...
Run Format

Source file src/mime/multipart/multipart.go

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

View as plain text