Vulnerabilities of the Week: OpenSSH, nginx, PHP, Linux

It has been a rough week for anyone who runs servers. One OpenSSH vulnerability is already being used to break in, a way to crash a Linux server remotely without logging in has turned up in the kernel, and nginx, Apache and PHP have all shipped large batches of fixes, including for critical bugs that risk remote code execution.
Below is a rundown of the week’s most important vulnerabilities with practical steps: how to tell whether your server is affected, how to update, and what to do if no update is available yet.
What to fix first
The rows are in the order you should patch.
| Vulnerability | Product | Impact | Exploited | Fixed in |
|---|---|---|---|---|
| SSHole, CVE-2026-123158 | OpenSSH on Alpine (musl) | login without a password | yes | 10.6p2 |
| CVE-2026-119618 | Apache, mod_http2 | code execution | no reports | 2.4.71 |
| CVE-2026-123007 and 9 more | nginx | crash, possibly more | no reports | 1.31.9, 1.30.7 |
| about 20 CVEs | PHP 8.2–8.5 | code execution | no reports | 8.5.13, 8.4.28, 8.3.37, 8.2.36 |
| NoisyTCP, CVE-2026-99288 | Linux kernel | server crash | no | 6.12.117, 6.18.63, 7.3.4 |
| Java4Shell, CVE-2026-114938 | Java (HotSpot) | code execution | no | 17.0.21.1, 21.0.13.1, 25.0.5.1, 27.0.1.1 |
If you run Alpine with SSH password logins, start with the first item right now. For everyone else: web server and PHP first, then the kernel and Java.
Quick version check
One command shows which of these programs are installed and at which versions. Lines for missing programs simply print an error:
ssh -V; nginx -v; apachectl -v 2>/dev/null || apache2 -v; php -v | head -n 1; java -version; uname -r
See which updates are already available:
apt update && apt list --upgradable # Debian, Ubuntu
dnf check-update # AlmaLinux, Rocky Linux, RHEL
apk update && apk version -l '<' # Alpine
1. SSHole: OpenSSH login without a password (CVE-2026-123158)
Status: actively exploited. CISA added it to its Known Exploited Vulnerabilities catalog on 9 November. Researchers report the first attack attempts within hours of publication.
How SSHole works
On systems using the musl C library (Alpine Linux first and foremost) where OpenSSH is built without PAM, a very long password makes crypt() return a special value. OpenSSH treated that value as a match for accounts whose password field holds a placeholder. As a result, an attacker can log in to such an account without knowing any password.
It gets worse: in some Alpine cloud images the passwordless account has doas rights — effectively root.
Who is affected by SSHole
All three conditions must hold at once:
- a musl-based system — Alpine Linux and images built on it, including containers running
sshd; - OpenSSH without PAM support — which is how Alpine’s standard
openssh-serverpackage is built; - password logins enabled.
Servers on Debian, Ubuntu, AlmaLinux and other glibc-based distributions are not affected.
How to fix SSHole
Update OpenSSH and restart the service:
apk update && apk upgrade
rc-service sshd restart
Alpine shipped fixed packages for every supported branch on the day of disclosure, 5 November. Note that the fixed packages keep their old version numbers, so scanners that compare version numbers alone may still flag the server as vulnerable. Custom builds and other musl systems need OpenSSH 10.6p2.
If you can’t update OpenSSH right away
Disable password logins — that closes the hole:
sed -i 's/^#\?PasswordAuthentication.*/PasswordAuthentication no/' /etc/ssh/sshd_config
sshd -t && rc-service sshd restart
sshd -T | grep -i passwordauthentication
Make sure SSH key login works first, or you will lock yourself out. Setting up keys is covered in the guide to connecting over SSH.
Check for signs of compromise
If the server was reachable from the internet with password logins enabled, review the login log for the past few weeks and the process list:
grep -i "accepted" /var/log/messages | tail -n 50
last -n 30
Unexpected logins under system accounts are a reason to treat the server as compromised and reinstall it from a clean image.
2. Apache HTTP Server: a critical bug in mod_http2 (CVE-2026-119618)
Status: fixed on 11 November, no reports of exploitation.
How the Apache flaw works
A synchronisation bug in mod_http2 header handling leads to memory corruption and may allow remote code execution. The Apache developers rate it critical. The same release fixes two more issues: a child process crash in mod_md (CVE-2026-106023) and a buffer overflow in mod_vhost_alias (CVE-2026-109163).
Who is affected by the Apache flaw
Apache HTTP Server 2.4.55 through 2.4.70 with HTTP/2 enabled.
How to fix the Apache flaw
Check the version:
apachectl -v
Update Apache to 2.4.71, or to your distribution’s package with the fix backported:
apt update && apt install --only-upgrade apache2 # Debian, Ubuntu
dnf update httpd # AlmaLinux, Rocky Linux
If you can’t update Apache right away
Disable HTTP/2 temporarily. On Debian and Ubuntu:
a2dismod http2 && systemctl restart apache2
Or allow only HTTP/1.1 in the configuration:
Protocols http/1.1
Sites keep working, just without the benefits of HTTP/2.
3. nginx: ten vulnerabilities, the main one in HTTP/2 (CVE-2026-123007)
Status: fixed on 18 November, no reports of exploitation.
What’s fixed in nginx
The nginx developers released 1.31.9 and 1.30.7, closing ten vulnerabilities. The most serious, CVE-2026-123007, is an HTTP/2 flow control window overflow: according to the developers, an attacker can crash a worker process, and other impact is possible. Other notable fixes:
| CVE | Where | Severity |
|---|---|---|
| CVE-2026-118899 | infinite loop reading a request body over TLS 1.3 | major |
| CVE-2026-113425 | use-after-free in the mail proxy | major |
| CVE-2026-123633 | use-after-free in HTTP/2 | medium |
| CVE-2026-121810 | use-after-free in HTTP/3 | medium |
| CVE-2026-123634 | use-after-free with the resolver directive |
medium |
| CVE-2026-120967 | SNI bypass with client certificate checks in stream |
medium |
| CVE-2026-125283 | buffer overflow with alias plus try_files |
minor |
Which nginx versions are vulnerable
Practically every nginx version before 1.31.9 and 1.30.7. The HTTP/2 flaw affects versions since 1.9.5 — almost any server with HTTP/2 enabled.
How to update nginx
Check the version:
nginx -v
If nginx comes from the official nginx.org repository:
apt update && apt install --only-upgrade nginx
nginx -t && systemctl restart nginx
If nginx comes from your distribution’s repository, wait for the fixed package in your distribution’s security advisory — its version number may stay the same.
If you can’t update nginx right away
Disable HTTP/2 temporarily. In nginx 1.25.1 and newer, with a directive in the server block:
http2 off;
In older versions, remove http2 from the listen directive. Then test and reload the configuration:
nginx -t && systemctl reload nginx
If you don’t use nginx as a mail proxy, CVE-2026-113425 does not affect you. If nginx is a reverse proxy in front of your applications, update it first: it takes all external traffic.
4. PHP: about twenty security fixes
Status: fixed on 19 November, no reports of exploitation.
What’s fixed in PHP
PHP 8.5.13, 8.4.28, 8.3.37 and 8.2.36 came out today. They close about twenty vulnerabilities, and several of them, according to the developers, can lead to remote code execution. The most important for hosting:
- CVE-2026-129289 — request smuggling in PHP-FPM through an unread request body. Default nginx and Apache configurations do not reuse connections to FPM, so the demonstrated attack does not work against them, but setups with keep-alive to FPM are exposed.
- CVE-2026-129291 — a use-after-free in PHP-FPM that can lead to local privilege escalation.
- CVE-2026-126714 and CVE-2026-115560 — session handling bugs that can lead to code execution.
- CVE-2026-129496 — a use-after-free in the timezone cache with code execution risk.
- CVE-2026-129474 — a use-after-free in magic method backtraces with code execution risk.
- CVE-2026-128090 — a PHAR signature check bypass through algorithm downgrade.
- CVE-2026-129538 — out-of-bounds read and write in PHP-CGI.
Which PHP versions are vulnerable
Every supported PHP branch before the versions above. PHP 8.1 and older no longer get security fixes — if they still run anywhere, it’s time to move to a supported branch.
How to update PHP
Check the version — both the CLI and the one serving your sites:
php -v
php-fpm8.3 -v
Update the packages and restart PHP-FPM for every installed version:
apt update && apt upgrade
systemctl restart php8.3-fpm
The service name depends on the PHP version; systemctl list-units 'php*-fpm*' lists the running ones. If you use HestiaCP or another panel with several PHP versions, restart FPM for each of them.
Distribution packages usually follow a few days later and sometimes keep the old version number with the fixes backported. Check your distribution’s security advisory.
5. NoisyTCP: remotely crashing the Linux kernel (CVE-2026-99288)
Status: details published on 15 November, no exploitation seen, fixes not yet in every distribution.
How NoisyTCP works
A race in the kernel’s TCP connection lookup ends in a NULL pointer dereference and a kernel panic. The attacker needs no credentials: access to any open TCP port — a web server, SSH, mail — is enough. The result is the whole server going down, with every container on it. Nobody has demonstrated code execution, but letting an attacker reboot your server at will is a serious availability problem.
Note that this is not the same bug as CVE-2026-98432, fixed in October. If you installed October’s fix, it does not protect you against NoisyTCP.
Who is affected by NoisyTCP
Linux kernels from 4.4 onward without the fix, on servers that spread incoming packets across several receive queues or have RPS enabled. That is a common setup on physical servers and cloud VMs.
Check whether the condition holds (replace eth0 with your interface name):
ethtool -l eth0
grep . /sys/class/net/eth0/queues/rx-*/rps_cpus
A Combined or RX value above 1 under the current settings, or a non-zero rps_cpus mask, means the server meets the conditions for the attack.
How to fix NoisyTCP
The fix landed in kernels 6.12.117, 6.18.63 and 7.3.4. Distribution status as of 16 November:
| Distribution | Status |
|---|---|
| Debian 12 and 13 | fixed kernels released |
| Oracle Linux (UEK) | fixed kernels released |
| RHEL 8, 9, 10 and rebuilds | no fix yet |
| Ubuntu | still evaluating |
| RHEL 7, CentOS 7 | not affected |
On Debian, update the kernel and reboot — the new kernel does not take effect without a reboot:
apt update && apt full-upgrade
reboot
After the reboot, check the kernel version with uname -r.
If no fixed kernel is available yet
There are two stopgaps, and both cost something:
- A single receive queue. Lowers network performance on busy servers:
ethtool -L eth0 combined 1
- Disabling SYN cookies. Removes the condition for the race but weakens SYN flood protection:
sysctl -w net.ipv4.tcp_syncookies=0
Neither setting survives a reboot — which is arguably a plus: once a fixed kernel is out, you reboot and return to normal settings.
6. Java4Shell: code execution through the Base64 decoder (CVE-2026-114938)
Status: fixed on 17 November, no reports of exploitation, CISA urges immediate patching.
How Java4Shell works
A memory corruption bug in the Base64 decoder of the HotSpot virtual machine on x86-64. Crafted Base64 input can crash the JVM and allow unauthenticated remote code execution. The name is a nod to the infamous Log4Shell: a great many Java applications decode Base64, often straight from user input.
Who is affected by Java4Shell
Current Java 17, 21, 25 and 27 releases on x86-64. Sources disagree on Java 8 and 11 — check your Java vendor’s advisory.
How to fix Java4Shell
Find every installed Java virtual machine. There are often several: the system one, ones bundled with applications, ones in containers:
java -version
find / -xdev -type f -name java -perm -u+x 2>/dev/null
Oracle’s fixed builds are 17.0.21.1, 21.0.13.1, 25.0.5.1 and 27.0.1.1. Vendor status as of 18 November:
| Vendor | Status |
|---|---|
| Ubuntu | fixed OpenJDK 17, 21, 25 and 27 packages released on 18 November |
| Amazon Corretto | fixed builds released |
| Eclipse Temurin | 21, 25 and 27 fixed |
| Debian | fixed only in unstable |
| Red Hat | no fixed packages yet |
After updating, restart your Java applications — a JVM that is already running keeps using the old code. Remember to rebuild Java-based container images.
If you can’t update, CISA advises restricting and validating Base64 input at the application boundary, enforcing size limits and limiting unauthenticated access.
Also in CISA’s exploited vulnerabilities catalog this week
CISA added these to its catalog as already used in attacks. If you run any of them, patch them first.
| Date | CVE | Product | What to do |
|---|---|---|---|
| 13 November | CVE-2026-114713 | UnrealIRCd | update per the developers’ advisory |
| 13 November | CVE-2026-123039 | Check Point, several products | install the vendor fix |
| 17 November | CVE-2026-115731 | Puppet Enterprise | update to 2023.8.13, 2025.12.1 or 2026.0.1 and restart Puppet Server |
| 18 November | CVE-2026-40217 | LiteLLM | update to 1.83.14 or newer |
| 18 November | CVE-2026-90436 | Apache Axis | update per the developers’ advisory |
| 18 November | CVE-2024-10441 | Synology DSM | install the DSM update |
For Puppet Enterprise, disable unverified certificate autosigning until you update — it is cited as one of the conditions for the attack. LiteLLM is a popular gateway to language models; if you deployed it on your own server, check its version first.
If you run a Windows server
Microsoft’s November update came out on 10 November, a few days before this week, but for a Windows VPS it remains the top priority:
- CVE-2026-124194 — an Outlook spoofing flaw, already exploited: attackers steal NTLM hashes.
- CVE-2026-119958 — code execution through Windows LDAP, CVSS 9.8.
- CVE-2026-88218 — code execution in the Windows DNS server, CVSS 9.8.
- CVE-2026-88243 — code execution in the Windows DHCP server, CVSS 9.8.
- CVE-2026-116945 — code execution in Microsoft SQL Server, CVSS 9.8.
- CVE-2026-118207 — code execution in Exchange Server: sending an email to the organisation’s server is enough.
Install the updates through Settings → Windows Update or from PowerShell and reboot. The DNS, DHCP and LDAP roles are a reason to update the same day.
How not to miss fixes like these
Most of this week’s vulnerabilities are closed by an ordinary package update. To stop depending on running it by hand, enable automatic installation of security fixes:
apt install unattended-upgrades && dpkg-reconfigure -plow unattended-upgrades # Debian, Ubuntu
dnf install dnf-automatic && systemctl enable --now dnf-automatic-install.timer # AlmaLinux, Rocky Linux
A few more rules that limit the damage from vulnerabilities like these:
- Restart services after updating. A program started before the update keeps running the old code. The
needrestartutility shows which services need a restart. - Reboot after a kernel update. Otherwise the fix is not applied.
- Don’t forget containers. Updating the system does not update Docker images — they must be rebuilt. Watchtower helps.
- Disable SSH password logins and restrict access: change the port, use Fail2Ban and two-factor authentication. SSHole is only dangerous where passwords are allowed.
- Back up before updating. If an update breaks something or the server was already compromised, a copy on separate storage lets you recover quickly.
FAQ
My server runs Ubuntu. Does SSHole affect me?
No. The flaw only works on musl-based systems without PAM — mainly Alpine Linux. Ubuntu, Debian and AlmaLinux use glibc and PAM. But if the server runs Alpine-based containers with sshd inside, check them.
Do I need to reboot the server?
Only for the kernel update (NoisyTCP). For OpenSSH, nginx, Apache, PHP and Java, restarting the service or application is enough.
A scanner still reports the vulnerability after I updated. Why?
Distributions often backport fixes into the old package version without changing its number. A scanner that compares version numbers alone sees an “old” program. Check your distribution’s security advisory: it lists the fixed package version.
What if there is no fix for my distribution yet?
Apply the stopgaps from the relevant section: disable HTTP/2 in nginx or Apache, disable SSH password logins, use a single receive queue for NoisyTCP. And watch your distribution’s advisory so you can update as soon as the package is out.
How do I hear about vulnerabilities like these in time?
Subscribe to the security mailing lists of your distribution and the software you run, and follow CISA’s Known Exploited Vulnerabilities catalog — it collects vulnerabilities that are already being used in attacks.


