CISA added CVE-2026-46300 to the Known Exploited Vulnerabilities catalog on 28 October, with a due date of 31 October and the forensic triage flag set 1. The CVE record for the same flaw carries a date of 23 May 2. CISA listed the flaw five months after the fix shipped, and tied the listing to evidence of active exploitation 3. The first thing I pulled was a host list.

I spent the first ten minutes on the wrong question. TuxCare described the patch as two lines in skb_try_coalesce(), and rereading them tells me nothing about which machines took the release 4. The questions that matter are which hosts sit below the fixed build on their branch, and what happens to the people who own them if the fallback mitigation goes out as a one line instruction.

The marker ESP trusts

skb_try_coalesce() attaches paged fragments from one sk_buff to another. When the source buffer carries SKBFL_SHARED_FRAG, the destination ends up holding the same page-cache backed fragments, and the coalescing path drops the marker 4. ESP input trusts that marker. Before it decides an uncloned nonlinear skb can skip skb_cow_data(), it calls skb_has_shared_frag(). A buffer that lost the marker answers no, the kernel skips the copy, and AES-GCM decryption writes into page-cache pages of a file the attacker picked 4.

The public exploit rewrites the cached copy of /usr/bin/su with an ELF stub that sets uid 0 and execs a shell. The copy on disk keeps its bytes and its hash, and the modified page stays in cache until eviction or reboot 4. That PoC needs CONFIG_INET_ESPINTCP to reach the bug, so a kernel built without it blocks that path, though TuxCare says the underlying skbuff defect may have other paths into it 4. Rapid7 ships a Metasploit module for the flaw 5. NVD scores it 7.8 on a CVSS 3.1 vector that starts at AV:L 6, which is accurate about the local account requirement and quiet about how cheap a local account is on a host that runs other people’s containers.

The chain in CISA’s short description

CISA writes the chain into the catalog entry. Its short description says the kernel flaw can be chained with CVE-2026-109618 and CVE-2026-109619, the two PHP remote code execution flaws from October, to plant PHP web shells and run commands as root 1. I wrote about that PHP pair when the details leaked ahead of the vendor release. Read side by side, the two entries describe a sequence I expect: a web-facing bug yields a shell as the web user, and a local kernel bug finishes the climb to root.

The entry sets forensic triage to Yes, and the directive behind the catalog asks agencies to establish whether an attacker compromised a system before the patch went on 3. The patching ticket does not answer that. A fixed kernel covers what happens after the patch and leaves open what ran on the host before it. The catalog binds federal agencies. CISA encourages everyone else to adopt risk-based vulnerability management 3.

Why I will not send the module block as a line item

Published guidance blocks esp4 and esp6 through modprobe and unloads them 7. On a host that carries IPsec, that is an outage wearing a security label. Red Hat puts a warning on the same page: blocking the modules breaks IPsec functionality 7. TuxCare names strongSwan and Libreswan as the deployments that stop working and notes that WireGuard is unaffected 4.

There is a second problem with the module block. rmmod refuses to unload a module in use, so a host that ran the command can have ESP loaded after the command returns 7. The ticket says mitigated, the module list says otherwise, and nobody reconciles the two until an audit.

Red Hat offers a targeted alternative: disable unprivileged user namespaces, which removes the namespace the exploit needs. The same page warns that rootless containers, sandboxed browsers and Flatpak depend on that setting 7. Both paths cost something. Neither one belongs in a bullet list with no owner named next to it.

What goes in the message

I send two facts and two questions together. The first fact is the fixed build on each branch, because “kernel updated” in a ticket names no release. The CVE record lists 5.10.257, 5.15.208, 6.1.174, 6.6.141, 6.12.91, 6.18.33 and 7.0.10 as the first unaffected builds on their branches, with 7.1.0 and later unaffected 2. The fix went into seven stable branches, so the running version is what settles a host.

The first question is whether ESP is loaded. The second is whether anything depends on it. The checks are the ones the vendors publish 8:

uname -r
lsmod | grep -E 'esp4|esp6'
ip xfrm state
grep -qE '^(esp4|esp6) ' /proc/modules && echo "modules loaded" || echo "modules not loaded"

Red Hat’s check is the lsmod line, which tells you whether IPsec is in play before anyone blocks a module 7. Ubuntu’s procedure adds a step that catches people: block the modules and regenerate the initramfs, or an image built earlier loads them at boot 8.

Detection where file hashes stay correct

This bug class defeats the tool most teams use for the question. AIDE and Tripwire compare hashes on disk, and the write lands in page cache, so the hashes stay correct while the running binary changes behavior 4. The signals that survive are behavioral: unshare(CLONE_NEWUSER | CLONE_NEWNET) from a process with no reason to create a namespace, NETLINK_XFRM traffic from a workload with no VPN role, AF_ALG crypto calls in a tight loop, and a splice from a regular file into a TCP socket followed by a ULP switch on that socket 4. Each one alone is noise on a busy host, and together they trace the exploit’s own sequence.

Clearing page cache removes the copy an attacker left in memory and says nothing about what else they did 4. TuxCare treats a successful exploit as full host compromise, which is the assumption that holds when the write primitive is this repeatable 4.

I send the branch table, the tunnel question and the triage question as one note, because the teams that own these hosts can answer the version and the tunnel in the same reply. The fix shipped in May. The work left is an inventory that names hosts and a decision about who can take the reboot.


  1. Cybersecurity and Infrastructure Security Agency. (2026, October 28). Known exploited vulnerabilities catalog (catalog version 2026.10.28). https://www.cisa.gov/sites/default/files/feeds/known_exploited_vulnerabilities.json ↩︎ ↩︎

  2. CVE Program. (2026, May 23). CVE-2026-46300 record [JSON]. MITRE. https://cveawg.mitre.org/api/cve/CVE-2026-46300 ↩︎ ↩︎

  3. Cybersecurity and Infrastructure Security Agency. (2026, October 28). CISA adds two known exploited vulnerabilities to catalog. CISA. https://www.cisa.gov/news-events/alerts/2026/10/28/cisa-adds-two-known-exploited-vulnerabilities-catalog ↩︎ ↩︎ ↩︎

  4. TuxCare. (2026, May 13). Fragnesia (CVE-2026-46300): a new Linux kernel LPE in the Dirty Frag family. TuxCare. https://tuxcare.com/blog/fragnesia-cve-2026-46300/ ↩︎ ↩︎ ↩︎ ↩︎ ↩︎ ↩︎ ↩︎ ↩︎ ↩︎ ↩︎

  5. Rapid7. (n.d.). Fragnesia LPE (CVE-2026-46300) [Metasploit module]. Rapid7 Vulnerability and Exploit Database. https://www.rapid7.com/db/modules/exploit/linux/local/cve_2026_46300_fragnesia/ ↩︎

  6. National Vulnerability Database. (2026, May 23). CVE-2026-46300 detail. National Institute of Standards and Technology. https://nvd.nist.gov/vuln/detail/CVE-2026-46300 ↩︎

  7. Red Hat. (2026). RHSB-2026-003: networking subsystem privilege escalation, Linux kernel (CVE-2026-43284, CVE-2026-43500, CVE-2026-46300). Red Hat Customer Portal. https://access.redhat.com/security/vulnerabilities/RHSB-2026-003 ↩︎ ↩︎ ↩︎ ↩︎ ↩︎

  8. Ubuntu. (2026). Fragnesia Linux kernel local privilege escalation vulnerability mitigations. Ubuntu Blog. https://ubuntu.com/blog/fragnesia-linux-vulnerability-fixes-available ↩︎ ↩︎