Disclaimer: This content was generated using AI. While we strive for accuracy, please verify critical information with official sources.
Sources: National Vulnerability Database, CISA KEV, Exploit-DB
CVE-2026-103649 Overview
CVE-2026-103649 is a denial-of-service vulnerability affecting the Linux builds of Progressive Robot hMailServer versions 6.3.0 through 6.3.5. The server applies socket timeouts in a Windows-specific format that Linux rejects, and its HTTPS clients perform reads without a deadline. A remote peer that accepts a TCP connection and then sends no data holds the waiting thread for the entire duration of the connection. Because the MTA-STS policy fetch runs during outbound delivery by default, any attacker who can induce the server to deliver mail to a domain they control can exhaust delivery threads and halt outbound mail. The flaw also affects DANE TLSA queries, OAuth2 token requests, the ACME client, and the ManageSieve and metrics listeners. Windows builds are not affected.
Critical Impact
A remote unauthenticated attacker can stop outbound mail delivery and disable auxiliary listeners by holding server threads indefinitely with silent TCP peers.
Affected Products
- Progressive Robot hMailServer 6.3.0 (Linux builds)
- Progressive Robot hMailServer 6.3.1 through 6.3.4 (Linux builds)
- Progressive Robot hMailServer 6.3.5 (Linux builds)
Discovery Timeline
- 2026-10-08 - CVE-2026-103649 published to NVD
- 2026-10-08 - Last updated in NVD database
Technical Details for CVE-2026-103649
Vulnerability Analysis
The root defect is a missing-network-timeout condition classified under [CWE-1088]. hMailServer sets socket timeouts using a structure layout that the Windows Winsock stack accepts, but the Linux kernel refuses the call and leaves sockets without any read deadline. Separately, the HTTPS client code paths read response data without applying an application-layer timeout. A remote peer that completes the TCP handshake and then stays silent causes the calling thread to block for as long as the connection remains open.
Several outbound and listening subsystems are affected. The MTA-STS policy fetch, enabled by default, contacts mta-sts.<recipient domain> during outbound delivery. The DANE TLSA query, the OAuth2 token request, the ACME client, the ManageSieve service, and the metrics listener share the same timeout flaw. Repeated silent connections exhaust worker threads and stop the server from processing legitimate work.
Root Cause
The root cause is platform-incorrect use of setsockopt timeout structures combined with HTTPS read calls that lack any deadline. Linux silently rejects the Windows-format option, so sockets inherit the default of no timeout and block indefinitely on peers that never send data.
Attack Vector
An unauthenticated remote attacker triggers outbound delivery to a domain they control, for example by sending a message that bounces so the server attempts a Delivery Status Notification. The server resolves mta-sts.attacker.tld, opens the HTTPS connection, and the attacker's listener accepts but sends nothing. The delivery thread blocks until worker exhaustion halts outbound mail. The ManageSieve and metrics listeners are reachable directly on the network and can be stalled without triggering a delivery flow. See the GitLab hMailServer Work Item #53 for upstream analysis.
Detection Methods for CVE-2026-103649
Indicators of Compromise
- Growing number of long-lived outbound TCP connections from the hMailServer host to mta-sts.* subdomains or recipient-controlled hosts on port 443.
- Delivery queues that stop draining while the service remains running and consumes no CPU.
- Stalled ManageSieve or metrics listener threads that fail to accept new client connections.
Detection Strategies
- Monitor hMailServer worker-thread counts and alert when active threads approach the configured pool maximum without matching message throughput.
- Correlate outbound connections to MTA-STS, DANE, OAuth2, and ACME endpoints against expected response times, flagging sessions that remain open beyond normal TLS handshake and HTTP response windows.
- Track bounce-message volume targeting external domains, since attacker-induced bounces are the primary trigger for the MTA-STS fetch path.
Monitoring Recommendations
- Log socket state (ESTABLISHED connections with no bytes received) on the mail server using ss or netstat snapshots at short intervals.
- Alert on sustained zero-throughput states for the SMTP outbound queue and the ManageSieve and metrics listeners.
- Capture strace or eBPF traces of blocked recv calls originating from hMailServer worker threads during incidents.
How to Mitigate CVE-2026-103649
Immediate Actions Required
- Upgrade affected Linux installations to hMailServer 6.3.6 or later, which corrects the socket timeout handling and adds HTTPS read deadlines.
- Restrict network reachability of the ManageSieve and metrics listeners to trusted management networks until the patch is applied.
- Disable the MTA-STS policy fetch where operationally acceptable to remove the default outbound trigger used by this attack.
Patch Information
Progressive Robot released hMailServer 6.3.6 to address CVE-2026-103649. The release corrects the Linux socket timeout structure usage and introduces explicit read deadlines on HTTPS client operations. Refer to the GitLab hMailServer Release v6.3.6 notes for upgrade guidance.
Workarounds
- Place the mail server behind an egress proxy that enforces connection and read timeouts on outbound HTTPS to third-party MTA-STS, OAuth2, and ACME endpoints.
- Firewall the ManageSieve and metrics ports so only authenticated internal clients can establish TCP connections.
- Lower the maximum delivery thread count and add out-of-band queue depth monitoring to surface thread exhaustion before it halts delivery.
# Example: restrict ManageSieve and metrics listeners to localhost via iptables
iptables -A INPUT -p tcp --dport 4190 ! -s 127.0.0.1 -j DROP
iptables -A INPUT -p tcp --dport 9000 ! -s 127.0.0.1 -j DROP
Disclaimer: This content was generated using AI. While we strive for accuracy, please verify critical information with official sources.
Sources: National Vulnerability Database, CISA KEV, Exploit-DB

