[Samba] samba3 upgrade misery

[Curtis Maloney]

Greetings, all

We've been running samba2 for far too long - I'd just never found the time to 
upgrade, and it wasn't a priority since it WORKED.  Our little Solaris9/SPARC64 
box happily served our few gigs of files to 20 users, and played the part of a PDC.

Then I upgraded CUPS, and discovered samba2 doesn't like cups-1.2 libs, and 
figured it was as good an excuse to make the time to upgrade.

I found the online doco about upgrading 2 to 3 (the lack of which previously had 
been another major reason to avoid it).  I downloaded the latest source, and it 
compiled _relatively_ OK.  There was a libtool issue (it assumes you're using 
GNU binutils if you're using GCC, and messes up Solaris' ld, but building static 
fixes that).

The actual upgrade seemed to go very smoothly.  The most burdensome part was 
having to explicitly tell all the client machines the IP of the WINS server - 
even though all the doco I could find said I shouldn't need or want to.

Then the user complaints started.  The network is _slow_.  Files can take 
minutes to load.  I tried eliminating all the variables - disable the virus 
checker, copy the file locally, etc etc.

It's samba.

Checking the logs, I can see why!  All the client machines are reconnecting 
periodically, most likely due to the smbd task choking off sockets:

[2007/08/16 09:13:41, 0] smbd/server.c:open_sockets_smbd(527)
   open_sockets_smbd: accept: Software caused connection abort
[2007/08/16 09:15:16, 0] lib/util_sock.c:set_socket_options(261)
   Failed to set socket option SO_KEEPALIVE (Error Invalid argument)
[2007/08/16 09:15:16, 0] lib/util_sock.c:get_peer_addr(1232)
   getpeername failed. Error was Transport endpoint is not connected
[2007/08/16 09:25:42, 0] smbd/server.c:open_sockets_smbd(527)
   open_sockets_smbd: accept: Software caused connection abort
[2007/08/16 09:36:42, 0] smbd/server.c:open_sockets_smbd(527)
   open_sockets_smbd: accept: Software caused connection abort
[2007/08/16 09:47:50, 0] smbd/server.c:open_sockets_smbd(527)
   open_sockets_smbd: accept: Software caused connection abort

First, the details:
Solaris 9
Samba 3.0.25b
GCC 3.4.6

1) Why are so many socket connections being aborted?  How can I track down the 
cause?

2) Why does set_socket_options have errors for ANY options I specify?  Samba2 
didn't.

3) And what can I say to my boss who keeps asking "What does samba3 give us over 
samba2?"  Because frankly, I'm coming up empty.

-- 
Curtis Maloney
cmaloney at cardgate.net