diff options
Diffstat (limited to 'design-documents/009-backup.rst')
-rw-r--r-- | design-documents/009-backup.rst | 105 |
1 files changed, 105 insertions, 0 deletions
diff --git a/design-documents/009-backup.rst b/design-documents/009-backup.rst new file mode 100644 index 00000000..7ec8be2c --- /dev/null +++ b/design-documents/009-backup.rst @@ -0,0 +1,105 @@ +Design Doc 009: Wallet Backup +############################# + +Summary +======= + +This document describes the backup system used by Taler wallets. +This is the second, simplified iteration of the proposal, which leaves +out multi-device synchronization. + + +Requirements +============ + +* Backup must work both with and without Anastasis + + * When not using Anastasis, the user is responsible for keeping + their wallet's root secret safe. + +* Arbitrary number of backup providers must be supported +* Minimize information leaks / timing side channels **if** requested by the user +* Minimize potential to lose money or important information +* Since real-time sync is not supported yet, wallets should have a feature + where their whole content is "emptied" to another wallet, and the wallet is + reset. +* Even without real-time sync, the backup data must support merging with old, existing wallet + state, as the device that the wallet runs on may be restored from backup or be offline + for a long time. + + +Solution Overview +================= + +Each wallet has a 64 byte wallet root secret, which is used to derive all other secrets +used during backup, which are currently: + +1. The sync account key for a sync provider, derived via the sync provider's base URL. +2. The symmetric key used to encrypt the backup blob + +If the user chooses to use Anastasis, the following information is backed up in Anastasis: + +* list of used backup providers +* wallet root secret + + +Supported Operations +-------------------- + +* **restore-from-anastasis**: Start Anastasis recovery process. This requires + the wallet backup state to be "uninitialized". +* **restore-from-recovery-secret**: This requires the wallet backup state to be uninitialized. +* **add-provider** / **remove-provider**: Add/remove a sync provider from the + list of providers. Adding a provider will cause payment(s) to the provider + to be scheduled according to the provider's terms. If the wallet backup + state is "uninitialized", adding a provider will set the backup state to + "initialized" with a fresh wallet root key. Changing the provider list will + also update the backup provider URL list in the anastasis core secret. +* **abandon** / **takeover**: When the user wants to stop using a wallet on a particular + device, another wallet can "take over" by reading the recovery secret of the abandoned wallet. + The abandoned wallet marks explicitly in its backup blob that it is abandoned. + Abandoning a wallet will set the backup state to "uninitialized". +* **backup**: Do a backup cycle. + + +Backup Format +------------- + +TBD. Considerations from :doc:`005-wallet-backup-sync` still apply, +especially regarding the CRDT. + + +Initial User Experience +----------------------- + +The user will be asked to set up backup&sync (by selecting a provider) +after the first withdrawal operation has been confirmed. After selecting +the backup&sync providers, the user will be presented with a "checklist" that +contains an option to (1) show/print the recovery secret and (2) set up Anastasis. + +The wallet will initially only withdraw enough money to pay the +backup&sync/anastasis providers. Only after successful backup of the wallet's +signed planchets, the full withdrawal will be completed. + + +Open Questions +============== + +* Should the wallet root secret and wallet database be locally encrypted + and protected via a passphrase? +* What happens if the same Anastasis user has multiple wallets? Can Anastasis somehow + support multiple "instances" per application? + +Future Work +=========== + +* Incremental backups? + + * Instead of one big blob that always needs to be read/written, we could have (1) a + limited length append-only journal and (2) a merkle tree so that the backup blob can + be updated incrementally once the journal is full. + * Leaks more information and is more complex. + +* Real-time synchronization, either over some signaling server + or P2P connectivity (WebRTC, etc.) + |