DEV Community

cl0q
cl0q

Posted on

Java4Shell, NGINX2Root, and a PHP security release: what 58 million crawled sites actually run

Three pre-auth remote code execution scares landed in three days. We run a web search engine, so we asked two questions in order: are we affected, and what does the internet actually run? Here are the answers, with our own crawl data as the evidence.

Java4Shell: a Base64 bug in OpenJDK

CVE-2026-114938 is a memory corruption bug in OpenJDK's Base64 decoder. In default configurations it gives pre-auth RCE against Keycloak, Jira, Confluence, Jenkins, and Artifactory. Oracle scores it 8.1. Calif scores it 9.2. Fixes shipped the same day, and no in the wild exploitation has been reported.

Are we affected? No. Nothing in our serving path runs Java. The site is Go plus Apache, the crawler and trafficd are Go, Milo is PHP, and the database is Postgres. Our database machine does have OpenJDK 17.0.21.0.3 installed, but nothing runs it: no Java process, no service. The box is also an old i5-2500 with no AVX-512 VBMI, so the buggy code path cannot execute there. It is worth patching or removing when convenient, not urgent. The CVE is not in CISA's Known Exploited Vulnerabilities catalog through the 2026.11.20 catalog, which lists 1,790 entries.

NGINX2Root: five nginx flaws

Disclosed November 18, NGINX2Root is five nginx flaws with unauthenticated RCE. The two big ones are rated 9.2. They are fixed in nginx 1.30.7 and 1.31.9. No exploitation in the wild has been reported, and none of the five is in CISA KEV.

Are we affected? No. Our serving path is Go plus Apache, with no nginx in it. The only nginx near us is the one bundled in the dispatcharr container on our database machine: version 1.24.0 on port 9191, plain HTTP, and its full config shows no HTTP/2, no QUIC, no resolver, and no mirror, so the flaws do not apply to it. Worth updating dispatcharr when convenient, not urgent. Two honest gaps: what triggers the fifth CVE was unverified, and whether port 9191 is forwarded through the router was unknown.

The PHP security release

November 19 brought PHP 8.6.0 plus security releases 8.5.13, 8.4.28, 8.3.37, and 8.2.36. The headline fix is CVE-2026-115560, a session serializer RCE, plus CVE-2026-129474, 129182, 129538, 129496, 129501, 129291, 129289, 129292, 128090, 15353, and 129506.

Are we affected? The site is not: no PHP in the serving path. Milo, our family dashboard, runs PHP 8.2.35, one patch behind 8.2.36. The worst triggers do not apply to it (it serves on PHP's built-in server, not FPM or CGI, and there is no serialize-handler switching in its code), but the core RCEs cannot be fully ruled out from outside. The fix is a rebuild of the Milo image plus a restart, and that rebuild is held by the owner for now.

What the crawl data says about exposure

Here is where the story gets interesting. Our index covers about 58.2 million domains and 38.4 million pages. The latest stats generation, November 21 at 21:35 MST, breaks down the servers those sites advertise:

  • Cloudflare: about 11.3 million
  • nginx: about 6.8 million
  • Apache: about 4.2 million

That is why NGINX2Root matters even if it does not touch us. nginx is the number two server in the index. When a 9.2 RCE lands in the second most common server on the internet, the blast radius is measured in millions of sites, and most of them will not patch this week.

Java is a different shape of problem. Java servers are identifiable from Server headers: Jetty, Tomcat, Coyote. But Java versions almost never leak in those headers, so there is no Java equivalent of the PHP version list we sell, which is built entirely on versions exposed in Server headers. What the crawl data can honestly show is which sites run Java servers at all, not which ones are vulnerable. That is the product limit, and it is also the security lesson: a version string in a Server header is free reconnaissance for an attacker, and the sites that hide it are the ones that force an attacker to guess.

The takeaway

Three disclosures, three verdicts of not affected, and one pattern worth stealing. Check your own Server header before someone else does. If it advertises a version, you have handed an attacker a shortlist. Patch the thing, hide the version, and check the exposure again next week.

Top comments (0)