static void locator_updatecache(enum locator_servicetype_t svc, char *key, char *resp)
{
RbtIterator handle;
cacheitm_t *newitm;
if (!havecache[svc]) return;
handle = rbtFind(locatorcache[svc], key);
if (handle == rbtEnd(locatorcache[svc])) {
newitm = (cacheitm_t *)calloc(1, sizeof(cacheitm_t));
newitm->key = strdup(key);
newitm->resp = strdup(resp);
if (rbtInsert(locatorcache[svc], newitm->key, newitm) != RBT_STATUS_OK) {
xfree(newitm->key);
xfree(newitm->resp);
xfree(newitm);
}
}
else {
newitm = (cacheitm_t *)gettreeitem(locatorcache[svc], handle);
if (newitm->resp) xfree(newitm->resp);
newitm->resp = strdup(resp);
newitm->tstamp = getcurrenttime(NULL);
}
}
static link_t *find_link(char *key)
{
link_t *l = NULL;
RbtIterator handle;
handle = rbtFind(linkstree, key);
if (handle != rbtEnd(linkstree)) l = (link_t *)gettreeitem(linkstree, handle);
return l;
}
void locator_flushcache(enum locator_servicetype_t svc, char *key)
{
RbtIterator handle;
if (!havecache[svc]) return;
if (key) {
handle = rbtFind(locatorcache[svc], key);
if (handle != rbtEnd(locatorcache[svc])) {
cacheitm_t *itm = (cacheitm_t *)gettreeitem(locatorcache[svc], handle);
itm->tstamp = 0;
}
}
else {
for (handle = rbtBegin(locatorcache[svc]); (handle != rbtEnd(locatorcache[svc])); handle = rbtNext(locatorcache[svc], handle)) {
cacheitm_t *itm = (cacheitm_t *)gettreeitem(locatorcache[svc], handle);
itm->tstamp = 0;
}
}
}
static char *locator_querycache(enum locator_servicetype_t svc, char *key)
{
RbtIterator handle;
cacheitm_t *itm;
if (!havecache[svc]) return NULL;
handle = rbtFind(locatorcache[svc], key);
if (handle == rbtEnd(locatorcache[svc])) return NULL;
itm = (cacheitm_t *)gettreeitem(locatorcache[svc], handle);
return (itm->tstamp + cachetimeout[svc]) > getcurrenttime(NULL) ? itm->resp : NULL;
}
int main(int argc, char *argv[])
{
int argi;
struct sigaction sa;
namelist_t *hostwalk;
time_t nexttimeout;
for (argi=1; (argi < argc); argi++) {
if (argnmatch(argv[argi], "--server=")) {
char *p = strchr(argv[argi], '=');
serverip = strdup(p+1);
}
else if (argnmatch(argv[argi], "--interval=")) {
char *p = strchr(argv[argi], '=');
pollinterval = atoi(p+1);
}
else if (argnmatch(argv[argi], "--log-interval=")) {
char *p = strchr(argv[argi], '=');
errorloginterval = atoi(p+1);
}
else if (argnmatch(argv[argi], "--id=")) {
char *p = strchr(argv[argi], '=');
serverid = atoi(p+1);
}
else if (strcmp(argv[argi], "--debug") == 0) {
debug = 1;
}
}
setup_signalhandler("hobbitfetch");
memset(&sa, 0, sizeof(sa));
sa.sa_handler = sigmisc_handler;
sigaction(SIGHUP, &sa, NULL);
sigaction(SIGTERM, &sa, NULL);
sigaction(SIGUSR1, &sa, NULL); /* SIGUSR1 triggers logging of active requests */
clients = rbtNew(name_compare);
nexttimeout = time(NULL) + 60;
{
/* Seed the random number generator */
struct timeval tv;
struct timezone tz;
gettimeofday(&tv, &tz);
srandom(tv.tv_usec);
}
do {
RbtIterator handle;
conn_t *connwalk, *cprev;
fd_set fdread, fdwrite;
int n, maxfd;
struct timeval tmo;
time_t now;
now = time(NULL);
if (now > reloadtime) {
/* Time to reload the bb-hosts file */
reloadtime = now + 600;
load_hostnames(xgetenv("BBHOSTS"), NULL, get_fqdn());
for (hostwalk = first_host(); (hostwalk); hostwalk = hostwalk->next) {
char *hname;
clients_t *newclient;
if (!bbh_item(hostwalk, BBH_FLAG_PULLDATA)) continue;
hname = bbh_item(hostwalk, BBH_HOSTNAME);
handle = rbtFind(clients, hname);
if (handle == rbtEnd(clients)) {
newclient = (clients_t *)calloc(1, sizeof(clients_t));
newclient->hostname = strdup(hname);
rbtInsert(clients, newclient->hostname, newclient);
whentoqueue = now;
}
}
}
now = time(NULL);
if (now > nexttimeout) {
/* Check for connections that have timed out */
nexttimeout = now + 60;
for (connwalk = chead; (connwalk); connwalk = connwalk->next) {
if ((connwalk->tstamp + 60) < now) {
if (debug || (connwalk->client->nexterrortxt < now)) {
errprintf("Timeout while talking to %s (req %lu): Aborting session\n",
addrstring(&connwalk->caddr), connwalk->seq);
connwalk->client->nexterrortxt = now + errorloginterval;
}
flag_cleanup(connwalk);
}
}
}
if (needcleanup) {
/* Remove any finished requests */
needcleanup = 0;
connwalk = chead; cprev = NULL;
dbgprintf("Doing cleanup\n");
while (connwalk) {
conn_t *zombie;
if ((connwalk->action == C_READING) || (connwalk->action == C_WRITING)) {
/* Active connection - skip to the next conn_t record */
cprev = connwalk;
connwalk = connwalk->next;
continue;
}
if (connwalk->action == C_CLEANUP) {
if (connwalk->ctype == C_CLIENT) {
/*
* Finished getting data from a client,
* flag idle and set next poll time.
*/
connwalk->client->busy = 0;
set_polltime(connwalk->client);
}
else if (connwalk->ctype == C_SERVER) {
/* Nothing needed for server cleanups */
}
}
/* Unlink the request from the list of active connections */
zombie = connwalk;
if (cprev == NULL) {
chead = zombie->next;
connwalk = chead;
cprev = NULL;
}
else {
cprev->next = zombie->next;
connwalk = zombie->next;
}
/* Purge the zombie */
dbgprintf("Request completed: req %lu, peer %s, action was %d, type was %d\n",
zombie->seq, addrstring(&zombie->caddr),
zombie->action, zombie->ctype);
close(zombie->sockfd);
freestrbuffer(zombie->msgbuf);
xfree(zombie);
}
/* Set the tail pointer correctly */
ctail = chead;
if (ctail) { while (ctail->next) ctail = ctail->next; }
}
if (dumpsessions) {
/* Set by SIGUSR1 - dump the list of active requests */
dumpsessions = 0;
for (connwalk = chead; (connwalk); connwalk = connwalk->next) {
char *ctypestr, *actionstr;
char timestr[30];
switch (connwalk->ctype) {
case C_CLIENT: ctypestr = "client"; break;
case C_SERVER: ctypestr = "server"; break;
}
switch (connwalk->action) {
case C_READING: actionstr = "reading"; break;
case C_WRITING: actionstr = "writing"; break;
case C_CLEANUP: actionstr = "cleanup"; break;
}
strftime(timestr, sizeof(timestr), "%Y-%m-%d %H:%M:%S",
localtime(&connwalk->tstamp));
errprintf("Request %lu: state %s/%s, peer %s, started %s (%lu secs ago)\n",
connwalk->seq, ctypestr, actionstr, addrstring(&connwalk->caddr),
timestr, (now - connwalk->tstamp));
}
}
now = time(NULL);
if (now >= whentoqueue) {
/* Scan host-tree for clients we need to contact */
for (handle = rbtBegin(clients); (handle != rbtEnd(clients)); handle = rbtNext(clients, handle)) {
clients_t *clientwalk;
char msgline[100];
strbuffer_t *request;
char *pullstr, *ip;
int port;
clientwalk = (clients_t *)gettreeitem(clients, handle);
if (clientwalk->busy) continue;
if (clientwalk->nextpoll > now) continue;
/* Deleted hosts stay in our tree - but should disappear from the known hosts */
hostwalk = hostinfo(clientwalk->hostname); if (!hostwalk) continue;
pullstr = bbh_item(hostwalk, BBH_FLAG_PULLDATA); if (!pullstr) continue;
ip = strchr(pullstr, '=');
port = atoi(xgetenv("BBPORT"));
if (!ip) {
ip = strdup(bbh_item(hostwalk, BBH_IP));
}
else {
/* There is an explicit IP setting in the pulldata tag */
char *p;
ip++; /* Skip the '=' */
ip = strdup(ip);
p = strchr(ip, ':');
if (p) { *p = '\0'; port = atoi(p+1); }
if (*ip == '\0') {
/* No IP given, just a port number */
xfree(ip);
ip = strdup(bbh_item(hostwalk, BBH_IP));
}
}
if (strcmp(ip, "0.0.0.0") == 0) {
struct hostent *hent;
xfree(ip); ip = NULL;
hent = gethostbyname(clientwalk->hostname);
if (hent) {
struct in_addr addr;
memcpy(&addr, *(hent->h_addr_list), sizeof(addr));
ip = strdup(inet_ntoa(addr));
}
}
if (!ip) continue;
/*
* Build the "pullclient" request, which includes the latest
* clientdata config we got from the server. Keep the clientdata
* here - we send "pullclient" requests more often that we actually
* contact the server, but we should provide the config data always.
*/
request = newstrbuffer(0);
sprintf(msgline, "pullclient %d\n", serverid);
addtobuffer(request, msgline);
if (clientwalk->clientdata) addtobuffer(request, clientwalk->clientdata);
/* Put the request on the connection queue */
addrequest(C_CLIENT, ip, port, request, clientwalk);
clientwalk->busy = 1;
xfree(ip);
}
}
/* Handle request queue */
FD_ZERO(&fdread);
FD_ZERO(&fdwrite);
maxfd = -1;
for (connwalk = chead; (connwalk); connwalk = connwalk->next) {
switch (connwalk->action) {
case C_READING:
FD_SET(connwalk->sockfd, &fdread);
if (connwalk->sockfd > maxfd) maxfd = connwalk->sockfd;
break;
case C_WRITING:
FD_SET(connwalk->sockfd, &fdwrite);
if (connwalk->sockfd > maxfd) maxfd = connwalk->sockfd;
break;
case C_CLEANUP:
break;
}
}
/* Do select with a 1 second timeout */
tmo.tv_sec = 1;
tmo.tv_usec = 0;
n = select(maxfd+1, &fdread, &fdwrite, NULL, &tmo);
if (n == -1) {
if (errno == EINTR) continue; /* Interrupted, e.g. a SIGHUP */
/* This is a "cannot-happen" failure. Bail out */
errprintf("select failure: %s\n", strerror(errno));
return 0;
}
if (n == 0) continue; /* Timeout */
for (connwalk = chead; (connwalk); connwalk = connwalk->next) {
switch (connwalk->action) {
case C_READING:
if (FD_ISSET(connwalk->sockfd, &fdread)) grabdata(connwalk);
break;
case C_WRITING:
if (FD_ISSET(connwalk->sockfd, &fdwrite)) senddata(connwalk);
break;
case C_CLEANUP:
break;
}
}
} while (running);
return 0;
}
int load_hostnames(char *bbhostsfn, char *extrainclude, int fqdn)
{
/* Return value: 0 for load OK, 1 for "No files changed since last load", -1 for error (file not found) */
static void *bbhfiles = NULL;
FILE *bbhosts;
int ip1, ip2, ip3, ip4, groupid, pageidx;
char hostname[4096];
strbuffer_t *inbuf;
pagelist_t *curtoppage, *curpage, *pgtail;
namelist_t *nametail = NULL;
RbtHandle htree;
/* First check if there were no modifications at all */
if (bbhfiles) {
if (!stackfmodified(bbhfiles)){
dbgprintf("No files modified, skipping reload of %s\n", bbhostsfn);
return 1;
}
else {
stackfclist(&bbhfiles);
bbhfiles = NULL;
}
}
MEMDEFINE(hostname);
MEMDEFINE(l);
configloaded = 1;
initialize_hostlist();
curpage = curtoppage = pgtail = pghead;
pageidx = groupid = 0;
bbhosts = stackfopen(bbhostsfn, "r", &bbhfiles);
if (bbhosts == NULL) return -1;
inbuf = newstrbuffer(0);
htree = rbtNew(name_compare);
while (stackfgets(inbuf, extrainclude)) {
sanitize_input(inbuf, 0, 0);
if (strncmp(STRBUF(inbuf), "page", 4) == 0) {
pagelist_t *newp;
char *name, *title;
pageidx = groupid = 0;
if (get_page_name_title(STRBUF(inbuf), "page", &name, &title) == 0) {
newp = (pagelist_t *)malloc(sizeof(pagelist_t));
newp->pagepath = strdup(name);
newp->pagetitle = (title ? strdup(title) : NULL);
newp->next = NULL;
pgtail->next = newp;
pgtail = newp;
curpage = curtoppage = newp;
}
}
else if (strncmp(STRBUF(inbuf), "subpage", 7) == 0) {
pagelist_t *newp;
char *name, *title;
pageidx = groupid = 0;
if (get_page_name_title(STRBUF(inbuf), "subpage", &name, &title) == 0) {
newp = (pagelist_t *)malloc(sizeof(pagelist_t));
newp->pagepath = malloc(strlen(curtoppage->pagepath) + strlen(name) + 2);
sprintf(newp->pagepath, "%s/%s", curtoppage->pagepath, name);
newp->pagetitle = malloc(strlen(curtoppage->pagetitle) + strlen(title) + 2);
sprintf(newp->pagetitle, "%s/%s", curtoppage->pagetitle, title);
newp->next = NULL;
pgtail->next = newp;
pgtail = newp;
curpage = newp;
}
}
else if (strncmp(STRBUF(inbuf), "subparent", 9) == 0) {
pagelist_t *newp, *parent;
char *pname, *name, *title;
pageidx = groupid = 0;
parent = NULL;
if (get_page_name_title(STRBUF(inbuf), "subparent", &pname, &title) == 0) {
for (parent = pghead; (parent && !pagematch(parent, pname)); parent = parent->next);
}
if (parent && (get_page_name_title(title, "", &name, &title) == 0)) {
newp = (pagelist_t *)malloc(sizeof(pagelist_t));
newp->pagepath = malloc(strlen(parent->pagepath) + strlen(name) + 2);
sprintf(newp->pagepath, "%s/%s", parent->pagepath, name);
newp->pagetitle = malloc(strlen(parent->pagetitle) + strlen(title) + 2);
sprintf(newp->pagetitle, "%s/%s", parent->pagetitle, title);
newp->next = NULL;
pgtail->next = newp;
pgtail = newp;
curpage = newp;
}
}
else if (strncmp(STRBUF(inbuf), "group", 5) == 0) {
groupid++;
}
else if (sscanf(STRBUF(inbuf), "%d.%d.%d.%d %s", &ip1, &ip2, &ip3, &ip4, hostname) == 5) {
char *startoftags, *tag, *delim;
int elemidx, elemsize;
char clientname[4096];
char downtime[4096];
char groupidstr[10];
RbtIterator handle;
namelist_t *newitem = calloc(1, sizeof(namelist_t));
namelist_t *iwalk, *iprev;
MEMDEFINE(clientname);
MEMDEFINE(downtime);
/* Hostname beginning with '@' are "no-display" hosts. But we still want them. */
if (*hostname == '@') memmove(hostname, hostname+1, strlen(hostname));
if (!fqdn) {
/* Strip any domain from the hostname */
char *p = strchr(hostname, '.');
if (p) *p = '\0';
}
sprintf(newitem->ip, "%d.%d.%d.%d", ip1, ip2, ip3, ip4);
sprintf(groupidstr, "%d", groupid);
newitem->groupid = strdup(groupidstr);
newitem->pageindex = pageidx++;
newitem->bbhostname = strdup(hostname);
if (ip1 || ip2 || ip3 || ip4) newitem->preference = 1; else newitem->preference = 0;
newitem->logname = strdup(newitem->bbhostname);
{ char *p = newitem->logname; while ((p = strchr(p, '.')) != NULL) { *p = '_'; } }
newitem->page = curpage;
newitem->defaulthost = defaulthost;
clientname[0] = downtime[0] = '\0';
startoftags = strchr(STRBUF(inbuf), '#');
if (startoftags == NULL) startoftags = ""; else startoftags++;
startoftags += strspn(startoftags, " \t\r\n");
newitem->allelems = strdup(startoftags);
elemsize = 5;
newitem->elems = (char **)malloc((elemsize+1)*sizeof(char *));
tag = newitem->allelems; elemidx = 0;
while (tag && *tag) {
if (elemidx == elemsize) {
elemsize += 5;
newitem->elems = (char **)realloc(newitem->elems, (elemsize+1)*sizeof(char *));
}
newitem->elems[elemidx] = tag;
/* Skip until we hit a whitespace or a quote */
tag += strcspn(tag, " \t\r\n\"");
if (*tag == '"') {
delim = tag;
/* Hit a quote - skip until the next matching quote */
tag = strchr(tag+1, '"');
if (tag != NULL) {
/* Found end-quote, NULL the item here and move on */
*tag = '\0'; tag++;
}
/* Now move quoted data one byte down (including the NUL) to kill quotechar */
memmove(delim, delim+1, strlen(delim));
}
else if (*tag) {
/* Normal end of item, NULL it and move on */
*tag = '\0'; tag++;
}
else {
/* End of line - no more to do. */
tag = NULL;
}
/*
* If we find a "noconn", drop preference value to 0.
* If we find a "prefer", up reference value to 2.
*/
if ((newitem->preference == 1) && (strcmp(newitem->elems[elemidx], "noconn") == 0))
newitem->preference = 0;
else if (strcmp(newitem->elems[elemidx], "prefer") == 0)
newitem->preference = 2;
/* Skip whitespace until start of next tag */
if (tag) tag += strspn(tag, " \t\r\n");
elemidx++;
}
newitem->elems[elemidx] = NULL;
/* See if this host is defined before */
handle = rbtFind(htree, newitem->bbhostname);
if (strcasecmp(newitem->bbhostname, ".default.") == 0) {
/* The pseudo DEFAULT host */
newitem->next = NULL;
defaulthost = newitem;
}
else if (handle == rbtEnd(htree)) {
/* New item, so add to end of list */
newitem->next = NULL;
if (namehead == NULL)
namehead = nametail = newitem;
else {
nametail->next = newitem;
nametail = newitem;
}
rbtInsert(htree, newitem->bbhostname, newitem);
}
else {
/* Find the existing record - compare the record pointer instead of the name */
namelist_t *existingrec = (namelist_t *)gettreeitem(htree, handle);
for (iwalk = namehead, iprev = NULL; ((iwalk != existingrec) && iwalk); iprev = iwalk, iwalk = iwalk->next) ;
if (newitem->preference <= iwalk->preference) {
/* Add after the existing (more preferred) entry */
newitem->next = iwalk->next;
iwalk->next = newitem;
}
else {
/* New item has higher preference, so add before the iwalk item (i.e. after iprev) */
if (iprev == NULL) {
newitem->next = namehead;
namehead = newitem;
}
else {
newitem->next = iprev->next;
iprev->next = newitem;
}
}
}
newitem->clientname = bbh_find_item(newitem, BBH_CLIENTALIAS);
if (newitem->clientname == NULL) newitem->clientname = newitem->bbhostname;
newitem->downtime = bbh_find_item(newitem, BBH_DOWNTIME);
MEMUNDEFINE(clientname);
MEMUNDEFINE(downtime);
}
}
stackfclose(bbhosts);
freestrbuffer(inbuf);
rbtDelete(htree);
MEMUNDEFINE(hostname);
MEMUNDEFINE(l);
build_hosttree();
return 0;
}
int main(int argc, char *argv[])
{
int daemonize = 0;
char *pidfile = NULL;
char *envarea = NULL;
int cnid = -1;
pcre *msgfilter = NULL;
pcre *stdfilter = NULL;
int argi;
struct sigaction sa;
RbtIterator handle;
/* Dont save the error buffer */
save_errbuf = 0;
/* Create the peer container */
peers = rbtNew(name_compare);
for (argi=1; (argi < argc); argi++) {
if (argnmatch(argv[argi], "--debug")) {
debug = 1;
}
else if (argnmatch(argv[argi], "--channel=")) {
char *cn = strchr(argv[argi], '=') + 1;
for (cnid = C_STATUS; (channelnames[cnid] && strcmp(channelnames[cnid], cn)); cnid++) ;
if (channelnames[cnid] == NULL) cnid = -1;
}
else if (argnmatch(argv[argi], "--daemon")) {
daemonize = 1;
}
else if (argnmatch(argv[argi], "--no-daemon")) {
daemonize = 0;
}
else if (argnmatch(argv[argi], "--pidfile=")) {
char *p = strchr(argv[argi], '=');
pidfile = strdup(p+1);
}
else if (argnmatch(argv[argi], "--log=")) {
char *p = strchr(argv[argi], '=');
logfn = strdup(p+1);
}
else if (argnmatch(argv[argi], "--env=")) {
char *p = strchr(argv[argi], '=');
loadenv(p+1, envarea);
}
else if (argnmatch(argv[argi], "--area=")) {
char *p = strchr(argv[argi], '=');
envarea = strdup(p+1);
}
else if (argnmatch(argv[argi], "--locator=")) {
char *p = strchr(argv[argi], '=');
locator_init(p+1);
locatorbased = 1;
}
else if (argnmatch(argv[argi], "--service=")) {
char *p = strchr(argv[argi], '=');
locatorservice = get_servicetype(p+1);
}
else if (argnmatch(argv[argi], "--filter=")) {
char *p = strchr(argv[argi], '=');
msgfilter = compileregex(p+1);
if (!msgfilter) {
errprintf("Invalid filter (bad expression): %s\n", p+1);
}
else {
stdfilter = compileregex("^@@(logrotate|shutdown|drophost|droptest|renamehost|renametest)");
}
}
else {
char *childcmd;
char **childargs;
int i = 0;
childcmd = argv[argi];
childargs = (char **) calloc((1 + argc - argi), sizeof(char *));
while (argi < argc) { childargs[i++] = argv[argi++]; }
addlocalpeer(childcmd, childargs);
}
}
/* Sanity checks */
if (cnid == -1) {
errprintf("No channel/unknown channel specified\n");
return 1;
}
if (locatorbased && (locatorservice == ST_MAX)) {
errprintf("Must specify --service when using locator\n");
return 1;
}
if (!locatorbased && (rbtBegin(peers) == rbtEnd(peers))) {
errprintf("Must specify command for local worker\n");
return 1;
}
/* Do cache responses to avoid doing too many lookups */
if (locatorbased) locator_prepcache(locatorservice, 0);
/* Go daemon */
if (daemonize) {
/* Become a daemon */
pid_t daemonpid = fork();
if (daemonpid < 0) {
/* Fork failed */
errprintf("Could not fork child\n");
exit(1);
}
else if (daemonpid > 0) {
/* Parent creates PID file and exits */
FILE *fd = NULL;
if (pidfile) fd = fopen(pidfile, "w");
if (fd) {
fprintf(fd, "%d\n", (int)daemonpid);
fclose(fd);
}
exit(0);
}
/* Child (daemon) continues here */
setsid();
}
/* Catch signals */
setup_signalhandler("xymond_channel");
memset(&sa, 0, sizeof(sa));
sa.sa_handler = sig_handler;
sigaction(SIGINT, &sa, NULL);
sigaction(SIGTERM, &sa, NULL);
sigaction(SIGCHLD, &sa, NULL);
signal(SIGALRM, SIG_IGN);
/* Switch stdout/stderr to the logfile, if one was specified */
freopen("/dev/null", "r", stdin); /* xymond_channel's stdin is not used */
if (logfn) {
freopen(logfn, "a", stdout);
freopen(logfn, "a", stderr);
}
/* Attach to the channel */
channel = setup_channel(cnid, CHAN_CLIENT);
if (channel == NULL) {
errprintf("Channel not available\n");
running = 0;
}
while (running) {
/*
* Wait for GOCLIENT to go up.
*
* Note that we use IPC_NOWAIT if there are messages in the
* queue, because then we just want to pick up a message if
* there is one, and if not we want to continue pushing the
* queued data to the worker.
*/
struct sembuf s;
int n;
s.sem_num = GOCLIENT; s.sem_op = -1; s.sem_flg = ((pendingcount > 0) ? IPC_NOWAIT : 0);
n = semop(channel->semid, &s, 1);
if (n == 0) {
/*
* GOCLIENT went high, and so we got alerted about a new
* message arriving. Copy the message to our own buffer queue.
*/
char *inbuf = NULL;
if (!msgfilter || matchregex(channel->channelbuf, msgfilter) || matchregex(channel->channelbuf, stdfilter)) {
inbuf = strdup(channel->channelbuf);
}
/*
* Now we have safely stored the new message in our buffer.
* Wait until any other clients on the same channel have picked up
* this message (GOCLIENT reaches 0).
*
* We wrap this into an alarm handler, because it can occasionally
* fail, causing the whole system to lock up. We dont want that....
* We'll set the alarm to trigger after 1 second. Experience shows
* that we'll either succeed in a few milliseconds, or fail completely
* and wait the full alarm-timer duration.
*/
gotalarm = 0; signal(SIGALRM, sig_handler); alarm(2);
do {
s.sem_num = GOCLIENT; s.sem_op = 0; s.sem_flg = 0;
n = semop(channel->semid, &s, 1);
} while ((n == -1) && (errno == EAGAIN) && running && (!gotalarm));
signal(SIGALRM, SIG_IGN);
if (gotalarm) {
errprintf("Gave up waiting for GOCLIENT to go low.\n");
}
/*
* Let master know we got it by downing BOARDBUSY.
* This should not block, since BOARDBUSY is upped
* by the master just before he ups GOCLIENT.
*/
do {
s.sem_num = BOARDBUSY; s.sem_op = -1; s.sem_flg = IPC_NOWAIT;
n = semop(channel->semid, &s, 1);
} while ((n == -1) && (errno == EINTR));
if (n == -1) {
errprintf("Tried to down BOARDBUSY: %s\n", strerror(errno));
}
if (inbuf) {
/*
* See if they want us to rotate logs. We pass this on to
* the worker module as well, but must handle our own logfile.
*/
if (strncmp(inbuf, "@@logrotate", 11) == 0) {
freopen(logfn, "a", stdout);
freopen(logfn, "a", stderr);
}
/*
* Put the new message on our outbound queue.
*/
if (addmessage(inbuf) != 0) {
/* Failed to queue message, free the buffer */
xfree(inbuf);
}
}
}
else {
if (errno != EAGAIN) {
dbgprintf("Semaphore wait aborted: %s\n", strerror(errno));
continue;
}
}
/*
* We've picked up messages from the master. Now we
* must push them to the worker process. Since there
* is no way to hang off both a semaphore and select(),
* we'll just push as much data as possible into the
* pipe. If we get to a point where we would block,
* then wait a teeny bit of time and restart the
* whole loop with checking for new messages from the
* master etc.
*
* In theory, this could become an almost busy-wait loop.
* In practice, however, the queue will be empty most
* of the time because we'll just shove the data to the
* worker child.
*/
for (handle = rbtBegin(peers); (handle != rbtEnd(peers)); handle = rbtNext(peers, handle)) {
int canwrite = 1, hasfailed = 0;
xymon_peer_t *pwalk;
time_t msgtimeout = gettimer() - MSGTIMEOUT;
int flushcount = 0;
pwalk = (xymon_peer_t *) gettreeitem(peers, handle);
if (pwalk->msghead == NULL) continue; /* Ignore peers with nothing queued */
switch (pwalk->peerstatus) {
case P_UP:
canwrite = 1;
break;
case P_DOWN:
openconnection(pwalk);
canwrite = (pwalk->peerstatus == P_UP);
break;
case P_FAILED:
canwrite = 0;
break;
}
/* See if we have stale messages queued */
while (pwalk->msghead && (pwalk->msghead->tstamp < msgtimeout)) {
flushmessage(pwalk);
flushcount++;
}
if (flushcount) {
errprintf("Flushed %d stale messages for %s:%d\n",
flushcount,
inet_ntoa(pwalk->peeraddr.sin_addr),
ntohs(pwalk->peeraddr.sin_port));
}
while (pwalk->msghead && canwrite) {
fd_set fdwrite;
struct timeval tmo;
/* Check that this peer is ready for writing. */
FD_ZERO(&fdwrite); FD_SET(pwalk->peersocket, &fdwrite);
tmo.tv_sec = 0; tmo.tv_usec = 2000;
n = select(pwalk->peersocket+1, NULL, &fdwrite, NULL, &tmo);
if (n == -1) {
errprintf("select() failed: %s\n", strerror(errno));
canwrite = 0;
hasfailed = 1;
continue;
}
else if ((n == 0) || (!FD_ISSET(pwalk->peersocket, &fdwrite))) {
canwrite = 0;
continue;
}
n = write(pwalk->peersocket, pwalk->msghead->bufp, pwalk->msghead->buflen);
if (n >= 0) {
pwalk->msghead->bufp += n;
pwalk->msghead->buflen -= n;
if (pwalk->msghead->buflen == 0) flushmessage(pwalk);
}
else if (errno == EAGAIN) {
/*
* Write would block ... stop for now.
*/
canwrite = 0;
}
else {
hasfailed = 1;
}
if (hasfailed) {
/* Write failed, or message grew stale */
errprintf("Peer at %s:%d failed: %s\n",
inet_ntoa(pwalk->peeraddr.sin_addr), ntohs(pwalk->peeraddr.sin_port),
strerror(errno));
canwrite = 0;
shutdownconnection(pwalk);
if (pwalk->peertype == P_NET) locator_serverdown(pwalk->peername, locatorservice);
pwalk->peerstatus = P_FAILED;
}
}
}
}
/* Detach from channels */
close_channel(channel, CHAN_CLIENT);
/* Close peer connections */
for (handle = rbtBegin(peers); (handle != rbtEnd(peers)); handle = rbtNext(peers, handle)) {
xymon_peer_t *pwalk = (xymon_peer_t *) gettreeitem(peers, handle);
shutdownconnection(pwalk);
}
/* Remove the PID file */
if (pidfile) unlink(pidfile);
return 0;
}
int addmessage(char *inbuf)
{
RbtIterator phandle;
xymon_peer_t *peer;
int bcastmsg = 0;
int inlen = strlen(inbuf);
if (locatorbased) {
char *hostname, *hostend, *peerlocation;
/* xymond sends us messages with the KEY in the first field, between a '/' and a '|' */
hostname = inbuf + strcspn(inbuf, "/|\r\n");
if (*hostname != '/') {
errprintf("No key field in message, dropping it\n");
return -1; /* Malformed input */
}
hostname++;
bcastmsg = (*hostname == '*');
if (!bcastmsg) {
/* Lookup which server handles this host */
hostend = hostname + strcspn(hostname, "|\r\n");
if (*hostend != '|') {
errprintf("No delimiter found in input, dropping it\n");
return -1; /* Malformed input */
}
*hostend = '\0';
peerlocation = locator_query(hostname, locatorservice, NULL);
/*
* If we get no response, or an empty response,
* then there is no server capable of handling this
* request.
*/
if (!peerlocation || (*peerlocation == '\0')) {
errprintf("No response from locator for %s/%s, dropping it\n",
servicetype_names[locatorservice], hostname);
return -1;
}
*hostend = '|';
phandle = rbtFind(peers, peerlocation);
if (phandle == rbtEnd(peers)) {
/* New peer - register it */
addnetpeer(peerlocation);
phandle = rbtFind(peers, peerlocation);
}
}
}
else {
phandle = rbtFind(peers, "");
}
if (bcastmsg) {
for (phandle = rbtBegin(peers); (phandle != rbtEnd(peers)); phandle = rbtNext(peers, phandle)) {
peer = (xymon_peer_t *)gettreeitem(peers, phandle);
addmessage_onepeer(peer, inbuf, inlen);
}
}
else {
if (phandle == rbtEnd(peers)) {
errprintf("No peer found to handle message, dropping it\n");
return -1;
}
peer = (xymon_peer_t *)gettreeitem(peers, phandle);
addmessage_onepeer(peer, inbuf, inlen);
}
return 0;
}
