You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Signature index: key by (target, signer) so honoring a drop cannot be crowded out by valid signatures from attacker keys #257
Since #254signatures::index_incoming and reindex index a signature only when its target is held and it verifies under the key it names (drop digest for tombstones), so forged or dangling signatures never enter the index. A signature made by a different key the attacker controls is still valid, so a writer to a shared remote can index more than MAX_SIG_READS (256) such signatures over a predictable tombstone id (it depends only on snapshot and transcript id) ahead of the real one. tombstone_honored reads the index in order and stops at 256, so the trusted signer's signature is never reached and the drop stays inert on every peer that ingests the junk. It is a denial of the drop, never an honored one.
Proposed: key the index per (target, signer) with dedupe, so a signer occupies one slot per target and honoring is a lookup over the trusted signers only (index lines gain the signer; old lines parse as unknown and are resolved by reading the object; reindex rebuilds). Honoring then reads at most one signature per trusted signer.
Related: ADR-0038 addendum (#230), THREAT-MODEL drop tombstones.
Activity
fix(repo): address review of the signature index and index lock (#257, …
fix(repo): hold the index lock across coupled passes, distinct busy e…
fix(repo): verify incoming signatures before taking the index lock (#257
fix(repo): skip trust re-evaluation while the legacy import is pending (
fix(repo): per-signer signature index and an index lock for serve-pat…