I did some further modifications to vms_tdb.c/tdb.h, available at
http://www.er6.eng.ohio-state.edu/~jonesd/samba/vms_tdb.zip.
Changes:
- Increase the maximum record size to 2036 bytes.
- Fixed a bug in the TDB_INTERNAL mode operation (only the [.locking]posix.c
module appears to use an 'internal' tdb.
- Save the last key as well as the last hash in the TDB_CONTEXT structure.
The optimization in fetch() now checks the last key is the same as
well as the last hash.
- The TDB_DATA items returned by the external functions now point to
dynamically allocated copies of the data rather have the dptr field point
to internal data buffers. The main samba code assumes the dptr field
contains an address returned by malloc() and can free() or realloc it.
- Redid the CLEAR_IF_FIRST handling, the initialization logic is
now:
Generate resource name (SMBTDB_'fname').
if CLEAR_IF_FIRST then
lock resource with CR (concurrent read) mode lock.
convert lock to PW (protected write) mode lock with NOQUEUE flag.
if conversion succeeds then
zero the database.
end
convert lock to PR (protected read) mode.
else
lock resource with PR (protected read) mode lock.
end
The idea is that the first person can get the PW mode lock because
there is no contention. Subsequent users will fail to get the PW
mode lock and then the conversion to PR mode will force them to wait
until the guy with the PW lock completes the file initialization (and
lowers his lock to PR mode. Passing through CR mode is probably
superflous.
--------------------------------------------------------------------------
David L. Jones | Phone: (614) 292-6929
Ohio State University | Internet:
140 W. 19th St. Rm. 231a | [EMAIL PROTECTED]
Columbus, OH 43210 | [EMAIL PROTECTED]
Disclaimer: I'm looking for marbles all day long.
PLEASE READ THIS IMPORTANT ETIQUETTE MESSAGE BEFORE POSTING:
http://www.catb.org/~esr/faqs/smart-questions.html
