fast-uri 4.0.1 → 4.1.1

package/index.js

@@ -1,6 +1,6 @@
 'use strict'
 
-const { normalizeIPv6, removeDotSegments, recomposeAuthority, normalizePercentEncoding, normalizePathEncoding, escapePreservingEscapes, reescapeHostDelimiters, isIPv4, nonSimpleDomain } = require('./lib/utils')
+const { normalizeIPv6, removeDotSegments, recomposeAuthority, normalizePercentEncoding, normalizePathEncoding, normalizeQueryFragmentEncoding, escapePreservingEscapes, reescapeHostDelimiters, isIPv4, nonSimpleDomain } = require('./lib/utils')
 const { SCHEMES, getSchemeHandler } = require('./lib/schemes')
 
 /**
@@ -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) {
@@ -325,12 +342,11 @@ function parseWithStatus (uri, opts) {
       if (parsed.path) {
         parsed.path = normalizePathEncoding(parsed.path)
       }
+      if (parsed.query) {
+        parsed.query = normalizeQueryFragmentEncoding(parsed.query)
+      }
       if (parsed.fragment) {
-        try {
-          parsed.fragment = encodeURI(decodeURIComponent(parsed.fragment))
-        } catch {
-          parsed.error = parsed.error || 'URI malformed'
-        }
+        parsed.fragment = normalizeQueryFragmentEncoding(parsed.fragment)
       }
     }
 

package/lib/utils.js

@@ -15,6 +15,9 @@ const isUnreserved = RegExp.prototype.test.bind(/^[\da-z\-._~]$/iu)
 /** @type {(value: string) => boolean} */
 const isPathCharacter = RegExp.prototype.test.bind(/^[\da-z\-._~!$&'()*+,;=:@/]$/iu)
 
+/** @type {(value: string) => boolean} */
+const isQueryFragmentCharacter = RegExp.prototype.test.bind(/^[\da-z\-._~!$&'()*+,;=:@/?]$/iu)
+
 /**
  * @param {Array<string>} input
  * @returns {string}
@@ -218,7 +221,7 @@ function removeDotSegments (path) {
           continue
         }
       } else if (input[0] === '/') {
-        if (input[1] === '.' || input[1] === '/') {
+        if (input[1] === '.') {
           output.push('/')
           break
         }
@@ -435,6 +438,63 @@ function normalizePathEncoding (input) {
   return output
 }
 
+/**
+ * Normalizes the percent-encoding of a query or fragment component.
+ *
+ * Like `normalizePathEncoding`, but uses the query/fragment character set
+ * (which additionally allows `?`) and decodes `.` since it has no dot-segment
+ * meaning outside of a path.
+ *
+ * @param {string} input
+ * @returns {string}
+ */
+function normalizeQueryFragmentEncoding (input) {
+  let output = ''
+
+  for (let i = 0; i < input.length; i++) {
+    const ch = input[i]
+    if (ch === '%' && i + 2 < input.length) {
+      const hex = input.slice(i + 1, i + 3)
+      if (isHexPair(hex)) {
+        const normalizedHex = hex.toUpperCase()
+        const decoded = String.fromCharCode(parseInt(normalizedHex, 16))
+
+        if (isUnreserved(decoded)) {
+          output += decoded
+        } else {
+          output += '%' + normalizedHex
+        }
+
+        i += 2
+        continue
+      }
+    }
+
+    if (isQueryFragmentCharacter(ch)) {
+      output += ch
+    } else {
+      const code = input.charCodeAt(i)
+      if (code < 0x80) {
+        output += isEscapeSafe(code) ? ch : BYTE_HEX[code]
+      } else if (code < 0xD800 || code > 0xDFFF) {
+        output += percentEncodeNonAscii(code)
+      } else if (code <= 0xDBFF && i + 1 < input.length) {
+        const low = input.charCodeAt(i + 1)
+        if (low >= 0xDC00 && low <= 0xDFFF) {
+          output += percentEncodeNonAscii(0x10000 + ((code - 0xD800) << 10) + (low - 0xDC00))
+          i++
+        } else {
+          output += percentEncodeNonAscii(0xFFFD)
+        }
+      } else {
+        output += percentEncodeNonAscii(0xFFFD)
+      }
+    }
+  }
+
+  return output
+}
+
 /**
  * Escapes a component while preserving existing valid percent escapes.
  *
@@ -515,6 +575,7 @@ module.exports = {
   reescapeHostDelimiters,
   normalizePercentEncoding,
   normalizePathEncoding,
+  normalizeQueryFragmentEncoding,
   escapePreservingEscapes,
   removeDotSegments,
   isIPv4,

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "fast-uri",
   "description": "Dependency-free RFC 3986 URI toolbox",
-  "version": "4.0.1",
+  "version": "4.1.1",
   "main": "index.js",
   "type": "commonjs",
   "types": "types/index.d.ts",

package/test/parse.test.js

@@ -150,10 +150,12 @@ test('URI parse', (t) => {
   t.equal(components.query, undefined, 'query')
   t.equal(components.fragment, '%0D', 'fragment')
 
-  // malformed percent-encoded fragment must not throw
+  // a fragment whose decoded bytes are not valid UTF-8 is still valid
+  // percent-encoding (RFC 3986 §2.1 is byte-level), so it is preserved as-is
+  // rather than being flagged as malformed
   components = fastURI.parse('http://example.com/#%E0%A4A')
-  t.equal(components.error, 'URI malformed', 'malformed fragment errors')
-  t.equal(components.fragment, '%E0%A4A', 'malformed fragment is preserved')
+  t.equal(components.error, undefined, 'valid percent-encoding is not flagged as malformed')
+  t.equal(components.fragment, '%E0%A4A', 'fragment is preserved')
 
   // all
   components = fastURI.parse('uri://user:pass@example.com:123/one/two.three?q1=a1&q2=a2#body')

package/test/query-fragment-normalization.test.js

@@ -0,0 +1,33 @@
+'use strict'
+
+const test = require('tape')
+const fastURI = require('..')
+
+test('query percent-encoding is normalized (RFC 3986 §6.2.2)', (t) => {
+  // hex digits are uppercased, like the path component already does
+  t.equal(fastURI.normalize('x://h/p?a=%2a'), 'x://h/p?a=%2A', 'uppercases hex in query')
+  // unreserved characters are decoded
+  t.equal(fastURI.normalize('x://h/p?a=%7e%2e'), 'x://h/p?a=~.', 'decodes unreserved in query')
+  // reserved characters stay percent-encoded
+  t.equal(fastURI.normalize('x://h/p?a=%2F'), 'x://h/p?a=%2F', 'keeps reserved encoded in query')
+  // raw characters that are not allowed are percent-encoded
+  t.equal(fastURI.normalize('x://h/p?a b'), 'x://h/p?a%20b', 'encodes a raw space in query')
+  t.equal(fastURI.normalize('x://h/p?café'), 'x://h/p?caf%C3%A9', 'encodes raw non-ASCII in query')
+  // `?` is allowed unencoded in a query
+  t.equal(fastURI.normalize('x://h/p?a?b'), 'x://h/p?a?b', 'keeps `?` raw in query')
+
+  t.end()
+})
+
+test('fragment percent-encoding is normalized without decoding reserved characters', (t) => {
+  // reserved characters must NOT be decoded — `%2F` is not `/`
+  t.equal(fastURI.normalize('x://h/p#a%2Fb%2A'), 'x://h/p#a%2Fb%2A', 'keeps reserved encoded in fragment')
+  // hex is uppercased and unreserved is decoded
+  t.equal(fastURI.normalize('x://h/p#a%2a%7e'), 'x://h/p#a%2A~', 'uppercases hex and decodes unreserved in fragment')
+  // raw characters are still encoded
+  t.equal(fastURI.normalize('x://h/p#a b'), 'x://h/p#a%20b', 'encodes a raw space in fragment')
+  // `?` and `/` are allowed unencoded in a fragment
+  t.equal(fastURI.normalize('x://h/p#f?x/y'), 'x://h/p#f?x/y', 'keeps `?` and `/` raw in fragment')
+
+  t.end()
+})

package/test/security.test.js

@@ -159,3 +159,65 @@ test('parse canonicalises IDN / Unicode hosts to their ASCII form', (t) => {
     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')
+})

package/test/util.test.js

@@ -29,6 +29,15 @@ test('removeDotSegments', (t) => {
   // https://github.com/fastify/fast-uri/issues/139
   testCases.push(['WS:/WS://1305G130505:1&%0D:1&C(XXXXX*)))))))XXX130505:UUVUaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa$aaaaaaaaaaaa13a',
     'WS:/WS://1305G130505:1&%0D:1&C(XXXXX*)))))))XXX130505:UUVUaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa$aaaaaaaaaaaa13a'])
+  // trailing empty segments must be preserved: RFC 3986 5.2.4 only removes
+  // complete "." / ".." segments, never an empty segment
+  testCases.push(['//', '//'])
+  testCases.push(['/foo//', '/foo//'])
+  testCases.push(['/a/b//', '/a/b//'])
+  testCases.push(['/x//y//z', '/x//y//z'])
+  // "." / ".." segments still collapse
+  testCases.push(['/a/b/c/./../../g', '/a/g'])
+  testCases.push(['/.', '/'])
 
   t.plan(testCases.length)