Mount helper doesn't handle all mount flags properly

As a suggestion, Tyler told me that we might call /"sbin/mount -i" using exec() instead of adding more flags into ecryptfs_generate_mount_flags().

The idea is a good solution to support all new flags from mount command without touching the ecryptfs code to add others ones.
I looked into some others mount helpers, such as nfs, cifs, smbfs, fuse, to see how they support these flags, but they use the same solution as ecryptfs. They uses mount() API function, and have a bigger list of mount flags that they check before call the mount() function.

mount.fuse is the only one that uses execl() to call a fuse command to mount a directory, but it does not call /sbin/mount; it uses its own fuse command instead; and flags are checked and ignored before calling the command.

Either way, I don't foresee any issues if we use exec() or execl() to call "/sbin/mount -i ecryptfs SOURCE DEST -o OPTS" in order to mount ecryptfs.

Any ideas or comments about this?

On 2012-04-19 16:27:54, Sergio Peña wrote:
> The idea is a good solution to support all new flags from mount
> command without touching the ecryptfs code to add others ones. I
> looked into some others mount helpers, such as nfs, cifs, smbfs, fuse,
> to see how they support these flags, but they use the same solution as
> ecryptfs. They uses mount() API function, and have a bigger list of
> mount flags that they check before call the mount() function.

Hmm... this kind of worries me. I can't imagine why no other mount
helpers would not just execl("/bin/mount", "-i", ...). It makes me think
that we're missing something.

Let's not do it for the setuid root helper (mount.ecryptfs_private) and
just do this for the regular mount helper (mount.ecryptfs) for now. I'd
need more confidence in this approach before rolling it out for a setuid
binary.

> mount.fuse is the only one that uses execl() to call a fuse command to
> mount a directory, but it does not call /sbin/mount; it uses its own
> fuse command instead; and flags are checked and ignored before calling
> the command.

Oddly, I see that fuse does execl("/bin/mount", "-i", "-f", ...) for a
fake mount to update mtab.

This was released back in ecryptfs-utils-97