block_meta_v2: set first-lba if needed when creating a new GPT
sfdisk sets first-lba to 2048 by default, which fails if one of
the partitions has an start smaller than that. Having a
partition start so early in the disk is a bit strange but it
works if you are preserving a partition (because we preserve
first-lba) so it should probably work when creating one too.
fix block meta 4k integration tests with recentish losetup
In recent releases of util-linux, --sector-size and --partscan don't
play well together, so drop the --partscan and call partprobe after
losetup returns. This is fixed in util-linux master but not the latest
release.
Today I discovered that there is a "curtin swap" command that just does
the "configure swap" part of curthooks and then I fairly quickly after
that discovered that probably noone has used it because it
unconditionally exits with code 2 even if everything succeeded.
re-apply e099e32 and update for the modified test codebase.
Drops using apt-key in favor of placing the file direct on disk.
Includes a fix for placing the key in the correct directory.
This refactor helps make way for discontinuing usage of apt-key.
When trying to do it all at once, there were problems with the file path
that needed to be used for the key file.
The major change is that these tests are run with a tmpdir as the
target, instead of '/'. This is also independently a nice improvement,
since misbehaved tests or code would really try to write to the host
system.
The tests were written assuming apt key files would have full paths, but
the examples (and presumably most real-world usage) were based on
filenames, so move the tests to use file names.
Add back a full path test, since that is supported behavior today.
