[karmic] [PATCH] slapd hangs at 100% cpu and is unkillable

[Andreas Hasenack]

slapd-2.4.18-0ubuntu1

Steps to reproduce:
1. apt-get install slapd
2. "ldapmodify -H ldapi:/// -Y EXTERNAL < base.ldif" (file attached to this ticket)
3. the ldapmodify hangs and slapd consumes all available cpu
4. slapd can now only be killed with the -9 signal

I tried to cut down the base.ldif as much as possible. I had a much bigger ldif file before.

I suspect a race or an ordering issue, because removing any one of the overlays makes the ldapmodify work.

[Andreas Hasenack (ahasenack)]

Interesting, looks like ordering is an issue.

In this order, slapd hangs:
adding new entry "olcOverlay=refint,olcDatabase={1}hdb,cn=config"
adding new entry "olcOverlay=syncprov,olcDatabase={1}hdb,cn=config"
adding new entry "o=example,c=us" <---- HANGS

In this other order, it works:
adding new entry "olcOverlay=syncprov,olcDatabase={1}hdb,cn=config"
adding new entry "olcOverlay=refint,olcDatabase={1}hdb,cn=config"
adding new entry "o=example,c=us"

Note: when testing again, you need to clear the database and the config, or just purge slapd and reinstall. Running that ldapmodify command twice in a row won't work.

[Andreas Hasenack (ahasenack)]

I just built a 2.4.19 package for karmic and it exhibits the same problem.

[Chuck Short (zulcss)]

I was able to reproduce this.

Regards
chuck

[Andreas Hasenack (ahasenack)]

It also happens with openldap-2.4.20 which was just released.

[Andreas Hasenack (ahasenack)]

Attached is a gdb backtrace with debugging symbols from the hung process. I'll also attach the output of slapd -d -1, and probably open a bug upstream.

[Andreas Hasenack (ahasenack)]

And now the slapd -d -1 output

[Andreas Hasenack (ahasenack)]

To help upstream to reproduce this bug, I'm also attaching a tarball containing the whole slapd.d contents as they are before running the ldapmodify command that triggers the issue. There is no database yet defined except the cn=config one.

[Andreas Hasenack (ahasenack)]

ITS 6408 filed upstream

[Howard Chu (hyc)]

Fixed in CVS slapd/bconfig.c 1.402

[Andreas Hasenack (ahasenack)]

I applied the diff the Howard mentioned (http://www.openldap.org/devel/cvsweb.cgi/servers/slapd/bconfig.c.diff?r1=1.401&r2=1.402&hideattic=1&sortbydate=0) to 2.4.20 and now it works!

Thanks!

[Andreas Hasenack (ahasenack)]

The same patch also worked in 2.4.18, but it applied with some offsets.

[Chuck Short (zulcss)]

This has been fixed in lucid.

Regards
chuck