Linux 6.6 Will Make It Easy To Disable IO_uring System-Wide

[stiiixy]

You do your cause a disservice if you pretend there's not a very real tradeoff, here. A security-at-all-costs approach would set us back nearly a decade on energy efficiency and performance. You're basically talking about overheads like that of a fully-deterministic RTOS, all for the sake of theoretical exploits. There has to be some balance and some willingness to blaze new trails and fix problems as they come up.

What we need is a way to make more calculated tradeoffs between security and performance. One approach, at least for side-channel attacks, is to asses a system's overall vulnerability, rather than always to pessimistically address each potential vulnerability with point-solutions, even if they overlap:

http://people.csail.mit.edu/emer/med...sca.metior.pdf

Buuls be a balls to the wall bleeding edge super dooper speedy gamey kernel?

I'm trading in all my chips and going for (El Dorado?) gold!

[microcode]

Is this normal or Jens Axboe from Facebook is not doing good coding practices?

"coding practices" are not the problem, but Jens and the reviewers of his patches do bear some responsibility for not considering the security implications as deeply as turns out was necessary. As for whether this is normal or not, this is totally normal in the Linux kernel development cycle. The first round of security issues found were the ones relating to a lack of LSM hooks, which would affect those who have strict LSM policies through SELinux and the like, io_uring had handlers that lacked the hooks that normally allow SELinux policies to filter and block operations. There are other classes of bug to work through now, relating to memory and reference handling (no surprise to any C developer), and there will probably be a couple other classes of bugs after that to work through. Eventually it'll be ironed out properly.