[Samba] Samba + Linux ext3/FS + As/400 + Quotas

[Nicolas Silvestre]

Hi,
Thank you for your help. You're right for the version : it's 2.2.3.
I checked on the ftp server of SuSE the compile options used to build the 
binaries. "--with-quotas" is mentionned. So, it's strange.
The last chance to go further is to compile Samba on the AS/400 but I'm not 
close to it for one month.

The next part of your mail deals about the strange behaviour of quotas. I also 
made tests on Samba 2.2.8/SuSE Linux 8.1 professional/Client NT4.
I got file truncation, and also differents displays between Linux and WinNT4 due 
to "copy on write" mode on WinNT4 I suppose

Nicolas Silvestre

Johannes Niess wrote:
>>>- samba 2.2.23
> This version does not exist.
> You have to recompile "./configure --with-quota" to get the quota
> displayed as share size. It is not enabled by most vendor binaries.
> 
> You need extensive testing for correctnes. On i386 linux I got silent
> file truncations when exceeding quota. I just copied files with W2k
> explorer. Unfortunately the Samba team could not reproduce it
> (yet). For me it happens even without quota when the real (floppy)
> disk is full.. Quota just triggers it earlier.
> 
> Reason seems to be "sparse" files. Sometimes Windows does a strange
> test for disk space: A new file is opened and seeked to the last byte
> of data to be written. Then the last byte is written and the return
> code used.
> 
> Unix tries to be smart and does a kind of run length encoding for all
> these unused bytes that have been seeked over. This is accounted with
> it's compressed size for quota. So the test succeeds and Windows does
> no further error checking. Samba is bound to Unix quota restrictions
> and the client has no way to reask the status. The  file is truncated
> at quota boundary and ext3 displays full file size to the client.
> 
> Setting "strict allocate = yes" in smb.conf is supposed to help, but
> it does not for me in this case.
> 
> Please report negative _and_ positive test results here to establish
> some statistics. Maybe we can find the pattern where things go wrong.
> 
> Johannes Nieß

-- 
  nicolas.silvestre at aliacom.fr    www.aliacom.fr
  Toulouse : 05 62 19 24 91
  Portable : 06 73 99 62 19
  ParisIdF : 01 48 25 53 13