Bug 225485

Summary: Intermittent segvs when prelink enabled
Product: [Fedora] Fedora Reporter: Ed Swierk <eswierk>
Component: kernelAssignee: Jakub Jelinek <jakub>
Status: CLOSED DUPLICATE QA Contact:
Severity: high Docs Contact:
Priority: medium    
Version: 6CC: n3npq
Target Milestone: ---   
Target Release: ---   
Hardware: i386   
OS: Linux   
Whiteboard:
Fixed In Version: Doc Type: Bug Fix
Doc Text:
Story Points: ---
Clone Of: Environment:
Last Closed: 2007-10-09 15:31:39 UTC Type: ---
Regression: --- Mount Type: ---
Documentation: --- CRM:
Verified Versions: Category: ---
oVirt Team: --- RHEL 7.3 requirements from Atomic Host:
Cloudforms Team: --- Target Upstream Version:
Embargoed:
Attachments:
Description Flags
/bin/ping and libraries before prelink -ua
none
/bin/ping and libraries after prelink -ua
none
/bin/ping coredump
none
/bin/su coredump none

Description Ed Swierk 2007-01-30 21:08:06 UTC 
On a few of our FC6 machines, various programs intermittently crash with
SIGSEGV.  The two easiest test cases are:

  while ping -c 1 localhost; do :; done >/dev/null

and

  while sudo su -c true; do :; done >/dev/null

Each fails after a few seconds to a minute.

All the machines in question are running 32-bit i386 FC6, and have
prelink-0.3.9-2, glibc-2.5-10.fc6 installed.  Some were upgraded from FC4, while
others were installed with FC6 from scratch.  Some are single-core Celerons and
others are dual-core Pentium Ds.  There is no apparent correlation between the
machine's configuration and whether the crashing occurs on the machine.

On one machine, I tarred up /bin/ping and its dependent shared libraries, then
disabled prelink and ran "prelink -ua".  After this, the above tests do not crash.
Comment 1 Ed Swierk 2007-01-30 21:08:06 UTC 
Created attachment 146963 [details]
/bin/ping and libraries before prelink -ua
Comment 2 Ed Swierk 2007-01-30 21:09:18 UTC 
Created attachment 146964 [details]
/bin/ping and libraries after prelink -ua
Comment 3 Ed Swierk 2007-01-30 21:10:36 UTC 
This issue seems related to (if not a duplicate of) bug 110665.
Comment 4 Ed Swierk 2007-01-30 21:15:41 UTC 
Actually, the crash occurs on all our FC6 machines.  The ones I thought were
okay are actually still running FC4.
Comment 5 Jakub Jelinek 2007-01-30 21:23:34 UTC 
ping is a PIE, my guess this is a dup of #217614.
What exact glibc do you have installed?  *.i686.rpm or *.i386.rpm?
Can you get a core dump?  ulimit -c unlimited; while ping -c 1 localhost; do :;
done >/dev/null
?  You'd need to run it as root, otherwise setuid ping won't create a core dump.
Comment 6 Ed Swierk 2007-01-30 21:45:34 UTC 
Created attachment 146967 [details]
/bin/ping coredump
Comment 7 Ed Swierk 2007-01-30 21:46:13 UTC 
Created attachment 146968 [details]
/bin/su coredump
Comment 8 Ed Swierk 2007-01-30 21:49:25 UTC 
# rpm -q --queryformat '%{NAME}-%{VERSION}-%{RELEASE}.%{ARCH}\n' glibc iputils
coreutils prelink
glibc-2.5-10.fc6.i686
iputils-20020927-41.fc6.i386
coreutils-5.97-12.3.fc6.i386
prelink-0.3.9-2.i386
Comment 9 Jeff Johnson 2007-04-23 15:49:12 UTC 
Yes, likely dupe of #217614.

I've just managed to figure a "fix" by unprelinking the libraries
that /bin/ping is linked too, cureing a previously observed segfault:

[root@skyline ~]# sh t
completed: 1867
[root@skyline ~]# prelink --undo /lib/libresolv.so.2
[root@skyline ~]# sh t
completed: 97
[root@skyline ~]# sh t
completed: 1746
[root@skyline ~]# prelink --undo /lib/tls/libc.so.6
[root@skyline ~]# sh t
completed: 11705
[root@skyline ~]# prelink --undo /lib/ld-linux.so.2
[root@skyline ~]# sh t
<... still running == likely fixed ...>

Here's the dinky t script:

#!/bin/sh

i=0
while `/bin/ping -c1 -w1 127.0.0.1 >& /dev/null`
do
    i="`expr $i + 1`"
done

echo "completed: $i"
Comment 10 Jeff Johnson 2007-04-23 15:52:25 UTC 
Note that there was another element in #217614, whether glibc.i386 or glibc.i686 is installed.

My comment #9 has these (FC6 kernel, CentOS4.4 glibc&ping) packages installed:

[root@skyline ~]# rpm -q glibc kernel iputils
glibc-2.3.4-2.25.i686
kernel-2.6.18-1.2849prerel3.0.0_60.44.0.i686
iputils-20020927-18.EL4.3.i386
Comment 11 Jeff Johnson 2007-04-23 16:34:50 UTC 
This appears the minimum necessary change to "cure" /bin/ping segfaults:

root@gt40 ~]# sh t
completed: 1076
[root@gt40 ~]# prelink --undo /lib/ld-linux.so.2
[root@gt40 ~]# sh t

< --- running almost an hour --- >

I can reproduce this problem on demand if that helps.
Comment 12 Jakub Jelinek 2007-10-09 15:31:39 UTC 

*** This bug has been marked as a duplicate of 246623 ***