Skip to content
KitploitKITPLOIT
ToolsExploitsBlog
Log in
Submit
ToolsExploitsBlog
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
jellyfin-svg-gzip — Jellyfin Server 12.2 - High - Remote - SSRF - gzip data:image/png leftover of GHSA-cf3c still fetches http and file | Kitploit
Tools/GitHubGitHub/abraxas/jellyfin-svg-gzip
Vulnerability AnalysisExploitationWeb Application ExploitationInformation GatheringPenetration TestingLabs & Practice
GitHubabraxas/jellyfin-svg-gzip

jellyfin-svg-gzip

Jellyfin Server 12.2 - High - Remote - SSRF - gzip data:image/png leftover of GHSA-cf3c still fetches http and file

View Repository
221 month agoNot yet reviewed
Share

Abraxas Labs - jellyfin-svg-gzip

abraxaslabs.tech  ·  github.com/abraxas  ·  @abraxas_null  ·  [email protected]  ·  jellyfin-svg-gzip

jellyfin-svg-gzip

Class: SSRF + file read + memex Reach: Remote (any authenticated user POST /UserImage; rasterize is unauth GET /UserImage)

Jellyfin Server 12.2 (ca0f16eb) - Jellyfin. Svg.Skia 3.7.0.

GHSA-cf3c-8m59-2vhx closed outer http and file hrefs on profile SVGs. 12.2 still pins Svg.Skia 3.7.0. SvgDocument.ResolveExternalImages = ExternalType.None is ignored by 3.7.0 ImageDrawable. SvgSecurityValidator.ValidateDataUri returns allow for any data URI whose media type is not exactly image/svg+xml, before the gzip walk. Svg.Skia GetImageFromDataUri: mimeType.StartsWith("image/") plus gzip magic calls LoadSvgz with no Jellyfin re-scan. The inner SVG's http(s) / file hrefs go to WebRequest.Create. That is the original GHSA-cf3c sink.

Any authenticated user uploads a gzip-png wrapped SVG. An unauthenticated GET of that profile image makes the server fetch attacker http / https / file:// while rasterizing.

IDno CVE yet
ClassSSRF + file read + memex
ReachRemote. Any user POST /UserImage. Rasterize is unauth GET /UserImage?userId=
CWECWE-918 / CWE-611 / CWE-400
CVSSHigh: 7.7 CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:C/C:H/I:N/A:L
ProductJellyfin Server
Affected12.2 (ca0f16eb7c195f720a1493ed469a27c4657db0c6), image jellyfin/jellyfin:12.2, Svg.Skia 3.7.0
AuthAuthenticated upload. Unauthenticated rasterize.
LicenseGNU Affero GPL v3.0
Lab127.0.0.1 only

Leftover #2 of the Jellyfin 12.2 leftovers series. Sibling #1 Suggestions IDOR. Same pin: playlist ItemIds, FFmpeg colon-slash.

What an attacker can do

Actor is any signed-in Jellyfin user. Profile image is not admin-only. Rasterize does not need a session.

AxisMethod + pathGrantsProvenance
ReconPOST /UserImage (Content-Type: image/svg+xml, BASE64 body) then unauth GET /UserImage?userId=Server GET of attacker http/https (loopback, RFC1918, link-local/IMDS) while Skia rasterizesLab-proven. Oracle hostname oracle:8080, hits=1, witness in path
Reconsamefile:// bytes painted into the PNG/WebP/JPEG the requester downloads, if they decode as an imageInferred. Same WebRequest.Create sink as GHSA-cf3c
Availabilitysame, inner file:///dev/urandomUnbounded read retained on the managed heap. Repeats on every later image GET without re-uploadInferred. Measured in GHSA-cf3c
IntegritynoneNo user-policy write, no plugin install, no session stealClosed

Stays closed: direct http/file hrefs on the outer SVG (lab: GET stayed image/svg+xml, oracle 0); nested data:image/svg+xml (re-scanned, including gzip); DTD SYSTEM; GHSA-vcmh inline XSS (Content-Disposition: attachment plus rasterize); plugin-image traversal; Magick/PDF. Not RCE. Not session XSS. SSRF body is what Skia paints, not the raw HTTP response.

How I found it

12.2 is the tag that already shipped the GHSA-cf3c pre-scan. I sat on it anyway. Patches that add a validator in front of a library that still fetches are a gift: you read what the validator walks, then you ask what the library still parses.

Directory.Packages.props still pins Svg.Skia 3.7.0. Upstream wieslawsoltes/Svg.Skia#448 (fixed in #501): ImageDrawable never consults ResolveExternalImages. Jellyfin's three ExternalType.None assignments are a no-op on this pin. Renovate #17076 is not in the tag.

The pre-scan is then the only gate. ValidateReference rejects every non-fragment, non-data: href. Nested image/svg+xml is re-parsed, including gzip. The comment at SvgSecurityValidator.cs:169-173 says any other media type is raster and returns allow before the 0x1F 0x8B walk. Svg.Skia 3.7.0 GetImageFromDataUri disagrees: any image/* with gzip magic is LoadSvgz.

Wrong turns already recorded: treating GHSA-cf3c as closed because the negative control stays image/svg+xml; nested data:image/svg+xml gzip (re-scanned); DTD SYSTEM; GHSA-vcmh inline XSS; plugin-image traversal; Magick/PDF; admin RemoteImages/Download (intended artwork fetch); a pixel-bomb DoS that is not this leftover; a reverse shell. Theatre.

Lab

python3 jellyfin-svg-gzip-Abraxas-Labs.py

or:

cd lab
./run.sh

Loopback only: Jellyfin http://127.0.0.1:18780, oracle http://127.0.0.1:18781. Image jellyfin/jellyfin:12.2. Oracle returns a 1x1 PNG and logs the request path. Compose hostname oracle:8080 is what Skia fetches.

Negative: POST a profile SVG whose image href is http://oracle:8080/negative.png. GET stays image/svg+xml, oracle 0. Positive: gzip of an inner SVG whose href is http://oracle:8080/JELLYFIN-SVG-SVGSKIA-WITNESS.png, wrapped as data:image/png;base64 on the outer SVG. GET is image/png, oracle 1.

SUCCESS jellyfin-svg-gzip gzip-png-ssrf hits=1 negative=0 JELLYFIN-SVG-SVGSKIA-WITNESS

Witness-only. No exploit payload. No reverse shell. Authorized lab. Loopback.

The fix

Walk gzip and nested data URIs in SvgSecurityValidator even when the declared media type is not image/svg+xml. Or bump Svg.Skia past the ImageDrawable ExternalType.None ignore (upstream #448 / #501; renovate #17076 is not in this tag). Both. The library flag is the intended kill switch. The validator should not lie about what the library will parse.

References

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.

    916
  • 2

    ghostlock-app

    yukonga/ghostlock-app

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

    558
  • 3

    redforge

    cybersecurity91/redforge

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

    370
  • 4

    ARTEX

    baarai/artex

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

    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.

    317
  • 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.

    289

Explore all tools

Browse our collection of tools

View all tools →