Linux 7.1 Drops UDP-Lite for Performance Boost
▼ Summary
– A null-pointer dereference bug in UDP-Lite, found in 2023, had persisted since 2016, indicating the protocol was likely unused for years.
– The bug was fixed and a deprecation notice was added, but no complaints were received, leading to the decision to officially retire UDP-Lite.
– Removing UDP-Lite code from the Linux kernel cleans up the networking stack and removes conditionals from the UDP fast path.
– This removal yields a performance increase of up to 10% in packets per second for standard UDP workloads on tested hardware.
– The retirement of UDP-Lite support is proceeding in the Linux kernel, following a deprecation plan announced in 2023 for removal by 2025.
The upcoming Linux 7.1 kernel will officially remove support for the UDP-Lite protocol, a move that cleans up the networking stack and delivers a measurable performance improvement for standard UDP traffic. This decision follows years of inactivity and a critical bug that went unnoticed due to a lack of real-world usage. The removal process eliminates unnecessary code paths, streamlining packet processing for a noticeable boost in efficiency.
Back in 2023, the automated testing system syzbot uncovered a significant flaw. It found a null pointer dereference bug that occurred when UDP-Lite tried to allocate socket buffer memory after the combined memory usage for both UDP and UDP-Lite surpassed a system-wide limit. The bug had existed since 2016, which strongly indicated that no practical applications were actively using UDP-Lite for at least seven years. The issue was specific to a shared memory threshold between the two protocols, meaning it would have been easily triggered if UDP-Lite saw any adoption.
Developers promptly fixed the bug and added a deprecation notice to the kernel, announcing a plan for future removal. In the time since that notice was posted, no users or developers have raised any objections or reported dependencies on the protocol. This silence confirmed the initial assessment of its obsolescence, paving the way for its complete removal in the current development cycle.
The retirement is being handled by Kuniyuki Iwashima of Google. The process involves systematically removing IPv4 and IPv6 UDP-Lite socket support before cleaning up the residual, now-dead code within the UDP stack. This cleanup is not just about removing unused features; it directly enhances performance. By stripping out numerous conditional checks for UDP-Lite from the critical fast data path, the kernel reduces overhead for standard UDP operations.
The performance impact is tangible. Benchmarks show that removing this legacy code results in a significant packets-per-second (pps) increase for UDP workloads. In one test using `udp_rr` with 20,000 concurrent flows, performance jumped by 10%, rising from 13.3 million pps to 14.7 million pps on a high-core-count AMD EPYC platform. Even with more advanced optimizations like Flow Director enabled, a solid 3% gain was observed, boosting performance from 20.1 million pps to 20.7 million pps.
The UDP-Lite protocol itself was a specialized variant of UDP introduced in the Linux 2.6.20 era. It allowed for partial checksums, enabling applications to receive packets even if parts of the data were corrupted, a feature intended for scenarios like streaming media over lossy networks. However, this niche capability never gained widespread adoption. The plan for removal was set in motion with a 2023 deprecation notice targeting 2025. With that deadline passed and the community’s lack of response, the Linux kernel is now proceeding to excise this unused component, resulting in a leaner, faster networking subsystem for all users.
(Source: Phoronix)
