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;
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;
}