neoneo-go/pkg/vm/vm.go

1644 lines
38 KiB
Go

package vm
import (
"encoding/binary"
"encoding/json"
"fmt"
"io/ioutil"
"math"
"math/big"
"os"
"text/tabwriter"
"unicode/utf8"
"github.com/nspcc-dev/neo-go/pkg/crypto/keys"
"github.com/nspcc-dev/neo-go/pkg/util"
"github.com/nspcc-dev/neo-go/pkg/vm/emit"
"github.com/nspcc-dev/neo-go/pkg/vm/opcode"
"github.com/pkg/errors"
)
type errorAtInstruct struct {
ip int
op opcode.Opcode
err interface{}
}
func (e *errorAtInstruct) Error() string {
return fmt.Sprintf("error encountered at instruction %d (%s): %s", e.ip, e.op, e.err)
}
func newError(ip int, op opcode.Opcode, err interface{}) *errorAtInstruct {
return &errorAtInstruct{ip: ip, op: op, err: err}
}
// StateMessage is a vm state message which could be used as additional info for example by cli.
type StateMessage string
// ScriptHashGetter defines an interface for getting calling, entry and current script hashes.
type ScriptHashGetter interface {
GetCallingScriptHash() util.Uint160
GetEntryScriptHash() util.Uint160
GetCurrentScriptHash() util.Uint160
}
const (
// MaxArraySize is the maximum array size allowed in the VM.
MaxArraySize = 1024
// MaxItemSize is the maximum item size allowed in the VM.
MaxItemSize = 1024 * 1024
// MaxInvocationStackSize is the maximum size of an invocation stack.
MaxInvocationStackSize = 1024
// MaxBigIntegerSizeBits is the maximum size of BigInt item in bits.
MaxBigIntegerSizeBits = 32 * 8
// MaxStackSize is the maximum number of items allowed to be
// on all stacks at once.
MaxStackSize = 2 * 1024
maxSHLArg = MaxBigIntegerSizeBits
)
// VM represents the virtual machine.
type VM struct {
state State
// callbacks to get interops.
getInterop []InteropGetterFunc
// callback to get interop price
getPrice func(*VM, opcode.Opcode, []byte) util.Fixed8
istack *Stack // invocation stack.
estack *Stack // execution stack.
astack *Stack // alt stack.
static *Slot
// Hash to verify in CHECKSIG/CHECKMULTISIG.
checkhash []byte
refs *refCounter
gasConsumed util.Fixed8
gasLimit util.Fixed8
// Public keys cache.
keys map[string]*keys.PublicKey
}
// New returns a new VM object ready to load .avm bytecode scripts.
func New() *VM {
vm := &VM{
getInterop: make([]InteropGetterFunc, 0, 3), // 3 functions is typical for our default usage.
state: haltState,
istack: NewStack("invocation"),
refs: newRefCounter(),
keys: make(map[string]*keys.PublicKey),
}
vm.estack = vm.newItemStack("evaluation")
vm.astack = vm.newItemStack("alt")
vm.RegisterInteropGetter(getDefaultVMInterop)
return vm
}
func (v *VM) newItemStack(n string) *Stack {
s := NewStack(n)
s.refs = v.refs
return s
}
// RegisterInteropGetter registers the given InteropGetterFunc into VM. There
// can be many interop getters and they're probed in LIFO order wrt their
// registration time.
func (v *VM) RegisterInteropGetter(f InteropGetterFunc) {
v.getInterop = append(v.getInterop, f)
}
// SetPriceGetter registers the given PriceGetterFunc in v.
// f accepts vm's Context, current instruction and instruction parameter.
func (v *VM) SetPriceGetter(f func(*VM, opcode.Opcode, []byte) util.Fixed8) {
v.getPrice = f
}
// GasConsumed returns the amount of GAS consumed during execution.
func (v *VM) GasConsumed() util.Fixed8 {
return v.gasConsumed
}
// SetGasLimit sets maximum amount of gas which v can spent.
// If max <= 0, no limit is imposed.
func (v *VM) SetGasLimit(max util.Fixed8) {
v.gasLimit = max
}
// Estack returns the evaluation stack so interop hooks can utilize this.
func (v *VM) Estack() *Stack {
return v.estack
}
// Astack returns the alt stack so interop hooks can utilize this.
func (v *VM) Astack() *Stack {
return v.astack
}
// Istack returns the invocation stack so interop hooks can utilize this.
func (v *VM) Istack() *Stack {
return v.istack
}
// SetPublicKeys sets internal key cache to the specified value (note
// that it doesn't copy them).
func (v *VM) SetPublicKeys(keys map[string]*keys.PublicKey) {
v.keys = keys
}
// GetPublicKeys returns internal key cache (note that it doesn't copy it).
func (v *VM) GetPublicKeys() map[string]*keys.PublicKey {
return v.keys
}
// LoadArgs loads in the arguments used in the Mian entry point.
func (v *VM) LoadArgs(method []byte, args []StackItem) {
if len(args) > 0 {
v.estack.PushVal(args)
}
if method != nil {
v.estack.PushVal(method)
}
}
// PrintOps prints the opcodes of the current loaded program to stdout.
func (v *VM) PrintOps() {
w := tabwriter.NewWriter(os.Stdout, 0, 0, 4, ' ', 0)
fmt.Fprintln(w, "INDEX\tOPCODE\tPARAMETER\t")
realctx := v.Context()
ctx := realctx.Copy()
ctx.ip = 0
ctx.nextip = 0
for {
cursor := ""
instr, parameter, err := ctx.Next()
if ctx.ip == realctx.ip {
cursor = "<<"
}
if err != nil {
fmt.Fprintf(w, "%d\t%s\tERROR: %s\t%s\n", ctx.ip, instr, err, cursor)
break
}
var desc = ""
if parameter != nil {
switch instr {
case opcode.JMP, opcode.JMPIF, opcode.JMPIFNOT, opcode.CALL:
offset := int16(binary.LittleEndian.Uint16(parameter))
desc = fmt.Sprintf("%d (%d/%x)", ctx.ip+int(offset), offset, parameter)
case opcode.PUSHA:
offset := int32(binary.LittleEndian.Uint32(parameter))
desc = fmt.Sprintf("%d (%x)", offset, parameter)
case opcode.SYSCALL:
desc = fmt.Sprintf("%q", parameter)
default:
if utf8.Valid(parameter) {
desc = fmt.Sprintf("%x (%q)", parameter, parameter)
} else {
desc = fmt.Sprintf("%x", parameter)
}
}
}
fmt.Fprintf(w, "%d\t%s\t%s\t%s\n", ctx.ip, instr, desc, cursor)
if ctx.nextip >= len(ctx.prog) {
break
}
}
w.Flush()
}
// AddBreakPoint adds a breakpoint to the current context.
func (v *VM) AddBreakPoint(n int) {
ctx := v.Context()
ctx.breakPoints = append(ctx.breakPoints, n)
}
// AddBreakPointRel adds a breakpoint relative to the current
// instruction pointer.
func (v *VM) AddBreakPointRel(n int) {
ctx := v.Context()
v.AddBreakPoint(ctx.ip + n)
}
// LoadFile loads a program from the given path, ready to execute it.
func (v *VM) LoadFile(path string) error {
b, err := ioutil.ReadFile(path)
if err != nil {
return err
}
v.Load(b)
return nil
}
// Load initializes the VM with the program given.
func (v *VM) Load(prog []byte) {
// Clear all stacks and state, it could be a reload.
v.istack.Clear()
v.estack.Clear()
v.astack.Clear()
v.state = noneState
v.gasConsumed = 0
v.LoadScript(prog)
}
// LoadScript loads a script from the internal script table. It
// will immediately push a new context created from this script to
// the invocation stack and starts executing it.
func (v *VM) LoadScript(b []byte) {
ctx := NewContext(b)
ctx.estack = v.estack
ctx.astack = v.astack
v.istack.PushVal(ctx)
}
// loadScriptWithHash if similar to the LoadScript method, but it also loads
// given script hash directly into the Context to avoid its recalculations. It's
// up to user of this function to make sure the script and hash match each other.
func (v *VM) loadScriptWithHash(b []byte, hash util.Uint160, hasDynamicInvoke bool) {
v.LoadScript(b)
ctx := v.Context()
ctx.scriptHash = hash
ctx.hasDynamicInvoke = hasDynamicInvoke
}
// Context returns the current executed context. Nil if there is no context,
// which implies no program is loaded.
func (v *VM) Context() *Context {
if v.istack.Len() == 0 {
return nil
}
return v.istack.Peek(0).Value().(*Context)
}
// PopResult is used to pop the first item of the evaluation stack. This allows
// us to test compiler and vm in a bi-directional way.
func (v *VM) PopResult() interface{} {
e := v.estack.Pop()
if e != nil {
return e.Value()
}
return nil
}
// Stack returns json formatted representation of the given stack.
func (v *VM) Stack(n string) string {
var s *Stack
if n == "astack" {
s = v.astack
}
if n == "istack" {
s = v.istack
}
if n == "estack" {
s = v.estack
}
b, _ := json.MarshalIndent(s.ToContractParameters(), "", " ")
return string(b)
}
// State returns string representation of the state for the VM.
func (v *VM) State() string {
return v.state.String()
}
// Ready returns true if the VM ready to execute the loaded program.
// Will return false if no program is loaded.
func (v *VM) Ready() bool {
return v.istack.Len() > 0
}
// Run starts the execution of the loaded program.
func (v *VM) Run() error {
if !v.Ready() {
v.state = faultState
return errors.New("no program loaded")
}
if v.state.HasFlag(faultState) {
// VM already ran something and failed, in general its state is
// undefined in this case so we can't run anything.
return errors.New("VM has failed")
}
// haltState (the default) or breakState are safe to continue.
v.state = noneState
for {
// check for breakpoint before executing the next instruction
ctx := v.Context()
if ctx != nil && ctx.atBreakPoint() {
v.state |= breakState
}
switch {
case v.state.HasFlag(faultState):
// Should be caught and reported already by the v.Step(),
// but we're checking here anyway just in case.
return errors.New("VM has failed")
case v.state.HasFlag(haltState), v.state.HasFlag(breakState):
// Normal exit from this loop.
return nil
case v.state == noneState:
if err := v.Step(); err != nil {
return err
}
default:
v.state = faultState
return errors.New("unknown state")
}
}
}
// Step 1 instruction in the program.
func (v *VM) Step() error {
ctx := v.Context()
op, param, err := ctx.Next()
if err != nil {
v.state = faultState
return newError(ctx.ip, op, err)
}
return v.execute(ctx, op, param)
}
// StepInto behaves the same as “step over” in case if the line does not contain a function. Otherwise
// the debugger will enter the called function and continue line-by-line debugging there.
func (v *VM) StepInto() error {
ctx := v.Context()
if ctx == nil {
v.state |= haltState
}
if v.HasStopped() {
return nil
}
if ctx != nil && ctx.prog != nil {
op, param, err := ctx.Next()
if err != nil {
v.state = faultState
return newError(ctx.ip, op, err)
}
vErr := v.execute(ctx, op, param)
if vErr != nil {
return vErr
}
}
cctx := v.Context()
if cctx != nil && cctx.atBreakPoint() {
v.state = breakState
}
return nil
}
// StepOut takes the debugger to the line where the current function was called.
func (v *VM) StepOut() error {
var err error
if v.state == breakState {
v.state = noneState
} else {
v.state = breakState
}
expSize := v.istack.len
for v.state == noneState && v.istack.len >= expSize {
err = v.StepInto()
}
return err
}
// StepOver takes the debugger to the line that will step over a given line.
// If the line contains a function the function will be executed and the result returned without debugging each line.
func (v *VM) StepOver() error {
var err error
if v.HasStopped() {
return err
}
if v.state == breakState {
v.state = noneState
} else {
v.state = breakState
}
expSize := v.istack.len
for {
err = v.StepInto()
if !(v.state == noneState && v.istack.len > expSize) {
break
}
}
if v.state == noneState {
v.state = breakState
}
return err
}
// HasFailed returns whether VM is in the failed state now. Usually used to
// check status after Run.
func (v *VM) HasFailed() bool {
return v.state.HasFlag(faultState)
}
// HasStopped returns whether VM is in Halt or Failed state.
func (v *VM) HasStopped() bool {
return v.state.HasFlag(haltState) || v.state.HasFlag(faultState)
}
// HasHalted returns whether VM is in Halt state.
func (v *VM) HasHalted() bool {
return v.state.HasFlag(haltState)
}
// AtBreakpoint returns whether VM is at breakpoint.
func (v *VM) AtBreakpoint() bool {
return v.state.HasFlag(breakState)
}
// SetCheckedHash sets checked hash for CHECKSIG and CHECKMULTISIG instructions.
func (v *VM) SetCheckedHash(h []byte) {
v.checkhash = make([]byte, len(h))
copy(v.checkhash, h)
}
// GetInteropID converts instruction parameter to an interop ID.
func GetInteropID(parameter []byte) uint32 {
return binary.LittleEndian.Uint32(parameter)
}
// GetInteropByID returns interop function together with price.
// Registered callbacks are checked in LIFO order.
func (v *VM) GetInteropByID(id uint32) *InteropFuncPrice {
for i := len(v.getInterop) - 1; i >= 0; i-- {
if ifunc := v.getInterop[i](id); ifunc != nil {
return ifunc
}
}
return nil
}
// execute performs an instruction cycle in the VM. Acting on the instruction (opcode).
func (v *VM) execute(ctx *Context, op opcode.Opcode, parameter []byte) (err error) {
// Instead of polluting the whole VM logic with error handling, we will recover
// each panic at a central point, putting the VM in a fault state and setting error.
defer func() {
if errRecover := recover(); errRecover != nil {
v.state = faultState
err = newError(ctx.ip, op, errRecover)
} else if v.refs.size > MaxStackSize {
v.state = faultState
err = newError(ctx.ip, op, "stack is too big")
}
}()
if v.getPrice != nil && ctx.ip < len(ctx.prog) {
v.gasConsumed += v.getPrice(v, op, parameter)
if v.gasLimit > 0 && v.gasConsumed > v.gasLimit {
panic("gas limit is exceeded")
}
}
if op <= opcode.PUSHINT256 {
v.estack.PushVal(emit.BytesToInt(parameter))
return
}
switch op {
case opcode.PUSHM1, opcode.PUSH0, opcode.PUSH1, opcode.PUSH2, opcode.PUSH3,
opcode.PUSH4, opcode.PUSH5, opcode.PUSH6, opcode.PUSH7,
opcode.PUSH8, opcode.PUSH9, opcode.PUSH10, opcode.PUSH11,
opcode.PUSH12, opcode.PUSH13, opcode.PUSH14, opcode.PUSH15,
opcode.PUSH16:
val := int(op) - int(opcode.PUSH0)
v.estack.PushVal(val)
case opcode.PUSHDATA1, opcode.PUSHDATA2, opcode.PUSHDATA4:
v.estack.PushVal(parameter)
case opcode.PUSHA:
n := int32(binary.LittleEndian.Uint32(parameter))
if n < 0 || int(n) > len(ctx.prog) {
panic(fmt.Sprintf("invalid pointer offset (%d)", n))
}
ptr := NewPointerItem(int(n), ctx.prog)
v.estack.PushVal(ptr)
case opcode.PUSHNULL:
v.estack.PushVal(NullItem{})
case opcode.ISNULL:
res := v.estack.Pop().value.Equals(NullItem{})
v.estack.PushVal(res)
case opcode.ISTYPE:
res := v.estack.Pop().Item()
v.estack.PushVal(res.Type() == StackItemType(parameter[0]))
case opcode.CONVERT:
typ := StackItemType(parameter[0])
item := v.estack.Pop().Item()
result, err := item.Convert(typ)
if err != nil {
panic(err)
}
v.estack.PushVal(result)
case opcode.INITSSLOT:
if v.static != nil {
panic("already initialized")
}
if parameter[0] == 0 {
panic("zero argument")
}
v.static = v.newSlot(int(parameter[0]))
case opcode.INITSLOT:
if ctx.local != nil || ctx.arguments != nil {
panic("already initialized")
}
if parameter[0] == 0 && parameter[1] == 0 {
panic("zero argument")
}
if parameter[0] > 0 {
ctx.local = v.newSlot(int(parameter[0]))
}
if parameter[1] > 0 {
sz := int(parameter[1])
ctx.arguments = v.newSlot(sz)
for i := 0; i < sz; i++ {
ctx.arguments.Set(i, v.estack.Pop().Item())
}
}
case opcode.LDSFLD0, opcode.LDSFLD1, opcode.LDSFLD2, opcode.LDSFLD3, opcode.LDSFLD4, opcode.LDSFLD5, opcode.LDSFLD6:
item := v.static.Get(int(op - opcode.LDSFLD0))
v.estack.PushVal(item)
case opcode.LDSFLD:
item := v.static.Get(int(parameter[0]))
v.estack.PushVal(item)
case opcode.STSFLD0, opcode.STSFLD1, opcode.STSFLD2, opcode.STSFLD3, opcode.STSFLD4, opcode.STSFLD5, opcode.STSFLD6:
item := v.estack.Pop().Item()
v.static.Set(int(op-opcode.STSFLD0), item)
case opcode.STSFLD:
item := v.estack.Pop().Item()
v.static.Set(int(parameter[0]), item)
case opcode.LDLOC0, opcode.LDLOC1, opcode.LDLOC2, opcode.LDLOC3, opcode.LDLOC4, opcode.LDLOC5, opcode.LDLOC6:
item := ctx.local.Get(int(op - opcode.LDLOC0))
v.estack.PushVal(item)
case opcode.LDLOC:
item := ctx.local.Get(int(parameter[0]))
v.estack.PushVal(item)
case opcode.STLOC0, opcode.STLOC1, opcode.STLOC2, opcode.STLOC3, opcode.STLOC4, opcode.STLOC5, opcode.STLOC6:
item := v.estack.Pop().Item()
ctx.local.Set(int(op-opcode.STLOC0), item)
case opcode.STLOC:
item := v.estack.Pop().Item()
ctx.local.Set(int(parameter[0]), item)
case opcode.LDARG0, opcode.LDARG1, opcode.LDARG2, opcode.LDARG3, opcode.LDARG4, opcode.LDARG5, opcode.LDARG6:
item := ctx.arguments.Get(int(op - opcode.LDARG0))
v.estack.PushVal(item)
case opcode.LDARG:
item := ctx.arguments.Get(int(parameter[0]))
v.estack.PushVal(item)
case opcode.STARG0, opcode.STARG1, opcode.STARG2, opcode.STARG3, opcode.STARG4, opcode.STARG5, opcode.STARG6:
item := v.estack.Pop().Item()
ctx.arguments.Set(int(op-opcode.STARG0), item)
case opcode.STARG:
item := v.estack.Pop().Item()
ctx.arguments.Set(int(parameter[0]), item)
case opcode.NEWBUFFER:
n := toInt(v.estack.Pop().BigInt())
if n < 0 || n > MaxItemSize {
panic("invalid size")
}
v.estack.PushVal(NewBufferItem(make([]byte, n)))
case opcode.MEMCPY:
n := toInt(v.estack.Pop().BigInt())
if n < 0 {
panic("invalid size")
}
si := toInt(v.estack.Pop().BigInt())
if si < 0 {
panic("invalid source index")
}
src := v.estack.Pop().Bytes()
if sum := si + n; sum < 0 || sum > len(src) {
panic("size is too big")
}
di := toInt(v.estack.Pop().BigInt())
if di < 0 {
panic("invalid destination index")
}
dst := v.estack.Pop().value.(*BufferItem).value
if sum := si + n; sum < 0 || sum > len(dst) {
panic("size is too big")
}
copy(dst[di:], src[si:si+n])
case opcode.CAT:
b := v.estack.Pop().Bytes()
a := v.estack.Pop().Bytes()
if l := len(a) + len(b); l > MaxItemSize {
panic(fmt.Sprintf("too big item: %d", l))
}
ab := append(a, b...)
v.estack.PushVal(NewBufferItem(ab))
case opcode.SUBSTR:
l := int(v.estack.Pop().BigInt().Int64())
if l < 0 {
panic("negative length")
}
o := int(v.estack.Pop().BigInt().Int64())
if o < 0 {
panic("negative index")
}
s := v.estack.Pop().Bytes()
last := l + o
if last > len(s) {
panic("invalid offset")
}
v.estack.PushVal(NewBufferItem(s[o:last]))
case opcode.LEFT:
l := int(v.estack.Pop().BigInt().Int64())
if l < 0 {
panic("negative length")
}
s := v.estack.Pop().Bytes()
if t := len(s); l > t {
l = t
}
v.estack.PushVal(NewBufferItem(s[:l]))
case opcode.RIGHT:
l := int(v.estack.Pop().BigInt().Int64())
if l < 0 {
panic("negative length")
}
s := v.estack.Pop().Bytes()
v.estack.PushVal(NewBufferItem(s[len(s)-l:]))
case opcode.DEPTH:
v.estack.PushVal(v.estack.Len())
case opcode.DROP:
if v.estack.Len() < 1 {
panic("stack is too small")
}
v.estack.Pop()
case opcode.NIP:
elem := v.estack.RemoveAt(1)
if elem == nil {
panic("no second element found")
}
case opcode.XDROP:
n := int(v.estack.Pop().BigInt().Int64())
if n < 0 {
panic("invalid length")
}
e := v.estack.RemoveAt(n)
if e == nil {
panic("bad index")
}
case opcode.CLEAR:
v.estack.Clear()
case opcode.DUP:
v.estack.Push(v.estack.Dup(0))
case opcode.OVER:
a := v.estack.Dup(1)
if a == nil {
panic("no second element found")
}
v.estack.Push(a)
case opcode.PICK:
n := int(v.estack.Pop().BigInt().Int64())
if n < 0 {
panic("negative stack item returned")
}
a := v.estack.Dup(n)
if a == nil {
panic("no nth element found")
}
v.estack.Push(a)
case opcode.TUCK:
a := v.estack.Dup(0)
if a == nil {
panic("no top-level element found")
}
if v.estack.Len() < 2 {
panic("can't TUCK with a one-element stack")
}
v.estack.InsertAt(a, 2)
case opcode.SWAP:
err := v.estack.Swap(1, 0)
if err != nil {
panic(err.Error())
}
case opcode.ROT:
err := v.estack.Roll(2)
if err != nil {
panic(err.Error())
}
case opcode.ROLL:
n := int(v.estack.Pop().BigInt().Int64())
err := v.estack.Roll(n)
if err != nil {
panic(err.Error())
}
case opcode.REVERSE3, opcode.REVERSE4, opcode.REVERSEN:
n := 3
switch op {
case opcode.REVERSE4:
n = 4
case opcode.REVERSEN:
n = int(v.estack.Pop().BigInt().Int64())
}
if err := v.estack.ReverseTop(n); err != nil {
panic(err.Error())
}
// Bit operations.
case opcode.INVERT:
// inplace
e := v.estack.Peek(0)
i := e.BigInt()
e.value = makeStackItem(i.Not(i))
case opcode.AND:
b := v.estack.Pop().BigInt()
a := v.estack.Pop().BigInt()
v.estack.PushVal(new(big.Int).And(b, a))
case opcode.OR:
b := v.estack.Pop().BigInt()
a := v.estack.Pop().BigInt()
v.estack.PushVal(new(big.Int).Or(b, a))
case opcode.XOR:
b := v.estack.Pop().BigInt()
a := v.estack.Pop().BigInt()
v.estack.PushVal(new(big.Int).Xor(b, a))
case opcode.EQUAL, opcode.NOTEQUAL:
b := v.estack.Pop()
if b == nil {
panic("no top-level element found")
}
a := v.estack.Pop()
if a == nil {
panic("no second-to-the-top element found")
}
v.estack.PushVal(a.value.Equals(b.value) == (op == opcode.EQUAL))
// Numeric operations.
case opcode.SIGN:
x := v.estack.Pop().BigInt()
v.estack.PushVal(x.Sign())
case opcode.ABS:
x := v.estack.Pop().BigInt()
v.estack.PushVal(x.Abs(x))
case opcode.NEGATE:
x := v.estack.Pop().BigInt()
v.estack.PushVal(x.Neg(x))
case opcode.INC:
x := v.estack.Pop().BigInt()
a := new(big.Int).Add(x, big.NewInt(1))
v.estack.PushVal(a)
case opcode.DEC:
x := v.estack.Pop().BigInt()
a := new(big.Int).Sub(x, big.NewInt(1))
v.estack.PushVal(a)
case opcode.ADD:
a := v.estack.Pop().BigInt()
b := v.estack.Pop().BigInt()
c := new(big.Int).Add(a, b)
v.estack.PushVal(c)
case opcode.SUB:
b := v.estack.Pop().BigInt()
a := v.estack.Pop().BigInt()
c := new(big.Int).Sub(a, b)
v.estack.PushVal(c)
case opcode.MUL:
a := v.estack.Pop().BigInt()
b := v.estack.Pop().BigInt()
c := new(big.Int).Mul(a, b)
v.estack.PushVal(c)
case opcode.DIV:
b := v.estack.Pop().BigInt()
a := v.estack.Pop().BigInt()
v.estack.PushVal(new(big.Int).Quo(a, b))
case opcode.MOD:
b := v.estack.Pop().BigInt()
a := v.estack.Pop().BigInt()
v.estack.PushVal(new(big.Int).Rem(a, b))
case opcode.SHL, opcode.SHR:
b := v.estack.Pop().BigInt().Int64()
if b == 0 {
return
} else if b < 0 || b > maxSHLArg {
panic(fmt.Sprintf("operand must be between %d and %d", 0, maxSHLArg))
}
a := v.estack.Pop().BigInt()
var item big.Int
if op == opcode.SHL {
item.Lsh(a, uint(b))
} else {
item.Rsh(a, uint(b))
}
v.estack.PushVal(&item)
case opcode.NOT:
x := v.estack.Pop().Bool()
v.estack.PushVal(!x)
case opcode.BOOLAND:
b := v.estack.Pop().Bool()
a := v.estack.Pop().Bool()
v.estack.PushVal(a && b)
case opcode.BOOLOR:
b := v.estack.Pop().Bool()
a := v.estack.Pop().Bool()
v.estack.PushVal(a || b)
case opcode.NZ:
x := v.estack.Pop().BigInt()
v.estack.PushVal(x.Sign() != 0)
case opcode.NUMEQUAL:
b := v.estack.Pop().BigInt()
a := v.estack.Pop().BigInt()
v.estack.PushVal(a.Cmp(b) == 0)
case opcode.NUMNOTEQUAL:
b := v.estack.Pop().BigInt()
a := v.estack.Pop().BigInt()
v.estack.PushVal(a.Cmp(b) != 0)
case opcode.LT:
b := v.estack.Pop().BigInt()
a := v.estack.Pop().BigInt()
v.estack.PushVal(a.Cmp(b) == -1)
case opcode.LTE:
b := v.estack.Pop().BigInt()
a := v.estack.Pop().BigInt()
v.estack.PushVal(a.Cmp(b) <= 0)
case opcode.GT:
b := v.estack.Pop().BigInt()
a := v.estack.Pop().BigInt()
v.estack.PushVal(a.Cmp(b) == 1)
case opcode.GTE:
b := v.estack.Pop().BigInt()
a := v.estack.Pop().BigInt()
v.estack.PushVal(a.Cmp(b) >= 0)
case opcode.MIN:
b := v.estack.Pop().BigInt()
a := v.estack.Pop().BigInt()
val := a
if a.Cmp(b) == 1 {
val = b
}
v.estack.PushVal(val)
case opcode.MAX:
b := v.estack.Pop().BigInt()
a := v.estack.Pop().BigInt()
val := a
if a.Cmp(b) == -1 {
val = b
}
v.estack.PushVal(val)
case opcode.WITHIN:
b := v.estack.Pop().BigInt()
a := v.estack.Pop().BigInt()
x := v.estack.Pop().BigInt()
v.estack.PushVal(a.Cmp(x) <= 0 && x.Cmp(b) == -1)
// Object operations
case opcode.NEWARRAY0:
v.estack.PushVal(&ArrayItem{[]StackItem{}})
case opcode.NEWARRAY, opcode.NEWARRAYT:
item := v.estack.Pop()
switch t := item.value.(type) {
case *StructItem:
arr := make([]StackItem, len(t.value))
copy(arr, t.value)
v.estack.PushVal(&ArrayItem{arr})
case *ArrayItem:
v.estack.PushVal(t)
default:
n := item.BigInt().Int64()
if n > MaxArraySize {
panic("too long array")
}
typ := AnyT
if op == opcode.NEWARRAYT {
typ = StackItemType(parameter[0])
}
items := makeArrayOfType(int(n), typ)
v.estack.PushVal(&ArrayItem{items})
}
case opcode.NEWSTRUCT0:
v.estack.PushVal(&StructItem{[]StackItem{}})
case opcode.NEWSTRUCT:
item := v.estack.Pop()
switch t := item.value.(type) {
case *ArrayItem:
arr := make([]StackItem, len(t.value))
copy(arr, t.value)
v.estack.PushVal(&StructItem{arr})
case *StructItem:
v.estack.PushVal(t)
default:
n := item.BigInt().Int64()
if n > MaxArraySize {
panic("too long struct")
}
items := makeArrayOfType(int(n), AnyT)
v.estack.PushVal(&StructItem{items})
}
case opcode.APPEND:
itemElem := v.estack.Pop()
arrElem := v.estack.Pop()
val := cloneIfStruct(itemElem.value)
switch t := arrElem.value.(type) {
case *ArrayItem:
arr := t.Value().([]StackItem)
if len(arr) >= MaxArraySize {
panic("too long array")
}
arr = append(arr, val)
t.value = arr
case *StructItem:
arr := t.Value().([]StackItem)
if len(arr) >= MaxArraySize {
panic("too long struct")
}
arr = append(arr, val)
t.value = arr
default:
panic("APPEND: not of underlying type Array")
}
v.refs.Add(val)
case opcode.PACK:
n := int(v.estack.Pop().BigInt().Int64())
if n < 0 || n > v.estack.Len() || n > MaxArraySize {
panic("OPACK: invalid length")
}
items := make([]StackItem, n)
for i := 0; i < n; i++ {
items[i] = v.estack.Pop().value
}
v.estack.PushVal(items)
case opcode.UNPACK:
a := v.estack.Pop().Array()
l := len(a)
for i := l - 1; i >= 0; i-- {
v.estack.PushVal(a[i])
}
v.estack.PushVal(l)
case opcode.PICKITEM:
key := v.estack.Pop()
validateMapKey(key)
obj := v.estack.Pop()
index := int(key.BigInt().Int64())
switch t := obj.value.(type) {
// Struct and Array items have their underlying value as []StackItem.
case *ArrayItem, *StructItem:
arr := t.Value().([]StackItem)
if index < 0 || index >= len(arr) {
panic("PICKITEM: invalid index")
}
item := arr[index].Dup()
v.estack.PushVal(item)
case *MapItem:
index := t.Index(key.Item())
if index < 0 {
panic("invalid key")
}
v.estack.Push(&Element{value: t.value[index].Value.Dup()})
default:
arr := obj.Bytes()
if index < 0 || index >= len(arr) {
panic("PICKITEM: invalid index")
}
item := arr[index]
v.estack.PushVal(int(item))
}
case opcode.SETITEM:
item := v.estack.Pop().value
key := v.estack.Pop()
validateMapKey(key)
obj := v.estack.Pop()
switch t := obj.value.(type) {
// Struct and Array items have their underlying value as []StackItem.
case *ArrayItem, *StructItem:
arr := t.Value().([]StackItem)
index := int(key.BigInt().Int64())
if index < 0 || index >= len(arr) {
panic("SETITEM: invalid index")
}
v.refs.Remove(arr[index])
arr[index] = item
v.refs.Add(arr[index])
case *MapItem:
if i := t.Index(key.value); i >= 0 {
v.refs.Remove(t.value[i].Value)
} else if len(t.value) >= MaxArraySize {
panic("too big map")
}
t.Add(key.value, item)
v.refs.Add(item)
case *BufferItem:
index := toInt(key.BigInt())
if index < 0 || index >= len(t.value) {
panic("invalid index")
}
bi, err := item.TryInteger()
b := toInt(bi)
if err != nil || b < math.MinInt8 || b > math.MaxUint8 {
panic("invalid value")
}
t.value[index] = byte(b)
default:
panic(fmt.Sprintf("SETITEM: invalid item type %s", t))
}
case opcode.REVERSEITEMS:
item := v.estack.Pop()
switch t := item.value.(type) {
case *ArrayItem, *StructItem:
a := t.Value().([]StackItem)
for i, j := 0, len(a)-1; i < j; i, j = i+1, j-1 {
a[i], a[j] = a[j], a[i]
}
case *BufferItem:
for i, j := 0, len(t.value)-1; i < j; i, j = i+1, j-1 {
t.value[i], t.value[j] = t.value[j], t.value[i]
}
default:
panic(fmt.Sprintf("invalid item type %s", t))
}
case opcode.REMOVE:
key := v.estack.Pop()
validateMapKey(key)
elem := v.estack.Pop()
switch t := elem.value.(type) {
case *ArrayItem:
a := t.value
k := int(key.BigInt().Int64())
if k < 0 || k >= len(a) {
panic("REMOVE: invalid index")
}
v.refs.Remove(a[k])
a = append(a[:k], a[k+1:]...)
t.value = a
case *StructItem:
a := t.value
k := int(key.BigInt().Int64())
if k < 0 || k >= len(a) {
panic("REMOVE: invalid index")
}
v.refs.Remove(a[k])
a = append(a[:k], a[k+1:]...)
t.value = a
case *MapItem:
index := t.Index(key.Item())
// NEO 2.0 doesn't error on missing key.
if index >= 0 {
v.refs.Remove(t.value[index].Value)
t.Drop(index)
}
default:
panic("REMOVE: invalid type")
}
case opcode.CLEARITEMS:
elem := v.estack.Pop()
switch t := elem.value.(type) {
case *ArrayItem:
for _, item := range t.value {
v.refs.Remove(item)
}
t.value = t.value[:0]
case *StructItem:
for _, item := range t.value {
v.refs.Remove(item)
}
t.value = t.value[:0]
case *MapItem:
for i := range t.value {
v.refs.Remove(t.value[i].Value)
}
t.value = t.value[:0]
default:
panic("CLEARITEMS: invalid type")
}
case opcode.SIZE:
elem := v.estack.Pop()
// Cause there is no native (byte) item type here, hence we need to check
// the type of the item for array size operations.
switch t := elem.Value().(type) {
case []StackItem:
v.estack.PushVal(len(t))
case []MapElement:
v.estack.PushVal(len(t))
default:
v.estack.PushVal(len(elem.Bytes()))
}
case opcode.JMP, opcode.JMPL, opcode.JMPIF, opcode.JMPIFL, opcode.JMPIFNOT, opcode.JMPIFNOTL,
opcode.JMPEQ, opcode.JMPEQL, opcode.JMPNE, opcode.JMPNEL,
opcode.JMPGT, opcode.JMPGTL, opcode.JMPGE, opcode.JMPGEL,
opcode.JMPLT, opcode.JMPLTL, opcode.JMPLE, opcode.JMPLEL:
offset := v.getJumpOffset(ctx, parameter, 0)
cond := true
switch op {
case opcode.JMP, opcode.JMPL:
case opcode.JMPIF, opcode.JMPIFL, opcode.JMPIFNOT, opcode.JMPIFNOTL:
cond = v.estack.Pop().Bool() == (op == opcode.JMPIF || op == opcode.JMPIFL)
default:
b := v.estack.Pop().BigInt()
a := v.estack.Pop().BigInt()
cond = getJumpCondition(op, a, b)
}
v.jumpIf(ctx, offset, cond)
case opcode.CALL, opcode.CALLL:
v.checkInvocationStackSize()
newCtx := ctx.Copy()
newCtx.local = nil
newCtx.arguments = nil
newCtx.rvcount = -1
v.istack.PushVal(newCtx)
offset := v.getJumpOffset(newCtx, parameter, 0)
v.jumpIf(newCtx, offset, true)
case opcode.CALLA:
ptr := v.estack.Pop().Item().(*PointerItem)
if ptr.hash != ctx.ScriptHash() {
panic("invalid script in pointer")
}
newCtx := ctx.Copy()
newCtx.local = nil
newCtx.arguments = nil
newCtx.rvcount = -1
v.istack.PushVal(newCtx)
v.jumpIf(newCtx, ptr.pos, true)
case opcode.SYSCALL:
interopID := GetInteropID(parameter)
ifunc := v.GetInteropByID(interopID)
if ifunc == nil {
panic(fmt.Sprintf("interop hook (%q/0x%x) not registered", parameter, interopID))
}
if err := ifunc.Func(v); err != nil {
panic(fmt.Sprintf("failed to invoke syscall: %s", err))
}
case opcode.RET:
oldCtx := v.istack.Pop().Value().(*Context)
rvcount := oldCtx.rvcount
oldEstack := v.estack
if rvcount > 0 && oldEstack.Len() < rvcount {
panic("missing some return elements")
}
if v.istack.Len() == 0 {
v.state = haltState
break
}
newEstack := v.Context().estack
if oldEstack != newEstack {
if rvcount < 0 {
rvcount = oldEstack.Len()
}
for i := rvcount; i > 0; i-- {
elem := oldEstack.RemoveAt(i - 1)
newEstack.Push(elem)
}
v.estack = newEstack
v.astack = v.Context().astack
}
case opcode.NEWMAP:
v.estack.Push(&Element{value: NewMapItem()})
case opcode.KEYS:
item := v.estack.Pop()
if item == nil {
panic("no argument")
}
m, ok := item.value.(*MapItem)
if !ok {
panic("not a Map")
}
arr := make([]StackItem, 0, len(m.value))
for k := range m.value {
arr = append(arr, m.value[k].Key.Dup())
}
v.estack.PushVal(arr)
case opcode.VALUES:
item := v.estack.Pop()
if item == nil {
panic("no argument")
}
var arr []StackItem
switch t := item.value.(type) {
case *ArrayItem, *StructItem:
src := t.Value().([]StackItem)
arr = make([]StackItem, len(src))
for i := range src {
arr[i] = cloneIfStruct(src[i])
}
case *MapItem:
arr = make([]StackItem, 0, len(t.value))
for k := range t.value {
arr = append(arr, cloneIfStruct(t.value[k].Value))
}
default:
panic("not a Map, Array or Struct")
}
v.estack.PushVal(arr)
case opcode.HASKEY:
key := v.estack.Pop()
validateMapKey(key)
c := v.estack.Pop()
if c == nil {
panic("no value found")
}
switch t := c.value.(type) {
case *ArrayItem, *StructItem:
index := key.BigInt().Int64()
if index < 0 {
panic("negative index")
}
v.estack.PushVal(index < int64(len(c.Array())))
case *MapItem:
v.estack.PushVal(t.Has(key.Item()))
case *BufferItem:
index := key.BigInt().Int64()
if index < 0 {
panic("negative index")
}
v.estack.PushVal(index < int64(len(t.value)))
default:
panic("wrong collection type")
}
case opcode.NOP:
// unlucky ^^
case opcode.THROW:
panic("THROW")
case opcode.ABORT:
panic("ABORT")
case opcode.ASSERT:
if !v.estack.Pop().Bool() {
panic("ASSERT failed")
}
default:
panic(fmt.Sprintf("unknown opcode %s", op.String()))
}
return
}
// getJumpCondition performs opcode specific comparison of a and b
func getJumpCondition(op opcode.Opcode, a, b *big.Int) bool {
cmp := a.Cmp(b)
switch op {
case opcode.JMPEQ, opcode.JMPEQL:
return cmp == 0
case opcode.JMPNE, opcode.JMPNEL:
return cmp != 0
case opcode.JMPGT, opcode.JMPGTL:
return cmp > 0
case opcode.JMPGE, opcode.JMPGEL:
return cmp >= 0
case opcode.JMPLT, opcode.JMPLTL:
return cmp < 0
case opcode.JMPLE, opcode.JMPLEL:
return cmp <= 0
default:
panic(fmt.Sprintf("invalid JMP* opcode: %s", op))
}
}
// jumpIf performs jump to offset if cond is true.
func (v *VM) jumpIf(ctx *Context, offset int, cond bool) {
if cond {
ctx.nextip = offset
}
}
// getJumpOffset returns instruction number in a current context
// to a which JMP should be performed.
// parameter should have length either 1 or 4 and
// is interpreted as little-endian.
func (v *VM) getJumpOffset(ctx *Context, parameter []byte, mod int) int {
var rOffset int32
switch l := len(parameter); l {
case 1:
rOffset = int32(int8(parameter[0]))
case 4:
rOffset = int32(binary.LittleEndian.Uint32(parameter))
default:
panic(fmt.Sprintf("invalid JMP* parameter length: %d", l))
}
offset := ctx.ip + int(rOffset) + mod
if offset < 0 || offset > len(ctx.prog) {
panic(fmt.Sprintf("JMP: invalid offset %d ip at %d", offset, ctx.ip))
}
return offset
}
// CheckMultisigPar checks if sigs contains sufficient valid signatures.
func CheckMultisigPar(v *VM, h []byte, pkeys [][]byte, sigs [][]byte) bool {
if len(sigs) == 1 {
return checkMultisig1(v, h, pkeys, sigs[0])
}
k1, k2 := 0, len(pkeys)-1
s1, s2 := 0, len(sigs)-1
type task struct {
pub *keys.PublicKey
signum int
}
type verify struct {
ok bool
signum int
}
worker := func(ch <-chan task, result chan verify) {
for {
t, ok := <-ch
if !ok {
return
}
result <- verify{
signum: t.signum,
ok: t.pub.Verify(sigs[t.signum], h),
}
}
}
const workerCount = 3
tasks := make(chan task, 2)
results := make(chan verify, len(sigs))
for i := 0; i < workerCount; i++ {
go worker(tasks, results)
}
tasks <- task{pub: v.bytesToPublicKey(pkeys[k1]), signum: s1}
tasks <- task{pub: v.bytesToPublicKey(pkeys[k2]), signum: s2}
sigok := true
taskCount := 2
loop:
for r := range results {
goingForward := true
taskCount--
if r.signum == s2 {
goingForward = false
}
if k1+1 == k2 {
sigok = r.ok && s1+1 == s2
if taskCount != 0 && sigok {
continue
}
break loop
} else if r.ok {
if s1+1 == s2 {
if taskCount != 0 && sigok {
continue
}
break loop
}
if goingForward {
s1++
} else {
s2--
}
}
var nextSig, nextKey int
if goingForward {
k1++
nextSig = s1
nextKey = k1
} else {
k2--
nextSig = s2
nextKey = k2
}
taskCount++
tasks <- task{pub: v.bytesToPublicKey(pkeys[nextKey]), signum: nextSig}
}
close(tasks)
return sigok
}
func checkMultisig1(v *VM, h []byte, pkeys [][]byte, sig []byte) bool {
for i := range pkeys {
pkey := v.bytesToPublicKey(pkeys[i])
if pkey.Verify(sig, h) {
return true
}
}
return false
}
func cloneIfStruct(item StackItem) StackItem {
switch it := item.(type) {
case *StructItem:
return it.Clone()
default:
return it
}
}
func makeArrayOfType(n int, typ StackItemType) []StackItem {
if !typ.IsValid() {
panic(fmt.Sprintf("invalid stack item type: %d", typ))
}
items := make([]StackItem, n)
for i := range items {
switch typ {
case BooleanT:
items[i] = NewBoolItem(false)
case IntegerT:
items[i] = NewBigIntegerItem(big.NewInt(0))
case ByteArrayT:
items[i] = NewByteArrayItem([]byte{})
default:
items[i] = NullItem{}
}
}
return items
}
func validateMapKey(key *Element) {
if key == nil {
panic("no key found")
}
if !isValidMapKey(key.Item()) {
panic("key can't be a collection")
}
}
func (v *VM) checkInvocationStackSize() {
if v.istack.len >= MaxInvocationStackSize {
panic("invocation stack is too big")
}
}
// bytesToPublicKey is a helper deserializing keys using cache and panicing on
// error.
func (v *VM) bytesToPublicKey(b []byte) *keys.PublicKey {
var pkey *keys.PublicKey
s := string(b)
if v.keys[s] != nil {
pkey = v.keys[s]
} else {
var err error
pkey, err = keys.NewPublicKeyFromBytes(b)
if err != nil {
panic(err.Error())
}
v.keys[s] = pkey
}
return pkey
}
// GetCallingScriptHash implements ScriptHashGetter interface
func (v *VM) GetCallingScriptHash() util.Uint160 {
return v.getContextScriptHash(1)
}
// GetEntryScriptHash implements ScriptHashGetter interface
func (v *VM) GetEntryScriptHash() util.Uint160 {
return v.getContextScriptHash(v.Istack().Len() - 1)
}
// GetCurrentScriptHash implements ScriptHashGetter interface
func (v *VM) GetCurrentScriptHash() util.Uint160 {
return v.getContextScriptHash(0)
}
// toInt converts an item to a 32-bit int.
func toInt(i *big.Int) int {
if !i.IsInt64() {
panic("not an int32")
}
n := i.Int64()
if n < math.MinInt32 || n > math.MaxInt32 {
panic("not an int32")
}
return int(n)
}