lambdaisland/uri `authority-regex` returns the wrong authority

Summary

authority-regex allows an attacker to send malicious URLs to be parsed by the lambdaisland/uri and return the wrong authority. This issue is similar to CVE-2020-8910.

Details

https://github.com/lambdaisland/uri/blob/d3355fcd3e235238f4dcd37be97787a84e580072/src/lambdaisland/uri.cljc#L9

This regex doesn’t handle the backslash (\) character in the username correctly, leading to a wrong output.
Payload: https://example.com\\@google.com
The returned host is google.com, but the correct host should be example.com.

urllib3 (Python) and google-closure-library (Javascript) return example.com as the host. Here the correct (or current) regex used by google-closure-library:

https://github.com/google/closure-library/blob/0e567abedb058e9b194a40cfa3ad4c507653bccf/closure/goog/uri/utils.js#L189

PoC

(ns poc.core)
(require '[lambdaisland.uri :refer (uri)])

(def myurl "https://example.com\\@google.com")

(defn -main
  []
   (println myurl)
   (println (:host (uri myurl)))
)

Impact

The library returns the wrong authority, and it can be abused to bypass host restrictions.

Reference

WHATWG Living URL spec, section 4.4 URL Parsing, host state: https://url.spec.whatwg.org/#url-parsing