...
Run Format

Source file src/archive/tar/reader.go

Documentation: archive/tar

     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 tar
     6  
     7  import (
     8  	"bytes"
     9  	"io"
    10  	"io/ioutil"
    11  	"strconv"
    12  	"strings"
    13  	"time"
    14  )
    15  
    16  // Reader provides sequential access to the contents of a tar archive.
    17  // Reader.Next advances to the next file in the archive (including the first),
    18  // and then Reader can be treated as an io.Reader to access the file's data.
    19  type Reader struct {
    20  	r    io.Reader
    21  	pad  int64      // Amount of padding (ignored) after current file entry
    22  	curr fileReader // Reader for current file entry
    23  	blk  block      // Buffer to use as temporary local storage
    24  
    25  	// err is a persistent error.
    26  	// It is only the responsibility of every exported method of Reader to
    27  	// ensure that this error is sticky.
    28  	err error
    29  }
    30  
    31  type fileReader interface {
    32  	io.Reader
    33  	fileState
    34  
    35  	WriteTo(io.Writer) (int64, error)
    36  }
    37  
    38  // NewReader creates a new Reader reading from r.
    39  func NewReader(r io.Reader) *Reader {
    40  	return &Reader{r: r, curr: &regFileReader{r, 0}}
    41  }
    42  
    43  // Next advances to the next entry in the tar archive.
    44  // The Header.Size determines how many bytes can be read for the next file.
    45  // Any remaining data in the current file is automatically discarded.
    46  //
    47  // io.EOF is returned at the end of the input.
    48  func (tr *Reader) Next() (*Header, error) {
    49  	if tr.err != nil {
    50  		return nil, tr.err
    51  	}
    52  	hdr, err := tr.next()
    53  	tr.err = err
    54  	return hdr, err
    55  }
    56  
    57  func (tr *Reader) next() (*Header, error) {
    58  	var paxHdrs map[string]string
    59  	var gnuLongName, gnuLongLink string
    60  
    61  	// Externally, Next iterates through the tar archive as if it is a series of
    62  	// files. Internally, the tar format often uses fake "files" to add meta
    63  	// data that describes the next file. These meta data "files" should not
    64  	// normally be visible to the outside. As such, this loop iterates through
    65  	// one or more "header files" until it finds a "normal file".
    66  	format := FormatUSTAR | FormatPAX | FormatGNU
    67  loop:
    68  	for {
    69  		// Discard the remainder of the file and any padding.
    70  		if err := discard(tr.r, tr.curr.PhysicalRemaining()); err != nil {
    71  			return nil, err
    72  		}
    73  		if _, err := tryReadFull(tr.r, tr.blk[:tr.pad]); err != nil {
    74  			return nil, err
    75  		}
    76  		tr.pad = 0
    77  
    78  		hdr, rawHdr, err := tr.readHeader()
    79  		if err != nil {
    80  			return nil, err
    81  		}
    82  		if err := tr.handleRegularFile(hdr); err != nil {
    83  			return nil, err
    84  		}
    85  		format.mayOnlyBe(hdr.Format)
    86  
    87  		// Check for PAX/GNU special headers and files.
    88  		switch hdr.Typeflag {
    89  		case TypeXHeader, TypeXGlobalHeader:
    90  			format.mayOnlyBe(FormatPAX)
    91  			paxHdrs, err = parsePAX(tr)
    92  			if err != nil {
    93  				return nil, err
    94  			}
    95  			if hdr.Typeflag == TypeXGlobalHeader {
    96  				mergePAX(hdr, paxHdrs)
    97  				return &Header{
    98  					Name:       hdr.Name,
    99  					Typeflag:   hdr.Typeflag,
   100  					Xattrs:     hdr.Xattrs,
   101  					PAXRecords: hdr.PAXRecords,
   102  					Format:     format,
   103  				}, nil
   104  			}
   105  			continue loop // This is a meta header affecting the next header
   106  		case TypeGNULongName, TypeGNULongLink:
   107  			format.mayOnlyBe(FormatGNU)
   108  			realname, err := ioutil.ReadAll(tr)
   109  			if err != nil {
   110  				return nil, err
   111  			}
   112  
   113  			var p parser
   114  			switch hdr.Typeflag {
   115  			case TypeGNULongName:
   116  				gnuLongName = p.parseString(realname)
   117  			case TypeGNULongLink:
   118  				gnuLongLink = p.parseString(realname)
   119  			}
   120  			continue loop // This is a meta header affecting the next header
   121  		default:
   122  			// The old GNU sparse format is handled here since it is technically
   123  			// just a regular file with additional attributes.
   124  
   125  			if err := mergePAX(hdr, paxHdrs); err != nil {
   126  				return nil, err
   127  			}
   128  			if gnuLongName != "" {
   129  				hdr.Name = gnuLongName
   130  			}
   131  			if gnuLongLink != "" {
   132  				hdr.Linkname = gnuLongLink
   133  			}
   134  			if hdr.Typeflag == TypeRegA && strings.HasSuffix(hdr.Name, "/") {
   135  				hdr.Typeflag = TypeDir // Legacy archives use trailing slash for directories
   136  			}
   137  
   138  			// The extended headers may have updated the size.
   139  			// Thus, setup the regFileReader again after merging PAX headers.
   140  			if err := tr.handleRegularFile(hdr); err != nil {
   141  				return nil, err
   142  			}
   143  
   144  			// Sparse formats rely on being able to read from the logical data
   145  			// section; there must be a preceding call to handleRegularFile.
   146  			if err := tr.handleSparseFile(hdr, rawHdr); err != nil {
   147  				return nil, err
   148  			}
   149  
   150  			// Set the final guess at the format.
   151  			if format.has(FormatUSTAR) && format.has(FormatPAX) {
   152  				format.mayOnlyBe(FormatUSTAR)
   153  			}
   154  			hdr.Format = format
   155  			return hdr, nil // This is a file, so stop
   156  		}
   157  	}
   158  }
   159  
   160  // handleRegularFile sets up the current file reader and padding such that it
   161  // can only read the following logical data section. It will properly handle
   162  // special headers that contain no data section.
   163  func (tr *Reader) handleRegularFile(hdr *Header) error {
   164  	nb := hdr.Size
   165  	if isHeaderOnlyType(hdr.Typeflag) {
   166  		nb = 0
   167  	}
   168  	if nb < 0 {
   169  		return ErrHeader
   170  	}
   171  
   172  	tr.pad = blockPadding(nb)
   173  	tr.curr = &regFileReader{r: tr.r, nb: nb}
   174  	return nil
   175  }
   176  
   177  // handleSparseFile checks if the current file is a sparse format of any type
   178  // and sets the curr reader appropriately.
   179  func (tr *Reader) handleSparseFile(hdr *Header, rawHdr *block) error {
   180  	var spd sparseDatas
   181  	var err error
   182  	if hdr.Typeflag == TypeGNUSparse {
   183  		spd, err = tr.readOldGNUSparseMap(hdr, rawHdr)
   184  	} else {
   185  		spd, err = tr.readGNUSparsePAXHeaders(hdr)
   186  	}
   187  
   188  	// If sp is non-nil, then this is a sparse file.
   189  	// Note that it is possible for len(sp) == 0.
   190  	if err == nil && spd != nil {
   191  		if isHeaderOnlyType(hdr.Typeflag) || !validateSparseEntries(spd, hdr.Size) {
   192  			return ErrHeader
   193  		}
   194  		sph := invertSparseEntries(spd, hdr.Size)
   195  		tr.curr = &sparseFileReader{tr.curr, sph, 0}
   196  	}
   197  	return err
   198  }
   199  
   200  // readGNUSparsePAXHeaders checks the PAX headers for GNU sparse headers.
   201  // If they are found, then this function reads the sparse map and returns it.
   202  // This assumes that 0.0 headers have already been converted to 0.1 headers
   203  // by the the PAX header parsing logic.
   204  func (tr *Reader) readGNUSparsePAXHeaders(hdr *Header) (sparseDatas, error) {
   205  	// Identify the version of GNU headers.
   206  	var is1x0 bool
   207  	major, minor := hdr.PAXRecords[paxGNUSparseMajor], hdr.PAXRecords[paxGNUSparseMinor]
   208  	switch {
   209  	case major == "0" && (minor == "0" || minor == "1"):
   210  		is1x0 = false
   211  	case major == "1" && minor == "0":
   212  		is1x0 = true
   213  	case major != "" || minor != "":
   214  		return nil, nil // Unknown GNU sparse PAX version
   215  	case hdr.PAXRecords[paxGNUSparseMap] != "":
   216  		is1x0 = false // 0.0 and 0.1 did not have explicit version records, so guess
   217  	default:
   218  		return nil, nil // Not a PAX format GNU sparse file.
   219  	}
   220  	hdr.Format.mayOnlyBe(FormatPAX)
   221  
   222  	// Update hdr from GNU sparse PAX headers.
   223  	if name := hdr.PAXRecords[paxGNUSparseName]; name != "" {
   224  		hdr.Name = name
   225  	}
   226  	size := hdr.PAXRecords[paxGNUSparseSize]
   227  	if size == "" {
   228  		size = hdr.PAXRecords[paxGNUSparseRealSize]
   229  	}
   230  	if size != "" {
   231  		n, err := strconv.ParseInt(size, 10, 64)
   232  		if err != nil {
   233  			return nil, ErrHeader
   234  		}
   235  		hdr.Size = n
   236  	}
   237  
   238  	// Read the sparse map according to the appropriate format.
   239  	if is1x0 {
   240  		return readGNUSparseMap1x0(tr.curr)
   241  	}
   242  	return readGNUSparseMap0x1(hdr.PAXRecords)
   243  }
   244  
   245  // mergePAX merges paxHdrs into hdr for all relevant fields of Header.
   246  func mergePAX(hdr *Header, paxHdrs map[string]string) (err error) {
   247  	for k, v := range paxHdrs {
   248  		if v == "" {
   249  			continue // Keep the original USTAR value
   250  		}
   251  		var id64 int64
   252  		switch k {
   253  		case paxPath:
   254  			hdr.Name = v
   255  		case paxLinkpath:
   256  			hdr.Linkname = v
   257  		case paxUname:
   258  			hdr.Uname = v
   259  		case paxGname:
   260  			hdr.Gname = v
   261  		case paxUid:
   262  			id64, err = strconv.ParseInt(v, 10, 64)
   263  			hdr.Uid = int(id64) // Integer overflow possible
   264  		case paxGid:
   265  			id64, err = strconv.ParseInt(v, 10, 64)
   266  			hdr.Gid = int(id64) // Integer overflow possible
   267  		case paxAtime:
   268  			hdr.AccessTime, err = parsePAXTime(v)
   269  		case paxMtime:
   270  			hdr.ModTime, err = parsePAXTime(v)
   271  		case paxCtime:
   272  			hdr.ChangeTime, err = parsePAXTime(v)
   273  		case paxSize:
   274  			hdr.Size, err = strconv.ParseInt(v, 10, 64)
   275  		default:
   276  			if strings.HasPrefix(k, paxSchilyXattr) {
   277  				if hdr.Xattrs == nil {
   278  					hdr.Xattrs = make(map[string]string)
   279  				}
   280  				hdr.Xattrs[k[len(paxSchilyXattr):]] = v
   281  			}
   282  		}
   283  		if err != nil {
   284  			return ErrHeader
   285  		}
   286  	}
   287  	hdr.PAXRecords = paxHdrs
   288  	return nil
   289  }
   290  
   291  // parsePAX parses PAX headers.
   292  // If an extended header (type 'x') is invalid, ErrHeader is returned
   293  func parsePAX(r io.Reader) (map[string]string, error) {
   294  	buf, err := ioutil.ReadAll(r)
   295  	if err != nil {
   296  		return nil, err
   297  	}
   298  	sbuf := string(buf)
   299  
   300  	// For GNU PAX sparse format 0.0 support.
   301  	// This function transforms the sparse format 0.0 headers into format 0.1
   302  	// headers since 0.0 headers were not PAX compliant.
   303  	var sparseMap []string
   304  
   305  	paxHdrs := make(map[string]string)
   306  	for len(sbuf) > 0 {
   307  		key, value, residual, err := parsePAXRecord(sbuf)
   308  		if err != nil {
   309  			return nil, ErrHeader
   310  		}
   311  		sbuf = residual
   312  
   313  		switch key {
   314  		case paxGNUSparseOffset, paxGNUSparseNumBytes:
   315  			// Validate sparse header order and value.
   316  			if (len(sparseMap)%2 == 0 && key != paxGNUSparseOffset) ||
   317  				(len(sparseMap)%2 == 1 && key != paxGNUSparseNumBytes) ||
   318  				strings.Contains(value, ",") {
   319  				return nil, ErrHeader
   320  			}
   321  			sparseMap = append(sparseMap, value)
   322  		default:
   323  			paxHdrs[key] = value
   324  		}
   325  	}
   326  	if len(sparseMap) > 0 {
   327  		paxHdrs[paxGNUSparseMap] = strings.Join(sparseMap, ",")
   328  	}
   329  	return paxHdrs, nil
   330  }
   331  
   332  // readHeader reads the next block header and assumes that the underlying reader
   333  // is already aligned to a block boundary. It returns the raw block of the
   334  // header in case further processing is required.
   335  //
   336  // The err will be set to io.EOF only when one of the following occurs:
   337  //	* Exactly 0 bytes are read and EOF is hit.
   338  //	* Exactly 1 block of zeros is read and EOF is hit.
   339  //	* At least 2 blocks of zeros are read.
   340  func (tr *Reader) readHeader() (*Header, *block, error) {
   341  	// Two blocks of zero bytes marks the end of the archive.
   342  	if _, err := io.ReadFull(tr.r, tr.blk[:]); err != nil {
   343  		return nil, nil, err // EOF is okay here; exactly 0 bytes read
   344  	}
   345  	if bytes.Equal(tr.blk[:], zeroBlock[:]) {
   346  		if _, err := io.ReadFull(tr.r, tr.blk[:]); err != nil {
   347  			return nil, nil, err // EOF is okay here; exactly 1 block of zeros read
   348  		}
   349  		if bytes.Equal(tr.blk[:], zeroBlock[:]) {
   350  			return nil, nil, io.EOF // normal EOF; exactly 2 block of zeros read
   351  		}
   352  		return nil, nil, ErrHeader // Zero block and then non-zero block
   353  	}
   354  
   355  	// Verify the header matches a known format.
   356  	format := tr.blk.GetFormat()
   357  	if format == FormatUnknown {
   358  		return nil, nil, ErrHeader
   359  	}
   360  
   361  	var p parser
   362  	hdr := new(Header)
   363  
   364  	// Unpack the V7 header.
   365  	v7 := tr.blk.V7()
   366  	hdr.Typeflag = v7.TypeFlag()[0]
   367  	hdr.Name = p.parseString(v7.Name())
   368  	hdr.Linkname = p.parseString(v7.LinkName())
   369  	hdr.Size = p.parseNumeric(v7.Size())
   370  	hdr.Mode = p.parseNumeric(v7.Mode())
   371  	hdr.Uid = int(p.parseNumeric(v7.UID()))
   372  	hdr.Gid = int(p.parseNumeric(v7.GID()))
   373  	hdr.ModTime = time.Unix(p.parseNumeric(v7.ModTime()), 0)
   374  
   375  	// Unpack format specific fields.
   376  	if format > formatV7 {
   377  		ustar := tr.blk.USTAR()
   378  		hdr.Uname = p.parseString(ustar.UserName())
   379  		hdr.Gname = p.parseString(ustar.GroupName())
   380  		hdr.Devmajor = p.parseNumeric(ustar.DevMajor())
   381  		hdr.Devminor = p.parseNumeric(ustar.DevMinor())
   382  
   383  		var prefix string
   384  		switch {
   385  		case format.has(FormatUSTAR | FormatPAX):
   386  			hdr.Format = format
   387  			ustar := tr.blk.USTAR()
   388  			prefix = p.parseString(ustar.Prefix())
   389  
   390  			// For Format detection, check if block is properly formatted since
   391  			// the parser is more liberal than what USTAR actually permits.
   392  			notASCII := func(r rune) bool { return r >= 0x80 }
   393  			if bytes.IndexFunc(tr.blk[:], notASCII) >= 0 {
   394  				hdr.Format = FormatUnknown // Non-ASCII characters in block.
   395  			}
   396  			nul := func(b []byte) bool { return int(b[len(b)-1]) == 0 }
   397  			if !(nul(v7.Size()) && nul(v7.Mode()) && nul(v7.UID()) && nul(v7.GID()) &&
   398  				nul(v7.ModTime()) && nul(ustar.DevMajor()) && nul(ustar.DevMinor())) {
   399  				hdr.Format = FormatUnknown // Numeric fields must end in NUL
   400  			}
   401  		case format.has(formatSTAR):
   402  			star := tr.blk.STAR()
   403  			prefix = p.parseString(star.Prefix())
   404  			hdr.AccessTime = time.Unix(p.parseNumeric(star.AccessTime()), 0)
   405  			hdr.ChangeTime = time.Unix(p.parseNumeric(star.ChangeTime()), 0)
   406  		case format.has(FormatGNU):
   407  			hdr.Format = format
   408  			var p2 parser
   409  			gnu := tr.blk.GNU()
   410  			if b := gnu.AccessTime(); b[0] != 0 {
   411  				hdr.AccessTime = time.Unix(p2.parseNumeric(b), 0)
   412  			}
   413  			if b := gnu.ChangeTime(); b[0] != 0 {
   414  				hdr.ChangeTime = time.Unix(p2.parseNumeric(b), 0)
   415  			}
   416  
   417  			// Prior to Go1.8, the Writer had a bug where it would output
   418  			// an invalid tar file in certain rare situations because the logic
   419  			// incorrectly believed that the old GNU format had a prefix field.
   420  			// This is wrong and leads to an output file that mangles the
   421  			// atime and ctime fields, which are often left unused.
   422  			//
   423  			// In order to continue reading tar files created by former, buggy
   424  			// versions of Go, we skeptically parse the atime and ctime fields.
   425  			// If we are unable to parse them and the prefix field looks like
   426  			// an ASCII string, then we fallback on the pre-Go1.8 behavior
   427  			// of treating these fields as the USTAR prefix field.
   428  			//
   429  			// Note that this will not use the fallback logic for all possible
   430  			// files generated by a pre-Go1.8 toolchain. If the generated file
   431  			// happened to have a prefix field that parses as valid
   432  			// atime and ctime fields (e.g., when they are valid octal strings),
   433  			// then it is impossible to distinguish between an valid GNU file
   434  			// and an invalid pre-Go1.8 file.
   435  			//
   436  			// See https://golang.org/issues/12594
   437  			// See https://golang.org/issues/21005
   438  			if p2.err != nil {
   439  				hdr.AccessTime, hdr.ChangeTime = time.Time{}, time.Time{}
   440  				ustar := tr.blk.USTAR()
   441  				if s := p.parseString(ustar.Prefix()); isASCII(s) {
   442  					prefix = s
   443  				}
   444  				hdr.Format = FormatUnknown // Buggy file is not GNU
   445  			}
   446  		}
   447  		if len(prefix) > 0 {
   448  			hdr.Name = prefix + "/" + hdr.Name
   449  		}
   450  	}
   451  	return hdr, &tr.blk, p.err
   452  }
   453  
   454  // readOldGNUSparseMap reads the sparse map from the old GNU sparse format.
   455  // The sparse map is stored in the tar header if it's small enough.
   456  // If it's larger than four entries, then one or more extension headers are used
   457  // to store the rest of the sparse map.
   458  //
   459  // The Header.Size does not reflect the size of any extended headers used.
   460  // Thus, this function will read from the raw io.Reader to fetch extra headers.
   461  // This method mutates blk in the process.
   462  func (tr *Reader) readOldGNUSparseMap(hdr *Header, blk *block) (sparseDatas, error) {
   463  	// Make sure that the input format is GNU.
   464  	// Unfortunately, the STAR format also has a sparse header format that uses
   465  	// the same type flag but has a completely different layout.
   466  	if blk.GetFormat() != FormatGNU {
   467  		return nil, ErrHeader
   468  	}
   469  	hdr.Format.mayOnlyBe(FormatGNU)
   470  
   471  	var p parser
   472  	hdr.Size = p.parseNumeric(blk.GNU().RealSize())
   473  	if p.err != nil {
   474  		return nil, p.err
   475  	}
   476  	s := blk.GNU().Sparse()
   477  	spd := make(sparseDatas, 0, s.MaxEntries())
   478  	for {
   479  		for i := 0; i < s.MaxEntries(); i++ {
   480  			// This termination condition is identical to GNU and BSD tar.
   481  			if s.Entry(i).Offset()[0] == 0x00 {
   482  				break // Don't return, need to process extended headers (even if empty)
   483  			}
   484  			offset := p.parseNumeric(s.Entry(i).Offset())
   485  			length := p.parseNumeric(s.Entry(i).Length())
   486  			if p.err != nil {
   487  				return nil, p.err
   488  			}
   489  			spd = append(spd, sparseEntry{Offset: offset, Length: length})
   490  		}
   491  
   492  		if s.IsExtended()[0] > 0 {
   493  			// There are more entries. Read an extension header and parse its entries.
   494  			if _, err := mustReadFull(tr.r, blk[:]); err != nil {
   495  				return nil, err
   496  			}
   497  			s = blk.Sparse()
   498  			continue
   499  		}
   500  		return spd, nil // Done
   501  	}
   502  }
   503  
   504  // readGNUSparseMap1x0 reads the sparse map as stored in GNU's PAX sparse format
   505  // version 1.0. The format of the sparse map consists of a series of
   506  // newline-terminated numeric fields. The first field is the number of entries
   507  // and is always present. Following this are the entries, consisting of two
   508  // fields (offset, length). This function must stop reading at the end
   509  // boundary of the block containing the last newline.
   510  //
   511  // Note that the GNU manual says that numeric values should be encoded in octal
   512  // format. However, the GNU tar utility itself outputs these values in decimal.
   513  // As such, this library treats values as being encoded in decimal.
   514  func readGNUSparseMap1x0(r io.Reader) (sparseDatas, error) {
   515  	var (
   516  		cntNewline int64
   517  		buf        bytes.Buffer
   518  		blk        block
   519  	)
   520  
   521  	// feedTokens copies data in blocks from r into buf until there are
   522  	// at least cnt newlines in buf. It will not read more blocks than needed.
   523  	feedTokens := func(n int64) error {
   524  		for cntNewline < n {
   525  			if _, err := mustReadFull(r, blk[:]); err != nil {
   526  				return err
   527  			}
   528  			buf.Write(blk[:])
   529  			for _, c := range blk {
   530  				if c == '\n' {
   531  					cntNewline++
   532  				}
   533  			}
   534  		}
   535  		return nil
   536  	}
   537  
   538  	// nextToken gets the next token delimited by a newline. This assumes that
   539  	// at least one newline exists in the buffer.
   540  	nextToken := func() string {
   541  		cntNewline--
   542  		tok, _ := buf.ReadString('\n')
   543  		return strings.TrimRight(tok, "\n")
   544  	}
   545  
   546  	// Parse for the number of entries.
   547  	// Use integer overflow resistant math to check this.
   548  	if err := feedTokens(1); err != nil {
   549  		return nil, err
   550  	}
   551  	numEntries, err := strconv.ParseInt(nextToken(), 10, 0) // Intentionally parse as native int
   552  	if err != nil || numEntries < 0 || int(2*numEntries) < int(numEntries) {
   553  		return nil, ErrHeader
   554  	}
   555  
   556  	// Parse for all member entries.
   557  	// numEntries is trusted after this since a potential attacker must have
   558  	// committed resources proportional to what this library used.
   559  	if err := feedTokens(2 * numEntries); err != nil {
   560  		return nil, err
   561  	}
   562  	spd := make(sparseDatas, 0, numEntries)
   563  	for i := int64(0); i < numEntries; i++ {
   564  		offset, err1 := strconv.ParseInt(nextToken(), 10, 64)
   565  		length, err2 := strconv.ParseInt(nextToken(), 10, 64)
   566  		if err1 != nil || err2 != nil {
   567  			return nil, ErrHeader
   568  		}
   569  		spd = append(spd, sparseEntry{Offset: offset, Length: length})
   570  	}
   571  	return spd, nil
   572  }
   573  
   574  // readGNUSparseMap0x1 reads the sparse map as stored in GNU's PAX sparse format
   575  // version 0.1. The sparse map is stored in the PAX headers.
   576  func readGNUSparseMap0x1(paxHdrs map[string]string) (sparseDatas, error) {
   577  	// Get number of entries.
   578  	// Use integer overflow resistant math to check this.
   579  	numEntriesStr := paxHdrs[paxGNUSparseNumBlocks]
   580  	numEntries, err := strconv.ParseInt(numEntriesStr, 10, 0) // Intentionally parse as native int
   581  	if err != nil || numEntries < 0 || int(2*numEntries) < int(numEntries) {
   582  		return nil, ErrHeader
   583  	}
   584  
   585  	// There should be two numbers in sparseMap for each entry.
   586  	sparseMap := strings.Split(paxHdrs[paxGNUSparseMap], ",")
   587  	if len(sparseMap) == 1 && sparseMap[0] == "" {
   588  		sparseMap = sparseMap[:0]
   589  	}
   590  	if int64(len(sparseMap)) != 2*numEntries {
   591  		return nil, ErrHeader
   592  	}
   593  
   594  	// Loop through the entries in the sparse map.
   595  	// numEntries is trusted now.
   596  	spd := make(sparseDatas, 0, numEntries)
   597  	for len(sparseMap) >= 2 {
   598  		offset, err1 := strconv.ParseInt(sparseMap[0], 10, 64)
   599  		length, err2 := strconv.ParseInt(sparseMap[1], 10, 64)
   600  		if err1 != nil || err2 != nil {
   601  			return nil, ErrHeader
   602  		}
   603  		spd = append(spd, sparseEntry{Offset: offset, Length: length})
   604  		sparseMap = sparseMap[2:]
   605  	}
   606  	return spd, nil
   607  }
   608  
   609  // Read reads from the current file in the tar archive.
   610  // It returns (0, io.EOF) when it reaches the end of that file,
   611  // until Next is called to advance to the next file.
   612  //
   613  // If the current file is sparse, then the regions marked as a hole
   614  // are read back as NUL-bytes.
   615  //
   616  // Calling Read on special types like TypeLink, TypeSymlink, TypeChar,
   617  // TypeBlock, TypeDir, and TypeFifo returns (0, io.EOF) regardless of what
   618  // the Header.Size claims.
   619  func (tr *Reader) Read(b []byte) (int, error) {
   620  	if tr.err != nil {
   621  		return 0, tr.err
   622  	}
   623  	n, err := tr.curr.Read(b)
   624  	if err != nil && err != io.EOF {
   625  		tr.err = err
   626  	}
   627  	return n, err
   628  }
   629  
   630  // writeTo writes the content of the current file to w.
   631  // The bytes written matches the number of remaining bytes in the current file.
   632  //
   633  // If the current file is sparse and w is an io.WriteSeeker,
   634  // then writeTo uses Seek to skip past holes defined in Header.SparseHoles,
   635  // assuming that skipped regions are filled with NULs.
   636  // This always writes the last byte to ensure w is the right size.
   637  //
   638  // TODO(dsnet): Re-export this when adding sparse file support.
   639  // See https://golang.org/issue/22735
   640  func (tr *Reader) writeTo(w io.Writer) (int64, error) {
   641  	if tr.err != nil {
   642  		return 0, tr.err
   643  	}
   644  	n, err := tr.curr.WriteTo(w)
   645  	if err != nil {
   646  		tr.err = err
   647  	}
   648  	return n, err
   649  }
   650  
   651  // regFileReader is a fileReader for reading data from a regular file entry.
   652  type regFileReader struct {
   653  	r  io.Reader // Underlying Reader
   654  	nb int64     // Number of remaining bytes to read
   655  }
   656  
   657  func (fr *regFileReader) Read(b []byte) (n int, err error) {
   658  	if int64(len(b)) > fr.nb {
   659  		b = b[:fr.nb]
   660  	}
   661  	if len(b) > 0 {
   662  		n, err = fr.r.Read(b)
   663  		fr.nb -= int64(n)
   664  	}
   665  	switch {
   666  	case err == io.EOF && fr.nb > 0:
   667  		return n, io.ErrUnexpectedEOF
   668  	case err == nil && fr.nb == 0:
   669  		return n, io.EOF
   670  	default:
   671  		return n, err
   672  	}
   673  }
   674  
   675  func (fr *regFileReader) WriteTo(w io.Writer) (int64, error) {
   676  	return io.Copy(w, struct{ io.Reader }{fr})
   677  }
   678  
   679  func (fr regFileReader) LogicalRemaining() int64 {
   680  	return fr.nb
   681  }
   682  
   683  func (fr regFileReader) PhysicalRemaining() int64 {
   684  	return fr.nb
   685  }
   686  
   687  // sparseFileReader is a fileReader for reading data from a sparse file entry.
   688  type sparseFileReader struct {
   689  	fr  fileReader  // Underlying fileReader
   690  	sp  sparseHoles // Normalized list of sparse holes
   691  	pos int64       // Current position in sparse file
   692  }
   693  
   694  func (sr *sparseFileReader) Read(b []byte) (n int, err error) {
   695  	finished := int64(len(b)) >= sr.LogicalRemaining()
   696  	if finished {
   697  		b = b[:sr.LogicalRemaining()]
   698  	}
   699  
   700  	b0 := b
   701  	endPos := sr.pos + int64(len(b))
   702  	for endPos > sr.pos && err == nil {
   703  		var nf int // Bytes read in fragment
   704  		holeStart, holeEnd := sr.sp[0].Offset, sr.sp[0].endOffset()
   705  		if sr.pos < holeStart { // In a data fragment
   706  			bf := b[:min(int64(len(b)), holeStart-sr.pos)]
   707  			nf, err = tryReadFull(sr.fr, bf)
   708  		} else { // In a hole fragment
   709  			bf := b[:min(int64(len(b)), holeEnd-sr.pos)]
   710  			nf, err = tryReadFull(zeroReader{}, bf)
   711  		}
   712  		b = b[nf:]
   713  		sr.pos += int64(nf)
   714  		if sr.pos >= holeEnd && len(sr.sp) > 1 {
   715  			sr.sp = sr.sp[1:] // Ensure last fragment always remains
   716  		}
   717  	}
   718  
   719  	n = len(b0) - len(b)
   720  	switch {
   721  	case err == io.EOF:
   722  		return n, errMissData // Less data in dense file than sparse file
   723  	case err != nil:
   724  		return n, err
   725  	case sr.LogicalRemaining() == 0 && sr.PhysicalRemaining() > 0:
   726  		return n, errUnrefData // More data in dense file than sparse file
   727  	case finished:
   728  		return n, io.EOF
   729  	default:
   730  		return n, nil
   731  	}
   732  }
   733  
   734  func (sr *sparseFileReader) WriteTo(w io.Writer) (n int64, err error) {
   735  	ws, ok := w.(io.WriteSeeker)
   736  	if ok {
   737  		if _, err := ws.Seek(0, io.SeekCurrent); err != nil {
   738  			ok = false // Not all io.Seeker can really seek
   739  		}
   740  	}
   741  	if !ok {
   742  		return io.Copy(w, struct{ io.Reader }{sr})
   743  	}
   744  
   745  	var writeLastByte bool
   746  	pos0 := sr.pos
   747  	for sr.LogicalRemaining() > 0 && !writeLastByte && err == nil {
   748  		var nf int64 // Size of fragment
   749  		holeStart, holeEnd := sr.sp[0].Offset, sr.sp[0].endOffset()
   750  		if sr.pos < holeStart { // In a data fragment
   751  			nf = holeStart - sr.pos
   752  			nf, err = io.CopyN(ws, sr.fr, nf)
   753  		} else { // In a hole fragment
   754  			nf = holeEnd - sr.pos
   755  			if sr.PhysicalRemaining() == 0 {
   756  				writeLastByte = true
   757  				nf--
   758  			}
   759  			_, err = ws.Seek(nf, io.SeekCurrent)
   760  		}
   761  		sr.pos += nf
   762  		if sr.pos >= holeEnd && len(sr.sp) > 1 {
   763  			sr.sp = sr.sp[1:] // Ensure last fragment always remains
   764  		}
   765  	}
   766  
   767  	// If the last fragment is a hole, then seek to 1-byte before EOF, and
   768  	// write a single byte to ensure the file is the right size.
   769  	if writeLastByte && err == nil {
   770  		_, err = ws.Write([]byte{0})
   771  		sr.pos++
   772  	}
   773  
   774  	n = sr.pos - pos0
   775  	switch {
   776  	case err == io.EOF:
   777  		return n, errMissData // Less data in dense file than sparse file
   778  	case err != nil:
   779  		return n, err
   780  	case sr.LogicalRemaining() == 0 && sr.PhysicalRemaining() > 0:
   781  		return n, errUnrefData // More data in dense file than sparse file
   782  	default:
   783  		return n, nil
   784  	}
   785  }
   786  
   787  func (sr sparseFileReader) LogicalRemaining() int64 {
   788  	return sr.sp[len(sr.sp)-1].endOffset() - sr.pos
   789  }
   790  func (sr sparseFileReader) PhysicalRemaining() int64 {
   791  	return sr.fr.PhysicalRemaining()
   792  }
   793  
   794  type zeroReader struct{}
   795  
   796  func (zeroReader) Read(b []byte) (int, error) {
   797  	for i := range b {
   798  		b[i] = 0
   799  	}
   800  	return len(b), nil
   801  }
   802  
   803  // mustReadFull is like io.ReadFull except it returns
   804  // io.ErrUnexpectedEOF when io.EOF is hit before len(b) bytes are read.
   805  func mustReadFull(r io.Reader, b []byte) (int, error) {
   806  	n, err := tryReadFull(r, b)
   807  	if err == io.EOF {
   808  		err = io.ErrUnexpectedEOF
   809  	}
   810  	return n, err
   811  }
   812  
   813  // tryReadFull is like io.ReadFull except it returns
   814  // io.EOF when it is hit before len(b) bytes are read.
   815  func tryReadFull(r io.Reader, b []byte) (n int, err error) {
   816  	for len(b) > n && err == nil {
   817  		var nn int
   818  		nn, err = r.Read(b[n:])
   819  		n += nn
   820  	}
   821  	if len(b) == n && err == io.EOF {
   822  		err = nil
   823  	}
   824  	return n, err
   825  }
   826  
   827  // discard skips n bytes in r, reporting an error if unable to do so.
   828  func discard(r io.Reader, n int64) error {
   829  	// If possible, Seek to the last byte before the end of the data section.
   830  	// Do this because Seek is often lazy about reporting errors; this will mask
   831  	// the fact that the stream may be truncated. We can rely on the
   832  	// io.CopyN done shortly afterwards to trigger any IO errors.
   833  	var seekSkipped int64 // Number of bytes skipped via Seek
   834  	if sr, ok := r.(io.Seeker); ok && n > 1 {
   835  		// Not all io.Seeker can actually Seek. For example, os.Stdin implements
   836  		// io.Seeker, but calling Seek always returns an error and performs
   837  		// no action. Thus, we try an innocent seek to the current position
   838  		// to see if Seek is really supported.
   839  		pos1, err := sr.Seek(0, io.SeekCurrent)
   840  		if pos1 >= 0 && err == nil {
   841  			// Seek seems supported, so perform the real Seek.
   842  			pos2, err := sr.Seek(n-1, io.SeekCurrent)
   843  			if pos2 < 0 || err != nil {
   844  				return err
   845  			}
   846  			seekSkipped = pos2 - pos1
   847  		}
   848  	}
   849  
   850  	copySkipped, err := io.CopyN(ioutil.Discard, r, n-seekSkipped)
   851  	if err == io.EOF && seekSkipped+copySkipped < n {
   852  		err = io.ErrUnexpectedEOF
   853  	}
   854  	return err
   855  }
   856  

View as plain text