teraWeb/backend/internal/serial/socket.go

package serial

import (
	"bufio"
	"bytes"
	"context"
	"encoding/json"
	"errors"
	"fmt"
	"io"
	"net"
	"time"
)

const DefaultSocketPath = "/run/tera-seriald/serial-ports.sock"

type socketRequest struct {
	Action   string          `json:"action"`
	Configs  []ChannelConfig `json:"configs,omitempty"`
	Config   *ChannelConfig  `json:"config,omitempty"`
	Captures []CaptureConfig `json:"captures,omitempty"`
}

type socketResponse struct {
	States   []State        `json:"states,omitempty"`
	Links    []LinkState    `json:"links,omitempty"`
	CRProbes []CRProbeState `json:"cr_probes,omitempty"`
	Probe    *ProbeResult   `json:"probe,omitempty"`
	Code     string         `json:"code,omitempty"`
	Error    string         `json:"error,omitempty"`
}

const maxSocketRequestBytes = 16 * 1024

// SocketServer exposes the seriald-only controller on a local Unix socket.
// It accepts only bounded state and line-configuration requests, never data.
type SocketServer struct {
	Controller Controller
	Capture    CaptureController
	CRProbe    CRProbeStateController
	Links      LinkStateController
}

func (s SocketServer) Serve(listener net.Listener) error {
	if s.Controller == nil {
		return errors.New("serial controller is required")
	}
	for {
		connection, err := listener.Accept()
		if err != nil {
			if errors.Is(err, net.ErrClosed) {
				return nil
			}
			if temporary, ok := err.(interface{ Temporary() bool }); ok && temporary.Temporary() {
				continue
			}
			return err
		}
		go s.handleConnection(connection)
	}
}

func (s SocketServer) handleConnection(connection net.Conn) {
	defer connection.Close()
	_ = connection.SetDeadline(time.Now().Add(10 * time.Second))
	payload, err := bufio.NewReader(io.LimitReader(connection, maxSocketRequestBytes+1)).ReadBytes('\n')
	if err != nil || len(payload) > maxSocketRequestBytes || !bytes.HasSuffix(payload, []byte{'\n'}) {
		s.writeResponse(connection, socketResponse{Code: "validation_failed", Error: "invalid serial control request"})
		return
	}
	decoder := json.NewDecoder(bytes.NewReader(payload))
	decoder.DisallowUnknownFields()
	var command socketRequest
	if err := decoder.Decode(&command); err != nil {
		s.writeResponse(connection, socketResponse{Code: "validation_failed", Error: "invalid serial control request"})
		return
	}
	var extra any
	if err := decoder.Decode(&extra); !errors.Is(err, io.EOF) {
		s.writeResponse(connection, socketResponse{Code: "validation_failed", Error: "serial control request must contain a single JSON value"})
		return
	}

	ctx, cancel := context.WithTimeout(context.Background(), 8*time.Second)
	defer cancel()
	switch command.Action {
	case "states":
		states, err := s.Controller.States(ctx)
		if err != nil {
			s.writeError(connection, err)
			return
		}
		s.writeResponse(connection, socketResponse{States: states})
	case "link_states":
		controller := s.Links
		if controller == nil {
			controller, _ = s.Controller.(LinkStateController)
		}
		if controller == nil {
			s.writeResponse(connection, socketResponse{Code: "hardware_unavailable", Error: "serial modem status is unavailable"})
			return
		}
		links, err := controller.LinkStates(ctx)
		if err != nil {
			s.writeError(connection, err)
			return
		}
		s.writeResponse(connection, socketResponse{Links: links})
	case "cr_probe_states":
		if s.CRProbe == nil {
			s.writeResponse(connection, socketResponse{Code: "hardware_unavailable", Error: "serial CR response probe is unavailable"})
			return
		}
		states, err := s.CRProbe.CRProbeStates(ctx)
		if err != nil {
			s.writeError(connection, err)
			return
		}
		s.writeResponse(connection, socketResponse{CRProbes: states})
	case "apply":
		states, err := s.Controller.Apply(ctx, command.Configs)
		if err != nil {
			s.writeError(connection, err)
			return
		}
		s.writeResponse(connection, socketResponse{States: states})
	case "apply_capture":
		if s.Capture == nil {
			s.writeResponse(connection, socketResponse{Code: "hardware_unavailable", Error: "serial payload capture control is unavailable"})
			return
		}
		if err := s.Capture.ApplyCapture(ctx, command.Captures); err != nil {
			s.writeError(connection, err)
			return
		}
		s.writeResponse(connection, socketResponse{})
	case "probe":
		if command.Config == nil {
			s.writeResponse(connection, socketResponse{Code: "validation_failed", Error: "serial line probe requires one config"})
			return
		}
		prober, ok := s.Controller.(interface {
			Probe(context.Context, ChannelConfig) (ProbeResult, error)
		})
		if !ok {
			s.writeResponse(connection, socketResponse{Code: "hardware_unavailable", Error: "serial line probe is unavailable"})
			return
		}
		result, err := prober.Probe(ctx, *command.Config)
		if err != nil {
			s.writeError(connection, err)
			return
		}
		s.writeResponse(connection, socketResponse{Probe: &result})
	default:
		s.writeResponse(connection, socketResponse{Code: "validation_failed", Error: "serial control action is not supported"})
	}
}

func (s SocketServer) writeError(connection net.Conn, err error) {
	code := "hardware_failed"
	if errors.Is(err, ErrStreamLeaseBusy) {
		code = "stream_lease_busy"
	} else if IsUnavailable(err) {
		code = "hardware_unavailable"
	}
	s.writeResponse(connection, socketResponse{Code: code, Error: err.Error()})
}

func (s SocketServer) writeResponse(connection net.Conn, value socketResponse) {
	_ = json.NewEncoder(connection).Encode(value)
}

// SocketClient implements Controller for the unprivileged tera-api process.
// Unix-socket permissions are the boundary between the API and seriald.
type SocketClient struct {
	path    string
	timeout time.Duration
}

func NewSocketClient(path string) *SocketClient {
	if path == "" {
		path = DefaultSocketPath
	}
	return &SocketClient{path: path, timeout: 10 * time.Second}
}

func (c *SocketClient) States(ctx context.Context) ([]State, error) {
	response, err := c.exchange(ctx, socketRequest{Action: "states"})
	if err != nil {
		return nil, err
	}
	return response.States, nil
}

func (c *SocketClient) LinkStates(ctx context.Context) ([]LinkState, error) {
	response, err := c.exchange(ctx, socketRequest{Action: "link_states"})
	if err != nil {
		return nil, err
	}
	return response.Links, nil
}

func (c *SocketClient) CRProbeStates(ctx context.Context) ([]CRProbeState, error) {
	response, err := c.exchange(ctx, socketRequest{Action: "cr_probe_states"})
	if err != nil {
		return nil, err
	}
	return response.CRProbes, nil
}

func (c *SocketClient) Apply(ctx context.Context, configs []ChannelConfig) ([]State, error) {
	response, err := c.exchange(ctx, socketRequest{Action: "apply", Configs: configs})
	if err != nil {
		return nil, err
	}
	return response.States, nil
}

func (c *SocketClient) ApplyCapture(ctx context.Context, configs []CaptureConfig) error {
	_, err := c.exchange(ctx, socketRequest{Action: "apply_capture", Captures: configs})
	return err
}

func (c *SocketClient) Probe(ctx context.Context, config ChannelConfig) (ProbeResult, error) {
	response, err := c.exchange(ctx, socketRequest{Action: "probe", Config: &config})
	if err != nil {
		return ProbeResult{}, err
	}
	if response.Probe == nil {
		return ProbeResult{}, errors.New("tera-seriald returned no probe result")
	}
	return *response.Probe, nil
}

func (c *SocketClient) exchange(ctx context.Context, command socketRequest) (socketResponse, error) {
	dialer := net.Dialer{Timeout: c.timeout}
	connection, err := dialer.DialContext(ctx, "unix", c.path)
	if err != nil {
		return socketResponse{}, &UnavailableError{Reason: "tera-seriald is not reachable"}
	}
	defer connection.Close()
	deadline := time.Now().Add(c.timeout)
	if contextDeadline, ok := ctx.Deadline(); ok && contextDeadline.Before(deadline) {
		deadline = contextDeadline
	}
	_ = connection.SetDeadline(deadline)
	if err := json.NewEncoder(connection).Encode(command); err != nil {
		return socketResponse{}, fmt.Errorf("serial control request failed")
	}
	var response socketResponse
	if err := json.NewDecoder(io.LimitReader(connection, 65536)).Decode(&response); err != nil {
		return socketResponse{}, &UnavailableError{Reason: "tera-seriald returned no response"}
	}
	if response.Error != "" {
		if response.Code == "hardware_unavailable" {
			return socketResponse{}, &UnavailableError{Reason: response.Error}
		}
		// The exact-message fallback keeps a newer API compatible with a
		// seriald that predates the stream_lease_busy protocol code.
		if response.Code == "stream_lease_busy" || response.Error == ErrStreamLeaseBusy.Error() {
			return socketResponse{}, ErrStreamLeaseBusy
		}
		return socketResponse{}, errors.New(response.Error)
	}
	return response, nil
}