Cross filesystem move behavior

[Michael Gerdts]

When a share crosses file systems, "moving" a file from one file system to
another fails from an NT client (ERRnaccess).  This appears to be a result
of conn->vfs_ops.rename() failing.  Since it eventually just calls
rename(), that is expected.  Relevant log entries are:

Is there any reason to *not* make it a bit friendlier to the poor windows
user?   I propose doing this by copying the file (or entire file
heirarchy, in the case of a directory move) then removing the file from the
source.  mv(1) in Solaris handles it this way, just in case anyone is
looking for a precedent.

More specifically the algorithm within vfswrap_rename() would look like:

	result = rename(old, new);
	if ( result != 0 ) {
		stat(old, oldbuf)
		stat(new, newbuf)

		if ( oldbuf.st_dev != newbuf.st_dev ) {
			result = move(old, new);
		}
	}

Then move would be:

	int move(char *old, char *new) {
		if ( old is a directory ) {
			create new/dir
			for each file in old
				move(old/file, new/file)
		} else {
			copy old to new
			unlink(old)
		}
	}

I am not promising to implement this, but if I am looking for a smallish
coding project I may implement it.  That is, assuming that I don't have
people screaming "That idea sucks!"

Mike