Linux kernel 6.9 is here, with many under-the-covers improvements that won't be very visible to users, but which tidy things up, fix bugs, and pave the way for future changes.

Linus Torvalds' release announcement summarizes what's new, with the man himself commenting that "last week has looked quite stable (and the whole release has felt pretty normal)". We looked at some of the upcoming changes back in March when the first release candidate came out. There are many more changes than we have room to go into, but here are some of the highlights.

There are tweaks and refinements to various filesystems. Some FUSE filesystems just filter the view they present of an underlying filesystem, and for those, the new FUSE passthrough mechanism will make them faster.

A more general change that's been in development for more than a decade is the new dm-vdo tooling, which has finally been merged. Developer Permabit was working on its storage de-duplication tech a decade ago, which it named Virtual Data Optimizer or VDO for short.

Red Hat acquired the company in 2017, open-sourced the tools, and now it's made it into the kernel. It brings some of the advanced functionality offered by ZFS right into the kernel and makes it available for all filesystems: not just inline block-level de-duplication, but also compression, zero-block elimination, and thin provisioning. The user-space tools are up on Github and contain some more info. This is a big patch - nearly 55 thousand lines of code - but the potential is rather impressive.

New support for Intel's FRED will make the switches between CPU protection rings quicker. The code patch summarizes the benefits, but itself is quite long. Over on the AMD side of the tracks, 6.9 supports AMD's Secure Nested Paging [PDF].

Developers can now access not just processes but individual threads using process file descriptors, or PIDFDs for short, and there's a new pseudo-filesystem for them called pidfdfs. This is invisible from user space, though - it's only for kernel developers, but it simplifies some things.

In-kernel programs written using the kernel's eBPF functionality can now communicate via a new shared-memory region called the BPF arena.

Kernel support for Rust upgraded to version 1.76.0, released in early February, and in-kernel Rust can now be used on Arm64 - but only if you compile it with Clang. Normally the kernel is built with GCC, and here, kernel 6.9 can use GCC's named address space functionality - which is used on some embedded processors, to help handle their differently-addressable types of memory. (If you're keen to know more, read the ISO's Draft Technical Report 18037, available as a zipped PDF. Ideal bedtime reading.)

Users of the latest Intel "Meteor Lake" CPUs and the slightly older AMD Epyc 4 ones should see some performance improvements, and slightly better power management for other recent CPU models as well. Torvalds noted that there will be more attention given to Arm64 in general:

Hurrah for new toys, eh? ®

https://www.theregister.com//2024/05/13/linux_kernel_69_released/