[Bug 254395] bsdinstall: fail script install after BETA3

[Chris]

On 2021-03-19 13:06, bugzilla-noreply at freebsd.org wrote:
> https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=254395
> 
> Nathan Whitehorn <nwhitehorn at FreeBSD.org> changed:
> 
>            What    |Removed                     |Added
> ----------------------------------------------------------------------------
>            Severity|Affects Only Me             |Affects Some People
>                  CC|                            |imp at FreeBSD.org
>            Priority|---                         |Normal
> 
> --- Comment #6 from Nathan Whitehorn <nwhitehorn at FreeBSD.org> ---
> Thanks for the suggestion about the documentation -- I've updated the man 
> page.
> 
> The core problem here is that our tar can't extract archives to FAT32 with
> default options, since it treats inability to set modification time as a 
> fatal
> error and FAT32 doesn't let you do that on the root directory. As such, any
> file in the release tarballs can't be extracted to FAT32. For interactive
> installations, the bsdinstall distextract tool, a CURSES-y frontend to
> libarchive, solves this by ignoring ctime/mtime errors.
> 
> Some extra commentary on solutions, so it can be in one place. Possibilities
> are:
> 
> 1. We drop /boot/efi from mtree. That will result in it not existing in
> base.txz, solving this issue, but will result in it not being in mtree. It 
> will
> also leave in place an identical bug that will break scripted installation 
> on
> bare-metal POWER8 and POWER9 systems, although that is a tier-2 platform.
> 
> 2. We add an option to tar to ignore failure in setting ctime/mtime, like 
> the
> interactive installer uses. This has the difficulty that the patch is hacky 
> and
> would have to go through upstream.
> 
> 3. We go back to using distextract for scripted installations as well as
> interactive ones, reverting d7640440fb644fde697f62fdff0b55aa3a4d5ef7. This
> fixes this issue but will result in installation failures for scripted 
> installs
> without a controlling tty. (It will also add nice progress bars to scripted
> installs).
> 
> 4. We do --exclude /boot/efi when running tar, then mkdir -p it by hand
> afterward. This is incredibly hacky and otherwise essentially functionally
> equivalent to #1. Like #1, it will fix this issue and has no obvious 
> functional
> downside, but leaves scripted installs bare-metal POWER8 and POWER9 broken.
> 
> 5. We patch the file system driver to (pretend to) allow setting times on 
> the
> mount point. I don't want to do this, since I don't want to solve this in 
> the
> kernel at RC3 and I don't like it pretending to do things it can't do.

  6. (my favorite) do NOT require that the efi/ partition be in strictly a
  fat32 format. I mean fat32 is not strictly required as the format for the 
efi
  partition. It is simply _assumed_ to be the required format and as such, the
  one used in so many cases.

Whould it actually be that much harder to use ffs/ufs?

You asked. ;-)

> 
> --------
> 
> Of these, 1, 3, and 4 are quite easy to implement, but all have some 
> downside.
> My temptation is to do 4 for 13.0, since it will definitely work but is just
> lame, then either do #2 or a variant on #3 where distextract notices there 
> is
> no tty and doesn't try to set up a dialog as a longer-term fix in HEAD. Any
> thoughts?