bug#18736: chroot regression - chroot avoids the chroot() call too eagerly.

[Rogier]

Hi,

Since a few months, it seems that chroot has started avoiding the 
chroot call if it can be determined to be idempotent. 
It looks like the new check is based on inode comparison - if the 
inode is the same, the chroot() call is considered idempotent, and not 
performed.

However, while two directories being in the same file system and having 
the same inode number implies that they are the same directory, it 
does not necessarily imply that they have same file system path (i.e. 
use the same mountpoint), so there is no guarantee that the entire 
directory trees rooted at the two directories are also identical even 
though the directories are.This means that chroot will fail to 
chroot() in cases when the call would in fact not be idempotent.

On my system (debian testing), I have / bind-mounted elsewhere, with a 
slightly different directory tree mounted beneath it, as is mounted 
beneath /. In my case, I need this so that I can make partial backups 
of the system - including some file systems but leaving out others. 
Undoubtly, there are other use-cases as well - I wouldn't be surprised 
if users of libpam-chroot can be affected by this, depending on exactly 
how they configure their chroot environments.

The new chroot behavior no longer allows chrooting into such alternate 
trees. In my case, that means that my backup fails.

Kind regards,

Rogier.