niko_trust/pkg/tce/decoder.go

package tce

import (
	"unicode/utf8"
)

// Decoder reads canonical TCE bytes.
//
// The decoder is strict: it accepts only the exact byte sequence the encoder
// would produce for a given object, and treats every other input as an error.
// In particular it never skips a field it does not understand. Skipping would
// mean two implementations computed different meanings for the same signed
// bytes while both saw a valid signature, so an old verifier could approve an
// object whose actual content it never examined.
//
// Every length prefix is checked against the remaining input and against the
// field's maximum before any allocation, so a hostile length cannot cause a
// large allocation or a long loop.
type Decoder struct {
	buf []byte
	off int
}

// NewDecoder returns a decoder reading b. The slice is not copied; the caller
// must not modify it while decoding.
func NewDecoder(b []byte) *Decoder { return &Decoder{buf: b} }

// Offset returns the current read position.
func (d *Decoder) Offset() int { return d.off }

// Remaining returns the number of unread bytes.
func (d *Decoder) Remaining() int { return len(d.buf) - d.off }

// End asserts that the input is fully consumed.
//
// Trailing bytes are an error rather than ignored data: an object with extra
// bytes appended has a different byte string, and therefore a different object
// ID and a different signature, from the object it appears to contain.
func (d *Decoder) End() error {
	if d.off != len(d.buf) {
		return ErrTrailing
	}
	return nil
}

// Header reads and validates the magic, object tag and object version.
func (d *Decoder) Header() (ObjectTag, error) {
	if d.Remaining() < MagicLen {
		return TagReserved, ErrTruncated
	}
	if string(d.buf[d.off:d.off+MagicLen]) != Magic {
		return TagReserved, ErrMagic
	}
	d.off += MagicLen

	if d.Remaining() < 1 {
		return TagReserved, ErrTruncated
	}
	tag := ObjectTag(d.buf[d.off])
	d.off++
	if !knownTag(tag) {
		return TagReserved, ErrObjectTag
	}

	ver, err := d.Uvarint()
	if err != nil {
		return TagReserved, fieldErr("version", err)
	}
	if ver != Version {
		return TagReserved, ErrVersion
	}
	return tag, nil
}

// Uvarint reads a canonical LEB128 unsigned varint.
//
// A multi-byte encoding whose final group is zero is a longer spelling of a
// shorter value and is rejected, because permitting it would give the same
// number two encodings and therefore the same object two byte strings.
func (d *Decoder) Uvarint() (uint64, error) {
	var n uint64
	var shift uint
	start := d.off
	for {
		if d.off >= len(d.buf) {
			return 0, ErrTruncated
		}
		if d.off-start >= MaxUvarintBytes {
			return 0, ErrUvarint
		}
		b := d.buf[d.off]
		d.off++

		if shift >= 64 || (shift == 63 && b > 1) {
			return 0, ErrOverflow
		}
		n |= uint64(b&0x7f) << shift

		if b&0x80 == 0 {
			if d.off-start > 1 && b == 0x00 {
				return 0, ErrNonMinimal
			}
			return n, nil
		}
		shift += 7
	}
}

// RawBytes reads a length-prefixed byte string, enforcing maxLen.
//
// The returned slice aliases the decoder's buffer. Callers that retain the
// data must copy it; the object constructors in the protocol package do.
func (d *Decoder) RawBytes(maxLen int) ([]byte, error) {
	n, err := d.Uvarint()
	if err != nil {
		return nil, err
	}
	// Compare against the remaining input first so that an absurd length is
	// rejected before it is compared with anything else.
	if n > uint64(d.Remaining()) {
		return nil, ErrTruncated
	}
	if n > uint64(maxLen) {
		return nil, ErrTooLong
	}
	b := d.buf[d.off : d.off+int(n)]
	d.off += int(n)
	return b, nil
}

// FixedBytes reads a length-prefixed byte string of exactly want bytes.
func (d *Decoder) FixedBytes(want int) ([]byte, error) {
	b, err := d.RawBytes(want)
	if err != nil {
		return nil, err
	}
	if len(b) != want {
		return nil, ErrFieldSize
	}
	return b, nil
}

// String reads a length-prefixed UTF-8 string and validates it.
func (d *Decoder) String(maxLen int) (string, error) {
	b, err := d.RawBytes(maxLen)
	if err != nil {
		return "", err
	}
	if !utf8.Valid(b) {
		return "", ErrUTF8
	}
	s := string(b)
	for _, r := range s {
		if r <= 0x1f || r == 0x7f || (r >= 0x80 && r <= 0x9f) {
			return "", ErrControlChar
		}
	}
	return s, nil
}

// Identity reads an identity field and returns a copy of the public key.
//
// The key is copied so that a decoded object does not alias the input buffer,
// which means a caller cannot alter an identity after the object containing it
// has been verified.
func (d *Decoder) Identity() ([]byte, error) {
	ver, err := d.Uvarint()
	if err != nil {
		return nil, err
	}
	if ver != AddressVersion {
		return nil, ErrAddressVersion
	}
	b, err := d.FixedBytes(PubKeySize)
	if err != nil {
		return nil, err
	}
	out := make([]byte, PubKeySize)
	copy(out, b)
	return out, nil
}

// Timestamp reads a timestamp and validates its range.
func (d *Decoder) Timestamp(allowZero bool) (uint64, error) {
	ts, err := d.Uvarint()
	if err != nil {
		return 0, err
	}
	if err := ValidateTimestamp(ts, allowZero); err != nil {
		return 0, err
	}
	return ts, nil
}

// Value reads a typed value.
//
// Reserved and unknown tags are rejected. This is the opposite of the
// "ignore what you do not understand" rule common in extensible formats, and
// it is deliberate: in a signed protocol, skipping an unrecognised value means
// the verifier's view of the object differs from the signer's.
func (d *Decoder) Value() (Value, error) {
	if d.Remaining() < 1 {
		return Value{}, ErrTruncated
	}
	tag := ValueTag(d.buf[d.off])
	d.off++

	switch tag {
	case ValNull:
		return Null(), nil
	case ValFalse:
		return Bool(false), nil
	case ValTrue:
		return Bool(true), nil
	case ValString:
		s, err := d.String(MaxStringValue)
		if err != nil {
			return Value{}, err
		}
		return String(s), nil
	case ValNumber:
		b, err := d.RawBytes(MaxNumberToken)
		if err != nil {
			return Value{}, err
		}
		tok := string(b)
		// A number must arrive in canonical form. Accepting "1.0" here would
		// mean two byte strings encoded the same value.
		if !IsCanonicalNumber(tok) {
			return Value{}, ErrNumberFormat
		}
		return Number(tok), nil
	default:
		return Value{}, ErrValueTag
	}
}

// Map reads a map, enforcing ascending key order, key uniqueness, the key
// grammar and the entry limit.
func (d *Decoder) Map(minEntries int) (map[string]Value, error) {
	n, err := d.Uvarint()
	if err != nil {
		return nil, err
	}
	if n > MaxMapEntries {
		return nil, ErrTooLong
	}
	if n < uint64(minEntries) {
		return nil, ErrEmptyMap
	}
	// Each entry costs at least two bytes, so a count larger than the
	// remaining input cannot be satisfied. Checking this before allocating
	// prevents a small input from reserving a large map.
	if n > uint64(d.Remaining()) {
		return nil, ErrTruncated
	}

	m := make(map[string]Value, n)
	prev := ""
	for i := uint64(0); i < n; i++ {
		kb, err := d.RawBytes(MaxKeyLen)
		if err != nil {
			return nil, err
		}
		key := string(kb)
		if err := ValidateKey(key); err != nil {
			return nil, err
		}
		if i > 0 {
			switch {
			case key == prev:
				return nil, ErrDuplicateKey
			case key < prev:
				// Out-of-order entries parse, but would give one map two
				// encodings, so they are rejected.
				return nil, ErrKeyOrder
			}
		}
		v, err := d.Value()
		if err != nil {
			return nil, err
		}
		m[key] = v
		prev = key
	}
	return m, nil
}