Fix for multi-user database corruption problems just checked in.

Steven French sfrench at
Wed Mar 13 21:08:03 GMT 2002

At some point you actually had to spin (on the server) in the good old days
because dos had no lock queuing mechanism (and client based lock retries
were done in a tight loop whose timing was based on a for loop running at
8080 processor speed not the speed of current screaming processors) so the
server was supposed to spin slowly on the 2nd lock of the same byte range
and then presumably more slowly ... doing this back off algorithm right ten
years ago in netservr and hpfs386 bought enormous performance and
functional gains on multi-user (DOS heritage) databases remotely accessing
the OS/2 SMB file servers.  It was probably a factor in some old smb

It is odd though that you had to add implicit server lock queueing before
returning the first failed lock rather than after the inevitable retry.

Steve French
Senior Software Engineer
Linux Technology Center - IBM Austin
phone: 512-838-2294
email: sfrench at

More information about the samba-technical mailing list