int ofi_mr_init(const struct fi_provider *in_prov, enum fi_mr_mode mode,
struct ofi_util_mr ** out_new_mr)
{
struct ofi_util_mr * new_mr = malloc(sizeof(struct ofi_util_mr));
if (!new_mr)
return -FI_ENOMEM;
assert((mode == FI_MR_SCALABLE) || (mode == FI_MR_BASIC));
new_mr->mr_type = mode;
new_mr->map_handle = rbtNew(compare_mr_keys);
if (!new_mr->map_handle) {
free(new_mr);
return -FI_ENOMEM;
}
new_mr->b_key = 0;
new_mr->prov = in_prov;
(*out_new_mr) = new_mr;
return 0;
}
void locator_prepcache(enum locator_servicetype_t svc, int timeout)
{
if (!havecache[svc]) {
locatorcache[svc] = rbtNew(name_compare);
havecache[svc] = 1;
}
else {
locator_flushcache(svc, NULL);
}
cachetimeout[svc] = ((timeout>0) ? timeout : DEFAULT_CACHETIMEOUT);
}
static void build_hosttree(void)
{
static int hosttree_exists = 0;
namelist_t *walk;
RbtStatus status;
char *tstr;
if (hosttree_exists) {
rbtDelete(rbhosts);
rbtDelete(rbclients);
}
rbhosts = rbtNew(name_compare);
rbclients = rbtNew(name_compare);
hosttree_exists = 1;
for (walk = namehead; (walk); walk = walk->next) {
status = rbtInsert(rbhosts, walk->bbhostname, walk);
if (walk->clientname) rbtInsert(rbclients, walk->clientname, walk);
switch (status) {
case RBT_STATUS_OK:
case RBT_STATUS_DUPLICATE_KEY:
break;
case RBT_STATUS_MEM_EXHAUSTED:
errprintf("loadhosts:build_hosttree - insert into tree failed (out of memory)\n");
break;
default:
errprintf("loadhosts:build_hosttree - insert into tree failed code %d\n", status);
break;
}
tstr = bbh_item(walk, BBH_NOTBEFORE);
walk->notbefore = (tstr ? timestr2timet(tstr) : 0);
if (walk->notbefore == -1) walk->notbefore = 0;
tstr = bbh_item(walk, BBH_NOTAFTER);
walk->notafter = (tstr ? timestr2timet(tstr) : INT_MAX);
if (walk->notafter == -1) walk->notafter = INT_MAX;
}
}
static int ofi_mr_cache_init_rbt_storage(struct ofi_mr_cache *cache)
{
cache->mr_storage.storage = rbtNew(cache->merge_regions ?
ofi_mr_find_overlap :
ofi_mr_find_within);
if (!cache->mr_storage.storage)
return -FI_ENOMEM;
cache->mr_storage.destroy = ofi_mr_rbt_storage_destroy;
cache->mr_storage.find = ofi_mr_rbt_storage_find;
cache->mr_storage.insert = ofi_mr_rbt_storage_insert;
cache->mr_storage.erase = ofi_mr_rbt_storage_erase;
return 0;
}
RbtHandle columnlist(RbtHandle statetree)
{
RbtHandle rbcolumns;
RbtIterator hhandle;
rbcolumns = rbtNew(name_compare);
for (hhandle = rbtBegin(statetree); (hhandle != rbtEnd(statetree)); hhandle = rbtNext(statetree, hhandle)) {
void *k1, *k2;
hstatus_t *itm;
RbtStatus status;
rbtKeyValue(statetree, hhandle, &k1, &k2);
itm = (hstatus_t *)k2;
status = rbtInsert(rbcolumns, itm->testname, NULL);
}
return rbcolumns;
}
void load_all_links(void)
{
char dirname[PATH_MAX];
char *p;
MEMDEFINE(dirname);
dbgprintf("load_all_links()\n");
linkstree = rbtNew(name_compare);
if (notesskin) { xfree(notesskin); notesskin = NULL; }
if (helpskin) { xfree(helpskin); helpskin = NULL; }
if (columndocurl) { xfree(columndocurl); columndocurl = NULL; }
if (xgetenv("XYMONNOTESSKIN")) notesskin = strdup(xgetenv("XYMONNOTESSKIN"));
else {
notesskin = (char *) malloc(strlen(xgetenv("XYMONWEB")) + strlen("/notes") + 1);
sprintf(notesskin, "%s/notes", xgetenv("XYMONWEB"));
}
if (xgetenv("XYMONHELPSKIN")) helpskin = strdup(xgetenv("XYMONHELPSKIN"));
else {
helpskin = (char *) malloc(strlen(xgetenv("XYMONWEB")) + strlen("/help") + 1);
sprintf(helpskin, "%s/help", xgetenv("XYMONWEB"));
}
if (xgetenv("COLUMNDOCURL")) columndocurl = strdup(xgetenv("COLUMNDOCURL"));
strcpy(dirname, xgetenv("XYMONNOTESDIR"));
load_links(dirname, notesskin);
/* Change xxx/xxx/xxx/notes into xxx/xxx/xxx/help */
p = strrchr(dirname, '/'); *p = '\0'; strcat(dirname, "/help");
load_links(dirname, helpskin);
linksloaded = 1;
MEMUNDEFINE(dirname);
}
static int psmx2_domain_init(struct psmx2_fid_domain *domain,
struct psmx2_ep_name *src_addr)
{
int err;
err = fastlock_init(&domain->mr_lock);
if (err) {
FI_WARN(&psmx2_prov, FI_LOG_CORE,
"fastlock_init(mr_lock) returns %d\n", err);
goto err_out;
}
domain->mr_map = rbtNew(&psmx2_key_compare);
if (!domain->mr_map) {
FI_WARN(&psmx2_prov, FI_LOG_CORE,
"rbtNew failed\n");
goto err_out_destroy_mr_lock;
}
domain->mr_reserved_key = 1;
domain->max_atomic_size = INT_MAX;
ofi_atomic_initialize32(&domain->sep_cnt, 0);
fastlock_init(&domain->sep_lock);
dlist_init(&domain->sep_list);
dlist_init(&domain->trx_ctxt_list);
fastlock_init(&domain->trx_ctxt_lock);
if (domain->progress_thread_enabled)
psmx2_domain_start_progress(domain);
return 0;
err_out_destroy_mr_lock:
fastlock_destroy(&domain->mr_lock);
err_out:
return err;
}
/*
* If a provider or app whose version is < 1.5, calls this function and passes
* FI_MR_UNSPEC as mode, it would be treated as MR scalable.
*/
int ofi_mr_map_init(const struct fi_provider *prov, int mode,
struct ofi_mr_map *map)
{
map->rbtree = rbtNew(compare_mr_keys);
if (!map->rbtree)
return -FI_ENOMEM;
switch (mode) {
case FI_MR_BASIC:
map->mode = OFI_MR_BASIC_MAP;
break;
case FI_MR_SCALABLE:
map->mode = 0;
break;
default:
map->mode = mode;
}
map->prov = prov;
map->key = 1;
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 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_nkconfig(char *fn)
{
static void *configfiles = NULL;
static int firsttime = 1;
FILE *fd;
strbuffer_t *inbuf;
/* Setup the default configuration filename */
if (!fn) {
if (!defaultfn) {
char *bbhome = xgetenv("BBHOME");
defaultfn = (char *)malloc(strlen(bbhome) + strlen(DEFAULTCONFIG) + 2);
sprintf(defaultfn, "%s/%s", bbhome, DEFAULTCONFIG);
}
fn = defaultfn;
}
if (configfn) xfree(configfn);
configfn = strdup(fn);
/* First check if there were no modifications at all */
if (configfiles) {
if (!stackfmodified(configfiles)){
dbgprintf("No files modified, skipping reload of %s\n", fn);
return 0;
}
else {
stackfclist(&configfiles);
configfiles = NULL;
}
}
if (!firsttime) {
/* Clean up existing datatree */
RbtHandle handle;
void *k1, *k2;
for (handle = rbtBegin(rbconf); (handle != rbtEnd(rbconf)); handle = rbtNext(rbconf, handle)) {
rbtKeyValue(rbconf, handle, &k1, &k2);
flushrec(k1, k2);
}
rbtDelete(rbconf);
}
firsttime = 0;
rbconf = rbtNew(name_compare);
fd = stackfopen(fn, "r", &configfiles);
if (fd == NULL) return 1;
inbuf = newstrbuffer(0);
while (stackfgets(inbuf, NULL)) {
/* Full record : Host service START END TIMESPEC TTPrio TTGroup TTExtra */
/* Clone record: Host =HOST */
char *ehost, *eservice, *estart, *eend, *etime, *ttgroup, *ttextra, *updinfo;
int ttprio = 0;
nkconf_t *newitem;
RbtStatus status;
int idx = 0;
ehost = gettok(STRBUF(inbuf), "|\n"); if (!ehost) continue;
eservice = gettok(NULL, "|\n"); if (!eservice) continue;
if (*eservice == '=') {
char *key = (char *)malloc(strlen(ehost) + 2);
char *pointsto = strdup(eservice+1);
sprintf(key, "%s=", ehost);
status = rbtInsert(rbconf, key, pointsto);
}
else {
estart = gettok(NULL, "|\n"); if (!estart) continue;
eend = gettok(NULL, "|\n"); if (!eend) continue;
etime = gettok(NULL, "|\n"); if (!etime) continue;
ttprio = atoi(gettok(NULL, "|\n")); if (ttprio == 0) continue;
ttgroup = gettok(NULL, "|\n");
ttextra = gettok(NULL, "|\n");
updinfo = gettok(NULL, "|\n");
newitem = (nkconf_t *)malloc(sizeof(nkconf_t));
newitem->key = (char *)malloc(strlen(ehost) + strlen(eservice) + 15);
sprintf(newitem->key, "%s|%s", ehost, eservice);
newitem->starttime= ((estart && *estart) ? atoi(estart) : 0);
newitem->endtime = ((eend && *eend) ? atoi(eend) : 0);
newitem->nktime = ((etime && *etime) ? strdup(etime) : NULL);
newitem->priority = ttprio;
newitem->ttgroup = strdup(ttgroup);
newitem->ttextra = strdup(ttextra);
newitem->updinfo = strdup(updinfo);
status = rbtInsert(rbconf, newitem->key, newitem);
while (status == RBT_STATUS_DUPLICATE_KEY) {
idx++;
sprintf(newitem->key, "%s|%s|%d", ehost, eservice, idx);
status = rbtInsert(rbconf, newitem->key, newitem);
}
}
}
stackfclose(fd);
freestrbuffer(inbuf);
if (debug) {
RbtHandle handle;
handle = rbtBegin(rbconf);
while (handle != rbtEnd(rbconf)) {
void *k1, *k2;
rbtKeyValue(rbconf, handle, &k1, &k2);
printf("%s\n", (char *)k1);
handle = rbtNext(rbconf, handle);
}
}
return 0;
}
static int psmx2_domain_init(struct psmx2_fid_domain *domain,
struct psmx2_src_name *src_addr)
{
int err;
psmx2_am_global_init();
psmx2_atomic_global_init();
domain->base_trx_ctxt = psmx2_trx_ctxt_alloc(domain, src_addr, -1);
if (!domain->base_trx_ctxt)
return -FI_ENODEV;
err = fastlock_init(&domain->mr_lock);
if (err) {
FI_WARN(&psmx2_prov, FI_LOG_CORE,
"fastlock_init(mr_lock) returns %d\n", err);
goto err_out_free_trx_ctxt;
}
domain->mr_map = rbtNew(&psmx2_key_compare);
if (!domain->mr_map) {
FI_WARN(&psmx2_prov, FI_LOG_CORE,
"rbtNew failed\n");
goto err_out_destroy_mr_lock;
}
domain->mr_reserved_key = 1;
err = fastlock_init(&domain->vl_lock);
if (err) {
FI_WARN(&psmx2_prov, FI_LOG_CORE,
"fastlock_init(vl_lock) returns %d\n", err);
goto err_out_delete_mr_map;
}
memset(domain->vl_map, 0, sizeof(domain->vl_map));
domain->vl_alloc = 0;
ofi_atomic_initialize32(&domain->sep_cnt, 0);
fastlock_init(&domain->sep_lock);
dlist_init(&domain->sep_list);
dlist_init(&domain->trx_ctxt_list);
fastlock_init(&domain->trx_ctxt_lock);
dlist_insert_before(&domain->base_trx_ctxt->entry, &domain->trx_ctxt_list);
/* Set active domain before psmx2_domain_enable_ep() installs the
* AM handlers to ensure that psmx2_active_fabric->active_domain
* is always non-NULL inside the handlers. Notice that the vlaue
* active_domain becomes NULL again only when the domain is closed.
* At that time the AM handlers are gone with the PSM endpoint.
*/
domain->fabric->active_domain = domain;
if (psmx2_domain_enable_ep(domain, NULL) < 0)
goto err_out_reset_active_domain;
if (domain->progress_thread_enabled)
psmx2_domain_start_progress(domain);
psmx2_am_init(domain->base_trx_ctxt);
return 0;
err_out_reset_active_domain:
domain->fabric->active_domain = NULL;
fastlock_destroy(&domain->vl_lock);
err_out_delete_mr_map:
rbtDelete(domain->mr_map);
err_out_destroy_mr_lock:
fastlock_destroy(&domain->mr_lock);
err_out_free_trx_ctxt:
psmx2_trx_ctxt_free(domain->base_trx_ctxt);
return err;
}
int loadstatus(int maxprio, time_t maxage, int mincolor, int wantacked)
{
int xymondresult;
char *board = NULL;
char *bol, *eol;
time_t now;
char msg[1024];
int i;
sendreturn_t *sres;
sprintf(msg, "xymondboard acklevel=%d fields=hostname,testname,color,lastchange,logtime,validtime,acklist color=%s", critacklevel,colorname(mincolor));
for (i=mincolor+1; (i < COL_COUNT); i++) sprintf(msg+strlen(msg), ",%s", colorname(i));
sres = newsendreturnbuf(1, NULL);
xymondresult = sendmessage(msg, NULL, XYMON_TIMEOUT, sres);
if (xymondresult != XYMONSEND_OK) {
freesendreturnbuf(sres);
errormsg("Unable to fetch current status\n");
return 1;
}
else {
board = getsendreturnstr(sres, 1);
freesendreturnbuf(sres);
}
now = getcurrenttime(NULL);
rbstate = rbtNew(name_compare);
bol = board;
while (bol && (*bol)) {
char *endkey;
RbtStatus status;
eol = strchr(bol, '\n'); if (eol) *eol = '\0';
/* Find the config entry */
endkey = strchr(bol, '|'); if (endkey) endkey = strchr(endkey+1, '|');
if (endkey) {
critconf_t *cfg;
char *ackstr, *ackrtimestr, *ackvtimestr, *acklevelstr, *ackbystr, *ackmsgstr;
*endkey = '\0';
cfg = get_critconfig(bol, CRITCONF_TIMEFILTER, NULL);
*endkey = '|';
if (cfg) {
hstatus_t *newitem = (hstatus_t *)calloc(1, sizeof(hstatus_t));
newitem->config = cfg;
newitem->hostname = gettok(bol, "|");
newitem->testname = gettok(NULL, "|");
newitem->color = parse_color(gettok(NULL, "|"));
newitem->lastchange = atoi(gettok(NULL, "|"));
newitem->logtime = atoi(gettok(NULL, "|"));
newitem->validtime = atoi(gettok(NULL, "|"));
ackstr = gettok(NULL, "|");
ackrtimestr = ackvtimestr = acklevelstr = ackbystr = ackmsgstr = NULL;
if (ackstr) {
nldecode(ackstr);
ackrtimestr = strtok(ackstr, ":");
if (ackrtimestr) ackvtimestr = strtok(NULL, ":");
if (ackvtimestr) acklevelstr = strtok(NULL, ":");
if (acklevelstr) ackbystr = strtok(NULL, ":");
if (ackbystr) ackmsgstr = strtok(NULL, ":");
}
if ( (hostinfo(newitem->hostname) == NULL) ||
(newitem->config->priority > maxprio) ||
((now - newitem->lastchange) > maxage) ||
(newitem->color < mincolor) ||
(ackmsgstr && !wantacked) ) {
xfree(newitem);
}
else {
if (ackvtimestr && ackbystr && ackmsgstr) {
newitem->acktime = atoi(ackvtimestr);
newitem->ackedby = strdup(ackbystr);
newitem->ackmsg = strdup(ackmsgstr);
}
newitem->key = (char *)malloc(strlen(newitem->hostname) + strlen(newitem->testname) + 2);
sprintf(newitem->key, "%s|%s", newitem->hostname, newitem->testname);
status = rbtInsert(rbstate, newitem->key, newitem);
}
}
}
bol = (eol ? (eol+1) : NULL);
}
return 0;
}
static int util_ns_map_init(struct util_ns *ns)
{
ns->ns_map = rbtNew(ns->service_cmp);
return ns->ns_map ? 0 : -FI_ENOMEM;
}
int psmx2_domain_open(struct fid_fabric *fabric, struct fi_info *info,
struct fid_domain **domain, void *context)
{
struct psmx2_fid_fabric *fabric_priv;
struct psmx2_fid_domain *domain_priv;
struct psm2_ep_open_opts opts;
int err;
FI_INFO(&psmx2_prov, FI_LOG_DOMAIN, "\n");
fabric_priv = container_of(fabric, struct psmx2_fid_fabric, fabric);
psmx2_fabric_acquire(fabric_priv);
if (fabric_priv->active_domain) {
psmx2_domain_acquire(fabric_priv->active_domain);
*domain = &fabric_priv->active_domain->domain;
return 0;
}
if (!info->domain_attr->name ||
strcmp(info->domain_attr->name, PSMX2_DOMAIN_NAME)) {
err = -FI_EINVAL;
goto err_out;
}
domain_priv = (struct psmx2_fid_domain *) calloc(1, sizeof *domain_priv);
if (!domain_priv) {
err = -FI_ENOMEM;
goto err_out;
}
domain_priv->domain.fid.fclass = FI_CLASS_DOMAIN;
domain_priv->domain.fid.context = context;
domain_priv->domain.fid.ops = &psmx2_fi_ops;
domain_priv->domain.ops = &psmx2_domain_ops;
domain_priv->domain.mr = &psmx2_mr_ops;
domain_priv->mr_mode = info->domain_attr->mr_mode;
domain_priv->mode = info->mode;
domain_priv->caps = info->caps;
domain_priv->fabric = fabric_priv;
domain_priv->progress_thread_enabled =
(info->domain_attr->data_progress == FI_PROGRESS_AUTO);
psm2_ep_open_opts_get_defaults(&opts);
FI_INFO(&psmx2_prov, FI_LOG_CORE,
"uuid: %s\n", psmx2_uuid_to_string(fabric_priv->uuid));
err = psm2_ep_open(fabric_priv->uuid, &opts,
&domain_priv->psm2_ep, &domain_priv->psm2_epid);
if (err != PSM2_OK) {
FI_WARN(&psmx2_prov, FI_LOG_CORE,
"psm2_ep_open returns %d, errno=%d\n", err, errno);
err = psmx2_errno(err);
goto err_out_free_domain;
}
FI_INFO(&psmx2_prov, FI_LOG_CORE,
"epid: 0x%016lx\n", domain_priv->psm2_epid);
err = psm2_mq_init(domain_priv->psm2_ep, PSM2_MQ_ORDERMASK_ALL,
NULL, 0, &domain_priv->psm2_mq);
if (err != PSM2_OK) {
FI_WARN(&psmx2_prov, FI_LOG_CORE,
"psm2_mq_init returns %d, errno=%d\n", err, errno);
err = psmx2_errno(err);
goto err_out_close_ep;
}
err = fastlock_init(&domain_priv->mr_lock);
if (err) {
FI_WARN(&psmx2_prov, FI_LOG_CORE,
"fastlock_init(mr_lock) returns %d\n", err);
goto err_out_finalize_mq;
}
domain_priv->mr_map = rbtNew(&psmx2_key_compare);
if (!domain_priv->mr_map) {
FI_WARN(&psmx2_prov, FI_LOG_CORE,
"rbtNew failed\n");
goto err_out_destroy_mr_lock;
}
domain_priv->mr_reserved_key = 1;
err = fastlock_init(&domain_priv->vl_lock);
if (err) {
FI_WARN(&psmx2_prov, FI_LOG_CORE,
"fastlock_init(vl_lock) returns %d\n", err);
goto err_out_delete_mr_map;
}
memset(domain_priv->vl_map, 0, sizeof(domain_priv->vl_map));
domain_priv->vl_alloc = 0;
err = fastlock_init(&domain_priv->poll_lock);
if (err) {
FI_WARN(&psmx2_prov, FI_LOG_CORE,
"fastlock_init(poll_lock) returns %d\n", err);
goto err_out_destroy_vl_lock;
}
/* Set active domain before psmx2_domain_enable_ep() installs the
* AM handlers to ensure that psmx2_active_fabric->active_domain
* is always non-NULL inside the handlers. Notice that the vlaue
* active_domain becomes NULL again only when the domain is closed.
* At that time the AM handlers are gone with the PSM endpoint.
*/
fabric_priv->active_domain = domain_priv;
if (psmx2_domain_enable_ep(domain_priv, NULL) < 0)
goto err_out_reset_active_domain;
if (domain_priv->progress_thread_enabled)
psmx2_domain_start_progress(domain_priv);
domain_priv->refcnt = 1;
*domain = &domain_priv->domain;
return 0;
err_out_reset_active_domain:
fabric_priv->active_domain = NULL;
fastlock_destroy(&domain_priv->poll_lock);
err_out_destroy_vl_lock:
fastlock_destroy(&domain_priv->vl_lock);
err_out_delete_mr_map:
rbtDelete(domain_priv->mr_map);
err_out_destroy_mr_lock:
fastlock_destroy(&domain_priv->mr_lock);
err_out_finalize_mq:
psm2_mq_finalize(domain_priv->psm2_mq);
err_out_close_ep:
if (psm2_ep_close(domain_priv->psm2_ep, PSM2_EP_CLOSE_GRACEFUL,
(int64_t) psmx2_env.timeout * 1000000000LL) != PSM2_OK)
psm2_ep_close(domain_priv->psm2_ep, PSM2_EP_CLOSE_FORCE, 0);
err_out_free_domain:
free(domain_priv);
err_out:
psmx2_fabric_release(fabric_priv);
return err;
}
static int psmx_domain_init(struct psmx_fid_domain *domain,
struct psmx_src_name *src_addr)
{
struct psmx_fid_fabric *fabric = domain->fabric;
struct psm_ep_open_opts opts;
int err;
psm_ep_open_opts_get_defaults(&opts);
FI_INFO(&psmx_prov, FI_LOG_CORE,
"uuid: %s\n", psmx_uuid_to_string(fabric->uuid));
if (src_addr) {
opts.unit = src_addr->unit;
opts.port = src_addr->port;
FI_INFO(&psmx_prov, FI_LOG_CORE,
"ep_open_opts: unit=%d port=%u\n", opts.unit, opts.port);
}
err = psm_ep_open(fabric->uuid, &opts,
&domain->psm_ep, &domain->psm_epid);
if (err != PSM_OK) {
FI_WARN(&psmx_prov, FI_LOG_CORE,
"psm_ep_open returns %d, errno=%d\n", err, errno);
err = psmx_errno(err);
goto err_out;
}
FI_INFO(&psmx_prov, FI_LOG_CORE,
"epid: 0x%016lx\n", domain->psm_epid);
err = psm_mq_init(domain->psm_ep, PSM_MQ_ORDERMASK_ALL,
NULL, 0, &domain->psm_mq);
if (err != PSM_OK) {
FI_WARN(&psmx_prov, FI_LOG_CORE,
"psm_mq_init returns %d, errno=%d\n", err, errno);
err = psmx_errno(err);
goto err_out_close_ep;
}
err = fastlock_init(&domain->mr_lock);
if (err) {
FI_WARN(&psmx_prov, FI_LOG_CORE,
"fastlock_init(mr_lock) returns %d\n", err);
goto err_out_finalize_mq;
}
domain->mr_map = rbtNew(&psmx_key_compare);
if (!domain->mr_map) {
FI_WARN(&psmx_prov, FI_LOG_CORE,
"rbtNew failed\n");
goto err_out_destroy_mr_lock;
}
domain->mr_reserved_key = 1;
err = fastlock_init(&domain->poll_lock);
if (err) {
FI_WARN(&psmx_prov, FI_LOG_CORE,
"fastlock_init(poll_lock) returns %d\n", err);
goto err_out_delete_mr_map;
}
/* Set active domain before psmx_domain_enable_ep() installs the
* AM handlers to ensure that psmx_active_fabric->active_domain
* is always non-NULL inside the handlers. Notice that the vlaue
* active_domain becomes NULL again only when the domain is closed.
* At that time the AM handlers are gone with the PSM endpoint.
*/
fabric->active_domain = domain;
if (psmx_domain_enable_ep(domain, NULL) < 0)
goto err_out_reset_active_domain;
if (domain->progress_thread_enabled)
psmx_domain_start_progress(domain);
return 0;
err_out_reset_active_domain:
fabric->active_domain = NULL;
fastlock_destroy(&domain->poll_lock);
err_out_delete_mr_map:
rbtDelete(domain->mr_map);
err_out_destroy_mr_lock:
fastlock_destroy(&domain->mr_lock);
err_out_finalize_mq:
psm_mq_finalize(domain->psm_mq);
err_out_close_ep:
if (psm_ep_close(domain->psm_ep, PSM_EP_CLOSE_GRACEFUL,
(int64_t) psmx_env.timeout * 1000000000LL) != PSM_OK)
psm_ep_close(domain->psm_ep, PSM_EP_CLOSE_FORCE, 0);
err_out:
return err;
}
static int psmx_ns_map_init(void)
{
psmx_ns_map = rbtNew(psmx_service_compare);
return psmx_ns_map ? 0 : -FI_ENOMEM;
}
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 sock_domain(struct fid_fabric *fabric, struct fi_info *info,
struct fid_domain **dom, void *context)
{
struct sock_domain *sock_domain;
struct sock_fabric *fab;
int ret;
fab = container_of(fabric, struct sock_fabric, fab_fid);
if (info && info->domain_attr) {
ret = sock_verify_domain_attr(info->domain_attr);
if (ret)
return -FI_EINVAL;
}
sock_domain = calloc(1, sizeof(*sock_domain));
if (!sock_domain)
return -FI_ENOMEM;
fastlock_init(&sock_domain->lock);
atomic_initialize(&sock_domain->ref, 0);
if (info) {
sock_domain->info = *info;
} else {
SOCK_LOG_ERROR("invalid fi_info\n");
goto err;
}
sock_domain->dom_fid.fid.fclass = FI_CLASS_DOMAIN;
sock_domain->dom_fid.fid.context = context;
sock_domain->dom_fid.fid.ops = &sock_dom_fi_ops;
sock_domain->dom_fid.ops = &sock_dom_ops;
sock_domain->dom_fid.mr = &sock_dom_mr_ops;
if (!info->domain_attr ||
info->domain_attr->data_progress == FI_PROGRESS_UNSPEC)
sock_domain->progress_mode = FI_PROGRESS_AUTO;
else
sock_domain->progress_mode = info->domain_attr->data_progress;
sock_domain->pe = sock_pe_init(sock_domain);
if (!sock_domain->pe) {
SOCK_LOG_ERROR("Failed to init PE\n");
goto err;
}
sock_domain->mr_heap = rbtNew(&sock_compare_mr_keys);
if (!sock_domain->mr_heap) {
goto err;
}
sock_domain->fab = fab;
*dom = &sock_domain->dom_fid;
if (info->domain_attr)
sock_domain->attr = *(info->domain_attr);
else
sock_domain->attr = sock_domain_attr;
sock_dom_add_to_list(sock_domain);
return 0;
err:
free(sock_domain);
return -FI_EINVAL;
}
