Black Lives Matter. Support the Equal Justice Initiative.

Source file src/mime/multipart/multipart.go

Documentation: mime/multipart

     1  // Copyright 2010 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  
     6  /*
     7  Package multipart implements MIME multipart parsing, as defined in RFC
     8  2046.
     9  
    10  The implementation is sufficient for HTTP (RFC 2388) and the multipart
    11  bodies generated by popular browsers.
    12  */
    13  package multipart
    14  
    15  import (
    16  	"bufio"
    17  	"bytes"
    18  	"fmt"
    19  	"io"
    20  	"mime"
    21  	"mime/quotedprintable"
    22  	"net/textproto"
    23  	"strings"
    24  )
    25  
    26  var emptyParams = make(map[string]string)
    27  
    28  // This constant needs to be at least 76 for this package to work correctly.
    29  // This is because \r\n--separator_of_len_70- would fill the buffer and it
    30  // wouldn't be safe to consume a single byte from it.
    31  const peekBufferSize = 4096
    32  
    33  // A Part represents a single part in a multipart body.
    34  type Part struct {
    35  	// The headers of the body, if any, with the keys canonicalized
    36  	// in the same fashion that the Go http.Request headers are.
    37  	// For example, "foo-bar" changes case to "Foo-Bar"
    38  	Header textproto.MIMEHeader
    39  
    40  	mr *Reader
    41  
    42  	disposition       string
    43  	dispositionParams map[string]string
    44  
    45  	// r is either a reader directly reading from mr, or it's a
    46  	// wrapper around such a reader, decoding the
    47  	// Content-Transfer-Encoding
    48  	r io.Reader
    49  
    50  	n       int   // known data bytes waiting in mr.bufReader
    51  	total   int64 // total data bytes read already
    52  	err     error // error to return when n == 0
    53  	readErr error // read error observed from mr.bufReader
    54  }
    55  
    56  // FormName returns the name parameter if p has a Content-Disposition
    57  // of type "form-data".  Otherwise it returns the empty string.
    58  func (p *Part) FormName() string {
    59  	// See https://tools.ietf.org/html/rfc2183 section 2 for EBNF
    60  	// of Content-Disposition value format.
    61  	if p.dispositionParams == nil {
    62  		p.parseContentDisposition()
    63  	}
    64  	if p.disposition != "form-data" {
    65  		return ""
    66  	}
    67  	return p.dispositionParams["name"]
    68  }
    69  
    70  // FileName returns the filename parameter of the Part's
    71  // Content-Disposition header.
    72  func (p *Part) FileName() string {
    73  	if p.dispositionParams == nil {
    74  		p.parseContentDisposition()
    75  	}
    76  	return p.dispositionParams["filename"]
    77  }
    78  
    79  func (p *Part) parseContentDisposition() {
    80  	v := p.Header.Get("Content-Disposition")
    81  	var err error
    82  	p.disposition, p.dispositionParams, err = mime.ParseMediaType(v)
    83  	if err != nil {
    84  		p.dispositionParams = emptyParams
    85  	}
    86  }
    87  
    88  // NewReader creates a new multipart Reader reading from r using the
    89  // given MIME boundary.
    90  //
    91  // The boundary is usually obtained from the "boundary" parameter of
    92  // the message's "Content-Type" header. Use mime.ParseMediaType to
    93  // parse such headers.
    94  func NewReader(r io.Reader, boundary string) *Reader {
    95  	b := []byte("\r\n--" + boundary + "--")
    96  	return &Reader{
    97  		bufReader:        bufio.NewReaderSize(&stickyErrorReader{r: r}, peekBufferSize),
    98  		nl:               b[:2],
    99  		nlDashBoundary:   b[:len(b)-2],
   100  		dashBoundaryDash: b[2:],
   101  		dashBoundary:     b[2 : len(b)-2],
   102  	}
   103  }
   104  
   105  // stickyErrorReader is an io.Reader which never calls Read on its
   106  // underlying Reader once an error has been seen. (the io.Reader
   107  // interface's contract promises nothing about the return values of
   108  // Read calls after an error, yet this package does do multiple Reads
   109  // after error)
   110  type stickyErrorReader struct {
   111  	r   io.Reader
   112  	err error
   113  }
   114  
   115  func (r *stickyErrorReader) Read(p []byte) (n int, _ error) {
   116  	if r.err != nil {
   117  		return 0, r.err
   118  	}
   119  	n, r.err = r.r.Read(p)
   120  	return n, r.err
   121  }
   122  
   123  func newPart(mr *Reader, rawPart bool) (*Part, error) {
   124  	bp := &Part{
   125  		Header: make(map[string][]string),
   126  		mr:     mr,
   127  	}
   128  	if err := bp.populateHeaders(); err != nil {
   129  		return nil, err
   130  	}
   131  	bp.r = partReader{bp}
   132  
   133  	// rawPart is used to switch between Part.NextPart and Part.NextRawPart.
   134  	if !rawPart {
   135  		const cte = "Content-Transfer-Encoding"
   136  		if strings.EqualFold(bp.Header.Get(cte), "quoted-printable") {
   137  			bp.Header.Del(cte)
   138  			bp.r = quotedprintable.NewReader(bp.r)
   139  		}
   140  	}
   141  	return bp, nil
   142  }
   143  
   144  func (bp *Part) populateHeaders() error {
   145  	r := textproto.NewReader(bp.mr.bufReader)
   146  	header, err := r.ReadMIMEHeader()
   147  	if err == nil {
   148  		bp.Header = header
   149  	}
   150  	return err
   151  }
   152  
   153  // Read reads the body of a part, after its headers and before the
   154  // next part (if any) begins.
   155  func (p *Part) Read(d []byte) (n int, err error) {
   156  	return p.r.Read(d)
   157  }
   158  
   159  // partReader implements io.Reader by reading raw bytes directly from the
   160  // wrapped *Part, without doing any Transfer-Encoding decoding.
   161  type partReader struct {
   162  	p *Part
   163  }
   164  
   165  func (pr partReader) Read(d []byte) (int, error) {
   166  	p := pr.p
   167  	br := p.mr.bufReader
   168  
   169  	// Read into buffer until we identify some data to return,
   170  	// or we find a reason to stop (boundary or read error).
   171  	for p.n == 0 && p.err == nil {
   172  		peek, _ := br.Peek(br.Buffered())
   173  		p.n, p.err = scanUntilBoundary(peek, p.mr.dashBoundary, p.mr.nlDashBoundary, p.total, p.readErr)
   174  		if p.n == 0 && p.err == nil {
   175  			// Force buffered I/O to read more into buffer.
   176  			_, p.readErr = br.Peek(len(peek) + 1)
   177  			if p.readErr == io.EOF {
   178  				p.readErr = io.ErrUnexpectedEOF
   179  			}
   180  		}
   181  	}
   182  
   183  	// Read out from "data to return" part of buffer.
   184  	if p.n == 0 {
   185  		return 0, p.err
   186  	}
   187  	n := len(d)
   188  	if n > p.n {
   189  		n = p.n
   190  	}
   191  	n, _ = br.Read(d[:n])
   192  	p.total += int64(n)
   193  	p.n -= n
   194  	if p.n == 0 {
   195  		return n, p.err
   196  	}
   197  	return n, nil
   198  }
   199  
   200  // scanUntilBoundary scans buf to identify how much of it can be safely
   201  // returned as part of the Part body.
   202  // dashBoundary is "--boundary".
   203  // nlDashBoundary is "\r\n--boundary" or "\n--boundary", depending on what mode we are in.
   204  // The comments below (and the name) assume "\n--boundary", but either is accepted.
   205  // total is the number of bytes read out so far. If total == 0, then a leading "--boundary" is recognized.
   206  // readErr is the read error, if any, that followed reading the bytes in buf.
   207  // scanUntilBoundary returns the number of data bytes from buf that can be
   208  // returned as part of the Part body and also the error to return (if any)
   209  // once those data bytes are done.
   210  func scanUntilBoundary(buf, dashBoundary, nlDashBoundary []byte, total int64, readErr error) (int, error) {
   211  	if total == 0 {
   212  		// At beginning of body, allow dashBoundary.
   213  		if bytes.HasPrefix(buf, dashBoundary) {
   214  			switch matchAfterPrefix(buf, dashBoundary, readErr) {
   215  			case -1:
   216  				return len(dashBoundary), nil
   217  			case 0:
   218  				return 0, nil
   219  			case +1:
   220  				return 0, io.EOF
   221  			}
   222  		}
   223  		if bytes.HasPrefix(dashBoundary, buf) {
   224  			return 0, readErr
   225  		}
   226  	}
   227  
   228  	// Search for "\n--boundary".
   229  	if i := bytes.Index(buf, nlDashBoundary); i >= 0 {
   230  		switch matchAfterPrefix(buf[i:], nlDashBoundary, readErr) {
   231  		case -1:
   232  			return i + len(nlDashBoundary), nil
   233  		case 0:
   234  			return i, nil
   235  		case +1:
   236  			return i, io.EOF
   237  		}
   238  	}
   239  	if bytes.HasPrefix(nlDashBoundary, buf) {
   240  		return 0, readErr
   241  	}
   242  
   243  	// Otherwise, anything up to the final \n is not part of the boundary
   244  	// and so must be part of the body.
   245  	// Also if the section from the final \n onward is not a prefix of the boundary,
   246  	// it too must be part of the body.
   247  	i := bytes.LastIndexByte(buf, nlDashBoundary[0])
   248  	if i >= 0 && bytes.HasPrefix(nlDashBoundary, buf[i:]) {
   249  		return i, nil
   250  	}
   251  	return len(buf), readErr
   252  }
   253  
   254  // matchAfterPrefix checks whether buf should be considered to match the boundary.
   255  // The prefix is "--boundary" or "\r\n--boundary" or "\n--boundary",
   256  // and the caller has verified already that bytes.HasPrefix(buf, prefix) is true.
   257  //
   258  // matchAfterPrefix returns +1 if the buffer does match the boundary,
   259  // meaning the prefix is followed by a dash, space, tab, cr, nl, or end of input.
   260  // It returns -1 if the buffer definitely does NOT match the boundary,
   261  // meaning the prefix is followed by some other character.
   262  // For example, "--foobar" does not match "--foo".
   263  // It returns 0 more input needs to be read to make the decision,
   264  // meaning that len(buf) == len(prefix) and readErr == nil.
   265  func matchAfterPrefix(buf, prefix []byte, readErr error) int {
   266  	if len(buf) == len(prefix) {
   267  		if readErr != nil {
   268  			return +1
   269  		}
   270  		return 0
   271  	}
   272  	c := buf[len(prefix)]
   273  	if c == ' ' || c == '\t' || c == '\r' || c == '\n' || c == '-' {
   274  		return +1
   275  	}
   276  	return -1
   277  }
   278  
   279  func (p *Part) Close() error {
   280  	io.Copy(io.Discard, p)
   281  	return nil
   282  }
   283  
   284  // Reader is an iterator over parts in a MIME multipart body.
   285  // Reader's underlying parser consumes its input as needed. Seeking
   286  // isn't supported.
   287  type Reader struct {
   288  	bufReader *bufio.Reader
   289  
   290  	currentPart *Part
   291  	partsRead   int
   292  
   293  	nl               []byte // "\r\n" or "\n" (set after seeing first boundary line)
   294  	nlDashBoundary   []byte // nl + "--boundary"
   295  	dashBoundaryDash []byte // "--boundary--"
   296  	dashBoundary     []byte // "--boundary"
   297  }
   298  
   299  // NextPart returns the next part in the multipart or an error.
   300  // When there are no more parts, the error io.EOF is returned.
   301  //
   302  // As a special case, if the "Content-Transfer-Encoding" header
   303  // has a value of "quoted-printable", that header is instead
   304  // hidden and the body is transparently decoded during Read calls.
   305  func (r *Reader) NextPart() (*Part, error) {
   306  	return r.nextPart(false)
   307  }
   308  
   309  // NextRawPart returns the next part in the multipart or an error.
   310  // When there are no more parts, the error io.EOF is returned.
   311  //
   312  // Unlike NextPart, it does not have special handling for
   313  // "Content-Transfer-Encoding: quoted-printable".
   314  func (r *Reader) NextRawPart() (*Part, error) {
   315  	return r.nextPart(true)
   316  }
   317  
   318  func (r *Reader) nextPart(rawPart bool) (*Part, error) {
   319  	if r.currentPart != nil {
   320  		r.currentPart.Close()
   321  	}
   322  	if string(r.dashBoundary) == "--" {
   323  		return nil, fmt.Errorf("multipart: boundary is empty")
   324  	}
   325  	expectNewPart := false
   326  	for {
   327  		line, err := r.bufReader.ReadSlice('\n')
   328  
   329  		if err == io.EOF && r.isFinalBoundary(line) {
   330  			// If the buffer ends in "--boundary--" without the
   331  			// trailing "\r\n", ReadSlice will return an error
   332  			// (since it's missing the '\n'), but this is a valid
   333  			// multipart EOF so we need to return io.EOF instead of
   334  			// a fmt-wrapped one.
   335  			return nil, io.EOF
   336  		}
   337  		if err != nil {
   338  			return nil, fmt.Errorf("multipart: NextPart: %v", err)
   339  		}
   340  
   341  		if r.isBoundaryDelimiterLine(line) {
   342  			r.partsRead++
   343  			bp, err := newPart(r, rawPart)
   344  			if err != nil {
   345  				return nil, err
   346  			}
   347  			r.currentPart = bp
   348  			return bp, nil
   349  		}
   350  
   351  		if r.isFinalBoundary(line) {
   352  			// Expected EOF
   353  			return nil, io.EOF
   354  		}
   355  
   356  		if expectNewPart {
   357  			return nil, fmt.Errorf("multipart: expecting a new Part; got line %q", string(line))
   358  		}
   359  
   360  		if r.partsRead == 0 {
   361  			// skip line
   362  			continue
   363  		}
   364  
   365  		// Consume the "\n" or "\r\n" separator between the
   366  		// body of the previous part and the boundary line we
   367  		// now expect will follow. (either a new part or the
   368  		// end boundary)
   369  		if bytes.Equal(line, r.nl) {
   370  			expectNewPart = true
   371  			continue
   372  		}
   373  
   374  		return nil, fmt.Errorf("multipart: unexpected line in Next(): %q", line)
   375  	}
   376  }
   377  
   378  // isFinalBoundary reports whether line is the final boundary line
   379  // indicating that all parts are over.
   380  // It matches `^--boundary--[ \t]*(\r\n)?$`
   381  func (mr *Reader) isFinalBoundary(line []byte) bool {
   382  	if !bytes.HasPrefix(line, mr.dashBoundaryDash) {
   383  		return false
   384  	}
   385  	rest := line[len(mr.dashBoundaryDash):]
   386  	rest = skipLWSPChar(rest)
   387  	return len(rest) == 0 || bytes.Equal(rest, mr.nl)
   388  }
   389  
   390  func (mr *Reader) isBoundaryDelimiterLine(line []byte) (ret bool) {
   391  	// https://tools.ietf.org/html/rfc2046#section-5.1
   392  	//   The boundary delimiter line is then defined as a line
   393  	//   consisting entirely of two hyphen characters ("-",
   394  	//   decimal value 45) followed by the boundary parameter
   395  	//   value from the Content-Type header field, optional linear
   396  	//   whitespace, and a terminating CRLF.
   397  	if !bytes.HasPrefix(line, mr.dashBoundary) {
   398  		return false
   399  	}
   400  	rest := line[len(mr.dashBoundary):]
   401  	rest = skipLWSPChar(rest)
   402  
   403  	// On the first part, see our lines are ending in \n instead of \r\n
   404  	// and switch into that mode if so. This is a violation of the spec,
   405  	// but occurs in practice.
   406  	if mr.partsRead == 0 && len(rest) == 1 && rest[0] == '\n' {
   407  		mr.nl = mr.nl[1:]
   408  		mr.nlDashBoundary = mr.nlDashBoundary[1:]
   409  	}
   410  	return bytes.Equal(rest, mr.nl)
   411  }
   412  
   413  // skipLWSPChar returns b with leading spaces and tabs removed.
   414  // RFC 822 defines:
   415  //    LWSP-char = SPACE / HTAB
   416  func skipLWSPChar(b []byte) []byte {
   417  	for len(b) > 0 && (b[0] == ' ' || b[0] == '\t') {
   418  		b = b[1:]
   419  	}
   420  	return b
   421  }
   422  

View as plain text