Unless it was something involving unmounted writes, and depending on how you do those, even then, there's still limited damage you can do that the kernel can't be modified to simply not allow.

It was all to do with writes done by fsck to a read-only mounted partition (and not, as it should have been, an unmounted partition). Suborning unlink() would not have saved these people; preventing direct partition writes on mounted filesystems might have, but realising that this was a potential problem equalled no more and no less than discovering the bug anyway.

Peter