#substrate #metadata #merkle-tree #chain #digest #shortener #frame-metadata

no-std metadata-shortener

Reference implementation for metadata shortener for Substrate chains, RFC46

2 releases

0.2.1 Feb 16, 2024
0.2.0 Feb 8, 2024
0.1.0 Jan 17, 2024

#2947 in Magic Beans

Download history 2/week @ 2024-01-14 2/week @ 2024-02-04 146/week @ 2024-02-11 38/week @ 2024-02-18 47/week @ 2024-02-25 6/week @ 2024-03-03 9/week @ 2024-03-10 1/week @ 2024-03-17 47/week @ 2024-03-31 113/week @ 2024-04-14

160 downloads per month

GPL-3.0-or-later

100KB
2K SLoC

metadata-shortener

Core functionality to shorten substrate metadata and provide proving mechanism in accordance with RFC0046

Supported metadata versions

Supported are metadata version RuntimeMetadataV15 and above.

RuntimeMetadataV14, while containing similarly structured types registry, is not supported, because the set of types itself is different, with RuntimeMetadataV14 having types not available in RuntimeMetadataV15 and vise versa. Additionally, extension types from SignedExtensionMetadata are referred through different id's in V14 and V15. While V14 and V15 could be both fetched from the node during the transitioning phase, supporting both would thus not be feasible. As V14 is becoming obsolete, it was decided to drop it altogether. Further versions (above V15) will retain compatibility.


lib.rs:

This crate is a shortener and digest generation tool for Substrate chains metadata.

Shortened metadata

During chain data parsing, only small fraction of the chain metadata is actually utilized.

Hardware signer devices with limited memory capability can encounter difficulties receiving and processing whole metadata which size is typically a few hundred kB. Receiving and using only the part required for decoding of particular data piece greatly simplifies the task, as the typical metadata part size decreases down to few kB.

Decoding of the signable transaction, or extrinsic, requires both information on extrinsic structure and the description of corresponding types. Signable transaction is built as a SCALE-encoded call and SCALE-encoded extensions concatenated to it. Call may or may not be double SCALE-encoded, i.e. preceded by compact of the call length.

Type describing all calls available is call_ty field in ExtrinsicMetadata. The extensions set is determined by signed_extensions in ExtrinsicMetadata.

ShortMetadata contains:

  • short types registry ShortRegistry with description of all types needed for signable transaction decoding (both for call and for extensions),
  • data from missing types, sufficient for Merkle tree root hash calculation (part of digest calculation, see below),
  • MetadataDescriptor with other relatively short data necessary for decoding and appropriate data representation

Note: chain specs (except base58 prefix in some cases) are a part of MetadataDescriptor, but are not in the full metadata, and should be fetched from chain and provided separately on ShortMetadata generation step, as ShortSpecs.

ShortRegistry is generated on the hot side, as the the transaction is preliminarily decoded and the types used are collected. Entries in ShortRegistry are PortableType values with unique id (same as in PortableRegistry) for type resolving and Type itself. For enums only the variants used in actual decoding are retained, all enum variants remain within a single entry.

ShortMetadata is generated with cut_metadata function for transactions with double SCALE-encoded call part (length-prefixed), and with cut_metadata_transaction_unmarked function for single SCALE-encoded call part.

ShortMetadata implements trait AsMetadata and could be used for chain data decoding using tools of substrate_parser crate.

SCALE-encoded ShortMetadata structure (as received by the cold side) is following:

  • ShortRegistry:
    • Compact of the number of types described in ShortRegistry
    • For each of the given number of types:
      • compact type id (same number as in original full metadata, for type resolving)
      • SCALE-encoded Type, encoded size is not known before decoding
  • Indices for Merkle tree leaves derived from types in ShortRegistry:
    • Compact of the number of indices for Merkle tree leaves derived from types in ShortRegistry
    • Given number of SCALE-encoded u32 indices, 4 bytes each
  • Merkle tree lemmas:
    • Compact of the number of lemmas for Merkle tree
    • Given number of lemmas, 32 bytes each
  • SCALE-encoded MetadataDescriptor:
    • 1-byte version of MetadataDescriptor (currently the only functioning variant is 1). For version 1:
      • id in types registry for the type describing all available calls
      • Signed extensions set:
        • Compact of the number of provided SignedExtensionMetadata entries
        • Given number of SCALE-encoded SignedExtensionMetadata, encoded size of each is not known before decoding
      • Compact length of the printed spec version followed by corresponding number of utf8 bytes
      • Compact length of the chain spec name followed by corresponding number of utf8 bytes
      • SCALE-encoded u16 base58 prefix value for the chain, 2 bytes
      • SCALE-encoded u8 decimals value for the chain, 1 byte
      • Compact length of the unit value for the chain followed by corresponding number of utf8 bytes

Example

use frame_metadata::v15::RuntimeMetadataV15;
use metadata_shortener::{
    traits::{Blake3Leaf, ExtendedMetadata},
    cut_metadata, ShortMetadata, ShortSpecs,
};
use parity_scale_codec::{Decode, Encode};
use primitive_types::H256;
use std::str::FromStr;
use substrate_parser::{parse_transaction, AsMetadata};

// Hex metadata string, read from file.
let meta_hex = std::fs::read_to_string("for_tests/westend1006001").unwrap();
let meta = hex::decode(meta_hex.trim()).unwrap();

// Full metadata is quite bulky. Check SCALE-encoded size here, for simplicity:
assert_eq!(291897, meta.len());

// Full `RuntimeMetadataV15`, ready to use.
let full_metadata = RuntimeMetadataV15::decode(&mut &meta[5..]).unwrap();

let specs_westend = ShortSpecs {
    base58prefix: 42,
    decimals: 12,
    unit: "WND".to_string(),
};

// Transaction for which the metadata is cut: utility batch call combining
// two staking calls.
let data = hex::decode("c901100208060007001b2c3ef70006050c0008264834504a64ace1373f0c8ed5d57381ddf54a2f67a318fa42b1352681606d00aebb0211dbb07b4d335a657257b8ac5e53794c901e4f616d4a254f2490c43934009ae581fef1fc06828723715731adcf810e42ce4dadad629b1b7fa5c3c144a81d55000800b1590f0007000000e143f23803ac50e8f6f8e62695d1ce9e4e1d68aa36c1cd2cfd15340213f3423e5b1d91c89d3de85a4d6eee76ecf3a303cf38b59e7d81522eb7cd24b02eb161ff").unwrap();

// Make short metadata here. It is sufficient to decode the transaction.
let short_metadata =
    cut_metadata(&data.as_ref(), &mut (), &full_metadata, &specs_westend).unwrap();

// `ShortMetadata` is substantially shorter. SCALE-encoded size:
assert_eq!(4486, short_metadata.encode().len());

// Now check that decoding result remains unchanged.

// Transaction parsed with shortened metadata, carded:
let parsed_with_short_meta = parse_transaction(
    &data.as_ref(),
    &mut (),
    &short_metadata,
    None,
)
.unwrap()
.card(
    &<ShortMetadata<Blake3Leaf, ()> as ExtendedMetadata<()>>::to_specs(&short_metadata)
        .unwrap(),
    &<ShortMetadata<Blake3Leaf, ()> as AsMetadata<()>>::spec_name_version(&short_metadata)
        .unwrap()
        .spec_name,
);

// Transaction parsed with full metadata, carded:
let parsed_with_full_meta = parse_transaction(
    &data.as_ref(),
    &mut (),
    &full_metadata,
    None,
)
.unwrap()
.card(
    &specs_westend,
    &<RuntimeMetadataV15 as AsMetadata<()>>::spec_name_version(&full_metadata)
        .unwrap()
        .spec_name,
);

// Call parsing result for short metadata (printed cards, without documentation):
let call_printed_short_meta = parsed_with_short_meta
    .call_result
    .unwrap()
    .iter()
    .map(|card| card.show())
    .collect::<Vec<String>>()
    .join("\n");

// Call parsing result for full metadata (printed cards, without documentation):
let call_printed_full_meta = parsed_with_full_meta
    .call_result
    .unwrap()
    .iter()
    .map(|card| card.show())
    .collect::<Vec<String>>()
    .join("\n");

// Call parsing results did not change.
assert_eq!(call_printed_short_meta, call_printed_full_meta);

// Extensions parsing result for short metadata (printed cards, without documentation):
let extensions_printed_short_meta = parsed_with_short_meta
    .extensions
    .iter()
    .map(|card| card.show())
    .collect::<Vec<String>>()
    .join("\n");

// Extensions parsing result for short metadata (printed cards, without documentation):
let extensions_printed_full_meta = parsed_with_full_meta
    .extensions
    .iter()
    .map(|card| card.show())
    .collect::<Vec<String>>()
    .join("\n");

// Extensions parsing results did not change.
assert_eq!(extensions_printed_short_meta, extensions_printed_full_meta);

Metadata digest

The decoding of the chain data is beneficial from safety viewpoint only if the metadata can be guaranteed to be authentic. A possible solution to that would be to produce a digest of the metadata and concat it to the signable transaction prior to signing, so that the signature would be valid only if the metadata used for decoding matches the one on chain. This crate generates such digest, both for complete and for shortened metadata.

Digest is generated by merging the root hash of the Merkle tree build over metadata's PortableRegistry with the hash of SCALE-encoded MetadataDescriptor.

Merkle tree for types data

Merkle tree is generated and processed using tools of merkle_cbt and merkle_cbt_lean crates. While providing the same outcome, merkle_cbt_lean is tailored for no_std environments with low internal memory capacity and external (streamed) data.

Merkle leaves are blake3-hashed SCALE-encoded individual PortableType values. In enums the same id is used for every retained variant, and every retained variant is placed as an individual enum with a single variant.

For full metadata RuntimeMetadataV15, all leaves are constructed, deterministically sorted, and processed to build the Merkle tree, and then the root hash. In ShortMetadata, the available types data is transformed into leaves and combined with MerkleProof to calculate the root hash.

Trait HashableRegistry for producing sorted set of Merkle tree leaves is implemented for PortableRegistry and for ShortRegistry.

Trait HashableMetadata for producing Merkle tree root hash is implemented both for RuntimeMetadataV15 and for ShortMetadata. Complete digest could be calculated for HashableMetadata if ShortSpecs are provided.

ShortMetadata also implements trait ExtendedMetadata for digest calculation and transaction parsing without providing additional data.

Metadata descriptor

MetadataDescriptor contains other relatively short data necessary for decoding and appropriate data representation:

  • id in types registry for the type describing all available calls
  • set of signed extension metadata entries SignedExtensionMetadata
  • chain spec name and spec version (extracted from Version constant of the System pallet)
  • chain specs (base58 prefix for in-chain Ss58 address representation, decimals and unit for balance values representation)

MetadataDescriptor is versioned to simplify version compatibility check on the hardware side.

Example

use frame_metadata::v15::RuntimeMetadataV15;
use metadata_shortener::{
    cut_metadata,
    traits::{Blake3Leaf, ExtendedMetadata, HashableMetadata},
    MetadataDescriptor, ShortMetadata, ShortSpecs,
};
use parity_scale_codec::Decode;
use substrate_parser::AsMetadata;

// Hex metadata string, read from file.
let meta_hex = std::fs::read_to_string("for_tests/westend1006001").unwrap();
let meta = hex::decode(meta_hex.trim()).unwrap();

// Full `RuntimeMetadataV15`, ready to use.
let full_metadata = RuntimeMetadataV15::decode(&mut &meta[5..]).unwrap();

let specs_westend = ShortSpecs {
    base58prefix: 42,
    decimals: 12,
    unit: "WND".to_string(),
};

// Full metadata digest:
let digest_full_metadata =
    <RuntimeMetadataV15 as HashableMetadata<()>>::digest_with_short_specs(
        &full_metadata,
        &specs_westend,
        &mut (),
    )
    .unwrap();

// Same transaction as in above example.
let data = hex::decode("c901100208060007001b2c3ef70006050c0008264834504a64ace1373f0c8ed5d57381ddf54a2f67a318fa42b1352681606d00aebb0211dbb07b4d335a657257b8ac5e53794c901e4f616d4a254f2490c43934009ae581fef1fc06828723715731adcf810e42ce4dadad629b1b7fa5c3c144a81d55000800d624000007000000e143f23803ac50e8f6f8e62695d1ce9e4e1d68aa36c1cd2cfd15340213f3423e5b1d91c89d3de85a4d6eee76ecf3a303cf38b59e7d81522eb7cd24b02eb161ff").unwrap();

// Generate short metadata:
let short_metadata =
    cut_metadata(&data.as_ref(), &mut (), &full_metadata, &specs_westend).unwrap();

// Short metadata digest:
let digest_short_metadata =
    <ShortMetadata<Blake3Leaf, ()> as ExtendedMetadata<()>>::digest(
        &short_metadata,
        &mut ()
    ).unwrap();

// Check that digest values match:
assert_eq!(digest_short_metadata, digest_full_metadata);

RuntimeMetadata versions support

RuntimeMetadataV14 implements trait AsMetadata and could be used for transactions decoding.

Trait HashableMetadata could be implemented for RuntimeMetadataV14 (and, in fact, was, in earlier editions of this crate), but intentionally is not.

The types registry of RuntimeMetadataV14 has structure similar to that of RuntimeMetadataV15, however, the types in registries for same spec_version are different in V14 and V15, with RuntimeMetadataV14 having types not available in RuntimeMetadataV15 and vise versa, thus making it not feasible to support both simultaneously during the transitioning phase.

V15 and above are expected to be supported.

Available features

  • merkle-standard: for calculating RuntimeMetadataV15 digest using tools of merkle_cbt crate. Intended for signature checking side. Digest is constant while metadata spec_version remains the same.

  • merkle-lean: for calculating ShortMetadata digest on cold signer side using tools of merkle_cbt_lean crate.

  • proof-gen: for generating ShortMetadata on wallet side, using tools of merkle_cbt_lean crate. proof-gen feature includes merkle-lean.

  • std

By default, all features are made available.

Dependencies

~8.5MB
~163K SLoC