CVE scanning enhancement#8437
Open
stutiibm wants to merge 7 commits into
Open
Conversation
Adds generalized_wheel_scanner.py — a single-script CVE scanner that runs
inside the build container after auditwheel repair to detect vulnerabilities
in bundled .so files that standard scanners cannot identify (metadata stripped,
hash-renamed filenames).
Scanner is invoked automatically from create_wheel_wrapper.sh after auditwheel
repair completes. Failure is non-blocking — build continues regardless.
Scanning strategy (three lanes + phase 1):
Phase 1 — grype scan directly on extracted wheel directory.
Results deduplicated against Lane 1 and Lane 2 before final report.
Lane 1 — Walks METADATA/egg-info inside the wheel to find all Python runtime
dependencies, then queries grype for CVEs.
Lane 2 — Strips auditwheel hash from each bundled .so filename, matches it
against a live RPM inventory (rpm -q --filesbypkg -a) to identify
the source package, then queries grype for CVEs.
Lane 3 — Parses the build script (git clone + git checkout lines) to extract
source-built C/C++ library names and versions. Queries NVD CPE API
to resolve a unique identifier, then NVD CVE API for findings.
All .so files not claimed by Lane 2 are attributed to Lane 3 libs.
Other details:
- grype self-installs at runtime (3 retries, 10s delay); non-blocking if it fails
- Shell variable substitution pre-pass handles $VAR / ${VAR:-default} in build scripts
- Layered NVD CPE search: raw repo name first, stripped name as fallback
- Language-binding CPE variants (-python, -java, etc.) filtered out
- Output: single unified JSON report per wheel saved to disk for COS upload
Fields in final_report:
total_unique_cve — globally distinct CVE IDs across all lanes
total_cve — per-library occurrence count
cpe_resolved — true = unique NVD CPE found; false = flagged, no CVE query
…per.sh
Adds the generalized_wheel_scanner.py invocation block to create_wheel_wrapper.sh.
The scanner is called after auditwheel repair completes and before SHA generation,
passing the repaired wheel and the original build script as arguments.
The call is non-blocking — if the scanner fails or the script is not found,
a warning is printed and the build continues normally.
Scanner invocation:
if [ -f "generalized_wheel_scanner.py" ]; then
python generalized_wheel_scanner.py "${wheel_final}" "${BUILD_SCRIPT_PATH}"
fi
Arguments passed:
$1 — repaired wheel file (manylinux-tagged .whl)
$2 — original build script path (used by Lane 3 to parse source-built libs)
Output: <wheel_stem>_cve_report.json written to the build working directory,
picked up and uploaded to COS by the workflow upload step.
Adds an "Upload CVE report to COS" step to all five wheel build jobs
(py310, py311, py312, py313, py314) in the build workflow.
The step runs after the wheel artifact upload and pushes the
<wheel_stem>_cve_report.json file produced by generalized_wheel_scanner.py
to the same COS bucket and path as the existing build logs:
ose-power-toolci-bucket-production/<PACKAGE_NAME>/<VERSION>/<wheel_stem>_cve_report.json
Example for numpy 2.2.5 (py312):
ose-power-toolci-bucket-production/numpy/2.2.5/
numpy-2.2.5-cp312-cp312-manylinux_2_34_ppc64le_cve_report.json
Design decisions:
- Reuses existing upload_file.sh — no new upload infrastructure needed
- Non-blocking — missing report only prints a warning, build continues
- One report per wheel per Python version, all landing under the same
PACKAGE_NAME/VERSION prefix, no extra collector job required
- No changes to create_wheel_wrapper.sh or any scanner script
Added CVE report renaming logic after wheel post-processing.
stutiibm
marked this pull request as ready for review
July 23, 2026 13:37
cdeepali
reviewed
Jul 24, 2026
| } | ||
|
|
||
| # Grype install URL | ||
| _GRYPE_INSTALL_URL = ( |
Contributor
There was a problem hiding this comment.
Add a step in workflow to install grype, instead of installing it in the python script?
| try: | ||
| # Step 1: one call — all package names + versions | ||
| meta_result = subprocess.run( | ||
| ['rpm', '-qa', '--queryformat', '%{NAME}|%{VERSION}-%{RELEASE}\n'], |
Contributor
There was a problem hiding this comment.
Can you try with, it will give package name, version and files all in one go.
rpm -qa --queryformat 'PKG:%{NAME}-%{VERSION}-%{RELEASE}.%{ARCH}\n[%{FILENAMES}\n]'
cdeepali
reviewed
Jul 24, 2026
| print(" Detected RPM-based system") | ||
| inventory['packages'] = self._get_rpm_packages() | ||
| # Check if DEB-based system | ||
| elif shutil.which('dpkg'): |
Contributor
There was a problem hiding this comment.
Is there a need to prepare inventory for DEB based system, considering that the build happens on RHEL container?
cdeepali
reviewed
Jul 24, 2026
| try: | ||
| # Get all packages | ||
| result = subprocess.run( | ||
| ['dpkg-query', '-W', '-f=${Package}|${Version}\n'], |
Contributor
There was a problem hiding this comment.
On Debian systems, dpkg stores the list of files for every installed package in plain text files under /var/lib/dpkg/info/.list. Reading these files directly in Python is better compared to executing the dpkg-query and then dpkg for each package.
cdeepali
reviewed
Jul 24, 2026
| 'basename': Path(file_path).name | ||
| }) | ||
| except: | ||
| pass |
Contributor
There was a problem hiding this comment.
Can use except Exception: here instead of passing silently.
cdeepali
reviewed
Jul 24, 2026
| # libgfortran-2758a8fd.so.5.0.0 -> libgfortran | ||
| # libopenblasp-r0-a8f83a82.3.32.so -> libopenblasp-r0 | ||
| # libabsl_base-eb206faa.so.2401.0.0 -> libabsl_base | ||
| base_name = re.sub(r'-[a-f0-9]{8,}', '', bundled_name) # Remove hex hash |
Contributor
There was a problem hiding this comment.
Can this be replaced with single regex?
cdeepali
reviewed
Jul 24, 2026
| site_dirs = [] | ||
| try: | ||
| result = subprocess.run( | ||
| ['find', '/', '-name', 'site-packages', '-type', 'd', |
Contributor
There was a problem hiding this comment.
Can the following be used instead?
import site
dirs = site.getsitepackages() + [site.getusersitepackages()]
| queue = [wheel_pkg_name] | ||
|
|
||
| while queue: | ||
| pkg = queue.pop(0) |
| queue = [wheel_pkg_name] | ||
|
|
||
| while queue: | ||
| pkg = queue.pop(0) |
| if result.returncode == 0 and shutil.which('grype'): | ||
| print(" grype installed successfully.") | ||
| return True | ||
| print(f" Attempt {attempt} failed: {result.stderr.strip()[:200]}") |
Contributor
There was a problem hiding this comment.
Use constant instead of magic number 200.
| with urllib.request.urlopen(req, timeout=30) as resp: | ||
| return json.loads(resp.read()) | ||
| except Exception as e: | ||
| print(f" Warning: NVD API call failed ({url}): {e}") |
Contributor
There was a problem hiding this comment.
Check that this does not print the Key here.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Checklist
set -eoption enabled and observe success ?