HostSplit: Exploitable Antipatterns in Unicode Normalization
summary
A new class of Unicode normalization exploits lets a URL parse as one hostname but resolve as another, enabling URL-filter bypass and domain impersonation. The key trick uses IDNA normalization + Punycode where characters normalize into syntax (e.g. full‑width slash → /), enabling HostSplit redirects that once let Edge/IE forward Office 365 OAuth tokens via a covert redirect chain (live.com → Dropbox example). Tests use DNS monitoring or wildcard domains; vulnerable platforms included .NET, Python, Java, Windows APIs and libidn (many later patched). HostBond is a related invisible‑character variant.
AI-written summary · about
chapters (7)
- 0:00Introduction
- 0:34Agenda
- 8:08HostSplit
- 15:58Whats vulnerable
- 26:12Whats not vulnerable
- 32:17How to test
- 36:37Best practices
related talks
54mUnicodes Gone Wild
59mBrowsers Gone Wild
31mNew Techniques for Split-Second DNS Rebinding
42mModern web application bugs
56mUnicode - going down the rabbit hole
47mA New Era of SSRF - Exploiting URL Parser in Trending Programming Languages!
40mUnicode: The hero or villain?
29mChar Wars: The Path Traversal Strikes Back
