Red Hat Bugzilla – Bug 19923
nss_ldap has a bug that can deadlock a machine
Last modified: 2007-03-26 23:37:02 EDT
From the firstname.lastname@example.org mailing list, Sat 21 Oct 2000:
Michael Shuey has fixed a race condition in nss_ldap which
caused nscd to lock up under Linux. The patch is in
See bugzilla.padl.com bug #49 for more info.
Luke Howard | email@example.com
PADL Software | www.padl.com
The gist of the bug is that nss_ldap has a lock which it takes. If you are
running nscd, it is possible to lock the entire machine by making all nscd
threads call getpw*.
This triggers a race condition, and locks *ALL* entity lookup - "ls -l",
ps, all login utilies - all will block. *IF* you happen to know the PID of
nscd, you can kill it, otherwise, you'll have to power-cycle the machine.
I could reliably trigger this bug using "tar" on a fast machine - the
lookups for the file username, combined with general system activity,
would lock the machine unrecoverably.
nss_ldap version 121 has a fix for this.
122 has just been placed into the errata testing pipeline. Please verify that
packages at http://people.redhat.com/nalin/test/ (1.6 for RHL 6.1 and 6.2, 1.7
for RHL 7) install and work correctly.