Block Tifinagh + Latin mix

components/url_formatter/idn_spoof_checker.cc

 // Copyright 2017 The Chromium Authors. All rights reserved.
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file.
 
 #include "components/url_formatter/idn_spoof_checker.h"
 
 #include "base/numerics/safe_conversions.h"
 #include "base/strings/string_split.h"
 #include "base/strings/string_util.h"
 #include "base/threading/thread_local_storage.h"
@@ skipping 149 matching lines @@
     //   Prolonged Sound) used out-of-context.
     // - Dislallow U+30FD/E (Katakana iteration mark/voiced iteration mark)
     //   unless they're preceded by a Katakana.
     // - Disallow three Hiragana letters (U+307[8-A]) or Katakana letters
     //   (U+30D[8-A]) that look exactly like each other when they're used in a
     //   label otherwise entirely in Katakna or Hiragana.
     // - Disallow U+0585 (Armenian Small Letter Oh) and U+0581 (Armenian Small
     //   Letter Co) to be next to Latin.
     // - Disallow Latin 'o' and 'g' next to Armenian.
     // - Disalow mixing of Latin and Canadian Syllabary.
+    // - Disalow mixing of Latin and Tifinagh.

[Peter Kasting, 2017/05/22 18:47:46]

Are there other "Latin + <script>" combos we should pre-emptively block?

     dangerous_pattern = new icu::RegexMatcher(
         icu::UnicodeString(
             R"([^\p{scx=kana}\p{scx=hira}\p{scx=hani}])"
             R"([\u30ce\u30f3\u30bd\u30be])"
             R"([^\p{scx=kana}\p{scx=hira}\p{scx=hani}]|)"
             R"([^\p{scx=kana}\p{scx=hira}]\u30fc|^\u30fc|)"
             R"([^\p{scx=kana}][\u30fd\u30fe]|^[\u30fd\u30fe]|)"
             R"(^[\p{scx=kana}]+[\u3078-\u307a][\p{scx=kana}]+$|)"
             R"(^[\p{scx=hira}]+[\u30d8-\u30da][\p{scx=hira}]+$|)"
             R"([a-z]\u30fb|\u30fb[a-z]|)"
             R"(^[\u0585\u0581]+[a-z]|[a-z][\u0585\u0581]+$|)"
             R"([a-z][\u0585\u0581]+[a-z]|)"
             R"(^[og]+[\p{scx=armn}]|[\p{scx=armn}][og]+$|)"
             R"([\p{scx=armn}][og]+[\p{scx=armn}]|)"
-            R"([\p{sc=cans}].*[a-z]|[a-z].*[\p{sc=cans}])",
+            R"([\p{sc=cans}].*[a-z]|[a-z].*[\p{sc=cans}]|)"
+            R"([\p{sc=tfng}].*[a-z]|[a-z].*[\p{sc=tfng}])",

[Peter Kasting, 2017/05/22 18:47:46]

Nit: Feels like there's maybe a less redundant way to write this regex, but it's
probably less readable.

             -1, US_INV),
         0, status);
     tls_index.Set(dangerous_pattern);
   }
   dangerous_pattern->reset(label_string);
   return !dangerous_pattern->find();
 }
 
 bool IDNSpoofChecker::IsMadeOfLatinAlikeCyrillic(
     const icu::UnicodeString& label) {
@@ skipping 92 matching lines @@
   allowed_set.remove(0x0F8Cu);
   allowed_set.remove(0x0F8Du);
   allowed_set.remove(0x0F8Eu);
   allowed_set.remove(0x0F8Fu);
 #endif
 
   uspoof_setAllowedUnicodeSet(checker_, &allowed_set, status);
 }
 
 }  // namespace url_formatter