When auto converting added TXT files with image references to TXTZ use a full markdown parser to detect markdown images

Fixes a DoS caused by an inefficient regex for detecting markdown's crazy inline image syntax.
Reference: GHSL-2021-112

src/calibre/customize/builtins.py

@@ -4,8 +4,7 @@
 __license__   = 'GPL v3'
 __copyright__ = '2008, Kovid Goyal <kovid at kovidgoyal.net>'
 
-import os, glob, re
-from calibre import guess_type
+import os, glob
 from calibre.customize import (FileTypePlugin, MetadataReaderPlugin,
     MetadataWriterPlugin, PreferencesPlugin, InterfaceActionBase, StoreBase)
 from calibre.constants import numeric_version
@@ -60,34 +59,8 @@ class TXT2TXTZ(FileTypePlugin):
     on_import = True
 
     def _get_image_references(self, txt, base_dir):
-        from calibre.ebooks.oeb.base import OEB_IMAGES
-
-        images = []
-
-        # Textile
-        for m in re.finditer(r'(?mu)(?:[\[{])?\!(?:\. )?(?P<path>[^\s(!]+)\s?(?:\(([^\)]+)\))?\!(?::(\S+))?(?:[\]}]|(?=\s|$))', txt):
-            path = m.group('path')
-            if path and not os.path.isabs(path) and guess_type(path)[0] in OEB_IMAGES and os.path.exists(os.path.join(base_dir, path)):
-                images.append(path)
-
-        # Markdown inline
-        for m in re.finditer(r'(?mu)\!\[([^\]\[]*(\[[^\]\[]*(\[[^\]\[]*(\[[^\]\[]*(\[[^\]\[]*(\[[^\]\[]*(\[[^\]\[]*\])*[^\]\[]*\])*[^\]\[]*\])*[^\]\[]*\])*[^\]\[]*\])*[^\]\[]*\])*[^\]\[]*)\]\s*\((?P<path>[^\)]*)\)', txt):  # noqa
-            path = m.group('path')
-            if path and not os.path.isabs(path) and guess_type(path)[0] in OEB_IMAGES and os.path.exists(os.path.join(base_dir, path)):
-                images.append(path)
-
-        # Markdown reference
-        refs = {}
-        for m in re.finditer(r'(?mu)^(\ ?\ ?\ ?)\[(?P<id>[^\]]*)\]:\s*(?P<path>[^\s]*)$', txt):
-            if m.group('id') and m.group('path'):
-                refs[m.group('id')] = m.group('path')
-        for m in re.finditer(r'(?mu)\!\[([^\]\[]*(\[[^\]\[]*(\[[^\]\[]*(\[[^\]\[]*(\[[^\]\[]*(\[[^\]\[]*(\[[^\]\[]*\])*[^\]\[]*\])*[^\]\[]*\])*[^\]\[]*\])*[^\]\[]*\])*[^\]\[]*\])*[^\]\[]*)\]\s*\[(?P<id>[^\]]*)\]', txt):  # noqa
-            path = refs.get(m.group('id'), None)
-            if path and not os.path.isabs(path) and guess_type(path)[0] in OEB_IMAGES and os.path.exists(os.path.join(base_dir, path)):
-                images.append(path)
-
-        # Remove duplicates
-        return list(set(images))
+        from calibre.ebooks.txt.processor import get_images_from_polyglot_text
+        return get_images_from_polyglot_text(txt, base_dir)
 
     def run(self, path_to_ebook):
         from calibre.ebooks.metadata.opf2 import metadata_to_opf

src/calibre/ebooks/txt/processor.py

@@ -341,3 +341,30 @@ def detect_formatting_type(txt):
             return 'textile'
 
     return 'heuristic'
+
+
+def get_images_from_polyglot_text(txt: str, base_dir: str = '') -> set:
+    from calibre.ebooks.oeb.base import OEB_IMAGES
+    from calibre import guess_type
+    if not base_dir:
+        base_dir = os.getcwd()
+    images = set()
+
+    def check_path(path: str) -> None:
+        if path and not os.path.isabs(path) and guess_type(path)[0] in OEB_IMAGES and os.path.exists(os.path.join(base_dir, path)):
+            images.add(path)
+
+    # Textile
+    for m in re.finditer(r'(?mu)(?:[\[{])?\!(?:\. )?(?P<path>[^\s(!]+)\s?(?:\(([^\)]+)\))?\!(?::(\S+))?(?:[\]}]|(?=\s|$))', txt):
+        path = m.group('path')
+        check_path(path)
+
+    # Markdown
+    from markdown import Markdown
+    html = HTML_TEMPLATE % ('', Markdown().convert(txt))
+    from html5_parser import parse
+    root = parse(html)
+    for img in root.iterdescendants('img'):
+        path = img.get('src')
+        check_path(path)
+    return images