OpenVPN 2.7.9 closes two privilege-escalation flaws in the Windows Interactive Service
On November 5, 2026, OpenVPN 2.7.9 fixed two flaws in the Windows Interactive Service that let a local user escalate privileges: one by inheriting a user-controlled environment, the other by bypassing command-line option validation. Update Windows clients and servers before you keep exposing your VPN.
November 5, 2026. OpenVPN ships version 2.7.9, closing two flaws in the Windows Interactive Service. CVE-2026-121720: the service was passing the user’s full environment to openvpn.exe. CVE-2026-115342: a mismatch between two command-line parsers let the service execute options it believed it had rejected. Both affect versions 2.4.0 through 2.7.8, and both resolve to a local privilege escalation. Why it matters: OpenVPN remains one of the most widely deployed enterprise VPNs, and its Windows Interactive Service is a prime target for anyone who already has a foothold on a workstation.
The user environment as an escalation vector
The first flaw, CVE-2026-121720, comes down to a design mistake in RunOpenvpn(), the Interactive Service routine that launches openvpn.exe. The service built the child process’s environment block from the requesting user’s full environment — including variables defined under HKCU\Environment.
Those variables are user-controlled, and some influence what code openvpn.exe loads. The textbook case is OPENSSL_CONF: by setting it, a local user can get OpenSSL to load provider modules of their choosing. Because the resulting openvpn.exe process has privileged operations through the --msg-channel connection to the service, the attacker runs code inside the VPN process and then uses it to request privileged service operations.
The fix inverts the logic: openvpn.exe’s environment is now built from system-wide variables, and only seven user variables are copied through an allowlist — APPDATA, LOCALAPPDATA, TEMP, TMP, USERDOMAIN, USERNAME, and USERPROFILE. Everything else is dropped.
Two parsers, two readings of the same command
The second flaw, CVE-2026-115342, is subtler. The Interactive Service validates the options it receives using CommandLineToArgvW(), while openvpn.exe splits its command line with the C runtime. Since Visual Studio 2008, the two parsers disagree on how to treat two consecutive double quotes inside an already-quoted string.
In practice, a crafted command line can hide restricted options from the service’s validator — which sees them as harmless text — while openvpn.exe still executes them. The direct consequence: you bypass the requirement that configuration files only load from OpenVPN’s locked-down configuration directory. A local user can force the loading of a configuration of their own making, outside the intended perimeter.
# Check the Windows client version and detect exposure
openvpn --version
# 2.7.8 and earlier are vulnerable; move to 2.7.9 (or newer). Both flaws share the same entry condition: an already-authenticated local user. This is not remote code execution — but it is exactly the link an attacker chains after phishing or an initial compromise to move from a user account to SYSTEM.
What an attacker does with them
Both flaws are exploitable from a local user account, without admin rights. The canonical scenario is the already-compromised endpoint: phishing or a malicious binary has produced a user shell, and the attacker wants the jump to SYSTEM. With CVE-2026-121720, they set an environment variable and reconnect the VPN; the Interactive Service runs it inside a process that already talks to the service with privileges. With CVE-2026-115342, they craft a command line that loads a configuration of their own making, outside the locked-down directory.
The severity comes from the fact that the Interactive Service runs as SYSTEM: once openvpn.exe is hijacked, the attacker no longer needs a separate kernel exploit to act privileged on the machine. That is why these flaws, though “local”, must be treated as full privilege escalations — the final link in a chain that often starts far away, with a simple email.
A surface that has been repeating since September
These two flaws don’t stand alone. Since September 2026, the project has published a run of Windows-focused advisories, especially around openvpnserv and the ovpn-dco-win driver. They include configuration-path validation bypasses (CVE-2026-81830, CVE-2026-78043), a binary planting via netsh.exe called without a full path (CVE-2026-84226), a use-after-free in the peer table (CVE-2026-82325), a local denial-of-service via a NULL DACL (CVE-2026-82312), and a kernel heap corruption (CVE-2026-11604).
The pattern is consistent: the Windows Interactive Service and the kernel driver have become the main attack surface, while the cross-platform protocol core has held. It’s a reminder that VPN security is not only about tunnel cryptography — it’s also about the code running as SYSTEM or in the kernel on the local machine, within reach of any user account.
Note that the cross-platform protocol core — the encrypted data channel — is not at fault. The recent flaws concentrate on the Windows layer: the service that manages local connections and the ovpn-dco-win driver. That is good news for tunnel cryptography, but bad news for fleets: most enterprise deployments mix Windows clients with a Linux server, and it is precisely the Windows side that now concentrates the risk.
Mitigation beyond the version bump is mostly about shrinking the local surface. Restrict who is allowed to use the Interactive Service — on shared or kiosk endpoints, that is the set of accounts an attacker would target first. Audit which configurations are actually loaded: the directory lock that CVE-2026-115342 bypasses exists precisely to stop a user from pointing the client at an arbitrary configuration. And if your fleet still runs the ovpn-dco-win kernel driver, weigh it against the client’s own user-space path — the recent advisories have repeatedly implicated the driver in memory-safety issues.
Seen in aggregate, the September-to-November run is a useful signal: a mature, widely-audited codebase can still harbor a long tail of local privilege bugs in the OS-specific glue. The core protocol held; the Windows service layer did not. For defenders, that argues for treating the VPN client as part of the endpoint’s local attack surface, not merely as an encrypted pipe.
Verdict
If you deploy OpenVPN on Windows — endpoints, access servers, or appliances — move to 2.7.9 without delay: both flaws are low-cost local privilege escalations, and the update requires no reconfiguration. The fix is clean (an allowlisted environment block, a parser alignment), so it is an update with no expected regression. Beyond the version, shrink the surface: restrict who can use the Interactive Service, and question the ovpn-dco-win driver if your fleet is heterogeneous. For a corporate VPN, a local host elevated to SYSTEM is often worth as much as a breach in the tunnel itself — because it is from that host that the attacker pivots back into the rest of the network. On access servers, keep the Interactive Service out of reach of ordinary accounts — that is the channel both flaws hijack to reach SYSTEM.
Références
- OpenVPN — Security Announcements (wiki, November 2026)
- OpenVPN — CVE-2026-121720: “Windows: Interactive service passes user environment to openvpn.exe”
- OpenVPN — CVE-2026-115342: “Windows: Interactive service option validation bypass via command-line parsing mismatch”
- OpenVPN — Release notes 2.7.9 (November 5, 2026)
- CVE.org — CVE-2026-121720
- CVE.org — CVE-2026-115342