Starlet (Perl web server) can be stalled by crafted chunked requests
A specially crafted web request can make the Starlet web server spend a lot of CPU time on one request, slowing or freezing that worker. An attacker can do this with a single request, without logging in.
| Severity | HighVulnerability |
|---|---|
| Being exploited? | UnclearNo reports either way yet |
| Fix available? | YesStarlet 0.33 or later · since 30 Oct 2026 |
| How sure is this? | Official, 2 sourcesBacked by an official source |
How the attack works simplified
- 1AttackerSends one crafted chunked request
- 2Chunk-size lineLong hex run followed by bare LF
- 3Regex engineBacktracks, cost grows quadratically
- 4Worker stalledCPU busy for about a minute
- ResultWorker is tied up, service degrades
What happened
Starlet, a Perl PSGI web server, uses a regular expression to decode chunk-size lines in chunked request bodies. Long runs of hex digits followed by a bare LF trigger catastrophic regex backtracking, causing CPU exhaustion.
Are you affected?
You are affected if you run Starlet from version 0.20 up to (not including) 0.33 for Perl and accept chunked request bodies from untrusted clients.
Starlet for Perl, versions from 0.20 before 0.33
What to do
- Upgrade Starlet to version 0.33 or later
- Check your Perl dependency lists (cpanfile, Makefile.PL, or deployment images) to confirm the installed version
- If you cannot upgrade right away, limit request sizes and timeouts at a reverse proxy in front of Starlet
Timeline
- 30 Oct 2026, 13:18 UTCPublicly disclosed
- 30 Oct 2026, 15:45 UTCReported by oss-security · oss-security
- 30 Oct 2026, 16:00 UTCPicked up by Zero Radar
- 30 Oct 2026, 17:22 UTCNVD reported on it · NVD
Sources 2 reports, oldest first
Official vendor, government CERT or advisory database
- CVE-2026-115866NVD · 30 Oct 2026, 13:18 UTC
Security researchers research teams and mailing lists
- Starlet versions before 0.33 for Perl (4 CVEs)oss-security · 30 Oct 2026, 15:45 UTC
Technical details
- ID
- CVE-2026-115866
- Disclosed
- 30 Oct 2026, 13:18 UTC
- On Zero Radar
- 30 Oct 2026, 16:00 UTC (3 h after disclosure)
- Last change
- 30 Oct 2026, 17:22 UTC
The plain-language parts and the attack diagram were written by an AI model (anthropic/claude-haiku-5.5) from the sources above. Facts like versions, exploitation and dates come from the sources themselves. How Zero Radar works