Fix infinite loop / duplicate rows on cyclic ESE leaf-page chain

[uziel110]

Summary

ESENT_DB.getNextRow() walks an ESE table's leaf pages by following each page's NextPageNumber. A healthy leaf chain is acyclic and terminates at NextPageNumber == 0.

However, a database in a dirty-shutdown / inconsistent state (e.g. an ntds.dit whose B-tree pointers were never committed) can contain a leaf chain whose NextPageNumber links back to an already-visited page. The current code follows that back-edge unconditionally, so it loops forever — re-emitting every row once per lap. In practice this hangs secretsdump.py LOCAL parsing and produces endless duplicate hashes.

Fix

Track the set of leaf pages already visited for each cursor (VisitedPages, plus CurrentPageNumber for accurate logging). When NextPageNumber points to a page that has already been walked, stop the table read and log an error noting the database may be inconsistent.

Because a valid ESE leaf chain never revisits a page, this only triggers on corrupt input and is a no-op for healthy databases. At the point the cycle is detected, every reachable page has already been read exactly once, so the table walk is complete.

Testing

• Verified against a dirty-shutdown ntds.dit that previously caused secretsdump.py -ntds ... LOCAL to loop and emit duplicate machine/user hashes — it now completes cleanly with the cycle-detection error logged.
• Healthy databases parse identically to before (no behavior change on the non-cyclic path).

🤖 Generated with Claude Code