M1ngdaXie/mini-docker
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

added image pull feature to pull from dockerhub

Browse Source
This commit is contained in:
M1ngdaXie
2026-07-26 15:48:06 +08:00
parent c7bbc92dbe
commit 5dee3d3bcb
2 changed files with 388 additions and 34 deletions
Download Patch File Download Diff File Expand all files Collapse all files
main.go
+94 -34
View File
@@ -6,9 +6,37 @@ import (
"os/exec"
"path/filepath"
"strconv"
"strings"
"syscall"
)
// ——— bridge ———
const bridgeName = "mdbr0"
const bridgeIP = "10.100.0.1"
const bridgeSubnet = "10.100.0.0/24"
func ensureBridge() {
if _, err := os.Stat("/sys/class/net/" + bridgeName); err == nil {
return
}
runCmd("ip", "link", "add", bridgeName, "type", "bridge")
runCmd("ip", "addr", "add", bridgeIP+"/24", "dev", bridgeName)
runCmd("ip", "link", "set", bridgeName, "up")
fmt.Printf("[bridge] %s created at %s\n", bridgeName, bridgeIP)
}
func allocateIP() string {
data, _ := os.ReadFile("/tmp/mdbr0-ip-last")
last := 1
if len(data) > 0 {
last, _ = strconv.Atoi(string(data))
}
next := last + 1
os.WriteFile("/tmp/mdbr0-ip-last", []byte(strconv.Itoa(next)), 0644)
return fmt.Sprintf("10.100.0.%d", next)
}
// 这个 demo 没有加 CLONE_NEWUSER(user namespace)。
// 加了之后容器里的"root"会映射成宿主机上一个无权限的普通用户,
// mount/mknod 这类系统调用会被内核直接拒绝——从根上堵死设备逃逸这条路。
@@ -34,11 +62,43 @@ func main() {
func run() {
args := os.Args[2:]
if len(args) == 0 {
args = []string{"/bin/sh"}
fmt.Fprintf(os.Stderr, "Usage: %s run <image>[:tag] [command...]\n", os.Args[0])
os.Exit(1)
}
// 解析 "alpine" 或 "python:3.11"
image := args[0]
tag := "latest"
if idx := strings.Index(image, ":"); idx != -1 {
tag = image[idx+1:]
image = image[:idx]
}
cmdArgs := args[1:]
// --no-entry: 跳过镜像的 entrypoint，直接跑用户命令
skipEntrypoint := false
if len(cmdArgs) > 0 && cmdArgs[0] == "--no-entry" {
skipEntrypoint = true
cmdArgs = cmdArgs[1:]
}
result, err := pullImage(image, tag)
must(err)
// 用户没指定命令，用镜像默认的
if len(cmdArgs) == 0 {
cmdArgs = result.Cmd
if len(cmdArgs) == 0 {
cmdArgs = []string{"/bin/sh"}
}
}
// 镜像有 entrypoint 且不跳过，就排在命令前面
if !skipEntrypoint {
cmdArgs = append(result.Entrypoint, cmdArgs...)
}
hostPID := os.Getpid()
fmt.Printf("[run] Running %v as PID %d\n", args, hostPID)
fmt.Printf("[run] %s:%s → %v as PID %d\n", image, tag, cmdArgs, hostPID)
overlayBase := fmt.Sprintf("/tmp/maingo-%d", hostPID)
upperDir := overlayBase + "/upper"
@@ -48,50 +108,52 @@ func run() {
vethCtr := fmt.Sprintf("veth1-%d", hostPID)
cgroupName := fmt.Sprintf("maingo-%d", hostPID)
ipSuffix := hostPID % 253
hostIP := fmt.Sprintf("10.0.%d.1", ipSuffix)
ctrIP := fmt.Sprintf("10.0.%d.2", ipSuffix)
subnet := fmt.Sprintf("10.0.%d.0/24", ipSuffix)
ctrIP := allocateIP()
must(os.MkdirAll(upperDir, 0755))
must(os.MkdirAll(workDir, 0755))
must(os.MkdirAll(mergedDir, 0755))
// 用 defer 而不是"cmd.Wait() 之后顺序执行"：
// must(cmd.Wait()) 在子进程非 0 退出时会 panic,
// 之前的写法一旦这里 panic，后面的清理代码一行都不会跑，
// 会在宿主机上留下孤儿 veth / iptables 规则 / cgroup / 临时目录。
// defer 保证不管中间从哪里 panic，清理都会按 LIFO 顺序跑完。
defer safeRemoveAll(overlayBase)
defer cleanupCgroup(cgroupName)
defer cleanupVeth(vethHost)
defer cleanupNAT(subnet)
r, w, err := os.Pipe()
must(err)
cmd := exec.Command("/proc/self/exe", append([]string{"child"}, args...)...)
cmd := exec.Command("/proc/self/exe", append([]string{"child"}, cmdArgs...)...)
cmd.Stdin = os.Stdin
cmd.Stdout = os.Stdout
cmd.Stderr = os.Stderr
cmd.ExtraFiles = []*os.File{r}
cmd.SysProcAttr = &syscall.SysProcAttr{
Cloneflags: syscall.CLONE_NEWUTS |
Cloneflags: syscall.CLONE_NEWUSER |
syscall.CLONE_NEWUTS |
syscall.CLONE_NEWPID |
syscall.CLONE_NEWNS |
syscall.CLONE_NEWNET,
}
cmd.Env = append(os.Environ(),
cmd.SysProcAttr.UidMappings = []syscall.SysProcIDMap{
{ContainerID: 0, HostID: 0, Size: 65536},
}
cmd.SysProcAttr.GidMappings = []syscall.SysProcIDMap{
{ContainerID: 0, HostID: 0, Size: 65536},
}
// 先加镜像的环境变量，再拼我们自己的——镜像的 Env 可能覆盖宿主机的
cmd.Env = append(os.Environ(), result.Env...)
cmd.Env = append(cmd.Env,
"MAINGO_MERGED="+mergedDir,
"MAINGO_UPPER="+upperDir,
"MAINGO_WORK="+workDir,
"MAINGO_LOWER="+result.LowerDirs,
"MAINGO_CGROUP="+cgroupName)
must(cmd.Start())
r.Close()
setupVeth(cmd.Process.Pid, vethHost, vethCtr, hostIP, ctrIP, subnet)
setupNAT(cmd.Process.Pid, ctrIP, subnet)
ensureBridge()
setupVeth(cmd.Process.Pid, vethHost, vethCtr, ctrIP)
setupNAT()
w.Write([]byte{1})
w.Close()
@@ -120,7 +182,7 @@ func child() {
upperDir := os.Getenv("MAINGO_UPPER")
workDir := os.Getenv("MAINGO_WORK")
lowerDir := "/home/mingdax/alpine-rootfs"
lowerDir := os.Getenv("MAINGO_LOWER")
opts := fmt.Sprintf("lowerdir=%s,upperdir=%s,workdir=%s", lowerDir, upperDir, workDir)
must(syscall.Mount("overlay", mergedDir, "overlay", 0, opts))
fmt.Printf("[overlay] mounted\n")
@@ -140,7 +202,7 @@ func child() {
must(syscall.Unmount("proc", 0))
}
func setupVeth(childPID int, vethHost, vethCtr, hostIP, ctrIP, subnet string) {
func setupVeth(childPID int, vethHost, vethCtr, ctrIP string) {
pid := strconv.Itoa(childPID)
runCmdSilent("ip", "link", "del", vethHost)
@@ -148,37 +210,35 @@ func setupVeth(childPID int, vethHost, vethCtr, hostIP, ctrIP, subnet string) {
runCmd("ip", "link", "add", vethHost, "type", "veth", "peer", "name", vethCtr)
runCmd("ip", "link", "set", vethCtr, "netns", pid)
runCmd("ip", "addr", "add", hostIP+"/24", "dev", vethHost)
// 宿主端插进网桥，不配 IP——bridge 自己有一个 IP 当网关
runCmd("ip", "link", "set", vethHost, "master", bridgeName)
runCmd("ip", "link", "set", vethHost, "up")
// 容器端：配 IP、设默认网关、启 lo
runCmd("nsenter", "-t", pid, "-n", "ip", "addr", "add", ctrIP+"/24", "dev", vethCtr)
runCmd("nsenter", "-t", pid, "-n", "ip", "link", "set", vethCtr, "up")
runCmd("nsenter", "-t", pid, "-n", "ip", "link", "set", "lo", "up")
runCmd("nsenter", "-t", pid, "-n", "ip", "route", "add", "default", "via", bridgeIP)
fmt.Printf("[veth] pair ready: host=%s, container=%s\n", hostIP, ctrIP)
fmt.Printf("[veth] %s → %s via %s\n", ctrIP, bridgeName, bridgeIP)
}
func setupNAT(childPID int, ctrIP, subnet string) {
pid := strconv.Itoa(childPID)
// setupNAT 确保共享子网能出公网。规则只加一次，多个容器复用。
func setupNAT() {
must(os.WriteFile("/proc/sys/net/ipv4/ip_forward", []byte("1"), 0644))
runCmd("iptables", "-t", "nat", "-A", "POSTROUTING", "-s", subnet, "-j", "MASQUERADE")
runCmd("nsenter", "-t", pid, "-n", "ip", "route", "add", "default", "via",
ctrIP[:len(ctrIP)-1]+"1")
fmt.Printf("[nat] NAT enabled: %s → internet\n", subnet)
if _, err := os.Stat("/tmp/mdbr0-nat-ready"); err == nil {
return
}
runCmd("iptables", "-t", "nat", "-A", "POSTROUTING", "-s", bridgeSubnet, "-j", "MASQUERADE")
os.WriteFile("/tmp/mdbr0-nat-ready", []byte("1"), 0644)
fmt.Printf("[nat] MASQUERADE %s → internet\n", bridgeSubnet)
}
func cleanupVeth(vethHost string) {
runCmdSilent("ip", "link", "del", vethHost)
}
func cleanupNAT(subnet string) {
runCmdSilent("iptables", "-t", "nat", "-D", "POSTROUTING", "-s", subnet, "-j",
"MASQUERADE")
fmt.Printf("[nat] cleaned up\n")
}
func waitForNetwork() {
f := os.NewFile(3, "syncpipe")
buf := make([]byte, 1)
pull.go
+294
View File
@@ -0,0 +1,294 @@
package main
import (
"archive/tar"
"compress/gzip"
"encoding/json"
"fmt"
"io"
"net/http"
"os"
"path/filepath"
"strings"
)
// ——— 1. 数据结构：把 curl 看到的 JSON 翻译成 Go struct ———
//
// curl /manifests/latest 返回的 JSON：
//
// { "manifests": [ { "digest": "sha256:xxx", "platform": {"architecture":"amd64","os":"linux"} }, ... ] }
//
// 这个叫 manifest list（多架构索引）。
type ManifestList struct {
SchemaVersion int `json:"schemaVersion"`
Manifests []ManifestEntry `json:"manifests"`
}
type ManifestEntry struct {
Digest string `json:"digest"` // 指向真正 manifest 的 sha256
Platform Platform `json:"platform"` // 比如 amd64/linux
}
type Platform struct {
Architecture string `json:"architecture"`
OS string `json:"os"`
}
// curl /manifests/sha256:79ff19... 返回的 JSON：
//
// { "config": {"digest":"sha256:xxx"}, "layers": [{"digest":"sha256:yyy"}] }
//
// 这个是单架构的 manifest——里面有 config 和 layers。
type ImageManifest struct {
SchemaVersion int `json:"schemaVersion"`
Config BlobRef `json:"config"` // 指向 config blob
Layers []BlobRef `json:"layers"` // 文件系统层，索引 0 是最底层
}
// BlobRef 是个"指针"——用 sha256 digest 指向 registry 上的一个文件。
// config blob 和每一层 layer blob 都用这个表示。
type BlobRef struct {
Digest string `json:"digest"`
Size int64 `json:"size"`
}
// curl /blobs/sha256:d529dd0c...（config blob）返回的 JSON：
//
// { "config": { "Env": ["PATH=..."], "Cmd": ["/bin/sh"] } }
//
// 只有 Env 和 Cmd 是我们需要的。
// ImageConfig 是 config blob 的 Go 表示。
// Docker 里实际执行的命令 = Entrypoint（如果有）+ Cmd
// 比如 redis: Entrypoint=["docker-entrypoint.sh"], Cmd=["redis-server"]
//
// → 实际执行: docker-entrypoint.sh redis-server
type ImageConfig struct {
Config struct {
Env []string `json:"Env"`
Cmd []string `json:"Cmd"`
Entrypoint []string `json:"Entrypoint"`
} `json:"config"`
}
// PullResult 是 pullImage() 的返回值——run() 只需要这三样
type PullResult struct {
LowerDirs string // 冒号分隔：/tmp/layers/.../sha1:/tmp/layers/.../sha2
Env []string // PATH, PYTHON_VERSION... 直接拼进 cmd.Env
Cmd []string // 镜像默认命令，用户没指定时用
Entrypoint []string // 镜像入口脚本，如果有就排在 Cmd 前面
}
// ——— 2. 注册表 HTTP 客户端 ———
// getToken 向 Docker Hub 要一个临时 token。
// 免费公开镜像不需要登录——传 repo 名就行，返回一个 Bearer token。
// 所有后续请求都要带 Authorization: Bearer <token>
func getToken(repo string) (string, error) {
url := fmt.Sprintf(
"https://auth.docker.io/token?service=registry.docker.io&scope=repository:%s:pull",
repo,
)
resp, err := http.Get(url)
if err != nil {
return "", fmt.Errorf("token request: %w", err)
}
defer resp.Body.Close()
var out struct {
Token string `json:"token"`
}
if err := json.NewDecoder(resp.Body).Decode(&out); err != nil {
return "", fmt.Errorf("token decode: %w", err)
}
return out.Token, nil
}
// doRequest 封装了"带 token 和 Accept 头发一个 GET，把 JSON 解析到 target"。
// manifest list、manifest、config blob 共用这个函数——它们都是 JSON。
// layer blob 不走这个（layer 是 tar.gz 二进制）。
func doRequest(url, token, accept string, target interface{}) error {
req, err := http.NewRequest("GET", url, nil)
if err != nil {
return err
}
req.Header.Set("Authorization", "Bearer "+token)
if accept != "" {
req.Header.Set("Accept", accept)
}
resp, err := http.DefaultClient.Do(req)
if err != nil {
return fmt.Errorf("%s: %w", url, err)
}
defer resp.Body.Close()
if resp.StatusCode != http.StatusOK {
return fmt.Errorf("%s: status %d", url, resp.StatusCode)
}
return json.NewDecoder(resp.Body).Decode(target)
}
// ——— 3. pullImage ——
// 输入 "alpine" + "latest" → 输出 PullResult{LowerDirs, Env, Cmd}
//
// 流程：
// 1. 解析镜像名（"alpine" → "library/alpine"）
// 2. 拿 token
// 3. GET /manifests/latest → ManifestList → 挑 amd64/linux
// 4. GET /manifests/sha256:xxx → ImageManifest → config.digest + layers[].digest
// 5. GET /blobs/config.digest → ImageConfig → Env + Cmd
// 6. 遍历 layers，每个 GET /blobs/digest → tar.gz → 解压到 /tmp/layers/
// 7. 返回
func pullImage(image, tag string) (*PullResult, error) {
if tag == "" {
tag = "latest"
}
repo := image
if !strings.Contains(image, "/") {
repo = "library/" + image // "alpine" → "library/alpine"
}
token, err := getToken(repo)
if err != nil {
return nil, err
}
baseURL := fmt.Sprintf("https://registry-1.docker.io/v2/%s", repo)
// 1. manifest list → 挑 amd64
var ml ManifestList
if err := doRequest(
fmt.Sprintf("%s/manifests/%s", baseURL, tag),
token,
"application/vnd.docker.distribution.manifest.v2+json",
&ml,
); err != nil {
return nil, fmt.Errorf("manifest list: %w", err)
}
var targetDigest string
for _, e := range ml.Manifests {
if e.Platform.Architecture == "amd64" && e.Platform.OS == "linux" {
targetDigest = e.Digest
break
}
}
if targetDigest == "" {
return nil, fmt.Errorf("no amd64/linux manifest for %s:%s", image, tag)
}
// 2. 真正的 manifest → config.digest + layers
var im ImageManifest
if err := doRequest(
fmt.Sprintf("%s/manifests/%s", baseURL, targetDigest),
token,
"application/vnd.oci.image.manifest.v1+json",
&im,
); err != nil {
return nil, fmt.Errorf("image manifest: %w", err)
}
// 3. config blob → Env + Cmd
var cfg ImageConfig
if err := doRequest(
fmt.Sprintf("%s/blobs/%s", baseURL, im.Config.Digest),
token,
"",
&cfg,
); err != nil {
return nil, fmt.Errorf("config blob: %w", err)
}
// 4. 逐层下载
layersBase := fmt.Sprintf("/tmp/layers/%s/%s", strings.ReplaceAll(repo, "/", "_"), tag)
var lowerDirs []string
for _, layer := range im.Layers {
safeName := strings.Replace(layer.Digest, ":", "_", 1)
layerDir := filepath.Join(layersBase, safeName)
// 目录已存在 = 之前下载过，跳过
if _, err := os.Stat(layerDir); os.IsNotExist(err) {
fmt.Printf("[pull] downloading %s (%d MB)...\n",
layer.Digest[:19], layer.Size/1024/1024)
if err := extractLayer(baseURL, layer.Digest, token, layerDir); err != nil {
fmt.Printf("[pull] WARNING: layer %s unavailable (%v), skipping\n", layer.Digest[:19], err)
continue // 404 或其他 registry 错误——跳过这一层
}
} else {
fmt.Printf("[pull] %s (cached)\n", layer.Digest[:19])
}
// OCI layers[0] = 基底层，overlay lowerdir 第一个 = 最上层
// 所以要反转——最新的层放在最前面
lowerDirs = append([]string{layerDir}, lowerDirs...)
}
return &PullResult{
LowerDirs: strings.Join(lowerDirs, ":"),
Env: cfg.Config.Env,
Cmd: cfg.Config.Cmd,
Entrypoint: cfg.Config.Entrypoint,
}, nil
}
// extractLayer 下载一个 blob（tar.gz 格式）并解压到 destDir。
// 处理普通文件、目录、符号链接。不处理硬链接（容器镜像很少用）。
func extractLayer(baseURL, digest, token, destDir string) error {
url := fmt.Sprintf("%s/blobs/%s", baseURL, digest)
req, err := http.NewRequest("GET", url, nil)
if err != nil {
return err
}
req.Header.Set("Authorization", "Bearer "+token)
resp, err := http.DefaultClient.Do(req)
if err != nil {
return err
}
defer resp.Body.Close()
if resp.StatusCode != http.StatusOK {
return fmt.Errorf("status %d", resp.StatusCode)
}
// 第一层：gzip 解压
gz, err := gzip.NewReader(resp.Body)
if err != nil {
return fmt.Errorf("gzip: %w", err)
}
defer gz.Close()
// 第二层：tar 遍历
tr := tar.NewReader(gz)
for {
hdr, err := tr.Next()
if err == io.EOF {
break
}
if err != nil {
return fmt.Errorf("tar: %w", err)
}
target := filepath.Join(destDir, hdr.Name)
switch hdr.Typeflag {
case tar.TypeDir:
os.MkdirAll(target, 0755)
os.Chmod(target, os.FileMode(hdr.Mode))
case tar.TypeReg:
os.MkdirAll(filepath.Dir(target), 0755)
f, err := os.Create(target)
if err != nil {
return fmt.Errorf("create %s: %w", target, err)
}
if _, err := io.Copy(f, tr); err != nil {
f.Close()
return err
}
f.Close()
os.Chmod(target, os.FileMode(hdr.Mode))
case tar.TypeSymlink:
os.Symlink(hdr.Linkname, target)
}
}
return nil
}