Problem
If the SCO OpenServer 5.0 NetBIOS cache is updated using the
dlnbhosts(ADMN) utility with a server "/etc/lmhosts" file that
contains more than 20 entries, the SCO AFPS server fails to respond
to connections. Symptoms include:The AFPS server does not respond to any client connections or to any "net" commands.
The server cannot be stopped. Existing server processes are seen running via the ps(C) command but must be killed. The server cannot be subsequently restarted. For example:
# net start server
The SERVER service is starting................................
......ASU6126: The SERVER service could not be started.
More help may be obtained by typing NET HELPMSG 6126.or:Unable to post the name on the network.
The "elfread" utility reports the following error:
# /usr/net/servers/lanman/bin/elfread -od system |pg...
...
--------------------------------------------------------------
DATE: 2/21/97 EVENT ID: 6035
TIME: 3:26:49PM SOURCE: SERVER
USER: N/A TYPE: Warning
COMPUTER: SCOTEST CATEGORY: NoneDESCRIPTION: ASU6035: Unable to post the name on any network:NOTE: The AFPS server may be started using both the TPI/NetBIOS and NetBEUI transports on system bootup. The server will respond correctly using the NetBEUI transport but if this is stopped, for example using the "#netbeui stop" command, the server will not respond on the remaining NetBIOS transport.
Cause
The SCO OpenServer 5.0 Operating System NetBIOS caching facility uses an internal buffer mechanism to store the "/etc/lmhosts" NetBIOS name and IP address entries. This buffer is set too small. The NetBIOS transport connection to the AFPS server is broken, if this buffer size is exceeded.
NOTE: The NetBIOS cache will be filled statically by "lmhosts" entries as
set in "/etc/lmhosts" via dlnbhosts(ADMN) and dynamically by the
system itself via name resolution on broadcasts. The latter entries,
however, have a finite life in the cache. Solution
Generally, a large amount of entries should not be required to be placed in the server "/etc/lmhosts" file. Entries are only required for other server NetBIOS names to facilitate server-to-server communication, in cases such as account replication to back up controllers in the same domain if on other subnets, and in file replication and trust relationships to other subnets. Entries for client NetBIOS names are also only required for server-to-client communication over different subnets in cases such as alerts, for example, print job notification, or server-to-client printing.
The number of "/etc/lmhosts" entries can be reduced or the NetBIOS buffer size can be increased, by adjusting the setting of "NB_NNSREQ" parameter within the file "/etc/conf/pack.d/nb/space.c". For example, modify:
#define NB_NNSREQ 64to:#define NB_NNSREQ 256NOTE: Increasing this parameter will contribute to an increase in kernel memory usage.
The kernel must be relinked and the system rebooted for this to take effect:
#cd /etc/conf/cf.d
#./link_unixThe system reboot also ensures that the NetBIOS cache is flushed and reloaded.
NOTES: Also run /usr/bin/nbstatus -d /dev/nbcots for more diagnostics.It may be possible that the "sam" database has become corrupted and AFPS may need to be re-configured or even re-installed.
It may be possible that the networking layer has become corrupted. In Single-User mode, try to remove and re-add the networking using "netconfig" and check:
/etc/conf/pack.d/tcp/space.cto ensure that there is the line:
int tcp_fastretx_max = 1;present in the file. The entry should be:int tcp_offer_big_mss = 1;
int tcp_delay_acks = 1;int tcp_fastretx_max = 1;/* default TIME_WAIT period */
int tcp_2msl = 30; See also
ta100037, "What is the "lmhosts" facility and how is it used?"
ta107704, "What is the SLS OSS449F Network Maintenance Supplement?"
ta100029, "The AFPS server is not starting; how can I troubleshoot this problem?"