OpenVPN/openvpn v2.7.7

on GitHub

Security fixes:

• reliability layer: avoid unbounded reliable TLS timeout, and ignore acks
  for packets that cannot be outstanding (CVE-2026-84732)

  Both reliability layer bugs found by Mark Bregman (Fox-IT)

• windows: fix CreateProcess() command line quoting for characters that
  are special to cmd.exe, where a combination of validation script plus
  rogue CA could lead to misbehavior (CVE-2026-84256)

  Bug found by Clouditera Security

• windows: fix tapctl to always call netsh.exe with full path
  (as we do elsewhere) (CVE-2026-84226)

  Bug found by BreachX Zero Day Labs (using Typhon AI Mil v2)

• windows: don't use NULL DACL with system objects, namely the --service
  exit event and the netsh.exe guard semaphore.

  The old approach was prone to a local DoS where one user could interfere
  with other users' openvpn processes by blocking the netsh semaphore or
  sending events. This only affects setups not using the iservice, or
  using the automatic service to start/stop openvpn (CVE-2026-82312)

  Bug found by DEBRAJ BASAK

• openvpnserv (windows): pass correct NRPT domains size - when IDN domains
  with UTF8 encoding were involved, a buffer overread could be achieved
  (CVE-2026-78221)

  Bug found by BreachX Zero Day Labs (using Typhon AI Mil v2)

• openvpnserv (windows): don't allow '/' in config paths.

  The APIs windows uses for path validation do not handle '/' as path
  separator, while the file open APIs do, so this could be used to
  circumvent our config path validation, leading to openvpn.exe starting
  a user-controlled config file even if administratively not allowed
  (CVE-2026-78043)

  Bug found by BreachX Zero Day Labs (using Typhon AI Mil v2)

• dhcp (windows): fix off-by-one in write_dhcp_search_str() temp buffer
  guard - suitable DHCP options could lead to a single-byte overflow of a
  temp buffer (CVE-2026-81738)

  Bug found by Andre Kropp (Nexory) and ChinhNguyen

• linux netlink: validate netlink replies against the request

  Suggested by Joshua Rogers as a security improvement

• openvpnserv (windows): fix off-by-one on input validation
  (discovered while fixing CVE-2026-78221)

• openvpnserv (windows): harden CheckConfigPath() a bit more
  (another improvement while working on CVE-2026-78043)

User-visible Changes:

• when using EPOCH data channel format, reduce the number of future keys
  from 16 to 4 - the previous calculation was wrong, and 4 spare keys are
  sufficient for 100+ Gbit/s links. This means less log spam in userland
  and fewer resources used in in-kernel implementations.

Bugfixes:

• work around a pubkey-handling bug in mbedTLS 4.1.0 and 4.2.0
  (supposedly fixed in 4.3.0)

• multi: don't let stale-routes-check delete permanent routes -
  --stale-routes-check did not only delete dynamic cached routes, but
  also routes installed by --iroute and --ifconfig-push. Fixed by
  introducing route flags and restraining the check on them
  (Github: OpenVPN/openvpn#1063)

• ssl: do not queue control ciphertext while a packet is still queued
  (fixes problems in TCP p2p handshake when both sides try to handshake
  at the same time)
  (Github: OpenVPN/openvpn#1089)

• reenable xmit_hold when using p2p tcp-server and tls-server - in TCP
  server mode the server is not expected to initiate the TLS handshake.
  This was introduced by the multisocket code checking the wrong variable
  for socket protocol
  (Github: OpenVPN/openvpn#1089)

• clinat: do not adjust UDP checksum if zero (as per RFC768)
  (Github: OpenVPN/openvpn#1037)

• openssl: avoid resetting the HMAC key on every packet
  (Github: OpenVPN/openvpn#1088)

• openvpnserv (windows): fix log lines format string - interface names with
  international characters printed in some error messages need to be
  converted from UTF8 to UCS16 first.

• fix format string specifier for size_t (%zu)

• fix test_misc compile issues with -Werror

Full Changelog: v2.7.6...v2.7.7