SSHole (CVE-2026-123158): passwordless login to OpenSSH on Alpine Linux

On 5 November 2026 the OpenSSH project closed a flaw that lets attackers log in to Linux servers without a password: an oversized password is all it takes. Affected are systems using the musl C library without PAM, above all Alpine Linux. The first attacks came less than nine hours after the CVE was published, on one decoy server all the way to root. This article explains the flaw, shows who is affected, what to do now and how CyberTell helps.

13 min read CyberTell
Key takeaways
  • SSHole (CVE-2026-123158, CVSS 8.1) allows login to accounts without a password using any password between 257 and 1024 bytes long.
  • Affected is OpenSSH before 10.6p2 on Linux with musl libc and without PAM, typically Alpine Linux. Debian, Ubuntu, Fedora and other glibc systems are not.
  • The first attack followed within nine hours. Install the update and restart sshd, or set PasswordAuthentication no right away.
  • CyberTell finds exposed SSH services, matches deployed versions against new CVEs every day and guides you through the incident if a login cannot be ruled out.

What is SSHole?

SSHole, also written *sshole, is a vulnerability in the OpenSSH server. It is tracked as CVE-2026-123158 with a CVSS score of 8.1 (high). It was discovered by security researcher Özgün Kültekin and fixed in Portable OpenSSH 10.6p2, released on 5 November 2026.

An attacker who can reach the SSH port can log in to certain accounts without knowing their password. No stolen credentials and no SSH key are needed. Any password between 257 and 1024 bytes long will do. The attack takes a single connection, succeeds on the first try and does not produce a noticeable series of failed attempts.

Only accounts whose password field in /etc/shadow contains a single * are affected. On Linux this is the usual entry for accounts without a password, for example users who only log in with a key. Accounts with a real password hash cannot be taken over this way. The attacker gets the rights of the compromised account. If that account may switch to root with doas and no password, the whole server is in the attacker's hands.

The OpenSSH team notes that the actual root cause lies outside sshd, in the behaviour of the C library. The bug is exploitable through sshd, however, so it was intercepted there.

Timeline: from 2016 to the first attack

The flaw is new, its roots are not. The code path it exploits was created in 2016 as protection against a different vulnerability. Between the patch and the first attack, on the other hand, only hours passed.

  1. 2016

    OpenSSH 7.3 fixes CVE-2016-6210

    Attackers could tell from the response time whether a user name existed. The fix makes sshd run a real hash computation even for accounts without a valid hash, using the salt of another account. The same release only accepts passwords up to 1024 characters.

  2. 2022

    Alpine ships root with "*"

    Since about Alpine 3.16 the base image contains root:* in /etc/shadow. The official cloud images from 3.18.11 to 3.24.3 also include a user alpine with "*" and passwordless doas.

  3. 6 Oct 2026

    OpenSSH 10.6 is released

    The release notes report a surge of security reports from AI tools. Because such findings are often rediscovered independently, the team announces more frequent releases.

  4. 5 Nov 2026

    OpenSSH 10.6p2 and fixed Alpine packages

    OpenSSH releases 10.6p2 with the countermeasure in sshd. On the same day Alpine publishes a security advisory and fixed packages for every branch from 3.16 to edge, including releases without regular support.

  5. 5 Nov 2026, 15:46 UTC

    CVE-2026-123158 is published

    The CVE record lists OpenSSH before 10.6p2 on Linux with musl and without PAM, CVSS 8.1. The researcher puts up a page with an explanation and an in-browser demo, and public lab setups for reproducing the flaw follow.

  6. within 9 h

    First attack, root on a decoy server

    Threat intelligence firm Previdian observes the first exploitation attempt. Root access succeeds on an Alpine honeypot, and eight attempts come from a single address.

  7. 6 Nov 2026

    CISA rating and first reports

    CISA rates the technical impact as total. Trade media report the first attacks on exposed servers.

How the attack works

Without PAM, sshd checks a password itself: it uses the library function crypt() to hash the input and compares the result with the entry in /etc/shadow. For an account with * there is no salt for that computation. Since the 2016 fix, sshd therefore borrows the salt of another account with a real hash ($5$ or $6$). The check takes as long as for a normal account, so attackers can no longer tell user names apart by response time.

This is where SSHole comes in. With this method the musl C library rejects passwords longer than 256 bytes and reports the error by having crypt() return the string *. That very character sits in the password field of every account without a password. sshd compares * with *, finds them equal and reports a successful login.

The upper limit of 1024 bytes comes from sshd itself: since 2016 it does not accept longer passwords at all. With glibc the bug does not occur, because there crypt() returns a normal hash even for long passwords, and that hash is never *.

A single account with a real password is enough as a salt donor. After that, every account with * is reachable.

01ssh client
$ ssh alice@server
alice@server's password:
AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
# 300 bytes
02sshd · /etc/shadow
root:*:20125:0:::::
alice:*:20125:0:::::
bob:$6$Qx9f…$…:20125:0:::::
# alice: "*" → no salt
salt = $6$Qx9f… # bob
03crypt(password, salt)
musl libc · no PAM
len(password) > 256
return "*"
glibc
return "$6$Qx9f…$Zk1…"
04result == shadow(alice)
"*" == "*"
Accepted password for alice
"$6$Qx9f…" != "*"
Failed password for alice
An SSHole login: with musl the error value of crypt() becomes a matching hash, with glibc a normal hash is produced and the login fails.

Who is affected and who is not

A server is only affected if all of the following conditions are met. Each line can be checked directly on the system.

Condition Check on the server
sshd uses musl libc ldd $(command -v sshd) lists ld-musl
sshd checks passwords without PAM package openssh-server instead of openssh-server-pam, or UsePAM no
Password login is on (default) sshd -T shows passwordauthentication yes
An account has * as its password field awk -F: '$2=="*"' /etc/shadow
Another account has a $5$ or $6$ hash grep -E '^[^:]+:\$[56]\$' /etc/shadow

Typical candidates

  • Alpine servers and VMs with the default openssh-server package, according to Alpine every release checked from 3.16 to edge
  • official Alpine cloud images: the user alpine has * and may become root with doas without a password. As soon as any account gets a password, for example for the console, the path via alpine leads to root
  • users from a directory service who come in through musl-nscd and have no entry in /etc/shadow
  • other musl systems with OpenSSH, such as custom builds with Buildroot or Yocto

On Alpine, root also has *, but it is protected by the OpenSSH default PermitRootLogin prohibit-password as long as nobody has set PermitRootLogin yes.

Not affected

Debian, Ubuntu, Fedora and other glibc distributions, systems with PAM (openssh-server-pam with UsePAM yes), Dropbear, BSD systems, containers without a running sshd and servers with PasswordAuthentication no. Locked accounts with ! in the password field are also safe.

What the fast exploitation shows

Previdian reports the first exploitation attempt less than nine hours after the CVE was published. Root access succeeded on a decoy server running Alpine Linux. In total the company counted eight attempts from one address geolocated to Japan (89.116.89[.]18). The attackers sent oversized passwords and ran reconnaissance commands such as id, uname and hostname in non-interactive sessions. That is how you check whether access works before using or reselling it.

Three lessons for SMEs:

  • The window is measured in hours. A weekly or monthly patch cycle is not enough for exposed services. If you do not know which of your servers have an open SSH port, you cannot act in time.
  • Block lists do not help. Tools such as Fail2ban react to series of failed attempts. SSHole needs a single connection.
  • The impact depends on the account. Security vendor Safe Security points out that after the login, deployment credentials, application secrets and connected systems are within reach. A server login can quickly turn into an incident across the whole network.

In its release notes the OpenSSH team writes that flaws found by AI tools are often rediscovered independently. It assumes attackers find such bugs too, and therefore ships fixes more often. For operators that means more updates at shorter intervals. Which reports concern your systems is shown by Threat Intelligence in the portal.

Immediate steps for operators

If you run Alpine or another musl system with OpenSSH, proceed in this order. The commands come from the Alpine security advisory. Fixed releases include Alpine 3.21.10, 3.22.8, 3.23.8 and 3.24.4.

  1. Install the update and restart sshd

    Update with apk upgrade -U -a, then run rc-service sshd restart. The package update alone does not replace the running daemon.

  2. If that is not possible right away: turn off password login

    Set PasswordAuthentication no and restart sshd. Check beforehand that key login works, and keep your current session open.

  3. Check the logs

    Search /var/log/messages for "Accepted password for" on accounts that have no password at all: alpine, root or users who only log in with a key.

  4. Look for traces

    On exposed servers, review doas use, newly created users, recently changed files and authorized_keys.

  5. When in doubt, treat it as compromised

    If an unauthorised login cannot be ruled out, the server counts as compromised: isolate it, preserve evidence, rotate credentials and rebuild it.

How CyberTell helps in a situation like this

A flaw like SSHole puts the same questions to every SME: Where do we run SSH servers? Which of them can be reached from the internet? Are they affected? And what do we do if someone is already in?

CyberTell answers these questions with modules that work together: attack surface, external pentest, vulnerability radar, asset register, Threat Intelligence and incident management. Patching the server and checking /etc/shadow remains the job of your IT. CyberTell makes sure no system is forgotten, the alert arrives in time and the response is documented.

  1. Attack surface: find exposed SSH services

    The module determines passively from public sources which addresses of your domains are reachable, with open ports, services and known CVEs. An open port 22 running OpenSSH stands out without CyberTell scanning your systems. The recurring run shows what has been added since last time.

  2. External pentest: take a closer look

    Via the Raspberry Pi, the external pentest actively scans your verified domains with nmap and detects services including their version. Findings appear in the portal with a severity.

  3. Vulnerability radar: match versions every day

    You record servers and software with product and version, optionally from suggestions out of the attack surface. Every night CyberTell matches the inventory against new CVEs. Hits above your threshold are reported by email and push, and actively exploited flaws from the CISA KEV catalogue always count as critical.

  4. Asset register: know what is critical

    If a system is listed as critical in the asset register, the severity of its radar findings goes up one level. The server holding customer data ends up at the top of the list.

  5. Threat Intelligence: the report that concerns you

    The news feed collects security reports from trusted sources. If a report mentions the CVE of an open finding, your admins receive a notification.

  6. Incident management, war room and playbooks: respond

    If a login cannot be ruled out, it becomes an incident with a ticket, a war room for coordination and a playbook with steps to tick off.

Attack surface Sample data
Attack surface in CyberTell: addresses, ports and services visible from outside Attack surface in CyberTell: addresses, ports and services visible from outside
Attack surface in CyberTell: addresses, ports and services visible from outside

Reading version numbers correctly

A match by version number is a hint, not a verdict. With SSHole this applies in both directions:

  • Patched despite an old number: Alpine backported the fix to older packages. Alpine 3.24 is safe with openssh 10.3_p1-r2 but still reports version 10.3. A check against "before 10.6p2" will list this server anyway.
  • Old number, but not affected: a Debian server with OpenSSH 9.2 is older than 10.6p2 but uses glibc and PAM. The conditions of the flaw do not apply.
  • What nobody sees from outside: whether a server uses musl and whether an account has * is not revealed by any banner. Only a check on the system answers that.

In the vulnerability radar you mark such hits as a false positive or an accepted risk after checking. The status persists across the daily runs, and a new version resolves the finding automatically at the next match. In the attack surface you also track findings with severity and status until the service is patched or closed.

Attack surface Sample data
Attack surface findings with severity and status Attack surface findings with severity and status
Attack surface findings with severity and status

When a login cannot be ruled out

Alpine puts it plainly: if you cannot rule out an unauthorised login, treat the machine as compromised. Then an orderly procedure matters, as described in the article Incident response for SMEs.

In incident management you open a ticket with priority and owner and create an investigation if needed. The war room brings together IT, management and service providers. A playbook from AI Playbooks & Tabletop guides you through the steps: isolate the server, preserve logs and authorized_keys, rotate every credential and key that was stored on the server, and rebuild the system from a clean source.

Check your reporting duties too. If personal data is affected and the affected persons are likely to face a high risk, the Swiss Data Protection Act requires a report to the FDPIC.

Attack surface Sample data
Investigating a security incident in CyberTell incident management Investigating a security incident in CyberTell incident management
Investigating a security incident in CyberTell incident management

Frequently asked questions

Is my Ubuntu or Debian server affected by SSHole?
No. SSHole only affects systems running OpenSSH with the musl C library and without PAM. Debian, Ubuntu, Fedora and other glibc distributions are not affected, nor are Dropbear and BSD systems.
Is updating the package enough?
No. After the update the old sshd keeps running until you restart it, on Alpine with rc-service sshd restart. Only then is the fix active.
Does Fail2ban protect against SSHole?
Hardly. The attack succeeds with a single connection on the first try, so a ban after several failed attempts comes too late. The update or PasswordAuthentication no are what work.
Are Docker containers based on Alpine affected?
Only if sshd runs inside the container and password logins are allowed. Most containers do not start an SSH server and are therefore not affected.
Why is SSHole rated high and not critical?
The CVSS score sets attack complexity to high because several conditions must coincide: musl, no PAM, password login and an account with "*" next to an account with a password. When all of that applies, the attack works reliably.
How does CyberTell learn about my SSH servers?
The attack surface lists the externally visible addresses of your domains with open ports and services. You record internal servers in the vulnerability radar with product and version, and CyberTell then checks them against new CVEs every day.
Topics

Sources

  1. OpenSSH 10.6p2 release notes (2026-11-05)
  2. Alpine Linux: Security advisory: openssh-server password authentication bypass (2026-11-05)
  3. CVE-2026-123158 (CVE Program)
  4. sshhole.com: explanation and in-browser demo by the researcher
  5. The Hacker News: Hackers Target OpenSSH SSHole Flaw Within Hours of Disclosure (2026-11-06)
  6. CVE-2016-6210 (CVE Program)