Skip to content
Exploit for CVE-2026

Exploit for CVE-2026

Sploitus • October 30, 2026

**Attackable over the network, with no login and no user action.** The bug is in the kernel code that handles every incoming IPv4 TCP connection.

This guide shows how to check a server, protect it until a patched kernel is available, and clean up afterwards. The commands below were used on Ubuntu 22.04 and 24.04 servers.

> Status in this document is as of **2026-10-30**. Patch status changes quickly, so always check your distribution's own security tracker.

- Upstream fix: *"ipv4: free inet_opt and ireq_opt after an RCU grace period"*, commit [`e3f33b0a1d89`](

**In plain words:** a connection request (SYN) can carry IPv4 **options**, rarely used extras in the packet header such as *Record Route* or *Timestamp*. When one arrives, the kernel keeps a copy of those options. If the attacker resends the SYN at just the right moment, one CPU can free that copy while another is still using it to build the . That is a *use-after-free* in `ip_options_build()`, and it can crash the server or corrupt kernel memory.

A second trigger exists when the server migrates pending connection requests between sockets (`net.ipv4.tcp_migrate_req=1`, or a `BPF_SK_REUSEPORT_SELECT_OR_MIGRATE` program). That setting is off by default.

| CVSS, kernel CNA | **9.8 Critical**: network, no authentication, no user interaction |

| CVSS, Red Hat | 8.1, because Red Hat considers the race harder to hit |

| IPv6 | Not affected by the main issue |

| Exploitation | None reported in the wild, and not in the CISA KEV catalog. A remote root exploit against servers with a **multipath TCP (MPTCP) listener** has been *reported* but not published. |

**Who is exposed:** any server with a TCP port reachable over IPv4, whether SSH, HTTP, HTTPS or anything else.

| Upstream kernel | ✅ Fixed | 5.10.272, 5.15.223, 6.1.190, 6.6.159, 6.12.113, 6.18.56, 7.2.10, 7.3 |

| Debian 12 / 13 | ✅ Fixed | `6.1.190-1` (or `linux-6.12 6.12.113-1~deb12u1`) / `6.12.113-1` |

| Ubuntu 26.10 / 26.04 / 24.04 / 22.04 | ❌ Fix in progress, not released | Planned: `7.3.0-10.10` / `7.0.0-43.43` / `6.8.0-149.149` / `5.15.0-203.213` |

| Ubuntu 20.04 / 18.04 / 16.04 | ❌ Vulnerable | Only through Ubuntu Pro (ESM), if at all |

| RHEL 8 / 9 / 10 | ❌ Reported as "fix deferred" | None |

| Oracle Linux (UEK) | ✅ Fixed | ELSA-2026-500397 / 500398 / 500399 |

| Amazon Linux 2023 | ⚠️ Partly fixed | `kernel6.18` and `kernel6.12` fixed; default kernel and AL2 pending |

| SUSE | ⚠️ Partly fixed | openSUSE Leap 16.0 fixed; SLES in progress |

Ubuntu cloud kernels (`-aws`, `-azure`, `-gcp`, `-oracle`) have their own version numbers. Distribution kernels backport fixes, so don't compare a distro version like `5.15.0-xxx` with the upstream numbers.

This only reads and reports. It doesn't install, change or restart anything. Paste it into a terminal on the server:

sudo bash -s /dev/null | sed 's|/boot/vmlinuz-| |'

[ -f /var/run/reboot-required ] && echo " !! reboot required"

echo "== Is the fix in the RUNNING kernel's changelog?"

dpkg-query -W -f=' package version: ${Version}\n' "linux-image-$KR" 2>/dev/null

zgrep -Eqs "$CVE|$FIX" /usr/ /doc/linux-*"$KR"*/changelog.Debian.gz \

&& echo " PATCHED" || echo " NOT FOUND -> likely VULNERABLE"

command -v pro >/dev/null && pro fix "$CVE" --dry-run /dev/null; then

PKG=$(rpm -qf "/lib/modules/$KR/vmlinuz" "/usr/lib/modules/$KR/vmlinuz" "/boot/vmlinuz-$KR" 2>/dev/null | grep -v 'not owned' | head -1)

rpm -q --changelog "$PKG" 2>/dev/null | grep -Eq "$CVE|$FIX" \

&& echo " PATCHED" || echo " NOT FOUND -> likely VULNERABLE"

if command -v zypper >/dev/null; then zypper -n -q lp -a --cve="$CVE" /dev/null; then dnf -q updateinfo list --all --cve="$CVE" /dev/null; then yum -q updateinfo list --cve="$CVE" /dev/null | sed 's/^/ /'

M=$(ss -lnMH 2>/dev/null); echo " MPTCP listeners (reported root-exploit path): ${M:-none}"

echo " TCP listeners not on loopback (attack surface):"

ss -ltnH | awk '$4 !~ /^(127\.|\[::1\])/ {print " "$4}' | sort -u

Paste it whole: the final `EOF` must be on its own line with no leading spaces.

| `PATCHED` | The running kernel already has the fix | You're done |

| `NOT FOUND -> likely VULNERABLE` | No fix in the running kernel | Step 2 |

| `!! reboot required`, or a newer kernel than the running one is listed | A newer kernel is installed but not running yet | Step 3 |

| `A fix is coming soon` (Ubuntu) | Ubuntu hasn't shipped the fix yet | Step 2 now, Step 5 later |

| `MPTCP listeners: none` | The reported root-exploit path doesn't apply right now | — |

| A database or cache port such as `*:3306`, `0.0.0.0:6379` or `0.0.0.0:11211` | It listens on all interfaces | Make sure the firewall blocks it (see the end of Step 4) |

The rule below **drops IPv4 TCP packets that carry IP options**, which is what an attacker needs to trigger the bug. Normal clients (browsers, SSH, phones, APIs, load balancers) never send IP options, so real traffic isn't affected.

**First check which firewall tool the server uses:**

systemctl is-active ufw nftables firewalld

If `nftables` or `firewalld` is `active`, reloading that firewall can wipe this rule. Use the nftables variant at the end of this step, or adapt it to your setup.

**Install the stopgap.** It's a small systemd unit, so the rule comes back after every reboot:

sudo tee /etc/systemd/system/cve-2026-98432.service >/dev/null >24&0xF=6:15" -m -- CVE-2026-98432 -j DROP

ExecStop=/usr/sbin/iptables -t raw -D PREROUTING -p tcp -m u32 --u32 "0>>24&0xF=6:15" -m -- CVE-2026-98432 -j DROP

sudo systemctl daemon-reload && sudo systemctl enable --now cve-2026-98432

The last command should list a `DROP` rule with the `CVE-2026-98432`. The `pkts` column counts packets it blocked; it should stay at 0 or close to it.

The `u32` test `0>>24&0xF=6:15` reads the IPv4 header-length field. A normal header is 5 × 4 = 20 bytes. Anything longer means options are attached, and that packet is dropped. The rule sits in the `raw` table, which ufw doesn't manage, so your existing firewall rules are untouched.

**Check SSH before going further:** keep your current session open and log in from a second terminal. Then load your website.

"Does this block connecting by IP address?" No.

Every packet has a header that carries the source and destination IP, and that's how you connect. The rule ignores the addresses. It only looks for the rarely used *options* extension inside the header, which makes the header longer than 20 bytes. Connecting with `ssh [email protected]` or ` works exactly as before.

Optional: turn off MPTCP if nothing uses it

If Step 1 showed `MPTCP listeners: none`, you can turn MPTCP off. This is harmless when nothing uses it:

echo 'net.mptcp.enabled = 0' | sudo tee /etc/sysctl.d/99-disable-mptcp.conf

sudo sysctl -p /etc/sysctl.d/99-disable-mptcp.conf

nftables variant for systems without iptables or the u32 match (not tested here)

sudo nft add chain ip cve98432 pre '{ type filter hook prerouting priority -300; policy accept; }'

sudo nft add rule ip cve98432 pre ip protocol tcp ip hdrlength gt 5 counter drop

# undo: sudo nft delete table ip cve98432

This rule does **not** survive a reboot. Add it to your persistent nftables config if you need it to.

### Step 3: Reboot (if Step 1 said "reboot required")

The reboot doesn't fix this CVE on Ubuntu yet, because no fixed kernel exists. It does load the newest kernel that's already installed, which usually carries months of other security fixes.

- Confirm SSH works from a second session (Step 2).

- Pick a moment when 1–2 minutes of downtime is acceptable.

- **Database primary and replica:** reboot them one at a time, never both at once.

- **Server up for a long time** (months or years): take a snapshot in your provider's panel, have its web console ready, and check for things someone started by hand. Those won't come back after the reboot:

systemctl list-units --type=service --state=running --no-legend | awk '{print $1}' | sort > /root/services-before.txt

> **No newer kernel installed, but the kernel is old?** Automatic updates may be off. Check with

> `sudo apt update && apt list --upgradable 2>/dev/null | grep -E '^linux-(generic|image)'`.

> If kernels are pending, install the updates and keep your existing config files:

> `sudo apt -o Dpkg::Options::="--force-confdef" -o Dpkg::Options::="--force-confold" upgrade`

> Then reboot. Turn automatic updates back on with:

> `printf 'APT::Periodic::Update-Package-Lists "1";\nAPT::Periodic::Unattended-Upgrade "1";\n' | sudo tee /etc/apt/apt.conf.d/20auto-upgrades`

| `ufw status` | `Default: deny (incoming)`, with only the ports you actually serve (usually 22, 80, 443) allowed from Anywhere |

| `uname -r` | The newer kernel version from Step 1 |

| `iptables …` | The `CVE-2026-98432` DROP rule is present |

| `systemctl --failed` | `0 loaded units listed` |

If you saved `services-before.txt` in Step 3, compare it with what's running now. Lines starting with `` files after a site was re-created. The reboot just made it visible. To see the error:

sudo journalctl -u logrotate --no-pager -n 20 | grep 'error:'

To move out files that are byte-for-byte copies of another file (keeping the first copy, which is the one logrotate already uses), run as root:

cd /etc/logrotate.d && md5sum forge-* | sort -k1,1 -k2,2 | awk 'seen[$1]++ {print $2}' | while read f; do mv "$f" "/root/$f.disabled" && echo "moved $f"; done; cd ~

logrotate -d /etc/logrotate.conf 2>&1 | grep 'error:' # should print nothing

systemctl start logrotate; systemctl --failed

If errors remain, the files overlap without being identical. Compare them by hand before moving anything.

**`systemd-networkd-wait-online.service` failed.** Run `networkctl`. If every link shows `unmanaged`, networkd isn't managing the network, so this service has nothing to wait for and gives up after 2 minutes:

sudo systemctl disable systemd-networkd-wait-online.service

sudo systemctl reset-failed systemd-networkd-wait-online.service

If some links *are* managed, an unused second network card is the usual cause. Mark it `optional: true` in netplan instead.

**`policy ACCEPT 0 packets` on Ubuntu 22.04.** This appears to be a display quirk of the older iptables version. What matters is that the DROP rule is listed.

**Ubuntu 20.04 or older, `This machine is not attached to an Ubuntu Pro subscription`.** Free security support for these releases has ended. Without Ubuntu Pro (ESM) the server gets no kernel or package fixes at all. Attach it with `sudo pro attach ` (free personal tier for up to 5 machines), or move the workload to a supported release.

These are community notes, not an official advisory. The stopgap is **not** endorsed by any vendor. It blocks the trigger described in the CVE, but nobody has said publicly whether it also covers the reported MPTCP exploit. Always install the vendor's fixed kernel when it ships. Use at your own risk.

- [Ubuntu: CVE-2026-98432](

- [Red Hat: CVE-2026-98432](

- [Debian security tracker](

- [Upstream commit e3f33b0a1d89](

- [Cyber Kendra: patch status across distributions](

- [Upwind threat post](

Extracted Entities

Attack Types (1)

CWE Weaknesses (1)