Last updated: November 20, 2026, 14:04 UTC. We'll update this post as vendors ship fixes.
On November 18, F5 shipped nginx 1.30.7 and 1.31.9, fixing five memory-safety bugs that Xint named NGINX2Root. The same release also fixed other nginx CVEs, so distribution advisories list eight to ten CVEs, not five. Each one lets an unauthenticated attacker run code as the nginx worker in a specific configuration. The one that matters to almost everyone is CVE-2026-123007, an HTTP/2 flow-control bug that has been in nginx since HTTP/2 support landed in 1.9.5 in 2015. If HTTP/2 is on, you're in scope, even if all nginx does is serve static files. Xint also chained the nginx bug with an unpatched Linux SCTP flaw to reach root on the host, including from inside a Docker container.
Patching your own nginx is the easy part. The harder question is where else nginx is running: inside firewalls, gateways, NAS boxes, ingress controllers, and appliances that ship their own builds on their own schedule. On the day this post went up, a firmware analysis from Threatbear showed Ubiquiti's current UniFi Cloud Gateway Ultra firmware still ships a vulnerable nginx with HTTP/2 enabled on four ports. There's no Ubiquiti advisory yet. That's the long tail, and it's where this one will stick around.
At a glance
| CVEs | CVE-2026-123007 (HTTP/2 flow control), CVE-2026-123633 (HTTP/2 use-after-free), CVE-2026-121810 (HTTP/3 use-after-free), CVE-2026-123634 (resolver use-after-free), CVE-2026-113425 (mail proxy use-after-free) |
|---|---|
| Severity | F5 rates all five CVSS 4.0 9.2 / CVSS 3.1 8.1. nginx.org rates CVE-2026-123007 and CVE-2026-113425 “major” and the other three “medium.” |
| Fixed versions | nginx Open Source 1.30.7 (stable) and 1.31.9 (mainline). NGINX Plus 37.0.8.1 and 37.3.0.1. |
| Affected range (CVE-2026-123007) | nginx 1.9.5 through 1.30.6, and 1.31.0 through 1.31.8 |
| Precondition | HTTP/2 enabled on a reachable listener |
| Exploitation | Lab only. Rapid7 published a working PoC for CVE-2026-123007 on November 18. Xint and Wiz have unreleased PoCs for all five. No confirmed exploitation in the wild. |
| CISA KEV | Not listed as of November 19 |
| EPSS | 6.36% (94th percentile) as of November 19 |
| Privilege | Code runs as the nginx worker. Root requires a second bug, such as the SCTP flaw Xint chained. |
| Biggest gap | Products that embed nginx: appliances, ingress controllers, forks. Many have no fix or statement yet. |
Am I affected?
- Yes, if you run nginx Open Source 1.9.5 through 1.30.6, or 1.31.0 through 1.31.8, with HTTP/2 on any listener. That includes static-file servers. CVE-2026-123633 adds exposure if that same server proxies to an upstream.
- Yes, if you run NGINX Plus R36 or earlier. F5 lists no fix for R34 through R36, and older releases are end of life. Move to a fixed 37.x build.
- Yes, if you run the community Kubernetes ingress-nginx controller. The project was archived in March 2026, turns HTTP/2 on by default, and won't get a fix.
- Probably, if you run a product that bundles nginx, such as a gateway, NAS, firewall, or management appliance. You can't patch it yourself. You need the vendor's firmware. See The nginx you didn't install below.
- Check the package, not the banner, if you're on a distribution build. Backports fix the bug without changing what
nginx -vprints. - Lower priority, if HTTP/2 is off everywhere. You still need to check the other four CVEs against your config (HTTP/3, resolver with mirror, mail proxy).
What to do now
- Find every nginx, including the ones you didn't install. Run
nginx -von hosts and container images, thennginx -T 2>/dev/null | grep -nE 'http2|listen'to see where HTTP/2 is on. Add ingress controllers, the official Docker image, and every appliance whose management UI runs on nginx. - Patch to 1.30.7 or 1.31.9, or your distribution's fixed package. Then restart. A config reload doesn't replace the running binary. If
ls -l /proc/$(cat /run/nginx.pid)/exeends in(deleted), the master is still running the old build. - Until you can patch, turn HTTP/2 off. Use
http2 off;on 1.25.1 and later, or removehttp2fromlistendirectives on 1.9.5 through 1.25.0. On 1.25.1 and later, also remove any leftoverhttp2parameter fromlisten. The deprecated parameter can still turn HTTP/2 on. Confirm it took withopenssl s_client -connect host:443 -servername host -alpn h2 </dev/null 2>&1 | grep -E 'ALPN protocol|No ALPN'. Check gRPC and anything else that depends on HTTP/2 before you flip it. - Don't stop at HTTP/2. Turning it off blocks CVE-2026-123007 and CVE-2026-123633 only. For HTTP/3 (CVE-2026-121810), remove QUIC listeners, set
http3 off, and consider blocking UDP 443. For the resolver (CVE-2026-123634), pointproxy_passat fixed upstream groups and turn offmirrorif you don't need it (SSI subrequests trigger it too). For the mail proxy (CVE-2026-113425), bind mail listeners to loopback or restrict them to trusted IPs. - On ingress-nginx, set
use-http2: "false"in the controller ConfigMap now and plan the move to a maintained controller. - Block SCTP where you don't use it. Add
install sctp /bin/falseto a file in/etc/modprobe.d/. This won't unload a module that's already loaded, and it won't help if SCTP is built into the kernel. There's no kernel fix yet. - Open a ticket with every vendor whose product runs nginx. Ask which nginx version they ship, whether HTTP/2 is on, and when a fixed build lands. Keep those management interfaces off the internet until you have answers.
- Watch for the exploit's footprint. Rapid7's exploit needs about 60 GB on a single connection to crash a worker and about 128 GB to reach code execution, per Upwind. A single HTTP/2 connection moving that much traffic is unusual. So are repeated worker segfaults in the error log. Treat both as worth a look, and preserve logs before you rebuild anything.
Is it being exploited?
Not in the wild, as far as anyone has reported. Every vendor and CERT we track describes lab exploitation only, and CISA hasn't added it to the KEV.
That's not much comfort. Rapid7's Stephen Fewer published a technical analysis and a PoC for CVE-2026-123007 with the disclosure, and it works against a stock nginx 1.31.8 Docker container. Xint and Wiz have built PoCs for all five flaws but haven't released them. Xint says its own exploit held up under normal concurrent traffic and across AWS, GCP, and Azure, including over a public HTTPS endpoint. The Netherlands NCSC expects proof-of-concept code to lead to widespread exploitation.
Here's what we're seeing in Mallory's data as of November 19, about a day after disclosure:
- 25 public exploit, scanner, and detection repos on GitHub. Most are scanners and log hunters, but several are research kits for the RCE itself.
- 44 vendor and CERT advisories, from F5, CERT-EU, NCSC-NL, Belgium's CCB, Debian, Ubuntu, Broadcom, and a long run of TuxCare package notices.
- 455 mentions in the past 30 days, with a single-day peak of 300.
The exploit isn't trivial. It's tuned for Debian 13 amd64 builds and depends on a specific x86-64 memory-copy implementation. That buys time for anything that doesn't match, like ARM appliances. It doesn't make them safe. The vulnerable code is the same, and porting is a matter of effort.
What the bugs are
Xint found CVE-2026-123007 while preparing for ZeroDay Cloud 2026, and the other four in an automated analysis run a few days later. All five are memory-safety bugs reachable without authentication.
| CVE | Component | Needs | First affected | Blocked by disabling HTTP/2? |
|---|---|---|---|---|
| CVE-2026-123007 | HTTP/2 flow control | HTTP/2 enabled | 1.9.5 | Yes |
| CVE-2026-123633 | HTTP/2 connection draining | HTTP/2 + proxying to an upstream | 1.19.7 | Yes |
| CVE-2026-121810 | HTTP/3 connection draining | HTTP/3 + proxying to an upstream | 1.25.0 | No |
| CVE-2026-123634 | Resolver | resolver, request-derived proxy_pass, mirror with mirror_request_body off, or SSI subrequests | 0.6.18 | No |
| CVE-2026-113425 | Mail proxy | proxy on with auth_http | 0.5.33 | No |
The lead bug is an accounting error. When HTTP/2 streams get cancelled, nginx hands their unsent data allowance back to the connection's send window without checking the protocol maximum. The window grows past what's allowed, a later check underflows, and a negative length turns into a huge memory copy in the TLS output path. That gives an attacker out-of-bounds heap reads and writes. Usually that crashes the worker. With ASLR bypassed, it's code execution, and Wiz reports lab bypasses.
CVE-2026-123633 is harder to reach. The attacker has to hold roughly 15/16 of worker_connections (about 960 of the default 1,024) to force nginx to drain a connection that's still in use. There's no public PoC for it yet.
The SCTP chain
Worker-level code execution isn't root. Xint closed that gap with a separate Linux kernel bug in SCTP, introduced in 4.15 in 2018: a waiting send can have its association moved to another socket, which leads to a use-after-free or double free. Chained with the nginx bug, Xint got a reliable root shell, and from a Docker container it reached root on the host. Whether that works on your systems depends on the kernel, which modules load, and your container restrictions. There's no kernel patch yet, which is why step 6 above matters.
The nginx you didn't install
Everyone's patching their web tier this week. That's the part you control. nginx is also one of the most embedded pieces of software on the internet. It's the management UI on gateways and NAS boxes, the front end on security appliances, and the core of every Kubernetes ingress controller built on it. When it ships inside someone else's product, you own the risk and the vendor owns the fix, on their patch clock.
That's a third-party risk problem, and it looks like this right now:
- Ubiquiti. Threatbear pulled Ubiquiti's current UniFi Cloud Gateway Ultra firmware (UniFi OS 6.0.17, released November 10) from Ubiquiti's own firmware API and unpacked it. It ships nginx 1.30.6 with the HTTP/2 module compiled in. Nine console templates set
http2 onacross ports 443, 8444, 11083, and 11084. The previous release, 5.1.35, ships 1.30.4, also with HTTP/2 on. Threatbear rates the gateway exposed to CVE-2026-123007 and likely exposed to CVE-2026-123633. Threatbear's caveats are fair ones: this was static analysis of published images, no device was exploited, and the gateway is ARM64 while Rapid7's PoC targets x86-64. The vulnerable code is still there. Thanks to Threatbear for bringing the Ubiquiti dependency to our attention. - Synology. Synology published advisory Synology-SA-26:20, rated critical. DSM 7.1 through 7.4, DSM Enterprise 1.0, BeeStation OS 1.5, and APM 1.2 and 2.0 are affected. Fixes for DSM 7.4 (7.4.1-90080-2) and 7.3 (7.3.2-86009-6) are scheduled by November 20. The rest are in progress. SRM isn't affected.
- Kubernetes ingress-nginx. Archived, HTTP/2 on by default, no fix coming. Chainguard's fork shipped fixes in controller-v1.15.15 and v1.15.16. RKE2 (which bundles it as rke2-ingress-nginx) and the AKS Application Routing add-on have no statement yet. F5's NGINX Ingress Controller 5.8.0 and NGINX Gateway Fabric 2.9.0 are fixed, though Gateway Fabric ships with HTTP/2 on.
- Forks. OpenResty, Angie, Tengine, and freenginx carry the same code. None had published a fix or statement as of November 19. The latest OpenResty is based on nginx 1.31.8.
- Broadcom. Its Enterprise Security Group is still investigating. No products are confirmed affected, and some remain under review.
- Splunk. Splunk is investigating (SVD-2026-1112) and hasn't named any affected products yet.
- Peplink. Balance and MAX routers up to firmware 8.6.1 are affected, but only when “Allow Starlink Dish API proxy access” is on. It's off by default. The fix is due in 8.6.2. Switches, access points, FusionHub, and InControl 2 aren't affected.
- Kong. Kong Gateway Enterprise and AI Gateway are affected by CVE-2026-123007 and CVE-2026-123633 when HTTP/2 is on, and fixed builds are out. The advisory doesn't cover Kong OSS or Kong Ingress Controller.
- Patched without a statement. Plesk (18.0.82 Update 4 and 18.0.81 Update 9) and Sentry self-hosted moved to fixed nginx builds without naming the CVEs.
- Ships nginx, no statement yet. Zimbra bundles nginx 1.24.0 as its mail and web proxy, with
auth_http. That's the exact setup CVE-2026-113425 needs. GitLab Omnibus bundles nginx 1.31.7, and Unraid 7.3.4 bundles 1.30.6. pfSense and the Cisco IOS XE Web UI run their management interfaces on nginx. Harbor, Discourse, Bitnami images, and Apache APISIX ship nginx or OpenResty too. - Not affected by default. QNAP QTS doesn't install nginx by default.
Ubiquiti, Synology, Peplink, and Kong are the ones with public data so far. They won't be the only ones. An unauthenticated RCE in the web server those appliances run on is exactly the shape of bug that gets picked up. The appliance with nginx inside it is the one most likely to sit on the internet, run as a privileged process, and get patched last.
So the question for this week isn't only “did we patch nginx.” It's “which of our vendors ship nginx, and do they know yet.” This is a good week to ask.
Why your scanner may not flag it
Three reasons, and they stack.
- Backports keep the old version string. Distributions fix the bug without changing the upstream version. AlmaLinux 9's fixed build still reports 1.20.1 from
nginx -v. Compare the full package version (rpm -q nginxordpkg -l nginx) against your distribution's advisory. AlmaLinux's first builds also fixed only the two HTTP/2 bugs, so check which build you have. - Appliances don't expose a package manager. A host-based scanner can't see the nginx inside a gateway or NAS. A network scanner sees a banner, if the vendor didn't strip it. You need the vendor to tell you, or someone to unpack the firmware the way Threatbear did.
- Exposure depends on config, not just version. A vulnerable build with HTTP/2 off isn't exposed to the lead bug. A fixed-looking build with HTTP/2 on might be. Appliance configs are often rendered at boot, so the config file in the image may not match what's running. Threatbear's suggestion for UCG-Ultra owners applies broadly: confirm on the device with
nginx -vandcurl -vk --http2 https://<gateway>/.
Who has shipped fixes
Status as of November 20, 2026.
| Product | Status | Fixed version or note |
|---|---|---|
| nginx Open Source stable | Fixed | 1.30.7 |
| nginx Open Source mainline | Fixed | 1.31.9 |
| NGINX Plus 37.x | Fixed | 37.0.8.1 / 37.3.0.1 |
| NGINX Plus R36 and earlier | No fix listed | Move to a fixed 37.x build |
| F5 NGINX Ingress Controller | Fixed | 5.8.0 |
| F5 NGINX Gateway Fabric | Fixed | 2.9.0 (HTTP/2 on by default) |
| Official Docker image | Fixed | Mainline and latest tags point to 1.31.9. Stable tags point to 1.30.7. Pull and redeploy. |
| Ubuntu 22.04, 24.04, 26.04 | Fixed | USN-9159-1 (backports) |
| Ubuntu 16.04, 18.04, 20.04 | Fixed for Ubuntu Pro only | USN-9159-1 |
| AlmaLinux 9, 10 | Fixed | Backports. Early builds covered HTTP/2 bugs only. |
| RHEL 8, 9, 10 / Rocky | Affected | RHSB-2026-019 confirms RHEL 8, 9, and 10 and 3scale APIcast are affected. No errata yet. |
| Debian 12, 13 | Fixed | Debian 13: DSA-6642-1 (1.26.3-3+deb13u11). Debian 12: DLA-4918-1 (1.22.1-9+deb12u11). |
| Synology DSM 7.3, 7.4 | Fix scheduled | By November 20 |
| Synology DSM 7.1, 7.2.x, others | In progress | Move to DSM 7.3 or 7.4 when those fixed builds ship |
| Ubiquiti UCG-Ultra | Affected (per Threatbear) | No advisory or fixed firmware |
| Community ingress-nginx | Won’t be fixed | Archived. Disable HTTP/2 and migrate. |
| OpenResty, Angie, Tengine, freenginx | No statement | Apply config stopgaps |
| F5 Distributed Cloud, NGINX One Console | Under investigation | Per F5’s CVE advisories |
| Broadcom, Splunk | Under investigation | No products confirmed affected (SVD-2026-1112 for Splunk) |
| Kong Gateway Enterprise, AI Gateway | Fixed | 3.4.3.30, 3.10.0.21, 3.13.0.13, 3.14.0.18, 3.15.0.8, 3.16.0.3. Advisory doesn’t cover Kong OSS or Kong Ingress Controller. |
| Chainguard ingress-nginx fork | Fixed | controller-v1.15.16 |
| Amazon Linux 2023, Elastic Beanstalk | Fixed | ALAS2023-2026-3494 (1.30.7) |
| Alpine | Fixed on 3.22 and later | 1.30.7-r0 (3.24, edge), 1.28.3-r8 (3.22, 3.23). 3.21 not fixed. |
| SUSE, openSUSE | Partly fixed | Tumbleweed 1.31.9-1.1. SLES and Leap in progress. |
| Plesk | Fixed | 18.0.82 Update 4 and 18.0.81 Update 9 (nginx 1.30.7) |
| Sentry self-hosted | Fixed | Moved to nginx 1.31.9 |
| Peplink Balance, MAX | Affected if Starlink Dish API proxy access is on | Off by default. Fix due in firmware 8.6.2. |
| Fedora, Oracle Linux | No advisory yet | Disable HTTP/2 until a package lands |
| Zimbra, GitLab, Unraid, pfSense, Cisco IOS XE | No statement | All ship nginx. Ask the vendor. |
| RKE2 ingress, AKS Application Routing | No statement | Set use-http2: "false" in the controller ConfigMap |
| Linux kernel (SCTP) | No fix | Block the module where unused |
If you know of a vendor statement we've missed, tell us and we'll add it.
Sources and method
We built this from Mallory's tracking of CVE-2026-123007 and the related NGINX2Root story (114 references and 455 mentions as of November 19), plus primary sources: the nginx.org security advisories, F5 articles K000163926 (CVE-2026-123007), K000163928 (CVE-2026-123633), K000163927 (CVE-2026-121810), K000163925 (CVE-2026-123634) and K000163931 (CVE-2026-113425), Xint's write-up, Rapid7's and Wiz's disclosures, CERT-EU Security Advisory 2026-022, NCSC-NL, Ubuntu USN-9159-1, the AlmaLinux announcement, Synology-SA-26:20, Broadcom's advisory, Debian DSA-6642-1 and DLA-4918-1, MS-ISAC Advisory 2026-136, Splunk SVD-2026-1112, Red Hat RHSB-2026-019, Kong's advisory, Peplink's advisory, ALAS2023-2026-3494, the SUSE and Alpine trackers, and Threatbear's UCG-Ultra firmware assessment. Exploit cost figures and Kubernetes findings come from Upwind. Fork and distribution status comes from GetPageSpeed and Qualys. Where sources disagree, we've said so.
This is what we're working on at Mallory: tying a bug like this to the products you actually run, including the ones you didn't install yourself.



