> Of course, and you'll be CC'ed on all the BPF token patches I will > resend after the holidays. > > And just to be clear for the future, by "core fs semantics" you also > mean any BPF UAPI FD field, right? Yes, because ultimately you end up with calling: fdget()/fdget_raw()/fget() to turn a userspace handle in the form of an fd and turn it into a struct file. And that is uniform across the kernel. And therein lies the beauty of it all imo. IMHO, a file descriptor is one of the most widely used generic abstraction we have across all of the kernel. It is almost literally used everywhere. And everyone has the same contract: a non-negative integer is a valid fd, a negative one is invalid. It's simple, there aren't corner cases, there aren't custom semantics. And it's also arguably one of the most successful ones as we keep implementing new apis on top of this abstraction (pidfd, seccomp, process_*(), memfd_*(), endless kvm ioctls etc etc).
