Skip to content

Follow-ups from the bug hunt of 2026-11-17 and the releases of 2026-11-18 #95

Open

Description

@maksimtech

Follow-ups for patchradar, in no particular order within a section.

Release notes

Scans

  • A judged scan can take hours: every MSRC- or KEV-only CVE costs one NVD request at 6 s each, with no cap (collectors/nvd.py records_for, with_records_for_the_rest). MSRC returned 1808 records for "edge" over 90 days, about 3 hours; the API's 600 s budget times out.
  • One non-CVE id stops the per-id lookups: MSRC returns ADV260001, NVD answers 404 "Invalid cveId", CollectorError ends the loop and the following CVEs are never asked. Send only CVE- ids, skip on 404.
  • No retry or backoff on NVD 403/429; pacing is per process, so a CLI scan and the API server pace independently.
  • A CVE stored before NVD scored it stays unscored: save_cve's ON CONFLICT refreshes EPSS and KEV only, not cvss_score / severity.
  • save_cve opens a connection and commits per row: about 133 ms per row on this disk; the Chrome target (2347 CVEs) took about 15 min, mostly saving.

Listings and UI

  • Listings apply LIMIT before the verdict: /api/cves?limit=200 returned 188 "not affected" rows while "not evaluable" ones were never shown; /api/stats counted CRITICAL 319, all not affecting; status showed 10 "not affected" of 20.
  • Rich markup in registry and tracker names: import with a name containing [/] crashes with MarkupError, [link=…] is interpreted (cli.py ~806, 820, 943; debian --file too).

Smaller

  • scripts/import-winget.ps1 skips 3 header lines where winget printed 2, losing the first package; 30 of 122 names rejected (+, (, ®); some names glued to version and id.
  • remove " 7-zip " does not strip the name (add and the API do).
  • _NUMERIC accepts any trailing letters, so "2.0b" sorts after 2.0 and is "not affected" by a CVE fixed in 2.0.
  • 7-Zip 24.08 prints "Updating to 26.02 closes 2 of 2" while "not evaluable" rows in the same table name fixes in 26.03 and 26.06.

Shared with the other Radars

  • OpenSSL CVE-2026-35190 (advisory of 2026-11-17, Low, fixed in 3.5.10: TLS 1.3 key-share with the Suite B flag, which Python's ssl never sets). No scanner reports it yet, so it has no exception: one written now would fail the gate as unused. When Docker Scout or Trivy reports it, add the entry like the other two of that advisory.
  • OpenSSL exceptions review on 2026-12-15 (the 2026-11-10 and 2026-11-17 entries). They also fail the gate as unused as soon as a rebuild picks up Debian's fixed openssl: then delete them.

Collected on 2026-11-18 from the bug hunt of 2026-11-17 (real inputs, evidence tests kept locally) and the releases of 2026-11-18. Items here were out of scope of the fixes already merged; each one gets its own pull request, test first.

🤖 Generated with Claude Code

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions