*sshole: the OpenSSH password bypass on Alpine and other musl servers
The OpenSSH *sshole flaw lets long passwords open locked accounts on musl systems such as Alpine, OpenWrt and Docker. How to check, patch and detect it.

On 5 November 2026 the OpenSSH team released 10.6p2 with a critical security fix. The flaw it fixes was published as “*sshole”: on Linux systems that use the musl C library and run without PAM, accounts with a locked password can be opened over SSH with any password longer than 256 characters.
Alpine-based servers, Docker images and embedded devices match this profile exactly. In this article we explain how the flaw works, how to check whether you are affected, what to do about it and how to look for signs of an attack.
This article is based on information public as of 6 November 2026. No official CVE has been published for the flaw yet; numbers quoted on some sites are not confirmed by primary sources.
What is the flaw?
On Linux, accounts with no password, or that must not log in with one, usually have * in the password field of /etc/shadow. It is a convention going back to the 1980s that means “no password matches this account”. Key-only SSH users, LDAP users and, by default on Alpine, the root account are set up this way.
*sshole comes from two unrelated, old decisions lining up:
- musl’s behaviour (2012): when musl’s
crypt()cannot compute a password hash, it returns*instead of an error. musl’s sha512 implementation has a 256-byte password limit, so a longer password produces*. - OpenSSH’s timing countermeasure (2016): to stop attackers from learning which users exist by measuring response times (CVE-2016-6210), OpenSSH began using the salt of a real account on the system during password checks. This behaviour also applies to real accounts whose password field is
*.
The result: when an attacker sends a password of 257 to 1,024 bytes to an account whose password field is *, crypt() returns *, which matches the stored value exactly, and the login is accepted. According to the research, the problem has existed since 2016.
The impact is whatever the compromised account can do. But the researcher notes that on Alpine’s official AWS and Azure images, the default alpine user has passwordless doas, so on those images the flaw means root access.
Who is affected?
Exploitation needs all of the following:
- The system uses musl: Alpine, OpenWrt, Buildroot, Yocto, postmarketOS, Void (musl), Chimera and Docker images based on them.
- sshd runs without PAM. Alpine’s standard
openssh-serverpackage is built without PAM. Systems using theopenssh-server-pampackage withUsePAM yesare not affected. - Password authentication is on (
PasswordAuthentication yes, the default). - At least one account has exactly
*as its password field. - According to the researcher’s check tool, at least one account with a real
$5$or$6$password hash also exists on the system.
Not affected: glibc distributions (Ubuntu, Debian, RHEL, Rocky, AlmaLinux and so on), sshd running with PAM, OpenBSD, FreeBSD, macOS and Dropbear. OpenWrt uses Dropbear by default; check it if you installed OpenSSH later.
Are you affected? Five commands
Run these as root on the server:
# 1. musl? (if a file is listed, you are on musl)
ls /lib/ld-musl-* 2>/dev/null
# 2. OpenSSH version (10.6p2 and later are fixed)
ssh -V
# 3. Effective sshd settings: password logins and PAM
sshd -T | grep -Ei '^(passwordauthentication|usepam)'
# 4. Accounts whose password field is exactly "*"
awk -F: '$2 == "*" { print $1 }' /etc/shadow
# 5. Accounts with a real password hash
awk -F: '$2 ~ /^\$[56]\$/ { print $1 }' /etc/shadow
How to read the output: if command 1 lists a file, command 2 shows a version older than 10.6p2, command 3 shows passwordauthentication yes without usepam yes, and commands 4 and 5 both return accounts, the system is exposed.
If you use Docker, run the same checks in every container and image that runs an SSH server (docker exec -it <container> sh). The researcher’s check.sh script runs these checks in one go; as with any script, read it before running it as root.
What to do, in order
1. Install the patch
Alpine packages are out and, according to the research, the fix has been backported to supported releases and several older Alpine branches:
apk update && apk upgrade openssh
rc-service sshd restart
ssh -V
For Docker, updating the running container is not enough: update the base image, rebuild your image and restart containers from it. Upgrade OpenSSH built from source or shipped on embedded devices to 10.6p2; if your device vendor has not released an update, contact them.
2. If you cannot patch now, turn off password logins
echo "PasswordAuthentication no" > /etc/ssh/sshd_config.d/00-no-passwords.conf
sshd -t && rc-service sshd restart
According to the researcher, this closes the hole completely. Before you do it, confirm in a separate session that you can log in with an SSH key, and keep your current session open; otherwise you may lose access to the server. If your sshd_config has no Include /etc/ssh/sshd_config.d/*.conf line, put the setting in /etc/ssh/sshd_config directly and confirm with sshd -T that it is in effect.
3. Do not leave SSH open to the internet
As long as the SSH port is open to the internet, anyone can try this flaw. Lasting measures:
- Limit SSH to a VPN or a list of allowed IP addresses (firewall or cloud security group).
- Name the accounts that may log in with
AllowUsersorAllowGroups. - Shrink the attack surface with settings such as
PermitRootLogin noandMaxAuthTries 3. - Turn off password logins on servers for good; use keys or hardware keys (FIDO2) only.
Detecting attacks: what to look for in the logs
To sshd, the attack looks like a successful password login; no special error is logged. The key to detection is therefore spotting password logins on accounts that should never log in with a password.
1. List password logins
# Alpine (syslog): /var/log/messages · OpenWrt: logread · systemd: journalctl -u sshd
grep -h "Accepted password for" /var/log/messages* 2>/dev/null
Then compare these users with the accounts from command 4 whose password field is *. If you see an “Accepted password” line for an account whose password field is *, there is no legitimate password for that account; treat it as an attack.
2. Look at the context
- Were there
Failed passwordorInvalid userlines from the same IP just before? - Do you recognise the source IP? What time was it?
- If you ship logs to a central system (SIEM or syslog server), query all
Accepted passwordlines per account over recent months. The bug existed since 2016 but only became public in November 2026, so looking back matters. - Use
LogLevel VERBOSEfor more detail; it records the method and key fingerprint for each login.
3. Look for traces on the system
If you find a suspicious login, check the same time window for:
lastandlastlogoutput and active sessions (who).- New keys in
~/.ssh/authorized_keysfiles (especiallyrootand service accounts). - Changes to
/etc/passwd,/etc/shadow,/etc/doas.confand/etc/sudoers. - New scheduled jobs (
/etc/crontabs/,/etc/periodic/) and startup scripts (/etc/local.d/,/etc/init.d/). - Unknown processes and outbound connections (
ps,netstat -tunaporss -tunap).
Incident response if you were compromised
- Preserve evidence: copy logs and suspicious files off the server, then act.
- Cut off access: isolate the server or restrict SSH to the management network.
- Rebuild: rather than trying to clean a compromised server, rebuild it from an up-to-date, patched image. For containers this is the normal approach anyway.
- Rotate secrets: change SSH keys, API keys, database passwords and cloud credentials stored on the server.
- Check for spread: review other systems that share the same keys or passwords.
The lesson for companies: images and defaults
*sshole shows once again that vulnerabilities do not always come from new code. Here two decisions, 14 and 10 years old, combined to unlock locked accounts. Three practical lessons:
- Asset inventory: if you do not know which servers and containers run which distribution, C library and OpenSSH version, you cannot respond to an advisory like this within hours.
- Image update pipeline: rebuilding container images regularly and scanning them is a core part of secure software and DevSecOps practice.
- Management access: management services such as SSH should never be open to the internet. Our server and IT management service can set servers up this way and keep them up to date.
Conclusion
*sshole only affects one specific setup: musl, sshd without PAM, password logins on and an account whose password is *. But that setup is very common on Alpine-based servers and Docker images, and on some cloud images it leads straight to root.
The to-do list is clear: upgrade OpenSSH to 10.6p2, turn off password logins if you cannot upgrade yet, keep SSH off the internet and search your logs for “Accepted password” lines on passwordless accounts. If you would like your servers and container images reviewed with this in mind, see our cybersecurity services or get in touch.
Sources: OpenSSH release notes (10.6p2), sshhole.com, The researcher’s GitHub repository and check script, Threat Landscape, Cybersecurity News
Bu yazı Türkçe olarak da mevcut — Türkçe oku →
Frequently Asked Questions
No. The flaw only affects OpenSSH servers built against the musl C library and running without PAM. Ubuntu, Debian, RHEL and their derivatives use glibc and run sshd with PAM; the OpenSSH release notes and the researcher state that these setups are not affected. The real exposure is Alpine-based servers, Docker images and embedded devices.
Turn off password logins in sshd (PasswordAuthentication no) and restart sshd. According to the researcher, this closes the hole completely. First confirm in a separate session that you can log in with an SSH key, or you may lock yourself out.
The strongest sign is an 'Accepted password for' line for an account that normally logs in only with an SSH key or whose password is locked. Search the sshd logs for these lines and check the same time window for new users, changed authorized_keys files, new scheduled jobs or privilege escalation.