diff --git a/fastboot/Android.mk b/fastboot/Android.mk
index 322fe5c..10bed6d 100644
--- a/fastboot/Android.mk
+++ b/fastboot/Android.mk
@@ -23,5 +23,5 @@
 my_dist_files += $(HOST_OUT_EXECUTABLES)/make_f2fs
 my_dist_files += $(HOST_OUT_EXECUTABLES)/make_f2fs_casefold
 my_dist_files += $(HOST_OUT_EXECUTABLES)/sload_f2fs
-$(call dist-for-goals,dist_files sdk win_sdk,$(my_dist_files))
+$(call dist-for-goals,dist_files sdk,$(my_dist_files))
 my_dist_files :=
diff --git a/fastboot/OWNERS b/fastboot/OWNERS
index 58b2a81..17b3466 100644
--- a/fastboot/OWNERS
+++ b/fastboot/OWNERS
@@ -1,3 +1,3 @@
 dvander@google.com
-hridya@google.com
+elsk@google.com
 enh@google.com
diff --git a/fastboot/device/flashing.cpp b/fastboot/device/flashing.cpp
index 3f9bcdc..7bef72a 100644
--- a/fastboot/device/flashing.cpp
+++ b/fastboot/device/flashing.cpp
@@ -172,7 +172,8 @@
         return -EOVERFLOW;
     } else if (data.size() < block_device_size &&
                (partition_name == "boot" || partition_name == "boot_a" ||
-                partition_name == "boot_b")) {
+                partition_name == "boot_b" || partition_name == "init_boot" ||
+                partition_name == "init_boot_a" || partition_name == "init_boot_b")) {
         CopyAVBFooter(&data, block_device_size);
     }
     if (android::base::GetProperty("ro.system.build.type", "") != "user") {
diff --git a/fastboot/fastboot.bash b/fastboot/fastboot.bash
index 5397455..e9bf9e9 100644
--- a/fastboot/fastboot.bash
+++ b/fastboot/fastboot.bash
@@ -109,7 +109,7 @@
 
     cur="${COMP_WORDS[COMP_CWORD]}"
     if [[ $i -eq $COMP_CWORD ]]; then
-        partitions="boot bootloader dtbo modem odm odm_dlkm oem product pvmfw radio recovery system system_dlkm vbmeta vendor vendor_dlkm"
+        partitions="boot bootloader dtbo init_boot modem odm odm_dlkm oem product pvmfw radio recovery system system_dlkm vbmeta vendor vendor_dlkm"
         COMPREPLY=( $(compgen -W "$partitions" -- $cur) )
     else
         _fastboot_util_complete_local_file "${cur}" '!*.img'
diff --git a/fastboot/fastboot.cpp b/fastboot/fastboot.cpp
index c8ef94f..fde1dab 100644
--- a/fastboot/fastboot.cpp
+++ b/fastboot/fastboot.cpp
@@ -141,6 +141,10 @@
 static Image images[] = {
         // clang-format off
     { "boot",     "boot.img",         "boot.sig",     "boot",     false, ImageType::BootCritical },
+    { "init_boot",
+                  "init_boot.img",    "init_boot.sig",
+                                                      "init_boot",
+                                                                  true,  ImageType::BootCritical },
     { nullptr,    "boot_other.img",   "boot.sig",     "boot",     true,  ImageType::Normal },
     { "cache",    "cache.img",        "cache.sig",    "cache",    true,  ImageType::Extra },
     { "dtbo",     "dtbo.img",         "dtbo.sig",     "dtbo",     true,  ImageType::BootCritical },
@@ -1021,7 +1025,7 @@
     return partition_size;
 }
 
-static void copy_boot_avb_footer(const std::string& partition, struct fastboot_buffer* buf) {
+static void copy_avb_footer(const std::string& partition, struct fastboot_buffer* buf) {
     if (buf->sz < AVB_FOOTER_SIZE) {
         return;
     }
@@ -1036,9 +1040,9 @@
     // In this case, partition_size will be zero.
     if (partition_size < buf->sz) {
         fprintf(stderr,
-                "Warning: skip copying boot image avb footer"
-                " (boot partition size: %" PRId64 ", boot image size: %" PRId64 ").\n",
-                partition_size, buf->sz);
+                "Warning: skip copying %s image avb footer"
+                " (%s partition size: %" PRId64 ", %s image size: %" PRId64 ").\n",
+                partition.c_str(), partition.c_str(), partition_size, partition.c_str(), buf->sz);
         return;
     }
 
@@ -1046,7 +1050,7 @@
     // Because buf->fd will still be used afterwards.
     std::string data;
     if (!android::base::ReadFdToString(buf->fd, &data)) {
-        die("Failed reading from boot");
+        die("Failed reading from %s", partition.c_str());
     }
 
     uint64_t footer_offset = buf->sz - AVB_FOOTER_SIZE;
@@ -1055,13 +1059,14 @@
         return;
     }
 
-    unique_fd fd(make_temporary_fd("boot rewriting"));
+    const std::string tmp_fd_template = partition + " rewriting";
+    unique_fd fd(make_temporary_fd(tmp_fd_template.c_str()));
     if (!android::base::WriteStringToFd(data, fd)) {
-        die("Failed writing to modified boot");
+        die("Failed writing to modified %s", partition.c_str());
     }
     lseek(fd.get(), partition_size - AVB_FOOTER_SIZE, SEEK_SET);
     if (!android::base::WriteStringToFd(data.substr(footer_offset), fd)) {
-        die("Failed copying AVB footer in boot");
+        die("Failed copying AVB footer in %s", partition.c_str());
     }
     buf->fd = std::move(fd);
     buf->sz = partition_size;
@@ -1072,8 +1077,9 @@
 {
     sparse_file** s;
 
-    if (partition == "boot" || partition == "boot_a" || partition == "boot_b") {
-        copy_boot_avb_footer(partition, buf);
+    if (partition == "boot" || partition == "boot_a" || partition == "boot_b" ||
+        partition == "init_boot" || partition == "init_boot_a" || partition == "init_boot_b") {
+        copy_avb_footer(partition, buf);
     }
 
     // Rewrite vbmeta if that's what we're flashing and modification has been requested.
diff --git a/fs_mgr/fs_mgr.cpp b/fs_mgr/fs_mgr.cpp
index 33dca58..8ce961b 100644
--- a/fs_mgr/fs_mgr.cpp
+++ b/fs_mgr/fs_mgr.cpp
@@ -170,6 +170,22 @@
             FS_STAT_SET_RESERVED_BLOCKS_FAILED | FS_STAT_ENABLE_ENCRYPTION_FAILED);
 }
 
+static bool umount_retry(const std::string& mount_point) {
+    int retry_count = 5;
+    bool umounted = false;
+
+    while (retry_count-- > 0) {
+        umounted = umount(mount_point.c_str()) == 0;
+        if (umounted) {
+            LINFO << __FUNCTION__ << "(): unmount(" << mount_point << ") succeeded";
+            break;
+        }
+        PERROR << __FUNCTION__ << "(): umount(" << mount_point << ") failed";
+        if (retry_count) sleep(1);
+    }
+    return umounted;
+}
+
 static void check_fs(const std::string& blk_device, const std::string& fs_type,
                      const std::string& target, int* fs_stat) {
     int status;
@@ -209,25 +225,12 @@
                         tmpmnt_opts.c_str());
             PINFO << __FUNCTION__ << "(): mount(" << blk_device << "," << target << "," << fs_type
                   << ")=" << ret;
-            if (!ret) {
-                bool umounted = false;
-                int retry_count = 5;
-                while (retry_count-- > 0) {
-                    umounted = umount(target.c_str()) == 0;
-                    if (umounted) {
-                        LINFO << __FUNCTION__ << "(): unmount(" << target << ") succeeded";
-                        break;
-                    }
-                    PERROR << __FUNCTION__ << "(): umount(" << target << ") failed";
-                    if (retry_count) sleep(1);
-                }
-                if (!umounted) {
-                    // boot may fail but continue and leave it to later stage for now.
-                    PERROR << __FUNCTION__ << "(): umount(" << target << ") timed out";
-                    *fs_stat |= FS_STAT_RO_UNMOUNT_FAILED;
-                }
-            } else {
+            if (ret) {
                 *fs_stat |= FS_STAT_RO_MOUNT_FAILED;
+            } else if (!umount_retry(target)) {
+                // boot may fail but continue and leave it to later stage for now.
+                PERROR << __FUNCTION__ << "(): umount(" << target << ") timed out";
+                *fs_stat |= FS_STAT_RO_UNMOUNT_FAILED;
             }
         }
 
@@ -268,12 +271,12 @@
             LINFO << "Running " << F2FS_FSCK_BIN << " -f -c 10000 --debug-cache "
                   << realpath(blk_device);
             ret = logwrap_fork_execvp(ARRAY_SIZE(f2fs_fsck_forced_argv), f2fs_fsck_forced_argv,
-                                      &status, false, LOG_KLOG | LOG_FILE, false, FSCK_LOG_FILE);
+                                      &status, false, LOG_KLOG | LOG_FILE, false, nullptr);
         } else {
             LINFO << "Running " << F2FS_FSCK_BIN << " -a -c 10000 --debug-cache "
                   << realpath(blk_device);
             ret = logwrap_fork_execvp(ARRAY_SIZE(f2fs_fsck_argv), f2fs_fsck_argv, &status, false,
-                                      LOG_KLOG | LOG_FILE, false, FSCK_LOG_FILE);
+                                      LOG_KLOG | LOG_FILE, false, nullptr);
         }
         if (ret < 0) {
             /* No need to check for error in fork, we can't really handle it now */
@@ -1009,12 +1012,11 @@
 // Check to see if a mountable volume has encryption requirements
 static int handle_encryptable(const FstabEntry& entry) {
     if (should_use_metadata_encryption(entry)) {
-        if (umount(entry.mount_point.c_str()) == 0) {
+        if (umount_retry(entry.mount_point)) {
             return FS_MGR_MNTALL_DEV_NEEDS_METADATA_ENCRYPTION;
-        } else {
-            PERROR << "Could not umount " << entry.mount_point << " - fail since can't encrypt";
-            return FS_MGR_MNTALL_FAIL;
         }
+        PERROR << "Could not umount " << entry.mount_point << " - fail since can't encrypt";
+        return FS_MGR_MNTALL_FAIL;
     } else if (entry.fs_mgr_flags.file_encryption) {
         LINFO << entry.mount_point << " is file encrypted";
         return FS_MGR_MNTALL_DEV_FILE_ENCRYPTED;
@@ -1807,9 +1809,13 @@
     auto& mount_point = alt_mount_point.empty() ? entry.mount_point : alt_mount_point;
 
     // Run fsck if needed
-    prepare_fs_for_mount(entry.blk_device, entry, mount_point);
+    int ret = prepare_fs_for_mount(entry.blk_device, entry, mount_point);
+    // Wiped case doesn't require to try __mount below.
+    if (ret & FS_STAT_INVALID_MAGIC) {
+      return FS_MGR_DOMNT_FAILED;
+    }
 
-    int ret = __mount(entry.blk_device, mount_point, entry);
+    ret = __mount(entry.blk_device, mount_point, entry);
     if (ret) {
       ret = (errno == EBUSY) ? FS_MGR_DOMNT_BUSY : FS_MGR_DOMNT_FAILED;
     }
diff --git a/fs_mgr/fs_mgr_overlayfs.cpp b/fs_mgr/fs_mgr_overlayfs.cpp
index 2b31119..2da5b0f 100644
--- a/fs_mgr/fs_mgr_overlayfs.cpp
+++ b/fs_mgr/fs_mgr_overlayfs.cpp
@@ -880,9 +880,14 @@
             errno = save_errno;
         }
         entry.flags &= ~MS_RDONLY;
+        entry.flags |= MS_SYNCHRONOUS;
+        entry.fs_options = "nodiscard";
         fs_mgr_set_blk_ro(device_path, false);
     }
-    entry.fs_mgr_flags.check = true;
+    // check_fs requires apex runtime library
+    if (fs_mgr_overlayfs_already_mounted("/data", false)) {
+        entry.fs_mgr_flags.check = true;
+    }
     auto save_errno = errno;
     if (mounted) mounted = fs_mgr_do_mount_one(entry) == 0;
     if (!mounted) {
diff --git a/fs_mgr/libfs_avb/avb_util.cpp b/fs_mgr/libfs_avb/avb_util.cpp
index e913d50..85dbb36 100644
--- a/fs_mgr/libfs_avb/avb_util.cpp
+++ b/fs_mgr/libfs_avb/avb_util.cpp
@@ -35,19 +35,6 @@
 namespace android {
 namespace fs_mgr {
 
-std::string GetAvbPropertyDescriptor(const std::string& key,
-                                     const std::vector<VBMetaData>& vbmeta_images) {
-    size_t value_size;
-    for (const auto& vbmeta : vbmeta_images) {
-        const char* value = avb_property_lookup(vbmeta.data(), vbmeta.size(), key.data(),
-                                                key.size(), &value_size);
-        if (value != nullptr) {
-            return {value, value_size};
-        }
-    }
-    return "";
-}
-
 // Constructs dm-verity arguments for sending DM_TABLE_LOAD ioctl to kernel.
 // See the following link for more details:
 // https://gitlab.com/cryptsetup/cryptsetup/wikis/DMVerity
@@ -130,64 +117,6 @@
     return true;
 }
 
-std::unique_ptr<FsAvbHashDescriptor> GetHashDescriptor(
-        const std::string& partition_name, const std::vector<VBMetaData>& vbmeta_images) {
-    bool found = false;
-    const uint8_t* desc_partition_name;
-    auto hash_desc = std::make_unique<FsAvbHashDescriptor>();
-
-    for (const auto& vbmeta : vbmeta_images) {
-        size_t num_descriptors;
-        std::unique_ptr<const AvbDescriptor*[], decltype(&avb_free)> descriptors(
-                avb_descriptor_get_all(vbmeta.data(), vbmeta.size(), &num_descriptors), avb_free);
-
-        if (!descriptors || num_descriptors < 1) {
-            continue;
-        }
-
-        for (size_t n = 0; n < num_descriptors && !found; n++) {
-            AvbDescriptor desc;
-            if (!avb_descriptor_validate_and_byteswap(descriptors[n], &desc)) {
-                LWARNING << "Descriptor[" << n << "] is invalid";
-                continue;
-            }
-            if (desc.tag == AVB_DESCRIPTOR_TAG_HASH) {
-                desc_partition_name = (const uint8_t*)descriptors[n] + sizeof(AvbHashDescriptor);
-                if (!avb_hash_descriptor_validate_and_byteswap((AvbHashDescriptor*)descriptors[n],
-                                                               hash_desc.get())) {
-                    continue;
-                }
-                if (hash_desc->partition_name_len != partition_name.length()) {
-                    continue;
-                }
-                // Notes that desc_partition_name is not NUL-terminated.
-                std::string hash_partition_name((const char*)desc_partition_name,
-                                                hash_desc->partition_name_len);
-                if (hash_partition_name == partition_name) {
-                    found = true;
-                }
-            }
-        }
-
-        if (found) break;
-    }
-
-    if (!found) {
-        LERROR << "Hash descriptor not found: " << partition_name;
-        return nullptr;
-    }
-
-    hash_desc->partition_name = partition_name;
-
-    const uint8_t* desc_salt = desc_partition_name + hash_desc->partition_name_len;
-    hash_desc->salt = BytesToHex(desc_salt, hash_desc->salt_len);
-
-    const uint8_t* desc_digest = desc_salt + hash_desc->salt_len;
-    hash_desc->digest = BytesToHex(desc_digest, hash_desc->digest_len);
-
-    return hash_desc;
-}
-
 std::unique_ptr<FsAvbHashtreeDescriptor> GetHashtreeDescriptor(
         const std::string& partition_name, const std::vector<VBMetaData>& vbmeta_images) {
     bool found = false;
diff --git a/fs_mgr/libfs_avb/avb_util.h b/fs_mgr/libfs_avb/avb_util.h
index e8f7c39..7941c70 100644
--- a/fs_mgr/libfs_avb/avb_util.h
+++ b/fs_mgr/libfs_avb/avb_util.h
@@ -37,12 +37,6 @@
         : partition_name(chain_partition_name), public_key_blob(chain_public_key_blob) {}
 };
 
-std::string GetAvbPropertyDescriptor(const std::string& key,
-                                     const std::vector<VBMetaData>& vbmeta_images);
-
-std::unique_ptr<FsAvbHashDescriptor> GetHashDescriptor(
-        const std::string& partition_name, const std::vector<VBMetaData>& vbmeta_images);
-
 // AvbHashtreeDescriptor to dm-verity table setup.
 std::unique_ptr<FsAvbHashtreeDescriptor> GetHashtreeDescriptor(
         const std::string& partition_name, const std::vector<VBMetaData>& vbmeta_images);
diff --git a/fs_mgr/libfs_avb/fs_avb.cpp b/fs_mgr/libfs_avb/fs_avb.cpp
index 1da7117..a288876 100644
--- a/fs_mgr/libfs_avb/fs_avb.cpp
+++ b/fs_mgr/libfs_avb/fs_avb.cpp
@@ -37,6 +37,7 @@
 
 #include "avb_ops.h"
 #include "avb_util.h"
+#include "fs_avb/fs_avb_util.h"
 #include "sha.h"
 #include "util.h"
 
diff --git a/fs_mgr/libfs_avb/fs_avb_util.cpp b/fs_mgr/libfs_avb/fs_avb_util.cpp
index 1c14cc0..5326226 100644
--- a/fs_mgr/libfs_avb/fs_avb_util.cpp
+++ b/fs_mgr/libfs_avb/fs_avb_util.cpp
@@ -74,6 +74,64 @@
     return GetHashtreeDescriptor(avb_partition_name, vbmeta_images);
 }
 
+std::unique_ptr<FsAvbHashDescriptor> GetHashDescriptor(
+        const std::string& partition_name, const std::vector<VBMetaData>& vbmeta_images) {
+    bool found = false;
+    const uint8_t* desc_partition_name;
+    auto hash_desc = std::make_unique<FsAvbHashDescriptor>();
+
+    for (const auto& vbmeta : vbmeta_images) {
+        size_t num_descriptors;
+        std::unique_ptr<const AvbDescriptor*[], decltype(&avb_free)> descriptors(
+                avb_descriptor_get_all(vbmeta.data(), vbmeta.size(), &num_descriptors), avb_free);
+
+        if (!descriptors || num_descriptors < 1) {
+            continue;
+        }
+
+        for (size_t n = 0; n < num_descriptors && !found; n++) {
+            AvbDescriptor desc;
+            if (!avb_descriptor_validate_and_byteswap(descriptors[n], &desc)) {
+                LWARNING << "Descriptor[" << n << "] is invalid";
+                continue;
+            }
+            if (desc.tag == AVB_DESCRIPTOR_TAG_HASH) {
+                desc_partition_name = (const uint8_t*)descriptors[n] + sizeof(AvbHashDescriptor);
+                if (!avb_hash_descriptor_validate_and_byteswap((AvbHashDescriptor*)descriptors[n],
+                                                               hash_desc.get())) {
+                    continue;
+                }
+                if (hash_desc->partition_name_len != partition_name.length()) {
+                    continue;
+                }
+                // Notes that desc_partition_name is not NUL-terminated.
+                std::string hash_partition_name((const char*)desc_partition_name,
+                                                hash_desc->partition_name_len);
+                if (hash_partition_name == partition_name) {
+                    found = true;
+                }
+            }
+        }
+
+        if (found) break;
+    }
+
+    if (!found) {
+        LERROR << "Hash descriptor not found: " << partition_name;
+        return nullptr;
+    }
+
+    hash_desc->partition_name = partition_name;
+
+    const uint8_t* desc_salt = desc_partition_name + hash_desc->partition_name_len;
+    hash_desc->salt = BytesToHex(desc_salt, hash_desc->salt_len);
+
+    const uint8_t* desc_digest = desc_salt + hash_desc->salt_len;
+    hash_desc->digest = BytesToHex(desc_digest, hash_desc->digest_len);
+
+    return hash_desc;
+}
+
 // Given a path, loads and verifies the vbmeta, to extract the Avb Hash descriptor.
 std::unique_ptr<FsAvbHashDescriptor> GetHashDescriptor(const std::string& avb_partition_name,
                                                        VBMetaData&& vbmeta) {
@@ -84,5 +142,18 @@
     return GetHashDescriptor(avb_partition_name, vbmeta_images);
 }
 
+std::string GetAvbPropertyDescriptor(const std::string& key,
+                                     const std::vector<VBMetaData>& vbmeta_images) {
+    size_t value_size;
+    for (const auto& vbmeta : vbmeta_images) {
+        const char* value = avb_property_lookup(vbmeta.data(), vbmeta.size(), key.data(),
+                                                key.size(), &value_size);
+        if (value != nullptr) {
+            return {value, value_size};
+        }
+    }
+    return "";
+}
+
 }  // namespace fs_mgr
 }  // namespace android
diff --git a/fs_mgr/libfs_avb/include/fs_avb/fs_avb_util.h b/fs_mgr/libfs_avb/include/fs_avb/fs_avb_util.h
index 3f37bd7..1b15db7 100644
--- a/fs_mgr/libfs_avb/include/fs_avb/fs_avb_util.h
+++ b/fs_mgr/libfs_avb/include/fs_avb/fs_avb_util.h
@@ -43,9 +43,15 @@
 std::unique_ptr<FsAvbHashtreeDescriptor> GetHashtreeDescriptor(
         const std::string& avb_partition_name, VBMetaData&& vbmeta);
 
+std::unique_ptr<FsAvbHashDescriptor> GetHashDescriptor(
+        const std::string& partition_name, const std::vector<VBMetaData>& vbmeta_images);
+
 // Gets the hash descriptor for avb_partition_name from the vbmeta.
 std::unique_ptr<FsAvbHashDescriptor> GetHashDescriptor(const std::string& avb_partition_name,
                                                        VBMetaData&& vbmeta);
 
+std::string GetAvbPropertyDescriptor(const std::string& key,
+                                     const std::vector<VBMetaData>& vbmeta_images);
+
 }  // namespace fs_mgr
 }  // namespace android
diff --git a/fs_mgr/libfs_avb/run_tests.sh b/fs_mgr/libfs_avb/run_tests.sh
index 5d2ce3d..3e945a4 100755
--- a/fs_mgr/libfs_avb/run_tests.sh
+++ b/fs_mgr/libfs_avb/run_tests.sh
@@ -1,8 +1,13 @@
 #!/bin/sh
 #
 # Run host tests
-atest libfs_avb_test                 # Tests public libfs_avb APIs.
-atest libfs_avb_internal_test        # Tests libfs_avb private APIs.
+atest --host libfs_avb_test          # Tests public libfs_avb APIs.
+
+# Tests libfs_avb private APIs.
+# The tests need more time to finish, so increase the timeout to 5 mins.
+# The default timeout is only 60 seconds.
+atest --host libfs_avb_internal_test -- --test-arg \
+    com.android.tradefed.testtype.HostGTest:native-test-timeout:5m
 
 # Run device tests
 atest libfs_avb_device_test          # Test public libfs_avb APIs on a device.
diff --git a/fs_mgr/libfs_avb/tests/avb_util_test.cpp b/fs_mgr/libfs_avb/tests/avb_util_test.cpp
index 6f874a6..2e34920 100644
--- a/fs_mgr/libfs_avb/tests/avb_util_test.cpp
+++ b/fs_mgr/libfs_avb/tests/avb_util_test.cpp
@@ -23,6 +23,7 @@
 #include <libavb/libavb.h>
 
 #include "avb_util.h"
+#include "fs_avb/fs_avb_util.h"
 #include "fs_avb_test_util.h"
 
 // Target classes or functions to test:
diff --git a/fs_mgr/libsnapshot/cow_reader.cpp b/fs_mgr/libsnapshot/cow_reader.cpp
index 20030b9..9b5fd2a 100644
--- a/fs_mgr/libsnapshot/cow_reader.cpp
+++ b/fs_mgr/libsnapshot/cow_reader.cpp
@@ -475,10 +475,7 @@
         std::sort(other_ops.begin(), other_ops.end(), std::greater<int>());
     }
 
-    merge_op_blocks->reserve(merge_op_blocks->size() + other_ops.size());
-    for (auto block : other_ops) {
-        merge_op_blocks->emplace_back(block);
-    }
+    merge_op_blocks->insert(merge_op_blocks->end(), other_ops.begin(), other_ops.end());
 
     num_total_data_ops_ = merge_op_blocks->size();
     if (header_.num_merge_ops > 0) {
diff --git a/fs_mgr/libsnapshot/include/libsnapshot/snapshot.h b/fs_mgr/libsnapshot/include/libsnapshot/snapshot.h
index 41c6ef5..120f95b 100644
--- a/fs_mgr/libsnapshot/include/libsnapshot/snapshot.h
+++ b/fs_mgr/libsnapshot/include/libsnapshot/snapshot.h
@@ -396,6 +396,17 @@
         DM_USER,
     };
 
+    // Add new public entries above this line.
+
+    // Helpers for failure injection.
+    using MergeConsistencyChecker =
+            std::function<MergeFailureCode(const std::string& name, const SnapshotStatus& status)>;
+
+    void set_merge_consistency_checker(MergeConsistencyChecker checker) {
+        merge_consistency_checker_ = checker;
+    }
+    MergeConsistencyChecker merge_consistency_checker() const { return merge_consistency_checker_; }
+
   private:
     FRIEND_TEST(SnapshotTest, CleanFirstStageMount);
     FRIEND_TEST(SnapshotTest, CreateSnapshot);
@@ -410,6 +421,7 @@
     FRIEND_TEST(SnapshotTest, NoMergeBeforeReboot);
     FRIEND_TEST(SnapshotTest, UpdateBootControlHal);
     FRIEND_TEST(SnapshotUpdateTest, AddPartition);
+    FRIEND_TEST(SnapshotUpdateTest, ConsistencyCheckResume);
     FRIEND_TEST(SnapshotUpdateTest, DaemonTransition);
     FRIEND_TEST(SnapshotUpdateTest, DataWipeAfterRollback);
     FRIEND_TEST(SnapshotUpdateTest, DataWipeRollbackInRecovery);
@@ -811,6 +823,7 @@
     std::unique_ptr<SnapuserdClient> snapuserd_client_;
     std::unique_ptr<LpMetadata> old_partition_metadata_;
     std::optional<bool> is_snapshot_userspace_;
+    MergeConsistencyChecker merge_consistency_checker_;
 };
 
 }  // namespace snapshot
diff --git a/fs_mgr/libsnapshot/snapshot.cpp b/fs_mgr/libsnapshot/snapshot.cpp
index e6e17bd..f3de2b4 100644
--- a/fs_mgr/libsnapshot/snapshot.cpp
+++ b/fs_mgr/libsnapshot/snapshot.cpp
@@ -87,6 +87,8 @@
 static constexpr char kRollbackIndicatorPath[] = "/metadata/ota/rollback-indicator";
 static constexpr auto kUpdateStateCheckInterval = 2s;
 
+MergeFailureCode CheckMergeConsistency(const std::string& name, const SnapshotStatus& status);
+
 // Note: IImageManager is an incomplete type in the header, so the default
 // destructor doesn't work.
 SnapshotManager::~SnapshotManager() {}
@@ -116,7 +118,9 @@
 }
 
 SnapshotManager::SnapshotManager(IDeviceInfo* device)
-    : dm_(device->GetDeviceMapper()), device_(device), metadata_dir_(device_->GetMetadataDir()) {}
+    : dm_(device->GetDeviceMapper()), device_(device), metadata_dir_(device_->GetMetadataDir()) {
+    merge_consistency_checker_ = android::snapshot::CheckMergeConsistency;
+}
 
 static std::string GetCowName(const std::string& snapshot_name) {
     return snapshot_name + "-cow";
@@ -1329,14 +1333,20 @@
                                                         const SnapshotStatus& status) {
     CHECK(lock);
 
+    return merge_consistency_checker_(name, status);
+}
+
+MergeFailureCode CheckMergeConsistency(const std::string& name, const SnapshotStatus& status) {
     if (!status.compression_enabled()) {
         // Do not try to verify old-style COWs yet.
         return MergeFailureCode::Ok;
     }
 
+    auto& dm = DeviceMapper::Instance();
+
     std::string cow_image_name = GetMappedCowDeviceName(name, status);
     std::string cow_image_path;
-    if (!dm_.GetDmDevicePathByName(cow_image_name, &cow_image_path)) {
+    if (!dm.GetDmDevicePathByName(cow_image_name, &cow_image_path)) {
         LOG(ERROR) << "Failed to get path for cow device: " << cow_image_name;
         return MergeFailureCode::GetCowPathConsistencyCheck;
     }
@@ -1400,9 +1410,11 @@
     }
 
     SnapshotUpdateStatus update_status = ReadSnapshotUpdateStatus(lock);
-    CHECK(update_status.state() == UpdateState::Merging);
+    CHECK(update_status.state() == UpdateState::Merging ||
+          update_status.state() == UpdateState::MergeFailed);
     CHECK(update_status.merge_phase() == MergePhase::FIRST_PHASE);
 
+    update_status.set_state(UpdateState::Merging);
     update_status.set_merge_phase(MergePhase::SECOND_PHASE);
     if (!WriteSnapshotUpdateStatus(lock, update_status)) {
         return MergeFailureCode::WriteStatus;
@@ -1455,6 +1467,14 @@
     }
 
     RemoveAllUpdateState(lock);
+
+    if (UpdateUsesUserSnapshots(lock) && !device()->IsTestDevice()) {
+        if (snapuserd_client_) {
+            snapuserd_client_->DetachSnapuserd();
+            snapuserd_client_->CloseConnection();
+            snapuserd_client_ = nullptr;
+        }
+    }
 }
 
 void SnapshotManager::AcknowledgeMergeFailure(MergeFailureCode failure_code) {
@@ -3188,7 +3208,7 @@
 
             // Terminate stale daemon if any
             std::unique_ptr<SnapuserdClient> snapuserd_client =
-                    SnapuserdClient::Connect(kSnapuserdSocket, 10s);
+                    SnapuserdClient::Connect(kSnapuserdSocket, 5s);
             if (snapuserd_client) {
                 snapuserd_client->DetachSnapuserd();
                 snapuserd_client->CloseConnection();
diff --git a/fs_mgr/libsnapshot/snapshot_test.cpp b/fs_mgr/libsnapshot/snapshot_test.cpp
index 14f2d45..d76558b 100644
--- a/fs_mgr/libsnapshot/snapshot_test.cpp
+++ b/fs_mgr/libsnapshot/snapshot_test.cpp
@@ -54,6 +54,8 @@
 #include <libsnapshot/mock_snapshot.h>
 
 DEFINE_string(force_config, "", "Force testing mode (dmsnap, vab, vabc) ignoring device config.");
+DEFINE_string(force_iouring_disable, "",
+              "Force testing mode (iouring_disabled) - disable io_uring");
 
 namespace android {
 namespace snapshot {
@@ -1394,6 +1396,93 @@
     }
 }
 
+// Test that a transient merge consistency check failure can resume properly.
+TEST_F(SnapshotUpdateTest, ConsistencyCheckResume) {
+    if (!ShouldUseCompression()) {
+        // b/179111359
+        GTEST_SKIP() << "Skipping Virtual A/B Compression test";
+    }
+
+    auto old_sys_size = GetSize(sys_);
+    auto old_prd_size = GetSize(prd_);
+
+    // Grow |sys| but shrink |prd|.
+    SetSize(sys_, old_sys_size * 2);
+    sys_->set_estimate_cow_size(8_MiB);
+    SetSize(prd_, old_prd_size / 2);
+    prd_->set_estimate_cow_size(1_MiB);
+
+    AddOperationForPartitions();
+
+    ASSERT_TRUE(sm->BeginUpdate());
+    ASSERT_TRUE(sm->CreateUpdateSnapshots(manifest_));
+    ASSERT_TRUE(WriteSnapshotAndHash("sys_b"));
+    ASSERT_TRUE(WriteSnapshotAndHash("vnd_b"));
+    ASSERT_TRUE(ShiftAllSnapshotBlocks("prd_b", old_prd_size));
+
+    sync();
+
+    // Assert that source partitions aren't affected.
+    for (const auto& name : {"sys_a", "vnd_a", "prd_a"}) {
+        ASSERT_TRUE(IsPartitionUnchanged(name));
+    }
+
+    ASSERT_TRUE(sm->FinishedSnapshotWrites(false));
+
+    // Simulate shutting down the device.
+    ASSERT_TRUE(UnmapAll());
+
+    // After reboot, init does first stage mount.
+    auto init = NewManagerForFirstStageMount("_b");
+    ASSERT_NE(init, nullptr);
+    ASSERT_TRUE(init->NeedSnapshotsInFirstStageMount());
+    ASSERT_TRUE(init->CreateLogicalAndSnapshotPartitions("super", snapshot_timeout_));
+
+    // Check that the target partitions have the same content.
+    for (const auto& name : {"sys_b", "vnd_b", "prd_b"}) {
+        ASSERT_TRUE(IsPartitionUnchanged(name));
+    }
+
+    auto old_checker = init->merge_consistency_checker();
+
+    init->set_merge_consistency_checker(
+            [](const std::string&, const SnapshotStatus&) -> MergeFailureCode {
+                return MergeFailureCode::WrongMergeCountConsistencyCheck;
+            });
+
+    // Initiate the merge and wait for it to be completed.
+    ASSERT_TRUE(init->InitiateMerge());
+    ASSERT_EQ(init->IsSnapuserdRequired(), ShouldUseUserspaceSnapshots());
+    {
+        // Check that the merge phase is FIRST_PHASE until at least one call
+        // to ProcessUpdateState() occurs.
+        ASSERT_TRUE(AcquireLock());
+        auto local_lock = std::move(lock_);
+        auto status = init->ReadSnapshotUpdateStatus(local_lock.get());
+        ASSERT_EQ(status.merge_phase(), MergePhase::FIRST_PHASE);
+    }
+
+    // Merge should have failed.
+    ASSERT_EQ(UpdateState::MergeFailed, init->ProcessUpdateState());
+
+    // Simulate shutting down the device and creating partitions again.
+    ASSERT_TRUE(UnmapAll());
+
+    // Restore the checker.
+    init->set_merge_consistency_checker(std::move(old_checker));
+
+    ASSERT_TRUE(init->CreateLogicalAndSnapshotPartitions("super", snapshot_timeout_));
+
+    // Complete the merge.
+    ASSERT_EQ(UpdateState::MergeCompleted, init->ProcessUpdateState());
+
+    // Check that the target partitions have the same content after the merge.
+    for (const auto& name : {"sys_b", "vnd_b", "prd_b"}) {
+        ASSERT_TRUE(IsPartitionUnchanged(name))
+                << "Content of " << name << " changes after the merge";
+    }
+}
+
 // Test that if new system partitions uses empty space in super, that region is not snapshotted.
 TEST_F(SnapshotUpdateTest, DirectWriteEmptySpace) {
     GTEST_SKIP() << "b/141889746";
@@ -2682,10 +2771,22 @@
         }
     }
 
+    if (FLAGS_force_iouring_disable == "iouring_disabled") {
+        if (!android::base::SetProperty("snapuserd.test.io_uring.force_disable", "1")) {
+            return testing::AssertionFailure()
+                   << "Failed to disable property: snapuserd.test.io_uring.disabled";
+        }
+    }
+
     int ret = RUN_ALL_TESTS();
 
     if (FLAGS_force_config == "dmsnap") {
         android::base::SetProperty("snapuserd.test.dm.snapshots", "0");
     }
+
+    if (FLAGS_force_iouring_disable == "iouring_disabled") {
+        android::base::SetProperty("snapuserd.test.io_uring.force_disable", "0");
+    }
+
     return ret;
 }
diff --git a/fs_mgr/libsnapshot/snapuserd/Android.bp b/fs_mgr/libsnapshot/snapuserd/Android.bp
index 84bcb94..bc2bceb 100644
--- a/fs_mgr/libsnapshot/snapuserd/Android.bp
+++ b/fs_mgr/libsnapshot/snapuserd/Android.bp
@@ -86,7 +86,9 @@
         "libsnapshot_cow",
         "libz",
         "libext4_utils",
+        "liburing",
     ],
+    include_dirs: ["bionic/libc/kernel"],
 }
 
 cc_binary {
@@ -182,7 +184,10 @@
         "libfs_mgr",
         "libdm",
         "libext4_utils",
+        "liburing",
+        "libgflags",
     ],
+    include_dirs: ["bionic/libc/kernel"],
     header_libs: [
         "libstorage_literals_headers",
         "libfiemap_headers",
diff --git a/fs_mgr/libsnapshot/snapuserd/user-space-merge/snapuserd_core.cpp b/fs_mgr/libsnapshot/snapuserd/user-space-merge/snapuserd_core.cpp
index 95d95cd..5109d82 100644
--- a/fs_mgr/libsnapshot/snapuserd/user-space-merge/snapuserd_core.cpp
+++ b/fs_mgr/libsnapshot/snapuserd/user-space-merge/snapuserd_core.cpp
@@ -16,6 +16,10 @@
 
 #include "snapuserd_core.h"
 
+#include <sys/utsname.h>
+
+#include <android-base/properties.h>
+#include <android-base/scopeguard.h>
 #include <android-base/strings.h>
 
 namespace android {
@@ -288,6 +292,136 @@
     return ReadMetadata();
 }
 
+void SnapshotHandler::FinalizeIouring() {
+    io_uring_queue_exit(ring_.get());
+}
+
+bool SnapshotHandler::InitializeIouring(int io_depth) {
+    ring_ = std::make_unique<struct io_uring>();
+
+    int ret = io_uring_queue_init(io_depth, ring_.get(), 0);
+    if (ret) {
+        LOG(ERROR) << "io_uring_queue_init failed with ret: " << ret;
+        return false;
+    }
+
+    LOG(INFO) << "io_uring_queue_init success with io_depth: " << io_depth;
+    return true;
+}
+
+bool SnapshotHandler::ReadBlocksAsync(const std::string& dm_block_device,
+                                      const std::string& partition_name, size_t size) {
+    // 64k block size with io_depth of 64 is optimal
+    // for a single thread. We just need a single thread
+    // to read all the blocks from all dynamic partitions.
+    size_t io_depth = 64;
+    size_t bs = (64 * 1024);
+
+    if (!InitializeIouring(io_depth)) {
+        return false;
+    }
+
+    LOG(INFO) << "ReadBlockAsync start "
+              << " Block-device: " << dm_block_device << " Partition-name: " << partition_name
+              << " Size: " << size;
+
+    auto scope_guard = android::base::make_scope_guard([this]() -> void { FinalizeIouring(); });
+
+    std::vector<std::unique_ptr<struct iovec>> vecs;
+    using AlignedBuf = std::unique_ptr<void, decltype(free)*>;
+    std::vector<AlignedBuf> alignedBufVector;
+
+    /*
+     * TODO: We need aligned memory for DIRECT-IO. However, if we do
+     * a DIRECT-IO and verify the blocks then we need to inform
+     * update-verifier that block verification has been done and
+     * there is no need to repeat the same. We are not there yet
+     * as we need to see if there are any boot time improvements doing
+     * a DIRECT-IO.
+     *
+     * Also, we could you the same function post merge for block verification;
+     * again, we can do a DIRECT-IO instead of thrashing page-cache and
+     * hurting other applications.
+     *
+     * For now, we will just create aligned buffers but rely on buffered
+     * I/O until we have perf numbers to justify DIRECT-IO.
+     */
+    for (int i = 0; i < io_depth; i++) {
+        auto iovec = std::make_unique<struct iovec>();
+        vecs.push_back(std::move(iovec));
+
+        struct iovec* iovec_ptr = vecs[i].get();
+
+        if (posix_memalign(&iovec_ptr->iov_base, BLOCK_SZ, bs)) {
+            LOG(ERROR) << "posix_memalign failed";
+            return false;
+        }
+
+        iovec_ptr->iov_len = bs;
+        alignedBufVector.push_back(
+                std::unique_ptr<void, decltype(free)*>(iovec_ptr->iov_base, free));
+    }
+
+    android::base::unique_fd fd(TEMP_FAILURE_RETRY(open(dm_block_device.c_str(), O_RDONLY)));
+    if (fd.get() == -1) {
+        SNAP_PLOG(ERROR) << "File open failed - block-device " << dm_block_device
+                         << " partition-name: " << partition_name;
+        return false;
+    }
+
+    loff_t offset = 0;
+    size_t remain = size;
+    size_t read_sz = io_depth * bs;
+
+    while (remain > 0) {
+        size_t to_read = std::min(remain, read_sz);
+        size_t queue_size = to_read / bs;
+
+        for (int i = 0; i < queue_size; i++) {
+            struct io_uring_sqe* sqe = io_uring_get_sqe(ring_.get());
+            if (!sqe) {
+                SNAP_LOG(ERROR) << "io_uring_get_sqe() failed";
+                return false;
+            }
+
+            struct iovec* iovec_ptr = vecs[i].get();
+
+            io_uring_prep_read(sqe, fd.get(), iovec_ptr->iov_base, iovec_ptr->iov_len, offset);
+            sqe->flags |= IOSQE_ASYNC;
+            offset += bs;
+        }
+
+        int ret = io_uring_submit(ring_.get());
+        if (ret != queue_size) {
+            SNAP_LOG(ERROR) << "submit got: " << ret << " wanted: " << queue_size;
+            return false;
+        }
+
+        for (int i = 0; i < queue_size; i++) {
+            struct io_uring_cqe* cqe;
+
+            int ret = io_uring_wait_cqe(ring_.get(), &cqe);
+            if (ret) {
+                SNAP_PLOG(ERROR) << "wait_cqe failed" << ret;
+                return false;
+            }
+
+            if (cqe->res < 0) {
+                SNAP_LOG(ERROR) << "io failed with res: " << cqe->res;
+                return false;
+            }
+            io_uring_cqe_seen(ring_.get(), cqe);
+        }
+
+        remain -= to_read;
+    }
+
+    LOG(INFO) << "ReadBlockAsync complete: "
+              << " Block-device: " << dm_block_device << " Partition-name: " << partition_name
+              << " Size: " << size;
+    return true;
+}
+
 void SnapshotHandler::ReadBlocksToCache(const std::string& dm_block_device,
                                         const std::string& partition_name, off_t offset,
                                         size_t size) {
@@ -344,17 +478,22 @@
         return;
     }
 
-    int num_threads = 2;
-    size_t num_blocks = dev_sz >> BLOCK_SHIFT;
-    size_t num_blocks_per_thread = num_blocks / num_threads;
-    size_t read_sz_per_thread = num_blocks_per_thread << BLOCK_SHIFT;
-    off_t offset = 0;
+    if (IsIouringSupported()) {
+        std::async(std::launch::async, &SnapshotHandler::ReadBlocksAsync, this, dm_block_device,
+                   partition_name, dev_sz);
+    } else {
+        int num_threads = 2;
+        size_t num_blocks = dev_sz >> BLOCK_SHIFT;
+        size_t num_blocks_per_thread = num_blocks / num_threads;
+        size_t read_sz_per_thread = num_blocks_per_thread << BLOCK_SHIFT;
+        off_t offset = 0;
 
-    for (int i = 0; i < num_threads; i++) {
-        std::async(std::launch::async, &SnapshotHandler::ReadBlocksToCache, this, dm_block_device,
-                   partition_name, offset, read_sz_per_thread);
+        for (int i = 0; i < num_threads; i++) {
+            std::async(std::launch::async, &SnapshotHandler::ReadBlocksToCache, this,
+                       dm_block_device, partition_name, offset, read_sz_per_thread);
 
-        offset += read_sz_per_thread;
+            offset += read_sz_per_thread;
+        }
     }
 }
 
@@ -513,5 +652,33 @@
     return ra_state;
 }
 
+bool SnapshotHandler::IsIouringSupported() {
+    struct utsname uts;
+    unsigned int major, minor;
+
+    if (android::base::GetBoolProperty("snapuserd.test.io_uring.force_disable", false)) {
+        SNAP_LOG(INFO) << "io_uring disabled for testing";
+        return false;
+    }
+
+    if ((uname(&uts) != 0) || (sscanf(uts.release, "%u.%u", &major, &minor) != 2)) {
+        SNAP_LOG(ERROR) << "Could not parse the kernel version from uname. "
+                        << " io_uring not supported";
+        return false;
+    }
+
+    // We will only support kernels from 5.6 onwards as IOSQE_ASYNC flag and
+    // IO_URING_OP_READ/WRITE opcodes were introduced only on 5.6 kernel
+    if (major >= 5) {
+        if (major == 5 && minor < 6) {
+            return false;
+        }
+    } else {
+        return false;
+    }
+
+    return android::base::GetBoolProperty("ro.virtual_ab.io_uring.enabled", false);
+}
+
 }  // namespace snapshot
 }  // namespace android
diff --git a/fs_mgr/libsnapshot/snapuserd/user-space-merge/snapuserd_core.h b/fs_mgr/libsnapshot/snapuserd/user-space-merge/snapuserd_core.h
index 1953316..b0f2d65 100644
--- a/fs_mgr/libsnapshot/snapuserd/user-space-merge/snapuserd_core.h
+++ b/fs_mgr/libsnapshot/snapuserd/user-space-merge/snapuserd_core.h
@@ -39,6 +39,7 @@
 #include <libdm/dm.h>
 #include <libsnapshot/cow_reader.h>
 #include <libsnapshot/cow_writer.h>
+#include <liburing.h>
 #include <snapuserd/snapuserd_buffer.h>
 #include <snapuserd/snapuserd_kernel.h>
 
@@ -113,6 +114,19 @@
     bool ReconstructDataFromCow();
     void CheckOverlap(const CowOperation* cow_op);
 
+    bool ReadAheadAsyncIO();
+    bool ReapIoCompletions(int pending_ios_to_complete);
+    bool ReadXorData(size_t block_index, size_t xor_op_index,
+                     std::vector<const CowOperation*>& xor_op_vec);
+    void ProcessXorData(size_t& block_xor_index, size_t& xor_index,
+                        std::vector<const CowOperation*>& xor_op_vec, void* buffer,
+                        loff_t& buffer_offset);
+    void UpdateScratchMetadata();
+
+    bool ReadAheadSyncIO();
+    bool InitializeIouring();
+    void FinalizeIouring();
+
     void* read_ahead_buffer_;
     void* metadata_buffer_;
 
@@ -131,7 +145,19 @@
     std::unordered_set<uint64_t> dest_blocks_;
     std::unordered_set<uint64_t> source_blocks_;
     bool overlap_;
+    std::vector<uint64_t> blocks_;
+    int total_blocks_merged_ = 0;
+    std::unique_ptr<uint8_t[]> ra_temp_buffer_;
+    std::unique_ptr<uint8_t[]> ra_temp_meta_buffer_;
     BufferSink bufsink_;
+
+    bool read_ahead_async_ = false;
+    // Queue depth of 32 seems optimal. We don't want
+    // to have a huge depth as it may put more memory pressure
+    // on the kernel worker threads given that we use
+    // IOSQE_ASYNC flag.
+    int queue_depth_ = 32;
+    std::unique_ptr<struct io_uring> ring_;
 };
 
 class Worker {
@@ -185,6 +211,7 @@
     // Merge related ops
     bool Merge();
     bool MergeOrderedOps(const std::unique_ptr<ICowOpIter>& cowop_iter);
+    bool MergeOrderedOpsAsync(const std::unique_ptr<ICowOpIter>& cowop_iter);
     bool MergeReplaceZeroOps(const std::unique_ptr<ICowOpIter>& cowop_iter);
     int PrepareMerge(uint64_t* source_offset, int* pending_ops,
                      const std::unique_ptr<ICowOpIter>& cowop_iter,
@@ -193,6 +220,9 @@
     sector_t ChunkToSector(chunk_t chunk) { return chunk << CHUNK_SHIFT; }
     chunk_t SectorToChunk(sector_t sector) { return sector >> CHUNK_SHIFT; }
 
+    bool InitializeIouring();
+    void FinalizeIouring();
+
     std::unique_ptr<CowReader> reader_;
     BufferSink bufsink_;
     XorSink xorsink_;
@@ -208,6 +238,14 @@
     unique_fd base_path_merge_fd_;
     unique_fd ctrl_fd_;
 
+    bool merge_async_ = false;
+    // Queue depth of 32 seems optimal. We don't want
+    // to have a huge depth as it may put more memory pressure
+    // on the kernel worker threads given that we use
+    // IOSQE_ASYNC flag.
+    int queue_depth_ = 32;
+    std::unique_ptr<struct io_uring> ring_;
+
     std::shared_ptr<SnapshotHandler> snapuserd_;
 };
 
@@ -292,6 +330,8 @@
     bool GetRABuffer(std::unique_lock<std::mutex>* lock, uint64_t block, void* buffer);
     MERGE_GROUP_STATE ProcessMergingBlock(uint64_t new_block, void* buffer);
 
+    bool IsIouringSupported();
+
   private:
     bool ReadMetadata();
     sector_t ChunkToSector(chunk_t chunk) { return chunk << CHUNK_SHIFT; }
@@ -304,6 +344,11 @@
     void ReadBlocksToCache(const std::string& dm_block_device, const std::string& partition_name,
                            off_t offset, size_t size);
 
+    bool InitializeIouring(int io_depth);
+    void FinalizeIouring();
+    bool ReadBlocksAsync(const std::string& dm_block_device, const std::string& partition_name,
+                         size_t size);
+
     // COW device
     std::string cow_device_;
     // Source device
@@ -352,6 +397,8 @@
     bool attached_ = false;
     bool is_socket_present_;
     bool scratch_space_ = false;
+
+    std::unique_ptr<struct io_uring> ring_;
 };
 
 }  // namespace snapshot
diff --git a/fs_mgr/libsnapshot/snapuserd/user-space-merge/snapuserd_merge.cpp b/fs_mgr/libsnapshot/snapuserd/user-space-merge/snapuserd_merge.cpp
index fa055b7..d4d4efe 100644
--- a/fs_mgr/libsnapshot/snapuserd/user-space-merge/snapuserd_merge.cpp
+++ b/fs_mgr/libsnapshot/snapuserd/user-space-merge/snapuserd_merge.cpp
@@ -72,16 +72,16 @@
 }
 
 bool Worker::MergeReplaceZeroOps(const std::unique_ptr<ICowOpIter>& cowop_iter) {
-    // Flush every 2048 ops. Since all ops are independent and there is no
+    // Flush every 8192 ops. Since all ops are independent and there is no
     // dependency between COW ops, we will flush the data and the number
-    // of ops merged in COW file for every 2048 ops. If there is a crash,
+    // of ops merged in COW file for every 8192 ops. If there is a crash,
     // we will end up replaying some of the COW ops which were already merged.
     // That is ok.
     //
-    // Why 2048 ops ? We can probably increase this to bigger value but just
-    // need to ensure that merge makes forward progress if there are
-    // crashes repeatedly which is highly unlikely.
-    int total_ops_merged_per_commit = (PAYLOAD_BUFFER_SZ / BLOCK_SZ) * 8;
+    // Why 8192 ops ? Increasing this may improve merge time 3-4 seconds but
+    // we need to make sure that we checkpoint; 8k ops seems optimal. In-case
+    // if there is a crash merge should always make forward progress.
+    int total_ops_merged_per_commit = (PAYLOAD_BUFFER_SZ / BLOCK_SZ) * 32;
     int num_ops_merged = 0;
 
     while (!cowop_iter->Done()) {
@@ -128,7 +128,7 @@
 
         num_ops_merged += linear_blocks;
 
-        if (num_ops_merged == total_ops_merged_per_commit) {
+        if (num_ops_merged >= total_ops_merged_per_commit) {
             // Flush the data
             if (fsync(base_path_merge_fd_.get()) < 0) {
                 SNAP_LOG(ERROR) << "Merge: ReplaceZeroOps: Failed to fsync merged data";
@@ -172,6 +172,173 @@
     return true;
 }
 
+bool Worker::MergeOrderedOpsAsync(const std::unique_ptr<ICowOpIter>& cowop_iter) {
+    void* mapped_addr = snapuserd_->GetMappedAddr();
+    void* read_ahead_buffer =
+            static_cast<void*>((char*)mapped_addr + snapuserd_->GetBufferDataOffset());
+    size_t block_index = 0;
+
+    SNAP_LOG(INFO) << "MergeOrderedOpsAsync started....";
+
+    while (!cowop_iter->Done()) {
+        const CowOperation* cow_op = &cowop_iter->Get();
+        if (!IsOrderedOp(*cow_op)) {
+            break;
+        }
+
+        SNAP_LOG(DEBUG) << "Waiting for merge begin...";
+        // Wait for RA thread to notify that the merge window
+        // is ready for merging.
+        if (!snapuserd_->WaitForMergeBegin()) {
+            snapuserd_->SetMergeFailed(block_index);
+            return false;
+        }
+
+        snapuserd_->SetMergeInProgress(block_index);
+
+        loff_t offset = 0;
+        int num_ops = snapuserd_->GetTotalBlocksToMerge();
+
+        int pending_sqe = queue_depth_;
+        int pending_ios_to_submit = 0;
+        bool flush_required = false;
+
+        SNAP_LOG(DEBUG) << "Merging copy-ops of size: " << num_ops;
+        while (num_ops) {
+            uint64_t source_offset;
+
+            int linear_blocks = PrepareMerge(&source_offset, &num_ops, cowop_iter);
+
+            if (linear_blocks != 0) {
+                size_t io_size = (linear_blocks * BLOCK_SZ);
+
+                // Get an SQE entry from the ring and populate the I/O variables
+                struct io_uring_sqe* sqe = io_uring_get_sqe(ring_.get());
+                if (!sqe) {
+                    SNAP_PLOG(ERROR) << "io_uring_get_sqe failed during merge-ordered ops";
+                    snapuserd_->SetMergeFailed(block_index);
+                    return false;
+                }
+
+                io_uring_prep_write(sqe, base_path_merge_fd_.get(),
+                                    (char*)read_ahead_buffer + offset, io_size, source_offset);
+
+                offset += io_size;
+                num_ops -= linear_blocks;
+
+                pending_sqe -= 1;
+                pending_ios_to_submit += 1;
+                sqe->flags |= IOSQE_ASYNC;
+            }
+
+            // Ring is full or no more COW ops to be merged in this batch
+            if (pending_sqe == 0 || num_ops == 0 || (linear_blocks == 0 && pending_ios_to_submit)) {
+                // If this is a last set of COW ops to be merged in this batch, we need
+                // to sync the merged data. We will try to grab an SQE entry
+                // and set the FSYNC command; additionally, make sure that
+                // the fsync is done after all the I/O operations queued
+                // in the ring is completed by setting IOSQE_IO_DRAIN.
+                //
+                // If there is no space in the ring, we will flush it later
+                // by explicitly calling fsync() system call.
+                if (num_ops == 0 || (linear_blocks == 0 && pending_ios_to_submit)) {
+                    if (pending_sqe != 0) {
+                        struct io_uring_sqe* sqe = io_uring_get_sqe(ring_.get());
+                        if (!sqe) {
+                            // very unlikely but let's continue and not fail the
+                            // merge - we will flush it later
+                            SNAP_PLOG(ERROR) << "io_uring_get_sqe failed during merge-ordered ops";
+                            flush_required = true;
+                        } else {
+                            io_uring_prep_fsync(sqe, base_path_merge_fd_.get(), 0);
+                            // Drain the queue before fsync
+                            io_uring_sqe_set_flags(sqe, IOSQE_IO_DRAIN);
+                            pending_sqe -= 1;
+                            flush_required = false;
+                            pending_ios_to_submit += 1;
+                            sqe->flags |= IOSQE_ASYNC;
+                        }
+                    } else {
+                        flush_required = true;
+                    }
+                }
+
+                // Submit the IO for all the COW ops in a single syscall
+                int ret = io_uring_submit(ring_.get());
+                if (ret != pending_ios_to_submit) {
+                    SNAP_PLOG(ERROR)
+                            << "io_uring_submit failed for read-ahead: "
+                            << " io submit: " << ret << " expected: " << pending_ios_to_submit;
+                    snapuserd_->SetMergeFailed(block_index);
+                    return false;
+                }
+
+                int pending_ios_to_complete = pending_ios_to_submit;
+                pending_ios_to_submit = 0;
+
+                // Reap I/O completions
+                while (pending_ios_to_complete) {
+                    struct io_uring_cqe* cqe;
+
+                    ret = io_uring_wait_cqe(ring_.get(), &cqe);
+                    if (ret) {
+                        SNAP_LOG(ERROR) << "Read-ahead - io_uring_wait_cqe failed: " << ret;
+                        snapuserd_->SetMergeFailed(block_index);
+                        return false;
+                    }
+
+                    if (cqe->res < 0) {
+                        SNAP_LOG(ERROR)
+                                << "Read-ahead - io_uring_Wait_cqe failed with res: " << cqe->res;
+                        snapuserd_->SetMergeFailed(block_index);
+                        return false;
+                    }
+
+                    io_uring_cqe_seen(ring_.get(), cqe);
+                    pending_ios_to_complete -= 1;
+                }
+
+                pending_sqe = queue_depth_;
+            }
+
+            if (linear_blocks == 0) {
+                break;
+            }
+        }
+
+        // Verify all ops are merged
+        CHECK(num_ops == 0);
+
+        // Flush the data
+        if (flush_required && (fsync(base_path_merge_fd_.get()) < 0)) {
+            SNAP_LOG(ERROR) << " Failed to fsync merged data";
+            snapuserd_->SetMergeFailed(block_index);
+            return false;
+        }
+
+        // Merge is done and data is on disk. Update the COW Header about
+        // the merge completion
+        if (!snapuserd_->CommitMerge(snapuserd_->GetTotalBlocksToMerge())) {
+            SNAP_LOG(ERROR) << " Failed to commit the merged block in the header";
+            snapuserd_->SetMergeFailed(block_index);
+            return false;
+        }
+
+        SNAP_LOG(DEBUG) << "Block commit of size: " << snapuserd_->GetTotalBlocksToMerge();
+        // Mark the block as merge complete
+        snapuserd_->SetMergeCompleted(block_index);
+
+        // Notify RA thread that the merge thread is ready to merge the next
+        // window
+        snapuserd_->NotifyRAForMergeReady();
+
+        // Get the next block
+        block_index += 1;
+    }
+
+    return true;
+}
+
 bool Worker::MergeOrderedOps(const std::unique_ptr<ICowOpIter>& cowop_iter) {
     void* mapped_addr = snapuserd_->GetMappedAddr();
     void* read_ahead_buffer =
@@ -260,15 +427,23 @@
 bool Worker::Merge() {
     std::unique_ptr<ICowOpIter> cowop_iter = reader_->GetMergeOpIter();
 
-    // Start with Copy and Xor ops
-    if (!MergeOrderedOps(cowop_iter)) {
-        SNAP_LOG(ERROR) << "Merge failed for ordered ops";
-        snapuserd_->MergeFailed();
-        return false;
+    if (merge_async_) {
+        if (!MergeOrderedOpsAsync(cowop_iter)) {
+            SNAP_LOG(ERROR) << "Merge failed for ordered ops";
+            snapuserd_->MergeFailed();
+            return false;
+        }
+        SNAP_LOG(INFO) << "MergeOrderedOpsAsync completed.....";
+    } else {
+        // Start with Copy and Xor ops
+        if (!MergeOrderedOps(cowop_iter)) {
+            SNAP_LOG(ERROR) << "Merge failed for ordered ops";
+            snapuserd_->MergeFailed();
+            return false;
+        }
+        SNAP_LOG(INFO) << "MergeOrderedOps completed.....";
     }
 
-    SNAP_LOG(INFO) << "MergeOrderedOps completed...";
-
     // Replace and Zero ops
     if (!MergeReplaceZeroOps(cowop_iter)) {
         SNAP_LOG(ERROR) << "Merge failed for replace/zero ops";
@@ -281,6 +456,31 @@
     return true;
 }
 
+bool Worker::InitializeIouring() {
+    if (!snapuserd_->IsIouringSupported()) {
+        return false;
+    }
+
+    ring_ = std::make_unique<struct io_uring>();
+
+    int ret = io_uring_queue_init(queue_depth_, ring_.get(), 0);
+    if (ret) {
+        LOG(ERROR) << "Merge: io_uring_queue_init failed with ret: " << ret;
+        return false;
+    }
+
+    merge_async_ = true;
+
+    LOG(INFO) << "Merge: io_uring initialized with queue depth: " << queue_depth_;
+    return true;
+}
+
+void Worker::FinalizeIouring() {
+    if (merge_async_) {
+        io_uring_queue_exit(ring_.get());
+    }
+}
+
 bool Worker::RunMergeThread() {
     SNAP_LOG(DEBUG) << "Waiting for merge begin...";
     if (!snapuserd_->WaitForMergeBegin()) {
@@ -296,10 +496,13 @@
         return false;
     }
 
+    InitializeIouring();
+
     if (!Merge()) {
         return false;
     }
 
+    FinalizeIouring();
     CloseFds();
     reader_->CloseCowFd();
 
diff --git a/fs_mgr/libsnapshot/snapuserd/user-space-merge/snapuserd_readahead.cpp b/fs_mgr/libsnapshot/snapuserd/user-space-merge/snapuserd_readahead.cpp
index 9e8ccfb..26c5f19 100644
--- a/fs_mgr/libsnapshot/snapuserd/user-space-merge/snapuserd_readahead.cpp
+++ b/fs_mgr/libsnapshot/snapuserd/user-space-merge/snapuserd_readahead.cpp
@@ -183,25 +183,311 @@
     return true;
 }
 
-bool ReadAhead::ReadAheadIOStart() {
-    // Check if the data has to be constructed from the COW file.
-    // This will be true only once during boot up after a crash
-    // during merge.
-    if (snapuserd_->ShouldReconstructDataFromCow()) {
-        return ReconstructDataFromCow();
-    }
+/*
+ * With io_uring, the data flow is slightly different.
+ *
+ * The data flow is as follows:
+ *
+ * 1: Queue the I/O requests to be read from backing source device.
+ * This is done by retrieving the SQE entry from ring and populating
+ * the SQE entry. Note that the I/O is not submitted yet.
+ *
+ * 2: Once the ring is full (aka queue_depth), we will submit all
+ * the queued I/O request with a single system call. This essentially
+ * cuts down "queue_depth" number of system calls to a single system call.
+ *
+ * 3: Once the I/O is submitted, user-space thread will now work
+ * on processing the XOR Operations. This happens in parallel when
+ * I/O requests are submitted to the kernel. This is ok because, for XOR
+ * operations, we first need to retrieve the compressed data form COW block
+ * device. Thus, we have offloaded the backing source I/O to the kernel
+ * and user-space is parallely working on fetching the data for XOR operations.
+ *
+ * 4: After the XOR operations are read from COW device, poll the completion
+ * queue for all the I/O submitted. If the I/O's were already completed,
+ * then user-space thread will just read the CQE requests from the ring
+ * without doing any system call. If none of the I/O were completed yet,
+ * user-space thread will do a system call and wait for I/O completions.
+ *
+ * Flow diagram:
+ *                                                    SQ-RING
+ *  SQE1 <----------- Fetch SQE1 Entry ---------- |SQE1||SQE2|SQE3|
+ *
+ *  SQE1  ------------ Populate SQE1 Entry ------> |SQE1-X||SQE2|SQE3|
+ *
+ *  SQE2 <----------- Fetch SQE2 Entry ---------- |SQE1-X||SQE2|SQE3|
+ *
+ *  SQE2  ------------ Populate SQE2 Entry ------> |SQE1-X||SQE2-X|SQE3|
+ *
+ *  SQE3 <----------- Fetch SQE3 Entry ---------- |SQE1-X||SQE2-X|SQE3|
+ *
+ *  SQE3  ------------ Populate SQE3 Entry ------> |SQE1-X||SQE2-X|SQE3-X|
+ *
+ *  Submit-IO ---------------------------------> |SQE1-X||SQE2-X|SQE3-X|
+ *     |                                                  |
+ *     |                                        Process I/O entries in kernel
+ *     |                                                  |
+ *  Retrieve XOR                                          |
+ *  data from COW                                         |
+ *     |                                                  |
+ *     |                                                  |
+ *  Fetch CQ completions
+ *     |                                              CQ-RING
+ *                                               |CQE1-X||CQE2-X|CQE3-X|
+ *                                                        |
+ *   CQE1 <------------Fetch CQE1 Entry          |CQE1||CQE2-X|CQE3-X|
+ *   CQE2 <------------Fetch CQE2 Entry          |CQE1||CQE2-|CQE3-X|
+ *   CQE3 <------------Fetch CQE3 Entry          |CQE1||CQE2-|CQE3-|
+ *    |
+ *    |
+ *  Continue Next set of operations in the RING
+ */
 
-    std::vector<uint64_t> blocks;
-
+bool ReadAhead::ReadAheadAsyncIO() {
     int num_ops = (snapuserd_->GetBufferDataSize()) / BLOCK_SZ;
     loff_t buffer_offset = 0;
-    int total_blocks_merged = 0;
+    total_blocks_merged_ = 0;
     overlap_ = false;
     dest_blocks_.clear();
     source_blocks_.clear();
+    blocks_.clear();
     std::vector<const CowOperation*> xor_op_vec;
 
-    auto ra_temp_buffer = std::make_unique<uint8_t[]>(snapuserd_->GetBufferDataSize());
+    int pending_sqe = queue_depth_;
+    int pending_ios_to_submit = 0;
+
+    size_t xor_op_index = 0;
+    size_t block_index = 0;
+
+    loff_t offset = 0;
+
+    bufsink_.ResetBufferOffset();
+
+    // Number of ops to be merged in this window. This is a fixed size
+    // except for the last window wherein the number of ops can be less
+    // than the size of the RA window.
+    while (num_ops) {
+        uint64_t source_offset;
+        struct io_uring_sqe* sqe;
+
+        int linear_blocks = PrepareNextReadAhead(&source_offset, &num_ops, blocks_, xor_op_vec);
+
+        if (linear_blocks != 0) {
+            size_t io_size = (linear_blocks * BLOCK_SZ);
+
+            // Get an SQE entry from the ring and populate the I/O variables
+            sqe = io_uring_get_sqe(ring_.get());
+            if (!sqe) {
+                SNAP_PLOG(ERROR) << "io_uring_get_sqe failed during read-ahead";
+                snapuserd_->ReadAheadIOFailed();
+                return false;
+            }
+
+            io_uring_prep_read(sqe, backing_store_fd_.get(),
+                               (char*)ra_temp_buffer_.get() + buffer_offset, io_size,
+                               source_offset);
+
+            buffer_offset += io_size;
+            num_ops -= linear_blocks;
+            total_blocks_merged_ += linear_blocks;
+
+            pending_sqe -= 1;
+            pending_ios_to_submit += 1;
+            sqe->flags |= IOSQE_ASYNC;
+        }
+
+        // pending_sqe == 0 : Ring is full
+        //
+        // num_ops == 0 : All the COW ops in this batch are processed - Submit
+        // pending I/O requests in the ring
+        //
+        // linear_blocks == 0 : All the COW ops processing is done. Submit
+        // pending I/O requests in the ring
+        if (pending_sqe == 0 || num_ops == 0 || (linear_blocks == 0 && pending_ios_to_submit)) {
+            // Submit the IO for all the COW ops in a single syscall
+            int ret = io_uring_submit(ring_.get());
+            if (ret != pending_ios_to_submit) {
+                SNAP_PLOG(ERROR) << "io_uring_submit failed for read-ahead: "
+                                 << " io submit: " << ret << " expected: " << pending_ios_to_submit;
+                snapuserd_->ReadAheadIOFailed();
+                return false;
+            }
+
+            int pending_ios_to_complete = pending_ios_to_submit;
+            pending_ios_to_submit = 0;
+
+            bool xor_processing_required = (xor_op_vec.size() > 0);
+
+            // Read XOR data from COW file in parallel when I/O's are in-flight
+            if (xor_processing_required && !ReadXorData(block_index, xor_op_index, xor_op_vec)) {
+                SNAP_LOG(ERROR) << "ReadXorData failed";
+                snapuserd_->ReadAheadIOFailed();
+                return false;
+            }
+
+            // Fetch I/O completions
+            if (!ReapIoCompletions(pending_ios_to_complete)) {
+                SNAP_LOG(ERROR) << "ReapIoCompletions failed";
+                snapuserd_->ReadAheadIOFailed();
+                return false;
+            }
+
+            // Retrieve XOR'ed data
+            if (xor_processing_required) {
+                ProcessXorData(block_index, xor_op_index, xor_op_vec, ra_temp_buffer_.get(),
+                               offset);
+            }
+
+            // All the I/O in the ring is processed.
+            pending_sqe = queue_depth_;
+        }
+
+        if (linear_blocks == 0) {
+            break;
+        }
+    }
+
+    // Done with merging ordered ops
+    if (RAIterDone() && total_blocks_merged_ == 0) {
+        return true;
+    }
+
+    CHECK(blocks_.size() == total_blocks_merged_);
+
+    UpdateScratchMetadata();
+
+    return true;
+}
+
+void ReadAhead::UpdateScratchMetadata() {
+    loff_t metadata_offset = 0;
+
+    struct ScratchMetadata* bm = reinterpret_cast<struct ScratchMetadata*>(
+            (char*)ra_temp_meta_buffer_.get() + metadata_offset);
+
+    bm->new_block = 0;
+    bm->file_offset = 0;
+
+    loff_t file_offset = snapuserd_->GetBufferDataOffset();
+
+    for (size_t block_index = 0; block_index < blocks_.size(); block_index++) {
+        uint64_t new_block = blocks_[block_index];
+        // Track the metadata blocks which are stored in scratch space
+        bm = reinterpret_cast<struct ScratchMetadata*>((char*)ra_temp_meta_buffer_.get() +
+                                                       metadata_offset);
+
+        bm->new_block = new_block;
+        bm->file_offset = file_offset;
+
+        metadata_offset += sizeof(struct ScratchMetadata);
+        file_offset += BLOCK_SZ;
+    }
+
+    // This is important - explicitly set the contents to zero. This is used
+    // when re-constructing the data after crash. This indicates end of
+    // reading metadata contents when re-constructing the data
+    bm = reinterpret_cast<struct ScratchMetadata*>((char*)ra_temp_meta_buffer_.get() +
+                                                   metadata_offset);
+    bm->new_block = 0;
+    bm->file_offset = 0;
+}
+
+bool ReadAhead::ReapIoCompletions(int pending_ios_to_complete) {
+    // Reap I/O completions
+    while (pending_ios_to_complete) {
+        struct io_uring_cqe* cqe;
+
+        int ret = io_uring_wait_cqe(ring_.get(), &cqe);
+        if (ret) {
+            SNAP_LOG(ERROR) << "Read-ahead - io_uring_wait_cqe failed: " << ret;
+            return false;
+        }
+
+        if (cqe->res < 0) {
+            SNAP_LOG(ERROR) << "Read-ahead - io_uring_Wait_cqe failed with res: " << cqe->res;
+            return false;
+        }
+
+        io_uring_cqe_seen(ring_.get(), cqe);
+        pending_ios_to_complete -= 1;
+    }
+
+    return true;
+}
+
+void ReadAhead::ProcessXorData(size_t& block_xor_index, size_t& xor_index,
+                               std::vector<const CowOperation*>& xor_op_vec, void* buffer,
+                               loff_t& buffer_offset) {
+    loff_t xor_buf_offset = 0;
+
+    while (block_xor_index < blocks_.size()) {
+        void* bufptr = static_cast<void*>((char*)buffer + buffer_offset);
+        uint64_t new_block = blocks_[block_xor_index];
+
+        if (xor_index < xor_op_vec.size()) {
+            const CowOperation* xor_op = xor_op_vec[xor_index];
+
+            // Check if this block is an XOR op
+            if (xor_op->new_block == new_block) {
+                // Pointer to the data read from base device
+                uint8_t* buffer = reinterpret_cast<uint8_t*>(bufptr);
+                // Get the xor'ed data read from COW device
+                uint8_t* xor_data = reinterpret_cast<uint8_t*>((char*)bufsink_.GetPayloadBufPtr() +
+                                                               xor_buf_offset);
+
+                for (size_t byte_offset = 0; byte_offset < BLOCK_SZ; byte_offset++) {
+                    buffer[byte_offset] ^= xor_data[byte_offset];
+                }
+
+                // Move to next XOR op
+                xor_index += 1;
+                xor_buf_offset += BLOCK_SZ;
+            }
+        }
+
+        buffer_offset += BLOCK_SZ;
+        block_xor_index += 1;
+    }
+
+    bufsink_.ResetBufferOffset();
+}
+
+bool ReadAhead::ReadXorData(size_t block_index, size_t xor_op_index,
+                            std::vector<const CowOperation*>& xor_op_vec) {
+    // Process the XOR ops in parallel - We will be reading data
+    // from COW file for XOR ops processing.
+    while (block_index < blocks_.size()) {
+        uint64_t new_block = blocks_[block_index];
+
+        if (xor_op_index < xor_op_vec.size()) {
+            const CowOperation* xor_op = xor_op_vec[xor_op_index];
+            if (xor_op->new_block == new_block) {
+                if (!reader_->ReadData(*xor_op, &bufsink_)) {
+                    SNAP_LOG(ERROR)
+                            << " ReadAhead - XorOp Read failed for block: " << xor_op->new_block;
+                    return false;
+                }
+
+                xor_op_index += 1;
+                bufsink_.UpdateBufferOffset(BLOCK_SZ);
+            }
+        }
+        block_index += 1;
+    }
+    return true;
+}
+
+bool ReadAhead::ReadAheadSyncIO() {
+    int num_ops = (snapuserd_->GetBufferDataSize()) / BLOCK_SZ;
+    loff_t buffer_offset = 0;
+    total_blocks_merged_ = 0;
+    overlap_ = false;
+    dest_blocks_.clear();
+    source_blocks_.clear();
+    blocks_.clear();
+    std::vector<const CowOperation*> xor_op_vec;
+
+    bufsink_.ResetBufferOffset();
 
     // Number of ops to be merged in this window. This is a fixed size
     // except for the last window wherein the number of ops can be less
@@ -209,7 +495,7 @@
     while (num_ops) {
         uint64_t source_offset;
 
-        int linear_blocks = PrepareNextReadAhead(&source_offset, &num_ops, blocks, xor_op_vec);
+        int linear_blocks = PrepareNextReadAhead(&source_offset, &num_ops, blocks_, xor_op_vec);
         if (linear_blocks == 0) {
             // No more blocks to read
             SNAP_LOG(DEBUG) << " Read-ahead completed....";
@@ -220,7 +506,7 @@
 
         // Read from the base device consecutive set of blocks in one shot
         if (!android::base::ReadFullyAtOffset(backing_store_fd_,
-                                              (char*)ra_temp_buffer.get() + buffer_offset, io_size,
+                                              (char*)ra_temp_buffer_.get() + buffer_offset, io_size,
                                               source_offset)) {
             SNAP_PLOG(ERROR) << "Ordered-op failed. Read from backing store: "
                              << backing_store_device_ << "at block :" << source_offset / BLOCK_SZ
@@ -233,21 +519,19 @@
         }
 
         buffer_offset += io_size;
-        total_blocks_merged += linear_blocks;
+        total_blocks_merged_ += linear_blocks;
         num_ops -= linear_blocks;
     }
 
     // Done with merging ordered ops
-    if (RAIterDone() && total_blocks_merged == 0) {
+    if (RAIterDone() && total_blocks_merged_ == 0) {
         return true;
     }
 
     loff_t metadata_offset = 0;
 
-    auto ra_temp_meta_buffer = std::make_unique<uint8_t[]>(snapuserd_->GetBufferMetadataSize());
-
     struct ScratchMetadata* bm = reinterpret_cast<struct ScratchMetadata*>(
-            (char*)ra_temp_meta_buffer.get() + metadata_offset);
+            (char*)ra_temp_meta_buffer_.get() + metadata_offset);
 
     bm->new_block = 0;
     bm->file_offset = 0;
@@ -255,12 +539,15 @@
     loff_t file_offset = snapuserd_->GetBufferDataOffset();
 
     loff_t offset = 0;
-    CHECK(blocks.size() == total_blocks_merged);
+    CHECK(blocks_.size() == total_blocks_merged_);
 
     size_t xor_index = 0;
-    for (size_t block_index = 0; block_index < blocks.size(); block_index++) {
-        void* bufptr = static_cast<void*>((char*)ra_temp_buffer.get() + offset);
-        uint64_t new_block = blocks[block_index];
+    BufferSink bufsink;
+    bufsink.Initialize(BLOCK_SZ * 2);
+
+    for (size_t block_index = 0; block_index < blocks_.size(); block_index++) {
+        void* bufptr = static_cast<void*>((char*)ra_temp_buffer_.get() + offset);
+        uint64_t new_block = blocks_[block_index];
 
         if (xor_index < xor_op_vec.size()) {
             const CowOperation* xor_op = xor_op_vec[xor_index];
@@ -268,17 +555,16 @@
             // Check if this block is an XOR op
             if (xor_op->new_block == new_block) {
                 // Read the xor'ed data from COW
-                if (!reader_->ReadData(*xor_op, &bufsink_)) {
+                if (!reader_->ReadData(*xor_op, &bufsink)) {
                     SNAP_LOG(ERROR)
                             << " ReadAhead - XorOp Read failed for block: " << xor_op->new_block;
                     snapuserd_->ReadAheadIOFailed();
                     return false;
                 }
-
                 // Pointer to the data read from base device
                 uint8_t* buffer = reinterpret_cast<uint8_t*>(bufptr);
                 // Get the xor'ed data read from COW device
-                uint8_t* xor_data = reinterpret_cast<uint8_t*>(bufsink_.GetPayloadBufPtr());
+                uint8_t* xor_data = reinterpret_cast<uint8_t*>(bufsink.GetPayloadBufPtr());
 
                 // Retrieve the original data
                 for (size_t byte_offset = 0; byte_offset < BLOCK_SZ; byte_offset++) {
@@ -292,7 +578,7 @@
 
         offset += BLOCK_SZ;
         // Track the metadata blocks which are stored in scratch space
-        bm = reinterpret_cast<struct ScratchMetadata*>((char*)ra_temp_meta_buffer.get() +
+        bm = reinterpret_cast<struct ScratchMetadata*>((char*)ra_temp_meta_buffer_.get() +
                                                        metadata_offset);
 
         bm->new_block = new_block;
@@ -308,11 +594,34 @@
     // This is important - explicitly set the contents to zero. This is used
     // when re-constructing the data after crash. This indicates end of
     // reading metadata contents when re-constructing the data
-    bm = reinterpret_cast<struct ScratchMetadata*>((char*)ra_temp_meta_buffer.get() +
+    bm = reinterpret_cast<struct ScratchMetadata*>((char*)ra_temp_meta_buffer_.get() +
                                                    metadata_offset);
     bm->new_block = 0;
     bm->file_offset = 0;
 
+    return true;
+}
+
+bool ReadAhead::ReadAheadIOStart() {
+    // Check if the data has to be constructed from the COW file.
+    // This will be true only once during boot up after a crash
+    // during merge.
+    if (snapuserd_->ShouldReconstructDataFromCow()) {
+        return ReconstructDataFromCow();
+    }
+
+    if (read_ahead_async_) {
+        if (!ReadAheadAsyncIO()) {
+            SNAP_LOG(ERROR) << "ReadAheadAsyncIO failed - io_uring processing failure.";
+            return false;
+        }
+    } else {
+        if (!ReadAheadSyncIO()) {
+            SNAP_LOG(ERROR) << "ReadAheadSyncIO failed";
+            return false;
+        }
+    }
+
     // Wait for the merge to finish for the previous RA window. We shouldn't
     // be touching the scratch space until merge is complete of previous RA
     // window. If there is a crash during this time frame, merge should resume
@@ -322,22 +631,22 @@
     }
 
     // Copy the data to scratch space
-    memcpy(metadata_buffer_, ra_temp_meta_buffer.get(), snapuserd_->GetBufferMetadataSize());
-    memcpy(read_ahead_buffer_, ra_temp_buffer.get(), total_blocks_merged * BLOCK_SZ);
+    memcpy(metadata_buffer_, ra_temp_meta_buffer_.get(), snapuserd_->GetBufferMetadataSize());
+    memcpy(read_ahead_buffer_, ra_temp_buffer_.get(), total_blocks_merged_ * BLOCK_SZ);
 
-    offset = 0;
+    loff_t offset = 0;
     std::unordered_map<uint64_t, void*>& read_ahead_buffer_map = snapuserd_->GetReadAheadMap();
     read_ahead_buffer_map.clear();
 
-    for (size_t block_index = 0; block_index < blocks.size(); block_index++) {
+    for (size_t block_index = 0; block_index < blocks_.size(); block_index++) {
         void* bufptr = static_cast<void*>((char*)read_ahead_buffer_ + offset);
-        uint64_t new_block = blocks[block_index];
+        uint64_t new_block = blocks_[block_index];
 
         read_ahead_buffer_map[new_block] = bufptr;
         offset += BLOCK_SZ;
     }
 
-    snapuserd_->SetMergedBlockCountForNextCommit(total_blocks_merged);
+    snapuserd_->SetMergedBlockCountForNextCommit(total_blocks_merged_);
 
     // Flush the data only if we have a overlapping blocks in the region
     // Notify the Merge thread to resume merging this window
@@ -350,6 +659,33 @@
     return true;
 }
 
+bool ReadAhead::InitializeIouring() {
+    if (!snapuserd_->IsIouringSupported()) {
+        return false;
+    }
+
+    ring_ = std::make_unique<struct io_uring>();
+
+    int ret = io_uring_queue_init(queue_depth_, ring_.get(), 0);
+    if (ret) {
+        SNAP_LOG(ERROR) << "io_uring_queue_init failed with ret: " << ret;
+        return false;
+    }
+
+    // For xor ops processing
+    bufsink_.Initialize(PAYLOAD_BUFFER_SZ * 2);
+    read_ahead_async_ = true;
+
+    SNAP_LOG(INFO) << "Read-ahead: io_uring initialized with queue depth: " << queue_depth_;
+    return true;
+}
+
+void ReadAhead::FinalizeIouring() {
+    if (read_ahead_async_) {
+        io_uring_queue_exit(ring_.get());
+    }
+}
+
 bool ReadAhead::RunThread() {
     if (!InitializeFds()) {
         return false;
@@ -363,14 +699,18 @@
 
     InitializeRAIter();
 
+    InitializeIouring();
+
     while (!RAIterDone()) {
         if (!ReadAheadIOStart()) {
             break;
         }
     }
 
+    FinalizeIouring();
     CloseFds();
     reader_->CloseCowFd();
+
     SNAP_LOG(INFO) << " ReadAhead thread terminating....";
     return true;
 }
@@ -434,8 +774,9 @@
     metadata_buffer_ =
             static_cast<void*>((char*)mapped_addr + snapuserd_->GetBufferMetadataOffset());
     read_ahead_buffer_ = static_cast<void*>((char*)mapped_addr + snapuserd_->GetBufferDataOffset());
-    // For xor ops
-    bufsink_.Initialize(PAYLOAD_BUFFER_SZ);
+
+    ra_temp_buffer_ = std::make_unique<uint8_t[]>(snapuserd_->GetBufferDataSize());
+    ra_temp_meta_buffer_ = std::make_unique<uint8_t[]>(snapuserd_->GetBufferMetadataSize());
 }
 
 }  // namespace snapshot
diff --git a/fs_mgr/libsnapshot/snapuserd/user-space-merge/snapuserd_test.cpp b/fs_mgr/libsnapshot/snapuserd/user-space-merge/snapuserd_test.cpp
index 1c3e04b..d670f1e 100644
--- a/fs_mgr/libsnapshot/snapuserd/user-space-merge/snapuserd_test.cpp
+++ b/fs_mgr/libsnapshot/snapuserd/user-space-merge/snapuserd_test.cpp
@@ -12,6 +12,9 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
+#include <android-base/strings.h>
+#include <gflags/gflags.h>
+
 #include <fcntl.h>
 #include <linux/fs.h>
 #include <linux/memfd.h>
@@ -27,6 +30,7 @@
 #include <string_view>
 
 #include <android-base/file.h>
+#include <android-base/properties.h>
 #include <android-base/unique_fd.h>
 #include <fs_mgr/file_wait.h>
 #include <gtest/gtest.h>
@@ -38,6 +42,8 @@
 
 #include "snapuserd_core.h"
 
+DEFINE_string(force_config, "", "Force testing mode with iouring disabled");
+
 namespace android {
 namespace snapshot {
 
@@ -857,5 +863,23 @@
 
 int main(int argc, char** argv) {
     ::testing::InitGoogleTest(&argc, argv);
-    return RUN_ALL_TESTS();
+
+    gflags::ParseCommandLineFlags(&argc, &argv, false);
+
+    android::base::SetProperty("ctl.stop", "snapuserd");
+
+    if (FLAGS_force_config == "iouring_disabled") {
+        if (!android::base::SetProperty("snapuserd.test.io_uring.force_disable", "1")) {
+            return testing::AssertionFailure()
+                   << "Failed to disable property: snapuserd.test.io_uring.disabled";
+        }
+    }
+
+    int ret = RUN_ALL_TESTS();
+
+    if (FLAGS_force_config == "iouring_disabled") {
+        android::base::SetProperty("snapuserd.test.io_uring.force_disable", "0");
+    }
+
+    return ret;
 }
diff --git a/healthd/OWNERS b/healthd/OWNERS
index d3f8758..e64c33d 100644
--- a/healthd/OWNERS
+++ b/healthd/OWNERS
@@ -1,2 +1 @@
 elsk@google.com
-hridya@google.com
diff --git a/init/builtins.cpp b/init/builtins.cpp
index cc445be..0eb894b 100644
--- a/init/builtins.cpp
+++ b/init/builtins.cpp
@@ -117,7 +117,7 @@
                         android::base::GetMinimumLogSeverity() > android::base::DEBUG) {}
 
     template <typename T>
-    operator android::base::expected<T, ResultError<int>>() {
+    operator android::base::expected<T, ResultError<android::base::Errno>>() {
         if (ignore_error_) {
             return {};
         }
diff --git a/libcutils/fs_config.cpp b/libcutils/fs_config.cpp
index e9497a8..a6835fc 100644
--- a/libcutils/fs_config.cpp
+++ b/libcutils/fs_config.cpp
@@ -211,6 +211,7 @@
     { 00755, AID_ROOT,      AID_ROOT,      0, "first_stage_ramdisk/system/bin/resize2fs" },
     { 00755, AID_ROOT,      AID_ROOT,      0, "first_stage_ramdisk/system/bin/snapuserd" },
     { 00755, AID_ROOT,      AID_ROOT,      0, "first_stage_ramdisk/system/bin/tune2fs" },
+    { 00755, AID_ROOT,      AID_ROOT,      0, "first_stage_ramdisk/system/bin/fsck.f2fs" },
     // generic defaults
     { 00755, AID_ROOT,      AID_ROOT,      0, "bin/*" },
     { 00640, AID_ROOT,      AID_SHELL,     0, "fstab.*" },
diff --git a/libqtaguid/Android.bp b/libqtaguid/Android.bp
deleted file mode 100644
index 64db095..0000000
--- a/libqtaguid/Android.bp
+++ /dev/null
@@ -1,60 +0,0 @@
-//
-// Copyright (C) 2017 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.
-//
-
-package {
-    default_applicable_licenses: ["Android-Apache-2.0"],
-}
-
-cc_library_headers {
-    name: "libqtaguid_headers",
-    vendor_available: false,
-    host_supported: false,
-    export_include_dirs: ["include"],
-    target: {
-        linux_bionic: {
-            enabled: true,
-        },
-    },
-}
-
-cc_library {
-    name: "libqtaguid",
-    vendor_available: false,
-    host_supported: false,
-    target: {
-        android: {
-            srcs: [
-                "qtaguid.c",
-            ],
-            sanitize: {
-                misc_undefined: ["integer"],
-            },
-        },
-    },
-
-    shared_libs: ["liblog"],
-    header_libs: [
-        "libqtaguid_headers",
-    ],
-    export_header_lib_headers: ["libqtaguid_headers"],
-    local_include_dirs: ["include"],
-
-    cflags: [
-        "-Werror",
-        "-Wall",
-        "-Wextra",
-    ],
-}
diff --git a/libqtaguid/include/qtaguid/qtaguid.h b/libqtaguid/include/qtaguid/qtaguid.h
deleted file mode 100644
index 72285e5..0000000
--- a/libqtaguid/include/qtaguid/qtaguid.h
+++ /dev/null
@@ -1,62 +0,0 @@
-/*
- * Copyright (C) 2011 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.
- */
-
-#ifndef __LEGACY_QTAGUID_H
-#define __LEGACY_QTAGUID_H
-
-#include <stdint.h>
-#include <sys/types.h>
-#include <unistd.h>
-
-#ifdef __cplusplus
-extern "C" {
-#endif
-
-/*
- * Set tags (and owning UIDs) for network sockets. The socket must be untagged
- * by calling qtaguid_untagSocket() before closing it, otherwise the qtaguid
- * module will keep a reference to it even after close.
- */
-extern int legacy_tagSocket(int sockfd, int tag, uid_t uid);
-
-/*
- * Untag a network socket before closing.
- */
-extern int legacy_untagSocket(int sockfd);
-
-/*
- * For the given uid, switch counter sets.
- * The kernel only keeps a limited number of sets.
- * 2 for now.
- */
-extern int legacy_setCounterSet(int counterSetNum, uid_t uid);
-
-/*
- * Delete all tag info that relates to the given tag an uid.
- * If the tag is 0, then ALL info about the uid is freeded.
- * The delete data also affects active tagged socketd, which are
- * then untagged.
- * The calling process can only operate on its own tags.
- * Unless it is part of the happy AID_NET_BW_ACCT group.
- * In which case it can clobber everything.
- */
-extern int legacy_deleteTagData(int tag, uid_t uid);
-
-#ifdef __cplusplus
-}
-#endif
-
-#endif /* __LEGACY_QTAGUID_H */
diff --git a/libqtaguid/qtaguid.c b/libqtaguid/qtaguid.c
deleted file mode 100644
index cd38bad..0000000
--- a/libqtaguid/qtaguid.c
+++ /dev/null
@@ -1,143 +0,0 @@
-/*
-** Copyright 2011, 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.
-*/
-
-// #define LOG_NDEBUG 0
-
-#define LOG_TAG "qtaguid"
-
-#include <errno.h>
-#include <fcntl.h>
-#include <inttypes.h>
-#include <pthread.h>
-#include <stdio.h>
-#include <string.h>
-#include <unistd.h>
-
-#include <log/log.h>
-#include <qtaguid/qtaguid.h>
-
-static const char* CTRL_PROCPATH = "/proc/net/xt_qtaguid/ctrl";
-static const int CTRL_MAX_INPUT_LEN = 128;
-
-/*
- * One per proccess.
- * Once the device is open, this process will have its socket tags tracked.
- * And on exit or untimely death, all socket tags will be removed.
- * A process can only open /dev/xt_qtaguid once.
- * It should not close it unless it is really done with all the socket tags.
- * Failure to open it will be visible when socket tagging will be attempted.
- */
-static int resTrackFd = -1;
-pthread_once_t resTrackInitDone = PTHREAD_ONCE_INIT;
-
-/* Only call once per process. */
-void legacy_resTrack(void) {
-    resTrackFd = TEMP_FAILURE_RETRY(open("/dev/xt_qtaguid", O_RDONLY | O_CLOEXEC));
-}
-
-/*
- * Returns:
- *   0 on success.
- *   -errno on failure.
- */
-static int write_ctrl(const char* cmd) {
-    int fd, res, savedErrno;
-
-    ALOGV("write_ctrl(%s)", cmd);
-
-    fd = TEMP_FAILURE_RETRY(open(CTRL_PROCPATH, O_WRONLY | O_CLOEXEC));
-    if (fd < 0) {
-        return -errno;
-    }
-
-    res = TEMP_FAILURE_RETRY(write(fd, cmd, strlen(cmd)));
-    if (res < 0) {
-        savedErrno = errno;
-    } else {
-        savedErrno = 0;
-    }
-    if (res < 0) {
-        // ALOGV is enough because all the callers also log failures
-        ALOGV("Failed write_ctrl(%s) res=%d errno=%d", cmd, res, savedErrno);
-    }
-    close(fd);
-    return -savedErrno;
-}
-
-int legacy_tagSocket(int sockfd, int tag, uid_t uid) {
-    char lineBuf[CTRL_MAX_INPUT_LEN];
-    int res;
-    uint64_t kTag = ((uint64_t)tag << 32);
-
-    pthread_once(&resTrackInitDone, legacy_resTrack);
-
-    snprintf(lineBuf, sizeof(lineBuf), "t %d %" PRIu64 " %d", sockfd, kTag, uid);
-
-    ALOGV("Tagging socket %d with tag %" PRIx64 "{%u,0} for uid %d", sockfd, kTag, tag, uid);
-
-    res = write_ctrl(lineBuf);
-    if (res < 0) {
-        ALOGI("Tagging socket %d with tag %" PRIx64 "(%d) for uid %d failed errno=%d", sockfd, kTag,
-              tag, uid, res);
-    }
-
-    return res;
-}
-
-int legacy_untagSocket(int sockfd) {
-    char lineBuf[CTRL_MAX_INPUT_LEN];
-    int res;
-
-    ALOGV("Untagging socket %d", sockfd);
-
-    snprintf(lineBuf, sizeof(lineBuf), "u %d", sockfd);
-    res = write_ctrl(lineBuf);
-    if (res < 0) {
-        ALOGI("Untagging socket %d failed errno=%d", sockfd, res);
-    }
-
-    return res;
-}
-
-int legacy_setCounterSet(int counterSetNum, uid_t uid) {
-    char lineBuf[CTRL_MAX_INPUT_LEN];
-    int res;
-
-    ALOGV("Setting counters to set %d for uid %d", counterSetNum, uid);
-
-    snprintf(lineBuf, sizeof(lineBuf), "s %d %d", counterSetNum, uid);
-    res = write_ctrl(lineBuf);
-    return res;
-}
-
-int legacy_deleteTagData(int tag, uid_t uid) {
-    char lineBuf[CTRL_MAX_INPUT_LEN];
-    int cnt = 0, res = 0;
-    uint64_t kTag = (uint64_t)tag << 32;
-
-    ALOGV("Deleting tag data with tag %" PRIx64 "{%d,0} for uid %d", kTag, tag, uid);
-
-    pthread_once(&resTrackInitDone, legacy_resTrack);
-
-    snprintf(lineBuf, sizeof(lineBuf), "d %" PRIu64 " %d", kTag, uid);
-    res = write_ctrl(lineBuf);
-    if (res < 0) {
-        ALOGI("Deleting tag data with tag %" PRIx64 "/%d for uid %d failed with cnt=%d errno=%d",
-              kTag, tag, uid, cnt, errno);
-    }
-
-    return res;
-}
diff --git a/libutils/Android.bp b/libutils/Android.bp
index bda9d6b..234638b 100644
--- a/libutils/Android.bp
+++ b/libutils/Android.bp
@@ -28,16 +28,18 @@
     min_sdk_version: "apex_inherit",
 
     header_libs: [
-        "liblog_headers",
-        "libsystem_headers",
+        "libbase_headers",
         "libcutils_headers",
+        "liblog_headers",
         "libprocessgroup_headers",
+        "libsystem_headers",
     ],
     export_header_lib_headers: [
-        "liblog_headers",
-        "libsystem_headers",
+        "libbase_headers",
         "libcutils_headers",
+        "liblog_headers",
         "libprocessgroup_headers",
+        "libsystem_headers",
     ],
     export_include_dirs: ["include"],
 
@@ -299,13 +301,14 @@
 
     srcs: [
         "BitSet_test.cpp",
+        "Errors_test.cpp",
         "FileMap_test.cpp",
         "LruCache_test.cpp",
         "Mutex_test.cpp",
         "SharedBuffer_test.cpp",
         "Singleton_test.cpp",
-        "String8_test.cpp",
         "String16_test.cpp",
+        "String8_test.cpp",
         "StrongPointer_test.cpp",
         "Timers_test.cpp",
         "Unicode_test.cpp",
@@ -364,6 +367,7 @@
         "-Wall",
         "-Werror",
     ],
+    header_libs: ["libutils_headers"],
 }
 
 cc_test_library {
@@ -376,6 +380,7 @@
         "-Werror",
     ],
     shared_libs: ["libutils_test_singleton1"],
+    header_libs: ["libutils_headers"],
 }
 
 cc_benchmark {
diff --git a/libutils/Errors_test.cpp b/libutils/Errors_test.cpp
new file mode 100644
index 0000000..873c994
--- /dev/null
+++ b/libutils/Errors_test.cpp
@@ -0,0 +1,110 @@
+/*
+ * Copyright (C) 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.
+ */
+
+#include "utils/ErrorsMacros.h"
+
+#include <android-base/result.h>
+
+#include <gtest/gtest.h>
+
+using namespace android;
+
+using android::base::Error;
+using android::base::Result;
+
+status_t success_or_fail(bool success) {
+    if (success)
+        return OK;
+    else
+        return PERMISSION_DENIED;
+}
+
+TEST(errors, unwrap_or_return) {
+    auto f = [](bool success, int* val) -> status_t {
+        OR_RETURN(success_or_fail(success));
+        *val = 10;
+        return OK;
+    };
+
+    int val;
+    status_t s = f(true, &val);
+    EXPECT_EQ(OK, s);
+    EXPECT_EQ(10, val);
+
+    val = 0;  // reset
+    status_t q = f(false, &val);
+    EXPECT_EQ(PERMISSION_DENIED, q);
+    EXPECT_EQ(0, val);
+}
+
+TEST(errors, unwrap_or_return_result) {
+    auto f = [](bool success) -> Result<std::string, StatusT> {
+        OR_RETURN(success_or_fail(success));
+        return "hello";
+    };
+
+    auto r = f(true);
+    EXPECT_TRUE(r.ok());
+    EXPECT_EQ("hello", *r);
+
+    auto s = f(false);
+    EXPECT_FALSE(s.ok());
+    EXPECT_EQ(PERMISSION_DENIED, s.error().code());
+    EXPECT_EQ("PERMISSION_DENIED", s.error().message());
+}
+
+TEST(errors, unwrap_or_return_result_int) {
+    auto f = [](bool success) -> Result<int, StatusT> {
+        OR_RETURN(success_or_fail(success));
+        return 10;
+    };
+
+    auto r = f(true);
+    EXPECT_TRUE(r.ok());
+    EXPECT_EQ(10, *r);
+
+    auto s = f(false);
+    EXPECT_FALSE(s.ok());
+    EXPECT_EQ(PERMISSION_DENIED, s.error().code());
+    EXPECT_EQ("PERMISSION_DENIED", s.error().message());
+}
+
+TEST(errors, unwrap_or_fatal) {
+    OR_FATAL(success_or_fail(true));
+
+    EXPECT_DEATH(OR_FATAL(success_or_fail(false)), "PERMISSION_DENIED");
+}
+
+TEST(errors, result_in_status) {
+    auto f = [](bool success) -> Result<std::string, StatusT> {
+        if (success)
+            return "OK";
+        else
+            return Error<StatusT>(PERMISSION_DENIED) << "custom error message";
+    };
+
+    auto g = [&](bool success) -> status_t {
+        std::string val = OR_RETURN(f(success));
+        EXPECT_EQ("OK", val);
+        return OK;
+    };
+
+    status_t a = g(true);
+    EXPECT_EQ(OK, a);
+
+    status_t b = g(false);
+    EXPECT_EQ(PERMISSION_DENIED, b);
+}
diff --git a/libutils/Unicode_test.cpp b/libutils/Unicode_test.cpp
index b92eef8..8b994d9 100644
--- a/libutils/Unicode_test.cpp
+++ b/libutils/Unicode_test.cpp
@@ -100,7 +100,7 @@
         0xF0, 0x90, 0x80, 0x80, // U+10000, 2 UTF-16 character
     };
 
-    char16_t output[1 + 1 + 1 + 2 + 1]; // Room for NULL
+    char16_t output[1 + 1 + 1 + 2 + 1];  // Room for null
 
     utf8_to_utf16(str, sizeof(str), output, sizeof(output) / sizeof(output[0]));
 
@@ -114,8 +114,7 @@
             << "should be first half of surrogate U+10000";
     EXPECT_EQ(0xDC00, output[4])
             << "should be second half of surrogate U+10000";
-    EXPECT_EQ(NULL, output[5])
-            << "should be NULL terminated";
+    EXPECT_EQ(0, output[5]) << "should be null terminated";
 }
 
 TEST_F(UnicodeTest, strstr16EmptyTarget) {
diff --git a/libutils/include/utils/ErrorsMacros.h b/libutils/include/utils/ErrorsMacros.h
new file mode 100644
index 0000000..048c538
--- /dev/null
+++ b/libutils/include/utils/ErrorsMacros.h
@@ -0,0 +1,86 @@
+/*
+ * Copyright (C) 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 "Errors.h"
+
+// It would have been better if this file (ErrorsMacros.h) is entirely in utils/Errors.h. However
+// that is infeasible as some (actually many) are using utils/Errors.h via the implicit include path
+// `system/core/include` [1].  Since such users are not guaranteed to specify the dependency to
+// libbase_headers, the following headers from libbase_headers can't be found.
+// [1] build/soong/cc/config/global.go#commonGlobalIncludes
+#include <android-base/errors.h>
+#include <android-base/result.h>
+
+#include <assert.h>
+
+namespace android {
+
+// StatusT is a wrapper class for status_t. Use this type instead of status_t when instantiating
+// Result<T, E> and Error<E> template classes. This is required to distinguish status_t from
+// other integer-based error code types like errno, and also to provide utility functions like
+// print().
+struct StatusT {
+    StatusT() : val_(OK) {}
+    StatusT(status_t s) : val_(s) {}
+    const status_t& value() const { return val_; }
+    operator status_t() const { return val_; }
+    std::string print() const { return statusToString(val_); }
+
+    status_t val_;
+};
+
+namespace base {
+
+// Specialization of android::base::OkOrFail<V> for V = status_t. This is used to use the OR_RETURN
+// and OR_FATAL macros with statements that yields a value of status_t. See android-base/errors.h
+// for the detailed contract.
+template <>
+struct OkOrFail<status_t> {
+    // Tests if status_t is a success value of not.
+    static bool IsOk(const status_t& s) { return s == OK; }
+
+    // Unwrapping status_t in the success case is just asserting that it is actually a success.
+    // We don't return OK because it would be redundant.
+    static void Unwrap([[maybe_unused]] status_t&& s) { assert(IsOk(s)); }
+
+    // Consumes status_t when it's a fail value
+    static OkOrFail<status_t> Fail(status_t&& s) {
+        assert(!IsOk(s));
+        return OkOrFail<status_t>{s};
+    }
+    status_t val_;
+
+    // And converts back into status_t. This is used when OR_RETURN is used in a function whose
+    // return type is status_t.
+    operator status_t() && { return val_; }
+
+    // Or converts into Result<T, StatusT>. This is used when OR_RETURN is used in a function whose
+    // return type is Result<T, StatusT>.
+    template <typename T, typename = std::enable_if_t<!std::is_same_v<T, status_t>>>
+    operator Result<T, StatusT>() && {
+        return Error<StatusT>(std::move(val_));
+    }
+
+    operator Result<int, StatusT>() && { return Error<StatusT>(std::move(val_)); }
+
+    // String representation of the error value.
+    static std::string ErrorMessage(const status_t& s) { return statusToString(s); }
+};
+
+}  // namespace base
+}  // namespace android
diff --git a/rootdir/Android.mk b/rootdir/Android.mk
index ce2ec0e..2ed9eec 100644
--- a/rootdir/Android.mk
+++ b/rootdir/Android.mk
@@ -75,9 +75,14 @@
   EXPORT_GLOBAL_GCOV_OPTIONS := export GCOV_PREFIX /data/misc/trace
 endif
 
-EXPORT_GLOBAL_CLANG_COVERAGE_OPTIONS :=
 ifeq ($(CLANG_COVERAGE),true)
-  EXPORT_GLOBAL_CLANG_COVERAGE_OPTIONS := export LLVM_PROFILE_FILE /data/misc/trace/clang-%20m.profraw
+  ifeq ($(BIONIC_COVERAGE),false)
+    # http://b/210012154 Disable continuous coverage if instrumentation is on
+    # for bionic/libc
+    EXPORT_GLOBAL_CLANG_COVERAGE_OPTIONS := export LLVM_PROFILE_FILE /data/misc/trace/clang%c-%20m.profraw
+  else
+    EXPORT_GLOBAL_CLANG_COVERAGE_OPTIONS := export LLVM_PROFILE_FILE /data/misc/trace/clang-%20m.profraw
+  endif
 endif
 
 # Put it here instead of in init.rc module definition,
diff --git a/rootdir/avb/Android.bp b/rootdir/avb/Android.bp
deleted file mode 100644
index cfc59a7..0000000
--- a/rootdir/avb/Android.bp
+++ /dev/null
@@ -1,31 +0,0 @@
-package {
-    default_applicable_licenses: ["Android-Apache-2.0"],
-}
-
-filegroup {
-    name: "q-gsi_avbpubkey",
-    srcs: [
-        "q-gsi.avbpubkey",
-    ],
-}
-
-filegroup {
-    name: "r-gsi_avbpubkey",
-    srcs: [
-        "r-gsi.avbpubkey",
-    ],
-}
-
-filegroup {
-    name: "s-gsi_avbpubkey",
-    srcs: [
-        "s-gsi.avbpubkey",
-    ],
-}
-
-filegroup {
-    name: "qcar-gsi_avbpubkey",
-    srcs: [
-        "qcar-gsi.avbpubkey",
-    ],
-}
diff --git a/rootdir/avb/Android.mk b/rootdir/avb/Android.mk
index 647cfa2..8cf3172 100644
--- a/rootdir/avb/Android.mk
+++ b/rootdir/avb/Android.mk
@@ -15,19 +15,6 @@
 endif
 
 #######################################
-# q-gsi.avbpubkey
-include $(CLEAR_VARS)
-
-LOCAL_MODULE := q-gsi.avbpubkey
-LOCAL_LICENSE_KINDS := SPDX-license-identifier-Apache-2.0
-LOCAL_LICENSE_CONDITIONS := notice
-LOCAL_MODULE_CLASS := ETC
-LOCAL_SRC_FILES := $(LOCAL_MODULE)
-LOCAL_MODULE_PATH := $(my_gsi_avb_keys_path)
-
-include $(BUILD_PREBUILT)
-
-#######################################
 # q-developer-gsi.avbpubkey
 include $(CLEAR_VARS)
 
@@ -41,19 +28,6 @@
 include $(BUILD_PREBUILT)
 
 #######################################
-# r-gsi.avbpubkey
-include $(CLEAR_VARS)
-
-LOCAL_MODULE := r-gsi.avbpubkey
-LOCAL_LICENSE_KINDS := SPDX-license-identifier-Apache-2.0
-LOCAL_LICENSE_CONDITIONS := notice
-LOCAL_MODULE_CLASS := ETC
-LOCAL_SRC_FILES := $(LOCAL_MODULE)
-LOCAL_MODULE_PATH := $(my_gsi_avb_keys_path)
-
-include $(BUILD_PREBUILT)
-
-#######################################
 # r-developer-gsi.avbpubkey
 include $(CLEAR_VARS)
 
@@ -67,19 +41,6 @@
 include $(BUILD_PREBUILT)
 
 #######################################
-# s-gsi.avbpubkey
-include $(CLEAR_VARS)
-
-LOCAL_MODULE := s-gsi.avbpubkey
-LOCAL_LICENSE_KINDS := SPDX-license-identifier-Apache-2.0
-LOCAL_LICENSE_CONDITIONS := notice
-LOCAL_MODULE_CLASS := ETC
-LOCAL_SRC_FILES := $(LOCAL_MODULE)
-LOCAL_MODULE_PATH := $(my_gsi_avb_keys_path)
-
-include $(BUILD_PREBUILT)
-
-#######################################
 # s-developer-gsi.avbpubkey
 include $(CLEAR_VARS)
 
@@ -92,17 +53,4 @@
 
 include $(BUILD_PREBUILT)
 
-#######################################
-# qcar-gsi.avbpubkey
-include $(CLEAR_VARS)
-
-LOCAL_MODULE := qcar-gsi.avbpubkey
-LOCAL_LICENSE_KINDS := SPDX-license-identifier-Apache-2.0
-LOCAL_LICENSE_CONDITIONS := notice
-LOCAL_MODULE_CLASS := ETC
-LOCAL_SRC_FILES := $(LOCAL_MODULE)
-LOCAL_MODULE_PATH := $(my_gsi_avb_keys_path)
-
-include $(BUILD_PREBUILT)
-
 my_gsi_avb_keys_path :=
diff --git a/rootdir/avb/q-gsi.avbpubkey b/rootdir/avb/q-gsi.avbpubkey
deleted file mode 100644
index 5ed7543..0000000
--- a/rootdir/avb/q-gsi.avbpubkey
+++ /dev/null
Binary files differ
diff --git a/rootdir/avb/qcar-gsi.avbpubkey b/rootdir/avb/qcar-gsi.avbpubkey
deleted file mode 100644
index ce56646..0000000
--- a/rootdir/avb/qcar-gsi.avbpubkey
+++ /dev/null
Binary files differ
diff --git a/rootdir/avb/r-gsi.avbpubkey b/rootdir/avb/r-gsi.avbpubkey
deleted file mode 100644
index 2609b30..0000000
--- a/rootdir/avb/r-gsi.avbpubkey
+++ /dev/null
Binary files differ
diff --git a/rootdir/avb/s-gsi.avbpubkey b/rootdir/avb/s-gsi.avbpubkey
deleted file mode 100644
index 9065fb8..0000000
--- a/rootdir/avb/s-gsi.avbpubkey
+++ /dev/null
Binary files differ