Please provide stack traces for each of these assertions. In this case it sounds like connection_closing() got called multiple times for the same connection, after it had already been fully closed. A trace would help confirm that.
This is a little odd:
(gdb) run -d 255 Starting program: /usr/sbin/slapd -d 255 must compile with LDAP_DEBUG for debugging
*** "CRASH" ***
Program exited normally. (gdb) ba No stack.
I have been running a script, while ldapsearch -x; do sleep 10; done, that eventually crashes slapd. But gdb provides little feedback, as noted above.
One things to note is that, in addition to being on MIPS32, my platform has only 32MB of memory. I also have a swap file that is 64MB.
In such an environment you won't have enough memory for more than 1 or 2 server threads. Have you already configured this?
I have just reconfigured slapd to use one thread and removed all indexes from the configuration. So far, I have not seen a crash. I am no using two clients to query slapd and have removed that sleep statements so that the server is continuously queried. I've been running this for about one hour and will report back if there are any negative results.