GHSA-3whf-vgf2-9w6g

Описание

Summary

NonFinalizedState::handle_reorg is a recursive, unbounded async function that traverses parent blocks until it finds a common ancestor on the main chain. It has no recursion depth limit and no cycle detection. A malicious or buggy validator can serve a block whose previous_block_hash points back to itself (or forms a cycle with other blocks), causing handle_reorg to infinite-loop, consuming 100% CPU and never making sync progress. Additionally, update() contains an .expect("empty snapshot impossible") that panics if the non-finalized snapshot becomes empty after trimming finalized blocks.

Details

Location: packages/zaino-state/src/chain_index/non_finalised_state.rs:443-489

Infinite loop via self-referencing block:

1. A compromised validator serves a block B where B.prev_hash == B.hash.
2. handle_reorg is called with B.
3. get_block_by_hash_bytes_in_serialized_order(B.prev_hash) finds B itself in working_snapshot.blocks.
4. Check: is B.hash in working_snapshot.heights_to_hashes? If B is a new chaintip not yet on the main chain, no.
5. Recurse with prev_block = B (the exact same block).
6. This repeats forever. The async recursion builds a new Box::pin future each iteration, consuming heap memory and CPU.

Stack exhaustion via deep reorg: A deep reorg of >1000 blocks would recurse >1000 times. Each async recursion creates a new Box::pin future on the heap. While this won't exhaust the native stack immediately, it will allocate unbounded heap memory and CPU time, effectively DoS-ing the sync task.

.expect("empty snapshot impossible") panic:

Location: packages/zaino-state/src/chain_index/non_finalised_state.rs:543-548

If finalized_height is greater than or equal to all blocks in new_snapshot.blocks, remove_finalized_blocks retains only blocks at or above that height. If none exist, new_snapshot.blocks becomes empty. The .expect() then panics. While the comment claims this is "impossible," defensive programming dictates it is reachable under corruption or edge-case sync conditions.

PoC

1. Run a regtest.
2. Serve a block where header.previous_block_hash == block.hash().
3. Zaino's NonFinalizedState::sync enters handle_reorg and infinite-loops.
4. Sync never completes. CPU usage pegs to 100%. No new blocks are served to clients.

Fix

1. Add an explicit recursion depth limit (e.g., max 1000 iterations) and return SyncError::ReorgFailure if exceeded:
   const MAX_REORG_DEPTH: usize = 1000;
2. Track visited hashes in a HashSet<BlockHash> during traversal to detect cycles and abort with an error.
3. Replace .expect("empty snapshot impossible") with a proper Err(UpdateError::DatabaseHole) or similar error return.

Additional Attack Vectors

• Deep reorg DoS: A miner with significant hash power (or a compromised validator) triggers a deep reorg. Zaino spends excessive CPU and memory in handle_reorg, starving the async runtime and stalling response serving.
• Fork-choice manipulation: By serving cyclic or very deep sidechains, an attacker can keep Zaino stuck in reorg handling indefinitely, preventing it from ever serving the real best chain.