This is a master-key flaw on the locked management hallway, not the front lobby
Tenable plugin 207464 maps to Broadcom advisory VMSA-2024-0019 and really means two bugs in VMware vCenter Server: CVE-2024-38812, an unauthenticated DCERPC heap overflow that can lead to remote code execution, and CVE-2024-38813, a privilege-escalation flaw that can push access to root. Affected versions are vCenter 7.0 before 7.0 U3t and vCenter 8.0 before the fixed trains 8.0 U2e or 8.0 U3d; the concrete safe builds are 7.0.3.02200 build 24322018, 8.0.2.00500 build 24321653, and 8.0.3.00400 build 24322831.
Vendor CRITICAL is technically defensible for the RCE half, but it slightly overstates real-world population risk because exploitation is only possible where an attacker can actually talk to vCenter's management/DCERPC surface. That reachability requirement is real friction in mature enterprises, so this lands as HIGH, not CRITICAL; the score stays near the ceiling because the issue is unauthenticated, KEV-listed, actively exploited, and hits the virtualization control plane.
4 steps from start to impact.
Find a reachable vCenter
Masscan, Nmap, or exposed-asset search to identify vCenter by /ui or :5480, then confirms the host is on a vulnerable build. Shadowserver explicitly notes its HTTP-side detection is version-based and separately tags systems where the DCERPC service is exposed, which is the difference between 'old version' and 'actually reachable exploit path.'- Network path to the vCenter management plane
- vCenter running a vulnerable 7.0 or 8.0 build
- DCERPC-relevant service reachable from the attacker's position
- Many enterprises keep vCenter off the public internet and behind jump hosts or management VLANs
- A version-only finding does not prove the exploitable DCERPC path is exposed from the attacker vantage point
- ZTNA, VPN segmentation, and ACLs commonly reduce reachable population
207464 is version-based; Shadowserver reports cve-2024-38812 and adds vcenter-dcerpc-exposed when the exploitable service is also reachable.Trigger CVE-2024-38812 for pre-auth code execution
- No authentication required
- Accurate packet delivery to the vulnerable service
- Target not already patched to
7.0 U3t,8.0 U2e, or8.0 U3dor later
- Exploit reliability on memory-corruption bugs is never as universal as the CVSS math implies
- Some environments expose the web UI but not the necessary service path to the attacker
- Network IDS/IPS or management-plane filtering can break direct reachability
Use CVE-2024-38813 or local follow-on to get root
CVE-2024-38813, which matters because root on vCenter is not just a single-host problem; it is control-plane ownership.- Initial code execution or equivalent network-triggerable path
- Vulnerable build still present
- Ability to interact with local services or privileged components
- This step assumes the attacker already reached the management plane or landed code locally
- EDR on appliance-like infrastructure is often absent, but hardened bastion access can still reduce attacker options
Exploit the control plane, not just the box
govc, vSphere APIs, guest operations, snapshots, credentials, and inventory access. This is why the blast radius is so severe: compromise of one management node can cascade to many ESXi hosts and VMs.- Administrative or root-level control of vCenter
- Managed ESXi hosts and VM inventory attached to that vCenter
- No compensating segmentation between vCenter and the rest of the estate
- RBAC, separate management domains, and limited linked-mode scope can reduce tenant or cluster blast radius
- Operational monitoring may catch disruptive post-exploitation like mass snapshots or VM tampering
The supporting signals.
| In-the-wild status | Confirmed exploited. Broadcom states exploitation occurred in the wild for CVE-2024-38813; NVD shows both CVE-2024-38812 and CVE-2024-38813 in CISA KEV. |
|---|---|
| KEV status and dates | KEV-listed on 2024-11-20 for both CVEs; NVD reflects CISA due date 2024-12-11 for federal remediation. |
| Proof-of-concept availability | Public barrier is lowered. GitHub references for CVE-2024-38812 exist in CVE aggregation sources, and Broadcom credits zbl and srs of team TZL from the 2024 Matrix Cup for discovery. Black Hat Asia 2025 research further documented the bug class and exploitation path. |
| EPSS | CVE-2024-38812 shows EPSS ~77.9% in Tenable's CVE page; CVE-2024-38813 is commonly tracked around 29.5%-31% in multiple vulnerability databases. Translation: the RCE half is in the 'expect real attacker attention' bucket. |
| CVSS reality check | CVE-2024-38812 is 9.8 with AV:N/AC:L/PR:N/UI:N and that matches reality well enough. CVE-2024-38813 is where nuance matters: VMware scores it 7.5 with AC:H/PR:L, while NVD inflates it to 9.8; for defenders, VMware's vector is the more believable operational read. |
| Affected versions | All vCenter 7.0 before 7.0 U3t and all vCenter 8.0 before 8.0 U2e or 8.0 U3d. Tenable plugin 207464 flags 7.x < 7.0 U3t and 8.x < 8.0 U3d. |
| Fixed versions | Use the corrected October 21, 2024 builds, not the incomplete September fix. Safe anchors: 7.0 U3t build 24322018, 8.0 U2e build 24321653, 8.0 U3d build 24322831. |
| Scanning and exposure data | Exposure is narrower than raw CVSS suggests. Shadowserver says patch status can be seen over HTTP, but *actual exploitation* requires the DCERPC service to be exposed and it separately tags those hosts as vcenter-dcerpc-exposed. Censys likewise warns exposed /ui and :5480 assets likely indicate vCenter components worth triaging. |
| Disclosure timeline | Disclosed 2024-09-17; advisory updated 2024-10-21. The update matters because Broadcom says the September 17 patch did not fully address CVE-2024-38812. |
| Reporting researcher / org | Reported by zbl and srs of team TZL working with the 2024 Matrix Cup contest, per Broadcom acknowledgment. |
noisgate verdict.
The single biggest reason this is HIGH instead of CRITICAL is reachability: the exploit path matters only where attackers can actually reach vCenter's management/DCERPC surface, and many enterprises intentionally keep that plane segmented. It still stays near the top of HIGH because no authentication is needed for the primary RCE and compromise lands directly on the virtualization control plane, where blast radius is far beyond one appliance.
Why this verdict
- Start at vendor 9.8 because
CVE-2024-38812is unauthenticated network RCE on a crown-jewel management server. - Adjust downward for attacker position because the bug is only valuable where the attacker can reach vCenter's management plane and, in practice, the relevant DCERPC service. That usually implies internet exposure or post-initial-access movement into admin or management segments, which shrinks the reachable population versus true edge software.
- Adjust upward again for exploitation evidence and blast radius because both CVEs hit KEV on
2024-11-20, Broadcom confirmed in-the-wild exploitation forCVE-2024-38813, and owning vCenter means owning the orchestration layer for many ESXi hosts and VMs.
Why not higher?
This is not a universal internet worm problem. Many enterprises keep vCenter off the public internet, restrict it to jump hosts, and will be flagged by version-based scanning even when the exact exploitable DCERPC path is not broadly reachable. That real-world reachability friction is enough to keep it out of CRITICAL.
Why not lower?
Once reachable, the primary flaw needs no authentication and the impact is not 'one server down' but potential compromise of the virtualization control plane. KEV status and confirmed in-the-wild exploitation remove any comfort that this is just a hard-to-use lab bug.
What to do — in priority order.
- Restrict vCenter reachability now — Limit
443,5480, and any management/DCERPC-relevant paths to dedicated admin jump hosts and management networks only. Because this issue is KEV-listed, apply this immediately, within hours, not on the normal HIGH cadence. - Block direct internet exposure — If any vCenter UI or VAMI is externally reachable, remove it from direct internet access and force administration through VPN plus bastion or ZTNA. For this KEV case, do it immediately, within hours.
- Monitor vCenter for abnormal API and root activity — Push vCenter audit logs, shell logs, and network telemetry into SIEM and alert on unusual VM operations, privilege changes, new services, and unexpected outbound connections from the appliance. Stand this up immediately, within hours where patching cannot be completed at once.
- Validate the October 21 fixed build — Do not assume the September 17 patch closed the hole; Broadcom explicitly said it was incomplete for
CVE-2024-38812. Confirm each appliance is on7.0 U3t,8.0 U2e,8.0 U3d, or later immediately, within hours.
- A WAF in front of
/uialone does not solve this, because the issue is in vCenter's DCERPC implementation and reachability may bypass simple HTTP-focused controls. - MFA on vSphere admins helps for account abuse but not for the unauthenticated RCE portion of
CVE-2024-38812. - Version-only 'patched in September' assumptions are dangerous here because Broadcom later stated that the September 17 fix did not fully address
CVE-2024-38812.
Crowdsourced verification payload.
Run this on each vCenter Server Appliance over SSH as root or an account with shell access to execute vpxd -v. Save it as check_vmsa_2024_0019.sh, then run bash check_vmsa_2024_0019.sh; it needs local shell access but no internet connectivity.
#!/usr/bin/env bash
# check_vmsa_2024_0019.sh
# Detects whether a VMware vCenter Server Appliance is vulnerable to VMSA-2024-0019
# Covers CVE-2024-38812 / CVE-2024-38813 fixed in:
# 7.0 U3t -> 7.0.3.02200 build 24322018
# 8.0 U2e -> 8.0.2.00500 build 24321653
# 8.0 U3d -> 8.0.3.00400 build 24322831
# Output: VULNERABLE / PATCHED / UNKNOWN
# Exit codes: 0=PATCHED, 1=VULNERABLE, 2=UNKNOWN
set -u
SAFE_BUILD_7_0_U3T=24322018
SAFE_BUILD_8_0_U2E=24321653
SAFE_BUILD_8_0_U3D=24322831
get_version_output() {
if command -v vpxd >/dev/null 2>&1; then
vpxd -v 2>/dev/null | head -n 1
return 0
fi
if [[ -r /var/log/vmware/vpxd/vpxd.log ]]; then
grep -m1 -E 'VMware VirtualCenter|build-' /var/log/vmware/vpxd/vpxd.log 2>/dev/null
return 0
fi
return 1
}
parse_version() {
local line="$1"
VERSION=$(echo "$line" | grep -Eo '[0-9]+\.[0-9]+\.[0-9]+' | head -n1)
BUILD=$(echo "$line" | grep -Eo 'build-?[0-9]+' | grep -Eo '[0-9]+' | head -n1)
}
version_ge() {
# returns 0 if $1 >= $2
local a="$1" b="$2"
local IFS=.
local i
read -r -a va <<< "$a"
read -r -a vb <<< "$b"
for ((i=${#va[@]}; i<3; i++)); do va[i]=0; done
for ((i=${#vb[@]}; i<3; i++)); do vb[i]=0; done
for i in 0 1 2; do
if ((10#${va[i]} > 10#${vb[i]})); then return 0; fi
if ((10#${va[i]} < 10#${vb[i]})); then return 1; fi
done
return 0
}
line="$(get_version_output)"
if [[ -z "${line:-}" ]]; then
echo "UNKNOWN: could not determine vCenter version/build (vpxd unavailable)"
exit 2
fi
parse_version "$line"
if [[ -z "${VERSION:-}" || -z "${BUILD:-}" ]]; then
echo "UNKNOWN: unable to parse version/build from: $line"
exit 2
fi
major_minor=$(echo "$VERSION" | awk -F. '{print $1"."$2}')
echo "Detected version=$VERSION build=$BUILD"
case "$major_minor" in
7.0)
# Future 7.0.4+ style versions do not exist for vCenter; build threshold is safest.
if (( BUILD >= SAFE_BUILD_7_0_U3T )); then
echo "PATCHED"
exit 0
else
echo "VULNERABLE"
exit 1
fi
;;
8.0)
# First handle clearly newer minor/update trains.
if version_ge "$VERSION" "8.0.4"; then
echo "PATCHED"
exit 0
fi
if version_ge "$VERSION" "8.0.3"; then
if (( BUILD >= SAFE_BUILD_8_0_U3D )); then
echo "PATCHED"
exit 0
else
echo "VULNERABLE"
exit 1
fi
fi
if version_ge "$VERSION" "8.0.2"; then
if (( BUILD >= SAFE_BUILD_8_0_U2E )); then
echo "PATCHED"
exit 0
else
echo "VULNERABLE"
exit 1
fi
fi
echo "VULNERABLE"
exit 1
;;
*)
echo "UNKNOWN: unsupported or unexpected vCenter version family '$major_minor'"
exit 2
;;
esac
If you remember one thing.
207464 hit as a management-plane emergency, but triage by reachability first: identify all vCenter instances, confirm whether the DCERPC-relevant path is reachable from anything outside tightly controlled admin networks, and for any reachable system patch / mitigate immediately, within hours because KEV and active exploitation override the normal noisgate mitigation SLA. For the formal patch program this is still a HIGH finding, so the noisgate remediation SLA is <=180 days, but do not consume that window on exposed or broadly reachable vCenter; move those to same-day or next-change execution and make sure the appliance is on the corrected October 21, 2024 builds, not the incomplete September 17, 2024 fix.Sources
What defenders are saying.
Crowdsourced verification outputs.
Results submitted by users who ran the verification payload against their environment.