diff --git a/debuggerd/libdebuggerd/test/tombstone_test.cpp b/debuggerd/libdebuggerd/test/tombstone_test.cpp
index b42d70c..7fe8f82 100644
--- a/debuggerd/libdebuggerd/test/tombstone_test.cpp
+++ b/debuggerd/libdebuggerd/test/tombstone_test.cpp
@@ -379,7 +379,7 @@
 TEST_F(TombstoneTest, gwp_asan_cause_uaf_exact) {
   gwp_asan::AllocationMetadata meta;
   meta.Addr = 0x1000;
-  meta.Size = 32;
+  meta.RequestedSize = 32;
 
   GwpAsanCrashDataTest crash_data(gwp_asan::Error::USE_AFTER_FREE, &meta);
   crash_data.SetCrashAddress(0x1000);
@@ -396,7 +396,7 @@
 TEST_F(TombstoneTest, gwp_asan_cause_double_free) {
   gwp_asan::AllocationMetadata meta;
   meta.Addr = 0x1000;
-  meta.Size = 32;
+  meta.RequestedSize = 32;
 
   GwpAsanCrashDataTest crash_data(gwp_asan::Error::DOUBLE_FREE, &meta);
   crash_data.SetCrashAddress(0x1000);
@@ -413,7 +413,7 @@
 TEST_F(TombstoneTest, gwp_asan_cause_overflow) {
   gwp_asan::AllocationMetadata meta;
   meta.Addr = 0x1000;
-  meta.Size = 32;
+  meta.RequestedSize = 32;
 
   GwpAsanCrashDataTest crash_data(gwp_asan::Error::BUFFER_OVERFLOW, &meta);
   crash_data.SetCrashAddress(0x1025);
@@ -432,7 +432,7 @@
 TEST_F(TombstoneTest, gwp_asan_cause_underflow) {
   gwp_asan::AllocationMetadata meta;
   meta.Addr = 0x1000;
-  meta.Size = 32;
+  meta.RequestedSize = 32;
 
   GwpAsanCrashDataTest crash_data(gwp_asan::Error::BUFFER_UNDERFLOW, &meta);
   crash_data.SetCrashAddress(0xffe);
@@ -451,7 +451,7 @@
 TEST_F(TombstoneTest, gwp_asan_cause_invalid_free_inside) {
   gwp_asan::AllocationMetadata meta;
   meta.Addr = 0x1000;
-  meta.Size = 32;
+  meta.RequestedSize = 32;
 
   GwpAsanCrashDataTest crash_data(gwp_asan::Error::INVALID_FREE, &meta);
   crash_data.SetCrashAddress(0x1001);
@@ -470,7 +470,7 @@
 TEST_F(TombstoneTest, gwp_asan_cause_invalid_free_outside) {
   gwp_asan::AllocationMetadata meta;
   meta.Addr = 0x1000;
-  meta.Size = 32;
+  meta.RequestedSize = 32;
 
   GwpAsanCrashDataTest crash_data(gwp_asan::Error::INVALID_FREE, &meta);
   crash_data.SetCrashAddress(0x1021);
diff --git a/debuggerd/proto/tombstone.proto b/debuggerd/proto/tombstone.proto
index 38a06f4..2c7156b 100644
--- a/debuggerd/proto/tombstone.proto
+++ b/debuggerd/proto/tombstone.proto
@@ -3,6 +3,9 @@
 option java_package = "com.android.server.os";
 option java_outer_classname = "TombstoneProtos";
 
+// NOTE TO OEMS:
+// If you add custom fields to this proto, do not use numbers in the reserved range.
+
 message Tombstone {
   Architecture arch = 1;
   string build_fingerprint = 2;
@@ -24,6 +27,8 @@
   repeated MemoryMapping memory_mappings = 17;
   repeated LogBuffer log_buffers = 18;
   repeated FD open_fds = 19;
+
+  reserved 20 to 999;
 }
 
 enum Architecture {
@@ -31,6 +36,8 @@
   ARM64 = 1;
   X86 = 2;
   X86_64 = 3;
+
+  reserved 4 to 999;
 }
 
 message Signal {
@@ -46,15 +53,21 @@
 
   bool has_fault_address = 8;
   uint64 fault_address = 9;
+
+  reserved 10 to 999;
 }
 
 message Cause {
   string human_readable = 1;
+
+  reserved 2 to 999;
 }
 
 message Register {
   string name = 1;
   uint64 u64 = 2;
+
+  reserved 3 to 999;
 }
 
 message Thread {
@@ -63,6 +76,8 @@
   repeated Register registers = 3;
   repeated BacktraceFrame current_backtrace = 4;
   repeated MemoryDump memory_dump = 5;
+
+  reserved 6 to 999;
 }
 
 message BacktraceFrame {
@@ -76,6 +91,8 @@
   string file_name = 6;
   uint64 file_map_offset = 7;
   string build_id = 8;
+
+  reserved 9 to 999;
 }
 
 message MemoryDump {
@@ -83,6 +100,8 @@
   string mapping_name = 2;
   uint64 begin_address = 3;
   bytes memory = 4;
+
+  reserved 5 to 999;
 }
 
 message MemoryMapping {
@@ -97,6 +116,8 @@
   string mapping_name = 7;
   string build_id = 8;
   uint64 load_bias = 9;
+
+  reserved 10 to 999;
 }
 
 message FD {
@@ -104,11 +125,15 @@
   string path = 2;
   string owner = 3;
   uint64 tag = 4;
+
+  reserved 5 to 999;
 }
 
 message LogBuffer {
   string name = 1;
   repeated LogMessage logs = 2;
+
+  reserved 3 to 999;
 }
 
 message LogMessage {
@@ -118,4 +143,6 @@
   uint32 priority = 4;
   string tag = 5;
   string message = 6;
+
+  reserved 7 to 999;
 }
diff --git a/fs_mgr/fs_mgr_overlayfs.cpp b/fs_mgr/fs_mgr_overlayfs.cpp
index 388c296..1134f14 100644
--- a/fs_mgr/fs_mgr_overlayfs.cpp
+++ b/fs_mgr/fs_mgr_overlayfs.cpp
@@ -125,10 +125,38 @@
 
 namespace {
 
+bool fs_mgr_in_recovery() {
+    // Check the existence of recovery binary instead of using the compile time
+    // macro, because first-stage-init is compiled with __ANDROID_RECOVERY__
+    // defined, albeit not in recovery. More details: system/core/init/README.md
+    return fs_mgr_access("/system/bin/recovery");
+}
+
+bool fs_mgr_is_dsu_running() {
+    // Since android::gsi::CanBootIntoGsi() or android::gsi::MarkSystemAsGsi() is
+    // never called in recovery, the return value of android::gsi::IsGsiRunning()
+    // is not well-defined. In this case, just return false as being in recovery
+    // implies not running a DSU system.
+    if (fs_mgr_in_recovery()) return false;
+    auto saved_errno = errno;
+    auto ret = android::gsi::IsGsiRunning();
+    errno = saved_errno;
+    return ret;
+}
+
 // list of acceptable overlayfs backing storage
 const auto kScratchMountPoint = "/mnt/scratch"s;
 const auto kCacheMountPoint = "/cache"s;
-const std::vector<const std::string> kOverlayMountPoints = {kScratchMountPoint, kCacheMountPoint};
+
+std::vector<const std::string> OverlayMountPoints() {
+    // Never fallback to legacy cache mount point if within a DSU system,
+    // because running a DSU system implies the device supports dynamic
+    // partitions, which means legacy cache mustn't be used.
+    if (fs_mgr_is_dsu_running()) {
+        return {kScratchMountPoint};
+    }
+    return {kScratchMountPoint, kCacheMountPoint};
+}
 
 // Return true if everything is mounted, but before adb is started.  Right
 // after 'trigger load_persist_props_action' is done.
@@ -166,26 +194,7 @@
     static constexpr unsigned long kSizeThreshold = 8 * 1024 * 1024;  // 8MB
 
     return (vst.f_bfree >= (vst.f_blocks * kPercentThreshold / 100)) &&
-           (vst.f_bfree * vst.f_bsize) >= kSizeThreshold;
-}
-
-bool fs_mgr_in_recovery() {
-    // Check the existence of recovery binary instead of using the compile time
-    // macro, because first-stage-init is compiled with __ANDROID_RECOVERY__
-    // defined, albeit not in recovery. More details: system/core/init/README.md
-    return fs_mgr_access("/system/bin/recovery");
-}
-
-bool fs_mgr_is_dsu_running() {
-    // Since android::gsi::CanBootIntoGsi() or android::gsi::MarkSystemAsGsi() is
-    // never called in recovery, the return value of android::gsi::IsGsiRunning()
-    // is not well-defined. In this case, just return false as being in recovery
-    // implies not running a DSU system.
-    if (fs_mgr_in_recovery()) return false;
-    auto saved_errno = errno;
-    auto ret = android::gsi::IsGsiRunning();
-    errno = saved_errno;
-    return ret;
+           (static_cast<uint64_t>(vst.f_bfree) * vst.f_frsize) >= kSizeThreshold;
 }
 
 const auto kPhysicalDevice = "/dev/block/by-name/"s;
@@ -300,7 +309,7 @@
 std::string fs_mgr_get_overlayfs_candidate(const std::string& mount_point) {
     if (!fs_mgr_is_dir(mount_point)) return "";
     const auto base = android::base::Basename(mount_point) + "/";
-    for (const auto& overlay_mount_point : kOverlayMountPoints) {
+    for (const auto& overlay_mount_point : OverlayMountPoints()) {
         auto dir = overlay_mount_point + kOverlayTopDir + "/" + base;
         auto upper = dir + kUpperName;
         if (!fs_mgr_is_dir(upper)) continue;
@@ -1344,7 +1353,7 @@
     if (candidates.empty()) return ret;
 
     std::string dir;
-    for (const auto& overlay_mount_point : kOverlayMountPoints) {
+    for (const auto& overlay_mount_point : OverlayMountPoints()) {
         if (backing && backing[0] && (overlay_mount_point != backing)) continue;
         if (overlay_mount_point == kScratchMountPoint) {
             if (!fs_mgr_overlayfs_setup_scratch(fstab, change)) continue;
@@ -1465,7 +1474,7 @@
         }
     }
     bool should_destroy_scratch = false;
-    for (const auto& overlay_mount_point : kOverlayMountPoints) {
+    for (const auto& overlay_mount_point : OverlayMountPoints()) {
         ret &= fs_mgr_overlayfs_teardown_one(
                 overlay_mount_point, mount_point ? fs_mgr_mount_point(mount_point) : "", change,
                 overlay_mount_point == kScratchMountPoint ? &should_destroy_scratch : nullptr);
@@ -1569,7 +1578,7 @@
     constexpr bool* ignore_change = nullptr;
 
     // Teardown legacy overlay mount points that's not backed by a scratch device.
-    for (const auto& overlay_mount_point : kOverlayMountPoints) {
+    for (const auto& overlay_mount_point : OverlayMountPoints()) {
         if (overlay_mount_point == kScratchMountPoint) {
             continue;
         }
diff --git a/fs_mgr/libfiemap/fiemap_writer.cpp b/fs_mgr/libfiemap/fiemap_writer.cpp
index 621031a..8acb885 100644
--- a/fs_mgr/libfiemap/fiemap_writer.cpp
+++ b/fs_mgr/libfiemap/fiemap_writer.cpp
@@ -52,7 +52,7 @@
 static constexpr const uint32_t kUnsupportedExtentFlags =
         FIEMAP_EXTENT_UNKNOWN | FIEMAP_EXTENT_UNWRITTEN | FIEMAP_EXTENT_DELALLOC |
         FIEMAP_EXTENT_NOT_ALIGNED | FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_DATA_TAIL |
-        FIEMAP_EXTENT_UNWRITTEN | FIEMAP_EXTENT_SHARED | FIEMAP_EXTENT_MERGED;
+        FIEMAP_EXTENT_UNWRITTEN | FIEMAP_EXTENT_SHARED;
 
 // Large file support must be enabled.
 static_assert(sizeof(off_t) == sizeof(uint64_t));
diff --git a/fs_mgr/libsnapshot/Android.bp b/fs_mgr/libsnapshot/Android.bp
index d36a7f0..678adf8 100644
--- a/fs_mgr/libsnapshot/Android.bp
+++ b/fs_mgr/libsnapshot/Android.bp
@@ -549,6 +549,7 @@
     ],
     srcs: [
         "cow_snapuserd_test.cpp",
+        "snapuserd.cpp",
     ],
     cflags: [
         "-Wall",
diff --git a/fs_mgr/libsnapshot/cow_reader.cpp b/fs_mgr/libsnapshot/cow_reader.cpp
index c15a05b..cf9f6ea 100644
--- a/fs_mgr/libsnapshot/cow_reader.cpp
+++ b/fs_mgr/libsnapshot/cow_reader.cpp
@@ -181,6 +181,7 @@
         ops_buffer->resize(current_op_num);
     }
 
+    LOG(DEBUG) << "COW file read complete. Total ops: " << ops_buffer->size();
     // To successfully parse a COW file, we need either:
     //  (1) a label to read up to, and for that label to be found, or
     //  (2) a valid footer.
@@ -236,6 +237,7 @@
                                [](CowOperation& op) { return IsMetadataOp(op); }),
                 ops_.get()->end());
 
+    set_total_data_ops(ops_->size());
     // We will re-arrange the vector in such a way that
     // kernel can batch merge. Ex:
     //
@@ -298,10 +300,9 @@
     // are contiguous. These are monotonically increasing numbers.
     //
     // When both (1) and (2) are true, kernel will batch merge the operations.
-    // However, we do not want copy operations to be batch merged as
-    // a crash or system reboot during an overlapping copy can drive the device
-    // to a corrupted state. Hence, merging of copy operations should always be
-    // done as a individual 4k block. In the above case, since the
+    // In the above case, we have to ensure that the copy operations
+    // are merged first before replace operations are done. Hence,
+    // we will not change the order of copy operations. Since,
     // cow_op->new_block numbers are contiguous, we will ensure that the
     // cow block numbers assigned in ReadMetadata() for these respective copy
     // operations are not contiguous forcing kernel to issue merge for each
@@ -328,10 +329,8 @@
     //
     // Merge sequence will look like:
     //
-    // Merge-1 - Copy-op-1
-    // Merge-2 - Copy-op-2
-    // Merge-3 - Copy-op-3
-    // Merge-4 - Batch-merge {Replace-op-7, Replace-op-6, Zero-op-8,
+    // Merge-1 - Batch-merge { Copy-op-1, Copy-op-2, Copy-op-3 }
+    // Merge-2 - Batch-merge {Replace-op-7, Replace-op-6, Zero-op-8,
     //                        Replace-op-4, Zero-op-9, Replace-op-5 }
     //==============================================================
 
diff --git a/fs_mgr/libsnapshot/cow_snapuserd_test.cpp b/fs_mgr/libsnapshot/cow_snapuserd_test.cpp
index 7fa23db..045d9db 100644
--- a/fs_mgr/libsnapshot/cow_snapuserd_test.cpp
+++ b/fs_mgr/libsnapshot/cow_snapuserd_test.cpp
@@ -36,6 +36,8 @@
 #include <libsnapshot/snapuserd_client.h>
 #include <storage_literals/storage_literals.h>
 
+#include "snapuserd.h"
+
 namespace android {
 namespace snapshot {
 
@@ -119,7 +121,6 @@
     void CreateDmUserDevice();
     void StartSnapuserdDaemon();
     void CreateSnapshotDevice();
-    unique_fd CreateTempFile(const std::string& name, size_t size);
 
     unique_ptr<LoopDevice> base_loop_;
     unique_ptr<TempDevice> dmuser_dev_;
@@ -140,7 +141,24 @@
     int total_base_size_;
 };
 
-unique_fd CowSnapuserdTest::CreateTempFile(const std::string& name, size_t size) {
+class CowSnapuserdMetadataTest final {
+  public:
+    void Setup();
+    void SetupPartialArea();
+    void ValidateMetadata();
+    void ValidatePartialFilledArea();
+
+  private:
+    void InitMetadata();
+    void CreateCowDevice();
+    void CreateCowPartialFilledArea();
+
+    std::unique_ptr<Snapuserd> snapuserd_;
+    std::unique_ptr<TemporaryFile> cow_system_;
+    size_t size_ = 1_MiB;
+};
+
+static unique_fd CreateTempFile(const std::string& name, size_t size) {
     unique_fd fd(syscall(__NR_memfd_create, name.c_str(), MFD_ALLOW_SEALING));
     if (fd < 0) {
         return {};
@@ -430,25 +448,299 @@
 }
 
 void CowSnapuserdTest::MergeInterrupt() {
+    // Interrupt merge at various intervals
     StartMerge();
-    std::this_thread::sleep_for(4s);
+    std::this_thread::sleep_for(250ms);
     SimulateDaemonRestart();
 
     StartMerge();
-    std::this_thread::sleep_for(3s);
+    std::this_thread::sleep_for(250ms);
     SimulateDaemonRestart();
 
     StartMerge();
-    std::this_thread::sleep_for(3s);
+    std::this_thread::sleep_for(150ms);
     SimulateDaemonRestart();
 
     StartMerge();
-    std::this_thread::sleep_for(1s);
+    std::this_thread::sleep_for(100ms);
+    SimulateDaemonRestart();
+
+    StartMerge();
+    std::this_thread::sleep_for(800ms);
+    SimulateDaemonRestart();
+
+    StartMerge();
+    std::this_thread::sleep_for(600ms);
     SimulateDaemonRestart();
 
     ASSERT_TRUE(Merge());
 }
 
+void CowSnapuserdMetadataTest::CreateCowPartialFilledArea() {
+    std::string path = android::base::GetExecutableDirectory();
+    cow_system_ = std::make_unique<TemporaryFile>(path);
+
+    CowOptions options;
+    options.compression = "gz";
+    CowWriter writer(options);
+
+    ASSERT_TRUE(writer.Initialize(cow_system_->fd));
+
+    // Area 0 is completely filled with 256 exceptions
+    for (int i = 0; i < 256; i++) {
+        ASSERT_TRUE(writer.AddCopy(i, 256 + i));
+    }
+
+    // Area 1 is partially filled with 2 copy ops and 10 zero ops
+    ASSERT_TRUE(writer.AddCopy(500, 1000));
+    ASSERT_TRUE(writer.AddCopy(501, 1001));
+
+    ASSERT_TRUE(writer.AddZeroBlocks(300, 10));
+
+    // Flush operations
+    ASSERT_TRUE(writer.Finalize());
+}
+
+void CowSnapuserdMetadataTest::ValidatePartialFilledArea() {
+    int area_sz = snapuserd_->GetMetadataAreaSize();
+
+    ASSERT_EQ(area_sz, 2);
+
+    size_t new_chunk = 263;
+    // Verify the partially filled area
+    void* buffer = snapuserd_->GetExceptionBuffer(1);
+    loff_t offset = 0;
+    struct disk_exception* de;
+    for (int i = 0; i < 12; i++) {
+        de = reinterpret_cast<struct disk_exception*>((char*)buffer + offset);
+        ASSERT_EQ(de->old_chunk, i);
+        ASSERT_EQ(de->new_chunk, new_chunk);
+        offset += sizeof(struct disk_exception);
+        new_chunk += 1;
+    }
+
+    de = reinterpret_cast<struct disk_exception*>((char*)buffer + offset);
+    ASSERT_EQ(de->old_chunk, 0);
+    ASSERT_EQ(de->new_chunk, 0);
+}
+
+void CowSnapuserdMetadataTest::SetupPartialArea() {
+    CreateCowPartialFilledArea();
+    InitMetadata();
+}
+
+void CowSnapuserdMetadataTest::CreateCowDevice() {
+    unique_fd rnd_fd;
+    loff_t offset = 0;
+
+    std::string path = android::base::GetExecutableDirectory();
+    cow_system_ = std::make_unique<TemporaryFile>(path);
+
+    rnd_fd.reset(open("/dev/random", O_RDONLY));
+    ASSERT_TRUE(rnd_fd > 0);
+
+    std::unique_ptr<uint8_t[]> random_buffer_1_ = std::make_unique<uint8_t[]>(size_);
+
+    // Fill random data
+    for (size_t j = 0; j < (size_ / 1_MiB); j++) {
+        ASSERT_EQ(ReadFullyAtOffset(rnd_fd, (char*)random_buffer_1_.get() + offset, 1_MiB, 0),
+                  true);
+
+        offset += 1_MiB;
+    }
+
+    CowOptions options;
+    options.compression = "gz";
+    CowWriter writer(options);
+
+    ASSERT_TRUE(writer.Initialize(cow_system_->fd));
+
+    size_t num_blocks = size_ / options.block_size;
+
+    // Overlapping region. This has to be split
+    // into two batch operations
+    ASSERT_TRUE(writer.AddCopy(23, 20));
+    ASSERT_TRUE(writer.AddCopy(22, 19));
+    ASSERT_TRUE(writer.AddCopy(21, 18));
+    ASSERT_TRUE(writer.AddCopy(20, 17));
+    ASSERT_TRUE(writer.AddCopy(19, 16));
+    ASSERT_TRUE(writer.AddCopy(18, 15));
+
+    // Contiguous region but blocks in ascending order
+    // Daemon has to ensure that these blocks are merged
+    // in a batch
+    ASSERT_TRUE(writer.AddCopy(50, 75));
+    ASSERT_TRUE(writer.AddCopy(51, 76));
+    ASSERT_TRUE(writer.AddCopy(52, 77));
+    ASSERT_TRUE(writer.AddCopy(53, 78));
+
+    // Dis-contiguous region
+    ASSERT_TRUE(writer.AddCopy(110, 130));
+    ASSERT_TRUE(writer.AddCopy(105, 125));
+    ASSERT_TRUE(writer.AddCopy(100, 120));
+
+    // Overlap
+    ASSERT_TRUE(writer.AddCopy(25, 30));
+    ASSERT_TRUE(writer.AddCopy(30, 31));
+
+    size_t source_blk = num_blocks;
+
+    ASSERT_TRUE(writer.AddRawBlocks(source_blk, random_buffer_1_.get(), size_));
+
+    size_t blk_zero_copy_start = source_blk + num_blocks;
+
+    ASSERT_TRUE(writer.AddZeroBlocks(blk_zero_copy_start, num_blocks));
+
+    // Flush operations
+    ASSERT_TRUE(writer.Finalize());
+}
+
+void CowSnapuserdMetadataTest::InitMetadata() {
+    snapuserd_ = std::make_unique<Snapuserd>("", cow_system_->path, "");
+    ASSERT_TRUE(snapuserd_->InitCowDevice());
+}
+
+void CowSnapuserdMetadataTest::Setup() {
+    CreateCowDevice();
+    InitMetadata();
+}
+
+void CowSnapuserdMetadataTest::ValidateMetadata() {
+    int area_sz = snapuserd_->GetMetadataAreaSize();
+    ASSERT_EQ(area_sz, 3);
+
+    size_t old_chunk;
+    size_t new_chunk;
+
+    for (int i = 0; i < area_sz; i++) {
+        void* buffer = snapuserd_->GetExceptionBuffer(i);
+        loff_t offset = 0;
+        if (i == 0) {
+            old_chunk = 256;
+            new_chunk = 2;
+        } else if (i == 1) {
+            old_chunk = 512;
+            new_chunk = 259;
+        }
+        for (int j = 0; j < 256; j++) {
+            struct disk_exception* de =
+                    reinterpret_cast<struct disk_exception*>((char*)buffer + offset);
+
+            if (i != 2) {
+                ASSERT_EQ(de->old_chunk, old_chunk);
+                ASSERT_EQ(de->new_chunk, new_chunk);
+                old_chunk += 1;
+                new_chunk += 1;
+            } else {
+                break;
+            }
+            offset += sizeof(struct disk_exception);
+        }
+
+        if (i == 2) {
+            // The first 5 copy operation is not batch merged
+            // as the sequence is discontiguous
+            struct disk_exception* de =
+                    reinterpret_cast<struct disk_exception*>((char*)buffer + offset);
+            ASSERT_EQ(de->old_chunk, 30);
+            ASSERT_EQ(de->new_chunk, 518);
+            offset += sizeof(struct disk_exception);
+
+            de = reinterpret_cast<struct disk_exception*>((char*)buffer + offset);
+            ASSERT_EQ(de->old_chunk, 25);
+            ASSERT_EQ(de->new_chunk, 520);
+            offset += sizeof(struct disk_exception);
+
+            de = reinterpret_cast<struct disk_exception*>((char*)buffer + offset);
+            ASSERT_EQ(de->old_chunk, 100);
+            ASSERT_EQ(de->new_chunk, 522);
+            offset += sizeof(struct disk_exception);
+
+            de = reinterpret_cast<struct disk_exception*>((char*)buffer + offset);
+            ASSERT_EQ(de->old_chunk, 105);
+            ASSERT_EQ(de->new_chunk, 524);
+            offset += sizeof(struct disk_exception);
+
+            de = reinterpret_cast<struct disk_exception*>((char*)buffer + offset);
+            ASSERT_EQ(de->old_chunk, 110);
+            ASSERT_EQ(de->new_chunk, 526);
+            offset += sizeof(struct disk_exception);
+
+            // The next 4 operations are batch merged as
+            // both old and new chunk are contiguous
+            de = reinterpret_cast<struct disk_exception*>((char*)buffer + offset);
+            ASSERT_EQ(de->old_chunk, 50);
+            ASSERT_EQ(de->new_chunk, 528);
+            offset += sizeof(struct disk_exception);
+
+            de = reinterpret_cast<struct disk_exception*>((char*)buffer + offset);
+            ASSERT_EQ(de->old_chunk, 51);
+            ASSERT_EQ(de->new_chunk, 529);
+            offset += sizeof(struct disk_exception);
+
+            de = reinterpret_cast<struct disk_exception*>((char*)buffer + offset);
+            ASSERT_EQ(de->old_chunk, 52);
+            ASSERT_EQ(de->new_chunk, 530);
+            offset += sizeof(struct disk_exception);
+
+            de = reinterpret_cast<struct disk_exception*>((char*)buffer + offset);
+            ASSERT_EQ(de->old_chunk, 53);
+            ASSERT_EQ(de->new_chunk, 531);
+            offset += sizeof(struct disk_exception);
+
+            // This is handling overlap operation with
+            // two batch merge operations.
+            de = reinterpret_cast<struct disk_exception*>((char*)buffer + offset);
+            ASSERT_EQ(de->old_chunk, 18);
+            ASSERT_EQ(de->new_chunk, 533);
+            offset += sizeof(struct disk_exception);
+
+            de = reinterpret_cast<struct disk_exception*>((char*)buffer + offset);
+            ASSERT_EQ(de->old_chunk, 19);
+            ASSERT_EQ(de->new_chunk, 534);
+            offset += sizeof(struct disk_exception);
+
+            de = reinterpret_cast<struct disk_exception*>((char*)buffer + offset);
+            ASSERT_EQ(de->old_chunk, 20);
+            ASSERT_EQ(de->new_chunk, 535);
+            offset += sizeof(struct disk_exception);
+
+            de = reinterpret_cast<struct disk_exception*>((char*)buffer + offset);
+            ASSERT_EQ(de->old_chunk, 21);
+            ASSERT_EQ(de->new_chunk, 537);
+            offset += sizeof(struct disk_exception);
+
+            de = reinterpret_cast<struct disk_exception*>((char*)buffer + offset);
+            ASSERT_EQ(de->old_chunk, 22);
+            ASSERT_EQ(de->new_chunk, 538);
+            offset += sizeof(struct disk_exception);
+
+            de = reinterpret_cast<struct disk_exception*>((char*)buffer + offset);
+            ASSERT_EQ(de->old_chunk, 23);
+            ASSERT_EQ(de->new_chunk, 539);
+            offset += sizeof(struct disk_exception);
+
+            // End of metadata
+            de = reinterpret_cast<struct disk_exception*>((char*)buffer + offset);
+            ASSERT_EQ(de->old_chunk, 0);
+            ASSERT_EQ(de->new_chunk, 0);
+            offset += sizeof(struct disk_exception);
+        }
+    }
+}
+
+TEST(Snapuserd_Test, Snapshot_Metadata) {
+    CowSnapuserdMetadataTest harness;
+    harness.Setup();
+    harness.ValidateMetadata();
+}
+
+TEST(Snapuserd_Test, Snapshot_Metadata_Overlap) {
+    CowSnapuserdMetadataTest harness;
+    harness.SetupPartialArea();
+    harness.ValidatePartialFilledArea();
+}
+
 TEST(Snapuserd_Test, Snapshot_Merge_Resume) {
     CowSnapuserdTest harness;
     ASSERT_TRUE(harness.Setup());
@@ -457,7 +749,7 @@
     harness.Shutdown();
 }
 
-TEST(Snapuserd_Test, Snapshot) {
+TEST(Snapuserd_Test, Snapshot_IO_TEST) {
     CowSnapuserdTest harness;
     ASSERT_TRUE(harness.Setup());
     harness.ReadSnapshotDeviceAndValidate();
@@ -465,7 +757,6 @@
     harness.ValidateMerge();
     harness.Shutdown();
 }
-
 }  // namespace snapshot
 }  // namespace android
 
diff --git a/fs_mgr/libsnapshot/cow_writer.cpp b/fs_mgr/libsnapshot/cow_writer.cpp
index c1a5f32..81edc79 100644
--- a/fs_mgr/libsnapshot/cow_writer.cpp
+++ b/fs_mgr/libsnapshot/cow_writer.cpp
@@ -491,7 +491,7 @@
     return true;
 }
 
-bool CowWriter::CommitMerge(int merged_ops, bool sync) {
+bool CowWriter::CommitMerge(int merged_ops) {
     CHECK(merge_in_progress_);
     header_.num_merge_ops += merged_ops;
 
@@ -506,11 +506,7 @@
         return false;
     }
 
-    // Sync only for merging of copy operations.
-    if (sync) {
-        return Sync();
-    }
-    return true;
+    return Sync();
 }
 
 bool CowWriter::Truncate(off_t length) {
diff --git a/fs_mgr/libsnapshot/include/libsnapshot/cow_reader.h b/fs_mgr/libsnapshot/include/libsnapshot/cow_reader.h
index 62b54f9..1de7473 100644
--- a/fs_mgr/libsnapshot/include/libsnapshot/cow_reader.h
+++ b/fs_mgr/libsnapshot/include/libsnapshot/cow_reader.h
@@ -142,6 +142,10 @@
 
     void InitializeMerge();
 
+    void set_total_data_ops(uint64_t size) { total_data_ops_ = size; }
+
+    uint64_t total_data_ops() { return total_data_ops_; }
+
   private:
     bool ParseOps(std::optional<uint64_t> label);
 
@@ -152,6 +156,7 @@
     uint64_t fd_size_;
     std::optional<uint64_t> last_label_;
     std::shared_ptr<std::vector<CowOperation>> ops_;
+    uint64_t total_data_ops_;
 };
 
 }  // namespace snapshot
diff --git a/fs_mgr/libsnapshot/include/libsnapshot/cow_writer.h b/fs_mgr/libsnapshot/include/libsnapshot/cow_writer.h
index 22ddfa6..6ffd5d8 100644
--- a/fs_mgr/libsnapshot/include/libsnapshot/cow_writer.h
+++ b/fs_mgr/libsnapshot/include/libsnapshot/cow_writer.h
@@ -101,7 +101,7 @@
     bool InitializeAppend(android::base::borrowed_fd fd, uint64_t label);
 
     void InitializeMerge(android::base::borrowed_fd fd, CowHeader* header);
-    bool CommitMerge(int merged_ops, bool sync);
+    bool CommitMerge(int merged_ops);
 
     bool Finalize() override;
 
diff --git a/fs_mgr/libsnapshot/include/libsnapshot/snapuserd_kernel.h b/fs_mgr/libsnapshot/include/libsnapshot/snapuserd_kernel.h
index 7941e68..2b6c8ef 100644
--- a/fs_mgr/libsnapshot/include/libsnapshot/snapuserd_kernel.h
+++ b/fs_mgr/libsnapshot/include/libsnapshot/snapuserd_kernel.h
@@ -47,8 +47,8 @@
 static constexpr uint32_t CHUNK_SIZE = 8;
 static constexpr uint32_t CHUNK_SHIFT = (__builtin_ffs(CHUNK_SIZE) - 1);
 
-static constexpr uint32_t BLOCK_SIZE = 4096;
-static constexpr uint32_t BLOCK_SHIFT = (__builtin_ffs(BLOCK_SIZE) - 1);
+static constexpr uint32_t BLOCK_SZ = 4096;
+static constexpr uint32_t BLOCK_SHIFT = (__builtin_ffs(BLOCK_SZ) - 1);
 
 #define DIV_ROUND_UP(n, d) (((n) + (d)-1) / (d))
 
diff --git a/fs_mgr/libsnapshot/snapshot.cpp b/fs_mgr/libsnapshot/snapshot.cpp
index 90225f8..eb3a501 100644
--- a/fs_mgr/libsnapshot/snapshot.cpp
+++ b/fs_mgr/libsnapshot/snapshot.cpp
@@ -1234,6 +1234,25 @@
     return true;
 }
 
+static bool DeleteDmDevice(const std::string& name, const std::chrono::milliseconds& timeout_ms) {
+    auto start = std::chrono::steady_clock::now();
+    auto& dm = DeviceMapper::Instance();
+    while (true) {
+        if (dm.DeleteDeviceIfExists(name)) {
+            break;
+        }
+        auto now = std::chrono::steady_clock::now();
+        auto elapsed = std::chrono::duration_cast<std::chrono::milliseconds>(now - start);
+        if (elapsed >= timeout_ms) {
+            LOG(ERROR) << "DeleteDevice timeout: " << name;
+            return false;
+        }
+        std::this_thread::sleep_for(250ms);
+    }
+
+    return true;
+}
+
 bool SnapshotManager::CollapseSnapshotDevice(const std::string& name,
                                              const SnapshotStatus& status) {
     auto& dm = DeviceMapper::Instance();
@@ -1292,10 +1311,11 @@
     if (!dm.DeleteDeviceIfExists(base_name)) {
         LOG(ERROR) << "Unable to delete base device for snapshot: " << base_name;
     }
-    auto source_name = GetSourceDeviceName(name);
-    if (!dm.DeleteDeviceIfExists(source_name)) {
-        LOG(ERROR) << "Unable to delete source device for snapshot: " << source_name;
+
+    if (!DeleteDmDevice(GetSourceDeviceName(name), 4000ms)) {
+        LOG(ERROR) << "Unable to delete source device for snapshot: " << GetSourceDeviceName(name);
     }
+
     return true;
 }
 
@@ -1387,9 +1407,6 @@
         }
 
         auto misc_name = user_cow_name;
-        if (transition == InitTransition::SELINUX_DETACH) {
-            misc_name += "-selinux";
-        }
 
         DmTable table;
         table.Emplace<DmTargetUser>(0, target.spec.length, misc_name);
@@ -1422,6 +1439,7 @@
         // Wait for ueventd to acknowledge and create the control device node.
         std::string control_device = "/dev/dm-user/" + misc_name;
         if (!WaitForDevice(control_device, 10s)) {
+            LOG(ERROR) << "dm-user control device no found:  " << misc_name;
             continue;
         }
 
@@ -2122,15 +2140,12 @@
     CHECK(lock);
     if (!EnsureImageManager()) return false;
 
-    auto& dm = DeviceMapper::Instance();
-
     if (UpdateUsesCompression(lock) && !UnmapDmUserDevice(name)) {
         return false;
     }
 
-    auto cow_name = GetCowName(name);
-    if (!dm.DeleteDeviceIfExists(cow_name)) {
-        LOG(ERROR) << "Cannot unmap " << cow_name;
+    if (!DeleteDmDevice(GetCowName(name), 4000ms)) {
+        LOG(ERROR) << "Cannot unmap: " << GetCowName(name);
         return false;
     }
 
@@ -2155,12 +2170,11 @@
         return false;
     }
 
-    if (!EnsureSnapuserdConnected()) {
-        return false;
-    }
-    if (!snapuserd_client_->WaitForDeviceDelete(dm_user_name)) {
-        LOG(ERROR) << "Failed to wait for " << dm_user_name << " control device to delete";
-        return false;
+    if (EnsureSnapuserdConnected()) {
+        if (!snapuserd_client_->WaitForDeviceDelete(dm_user_name)) {
+            LOG(ERROR) << "Failed to wait for " << dm_user_name << " control device to delete";
+            return false;
+        }
     }
 
     // Ensure the control device is gone so we don't run into ABA problems.
diff --git a/fs_mgr/libsnapshot/snapuserd.cpp b/fs_mgr/libsnapshot/snapuserd.cpp
index 82db0d3..d620300 100644
--- a/fs_mgr/libsnapshot/snapuserd.cpp
+++ b/fs_mgr/libsnapshot/snapuserd.cpp
@@ -17,6 +17,8 @@
 #include "snapuserd.h"
 
 #include <csignal>
+#include <optional>
+#include <set>
 
 #include <libsnapshot/snapuserd_client.h>
 
@@ -32,7 +34,7 @@
 
 static constexpr size_t PAYLOAD_SIZE = (1UL << 20);
 
-static_assert(PAYLOAD_SIZE >= BLOCK_SIZE);
+static_assert(PAYLOAD_SIZE >= BLOCK_SZ);
 
 void BufferSink::Initialize(size_t size) {
     buffer_size_ = size;
@@ -78,10 +80,10 @@
 // request will always be 4k. After constructing
 // the header, zero out the remaining block.
 void Snapuserd::ConstructKernelCowHeader() {
-    void* buffer = bufsink_.GetPayloadBuffer(BLOCK_SIZE);
+    void* buffer = bufsink_.GetPayloadBuffer(BLOCK_SZ);
     CHECK(buffer != nullptr);
 
-    memset(buffer, 0, BLOCK_SIZE);
+    memset(buffer, 0, BLOCK_SZ);
 
     struct disk_header* dh = reinterpret_cast<struct disk_header*>(buffer);
 
@@ -106,13 +108,13 @@
 // Start the copy operation. This will read the backing
 // block device which is represented by cow_op->source.
 bool Snapuserd::ProcessCopyOp(const CowOperation* cow_op) {
-    void* buffer = bufsink_.GetPayloadBuffer(BLOCK_SIZE);
+    void* buffer = bufsink_.GetPayloadBuffer(BLOCK_SZ);
     CHECK(buffer != nullptr);
 
     // Issue a single 4K IO. However, this can be optimized
     // if the successive blocks are contiguous.
-    if (!android::base::ReadFullyAtOffset(backing_store_fd_, buffer, BLOCK_SIZE,
-                                          cow_op->source * BLOCK_SIZE)) {
+    if (!android::base::ReadFullyAtOffset(backing_store_fd_, buffer, BLOCK_SZ,
+                                          cow_op->source * BLOCK_SZ)) {
         SNAP_PLOG(ERROR) << "Copy-op failed. Read from backing store: " << backing_store_device_
                          << "at block :" << cow_op->source;
         return false;
@@ -123,10 +125,10 @@
 
 bool Snapuserd::ProcessZeroOp() {
     // Zero out the entire block
-    void* buffer = bufsink_.GetPayloadBuffer(BLOCK_SIZE);
+    void* buffer = bufsink_.GetPayloadBuffer(BLOCK_SZ);
     CHECK(buffer != nullptr);
 
-    memset(buffer, 0, BLOCK_SIZE);
+    memset(buffer, 0, BLOCK_SZ);
     return true;
 }
 
@@ -173,11 +175,11 @@
         struct dm_user_message* msg = (struct dm_user_message*)(&(buffer[0]));
 
         memmove(msg->payload.buf, (char*)msg->payload.buf + skip_sector_size,
-                (BLOCK_SIZE - skip_sector_size));
+                (BLOCK_SZ - skip_sector_size));
     }
 
     bufsink_.ResetBufferOffset();
-    return std::min(size, (BLOCK_SIZE - skip_sector_size));
+    return std::min(size, (BLOCK_SZ - skip_sector_size));
 }
 
 /*
@@ -234,7 +236,7 @@
         return ReadUnalignedSector(sector, size, it);
     }
 
-    int num_ops = DIV_ROUND_UP(size, BLOCK_SIZE);
+    int num_ops = DIV_ROUND_UP(size, BLOCK_SZ);
     while (num_ops) {
         if (!ProcessCowOp(it->second)) {
             return -1;
@@ -242,7 +244,7 @@
         num_ops -= 1;
         it++;
         // Update the buffer offset
-        bufsink_.UpdateBufferOffset(BLOCK_SIZE);
+        bufsink_.UpdateBufferOffset(BLOCK_SZ);
 
         SNAP_LOG(DEBUG) << "ReadData at sector: " << sector << " size: " << size;
     }
@@ -344,7 +346,7 @@
 }
 
 int Snapuserd::GetNumberOfMergedOps(void* merged_buffer, void* unmerged_buffer, loff_t offset,
-                                    int unmerged_exceptions, bool* copy_op) {
+                                    int unmerged_exceptions) {
     int merged_ops_cur_iter = 0;
 
     // Find the operations which are merged in this cycle.
@@ -362,10 +364,8 @@
             offset += sizeof(struct disk_exception);
             const CowOperation* cow_op = chunk_map_[ChunkToSector(cow_de->new_chunk)];
             CHECK(cow_op != nullptr);
+
             CHECK(cow_op->new_block == cow_de->old_chunk);
-            if (cow_op->type == kCowCopyOp) {
-                *copy_op = true;
-            }
             // zero out to indicate that operation is merged.
             cow_de->old_chunk = 0;
             cow_de->new_chunk = 0;
@@ -389,10 +389,6 @@
             return -1;
         }
     }
-
-    if (*copy_op) {
-        CHECK(merged_ops_cur_iter == 1);
-    }
     return merged_ops_cur_iter;
 }
 
@@ -414,47 +410,21 @@
     int unmerged_exceptions = 0;
     loff_t offset = GetMergeStartOffset(buffer, vec_[divresult.quot].get(), &unmerged_exceptions);
 
-    bool copy_op = false;
-    // Check if the merged operation is a copy operation. If so, then we need
-    // to explicitly sync the metadata before initiating the next merge.
-    // For ex: Consider a following sequence of copy operations in the COW file:
-    //
-    // Op-1: Copy 2 -> 3
-    // Op-2: Copy 1 -> 2
-    // Op-3: Copy 5 -> 10
-    //
-    // Op-1 and Op-2 are overlapping copy operations. The merge sequence will
-    // look like:
-    //
-    // Merge op-1: Copy 2 -> 3
-    // Merge op-2: Copy 1 -> 2
-    // Merge op-3: Copy 5 -> 10
-    //
-    // Now, let's say we have a crash _after_ Merge op-2; Block 2 contents would
-    // have been over-written by Block-1 after merge op-2. During next reboot,
-    // kernel will request the metadata for all the un-merged blocks. If we had
-    // not sync the metadata after Merge-op 1 and Merge op-2, snapuser daemon
-    // will think that these merge operations are still pending and hence will
-    // inform the kernel that Op-1 and Op-2 are un-merged blocks. When kernel
-    // resumes back the merging process, it will attempt to redo the Merge op-1
-    // once again. However, block 2 contents are wrong as it has the contents
-    // of block 1 from previous merge cycle. Although, merge will silently succeed,
-    // this will lead to silent data corruption.
-    //
-    int merged_ops_cur_iter = GetNumberOfMergedOps(buffer, vec_[divresult.quot].get(), offset,
-                                                   unmerged_exceptions, &copy_op);
+    int merged_ops_cur_iter =
+            GetNumberOfMergedOps(buffer, vec_[divresult.quot].get(), offset, unmerged_exceptions);
 
     // There should be at least one operation merged in this cycle
     CHECK(merged_ops_cur_iter > 0);
 
     header.num_merge_ops += merged_ops_cur_iter;
     reader_->UpdateMergeProgress(merged_ops_cur_iter);
-    if (!writer_->CommitMerge(merged_ops_cur_iter, copy_op)) {
+    if (!writer_->CommitMerge(merged_ops_cur_iter)) {
         SNAP_LOG(ERROR) << "CommitMerge failed... merged_ops_cur_iter: " << merged_ops_cur_iter;
         return false;
     }
 
     SNAP_LOG(DEBUG) << "Merge success: " << merged_ops_cur_iter << "chunk: " << chunk;
+    merge_initiated_ = true;
     return true;
 }
 
@@ -476,6 +446,19 @@
     return next_chunk;
 }
 
+void Snapuserd::CheckMergeCompletionStatus() {
+    CowHeader header;
+
+    if (merge_initiated_) {
+        reader_->GetHeader(&header);
+        SNAP_LOG(INFO) << "Merge-status: Total-Merged-ops: " << header.num_merge_ops
+                       << " Total-data-ops: " << reader_->total_data_ops();
+    } else {
+        SNAP_LOG(INFO) << "Merge was not initiated. Total-Merged-ops: " << header.num_merge_ops
+                       << " Total-data-ops: " << reader_->total_data_ops();
+    }
+}
+
 /*
  * Read the metadata from COW device and
  * construct the metadata as required by the kernel.
@@ -506,12 +489,13 @@
  *      during merge; specifically when the merge operation has dependency.
  *      These dependencies can only happen during copy operations.
  *
- *      To avoid this problem, we make sure that no two copy-operations
- *      do not have contiguous chunk IDs. Additionally, we make sure
- *      that each copy operation is merged individually.
+ *      To avoid this problem, we make sure overlap copy operations
+ *      are not batch merged.
  * 6: Use a monotonically increasing chunk number to assign the
  *    new_chunk
- * 7: Each chunk-id represents either a: Metadata page or b: Data page
+ * 7: Each chunk-id represents either
+ *        a: Metadata page or
+ *        b: Data page
  * 8: Chunk-id representing a data page is stored in a map.
  * 9: Chunk-id representing a metadata page is converted into a vector
  *    index. We store this in vector as kernel requests metadata during
@@ -531,10 +515,10 @@
     reader_ = std::make_unique<CowReader>();
     CowHeader header;
     CowOptions options;
-    bool prev_copy_op = false;
     bool metadata_found = false;
+    int replace_ops = 0, zero_ops = 0, copy_ops = 0;
 
-    SNAP_LOG(DEBUG) << "ReadMetadata Start...";
+    SNAP_LOG(DEBUG) << "ReadMetadata: Parsing cow file";
 
     if (!reader_->Parse(cow_fd_)) {
         SNAP_LOG(ERROR) << "Failed to parse";
@@ -546,10 +530,10 @@
         return false;
     }
 
-    CHECK(header.block_size == BLOCK_SIZE);
+    CHECK(header.block_size == BLOCK_SZ);
 
-    SNAP_LOG(DEBUG) << "Merge-ops: " << header.num_merge_ops;
     reader_->InitializeMerge();
+    SNAP_LOG(DEBUG) << "Merge-ops: " << header.num_merge_ops;
 
     writer_ = std::make_unique<CowWriter>(options);
     writer_->InitializeMerge(cow_fd_.get(), &header);
@@ -584,17 +568,26 @@
         }
 
         metadata_found = true;
-        if ((cow_op->type == kCowCopyOp || prev_copy_op)) {
+        // This loop will handle all the replace and zero ops.
+        // We will handle the copy ops later as it requires special
+        // handling of assigning chunk-id's. Furthermore, we make
+        // sure that replace/zero and copy ops are not batch merged; hence,
+        // the bump in the chunk_id before break of this loop
+        if (cow_op->type == kCowCopyOp) {
             data_chunk_id = GetNextAllocatableChunkId(data_chunk_id);
+            break;
         }
 
-        prev_copy_op = (cow_op->type == kCowCopyOp);
+        if (cow_op->type == kCowReplaceOp) {
+            replace_ops++;
+        } else if (cow_op->type == kCowZeroOp) {
+            zero_ops++;
+        }
 
         // Construct the disk-exception
         de->old_chunk = cow_op->new_block;
         de->new_chunk = data_chunk_id;
 
-        SNAP_LOG(DEBUG) << "Old-chunk: " << de->old_chunk << "New-chunk: " << de->new_chunk;
 
         // Store operation pointer.
         chunk_map_[ChunkToSector(data_chunk_id)] = cow_op;
@@ -602,6 +595,9 @@
         offset += sizeof(struct disk_exception);
         cowop_riter_->Next();
 
+        SNAP_LOG(DEBUG) << num_ops << ":"
+                        << " Old-chunk: " << de->old_chunk << " New-chunk: " << de->new_chunk;
+
         if (num_ops == exceptions_per_area_) {
             // Store it in vector at the right index. This maps the chunk-id to
             // vector index.
@@ -616,13 +612,213 @@
 
             if (cowop_riter_->Done()) {
                 vec_.push_back(std::move(de_ptr));
-                SNAP_LOG(DEBUG) << "ReadMetadata() completed; Number of Areas: " << vec_.size();
             }
         }
 
         data_chunk_id = GetNextAllocatableChunkId(data_chunk_id);
     }
 
+    std::optional<chunk_t> prev_id = {};
+    std::map<uint64_t, const CowOperation*> map;
+    std::set<uint64_t> dest_blocks;
+    size_t pending_copy_ops = exceptions_per_area_ - num_ops;
+    SNAP_LOG(INFO) << " Processing copy-ops at Area: " << vec_.size()
+                   << " Number of replace/zero ops completed in this area: " << num_ops
+                   << " Pending copy ops for this area: " << pending_copy_ops;
+    while (!cowop_riter_->Done()) {
+        do {
+            const CowOperation* cow_op = &cowop_riter_->Get();
+            if (IsMetadataOp(*cow_op)) {
+                cowop_riter_->Next();
+                continue;
+            }
+
+            // We have two cases specific cases:
+            //
+            // =====================================================
+            // Case 1: Overlapping copy regions
+            //
+            // Ex:
+            //
+            // Source -> Destination
+            //
+            // 1: 15 -> 18
+            // 2: 16 -> 19
+            // 3: 17 -> 20
+            // 4: 18 -> 21
+            // 5: 19 -> 22
+            // 6: 20 -> 23
+            //
+            // We have 6 copy operations to be executed in OTA and there is a overlap. Update-engine
+            // will write to COW file as follows:
+            //
+            // Op-1: 20 -> 23
+            // Op-2: 19 -> 22
+            // Op-3: 18 -> 21
+            // Op-4: 17 -> 20
+            // Op-5: 16 -> 19
+            // Op-6: 15 -> 18
+            //
+            // Note that the blocks numbers are contiguous. Hence, all 6 copy
+            // operations can potentially be batch merged. However, that will be
+            // problematic if we have a crash as block 20, 19, 18 would have
+            // been overwritten and hence subsequent recovery may end up with
+            // a silent data corruption when op-1, op-2 and op-3 are
+            // re-executed.
+            //
+            // We will split these 6 operations into two batches viz:
+            //
+            // Batch-1:
+            // ===================
+            // Op-1: 20 -> 23
+            // Op-2: 19 -> 22
+            // Op-3: 18 -> 21
+            // ===================
+            //
+            // Batch-2:
+            // ==================
+            // Op-4: 17 -> 20
+            // Op-5: 16 -> 19
+            // Op-6: 15 -> 18
+            // ==================
+            //
+            // Now, merge sequence will look like:
+            //
+            // 1: Merge Batch-1 { op-1, op-2, op-3 }
+            // 2: Update Metadata in COW File that op-1, op-2, op-3 merge is
+            // done.
+            // 3: Merge Batch-2
+            // 4: Update Metadata in COW File that op-4, op-5, op-6 merge is
+            // done.
+            //
+            // Note, that the order of block operations are still the same.
+            // However, we have two batch merge operations. Any crash between
+            // either of this sequence should be safe as each of these
+            // batches are self-contained.
+            //
+            //===========================================================
+            //
+            // Case 2:
+            //
+            // Let's say we have three copy operations written to COW file
+            // in the following order:
+            //
+            // op-1: 15 -> 18
+            // op-2: 16 -> 19
+            // op-3: 17 -> 20
+            //
+            // As aforementioned, kernel will initiate merge in reverse order.
+            // Hence, we will read these ops in reverse order so that all these
+            // ops are exectued in the same order as requested. Thus, we will
+            // read the metadata in reverse order and for the kernel it will
+            // look like:
+            //
+            // op-3: 17 -> 20
+            // op-2: 16 -> 19
+            // op-1: 15 -> 18   <-- Merge starts here in the kernel
+            //
+            // Now, this is problematic as kernel cannot batch merge them.
+            //
+            // Merge sequence will look like:
+            //
+            // Merge-1: op-1: 15 -> 18
+            // Merge-2: op-2: 16 -> 19
+            // Merge-3: op-3: 17 -> 20
+            //
+            // We have three merge operations.
+            //
+            // Even though the blocks are contiguous, kernel can batch merge
+            // them if the blocks are in descending order. Update engine
+            // addresses this issue partially for overlapping operations as
+            // we see that op-1 to op-3 and op-4 to op-6 operatiosn are in
+            // descending order. However, if the copy operations are not
+            // overlapping, update engine cannot write these blocks
+            // in descending order. Hence, we will try to address it.
+            // Thus, we will send these blocks to the kernel and it will
+            // look like:
+            //
+            // op-3: 15 -> 18
+            // op-2: 16 -> 19
+            // op-1: 17 -> 20  <-- Merge starts here in the kernel
+            //
+            // Now with this change, we can batch merge all these three
+            // operations. Merge sequence will look like:
+            //
+            // Merge-1: {op-1: 17 -> 20, op-2: 16 -> 19, op-3: 15 -> 18}
+            //
+            // Note that we have changed the ordering of merge; However, this
+            // is ok as each of these copy operations are independent and there
+            // is no overlap.
+            //
+            //===================================================================
+            if (prev_id.has_value()) {
+                chunk_t diff = (cow_op->new_block > prev_id.value())
+                                       ? (cow_op->new_block - prev_id.value())
+                                       : (prev_id.value() - cow_op->new_block);
+                if (diff != 1) {
+                    break;
+                }
+                if (dest_blocks.count(cow_op->new_block) || map.count(cow_op->source) > 0) {
+                    break;
+                }
+            }
+            metadata_found = true;
+            pending_copy_ops -= 1;
+            map[cow_op->new_block] = cow_op;
+            dest_blocks.insert(cow_op->source);
+            prev_id = cow_op->new_block;
+            cowop_riter_->Next();
+        } while (!cowop_riter_->Done() && pending_copy_ops);
+
+        data_chunk_id = GetNextAllocatableChunkId(data_chunk_id);
+        SNAP_LOG(DEBUG) << "Batch Merge copy-ops of size: " << map.size()
+                        << " Area: " << vec_.size() << " Area offset: " << offset
+                        << " Pending-copy-ops in this area: " << pending_copy_ops;
+
+        for (auto it = map.begin(); it != map.end(); it++) {
+            struct disk_exception* de =
+                    reinterpret_cast<struct disk_exception*>((char*)de_ptr.get() + offset);
+            de->old_chunk = it->first;
+            de->new_chunk = data_chunk_id;
+
+            // Store operation pointer.
+            chunk_map_[ChunkToSector(data_chunk_id)] = it->second;
+            offset += sizeof(struct disk_exception);
+            num_ops += 1;
+            copy_ops++;
+
+            SNAP_LOG(DEBUG) << num_ops << ":"
+                            << " Copy-op: "
+                            << " Old-chunk: " << de->old_chunk << " New-chunk: " << de->new_chunk;
+
+            if (num_ops == exceptions_per_area_) {
+                // Store it in vector at the right index. This maps the chunk-id to
+                // vector index.
+                vec_.push_back(std::move(de_ptr));
+                num_ops = 0;
+                offset = 0;
+
+                // Create buffer for next area
+                de_ptr = std::make_unique<uint8_t[]>(exceptions_per_area_ *
+                                                     sizeof(struct disk_exception));
+                memset(de_ptr.get(), 0, (exceptions_per_area_ * sizeof(struct disk_exception)));
+
+                if (cowop_riter_->Done()) {
+                    vec_.push_back(std::move(de_ptr));
+                    SNAP_LOG(DEBUG) << "ReadMetadata() completed; Number of Areas: " << vec_.size();
+                }
+
+                CHECK(pending_copy_ops == 0);
+                pending_copy_ops = exceptions_per_area_;
+            }
+
+            data_chunk_id = GetNextAllocatableChunkId(data_chunk_id);
+        }
+        map.clear();
+        dest_blocks.clear();
+        prev_id.reset();
+    }
+
     // Partially filled area or there is no metadata
     // If there is no metadata, fill with zero so that kernel
     // is aware that merge is completed.
@@ -632,12 +828,17 @@
                         << "Areas : " << vec_.size();
     }
 
-    SNAP_LOG(DEBUG) << "ReadMetadata() completed. Final_chunk_id: " << data_chunk_id
-                    << "Num Sector: " << ChunkToSector(data_chunk_id);
+    SNAP_LOG(INFO) << "ReadMetadata completed. Final-chunk-id: " << data_chunk_id
+                   << " Num Sector: " << ChunkToSector(data_chunk_id)
+                   << " Replace-ops: " << replace_ops << " Zero-ops: " << zero_ops
+                   << " Copy-ops: " << copy_ops << " Areas: " << vec_.size()
+                   << " Num-ops-merged: " << header.num_merge_ops
+                   << " Total-data-ops: " << reader_->total_data_ops();
 
     // Total number of sectors required for creating dm-user device
     num_sectors_ = ChunkToSector(data_chunk_id);
     metadata_read_done_ = true;
+    merge_initiated_ = false;
     return true;
 }
 
@@ -742,7 +943,7 @@
 
     size_t remaining_size = header->len;
     size_t read_size = std::min(PAYLOAD_SIZE, remaining_size);
-    CHECK(read_size == BLOCK_SIZE);
+    CHECK(read_size == BLOCK_SZ);
 
     CHECK(header->sector > 0);
     chunk_t chunk = SectorToChunk(header->sector);
@@ -793,11 +994,11 @@
         // will always be a single 4k.
         if (header->sector == 0) {
             CHECK(metadata_read_done_ == true);
-            CHECK(read_size == BLOCK_SIZE);
+            CHECK(read_size == BLOCK_SZ);
             ConstructKernelCowHeader();
             SNAP_LOG(DEBUG) << "Kernel header constructed";
         } else {
-            if (!offset && (read_size == BLOCK_SIZE) &&
+            if (!offset && (read_size == BLOCK_SZ) &&
                 chunk_map_.find(header->sector) == chunk_map_.end()) {
                 if (!ReadDiskExceptions(chunk, read_size)) {
                     SNAP_LOG(ERROR) << "ReadDiskExceptions failed for chunk id: " << chunk
diff --git a/fs_mgr/libsnapshot/snapuserd.h b/fs_mgr/libsnapshot/snapuserd.h
index c01fee3..518d08b 100644
--- a/fs_mgr/libsnapshot/snapuserd.h
+++ b/fs_mgr/libsnapshot/snapuserd.h
@@ -70,6 +70,14 @@
     const std::string& GetMiscName() { return misc_name_; }
     uint64_t GetNumSectors() { return num_sectors_; }
     bool IsAttached() const { return ctrl_fd_ >= 0; }
+    void CheckMergeCompletionStatus();
+    void CloseFds() {
+        ctrl_fd_ = {};
+        cow_fd_ = {};
+        backing_store_fd_ = {};
+    }
+    size_t GetMetadataAreaSize() { return vec_.size(); }
+    void* GetExceptionBuffer(size_t i) { return vec_[i].get(); }
 
   private:
     bool DmuserReadRequest();
@@ -96,11 +104,11 @@
     loff_t GetMergeStartOffset(void* merged_buffer, void* unmerged_buffer,
                                int* unmerged_exceptions);
     int GetNumberOfMergedOps(void* merged_buffer, void* unmerged_buffer, loff_t offset,
-                             int unmerged_exceptions, bool* copy_op);
+                             int unmerged_exceptions);
     bool ProcessMergeComplete(chunk_t chunk, void* buffer);
     sector_t ChunkToSector(chunk_t chunk) { return chunk << CHUNK_SHIFT; }
     chunk_t SectorToChunk(sector_t sector) { return sector >> CHUNK_SHIFT; }
-    bool IsBlockAligned(int read_size) { return ((read_size & (BLOCK_SIZE - 1)) == 0); }
+    bool IsBlockAligned(int read_size) { return ((read_size & (BLOCK_SZ - 1)) == 0); }
 
     std::string cow_device_;
     std::string backing_store_device_;
@@ -134,6 +142,7 @@
     std::map<sector_t, const CowOperation*> chunk_map_;
 
     bool metadata_read_done_ = false;
+    bool merge_initiated_ = false;
     BufferSink bufsink_;
 };
 
diff --git a/fs_mgr/libsnapshot/snapuserd_server.cpp b/fs_mgr/libsnapshot/snapuserd_server.cpp
index 38abaec..017de3b 100644
--- a/fs_mgr/libsnapshot/snapuserd_server.cpp
+++ b/fs_mgr/libsnapshot/snapuserd_server.cpp
@@ -210,8 +210,11 @@
         }
     }
 
+    handler->snapuserd()->CloseFds();
+
     auto misc_name = handler->misc_name();
     LOG(INFO) << "Handler thread about to exit: " << misc_name;
+    handler->snapuserd()->CheckMergeCompletionStatus();
 
     {
         std::lock_guard<std::mutex> lock(lock_);
diff --git a/fs_mgr/tests/adb-remount-test.sh b/fs_mgr/tests/adb-remount-test.sh
index ec73253..242fa93 100755
--- a/fs_mgr/tests/adb-remount-test.sh
+++ b/fs_mgr/tests/adb-remount-test.sh
@@ -881,10 +881,10 @@
 [ -z "${D}" -o -n "${ANDROID_SERIAL}" ] || ANDROID_SERIAL=${D}
 USB_SERIAL=
 if [ -n "${ANDROID_SERIAL}" -a "Darwin" != "${HOSTOS}" ]; then
-  USB_SERIAL="`find /sys/devices -name serial | grep usb`"
+  USB_SERIAL="`find /sys/devices -name serial | grep usb || true`"
   if [ -n "${USB_SERIAL}" ]; then
     USB_SERIAL=`echo "${USB_SERIAL}" |
-                  xargs grep -l ${ANDROID_SERIAL}`
+                  xargs grep -l ${ANDROID_SERIAL} || true`
   fi
 fi
 USB_ADDRESS=
@@ -1315,7 +1315,7 @@
 [ -n "${VENDOR_DEVT%[0-9a-fA-F][0-9a-fA-F]}" ] ||
   echo "${YELLOW}[  WARNING ]${NORMAL} vendor devt ${VENDOR_DEVT} major 0" >&2
 
-# Download libc.so, append some gargage, push back, and check if the file
+# Download libc.so, append some garbage, push back, and check if the file
 # is updated.
 tempdir="`mktemp -d`"
 cleanup() {
@@ -1323,8 +1323,8 @@
 }
 adb pull /system/lib/bootstrap/libc.so ${tempdir} >/dev/null ||
   die "pull libc.so from device"
-garbage="`hexdump -n 16 -e '4/4 "%08X" 1 "\n"' /dev/random`"
-echo ${garbage} >> ${tempdir}/libc.so
+garbage="D105225BBFCB1EB8AB8EBDB7094646F0"
+echo "${garbage}" >> ${tempdir}/libc.so
 adb push ${tempdir}/libc.so /system/lib/bootstrap/libc.so >/dev/null ||
   die "push libc.so to device"
 adb pull /system/lib/bootstrap/libc.so ${tempdir}/libc.so.fromdevice >/dev/null ||
diff --git a/gatekeeperd/gatekeeperd.cpp b/gatekeeperd/gatekeeperd.cpp
index 781b4af..f9c0cdd 100644
--- a/gatekeeperd/gatekeeperd.cpp
+++ b/gatekeeperd/gatekeeperd.cpp
@@ -318,6 +318,7 @@
 
                     authToken.timestamp.milliSeconds = betoh64(hwAuthToken->timestamp);
                     authToken.challenge = hwAuthToken->challenge;
+                    authToken.userId = hwAuthToken->user_id;
                     authToken.authenticatorId = hwAuthToken->authenticator_id;
                     authToken.authenticatorType = static_cast<HardwareAuthenticatorType>(
                             betoh32(hwAuthToken->authenticator_type));
diff --git a/healthd/BatteryMonitor.cpp b/healthd/BatteryMonitor.cpp
index fd810cb..377acb7 100644
--- a/healthd/BatteryMonitor.cpp
+++ b/healthd/BatteryMonitor.cpp
@@ -349,9 +349,14 @@
 }
 
 void BatteryMonitor::logValues(void) {
+    logValues(*mHealthInfo, *mHealthdConfig);
+}
+
+void BatteryMonitor::logValues(const android::hardware::health::V2_1::HealthInfo& health_info,
+                               const struct healthd_config& healthd_config) {
     char dmesgline[256];
     size_t len;
-    const HealthInfo_1_0& props = mHealthInfo->legacy.legacy;
+    const HealthInfo_1_0& props = health_info.legacy.legacy;
     if (props.batteryPresent) {
         snprintf(dmesgline, sizeof(dmesgline), "battery l=%d v=%d t=%s%d.%d h=%d st=%d",
                  props.batteryLevel, props.batteryVoltage, props.batteryTemperature < 0 ? "-" : "",
@@ -359,17 +364,17 @@
                  props.batteryHealth, props.batteryStatus);
 
         len = strlen(dmesgline);
-        if (!mHealthdConfig->batteryCurrentNowPath.isEmpty()) {
+        if (!healthd_config.batteryCurrentNowPath.isEmpty()) {
             len += snprintf(dmesgline + len, sizeof(dmesgline) - len, " c=%d",
                             props.batteryCurrent);
         }
 
-        if (!mHealthdConfig->batteryFullChargePath.isEmpty()) {
+        if (!healthd_config.batteryFullChargePath.isEmpty()) {
             len += snprintf(dmesgline + len, sizeof(dmesgline) - len, " fc=%d",
                             props.batteryFullCharge);
         }
 
-        if (!mHealthdConfig->batteryCycleCountPath.isEmpty()) {
+        if (!healthd_config.batteryCycleCountPath.isEmpty()) {
             len += snprintf(dmesgline + len, sizeof(dmesgline) - len, " cc=%d",
                             props.batteryCycleCount);
         }
diff --git a/healthd/include/healthd/BatteryMonitor.h b/healthd/include/healthd/BatteryMonitor.h
index fadb5a5..3cda727 100644
--- a/healthd/include/healthd/BatteryMonitor.h
+++ b/healthd/include/healthd/BatteryMonitor.h
@@ -66,6 +66,9 @@
     void logValues(void);
     bool isChargerOnline();
 
+    static void logValues(const android::hardware::health::V2_1::HealthInfo& health_info,
+                          const struct healthd_config& healthd_config);
+
   private:
     struct healthd_config *mHealthdConfig;
     Vector<String8> mChargerNames;
diff --git a/init/Android.bp b/init/Android.bp
index b0a59b1..5da8e36 100644
--- a/init/Android.bp
+++ b/init/Android.bp
@@ -235,6 +235,119 @@
     visibility: ["//packages/modules/Virtualization/microdroid"],
 }
 
+// This currently is only for the VM usecase.
+// TODO(jiyong): replace init_first_stage in Android.mk with this
+cc_binary {
+    name: "init_first_stage_soong",
+    stem: "init_vendor",
+
+    srcs: [
+        "block_dev_initializer.cpp",
+        "devices.cpp",
+        "first_stage_console.cpp",
+        "first_stage_init.cpp",
+        "first_stage_main.cpp",
+        "first_stage_mount.cpp",
+        "reboot_utils.cpp",
+        "selabel.cpp",
+        "selinux.cpp",
+        "service_utils.cpp",
+        "snapuserd_transition.cpp",
+        "switch_root.cpp",
+        "uevent_listener.cpp",
+        "util.cpp",
+    ],
+
+    static_libs: [
+        "libc++fs",
+        "libfs_avb",
+        "libfs_mgr",
+        "libfec",
+        "libfec_rs",
+        "libsquashfs_utils",
+        "liblogwrap",
+        "libext4_utils",
+        "libcrypto_utils",
+        "libsparse",
+        "libavb",
+        "libkeyutils",
+        "liblp",
+        "libcutils",
+        "libbase",
+        "liblog",
+        "libcrypto_static",
+        "libdl",
+        "libz",
+        "libselinux",
+        "libcap",
+        "libgsi",
+        "libcom.android.sysprop.apex",
+        "liblzma",
+        "libunwindstack_no_dex",
+        "libbacktrace_no_dex",
+        "libmodprobe",
+        "libext2_uuid",
+        "libprotobuf-cpp-lite",
+        "libsnapshot_cow",
+        "libsnapshot_init",
+        "update_metadata-protos",
+    ],
+
+    static_executable: true,
+
+    cflags: [
+        "-Wall",
+        "-Wextra",
+        "-Wno-unused-parameter",
+        "-Werror",
+        "-DALLOW_FIRST_STAGE_CONSOLE=0",
+        "-DALLOW_LOCAL_PROP_OVERRIDE=0",
+        "-DALLOW_PERMISSIVE_SELINUX=0",
+        "-DREBOOT_BOOTLOADER_ON_PANIC=0",
+        "-DWORLD_WRITABLE_KMSG=0",
+        "-DDUMP_ON_UMOUNT_FAILURE=0",
+        "-DSHUTDOWN_ZERO_TIMEOUT=0",
+        "-DLOG_UEVENTS=0",
+        "-DSEPOLICY_VERSION=30", // TODO(jiyong): externalize the version number
+    ],
+
+    product_variables: {
+        debuggable: {
+            cflags: [
+                "-UALLOW_FIRST_STAGE_CONSOLE",
+                "-DALLOW_FIRST_STAGE_CONSOLE=1",
+
+                "-UALLOW_LOCAL_PROP_OVERRIDE",
+                "-DALLOW_LOCAL_PROP_OVERRIDE=1",
+
+                "-UALLOW_PERMISSIVE_SELINUX",
+                "-DALLOW_PERMISSIVE_SELINUX=1",
+
+                "-UREBOOT_BOOTLOADER_ON_PANIC",
+                "-DREBOOT_BOOTLOADER_ON_PANIC=1",
+
+                "-UWORLD_WRITABLE_KMSG",
+                "-DWORLD_WRITABLE_KMSG=1",
+
+                "-UDUMP_ON_UMOUNT_FAILURE",
+                "-DDUMP_ON_UMOUNT_FAILURE=1",
+            ],
+        },
+
+        eng: {
+            cflags: [
+                "-USHUTDOWN_ZERO_TIMEOUT",
+                "-DSHUTDOWN_ZERO_TIMEOUT=1",
+            ],
+        },
+    },
+
+    sanitize: {
+        misc_undefined: ["signed-integer-overflow"],
+        hwaddress: false,
+    },
+}
+
 // Tests
 // ------------------------------------------------------------------------------
 
diff --git a/init/init.cpp b/init/init.cpp
index 0c752a9..b08037a 100644
--- a/init/init.cpp
+++ b/init/init.cpp
@@ -723,37 +723,6 @@
     }
 }
 
-static Result<void> TransitionSnapuserdAction(const BuiltinArguments&) {
-    if (!SnapshotManager::IsSnapshotManagerNeeded() ||
-        !android::base::GetBoolProperty(android::snapshot::kVirtualAbCompressionProp, false)) {
-        return {};
-    }
-
-    auto sm = SnapshotManager::New();
-    if (!sm) {
-        LOG(FATAL) << "Failed to create SnapshotManager, will not transition snapuserd";
-        return {};
-    }
-
-    ServiceList& service_list = ServiceList::GetInstance();
-    auto svc = service_list.FindService("snapuserd");
-    if (!svc) {
-        LOG(FATAL) << "Failed to find snapuserd service, aborting transition";
-        return {};
-    }
-    svc->Start();
-    svc->SetShutdownCritical();
-
-    if (!sm->PerformSecondStageInitTransition()) {
-        LOG(FATAL) << "Failed to transition snapuserd to second-stage";
-    }
-
-    if (auto pid = GetSnapuserdFirstStagePid()) {
-        KillFirstStageSnapuserd(pid.value());
-    }
-    return {};
-}
-
 int SecondStageMain(int argc, char** argv) {
     if (REBOOT_BOOTLOADER_ON_PANIC) {
         InstallRebootSignalHandlers();
@@ -900,7 +869,6 @@
 
     // Queue an action that waits for coldboot done so we know ueventd has set up all of /dev...
     am.QueueBuiltinAction(wait_for_coldboot_done_action, "wait_for_coldboot_done");
-    am.QueueBuiltinAction(TransitionSnapuserdAction, "TransitionSnapuserd");
     // ... so that we can start queuing up actions that require stuff from /dev.
     am.QueueBuiltinAction(SetMmapRndBitsAction, "SetMmapRndBits");
     Keychords keychords;
diff --git a/libcutils/trace-dev.inc b/libcutils/trace-dev.inc
index 821f1ea..3b459e0 100644
--- a/libcutils/trace-dev.inc
+++ b/libcutils/trace-dev.inc
@@ -21,6 +21,7 @@
 
 #include <errno.h>
 #include <fcntl.h>
+#include <fnmatch.h>
 #include <limits.h>
 #include <pthread.h>
 #include <stdatomic.h>
@@ -106,7 +107,7 @@
     for (int i = 0; i < count; i++) {
         snprintf(buf, sizeof(buf), "debug.atrace.app_%d", i);
         property_get(buf, value, "");
-        if (strcmp(value, "*") == 0 || strcmp(value, cmdline) == 0) {
+        if (fnmatch(value, cmdline, FNM_NOESCAPE) == 0) {
             return true;
         }
     }
diff --git a/libprocessgroup/include/processgroup/processgroup.h b/libprocessgroup/include/processgroup/processgroup.h
index 1cadc9f..1aaed7b 100644
--- a/libprocessgroup/include/processgroup/processgroup.h
+++ b/libprocessgroup/include/processgroup/processgroup.h
@@ -36,7 +36,8 @@
 
 static constexpr const char* CGROUPS_RC_PATH = "/dev/cgroup_info/cgroup.rc";
 // Path to test against for freezer support
-static constexpr const char* CGROUP_FREEZE_PATH = "/sys/fs/cgroup/freezer/cgroup.freeze";
+// TODO: remove and replace with a function call, see http://b/180056337
+static constexpr const char* CGROUP_FREEZE_PATH = "/sys/fs/cgroup/uid_0/cgroup.freeze";
 
 bool UsePerAppMemcg();
 
diff --git a/libprocessgroup/processgroup.cpp b/libprocessgroup/processgroup.cpp
index d669ebe..209ccd9 100644
--- a/libprocessgroup/processgroup.cpp
+++ b/libprocessgroup/processgroup.cpp
@@ -131,13 +131,25 @@
     return StringPrintf("%s/uid_%d/pid_%d", cgroup, uid, pid);
 }
 
-static int RemoveProcessGroup(const char* cgroup, uid_t uid, int pid) {
-    int ret;
-
+static int RemoveProcessGroup(const char* cgroup, uid_t uid, int pid, unsigned int retries) {
+    int ret = 0;
     auto uid_pid_path = ConvertUidPidToPath(cgroup, uid, pid);
-    ret = rmdir(uid_pid_path.c_str());
-
     auto uid_path = ConvertUidToPath(cgroup, uid);
+
+    if (retries == 0) {
+        retries = 1;
+    }
+
+    while (retries--) {
+        ret = rmdir(uid_pid_path.c_str());
+        if (!ret || errno != EBUSY) break;
+        std::this_thread::sleep_for(5ms);
+    }
+
+    // With the exception of boot or shutdown, system uid_ folders are always populated. Spinning
+    // here would needlessly delay most pid removals. Additionally, once empty a uid_ cgroup won't
+    // have processes hanging on it (we've already spun for all its pid_), so there's no need to
+    // spin anyway.
     rmdir(uid_path.c_str());
 
     return ret;
@@ -176,7 +188,7 @@
     std::vector<std::string> cgroups;
     std::string path;
 
-    if (CgroupGetControllerPath("cpuacct", &path)) {
+    if (CgroupGetControllerPath(CGROUPV2_CONTROLLER_NAME, &path)) {
         cgroups.push_back(path);
     }
     if (CgroupGetControllerPath("memory", &path)) {
@@ -212,19 +224,49 @@
     }
 }
 
+/**
+ * Process groups are primarily created by the Zygote, meaning that uid/pid groups are created by
+ * the user root. Ownership for the newly created cgroup and all of its files must thus be
+ * transferred for the user/group passed as uid/gid before system_server can properly access them.
+ */
 static bool MkdirAndChown(const std::string& path, mode_t mode, uid_t uid, gid_t gid) {
     if (mkdir(path.c_str(), mode) == -1 && errno != EEXIST) {
         return false;
     }
 
-    if (chown(path.c_str(), uid, gid) == -1) {
-        int saved_errno = errno;
-        rmdir(path.c_str());
-        errno = saved_errno;
-        return false;
+    auto dir = std::unique_ptr<DIR, decltype(&closedir)>(opendir(path.c_str()), closedir);
+
+    if (dir == NULL) {
+        PLOG(ERROR) << "opendir failed for " << path;
+        goto err;
+    }
+
+    struct dirent* dir_entry;
+    while ((dir_entry = readdir(dir.get()))) {
+        if (!strcmp("..", dir_entry->d_name)) {
+            continue;
+        }
+
+        std::string file_path = path + "/" + dir_entry->d_name;
+
+        if (lchown(file_path.c_str(), uid, gid) < 0) {
+            PLOG(ERROR) << "lchown failed for " << file_path;
+            goto err;
+        }
+
+        if (fchmodat(AT_FDCWD, file_path.c_str(), mode, AT_SYMLINK_NOFOLLOW) != 0) {
+            PLOG(ERROR) << "fchmodat failed for " << file_path;
+            goto err;
+        }
     }
 
     return true;
+err:
+    int saved_errno = errno;
+    rmdir(path.c_str());
+    errno = saved_errno;
+
+    return false;
 }
 
 // Returns number of processes killed on success
@@ -302,17 +344,9 @@
 
 static int KillProcessGroup(uid_t uid, int initialPid, int signal, int retries,
                             int* max_processes) {
-    std::string cpuacct_path;
-    std::string memory_path;
-
-    CgroupGetControllerPath("cpuacct", &cpuacct_path);
-    CgroupGetControllerPath("memory", &memory_path);
-    memory_path += "/apps";
-
-    const char* cgroup =
-            (!access(ConvertUidPidToPath(cpuacct_path.c_str(), uid, initialPid).c_str(), F_OK))
-                    ? cpuacct_path.c_str()
-                    : memory_path.c_str();
+    std::string hierarchy_root_path;
+    CgroupGetControllerPath(CGROUPV2_CONTROLLER_NAME, &hierarchy_root_path);
+    const char* cgroup = hierarchy_root_path.c_str();
 
     std::chrono::steady_clock::time_point start = std::chrono::steady_clock::now();
 
@@ -355,7 +389,17 @@
             LOG(INFO) << "Successfully killed process cgroup uid " << uid << " pid " << initialPid
                       << " in " << static_cast<int>(ms) << "ms";
         }
-        return RemoveProcessGroup(cgroup, uid, initialPid);
+
+        int err = RemoveProcessGroup(cgroup, uid, initialPid, retries);
+
+        if (isMemoryCgroupSupported() && UsePerAppMemcg()) {
+            std::string memory_path;
+            CgroupGetControllerPath("memory", &memory_path);
+            memory_path += "/apps";
+            if (RemoveProcessGroup(memory_path.c_str(), uid, initialPid, retries)) return -1;
+        }
+
+        return err;
     } else {
         if (retries > 0) {
             LOG(ERROR) << "Failed to kill process cgroup uid " << uid << " pid " << initialPid
@@ -374,25 +418,30 @@
     return KillProcessGroup(uid, initialPid, signal, 0 /*retries*/, max_processes);
 }
 
-int createProcessGroup(uid_t uid, int initialPid, bool memControl) {
-    std::string cgroup;
-    if (isMemoryCgroupSupported() && (memControl || UsePerAppMemcg())) {
-        CgroupGetControllerPath("memory", &cgroup);
-        cgroup += "/apps";
-    } else {
-        CgroupGetControllerPath("cpuacct", &cgroup);
-    }
-
+static int createProcessGroupInternal(uid_t uid, int initialPid, std::string cgroup) {
     auto uid_path = ConvertUidToPath(cgroup.c_str(), uid);
 
-    if (!MkdirAndChown(uid_path, 0750, AID_SYSTEM, AID_SYSTEM)) {
+    struct stat cgroup_stat;
+    mode_t cgroup_mode = 0750;
+    gid_t cgroup_uid = AID_SYSTEM;
+    uid_t cgroup_gid = AID_SYSTEM;
+
+    if (stat(cgroup.c_str(), &cgroup_stat) == 1) {
+        PLOG(ERROR) << "Failed to get stats for " << cgroup;
+    } else {
+        cgroup_mode = cgroup_stat.st_mode;
+        cgroup_uid = cgroup_stat.st_uid;
+        cgroup_gid = cgroup_stat.st_gid;
+    }
+
+    if (!MkdirAndChown(uid_path, cgroup_mode, cgroup_uid, cgroup_gid)) {
         PLOG(ERROR) << "Failed to make and chown " << uid_path;
         return -errno;
     }
 
     auto uid_pid_path = ConvertUidPidToPath(cgroup.c_str(), uid, initialPid);
 
-    if (!MkdirAndChown(uid_pid_path, 0750, AID_SYSTEM, AID_SYSTEM)) {
+    if (!MkdirAndChown(uid_pid_path, cgroup_mode, cgroup_uid, cgroup_gid)) {
         PLOG(ERROR) << "Failed to make and chown " << uid_pid_path;
         return -errno;
     }
@@ -408,6 +457,27 @@
     return ret;
 }
 
+int createProcessGroup(uid_t uid, int initialPid, bool memControl) {
+    std::string cgroup;
+
+    if (memControl && !UsePerAppMemcg()) {
+        PLOG(ERROR) << "service memory controls are used without per-process memory cgroup support";
+        return -EINVAL;
+    }
+
+    if (isMemoryCgroupSupported() && UsePerAppMemcg()) {
+        CgroupGetControllerPath("memory", &cgroup);
+        cgroup += "/apps";
+        int ret = createProcessGroupInternal(uid, initialPid, cgroup);
+        if (ret != 0) {
+            return ret;
+        }
+    }
+
+    CgroupGetControllerPath(CGROUPV2_CONTROLLER_NAME, &cgroup);
+    return createProcessGroupInternal(uid, initialPid, cgroup);
+}
+
 static bool SetProcessGroupValue(int tid, const std::string& attr_name, int64_t value) {
     if (!isMemoryCgroupSupported()) {
         PLOG(ERROR) << "Memcg is not mounted.";
diff --git a/libprocessgroup/profiles/cgroups.json b/libprocessgroup/profiles/cgroups.json
index 5b7a28a..7bcb94b 100644
--- a/libprocessgroup/profiles/cgroups.json
+++ b/libprocessgroup/profiles/cgroups.json
@@ -15,11 +15,6 @@
       "GID": "system"
     },
     {
-      "Controller": "cpuacct",
-      "Path": "/acct",
-      "Mode": "0555"
-    },
-    {
       "Controller": "cpuset",
       "Path": "/dev/cpuset",
       "Mode": "0755",
diff --git a/libprocessgroup/profiles/cgroups.recovery.json b/libprocessgroup/profiles/cgroups.recovery.json
index f0bf5fd..e275252 100644
--- a/libprocessgroup/profiles/cgroups.recovery.json
+++ b/libprocessgroup/profiles/cgroups.recovery.json
@@ -1,9 +1,8 @@
 {
-  "Cgroups": [
-    {
-      "Controller": "cpuacct",
-      "Path": "/acct",
-      "Mode": "0555"
-    }
-  ]
+  "Cgroups2": {
+    "Path": "/sys/fs/cgroup",
+    "Mode": "0755",
+    "UID": "root",
+    "GID": "root"
+  }
 }
diff --git a/rootdir/init.rc b/rootdir/init.rc
index 03af4f3..863cf6c 100644
--- a/rootdir/init.rc
+++ b/rootdir/init.rc
@@ -655,6 +655,7 @@
     mkdir /data/apex 0755 root system encryption=None
     mkdir /data/apex/active 0755 root system
     mkdir /data/apex/backup 0700 root system
+    mkdir /data/apex/decompressed 0700 root system encryption=Require
     mkdir /data/apex/hashtree 0700 root system
     mkdir /data/apex/sessions 0700 root system
     mkdir /data/app-staging 0751 system system encryption=DeleteIfNecessary
@@ -722,6 +723,8 @@
     mkdir /data/misc/trace 0700 root root
     # create location to store surface and window trace files
     mkdir /data/misc/wmtrace 0700 system system
+    # create location to store accessibility trace files
+    mkdir /data/misc/a11ytrace 0700 system system
     # profile file layout
     mkdir /data/misc/profiles 0771 system system
     mkdir /data/misc/profiles/cur 0771 system system
@@ -1169,6 +1172,8 @@
     chmod 0773 /data/misc/trace
     # Give reads to anyone for the window trace folder on debug builds.
     chmod 0775 /data/misc/wmtrace
+    # Give reads to anyone for the accessibility trace folder on debug builds.
+    chmod 0775 /data/misc/a11ytrace
 
 on init && property:ro.debuggable=1
     start console
diff --git a/rootdir/ueventd.rc b/rootdir/ueventd.rc
index e9293b5..65e29c1 100644
--- a/rootdir/ueventd.rc
+++ b/rootdir/ueventd.rc
@@ -37,8 +37,6 @@
 /dev/tty                  0666   root       root
 /dev/random               0666   root       root
 /dev/urandom              0666   root       root
-# Make HW RNG readable by group system to let EntropyMixer read it.
-/dev/hw_random            0440   root       system
 /dev/ashmem*              0666   root       root
 /dev/binder               0666   root       root
 /dev/hwbinder             0666   root       root
diff --git a/trusty/confirmationui/Android.bp b/trusty/confirmationui/Android.bp
index 60e0e71..09d48ad 100644
--- a/trusty/confirmationui/Android.bp
+++ b/trusty/confirmationui/Android.bp
@@ -54,6 +54,7 @@
         "android.hardware.confirmationui@1.0",
         "android.hardware.keymaster@4.0",
         "libbase",
+        "libdmabufheap",
         "libhidlbase",
         "libteeui_hal_support",
         "libtrusty",
@@ -92,4 +93,4 @@
         "-Werror",
         "-DTEEUI_USE_STD_VECTOR",
     ],
-}
\ No newline at end of file
+}
diff --git a/trusty/confirmationui/TrustyApp.cpp b/trusty/confirmationui/TrustyApp.cpp
index e4c68f9..0e84b19 100644
--- a/trusty/confirmationui/TrustyApp.cpp
+++ b/trusty/confirmationui/TrustyApp.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright 2020, The Android Open Source Project
+ * Copyright 2021, 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.
@@ -15,140 +15,155 @@
  */
 
 #include "TrustyApp.h"
+#include "TrustyIpc.h"
 
+#include <BufferAllocator/BufferAllocator.h>
 #include <android-base/logging.h>
+#include <sys/mman.h>
 #include <sys/uio.h>
 #include <trusty/tipc.h>
 
+#define countof(arr) (sizeof(arr) / sizeof(arr[0]))
+
 namespace android {
 namespace trusty {
 
-// 0x1000 is the message buffer size but we need to leave some space for a protocol header.
-// This assures that packets can always be read/written in one read/write operation.
-static constexpr const uint32_t kPacketSize = 0x1000 - 32;
+using ::android::base::unique_fd;
 
-enum class PacketType : uint32_t {
-    SND,
-    RCV,
-    ACK,
-};
-
-struct PacketHeader {
-    PacketType type;
-    uint32_t remaining;
-};
-
-const char* toString(PacketType t) {
-    switch (t) {
-    case PacketType::SND:
-        return "SND";
-    case PacketType::RCV:
-        return "RCV";
-    case PacketType::ACK:
-        return "ACK";
-    default:
-        return "UNKNOWN";
-    }
+static inline uintptr_t RoundPageUp(uintptr_t val) {
+    return (val + (PAGE_SIZE - 1)) & ~(PAGE_SIZE - 1);
 }
 
-static constexpr const uint32_t kHeaderSize = sizeof(PacketHeader);
-static constexpr const uint32_t kPayloadSize = kPacketSize - kHeaderSize;
+ssize_t TrustyApp::TrustyRpc(const uint8_t* obegin, const uint8_t* oend, uint8_t* ibegin,
+                             uint8_t* iend) {
+    uint32_t olen = oend - obegin;
 
-ssize_t TrustyRpc(int handle, const uint8_t* obegin, const uint8_t* oend, uint8_t* ibegin,
-                  uint8_t* iend) {
-    while (obegin != oend) {
-        PacketHeader header = {
-            .type = PacketType::SND,
-            .remaining = uint32_t(oend - obegin),
-        };
-        uint32_t body_size = std::min(kPayloadSize, header.remaining);
-        iovec iov[] = {
-            {
-                .iov_base = &header,
-                .iov_len = kHeaderSize,
-            },
-            {
-                .iov_base = const_cast<uint8_t*>(obegin),
-                .iov_len = body_size,
-            },
-        };
-        int rc = writev(handle, iov, 2);
-        if (!rc) {
-            PLOG(ERROR) << "Error sending SND message. " << rc;
-            return rc;
-        }
-
-        obegin += body_size;
-
-        rc = read(handle, &header, kHeaderSize);
-        if (!rc) {
-            PLOG(ERROR) << "Error reading ACK. " << rc;
-            return rc;
-        }
-
-        if (header.type != PacketType::ACK || header.remaining != oend - obegin) {
-            LOG(ERROR) << "malformed ACK";
-            return -1;
-        }
+    if (olen > shm_len_) {
+        LOG(ERROR) << AT << "request message too long to fit in shared memory";
+        return -1;
     }
 
-    ssize_t remaining = 0;
-    auto begin = ibegin;
-    do {
-        PacketHeader header = {
-            .type = PacketType::RCV,
-            .remaining = 0,
-        };
+    memcpy(shm_base_, obegin, olen);
 
-        iovec iov[] = {
-            {
-                .iov_base = &header,
-                .iov_len = kHeaderSize,
-            },
-            {
-                .iov_base = begin,
-                .iov_len = uint32_t(iend - begin),
-            },
-        };
+    confirmationui_hdr hdr = {
+        .cmd = CONFIRMATIONUI_CMD_MSG,
+    };
+    confirmationui_msg_args args = {
+        .msg_len = olen,
+    };
+    iovec iov[] = {
+        {
+            .iov_base = &hdr,
+            .iov_len = sizeof(hdr),
+        },
+        {
+            .iov_base = &args,
+            .iov_len = sizeof(args),
+        },
+    };
 
-        ssize_t rc = writev(handle, iov, 1);
-        if (!rc) {
-            PLOG(ERROR) << "Error sending RCV message. " << rc;
-            return rc;
-        }
+    int rc = tipc_send(handle_, iov, countof(iov), NULL, 0);
+    if (rc != static_cast<int>(sizeof(hdr) + sizeof(args))) {
+        LOG(ERROR) << AT << "failed to send MSG request";
+        return -1;
+    }
 
-        rc = readv(handle, iov, 2);
-        if (rc < 0) {
-            PLOG(ERROR) << "Error reading response. " << rc;
-            return rc;
-        }
+    rc = readv(handle_, iov, countof(iov));
+    if (rc != static_cast<int>(sizeof(hdr) + sizeof(args))) {
+        LOG(ERROR) << AT << "failed to receive MSG response";
+        return -1;
+    }
 
-        uint32_t body_size = std::min(kPayloadSize, header.remaining);
-        if (body_size != rc - kHeaderSize) {
-            LOG(ERROR) << "Unexpected amount of data: " << rc;
-            return -1;
-        }
+    if (hdr.cmd != (CONFIRMATIONUI_CMD_MSG | CONFIRMATIONUI_RESP_BIT)) {
+        LOG(ERROR) << AT << "unknown response command: " << hdr.cmd;
+        return -1;
+    }
 
-        remaining = header.remaining - body_size;
-        begin += body_size;
-    } while (remaining);
+    uint32_t ilen = iend - ibegin;
+    if (args.msg_len > ilen) {
+        LOG(ERROR) << AT << "response message too long to fit in return buffer";
+        return -1;
+    }
 
-    return begin - ibegin;
+    memcpy(ibegin, shm_base_, args.msg_len);
+
+    return args.msg_len;
 }
 
 TrustyApp::TrustyApp(const std::string& path, const std::string& appname)
     : handle_(kInvalidHandle) {
-    handle_ = tipc_connect(path.c_str(), appname.c_str());
-    if (handle_ == kInvalidHandle) {
+    unique_fd tipc_handle(tipc_connect(path.c_str(), appname.c_str()));
+    if (tipc_handle < 0) {
         LOG(ERROR) << AT << "failed to connect to Trusty TA \"" << appname << "\" using dev:"
                    << "\"" << path << "\"";
+        return;
     }
+
+    uint32_t shm_len = RoundPageUp(CONFIRMATIONUI_MAX_MSG_SIZE);
+    BufferAllocator allocator;
+    unique_fd dma_buf(allocator.Alloc("system", shm_len));
+    if (dma_buf < 0) {
+        LOG(ERROR) << AT << "failed to allocate shared memory buffer";
+        return;
+    }
+
+    if (dma_buf < 0) {
+        LOG(ERROR) << AT << "failed to allocate shared memory buffer";
+        return;
+    }
+
+    confirmationui_hdr hdr = {
+        .cmd = CONFIRMATIONUI_CMD_INIT,
+    };
+    confirmationui_init_req args = {
+        .shm_len = shm_len,
+    };
+    iovec iov[] = {
+        {
+            .iov_base = &hdr,
+            .iov_len = sizeof(hdr),
+        },
+        {
+            .iov_base = &args,
+            .iov_len = sizeof(args),
+        },
+    };
+    trusty_shm shm = {
+        .fd = dma_buf,
+        .transfer = TRUSTY_SHARE,
+    };
+
+    int rc = tipc_send(tipc_handle, iov, 2, &shm, 1);
+    if (rc != static_cast<int>(sizeof(hdr) + sizeof(args))) {
+        LOG(ERROR) << AT << "failed to send INIT request";
+        return;
+    }
+
+    rc = read(tipc_handle, &hdr, sizeof(hdr));
+    if (rc != static_cast<int>(sizeof(hdr))) {
+        LOG(ERROR) << AT << "failed to receive INIT response";
+        return;
+    }
+
+    if (hdr.cmd != (CONFIRMATIONUI_CMD_INIT | CONFIRMATIONUI_RESP_BIT)) {
+        LOG(ERROR) << AT << "unknown response command: " << hdr.cmd;
+        return;
+    }
+
+    void* shm_base = mmap(0, shm_len, PROT_READ | PROT_WRITE, MAP_SHARED, dma_buf, 0);
+    if (shm_base == MAP_FAILED) {
+        LOG(ERROR) << AT << "failed to mmap() shared memory buffer";
+        return;
+    }
+
+    handle_ = std::move(tipc_handle);
+    shm_base_ = shm_base;
+    shm_len_ = shm_len;
+
     LOG(INFO) << AT << "succeeded to connect to Trusty TA \"" << appname << "\"";
 }
+
 TrustyApp::~TrustyApp() {
-    if (handle_ != kInvalidHandle) {
-        tipc_close(handle_);
-    }
     LOG(INFO) << "Done shutting down TrustyApp";
 }
 
diff --git a/trusty/confirmationui/TrustyApp.h b/trusty/confirmationui/TrustyApp.h
index 05a25f6..406f439 100644
--- a/trusty/confirmationui/TrustyApp.h
+++ b/trusty/confirmationui/TrustyApp.h
@@ -16,7 +16,10 @@
 
 #pragma once
 
+#include "TrustyIpc.h"
+
 #include <android-base/logging.h>
+#include <android-base/unique_fd.h>
 #include <errno.h>
 #include <poll.h>
 #include <stdio.h>
@@ -60,19 +63,11 @@
     MSG_TOO_LONG = -2,
 };
 
-/*
- * There is a hard limitation of 0x1800 bytes for the to-be-signed message size. The protocol
- * overhead is limited, so that 0x2000 is a buffer size that will be sufficient in any benign
- * mode of operation.
- */
-static constexpr const size_t kSendBufferSize = 0x2000;
-
-ssize_t TrustyRpc(int handle, const uint8_t* obegin, const uint8_t* oend, uint8_t* ibegin,
-                  uint8_t* iend);
-
 class TrustyApp {
   private:
-    int handle_;
+    android::base::unique_fd handle_;
+    void* shm_base_;
+    size_t shm_len_;
     static constexpr const int kInvalidHandle = -1;
     /*
      * This mutex serializes communication with the trusted app, not handle_.
@@ -84,6 +79,8 @@
     TrustyApp(const std::string& path, const std::string& appname);
     ~TrustyApp();
 
+    ssize_t TrustyRpc(const uint8_t* obegin, const uint8_t* oend, uint8_t* ibegin, uint8_t* iend);
+
     template <typename Request, typename Response, typename... T>
     std::tuple<TrustyAppError, msg2tuple_t<Response>> issueCmd(const T&... args) {
         std::lock_guard<std::mutex> lock(mutex_);
@@ -93,7 +90,7 @@
             return {TrustyAppError::ERROR, {}};
         }
 
-        uint8_t buffer[kSendBufferSize];
+        uint8_t buffer[CONFIRMATIONUI_MAX_MSG_SIZE];
         WriteStream out(buffer);
 
         out = write(Request(), out, args...);
@@ -102,8 +99,8 @@
             return {TrustyAppError::MSG_TOO_LONG, {}};
         }
 
-        auto rc = TrustyRpc(handle_, &buffer[0], const_cast<const uint8_t*>(out.pos()), &buffer[0],
-                            &buffer[kSendBufferSize]);
+        auto rc = TrustyRpc(&buffer[0], const_cast<const uint8_t*>(out.pos()), &buffer[0],
+                            &buffer[CONFIRMATIONUI_MAX_MSG_SIZE]);
         if (rc < 0) return {TrustyAppError::ERROR, {}};
 
         ReadStream in(&buffer[0], rc);
@@ -125,7 +122,7 @@
             return TrustyAppError::ERROR;
         }
 
-        uint8_t buffer[kSendBufferSize];
+        uint8_t buffer[CONFIRMATIONUI_MAX_MSG_SIZE];
         WriteStream out(buffer);
 
         out = write(Request(), out, args...);
@@ -134,8 +131,8 @@
             return TrustyAppError::MSG_TOO_LONG;
         }
 
-        auto rc = TrustyRpc(handle_, &buffer[0], const_cast<const uint8_t*>(out.pos()), &buffer[0],
-                            &buffer[kSendBufferSize]);
+        auto rc = TrustyRpc(&buffer[0], const_cast<const uint8_t*>(out.pos()), &buffer[0],
+                            &buffer[CONFIRMATIONUI_MAX_MSG_SIZE]);
         if (rc < 0) {
             LOG(ERROR) << "send command failed: " << strerror(errno) << " (" << errno << ")";
             return TrustyAppError::ERROR;
diff --git a/trusty/confirmationui/TrustyConfirmationUI.cpp b/trusty/confirmationui/TrustyConfirmationUI.cpp
index 6b25893..c8b24e3 100644
--- a/trusty/confirmationui/TrustyConfirmationUI.cpp
+++ b/trusty/confirmationui/TrustyConfirmationUI.cpp
@@ -71,7 +71,7 @@
 using TeeuiRc = ::teeui::ResponseCode;
 
 constexpr const char kTrustyDeviceName[] = "/dev/trusty-ipc-dev0";
-constexpr const char kConfirmationuiAppName[] = "com.android.trusty.confirmationui";
+constexpr const char kConfirmationuiAppName[] = CONFIRMATIONUI_PORT;
 
 namespace {
 
diff --git a/trusty/confirmationui/TrustyIpc.h b/trusty/confirmationui/TrustyIpc.h
new file mode 100644
index 0000000..eb764bc
--- /dev/null
+++ b/trusty/confirmationui/TrustyIpc.h
@@ -0,0 +1,80 @@
+/*
+ * Copyright 2021 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
+
+#include <stdint.h>
+
+/*
+ * This interface is shared between Android and Trusty. There is a copy in each
+ * repository. They must be kept in sync.
+ */
+
+#define CONFIRMATIONUI_PORT "com.android.trusty.confirmationui"
+
+/**
+ * enum confirmationui_cmd - command identifiers for ConfirmationUI interface
+ * @CONFIRMATIONUI_RESP_BIT:  response bit set as part of response
+ * @CONFIRMATIONUI_REQ_SHIFT: number of bits used by response bit
+ * @CONFIRMATIONUI_CMD_INIT:  command to initialize session
+ * @CONFIRMATIONUI_CMD_MSG:   command to send ConfirmationUI messages
+ */
+enum confirmationui_cmd : uint32_t {
+    CONFIRMATIONUI_RESP_BIT = 1,
+    CONFIRMATIONUI_REQ_SHIFT = 1,
+
+    CONFIRMATIONUI_CMD_INIT = (1 << CONFIRMATIONUI_REQ_SHIFT),
+    CONFIRMATIONUI_CMD_MSG = (2 << CONFIRMATIONUI_REQ_SHIFT),
+};
+
+/**
+ * struct confirmationui_hdr - header for ConfirmationUI messages
+ * @cmd: command identifier
+ *
+ * Note that no messages return a status code. Any error on the server side
+ * results in the connection being closed. So, operations can be assumed to be
+ * successful if they return a response.
+ */
+struct confirmationui_hdr {
+    uint32_t cmd;
+};
+
+/**
+ * struct confirmationui_init_req - arguments for request to initialize a
+ *                                  session
+ * @shm_len: length of memory region being shared
+ *
+ * A handle to a memory region must be sent along with this message. This memory
+ * is send to ConfirmationUI messages.
+ */
+struct confirmationui_init_req {
+    uint32_t shm_len;
+};
+
+/**
+ * struct confirmationui_msg_args - arguments for sending a message
+ * @msg_len: length of message being sent
+ *
+ * Contents of the message are located in the shared memory region that is
+ * established using %CONFIRMATIONUI_CMD_INIT.
+ *
+ * ConfirmationUI messages can travel both ways.
+ */
+struct confirmationui_msg_args {
+    uint32_t msg_len;
+};
+
+#define CONFIRMATIONUI_MAX_MSG_SIZE 0x2000
diff --git a/trusty/confirmationui/android.hardware.confirmationui@1.0-service.trusty.rc b/trusty/confirmationui/android.hardware.confirmationui@1.0-service.trusty.rc
index dc7a03b..3ba6fc0 100644
--- a/trusty/confirmationui/android.hardware.confirmationui@1.0-service.trusty.rc
+++ b/trusty/confirmationui/android.hardware.confirmationui@1.0-service.trusty.rc
@@ -1,4 +1,4 @@
 service confirmationui-1-0 /vendor/bin/hw/android.hardware.confirmationui@1.0-service.trusty
     class hal
-    user nobody
-    group drmrpc input
+    user system
+    group drmrpc input system
diff --git a/trusty/coverage/coverage.cpp b/trusty/coverage/coverage.cpp
index 5eccdc5..3c6b5c5 100644
--- a/trusty/coverage/coverage.cpp
+++ b/trusty/coverage/coverage.cpp
@@ -29,6 +29,7 @@
 #include <trusty/coverage/record.h>
 #include <trusty/coverage/tipc.h>
 #include <trusty/tipc.h>
+#include <iostream>
 
 #define COVERAGE_CLIENT_PORT "com.android.trusty.coverage.client"
 
@@ -122,7 +123,9 @@
 
     int fd = tipc_connect(tipc_dev_.c_str(), COVERAGE_CLIENT_PORT);
     if (fd < 0) {
-        return ErrnoError() << "failed to connect to Trusty coverarge server: ";
+        // Don't error out to support fuzzing builds without coverage, e.g. for repros.
+        std::cerr << "WARNING!!! Failed to connect to Trusty coverarge server." << std::endl;
+        return {};
     }
     coverage_srv_fd_.reset(fd);
 
@@ -130,7 +133,7 @@
     req.open_args.uuid = uuid_;
     auto ret = Rpc(&req, -1, &resp);
     if (!ret.ok()) {
-        return Error() << "failed to open coverage client: ";
+        return Error() << "failed to open coverage client: " << ret.error();
     }
     record_len_ = resp.open_args.record_len;
     shm_len_ = RoundPageUp(record_len_);
@@ -153,13 +156,17 @@
     req.share_record_args.shm_len = shm_len_;
     ret = Rpc(&req, dma_buf, &resp);
     if (!ret.ok()) {
-        return Error() << "failed to send shared memory: ";
+        return Error() << "failed to send shared memory: " << ret.error();
     }
 
     shm_ = shm;
     return {};
 }
 
+bool CoverageRecord::IsOpen() {
+    return shm_;
+}
+
 void CoverageRecord::ResetFullRecord() {
     auto header_region = GetRegionBounds(COV_START);
     if (!header_region.ok()) {
diff --git a/trusty/coverage/include/trusty/coverage/coverage.h b/trusty/coverage/include/trusty/coverage/coverage.h
index 5da68da..9ccc981 100644
--- a/trusty/coverage/include/trusty/coverage/coverage.h
+++ b/trusty/coverage/include/trusty/coverage/coverage.h
@@ -47,6 +47,7 @@
 
     ~CoverageRecord();
     Result<void> Open();
+    bool IsOpen();
     void ResetFullRecord();
     void ResetCounts();
     void ResetPCs();
diff --git a/trusty/fuzz/counters.cpp b/trusty/fuzz/counters.cpp
index 1e863ac..4d34059 100644
--- a/trusty/fuzz/counters.cpp
+++ b/trusty/fuzz/counters.cpp
@@ -41,6 +41,10 @@
 namespace fuzz {
 
 ExtraCounters::ExtraCounters(coverage::CoverageRecord* record) : record_(record) {
+    if (!record_->IsOpen()) {
+        return;
+    }
+
     assert(fuzzer::ExtraCountersBegin());
     assert(fuzzer::ExtraCountersEnd());
 
@@ -51,10 +55,18 @@
 }
 
 ExtraCounters::~ExtraCounters() {
+    if (!record_->IsOpen()) {
+        return;
+    }
+
     Flush();
 }
 
 void ExtraCounters::Reset() {
+    if (!record_->IsOpen()) {
+        return;
+    }
+
     record_->ResetCounts();
     fuzzer::ClearExtraCounters();
 }
diff --git a/trusty/fuzz/utils.cpp b/trusty/fuzz/utils.cpp
index f4cf0b6..3526337 100644
--- a/trusty/fuzz/utils.cpp
+++ b/trusty/fuzz/utils.cpp
@@ -113,7 +113,7 @@
     int rc = write(ta_fd_, buf, len);
     alarm(0);
     if (rc < 0) {
-        return Error() << "failed to read TIPC message from TA: ";
+        return Error() << "failed to write TIPC message to TA: ";
     }
 
     return {};