libsed in mdadm

Hello,
IMSM/VROC is going to support self-encrypted drives. With this feature you need to unlock the drives during boot-up in UEFI first. It is kind of protection from physical stealing.

To ensure security, Linux have to respect that. It means that we need to determine if the drive support locking and do not allow to mix locked and unlocked drives in one IMSM array.

To grab that information we will need to impose the "magic commands" to the drives. There is a libsed library, designed for such purposes:
https://github.com/sedcli/sedcli

So far I know, this library is not released under distributions (not handled by package managers) and that will bring not user friendly dependency- you will need to compile and install the lib first to build mdadm.

The sedcli project is maintained in Intel, currently it is not in active development but there are no plans to drop it, interest around it is growing as you can see. It seems to be great opportunity for this project to become integrated with mainstream distributions when mdadm will start to require it.

So, my questions are: Are we fine with adding this dependency? Are there big cons you see?
Obviously, I will make it optional like libudev is.

I can try to re-implement the functionality I need in mdadm but it is like reinventing the wheel.

Any feedback will be appreciated.

Quoted 30 lines · Mariusz Tkaczyk

Hello,
IMSM/VROC is going to support self-encrypted drives. With this feature you need to unlock the drives during boot-up in UEFI first. It is kind of protection from physical stealing.

To ensure security, Linux have to respect that. It means that we need to determine if the drive support locking and do not allow to mix locked and unlocked drives in one IMSM array.

To grab that information we will need to impose the "magic commands" to the drives. There is a libsed library, designed for such purposes:
https://github.com/sedcli/sedcli

So far I know, this library is not released under distributions (not handled by package managers) and that will bring not user friendly dependency- you will need to compile and install the lib first to build mdadm.

The sedcli project is maintained in Intel, currently it is not in active development but there are no plans to drop it, interest around it is growing as you can see. It seems to be great opportunity for this project to become integrated with mainstream distributions when mdadm will start to require it.

So, my questions are: Are we fine with adding this dependency? Are there big cons you see?
Obviously, I will make it optional like libudev is.

I can try to re-implement the functionality I need in mdadm but it is like reinventing the wheel.

Quoted 1 line

Any feedback will be appreciated.

Hi Mariusz,

I am not against adding it to mdadm, though I think a better approach is to try and get the library built as a package for the distros.

Did you look into that yet?

Quoted 39 lines · Jes Sorensen

Quoted 24 lines · Mariusz Tkaczyk

Hello,
IMSM/VROC is going to support self-encrypted drives. With this feature you need to unlock the drives during boot-up in UEFI first. It is kind of protection from physical stealing.

To ensure security, Linux have to respect that. It means that we need to determine if the drive support locking and do not allow to mix locked and unlocked drives in one IMSM array.

To grab that information we will need to impose the "magic commands" to the drives. There is a libsed library, designed for such purposes:
https://github.com/sedcli/sedcli

So far I know, this library is not released under distributions (not handled by package managers) and that will bring not user friendly dependency- you will need to compile and install the lib first to build mdadm.

The sedcli project is maintained in Intel, currently it is not in active development but there are no plans to drop it, interest around it is growing as you can see. It seems to be great opportunity for this project to become integrated with mainstream distributions when mdadm will start to require it.

Quoted 8 lines

So, my questions are: Are we fine with adding this dependency? Are there big cons you see?
Obviously, I will make it optional like libudev is.

I can try to re-implement the functionality I need in mdadm but it is like reinventing the wheel.

Any feedback will be appreciated.

Hi Mariusz,

I am not against adding it to mdadm, though I think a better approach is to try and get the library built as a package for the distros.

Quoted 2 lines

Did you look into that yet?

We (as in 'We as an OS distributor') actually evaluated packaging libsed some time ago, but decided against it as the original authors (namely Intel) apparently disbanded it. So before adding it to a distro there needs to be an active maintainer, and one would be looking to Intel here.

Quoted 42 lines · On Wed, 23 Aug 2023 08:00:51 +0200 Hannes Reinecke <[email protected]> wrote:

Quoted 41 lines · Jes Sorensen

Quoted 24 lines · Mariusz Tkaczyk

Hello,
IMSM/VROC is going to support self-encrypted drives. With this feature you need to unlock the drives during boot-up in UEFI first. It is kind of protection from physical stealing.

To ensure security, Linux have to respect that. It means that we need to determine if the drive support locking and do not allow to mix locked and unlocked drives in one IMSM array.

To grab that information we will need to impose the "magic commands" to the drives. There is a libsed library, designed for such purposes:
https://github.com/sedcli/sedcli

So far I know, this library is not released under distributions (not handled by package managers) and that will bring not user friendly dependency- you will need to compile and install the lib first to build mdadm.

The sedcli project is maintained in Intel, currently it is not in active development but there are no plans to drop it, interest around it is growing as you can see. It seems to be great opportunity for this project to become integrated with mainstream distributions when mdadm will start to require it.

Quoted 8 lines

So, my questions are: Are we fine with adding this dependency? Are there big cons you see?
Obviously, I will make it optional like libudev is.

I can try to re-implement the functionality I need in mdadm but it is like reinventing the wheel.

Any feedback will be appreciated.

Hi Mariusz,

I am not against adding it to mdadm, though I think a better approach is to try and get the library built as a package for the distros.

Did you look into that yet?

Quoted 5 lines

We (as in 'We as an OS distributor') actually evaluated packaging libsed some time ago, but decided against it as the original authors (namely Intel) apparently disbanded it. So before adding it to a distro there needs to be an active maintainer, and one would be looking to Intel here.

Thanks Hannes for feedback. I totally agree with you. I will raise it internally.