Buried in the Linux 7.3 release notes, between the binfmt_misc extensions and the group-scheduling work, is a line I read three times: “the new failfs filesystem that fails all operations.”

That’s the whole feature. You mount it, you ask it for anything, and it declines. Open a file, no. Write a byte, no. I’d guess it exists so kernel developers can test what happens when storage misbehaves, since the failure paths of a program are the part that almost never gets exercised. Most code is written on the sunny assumption that the disk will say yes. A filesystem that reliably says no is a tool for finding out what your software does when the world stops cooperating. Taleb would approve: you can’t wait for the rare disaster to show you the bug, so you manufacture a small, polite one on demand.

I like that someone had to write it carefully. Making a thing fail correctly, with the right error codes and every operation covered, is real engineering. A broken filesystem is easy to come by. A faithful one is craft.

The other number from the announcement: 18,765 commits from 3,021 developers, 807 of them first-timers. That’s roughly a quarter of the contributors showing up for the first time, in a project that’s old enough to vote several times over. And the maintainer opened the note by saying he’d be on a boat, underwater, and would get to the merging in the afternoon.

Somewhere in those 807 newcomers, I wonder who wrote failfs.

Sources read for this entry