Re: [PATCH] files: improve the backup procedure to ensure no data is lost

[Michalis Kokologiannakis]

On Wed 01 Jul 2020 at 20:58, Benno Schulenberg 
<bensberg@telfort.nl> wrote:

> Op 01-07-2020 om 12:15 schreef Michalis Kokologiannakis:
> > On Wed 01 Jul 2020 at 10:44, Benno Schulenberg 
> > <bensberg@telfort.nl> wrote:
> > > So... I think we should skip the fastpath and just always copy
> > > the current file to create a backup.  Or at least skip the
> > > fastpath when the current user is not the owner of the file.
> >
> > I can send out another version that implements either of 
> > these. I think keeping
> > the fastpath would be useful for when editing large files (and 
> > should cover a lot
> > of cases), but then special care would be required to make sure 
> > we also get
> > correct things like ACLs, right?
>
> Oof.  More complications we hadn't thought of.  :|  Another one 
> was that
> when 'backup_by_move' is true, the open() for the writing of the 
> actual
> file should use O_EXCL, not O_TRUNC.

And one more, just for the record. If the original file had hard 
links pointing to it, renaming the file to create the backup would 
make the links point to the backup...

> So, let's take the simpler route: stay closest to what we have 
> now, and
> copy the file to make a backup.  This way permissions of the 
> original
> file are guaranteed to stay the same.  In bad situations nano 
> may still
> truncate the file, but then there will be a backup somewhere, 
> and that
> is what you set out to achieve: data resilience.

Agreed. I will follow up with an updated version.

Thank you,
Michalis

> Benno