BlockchainProvider2::block_state_provider may return inconsistent state provider

[nekomoto911]

Describe the bug

gravity-reth/crates/storage/provider/src/providers/blockchain_provider.rs

Lines 1116 to 1119 in 3c51ce0

	} else if let Some(state) = self.canonical_in_memory_state.state_by_hash(block_hash) {
	// ... or this could be tracked by the in memory state
	let state_provider = self.block_state_provider(state, opts)?;
	Ok(Box::new(state_provider))

gravity-reth/crates/storage/provider/src/providers/blockchain_provider.rs

Lines 174 to 183 in 3c51ce0

	fn block_state_provider(
	&self,
	state: impl AsRef<BlockState>,
	opts: StateProviderOptions,
	) -> ProviderResult<MemoryOverlayStateProvider> {
	let state = state.as_ref();
	let anchor_hash = state.anchor().hash;
	let latest_historical = self.database.history_by_block_hash(anchor_hash, opts)?;
	Ok(self.canonical_in_memory_state.state_provider(state.hash(), latest_historical))
	}

gravity-reth/crates/chain-state/src/in_memory.rs

Lines 510 to 522 in 3c51ce0

	pub fn state_provider(
	&self,
	hash: B256,
	historical: StateProviderBox,
	) -> MemoryOverlayStateProvider {
	let in_memory = if let Some(state) = self.state_by_hash(hash) {
	state.chain().into_iter().map(|block_state| block_state.block()).collect()
	} else {
	Vec::new()
	};
	MemoryOverlayStateProvider::new(historical, in_memory)
	}

In the block_state_provider, the in memory BlockState is first obtained from CanonicalInMemoryState, and with the anchor hash of this BlockState, an already written state provider in the database is retrieved, attempting to construct a MemoryOverlayStateProvider by combining the in memory BlockState and the state provider returned from the database. However, when constructing the MemoryOverlayStateProvider, the BlockState used is obtained again from CanonicalInMemoryState by hash, where this new BlockState may have an anchor hash different from the state provider obtained from the database (for example, the memory layout of CanonicalInMemoryState may have been modified by advancing persistent block between two calls to CanonicalInMemoryState::state_by_hash), resulting in an inconsistent CanonicalInMemoryState between the memory portion and the database portion.

Platform(s)

Linux (x86)

What version/commit are you on?

v1.0.7-gravity

Which chain / network are you on?

reth --chain

What type of node are you running?

Full via --full flag

Code of Conduct

• I agree to follow the Code of Conduct

[nekomoto911]

The latest version of reth has solved this problem. Merge the latest version of reth will solve it.

[nekomoto911]

paradigmxyz/reth#11338