On Sat, Jun 08, 2002 at 10:47:26PM +0200, Andreas Bombe wrote: > On Fri, Jun 07, 2002 at 08:52:52AM -0500, Steve Langasek wrote: > > On Fri, Jun 07, 2002 at 04:13:11AM +0200, Andreas Bombe wrote: > > > I'm just going by the Documentation/initrd.txt in the Linux sources > > > (I looked in 2.4.18 here). pivot_root(1) also recommends a chroot. > > > In any case much is implementation defined (like whether the shell's > > > root is implicitly changed), so it's better to stick to the protocol. > > > It might change, and a simple cd might not be sufficient anymore. > > > Anyway, can you umount the initrd if you only did a 'cd /'? > > Well, no, because the script itself, if still running, has files open on > > that device: the executable and all the libraries it loads. You have to > > exec a different program/script on your main root in order to free up > > the mount point, IIRC. > Sure. Um, let's word it a bit differently. Is it possible to umount > the initrd if /sbin/init in the new root is exec'd without chroot? Yes, it is. > Maybe it is, but the documentation gives no guarantees about that, so > that behaviour is fair game for kernel changes. I wish it would give such a guarantee. The current behavior is far more intuitive, IMHO. :) Steve Langasek postmodern programmer
Attachment:
pgpQBOqo9Td9c.pgp
Description: PGP signature