April 2006
Most Linux and Unix file systems don't allow hard links to directories (except for the . and .. entries that mkdir creates itself). The reasons are are pretty obvious: you could really confuse programs like ls (ls -R), find and of course fsck if you created links that recursed back to themselves. If there was a compelling reason to allow directory hard links, you'd need to rewrite any program that wants to walk a file system tree to be aware of the possible problems..
So instead we have symlinks. You've probably used them for things like shifting around disk space or to give more convenient access to a directory. For example, Mac OS X creates /tmp as a symbolic link to "private/tmp". We use symlinks to make other directories visible under Apache's htdocs directory (though the same thing can be accomplished with Apache's configuration files).
One problem with symbolic links is that really they are just files. A special kind of file, yes, but a symlink only points at a directory - it doesn't act like one. So, for example, if you put a symlink to /xyz in a users home directory, and the user has write permission to his home (as he ordinarily would), he can remove your symlink. Nothing you can do with ordinary permissions can prevent that. You can do a "chattr +i" on your symlink, but because it is a symlink, that passes through to the actual directory, making it unusable. If you use "+u" (undeletable), that again passes through, and the user still can delete your symlink.
This can be extremely annoying, especially when users accidentally delete a symlink they need to have. Of course your real directory is still safe, but you need to recreate the symlink. In the mean time, your user is confused or maybe even broken.
There is at least one way around this. If the thing you want to link to is a mounted file system, you can use the "--bind" option of mount to create an unbreakable link.
If the thing to link to it isn't a separate fs, you can almost always make it be one.
Here's how it works. Let's say we have /dev/foo mounted at /foo and I want a "link" to that under /home/fred. All I have to do is:
mount --bind /foo /home/fred/foo
Fred can have full write permissions on /foo if he needs it, but he will not be able to remove /home/fred/foo. Not even root can:
# rm -rf /home/fred/foo
rm: cannot remove directory '/home/fred/foo': Device or resource busy
Some related links you might find interesting:
http://lwn.net/Articles/159077/
http://linuxgazette.net/issue93/tag/2.html