*sshole: OpenSSH Flaw Unlocks Locked Accounts

A flaw in OpenSSH on Alpine and other musl-based Linux systems lets anyone log into accounts that are supposed to reject passwords entirely, using any password over 256 characters. It has been hiding in plain sight since 2016.

A locked door that opens for a 257-character password

There is an account on your server that should never take a password. Root. Your key-only deploy user. The default "alpine" user on a cloud image. Someone locked it on purpose, which in the Unix world usually means the password field holds a single asterisk instead of a real hash.

An asterisk is not a hash. Nothing you type will ever match it. That has been the deal since 1980, when Bill Joy put a * next to the "daemon" account in BSD so nobody could ever log into it. Unix still works this way today.

Except on Alpine Linux and other musl-based systems, there is now a way in. Type any password longer than 256 characters and the door swings open. Any password. Yes, any. That is the *sshole flaw, disclosed on November 5, 2026, and it has been sitting there quietly since 2016.

How three small decisions added up to a hole

The flaw is a chain of perfectly reasonable decisions that, laid end to end, collapse into a security disaster.

First, in 2012, the musl C library changed its crypt() function so that when password hashing fails, it returns a single character instead of NULL. That character was picked to never match a real password. You can guess which one. The asterisk. Again.

A few weeks later, musl added SHA-512 password hashing and capped passwords at 256 bytes to stop someone from sending a giant password and grinding the CPU. Anything longer simply fails, and a failure returns that same asterisk.

Then in 2016, OpenSSH fixed a username enumeration bug (CVE-2016-6210) by making the server hash your password even when an account has no usable hash. That way an attacker cannot tell, from timing, which usernames are real. To do this, sshd borrows another account's salt and runs the hash anyway.

Now put the pieces together. A locked account stores *. An attacker sends a password longer than 256 characters. musl's crypt() hits its length limit, fails, and returns *. OpenSSH compares the two asterisks. They match. You are in.

There is no single patch for the thinking that got here. Three teams, years apart, each made a call that looked sensible on its own.

Who is actually exposed

This is not every Linux box. If you run Debian, Ubuntu, Fedora, or anything on glibc, you are fine. If your SSH goes through PAM, you are fine. The hole only opens when a few conditions line up:

  • The server runs a musl-based distro (Alpine, OpenWrt, Buildroot, Yocto).
  • PAM authentication is off, which is the default on Alpine's openssh-server.
  • Password logins are still enabled.
  • At least one account is locked with a *, and another has a real password.

The sharpest edge here is containers. Alpine is the most popular base image in the world, which means a lot of the Docker images you have already pulled are built on musl. The researcher also warns that Alpine's official AWS and Azure images ship a default "alpine" account with passwordless doas, which is effectively root for anyone who gets in.

CVSS puts this at 8.1, High. Whether that is "interesting" or "emergency" for you comes down to one question: do you actually know which of your systems, and which of your container images, are built on musl?

Patch it, then go look for evidence

OpenSSH 10.6p2, released the same day, fixes it. On Alpine, updating is one command:

apk upgrade openssh

If you cannot patch right now, turn off password authentication and restart SSH:

echo "PasswordAuthentication no" > /etc/ssh/sshd_config.d/00-no-passwords.conf

Then check whether anyone already used this against you. A successful password login for an account that should never accept one is the tell:

grep "Accepted password" /var/log/messages

If you see a password login for a locked account, treat it as a compromise, not a curiosity.

Most teams will not know they were exposed until they go looking. If you would rather not learn about a ten-year-old hole the hard way, let's check.


Book a 15-minute Linux security check
We'll map which systems and container images need patching first, with no obligation.