[Samba] Samba over PVFS: Corrupted Data

Justin Mazzola Paluska jmp at MIT.EDU
Thu Jan 5 19:48:02 GMT 2006

On Tue, Dec 20, 2005 at 08:08:24AM -0800, Jeremy Allison wrote:
> Ok, what you need is to create files locally on the windows client
> of 512 bytes, 1024 bytes etc. and then just copy them onto the
> Samba drive. Check what the smallest size is that the first file becomes
> corrupted. That will show the simplest problem that reproduces the
> issue. At *that* point then send in logs etc. Without an *exact*
> method of reproducing this with detailed information on what is
> going wrong (ie. the last 20 bytes of a 512 byte file get corrupted,
> but a 256 byte file is fine - here is the comparitive copy) it will
> be a massive investment of time to track this down, that we currently
> don't have. You're going to have to do more work on this I'm afraid.

OK, I made a bunch of files consisting of the string
"123456789abcdef\n" where \n is the UNIX newline (ASCII 0xA) and
performed the binary search.  Files 64KB (=65536 bytes) and higher are
corrupted and files under 64KB are fine.

In the attached tarball area the results of my tests for two files,
one 64KB in length and one 65535B (64K-1B) in length.  In more detail:

 - samba logs with the log level = 10

 - original files

 - files as put on the share by our Windows client.  The 64k file from
   the share starts to differ from the original file at offset
   0xF000.  The file from the share is filled with 0 bytes.

 - whole-packet ethereal traces.

 - the Samba configuration for this test.  The client was accessing
   the Cluster share.

I did these tests with just-released Samba 3.0.21a on a Mandrake
2005LE (10.2) system.

Sorry for the delay in getting this extra information -- holidays and
such.  If there is any more information you need, please let me know!

More information about the samba mailing list