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-2026-83663 Overview
CVE-2026-83663 is an uncontrolled recursion vulnerability [CWE-674] affecting the Go bindings of Apache Thrift before version 0.25.0. A remote peer can send a crafted frame that forces the TFramedTransport or THeaderTransport reader into unbounded recursion. The Go runtime raises a fatal error when the stack limit is reached, and recover() cannot intercept this condition, so the entire process terminates. The flaw is reachable over the network without authentication or user interaction, which makes any Thrift service built on the Go bindings exposed to an availability attack.
Critical Impact
An unauthenticated remote attacker can crash any Apache Thrift Go service by sending a 4-byte or 18-byte malformed frame, causing a denial-of-service condition that recover() cannot mitigate.
Affected Products
- Apache Thrift Go bindings — all versions before 0.25.0
- Services using TFramedTransport in the Thrift Go library
- Services using THeaderTransport in the Thrift Go library
Discovery Timeline
- 2026-10-02 - CVE-2026-83663 published to NVD
- 2026-10-02 - Last updated in NVD database
Technical Details for CVE-2026-83663
Vulnerability Analysis
The vulnerability resides in how the Go bindings for Apache Thrift satisfy a Read call that is backed by a buffered frame. When the current frame yields zero payload bytes, both transports read the next frame and then recursively invoke Read instead of looping. A hostile peer can keep producing such empty-payload frames indefinitely. Each recursive call consumes additional Go stack space.
In TFramedTransport, a 4-byte frame declaring a size of zero is sufficient to trigger the recursive step. In THeaderTransport, an 18-byte frame consisting entirely of a header block (and therefore no payload bytes) achieves the same effect. Because the terminating condition depends on data the attacker controls, the recursion depth is unbounded.
Once the Go stack soft limit is exceeded, the runtime emits a fatal error: stack overflow. Unlike a normal panic, a fatal runtime error is not recoverable through defer/recover(), so the entire process exits. The impact is loss of availability for the Thrift service and any workloads sharing the process.
Root Cause
The root cause is a control-flow defect: a recursive Read call where a bounded loop should have been used. The implementation assumed that a drained frame would be followed by a frame carrying payload bytes, but a malicious peer never has to send one. This matches [CWE-674: Uncontrolled Recursion].
Attack Vector
Exploitation is purely network-based. An attacker opens a TCP connection to a Thrift Go endpoint and transmits a sequence of empty-payload frames conforming to TFramedTransport or THeaderTransport. No credentials and no prior session state are required. The server consumes stack frames with each recursive Read, hits the Go stack limit, and aborts. The vendor advisories describe the triggering frame structures in detail. See the Apache Mailing List Discussion and the Apache Mailing List Thread for the technical write-up and patch references.
// No verified exploit code is published for CVE-2026-83663.
// Refer to the Apache advisories above for frame-level details.
Detection Methods for CVE-2026-83663
Indicators of Compromise
- Thrift Go processes terminating with fatal error: stack overflow in standard error or container logs.
- Repeated short-lived TCP connections to Thrift ports that close immediately before any application-level response.
- Service supervisors (systemd, Kubernetes) reporting rapid restart loops of Thrift Go workers.
Detection Strategies
- Inspect network traffic to Thrift endpoints for streams of 4-byte frames declaring a zero size, or 18-byte THeaderTransport frames with no payload.
- Correlate client source IPs with crash events on backend Thrift Go services to identify abuse patterns.
- Build identification rules that fire when the string fatal error: stack overflow appears in logs from Go processes exposing Thrift transports.
Monitoring Recommendations
- Alert on abnormal restart rates or crash loops of Thrift Go services via container orchestrator and process supervisor metrics.
- Track inbound frame-size distributions on Thrift listeners and flag sustained traffic composed of minimum-size frames.
- Centralize Go runtime stderr output into the SIEM so stack-overflow fatal errors are preserved after the process dies.
How to Mitigate CVE-2026-83663
Immediate Actions Required
- Upgrade the Apache Thrift Go library dependency to version 0.25.0 or later across all affected services.
- Rebuild and redeploy any binaries or container images that embed vulnerable Thrift Go code; dependency updates alone do not patch running processes.
- Restrict network exposure of Thrift endpoints to trusted peers until patched builds are rolled out.
Patch Information
Apache Thrift version 0.25.0 fixes the uncontrolled recursion in both TFramedTransport and THeaderTransport by replacing the recursive Read call with a bounded loop. Consumers must update the github.com/apache/thrift module in go.mod, run go mod tidy, and rebuild. Transitive dependencies in vendored trees must also be refreshed. Review the Apache Mailing List Discussion for the authoritative fix announcement.
Workarounds
- Place Thrift Go services behind a reverse proxy or L7 gateway that enforces a minimum useful frame size and rejects malformed Thrift framing.
- Apply network ACLs or service-mesh policies that limit Thrift ports to authenticated internal clients.
- Enable aggressive connection rate limiting on Thrift listeners to reduce the attack surface while patches are staged.
# Update the Apache Thrift Go dependency to the fixed release
go get github.com/apache/thrift@v0.25.0
go mod tidy
go build ./...
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

