# Ordering security patches by entry point, privilege and evidence

> Node.js, PHP2Shell and the kernel needed patches in October 2026. I weigh entry points, privilege and evidence to set the order, and I look past the CVSS Base score.

Source: https://sheng.page/en/posts/patch-priority-by-entry-point/ · Published: 2026-11-15 · Author: Sheng

In the second half of October 2026, I read the Node.js and PHP security advisories alongside two Linux kernel vulnerability reports. A patch list sorted by CVSS Base score often misses the risk in my environment. The Base score describes the bug. My priorities depend on what it can do where I run it.

I use three questions to judge urgency, then weigh the impact of an incident and the risk of the fix before I schedule the patches. This post runs October's batch and my blog's build pipeline through them, unfinished work included. I drafted it on 23 October and have since folded in the changes up to 8 November, such as the KEV listings and the Unit 42 update.

## 1. Three questions before the score

```text
1. How does untrusted input reach the vulnerable code?
2. What can the environment running that code do?
3. How urgent does the current evidence say it is?
```

"Input" includes incoming requests, fetched responses, external files processed by CI and package installation scripts. Privilege covers the ability to change release artefacts or hold deployment credentials, as well as access to other tenants. Evidence includes official fixes, a working PoC, honeypot observations and confirmed compromises.

I have no quarrel with CVSS itself. FIRST's own v4.0 implementation guide states that "CVSS Base scores alone should not be used for customer CVSS scoring and vulnerability program management." Threat metrics lower the score when there is no known exploit or only a PoC. Environmental metrics need to reflect your systems. Many teams stop at the Base score. As Wikipedia puts it, CVSS "is not intended as a method for patch prioritization, but is often used as one regardless".

## 2. October's batch

The Node.js security release of 21 October fixes 10 CVEs, 4 of them High, including an out-of-bounds write in the Brotli decoder that can lead to remote code execution (CVE-2026-111497) and two High-rated bugs that can bypass TLS hostname verification. The fixed versions are v22.23.4, v24.22.1 and v26.12.1. The project gives no CVSS figure and says nothing about exploitation in the wild. When I drafted this on 23 October, that left little evidence for the third question. On 27 October, CISA added the same CVE to its KEV catalogue, listed under Google's Brotli library, and NVD has now scored it 9.8 under CVSS 3.1. The KEV entry does not say which product was exploited. A KEV listing is evidence of exploitation in the wild, and under the third question it makes patching urgent.

On the first question, Node does not decompress every incoming request. Its built-in fetch decompresses responses according to their Content-Encoding. I checked this on Node v22.23.4 and v26.13.0: when a local server answers with `Content-Encoding: br`, `res.text()` returns the decoded body. Red Hat's CVE page also names `fetch()` among the APIs that decode Brotli. Vercel's bulletin states: "Your application is exposed when its code, or a library it uses, decompresses Brotli data an attacker controls", and responses your code fetches count too. It also notes that its WAF "does not cover responses your functions fetch from other servers" and "reduces exposure but does not replace patching".

PHP2Shell (CVE-2026-109618 and CVE-2026-109619) follows a different path. PHP rates both at 9.2 under CVSS v4.0, with fixes in 8.5.12, 8.4.27, 8.3.36 and 8.2.35. One bug leaves half-initialised data when allocation failure aborts a request. The other lets the same worker's next request use leftover state pointing at freed memory. Exploit conditions depend on the SAPI. The official stale-state advisory confirms RCE paths for mod_php and PHP-FPM. The FPM path uses session upload-progress, verified on a typical dynamic pool, and it still requires suitable application behaviour and both requests reaching the same worker.

The evidence on the third question is stronger here. AWS's MadPot honeypots logged exploitation attempts on 17 October, four days before disclosure. AWS writes that "PHP2Shell is now seeing broad exploitation attempts in the wild", but of the pre-disclosure attempts it cautions that "it’s unclear if any of these attempts would have succeeded in real-world systems". Rapid7 has not itself observed exploitation in the wild, and it considers the public PoCs to be of limited reliability. CISA added both CVEs to the KEV catalogue on 26 October, with a 29 October deadline for Federal Civilian Executive Branch agencies. The 29 October update from Palo Alto Networks' Unit 42 adds evidence of successful compromise: PHP2Shell is being actively exploited against internet-facing PAN-OS management interfaces. Dream Research reported attackers chaining it with a Linux kernel privilege-escalation bug to get root, and Unit 42 verified the same tooling and the same results. That update does not identify the specific PHP CVE or the kernel vulnerability. The PHP project states: "Updating to a release containing the complete state reset patch is the only reliable mitigation." If you run PHP that faces the internet, upgrade.

For the kernel, Vghost (CVE-2026-74580) lets a local user with access to a vhost device reach root, and PwnEscape (CVE-2026-99999) lets code inside a guest VM read host memory. The CVE.org record scores PwnEscape at 9.3 under CVSS 3.1, the figure The Hacker News reported. Red Hat rates it 8.8 under CVSS 3.1. The difference is Privileges Required. For PwnEscape, the evidence conflicts. The Hacker News reported on 15 October that it was absent from CISA's KEV catalogue, with no sign of exploitation, while Red Hat tracks it as a "KEV active-exploit case". On 8 November, I checked the catalogue (the 6 November release). PwnEscape and Vghost are absent from that release. Their absence does not establish whether exploitation has occurred. Red Hat's statement remains unresolved. For both, the first question is who can already run code on the machine or inside its VMs. Multi-tenant hosts, CI runners, shared development boxes and sandboxes that run other people's code go first. Check heavily marketed vulnerability pages against your distribution's advisory.

ChainDrop, from early August, is my comparison. In the npm supply chain, installation is the entry point. Datadog's report on ChainDrop describes malicious versions of popular packages such as keyv, flat-cache and cacheable-request, carrying a preinstall script that steals GitHub, npm and cloud tokens. If the package manager allows lifecycle scripts, the script runs at installation, before any request arrives.

## 3. My own site

This blog is an assets-only static site, deployed through Cloudflare Workers Builds. Visitor requests never reach Node or PHP code that I manage, so on the visitor path the Brotli bug has no way in.

The build runs Node with deployment permissions, a poor answer to the second question. The repo's `.nvmrc` specifies the fixed version `22.23.4`, but the project notes say Workers Builds has `NODE_VERSION=22` set. Only a build log can establish the version Workers Builds actually used. For now I can confirm the repo's setting but not the running version, and the advisory cannot settle it. On 3 November, Node announced a security release for 11 November or shortly after, with High as the top severity on the 22.x, 24.x and 26.x lines, so the pinned version will soon be out of date.

On 23 October, `npm audit` reported 6 high and 2 moderate findings, all in the build and development toolchain, such as vite's dev server and the sharp that wrangler pulls in. I applied `npm audit fix` that day, without `--force`. The tests and the build passed, and `npm audit` reported nothing. A cheap, verifiable fix need not wait for severity classification and a maintenance window. That clean result was only a snapshot. On 8 November, with the lockfile untouched, it found 6 again (2 high and 4 moderate) in undici, zod, fast-uri and satori, all in the toolchain. For satori it recommends `--force` and a breaking upgrade. On 23 October I also scanned my local lockfiles and a few known persistence locations, and found no ChainDrop versions or residual files within that scope. That, too, is a snapshot from that day.

## 4. Where the ordering goes wrong

A vulnerability at build time can be as urgent as one at runtime. CI that processes external pull request files, and that holds deployment credentials or permission to change release artefacts, combines untrusted input with high privilege. A compromised CI job can alter deployments, and putting the package in devDependencies does not reduce that risk.

"I could not find a path" does not establish that "it cannot be reached". Brotli decompression may be buried in a dependency. If a fix needs only a patch-level bump and a build to confirm it, I apply it. The analysis time I save is worth more than a precise classification.

I pin versions in the lockfile and delay routine dependency updates for a few days, while watching lifecycle scripts such as preinstall. Security fixes are the exception, and they go through the three questions above. These habits cannot eliminate risk. They keep me from installing a release minutes after publication, before others have had time to check it.

## Notes

- The three questions cover input paths, privilege and evidence. FIRST treats the Base score as a starting point.
- The Node Brotli bug requires decompression of attacker-controlled data. Built-in fetch decodes `br`, and Vercel's WAF does not cover fetched responses. The CVE joined CISA KEV on 27 October.
- PHP2Shell's conditions depend on the SAPI. Both CVEs entered CISA KEV on 26 October. On 29 October, Unit 42 confirmed chaining with an unnamed kernel bug to get root. The PHP project calls upgrading the only reliable mitigation.
- For kernel bugs, existing code execution on the host or its VMs matters. Prioritise multi-tenant hosts, CI runners and shared machines. PwnEscape remained absent from the 6 November KEV catalogue.
- My visitor path cannot reach Node. The build has deployment rights. Its Node version still needs checking in the build log. `npm audit` is a snapshot: clean on 23 October, 6 findings on 8 November.

## Further reading

- Node.js, [Wednesday, October 21, 2026 Security Releases](https://nodejs.org/en/blog/vulnerability/october-2026-security-releases): the official release note and primary source for versions and CVEs, with no CVSS figures or exploitation status.
- Node.js, [Wednesday, November 11, 2026 Security Releases](https://nodejs.org/en/blog/vulnerability/november-2026-security-releases): the pre-announcement, published on 3 November, of a further security release on or shortly after 11 November, with High as the top severity on the 22.x, 24.x and 26.x lines.
- Vercel, [Node.js security advisory: Brotli decoder RCE (CVE-2026-111497)](https://vercel.com/kb/bulletin/node-js-security-advisory): the platform's description of the Brotli exposure condition and its WAF coverage, quoted above.
- Red Hat, [CVE-2026-111497](https://access.redhat.com/security/cve/cve-2026-111497): the distribution's view of the Brotli bug, naming Node.js zlib and `fetch()` as decoders and noting which products use system or bundled brotli.
- NVD, [CVE-2026-111497](https://nvd.nist.gov/vuln/detail/cve-2026-111497): the NVD record, the source of the CVSS 3.1 9.8 score quoted above.
- CISA, [Known Exploited Vulnerabilities Catalog](https://www.cisa.gov/known-exploited-vulnerabilities-catalog): the catalogue itself, where the KEV dates and deadlines above were checked.
- Rapid7, [CVE-2026-109618 and CVE-2026-109619: Critical Unauthenticated Remote Code Execution in PHP](https://www.rapid7.com/blog/post/etr-cve-2026-109618-and-cve-2026-109619-critical-unauthenticated-remote-code-execution-in-php/): a security vendor's analysis of the root causes, exploitation status and interim mitigations.
- PHP, [RCE via stale global state (GHSA-6wmf-4rv5-p3rp)](https://github.com/php/php-src/security/advisories/GHSA-6wmf-4rv5-p3rp): the official advisory and the authoritative source on per-SAPI exploit conditions and mitigations.
- PHP, [RCE via OOM corruption (GHSA-w8hw-w9fx-j79w)](https://github.com/php/php-src/security/advisories/GHSA-w8hw-w9fx-j79w): the second official advisory, for CVE-2026-109618, and the source for the allocation-failure root cause.
- AWS, [Amazon Threat Intelligence identifies PHP2Shell exploitation prior to public disclosure in MadPot](https://aws.amazon.com/blogs/security/amazon-threat-intelligence-identifies-php2shell-exploitation-prior-to-public-disclosure-in-madpot/): honeypot data showing attempts before disclosure, with success in real systems unconfirmed for those pre-disclosure attempts.
- Unit 42, [Threat Brief: Critical Vulnerabilities in PHP: CVE-2026-109618, CVE-2026-109619](https://unit42.paloaltonetworks.com/cve-2026-109618-cve-2026-109619/): the 29 October update confirming PHP2Shell attacks on PAN-OS management interfaces, chained with an unnamed kernel privilege-escalation vulnerability to reach root.
- The Hacker News, [AI-Found Linux Kernel Flaw Gives Local Users Root, and Can Break a Guest Out to the Host](https://thehackernews.com/2026/10/ai-found-linux-kernel-flaw-gives-local.html): news coverage of Vghost; check your distribution's advisory for specifics.
- The Hacker News, [Critical PwnEscape Linux Kernel Vulnerability Lets Guest VMs Read Host Memory](https://thehackernews.com/2026/10/critical-pwnescape-linux-kernel.html): news coverage of PwnEscape, quoting the 9.3 score from the CVE.org record.
- Red Hat, [CVE-2026-99999](https://access.redhat.com/security/cve/cve-2026-99999): the distribution's CVE page, with attack prerequisites, Red Hat's own score (8.8, CVSS 3.1), the CVE.org record's CVSS 3.1 score (9.3) and a mitigation; the KEV tracking it mentions has no matching entry in the catalogue.
- Datadog Security Labs, ['ChainDrop' worm compromises hundreds of popular npm packages](https://securitylabs.datadoghq.com/articles/npm-worm-compromises-popular-npm-packages/): incident analysis listing affected packages and the preinstall behaviour.
- FIRST, [CVSS v4.0 Consumer Implementation Guide](https://www.first.org/cvss/v4.0/implementation-guide): the official guidance stating that Base scores alone should not drive vulnerability management.
- Wikipedia, [Common Vulnerability Scoring System](https://en.wikipedia.org/wiki/Common_Vulnerability_Scoring_System): background on what CVSS is for and how EPSS complements it.

> Ideas and technical judgement by Sheng; drafted with Claude · fact update and revision on 8 November by Claude (Opus 5.5), second review by Codex (gpt-6-astra), tone by Grok (grok-4.7), coordination and final edit by Claude Fable; public sources checked on 8 November 2026, with KEV as of the 6 November catalogue release, still current when rechecked early on 9 November. Built-in fetch decoding Brotli was tested on Node v22.23.4 (undici 6.29.1, Linux) and v26.13.0 (undici 8.12.2, macOS 27.2); the 8 November `npm audit` ran on npm 11.20.0.