->iff_plen == 0 && np->rip6_plen > 0)
continue;
ia = np->rip6_dest;
applyplen(&ia, iffp->iff_plen);
if (IN6_ARE_ADDR_EQUAL(&ia, &iffp->iff_addr)) {
ok = 1;
break;
}
}
if (!ok) {
trace(2, " (filtered)\n");
continue;
}
trace(2, "\n");
np->rip6_metric++;
np->rip6_metric += ifcp->ifc_metric;
if (np->rip6_metric > HOPCNT_INFINITY6)
np->rip6_metric = HOPCNT_INFINITY6;
applyplen(&np->rip6_dest, np->rip6_plen);
if ((rrt = rtsearch(np)) != NULL) {
if (rrt->rrt_t == 0)
continue; /* Intf route has priority */
nq = &rrt->rrt_info;
if (nq->rip6_metric > np->rip6_metric) {
if (rrt->rrt_index == ifcp->ifc_index &&
IN6_ARE_ADDR_EQUAL(&nh, &rrt->rrt_gw)) {
/* Small metric from the same gateway */
nq->rip6_metric = np->rip6_metric;
} else {
/* Better route found */
rrt->rrt_index = ifcp->ifc_index;
/* Update routing table */
delroute(nq, &rrt->rrt_gw);
rrt->rrt_gw = nh;
*nq = *np;
addroute(rrt, &nh, ifcp);
}
rrt->rrt_rflags |= RRTF_CHANGED;
rrt->rrt_t = t;
need_trigger = 1;
} else if (nq->rip6_metric < np->rip6_metric &&
rrt->rrt_index == ifcp->ifc_index &&
IN6_ARE_ADDR_EQUAL(&nh, &rrt->rrt_gw)) {
/* Got worse route from same gw */
nq->rip6_metric = np->rip6_metric;
rrt->rrt_t = t;
rrt->rrt_rflags |= RRTF_CHANGED;
need_trigger = 1;
} else if (nq->rip6_metric == np->rip6_metric &&
rrt->rrt_index == ifcp->ifc_index &&
IN6_ARE_ADDR_EQUAL(&nh, &rrt->rrt_gw) &&
np->rip6_metric < HOPCNT_INFINITY6) {
/* same metric, same route from same gw */
rrt->rrt_t = t;
}
/*
* if nq->rip6_metric == HOPCNT_INFINITY6 then
* do not update age value. Do nothing.
*/
} else if (np->rip6_metric < HOPCNT_INFINITY6) {
/* Got a new valid route */
if ((rrt = MALLOC(struct riprt)) == NULL)
fatal("malloc: struct riprt");
memset(rrt, 0, sizeof(*rrt));
nq = &rrt->rrt_info;
rrt->rrt_same = NULL;
rrt->rrt_index = ifcp->ifc_index;
rrt->rrt_flags = RTF_UP|RTF_GATEWAY;
rrt->rrt_gw = nh;
*nq = *np;
applyplen(&nq->rip6_dest, nq->rip6_plen);
if (nq->rip6_plen == sizeof(struct in6_addr) * 8)
rrt->rrt_flags |= RTF_HOST;
/* Put the route to the list */
rrt->rrt_next = riprt;
riprt = rrt;
/* Update routing table */
addroute(rrt, &nh, ifcp);
rrt->rrt_rflags |= RRTF_CHANGED;
need_trigger = 1;
rrt->rrt_t = t;
}
}
/* XXX need to care the interval between triggered updates */
if (need_trigger) {
if (nextalarm > time(NULL) + RIP_TRIG_INT6_MAX) {
for (ic = ifc; ic; ic = ic->ifc_next) {
if (ifcp->ifc_index == ic->ifc_index)
continue;
if (ic->ifc_flags & IFF_UP)
ripsend(ic, &ic->ifc_ripsin,
RRTF_CHANGED);
}
}
/* Reset the flag */
for (rrt = riprt; rrt; rrt = rrt->rrt_next)
rrt->rrt_rflags &= ~RRTF_CHANGED;
}
}
/*
* Send all routes request packet to the specified interface.
*/
void
sendrequest(ifcp)
struct ifc *ifcp;
{
struct netinfo6 *np;
int error;
if (ifcp->ifc_flags & IFF_LOOPBACK)
return;
ripbuf->rip6_cmd = RIP6_REQUEST;
np = ripbuf->rip6_nets;
memset(np, 0, sizeof(struct netinfo6));
np->rip6_metric = HOPCNT_INFINITY6;
tracet(1, "Send rtdump Request to %s (%s)\n",
ifcp->ifc_name, inet6_n2p(&ifcp->ifc_ripsin.sin6_addr));
error = sendpacket(&ifcp->ifc_ripsin, RIPSIZE(1));
if (error == EAFNOSUPPORT) {
/* Protocol not supported */
tracet(1, "Could not send rtdump Request to %s (%s): "
"set IFF_UP to 0\n",
ifcp->ifc_name, inet6_n2p(&ifcp->ifc_ripsin.sin6_addr));
ifcp->ifc_flags &= ~IFF_UP; /* As if down for AF_INET6 */
}
ripbuf->rip6_cmd = RIP6_RESPONSE;
}
/*
* Process a RIP6_REQUEST packet.
*/
void
riprequest(ifcp, np, nn, sin)
struct ifc *ifcp;
struct netinfo6 *np;
int nn;
struct sockaddr_in6 *sin;
{
int i;
struct riprt *rrt;
if (!(nn == 1 && IN6_IS_ADDR_UNSPECIFIED(&np->rip6_dest) &&
np->rip6_plen == 0 && np->rip6_metric == HOPCNT_INFINITY6)) {
/* Specific response, don't split-horizon */
trace(1, "\tRIP Request\n");
for (i = 0; i < nn; i++, np++) {
rrt = rtsearch(np);
if (rrt)
np->rip6_metric = rrt->rrt_info.rip6_metric;
else
np->rip6_metric = HOPCNT_INFINITY6;
}
(void)sendpacket(sin, RIPSIZE(nn));
return;
}
/* Whole routing table dump */
trace(1, "\tRIP Request -- whole routing table\n");
ripsend(ifcp, sin, RRTF_SENDANYWAY);
}
/*
* Get information of each interface.
*/
void
ifconfig()
{
#ifdef HAVE_GETIFADDRS
struct ifaddrs *ifap, *ifa;
struct ifc *ifcp;
struct ipv6_mreq mreq;
int s;
if ((s = socket(AF_INET6, SOCK_DGRAM, 0)) < 0)
fatal("socket");
if (getifaddrs(&ifap) != 0)
fatal("getifaddrs");
for (ifa = ifap; ifa; ifa = ifa->ifa_next) {
if (ifa->ifa_addr->sa_family != AF_INET6)
continue;
ifcp = ifc_find(ifa->ifa_name);
/* we are interested in multicast-capable interfaces */
if ((ifa->ifa_flags & IFF_MULTICAST) == 0)
continue;
if (!ifcp) {
/* new interface */
if ((ifcp = MALLOC(struct ifc)) == NULL)
fatal("malloc: struct ifc");
memset(ifcp, 0, sizeof(*ifcp));
ifcp->ifc_index = -1;
ifcp->ifc_next = ifc;
ifc = ifcp;
nifc++;
ifcp->ifc_name = allocopy(ifa->ifa_name);
ifcp->ifc_addr = 0;
ifcp->ifc_filter = 0;
ifcp->ifc_flags = ifa->ifa_flags;
trace(1, "newif %s <%s>\n", ifcp->ifc_name,
ifflags(ifcp->ifc_flags));
if (!strcmp(ifcp->ifc_name, LOOPBACK_IF))
loopifcp = ifcp;
} else {
/* update flag, this may be up again */
if (ifcp->ifc_flags != ifa->ifa_flags) {
trace(1, "%s: <%s> -> ", ifcp->ifc_name,
ifflags(ifcp->ifc_flags));
trace(1, "<%s>\n", ifflags(ifa->ifa_flags));
}
ifcp->ifc_flags = ifa->ifa_flags;
}
ifconfig1(ifa->ifa_name, ifa->ifa_addr, ifcp, s);
if ((ifcp->ifc_flags & (IFF_LOOPBACK | IFF_UP)) == IFF_UP
&& 0 < ifcp->ifc_index && !ifcp->ifc_joined) {
mreq.ipv6mr_multiaddr = ifcp->ifc_ripsin.sin6_addr;
mreq.ipv6mr_interface = ifcp->ifc_index;
if (setsockopt(ripsock, IPPROTO_IPV6,
IPV6_JOIN_GROUP, &mreq, sizeof(mreq)) < 0)
fatal("IPV6_JOIN_GROUP");
trace(1, "join %s %s\n", ifcp->ifc_name, RIP6_DEST);
ifcp->ifc_joined++;
}
}
close(s);
freeifaddrs(ifap);
#else
int s, i;
char *buf;
struct ifconf ifconf;
struct ifreq *ifrp, ifr;
struct ifc *ifcp;
struct ipv6_mreq mreq;
int bufsiz;
if ((s = socket(AF_INET6, SOCK_DGRAM, 0)) < 0)
fatal("socket");
/* wild guess - v4, media, link, v6 * 3 */
bufsiz = if_maxindex() * sizeof(struct ifreq) * 6;
if ((buf = (char *)malloc(bufsiz)) == NULL)
fatal("malloc");
/*
* ioctl(SIOCGIFCONF) does not return error on buffer size.
* we'll try to guess the buffer size by trying it twice, with
* different buffer size.
*/
ifconf.ifc_buf = buf;
ifconf.ifc_len = bufsiz / 2;
if (ioctl(s, SIOCGIFCONF, (char *)&ifconf) < 0)
fatal("ioctl: SIOCGIFCONF");
i = ifconf.ifc_len;
while (1) {
char *newbuf;
ifconf.ifc_buf = buf;
ifconf.ifc_len = bufsiz;
if (ioctl(s, SIOCGIFCONF, (char *)&ifconf) < 0)
fatal("ioctl: SIOCGIFCONF");
if (i == ifconf.ifc_len)
break;
i = ifconf.ifc_len;
bufsiz *= 2;
if ((newbuf = (char *)realloc(buf, bufsiz)) == NULL) {
free(buf);
fatal("realloc");
}
buf = newbuf;
}
for (i = 0; i < ifconf.ifc_len; ) {
ifrp = (struct ifreq *)(buf + i);
if (ifrp->ifr_addr.sa_family != AF_INET6)
goto skip;
ifcp = ifc_find(ifrp->ifr_name);
strcpy(ifr.ifr_name, ifrp->ifr_name);
if (ioctl(s, SIOCGIFFLAGS, (char *)&ifr) < 0)
fatal("ioctl: SIOCGIFFLAGS");
/* we are interested in multicast-capable interfaces */
if ((ifr.ifr_flags & IFF_MULTICAST) == 0)
goto skip;
if (!ifcp) {
/* new interface */
if ((ifcp = MALLOC(struct ifc)) == NULL)
fatal("malloc: struct ifc");
memset(ifcp, 0, sizeof(*ifcp));
ifcp->ifc_index = -1;
ifcp->ifc_next = ifc;
ifc = ifcp;
nifc++;
ifcp->ifc_name = allocopy(ifrp->ifr_name);
ifcp->ifc_addr = 0;
ifcp->ifc_filter = 0;
ifcp->ifc_flags = ifr.ifr_flags;
trace(1, "newif %s <%s>\n", ifcp->ifc_name,
ifflags(ifcp->ifc_flags));
if (!strcmp(ifcp->ifc_name, LOOPBACK_IF))
loopifcp = ifcp;
} else {
/* update flag, this may be up again */
if (ifcp->ifc_flags != ifr.ifr_flags) {
trace(1, "%s: <%s> -> ", ifcp->ifc_name,
ifflags(ifcp->ifc_flags));
trace(1, "<%s>\n", ifflags(ifr.ifr_flags));
}
ifcp->ifc_flags = ifr.ifr_flags;
}
ifconfig1(ifrp->ifr_name, &ifrp->ifr_addr, ifcp, s);
if ((ifcp->ifc_flags & (IFF_LOOPBACK | IFF_UP)) == IFF_UP
&& 0 < ifcp->ifc_index && !ifcp->ifc_joined) {
mreq.ipv6mr_multiaddr = ifcp->ifc_ripsin.sin6_addr;
mreq.ipv6mr_interface = ifcp->ifc_index;
if (setsockopt(ripsock, IPPROTO_IPV6,
IPV6_JOIN_GROUP, &mreq, sizeof(mreq)) < 0)
fatal("IPV6_JOIN_GROUP");
trace(1, "join %s %s\n", ifcp->ifc_name, RIP6_DEST);
ifcp->ifc_joined++;
}
skip:
i += IFNAMSIZ;
if (ifrp->ifr_addr.sa_len > sizeof(struct sockaddr))
i += ifrp->ifr_addr.sa_len;
else
i += sizeof(struct sockaddr);
}
close(s);
free(buf);
#endif
}
void
ifconfig1(name, sa, ifcp, s)
const char *name;
const struct sockaddr *sa;
struct ifc *ifcp;
int s;
{
struct in6_ifreq ifr;
struct sockaddr_in6 *sin;
struct ifac *ifa;
int plen;
char buf[BUFSIZ];
sin = (struct sockaddr_in6 *)sa;
ifr.ifr_addr = *sin;
strcpy(ifr.ifr_name, name);
if (ioctl(s, SIOCGIFNETMASK_IN6, (char *)&ifr) < 0)
fatal("ioctl: SIOCGIFNETMASK_IN6");
plen = mask2len(&ifr.ifr_addr.sin6_addr, 16);
if ((ifa = ifa_match(ifcp, &sin->sin6_addr, plen)) != NULL) {
/* same interface found */
/* need check if something changed */
/* XXX not yet implemented */
return;
}
/*
* New address is found
*/
if ((ifa = MALLOC(struct ifac)) == NULL)
fatal("malloc: struct ifac");
memset(ifa, 0, sizeof(*ifa));
ifa->ifa_conf = ifcp;
ifa->ifa_next = ifcp->ifc_addr;
ifcp->ifc_addr = ifa;
ifa->ifa_addr = sin->sin6_addr;
ifa->ifa_plen = plen;
if (ifcp->ifc_flags & IFF_POINTOPOINT) {
ifr.ifr_addr = *sin;
if (ioctl(s, SIOCGIFDSTADDR_IN6, (char *)&ifr) < 0)
fatal("ioctl: SIOCGIFDSTADDR_IN6");
ifa->ifa_raddr = ifr.ifr_dstaddr.sin6_addr;
inet_ntop(AF_INET6, (void *)&ifa->ifa_raddr, buf, sizeof(buf));
trace(1, "found address %s/%d -- %s\n",
inet6_n2p(&ifa->ifa_addr), ifa->ifa_plen, buf);
} else {
trace(1, "found address %s/%d\n",
inet6_n2p(&ifa->ifa_addr), ifa->ifa_plen);
}
if (ifcp->ifc_index < 0 && IN6_IS_ADDR_LINKLOCAL(&ifa->ifa_addr)) {
ifcp->ifc_mylladdr = ifa->ifa_addr;
ifcp->ifc_index = IN6_LINKLOCAL_IFINDEX(ifa->ifa_addr);
memcpy(&ifcp->ifc_ripsin, &ripsin, ripsin.ss_len);
SET_IN6_LINKLOCAL_IFINDEX(ifcp->ifc_ripsin.sin6_addr,
ifcp->ifc_index);
setindex2ifc(ifcp->ifc_index, ifcp);
ifcp->ifc_mtu = getifmtu(ifcp->ifc_index);
if (ifcp->ifc_mtu > RIP6_MAXMTU)
ifcp->ifc_mtu = RIP6_MAXMTU;
if (ioctl(s, SIOCGIFMETRIC, (char *)&ifr) < 0)
fatal("ioctl: SIOCGIFMETRIC");
ifcp->ifc_metric = ifr.ifr_metric;
trace(1, "\tindex: %d, mtu: %d, metric: %d\n",
ifcp->ifc_index, ifcp->ifc_mtu, ifcp->ifc_metric);
}
}
/*
* Receive and process routing messages.
* Update interface information as necesssary.
*/
void
rtrecv()
{
char buf[BUFSIZ];
char *p, *q;
struct rt_msghdr *rtm;
struct ifa_msghdr *ifam;
struct if_msghdr *ifm;
int len;
struct ifc *ifcp;
int iface = 0, rtable = 0;
struct sockaddr_in6 *rta[RTAX_MAX];
int i, addrs;
if ((len = read(rtsock, buf, sizeof(buf))) < 0) {
perror("read from rtsock");
exit(-1);
}
if (len < sizeof(*rtm)) {
trace(1, "short read from rtsock: %d (should be > %lu)\n",
len, (u_long)sizeof(*rtm));
return;
}
for (p = buf; p - buf < len; p += ((struct rt_msghdr *)p)->rtm_msglen) {
/* safety against bogus message */
if (((struct rt_msghdr *)p)->rtm_msglen <= 0) {
trace(1, "bogus rtmsg: length=%d\n",
((struct rt_msghdr *)p)->rtm_msglen);
break;
}
rtm = NULL;
ifam = NULL;
ifm = NULL;
switch (((struct rt_msghdr *)p)->rtm_type) {
case RTM_NEWADDR:
case RTM_DELADDR:
ifam = (struct ifa_msghdr *)p;
addrs = ifam->ifam_addrs;
q = (char *)(ifam + 1);
break;
case RTM_IFINFO:
ifm = (struct if_msghdr *)p;
addrs = ifm->ifm_addrs;
q = (char *)(ifm + 1);
break;
default:
rtm = (struct rt_msghdr *)p;
addrs = rtm->rtm_addrs;
q = (char *)(rtm + 1);
if (rtm->rtm_version != RTM_VERSION) {
trace(1, "unexpected rtmsg version %d "
"(should be %d)\n",
rtm->rtm_version, RTM_VERSION);
continue;
}
if (rtm->rtm_pid == pid) {
#if 0
trace(1, "rtmsg looped back to me, ignored\n");
#endif
continue;
}
break;
}
memset(&rta, 0, sizeof(rta));
for (i = 0; i < RTAX_MAX; i++) {
if (addrs & (1 << i)) {
rta[i] = (struct sockaddr_in6 *)q;
q += ROUNDUP(rta[i]->sin6_len);
}
}
trace(1, "rtsock: %s (addrs=%x)\n",
rttypes((struct rt_msghdr *)p), addrs);
if (dflag >= 2) {
int i;
for (i = 0;
i < ((struct rt_msghdr *)p)->rtm_msglen;
i++) {
fprintf(stderr, "%02x ", p[i] & 0xff);
if (i % 16 == 15) fprintf(stderr, "\n");
}
fprintf(stderr, "\n");
}
/*
* Easy ones first.
*
* We may be able to optimize by using ifm->ifm_index or
* ifam->ifam_index. For simplicity we don't do that here.
*/
switch (((struct rt_msghdr *)p)->rtm_type) {
case RTM_NEWADDR:
case RTM_IFINFO:
iface++;
continue;
case RTM_ADD:
rtable++;
continue;
case RTM_LOSING:
case RTM_MISS:
case RTM_RESOLVE:
case RTM_GET:
case RTM_LOCK:
/* nothing to be done here */
trace(1, "\tnothing to be done, ignored\n");
continue;
}
#if 0
if (rta[RTAX_DST] == NULL) {
trace(1, "\tno destination, ignored\n");
continue;
}
if (rta[RTAX_DST]->sin6_family != AF_INET6) {
trace(1, "\taf mismatch, ignored\n");
continue;
}
if (IN6_IS_ADDR_LINKLOCAL(&rta[RTAX_DST]->sin6_addr)) {
trace(1, "\tlinklocal destination, ignored\n");
continue;
}
if (IN6_ARE_ADDR_EQUAL(&rta[RTAX_DST]->sin6_addr, &in6addr_loopback)) {
trace(1, "\tloopback destination, ignored\n");
continue; /* Loopback */
}
if (IN6_IS_ADDR_MULTICAST(&rta[RTAX_DST]->sin6_addr)) {
trace(1, "\tmulticast destination, ignored\n");
continue;
}
#endif
/* hard ones */
switch (((struct rt_msghdr *)p)->rtm_type) {
case RTM_NEWADDR:
case RTM_IFINFO:
case RTM_ADD:
case RTM_LOSING:
case RTM_MISS:
case RTM_RESOLVE:
case RTM_GET:
case RTM_LOCK:
/* should already be handled */
fatal("rtrecv: never reach here");
case RTM_DELETE:
if (!rta[RTAX_DST] || !rta[RTAX_GATEWAY]
|| !rta[RTAX_NETMASK]) {
trace(1, "\tsome of dst/gw/netamsk are unavailable, ignored\n");
break;
}
if (rt_del(rta[RTAX_DST], rta[RTAX_GATEWAY], rta[RTAX_NETMASK]) == 0) {
rtable++; /*just to be sure*/
}
break;
case RTM_CHANGE:
case RTM_REDIRECT:
trace(1, "\tnot supported yet, ignored\n");
break;
case RTM_DELADDR:
if (!rta[RTAX_NETMASK] || !rta[RTAX_IFA]) {
trace(1, "\tno netmask or ifa given, ignored\n");
break;
}
if (ifam->ifam_index < nindex2ifc)
ifcp = index2ifc[ifam->ifam_index];
else
ifcp = NULL;
if (!ifcp) {
trace(1, "\tinvalid ifam_index %d, ignored\n",
ifam->ifam_index);
break;
}
rt_deladdr(ifcp, rta[RTAX_IFA], rta[RTAX_NETMASK]);
iface++;
break;
case RTM_OLDADD:
case RTM_OLDDEL:
trace(1, "\tnot supported yet, ignored\n");
break;
}
}
if (iface) {
trace(1, "rtsock: reconfigure interfaces, refresh interface routes\n");
ifconfig();
for (ifcp = ifc; ifcp; ifcp = ifcp->ifc_next)
ifrt(ifcp, 1);
}
if (rtable) {
trace(1, "rtsock: read routing table again\n");
krtread(1);
}
}
/*
* remove specified route from the internal routing table.
*/
int
rt_del(sdst, sgw, smask)
const struct sockaddr_in6 *sdst;
const struct sockaddr_in6 *sgw;
const struct sockaddr_in6 *smask;
{
const struct in6_addr *dst = NULL;
const struct in6_addr *gw = NULL;
int prefix;
struct netinfo6 ni6;
struct riprt *rrt = NULL;
time_t t_lifetime;
if (sdst->sin6_family != AF_INET6) {
trace(1, "\tother AF, ignored\n");
return -1;
}
if (IN6_IS_ADDR_LINKLOCAL(&sdst->sin6_addr)
|| IN6_ARE_ADDR_EQUAL(&sdst->sin6_addr, &in6addr_loopback)
|| IN6_IS_ADDR_MULTICAST(&sdst->sin6_addr)) {
trace(1, "\taddress %s not interesting, ignored\n",
inet6_n2p(&sdst->sin6_addr));
return -1;
}
dst = &sdst->sin6_addr;
if (sgw->sin6_family == AF_INET6
&& smask->sin6_family == AF_INET6) {
/* easy case */
gw = &sgw->sin6_addr;
prefix = mask2len(&smask->sin6_addr, 16);
} else if (sgw->sin6_family == AF_LINK) {
/*
* Interface route... a hard case. We need to get the prefix
* length from the kernel, but we now are parsing rtmsg.
* We'll purge matching routes from my list, then get the
* fresh list.
*/
struct riprt *longest;
trace(1, "\t%s is a interface route, guessing prefixlen\n",
inet6_n2p(dst));
longest = NULL;
for (rrt = riprt; rrt; rrt = rrt->rrt_next) {
if (IN6_ARE_ADDR_EQUAL(&rrt->rrt_info.rip6_dest,
&sdst->sin6_addr)
&& IN6_IS_ADDR_LOOPBACK(&rrt->rrt_gw)) {
if (!longest
|| longest->rrt_info.rip6_plen <
rrt->rrt_info.rip6_plen) {
longest = rrt;
}
}
}
rrt = longest;
if (!rrt) {
trace(1, "\tno matching interface route found\n");
return -1;
}
gw = &in6addr_loopback;
prefix = rrt->rrt_info.rip6_plen;
} else {
trace(1, "\tunsupported af: (gw=%d, mask=%d)\n",
sgw->sin6_family, smask->sin6_family);
return -1;
}
trace(1, "\tdeleting %s/%d ", inet6_n2p(dst), prefix);
trace(1, "gw %s\n", inet6_n2p(gw));
t_lifetime = time(NULL) - RIP_LIFETIME;
/* age route for interface address */
memset(&ni6, 0, sizeof(ni6));
ni6.rip6_dest = *dst;
ni6.rip6_plen = prefix;
applyplen(&ni6.rip6_dest, ni6.rip6_plen); /*to be sure*/
trace(1, "\tfind route %s/%d\n", inet6_n2p(&ni6.rip6_dest),
ni6.rip6_plen);
if (!rrt && (rrt = rtsearch(&ni6)) == NULL) {
trace(1, "\tno route found\n");
return -1;
}
if ((rrt->rrt_flags & RTF_STATIC) == 0) {
trace(1, "\tyou can delete static routes only\n");
} else if (memcmp(&rrt->rrt_gw, gw, sizeof(rrt->rrt_gw)) != 0) {
trace(1, "\tgw mismatch: %s <-> ",
inet6_n2p(&rrt->rrt_gw));
trace(1, "%s\n", inet6_n2p(gw));
} else {
trace(1, "\troute found, age it\n");
if (rrt->rrt_t == 0 || rrt->rrt_t > t_lifetime) {
rrt->rrt_t = t_lifetime;
rrt->rrt_info.rip6_metric = HOPCNT_INFINITY6;
}
}
return 0;
}
/*
* remove specified address from internal interface/routing table.
*/
int
rt_deladdr(ifcp, sifa, smask)
struct ifc *ifcp;
const struct sockaddr_in6 *sifa;
const struct sockaddr_in6 *smask;
{
const struct in6_addr *addr = NULL;
int prefix;
struct ifac *ifa = NULL;
struct netinfo6 ni6;
struct riprt *rrt = NULL;
time_t t_lifetime;
int updated = 0;
if (sifa->sin6_family != AF_INET6 || smask->sin6_family != AF_INET6) {
trace(1, "\tother AF, ignored\n");
return -1;
}
addr = &sifa->sin6_addr;
prefix = mask2len(&smask->sin6_addr, 16);
trace(1, "\tdeleting %s/%d from %s\n",
inet6_n2p(addr), prefix, ifcp->ifc_name);
ifa = ifa_match(ifcp, addr, prefix);
if (!ifa) {
trace(1, "\tno matching ifa found for %s/%d on %s\n",
inet6_n2p(addr), prefix, ifcp->ifc_name);
return -1;
}
if (ifa->ifa_conf != ifcp) {
trace(1, "\taddress table corrupt: back pointer does not match "
"(%s != %s)\n",
ifcp->ifc_name, ifa->ifa_conf->ifc_name);
return -1;
}
/* remove ifa from interface */
if (ifcp->ifc_addr == ifa)
ifcp->ifc_addr = ifa->ifa_next;
else {
struct ifac *p;
for (p = ifcp->ifc_addr; p; p = p->ifa_next) {
if (p->ifa_next == ifa) {
p->ifa_next = ifa->ifa_next;
break;
}
}
}
ifa->ifa_next = NULL;
ifa->ifa_conf = NULL;
t_lifetime = time(NULL) - RIP_LIFETIME;
/* age route for interface address */
memset(&ni6, 0, sizeof(ni6));
ni6.rip6_dest = ifa->ifa_addr;
ni6.rip6_plen = ifa->ifa_plen;
applyplen(&ni6.rip6_dest, ni6.rip6_plen);
trace(1, "\tfind interface route %s/%d on %d\n",
inet6_n2p(&ni6.rip6_dest), ni6.rip6_plen, ifcp->ifc_index);
if ((rrt = rtsearch(&ni6)) != NULL) {
struct in6_addr none;
memset(&none, 0, sizeof(none));
if (rrt->rrt_index == ifcp->ifc_index
&& memcmp(&rrt->rrt_gw, &none, sizeof(rrt->rrt_gw)) == 0) {
trace(1, "\troute found, age it\n");
if (rrt->rrt_t == 0 || rrt->rrt_t > t_lifetime) {
rrt->rrt_t = t_lifetime;
rrt->rrt_info.rip6_metric = HOPCNT_INFINITY6;
}
updated++;
} else {
trace(1, "\tnon-interface route found: %s/%d on %d\n",
inet6_n2p(&rrt->rrt_info.rip6_dest),
rrt->rrt_info.rip6_plen,
rrt->rrt_index);
}
} else
trace(1, "\tno interface route found\n");
/* age route for p2p destination */
if (ifcp->ifc_flags & IFF_POINTOPOINT) {
memset(&ni6, 0, sizeof(ni6));
ni6.rip6_dest = ifa->ifa_raddr;
ni6.rip6_plen = 128;
applyplen(&ni6.rip6_dest, ni6.rip6_plen); /*to be sure*/
trace(1, "\tfind p2p route %s/%d on %d\n",
inet6_n2p(&ni6.rip6_dest), ni6.rip6_plen,
ifcp->ifc_index);
if ((rrt = rtsearch(&ni6)) != NULL) {
if (rrt->rrt_index == ifcp->ifc_index
&& memcmp(&rrt->rrt_gw, &ifa->ifa_addr,
sizeof(rrt->rrt_gw)) == 0) {
trace(1, "\troute found, age it\n");
if (rrt->rrt_t == 0 || rrt->rrt_t > t_lifetime) {
rrt->rrt_t = t_lifetime;
rrt->rrt_info.rip6_metric =
HOPCNT_INFINITY6;
updated++;
}
} else {
trace(1, "\tnon-p2p route found: %s/%d on %d\n",
inet6_n2p(&rrt->rrt_info.rip6_dest),
rrt->rrt_info.rip6_plen,
rrt->rrt_index);
}
} else
trace(1, "\tno p2p route found\n");
}
return updated ? 0 : -1;
}
/*
* Get each interface address and put those interface routes to the route
* list.
*/
void
ifrt(ifcp, again)
struct ifc *ifcp;
int again;
{
struct ifac *ifa;
struct riprt *rrt;
struct netinfo6 *np;
if (ifcp->ifc_flags & IFF_LOOPBACK)
return; /* ignore loopback */
if (ifcp->ifc_flags & IFF_POINTOPOINT) {
ifrt_p2p(ifcp, again);
return;
}
for (ifa = ifcp->ifc_addr; ifa; ifa = ifa->ifa_next) {
if (IN6_IS_ADDR_LINKLOCAL(&ifa->ifa_addr)) {
#if 0
trace(1, "route: %s on %s: "
"skip linklocal interface address\n",
inet6_n2p(&ifa->ifa_addr), ifcp->ifc_name);
#endif
continue;
}
if (IN6_IS_ADDR_UNSPECIFIED(&ifa->ifa_addr)) {
#if 0
trace(1, "route: %s: skip unspec interface address\n",
ifcp->ifc_name);
#endif
continue;
}
if ((rrt = MALLOC(struct riprt)) == NULL)
fatal("malloc: struct riprt");
memset(rrt, 0, sizeof(*rrt));
rrt->rrt_same = NULL;
rrt->rrt_index = ifcp->ifc_index;
rrt->rrt_t = 0; /* don't age */
rrt->rrt_info.rip6_dest = ifa->ifa_addr;
rrt->rrt_info.rip6_tag = htons(routetag & 0xffff);
rrt->rrt_info.rip6_metric = 1 + ifcp->ifc_metric;
rrt->rrt_info.rip6_plen = ifa->ifa_plen;
applyplen(&rrt->rrt_info.rip6_dest, ifa->ifa_plen);
memset(&rrt->rrt_gw, 0, sizeof(struct in6_addr));
np = &rrt->rrt_info;
if (rtsearch(np) == NULL) {
/* Attach the route to the list */
trace(1, "route: %s/%d: register route (%s)\n",
inet6_n2p(&np->rip6_dest), np->rip6_plen,
ifcp->ifc_name);
rrt->rrt_next = riprt;
riprt = rrt;
} else {
/* Already found */
if (!again) {
trace(1, "route: %s/%d: "
"already registered (%s)\n",
inet6_n2p(&np->rip6_dest), np->rip6_plen,
ifcp->ifc_name);
}
free(rrt);
}
}
}
/*
* there are couple of p2p interface routing models. "behavior" lets
* you pick one. it looks that gated behavior fits best with BSDs,
* since BSD kernels does not look at prefix length on p2p interfaces.
*/
void
ifrt_p2p(ifcp, again)
struct ifc *ifcp;
int again;
{
struct ifac *ifa;
struct riprt *rrt;
struct netinfo6 *np;
struct in6_addr addr, dest;
int advert, ignore, i;
#define P2PADVERT_NETWORK 1
#define P2PADVERT_ADDR 2
#define P2PADVERT_DEST 4
#define P2PADVERT_MAX 4
const enum { CISCO, GATED, ROUTE6D } behavior = GATED;
const char *category;
const char *noadv;
for (ifa = ifcp->ifc_addr; ifa; ifa = ifa->ifa_next) {
addr = ifa->ifa_addr;
dest = ifa->ifa_raddr;
applyplen(&addr, ifa->ifa_plen);
applyplen(&dest, ifa->ifa_plen);
advert = ignore = 0;
switch (behavior) {
case CISCO:
/*
* honor addr/plen, just like normal shared medium
* interface. this may cause trouble if you reuse
* addr/plen on other interfaces.
*
* advertise addr/plen.
*/
advert |= P2PADVERT_NETWORK;
break;
case GATED:
/*
* prefixlen on p2p interface is meaningless.
* advertise addr/128 and dest/128.
*
* do not install network route to route6d routing
* table (if we do, it would prevent route installation
* for other p2p interface that shares addr/plen).
*/
advert |= P2PADVERT_ADDR;
advert |= P2PADVERT_DEST;
ignore |= P2PADVERT_NETWORK;
break;
case ROUTE6D:
/*
* just for testing...
*/
if (IN6_ARE_ADDR_EQUAL(&addr, &dest))
advert |= P2PADVERT_NETWORK;
else {
advert |= P2PADVERT_ADDR;
advert |= P2PADVERT_DEST;
ignore |= P2PADVERT_NETWORK;
}
break;
}
for (i = 1; i <= P2PADVERT_MAX; i *= 2) {
if ((ignore & i) != 0)
continue;
if ((rrt = MALLOC(struct riprt)) == NULL)
fatal("malloc: struct riprt");
memset(rrt, 0, sizeof(*rrt));
rrt->rrt_same = NULL;
rrt->rrt_index = ifcp->ifc_index;
rrt->rrt_t = 0; /* don't age */
switch (i) {
case P2PADVERT_NETWORK:
rrt->rrt_info.rip6_dest = ifa->ifa_addr;
rrt->rrt_info.rip6_plen = ifa->ifa_plen;
applyplen(&rrt->rrt_info.rip6_dest,
ifa->ifa_plen);
category = "network";
break;
case P2PADVERT_ADDR:
rrt->rrt_info.rip6_dest = ifa->ifa_addr;
rrt->rrt_info.rip6_plen = 128;
category = "addr";
break;
case P2PADVERT_DEST:
rrt->rrt_info.rip6_dest = ifa->ifa_raddr;
rrt->rrt_info.rip6_plen = 128;
category = "dest";
break;
}
if (IN6_IS_ADDR_UNSPECIFIED(&rrt->rrt_info.rip6_dest) ||
IN6_IS_ADDR_LINKLOCAL(&rrt->rrt_info.rip6_dest)) {
#if 0
trace(1, "route: %s: skip unspec/linklocal "
"(%s on %s)\n", category, ifcp->ifc_name);
#endif
free(rrt);
continue;
}
if ((advert & i) == 0) {
rrt->rrt_rflags |= RRTF_NOADVERTISE;
noadv = ", NO-ADV";
} else
noadv = "";
rrt->rrt_info.rip6_tag = htons(routetag & 0xffff);
rrt->rrt_info.rip6_metric = 1 + ifcp->ifc_metric;
memset(&rrt->rrt_gw, 0, sizeof(struct in6_addr));
np = &rrt->rrt_info;
if (rtsearch(np) == NULL) {
/* Attach the route to the list */
trace(1, "route: %s/%d: register route "
"(%s on %s%s)\n",
inet6_n2p(&np->rip6_dest), np->rip6_plen,
category, ifcp->ifc_name, noadv);
rrt->rrt_next = riprt;
riprt = rrt;
} else {
/* Already found */
if (!again) {
trace(1, "route: %s/%d: "
"already registered (%s on %s%s)\n",
inet6_n2p(&np->rip6_dest),
np->rip6_plen, category,
ifcp->ifc_name, noadv);
}
free(rrt);
}
}
}
#undef P2PADVERT_NETWORK
#undef P2PADVERT_ADDR
#undef P2PADVERT_DEST
#undef P2PADVERT_MAX
}
int
getifmtu(ifindex)
int ifindex;
{
int mib[6];
char *buf;
size_t msize;
struct if_msghdr *ifm;
int mtu;
mib[0] = CTL_NET;
mib[1] = PF_ROUTE;
mib[2] = 0;
mib[3] = AF_INET6;
mib[4] = NET_RT_IFLIST;
mib[5] = ifindex;
if (sysctl(mib, 6, NULL, &msize, NULL, 0) < 0)
fatal("sysctl estimate NET_RT_IFLIST");
if ((buf = malloc(msize)) == NULL)
fatal("malloc");
if (sysctl(mib, 6, buf, &msize, NULL, 0) < 0)
fatal("sysctl NET_RT_IFLIST");
ifm = (struct if_msghdr *)buf;
mtu = ifm->ifm_data.ifi_mtu;
#ifdef __FREEBSD__
if (ifindex != ifm->ifm_index)
fatal("ifindex does not match with ifm_index");
#endif /* __FREEBSD__ */
free(buf);
return mtu;
}
const char *
rttypes(rtm)
struct rt_msghdr *rtm;
{
#define RTTYPE(s, f) \
do { \
if (rtm->rtm_type == (f)) \
return (s); \
} while (0)
RTTYPE("ADD", RTM_ADD);
RTTYPE("DELETE", RTM_DELETE);
RTTYPE("CHANGE", RTM_CHANGE);
RTTYPE("GET", RTM_GET);
RTTYPE("LOSING", RTM_LOSING);
RTTYPE("REDIRECT", RTM_REDIRECT);
RTTYPE("MISS", RTM_MISS);
RTTYPE("LOCK", RTM_LOCK);
RTTYPE("OLDADD", RTM_OLDADD);
RTTYPE("OLDDEL", RTM_OLDDEL);
RTTYPE("RESOLVE", RTM_RESOLVE);
RTTYPE("NEWADDR", RTM_NEWADDR);
RTTYPE("DELADDR", RTM_DELADDR);
RTTYPE("IFINFO", RTM_IFINFO);
#undef RTTYPE
return NULL;
}
const char *
rtflags(rtm)
struct rt_msghdr *rtm;
{
static char buf[BUFSIZ];
strcpy(buf, "");
#define RTFLAG(s, f) \
do { \
if (rtm->rtm_flags & (f)) \
strcat(buf, (s)); \
} while (0)
RTFLAG("U", RTF_UP);
RTFLAG("G", RTF_GATEWAY);
RTFLAG("H", RTF_HOST);
RTFLAG("R", RTF_REJECT);
RTFLAG("D", RTF_DYNAMIC);
RTFLAG("M", RTF_MODIFIED);
RTFLAG("d", RTF_DONE);
#ifdef RTF_MASK
RTFLAG("m", RTF_MASK);
#endif
RTFLAG("C", RTF_CLONING);
RTFLAG("X", RTF_XRESOLVE);
RTFLAG("L", RTF_LLINFO);
RTFLAG("S", RTF_STATIC);
RTFLAG("B", RTF_BLACKHOLE);
RTFLAG("2", RTF_PROTO2);
RTFLAG("1", RTF_PROTO1);
#undef RTFLAG
return buf;
}
const char *
ifflags(flags)
int flags;
{
static char buf[BUFSIZ];
strcpy(buf, "");
#define IFFLAG(s, f) \
do { \
if (flags & f) { \
if (buf[0]) \
strcat(buf, ","); \
strcat(buf, s); \
} \
} while (0)
IFFLAG("UP", IFF_UP);
IFFLAG("BROADCAST", IFF_BROADCAST);
IFFLAG("DEBUG", IFF_DEBUG);
IFFLAG("LOOPBACK", IFF_LOOPBACK);
IFFLAG("POINTOPOINT", IFF_POINTOPOINT);
#ifdef IFF_NOTRAILERS
IFFLAG("NOTRAILERS", IFF_NOTRAILERS);
#endif
IFFLAG("RUNNING", IFF_RUNNING);
IFFLAG("NOARP", IFF_NOARP);
IFFLAG("PROMISC", IFF_PROMISC);
IFFLAG("ALLMULTI", IFF_ALLMULTI);
IFFLAG("OACTIVE", IFF_OACTIVE);
IFFLAG("SIMPLEX", IFF_SIMPLEX);
IFFLAG("LINK0", IFF_LINK0);
IFFLAG("LINK1", IFF_LINK1);
IFFLAG("LINK2", IFF_LINK2);
IFFLAG("MULTICAST", IFF_MULTICAST);
#undef IFFLAG
return buf;
}
void
krtread(again)
int again;
{
int mib[6];
size_t msize;
char *buf, *p, *lim;
struct rt_msghdr *rtm;
int retry;
const char *errmsg;
retry = 0;
buf = NULL;
mib[0] = CTL_NET;
mib[1] = PF_ROUTE;
mib[2] = 0;
mib[3] = AF_INET6; /* Address family */
mib[4] = NET_RT_DUMP; /* Dump the kernel routing table */
mib[5] = 0; /* No flags */
do {
retry++;
errmsg = NULL;
if (buf)
free(buf);
if (sysctl(mib, 6, NULL, &msize, NULL, 0) < 0) {
errmsg = "sysctl estimate";
continue;
}
if ((buf = malloc(msize)) == NULL) {
errmsg = "malloc";
continue;
}
if (sysctl(mib, 6, buf, &msize, NULL, 0) < 0) {
errmsg = "sysctl NET_RT_DUMP";
continue;
}
} while (retry < 5 && errmsg != NULL);
if (errmsg)
fatal("%s (with %d retries, msize=%lu)", errmsg, retry,
(u_long)msize);
else if (1 < retry)
syslog(LOG_INFO, "NET_RT_DUMP %d retires", retry);
lim = buf + msize;
for (p = buf; p < lim; p += rtm->rtm_msglen) {
rtm = (struct rt_msghdr *)p;
rt_entry(rtm, again);
}
free(buf);
}
void
rt_entry(rtm, again)
struct rt_msghdr *rtm;
int again;
{
struct sockaddr_in6 *sin6_dst, *sin6_gw, *sin6_mask;
struct sockaddr_in6 *sin6_genmask, *sin6_ifp;
char *rtmp, *ifname = NULL;
struct riprt *rrt;
struct netinfo6 *np;
int s;
sin6_dst = sin6_gw = sin6_mask = sin6_genmask = sin6_ifp = 0;
if ((rtm->rtm_flags & RTF_UP) == 0 || rtm->rtm_flags &
(RTF_CLONING|RTF_XRESOLVE|RTF_LLINFO|RTF_BLACKHOLE)) {
return; /* not interested in the link route */
}
/* do not look at cloned routes */
#ifdef RTF_WASCLONED
if (rtm->rtm_flags & RTF_WASCLONED)
return;
#endif
#ifdef RTF_CLONED
if (rtm->rtm_flags & RTF_CLONED)
return;
#endif
/*
* do not look at dynamic routes.
* netbsd/openbsd cloned routes have UGHD.
*/
if (rtm->rtm_flags & RTF_DYNAMIC)
return;
rtmp = (char *)(rtm + 1);
/* Destination */
if ((rtm->rtm_addrs & RTA_DST) == 0)
return; /* ignore routes without destination address */
sin6_dst = (struct sockaddr_in6 *)rtmp;
rtmp += ROUNDUP(sin6_dst->sin6_len);
if (rtm->rtm_addrs & RTA_GATEWAY) {
sin6_gw = (struct sockaddr_in6 *)rtmp;
rtmp += ROUNDUP(sin6_gw->sin6_len);
}
if (rtm->rtm_addrs & RTA_NETMASK) {
sin6_mask = (struct sockaddr_in6 *)rtmp;
rtmp += ROUNDUP(sin6_mask->sin6_len);
}
if (rtm->rtm_addrs & RTA_GENMASK) {
sin6_genmask = (struct sockaddr_in6 *)rtmp;
rtmp += ROUNDUP(sin6_genmask->sin6_len);
}
if (rtm->rtm_addrs & RTA_IFP) {
sin6_ifp = (struct sockaddr_in6 *)rtmp;
rtmp += ROUNDUP(sin6_ifp->sin6_len);
}
/* Destination */
if (sin6_dst->sin6_family != AF_INET6)
return;
if (IN6_IS_ADDR_LINKLOCAL(&sin6_dst->sin6_addr))
return; /* Link-local */
if (IN6_ARE_ADDR_EQUAL(&sin6_dst->sin6_addr, &in6addr_loopback))
return; /* Loopback */
if (IN6_IS_ADDR_MULTICAST(&sin6_dst->sin6_addr))
return;
if ((rrt = MALLOC(struct riprt)) == NULL)
fatal("malloc: struct riprt");
memset(rrt, 0, sizeof(*rrt));
np = &rrt->rrt_info;
rrt->rrt_same = NULL;
rrt->rrt_t = time(NULL);
if (aflag == 0 && (rtm->rtm_flags & RTF_STATIC))
rrt->rrt_t = 0; /* Don't age static routes */
#if 0
np->rip6_tag = htons(routetag & 0xffff);
#else
np->rip6_tag = 0;
#endif
np->rip6_metric = rtm->rtm_rmx.rmx_hopcount;
if (np->rip6_metric < 1)
np->rip6_metric = 1;
rrt->rrt_flags = rtm->rtm_flags;
np->rip6_dest = sin6_dst->sin6_addr;
/* Mask or plen */
if (rtm->rtm_flags & RTF_HOST)
np->rip6_plen = 128; /* Host route */
else if (sin6_mask) {
np->rip6_plen = mask2len(&sin6_mask->sin6_addr,
sin6_mask->sin6_len - offsetof(struct sockaddr_in6, sin6_addr));
} else
np->rip6_plen = 0;
if (rtsearch(np)) {
/* Already found */
if (!again) {
trace(1, "route: %s/%d flags %s: already registered\n",
inet6_n2p(&np->rip6_dest), np->rip6_plen,
rtflags(rtm));
}
free(rrt);
return;
}
/* Gateway */
if (!sin6_gw)
memset(&rrt->rrt_gw, 0, sizeof(struct in6_addr));
else {
if (sin6_gw->sin6_family == AF_INET6)
rrt->rrt_gw = sin6_gw->sin6_addr;
else if (sin6_gw->sin6_family == AF_LINK) {
/* XXX in case ppp link? */
rrt->rrt_gw = in6addr_loopback;
} else
memset(&rrt->rrt_gw, 0, sizeof(struct in6_addr));
}
trace(1, "route: %s/%d flags %s",
inet6_n2p(&np->rip6_dest), np->rip6_plen, rtflags(rtm));
trace(1, " gw %s", inet6_n2p(&rrt->rrt_gw));
/* Interface */
s = rtm->rtm_index;
if (s < nindex2ifc && index2ifc[s])
ifname = index2ifc[s]->ifc_name;
else {
trace(1, " not configured\n");
free(rrt);
return;
}
trace(1, " if %s sock %d", ifname, s);
rrt->rrt_index = s;
trace(1, "\n");
/* Check gateway */
if (!IN6_IS_ADDR_LINKLOCAL(&rrt->rrt_gw) &&
!IN6_IS_ADDR_LOOPBACK(&rrt->rrt_gw)
#ifdef __FreeBSD__
&& (rrt->rrt_flags & RTF_LOCAL) == 0
#endif
) {
trace(0, "***** Gateway %s is not a link-local address.\n",
inet6_n2p(&rrt->rrt_gw));
trace(0, "***** dest(%s) if(%s) -- Not optimized.\n",
inet6_n2p(&rrt->rrt_info.rip6_dest), ifname);
rrt->rrt_rflags |= RRTF_NH_NOT_LLADDR;
}
/* Put it to the route list */
rrt->rrt_next = riprt;
riprt = rrt;
}
int
addroute(rrt, gw, ifcp)
struct riprt *rrt;
const struct in6_addr *gw;
struct ifc *ifcp;
{
struct netinfo6 *np;
u_char buf[BUFSIZ], buf1[BUFSIZ], buf2[BUFSIZ];
struct rt_msghdr *rtm;
struct sockaddr_in6 *sin;
int len;
np = &rrt->rrt_info;
inet_ntop(AF_INET6, (void *)gw, (char *)buf1, sizeof(buf1));
inet_ntop(AF_INET6, (void *)&ifcp->ifc_mylladdr, (char *)buf2, sizeof(buf2));
tracet(1, "ADD: %s/%d gw %s [%d] ifa %s\n",
inet6_n2p(&np->rip6_dest), np->rip6_plen, buf1,
np->rip6_metric - 1, buf2);
if (rtlog)
fprintf(rtlog, "%s: ADD: %s/%d gw %s [%d] ifa %s\n", hms(),
inet6_n2p(&np->rip6_dest), np->rip6_plen, buf1,
np->rip6_metric - 1, buf2);
if (nflag)
return 0;
memset(buf, 0, sizeof(buf));
rtm = (struct rt_msghdr *)buf;
rtm->rtm_type = RTM_ADD;
rtm->rtm_version = RTM_VERSION;
rtm->rtm_seq = ++seq;
rtm->rtm_pid = pid;
rtm->rtm_flags = rrt->rrt_flags;
rtm->rtm_addrs = RTA_DST | RTA_GATEWAY | RTA_NETMASK;
rtm->rtm_rmx.rmx_hopcount = np->rip6_metric - 1;
rtm->rtm_inits = RTV_HOPCOUNT;
sin = (struct sockaddr_in6 *)&buf[sizeof(struct rt_msghdr)];
/* Destination */
sin->sin6_len = sizeof(struct sockaddr_in6);
sin->sin6_family = AF_INET6;
sin->sin6_addr = np->rip6_dest;
sin = (struct sockaddr_in6 *)((char *)sin + ROUNDUP(sin->sin6_len));
/* Gateway */
sin->sin6_len = sizeof(struct sockaddr_in6);
sin->sin6_family = AF_INET6;
sin->sin6_addr = *gw;
sin = (struct sockaddr_in6 *)((char *)sin + ROUNDUP(sin->sin6_len));
/* Netmask */
sin->sin6_len = sizeof(struct sockaddr_in6);
sin->sin6_family = AF_INET6;
sin->sin6_addr = *(plen2mask(np->rip6_plen));
sin = (struct sockaddr_in6 *)((char *)sin + ROUNDUP(sin->sin6_len));
len = (char *)sin - (char *)buf;
rtm->rtm_msglen = len;
if (write(rtsock, buf, len) > 0)
return 0;
if (errno == EEXIST) {
trace(0, "ADD: Route already exists %s/%d gw %s\n",
inet6_n2p(&np->rip6_dest), np->rip6_plen, buf1);
if (rtlog)
fprintf(rtlog, "ADD: Route already exists %s/%d gw %s\n",
inet6_n2p(&np->rip6_dest), np->rip6_plen, buf1);
} else {
trace(0, "Can not write to rtsock (addroute): %s\n",
strerror(errno));
if (rtlog)
fprintf(rtlog, "\tCan not write to rtsock: %s\n",
strerror(errno));
}
return -1;
}
int
delroute(np, gw)
struct netinfo6 *np;
struct in6_addr *gw;
{
u_char buf[BUFSIZ], buf2[BUFSIZ];
struct rt_msghdr *rtm;
struct sockaddr_in6 *sin;
int len;
inet_ntop(AF_INET6, (void *)gw, (char *)buf2, sizeof(buf2));
tracet(1, "DEL: %s/%d gw %s\n", inet6_n2p(&np->rip6_dest),
np->rip6_plen, buf2);
if (rtlog)
fprintf(rtlog, "%s: DEL: %s/%d gw %s\n",
hms(), inet6_n2p(&np->rip6_dest), np->rip6_plen, buf2);
if (nflag)
return 0;
memset(buf, 0, sizeof(buf));
rtm = (struct rt_msghdr *)buf;
rtm->rtm_type = RTM_DELETE;
rtm->rtm_version = RTM_VERSION;
rtm->rtm_seq = ++seq;
rtm->rtm_pid = pid;
rtm->rtm_flags = RTF_UP | RTF_GATEWAY;
rtm->rtm_addrs = RTA_DST | RTA_GATEWAY | RTA_NETMASK;
sin = (struct sockaddr_in6 *)&buf[sizeof(struct rt_msghdr)];
/* Destination */
sin->sin6_len = sizeof(struct sockaddr_in6);
sin->sin6_family = AF_INET6;
sin->sin6_addr = np->rip6_dest;
sin = (struct sockaddr_in6 *)((char *)sin + ROUNDUP(sin->sin6_len));
/* Gateway */
sin->sin6_len = sizeof(struct sockaddr_in6);
sin->sin6_family = AF_INET6;
sin->sin6_addr = *gw;
sin = (struct sockaddr_in6 *)((char *)sin + ROUNDUP(sin->sin6_len));
/* Netmask */
sin->sin6_len = sizeof(struct sockaddr_in6);
sin->sin6_family = AF_INET6;
sin->sin6_addr = *(plen2mask(np->rip6_plen));
sin = (struct sockaddr_in6 *)((char *)sin + ROUNDUP(sin->sin6_len));
len = (char *)sin - (char *)buf;
rtm->rtm_msglen = len;
if (write(rtsock, buf, len) >= 0)
return 0;
if (errno == ESRCH) {
trace(0, "RTDEL: Route does not exist: %s/%d gw %s\n",
inet6_n2p(&np->rip6_dest), np->rip6_plen, buf2);
if (rtlog)
fprintf(rtlog, "RTDEL: Route does not exist: %s/%d gw %s\n",
inet6_n2p(&np->rip6_dest), np->rip6_plen, buf2);
} else {
trace(0, "Can not write to rtsock (delroute): %s\n",
strerror(errno));
if (rtlog)
fprintf(rtlog, "\tCan not write to rtsock: %s\n",
strerror(errno));
}
return -1;
}
struct in6_addr *
getroute(np, gw)
struct netinfo6 *np;
struct in6_addr *gw;
{
u_char buf[BUFSIZ];
u_long myseq;
int len;
struct rt_msghdr *rtm;
struct sockaddr_in6 *sin;
rtm = (struct rt_msghdr *)buf;
len = sizeof(struct rt_msghdr) + sizeof(struct sockaddr_in6);
memset(rtm, 0, len);
rtm->rtm_type = RTM_GET;
rtm->rtm_version = RTM_VERSION;
myseq = ++seq;
rtm->rtm_seq = myseq;
rtm->rtm_addrs = RTA_DST;
rtm->rtm_msglen = len;
sin = (struct sockaddr_in6 *)&buf[sizeof(struct rt_msghdr)];
sin->sin6_len = sizeof(struct sockaddr_in6);
sin->sin6_family = AF_INET6;
sin->sin6_addr = np->rip6_dest;
if (write(rtsock, buf, len) < 0) {
if (errno == ESRCH) /* No such route found */
return NULL;
perror("write to rtsock");
exit(-1);
}
do {
if ((len = read(rtsock, buf, sizeof(buf))) < 0) {
perror("read from rtsock");
exit(-1);
}
rtm = (struct rt_msghdr *)buf;
} while (rtm->rtm_seq != myseq || rtm->rtm_pid != pid);
sin = (struct sockaddr_in6 *)&buf[sizeof(struct rt_msghdr)];
if (rtm->rtm_addrs & RTA_DST) {
sin = (struct sockaddr_in6 *)
((char *)sin + ROUNDUP(sin->sin6_len));
}
if (rtm->rtm_addrs & RTA_GATEWAY) {
*gw = sin->sin6_addr;
return gw;
}
return NULL;
}
const char *
inet6_n2p(p)
const struct in6_addr *p;
{
static char buf[BUFSIZ];
return inet_ntop(AF_INET6, (void *)p, buf, sizeof(buf));
}
void
ifrtdump(sig)
int sig;
{
ifdump(sig);
rtdump(sig);
}
void
ifdump(sig)
int sig;
{
struct ifc *ifcp;
FILE *dump;
int i;
if (sig == 0)
dump = stderr;
else
if ((dump = fopen(ROUTE6D_DUMP, "a")) == NULL)
dump = stderr;
fprintf(dump, "%s: Interface Table Dump\n", hms());
fprintf(dump, " Number of interfaces: %d\n", nifc);
for (i = 0; i < 2; i++) {
fprintf(dump, " %sadvertising interfaces:\n", i ? "non-" : "");
for (ifcp = ifc; ifcp; ifcp = ifcp->ifc_next) {
if (i == 0) {
if ((ifcp->ifc_flags & IFF_UP) == 0)
continue;
if (iff_find(ifcp, 'N') != NULL)
continue;
} else {
if (ifcp->ifc_flags & IFF_UP)
continue;
}
ifdump0(dump, ifcp);
}
}
fprintf(dump, "\n");
if (dump != stderr)
fclose(dump);
}
void
ifdump0(dump, ifcp)
FILE *dump;
const struct ifc *ifcp;
{
struct ifac *ifa;
struct iff *iffp;
char buf[BUFSIZ];
const char *ft;
int addr;
fprintf(dump, " %s: index(%d) flags(%s) addr(%s) mtu(%d) metric(%d)\n",
ifcp->ifc_name, ifcp->ifc_index, ifflags(ifcp->ifc_flags),
inet6_n2p(&ifcp->ifc_mylladdr),
ifcp->ifc_mtu, ifcp->ifc_metric);
for (ifa = ifcp->ifc_addr; ifa; ifa = ifa->ifa_next) {
if (ifcp->ifc_flags & IFF_POINTOPOINT) {
inet_ntop(AF_INET6, (void *)&ifa->ifa_raddr,
buf, sizeof(buf));
fprintf(dump, "\t%s/%d -- %s\n",
inet6_n2p(&ifa->ifa_addr),
ifa->ifa_plen, buf);
} else {
fprintf(dump, "\t%s/%d\n",
inet6_n2p(&ifa->ifa_addr),
ifa->ifa_plen);
}
}
if (ifcp->ifc_filter) {
fprintf(dump, "\tFilter:");
for (iffp = ifcp->ifc_filter; iffp; iffp = iffp->iff_next) {
addr = 0;
switch (iffp->iff_type) {
case 'A':
ft = "Aggregate"; addr++; break;
case 'N':
ft = "No-advertise"; break;
case 'O':
ft = "Advertise-only"; addr++; break;
case 'T':
ft = "Default-only"; break;
case 'L':
ft = "Listen-only"; addr++; break;
default:
snprintf(buf, sizeof(buf), "Unknown-%c", iffp->iff_type);
ft = buf;
addr++;
break;
}
fprintf(dump, " %s", ft);
if (addr) {
fprintf(dump, "(%s/%d)", inet6_n2p(&iffp->iff_addr),
iffp->iff_plen);
}
}
fprintf(dump, "\n");
}
}
void
rtdump(sig)
int sig;
{
struct riprt *rrt;
char buf[BUFSIZ];
FILE *dump;
time_t t, age;
if (sig == 0)
dump = stderr;
else
if ((dump = fopen(ROUTE6D_DUMP, "a")) == NULL)
dump = stderr;
t = time(NULL);
fprintf(dump, "\n%s: Routing Table Dump\n", hms());
for (rrt = riprt; rrt; rrt = rrt->rrt_next) {
if (rrt->rrt_t == 0)
age = 0;
else
age = t - rrt->rrt_t;
inet_ntop(AF_INET6, (void *)&rrt->rrt_info.rip6_dest,
buf, sizeof(buf));
fprintf(dump, " %s/%d if(%d:%s) gw(%s) [%d] age(%ld)",
buf, rrt->rrt_info.rip6_plen, rrt->rrt_index,
index2ifc[rrt->rrt_index]->ifc_name,
inet6_n2p(&rrt->rrt_gw),
rrt->rrt_info.rip6_metric, (long)age);
if (rrt->rrt_info.rip6_tag) {
fprintf(dump, " tag(0x%04x)",
ntohs(rrt->rrt_info.rip6_tag) & 0xffff);
}
if (rrt->rrt_rflags & RRTF_NH_NOT_LLADDR)
fprintf(dump, " NOT-LL");
if (rrt->rrt_rflags & RRTF_NOADVERTISE)
fprintf(dump, " NO-ADV");
fprintf(dump, "\n");
}
fprintf(dump, "\n");
if (dump != stderr)
fclose(dump);
}
/*
* Parse the -A (and -O) options and put corresponding filter object to the
* specified interface structures. Each of the -A/O option has the following
* syntax: -A 5f09:c400::/32,ef0,ef1 (aggregate)
* -O 5f09:c400::/32,ef0,ef1 (only when match)
*/
void
filterconfig()
{
int i;
char *p, *ap, *iflp, *ifname;
struct iff ftmp, *iff_obj;
struct ifc *ifcp;
struct riprt *rrt;
#if 0
struct in6_addr gw;
#endif
for (i = 0; i < nfilter; i++) {
ap = filter[i];
iflp = NULL;
ifcp = NULL;
if (filtertype[i] == 'N' || filtertype[i] == 'T') {
iflp = ap;
goto ifonly;
}
if ((p = index(ap, ',')) != NULL) {
*p++ = '\0';
iflp = p;
}
if ((p = index(ap, '/')) == NULL)
fatal("no prefixlen specified for '%s'", ap);
*p++ = '\0';
if (inet_pton(AF_INET6, ap, &ftmp.iff_addr) != 1)
fatal("invalid prefix specified for '%s'", ap);
ftmp.iff_plen = atoi(p);
ftmp.iff_next = NULL;
applyplen(&ftmp.iff_addr, ftmp.iff_plen);
ifonly:
ftmp.iff_type = filtertype[i];
if (iflp == NULL || *iflp == '\0')
fatal("no interface specified for '%s'", ap);
/* parse the interface listing portion */
while (iflp) {
ifname = iflp;
if ((iflp = index(iflp, ',')) != NULL)
*iflp++ = '\0';
ifcp = ifc_find(ifname);
if (ifcp == NULL)
fatal("no interface %s exists", ifname);
iff_obj = (struct iff *)malloc(sizeof(struct iff));
if (iff_obj == NULL)
fatal("malloc of iff_obj");
memcpy((void *)iff_obj, (void *)&ftmp,
sizeof(struct iff));
/* link it to the interface filter */
iff_obj->iff_next = ifcp->ifc_filter;
ifcp->ifc_filter = iff_obj;
}
if (filtertype[i] != 'A')
continue;
/* put the aggregate to the kernel routing table */
rrt = (struct riprt *)malloc(sizeof(struct riprt));
if (rrt == NULL)
fatal("malloc: rrt");
memset(rrt, 0, sizeof(struct riprt));
rrt->rrt_info.rip6_dest = ftmp.iff_addr;
rrt->rrt_info.rip6_plen = ftmp.iff_plen;
rrt->rrt_info.rip6_metric = 1;
rrt->rrt_info.rip6_tag = htons(routetag & 0xffff);
rrt->rrt_gw = in6addr_loopback;
rrt->rrt_flags = RTF_UP | RTF_REJECT;
rrt->rrt_rflags = RRTF_AGGREGATE;
rrt->rrt_t = 0;
rrt->rrt_index = loopifindex;
#if 0
if (getroute(&rrt->rrt_info, &gw)) {
#if 0
/*
* When the address has already been registered in the
* kernel routing table, it should be removed
*/
delroute(&rrt->rrt_info, &gw);
#else
/* it is more safe behavior */
errno = EINVAL;
fatal("%s/%u already in routing table, "
"cannot aggregate",
inet6_n2p(&rrt->rrt_info.rip6_dest),
rrt->rrt_info.rip6_plen);
#endif
}
#endif
/* Put the route to the list */
rrt->rrt_next = riprt;
riprt = rrt;
trace(1, "Aggregate: %s/%d for %s\n",
inet6_n2p(&ftmp.iff_addr), ftmp.iff_plen,
ifcp->ifc_name);
/* Add this route to the kernel */
if (nflag) /* do not modify kernel routing table */
continue;
addroute(rrt, &in6addr_loopback, loopifcp);
}
}
/***************** utility functions *****************/
/*
* Returns a pointer to ifac whose address and prefix length matches
* with the address and prefix length specified in the arguments.
*/
struct ifac *
ifa_match(ifcp, ia, plen)
const struct ifc *ifcp;
const struct in6_addr *ia;
int plen;
{
struct ifac *ifa;
for (ifa = ifcp->ifc_addr; ifa; ifa = ifa->ifa_next) {
if (IN6_ARE_ADDR_EQUAL(&ifa->ifa_addr, ia) &&
ifa->ifa_plen == plen)
break;
}
return ifa;
}
/*
* Return a pointer to riprt structure whose address and prefix length
* matches with the address and prefix length found in the argument.
* Note: This is not a rtalloc(). Therefore exact match is necessary.
*/
struct riprt *
rtsearch(np)
struct netinfo6 *np;
{
struct riprt *rrt;
for (rrt = riprt; rrt; rrt = rrt->rrt_next) {
if (rrt->rrt_info.rip6_plen == np->rip6_plen &&
IN6_ARE_ADDR_EQUAL(&rrt->rrt_info.rip6_dest,
&np->rip6_dest))
return rrt;
}
return 0;
}
int
mask2len(addr, lenlim)
const struct in6_addr *addr;
int lenlim;
{
int i = 0, j;
u_char *p = (u_char *)addr;
for (j = 0; j < lenlim; j++, p++) {
if (*p != 0xff)
break;
i += 8;
}
if (j < lenlim) {
switch (*p) {
#define MASKLEN(m, l) case m: do { i += l; break; } while (0)
MASKLEN(0xfe, 7); break;
MASKLEN(0xfc, 6); break;
MASKLEN(0xf8, 5); break;
MASKLEN(0xf0, 4); break;
MASKLEN(0xe0, 3); break;
MASKLEN(0xc0, 2); break;
MASKLEN(0x80, 1); break;
#undef MASKLEN
}
}
return i;
}
void
applymask(addr, mask)
struct in6_addr *addr, *mask;
{
int i;
u_long *p, *q;
p = (u_long *)addr; q = (u_long *)mask;
for (i = 0; i < 4; i++)
*p++ &= *q++;
}
static const u_char plent[8] = {
0x00, 0x80, 0xc0, 0xe0, 0xf0, 0xf8, 0xfc, 0xfe
};
void
applyplen(ia, plen)
struct in6_addr *ia;
int plen;
{
u_char *p;
int i;
p = ia->s6_addr;
for (i = 0; i < 16; i++) {
if (plen <= 0)
*p = 0;
else if (plen < 8)
*p &= plent[plen];
p++, plen -= 8;
}
}
static const int pl2m[9] = {
0x00, 0x80, 0xc0, 0xe0, 0xf0, 0xf8, 0xfc, 0xfe, 0xff
};
struct in6_addr *
plen2mask(n)
int n;
{
static struct in6_addr ia;
u_char *p;
int i;
memset(&ia, 0, sizeof(struct in6_addr));
p = (u_char *)&ia;
for (i = 0; i < 16; i++, p++, n -= 8) {
if (n >= 8) {
*p = 0xff;
continue;
}
*p = pl2m[n];
break;
}
return &ia;
}
char *
allocopy(p)
char *p;
{
char *q = (char *)malloc(strlen(p) + 1);
strcpy(q, p);
return q;
}
char *
hms()
{
static char buf[BUFSIZ];
time_t t;
struct tm *tm;
t = time(NULL);
if ((tm = localtime(&t)) == 0)
fatal("localtime");
snprintf(buf, sizeof(buf), "%02d:%02d:%02d", tm->tm_hour, tm->tm_min, tm->tm_sec);
return buf;
}
#define RIPRANDDEV 1.0 /* 30 +- 15, max - min = 30 */
int
ripinterval(timer)
int timer;
{
double r = rand();
interval = (int)(timer + timer * RIPRANDDEV * (r / RAND_MAX - 0.5));
nextalarm = time(NULL) + interval;
return interval;
}
time_t
ripsuptrig()
{
time_t t;
double r = rand();
t = (int)(RIP_TRIG_INT6_MIN +
(RIP_TRIG_INT6_MAX - RIP_TRIG_INT6_MIN) * (r / RAND_MAX ));
sup_trig_update = time(NULL) + t;
return t;
}
void
#ifdef __STDC__
fatal(const char *fmt, ...)
#else
fatal(fmt, va_alist)
char *fmt;
va_dcl
#endif
{
va_list ap;
char buf[1024];
#ifdef __STDC__
va_start(ap, fmt);
#else
va_start(ap);
#endif
vsnprintf(buf, sizeof(buf), fmt, ap);
perror(buf);
syslog(LOG_ERR, "%s: %s", buf, strerror(errno));
rtdexit(0);
va_end(ap);
}
void
#ifdef __STDC__
tracet(int level, const char *fmt, ...)
#else
tracet(level, fmt, va_alist)
int level;
char *fmt;
va_dcl
#endif
{
va_list ap;
#ifdef __STDC__
va_start(ap, fmt);
#else
va_start(ap);
#endif
if (level <= dflag) {
fprintf(stderr, "%s: ", hms());
vfprintf(stderr, fmt, ap);
}
if (dflag) {
if (level > 0)
vsyslog(LOG_DEBUG, fmt, ap);
else
vsyslog(LOG_WARNING, fmt, ap);
}
va_end(ap);
}
void
#ifdef __STDC__
trace(int level, const char *fmt, ...)
#else
trace(level, fmt, va_alist)
int level;
char *fmt;
va_dcl
#endif
{
va_list ap;
#ifdef __STDC__
va_start(ap, fmt);
#else
va_start(ap);
#endif
if (level <= dflag)
vfprintf(stderr, fmt, ap);
if (dflag) {
if (level > 0)
vsyslog(LOG_DEBUG, fmt, ap);
else
vsyslog(LOG_WARNING, fmt, ap);
}
va_end(ap);
}
unsigned int
if_maxindex()
{
struct if_nameindex *p, *p0;
unsigned int max = 0;
p0 = if_nameindex();
for (p = p0; p && p->if_index && p->if_name; p++) {
if (max < p->if_index)
max = p->if_index;
}
if_freenameindex(p0);
return max;
}
struct ifc *
ifc_find(name)
char *name;
{
struct ifc *ifcp;
for (ifcp = ifc; ifcp; ifcp = ifcp->ifc_next) {
if (strcmp(name, ifcp->ifc_name) == 0)
return ifcp;
}
return (struct ifc *)NULL;
}
struct iff *
iff_find(ifcp, type)
struct ifc *ifcp;
int type;
{
struct iff *iffp;
for (iffp = ifcp->ifc_filter; iffp; iffp = iffp->iff_next) {
if (iffp->iff_type == type)
return iffp;
}
return NULL;
}
void
setindex2ifc(index, ifcp)
int index;
struct ifc *ifcp;
{
int n;
struct ifc **p;
if (!index2ifc) {
nindex2ifc = 5; /*initial guess*/
index2ifc = (struct ifc **)
malloc(sizeof(*index2ifc) * nindex2ifc);
if (index2ifc == NULL)
fatal("malloc");
memset(index2ifc, 0, sizeof(*index2ifc) * nindex2ifc);
}
n = nindex2ifc;
while (nindex2ifc <= index)
nindex2ifc *= 2;
if (n != nindex2ifc) {
p = (struct ifc **)realloc(index2ifc,
sizeof(*index2ifc) * nindex2ifc);
if (p == NULL)
fatal("realloc");
index2ifc = p;
}
index2ifc[index] = ifcp;
}