Summary

The Hickory DNS project's experimental hickory-recursor crate's record cache (DnsLru) stores records from DNS responses keyed by each record's own (name, type), not by the query that triggered the response. cache_response() in crates/recursor/src/lib.rs chains ANSWER, AUTHORITY, and ADDITIONAL sections into one record iterator before insertion. The bailiwick filter it applies uses the zone context of the NS pool that serviced the lookup, not the zone being queried.

This creates a cross-zone poisoning path. When Hickory builds the NS pool for attacker.poc. it uses the parent poc. NS pool (ns.zone() = "poc."). If the poc. nameserver under the attacker's control includes in its response's AUTHORITY section a record for a sibling zone like victim.poc. NS ns.evil.poc., the bailiwick check is_subzone("poc.", "victim.poc.") passes (victim.poc. is a subdomain of poc.). The record is stored under (victim.poc., NS) in the shared cache.

Subsequently, any client querying a name in victim.poc. causes Hickory to build its NS pool from the poisoned cache entry, routing queries to the attacker's nameserver (ns.evil.poc.) rather than to the legitimate nameserver for victim.poc.. The legitimate NS for that zone receives zero queries.

This issue is fixed in hickory-resolver 0.26.0 with the recursor feature through an architectural change to response-level caching: responses are stored keyed by the originating query (name, type). A response to (attacker.poc. NS) is stored only under that key and cannot affect the (victim.poc., NS) cache entry.

Hickory DNS believes this issue has been present in all published versions of the experimental hickory-recursor crate, which has now been folded into the hickory-resolver crate under the non-default recursor feature flag. The hickory-recursor crate will not receive any updates going forward and all users should migrate to hickory-resolver with the recursor feature....