fix(tar-extract): bound pax parsing and enforce strict tar framing (PHARN-18)

[PrzemekGalarowicz]

Problem

• readPaxKeywords parsed g/x payloads of any size. A 192 KB archive whose payload expands to about 128 MB took ~13 s of CPU and ~1.1 GB of memory. Under a 512 MB heap the process crashed and left the temp clone behind.
• Framing was lenient:
  • a zero block in the middle of the archive silently dropped the rest of it;
  • a missing end marker, trailing garbage and a missing ustar magic were all accepted;
  • a//b and a/./b segments were caught only by safeJoin.

Fix

• A g/x payload over 64 KiB is refused on its size field alone, before any parsing. The codeload header is about 52 bytes.
• After the first zero block, every remaining byte must be zero.
• An archive with no end-of-archive marker is refused.
• The header must start with ustar (both POSIX ustar\0 and old GNU ustar are accepted).
• Empty and . path segments below the root are refused.

Tests

• 7 cases that the old code accepted are now refused. The codeload-shaped fixture still extracts.
• Local gates, run as a non-root user on node 22: all green. 1472/1472 tests pass and statement coverage is 97.26%.
• Pipeline artifacts are in .dev/features/tar-strict-framing/.

Advisory

THREAT-MODEL.md §2 (the list of extractor refusals) should gain these rules. That file is write-protected, so a maintainer has to add them.

🤖 Generated with Claude Code

https://claude.ai/code/session_01TvcuVhk8hTeDskp5pAJhnc

---

Generated by Claude Code

[coderabbitai]

Important

• 🔍 Trigger review

This repository does not receive automatic reviews because it has fewer than 10 stars.

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Advanced

Run ID: 53d21815-3004-41c9-aab4-81b22e2af51b

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}