[Samba] Samba over PVFS: Corrupted Data

Justin Mazzola Paluska jmp at MIT.EDU
Mon Dec 19 15:11:10 GMT 2005

On Sat, Dec 17, 2005 at 09:46:08AM -0800, Andrew Bartlett wrote:
> The Samba 3.0.13 would be the first thing I would fix.  Samba 3.0.21 is
> about to be released, and with oplock rewrites and other things since
> 3.0.13, it should provide a better basis for distributed filesystem
> work.  
> My guess is that the lack of posix locking is causing Word to fail, as
> it uses a lot of locks.

I upgraded to 3.0.21rc2 last night and still have the same problems,
though I kept much better records of what's going on, so maybe that

1.  Corruption of PDF files.

    Copying PDF files from my Windows desktop to the Samba share
    results in corruption.  It looks like only the beginning of the
    file is getting corrupted.  For example, the file normally starts

      5 0 obj <<
      /Length 3232
      /Filter /FlateDecode

    But after the copy from the windows desktop, the same file starts

       678 0 729 562 716 0 0 0 0 0 0 0 0 0 0 0 0 511 460 460 511 460 307 460
       511 307 0
       460 256 818 562 511 0 460 422 409 332 537 460 0 0 486 ]
      25 0 obj <<
      /Length1 1997
      /Length2 14184
      /Length3 532
      /Length 15276
      /Filter /FlateDecode

    The binary data that follows is different too.

2.  We can copy Word files to and from the share without any problems.
    However, Word refuses to save to the share.  It reports the error
    "The save failed due to out of memory or disk space."  The share
    has terabytes of space available (as reported by du), so this
    shouldn't be a problem.

3.  We also use Avid -- a video editing program -- that complains
    "Assertion Failed: ReadSize > 0,
    file/coresw/core/filesys/diskrtnsWIN.c,line 444" and then locks

We've run some other tests:

1.  using the same configuration, but writing to an ext3 or xfs
    partition works normally with no problems.

2.  Writing to and from the PVFS partition from within Linux also
    works with no problems.

All of the above lead me to believe that there's something about PVFS
that samba doesn't like.

We have logs for these tests, at log level 3 and at log level 10.
They're quite big, so I can put them on a website instead of posting
them to the list if they would be helpful.


More information about the samba mailing list