Free and open source · Apache-2.0
Patch Proof: a free, open-source CVE scanner for Linux servers
See which CVEs are really open on an Ubuntu, Debian, AlmaLinux or Rocky Linux server, counting the fixes your distro backported and the kernel that's actually running.
Patch Proof is a free, open-source (Apache-2.0) vulnerability scanner that shows which CVEs are really open on an Ubuntu, Debian, AlmaLinux or Rocky Linux server. It compares every installed package with the distribution's own security data, so fixes the distro backported count as fixed, and it judges kernel CVEs against the kernel that's actually running.
It's read-only. A short shell script reads the package list over SSH, needs no root and installs nothing, and the report is built on your own machine as one HTML file (plus JSON). On our fully patched Debian 12 and AlmaLinux 9.8 test servers it found 0 CVEs with a patch waiting. It also listed 430 and 1,530 CVEs as fixed by a distro backport, each with the build that fixed it, as evidence you can hand an auditor. In our study, Trivy, Grype and OSV-Scanner didn't report any of those against the build that carries the fix either; the differences were the running kernel, the backport evidence and that nothing is installed on the server.
pipx install git+https://github.com/HostCrust/patch-proof@v1.0.0
hc-proof collect --script | ssh user@server sh -s | hc-proof report --input - --out report.html --label web-01
- License
- Apache-2.0
- Version
- 1.0.0 (November 14, 2026)
- Runs on
- Linux, macOS
- Needs
- Python 3.10+ and SSH. No root on the server
- Written in
- Python
What is Patch Proof?
HostCrust Patch Proof is a command-line CVE checker for Linux servers. A short POSIX shell script, the collector, prints the server's OS release, running kernel and installed packages. The hc-proof command on your own machine matches that list against the security data each distribution publishes through OSV.dev, then writes one self-contained HTML report. Every CVE lands in one of five classes: patch available but not installed, patched on disk but not running, no distro fix yet, fixed by backport, and fixed normally. Only the first two need action, and each comes with the fixed version and a link to the distribution's own advisory. Version 1.0.0 was released on November 14, 2026 under the Apache-2.0 license. It's written in Python with nothing outside the standard library, runs on Linux and macOS, and checks Ubuntu 20.04 to 24.04, Debian 11 and 12, and AlmaLinux and Rocky Linux 8 and 9.
| Report section | What it means | What to do |
|---|---|---|
| Patch available, not installed | The distro has published a fixed build and the installed one is older | Install the update |
| Patched on disk, not running | A fixed kernel is installed, but the server still runs an older, affected one | Reboot into the newest kernel |
| No distro fix yet | The distro lists the installed version as affected and has no fix (including issues it deferred or rated unimportant) | Track it, or mitigate if it applies to how the server is used |
| Fixed by backport | The installed build has the distro's fix, made in a new build of the same upstream version | Nothing. Keep the row as evidence if a scanner flags it |
| Fixed, normal | Fixed by a newer upstream version | Nothing (counted in the summary only) |
Why do vulnerability scanners flag CVEs that Debian, Ubuntu and RHEL already fixed?
Distributions fix most security bugs by backporting: they patch the version they already ship instead of moving to the new upstream release. Debian 12 still ships OpenSSH 9.2p1, but fixed CVE-2024-6387 (“regreSSHion”) in its own build 1:9.2p1-2+deb12u3, while upstream lists 8.5p1 to 9.7p1 as affected and fixed it in 9.8p1. A tool that matched the upstream version alone would call every Debian 12 server vulnerable. That's what matching against the upstream version ranges in CVE records, such as NVD's, does, and network scanners that judge a service by its version banner can do the same. The package scanners we tested don't do it for installed packages: Trivy, Grype and OSV-Scanner all compare the installed build with the distribution's own fixed build, and none of them reported CVE-2024-6387 for web-01's installed openssh-server. (Grype did list it once, from an old openssh-server_1:9.2p1-2 .deb file left in the apt cache, which is a different question: that file really is the unfixed build.) Patch Proof uses the distribution's data too, and labels each such CVE “Fixed by backport” with the build that fixed it, so you can hand the evidence to an auditor. Where the scanners did differ on 14 Nov 2026 was the kernel, the evidence they show and what they need on the server. The sections below show the measured differences, and the limits.
Here are four results our report files as “Fixed by backport”. A version-only match would call each of them open; the three scanners we ran did not report them for installed packages.
| Server | Installed | CVE | Upstream fixed in | Distro fixed in |
|---|---|---|---|---|
| web-01, Debian 12 | openssh 1:9.2p1-2+deb12u10 | CVE-2024-6387 | OpenSSH 9.8p1 | 1:9.2p1-2+deb12u3 (DSA-5724-1) |
| web-01, Debian 12 | curl 7.88.1-10+deb12u15 | CVE-2023-38545 | curl 8.4.0 | 7.88.1-10+deb12u4 (DSA-5523-1) |
| web-02, AlmaLinux 9.8 | curl 7.76.1-40.el9_8.9 | CVE-2023-38545 | curl 8.4.0 | 7.76.1-26.el9_3.2 (ALSA-2023:6745) |
| web-02, AlmaLinux 9.8 | kernel 5.14.0-687.66.2.el9_8 (running) | CVE-2024-1086 | A newer upstream kernel; RHEL 9 stays on 5.14.0 | 5.14.0-427.13.1.el9_4 (ALSA-2024:2394) |
The kernel is where the scanners we ran differed most. On Debian 12, Trivy, OSV-Scanner and Grype's package matcher reported no kernel CVEs, because Debian builds linux-image-* from the source package linux-signed-amd64 while its tracker files kernel CVEs under linux. Patch Proof maps the running kernel to linux and reported 1,994 CVEs with no Debian fix for 6.1.190-1; Debian's tracker, checked about five hours after Patch Proof's data snapshot, still listed 1,984 of them open. Grype matched the /boot kernel image files against NVD instead and reported 6,112 CVEs: 5,262 are fixed in the running kernel, 136 never affected Debian 12, 708 are open and 6 aren't in Debian's tracker under linux. On AlmaLinux, Trivy reported 795 CVEs, all on an old kernel package that isn't running. Trivy's request to filter kernels that aren't running (#3764) is still open. See our full comparison. Patch Proof checks kernel CVEs against the running kernel's exact package version, and reports a fixed kernel that hasn't been booted as “Patched on disk, not running”. Our guide to checking the running and installed kernel version shows how to do the same check by hand.
What did Patch Proof find on two fully patched servers?
No patch waiting. On November 14, 2026 (UTC) we ran Patch Proof 1.0.0 against two HostCrust test servers that had just been fully patched and rebooted: web-01 runs Debian 12.15 and web-02 runs AlmaLinux 9.8. Both reported 0 CVEs with a patch available but not installed, and 0 kernel fixes waiting for a reboot, which is what a fully patched server should show. The interesting numbers are the backports. On web-01, 430 CVEs in 53 source packages are fixed in Debian builds of an older upstream version, including CVE-2024-6387 in OpenSSH and CVE-2023-38545 in curl. On web-02 the figure is 1,530 CVEs in 46 packages, and 1,246 of them are kernel CVEs, because RHEL-family kernels keep the version 5.14.0 for the whole release. A tool that compared upstream version strings alone would report all of those as open; Trivy, Grype and OSV-Scanner didn't report any of them against the build that carries the fix (see the comparison below). Neither server needs any action for them.
These are the commands and output of one run on 14 Nov 2026 (UTC), from a Mac. The first run downloads the 41 MB index; later runs use the cache.
$ hc-proof collect --script | ssh web-01 sh -s | hc-proof report --input - --out web-01.html --label web-01 --json web-01.json
Downloading the prebuilt index from https://github.com/HostCrust/patch-proof/releases/download/index-latest/index.sqlite.gz ...
6 of 41 MB
9 of 41 MB
13 of 41 MB
17 of 41 MB
21 of 41 MB
26 of 41 MB
33 of 41 MB
40 of 41 MB
verified SHA-256 c237a39de3f84f5040a6bc7b6b4a0b6e2d45d7c4090788670cc4ec60cbd3dcf5; index built 2026-11-14T14:10:05Z
Using the prebuilt index built 2026-11-14T14:10:05Z (OSV Debian snapshot 2026-11-14T14:03:09Z).
Debian GNU/Linux 12 (bookworm): open 0, notrunning 0, nofix 2643, backport 430, fixed 16076. Wrote web-01.html in 25.8 s.
$ hc-proof collect --script | ssh web-02 sh -s | hc-proof report --input - --out web-02.html --label web-02 --json web-02.json
Using the prebuilt index built 2026-11-14T14:10:05Z (OSV AlmaLinux snapshot 2026-11-13T16:03:25Z).
AlmaLinux 9.8 (Olive Jaguar): open 0, notrunning 0, nofix 0, backport 1530, fixed 351. Wrote web-02.html in 4.6 s.
| Result (distinct CVEs) | web-01: Debian 12.15 | web-02: AlmaLinux 9.8 |
|---|---|---|
| Patch available, not installed | 0 | 0 |
| Patched on disk, not running | 0 (running 6.1.190-1, the newest installed) | 0 (running 5.14.0-687.66.2.el9_8, the newest installed) |
| Fixed by backport | 430 in 53 packages | 1,530 in 46 packages (1,246 kernel) |
| No distro fix yet | 2,643 in 82 packages (1,994 kernel) | Not in the data (see below) |
| Fixed, normal | 16,076 (13,220 kernel) | 351 |
| Installed packages | 372 | 346 (rpm -qa lists 347; the extra entry, gpg-pubkey, is AlmaLinux's imported signing key) |
Three things in these numbers are worth knowing before you run it on your own servers:
- Debian's kernel fixes are mostly “normal”, not backports. Debian 12 moves to new upstream 6.1.y stable releases (the running kernel is 6.1.190), so its 13,220 fixed kernel CVEs came with newer upstream versions. RHEL-family kernels stay at 5.14.0, which is why AlmaLinux shows the opposite pattern.
- “No distro fix yet” is a tracking list, not an alarm. Debian's tracker lists 2,643 CVEs as unfixed in bookworm for packages on web-01. Debian rates 292 of them “unimportant”, and 1,994 are kernel CVEs. The report shows the distro's rating and CVSS score for each, so you can pick out the few that apply to how the server is used.
- AlmaLinux and Rocky Linux data lists released fixes only. So the “No distro fix yet” class can't be filled for them, and the report says so at the top: “The AlmaLinux OSV data lists released fixes (security advisories) only.”
How do I install and run Patch Proof?
You need Python 3.10 or newer on your own machine (Linux or macOS) and SSH access to the server. Install the hc-proof command, pinned to the release:
pipx install git+https://github.com/HostCrust/patch-proof@v1.0.0
hc-proof --version # hc-proof 1.0.0
No pipx? A virtual environment works the same way, and it's what we used for this page:
python3 -m venv ~/.venvs/hc-proof
~/.venvs/hc-proof/bin/pip install git+https://github.com/HostCrust/patch-proof@v1.0.0
Then use one of these ways to run it:
- Over SSH, from your machine:
hc-proof collect --script | ssh user@server sh -s | hc-proof report --input - --out report.html --label web-01. The collector runs on the server; the report is built locally. Read the script first withhc-proof collect --script | less. - On the server itself:
hc-proof collect | hc-proof report --input - --out report.html. - From a saved inventory: save the script with
hc-proof collect --script > collect.sh, collect withssh web-01 sh -s < collect.sh > web-01.txt, then runhc-proof report --input web-01.txt --out web-01.html --offline. With the index cached, the report for web-01 took 0.5 seconds. - For scripts: add
--json web-01.jsonto get every finding as JSON next to the HTML. - In a browser: the same engine runs at hostcrust.com/proof. You paste or upload the collector output, and the page says the report is built in memory and the inventory isn't stored. Use the command-line tool if the inventory must stay on your machine.
By default the report uses a prebuilt index (about 41 MB), rebuilt daily from OSV.dev, checked against its published SHA-256 and cached for 24 hours. --offline uses only the cache, --index FILE uses a file you fetched or built yourself, and --osv builds the index from each distribution's OSV bulk file instead (816 MB for Ubuntu). The SHA-256 check catches a broken download; it isn't a signature, so build the index yourself with hc-proof index build if you need that guarantee.
What does the collector read on the server?
The collector is a 49-line POSIX sh script. It needs no root and no network, writes nothing, and prints only:
/etc/os-release- the running kernel (
uname -r,uname -v,uname -m, and/proc/version_signatureon Ubuntu), plus the file names of the kernel images in/boot - the reboot-required flag and the packages that asked for a reboot, where present
- the package list, from
dpkg-query(status, name, version, architecture, source package) orrpm -qa(name, epoch, version, release, architecture, source RPM, vendor)
It collects no hostname, IP address, user name or file contents other than /etc/os-release. The report's header shows the SHA-256 of the input as a receipt. To check the “no root” claim, we ran it as root and as the unprivileged nobody user on both servers. The hashes match:
$ hc-proof collect --script | ssh web-01 sh -s | shasum -a 256
45f6762e0f243efbe9925dcb2b2c8a656cc221c6c16d9ec62404bb01bdb55cfd -
$ hc-proof collect --script | ssh web-01 'runuser -u nobody -- sh -s' | shasum -a 256
45f6762e0f243efbe9925dcb2b2c8a656cc221c6c16d9ec62404bb01bdb55cfd -
$ hc-proof collect --script | ssh web-02 sh -s | shasum -a 256
b55d34b5e928761fd04ba6d9058ab5a781b1db1c61241a2c555216bb411a7a12 -
$ hc-proof collect --script | ssh web-02 'runuser -u nobody -- sh -s' | shasum -a 256
b55d34b5e928761fd04ba6d9058ab5a781b1db1c61241a2c555216bb411a7a12 -
Packages from third-party repositories, such as deb.sury.org PHP, nginx.org, NodeSource, PGDG, PPAs or Debian backports, are listed as “not assessed”, because the distribution's data doesn't describe their builds.
Which Linux distributions does Patch Proof support?
| Distribution | Versions | Data it reads (through OSV.dev) | Notes |
|---|---|---|---|
| Ubuntu | 20.04, 22.04, 24.04 LTS | Ubuntu's per-CVE records, plus Ubuntu Pro (ESM) | Running kernel from /proc/version_signature; Pro-only fixes are labelled, even on hosts without Pro |
| Debian | 11, 12 | Debian security tracker records, then DSA/DLA advisories | Debian 11 data is incomplete since it left Debian LTS in August 2026; the report says so |
| AlmaLinux | 8, 9 | AlmaLinux security advisories (ALSA) | Released fixes only, so no “No distro fix yet” class |
| Rocky Linux | 8, 9 | Rocky Linux security advisories | Covers fewer packages and CVEs than the AlmaLinux data for the same RHEL release |
Debian 13, Ubuntu 26.04, AlmaLinux 10 and Rocky Linux 10 are accepted but untested in 1.0.0; if the prebuilt index doesn't cover a release, report says so and falls back to --osv. Other distributions aren't supported. Matching uses the source package on Debian and Ubuntu and the binary package on AlmaLinux and Rocky Linux, because that's how each distribution publishes its data. Version comparison is a pure-Python implementation of dpkg's and rpm's ordering rules, tested against shared vectors that include epochs, ~, ^, +deb12u1 and .el9_4.
How does Patch Proof compare with Trivy, Grype, OSV-Scanner, OpenSCAP, Vuls and Lynis?
They're good at different jobs, and Patch Proof is the narrowest of them. It answers one question about one Linux server's distro packages: which CVEs are really open right now, counting backports and the running kernel. Trivy and Grype are broad scanners for container images, filesystems and software bills of materials, and both use distribution security data for OS packages; in Trivy's documentation's words, “Trivy trusts the OS vendor's advisory”. OSV-Scanner reads the same OSV.dev data and is built for language dependencies and container images. Vuls is the closest match in purpose, an agentless SSH scanner with running-kernel handling on Debian and Ubuntu, plus far wider OS and exploit-data coverage. OpenSCAP and Lynis are compliance and hardening auditors that also touch vulnerabilities. Trivy, Grype and OSV-Scanner were run side by side with Patch Proof on 14 Nov 2026 (UTC) (see the study); the other cells come from each project's own documentation, checked in November 2026.
| Tool (version) | Built for | How it runs on a server | Distro backports | Running kernel | Where it's stronger | License |
|---|---|---|---|---|---|---|
| Patch Proof 1.0.0 | Which CVEs in distro packages are open on one Linux server | Read-only sh script over SSH; no install, no root | Uses distro data; labels each backport with the build that fixed it | Yes, including “patched on disk, not running”; says nothing about older kernels that aren't running | Narrow by design: distro packages on four distributions only | Apache-2.0 |
| Trivy 0.76.2 | Images, filesystems, repos, VM images and Kubernetes: CVEs, IaC, secrets, licenses, SBOMs | The trivy binary on the host, for example trivy rootfs --pkg-types os --scanners vuln / | Uses the OS vendor's advisories for OS packages | No kernel CVEs on Debian 12 in our test; on AlmaLinux it judged an old installed kernel package. Filtering kernels that aren't running is an open request (#3764) | Containers, Kubernetes, IaC, secrets, language packages, SBOMs | Apache-2.0 |
| Grype 0.121.3 | Container images, filesystems and SBOMs | The grype binary, pointed at a directory or an SBOM | Matches OS packages against distribution security feeds | Not described in its ecosystem docs. In our test its whole-server scan also matched /boot kernel images through NVD (6,112 CVEs on Debian) | SBOM workflows with Syft, language packages, EPSS and KEV risk data | Apache-2.0 |
| OSV-Scanner 2.7.1 | Dependencies in source projects and container images | Documented: lockfiles and directories, and OS packages inside container images (Alpine, Debian and Ubuntu). In our test its directory scan read the dpkg database on Debian but found no packages on AlmaLinux | Uses the same OSV.dev distro data | Not documented for hosts; no kernel CVEs on Debian 12 in our test | Many language ecosystems and lockfile formats, layer-aware image scans, offline database | Apache-2.0 |
| OpenSCAP 1.4.6 | SCAP compliance (XCCDF profiles such as CIS, STIG, PCI DSS) and OVAL checks | oscap on the target; oscap-ssh needs OpenSCAP installed on the remote machine | Uses the distro's own OVAL definitions | Depends on the OVAL content | Configuration compliance profiles and HTML reports | LGPL-2.1 |
| Vuls 0.42.0 | Agentless vulnerability scanning for Linux, FreeBSD, Windows and macOS | One central machine connects over SSH; fast scan needs no root on most distributions; offline mode | Uses Debian, Ubuntu and Red Hat trackers, plus NVD, JVN and OVAL | Collects the running kernel's source package on Debian and Ubuntu (PR #1935) | More operating systems, language libraries, exploit and PoC data, processes that need a restart | GPL-3.0 |
| Lynis 3.1.7 | Security auditing, hardening and compliance testing on Unix-like systems | A shell script run on the host itself; installation optional | Not a per-CVE check: test PKGS-7392 flags packages with a security update available | Not a CVE check | Broad hardening audit across Linux, the BSDs, macOS, AIX and Solaris | GPL-3.0 |
If you already run Trivy or Grype on images, keep them; Patch Proof is for the server underneath. If a scanner you must use for compliance flags CVEs your distro fixed, the “Fixed by backport” rows are your evidence. To check that patches keep arriving, pair it with unattended-upgrades-check. For the manual commands, see checking for security updates on Ubuntu and Debian.
What are Patch Proof's limits?
- Distribution packages only. No containers, images, language packages (pip, npm, gems), configuration or network exposure. Use Trivy, Grype or OSV-Scanner for those.
- The data decides. A result is only as current as the snapshot, and the distribution's data decides what counts as affected and fixed. The report states what that data says about the installed versions; it isn't a guarantee that a server is safe.
- It says nothing about kernels that aren't running. In our 14 Nov 2026 test, Debian's older installed kernel (6.1.128-1) was still affected by 2,444 CVEs that Debian fixed only in later builds, such as the running one, including “Copy Fail” (CVE-2026-31431), and AlmaLinux had an old kernel package (Trivy found 795 CVEs on it) and a 2022 rescue kernel affected by at least 680. All of them can be booted from GRUB. Trivy and Grype judge installed kernel packages and reported them; Patch Proof judges only the running kernel.
- No unfixed CVEs on AlmaLinux and Rocky Linux. Their data lists released fixes only. On our AlmaLinux 9.8 server Patch Proof reported 0 unfixed CVEs, while Grype, using Red Hat's data, listed 7,198 that Red Hat marks as affected with no fix, or won't fix (6,686 on the running kernel, 512 in user space).
- Data freshness. In our test the prebuilt index was built at 14:10 UTC (Debian data from 14:03 UTC, AlmaLinux data from 16:03 UTC the day before). OSV-Scanner, querying OSV.dev live at 18:44 UTC, found 4 Debian CVEs published or updated after that. About five hours after the snapshot, Debian's tracker no longer listed 12 of the CVEs Patch Proof reported (10 had no package entry, 2 were marked not affected).
- Known data gaps (November 2026). Debian 11 results are incomplete. Ubuntu's “deferred”, “needs evaluation” and “ignored”, and Debian's
<ignored>and<postponed>, all show as “No distro fix yet”. Ubuntu FIPS, Realtime and BlueField kernels, and live kernel patching (Livepatch, kpatch), aren't covered. - RPM minor-release streams. When the same CVE is fixed separately in several streams (for example
el9_7andel9_8kernels), it compares against the lowest fixed version, so an early build of the newer stream could be shown as fixed. - Pasted package lists are less accurate. A pasted
dpkg -lhas no source package names and a pastedrpm -qausually has no epochs, and neither says which kernel is running. Use the collector when you can. - One server per report. To cover a fleet, run it in a loop or from your scheduler.
If a report shows kernel fixes waiting, our guides on updating the kernel on Ubuntu and on rebooting and checking the server came back cover the next step. For releases near the end of support, see Linux end-of-life dates.
Changelog
1.0.0, November 14, 2026. First public release.
hc-proof report: which CVEs are really open on a Debian, Ubuntu, AlmaLinux or Rocky Linux server, counting backported fixes, Ubuntu Pro (ESM) fixes and the running kernel. One self-contained HTML report, plus JSON.- A read-only collector (
collect.sh, POSIX sh): no install and no root on the server. - A prebuilt, SHA-256-checked index (about 41 MB, rebuilt daily from OSV.dev) instead of the 816 MB Ubuntu OSV file;
--osv,--offlineand--indexremain. hc-proof sanitisereduces collector output to package and OS data, for sharing.
Results were spot-checked by hand against ubuntu.com/security and the Debian security tracker before release; the method and every checked item are in VERIFICATION.md.
How we checked this: Patch Proof 1.0.0, installed from the v1.0.0 tag into a Python 3.14 virtual environment on macOS, run over SSH on 14 Nov 2026 (UTC) against two fully patched test servers: Debian 12.15 (running kernel 6.1.0-54-amd64, package 6.1.190-1) and AlmaLinux 9.8 (running kernel 5.14.0-687.66.2.el9_8), with the prebuilt index built 2026-11-14T14:10:05Z. We also ran the collector as the unprivileged nobody user on both servers and compared SHA-256 hashes with the root run. We ran Trivy 0.76.2, Grype 0.121.3 and OSV-Scanner 2.7.1 on the same two servers in a separate study the same day (18:41 to 19:20 UTC). The OpenSCAP, Vuls and Lynis rows of the comparison table come from their documentation. Anything we didn't run ourselves comes from the official documentation listed under Sources.
Check which CVEs are really open on your server
HostCrust Patch Proof compares every installed package with your distribution's own security data. It counts fixes your distro backported and checks the kernel that is actually running, not just the one on disk. Read-only, no root, free.
Run this check on every server, every day
HostCrust connects your servers with a small outbound-only agent and runs Patch Proof on each one daily. Free for up to 3 servers.
FAQ
Is Patch Proof free?
Yes. It's open source under the Apache-2.0 license, with no account, no sign-up and no usage limit. The command-line tool runs on your own machine. Its only network call is downloading the public vulnerability index from GitHub, and --offline skips even that once the index is cached.
Does Patch Proof need root or an agent on the server?
No. The collector is a 49-line POSIX shell script that you pipe over SSH, and it runs as any user. On both of our test servers, its output as the unprivileged nobody user had the same SHA-256 hash as its output as root. It writes nothing and installs nothing on the server.
Why does my scanner say my server is vulnerable to a CVE that Ubuntu or Debian already fixed?
Usually for one of four reasons. The scanner compares upstream version numbers while the distro fixed the bug in its own build of an older version (a backport); network scanners and NVD-range matching do this. It matched a kernel image file or a package file that isn't an installed package: in our test, Grype's whole-server scan matched /boot kernel images and old .deb files in the apt cache. It reports a kernel that is installed but not running (a real finding, since that kernel can still be booted, but it isn't what the server runs now). Or the fix exists only in Ubuntu Pro (ESM). Trivy, Grype and OSV-Scanner did not make the first mistake on installed packages in our test. Patch Proof labels backports “Fixed by backport” and Pro-only fixes “Fix available with Ubuntu Pro (ESM)”, judges the kernel that's running, and reports a fixed kernel that hasn't been booted as “Patched on disk, not running”. It says nothing about older kernels left installed.
Can Nmap detect vulnerabilities?
Partly. Nmap's vulners script takes the software version that service detection (-sV) finds on each open port and looks up known CVEs for that product and version on vulners.com. That's a version match from outside the server, and the script's documentation says nothing about distro backports, so on Debian, Ubuntu and RHEL-family servers it may list CVEs the distro already fixed in its own build. It can't see software that doesn't listen on a port. Use it to see what the network sees, and a package-level check to see what's installed and fixed.
What software is used for vulnerability scanning on Linux servers?
Three kinds of tool, and most teams use more than one. Network scanners probe open ports from outside. Package-level CVE scanners, such as Trivy, Grype, Vuls and Patch Proof, compare installed packages with vulnerability data. Compliance and hardening auditors, such as OpenSCAP and Lynis, check configuration against a policy. The comparison table on this page shows what each open-source option is best at.
Does Patch Proof work offline or in an air-gapped network?
Yes. On a machine with internet access, run hc-proof index fetch --out index.sqlite, copy the file across, and pass --index index.sqlite to hc-proof report. That uses no network at all. On a machine that has run it before, --offline uses the cached index.
How often is the vulnerability data updated?
A GitHub Actions workflow in the repository rebuilds the prebuilt index from OSV.dev every day. The tool re-checks for a newer index after 24 hours (change it with --max-age), and every report states the snapshot time of the data it used, so an old report can't pass for a new one.
Sources
- HostCrust Patch Proof v1.0.0: README (install, classes, collector, limits)
- HostCrust Patch Proof v1.0.0: VERIFICATION.md (results hand-checked against ubuntu.com and the Debian tracker)
- HostCrust Patch Proof: v1.0.0 and index-latest releases
- Debian security tracker: CVE-2024-6387 (bookworm fixed in 1:9.2p1-2+deb12u3, DSA-5724-1)
- OpenSSH 9.8 release notes: sshd race condition, Portable OpenSSH 8.5p1 to 9.7p1 affected
- NVD: CVE-2024-6387 (affected software listed by upstream OpenSSH version)
- curl: CVE-2023-38545 (libcurl 7.69.0 to 8.3.0 affected, fixed in 8.4.0)
- Debian security tracker: CVE-2023-38545 (bookworm fixed in 7.88.1-10+deb12u4, DSA-5523-1)
- AlmaLinux ALSA-2023:6745: curl 7.76.1-26.el9_3.2 fixes CVE-2023-38545
- AlmaLinux ALSA-2024:2394: kernel 5.14.0-427.13.1.el9_4 fixes CVE-2024-1086
- Trivy v0.76.2 docs: vulnerability scanning (OS vendor advisories, backports, detection priority)
- Trivy v0.76.2 docs: rootfs target (scanning a host)
- Trivy issue #3764: filter vulnerabilities found in old kernel versions (open)
- Grype README (images, filesystems, SBOMs; EPSS and KEV)
- Anchore docs: how Grype matches OS packages against distribution feeds
- OSV-Scanner README (source and container image scanning, offline database)
- OSV-Scanner docs: supported artifacts (OS packages in container images)
- OpenSCAP 1.4.1 user manual (local scans, oscap-ssh, compliance profiles)
- Ubuntu OVAL: scanning an Ubuntu system with oscap
- Vuls docs: main features (agentless over SSH, data sources, offline mode)
- Vuls docs: fast scan (no root on most distributions)
- Vuls PR #1935: collect the running kernel's source package on Debian and Ubuntu (merged)
- CISOfy: Lynis, security auditing and hardening tool
- CISOfy: Lynis control PKGS-7392 (vulnerable packages with an update available)
- Nmap NSE: vulners script
- OSV.dev: open source vulnerability database