Debian Bug report logs - #837123
[anna] segfault in wheezy installer

Reply or subscribe to this bug.

Toggle useless messages

Message #5 received at submit@bugs.debian.org (full text, mbox, reply):

From: Vincent McIntyre <vincent.mcintyre@csiro.au>

To: <submit@bugs.debian.org>

Subject: [anna] segfault in wheezy installer

Date: Fri, 9 Sep 2016 10:27:05 +1000

Package: libc6-udeb
Version: 2.13-38+deb7u10
Severity: grave
Justification: breaks installation entirely

The wheezy installer fails with anna reporting a segfault:

...
anna[5033]: DEBUG: retrieving libc6-udeb 2.13-38+deb7u10
anna[5033]: DEBUG: retrieving finish-install 2.41wheezy1
anna[5033]: DEBUG: Segmentation fault
anna[5033]: WARNING **: package retrieval failed
kernel: [   66.427372] wget[5382]: segfault a 0 ip 00007176c922c0ca sp 00007ffc8ca83890 error 6 in libresolv-2.13.so[7f76c9222000+13000]

This occurs shortly after libc6-udev is downloaded,
according to the installer interactive display.

Background
----------
I was trying to reinstall an older system with wheezy.
I was using the PXE install image available on 2016-09-08 UTC, at
ftp.au.debian.org/debian/dists/wheezy/main/installer-amd64/current/images/netboot

The mirror shows the last update timestamp as 2016-05-31.
debian-installer/amd64/boot-screens/f1.txt has the fairly useless
build date stamp 20130613+deb7u3+b2.

This installation method was working fine with earlier installer
versions. The breakage has been there some months I'd say.
I only became fully aware of it yesterday.

I don't think the mirror is out of date; I got the same result
with the netboot installer image downloaded from ftp.us.debian.org.


The log above was copied by eye from the console,
I was unable to install the openssh-client=udeb to copy the logs,
thanks to the segfault.

The core of the issue is probably screwed up libc6 versions,
see bugs 740068, 833432

Other package versions that may be of interest
 base-installer: 1.130
 cdebconf-udeb: 0.182
 di-utils: 1.92+deb7u1
 libc6-udeb: 2.13-38+deb7u10
 anna: 1.44+deb7u1

It's really unfortunate to have the oldstable installer break like this.
I hope it can be fixed, age notwithstanding.

Kind regards
Vince
-- 

Message #10 received at 837123@bugs.debian.org (full text, mbox, reply):

From: Vincent McIntyre <vincent.mcintyre@csiro.au>

To: <837123@bugs.debian.org>

Cc: <debian-boot@lists.debian.org>

Subject: Re: Bug#837123: ([anna] segfault in wheezy installer)

Date: Fri, 9 Sep 2016 10:39:12 +1000

I just realised this went to the libc maintainers;
I was expecting it would go to the debian-installer team.

This might be an issue in the way libc6-udeb is being used
within debian-installer, rather than libc6-udeb itself.

I don't know how to figure out if that is the case;
if it is the case, please do reassign.

Kind regards
Vince

Message #15 received at 837123@bugs.debian.org (full text, mbox, reply):

From: Aurelien Jarno <aurelien@aurel32.net>

To: Vincent McIntyre <vincent.mcintyre@csiro.au>, 837123@bugs.debian.org

Subject: Re: Bug#837123: [anna] segfault in wheezy installer

Date: Fri, 9 Sep 2016 11:46:30 +0200

Hi,

On 2016-09-09 10:27, Vincent McIntyre wrote:
> 
> Package: libc6-udeb
> Version: 2.13-38+deb7u10
> Severity: grave
> Justification: breaks installation entirely
> 
> The wheezy installer fails with anna reporting a segfault:
>
> ...
> anna[5033]: DEBUG: retrieving libc6-udeb 2.13-38+deb7u10
> anna[5033]: DEBUG: retrieving finish-install 2.41wheezy1
> anna[5033]: DEBUG: Segmentation fault
> anna[5033]: WARNING **: package retrieval failed
> kernel: [   66.427372] wget[5382]: segfault a 0 ip 00007176c922c0ca sp 00007ffc8ca83890 error 6 in libresolv-2.13.so[7f76c9222000+13000]
> 
> This occurs shortly after libc6-udev is downloaded,
> according to the installer interactive display.

Unfortunately I am not able to reproduce the issue here. Given it
involves name resolving, I wouldn't be surprised it depends on the DNS
servers being used.


> Background
> ----------
> I was trying to reinstall an older system with wheezy.
> I was using the PXE install image available on 2016-09-08 UTC, at
> ftp.au.debian.org/debian/dists/wheezy/main/installer-amd64/current/images/netboot
> 
> The mirror shows the last update timestamp as 2016-05-31.
> debian-installer/amd64/boot-screens/f1.txt has the fairly useless
> build date stamp 20130613+deb7u3+b2.
> 
> This installation method was working fine with earlier installer
> versions. The breakage has been there some months I'd say.
> I only became fully aware of it yesterday.
> 
> I don't think the mirror is out of date; I got the same result
> with the netboot installer image downloaded from ftp.us.debian.org.
> 
> 
> The log above was copied by eye from the console,
> I was unable to install the openssh-client=udeb to copy the logs,
> thanks to the segfault.
> 
> The core of the issue is probably screwed up libc6 versions,
> see bugs 740068, 833432

It indeeds looks like that, that said the build logs actually show that
debian-installer 20130613+deb7u3+b2 has been built against glibc
2.13-38+deb7u10, so there is no version skew in this case.

Unfortunately, without being able to reproduce the problem, it will be
very difficult to debug it.

Aurelien

-- 
Aurelien Jarno                          GPG: 4096R/1DDD8C9B
aurelien@aurel32.net                 http://www.aurel32.net

Message #20 received at 837123@bugs.debian.org (full text, mbox, reply):

From: Vincent McIntyre <vincent.mcintyre@csiro.au>

To: Aurelien Jarno <aurelien@aurel32.net>

Cc: <837123@bugs.debian.org>

Subject: Re: Bug#837123: [anna] segfault in wheezy installer

Date: Fri, 9 Sep 2016 23:22:04 +1000

On Fri, Sep 09, 2016 at 11:46:30AM +0200, Aurelien Jarno wrote:
> Hi,
> 
> On 2016-09-09 10:27, Vincent McIntyre wrote:
> > 
> > Package: libc6-udeb
> > Version: 2.13-38+deb7u10
> > Severity: grave
> > Justification: breaks installation entirely
> > 
> > The wheezy installer fails with anna reporting a segfault:
> >
> > ...
> > anna[5033]: DEBUG: retrieving libc6-udeb 2.13-38+deb7u10
> > anna[5033]: DEBUG: retrieving finish-install 2.41wheezy1
> > anna[5033]: DEBUG: Segmentation fault
> > anna[5033]: WARNING **: package retrieval failed
> > kernel: [   66.427372] wget[5382]: segfault a 0 ip 00007176c922c0ca sp 00007ffc8ca83890 error 6 in libresolv-2.13.so[7f76c9222000+13000]
> > 
> > This occurs shortly after libc6-udev is downloaded,
> > according to the installer interactive display.
> 
> Unfortunately I am not able to reproduce the issue here. Given it
> involves name resolving, I wouldn't be surprised it depends on the DNS
> servers being used.

Seriously??? I have to use the exactly right flavour of DNS server?
I think this would have been using microsoft DNS servers.
I preseume you're using BIND?

Could this be a locale thing?

> > The core of the issue is probably screwed up libc6 versions,
> > see bugs 740068, 833432
> 
> It indeeds looks like that, that said the build logs actually show
> that debian-installer 20130613+deb7u3+b2 has been built against
> glibc
> 2.13-38+deb7u10, so there is no version skew in this case.
> 
> Unfortunately, without being able to reproduce the problem, it
> will be very difficult to debug it.

Lovely. Thanks for bothering to reply.
Vince

Message #25 received at 837123@bugs.debian.org (full text, mbox, reply):

From: Aurelien Jarno <aurelien@aurel32.net>

To: Vincent McIntyre <vincent.mcintyre@csiro.au>

Cc: 837123@bugs.debian.org

Subject: Re: Bug#837123: [anna] segfault in wheezy installer

Date: Fri, 9 Sep 2016 15:59:17 +0200

On 2016-09-09 23:22, Vincent McIntyre wrote:
> On Fri, Sep 09, 2016 at 11:46:30AM +0200, Aurelien Jarno wrote:
> > Hi,
> > 
> > On 2016-09-09 10:27, Vincent McIntyre wrote:
> > > 
> > > Package: libc6-udeb
> > > Version: 2.13-38+deb7u10
> > > Severity: grave
> > > Justification: breaks installation entirely
> > > 
> > > The wheezy installer fails with anna reporting a segfault:
> > >
> > > ...
> > > anna[5033]: DEBUG: retrieving libc6-udeb 2.13-38+deb7u10
> > > anna[5033]: DEBUG: retrieving finish-install 2.41wheezy1
> > > anna[5033]: DEBUG: Segmentation fault
> > > anna[5033]: WARNING **: package retrieval failed
> > > kernel: [   66.427372] wget[5382]: segfault a 0 ip 00007176c922c0ca sp 00007ffc8ca83890 error 6 in libresolv-2.13.so[7f76c9222000+13000]
> > > 
> > > This occurs shortly after libc6-udev is downloaded,
> > > according to the installer interactive display.
> > 
> > Unfortunately I am not able to reproduce the issue here. Given it
> > involves name resolving, I wouldn't be surprised it depends on the DNS
> > servers being used.
> 
> Seriously??? I have to use the exactly right flavour of DNS server?
> I think this would have been using microsoft DNS servers.
> I preseume you're using BIND?

I don't talk about the software running on your DNS servers, but rather
how they behave when they get queried. It might depends on many other
things, like if your network has IPv6 or not.

Note that's only one explanation, not sure it's the right one, but given
I am unable to reproduce the issue, it's the only one that come to my
mind.

Aurelien

-- 
Aurelien Jarno                          GPG: 4096R/1DDD8C9B
aurelien@aurel32.net                 http://www.aurel32.net

Message #30 received at 837123@bugs.debian.org (full text, mbox, reply):

From: Vincent McIntyre <vincent.mcintyre@csiro.au>

To: Aurelien Jarno <aurelien@aurel32.net>

Cc: <837123@bugs.debian.org>

Subject: Re: Bug#837123: [anna] segfault in wheezy installer

Date: Sat, 10 Sep 2016 11:30:32 +1000

On Fri, Sep 09, 2016 at 03:59:17PM +0200, Aurelien Jarno wrote:
> 
> I don't talk about the software running on your DNS servers, but
> rather how they behave when they get queried. It might depends on
> many other things, like if your network has IPv6 or not.
> 
> Note that's only one explanation, not sure it's the right one, but
> given I am unable to reproduce the issue, it's the only one that
> come to my mind.

Thank you for explaining. I am struggling with this idea since AFAIK
the DNS servers have not been changed in the last six months or more
so I can think of no reason for the behaviour to have changed.
That history is what makes me think something must have changed in
the installer around the time of the last point release (May).
But from what you say that isn't the case.

You've given me a few things to try out
 - tell DHCP to supply different DNS servers (running bind)
 - make sure all ipv6 related options are disabled
   and no ipv6 DNS entries exist for the target host
 - make sure all ipv6 related options are enabled
   and valid ipv6 DNS entries exist for the target host

Also I can try installing oldstable with the jessie installer.
 
If all of that makes no difference, what would be the next step?
I assume I'd have to build a debug version of the installer
and try to get a backtrace?
My concern is there's a segfault still lurking in the stretch
version. If we can squash it here that fix could be forward-ported.

Kind regards
Vince

Message #35 received at 837123@bugs.debian.org (full text, mbox, reply):

From: Vincent McIntyre <vincent.mcintyre@csiro.au>

To: Aurelien Jarno <aurelien@aurel32.net>

Cc: <837123@bugs.debian.org>

Subject: Re: Bug#837123: [anna] segfault in wheezy installer

Date: Mon, 12 Sep 2016 12:28:55 +1000

On Sat, Sep 10, 2016 at 11:30:32AM +1000, Vincent McIntyre wrote:
> 
> You've given me a few things to try out
>  - tell DHCP to supply different DNS servers (running bind)

   Makes no difference. These servers are not configured for v6,
   while the first ones I used were. There are no v6 entries for
   the host in any case.

>  - make sure all ipv6 related options are disabled
>    and no ipv6 DNS entries exist for the target host

   Didn't try it. The failure happens really early, before the
   preseed file is downloaded.

>  - make sure all ipv6 related options are enabled
>    and valid ipv6 DNS entries exist for the target host
 
   Didn't try
 
> Also I can try installing oldstable with the jessie installer.
  
  This worked. Even with mirror/udeb/suite set to 'stable'.
  For anyone who comes along later, the udeb versions were
    anna 1.52
    base-installer 1.154
    libc6-udeb 2.19-18+deb8u4
    netcfg 1.131+deb8u1

I think this bug can probably be downgraded until someone else can
reproduce it, or at least marked as 'not found' in the later version
of libc6-udeb. If I get time I will try with the daily installer and
report on that.

Vince

Message #40 received at 837123@bugs.debian.org (full text, mbox, reply):

From: Vincent McIntyre <vincent.mcintyre@csiro.au>

To: Aurelien Jarno <aurelien@aurel32.net>

Cc: <837123@bugs.debian.org>

Subject: Re: Bug#837123: [anna] segfault in wheezy installer

Date: Mon, 12 Sep 2016 12:51:23 +1000

On Mon, Sep 12, 2016 at 12:28:55PM +1000, Vincent McIntyre wrote:
> 
> >  - make sure all ipv6 related options are disabled
> >    and no ipv6 DNS entries exist for the target host
> 
>    Didn't try it. The failure happens really early, before the
>    preseed file is downloaded.

Incorrect. The preseed is downloaded ok but it doesn't matter, as far
as I can see, the fault happens before any network configuration
directives in the preseed file can be used.

Vince

Message #45 received at 837123@bugs.debian.org (full text, mbox, reply):

From: Aurelien Jarno <aurelien@aurel32.net>

To: Vincent McIntyre <vincent.mcintyre@csiro.au>, 837123@bugs.debian.org

Subject: Re: Bug#837123: [anna] segfault in wheezy installer

Date: Tue, 13 Sep 2016 23:03:19 +0200

On 2016-09-10 11:30, Vincent McIntyre wrote:
> On Fri, Sep 09, 2016 at 03:59:17PM +0200, Aurelien Jarno wrote:
> > 
> > I don't talk about the software running on your DNS servers, but
> > rather how they behave when they get queried. It might depends on
> > many other things, like if your network has IPv6 or not.
> > 
> > Note that's only one explanation, not sure it's the right one, but
> > given I am unable to reproduce the issue, it's the only one that
> > come to my mind.
> 
> Thank you for explaining. I am struggling with this idea since AFAIK
> the DNS servers have not been changed in the last six months or more
> so I can think of no reason for the behaviour to have changed.

Thanks for the details, that's an important information.

> That history is what makes me think something must have changed in
> the installer around the time of the last point release (May).
> But from what you say that isn't the case.

The installer did change, and has been built against a new libc which
includes changes to the resolver, fixing security issues. I am not aware
of any issue introduced by these patches (this libc is used on regular
systems, not only in debian-installer), except #816669 which involves
IPv6 and has different symptoms.

> You've given me a few things to try out
>  - tell DHCP to supply different DNS servers (running bind)
>  - make sure all ipv6 related options are disabled
>    and no ipv6 DNS entries exist for the target host
>  - make sure all ipv6 related options are enabled
>    and valid ipv6 DNS entries exist for the target host
> 
> Also I can try installing oldstable with the jessie installer.
>  
> If all of that makes no difference, what would be the next step?

What would be interesting would be to try to reproduce the issue in
qemu or virtualbox, with as many things as possible close to your
system.

Also you might want to use the console (alt+f2) to run wget by hand and
see if the issue happen with all hosts or only some of them.

> I assume I'd have to build a debug version of the installer
> and try to get a backtrace?

I don't know if it's really something doable, especially the wheezy
installer use a shared library reduction system to strip some symbols 
from the shared libraries.

> My concern is there's a segfault still lurking in the stretch
> version. If we can squash it here that fix could be forward-ported.

This is very unlikely, as both debian-installer and glibc are quite
different in stretch and do not use the shared library reduction
anymore.

Aurelien

-- 
Aurelien Jarno                          GPG: 4096R/1DDD8C9B
aurelien@aurel32.net                 http://www.aurel32.net

Message #50 received at 837123@bugs.debian.org (full text, mbox, reply):

From: Vincent McIntyre <vincent.mcintyre@csiro.au>

To: Aurelien Jarno <aurelien@aurel32.net>

Cc: <837123@bugs.debian.org>

Subject: Re: Bug#837123: [anna] segfault in wheezy installer

Date: Wed, 14 Sep 2016 09:29:18 +1000

On Tue, Sep 13, 2016 at 11:03:19PM +0200, Aurelien Jarno wrote:

...

> > If all of that makes no difference, what would be the next step?
> 
> What would be interesting would be to try to reproduce the issue in
> qemu or virtualbox, with as many things as possible close to your
> system.
> 

Just to clarify - you mean try to run the installer using a qemu VM
as the target for installation? I can certainly try that.

More details.
The target system is pxe booted and next-server takes it to a (debian)
system running tftpd-hpa. The defaults.cfg has lots of boot targets
but the one I have been testing with is the netboot image, in manual
install mode. The only boot options it is given are 
 'append vga=normal initrd=yadayada'
It also falls over if I feed it a preseed file, where we use
 'append auto=true priority=critical vga=normal initrd=yadayda 
  url=blahdeblah'

> Also you might want to use the console (alt+f2) to run wget by hand and
> see if the issue happen with all hosts or only some of them.

I tried to wget pages from a few web sites from the alt+f2 console.
It segfaulted every time when I used a DNS name in the URL,
but worked if I used an IP address in the URL.
ping does the same thing; segfaults only when using domain names.

If I put an entry in /etc/hosts and try to access that hostname,
wget and ping also segfault, until I add this line to nsswitch.conf:

  hosts: files dns

Then they both work for that hostname.
The only other nsswitch.conf lines are for passwd, group & shadow.

> > My concern is there's a segfault still lurking in the stretch
> > version. If we can squash it here that fix could be forward-ported.
> 
> This is very unlikely, as both debian-installer and glibc are quite
> different in stretch and do not use the shared library reduction
> anymore.

Ok that's good to know.

Message #55 received at 837123@bugs.debian.org (full text, mbox, reply):

From: Vincent McIntyre <vincent.mcintyre@csiro.au>

To: Aurelien Jarno <aurelien@aurel32.net>

Cc: <837123@bugs.debian.org>

Subject: Re: Bug#837123: [anna] segfault in wheezy installer

Date: Wed, 14 Sep 2016 16:08:54 +1000

On Wed, Sep 14, 2016 at 09:29:18AM +1000, Vincent McIntyre wrote:
> > Also you might want to use the console (alt+f2) to run wget by hand and
> > see if the issue happen with all hosts or only some of them.
> 
> I tried to wget pages from a few web sites from the alt+f2 console.
> It segfaulted every time when I used a DNS name in the URL,
> but worked if I used an IP address in the URL.
> ping does the same thing; segfaults only when using domain names.
> 
> If I put an entry in /etc/hosts and try to access that hostname,
> wget and ping also segfault, until I add this line to nsswitch.conf:
> 
>   hosts: files dns
> 
> Then they both work for that hostname.
> The only other nsswitch.conf lines are for passwd, group & shadow.
> 

I was able to netcat busybox & the coredump to a wheezy machine
and got this backtrace, which does not look like it is much help...

% gdb ./ping ./ping.core
GNU gdb (GDB) 7.4.1-debian
...
warning: Can't read pathname for load map: Input/output error.

warning: Could not load shared library symbols for 4 libraries, e.g. /lib/libc.so.6.
Use the "info sharedlibrary" command to see the complete listing.
Do you need "set solib-search-path" or "set sysroot"?
Core was generated by `ping ftp.au.debian.org'.
Program terminated with signal 11, Segmentation fault.
#0  0x00007f826609d0ca in ?? ()
(gdb) bt full
#0  0x00007f826609d0ca in ?? ()
No symbol table info available.
#1  0x0000000000000000 in ?? ()
No symbol table info available.
(gdb) q


Vince
-- 

Message #60 received at 837123@bugs.debian.org (full text, mbox, reply):

From: Aurelien Jarno <aurelien@aurel32.net>

To: Vincent McIntyre <vincent.mcintyre@csiro.au>

Cc: 837123@bugs.debian.org

Subject: Re: Bug#837123: [anna] segfault in wheezy installer

Date: Wed, 14 Sep 2016 08:59:52 +0200

On 2016-09-14 09:29, Vincent McIntyre wrote:
> On Tue, Sep 13, 2016 at 11:03:19PM +0200, Aurelien Jarno wrote:
> 
> ...
> 
> > > If all of that makes no difference, what would be the next step?
> > 
> > What would be interesting would be to try to reproduce the issue in
> > qemu or virtualbox, with as many things as possible close to your
> > system.
> > 
> 
> Just to clarify - you mean try to run the installer using a qemu VM
> as the target for installation? I can certainly try that.

Yes, that's the idea, to see if by using the same network environment
you have the same issue.

> More details.
> The target system is pxe booted and next-server takes it to a (debian)
> system running tftpd-hpa. The defaults.cfg has lots of boot targets
> but the one I have been testing with is the netboot image, in manual
> install mode. The only boot options it is given are 
>  'append vga=normal initrd=yadayada'
> It also falls over if I feed it a preseed file, where we use
>  'append auto=true priority=critical vga=normal initrd=yadayda 
>   url=blahdeblah'

Thanks for the details. Here in a QEMU VM, both work fine.

> > Also you might want to use the console (alt+f2) to run wget by hand and
> > see if the issue happen with all hosts or only some of them.
> 
> I tried to wget pages from a few web sites from the alt+f2 console.
> It segfaulted every time when I used a DNS name in the URL,
> but worked if I used an IP address in the URL.
> ping does the same thing; segfaults only when using domain names.
> 
> If I put an entry in /etc/hosts and try to access that hostname,
> wget and ping also segfault, until I add this line to nsswitch.conf:
> 
>   hosts: files dns
> 
> Then they both work for that hostname.
> The only other nsswitch.conf lines are for passwd, group & shadow.

That means the segmentation fault happen as soon as there is a name
resolution.

Aurelien

-- 
Aurelien Jarno                          GPG: 4096R/1DDD8C9B
aurelien@aurel32.net                 http://www.aurel32.net

Message #65 received at 837123@bugs.debian.org (full text, mbox, reply):

From: Aurelien Jarno <aurelien@aurel32.net>

To: Vincent McIntyre <vincent.mcintyre@csiro.au>

Cc: 837123@bugs.debian.org

Subject: Re: Bug#837123: [anna] segfault in wheezy installer

Date: Wed, 14 Sep 2016 09:01:20 +0200

On 2016-09-14 16:08, Vincent McIntyre wrote:
> On Wed, Sep 14, 2016 at 09:29:18AM +1000, Vincent McIntyre wrote:
> > > Also you might want to use the console (alt+f2) to run wget by hand and
> > > see if the issue happen with all hosts or only some of them.
> > 
> > I tried to wget pages from a few web sites from the alt+f2 console.
> > It segfaulted every time when I used a DNS name in the URL,
> > but worked if I used an IP address in the URL.
> > ping does the same thing; segfaults only when using domain names.
> > 
> > If I put an entry in /etc/hosts and try to access that hostname,
> > wget and ping also segfault, until I add this line to nsswitch.conf:
> > 
> >   hosts: files dns
> > 
> > Then they both work for that hostname.
> > The only other nsswitch.conf lines are for passwd, group & shadow.
> > 
> 
> I was able to netcat busybox & the coredump to a wheezy machine
> and got this backtrace, which does not look like it is much help...
> 
> % gdb ./ping ./ping.core
> GNU gdb (GDB) 7.4.1-debian
> ...
> warning: Can't read pathname for load map: Input/output error.
> 
> warning: Could not load shared library symbols for 4 libraries, e.g. /lib/libc.so.6.
> Use the "info sharedlibrary" command to see the complete listing.
> Do you need "set solib-search-path" or "set sysroot"?
> Core was generated by `ping ftp.au.debian.org'.
> Program terminated with signal 11, Segmentation fault.
> #0  0x00007f826609d0ca in ?? ()
> (gdb) bt full
> #0  0x00007f826609d0ca in ?? ()
> No symbol table info available.
> #1  0x0000000000000000 in ?? ()
> No symbol table info available.
> (gdb) q

Have you installed libc6-dbg on this wheezy machine? Anyway I am afraid
that gdb is confused as the libc6 in debian-installer got some symbols
removed...

Aurelien

-- 
Aurelien Jarno                          GPG: 4096R/1DDD8C9B
aurelien@aurel32.net                 http://www.aurel32.net

Message #70 received at 837123@bugs.debian.org (full text, mbox, reply):

From: Vincent McIntyre <vincent.mcintyre@csiro.au>

To: Aurelien Jarno <aurelien@aurel32.net>

Cc: <837123@bugs.debian.org>

Subject: Re: Bug#837123: [anna] segfault in wheezy installer

Date: Thu, 15 Sep 2016 08:26:50 +1000

On Wed, Sep 14, 2016 at 09:01:20AM +0200, Aurelien Jarno wrote:
> > ...
> > warning: Can't read pathname for load map: Input/output error.
> > 
> > warning: Could not load shared library symbols for 4 libraries, e.g. /lib/libc.so.6.
> > Use the "info sharedlibrary" command to see the complete listing.
> > Do you need "set solib-search-path" or "set sysroot"?
> > Core was generated by `ping ftp.au.debian.org'.
> > Program terminated with signal 11, Segmentation fault.
> > #0  0x00007f826609d0ca in ?? ()
> > (gdb) bt full
> > #0  0x00007f826609d0ca in ?? ()
> > No symbol table info available.
> > #1  0x0000000000000000 in ?? ()
> > No symbol table info available.
> > (gdb) q
> 
> Have you installed libc6-dbg on this wheezy machine? Anyway I am afraid
> that gdb is confused as the libc6 in debian-installer got some symbols
> removed...

I hadn't in the example above but it makes no difference if I have
libc6-dbg installed, the bt is the same.

Vince

Message #75 received at 837123@bugs.debian.org (full text, mbox, reply):

From: Vincent McIntyre <vincent.mcintyre@csiro.au>

To: Aurelien Jarno <aurelien@aurel32.net>

Cc: <837123@bugs.debian.org>

Subject: Re: Bug#837123: [anna] segfault in wheezy installer

Date: Thu, 15 Sep 2016 18:11:37 +1000

[Message part 1 (text/plain, inline)]

On Wed, Sep 14, 2016 at 08:59:52AM +0200, Aurelien Jarno wrote:
> 
> > More details.
> > The target system is pxe booted and next-server takes it to a (debian)
> > system running tftpd-hpa. The defaults.cfg has lots of boot targets
> > but the one I have been testing with is the netboot image, in manual
> > install mode. The only boot options it is given are 
> >  'append vga=normal initrd=yadayada'
> > It also falls over if I feed it a preseed file, where we use
> >  'append auto=true priority=critical vga=normal initrd=yadayda 
> >   url=blahdeblah'
> 
> Thanks for the details. Here in a QEMU VM, both work fine.

I learned today that this whole goose chase started when
a colleague hit this using qemu and bridged networking.
Before today I thought it was real hardware.

I tried this with vmware fusion, setting next-server to the pxe
server used above and the manual/interactive installer target.
I was using NATted networking here; resolv.conf lists only the
vmware nameserver (172.16.119.2).

busybox segfaults at the same time and in the same way.
If I test with wget or ping, I get a segfault unless I use IP
addresses or add the hostname I want to resolve to /etc/hosts.

I tried patching the initrd so that the hostname and lookup
could be pre-configured. /etc/nsswitch.conf comes through
unscathed but /etc/hosts is overwritten. When I hit the segfault
I added the necessary /etc/hosts entry and retried. That worked;
I was able to do a complete installation.

I also figured out how to use netcat to get the log file.
I've attached the syslog up to the first segfault,
with DEBCONF_DEBUG=5.
 
Vince

[syslog.debug=5.gz (application/gzip, attachment)]

Send a report that this bug log contains spam.