New clients stop connecting, whilst existing OK.

[Martin Rootes]

Dear All,

	I'm seeing problems with samba 2.0.6 running on a Sun E450 running Solaris 2.7, what is 
happening is that people will start reporting problems with students attempting to connect
to the server, whilst at the same time existing connections seem to be unaffected apart from
a performance slow down. uptime reports a high load, and doing 'pgrep smbd | wc -l' shows 
~2000 smbd processes running, but smbstatus -b only shows ~200 or so (we can have up to 
1000 connections in normal use). This situation last for about 30 minutes then the extra smbd 
processes die of and the users can then get connections without any problem.

The logs show this kind of activity when the problem is occurring:-

[2000/10/30 16:06:47, 0] lib/util_sock.c:write_socket_data(537)
  write_socket_data: write failure. Error = Broken pipe
[2000/10/30 16:06:47, 0] lib/util_sock.c:write_socket(563)
  write_socket: Error writing 4 bytes to socket 7: ERRNO = Broken pipe
[2000/10/30 16:06:47, 0] lib/util_sock.c:send_smb(751)
  Error writing 4 bytes to client. -1. Exiting
[2000/10/30 16:06:50, 0] lib/util_sock.c:client_addr(1035)
  getpeername failed. Error was Transport endpoint is not connected
[2000/10/30 16:06:50, 0] lib/util_sock.c:write_socket_data(537)
  write_socket_data: write failure. Error = Broken pipe
[2000/10/30 16:06:50, 0] lib/util_sock.c:write_socket(563)
  write_socket: Error writing 4 bytes to socket 7: ERRNO = Broken pipe
[2000/10/30 16:06:50, 0] lib/util_sock.c:send_smb(751)
  Error writing 4 bytes to client. -1. Exiting
[2000/10/30 16:06:52, 0] lib/util_sock.c:client_addr(1035)
  getpeername failed. Error was Transport endpoint is not connected
[2000/10/30 16:06:52, 0] lib/util_sock.c:write_socket_data(537)
  write_socket_data: write failure. Error = Broken pipe

There also seems to be corruption of the log file with lines with multiple dates etc.


My take on it is that new connections are getting as far as firing off a smbd process but
not completing the connection, as students seem perfectly willing to sit there clicking
on the reconnect icon we then get the ~2000 smbd processes appearing causing the
server to slow down and probably causing extra problems. But what I'm not sure what
causes the problem or how to prevent it recurring . Anyone out there with any bright
ideas as to what is happening here? Does it indicate a problem with the Samba server
or could it be caused by problems with the network? Any ideas as to how to prevent it
recurring?

	Martin Rootes
	Systems Support,
	C&IT Services,
	Sheffield Hallam University.

------------------------------------------------------------------------------
Martin Rootes - Senior Systems Programmer/Analyst, Sheffield Hallam University
Email :         M.J.Rootes at shu.ac.uk                      Phone: 0114 225 3828
------------------------------------------------------------------------------