Improve Vault integration and validation

command/agent/config-test-fixtures/basic.hcl

@@ -121,6 +121,7 @@ vault {
     key_file = "/path/to/key/file"
     tls_server_name = "foobar"
     tls_skip_verify = true
+    create_from_role = "test_role"
 }
 tls {
     http = true

command/agent/config_parse.go

@@ -710,6 +710,7 @@ func parseVaultConfig(result **config.VaultConfig, list *ast.ObjectList) error {
 		"ca_file",
 		"ca_path",
 		"cert_file",
+		"create_from_role",
 		"key_file",
 		"tls_server_name",
 		"tls_skip_verify",

command/agent/config_parse_test.go

@@ -129,6 +129,7 @@ func TestConfig_Parse(t *testing.T) {
 					Addr:                 "127.0.0.1:9500",
 					AllowUnauthenticated: &trueValue,
 					Enabled:              &falseValue,
+					Role:                 "test_role",
 					TLSCaFile:            "/path/to/ca/file",
 					TLSCaPath:            "/path/to/ca",
 					TLSCertFile:          "/path/to/cert/file",

nomad/structs/config/vault.go

@@ -30,6 +30,13 @@ type VaultConfig struct {
 	// lifetime.
 	Token string `mapstructure:"token"`
 
+	// Role sets the role in which to create tokens from. The Token given to
+	// Nomad does not have to be created from this role but must have "update"
+	// capability on "auth/token/create/<create_from_role>". If this value is
+	// unset and the token is created from a role, the value is defaulted to the
+	// role the token is from.
+	Role string `mapstructure:"create_from_role"`
+
 	// AllowUnauthenticated allows users to submit jobs requiring Vault tokens
 	// without providing a Vault token proving they have access to these
 	// policies.
@@ -99,6 +106,9 @@ func (a *VaultConfig) Merge(b *VaultConfig) *VaultConfig {
 	if b.Token != "" {
 		result.Token = b.Token
 	}
+	if b.Role != "" {
+		result.Role = b.Role
+	}
 	if b.TaskTokenTTL != "" {
 		result.TaskTokenTTL = b.TaskTokenTTL
 	}

nomad/vault.go

@@ -47,11 +47,68 @@ const (
 	// vaultRevocationIntv is the interval at which Vault tokens that failed
 	// initial revocation are retried
 	vaultRevocationIntv = 5 * time.Minute
+
+	// vaultCapabilitiesLookupPath is the path to lookup the capabilities of
+	// ones token.
+	vaultCapabilitiesLookupPath = "/sys/capabilities-self"
+
+	// vaultTokenRenewPath is the path used to renew our token
+	vaultTokenRenewPath = "auth/token/renew-self"
+
+	// vaultTokenLookupPath is the path used to lookup a token
+	vaultTokenLookupPath = "auth/token/lookup"
+
+	// vaultTokenLookupSelfPath is the path used to lookup self token
+	vaultTokenLookupSelfPath = "auth/token/lookup-self"
+
+	// vaultTokenRevokePath is the path used to revoke a token
+	vaultTokenRevokePath = "auth/token/revoke-accessor"
+
+	// vaultRoleLookupPath is the path to lookup a role
+	vaultRoleLookupPath = "auth/token/roles/%s"
+
+	// vaultRoleCreatePath is the path to create a token from a role
+	vaultTokenRoleCreatePath = "auth/token/create/%s"
 )
 
 var (
 	// vaultUnrecoverableError matches unrecoverable errors
 	vaultUnrecoverableError = regexp.MustCompile(`Code:\s+40(0|3|4)`)
+
+	// vaultCapabilitiesCapability is the expected capability of Nomad's Vault
+	// token on the the path. The token must have at least one of the
+	// capabilities.
+	vaultCapabilitiesCapability = []string{"update", "root"}
+
+	// vaultTokenRenewCapability is the expected capability Nomad's
+	// Vault token should have on the path. The token must have at least one of
+	// the capabilities.
+	vaultTokenRenewCapability = []string{"update", "root"}
+
+	// vaultTokenLookupCapability is the expected capability Nomad's
+	// Vault token should have on the path. The token must have at least one of
+	// the capabilities.
+	vaultTokenLookupCapability = []string{"update", "root"}
+
+	// vaultTokenLookupSelfCapability is the expected capability Nomad's
+	// Vault token should have on the path. The token must have at least one of
+	// the capabilities.
+	vaultTokenLookupSelfCapability = []string{"update", "root"}
+
+	// vaultTokenRevokeCapability is the expected capability Nomad's
+	// Vault token should have on the path. The token must have at least one of
+	// the capabilities.
+	vaultTokenRevokeCapability = []string{"update", "root"}
+
+	// vaultRoleLookupCapability is the the expected capability Nomad's Vault
+	// token should have on the path. The token must have at least one of the
+	// capabilities.
+	vaultRoleLookupCapability = []string{"read", "root"}
+
+	// vaultTokenRoleCreateCapability is the the expected capability Nomad's Vault
+	// token should have on the path. The token must have at least one of the
+	// capabilities.
+	vaultTokenRoleCreateCapability = []string{"update", "root"}
 )
 
 // VaultClient is the Servers interface for interfacing with Vault
@@ -417,7 +474,7 @@ func (v *vaultClient) renewalLoop() {
 
 			// Set base values and add some backoff
 
-			v.logger.Printf("[DEBUG] vault: got error or bad auth, so backing off: %v", err)
+			v.logger.Printf("[WARN] vault: got error or bad auth, so backing off: %v", err)
 			switch {
 			case backoff < 5:
 				backoff = 5
@@ -477,8 +534,9 @@ func (v *vaultClient) renew() error {
 // getWrappingFn returns an appropriate wrapping function for Nomad Servers
 func (v *vaultClient) getWrappingFn() func(operation, path string) string {
 	createPath := "auth/token/create"
-	if !v.tokenData.Root {
-		createPath = fmt.Sprintf("auth/token/create/%s", v.tokenData.Role)
+	role := v.getRole()
+	if role != "" {
+		createPath = fmt.Sprintf("auth/token/create/%s", role)
 	}
 
 	return func(operation, path string) string {
@@ -497,10 +555,18 @@ func (v *vaultClient) getWrappingFn() func(operation, path string) string {
 func (v *vaultClient) parseSelfToken() error {
 	// Get the initial lease duration
 	auth := v.client.Auth().Token()
-	self, err := auth.LookupSelf()
+	var self *vapi.Secret
+
+	// Try looking up the token using the self endpoint
+	secret, err := auth.LookupSelf()
 	if err != nil {
-		return fmt.Errorf("failed to lookup Vault periodic token: %v", err)
+		// Try looking up our token directly
+		self, err = auth.Lookup(v.client.Token())
+		if err != nil {
+			return fmt.Errorf("failed to lookup Vault periodic token: %v", err)
+		}
 	}
+	self = secret
 
 	// Read and parse the fields
 	var data tokenData
@@ -516,7 +582,28 @@ func (v *vaultClient) parseSelfToken() error {
 		}
 	}
 
+	// Store the token data
+	data.Root = root
+	v.tokenData = &data
+
+	// The criteria that must be met for the token to be valid are as follows:
+	// 1) If token is non-root or is but has a creation ttl
+	//   a) The token must be renewable
+	//   b) Token must have a non-zero TTL
+	// 2) Must have update capability for "auth/token/lookup/" (used to verify incoming tokens)
+	// 3) Must have update capability for "/auth/token/revoke-accessor/" (used to revoke unneeded tokens)
+	// 4) If configured to create tokens against a role:
+	//   a) Must have read capability for "auth/token/roles/<role_name" (Can just attemp a read)
+	//   b) Must have update capability for path "auth/token/create/<role_name>"
+	//   c) Role must:
+	//     1) Not allow orphans
+	//     2) Must allow tokens to be renewed
+	//     3) Must not have an explicit max TTL
+	//     4) Must have non-zero period
+	// 5) If not configured against a role, the token must be root
+
 	var mErr multierror.Error
+	role := v.getRole()
 	if !root {
 		// All non-root tokens must be renewable
 		if !data.Renewable {
@@ -534,7 +621,7 @@ func (v *vaultClient) parseSelfToken() error {
 		}
 
 		// There must be a valid role since we aren't root
-		if data.Role == "" {
+		if role == "" {
 			multierror.Append(&mErr, fmt.Errorf("token role name must be set when not using a root token"))
 		}
 
@@ -548,18 +635,106 @@ func (v *vaultClient) parseSelfToken() error {
 		}
 	}
 
+	// Check we have the correct capabilities
+	if err := v.validateCapabilities(role, root); err != nil {
+		multierror.Append(&mErr, err)
+	}
+
 	// If given a role validate it
-	if data.Role != "" {
-		if err := v.validateRole(data.Role); err != nil {
+	if role != "" {
+		if err := v.validateRole(role); err != nil {
 			multierror.Append(&mErr, err)
 		}
 	}
 
-	data.Root = root
-	v.tokenData = &data
 	return mErr.ErrorOrNil()
 }
 
+// getRole returns the role name to be used when creating tokens
+func (v *vaultClient) getRole() string {
+	if v.config.Role != "" {
+		return v.config.Role
+	}
+
+	return v.tokenData.Role
+}
+
+// validateCapabilities checks that Nomad's Vault token has the correct
+// capabilities.
+func (v *vaultClient) validateCapabilities(role string, root bool) error {
+	// Check if the token can lookup capabilities.
+	var mErr multierror.Error
+	_, _, err := v.hasCapability(vaultCapabilitiesLookupPath, vaultCapabilitiesCapability)
+	if err != nil {
+		// Check if there is a permission denied
+		if vaultUnrecoverableError.MatchString(err.Error()) {
+			// Since we can't read permissions, we just log a warning that we
+			// can't tell if the Vault token will work
+			msg := fmt.Sprintf("Can not lookup token capabilities. "+
+				"As such certain operations may fail in the future. "+
+				"Please give Nomad a Vault token with one of the following "+
+				"capabilities %q on %q so that the required capabilities can be verified",
+				vaultCapabilitiesCapability, vaultCapabilitiesLookupPath)
+			v.logger.Printf("[WARN] vault: %s", msg)
+			return nil
+		} else {
+			multierror.Append(&mErr, err)
+		}
+	}
+
+	// verify is a helper function that verifies the token has one of the
+	// capabilities on the given path and adds an issue to the error
+	verify := func(path string, requiredCaps []string) {
+		ok, caps, err := v.hasCapability(path, requiredCaps)
+		if err != nil {
+			multierror.Append(&mErr, err)
+		} else if !ok {
+			multierror.Append(&mErr,
+				fmt.Errorf("token must have one of the following capabilities %q on %q; has %v", requiredCaps, path, caps))
+		}
+	}
+
+	// Check if we are verifying incoming tokens
+	if !v.config.AllowsUnauthenticated() {
+		verify(vaultTokenLookupPath, vaultTokenLookupCapability)
+	}
+
+	// Verify we can renew our selves tokens
+	verify(vaultTokenRenewPath, vaultTokenRenewCapability)
+
+	// Verify we can revoke tokens
+	verify(vaultTokenRevokePath, vaultTokenRevokeCapability)
+
+	// If we are using a role verify the capability
+	if role != "" {
+		// Verify we can read the role
+		verify(fmt.Sprintf(vaultRoleLookupPath, role), vaultRoleLookupCapability)
+
+		// Verify we can create from the role
+		verify(fmt.Sprintf(vaultTokenRoleCreatePath, role), vaultTokenRoleCreateCapability)
+	}
+
+	return mErr.ErrorOrNil()
+}
+
+// hasCapability takes a path and returns whether the token has at least one of
+// the required capabilities on the given path. It also returns the set of
+// capabilities the token does have as well as any error that occured.
+func (v *vaultClient) hasCapability(path string, required []string) (bool, []string, error) {
+	caps, err := v.client.Sys().CapabilitiesSelf(path)
+	if err != nil {
+		return false, nil, err
+	}
+	for _, c := range caps {
+		for _, r := range required {
+			if c == r {
+				return true, caps, nil
+			}
+		}
+	}
+	return false, caps, nil
+}
+
 // validateRole contacts Vault and checks that the given Vault role is valid for
 // the purposes of being used by Nomad
 func (v *vaultClient) validateRole(role string) error {
@@ -679,12 +854,13 @@ func (v *vaultClient) CreateToken(ctx context.Context, a *structs.Allocation, ta
 	// token or a role based token
 	var secret *vapi.Secret
 	var err error
-	if v.tokenData.Root {
+	role := v.getRole()
+	if v.tokenData.Root && role == "" {
 		req.Period = v.childTTL
 		secret, err = v.auth.Create(req)
 	} else {
 		// Make the token using the role
-		secret, err = v.auth.CreateWithRole(req, v.tokenData.Role)
+		secret, err = v.auth.CreateWithRole(req, v.getRole())
 	}
 
 	// Determine whether it is unrecoverable