diff --git a/common/pkgs/ec/lrc/lrc.go b/common/pkgs/ec/lrc/lrc.go
new file mode 100644
index 0000000..6ef6c5a
--- /dev/null
+++ b/common/pkgs/ec/lrc/lrc.go
@@ -0,0 +1,36 @@
+package lrc
+
+import "github.com/klauspost/reedsolomon"
+
+type LRC struct {
+	n      int   // 总块数，包括局部块
+	k      int   // 数据块数量
+	groups []int // 分组校验块生成时使用的数据块
+	l      *reedsolomon.LRC
+}
+
+func New(n int, k int, groups []int) (*LRC, error) {
+	lrc := &LRC{
+		n:      n,
+		k:      k,
+		groups: groups,
+	}
+
+	l, err := reedsolomon.NewLRC(k, n-k, groups)
+	if err != nil {
+		return nil, err
+	}
+
+	lrc.l = l
+	return lrc, nil
+}
+
+// 根据全局修复的原理，生成根据输入修复指定块的矩阵。要求input内元素的值<n-len(r)，且至少包含k个。
+func (l *LRC) GenerateMatrix(inputIdxs []int, outputIdxs []int) ([][]byte, error) {
+	return l.l.GenerateMatrix(inputIdxs, outputIdxs)
+}
+
+// 生成修复组内某个块的矩阵。只支持组内缺少一个块的情况，且默认组内的其他块都存在。
+func (l *LRC) GenerateGroupMatrix(outputIdx int) ([][]byte, error) {
+	return l.l.GenerateGroupMatrix(outputIdx)
+}
diff --git a/common/pkgs/ioswitchlrc/agent_worker.go b/common/pkgs/ioswitchlrc/agent_worker.go
new file mode 100644
index 0000000..67b084f
--- /dev/null
+++ b/common/pkgs/ioswitchlrc/agent_worker.go
@@ -0,0 +1,61 @@
+package ioswitchlrc
+
+import (
+	"context"
+	"io"
+
+	"gitlink.org.cn/cloudream/common/pkgs/ioswitch/exec"
+	cdssdk "gitlink.org.cn/cloudream/common/sdks/storage"
+	stgglb "gitlink.org.cn/cloudream/storage/common/globals"
+	agtrpc "gitlink.org.cn/cloudream/storage/common/pkgs/grpc/agent"
+)
+
+type AgentWorker struct {
+	Node cdssdk.Node
+}
+
+func (w *AgentWorker) NewClient() (exec.WorkerClient, error) {
+	cli, err := stgglb.AgentRPCPool.Acquire(stgglb.SelectGRPCAddress(&w.Node))
+	if err != nil {
+		return nil, err
+	}
+
+	return &AgentWorkerClient{cli: cli}, nil
+}
+
+func (w *AgentWorker) String() string {
+	return w.Node.String()
+}
+
+func (w *AgentWorker) Equals(worker exec.WorkerInfo) bool {
+	aw, ok := worker.(*AgentWorker)
+	if !ok {
+		return false
+	}
+
+	return w.Node.NodeID == aw.Node.NodeID
+}
+
+type AgentWorkerClient struct {
+	cli *agtrpc.PoolClient
+}
+
+func (c *AgentWorkerClient) ExecutePlan(ctx context.Context, plan exec.Plan) error {
+	return c.cli.ExecuteIOPlan(ctx, plan)
+}
+func (c *AgentWorkerClient) SendStream(ctx context.Context, planID exec.PlanID, v *exec.StreamVar, str io.ReadCloser) error {
+	return c.cli.SendStream(ctx, planID, v.ID, str)
+}
+func (c *AgentWorkerClient) SendVar(ctx context.Context, planID exec.PlanID, v exec.Var) error {
+	return c.cli.SendVar(ctx, planID, v)
+}
+func (c *AgentWorkerClient) GetStream(ctx context.Context, planID exec.PlanID, v *exec.StreamVar, signal *exec.SignalVar) (io.ReadCloser, error) {
+	return c.cli.GetStream(ctx, planID, v.ID, signal)
+}
+func (c *AgentWorkerClient) GetVar(ctx context.Context, planID exec.PlanID, v exec.Var, signal *exec.SignalVar) error {
+	return c.cli.GetVar(ctx, planID, v, signal)
+}
+func (c *AgentWorkerClient) Close() error {
+	stgglb.AgentRPCPool.Release(c.cli)
+	return nil
+}
diff --git a/common/pkgs/ioswitchlrc/fromto.go b/common/pkgs/ioswitchlrc/fromto.go
new file mode 100644
index 0000000..ef3ad0e
--- /dev/null
+++ b/common/pkgs/ioswitchlrc/fromto.go
@@ -0,0 +1,134 @@
+package ioswitchlrc
+
+import (
+	"gitlink.org.cn/cloudream/common/pkgs/ioswitch/exec"
+	cdssdk "gitlink.org.cn/cloudream/common/sdks/storage"
+)
+
+type From interface {
+	GetDataIndex() int
+}
+
+type To interface {
+	// To所需要的文件流的范围。具体含义与DataIndex有关系：
+	// 如果DataIndex == -1，则表示在整个文件的范围。
+	// 如果DataIndex >= 0，则表示在文件的某个分片的范围。
+	GetRange() exec.Range
+	GetDataIndex() int
+}
+
+type FromDriver struct {
+	Handle    *exec.DriverWriteStream
+	DataIndex int
+}
+
+func NewFromDriver(dataIndex int) (*FromDriver, *exec.DriverWriteStream) {
+	handle := &exec.DriverWriteStream{
+		RangeHint: &exec.Range{},
+	}
+	return &FromDriver{
+		Handle:    handle,
+		DataIndex: dataIndex,
+	}, handle
+}
+
+func (f *FromDriver) GetDataIndex() int {
+	return f.DataIndex
+}
+
+type FromNode struct {
+	FileHash  string
+	Node      *cdssdk.Node
+	DataIndex int
+}
+
+func NewFromNode(fileHash string, node *cdssdk.Node, dataIndex int) *FromNode {
+	return &FromNode{
+		FileHash:  fileHash,
+		Node:      node,
+		DataIndex: dataIndex,
+	}
+}
+
+func (f *FromNode) GetDataIndex() int {
+	return f.DataIndex
+}
+
+type ToDriver struct {
+	Handle    *exec.DriverReadStream
+	DataIndex int
+	Range     exec.Range
+}
+
+func NewToDriver(dataIndex int) (*ToDriver, *exec.DriverReadStream) {
+	str := exec.DriverReadStream{}
+	return &ToDriver{
+		Handle:    &str,
+		DataIndex: dataIndex,
+	}, &str
+}
+
+func NewToDriverWithRange(dataIndex int, rng exec.Range) (*ToDriver, *exec.DriverReadStream) {
+	str := exec.DriverReadStream{}
+	return &ToDriver{
+		Handle:    &str,
+		DataIndex: dataIndex,
+		Range:     rng,
+	}, &str
+}
+
+func (t *ToDriver) GetDataIndex() int {
+	return t.DataIndex
+}
+
+func (t *ToDriver) GetRange() exec.Range {
+	return t.Range
+}
+
+type ToNode struct {
+	Node             cdssdk.Node
+	DataIndex        int
+	Range            exec.Range
+	FileHashStoreKey string
+}
+
+func NewToNode(node cdssdk.Node, dataIndex int, fileHashStoreKey string) *ToNode {
+	return &ToNode{
+		Node:             node,
+		DataIndex:        dataIndex,
+		FileHashStoreKey: fileHashStoreKey,
+	}
+}
+
+func NewToNodeWithRange(node cdssdk.Node, dataIndex int, fileHashStoreKey string, rng exec.Range) *ToNode {
+	return &ToNode{
+		Node:             node,
+		DataIndex:        dataIndex,
+		FileHashStoreKey: fileHashStoreKey,
+		Range:            rng,
+	}
+}
+
+func (t *ToNode) GetDataIndex() int {
+	return t.DataIndex
+}
+
+func (t *ToNode) GetRange() exec.Range {
+	return t.Range
+}
+
+// type ToStorage struct {
+// 	Storage   cdssdk.Storage
+// 	DataIndex int
+// }
+
+// func NewToStorage(storage cdssdk.Storage, dataIndex int) *ToStorage {
+// 	return &ToStorage{
+// 		Storage:   storage,
+// 		DataIndex: dataIndex,
+// 	}
+// }
+
+// func (t *ToStorage) GetDataIndex() int {
+// 	return t.DataIndex
+// }
diff --git a/common/pkgs/ioswitchlrc/ioswitch.go b/common/pkgs/ioswitchlrc/ioswitch.go
new file mode 100644
index 0000000..b6198a8
--- /dev/null
+++ b/common/pkgs/ioswitchlrc/ioswitch.go
@@ -0,0 +1,23 @@
+package ioswitchlrc
+
+import (
+	"gitlink.org.cn/cloudream/common/pkgs/ioswitch/exec"
+)
+
+type NodeProps struct {
+	From From
+	To   To
+}
+
+type ValueVarType int
+
+const (
+	StringValueVar ValueVarType = iota
+	SignalValueVar
+)
+
+type VarProps struct {
+	StreamIndex int          // 流的编号，只在StreamVar上有意义
+	ValueType   ValueVarType // 值类型，只在ValueVar上有意义
+	Var         exec.Var     // 生成Plan的时候创建的对应的Var
+}
diff --git a/common/pkgs/ioswitchlrc/ops2/chunked.go b/common/pkgs/ioswitchlrc/ops2/chunked.go
new file mode 100644
index 0000000..18090a4
--- /dev/null
+++ b/common/pkgs/ioswitchlrc/ops2/chunked.go
@@ -0,0 +1,137 @@
+package ops2
+
+import (
+	"context"
+	"fmt"
+	"io"
+
+	"github.com/samber/lo"
+	"gitlink.org.cn/cloudream/common/pkgs/future"
+	"gitlink.org.cn/cloudream/common/pkgs/ioswitch/dag"
+	"gitlink.org.cn/cloudream/common/pkgs/ioswitch/exec"
+	"gitlink.org.cn/cloudream/common/utils/io2"
+	"gitlink.org.cn/cloudream/storage/common/pkgs/ioswitch2"
+	"golang.org/x/sync/semaphore"
+)
+
+func init() {
+	exec.UseOp[*ChunkedSplit]()
+	exec.UseOp[*ChunkedJoin]()
+}
+
+type ChunkedSplit struct {
+	Input        *exec.StreamVar   `json:"input"`
+	Outputs      []*exec.StreamVar `json:"outputs"`
+	ChunkSize    int               `json:"chunkSize"`
+	PaddingZeros bool              `json:"paddingZeros"`
+}
+
+func (o *ChunkedSplit) Execute(ctx context.Context, e *exec.Executor) error {
+	err := e.BindVars(ctx, o.Input)
+	if err != nil {
+		return err
+	}
+	defer o.Input.Stream.Close()
+
+	outputs := io2.ChunkedSplit(o.Input.Stream, o.ChunkSize, len(o.Outputs), io2.ChunkedSplitOption{
+		PaddingZeros: o.PaddingZeros,
+	})
+
+	sem := semaphore.NewWeighted(int64(len(outputs)))
+	for i := range outputs {
+		sem.Acquire(ctx, 1)
+
+		o.Outputs[i].Stream = io2.AfterReadClosedOnce(outputs[i], func(closer io.ReadCloser) {
+			sem.Release(1)
+		})
+	}
+	exec.PutArrayVars(e, o.Outputs)
+
+	return sem.Acquire(ctx, int64(len(outputs)))
+}
+
+type ChunkedJoin struct {
+	Inputs    []*exec.StreamVar `json:"inputs"`
+	Output    *exec.StreamVar   `json:"output"`
+	ChunkSize int               `json:"chunkSize"`
+}
+
+func (o *ChunkedJoin) Execute(ctx context.Context, e *exec.Executor) error {
+	err := exec.BindArrayVars(e, ctx, o.Inputs)
+	if err != nil {
+		return err
+	}
+
+	var strReaders []io.Reader
+	for _, s := range o.Inputs {
+		strReaders = append(strReaders, s.Stream)
+	}
+	defer func() {
+		for _, str := range o.Inputs {
+			str.Stream.Close()
+		}
+	}()
+
+	fut := future.NewSetVoid()
+	o.Output.Stream = io2.AfterReadClosedOnce(io2.BufferedChunkedJoin(strReaders, o.ChunkSize), func(closer io.ReadCloser) {
+		fut.SetVoid()
+	})
+	e.PutVars(o.Output)
+
+	return fut.Wait(ctx)
+}
+
+type ChunkedSplitType struct {
+	OutputCount int
+	ChunkSize   int
+}
+
+func (t *ChunkedSplitType) InitNode(node *dag.Node) {
+	dag.NodeDeclareInputStream(node, 1)
+	for i := 0; i < t.OutputCount; i++ {
+		dag.NodeNewOutputStream(node, &ioswitch2.VarProps{
+			StreamIndex: i,
+		})
+	}
+}
+
+func (t *ChunkedSplitType) GenerateOp(op *dag.Node) (exec.Op, error) {
+	return &ChunkedSplit{
+		Input: op.InputStreams[0].Var,
+		Outputs: lo.Map(op.OutputStreams, func(v *dag.StreamVar, idx int) *exec.StreamVar {
+			return v.Var
+		}),
+		ChunkSize:    t.ChunkSize,
+		PaddingZeros: true,
+	}, nil
+}
+
+func (t *ChunkedSplitType) String(node *dag.Node) string {
+	return fmt.Sprintf("ChunkedSplit[%v]%v%v", t.ChunkSize, formatStreamIO(node), formatValueIO(node))
+}
+
+type ChunkedJoinType struct {
+	InputCount int
+	ChunkSize  int
+}
+
+func (t *ChunkedJoinType) InitNode(node *dag.Node) {
+	dag.NodeDeclareInputStream(node, t.InputCount)
+	dag.NodeNewOutputStream(node, &ioswitch2.VarProps{
+		StreamIndex: -1,
+	})
+}
+
+func (t *ChunkedJoinType) GenerateOp(op *dag.Node) (exec.Op, error) {
+	return &ChunkedJoin{
+		Inputs: lo.Map(op.InputStreams, func(v *dag.StreamVar, idx int) *exec.StreamVar {
+			return v.Var
+		}),
+		Output:    op.OutputStreams[0].Var,
+		ChunkSize: t.ChunkSize,
+	}, nil
+}
+
+func (t *ChunkedJoinType) String(node *dag.Node) string {
+	return fmt.Sprintf("ChunkedJoin[%v]%v%v", t.ChunkSize, formatStreamIO(node), formatValueIO(node))
+}
diff --git a/common/pkgs/ioswitchlrc/ops2/clone.go b/common/pkgs/ioswitchlrc/ops2/clone.go
new file mode 100644
index 0000000..d53e29b
--- /dev/null
+++ b/common/pkgs/ioswitchlrc/ops2/clone.go
@@ -0,0 +1,112 @@
+package ops2
+
+import (
+	"context"
+	"fmt"
+	"io"
+
+	"github.com/samber/lo"
+	"gitlink.org.cn/cloudream/common/pkgs/ioswitch/dag"
+	"gitlink.org.cn/cloudream/common/pkgs/ioswitch/exec"
+	"gitlink.org.cn/cloudream/common/utils/io2"
+	"golang.org/x/sync/semaphore"
+)
+
+func init() {
+	exec.UseOp[*CloneStream]()
+	exec.UseOp[*CloneVar]()
+}
+
+type CloneStream struct {
+	Input   *exec.StreamVar   `json:"input"`
+	Outputs []*exec.StreamVar `json:"outputs"`
+}
+
+func (o *CloneStream) Execute(ctx context.Context, e *exec.Executor) error {
+	err := e.BindVars(ctx, o.Input)
+	if err != nil {
+		return err
+	}
+	defer o.Input.Stream.Close()
+
+	cloned := io2.Clone(o.Input.Stream, len(o.Outputs))
+
+	sem := semaphore.NewWeighted(int64(len(o.Outputs)))
+	for i, s := range cloned {
+		sem.Acquire(ctx, 1)
+
+		o.Outputs[i].Stream = io2.AfterReadClosedOnce(s, func(closer io.ReadCloser) {
+			sem.Release(1)
+		})
+	}
+	exec.PutArrayVars(e, o.Outputs)
+
+	return sem.Acquire(ctx, int64(len(o.Outputs)))
+}
+
+type CloneVar struct {
+	Raw     exec.Var   `json:"raw"`
+	Cloneds []exec.Var `json:"cloneds"`
+}
+
+func (o *CloneVar) Execute(ctx context.Context, e *exec.Executor) error {
+	err := e.BindVars(ctx, o.Raw)
+	if err != nil {
+		return err
+	}
+
+	for _, v := range o.Cloneds {
+		if err := exec.AssignVar(o.Raw, v); err != nil {
+			return fmt.Errorf("clone var: %w", err)
+		}
+	}
+	e.PutVars(o.Cloneds...)
+
+	return nil
+}
+
+type CloneStreamType struct{}
+
+func (t *CloneStreamType) InitNode(node *dag.Node) {
+	dag.NodeDeclareInputStream(node, 1)
+}
+
+func (t *CloneStreamType) GenerateOp(op *dag.Node) (exec.Op, error) {
+	return &CloneStream{
+		Input: op.InputStreams[0].Var,
+		Outputs: lo.Map(op.OutputStreams, func(v *dag.StreamVar, idx int) *exec.StreamVar {
+			return v.Var
+		}),
+	}, nil
+}
+
+func (t *CloneStreamType) NewOutput(node *dag.Node) *dag.StreamVar {
+	return dag.NodeNewOutputStream(node, nil)
+}
+
+func (t *CloneStreamType) String(node *dag.Node) string {
+	return fmt.Sprintf("CloneStream[]%v%v", formatStreamIO(node), formatValueIO(node))
+}
+
+type CloneVarType struct{}
+
+func (t *CloneVarType) InitNode(node *dag.Node) {
+	dag.NodeDeclareInputValue(node, 1)
+}
+
+func (t *CloneVarType) GenerateOp(op *dag.Node) (exec.Op, error) {
+	return &CloneVar{
+		Raw: op.InputValues[0].Var,
+		Cloneds: lo.Map(op.OutputValues, func(v *dag.ValueVar, idx int) exec.Var {
+			return v.Var
+		}),
+	}, nil
+}
+
+func (t *CloneVarType) NewOutput(node *dag.Node) *dag.ValueVar {
+	return dag.NodeNewOutputValue(node, nil)
+}
+
+func (t *CloneVarType) String(node *dag.Node) string {
+	return fmt.Sprintf("CloneVar[]%v%v", formatStreamIO(node), formatValueIO(node))
+}
diff --git a/common/pkgs/ioswitchlrc/ops2/ec.go b/common/pkgs/ioswitchlrc/ops2/ec.go
new file mode 100644
index 0000000..7d644d4
--- /dev/null
+++ b/common/pkgs/ioswitchlrc/ops2/ec.go
@@ -0,0 +1,189 @@
+package ops2
+
+import (
+	"context"
+	"fmt"
+	"io"
+
+	"github.com/samber/lo"
+	"gitlink.org.cn/cloudream/common/pkgs/future"
+	"gitlink.org.cn/cloudream/common/pkgs/ioswitch/dag"
+	"gitlink.org.cn/cloudream/common/pkgs/ioswitch/exec"
+	cdssdk "gitlink.org.cn/cloudream/common/sdks/storage"
+	"gitlink.org.cn/cloudream/common/utils/io2"
+	"gitlink.org.cn/cloudream/common/utils/sync2"
+	"gitlink.org.cn/cloudream/storage/common/pkgs/ec"
+	"gitlink.org.cn/cloudream/storage/common/pkgs/ec/lrc"
+	"gitlink.org.cn/cloudream/storage/common/pkgs/ioswitch2"
+	"gitlink.org.cn/cloudream/storage/common/pkgs/ioswitchlrc"
+)
+
+func init() {
+	exec.UseOp[*GalMultiply]()
+}
+
+type GalMultiply struct {
+	Coef      [][]byte          `json:"coef"`
+	Inputs    []*exec.StreamVar `json:"inputs"`
+	Outputs   []*exec.StreamVar `json:"outputs"`
+	ChunkSize int               `json:"chunkSize"`
+}
+
+func (o *GalMultiply) Execute(ctx context.Context, e *exec.Executor) error {
+	err := exec.BindArrayVars(e, ctx, o.Inputs)
+	if err != nil {
+		return err
+	}
+	defer func() {
+		for _, s := range o.Inputs {
+			s.Stream.Close()
+		}
+	}()
+
+	outputWrs := make([]*io.PipeWriter, len(o.Outputs))
+
+	for i := range o.Outputs {
+		rd, wr := io.Pipe()
+		o.Outputs[i].Stream = rd
+		outputWrs[i] = wr
+	}
+
+	fut := future.NewSetVoid()
+	go func() {
+		mul := ec.GaloisMultiplier().BuildGalois()
+
+		inputChunks := make([][]byte, len(o.Inputs))
+		for i := range o.Inputs {
+			inputChunks[i] = make([]byte, o.ChunkSize)
+		}
+		outputChunks := make([][]byte, len(o.Outputs))
+		for i := range o.Outputs {
+			outputChunks[i] = make([]byte, o.ChunkSize)
+		}
+
+		for {
+			err := sync2.ParallelDo(o.Inputs, func(s *exec.StreamVar, i int) error {
+				_, err := io.ReadFull(s.Stream, inputChunks[i])
+				return err
+			})
+			if err == io.EOF {
+				fut.SetVoid()
+				return
+			}
+			if err != nil {
+				fut.SetError(err)
+				return
+			}
+
+			err = mul.Multiply(o.Coef, inputChunks, outputChunks)
+			if err != nil {
+				fut.SetError(err)
+				return
+			}
+
+			for i := range o.Outputs {
+				err := io2.WriteAll(outputWrs[i], outputChunks[i])
+				if err != nil {
+					fut.SetError(err)
+					return
+				}
+			}
+		}
+	}()
+
+	exec.PutArrayVars(e, o.Outputs)
+	err = fut.Wait(ctx)
+	if err != nil {
+		for _, wr := range outputWrs {
+			wr.CloseWithError(err)
+		}
+		return err
+	}
+
+	for _, wr := range outputWrs {
+		wr.Close()
+	}
+	return nil
+}
+
+type LRCConstructAnyType struct {
+	LRC cdssdk.LRCRedundancy
+}
+
+func (t *LRCConstructAnyType) InitNode(node *dag.Node) {}
+
+func (t *LRCConstructAnyType) GenerateOp(op *dag.Node) (exec.Op, error) {
+	var inputIdxs []int
+	var outputIdxs []int
+	for _, in := range op.InputStreams {
+		inputIdxs = append(inputIdxs, ioswitch2.SProps(in).StreamIndex)
+	}
+	for _, out := range op.OutputStreams {
+		outputIdxs = append(outputIdxs, ioswitch2.SProps(out).StreamIndex)
+	}
+
+	l, err := lrc.New(t.LRC.N, t.LRC.K, t.LRC.Groups)
+	if err != nil {
+		return nil, err
+	}
+	coef, err := l.GenerateMatrix(inputIdxs, outputIdxs)
+	if err != nil {
+		return nil, err
+	}
+
+	return &GalMultiply{
+		Coef:      coef,
+		Inputs:    lo.Map(op.InputStreams, func(v *dag.StreamVar, idx int) *exec.StreamVar { return v.Var }),
+		Outputs:   lo.Map(op.OutputStreams, func(v *dag.StreamVar, idx int) *exec.StreamVar { return v.Var }),
+		ChunkSize: t.LRC.ChunkSize,
+	}, nil
+}
+
+func (t *LRCConstructAnyType) AddInput(node *dag.Node, str *dag.StreamVar) {
+	node.InputStreams = append(node.InputStreams, str)
+	str.To(node, len(node.InputStreams)-1)
+}
+
+func (t *LRCConstructAnyType) NewOutput(node *dag.Node, dataIndex int) *dag.StreamVar {
+	return dag.NodeNewOutputStream(node, &ioswitch2.VarProps{StreamIndex: dataIndex})
+}
+
+func (t *LRCConstructAnyType) String(node *dag.Node) string {
+	return fmt.Sprintf("LRCAny[]%v%v", formatStreamIO(node), formatValueIO(node))
+}
+
+type LRCConstructGroupType struct {
+	LRC              cdssdk.LRCRedundancy
+	TargetBlockIndex int
+}
+
+func (t *LRCConstructGroupType) InitNode(node *dag.Node) {
+	dag.NodeNewOutputStream(node, &ioswitchlrc.VarProps{
+		StreamIndex: t.TargetBlockIndex,
+	})
+
+	grpIdx := t.LRC.FindGroup(t.TargetBlockIndex)
+	dag.NodeDeclareInputStream(node, t.LRC.Groups[grpIdx])
+}
+
+func (t *LRCConstructGroupType) GenerateOp(op *dag.Node) (exec.Op, error) {
+	l, err := lrc.New(t.LRC.N, t.LRC.K, t.LRC.Groups)
+	if err != nil {
+		return nil, err
+	}
+	coef, err := l.GenerateGroupMatrix(t.TargetBlockIndex)
+	if err != nil {
+		return nil, err
+	}
+
+	return &GalMultiply{
+		Coef:      coef,
+		Inputs:    lo.Map(op.InputStreams, func(v *dag.StreamVar, idx int) *exec.StreamVar { return v.Var }),
+		Outputs:   lo.Map(op.OutputStreams, func(v *dag.StreamVar, idx int) *exec.StreamVar { return v.Var }),
+		ChunkSize: t.LRC.ChunkSize,
+	}, nil
+}
+
+func (t *LRCConstructGroupType) String(node *dag.Node) string {
+	return fmt.Sprintf("LRCGroup[]%v%v", formatStreamIO(node), formatValueIO(node))
+}
diff --git a/common/pkgs/ioswitchlrc/ops2/ipfs.go b/common/pkgs/ioswitchlrc/ops2/ipfs.go
new file mode 100644
index 0000000..fa02b6d
--- /dev/null
+++ b/common/pkgs/ioswitchlrc/ops2/ipfs.go
@@ -0,0 +1,129 @@
+package ops2
+
+import (
+	"context"
+	"fmt"
+	"io"
+
+	"gitlink.org.cn/cloudream/common/pkgs/future"
+	"gitlink.org.cn/cloudream/common/pkgs/ioswitch/dag"
+	"gitlink.org.cn/cloudream/common/pkgs/ioswitch/exec"
+	"gitlink.org.cn/cloudream/common/pkgs/ipfs"
+	"gitlink.org.cn/cloudream/common/pkgs/logger"
+	"gitlink.org.cn/cloudream/common/utils/io2"
+	stgglb "gitlink.org.cn/cloudream/storage/common/globals"
+	"gitlink.org.cn/cloudream/storage/common/pkgs/ioswitch2"
+)
+
+func init() {
+	exec.UseOp[*IPFSRead]()
+	exec.UseOp[*IPFSWrite]()
+}
+
+type IPFSRead struct {
+	Output   *exec.StreamVar `json:"output"`
+	FileHash string          `json:"fileHash"`
+	Option   ipfs.ReadOption `json:"option"`
+}
+
+func (o *IPFSRead) Execute(ctx context.Context, e *exec.Executor) error {
+	logger.
+		WithField("FileHash", o.FileHash).
+		Debugf("ipfs read op")
+	defer logger.Debugf("ipfs read op finished")
+
+	ipfsCli, err := stgglb.IPFSPool.Acquire()
+	if err != nil {
+		return fmt.Errorf("new ipfs client: %w", err)
+	}
+	defer stgglb.IPFSPool.Release(ipfsCli)
+
+	file, err := ipfsCli.OpenRead(o.FileHash, o.Option)
+	if err != nil {
+		return fmt.Errorf("reading ipfs: %w", err)
+	}
+	defer file.Close()
+
+	fut := future.NewSetVoid()
+	o.Output.Stream = io2.AfterReadClosedOnce(file, func(closer io.ReadCloser) {
+		fut.SetVoid()
+	})
+	e.PutVars(o.Output)
+
+	return fut.Wait(ctx)
+}
+
+type IPFSWrite struct {
+	Input    *exec.StreamVar `json:"input"`
+	FileHash *exec.StringVar `json:"fileHash"`
+}
+
+func (o *IPFSWrite) Execute(ctx context.Context, e *exec.Executor) error {
+	logger.
+		WithField("Input", o.Input.ID).
+		WithField("FileHashVar", o.FileHash.ID).
+		Debugf("ipfs write op")
+
+	ipfsCli, err := stgglb.IPFSPool.Acquire()
+	if err != nil {
+		return fmt.Errorf("new ipfs client: %w", err)
+	}
+	defer stgglb.IPFSPool.Release(ipfsCli)
+
+	err = e.BindVars(ctx, o.Input)
+	if err != nil {
+		return err
+	}
+	defer o.Input.Stream.Close()
+
+	o.FileHash.Value, err = ipfsCli.CreateFile(o.Input.Stream)
+	if err != nil {
+		return fmt.Errorf("creating ipfs file: %w", err)
+	}
+
+	e.PutVars(o.FileHash)
+
+	return nil
+}
+
+type IPFSReadType struct {
+	FileHash string
+	Option   ipfs.ReadOption
+}
+
+func (t *IPFSReadType) InitNode(node *dag.Node) {
+	dag.NodeNewOutputStream(node, &ioswitch2.VarProps{})
+}
+
+func (t *IPFSReadType) GenerateOp(n *dag.Node) (exec.Op, error) {
+	return &IPFSRead{
+		Output:   n.OutputStreams[0].Var,
+		FileHash: t.FileHash,
+		Option:   t.Option,
+	}, nil
+}
+
+func (t *IPFSReadType) String(node *dag.Node) string {
+	return fmt.Sprintf("IPFSRead[%s,%v+%v]%v%v", t.FileHash, t.Option.Offset, t.Option.Length, formatStreamIO(node), formatValueIO(node))
+}
+
+type IPFSWriteType struct {
+	FileHashStoreKey string
+	Range            exec.Range
+}
+
+func (t *IPFSWriteType) InitNode(node *dag.Node) {
+	dag.NodeDeclareInputStream(node, 1)
+	dag.NodeNewOutputValue(node, &ioswitch2.VarProps{})
+}
+
+func (t *IPFSWriteType) GenerateOp(op *dag.Node) (exec.Op, error) {
+	return &IPFSWrite{
+		Input:    op.InputStreams[0].Var,
+		FileHash: op.OutputValues[0].Var.(*exec.StringVar),
+	}, nil
+}
+
+func (t *IPFSWriteType) String(node *dag.Node) string {
+	return fmt.Sprintf("IPFSWrite[%s,%v+%v]%v%v", t.FileHashStoreKey, t.Range.Offset, t.Range.Length, formatStreamIO(node), formatValueIO(node))
+}
diff --git a/common/pkgs/ioswitchlrc/ops2/ops.go b/common/pkgs/ioswitchlrc/ops2/ops.go
new file mode 100644
index 0000000..c7309b6
--- /dev/null
+++ b/common/pkgs/ioswitchlrc/ops2/ops.go
@@ -0,0 +1,75 @@
+package ops2
+
+import (
+	"fmt"
+
+	"gitlink.org.cn/cloudream/common/pkgs/ioswitch/dag"
+)
+
+func formatStreamIO(node *dag.Node) string {
+	is := ""
+	for i, in := range node.InputStreams {
+		if i > 0 {
+			is += ","
+		}
+
+		if in == nil {
+			is += "."
+		} else {
+			is += fmt.Sprintf("%v", in.ID)
+		}
+	}
+
+	os := ""
+	for i, out := range node.OutputStreams {
+		if i > 0 {
+			os += ","
+		}
+
+		if out == nil {
+			os += "."
+		} else {
+			os += fmt.Sprintf("%v", out.ID)
+		}
+	}
+
+	if is == "" && os == "" {
+		return ""
+	}
+
+	return fmt.Sprintf("S{%s>%s}", is, os)
+}
+
+func formatValueIO(node *dag.Node) string {
+	is := ""
+	for i, in := range node.InputValues {
+		if i > 0 {
+			is += ","
+		}
+
+		if in == nil {
+			is += "."
+		} else {
+			is += fmt.Sprintf("%v", in.ID)
+		}
+	}
+
+	os := ""
+	for i, out := range node.OutputValues {
+		if i > 0 {
+			os += ","
+		}
+
+		if out == nil {
+			os += "."
+		} else {
+			os += fmt.Sprintf("%v", out.ID)
+		}
+	}
+
+	if is == "" && os == "" {
+		return ""
+	}
+
+	return fmt.Sprintf("V{%s>%s}", is, os)
+}
diff --git a/common/pkgs/ioswitchlrc/ops2/range.go b/common/pkgs/ioswitchlrc/ops2/range.go
new file mode 100644
index 0000000..383b3c0
--- /dev/null
+++ b/common/pkgs/ioswitchlrc/ops2/range.go
@@ -0,0 +1,95 @@
+package ops2
+
+import (
+	"context"
+	"fmt"
+	"io"
+
+	"gitlink.org.cn/cloudream/common/pkgs/future"
+	"gitlink.org.cn/cloudream/common/pkgs/ioswitch/dag"
+	"gitlink.org.cn/cloudream/common/pkgs/ioswitch/exec"
+	"gitlink.org.cn/cloudream/common/utils/io2"
+	"gitlink.org.cn/cloudream/common/utils/math2"
+	"gitlink.org.cn/cloudream/storage/common/pkgs/ioswitch2"
+)
+
+func init() {
+	exec.UseOp[*Range]()
+}
+
+type Range struct {
+	Input  *exec.StreamVar `json:"input"`
+	Output *exec.StreamVar `json:"output"`
+	Offset int64           `json:"offset"`
+	Length *int64          `json:"length"`
+}
+
+func (o *Range) Execute(ctx context.Context, e *exec.Executor) error {
+	err := e.BindVars(ctx, o.Input)
+	if err != nil {
+		return err
+	}
+	defer o.Input.Stream.Close()
+
+	buf := make([]byte, 1024*16)
+
+	// 跳过前Offset个字节
+	for o.Offset > 0 {
+		rdCnt := math2.Min(o.Offset, int64(len(buf)))
+		rd, err := o.Input.Stream.Read(buf[:rdCnt])
+		if err == io.EOF {
+			// 输入流不够长度也不报错，只是产生一个空的流
+			break
+		}
+		if err != nil {
+			return err
+		}
+		o.Offset -= int64(rd)
+	}
+
+	fut := future.NewSetVoid()
+
+	if o.Length == nil {
+		o.Output.Stream = io2.AfterEOF(o.Input.Stream, func(closer io.ReadCloser, err error) {
+			fut.SetVoid()
+		})
+
+		e.PutVars(o.Output)
+		return fut.Wait(ctx)
+	}
+
+	o.Output.Stream = io2.AfterEOF(io2.Length(o.Input.Stream, *o.Length), func(closer io.ReadCloser, err error) {
+		fut.SetVoid()
+	})
+
+	e.PutVars(o.Output)
+	err = fut.Wait(ctx)
+	if err != nil {
+		return err
+	}
+
+	io2.DropWithBuf(o.Input.Stream, buf)
+	return nil
+}
+
+type RangeType struct {
+	Range exec.Range
+}
+
+func (t *RangeType) InitNode(node *dag.Node) {
+	dag.NodeDeclareInputStream(node, 1)
+	dag.NodeNewOutputStream(node, &ioswitch2.VarProps{})
+}
+
+func (t *RangeType) GenerateOp(n *dag.Node) (exec.Op, error) {
+	return &Range{
+		Input:  n.InputStreams[0].Var,
+		Output: n.OutputStreams[0].Var,
+		Offset: t.Range.Offset,
+		Length: t.Range.Length,
+	}, nil
+}
+
+func (t *RangeType) String(node *dag.Node) string {
+	return fmt.Sprintf("Range[%v+%v]%v%v", t.Range.Offset, t.Range.Length, formatStreamIO(node), formatValueIO(node))
+}
diff --git a/common/pkgs/ioswitchlrc/parser/generator.go b/common/pkgs/ioswitchlrc/parser/generator.go
new file mode 100644
index 0000000..97b4b8c
--- /dev/null
+++ b/common/pkgs/ioswitchlrc/parser/generator.go
@@ -0,0 +1,297 @@
+package parser
+
+import (
+	"fmt"
+
+	"gitlink.org.cn/cloudream/common/pkgs/ioswitch/dag"
+	"gitlink.org.cn/cloudream/common/pkgs/ioswitch/exec"
+	"gitlink.org.cn/cloudream/common/pkgs/ioswitch/plan"
+	cdssdk "gitlink.org.cn/cloudream/common/sdks/storage"
+	"gitlink.org.cn/cloudream/storage/common/pkgs/ioswitchlrc"
+	"gitlink.org.cn/cloudream/storage/common/pkgs/ioswitchlrc/ops2"
+)
+
+type GenerateContext struct {
+	LRC         cdssdk.LRCRedundancy
+	DAG         *dag.Graph
+	Toes        []ioswitchlrc.To
+	StreamRange exec.Range
+}
+
+// 输入一个完整文件，从这个完整文件产生任意文件块（也可再产生完整文件）。
+func Encode(fr ioswitchlrc.From, toes []ioswitchlrc.To, blder *exec.PlanBuilder) error {
+	if fr.GetDataIndex() != -1 {
+		return fmt.Errorf("from data is not a complete file")
+	}
+
+	ctx := GenerateContext{
+		LRC:  cdssdk.DefaultLRCRedundancy,
+		DAG:  dag.NewGraph(),
+		Toes: toes,
+	}
+
+	calcStreamRange(&ctx)
+	err := buildDAGEncode(&ctx, fr, toes)
+	if err != nil {
+		return err
+	}
+
+	// 确定指令执行位置的过程，也需要反复进行，直到没有变化为止。
+	for pin(&ctx) {
+	}
+
+	// 下面这些只需要执行一次，但需要按顺序
+	dropUnused(&ctx)
+	storeIPFSWriteResult(&ctx)
+	generateClone(&ctx)
+	generateRange(&ctx)
+
+	return plan.Generate(ctx.DAG, blder)
+}
+
+func buildDAGEncode(ctx *GenerateContext, fr ioswitchlrc.From, toes []ioswitchlrc.To) error {
+	frNode, err := buildFromNode(ctx, fr)
+	if err != nil {
+		return fmt.Errorf("building from node: %w", err)
+	}
+
+	var dataToes []ioswitchlrc.To
+	var parityToes []ioswitchlrc.To
+
+	// 先创建需要完整文件的To节点，同时统计一下需要哪些文件块
+	for _, to := range toes {
+		if to.GetDataIndex() != -1 {
+			continue
+		}
+
+		toNode, err := buildToNode(ctx, to)
+		if err != nil {
+			return fmt.Errorf("building to node: %w", err)
+		}
+
+		idx := to.GetDataIndex()
+		if idx == -1 {
+			frNode.OutputStreams[0].To(toNode, 0)
+		} else if idx < ctx.LRC.K {
+			dataToes = append(dataToes, to)
+		} else {
+			parityToes = append(parityToes, to)
+		}
+	}
+
+	if len(dataToes) == 0 && len(parityToes) == 0 {
+		return nil
+	}
+
+	// 需要文件块，则生成Split指令
+	splitNode := ctx.DAG.NewNode(&ops2.ChunkedSplitType{
+		OutputCount: ctx.LRC.K,
+		ChunkSize:   ctx.LRC.ChunkSize,
+	}, nil)
+
+	for _, to := range dataToes {
+		toNode, err := buildToNode(ctx, to)
+		if err != nil {
+			return fmt.Errorf("building to node: %w", err)
+		}
+
+		splitNode.OutputStreams[to.GetDataIndex()].To(toNode, 0)
+	}
+
+	if len(parityToes) == 0 {
+		return nil
+	}
+
+	// 需要校验块，则进一步生成Construct指令
+
+	conNode, conType := dag.NewNode(ctx.DAG, &ops2.LRCConstructAnyType{
+		LRC: ctx.LRC,
+	}, nil)
+
+	for _, out := range splitNode.OutputStreams {
+		conType.AddInput(conNode, out)
+	}
+
+	for _, to := range parityToes {
+		toNode, err := buildToNode(ctx, to)
+		if err != nil {
+			return fmt.Errorf("building to node: %w", err)
+		}
+
+		conType.NewOutput(conNode, to.GetDataIndex()).To(toNode, 0)
+	}
+	return nil
+}
+
+// 提供数据块+编码块中的k个块，重建任意块，包括完整文件。
+func ReconstructAny(frs []ioswitchlrc.From, toes []ioswitchlrc.To, blder *exec.PlanBuilder) error {
+	ctx := GenerateContext{
+		LRC:  cdssdk.DefaultLRCRedundancy,
+		DAG:  dag.NewGraph(),
+		Toes: toes,
+	}
+
+	calcStreamRange(&ctx)
+	err := buildDAGReconstructAny(&ctx, frs, toes)
+	if err != nil {
+		return err
+	}
+
+	// 确定指令执行位置的过程，也需要反复进行，直到没有变化为止。
+	for pin(&ctx) {
+	}
+
+	// 下面这些只需要执行一次，但需要按顺序
+	dropUnused(&ctx)
+	storeIPFSWriteResult(&ctx)
+	generateClone(&ctx)
+	generateRange(&ctx)
+
+	return plan.Generate(ctx.DAG, blder)
+}
+
+func buildDAGReconstructAny(ctx *GenerateContext, frs []ioswitchlrc.From, toes []ioswitchlrc.To) error {
+	frNodes := make(map[int]*dag.Node)
+	for _, fr := range frs {
+		frNode, err := buildFromNode(ctx, fr)
+		if err != nil {
+			return fmt.Errorf("building from node: %w", err)
+		}
+
+		frNodes[fr.GetDataIndex()] = frNode
+	}
+
+	var completeToes []ioswitchlrc.To
+	var missedToes []ioswitchlrc.To
+
+	// 先创建需要完整文件的To节点，同时统计一下需要哪些文件块
+	for _, to := range toes {
+		toIdx := to.GetDataIndex()
+		fr := frNodes[toIdx]
+		if fr != nil {
+			node, err := buildToNode(ctx, to)
+			if err != nil {
+				return fmt.Errorf("building to node: %w", err)
+			}
+
+			fr.OutputStreams[0].To(node, 0)
+			continue
+		}
+
+		if toIdx == -1 {
+			completeToes = append(completeToes, to)
+		} else {
+			missedToes = append(missedToes, to)
+		}
+	}
+
+	if len(completeToes) == 0 && len(missedToes) == 0 {
+		return nil
+	}
+
+	// 生成Construct指令来恢复缺少的块
+
+	conNode, conType := dag.NewNode(ctx.DAG, &ops2.LRCConstructAnyType{
+		LRC: ctx.LRC,
+	}, nil)
+
+	for _, fr := range frNodes {
+		conType.AddInput(conNode, fr.OutputStreams[0])
+	}
+
+	for _, to := range missedToes {
+		toNode, err := buildToNode(ctx, to)
+		if err != nil {
+			return fmt.Errorf("building to node: %w", err)
+		}
+
+		conType.NewOutput(conNode, to.GetDataIndex()).To(toNode, 0)
+	}
+
+	if len(completeToes) == 0 {
+		return nil
+	}
+
+	// 需要完整文件，则生成Join指令
+
+	joinNode := ctx.DAG.NewNode(&ops2.ChunkedJoinType{
+		InputCount: ctx.LRC.K,
+		ChunkSize:  ctx.LRC.ChunkSize,
+	}, nil)
+
+	for i := 0; i < ctx.LRC.K; i++ {
+		n := frNodes[i]
+		if n == nil {
+			conType.NewOutput(conNode, i).To(joinNode, i)
+		} else {
+			n.OutputStreams[0].To(joinNode, i)
+		}
+	}
+
+	for _, to := range completeToes {
+		toNode, err := buildToNode(ctx, to)
+		if err != nil {
+			return fmt.Errorf("building to node: %w", err)
+		}
+
+		joinNode.OutputStreams[0].To(toNode, 0)
+	}
+
+	return nil
+}
+
+// 输入同一组的多个块，恢复出剩下缺少的一个块。
+func ReconstructGroup(frs []ioswitchlrc.From, toes []ioswitchlrc.To, blder *exec.PlanBuilder) error {
+	ctx := GenerateContext{
+		LRC:  cdssdk.DefaultLRCRedundancy,
+		DAG:  dag.NewGraph(),
+		Toes: toes,
+	}
+
+	calcStreamRange(&ctx)
+	err := buildDAGReconstructGroup(&ctx, frs, toes)
+	if err != nil {
+		return err
+	}
+
+	// 确定指令执行位置的过程，也需要反复进行，直到没有变化为止。
+	for pin(&ctx) {
+	}
+
+	// 下面这些只需要执行一次，但需要按顺序
+	dropUnused(&ctx)
+	storeIPFSWriteResult(&ctx)
+	generateClone(&ctx)
+	generateRange(&ctx)
+
+	return plan.Generate(ctx.DAG, blder)
+}
+
+func buildDAGReconstructGroup(ctx *GenerateContext, frs []ioswitchlrc.From, toes []ioswitchlrc.To) error {
+	missedGrpIdx := toes[0].GetDataIndex()
+
+	conNode := ctx.DAG.NewNode(&ops2.LRCConstructGroupType{
+		LRC:              ctx.LRC,
+		TargetBlockIndex: missedGrpIdx,
+	}, nil)
+
+	for i, fr := range frs {
+		frNode, err := buildFromNode(ctx, fr)
+		if err != nil {
+			return fmt.Errorf("building from node: %w", err)
+		}
+
+		frNode.OutputStreams[0].To(conNode, i)
+	}
+
+	for _, to := range toes {
+		toNode, err := buildToNode(ctx, to)
+		if err != nil {
+			return fmt.Errorf("building to node: %w", err)
+		}
+
+		conNode.OutputStreams[0].To(toNode, 0)
+	}
+
+	return nil
+}
diff --git a/common/pkgs/ioswitchlrc/parser/passes.go b/common/pkgs/ioswitchlrc/parser/passes.go
new file mode 100644
index 0000000..ed4b1c9
--- /dev/null
+++ b/common/pkgs/ioswitchlrc/parser/passes.go
@@ -0,0 +1,309 @@
+package parser
+
+import (
+	"fmt"
+	"math"
+
+	"gitlink.org.cn/cloudream/common/pkgs/ioswitch/dag"
+	"gitlink.org.cn/cloudream/common/pkgs/ioswitch/exec"
+	"gitlink.org.cn/cloudream/common/pkgs/ioswitch/plan/ops"
+	"gitlink.org.cn/cloudream/common/pkgs/ipfs"
+	"gitlink.org.cn/cloudream/common/utils/math2"
+	"gitlink.org.cn/cloudream/storage/common/pkgs/ioswitchlrc"
+	"gitlink.org.cn/cloudream/storage/common/pkgs/ioswitchlrc/ops2"
+)
+
+// 计算输入流的打开范围。会把流的范围按条带大小取整
+func calcStreamRange(ctx *GenerateContext) {
+	stripSize := int64(ctx.LRC.ChunkSize * ctx.LRC.K)
+
+	rng := exec.Range{
+		Offset: math.MaxInt64,
+	}
+
+	for _, to := range ctx.Toes {
+		if to.GetDataIndex() == -1 {
+			toRng := to.GetRange()
+			rng.ExtendStart(math2.Floor(toRng.Offset, stripSize))
+			if toRng.Length != nil {
+				rng.ExtendEnd(math2.Ceil(toRng.Offset+*toRng.Length, stripSize))
+			} else {
+				rng.Length = nil
+			}
+
+		} else {
+			toRng := to.GetRange()
+
+			blkStartIndex := math2.FloorDiv(toRng.Offset, int64(ctx.LRC.ChunkSize))
+			rng.ExtendStart(blkStartIndex * stripSize)
+			if toRng.Length != nil {
+				blkEndIndex := math2.CeilDiv(toRng.Offset+*toRng.Length, int64(ctx.LRC.ChunkSize))
+				rng.ExtendEnd(blkEndIndex * stripSize)
+			} else {
+				rng.Length = nil
+			}
+		}
+	}
+
+	ctx.StreamRange = rng
+}
+
+func buildFromNode(ctx *GenerateContext, f ioswitchlrc.From) (*dag.Node, error) {
+	var repRange exec.Range
+	var blkRange exec.Range
+
+	repRange.Offset = ctx.StreamRange.Offset
+	blkRange.Offset = ctx.StreamRange.Offset / int64(ctx.LRC.ChunkSize*ctx.LRC.K) * int64(ctx.LRC.ChunkSize)
+	if ctx.StreamRange.Length != nil {
+		repRngLen := *ctx.StreamRange.Length
+		repRange.Length = &repRngLen
+
+		blkRngLen := *ctx.StreamRange.Length / int64(ctx.LRC.ChunkSize*ctx.LRC.K) * int64(ctx.LRC.ChunkSize)
+		blkRange.Length = &blkRngLen
+	}
+
+	switch f := f.(type) {
+	case *ioswitchlrc.FromNode:
+		n, t := dag.NewNode(ctx.DAG, &ops2.IPFSReadType{
+			FileHash: f.FileHash,
+			Option: ipfs.ReadOption{
+				Offset: 0,
+				Length: -1,
+			},
+		}, &ioswitchlrc.NodeProps{
+			From: f,
+		})
+		ioswitchlrc.SProps(n.OutputStreams[0]).StreamIndex = f.DataIndex
+
+		if f.DataIndex == -1 {
+			t.Option.Offset = repRange.Offset
+			if repRange.Length != nil {
+				t.Option.Length = *repRange.Length
+			}
+		} else {
+			t.Option.Offset = blkRange.Offset
+			if blkRange.Length != nil {
+				t.Option.Length = *blkRange.Length
+			}
+		}
+
+		if f.Node != nil {
+			n.Env.ToEnvWorker(&ioswitchlrc.AgentWorker{Node: *f.Node})
+		}
+
+		return n, nil
+
+	case *ioswitchlrc.FromDriver:
+		n, _ := dag.NewNode(ctx.DAG, &ops.FromDriverType{Handle: f.Handle}, &ioswitchlrc.NodeProps{From: f})
+		n.Env.ToEnvDriver()
+		ioswitchlrc.SProps(n.OutputStreams[0]).StreamIndex = f.DataIndex
+
+		if f.DataIndex == -1 {
+			f.Handle.RangeHint.Offset = repRange.Offset
+			f.Handle.RangeHint.Length = repRange.Length
+		} else {
+			f.Handle.RangeHint.Offset = blkRange.Offset
+			f.Handle.RangeHint.Length = blkRange.Length
+		}
+
+		return n, nil
+
+	default:
+		return nil, fmt.Errorf("unsupported from type %T", f)
+	}
+}
+
+func buildToNode(ctx *GenerateContext, t ioswitchlrc.To) (*dag.Node, error) {
+	switch t := t.(type) {
+	case *ioswitchlrc.ToNode:
+		n, _ := dag.NewNode(ctx.DAG, &ops2.IPFSWriteType{
+			FileHashStoreKey: t.FileHashStoreKey,
+			Range:            t.Range,
+		}, &ioswitchlrc.NodeProps{
+			To: t,
+		})
+
+		return n, nil
+
+	case *ioswitchlrc.ToDriver:
+		n, _ := dag.NewNode(ctx.DAG, &ops.ToDriverType{Handle: t.Handle, Range: t.Range}, &ioswitchlrc.NodeProps{To: t})
+		n.Env.ToEnvDriver()
+
+		return n, nil
+
+	default:
+		return nil, fmt.Errorf("unsupported to type %T", t)
+	}
+}
+
+// 通过流的输入输出位置来确定指令的执行位置。
+// To系列的指令都会有固定的执行位置，这些位置会随着pin操作逐步扩散到整个DAG，
+// 所以理论上不会出现有指令的位置始终无法确定的情况。
+func pin(ctx *GenerateContext) bool {
+	changed := false
+	ctx.DAG.Walk(func(node *dag.Node) bool {
+		var toEnv *dag.NodeEnv
+		for _, out := range node.OutputStreams {
+			for _, to := range out.Toes {
+				if to.Node.Env.Type == dag.EnvUnknown {
+					continue
+				}
+
+				if toEnv == nil {
+					toEnv = &to.Node.Env
+				} else if !toEnv.Equals(to.Node.Env) {
+					toEnv = nil
+					break
+				}
+			}
+		}
+
+		if toEnv != nil {
+			if !node.Env.Equals(*toEnv) {
+				changed = true
+			}
+
+			node.Env = *toEnv
+			return true
+		}
+
+		// 否则根据输入流的始发地来固定
+		var fromEnv *dag.NodeEnv
+		for _, in := range node.InputStreams {
+			if in.From.Node.Env.Type == dag.EnvUnknown {
+				continue
+			}
+
+			if fromEnv == nil {
+				fromEnv = &in.From.Node.Env
+			} else if !fromEnv.Equals(in.From.Node.Env) {
+				fromEnv = nil
+				break
+			}
+		}
+
+		if fromEnv != nil {
+			if !node.Env.Equals(*fromEnv) {
+				changed = true
+			}
+
+			node.Env = *fromEnv
+		}
+		return true
+	})
+
+	return changed
+}
+
+// 对于所有未使用的流，增加Drop指令
+func dropUnused(ctx *GenerateContext) {
+	ctx.DAG.Walk(func(node *dag.Node) bool {
+		for _, out := range node.OutputStreams {
+			if len(out.Toes) == 0 {
+				n := ctx.DAG.NewNode(&ops.DropType{}, &ioswitchlrc.NodeProps{})
+				n.Env = node.Env
+				out.To(n, 0)
+			}
+		}
+		return true
+	})
+}
+
+// 为IPFS写入指令存储结果
+func storeIPFSWriteResult(ctx *GenerateContext) {
+	dag.WalkOnlyType[*ops2.IPFSWriteType](ctx.DAG, func(node *dag.Node, typ *ops2.IPFSWriteType) bool {
+		if typ.FileHashStoreKey == "" {
+			return true
+		}
+
+		n := ctx.DAG.NewNode(&ops.StoreType{
+			StoreKey: typ.FileHashStoreKey,
+		}, &ioswitchlrc.NodeProps{})
+		n.Env.ToEnvDriver()
+
+		node.OutputValues[0].To(n, 0)
+		return true
+	})
+}
+
+// 生成Range指令。StreamRange可能超过文件总大小，但Range指令会在数据量不够时不报错而是正常返回
+func generateRange(ctx *GenerateContext) {
+	ctx.DAG.Walk(func(node *dag.Node) bool {
+		props := ioswitchlrc.NProps(node)
+		if props.To == nil {
+			return true
+		}
+
+		toDataIdx := props.To.GetDataIndex()
+		toRng := props.To.GetRange()
+
+		if toDataIdx == -1 {
+			n := ctx.DAG.NewNode(&ops2.RangeType{
+				Range: exec.Range{
+					Offset: toRng.Offset - ctx.StreamRange.Offset,
+					Length: toRng.Length,
+				},
+			}, &ioswitchlrc.NodeProps{})
+			n.Env = node.InputStreams[0].From.Node.Env
+
+			node.InputStreams[0].To(n, 0)
+			node.InputStreams[0].NotTo(node)
+			n.OutputStreams[0].To(node, 0)
+
+		} else {
+			stripSize := int64(ctx.LRC.ChunkSize * ctx.LRC.K)
+			blkStartIdx := ctx.StreamRange.Offset / stripSize
+
+			blkStart := blkStartIdx * int64(ctx.LRC.ChunkSize)
+
+			n := ctx.DAG.NewNode(&ops2.RangeType{
+				Range: exec.Range{
+					Offset: toRng.Offset - blkStart,
+					Length: toRng.Length,
+				},
+			}, &ioswitchlrc.NodeProps{})
+			n.Env = node.InputStreams[0].From.Node.Env
+
+			node.InputStreams[0].To(n, 0)
+			node.InputStreams[0].NotTo(node)
+			n.OutputStreams[0].To(node, 0)
+		}
+
+		return true
+	})
+}
+
+// 生成Clone指令
+func generateClone(ctx *GenerateContext) {
+	ctx.DAG.Walk(func(node *dag.Node) bool {
+		for _, out := range node.OutputStreams {
+			if len(out.Toes) <= 1 {
+				continue
+			}
+
+			n, t := dag.NewNode(ctx.DAG, &ops2.CloneStreamType{}, &ioswitchlrc.NodeProps{})
+			n.Env = node.Env
+			for _, to := range out.Toes {
+				t.NewOutput(node).To(to.Node, to.SlotIndex)
+			}
+			out.Toes = nil
+			out.To(n, 0)
+		}
+
+		for _, out := range node.OutputValues {
+			if len(out.Toes) <= 1 {
+				continue
+			}
+
+			n, t := dag.NewNode(ctx.DAG, &ops2.CloneVarType{}, &ioswitchlrc.NodeProps{})
+			n.Env = node.Env
+			for _, to := range out.Toes {
+				t.NewOutput(node).To(to.Node, to.SlotIndex)
+			}
+			out.Toes = nil
+			out.To(n, 0)
+		}
+
+		return true
+	})
+}
diff --git a/common/pkgs/ioswitchlrc/utils.go b/common/pkgs/ioswitchlrc/utils.go
new file mode 100644
index 0000000..aa162ee
--- /dev/null
+++ b/common/pkgs/ioswitchlrc/utils.go
@@ -0,0 +1,17 @@
+package ioswitchlrc
+
+import (
+	"gitlink.org.cn/cloudream/common/pkgs/ioswitch/dag"
+)
+
+func NProps(n *dag.Node) *NodeProps {
+	return dag.NProps[*NodeProps](n)
+}
+
+func SProps(str *dag.StreamVar) *VarProps {
+	return dag.SProps[*VarProps](str)
+}
+
+func VProps(v *dag.ValueVar) *VarProps {
+	return dag.VProps[*VarProps](v)
+}