OpenSSH 'SSHole' (CVE-2026-123158): When a Long Password Becomes a Skeleton Key

SSHole (CVE-2026-123158) is an OpenSSH authentication bypass on Linux systems that use musl libc without PAM. An oversized password makes the hash computation fail and return an asterisk, which matches the locked-account placeholder in the shadow file, so sshd accepts the login. It was exploited within about nine hours of disclosure. Here's the mechanism, who is affected, and the fix.

3 min readAdvanced

Executive summary. "SSHole" (CVE-2026-123158) is an authentication bypass in OpenSSH's sshd on Linux systems built with musl libc and without PAM. The trick is almost absurd: an oversized password (roughly 257–1024 bytes) makes the password-hash computation fail and return * — and * is the exact placeholder used for locked accounts in the shadow file, so sshd compares the two, sees a match, and lets the attacker in. Disclosed on 5 November 2026, it was exploited in the wild within about nine hours, with researchers confirming root on an Alpine honeypot. CVSS is 8.1; the fix is OpenSSH 10.6p2. Because musl-based images are everywhere — Alpine, OpenWrt, Buildroot, Yocto, and a great many Docker images — this one reaches far beyond "a Linux bug."

SSHole disclosure-to-exploitation timeline

What it is

Field Detail
CVE / name CVE-2026-123158 — "SSHole"
Component OpenSSH sshd on musl libc without PAM
Mechanism Oversized password hash returns *, matching the locked-account placeholder
At risk Locked / service accounts; root if PermitRootLogin is enabled
Severity CVSS 8.1
Fixed in OpenSSH 10.6p2
Reality Exploited ~9h after disclosure; botnet + cryptomining use reported

How it works

text
 Attacker sends a very long password (≈257–1024 bytes)
          │
          ▼
 musl crypt() can't hash it  ──►  returns "*"
          │
          ▼
 shadow file stores "*" for LOCKED accounts
          │
          ▼
 sshd compares "*" == "*"  ──►  MATCH  ──►  login accepted

The elegance (and horror) is that nothing is injected. The bug is a confusion between "hashing failed" and "this is a valid hash". On a system where locked or service accounts carry * in /etc/shadow, an attacker who sends a password long enough to break crypt() lands on a value that equals those placeholders. Some reports describe a two-attempt variant (a first attempt primes a cached fake salt), which is why severity isn't a flat 10 — but the outcome is a login the account was never meant to allow.

Who is affected

  • Systems using musl libc (not glibc) and not using PAM for password auth.
  • That disproportionately means containers and appliances: Alpine images, OpenWrt routers, Buildroot/Yocto embedded builds, and the long tail of Docker base images built on Alpine.
  • Locked and service accounts are the prize, and root is reachable where PermitRootLogin is on.

Why it needs industry attention

The scary part isn't one honeypot — it's blast radius plus speed. Alpine is one of the most common container bases in the world, and "disabled" or service accounts are everywhere. Attackers turned this into cryptomining and botnet activity, with persistence via implanted authorized_keys, within hours. That is faster than most patch cycles.

Fix and mitigation

  1. Upgrade OpenSSH to 10.6p2+ across hosts and, crucially, rebuild container images — a patched host with vulnerable images still ships the bug.
  2. Prefer key-based or MFA authentication; both would have blocked exploitation outright. Set PasswordAuthentication no where you can.
  3. Audit /etc/shadow for accounts with */! that nonetheless permit login, and make sure locked means locked.
  4. Disable PermitRootLogin unless you have a hard reason not to.

How to detect it

  • Hunt authorized_keys changes you didn't make — the common persistence step.
  • Successful logins to accounts that should be locked / service-only, especially from unexpected IPs.
  • sshd child processes that are shells, miners, or curl/wget pulling payloads.
  • Oversized password attempts in verbose auth logs where available.

Conclusion

SSHole is a textbook case of failing closed vs failing open: a hash function that returns a sentinel on error, meeting a shadow file that uses that same sentinel to mean "locked." Patch sshd to 10.6p2, rebuild your musl-based images, move to keys/MFA, and hunt for planted authorized_keys.

Do this now: inventory musl-based hosts and images (Alpine especially), upgrade OpenSSH to 10.6p2, disable password auth where possible, and sweep for unexpected authorized_keys entries.

Related: our network pentesting manual and the lesson on Linux privilege escalation.

Sources: Help Net Security, The Hacker News, BleepingComputer, Safe Security, Wiz. KEV/EPSS status varied by source at publication; confirm against the OpenSSH advisory and CISA KEV.

HackersRoot Research · Research team

The HackersRoot research team investigates attack techniques across AI systems, cloud and identity, and enterprise networks, then publishes the detections and hardening guidance that stop them.

More from HackersRoot →

Related research

Vulnerability Intel

FreeSWITCH CVE-2026-116779: One SIP Message to Command Execution

FreeSWITCH powers a huge share of the world's VoIP. A single crafted inbound SIP MESSAGE can select an 'api' chat protocol that the server then executes as FreeSWITCH API commands, several of which reach the operating system. The result is unauthenticated command execution, rated CVSS 9.8 and already exploited in the wild. Here's the mechanism, the fix in 1.11.3, and how to catch it.

Vulnerability Intel

Oracle PeopleSoft CVE-2026-35273: An Unauthenticated Door Into the HR Mainframe of the Enterprise

Oracle issued an out-of-band alert for CVE-2026-35273, a missing-authentication flaw in PeopleSoft PeopleTools' Environment Management hub (PSEMHUB). Reachable over HTTP with no credentials, it can lead to server-side request forgery and remote code execution. Mandiant tied zero-day exploitation to ShinyHunters (UNC6240) against higher education. Here's the mechanism and the response.

Vulnerability Intel

Zyxel GS1900 CVE-2026-107441: The Patch That Didn't Hold, and 996 Switches That Paid for It

In June, Zyxel patched a GS1900 switch flaw (CVE-2026-7273). But the fix didn't fully close it: Binarly found the exploit still worked, and GreyNoise saw data taken from around 996 switches in August. Zyxel reclassified the missing parameter-length validation as a new CVE, CVE-2026-107441, an out-of-bounds write giving unauthenticated root command execution. The real fix is 2.90 Patch 3.