Linux 7.3
what really changes for a server

What was released, and when

Linux 7.3 was released on October 18, 2026 by Linus Torvalds, and the first stable patched version, 7.3.1, has been available since October 30, 2026. The 6.18 branch remains maintained as longterm that is the one we keep on servers that do not need new features.

The cycle is a record: approximately 18,800 commits non-merge, 3,021 developers of whom 807 were first-time contributors. This is not anecdotal information: a kernel that attracts this many new contributors remains capable of fixing things fast.

What interests us in operations

On a server, a new feature only matters if it fixes a problem you already have. Four points in this version fix one.

• Cgroup scheduling with sched_ext the ability to attach a BPF scheduler to a cgroup is now complete. Concretely: you can give a container a different scheduler from the rest of the machine, without patching the kernel. For a voice gateway or a real-time service placed next to batch processing, this is exactly what was missing.
• failfs and fchroot(2) a filesystem where all operations fail, and a system call that changes a process root after the fact. Two sandboxing building blocks: you can neutralize a path before exposing a service, rather than relying on correct application configuration.
• RWF_DONTCACHE and btrfs / ext4 performance an alternative to Direct I/O for write-heavy workloads (backups, databases, caches), and gains on the two filesystems we use most. On backup servers, this translates to minutes of nightly window.
• SMB server and Time Machine extensions the kernel SMB server now understands Apple extensions. A backup share for a fleet of Macs becomes viable without going through a third-party service.

The rest, better VRAM management for entry-level GPUs, more expressive binfmt_misc, page table tracking for virtual machines, faster io_uring for FUSE, mainly concerns workstations and development environments.

What we do with it

Nothing, right away. Our rule is simple: a kernel does not move on a production server until the stable distribution version has integrated it, or until a measured problem requires it. 7.3.1 just came out; we observe, we test on a bench, and we will deploy when the gain is proven, not because the number is higher.

On admin workstations and lab machines, on the other hand, this is where these new features get tested risk-free. That is where sched_ext and failfs will be evaluated first.