pypdf possibly has long runtimes for malformed FlateDecode streams

@@ -17,6 +17,9 @@ aware of the possible side effects, you can modify the following constants which
 For JBIG2 images, there is a similar parameter to limit the memory usage during decoding: `pypdf.filters.JBIG2_MAX_OUTPUT_LENGTH`
 It defaults to 75 MB as well.
 
+For the *FlateDecode* filter, the number of bytes to attempt recovery with can be set by `pypdf.filters.ZLIB_MAX_RECOVERY_INPUT_LENGTH`.
+It defaults to 5 MB due to the much more complex recovery approach.
+
 ### Reading
 
 *pypdf* currently employs the following reading limits on *PdfReader* instances:

@@ -74,7 +74,10 @@
 JBIG2_MAX_OUTPUT_LENGTH = 75_000_000
 LZW_MAX_OUTPUT_LENGTH = 75_000_000
 ZLIB_MAX_OUTPUT_LENGTH = 75_000_000
+ZLIB_MAX_RECOVERY_INPUT_LENGTH = 5_000_000
 
+# Reuse cached 1-byte values in the fallback loop to avoid per-byte allocations.
+_SINGLE_BYTES = tuple(bytes((i,)) for i in range(256))
 
 
 def _decompress_with_limit(data: bytes) -> bytes:
@@ -133,18 +136,23 @@ def decompress(data: bytes) -> bytes:
         decompressor = zlib.decompressobj(zlib.MAX_WBITS | 32)
         result_str = b""
         remaining_limit = ZLIB_MAX_OUTPUT_LENGTH
-        data_single_bytes = [data[i : i + 1] for i in range(len(data))]
+        data_length = len(data)
         known_errors = set()
-        for index, b in enumerate(data_single_bytes):
+        for index in range(data_length):
+            chunk = _SINGLE_BYTES[data[index]]
             try:
-                decompressed = decompressor.decompress(b, max_length=remaining_limit)
+                decompressed = decompressor.decompress(chunk, max_length=remaining_limit)
                 result_str += decompressed
                 remaining_limit -= len(decompressed)
                 if remaining_limit <= 0:
                     raise LimitReachedError(
-                        f"Limit reached while decompressing. {len(data_single_bytes) - index} bytes remaining."
+                        f"Limit reached while decompressing. {data_length - index} bytes remaining."
                     )
             except zlib.error as error:
+                if index > ZLIB_MAX_RECOVERY_INPUT_LENGTH:
+                    raise LimitReachedError(
+                        f"Recovery limit reached while decompressing. {data_length - index} bytes remaining."
+                    )
                 error_str = str(error)
                 if error_str in known_errors:
                     continue

@@ -876,11 +876,11 @@ def test_decompress():
     data = string.printable.encode("utf-8") + string.printable[::-1].encode("utf-8")
     compressed = FlateDecode.encode(data)
 
-    # # Decompress regularly.
+    # Decompress regularly.
     decompressed = decompress(compressed)
     assert decompressed == data
 
-    # # Decompress byte-wise.
+    # Decompress byte-wise.
     with mock.patch("pypdf.filters._decompress_with_limit", side_effect=zlib.error):
         decompressed = decompress(compressed)
         assert decompressed == data
@@ -889,10 +889,17 @@ def test_decompress():
     with mock.patch("pypdf.filters._decompress_with_limit", side_effect=zlib.error), \
             mock.patch("pypdf.filters.ZLIB_MAX_OUTPUT_LENGTH", len(compressed) - 13), \
             pytest.raises(
-                LimitReachedError, match=r"^Limit reached while decompressing\. 12 bytes remaining\."
+                LimitReachedError, match=r"^Limit reached while decompressing\. 12 bytes remaining\.$"
             ):
         decompress(compressed)
 
+    # Decompress byte-wise with input limit.
+    with mock.patch("pypdf.filters.ZLIB_MAX_RECOVERY_INPUT_LENGTH", 1000), \
+            pytest.raises(
+                LimitReachedError, match=r"^Recovery limit reached while decompressing\. 336 bytes remaining\.$"
+            ):
+        decompress(b"A" * 1337)
+
 
 def test_decompress__logging_on_invalid_data(caplog):
     """We do not like suddenly getting empty outputs for non-empty inputs without a warning."""