Linux Security

Why /etc/hosts suddenly stops being respected after a kernel upgrade

When a kernel upgrade lands, you expect only low‑level changes. In practice, the upgrade can ripple through the entire userland, especially when systemd‑resolved is involved. A common symptom is that hostname lookups that used to resolve via /etc/hosts now fall back to DNS or fail entirely. The culprit is often a subtle change in how systemd‑resolved loads its configuration or how the Name Service Switch (NSS) library consults /etc/hosts.

Why per‑interface search domains matter

If a box has more than one NIC—like a wired LAN that talks to a corporate DNS and a Wi‑Fi that talks to a public ISP—mixing the two can bite you. A query for server will be sent to the DNS server of the interface that sent the request, but the search list that the resolver appends is taken from the global configuration unless you tell it otherwise. That means you can end up asking your public DNS for an internal host or, even worse, leaking internal names to the Internet. Per‑interface search domains keep the resolver tidy and stop accidental leakage.

When a kernel upgrade silently breaks your network

Kernel upgrades are usually painless, but on a few systems they can trigger a cascade of “carrier lost” messages that make the interface drop and come back up repeatedly. The messages are buried deep in the system journal, so you may not notice them until the network flaps start causing outages. This post shows how to locate those hidden messages with journalctl, correlate them with real‑world symptoms, and apply a fix that keeps the interface stable.

Rebuilding the initramfs is the most reliable way to escape an emergency shell after a kernel upgrade that leaves the system stuck in emergency mode.
The problem usually stems from a mismatch between the running kernel and the modules or hooks that the initramfs contains.
Below is a step‑by‑step guide that covers the common causes, the exact commands you’ll run, and the security‑aware trade‑offs you should keep in mind.

Why Emergency Mode Happens After a Kernel Update

When you install a new kernel on Ubuntu 24.04, GRUB points to the new image, but the initramfs that ships with it must contain:

Extracting a Single File from a Borg Archive Without Recreating the Entire Directory Tree

When a backup archive grows to several terabytes, restoring a single configuration file or a database dump can feel like pulling a needle from a haystack. Borg’s default extract command rebuilds the whole directory tree, which is wasteful when you only need one file. The following guide shows how to pull a single file efficiently, keeps metadata intact, and keeps your system secure.

I’d be happy to help polish it, but I’ll need the original article text first. Could you paste it here?

Understanding the 2775 Permission Set

The octal mode 2775 is the go‑to for giving a directory shared write access while keeping new files in the same group.

• 2 – set‑gid bit: new files inherit the directory’s group.
• 7 – owner gets rwx.
• 7 – group gets rwx.
• 5 – others get r-x.

On a path like /srv/shared it looks tidy: any member of the group can drop files, and those files stay in the same group for later collaboration. The flip side? Group write is a double‑edged sword. If the group contains people you don’t fully trust, or if the directory is exposed to a wider audience, the set‑gid bit can become a vector for privilege escalation or accidental data exposure.

Why KDE Plasma Appeared During Upgrade

When Debian 12 “Bookworm” landed in early 2025, the default desktop stack for the desktop task was GNOME. A lot of folks, myself included, still had KDE Plasma sitting around from a previous release or a custom install. During a normal apt full-upgrade, the package manager pulls in the newest kde-plasma-desktop meta‑package because it’s part of the kde-standard task, which is automatically enabled by the tasksel configuration that ships with Debian. The result? A silent, automatic installation of a sizeable KDE stack even if you never intended to use it.

When a Windows machine syncs a shared folder that contains /etc/hosts, the file often ends up with Windows‑style CRLF line endings.

If you’ve ever pulled a hosts file from a Windows share, you’ve probably noticed that the resolver starts acting weird. The glibc resolver stops at the carriage return (\r) and treats the rest of the line as part of the hostname. That means lookups for localhost, myserver, or any entry that follows a stray CR fail, and some services silently fall back to DNS—exposing the system to name‑resolution attacks.

Why a service can stay “activating” after a kernel upgrade

When you bump the kernel, systemd does a hard reset of almost everything.
If a unit you depend on never leaves the activating state, the rest of the boot can stall or the service simply never runs.
The usual suspect is a dependency on a kernel‑provided resource that is missing or delayed after the upgrade.

Below is a practical checklist that shows how to diagnose the issue, why it happens, and how to fix it without compromising security or stability.