Is there any practical difference between using ln -s or mount --bind?
I want to move some folders to another partition, without changing their daemon setting, and wonder what approach I should take.
I prefer ln -s as it requires minimum setup (no /etc/fstab modifications), but perhaps there is a reason why it's not common?
Hell yes. If you execute the ln -s you create a symbolic link, which is an inode pointing to a certain filesystem object, which is why symlinks can traverse filesystems and hard links cannot: hard links do not have their own inode.
If you mount a filesystem with --bind, you create a second mountpoint for a device or filesystem.
If you envision a symlink as a redirect, then envision a --bind mounted filesystem as creating another gateway to data.
Symlinks and bind mounts are a whole different ballgame.
The --bind mount seems a bit more robust to me and it probably is a bit faster than working with a symlink. On the other hand, there are no serious drawbacks to using the symlink, as the performance hit will be small (if it at all exists).
Edit: I've been thinking about this, and the performance hit might be a bit bigger than I originally thought. If you have an application that reads a lot of different files, then every new file that is opened will require an extra read. Some research here suggests that my assumption is correct, so if you have an IO heavy application running there, consider the --bind option to mount above the symlink solution.
The reason it is not common, is probably the fact that a symlink is visible in an ls, whereas a bind mount is only visible when looking at /proc/mounts or /etc/mtab (which is what the mount command does, if it is executed without parameters). Other than that, I don't think there are any issues. I'd be interested if there are, though.
Addition: another issue with ln -s is that for some applications, when the path gets dereferenced, it may cause the application to balk if it "expects" certain items to be in specific places.
mount, when invoked without any arguments, prints the contents of /etc/mtab, which has slightly different information than /proc/mounts. (In particular, /proc/mounts (symlink to /proc/self/mounts) always shows the mountpoints visible to the process reading it.)
Dec 31, 2009 at 14:54
One of the big differences between ln -s and a bind mount is that you can use a bind mount to "modify" a read-only filesystem. For example, if there were a CD mounted on /mnt/application, and you wanted to replace /mnt/application/badconfigfile.conf with a correct version, you could do this:
mount -o bind /path/to/correct/file.conf /mnt/application/badconfigfile.conf
It would not be possible to affect the same change using a symlink because you're not able to modify the target filesystem.
This can also be used to good affect if you distributed a common suite of software via NFS (or some sort of cluster filesystem), and you want to make host-specific changes on one system. You can simply use a bind mount on the target system to override the files or directories as necessary.
Practial difference #1 for me between ln -s and mount --bind :
vsftpd doesn't not allow to browse a directory through a symbolic link, but allows when mounted.
I don't know what daemon you're using, but it may behave similarly.
One might note that as a consequence of binding to a mount, which is itself a binding, that is later rebound, the original binding remains intact, assuming everything physically stay connected.
That is, if bind A to B and bind B to C, and then bind D to B, C will still be bound to A. That might be what you want, or not. If one desires C to follow B then remount using the same targets, i.e. mount -o remount B C, or use --rbind instead. There is no --rebind option.
