snapuserd: Read partition blocks to memory

When the device is rebooted after OTA is done,
daemon will read through all the dynamic partitions
to bring the blocks to memory. When update-verifier
runs, all the blocks would be in page-cache thereby
cutting down the boot time.

Boot time improvements on Pixel:

Full OTA:
===========

VABC (Without this patch): 37.308 seconds
VABC (With this patch): 28.604 seconds

Incremental OTA:
=================

VABC (Without this patch): 39.072 seconds
VABC (With this patch): 27.523 seconds

We read the blocks only during second stage transition. Thus,
it shouldn't impact when snapuserd is spin up during
post-install phase or during first-stage init.

Bug: 193863442
Test: Full and Incremental OTA on pixel
Signed-off-by: Akilesh Kailash <akailash@google.com>
Change-Id: Id654449238e22125f7d6288c7100fde512cc2ced
diff --git a/fs_mgr/libdm/dm.cpp b/fs_mgr/libdm/dm.cpp
index a5eda29..8e3d5a5 100644
--- a/fs_mgr/libdm/dm.cpp
+++ b/fs_mgr/libdm/dm.cpp
@@ -651,5 +651,61 @@
     return spec.target_type == "snapshot"s && data == "Overflow"s;
 }
 
+// Find directories in format of "/sys/block/dm-X".
+static int DmNameFilter(const dirent* de) {
+    if (android::base::StartsWith(de->d_name, "dm-")) {
+        return 1;
+    }
+    return 0;
+}
+
+std::map<std::string, std::string> DeviceMapper::FindDmPartitions() {
+    static constexpr auto DM_PATH_PREFIX = "/sys/block/";
+    dirent** namelist;
+    int n = scandir(DM_PATH_PREFIX, &namelist, DmNameFilter, alphasort);
+    if (n == -1) {
+        PLOG(ERROR) << "Failed to scan dir " << DM_PATH_PREFIX;
+        return {};
+    }
+    if (n == 0) {
+        LOG(ERROR) << "No dm block device found.";
+        free(namelist);
+        return {};
+    }
+
+    static constexpr auto DM_PATH_SUFFIX = "/dm/name";
+    static constexpr auto DEV_PATH = "/dev/block/";
+    std::map<std::string, std::string> dm_block_devices;
+    while (n--) {
+        std::string path = DM_PATH_PREFIX + std::string(namelist[n]->d_name) + DM_PATH_SUFFIX;
+        std::string content;
+        if (!android::base::ReadFileToString(path, &content)) {
+            PLOG(WARNING) << "Failed to read " << path;
+        } else {
+            std::string dm_block_name = android::base::Trim(content);
+            // AVB is using 'vroot' for the root block device but we're expecting 'system'.
+            if (dm_block_name == "vroot") {
+                dm_block_name = "system";
+            } else if (android::base::EndsWith(dm_block_name, "-verity")) {
+                auto npos = dm_block_name.rfind("-verity");
+                dm_block_name = dm_block_name.substr(0, npos);
+            } else if (!android::base::GetProperty("ro.boot.avb_version", "").empty()) {
+                // Verified Boot 1.0 doesn't add a -verity suffix. On AVB 2 devices,
+                // if DAP is enabled, then a -verity suffix must be used to
+                // differentiate between dm-linear and dm-verity devices. If we get
+                // here, we're AVB 2 and looking at a non-verity partition.
+                free(namelist[n]);
+                continue;
+            }
+
+            dm_block_devices.emplace(dm_block_name, DEV_PATH + std::string(namelist[n]->d_name));
+        }
+        free(namelist[n]);
+    }
+    free(namelist);
+
+    return dm_block_devices;
+}
+
 }  // namespace dm
 }  // namespace android
diff --git a/fs_mgr/libdm/include/libdm/dm.h b/fs_mgr/libdm/include/libdm/dm.h
index 8fcdf74..31c3003 100644
--- a/fs_mgr/libdm/include/libdm/dm.h
+++ b/fs_mgr/libdm/include/libdm/dm.h
@@ -17,6 +17,7 @@
 #ifndef _LIBDM_DM_H_
 #define _LIBDM_DM_H_
 
+#include <dirent.h>
 #include <fcntl.h>
 #include <linux/dm-ioctl.h>
 #include <linux/kdev_t.h>
@@ -26,6 +27,7 @@
 #include <unistd.h>
 
 #include <chrono>
+#include <map>
 #include <memory>
 #include <optional>
 #include <string>
@@ -255,6 +257,12 @@
     //  * A failure occurred.
     std::optional<std::string> GetParentBlockDeviceByPath(const std::string& path);
 
+    // Iterate the content over "/sys/block/dm-x/dm/name" and find
+    // all the dm-wrapped block devices.
+    //
+    // Returns mapping <partition-name, /dev/block/dm-x>
+    std::map<std::string, std::string> FindDmPartitions();
+
   private:
     // Maximum possible device mapper targets registered in the kernel.
     // This is only used to read the list of targets from kernel so we allocate
diff --git a/fs_mgr/libsnapshot/snapuserd/Android.bp b/fs_mgr/libsnapshot/snapuserd/Android.bp
index bc97afc..47268d4 100644
--- a/fs_mgr/libsnapshot/snapuserd/Android.bp
+++ b/fs_mgr/libsnapshot/snapuserd/Android.bp
@@ -78,6 +78,7 @@
         "liblog",
         "libsnapshot_cow",
         "libz",
+        "libext4_utils",
     ],
 }
 
@@ -121,6 +122,7 @@
         "libz",
         "libfs_mgr",
         "libdm",
+        "libext4_utils",
     ],
     header_libs: [
         "libstorage_literals_headers",
diff --git a/fs_mgr/libsnapshot/snapuserd/include/snapuserd/snapuserd_kernel.h b/fs_mgr/libsnapshot/snapuserd/include/snapuserd/snapuserd_kernel.h
index 6bb7a39..c592257 100644
--- a/fs_mgr/libsnapshot/snapuserd/include/snapuserd/snapuserd_kernel.h
+++ b/fs_mgr/libsnapshot/snapuserd/include/snapuserd/snapuserd_kernel.h
@@ -47,8 +47,6 @@
 static constexpr uint32_t CHUNK_SIZE = 8;
 static constexpr uint32_t CHUNK_SHIFT = (__builtin_ffs(CHUNK_SIZE) - 1);
 
-#define DIV_ROUND_UP(n, d) (((n) + (d)-1) / (d))
-
 // This structure represents the kernel COW header.
 // All the below fields should be in Little Endian format.
 struct disk_header {
diff --git a/fs_mgr/libsnapshot/snapuserd/snapuserd.cpp b/fs_mgr/libsnapshot/snapuserd/snapuserd.cpp
index 31d0221..b32dd8a 100644
--- a/fs_mgr/libsnapshot/snapuserd/snapuserd.cpp
+++ b/fs_mgr/libsnapshot/snapuserd/snapuserd.cpp
@@ -16,10 +16,22 @@
 
 #include "snapuserd.h"
 
+#include <dirent.h>
+#include <fcntl.h>
+#include <linux/fs.h>
+#include <unistd.h>
+#include <algorithm>
+
 #include <csignal>
 #include <optional>
 #include <set>
 
+#include <android-base/file.h>
+#include <android-base/logging.h>
+#include <android-base/parseint.h>
+#include <android-base/properties.h>
+#include <android-base/strings.h>
+#include <android-base/unique_fd.h>
 #include <snapuserd/snapuserd_client.h>
 
 namespace android {
@@ -659,6 +671,74 @@
     return ReadMetadata();
 }
 
+void Snapuserd::ReadBlocksToCache(const std::string& dm_block_device,
+                                  const std::string partition_name, off_t offset, size_t size) {
+    android::base::unique_fd fd(TEMP_FAILURE_RETRY(open(dm_block_device.c_str(), O_RDONLY)));
+    if (fd.get() == -1) {
+        SNAP_PLOG(ERROR) << "Error reading " << dm_block_device
+                         << " partition-name: " << partition_name;
+        return;
+    }
+
+    size_t remain = size;
+    off_t file_offset = offset;
+    // We pick 4M I/O size based on the fact that the current
+    // update_verifier has a similar I/O size.
+    size_t read_sz = 1024 * BLOCK_SZ;
+    std::vector<uint8_t> buf(read_sz);
+
+    while (remain > 0) {
+        size_t to_read = std::min(remain, read_sz);
+
+        if (!android::base::ReadFullyAtOffset(fd.get(), buf.data(), to_read, file_offset)) {
+            SNAP_PLOG(ERROR) << "Failed to read block from block device: " << dm_block_device
+                             << " at offset: " << file_offset
+                             << " partition-name: " << partition_name << " total-size: " << size
+                             << " remain_size: " << remain;
+            return;
+        }
+
+        file_offset += to_read;
+        remain -= to_read;
+    }
+
+    SNAP_LOG(INFO) << "Finished reading block-device: " << dm_block_device
+                   << " partition: " << partition_name << " size: " << size
+                   << " offset: " << offset;
+}
+
+void Snapuserd::ReadBlocks(const std::string partition_name, const std::string& dm_block_device) {
+    SNAP_LOG(DEBUG) << "Reading partition: " << partition_name
+                    << " Block-Device: " << dm_block_device;
+
+    uint64_t dev_sz = 0;
+
+    unique_fd fd(TEMP_FAILURE_RETRY(open(dm_block_device.c_str(), O_RDONLY | O_CLOEXEC)));
+    if (fd < 0) {
+        SNAP_LOG(ERROR) << "Cannot open block device";
+        return;
+    }
+
+    dev_sz = get_block_device_size(fd.get());
+    if (!dev_sz) {
+        SNAP_PLOG(ERROR) << "Could not determine block device size: " << dm_block_device;
+        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;
+
+    for (int i = 0; i < num_threads; i++) {
+        std::async(std::launch::async, &Snapuserd::ReadBlocksToCache, this, dm_block_device,
+                   partition_name, offset, read_sz_per_thread);
+
+        offset += read_sz_per_thread;
+    }
+}
+
 /*
  * Entry point to launch threads
  */
@@ -687,6 +767,39 @@
                 std::async(std::launch::async, &WorkerThread::RunThread, worker_threads_[i].get()));
     }
 
+    bool second_stage_init = true;
+
+    // We don't want to read the blocks during first stage init.
+    if (android::base::EndsWith(misc_name_, "-init") || is_socket_present_) {
+        second_stage_init = false;
+    }
+
+    if (second_stage_init) {
+        SNAP_LOG(INFO) << "Reading blocks to cache....";
+        auto& dm = DeviceMapper::Instance();
+        auto dm_block_devices = dm.FindDmPartitions();
+        if (dm_block_devices.empty()) {
+            SNAP_LOG(ERROR) << "No dm-enabled block device is found.";
+        } else {
+            auto parts = android::base::Split(misc_name_, "-");
+            std::string partition_name = parts[0];
+
+            const char* suffix_b = "_b";
+            const char* suffix_a = "_a";
+
+            partition_name.erase(partition_name.find_last_not_of(suffix_b) + 1);
+            partition_name.erase(partition_name.find_last_not_of(suffix_a) + 1);
+
+            if (dm_block_devices.find(partition_name) == dm_block_devices.end()) {
+                SNAP_LOG(ERROR) << "Failed to find dm block device for " << partition_name;
+            } else {
+                ReadBlocks(partition_name, dm_block_devices.at(partition_name));
+            }
+        }
+    } else {
+        SNAP_LOG(INFO) << "Not reading block device into cache";
+    }
+
     bool ret = true;
     for (auto& t : threads) {
         ret = t.get() && ret;
diff --git a/fs_mgr/libsnapshot/snapuserd/snapuserd.h b/fs_mgr/libsnapshot/snapuserd/snapuserd.h
index 95d2f77..f4c296a 100644
--- a/fs_mgr/libsnapshot/snapuserd/snapuserd.h
+++ b/fs_mgr/libsnapshot/snapuserd/snapuserd.h
@@ -38,6 +38,7 @@
 #include <android-base/logging.h>
 #include <android-base/stringprintf.h>
 #include <android-base/unique_fd.h>
+#include <ext4_utils/ext4_utils.h>
 #include <libdm/dm.h>
 #include <libsnapshot/cow_reader.h>
 #include <libsnapshot/cow_writer.h>
@@ -284,6 +285,7 @@
     // Total number of blocks to be merged in a given read-ahead buffer region
     void SetTotalRaBlocksMerged(int x) { total_ra_blocks_merged_ = x; }
     int GetTotalRaBlocksMerged() { return total_ra_blocks_merged_; }
+    void SetSocketPresent(bool socket) { is_socket_present_ = socket; }
 
   private:
     bool IsChunkIdMetadata(chunk_t chunk);
@@ -296,6 +298,10 @@
     bool IsBlockAligned(int read_size) { return ((read_size & (BLOCK_SZ - 1)) == 0); }
     struct BufferState* GetBufferState();
 
+    void ReadBlocks(const std::string partition_name, const std::string& dm_block_device);
+    void ReadBlocksToCache(const std::string& dm_block_device, const std::string partition_name,
+                           off_t offset, size_t size);
+
     std::string cow_device_;
     std::string backing_store_device_;
     std::string control_device_;
@@ -335,6 +341,7 @@
 
     bool merge_initiated_ = false;
     bool attached_ = false;
+    bool is_socket_present_;
 };
 
 }  // namespace snapshot
diff --git a/fs_mgr/libsnapshot/snapuserd/snapuserd_server.cpp b/fs_mgr/libsnapshot/snapuserd/snapuserd_server.cpp
index 672c13c..2f87557 100644
--- a/fs_mgr/libsnapshot/snapuserd/snapuserd_server.cpp
+++ b/fs_mgr/libsnapshot/snapuserd/snapuserd_server.cpp
@@ -227,6 +227,7 @@
 void SnapuserdServer::RunThread(std::shared_ptr<DmUserHandler> handler) {
     LOG(INFO) << "Entering thread for handler: " << handler->misc_name();
 
+    handler->snapuserd()->SetSocketPresent(is_socket_present_);
     if (!handler->snapuserd()->Start()) {
         LOG(ERROR) << " Failed to launch all worker threads";
     }
@@ -290,6 +291,7 @@
     }
 
     AddWatchedFd(sockfd_, POLLIN);
+    is_socket_present_ = true;
 
     // If started in first-stage init, the property service won't be online.
     if (access("/dev/socket/property_service", F_OK) == 0) {
diff --git a/fs_mgr/libsnapshot/snapuserd/snapuserd_server.h b/fs_mgr/libsnapshot/snapuserd/snapuserd_server.h
index 846f848..3b6ff15 100644
--- a/fs_mgr/libsnapshot/snapuserd/snapuserd_server.h
+++ b/fs_mgr/libsnapshot/snapuserd/snapuserd_server.h
@@ -96,6 +96,7 @@
     bool terminating_;
     volatile bool received_socket_signal_ = false;
     std::vector<struct pollfd> watched_fds_;
+    bool is_socket_present_ = false;
 
     std::mutex lock_;