<?xml version="1.0" encoding="utf-8" standalone="yes"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
	<channel>
		<title>EBPF on Catscrdl</title>
		<link>https://catscrdl.io/tags/ebpf/</link>
		<description>Recent content in EBPF on Catscrdl</description>
		<generator>Hugo</generator>
		<language>en</language>
		
		
		
		
			<lastBuildDate>Tue, 31 Mar 2026 00:00:00 +0000</lastBuildDate>
		
			<atom:link href="https://catscrdl.io/tags/ebpf/index.xml" rel="self" type="application/rss+xml" />
			<item>
				<title>Container Escape Telemetry, Part 1: Isolation Primitives and the eBPF Observability Model</title>
				<link>https://catscrdl.io/blog/containerescapetelemetry/isolation/</link>
				<pubDate>Tue, 31 Mar 2026 00:00:00 +0000</pubDate>
				<guid>https://catscrdl.io/blog/containerescapetelemetry/isolation/</guid>
				<description>&lt;p&gt;This is Part 1 of the container escape telemetry series. The &lt;a href=&#34;../intro&#34;&gt;series overview&lt;/a&gt; has the project goals, key findings, and reading guide. If you&amp;rsquo;re already comfortable with namespaces, cgroups, capabilities, and how eBPF ring buffers work, skip ahead to &lt;a href=&#34;../methodology&#34;&gt;Part 2: Methodology and Tool Architecture&lt;/a&gt;.&lt;/p&gt;&#xA;&lt;p&gt;This post is the primer. It covers what containers actually are at the kernel level, what &amp;ldquo;escape&amp;rdquo; means in concrete terms, and how eBPF-based tools get telemetry out of the kernel. That last part matters more than you might think, because the mechanism that delivers telemetry to your security tool has its own failure modes, and an attacker who understands those failure modes can exploit them.&lt;/p&gt;</description>
			</item>
			<item>
				<title>Container Escape Telemetry, Part 2: Methodology and Tool Architecture</title>
				<link>https://catscrdl.io/blog/containerescapetelemetry/methodology/</link>
				<pubDate>Tue, 31 Mar 2026 00:00:00 +0000</pubDate>
				<guid>https://catscrdl.io/blog/containerescapetelemetry/methodology/</guid>
				<description>&lt;p&gt;This is Part 2 of the container escape telemetry series (&lt;a href=&#34;../intro&#34;&gt;overview&lt;/a&gt;). &lt;a href=&#34;../isolation&#34;&gt;Part 1&lt;/a&gt; covered the isolation primitives and eBPF observability model. This post covers the lab architecture, the three tools under test, the 15 escape scenarios, and the detection coverage matrix. If you want to skip straight to what the telemetry actually looks like, jump to &lt;a href=&#34;../deepdives&#34;&gt;Part 3: Per-Scenario Deep Dives&lt;/a&gt;.&lt;/p&gt;&#xA;&lt;h1 id=&#34;the-lab&#34;&gt;The Lab&lt;/h1&gt;&#xA;&lt;p&gt;Each tool ran on its own dedicated Hyper-V VM: 3 GB RAM, 2 vCPUs, Ubuntu 22.04 on kernel 5.15.0-91-generic. One tool per VM. This was non-negotiable. eBPF programs from different tools will contend for the same kernel hooks, and if you&amp;rsquo;re trying to measure event ordering and timing, that contention corrupts your data. You can&amp;rsquo;t draw conclusions about &amp;ldquo;tool A generated event X before tool B generated event Y&amp;rdquo; if both tools are fighting over the same kprobe attachment points on the same kernel.&lt;/p&gt;</description>
			</item>
			<item>
				<title>Container Escape Telemetry, Part 3: What Each Tool Actually Captured</title>
				<link>https://catscrdl.io/blog/containerescapetelemetry/deepdives/</link>
				<pubDate>Tue, 31 Mar 2026 00:00:00 +0000</pubDate>
				<guid>https://catscrdl.io/blog/containerescapetelemetry/deepdives/</guid>
				<description>&lt;p&gt;This is Part 3 of the container escape telemetry series (&lt;a href=&#34;../intro&#34;&gt;overview&lt;/a&gt;). &lt;a href=&#34;../isolation&#34;&gt;Part 1&lt;/a&gt; covered isolation primitives and the eBPF observability model. &lt;a href=&#34;../methodology&#34;&gt;Part 2&lt;/a&gt; covered the lab, the tools, and the detection coverage matrix. This post is the meat: per-scenario telemetry breakdowns showing what each tool actually captured, where the qualitative differences behind the checkmarks become concrete.&lt;/p&gt;&#xA;&lt;p&gt;Not every scenario gets a deep dive. Some produced straightforward results that the coverage matrix already summarizes. I&amp;rsquo;m focusing on the scenarios where the cross-tool comparison reveals something interesting about how these tools work, where they disagree, and what the disagreements mean for detection engineering.&lt;/p&gt;</description>
			</item>
			<item>
				<title>Container Escape Telemetry, Part 4: Volume, Signal-to-Noise, and Choosing a Tool</title>
				<link>https://catscrdl.io/blog/containerescapetelemetry/production/</link>
				<pubDate>Tue, 31 Mar 2026 00:00:00 +0000</pubDate>
				<guid>https://catscrdl.io/blog/containerescapetelemetry/production/</guid>
				<description>&lt;p&gt;This is Part 4 of the container escape telemetry series (&lt;a href=&#34;../intro&#34;&gt;overview&lt;/a&gt;). &lt;a href=&#34;../isolation&#34;&gt;Part 1&lt;/a&gt; covered isolation primitives and the eBPF observability model. &lt;a href=&#34;../methodology&#34;&gt;Part 2&lt;/a&gt; covered the lab and detection coverage matrix. &lt;a href=&#34;../deepdives&#34;&gt;Part 3&lt;/a&gt; walked through per-scenario telemetry. This post is about the operational questions: how much data do these tools produce, what percentage of it matters, and which tool should you actually deploy?&lt;/p&gt;&#xA;&lt;h1 id=&#34;the-volume-question&#34;&gt;The Volume Question&lt;/h1&gt;&#xA;&lt;p&gt;Parts 2 and 3 focused on what each tool sees. But there&amp;rsquo;s an equally important dimension: how &lt;em&gt;much&lt;/em&gt; does each tool generate, and what&amp;rsquo;s the signal-to-noise ratio of that output? If you&amp;rsquo;re going to deploy any of these tools in production, the telemetry volume has direct implications for storage, bandwidth, and whether an analyst can actually find the escape signal in the noise.&lt;/p&gt;</description>
			</item>
			<item>
				<title>Container Escape Telemetry, Part 5: Tuning eBPF Tools From Defaults to Detection</title>
				<link>https://catscrdl.io/blog/containerescapetelemetry/tuning/</link>
				<pubDate>Tue, 31 Mar 2026 00:00:00 +0000</pubDate>
				<guid>https://catscrdl.io/blog/containerescapetelemetry/tuning/</guid>
				<description>&lt;p&gt;This is Part 5 of the container escape telemetry series (&lt;a href=&#34;../intro&#34;&gt;overview&lt;/a&gt;). &lt;a href=&#34;../isolation&#34;&gt;Part 1&lt;/a&gt; covered isolation primitives. &lt;a href=&#34;../methodology&#34;&gt;Part 2&lt;/a&gt; covered the lab and tools. &lt;a href=&#34;../deepdives&#34;&gt;Part 3&lt;/a&gt; was the per-scenario data. &lt;a href=&#34;../production&#34;&gt;Part 4&lt;/a&gt; covered volume, signal-to-noise, and tool selection. This post is for the practitioner who just installed one of these tools and wants to know what to do next.&lt;/p&gt;&#xA;&lt;p&gt;The gap between &amp;ldquo;tool installed&amp;rdquo; and &amp;ldquo;tool detecting container escapes&amp;rdquo; is larger than any of these projects&amp;rsquo; documentation suggests. Here&amp;rsquo;s what that gap actually looks like and how to close it.&lt;/p&gt;</description>
			</item>
			<item>
				<title>Container Escape Telemetry, Part 6: TeamPCP and What the Lab Predicted</title>
				<link>https://catscrdl.io/blog/containerescapetelemetry/teampcp/</link>
				<pubDate>Tue, 31 Mar 2026 00:00:00 +0000</pubDate>
				<guid>https://catscrdl.io/blog/containerescapetelemetry/teampcp/</guid>
				<description>&lt;p&gt;This is Part 6 of the container escape telemetry series (&lt;a href=&#34;../intro&#34;&gt;overview&lt;/a&gt;). Parts &lt;a href=&#34;../isolation&#34;&gt;1&lt;/a&gt;-&lt;a href=&#34;../tuning&#34;&gt;5&lt;/a&gt; covered isolation primitives, methodology, per-scenario telemetry, production considerations, and tuning. This post takes the lab findings and pressure-tests them against a real threat actor operating in the wild right now.&lt;/p&gt;&#xA;&lt;p&gt;TeamPCP compromised Trivy on March 19, 2026. Five days later they backdoored LiteLLM&amp;rsquo;s PyPI package. As of late March 2026, incident response teams across the industry are still assessing the blast radius. But the container escape techniques TeamPCP uses aren&amp;rsquo;t novel. They&amp;rsquo;re the same patterns our 15 lab scenarios were designed to detect. That&amp;rsquo;s either a validation of the research or a coincidence. Having spent weeks staring at this telemetry, I don&amp;rsquo;t think it&amp;rsquo;s a coincidence.&lt;/p&gt;</description>
			</item>
			<item>
				<title>Container Escape Telemetry: Series Overview</title>
				<link>https://catscrdl.io/blog/containerescapetelemetry/intro/</link>
				<pubDate>Tue, 31 Mar 2026 00:00:00 +0000</pubDate>
				<guid>https://catscrdl.io/blog/containerescapetelemetry/intro/</guid>
				<description>&lt;p&gt;As a detection engineer, container escapes have always fascinated me. Detections focused on control bypasses can be both extremely difficult to make and extremely valuable when they fire. The challenge is that most discussions about container runtime security focus on &lt;em&gt;whether&lt;/em&gt; a tool detected an escape &amp;ndash; a binary yes/no that doesn&amp;rsquo;t tell you much about the underlying telemetry that makes detection possible in the first place.&lt;/p&gt;&#xA;&lt;p&gt;There&amp;rsquo;s also a marketing problem. I&amp;rsquo;ve seen vendors demonstrate container escape detection against one or two scenarios &amp;ndash; usually cgroup release_agent or nsenter &amp;ndash; and present that as comprehensive coverage. It isn&amp;rsquo;t. A tool that detects cgroup escapes but misses DirtyPipe, or catches namespace transitions but is blind to capability abuse, provides a false sense of security. You can&amp;rsquo;t evaluate coverage without testing breadth.&lt;/p&gt;</description>
			</item>
	</channel>
</rss>
