1
0
mirror of https://git.FreeBSD.org/src.git synced 2024-12-22 11:17:19 +00:00
freebsd/usr.sbin/mrouted/main.c

1065 lines
24 KiB
C
Raw Normal View History

/*
* The mrouted program is covered by the license in the accompanying file
* named "LICENSE". Use of the mrouted program represents acceptance of
* the terms and conditions listed in that file.
*
* The mrouted program is COPYRIGHT 1989 by The Board of Trustees of
* Leland Stanford Junior University.
*
*
* main.c,v 3.8.4.29 1998/03/01 01:49:00 fenner Exp
*/
/*
* Written by Steve Deering, Stanford University, February 1989.
*
* (An earlier version of DVMRP was implemented by David Waitzman of
* BBN STC by extending Berkeley's routed program. Some of Waitzman's
* extensions have been incorporated into mrouted, but none of the
* original routed code has been adopted.)
*/
#ifndef lint
static const char rcsid[] =
1999-08-28 01:35:59 +00:00
"$FreeBSD$";
#endif
#include <err.h>
#include "defs.h"
#ifdef __STDC__
#include <stdarg.h>
#else
#include <varargs.h>
#endif
#include <fcntl.h>
#include <paths.h>
#ifdef SNMP
#include "snmp.h"
#endif
extern char *configfilename;
char versionstring[100];
static char pidfilename[] = _PATH_MROUTED_PID;
static char dumpfilename[] = _PATH_MROUTED_DUMP;
static char cachefilename[] = _PATH_MROUTED_CACHE;
static char genidfilename[] = _PATH_MROUTED_GENID;
static int haveterminal = 1;
int did_final_init = 0;
static int sighandled = 0;
#define GOT_SIGINT 0x01
#define GOT_SIGHUP 0x02
#define GOT_SIGUSR1 0x04
#define GOT_SIGUSR2 0x08
int cache_lifetime = DEFAULT_CACHE_LIFETIME;
int prune_lifetime = AVERAGE_PRUNE_LIFETIME;
int debug = 0;
char *progname;
time_t mrouted_init_time;
#ifdef SNMP
#define NHANDLERS 34
#else
#define NHANDLERS 2
#endif
static struct ihandler {
int fd; /* File descriptor */
ihfunc_t func; /* Function to call with &fd_set */
} ihandlers[NHANDLERS];
static int nhandlers = 0;
static struct debugname {
char *name;
int level;
int nchars;
} debugnames[] = {
{ "packet", DEBUG_PKT, 2 },
{ "pkt", DEBUG_PKT, 3 },
{ "pruning", DEBUG_PRUNE, 1 },
{ "prunes", DEBUG_PRUNE, 1 },
{ "routing", DEBUG_ROUTE, 1 },
{ "routes", DEBUG_ROUTE, 1 },
{ "route_detail", DEBUG_RTDETAIL, 6 },
{ "rtdetail", DEBUG_RTDETAIL, 2 },
{ "peers", DEBUG_PEER, 2 },
{ "neighbors", DEBUG_PEER, 1 },
{ "cache", DEBUG_CACHE, 1 },
{ "timeout", DEBUG_TIMEOUT, 1 },
{ "callout", DEBUG_TIMEOUT, 2 },
{ "interface", DEBUG_IF, 2 },
{ "vif", DEBUG_IF, 1 },
{ "membership", DEBUG_MEMBER, 1 },
{ "groups", DEBUG_MEMBER, 1 },
{ "traceroute", DEBUG_TRACE, 2 },
{ "mtrace", DEBUG_TRACE, 2 },
{ "igmp", DEBUG_IGMP, 1 },
{ "icmp", DEBUG_ICMP, 2 },
{ "rsrr", DEBUG_RSRR, 2 },
{ "3", 0xffffffff, 1 } /* compat. */
};
/*
* Forward declarations.
*/
static void final_init __P((void *));
static void fasttimer __P((void *));
static void timer __P((void *));
static void dump __P((void));
static void dump_version __P((FILE *));
static void fdump __P((void));
static void cdump __P((void));
static void restart __P((void));
static void handler __P((int));
static void cleanup __P((void));
static void resetlogging __P((void *));
static void usage __P((void));
/* To shut up gcc -Wstrict-prototypes */
int main __P((int argc, char **argv));
int
register_input_handler(fd, func)
int fd;
ihfunc_t func;
{
if (nhandlers >= NHANDLERS)
return -1;
ihandlers[nhandlers].fd = fd;
ihandlers[nhandlers++].func = func;
return 0;
}
int
main(argc, argv)
int argc;
char *argv[];
{
register int recvlen;
int dummy;
FILE *fp;
struct timeval tv, difftime, curtime, lasttime, *timeout;
u_int32 prev_genid;
int vers;
fd_set rfds, readers;
int nfds, n, i, secs;
extern char todaysversion[];
struct sigaction sa;
#ifdef SNMP
struct timeval timeout, *tvp = &timeout;
struct timeval sched, *svp = &sched, now, *nvp = &now;
int index, block;
#endif
setlinebuf(stderr);
if (geteuid() != 0)
errx(1, "must be root");
progname = strrchr(argv[0], '/');
if (progname)
progname++;
else
progname = argv[0];
argv++, argc--;
while (argc > 0 && *argv[0] == '-') {
if (strcmp(*argv, "-d") == 0) {
if (argc > 1 && *(argv + 1)[0] != '-') {
char *p,*q;
int i, len;
struct debugname *d;
argv++, argc--;
debug = 0;
p = *argv; q = NULL;
while (p) {
q = strchr(p, ',');
if (q)
*q++ = '\0';
len = strlen(p);
for (i = 0, d = debugnames;
i < sizeof(debugnames) / sizeof(debugnames[0]);
i++, d++)
if (len >= d->nchars && strncmp(d->name, p, len) == 0)
break;
if (i == sizeof(debugnames) / sizeof(debugnames[0])) {
int j = 0xffffffff;
int k = 0;
fprintf(stderr, "Valid debug levels: ");
for (i = 0, d = debugnames;
i < sizeof(debugnames) / sizeof(debugnames[0]);
i++, d++) {
if ((j & d->level) == d->level) {
if (k++)
putc(',', stderr);
fputs(d->name, stderr);
j &= ~d->level;
}
}
putc('\n', stderr);
usage();
}
debug |= d->level;
p = q;
}
} else
debug = DEFAULT_DEBUG;
} else if (strcmp(*argv, "-c") == 0) {
if (argc > 1) {
argv++, argc--;
configfilename = *argv;
} else
usage();
} else if (strcmp(*argv, "-p") == 0) {
log(LOG_WARNING, 0, "disabling pruning is no longer supported");
#ifdef SNMP
} else if (strcmp(*argv, "-P") == 0) {
if (argc > 1 && isdigit(*(argv + 1)[0])) {
argv++, argc--;
dest_port = atoi(*argv);
} else
dest_port = DEFAULT_PORT;
#endif
} else
usage();
argv++, argc--;
}
if (argc > 0)
usage();
if (debug != 0) {
struct debugname *d;
char c;
int tmpd = debug;
fprintf(stderr, "debug level 0x%x ", debug);
c = '(';
for (d = debugnames; d < debugnames +
sizeof(debugnames) / sizeof(debugnames[0]); d++) {
if ((tmpd & d->level) == d->level) {
tmpd &= ~d->level;
fprintf(stderr, "%c%s", c, d->name);
c = ',';
}
}
fprintf(stderr, ")\n");
}
#ifdef LOG_DAEMON
(void)openlog("mrouted", LOG_PID, LOG_DAEMON);
(void)setlogmask(LOG_UPTO(LOG_NOTICE));
#else
(void)openlog("mrouted", LOG_PID);
#endif
sprintf(versionstring, "mrouted version %s", todaysversion);
log(LOG_DEBUG, 0, "%s starting", versionstring);
#ifdef SYSV
srand48(time(NULL));
#endif
/*
* Get generation id
*/
gettimeofday(&tv, 0);
dvmrp_genid = tv.tv_sec;
fp = fopen(genidfilename, "r");
if (fp != NULL) {
fscanf(fp, "%d", &prev_genid);
if (prev_genid == dvmrp_genid)
dvmrp_genid++;
(void) fclose(fp);
}
fp = fopen(genidfilename, "w");
if (fp != NULL) {
fprintf(fp, "%d", dvmrp_genid);
(void) fclose(fp);
}
/* Start up the log rate-limiter */
resetlogging(NULL);
callout_init();
init_igmp();
init_icmp();
init_ipip();
init_routes();
init_ktable();
#ifndef OLD_KERNEL
/*
* Unfortunately, you can't k_get_version() unless you've
* k_init_dvmrp()'d. Now that we want to move the
* k_init_dvmrp() to later in the initialization sequence,
* we have to do the disgusting hack of initializing,
* getting the version, then stopping the kernel multicast
* forwarding.
*/
k_init_dvmrp();
vers = k_get_version();
k_stop_dvmrp();
/*XXX
* This function must change whenever the kernel version changes
*/
if ((((vers >> 8) & 0xff) != 3) ||
((vers & 0xff) != 5))
log(LOG_ERR, 0, "kernel (v%d.%d)/mrouted (v%d.%d) version mismatch",
(vers >> 8) & 0xff, vers & 0xff,
PROTOCOL_VERSION, MROUTED_VERSION);
#endif
#ifdef SNMP
if (i = snmp_init())
return i;
gettimeofday(nvp, 0);
if (nvp->tv_usec < 500000L){
svp->tv_usec = nvp->tv_usec + 500000L;
svp->tv_sec = nvp->tv_sec;
} else {
svp->tv_usec = nvp->tv_usec - 500000L;
svp->tv_sec = nvp->tv_sec + 1;
}
#endif /* SNMP */
init_vifs();
#ifdef RSRR
rsrr_init();
#endif /* RSRR */
sa.sa_handler = handler;
sa.sa_flags = 0; /* Interrupt system calls */
sigemptyset(&sa.sa_mask);
sigaction(SIGHUP, &sa, NULL);
sigaction(SIGTERM, &sa, NULL);
sigaction(SIGINT, &sa, NULL);
sigaction(SIGUSR1, &sa, NULL);
sigaction(SIGUSR2, &sa, NULL);
if (igmp_socket >= FD_SETSIZE)
log(LOG_ERR, 0, "descriptor too big");
FD_ZERO(&readers);
FD_SET(igmp_socket, &readers);
nfds = igmp_socket + 1;
for (i = 0; i < nhandlers; i++) {
if (ihandlers[i].fd >= FD_SETSIZE)
log(LOG_ERR, 0, "descriptor too big");
FD_SET(ihandlers[i].fd, &readers);
if (ihandlers[i].fd >= nfds)
nfds = ihandlers[i].fd + 1;
}
IF_DEBUG(DEBUG_IF)
dump_vifs(stderr);
IF_DEBUG(DEBUG_ROUTE)
dump_routes(stderr);
/* schedule first timer interrupt */
timer_setTimer(1, fasttimer, NULL);
timer_setTimer(TIMER_INTERVAL, timer, NULL);
if (debug == 0) {
/*
* Detach from the terminal
*/
int t;
haveterminal = 0;
if (fork()) exit(0);
(void)close(0);
(void)close(1);
(void)close(2);
(void)open("/", 0);
(void)dup2(0, 1);
(void)dup2(0, 2);
#if defined(SYSV) || defined(linux)
(void)setpgrp();
#else
#ifdef TIOCNOTTY
t = open(_PATH_TTY, 2);
if (t >= 0) {
(void)ioctl(t, TIOCNOTTY, (char *)0);
(void)close(t);
}
#else
if (setsid() < 0)
perror("setsid");
#endif
#endif
}
fp = fopen(pidfilename, "w");
if (fp != NULL) {
fprintf(fp, "%d\n", (int)getpid());
(void) fclose(fp);
}
/* XXX HACK
* This will cause black holes for the first few seconds after startup,
* since we are exchanging routes but not actually forwarding.
* However, it eliminates much of the startup transient.
*
* It's possible that we can set a flag which says not to report any
* routes (just accept reports) until this timer fires, and then
* do a report_to_all_neighbors(ALL_ROUTES) immediately before
* turning on DVMRP.
*/
timer_setTimer(10, final_init, NULL);
/*
* Main receive loop.
*/
dummy = 0;
difftime.tv_usec = 0;
gettimeofday(&curtime, NULL);
lasttime = curtime;
for(;;) {
bcopy((char *)&readers, (char *)&rfds, sizeof(rfds));
secs = timer_nextTimer();
if (secs == -1)
timeout = NULL;
else {
timeout = &tv;
timeout->tv_sec = secs;
timeout->tv_usec = 0;
}
#ifdef SNMP
THIS IS BROKEN
if (nvp->tv_sec > svp->tv_sec
|| (nvp->tv_sec == svp->tv_sec && nvp->tv_usec > svp->tv_usec)){
alarmTimer(nvp);
eventTimer(nvp);
if (nvp->tv_usec < 500000L){
svp->tv_usec = nvp->tv_usec + 500000L;
svp->tv_sec = nvp->tv_sec;
} else {
svp->tv_usec = nvp->tv_usec - 500000L;
svp->tv_sec = nvp->tv_sec + 1;
}
}
tvp = &timeout;
tvp->tv_sec = 0;
tvp->tv_usec = 500000L;
block = 0;
snmp_select_info(&nfds, &rfds, tvp, &block);
if (block == 1)
tvp = NULL; /* block without timeout */
if ((n = select(nfds, &rfds, NULL, NULL, tvp)) < 0)
#endif
if (sighandled) {
if (sighandled & GOT_SIGINT) {
sighandled &= ~GOT_SIGINT;
break;
}
if (sighandled & GOT_SIGHUP) {
sighandled &= ~GOT_SIGHUP;
restart();
}
if (sighandled & GOT_SIGUSR1) {
sighandled &= ~GOT_SIGUSR1;
fdump();
}
if (sighandled & GOT_SIGUSR2) {
sighandled &= ~GOT_SIGUSR2;
cdump();
}
}
if ((n = select(nfds, &rfds, NULL, NULL, timeout)) < 0) {
if (errno != EINTR)
log(LOG_WARNING, errno, "select failed");
continue;
}
if (n > 0) {
if (FD_ISSET(igmp_socket, &rfds)) {
recvlen = recvfrom(igmp_socket, recv_buf, RECV_BUF_SIZE,
0, NULL, &dummy);
if (recvlen < 0) {
if (errno != EINTR) log(LOG_ERR, errno, "recvfrom");
continue;
}
accept_igmp(recvlen);
}
for (i = 0; i < nhandlers; i++) {
if (FD_ISSET(ihandlers[i].fd, &rfds)) {
(*ihandlers[i].func)(ihandlers[i].fd, &rfds);
}
}
}
#ifdef SNMP
THIS IS BROKEN
snmp_read(&rfds);
snmp_timeout(); /* poll */
#endif
/*
* Handle timeout queue.
*
* If select + packet processing took more than 1 second,
* or if there is a timeout pending, age the timeout queue.
*
* If not, collect usec in difftime to make sure that the
* time doesn't drift too badly.
*
* If the timeout handlers took more than 1 second,
* age the timeout queue again. XXX This introduces the
* potential for infinite loops!
*/
do {
/*
* If the select timed out, then there's no other
* activity to account for and we don't need to
* call gettimeofday.
*/
if (n == 0) {
curtime.tv_sec = lasttime.tv_sec + secs;
curtime.tv_usec = lasttime.tv_usec;
n = -1; /* don't do this next time through the loop */
} else
gettimeofday(&curtime, NULL);
difftime.tv_sec = curtime.tv_sec - lasttime.tv_sec;
difftime.tv_usec += curtime.tv_usec - lasttime.tv_usec;
2000-03-26 14:44:34 +00:00
while (difftime.tv_usec >= 1000000) {
difftime.tv_sec++;
difftime.tv_usec -= 1000000;
}
if (difftime.tv_usec < 0) {
difftime.tv_sec--;
difftime.tv_usec += 1000000;
}
lasttime = curtime;
if (secs == 0 || difftime.tv_sec > 0)
age_callout_queue(difftime.tv_sec);
secs = -1;
} while (difftime.tv_sec > 0);
}
log(LOG_NOTICE, 0, "%s exiting", versionstring);
cleanup();
exit(0);
}
static void
usage()
{
fprintf(stderr,
"usage: mrouted [-p] [-c configfile] [-d [debug_level]]\n");
exit(1);
}
static void
final_init(i)
void *i;
{
char *s = (char *)i;
log(LOG_NOTICE, 0, "%s%s", versionstring, s ? s : "");
if (s)
free(s);
k_init_dvmrp(); /* enable DVMRP routing in kernel */
/*
* Install the vifs in the kernel as late as possible in the
* initialization sequence.
*/
init_installvifs();
time(&mrouted_init_time);
did_final_init = 1;
}
/*
* routine invoked every second. Its main goal is to cycle through
* the routing table and send partial updates to all neighbors at a
* rate that will cause the entire table to be sent in ROUTE_REPORT_INTERVAL
* seconds. Also, every TIMER_INTERVAL seconds it calls timer() to
* do all the other time-based processing.
*/
static void
fasttimer(i)
void *i;
{
static unsigned int tlast;
static unsigned int nsent;
register unsigned int t = tlast + 1;
register int n;
/*
* if we're in the last second, send everything that's left.
* otherwise send at least the fraction we should have sent by now.
*/
if (t >= ROUTE_REPORT_INTERVAL) {
register int nleft = nroutes - nsent;
while (nleft > 0) {
if ((n = report_next_chunk()) <= 0)
break;
nleft -= n;
}
tlast = 0;
nsent = 0;
} else {
register unsigned int ncum = nroutes * t / ROUTE_REPORT_INTERVAL;
while (nsent < ncum) {
if ((n = report_next_chunk()) <= 0)
break;
nsent += n;
}
tlast = t;
}
timer_setTimer(1, fasttimer, NULL);
}
/*
* The 'virtual_time' variable is initialized to a value that will cause the
* first invocation of timer() to send a probe or route report to all vifs
* and send group membership queries to all subnets for which this router is
* querier. This first invocation occurs approximately TIMER_INTERVAL seconds
* after the router starts up. Note that probes for neighbors and queries
* for group memberships are also sent at start-up time, as part of initial-
* ization. This repetition after a short interval is desirable for quickly
* building up topology and membership information in the presence of possible
* packet loss.
*
* 'virtual_time' advances at a rate that is only a crude approximation of
* real time, because it does not take into account any time spent processing,
* and because the timer intervals are sometimes shrunk by a random amount to
* avoid unwanted synchronization with other routers.
*/
u_long virtual_time = 0;
/*
* Timer routine. Performs periodic neighbor probing, route reporting, and
* group querying duties, and drives various timers in routing entries and
* virtual interface data structures.
*/
static void
timer(i)
void *i;
{
age_routes(); /* Advance the timers in the route entries */
age_vifs(); /* Advance the timers for neighbors */
age_table_entry(); /* Advance the timers for the cache entries */
if (virtual_time % IGMP_QUERY_INTERVAL == 0) {
/*
* Time to query the local group memberships on all subnets
* for which this router is the elected querier.
*/
query_groups();
}
if (virtual_time % NEIGHBOR_PROBE_INTERVAL == 0) {
/*
* Time to send a probe on all vifs from which no neighbors have
* been heard. Also, check if any inoperative interfaces have now
* come up. (If they have, they will also be probed as part of
* their initialization.)
*/
probe_for_neighbors();
if (vifs_down)
check_vif_state();
}
delay_change_reports = FALSE;
if (routes_changed) {
/*
* Some routes have changed since the last timer interrupt, but
* have not been reported yet. Report the changed routes to all
* neighbors.
*/
report_to_all_neighbors(CHANGED_ROUTES);
}
#ifdef SNMP
sync_timer();
#endif
/*
* Advance virtual time
*/
virtual_time += TIMER_INTERVAL;
timer_setTimer(TIMER_INTERVAL, timer, NULL);
}
static void
cleanup()
{
static in_cleanup = 0;
if (!in_cleanup) {
in_cleanup++;
#ifdef RSRR
rsrr_clean();
#endif /* RSRR */
expire_all_routes();
report_to_all_neighbors(ALL_ROUTES);
if (did_final_init)
k_stop_dvmrp();
}
}
/*
* Signal handler. Take note of the fact that the signal arrived
* so that the main loop can take care of it.
*/
static void
handler(sig)
int sig;
{
switch (sig) {
case SIGINT:
case SIGTERM:
sighandled |= GOT_SIGINT;
break;
case SIGHUP:
sighandled |= GOT_SIGHUP;
break;
case SIGUSR1:
sighandled |= GOT_SIGUSR1;
break;
case SIGUSR2:
sighandled |= GOT_SIGUSR2;
break;
}
}
/*
* Dump internal data structures to stderr.
*/
static void
dump()
{
dump_vifs(stderr);
dump_routes(stderr);
}
static void
dump_version(fp)
FILE *fp;
{
time_t t;
time(&t);
fprintf(fp, "%s ", versionstring);
if (did_final_init)
fprintf(fp, "up %s",
scaletime(t - mrouted_init_time));
else
fprintf(fp, "(not yet initialized)");
fprintf(fp, " %s\n", ctime(&t));
}
/*
* Dump internal data structures to a file.
*/
static void
fdump()
{
FILE *fp;
fp = fopen(dumpfilename, "w");
if (fp != NULL) {
dump_version(fp);
dump_vifs(fp);
dump_routes(fp);
(void) fclose(fp);
}
}
/*
* Dump local cache contents to a file.
*/
static void
cdump()
{
FILE *fp;
fp = fopen(cachefilename, "w");
if (fp != NULL) {
dump_version(fp);
dump_cache(fp);
(void) fclose(fp);
}
}
/*
* Restart mrouted
*/
static void
restart()
{
char *s;
s = (char *)malloc(sizeof(" restart"));
if (s == NULL)
log(LOG_ERR, 0, "out of memory");
strcpy(s, " restart");
/*
* reset all the entries
*/
free_all_prunes();
free_all_routes();
free_all_callouts();
stop_all_vifs();
k_stop_dvmrp();
close(igmp_socket);
close(udp_socket);
did_final_init = 0;
/*
* start processing again
*/
dvmrp_genid++;
init_igmp();
init_routes();
init_ktable();
init_vifs();
/*XXX Schedule final_init() as main does? */
final_init(s);
/* schedule timer interrupts */
timer_setTimer(1, fasttimer, NULL);
timer_setTimer(TIMER_INTERVAL, timer, NULL);
}
#define LOG_MAX_MSGS 20 /* if > 20/minute then shut up for a while */
#define LOG_SHUT_UP 600 /* shut up for 10 minutes */
static int log_nmsgs = 0;
static void
resetlogging(arg)
void *arg;
{
int nxttime = 60;
void *narg = NULL;
if (arg == NULL && log_nmsgs > LOG_MAX_MSGS) {
nxttime = LOG_SHUT_UP;
narg = (void *)&log_nmsgs; /* just need some valid void * */
syslog(LOG_WARNING, "logging too fast, shutting up for %d minutes",
LOG_SHUT_UP / 60);
} else {
log_nmsgs = 0;
}
timer_setTimer(nxttime, resetlogging, narg);
}
char *
scaletime(t)
u_long t;
{
#define SCALETIMEBUFLEN 20
static char buf1[20];
static char buf2[20];
static char *buf = buf1;
char *p;
p = buf;
if (buf == buf1)
buf = buf2;
else
buf = buf1;
/* XXX snprintf */
sprintf(p, "%2ld:%02ld:%02ld", t / 3600, (t % 3600) / 60, t % 60);
p[SCALETIMEBUFLEN - 1] = '\0';
return p;
}
#ifdef RINGBUFFER
#define NLOGMSGS 10000
#define LOGMSGSIZE 200
char *logmsg[NLOGMSGS];
static int logmsgno = 0;
void
printringbuf()
{
FILE *f;
int i;
f = fopen("/var/tmp/mrouted.log", "a");
if (f == NULL) {
log(LOG_ERR, errno, "can't open /var/tmp/mrouted.log");
/*NOTREACHED*/
}
fprintf(f, "--------------------------------------------\n");
i = (logmsgno + 1) % NLOGMSGS;
while (i != logmsgno) {
if (*logmsg[i]) {
fprintf(f, "%s\n", logmsg[i]);
*logmsg[i] = '\0';
}
i = (i + 1) % NLOGMSGS;
}
fclose(f);
}
#endif
/*
* Log errors and other messages to the system log daemon and to stderr,
* according to the severity of the message and the current debug level.
* For errors of severity LOG_ERR or worse, terminate the program.
*/
#ifdef __STDC__
void
log(int severity, int syserr, char *format, ...)
{
va_list ap;
static char fmt[211] = "warning - ";
char *msg;
struct timeval now;
1998-06-29 17:38:35 +00:00
time_t now_sec;
struct tm *thyme;
#ifdef RINGBUFFER
static int ringbufinit = 0;
#endif
va_start(ap, format);
#else
/*VARARGS3*/
void
log(severity, syserr, format, va_alist)
int severity, syserr;
char *format;
va_dcl
{
va_list ap;
static char fmt[311] = "warning - ";
char *msg;
char tbuf[20];
struct timeval now;
1998-06-29 17:38:35 +00:00
time_t now_sec;
struct tm *thyme;
#ifdef RINGBUFFER
static int ringbufinit = 0;
#endif
va_start(ap);
#endif
vsnprintf(&fmt[10], sizeof(fmt) - 10, format, ap);
va_end(ap);
msg = (severity == LOG_WARNING) ? fmt : &fmt[10];
#ifdef RINGBUFFER
if (!ringbufinit) {
int i;
for (i = 0; i < NLOGMSGS; i++) {
logmsg[i] = malloc(LOGMSGSIZE);
if (logmsg[i] == 0) {
syslog(LOG_ERR, "out of memory");
exit(-1);
}
*logmsg[i] = 0;
}
ringbufinit = 1;
}
gettimeofday(&now,NULL);
now_sec = now.tv_sec;
thyme = localtime(&now_sec);
snprintf(logmsg[logmsgno++], LOGMSGSIZE, "%02d:%02d:%02d.%03ld %s err %d",
thyme->tm_hour, thyme->tm_min, thyme->tm_sec,
now.tv_usec / 1000, msg, syserr);
logmsgno %= NLOGMSGS;
if (severity <= LOG_NOTICE)
#endif
/*
* Log to stderr if we haven't forked yet and it's a warning or worse,
* or if we're debugging.
*/
if (haveterminal && (debug || severity <= LOG_WARNING)) {
gettimeofday(&now,NULL);
now_sec = now.tv_sec;
thyme = localtime(&now_sec);
if (!debug)
fprintf(stderr, "%s: ", progname);
fprintf(stderr, "%02d:%02d:%02d.%03ld %s", thyme->tm_hour,
thyme->tm_min, thyme->tm_sec, now.tv_usec / 1000, msg);
if (syserr == 0)
fprintf(stderr, "\n");
else if (syserr < sys_nerr)
fprintf(stderr, ": %s\n", sys_errlist[syserr]);
else
fprintf(stderr, ": errno %d\n", syserr);
}
/*
* Always log things that are worse than warnings, no matter what
* the log_nmsgs rate limiter says.
* Only count things worse than debugging in the rate limiter
* (since if you put daemon.debug in syslog.conf you probably
* actually want to log the debugging messages so they shouldn't
* be rate-limited)
*/
if ((severity < LOG_WARNING) || (log_nmsgs < LOG_MAX_MSGS)) {
if (severity < LOG_DEBUG)
log_nmsgs++;
if (syserr != 0) {
errno = syserr;
syslog(severity, "%s: %m", msg);
} else
syslog(severity, "%s", msg);
}
if (severity <= LOG_ERR) exit(-1);
}
#ifdef DEBUG_MFC
void
md_log(what, origin, mcastgrp)
int what;
u_int32 origin, mcastgrp;
{
static FILE *f = NULL;
struct timeval tv;
u_int32 buf[4];
if (!f) {
if ((f = fopen("/tmp/mrouted.clog", "w")) == NULL) {
log(LOG_ERR, errno, "open /tmp/mrouted.clog");
}
}
gettimeofday(&tv, NULL);
buf[0] = tv.tv_sec;
buf[1] = what;
buf[2] = origin;
buf[3] = mcastgrp;
fwrite(buf, sizeof(u_int32), 4, f);
}
#endif