Crate craft_codec

Expand description

§CRAFT Codec

Cryptographic Random-Access Framing Toolkit.

With CRAFT, a Rust library, data streams can be compressed and encrypted as they are written. They can then be read from any byte position without earlier data being read or decoded.

Data is written sequentially.
A read starts at a chosen position and continues sequentially.
A new read can be started to access another position.

Compression and encryption are optional.

§How it works

Data is split into independent blocks called frames. Each frame can be compressed, encrypted, and read on its own.

Separate metadata records where the frames are stored. To read part of the original data, only the frames containing that part are retrieved and decoded.

For example, with 64 KiB frames, reading 4 KiB from the middle of a file requires decoding only one or two frames. Everything before them can be skipped.

The metadata must be saved for later reads. It contains the frame index, codec parameters and encryption key, so it must be kept in trusted, confidential storage.

§Installation

In Cargo.toml:

[dependencies]
craft-codec = { version = "0.2", features = ["lz4", "serde"] }

Requires Rust 1.85 or newer. AES-256-GCM encryption is enabled by default; the lz4 feature adds LZ4 compression. The optional serde feature implements Serialize and Deserialize for Metadata.

§Basic usage

Metadata contains the complete configuration for an object. Encoder and Decoder are constructed from it with TryInto, without separate codec or key arguments:

use craft_codec::{Compression, Config, Decoder, Encoder, Encryption, Framing, Metadata};

let mut metadata = Metadata::new(Config::new(
    Framing::Fixed(8),
    Compression::None,
    Encryption::None,
)?);
let mut encoder: Encoder = (&metadata).try_into()?;
let mut buffer = b"abcdefgh".to_vec();
let sizes = encoder.encode_frame(0, &mut buffer)?;
// Write the complete encoded frame successfully before recording its sizes.
metadata.push(sizes)?;

let (_stored_range, frames) = metadata.range(3, Some(7))?;
let mut decoder: Decoder = (&metadata).try_into()?;
for frame in frames {
    decoder.decode_frame(frame.spec, &mut buffer)?;
    let selected = buffer.get(frame.selected).ok_or(craft_codec::Error::InvalidRange)?;
    assert_eq!(selected, b"defg");
}

Compression::Lz4 enables compression. Encryption::Aes256Gcm { key } stores all encryption initialization data in the metadata. Deserializing Metadata from a database restores everything needed by TryInto. Codecs own their initialized state and retain no borrow of the metadata.

Read positions refer to the original data, before compression or encryption. For example, this selects bytes 3 through 18:

let (stored_range, frames) = metadata.range(3, Some(19))?;

The result identifies the stored range to read and the frames to decode. metadata.range(start, None) selects everything from start to the end.

The complete example writes data to memory and reads a selected range with compression and encryption disabled.

cargo run --example memory --no-default-features

The examples guide also covers integration with CARBON I/O for concurrent reads and writes.

§Frame size and compression

Frames can have a fixed size or vary up to a chosen maximum. Smaller frames reduce the extra data decoded for a small read. Larger frames reduce the space used by metadata and encryption tags.

Compression happens before encryption. If compression does not make a frame smaller, the original bytes are kept and encrypted if encryption is enabled.

§Encryption and data safety

Each file or blob must have a fresh independent secret key. Once encrypted, its contents must remain unchanged. Different data must never be encrypted with the same key and frame index. Retries may resend the same encrypted bytes; changed data requires a new key.

Metadata includes the encryption key and must be kept in trusted, confidential storage, linked to the correct file or blob. Encryption detects changes to the frames being read, but does not protect metadata or detect missing final frames on its own. Expected lengths and frame counts must also be checked by the application.

Without encryption, CRAFT does not detect data tampering. Data lengths and compression ratios remain visible even with encryption. Sensitive data is not securely erased from memory by CRAFT.

The format has not undergone a security audit. Encryption limits and detailed requirements are covered in the format specification.

§Metadata schema

Metadata carries a mandatory schema_version, currently 2, separate from the frame format version 1. Only schema 2, introduced in craft-codec 0.2.0, is supported. Metadata from releases before 0.2.0, missing schema versions and unknown schema versions are rejected. No legacy decoder or migration is provided.

The metadata enums can be deserialized even when a codec feature is disabled. Conversion to Encoder or Decoder then returns CompressionUnavailable or EncryptionUnavailable if that codec is required.

§More information

MIT licensed.

Structs§

Config: Validated object-wide settings, including all codec initialization parameters.
Decoder: Synchronous frame decoder. It knows no logical ranges or I/O offsets. Authentication precedes decompression. Reuse one instance and the caller’s buffer to amortize allocations. Compression may swap the buffer allocation.
Encoder: Synchronous frame encoder constructed entirely from metadata with reusable codec storage.
FrameIter: Borrowing, allocation-free iterator over a selected logical range.
FrameRead: One complete frame to decode, followed by a slice selected by the caller.
FrameSizes: Lengths returned by an encoder, before the authentication tag.
FrameSpec: Codec input derived from validated metadata, independent of logical slicing.
Metadata: Validated compact frame metadata, with cached scalar totals only. Append sizes after the corresponding frame has been written successfully.
MetadataParts: Self-contained external metadata, including keys. It is not self-authenticating. Persist it in trusted storage and restore with Metadata::from_parts.

Enums§

Compression: Compression identifier in trusted external metadata.
Encryption: Complete encryption parameters stored in trusted external metadata.
Error: A format, resource, or transformation failure. No I/O is performed by CRAFT.
Framing: Logical frame boundaries, before compression or encryption.
Lengths: Compact positive lengths, stored as length - 1.

Constants§

AES_MAX_BYTES: Upper bound on raw bytes covered by the AES profile’s frame slots per key. This is a library usage policy, not a claim of 128-bit security at this limit.
AES_MAX_FRAMES: Maximum number of AES-GCM frame slots per object/key.
AES_MAX_FRAME_LEN: Maximum configured raw frame length for the AES-GCM profile, 16 MiB.
FORMAT_VERSION: Version of the headerless frame data format.
METADATA_SCHEMA_VERSION: Metadata schema introduced in craft-codec 0.2.0. Older schemas are unsupported.
TAG_LEN: AES-GCM authentication tag length in bytes.

Type Aliases§

Result: Result of a CRAFT operation.