Merge changes I0f1d59ae,I2f63bffb

* changes:
  fdtrack: GC before checking fd count.
  Move system_server fdtrack implementation to Java.

2 files changed, 75 insertions, 41 deletions

diff --git a/services/core/jni/com_android_server_SystemServer.cpp b/services/core/jni/com_android_server_SystemServer.cpp
index 43f50bfc33d5..729fa71af169 100644
--- a/services/core/jni/com_android_server_SystemServer.cpp
+++ b/services/core/jni/com_android_server_SystemServer.cpp
@@ -99,47 +99,17 @@ static void android_server_SystemServer_initZygoteChildHeapProfiling(JNIEnv* /*
     android_mallopt(M_INIT_ZYGOTE_CHILD_PROFILING, nullptr, 0);
 }
 
-static int get_current_max_fd() {
-    // Not actually guaranteed to be the max, but close enough for our purposes.
-    int fd = open("/dev/null", O_RDONLY | O_CLOEXEC);
-    LOG_ALWAYS_FATAL_IF(fd == -1, "failed to open /dev/null: %s", strerror(errno));
-    close(fd);
-    return fd;
-}
+static void android_server_SystemServer_fdtrackAbort(JNIEnv*, jobject) {
+    raise(BIONIC_SIGNAL_FDTRACK);
 
-static const char kFdLeakEnableThresholdProperty[] = "persist.sys.debug.fdtrack_enable_threshold";
-static const char kFdLeakAbortThresholdProperty[] = "persist.sys.debug.fdtrack_abort_threshold";
-static const char kFdLeakCheckIntervalProperty[] = "persist.sys.debug.fdtrack_interval";
+    // Wait for a bit to allow fdtrack to dump backtraces to logcat.
+    std::this_thread::sleep_for(5s);
 
-static void android_server_SystemServer_spawnFdLeakCheckThread(JNIEnv*, jobject) {
+    // Abort on a different thread to avoid ART dumping runtime stacks.
     std::thread([]() {
-        pthread_setname_np(pthread_self(), "FdLeakCheckThread");
-        bool loaded = false;
-        while (true) {
-            const int enable_threshold = GetIntProperty(kFdLeakEnableThresholdProperty, 1024);
-            const int abort_threshold = GetIntProperty(kFdLeakAbortThresholdProperty, 2048);
-            const int check_interval = GetIntProperty(kFdLeakCheckIntervalProperty, 120);
-            int max_fd = get_current_max_fd();
-            if (max_fd > enable_threshold && !loaded) {
-                loaded = true;
-                ALOGE("fd count above threshold of %d, starting fd backtraces", enable_threshold);
-                if (dlopen("libfdtrack.so", RTLD_GLOBAL) == nullptr) {
-                    ALOGE("failed to load libfdtrack.so: %s", dlerror());
-                }
-            } else if (max_fd > abort_threshold) {
-                raise(BIONIC_SIGNAL_FDTRACK);
-
-                // Wait for a bit to allow fdtrack to dump backtraces to logcat.
-                std::this_thread::sleep_for(5s);
-
-                LOG_ALWAYS_FATAL(
-                    "b/140703823: aborting due to fd leak: check logs for fd "
-                    "backtraces");
-            }
-
-            std::this_thread::sleep_for(std::chrono::seconds(check_interval));
-        }
-    }).detach();
+        LOG_ALWAYS_FATAL("b/140703823: aborting due to fd leak: check logs for fd "
+                         "backtraces");
+    }).join();
 }
 
 static jlong android_server_SystemServer_startIncrementalService(JNIEnv* env, jclass klass,
@@ -161,8 +131,7 @@ static const JNINativeMethod gMethods[] = {
         {"startHidlServices", "()V", (void*)android_server_SystemServer_startHidlServices},
         {"initZygoteChildHeapProfiling", "()V",
          (void*)android_server_SystemServer_initZygoteChildHeapProfiling},
-        {"spawnFdLeakCheckThread", "()V",
-         (void*)android_server_SystemServer_spawnFdLeakCheckThread},
+        {"fdtrackAbort", "()V", (void*)android_server_SystemServer_fdtrackAbort},
         {"startIncrementalService", "()J",
          (void*)android_server_SystemServer_startIncrementalService},
         {"setIncrementalServiceSystemReady", "(J)V",
diff --git a/services/java/com/android/server/SystemServer.java b/services/java/com/android/server/SystemServer.java
index 4cd1348a0f7a..516c64217177 100644
--- a/services/java/com/android/server/SystemServer.java
+++ b/services/java/com/android/server/SystemServer.java
@@ -23,6 +23,8 @@ import static android.os.IServiceManager.DUMP_FLAG_PRIORITY_NORMAL;
 import static android.os.IServiceManager.DUMP_FLAG_PROTO;
 import static android.os.Process.SYSTEM_UID;
 import static android.os.Process.myPid;
+import static android.system.OsConstants.O_CLOEXEC;
+import static android.system.OsConstants.O_RDONLY;
 import static android.view.Display.DEFAULT_DISPLAY;
 
 import static com.android.server.utils.TimingsTraceAndSlog.SYSTEM_SERVER_TIMING_TAG;
@@ -75,6 +77,8 @@ import android.provider.DeviceConfig;
 import android.provider.Settings;
 import android.server.ServerProtoEnums;
 import android.sysprop.VoldProperties;
+import android.system.ErrnoException;
+import android.system.Os;
 import android.text.TextUtils;
 import android.util.DisplayMetrics;
 import android.util.EventLog;
@@ -188,6 +192,7 @@ import dalvik.system.VMRuntime;
 import com.google.android.startop.iorap.IorapForwardingService;
 
 import java.io.File;
+import java.io.FileDescriptor;
 import java.io.IOException;
 import java.util.LinkedList;
 import java.util.Locale;
@@ -394,11 +399,71 @@ public final class SystemServer {
      */
     private static native void initZygoteChildHeapProfiling();
 
+    private static final String SYSPROP_FDTRACK_ENABLE_THRESHOLD =
+            "persist.sys.debug.fdtrack_enable_threshold";
+    private static final String SYSPROP_FDTRACK_ABORT_THRESHOLD =
+            "persist.sys.debug.fdtrack_abort_threshold";
+    private static final String SYSPROP_FDTRACK_INTERVAL =
+            "persist.sys.debug.fdtrack_interval";
+
+    private static int getMaxFd() {
+        FileDescriptor fd = null;
+        try {
+            fd = Os.open("/dev/null", O_RDONLY | O_CLOEXEC, 0);
+            return fd.getInt$();
+        } catch (ErrnoException ex) {
+            Slog.e("System", "Failed to get maximum fd: " + ex);
+        } finally {
+            if (fd != null) {
+                try {
+                    Os.close(fd);
+                } catch (ErrnoException ex) {
+                    // If Os.close threw, something went horribly wrong.
+                    throw new RuntimeException(ex);
+                }
+            }
+        }
+
+        return Integer.MAX_VALUE;
+    }
+
+    private static native void fdtrackAbort();
 
     /**
      * Spawn a thread that monitors for fd leaks.
      */
-    private static native void spawnFdLeakCheckThread();
+    private static void spawnFdLeakCheckThread() {
+        final int enableThreshold = SystemProperties.getInt(SYSPROP_FDTRACK_ENABLE_THRESHOLD, 1024);
+        final int abortThreshold = SystemProperties.getInt(SYSPROP_FDTRACK_ABORT_THRESHOLD, 2048);
+        final int checkInterval = SystemProperties.getInt(SYSPROP_FDTRACK_INTERVAL, 120);
+
+        new Thread(() -> {
+            boolean enabled = false;
+            while (true) {
+                int maxFd = getMaxFd();
+                if (maxFd > enableThreshold) {
+                    // Do a manual GC to clean up fds that are hanging around as garbage.
+                    System.gc();
+                    maxFd = getMaxFd();
+                }
+
+                if (maxFd > enableThreshold && !enabled) {
+                    Slog.i("System", "fdtrack enable threshold reached, enabling");
+                    System.loadLibrary("fdtrack");
+                    enabled = true;
+                } else if (maxFd > abortThreshold) {
+                    Slog.i("System", "fdtrack abort threshold reached, dumping and aborting");
+                    fdtrackAbort();
+                }
+
+                try {
+                    Thread.sleep(checkInterval);
+                } catch (InterruptedException ex) {
+                    continue;
+                }
+            }
+        }).start();
+    }
 
     /**
      * Start native Incremental Service and get its handle.