Linux 7.3 Cleans Up Three Filesystems — FailFS Enters, FreeVxFS Exits, NULLFS Paved the Way

August 3, 2026. The Linux 7.3 merge window is open, and the kernel is about to clean house in its filesystem layer. Three drivers — for radically different reasons — illustrate how the world’s most deployed kernel handles its legacy. FailFS enters as a tongue-in-cheek testing tool, FreeVxFS exits after two decades without real maintenance, and NULLFS had already opened the door in the previous cycle. This isn’t cosmetic cleanup: it’s an architectural decision that protects millions of servers.

What the Linux 7.3 cycle says about the kernel goes far beyond the files themselves.

FailFS: the filesystem that fails on purpose

FailFS is the twin sibling of NULLFS, merged in Linux 7.2 back in July 2026. Where NULLFS does absolutely nothing — every operation succeeds with no effect — FailFS systematically fails. A successful mount, then every read, write, or stat attempt returns an error.

The code is minimal: under 300 lines, essentially stubs that return -ENOSYS or -EIO. The value isn’t functional — nobody deploys FailFS in production — but behavioral. Kernel developers use these pseudo-filesystems to validate the robustness of upper layers (VFS, page cache, memory management) against explicit failures.

The logic mirrors the chaos engineering practiced at Netflix or AWS: if your code can’t survive a filesystem that refuses everything, it won’t survive a dying disk, a timing-out NFS mount, or a segfaulting FUSE daemon. FailFS turns those scenarios into reproducible unit tests.

Michael Larabel reports at Phoronix that the proposal was warmly received by Linus Torvalds himself, who noted that “this is exactly the kind of thing people write internally and never submit.” Upstreaming FailFS makes this behavior available to all maintainers, without each having to reinvent a stub in their own corner.

FreeVxFS: twenty years of grace, zero users

At the other end of the spectrum, Linux 7.3 is set to remove FreeVxFS, the read-only driver for the Veritas VxFS filesystem historically used by HP-UX and SCO UnixWare.

The driver was introduced in the early 2000s. It never progressed beyond read-only, never supported recent VxFS format versions, and received no functional contribution since 2008. The only changes in the last fifteen years are mechanical fixes tied to VFS internal reworks — changes that affect every driver, not FreeVxFS specifically.

The removal patch, submitted by subsystem maintainer Christian Brauner, is clinical: deletion of the entire fs/freevxfs/ directory and removal of the corresponding Kconfig entries. No debate, no pushback. The silence itself is a verdict.

What’s remarkable isn’t that FreeVxFS is being removed, but that it survived this long. The Linux kernel has a culture of near-absolute backward compatibility — you don’t break userspace — but this rule applies to interfaces, not dead code. FreeVxFS wasn’t a used interface; it was a fossil riding through release cycles without justification.

The removal of FreeVxFS affects no one. If you need to read a VxFS volume in 2026, you’re probably using an emulated HP-UX 11i v3 virtual machine, not a direct mount on a recent Linux kernel. And if you’re still storing active data on VxFS, you have a more urgent problem than the Linux 7.3 release.

NULLFS: the precedent that normalized cleanup

NULLFS, merged in Linux 7.2, set the precedent. This “completely catatonic” filesystem — to quote its own documentation — does nothing, consumes zero resources, and exists solely to provide a harmless mount point for VFS layer testing.

Its acceptance into the mainline kernel sent a signal: the kernel welcomes drivers whose utility is internal to development, not just those serving a user-facing use case. It’s a subtle but important cultural shift — the kernel isn’t just a product; it’s also a development platform for its own maintainers.

FailFS is the direct extension of that logic. One does nothing and succeeds; the other does nothing and fails. Together, they cover both extremes of filesystem behavior, giving developers a standardized way to test kernel robustness.

Verdict

Linux 7.3 won’t change end users’ lives. No flashy new feature, no performance optimization that will headline benchmark charts. But it’s precisely this kind of cycle — where dead code nobody dared touch gets deleted and testing tools nobody dared submit get upstreamed — that maintains the kernel’s long-term sustainability.

If you maintain Linux systems in production, your action plan is simple: do nothing. FreeVxFS doesn’t concern you, FailFS will never be exposed to you, NULLFS is transparent. The lesson lies elsewhere: the Linux kernel does the housekeeping your own codebases deserve. When did you last delete a module nobody has used since 2008?

References

• Phoronix, “Linux 7.3 Looks Like It Will Upstream FailFS,” August 3, 2026.
• Phoronix, “Linux 7.3 Expected To Drop The FreeVxFS File-System Driver,” August 3, 2026.
• Phoronix, “NULLFS Merged For Linux 7.2,” July 2026.
• LKML, patch series “fs: remove freevxfs,” Christian Brauner, August 2026.