cluster: auto session certs

Signed-off-by: xhe <[email protected]>

Author: xhebox

pkg/cluster/manager/builder.go

@@ -15,7 +15,9 @@ package manager
 
 import (
 	"context"
+	"encoding/pem"
 	"fmt"
+	"os"
 	"path/filepath"
 	"strings"
 
@@ -835,6 +837,36 @@ func buildTLSTask(
 	return builder.Build(), nil
 }
 
+func genTiProxySessionCerts(dir string) error {
+	if _, err := os.Stat(filepath.Join(dir, "tiproxy-session.crt")); err == nil {
+		return nil
+	}
+
+	ca, err := crypto.NewCA("tiproxy")
+	if err != nil {
+		return err
+	}
+	privKey, err := crypto.NewKeyPair(crypto.KeyTypeRSA, crypto.KeySchemeRSASSAPSSSHA256)
+	if err != nil {
+		return err
+	}
+	csr, err := privKey.CSR("tiproxy", "tiproxy", nil, nil)
+	if err != nil {
+		return err
+	}
+	cert, err := ca.Sign(csr)
+	if err != nil {
+		return err
+	}
+	if err := utils.SaveFileWithBackup(filepath.Join(dir, "tiproxy-session.key"), privKey.Pem(), ""); err != nil {
+		return err
+	}
+	return utils.SaveFileWithBackup(filepath.Join(dir, "tiproxy-session.crt"), pem.EncodeToMemory(&pem.Block{
+		Type:  "CERTIFICATE",
+		Bytes: cert,
+	}), "")
+}
+
 // buildCertificateTasks generates certificate for instance and transfers it to the server
 func buildCertificateTasks(
 	m *Manager,
@@ -848,37 +880,61 @@ func buildCertificateTasks(
 		certificateTasks []*task.StepDisplay // tasks which are used to copy certificate to remote host
 	)
 
-	if topo.BaseTopo().GlobalOptions.TLSEnabled {
-		// copy certificate to remote host
-		topo.IterInstance(func(inst spec.Instance) {
-			deployDir := spec.Abs(base.User, inst.DeployDir())
-			tlsDir := filepath.Join(deployDir, spec.TLSCertKeyDir)
+	// check if there is tiproxy
+	// if there is tiproxy, whether or not TLS, we must issue a self-signed cert
+	hasTiProxy := false
+	topo.IterInstance(func(inst spec.Instance) {
+		if inst.ComponentName() == spec.ComponentTiProxy {
+			hasTiProxy = true
+		}
+	})
+	if hasTiProxy {
+		if err := genTiProxySessionCerts(m.specManager.Path(name, spec.TempConfigPath)); err != nil {
+			return certificateTasks, err
+		}
+	}
 
+	// copy certificate to remote host
+	topo.IterInstance(func(inst spec.Instance) {
+		deployDir := spec.Abs(base.User, inst.DeployDir())
+		tlsDir := filepath.Join(deployDir, spec.TLSCertKeyDir)
+
+		needSessionCert := hasTiProxy && inst.ComponentName() == spec.ComponentTiDB
+		if needSessionCert || topo.BaseTopo().GlobalOptions.TLSEnabled {
 			tb := task.NewSimpleUerSSH(m.logger, inst.GetManageHost(), inst.GetSSHPort(), base.User, gOpt, p, topo.BaseTopo().GlobalOptions.SSHType).
 				Mkdir(base.User, inst.GetManageHost(), topo.BaseTopo().GlobalOptions.SystemdMode != spec.UserMode, deployDir, tlsDir)
 
-			ca, err := crypto.ReadCA(
-				name,
-				m.specManager.Path(name, spec.TLSCertKeyDir, spec.TLSCACert),
-				m.specManager.Path(name, spec.TLSCertKeyDir, spec.TLSCAKey),
-			)
-			if err != nil {
-				iterErr = err
-				return
+			if needSessionCert {
+				tb = tb.
+					CopyFile(filepath.Join(m.specManager.Path(name, spec.TempConfigPath), "tiproxy-session.key"), filepath.Join(deployDir, spec.TLSCertKeyDir, "tiproxy-session.key"), inst.GetHost(), false, 0, false).
+					CopyFile(filepath.Join(m.specManager.Path(name, spec.TempConfigPath), "tiproxy-session.crt"), filepath.Join(deployDir, spec.TLSCertKeyDir, "tiproxy-session.crt"), inst.GetHost(), false, 0, false)
 			}
-			t := tb.TLSCert(
-				inst.GetHost(),
-				inst.ComponentName(),
-				inst.Role(),
-				inst.GetMainPort(),
-				ca,
-				meta.DirPaths{
-					Deploy: deployDir,
-					Cache:  m.specManager.Path(name, spec.TempConfigPath),
-				}).
-				BuildAsStep(fmt.Sprintf("  - Generate certificate %s -> %s", inst.ComponentName(), inst.ID()))
+
+			if topo.BaseTopo().GlobalOptions.TLSEnabled {
+				ca, err := crypto.ReadCA(
+					name,
+					m.specManager.Path(name, spec.TLSCertKeyDir, spec.TLSCACert),
+					m.specManager.Path(name, spec.TLSCertKeyDir, spec.TLSCAKey),
+				)
+				if err != nil {
+					iterErr = err
+					return
+				}
+				tb = tb.TLSCert(
+					inst.GetHost(),
+					inst.ComponentName(),
+					inst.Role(),
+					inst.GetMainPort(),
+					ca,
+					meta.DirPaths{
+						Deploy: deployDir,
+						Cache:  m.specManager.Path(name, spec.TempConfigPath),
+					})
+			}
+
+			t := tb.BuildAsStep(fmt.Sprintf("  - Generate certificate %s -> %s", inst.ComponentName(), inst.ID()))
 			certificateTasks = append(certificateTasks, t)
-		})
-	}
+		}
+	})
 	return certificateTasks, iterErr
 }

pkg/cluster/spec/tidb.go

@@ -241,6 +241,11 @@ func (i *TiDBInstance) InitConfig(
 		}
 	}
 
+	spec.Config, err = i.setTiProxyConfig(ctx, topo, spec.Config, paths)
+	if err != nil {
+		return err
+	}
+
 	// set TLS configs
 	spec.Config, err = i.setTLSConfig(ctx, enableTLS, spec.Config, paths)
 	if err != nil {
@@ -254,6 +259,39 @@ func (i *TiDBInstance) InitConfig(
 	return checkConfig(ctx, e, i.ComponentName(), i.ComponentSource(), version, i.OS(), i.Arch(), i.ComponentName()+".toml", paths)
 }
 
+// setTiProxyConfig sets tiproxy session certs
+func (i *TiDBInstance) setTiProxyConfig(ctx context.Context, topo *Specification, configs map[string]any, paths meta.DirPaths) (map[string]any, error) {
+	hasTiProxy := false
+	topo.IterInstance(func(instance Instance) {
+		if instance.ComponentName() == ComponentTiProxy {
+			hasTiProxy = true
+		}
+	})
+
+	if hasTiProxy {
+		if configs == nil {
+			configs = make(map[string]any)
+		}
+		configs["security.session-token-signing-cert"] = fmt.Sprintf(
+			"%s/tls/tiproxy-session.crt",
+			paths.Deploy)
+		configs["security.session-token-signing-key"] = fmt.Sprintf(
+			"%s/tls/tiproxy-session.key",
+			paths.Deploy)
+	} else {
+		tlsConfigs := []string{
+			"security.session-token-signing-cert",
+			"security.session-token-signing-key",
+		}
+		if configs != nil {
+			for _, config := range tlsConfigs {
+				delete(configs, config)
+			}
+		}
+	}
+	return configs, nil
+}
+
 // setTLSConfig set TLS Config to support enable/disable TLS
 func (i *TiDBInstance) setTLSConfig(ctx context.Context, enableTLS bool, configs map[string]any, paths meta.DirPaths) (map[string]any, error) {
 	// set TLS configs