nmbd redesign ideas [proxy, userspace sock_struct af_netbios]

[Luke Kenneth Casson Leighton]

On Mon, Dec 13, 2004 at 10:04:25PM +1100, tridge at samba.org wrote:

> nmbd is not in Samba4 yet. When it is added, I expect it will contain
> very little, if any, of the Samba3 code, regardless of who contributed
> the original code.
 
 is there any possibility of adding in a proxying system for 137-138?

 such that, for example, the Wine team can implement the Win32
 NETBIOS API, register NetBIOS names, and samba will "defend" the names
 on behalf of Wine?

 also, so that a NetBIOS name query made _by_ say a future Wine NetBIOS
 API, the answer will actually come back, because it's proxied back via
 samba?

 i.e. is there any possibility to create something like a userland
 netbios socket API, complete with sock_struct af_netbios, such that
 both a future redesign of nmbd _and_ any other projects could benefit
 from much cleaner and simpler code compartmentalisation?


 i did give serious consideration to splitting down nmbd into several
 smaller daemons - one for winsd, one for each NetBIOS nameset
 registered [so one daemon for running as a member of a workgroup
 handling WORKSTATION#0 and WORKGROUP#0, another one for running as a PDC
 and registering WORKSTATION#20 and WORKGROUP#20 etc.]


 if you recall i partially-successfully implemented a NetBIOS proxy in
 1998: the sticking issue i had was whether to multiplex out to several
 eth transports.

 if i _had_ done that, then there was the risk of suffering from exactly
 the same NetBIOS bug that NT suffers from: if i didn't... well, anyway,
 i stopped researching it.
 
 [as you are no doubt aware, the problem with nt is that if you
 have two network cards, you must switch off binding NetBIOS to
 one of them because otherwise you get NetBIOS name registration
 on one interface broadcasting out to two interfaces, the
 stupid code then goes and receives back two responses from
 the UDP broadcast on the two separate interfaces, _both_
 of those responses get fed back up to the NetBIOS layer,
 at which point some idiot's code goes "if num_responses >
 1 then NETBIOS_ERROR(NAME_ALREADY_TAKEN);" ]

 duh.

 l.