> > >  You are right - we do have races there.  Always had.
quoted text
> > > And nfsd_permission() is the next target for continuation of ro-bind
> > > series.  Assuming that we don't simply make r/w export to hold will_write
> > > all along, in which case all these checks around calls of vfs_...() in
> > > there simply go away - that's also an arguable option.
> > 
> > Yes.  And that _still_ doesn't make the path_*() interface wrong.
> 
> It would make it bloody useless for nfsd.  With ecryptfs being a piss-poor
> argument in favour of anything other than git rm, what's left?  Another
> stack frame in fs/namei.c syscalls?

Since all the vfs_* functions will become static with path_* being the
only caller, the compiler will be happy to get rid of that stack frame
too.

What is left is the guarantee, that the race-free r/o remounts will
always work and some obscure caller didn't forget to surround it with
the r/o checks.

I think it's definitely worth it.

Miklos
--
unsubscribe notice
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/