IO_uring Enjoys Hybrid IO Polling & Ring Resizing With Linux 6.13

[phoronix]

Phoronix: IO_uring Enjoys Hybrid IO Polling & Ring Resizing With Linux 6.13

Merged last week back toward the start of the Linux 6.13 merge window were a number of interesting IO_uring enhancements for this first major kernel version of 2025...

https://www.phoronix.com/news/Linux-6.13-IO_uring

[uid313]

But is IO_uring still insecure?

Maybe IO_uring would be a good candidate for something to rewrite in Rust.

[jonkoops]

But is IO_uring still insecure?

Define 'insecure', that is a very broad definition.

[uid313]

Define 'insecure', that is a very broad definition.

I don't know but it does have a reputation for being insecure. Google even banned it from Android and ChromeOS because it was deemed to be just way too insecure.

[jonkoops]

I don't know but it does have a reputation for being insecure. Google even banned it from Android and ChromeOS because it was deemed to be just way too insecure.

I guess you are referring to this previous article, which seems to adequately describe the security concerns. From reading the Google blog post it sounds mostly like they are waiting for this code to be vetted, but are not inherently against its design. That said, it does sound like something that could benefit from a safe language, but perhaps its concerns are so low level you cannot escape a degree of unsafe code.

[JEBjames]

Michael

Typo

"sizing the the busiest". I think you meant "sizing for the busiest"

[Developer12]

I guess you are referring to this previous article, which seems to adequately describe the security concerns. From reading the Google blog post it sounds mostly like they are waiting for this code to be vetted, but are not inherently against its design. That said, it does sound like something that could benefit from a safe language, but perhaps its concerns are so low level you cannot escape a degree of unsafe code.

I doubt there's anything low level about it at all. It's just exchanging messages in a ring in shared memory.

What might be tricky is coming up with ownership models that make sense, since technically at any moment somebody else could write into your buffer. This is also probably where a lot of the security issues come from. In that sort of environment it's easy to make implicit assumptions that something is static and unchanging....just in time for it to be partially overwritten.