[BUG] possibly in tdb

[Luke Kenneth Casson Leighton]

andrew, yesterday's rpctorture test had 50 simultaneous processes making
100 sequential logins to a single server.  number of processes went up
from 30 to 185, and pretty much stayed there.  loadavg up to 5-7.

1/3 way thorugh, i saw a single login fail.  thought nothing of it.

then a whole sequence of fails came through - about 8 in succession.  this
happened frequently.  presumably, some of the processes all failed at the
same time.  they kept sending login requests.

then a _really_ large batch of fails came in.  load averg went up to
10-14, number of processes went down to 150, and the fails kept coming in.

anyway, i started trying to work on testing logins, today, and couldn't. i
saw that the cred_store() function was failing, which is _really_ odd.

i fixed it by theleting netlogoncreds.tdb, and then realised what had
happened because this fixed the problem.

i think that there may be some corruption in tdb, and a large batch of
simultaneous accesses, pushing the limits of this portable, triggered it.

given that i am now heavily reliant on tdb in tng, this is kinda...
critical.

i'll try the torture test again and see if i can repro this.

luke