Source file src/encoding/base64/base64.go

Documentation: encoding/base64

     1  // Copyright 2009 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 base64 implements base64 encoding as specified by RFC 4648.
     6  package base64
     7  
     8  import (
     9  	"encoding/binary"
    10  	"io"
    11  	"strconv"
    12  )
    13  
    14  /*
    15   * Encodings
    16   */
    17  
    18  // An Encoding is a radix 64 encoding/decoding scheme, defined by a
    19  // 64-character alphabet. The most common encoding is the "base64"
    20  // encoding defined in RFC 4648 and used in MIME (RFC 2045) and PEM
    21  // (RFC 1421).  RFC 4648 also defines an alternate encoding, which is
    22  // the standard encoding with - and _ substituted for + and /.
    23  type Encoding struct {
    24  	encode    [64]byte
    25  	decodeMap [256]byte
    26  	padChar   rune
    27  	strict    bool
    28  }
    29  
    30  const (
    31  	StdPadding rune = '=' // Standard padding character
    32  	NoPadding  rune = -1  // No padding
    33  )
    34  
    35  const encodeStd = "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789+/"
    36  const encodeURL = "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789-_"
    37  
    38  // NewEncoding returns a new padded Encoding defined by the given alphabet,
    39  // which must be a 64-byte string that does not contain the padding character
    40  // or CR / LF ('\r', '\n').
    41  // The resulting Encoding uses the default padding character ('='),
    42  // which may be changed or disabled via WithPadding.
    43  func NewEncoding(encoder string) *Encoding {
    44  	if len(encoder) != 64 {
    45  		panic("encoding alphabet is not 64-bytes long")
    46  	}
    47  	for i := 0; i < len(encoder); i++ {
    48  		if encoder[i] == '\n' || encoder[i] == '\r' {
    49  			panic("encoding alphabet contains newline character")
    50  		}
    51  	}
    52  
    53  	e := new(Encoding)
    54  	e.padChar = StdPadding
    55  	copy(e.encode[:], encoder)
    56  
    57  	for i := 0; i < len(e.decodeMap); i++ {
    58  		e.decodeMap[i] = 0xFF
    59  	}
    60  	for i := 0; i < len(encoder); i++ {
    61  		e.decodeMap[encoder[i]] = byte(i)
    62  	}
    63  	return e
    64  }
    65  
    66  // WithPadding creates a new encoding identical to enc except
    67  // with a specified padding character, or NoPadding to disable padding.
    68  // The padding character must not be '\r' or '\n', must not
    69  // be contained in the encoding's alphabet and must be a rune equal or
    70  // below '\xff'.
    71  func (enc Encoding) WithPadding(padding rune) *Encoding {
    72  	if padding == '\r' || padding == '\n' || padding > 0xff {
    73  		panic("invalid padding")
    74  	}
    75  
    76  	for i := 0; i < len(enc.encode); i++ {
    77  		if rune(enc.encode[i]) == padding {
    78  			panic("padding contained in alphabet")
    79  		}
    80  	}
    81  
    82  	enc.padChar = padding
    83  	return &enc
    84  }
    85  
    86  // Strict creates a new encoding identical to enc except with
    87  // strict decoding enabled. In this mode, the decoder requires that
    88  // trailing padding bits are zero, as described in RFC 4648 section 3.5.
    89  func (enc Encoding) Strict() *Encoding {
    90  	enc.strict = true
    91  	return &enc
    92  }
    93  
    94  // StdEncoding is the standard base64 encoding, as defined in
    95  // RFC 4648.
    96  var StdEncoding = NewEncoding(encodeStd)
    97  
    98  // URLEncoding is the alternate base64 encoding defined in RFC 4648.
    99  // It is typically used in URLs and file names.
   100  var URLEncoding = NewEncoding(encodeURL)
   101  
   102  // RawStdEncoding is the standard raw, unpadded base64 encoding,
   103  // as defined in RFC 4648 section 3.2.
   104  // This is the same as StdEncoding but omits padding characters.
   105  var RawStdEncoding = StdEncoding.WithPadding(NoPadding)
   106  
   107  // RawURLEncoding is the unpadded alternate base64 encoding defined in RFC 4648.
   108  // It is typically used in URLs and file names.
   109  // This is the same as URLEncoding but omits padding characters.
   110  var RawURLEncoding = URLEncoding.WithPadding(NoPadding)
   111  
   112  /*
   113   * Encoder
   114   */
   115  
   116  // Encode encodes src using the encoding enc, writing
   117  // EncodedLen(len(src)) bytes to dst.
   118  //
   119  // The encoding pads the output to a multiple of 4 bytes,
   120  // so Encode is not appropriate for use on individual blocks
   121  // of a large data stream. Use NewEncoder() instead.
   122  func (enc *Encoding) Encode(dst, src []byte) {
   123  	if len(src) == 0 {
   124  		return
   125  	}
   126  	// enc is a pointer receiver, so the use of enc.encode within the hot
   127  	// loop below means a nil check at every operation. Lift that nil check
   128  	// outside of the loop to speed up the encoder.
   129  	_ = enc.encode
   130  
   131  	di, si := 0, 0
   132  	n := (len(src) / 3) * 3
   133  	for si < n {
   134  		// Convert 3x 8bit source bytes into 4 bytes
   135  		val := uint(src[si+0])<<16 | uint(src[si+1])<<8 | uint(src[si+2])
   136  
   137  		dst[di+0] = enc.encode[val>>18&0x3F]
   138  		dst[di+1] = enc.encode[val>>12&0x3F]
   139  		dst[di+2] = enc.encode[val>>6&0x3F]
   140  		dst[di+3] = enc.encode[val&0x3F]
   141  
   142  		si += 3
   143  		di += 4
   144  	}
   145  
   146  	remain := len(src) - si
   147  	if remain == 0 {
   148  		return
   149  	}
   150  	// Add the remaining small block
   151  	val := uint(src[si+0]) << 16
   152  	if remain == 2 {
   153  		val |= uint(src[si+1]) << 8
   154  	}
   155  
   156  	dst[di+0] = enc.encode[val>>18&0x3F]
   157  	dst[di+1] = enc.encode[val>>12&0x3F]
   158  
   159  	switch remain {
   160  	case 2:
   161  		dst[di+2] = enc.encode[val>>6&0x3F]
   162  		if enc.padChar != NoPadding {
   163  			dst[di+3] = byte(enc.padChar)
   164  		}
   165  	case 1:
   166  		if enc.padChar != NoPadding {
   167  			dst[di+2] = byte(enc.padChar)
   168  			dst[di+3] = byte(enc.padChar)
   169  		}
   170  	}
   171  }
   172  
   173  // EncodeToString returns the base64 encoding of src.
   174  func (enc *Encoding) EncodeToString(src []byte) string {
   175  	buf := make([]byte, enc.EncodedLen(len(src)))
   176  	enc.Encode(buf, src)
   177  	return string(buf)
   178  }
   179  
   180  type encoder struct {
   181  	err  error
   182  	enc  *Encoding
   183  	w    io.Writer
   184  	buf  [3]byte    // buffered data waiting to be encoded
   185  	nbuf int        // number of bytes in buf
   186  	out  [1024]byte // output buffer
   187  }
   188  
   189  func (e *encoder) Write(p []byte) (n int, err error) {
   190  	if e.err != nil {
   191  		return 0, e.err
   192  	}
   193  
   194  	// Leading fringe.
   195  	if e.nbuf > 0 {
   196  		var i int
   197  		for i = 0; i < len(p) && e.nbuf < 3; i++ {
   198  			e.buf[e.nbuf] = p[i]
   199  			e.nbuf++
   200  		}
   201  		n += i
   202  		p = p[i:]
   203  		if e.nbuf < 3 {
   204  			return
   205  		}
   206  		e.enc.Encode(e.out[:], e.buf[:])
   207  		if _, e.err = e.w.Write(e.out[:4]); e.err != nil {
   208  			return n, e.err
   209  		}
   210  		e.nbuf = 0
   211  	}
   212  
   213  	// Large interior chunks.
   214  	for len(p) >= 3 {
   215  		nn := len(e.out) / 4 * 3
   216  		if nn > len(p) {
   217  			nn = len(p)
   218  			nn -= nn % 3
   219  		}
   220  		e.enc.Encode(e.out[:], p[:nn])
   221  		if _, e.err = e.w.Write(e.out[0 : nn/3*4]); e.err != nil {
   222  			return n, e.err
   223  		}
   224  		n += nn
   225  		p = p[nn:]
   226  	}
   227  
   228  	// Trailing fringe.
   229  	for i := 0; i < len(p); i++ {
   230  		e.buf[i] = p[i]
   231  	}
   232  	e.nbuf = len(p)
   233  	n += len(p)
   234  	return
   235  }
   236  
   237  // Close flushes any pending output from the encoder.
   238  // It is an error to call Write after calling Close.
   239  func (e *encoder) Close() error {
   240  	// If there's anything left in the buffer, flush it out
   241  	if e.err == nil && e.nbuf > 0 {
   242  		e.enc.Encode(e.out[:], e.buf[:e.nbuf])
   243  		_, e.err = e.w.Write(e.out[:e.enc.EncodedLen(e.nbuf)])
   244  		e.nbuf = 0
   245  	}
   246  	return e.err
   247  }
   248  
   249  // NewEncoder returns a new base64 stream encoder. Data written to
   250  // the returned writer will be encoded using enc and then written to w.
   251  // Base64 encodings operate in 4-byte blocks; when finished
   252  // writing, the caller must Close the returned encoder to flush any
   253  // partially written blocks.
   254  func NewEncoder(enc *Encoding, w io.Writer) io.WriteCloser {
   255  	return &encoder{enc: enc, w: w}
   256  }
   257  
   258  // EncodedLen returns the length in bytes of the base64 encoding
   259  // of an input buffer of length n.
   260  func (enc *Encoding) EncodedLen(n int) int {
   261  	if enc.padChar == NoPadding {
   262  		return (n*8 + 5) / 6 // minimum # chars at 6 bits per char
   263  	}
   264  	return (n + 2) / 3 * 4 // minimum # 4-char quanta, 3 bytes each
   265  }
   266  
   267  /*
   268   * Decoder
   269   */
   270  
   271  type CorruptInputError int64
   272  
   273  func (e CorruptInputError) Error() string {
   274  	return "illegal base64 data at input byte " + strconv.FormatInt(int64(e), 10)
   275  }
   276  
   277  // decodeQuantum decodes up to 4 base64 bytes. The received parameters are
   278  // the destination buffer dst, the source buffer src and an index in the
   279  // source buffer si.
   280  // It returns the number of bytes read from src, the number of bytes written
   281  // to dst, and an error, if any.
   282  func (enc *Encoding) decodeQuantum(dst, src []byte, si int) (nsi, n int, err error) {
   283  	// Decode quantum using the base64 alphabet
   284  	var dbuf [4]byte
   285  	dlen := 4
   286  
   287  	// Lift the nil check outside of the loop.
   288  	_ = enc.decodeMap
   289  
   290  	for j := 0; j < len(dbuf); j++ {
   291  		if len(src) == si {
   292  			switch {
   293  			case j == 0:
   294  				return si, 0, nil
   295  			case j == 1, enc.padChar != NoPadding:
   296  				return si, 0, CorruptInputError(si - j)
   297  			}
   298  			dlen = j
   299  			break
   300  		}
   301  		in := src[si]
   302  		si++
   303  
   304  		out := enc.decodeMap[in]
   305  		if out != 0xff {
   306  			dbuf[j] = out
   307  			continue
   308  		}
   309  
   310  		if in == '\n' || in == '\r' {
   311  			j--
   312  			continue
   313  		}
   314  
   315  		if rune(in) != enc.padChar {
   316  			return si, 0, CorruptInputError(si - 1)
   317  		}
   318  
   319  		// We've reached the end and there's padding
   320  		switch j {
   321  		case 0, 1:
   322  			// incorrect padding
   323  			return si, 0, CorruptInputError(si - 1)
   324  		case 2:
   325  			// "==" is expected, the first "=" is already consumed.
   326  			// skip over newlines
   327  			for si < len(src) && (src[si] == '\n' || src[si] == '\r') {
   328  				si++
   329  			}
   330  			if si == len(src) {
   331  				// not enough padding
   332  				return si, 0, CorruptInputError(len(src))
   333  			}
   334  			if rune(src[si]) != enc.padChar {
   335  				// incorrect padding
   336  				return si, 0, CorruptInputError(si - 1)
   337  			}
   338  
   339  			si++
   340  		}
   341  
   342  		// skip over newlines
   343  		for si < len(src) && (src[si] == '\n' || src[si] == '\r') {
   344  			si++
   345  		}
   346  		if si < len(src) {
   347  			// trailing garbage
   348  			err = CorruptInputError(si)
   349  		}
   350  		dlen = j
   351  		break
   352  	}
   353  
   354  	// Convert 4x 6bit source bytes into 3 bytes
   355  	val := uint(dbuf[0])<<18 | uint(dbuf[1])<<12 | uint(dbuf[2])<<6 | uint(dbuf[3])
   356  	dbuf[2], dbuf[1], dbuf[0] = byte(val>>0), byte(val>>8), byte(val>>16)
   357  	switch dlen {
   358  	case 4:
   359  		dst[2] = dbuf[2]
   360  		dbuf[2] = 0
   361  		fallthrough
   362  	case 3:
   363  		dst[1] = dbuf[1]
   364  		if enc.strict && dbuf[2] != 0 {
   365  			return si, 0, CorruptInputError(si - 1)
   366  		}
   367  		dbuf[1] = 0
   368  		fallthrough
   369  	case 2:
   370  		dst[0] = dbuf[0]
   371  		if enc.strict && (dbuf[1] != 0 || dbuf[2] != 0) {
   372  			return si, 0, CorruptInputError(si - 2)
   373  		}
   374  	}
   375  
   376  	return si, dlen - 1, err
   377  }
   378  
   379  // DecodeString returns the bytes represented by the base64 string s.
   380  func (enc *Encoding) DecodeString(s string) ([]byte, error) {
   381  	dbuf := make([]byte, enc.DecodedLen(len(s)))
   382  	n, err := enc.Decode(dbuf, []byte(s))
   383  	return dbuf[:n], err
   384  }
   385  
   386  type decoder struct {
   387  	err     error
   388  	readErr error // error from r.Read
   389  	enc     *Encoding
   390  	r       io.Reader
   391  	buf     [1024]byte // leftover input
   392  	nbuf    int
   393  	out     []byte // leftover decoded output
   394  	outbuf  [1024 / 4 * 3]byte
   395  }
   396  
   397  func (d *decoder) Read(p []byte) (n int, err error) {
   398  	// Use leftover decoded output from last read.
   399  	if len(d.out) > 0 {
   400  		n = copy(p, d.out)
   401  		d.out = d.out[n:]
   402  		return n, nil
   403  	}
   404  
   405  	if d.err != nil {
   406  		return 0, d.err
   407  	}
   408  
   409  	// This code assumes that d.r strips supported whitespace ('\r' and '\n').
   410  
   411  	// Refill buffer.
   412  	for d.nbuf < 4 && d.readErr == nil {
   413  		nn := len(p) / 3 * 4
   414  		if nn < 4 {
   415  			nn = 4
   416  		}
   417  		if nn > len(d.buf) {
   418  			nn = len(d.buf)
   419  		}
   420  		nn, d.readErr = d.r.Read(d.buf[d.nbuf:nn])
   421  		d.nbuf += nn
   422  	}
   423  
   424  	if d.nbuf < 4 {
   425  		if d.enc.padChar == NoPadding && d.nbuf > 0 {
   426  			// Decode final fragment, without padding.
   427  			var nw int
   428  			nw, d.err = d.enc.Decode(d.outbuf[:], d.buf[:d.nbuf])
   429  			d.nbuf = 0
   430  			d.out = d.outbuf[:nw]
   431  			n = copy(p, d.out)
   432  			d.out = d.out[n:]
   433  			if n > 0 || len(p) == 0 && len(d.out) > 0 {
   434  				return n, nil
   435  			}
   436  			if d.err != nil {
   437  				return 0, d.err
   438  			}
   439  		}
   440  		d.err = d.readErr
   441  		if d.err == io.EOF && d.nbuf > 0 {
   442  			d.err = io.ErrUnexpectedEOF
   443  		}
   444  		return 0, d.err
   445  	}
   446  
   447  	// Decode chunk into p, or d.out and then p if p is too small.
   448  	nr := d.nbuf / 4 * 4
   449  	nw := d.nbuf / 4 * 3
   450  	if nw > len(p) {
   451  		nw, d.err = d.enc.Decode(d.outbuf[:], d.buf[:nr])
   452  		d.out = d.outbuf[:nw]
   453  		n = copy(p, d.out)
   454  		d.out = d.out[n:]
   455  	} else {
   456  		n, d.err = d.enc.Decode(p, d.buf[:nr])
   457  	}
   458  	d.nbuf -= nr
   459  	copy(d.buf[:d.nbuf], d.buf[nr:])
   460  	return n, d.err
   461  }
   462  
   463  // Decode decodes src using the encoding enc. It writes at most
   464  // DecodedLen(len(src)) bytes to dst and returns the number of bytes
   465  // written. If src contains invalid base64 data, it will return the
   466  // number of bytes successfully written and CorruptInputError.
   467  // New line characters (\r and \n) are ignored.
   468  func (enc *Encoding) Decode(dst, src []byte) (n int, err error) {
   469  	if len(src) == 0 {
   470  		return 0, nil
   471  	}
   472  
   473  	// Lift the nil check outside of the loop. enc.decodeMap is directly
   474  	// used later in this function, to let the compiler know that the
   475  	// receiver can't be nil.
   476  	_ = enc.decodeMap
   477  
   478  	si := 0
   479  	for strconv.IntSize >= 64 && len(src)-si >= 8 && len(dst)-n >= 8 {
   480  		if dn, ok := assemble64(
   481  			enc.decodeMap[src[si+0]],
   482  			enc.decodeMap[src[si+1]],
   483  			enc.decodeMap[src[si+2]],
   484  			enc.decodeMap[src[si+3]],
   485  			enc.decodeMap[src[si+4]],
   486  			enc.decodeMap[src[si+5]],
   487  			enc.decodeMap[src[si+6]],
   488  			enc.decodeMap[src[si+7]],
   489  		); ok {
   490  			binary.BigEndian.PutUint64(dst[n:], dn)
   491  			n += 6
   492  			si += 8
   493  		} else {
   494  			var ninc int
   495  			si, ninc, err = enc.decodeQuantum(dst[n:], src, si)
   496  			n += ninc
   497  			if err != nil {
   498  				return n, err
   499  			}
   500  		}
   501  	}
   502  
   503  	for len(src)-si >= 4 && len(dst)-n >= 4 {
   504  		if dn, ok := assemble32(
   505  			enc.decodeMap[src[si+0]],
   506  			enc.decodeMap[src[si+1]],
   507  			enc.decodeMap[src[si+2]],
   508  			enc.decodeMap[src[si+3]],
   509  		); ok {
   510  			binary.BigEndian.PutUint32(dst[n:], dn)
   511  			n += 3
   512  			si += 4
   513  		} else {
   514  			var ninc int
   515  			si, ninc, err = enc.decodeQuantum(dst[n:], src, si)
   516  			n += ninc
   517  			if err != nil {
   518  				return n, err
   519  			}
   520  		}
   521  	}
   522  
   523  	for si < len(src) {
   524  		var ninc int
   525  		si, ninc, err = enc.decodeQuantum(dst[n:], src, si)
   526  		n += ninc
   527  		if err != nil {
   528  			return n, err
   529  		}
   530  	}
   531  	return n, err
   532  }
   533  
   534  // assemble32 assembles 4 base64 digits into 3 bytes.
   535  // Each digit comes from the decode map, and will be 0xff
   536  // if it came from an invalid character.
   537  func assemble32(n1, n2, n3, n4 byte) (dn uint32, ok bool) {
   538  	// Check that all the digits are valid. If any of them was 0xff, their
   539  	// bitwise OR will be 0xff.
   540  	if n1|n2|n3|n4 == 0xff {
   541  		return 0, false
   542  	}
   543  	return uint32(n1)<<26 |
   544  			uint32(n2)<<20 |
   545  			uint32(n3)<<14 |
   546  			uint32(n4)<<8,
   547  		true
   548  }
   549  
   550  // assemble64 assembles 8 base64 digits into 6 bytes.
   551  // Each digit comes from the decode map, and will be 0xff
   552  // if it came from an invalid character.
   553  func assemble64(n1, n2, n3, n4, n5, n6, n7, n8 byte) (dn uint64, ok bool) {
   554  	// Check that all the digits are valid. If any of them was 0xff, their
   555  	// bitwise OR will be 0xff.
   556  	if n1|n2|n3|n4|n5|n6|n7|n8 == 0xff {
   557  		return 0, false
   558  	}
   559  	return uint64(n1)<<58 |
   560  			uint64(n2)<<52 |
   561  			uint64(n3)<<46 |
   562  			uint64(n4)<<40 |
   563  			uint64(n5)<<34 |
   564  			uint64(n6)<<28 |
   565  			uint64(n7)<<22 |
   566  			uint64(n8)<<16,
   567  		true
   568  }
   569  
   570  type newlineFilteringReader struct {
   571  	wrapped io.Reader
   572  }
   573  
   574  func (r *newlineFilteringReader) Read(p []byte) (int, error) {
   575  	n, err := r.wrapped.Read(p)
   576  	for n > 0 {
   577  		offset := 0
   578  		for i, b := range p[:n] {
   579  			if b != '\r' && b != '\n' {
   580  				if i != offset {
   581  					p[offset] = b
   582  				}
   583  				offset++
   584  			}
   585  		}
   586  		if offset > 0 {
   587  			return offset, err
   588  		}
   589  		// Previous buffer entirely whitespace, read again
   590  		n, err = r.wrapped.Read(p)
   591  	}
   592  	return n, err
   593  }
   594  
   595  // NewDecoder constructs a new base64 stream decoder.
   596  func NewDecoder(enc *Encoding, r io.Reader) io.Reader {
   597  	return &decoder{enc: enc, r: &newlineFilteringReader{r}}
   598  }
   599  
   600  // DecodedLen returns the maximum length in bytes of the decoded data
   601  // corresponding to n bytes of base64-encoded data.
   602  func (enc *Encoding) DecodedLen(n int) int {
   603  	if enc.padChar == NoPadding {
   604  		// Unpadded data may end with partial block of 2-3 characters.
   605  		return n * 6 / 8
   606  	}
   607  	// Padded base64 should always be a multiple of 4 characters in length.
   608  	return n / 4 * 3
   609  }
   610  

View as plain text