fast-uri 4.1.2 → 4.1.3

package/lib/schemes.js

@@ -1,10 +1,10 @@
 'use strict'
 
-const { isUUID } = require('./utils')
-const URN_REG = /([\da-z][\d\-a-z]{0,31}):((?:[\w!$'()*+,\-.:;=@]|%[\da-f]{2})+)/iu
+const { isUUID, BYTE_HEX, percentEncodeNonAscii, nonSimpleMailtoDomain } = require('./utils')
+const URN_REG = /^([\da-z][\d\-a-z]{0,31}):((?:[\w!$'()*+,\-./:;=@]|%[\da-f]{2})+)$/iu
 
 const supportedSchemeNames = /** @type {const} */ (['http', 'https', 'ws',
-  'wss', 'urn', 'urn:uuid'])
+  'wss', 'urn', 'urn:uuid', 'mailto'])
 
 /** @typedef {supportedSchemeNames[number]} SchemeName */
 
@@ -113,9 +113,14 @@ function wsSerialize (wsComponent) {
 
   // reconstruct path from resource name
   if (wsComponent.resourceName) {
-    const [path, query] = wsComponent.resourceName.split('?')
+    const queryIndex = wsComponent.resourceName.indexOf('?')
+    const path = queryIndex === -1
+      ? wsComponent.resourceName
+      : wsComponent.resourceName.slice(0, queryIndex)
     wsComponent.path = (path && path !== '/' ? path : undefined)
-    wsComponent.query = query
+    wsComponent.query = queryIndex === -1
+      ? undefined
+      : wsComponent.resourceName.slice(queryIndex + 1)
     wsComponent.resourceName = undefined
   }
 
@@ -132,7 +137,7 @@ function urnParse (urnComponent, options) {
     return urnComponent
   }
   const matches = urnComponent.path.match(URN_REG)
-  if (matches) {
+  if (matches && matches[0] === urnComponent.path) {
     const scheme = options.scheme || urnComponent.scheme || 'urn'
     urnComponent.nid = matches[1].toLowerCase()
     urnComponent.nss = matches[2]
@@ -235,13 +240,385 @@ const urnuuid = /** @type {SchemeHandler} */ ({
   skipNormalize: true
 })
 
+/**
+ * Allow-sets for `encodeWithAllow`, each a `{ all, table, allowNonAscii }`
+ * descriptor built by `allowSet`:
+ *
+ * - `LOCAL_PART` is the intersection of upstream uri-js's VCHAR and
+ *   NOT_PATH_NOSCHEME, minus ",": the serializer joins recipients with "," and
+ *   `mailtoParse` splits the path on it, so a comma left literal inside a local
+ *   part would silently become a recipient delimiter.
+ * - `HFNAME` is the qchar allow-set used for header names/values.
+ * - `DTEXT` is RFC 6068 dtext-no-obs restricted to characters that are also
+ *   safe in a URI path; `DTEXT_IRI` additionally permits non-ASCII for
+ *   `unicodeSupport`.
+ */
+const LOCAL_PART_ALLOWED = "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789-._~!$'()*+="
+const HFNAME_ALLOWED = "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789-._~!$'()*+,;:@"
+const DTEXT_ALLOWED = "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789-._~!$'()*+;:"
+
+/**
+ * @typedef {Object} AllowSet
+ * @property {RegExp} all - matches a whole string needing no encoding at all.
+ * @property {Uint8Array} table - per-ASCII-code allow lookup for the slow path.
+ * @property {boolean} allowNonAscii - pass non-ASCII code points through (IRI).
+ */
+
+/**
+ * `all` deliberately excludes "%" and (for the IRI variant) surrogates, so any
+ * input needing escape normalization or surrogate repair falls to the slow path.
+ *
+ * @param {string} chars
+ * @param {boolean} [allowNonAscii=false]
+ * @returns {AllowSet}
+ */
+function allowSet (chars, allowNonAscii = false) {
+  const table = new Uint8Array(128)
+  for (let i = 0; i < chars.length; i++) {
+    table[chars.charCodeAt(i)] = 1
+  }
+  const escaped = chars.replace(/[\\\]^-]/gu, '\\$&')
+  const cls = allowNonAscii ? `[${escaped}]|[^\\0-\\x7F\\uD800-\\uDFFF]` : `[${escaped}]`
+  return { all: new RegExp(`^(?:${cls})*$`, 'u'), table, allowNonAscii }
+}
+
+const LOCAL_PART = allowSet(LOCAL_PART_ALLOWED)
+const HFNAME = allowSet(HFNAME_ALLOWED)
+const DTEXT = allowSet(DTEXT_ALLOWED)
+const DTEXT_IRI = allowSet(DTEXT_ALLOWED, true)
+
+const HEX_PAIR = /^[\da-f]{2}$/iu
+const MAILTO_DOMAIN_LITERAL = /^\[[\x21-\x5A\x5E-\x7E]*\]$/u
+const MAILTO_DOMAIN_ERROR = 'URI mailto has an invalid recipient domain.'
+const HAS_SURROGATE = /[\uD800-\uDFFF]/u
+
+/**
+ * @param {string} str
+ * @returns {string}
+ */
+function decodeHex (str) {
+  if (typeof str !== 'string' || str.indexOf('%') === -1) {
+    return str
+  }
+  try {
+    return decodeURIComponent(str)
+  } catch {
+    return str
+  }
+}
+
+/**
+ * Replaces lone surrogates with U+FFFD, leaving valid pairs intact. Only called
+ * when `HAS_SURROGATE` matched, so the scan cost is not on the common path.
+ * (`String.prototype.toWellFormed` would do this natively but is Node 20+.)
+ *
+ * @param {string} input
+ * @returns {string}
+ */
+function replaceLoneSurrogates (input) {
+  let result = ''
+  for (let i = 0; i < input.length; i++) {
+    const code = input.charCodeAt(i)
+    if (code >= 0xD800 && code <= 0xDBFF && i + 1 < input.length) {
+      const low = input.charCodeAt(i + 1)
+      if (low >= 0xDC00 && low <= 0xDFFF) {
+        result += input[i] + input[i + 1]
+        i++
+        continue
+      }
+    }
+    result += (code >= 0xD800 && code <= 0xDFFF) ? '\uFFFD' : input[i]
+  }
+  return result
+}
+
+/**
+ * Percent-encodes everything outside `set`, preserving existing valid escapes
+ * (uppercased) and repairing lone surrogates.
+ *
+ * ASCII uses `BYTE_HEX` rather than `encodeURIComponent`. That is equivalent
+ * only because every character `encodeURIComponent` leaves unescaped
+ * (`A-Za-z0-9-_.!~*'()`) is present in all of the allow-sets above, so no such
+ * character ever reaches the encode branch. Re-check if a set is narrowed.
+ *
+ * @param {string} input
+ * @param {AllowSet} set
+ * @returns {string}
+ */
+function encodeWithAllow (input, set) {
+  if (set.all.test(input)) {
+    return input
+  }
+
+  const table = set.table
+  let result = ''
+  for (let i = 0; i < input.length; i++) {
+    const code = input.charCodeAt(i)
+
+    if (code < 0x80) {
+      if (table[code] === 1) {
+        result += input[i]
+      } else if (code === 0x25 && i + 2 < input.length && HEX_PAIR.test(input.slice(i + 1, i + 3))) {
+        result += '%' + input.slice(i + 1, i + 3).toUpperCase()
+        i += 2
+      } else {
+        result += BYTE_HEX[code]
+      }
+      continue
+    }
+
+    if (code < 0xD800 || code > 0xDFFF) {
+      result += set.allowNonAscii ? input[i] : percentEncodeNonAscii(code)
+      continue
+    }
+
+    if (code <= 0xDBFF && i + 1 < input.length) {
+      const low = input.charCodeAt(i + 1)
+      if (low >= 0xDC00 && low <= 0xDFFF) {
+        result += set.allowNonAscii
+          ? input[i] + input[i + 1]
+          : percentEncodeNonAscii(0x10000 + ((code - 0xD800) << 10) + (low - 0xDC00))
+        i++
+        continue
+      }
+    }
+
+    result += set.allowNonAscii ? '\uFFFD' : percentEncodeNonAscii(0xFFFD)
+  }
+  return result
+}
+
+/**
+ * @param {string} domain
+ * @param {import('../types/index').Options} [options]
+ * @param {{error?:string}} [component]
+ * @returns {string}
+ */
+function mailtoNormalizeDomain (domain, options, component) {
+  const normalizedDomain = String(domain).toLowerCase()
+
+  if (MAILTO_DOMAIN_LITERAL.test(normalizedDomain)) {
+    return normalizedDomain
+  }
+
+  // `hostname` is the identity for these, so skip constructing a WHATWG URL.
+  if (normalizedDomain !== '' && !nonSimpleMailtoDomain(normalizedDomain)) {
+    return normalizedDomain
+  }
+
+  try {
+    const parsedDomain = new URL('http://' + normalizedDomain)
+    if (
+      parsedDomain.username ||
+      parsedDomain.password ||
+      parsedDomain.port ||
+      parsedDomain.pathname !== '/' ||
+      parsedDomain.search ||
+      parsedDomain.hash ||
+      !parsedDomain.hostname
+    ) {
+      throw new Error(MAILTO_DOMAIN_ERROR)
+    }
+
+    return options && options.unicodeSupport
+      ? normalizedDomain
+      : parsedDomain.hostname
+  } catch {
+    if (component) {
+      component.error = component.error || MAILTO_DOMAIN_ERROR
+    }
+    return normalizedDomain
+  }
+}
+
+/**
+ * Percent-encodes everything in a recipient domain that is not `dtext-no-obs`
+ * (RFC 6068). Serialize-only: `mailtoNormalizeDomain` returns the domain
+ * verbatim when validation fails, and `MAILTO_DOMAIN_LITERAL` accepts the full
+ * RFC 5321 dcontent range (which includes "?", "#", "&" and "/"), so without
+ * this an attacker-supplied recipient could inject header fields or a fragment
+ * into the serialized URI. `mailtoParse` deliberately reports the raw domain
+ * instead, so it must not use this.
+ * @param {string} domain
+ * @param {import('../types/index').Options} [options]
+ * @returns {string}
+ */
+function mailtoEncodeDomain (domain, options) {
+  const set = options && options.unicodeSupport ? DTEXT_IRI : DTEXT
+
+  // Keep the delimiters of a domain literal, encode its contents.
+  if (domain.length > 1 && domain[0] === '[' && domain[domain.length - 1] === ']') {
+    return '[' + encodeWithAllow(domain.slice(1, -1), set) + ']'
+  }
+
+  return encodeWithAllow(domain, set)
+}
+
+/**
+ * @param {import('../types/index').URIComponent} component
+ * @param {import('../types/index').Options} options
+ * @returns {import('../types/index').URIComponent}
+ */
+function mailtoParse (component, options) {
+  const mailtoComponent = component
+
+  // The handler sets `skipNormalize`, so path and query arrive raw. Everything
+  // here goes through `decodeHex`, which subsumes the generic normalizers --
+  // except that they also fold lone surrogates to U+FFFD, so do that here.
+  let rawPath = mailtoComponent.path
+  let rawQuery = mailtoComponent.query
+  if (rawPath && HAS_SURROGATE.test(rawPath)) rawPath = replaceLoneSurrogates(rawPath)
+  if (rawQuery && HAS_SURROGATE.test(rawQuery)) rawQuery = replaceLoneSurrogates(rawQuery)
+
+  const to = rawPath
+    ? (rawPath.indexOf(',') === -1 ? [rawPath] : rawPath.split(','))
+    : []
+  mailtoComponent.path = undefined
+
+  if (rawQuery) {
+    // Null prototype: header names come from untrusted input, so `headers[name]`
+    // must not resolve to inherited members ("constructor", "toString", ...),
+    // and "__proto__" has to land as a plain own property instead of hitting
+    // Object.prototype's setter and being silently dropped. Allocated lazily so
+    // the common subject/body-only URI does not pay for it.
+    /** @type {Record<string,string>|null} */
+    let headers = null
+    let start = 0
+    while (start <= rawQuery.length) {
+      let end = rawQuery.indexOf('&', start)
+      if (end === -1) end = rawQuery.length
+      const token = rawQuery.slice(start, end)
+      start = end + 1
+
+      const eqIdx = token.indexOf('=')
+      if (eqIdx !== token.lastIndexOf('=')) {
+        mailtoComponent.error = mailtoComponent.error || 'URI mailto has malformed header fields.'
+        continue
+      }
+      const name = eqIdx === -1 ? token : token.slice(0, eqIdx)
+      const value = eqIdx === -1 ? '' : token.slice(eqIdx + 1)
+      if (name === 'to') {
+        const addrs = value.split(',')
+        for (let j = 0; j < addrs.length; j++) to.push(addrs[j])
+        continue
+      }
+      if (name === 'subject') {
+        mailtoComponent.subject = decodeHex(value)
+        continue
+      }
+      if (name === 'body') {
+        mailtoComponent.body = decodeHex(value)
+        continue
+      }
+      if (headers === null) headers = /** @type {Record<string,string>} */ (Object.create(null))
+      headers[decodeHex(name)] = decodeHex(value)
+    }
+    if (headers !== null) mailtoComponent.headers = headers
+  }
+
+  mailtoComponent.query = undefined
+
+  for (let i = 0; i < to.length; i++) {
+    const rawAddr = to[i]
+    const atIdx = rawAddr.lastIndexOf('@')
+    if (atIdx < 0) {
+      mailtoComponent.error = mailtoComponent.error || MAILTO_DOMAIN_ERROR
+      to[i] = decodeHex(rawAddr)
+      continue
+    }
+    const local = decodeHex(rawAddr.slice(0, atIdx))
+    const domain = mailtoNormalizeDomain(decodeHex(rawAddr.slice(atIdx + 1)), options, mailtoComponent)
+    to[i] = local + '@' + domain
+  }
+
+  if (to.length) mailtoComponent.to = to
+
+  return mailtoComponent
+}
+
+/**
+ * @param {import('../types/index').URIComponent} component
+ * @param {import('../types/index').Options} options
+ * @returns {import('../types/index').URIComponent}
+ */
+function mailtoSerialize (component, options) {
+  const mailtoComponent = component
+  const to = Array.isArray(mailtoComponent.to) ? mailtoComponent.to.slice() : []
+  if (to.length) {
+    for (let i = 0; i < to.length; i++) {
+      const addr = String(to[i])
+      const atIdx = addr.lastIndexOf('@')
+      const rawLocal = atIdx >= 0 ? addr.slice(0, atIdx) : addr
+      const rawDomain = atIdx >= 0 ? addr.slice(atIdx + 1) : ''
+      const local = encodeWithAllow(rawLocal, LOCAL_PART)
+      const decodedDomain = decodeHex(rawDomain)
+      // `component` is intentionally not passed: index.js's `serialize` writes
+      // the error onto a private copy and returns only a string, so an error
+      // recorded here would be unreachable. Safety comes from encoding instead.
+      const normalizedDomain = mailtoNormalizeDomain(decodedDomain, options)
+      to[i] = local + '@' + mailtoEncodeDomain(normalizedDomain, options)
+    }
+    // Every recipient is now fully encoded and only the joining commas are
+    // literal, so bypass index.js's generic encoder, whose allow-set is
+    // narrower than upstream's and would over-encode.
+    mailtoComponent.path = to.join(',')
+    options.skipEscape = true
+  } else {
+    // No recipients: drop any caller-supplied path, matching upstream uri-js.
+    // `to` is the only recipient source, and `mailtoParse` always moves the
+    // path into it, so a mailto component never legitimately carries one.
+    mailtoComponent.path = undefined
+  }
+
+  const headers = mailtoComponent.headers && typeof mailtoComponent.headers === 'object'
+    ? Object.assign(Object.create(null), mailtoComponent.headers)
+    : Object.create(null)
+
+  if (mailtoComponent.subject) headers.subject = mailtoComponent.subject
+  if (mailtoComponent.body) headers.body = mailtoComponent.body
+  mailtoComponent.headers = headers
+
+  // `headers` has a null prototype and was filled via Object.assign, so every
+  // enumerable key is an own property and no hasOwnProperty guard is needed.
+  let query = ''
+  let count = 0
+  for (const name in headers) {
+    if (count++ !== 0) query += '&'
+    query += encodeWithAllow(name, HFNAME) + '=' + encodeWithAllow(String(headers[name]), HFNAME)
+  }
+
+  if (count !== 0) {
+    mailtoComponent.query = query
+  } else {
+    mailtoComponent.headers = undefined
+  }
+
+  return mailtoComponent
+}
+
+const mailto = /** @type {SchemeHandler} */ ({
+  scheme: 'mailto',
+  parse: mailtoParse,
+  serialize: mailtoSerialize,
+  domainHost: false,
+  unicodeSupport: true,
+  // `mailtoParse` re-derives every component from the raw path/query via
+  // `decodeHex`, which subsumes the generic normalizers, so running them first
+  // is wasted work (~28% of parsing a URI with header fields).
+  skipNormalize: true,
+  // A recipient list has no dot segments to remove, and `removeDotSegments`
+  // would rewrite a "./"-prefixed local part.
+  absolutePath: true
+})
+
 const SCHEMES = /** @type {Record<SchemeName, SchemeHandler>} */ ({
   http,
   https,
   ws,
   wss,
   urn,
-  'urn:uuid': urnuuid
+  'urn:uuid': urnuuid,
+  mailto
 })
 
 Object.setPrototypeOf(SCHEMES, null)