Re: [PATCH 0/4][RFC v2] Introduce the in-kernel hibernation encryption

[Yu Chen]

On Mon, Aug 06, 2018 at 06:39:58PM +0800, joeyli wrote:
> On Mon, Aug 06, 2018 at 04:45:34PM +0800, Yu Chen wrote:
> > Hi Pavel,
> > On Sun, Aug 05, 2018 at 12:02:00PM +0200, Pavel Machek wrote:
> > > Hi!
> > > 
> > > > > User space doesn't need to involve. The EFI root key is generated by
> > > > > EFI boot stub and be transfer to kernel. It's stored in EFI boot service
> > > > > variable that it can only be accessed by trusted EFI binary when
> > > > > secure boot is enabled.
> > > > >
> > > > Okay, this apply to the 'suspend' phase, right?
> > > > I'm still a little confused about the 'resume' phase.
> > > > Taking encryption as example(not signature),
> > > > the purpose of doing hibernation encryption is to prevent other users
> > > > from stealing ram content. Say, user A uses a  passphrase to generate the
> > > 
> > > No, I don't think that's purpose here.
> > > 
> > > Purpose here is to prevent user from reading/modifying kernel memory
> > > content on machine he owns.
> > >
> > Say, A puts his laptop into hibernation and walks away,
> > and B walks by, and opens A's laptop and wakes up the system and he
> > can do what he wants. Although EFI key/TPM trusted key is enabled,
> > currently there's no certification during resume, which sounds
> > unsafe to me. Afterall, the original requirement is to probe
> > user for password during resume, which sounds more natural.
> 
> OK, I saw your case. This is a physical accessing.
> 
> I have a question: The suspend to memory also has the same behavior
> and more people are using suspend. Should we think a common solution
> to cover S3 and S4? 
>
Since STD behaves more likely a boot up, STR does not have solid
requirement for certification.
> > > Strange as it may sound, that is what "secure" boot requires (and what
> > > Disney wants).
> > > 
> > Ok, I understand this requirement, and I'm also concerning how to
> > distinguish different users from seeing data of each other.
> > 
> > Joey,
> > I'm thinking of a possible direction which could take advantage
> > of the password.  It leverages either EFI key or TPM
> > trusted key to get it done. Does it make sense?
> > 
> > 1. The user space generates a symetric key key_user using
> >    the password, and passes the key_user to the kernel as the master
> >    key.
> > 2. The kernel uses the EFI key or TPM trusted key to encrypt
> >    the key_user thus gets a encrypt_key.
> > 3. Uses the encrypt_key to do snapshot encryption
> > 4. During resume, the same encrypt_key is generated following
> >    the same steps(I assume the same EFI key or TPM key could be fetched
> >    during resumed, right?) and do the snapshot decryption.
> >
> 
> Yes, we can use TPM key to protect the user key. But I suggest that we
> should give user a function to disable the user key because not everyone
> want to key-in a password for hibernate/resume and also snapshot image
> encryption.
> 
> Two policies:
> - When user key-in user key, the snapshot image must be encryption.
> - Without key-in user key, I still want the snapshot image can be encryption.
> 
> No matter that the user key be key-in or not, the snapshot image must be
> encrypted by a kernel key. So I suggest that we treat the user key as a salt
> for snapshot image encryption and authentication. If the user key
> be disabled, then kernel just generates a random number as a salt.
> 
> Actually, the kernel must compares the user key before snapshot decryption.
> If the user key doesn't match but user space still triggers furture resume
> process. Then kernel direct drops the snapshot image. 
>  
Anyway I'm ok with using TPM for major 'security', please feel free
to send a second version out, and for certification implementation
we can have further discussion on that later.

Best,
Yu
> > And this is what fscrypt is doing:
> > Documentation/filesystems/fscrypt.rst
> >
> The use case is different. We have two key for two purposes. And the two
> functions can be separated.
> 
> Thanks
> Joey Lee