base32.go

     1  // Copyright 2011 The Go Authors. All rights reserved.
     2  // Use of this source code is governed by a BSD-style
     3  // license that can be found in the LICENSE file.
     4  
     5  // Package base32 implements base32 encoding as specified by RFC 4648.
     6  package base32
     7  
     8  import (
     9  	"io"
    10  	"slices"
    11  	"strconv"
    12  )
    13  
    14  /*
    15   * Encodings
    16   */
    17  
    18  // An Encoding is a radix 32 encoding/decoding scheme, defined by a
    19  // 32-character alphabet. The most common is the "base32" encoding
    20  // introduced for SASL GSSAPI and standardized in RFC 4648.
    21  // The alternate "base32hex" encoding is used in DNSSEC.
    22  type Encoding struct {
    23  	encode    [32]byte   // mapping of symbol index to symbol byte value
    24  	decodeMap [256]uint8 // mapping of symbol byte value to symbol index
    25  	padChar   rune
    26  }
    27  
    28  const (
    29  	StdPadding rune = '=' // Standard padding character
    30  	NoPadding  rune = -1  // No padding
    31  )
    32  
    33  const (
    34  	decodeMapInitialize = "" +
    35  		"\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff" +
    36  		"\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff" +
    37  		"\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff" +
    38  		"\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff" +
    39  		"\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff" +
    40  		"\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff" +
    41  		"\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff" +
    42  		"\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff" +
    43  		"\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff" +
    44  		"\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff" +
    45  		"\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff" +
    46  		"\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff" +
    47  		"\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff" +
    48  		"\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff" +
    49  		"\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff" +
    50  		"\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff"
    51  	invalidIndex = '\xff'
    52  )
    53  
    54  // NewEncoding returns a new padded Encoding defined by the given alphabet,
    55  // which must be a 32-byte string that contains unique byte values and
    56  // does not contain the padding character or CR / LF ('\r', '\n').
    57  // The alphabet is treated as a sequence of byte values
    58  // without any special treatment for multi-byte UTF-8.
    59  // The resulting Encoding uses the default padding character ('='),
    60  // which may be changed or disabled via [Encoding.WithPadding].
    61  func NewEncoding(encoder string) *Encoding {
    62  	if len(encoder) != 32 {
    63  		panic("encoding alphabet is not 32-bytes long")
    64  	}
    65  
    66  	e := new(Encoding)
    67  	e.padChar = StdPadding
    68  	copy(e.encode[:], encoder)
    69  	copy(e.decodeMap[:], decodeMapInitialize)
    70  
    71  	for i := 0; i < len(encoder); i++ {
    72  		// Note: While we document that the alphabet cannot contain
    73  		// the padding character, we do not enforce it since we do not know
    74  		// if the caller intends to switch the padding from StdPadding later.
    75  		switch {
    76  		case encoder[i] == '\n' || encoder[i] == '\r':
    77  			panic("encoding alphabet contains newline character")
    78  		case e.decodeMap[encoder[i]] != invalidIndex:
    79  			panic("encoding alphabet includes duplicate symbols")
    80  		}
    81  		e.decodeMap[encoder[i]] = uint8(i)
    82  	}
    83  	return e
    84  }
    85  
    86  // StdEncoding is the standard base32 encoding, as defined in RFC 4648.
    87  var StdEncoding = NewEncoding("ABCDEFGHIJKLMNOPQRSTUVWXYZ234567")
    88  
    89  // HexEncoding is the “Extended Hex Alphabet” defined in RFC 4648.
    90  // It is typically used in DNS.
    91  var HexEncoding = NewEncoding("0123456789ABCDEFGHIJKLMNOPQRSTUV")
    92  
    93  // WithPadding creates a new encoding identical to enc except
    94  // with a specified padding character, or NoPadding to disable padding.
    95  // The padding character must not be '\r' or '\n',
    96  // must not be contained in the encoding's alphabet,
    97  // must not be negative, and must be a rune equal or below '\xff'.
    98  // Padding characters above '\x7f' are encoded as their exact byte value
    99  // rather than using the UTF-8 representation of the codepoint.
   100  func (enc Encoding) WithPadding(padding rune) *Encoding {
   101  	switch {
   102  	case padding < NoPadding || padding == '\r' || padding == '\n' || padding > 0xff:
   103  		panic("invalid padding")
   104  	case padding != NoPadding && enc.decodeMap[byte(padding)] != invalidIndex:
   105  		panic("padding contained in alphabet")
   106  	}
   107  	enc.padChar = padding
   108  	return &enc
   109  }
   110  
   111  /*
   112   * Encoder
   113   */
   114  
   115  // Encode encodes src using the encoding enc,
   116  // writing [Encoding.EncodedLen](len(src)) bytes to dst.
   117  //
   118  // The encoding pads the output to a multiple of 8 bytes,
   119  // so Encode is not appropriate for use on individual blocks
   120  // of a large data stream. Use [NewEncoder] instead.
   121  func (enc *Encoding) Encode(dst, src []byte) {
   122  	if len(src) == 0 {
   123  		return
   124  	}
   125  	// enc is a pointer receiver, so the use of enc.encode within the hot
   126  	// loop below means a nil check at every operation. Lift that nil check
   127  	// outside of the loop to speed up the encoder.
   128  	_ = enc.encode
   129  
   130  	di, si := 0, 0
   131  	n := (len(src) / 5) * 5
   132  	for si < n {
   133  		// Combining two 32 bit loads allows the same code to be used
   134  		// for 32 and 64 bit platforms.
   135  		hi := uint32(src[si+0])<<24 | uint32(src[si+1])<<16 | uint32(src[si+2])<<8 | uint32(src[si+3])
   136  		lo := hi<<8 | uint32(src[si+4])
   137  
   138  		dst[di+0] = enc.encode[(hi>>27)&0x1F]
   139  		dst[di+1] = enc.encode[(hi>>22)&0x1F]
   140  		dst[di+2] = enc.encode[(hi>>17)&0x1F]
   141  		dst[di+3] = enc.encode[(hi>>12)&0x1F]
   142  		dst[di+4] = enc.encode[(hi>>7)&0x1F]
   143  		dst[di+5] = enc.encode[(hi>>2)&0x1F]
   144  		dst[di+6] = enc.encode[(lo>>5)&0x1F]
   145  		dst[di+7] = enc.encode[(lo)&0x1F]
   146  
   147  		si += 5
   148  		di += 8
   149  	}
   150  
   151  	// Add the remaining small block
   152  	remain := len(src) - si
   153  	if remain == 0 {
   154  		return
   155  	}
   156  
   157  	// Encode the remaining bytes in reverse order.
   158  	val := uint32(0)
   159  	switch remain {
   160  	case 4:
   161  		val |= uint32(src[si+3])
   162  		dst[di+6] = enc.encode[val<<3&0x1F]
   163  		dst[di+5] = enc.encode[val>>2&0x1F]
   164  		fallthrough
   165  	case 3:
   166  		val |= uint32(src[si+2]) << 8
   167  		dst[di+4] = enc.encode[val>>7&0x1F]
   168  		fallthrough
   169  	case 2:
   170  		val |= uint32(src[si+1]) << 16
   171  		dst[di+3] = enc.encode[val>>12&0x1F]
   172  		dst[di+2] = enc.encode[val>>17&0x1F]
   173  		fallthrough
   174  	case 1:
   175  		val |= uint32(src[si+0]) << 24
   176  		dst[di+1] = enc.encode[val>>22&0x1F]
   177  		dst[di+0] = enc.encode[val>>27&0x1F]
   178  	}
   179  
   180  	// Pad the final quantum
   181  	if enc.padChar != NoPadding {
   182  		nPad := (remain * 8 / 5) + 1
   183  		for i := nPad; i < 8; i++ {
   184  			dst[di+i] = byte(enc.padChar)
   185  		}
   186  	}
   187  }
   188  
   189  // AppendEncode appends the base32 encoded src to dst
   190  // and returns the extended buffer.
   191  func (enc *Encoding) AppendEncode(dst, src []byte) []byte {
   192  	n := enc.EncodedLen(len(src))
   193  	dst = slices.Grow(dst, n)
   194  	enc.Encode(dst[len(dst):][:n], src)
   195  	return dst[:len(dst)+n]
   196  }
   197  
   198  // EncodeToString returns the base32 encoding of src.
   199  func (enc *Encoding) EncodeToString(src []byte) string {
   200  	buf := make([]byte, enc.EncodedLen(len(src)))
   201  	enc.Encode(buf, src)
   202  	return string(buf)
   203  }
   204  
   205  type encoder struct {
   206  	err  error
   207  	enc  *Encoding
   208  	w    io.Writer
   209  	buf  [5]byte    // buffered data waiting to be encoded
   210  	nbuf int        // number of bytes in buf
   211  	out  [1024]byte // output buffer
   212  }
   213  
   214  func (e *encoder) Write(p []byte) (n int, err error) {
   215  	if e.err != nil {
   216  		return 0, e.err
   217  	}
   218  
   219  	// Leading fringe.
   220  	if e.nbuf > 0 {
   221  		var i int
   222  		for i = 0; i < len(p) && e.nbuf < 5; i++ {
   223  			e.buf[e.nbuf] = p[i]
   224  			e.nbuf++
   225  		}
   226  		n += i
   227  		p = p[i:]
   228  		if e.nbuf < 5 {
   229  			return
   230  		}
   231  		e.enc.Encode(e.out[0:], e.buf[0:])
   232  		if _, e.err = e.w.Write(e.out[0:8]); e.err != nil {
   233  			return n, e.err
   234  		}
   235  		e.nbuf = 0
   236  	}
   237  
   238  	// Large interior chunks.
   239  	for len(p) >= 5 {
   240  		nn := len(e.out) / 8 * 5
   241  		if nn > len(p) {
   242  			nn = len(p)
   243  			nn -= nn % 5
   244  		}
   245  		e.enc.Encode(e.out[0:], p[0:nn])
   246  		if _, e.err = e.w.Write(e.out[0 : nn/5*8]); e.err != nil {
   247  			return n, e.err
   248  		}
   249  		n += nn
   250  		p = p[nn:]
   251  	}
   252  
   253  	// Trailing fringe.
   254  	copy(e.buf[:], p)
   255  	e.nbuf = len(p)
   256  	n += len(p)
   257  	return
   258  }
   259  
   260  // Close flushes any pending output from the encoder.
   261  // It is an error to call Write after calling Close.
   262  func (e *encoder) Close() error {
   263  	// If there's anything left in the buffer, flush it out
   264  	if e.err == nil && e.nbuf > 0 {
   265  		e.enc.Encode(e.out[0:], e.buf[0:e.nbuf])
   266  		encodedLen := e.enc.EncodedLen(e.nbuf)
   267  		e.nbuf = 0
   268  		_, e.err = e.w.Write(e.out[0:encodedLen])
   269  	}
   270  	return e.err
   271  }
   272  
   273  // NewEncoder returns a new base32 stream encoder. Data written to
   274  // the returned writer will be encoded using enc and then written to w.
   275  // Base32 encodings operate in 5-byte blocks; when finished
   276  // writing, the caller must Close the returned encoder to flush any
   277  // partially written blocks.
   278  func NewEncoder(enc *Encoding, w io.Writer) io.WriteCloser {
   279  	return &encoder{enc: enc, w: w}
   280  }
   281  
   282  // EncodedLen returns the length in bytes of the base32 encoding
   283  // of an input buffer of length n.
   284  func (enc *Encoding) EncodedLen(n int) int {
   285  	if enc.padChar == NoPadding {
   286  		return n/5*8 + (n%5*8+4)/5
   287  	}
   288  	return (n + 4) / 5 * 8
   289  }
   290  
   291  /*
   292   * Decoder
   293   */
   294  
   295  type CorruptInputError int64
   296  
   297  func (e CorruptInputError) Error() string {
   298  	return "illegal base32 data at input byte " + strconv.FormatInt(int64(e), 10)
   299  }
   300  
   301  // decode is like Decode but returns an additional 'end' value, which
   302  // indicates if end-of-message padding was encountered and thus any
   303  // additional data is an error. This method assumes that src has been
   304  // stripped of all supported whitespace ('\r' and '\n').
   305  func (enc *Encoding) decode(dst, src []byte) (n int, end bool, err error) {
   306  	// Lift the nil check outside of the loop.
   307  	_ = enc.decodeMap
   308  
   309  	dsti := 0
   310  	olen := len(src)
   311  
   312  	for len(src) > 0 && !end {
   313  		// Decode quantum using the base32 alphabet
   314  		var dbuf [8]byte
   315  		dlen := 8
   316  
   317  		for j := 0; j < 8; {
   318  
   319  			if len(src) == 0 {
   320  				if enc.padChar != NoPadding {
   321  					// We have reached the end and are missing padding
   322  					return n, false, CorruptInputError(olen - len(src) - j)
   323  				}
   324  				// We have reached the end and are not expecting any padding
   325  				dlen, end = j, true
   326  				break
   327  			}
   328  			in := src[0]
   329  			src = src[1:]
   330  			if in == byte(enc.padChar) && j >= 2 && len(src) < 8 {
   331  				// We've reached the end and there's padding
   332  				if len(src)+j < 8-1 {
   333  					// not enough padding
   334  					return n, false, CorruptInputError(olen)
   335  				}
   336  				for k := 0; k < 8-1-j; k++ {
   337  					if len(src) > k && src[k] != byte(enc.padChar) {
   338  						// incorrect padding
   339  						return n, false, CorruptInputError(olen - len(src) + k - 1)
   340  					}
   341  				}
   342  				dlen, end = j, true
   343  				// 7, 5 and 2 are not valid padding lengths, and so 1, 3 and 6 are not
   344  				// valid dlen values. See RFC 4648 Section 6 "Base 32 Encoding" listing
   345  				// the five valid padding lengths, and Section 9 "Illustrations and
   346  				// Examples" for an illustration for how the 1st, 3rd and 6th base32
   347  				// src bytes do not yield enough information to decode a dst byte.
   348  				if dlen == 1 || dlen == 3 || dlen == 6 {
   349  					return n, false, CorruptInputError(olen - len(src) - 1)
   350  				}
   351  				break
   352  			}
   353  			dbuf[j] = enc.decodeMap[in]
   354  			if dbuf[j] == 0xFF {
   355  				return n, false, CorruptInputError(olen - len(src) - 1)
   356  			}
   357  			j++
   358  		}
   359  
   360  		// Pack 8x 5-bit source blocks into 5 byte destination
   361  		// quantum
   362  		switch dlen {
   363  		case 8:
   364  			dst[dsti+4] = dbuf[6]<<5 | dbuf[7]
   365  			n++
   366  			fallthrough
   367  		case 7:
   368  			dst[dsti+3] = dbuf[4]<<7 | dbuf[5]<<2 | dbuf[6]>>3
   369  			n++
   370  			fallthrough
   371  		case 5:
   372  			dst[dsti+2] = dbuf[3]<<4 | dbuf[4]>>1
   373  			n++
   374  			fallthrough
   375  		case 4:
   376  			dst[dsti+1] = dbuf[1]<<6 | dbuf[2]<<1 | dbuf[3]>>4
   377  			n++
   378  			fallthrough
   379  		case 2:
   380  			dst[dsti+0] = dbuf[0]<<3 | dbuf[1]>>2
   381  			n++
   382  		}
   383  		dsti += 5
   384  	}
   385  	return n, end, nil
   386  }
   387  
   388  // Decode decodes src using the encoding enc. It writes at most
   389  // [Encoding.DecodedLen](len(src)) bytes to dst and returns the number of bytes
   390  // written. If src contains invalid base32 data, it will return the
   391  // number of bytes successfully written and [CorruptInputError].
   392  // Newline characters (\r and \n) are ignored.
   393  func (enc *Encoding) Decode(dst, src []byte) (n int, err error) {
   394  	buf := make([]byte, len(src))
   395  	l := stripNewlines(buf, src)
   396  	n, _, err = enc.decode(dst, buf[:l])
   397  	return
   398  }
   399  
   400  // AppendDecode appends the base32 decoded src to dst
   401  // and returns the extended buffer.
   402  // If the input is malformed, it returns the partially decoded src and an error.
   403  func (enc *Encoding) AppendDecode(dst, src []byte) ([]byte, error) {
   404  	// Compute the output size without padding to avoid over allocating.
   405  	n := len(src)
   406  	for n > 0 && rune(src[n-1]) == enc.padChar {
   407  		n--
   408  	}
   409  	n = decodedLen(n, NoPadding)
   410  
   411  	dst = slices.Grow(dst, n)
   412  	n, err := enc.Decode(dst[len(dst):][:n], src)
   413  	return dst[:len(dst)+n], err
   414  }
   415  
   416  // DecodeString returns the bytes represented by the base32 string s.
   417  func (enc *Encoding) DecodeString(s string) ([]byte, error) {
   418  	buf := []byte(s)
   419  	l := stripNewlines(buf, buf)
   420  	n, _, err := enc.decode(buf, buf[:l])
   421  	return buf[:n], err
   422  }
   423  
   424  type decoder struct {
   425  	err    error
   426  	enc    *Encoding
   427  	r      io.Reader
   428  	end    bool       // saw end of message
   429  	buf    [1024]byte // leftover input
   430  	nbuf   int
   431  	out    []byte // leftover decoded output
   432  	outbuf [1024 / 8 * 5]byte
   433  }
   434  
   435  func readEncodedData(r io.Reader, buf []byte, min int, expectsPadding bool) (n int, err error) {
   436  	for n < min && err == nil {
   437  		var nn int
   438  		nn, err = r.Read(buf[n:])
   439  		n += nn
   440  	}
   441  	// data was read, less than min bytes could be read
   442  	if n < min && n > 0 && err == io.EOF {
   443  		err = io.ErrUnexpectedEOF
   444  	}
   445  	// no data was read, the buffer already contains some data
   446  	// when padding is disabled this is not an error, as the message can be of
   447  	// any length
   448  	if expectsPadding && min < 8 && n == 0 && err == io.EOF {
   449  		err = io.ErrUnexpectedEOF
   450  	}
   451  	return
   452  }
   453  
   454  func (d *decoder) Read(p []byte) (n int, err error) {
   455  	// Use leftover decoded output from last read.
   456  	if len(d.out) > 0 {
   457  		n = copy(p, d.out)
   458  		d.out = d.out[n:]
   459  		if len(d.out) == 0 {
   460  			return n, d.err
   461  		}
   462  		return n, nil
   463  	}
   464  
   465  	if d.err != nil {
   466  		return 0, d.err
   467  	}
   468  
   469  	// Read a chunk.
   470  	nn := (len(p) + 4) / 5 * 8
   471  	if nn < 8 {
   472  		nn = 8
   473  	}
   474  	if nn > len(d.buf) {
   475  		nn = len(d.buf)
   476  	}
   477  
   478  	// Minimum amount of bytes that needs to be read each cycle
   479  	var min int
   480  	var expectsPadding bool
   481  	if d.enc.padChar == NoPadding {
   482  		min = 1
   483  		expectsPadding = false
   484  	} else {
   485  		min = 8 - d.nbuf
   486  		expectsPadding = true
   487  	}
   488  
   489  	nn, d.err = readEncodedData(d.r, d.buf[d.nbuf:nn], min, expectsPadding)
   490  	d.nbuf += nn
   491  	if d.nbuf < min {
   492  		return 0, d.err
   493  	}
   494  	if nn > 0 && d.end {
   495  		return 0, CorruptInputError(0)
   496  	}
   497  
   498  	// Decode chunk into p, or d.out and then p if p is too small.
   499  	var nr int
   500  	if d.enc.padChar == NoPadding {
   501  		nr = d.nbuf
   502  	} else {
   503  		nr = d.nbuf / 8 * 8
   504  	}
   505  	nw := d.enc.DecodedLen(d.nbuf)
   506  
   507  	if nw > len(p) {
   508  		nw, d.end, err = d.enc.decode(d.outbuf[0:], d.buf[0:nr])
   509  		d.out = d.outbuf[0:nw]
   510  		n = copy(p, d.out)
   511  		d.out = d.out[n:]
   512  	} else {
   513  		n, d.end, err = d.enc.decode(p, d.buf[0:nr])
   514  	}
   515  	d.nbuf -= nr
   516  	for i := 0; i < d.nbuf; i++ {
   517  		d.buf[i] = d.buf[i+nr]
   518  	}
   519  
   520  	if err != nil && (d.err == nil || d.err == io.EOF) {
   521  		d.err = err
   522  	}
   523  
   524  	if len(d.out) > 0 {
   525  		// We cannot return all the decoded bytes to the caller in this
   526  		// invocation of Read, so we return a nil error to ensure that Read
   527  		// will be called again.  The error stored in d.err, if any, will be
   528  		// returned with the last set of decoded bytes.
   529  		return n, nil
   530  	}
   531  
   532  	return n, d.err
   533  }
   534  
   535  type newlineFilteringReader struct {
   536  	wrapped io.Reader
   537  }
   538  
   539  // stripNewlines removes newline characters and returns the number
   540  // of non-newline characters copied to dst.
   541  func stripNewlines(dst, src []byte) int {
   542  	offset := 0
   543  	for _, b := range src {
   544  		if b == '\r' || b == '\n' {
   545  			continue
   546  		}
   547  		dst[offset] = b
   548  		offset++
   549  	}
   550  	return offset
   551  }
   552  
   553  func (r *newlineFilteringReader) Read(p []byte) (int, error) {
   554  	n, err := r.wrapped.Read(p)
   555  	for n > 0 {
   556  		s := p[0:n]
   557  		offset := stripNewlines(s, s)
   558  		if err != nil || offset > 0 {
   559  			return offset, err
   560  		}
   561  		// Previous buffer entirely whitespace, read again
   562  		n, err = r.wrapped.Read(p)
   563  	}
   564  	return n, err
   565  }
   566  
   567  // NewDecoder constructs a new base32 stream decoder.
   568  func NewDecoder(enc *Encoding, r io.Reader) io.Reader {
   569  	return &decoder{enc: enc, r: &newlineFilteringReader{r}}
   570  }
   571  
   572  // DecodedLen returns the maximum length in bytes of the decoded data
   573  // corresponding to n bytes of base32-encoded data.
   574  func (enc *Encoding) DecodedLen(n int) int {
   575  	return decodedLen(n, enc.padChar)
   576  }
   577  
   578  func decodedLen(n int, padChar rune) int {
   579  	if padChar == NoPadding {
   580  		return n/8*5 + n%8*5/8
   581  	}
   582  	return n / 8 * 5
   583  }
   584
View as plain text