Re: kern/44208

[Antti Kantee <pooka%cs.hut.fi@localhost>]

The following reply was made to PR lib/44208; it has been noted by GNATS.

From: Antti Kantee <pooka%cs.hut.fi@localhost>
To: YAMAMOTO Takashi <yamt%mwd.biglobe.ne.jp@localhost>
Cc: gnats-bugs%NetBSD.org@localhost, lib-bug-people%netbsd.org@localhost,
        gnats-admin%netbsd.org@localhost, netbsd-bugs%netbsd.org@localhost, 
yamt%NetBSD.org@localhost
Subject: Re: kern/44208
Date: Mon, 13 Dec 2010 11:07:58 +0200

 On Mon Dec 13 2010 at 00:14:34 +0000, YAMAMOTO Takashi wrote:
 > hi,
 > 
 > >  That's because puffs_advlock is broken ;)
 > >  
 > >  See kern/43321
 > 
 > i didn't notice it.  thanks.
 > how about abortop?
 
 Well, abortop is kind of an afterthought to be able to run kernel file
 servers from libp2k.  For "normal" file servers componentname is always
 what is passed down from the kernel, so there is no automatic state in
 lookup in userspace.  I guess I could fix protocol so that componentname
 would be cached in libpuffs, but that's not a high priority, and that
 code is supposed to be changing in weird and wonderous ways anyway.
 
 I added some explanation on the manpage.
 
 > >  I guess an easy way to "fix" it is to just make it call lf_advlock()
 > >  in the kernel.  That would at least fix running vi to not give a warning
 > >  when run on puffs file systems.
 > 
 > is there anything which prevents a file system from implementing
 > an lf_advlock equivalent (or a more complicated one like NLM) by their own?
 
 I'm under the recollection that it's not possible for some fundamental
 reason, but I might misremember.  In any case, the kernel vop should
 stop trying to copyin struct flock for anything to even remotely work ;)
 
 -- 
 älä karot toivorikkauttas, kyl rätei ja lumpui piisaa