> > Meh, new opcode would suck. Don't deviate from the standard apis then. > > Not a new opcode, it'd just be a flag for that opcode. We default to > O_CLOEXEC is nothing is given, and you can do: > > io_uring_prep_fixed_fd_install(sqe, fixed_index, IORING_FIXED_FD_NO_CLOEXEC); > > to simply set that flag to turn it off. Only reason I bring it up as a > bit annoying is that it'd be cleaner to have it be part of the O_* > namespace as O_NOCLOEXEC, but it's not a huge deal. > > It retains the part you cared about, which is making O_CLOEXEC the > default, but retains the option of turning it off rather than needing to > do an fcntl() to retrieve flags, mask it, then another fcntl(). Ok, sounds good.
