diff --git a/weed/storage/blockvol/iscsi/cmd/blockbench/main.go b/weed/storage/blockvol/iscsi/cmd/blockbench/main.go
new file mode 100644
index 000000000..869d93edd
--- /dev/null
+++ b/weed/storage/blockvol/iscsi/cmd/blockbench/main.go
@@ -0,0 +1,189 @@
+// blockbench is a simple block device benchmark tool.
+// Usage: blockbench [-f path] [-size 256M] [-bs 4K] [-duration 10s] [-pattern seq|rand] [-rw read|write|mixed]
+package main
+
+import (
+	"crypto/rand"
+	"flag"
+	"fmt"
+	"math/big"
+	"os"
+	"strings"
+	"sync/atomic"
+	"time"
+)
+
+func main() {
+	filePath := flag.String("f", "", "test file path (e.g. F:\\testfile.dat)")
+	fileSize := flag.String("size", "256M", "test file size")
+	blockSize := flag.String("bs", "4K", "block size")
+	duration := flag.Duration("duration", 10*time.Second, "test duration")
+	pattern := flag.String("pattern", "seq", "access pattern: seq or rand")
+	mode := flag.String("rw", "write", "read, write, or mixed")
+	workers := flag.Int("workers", 1, "number of concurrent workers")
+	direct := flag.Bool("direct", false, "use O_DIRECT (not supported on all OS)")
+	flag.Parse()
+
+	if *filePath == "" {
+		fmt.Fprintln(os.Stderr, "error: -f is required")
+		flag.Usage()
+		os.Exit(1)
+	}
+
+	fSize, err := parseSize(*fileSize)
+	if err != nil {
+		fmt.Fprintf(os.Stderr, "bad -size: %v\n", err)
+		os.Exit(1)
+	}
+	bSize, err := parseSize(*blockSize)
+	if err != nil {
+		fmt.Fprintf(os.Stderr, "bad -bs: %v\n", err)
+		os.Exit(1)
+	}
+
+	_ = *direct // placeholder for future O_DIRECT support
+
+	fmt.Printf("=== BlockBench ===\n")
+	fmt.Printf("file=%s size=%s bs=%s duration=%s pattern=%s mode=%s workers=%d\n",
+		*filePath, *fileSize, *blockSize, *duration, *pattern, *mode, *workers)
+
+	// Create/open file
+	f, err := os.OpenFile(*filePath, os.O_RDWR|os.O_CREATE, 0644)
+	if err != nil {
+		fmt.Fprintf(os.Stderr, "open: %v\n", err)
+		os.Exit(1)
+	}
+	defer f.Close()
+
+	// Pre-allocate
+	if err := f.Truncate(int64(fSize)); err != nil {
+		fmt.Fprintf(os.Stderr, "truncate: %v\n", err)
+		os.Exit(1)
+	}
+
+	totalBlocks := fSize / bSize
+
+	// Prepare write buffer
+	writeBuf := make([]byte, bSize)
+	rand.Read(writeBuf)
+
+	var ops atomic.Int64
+	var bytes atomic.Int64
+	var maxLatUs atomic.Int64
+
+	isRand := *pattern == "rand"
+	doWrite := *mode == "write" || *mode == "mixed"
+	doRead := *mode == "read" || *mode == "mixed"
+
+	stop := make(chan struct{})
+
+	fmt.Printf("running %s %s for %s...\n", *pattern, *mode, *duration)
+
+	start := time.Now()
+
+	for w := 0; w < *workers; w++ {
+		wf, err := os.OpenFile(*filePath, os.O_RDWR, 0644)
+		if err != nil {
+			fmt.Fprintf(os.Stderr, "worker open: %v\n", err)
+			os.Exit(1)
+		}
+		defer wf.Close()
+
+		go func(wf *os.File, workerID int) {
+			buf := make([]byte, bSize)
+			copy(buf, writeBuf)
+			var seqBlock uint64
+			readBuf := make([]byte, bSize)
+
+			for {
+				select {
+				case <-stop:
+					return
+				default:
+				}
+
+				var blockNum uint64
+				if isRand {
+					n, _ := rand.Int(rand.Reader, big.NewInt(int64(totalBlocks)))
+					blockNum = n.Uint64()
+				} else {
+					blockNum = seqBlock % totalBlocks
+					seqBlock++
+				}
+
+				off := int64(blockNum * bSize)
+				opStart := time.Now()
+
+				if doWrite && (!doRead || seqBlock%2 == 0) {
+					_, err = wf.WriteAt(buf, off)
+				} else if doRead {
+					_, err = wf.ReadAt(readBuf, off)
+				}
+
+				us := time.Since(opStart).Microseconds()
+				if err == nil {
+					ops.Add(1)
+					bytes.Add(int64(bSize))
+					for {
+						old := maxLatUs.Load()
+						if us <= old || maxLatUs.CompareAndSwap(old, us) {
+							break
+						}
+					}
+				}
+			}
+		}(wf, w)
+	}
+
+	time.Sleep(*duration)
+	close(stop)
+	elapsed := time.Since(start)
+
+	// Small sleep to let workers finish
+	time.Sleep(10 * time.Millisecond)
+
+	totalOps := ops.Load()
+	totalBytes := bytes.Load()
+	maxLat := maxLatUs.Load()
+
+	iops := float64(totalOps) / elapsed.Seconds()
+	mbps := float64(totalBytes) / (1024 * 1024) / elapsed.Seconds()
+	avgUs := int64(0)
+	if totalOps > 0 {
+		avgUs = int64(elapsed.Microseconds()) / totalOps * int64(*workers)
+		// Rough approximation — actual avg needs per-op tracking
+		avgUs = int64(float64(elapsed.Microseconds()) / float64(totalOps) * float64(*workers))
+	}
+
+	fmt.Printf("\n=== Results ===\n")
+	fmt.Printf("ops:       %d\n", totalOps)
+	fmt.Printf("IOPS:      %.0f\n", iops)
+	fmt.Printf("throughput: %.1f MB/s\n", mbps)
+	fmt.Printf("max_lat:   %d us (%.1f ms)\n", maxLat, float64(maxLat)/1000)
+	fmt.Printf("elapsed:   %.1fs\n", elapsed.Seconds())
+	if totalOps > 0 {
+		fmt.Printf("avg_lat:   ~%d us (%.2f ms) [estimated]\n", avgUs, float64(avgUs)/1000)
+	}
+}
+
+func parseSize(s string) (uint64, error) {
+	s = strings.TrimSpace(s)
+	if len(s) == 0 {
+		return 0, fmt.Errorf("empty size")
+	}
+	mul := uint64(1)
+	switch s[len(s)-1] {
+	case 'K', 'k':
+		mul = 1024
+		s = s[:len(s)-1]
+	case 'M', 'm':
+		mul = 1024 * 1024
+		s = s[:len(s)-1]
+	case 'G', 'g':
+		mul = 1024 * 1024 * 1024
+		s = s[:len(s)-1]
+	}
+	var n uint64
+	_, err := fmt.Sscanf(s, "%d", &n)
+	return n * mul, err
+}
diff --git a/weed/storage/blockvol/iscsi/cmd/iscsi-target/admin.go b/weed/storage/blockvol/iscsi/cmd/iscsi-target/admin.go
index 0e7469a0a..39aac99fe 100644
--- a/weed/storage/blockvol/iscsi/cmd/iscsi-target/admin.go
+++ b/weed/storage/blockvol/iscsi/cmd/iscsi-target/admin.go
@@ -12,6 +12,7 @@ import (
 	"log"
 	"net"
 	"net/http"
+	_ "net/http/pprof" // registers /debug/pprof/* handlers on DefaultServeMux
 	"time"
 
 	"github.com/seaweedfs/seaweedfs/weed/storage/blockvol"
@@ -189,13 +190,21 @@ func (a *adminServer) handleRebuild(w http.ResponseWriter, r *http.Request) {
 
 // startAdminServer starts the HTTP admin server in a background goroutine.
 // Returns the listener so tests can determine the actual bound port.
+// Includes /debug/pprof/* endpoints for profiling.
 func startAdminServer(addr string, srv *adminServer) (net.Listener, error) {
 	ln, err := net.Listen("tcp", addr)
 	if err != nil {
 		return nil, fmt.Errorf("admin listen %s: %w", addr, err)
 	}
+	mux := http.NewServeMux()
+	mux.Handle("/assign", srv)
+	mux.Handle("/status", srv)
+	mux.Handle("/replica", srv)
+	mux.Handle("/rebuild", srv)
+	// pprof handlers registered on DefaultServeMux by net/http/pprof import.
+	mux.Handle("/debug/pprof/", http.DefaultServeMux)
 	go func() {
-		if err := http.Serve(ln, srv); err != nil && !isClosedErr(err) {
+		if err := http.Serve(ln, mux); err != nil && !isClosedErr(err) {
 			srv.logger.Printf("admin server error: %v", err)
 		}
 	}()
diff --git a/weed/storage/blockvol/iscsi/cmd/iscsi-target/main.go b/weed/storage/blockvol/iscsi/cmd/iscsi-target/main.go
index d183d2de4..8a1979ffb 100644
--- a/weed/storage/blockvol/iscsi/cmd/iscsi-target/main.go
+++ b/weed/storage/blockvol/iscsi/cmd/iscsi-target/main.go
@@ -13,7 +13,9 @@ import (
 	"os/signal"
 	"strconv"
 	"strings"
+	"sync/atomic"
 	"syscall"
+	"time"
 
 	"github.com/seaweedfs/seaweedfs/weed/storage/blockvol"
 	"github.com/seaweedfs/seaweedfs/weed/storage/blockvol/iscsi"
@@ -98,19 +100,33 @@ func main() {
 		logger.Printf("admin server: %s", ln.Addr())
 	}
 
-	// Create adapter
-	adapter := &blockVolAdapter{vol: vol}
+	// Create adapter with latency instrumentation
+	adapter := &instrumentedAdapter{
+		inner:  &blockVolAdapter{vol: vol},
+		logger: logger,
+	}
 
 	// Create target server
 	config := iscsi.DefaultTargetConfig()
 	config.TargetName = *iqn
 	config.TargetAlias = "SeaweedFS BlockVol"
+	if *portal != "" {
+		// Parse portal group tag from "addr:port,tpgt" format
+		if idx := strings.LastIndex(*portal, ","); idx >= 0 {
+			if tpgt, err := strconv.Atoi((*portal)[idx+1:]); err == nil {
+				config.TargetPortalGroupTag = tpgt
+			}
+		}
+	}
 	ts := iscsi.NewTargetServer(*addr, config, logger)
 	if *portal != "" {
 		ts.SetPortalAddr(*portal)
 	}
 	ts.AddVolume(*iqn, adapter)
 
+	// Start periodic performance stats logging (every 5 seconds).
+	adapter.StartStatsLogger(5 * time.Second)
+
 	// Graceful shutdown on signal
 	sigCh := make(chan os.Signal, 1)
 	signal.Notify(sigCh, syscall.SIGINT, syscall.SIGTERM)
@@ -148,6 +164,119 @@ func (a *blockVolAdapter) BlockSize() uint32  { return a.vol.Info().BlockSize }
 func (a *blockVolAdapter) VolumeSize() uint64 { return a.vol.Info().VolumeSize }
 func (a *blockVolAdapter) IsHealthy() bool    { return a.vol.Info().Healthy }
 
+// instrumentedAdapter wraps a BlockDevice and logs latency stats periodically.
+type instrumentedAdapter struct {
+	inner  iscsi.BlockDevice
+	logger *log.Logger
+
+	// Counters (atomic, lock-free)
+	writeOps   atomic.Int64
+	readOps    atomic.Int64
+	syncOps    atomic.Int64
+	writeTotUs atomic.Int64 // total microseconds
+	readTotUs  atomic.Int64
+	syncTotUs  atomic.Int64
+	writeMaxUs atomic.Int64
+	readMaxUs  atomic.Int64
+	syncMaxUs  atomic.Int64
+	writeBytes atomic.Int64
+	readBytes  atomic.Int64
+}
+
+func (a *instrumentedAdapter) ReadAt(lba uint64, length uint32) ([]byte, error) {
+	start := time.Now()
+	data, err := a.inner.ReadAt(lba, length)
+	us := time.Since(start).Microseconds()
+	a.readOps.Add(1)
+	a.readTotUs.Add(us)
+	a.readBytes.Add(int64(length))
+	atomicMax(&a.readMaxUs, us)
+	return data, err
+}
+
+func (a *instrumentedAdapter) WriteAt(lba uint64, data []byte) error {
+	start := time.Now()
+	err := a.inner.WriteAt(lba, data)
+	us := time.Since(start).Microseconds()
+	a.writeOps.Add(1)
+	a.writeTotUs.Add(us)
+	a.writeBytes.Add(int64(len(data)))
+	atomicMax(&a.writeMaxUs, us)
+	return err
+}
+
+func (a *instrumentedAdapter) Trim(lba uint64, length uint32) error {
+	return a.inner.Trim(lba, length)
+}
+
+func (a *instrumentedAdapter) SyncCache() error {
+	start := time.Now()
+	err := a.inner.SyncCache()
+	us := time.Since(start).Microseconds()
+	a.syncOps.Add(1)
+	a.syncTotUs.Add(us)
+	atomicMax(&a.syncMaxUs, us)
+	return err
+}
+
+func (a *instrumentedAdapter) BlockSize() uint32  { return a.inner.BlockSize() }
+func (a *instrumentedAdapter) VolumeSize() uint64 { return a.inner.VolumeSize() }
+func (a *instrumentedAdapter) IsHealthy() bool    { return a.inner.IsHealthy() }
+
+// StartStatsLogger runs a goroutine that logs performance stats every interval.
+func (a *instrumentedAdapter) StartStatsLogger(interval time.Duration) {
+	go func() {
+		ticker := time.NewTicker(interval)
+		defer ticker.Stop()
+		for range ticker.C {
+			wr := a.writeOps.Swap(0)
+			rd := a.readOps.Swap(0)
+			sy := a.syncOps.Swap(0)
+			wrUs := a.writeTotUs.Swap(0)
+			rdUs := a.readTotUs.Swap(0)
+			syUs := a.syncTotUs.Swap(0)
+			wrMax := a.writeMaxUs.Swap(0)
+			rdMax := a.readMaxUs.Swap(0)
+			syMax := a.syncMaxUs.Swap(0)
+			wrBytes := a.writeBytes.Swap(0)
+			rdBytes := a.readBytes.Swap(0)
+
+			if wr+rd+sy == 0 {
+				continue // quiet when idle
+			}
+
+			wrAvg, rdAvg, syAvg := int64(0), int64(0), int64(0)
+			if wr > 0 {
+				wrAvg = wrUs / wr
+			}
+			if rd > 0 {
+				rdAvg = rdUs / rd
+			}
+			if sy > 0 {
+				syAvg = syUs / sy
+			}
+			a.logger.Printf("PERF[%ds] wr=%d(%.1fMB avg=%dus max=%dus) rd=%d(%.1fMB avg=%dus max=%dus) sync=%d(avg=%dus max=%dus)",
+				int(interval.Seconds()),
+				wr, float64(wrBytes)/(1024*1024), wrAvg, wrMax,
+				rd, float64(rdBytes)/(1024*1024), rdAvg, rdMax,
+				sy, syAvg, syMax,
+			)
+		}
+	}()
+}
+
+func atomicMax(addr *atomic.Int64, val int64) {
+	for {
+		old := addr.Load()
+		if val <= old {
+			return
+		}
+		if addr.CompareAndSwap(old, val) {
+			return
+		}
+	}
+}
+
 func parseSize(s string) (uint64, error) {
 	s = strings.TrimSpace(s)
 	if len(s) == 0 {
diff --git a/weed/storage/blockvol/iscsi/dataio.go b/weed/storage/blockvol/iscsi/dataio.go
index 94873075f..fbbef95fa 100644
--- a/weed/storage/blockvol/iscsi/dataio.go
+++ b/weed/storage/blockvol/iscsi/dataio.go
@@ -27,16 +27,21 @@ func NewDataInWriter(maxSegLen uint32) *DataInWriter {
 }
 
 // WriteDataIn splits data into Data-In PDUs and writes them to w.
-// itt is the initiator task tag, ttt is the target transfer tag.
+// itt is the initiator task tag, edtl is expected data transfer length.
 // statSN is the current StatSN (incremented when S-bit is set).
 // Returns the number of PDUs written.
-func (d *DataInWriter) WriteDataIn(w io.Writer, data []byte, itt uint32, expCmdSN, maxCmdSN uint32, statSN *uint32) (int, error) {
+func (d *DataInWriter) WriteDataIn(w io.Writer, data []byte, itt, edtl uint32, expCmdSN, maxCmdSN uint32, statSN *uint32) (int, error) {
 	totalLen := uint32(len(data))
 	if totalLen == 0 {
 		// Zero-length read -- send single Data-In with S-bit, no data
 		pdu := &PDU{}
 		pdu.SetOpcode(OpSCSIDataIn)
-		pdu.SetOpSpecific1(FlagF | FlagS) // Final + Status
+		flags := uint8(FlagF | FlagS)
+		if edtl > 0 {
+			flags |= FlagU
+			pdu.SetResidualCount(edtl)
+		}
+		pdu.SetOpSpecific1(flags)
 		pdu.SetInitiatorTaskTag(itt)
 		pdu.SetTargetTransferTag(0xFFFFFFFF)
 		pdu.SetStatSN(*statSN)
@@ -74,7 +79,15 @@ func (d *DataInWriter) WriteDataIn(w io.Writer, data []byte, itt uint32, expCmdS
 		pdu.DataSegment = data[offset : offset+segLen]
 
 		if isFinal {
-			pdu.SetOpSpecific1(FlagF | FlagS) // Final + Status
+			flags := uint8(FlagF | FlagS)
+			if totalLen < edtl {
+				flags |= FlagU
+				pdu.SetResidualCount(edtl - totalLen)
+			} else if totalLen > edtl {
+				flags |= FlagO
+				pdu.SetResidualCount(totalLen - edtl)
+			}
+			pdu.SetOpSpecific1(flags)
 			pdu.SetStatSN(*statSN)
 			*statSN++
 			pdu.SetSCSIStatus(SCSIStatusGood)
@@ -219,14 +232,20 @@ func BuildSCSIResponse(result SCSIResult, itt uint32, expCmdSN, maxCmdSN uint32)
 }
 
 // BuildDataInPDUs splits data into Data-In PDUs and returns them.
+// edtl is the Expected Data Transfer Length from the SCSI command PDU.
 // StatSN is NOT set on the final PDU -- the txLoop assigns it.
 // Intermediate PDUs (without S-bit) never carry StatSN.
-func (d *DataInWriter) BuildDataInPDUs(data []byte, itt uint32, expCmdSN, maxCmdSN uint32) []*PDU {
+func (d *DataInWriter) BuildDataInPDUs(data []byte, itt, edtl uint32, expCmdSN, maxCmdSN uint32) []*PDU {
 	totalLen := uint32(len(data))
 	if totalLen == 0 {
 		pdu := &PDU{}
 		pdu.SetOpcode(OpSCSIDataIn)
-		pdu.SetOpSpecific1(FlagF | FlagS)
+		flags := uint8(FlagF | FlagS)
+		if edtl > 0 {
+			flags |= FlagU // underflow: expected data but got none
+			pdu.SetResidualCount(edtl)
+		}
+		pdu.SetOpSpecific1(flags)
 		pdu.SetInitiatorTaskTag(itt)
 		pdu.SetTargetTransferTag(0xFFFFFFFF)
 		pdu.SetExpCmdSN(expCmdSN)
@@ -262,7 +281,16 @@ func (d *DataInWriter) BuildDataInPDUs(data []byte, itt uint32, expCmdSN, maxCmd
 		pdu.DataSegment = seg
 
 		if isFinal {
-			pdu.SetOpSpecific1(FlagF | FlagS)
+			flags := uint8(FlagF | FlagS)
+			// RFC 7143: set underflow/overflow residual on final Data-In with S-bit.
+			if totalLen < edtl {
+				flags |= FlagU
+				pdu.SetResidualCount(edtl - totalLen)
+			} else if totalLen > edtl {
+				flags |= FlagO
+				pdu.SetResidualCount(totalLen - edtl)
+			}
+			pdu.SetOpSpecific1(flags)
 			pdu.SetSCSIStatus(SCSIStatusGood)
 		} else {
 			pdu.SetOpSpecific1(0)
diff --git a/weed/storage/blockvol/iscsi/dataio_test.go b/weed/storage/blockvol/iscsi/dataio_test.go
index ab314f9d2..7aa0780e4 100644
--- a/weed/storage/blockvol/iscsi/dataio_test.go
+++ b/weed/storage/blockvol/iscsi/dataio_test.go
@@ -40,7 +40,7 @@ func testDataInSinglePDU(t *testing.T) {
 	data := bytes.Repeat([]byte{0xAA}, 4096)
 	statSN := uint32(1)
 
-	n, err := dw.WriteDataIn(w, data, 0x100, 1, 10, &statSN)
+	n, err := dw.WriteDataIn(w, data, 0x100, uint32(len(data)), 1, 10, &statSN)
 	if err != nil {
 		t.Fatal(err)
 	}
@@ -75,7 +75,7 @@ func testDataInMultiPDU(t *testing.T) {
 	data := bytes.Repeat([]byte{0xBB}, 3000)
 	statSN := uint32(1)
 
-	n, err := dw.WriteDataIn(w, data, 0x200, 1, 10, &statSN)
+	n, err := dw.WriteDataIn(w, data, 0x200, uint32(len(data)), 1, 10, &statSN)
 	if err != nil {
 		t.Fatal(err)
 	}
@@ -121,7 +121,7 @@ func testDataInExactBoundary(t *testing.T) {
 	data := bytes.Repeat([]byte{0xCC}, 2048) // exact 2 PDUs
 	statSN := uint32(1)
 
-	n, err := dw.WriteDataIn(w, data, 0x300, 1, 10, &statSN)
+	n, err := dw.WriteDataIn(w, data, 0x300, uint32(len(data)), 1, 10, &statSN)
 	if err != nil {
 		t.Fatal(err)
 	}
@@ -135,7 +135,7 @@ func testDataInZeroLength(t *testing.T) {
 	dw := NewDataInWriter(8192)
 	statSN := uint32(5)
 
-	n, err := dw.WriteDataIn(w, nil, 0x400, 1, 10, &statSN)
+	n, err := dw.WriteDataIn(w, nil, 0x400, 0, 1, 10, &statSN)
 	if err != nil {
 		t.Fatal(err)
 	}
@@ -153,7 +153,7 @@ func testDataInDataSNOrdering(t *testing.T) {
 	data := bytes.Repeat([]byte{0xDD}, 2048) // 4 PDUs
 	statSN := uint32(1)
 
-	dw.WriteDataIn(w, data, 0x500, 1, 10, &statSN)
+	dw.WriteDataIn(w, data, 0x500, uint32(len(data)), 1, 10, &statSN)
 
 	for i := 0; i < 4; i++ {
 		pdu, _ := ReadPDU(w)
@@ -173,7 +173,7 @@ func testDataInFbitSbit(t *testing.T) {
 	data := bytes.Repeat([]byte{0xEE}, 2500)
 	statSN := uint32(1)
 
-	dw.WriteDataIn(w, data, 0x600, 1, 10, &statSN)
+	dw.WriteDataIn(w, data, 0x600, uint32(len(data)), 1, 10, &statSN)
 
 	for i := 0; i < 3; i++ {
 		pdu, _ := ReadPDU(w)
@@ -399,7 +399,7 @@ func testDataInStatSNIncrement(t *testing.T) {
 	data := bytes.Repeat([]byte{0x00}, 3072) // 3 PDUs
 	statSN := uint32(100)
 
-	dw.WriteDataIn(w, data, 0x700, 1, 10, &statSN)
+	dw.WriteDataIn(w, data, 0x700, uint32(len(data)), 1, 10, &statSN)
 	// Only the final PDU has S-bit, so StatSN increments once
 	if statSN != 101 {
 		t.Fatalf("StatSN should be 101, got %d", statSN)
diff --git a/weed/storage/blockvol/iscsi/login.go b/weed/storage/blockvol/iscsi/login.go
index 9d6f22d8e..19062915d 100644
--- a/weed/storage/blockvol/iscsi/login.go
+++ b/weed/storage/blockvol/iscsi/login.go
@@ -58,6 +58,7 @@ const (
 type TargetConfig struct {
 	TargetName              string
 	TargetAlias             string
+	TargetPortalGroupTag    int // TPGT (1-65535); 0 = use default 1
 	MaxRecvDataSegmentLength int
 	MaxBurstLength          int
 	FirstBurstLength        int
@@ -272,6 +273,13 @@ func (ln *LoginNegotiator) HandleLoginPDU(req *PDU, resolver TargetResolver) *PD
 	}
 	resp.SetTSIH(ln.tsih)
 
+	// RFC 7143 Section 13.9: All Login Responses MUST carry TargetPortalGroupTag.
+	tpgt := ln.config.TargetPortalGroupTag
+	if tpgt <= 0 {
+		tpgt = 1
+	}
+	respParams.Set("TargetPortalGroupTag", strconv.Itoa(tpgt))
+
 	// Encode response params
 	if respParams.Len() > 0 {
 		resp.DataSegment = respParams.Encode()
diff --git a/weed/storage/blockvol/iscsi/scsi.go b/weed/storage/blockvol/iscsi/scsi.go
index f42d0ee78..281676493 100644
--- a/weed/storage/blockvol/iscsi/scsi.go
+++ b/weed/storage/blockvol/iscsi/scsi.go
@@ -6,21 +6,28 @@ import (
 
 // SCSI opcode constants (SPC-5 / SBC-4)
 const (
-	ScsiTestUnitReady  uint8 = 0x00
-	ScsiInquiry        uint8 = 0x12
-	ScsiModeSense6     uint8 = 0x1a
-	ScsiModeSense10    uint8 = 0x5a
-	ScsiReadCapacity10 uint8 = 0x25
-	ScsiRead10         uint8 = 0x28
-	ScsiWrite10        uint8 = 0x2a
-	ScsiSyncCache10    uint8 = 0x35
-	ScsiUnmap          uint8 = 0x42
-	ScsiReportLuns     uint8 = 0xa0
-	ScsiRead16         uint8 = 0x88
-	ScsiWrite16        uint8 = 0x8a
-	ScsiReadCapacity16 uint8 = 0x9e // SERVICE ACTION IN (16), SA=0x10
-	ScsiSyncCache16    uint8 = 0x91
-	ScsiWriteSame16    uint8 = 0x93
+	ScsiTestUnitReady      uint8 = 0x00
+	ScsiRequestSense       uint8 = 0x03
+	ScsiInquiry            uint8 = 0x12
+	ScsiModeSelect6        uint8 = 0x15
+	ScsiModeSense6         uint8 = 0x1a
+	ScsiStartStopUnit      uint8 = 0x1b
+	ScsiReadCapacity10     uint8 = 0x25
+	ScsiRead10             uint8 = 0x28
+	ScsiWrite10            uint8 = 0x2a
+	ScsiSyncCache10        uint8 = 0x35
+	ScsiUnmap              uint8 = 0x42
+	ScsiModeSelect10       uint8 = 0x55
+	ScsiModeSense10        uint8 = 0x5a
+	ScsiPersistReserveIn   uint8 = 0x5e
+	ScsiPersistReserveOut  uint8 = 0x5f
+	ScsiRead16             uint8 = 0x88
+	ScsiWrite16            uint8 = 0x8a
+	ScsiSyncCache16        uint8 = 0x91
+	ScsiWriteSame16        uint8 = 0x93
+	ScsiServiceActionIn16  uint8 = 0x9e // READ CAPACITY(16), etc.
+	ScsiReportLuns         uint8 = 0xa0
+	ScsiMaintenanceIn      uint8 = 0xa3 // REPORT SUPPORTED OPCODES, etc.
 )
 
 // Service action for READ CAPACITY (16)
@@ -93,38 +100,52 @@ func (h *SCSIHandler) HandleCommand(cdb [16]byte, dataOut []byte) SCSIResult {
 	switch opcode {
 	case ScsiTestUnitReady:
 		return h.testUnitReady()
+	case ScsiRequestSense:
+		return h.requestSense(cdb)
 	case ScsiInquiry:
 		return h.inquiry(cdb)
+	case ScsiModeSelect6:
+		return h.modeSelect6(cdb, dataOut)
 	case ScsiModeSense6:
 		return h.modeSense6(cdb)
-	case ScsiModeSense10:
-		return h.modeSense10(cdb)
+	case ScsiStartStopUnit:
+		return h.startStopUnit(cdb)
 	case ScsiReadCapacity10:
 		return h.readCapacity10()
-	case ScsiReadCapacity16:
-		sa := cdb[1] & 0x1f
-		if sa == ScsiSAReadCapacity16 {
-			return h.readCapacity16(cdb)
-		}
-		return illegalRequest(ASCInvalidOpcode, ASCQLuk)
-	case ScsiReportLuns:
-		return h.reportLuns(cdb)
 	case ScsiRead10:
 		return h.read10(cdb)
-	case ScsiRead16:
-		return h.read16(cdb)
 	case ScsiWrite10:
 		return h.write10(cdb, dataOut)
-	case ScsiWrite16:
-		return h.write16(cdb, dataOut)
 	case ScsiSyncCache10:
 		return h.syncCache()
-	case ScsiSyncCache16:
-		return h.syncCache()
 	case ScsiUnmap:
 		return h.unmap(cdb, dataOut)
+	case ScsiModeSelect10:
+		return h.modeSelect10(cdb, dataOut)
+	case ScsiModeSense10:
+		return h.modeSense10(cdb)
+	case ScsiPersistReserveIn:
+		return h.persistReserveIn(cdb)
+	case ScsiPersistReserveOut:
+		return h.persistReserveOut(cdb, dataOut)
+	case ScsiRead16:
+		return h.read16(cdb)
+	case ScsiWrite16:
+		return h.write16(cdb, dataOut)
+	case ScsiSyncCache16:
+		return h.syncCache()
 	case ScsiWriteSame16:
 		return h.writeSame16(cdb, dataOut)
+	case ScsiServiceActionIn16:
+		sa := cdb[1] & 0x1f
+		if sa == ScsiSAReadCapacity16 {
+			return h.readCapacity16(cdb)
+		}
+		return illegalRequest(ASCInvalidOpcode, ASCQLuk)
+	case ScsiReportLuns:
+		return h.reportLuns(cdb)
+	case ScsiMaintenanceIn:
+		return h.maintenanceIn(cdb)
 	default:
 		return illegalRequest(ASCInvalidOpcode, ASCQLuk)
 	}
@@ -144,6 +165,64 @@ func (h *SCSIHandler) testUnitReady() SCSIResult {
 	return SCSIResult{Status: SCSIStatusGood}
 }
 
+func (h *SCSIHandler) requestSense(cdb [16]byte) SCSIResult {
+	allocLen := cdb[4]
+	if allocLen == 0 {
+		allocLen = 18
+	}
+
+	// Return fixed-format sense data with NO SENSE (no pending error).
+	data := BuildSenseData(SenseNoSense, 0x00, 0x00)
+	if int(allocLen) < len(data) {
+		data = data[:allocLen]
+	}
+	return SCSIResult{Status: SCSIStatusGood, Data: data}
+}
+
+// startStopUnit handles START STOP UNIT (0x1B).
+// Windows sends this during disk init. We accept and ignore.
+func (h *SCSIHandler) startStopUnit(cdb [16]byte) SCSIResult {
+	return SCSIResult{Status: SCSIStatusGood}
+}
+
+// modeSelect6 handles MODE SELECT(6) (0x15).
+// Windows sends this to set caching mode pages. We accept and ignore the data.
+func (h *SCSIHandler) modeSelect6(cdb [16]byte, dataOut []byte) SCSIResult {
+	return SCSIResult{Status: SCSIStatusGood}
+}
+
+// modeSelect10 handles MODE SELECT(10) (0x55).
+// Same as modeSelect6 but with 10-byte CDB.
+func (h *SCSIHandler) modeSelect10(cdb [16]byte, dataOut []byte) SCSIResult {
+	return SCSIResult{Status: SCSIStatusGood}
+}
+
+// persistReserveIn handles PERSISTENT RESERVE IN (0x5E).
+// Windows Cluster and MPIO use this. We don't support reservations,
+// so return ILLEGAL_REQUEST with a specific ASC that tells Windows
+// "not supported" rather than "broken device".
+func (h *SCSIHandler) persistReserveIn(cdb [16]byte) SCSIResult {
+	return illegalRequest(ASCInvalidOpcode, ASCQLuk)
+}
+
+// persistReserveOut handles PERSISTENT RESERVE OUT (0x5F).
+func (h *SCSIHandler) persistReserveOut(cdb [16]byte, dataOut []byte) SCSIResult {
+	return illegalRequest(ASCInvalidOpcode, ASCQLuk)
+}
+
+// maintenanceIn handles MAINTENANCE IN (0xA3).
+// Service action 0x0C = REPORT SUPPORTED OPERATION CODES.
+// Windows sends this to discover which commands we support.
+func (h *SCSIHandler) maintenanceIn(cdb [16]byte) SCSIResult {
+	sa := cdb[1] & 0x1f
+	if sa == 0x0c {
+		// REPORT SUPPORTED OPERATION CODES -- return empty (not supported).
+		// This tells Windows to probe commands individually.
+		return illegalRequest(ASCInvalidOpcode, ASCQLuk)
+	}
+	return illegalRequest(ASCInvalidOpcode, ASCQLuk)
+}
+
 func (h *SCSIHandler) inquiry(cdb [16]byte) SCSIResult {
 	evpd := cdb[1] & 0x01
 	pageCode := cdb[2]
@@ -353,17 +432,22 @@ func (h *SCSIHandler) readCapacity16(cdb [16]byte) SCSIResult {
 }
 
 func (h *SCSIHandler) modeSense6(cdb [16]byte) SCSIResult {
-	// Minimal MODE SENSE(6) response -- no mode pages
 	allocLen := cdb[4]
 	if allocLen == 0 {
 		allocLen = 4
 	}
+	pageCode := cdb[2] & 0x3f
 
-	data := make([]byte, 4)
-	data[0] = 3    // Mode data length (3 bytes follow)
-	data[1] = 0x00 // Medium type: default
-	data[2] = 0x00 // Device-specific parameter (no write protect)
-	data[3] = 0x00 // Block descriptor length = 0
+	// Build mode page data based on requested page.
+	pages := h.buildModePages(pageCode)
+
+	// MODE SENSE(6) header = 4 bytes + pages
+	data := make([]byte, 4+len(pages))
+	data[0] = byte(3 + len(pages)) // Mode data length (everything after byte 0)
+	data[1] = 0x00                 // Medium type: default
+	data[2] = 0x00                 // Device-specific parameter (no write protect)
+	data[3] = 0x00                 // Block descriptor length = 0
+	copy(data[4:], pages)
 
 	if int(allocLen) < len(data) {
 		data = data[:allocLen]
@@ -372,21 +456,24 @@ func (h *SCSIHandler) modeSense6(cdb [16]byte) SCSIResult {
 }
 
 func (h *SCSIHandler) modeSense10(cdb [16]byte) SCSIResult {
-	// MODE SENSE(10) -- 8-byte header, no mode pages
 	allocLen := binary.BigEndian.Uint16(cdb[7:9])
 	if allocLen == 0 {
 		allocLen = 8
 	}
+	pageCode := cdb[2] & 0x3f
 
-	data := make([]byte, 8)
-	// Mode data length = 6 (8-byte header minus the 2-byte length field)
-	binary.BigEndian.PutUint16(data[0:2], 6)
+	// Build mode page data based on requested page.
+	pages := h.buildModePages(pageCode)
+
+	// MODE SENSE(10) header = 8 bytes + pages
+	data := make([]byte, 8+len(pages))
+	binary.BigEndian.PutUint16(data[0:2], uint16(6+len(pages))) // Mode data length
 	data[2] = 0x00 // Medium type: default
 	data[3] = 0x00 // Device-specific parameter (no write protect)
 	data[4] = 0x00 // Reserved (LONGLBA=0)
 	data[5] = 0x00 // Reserved
-	// Block descriptor length = 0 (bytes 6-7)
-	binary.BigEndian.PutUint16(data[6:8], 0)
+	binary.BigEndian.PutUint16(data[6:8], 0) // Block descriptor length = 0
+	copy(data[8:], pages)
 
 	if int(allocLen) < len(data) {
 		data = data[:allocLen]
@@ -394,6 +481,42 @@ func (h *SCSIHandler) modeSense10(cdb [16]byte) SCSIResult {
 	return SCSIResult{Status: SCSIStatusGood, Data: data}
 }
 
+// buildModePages returns the mode page data for the requested page code.
+// Page 0x3F = return all pages.
+func (h *SCSIHandler) buildModePages(pageCode uint8) []byte {
+	switch pageCode {
+	case 0x08: // Caching mode page
+		return h.modePage08()
+	case 0x0a: // Control mode page
+		return h.modePage0A()
+	case 0x3f: // All pages
+		var pages []byte
+		pages = append(pages, h.modePage08()...)
+		pages = append(pages, h.modePage0A()...)
+		return pages
+	default:
+		return nil
+	}
+}
+
+// modePage08 returns the Caching mode page (SBC-4, Section 7.5.5).
+// Tells Windows we support write caching.
+func (h *SCSIHandler) modePage08() []byte {
+	page := make([]byte, 20)
+	page[0] = 0x08 // Page code
+	page[1] = 18   // Page length (20 - 2)
+	page[2] = 0x04 // WCE=1 (Write Cache Enable), RCD=0 (Read Cache Disable)
+	return page
+}
+
+// modePage0A returns the Control mode page (SPC-5, Section 8.4.8).
+func (h *SCSIHandler) modePage0A() []byte {
+	page := make([]byte, 12)
+	page[0] = 0x0a // Page code
+	page[1] = 10   // Page length (12 - 2)
+	return page
+}
+
 func (h *SCSIHandler) reportLuns(cdb [16]byte) SCSIResult {
 	allocLen := binary.BigEndian.Uint32(cdb[6:10])
 	if allocLen < 16 {
diff --git a/weed/storage/blockvol/iscsi/scsi_test.go b/weed/storage/blockvol/iscsi/scsi_test.go
index e6697bf91..655763f09 100644
--- a/weed/storage/blockvol/iscsi/scsi_test.go
+++ b/weed/storage/blockvol/iscsi/scsi_test.go
@@ -311,7 +311,7 @@ func testReadCapacity16(t *testing.T) {
 	dev := newMockDevice(3 * 1024 * 1024 * 1024 * 1024) // 3 TB
 	h := NewSCSIHandler(dev)
 	var cdb [16]byte
-	cdb[0] = ScsiReadCapacity16
+	cdb[0] = ScsiServiceActionIn16
 	cdb[1] = ScsiSAReadCapacity16
 	binary.BigEndian.PutUint32(cdb[10:14], 32)
 	r := h.HandleCommand(cdb, nil)
@@ -329,7 +329,7 @@ func testReadCapacity16LBPME(t *testing.T) {
 	dev := newMockDevice(100 * 4096)
 	h := NewSCSIHandler(dev)
 	var cdb [16]byte
-	cdb[0] = ScsiReadCapacity16
+	cdb[0] = ScsiServiceActionIn16
 	cdb[1] = ScsiSAReadCapacity16
 	binary.BigEndian.PutUint32(cdb[10:14], 32)
 	r := h.HandleCommand(cdb, nil)
@@ -925,7 +925,7 @@ func testReadCapacity16InvalidSA(t *testing.T) {
 	dev := newMockDevice(100 * 4096)
 	h := NewSCSIHandler(dev)
 	var cdb [16]byte
-	cdb[0] = ScsiReadCapacity16
+	cdb[0] = ScsiServiceActionIn16
 	cdb[1] = 0x05 // wrong service action
 	r := h.HandleCommand(cdb, nil)
 	if r.Status != SCSIStatusCheckCond {
diff --git a/weed/storage/blockvol/iscsi/session.go b/weed/storage/blockvol/iscsi/session.go
index d604700e7..fc12b6927 100644
--- a/weed/storage/blockvol/iscsi/session.go
+++ b/weed/storage/blockvol/iscsi/session.go
@@ -269,6 +269,10 @@ func (s *Session) close() {
 func (s *Session) dispatch(pdu *PDU) error {
 	op := pdu.Opcode()
 
+	if op != OpSCSICmd && op != OpSCSIDataOut {
+		s.logger.Printf("dispatch: opcode=%s(0x%02x)", OpcodeName(op), op)
+	}
+
 	switch op {
 	case OpLoginReq:
 		return s.handleLogin(pdu)
@@ -337,6 +341,8 @@ func (s *Session) handleText(pdu *PDU) error {
 		targets = lister.ListTargets()
 	}
 
+	s.logger.Printf("text request: params=%q targets=%d", string(pdu.DataSegment), len(targets))
+
 	resp := HandleTextRequest(pdu, targets)
 	// ExpCmdSN/MaxCmdSN are set by txLoop via pduNeedsStatSN.
 	s.enqueue(resp)
@@ -356,6 +362,9 @@ func (s *Session) handleSCSICmd(pdu *PDU) error {
 	itt := pdu.InitiatorTaskTag()
 	flags := pdu.OpSpecific1()
 
+	s.logger.Printf("SCSI CDB: opcode=0x%02x cdb=%x itt=0x%08x flags=0x%02x edtl=%d",
+		cdb[0], cdb[:], itt, flags, pdu.ExpectedDataTransferLength())
+
 	// CmdSN validation for non-immediate commands
 	if !pdu.Immediate() {
 		cmdSN := pdu.CmdSN()
@@ -400,7 +409,7 @@ func (s *Session) handleSCSICmd(pdu *PDU) error {
 
 	if isRead && result.Status == SCSIStatusGood && len(result.Data) > 0 {
 		// Build Data-In PDUs and enqueue them all.
-		pdus := s.dataInWriter.BuildDataInPDUs(result.Data, itt, s.expCmdSN.Load(), s.maxCmdSN.Load())
+		pdus := s.dataInWriter.BuildDataInPDUs(result.Data, itt, expectedLen, s.expCmdSN.Load(), s.maxCmdSN.Load())
 		for _, p := range pdus {
 			s.enqueue(p)
 		}
diff --git a/weed/storage/blockvol/qa_phase4a_cp4b4_test.go b/weed/storage/blockvol/qa_phase4a_cp4b4_test.go
new file mode 100644
index 000000000..fdabc193f
--- /dev/null
+++ b/weed/storage/blockvol/qa_phase4a_cp4b4_test.go
@@ -0,0 +1,755 @@
+package blockvol
+
+import (
+	"fmt"
+	"net"
+	"path/filepath"
+	"sync"
+	"testing"
+	"time"
+)
+
+// TestQAPhase4ACP4b4 tests Phase 4A CP4b-4 adversarial scenarios:
+// partial channel failures, protocol edge cases, rebuild scenarios,
+// edge interactions, and error injection.
+func TestQAPhase4ACP4b4(t *testing.T) {
+	tests := []struct {
+		name string
+		run  func(t *testing.T)
+	}{
+		// Group A: Partial Channel Failures
+		{name: "barrier_stale_lsn_after_gap", run: testQA4b4BarrierStaleLSNAfterGap},
+		{name: "ship_data_ok_ctrl_down", run: testQA4b4ShipDataOkCtrlDown},
+		{name: "ship_ctrl_ok_data_down", run: testQA4b4ShipCtrlOkDataDown},
+
+		// Group B: Protocol Edge Cases
+		{name: "ship_large_entry_near_max", run: testQA4b4ShipLargeEntryNearMax},
+		{name: "ship_entries_across_wal_wrap", run: testQA4b4ShipEntriesAcrossWALWrap},
+
+		// Group C: Rebuild Scenarios
+		{name: "rebuild_catchup_concurrent_writes", run: testQA4b4RebuildCatchupConcurrentWrites},
+		{name: "rebuild_full_extent_midcopy_writes", run: testQA4b4RebuildFullExtentMidcopyWrites},
+		{name: "rebuild_interrupted_retry", run: testQA4b4RebuildInterruptedRetry},
+
+		// Group D: Edge Interactions
+		{name: "read_during_wal_reclaim", run: testQA4b4ReadDuringWALReclaim},
+		{name: "trim_blocks_survive_rebuild", run: testQA4b4TrimBlocksSurviveRebuild},
+		{name: "demote_while_wal_full", run: testQA4b4DemoteWhileWALFull},
+
+		// Group E: Error Injection
+		{name: "rebuild_server_epoch_mismatch", run: testQA4b4RebuildServerEpochMismatch},
+	}
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			tt.run(t)
+		})
+	}
+}
+
+// ========== Group A: Partial Channel Failures ==========
+
+// A1: Ship LSN 1,2,4 (skip 3) to replica receiver, then barrier(4).
+// Barrier must time out because receiver blocks waiting for contiguous LSN 3.
+func testQA4b4BarrierStaleLSNAfterGap(t *testing.T) {
+	dir := t.TempDir()
+	cfg := DefaultConfig()
+	cfg.FlushInterval = 5 * time.Millisecond
+
+	vol, err := CreateBlockVol(filepath.Join(dir, "replica.blockvol"), CreateOptions{
+		VolumeSize: 1 * 1024 * 1024, BlockSize: 4096, WALSize: 256 * 1024,
+	}, cfg)
+	if err != nil {
+		t.Fatalf("CreateBlockVol: %v", err)
+	}
+	defer vol.Close()
+	if err := vol.SetEpoch(1); err != nil {
+		t.Fatalf("SetEpoch: %v", err)
+	}
+
+	recv, err := NewReplicaReceiver(vol, "127.0.0.1:0", "127.0.0.1:0")
+	if err != nil {
+		t.Fatalf("NewReplicaReceiver: %v", err)
+	}
+	recv.barrierTimeout = 200 * time.Millisecond // fast timeout for test
+	recv.Serve()
+	defer recv.Stop()
+
+	dataConn, err := net.Dial("tcp", recv.DataAddr())
+	if err != nil {
+		t.Fatalf("dial data: %v", err)
+	}
+	defer dataConn.Close()
+
+	// Send LSN 1, 2, 4 (skip 3).
+	for _, lsn := range []uint64{1, 2, 4} {
+		entry := &WALEntry{LSN: lsn, Epoch: 1, Type: EntryTypeWrite, LBA: lsn - 1, Length: 4096, Data: makeBlock(byte('A' + lsn))}
+		encoded, _ := entry.Encode()
+		if err := WriteFrame(dataConn, MsgWALEntry, encoded); err != nil {
+			t.Fatalf("send LSN=%d: %v", lsn, err)
+		}
+	}
+	time.Sleep(50 * time.Millisecond)
+
+	// receivedLSN should be 2 (gap at 3 blocks further progress).
+	if got := recv.ReceivedLSN(); got != 2 {
+		t.Fatalf("ReceivedLSN = %d after gap, want 2", got)
+	}
+
+	// Barrier(4) must time out since LSN 3 is missing.
+	ctrlConn, err := net.Dial("tcp", recv.CtrlAddr())
+	if err != nil {
+		t.Fatalf("dial ctrl: %v", err)
+	}
+	defer ctrlConn.Close()
+
+	req := EncodeBarrierRequest(BarrierRequest{LSN: 4, Epoch: 1})
+	if err := WriteFrame(ctrlConn, MsgBarrierReq, req); err != nil {
+		t.Fatalf("send barrier: %v", err)
+	}
+
+	ctrlConn.SetReadDeadline(time.Now().Add(3 * time.Second))
+	_, payload, err := ReadFrame(ctrlConn)
+	if err != nil {
+		t.Fatalf("read barrier response: %v", err)
+	}
+	if payload[0] == BarrierOK {
+		t.Error("barrier returned OK despite gap at LSN 3, expected timeout")
+	}
+	// receivedLSN should still be 2.
+	if got := recv.ReceivedLSN(); got != 2 {
+		t.Errorf("ReceivedLSN = %d after barrier timeout, want 2", got)
+	}
+}
+
+// A2: Data channel works, ctrl listener closed before Barrier.
+// Barrier must return error. Data entries still applied.
+func testQA4b4ShipDataOkCtrlDown(t *testing.T) {
+	primary, replica := createReplicaPair(t)
+	defer primary.Close()
+	defer replica.Close()
+
+	// Write data through primary -> shipped to replica via data channel.
+	for i := 0; i < 5; i++ {
+		if err := primary.WriteLBA(uint64(i), makeBlock(byte('A'+i))); err != nil {
+			t.Fatalf("WriteLBA(%d): %v", i, err)
+		}
+	}
+	time.Sleep(50 * time.Millisecond)
+
+	// Verify data entries were applied to replica.
+	replicaLSN := replica.replRecv.ReceivedLSN()
+	if replicaLSN < 5 {
+		t.Fatalf("replica receivedLSN = %d, want >= 5", replicaLSN)
+	}
+
+	// Kill ctrl listener to make barrier fail.
+	replica.replRecv.ctrlListener.Close()
+
+	// Force primary shipper to reconnect ctrl.
+	primary.shipper.ctrlMu.Lock()
+	if primary.shipper.ctrlConn != nil {
+		primary.shipper.ctrlConn.Close()
+		primary.shipper.ctrlConn = nil
+	}
+	primary.shipper.ctrlMu.Unlock()
+
+	// Barrier must fail (ctrl down).
+	err := primary.shipper.Barrier(replicaLSN)
+	if err == nil {
+		t.Error("Barrier should fail when ctrl listener is closed")
+	}
+
+	// Data entries should still be applied (receivedLSN unchanged or higher).
+	if got := replica.replRecv.ReceivedLSN(); got < replicaLSN {
+		t.Errorf("receivedLSN regressed from %d to %d after ctrl failure", replicaLSN, got)
+	}
+}
+
+// A3: Ctrl is up, data listener closed. Ship must fail, barrier may timeout.
+func testQA4b4ShipCtrlOkDataDown(t *testing.T) {
+	dir := t.TempDir()
+	cfg := DefaultConfig()
+	cfg.FlushInterval = 5 * time.Millisecond
+
+	vol, err := CreateBlockVol(filepath.Join(dir, "replica.blockvol"), CreateOptions{
+		VolumeSize: 1 * 1024 * 1024, BlockSize: 4096, WALSize: 256 * 1024,
+	}, cfg)
+	if err != nil {
+		t.Fatalf("CreateBlockVol: %v", err)
+	}
+	defer vol.Close()
+	if err := vol.SetEpoch(1); err != nil {
+		t.Fatalf("SetEpoch: %v", err)
+	}
+
+	recv, err := NewReplicaReceiver(vol, "127.0.0.1:0", "127.0.0.1:0")
+	if err != nil {
+		t.Fatalf("NewReplicaReceiver: %v", err)
+	}
+	recv.barrierTimeout = 200 * time.Millisecond
+	recv.Serve()
+	defer recv.Stop()
+
+	ctrlAddr := recv.CtrlAddr()
+
+	// Close data listener -- data conn will fail.
+	recv.dataListener.Close()
+
+	// Create shipper pointed at closed data port, working ctrl port.
+	shipper := NewWALShipper(recv.DataAddr(), ctrlAddr, func() uint64 { return 1 })
+	defer shipper.Stop()
+
+	// Ship should fail (can't connect data), shipper degrades.
+	entry := &WALEntry{LSN: 1, Epoch: 1, Type: EntryTypeWrite, LBA: 0, Length: 4096, Data: makeBlock('X')}
+	_ = shipper.Ship(entry)
+
+	// receivedLSN should stay 0.
+	if got := recv.ReceivedLSN(); got != 0 {
+		t.Errorf("ReceivedLSN = %d, want 0 (data channel was down)", got)
+	}
+}
+
+// ========== Group B: Protocol Edge Cases ==========
+
+// B1: Ship a large WAL entry through the frame codec and verify it round-trips.
+// Uses WriteFrame/ReadFrame directly to test near-max frame payload.
+// Then ships a multi-block entry through a real replica receiver.
+func testQA4b4ShipLargeEntryNearMax(t *testing.T) {
+	// Part 1: Frame codec round-trip with a large payload (1MB).
+	const largeSize = 1 * 1024 * 1024
+	largePayload := make([]byte, largeSize)
+	for i := range largePayload {
+		largePayload[i] = byte('L')
+	}
+
+	server, client := net.Pipe()
+	defer server.Close()
+	defer client.Close()
+
+	writeDone := make(chan error, 1)
+	go func() {
+		writeDone <- WriteFrame(client, MsgWALEntry, largePayload)
+	}()
+
+	msgType, payload, err := ReadFrame(server)
+	if err != nil {
+		t.Fatalf("ReadFrame large: %v", err)
+	}
+	if msgType != MsgWALEntry {
+		t.Errorf("msgType = 0x%02x, want MsgWALEntry", msgType)
+	}
+	if len(payload) != largeSize {
+		t.Errorf("payload len = %d, want %d", len(payload), largeSize)
+	}
+	if payload[0] != 'L' || payload[largeSize-1] != 'L' {
+		t.Error("payload data corrupted")
+	}
+
+	if err := <-writeDone; err != nil {
+		t.Fatalf("WriteFrame: %v", err)
+	}
+
+	// Part 2: Ship a multi-block entry (8 blocks = 32KB) through a real receiver.
+	dir := t.TempDir()
+	cfg := DefaultConfig()
+	cfg.FlushInterval = 5 * time.Millisecond
+
+	vol, err := CreateBlockVol(filepath.Join(dir, "replica.blockvol"), CreateOptions{
+		VolumeSize: 1 * 1024 * 1024, BlockSize: 4096, WALSize: 512 * 1024,
+	}, cfg)
+	if err != nil {
+		t.Fatalf("CreateBlockVol: %v", err)
+	}
+	defer vol.Close()
+	if err := vol.SetEpoch(1); err != nil {
+		t.Fatalf("SetEpoch: %v", err)
+	}
+
+	recv, err := NewReplicaReceiver(vol, "127.0.0.1:0", "127.0.0.1:0")
+	if err != nil {
+		t.Fatalf("NewReplicaReceiver: %v", err)
+	}
+	recv.Serve()
+	defer recv.Stop()
+
+	dataConn, err := net.Dial("tcp", recv.DataAddr())
+	if err != nil {
+		t.Fatalf("dial data: %v", err)
+	}
+	defer dataConn.Close()
+
+	// 8 blocks = 32KB data.
+	const blockCount = 8
+	bigData := make([]byte, blockCount*4096)
+	for i := range bigData {
+		bigData[i] = byte('B')
+	}
+
+	entry := &WALEntry{LSN: 1, Epoch: 1, Type: EntryTypeWrite, LBA: 0, Length: uint32(len(bigData)), Data: bigData}
+	encoded, err := entry.Encode()
+	if err != nil {
+		t.Fatalf("Encode: %v", err)
+	}
+
+	if err := WriteFrame(dataConn, MsgWALEntry, encoded); err != nil {
+		t.Fatalf("WriteFrame: %v", err)
+	}
+
+	time.Sleep(100 * time.Millisecond)
+
+	if got := recv.ReceivedLSN(); got != 1 {
+		t.Errorf("ReceivedLSN = %d, want 1 (multi-block entry should be applied)", got)
+	}
+}
+
+// B2: Write enough entries to force WAL wrap, ship all, verify replica has data.
+func testQA4b4ShipEntriesAcrossWALWrap(t *testing.T) {
+	primary, replica := createReplicaPair(t)
+	defer primary.Close()
+	defer replica.Close()
+
+	// createReplicaPair uses WALSize=512KB, VolumeSize=1MB.
+	// Each write is 4KB WAL entry + header. Fill enough to wrap.
+	// ~512KB WAL / ~4KB per entry ≈ 128 entries before wrap, but flusher
+	// reclaims space so we can write more.
+	const numWrites = 50
+	for i := 0; i < numWrites; i++ {
+		if err := primary.WriteLBA(uint64(i%256), makeBlock(byte('W'+i%26))); err != nil {
+			// WAL full is acceptable if flusher can't keep up.
+			t.Logf("WriteLBA(%d): %v (expected if WAL fills)", i, err)
+			break
+		}
+	}
+
+	// SyncCache to push barrier.
+	if err := primary.SyncCache(); err != nil {
+		// Degraded replica is ok -- we test that entries shipped correctly.
+		t.Logf("SyncCache: %v (may degrade)", err)
+	}
+
+	time.Sleep(100 * time.Millisecond)
+
+	// Replica should have received entries.
+	replicaLSN := replica.replRecv.ReceivedLSN()
+	if replicaLSN == 0 {
+		t.Error("replica receivedLSN = 0, expected > 0 after writes across WAL")
+	}
+
+	// Verify first few LBAs on replica.
+	for i := 0; i < 5 && uint64(i) <= replicaLSN; i++ {
+		data, err := replica.ReadLBA(uint64(i), 4096)
+		if err != nil {
+			t.Fatalf("replica ReadLBA(%d): %v", i, err)
+		}
+		if data[0] == 0 {
+			t.Errorf("replica LBA %d is zero, expected non-zero data", i)
+		}
+	}
+}
+
+// ========== Group C: Rebuild Scenarios ==========
+
+// C1: Start rebuild while primary receives concurrent writes.
+func testQA4b4RebuildCatchupConcurrentWrites(t *testing.T) {
+	primary := cp3Primary(t, "rb_cc_pri.bv", 1)
+	defer primary.Close()
+
+	// Write initial data.
+	for i := 0; i < 5; i++ {
+		primary.WriteLBA(uint64(i), makeBlock(byte('A'+i)))
+	}
+
+	primary.StartRebuildServer("127.0.0.1:0")
+	defer primary.StopRebuildServer()
+
+	// Start concurrent writes on primary while rebuild happens.
+	var wg sync.WaitGroup
+	stopWriting := make(chan struct{})
+	wg.Add(1)
+	go func() {
+		defer wg.Done()
+		lba := uint64(10)
+		for {
+			select {
+			case <-stopWriting:
+				return
+			default:
+			}
+			primary.WriteLBA(lba, makeBlock(byte('C')))
+			lba++
+			if lba > 14 {
+				lba = 10
+			}
+			time.Sleep(1 * time.Millisecond)
+		}
+	}()
+
+	// Start rebuild client.
+	replica := cp3Rebuilding(t, "rb_cc_rep.bv", 1)
+	defer replica.Close()
+
+	err := StartRebuild(replica, primary.rebuildServer.Addr(), 1, 1)
+	close(stopWriting)
+	wg.Wait()
+
+	if err != nil {
+		t.Fatalf("StartRebuild: %v", err)
+	}
+	if replica.Role() != RoleReplica {
+		t.Errorf("role = %s, want Replica", replica.Role())
+	}
+
+	// Verify initial data on replica.
+	for i := 0; i < 5; i++ {
+		data, err := replica.ReadLBA(uint64(i), 4096)
+		if err != nil {
+			t.Fatalf("ReadLBA(%d): %v", i, err)
+		}
+		if data[0] != byte('A'+i) {
+			t.Errorf("LBA %d: got %c, want %c", i, data[0], byte('A'+i))
+		}
+	}
+}
+
+// C2: Write initial data, start full extent rebuild, concurrent writes during copy.
+func testQA4b4RebuildFullExtentMidcopyWrites(t *testing.T) {
+	primary := cp3Primary(t, "rb_fe_pri.bv", 1)
+	defer primary.Close()
+
+	// Write and flush to make data in extent.
+	for i := 0; i < 5; i++ {
+		primary.WriteLBA(uint64(i), makeBlock(byte('D'+i)))
+	}
+	primary.SyncCache()
+	time.Sleep(30 * time.Millisecond) // let flusher move to extent
+
+	primary.StartRebuildServer("127.0.0.1:0")
+	defer primary.StopRebuildServer()
+
+	// Start concurrent writes during rebuild.
+	var wg sync.WaitGroup
+	stopWriting := make(chan struct{})
+	wg.Add(1)
+	go func() {
+		defer wg.Done()
+		lba := uint64(10)
+		for {
+			select {
+			case <-stopWriting:
+				return
+			default:
+			}
+			primary.WriteLBA(lba, makeBlock(byte('M')))
+			lba++
+			if lba > 14 {
+				lba = 10
+			}
+			time.Sleep(1 * time.Millisecond)
+		}
+	}()
+
+	// Force full extent rebuild (fromLSN=0 below checkpoint).
+	replica := cp3Rebuilding(t, "rb_fe_rep.bv", 1)
+	defer replica.Close()
+
+	err := StartRebuild(replica, primary.rebuildServer.Addr(), 0, 1)
+	close(stopWriting)
+	wg.Wait()
+
+	if err != nil {
+		t.Fatalf("StartRebuild: %v", err)
+	}
+	if replica.Role() != RoleReplica {
+		t.Errorf("role = %s, want Replica", replica.Role())
+	}
+
+	// Verify flushed data on replica (from extent).
+	for i := 0; i < 5; i++ {
+		data, err := replica.ReadLBA(uint64(i), 4096)
+		if err != nil {
+			t.Fatalf("ReadLBA(%d): %v", i, err)
+		}
+		if data[0] != byte('D'+i) {
+			t.Errorf("LBA %d: got %c, want %c", i, data[0], byte('D'+i))
+		}
+	}
+}
+
+// C3: Start rebuild, read half the entries, close connection, retry from start.
+func testQA4b4RebuildInterruptedRetry(t *testing.T) {
+	primary := cp3Primary(t, "rb_int_pri.bv", 1)
+	defer primary.Close()
+
+	for i := 0; i < 10; i++ {
+		primary.WriteLBA(uint64(i), makeBlock(byte('R'+i%26)))
+	}
+
+	primary.StartRebuildServer("127.0.0.1:0")
+	defer primary.StopRebuildServer()
+
+	// First attempt: connect, read a few entries, then close.
+	conn1, err := net.Dial("tcp", primary.rebuildServer.Addr())
+	if err != nil {
+		t.Fatalf("dial 1: %v", err)
+	}
+	req := EncodeRebuildRequest(RebuildRequest{Type: RebuildWALCatchUp, FromLSN: 1, Epoch: 1})
+	WriteFrame(conn1, MsgRebuildReq, req)
+
+	// Read a few entries then abort.
+	for i := 0; i < 3; i++ {
+		_, _, err := ReadFrame(conn1)
+		if err != nil {
+			t.Fatalf("read entry %d: %v", i, err)
+		}
+	}
+	conn1.Close() // simulate crash
+
+	// Second attempt: full rebuild should succeed.
+	replica := cp3Rebuilding(t, "rb_int_rep.bv", 1)
+	defer replica.Close()
+
+	if err := StartRebuild(replica, primary.rebuildServer.Addr(), 1, 1); err != nil {
+		t.Fatalf("StartRebuild (retry): %v", err)
+	}
+	if replica.Role() != RoleReplica {
+		t.Errorf("role = %s, want Replica", replica.Role())
+	}
+
+	// Verify all data.
+	for i := 0; i < 10; i++ {
+		data, err := replica.ReadLBA(uint64(i), 4096)
+		if err != nil {
+			t.Fatalf("ReadLBA(%d): %v", i, err)
+		}
+		expected := byte('R' + i%26)
+		if data[0] != expected {
+			t.Errorf("LBA %d: got %c, want %c", i, data[0], expected)
+		}
+	}
+}
+
+// ========== Group D: Edge Interactions ==========
+
+// D1: Write blocks, flush to reclaim WAL, concurrent ReadLBA on flushed blocks.
+func testQA4b4ReadDuringWALReclaim(t *testing.T) {
+	vol := cp3Primary(t, "read_reclaim.bv", 1)
+	defer vol.Close()
+
+	// Write data.
+	for i := 0; i < 10; i++ {
+		if err := vol.WriteLBA(uint64(i), makeBlock(byte('R'+i%26))); err != nil {
+			t.Fatalf("WriteLBA(%d): %v", i, err)
+		}
+	}
+
+	// SyncCache + wait for flusher to reclaim WAL.
+	vol.SyncCache()
+	time.Sleep(50 * time.Millisecond)
+
+	// Concurrent reads on recently-flushed blocks.
+	var wg sync.WaitGroup
+	errs := make([]error, 10)
+	for i := 0; i < 10; i++ {
+		wg.Add(1)
+		go func(lba int) {
+			defer wg.Done()
+			data, err := vol.ReadLBA(uint64(lba), 4096)
+			if err != nil {
+				errs[lba] = err
+				return
+			}
+			expected := byte('R' + lba%26)
+			if data[0] != expected {
+				errs[lba] = fmt.Errorf("LBA %d: got %c, want %c", lba, data[0], expected)
+			}
+		}(i)
+	}
+	wg.Wait()
+
+	for i, err := range errs {
+		if err != nil {
+			t.Errorf("read LBA %d: %v", i, err)
+		}
+	}
+}
+
+// D2: Write blocks, TRIM some, rebuild replica, verify trimmed blocks read as zeros.
+func testQA4b4TrimBlocksSurviveRebuild(t *testing.T) {
+	primary := cp3Primary(t, "trim_rb_pri.bv", 1)
+	defer primary.Close()
+
+	// Write 10 blocks.
+	for i := 0; i < 10; i++ {
+		primary.WriteLBA(uint64(i), makeBlock(byte('T'+i%26)))
+	}
+
+	// TRIM blocks 3-5.
+	for i := 3; i <= 5; i++ {
+		if err := primary.Trim(uint64(i), 4096); err != nil {
+			t.Fatalf("Trim(%d): %v", i, err)
+		}
+	}
+
+	primary.StartRebuildServer("127.0.0.1:0")
+	defer primary.StopRebuildServer()
+
+	replica := cp3Rebuilding(t, "trim_rb_rep.bv", 1)
+	defer replica.Close()
+
+	if err := StartRebuild(replica, primary.rebuildServer.Addr(), 1, 1); err != nil {
+		t.Fatalf("StartRebuild: %v", err)
+	}
+
+	// Trimmed blocks should read as zeros.
+	for i := 3; i <= 5; i++ {
+		data, err := replica.ReadLBA(uint64(i), 4096)
+		if err != nil {
+			t.Fatalf("ReadLBA(%d): %v", i, err)
+		}
+		nonZero := false
+		for _, b := range data {
+			if b != 0 {
+				nonZero = true
+				break
+			}
+		}
+		if nonZero {
+			t.Errorf("LBA %d should be zero after TRIM+rebuild, got non-zero data", i)
+		}
+	}
+
+	// Non-trimmed blocks should still have data.
+	data0, _ := replica.ReadLBA(0, 4096)
+	if data0[0] != 'T' {
+		t.Errorf("LBA 0: got %c, want 'T'", data0[0])
+	}
+}
+
+// D3: Fill WAL, start WriteLBA (blocks in WAL-full retry), concurrently demote.
+func testQA4b4DemoteWhileWALFull(t *testing.T) {
+	// Create volume with tiny WAL.
+	dir := t.TempDir()
+	cfg := DefaultConfig()
+	cfg.FlushInterval = 100 * time.Millisecond // slow flusher
+	cfg.WALFullTimeout = 2 * time.Second
+
+	vol, err := CreateBlockVol(filepath.Join(dir, "demote_walful.bv"), CreateOptions{
+		VolumeSize: 64 * 1024,
+		BlockSize:  4096,
+		WALSize:    16 * 1024, // tiny WAL: ~4 entries
+	}, cfg)
+	if err != nil {
+		t.Fatalf("CreateBlockVol: %v", err)
+	}
+	defer vol.Close()
+
+	if err := vol.HandleAssignment(1, RolePrimary, 30*time.Second); err != nil {
+		t.Fatalf("assign primary: %v", err)
+	}
+	vol.drainTimeout = 3 * time.Second
+
+	// Fill WAL.
+	writeCount := 0
+	for i := 0; i < 20; i++ {
+		if err := vol.WriteLBA(uint64(i%16), makeBlock(byte('F'))); err != nil {
+			break
+		}
+		writeCount++
+	}
+	if writeCount < 2 {
+		t.Fatalf("only wrote %d entries before WAL full", writeCount)
+	}
+
+	// Start a WriteLBA that will likely block in WAL-full retry.
+	writeDone := make(chan error, 1)
+	go func() {
+		writeDone <- vol.WriteLBA(0, makeBlock('Z'))
+	}()
+
+	// Give the write a moment to start retrying.
+	time.Sleep(10 * time.Millisecond)
+
+	// Concurrently demote.
+	demoteDone := make(chan error, 1)
+	go func() {
+		demoteDone <- vol.HandleAssignment(2, RoleStale, 0)
+	}()
+
+	// Both should complete within timeout (no deadlock).
+	timer := time.NewTimer(5 * time.Second)
+	defer timer.Stop()
+
+	select {
+	case err := <-demoteDone:
+		// Demote may succeed or fail with drain timeout -- both OK.
+		t.Logf("demote result: %v", err)
+	case <-timer.C:
+		t.Fatal("demote hung for 5s -- deadlock between WAL-full retry and demote drain")
+	}
+
+	select {
+	case err := <-writeDone:
+		// Write may fail (demoted, closed, WAL full timeout) -- all acceptable.
+		t.Logf("write result: %v", err)
+	case <-timer.C:
+		t.Fatal("write hung for 5s after demote")
+	}
+}
+
+// ========== Group E: Error Injection ==========
+
+// E1: Start rebuild server on primary (epoch=1), client sends epoch=2.
+// Server should return EPOCH_MISMATCH.
+func testQA4b4RebuildServerEpochMismatch(t *testing.T) {
+	vol := cp3Primary(t, "rbs_epm.bv", 1)
+	defer vol.Close()
+
+	vol.StartRebuildServer("127.0.0.1:0")
+	defer vol.StopRebuildServer()
+
+	// Client sends mismatched epoch=2.
+	conn, err := net.Dial("tcp", vol.rebuildServer.Addr())
+	if err != nil {
+		t.Fatalf("dial: %v", err)
+	}
+	defer conn.Close()
+
+	req := EncodeRebuildRequest(RebuildRequest{Type: RebuildWALCatchUp, FromLSN: 1, Epoch: 2})
+	if err := WriteFrame(conn, MsgRebuildReq, req); err != nil {
+		t.Fatalf("send request: %v", err)
+	}
+
+	conn.SetReadDeadline(time.Now().Add(3 * time.Second))
+	msgType, payload, err := ReadFrame(conn)
+	if err != nil {
+		t.Fatalf("read response: %v", err)
+	}
+	if msgType != MsgRebuildError {
+		t.Errorf("msgType = 0x%02x, want MsgRebuildError(0x%02x)", msgType, MsgRebuildError)
+	}
+	if string(payload) != "EPOCH_MISMATCH" {
+		t.Errorf("payload = %q, want EPOCH_MISMATCH", string(payload))
+	}
+
+	// Also test epoch=0 (lower than server's epoch=1).
+	conn2, err := net.Dial("tcp", vol.rebuildServer.Addr())
+	if err != nil {
+		t.Fatalf("dial 2: %v", err)
+	}
+	defer conn2.Close()
+
+	req2 := EncodeRebuildRequest(RebuildRequest{Type: RebuildFullExtent, Epoch: 0})
+	if err := WriteFrame(conn2, MsgRebuildReq, req2); err != nil {
+		t.Fatalf("send request 2: %v", err)
+	}
+
+	conn2.SetReadDeadline(time.Now().Add(3 * time.Second))
+	msgType2, payload2, err := ReadFrame(conn2)
+	if err != nil {
+		t.Fatalf("read response 2: %v", err)
+	}
+	if msgType2 != MsgRebuildError || string(payload2) != "EPOCH_MISMATCH" {
+		t.Errorf("epoch=0: type=0x%02x payload=%q, want EPOCH_MISMATCH", msgType2, payload2)
+	}
+}
+
diff --git a/weed/storage/blockvol/test/ha_test.go b/weed/storage/blockvol/test/ha_test.go
index 5ce0e0044..4f4d54224 100644
--- a/weed/storage/blockvol/test/ha_test.go
+++ b/weed/storage/blockvol/test/ha_test.go
@@ -24,10 +24,10 @@ const (
 	haISCSIPort2   = 3261 // replica iSCSI (used after promotion)
 	haAdminPort1   = 8080 // primary admin
 	haAdminPort2   = 8081 // replica admin
-	haReplData1    = 9001 // replica receiver data (on replica node)
-	haReplCtrl1    = 9002 // replica receiver ctrl (on replica node)
-	haRebuildPort1 = 9003 // rebuild server (primary)
-	haRebuildPort2 = 9004 // rebuild server (replica, after promotion)
+	haReplData1    = 9011 // replica receiver data (on replica node)
+	haReplCtrl1    = 9012 // replica receiver ctrl (on replica node)
+	haRebuildPort1 = 9013 // rebuild server (primary)
+	haRebuildPort2 = 9014 // rebuild server (replica, after promotion)
 )
 
 // newHAPair creates a primary HATarget on targetNode and a replica HATarget