Swift Homomorphic Encryption
Swift Homomorphic Encryption is a Swift implementation of homomorphic encryption (HE) and applications including Private Information Retrieval (PIR).
Applications of Swift Homomorphic Encryption include:
* See pir-service-example for a sample serviceOverview
Swift Homomorphic Encryption is a collection of libraries and executables. For more information, refer to documentation for the libraries:- HomomorphicEncryptionProtobuf
- HomomorphicEncryption
- ApplicationProtobuf
- PrivateInformationRetrieval
- PrivateNearestNeighborSearch
Background
Homomorphic Encryption (HE)
Swift Homomorphic Encryption implements a special form of cryptography called homomorphic encryption (HE). HE is a cryptosystem which enables computation on encrypted data. The computation is performed directly on the encrypted data, without revealing the plaintext of that data to the operating process. HE computations therefore happen without decryption or access to the decryption key.HE thereby allows a client to enable a server to perform operations on encrypted data, and therefore without revealing the data to server. A typical HE workflow might be:
- The client encrypts its sensitive data and sends the resulting ciphertext to the server.
- The server performs HE computation on the ciphertext (and perhaps its own plaintext inputs), without learning what any ciphertext decrypts to.
- The server sends the resulting ciphertext response to the client.
- The client decrypts to learn the response.
[!WARNING]
BFV does not provide IND-CCA security, nor does it provide IND-CPAD security when there is a non-negligible decryption error probability. BFV should be used accordingly.
Therefore, no information about each decrypted ciphertext should be sent back to the server.
In particular, the noise budget of a ciphertext (computed via noiseBudget) must not be forwarded to any other party.
Sharing the decrypted ciphertext or noise budget acts as an oracle that can be used to recover the secret key.
To protect against a malicious server, the client should also validate the decrypted content is in the expected format.
> Consult a cryptography expert when developing and deploying homomorphic encryption applications.
Private Information Retrieval (PIR)
Private information retrieval (PIR) is one application of HE. PIR enables a client to perform a database lookup from a server hosting a keyword-value database, without the server learning the keyword in the client's query.. Each row in the database is a keyword with an associated value. During the PIR protocol, the client issues a query using its private keyword, and learns the value associated with the keyword.A trivial implementation of PIR is to have the client issue a generic "fetch database" request, independent of its private keyword. Then the server sends the entire database to the client. While this trivial PIR protocol satisfies the privacy and correctness requirements of PIR, it is only feasible for small databases.
The PIR implementation in Swift Homomorphic Encryption uses HE to improve upon the trivial PIR protocol.
[!WARNING]
PIR is asymmetric, meaning the client may learn keyword-value pairs not requested, as happens in trivial PIR for instance.
A variant of PIR, known as symmetric PIR, would be required to ensure the client does not learn anything about values it did not request.
Private Nearest Neighbor Search (PNNS)
Private nearest neighbor search (PNNS) enables a client with a private vector to search for the nearest vectors in a database hosted by a server, without the server learning the client's vector.. Each row in the database is a vector with an associated entry identifier and entry metadata. During the PNNS protocol, the client issues a query using its private vector, and learns the nearest neighbor according to a `DistanceMetric.
Specifically, the client learns the distances between the client's query vector to the nearest neighbor, as well as the entry identifier and entry metadata of the nearest neighbor.
A trivial implementation of PNNS is to have the client issue a generic "fetch database" request, independent of its private vector.
Then the server sends the entire database to the client, who computes the distances locally.
While this trivial PNNS protocol satisfies the privacy and correctness requirements of PNNS, it is only feasible for small databases.
The PNNS implementation in Swift Homomorphic Encryption uses homomorphic encryption to improve upon the trivial PNNS protocol.
Using Swift Homomorphic Encryption
Swift Homomorphic Encryption is available as a Swift Package Manager package.
To use Swift Homomorphic Encryption, choose a tag.
Then, add the following dependency in your Package.swift
.package(
url: "https://github.com/apple/swift-homomorphic-encryption",
from: "tag"),
, replacing tag with your chosen tag, e.g. 1.0.0.
To use the
HomomorphicEncryption library, add
.product(name: "HomomorphicEncryption", package: "swift-homomorphic-encryption"),
to your target's dependencies.
[!IMPORTANT]
When linking your executable, make sure to set
-cross-module-optimization.
Without this flag, performance of Swift Homomorphic Encryption degrades dramatically,
due to failure to specialize generics. For example,
> .executableTarget(
> name: "YourTarget",
> dependencies: [
> .product(name: "HomomorphicEncryption", package: "swift-homomorphic-encryption"),
> ],
> swiftSettings: [.unsafeFlags(["-cross-module-optimization"],
> .when(configuration: .release))]
> )
>
You can then add
import HomomorphicEncryption
to your Swift code to access the functionality in the HomomorphicEncryption library.
[!NOTE]
If you are using Swift Homomorphic Encryption for research, please cite using the
CITATION.cff file.
Examples
See the Snippets for examples using HomomorphicEncryption.
To list available snippets,
swift package show-executables | grep 'Snippet'
To run the
EncryptionParametersSnippet, run
swift run -c release EncryptionParametersSnippet
[!NOTE]
You may get an error saying that a testing framework was not found, e.g.
dyld[12358]: Library not loaded: @rpath/Testing.framework/Versions/A/Testing.
To fix on macOS, try one of the below workarounds, replacing
/Applications/Xcode.app as needed.
*
DYLD_INSERT_LIBRARIES=/Applications/Xcode.app/Contents/Developer/Platforms/MacOSX.platform/Developer/Library/Frameworks/Testing.framework/Versions/A/Testing swift run -c release EncryptionParametersSnippet
*
swift run -Xlinker -rpath -Xlinker /Applications/Xcode.app/Contents/Developer/Platforms/MacOSX.platform/Developer/Library/Frameworks/ -c release EncryptionParametersSnippet
Supported Platforms
Swift Homomorphic Encryption aims to support all of the platforms where Swift is supported.
[!NOTE]
Swift Homomorphic Encryption relies on SystemRandomNumberGenerator as a cryptographically secure random number generator, which may have platform-dependent behavior.
Swift / Xcode versions
The following table maps Swift Homomorphic Encryption package versions to required Swift and Xcode versions:
Package version | Swift version | Xcode version
----------------|---------------|-----------------------------------------
1.0.x | >= Swift 5.10 | >= Xcode 15.3
main | >= Swift 6.2 | >= Xcode 26
Source Stability
Swift Homomorphic Encryption follows Semantic Versioning 2.0.0. Source breaking changes to the public API can only land in a new major version, with the following exception:
- Adding a new
case to a public enum type will require only a minor version bump. For instance, we may add a new enum to an HeError. To avoid breaking source code, add a default case when adding a switch on the enum values.
Future minor versions of the package may introduce changes to these rules as needed.
We'd like this package to quickly embrace Swift language and toolchain improvements that are relevant to its mandate. Accordingly, from time to time, we expect that new versions of this package will require clients to upgrade to a more recent Swift toolchain release. Requiring a new Swift release will only require a minor version bump.
[!WARNING]
Any symbol beginning with an underscore, and any product beginning with an underscore, is not subject to semantic versioning: these APIs may change without warning.
Developing Swift Homomorphic Encryption
Dependencies
Developing Swift Homomorphic Encryption requires:
- Nick Lockwood SwiftFormat, 0.60.0
- pre-commit
- swift-format, 602.0.0
- swift-protobuf, 1.31.1
- SwiftLint, 0.63.2
Building
You can build Swift Homomorphic Encryption either via Xcode or via command line in a terminal.
After cloning the repository, run
cd swift-homomorphic-encryption
git submodule update --init --recursive
Xcode
To build Swift Homomorphic Encryption from Xcode, simply open the root directory in Xcode.
See the Xcode documentation for more details on developing with Xcode.
Command line
To build Swift Homomorphic Encryption from command line, open the root directory (i.e., the swift-homomorphic-encryption directory) of the cloned repository in a terminal, and run
swift build -c release
The build products will be in the .build/release/ folder.
To build in debug mode, run
swift build
The build products will be in the .build/debug/ folder.
[!WARNING]
Runtimes may be much slower in debug mode.
Installing
To install Swift Homomorphic Encryption targets, use the experimental-install feature of Swift Package Manager.
First ensure that the
~/.swiftpm/bin directory is on your $PATH.
For example, if using the zsh shell, add the following line to your ~/.zshrc
export PATH="$HOME/.swiftpm/bin:$PATH"
Make sure to reload the path via (source ~/.zshrc) or by restarting your terminal emulator.
Then, to install the
PIRProcessDatabase, executable, e.g., run
swift package experimental-install -c release --product PIRProcessDatabase
Testing
Run unit tests via
swift test -c release
To run tests in debug mode, run
swift test
[!WARNING]
Tests will be slow in debug mode.
Benchmarking
Swift homomorphic encryption uses Benchmark for benchmarking.
To enable benchmarking, set the environment variable SWIFT_HOMOMORPHIC_ENCRYPTION_ENABLE_BENCHMARKING=1.
By default, benchmarking requires the jemalloc dependency.
[!WARNING]
The benchmark may crash intermittently due to a known issue.
For reliable execution, benchmark can be run without
jemalloc as described here.
Two ways to run the benchmarks are:
- Xcode
* Open the swift-homomorphic-encryption folder in Xcode with SWIFT_HOMOMORPHIC_ENCRYPTION_ENABLE_BENCHMARKING=1 set, e.g.:
* open --env SWIFT_HOMOMORPHIC_ENCRYPTION_ENABLE_BENCHMARKING=1 Package.swift
* Switch to a benchmark target.
* Run the target, e.g., via the Product menu.
- Command line
* Run SWIFT_HOMOMORPHIC_ENCRYPTION_ENABLE_BENCHMARKING=1 swift package benchmark.
* See the Benchmark documentation for more information on running benchmarks.
Contributing
If you are interested in making a contribution to Swift Homomorphic Encryption, see our contributing guide.
Documentation
Swift Homomorphic Encryption uses DocC for documentation.
For more information, refer to the DocC documentation and the Swift-DocC Plugin.
Xcode
The documentation can be built from Xcode via Product -> Build Documentation`.
Command line
The documentation can be built from command line by runningSWIFT_HOMOMORPHIC_ENCRYPTION_ENABLE_DOCCPLUGIN=1 swift package generate-documentation
and previewed by running
SWIFT_HOMOMORPHIC_ENCRYPTION_ENABLE_DOCCPLUGIN=1 swift package --disable-sandbox preview-documentation --target HomomorphicEncryption