Move across mount,sb

Move across mount,sb

[Jan Engelhardt]

Hello,


	touch /tmp/foo;
	mount /tmp /mnt --bind;
	strace -e rename mv /tmp/foo /mnt/bar

Ideally, I would expect, that since /tmp and /mnt are the same 
filesystem, that the move operation would complete without _copying_ the 
file. But strace returns

	rename("/tmp/foo", "/mnt/bar") = -1 EXDEV (Invalid cross-device link)

The piece of code identified is fs/namei.c:2578:do_rename()

	error = -EXDEV;
	if (oldnd.mnt != newnd.mnt)
		goto exit2;

Why is EXDEV returned when _vfs mountpoints_ are crossed? Should not it 
be more like the following?

	error = -EXDEV;
	if (oldnd.mnt->mnt_sb != newnd.mnt->mnt_sb)
		goto exit2;




Jan
-- 

Re: Move across mount,sb

[Al Viro]

On Thu, Mar 15, 2007 at 02:20:25AM +0100, Jan Engelhardt wrote:
> Why is EXDEV returned when _vfs mountpoints_ are crossed? Should not it 
> be more like the following?
> 
> 	error = -EXDEV;
> 	if (oldnd.mnt->mnt_sb != newnd.mnt->mnt_sb)
> 		goto exit2;
> 

	No.  This is absolutely deliberate - mountpoint creates a boundary and
such boundaries are very useful for restricting modifications of filesystem.
IOW, it's not a bug and it applies to other operations as well (link(), for
example).