All you never wanted to know about creating fast socket I/O on #Linux. Unless you are a real geek.

A small, ultra-high performance publish-subscribe server doing kind of TCP #multicast using #sendfile, #io_uring, #splice, (mapped) files, and #fallocate for file hole punching.

Saturating a 10 Gbps link with useful data with just 8 cores.

A beautiful writeup for anyone into fast Linux networking. Not just for #ATproto

h/t @nohillside
#PubSub #Kernel #Hacking
https://www.asayers.com/jetrelay

https://blog.gslin.org/archives/2025/05/08/12389/postgresql-18-%e7%9a%84-async-i-o/

PostgreSQL 18 的 async i/o

#async #database #i/o #input #io #io_uring #linux #oltp #output #performance #pg #pgsql #postgresql #rdbms #read #speed #stream

Rootkits invisibles: el nuevo desafío para las herramientas de seguridad en Linux #io_uring #seguridad #linux https://blog.desdelinux.net/rootkits-invisibles-el-nuevo-desafio-para-las-herramientas-de-seguridad-en-linux/

Linux: Kritische Sicherheitslücke in Kernel-API entdeckt
https://winfuture.de/news,150557.html #Linux #LinuxKernel #Sicherheitslücke #io_uring

#Linux '#io_uring' security blindspot allows stealthy #rootkit attacks

https://www.bleepingcomputer.com/news/security/linux-io-uring-security-blindspot-allows-stealthy-rootkit-attacks/

#cybersecurity #FOSS #malware

Do any operating systems other than #Linux (#Windows, #macOS, #FreeBSD, #OpenBSD, etc) have an API for non-blocking file IO?

I know Linux has that in #io_uring, which can do almost any IO operation (even fsync) in the background and tell you when it's done, but is that the only OS with such a feature?

#programming #async

Does anyone happen to have any information on the relevance of UIO_FASTIOV (defined as 8 on #Linux). I understand that submitting vectored io with within this limit can avoid dynamic memory allocation in the kernel, but is this relevant in any way in practice? Is it relevant with #io_uring ?

Support for #io_uring communication between #kernel and [#fuse] userspace using IORING_OP_URING_CMD is the highlight from the main fuse (Filesystem in Userspace) updates merged for #Linux 6.14:

https://git.kernel.org/torvalds/c/92cc9acff7194b1b9db078901f2a83182bb73202

Miklos furthermore writes:

'"Following features enable gains in performance compared to the regular interface:

- Allow processing multiple requests with less syscall overhead

- Combine commit of old and fetch of new fuse request

- CPU/NUMA affinity of queues"'

#LinuxKernel

My favorite talk from #38c3 so far: Harald Welte on #io_uring https://media.ccc.de/v/38c3-iouring-ebpf-xdp-and-afxdp#t=3412https://media.ccc.de/v/38c3-iouring-ebpf-xdp-and-afxdp#t=3412

Just watching the #38c3 #io_uring
talk. That's the #linux subsystem I recently fixed bugs around incorrect #CPU affinities. https://events.ccc.de/congress/2024/hub/en/event/iouring-ebpf-xdp-and-afxdp/

🌘 io_uring 中的程序創建 [LWN.net]
➤ Io_uring 補丁系列更新，加入新的程序創建操作。
✤ https://lwn.net/SubscriberLink/1002371/0ff2be6a2c7624ca/
在 Linux 內核中，io_uring 通常用於異步操作，最新的補丁系列使其能夠創建新程序，雖然目前仍存在限制。
+ 這項新功能或許將為 Linux 內核帶來更大的彈性和廣泛的應用。
+ 聚焦於程序創建操作對於內核功能的進一步擴展是一個值得期待的方向。
#Linux內核 #io_uring #程序創建 #系統調用 #異步操作

as i am currently working on making #varnishcache client side io #async'hronous with #io_uring, i must say that https://lucumr.pocoo.org/2024/11/18/threads-beat-async-await/ made me think if we need more facilities for forensic analysis…

This is probably a ridiculous question: is #io_uring an event loop or does it resemble one?

And is it possible to make all #linux #kernel subsystems use it? E.g. GPU drivers

If so, would it be possible to write something like a #microkernel with good performance where internally subsystems submit things to #io_uring?

🌘 異步 Rust 在 io_uring 中並不安全
➤ 異步 Rust 開發者在使用 io_uring 時需留意 TCP 連線洩漏的問題
✤ https://tonbo.io/blog/async-rust-is-not-safe-with-io-uring
文章探討使用 io_uring 驅動遇到的 TCP 連線洩漏問題，揭示了這個在各種 io_uring 運行時都普遍存在的問題。
+ 對於 Rust 開發中的新挑戰有了更深入的瞭解。
+ 瞭解了 io_uring 可能存在的問題後，如何改進以確保程式碼的安全性。
#異步 Rust #io_uring

using #io_uring for storage in #varnishcache

#allsystemsgo #ASG2024

https://media.ccc.de/v/all-systems-go-2024-305-using-iouring-for-storage

From: @mediacccde
https://botsin.space/@mediacccde/113203506819669014

A new way to perform syscalls in the linux kernel, both batched and asynchronous : https://matklad.github.io/2024/09/32/-what-is-io-uring.html

There is something I don't understand though : if it's a ringbuffer, it means that some syscalls get discarded if the buffer gets full? 🤔 Or maybe it refuses more calls then, but then why make it a ringbuffer rather than a classic FIFO?

#io_uring #linux
(also, why is this article timestamp in the future, on Oct 2, 2024? ^^ )

Anyone familiar with #io_uring have any idea what this bit in mini_liburing.h is doing? I think I understand the rest of io_uring_submit(), but I can't fathom what this bit is doing.

if (*sq->khead != *sq->ktail) {
submitted = *sq->kring_entries;
goto submit;
}

(Also, isn't the IORING_ENTER_GETEVENTS flag at the end no-op, since min_complete=0 ?)

babysteps, but something is happening #io_uring #varnishcache

the #io_uring example/proxy.c code really is super helpful. thank you, @axboe !

@andreclaassen @derralf in dem bereich, in dem ich zu hause bin, ist es aber definitiv falsch, hier kommen quasi alle software innovationen zuerst als #foss. einige beispiele: bsd jails als blaupause für solaris zonen und linux container, das dateisystem #zfs, #varnishcache mit vcl als #edgecomputing oder aktueller das #io_uring kernel interface, das nun microsoft für windows kopiert hat.
🧵