CVE-2026-119618: Triage the Apache HTTP/2 RCE on RHEL

Most teams meet a vulnerability the same way: someone pastes a bare link into a channel. This time it is Red Hat's page for CVE-2026-119618, with no context and a 9.8 score at the top. The next question is "are we affected?", and answering it is slow when nobody knows where every httpd runs.
This post walks through a triage loop you can finish before your next patch window: confirm what is affected, find the hosts and images that are actually exposed, decide whether to patch or mitigate, and leave behind an inventory query so the next CVE is a lookup.
What CVE-2026-119618 Is
The facts below come from the Apache httpd 2.4 security page, the oss-security announcement, the CVE record, and Red Hat's page.
- Component:
assign_headers()in mod_http2, the module that implements HTTP/2 in Apache HTTP Server. Apache describes it as a synchronization flaw on a shared resource that "can lead to RCE." Red Hat says a remote, unauthenticated attacker could execute arbitrary commands. - Affected upstream: httpd 2.4.55 through 2.4.70, and the standalone mod_h2 module 2.0.8 through 2.0.42.
- Fixed upstream: httpd 2.4.71, released on November 11, 2026.
- Condition: Red Hat states that "only servers that have HTTP/2 enabled are affected; servers configured to serve only HTTP/1.1 are not vulnerable."
- Credit: Anthropic's OSS Scanner and Google Cloud Vulnerability Research.
The scores disagree, which is normal for a fresh CVE. Red Hat gives a CVSS 3.1 base of 9.8 and an overall rating of Critical, but rates RHEL 9 and RHEL 10 as Important, because each HTTP/2 stream gets its own allocator there and that makes exploitation harder. CISA's enrichment in the CVE record and Ubuntu both use 8.1, with high attack complexity. Do not spend the triage meeting on which number is right. Your exposure decides the priority, not the score.
On exploitation: Red Hat says public research "has demonstrated a reliable exploit against a default-configured server with the mod_http2 module enabled." Wiz wrote on November 11 that no exploit was available and there were no reports of exploitation in the wild, and the CISA SSVC entry in the CVE record also says exploitation "none". We did not find a public proof of concept. Treat that as "not yet", not as "safe".
Step 1: Confirm the Red Hat Products and Fix Versions
Red Hat's security data API lists the package state per product:
| Product | State | Advisory | Fixed package |
|---|---|---|---|
| RHEL 10 | Fixed | RHSA-2026:98978 | mod_http2-2.0.29-5.el10_3.1 |
| RHEL 10.0 EUS / 10.2 EUS | Fixed | RHSA-2026:98953 / 98979 | mod_http2-2.0.29-2.el10_0.3 / 2.0.29-4.el10_2.3 |
| RHEL 9 | Fixed | RHSA-2026:98981 | mod_http2-2.0.26-7.el9_9.1 |
| RHEL 9.4 E4S, 9.6 EUS, 9.8 EUS | Fixed | RHSA-2026:98968 / 98954 / 98982 | 2.0.26-2.el9_4.4 / 2.0.26-4.el9_6.3 / 2.0.26-6.el9_8.3 |
| JBoss Core Services 2.4.62.SP7 | Fixed | RHSA-2026:100552 | jbcs-httpd24-mod_http2 |
| RHEL 6, 7, 8 | Not affected | ||
| Red Hat Hardened Images | Not affected |
Note the package versions: 2.0.29 and 2.0.26. That is the first trap. Red Hat backports the fix, so a fixed RHEL host still reports a mod_http2 version lower than the upstream 2.0.42. A scanner or script that compares version strings against the upstream range flags every patched RHEL box as vulnerable, and the real findings drown in the noise. Ubuntu does the same thing: on 24.04 LTS the fix is apache2 2.4.58-1ubuntu8.19, which does not look like 2.4.71. Always compare against the distro's fixed build, never the upstream version.
On RHEL, ask the package manager instead of reading version numbers:
dnf updateinfo list --cve CVE-2026-119618 # lists the advisory while the fix is missing
rpm -q mod_http2
Step 2: Find What Actually Speaks HTTP/2
An installed mod_http2 package is not an exposure. The module must be loaded, and the Protocols directive must offer h2 or h2c. The core docs give the default as Protocols http/1.1, and the mod_http2 docs say you "must enable HTTP/2 via Protocols" to use it. So check three things on each host:
rpm -q mod_http2 # installed?
httpd -M 2>/dev/null | grep http2_module # loaded?
grep -RiE '^\s*Protocols' /etc/httpd/ # offered, and in which vhosts?
Then check from the outside, because config files lie about what the running process negotiates:
curl -s -o /dev/null -w '%{http_version}\n' --http2 https://app.example.com/ # 2 means h2
curl -s -o /dev/null -w '%{http_version}\n' --http2-prior-knowledge http://10.0.4.12/ # 2 means h2c
The second trap is the edge. If a load balancer or CDN terminates TLS, the probe above tells you what the edge speaks, not what httpd speaks. Probe the origin directly from inside the network as well. An origin with h2c enabled behind a proxy is still reachable by anything on that network segment.
Images need the same checks. For every image that runs httpd, run the package query inside it:
podman run --rm --entrypoint rpm registry.internal/portal:2026.2 -q mod_http2
podman run --rm --entrypoint httpd registry.internal/portal:2026.2 -M | grep http2
A RHEL 9 or RHEL 10 image built before November 11 that installs mod_http2 is vulnerable until it is rebuilt, whatever the hosts underneath it run. Your image scanner, such as Trivy, catches the package, but only the config tells you whether h2 is on.
Step 3: Patch or Mitigate
The fix is the vendor package plus a restart. The mitigation, if you cannot patch in time, is to stop offering HTTP/2. Wiz recommends exactly that as a temporary step: remove h2 and h2c and set Protocols http/1.1. This fits Red Hat's statement that HTTP/1.1-only servers are not vulnerable.
| mod_http2 loaded | h2 or h2c offered | Reachable from | Fixed package installed | Action |
|---|---|---|---|---|
| Yes | Yes | Internet | No | Patch today. If you cannot, set Protocols http/1.1 today and patch in the next window |
| Yes | Yes | Internal network or VPN | No | Patch in this window, owner and date on the ticket |
| Yes | No | Any | No | Patch on the normal cadence; add a check that h2 stays off |
| No | No | Any | No | Patch on the normal cadence, so a later config change does not expose it |
| Any | Any | Any | Yes, but process older than the package | Restart httpd; the old code is still in memory |
| Image built before the fix | Any | Any | No | Rebuild and redeploy; patching hosts does not fix the image |
| JBCS or vendor bundle | Yes | Any | Unknown | Check the vendor advisory; JBCS has RHSA-2026:100552 |
Three operational notes:
- Restart, then verify. A package update does not replace a running process.
needs-restarting -s(fromdnf-utils) lists services still running old libraries. - Disabling h2 has a cost. Clients fall back to HTTP/1.1 and lose multiplexing. For most backends that is fine for a week. For gRPC over httpd it is not, so find those vhosts before you push one
Protocolsline everywhere. - Write the mitigation down as temporary. Put a date on the ticket to remove it after the patch, or HTTP/1.1 becomes the permanent config and nobody remembers why.
Step 4: Keep the Inventory Query
The expensive part of this triage was not the patch. It was answering "where does httpd run, with which modules, offering which protocols?" Make that answer a scheduled job, not a scramble. A minimal per-host record:
printf '%s,%s,%s,%s\n' \
"$(hostname)" \
"$(rpm -q mod_http2 2>/dev/null || echo none)" \
"$(httpd -M 2>/dev/null | grep -c http2_module)" \
"$(grep -RhiE '^\s*Protocols' /etc/httpd/ 2>/dev/null | tr -s ' ' | paste -sd ';')"
Collect it across the fleet, add the image list from your registry, and keep owners next to each row. The next mod_http2, mod_ssl or mod_proxy CVE then starts with a filter on a table you already have. We made the same argument for runtimes in the Java4Shell inventory post and for edge proxies in the NGINX Rift post: the inventory outlives the CVE.
Where an Agent Helps
This loop is mostly reading and cross-referencing, which an agent does well. In Akmatori you can write a skill that runs steps 1 to 3 and reports the decision table. The SSH connector is read-only by default. For commands outside its diagnostic set, you add allowed command patterns on the connector, such as rpm -q * or httpd -M. An allowed command can never contain $, a backtick or a redirect, so the skill should run each check as a plain separate command rather than the printf one-liner above. A cron job with connector tools allowed can run the inventory on a schedule. With write operations turned on in the Jira connector, the agent can open the remediation issues itself. A human still decides when to restart production httpd.
Conclusion
CVE-2026-119618 matters only where httpd actually negotiates HTTP/2, and that is the question most teams cannot answer quickly. Check the distro fixed build, not the upstream version. Probe both the edge and the origin. Patch and restart, or turn off h2 with a dated ticket. Then keep the query, so the next bare CVE link takes minutes.
If your team wants AI-assisted incident workflows on your own infrastructure, Akmatori helps SRE teams investigate alerts, coordinate response, and automate safe infrastructure actions. Powered by Gcore for global infrastructure reliability.
