Re: more on proff crash building native

[Anders Magnusson <ragge%ludd.ltu.se@localhost>]

David Brownlee wrote:
> On Sat, 14 Feb 2009, Anders Magnusson wrote:
> 
>> Michael L. Hitch wrote:
>> ...
>>>
>>>   It seems that ssh corrupts inbound data.  It's been mentioned a couple
>>> of times on port-vax.  I had seen it quite some time ago when trying to
>>> run a cvs update, but wasn't aware at the time it was an ssh problem
>>> instead of hardware/network.  Running the ssh on the remote system
>>> (inbound through sshd instead of ssh) seems to work, as well as non-ssh
>>> transfers.  Again, I've been collecting information on that problem and
>>> will be submitting another PR.  I hit this again when trying to copy a
>>> test kernel and didn't realize that was the problem for a while.
>>>
>> This unfortunate bug has been around since ~forever :-(  Even really long
>> time ago (> 10 years) this bug would hit me when for example trying to do
>> "cvs co" or "cvs update"; suddenly cvs would bail out complaining about
>> protocol error.   Because of this I usually use rsh instead (also the
>> encryption is not too fast on VAXen).
>>
>> Hm, maybe it is some vax-tailored code in the encryption routines that
>> is incorrect?  Would be an good place to start hunting.
> 
>     Has anyone confirmed it is inbound only - sending lots of data
>     from the vax never hits the issue? Also, does it have the
>     same issue via loopback?
> 
>     I suspect there could be some issue at a lower level which
>     ssh's crc is picking up. Quit a while back I had to diagnose
>     an issue with mozilla having issues downloading mailboxes.
>     Turned out to be an ISDN box which would fixup any invalid
>     checksums on inbound data when performing NAT. Vile piece of
>     electronic money offal.
> 
>     Anyway - the epiphany came when ssh aborted with a CRC error...
> 
In this case ssh is never aborted, so the error occures later than its
CRC check.  And, I have never seen it on outbound traffic, only inbound.

-- Ragge