Jim Rees wrote:
> There are quite a few lock order reversals in the file server, and my patch
> does not attempt to address them.  I haven't completely figured out how
> locking is supposed to work.  If deadlock prevention depends entirely on
> lock order, we're in big trouble.

Deadlock prevention MUST depend entirely on lock order.   When I started
working on the Windows code there the documented lock ordering had been
completely ignored.   It took two years and the development of the
stress tests to be able to remove all of the deadlocks that we were able
to find.

If the server code is in the same shape, then we have a lot of auditing
to do.

> I'll post what I have tomorrow but I don't think it will fix this deadlock.

It may not fix the deadlock but if your patch prevents the holding of
host locks across RPCs, then it should avoid the server coming to a halt
simply because one client isn't responding.

I look forward to seeing the patch.

Jeffrey Altman

Attachment: smime.p7s
Description: S/MIME Cryptographic Signature

Reply via email to