稀疏镜像在OpenHarmony上的应用

一、稀疏镜像升级背景

常用系统镜像格式为原始镜像，即RAW格式。镜像体积比较大，在烧录固件或者升级固件时比较耗时，而且在移动设备升级过程时比较耗费流量。为此，将原始镜像用稀疏描述，可以大大地缩减镜像体积，省时省流量。

二、稀疏镜像原理

1、稀疏镜像的概念

原始镜像：即raw image，完整的ext4分区镜像，包含很多全零的无效填充区

稀疏镜像：即sparse image，将raw ext4进行稀疏描述，因此尺寸比较小，制作目录有多少文件就计算多少，没有全零填充

2、稀疏镜像格式

稀疏镜像数据格式：首先是sparse_header占用28byte，然后是12byte的chunk_header，同样这chunk_header的类型决定了后面跟着的数据，如果读到数据是0xCAC1意味着后面是本身的raw_data，如果是0xCAC3，则后面num为0，接着再0xCAC2意味着后面填充4byte的内容。

三、实现稀疏镜像升级方案

版本基线：

OpenAtom OpenHarmony（以下简称“OpenHarmony”） 3.1 Release

代码路径：

https://gitee.com/openharmony/docs/blob/master/zh-cn/release-notes/OpenHarmony-v3.1-release.md

1、稀疏镜像烧录

（1）生成稀疏格式镜像

有2种方法可以生成稀疏镜像：

1）修改文件build/ohos_var.gni中，sparse_image=true

2）编译命令增加--sparse-image字段，如./build.sh --product-name=xxx --sparse-image

（2）增加稀疏格式转换工具

在目录build/ohos/images/mkimage中增加文件img2simg，该工具用于编译完成后将raw镜像转换为sparse格式，并设置权限为777。

（3）编译后的镜像对比

编译出的镜像格式为sparse格式，镜像大小相比raw格式明显变小。

（4）烧录稀疏镜像

烧录稀疏镜像方法和烧录原始镜像方法一致。

稀疏镜像本身是不能直接挂载的，在烧录过程中通过uboot将稀疏格式镜像还原为原始镜像，然后写到磁盘中，系统启动后可挂载对应的镜像。

2、稀疏镜像升级

升级包采用稀疏镜像制作。

（1）修改升级包制作工具

官方升级包工具不支持生成稀疏镜像的升级包，修改升级包工具，生成稀疏格式的升级包。.\base\update\packaging_tools\image_class.py

按照上图所示注释代码

（2）生成稀疏镜像升级包

和全量镜像升级包制作方法一致。

参考：

https://gitee.com/openharmony/docs/blob/master/zh-cn/device-dev/subsystems/subsys-ota-guide.md#%E6%A0%87%E5%87%86%E7%B3%BB%E7%BB%9F%E5%8D%87%E7%BA%A7%E5%8C%85%E5%88%B6%E4%BD%9C

（3）适配updater组件中稀疏镜像功能

● 增加写稀疏镜像分支

.\base\update\updater\services\applypatch\data_writer.cpp

写数据函数CreateDataWriter增加写稀疏镜像分支

case WRITE_SPARSE:
{
    std::unique_ptr<SparseWriter> writer(std::make_unique<SparseWriter>(partitionName));
    return std::move(writer);
}

● 增加稀疏镜像类声明

.\base\update\updater\services\applypatch\raw_writer.h

增加稀疏镜像类声明及相关变量定义

typedef struct sparse_header {
  uint32_t  magic;      /* 0xed26ff3a */
  uint16_t  major_version;  /* (0x1) - reject images with higher major versions */
  uint16_t  minor_version;  /* (0x0) - allow images with higer minor versions */
  uint16_t  file_hdr_sz;    /* 28 bytes for first revision of the file format */
  uint16_t  chunk_hdr_sz;   /* 12 bytes for first revision of the file format */
  uint32_t  blk_sz;     /* block size in bytes, must be a multiple of 4 (4096) */
  uint32_t  total_blks; /* total blocks in the non-sparse output image */
  uint32_t  total_chunks;   /* total chunks in the sparse input image */
  uint32_t  image_checksum; /* CRC32 checksum of the original data, counting "don't care" */
                /* as 0. Standard 802.3 polynomial, use a Public Domain */
                /* table implementation */
} sparse_header_t;

#define SPARSE_HEADER_MAGIC 0xed26ff3a

#define CHUNK_TYPE_RAW      0xCAC1
#define CHUNK_TYPE_FILL     0xCAC2
#define CHUNK_TYPE_DONT_CARE    0xCAC3
#define CHUNK_TYPE_CRC32    0xCAC4

typedef struct chunk_header {
  uint16_t  chunk_type; /* 0xCAC1 -> raw; 0xCAC2 -> fill; 0xCAC3 -> don't care */
  uint16_t  reserved1;
  uint32_t  chunk_sz;   /* in blocks in output image */
  uint32_t  total_sz;   /* in bytes of chunk input file including chunk header and data */
} chunk_header_t;

class SparseWriter : public DataWriter {
public:
    virtual bool Write(const uint8_t *addr, size_t len, WriteMode mode, const std::string &partitionName);

    explicit SparseWriter(const std::string partitionName) : offset_(0), fd_(-1), partitionName_(partitionName) {}

    virtual ~SparseWriter()
    {
        offset_ = 0;
        if (fd_ > 0) {
            fsync(fd_);
            close(fd_);
        }
        fd_ = -1;
    }
private:
    int WriteInternal(int fd, const uint8_t *data, size_t len, const std::string &partitionName);

    SparseWriter(const SparseWriter&) = delete;

    const SparseWriter& operator=(const SparseWriter&) = delete;
    off64_t offset_;
    int fd_;
    std::string partitionName_;
};

● 增加稀疏镜像类实现

.\base\update\updater\services\applypatch\raw_writer.cpp

增加稀疏镜像类实现及相关变量定义，原有代码不变

bool SparseWriter::Write(const uint8_t *addr, size_t len, WriteMode mode, const std::string &partitionName)
{
    if (addr == nullptr) {
        LOG(ERROR) << "SparseWriter: invalid address.";
        return false;
    }
    if (len == 0) {
        LOG(INFO) << "SparseWriter: write length is 0, skip.";
        return false;
    }
    if (fd_ < 0) {
        fd_ = OpenPartition(partitionName_);
        if (fd_ < 0) {
            return false;
        }
    }

    UPDATER_CHECK_ONLY_RETURN(WriteInternal(fd_, addr, len, partitionName_) >= 0, return false);
    return true;
}


int SparseWriter::WriteInternal(int fd, const uint8_t *data, size_t len, const std::string &partitionName)
{
    uint32_t written = 0;
    sparse_header_t *sparse_header;
    chunk_header_t *chunk_header;
    unsigned int chunk;
    void *membuf = NULL;
    uint32_t *fill_buf = NULL;
    uint32_t fill_val;
    uint32_t bytes_written = 0;
    uint32_t total_bytes = 0;
    uint32_t blk = 0;
    uint32_t chunk_data_sz = 0;
    uint32_t blkcnt = 0;
    uint32_t blks = 0;
    uint32_t total_blocks = 0;
    uint32_t addr_offset = 0;
    uint32_t fill_buf_num_blks = 0;


    uint32_t block_size = 4096;
    uint32_t block_count = 524288;
    uint32_t i;
    uint32_t j;
    int ret = lseek64(fd, offset_, SEEK_SET);
    UPDATER_FILE_CHECK(ret != -1, "RawWriter: failed to seek file to " << offset_, return -1);
    fill_buf_num_blks = CONFIG_FASTBOOT_FLASH_FILLBUF_SIZE / block_size; 
    LOG(INFO) << "WriteInternal offset_ " << offset_;
    /* Read and skip over sparse image header */
    sparse_header = (sparse_header_t *)data;
    data += sparse_header->file_hdr_sz;
    if (sparse_header->file_hdr_sz > sizeof(sparse_header_t)) {
        /*
         * Skip the remaining bytes in a header that is longer than
         * we expected.
         */
        data += (sparse_header->file_hdr_sz - sizeof(sparse_header_t));
    }
    LOG(INFO) << "=== Sparse Image Header ===";
    LOG(INFO) << "magic: "  << sparse_header->magic;
    LOG(INFO) << "major_version: " << sparse_header->major_version;
    LOG(INFO) << "minor_version: " << sparse_header->minor_version;
    LOG(INFO) << "file_hdr_sz: " << sparse_header->file_hdr_sz;
    LOG(INFO) << "chunk_hdr_sz: " << sparse_header->chunk_hdr_sz;
    LOG(INFO) << "blk_sz: " << sparse_header->blk_sz;
    LOG(INFO) << "total_blks: " << sparse_header->total_blks;
    LOG(INFO) << "total_chunks: " << sparse_header->total_chunks;


    LOG(INFO) << "Flashing Sparse Image";
    blk = 0;
    for (chunk = 0; chunk < sparse_header->total_chunks; chunk++) {
        /* Read and skip over chunk header */
        chunk_header = (chunk_header_t *)data;
        data += sizeof(chunk_header_t);
        if (chunk_header->chunk_type != CHUNK_TYPE_RAW) 
        {
            LOG(INFO) << "=== Chunk Header ===";
            LOG(INFO) << "chunk_type: " << chunk_header->chunk_type;
            LOG(INFO) << "chunk_sz: " << chunk_header->chunk_sz;
            LOG(INFO) << "total_sz: " << chunk_header->total_sz;
        }
        if (sparse_header->chunk_hdr_sz > sizeof(chunk_header_t)) {
            /*
             * Skip the remaining bytes in a header that is longer
             * than we expected.
             */
            data += (sparse_header->chunk_hdr_sz -
                 sizeof(chunk_header_t));
        }
        chunk_data_sz = sparse_header->blk_sz * chunk_header->chunk_sz;
        blkcnt = chunk_data_sz / block_size;
        switch (chunk_header->chunk_type) {
        case CHUNK_TYPE_RAW:
            if (chunk_header->total_sz !=
                (sparse_header->chunk_hdr_sz + chunk_data_sz)) {
                LOG(ERROR) << "Bogus chunk size for chunk type Raw";
                return -1;
            }
            if (blk + blkcnt > 0 + block_count) {
                LOG(ERROR) << "Request would exceed partition size!";
                return -1;
            }
            addr_offset = blk * block_size;
            ret = lseek64(fd, offset_ + addr_offset, SEEK_SET);
            if (ret < 0) {
                LOG(ERROR) << "failed to seek file to " << addr_offset << " error=" << strerror(errno);
                return -1;
            }
            written = write(fd, data, blkcnt * block_size);
            if (written < 0) {
                LOG(ERROR) << "SparseWriter: failed to write data of len ";
                return -1;
            }
            total_bytes = total_bytes + blkcnt * block_size;
            blks = written / block_size;
            blk += blks;
            bytes_written += blkcnt * block_size;
            total_blocks += chunk_header->chunk_sz;
            data += chunk_data_sz;
            break;
        case CHUNK_TYPE_FILL:
            if (chunk_header->total_sz !=
                (sparse_header->chunk_hdr_sz + sizeof(uint32_t))) {
                LOG(ERROR) << "Bogus chunk size for chunk type FILL total_sz err " << chunk_header->total_sz << "\n";
                return -1;
            }
            ret = posix_memalign (&membuf, 64, 
                        ROUNDUP(
                        block_size * fill_buf_num_blks,
                        64));
            if (ret) {
                LOG(ERROR) << "posix_memalign:" << strerror (errno);
                return -1;
            }
            fill_buf = (uint32_t *)membuf;
            if (!fill_buf) {
                LOG(ERROR) << "Malloc failed for: CHUNK_TYPE_FILL";
                return -1;
            }
            fill_val = *(uint32_t *)data;
            data = data + sizeof(uint32_t);
            for (i = 0;
                 i < (block_size * fill_buf_num_blks /
                  sizeof(fill_val));
                 i++)
                fill_buf[i] = fill_val;
            if (blk + blkcnt > 0 + block_count) {
                LOG(ERROR) << "Request would exceed partition size!";
                return -1;
            }
            for (i = 0; i < blkcnt;) {
                j = blkcnt - i;
                if (j > fill_buf_num_blks)
                    j = fill_buf_num_blks;
                addr_offset = blk * block_size;
                ret = lseek64(fd, offset_ + addr_offset, SEEK_SET);
                if (ret < 0) {
                    LOG(ERROR) << "failed to lseek file to " << addr_offset << " error=" << strerror(errno);
                    return -1;
                }
                written = write(fd, fill_buf, j * block_size);
                if (written < 0) {
                    LOG(ERROR) << "SparseWriter: failed to write data of len ";
                    return -1;
                }
                total_bytes = total_bytes + j * block_size;
                blks = written / block_size;
                if (blks < j) {
                    LOG(ERROR) << "Write failed, block";
                    free(fill_buf);
                    return -1;
                }
                blk += blks;
                i += j;
            }
            bytes_written += blkcnt * block_size;
            total_blocks += chunk_data_sz / sparse_header->blk_sz;
            free(fill_buf);
            break;
        case CHUNK_TYPE_DONT_CARE:
            blk += blkcnt;
            total_blocks += chunk_header->chunk_sz;
            break;
        case CHUNK_TYPE_CRC32:
            if (chunk_header->total_sz !=
                sparse_header->chunk_hdr_sz) {
                LOG(ERROR) << "Bogus chunk size for chunk type CRC32 total_sz err " << chunk_header->total_sz;
                return -1;
            }
            total_blocks += chunk_header->chunk_sz;
            data += chunk_data_sz;
            break;
        default:
            LOG(INFO) << __func__ << ": Unknown chunk type: " << chunk_header->chunk_type;
            return -1;
        }
    }
    LOG(INFO) << "Wrote "<< chunk <<"blocks, expected to write " << sparse_header->total_blks << "blocks\n";
    LOG(INFO) << "........ wrote "<< bytes_written <<"bytes to " << partitionName << "\n";
    LOG(INFO) << "total_bytes=" << total_bytes;
    return 0;
}