Skip to content

v0.3.2

Choose a tag to compare

@github-actions github-actions released this 06 Nov 19:39
· 35 commits to main since this release

Closes three gaps that let content pass unscanned, shows incomplete runs as
such, and is the first release tested against popular and known malicious
images: see docs/popular-images.md.

Upgrade

A client of the API that treats every status other than infected and error
as clean has to learn the new status incomplete. Scans of images with large
archives take longer with the new limits. The operator scans its images again,
because the rules changed. With the CLI against your own clamd, add
meles-own.yar to clamd's database, or meles selftest --yara fails.

Security

  • The server stored runs with skipped files as clean and showed "No malware
    found". They now have the status incomplete in the UI and the API, like in
    the CLI and the operator, and count as needing attention. Stored runs are
    corrected at the first start.
  • Whiteout entries with content were not scanned: a file named .wh.<name>
    or .wh..wh..opq could carry malware through a scan as clean. A runtime
    never unpacks such an entry, but the image ships it. Its content is now
    scanned and a finding has the status deleted. The selftest has a new case
    for it, eicar-whiteout-content.
  • Data after the end of a layer's tar archive was read but not scanned. It now
    makes the scan Incomplete (meles.Unsupported.TrailingData), and it
    counts against --max-layer-size.
  • Sparse files could expand far beyond --max-layer-size, because the limit
    only counted the bytes read. It now counts what the entries declare, across
    the whole layer.
  • The mirror installs cvdupdate and its dependencies with pinned hashes
    (charts/meles/files/mirror/requirements.txt) instead of the newest
    matching version from PyPI. Dependabot updates the file.

Added

  • Fuzz tests for the layer extractor: make fuzz.

  • A refresh button with an interval in the header of the web UI. The browser
    remembers the interval; Auto refreshes while a run is active, as before.

  • Chart value scan.maxLayerSize for --max-layer-size. Machine learning
    images can have layers above the default of 4G.

  • YARA rules of our own in meles-own.yar, for xmrig-proxy, ariominer and the
    JSON configuration of XMRig and xmrig-proxy.

Changed

  • Higher default limits in the chart, because popular images stayed
    Incomplete with the old ones: clamd.maxFiles 100000 (was 10000),
    clamd.maxScanTime 600000 (was 120000), clamd.maxScanSize 2047M (in effect
    768M before). Images with large archives take up to twice as long, in the
    worst case 3.6 times, and clamd can unpack more into its ephemeral storage.

Fixed

  • A long scan could end with an error: a resumed layer download failed with
    403 once the signed CDN URL had expired, and with 401 once the token of the
    registry had. meles now asks the registry for new access.

Changelog

Features

  • a3bdc0c: feat(chart): higher default limits and scan.maxLayerSize (@movray)
  • ff6b058: feat(rules): own YARA rules for xmrig-proxy, ariominer and XMRig configs (@movray)
  • b09669b: feat(server): refresh button with an interval in the web UI (@movray)

Fixes

  • a78f6bf: fix(chart): install the mirror's Python packages with pinned hashes (@movray)
  • 5098af9: fix(ci): accept rule files that YARA-X passes without warnings (@movray)
  • e9644d1: fix(extract): scan whiteout content and limit what a layer can hide (@movray)
  • 57284a0: fix(server): show runs with skipped files as incomplete (@movray)
  • 8297b8f: fix(source): renew expired access when resuming a download (@movray)

Image: ghcr.io/movray/meles:0.3.2 (linux/amd64, linux/arm64)
Helm chart: helm install meles oci://ghcr.io/movray/charts/meles --version 0.3.2

Binaries, image and chart carry build provenance; verify with gh attestation verify
(see the README, section Install).