Home/ Blog/ Security news/ Article
Blog · Security news

nginx CVE-2026-123007: Critical HTTP/2 RCE, Patch to 1.30.7 Now

nginx CVE-2026-123007 lets an unauthenticated client corrupt memory over HTTP/2 (1.9.5 to 1.31.8). Public PoC, distro packages lag. Patch or disable HTTP/2.

Security brief card reading nginx HTTP/2 flaw can let attackers run code

Upgrade nginx to 1.30.7 (stable) or 1.31.9 (mainline) and fully restart it: every open-source release from 1.9.5 through 1.31.8 with HTTP/2 switched on carries CVE-2026-123007, a memory-corruption bug that can let an unauthenticated stranger run code inside an nginx worker process. F5 scores it 9.2 under CVSS v4.0. Rapid7 has put a working proof of concept on GitHub. The catch for most Linux hosting fleets is that the upstream fix is a day old and the distribution packages have not caught up.

The lead bug arrived alongside nine other nginx CVEs in one coordinated release on November 18. Researchers are calling the set NGINX2Root. Qualys groups five of them as the code-execution set, and only two of those five live in HTTP/2, which is why "turn off HTTP/2" is a partial answer and not a fix.

Which nginx versions does CVE-2026-123007 affect?

nginx Open Source 1.9.5 through 1.31.8 is affected whenever HTTP/2 is enabled; the fixed releases are 1.30.7 on the stable branch and 1.31.9 on mainline, both published on nginx.org on November 18. NGINX Plus customers should move to 37.0.8.1 or 37.3.0.1. Builds without the HTTP/2 module are not exposed to this CVE.

The flaw is an integer underflow (CWE-191) in HTTP/2 flow control, the bookkeeping that tracks how much response data a client has said it can accept. According to Rapid7, a client that cancels queued data can push that counter into an invalid state, so nginx later copies a response with a bad length and corrupts memory. The result ranges from a crashed worker to code execution with the worker's privileges. The bug dates to 2015, which is why the affected range spans more than a decade of releases.

Its sibling, CVE-2026-123633, is a use-after-free in HTTP/2 connection handling that Wiz says fires when nginx proxies to a backend and runs low on connection slots. That second condition matters for hosting: an nginx front end proxying to Apache or PHP-FPM is exactly that shape.

CVEFlawExposed whenBlocked by turning HTTP/2 off?
CVE-2026-123007Underflow in HTTP/2 window accountingHTTP/2 enabledYes
CVE-2026-123633Freed connection reused while proxyingHTTP/2 plus proxying to a backendYes
CVE-2026-123634Resolver use-after-freeVariable upstream names with SSI or mirror subrequestsNo
CVE-2026-121810HTTP/3 use-after-freeHTTP/3 enabledNo
CVE-2026-113425Mail proxy use-after-freeMail proxy in useNo
Source: nginx.org security advisories, Wiz, Qualys TRU (November 18, 2026).

Three of the remaining code-execution candidates need features many servers never touch: a resolver use-after-free (CVE-2026-123634) tied to variable upstream names plus SSI or mirror subrequests, an HTTP/3 use-after-free (CVE-2026-121810), and a mail-proxy use-after-free (CVE-2026-113425). Disabling HTTP/2 does nothing for any of them. The upgrade covers all ten.

How real is the exploit risk today?

Real, but gated by bandwidth and reachability. Rapid7's published proof of concept reached code execution against a Docker image of nginx 1.31.8 with HTTP/2 and TLS on, and the firm is explicit that the technique is tuned to that lab target. It needs roughly 128 GB of HTTP/2 traffic over a single connection: about two minutes at 10 Gbit/s and about seventeen at 1 Gbit/s. Below roughly 300 Mbit/s, nginx's default one-hour connection lifetime cuts the attempt off. The demonstrated path also depends on a specific glibc memory-copy routine selected on x86-64 CPUs with AVX-512 support.

Those limits slow a spray-and-pray campaign; they do not protect a server that a capable attacker picks deliberately. Qualys reports that Xint, one of the independent finders, has its own exploit working against public cloud addresses and is holding back the final steps. Wiz says it is not releasing its exploits. As of November 19, nobody has reported exploitation in the wild, and CISA has not added the bug to its Known Exploited Vulnerabilities (KEV) list.

nginx HTTP/2 flaws: first report to public PoC2026-10-21: Wiz reports UAF. 2026-11-11: Rapid7 contacts F5. 2026-11-16: Wiz reports underflow. 2026-11-18: 1.30.7 and 1.31.9 ship. 2026-11-18: Rapid7 PoC public.nginx HTTP/2 flaws: first report to public PoC2026-10-21Wiz reports UAF2026-11-11Rapid7 contactsF52026-11-16Wiz reportsunderflow2026-11-181.30.7 and1.31.9 ship2026-11-18Rapid7 PoCpublic
Source: Wiz and Rapid7 disclosure timelines, nginx.org.

Our read on the bandwidth requirement: it turns this into an origin-exposure question. Rapid7 expects its technique to fail when the attacker's HTTP/2 stream passes through a proxy first, and Wiz says a content delivery network (CDN) or web application firewall (WAF) that terminates HTTP/2 may partly mitigate it. A site behind a CDN whose origin IP still answers HTTP/2 directly gets none of that cover. Finding those leaked origins is the first triage pass, before you touch a version number.

The "root" in NGINX2Root does not come from nginx. Qualys says none of the five flaws grants root on its own; Xint reached root by chaining a separate Linux kernel flaw in the SCTP networking protocol that, per Qualys, has no CVE and no patch yet. If your web servers never use SCTP, which is the case for almost every hosting box, stopping that kernel module from loading removes a step an attacker would need and costs you nothing.

Why the distro package you run is probably still vulnerable

Upstream shipped a fix; your package manager may not have one. As of November 19, the Debian security tracker lists nginx as vulnerable in every release, from bookworm through sid, with no fixed package and no Debian Security Advisory. Qualys reported Ubuntu in the same state on disclosure day. Most hosting fleets install nginx from the distribution, so for them "patch now" today means switching to the nginx.org package repository or applying the HTTP/2 workaround until the distro catches up.

Watch for the reverse trap too. When the distro fix does land, it will likely be a backport under an older version string, so a server that reports 1.22.1 may be fixed and one reporting 1.26.3 may not. Trust the tracker entry for CVE-2026-123007, not nginx -v.

Control panels add a third patch path. If your panel runs nginx as the front-end proxy for Apache (Plesk does this by default, and cPanel offers it), the nginx build comes from the panel vendor's update channel, and that vendor's release notes are what you check. Container images follow their own path: docker pull only helps after you redeploy the running containers.

This is the second major HTTP/2 code-execution bug in a hosting web server in a week, after Apache's mod_http2 race (CVE-2026-119618). A typical cPanel or Plesk box running nginx in front of Apache needed both fixes. It also follows the two 9.2-scored nginx bugs from June that, on a default server, gave a crash rather than a shell, and the July map regex heap overflow. The difference this time: Wiz says the lead bug is reachable in default configurations once HTTP/2 is on, with no backend or third-party module required, so the "is my config even reachable" filter removes far fewer servers.

Check a server: build, config, origin and crash log

Four checks answer whether a host is in scope, and one log search shows whether something has been pushing on it. Memory corruption in a worker usually shows up first as workers dying on a signal, which nginx's master process records at the alert level in the error log.

nginx: exposure checks and worker-crash signature (adjust log path for your distro or panel)
nginx -V 2>&1 | grep -oE 'nginx/[0-9.]+|with-http_v2_module'
nginx -T 2>/dev/null | grep -nE '^\s*(listen\s.*\bhttp2\b|http2\s+on)'
openssl s_client -connect ORIGIN_IP:443 -servername www.example.com -alpn h2 </dev/null 2>/dev/null | grep 'ALPN protocol'
ps -o pid,lstart,cmd -C nginx
grep -E 'worker process [0-9]+ exited on signal (11|6|7)' /var/log/nginx/error.log

A host is in scope if the build is unpatched, the HTTP/2 module is compiled in, and either listen ... http2 or http2 on appears in the effective config. Run the ALPN check against the origin IP, not the CDN hostname; an answer of h2 means HTTP/2 is reachable directly. The ps start time tells you whether the running master predates your upgrade. Workers die on signal 11 for plenty of dull reasons; on an exposed, unpatched server, a cluster of them is still the event to chase first. Wiz adds that the underflow needs a flood of HTTP/2 WINDOW_UPDATE frames on one long connection, so a single client holding a connection open for minutes while moving tens of gigabytes is worth an alert on its own.

Restart into 1.30.7 or 1.31.9, or switch HTTP/2 off

Install 1.30.7 or 1.31.9 (or a distro, panel or NGINX Plus build that carries the fix) and do a full restart. A reload only re-reads configuration; the old binary keeps serving until the master process is replaced. Confirm with nginx -v and the ps start time above.

If you cannot upgrade today, stop serving HTTP/2: on nginx 1.25.1 or later set http2 off; in each affected server block, and remove the http2 parameter from every listen line on any version. Reload, and check that old workers have exited. Clients fall back to HTTP/1.1, which costs page-load speed and can break gRPC, so test anything that depends on HTTP/2 first. That shuts the door on the two HTTP/2 bugs and nothing else; the resolver, HTTP/3 and mail-proxy bugs still need the upgrade. Turn HTTP/2 back on once the fixed build is running.

An upgrade says nothing about the weeks before it. On servers that sat exposed, keep the error logs from the past few weeks before rotation removes them. Look for worker crashes that do not line up with a deploy or a traffic spike, and for any process spawned by the nginx user that is not nginx. If the slow part is knowing which of your servers run which nginx build, our vulnerability detection page explains how we approach that inventory.

Topics

Frequently asked questions

Which nginx versions are affected by CVE-2026-123007?

nginx Open Source 1.9.5 through 1.31.8 is affected when HTTP/2 is enabled. The fixes are 1.30.7 on the stable branch and 1.31.9 on mainline, released November 18, 2026. NGINX Plus is fixed in 37.0.8.1 and 37.3.0.1. Builds without the HTTP/2 module are not exposed.

Is there a public exploit for the nginx HTTP/2 flaw?

Yes. Rapid7 published a proof of concept on GitHub that reached code execution on a Docker image of nginx 1.31.8. It needs roughly 128 GB of HTTP/2 traffic on one connection and a specific glibc copy routine on AVX-512 CPUs. No exploitation in the wild had been reported as of November 19, 2026.

Does disabling HTTP/2 fix all the NGINX2Root vulnerabilities?

No. Turning HTTP/2 off blocks CVE-2026-123007 and CVE-2026-123633 only. The resolver flaw (CVE-2026-123634), the HTTP/3 flaw (CVE-2026-121810) and the mail-proxy flaw (CVE-2026-113425) need the upgrade to 1.30.7 or 1.31.9. Use the HTTP/2 switch-off only as a stopgap.

Is the Debian or Ubuntu nginx package patched for CVE-2026-123007?

As of November 19, 2026, not yet. The Debian security tracker listed nginx as vulnerable in every release with no fixed package, and Qualys reported Ubuntu in the same state on disclosure day. Check your distro's tracker; until a fix lands, use the nginx.org packages or disable HTTP/2.

Does a CDN in front of nginx protect against CVE-2026-123007?

Partly. Rapid7 says a proxied HTTP/2 connection will likely not satisfy its technique, and Wiz says a CDN or WAF that terminates HTTP/2 may reduce the risk. That cover is gone if the origin server still answers HTTP/2 directly on its own IP address, so check the origin too.

Does NGINX2Root give attackers root access?

Not by itself. Qualys says none of the five nginx flaws grants root directly. Researchers at Xint reached root by chaining a separate, unpatched Linux kernel SCTP flaw after gaining code execution as the nginx worker, including from inside a Docker container.

See what Suriq finds on your own servers.

Agentless checks for uptime and your cloud, a lightweight agent for detection on each server, and one console that ranks and explains what it finds.