Six Years, 360 Patches: Linux Finally Buries a Dangerous String Function

After six years of sustained effort and 360 separate patches, the Linux kernel has officially removed the strncpy API. The C string-copying function has existed since the earliest days of the language and carries a subtle but dangerous flaw: when the source string is longer than the destination buffer, strncpy writes up to n bytes but does not guarantee null termination. Code that subsequently reads from that buffer can walk past the end of valid memory, a pattern that has produced buffer overflow vulnerabilities for decades.

The replacement function, strscpy, guarantees null termination and returns an error code when truncation occurs, making the dangerous case visible rather than silent. The coordination required to get there was extraordinary. The Linux kernel contains somewhere between 35 and 40 million lines of code, and systematically locating every strncpy call, understanding its context, writing a safe replacement, and shepherding each change through review across dozens of maintainers over six years represents an immense accumulation of unglamorous work.

One subtlety complicated the effort: some uses of strncpy were intentional. Certain binary data structures genuinely wanted a fixed-size field without null termination, and those cases required not mechanical substitution but actual reasoning about intent. Kernel developers in the Hacker News discussion noted that this kind of systematic cleanup is what distinguishes actively maintained infrastructure from legacy codebases that accumulate risk invisibly over time. Critics, however, raised the Rust-in-Linux project as a counterpoint — arguing that language-level memory safety guarantees may ultimately do more than any API hygiene campaign for the C code that remains.

Elsewhere in low-level systems, a comparison of epoll and io_uring drew attention on Hacker News. Epoll has powered event-driven network programming on Linux for roughly twenty years, letting programs monitor many file descriptors and receive notifications when any become ready. io_uring, introduced in Linux 5.1 in 2019, uses shared ring buffers between the kernel and user space so that the common path requires zero system calls, eliminating significant overhead in I/O-intensive applications. Benchmark results favored io_uring under high concurrency with many small I/O operations, though commenters noted that epoll's two decades of battle-hardening remains a practical argument for most production deployments. A separate post from zeux.io demonstrated AVX-512 SIMD optimization for zigzag decoding — processing 64 bytes simultaneously to accelerate deserialization of Protocol Buffer-encoded data — showing the full reasoning behind cases where compilers fail to auto-vectorize.