Jan Kara <jack@suse.cz> writes:> But there's the second race I describe making it possible> for new IO to be created after io_destroy() has waited for all IO to> finish...Can't that be solved by introducing memory barriers around the accessesto ->dead?Cheers,Jeff