Merge pull request from GHSA-x345-32rc-8h85

* tests for push rule pattern matching

* tests for acl pattern matching

* factor out common `re.escape`

* Factor out common re.compile

* Factor out common anchoring code

* add word_boundary support to `glob_to_regex`

* Use `glob_to_regex` in push rule evaluator

NB that this drops support for character classes. I don't think anyone ever
used them.

* Improve efficiency of globs with multiple wildcards

The idea here is that we compress multiple `*` globs into a single `.*`. We
also need to consider `?`, since `*?*` is as hard to implement efficiently as
`**`.

* add assertion on regex pattern

* Fix mypy

* Simplify glob_to_regex

* Inline the glob_to_regex helper function

Signed-off-by: Dan Callahan <danc@element.io>

* Moar comments

Signed-off-by: Dan Callahan <danc@element.io>

Co-authored-by: Dan Callahan <danc@element.io>

1 files changed, 47 insertions, 14 deletions

diff --git a/synapse/util/__init__.py b/synapse/util/__init__.py

index 0f84fa3f4e..b69f562ca5 100644
--- a/synapse/util/__init__.py
+++ b/synapse/util/__init__.py
@@ -15,6 +15,7 @@
 import json
 import logging
 import re
+from typing import Pattern
 
 import attr
 from frozendict import frozendict
@@ -26,6 +27,9 @@ from synapse.logging import context
 logger = logging.getLogger(__name__)
 
 
+_WILDCARD_RUN = re.compile(r"([\?\*]+)")
+
+
 def _reject_invalid_json(val):
     """Do not allow Infinity, -Infinity, or NaN values in JSON."""
     raise ValueError("Invalid JSON value: '%s'" % val)
@@ -158,25 +162,54 @@ def log_failure(failure, msg, consumeErrors=True):
         return failure
 
 
-def glob_to_regex(glob):
+def glob_to_regex(glob: str, word_boundary: bool = False) -> Pattern:
     """Converts a glob to a compiled regex object.
 
-    The regex is anchored at the beginning and end of the string.
-
     Args:
-        glob (str)
+        glob: pattern to match
+        word_boundary: If True, the pattern will be allowed to match at word boundaries
+           anywhere in the string. Otherwise, the pattern is anchored at the start and
+           end of the string.
 
     Returns:
-        re.RegexObject
+        compiled regex pattern
     """
-    res = ""
-    for c in glob:
-        if c == "*":
-            res = res + ".*"
-        elif c == "?":
-            res = res + "."
+
+    # Patterns with wildcards must be simplified to avoid performance cliffs
+    # - The glob `?**?**?` is equivalent to the glob `???*`
+    # - The glob `???*` is equivalent to the regex `.{3,}`
+    chunks = []
+    for chunk in _WILDCARD_RUN.split(glob):
+        # No wildcards? re.escape()
+        if not _WILDCARD_RUN.match(chunk):
+            chunks.append(re.escape(chunk))
+            continue
+
+        # Wildcards? Simplify.
+        qmarks = chunk.count("?")
+        if "*" in chunk:
+            chunks.append(".{%d,}" % qmarks)
         else:
-            res = res + re.escape(c)
+            chunks.append(".{%d}" % qmarks)
+
+    res = "".join(chunks)
 
-    # \A anchors at start of string, \Z at end of string
-    return re.compile(r"\A" + res + r"\Z", re.IGNORECASE)
+    if word_boundary:
+        res = re_word_boundary(res)
+    else:
+        # \A anchors at start of string, \Z at end of string
+        res = r"\A" + res + r"\Z"
+
+    return re.compile(res, re.IGNORECASE)
+
+
+def re_word_boundary(r: str) -> str:
+    """
+    Adds word boundary characters to the start and end of an
+    expression to require that the match occur as a whole word,
+    but do so respecting the fact that strings starting or ending
+    with non-word characters will change word boundaries.
+    """
+    # we can't use \b as it chokes on unicode. however \W seems to be okay
+    # as shorthand for [^0-9A-Za-z_].
+    return r"(^|\W)%s(\W|$)" % (r,)