This section documents some of our team's thinking regarding operating the Coldcard in HSM mode and the CKBunker. These notes come from our internal discussions while the feature was being developed.
It's important that entry into HSM mode must always require a user OK on the device screen. It would be bad if a desktop program could program a bogus policy, and then start HSM mode and steal your funds.
The Coldcard should not be generous and share what the current/proposed HSM policy is. Knowing the rules allows you to stretch them... for example, the web page, how much does it show about the policy (after it's setup)? Or the list of usernames, is that a semi-secret, since it identifies humans who might be pressured?
Since our model is normally that we don't trust the desktop, initial
thinking was we should only show these things on the Coldcard screen,
and maybe during setup process. However, as the CKBunker was
developed, it became obvious that this semi-private data was
pretty important, so we made this choice a setting:
Some USB commands are inappropriate in HSM mode: for example, firmware upgrade should not be possible. We use a fixed whitelist to check this.
What if: An operator has the PIN, but not the authority to spend all the money. She reboots the Coldcard and enters PIN. Then uploads a generous new HSM policy file that allows her to do anything. Puts CC into HSM mode, and steals the money later. Or she simply copies the seed words, or makes a backup, etc.
Conclusion: We assume those with the PIN have full authority over all the funds (as we did before) but we implemented the Boot to HSM feature, with possibility to completely lock down operation to always be in HSM mode.
The 60-second timeout on entering the escape code is there because it if a local operator learns the escape code, but doesn't have the master PIN, they shouldn't be able to escape HSM mode. A person with knowledge of the master PIN, can safely leave the Coldcard after unlocking it with master PIN and waiting 60 seconds.
A policy involving user accounts that don't exist is invalid. Therefore, deleting a user can impact HSM mode, so we don't allow that during operation.
Why SHA1 inside our TOTP implementation? Because Google Authenticator only supports that, based on this document.
FreeOTP is better and supports SHA256, but it does not seem to be widely used. We also don't have space in our QR code to indicate alternative hashing algorithms.
If the bunker is captured while the Coldcard is unlocked, the attacker will learn your master xpub. We cannot disable that because the master xpub is used to protect the USB communications privacy (MitM protection).
Therefore to stop the bad guys from finding all your transactions; past and future, just don't use the master xpub directly, and instead, use a derived path (like all wallets do already).
We crash the Coldcard (secure logout) if there are more than 100 refused transactions. This is our main protection against any sort of brute-forcing of the policy rules.
There is a potential race condition with local user confirmation and multiple remote users: you call in and tell the operator to do code 378555 (which is the next code) but before you can upload your PSBT, your evil co-worker uploads his PSBT which uses that local confirmation code first. Your PSBT fails (wrong code) and evil coworker has avoided the local-confirmation check.
To address this, the local confirmation code is a
function of the PSBT being approved and a random value (seed/salt)
picked by the Coldcard. The PIN is effectively:
BE32(31 LSB) of HMAC(msg=psbt hash, key=bytes from CC, sha256) mod 1000000
share_addrs is effectively a proper superset of
whenever you sign a message, you are sharing the public key for
that path, and from the public key, you can deduce the segwit or
classic payment address which corresponds.
HMAC-based passwords won't work after backup/restore to a different Coldcard.
We used the original Coldcard's serial number in the key stretching. So although we restore that value, unless the backup is restored onto same Coldcard, it won't be right. TOTP users won't have same issue because shared secret is stored raw.
There is no way to detect if this situation is happening, so the Coldcard cannot handle this case. Please use TOTP instead of passwords or reset user passwords after a restore.
If we assume your USB-attached computer (CKBunker or otherwise) is fully compromised by attackers, we might worry that it's poised and ready to apply a number of USB commands as soon as you unlock your Coldcard, and before you enter HSM mode.
That's why you are prompted at power-up to go into HSM mode, if a policy file is detected. The USB port is not yet enabled at that point, so if you do go to HSM mode, there is no window in which unprotected USB commands can be launched.
boot_to_hsm mode also addresses the same concern, since it assumes
the answer is Yes to that boot-up question.
When "boot to HSM" feature is enabled, and there are any per-period limits, we start off with the current period active, and all maximum amounts spent. In other words, after boot up, they will have to wait-out the velocity period before those rules can be used.
This is needed because the "remote hands" may have access to the master PIN code, but not have full spending authority. Therefore, we cannot allow those hands to reset the spending period because they could circumvent the velocity limits that way.
If the "boot to HSM" feature isn't enabled, we don't need to have this policy because anyone who can reboot the device has the master PIN, and could just as well disable the HSM entirely... so velocity limits don't apply to them.
Any HSM which can perform unattended signing operations is at risk of leaking key information through side channels. Since multiple requests can be performed, even a "statistical" leak of a bit or two of the private key could add-up over time to be a big risk.
Even with all possible countermeasures in software, we still think there is a risk of side channel data leaks through the USB power or data lines. Mk3 Coldcard has a new filtering circuit, but it's hard to cost-effectively mitigate this risk when the budget of the receiving equipment is effectively unlimited.
Therefore, if this type of attack is a concern, we recommend putting your Coldcard HSM into a physically-secure (locked) EMI-shielded box, such as this experiment box from Holland Shielding and isolating the USB power and signal using an adapter such as this 5kV USB Isolator and a DC to DC convertor.
If there is enough interest, we will make a lockable metal box, with suitable EMI gasketing, and specialized power isolation circuit inside for the USB connection. Please contact firstname.lastname@example.org if you'd like more information.
We've found relevant search results.