Passphrase identification

[andrewkozlik]

Trezor allows handling multiple passphrases at the same time. If the user enters two different passphrases and then sees a GetAddress dialog, then they can see the path of the address, but not the passphrase it belongs to.

If more than one passphrase is active in the session cache, then Trezor should show some indication of which one is being used in GetAddress, SignTx and when confirming access to a SLIP-25 account.

How should the passphrases be identified:

• Temporary numeric label. The sequence number of the passphrase as it was entered on Trezor. Trezor should be able to tell if the user entered the same passphrase as in another session. In that case both sessions should have the same passphrase number. Requires adding one uint8_t to the session cache. The assigned sequence number would need to be shared with Suite, so that both Suite and Trezor use the same number (e.g. Electrum might be using Trezor with passphrase 1 and then Suite connects to Trezor with passphrase 2).
• Temporary text label from Suite. Suite can send a text label for the entered passphrase and the user will just confirm that they are happy with referring to the entered passphrase using that label. The label will then either:
  • be stored in the session cache or
  • linked with the session ID using a temporary MAC via a temporary key generated in Trezor. In every Initialize the label and MAC will be sent to Trezor and the MAC verified.
• Permanent text label. A label that is configured on Trezor only once with a long-term MAC that can be stored and used at any time in the future, see #1244.

Related issues:

• https://github.com/trezor/trezor-firmware/pull/2289
• https://github.com/trezor/trezor-firmware/issues/1244
• https://github.com/trezor/trezor-firmware/issues/2151

[Hannsek]

We can add the "name" of the wallet on the screen where the user sees the passphrase. Then we can add the "name" of the wallet also on the account detail screen in receive flow.

"name" here stands for "hidden wallet #X"

[matejcik]

That looks reasonable, and I can't foresee any good attack scenarios either.

The only thing comes to mind, malware wants you to spend from account A, so while unlocking account A, the label shown will be account B, we cross our fingers that the user doesn't notice, then we get them to unlock account B and spend from account B, but the transaction actually spends from A (which now has the same label). This already assumes the user is going to send funds to the attacker, so it doesn't actually seem to achieve much. Maybe prove that account A exists, but not in a communicable way: given that accounts A and B can't be linked, it could have been anyone who sent that transaction.

Malware could also cause user confusion by, e.g., forcing the same label for all sessions, but that's not worse than what we have now. (and, again, banking on the user not noticing, which is kind of the same as typo-attacking the passphrase entry).

(defending against the same label used by multiple passphrases does not seem worth the effort)

With temporary labels provided by Suite, we can even sidestep the whole labeling issue and trust Suite with whatever text it submits, so if your hidden wallet is named, Trezor can show that name.

[andrewkozlik]

Do I understand correctly, that you are in favor of the proposal which I referred to as "Temporary text label from Suite" above? If so then there is still the question of how to store the passphrase (in the session cache or using the temporary MAC solution) and the question of what to do if the host application does not provide a label, e.g. old version of Suite or some 3rd party tool.

[matejcik]

yes from me, how about @Hannsek

still several open questions:

• how to provide the string in the first place?
  • probably in a PassphraseAck field
• should firmware recognize standard wallet by itself; i.e., if passphrase is empty but the host claims this is "Hidden Wallet # 1", should we trust it?
  • probably yes, the user could have a custom label for their standard wallet?

if the string is not provided, we can use "Standard wallet" as a default for no-passphrase, "Hidden wallet" for passphrased. we could have an internal counter, but I don't like scanning open sessions for the same passphrase so this counter would increment every time and you'd get "hidden wallet #7" if it's the 7th time you are opening the same passphrase.

Temporary MAC would be better in that Trezor doesn't need to spend cache memory on this. A significant drawback is that this relies on the host to provide the string every time we need it, i.e., for every currency that wants to show which wallet you're spending from. (so in other words all of them). So unfortunately putting this into session cache seems like the only viable solution.

[andrewkozlik]

• how to provide the string in the first place?
  • probably in a PassphraseAck field

Sounds good.

• should firmware recognize standard wallet by itself; i.e., if passphrase is empty but the host claims this is "Hidden Wallet Redesign Passphrase #1", should we trust it?
  • probably yes, the user could have a custom label for their standard wallet?

Agreed, makes sense.

if the string is not provided, we can use "Standard wallet" as a default for no-passphrase, "Hidden wallet" for passphrased. we could have an internal counter, but I don't like scanning open sessions for the same passphrase so this counter would increment every time and you'd get "hidden wallet #7" if it's the 7th time you are opening the same passphrase.

Also makes sense, the only problem I see is collision with an already provided label. For example Suite provides the label "Hidden wallet #2", then another app that doesn't provide labels opens a passphrased wallet and we automatically assign the same label, because it's the second passphrased wallet.

Temporary MAC would be better in that Trezor doesn't need to spend cache memory on this. A significant drawback is that this relies on the host to provide the string every time we need it, i.e., for every currency that wants to show which wallet you're spending from. (so in other words all of them). So unfortunately putting this into session cache seems like the only viable solution.

I think it's sufficient to send the label and MAC only when switching between sessions, so only in Initialize. Then it can be stored only for the active session, not for each one individually.

[matejcik]

more significant: how to return the MAC to the host?

if the host provides label in PassphraseAck, there is no standard response that could carry the MAC back. we'd either need to insert an aux flow, like LabelMacRequest as a response to PassphraseAck (which will break every 3rd party)

or create a special flow, say, SetWalletLabel(label="what") -> PassphraseRequest -> PassphraseAck -> WalletLabel(mac=0xabcdef) which causes some corner cases:

• what if the user already entered the passphrase?
  • in that case the user did not see wallet label confirmation while entering, so we'd be introducing "really set wallet label??" which is just ugly ux
  • we could fail and force a new Initialize?
• what if you call SetWalletLabel multiple times in the same session?
  • detect and reject, probably, otherwise it's asking for the "really set label" thing

OTOH, we could do SetWalletLabel(label="what", mac=0xabcdef) as the obvious place to provide the mac

[andrewkozlik]

if the host provides label in PassphraseAck, there is no standard response that could carry the MAC back. we'd either need to insert an aux flow, like LabelMacRequest as a response to PassphraseAck (which will break every 3rd party)

if the host provides label in PassphraseAck, there is no standard response that could carry the MAC back. we'd either need to insert an aux flow, like LabelMacRequest as a response to PassphraseAck (which will break every 3rd party)

Nit: I think the LabelMacRequest would have to come after the ButtonAck, where the user confirms the passphrase on the Trezor screen.

I don't think it has to break 3rd parties. The LabelMacRequest can be sent to the host only if the host provided a label. (For unlabeled walets, i.e. for hosts that don't support the functionality, we'd have to store the wallet sequence number in the cache.)

I think the above can work fine in the scenario where the user is opening a wallet that is remembered in Suite. What seems more problematic are the following:

1. What to do if the user is creating a new wallet but actually enters the passphrase of an already labelled wallet? ("Passphrase duplicated" -> "Continue to discovered wallet" in Suite)
2. What to do if the user changes the wallet label in Suite after the wallet is loaded on Trezor? (This may also answer point 1.)

OTOH, we could do SetWalletLabel(label="what", mac=0xabcdef) as the obvious place to provide the mac

I don't understand. The label is coming from Suite whereas the mac is being returned by Trezor.

[Hannsek]

todo: https://github.com/trezor/trezor-firmware/pull/2925#issuecomment-1494088275

[SecurityIsHard69420666]

Another solution is to use a deterministic graphical icon (and/or accompanying text), just like my Github icon here.