[distcc] sendfile() not working on tmpfs

[Martin Pool]

On 28 Jun 2002, Eivind Magnus Hvidevold <emh at hvidevold.cjb.net> wrote:
> 
> On Fri, Jun 28, 2002 at 04:19:50PM +1000, Martin Pool wrote:
> > I think your problem is fixed in CVS HEAD, or by the attached patch.
> > If you get a chance to try it please let me know.
> 
> The patch works perfectly! Thanks!
> 
> PS! What I meant by "fixable" was related to whether the problem with
> tmpfs was in distcc's usage of sendfile() or in the sendfile() 
> function itself. I'd call it "workaroundable" per my definition.

I basically implemented this

  http://banyan.dlut.edu.cn/news/052201/0043.html

which is what I'd always known I ought to do, but I hadn't got around
to it.

I kind of think that it can be done in the way Linus says, but I don't
care enough to do it.  sendfile() failing is not such a problem, but
loop devices are.  I was looking just the other day at a system that
builds a disk image in /tmp by loopback-mounting it, and this would
break if that machine's /tmp ever became tmpfs.

It looks like doing LZO compression will be worthwhile in most cases,
so sendfile will then not be so important.

-- 
Martin