ivan

From 65a0e76b06757613c0156ee24fae2646b846e494 Mon Sep 17 00:00:00 2001
From: Ivan Kozik <ivan@ludios.org>
Date: Sat, 7 Nov 2026 10:31:38 +0000
Subject: [PATCH] btrfs-progs: scrub: keep device limits unless --limit is
 given

[BUG]
A `btrfs scrub start` or `btrfs scrub resume` without --limit runs
unthrottled even when the devices have a throughput limit:

ivan

From 782cfff08973ded7ea655c58513ebd2c0f428325 Mon Sep 17 00:00:00 2001
From: Ivan Kozik <ivan@ludios.org>
Date: Sat, 24 Oct 2026 07:24:02 +0000
Subject: [PATCH] nix-serve-ng: don't flush after every NAR piece

Nix's NAR dumper hands over many tiny pieces per file (names, lengths,
padding), and nix-serve-ng flushed after each one, so Warp sent each as
its own packet. Over a ~190 ms RTT link, small-file-heavy parts of a
NAR then dropped from ~11 MB/s to ~1.4 MB/s, stalling remote Hydra
builders on "Sending inputs" for minutes.

ivan

#!/usr/bin/env bash
# Model-output: Claude Opus 5.5
#
# Read-only triage for CVE-2026-111497 (brotli decoder out-of-bounds write, fixed upstream in
# google/brotli@0ed4268a): which running or installed binaries carry an unfixed decoder, and
# whether anything crashed the way a failed exploit attempt would.
#
# Usage: ssh root@HOST bash -s -- [--closure] [SINCE] < check-brotli-cve.sh
#   --closure  also scan the /run/current-system closure (about 1 minute per 2000 store paths)
#   SINCE      start of the crash search; default 2026-02-27, when the upstream fix went public

ivan

Codex found a kernel edge case where dedupe does update mtime and ctime.

What goes wrong:

• btrfs dedupes in 16 MiB chunks. If the file being kept has a hole (a sparse, never-written region) at the end of a chunk, btrfs punches a matching hole in the other file. It does this through a helper call that doesn't tell the helper this is a dedupe (reflink.c:660-661, the NULL argument).
• That helper works from a small metadata budget. If removing the other file's extents uses it up, the helper goes round a retry loop (file.c:2412, block-rsv.c:94-101).
• The retry loop sets mtime and ctime to "now" unless it was told this is a dedupe (file.c:2496-2498). The comment right above that line says dedupe is supposed t

ivan

After making changes

Automatically commit your changes with this commit template:

subsystem: short one-line description; semicolon if multiple changes

Model-output: model name e.g. Claude Fable 5

ivan

diff --git a/nixos/modules/system/boot/loader/grub/grub.nix b/nixos/modules/system/boot/loader/grub/grub.nix
index 042fd63cd254..6d319a3d3ca3 100644
--- a/nixos/modules/system/boot/loader/grub/grub.nix
+++ b/nixos/modules/system/boot/loader/grub/grub.nix
@@ -90,6 +90,9 @@ let
           else
             args.efiBootloaderId;
         timeout = if config.boot.loader.timeout == null then -1 else config.boot.loader.timeout;
+        dynamicLimit               = cfg.dynamicConfigurationLimit.enable;
+        dynamicLimitMarginMB       = cfg.dynamicConfigurationLimit.marginMB;

ivan

#!/usr/bin/env python3

# Model-output: ChatGPT 5.6 Sol (High)

"""Patch The Witness stair movement and bypass its startup corruption guard.

The movement patch rescales a player-only walk-manifold request when traversal adds
vertical displacement. The startup patch removes the conditional jump that routes
a zero critical-asset sentinel to ``fatal_error_data_corruption``. The PE file size
and section count remain unchanged.

ivan

// ==UserScript==
// @name         leave my text alone (chatgpt.com)
// @namespace    https://chatgpt.com/
// @version      1.0.0
// @description  Keep every chat turn mounted in the DOM (defeats IntersectionObserver-driven unloading) so find-in-page and SingleFile capture whole conversations.
// @match        https://chatgpt.com/*
// @match        https://chat.openai.com/*
// @run-at       document-start
// @grant        none
// @inject-into  page

ivan

// ==UserScript==
// @name         leave my text alone (claude.ai)
// @namespace    https://claude.ai/
// @version      1.0.0
// @description  Keep every chat message mounted in the DOM (defeats list virtualization) so find-in-page and SingleFile capture whole conversations.
// @match        https://claude.ai/*
// @run-at       document-idle
// @grant        none
// @inject-into  page
// @noframes

ivan

rsync -av -@-1 as the baseline thing
or with --delete to remove things from DEST, and probably
--delete-excluded when using --exclude
rsync -@-1 compares timestamps by nanoseconds instead of seconds, which is important for transferring e.g. PostgreSQL data; rsync doesn't read the data inside each file if the (mtime, size) matches on both ends, unless you:
rsync -c
rsync -X to also copy xattrs (slower)
rsync -H for hard link preservation (uses memory to track all the inodes)
rsync -S to preserve sparseness in sparse file
rsync --zc=zstd
rsync --zc=zstd --compress-level=6  or maybe 8 where the link is slow and extra compression is useful