diff --git a/apps-baosec/vault2/src/main.rs b/apps-baosec/vault2/src/main.rs index 7d2a594b1..b3eb30098 100644 --- a/apps-baosec/vault2/src/main.rs +++ b/apps-baosec/vault2/src/main.rs @@ -34,80 +34,27 @@ use crate::vendor_commands::VendorSession; /* Dev status & notes -- -Aug 3 2025: +Nov 12 2025: -`cargo xtask baosec-emu` will launch the emulated UI. The aspect currently -in progress is the user interaction. Some firm conclusions so far are: +`cargo xtask baosec-emu` will launch the emulated UI. -- There is only going to be a TOTP and a password flow. FIDO interactions -will be through modal pop-ups only that alert of the relying party (if FIDO2). -FIDO-1 will notify that this is a legacy transaction with no identifier. -- We will use the same PDDB record format as vault original -- At the moment all "actions" are commented out in this app, so it is just a shell -used to refine the UI flow. Once the UI flow is refined we will add in the -live functionality -- Likewise there is no HID handler loop or camera loop -- Menu interactions will happen by pressing the "home" key and will be -done with an explicitly coded menu routine. This is because we don't have -a GAM in this implementation to handle layering/compositing in this UI framework. + -[ ] implement RTC + -[ ] implement QR code scanning loop + -[x] implement HID loop - needs testing, but have to implement RTC first before can test on HW + -[x] implement "action" loop --[ ] harmonized TOTP & password interactions? It is not yet clear if we are -going to have two separate UI flows, or if we will share elements between the -two. I am inclined to harmonize the two, which means reworking the TOTP flow -to have the search entry at the bottom. - -[x] TOTP interaction prototyped - -[ ] password interaction prototyped --[x] implement "home" button press menu mode --[ ] implement search by text entry - activated via both menu pop-up, and by -selecting >search prefix< on the bottom line of UI + -[ ] figure out interaction for searching for passwords + -[ ] search by QR code + -[ ] search by text - is this even a good idea? i think the UX is not great, so maybe we omit it entirely. -[ ] implement PIN management UI - entering a pin unlocks the deniable basis automatically -[ ] remove basis management calls when replacing with PIN calls -There are three action loops that need to be implemented. These are not -necessarily listed in order of implementation: - -[ ] implement QR code scanning loop - -[ ] implement HID loop - -[ ] implement "action" loop +-[ ] Consider the side channel resistance of the AES implementation in the keystore. We may want + to use an AES API that explicitly wraps the SCE's masked AES implementation to reduce side channels, + versus the Vex CPU core's AES implementation --[x] implement keystore - -[ ] Consider the side channel resistance of the AES implementation in the keystore. We may want - to use an AES API that explicitly wraps the SCE's masked AES implementation to reduce side channels, - versus the Vex CPU core's AES implementation - -[x] PDDB should have a base 256-bit key to protect all entries, stored in key store - This is equivalent to the "Backup key" in precursor - -[x] Base 256-bit key is derived from hash of 64kbits raw data scattered - across the RRAM array. Locations are chosen to be diverse, with the goal of reducing - voltage contrast due to parasitic leakage masking bits - -[x] 64kbit array is split into four 16kbit chunks, which are XOR'd together to create - a 16kbit number. The four 16kbit chunks are read out 256-bits at a time, with an order - determined by a random number on boot. The random ordering frustrates side channel attacks - on read-out of the array. Thus the format is: - - Kn, where n=[0..=3] - - Each Kn is 16384 bits long, composed of 64 256-bit blocks: Bm = {B0, B1, ..Bm}, - where m=[0..=255] - - The pre-hash key matter P is composed of K0 ^ K1 ^ K2 ^ K3, and the final key - is SHA512/256(P) - - P is derived specifically by visiting every one of KnBm in a random order and XOR'ing - them together, such that: - - The 16384-bit P is initialized to all 0's - - Break P into 64 separate 256-bit long blocks. Fill each block as follows: - - R is a 5-bit random number from the TRNG. It is rejection-sampled to generate a number - Rk that is [0..=23] (i.e. any number 24-31 is rejected and the TRNG is run again). - - Rk is turned into a permutation using the factoradic system: - - Derive factoradic digits [d₁, d₂, d₃, d₄] where: - - d1=k÷3!d1​=k÷3! - - d2=(kmod  3!)÷2!d2​=(kmod3!)÷2! - - d3=(kmod  2!)÷1!d3​=(kmod2!)÷1! - - d4=0d4​=0 (there should only be one element left at this point) - - Starting with the ordered list [A,B,C,D], at step i, - pick the element at position dᵢ and remove it. - - [A, B, C, D] represent visiting one of bank K0..=3 - - Fetching 256 bits from the respective bank and XOR it into the block position in P - - The final backup key is computed as SHA512/256(DS | P), where DS is a domain separator consisting - of 8 bits of 0's (8 bits is chosen simply because it's convenient, not to imply we could have only - 256 domains - you can always add more bits at the top of the hash). - - Other keys may be derived by changing the domain separator. - -[ ] The PIN may be optionally set to one of three values: +-[ ] deniable basis implementation + -[ ] The PIN may be optionally set to one of three values: -[ ] nothing (implemented as all 0's in the cryptographic key) -[ ] a passcode (4-8 digits entered via keyboard) -[ ] Configurable self-destruct after # of incorrect PIN guesses. Set at 8 by default. @@ -118,6 +65,7 @@ necessarily listed in order of implementation: -[ ] One deniable basis at a time is allowed. It is mounted using using only: -[ ] a PIN (4-8 digits entered via keyboard) -[ ] a QR code (generated using browser & managed by user) + -[ ] implement backups -[ ] PIN confirmation required to set system into backup mode -[ ] Backup key is displayed. This is 256-bit key, displayed as QR code or Bip-39.