Re: [PATCH][RFC] Re: 2.4.15-pre5: /proc/cpuinfo broken

["H. Peter Anvin" <hpa@zytor.com>]

Followup to:  <Pine.GSO.4.21.0111171359410.11475-100000@weyl.math.psu.edu>
By author:    Alexander Viro <viro@math.psu.edu>
In newsgroup: linux.dev.kernel
> 
> On Sat, 17 Nov 2001, Linus Torvalds wrote:
> 
> > 
> > On Sat, 17 Nov 2001, Alexander Viro wrote:
> > >
> > > Frankly, I'd prefer to try (b) before reverting to (a).  Patch doing that
> > > variant follows.  Linus, your opinion?
> > 
> > (d) make seq_file have my originally suggested "subposition" code.
> > 
> > Ie make the X low bits of "pos" be the position in the record, with the
> > high bits of "pos" being the current "record index" kind of thing.
> > 
> > That makes lseek() happy.
> 
> It will not help.  lseek() in question is relative and crosses the
> record boundary.  I.e. we have
> 
> 	n = read(fd, buf, ...);
> 	/* process k bytes */
> 	lseek(fd, k-n, SEEK_CUR);
> 
> and that will break just as the current variant does.  It's not about
> seek to remembered position - it's a relative seek to calculated offset.
> Calculated from number of bytes returned by read().
> 

We may really want to consider if we want /proc entries to be
S_IFREG().  The closest equivalent I can think of is really a
character device node (S_IFCHR) more so that S_IFIFO.

	  -hpa

-- 
<hpa@transmeta.com> at work, <hpa@zytor.com> in private!
"Unix gives you enough rope to shoot yourself in the foot."
http://www.zytor.com/~hpa/puzzle.txt	<amsp@zytor.com>