Re: [PATCH 2/2] exec: open code copy_string_kernel

[Al Viro <viro@zeniv.linux.org.uk>]

On Fri, May 01, 2020 at 02:40:13PM -0700, Andrew Morton wrote:
> On Fri, 1 May 2020 22:30:48 +0100 Al Viro <viro@zeniv.linux.org.uk> wrote:
> 
> > On Fri, May 01, 2020 at 02:19:03PM -0700, Andrew Morton wrote:
> > > On Fri,  1 May 2020 12:41:05 +0200 Christoph Hellwig <hch@lst.de> wrote:
> > > 
> > > > Currently copy_string_kernel is just a wrapper around copy_strings that
> > > > simplifies the calling conventions and uses set_fs to allow passing a
> > > > kernel pointer.  But due to the fact the we only need to handle a single
> > > > kernel argument pointer, the logic can be sigificantly simplified while
> > > > getting rid of the set_fs.
> > > > 
> > > 
> > > I don't get why this is better?  copy_strings() is still there and
> > > won't be going away - what's wrong with simply reusing it in this
> > > fashion?
> > > 
> > > I guess set_fs() is a bit hacky, but there's the benefit of not having
> > > to maintain two largely similar bits of code?
> > 
> > Killing set_fs() would be a very good thing...
> 
> Why is that?  And is there a project afoot to do this?

Long story - basically, it's been a source of massive headache too many times.
No formal project, but there are several people (me, Arnd, Christoph) who'd
been reducing its use.  For more than a decade now, I think...

FWIW, I doubt that it will be entirely killable; Christoph appears to be
more optimistic.  In any case, its use has been greatly reduced and having
it narrowed down to even fewer places would be a good thing.

In the same direction: use_mm()/unuse_mm() regularization wrt set_fs(), getting
rid of it in coredump code, some movements towards killing ioctl_by_bdev();
not sure if I've spotted everything - Christoph?