BIND 4.8[.1] TCP domain errors

[Mark Tinguely]

 We have been having a frequent BIND failure on both our VAX and
 Solbourne that is traced to TCP domain queries from an IBM NSMAIN
 nameserver running in cache mode (UDP queries do not cause this
 problem, though it is usually a UDP resolution that is active upon the
 crash -- this resolution is an innocent victim).

 I have discovered that something is trashing the hash areas (sometimes
 even as it is being recursively used in a resolution). Also, occasionally
 the socket/file descriptor for the TCP connection is changed to invalid
 entries causing a reply write fail (though this is not necessarially fatal,
 and the rest of the structure is not apparently altered).

 One dump I have points out in nlookup (dblookup.c), we pass, but do not
 modify the double pointer htpp, one of the first steps is to do a
 one level reference and put the answer in htp. Between the second and
 third call to nlookup, contents of *htpp (htp) changes. The only thing
 that could change this value (other than evil gods), is a signal processing
 routine, but the signal processing routines just set a flag and exit.

 Has any one else have frequent BIND failures (especially major domain
 sites that have heavy TCP domain loads).

-- 
Mark Tinguely           North Dakota State University,  Fargo, ND  58105
  UUCP:       		...!uunet!plains!tinguely
  BITNET:      		tinguely at plains.bitnet
  INTERNET:   		tinguely at plains.NoDak.edu