Skip to content

Mitigating NGINX2Root in Exchange4all 8.9.5

On 2026-11-18 Xint published NGINX2Root, five nginx vulnerabilities that range from unauthenticated remote code execution in the nginx worker to root on the host, when combined with a separate Linux kernel bug. The fix is nginx 1.31.9.

APISIX, the reverse proxy in front of every HTTP, IMAP and POP3 port of the container, runs on OpenResty and therefore on nginx. Exchange4all 8.9.5 bundles OpenResty 1.29.2.4 with nginx 1.29.2, inside the affected range of 1.9.5 to 1.31.8, and so does every earlier release. Until a release with the fixed nginx is available, disable HTTP/2 in APISIX as described below. This closes the two findings that apply to the default configuration.

Which findings apply

Finding Needs Exchange4all 8.9.5
CVE-2026-123007 HTTP/2 Affected. HTTP/2 is enabled on port 443 and, as cleartext h2c, on port 80.
CVE-2026-123633 HTTP/2 and proxying to an upstream Affected. Every route proxies to a backend service.
CVE-2026-121810 HTTP/3 (QUIC) and proxying to an upstream Not affected, unless you set enable_http3: true in /storage/apisix-config.yaml.
CVE-2026-123634 resolver, proxy_pass, mirror and mirror_request_body off together Not affected, unless you added the proxy-mirror plugin to /storage/apisix-config.yaml.
CVE-2026-113425 nginx mail proxy with auth_http Not affected. nginx is built without the mail modules; IMAP and POP3 pass through the stream proxy.
Linux kernel (SCTP) Code execution on the host Not part of the container. The container shares the kernel of the host, see The host kernel.

Before you disable HTTP/2

ad-connect and the Kopano importer stop working without HTTP/2

Both talk gRPC through APISIX, and gRPC requires HTTP/2. With HTTP/2 disabled:

  • ad-connect can no longer reach the server. That applies to the e4a-ad-connect service inside the container as well, because it connects through https://<E4AFQDN>:443. It logs tls: no application protocol every few seconds, no users are imported, and password checks that are delegated to your directory fail, so those users cannot sign in.
  • The Kopano importer at /e4a.importer_d1.v1.importer/ stops accepting data.

Web, Outlook (MAPI over HTTP and RPC over HTTP), ActiveSync, EWS, IMAP and POP3 keep working over HTTP/1.1.

If your users sign in through ad-connect, you have to choose until the fixed release is out: either keep HTTP/2 and allow only trusted networks, such as your VPN, to reach ports 80 and 443, or disable HTTP/2 and lose the directory sign-in. A firewall restriction also locks out mobile devices and Outlook clients outside those networks.

Disabling HTTP/2

Run these commands in the directory of your docker-compose.yml. ./setup.sh -q --exec instead of docker compose exec exchange4all works as well, see Checking the status and running commands.

  1. Check the current settings. enable_http2 is true by default, enable_http3 must be false and proxy-mirror must not be in the list of plugins:

    docker compose exec exchange4all yq '.apisix.enable_http2, .apisix.ssl.listen, .plugins' /storage/apisix-config.yaml
    
  2. Disable HTTP/2:

    docker compose exec exchange4all yq -i '.apisix.enable_http2 = false' /storage/apisix-config.yaml
    

    If step 1 showed enable_http3: true, disable HTTP/3 as well:

    docker compose exec exchange4all yq -i '.apisix.ssl.listen[].enable_http3 = false' /storage/apisix-config.yaml
    
  3. Restart APISIX:

    docker compose exec exchange4all sv restart apisix
    

    Only a restart generates a new nginx configuration and closes the HTTP/2 connections that are already open. apisix-refresh-config.sh is not enough: it reloads the configuration that was generated at the last start, which still has HTTP/2 enabled. The restart drops all open HTTP, IMAP and POP3 connections for a few seconds; clients reconnect on their own.

  4. Check that the generated configuration no longer enables HTTP/2. The command prints 0:

    docker compose exec exchange4all grep -c 'http2 on' /run/apisix/nginx-run/nginx.conf
    

Verifying from outside

Run these checks from a machine outside your network, against the FQDN of your deployment. Replace mail.example.com with it.

  1. TLS no longer negotiates HTTP/2. Before the change this printed ALPN protocol: h2:

    openssl s_client -connect mail.example.com:443 -servername mail.example.com -alpn h2,http/1.1 </dev/null 2>/dev/null | grep ALPN
    
    ALPN protocol: http/1.1
    

    The result only means something if the TLS connection itself succeeds. No output at all points to a connection or certificate problem, not to a disabled HTTP/2.

  2. HTTPS still serves the WebApp, over HTTP/1.1. This prints 1.1 200:

    curl -sS --http2 -o /dev/null -w '%{http_version} %{http_code}\n' https://mail.example.com/webapp/
    
  3. Port 80 no longer accepts cleartext HTTP/2. Before the change this printed 2, now curl fails with exit code 16 and an HTTP/2 error such as Remote peer returned unexpected data while we expected SETTINGS frame:

    curl -sS --http2-prior-knowledge -o /dev/null -w '%{http_version}\n' http://mail.example.com/
    

Keeping the setting across updates

When a new release ships a newer APISIX configuration, the container merges it into /storage/apisix-config.yaml on start. The shipped value wins, so the update sets enable_http2 back to true. A previous version of the file is kept as /storage/apisix-config.yaml.e4a-old.<timestamp>.

After every update, until you run a release that contains the fixed nginx, check the value and repeat Disabling HTTP/2 if it prints true:

docker compose exec exchange4all yq '.apisix.enable_http2' /storage/apisix-config.yaml

Leave auto_update alone

Setting e4a_config.auto_update: false in the file would keep your value, but it also keeps every other configuration change of future releases away from your deployment.

After updating to the fixed release

The release notes of the fixed release mention NGINX2Root. Once you run it, enable HTTP/2 again so that ad-connect and the importer work, and restart APISIX:

docker compose exec exchange4all yq -i '.apisix.enable_http2 = true' /storage/apisix-config.yaml
docker compose exec exchange4all sv restart apisix

The checks from outside show ALPN protocol: h2 and 2 200 again. If you run ad-connect, its log shows got hello once it has reconnected.

The host kernel

The step from the nginx worker to root on the host uses a separate Linux kernel bug in SCTP. The container runs on the kernel of the host, so no Exchange4all release can fix it. Update the kernel of the host according to your distribution's advisory and reboot.

Looking for signs of an attack

Failed exploitation attempts can crash nginx workers, and repeated crashes are one of the signs Xint names. Each crash is logged to /storage/log/apisix/apisix-error.log, on the host /srv/exchange4all/e4a/log/apisix/apisix-error.log by default. This counts them in the current file:

docker compose exec exchange4all grep -c 'exited on signal' /storage/log/apisix/apisix-error.log

Any match warrants a closer look. Commands an attacker starts through nginx run as child processes of nginx. On a clean system this prints only the header line:

docker compose exec exchange4all sh -c 'ps -o pid,ppid,user,args --ppid "$(pgrep -d, -x openresty)" | grep -v "nginx: "'

If you find signs of a compromise, patching does not undo it: rotate the credentials the container had access to, among them the certificates and the secrets in /storage, and rebuild the host from a trusted state.