[Pkg-openssl-devel] Bug#665836: Bug#665836: openssh-server: segfault error 6 in libcrypto.so.1.0.0

[Michael Welsh Duggan]

Kurt Roeckx <kurt at roeckx.be> writes:

> On Mon, Mar 26, 2012 at 05:22:36PM -0400, Michael Welsh Duggan wrote:
>> > Can you attach gdb to the sshd, and then make it crash and
>> > send me a backtrace?
>> 
>> Unfortunately, no.  Here is the sequence of events I tested:
>> 
>> 1) I re-installed libssl-1.0.0 version 1.0.1-1
>> 2) I started gdb on /usr/bin/sshd as root
>> 3) I ran sshd from gdb using -d -p 2222
>> 4) I connected to that port from my Redhat box.
>> 5) I started an xterm.
>> 6) I exited the exterm with ^D.
>
> Which is exactly what I tried before.
>
>> At the instant step 6 happened, I got an ordinary shutdown of sshd from
>> within gdb, with the following messages:
>> 
>> debug1: do_cleanup
>> debug1: PAM: cleanup
>> debug1: PAM: closing session
>> debug1: PAM: deleting credentials
>> debug1: session_pty_cleanup: session 0 release /dev/pts/5
>> [Inferior 1 (process 21113) exited with code 0377]
>> 
>> In my syslog, I get the following line:
>> 
>> Mar 26 17:09:47 maru kernel: sshd[21123]: segfault at 800ef000 ip
>> b7e2d678 sp bfffeec0 error 6 in libcrypto.so.1.0.0[b7dc4000+1a3000]
>
> 21123 != 21113.
>
> Notice that it says this in the log:
>> User child is on pid 21123
>
> You need to attach to the right pid.

Mea culpa.  Digits looked too similar.  I have a backtrace, but it's not
really useful without symbols.  I didn't see a -dbg package, so I guess
I have to rebuild with debugging turned on.  I'll start working on that,
but if you have a better solution, let me know.

-- 
Michael Welsh Duggan
(mwd at cert.org)