Skip to content
KitploitKITPLOIT
ToolsBlog
Log in
Submit
ToolsBlog
Submit

Hacking, PenTest, and Cybersecurity Tools for Your Security Arsenal!

Kitploit is a directory of hacking, cybersecurity, and pentesting tools. Discover the latest project updates to find vulnerabilities, analyze systems, automate testing, and strengthen your security.

FeedsContactPrivacy© 2026 Kitploit

Tool Directory

Categories

View all categories
Loading categories
VsockDrop-Adapted-CVE-2026-53365 — A slightly more robust version of MaherAzzouzi's POC, created by my agent on the fly during testing. | Kitploit
Tools/GitHubGitHub/hoodoer/vsockdrop-adapted-cve-2026-53365
Privilege EscalationMemory ForensicsExploitationCTFPenetration TestingBinary Exploitation
GitHubhoodoer/vsockdrop-adapted-cve-2026-53365

VsockDrop-Adapted-CVE-2026-53365

A slightly more robust version of MaherAzzouzi's POC, created by my agent on the fly during testing.

Share
View Repository
11 month agoNot yet reviewed

VsockDrop-Adapted

An adapted version of MaherAzzouzi's VsockDrop exploit for CVE-2026-53365, modified for reliability on busy systems where the original PoC's page reclaim race is unreliable.

These modifications were written autonomously by an AI model during a penetration testing engagement. No human wrote or reviewed the exploit code before it was compiled and executed. See How This Came About for the full story.

CVE-2026-53365

A reference-count underflow in the Linux kernel's io_uring zerocopy send path over AF_VSOCK. When a large message is fragmented into multiple skbs, the zerocopy completion tracking (uarg) is only attached to the final skb. The earlier skbs carry pinned user pages with no completion callback, so each send-and-drain cycle applies one extra put_page to those pages.

After enough iterations the pin bias drops to zero, the pages hit the per-cpu page (PCP) free list, and a cold read of a setuid binary (e.g. /usr/bin/su) reclaims one as pagecache. The attacker still holds an alias through the io_uring fixed buffer and rewrites the ELF .interp string to point at a controlled stub, which runs as root when su is exec'd.

  • Affected: Linux 6.7 through 7.0.10 (and corresponding stable branches)
  • Fixed: 7.0.11, 6.12.97, 6.8.0-147 (Ubuntu), and vendor backports
  • Requirements: io_uring enabled, vsock_loopback module loaded, /dev/vsock accessible (typically mode 0666)

What's Different

The original VsockDrop PoC hardcodes 1024 sends to exhaust the io_uring pin refcount and assumes the freed page lands cleanly in the su pagecache on the first try. On a quiet VM this works reliably. On a busy system (multiple services, memory pressure, PCP churn) two things go wrong:

1. Drain point varies

The refcount bias at pin time depends on kernel config and runtime state. On Ubuntu 6.8.0-136 with default settings, the drain point was ~1994 sends rather than 1024. This version probes dynamically: each iteration dumps the io_uring fixed buffer through a snapshot file and scans for an ELF header, stopping at the exact iteration where reclaim succeeds.

2. Sacrificial page dig

After the page is freed to PCP, another allocation (not the su pagecache read) may grab it first. The original PoC has no recovery path for this. This version implements a page-level heap feng shui technique:

  • Each send iteration allocates a small "sacrificial" page via mmap MAP_ANONYMOUS, tagged with a unique byte (0x01 + (n & 0xFF))
  • After each send, the fixed buffer is dumped and checked for non-0xAA pages (the buffer was initialized to 0xAA)
  • If a sacrificial page grabbed the freed page instead of su, the exploit identifies which one by scanning the dump for the tag byte
  • It frees only that sacrificial page, pushing the target page back to PCP-top
  • It then evicts and re-reads su's page 0, which now reclaims the correct page

This makes the exploit reliable on systems where PCP has active churn from other allocators.

3. Staging changes

  • The setuid helper is placed at /var/tmp/.s (the original uses /tmp), since Ubuntu 26.04 mounts /tmp with nosuid
  • The interp stub and rootsh helper are embedded as C arrays at compile time (via od/sed, no xxd dependency) so the binary is fully self-contained
  • The entire bug runs in a forked worker process; the worker's io_uring ring and page pins are torn down on exit before the clean parent touches the setuid helper, avoiding use-after-free faults on kernel 7.0+

How This Came About

This exploit was not written by a human. It was produced autonomously by Qwen3.8-Flash-Next (~170B parameter model, uncensored) running on a Halogen Flash Server backend, during an unassisted penetration testing engagement against a Hack The Box challenge machine.

The model was operating inside an autonomous pentesting harness (OpenCode + custom plugins) with no human guidance during the exploitation phase. After gaining an unprivileged shell on a Hack The Box target, it found CVE-2026-53365 via web search, downloaded the original PoC, identified that the target kernel was in the affected range, and ran it. When the original failed (drain point mismatch), the model diagnosed the issue, adapted the drain logic, added the sacrificial page technique, and iterated until it achieved reliable root — all without human input. The exploit development took approximately 4 hours of autonomous operation.

Building

Requires liburing (static library + headers). On Ubuntu/Debian:

apt-get install liburing-dev

Or build from source:

git clone --depth 1 https://github.com/axboe/liburing.git
cd liburing && ./configure && make -j$(nproc) && cd ..

Then:

# If liburing is installed system-wide:
make LIBURING=/usr

# If built from source alongside this directory:
make LIBURING=./liburing

This produces a single static binary exploit with the interp stub and rootsh helper embedded.

Usage

# Default: probe up to 1024 iterations
./exploit

# Specify max iterations (use higher values for kernels with larger pin bias)
./exploit 2500

The exploit will print HIT at send <N> when it successfully reclaims su's page 0, or DIG aa=<N> elf=<page> when the sacrificial page technique fires. On success, you get a root shell. On failure, it prints a diagnostic (no elf page, no interp, etc.) and exits non-zero.

After the exploit runs, the target's su binary has a corrupted pagecache and the marimo/application server may OOM. Run this only in disposable environments (CTF boxes, lab VMs).

Credits

  • Original exploit: Maher Azzouzi (vsockdrop) — discovered and reported CVE-2026-53365 to [email protected], wrote the original PoC
  • Adaptations: Qwen3.8-Flash-Next (autonomous AI agent), during an engagement operated by hoodoer.

License

MIT — see LICENSE. Same license as the original VsockDrop.

Disclaimer

This software is provided for authorized security testing, research, and educational purposes only. Do not use it against systems you do not own or have explicit written permission to test. The authors are not responsible for misuse.

Download Tool

Most Popular

View all →

Discover the most used tools by our community.

  • 1

    OsintHunter

    letox4/osinthunter

    🎯 The ultimate OSINT toolkit: investigation graph, confidence scoring, alias families, map, timeline and reports. 49 modules, 400+ sites, no API keys required.

    867
  • 2

    ghostlock-app

    yukonga/ghostlock-app

    GhostLock One-Tap Execution App (CVE-2026-43499)

    544
  • 3

    ARTEX

    baarai/artex

    LLM multi-agent autonomous penetration testing platform with planner/worker agents, asset and exploration graphs, MITM traffic recording, and vulnerability retesting.

    385
  • 4

    redforge

    cybersecurity91/redforge

    Browser-side red team toolkit with 102 offline tools for recon, payload generation, credential attacks, crypto, encoding and engagement reporting.

    360
  • 5

    looking-glass

    anonghosty/looking-glass

    CSIRT platform that ingests Shadowserver reports into MongoDB and replays them as interactive maps, analysis, findings, advisories and executive dashboards.

    302
  • 6

    CyberStrikeAI

    ed1s0nz/cyberstrikeai

    The system of action for AI-native cybersecurity—where intent becomes governed execution, evidence becomes operational memory, and every operation improves the next.

    279

Explore all tools

Browse our collection of tools

View all tools →