Nov 09, 2026 · 8 min read
OpenSSH CVE-2026-123158: Long Passwords Open Locked Accounts
Researcher Özgün Kültekin's "*sshole" bug lets anyone reach an SSH port on Alpine and other musl systems and log into accounts that were never supposed to take a password, root included. OpenSSH shipped a workaround in 10.6p2 on November 5, 2026.
An account whose password field is a single asterisk is supposed to be locked to password logins. On Alpine Linux, that is how root ships. Since 2016, typing a password longer than 256 characters into such an account over SSH has worked, as long as the box runs musl libc without PAM.
Alpine's own advisory puts it bluntly: "It takes one connection, works every time, and needs no retries." No memory corruption and no brute force. Just a long string.
Key Takeaways
- CVE-2026-123158, published November 5, 2026 with a CVSS 3.1 score of 8.1, covers sshd in OpenSSH before 10.6p2 on Linux when musl libc is used without PAM.
- Accounts whose shadow password field is
*accept any password longer than 256 characters, up to sshd's 1,024 character cap, once another account on the host has a real SHA crypt hash. - Alpine Linux is the main exposure because its openssh-server package runs without PAM by default; glibc distributions such as Debian, Ubuntu and Fedora are not affected.
- OpenSSH 10.6p2 fixes it, Alpine has patched packages for branches 3.16 through edge, and setting
PasswordAuthentication noblocks it until you update. - CISA's record marks exploitation as "poc"; no source reports attacks in the wild so far.
What Is CVE-2026-123158?
CVE-2026-123158 is a password authentication bypass in OpenSSH's sshd that only appears on musl based Linux builds with PAM disabled. The NVD record says sshd "lacks a workaround for a password authentication bypass (related to misinterpretation of whether a string is intended to be a password hash)." It names Alpine Linux as the example platform.
Özgün Kültekin of Wakefield found it and published the write up on sshhole.com, nicknaming it "*sshole." He reports throwing "like 17,600 weird logins" at builds with glibc, musl, PAM, no PAM, shadow and no shadow. Every build agreed on every login except one combination: musl without PAM.
The researcher lists musl systems at risk as Alpine, OpenWrt, Buildroot, Yocto and many Docker images. Dropbear, OpenBSD and any setup running through PAM are not affected, per his page.
How Does a Long Password Unlock a Locked Account?
A long password makes musl's crypt() fail, musl reports that failure by returning the string "*", and sshd then compares that string to the account's stored * and sees a match. Three decisions from three different years line up to make it work.
- musl, 2012. musl's SHA256 and SHA512 crypt code sets
#define KEY_MAX 256and gives up on longer keys. The__crypt_sha512wrapper then returns"*"instead of the null pointer the 1997 Single UNIX Specification describes for errors. - OpenSSH, 2016. To fix the username enumeration bug CVE-2016-6210, sshd started hashing every password attempt, even for accounts with no usable hash. When the stored value is too short to be a salt, the
pick_salt()helper borrows the salt prefix of the first crypt style hash it finds in the password database. Alpine says a$5$or$6$hash is what makes the attack work, which is why it needs a second account with a real password. - Alpine, 2022. Alpine 3.16 switched root's password field from
!to*, according to Kültekin's timeline. That change kept root key only, and also made it a target.
The last step lives in sys_auth_passwd() in auth-passwd.c, which accepted the login when strcmp(encrypted_password, pw_password) == 0. A failed hash of "*" equals a stored *. sshd separately rejects passwords over its MAX_PASSWORD_LEN of 1,024, which sets the ceiling. Accounts with a valid hash are safe; BleepingComputer notes the flaw "does not allow attackers to bypass passwords on accounts with valid password hashes."
Damien Miller's fix commit, dated November 4, 2026, adds #define MIN_PASSWORD_HASH_LEN 4 and refuses any match against a stored value shorter than that. Its message is pointed: "We do not consider the root cause to be within OpenSSH but we are well equipped to avoid it." musl's behavior is unchanged, per the researcher's page.
Which Systems Are Exposed?
A host is exposed only when four conditions hold at once. The Alpine advisory and the researcher agree on them:
- sshd is linked against musl libc.
- PAM is off. That is the default in Alpine's openssh-server package, or anywhere
UsePAM nois set. PasswordAuthenticationisyes, which is the OpenSSH default.- One account has
*as its password field and at least one other has a$5$or$6$hash.
Cloud images are the sharp edge. Alpine says its tiny-cloud images from 3.18.11 through 3.24.3 ship an alpine user with a * password and passwordless doas. They are "safe on first boot" and "become exploitable as soon as any $6$ hash exists on the machine." One passwd command for a new admin account turns a locked default user into a root shell for anyone who can reach port 22.
Why the 8.1 Score Undersells It
The CVE record rates attack complexity as High, which drags the score below critical. That rating describes the preconditions, not the attack. Once a host matches, exploitation is a single deterministic login with no credentials, against root on a stock Alpine install. OpenSSH's own release notes call 10.6p2 "a critical security workaround."
This is also the second time in seven years that Alpine root's password field has been the bug. In 2019, CVE-2019-5021 found Alpine Docker images with no root password at all; Alpine fixed it with !, then moved to * in 2022. The asterisk was the safer marker on paper. It only became dangerous because sshd trusted a string comparison with a library that uses the same character to mean "error." Public lab repos for reproducing the bug are already listed in the CVE references, so assume scanners will follow. We saw the same pattern after the libssh2 CVE-2026-55200 proof of concept dropped.
What This Means for Your Email Infrastructure
If any part of your mail stack runs on Alpine with sshd listening, treat it as in scope. A small SMTP relay, a webmail container or the host that stores your DKIM private key are all just Linux boxes with an SSH port. Root on one of them means reading every queued message and every mailbox it serves.
The DKIM key is the worse loss. Mail signed with your own key passes authentication checks, so an intruder can send phishing that looks perfectly legitimate to Gmail and Outlook filters, from your domain. Rotating that key and republishing the DNS record is the cleanup step teams forget. SSH access to developer and ops machines has been a recurring target this month; see how a tampered Claude Code extension planted SSH keys on Cursor users.
How Do You Check and Fix Your Servers?
Update OpenSSH to 10.6p2 or your distribution's patched package, then restart sshd. On Alpine the fixed builds include 10.6_p2-r0 on edge, 10.3_p1-r2 on 3.24, 10.2_p1-r1 on 3.23 and 10.0_p1-r11 on 3.22, per the Alpine security tracker. Run these checks as root:
- Confirm the libc.
ldd /usr/sbin/sshd | grep -i musl. No output means glibc, and you are not affected. - Check the version.
ssh -Vshows the client; on Alpine,apk info -v openssh-servershows the package release to compare against the list above. - Read the live config.
sshd -T | grep -Ei '^(usepam|passwordauthentication)'. Ifusepamis missing, sshd was built without PAM. Ifpasswordauthentication yesappears, the door is open. - Find asterisk and real hashes.
awk -F: '$2=="*"' /etc/shadowlists exposed accounts, andawk -F: '$2 ~ /^\$[56]\$/' /etc/shadowlists the hashes that enable the attack. - Patch and restart.
apk upgrade -U -a, thenrc-service sshd restart. Alpine warns that "Upgrading the package does not restart a runningsshd." - Mitigate if you cannot patch. Add
PasswordAuthentication noto sshd_config and restart. While there, setKbdInteractiveAuthentication notoo. - Hunt for past use.
grep "Accepted password" /var/log/messages. A password login for an account whose field is*means the host should be treated as compromised.
Kernel and SSH bugs keep landing on the same Linux fleets; our DirtyClone CVE-2026-43503 breakdown covers the local root side of that problem.
What Comes Next?
OpenSSH closed the sshd path, but musl still returns "*" on a failed hash. Any other program on a musl system that calls crypt() and compares the result to a stored field can inherit the same mistake. Expect audits of embedded login code on OpenWrt and Yocto builds, where updates move slowly.
Miller's 10.6p2 announcement also says OpenSSH will ship more frequent releases because AI assisted tools are finding bugs that others may find too. BleepingComputer reports Kültekin found this one using AI agents that ran about 17,600 authentication tests. Plan for shorter patch windows on SSH, and disable password logins anywhere keys already work.