[Samba] Windows Errors on Long, Large Writes...

[Clay Goss]

Problem:

Windows machines (some W2K-Pro, some XP-Pro) through a delayed write error
when doing backup to Samba share.

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Configuration:

Fedora Core 3 on Dell PIII-450 with 320MB ram, 100T NIC.

[root at GCPWS2 ~]# df
Filesystem           1K-blocks      Used Available Use% Mounted on
/dev/hda2             12729096   8247728   3834752  69% /
/dev/hda1               101089      9400     86470  10% /boot
none                    160572         0    160572   0% /dev/shm
/dev/hdb1            240362656 150851676  77301180  67% /disk2
[root at GCPWS2 ~]# 

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Logs:

<some-pc-name>.log

[2004/12/28 21:33:09, 0] smbd/nttrans.c:call_nt_transact_ioctl(2276)
  call_nt_transact_ioctl(0x9003c): Currently not implemented.
[2004/12/28 22:40:34, 0] lib/util_sock.c:read_socket_data(384)
  read_socket_data: recv failure for 17424. Error = Connection timed out

Samba.log

[2004/12/28 19:40:01, 0] lib/util_sock.c:get_peer_addr(1000)
  getpeername failed. Error was Transport endpoint is not connected
[2004/12/28 20:40:17, 0] lib/util_sock.c:get_peer_addr(1000)
  getpeername failed. Error was Transport endpoint is not connected
[2004/12/28 21:29:09, 0] lib/util_sock.c:get_peer_addr(1000)
  getpeername failed. Error was Transport endpoint is not connected
[2004/12/29 04:03:48, 1] smbd/server.c:open_sockets_smbd(348)
  Reloading services after SIGHUP
[2004/12/29 07:10:19, 0] lib/util_sock.c:get_peer_addr(1000)
  getpeername failed. Error was Transport endpoint is not connected
[2004/12/29 16:32:54, 0] smbd/server.c:main(760)
  smbd version 3.0.10-1.fc3 started.
  Copyright Andrew Tridgell and the Samba Team 1992-2004
[2004/12/29 18:15:13, 0] lib/util_sock.c:get_peer_addr(1000)
  getpeername failed. Error was Transport endpoint is not connected
[2004/12/29 18:43:38, 0] lib/util_sock.c:get_peer_addr(1000)
  getpeername failed. Error was Transport endpoint is not connected
[2004/12/29 20:49:51, 0] lib/util_sock.c:get_peer_addr(1000)
  getpeername failed. Error was Transport endpoint is not connected
[2004/12/29 23:55:49, 0] lib/util_sock.c:get_peer_addr(1000)
  getpeername failed. Error was Transport endpoint is not connected


~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

Thoughts:

It appears smbd services are being restarted from time to time while the
large transfers are going on and Windows then gives up and throws an error.

Conclusion: 

So, all this brings me to the question, why are samba services restarting?

Any help would be appreciated.

Thanks, 
Clay