Among the changes expected for Linux 7.3 is the removal of a piece of history: the EFS file-system driver, left unmaintained for more than twenty years. Originally used on SGI IRIX systems for non-ISO9660 CD-ROMs and disk partitions before IRIX 6.0 switched to XFS, EFS made its way into the Linux kernel as a read-only module with no active maintainer for over two decades.

The decision, while long overdue, matters for anyone managing infrastructure that relies on kernel components: every line of unmaintained code is a potential security or stability risk. In EFS's case, its almost complete lack of modern usage makes the removal uncontroversial, but it serves as a reminder of the importance of system software maintenance, especially in environments that heavily use containers and orchestration, often running very recent kernels.

Technically, EFS (Extent File System) was a 64-bit file system with some advanced features for its time, such as extent-based allocation, but it was quickly eclipsed by XFS, which offered better scalability and performance and remains actively maintained to this day. The Linux EFS driver was read-only, so the risk of data corruption was limited, but its presence in the kernel represented technical debt that is now being paid off.

The removal of EFS is not an isolated event; the Linux kernel periodically expunges legacy drivers, a process that accelerates as release cycles shorten. For those managing bare-metal servers for LLMs, this can mean more predictable updates and less need to test unused code paths.

What this means for on-premise deployment evaluators

For sysadmins overseeing on-premise infrastructure, cleaning obsolete code from the kernel is generally a positive sign. A leaner kernel reduces the attack surface and simplifies testing. Those using modern Linux distributions to host AI workloads—such as LLM inference servers or high-performance storage—will indirectly benefit from this tidy-up. No one today would mount an EFS CD-ROM on an AI server, but the removal signals healthy maintenance at the core of the operating system.

The choice to delete unused code also touches a nerve in the open-source "technical debt" debate: some argue that keeping everything guarantees infinite backward compatibility, while others see such drivers as unvetted burdens. The Linux kernel's approach is pragmatic: if nobody steps up to maintain a component for decades, it's probably no longer needed. For production deployments, especially in regulated contexts where compliance and certification are critical (healthcare, defense, finance), knowing that the underlying OS actively removes unmaintained code can be an asset, not a worry.

The EFS story reminds us that IT infrastructures, even the most modern ones for on-premise LLMs, rest on foundations shaped by choices made thirty years ago. The ability of an ecosystem to clean house is a mark of maturity.

Article with original AI-RADAR analysis.