Bug 6779 - Domain browsing loses entries from large browse lists
Summary: Domain browsing loses entries from large browse lists
Status: NEW
Alias: None
Product: Samba 3.3
Classification: Unclassified
Component: Nmbd (show other bugs)
Version: 3.3.6
Hardware: x86 Linux
: P3 normal
Target Milestone: ---
Assignee: Jeremy Allison
QA Contact: Samba QA Contact
Depends on:
Reported: 2009-10-04 18:55 UTC by Lesley Walker
Modified: 2009-10-15 19:18 UTC (History)
0 users

See Also:

Packet trace of browse list failure (76.13 KB, application/cap)
2009-10-04 18:58 UTC, Lesley Walker
no flags Details
Level 10 log from BDC (975.81 KB, application/octet-stream)
2009-10-04 18:59 UTC, Lesley Walker
no flags Details
Level 10 log from domain master browser (289.52 KB, application/octet-stream)
2009-10-04 19:02 UTC, Lesley Walker
no flags Details
Configs (4.07 KB, application/x-gtar)
2009-10-04 19:03 UTC, Lesley Walker
no flags Details

Note You need to log in before you can comment on or make changes to this bug.
Description Lesley Walker 2009-10-04 18:55:45 UTC
When the domain browse list is too large to fit into a 64k packet, a Windows PC sends NetServerEnum3 request, and the Samba server responds with a "malformed packet".

The browse list on our (Samba) PDC is currently over 2700 entries. (This numbers was obtained by sending a -1 signal to nmbd and retrieving info from the log). When the (Samba) BDCs run the sync_with_dmb function, only part of the list can be retrieved. The BDCs have approx 2100-2300 entries.

I found a commit log entry from two years ago:
This seems to imply that NetServerEnum3 has been implemented in the Samba client code, but it is clearly not working in the server code.

I have tried turning on enhanced browsing, but it made no difference.

We are taking steps to reduce the size of the browse list (it's mostly PCs that don't need to be listed) but in the meantime we have problems because certain hosts that need to visible domain-wide are not.

I intend to attach an example of a packet trace showing the replication failure, level 10 logs from a BDC and from the PDC, and also the full config information. The logs and packet trace are from separate replication instances, not simultaneous. Please let me know if there's any further information you require.
Comment 1 Lesley Walker 2009-10-04 18:58:53 UTC
Created attachment 4779 [details]
Packet trace of browse list failure

This was captured on a BDC (hostname = CTSV01) and shows an attempt to synchronise with the PDC (hostname = COM).
Comment 2 Lesley Walker 2009-10-04 18:59:44 UTC
Created attachment 4780 [details]
Level 10 log from BDC
Comment 3 Lesley Walker 2009-10-04 19:02:00 UTC
Created attachment 4781 [details]
Level 10 log from domain master browser

This was captured on the PDC/Domain master browser (hostname = COM) and shows a complete instance of attempted replication to a BDC (hostname = MLSV01).
Comment 4 Lesley Walker 2009-10-04 19:03:43 UTC
Created attachment 4782 [details]

Gzipped tar file of the configs from both the PDC and a typical BDC.
Comment 5 Lesley Walker 2009-10-15 19:18:08 UTC
Updated from 3.3.6 to 3.3.7, this behaviour still occurs.