PaperVault

Overview

PaperVault was written with the goal of handling sensitive information. PaperVault follows the principle information at rest should be securely encrypted using a secret only known to the user. This includes both the documents stored on the application itself in the user's device, and the printed documents. It's important to note there is no recovery mechanism for documents either stored on the device or printed. If the user forgets the password, the data is considered lost.

All the encryption algorithms are implemented by Apple CryptoKit framework, which is part of Apple operating systems. PaperVault uses CryptoKit APIs to perform encryption, decryption and related security activities.

This application does not use the network, does not perform any network requests nor does it store data in any cloud service. Data is encrypted and encoded on the printed QR Codes themselves. Per the normal behaviour of the operating system, encrypted data may still travel across the network when backed up and printed to networked printers. See the Network section for more details.

Device document storage

As a quick overview, each document is encrypted with a document-specific key. The key is itself encrypted with a global key, and stored in the file system. The global key is generated during the app initial setup, encrypted by a key derived from the main password provided by the user, and also stored in the file system.

Global key

During the initial setup, a random 256-bit key is generated. This is the global file encryption key. To protect that key, a temporary key is derived from the main password the user provides. The derivation of that temporary key is done through the following algorithm, using a random 128-bit salt and 10 million derivation rounds:

CCKeyDerivationPBKDF(
    CCPBKDFAlgorithm(kCCPBKDF2),
    passwordPtr, strlen(passwordPtr),
    saltBuf.baseAddress!.assumingMemoryBound(to: UInt8.self),
    saltBuf.count,
    CCPseudoRandomAlgorithm(kCCPRFHmacAlgSHA256),
    rounds,
    &output, output.count)

The global encryption key is then encrypted (wrapped, using AES.KeyWrap) by the temporary key, and stored on the file system, together with the salt and number of derivation rounds. The temporary key is discarded. This means the only way to obtain the key necessary to unwrap the global key is performing a similar derivation of the temporary key from the main password, the saved salt and number of rounds.

Main Password

During app setup (and any time when changing the main password) the user may choose to store the main password in the device keychain. This is only possible if the user has configured authentication on the device (like Face ID or Touch ID, or a Pin Code), something the vast majority of the users do. On macOS, if biometrics are not available, the password cannot be stored in the keychain.

If the user opts to store the main password in this way, the device will require authentication when launching the application, through biometrics if configured, falling back to the Pin Code (on iOS) if no biometrics are available or the biometric-based authentication fails. If the user does not provide a valid biometric or Pin Code, the application falls back to requiring the user to type the main password, as if no password had been stored in the keychain. If the user opts for not storing the main password in the keychain, the application always asks for the main password when launching.

Either way, if authentication is not provided, the application wont proceed since it would not be possible to decrypt the stored content, or encrypt new content.

When the user changes the main password, the global key is unwrapped using a temporary key derived from the old password as described above, a new temporary key is derived from the new password, the global key is wrapped by the new temporary key, and stored on disk, replacing the old stored key.

Document specific key

When a document is created, a new random 256-bit key is generated specifically for that document. That key is wrapped by the global key, and stored in the file system. Each document's title and content is encrypted with its document specific key using the ChaCha20-Poly1305 algorithm. Before encryption, a random number of random bytes is added to the plain-text. This is used to intentionally increase the size of the data by an arbitrary amount, preventing an attacker from knowing the real size of the encrypted data. This is useful in situations where a message contains only a password, where knowing the exact size of the content can reduce the search space in a relevant amount. The padded data is added to the pain-text message itself, so it's subject to ChaCha20 encryption and Poly1305 authentication, preventing certain classes of attacks. When a document is deleted, its document-specific key is deleted from the file system as well. This means that even if an attacker is able to obtain any left-over information related to the document, they wont be able to decrypt it since the document-specific key is gone.

When the user changes the main password, the file specific keys are not affected, since they are wrapped by the global key. Even if the global key is re-wrapped by the temporary key derived from the new password so it can be stored safely in the file system, the global key itself never changes.

Printed documents

The information printed on the QR Codes consists of an encrypted data payload split by one or more segments (one per QR Code). Each QR Code contains that encrypted payload segment, and some non-encrypted information that guides PaperVault while scanning (like the total number of codes and a document identifier). More information about the data structures can be found in the data format page. The guiding data contains no sensitive information. For the purposes of this page, that information will be ignored and the focus will be on the data payload.

Before proceeding, it's important to note a document printed by PaperVault includes the document title printed in human-readable text in the page(s) header. The title is also included in the encrypted stream, so it can be scanned as part of the document. However, since the title is printed in human-readable form, it should not contain sensitive information.

The payload starts with one byte used for control flags. These flags are not part of the encrypted data stream, and are reserved for future information that needs to be present before decryption may happen. The rest of the payload is the encrypted data stream containing the document (content and title) plus another set of control flags (these ones encrypted themselves as part of the encrypted data stream, with information needed to decode the document after decryption takes place). The data format page contains all the details. For the purposes of this page, the important bit is how this information is encrypted.

When printing a document, the user may choose between using their main password (the same password used to encrypt the storage encryption key) or a different, document-specific password. Regardless of which password is used, similarly to what happens with documents stored on the device, a PBKDF2 function is used to derive an encryption key from the user-provided password, and a random salt. Also similarly to what was described before, 10 million rounds are used to derive the key. This key is used directly as the encryption key for the printed document's encrypted stream. Just like stored documents, a random number of random bytes is added to the plain-text stream before encryption. The stream is encrypted using the ChaCha20-Poly1305 algorithm as well, and prepended with the information needed to decrypt it (number of rounds and salt).

When the user scans a document, the payload is decrypted deriving the key from the user-provided password in the same way as during the encryption process. PaperVault will try to decrypt using the current user's main password. If that fails, the user will be prompted for the correct password for the scanned document.

Network

PaperVault does not make any network calls. As stated before, when a document is printed, the data is encrypted and encoded in the QR Codes themselves, not in any cloud service. On macOS, PaperVault is sandboxed without network access. That said, there are situations when encrypted data may be transmitted across the network or to different devices as part of normal operation of your devices. In these situations, the data is always encrypted, be it in printed form (QR Codes) or data stored on device, except for the document title when printing.

These situations include backups (iCloud backups or backups to a Mac or PC on iOS, and any backup system used on macOS, like Time Machine and other third party backup products). You can find more about what is backed up below. Other situations include:

Migrations to new devices. In this situation, the stored documents and keys will be migrated. If the main password is stored in the keychain, it will not be migrated to the new device (read the backups section below for more details on keychain management).
Printing to networked printers. This is handled directly by the operating system, and PaperVault does not control nor implement communication with printers. The data sent to a printer is always encrypted, and encoded in the QR Codes, except for the document title. Keep in mind the document title is printed in human readable form without any protection, so don't store sensitive information in the document title.
Continuity camera. This is the process of using an iPhone camera to scan printed documents that are decoded by the macOS version of PaperVault. During usage, the image captured by the iPhone camera is streamed to the Mac through any available means (USB cable, wi-fi networks, ad-hoc network established automatically between the iPhone and the Mac, etc). This is implemented by the operating systems, not PaperVault, and implementation details (including security) may change without PaperVault's author control.
Explicit user action. This include shared clipboard across devices, moving files to external store or through the network, etc. Users are ultimately responsible for the security of their data and understanding the consequences of their actions. For example, if you copy unencrypted data from PaperVault's document editor and paste it on a different device using Universal Clipboard, you are exposing unencrypted data through that mechanism. PaperVault's author is not responsible for the security of that or any other operating systems' features.

Backups

Many users configure their devices to regularly back up their data. On iOS, this includes backing up to iCloud or a computer (Mac or PC). On macOS, there are quite a few options, from Time Machine to locally attached or network volumes, to third party products that can use a panoplia of backup destinations, like multiple cloud services, SFTP servers, etc. Read this section to learn how information is backed up and how that affects security.

On iOS, if iCloud backups are turned on, PaperVault data can be excluded from backups by configuring the device appropriately using iOS Settings. As far as the author is aware, there is no way to do the same when backing up to a computer.

Here's what is backed up:

If the user opts for storing the main password in the device keychain, protected by biometrics or Pin Code, the password is stored using kSecAttrAccessibleWhenPasscodeSetThisDeviceOnly. This tells the operating system to protect the password with a device specific key. The password will be backed up as part of the device's keychain, but can only be restored to the same device. If the backup is restored to a different device, the user must type their main password again. This protects against an attacker who has access to a backup from restoring it to a different device, since, without the password, the restored PaperVault data will be useless. If you're a developer, here's a good summary of the different keychain accessibility values.
The database containing the documents, the document-specific encrypted keys, and the global encrypted key are fully backed up. Since the main password is necessary to decrypt the global key, which is itself necessary to decrypt the document specific keys (which are themselves necessary to decrypt the documents' title and content), the backed up data will be useless for anyone who doesn't know the main password.

Remember that, by default, iCloud backups are not protected by end-to-end encryption, meaning your backed up data can be obtained by third parties, including surrendered to authorities, in specific situations. Apple offers Advanced Data Protection in most jurisdictions. Advanced Data Protection enables end-to-end encryption, protecting your backups. Read more about Advanced Data Protection on Apple's website to understand how to turn it on and its implications in your devices usage, specifically recovery methods.

Rationale and Considerations

PaperVault security is based on a user typed password. The rationale for this is: One main use case of PaperVault is being able to recover information after a total loss of the user's digital devices and cloud services (this happens more often than people realize). Therefore, PaperVault printed documents cannot rely on anything that requires non-memorable keys, hardware keys, or any kind of device-specific key.
Stored documents could be protected with more sophisticated security schemes, like a device-specific key stored in the device's Secure Enclave, and even be password-less. However, that would complicate and likely make backups impossible. Since the user needs to deal with passwords anyway (see previous item), and can securely delete a document right after printing it (or opt for not save a scanned one), this seemed like a good compromise between functionality, usability and security. Different options (a more secure mode, without backup capability, for example) may be offered in the future, since this decision is not written in stone.
The stored documents are kept in SwiftData, which is a wrapper for CoreData, and uses SQLite underneath. This means document data may not be deleted immediately after the document is discarded, or old versions of the document may remain after the document is edited a newer version is stored. The document-specific key protects against the first situation: even if the bits remain in SQLite for some time, the key to decrypt them should be securely deleted by iOS, rendering the bits useless. There is no protection about the second situation.
The temporary key derivations from passwords, both for stored and printed documents, use 10 million rounds. Although this makes a brute force attack extremely computationally expensive, attackers (especially government backed actors) may have access to massive computational resources. Although extremely small (negligible in most situations), there is theoretically a risk a brute force attack could succeed. The use of strong, complex passwords (using lower and upper case characters, numbers, and symbols) mitigate this. Remember, no security method in computer science protects against weak passwords like "qwerty" or "12345". It's your responsibility to pick a long, complex password you can remember.
Since the secure enclave is not used for security, the unencrypted keys remain in the operating system accessible memory during execution as needed. Due to the extremely well designed iOS/macOS security and application sandboxing, it's very unlikely an attacker may obtain the keys this way. A very small risk exists of a compromised app being able to steal data in memory from another app, although from the moment a device is compromised by an attacker, all bets are off and nothing running on that hardware, currently and in the future, should be considered secure. Keep also in mind that this applies to iOS os macOS devices running the latest version of the OS, unmodified. On a jail-broken device, you should consider the typical iOS security guarantees provided by sandboxing and process isolation simply do not exist. You should also consider macOS is a much more powerful platform than iOS that comes with additional security risks. Be mindful of the software you install and its origin.

PaperVault Security

A note about security