fstab rootfs / switch_root

[pwFoo]

Hi @troglobit,

I've done some first steps to build my own running linux system based on finit and busybox / alpine. Looks really great and works fine so far.
Great job with finit 👍

Now I thinking about rootfs and found an example to mount "/" from /etc/fstab. Is it something like switch_root or just mount a partition to "/"?

My idea was to mount squashfs and a overlayfs to / or squashfs with some persistent additional / mounts.

Is there an example with fstab rootfs mount / switch_root or squashfs / overlayfs?

Many thanks!

[troglobit]

Thanks, that's very kind of you :)

About 10-15 years ago there was a pivot root implementation in Finit. Since then nobody's use that. What works, and is in use today, is a basic mount -a which uses /etc/fstab by default. We recently added an option to use a different fstab from the finit command line (kernel command line with the -- separator from the boot loader).

I have setups with Alpine, Debian working with Finit, there use the native ext2 partitions. However, for embedded systems I employ Buildroot and there's an example of using a squashfs root in https://github.com/troglobit/myLinux -- specifically, see https://github.com/troglobit/myLinux/blob/main/board/common/rootfs/etc/fstab.ro which is used for make run when an amd64 build is started in Qemu. Another (very similar) example is https://github.com/westermo/netbox/blob/master/board/common/rootfs/etc/fstab which is always used. Both use a BusyBox mount helper to actually mount the things, but I think you'll get the gist of it from the fstab file.

[aanderse]

sorry to revive an old discussion but i was hoping for some guidance on the topic

warning: i am no expert on building an os from ground up so pardon my ignorance where applicable 😅

how does this work for a standard debian install (for example) which has an initramfs and wants to run finit? i have seen where systemd is run in an initramfs and has explicit switch root functionally so it can transition from initramfs to a "real" mounted root... but i don't understand how i can do this with finit without resorting to the solutions you posted above (but those solutions don't seem appropriate for a "proper" desktop for example)

my thought is that switch root functionality would be immensely useful to start finit have it do some basic startup to allow mounting of the root filesystem (to an alternate location) and then switch root to it... all before reading /etc/finit.conf which of course resides on the mounted root

am i missing something obvious or is it simply that finit isn't commonly used in the scenario i am imagining?

any and all guidance is much appreciated, thank you 🙇‍♂️

[aanderse]

one option that appeals to me

i will emphasize "one option" and "me", clarifying that i just wanted to add some ideas in an effort to brainstorm - regardless of what you end up thinking is a good idea i appreciate you simply listening and considering my thoughts regarding your project - i want to be respectful of your time and effort 🙇‍♂️

---

i watched a really cool talk where this guy was doing some neat things in initramfs (which he calls "stage 1"), just like @wkz had considered

the speaker in that linked video gave an example of a motivating case: "using systemd stage 1 to unlock drives with a TPM and Tailscale", though the demo code is a bit convoluted to those unfamiliar with nix lang

---

how do you see the switch root working?

[troglobit]

i want to be respectful of your time and effort 🙇‍♂️

Much appreciated 🙏

i watched a really cool talk where this guy was doing some neat things in initramfs (which he calls "stage 1"), just like @wkz had considered the speaker in that linked video gave an example of a motivating case: "using systemd stage 1 to unlock drives with a TPM and Tailscale"

Very interesting, thanks for sharing that!

how do you see the switch root working?

There are many ways to do this, from what I can see, systemd supports this over at least D-Bus and systemctl switch-root and the latter should be fairly straight forward to implement, we can also have a peek at switch-root.c

Please keep in mind, we don't have the massive feature set that systemd does (in part intentional), so the impressive systemd-ask-password feature mentioned in the talk is not something we have. I'd like to say yet, because if there's a good use-case for it I see no reason for not carrying our own implementations of the same ideas, and considering it could be integrated with ssh made it very interesting.

[aanderse]

sounds good!

Please keep in mind, we don't have the massive feature set that systemd does (in part intentional)

yes, i think you're doing a wonderful job at striking a balance between easy to use/understand + capable feature set so please keep it up!

how would plugins and events work? would events like filesystem mounted fire twice and trigger plugins multiple times? that sounds like it could be useful - or perhaps separate events to hook into?

---

if you think this is fairly straightforward to implement and are willing to i am very excited to do thorough testing for you because i am very interested in getting this running on an old laptop i have

thanks

[troglobit]

Please keep in mind, we don't have the massive feature set that systemd does (in part intentional)
yes, i think you're doing a wonderful job at striking a balance between easy to use/understand + capable feature set so please keep it up!

Thanks, it's a struggle ... ☺️

how would plugins and events work? would events like filesystem mounted fire twice and trigger plugins multiple times? that sounds like it could be useful - or perhaps separate events to hook into?

The way I see it is that you have one finit in your initramfs that runs until someone/somescript calls initctl swith-root /path (optionally with a second argument like the corresponding systemctl call) -- this would trigger the first finit instance to invoke the kernel switch_root(8) syscall, which effectively replaces the first finit with the second (or another init process). They will both be PID 1, weirdly enough.

if you think this is fairly straightforward to implement and are willing to i am very excited to do thorough testing for you because i am very interested in getting this running on an old laptop i have

Finit is one of my favorite personal projects, that is also used in other projects I care about, but these days I have very little time in general to spend on implementing requests from others, for free. This particular feature is interesting but unfortunately not something I need myself at the moment.

[aanderse]

ah, understood - at least this discussion serves as a record if anyone else had the same question

thanks for your time and explanation 🙇‍♂️

[troglobit]

This discussion has a related issue, #74

[aanderse]

@troglobit - i asked my friend claude to take a look at what it would take to add initramfs functionality to finit and after a bit of iterating we came up with this

i'm still in the highly experimental stages of playing with this branch, but i wanted to reach out and see if this seems reasonable to you - after going over the code a bit and running some qemu tests i think i'm on the right track

if you end up with some extra time and can take a quick look i would appreciate it, but no rush at all... holidays and all, enjoy your time off 😄

[troglobit]

This looks more than reasonable! Very cool indeed!

[troglobit]

Merged to main in 1e93fc1