(ITS#7705) mdb back-end segfaults sporadically with paged searches

[openldap@semyon.org]

Full_Name: Semyon Chaichenets
Version: 2.4.36
OS: Linux 3.2.0-51-generic #77-Ubuntu SMP Wed Jul 24 20:18:19 UTC 2013 x86_64 x86_64 x86_64 GNU/Linux
URL: ftp://ftp.openldap.org/incoming/
Submission from: (NULL) (129.128.11.113)


We have an issue whereby a paged-search operation causes slapd to segfault. At
this point, I can only reproduce the issue on our production data - for which I
cannot share coredumps. I am working on isolating the issue but would appreciate
any help from openLDAP developers.

The issue is triggered by the following search:

ldapsearch -H ldap://ldaphost -b 'ou=people,dc=our,dc=org' -x -E pr=126/noprompt
'(uid=m*)'

We have ~59k entries starting matching uid=m*; the issue occurs unpredictably in
the sense that it may or may not happen for filters matching other sets, but it
can be reproduced reliably for any given filter/page size combination. On our
setup, I could reproduce it for page-sizes 81,84,87,96,112,114,116,126.

The issue is specific to back-mdb; backtrace points to dn2id.c:

#0  0x00007f6d6a5ea6a1 in mdb_idscopes (op=0x7f695c110df0, isc=0x7f69692ca670)
    at ../../../../../servers/slapd/back-mdb/dn2id.c:738
#1  0x00007f6d6a5e399b in mdb_search (op=0x7f695c110df0, rs=0x7f69692dba00)
    at ../../../../../servers/slapd/back-mdb/search.c:747

[..]

I would greatly appreciate any help you could give me with this problem.


Best,
Semyon