fast-uri 3.1.2 → 3.1.4

package/index.js

@@ -202,6 +202,10 @@ function serialize (cmpts, opts) {
 
 const URI_PARSE = /^(?:([^#/:?]+):)?(?:\/\/((?:([^#/?@]*)@)?(\[[^#/?\]]+\]|[^#/:?]*)(?::(\d*))?))?([^#?]*)(?:\?([^#]*))?(?:#((?:.|[\n\r])*))?/u
 
+// Captures the authority component (between "//" and the next "/", "?" or "#"),
+// with or without a scheme prefix, for the literal-backslash rejection below.
+const AUTHORITY_PREFIX = /^(?:[^#/:?]+:)?\/\/([^/?#]*)/
+
 /**
  * @param {import('./types/index').URIComponent} parsed
  * @param {RegExpMatchArray} matches
@@ -248,6 +252,19 @@ function parseWithStatus (uri, opts) {
     }
   }
 
+  // A literal backslash (U+005C) is not a valid RFC 3986 URI character and is
+  // not an authority delimiter. Reject it in the authority rather than
+  // rewriting it: normalizing "\" -> "/" (WHATWG error recovery) could silently
+  // change the resource identified by an otherwise-invalid input, and lets "\"
+  // act as a host delimiter here while Node's native URL parses a different
+  // host (SSRF / redirect / origin-allowlist bypass). Percent-encoded %5C is
+  // untouched and remains valid encoded data.
+  const authorityMatch = uri.match(AUTHORITY_PREFIX)
+  if (authorityMatch !== null && authorityMatch[1].indexOf('\\') !== -1) {
+    parsed.error = 'URI authority must not contain a literal backslash.'
+    malformedAuthorityOrPort = true
+  }
+
   const matches = uri.match(URI_PARSE)
 
   if (matches) {
@@ -305,7 +322,7 @@ function parseWithStatus (uri, opts) {
       if (parsed.host && (options.domainHost || (schemeHandler && schemeHandler.domainHost)) && isIP === false && nonSimpleDomain(parsed.host)) {
         // convert Unicode IDN -> ASCII IDN
         try {
-          parsed.host = URL.domainToASCII(parsed.host.toLowerCase())
+          parsed.host = new URL('http://' + parsed.host).hostname
         } catch (e) {
           parsed.error = parsed.error || "Host's domain name can not be converted to ASCII: " + e
         }

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "fast-uri",
   "description": "Dependency-free RFC 3986 URI toolbox",
-  "version": "3.1.2",
+  "version": "3.1.4",
   "main": "index.js",
   "type": "commonjs",
   "types": "types/index.d.ts",

package/test/security.test.js

@@ -131,3 +131,93 @@ test('normalize does not double-decode %2540 into a live @', (t) => {
   t.plan(1)
   t.notEqual(parsed.host, 'trusted.com@evil.com', 'http://trusted.com%2540evil.com/')
 })
+
+test('parse canonicalises IDN / Unicode hosts to their ASCII form', (t) => {
+  const cases = [
+    {
+      input: 'http://127。0。0。1/',
+      expectedHost: '127.0.0.1',
+      description: 'full-width ideographic stops as octet separators'
+    },
+    {
+      input: 'http://ｅxample.com/',
+      expectedHost: 'example.com',
+      description: 'fullwidth e as first letter'
+    },
+    {
+      input: 'http://納豆.example.org/',
+      expectedHost: 'xn--99zt52a.example.org',
+      description: 'CJK label requiring punycode'
+    }
+  ]
+
+  t.plan(cases.length * 2)
+
+  cases.forEach(({ input, expectedHost, description }) => {
+    const parsed = fastURI.parse(input)
+    t.notOk(parsed.error, `parse should not set error: ${description}`)
+    t.equal(parsed.host, expectedHost, `host canonicalised to ASCII: ${description}`)
+  })
+})
+
+test('parse rejects a literal backslash in the authority as malformed (RFC 3986)', (t) => {
+  // Regression for the host-confusion bypass: a literal "\" is invalid RFC 3986
+  // syntax and must be flagged malformed, not silently rewritten. Otherwise "\"
+  // acts as a host delimiter here while Node's native URL parses a different
+  // host, defeating a host-based SSRF/redirect/origin allowlist.
+  const cases = [
+    'http://evil.com\\@allowed.com',
+    'https://169.254.169.254\\@trusted.example.com',
+    'http://127.0.0.1\\@public.example.com',
+    'https://attacker.com\\@api.internal',
+    'http://a\\@b',
+    'ws://evil.com\\@allowed.com/chat',
+    'wss://evil.com\\@allowed.com/chat',
+    'http://evil.com\\%40allowed.com',
+    '//evil.com\\@allowed.com'
+  ]
+
+  t.plan(cases.length)
+
+  cases.forEach((input) => {
+    t.equal(
+      fastURI.parse(input).error,
+      'URI authority must not contain a literal backslash.',
+      input
+    )
+  })
+})
+
+test('normalize does not canonicalize a literal-backslash URI into a different valid URL', (t) => {
+  const cases = [
+    'http://evil.com\\@allowed.com',
+    'https://attacker.com\\@api.internal'
+  ]
+
+  t.plan(cases.length)
+
+  cases.forEach((input) => {
+    t.equal(fastURI.normalize(input), input, input)
+  })
+})
+
+test('parse leaves percent-encoded %5C untouched as encoded data (not rejected)', (t) => {
+  // Only the literal "\" byte is rejected; %5C stays valid encoded data and
+  // does not diverge from the native URL parser, so it must not be flagged.
+  const input = 'http://evil.com%5C@allowed.com'
+  const parsed = fastURI.parse(input)
+
+  t.plan(2)
+  t.notOk(parsed.error, '%5C is valid encoded data, not malformed')
+  t.equal(parsed.host, new URL(input).hostname, '%5C host matches native URL (no divergence)')
+})
+
+test('parse does not reject a literal backslash in the query or fragment', (t) => {
+  // The rejection is scoped to the authority/path (the host-confusion surface);
+  // a backslash after "?"/"#" is normalized as encoded data as before.
+  const parsed = fastURI.parse('http://host.example.com/?x=\\y#z\\w')
+
+  t.plan(2)
+  t.notOk(parsed.error, 'backslash in query/fragment does not mark the URI malformed')
+  t.equal(parsed.host, 'host.example.com', 'host parsed normally')
+})