Files
pangolin/server/internal/agentd/enroll.go
T
wangjia f03d2dc8a6 feat(agent): node agent — enroll/mTLS, heartbeat, command stream, sing-box 用户表 (tsk__R8M4jEw43JR)
与控制面同仓同 go.mod,新增节点 agent 实现:

- proto/agent/v1/agent.proto + internal/pb/agentv1:冻结的控制面↔agent gRPC 契约
  (Enroll/Register/Heartbeat/Subscribe/Ack/ReportUsage)。仓库尚无 protoc 流水线,
  暂以手写 Go 类型 + JSON gRPC codec 实现,与 proto 1:1 对应,待 protoc 接入即可替换。
- internal/agentd:
  - enroll.go:首启生成 EC 密钥+CSR,持 bootstrap token 调 Enroll 换 90d 节点证书
    (CN=node_uuid),落 /etc/pangolin-agent/,此后 mTLS。
  - conn.go(agent.go)+creds.go:mTLS 主动拨号 + 指数退避重连;重连携带 last_command_id;
    Register 取 ConfigSnapshot 全量配置覆盖本地。
  - heartbeat.go:30s 上报 peer/带宽/CPU + config_version;need_full_resync→全量同步。
  - command.go:消费 Subscribe,Upsert/Revoke/Rotate/ApplyConfig/Lifecycle 幂等处理后
    Ack(at-least-once,按 command_id 去重)。
  - singbox.go+render.go:内存用户表 + 落盘 state.json(仅 dp_uuid+expires_at);任何变更
    渲染完整 sing-box 配置(REALITY users[uuid,flow] + Hy2 users[派生口令])→ 500ms 去抖
    合并 → systemd 重启。
  - ttl.go:凭证 TTL 定时移除并上报。
  - usage.go:按 dp_uuid 聚合上报,绝无 user_id/email/目的地址。
  - derive.go:Hy2 口令 = HMAC-SHA256(key, dp_uuid),与控制面同源派生。
- cmd/agent:入口(flag/env 配置)。
- infra/cloud-init/{node.yaml.tmpl,install-node.sh,README.md}:一段式安装,下载锁定版本
  二进制并校验 SHA-256,systemd 拉管,首启即 Enroll/Register。shellcheck -S warning 通过。

测试(bufconn mock 控制面,无需 docker):Enroll→Register→Heartbeat 全流转;Upsert/Revoke
渲染正确;Rotate 宽限期新旧并存到点移除;TTL 自动移除并上报;断流重连 last_command_id
续发不丢不重;need_full_resync 触发重注册;state.json 恢复;去抖合并;扫描确认无身份字段。
go test -race ./internal/agentd/... ./internal/pb/... 通过;go vet ./... 通过。

落实 doc/04 §2 节点无状态化与 doc/06 §3 数据面红线(节点仅见 dp_uuid)。

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-06-13 12:26:58 +08:00

140 lines
4.3 KiB
Go

package agentd
import (
"context"
"crypto/ecdsa"
"crypto/elliptic"
"crypto/rand"
"crypto/x509"
"crypto/x509/pkix"
"encoding/pem"
"errors"
"fmt"
"io/fs"
"os"
agentv1 "github.com/wangjia/pangolin/server/internal/pb/agentv1"
"google.golang.org/grpc"
)
// EnrollDialer opens a connection suitable for the (un-authenticated) Enroll RPC.
// The agent has no client certificate yet, so this connection presents none; the
// bootstrap token authenticates the call instead.
type EnrollDialer func(ctx context.Context) (*grpc.ClientConn, error)
// EnsureEnrolled guarantees the node holds a valid client certificate.
//
// - If node.crt/node.key already exist, it loads and returns the certificate's
// CN (the node UUID) without contacting the control plane (idempotent first
// start vs. restart).
// - Otherwise it generates an EC P-256 key + CSR, calls Enroll with the
// bootstrap token, and persists node.key (0600), node.crt and ca.crt.
//
// The private key is generated on the node and never leaves it.
func EnsureEnrolled(ctx context.Context, cfg Config, dial EnrollDialer) (string, error) {
cfg = cfg.withDefaults()
if uuid, ok, err := loadEnrolledUUID(cfg); err != nil {
return "", err
} else if ok {
return uuid, nil
}
if cfg.BootstrapToken == "" {
return "", errors.New("agentd: not enrolled and no bootstrap token provided")
}
keyPEM, csrPEM, err := generateKeyAndCSR()
if err != nil {
return "", err
}
conn, err := dial(ctx)
if err != nil {
return "", fmt.Errorf("agentd: dial for enroll: %w", err)
}
defer conn.Close()
client := agentv1.NewAgentServiceClient(conn)
resp, err := client.Enroll(ctx, &agentv1.EnrollRequest{
BootstrapToken: cfg.BootstrapToken,
CSRPEM: csrPEM,
AgentVersion: cfg.AgentVersion,
})
if err != nil {
return "", fmt.Errorf("agentd: enroll rpc: %w", err)
}
if resp.NodeUUID == "" || len(resp.CertPEM) == 0 {
return "", errors.New("agentd: enroll response missing node_uuid/cert")
}
if err := os.MkdirAll(cfg.StateDir, 0o700); err != nil {
return "", fmt.Errorf("agentd: mkdir state dir: %w", err)
}
if err := atomicWrite(cfg.KeyPath(), keyPEM, 0o600); err != nil {
return "", err
}
if err := atomicWrite(cfg.CertPath(), resp.CertPEM, 0o644); err != nil {
return "", err
}
if len(resp.CAPEM) > 0 {
if err := atomicWrite(cfg.CAPath(), resp.CAPEM, 0o644); err != nil {
return "", err
}
}
logf("enrolled successfully (cert valid until unix=%d)", resp.NotAfterUnix)
return resp.NodeUUID, nil
}
// loadEnrolledUUID returns the CN of the persisted node certificate, if present.
func loadEnrolledUUID(cfg Config) (string, bool, error) {
certPEM, err := os.ReadFile(cfg.CertPath())
if errors.Is(err, fs.ErrNotExist) {
return "", false, nil
}
if err != nil {
return "", false, fmt.Errorf("agentd: read node cert: %w", err)
}
if _, err := os.Stat(cfg.KeyPath()); err != nil {
// Cert without key is unusable — treat as not enrolled.
return "", false, nil
}
block, _ := pem.Decode(certPEM)
if block == nil {
return "", false, errors.New("agentd: node cert has no PEM block")
}
cert, err := x509.ParseCertificate(block.Bytes)
if err != nil {
return "", false, fmt.Errorf("agentd: parse node cert: %w", err)
}
if cert.Subject.CommonName == "" {
return "", false, errors.New("agentd: node cert has empty CN")
}
return cert.Subject.CommonName, true, nil
}
// generateKeyAndCSR creates an EC P-256 private key and a PKCS#10 CSR. The CSR's
// CN is a placeholder; the CA overrides it with the authoritative node UUID.
func generateKeyAndCSR() (keyPEM, csrPEM []byte, err error) {
key, err := ecdsa.GenerateKey(elliptic.P256(), rand.Reader)
if err != nil {
return nil, nil, fmt.Errorf("agentd: generate key: %w", err)
}
keyDER, err := x509.MarshalECPrivateKey(key)
if err != nil {
return nil, nil, fmt.Errorf("agentd: marshal key: %w", err)
}
keyPEM = pem.EncodeToMemory(&pem.Block{Type: "EC PRIVATE KEY", Bytes: keyDER})
tmpl := &x509.CertificateRequest{
Subject: pkix.Name{CommonName: "pangolin-node-pending"},
SignatureAlgorithm: x509.ECDSAWithSHA256,
}
csrDER, err := x509.CreateCertificateRequest(rand.Reader, tmpl, key)
if err != nil {
return nil, nil, fmt.Errorf("agentd: create CSR: %w", err)
}
csrPEM = pem.EncodeToMemory(&pem.Block{Type: "CERTIFICATE REQUEST", Bytes: csrDER})
return keyPEM, csrPEM, nil
}