int main(int argc, char *argv[])
{
char *msg;
int seq;
int argi;
int alertcolors, alertinterval;
char *configfn = NULL;
char *checkfn = NULL;
int checkpointinterval = 900;
char acklogfn[PATH_MAX];
FILE *acklogfd = NULL;
char notiflogfn[PATH_MAX];
FILE *notiflogfd = NULL;
char *tracefn = NULL;
struct sigaction sa;
int configchanged;
time_t lastxmit = 0;
MEMDEFINE(acklogfn);
MEMDEFINE(notiflogfn);
libxymon_init(argv[0]);
/* Dont save the error buffer */
save_errbuf = 0;
/* Load alert config */
alertcolors = colorset(xgetenv("ALERTCOLORS"), ((1 << COL_GREEN) | (1 << COL_BLUE)));
alertinterval = 60*atoi(xgetenv("ALERTREPEAT"));
/* Create our loookup-trees */
hostnames = xtreeNew(strcasecmp);
testnames = xtreeNew(strcasecmp);
locations = xtreeNew(strcasecmp);
for (argi=1; (argi < argc); argi++) {
if (argnmatch(argv[argi], "--config=")) {
configfn = strdup(strchr(argv[argi], '=')+1);
}
else if (argnmatch(argv[argi], "--checkpoint-file=")) {
checkfn = strdup(strchr(argv[argi], '=')+1);
}
else if (argnmatch(argv[argi], "--checkpoint-interval=")) {
char *p = strchr(argv[argi], '=') + 1;
checkpointinterval = atoi(p);
}
else if (argnmatch(argv[argi], "--dump-config")) {
load_alertconfig(configfn, alertcolors, alertinterval);
dump_alertconfig(1);
return 0;
}
else if (argnmatch(argv[argi], "--cfid")) {
include_configid = 1;
}
else if (argnmatch(argv[argi], "--test")) {
char *testhost = NULL, *testservice = NULL, *testpage = NULL,
*testcolor = "red", *testgroups = NULL;
void *hinfo;
int testdur = 0;
FILE *logfd = NULL;
activealerts_t *awalk = NULL;
int paramno = 0;
argi++; if (argi < argc) testhost = argv[argi];
argi++; if (argi < argc) testservice = argv[argi];
argi++;
while (argi < argc) {
if (strncasecmp(argv[argi], "--duration=", 11) == 0) {
testdur = durationvalue(strchr(argv[argi], '=')+1);
}
else if (strncasecmp(argv[argi], "--color=", 8) == 0) {
testcolor = strchr(argv[argi], '=')+1;
}
else if (strncasecmp(argv[argi], "--group=", 8) == 0) {
testgroups = strchr(argv[argi], '=')+1;
}
else if (strncasecmp(argv[argi], "--time=", 7) == 0) {
fakestarttime = (time_t)atoi(strchr(argv[argi], '=')+1);
}
else {
paramno++;
if (paramno == 1) testdur = atoi(argv[argi]);
else if (paramno == 2) testcolor = argv[argi];
else if (paramno == 3) fakestarttime = (time_t) atoi(argv[argi]);
}
argi++;
}
if ((testhost == NULL) || (testservice == NULL)) {
printf("Usage: xymond_alert --test HOST SERVICE [options]\n");
printf("Possible options:\n\t[--duration=MINUTES]\n\t[--color=COLOR]\n\t[--group=GROUPNAME]\n\t[--time=TIMESPEC]\n");
return 1;
}
load_hostnames(xgetenv("HOSTSCFG"), NULL, get_fqdn());
hinfo = hostinfo(testhost);
if (hinfo) {
testpage = strdup(xmh_item(hinfo, XMH_ALLPAGEPATHS));
}
else {
errprintf("Host not found in hosts.cfg - assuming it is on the top page\n");
testpage = "";
}
awalk = (activealerts_t *)calloc(1, sizeof(activealerts_t));
awalk->hostname = find_name(hostnames, testhost);
awalk->testname = find_name(testnames, testservice);
awalk->location = find_name(locations, testpage);
awalk->ip = strdup("127.0.0.1");
awalk->color = awalk->maxcolor = parse_color(testcolor);
awalk->pagemessage = "Test of the alert configuration";
awalk->eventstart = getcurrenttime(NULL) - testdur*60;
awalk->groups = (testgroups ? strdup(testgroups) : NULL);
awalk->state = A_PAGING;
awalk->cookie = 12345;
awalk->next = NULL;
logfd = fopen("/dev/null", "w");
starttrace(NULL);
testonly = 1;
load_alertconfig(configfn, alertcolors, alertinterval);
load_holidays(0);
send_alert(awalk, logfd);
return 0;
}
else if (argnmatch(argv[argi], "--trace=")) {
tracefn = strdup(strchr(argv[argi], '=')+1);
starttrace(tracefn);
}
else if (net_worker_option(argv[argi])) {
/* Handled in the subroutine */
}
else if (standardoption(argv[argi])) {
if (showhelp) return 0;
}
else {
errprintf("Unknown option '%s'\n", argv[argi]);
}
}
/* Do the network stuff if needed */
net_worker_run(ST_ALERT, LOC_SINGLESERVER, NULL);
if (checkfn) {
load_checkpoint(checkfn);
nextcheckpoint = gettimer() + checkpointinterval;
dbgprintf("Next checkpoint at %d, interval %d\n", (int) nextcheckpoint, checkpointinterval);
}
setup_signalhandler("xymond_alert");
/* Need to handle these ourselves, so we can shutdown and save state-info */
memset(&sa, 0, sizeof(sa));
sa.sa_handler = sig_handler;
sigaction(SIGPIPE, &sa, NULL);
sigaction(SIGTERM, &sa, NULL);
sigaction(SIGINT, &sa, NULL);
sigaction(SIGCHLD, &sa, NULL);
sigaction(SIGUSR1, &sa, NULL);
sigaction(SIGHUP, &sa, NULL);
if (xgetenv("XYMONSERVERLOGS")) {
sprintf(acklogfn, "%s/acknowledge.log", xgetenv("XYMONSERVERLOGS"));
acklogfd = fopen(acklogfn, "a");
sprintf(notiflogfn, "%s/notifications.log", xgetenv("XYMONSERVERLOGS"));
notiflogfd = fopen(notiflogfn, "a");
}
/*
* The general idea here is that this loop handles receiving of alert-
* and ack-messages from the master daemon, and maintains a list of
* host+test combinations that may have alerts going out.
*
* This module does not deal with any specific alert-configuration,
* it just picks up the alert messages, maintains the list of
* known tests that are in some sort of critical condition, and
* periodically pushes alerts to the do_alert.c module for handling.
*
* The only modification of alerts that happen here is the handling
* of when the next alert is due. It calls into the next_alert()
* routine to learn when an alert should be repeated, and also
* deals with Acknowledgments that stop alerts from going out for
* a period of time.
*/
while (running) {
char *eoln, *restofmsg;
char *metadata[20];
char *p;
int metacount;
char *hostname = NULL, *testname = NULL;
struct timespec timeout;
time_t now, nowtimer;
int anytogo;
activealerts_t *awalk;
int childstat;
nowtimer = gettimer();
if (checkfn && (nowtimer > nextcheckpoint)) {
dbgprintf("Saving checkpoint\n");
nextcheckpoint = nowtimer + checkpointinterval;
save_checkpoint(checkfn);
if (acklogfd) acklogfd = freopen(acklogfn, "a", acklogfd);
if (notiflogfd) notiflogfd = freopen(notiflogfn, "a", notiflogfd);
}
timeout.tv_sec = 60; timeout.tv_nsec = 0;
msg = get_xymond_message(C_PAGE, "xymond_alert", &seq, &timeout);
if (msg == NULL) {
running = 0;
continue;
}
/* See what time it is - must happen AFTER the timeout */
now = getcurrenttime(NULL);
/* Split the message in the first line (with meta-data), and the rest */
eoln = strchr(msg, '\n');
if (eoln) {
*eoln = '\0';
restofmsg = eoln+1;
}
else {
restofmsg = "";
}
/*
* Now parse the meta-data into elements.
* We use our own "gettok()" routine which works
* like strtok(), but can handle empty elements.
*/
metacount = 0;
memset(&metadata, 0, sizeof(metadata));
p = gettok(msg, "|");
while (p && (metacount < 19)) {
metadata[metacount] = p;
metacount++;
p = gettok(NULL, "|");
}
metadata[metacount] = NULL;
if (metacount > 3) hostname = metadata[3];
if (metacount > 4) testname = metadata[4];
if ((metacount > 10) && (strncmp(metadata[0], "@@page", 6) == 0)) {
/* @@page|timestamp|sender|hostname|testname|hostip|expiretime|color|prevcolor|changetime|location|cookie|osname|classname|grouplist|modifiers */
int newcolor, newalertstatus, oldalertstatus;
dbgprintf("Got page message from %s:%s\n", hostname, testname);
traceprintf("@@page %s:%s:%s=%s\n", hostname, testname, metadata[10], metadata[7]);
awalk = find_active(hostname, testname);
if (awalk == NULL) {
char *hwalk = find_name(hostnames, hostname);
char *twalk = find_name(testnames, testname);
char *pwalk = find_name(locations, metadata[10]);
awalk = (activealerts_t *)calloc(1, sizeof(activealerts_t));
awalk->hostname = hwalk;
awalk->testname = twalk;
awalk->location = pwalk;
awalk->cookie = -1;
awalk->state = A_DEAD;
/*
* Use changetime here, if we restart the alert module then
* this gets the duration values more right than using "now".
* Also, define this only when a new alert arrives - we should
* NOT clear this when a status goes yellow->red, or if it
* flaps between yellow and red.
*/
awalk->eventstart = atoi(metadata[9]);
add_active(awalk->hostname, awalk);
traceprintf("New record\n");
}
newcolor = parse_color(metadata[7]);
oldalertstatus = ((alertcolors & (1 << awalk->color)) != 0);
newalertstatus = ((alertcolors & (1 << newcolor)) != 0);
traceprintf("state %d->%d\n", oldalertstatus, newalertstatus);
if (newalertstatus) {
/* It's in an alert state. */
awalk->color = newcolor;
awalk->state = A_PAGING;
if (newcolor > awalk->maxcolor) {
if (awalk->maxcolor != 0) {
/*
* Severity has increased (yellow -> red).
* Clear the repeat-interval, and set maxcolor to
* the new color. If it drops to yellow again,
* maxcolor stays at red, so a test that flaps
* between yellow and red will only alert on red
* the first time, and then follow the repeat
* interval.
*/
dbgprintf("Severity increased, cleared repeat interval: %s/%s %s->%s\n",
awalk->hostname, awalk->testname,
colorname(awalk->maxcolor), colorname(newcolor));
clear_interval(awalk);
}
awalk->maxcolor = newcolor;
}
}
else {
/*
* Send one "recovered" message out now, then go to A_DEAD.
* Dont update the color here - we want recoveries to go out
* only if the alert color triggered an alert
*/
awalk->state = (newcolor == COL_BLUE) ? A_DISABLED : A_RECOVERED;
}
if (oldalertstatus != newalertstatus) {
dbgprintf("Alert status changed from %d to %d\n", oldalertstatus, newalertstatus);
clear_interval(awalk);
}
if (awalk->ip) xfree(awalk->ip);
awalk->ip = strdup(metadata[5]);
awalk->cookie = atoi(metadata[11]);
if (awalk->osname) xfree(awalk->osname);
awalk->osname = (metadata[12] ? strdup(metadata[12]) : NULL);
if (awalk->classname) xfree(awalk->classname);
awalk->classname = (metadata[13] ? strdup(metadata[13]) : NULL);
if (awalk->groups) xfree(awalk->groups);
awalk->groups = (metadata[14] ? strdup(metadata[14]) : NULL);
if (awalk->pagemessage) xfree(awalk->pagemessage);
if (metadata[15]) {
/* Modifiers are more interesting than the message itself */
awalk->pagemessage = (char *)malloc(strlen(awalk->hostname) + strlen(awalk->testname) + strlen(colorname(awalk->color)) + strlen(metadata[15]) + strlen(restofmsg) + 10);
sprintf(awalk->pagemessage, "%s:%s %s\n%s\n%s",
awalk->hostname, awalk->testname, colorname(awalk->color), metadata[15], restofmsg);
}
else {
awalk->pagemessage = strdup(restofmsg);
}
}
else if ((metacount > 5) && (strncmp(metadata[0], "@@ack", 5) == 0)) {
/* @@ack|timestamp|sender|hostname|testname|hostip|expiretime */
/*
* An ack is handled simply by setting the next
* alert-time to when the ack expires.
*/
time_t nextalert = atoi(metadata[6]);
dbgprintf("Got ack message from %s:%s\n", hostname, testname);
traceprintf("@@ack: %s:%s now=%d, ackeduntil %d\n",
hostname, testname, (int)now, (int)nextalert);
awalk = find_active(hostname, testname);
if (acklogfd) {
int cookie = (awalk ? awalk->cookie : -1);
int color = (awalk ? awalk->color : 0);
fprintf(acklogfd, "%d\t%d\t%d\t%d\t%s\t%s.%s\t%s\t%s\n",
(int)now, cookie,
(int)((nextalert - now) / 60), cookie,
"np_filename_not_used",
hostname, testname,
colorname(color),
nlencode(restofmsg));
fflush(acklogfd);
}
if (awalk && (awalk->state == A_PAGING)) {
traceprintf("Record updated\n");
awalk->state = A_ACKED;
awalk->nextalerttime = nextalert;
if (awalk->ackmessage) xfree(awalk->ackmessage);
awalk->ackmessage = strdup(restofmsg);
}
else {
traceprintf("No record\n");
}
}
else if ((metacount > 4) && (strncmp(metadata[0], "@@notify", 5) == 0)) {
/* @@notify|timestamp|sender|hostname|testname|pagepath */
char *hwalk = find_name(hostnames, hostname);
char *twalk = find_name(testnames, testname);
char *pwalk = find_name(locations, (metadata[5] ? metadata[5] : ""));
awalk = (activealerts_t *)calloc(1, sizeof(activealerts_t));
awalk->hostname = hwalk;
awalk->testname = twalk;
awalk->location = pwalk;
awalk->cookie = -1;
awalk->pagemessage = strdup(restofmsg);
awalk->eventstart = getcurrenttime(NULL);
awalk->state = A_NOTIFY;
add_active(awalk->hostname, awalk);
}
else if ((metacount > 3) &&
((strncmp(metadata[0], "@@drophost", 10) == 0) || (strncmp(metadata[0], "@@dropstate", 11) == 0))) {
/* @@drophost|timestamp|sender|hostname */
/* @@dropstate|timestamp|sender|hostname */
xtreePos_t handle;
handle = xtreeFind(hostnames, hostname);
if (handle != xtreeEnd(hostnames)) {
alertanchor_t *anchor = (alertanchor_t *)xtreeData(hostnames, handle);
for (awalk = anchor->head; (awalk); awalk = awalk->next) awalk->state = A_DEAD;
}
}
else if ((metacount > 4) && (strncmp(metadata[0], "@@droptest", 10) == 0)) {
/* @@droptest|timestamp|sender|hostname|testname */
awalk = find_active(hostname, testname);
if (awalk) awalk->state = A_DEAD;
}
else if ((metacount > 4) && (strncmp(metadata[0], "@@renamehost", 12) == 0)) {
/* @@renamehost|timestamp|sender|hostname|newhostname */
/*
* We handle rename's simply by dropping the alert. If there is still an
* active alert for the host, it will have to be dealt with when the next
* status update arrives.
*/
xtreePos_t handle;
handle = xtreeFind(hostnames, hostname);
if (handle != xtreeEnd(hostnames)) {
alertanchor_t *anchor = (alertanchor_t *)xtreeData(hostnames, handle);
for (awalk = anchor->head; (awalk); awalk = awalk->next) awalk->state = A_DEAD;
}
}
else if ((metacount > 5) && (strncmp(metadata[0], "@@renametest", 12) == 0)) {
/* @@renametest|timestamp|sender|hostname|oldtestname|newtestname */
/*
* We handle rename's simply by dropping the alert. If there is still an
* active alert for the host, it will have to be dealt with when the next
* status update arrives.
*/
awalk = find_active(hostname, testname);
if (awalk) awalk->state = A_DEAD;
}
else if (strncmp(metadata[0], "@@shutdown", 10) == 0) {
running = 0;
errprintf("Got a shutdown message\n");
continue;
}
else if (strncmp(metadata[0], "@@logrotate", 11) == 0) {
char *fn = xgetenv("XYMONCHANNEL_LOGFILENAME");
if (fn && strlen(fn)) {
reopen_file(fn, "a", stdout);
reopen_file(fn, "a", stderr);
if (tracefn) {
stoptrace();
starttrace(tracefn);
}
}
continue;
}
else if (strncmp(metadata[0], "@@reload", 8) == 0) {
/* Nothing ... right now */
}
else if (strncmp(metadata[0], "@@idle", 6) == 0) {
/* Timeout */
}
/*
* When a burst of alerts happen, we get lots of alert messages
* coming in quickly. So lets handle them in bunches and only
* do the full alert handling once every 10 secs - that lets us
* combine a bunch of alerts into one transmission process.
*/
if (nowtimer < (lastxmit+10)) continue;
lastxmit = nowtimer;
/*
* Loop through the activealerts list and see if anything is pending.
* This is an optimization, we could just as well just fork off the
* notification child and let it handle all of it. But there is no
* reason to fork a child process unless it is going to do something.
*/
configchanged = load_alertconfig(configfn, alertcolors, alertinterval);
configchanged += load_holidays(0);
anytogo = 0;
for (awalk = alistBegin(); (awalk); awalk = alistNext()) {
int anymatch = 0;
switch (awalk->state) {
case A_NORECIP:
if (!configchanged) break;
/* The configuration has changed - switch NORECIP -> PAGING */
awalk->state = A_PAGING;
clear_interval(awalk);
/* Fall through */
case A_PAGING:
if (have_recipient(awalk, &anymatch)) {
if (awalk->nextalerttime <= now) anytogo++;
}
else {
if (!anymatch) {
awalk->state = A_NORECIP;
cleanup_alert(awalk);
}
}
break;
case A_ACKED:
if (awalk->nextalerttime <= now) {
/* An ack has expired, so drop the ack message and switch to A_PAGING */
anytogo++;
if (awalk->ackmessage) xfree(awalk->ackmessage);
awalk->state = A_PAGING;
}
break;
case A_RECOVERED:
case A_DISABLED:
case A_NOTIFY:
anytogo++;
break;
case A_DEAD:
break;
}
}
dbgprintf("%d alerts to go\n", anytogo);
if (anytogo) {
pid_t childpid;
childpid = fork();
if (childpid == 0) {
/* The child */
start_alerts();
for (awalk = alistBegin(); (awalk); awalk = alistNext()) {
switch (awalk->state) {
case A_PAGING:
if (awalk->nextalerttime <= now) {
send_alert(awalk, notiflogfd);
}
break;
case A_ACKED:
/* Cannot be A_ACKED unless the ack is still valid, so no alert. */
break;
case A_RECOVERED:
case A_DISABLED:
case A_NOTIFY:
send_alert(awalk, notiflogfd);
break;
case A_NORECIP:
case A_DEAD:
break;
}
}
finish_alerts();
/* Child does not continue */
exit(0);
}
else if (childpid < 0) {
errprintf("Fork failed, cannot send alerts: %s\n", strerror(errno));
}
}
/* Update the state flag and the next-alert timestamp */
for (awalk = alistBegin(); (awalk); awalk = alistNext()) {
switch (awalk->state) {
case A_PAGING:
if (awalk->nextalerttime <= now) awalk->nextalerttime = next_alert(awalk);
break;
case A_NORECIP:
break;
case A_ACKED:
/* Still cannot get here except if ack is still valid */
break;
case A_RECOVERED:
case A_DISABLED:
case A_NOTIFY:
awalk->state = A_DEAD;
/* Fall through */
case A_DEAD:
cleanup_alert(awalk);
break;
}
}
clean_all_active();
/* Pickup any finished child processes to avoid zombies */
while (wait3(&childstat, WNOHANG, NULL) > 0) ;
}
if (checkfn) save_checkpoint(checkfn);
if (acklogfd) fclose(acklogfd);
if (notiflogfd) fclose(notiflogfd);
stoptrace();
MEMUNDEFINE(notiflogfn);
MEMUNDEFINE(acklogfn);
if (termsig >= 0) {
errprintf("Terminated by signal %d\n", termsig);
}
return 0;
}
int main(int argc, char *argv[])
{
char *msg;
int argi;
struct sigaction sa;
char *exthandler = NULL;
char *extids = NULL;
char *processor = NULL;
struct sockaddr_un ctlsockaddr;
int ctlsocket;
/* Handle program options. */
for (argi = 1; (argi < argc); argi++) {
if (strcmp(argv[argi], "--debug") == 0) {
debug = 1;
}
else if (argnmatch(argv[argi], "--rrddir=")) {
char *p = strchr(argv[argi], '=');
rrddir = strdup(p+1);
}
else if (argnmatch(argv[argi], "--extra-script=")) {
char *p = strchr(argv[argi], '=');
exthandler = strdup(p+1);
}
else if (argnmatch(argv[argi], "--extra-tests=")) {
char *p = strchr(argv[argi], '=');
extids = strdup(p+1);
}
else if (argnmatch(argv[argi], "--processor=")) {
char *p = strchr(argv[argi], '=');
processor = strdup(p+1);
}
else if (strcmp(argv[argi], "--no-cache") == 0) {
use_rrd_cache = 0;
}
else if (net_worker_option(argv[argi])) {
/* Handled in the subroutine */
}
}
save_errbuf = 0;
if ((rrddir == NULL) && xgetenv("XYMONRRDS")) {
rrddir = strdup(xgetenv("XYMONRRDS"));
}
if (exthandler && extids) setup_exthandler(exthandler, extids);
/* Do the network stuff if needed */
net_worker_run(ST_RRD, LOC_STICKY, update_locator_hostdata);
setup_signalhandler("xymond_rrd");
memset(&sa, 0, sizeof(sa));
sa.sa_handler = sig_handler;
sigaction(SIGHUP, &sa, NULL);
sigaction(SIGCHLD, &sa, NULL);
sigaction(SIGTERM, &sa, NULL);
sigaction(SIGINT, &sa, NULL);
signal(SIGPIPE, SIG_DFL);
/* Setup the control socket that receives cache-flush commands */
memset(&ctlsockaddr, 0, sizeof(ctlsockaddr));
sprintf(ctlsockaddr.sun_path, "%s/rrdctl.%d", xgetenv("XYMONTMP"), getpid());
unlink(ctlsockaddr.sun_path); /* In case it was accidentally left behind */
ctlsockaddr.sun_family = AF_UNIX;
ctlsocket = socket(AF_UNIX, SOCK_DGRAM, 0);
if (ctlsocket == -1) {
errprintf("Cannot create cache-control socket (%s)\n", strerror(errno));
return 1;
}
fcntl(ctlsocket, F_SETFL, O_NONBLOCK);
if (bind(ctlsocket, (struct sockaddr *)&ctlsockaddr, sizeof(ctlsockaddr)) == -1) {
errprintf("Cannot bind to cache-control socket (%s)\n", strerror(errno));
return 1;
}
/* Linux obeys filesystem permissions on the socket file, so make it world-accessible */
if (chmod(ctlsockaddr.sun_path, S_IRUSR|S_IWUSR|S_IRGRP|S_IWGRP|S_IROTH|S_IWOTH) == -1) {
errprintf("Setting permissions on cache-control socket failed: %s\n", strerror(errno));
}
/* Load the RRD definitions */
load_rrddefs();
/* If we are passing data to an external processor, create the pipe to it */
setup_extprocessor(processor);
running = 1;
while (running) {
char *eoln, *restofmsg = NULL;
char *metadata[MAX_META+1];
int metacount;
char *p;
char *hostname = NULL, *testname = NULL, *sender = NULL, *classname = NULL, *pagepaths = NULL;
xymonrrd_t *ldef = NULL;
time_t tstamp;
int childstat;
ssize_t n;
char ctlbuf[PATH_MAX];
int gotcachectlmessage;
time_t now;
/* See if we have any cache-control messages pending */
do {
n = recv(ctlsocket, ctlbuf, sizeof(ctlbuf), 0);
gotcachectlmessage = (n > 0);
if (gotcachectlmessage) {
/* We have a control message */
char *bol, *eol;
ctlbuf[n] = '\0';
bol = ctlbuf;
do {
eol = strchr(bol, '\n'); if (eol) *eol = '\0';
rrdcacheflushhost(bol);
if (eol) { bol = eol+1; } else bol = NULL;
} while (bol && *bol);
}
} while (gotcachectlmessage);
/* Get next message */
msg = get_xymond_message(C_LAST, argv[0], &seq, NULL);
if (msg == NULL) {
running = 0;
continue;
}
now = gettimer();
if (reloadtime < now) {
/* Reload configuration files */
load_hostnames(xgetenv("HOSTSCFG"), NULL, get_fqdn());
load_client_config(NULL);
reloadtime = now + 600;
}
/* Split the message in the first line (with meta-data), and the rest */
eoln = strchr(msg, '\n');
if (eoln) {
*eoln = '\0';
restofmsg = eoln+1;
}
/* Parse the meta-data */
metacount = 0;
memset(&metadata, 0, sizeof(metadata));
p = gettok(msg, "|");
while (p && (metacount < MAX_META)) {
metadata[metacount++] = p;
p = gettok(NULL, "|");
}
metadata[metacount] = NULL;
if ((metacount >= 14) && (strncmp(metadata[0], "@@status", 8) == 0)) {
/*
* @@status|timestamp|sender|origin|hostname|testname|expiretime|color|testflags|\
* prevcolor|changetime|ackexpiretime|ackmessage|disableexpiretime|disablemessage|\
* clienttstamp|flapping|classname|pagepaths
*/
int color = parse_color(metadata[7]);
switch (color) {
case COL_GREEN:
case COL_YELLOW:
case COL_RED:
case COL_BLUE: /* Blue is OK, because it only arrives here when an update is sent */
tstamp = atoi(metadata[1]);
sender = metadata[2];
hostname = metadata[4];
testname = metadata[5];
classname = (metadata[17] ? metadata[17] : "");
pagepaths = (metadata[18] ? metadata[18] : "");
ldef = find_xymon_rrd(testname, metadata[8]);
update_rrd(hostname, testname, restofmsg, tstamp, sender, ldef, classname, pagepaths);
break;
default:
/* Ignore reports with purple, blue or clear - they have no data we want. */
break;
}
}
else if ((metacount > 5) && (strncmp(metadata[0], "@@data", 6) == 0)) {
/* @@data|timestamp|sender|origin|hostname|testname|classname|pagepaths */
tstamp = atoi(metadata[1]);
sender = metadata[2];
hostname = metadata[4];
testname = metadata[5];
classname = (metadata[6] ? metadata[6] : "");
pagepaths = (metadata[7] ? metadata[7] : "");
ldef = find_xymon_rrd(testname, "");
update_rrd(hostname, testname, restofmsg, tstamp, sender, ldef, classname, pagepaths);
}
else if (strncmp(metadata[0], "@@shutdown", 10) == 0) {
running = 0;
continue;
}
else if (strncmp(metadata[0], "@@idle", 6) == 0) {
/* Ignored */
continue;
}
else if (strncmp(metadata[0], "@@logrotate", 11) == 0) {
char *fn = xgetenv("XYMONCHANNEL_LOGFILENAME");
if (fn && strlen(fn)) {
freopen(fn, "a", stdout);
freopen(fn, "a", stderr);
}
continue;
}
else if (strncmp(metadata[0], "@@reload", 8) == 0) {
reloadtime = 0;
}
else if ((metacount > 3) && (strncmp(metadata[0], "@@drophost", 10) == 0)) {
char hostdir[PATH_MAX];
hostname = metadata[3];
MEMDEFINE(hostdir);
sprintf(hostdir, "%s/%s", rrddir, hostname);
dropdirectory(hostdir, 1);
MEMUNDEFINE(hostdir);
}
else if ((metacount > 4) && (strncmp(metadata[0], "@@droptest", 10) == 0)) {
/*
* Not implemented. Mappings of testnames -> rrd files is
* too complex, so on the rare occasion that a single test
* is deleted, they will have to delete the rrd files themselves.
*/
}
else if ((metacount > 4) && (strncmp(metadata[0], "@@renamehost", 12) == 0)) {
char oldhostdir[PATH_MAX];
char newhostdir[PATH_MAX];
char *newhostname;
MEMDEFINE(oldhostdir);
MEMDEFINE(newhostdir);
hostname = metadata[3];
newhostname = metadata[4];
sprintf(oldhostdir, "%s/%s", rrddir, hostname);
sprintf(newhostdir, "%s/%s", rrddir, newhostname);
rename(oldhostdir, newhostdir);
if (net_worker_locatorbased()) locator_rename_host(hostname, newhostname, ST_RRD);
MEMUNDEFINE(newhostdir);
MEMUNDEFINE(oldhostdir);
}
else if ((metacount > 5) && (strncmp(metadata[0], "@@renametest", 12) == 0)) {
/* Not implemented. See "droptest". */
}
/*
* We fork a subprocess when processing drophost requests.
* Pickup any finished child processes to avoid zombies
*/
while (wait3(&childstat, WNOHANG, NULL) > 0) ;
}
/* Flush all cached updates to disk */
errprintf("Shutting down, flushing cached updates to disk\n");
rrdcacheflushall();
errprintf("Cache flush completed\n");
/* Close the external processor */
shutdown_extprocessor();
/* Close the control socket */
close(ctlsocket);
unlink(ctlsockaddr.sun_path);
return 0;
}
int main(int argc, char *argv[])
{
char *msg;
int running;
int argi, seq;
char newmsg[4096];
/* Handle program options. */
for (argi = 1; (argi < argc); argi++) {
if (strcmp(argv[argi], "--debug") == 0) {
debug = 1;
}
else if (strncmp(argv[argi], "--peer=", 7) == 0) {
char *ip = strchr(argv[argi], '=') + 1;
if (!peers) {
peercount = 1;
peers = (char **)calloc((peercount + 1), sizeof(char *));
peers[peercount-1] = strdup(ip);
peers[peercount] = NULL;
}
else {
peercount++;
peers = (char **)realloc(peers, (peercount + 1)*sizeof(char *));
peers[peercount-1] = strdup(ip);
peers[peercount] = NULL;
}
}
else if (strncmp(argv[argi], "--channel=", 10) == 0) {
channelname = strdup(strchr(argv[argi], '=') + 1);
}
}
if (!peers) {
errprintf("No peers specified, aborting\n");
return 1;
}
save_errbuf = 0;
signal(SIGCHLD, SIG_IGN);
running = 1;
while (running) {
char *eoln, *p;
char *metadata[MAX_META+1];
int metacount;
*newmsg = '\0';
msg = get_xymond_message(C_LAST, argv[0], &seq, NULL);
/* Split the message in the first line (with meta-data), and the rest. We're only interested in the first line. */
eoln = strchr(msg, '\n');
if (eoln) *eoln = '\0';
metacount = 0;
memset(&metadata, 0, sizeof(metadata));
p = gettok(msg, "|");
while (p && (metacount < MAX_META)) {
metadata[metacount++] = p;
p = gettok(NULL, "|");
}
metadata[metacount] = NULL;
if ((msg == NULL) || (strncmp(metadata[0], "@@shutdown", 10) == 0)) {
printf("Shutting down\n");
running = 0;
continue;
}
else if (strncmp(metadata[0], "@@logrotate", 11) == 0) {
char *fn = xgetenv("XYMONCHANNEL_LOGFILENAME");
if (fn && strlen(fn)) {
reopen_file(fn, "a", stdout);
reopen_file(fn, "a", stderr);
}
continue;
}
else if ((metacount > 3) && (strncmp(metadata[0], "@@drophost", 10) == 0)) {
snprintf(newmsg, sizeof(newmsg)-1, "drop %s", metadata[3]);
}
else if ((metacount > 4) && (strncmp(metadata[0], "@@droptest", 10) == 0)) {
snprintf(newmsg, sizeof(newmsg)-1, "drop %s %s", metadata[3], metadata[4]);
}
else if ((metacount > 4) && (strncmp(metadata[0], "@@renamehost", 12) == 0)) {
snprintf(newmsg, sizeof(newmsg)-1, "rename %s %s", metadata[3], metadata[4]);
}
else if ((metacount > 5) && (strncmp(metadata[0], "@@renametest", 12) == 0)) {
snprintf(newmsg, sizeof(newmsg)-1, "rename %s %s %s", metadata[3], metadata[4], metadata[5]);
}
else if ((metacount > 5) && (strncmp(metadata[0], "@@enadis", 8) == 0)) {
/* @@enadis|timestamp|sender|hostname|testname|expiretime|message */
if (strcmp(metadata[5], "0") == 0) {
snprintf(newmsg, sizeof(newmsg)-1, "enable %s.%s", metadata[3], metadata[4]);
}
else {
long int distime;
/* Disable until OK has time -1; normal disables has a count of minutes */
distime = (strcmp(metadata[5], "-1") == 0) ? -1 : ((atol(metadata[5]) - time(NULL)) / 60);
snprintf(newmsg, sizeof(newmsg)-1, "disable %s.%s %ld", metadata[3], metadata[4], distime);
if (metadata[6] && strlen(metadata[6])) {
nldecode(metadata[6]);
sprintf(newmsg + strlen(newmsg), " %s", metadata[6]);
}
dbgprintf("Disable: %s\n", newmsg);
}
}
if (strlen(newmsg) > 0) {
int i;
for (i = 0; (i < peercount); i++)
sendmessage(newmsg, peers[i], XYMON_TIMEOUT, NULL);
}
}
return 0;
}
int main(int argc, char *argv[])
{
char *msg;
int running;
int argi, seq;
int recentperiod = 3600;
int maxrecentcount = 5;
int logdirfull = 0;
int minlogspace = 5;
struct sigaction sa;
/* Handle program options. */
for (argi = 1; (argi < argc); argi++) {
if (argnmatch(argv[argi], "--logdir=")) {
clientlogdir = strchr(argv[argi], '=')+1;
}
else if (argnmatch(argv[argi], "--recent-period=")) {
char *p = strchr(argv[argi], '=');
recentperiod = 60*atoi(p+1);
}
else if (argnmatch(argv[argi], "--recent-count=")) {
char *p = strchr(argv[argi], '=');
maxrecentcount = atoi(p+1);
}
else if (argnmatch(argv[argi], "--minimum-free=")) {
minlogspace = atoi(strchr(argv[argi], '=')+1);
}
else if (strcmp(argv[argi], "--debug") == 0) {
/*
* A global "debug" variable is available. If
* it is set, then "dbgprintf()" outputs debug messages.
*/
debug = 1;
}
else if (net_worker_option(argv[argi])) {
/* Handled in the subroutine */
}
}
if (clientlogdir == NULL) clientlogdir = xgetenv("CLIENTLOGS");
if (clientlogdir == NULL) {
clientlogdir = (char *)malloc(strlen(xgetenv("XYMONVAR")) + 10);
sprintf(clientlogdir, "%s/hostdata", xgetenv("XYMONVAR"));
}
save_errbuf = 0;
/* Do the network stuff if needed */
net_worker_run(ST_HOSTDATA, LOC_STICKY, update_locator_hostdata);
setup_signalhandler("xymond_hostdata");
memset(&sa, 0, sizeof(sa));
sa.sa_handler = sig_handler;
signal(SIGCHLD, SIG_IGN);
sigaction(SIGHUP, &sa, NULL);
signal(SIGPIPE, SIG_DFL);
savetimes = xtreeNew(strcasecmp);
running = 1;
while (running) {
char *eoln, *restofmsg, *p;
char *metadata[MAX_META+1];
int metacount;
msg = get_xymond_message(C_CLICHG, "xymond_hostdata", &seq, NULL);
if (msg == NULL) {
/*
* get_xymond_message will return NULL if xymond_channel closes
* the input pipe. We should shutdown when that happens.
*/
running = 0;
continue;
}
if (nextfscheck < gettimer()) {
logdirfull = (chkfreespace(clientlogdir, minlogspace, minlogspace) != 0);
if (logdirfull) errprintf("Hostdata directory %s has less than %d%% free space - disabling save of data for 5 minutes\n", clientlogdir, minlogspace);
nextfscheck = gettimer() + 300;
}
/* Split the message in the first line (with meta-data), and the rest */
eoln = strchr(msg, '\n');
if (eoln) {
*eoln = '\0';
restofmsg = eoln+1;
}
else {
restofmsg = "";
}
metacount = 0;
memset(&metadata, 0, sizeof(metadata));
p = gettok(msg, "|");
while (p && (metacount < MAX_META)) {
metadata[metacount++] = p;
p = gettok(NULL, "|");
}
metadata[metacount] = NULL;
if (strncmp(metadata[0], "@@clichg", 8) == 0) {
xtreePos_t handle;
savetimes_t *itm;
int i, recentcount;
time_t now = gettimer();
char hostdir[PATH_MAX];
char fn[PATH_MAX];
FILE *fd;
/* metadata[3] is the hostname */
handle = xtreeFind(savetimes, metadata[3]);
if (handle != xtreeEnd(savetimes)) {
itm = (savetimes_t *)xtreeData(savetimes, handle);
}
else {
itm = (savetimes_t *)calloc(1, sizeof(savetimes_t));
itm->hostname = strdup(metadata[3]);
xtreeAdd(savetimes, itm->hostname, itm);
}
/* See how many times we've saved the hostdata recently (within the past 'recentperiod' seconds) */
for (i=0, recentcount=0; ((i < 12) && (itm->tstamp[i] > (now - recentperiod))); i++) recentcount++;
/* If it's been saved less than 'maxrecentcount' times, then save it. Otherwise just drop it */
if (!logdirfull && (recentcount < maxrecentcount)) {
int written, closestatus, ok = 1;
for (i = 10; (i > 0); i--) itm->tstamp[i+1] = itm->tstamp[i];
itm->tstamp[0] = now;
sprintf(hostdir, "%s/%s", clientlogdir, metadata[3]);
mkdir(hostdir, S_IRWXU|S_IRGRP|S_IXGRP|S_IROTH|S_IXOTH);
sprintf(fn, "%s/%s", hostdir, metadata[4]);
fd = fopen(fn, "w");
if (fd == NULL) {
errprintf("Cannot create file %s: %s\n", fn, strerror(errno));
continue;
}
written = fwrite(restofmsg, 1, strlen(restofmsg), fd);
if (written != strlen(restofmsg)) {
errprintf("Cannot write hostdata file %s: %s\n", fn, strerror(errno));
closestatus = fclose(fd); /* Ignore any close errors */
ok = 0;
}
else {
closestatus = fclose(fd);
if (closestatus != 0) {
errprintf("Cannot write hostdata file %s: %s\n", fn, strerror(errno));
ok = 0;
}
}
if (!ok) remove(fn);
}
}
/*
* A "shutdown" message is sent when the master daemon
* terminates. The child workers should shutdown also.
*/
else if (strncmp(metadata[0], "@@shutdown", 10) == 0) {
running = 0;
continue;
}
else if (strncmp(metadata[0], "@@idle", 6) == 0) {
/* Ignored */
continue;
}
/*
* A "logrotate" message is sent when the Xymon logs are
* rotated. The child workers must re-open their logfiles,
* typically stdin and stderr - the filename is always
* provided in the XYMONCHANNEL_LOGFILENAME environment.
*/
else if (strncmp(metadata[0], "@@logrotate", 11) == 0) {
char *fn = xgetenv("XYMONCHANNEL_LOGFILENAME");
if (fn && strlen(fn)) {
reopen_file(fn, "a", stdout);
reopen_file(fn, "a", stderr);
}
continue;
}
else if ((metacount > 3) && (strncmp(metadata[0], "@@drophost", 10) == 0)) {
/* @@drophost|timestamp|sender|hostname */
char hostdir[PATH_MAX];
snprintf(hostdir, sizeof(hostdir), "%s/%s", clientlogdir, basename(metadata[3]));
dropdirectory(hostdir, 1);
}
else if ((metacount > 4) && (strncmp(metadata[0], "@@renamehost", 12) == 0)) {
/* @@renamehost|timestamp|sender|hostname|newhostname */
char oldhostdir[PATH_MAX], newhostdir[PATH_MAX];
snprintf(oldhostdir, sizeof(oldhostdir), "%s/%s", clientlogdir, basename(metadata[3]));
snprintf(newhostdir, sizeof(newhostdir), "%s/%s", clientlogdir, basename(metadata[4]));
rename(oldhostdir, newhostdir);
if (net_worker_locatorbased()) locator_rename_host(metadata[3], metadata[4], ST_HOSTDATA);
}
else if (strncmp(metadata[0], "@@reload", 8) == 0) {
/* Do nothing */
}
}
return 0;
}
int main(int argc, char *argv[])
{
time_t starttime = gettimer();
char *histdir = NULL;
char *histlogdir = NULL;
char *msg;
int argi, seq;
int save_allevents = 1;
int save_hostevents = 1;
int save_statusevents = 1;
int save_histlogs = 1, defaultsaveop = 1;
FILE *alleventsfd = NULL;
int running = 1;
struct sigaction sa;
char newcol2[3];
char oldcol2[3];
char alleventsfn[PATH_MAX];
char pidfn[PATH_MAX];
int logdirfull = 0;
int minlogspace = 5;
MEMDEFINE(pidfn);
MEMDEFINE(alleventsfn);
MEMDEFINE(newcol2);
MEMDEFINE(oldcol2);
/* Dont save the error buffer */
save_errbuf = 0;
sprintf(pidfn, "%s/xymond_history.pid", xgetenv("XYMONSERVERLOGS"));
if (xgetenv("XYMONALLHISTLOG")) save_allevents = (strcmp(xgetenv("XYMONALLHISTLOG"), "TRUE") == 0);
if (xgetenv("XYMONHOSTHISTLOG")) save_hostevents = (strcmp(xgetenv("XYMONHOSTHISTLOG"), "TRUE") == 0);
if (xgetenv("SAVESTATUSLOG")) save_histlogs = (strncmp(xgetenv("SAVESTATUSLOG"), "FALSE", 5) != 0);
for (argi = 1; (argi < argc); argi++) {
if (argnmatch(argv[argi], "--histdir=")) {
histdir = strchr(argv[argi], '=')+1;
}
else if (argnmatch(argv[argi], "--histlogdir=")) {
histlogdir = strchr(argv[argi], '=')+1;
}
else if (argnmatch(argv[argi], "--pidfile=")) {
strcpy(pidfn, strchr(argv[argi], '=')+1);
}
else if (argnmatch(argv[argi], "--minimum-free=")) {
minlogspace = atoi(strchr(argv[argi], '=')+1);
}
else if (argnmatch(argv[argi], "--debug")) {
debug = 1;
}
}
if (xgetenv("XYMONHISTDIR") && (histdir == NULL)) {
histdir = strdup(xgetenv("XYMONHISTDIR"));
}
if (histdir == NULL) {
errprintf("No history directory given, aborting\n");
return 1;
}
if (save_histlogs && (histlogdir == NULL) && xgetenv("XYMONHISTLOGS")) {
histlogdir = strdup(xgetenv("XYMONHISTLOGS"));
}
if (save_histlogs && (histlogdir == NULL)) {
errprintf("No history-log directory given, aborting\n");
return 1;
}
columndefs = xtreeNew(strcmp);
{
char *defaultsave, *tok;
char *savelist;
columndef_t *newrec;
savelist = strdup(xgetenv("SAVESTATUSLOG"));
defaultsave = strtok(savelist, ",");
/*
* TRUE: Save everything by default; may list some that are not saved.
* ONLY: Save nothing by default; may list some that are saved.
* FALSE: Save nothing.
*/
defaultsaveop = (strcasecmp(defaultsave, "TRUE") == 0);
tok = strtok(NULL, ",");
while (tok) {
newrec = (columndef_t *)malloc(sizeof(columndef_t));
if (*tok == '!') {
newrec->saveit = 0;
newrec->name = strdup(tok+1);
}
else {
newrec->saveit = 1;
newrec->name = strdup(tok);
}
xtreeAdd(columndefs, newrec->name, newrec);
tok = strtok(NULL, ",");
}
xfree(savelist);
}
{
FILE *pidfd = fopen(pidfn, "w");
if (pidfd) {
fprintf(pidfd, "%lu\n", (unsigned long)getpid());
fclose(pidfd);
}
}
sprintf(alleventsfn, "%s/allevents", histdir);
if (save_allevents) {
alleventsfd = fopen(alleventsfn, "a");
if (alleventsfd == NULL) {
errprintf("Cannot open the all-events file '%s'\n", alleventsfn);
}
setvbuf(alleventsfd, (char *)NULL, _IOLBF, 0);
}
/* For picking up lost children */
setup_signalhandler("xymond_history");
memset(&sa, 0, sizeof(sa));
sa.sa_handler = sig_handler;
sigaction(SIGCHLD, &sa, NULL);
sigaction(SIGHUP, &sa, NULL);
signal(SIGPIPE, SIG_DFL);
while (running) {
char *metadata[20] = { NULL, };
int metacount;
char *p;
char *statusdata = "";
char *hostname, *hostnamecommas, *testname, *dismsg, *modifiers;
time_t tstamp, lastchg, disabletime, clienttstamp;
int tstamp_i, lastchg_i;
int newcolor, oldcolor;
int downtimeactive;
struct tm tstamptm;
int trend;
int childstat;
/* Pickup any finished child processes to avoid zombies */
while (wait3(&childstat, WNOHANG, NULL) > 0) ;
if (rotatefiles && alleventsfd) {
fclose(alleventsfd);
alleventsfd = fopen(alleventsfn, "a");
if (alleventsfd == NULL) {
errprintf("Cannot re-open the all-events file '%s'\n", alleventsfn);
}
else {
setvbuf(alleventsfd, (char *)NULL, _IOLBF, 0);
}
}
msg = get_xymond_message(C_STACHG, "xymond_history", &seq, NULL);
if (msg == NULL) {
running = 0;
continue;
}
if (nextfscheck < gettimer()) {
logdirfull = (chkfreespace(histlogdir, minlogspace, minlogspace) != 0);
if (logdirfull) errprintf("Historylog directory %s has less than %d%% free space - disabling save of data for 5 minutes\n", histlogdir, minlogspace);
nextfscheck = gettimer() + 300;
}
p = strchr(msg, '\n');
if (p) {
*p = '\0';
statusdata = msg_data(p+1, 0);
}
metacount = 0;
memset(&metadata, 0, sizeof(metadata));
p = gettok(msg, "|");
while (p && (metacount < 20)) {
metadata[metacount++] = p;
p = gettok(NULL, "|");
}
if ((metacount > 9) && (strncmp(metadata[0], "@@stachg", 8) == 0)) {
xtreePos_t handle;
columndef_t *saveit = NULL;
/* @@stachg#seq|timestamp|sender|origin|hostname|testname|expiretime|color|prevcolor|changetime|disabletime|disablemsg|downtimeactive|clienttstamp|modifiers */
sscanf(metadata[1], "%d.%*d", &tstamp_i); tstamp = tstamp_i;
hostname = metadata[4];
testname = metadata[5];
newcolor = parse_color(metadata[7]);
oldcolor = parse_color(metadata[8]);
lastchg = atoi(metadata[9]);
disabletime = atoi(metadata[10]);
dismsg = metadata[11];
downtimeactive = (atoi(metadata[12]) > 0);
clienttstamp = atoi(metadata[13]);
modifiers = metadata[14];
if (newcolor == -1) {
errprintf("Bad message: newcolor is unknown '%s'\n", metadata[7]);
continue;
}
p = hostnamecommas = strdup(hostname); while ((p = strchr(p, '.')) != NULL) *p = ',';
handle = xtreeFind(columndefs, testname);
if (handle == xtreeEnd(columndefs)) {
saveit = (columndef_t *)malloc(sizeof(columndef_t));
saveit->name = strdup(testname);
saveit->saveit = defaultsaveop;
xtreeAdd(columndefs, saveit->name, saveit);
}
else {
saveit = (columndef_t *) xtreeData(columndefs, handle);
}
if (save_statusevents) {
char statuslogfn[PATH_MAX];
int logexists;
FILE *statuslogfd;
char oldcol[100];
char timestamp[40];
struct stat st;
MEMDEFINE(statuslogfn);
MEMDEFINE(oldcol);
MEMDEFINE(timestamp);
sprintf(statuslogfn, "%s/%s.%s", histdir, hostnamecommas, testname);
stat(statuslogfn, &st);
statuslogfd = fopen(statuslogfn, "r+");
logexists = (statuslogfd != NULL);
*oldcol = '\0';
if (logexists) {
/*
* There is a fair chance xymond has not been
* running all the time while this system was monitored.
* So get the time of the latest status change from the file,
* instead of relying on the "lastchange" value we get
* from xymond. This is also needed when migrating from
* standard bbd to xymond.
*/
off_t pos = -1;
char l[1024];
int gotit;
MEMDEFINE(l);
fseeko(statuslogfd, 0, SEEK_END);
if (ftello(statuslogfd) > 512) {
/* Go back 512 from EOF, and skip to start of a line */
fseeko(statuslogfd, -512, SEEK_END);
gotit = (fgets(l, sizeof(l)-1, statuslogfd) == NULL);
}
else {
/* Read from beginning of file */
fseeko(statuslogfd, 0, SEEK_SET);
gotit = 0;
}
while (!gotit) {
off_t tmppos = ftello(statuslogfd);
int dur_i;
if (fgets(l, sizeof(l)-1, statuslogfd)) {
/* Sun Oct 10 06:49:42 2004 red 1097383782 602 */
if ((strlen(l) > 24) &&
(sscanf(l+24, " %s %d %d", oldcol, &lastchg_i, &dur_i) == 2) &&
(parse_color(oldcol) != -1)) {
/*
* Record the start location of the line
*/
pos = tmppos;
lastchg = lastchg_i;
}
}
else {
gotit = 1;
}
}
if (pos == -1) {
/*
* Couldnt find anything in the log.
* Take lastchg from the timestamp of the logfile,
* and just append the data.
*/
lastchg = st.st_mtime;
fseeko(statuslogfd, 0, SEEK_END);
}
else {
/*
* lastchg was updated above.
* Seek to where the last line starts.
*/
fseeko(statuslogfd, pos, SEEK_SET);
}
MEMUNDEFINE(l);
}
else {
/*
* Logfile does not exist.
*/
lastchg = tstamp;
statuslogfd = fopen(statuslogfn, "a");
if (statuslogfd == NULL) {
errprintf("Cannot open status historyfile '%s' : %s\n",
statuslogfn, strerror(errno));
}
}
if (strcmp(oldcol, colorname(newcolor)) == 0) {
/* We wont update history unless the color did change. */
if ((gettimer() - starttime) > 300) {
errprintf("Will not update %s - color unchanged (%s)\n",
statuslogfn, oldcol);
}
if (hostnamecommas) xfree(hostnamecommas);
if (statuslogfd) fclose(statuslogfd);
MEMUNDEFINE(statuslogfn);
MEMUNDEFINE(oldcol);
MEMUNDEFINE(timestamp);
continue;
}
if (statuslogfd) {
if (logexists) {
struct tm oldtm;
/* Re-print the old record, now with the final duration */
memcpy(&oldtm, localtime(&lastchg), sizeof(oldtm));
strftime(timestamp, sizeof(timestamp), "%a %b %e %H:%M:%S %Y", &oldtm);
fprintf(statuslogfd, "%s %s %d %d\n",
timestamp, oldcol, (int)lastchg, (int)(tstamp - lastchg));
}
/* And the new record. */
memcpy(&tstamptm, localtime(&tstamp), sizeof(tstamptm));
strftime(timestamp, sizeof(timestamp), "%a %b %e %H:%M:%S %Y", &tstamptm);
fprintf(statuslogfd, "%s %s %d", timestamp, colorname(newcolor), (int)tstamp);
fclose(statuslogfd);
}
MEMUNDEFINE(statuslogfn);
MEMUNDEFINE(oldcol);
MEMUNDEFINE(timestamp);
}
if (save_histlogs && saveit->saveit && !logdirfull) {
char *hostdash;
char fname[PATH_MAX];
FILE *histlogfd;
MEMDEFINE(fname);
p = hostdash = strdup(hostname); while ((p = strchr(p, '.')) != NULL) *p = '_';
sprintf(fname, "%s/%s", histlogdir, hostdash);
mkdir(fname, S_IRWXU|S_IRGRP|S_IXGRP|S_IROTH|S_IXOTH);
p = fname + sprintf(fname, "%s/%s/%s", histlogdir, hostdash, testname);
mkdir(fname, S_IRWXU|S_IRGRP|S_IXGRP|S_IROTH|S_IXOTH);
p += sprintf(p, "/%s", histlogtime(tstamp));
histlogfd = fopen(fname, "w");
if (histlogfd) {
/*
* When a host gets disabled or goes purple, the status
* message data is not changed - so it will include a
* wrong color as the first word of the message.
* Therefore we need to fixup this so it matches the
* newcolor value.
*/
int txtcolor = parse_color(statusdata);
char *origstatus = statusdata;
char *eoln, *restofdata;
int written, closestatus, ok = 1;
if (txtcolor != -1) {
fprintf(histlogfd, "%s", colorname(newcolor));
statusdata += strlen(colorname(txtcolor));
}
if (dismsg && *dismsg) nldecode(dismsg);
if (disabletime > 0) {
fprintf(histlogfd, " Disabled until %s\n%s\n\n",
ctime(&disabletime), (dismsg ? dismsg : ""));
fprintf(histlogfd, "Status message when disabled follows:\n\n");
statusdata = origstatus;
}
else if (dismsg && *dismsg) {
fprintf(histlogfd, " Planned downtime: %s\n\n", dismsg);
fprintf(histlogfd, "Original status message follows:\n\n");
statusdata = origstatus;
}
restofdata = statusdata;
if (modifiers && *modifiers) {
char *modtxt;
/* We must finish writing the first line before putting in the modifiers */
eoln = strchr(restofdata, '\n');
if (eoln) {
restofdata = eoln+1;
*eoln = '\0';
fprintf(histlogfd, "%s\n", statusdata);
}
nldecode(modifiers);
modtxt = strtok(modifiers, "\n");
while (modtxt) {
fprintf(histlogfd, "%s\n", modtxt);
modtxt = strtok(NULL, "\n");
}
}
written = fwrite(restofdata, 1, strlen(restofdata), histlogfd);
if (written != strlen(restofdata)) {
ok = 0;
errprintf("Error writing to file %s: %s\n", fname, strerror(errno));
closestatus = fclose(histlogfd); /* Ignore any errors on close */
}
else {
fprintf(histlogfd, "Status unchanged in 0.00 minutes\n");
fprintf(histlogfd, "Message received from %s\n", metadata[2]);
if (clienttstamp) fprintf(histlogfd, "Client data ID %d\n", (int) clienttstamp);
closestatus = fclose(histlogfd);
if (closestatus != 0) {
ok = 0;
errprintf("Error writing to file %s: %s\n", fname, strerror(errno));
}
}
if (!ok) remove(fname);
}
else {
errprintf("Cannot create histlog file '%s' : %s\n", fname, strerror(errno));
}
xfree(hostdash);
MEMUNDEFINE(fname);
}
strncpy(oldcol2, ((oldcolor >= 0) ? colorname(oldcolor) : "-"), 2);
strncpy(newcol2, colorname(newcolor), 2);
newcol2[2] = oldcol2[2] = '\0';
if (oldcolor == -1) trend = -1; /* we dont know how bad it was */
else if (newcolor > oldcolor) trend = 2; /* It's getting worse */
else if (newcolor < oldcolor) trend = 1; /* It's getting better */
else trend = 0; /* Shouldn't happen ... */
if (save_hostevents) {
char hostlogfn[PATH_MAX];
FILE *hostlogfd;
MEMDEFINE(hostlogfn);
sprintf(hostlogfn, "%s/%s", histdir, hostname);
hostlogfd = fopen(hostlogfn, "a");
if (hostlogfd) {
fprintf(hostlogfd, "%s %d %d %d %s %s %d\n",
testname, (int)tstamp, (int)lastchg, (int)(tstamp - lastchg),
newcol2, oldcol2, trend);
fclose(hostlogfd);
}
else {
errprintf("Cannot open host logfile '%s' : %s\n", hostlogfn, strerror(errno));
}
MEMUNDEFINE(hostlogfn);
}
if (save_allevents) {
fprintf(alleventsfd, "%s %s %d %d %d %s %s %d\n",
hostname, testname, (int)tstamp, (int)lastchg, (int)(tstamp - lastchg),
newcol2, oldcol2, trend);
fflush(alleventsfd);
}
xfree(hostnamecommas);
}
else if ((metacount > 3) && ((strncmp(metadata[0], "@@drophost", 10) == 0))) {
/* @@drophost|timestamp|sender|hostname */
hostname = metadata[3];
if (save_histlogs) {
char *hostdash;
char testdir[PATH_MAX];
MEMDEFINE(testdir);
/* Remove all directories below the host-specific histlog dir */
p = hostdash = strdup(hostname); while ((p = strchr(p, '.')) != NULL) *p = '_';
sprintf(testdir, "%s/%s", histlogdir, hostdash);
dropdirectory(testdir, 1);
xfree(hostdash);
MEMUNDEFINE(testdir);
}
if (save_hostevents) {
char hostlogfn[PATH_MAX];
struct stat st;
MEMDEFINE(hostlogfn);
sprintf(hostlogfn, "%s/%s", histdir, hostname);
if ((stat(hostlogfn, &st) == 0) && S_ISREG(st.st_mode)) {
unlink(hostlogfn);
}
MEMUNDEFINE(hostlogfn);
}
if (save_statusevents) {
DIR *dirfd;
struct dirent *de;
char *hostlead;
char statuslogfn[PATH_MAX];
struct stat st;
MEMDEFINE(statuslogfn);
/* Remove $XYMONVAR/hist/host,name.* */
p = hostnamecommas = strdup(hostname); while ((p = strchr(p, '.')) != NULL) *p = ',';
hostlead = malloc(strlen(hostname) + 2);
strcpy(hostlead, hostnamecommas); strcat(hostlead, ".");
dirfd = opendir(histdir);
if (dirfd) {
while ((de = readdir(dirfd)) != NULL) {
if (strncmp(de->d_name, hostlead, strlen(hostlead)) == 0) {
sprintf(statuslogfn, "%s/%s", histdir, de->d_name);
if ((stat(statuslogfn, &st) == 0) && S_ISREG(st.st_mode)) {
unlink(statuslogfn);
}
}
}
closedir(dirfd);
}
xfree(hostlead);
xfree(hostnamecommas);
MEMUNDEFINE(statuslogfn);
}
}
else if ((metacount > 4) && ((strncmp(metadata[0], "@@droptest", 10) == 0))) {
/* @@droptest|timestamp|sender|hostname|testname */
hostname = metadata[3];
testname = metadata[4];
if (save_histlogs) {
char *hostdash;
char testdir[PATH_MAX];
MEMDEFINE(testdir);
p = hostdash = strdup(hostname); while ((p = strchr(p, '.')) != NULL) *p = '_';
sprintf(testdir, "%s/%s/%s", histlogdir, hostdash, testname);
dropdirectory(testdir, 1);
xfree(hostdash);
MEMUNDEFINE(testdir);
}
if (save_statusevents) {
char *hostnamecommas;
char statuslogfn[PATH_MAX];
struct stat st;
MEMDEFINE(statuslogfn);
p = hostnamecommas = strdup(hostname); while ((p = strchr(p, '.')) != NULL) *p = ',';
sprintf(statuslogfn, "%s/%s.%s", histdir, hostnamecommas, testname);
if ((stat(statuslogfn, &st) == 0) && S_ISREG(st.st_mode)) unlink(statuslogfn);
xfree(hostnamecommas);
MEMUNDEFINE(statuslogfn);
}
}
else if ((metacount > 4) && ((strncmp(metadata[0], "@@renamehost", 12) == 0))) {
/* @@renamehost|timestamp|sender|hostname|newhostname */
char *newhostname;
hostname = metadata[3];
newhostname = metadata[4];
if (save_histlogs) {
char *hostdash;
char *newhostdash;
char olddir[PATH_MAX];
char newdir[PATH_MAX];
MEMDEFINE(olddir); MEMDEFINE(newdir);
p = hostdash = strdup(hostname); while ((p = strchr(p, '.')) != NULL) *p = '_';
p = newhostdash = strdup(newhostname); while ((p = strchr(p, '.')) != NULL) *p = '_';
sprintf(olddir, "%s/%s", histlogdir, hostdash);
sprintf(newdir, "%s/%s", histlogdir, newhostdash);
rename(olddir, newdir);
xfree(newhostdash);
xfree(hostdash);
MEMUNDEFINE(newdir); MEMUNDEFINE(olddir);
}
if (save_hostevents) {
char hostlogfn[PATH_MAX];
char newhostlogfn[PATH_MAX];
MEMDEFINE(hostlogfn); MEMDEFINE(newhostlogfn);
sprintf(hostlogfn, "%s/%s", histdir, hostname);
sprintf(newhostlogfn, "%s/%s", histdir, newhostname);
rename(hostlogfn, newhostlogfn);
MEMUNDEFINE(hostlogfn); MEMUNDEFINE(newhostlogfn);
}
if (save_statusevents) {
DIR *dirfd;
struct dirent *de;
char *hostlead;
char *newhostnamecommas;
char statuslogfn[PATH_MAX];
char newlogfn[PATH_MAX];
MEMDEFINE(statuslogfn); MEMDEFINE(newlogfn);
p = hostnamecommas = strdup(hostname); while ((p = strchr(p, '.')) != NULL) *p = ',';
hostlead = malloc(strlen(hostname) + 2);
strcpy(hostlead, hostnamecommas); strcat(hostlead, ".");
p = newhostnamecommas = strdup(newhostname); while ((p = strchr(p, '.')) != NULL) *p = ',';
dirfd = opendir(histdir);
if (dirfd) {
while ((de = readdir(dirfd)) != NULL) {
if (strncmp(de->d_name, hostlead, strlen(hostlead)) == 0) {
char *testname = strchr(de->d_name, '.');
sprintf(statuslogfn, "%s/%s", histdir, de->d_name);
sprintf(newlogfn, "%s/%s%s", histdir, newhostnamecommas, testname);
rename(statuslogfn, newlogfn);
}
}
closedir(dirfd);
}
xfree(newhostnamecommas);
xfree(hostlead);
xfree(hostnamecommas);
MEMUNDEFINE(statuslogfn); MEMUNDEFINE(newlogfn);
}
}
else if ((metacount > 5) && (strncmp(metadata[0], "@@renametest", 12) == 0)) {
/* @@renametest|timestamp|sender|hostname|oldtestname|newtestname */
char *newtestname;
hostname = metadata[3];
testname = metadata[4];
newtestname = metadata[5];
if (save_histlogs) {
char *hostdash;
char olddir[PATH_MAX];
char newdir[PATH_MAX];
MEMDEFINE(olddir); MEMDEFINE(newdir);
p = hostdash = strdup(hostname); while ((p = strchr(p, '.')) != NULL) *p = '_';
sprintf(olddir, "%s/%s/%s", histlogdir, hostdash, testname);
sprintf(newdir, "%s/%s/%s", histlogdir, hostdash, newtestname);
rename(olddir, newdir);
xfree(hostdash);
MEMUNDEFINE(newdir); MEMUNDEFINE(olddir);
}
if (save_statusevents) {
char *hostnamecommas;
char statuslogfn[PATH_MAX];
char newstatuslogfn[PATH_MAX];
MEMDEFINE(statuslogfn); MEMDEFINE(newstatuslogfn);
p = hostnamecommas = strdup(hostname); while ((p = strchr(p, '.')) != NULL) *p = ',';
sprintf(statuslogfn, "%s/%s.%s", histdir, hostnamecommas, testname);
sprintf(newstatuslogfn, "%s/%s.%s", histdir, hostnamecommas, newtestname);
rename(statuslogfn, newstatuslogfn);
xfree(hostnamecommas);
MEMUNDEFINE(newstatuslogfn); MEMUNDEFINE(statuslogfn);
}
}
else if (strncmp(metadata[0], "@@idle", 6) == 0) {
/* Nothing */
}
else if (strncmp(metadata[0], "@@shutdown", 10) == 0) {
running = 0;
}
else if (strncmp(metadata[0], "@@logrotate", 11) == 0) {
char *fn = xgetenv("XYMONCHANNEL_LOGFILENAME");
if (fn && strlen(fn)) {
reopen_file(fn, "a", stdout);
reopen_file(fn, "a", stderr);
}
continue;
}
else if (strncmp(metadata[0], "@@reload", 8) == 0) {
/* Do nothing */
}
}
MEMUNDEFINE(newcol2);
MEMUNDEFINE(oldcol2);
MEMUNDEFINE(alleventsfn);
MEMUNDEFINE(pidfn);
fclose(alleventsfd);
unlink(pidfn);
return 0;
}
int main(int argc, char *argv[])
{
char *msg;
int running;
int argi, seq;
struct timespec *timeout = NULL;
pcre *hostexp = NULL;
pcre *exhostexp = NULL;
pcre *testexp = NULL;
pcre *extestexp = NULL;
pcre *colorexp = NULL;
const char *errmsg = NULL;
int errofs = 0;
FILE *logfd = stdout;
int batchtimeout = 30;
char *batchcmd = NULL;
strbuffer_t *batchbuf = NULL;
time_t lastmsgtime = 0;
int hostnameitem = 4, testnameitem = 5, coloritem = 7;
/* Handle program options. */
for (argi = 1; (argi < argc); argi++) {
if (strcmp(argv[argi], "--debug") == 0) {
/*
* A global "debug" variable is available. If
* it is set, then "dbgprintf()" outputs debug messages.
*/
debug = 1;
}
else if (strncmp(argv[argi], "--timeout=", 10) == 0) {
/*
* You can have a timeout when waiting for new
* messages. If it happens, you will get a "@@idle\n"
* message with sequence number 0.
* If you dont want a timeout, just pass a NULL for the timeout parameter.
*/
timeout = (struct timespec *)(malloc(sizeof(struct timespec)));
timeout->tv_sec = (atoi(argv[argi]+10));
timeout->tv_nsec = 0;
}
else if (strcmp(argv[argi], "--client") == 0) {
hostnameitem = 3;
testnameitem = 4;
errprintf("Expecting to be fed from 'client' channel\n");
}
else if (argnmatch(argv[argi], "--hosts=")) {
char *exp = strchr(argv[argi], '=') + 1;
hostexp = pcre_compile(exp, PCRE_CASELESS, &errmsg, &errofs, NULL);
if (hostexp == NULL) printf("Invalid expression '%s'\n", exp);
}
else if (argnmatch(argv[argi], "--exhosts=")) {
char *exp = strchr(argv[argi], '=') + 1;
exhostexp = pcre_compile(exp, PCRE_CASELESS, &errmsg, &errofs, NULL);
if (exhostexp == NULL) printf("Invalid expression '%s'\n", exp);
}
else if (argnmatch(argv[argi], "--tests=")) {
char *exp = strchr(argv[argi], '=') + 1;
testexp = pcre_compile(exp, PCRE_CASELESS, &errmsg, &errofs, NULL);
if (testexp == NULL) printf("Invalid expression '%s'\n", exp);
}
else if (argnmatch(argv[argi], "--extests=")) {
char *exp = strchr(argv[argi], '=') + 1;
extestexp = pcre_compile(exp, PCRE_CASELESS, &errmsg, &errofs, NULL);
if (extestexp == NULL) printf("Invalid expression '%s'\n", exp);
}
else if (argnmatch(argv[argi], "--colors=")) {
char *exp = strchr(argv[argi], '=') + 1;
colorexp = pcre_compile(exp, PCRE_CASELESS, &errmsg, &errofs, NULL);
if (colorexp == NULL) printf("Invalid expression '%s'\n", exp);
}
else if (argnmatch(argv[argi], "--outfile=")) {
char *fn = strchr(argv[argi], '=') + 1;
logfd = fopen(fn, "a");
if (logfd == NULL) {
printf("Cannot open logfile %s: %s\n", fn, strerror(errno));
logfd = stdout;
}
}
else if (argnmatch(argv[argi], "--batch-timeout=")) {
char *p = strchr(argv[argi], '=');
batchtimeout = atoi(p+1);
timeout = (struct timespec *)(malloc(sizeof(struct timespec)));
timeout->tv_sec = batchtimeout;
timeout->tv_nsec = 0;
}
else if (argnmatch(argv[argi], "--batch-command=")) {
char *p = strchr(argv[argi], '=');
batchcmd = strdup(p+1);
batchbuf = newstrbuffer(0);
}
else {
printf("Unknown option %s\n", argv[argi]);
printf("Usage: %s [--hosts=EXP] [--tests=EXP] [--exhosts=EXP] [--extests=EXP] [--color=EXP] [--outfile=FILENAME] [--batch-timeout=N] [--batch-command=COMMAND]\n", argv[0]);
return 0;
}
}
signal(SIGCHLD, SIG_IGN);
running = 1;
while (running) {
char *eoln, *restofmsg, *p;
char *metadata[MAX_META+1];
int metacount;
msg = get_xymond_message(C_LAST, argv[0], &seq, timeout);
if (msg == NULL) {
/*
* get_xymond_message will return NULL if xymond_channel closes
* the input pipe. We should shutdown when that happens.
*/
running = 0;
continue;
}
/*
* Now we have a message. So do something with it.
*
* The first line of the message is always a '|' separated
* list of meta-data about the message. After the first
* line, the content varies by channel.
*/
/* Split the message in the first line (with meta-data), and the rest */
eoln = strchr(msg, '\n');
if (eoln) {
*eoln = '\0';
restofmsg = eoln+1;
}
else {
restofmsg = "";
}
/*
* Now parse the meta-data into elements.
* We use our own "gettok()" routine which works
* like strtok(), but can handle empty elements.
*/
metacount = 0;
memset(&metadata, 0, sizeof(metadata));
p = gettok(msg, "|");
while (p && (metacount < MAX_META)) {
metadata[metacount++] = p;
p = gettok(NULL, "|");
}
metadata[metacount] = NULL;
/*
* A "shutdown" message is sent when the master daemon
* terminates. The child workers should shutdown also.
*/
if (strncmp(metadata[0], "@@shutdown", 10) == 0) {
printf("Shutting down\n");
running = 0;
continue;
}
/*
* A "logrotate" message is sent when the Xymon logs are
* rotated. The child workers must re-open their logfiles,
* typically stdin and stderr - the filename is always
* provided in the XYMONDHANNEL_LOGFILENAME environment.
*/
else if (strncmp(metadata[0], "@@logrotate", 11) == 0) {
char *fn = xgetenv("XYMONCHANNEL_LOGFILENAME");
if (fn && strlen(fn)) {
freopen(fn, "a", stdout);
freopen(fn, "a", stderr);
}
continue;
}
/*
* A "reload" means the hosts.cfg file has changed.
*/
else if (strncmp(metadata[0], "@@reload", 8) == 0) {
/* Nothing ... right now */
}
/*
* An "idle" message appears when get_xymond_message()
* exceeds the timeout setting (ie. you passed a timeout
* value). This allows your worker module to perform
* some internal processing even though no messages arrive.
*/
else if (strncmp(metadata[0], "@@idle", 6) == 0) {
dbgprintf("Got an 'idle' message\n");
}
/*
* The "drophost", "droptest", "renamehost" and "renametst"
* indicates that a host/test was deleted or renamed. If the
* worker module maintains some internal storage (in memory
* or persistent file-storage), it should act on these
* messages to maintain data consistency.
*/
else if ((metacount > 3) && (strncmp(metadata[0], "@@drophost", 10) == 0)) {
dbgprintf("Got a 'drophost' message for host '%s'\n", metadata[3]);
}
else if ((metacount > 3) && (strncmp(metadata[0], "@@dropstate", 11) == 0)) {
dbgprintf("Got a 'dropstate' message for host '%s'\n", metadata[3]);
}
else if ((metacount > 4) && (strncmp(metadata[0], "@@droptest", 10) == 0)) {
dbgprintf("Got a 'droptest' message for host '%s' test '%s'\n", metadata[3], metadata[4]);
}
else if ((metacount > 4) && (strncmp(metadata[0], "@@renamehost", 12) == 0)) {
dbgprintf("Got a 'renamehost' message for host '%s' -> '%s'\n", metadata[3], metadata[4]);
}
else if ((metacount > 5) && (strncmp(metadata[0], "@@renametest", 12) == 0)) {
dbgprintf("Got a 'renametest' message for host '%s' test '%s' -> '%s'\n",
metadata[3], metadata[4], metadata[5]);
}
/*
* Process this message.
*/
else {
int ovector[30];
int match, i;
char *hostname = metadata[hostnameitem];
char *testname = metadata[testnameitem];
char *color = metadata[coloritem];
/* See if we should handle the batched messages we've got */
if (batchcmd && ((lastmsgtime + batchtimeout) < gettimer()) && (STRBUFLEN(batchbuf) > 0)) {
pid_t childpid = fork();
int childres = 0;
if (childpid < 0) {
/* Fork failed! */
errprintf("Fork failed: %s\n", strerror(errno));
}
else if (childpid == 0) {
/* Child */
FILE *cmdpipe = popen(batchcmd, "w");
if (cmdpipe) {
/* Write the data to the batch command pipe */
int n, bytesleft = STRBUFLEN(batchbuf);
char *outp = STRBUF(batchbuf);
while (bytesleft) {
n = fwrite(outp, 1, bytesleft, cmdpipe);
if (n >= 0) {
bytesleft -= n;
outp += n;
}
else {
errprintf("Error while writing data to batch command\n");
bytesleft = 0;
}
}
childres = pclose(cmdpipe);
}
else {
errprintf("Could not open pipe to batch command '%s'\n", batchcmd);
childres = 127;
}
exit(childres);
}
else if (childpid > 0) {
/* Parent continues */
}
clearstrbuffer(batchbuf);
}
if (hostexp) {
match = (pcre_exec(hostexp, NULL, hostname, strlen(hostname), 0, 0, ovector, (sizeof(ovector)/sizeof(int))) >= 0);
if (!match) continue;
}
if (exhostexp) {
match = (pcre_exec(exhostexp, NULL, hostname, strlen(hostname), 0, 0, ovector, (sizeof(ovector)/sizeof(int))) >= 0);
if (match) continue;
}
if (testexp) {
match = (pcre_exec(testexp, NULL, testname, strlen(testname), 0, 0, ovector, (sizeof(ovector)/sizeof(int))) >= 0);
if (!match) continue;
}
if (exhostexp) {
match = (pcre_exec(extestexp, NULL, testname, strlen(testname), 0, 0, ovector, (sizeof(ovector)/sizeof(int))) >= 0);
if (match) continue;
}
if (colorexp) {
match = (pcre_exec(colorexp, NULL, color, strlen(color), 0, 0, ovector, (sizeof(ovector)/sizeof(int))) >= 0);
if (!match) continue;
}
lastmsgtime = gettimer();
if (batchcmd) {
addtobuffer(batchbuf, "## ");
for (i=0; (i < metacount); i++) {
addtobuffer(batchbuf, metadata[i]);
addtobuffer(batchbuf, " ");
}
addtobuffer(batchbuf, "\n");
addtobuffer(batchbuf, restofmsg);
addtobuffer(batchbuf, "\n");
}
else {
fprintf(logfd, "## ");
for (i=0; (i < metacount); i++) fprintf(logfd, "%s ", metadata[i]);
fprintf(logfd, "\n");
fprintf(logfd, "%s\n", restofmsg);
}
}
}
return 0;
}
int main(int argc, char *argv[])
{
char *filedir = NULL;
char *htmldir = NULL;
char *htmlextension = "html";
char *onlytests = NULL;
char *msg;
enum role_t role = ROLE_STATUS;
enum msgchannels_t chnid = C_STATUS;
int argi;
int seq;
int running = 1;
/* Dont save the error buffer */
save_errbuf = 0;
for (argi = 1; (argi < argc); argi++) {
if (strcmp(argv[argi], "--status") == 0) {
role = ROLE_STATUS;
chnid = C_STATUS;
if (!filedir) filedir = xgetenv("XYMONRAWSTATUSDIR");
}
else if (strcmp(argv[argi], "--html") == 0) {
role = ROLE_STATUS;
chnid = C_STATUS;
if (!htmldir) htmldir = xgetenv("XYMONHTMLSTATUSDIR");
}
else if (strcmp(argv[argi], "--data") == 0) {
role = ROLE_DATA;
chnid = C_DATA;
if (!filedir) filedir = xgetenv("XYMONDATADIR");
}
else if (strcmp(argv[argi], "--notes") == 0) {
role = ROLE_NOTES;
chnid = C_NOTES;
if (!filedir) filedir = xgetenv("XYMONNOTESDIR");
}
else if (strcmp(argv[argi], "--enadis") == 0) {
role = ROLE_ENADIS;
chnid = C_ENADIS;
if (!filedir) filedir = xgetenv("XYMONDISABLEDDIR");
}
else if (strcmp(argv[argi], "--debug") == 0) {
debug = 1;
}
else if (argnmatch(argv[argi], "--dir=")) {
filedir = strchr(argv[argi], '=')+1;
}
else if (argnmatch(argv[argi], "--htmldir=")) {
htmldir = strchr(argv[argi], '=')+1;
}
else if (argnmatch(argv[argi], "--htmlext=")) {
htmlextension = strchr(argv[argi], '=')+1;
}
else if (argnmatch(argv[argi], "--only=")) {
char *p = strchr(argv[argi], '=') + 1;
onlytests = (char *)malloc(3 + strlen(p));
sprintf(onlytests, ",%s,", p);
}
else if (argnmatch(argv[argi], "--multigraphs=")) {
char *p = strchr(argv[argi], '=');
multigraphs = (char *)malloc(strlen(p+1) + 3);
sprintf(multigraphs, ",%s,", p+1);
}
else if (argnmatch(argv[argi], "--locator=")) {
char *p = strchr(argv[argi], '=');
locator_init(p+1);
locatorbased = 1;
}
}
if (filedir == NULL) {
errprintf("No directory given, aborting\n");
return 1;
}
/* For picking up lost children */
setup_signalhandler("xymond_filestore");
signal(SIGPIPE, SIG_DFL);
if (onlytests) dbgprintf("Storing tests '%s' only\n", onlytests);
else dbgprintf("Storing all tests\n");
while (running) {
char *metadata[20] = { NULL, };
char *statusdata = "";
char *p;
int metacount;
char *hostname, *testname;
time_t expiretime = 0;
char logfn[PATH_MAX];
MEMDEFINE(logfn);
msg = get_xymond_message(chnid, "filestore", &seq, NULL);
if (msg == NULL) {
running = 0;
MEMUNDEFINE(logfn);
continue;
}
p = strchr(msg, '\n');
if (p) {
*p = '\0';
statusdata = p+1;
}
metacount = 0;
memset(&metadata, 0, sizeof(metadata));
p = gettok(msg, "|");
while (p && (metacount < 20)) {
metadata[metacount++] = p;
p = gettok(NULL, "|");
}
if ((role == ROLE_STATUS) && (metacount >= 14) && (strncmp(metadata[0], "@@status", 8) == 0)) {
/* @@status|timestamp|sender|origin|hostname|testname|expiretime|color|testflags|prevcolor|changetime|ackexpiretime|ackmessage|disableexpiretime|disablemessage|clientmsgtstamp|flapping */
int ltime, flapping = 0;
time_t logtime = 0, timesincechange = 0, acktime = 0, disabletime = 0;
hostname = metadata[4];
testname = metadata[5];
if (!wantedtest(onlytests, testname)) {
dbgprintf("Status dropped - not wanted\n");
MEMUNDEFINE(logfn);
continue;
}
sprintf(logfn, "%s/%s.%s", filedir, commafy(hostname), testname);
expiretime = atoi(metadata[6]);
statusdata = msg_data(statusdata);
sscanf(metadata[1], "%d.%*d", <ime); logtime = ltime;
timesincechange = logtime - atoi(metadata[10]);
update_file(logfn, "w", statusdata, expiretime, metadata[2], timesincechange, seq);
if (htmldir) {
char *ackmsg = NULL;
char *dismsg = NULL;
char htmllogfn[PATH_MAX];
MEMDEFINE(htmllogfn);
if (metadata[11]) acktime = atoi(metadata[11]);
if (metadata[12] && strlen(metadata[12])) ackmsg = metadata[12];
if (ackmsg) nldecode(ackmsg);
if (metadata[13]) disabletime = atoi(metadata[13]);
if (metadata[14] && strlen(metadata[14]) && (disabletime > 0)) dismsg = metadata[14];
if (dismsg) nldecode(dismsg);
flapping = (metadata[16] ? (*metadata[16] == '1') : 0);
sprintf(htmllogfn, "%s/%s.%s.%s", htmldir, hostname, testname, htmlextension);
update_htmlfile(htmllogfn, statusdata, hostname, testname, parse_color(metadata[7]), flapping,
metadata[2], metadata[8], logtime, timesincechange,
acktime, ackmsg,
disabletime, dismsg);
MEMUNDEFINE(htmllogfn);
}
}
else if ((role == ROLE_DATA) && (metacount > 5) && (strncmp(metadata[0], "@@data", 6) == 0)) {
/* @@data|timestamp|sender|hostname|testname */
p = hostname = metadata[4]; while ((p = strchr(p, '.')) != NULL) *p = ',';
testname = metadata[5];
if (!wantedtest(onlytests, testname)) {
dbgprintf("data dropped - not wanted\n");
MEMUNDEFINE(logfn);
continue;
}
statusdata = msg_data(statusdata); if (*statusdata == '\n') statusdata++;
sprintf(logfn, "%s/%s.%s", filedir, hostname, testname);
expiretime = 0;
update_file(logfn, "a", statusdata, expiretime, NULL, -1, seq);
}
else if ((role == ROLE_NOTES) && (metacount > 3) && (strncmp(metadata[0], "@@notes", 7) == 0)) {
/* @@notes|timestamp|sender|hostname */
hostname = metadata[3];
statusdata = msg_data(statusdata); if (*statusdata == '\n') statusdata++;
sprintf(logfn, "%s/%s", basename(filedir), hostname);
expiretime = 0;
update_file(logfn, "w", statusdata, expiretime, NULL, -1, seq);
}
else if ((role == ROLE_ENADIS) && (metacount > 5) && (strncmp(metadata[0], "@@enadis", 8) == 0)) {
/* @@enadis|timestamp|sender|hostname|testname|expiretime */
p = hostname = metadata[3]; while ((p = strchr(p, '.')) != NULL) *p = ',';
testname = metadata[4];
expiretime = atoi(metadata[5]);
sprintf(logfn, "%s/%s.%s", filedir, hostname, testname);
update_enable(logfn, expiretime);
}
else if (((role == ROLE_STATUS) || (role == ROLE_DATA) || (role == ROLE_ENADIS)) && (metacount > 3) && (strncmp(metadata[0], "@@drophost", 10) == 0)) {
/* @@drophost|timestamp|sender|hostname */
DIR *dirfd;
struct dirent *de;
char *hostlead;
p = hostname = metadata[3]; while ((p = strchr(p, '.')) != NULL) *p = ',';
hostlead = malloc(strlen(hostname) + 2);
strcpy(hostlead, hostname); strcat(hostlead, ".");
dirfd = opendir(filedir);
if (dirfd) {
while ( (de = readdir(dirfd)) != NULL) {
if (strncmp(de->d_name, hostlead, strlen(hostlead)) == 0) {
sprintf(logfn, "%s/%s", filedir, de->d_name);
unlink(logfn);
}
}
closedir(dirfd);
}
xfree(hostlead);
}
else if (((role == ROLE_STATUS) || (role == ROLE_DATA) || (role == ROLE_ENADIS)) && (metacount > 4) && (strncmp(metadata[0], "@@droptest", 10) == 0)) {
/* @@droptest|timestamp|sender|hostname|testname */
p = hostname = metadata[3]; while ((p = strchr(p, '.')) != NULL) *p = ',';
testname = metadata[4];
sprintf(logfn, "%s/%s.%s", filedir, hostname, testname);
unlink(logfn);
}
else if (((role == ROLE_STATUS) || (role == ROLE_DATA) || (role == ROLE_ENADIS)) && (metacount > 4) && (strncmp(metadata[0], "@@renamehost", 12) == 0)) {
/* @@renamehost|timestamp|sender|hostname|newhostname */
DIR *dirfd;
struct dirent *de;
char *hostlead;
char *newhostname;
char newlogfn[PATH_MAX];
MEMDEFINE(newlogfn);
p = hostname = metadata[3]; while ((p = strchr(p, '.')) != NULL) *p = ',';
hostlead = malloc(strlen(hostname) + 2);
strcpy(hostlead, hostname); strcat(hostlead, ".");
p = newhostname = metadata[4]; while ((p = strchr(p, '.')) != NULL) *p = ',';
dirfd = opendir(filedir);
if (dirfd) {
while ( (de = readdir(dirfd)) != NULL) {
if (strncmp(de->d_name, hostlead, strlen(hostlead)) == 0) {
char *testname = strchr(de->d_name, '.');
sprintf(logfn, "%s/%s", filedir, de->d_name);
sprintf(newlogfn, "%s/%s%s", filedir, newhostname, testname);
rename(logfn, newlogfn);
}
}
closedir(dirfd);
}
xfree(hostlead);
MEMUNDEFINE(newlogfn);
}
else if (((role == ROLE_STATUS) || (role == ROLE_DATA) || (role == ROLE_ENADIS)) && (metacount > 5) && (strncmp(metadata[0], "@@renametest", 12) == 0)) {
/* @@renametest|timestamp|sender|hostname|oldtestname|newtestname */
char *newtestname;
char newfn[PATH_MAX];
MEMDEFINE(newfn);
p = hostname = metadata[3]; while ((p = strchr(p, '.')) != NULL) *p = ',';
testname = metadata[4];
newtestname = metadata[5];
sprintf(logfn, "%s/%s.%s", filedir, hostname, testname);
sprintf(newfn, "%s/%s.%s", filedir, hostname, newtestname);
rename(logfn, newfn);
MEMUNDEFINE(newfn);
}
else if (strncmp(metadata[0], "@@shutdown", 10) == 0) {
running = 0;
}
else if (strncmp(metadata[0], "@@logrotate", 11) == 0) {
char *fn = xgetenv("XYMONCHANNEL_LOGFILENAME");
if (fn && strlen(fn)) {
freopen(fn, "a", stdout);
freopen(fn, "a", stderr);
}
continue;
}
else if (strncmp(metadata[0], "@@idle", 6) == 0) {
/* Ignored */
}
else {
errprintf("Dropping message type %s, metacount=%d\n", metadata[0], metacount);
}
MEMUNDEFINE(logfn);
}
return 0;
}
