@jkbs0

No Keyboard Detected!

codecave.cc/
Joined October 2015
  • Tweets157
  • Following150
  • Followers523
24Media
Simple writing is the no.1 business skill. Warren Buffett writes his shareholder letters at a 4th-grade reading level. I discovered it by accident. Here are 5 proven methods for simple writing I wish I knew earlier:
16
38
3
243
Slides and pictures from BPF track of LSFMMBPF 2022 conference: vger.kernel.org/bpfconf2022.…
1
22
46
Hello Twitter friends! In ~30 minutes, we'll host an exciting video segment by @lzbauer @jkbs0 Arthur and me about Linux's sk_lookup watch it on CF TV cloudflare.tv/event/4yveCaDT… Blog posts we'll mention: blog.cloudflare.com/how-we-b… blog.cloudflare.com/its-crow… blog.cloudflare.com/tubular-…

Making socket lookup in Linux programmable with BPF - a retrospective...

Cloudflare outgrew the Linux network stack, and fixing it took three years. Join us to discover why it took so long, what mistakes were made along the way, and why we should've used an ioctl() after...

cloudflare.tv
9
29
A conclusion to a very long saga of "fixing" BSD Sockets API: First @akajibi wrote SO_BINDTOPREFIX patch. Then @jkbs0 did more generic eBPF sk_lookup that was more flexible. Now @lzbauer augmented it with real nice userspace management component - tubular (and tubectl).
Cloudflare
@Cloudflare17 Feb 2022
Production ready eBPF, or how we fixed the BSD socket API. cfl.re/3Bw1f17

Production ready eBPF, or how we fixed the BSD socket API

We are open sourcing the production tooling we’ve built for the sk_lookup hook we contributed to the Linux kernel, called tubular

blog.cloudflare.com
3
1
21
blog.cloudflare.com/how-to-s… "How to stop running out of ephemeral ports and start to love long-lived connections" Or "Userspace Connectx() on Linux" Or "This is why we need kernel-side connectx()" Or "Why you can have tops 28K outbound UDP connections?" Code github.com/cloudflare/cloudf…

How to stop running out of ephemeral ports and start to love long-lived connections

Often programmers have assumptions that turn out, to their surprise, to be invalid. From my experience this happens a lot. Every API, technology or system can be abused beyond its limits and break in...

blog.cloudflare.com
5
27
63
If anyone wants to get started with task isolation on Linux, here’s an article I wrote, codeblueprint.co.uk/2020/05/…

Reducing jitter on Linux with task isolation

codeblueprint.co.uk
Paul Khuong ([email protected])@pkhuong3 Sep 2021
Replying to @pkhuong @chriseberly
When I cared about noise (ubench), I would also cpuset pid 1 + non-kernel threads to core 0 on startup, and explicitly opt into the remainder. Linuxers who *really really* care (@fleming_matt ) use boot-time isolcpu. I did for a while, but ran into (now fixed) kernel lockups.
4
18
1
53
Really good article about non-trivial gdb features: interrupt.memfault.com/blog/… #gdb #debugging #linux – at Marechal Deodoro, Brasil

Advanced GDB Usage

A collection of advanced GDB tips, extensions, and .gdbinit macros to speed up your debugging experience with the GNU debugger.

interrupt.memfault.com
8
33
Two magical words "Applied. Thanks" Jakub's 12 months of work is completed. Hurray! BPF Socket Dispatch feature landed in bpf-next tree! lore.kernel.org/netdev/CAADn… kernel.googlesource.com/pub/…
15
3
85
The CFP for the Networking and BPF track at LPC 2020 is out! Show us what you've got! lore.kernel.org/netdev/20200… #LinuxPlumbers @linuxplumbers #linux #bpf #netdev
1
17
34
Huge milestone on the "BPF program on a lookup socket" project. @jkbs0 just published first non-rfc patch series with the code: lore.kernel.org/netdev/20200… It's about a year of Jakubs work...
12
48
jkbs retweeted
jkbs@jkbs013 Mar 2020
Replying to @packagecloudio
@packagecloudio blogs about Monitoring and Tuning the Linux Networking Stack are legendary. But, there is one about syscalls too! Just as epic blog.packagecloud.io/eng/201… Do you know 3 methods of making a syscall on x86? What is located after envp? And what it has to do with vDSO?
5
1
18
We decided to figure out what TCP_USER_TIMEOUT and SO_KEEPALIVE socket options actually do... Ten days later I'm confident nobody really understands it (and I didn't even look at the funny states like FIN-WAIT, LAST-ACK) Code: github.com/cloudflare/cloudf… blog.cloudflare.com/when-tcp…
Cloudflare
@Cloudflare20 Sep 2019
When TCP sockets refuse to die - cfl.re/32TuOaA

When TCP sockets refuse to die

We noticed something weird - the TCP sockets which we thought should have been closed - were lingering around. We realized we don't really understand when TCP sockets are supposed to time out! We...

blog.cloudflare.com
5
57
5
143
The demo worked flawlessly! Here is a video of the same thing: youtube.com/watch?v=4sRECNvn… inet-tool: github.com/majek/inet-tool#r… kernel patches: github.com/jsitnicki/linux/c…
David Godfather Miller
@davem_dokebi9 Sep 2019
#ebpf programmable socket lookup demo! #LinuxPlumbers #lpc2019
5
32
nsenter is a pretty underrated tool for debugging container networking. If you know the pause PID, you can effectively use any tools installed on the host (not the container) against the pod’s network namespace. For example: $ nsenter -t <pause-pid> -n tcpdump -i eth0
7
34
6
166
We are pleased to announce our first accepted Microconference for 2019 Linux Plumbers Conference! Welcome BPF! linuxplumbersconf.org/blog/2…
1
17
67
New blog post: "SOCKMAP - TCP splicing of the future" blog.cloudflare.com/sockmap-… The code: github.com/cloudflare/cloudf… We had to: - write custom bpf loader, I hate dependencies - understand what SOCKMAP is supposed to do (zero docs) - debug it - wonder why splice is slow

SOCKMAP - TCP splicing of the future

cccccblgtbhdtvkdkbdfcdvgiicuckkdkruvfceiuiur

blog.cloudflare.com
2
46
129
How to check if some odd syscall is supported by your kernel? Nothing simpler! Use Perl! strace -e trace=/syscall perl -e 'printf "ret=%d err=%s\n", syscall(425), $!' Found by @jkbs0; For the record syscall 425 is io_uring_setup :)
2
6
23
IOCB_CMD_POLL usage is picking up! blog.cloudflare.com/io_submi… by @Cloudflare
7
18
My XDP talk covered in blogpost at #KR2018 super very nice write-up by @Aissn kernel-recipes.org/en/2018/l…
5
10
1/2 Joel Jaeggli from Fastly is saying *not* to use flow-label for ECMP hashing. The problem? Flow-label is created *especially* in order to be used for hashing. Arguments: - draft-fioccola-spring-flow-label-alt-mark - some firewalls rewrite flow-label blog.apnic.net/2018/01/11/ip…
1
2
3
Load more