| 1 | package sidecar |
| 2 | |
| 3 | import ( |
| 4 | "errors" |
| 5 | "fmt" |
| 6 | "io" |
| 7 | "os" |
| 8 | "os/exec" |
| 9 | "path/filepath" |
| 10 | "strings" |
| 11 | "sync" |
| 12 | "time" |
| 13 | |
| 14 | "reasonix/internal/pluginpkg" |
| 15 | "reasonix/internal/proc" |
| 16 | "reasonix/internal/secrets" |
| 17 | ) |
| 18 | |
| 19 | // Bounded-close budgets, mirrored from internal/plugin's stdio transport: a |
| 20 | // short stdin-EOF grace for protocol-aware sidecars, then a hard tree kill |
| 21 | // with a bounded reap so one wedged sidecar can never stall teardown. |
| 22 | const ( |
| 23 | gracefulCloseWaitBudget = 750 * time.Millisecond |
| 24 | closeWaitBudget = 5 * time.Second |
| 25 | // stderrTailBytes bounds the ring of sidecar stderr retained for |
| 26 | // diagnostics, mirrored from internal/plugin's tailBuffer. |
| 27 | stderrTailBytes = 16 << 10 |
| 28 | ) |
| 29 | |
| 30 | // pluginEnvVarPrefix is the well-known environment block every sidecar sees. |
| 31 | const ( |
| 32 | envPluginRoot = "REASONIX_PLUGIN_ROOT" |
| 33 | envPluginName = "REASONIX_PLUGIN_NAME" |
| 34 | envPluginVersion = "REASONIX_PLUGIN_VERSION" |
| 35 | ) |
| 36 | |
| 37 | // shellExecutables are interpreter names a runtime command may not resolve |
| 38 | // to. The runtime contract is exec form: the command IS the extension |
| 39 | // executable and args are its argv. Routing through a shell would smuggle |
| 40 | // shell semantics (pipes, &&, $ expansion) into a contract that promises |
| 41 | // there are none. |
| 42 | var shellExecutables = map[string]bool{ |
| 43 | "sh": true, "bash": true, "zsh": true, "fish": true, "dash": true, "ksh": true, |
| 44 | "cmd": true, "cmd.exe": true, "powershell": true, "powershell.exe": true, |
| 45 | "pwsh": true, "pwsh.exe": true, |
| 46 | } |
| 47 | |
| 48 | // startupFailure carries bounded diagnostics for a sidecar that failed to |
| 49 | // start or hand shake: the stage, elapsed time, and the redacted stderr tail. |
| 50 | // The shape mirrors internal/plugin's startupFailure. |
| 51 | type startupFailure struct { |
| 52 | Stage string |
| 53 | Elapsed time.Duration |
| 54 | Stderr string |
| 55 | Err error |
| 56 | } |
| 57 | |
| 58 | func (e *startupFailure) Error() string { |
| 59 | if e == nil { |
| 60 | return "extension sidecar startup failed" |
| 61 | } |
| 62 | stage := strings.TrimSpace(e.Stage) |
| 63 | if stage == "" { |
| 64 | stage = "unknown" |
| 65 | } |
| 66 | msg := fmt.Sprintf("extension sidecar startup %s failed after %s: %s", stage, formatElapsed(e.Elapsed), secrets.RedactError(e.Err)) |
| 67 | if stderr := strings.TrimSpace(e.Stderr); stderr != "" { |
| 68 | msg += "; stderr: " + stderr |
| 69 | } |
| 70 | return msg |
| 71 | } |
| 72 | |
| 73 | func (e *startupFailure) Unwrap() error { |
| 74 | if e == nil { |
| 75 | return nil |
| 76 | } |
| 77 | return e.Err |
| 78 | } |
| 79 | |
| 80 | func newStartupFailure(stage string, started time.Time, stderr string, err error) error { |
| 81 | if err == nil { |
| 82 | return nil |
| 83 | } |
| 84 | var existing *startupFailure |
| 85 | if errors.As(err, &existing) { |
| 86 | return err |
| 87 | } |
| 88 | elapsed := time.Since(started) |
| 89 | if elapsed < 0 { |
| 90 | elapsed = 0 |
| 91 | } |
| 92 | return &startupFailure{ |
| 93 | Stage: strings.TrimSpace(stage), |
| 94 | Elapsed: elapsed, |
| 95 | Stderr: secrets.RedactCredentials(strings.TrimSpace(stderr)), |
| 96 | Err: err, |
| 97 | } |
| 98 | } |
| 99 | |
| 100 | func formatElapsed(elapsed time.Duration) string { |
| 101 | if elapsed < time.Millisecond { |
| 102 | return elapsed.String() |
| 103 | } |
| 104 | return elapsed.Round(time.Millisecond).String() |
| 105 | } |
| 106 | |
| 107 | // tailBuffer is a bounded ring holding the most recent stderr bytes. Writes |
| 108 | // never block the child; only the tail is ever surfaced, and only after |
| 109 | // credential redaction. |
| 110 | type tailBuffer struct { |
| 111 | mu sync.Mutex |
| 112 | limit int |
| 113 | buf []byte |
| 114 | } |
| 115 | |
| 116 | func (b *tailBuffer) Write(p []byte) (int, error) { |
| 117 | b.mu.Lock() |
| 118 | defer b.mu.Unlock() |
| 119 | b.buf = append(b.buf, p...) |
| 120 | if b.limit > 0 && len(b.buf) > b.limit { |
| 121 | b.buf = append([]byte(nil), b.buf[len(b.buf)-b.limit:]...) |
| 122 | } |
| 123 | return len(p), nil |
| 124 | } |
| 125 | |
| 126 | func (b *tailBuffer) String() string { |
| 127 | b.mu.Lock() |
| 128 | defer b.mu.Unlock() |
| 129 | return strings.TrimSpace(string(b.buf)) |
| 130 | } |
| 131 | |
| 132 | // process is one spawned sidecar OS process with its pipes and tracked |
| 133 | // process-tree handle. |
| 134 | type process struct { |
| 135 | pluginID string |
| 136 | cmd *exec.Cmd |
| 137 | job uintptr |
| 138 | stdin io.WriteCloser |
| 139 | stdout io.ReadCloser |
| 140 | stderr *tailBuffer |
| 141 | |
| 142 | waitOnce sync.Once |
| 143 | waitDone chan struct{} |
| 144 | jobOnce sync.Once |
| 145 | } |
| 146 | |
| 147 | // resolveRuntimeCommand expands ${REASONIX_PLUGIN_ROOT} and enforces the exec |
| 148 | // contract: the resolved command must be an absolute path to the extension |
| 149 | // executable itself, never a relative name (no PATH lookup — the package must |
| 150 | // know exactly what it runs) and never a shell. |
| 151 | func resolveRuntimeCommand(rt *pluginpkg.RuntimeSpec, root string) (string, error) { |
| 152 | command := strings.TrimSpace(pluginpkg.ExpandRuntimeCommand(rt.Command, root)) |
| 153 | if command == "" { |
| 154 | return "", errors.New("runtime command is empty after expansion") |
| 155 | } |
| 156 | if !filepath.IsAbs(command) { |
| 157 | return "", fmt.Errorf("runtime command %q is not an absolute path after expansion (use %s to address the installed package)", rt.Command, pluginpkg.PluginRootEnvVar) |
| 158 | } |
| 159 | base := strings.ToLower(filepath.Base(command)) |
| 160 | if shellExecutables[base] { |
| 161 | return "", fmt.Errorf("runtime command %q is a shell; the runtime contract is exec form (the command is the extension executable, args are its argv)", rt.Command) |
| 162 | } |
| 163 | return command, nil |
| 164 | } |
| 165 | |
| 166 | // runtimeEnv builds the sidecar's environment: the UNFILTERED inherited |
| 167 | // process environment (full-trust contract — see the package doc), the |
| 168 | // manifest's env, and the well-known plugin identity variables. Later entries |
| 169 | // win over earlier ones for duplicate keys (exec.Cmd.Env semantics), so the |
| 170 | // manifest can tune but the identity variables cannot be forged by it. |
| 171 | func runtimeEnv(rt *pluginpkg.RuntimeSpec, pkg pluginpkg.Package, installed pluginpkg.InstalledPlugin) []string { |
| 172 | env := append([]string(nil), os.Environ()...) |
| 173 | for key, value := range rt.Env { |
| 174 | env = append(env, key+"="+value) |
| 175 | } |
| 176 | version := strings.TrimSpace(installed.Version) |
| 177 | if version == "" { |
| 178 | version = strings.TrimSpace(pkg.Manifest.Version) |
| 179 | } |
| 180 | env = append(env, |
| 181 | envPluginRoot+"="+pkg.Root, |
| 182 | envPluginName+"="+installed.Name, |
| 183 | envPluginVersion+"="+version, |
| 184 | ) |
| 185 | return env |
| 186 | } |
| 187 | |
| 188 | // startProcess spawns the sidecar process. It never goes through a shell: |
| 189 | // exec.Command takes the resolved executable and the argv vector directly. |
| 190 | func startProcess(pkg pluginpkg.Package, installed pluginpkg.InstalledPlugin) (*process, error) { |
| 191 | started := time.Now() |
| 192 | rt := pkg.Manifest.Runtime |
| 193 | if rt == nil { |
| 194 | return nil, fmt.Errorf("plugin %q declares no runtime", installed.Name) |
| 195 | } |
| 196 | command, err := resolveRuntimeCommand(rt, pkg.Root) |
| 197 | if err != nil { |
| 198 | return nil, newStartupFailure("resolve", started, "", err) |
| 199 | } |
| 200 | cmd := exec.Command(command, rt.Args...) |
| 201 | cmd.Env = runtimeEnv(rt, pkg, installed) |
| 202 | proc.HideWindow(cmd) |
| 203 | |
| 204 | stdin, err := cmd.StdinPipe() |
| 205 | if err != nil { |
| 206 | return nil, newStartupFailure("pipes", started, "", err) |
| 207 | } |
| 208 | stdout, err := cmd.StdoutPipe() |
| 209 | if err != nil { |
| 210 | return nil, newStartupFailure("pipes", started, "", err) |
| 211 | } |
| 212 | stderr := &tailBuffer{limit: stderrTailBytes} |
| 213 | cmd.Stderr = stderr |
| 214 | |
| 215 | job, err := proc.StartTracked(cmd) |
| 216 | if err != nil { |
| 217 | return nil, newStartupFailure("spawn", started, stderr.String(), err) |
| 218 | } |
| 219 | p := &process{ |
| 220 | pluginID: installed.Name, |
| 221 | cmd: cmd, |
| 222 | job: job, |
| 223 | stdin: stdin, |
| 224 | stdout: stdout, |
| 225 | stderr: stderr, |
| 226 | waitDone: make(chan struct{}), |
| 227 | } |
| 228 | return p, nil |
| 229 | } |
| 230 | |
| 231 | // wait blocks until the process exits, exactly once; later callers observe |
| 232 | // the same completed wait. Safe to abandon: the first caller owns cmd.Wait. |
| 233 | func (p *process) wait() { |
| 234 | p.waitOnce.Do(func() { |
| 235 | if p.cmd != nil && p.cmd.Process != nil { |
| 236 | _ = p.cmd.Wait() |
| 237 | } |
| 238 | p.finishJob() |
| 239 | close(p.waitDone) |
| 240 | }) |
| 241 | } |
| 242 | |
| 243 | // finishJob releases the Windows Job Object once the process is known to be |
| 244 | // gone (a no-op off Windows and after KillTracked already released it). |
| 245 | func (p *process) finishJob() { |
| 246 | p.jobOnce.Do(func() { proc.FinishTracked(p.job) }) |
| 247 | } |
| 248 | |
| 249 | // kill terminates the whole process tree. |
| 250 | func (p *process) kill() { |
| 251 | if p.cmd == nil || p.cmd.Process == nil { |
| 252 | return |
| 253 | } |
| 254 | proc.KillTracked(p.cmd, p.job) |
| 255 | p.finishJob() |
| 256 | } |
| 257 | |
| 258 | // close stops the sidecar with the bounded sequence: close stdin, grant a |
| 259 | // short EOF grace for protocol-aware processes, kill the tree, and wait a |
| 260 | // bounded time for the reap. It is idempotent and never blocks longer than |
| 261 | // gracefulCloseWaitBudget + closeWaitBudget. |
| 262 | func (p *process) close() { |
| 263 | if p.stdin != nil { |
| 264 | _ = p.stdin.Close() |
| 265 | } |
| 266 | if p.cmd == nil || p.cmd.Process == nil { |
| 267 | return |
| 268 | } |
| 269 | if waitFinishedWithinBudget(p.wait, gracefulCloseWaitBudget) { |
| 270 | return |
| 271 | } |
| 272 | p.kill() |
| 273 | waitWithBudget(p.wait, closeWaitBudget) |
| 274 | } |
| 275 | |
| 276 | func waitWithBudget(wait func(), budget time.Duration) { |
| 277 | _ = waitFinishedWithinBudget(wait, budget) |
| 278 | } |
| 279 | |
| 280 | func waitFinishedWithinBudget(wait func(), budget time.Duration) bool { |
| 281 | done := make(chan struct{}) |
| 282 | go func() { wait(); close(done) }() |
| 283 | select { |
| 284 | case <-done: |
| 285 | return true |
| 286 | case <-time.After(budget): |
| 287 | return false |
| 288 | } |
| 289 | } |
| 290 |