/me is installing OS to my new server with the heretical approach: pv disk.img | ssh root@server "cat > /dev/sda"
Remote
niconiconi
@niconiconi@mk.absturztau.be
sharkey 2025.5.2-dev+g8e80ed6771Previously: @niconiconi@cybre.space@mk.absturztau.be / Code monkey and sysadmin / No nations, no flags, no patriots. / Chaotic Neutral / Now Accelerationist / currently NEET + hikikomori / 🔐 “Onii-chan is watching you!", use OpenPGP: FAD3EB05E88E8D6D / biologically male, self-identified as '; DROP TABLE genders;
2739 Followers
182 Following
50 Posts
OpenPGP:
0x255211B2395A5A3E0E48A0F1FAD3EB05E88E8D6D
Lucky Number:
9629 · 2^943047 + 1
Public Posts:
(U)
Other Posts:
(TS//SI//NF)
Gen X, Gen Y, Gen Z, Gen Z 2×2 (20 Gbps)
The future of Unix
- Delete all system calls, keep only 1: ioctl(), because it's already used for everything like sending rendering commands to GPUs. So why not to use it for controlling the whole kernel via the file descriptor FD_UNIX (/dev/unix)?
- Instead of system calls, make ioctl() pass a message to the system's "RPC server".
- Move the RPC server to userspace "for security", and call it "systemd-calld".
- Read system call definitions from /usr/share/systemd/calls/ on startup, so you get ioctl() type safety. You can create a new system call by adding a file and run "callctl reload".
- Rewrite each driver in userspace for even better security, and name them "systemd-ext4d", "systemd-ethd", etc.
- Support all calls over D-Bus instead of ioctl() because it's what systemd uses.
- Rename LKML to Lennart's Kernel Mailing List.
- For all that security, rm -rf /* still works, and your shell still blows up when someone forgot to "${quote}" because the userspace is still POSIX.
Q: What do you call a training course on BIOS and UEFI development?
A: A startup bootcamp.
"Diskless workstation"
*look inside: uses someone else's disk.*
"Serverless application"
*look inside: uses someone else's server*
Grandpa on Twitter? The Elder Scrolls.
NFS version 3 protocol servers must be able to recover without data loss from multiple power failures (including cascading power failures, that is, several power failures in quick succession), operating system failures, and hardware failure of components other than the storage medium itself (for example, disk, nonvolatile RAM).TIL If your file server is too unstable, it's technically an RFC violation.
Boosted by @GroupNebula563@mastodon.social
"Having the same IQ as the temperature of the room" means you're a genius, all physicists know the room temperature is 300 K.
Boosted by @GroupNebula563@mastodon.social
"--repeating-payload: Use repeating pattern in payload, instead of random bytes. The same payload is used in iperf2 (ASCII ’0..9’ repeating). It might help to test and reveal problems in networking gear with hardware compression (including some WiFi access points), where iperf2 and iperf3 perform differently, just based on payload entropy."Aha, good old transparent compression cheat. Becomes "infinitely" fast when you send a stream of zeros.
Replying to
I only get 4 Gbps from my 10 Gbps server, I'll never ask "why" online, because I'm 100% guaranteed to get the silly answer of "enable Jumbo Frame". 10 GbE servers worked over Internet for a decade with no jumbo frames. Actually solving it requires testing a bunch of hypervisor, guest, routing, bridge, netfilters combinations to find the PPS bottleneck (conveniently worked around by jumbo frames).
This is ridiculous. Is there really no way to check if your OpenSSL is built with KTLS support, other than "write a small application and call BIO_get_ktls_send() / recv()" or "establish a TLS connection and check the kernel counters"?!
Smartphones be like
Past: "There's an app for that!"
Now: "There's only an app for that!"
There's a "hardcore gamer online" stereotype according to which, if you play a game in easy mode, you're not a true gamer to them. I'm now starting to wonder if this started as a Touhou Project meme. In 2007, IOSYS's Flash videos went viral online. Overdrive had a meme with Reisen and Tewi saying: easy mode, how lame! It instantly became a must-post gaming meme on 4chan for years (the IOSYS meme itself was a SFW and Touhou-fied version of 2ch's "virgin insult" meme from an eromanga). So this somewhat niche mentality was boosted by a meme reaction picture with a catchphrase.
The Japanese text was actually "初月", a difficulty level specific to Imperishable Night with its moon theme (but assigned a pronunciation 初月 イージモード in the lyrics). The naturally became "easy mode" in English, and unintentionally made it exploitable in all gaming threads. #2hu #touhou
"The problem is that for decades, people have used echo something | fdisk, so we need to be careful not to change the expected default.":woozy_baa:
Replying to
I should do another one with a quantum physics book.
Replying to
Until that happens, Everyone needs to keep using this hack:
dhclient eth0
wg-quick up wg0
mount -o ro 10.0.0.1:/my_root /mnt
cd /mnt
pivot_root . old_root
exec chroot . sh -c 'umount /old_root; exec /sbin/init' \
dev/console 2>&1
This doesn't even work well in 2026, because /sbin/init is not PID 1. Not a problem for old SysV init which is just a collection of shell scripts, but a huge problem today. You can start a PID namespace to work around it, using the existing system userspace as a do-nothing Dom0, but I hope someone else already did the hack for me in a distro.
NSA 的领导在听证会上发表慷慨激昂的讲话：“随着 Clipper 加密芯片在美国的全面推广，我们很快就能实现通话安全了！”台下突然传出一个公民的声音：“那我们怎么办？”
Twitter 被收购后的第一次全体员工大会上，马斯克激情演讲：“我们将构建一个开放的公共广场！”台下员工举手：“是开放标准，开放源码还是开放 API？”马斯克：“开放注册。”
Linux 内核邮件列表上，Nvidia 宣布：“我们将为这个新 GPU 提供完整的主线内核支持。”
一位开发者问：“驱动代码会遵守 GPL 吗？”
Nvidia 回答：“当然，不遵守 GPL 的早就放在 /lib/firmware 里了。”
在 Linux 基金会的一次大会上，主持人突然说：下面请认为自由软件好的来宾坐到会场的左边，认为私有软件好的来宾坐到会场右边。大部分人坐到了左边，少数人坐到右边，只有一个人还坐在中间不动。主持人：这位来宾，你到底认为自由软件好还是私有软件好？参会者回答：我认为自由软件好，但我以此为平台开发私有软件。主持人慌忙说：那请您赶快坐到主席台上来！
在技术峰会上，主持人问：“请问各位，你们认为什么是最好的开源许可？”
一位开发者说：“GPL，因为它保证了自由。”
另一位说：“MIT，因为它最宽容。”
第三位说：“AGPL，因为它让云厂商害怕。”
这时，一位大厂法务站起来说：“最好的许可，是别人用了开源许可，也绕不开我的额外许可。”
主持人立刻说：“麻烦您到旁边的 MPEG 分会场去讲。”
"For a start lets define some TERMS. I use UPPERCASE not to shout - but to underline a part of the sentence that is needed to be underlined - this is plain email - I do not have Bold/Underscore/Italic options - keep that in mind - now - to the core of the issue ..."Hint: Where did the original Markdown come from? 
Q: What happens if you made a groundbreaking discovery in abstract algebra?
A: You get the "Field's Medal".
// Additional sanity checks: we're about to rm -rf /, so be REALLY SURE we mean it. I could make this a CONFIG option, but I would get email from all the people who WILL destroy their filesystems.:blobcatlul:
if (stat("/init", &st) != 0 || !S_ISREG(st.st_mode)) {
bb_error_msg_and_die("/init is not a regular file");
}
NSA 的领导在听证会上发表慷慨激昂的讲话：“随着 Clipper 加密芯片的全面推广，美国很快就能实现通话安全了！”台下突然传出一个公民的声音：“那我们怎么办？”
Replying to
@ProjectPhysX@mast.hpc.social I don't think it's a scam, more like insider trading. Compute-unlock PoC exists, I have not tested it yet, but it looks definitely real. But memory, only insiders know...
Replying to
@thedarktangent@defcon.social @drwho@masto.hackers.town Even a 2011 Core i7-2600K (Sandy Bridge) can do 10 GbE at line rate over a single TCP stream on a bare-metal system, no jumbo frames needed, no problem. It becomes a problem if there's a packet-per-second bottleneck somewhere in the stack, and there are a lot of them for a VM. But all forum users only know how to enable jumbo frame.
Replying to
@ProjectPhysX@mast.hpc.social The paper said 80 GB VRAM was unlocked, but with errors, while 40 GB was error-free, this is consistent with the 40 GB A100 SKU that the CMP 170HX was based on. But on the forums, there are doubts about whether all cards have 40 GB of error-free memory (8 GB vs. 10 GB SKUs), and there's also no PoC at the moment (PoC is compute-only), so some suspect it's a GPU pump-and-dump scheme. Just wait and see.
问：Perl 的包管理器为何叫 CPAN？
答：如果你装太多软件包，你的 C 盘就满了。
问：为何曹操能“挟天子以令诸侯”？
答：因为他的名字叫 cmd。
Replying to
@jernej__s@infosec.exchange MusicBrainz distinguishes Release and Release Group. Release Group is the album, a Release is a particular version (US, EU, JP, re-release) of that. The hardware database should have "hardware" and "hardware group".
Replying to
@dramforever@mastodon.social No systemd in the boot disk, because that would be too easy. Fun fact: systemd has native support for userspace reboot: systemctl soft-reboot, which reboots the userspace and starts a new rootfs at /run/nextroot.
Enabling X25519MLKEM768 on my Web server so the NSA won't be able to decrypt my cat photos and anime memes in 2050.
Q: What happens if you exploit a FreeBSD service?
A: You go to jail(8).
Fediverse in a nutshell. (src: https://www.pixiv.net/artworks/118376737) #animeirl
/etc/passwd is only a shadow of its former self.
Replying to
5 years later a random hacker would fix it, and suddenly it becomes a recommended chipset, but only on the latest mainline Linux kernel with linux-firmware from Git, it's still broken on BSDs, Windows, OpenWrt, Xen and Proxmox.
It must be DNS.
There's no way it's not DNS.
It's Router Advertisement.
It must be RA.
There's no way it's not RA
It's the firewall.
It must be the firewall.
There's no way it's not the firewall.
It's MTU.
Spent two days troubleshooting this silly issue: https://serverfault.com/questions/1199535/lan-server-is-unreachable-via-ipv6-mtu-1500-capped-to-1492-but-ipv4-works-o/1199536
Replying to
@higuma@sakurajima.moe My rootfs is on the other side of the VPN. No standard bootloader, however powerful, would allow me to do this. The only bootloader that can do this, is known as "a userspace in initramfs".
Replying to
I've found a solution to the "init is not PID 1" problem on Alpine Linux. Now I can use a normal Alpine Linux as a bootloader! It turns out that "userspace-only reboot" is also supported by BusyBox init, after all. When you send SIGQUIT to PID 1, BusyBox shuts down all services, and relaunches a new init according to /etc/inittab's ::reboot:/target.
So, to chainload your next system from an Alpine Live CD/USB. You just need to do these steps. First, get your networking, VPN (kernel-mode only) and NFS ready. Use IPv6 SLAAC for network, so the network survives in multiple userspaces even without a running DHCP client.
/etc/init.d/networking start
wg-quick up wg0
mount -t nfs4 -o vers=4.2,sec=sys 10.10.20.1:/client1/debian_root /mnt/
Then,
echo "::restart:/sbin/switch_root /mnt /sbin/init" >> /etc/inittab
# reload inittab
kill -s SIGHUP 1
# disable sanity check
# warning: switch_root will delete everything in rootfs
touch /init
# the king is dead, long live the king
kill -s SIGQUIT 1
OpenRC shuts down the Alpine Linux, and BusyBox init replaces itself with systemd, and starts Debian in /mnt via NFS + VPN! The kernels don't match, new Debian kernel modules WILL NOT work, so you need a "double jump" with a kexec step to reload the kernels. This is currently in development, I'll write a blog post soon.
Replying to
The definition of science is doing the same thing over and over again and expecting a different publishable p < 0.05.
问：什么叫做“把权力关进制度的笼子里”？
答：电脑玩家就有一个外行人看来很奇怪的做法。他们平时绝不主动使用“系统管理员”账号，只使用普通账号。但是他们自己手里却有“系统管理员”密码，安装软件时输入密码，就摇身变为了“管理员”。一方面，他们原本就拥有整台电脑，自然可以随意操作。但“系统管理员”权力大，日常办公、上网、打游戏时，因误操作运行恶意软件或者删除系统文件，系统就轻易破坏了，所以需要“自己给自己设密码”，防止误操作。而“把权力关进制度的笼子”的原理也是如此：法律和机关本身都是被统治集团掌控的，权力自然可以滥用。但是频繁滥用权力容易扰乱社会秩序，一个贪官就能制造很多麻烦。因此，当统治集团觉得没必要滥用权力时，就会走进笼子把自己锁起来，表面上遵守规则。当统治集团想要滥用权力时，又会自己用钥匙打开笼子走出来。
Replying to
@brouhaha@mastodon.social I don't think they're instructions. Probably they're setup scripts.
The community should create a MusicBrainz for computer peripherals. You buy a product, add its barcode, PCB photos, revisions, part numbers, VID/PID to the database. How many times have you purchased multiple USB Hubs or card readers due to stability issues, before realizing that they are powered by the same controller... As a bonus, accept packaging photos to archive their artworks as well.
Q: How did Wolfenstein 3D perform collision detection?
A: With axis-aligned bounding boxes.
Q: How do computer display makers ensure their products don't have too many dead pixels?
A: They screen the panels.
Every PC hardware reviewer complains about "SSD is running out of pSLC cache" after writing a few GiBs of data. But it's a pointless exercise unless you are write-heavy (e.g. when you're a YouTuber with 4K videos). Meanwhile, nobody talks about low-depth IOPS. Where are the Windows Update, tar -xf chromium.tar.gz, and npm install simulations? Better, try resilvering a super fragmented RAID-Z2 using an old version of OpenZFS without the sequential resilvering patch.
Q: What do you call a boy who's calculating the tensile stresses on a bridge using a computer?
A: FEMboy.
The Internet be like: there is only one protocol in the world, it's called TLS. There is only one port in the world, it's called 443. Just run everything on port 443.