To: None <tech-kern@NetBSD.ORG>
From: der Mouse <mouse@Collatz.McRCIM.McGill.EDU>
Date: 11/22/1995 09:33:06
>> Try running "tail -f" on an NFS-mounted filesystem.
>> (How does one work around that?)
> by not using NFS?
Yeah, the problem is, that's about the only way to.
> the fact is, while NFS is bloody useful, you can get by without using
> it in many environments. (*chuckle* and some would say "you
_I_ think that if NetBSD's NFS implementation doesn't work, it should
be removed. Whether it's good or not, if it doesn't work, it shouldn't
> by fixing tail? the problem is cause by the fact that tail uses
> mmap, and other mmap-using programs (like 'cat') have been changed to
> not use it, to avoid just this problem!
This is hardly a fix. If mmap() doesn't work, it should be removed, or
at the _very_ least, it should be documented as "broken, use at your
own risk". And supplied programs (tail and vi come to mind
immediately) should never go near it. But shipping an mmap() that
doesn't work the way it should, without a warning, and especially using
this broken mmap() in stock code, is horrible.
But the problem is worse than that, as I understand it; it's not just
mmap(), but anything that pages to a file. And that includes almost
all executing binaries. After I did my tfs, I tried using it for
/usr/src and doing a "make build"; it died horribly because so many
things broke on NFS filesystems.