[Samba] lib/util_sock.c:read_data(436) ...

[Hannes Seidel]

Hello,

I am administering several Samba PDCs (2.2.7a - 3.0.1) in different networks.
The underlying OS'es are nearly the same on all machines, Debian 3.x and
SuSE 8.x. The machines are also providing the following services to the
clients: ISC dhcpd3, bind9 for dynamic DNS and lprng. Clients are running 
Win2K/XP.

So far, everything works: domain logons, printing, file sharing.

But I realized, that there are a lot of crashes from applications, that
need to keep open connections to the SambaServer. So I browsed the logs
and found the following repeating errors in all my networks:

  [2004/01/30 18:20:12, 0] lib/util_sock.c:read_data(436)
    read_data: read failure for 4. Error = No route to host

The above  message appears (mostly) once a day, on all Clients at different
times in all my networks. The second message below appears randomly. I
think, this happens, when a client machine died:

   read_data: read failure for 4. Error = Connection reset by peer
   [2004/02/03 09:17:02, 0] lib/util_sock.c:read_data(436)

Another issue (I think, it has the same cause) is, that I often lose
printjobs or parts of a printjob. Maybe 95% will be finished, the rest is lost.

So here is, what I tried to get rid of that issue:

- first of all, I thought, it could be something in conjunction with
  "Opportunistic locking" so I disabled it in smb.conf
  (oplocks = false ; level2 oplocks = false)

- Many people said, that the physical connection must be buggy somehow.
  But in 6 different locations with switched 100Mbit twisted-pair network?
  So I exchanged switches, network controllers, downgraded to 10MBit -
  still no improvement.

- I tried to use static IP's on the client machines, because I thought,
  it could be something in conjunction with the renewal of dynamic IP's
  from the DHCP server.

None of the upper actions solved the problem.

Does anyone has an idea?
I'll post a sample smb.conf, when needed.

--
Hannes Seidel
IUNDS AG

h.seidel at iunds.com