Good morning,
Yes  it  seems  if we can’t ensure root would be read‐only there’s a risk.
It’s
limited though since first few 100ms are done through ROM, the filesystem
 start
mount  a  bit  afterwards  but the solution i should design must ensure it
works
even in those cases since this is an improvement over  the  base  which
 handles
this gracefully with journal rebuild on each reboot at the moment.

Regards
Jean-François

Le mar. 25 oct. 2022 à 11:06, Stuart Henderson <s...@spacehopper.org> a
écrit :

> On 2022/10/24 21:35, Jean-François SIMON wrote:
> > Hi,
> >
> > Thanks i believe this is very valuable approach, i didn't think of it.
> > Indeed i can run tests on that way, obviously a r/o system is much
> > better way.
> >
> > I believe i should do testing that way to validate it.
> >
> > All right i'll be working on it as time allow, thx for this idea.
>
> It should be possible to remount / as read-only after reorder_kernel is
> done, or immediately in rc.local if you disable reorder_kernel.
>
> (This is still a bit problematic though as IME there is a fairly high
> risk of a second power failure shortly after power has returned -
> circuit breakers are often designed so that they don't trip immediately
> when current limits have been exceeded, only if they are sustained,
> so actually the first few minutes after power is restored are quite
> high risk).
>
> /dev, /tmp and at least parts of /var need to be writable, this can be
> done via mfs if you have enough RAM (see the -P flag to mount_mfs to
> populate with the files you need) but some of the /var files may need
> syncing back to permanent storage, and mfs /dev is a bit of a nuisance
> for upgrades.
>
> For logs, memory buffer logging is usually the way to go (see syslogc)
> with network logging if you want permanent storage.
>
> None of this is arm-specific, you might get some other ideas on a more
> general list like misc@. There are also projects like flashrd which
> might give you some other ideas. Though in general it is enough of a
> pain that it's often easier to improve power reliability...
>
>

Reply via email to