control/controlclient, ipn: add client audit logging (#14950)

updates tailscale/corp#26435

Adds client support for sending audit logs to control via /machine/audit-log.
Specifically implements audit logging for user initiated disconnections.

This will require further work to optimize the peristant storage and exclusion
via build tags for mobile:
tailscale/corp#27011
tailscale/corp#27012

Signed-off-by: Jonathan Nobels <jonathan@tailscale.com>

control/controlclient/auto.go

@@ -119,6 +119,7 @@ type Auto struct {
 	updateCh      chan struct{} // readable when we should inform the server of a change
 	observer      Observer      // called to update Client status; always non-nil
 	observerQueue execqueue.ExecQueue
+	shutdownFn    func() // to be called prior to shutdown or nil
 
 	unregisterHealthWatch func()
 
@@ -189,6 +190,7 @@ func NewNoStart(opts Options) (_ *Auto, err error) {
 		mapDone:    make(chan struct{}),
 		updateDone: make(chan struct{}),
 		observer:   opts.Observer,
+		shutdownFn: opts.Shutdown,
 	}
 	c.authCtx, c.authCancel = context.WithCancel(context.Background())
 	c.authCtx = sockstats.WithSockStats(c.authCtx, sockstats.LabelControlClientAuto, opts.Logf)
@@ -755,6 +757,7 @@ func (c *Auto) Shutdown() {
 		return
 	}
 	c.logf("client.Shutdown ...")
+	shutdownFn := c.shutdownFn
 
 	direct := c.direct
 	c.closed = true
@@ -767,6 +770,10 @@ func (c *Auto) Shutdown() {
 	c.unpauseWaiters = nil
 	c.mu.Unlock()
 
+	if shutdownFn != nil {
+		shutdownFn()
+	}
+
 	c.unregisterHealthWatch()
 	<-c.authDone
 	<-c.mapDone

control/controlclient/controlclient_test.go

@@ -4,6 +4,8 @@
 package controlclient
 
 import (
+	"errors"
+	"fmt"
 	"io"
 	"reflect"
 	"slices"
@@ -147,3 +149,42 @@ func TestCanSkipStatus(t *testing.T) {
 		t.Errorf("Status fields = %q; this code was only written to handle fields %q", f, want)
 	}
 }
+
+func TestRetryableErrors(t *testing.T) {
+	errorTests := []struct {
+		err  error
+		want bool
+	}{
+		{errNoNoiseClient, true},
+		{errNoNodeKey, true},
+		{fmt.Errorf("%w: %w", errNoNoiseClient, errors.New("no noise")), true},
+		{fmt.Errorf("%w: %w", errHTTPPostFailure, errors.New("bad post")), true},
+		{fmt.Errorf("%w: %w", errNoNodeKey, errors.New("not node key")), true},
+		{errBadHTTPResponse(429, "too may requests"), true},
+		{errBadHTTPResponse(500, "internal server eror"), true},
+		{errBadHTTPResponse(502, "bad gateway"), true},
+		{errBadHTTPResponse(503, "service unavailable"), true},
+		{errBadHTTPResponse(504, "gateway timeout"), true},
+		{errBadHTTPResponse(1234, "random error"), false},
+	}
+
+	for _, tt := range errorTests {
+		t.Run(tt.err.Error(), func(t *testing.T) {
+			if isRetryableErrorForTest(tt.err) != tt.want {
+				t.Fatalf("retriable: got %v, want %v", tt.err, tt.want)
+			}
+		})
+	}
+}
+
+type retryableForTest interface {
+	Retryable() bool
+}
+
+func isRetryableErrorForTest(err error) bool {
+	var ae retryableForTest
+	if errors.As(err, &ae) {
+		return ae.Retryable()
+	}
+	return false
+}

control/controlclient/direct.go

@@ -156,6 +156,11 @@ type Options struct {
 	// If we receive a new DialPlan from the server, this value will be
 	// updated.
 	DialPlan ControlDialPlanner
+
+	// Shutdown is an optional function that will be called before client shutdown is
+	// attempted. It is used to allow the client to clean up any resources or complete any
+	// tasks that are dependent on a live client.
+	Shutdown func()
 }
 
 // ControlDialPlanner is the interface optionally supplied when creating a
@@ -1662,11 +1667,11 @@ func (c *Auto) SetDeviceAttrs(ctx context.Context, attrs tailcfg.AttrUpdate) err
 func (c *Direct) SetDeviceAttrs(ctx context.Context, attrs tailcfg.AttrUpdate) error {
 	nc, err := c.getNoiseClient()
 	if err != nil {
-		return err
+		return fmt.Errorf("%w: %w", errNoNoiseClient, err)
 	}
 	nodeKey, ok := c.GetPersist().PublicNodeKeyOK()
 	if !ok {
-		return errors.New("no node key")
+		return errNoNodeKey
 	}
 	if c.panicOnUse {
 		panic("tainted client")
@@ -1697,6 +1702,47 @@ func (c *Direct) SetDeviceAttrs(ctx context.Context, attrs tailcfg.AttrUpdate) e
 	return nil
 }
 
+// SendAuditLog implements [auditlog.Transport] by sending an audit log synchronously to the control plane.
+//
+// See docs on [tailcfg.AuditLogRequest] and [auditlog.Logger] for background.
+func (c *Auto) SendAuditLog(ctx context.Context, auditLog tailcfg.AuditLogRequest) (err error) {
+	return c.direct.sendAuditLog(ctx, auditLog)
+}
+
+func (c *Direct) sendAuditLog(ctx context.Context, auditLog tailcfg.AuditLogRequest) (err error) {
+	nc, err := c.getNoiseClient()
+	if err != nil {
+		return fmt.Errorf("%w: %w", errNoNoiseClient, err)
+	}
+
+	nodeKey, ok := c.GetPersist().PublicNodeKeyOK()
+	if !ok {
+		return errNoNodeKey
+	}
+
+	req := &tailcfg.AuditLogRequest{
+		Version: tailcfg.CurrentCapabilityVersion,
+		NodeKey: nodeKey,
+		Action:  auditLog.Action,
+		Details: auditLog.Details,
+	}
+
+	if c.panicOnUse {
+		panic("tainted client")
+	}
+
+	res, err := nc.post(ctx, "/machine/audit-log", nodeKey, req)
+	if err != nil {
+		return fmt.Errorf("%w: %w", errHTTPPostFailure, err)
+	}
+	defer res.Body.Close()
+	if res.StatusCode != 200 {
+		all, _ := io.ReadAll(res.Body)
+		return errBadHTTPResponse(res.StatusCode, string(all))
+	}
+	return nil
+}
+
 func addLBHeader(req *http.Request, nodeKey key.NodePublic) {
 	if !nodeKey.IsZero() {
 		req.Header.Add(tailcfg.LBHeader, nodeKey.String())

control/controlclient/errors.go

@@ -0,0 +1,51 @@
+// Copyright (c) Tailscale Inc & AUTHORS
+// SPDX-License-Identifier: BSD-3-Clause
+
+package controlclient
+
+import (
+	"errors"
+	"fmt"
+	"net/http"
+)
+
+// apiResponseError is an error type that can be returned by controlclient
+// api requests.
+//
+// It wraps an underlying error and a flag for clients to query if the
+// error is retryable via the Retryable() method.
+type apiResponseError struct {
+	err       error
+	retryable bool
+}
+
+// Error implements [error].
+func (e *apiResponseError) Error() string {
+	return e.err.Error()
+}
+
+// Retryable reports whether the error is retryable.
+func (e *apiResponseError) Retryable() bool {
+	return e.retryable
+}
+
+func (e *apiResponseError) Unwrap() error { return e.err }
+
+var (
+	errNoNodeKey       = &apiResponseError{errors.New("no node key"), true}
+	errNoNoiseClient   = &apiResponseError{errors.New("no noise client"), true}
+	errHTTPPostFailure = &apiResponseError{errors.New("http failure"), true}
+)
+
+func errBadHTTPResponse(code int, msg string) error {
+	retryable := false
+	switch code {
+	case http.StatusTooManyRequests,
+		http.StatusInternalServerError,
+		http.StatusBadGateway,
+		http.StatusServiceUnavailable,
+		http.StatusGatewayTimeout:
+		retryable = true
+	}
+	return &apiResponseError{fmt.Errorf("http error %d: %s", code, msg), retryable}
+}