Hello,
This bug/leak becomes noticeable (and pretty harmful) when you make
frequent updates of the databases and have a long uptime.
The patch below is not suggested as-is, rather as an illustration.
It is though syntactically and semantically correct AFAICT.
It looks like the "checking" code was written with server notification
via a signal in mind and then lost the corresponding fclose() when the
signal sending part was replaced with an RPC.
----
--- coda-src/update/updateclnt.cc.ori 2014-06-27 12:38:19.271390517 +0200
+++ coda-src/update/updateclnt.cc 2014-06-27 12:48:03.034448819 +0200
@@ -245,6 +245,8 @@
if (nservers != 1)
vice_dir_init (vicedir, i+1);
/* signal file server to check data bases */
+/* here we badly leaked open file descriptors without any reason -- rl */
+#if 0 /* this check is irrelevant */
file = fopen(vice_file("srv/pid"), "r");
if (file == NULL) {
LogMsg(0, SrvDebugLevel, stdout,
@@ -252,6 +254,9 @@
vice_file("srv/pid"),
ViceErrorMsg(errno));
} else {
+#else /* just go ahead and try binding */
+ {
+#endif
RPC2_Handle rpcid;
if (U_BindToServer(hostlist[i], &rpcid) == RPC2_SUCCESS) {
if (VolUpdateDB(rpcid) == RPC2_SUCCESS) {
----
Regards,
Rune
