FreeBSD 7.2-BETA1 tcp retransmit crash

[Bjoern A. Zeeb]

On Sun, 5 Apr 2009, Bjoern A. Zeeb wrote:

Hi,

> On Sun, 5 Apr 2009, Blapp, Martin wrote:
>
> Hi Martin,
>
>> Looks like the same problem as PR 129197 (FreeBSD 7 panic)
>> 
>> http://www.freebsd.org/cgi/query-pr.cgi?pr=129197
>> 
>> OS:     FreeBSD 7.2 BETA1
>> PF:      Enabled
>> SACK: net.inet.tcp.sack.enable: 1
>> 
>> Happens after some/many soabort calls ...  I can reproduce it
>> after 3-4 hours running time. Currently I'm testing a workaround
>> but I guess the underlying problem should be fixed.
>
> I had added a few assertions to HEAD last autumn to catch (possibly
> different) but still same panics that I haven't MFCed yet.
>
> I'll try to get the list of revisions together and get you a diff.
> With that I'd expect you to either hit a KASSERT or a pnic.

Here's the patch of what I found was left un-MFCed:

http://people.freebsd.org/~bz/20090405-mfc-r182841-r182842.diff

According to your gdb output you are not getting to sbsndptr() with an
invalid length so it's less likely to hit the KASSERT().

It would be interesting to see if at least the sbsndptr() patch works and
gives you the panic. In that case I'd finally have a positive confirmation
that those changes are right and MFC them;-)


The ways I had found to hit this had been fixed a year ago and
properly MFCed.  This one sounds more like a race and other memory
problem.

What kind of patch are you testing?

/bz

-- 
Bjoern A. Zeeb                      The greatest risk is not taking one.