CVE-2026-46300: Linux Kernel, is it exploitable where you run it
Code called: openReached: yesControl in front: noExploitation known: yesWalk it through the four questions
What the flaw is
Linux keeps file contents in memory in a page cache. Network code marks packet buffers that point into those pages as shared, so that nobody modifies them in place. The function skb_try_coalesce() merges receive buffers and, before the fix, dropped that marker when it moved the page fragments into the merged buffer.
The IPsec ESP input code trusts the marker. When it sees none, it decrypts in place and skips the copy it would otherwise make. The decrypted bytes then land in the page cache of a file that was only read, for example a setuid binary. The file on disk is unchanged, but the next program run from the cache uses the altered copy.
A low-privilege local user can use this to gain root. A public proof of concept named Fragnesia appeared on 13 May 2026 and belongs to the Dirty Frag class of page-cache bugs. This page gives no exploit steps.
Are you exposed: each condition and how to check it
| Condition for exposure | How to check it on your system |
|---|---|
| the running kernel lacks the fix or a distribution backport | Run uname -r. Compare it with the versions table; kernels older than 5.6 (Ubuntu 18.04 and 20.04 kernels) lack espintcp, which the public proofs of concept need, so they cannot run them (Ubuntu). On RPM systems run rpm -q --changelog kernel | grep CVE-2026-46300. On Debian or Ubuntu run apt changelog linux-image-$(uname -r) | grep CVE-2026-46300 or look up the package in the Debian and Ubuntu trackers. |
| a local user can run code, and esp4 or esp6 can load with unprivileged namespaces allowed | Foothold: list who can log in or run code with getent passwd, last, web application accounts, CI runners and container users; without a local foothold the flaw is not reachable (CVSS AV:L, PR:L). ESP modules: Red Hat documents lsmod | grep -E 'esp4|esp6'; no output means IPsec is not in active use, but it does not show that the modules cannot load later. Run modinfo esp4 to see whether the module exists for the running kernel, and cat /etc/modprobe.d/dirtyfrag.conf to see whether the Red Hat block is in place. Namespaces: sysctl user.max_user_namespaces; 0 blocks creation of new user namespaces (Red Hat). Red Hat documents the module block and the namespace setting as separate mitigations, and either one stops the path, so a host is exposed only while both are open. |
Affected and fixed versions
| Product or line | Affected | First fixed |
|---|---|---|
| Linux 3.9 to 5.10 | 3.9 through 5.10.257 (NVD) | No fixed upstream 5.10 stable release through 5.10.274. Debian bullseye package 5.10.257-1 (DLA-4606-1) is fixed (see note) |
| Linux 5.11 to 5.15 | 5.11 before 5.15.208 | 5.15.208 |
| Linux 5.16 to 6.1 | 5.16 before 6.1.174 | 6.1.174 |
| Linux 6.2 to 6.6 | 6.2 before 6.6.141 | 6.6.141 |
| Linux 6.7 to 6.12 | 6.7 before 6.12.91 | 6.12.91 |
| Linux 6.13 to 6.18 | 6.13 before 6.18.33 | 6.18.33 |
| Linux 6.19 to 7.0 | 6.19 before 7.0.10 | 7.0.10 |
| Linux 7.1 candidates | 7.1-rc1 to 7.1-rc4 | 7.1-rc5 (mainline) |
| RHEL 10 / 10.0 EUS | All earlier kernels | kernel-6.12.0-211.16.1.el10_2 (RHSA-2026:19569, 20 May 2026) / kernel-6.12.0-55.75.1.el10_0 (RHSA-2026:20299) |
| RHEL 9 / 9.6 EUS / 9.4 EUS | All earlier kernels | kernel-5.14.0-687.10.1.el9_8 (RHSA-2026:19568) / 5.14.0-570.116.1.el9_6 (RHSA-2026:20129) / 5.14.0-427.126.1.el9_4 (RHSA-2026:20054) |
| RHEL 8 / 8.8 / 8.6 / 8.4 extended support | All earlier kernels | kernel-4.18.0-553.125.1.el8_10 (RHSA-2026:19666) / 4.18.0-477.143.1.el8_8 (RHSA-2026:19521) / 4.18.0-372.193.1.el8_6 (RHSA-2026:20051) / 4.18.0-305.192.1.el8_4 (RHSA-2026:20130) |
| RHEL 6 and 7 | Not affected (Red Hat) | Not applicable |
| OpenShift Container Platform 4.12 to 4.21 | RHCOS images before the errata | One RHSA per minor version, for example 4.18: RHSA-2026:21656 (3 Jun 2026) and 4.21: RHSA-2026:23240 (11 Jun 2026) |
| Debian | Earlier source packages | bullseye 5.10.257-1 (DLA-4606-1), bookworm 6.1.174-1 (DSA-6306-1), trixie 6.12.90-1 (DSA-6295-1), unstable 7.0.9-1; later builds in the tracker on 19 Nov 2026: bookworm (security) 6.1.192-2, trixie (security) 6.12.117-1 |
| Ubuntu linux package | Earlier kernels | focal 5.4.0-233.253, jammy 5.15.0-181.191, noble 6.8.0-124.124, questing 6.17.0-35.35, resolute 7.0.0-22.22 |
| Siemens SIMATIC S7-1500 CPU 1518(F)-4 PN/DP MFP and SIPLUS variants (a separate product, not a Linux distribution: read the CPU firmware version in TIA Portal online and diagnostics or on the CPU web server) | V3.1.5 (no fix) and V3.1.6 | V3.1.7 |
If the fix cannot go in yet
- Install the fixed kernel from your distribution and reboot. Cost: a reboot. Red Hat and Siemens also ship the fix, and Red Hat offers kpatch for some streams, which avoids the reboot.
- Block the IPsec ESP modules (Red Hat's broader method). Write install lines for esp4 and esp6 that run /bin/false into /etc/modprobe.d/dirtyfrag.conf and run rmmod esp4 esp6. Cost: IPsec stops working after the next reboot. Revert by deleting the file. Check first with lsmod | grep -E 'esp4|esp6': if neither module is loaded, IPsec is not in use and blocking is safe.
- Disable unprivileged user namespaces (Red Hat's targeted method when IPsec must stay). Set user.max_user_namespaces=0 in /etc/sysctl.d/dirtyfrag.conf and run sysctl --system. Cost: rootless containers, sandboxed browsers and Flatpak can stop working.
- Limit local access (Red Hat). Turn off unneeded SSH access, keep SELinux in enforcing mode, use the default OpenShift Security Context Constraints, run workloads as non-root and restrict oc debug to trusted administrators. Red Hat notes this list is not exhaustive and closing one access path does not close the others.
- Keep the module mitigation in place after patching for now. The researcher who posted the later Dirty Frag fixes advised on 22 May 2026 that more variants in the ESP in-place path are expected.
- On the Siemens CPU 1518(F)-4 PN/DP MFP, update to V3.1.7. Siemens also advises building and running only applications from trusted sources and limiting access to the interactive shell of the additional GNU/Linux subsystem to trusted staff.
What to look for
- CISA's KEV entry says the flaw was chained with CVE-2026-109618 and CVE-2026-109619 to implant PHP webshells and run commands as root, and marks forensic triage as required. CISA publishes no file names, hashes or addresses in the entry.
- Reads go through the page cache, so a package verification run before a reboot or cache drop can show the change: run
rpm -V PACKAGE(RPM systems) ordebsums -s PACKAGE(Debian and Ubuntu) for the package that owns each setuid binary, for examplerpm -qf /usr/bin/su, and compare asha256sumof the running binary with the one in a clean copy of the package. A mismatch means the cached copy was altered. Dropping the cache withecho 3 > /proc/sys/vm/drop_cachesor rebooting restores the on-disk bytes and erases that evidence, so record the hashes first. Red Hat, Ubuntu and Siemens publish no indicators of compromise for this CVE. - Review unexpected PHP files in web roots and new root-level processes on hosts where an untrusted local user or web application account could have run code.
The earliest dated mark of exploitation is CISA's, on 14 May 2026, 1 day after disclosure; exploitation may have begun earlier.
Exploitation reported so far
The 14 May 2026 date is CISA's own assessment date on the SSVC entry, which NVD published nine days later, on 23 May 2026. The entry gives no source for the finding, so it is not a dated incident report. CISA's first public statement of use is the KEV listing. The KEV entry says the flaw can be chained with CVE-2026-109618 and CVE-2026-109619 to implant PHP webshells and run commands as root, and lists ransomware use as unknown.
A working public proof of concept has existed since 13 May 2026. Neither CISA nor Red Hat, Ubuntu or Siemens publishes the actor, victims or indicators.
Why runtime reachability settles it
Exposure needs a local foothold plus a kernel that can load the ESP code and lets the user create namespaces. Those are facts about the running host, so checking the kernel, the loaded modules and the namespace setting settles it, and a severity score or the version alone cannot. Known-exploited status tells you to act now, but not which hosts are reachable. See runtime reachability for what the method proves and misses.
Related pages
Sources
- CVE-2026-46300: NVD record, read 19 Nov 2026
- CVE-2026-46300: CISA KEV feed (catalog version 2026.11.18), read 19 Nov 2026
- CVE-2026-46300: vendor advisory (Red Hat CVE page), read 19 Nov 2026
- CVE-2026-46300: vendor advisory data (Red Hat Security Data API: errata and package versions), read 19 Nov 2026
- CVE-2026-46300: vendor advisory (Red Hat security bulletin RHSB-2026-003, mitigation), read 19 Nov 2026
- CVE-2026-46300: vendor advisory (Siemens SSA-019113, firmware V3.1.6), read 19 Nov 2026
- CVE-2026-46300: vendor advisory (Siemens SSA-082556, firmware V3.1.5), read 19 Nov 2026
- CVE-2026-46300: oss-security disclosure and CVE assignment, read 19 Nov 2026
- CVE-2026-46300: oss-security post on the Dirty Frag variant fixes, read 19 Nov 2026
- CVE-2026-46300: oss-security reply in the same thread, read 19 Nov 2026
- CVE-2026-46300: oss-security reply in the same thread, read 19 Nov 2026
- CVE-2026-46300: upstream fix commit (GitHub mirror of the Linux kernel), read 19 Nov 2026
- CVE-2026-46300: kernel.org stable change log 6.1.174 (also read for 5.15.208, 6.6.141, 6.12.91, 6.18.33, 7.0.10 and 5.10.258 to 5.10.274), read 19 Nov 2026
- CVE-2026-46300: Debian security tracker, read 19 Nov 2026
- CVE-2026-46300: Ubuntu CVE tracker data, read 19 Nov 2026
- CVE-2026-46300: GitHub advisory database entry GHSA-47jg-vqrv-5f8v, read 19 Nov 2026
Last reviewed 19 Nov 2026.