Cloudflare says its proxy shields origin servers from the HTTP/2 and HTTP/3 attacks among five newly disclosed nginx vulnerabilities, but administrators still need to patch. F5 and nginx.org disclosed the flaws on November 18, 2026, as part of a wider announcement covering ten nginx vulnerabilities.
The five bugs detailed by research team Xint can crash an nginx worker or, according to Xint, run commands on a server. Xint also found a Linux kernel flaw that can turn that access into root privileges, including inside a Docker container. The researchers published their work as nginx2root.
Cloudflare says attackers connecting through its HTTP proxy cannot send the affected HTTP/2 and HTTP/3 frames directly to a protected origin. The edge terminates those client connections and opens a separate connection to the origin. Cloudflare says its own edge is unaffected and that it updated related paths in its modified nginx code before public disclosure.

nginx Vulnerabilities: Cloudflare Protection Limits
These nginx vulnerabilities include attacks the proxy can limit, but the protection applies only to proxied hostnames and the HTTP/2 and HTTP/3 attacks. It does not address every flaw or configuration. The resolver and mail proxy vulnerabilities depend on particular nginx features, and administrators should update even when Cloudflare sits in front of a server.
Cloudflare recommends nginx 1.30.7, 1.31.9 or a fixed build from the server vendor, followed by a restart. Operators can also disable affected features where practical. Cloudflare’s nginx2root analysis describes the proxy’s limits and the five flaws.
The nginx vulnerabilities have different prerequisites. CVE-2026-123007 affects HTTP/2 and involves an integer overflow. CVE-2026-123633 involves HTTP/2 proxying when worker connection slots run low. CVE-2026-121810 affects HTTP/3 with QUIC and proxying. CVE-2026-123634 involves resolver lookups, variable-based proxying and a subrequest. CVE-2026-113425 affects mail proxy configurations using auth_http.
Four are use-after-free bugs, in which code releases memory another part of the program still uses. The remaining flaw corrupts HTTP/2 flow-control accounting. F5 rates the use-after-free flaws as potential code execution when address-space layout randomization is disabled or bypassed. Xint reported code execution with nginx worker privileges in testing.
What Administrators Should Check
Known attacks for CVE-2026-123007 require very large volumes of HTTP/2 traffic. Rapid7 estimates its proof of concept sent approximately 128 GB, including about 15 GiB of other request traffic. That describes its approach, not a proven minimum. A single long-lived connection receiving far more data than access logs explain may merit investigation.
Other indicators vary by flaw and logging configuration. A worker crash can accompany the HTTP/2 or HTTP/3 use-after-free bugs, though a crash is not guaranteed. The resolver issue may produce a “could not cancel” alert on vulnerable versions, but its absence does not rule out exposure. Unexpected programs started by an nginx worker are a stronger sign of successful exploitation.
Cloudflare’s Zero Wire vulnerability feed explains how its briefings help operators investigate internet flaws. The publication has also covered a Linux kernel flaw, offering related context for teams tracking vulnerabilities.
To address these nginx vulnerabilities, administrators should install a fixed release and restart services so old connections close. They should restrict mail proxy listeners to trusted networks where that feature is used, and block SCTP if unnecessary while awaiting a kernel fix. Cloudflare says origin access should be limited so its proxy is the only route to the server. A PHP security update is another reminder to track fixes across the stack.
The Cloudflare Radar Zero Wire entry includes a checklist and links to vendor advisories. Cloudflare recommends using it to assess affected systems while applying nginx and kernel updates as they become available.









