smdb warp-around after 4 GB

[Jeremy Allison]

On Tue, Nov 06, 2001 at 05:40:40PM +0100, Andreas.Karrer at Switzerland.org wrote:
> 
> Hi Jeremy,
> 
> >Exactly what form does the corruption take ? I've transferred several
> >>4GB files between W2k clients and Samba servers as part of the tests
> >in shipping a release.
> 
> When I monitor the growing file on the (Solaris) Samba server with
> "ls -lsn" every second, I get around the critical size:
> 
> 8377920 -rwxr--r--   1 1042   210   4287384716 Nov  6 17:00 bigfile
> 8383520 -rwxr--r--   1 1042   210   4290310796 Nov  6 17:00 bigfile
> 8389344 -rwxr--r--   1 1042   210   4293236876 Nov  6 17:00 bigfile
>      50 -rwxr--r--   1 1042   210        25228 Nov  6 17:00 bigfile [1]
>     176 -rwxr--r--   1 1042   210   4295122944 Nov  6 17:00 bigfile [2]
>     688 -rwxr--r--   1 1042   210   4296748172 Nov  6 17:00 bigfile
>    1088 -rwxr--r--   1 1042   210   4298373772 Nov  6 17:00 bigfile
>    1712 -rwxr--r--   1 1042   210   4301365248 Nov  6 17:00 bigfile
>    2160 -rwxr--r--   1 1042   210   4302730380 Nov  6 17:00 bigfile
>    2336 -rwxr--r--   1 1042   210      9193612 Nov  6 17:00 bigfile [3]
>    2832 -rwxr--r--   1 1042   210   4305461388 Nov  6 17:00 bigfile [4]
>    2960 -rwxr--r--   1 1042   210     11729548 Nov  6 17:00 bigfile [5]
>    3456 -rwxr--r--   1 1042   210   4307997324 Nov  6 17:00 bigfile [6]
>    3760 -rwxr--r--   1 1042   210   4309232780 Nov  6 17:00 bigfile
> 
> So apparently the file is overwritten from BOF [1], but shortly
> afterwards we have a sparse file of approximately the right EOF offset
> [2], but which occupies only a few MB on disk.
> Some time later, the EOF jumps back and forth again ([3] .. [6]).
> 
> To me, this looks as if the file was truncated to (size % 4294967296)
> but the extended to (size) again.

I have checked in a fix for this in 2.2.x and HEAD. Please CVS
update and try it again. It works here.

This one was *nasty* to fix......

Don't ask :-).

Jeremy.