On Wed, Jan 31, 2018 at 2:44 PM, Johan Corveleyn <jcorvel_at_gmail.com> wrote:
> On Wed, Jan 31, 2018 at 9:28 AM, Santosh Kondapuram
> <SKondapuram_at_vitechinc.com> wrote:
>> Hi Johan,
>>
>> Sorry for the delayed response as I was in long vacation.
>> Yes, I don’t think we are hitting the real sha-1 collision in our repository and as you said it might be another bug in the sha-1 collision detection code.
>> I am sure that no one committed the sha-1-colliding files from https://shattered.io
>> Our repository is very huge with 1 Million + revisions and I am seeing the issue only while loading the specific revision 724413. I am able to load the revisions before and after it.
>> As you suggested I have now loaded the dump till 724412 which is the revision before the problematic one with rep sharing enabled.
>> It's not allowing me to load the problematic revision with rep-sharing disabled and getting below message.
>>
>> FYI,
>>
>> [root_at_vitech-svn-slave-test-01 svndump]# svnadmin load /u01/svn/repos < incdump724413.txt
>> svnadmin: E200002: Error while parsing config file: /u01/svn/repos/db/fsfs.conf:
>> svnadmin: E200002: line 39: Option expected
>
> Hm, that indicates that there was a syntax error on line 39 in the
> db/fsfs.conf file you edited (when you were trying to disable
> rep-sharing). Can you take another look and try again with rep-sharing
> disabled? That line in fsfs.conf should be (without leading spaces):
>
> enable-rep-sharing = false
>
> Now, I'm not sure what we'll be able to do next to figure this out.
> First step is probably find out which two files are colliding, and
> what their contents is exactly. It's extremely unlikely that there is
> a real sha1 collision, but we have to rule it out I suppose.
Santosh,
Can you double-check that using -M 0 for the 'svnadmin load' operation
doesn't solve the problem (while keeping rep-sharing enabled)?
So:
svnadmin load -M 0 /u01/svn/repos < incdump724413.txt
It's just that this works around the only currently known bug in the
sha1-collision-detection code, so I want to be really sure that this
workaround doesn't help in your case and we need to look further.
--
Johan
Received on 2018-01-31 15:03:24 CET