[Pkg-sysvinit-devel] Bug#359717: Binding Mount to Itself Breaks Things

[Askar Safin]

> The problem is binding mounts *will* break things if used improperly
No, I completely disagree with you.
If you really think that bind mounts are bad, then go to kernel developers and tell them to remove bind mount support.
Bind mount are present in kernel. People use them. So, Debian should fully support them.
Debian should not force any policy to users. If users want to use bind mounts, let them use it. And Debian's "mountpoint" should respect users and should show bind mounts.
Yes, this is not UNIX-way. But modern GNU/Linux is not old UNIX. GNU/Linux is modern OS with a lot of not-UNIX features:
* bind mounts (also rbind mounts etc)
* FUS (file system in user space). FUS can potentially be infinite, for example, say, /sshfs/some-host/some-port/remote-fs/sshfs/some-host/some-port/remote-fs/...
* Linux containers (also OPENVZ), this includes PID namespaces, mount namespaces, etc
* Even minimal GNU/Linux installation has infinite fs, for example, /proc/1/root/proc/1/root/proc/1/root...
* System capabilities (man 7 capabilities)
* ...
Do you want to go back to past? Of course, no. So, please agree with modern GNU/Linux features. We should support them as more as we can. Of course, there are a lot of programs, which still assume old UNIX system. But in core programs (such as mountpoint) we should be very close to kernel and should write unportable but right code. Yes, low-level programs should be unportable, they, of course, should depend on kernels. So, they may contain a lot of #ifdefs and they should support all features that kernel gives.

Askar Safin