Re: [PATCH] ext3 [linux-2.6.2.]: accessing already freed inodes when under memory pressure

[viro@parcelfarce.linux.theplanet.co.uk]

On Fri, Apr 02, 2004 at 10:52:09AM -0800, Linus Torvalds wrote:
> 
> 
> On Fri, 2 Apr 2004 viro@parcelfarce.linux.theplanet.co.uk wrote:
> > 
> > Argh...  It's nastier than I thought - the corner case that can bite us
> > is
> > 
> > mount --bind path/foo bar ln bar baz
> > 
> > What should happen here?  Current code would make baz a link to
> > path/foo. Which is fine and should cause no problems, except that we
> > have no idea what the parent is.  Ignoring the mountpoint and linking to
> > underlying bar is clearly bogus.  Failing with -EXDEV might be more or
> > less OK, though...
> 
> I think that -EXDEV is not only OK, it's _required_. It may be the same 
> physical filesystem, but it's a different mount and thus a different 
> filesystem.

Yes - as long as we don't allow links to directories.  If we do, that's
much, much worse mess anyway.

Error values for link(2) are interesting, to put it mildly.  E.g. according
to Linux manpages a dangling symlink as a target must give -ENOENT; however,
link(2) does _NOT_ follow symlinks.  Since current tree doesn't agree with
that idiocy (we simply create a link to dangling symlink), I'd say "screw
the manpages".  Especially since SuS for once does not mandate that idiocy...