Prevent bcrypt from raising a ValueError and log

changelog.d/19078.bugfix

@@ -0,0 +1 @@
+Fix a bug introduced in 1.140.0 where an internal server error could be raised when hashing user passwords that are too long.

synapse/_scripts/hash_password.py

@@ -73,8 +73,18 @@ def main() -> None:
 
     pw = unicodedata.normalize("NFKC", password)
 
+    bytes_to_hash = pw.encode("utf8") + password_pepper.encode("utf8")
+    if len(bytes_to_hash) > 72:
+        # bcrypt only looks at the first 72 bytes
+        print(
+            f"Password is too long ({len(bytes_to_hash)} bytes); truncating to 72 bytes for bcrypt. "
+            "This is expected behaviour and will not affect a user's ability to log in. 72 bytes is "
+            "sufficient entropy for a password."
+        )
+        bytes_to_hash = bytes_to_hash[:72]
+
     hashed = bcrypt.hashpw(
-        pw.encode("utf8") + password_pepper.encode("utf8"),
+        bytes_to_hash,
         bcrypt.gensalt(bcrypt_rounds),
     ).decode("ascii")
 

synapse/handlers/auth.py

@@ -1683,8 +1683,21 @@ def _do_hash() -> str:
             # Normalise the Unicode in the password
             pw = unicodedata.normalize("NFKC", password)
 
+            bytes_to_hash = pw.encode(
+                "utf8"
+            ) + self.hs.config.auth.password_pepper.encode("utf8")
+            if len(bytes_to_hash) > 72:
+                # bcrypt only looks at the first 72 bytes
+                logger.debug(
+                    "Password is too long (%d bytes); truncating to 72 bytes for bcrypt. "
+                    "This is expected behaviour and will not affect a user's ability to log in. 72 bytes is "
+                    "sufficient entropy for a password.",
+                    len(bytes_to_hash),
+                )
+                bytes_to_hash = bytes_to_hash[:72]
+
             return bcrypt.hashpw(
-                pw.encode("utf8") + self.hs.config.auth.password_pepper.encode("utf8"),
+                bytes_to_hash,
                 bcrypt.gensalt(self.bcrypt_rounds),
             ).decode("ascii")