Re: [Tim Waugh <address@hidden>] directory mounted with --bind causes cp -R to fail

[Jim Meyering]

>   Date: Mon, 31 Mar 2003 14:17:08 +0100
>   From: Tim Waugh <address@hidden>
>   To: address@hidden
>   Subject: directory mounted with --bind causes cp -R to fail
>   Message-ID: <address@hidden>
>
>   Original bug report at:
>   https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=86302
>
>   Under Linux it is possible to have a special 'bind' mount type, which
>   has the effect of remounting part of the file hierarchy somewhere
>   else.  But this confuses 'cp --preserve=links':
>
>   Run these commands as root:
>     mkdir -p src/{x,y}
>     mount --bind src/x src/y
>     cp --archive --one-file-system src dst
>
>   Is this a cp bug, or expected behaviour?
>
>   Tim.

Hi Tim,

Thanks for forwarding that.
Could it be that there's a problem with the way `mount --bind' works?
With linux-2.4.19, I see that the two directories created above
have the same dev/inode numbers.  That doesn't seem right.
I would have expected the device numbers to be different.

This shows they have the same device numbers
(using your example):

  # cd src; stat --format='%d %n' [xy]
  773 x
  773 y

But are they really both on the same `device'?
If so, I should be able to make a hard link from
a file in one to the other:

  # touch x/foo
  # ln x/foo y/bar
  ln: creating hard link `y/bar' to `x/foo': Invalid cross-device link
  [Exit 1]

Whoops.  We can't do that.
And the failure suggests that they really are on separate devices.
So what's the deal?  I suspect that this is all due to a deeper problem.

BTW, the fact that rsync succeeds merely shows that
rsync doesn't complain about hard-linked directories.
`cp -a' could do the same thing, but then we wouldn't
have discovered this inconsistency :-)