Files
send-mail/node_modules/nodemailer/dist/esm/shared/url.js
T

215 lines
10 KiB
JavaScript
Raw Normal View History

2026-09-15 13:00:32 +02:00
// URL parsing wrapper around the WHATWG `URL` class. It only falls back to the
// legacy, deprecation-warning-emitting `url.parse()` / `url.resolve()` for input
// the WHATWG parser rejects.
//
// The WHATWG `URL` exposes a different shape than the legacy parser, so results
// are normalized back into the legacy field names the rest of the codebase reads
// (`protocol`, `hostname`, `port`, `pathname`, `path`, `search`, `auth`, `query`,
// `href`). This keeps every existing call site unchanged.
//
// Known, accepted divergences from the legacy parser:
// - non-special schemes (smtp:/smtps:/direct:) are not host-lowercased by
// WHATWG; cosmetic only, SMTP/DNS hosts are case-insensitive. (IDNA mapping
// and IPv6 brackets are normalized back by normalizeHostname below.)
// - a literal unescaped ':' inside a password is percent-encoded by WHATWG;
// such passwords should be percent-encoded by the caller anyway.
import net from 'node:net';
import urllib from 'node:url';
import * as punycode from '../punycode/index.js';
// Matches a "scheme:" not followed by "//" (and with something after it), used
// to re-insert the authority separator the legacy parser did not require.
const SLASHLESS_AUTHORITY = /^([a-zA-Z][a-zA-Z0-9+.-]*:)(?!\/\/)([\s\S]+)$/;
// Leading and trailing C0 controls and spaces, which the WHATWG parser strips before it
// looks at the input. Stripped up front so that the slash-less form is recognized in a
// value read from a file with a trailing newline as well.
const SURROUNDING_WHITESPACE = /^[\x00-\x20]+|[\x00-\x20]+$/g;
// Leading characters legacy url.parse() skips before it reads the scheme
const LEGACY_TRIM = /^[\x00-\x20\u00a0\ufeff]+/;
// The authority of a "scheme://authority/..." or a scheme-relative "//authority/..."
// string: the scheme, if any (anything the legacy parser takes for one, it is less strict
// than WHATWG about the first character), and what the legacy parser has to report as the
// host once the userinfo is removed, see legacyParse. The legacy parser treats a backslash
// as a slash.
const AUTHORITY = /^([a-zA-Z0-9+.-]+:)?[\\/]{2}([^\\/?#]*)/;
// The WHATWG forbidden domain code points, except '%' which an opaque host may carry: the
// C0 control characters, space, DEL and the URL delimiters. CONTROL_CHARS is the C0 and DEL
// subset, checked on its own in legacyParse where the host string still carries the port
// and IPv6 brackets.
const FORBIDDEN_HOST_CHARS = /[\x00-\x20#/:<>?@[\\\]^|\x7f]/;
const CONTROL_CHARS = /[\x00-\x1f\x7f]/;
// The error the WHATWG parser throws, for a host that only fails the checks in this module
function invalidUrl(input) {
const err = new TypeError('Invalid URL');
err.code = 'ERR_INVALID_URL';
err.input = input;
return err;
}
// Legacy url.parse() for input the WHATWG parser refused. The legacy parser does not
// reject a host it can not represent: a NUL byte, a percent-encoded byte, a space or a
// '<' inside the host ends the host early and the rest becomes the path, so
// 'localhost%00.example.com' silently turns into a request to 'localhost'. When the
// input has an authority that the legacy parser reads as the host (always for
// "scheme://", for "//host" only when slashesDenoteHost is set, as url.resolve() does
// for its target), the legacy result is accepted only if that host is the whole written
// authority, lowercased and IDNA mapped the way the legacy parser does it, and carries
// no control character. A legacy result with a host (an empty one included, 'http:///x'
// is a request to localhost) that no such authority accounts for is refused as well.
// Otherwise the WHATWG error is reported. Relative input has no authority to check and
// keeps the legacy behavior.
function legacyParse(input, parseQueryString, whatwgError, slashesDenoteHost) {
const parsed = urllib.parse(input, parseQueryString, slashesDenoteHost);
const authority = AUTHORITY.exec(input.replace(LEGACY_TRIM, ''));
if (authority && (authority[1] || parsed.hostname !== null)) {
const written = authority[2].slice(authority[2].lastIndexOf('@') + 1);
if (!written || CONTROL_CHARS.test(written) || (parsed.host || '').toLowerCase() !== punycode.toASCII(written.toLowerCase())) {
throw whatwgError;
}
// the legacy parser takes any bracketed value for an IPv6 literal
if (written.charAt(0) === '[' && !net.isIPv6(written.slice(1, written.indexOf(']')))) {
throw whatwgError;
}
}
else if (parsed.hostname !== null) {
throw whatwgError;
}
// the legacy parser only offers the joined form, split it on the first colon
const legacyAuth = parsed.auth === null || parsed.auth === undefined ? null : parsed.auth.split(':');
const result = parsed;
result.username = legacyAuth ? legacyAuth.shift() : null;
result.password = legacyAuth && legacyAuth.length ? legacyAuth.join(':') : null;
return result;
}
// decodeURIComponent that never throws. Legacy url.parse() decodes the auth
// component but tolerates malformed percent sequences, so mirror that.
function safeDecode(str) {
try {
return decodeURIComponent(str);
}
catch (_err) {
return str;
}
}
// Derives the legacy-shaped bare hostname from a WHATWG URL. WHATWG keeps IPv6
// literals bracketed ('[::1]') and, for non-special schemes (smtp:/smtps:/socks:),
// percent-encodes a non-ASCII host instead of IDNA-mapping it. Both forms are
// un-resolvable when handed to net/dns/http.request, which is what every call
// site does, so map them back to what legacy url.parse() returned: the bare
// address and the IDNA mapped (lowercased, punycode) form. Idempotent on plain
// ASCII and already-punycode hosts, so special-scheme hosts (already IDNA-mapped
// by WHATWG) pass through.
function normalizeHostname(raw, href) {
const hostname = raw || '';
if (!hostname) {
// Host-less URL (e.g. 'direct:'): legacy returned '' here, not null;
// consumers do `hostname.length` / `'.' + hostname`, so keep it a string.
return '';
}
if (hostname.charAt(0) === '[' && hostname.charAt(hostname.length - 1) === ']') {
return hostname.slice(1, -1);
}
const decoded = safeDecode(hostname);
// domainToASCII applies the WHATWG host rules (IDNA mapping included) and returns an
// empty string for a host it refuses, the forbidden characters among them
const mapped = FORBIDDEN_HOST_CHARS.test(decoded) ? '' : urllib.domainToASCII(decoded);
if (!mapped) {
throw invalidUrl(href);
}
return mapped;
}
export const parse = (input, parseQueryString) => {
input = (input || '').replace(SURROUNDING_WHITESPACE, '');
// Legacy url.parse() parses a "user:pass@host:port" authority that follows
// the scheme even without the "//" separator, for schemes outside its
// built-in slashed-protocol list (smtp:/smtps:/socks:/...). The WHATWG
// parser instead treats a scheme not followed by "//" as an opaque path.
// Re-insert the "//" so slash-less connection/proxy URLs keep resolving to
// an authority, as they did before. This assumes a slash-authority scheme,
// which every consumer here uses (http/https/smtp/smtps/socks/direct); an
// opaque scheme like mailto:/data:/tel: would be mis-split, but none reach
// this module.
const slashless = SLASHLESS_AUTHORITY.exec(input);
const normalized = slashless ? slashless[1] + '//' + slashless[2] : input;
let u;
try {
u = new URL(normalized);
}
catch (err) {
// WHATWG rejects some input the legacy parser tolerated (empty/relative
// strings, scheme-relative '//host/path', out-of-range ports, ...). Fall
// back to the legacy parser so behavior, including the downstream errors
// callers rely on, is preserved. This is the only path that can still
// emit a deprecation warning; it fires for anything WHATWG cannot
// represent, including legitimate relative URLs, not just malformed input.
return legacyParse(normalized, parseQueryString, err);
}
const hostname = normalizeHostname(u.hostname, u.href);
const port = u.port || null;
const pathname = u.pathname || null;
const search = u.search || null;
// Legacy `.auth` is the decoded "user[:pass]" string; WHATWG keeps the
// username/password percent-encoded, so decode to stay byte-compatible with
// existing consumers (parseConnectionUrl, Basic/Proxy-Authorization headers).
let auth = null;
let username = null;
let password = null;
if (u.username || u.password) {
// Gate on password too: legacy url.parse('smtps://:pass@host').auth was
// ':pass'. Dropping it would silently connect unauthenticated.
username = safeDecode(u.username);
password = u.password ? safeDecode(u.password) : null;
// the joined form is ambiguous once the user name contains a colon, so
// consumers that need the parts read username and password instead
auth = username + (password !== null ? ':' + password : '');
}
let query;
if (parseQueryString) {
// Mirror querystring.parse(): null-prototype object, repeated keys become an array.
const parsed = Object.create(null);
u.searchParams.forEach((value, key) => {
if (Object.prototype.hasOwnProperty.call(parsed, key)) {
const existing = parsed[key];
if (Array.isArray(existing)) {
existing.push(value);
}
else {
parsed[key] = [existing, value];
}
}
else {
parsed[key] = value;
}
});
query = parsed;
}
else {
query = search ? search.slice(1) : null;
}
return {
protocol: u.protocol || null,
host: u.host || null,
hostname,
port,
pathname,
search,
path: (pathname || '') + (search || '') || null,
href: u.href,
auth,
username,
password,
query
};
};
export const resolve = (from, to) => {
try {
return new URL(to, from).href;
}
catch (err) {
// Malformed target, fall back to the legacy resolver, but only when the legacy
// parser reads the same host out of both inputs that was written. The target
// decides the host when it is absolute or scheme-relative, the base otherwise
legacyParse(from, false, err, true);
legacyParse(to, false, err, true);
return urllib.resolve(from, to);
}
};