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;
}
3、验证稀疏镜像升级
(1)拷贝升级包
将制作的稀疏镜像升级包通过HDC工具推进系统data目录,并修改文件名为updater.zip,路径如下:
/data/updater/updater.zip
(2)启动升级
系统启动后,在命令行工具输入如下命令启动升级
reboot updater:–update_package=/data/updater/updater.zip
输入命令后,系统重启,进入升级页面,等待升级完成。
本文介绍了OpenHarmony系统中实现稀疏镜像升级的方法,理解稀疏镜像原理及稀疏镜像还原方法可以快速在自己的系统中应用稀疏镜像升级,提高系统升级速度。
写在最后
有很多小伙伴不知道学习哪些鸿蒙开发技术?不知道需要重点掌握哪些鸿蒙应用开发知识点?而且学习时频繁踩坑,最终浪费大量时间。所以有一份实用的鸿蒙(HarmonyOS NEXT)文档用来跟着学习是非常有必要的。
这份鸿蒙(HarmonyOS NEXT)文档包含了鸿蒙开发必掌握的核心知识要点,内容包含了(ArkTS、ArkUI开发组件、Stage模型、多端部署、分布式应用开发、音频、视频、WebGL、OpenHarmony多媒体技术、Napi组件、OpenHarmony内核、OpenHarmony南向开发、鸿蒙项目实战等等)鸿蒙(HarmonyOS NEXT)技术知识点。
希望这一份鸿蒙学习文档能够给大家带来帮助,有需要的小伙伴自行领取,限时开源,先到先得~无套路领取!!
获取这份完整版高清学习路线,请点击→纯血版全套鸿蒙HarmonyOS学习文档
鸿蒙(HarmonyOS NEXT)5.0最新学习路线
有了路线图,怎么能没有学习文档呢,小编也准备了一份联合鸿蒙官方发布笔记整理收纳的一套系统性的鸿蒙(OpenHarmony )学习手册(共计1236页)与鸿蒙(OpenHarmony )开发入门教学视频,内容包含:ArkTS、ArkUI、Web开发、应用模型、资源分类…等知识点。
获取以上完整版高清学习路线,请点击→纯血版全套鸿蒙HarmonyOS学习文档
《鸿蒙 (OpenHarmony)开发入门教学视频》
《鸿蒙生态应用开发V3.0白皮书》
《鸿蒙 (OpenHarmony)开发基础到实战手册》
OpenHarmony北向、南向开发环境搭建
《鸿蒙开发基础》
●ArkTS语言
●安装DevEco Studio
●运用你的第一个ArkTS应用
●ArkUI声明式UI开发
.……
《鸿蒙开发进阶》
●Stage模型入门
●网络管理
●数据管理
●电话服务
●分布式应用开发
●通知与窗口管理
●多媒体技术
●安全技能
●任务管理
●WebGL
●国际化开发
●应用测试
●DFX面向未来设计
●鸿蒙系统移植和裁剪定制
……
《鸿蒙进阶实战》
●ArkTS实践
●UIAbility应用
●网络案例
……