Date: Thu, 26 Feb 1998 07:26:00 +0000 (GMT) From: Terry Lambert <tlambert@primenet.com> To: mark@vmunix.com (Mark Mayo) Cc: Matthew.Thyer@dsto.defence.gov.au, perlsta@cs.sunyit.edu, current@FreeBSD.ORG Subject: Re: Build still broken - NOT! - Oh yes it is (was) Message-ID: <199802260726.AAA26006@usr08.primenet.com> In-Reply-To: <19980226000847.50861@vmunix.com> from "Mark Mayo" at Feb 26, 98 00:08:47 am
next in thread | previous in thread | raw e-mail | index | archive | help
> Well, I've personaly given Terry's patches a whirl and they seemed > to fix up a lot of locking problems with my 3.0-CURRENT NFSv3 > cliient. I test against an Auspex filer. This is a false sense of security you're getting here. The locks are locally asserted. As I said in another posting, inter-client locking won't work (ie: you're locks will be granted, even if another NFS client thinks it holds them) because they aren't proxied to the server. I've only got one collision domain implemented here, and then only as a preparation for rpc.statd to tell me the NFS server rebooted and I need to recreate the locks (locks held by processes on files where the lock can't be reasserted should actually get ESTALE on access attempts, I think, to be safe). In other words, they are necessary, but not sufficient for a full NFS client locking fix. > I haven't tried mounting both /usr/src and /usr/obj from the NFS > server, but I have successfully built the world with a /usr/src > NFS mounted. The majority of current NFS problems are "write"-based. You wouldn't see improvement in anything but locking tests run on one client against the server (but in that case, it would look like it's working). > I don't know if Terry's patches would help you since I belive it > was mostly just to implement locking correctly (I think), but it > may be worth trying out for a comparison if your latest cvsup > still barfs. http://www.freebsd.org/~terry for the patches. The more testers, the merrier. But I'm already confident from the people who've run it so far that I'm not leaking memory, and that the lock states are held correctly before and after the coelesce, and the lock problems and the write problems are two different animals (John Dyson hinted that he may be jumping into this arena soon, now that he has time). If you commit the patches, do it for what they do, not for what they don't do. 8-). Terry Lambert terry@lambert.org --- Any opinions in this posting are my own and not those of my present or previous employers. To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-current" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199802260726.AAA26006>