RPC Binder: increase transaction size
One of the RPC Binder usecases requires a larger transaction limit. This
change increases the limit, and it unifies the 'too large transaction'
logging from regular and RPC binder. This hopefully makes it more clear
why there is a limit there, we want to keep code compatible between the
two transports.
A test is added to show the current behavior. When a transaction which
is sent is too large, the server closes the session. This is probably
the correct behavior for too large replies, but for too large
transactions, the client could handle these errors.
b/392717039 is filed to investigate inconsistencies raised in the test
more deeply.
Fixes: 392575419
Test: atest binderRpcTest --test-filter="*LargeVector*"
Change-Id: I2eeb08818c10371c7f77a35abee7d4e46bb63d72
diff --git a/libs/binder/Constants.h b/libs/binder/Constants.h
new file mode 100644
index 0000000..b75493c
--- /dev/null
+++ b/libs/binder/Constants.h
@@ -0,0 +1,39 @@
+/*
+ * Copyright (C) 2025 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#pragma once
+
+namespace android::binder {
+
+/**
+ * See also BINDER_VM_SIZE. In kernel binder, the sum of all transactions must be allocated in this
+ * space. Large transactions are very error prone. In general, we should work to reduce this limit.
+ * The same limit is used in RPC binder for consistency.
+ */
+constexpr size_t kLogTransactionsOverBytes = 300 * 1024;
+
+/**
+ * See b/392575419 - this limit is chosen for a specific usecase, because RPC binder does not have
+ * support for shared memory in the Android Baklava timeframe. This was 100 KB during and before
+ * Android V.
+ *
+ * Keeping this low helps preserve overall system performance. Transactions of this size are far too
+ * expensive to make multiple copies over binder or sockets, and they should be avoided if at all
+ * possible and transition to shared memory.
+ */
+constexpr size_t kRpcTransactionLimitBytes = 600 * 1024;
+
+} // namespace android::binder