From: Linus Nordberg <linus@nordu.net>
Date: Tue, 3 Apr 2012 14:56:23 +0000 (+0200)
Subject: Keep track of a failing dynamic server and don't use it while failing.
X-Git-Tag: radsecproxy-1.6-rc0~16
X-Git-Url: http://www.project-moonshot.org/gitweb/?p=libradsec.git;a=commitdiff_plain;h=39e90a8cfdfe5bc95d2d01aa680e328c3476ff33

Keep track of a failing dynamic server and don't use it while failing.

Also, sleep less than 15 minutes (900s), mainly for testing.  This
number will change.

Also, die hard and explicitly if freeing an already freed config in
freeclsrvconf().

This is part of fixing RADSECPROXY-33.
---

diff --git a/radsecproxy.c b/radsecproxy.c
index 88c198f..5248c41 100644
--- a/radsecproxy.c
+++ b/radsecproxy.c
@@ -64,6 +64,7 @@
 #include <libgen.h>
 #include <pthread.h>
 #include <errno.h>
+#include <assert.h>
 #include <openssl/ssl.h>
 #include <openssl/rand.h>
 #include <openssl/err.h>
@@ -1306,6 +1307,8 @@ struct clsrvconf *choosesrvconf(struct list *srvconfs) {
 	server = (struct clsrvconf *)entry->data;
 	if (!server->servers)
 	    return server;
+        if (server->servers->dynfailing)
+            continue;
 	if (!first)
 	    first = server;
 	if (!server->servers->connectionok && !server->servers->dynstartup)
@@ -1344,11 +1347,15 @@ struct server *findserver(struct realm **realm, struct tlv *username, uint8_t ac
 	    pthread_mutex_unlock(&(*realm)->mutex);
 	    freerealm(*realm);
 	    *realm = subrealm;
+            debug(DBG_DBG, "added realm: %s", (*realm)->name);
 	    srvconf = choosesrvconf(acc ? (*realm)->accsrvconfs : (*realm)->srvconfs);
+            debug(DBG_DBG, "found conf for new realm: %s", srvconf->name);
 	}
     }
-    if (srvconf)
+    if (srvconf) {
+        debug(DBG_DBG, "found matching conf: %s", srvconf->name);
 	server = srvconf->servers;
+    }
 
 exit:
     free(id);
@@ -1751,18 +1758,25 @@ void *clientwr(void *arg) {
 
     conf = server->conf;
 
+#define ZZZ 60
+
     if (server->dynamiclookuparg && !dynamicconfig(server)) {
 	dynconffail = 1;
 	server->dynstartup = 0;
-	sleep(900);
+	server->dynfailing = 1;
+	debug(DBG_WARN, "%s: dynamicconfig(%s) failed, sleeping %ds",
+              __func__, server->conf->name, ZZZ);
+	sleep(ZZZ);
 	goto errexit;
     }
 
+    /* FIXME: Is resolving not always done by compileserverconfig(),
+     * either as part of static configuration setup or by
+     * dynamicconfig() above?  */
     if (!resolvehostports(conf->hostports, conf->pdef->socktype)) {
-	debug(DBG_WARN, "clientwr: resolve failed");
-	server->dynstartup = 0;
-	sleep(900);
-	goto errexit;
+        debug(DBG_WARN, "%s: resolve failed, sleeping %ds", __func__, ZZZ);
+        sleep(ZZZ);
+        goto errexit;
     }
 
     memset(&timeout, 0, sizeof(struct timespec));
@@ -1775,8 +1789,11 @@ void *clientwr(void *arg) {
     if (conf->pdef->connecter) {
 	if (!conf->pdef->connecter(server, NULL, server->dynamiclookuparg ? 5 : 0, "clientwr")) {
 	    if (server->dynamiclookuparg) {
-		server->dynstartup = 0;
-		sleep(900);
+                server->dynstartup = 0;
+		server->dynfailing = 1;
+                debug(DBG_WARN, "%s: connect failed, sleeping %ds",
+                      __func__, ZZZ);
+		sleep(ZZZ);
 	    }
 	    goto errexit;
 	}
@@ -2173,10 +2190,17 @@ struct list *createsubrealmservers(struct realm *realm, struct list *srvconfs) {
 		debug(DBG_ERR, "malloc failed");
 		continue;
 	    }
+            debug(DBG_DBG, "%s: copying config %s", __func__, conf->name);
 	    *srvconf = *conf;
+            /* Shallow copy -- sharing all the pointers.  addserver()
+             * will take care of servers (which btw has to be NUL) but
+             * the rest of them are shared with the config found in
+             * the srvconfs list.  */
 	    if (addserver(srvconf)) {
 		srvconf->servers->dynamiclookuparg = stringcopy(realm->name, 0);
 		srvconf->servers->dynstartup = 1;
+                debug(DBG_DBG, "%s: new client writer for %s",
+                      __func__, srvconf->servers->conf->name);
 		if (pthread_create(&clientth, NULL, clientwr, (void *)(srvconf->servers))) {
 		    debugerrno(errno, DBG_ERR, "pthread_create failed");
 		    freeserver(srvconf->servers, 1);
@@ -2544,6 +2568,9 @@ int setttlattr(struct options *opts, char *defaultattr) {
 }
 
 void freeclsrvconf(struct clsrvconf *conf) {
+    assert(conf);
+    assert(conf->name);
+    debug(DBG_DBG, "%s: freeing %p (%s)", __func__, conf, conf->name);
     free(conf->name);
     if (conf->hostsrc)
 	freegconfmstr(conf->hostsrc);
diff --git a/radsecproxy.h b/radsecproxy.h
index 680a58d..dc09b1e 100644
--- a/radsecproxy.h
+++ b/radsecproxy.h
@@ -160,6 +160,7 @@ struct server {
     uint8_t connectionok;
     uint8_t lostrqs;
     uint8_t dynstartup;
+    uint8_t dynfailing;
     char *dynamiclookuparg;
     int nextid;
     struct timeval lastrcv;