[Samba] Re: How Samba let us down

[John H Terpstra]

On Wed, 23 Oct 2002, Jay Ts wrote:

> John H Terpstra wrote:
> >
> > For the record, I thouroughly test samba pre-releases before we ever ship.
> > To the best of my knowledge, NOT ONE version of samba we have released
> > ever CAUSED (or resulted in) file/data corruption. If I sound defensive -
> > that's is exactly correct because file corruption is a DEATH issue!
>
> Ok, first of all, check my wording carefully:
>
> > > I've seen problems in the early 2.2.x releases (when transferring
> > > large files) that could be perceived as (or called) "file corruption",
> > > but the problem went away sometime before 2.2.4.
>
> The key words are "perceived" and "called". The idea there was that
> the guy who was having problems was referring to data corruption,
> while not saying _exactly_ what the problems were. Maybe there can
> be things that some people would call "data corruption" that actually
> don't result in loss of data.

Please consider how people use information using web searches. If they are
at all concerned about data integrity and a search pulls up incident
reports claiming data corruption with samba, then the rule of "Guilt by
association" may prevail. Witness perceptions have been enough to get
someone convicted of murder. Please understand our extreem sensitivity to
any reflection on data integrity.

>
> My own experience was that when working with early 2.2.x releases,
> (I think it was 2.2.2 or earlier).  Every once in a while there would
> be an error while copying large files over the network.  This happened
> maybe 5 times within a period of some months. It happened a couple
> of times while I was copying CD-ROM (iso9660) images from Windows
> 98 to Linux/Samba, and about 4 times while installing an application
> which had it's install CD residing on a Samba share.  (I got tired
> of searching for installation CDs every time I reinstalled Windows,
> so threw them on the server's hard disk.)
>
> In each case, I "fixed" the problem simply by redoing the operation,
> and it succeeded on the next try.  And the problems haven't happened
> since upgrading to 2.2.3, IIRC.  I haven't seen any problems at
> all in a long time.

Something was causing the problem. It is important to duplicate the
problem and then bring to the attention of the samba-team how to
demonstrate the problem. We will drop everything to fix any significant
problem - and nothing is more important than data integrity.

>
> BTW, I remember reading messages on the samba list from other
> users who were reporting similar problems around the same time.
>
> > I can make NO assertions regarding the integrity of smbfs as I
> > regard this as most undesirable technology.
>
> Wow, this is news to me.  Why?  And do your comments also apply
> to Steve French's Linux CIFS filesystem?

Because NT and upwards has user level security. SMBFS shares the data from
MS Windows as a single user ID (SID on the NT system). While it may be
"nice" to be able to share data this way, it is not architecturally the
most desirable way to do it. Steve French is doing a great job at making
this technology more mature and more usable. The current smbfs is in my
opinion NOT something that should be widely deployed for multiple
concurrent Linux/Unix user access.

>
> My experience here is that smbfs isn't perfect, but works pretty
> well, and I *really* like it!  Without smbfs, I end up having to
> run to the Windows system to transfer files. (Sorry, but smbclient
> just doesn't "do it" for me. It works, but is really inconvenient.)
>
> > Every reported case of file corruption I have looked at has been due to:
> >
> > 	1. Bad or defective or low grade ethernet cards
>
> I'm using 3com 3c905's on the systems in question, along with ...
>
> > 	2. Defective HUBs / Ether-Switches
>
> Netgear FS-105 5-port fast ethernet switch.
>
> > 	3. Defective Hardware on the Server
>
> No problems outside of those I described. IIRC, the two systems
> were Athlon/1GHz and Pentium/200, with ASUS A7V and Intel motherboards.
>
> > 	4. Incorrect Protocol Stack configuration on the MS Windows client
>
> I just set up the Windows system as usual, with IP address, workgroup,
> computer name, WINS and DNS servers, etc.  But, being Windows 98,
> the installation that was on the system might well have been corrupted
> somehow!  That would not be unusual, after all.

Are you installing multiple protocols? If so the load order is critical.
If you want integrity you need to make absolutely certain that your LANA
adaptor ordering is identical on all clients. If not, you can expect
problems.

- John T.
-- 
John H Terpstra
Email: jht at samba.org