| Age | Commit message (Collapse) | Author | Files | Lines |
|---|
|
server. The real problem is all the special cases we had for when we
are a wins server as opposed to when we are using a 'real' wins
server.
This patch removes the special cases. We now accept non-broadcast
packets from ourselves and we use ourselves as a wins server when we
are one. This gets rid of the special cases and simplifies things
quite a bit.
It all seems to work, but there are bound to be problems found later.
(This used to be commit 3e843d30158d05cdfba716bac7e5c0a75ae7a79c)
|
|
(This used to be commit 9a3e323ec261a1ee3a83f8c558583c3d4a53e06a)
|
|
We now cope wiith multiple WINS groups and multiple failover servers
for release and refresh as well as registration. We also do the regitrations
in the same fashion as W2K does, where we don't try to register the next
IP in the list for a name until the WINS server has acked the previos IP.
This prevents us flooding the WINS server and also seems to make for much
more reliable multi-homed registration.
I also changed the dead WINS server code to mark pairs of IPs dead,
not individual IPs. The idea is that a WINS server might be dead from
the point of view of one of our interfaces, but not another, so we
need to keep talking to it on one while moving onto a failover WINS
server on the other interface. This copes much better with partial
LAN outages and weird routing tables.
(This used to be commit 313f2c9ff7a513802e4f893324865e70912d419e)
|
|
accept an extended syntax for 'wins server' like this:
wins server = group1:192.168.2.10 group2:192.168.3.99 group1:192.168.0.1
The tags before the IPs don't mean anything, they are just a way of
grouping IPs together. If you use the old syntax (ie. no ':') then
an implicit group name of '*' is used. In general I'd recommend people
use interface names for the group names, but it doesn't matter much.
When we register in nmbd we try to register all our IPs with each group
of WINS servers. We keep trying until all of them are registered with
every group, falling back to the failover WINS servers for each group
as we go.
When we do a WINS lookup we try each of the WINS servers for each group.
If a WINS server for a group gives a negative answer then we give up
on that group and move to the next group. If it times out then
we move to the next failover wins server in the group.
In either case, if a WINS server doesn't respond then we mark it dead
for 10 minutes, to prevent lengthy waits for dead servers.
(This used to be commit e125f06058b6b51382cf046b1dbb30728b8aeda5)
|
|
(This used to be commit d03efabc4dca842cafcd0edfa1eaa5b4d3e821b6)
|
|
gives us a good grounding to properly support multiple wins servers
for different interfaces (which will be coming soon ...)
- fixed our wins registration failover code to actually do failover!
We were not trying to register with a secondary wins server at all
when the primary was down. We now fallback correctly.
- fixed the multi-homed name registration packets so that they work
even in a non-connected network (ie. when one of our interfaces is not
routable from the wins server. Yes, this really happens in the real
world).
(This used to be commit a049360d5b0d95a935b06aad43efc17d34de46dc)
|
|
Changed "SMB/Netbios" to "SMB/CIFS" in file header.
(This used to be commit 6a58c9bd06d0d7502a24bf5ce5a2faf0a146edfa)
|
|
and replaced with two functions:
void zero_ip(struct in_adder *ip);
BOOL is_zero_ip(struct in_addr ip);
(This used to be commit 778f5f77a66cda76348a7c6f64cd63afe2bfe077)
|
|
(This used to be commit 60e907b7e8e1c008463a88ed2b076344278986ef)
|
|
My plan is to change the lp_wins_server() function to lp_wins_server_list().
My reason being: With WINS failover the 'wins server' parameter may take a
list of WINS server names/IPs instead of just one. If it's a list, then
calling lp_wins_server() won't give you what you expect (that is, a single
WINS server name or IP). Instead, the functions in wins_srv.c should be
used. You can get either the name or IP of the 'current' working WINS
server in the list.
Chris -)-----
(This used to be commit efaa9ef5e72c3748d97e43bba38a0ba47ff3438d)
|
|
Jeremy.
(This used to be commit 448ff58fd5a88b1fdadfb9a8e0bc5f38b707a4d3)
|
|
string), the wins_srv module now hands back a struct in_addr when it's
called. It caches the IP address once it has been looked up. The IP
is cleared (and must be looked up again) if the 'wins server' parameter
is reread, or if the node is marked 'dead'. A dead node will not be
re-tried for 10 minutes (per a #define in wins_srv.c).
As it was, the code was reading the WINS server name or IP directly from
lp_wins_server. That's okay, except that if the value was expressed as
a name, then a DNS lookup would be done every time the client wanted to
talk to the server.
I still need to work out the implications of failover regarding the
'unicast subnet' list.
Chris -)-----
(This used to be commit 73aa188320fd3bf10b5dfc057323f40aff2c13bd)
|
|
*Note: failover doesn't actually work yet!* It's just that the code I'm
adding provides all of the pieces necessary.
I do have one big question. Something that I'll have to ask Jeremy, I'm
thinkin'. In nmbd/nmbd_subnetdb.c the IP of the WINS server is used to
set up the Unicast subnet.
...so what happens if the WINS server changes?
My guess is either:
a) nothing.
b) I'd have to change the unicast subnet entry whenever the WINS server
changes.
Urq.
BTW, the lp_wins_server() function no longer returns the WINS server name
or IP. It returns the list of WINS servers entered in smb.conf. To get
the currently 'live' WINS server, use the wins_srv() function.
Fun, eh?
Chris -)-----
(This used to be commit cc08bdc74f4cd111fdc582ee7babef47ed8a950d)
|
