"sync"
)
+// The default bencode string length limit. This is a poor attempt to prevent excessive memory
+// allocation when parsing, but also leaves the window open to implement a better solution.
+const DefaultDecodeMaxStrLen = 1<<27 - 1 // ~128MiB
+
+type MaxStrLen = int64
+
type Decoder struct {
+ // Maximum parsed bencode string length. Defaults to DefaultMaxStrLen if zero.
+ MaxStrLen MaxStrLen
+
r interface {
io.ByteScanner
io.Reader
return
}
r := recover()
+ if r == nil {
+ return
+ }
_, ok := r.(runtime.Error)
if ok {
panic(r)
}
- err, ok = r.(error)
- if !ok && r != nil {
+ if err, ok = r.(error); !ok {
panic(r)
}
+ // Errors thrown from deeper in parsing are unexpected. At value boundaries, errors should
+ // be returned directly (at least until all the panic nonsense is removed entirely).
+ if err == io.EOF {
+ err = io.ErrUnexpectedEOF
+ }
}()
pv := reflect.ValueOf(v)
})
}
-// called when 'i' was consumed
-func (d *Decoder) parseInt(v reflect.Value) {
- start := d.Offset - 1
+// Assume the 'i' is already consumed. Read and validate the rest of an int into the buffer.
+func (d *Decoder) readInt() error {
+ // start := d.Offset - 1
d.readUntil('e')
- if d.buf.Len() == 0 {
- panic(&SyntaxError{
- Offset: start,
- What: errors.New("empty integer value"),
- })
- }
+ if err := d.checkBufferedInt(); err != nil {
+ return err
+ }
+ // if d.buf.Len() == 0 {
+ // panic(&SyntaxError{
+ // Offset: start,
+ // What: errors.New("empty integer value"),
+ // })
+ // }
+ return nil
+}
+// called when 'i' was consumed, for the integer type in v.
+func (d *Decoder) parseInt(v reflect.Value) error {
+ start := d.Offset - 1
+
+ if err := d.readInt(); err != nil {
+ return err
+ }
s := bytesAsString(d.buf.Bytes())
switch v.Kind() {
checkForIntParseError(err, start)
if v.OverflowInt(n) {
- panic(&UnmarshalTypeError{
- Value: "integer " + s,
- Type: v.Type(),
- })
+ return &UnmarshalTypeError{
+ BencodeTypeName: "int",
+ UnmarshalTargetType: v.Type(),
+ }
}
v.SetInt(n)
case reflect.Uint, reflect.Uint8, reflect.Uint16, reflect.Uint32, reflect.Uint64:
checkForIntParseError(err, start)
if v.OverflowUint(n) {
- panic(&UnmarshalTypeError{
- Value: "integer " + s,
- Type: v.Type(),
- })
+ return &UnmarshalTypeError{
+ BencodeTypeName: "int",
+ UnmarshalTargetType: v.Type(),
+ }
}
v.SetUint(n)
case reflect.Bool:
v.SetBool(s != "0")
default:
- panic(&UnmarshalTypeError{
- Value: "integer " + s,
- Type: v.Type(),
- })
+ return &UnmarshalTypeError{
+ BencodeTypeName: "int",
+ UnmarshalTargetType: v.Type(),
+ }
}
d.buf.Reset()
+ return nil
}
-func (d *Decoder) parseString(v reflect.Value) error {
- start := d.Offset - 1
+func (d *Decoder) checkBufferedInt() error {
+ b := d.buf.Bytes()
+ if len(b) <= 1 {
+ return nil
+ }
+ if b[0] == '-' {
+ b = b[1:]
+ }
+ if b[0] < '1' || b[0] > '9' {
+ return errors.New("invalid leading digit")
+ }
+ return nil
+}
- // read the string length first
+func (d *Decoder) parseStringLength() (int, error) {
+ // We should have already consumed the first byte of the length into the Decoder buf.
+ start := d.Offset - 1
d.readUntil(':')
+ if err := d.checkBufferedInt(); err != nil {
+ return 0, err
+ }
+ // Really the limit should be the uint size for the platform. But we can't pass in an allocator,
+ // or limit total memory use in Go, the best we might hope to do is limit the size of a single
+ // decoded value (by reading it in in-place and then operating on a view).
length, err := strconv.ParseInt(bytesAsString(d.buf.Bytes()), 10, 0)
checkForIntParseError(err, start)
+ if int64(length) > d.getMaxStrLen() {
+ err = fmt.Errorf("parsed string length %v exceeds limit (%v)", length, DefaultDecodeMaxStrLen)
+ }
+ d.buf.Reset()
+ return int(length), err
+}
+func (d *Decoder) parseString(v reflect.Value) error {
+ length, err := d.parseStringLength()
+ if err != nil {
+ return err
+ }
defer d.buf.Reset()
-
read := func(b []byte) {
n, err := io.ReadFull(d.r, b)
d.Offset += int64(n)
if v.Type().Elem().Kind() != reflect.Uint8 {
break
}
- d.buf.Grow(int(length))
+ d.buf.Grow(length)
b := d.buf.Bytes()[:length]
read(b)
reflect.Copy(v, reflect.ValueOf(b))
return nil
+ case reflect.Bool:
+ d.buf.Grow(length)
+ b := d.buf.Bytes()[:length]
+ read(b)
+ x, err := strconv.ParseBool(bytesAsString(b))
+ if err != nil {
+ x = length != 0
+ }
+ v.SetBool(x)
+ return nil
}
- d.buf.Grow(int(length))
+ // Can't move this into default clause because some cases above fail through to here after
+ // additional checks.
+ d.buf.Grow(length)
read(d.buf.Bytes()[:length])
// I believe we return here to support "ignore_unmarshal_type_error".
return &UnmarshalTypeError{
- Value: "string",
- Type: v.Type(),
+ BencodeTypeName: "string",
+ UnmarshalTargetType: v.Type(),
}
}
// Info for parsing a dict value.
type dictField struct {
- Value reflect.Value // Storage for the parsed value.
- // True if field value should be parsed into Value. If false, the value
- // should be parsed and discarded.
- Ok bool
- Set func() // Call this after parsing into Value.
- IgnoreUnmarshalTypeError bool
+ Type reflect.Type
+ Get func(value reflect.Value) func(reflect.Value)
+ Tags tag
}
// Returns specifics for parsing a dict field value.
-func getDictField(dict reflect.Value, key string) dictField {
+func getDictField(dict reflect.Type, key string) (_ dictField, err error) {
// get valuev as a map value or as a struct field
- switch dict.Kind() {
+ switch k := dict.Kind(); k {
case reflect.Map:
- value := reflect.New(dict.Type().Elem()).Elem()
return dictField{
- Value: value,
- Ok: true,
- Set: func() {
- if dict.IsNil() {
- dict.Set(reflect.MakeMap(dict.Type()))
+ Type: dict.Elem(),
+ Get: func(mapValue reflect.Value) func(reflect.Value) {
+ return func(value reflect.Value) {
+ if mapValue.IsNil() {
+ mapValue.Set(reflect.MakeMap(dict))
+ }
+ // Assigns the value into the map.
+ // log.Printf("map type: %v", mapValue.Type())
+ mapValue.SetMapIndex(reflect.ValueOf(key).Convert(dict.Key()), value)
}
- // Assigns the value into the map.
- dict.SetMapIndex(reflect.ValueOf(key).Convert(dict.Type().Key()), value)
},
- }
+ }, nil
case reflect.Struct:
- sf, ok := getStructFieldForKey(dict.Type(), key)
- if !ok {
- return dictField{}
- }
- if sf.r.PkgPath != "" {
- panic(&UnmarshalFieldError{
- Key: key,
- Type: dict.Type(),
- Field: sf.r,
- })
- }
- return dictField{
- Value: dict.FieldByIndex(sf.r.Index),
- Ok: true,
- Set: func() {},
- IgnoreUnmarshalTypeError: sf.tag.IgnoreUnmarshalTypeError(),
- }
+ return getStructFieldForKey(dict, key), nil
+ // if sf.r.PkgPath != "" {
+ // panic(&UnmarshalFieldError{
+ // Key: key,
+ // Type: dict.Type(),
+ // Field: sf.r,
+ // })
+ // }
default:
- return dictField{}
+ err = fmt.Errorf("can't assign bencode dict items into a %v", k)
+ return
}
}
-type structField struct {
- r reflect.StructField
- tag tag
-}
-
var (
structFieldsMu sync.Mutex
- structFields = map[reflect.Type]map[string]structField{}
+ structFields = map[reflect.Type]map[string]dictField{}
)
-func parseStructFields(struct_ reflect.Type, each func(string, structField)) {
- for i, n := 0, struct_.NumField(); i < n; i++ {
+func parseStructFields(struct_ reflect.Type, each func(key string, df dictField)) {
+ for _i, n := 0, struct_.NumField(); _i < n; _i++ {
+ i := _i
f := struct_.Field(i)
if f.Anonymous {
+ t := f.Type
+ if t.Kind() == reflect.Ptr {
+ t = t.Elem()
+ }
+ parseStructFields(t, func(key string, df dictField) {
+ innerGet := df.Get
+ df.Get = func(value reflect.Value) func(reflect.Value) {
+ anonPtr := value.Field(i)
+ if anonPtr.Kind() == reflect.Ptr && anonPtr.IsNil() {
+ anonPtr.Set(reflect.New(f.Type.Elem()))
+ anonPtr = anonPtr.Elem()
+ }
+ return innerGet(anonPtr)
+ }
+ each(key, df)
+ })
continue
}
tagStr := f.Tag.Get("bencode")
if key == "" {
key = f.Name
}
- each(key, structField{f, tag})
+ each(key, dictField{f.Type, func(value reflect.Value) func(reflect.Value) {
+ return value.Field(i).Set
+ }, tag})
}
}
func saveStructFields(struct_ reflect.Type) {
- m := make(map[string]structField)
- parseStructFields(struct_, func(key string, sf structField) {
+ m := make(map[string]dictField)
+ parseStructFields(struct_, func(key string, sf dictField) {
m[key] = sf
})
structFields[struct_] = m
}
-func getStructFieldForKey(struct_ reflect.Type, key string) (f structField, ok bool) {
+func getStructFieldForKey(struct_ reflect.Type, key string) (f dictField) {
structFieldsMu.Lock()
if _, ok := structFields[struct_]; !ok {
saveStructFields(struct_)
}
- f, ok = structFields[struct_][key]
+ f, ok := structFields[struct_][key]
structFieldsMu.Unlock()
+ if !ok {
+ var discard interface{}
+ return dictField{
+ Type: reflect.TypeOf(discard),
+ Get: func(reflect.Value) func(reflect.Value) { return func(reflect.Value) {} },
+ Tags: nil,
+ }
+ }
return
}
func (d *Decoder) parseDict(v reflect.Value) error {
- // so, at this point 'd' byte was consumed, let's just read key/value
- // pairs one by one
+ // At this point 'd' byte was consumed, now read key/value pairs
for {
var keyStr string
keyValue := reflect.ValueOf(&keyStr).Elem()
ok, err := d.parseValue(keyValue)
if err != nil {
- return fmt.Errorf("error parsing dict key: %s", err)
+ return fmt.Errorf("error parsing dict key: %w", err)
}
if !ok {
return nil
}
- df := getDictField(v, keyStr)
+ df, err := getDictField(v.Type(), keyStr)
+ if err != nil {
+ return fmt.Errorf("parsing bencode dict into %v: %w", v.Type(), err)
+ }
// now we need to actually parse it
- if df.Ok {
- // log.Printf("parsing ok struct field for key %q", keyStr)
- ok, err = d.parseValue(df.Value)
- } else {
+ if df.Type == nil {
// Discard the value, there's nowhere to put it.
var if_ interface{}
if_, ok = d.parseValueInterface()
if if_ == nil {
- err = fmt.Errorf("error parsing value for key %q", keyStr)
+ return fmt.Errorf("error parsing value for key %q", keyStr)
}
+ if !ok {
+ return fmt.Errorf("missing value for key %q", keyStr)
+ }
+ continue
}
+ setValue := reflect.New(df.Type).Elem()
+ // log.Printf("parsing into %v", setValue.Type())
+ ok, err = d.parseValue(setValue)
if err != nil {
- if _, ok := err.(*UnmarshalTypeError); !ok || !df.IgnoreUnmarshalTypeError {
- return fmt.Errorf("parsing value for key %q: %s", keyStr, err)
+ var target *UnmarshalTypeError
+ if !(errors.As(err, &target) && df.Tags.IgnoreUnmarshalTypeError()) {
+ return fmt.Errorf("parsing value for key %q: %w", keyStr, err)
}
}
if !ok {
return fmt.Errorf("missing value for key %q", keyStr)
}
- if df.Ok {
- df.Set()
- }
+ df.Get(v)(setValue)
}
}
func (d *Decoder) parseList(v reflect.Value) error {
switch v.Kind() {
- case reflect.Array, reflect.Slice:
default:
- return &UnmarshalTypeError{
- Value: "list",
- Type: v.Type(),
+ // If the list is a singleton of the expected type, use that value. See
+ // https://github.com/anacrolix/torrent/issues/297.
+ l := reflect.New(reflect.SliceOf(v.Type()))
+ if err := d.parseList(l.Elem()); err != nil {
+ return err
+ }
+ if l.Elem().Len() != 1 {
+ return &UnmarshalTypeError{
+ BencodeTypeName: "list",
+ UnmarshalTargetType: v.Type(),
+ }
}
+ v.Set(l.Elem().Index(0))
+ return nil
+ case reflect.Array, reflect.Slice:
+ // We can work with this. Normal case, fallthrough.
}
i := 0
}
return true
-
}
func (d *Decoder) parseUnmarshaler(v reflect.Value) bool {
b, err := d.r.ReadByte()
if err != nil {
- panic(err)
+ return false, err
}
d.Offset++
case 'l':
return true, d.parseList(v)
case 'i':
- d.parseInt(v)
- return true, nil
+ return true, d.parseInt(v)
default:
if b >= '0' && b <= '9' {
// It's a string.
}
}
+// Called after 'i', for an arbitrary integer size.
func (d *Decoder) parseIntInterface() (ret interface{}) {
start := d.Offset - 1
- d.readUntil('e')
- if d.buf.Len() == 0 {
- panic(&SyntaxError{
- Offset: start,
- What: errors.New("empty integer value"),
- })
- }
+ if err := d.readInt(); err != nil {
+ panic(err)
+ }
n, err := strconv.ParseInt(d.buf.String(), 10, 64)
if ne, ok := err.(*strconv.NumError); ok && ne.Err == strconv.ErrRange {
i := new(big.Int)
return
}
-func (d *Decoder) parseStringInterface() interface{} {
- start := d.Offset - 1
-
- // read the string length first
- d.readUntil(':')
- length, err := strconv.ParseInt(d.buf.String(), 10, 64)
- checkForIntParseError(err, start)
-
- d.buf.Reset()
- n, err := io.CopyN(&d.buf, d.r, length)
- d.Offset += n
+func (d *Decoder) readBytes(length int) []byte {
+ b, err := io.ReadAll(io.LimitReader(d.r, int64(length)))
if err != nil {
- checkForUnexpectedEOF(err, d.Offset)
- panic(&SyntaxError{
- Offset: d.Offset,
- What: errors.New("unexpected I/O error: " + err.Error()),
- })
+ panic(err)
}
+ if len(b) != length {
+ panic(fmt.Errorf("read %v bytes expected %v", len(b), length))
+ }
+ return b
+}
- s := d.buf.String()
- d.buf.Reset()
- return s
+func (d *Decoder) parseStringInterface() string {
+ length, err := d.parseStringLength()
+ if err != nil {
+ panic(err)
+ }
+ b := d.readBytes(int(length))
+ d.Offset += int64(len(b))
+ if err != nil {
+ panic(&SyntaxError{Offset: d.Offset, What: err})
+ }
+ return bytesAsString(b)
}
func (d *Decoder) parseDictInterface() interface{} {
dict := make(map[string]interface{})
+ var lastKey string
+ lastKeyOk := false
for {
+ start := d.Offset
keyi, ok := d.parseValueInterface()
if !ok {
break
What: errors.New("non-string key in a dict"),
})
}
-
+ if lastKeyOk && key <= lastKey {
+ d.throwSyntaxError(start, fmt.Errorf("dict keys unsorted: %q <= %q", key, lastKey))
+ }
+ start = d.Offset
valuei, ok := d.parseValueInterface()
if !ok {
- break
+ d.throwSyntaxError(start, fmt.Errorf("dict elem missing value [key=%v]", key))
}
+ lastKey = key
+ lastKeyOk = true
dict[key] = valuei
}
return dict
}
-func (d *Decoder) parseListInterface() interface{} {
- var list []interface{}
- for {
- valuei, ok := d.parseValueInterface()
- if !ok {
- break
- }
-
+func (d *Decoder) parseListInterface() (list []interface{}) {
+ list = []interface{}{}
+ valuei, ok := d.parseValueInterface()
+ for ok {
list = append(list, valuei)
+ valuei, ok = d.parseValueInterface()
}
- if list == nil {
- list = make([]interface{}, 0, 0)
+ return
+}
+
+func (d *Decoder) getMaxStrLen() int64 {
+ if d.MaxStrLen == 0 {
+ return DefaultDecodeMaxStrLen
}
- return list
+ return d.MaxStrLen
}