<?xml version="1.0" encoding="utf-8"?><feed xmlns="http://www.w3.org/2005/Atom" ><generator uri="https://jekyllrb.com/" version="4.4.1">Jekyll</generator><link href="https://binary0101devil.pages.dev/feed.xml" rel="self" type="application/atom+xml" /><link href="https://binary0101devil.pages.dev/" rel="alternate" type="text/html" /><updated>2026-11-21T00:30:57+05:30</updated><id>https://binary0101devil.pages.dev/feed.xml</id><title type="html">Aniket Tomar · Binary0101Devil</title><subtitle>Aniket Tomar (Binary0101Devil): Senior Security Consultant, Red Team &amp; Offensive Security, OSINT and Dark Web researcher. Portfolio and daily security blog.</subtitle><author><name>Aniket Tomar</name></author><entry><title type="html">CVE-2026-129448: unauthenticated RCE in Apache Roller, exploited in the wild</title><link href="https://binary0101devil.pages.dev/blog/2026/11/20/cve-2026-129448-apache-roller-rce/" rel="alternate" type="text/html" title="CVE-2026-129448: unauthenticated RCE in Apache Roller, exploited in the wild" /><published>2026-11-20T00:00:00+05:30</published><updated>2026-11-20T00:00:00+05:30</updated><id>https://binary0101devil.pages.dev/blog/2026/11/20/cve-2026-129448-apache-roller-rce</id><content type="html" xml:base="https://binary0101devil.pages.dev/blog/2026/11/20/cve-2026-129448-apache-roller-rce/"><![CDATA[<p>The first of three critical RCE flaws I’m covering this week, and the only one already under active exploitation. If you run Apache Roller, this is a drop-everything patch. For the other two, see the <a href="/blog/2026/11/18/cve-2026-119618-apache-http2-rce/">Apache HTTP/2 race condition</a> and the <a href="/blog/2026/11/16/cve-2026-98432-linux-kernel-tcp-rce/">Linux kernel TCP use-after-free</a>; for the wider picture, the <a href="/blog/2026/11/14/top-rce-cves-2026/">2026 RCE field guide</a> lists ten.</p>

<blockquote>
  <p><strong>Scope and ethics.</strong> The steps below are exposure <strong>detection</strong> and remediation for systems you own or are authorized to assess — not a weaponized exploit. Working PoCs live behind the references, for use in a lab you control.</p>
</blockquote>

<h2 id="description">Description</h2>

<p>Apache Roller is a Java blogging platform. Its XML-RPC endpoint <strong>deserializes vendor-extension types before it authenticates the request</strong>. Deserializing untrusted data on a pre-auth code path is a classic road to remote code execution: an attacker who can reach the endpoint controls the object graph that gets instantiated. The Apache Software Foundation notes this is the same underlying weakness as <strong>CVE-2026-82384</strong> (patched September 2026) that was incompletely fixed — CVE-2026-129448 is the proper closure.</p>

<h2 id="impact">Impact</h2>

<p>Full remote code execution as the Roller service account: site defacement, theft of posts, drafts and user data, and a foothold for lateral movement into whatever network the blog host sits in. The ASF confirms the flaw is <strong>being actively exploited</strong>, so any exposed, unpatched instance should be treated as potentially already compromised.</p>

<h2 id="prerequisites-for-the-attack-to-work">Prerequisites (for the attack to work)</h2>

<ul>
  <li>Network reachability to the Roller app, specifically the <code class="language-plaintext highlighter-rouge">/roller-services/xmlrpc</code> servlet.</li>
  <li>A vulnerable version: <strong>Apache Roller 4.0.0 through 6.1.4</strong>.</li>
  <li><strong>No credentials. No user interaction.</strong></li>
</ul>

<h2 id="steps-to-reproduce-exposure-validation">Steps to reproduce (exposure validation)</h2>

<p>On a host you administer, confirm whether the endpoint is reachable and which version is deployed:</p>

<div class="language-bash highlighter-rouge"><div class="highlight"><pre class="highlight"><code><span class="c"># 1) Is the XML-RPC endpoint exposed? A 200/405 to POST means it's present.</span>
curl <span class="nt">-s</span> <span class="nt">-o</span> /dev/null <span class="nt">-w</span> <span class="s2">"status=%{http_code}</span><span class="se">\n</span><span class="s2">"</span> <span class="se">\</span>
  <span class="nt">-X</span> POST https://blog.example.com/roller-services/xmlrpc

<span class="c"># 2) Confirm the deployed version from the WAR manifest on your own server</span>
unzip <span class="nt">-p</span> /opt/roller/roller.war META-INF/MANIFEST.MF | <span class="nb">grep</span> <span class="nt">-i</span> version

<span class="c"># 3) Spot unexpected children of the Roller/JVM process (post-exploitation signal)</span>
ps <span class="nt">--ppid</span> <span class="s2">"</span><span class="si">$(</span>pgrep <span class="nt">-f</span> roller | <span class="nb">head</span> <span class="nt">-n1</span><span class="si">)</span><span class="s2">"</span> <span class="nt">-o</span> pid,cmd 2&gt;/dev/null
</code></pre></div></div>

<p>If the endpoint answers <strong>and</strong> the version is ≤ 6.1.4, you are exposed. For a full proof-of-concept, build a matching version in an isolated VM and use the researcher references below — never against production or third-party systems.</p>

<h2 id="remediation">Remediation</h2>

<ol>
  <li><strong>Upgrade to Apache Roller 6.1.6 or later.</strong> This is the real fix.</li>
  <li>If you genuinely cannot patch yet, apply an official workaround and <strong>restart Roller</strong>:
    <ul>
      <li>Set <code class="language-plaintext highlighter-rouge">enabledForExtensions</code> to <code class="language-plaintext highlighter-rouge">false</code> in <code class="language-plaintext highlighter-rouge">WEB-INF/web.xml</code>, <strong>or</strong></li>
      <li>Remove the <code class="language-plaintext highlighter-rouge">/roller-services/xmlrpc</code> servlet mapping entirely.</li>
      <li>Do <strong>not</strong> rely on disabling XML-RPC in the admin UI — older versions deserialize the request <em>before</em> checking that setting.</li>
    </ul>
  </li>
  <li>Because exploitation is active, <strong>hunt before you trust the patch</strong>: review the Roller account for unexpected child processes, new admin users, modified templates, and outbound connections from the app host.</li>
</ol>

<h2 id="references">References</h2>

<ul>
  <li><a href="https://www.cve.org/CVERecord?id=CVE-2026-129448">CVE-2026-129448 — cve.org</a></li>
  <li><a href="https://www.cve.org/CVERecord?id=CVE-2026-82384">CVE-2026-82384 — related, cve.org</a></li>
  <li><a href="https://lists.apache.org/thread/lnz4s26tv6hdo03wngtdbpfj4hg4gtds">Apache Software Foundation advisory thread</a></li>
  <li><a href="https://thehackernews.com/2026/11/critical-apache-roller-rce-flaw-under.html">The Hacker News — active exploitation write-up</a></li>
  <li><a href="https://socprime.com/blog/cve-2026-129448-critical-apache-roller-rce-flaw-exploited-in-the-wild/">SOC Prime analysis</a></li>
</ul>

<p><em>Verify every version against the vendor advisory before acting. Published 20 Nov 2026.</em></p>]]></content><author><name>Aniket Tomar</name></author><category term="cve" /><category term="rce" /><category term="apache" /><category term="deserialization" /><category term="blue-team" /><summary type="html"><![CDATA[A pre-auth XML-RPC deserialization flaw (CVSS 9.8) gives attackers code execution on Apache Roller. Impact, prerequisites, exposure checks and the exact fix.]]></summary><media:thumbnail xmlns:media="http://search.yahoo.com/mrss/" url="https://binary0101devil.pages.dev/assets/img/cve-2026-129448-apache-roller.svg" /><media:content medium="image" url="https://binary0101devil.pages.dev/assets/img/cve-2026-129448-apache-roller.svg" xmlns:media="http://search.yahoo.com/mrss/" /></entry><entry><title type="html">Java2Shell / Java4Shell (CVE-2026-114938): heap overflow in the JVM’s AVX-512 Base64 intrinsic</title><link href="https://binary0101devil.pages.dev/blog/2026/11/20/java2shell-cve-2026-114938-jvm-base64/" rel="alternate" type="text/html" title="Java2Shell / Java4Shell (CVE-2026-114938): heap overflow in the JVM’s AVX-512 Base64 intrinsic" /><published>2026-11-20T00:00:00+05:30</published><updated>2026-11-20T00:00:00+05:30</updated><id>https://binary0101devil.pages.dev/blog/2026/11/20/java2shell-cve-2026-114938-jvm-base64</id><content type="html" xml:base="https://binary0101devil.pages.dev/blog/2026/11/20/java2shell-cve-2026-114938-jvm-base64/"><![CDATA[<p>Java2Shell (Harness’s name) or Java4Shell (Calif.io’s name, the original finders) is a memory-corruption bug <strong>inside HotSpot itself</strong>: the hand-written AVX-512 machine code the JIT uses to speed up <code class="language-plaintext highlighter-rouge">java.util.Base64</code> decoding. It was disclosed and fixed on <strong>17 November 2026</strong>. The headlines say “unauthenticated RCE in Java.” The reality is narrower but still serious, and this post is mostly about telling the two apart. Companion posts: <a href="/blog/2026/11/20/php2shell-cve-2026-109618-109619/">PHP2Shell</a> and <a href="/blog/2026/11/20/node2shell-cve-2026-111497-brotli-rce/">Node2Shell</a>.</p>

<blockquote>
  <p><strong>Scope and ethics.</strong> This post covers architecture, exposure checks, detection and remediation. It contains no trigger strings. Run any validation on JVMs you own, in an isolated lab.</p>
</blockquote>

<h2 id="at-a-glance">At a glance</h2>

<table>
  <thead>
    <tr>
      <th> </th>
      <th> </th>
    </tr>
  </thead>
  <tbody>
    <tr>
      <td><strong>CVE</strong></td>
      <td>CVE-2026-114938</td>
    </tr>
    <tr>
      <td><strong>Severity</strong></td>
      <td>CVSS 3.1 <strong>8.1 High</strong>, <code class="language-plaintext highlighter-rouge">AV:N/AC:H/PR:N/UI:N/S:U/C:H/I:H/A:H</code> (Oracle, via NVD). Calif.io scores it CVSS 4.0 9.2</td>
    </tr>
    <tr>
      <td><strong>CWE</strong></td>
      <td>CWE-122 Heap-based Buffer Overflow</td>
    </tr>
    <tr>
      <td><strong>Component</strong></td>
      <td>HotSpot C2 compiler, x86-64 AVX-512 Base64 decode intrinsic</td>
    </tr>
    <tr>
      <td><strong>Affected (Oracle)</strong></td>
      <td>Java SE 17.0.21, 21.0.13, 25.0.5, 27.0.1 and earlier on those lines (17 from <strong>17.0.6</strong> onward), Oracle Java 8u511b33-<strong>perf</strong> only, GraalVM for JDK 17/21, GraalVM EE 21.3.20</td>
    </tr>
    <tr>
      <td><strong>Fixed in</strong></td>
      <td>November 2026 update: <strong>17.0.21.1 / 21.0.13.1 / 25.0.5.1 / 27.0.1.1</strong> (OpenJDK distributors’ numbering may differ, so check yours)</td>
    </tr>
    <tr>
      <td><strong>Exploited?</strong></td>
      <td><strong>No in-the-wild exploitation reported</strong> as of 20 Nov. Not in CISA KEV. CISA SSVC: exploitation <em>none</em>, automatable <em>no</em>, technical impact <em>total</em>. A memory-corruption PoC is public, but product RCE chains are not</td>
    </tr>
  </tbody>
</table>

<h2 id="description-how-the-bug-works">Description: how the bug works</h2>

<p><code class="language-plaintext highlighter-rouge">java.util.Base64.Decoder</code> works out the exact decoded length up front and allocates an output array of exactly that size. The pure-Java decoder then fills it with bounds-checked writes.</p>

<p>Hot code gets compiled by the C2 JIT, and on CPUs with the right AVX-512 extensions (VBMI, plus AVX512F/BW and BMI2) HotSpot swaps the inner loop for an <strong>intrinsic</strong>: hand-tuned vector assembly that decodes 64 bytes per instruction. The intrinsic <strong>isn’t given the destination length and doesn’t recheck it</strong>. It trusts that the Java caller sized everything correctly.</p>

<p>The bug is in the intrinsic’s <strong>padding path</strong>. Given a padded tail (<code class="language-plaintext highlighter-rouge">=</code>) shorter than one full vector block, the native code computes a <strong>negative</strong> “bytes remaining” count. That count turns into an AVX-512 store mask, the negative value saturates so <strong>every lane is enabled</strong>, and a full 64-byte vector store goes out starting near the end of the output array. Public lab work measured <strong>62 bytes written past a 50-byte output</strong>. Java then notices the input is malformed and throws <code class="language-plaintext highlighter-rouge">IllegalArgumentException</code>, <strong>but the out-of-bounds write has already happened.</strong> An HTTP 401 or 400 response doesn’t mean nothing was corrupted.</p>

<p>The upstream fix (OpenJDK commit <code class="language-plaintext highlighter-rouge">5177a70</code>, per public analyses) sends short padded tails back to the safe Java code path.</p>

<h2 id="why-its-narrower-than-the-headlines">Why it’s narrower than the headlines</h2>

<p>All of these must be true for a JVM to be exploitable:</p>

<ol>
  <li><strong>x86-64 with AVX-512 VBMI.</strong> That means Intel Ice Lake and later server and client chips, and AMD Zen 4/5. <strong>ARM64 (Graviton, Ampere, Apple silicon) and older x86 aren’t affected.</strong> In the cloud, whether you’re exposed can depend on the instance family your autoscaler happens to pick.</li>
  <li><strong>An affected JDK build</strong> (17.0.6+, 21, 25, 27; not generic 8 or 11). Remember <strong>bundled JREs</strong> inside vendor products.</li>
  <li><strong>The intrinsic is hot.</strong> Interpreted code takes the safe path, and the overflow only happens after C2 compiles the decode routine. Public tooling needed a warm-up of many valid decodes first. The exception can also deoptimize the method, so <strong>a one-shot scan can miss a vulnerable JVM</strong>.</li>
  <li><strong>Attacker-controlled Base64 reaches <code class="language-plaintext highlighter-rouge">java.util.Base64</code>.</strong> Unfortunately that’s everywhere: <code class="language-plaintext highlighter-rouge">Authorization: Basic</code> headers, JWT segments, SAML/XML, SASL, API fields and queue messages.</li>
  <li><strong>Turning a 62-byte heap overwrite into code execution is application-specific.</strong> The public lab shows input → out-of-bounds write and nothing more (<code class="language-plaintext highlighter-rouge">shell_proven=false</code>). Calif.io and Immersive say they have private product chains, with the pre-auth list including <strong>Keycloak, Jira, Confluence, Jenkins, PingFederate</strong> and possibly Artifactory, and authenticated paths in Kafka, Solr, Elasticsearch and Gerrit. None are public or independently verified.</li>
</ol>

<p>So: patch with high priority on AVX-512 x86 hosts running identity, SSO and CI systems. Treat ARM fleets as not affected. Don’t page the whole company over a Java 8 desktop app.</p>

<h2 id="identifying-affected-systems">Identifying affected systems</h2>

<p><strong>The running PID is the source of truth.</strong> An SBOM, image tag or package manager can say “patched” while a JVM started three weeks ago is still running the old code.</p>

<div class="language-bash highlighter-rouge"><div class="highlight"><pre class="highlight"><code><span class="c"># 1) CPU gate: no avx512vbmi = not exploitable on this host</span>
<span class="nb">grep</span> <span class="nt">-o</span> <span class="nt">-m1</span> <span class="nt">-w</span> avx512vbmi /proc/cpuinfo <span class="o">||</span> <span class="nb">echo</span> <span class="s2">"no AVX-512 VBMI: not affected"</span>
<span class="nb">grep</span> <span class="nt">-m1</span> <span class="s1">'^flags'</span> /proc/cpuinfo | <span class="nb">grep</span> <span class="nt">-ow</span> <span class="nt">-E</span> <span class="s1">'avx512f|avx512bw|avx512vbmi|bmi2'</span> | <span class="nb">sort</span> <span class="nt">-u</span>

<span class="c"># 2) Each running JVM: exact runtime version and whether the intrinsic is enabled</span>
<span class="k">for </span>pid <span class="k">in</span> <span class="si">$(</span>pgrep <span class="nt">-f</span> java<span class="si">)</span><span class="p">;</span> <span class="k">do
  </span><span class="nb">echo</span> <span class="s2">"== PID </span><span class="nv">$pid</span><span class="s2">: </span><span class="si">$(</span><span class="nb">tr</span> <span class="s1">'\0'</span> <span class="s1">' '</span> &lt;/proc/<span class="nv">$pid</span>/cmdline | <span class="nb">cut</span> <span class="nt">-c1-120</span><span class="si">)</span><span class="s2">"</span>
  jcmd <span class="s2">"</span><span class="nv">$pid</span><span class="s2">"</span> VM.version 2&gt;/dev/null | <span class="nb">sed</span> <span class="nt">-n</span> <span class="s1">'2,3p'</span>
  jcmd <span class="s2">"</span><span class="nv">$pid</span><span class="s2">"</span> VM.flags <span class="nt">-all</span> 2&gt;/dev/null | <span class="nb">grep</span> <span class="nt">-E</span> <span class="s1">'\bUseBASE64Intrinsics\b|\bUseAVX\b'</span>
<span class="k">done</span>

<span class="c"># 3) Installed JDKs and container images</span>
java <span class="nt">-version</span> 2&gt;&amp;1
java <span class="nt">-XX</span>:+PrintFlagsFinal <span class="nt">-version</span> 2&gt;/dev/null | <span class="nb">grep</span> <span class="nt">-E</span> <span class="s1">'\bUseBASE64Intrinsics\b|\bUseAVX\b'</span>
docker ps <span class="nt">--format</span> <span class="s1">'{{.Image}}'</span> | <span class="nb">sort</span> <span class="nt">-u</span> | xargs <span class="nt">-I</span><span class="o">{}</span> sh <span class="nt">-c</span> <span class="s1">'echo "{}:"; docker run --rm --entrypoint java {} -version 2&gt;&amp;1 | head -1'</span>

<span class="c"># 4) Find bundled JREs inside vendor products</span>
find / <span class="nt">-xdev</span> <span class="nt">-type</span> f <span class="nt">-name</span> <span class="s1">'libjvm.so'</span> 2&gt;/dev/null
</code></pre></div></div>

<p>A JVM is <strong>exposed</strong> if the CPU check passes, the version is in the affected range, and <code class="language-plaintext highlighter-rouge">UseBASE64Intrinsics</code> is <code class="language-plaintext highlighter-rouge">true</code> (it’s on by default on capable CPUs). <code class="language-plaintext highlighter-rouge">UseAVX</code> below 3 also turns off the AVX-512 path.</p>

<h2 id="safe-validation-in-an-isolated-lab">Safe validation in an isolated lab</h2>

<ol>
  <li>Use an AVX-512 VBMI host (or a cloud instance type that has it) on an isolated network.</li>
  <li>Install an affected build (e.g. Temurin 25.0.5) and the fixed build side by side.</li>
  <li>Run the checks above against each, and confirm your inventory tooling reports the <strong>live process version</strong>, not just the package.</li>
  <li>Start the affected JVM with <code class="language-plaintext highlighter-rouge">-XX:-UseBASE64Intrinsics</code> and confirm <code class="language-plaintext highlighter-rouge">jcmd &lt;pid&gt; VM.flags</code> shows it. That’s how you check the workaround actually reached the process.</li>
  <li>If you want to go further, the open-source <a href="https://github.com/zake1god/java2shell">zake1god/java2shell</a> checker writes into a <strong>sentinel array</strong> and compares vulnerable and mitigated JVMs <strong>in-process</strong>, without targeting any network service. Use it only inside your lab.</li>
</ol>

<h2 id="detection">Detection</h2>

<p>There’s little to see on the wire. The trigger is just a slightly malformed Base64 value, so detection is mostly about <strong>effects</strong>:</p>

<ul>
  <li><strong>JVM crashes:</strong> <code class="language-plaintext highlighter-rouge">hs_err_pid*.log</code> files mentioning <code class="language-plaintext highlighter-rouge">StubRoutines</code> / Base64 stubs, or <code class="language-plaintext highlighter-rouge">SIGSEGV</code> in compiled code on identity, CI and SSO hosts. Collect them centrally.</li>
  <li><strong>Spikes in Base64 decode errors:</strong> bursts of <code class="language-plaintext highlighter-rouge">IllegalArgumentException: Illegal base64 character</code> / <code class="language-plaintext highlighter-rouge">Input byte array has incorrect ending byte</code> in app logs, especially on auth endpoints (Basic auth, JWT parsing).</li>
  <li><strong>Odd auth failures:</strong> many 400/401s with unusually long <code class="language-plaintext highlighter-rouge">Authorization: Basic</code> values or malformed JWT segments from a single source.</li>
  <li><strong>Process lineage:</strong> <code class="language-plaintext highlighter-rouge">java</code> spawning <code class="language-plaintext highlighter-rouge">sh</code>, <code class="language-plaintext highlighter-rouge">bash</code>, <code class="language-plaintext highlighter-rouge">curl</code>, <code class="language-plaintext highlighter-rouge">wget</code> or <code class="language-plaintext highlighter-rouge">cmd.exe</code>. The usual signal that any Java RCE has been chained to a shell.</li>
</ul>

<h2 id="remediation">Remediation</h2>

<ol>
  <li><strong>Patch the JDK/JRE</strong> to the November 2026 update (Oracle 17.0.21.1 / 21.0.13.1 / 25.0.5.1 / 27.0.1.1, or your OpenJDK vendor’s equivalent: Temurin, Corretto, Zulu, Red Hat, Microsoft and so on), plus GraalVM where used.</li>
  <li><strong>Restart every JVM.</strong> This is the step most often missed. Then re-check with <code class="language-plaintext highlighter-rouge">jcmd &lt;pid&gt; VM.version</code>.</li>
  <li><strong>Chase bundled runtimes.</strong> Keycloak, Atlassian, Jenkins, PingFederate and similar appliance or container images ship their own JREs, so apply the <em>vendor’s</em> update.</li>
</ol>

<h3 id="workarounds-stopgaps-need-a-restart">Workarounds (stopgaps, need a restart)</h3>

<table>
  <thead>
    <tr>
      <th>Option</th>
      <th>Effect</th>
      <th>Cost</th>
    </tr>
  </thead>
  <tbody>
    <tr>
      <td><code class="language-plaintext highlighter-rouge">-XX:-UseBASE64Intrinsics</code></td>
      <td>Forces the pure-Java decoder. Lab tests show <strong>0 bytes</strong> written out of bounds</td>
      <td>Slower Base64 only. Narrowest option</td>
    </tr>
    <tr>
      <td><code class="language-plaintext highlighter-rouge">-XX:UseAVX=2</code></td>
      <td>Stops the JVM using AVX-512 at all</td>
      <td>Broader slowdown in crypto, array and string intrinsics</td>
    </tr>
  </tbody>
</table>

<p>Add the flag to the service’s own JVM options (<code class="language-plaintext highlighter-rouge">JAVA_OPTS</code>, <code class="language-plaintext highlighter-rouge">setenv.sh</code>, systemd unit, Kubernetes manifest). <code class="language-plaintext highlighter-rouge">JAVA_TOOL_OPTIONS</code> works but applies to <strong>every</strong> JVM in that environment. Public write-ups report that the flag isn’t recognized on ARM and may stop the JVM from starting there. ARM isn’t affected anyway, so scope the flag to x86 hosts and test it before rolling out.</p>

<h2 id="references">References</h2>

<ul>
  <li><a href="https://nvd.nist.gov/vuln/detail/CVE-2026-114938">CVE-2026-114938, NVD</a></li>
  <li><a href="https://openjdk.org/groups/vulnerability/advisories/2026-11-17">OpenJDK vulnerability advisory, 17 Nov 2026</a></li>
  <li><a href="https://www.oracle.com/security-alerts/cspunov2026.html">Oracle Security Alert, November 2026</a></li>
  <li><a href="https://calif.io/java4shell/">Calif.io: Java4Shell disclosure</a></li>
  <li><a href="https://www.harness.io/blog/past-the-end-of-the-buffer-cve-2026-114938">Harness: Past the end of the buffer</a></li>
  <li><a href="https://access.redhat.com/security/vulnerabilities/RHSB-2026-018">Red Hat bulletin RHSB-2026-018</a></li>
  <li><a href="https://www.tenable.com/blog/java4shell-cve-2026-114938-frequently-asked-questions-about-openjdk-rce-vulnerability">Tenable FAQ</a></li>
  <li><a href="https://securelayer7.net/learn/web-exploitation/what-is-java2shell">SecureLayer7: What is Java2Shell</a> · <a href="https://securitycipher.com/2026/11/18/java2shell-cve-2026-114938-lab/">SecurityCipher lab analysis</a></li>
</ul>

<p><em>Verify every version against the vendor advisory before acting. Published 20 Nov 2026.</em></p>]]></content><author><name>Aniket Tomar</name></author><category term="cve" /><category term="rce" /><category term="java" /><category term="heap-overflow" /><category term="blue-team" /><summary type="html"><![CDATA[A JIT-compiled Base64 decoder on AVX-512 CPUs writes past its output buffer (CVSS 3.1 8.1, researcher CVSS 4.0 9.2). Who is really exposed, how to check a live JVM, the flag workaround and the fix.]]></summary><media:thumbnail xmlns:media="http://search.yahoo.com/mrss/" url="https://binary0101devil.pages.dev/assets/img/java2shell-cve-2026-114938.svg" /><media:content medium="image" url="https://binary0101devil.pages.dev/assets/img/java2shell-cve-2026-114938.svg" xmlns:media="http://search.yahoo.com/mrss/" /></entry><entry><title type="html">Node2Shell (CVE-2026-111497): Brotli decoder out-of-bounds write, exploited in the wild</title><link href="https://binary0101devil.pages.dev/blog/2026/11/20/node2shell-cve-2026-111497-brotli-rce/" rel="alternate" type="text/html" title="Node2Shell (CVE-2026-111497): Brotli decoder out-of-bounds write, exploited in the wild" /><published>2026-11-20T00:00:00+05:30</published><updated>2026-11-20T00:00:00+05:30</updated><id>https://binary0101devil.pages.dev/blog/2026/11/20/node2shell-cve-2026-111497-brotli-rce</id><content type="html" xml:base="https://binary0101devil.pages.dev/blog/2026/11/20/node2shell-cve-2026-111497-brotli-rce/"><![CDATA[<p>Node2Shell (also called HTTP2Shell, Broccoli2Shell or Brotli2Shell) isn’t really a Node.js bug. It’s a bug in <strong>Google’s Brotli C decoder</strong>, which Node bundles, and so do Bun, Python’s <code class="language-plaintext highlighter-rouge">brotli</code>/<code class="language-plaintext highlighter-rouge">brotlicffi</code>, Envoy, Istio, .NET, curl and others. Any process that <strong>decompresses Brotli data an attacker can influence</strong> is in scope. CISA added it to the KEV catalog on 27 October 2026. Companion posts: <a href="/blog/2026/11/20/php2shell-cve-2026-109618-109619/">PHP2Shell</a> and <a href="/blog/2026/11/20/java2shell-cve-2026-114938-jvm-base64/">Java2Shell</a>.</p>

<blockquote>
  <p><strong>Scope and ethics.</strong> This post covers architecture, exposure checks, detection and remediation. It contains no malicious Brotli streams or exploit offsets. Run any validation on systems you own, in an isolated lab.</p>
</blockquote>

<h2 id="at-a-glance">At a glance</h2>

<table>
  <thead>
    <tr>
      <th> </th>
      <th> </th>
    </tr>
  </thead>
  <tbody>
    <tr>
      <td><strong>CVE</strong></td>
      <td>CVE-2026-111497</td>
    </tr>
    <tr>
      <td><strong>Severity</strong></td>
      <td>CVSS 4.0 <strong>9.2 Critical</strong>, <code class="language-plaintext highlighter-rouge">AV:N/AC:L/AT:P/PR:N/UI:N/VC:H/VI:H/VA:H</code> (Google CNA, via NVD). Node.js rates it <em>High</em> on its own scale</td>
    </tr>
    <tr>
      <td><strong>CWE</strong></td>
      <td>CWE-252 Unchecked Return Value → CWE-787 Out-of-bounds Write</td>
    </tr>
    <tr>
      <td><strong>Affected</strong></td>
      <td>Brotli <strong>1.0.3 – 1.2.0</strong> (bug shipped March 2018)</td>
    </tr>
    <tr>
      <td><strong>Fixed in</strong></td>
      <td>Brotli <strong>1.0.10 / 1.1.1 / 1.2.1</strong>. Node.js <strong>22.23.4 / 24.22.1 / 26.12.1</strong> (21 Oct 2026). Bun <strong>1.4.4</strong></td>
    </tr>
    <tr>
      <td><strong>Exploited?</strong></td>
      <td><strong>Yes.</strong> CISA KEV, added 27 Oct 2026 (due 10 Nov)</td>
    </tr>
    <tr>
      <td><strong>Auth / interaction</strong></td>
      <td>None / none</td>
    </tr>
  </tbody>
</table>

<h2 id="description-how-the-bug-works">Description: how the bug works</h2>

<p>A Brotli stream is split into <strong>metablocks</strong>. Within each metablock the decoder tracks three kinds of block (literals, insert-and-copy commands, and distances), each with a counter. When a counter reaches zero, the decoder reads a <strong>block switch</strong> that says which block type comes next.</p>

<p>If a metablock declares only <em>one</em> block type for a kind, a block switch is invalid. That’s a malformed stream and should be a hard error. The helper that decodes block switches returned a plain boolean, though, and <code class="language-plaintext highlighter-rouge">false</code> meant both “this stream is invalid” <strong>and</strong> “I ran out of input, feed me more.” In streaming mode the caller read <code class="language-plaintext highlighter-rouge">false</code> as <em>need more input</em>, so it kept buffering bytes into a small fixed internal buffer (8 bytes), past its end. The result is an <strong>out-of-bounds write into adjacent decoder state</strong>, with the contents controlled by the attacker.</p>

<p>The upstream fix (google/brotli commit <code class="language-plaintext highlighter-rouge">0ed4268a</code>, “Detect and report invalid stream earlier”) adds a dedicated error code, <code class="language-plaintext highlighter-rouge">BROTLI_DECODER_ERROR_FORMAT_BLOCK_SWITCH</code>, so invalid switches fail immediately. The commit landed on master on <strong>27 Feb 2026 with no release or advisory</strong>. The patched releases and CVE followed in October. That eight-month gap gave attackers a long time to diff the fix, and ELLIO saw a <code class="language-plaintext highlighter-rouge">node2shell-check</code> scanner as early as <strong>13 June 2026</strong>.</p>

<h2 id="impact">Impact</h2>

<p>Per NVD and CISA: corruption of decoder state, <strong>memory disclosure</strong>, <strong>denial of service</strong> (the usual outcome of a failed attempt, which crashes the worker), and <strong>potential arbitrary code execution</strong>. Imperva showed RCE on the official Node.js 24.22.0 image via both <code class="language-plaintext highlighter-rouge">fetch()</code> and Express 5. ELLIO reproduced it on x86-64.</p>

<h2 id="which-applications-are-reachable">Which applications are reachable</h2>

<p>The Node.js advisory frames exposure as “decompresses with <code class="language-plaintext highlighter-rouge">node:zlib</code>”. In practice, that covers far more code than direct zlib calls:</p>

<table>
  <thead>
    <tr>
      <th>Path</th>
      <th>Reaches the decoder?</th>
    </tr>
  </thead>
  <tbody>
    <tr>
      <td><code class="language-plaintext highlighter-rouge">zlib.brotliDecompress*()</code> / <code class="language-plaintext highlighter-rouge">createBrotliDecompress()</code> on untrusted data</td>
      <td><strong>Yes</strong></td>
    </tr>
    <tr>
      <td><strong>Inbound</strong> bodies with <code class="language-plaintext highlighter-rouge">Content-Encoding: br</code>: Express 5 <code class="language-plaintext highlighter-rouge">express.json()/raw()/text()/urlencoded()</code> via body-parser 2 (<code class="language-plaintext highlighter-rouge">inflate: true</code> by default), Koa <code class="language-plaintext highlighter-rouge">@koa/bodyparser</code>, <code class="language-plaintext highlighter-rouge">@fastify/compress</code> request decompression</td>
      <td><strong>Yes.</strong> Any client can send one</td>
    </tr>
    <tr>
      <td><strong>Outbound</strong> responses: built-in <code class="language-plaintext highlighter-rouge">fetch()</code>/undici, axios, got, SuperAgent, server-side fetch in Next.js</td>
      <td><strong>Yes</strong>, if an attacker controls or influences the URL or the upstream response (SSRF, webhooks, link previews)</td>
    </tr>
    <tr>
      <td>A bare <code class="language-plaintext highlighter-rouge">node:http</code> server that never decodes bodies</td>
      <td>No</td>
    </tr>
    <tr>
      <td>Python <code class="language-plaintext highlighter-rouge">requests</code>/<code class="language-plaintext highlighter-rouge">urllib3</code>/<code class="language-plaintext highlighter-rouge">httpx</code>/<code class="language-plaintext highlighter-rouge">aiohttp</code>/Scrapy with <code class="language-plaintext highlighter-rouge">brotli</code> or <code class="language-plaintext highlighter-rouge">brotlicffi</code> installed</td>
      <td>Yes, via response <code class="language-plaintext highlighter-rouge">Content-Encoding</code></td>
    </tr>
    <tr>
      <td>Envoy / Istio with the Brotli decompressor filter configured</td>
      <td>Yes</td>
    </tr>
    <tr>
      <td>.NET with Brotli request decompression enabled (off by default)</td>
      <td>Yes</td>
    </tr>
  </tbody>
</table>

<p><strong>Prerequisites:</strong> the attacker can deliver Brotli bytes that your process decompresses. No credentials and no user interaction are needed. Building a reliable exploit is build-specific, but crashing a worker is not.</p>

<h2 id="identifying-affected-systems">Identifying affected systems</h2>

<p><strong>Don’t trust <code class="language-plaintext highlighter-rouge">process.versions.brotli</code>.</strong> It reports <code class="language-plaintext highlighter-rouge">1.2.0</code> on both patched and unpatched Node builds, because Node cherry-picked the fix without bumping the string. Go by the <strong>Node.js version</strong> instead:</p>

<div class="language-bash highlighter-rouge"><div class="highlight"><pre class="highlight"><code><span class="c"># 1) Node runtime version: vulnerable if below 22.23.4 / 24.22.1 / 26.12.1, or on 20.x / 23.x / 25.x (EOL, no fix)</span>
node <span class="nt">-v</span>

<span class="c"># 2) Does this node binary link the SYSTEM libbrotli? If so, the distro package matters too.</span>
ldd <span class="s2">"</span><span class="si">$(</span><span class="nb">command</span> <span class="nt">-v</span> node<span class="si">)</span><span class="s2">"</span> | <span class="nb">grep</span> <span class="nt">-i</span> brotli <span class="o">||</span> <span class="nb">echo</span> <span class="s2">"bundled brotli (go by node version)"</span>
dpkg <span class="nt">-l</span> | <span class="nb">grep</span> <span class="nt">-i</span> brotli 2&gt;/dev/null <span class="o">||</span> rpm <span class="nt">-qa</span> | <span class="nb">grep</span> <span class="nt">-i</span> brotli

<span class="c"># 3) Every container image's runtime, not just the host</span>
docker ps <span class="nt">--format</span> <span class="s1">'{{.Image}}'</span> | <span class="nb">sort</span> <span class="nt">-u</span> | xargs <span class="nt">-I</span><span class="o">{}</span> sh <span class="nt">-c</span> <span class="s1">'echo -n "{}: "; docker run --rm --entrypoint node {} -v 2&gt;/dev/null || echo n/a'</span>

<span class="c"># 4) Python bindings and Bun</span>
pip list 2&gt;/dev/null | <span class="nb">grep</span> <span class="nt">-iE</span> <span class="s1">'^(brotli|brotlicffi) '</span>   <span class="c"># need Brotli&gt;=1.2.1 / brotlicffi&gt;=1.2.1.0</span>
bun <span class="nt">--version</span> 2&gt;/dev/null                                  <span class="c"># need &gt;=1.4.4</span>

<span class="c"># 5) Is your app actually decoding inbound br? (look for body-parser 2 / Express 5)</span>
npm <span class="nb">ls </span>express body-parser @koa/bodyparser @fastify/compress 2&gt;/dev/null
</code></pre></div></div>

<p>Fixed distro packages include Debian bookworm <code class="language-plaintext highlighter-rouge">brotli 1.0.9-2+deb12u1</code> (DLA-4855-1), trixie <code class="language-plaintext highlighter-rouge">1.1.0-2+deb13u1</code> (DSA-6582-1) and Alpine <code class="language-plaintext highlighter-rouge">brotli 1.2.1-r0</code>. Envoy is fixed in 1.40.0 / 1.39.3 / 1.38.6 / 1.37.8 / 1.36.12, and the Ruby <code class="language-plaintext highlighter-rouge">brotli</code> gem in 0.9.1.</p>

<h2 id="safe-validation-in-an-isolated-lab">Safe validation in an isolated lab</h2>

<ol>
  <li>Run <code class="language-plaintext highlighter-rouge">node:24.22.0</code> and <code class="language-plaintext highlighter-rouge">node:24.22.1</code> containers on an isolated network, each with a minimal Express 5 app using <code class="language-plaintext highlighter-rouge">express.json()</code>.</li>
  <li>Send a <strong>benign, valid</strong> Brotli-compressed JSON body (<code class="language-plaintext highlighter-rouge">echo '{"ok":1}' | brotli -c | curl --data-binary @- -H 'Content-Encoding: br' -H 'Content-Type: application/json' ...</code>) to both. They should both accept it, which proves the decode path is reachable.</li>
  <li>Enable your interim control (body-parser <code class="language-plaintext highlighter-rouge">inflate: false</code>, or a proxy rule rejecting <code class="language-plaintext highlighter-rouge">br</code>) and confirm the same request now gets <strong>415</strong> or is blocked.</li>
  <li>Confirm your scanner and SBOM tooling flags <code class="language-plaintext highlighter-rouge">24.22.0</code> and passes <code class="language-plaintext highlighter-rouge">24.22.1</code>, and that it <strong>doesn’t</strong> rely on <code class="language-plaintext highlighter-rouge">process.versions.brotli</code>.</li>
</ol>

<h2 id="detection-and-threat-hunting">Detection and threat hunting</h2>

<ul>
  <li><strong>Worker crashes.</strong> Segfaults in <code class="language-plaintext highlighter-rouge">node</code>, <code class="language-plaintext highlighter-rouge">bun</code> or <code class="language-plaintext highlighter-rouge">python</code> with <code class="language-plaintext highlighter-rouge">BrotliDecoderDecompressStream</code> / <code class="language-plaintext highlighter-rouge">libbrotlidec</code> on the stack. Forward <code class="language-plaintext highlighter-rouge">dmesg</code>/kernel messages and <code class="language-plaintext highlighter-rouge">systemd-coredump</code> events to your SIEM. Repeated PM2 or Kubernetes restarts of Node pods (<code class="language-plaintext highlighter-rouge">CrashLoopBackOff</code>) after inbound POSTs are a strong signal.</li>
  <li><strong>Request telemetry.</strong> Inbound requests with <code class="language-plaintext highlighter-rouge">Content-Encoding: br</code> (case-insensitive, including lists like <code class="language-plaintext highlighter-rouge">gzip, br</code>) to APIs that no legitimate client sends compressed. Most browsers never compress <em>request</em> bodies, so this traffic is rare in normal use.</li>
  <li><strong>Process lineage.</strong> <code class="language-plaintext highlighter-rouge">node</code>/<code class="language-plaintext highlighter-rouge">bun</code>/<code class="language-plaintext highlighter-rouge">python</code> spawning <code class="language-plaintext highlighter-rouge">sh</code>, <code class="language-plaintext highlighter-rouge">bash</code>, <code class="language-plaintext highlighter-rouge">curl</code>, <code class="language-plaintext highlighter-rouge">wget</code>, <code class="language-plaintext highlighter-rouge">nc</code>, <code class="language-plaintext highlighter-rouge">cat /etc/passwd</code>, <code class="language-plaintext highlighter-rouge">cmd.exe</code> or <code class="language-plaintext highlighter-rouge">powershell.exe</code>.</li>
  <li><strong>IOCs from ELLIO (26 Oct 2026):</strong> exploit traffic from <code class="language-plaintext highlighter-rouge">94.154.43.7</code>, request body SHA-256 <code class="language-plaintext highlighter-rouge">7fcbdaaecd4419e2930927844f3b772668204737a4e7602a2290c991c93c5c4b</code>. A scanner from AS44620 was active since June.</li>
  <li><strong>Core dumps.</strong> A crashed worker’s core file can contain session tokens and secrets. Restrict who can read core files and how long they’re kept.</li>
</ul>

<h2 id="remediation">Remediation</h2>

<ol>
  <li><strong>Upgrade Node.js</strong> to <strong>22.23.4 / 24.22.1 / 26.12.1</strong> or later (take the latest patch on your line, since later releases fixed other High-severity issues too). <strong>Move off EOL 20.x / 23.x / 25.x</strong>, which won’t get a fix.</li>
  <li><strong>Update Brotli everywhere</strong>: distro <code class="language-plaintext highlighter-rouge">libbrotli</code>, PyPI <code class="language-plaintext highlighter-rouge">Brotli&gt;=1.2.1</code> / <code class="language-plaintext highlighter-rouge">brotlicffi&gt;=1.2.1.0</code>, Bun ≥ 1.4.4, Envoy, Istio, .NET and the Ruby gem, then <strong>refresh lockfiles</strong>.</li>
  <li><strong>Rebuild and restart.</strong> Dockerfiles pinning <code class="language-plaintext highlighter-rouge">node:24.22.0</code>, <code class="language-plaintext highlighter-rouge">bun build --compile</code> binaries and self-contained .NET apps keep the bug until rebuilt. Running processes keep it until restarted.</li>
</ol>

<h3 id="interim-mitigations-not-fixes">Interim mitigations (not fixes)</h3>

<ul>
  <li><strong>Express / body-parser:</strong> <code class="language-plaintext highlighter-rouge">express.json({ inflate: false })</code> (and the same for <code class="language-plaintext highlighter-rouge">raw</code>/<code class="language-plaintext highlighter-rouge">text</code>/<code class="language-plaintext highlighter-rouge">urlencoded</code>). Compressed bodies are then rejected with 415.</li>
  <li><strong>Edge / WAF / proxy:</strong> reject inbound requests whose <code class="language-plaintext highlighter-rouge">Content-Encoding</code> contains <code class="language-plaintext highlighter-rouge">br</code>. This does <strong>not</strong> protect outbound <code class="language-plaintext highlighter-rouge">fetch()</code> responses.</li>
  <li><strong>Outbound:</strong> send <code class="language-plaintext highlighter-rouge">Accept-Encoding: gzip, deflate</code> (no <code class="language-plaintext highlighter-rouge">br</code>) on server-side HTTP clients that fetch attacker-influenced URLs, and tighten SSRF egress controls.</li>
  <li>Microsegment Node services so a compromised worker can’t reach much.</li>
</ul>

<h2 id="references">References</h2>

<ul>
  <li><a href="https://nvd.nist.gov/vuln/detail/CVE-2026-111497">CVE-2026-111497, NVD</a> · <a href="https://www.cisa.gov/known-exploited-vulnerabilities-catalog">CISA KEV catalog</a></li>
  <li><a href="https://nodejs.org/en/blog/vulnerability/october-2026-security-releases">Node.js October 2026 security releases</a></li>
  <li><a href="https://github.com/google/brotli/releases/tag/v1.2.1">google/brotli v1.2.1 release</a> · <a href="https://github.com/google/brotli/commit/0ed4268a4401abcb4472096340477cf4af7d1a2b">Fix commit 0ed4268a</a></li>
  <li><a href="https://www.akamai.com/blog/security-research/inside-node2shell-closer-look-cve-2026-111497">Akamai: Inside Node2Shell</a></li>
  <li><a href="https://ellio.tech/en/blog/node2shell-cve-2026-111497-rce-on-stock-nodejs-builds/">ELLIO: RCE on stock Node.js builds, in-the-wild IOCs</a></li>
  <li><a href="https://socprime.com/blog/node2shell-http2shell-detection/">SOC Prime: detection content</a></li>
  <li><a href="https://cybersecuritynews.com/node2shell-vulnerability/">Cyber Security News: KEV listing coverage</a></li>
</ul>

<p><em>Verify every version against the vendor advisory before acting. Published 20 Nov 2026.</em></p>]]></content><author><name>Aniket Tomar</name></author><category term="cve" /><category term="rce" /><category term="nodejs" /><category term="memory-corruption" /><category term="blue-team" /><summary type="html"><![CDATA[An unchecked return value in Google's Brotli C decoder (CVSS 4.0 9.2, on CISA KEV) lets a crafted br stream corrupt memory in Node.js, Bun, Python, Envoy and more. Exposure, detection and the fix.]]></summary><media:thumbnail xmlns:media="http://search.yahoo.com/mrss/" url="https://binary0101devil.pages.dev/assets/img/node2shell-cve-2026-111497.svg" /><media:content medium="image" url="https://binary0101devil.pages.dev/assets/img/node2shell-cve-2026-111497.svg" xmlns:media="http://search.yahoo.com/mrss/" /></entry><entry><title type="html">PHP2Shell (CVE-2026-109618 / CVE-2026-109619): pre-auth RCE in the PHP engine</title><link href="https://binary0101devil.pages.dev/blog/2026/11/20/php2shell-cve-2026-109618-109619/" rel="alternate" type="text/html" title="PHP2Shell (CVE-2026-109618 / CVE-2026-109619): pre-auth RCE in the PHP engine" /><published>2026-11-20T00:00:00+05:30</published><updated>2026-11-20T00:00:00+05:30</updated><id>https://binary0101devil.pages.dev/blog/2026/11/20/php2shell-cve-2026-109618-109619</id><content type="html" xml:base="https://binary0101devil.pages.dev/blog/2026/11/20/php2shell-cve-2026-109618-109619/"><![CDATA[<p>PHP2Shell is the name for two engine-level PHP flaws disclosed on 21 October 2026. Both are in CISA’s Known Exploited Vulnerabilities catalog. Because the bugs are in the interpreter and not in any framework, <strong>your application code doesn’t matter</strong>: a patched WordPress or Laravel on an unpatched PHP binary is still exposed. This is one of three “2Shell” bugs I’m covering today. The other two are <a href="/blog/2026/11/20/node2shell-cve-2026-111497-brotli-rce/">Node2Shell</a> and <a href="/blog/2026/11/20/java2shell-cve-2026-114938-jvm-base64/">Java2Shell</a>.</p>

<blockquote>
  <p><strong>Scope and ethics.</strong> This post covers architecture, exposure checks, detection and remediation for systems you own or are authorized to assess. It contains no trigger inputs or exploit code. Lab validation means confirming a version and config on a VM you control, not firing a PoC at someone else’s server.</p>
</blockquote>

<h2 id="at-a-glance">At a glance</h2>

<table>
  <thead>
    <tr>
      <th> </th>
      <th> </th>
    </tr>
  </thead>
  <tbody>
    <tr>
      <td><strong>CVEs</strong></td>
      <td>CVE-2026-109618 (OOM memory corruption), CVE-2026-109619 (stale request state)</td>
    </tr>
    <tr>
      <td><strong>Severity</strong></td>
      <td>CVSS 4.0 <strong>9.2 Critical</strong>, <code class="language-plaintext highlighter-rouge">AV:N/AC:H/AT:P/PR:N/UI:N/VC:H/VI:H/VA:H</code> (PHP CNA, via NVD)</td>
    </tr>
    <tr>
      <td><strong>CWE</strong></td>
      <td>CWE-416 Use After Free (NVD). CISA additionally lists CWE-755, -824, -1325, -1419</td>
    </tr>
    <tr>
      <td><strong>Fixed in</strong></td>
      <td><strong>PHP 8.5.12, 8.4.27, 8.3.36, 8.2.35</strong> (21 Oct 2026). Minimal source backports exist for 7.4, 8.0, 8.1</td>
    </tr>
    <tr>
      <td><strong>Exploited?</strong></td>
      <td><strong>Yes.</strong> CISA KEV, added 26 Oct 2026 (federal due date 29 Oct), with forensic triage required</td>
    </tr>
    <tr>
      <td><strong>Auth / interaction</strong></td>
      <td>None / none</td>
    </tr>
  </tbody>
</table>

<h2 id="description-how-the-bugs-work">Description: how the bugs work</h2>

<p>Both bugs are in how PHP handles a request from setup to teardown. Each request gets its own memory arena. When the request ends, PHP’s cleanup routines walk the request’s internal structures and free them, and the next request starts from a clean slate. Both bugs break the assumption that the slate really is clean.</p>

<p><strong>CVE-2026-109618: out-of-memory corruption.</strong> When PHP runs out of memory it doesn’t return an error up the call stack. It uses <code class="language-plaintext highlighter-rouge">longjmp()</code> to jump straight out of whatever it was doing and abort the request. NVD’s description: some code paths “leave request memory in an uninitialized state before performing an allocation.” If an OOM hits in that window, the structure is only half built, and the cleanup that runs next reads or branches on uninitialized memory as if it were valid. The attacker controls the request, so they can choose where the OOM lands. Qualys’s analysis points at multipart form handling as the practical trigger area.</p>

<p><strong>CVE-2026-109619: stale request state (use-after-free).</strong> PHP keeps several <em>global</em> pointers that temporarily point into request memory. In some paths a global isn’t reset at the end of a request, or it gets repopulated after being reset. The request memory is freed, but the global still points at it. <strong>The next request handled by the same worker</strong> reads or branches on that stale pointer before it’s reinitialized. That’s a classic cross-request use-after-free.</p>

<p>Both bugs fire during request setup and teardown, before your <code class="language-plaintext highlighter-rouge">index.php</code> runs. That’s why a WAF rule tuned to application routes won’t see them, and why patching the app does nothing.</p>

<h2 id="impact">Impact</h2>

<ul>
  <li><strong>Remote code execution</strong> as the web server or PHP worker user (<code class="language-plaintext highlighter-rouge">www-data</code>, <code class="language-plaintext highlighter-rouge">apache</code>, <code class="language-plaintext highlighter-rouge">nginx</code>, and so on).</li>
  <li><strong>Memory disclosure.</strong> Researchers showed unauthenticated reads of PHP worker memory, which can leak session tokens, database passwords and API keys from <em>other users’</em> requests.</li>
  <li><strong>Privilege escalation chains.</strong> Unit 42 reports PHP2Shell chained with a Linux kernel LPE to reach root, including against internet-facing PAN-OS management interfaces (PAN-OS embeds PHP).</li>
</ul>

<h2 id="which-deployments-are-exposed">Which deployments are exposed</h2>

<table>
  <thead>
    <tr>
      <th>SAPI / server</th>
      <th>Status</th>
    </tr>
  </thead>
  <tbody>
    <tr>
      <td><strong>Apache <code class="language-plaintext highlighter-rouge">mod_php</code></strong></td>
      <td><strong>Highest risk.</strong> PHP says a single unauthenticated request can trigger exploitation before app code runs. The demonstrated config is prefork MPM with default limits</td>
    </tr>
    <tr>
      <td><strong>PHP-FPM</strong> (nginx, Apache <code class="language-plaintext highlighter-rouge">proxy_fcgi</code>, Caddy)</td>
      <td>Affected. Startup behavior blocks one OOM path, but PHP validated RCE through the session lifecycle on FPM “under suitable application conditions”</td>
    </tr>
    <tr>
      <td><code class="language-plaintext highlighter-rouge">php-cgi</code>, FrankenPHP, LiteSpeed</td>
      <td>Listed as affected by NCSC-NL and Help Net Security</td>
    </tr>
    <tr>
      <td>Appliances with embedded PHP</td>
      <td>Check the vendor. PAN-OS/Panorama, Trend Micro Apex One/Central, Xerox and X-Cart have published advisories</td>
    </tr>
  </tbody>
</table>

<p><strong>Prerequisites:</strong> network reachability to a PHP endpoint, a pre-fix PHP build, no credentials, and no user interaction. Attack complexity is rated <em>High</em>. Reliability depends on allocator layout, build and worker scheduling, and the stale-state path needs two requests to land on the same worker. Don’t let “high complexity” slow you down: exploitation is confirmed, and Amazon’s MadPot honeypot logged an attempt <strong>four days before disclosure</strong> (17 Oct).</p>

<h2 id="identifying-affected-systems">Identifying affected systems</h2>

<p>Check <strong>the PHP that actually serves requests</strong>, not whatever <code class="language-plaintext highlighter-rouge">php</code> happens to be on <code class="language-plaintext highlighter-rouge">$PATH</code>. Qualys specifically warns that the CLI binary often differs from the FPM or <code class="language-plaintext highlighter-rouge">mod_php</code> build.</p>

<div class="language-bash highlighter-rouge"><div class="highlight"><pre class="highlight"><code><span class="c"># 1) Every PHP binary/package on the host (CLI, FPM, Apache module)</span>
php <span class="nt">-v</span>
<span class="nb">ls</span> /usr/sbin/php-fpm<span class="k">*</span> /usr/lib/apache2/modules/libphp<span class="k">*</span>.so 2&gt;/dev/null
dpkg <span class="nt">-l</span> | <span class="nb">grep</span> <span class="nt">-E</span> <span class="s1">'^ii\s+(php|libapache2-mod-php)'</span> 2&gt;/dev/null <span class="o">||</span> rpm <span class="nt">-qa</span> | <span class="nb">grep</span> <span class="nt">-i</span> <span class="s1">'^php'</span>

<span class="c"># 2) Which SAPI is live? (mod_php shows up as php_module / php7_module)</span>
apachectl <span class="nt">-M</span> 2&gt;/dev/null | <span class="nb">grep</span> <span class="nt">-i</span> php
systemctl list-units <span class="nt">--type</span><span class="o">=</span>service | <span class="nb">grep</span> <span class="nt">-i</span> fpm

<span class="c"># 3) Version as seen by the web tier: drop a temporary file you DELETE afterwards</span>
<span class="nb">echo</span> <span class="s1">'&lt;?php echo PHP_VERSION, " ", PHP_SAPI;'</span> | <span class="nb">sudo tee</span> /var/www/html/.v.php <span class="o">&gt;</span>/dev/null
curl <span class="nt">-s</span> http://127.0.0.1/.v.php<span class="p">;</span> <span class="nb">sudo rm</span> <span class="nt">-f</span> /var/www/html/.v.php

<span class="c"># 4) Containers: check the image, not the host</span>
docker ps <span class="nt">--format</span> <span class="s1">'{{.Image}}'</span> | <span class="nb">sort</span> <span class="nt">-u</span> | xargs <span class="nt">-I</span><span class="o">{}</span> sh <span class="nt">-c</span> <span class="s1">'echo -n "{}: "; docker run --rm --entrypoint php {} -r "echo PHP_VERSION;" 2&gt;/dev/null; echo'</span>
</code></pre></div></div>

<p><strong>Distro backports:</strong> RHEL, Debian and Ubuntu often fix PHP <strong>without changing the upstream version string</strong>. A RHEL 9 host can say <code class="language-plaintext highlighter-rouge">8.4.x</code> and still be patched or unpatched. Check the package changelog or advisory instead: <code class="language-plaintext highlighter-rouge">rpm -q --changelog php-common | grep -E 'CVE-2026-10961[89]'</code>, or <code class="language-plaintext highlighter-rouge">apt changelog php8.x-common | grep ...</code>. Known errata include RHSA-2026:97880, USN-8948-1/2, DLA/DSA for php8.2/8.4, ALAS2023-2026-3301..3304, and SUSE-SU-2026:4843-1/4844-1.</p>

<p>Fleet scanners: Qualys QID 533086 (PHP2Shell). Qualys warns that <strong>QID 533068 is intrusive and may crash unpatched systems</strong>, so don’t run it against production.</p>

<h2 id="safe-validation-in-an-isolated-lab">Safe validation in an isolated lab</h2>

<p>You don’t need to trigger the bug to confirm you’re fixed. A useful lab exercise is checking that your <strong>detection and patch process</strong> works end to end:</p>

<ol>
  <li>On an isolated VM with no route to production, deploy a pre-fix image (for example <code class="language-plaintext highlighter-rouge">php:8.4.26-apache</code>) and the fixed one (<code class="language-plaintext highlighter-rouge">php:8.4.27-apache</code>) side by side.</li>
  <li>Run your version checks and scanner against both. The scanner should flag only the old one. If it flags neither, your tooling is reading the wrong binary.</li>
  <li>Apply the interim config below to the old one and confirm it shows up in <code class="language-plaintext highlighter-rouge">php -i</code> / <code class="language-plaintext highlighter-rouge">phpinfo()</code>.</li>
  <li>Generate benign web-shell-like activity (e.g. <code class="language-plaintext highlighter-rouge">www-data</code> running <code class="language-plaintext highlighter-rouge">id</code> via a test script) and confirm your EDR or SIEM rule from the detection section fires.</li>
</ol>

<h2 id="detection-and-threat-hunting">Detection and threat hunting</h2>

<p>PHP2Shell traffic hits the engine before any app routing, so focus on <strong>post-exploitation behavior</strong> of the PHP worker.</p>

<ul>
  <li><strong>Child processes of PHP workers.</strong> <code class="language-plaintext highlighter-rouge">php-fpm</code>, <code class="language-plaintext highlighter-rouge">apache2</code>/<code class="language-plaintext highlighter-rouge">httpd</code> (<code class="language-plaintext highlighter-rouge">mod_php</code>) or <code class="language-plaintext highlighter-rouge">www-data</code> spawning <code class="language-plaintext highlighter-rouge">sh</code>, <code class="language-plaintext highlighter-rouge">bash</code>, <code class="language-plaintext highlighter-rouge">curl</code>, <code class="language-plaintext highlighter-rouge">wget</code>, <code class="language-plaintext highlighter-rouge">python</code>, <code class="language-plaintext highlighter-rouge">perl</code>, <code class="language-plaintext highlighter-rouge">nc</code>, <code class="language-plaintext highlighter-rouge">base64 -d</code>, or <code class="language-plaintext highlighter-rouge">chmod +x</code> in <code class="language-plaintext highlighter-rouge">/tmp</code>, <code class="language-plaintext highlighter-rouge">/dev/shm</code> or <code class="language-plaintext highlighter-rouge">/var/tmp</code>. Unit 42’s Cortex XQL hunting query follows this model. They note it isn’t PHP2Shell-specific.
    <div class="language-bash highlighter-rouge"><div class="highlight"><pre class="highlight"><code><span class="c"># quick live check: anything running under a PHP worker that isn't a PHP worker</span>
<span class="k">for </span>p <span class="k">in</span> <span class="si">$(</span>pgrep <span class="nt">-f</span> <span class="s1">'php-fpm|apache2|httpd'</span><span class="si">)</span><span class="p">;</span> <span class="k">do </span>ps <span class="nt">--ppid</span> <span class="s2">"</span><span class="nv">$p</span><span class="s2">"</span> <span class="nt">-o</span> <span class="nv">pid</span><span class="o">=</span>,user<span class="o">=</span>,cmd<span class="o">=</span> <span class="p">;</span> <span class="k">done</span> | <span class="nb">grep</span> <span class="nt">-vE</span> <span class="s1">'php-fpm|apache2|httpd'</span>
</code></pre></div>    </div>
  </li>
  <li><strong>Session files.</strong> Public PoCs stage attacker-controlled data into a real session file. Review <code class="language-plaintext highlighter-rouge">session.save_path</code> (often <code class="language-plaintext highlighter-rouge">/var/lib/php/sessions</code>) for unusually large or binary session files.</li>
  <li><strong>Worker crashes.</strong> Bursts of <code class="language-plaintext highlighter-rouge">child exited on signal 11 (SIGSEGV)</code> in FPM logs, or <code class="language-plaintext highlighter-rouge">AH00052: child pid ... exit signal Segmentation fault</code> in Apache’s error log, can mean failed attempts.</li>
  <li><strong>Network.</strong> Palo Alto NGFW threat ID 543004, plus Cloudflare’s free managed WAF rules for PHP2Shell. Watch for oversized or unusual <code class="language-plaintext highlighter-rouge">multipart/form-data</code> POSTs to endpoints that never take uploads.</li>
  <li><strong>Cloud credentials.</strong> Requests to <code class="language-plaintext highlighter-rouge">169.254.169.254</code> from the web tier right after anomalous POSTs.</li>
</ul>

<p>Because exploitation is active and pre-dates disclosure, <strong>treat any internet-facing instance unpatched since 17 October as potentially compromised</strong>, and hunt before you declare it clean.</p>

<h2 id="remediation">Remediation</h2>

<ol>
  <li><strong>Upgrade PHP</strong> to 8.5.12 / 8.4.27 / 8.3.36 / 8.2.35 or later (later monthly releases include these fixes), or install your distro’s errata. PHP’s own words: there are <strong>no complete mitigations</strong> short of updating.</li>
  <li><strong>Restart every PHP worker.</strong> A new package on disk does nothing for an FPM master or Apache parent started last month: <code class="language-plaintext highlighter-rouge">systemctl restart php*-fpm apache2</code> (or <code class="language-plaintext highlighter-rouge">httpd</code>).</li>
  <li><strong>Rebuild containers</strong> from patched base images and redeploy. Pinned tags like <code class="language-plaintext highlighter-rouge">php:8.4.26-*</code> keep the bug.</li>
  <li><strong>Get off EOL PHP (≤ 8.1).</strong> Source backports exist, but no official binaries do.</li>
  <li><strong>Appliances:</strong> apply vendor fixes (PAN-OS/Panorama, Trend Micro, Xerox and others) and take management interfaces off the internet.</li>
</ol>

<h3 id="interim-hardening-if-you-truly-cant-patch-today">Interim hardening (if you truly can’t patch today)</h3>

<p>These reduce attack surface <strong>but aren’t fixes</strong>:</p>

<div class="language-ini highlighter-rouge"><div class="highlight"><pre class="highlight"><code><span class="c">; php.ini: reduce exposure until the update is applied
</span><span class="py">session.upload_progress.enabled</span><span class="w"> </span><span class="p">=</span><span class="w"> </span><span class="s">0   ; NCSC-NL: keep 0 until the next regular PHP release</span>
<span class="py">post_max_size</span><span class="w"> </span><span class="p">=</span><span class="w"> </span><span class="s">8M                    ; or 32M if you accept uploads, with upload_max_filesize below it</span>
<span class="c">; PHP 8.5 only:
</span><span class="py">fatal_error_backtraces</span><span class="w"> </span><span class="p">=</span><span class="w"> </span><span class="s">Off</span>
</code></pre></div></div>

<ul>
  <li>For FPM, <code class="language-plaintext highlighter-rouge">pm.max_requests = 1</code> recycles a worker after each request, which breaks the cross-request stale-state path. Expect a significant performance cost.</li>
  <li>Block <code class="language-plaintext highlighter-rouge">multipart/form-data</code> at the proxy for routes that never accept uploads.</li>
  <li>Note that the <code class="language-plaintext highlighter-rouge">session.upload_progress</code> setting is tied to companion CVE-2026-115560 in PHP’s 23 Oct advisory, so treat it as defense in depth and not a fix for 109618/109619.</li>
  <li>Hiding errors (<code class="language-plaintext highlighter-rouge">display_errors=Off</code>, <code class="language-plaintext highlighter-rouge">@</code>) does <strong>not</strong> prevent the underlying memory access.</li>
</ul>

<h2 id="references">References</h2>

<ul>
  <li><a href="https://nvd.nist.gov/vuln/detail/CVE-2026-109618">CVE-2026-109618, NVD</a> · <a href="https://nvd.nist.gov/vuln/detail/CVE-2026-109619">CVE-2026-109619, NVD</a></li>
  <li><a href="https://www.cisa.gov/known-exploited-vulnerabilities-catalog">CISA KEV catalog</a> · <a href="https://www.cisa.gov/news-events/alerts/2026/10/21/critical-vulnerabilities-php-require-immediate-patching">CISA alert, 21 Oct 2026</a></li>
  <li><a href="https://www.php.net/archive/2026.php#2026-10-21-5">PHP security release announcement</a> · <a href="https://www.php.net/archive/2026.php#2026-10-23-1">PHP follow-up advisory</a></li>
  <li><a href="https://github.com/NCSC-NL/php2shell">NCSC-NL operational info (versions, vendor status)</a></li>
  <li><a href="https://unit42.paloaltonetworks.com/cve-2026-109618-cve-2026-109619/">Unit 42 threat brief (exploitation, hunting query)</a></li>
  <li><a href="https://blog.qualys.com/vulnerabilities-threat-research/2026/10/20/php-announces-critical-security-release-for-two-vulnerabilities">Qualys: find, mitigate and patch</a></li>
  <li><a href="https://www.helpnetsecurity.com/2026/10/21/php2shell-php-vulnerabilities-patched-cve-2026-109618-cve-2026-109619/">Help Net Security coverage</a></li>
  <li><a href="https://security.paloaltonetworks.com/CVE-2026-109618">Palo Alto Networks advisory</a></li>
</ul>

<p><em>Verify every version against the vendor advisory before acting. Published 20 Nov 2026.</em></p>]]></content><author><name>Aniket Tomar</name></author><category term="cve" /><category term="rce" /><category term="php" /><category term="use-after-free" /><category term="blue-team" /><summary type="html"><![CDATA[Two memory-safety flaws in PHP's request lifecycle (CVSS 4.0 9.2, on CISA KEV) allow unauthenticated RCE before any application code runs. Root cause, exposure checks, detection and the fix.]]></summary><media:thumbnail xmlns:media="http://search.yahoo.com/mrss/" url="https://binary0101devil.pages.dev/assets/img/php2shell-cve-2026-109618-109619.svg" /><media:content medium="image" url="https://binary0101devil.pages.dev/assets/img/php2shell-cve-2026-109618-109619.svg" xmlns:media="http://search.yahoo.com/mrss/" /></entry><entry><title type="html">CVE-2026-40217: LiteLLM guardrail sandbox escape, now on CISA KEV</title><link href="https://binary0101devil.pages.dev/blog/2026/11/19/cve-2026-40217-litellm/" rel="alternate" type="text/html" title="CVE-2026-40217: LiteLLM guardrail sandbox escape, now on CISA KEV" /><published>2026-11-19T00:00:00+05:30</published><updated>2026-11-19T00:00:00+05:30</updated><id>https://binary0101devil.pages.dev/blog/2026/11/19/cve-2026-40217-litellm</id><content type="html" xml:base="https://binary0101devil.pages.dev/blog/2026/11/19/cve-2026-40217-litellm/"><![CDATA[<p>LiteLLM is the open-source proxy a lot of teams put in front of their LLM providers: one OpenAI-compatible API, with virtual keys, budgets and logging, fanning out to OpenAI, Anthropic, Bedrock, Azure and the rest. That makes the box running it a credential vault. It holds every upstream provider key, plus the master key and the database that issues all the virtual keys. CVE-2026-40217 is code execution on that box. It was disclosed in April, but CISA only added it to the KEV catalog on 18 November 2026, so plenty of instances that were “patch later” because it needs admin rights are now overdue.</p>

<blockquote>
  <p><strong>Scope and ethics.</strong> This post covers architecture, exposure checks, detection and remediation for LiteLLM deployments you own or are authorized to assess. It contains no sandbox-escape code and no requests that trigger the bug. The only network check below is an unauthenticated request that confirms whether a route is reachable on your own proxy.</p>
</blockquote>

<h2 id="at-a-glance">At a glance</h2>

<table>
  <thead>
    <tr>
      <th> </th>
      <th> </th>
    </tr>
  </thead>
  <tbody>
    <tr>
      <td><strong>CVE</strong></td>
      <td>CVE-2026-40217 (GHSA-wxxx-gvqv-xp7p, X41-2026-001)</td>
    </tr>
    <tr>
      <td><strong>Severity</strong></td>
      <td>CVSS 3.1 <strong>8.8 High</strong>, <code class="language-plaintext highlighter-rouge">AV:N/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H</code> (NVD and GitHub advisory). The GitHub advisory also gives CVSS 4.0 <strong>7.5</strong> with <code class="language-plaintext highlighter-rouge">PR:H</code>; X41 gives CVSS 4.0 <strong>8.7</strong> with <code class="language-plaintext highlighter-rouge">PR:L</code></td>
    </tr>
    <tr>
      <td><strong>CWE</strong></td>
      <td>CWE-420 Unprotected Alternate Channel, CWE-94 Code Injection (NVD); CWE-913 (GitHub advisory)</td>
    </tr>
    <tr>
      <td><strong>Affected</strong></td>
      <td><code class="language-plaintext highlighter-rouge">litellm</code> <strong>1.81.8 up to, not including, 1.83.10</strong> (GitHub advisory and NVD CPE data)</td>
    </tr>
    <tr>
      <td><strong>Fixed in</strong></td>
      <td><strong>1.83.10</strong> per the advisory’s structured data; the advisory text says 1.83.11. Because of related bugs (below), I’d target <strong>1.83.14 or later</strong>, and ideally current</td>
    </tr>
    <tr>
      <td><strong>Exploited?</strong></td>
      <td><strong>Yes</strong>, per CISA KEV (added 18 Nov 2026, federal due date 2 Dec 2026). I found no public report describing the attacks themselves</td>
    </tr>
    <tr>
      <td><strong>Auth / interaction</strong></td>
      <td>Requires a credential that can call the endpoint, which is a proxy-admin / master key in default configs / none</td>
    </tr>
  </tbody>
</table>

<h2 id="description-how-the-bug-works">Description: how the bug works</h2>

<p>LiteLLM’s <strong>guardrails</strong> let you run checks on prompts and responses. One kind is a <strong>custom-code guardrail</strong>: an admin supplies Python that inspects or transforms traffic. To help admins write those, the proxy exposes <code class="language-plaintext highlighter-rouge">POST /guardrails/test_custom_code</code>, which <strong>runs the submitted Python on the server</strong> so you can see what it does.</p>

<p>Running user-supplied Python safely is very hard, and LiteLLM’s original approach was a hand-rolled sandbox. According to X41 D-Sec, who found the bug, the sandbox decided what was allowed by applying <strong>regex filters to the source text</strong>, blocking strings for dangerous attributes and built-ins. That’s the design flaw. The filter only sees the literal text you submit, while Python can build names at runtime and manipulate its own code objects. X41 showed that by rewriting <strong>CPython bytecode</strong>, submitted code can reach attributes the text filter never saw, giving arbitrary code execution in the proxy process. CSA’s later write-up adds that the guardrail code ran through <code class="language-plaintext highlighter-rouge">exec()</code> with built-ins still reachable.</p>

<p>That’s why CISA and NVD classify it as an <strong>unprotected alternate channel</strong> (CWE-420): a debug/test endpoint that offers a second route into server-side code execution next to the “real” guardrail path.</p>

<p>The fix, per the GitHub advisory, <strong>replaced the hand-rolled sandbox with <code class="language-plaintext highlighter-rouge">RestrictedPython</code></strong>.</p>

<h2 id="impact">Impact</h2>

<ul>
  <li><strong>Code execution as the LiteLLM process.</strong> In the default Docker image that’s <strong>root</strong> (GitHub advisory, X41).</li>
  <li><strong>Secrets exposure.</strong> The process has every upstream provider API key, the LiteLLM master key and salt key, and the database connection string, usually as environment variables. Assume all of them are readable once code runs.</li>
  <li><strong>Abuse of the AI estate.</strong> Stolen provider keys mean attacker-paid-by-you inference, access to fine-tuned models, files and assistants on those accounts, and a pivot into whatever cloud account the gateway runs in.</li>
  <li><strong>Tampering.</strong> Someone with code execution on the gateway can alter guardrails, logging and routing, so it can read or modify the prompts and responses of every application behind it.</li>
</ul>

<h2 id="prerequisites-and-the-authentication-question">Prerequisites and the authentication question</h2>

<p>Sources describe the privilege needed differently, so be careful:</p>

<ul>
  <li><strong>GitHub advisory (vendor):</strong> “Reaching the endpoint requires a proxy-admin credential in default configurations.” Its CVSS 4.0 vector says <code class="language-plaintext highlighter-rouge">PR:H</code>.</li>
  <li><strong>NVD and the advisory’s CVSS 3.1 vector:</strong> <code class="language-plaintext highlighter-rouge">PR:L</code>. NVD’s text just says “remote attackers”.</li>
  <li><strong>X41:</strong> analysis says “authenticated users” with <code class="language-plaintext highlighter-rouge">PR:L</code>, but its workaround section says the endpoint “requires authentication with the master key”.</li>
</ul>

<p>The practical answer is that admin access to LiteLLM is <strong>easier to get than it sounds</strong>:</p>

<ul>
  <li>The Cloud Security Alliance documented a chain in which <strong>CVE-2026-47101</strong> (virtual keys created with unvalidated <code class="language-plaintext highlighter-rouge">allowed_routes</code>) and <strong>CVE-2026-47102</strong> (users promoting their own role via <code class="language-plaintext highlighter-rouge">/user/update</code>) let a low-privilege user reach admin, with CVE-2026-40217 as the final step to code execution. CSA lists LiteLLM <strong>v1.83.14-stable</strong> as the fix for all three.</li>
  <li>LiteLLM has had other KEV-listed bugs this year: <strong>CVE-2026-42208</strong>, a pre-auth SQL injection in API-key verification that exposes the key database (KEV, 8 May 2026), and <strong>CVE-2026-42271</strong>, command injection in the MCP test endpoints (KEV, 8 June 2026).</li>
  <li>Master keys tend to leak: they’re in <code class="language-plaintext highlighter-rouge">.env</code> files, CI variables, Helm values and notebooks.</li>
</ul>

<p>So treat this as reachable for anyone who can get <strong>any</strong> foothold on your proxy, not only your admins.</p>

<h2 id="identifying-affected-systems">Identifying affected systems</h2>

<p>Inventory every LiteLLM deployment, including the “temporary” ones on dev laptops and GPU boxes.</p>

<div class="language-bash highlighter-rouge"><div class="highlight"><pre class="highlight"><code><span class="c"># 1) Python installs</span>
pip show litellm 2&gt;/dev/null | <span class="nb">grep</span> <span class="nt">-E</span> <span class="s1">'^(Name|Version|Location)'</span>
python3 <span class="nt">-c</span> <span class="s2">"import importlib.metadata as m; print(m.version('litellm'))"</span> 2&gt;/dev/null

<span class="c"># 2) Containers: which LiteLLM images are running, and what version is inside?</span>
docker ps <span class="nt">--format</span> <span class="s1">'{{.ID}} {{.Image}}'</span> | <span class="nb">grep</span> <span class="nt">-i</span> litellm | <span class="k">while </span><span class="nb">read</span> <span class="nt">-r</span> <span class="nb">id </span>img<span class="p">;</span> <span class="k">do
  </span><span class="nb">echo</span> <span class="nt">-n</span> <span class="s2">"</span><span class="nv">$img</span><span class="s2"> -&gt; "</span><span class="p">;</span> docker <span class="nb">exec</span> <span class="s2">"</span><span class="nv">$id</span><span class="s2">"</span> pip show litellm 2&gt;/dev/null | <span class="nb">awk</span> <span class="s1">'/^Version/{print $2}'</span>
<span class="k">done</span>

<span class="c"># 3) Kubernetes</span>
kubectl get pods <span class="nt">-A</span> <span class="nt">-o</span> <span class="nv">jsonpath</span><span class="o">=</span><span class="s1">'{range .items[*]}{.metadata.namespace}/{.metadata.name}{"\t"}{range .spec.containers[*]}{.image}{" "}{end}{"\n"}{end}'</span> | <span class="nb">grep</span> <span class="nt">-i</span> litellm

<span class="c"># 4) Does the process run as root? (affects impact)</span>
docker ps <span class="nt">-q</span> | xargs <span class="nt">-r</span> docker inspect <span class="nt">--format</span> <span class="s1">'{{.Name}} user={{.Config.User}}'</span> | <span class="nb">grep</span> <span class="nt">-i</span> litellm
</code></pre></div></div>

<p>Vulnerable range: <strong>1.81.8 ≤ version &lt; 1.83.10</strong>. Anything below 1.83.14 is also exposed to the related privilege-escalation bugs CSA describes.</p>

<p><strong>Red Hat products</strong> bundle LiteLLM. NVD lists Ansible Automation Platform 2.0 up to 2.6-9, OpenShift AI 3.3 up to 3.3.4 and OpenShift Lightspeed as affected; see RHSA-2026:24866 and RHSA-2026:30056.</p>

<p><strong>Is the endpoint reachable?</strong> On a proxy you own, send a request <strong>with no credentials and no body</strong>. You’re only checking routing and auth, not running anything:</p>

<div class="language-bash highlighter-rouge"><div class="highlight"><pre class="highlight"><code><span class="c"># Expect 401/403 (route exists, auth enforced) or 404 (blocked at your proxy / not present).</span>
<span class="c"># A 2xx here would mean auth isn't being enforced: investigate immediately.</span>
curl <span class="nt">-s</span> <span class="nt">-o</span> /dev/null <span class="nt">-w</span> <span class="s2">"test_custom_code: %{http_code}</span><span class="se">\n</span><span class="s2">"</span> <span class="se">\</span>
  <span class="nt">-X</span> POST https://litellm.internal.example.com/guardrails/test_custom_code
</code></pre></div></div>

<p>Also check where the admin API and UI are reachable from. If the proxy port (4000 by default) is open to the internet or to every workload in the cluster, fix that regardless of version.</p>

<h2 id="safe-validation-in-an-isolated-lab">Safe validation in an isolated lab</h2>

<ol>
  <li>In an isolated network namespace, run an image with a vulnerable version (anything from 1.81.8 to 1.83.9) and a current one with a throwaway master key and <strong>no real provider keys</strong>.</li>
  <li>Confirm your SCA tooling (Dependabot, Trivy, Grype, pip-audit) flags <code class="language-plaintext highlighter-rouge">litellm</code> in the old image for GHSA-wxxx-gvqv-xp7p / CVE-2026-40217. If it doesn’t, check whether it scans Python packages inside container images.</li>
  <li>Add the reverse-proxy block below in front of the old one and re-run the unauthenticated <code class="language-plaintext highlighter-rouge">curl</code> above. It should now return 403/404 at the proxy layer, never reaching LiteLLM.</li>
  <li>Make a normal admin API call and confirm it lands in your SIEM with the key alias, source IP and route, which is what you’ll need for the hunt below.</li>
</ol>

<h2 id="detection-and-threat-hunting">Detection and threat hunting</h2>

<p>No IOCs have been published for the November exploitation that put this on KEV, so hunt on behavior.</p>

<ul>
  <li><strong>Requests to the test endpoint.</strong> Search reverse-proxy, ingress and LiteLLM logs for <code class="language-plaintext highlighter-rouge">POST /guardrails/test_custom_code</code>. Legitimate use should be rare and come from known admins. Calls from unexpected IPs, at odd hours or with large bodies are worth a look.</li>
  <li><strong>Neighboring endpoints from the same chain.</strong> <code class="language-plaintext highlighter-rouge">/user/update</code> and <code class="language-plaintext highlighter-rouge">/user/bulk_update</code> calls that change roles, <code class="language-plaintext highlighter-rouge">/key/generate</code> or key updates with unusual <code class="language-plaintext highlighter-rouge">allowed_routes</code>, and the MCP test routes <code class="language-plaintext highlighter-rouge">/mcp-rest/test/connection</code> and <code class="language-plaintext highlighter-rouge">/mcp-rest/test/tools/list</code>.</li>
  <li><strong>Process lineage.</strong> The LiteLLM Python process (usually <code class="language-plaintext highlighter-rouge">uvicorn</code>/<code class="language-plaintext highlighter-rouge">python ... litellm</code>) spawning <code class="language-plaintext highlighter-rouge">sh</code>, <code class="language-plaintext highlighter-rouge">bash</code>, <code class="language-plaintext highlighter-rouge">curl</code>, <code class="language-plaintext highlighter-rouge">wget</code>, <code class="language-plaintext highlighter-rouge">nc</code> or <code class="language-plaintext highlighter-rouge">python -c</code>, or writing files under <code class="language-plaintext highlighter-rouge">/tmp</code>. In Kubernetes, watch for shells inside the LiteLLM pod that no <code class="language-plaintext highlighter-rouge">kubectl exec</code> session explains, and for unexpected outbound connections from the pod.</li>
  <li><strong>Environment and secret access.</strong> Reads of <code class="language-plaintext highlighter-rouge">/proc/&lt;pid&gt;/environ</code>, the <code class="language-plaintext highlighter-rouge">.env</code> file or mounted secrets by anything other than the main process.</li>
  <li><strong>API key use.</strong> This is the most important hunt for an AI gateway:
    <ul>
      <li>In LiteLLM’s spend/usage logs, look for virtual keys used from new IPs, new models, or volumes far outside their history, and keys you don’t recognize.</li>
      <li><strong>At each upstream provider</strong> (OpenAI, Anthropic, Azure, Bedrock, Vertex), review usage and audit logs for the keys the gateway holds. Calls that didn’t come through your gateway’s egress IPs mean the key has left the building.</li>
      <li>Watch for new provider-side assets: files, fine-tunes, assistants or batch jobs nobody created.</li>
    </ul>
  </li>
  <li><strong>Supply chain, separately.</strong> CSA notes that malicious <code class="language-plaintext highlighter-rouge">litellm</code> <strong>1.82.7 and 1.82.8</strong> were published to PyPI on 24 March 2026. If either was ever installed, treat the host’s secrets as compromised, independently of this CVE.</li>
</ul>

<h2 id="remediation">Remediation</h2>

<ol>
  <li><strong>Upgrade LiteLLM.</strong> The advisory marks 1.83.10 as the first patched release (its text says 1.83.11). Given the related CVE-2026-47101/-47102/-42271 fixes, go to <strong>1.83.14 or later</strong>, and preferably the current release. Rebuild and redeploy images; don’t just <code class="language-plaintext highlighter-rouge">pip install</code> inside a running container.</li>
  <li><strong>Block the endpoint at the edge anyway.</strong> Even on fixed versions, few teams need to test guardrail code in production:
    <div class="language-nginx highlighter-rouge"><div class="highlight"><pre class="highlight"><code><span class="c1"># nginx in front of LiteLLM</span>
<span class="k">location</span> <span class="p">=</span> <span class="n">/guardrails/test_custom_code</span> <span class="p">{</span> <span class="kn">return</span> <span class="mi">403</span><span class="p">;</span> <span class="p">}</span>
</code></pre></div>    </div>
  </li>
  <li><strong>Rotate secrets</strong> if the proxy was on a vulnerable version and reachable by anyone beyond a small admin group:
    <ul>
      <li>every upstream provider key the proxy holds,</li>
      <li>the LiteLLM master key and salt key (check LiteLLM’s docs before rotating the salt key, since it encrypts stored credentials),</li>
      <li>the database password and any cloud credentials in the environment,</li>
      <li>virtual keys issued to applications, prioritizing high-budget ones.</li>
    </ul>
  </li>
  <li><strong>Reduce blast radius.</strong> Run the container as a non-root user (X41 suggests <code class="language-plaintext highlighter-rouge">--user</code>), keep the admin API and UI on a management network, use a secrets manager instead of plain env vars where you can, and set per-key budgets and rate limits so a stolen key is noisy.</li>
  <li><strong>Audit admins.</strong> Review who holds proxy-admin roles and the master key, and remove anything that shouldn’t be there.</li>
</ol>

<h3 id="interim-mitigation-if-you-cant-upgrade-today">Interim mitigation (if you can’t upgrade today)</h3>

<p>Both the vendor and X41 give the same workaround: <strong>block <code class="language-plaintext highlighter-rouge">POST /guardrails/test_custom_code</code> at your reverse proxy or API gateway</strong>. Pair it with network restrictions on the admin interface and limiting master-key distribution. That narrows this specific route; it doesn’t fix the related privilege-escalation bugs, so keep the upgrade on the schedule.</p>

<h2 id="references">References</h2>

<ul>
  <li><a href="https://nvd.nist.gov/vuln/detail/CVE-2026-40217">CVE-2026-40217, NVD</a> · <a href="https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-40217">CISA KEV entry</a></li>
  <li><a href="https://github.com/BerriAI/litellm/security/advisories/GHSA-wxxx-gvqv-xp7p">LiteLLM security advisory GHSA-wxxx-gvqv-xp7p</a> · <a href="https://github.com/advisories/GHSA-wxxx-gvqv-xp7p">GitHub Advisory Database entry</a></li>
  <li><a href="https://www.x41-dsec.de/lab/advisories/x41-2026-001-litellm/">X41 D-Sec advisory X41-2026-001 (root cause, timeline)</a></li>
  <li><a href="https://labs.cloudsecurityalliance.org/research/csa-research-note-litellm-ai-gateway-attack-chain-20260617-c/">Cloud Security Alliance research note: LiteLLM AI gateway attack chain</a></li>
  <li><a href="https://access.redhat.com/security/cve/CVE-2026-40217">Red Hat CVE page</a></li>
  <li><a href="https://github.com/BerriAI/litellm/releases/tag/v1.83.10-stable">LiteLLM v1.83.10-stable release</a></li>
</ul>

<p><em>Verify every version against the vendor advisory before acting. Published 19 Nov 2026.</em></p>]]></content><author><name>Aniket Tomar</name></author><category term="cve" /><category term="rce" /><category term="python" /><category term="ai-security" /><category term="blue-team" /><summary type="html"><![CDATA[A sandbox escape in LiteLLM's custom-code guardrail test endpoint gives code execution on the AI gateway (CVSS 8.8, KEV). Exposure checks, key rotation, hunting and the fix.]]></summary><media:thumbnail xmlns:media="http://search.yahoo.com/mrss/" url="https://binary0101devil.pages.dev/assets/img/cve-2026-40217-litellm.svg" /><media:content medium="image" url="https://binary0101devil.pages.dev/assets/img/cve-2026-40217-litellm.svg" xmlns:media="http://search.yahoo.com/mrss/" /></entry><entry><title type="html">CVE-2026-119618: Apache HTTP/2 race condition leading to RCE</title><link href="https://binary0101devil.pages.dev/blog/2026/11/18/cve-2026-119618-apache-http2-rce/" rel="alternate" type="text/html" title="CVE-2026-119618: Apache HTTP/2 race condition leading to RCE" /><published>2026-11-18T00:00:00+05:30</published><updated>2026-11-18T00:00:00+05:30</updated><id>https://binary0101devil.pages.dev/blog/2026/11/18/cve-2026-119618-apache-http2-rce</id><content type="html" xml:base="https://binary0101devil.pages.dev/blog/2026/11/18/cve-2026-119618-apache-http2-rce/"><![CDATA[<p>The second of three critical RCE flaws this week. This one needs HTTP/2 enabled, and there’s no confirmed in-the-wild exploitation yet — but it’s unauthenticated and rated 9.8, so it belongs on the short list. See also the actively-exploited <a href="/blog/2026/11/20/cve-2026-129448-apache-roller-rce/">Apache Roller RCE</a>, the <a href="/blog/2026/11/16/cve-2026-98432-linux-kernel-tcp-rce/">Linux kernel TCP flaw</a>, and the <a href="/blog/2026/11/14/top-rce-cves-2026/">2026 RCE field guide</a>.</p>

<blockquote>
  <p><strong>Scope and ethics.</strong> Exposure <strong>detection</strong> and remediation only, for systems you own or are authorized to test. No weaponized exploit here.</p>
</blockquote>

<h2 id="description">Description</h2>

<p>A race condition (CWE-362) in the <code class="language-plaintext highlighter-rouge">assign_headers()</code> function of <strong>mod_http2</strong>. Concurrent worker threads access a shared header table without proper synchronization, which corrupts heap memory. Crafted concurrent HTTP/2 requests can turn that corruption into remote code execution in the web-server process.</p>

<h2 id="impact">Impact</h2>

<p>Heap corruption in <code class="language-plaintext highlighter-rouge">httpd</code> worker processes, leading to potential RCE and denial of service. The blast radius is every internet-facing Apache front end and reverse proxy that speaks HTTP/2 — often the most exposed box in the estate.</p>

<h2 id="prerequisites-for-the-attack-to-work">Prerequisites (for the attack to work)</h2>

<ul>
  <li><strong>HTTP/2 enabled</strong> via the <code class="language-plaintext highlighter-rouge">Protocols</code> directive (<code class="language-plaintext highlighter-rouge">h2</code> over TLS or <code class="language-plaintext highlighter-rouge">h2c</code> cleartext). HTTP/1.1-only servers are <strong>not</strong> affected.</li>
  <li>A vulnerable build: <strong>Apache httpd 2.4.55 – 2.4.70</strong>, or standalone <strong>mod_h2 2.0.8 – 2.0.42</strong>. RHEL 9/10 and third-party products that bundle Apache with HTTP/2 are also in scope.</li>
  <li><strong>No authentication, no user interaction.</strong></li>
</ul>

<h2 id="steps-to-reproduce-exposure-validation">Steps to reproduce (exposure validation)</h2>

<div class="language-bash highlighter-rouge"><div class="highlight"><pre class="highlight"><code><span class="c"># 1) Version of the running server</span>
httpd <span class="nt">-v</span> 2&gt;/dev/null <span class="o">||</span> apachectl <span class="nt">-v</span>

<span class="c"># 2) Is mod_http2 actually loaded?</span>
apachectl <span class="nt">-M</span> 2&gt;/dev/null | <span class="nb">grep</span> <span class="nt">-i</span> http2

<span class="c"># 3) Any vhost negotiating h2/h2c anywhere in the config tree?</span>
<span class="nb">grep</span> <span class="nt">-RniE</span> <span class="s2">"Protocols .*h2c?</span><span class="se">\b</span><span class="s2">"</span> /etc/httpd /etc/apache2 2&gt;/dev/null

<span class="c"># 4) From the outside: does the host negotiate HTTP/2?</span>
curl <span class="nt">-sI</span> <span class="nt">--http2</span> https://www.example.com | <span class="nb">head</span> <span class="nt">-n1</span>   <span class="c"># "HTTP/2 200" = negotiated</span>
</code></pre></div></div>

<p>Module loaded <strong>+</strong> a <code class="language-plaintext highlighter-rouge">Protocols</code> line with <code class="language-plaintext highlighter-rouge">h2</code>/<code class="language-plaintext highlighter-rouge">h2c</code> <strong>+</strong> version ≤ 2.4.70 = exposed. To study the memory corruption itself, reproduce it against your own 2.4.70 build in a lab using the public research linked below.</p>

<h2 id="remediation">Remediation</h2>

<ol>
  <li><strong>Upgrade to Apache HTTP Server 2.4.71</strong> (released 2026-11-11). The same release also fixes CVE-2026-106023 (mod_md) and CVE-2026-109163 (mod_vhost_alias).</li>
  <li>Interim mitigations — they reduce exposure but are <strong>not</strong> a substitute for the patch:
    <ul>
      <li>Restrict <code class="language-plaintext highlighter-rouge">Protocols</code> to <code class="language-plaintext highlighter-rouge">http/1.1</code>.</li>
      <li>Stop loading <code class="language-plaintext highlighter-rouge">mod_http2</code> if you don’t actually need HTTP/2.</li>
      <li>Terminate HTTP/2 at an unaffected load balancer or CDN in front of the origin.</li>
      <li>Apply least privilege: restrict outbound access from web servers and don’t mount unnecessary credentials into web containers.</li>
    </ul>
  </li>
  <li><strong>Monitoring:</strong> Apache published no IOCs. Treat <code class="language-plaintext highlighter-rouge">httpd</code> child <strong>segfaults in the error log</strong>, anomalous HTTP/2 traffic patterns, and unexpected processes under the web-server account as possible exploitation signals.</li>
</ol>

<h2 id="references">References</h2>

<ul>
  <li><a href="https://nvd.nist.gov/vuln/detail/cve-2026-119618">NVD — CVE-2026-119618</a></li>
  <li><a href="https://httpd.apache.org/security/vulnerabilities_24.html">Apache httpd 2.4 security advisories</a></li>
  <li><a href="https://github.com/apache/httpd/commit/824beab9b9424d6cfb29595248fdbf6c3dddeab1">Fix commit — apache/httpd</a></li>
  <li><a href="https://access.redhat.com/security/cve/cve-2026-119618">Red Hat — CVE-2026-119618</a></li>
  <li><a href="https://www.netspi.com/blog/executive-blog/critical-vulnerability/apache-http-server-critical-rce-vulnerability-cve-2026-119618/">NetSPI analysis</a></li>
</ul>

<p><em>Verify every version against the vendor advisory before acting. Published 18 Nov 2026.</em></p>]]></content><author><name>Aniket Tomar</name></author><category term="cve" /><category term="rce" /><category term="apache" /><category term="http2" /><category term="blue-team" /><summary type="html"><![CDATA[A race condition in mod_http2 (CVSS 9.8) corrupts heap memory and can lead to remote code execution. Who's exposed, how to check, and the fix in 2.4.71.]]></summary><media:thumbnail xmlns:media="http://search.yahoo.com/mrss/" url="https://binary0101devil.pages.dev/assets/img/cve-2026-119618-apache-http2.svg" /><media:content medium="image" url="https://binary0101devil.pages.dev/assets/img/cve-2026-119618-apache-http2.svg" xmlns:media="http://search.yahoo.com/mrss/" /></entry><entry><title type="html">CVE-2026-98432: pre-auth use-after-free in the Linux kernel TCP stack</title><link href="https://binary0101devil.pages.dev/blog/2026/11/16/cve-2026-98432-linux-kernel-tcp-rce/" rel="alternate" type="text/html" title="CVE-2026-98432: pre-auth use-after-free in the Linux kernel TCP stack" /><published>2026-11-16T00:00:00+05:30</published><updated>2026-11-16T00:00:00+05:30</updated><id>https://binary0101devil.pages.dev/blog/2026/11/16/cve-2026-98432-linux-kernel-tcp-rce</id><content type="html" xml:base="https://binary0101devil.pages.dev/blog/2026/11/16/cve-2026-98432-linux-kernel-tcp-rce/"><![CDATA[<p>The third of three critical RCE flaws this week — and the broadest, because the bug lives in the core TCP path rather than an optional module. See also the actively-exploited <a href="/blog/2026/11/20/cve-2026-129448-apache-roller-rce/">Apache Roller RCE</a>, the <a href="/blog/2026/11/18/cve-2026-119618-apache-http2-rce/">Apache HTTP/2 race condition</a>, and the <a href="/blog/2026/11/14/top-rce-cves-2026/">2026 RCE field guide</a>.</p>

<blockquote>
  <p><strong>Scope and ethics.</strong> Exposure <strong>detection</strong> and remediation only, for systems you own or are authorized to assess.</p>
</blockquote>

<h2 id="description">Description</h2>

<p>A use-after-free in the IPv4 TCP stack. IP options stored with a pending connection’s request socket can be freed while another CPU still references them (<code class="language-plaintext highlighter-rouge">ip_options_build()</code>); the child socket takes ownership of those options <strong>without copying them</strong> in <code class="language-plaintext highlighter-rouge">tcp_v4_syn_recv_sock()</code>, and socket migration follows a similar path. IPv6 largely escapes the issue because the kernel copies IPv6 options. The defect dates back to kernel <strong>4.4</strong>.</p>

<h2 id="impact">Impact</h2>

<p>Remote, pre-authentication kernel memory corruption. The worst-case outcome is <strong>root-level code execution on the host</strong>. Because the vulnerable code is in the standard TCP handshake path, this is a broad exposure across servers, appliances and containers hosts — not a niche add-on.</p>

<h2 id="prerequisites-for-the-attack-to-work">Prerequisites (for the attack to work)</h2>

<ul>
  <li>Network reachability to a listening TCP service.</li>
  <li>A vulnerable kernel: anything from <strong>4.4</strong> up to the fixed stable releases listed below.</li>
  <li><strong>MPTCP caveat — read this carefully:</strong> the public research exploit used an MPTCP listener, <strong>but Red Hat states MPTCP is not required</strong>, and nodes with MPTCP disabled remain affected. Do not treat “MPTCP off” as “safe.”</li>
</ul>

<h2 id="steps-to-reproduce-exposure-validation">Steps to reproduce (exposure validation)</h2>

<div class="language-bash highlighter-rouge"><div class="highlight"><pre class="highlight"><code><span class="c"># 1) Running kernel — compare against your distro's fixed version AFTER a reboot</span>
<span class="nb">uname</span> <span class="nt">-r</span>

<span class="c"># 2) Is MPTCP enabled? (reduces, but does NOT eliminate, the surface)</span>
sysctl net.mptcp.enabled 2&gt;/dev/null

<span class="c"># 3) RHEL / derivatives: installed vs. running kernel</span>
rpm <span class="nt">-q</span> kernel<span class="p">;</span> <span class="nb">echo</span> <span class="s2">"running: </span><span class="si">$(</span><span class="nb">uname</span> <span class="nt">-r</span><span class="si">)</span><span class="s2">"</span>

<span class="c"># 4) Debian / Ubuntu: installed kernel images</span>
dpkg <span class="nt">-l</span> <span class="s1">'linux-image*'</span> | <span class="nb">awk</span> <span class="s1">'/^ii/{print $2, $3}'</span>
</code></pre></div></div>

<p>Compare the running kernel to your vendor’s fixed build. Exploit research is linked below for lab use against a kernel you control.</p>

<h2 id="remediation">Remediation</h2>

<ol>
  <li><strong>Patch the kernel and reboot.</strong> Upstream fixes include <strong>5.10.272, 5.15.223, 6.1.190, 6.6.159, 6.12.113, 6.18.56, 7.2.10 and 7.3</strong>; the kernel CVE team recommends moving to the latest stable rather than cherry-picking the patch. Distro fixes: Red Hat <strong>RHSA-2026:93412 / 93413 / 93414</strong> (RHEL 8/9/10), Ubuntu <strong>7.3.0-10.10</strong> on 26.10 (others in <code class="language-plaintext highlighter-rouge">-proposed</code>), and patched Amazon Linux streams.</li>
  <li>Mitigations until you can reboot — <strong>not</strong> a fix:
    <ul>
      <li><code class="language-plaintext highlighter-rouge">sysctl -w net.mptcp.enabled=0</code> (and <code class="language-plaintext highlighter-rouge">GODEBUG=multipathtcp=0</code> for Go 1.24+ services), then restart affected services.</li>
      <li>Drop inbound IPv4 packets carrying IP options (AWS publishes an iptables rule for this).</li>
      <li>Front the host with a <strong>separate</strong> load balancer or managed service that terminates TCP — a proxy on the <em>same</em> host does not count.</li>
    </ul>
  </li>
  <li>Watch for the reported scanning: researchers noted probing of <strong>port 443 over MPTCP</strong> that may be hunting for this bug.</li>
</ol>

<h2 id="references">References</h2>

<ul>
  <li><a href="https://access.redhat.com/security/cve/CVE-2026-98432">Red Hat — CVE-2026-98432</a> · <a href="https://access.redhat.com/security/vulnerabilities/RHSB-2026-016">RHSB-2026-016</a></li>
  <li><a href="https://ubuntu.com/security/CVE-2026-98432">Ubuntu — CVE-2026-98432</a></li>
  <li><a href="https://git.kernel.org/linus/e3f33b0a1d89d738e2b913dab519226eaf76e15b">Upstream fix commit — git.kernel.org</a></li>
  <li><a href="https://aws.amazon.com/security/security-bulletins/2026-160-aws/">AWS security bulletin</a></li>
  <li><a href="https://thehackernews.com/2026/10/linux-kernel-tcp-flaw-with-98-cvss.html">The Hacker News write-up</a></li>
</ul>

<p><em>Verify every version against the vendor advisory before acting. Published 16 Nov 2026.</em></p>]]></content><author><name>Aniket Tomar</name></author><category term="cve" /><category term="rce" /><category term="linux-kernel" /><category term="use-after-free" /><category term="blue-team" /><summary type="html"><![CDATA[A use-after-free in the IPv4 TCP path (CVSS 9.8) can give a remote attacker root. Affected kernels, the MPTCP myth, exposure checks and remediation.]]></summary><media:thumbnail xmlns:media="http://search.yahoo.com/mrss/" url="https://binary0101devil.pages.dev/assets/img/cve-2026-98432-linux-tcp.svg" /><media:content medium="image" url="https://binary0101devil.pages.dev/assets/img/cve-2026-98432-linux-tcp.svg" xmlns:media="http://search.yahoo.com/mrss/" /></entry><entry><title type="html">The 2026 RCE field guide: 10 shell-access CVEs, triaged and ranked</title><link href="https://binary0101devil.pages.dev/blog/2026/11/14/top-rce-cves-2026/" rel="alternate" type="text/html" title="The 2026 RCE field guide: 10 shell-access CVEs, triaged and ranked" /><published>2026-11-14T00:00:00+05:30</published><updated>2026-11-14T00:00:00+05:30</updated><id>https://binary0101devil.pages.dev/blog/2026/11/14/top-rce-cves-2026</id><content type="html" xml:base="https://binary0101devil.pages.dev/blog/2026/11/14/top-rce-cves-2026/"><![CDATA[<p>CVSS tells you how bad a bug <em>could</em> be. It doesn’t tell you what to patch first on a Monday morning. This guide ranks ten 2026 remote-code-execution flaws on three questions I actually triage by:</p>

<ol>
  <li><strong>Is it being exploited?</strong> A CISA KEV listing or a vendor confirmation beats any score.</li>
  <li><strong>Does the attacker need credentials or a click?</strong> Pre-auth and zero-click go to the top.</li>
  <li><strong>How much of a typical estate runs it?</strong> A bug in PHP or Node.js reaches more boxes than one in a niche appliance module.</li>
</ol>

<p>Each entry links to a full write-up covering root cause, exposure checks, detection and remediation.</p>

<blockquote>
  <p><strong>Scope and ethics.</strong> These are defensive summaries for prioritizing patching and hunting on systems you own or are authorized to assess. No exploit code here or in the linked posts.</p>
</blockquote>

<p><em>Updated 20 Nov 2026: entries now link to the full write-ups published since this guide went up.</em></p>

<h2 id="the-ranking">The ranking</h2>

<table>
  <thead>
    <tr>
      <th>#</th>
      <th>Vulnerability</th>
      <th>Score</th>
      <th>Pre-auth?</th>
      <th>Exploited?</th>
      <th>Fix</th>
    </tr>
  </thead>
  <tbody>
    <tr>
      <td>1</td>
      <td><a href="/blog/2026/11/20/php2shell-cve-2026-109618-109619/">PHP2Shell, CVE-2026-109618 / 109619</a></td>
      <td>9.2 (v4.0)</td>
      <td>Yes</td>
      <td><strong>KEV</strong></td>
      <td>PHP 8.5.12 / 8.4.27 / 8.3.36 / 8.2.35</td>
    </tr>
    <tr>
      <td>2</td>
      <td><a href="/blog/2026/11/20/node2shell-cve-2026-111497-brotli-rce/">Node2Shell, CVE-2026-111497</a></td>
      <td>9.2 (v4.0)</td>
      <td>Yes</td>
      <td><strong>KEV</strong></td>
      <td>Node 22.23.4 / 24.22.1 / 26.12.1; Brotli 1.2.1</td>
    </tr>
    <tr>
      <td>3</td>
      <td><a href="/blog/2026/11/04/cve-2026-94127-f5-big-ip-apm/">F5 BIG-IP APM, CVE-2026-94127</a></td>
      <td>9.8</td>
      <td>Yes</td>
      <td><strong>KEV</strong></td>
      <td>F5 engineering hotfixes for 21.1 / 17.5 / 17.1</td>
    </tr>
    <tr>
      <td>4</td>
      <td><a href="/blog/2026/11/03/citrix-netscaler-cve-2026-88771-88772-88779/">Citrix NetScaler, CVE-2026-88771 / 88772</a></td>
      <td>9.5 (v4.0)</td>
      <td>Yes</td>
      <td><strong>KEV</strong></td>
      <td>14.1-73.41 / 13.1-64.28 (also covers 88779)</td>
    </tr>
    <tr>
      <td>5</td>
      <td><a href="/blog/2026/11/08/ivanti-connect-secure-cve-2026-111357-118527/">Ivanti Connect Secure chain, CVE-2026-111357 / 118527</a></td>
      <td>8.8</td>
      <td>Yes, chained</td>
      <td><strong>KEV</strong></td>
      <td>25.1.1.3 / 22.8R2.6</td>
    </tr>
    <tr>
      <td>6</td>
      <td><a href="/blog/2026/11/20/cve-2026-129448-apache-roller-rce/">Apache Roller, CVE-2026-129448</a></td>
      <td>9.8</td>
      <td>Yes</td>
      <td>Yes (ASF)</td>
      <td>Roller 6.1.6</td>
    </tr>
    <tr>
      <td>7</td>
      <td><a href="/blog/2026/11/06/plex2shell-plex-media-server-cve-2026-96652/">Plex2Shell, CVE-2026-96652 + 3</a></td>
      <td>7.2 max per CVE</td>
      <td>Yes, chained</td>
      <td><strong>KEV</strong></td>
      <td>Plex Media Server 1.43.3.10861</td>
    </tr>
    <tr>
      <td>8</td>
      <td><a href="/blog/2026/11/16/cve-2026-98432-linux-kernel-tcp-rce/">Linux kernel TCP, CVE-2026-98432</a></td>
      <td>9.8</td>
      <td>Yes</td>
      <td>See post</td>
      <td>Distro kernel updates + reboot</td>
    </tr>
    <tr>
      <td>9</td>
      <td><a href="/blog/2026/11/18/cve-2026-119618-apache-http2-rce/">Apache HTTP/2, CVE-2026-119618</a></td>
      <td>9.8</td>
      <td>Yes</td>
      <td>Not confirmed</td>
      <td>httpd 2.4.71</td>
    </tr>
    <tr>
      <td>10</td>
      <td><a href="/blog/2026/11/20/java2shell-cve-2026-114938-jvm-base64/">Java2Shell, CVE-2026-114938</a></td>
      <td>8.1 (v3.1)</td>
      <td>Yes</td>
      <td>Not reported</td>
      <td>Nov 2026 JDK updates</td>
    </tr>
  </tbody>
</table>

<h2 id="why-this-order">Why this order</h2>

<p><strong>1–2: the runtime bugs.</strong> PHP2Shell and Node2Shell live in the <em>language runtime</em>, not an app or plugin. Every PHP site and every Node service that decodes Brotli inherits them, whatever framework sits on top. Both are on KEV, and PHP2Shell was seen in a honeypot four days before disclosure. If you only do two things this week, patch these two and <strong>restart the workers</strong>.</p>

<p><strong>3–5: the edge appliances.</strong> F5 APM, NetScaler and Ivanti Connect Secure sit on the network edge by design, run as root, and are favorite initial-access targets for state-linked and ransomware crews. Each was exploited as a zero-day or within days. Patching isn’t enough: <strong>hunt for persistence first</strong>, because these appliances have a history of implants that survive upgrades. Ivanti’s own Integrity Checker missed the in-memory activity in this chain.</p>

<p><strong>6–7: exploited, but narrower footprints.</strong> Apache Roller and Plex Media Server are less common in enterprise estates, but both were exploited. Plex is a reminder that “home” software ends up on corporate laptops and lab servers. No single Plex CVE scores above 7.2. Chained, they hand over the box, which is why per-CVE scores can mislead.</p>

<p><strong>8–10: severe on paper, gated in practice.</strong> The kernel TCP bug and Apache HTTP/2 race are pre-auth 9.8s with huge potential reach, but reliable exploitation is hard, so they rank by exposure rather than evidence of attacks. Java2Shell needs an AVX-512 x86 CPU, a JIT-hot decoder and an application-specific chain to turn a 62-byte overwrite into a shell. Patch all three on your normal fast track. Don’t let them jump ahead of anything on KEV.</p>

<h2 id="also-worth-your-attention">Also worth your attention</h2>

<p>These didn’t make the top ten because they need credentials, a user action, or aren’t strictly remote code execution. Several are still on KEV and urgent:</p>

<ul>
  <li><a href="/blog/2026/11/01/cve-2026-65660-sharepoint-code-injection/">SharePoint CVE-2026-65660</a>: authenticated code injection, exploited, KEV.</li>
  <li><a href="/blog/2026/11/02/cve-2026-0322-pan-os-code-injection/">PAN-OS CVE-2026-0322</a>: management-plane code injection to root. Needs an admin login, but exploited.</li>
  <li><a href="/blog/2026/11/05/cve-2026-84869-screenconnect/">ScreenConnect CVE-2026-84869</a>: clients run files pushed by guests. Aimed at MSP technicians.</li>
  <li><a href="/blog/2026/11/07/cve-2026-48356-adobe-commerce-magento-upload/">Adobe Commerce “PolyShell” CVE-2026-48356</a>: unrestricted upload, mass-exploited against stores.</li>
  <li><a href="/blog/2026/11/09/cve-2026-119977-wordpress-core-toctou/">WordPress CVE-2026-119977</a>: an upgrade-window race. Read the post before you click “Update”.</li>
  <li><a href="/blog/2026/11/10/cve-2026-123158-openssh/">OpenSSH CVE-2026-123158</a>: password bypass on musl, no-PAM builds.</li>
  <li><a href="/blog/2026/11/11/cve-2026-125350-chrome-use-after-free/">Chrome CVE-2026-125350</a>: browser use-after-free zero-day.</li>
  <li><a href="/blog/2026/11/19/cve-2026-40217-litellm/">LiteLLM CVE-2026-40217</a>: AI-gateway sandbox escape with API keys at stake.</li>
</ul>

<h2 id="a-triage-routine-that-works">A triage routine that works</h2>

<div class="language-bash highlighter-rouge"><div class="highlight"><pre class="highlight"><code><span class="c"># 1) Pull the current CISA KEV feed and list everything added in the last 30 days</span>
curl <span class="nt">-s</span> https://www.cisa.gov/sites/default/files/feeds/known_exploited_vulnerabilities.json <span class="se">\</span>
 | python3 <span class="nt">-c</span> <span class="s1">'import json,sys,datetime as d
c=(d.date.today()-d.timedelta(days=30)).isoformat()
for v in json.load(sys.stdin)["vulnerabilities"]:
  if v["dateAdded"]&gt;=c: print(v["dateAdded"],v["cveID"],v["vendorProject"],v["product"])'</span> | <span class="nb">sort</span> <span class="nt">-r</span>

<span class="c"># 2) Cross-check against what you actually run (examples)</span>
php <span class="nt">-v</span> 2&gt;/dev/null | <span class="nb">head</span> <span class="nt">-1</span><span class="p">;</span> node <span class="nt">-v</span> 2&gt;/dev/null<span class="p">;</span> java <span class="nt">-version</span> 2&gt;&amp;1 | <span class="nb">head</span> <span class="nt">-1</span>
httpd <span class="nt">-v</span> 2&gt;/dev/null <span class="o">||</span> apache2 <span class="nt">-v</span> 2&gt;/dev/null<span class="p">;</span> <span class="nb">uname</span> <span class="nt">-r</span><span class="p">;</span> ssh <span class="nt">-V</span>
</code></pre></div></div>

<p>Then for every match:</p>

<ol>
  <li><strong>Patch and restart.</strong> Runtimes, JVMs, PHP workers and kernels keep the old code in memory until restarted or rebooted.</li>
  <li><strong>Hunt before declaring clean.</strong> For anything exploited pre-disclosure, look at process lineage from the vulnerable service, new admin accounts, web shells and outbound connections back to the date the vendor says exploitation began.</li>
  <li><strong>Shrink exposure.</strong> Management interfaces, VPN admin portals and media servers shouldn’t face the internet.</li>
  <li><strong>Verify from the outside.</strong> Re-scan after the change. A package on disk isn’t a patched process.</li>
</ol>

<h2 id="references">References</h2>

<ul>
  <li><a href="https://www.cisa.gov/known-exploited-vulnerabilities-catalog">CISA Known Exploited Vulnerabilities catalog</a></li>
  <li><a href="https://nvd.nist.gov/">NVD</a>, for CVSS vectors and affected configurations</li>
  <li><a href="https://www.first.org/cvss/v4.0/specification-document">FIRST CVSS v4.0 specification</a>, on why v3.1 and v4.0 scores differ</li>
  <li>Vendor advisories are linked from each individual write-up.</li>
</ul>

<p><em>Verify every version against the vendor advisory before acting. Published 14 Nov 2026.</em></p>]]></content><author><name>Aniket Tomar</name></author><category term="cve" /><category term="rce" /><category term="blue-team" /><category term="vulnerability-management" /><summary type="html"><![CDATA[Ten remote-code-execution flaws from 2026, ranked by what actually matters for defenders: exploitation status, authentication, and how much of your estate is exposed.]]></summary><media:thumbnail xmlns:media="http://search.yahoo.com/mrss/" url="https://binary0101devil.pages.dev/assets/img/rce-cve-roundup-2026.svg" /><media:content medium="image" url="https://binary0101devil.pages.dev/assets/img/rce-cve-roundup-2026.svg" xmlns:media="http://search.yahoo.com/mrss/" /></entry><entry><title type="html">CVE-2026-125350: Chrome XML use-after-free zero-day, exploited in the wild</title><link href="https://binary0101devil.pages.dev/blog/2026/11/11/cve-2026-125350-chrome-use-after-free/" rel="alternate" type="text/html" title="CVE-2026-125350: Chrome XML use-after-free zero-day, exploited in the wild" /><published>2026-11-11T00:00:00+05:30</published><updated>2026-11-11T00:00:00+05:30</updated><id>https://binary0101devil.pages.dev/blog/2026/11/11/cve-2026-125350-chrome-use-after-free</id><content type="html" xml:base="https://binary0101devil.pages.dev/blog/2026/11/11/cve-2026-125350-chrome-use-after-free/"><![CDATA[<p>Another Chrome zero-day, and another one where the public technical detail amounts to a single line. Google shipped the fix on 9 November with the familiar note that it “is aware that an exploit for CVE-2026-125350 exists in the wild”, and CISA added it to KEV the next day. This post is mostly about fleet work: knowing which browsers you have, making sure they actually restarted onto the fixed build, and covering the other Chromium-based browsers. I won’t speculate on the exploit chain, because nothing reliable about it has been published.</p>

<blockquote>
  <p><strong>Scope and ethics.</strong> This post covers what’s publicly known, version inventory, detection and remediation for endpoints you manage. There’s no PoC, no crafted page and no speculation about exploit internals. Lab validation means checking version reporting and update policy on test machines, not loading malicious content.</p>
</blockquote>

<h2 id="at-a-glance">At a glance</h2>

<table>
  <thead>
    <tr>
      <th> </th>
      <th> </th>
    </tr>
  </thead>
  <tbody>
    <tr>
      <td><strong>CVE</strong></td>
      <td>CVE-2026-125350, “Use after free in XML”</td>
    </tr>
    <tr>
      <td><strong>Severity</strong></td>
      <td>CVSS 3.1 <strong>8.8 High</strong>, <code class="language-plaintext highlighter-rouge">AV:N/AC:L/PR:N/UI:R/S:U/C:H/I:H/A:H</code> (CISA-ADP, via NVD). Chromium severity: High</td>
    </tr>
    <tr>
      <td><strong>CWE</strong></td>
      <td>CWE-416 Use After Free</td>
    </tr>
    <tr>
      <td><strong>Fixed in</strong></td>
      <td>Chrome desktop <strong>157.0.8103.50/.51</strong> (Windows, Mac) and <strong>157.0.8103.50</strong> (Linux), 9 Nov 2026. Extended Stable <strong>156.0.8078.96</strong> (Windows, Mac)</td>
    </tr>
    <tr>
      <td><strong>Edge</strong></td>
      <td>Stable <strong>157.0.4341.46</strong> (9 Nov), Extended Stable <strong>156.0.4314.80</strong> (10 Nov), Android <strong>157.0.4341.53</strong> (11 Nov), per Microsoft</td>
    </tr>
    <tr>
      <td><strong>Exploited?</strong></td>
      <td><strong>Yes.</strong> Google confirms an in-the-wild exploit. CISA KEV, added 10 Nov 2026 (federal due date 24 Nov)</td>
    </tr>
    <tr>
      <td><strong>Auth / interaction</strong></td>
      <td>None / the victim has to load attacker-controlled content</td>
    </tr>
  </tbody>
</table>

<h2 id="description-whats-actually-known">Description: what’s actually known</h2>

<p>Chrome is Google’s browser, built on the open-source Chromium project, which also underpins Microsoft Edge, Opera, Brave, Vivaldi and many embedded browsers. CISA’s KEV entry explicitly notes that the flaw “could affect multiple web browsers that utilize Chromium”.</p>

<p>The public description, from NVD and Google’s release post, is:</p>

<blockquote>
  <p>Use after free in XML in Google Chrome prior to 157.0.8103.50 allowed a remote attacker to execute arbitrary code inside the sandbox via a crafted HTML page.</p>
</blockquote>

<p>What that tells us:</p>

<ul>
  <li><strong>Bug class:</strong> use-after-free. Code keeps a pointer to an object after the memory has been freed. If an attacker can get new, controlled data allocated into that slot, the stale pointer then operates on the attacker’s data. In a browser this is a well-worn path from memory corruption to code execution.</li>
  <li><strong>Component:</strong> Google labels it “XML”, meaning the code that handles XML content in the rendering pipeline. Google hasn’t said which code path, and I’m not going to guess.</li>
  <li><strong>Where code runs:</strong> “inside the sandbox”, i.e. in a sandboxed renderer process. On its own, that gives an attacker code execution in a heavily restricted process. Full device compromise normally needs a second, separate bug to escape the sandbox. Google hasn’t said whether this exploit was paired with one.</li>
  <li><strong>Trigger:</strong> a crafted HTML page. Visiting a malicious or compromised site, or rendering attacker content in a malicious ad or embedded frame, is enough. No other interaction needed.</li>
</ul>

<p><strong>What isn’t public:</strong> the vulnerable code path, the exploit, the attacker, the targets and any IOCs. The Chromium bug entries are restricted. Google’s standard note says bug details may stay restricted “until a majority of users are updated with a fix”, and also while a bug in a third-party library that other projects depend on remains unfixed. That second sentence is boilerplate in every Chrome release post, so don’t read it as confirmation that a third-party library is involved. The release post credits Arni Hardarson (Neonix Security), who reported on 12 October, and an anonymous reporter on 6 November.</p>

<p>This is the <strong>eighth</strong> exploited Chrome zero-day Google has patched in 2026, according to BleepingComputer’s count.</p>

<h2 id="impact">Impact</h2>

<ul>
  <li>Arbitrary code execution in the renderer sandbox from simply loading a page.</li>
  <li>Used in real attacks before the patch was available.</li>
  <li>Highest risk: users who browse untrusted content on privileged machines (admins, developers, finance), and any service that renders untrusted HTML with headless Chromium, such as PDF generators, link preview or scraping workers, and test harnesses. Those often run older bundled builds with weaker sandboxing.</li>
</ul>

<h2 id="which-browsers-are-exposed">Which browsers are exposed</h2>

<table>
  <thead>
    <tr>
      <th>Browser / channel</th>
      <th>Fixed build</th>
      <th>Source</th>
    </tr>
  </thead>
  <tbody>
    <tr>
      <td>Chrome Stable, Windows / Mac</td>
      <td>157.0.8103.50 / .51 or later</td>
      <td>Chrome Releases, 9 Nov</td>
    </tr>
    <tr>
      <td>Chrome Stable, Linux</td>
      <td>157.0.8103.50 or later</td>
      <td>Chrome Releases, 9 Nov</td>
    </tr>
    <tr>
      <td>Chrome Extended Stable, Windows / Mac</td>
      <td>156.0.8078.96 or later</td>
      <td>Chrome Releases, 9 Nov</td>
    </tr>
    <tr>
      <td>ChromeOS Stable</td>
      <td>16836.33.0 (browser 156.0.8078.102) lists the fix</td>
      <td>Chrome Releases, 13 Nov</td>
    </tr>
    <tr>
      <td>Edge Stable</td>
      <td>157.0.4341.46 or later</td>
      <td>Microsoft Edge security release notes</td>
    </tr>
    <tr>
      <td>Edge Extended Stable</td>
      <td>156.0.4314.80 or later</td>
      <td>Microsoft Edge security release notes</td>
    </tr>
    <tr>
      <td>Edge for Android</td>
      <td>157.0.4341.53 or later</td>
      <td>Microsoft Edge security release notes</td>
    </tr>
    <tr>
      <td>Chrome for Android / iOS, Opera, Brave, Vivaldi, Electron/CEF apps, distro <code class="language-plaintext highlighter-rouge">chromium</code> packages</td>
      <td><strong>Not confirmed here.</strong> Check each vendor</td>
      <td>—</td>
    </tr>
  </tbody>
</table>

<p>Chrome 158 has since been promoted to Stable (17 Nov), so the practical target is “current Stable”, and anything at or above those numbers in the same branch is fine. The Chrome for Android 157.0.8103.50 release post on 9 November only mentions “stability and performance improvements”, so I couldn’t confirm the Android fixed build from Google’s posts.</p>

<h2 id="identifying-affected-systems">Identifying affected systems</h2>

<p>The number that matters is the version <strong>running</strong> in the browser process. Chrome downloads updates in the background, but they only take effect after a relaunch, so many “updated” machines are still running the old code.</p>

<div class="language-bash highlighter-rouge"><div class="highlight"><pre class="highlight"><code><span class="c"># Linux</span>
google-chrome <span class="nt">--version</span><span class="p">;</span> chromium <span class="nt">--version</span> 2&gt;/dev/null
dpkg <span class="nt">-l</span> google-chrome-stable 2&gt;/dev/null | <span class="nb">tail</span> <span class="nt">-n1</span> <span class="o">||</span> rpm <span class="nt">-q</span> google-chrome-stable
<span class="c"># was the browser restarted after the update? compare start time with package install time</span>
ps <span class="nt">-eo</span> pid,lstart,cmd | <span class="nb">grep</span> <span class="nt">-E</span> <span class="s1">'[c]hrome'</span> | <span class="nb">grep</span> <span class="nt">-v</span> <span class="nt">--</span> <span class="s1">'--type='</span> | <span class="nb">head</span> <span class="nt">-n3</span>

<span class="c"># macOS</span>
defaults <span class="nb">read</span> <span class="s2">"/Applications/Google Chrome.app/Contents/Info.plist"</span> CFBundleShortVersionString
defaults <span class="nb">read</span> <span class="s2">"/Applications/Microsoft Edge.app/Contents/Info.plist"</span> CFBundleShortVersionString
</code></pre></div></div>

<div class="language-powershell highlighter-rouge"><div class="highlight"><pre class="highlight"><code><span class="c"># Windows</span><span class="w">
</span><span class="p">(</span><span class="n">Get-Item</span><span class="w"> </span><span class="s2">"</span><span class="nv">$</span><span class="nn">env</span><span class="p">:</span><span class="nv">ProgramFiles</span><span class="s2">\Google\Chrome\Application\chrome.exe"</span><span class="p">)</span><span class="o">.</span><span class="nf">VersionInfo</span><span class="o">.</span><span class="nf">ProductVersion</span><span class="w">
</span><span class="p">(</span><span class="n">Get-Item</span><span class="w"> </span><span class="s2">"</span><span class="nv">${env:ProgramFiles(x86)}</span><span class="s2">\Microsoft\Edge\Application\msedge.exe"</span><span class="p">)</span><span class="o">.</span><span class="nf">VersionInfo</span><span class="o">.</span><span class="nf">ProductVersion</span><span class="w">
</span><span class="c"># per-user installs live under %LOCALAPPDATA%\Google\Chrome\Application</span><span class="w">
</span></code></pre></div></div>

<p>For a fleet, use what you already have:</p>

<ul>
  <li><strong>Chrome Enterprise Core (Browser Cloud Management)</strong> and the Edge management service both report browser versions per device. Filter for anything below the fixed build.</li>
  <li><strong>EDR, MDM or software inventory</strong> (Intune, Jamf, SCCM, osquery). In osquery, the <code class="language-plaintext highlighter-rouge">programs</code> (Windows) and <code class="language-plaintext highlighter-rouge">apps</code> (macOS) tables carry installed versions.</li>
  <li><strong>Headless and embedded Chromium:</strong> search build manifests and images for <code class="language-plaintext highlighter-rouge">puppeteer</code>, <code class="language-plaintext highlighter-rouge">playwright</code>, <code class="language-plaintext highlighter-rouge">chromium</code>, <code class="language-plaintext highlighter-rouge">electron</code> and <code class="language-plaintext highlighter-rouge">cef</code>, and check what Chromium version each ships.</li>
  <li>Users can confirm at <code class="language-plaintext highlighter-rouge">chrome://version</code> / <code class="language-plaintext highlighter-rouge">edge://version</code>.</li>
</ul>

<h2 id="safe-validation-in-an-isolated-lab">Safe validation in an isolated lab</h2>

<p>Nothing here needs a malicious page:</p>

<ol>
  <li>On a test VM, install a pre-fix build (156.x Stable or 157 before .50) and the current build side by side, or on two VMs.</li>
  <li>Confirm your inventory or scanner flags only the old one, and that it reads the <strong>running</strong> version, not just the file on disk. Update the old one without relaunching and see what each tool reports.</li>
  <li>Apply your relaunch policy (below) to the test OU and confirm the browser actually restarts within the deadline.</li>
  <li>If you have headless Chromium in CI or services, run the version check inside those containers and confirm they’re part of your patch pipeline.</li>
</ol>

<h2 id="detection-and-threat-hunting">Detection and threat hunting</h2>

<p>No IOCs, payload hashes or attacker infrastructure have been published for this CVE, so hunting is generic:</p>

<ul>
  <li><strong>Renderer crashes.</strong> Clusters of Chrome or Edge renderer crashes on the same domain across several users, especially before the patch landed. Check crash reporting or <code class="language-plaintext highlighter-rouge">chrome://crashes</code> on suspected machines.</li>
  <li><strong>Unexpected child processes of the browser.</strong> <code class="language-plaintext highlighter-rouge">chrome.exe</code> / <code class="language-plaintext highlighter-rouge">msedge.exe</code> / <code class="language-plaintext highlighter-rouge">chrome</code> spawning <code class="language-plaintext highlighter-rouge">cmd.exe</code>, <code class="language-plaintext highlighter-rouge">powershell.exe</code>, <code class="language-plaintext highlighter-rouge">rundll32.exe</code>, <code class="language-plaintext highlighter-rouge">mshta.exe</code>, <code class="language-plaintext highlighter-rouge">sh</code>, <code class="language-plaintext highlighter-rouge">bash</code>, <code class="language-plaintext highlighter-rouge">curl</code> or <code class="language-plaintext highlighter-rouge">osascript</code> is abnormal and would suggest a successful sandbox escape on top of the renderer bug. Most EDRs have a built-in rule for this; make sure it’s enabled and alerting.</li>
  <li><strong>Proxy and DNS logs.</strong> Look for newly registered or rare domains visited just before any such alert, and for unusual frames or ad-delivery chains.</li>
  <li><strong>Unpatched-window exposure.</strong> Prioritize investigating high-value users whose browsers stayed below the fixed build after 9 November.</li>
</ul>

<h2 id="remediation">Remediation</h2>

<ol>
  <li><strong>Update and relaunch.</strong> Chrome to 157.0.8103.50/.51 or later (now 158 Stable), Extended Stable to 156.0.8078.96 or later, Edge to the builds above. The relaunch is the step people forget.</li>
  <li><strong>Enforce relaunch with policy.</strong> Chrome’s <code class="language-plaintext highlighter-rouge">RelaunchNotification</code> (set to required) and <code class="language-plaintext highlighter-rouge">RelaunchNotificationPeriod</code>, or Edge’s equivalent policies, put a deadline on it.</li>
  <li><strong>Check for update blockers.</strong> Remove or raise any <code class="language-plaintext highlighter-rouge">TargetVersionPrefix</code> pin, and make sure no GPO has disabled Google Update / Edge Update.</li>
  <li><strong>Other Chromium browsers.</strong> Check Opera, Brave, Vivaldi and distro <code class="language-plaintext highlighter-rouge">chromium</code> packages against their vendors’ notes.</li>
  <li><strong>Embedded Chromium.</strong> Rebuild services and CI images that bundle Chromium (Puppeteer, Playwright, Electron, CEF) once their upstream ships the fix, and don’t render untrusted HTML in them in the meantime unless the process is isolated.</li>
</ol>

<h3 id="interim-measures">Interim measures</h3>

<p>There’s no configuration workaround from Google. Until devices relaunch, reduce exposure where you can: block uncategorized and newly registered domains at the proxy, keep ad blocking enabled in managed browser policy if that fits your environment, and don’t let admins browse the web from privileged workstations.</p>

<h2 id="references">References</h2>

<ul>
  <li><a href="https://nvd.nist.gov/vuln/detail/CVE-2026-125350">CVE-2026-125350, NVD</a> · <a href="https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-125350">CISA KEV entry</a></li>
  <li><a href="https://chromereleases.googleblog.com/2026/11/stable-channel-update-for-desktop_018223192.html">Chrome Releases: Stable Channel Update for Desktop, 9 Nov 2026</a></li>
  <li><a href="https://chromereleases.googleblog.com/">Chrome Releases blog (Extended Stable and ChromeOS posts)</a></li>
  <li><a href="https://learn.microsoft.com/en-us/deployedge/microsoft-edge-relnotes-security">Microsoft Edge security release notes</a></li>
  <li><a href="https://www.bleepingcomputer.com/news/security/google-patches-eighth-chrome-zero-day-exploited-in-attacks-this-year/">BleepingComputer: eighth exploited Chrome zero-day of 2026</a></li>
</ul>

<p><em>Verify every version against the vendor advisory before acting. Published 11 Nov 2026.</em></p>]]></content><author><name>Aniket Tomar</name></author><category term="cve" /><category term="chrome" /><category term="use-after-free" /><category term="memory-corruption" /><category term="blue-team" /><summary type="html"><![CDATA[Google's eighth exploited Chrome zero-day of 2026 is a use-after-free in the XML component (CVSS 8.8, on CISA KEV). Fixed builds for Chrome and Edge, fleet checks, hunting and policy.]]></summary><media:thumbnail xmlns:media="http://search.yahoo.com/mrss/" url="https://binary0101devil.pages.dev/assets/img/cve-2026-125350-chrome-use-after-free.svg" /><media:content medium="image" url="https://binary0101devil.pages.dev/assets/img/cve-2026-125350-chrome-use-after-free.svg" xmlns:media="http://search.yahoo.com/mrss/" /></entry><entry><title type="html">CVE-2026-123158 (SSHole): OpenSSH password bypass on musl, no-PAM Linux</title><link href="https://binary0101devil.pages.dev/blog/2026/11/10/cve-2026-123158-openssh/" rel="alternate" type="text/html" title="CVE-2026-123158 (SSHole): OpenSSH password bypass on musl, no-PAM Linux" /><published>2026-11-10T00:00:00+05:30</published><updated>2026-11-10T00:00:00+05:30</updated><id>https://binary0101devil.pages.dev/blog/2026/11/10/cve-2026-123158-openssh</id><content type="html" xml:base="https://binary0101devil.pages.dev/blog/2026/11/10/cve-2026-123158-openssh/"><![CDATA[<p>This one is narrow, but where it applies it’s about as bad as SSH bugs get: a password login that succeeds without the password. It only hits Linux systems where <code class="language-plaintext highlighter-rouge">sshd</code> is built against <strong>musl libc with PAM disabled</strong>, which in practice means <strong>Alpine Linux</strong> hosts, Alpine-based containers that run <code class="language-plaintext highlighter-rouge">sshd</code>, and anything else built the same way. Glibc distros with PAM (most Debian, Ubuntu and RHEL installs) are not in the described scope. It was fixed on 5 November, exploited within two days, and added to CISA’s KEV catalog on 9 November.</p>

<blockquote>
  <p><strong>Scope and ethics.</strong> This post covers root cause, exposure checks, detection and remediation for systems you own or are authorized to assess. It contains no exploit code or trigger inputs. Lab validation here means confirming package versions and effective <code class="language-plaintext highlighter-rouge">sshd</code> configuration on a VM you control, not attempting logins against anyone’s server.</p>
</blockquote>

<h2 id="at-a-glance">At a glance</h2>

<table>
  <thead>
    <tr>
      <th> </th>
      <th> </th>
    </tr>
  </thead>
  <tbody>
    <tr>
      <td><strong>CVE</strong></td>
      <td>CVE-2026-123158, nicknamed “SSHole” by researchers and GreyNoise</td>
    </tr>
    <tr>
      <td><strong>Severity</strong></td>
      <td>CVSS 3.1 <strong>8.1 High</strong>, <code class="language-plaintext highlighter-rouge">AV:N/AC:H/PR:N/UI:N/S:U/C:H/I:H/A:H</code> (MITRE CNA, via NVD)</td>
    </tr>
    <tr>
      <td><strong>CWE</strong></td>
      <td>CWE-394 Unexpected Status Code or Return Value (NVD, CISA)</td>
    </tr>
    <tr>
      <td><strong>Affected</strong></td>
      <td><code class="language-plaintext highlighter-rouge">sshd</code> in OpenSSH <strong>before 10.6p2</strong> on Linux, <strong>built with musl libc and without PAM</strong> (NVD)</td>
    </tr>
    <tr>
      <td><strong>Fixed in</strong></td>
      <td><strong>OpenSSH 10.6p2</strong> (5 Nov 2026). Alpine backported the fix to every branch from 3.16 to edge (table below)</td>
    </tr>
    <tr>
      <td><strong>Exploited?</strong></td>
      <td><strong>Yes.</strong> CISA KEV, added 9 Nov 2026 (federal due date 12 Nov), with forensic triage required. GreyNoise saw the first successful exploitation on 7 Nov</td>
    </tr>
    <tr>
      <td><strong>Auth / interaction</strong></td>
      <td>None / none, but only for specific account configurations</td>
    </tr>
  </tbody>
</table>

<h2 id="description-how-the-bug-works">Description: how the bug works</h2>

<p>OpenSSH’s own release note is short: 10.6p2 adds a workaround for “a password authentication bypass that is exploitable through sshd(8) that affects some Linux systems when built and run with PAM support disabled.” The maintainers say the root cause is <strong>outside sshd</strong>, but that sshd is where it can be blocked.</p>

<p>Alpine’s advisory explains the mechanism:</p>

<ol>
  <li><strong>Without PAM, sshd checks passwords itself.</strong> It hashes the attempt with <code class="language-plaintext highlighter-rouge">crypt()</code> using the salt from the account’s shadow entry and compares the output to the stored field.</li>
  <li><strong>Some accounts have no real hash.</strong> Their shadow field is just <code class="language-plaintext highlighter-rouge">*</code>: no password set. That’s the default for <code class="language-plaintext highlighter-rouge">root</code> on recent Alpine base images, key-only users created without a password, the <code class="language-plaintext highlighter-rouge">alpine</code> user in tiny-cloud images, and directory-service users with no shadow entry. Because <code class="language-plaintext highlighter-rouge">*</code> is too short to hold a salt, sshd borrows the salt from another account on the system.</li>
  <li><strong>musl’s <code class="language-plaintext highlighter-rouge">crypt()</code> signals errors with <code class="language-plaintext highlighter-rouge">*</code>.</strong> When the supplied key is longer than 256 bytes, musl’s <code class="language-plaintext highlighter-rouge">crypt()</code> returns the string <code class="language-plaintext highlighter-rouge">*</code> to mean “error”.</li>
  <li><strong>The error string matches the stored value.</strong> sshd compares <code class="language-plaintext highlighter-rouge">*</code> (error) with <code class="language-plaintext highlighter-rouge">*</code> (no password), they’re equal, and the login succeeds.</li>
</ol>

<p>So the trigger condition is an account whose password field is <code class="language-plaintext highlighter-rouge">*</code>, <strong>plus</strong> at least one other local account with a <code class="language-plaintext highlighter-rouge">$5$</code> or <code class="language-plaintext highlighter-rouge">$6$</code> hash to provide a salt, plus a long password attempt. NVD describes the length as 256 to 1024 characters and Alpine as 257 to 1024 bytes. Alpine notes that the attack takes a single connection and needs no retries. That’s why “high complexity” in the CVSS vector understates the risk once your configuration matches.</p>

<p>This is a textbook CWE-394: a caller treated an error return value as a legitimate result.</p>

<h2 id="impact">Impact</h2>

<ul>
  <li><strong>Unauthenticated interactive login</strong> as any matching account, over the network.</li>
  <li>On images that give passwordless <code class="language-plaintext highlighter-rouge">doas</code> to the login user (Alpine’s tiny-cloud images ship an <code class="language-plaintext highlighter-rouge">alpine</code> user with <code class="language-plaintext highlighter-rouge">*</code> and passwordless <code class="language-plaintext highlighter-rouge">doas</code>), that becomes <strong>root</strong>.</li>
  <li><code class="language-plaintext highlighter-rouge">root</code> itself has <code class="language-plaintext highlighter-rouge">*</code> on recent base images, but Alpine notes it’s protected by the default <code class="language-plaintext highlighter-rouge">PermitRootLogin prohibit-password</code> <strong>unless you’ve overridden it</strong>.</li>
  <li>One account with a real password hash is enough to expose <strong>every</strong> <code class="language-plaintext highlighter-rouge">*</code> account on the host.</li>
</ul>

<h2 id="who-is-exposed">Who is exposed</h2>

<table>
  <thead>
    <tr>
      <th>Configuration</th>
      <th>Status (per Alpine)</th>
    </tr>
  </thead>
  <tbody>
    <tr>
      <td>Alpine <code class="language-plaintext highlighter-rouge">openssh-server</code> (built without PAM), <code class="language-plaintext highlighter-rouge">PasswordAuthentication yes</code> (default)</td>
      <td><strong>Affected</strong></td>
    </tr>
    <tr>
      <td>Alpine <code class="language-plaintext highlighter-rouge">openssh-server-pam</code> with <code class="language-plaintext highlighter-rouge">UsePAM no</code></td>
      <td><strong>Affected</strong></td>
    </tr>
    <tr>
      <td><code class="language-plaintext highlighter-rouge">openssh-server-pam</code> with <code class="language-plaintext highlighter-rouge">UsePAM yes</code></td>
      <td>Not affected</td>
    </tr>
    <tr>
      <td><code class="language-plaintext highlighter-rouge">PasswordAuthentication no</code></td>
      <td>Not affected</td>
    </tr>
    <tr>
      <td>No account on the host has a <code class="language-plaintext highlighter-rouge">$</code>-style hash</td>
      <td>Not affected</td>
    </tr>
    <tr>
      <td>Accounts locked with <code class="language-plaintext highlighter-rouge">!</code> (rather than <code class="language-plaintext highlighter-rouge">*</code>)</td>
      <td>Not affected</td>
    </tr>
    <tr>
      <td>Cloud-init images that write <code class="language-plaintext highlighter-rouge">PasswordAuthentication no</code></td>
      <td>Not affected</td>
    </tr>
    <tr>
      <td>Containers that don’t run <code class="language-plaintext highlighter-rouge">sshd</code> at all</td>
      <td>Not affected</td>
    </tr>
  </tbody>
</table>

<p>Alpine checked releases 3.21 to 3.24 and edge, plus unsupported 3.16 to 3.20; all were affected. Alpine tiny-cloud images from 3.18.11 through 3.24.3 are safe on first boot but become exploitable as soon as any <code class="language-plaintext highlighter-rouge">$6$</code> hash exists on the machine.</p>

<p>Outside Alpine, think about <strong>any musl-based build of OpenSSH without PAM</strong>: custom embedded or appliance images, and Docker images that bake in an Alpine <code class="language-plaintext highlighter-rouge">sshd</code> (NVD’s references include issue threads from such projects). Ask your vendors.</p>

<h2 id="identifying-affected-systems">Identifying affected systems</h2>

<p>You need three facts per host: which libc <code class="language-plaintext highlighter-rouge">sshd</code> uses, whether PAM is in play, and whether the package includes the fix.</p>

<div class="language-bash highlighter-rouge"><div class="highlight"><pre class="highlight"><code><span class="c"># 1) Version. Note: distros backport, so the upstream number alone isn't the answer</span>
ssh <span class="nt">-V</span>
sshd <span class="nt">-V</span> 2&gt;&amp;1 | <span class="nb">head</span> <span class="nt">-n1</span>     <span class="c"># newer sshd builds; otherwise rely on the package version</span>

<span class="c"># 2) Is sshd linked against musl?</span>
ldd <span class="s2">"</span><span class="si">$(</span><span class="nb">command</span> <span class="nt">-v</span> sshd<span class="si">)</span><span class="s2">"</span> 2&gt;/dev/null | <span class="nb">grep</span> <span class="nt">-i</span> musl

<span class="c"># 3) Effective config (run as root): what sshd will actually do</span>
sshd <span class="nt">-T</span> | <span class="nb">grep</span> <span class="nt">-Ei</span> <span class="s1">'^(passwordauthentication|usepam|permitrootlogin) '</span>

<span class="c"># 4) Alpine: which openssh package and revision is installed?</span>
apk list <span class="nt">-I</span> <span class="s1">'openssh*'</span> 2&gt;/dev/null

<span class="c"># 5) Precondition check: accounts with '*' and whether any '$5$'/'$6$' hash exists</span>
<span class="nb">awk</span> <span class="nt">-F</span>: <span class="s1">'$2=="*" {print "no-password:", $1}'</span> /etc/shadow
<span class="nb">awk</span> <span class="nt">-F</span>: <span class="s1">'$2 ~ /^\$[56]\$/ {n++} END {print "accounts with $5/$6 hash:", n+0}'</span> /etc/shadow
</code></pre></div></div>

<p>A host is exposed when <code class="language-plaintext highlighter-rouge">sshd</code> is musl-linked, PAM is off (no PAM build, or <code class="language-plaintext highlighter-rouge">usepam no</code>), <code class="language-plaintext highlighter-rouge">passwordauthentication yes</code>, the package is older than the fixed revision, and step 5 shows at least one <code class="language-plaintext highlighter-rouge">*</code> account <strong>and</strong> at least one <code class="language-plaintext highlighter-rouge">$5$</code>/<code class="language-plaintext highlighter-rouge">$6$</code> hash.</p>

<p><strong>Alpine fixed package revisions</strong> (from Alpine’s advisory):</p>

<table>
  <thead>
    <tr>
      <th>Branch</th>
      <th>Fixed <code class="language-plaintext highlighter-rouge">openssh</code></th>
    </tr>
  </thead>
  <tbody>
    <tr>
      <td>edge</td>
      <td>10.6_p2-r0</td>
    </tr>
    <tr>
      <td>3.24</td>
      <td>10.3_p1-r2</td>
    </tr>
    <tr>
      <td>3.23</td>
      <td>10.2_p1-r1</td>
    </tr>
    <tr>
      <td>3.22</td>
      <td>10.0_p1-r11</td>
    </tr>
    <tr>
      <td>3.21</td>
      <td>9.9_p2-r1</td>
    </tr>
    <tr>
      <td>3.20</td>
      <td>9.7_p1-r6</td>
    </tr>
    <tr>
      <td>3.19</td>
      <td>9.6_p1-r3</td>
    </tr>
    <tr>
      <td>3.18</td>
      <td>9.3_p2-r4</td>
    </tr>
    <tr>
      <td>3.17</td>
      <td>9.1_p1-r7</td>
    </tr>
    <tr>
      <td>3.16</td>
      <td>9.0_p1-r6</td>
    </tr>
  </tbody>
</table>

<p>Alpine point releases 3.21.10, 3.22.8, 3.23.8 and 3.24.4 include the fixed package. Note how this works with backports: a 3.24 host on <code class="language-plaintext highlighter-rouge">10.3_p1-r2</code> is <strong>fixed</strong> even though <code class="language-plaintext highlighter-rouge">ssh -V</code> reports a version far below 10.6p2.</p>

<p><strong>Containers:</strong> check images, not just hosts. List images whose entrypoint or process list includes <code class="language-plaintext highlighter-rouge">sshd</code>, then run step 4 inside each:</p>

<div class="language-bash highlighter-rouge"><div class="highlight"><pre class="highlight"><code>docker ps <span class="nt">--format</span> <span class="s1">'{{.ID}} {{.Image}}'</span> | <span class="k">while </span><span class="nb">read</span> <span class="nt">-r</span> <span class="nb">id </span>img<span class="p">;</span> <span class="k">do
  </span>docker <span class="nb">exec</span> <span class="s2">"</span><span class="nv">$id</span><span class="s2">"</span> sh <span class="nt">-c</span> <span class="s1">'pgrep -x sshd &gt;/dev/null &amp;&amp; apk list -I "openssh*" 2&gt;/dev/null'</span> <span class="se">\</span>
    | <span class="nb">sed</span> <span class="s2">"s|^|</span><span class="nv">$img</span><span class="s2">: |"</span>
<span class="k">done</span>
</code></pre></div></div>

<h2 id="safe-validation-in-an-isolated-lab">Safe validation in an isolated lab</h2>

<p>You don’t need to attempt a bypass to prove you’re fixed or mitigated:</p>

<ol>
  <li>On an isolated VM or container network, build two Alpine images: one pinned to a pre-fix <code class="language-plaintext highlighter-rouge">openssh-server</code> revision and one on the fixed revision for the same branch.</li>
  <li>Run the checks above against both. Your inventory or vulnerability scanner should flag only the old one. If it flags neither, it’s probably matching on upstream version strings and missing Alpine revisions.</li>
  <li>On the old one, apply the interim mitigation (<code class="language-plaintext highlighter-rouge">PasswordAuthentication no</code>) and confirm <code class="language-plaintext highlighter-rouge">sshd -T</code> reports it. Restart <code class="language-plaintext highlighter-rouge">sshd</code> and confirm key-based login still works.</li>
  <li>Make a normal, legitimate password login to a test account and confirm the <code class="language-plaintext highlighter-rouge">Accepted password for</code> line reaches your SIEM. That’s the event you’ll hunt on below.</li>
</ol>

<h2 id="detection-and-threat-hunting">Detection and threat hunting</h2>

<p>SSH is encrypted, and GreyNoise points out that spotting attempts from network traffic alone is hard. Focus on <strong>authentication logs and host artifacts</strong>.</p>

<ul>
  <li><strong>Successful password logins to accounts that have no password.</strong> On Alpine, search <code class="language-plaintext highlighter-rouge">/var/log/messages</code> (or wherever syslog goes) for <code class="language-plaintext highlighter-rouge">Accepted password for &lt;user&gt;</code> where <code class="language-plaintext highlighter-rouge">&lt;user&gt;</code> is one of the <code class="language-plaintext highlighter-rouge">*</code> accounts from step 5. Any hit is a strong signal, because those accounts can’t legitimately log in with a password.
    <div class="language-bash highlighter-rouge"><div class="highlight"><pre class="highlight"><code><span class="nv">NOPW</span><span class="o">=</span><span class="si">$(</span><span class="nb">awk</span> <span class="nt">-F</span>: <span class="s1">'$2=="*" {print $1}'</span> /etc/shadow | <span class="nb">paste</span> <span class="nt">-sd</span><span class="s1">'|'</span><span class="si">)</span>
<span class="nb">grep</span> <span class="nt">-E</span> <span class="s2">"Accepted password for (</span><span class="nv">$NOPW</span><span class="s2">) "</span> /var/log/messages<span class="k">*</span> 2&gt;/dev/null
</code></pre></div>    </div>
  </li>
  <li><strong>Post-login activity.</strong> Alpine recommends reviewing <code class="language-plaintext highlighter-rouge">doas</code> use, newly created users, recently changed files and <code class="language-plaintext highlighter-rouge">authorized_keys</code> files. If you can’t rule out a login, treat the machine as compromised.</li>
  <li><strong>Published IOCs (GreyNoise, cluster “GT10”).</strong> C2 domain <code class="language-plaintext highlighter-rouge">tac.vpnla.org</code>; host artifacts <code class="language-plaintext highlighter-rouge">/opt/py3</code>, <code class="language-plaintext highlighter-rouge">/tmp/py3.tar.gz</code>, <code class="language-plaintext highlighter-rouge">/usr/local/bin/tac_agent.py</code>, <code class="language-plaintext highlighter-rouge">/etc/init.d/tac-agent</code>, and an extra <code class="language-plaintext highlighter-rouge">sshd</code> listening on <strong>port 2222</strong>; HASSH client fingerprints <code class="language-plaintext highlighter-rouge">a2de0f306611e0957be704f5b0e35a82</code> and <code class="language-plaintext highlighter-rouge">38da7a9dfb196c72a7354bb4cdce64d9</code>. GreyNoise also lists source IPs on its tracker page. GreyNoise notes that not all SSHole activity is attributable to GT10, so absence of these IOCs isn’t proof of a clean host.
    <div class="language-bash highlighter-rouge"><div class="highlight"><pre class="highlight"><code><span class="nb">ls</span> <span class="nt">-la</span> /opt/py3 /tmp/py3.tar.gz /usr/local/bin/tac_agent.py /etc/init.d/tac-agent 2&gt;/dev/null
ss <span class="nt">-ltnp</span> | <span class="nb">grep</span> <span class="nt">-E</span> <span class="s1">':2222\b'</span>
</code></pre></div>    </div>
  </li>
  <li><strong>Timeline.</strong> Per GreyNoise: fix published 5 Nov 09:00 UTC, first public exploit on GitHub 5 Nov 13:16 UTC, first successful exploitation on their sensors 7 Nov 00:46 UTC. <strong>Hunt back to at least 5 November</strong> on anything that was exposed.</li>
</ul>

<p>CISA set forensic triage to “Yes” for this entry, which tells you how it expects agencies to treat exposed hosts: look for compromise, don’t just patch.</p>

<h2 id="remediation">Remediation</h2>

<ol>
  <li><strong>Upgrade</strong> to OpenSSH 10.6p2, or on Alpine to the fixed revision for your branch: <code class="language-plaintext highlighter-rouge">apk upgrade -U -a</code>.</li>
  <li><strong>Restart sshd.</strong> Alpine stresses that upgrading alone leaves the old daemon running: <code class="language-plaintext highlighter-rouge">rc-service sshd restart</code>.</li>
  <li><strong>Rebuild and redeploy</strong> any container image that runs <code class="language-plaintext highlighter-rouge">sshd</code> on an Alpine base. Pinned base tags keep the bug.</li>
  <li><strong>Refresh cloud images.</strong> For tiny-cloud deployments, the fixed images start at 3.21.10, 3.22.8, 3.23.8 and 3.24.4. Existing instances need the package update or the mitigation below.</li>
  <li><strong>Hunt</strong> using the section above, and rotate credentials on any host you can’t clear.</li>
</ol>

<h3 id="interim-mitigation-if-you-cant-patch-today">Interim mitigation (if you can’t patch today)</h3>

<p>Turn off password authentication. It’s the documented workaround, and on most servers it’s what you want anyway:</p>

<div class="language-bash highlighter-rouge"><div class="highlight"><pre class="highlight"><code><span class="c"># Keep an existing session open while you do this, and confirm key login works afterwards</span>
<span class="nb">echo</span> <span class="s1">'PasswordAuthentication no'</span> <span class="o">&gt;</span> /etc/ssh/sshd_config.d/00-no-password.conf
<span class="nb">grep</span> <span class="nt">-q</span> <span class="s1">'^Include /etc/ssh/sshd_config.d/'</span> /etc/ssh/sshd_config <span class="o">||</span> <span class="nb">echo</span> <span class="s1">'check that your sshd_config includes sshd_config.d'</span>
sshd <span class="nt">-t</span> <span class="o">&amp;&amp;</span> rc-service sshd restart
sshd <span class="nt">-T</span> | <span class="nb">grep</span> <span class="nt">-i</span> <span class="s1">'^passwordauthentication'</span>
</code></pre></div></div>

<p><code class="language-plaintext highlighter-rouge">sshd</code> uses the <strong>first</strong> value it reads for each keyword, so if the drop-in isn’t included, or an earlier line sets <code class="language-plaintext highlighter-rouge">PasswordAuthentication yes</code>, fix <code class="language-plaintext highlighter-rouge">sshd_config</code> itself. The final <code class="language-plaintext highlighter-rouge">sshd -T</code> line is what counts. Alpine’s advisory has a fuller version with a backup and rollback. Also check that <code class="language-plaintext highlighter-rouge">PermitRootLogin</code> is at its <code class="language-plaintext highlighter-rouge">prohibit-password</code> default, and restrict port 22 to management networks where you can.</p>

<h2 id="references">References</h2>

<ul>
  <li><a href="https://nvd.nist.gov/vuln/detail/CVE-2026-123158">CVE-2026-123158, NVD</a> · <a href="https://www.cisa.gov/known-exploited-vulnerabilities-catalog?field_cve=CVE-2026-123158">CISA KEV entry</a></li>
  <li><a href="https://www.openssh.org/releasenotes.html#10.6p2">OpenSSH 10.6p2 release notes</a></li>
  <li><a href="https://github.com/openssh/openssh-portable/commit/4353fdc43ef646c43a8dda7025cc1a898caf1482">OpenSSH portable fix commit</a></li>
  <li><a href="https://alpinelinux.org/posts/2026-11-05-openssh-password-authentication-bypass/">Alpine Linux advisory: openssh password authentication bypass</a></li>
  <li><a href="https://sshole.events.greynoise.io/">GreyNoise SSHole tracker (timeline, IOCs)</a></li>
  <li><a href="http://www.openwall.com/lists/oss-security/2026/11/09/27">oss-security discussion</a></li>
</ul>

<p><em>Verify every version against the vendor advisory before acting. Published 10 Nov 2026.</em></p>]]></content><author><name>Aniket Tomar</name></author><category term="cve" /><category term="openssh" /><category term="linux" /><category term="authentication-bypass" /><category term="blue-team" /><summary type="html"><![CDATA[On musl-based Linux builds without PAM, sshd accepts any very long password for some accounts (CVSS 8.1, on CISA KEV). Root cause, who is exposed, detection, IOCs and the fix.]]></summary><media:thumbnail xmlns:media="http://search.yahoo.com/mrss/" url="https://binary0101devil.pages.dev/assets/img/cve-2026-123158-openssh.svg" /><media:content medium="image" url="https://binary0101devil.pages.dev/assets/img/cve-2026-123158-openssh.svg" xmlns:media="http://search.yahoo.com/mrss/" /></entry></feed>