feat: add socket auditor for forwarding logs to coder agent

Add SocketAuditor that sends audit logs to the Coder workspace agent
via a Unix socket. This enables boundary audit events to be forwarded
to coderd for centralized logging.

Implementation notes:
- Batching: 10 logs or 5 seconds, whichever comes first
- Wire format: length-prefixed protobuf. proto imported from AgentAPI to
  simplify boundary -> agent -> coderd forwarding to start.

Author: zedkipp

audit/multi_auditor.go

@@ -0,0 +1,18 @@
+package audit
+
+// MultiAuditor wraps multiple auditors and sends audit events to all of them.
+type MultiAuditor struct {
+	auditors []Auditor
+}
+
+// NewMultiAuditor creates a new MultiAuditor that sends to all provided auditors.
+func NewMultiAuditor(auditors ...Auditor) *MultiAuditor {
+	return &MultiAuditor{auditors: auditors}
+}
+
+// AuditRequest sends the request to all wrapped auditors.
+func (m *MultiAuditor) AuditRequest(req Request) {
+	for _, a := range m.auditors {
+		a.AuditRequest(req)
+	}
+}

audit/socket_auditor.go

@@ -0,0 +1,240 @@
+package audit
+
+import (
+	"context"
+	"encoding/binary"
+	"fmt"
+	"log/slog"
+	"net"
+	"time"
+
+	"google.golang.org/protobuf/proto"
+	"google.golang.org/protobuf/types/known/timestamppb"
+
+	agentproto "github.com/coder/coder/v2/agent/proto"
+)
+
+const (
+	defaultBatchSize          = 10
+	defaultBatchTimerDuration = 5 * time.Second
+	// DefaultAuditSocketPath is the well-known path for the boundary audit socket.
+	// The expectation is the Coder agent listens on this socket to receive audit logs.
+	DefaultAuditSocketPath = "/tmp/boundary-audit.sock"
+)
+
+// SocketAuditor implements the Auditor interface. It sends logs to the
+// workspace agent's boundary log proxy socket. It queues logs and sends
+// them in batches using a batch size and timer. The internal queue operates
+// as a FIFO i.e., logs are sent in the order they are received and dropped
+// if the queue is full.
+//
+// The proto messages sent to the agent are framed as follows:
+// [4 bit tag][28 bit length][length bytes of encoded protobuf]
+//
+// The tag is currently always 1, but may be extended in the future (e.g.
+// to support compression).
+type SocketAuditor struct {
+	socketPath         string
+	logger             *slog.Logger
+	logCh              chan *agentproto.BoundaryLog
+	batchSize          int
+	batchTimerDuration time.Duration
+
+	// onFlushAttempt is called after each flush attempt (intended for testing).
+	onFlushAttempt func()
+}
+
+// NewSocketAuditor creates a new SocketAuditor that sends logs to the agent's
+// boundary log proxy socket at DefaultAuditSocketPath after SocketAuditor.Loop
+// is called.
+func NewSocketAuditor(logger *slog.Logger) *SocketAuditor {
+	return &SocketAuditor{
+		socketPath:         DefaultAuditSocketPath,
+		logger:             logger,
+		logCh:              make(chan *agentproto.BoundaryLog, 2*defaultBatchSize),
+		batchSize:          defaultBatchSize,
+		batchTimerDuration: defaultBatchTimerDuration,
+	}
+}
+
+// AuditRequest implements the Auditor interface. It queues the log to be sent to the
+// agent in a batch.
+func (s *SocketAuditor) AuditRequest(req Request) {
+	httpReq := &agentproto.BoundaryLog_HttpRequest{
+		Method: req.Method,
+		Url:    req.URL,
+	}
+	// Only include the matched rule for denied requests, as documented in
+	// the proto schema.
+	if !req.Allowed {
+		httpReq.MatchedRule = req.Rule
+	}
+
+	log := &agentproto.BoundaryLog{
+		Allowed:  req.Allowed,
+		Time:     timestamppb.Now(),
+		Resource: &agentproto.BoundaryLog_HttpRequest_{HttpRequest: httpReq},
+	}
+
+	select {
+	case s.logCh <- log:
+	default:
+		s.logger.Warn("audit log dropped, channel full")
+	}
+}
+
+// flushErr represents an error from flush, distinguishing between
+// permanent errors (bad data) and transient errors (network issues).
+type flushErr struct {
+	err       error
+	permanent bool
+}
+
+func (e *flushErr) Error() string { return e.err.Error() }
+
+// flush sends the current batch of logs to the given connection.
+func flush(conn net.Conn, logs []*agentproto.BoundaryLog) *flushErr {
+	if len(logs) == 0 {
+		return nil
+	}
+
+	req := &agentproto.ReportBoundaryLogsRequest{
+		Logs: logs,
+	}
+
+	data, err := proto.Marshal(req)
+	if err != nil {
+		return &flushErr{err: err, permanent: true}
+	}
+
+	if len(data) > 1<<28 {
+		return &flushErr{err: fmt.Errorf("data too large: %d bytes", len(data)), permanent: true}
+	}
+
+	var header uint32
+	header |= uint32(len(data))
+	header |= 1 << 28
+
+	if err := binary.Write(conn, binary.BigEndian, header); err != nil {
+		return &flushErr{err: err}
+	}
+	if _, err := conn.Write(data); err != nil {
+		return &flushErr{err: err}
+	}
+	return nil
+}
+
+// Loop handles the I/O to send audit logs to the agent.
+func (s *SocketAuditor) Loop(ctx context.Context) {
+	var conn net.Conn
+	batch := make([]*agentproto.BoundaryLog, 0, s.batchSize)
+	t := time.NewTimer(0)
+	t.Stop()
+
+	// connect attempts to establish a connection to the socket.
+	connect := func() {
+		if conn != nil {
+			return
+		}
+		var err error
+		conn, err = net.Dial("unix", s.socketPath)
+		if err != nil {
+			s.logger.Warn("failed to connect to audit socket", "path", s.socketPath, "error", err)
+			conn = nil
+		}
+	}
+
+	// closeConn closes the current connection if open.
+	closeConn := func() {
+		if conn != nil {
+			_ = conn.Close()
+			conn = nil
+		}
+	}
+
+	// clearBatch resets the length of the batch and frees memory while preserving
+	// the batch slice backing array.
+	clearBatch := func() {
+		for i := range len(batch) {
+			batch[i] = nil
+		}
+		batch = batch[:0]
+	}
+
+	// doFlush flushes the batch and handles errors by reconnecting.
+	doFlush := func() {
+		t.Stop()
+		defer func() {
+			if s.onFlushAttempt != nil {
+				s.onFlushAttempt()
+			}
+		}()
+		if len(batch) == 0 {
+			return
+		}
+		connect()
+		if conn == nil {
+			// No connection: logs will be retried on next flush.
+			return
+		}
+
+		if err := flush(conn, batch); err != nil {
+			s.logger.Warn("failed to flush audit logs", "error", err)
+			if err.permanent {
+				// Data error: discard batch to avoid infinite retries.
+				clearBatch()
+			} else {
+				// Network error: close connection but keep batch for a future retry.
+				closeConn()
+			}
+			return
+		}
+
+		clearBatch()
+	}
+
+	connect()
+
+	for {
+		select {
+		case <-ctx.Done():
+			// Drain any pending logs before the last flush. Not concerned about
+			// growing the batch slice here since we're exiting.
+		drain:
+			for {
+				select {
+				case log := <-s.logCh:
+					batch = append(batch, log)
+				default:
+					break drain
+				}
+			}
+
+			doFlush()
+			closeConn()
+			return
+		case <-t.C:
+			doFlush()
+		case log := <-s.logCh:
+			// If batch is at capacity, attempt flushing first and drop the log if
+			// the batch still full.
+			if len(batch) >= s.batchSize {
+				doFlush()
+				if len(batch) >= s.batchSize {
+					s.logger.Warn("audit log dropped, batch full")
+					continue
+				}
+			}
+
+			batch = append(batch, log)
+
+			if len(batch) == 1 {
+				t.Reset(s.batchTimerDuration)
+			}
+
+			if len(batch) >= s.batchSize {
+				doFlush()
+			}
+		}
+	}
+}