AIOHTTP's Unicode processing of header values could cause parsing discrepancies

@@ -238,7 +238,9 @@ def parse_headers(
 
 def _is_supported_upgrade(headers: CIMultiDictProxy[str]) -> bool:
     """Check if the upgrade header is supported."""
-    return headers.get(hdrs.UPGRADE, "").lower() in {"tcp", "websocket"}
+    u = headers.get(hdrs.UPGRADE, "")
+    # .lower() can transform non-ascii characters.
+    return u.isascii() and u.lower() in {"tcp", "websocket"}
 
 
 class HttpParser(abc.ABC, Generic[_MsgT]):
@@ -549,11 +551,9 @@ def parse_headers(
                 upgrade = True
 
         # encoding
-        enc = headers.get(hdrs.CONTENT_ENCODING)
-        if enc:
-            enc = enc.lower()
-            if enc in ("gzip", "deflate", "br", "zstd"):
-                encoding = enc
+        enc = headers.get(hdrs.CONTENT_ENCODING, "")
+        if enc.isascii() and enc.lower() in {"gzip", "deflate", "br", "zstd"}:
+            encoding = enc
 
         # chunking
         te = headers.get(hdrs.TRANSFER_ENCODING)
@@ -670,7 +670,9 @@ def parse_message(self, lines: List[bytes]) -> RawRequestMessage:
         )
 
     def _is_chunked_te(self, te: str) -> bool:
-        if te.rsplit(",", maxsplit=1)[-1].strip(" \t").lower() == "chunked":
+        te = te.rsplit(",", maxsplit=1)[-1].strip(" \t")
+        # .lower() transforms some non-ascii chars, so must check first.
+        if te.isascii() and te.lower() == "chunked":
             return True
         # https://www.rfc-editor.org/rfc/rfc9112#section-6.3-2.4.3
         raise BadHttpMessage("Request has invalid `Transfer-Encoding`")

@@ -419,7 +419,8 @@ cdef class HttpParser:
         headers = CIMultiDictProxy(CIMultiDict(self._headers))
 
         if self._cparser.type == cparser.HTTP_REQUEST:
-            allowed = upgrade and headers.get("upgrade", "").lower() in ALLOWED_UPGRADES
+            h_upg = headers.get("upgrade", "")
+            allowed = upgrade and h_upg.isascii() and h_upg.lower() in ALLOWED_UPGRADES
             if allowed or self._cparser.method == cparser.HTTP_CONNECT:
                 self._upgraded = True
         else:
@@ -434,8 +435,7 @@ cdef class HttpParser:
         enc = self._content_encoding
         if enc is not None:
             self._content_encoding = None
-            enc = enc.lower()
-            if enc in ('gzip', 'deflate', 'br', 'zstd'):
+            if enc.isascii() and enc.lower() in {"gzip", "deflate", "br", "zstd"}:
                 encoding = enc
 
         if self._cparser.type == cparser.HTTP_REQUEST:

@@ -476,7 +476,21 @@ def test_request_chunked(parser) -> None:
     assert isinstance(payload, streams.StreamReader)
 
 
-def test_request_te_chunked_with_content_length(parser: Any) -> None:
+def test_te_header_non_ascii(parser: HttpRequestParser) -> None:
+    # K = Kelvin sign, not valid ascii.
+    text = "GET /test HTTP/1.1\r\nTransfer-Encoding: chunKed\r\n\r\n"
+    with pytest.raises(http_exceptions.BadHttpMessage):
+        parser.feed_data(text.encode())
+
+
+def test_upgrade_header_non_ascii(parser: HttpRequestParser) -> None:
+    # K = Kelvin sign, not valid ascii.
+    text = "GET /test HTTP/1.1\r\nUpgrade: websocKet\r\n\r\n"
+    messages, upgrade, tail = parser.feed_data(text.encode())
+    assert not upgrade
+
+
+def test_request_te_chunked_with_content_length(parser: HttpRequestParser) -> None:
     text = (
         b"GET /test HTTP/1.1\r\n"
         b"content-length: 1234\r\n"
@@ -574,6 +588,21 @@ def test_compression_zstd(parser: HttpRequestParser) -> None:
     assert msg.compression == "zstd"
 
 
+@pytest.mark.parametrize(
+    "enc",
+    (
+        "zstd".encode(),  # "st".upper() == "ST"
+        "deflate".encode(),  # "fl".upper() == "FL"
+    ),
+)
+def test_compression_non_ascii(parser: HttpRequestParser, enc: bytes) -> None:
+    text = b"GET /test HTTP/1.1\r\ncontent-encoding: " + enc + b"\r\n\r\n"
+    messages, upgrade, tail = parser.feed_data(text)
+    msg = messages[0][0]
+    # Non-ascii input should not evaluate to a valid encoding scheme.
+    assert msg.compression is None
+
+
 def test_compression_unknown(parser: HttpRequestParser) -> None:
     text = b"GET /test HTTP/1.1\r\ncontent-encoding: compress\r\n\r\n"
     messages, upgrade, tail = parser.feed_data(text)