Lennart Poettering

well, sure, but if you encrypt /usr/ then I am not sure a "golden image" model really is in focus? i.e. pick one: 1. golden images that encrypt rootfs on...

This isn't material to merge, since deactivating plymouth like this also disables all kind of kbd input. Which sucks a bit too much I guess. Just wanted to push this...

We can certainly do something like this, but only if the relevant kernel subsystem maintainers sign off on this, that this is a safe thing to do. @jwrdegoede iirc you...

Hmm, the pre-existing code is indeed really broken about this. By passing over the map fd things certainly will be better, but this still means that over the reexec we'll...

Maybe a compromise that is relatively simple to implement is this: 1. Pin both program and map for now. 2. Consider the name you pick a version identifier for program...

> For Debian cloud images, we currently need to have the root partition both as number 1 (or how you count it), aka both the first partition but also physically...

> It expects files to be aligned to 512 bytes, which is properly documented. However nowhere seems to exist a technical limitation that warrants this limitation. Linux block devices have...

I am not convinced we should do that. ".." is something we cannot just resolve offline, because "/foobar/.." might be equivalent on "/" if "/foobar" is a dir, but might...

hmm, not sure how much I like this: i am pretty sure that various calls of ours inside the loop that applies transfer_table[] actually might end up using /proc/ to...

This change was already made in v254, why does this only come up now? Given this is just a workaround for a limitation of AA, why not just add a...