Wiping disk vs. initializing container
Hi there,
does overwriting an entire hard disk with random data before setting up encryption make initializing a LUKS container (as described in section 5.3 of the FAQ) obsolete?
Thanks
Michael
If it is crypto-grade randomness, yes.
But note that Section 5.3 may still be faster.
Regards,
Arno
Quoted 10 lines · On Tue, Jun 06, 2023 at 23:08:50 CEST, [email protected] wrote:
Hi there,
does overwriting an entire hard disk with random data before setting up encryption make initializing a LUKS container (as described in section 5.3 of the FAQ) obsolete?
Thanks
Michael
Signature · 7 lines
Arno Wagner, Dr. sc. techn., Dipl. Inform., Email: [email protected] GnuPG: ID: CB5D9718 FP: 12D6 C03B 1B30 33BB 13CF B774 E35C 5FA1 CB5D 9718 ---- A good decision is based on knowledge and not on numbers. -- Plato If it's in the news, don't worry about it. The very definition of "news" is "something that hardly ever happens." -- Bruce Schneier
Am Wed, 7 Jun 2023 05:30:38 +0200
schrieb Arno Wagner <[email protected]>:
Quoted 2 lines
If it is crypto-grade randomness, yes.
But note that Section 5.3 may still be faster.
I see.
Now I have some follow-up questions:
1. Would cat /dev/urandom > /dev/sdX give me crypto-grade randomness?
2. Is initializing a LUKS container with zeroes equivalent to filling a whole drive with crypto-grade random data if the LUKS container spans the whole disk?
3. The FAQ says:
If the target was in use previously, it is a good idea to wipe it before creating the LUKS container in order to remove any trace of old file systems and data.
So, isn't just filling the whole disk with random data before setting up the LUKS container the simplest solution if you want to a) destroy old data reliably, b) put the disk into a clean state, and c) make sure that parts of the LUKS container that have not been written to cannot be distinguished from those that have?
4. Should new hard disks that have not been used previously also be filled with random data in order to achieve c)?
Thanks
Michael
On 7 Jun 2023 12:03 +0200, from [email protected]:
Quoted 1 line
1. Would
cat /dev/urandom > /dev/sdXgive me crypto-grade randomness?
With a modern kernel, it should, yes.
Or you could set up a temporary mapping with a throwaway key and write from /dev/zero through that. The result should be much the same but my experience is that the latter is more performant.
Quoted 3 lines
2. Is initializing a LUKS container with zeroes equivalent to filling a whole drive with crypto-grade random data if the LUKS container spans the whole disk?
It should be. If not, that would be a more general vulnerability.
Quoted 11 lines
3. The FAQ says:
If the target was in use previously, it is a good idea to wipe it before creating the LUKS container in order to remove any trace of old file systems and data.
So, isn't just filling the whole disk with random data before setting up the LUKS container the simplest solution if you want to a) destroy old data reliably, b) put the disk into a clean state, and c) make sure that parts of the LUKS container that have not been written to cannot be distinguished from those that have?
Generally, absent the key, ciphertext should be indistinguishable from random data. So for an adversary who does not have access to the key, the disk having been filled with random data (e.g. cat from /dev/urandom) or a container having been filled with predictable data (e.g. cat from /dev/zero) should be largely equivalent.
Also, any random data should be as good as any other random data, so there would be no easy way to tell whether a particular chunk of random data is there because of an initial overwrite with random data, or because of useful data being stored through the later-created LUKS container.
Quoted 2 lines
4. Should new hard disks that have not been used previously also be filled with random data in order to achieve c)?
I would argue that it should, and by extension also for another
reason: allocation patterns can provide very strong clues both about the amount of data stored, and the type of file system within the container. A NTFS file system looks very different in terms of allocation patterns from an ext4 or Btrfs or exfat file system.
SSDs introduce other factors, particularly relating to wear leveling, where a different trade-off might be reasonable.
Signature · 2 lines
Michael Kjörling 🔗 https://michael.kjorling.se “Remember when, on the Internet, nobody cared that you were a dog?”
Quoted 6 lines · Michael Kjörling
On 7 Jun 2023 12:03 +0200, from [email protected]:
Quoted 3 lines
2. Is initializing a LUKS container with zeroes equivalent to filling a whole drive with crypto-grade random data if the LUKS container spans the whole disk?
It should be. If not, that would be a more general vulnerability.
Except for the header. There may be some space in the header or between header and data that does not get written this way. Hence I think I recommend using plain dm-crypt for a cyptsetup based wipe. The header and empty space is not a concern for data-patterns leaking from the container though. It is just a concern for removing old data.
Quoted 12 lines
Quoted 11 lines
3. The FAQ says:
If the target was in use previously, it is a good idea to wipe it before creating the LUKS container in order to remove any trace of old file systems and data.
So, isn't just filling the whole disk with random data before setting up the LUKS container the simplest solution if you want to a) destroy old data reliably, b) put the disk into a clean state, and c) make sure that parts of the LUKS container that have not been written to cannot be distinguished from those that have?
Quoted 5 lines
Generally, absent the key, ciphertext should be indistinguishable from random data. So for an adversary who does not have access to the key, the disk having been filled with random data (e.g. cat from
/dev/urandom) or a container having been filled with predictable data (e.g. cat from/dev/zero) should be largely equivalent.
It is. Anything else would allow you to break the cipher used.
Quoted 5 lines
Also, any random data should be as good as any other random data, so there would be no easy way to tell whether a particular chunk of random data is there because of an initial overwrite with random data, or because of useful data being stored through the later-created LUKS container.
That depends. Good quality non-crypto random data often allows you to determine the generator state and may even have blatantly obvious patterns. For example, I remember one generator that just toggled the smallest bit in the output stream. In that case, you can still determine what was written within the container later, because the ovwerwrite randomness has those patterns while the encrypted data does not.
Quoted 8 lines
Quoted 2 lines
4. Should new hard disks that have not been used previously also be filled with random data in order to achieve c)?
I would argue that it should, and by extension also for another
reason: allocation patterns can provide very strong clues both about the amount of data stored, and the type of file system within the container. A NTFS file system looks very different in terms of allocation patterns from an ext4 or Btrfs or exfat file system.
You put in the random data to reduce data leaking from the encrypted container. A new disk should hence be treated the same.
Quoted 2 lines
SSDs introduce other factors, particularly relating to wear leveling, where a different trade-off might be reasonable.
I would recommend to do it for SSDs also, just be aware it is not as good as for spinning disks. Still should help. Maybe do several random overwrites with SSD. Usually you have a few hundred before the disks becomes worn out.
Arno
Signature · 7 lines
Arno Wagner, Dr. sc. techn., Dipl. Inform., Email: [email protected] GnuPG: ID: CB5D9718 FP: 12D6 C03B 1B30 33BB 13CF B774 E35C 5FA1 CB5D 9718 ---- A good decision is based on knowledge and not on numbers. -- Plato If it's in the news, don't worry about it. The very definition of "news" is "something that hardly ever happens." -- Bruce Schneier