Re: svn segfault when using svn+ssh schema

[Jon Turney <jon.turney@dronecode.org.uk>]

On 16/01/2017 20:07, Sam Edge wrote:
> On 09/01/2017 19:07, Sam Edge wrote:
>> On 09/01/2017 02:04, Eliot Moss wrote:
>>> On 1/8/2017 3:45 PM, David Rothenberger wrote:
>>>> On 1/8/2017 6:12 AM, Sam Edge wrote:
>>>>> I've seen a number of 'svn segfault' threads on the mailing list
>>>>> archive
>>>>> but none of them seem to cover this specific failure mode.
>>>>> (Apologies if
>>>>> one of them does!)
>>>>>
>>>>> I'm getting segfaults from svn but only when using ssh as the schema.
>>>>> The following three commands are all accessing the same server, the
>>>>> name
>>>>> of which has been changed to protect the guilty.
>>>>>
>>>>> --------
>>>>> SamEdge@thor ~
>>>>> $ ssh cmserver.XXXXXX.com
>>>>> ( success ( 2 2 ( ) ( edit-pipeline svndiff1 absent-entries
>>>>> commit-revprops depth log-revprops atomic-revprops partial-replay
>>>>> inherited-props ephemeral-txnprops file-revs-reverse ) ) )
>>>>> SamEdge@thor ~
>>>>> $ svn ls svn+ssh://cmserver.XXXXXX.com/edl/
>>>>> Segmentation fault (core dumped)
>>> Assuming this is repeatable behavior, perhaps getting a trace using
>>> strace would reveal some more useful information.
>>>
>>> Regards - Eliot Moss
>>>
>> Good idea.
>>
>> Strangely, on another Windows 7 64-bit machine svn+ssh to the same
>> server is working fine for me. The most significant differences are that
>> the 'working' machine is running Windows 7 Pro on an Intel CPU with
>> Sophos AV and domain accounts whereas the 'failing' machine is Windows 7
>> Home Premium on an AMD with MS Security Essentials and local accounts.
>> These differences may or may not be relevant of course.
>>
>> Anyway, attached are the two strace outputs. Diffing them suggests that
>> it's segfaulting in the fork() "system" call.

Not quite, this strace has...

> --- Process 5436 loaded C:\cygwin64\lib\sasl2_3\cygcrammd5-3.dll at 00000003EA090000
> --- Process 5436, exception c0000005 at 00000003EA023780
> --- Process 5436 unloaded DLL at 00000003EA090000
>   333   52465 [main] svn 5436 seterrno_from_win_error: /home/corinna/src/cygwin/cygwin-2.6.1/cygwin-2.6.1-1.x86_64/src/newlib-cygwin/winsup/cygwin/dlfcn.cc:304 windows error 998
>    24   52489 [main] svn 5436 geterrno_from_win_error: windows error 998 == errno 14
>    20   52509 [main] svn 5436 dlopen: ret 0x0

and then a bit later

>    21   91851 [main] svn 5436 fork: entering
>   168   92019 [main] svn 5436 sig_send: sendsig 0x80, pid 5436, signal -72, its_me 1
>    19   92038 [main] svn 5436 sig_send: wakeup 0x28C
>    24   92062 [main] svn 5436 sig_send: Waiting for pack.wakeup 0x28C
>    36   92098 [sig] svn 5436 wait_sig: signalling pack.wakeup 0x28C
>    27   92125 [main] svn 5436 sig_send: returning 0x0 from sending signal -72
> --- Process 5436, exception c0000005 at 0000000180053D3F

So it looks like perhaps cygcrammd5 is faulting during it's DllMain, and 
perhaps this leaves things in a state that dll_list::topsort can't handle...

>> I don't have time in the week to run it in a debug session but I'll try
>> to do so at the weekend given the advice I've been given here. I will
>> also do a rebaseall after I've tracked down the
>>
>> Thanks for the feedback folks. Much appreciated.
>>
>> BR
>
> Hi again.
>
> I've downloaded the debug symbols for subversion & the Cygwin dll and
> run the command in gdb. Output attached. Not really giving me any clues
> so far.

You also need the symbols for the DLL loaded at the faulting address.

 From previous information I'd guess that is cyrus-sasl-debuginfo.

(you can check this using gdb's 'info shared' command)

> I've also run 'rebaseall' on the off chance that it would help - it didn't!
>
> Still open to suggestions if anyone has any "light bulb" notions. :-)

--
Problem reports:       http://cygwin.com/problems.html
FAQ:                   http://cygwin.com/faq/
Documentation:         http://cygwin.com/docs.html
Unsubscribe info:      http://cygwin.com/ml/#unsubscribe-simple