mirror of
https://github.com/lightningnetwork/lnd.git
synced 2024-11-19 01:43:16 +01:00
338 lines
10 KiB
Go
338 lines
10 KiB
Go
package tlv
|
|
|
|
import (
|
|
"bytes"
|
|
"errors"
|
|
"io"
|
|
"io/ioutil"
|
|
"math"
|
|
)
|
|
|
|
// MaxRecordSize is the maximum size of a particular record that will be parsed
|
|
// by a stream decoder. This value is currently chosen to the be equal to the
|
|
// maximum message size permitted by BOLT 1, as no record should be bigger than
|
|
// an entire message.
|
|
const MaxRecordSize = 65535 // 65KB
|
|
|
|
// ErrStreamNotCanonical signals that a decoded stream does not contain records
|
|
// sorting by monotonically-increasing type.
|
|
var ErrStreamNotCanonical = errors.New("tlv stream is not canonical")
|
|
|
|
// ErrRecordTooLarge signals that a decoded record has a length that is too
|
|
// long to parse.
|
|
var ErrRecordTooLarge = errors.New("record is too large")
|
|
|
|
// Stream defines a TLV stream that can be used for encoding or decoding a set
|
|
// of TLV Records.
|
|
type Stream struct {
|
|
records []Record
|
|
buf [8]byte
|
|
}
|
|
|
|
// NewStream creates a new TLV Stream given an encoding codec, a decoding codec,
|
|
// and a set of known records.
|
|
func NewStream(records ...Record) (*Stream, error) {
|
|
// Assert that the ordering of the Records is canonical and appear in
|
|
// ascending order of type.
|
|
var (
|
|
min Type
|
|
overflow bool
|
|
)
|
|
for _, record := range records {
|
|
if overflow || record.typ < min {
|
|
return nil, ErrStreamNotCanonical
|
|
}
|
|
if record.encoder == nil {
|
|
record.encoder = ENOP
|
|
}
|
|
if record.decoder == nil {
|
|
record.decoder = DNOP
|
|
}
|
|
if record.typ == math.MaxUint64 {
|
|
overflow = true
|
|
}
|
|
min = record.typ + 1
|
|
}
|
|
|
|
return &Stream{
|
|
records: records,
|
|
}, nil
|
|
}
|
|
|
|
// MustNewStream creates a new TLV Stream given an encoding codec, a decoding
|
|
// codec, and a set of known records. If an error is encountered in creating the
|
|
// stream, this method will panic instead of returning the error.
|
|
func MustNewStream(records ...Record) *Stream {
|
|
stream, err := NewStream(records...)
|
|
if err != nil {
|
|
panic(err.Error())
|
|
}
|
|
return stream
|
|
}
|
|
|
|
// Encode writes a Stream to the passed io.Writer. Each of the Records known to
|
|
// the Stream is written in ascending order of their type so as to be canonical.
|
|
//
|
|
// The stream is constructed by concatenating the individual, serialized Records
|
|
// where each record has the following format:
|
|
//
|
|
// [varint: type]
|
|
// [varint: length]
|
|
// [length: value]
|
|
//
|
|
// An error is returned if the io.Writer fails to accept bytes from the
|
|
// encoding, and nothing else. The ordering of the Records is asserted upon the
|
|
// creation of a Stream, and thus the output will be by definition canonical.
|
|
func (s *Stream) Encode(w io.Writer) error {
|
|
// Iterate through all known records, if any, serializing each record's
|
|
// type, length and value.
|
|
for i := range s.records {
|
|
rec := &s.records[i]
|
|
|
|
// Write the record's type as a varint.
|
|
err := WriteVarInt(w, uint64(rec.typ), &s.buf)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
|
|
// Write the record's length as a varint.
|
|
err = WriteVarInt(w, rec.Size(), &s.buf)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
|
|
// Encode the current record's value using the stream's codec.
|
|
err = rec.encoder(w, rec.value, &s.buf)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
}
|
|
|
|
return nil
|
|
}
|
|
|
|
// Decode deserializes TLV Stream from the passed io.Reader for non-P2P
|
|
// settings. The Stream will inspect each record that is parsed and check to
|
|
// see if it has a corresponding Record to facilitate deserialization of that
|
|
// field. If the record is unknown, the Stream will discard the record's bytes
|
|
// and proceed to the subsequent record.
|
|
//
|
|
// Each record has the following format:
|
|
//
|
|
// [varint: type]
|
|
// [varint: length]
|
|
// [length: value]
|
|
//
|
|
// A series of (possibly zero) records are concatenated into a stream, this
|
|
// example contains two records:
|
|
//
|
|
// (t: 0x01, l: 0x04, v: 0xff, 0xff, 0xff, 0xff)
|
|
// (t: 0x02, l: 0x01, v: 0x01)
|
|
//
|
|
// This method asserts that the byte stream is canonical, namely that each
|
|
// record is unique and that all records are sorted in ascending order. An
|
|
// ErrNotCanonicalStream error is returned if the encoded TLV stream is not.
|
|
//
|
|
// We permit an io.EOF error only when reading the type byte which signals that
|
|
// the last record was read cleanly and we should stop parsing. All other io.EOF
|
|
// or io.ErrUnexpectedEOF errors are returned.
|
|
func (s *Stream) Decode(r io.Reader) error {
|
|
_, err := s.decode(r, nil, false)
|
|
return err
|
|
}
|
|
|
|
// DecodeP2P is identical to Decode except that the maximum record size is
|
|
// capped at 65535.
|
|
func (s *Stream) DecodeP2P(r io.Reader) error {
|
|
_, err := s.decode(r, nil, true)
|
|
return err
|
|
}
|
|
|
|
// DecodeWithParsedTypes is identical to Decode, but if successful, returns a
|
|
// TypeMap containing the types of all records that were decoded or ignored from
|
|
// the stream.
|
|
func (s *Stream) DecodeWithParsedTypes(r io.Reader) (TypeMap, error) {
|
|
return s.decode(r, make(TypeMap), false)
|
|
}
|
|
|
|
// DecodeWithParsedTypesP2P is identical to DecodeWithParsedTypes except that
|
|
// the record size is capped at 65535. This should only be called from a p2p
|
|
// setting where untrusted input is being deserialized.
|
|
func (s *Stream) DecodeWithParsedTypesP2P(r io.Reader) (TypeMap, error) {
|
|
return s.decode(r, make(TypeMap), true)
|
|
}
|
|
|
|
// decode is a helper function that performs the basis of stream decoding. If
|
|
// the caller needs the set of parsed types, it must provide an initialized
|
|
// parsedTypes, otherwise the returned TypeMap will be nil. If the p2p bool is
|
|
// true, then the record size is capped at 65535. Otherwise, it is not capped.
|
|
func (s *Stream) decode(r io.Reader, parsedTypes TypeMap, p2p bool) (TypeMap,
|
|
error) {
|
|
|
|
var (
|
|
typ Type
|
|
min Type
|
|
recordIdx int
|
|
overflow bool
|
|
)
|
|
|
|
// Iterate through all possible type identifiers. As types are read from
|
|
// the io.Reader, min will skip forward to the last read type.
|
|
for {
|
|
// Read the next varint type.
|
|
t, err := ReadVarInt(r, &s.buf)
|
|
switch {
|
|
|
|
// We'll silence an EOF when zero bytes remain, meaning the
|
|
// stream was cleanly encoded.
|
|
case err == io.EOF:
|
|
return parsedTypes, nil
|
|
|
|
// Other unexpected errors.
|
|
case err != nil:
|
|
return nil, err
|
|
}
|
|
|
|
typ = Type(t)
|
|
|
|
// Assert that this type is greater than any previously read.
|
|
// If we've already overflowed and we parsed another type, the
|
|
// stream is not canonical. This check prevents us from accepts
|
|
// encodings that have duplicate records or from accepting an
|
|
// unsorted series.
|
|
if overflow || typ < min {
|
|
return nil, ErrStreamNotCanonical
|
|
}
|
|
|
|
// Read the varint length.
|
|
length, err := ReadVarInt(r, &s.buf)
|
|
switch {
|
|
|
|
// We'll convert any EOFs to ErrUnexpectedEOF, since this
|
|
// results in an invalid record.
|
|
case err == io.EOF:
|
|
return nil, io.ErrUnexpectedEOF
|
|
|
|
// Other unexpected errors.
|
|
case err != nil:
|
|
return nil, err
|
|
}
|
|
|
|
// Place a soft limit on the size of a sane record, which
|
|
// prevents malicious encoders from causing us to allocate an
|
|
// unbounded amount of memory when decoding variable-sized
|
|
// fields. This is only checked when the p2p bool is true.
|
|
if p2p && length > MaxRecordSize {
|
|
return nil, ErrRecordTooLarge
|
|
}
|
|
|
|
// Search the records known to the stream for this type. We'll
|
|
// begin the search and recordIdx and walk forward until we find
|
|
// it or the next record's type is larger.
|
|
rec, newIdx, ok := s.getRecord(typ, recordIdx)
|
|
switch {
|
|
|
|
// We know of this record type, proceed to decode the value.
|
|
// This method asserts that length bytes are read in the
|
|
// process, and returns an error if the number of bytes is not
|
|
// exactly length.
|
|
case ok:
|
|
err := rec.decoder(r, rec.value, &s.buf, length)
|
|
switch {
|
|
|
|
// We'll convert any EOFs to ErrUnexpectedEOF, since this
|
|
// results in an invalid record.
|
|
case err == io.EOF:
|
|
return nil, io.ErrUnexpectedEOF
|
|
|
|
// Other unexpected errors.
|
|
case err != nil:
|
|
return nil, err
|
|
}
|
|
|
|
// Record the successfully decoded type if the caller
|
|
// provided an initialized TypeMap.
|
|
if parsedTypes != nil {
|
|
parsedTypes[typ] = nil
|
|
}
|
|
|
|
// Otherwise, the record type is unknown and is odd, discard the
|
|
// number of bytes specified by length.
|
|
default:
|
|
// If the caller provided an initialized TypeMap, record
|
|
// the encoded bytes.
|
|
var b *bytes.Buffer
|
|
writer := ioutil.Discard
|
|
if parsedTypes != nil {
|
|
b = bytes.NewBuffer(make([]byte, 0, length))
|
|
writer = b
|
|
}
|
|
|
|
_, err := io.CopyN(writer, r, int64(length))
|
|
switch {
|
|
|
|
// We'll convert any EOFs to ErrUnexpectedEOF, since this
|
|
// results in an invalid record.
|
|
case err == io.EOF:
|
|
return nil, io.ErrUnexpectedEOF
|
|
|
|
// Other unexpected errors.
|
|
case err != nil:
|
|
return nil, err
|
|
}
|
|
|
|
if parsedTypes != nil {
|
|
parsedTypes[typ] = b.Bytes()
|
|
}
|
|
}
|
|
|
|
// Update our record index so that we can begin our next search
|
|
// from where we left off.
|
|
recordIdx = newIdx
|
|
|
|
// If we've parsed the largest possible type, the next loop will
|
|
// overflow back to zero. However, we need to attempt parsing
|
|
// the next type to ensure that the stream is empty.
|
|
if typ == math.MaxUint64 {
|
|
overflow = true
|
|
}
|
|
|
|
// Finally, set our lower bound on the next accepted type.
|
|
min = typ + 1
|
|
}
|
|
}
|
|
|
|
// getRecord searches for a record matching typ known to the stream. The boolean
|
|
// return value indicates whether the record is known to the stream. The integer
|
|
// return value carries the index from where getRecord should be invoked on the
|
|
// subsequent call. The first call to getRecord should always use an idx of 0.
|
|
func (s *Stream) getRecord(typ Type, idx int) (Record, int, bool) {
|
|
for idx < len(s.records) {
|
|
record := s.records[idx]
|
|
switch {
|
|
|
|
// Found target record, return it to the caller. The next index
|
|
// returned points to the immediately following record.
|
|
case record.typ == typ:
|
|
return record, idx + 1, true
|
|
|
|
// This record's type is lower than the target. Advance our
|
|
// index and continue to the next record which will have a
|
|
// strictly higher type.
|
|
case record.typ < typ:
|
|
idx++
|
|
continue
|
|
|
|
// This record's type is larger than the target, hence we have
|
|
// no record matching the current type. Return the current index
|
|
// so that we can start our search from here when processing the
|
|
// next tlv record.
|
|
default:
|
|
return Record{}, idx, false
|
|
}
|
|
}
|
|
|
|
// All known records are exhausted.
|
|
return Record{}, idx, false
|
|
}
|