Source file src/archive/zip/reader.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  package zip
     6  
     7  import (
     8  	"bufio"
     9  	"encoding/binary"
    10  	"errors"
    11  	"hash"
    12  	"hash/crc32"
    13  	"internal/godebug"
    14  	"io"
    15  	"io/fs"
    16  	"os"
    17  	"path"
    18  	"path/filepath"
    19  	"slices"
    20  	"strings"
    21  	"sync"
    22  	"time"
    23  )
    24  
    25  var zipinsecurepath = godebug.New("zipinsecurepath")
    26  
    27  var (
    28  	ErrFormat       = errors.New("zip: not a valid zip file")
    29  	ErrAlgorithm    = errors.New("zip: unsupported compression algorithm")
    30  	ErrChecksum     = errors.New("zip: checksum error")
    31  	ErrInsecurePath = errors.New("zip: insecure file path")
    32  )
    33  
    34  // A Reader serves content from a ZIP archive.
    35  type Reader struct {
    36  	r             io.ReaderAt
    37  	File          []*File
    38  	Comment       string
    39  	decompressors map[uint16]Decompressor
    40  
    41  	// Some JAR files are zip files with a prefix that is a bash script.
    42  	// The baseOffset field is the start of the zip file proper.
    43  	baseOffset int64
    44  
    45  	// fileList is a list of files sorted by ename,
    46  	// for use by the Open method.
    47  	fileListOnce sync.Once
    48  	fileList     []fileListEntry
    49  }
    50  
    51  // A ReadCloser is a [Reader] that must be closed when no longer needed.
    52  type ReadCloser struct {
    53  	f *os.File
    54  	Reader
    55  }
    56  
    57  // A File is a single file in a ZIP archive.
    58  // The file information is in the embedded [FileHeader].
    59  // The file content can be accessed by calling [File.Open].
    60  type File struct {
    61  	FileHeader
    62  	zip          *Reader
    63  	zipr         io.ReaderAt
    64  	headerOffset int64 // includes overall ZIP archive baseOffset
    65  	zip64        bool  // zip64 extended information extra field presence
    66  }
    67  
    68  // OpenReader will open the Zip file specified by name and return a ReadCloser.
    69  //
    70  // If any file inside the archive uses a non-local name
    71  // (as defined by [filepath.IsLocal]) or a name containing backslashes
    72  // and the GODEBUG environment variable contains `zipinsecurepath=0`,
    73  // OpenReader returns the reader with an ErrInsecurePath error.
    74  // A future version of Go may introduce this behavior by default.
    75  // Programs that want to accept non-local names can ignore
    76  // the ErrInsecurePath error and use the returned reader.
    77  func OpenReader(name string) (*ReadCloser, error) {
    78  	f, err := os.Open(name)
    79  	if err != nil {
    80  		return nil, err
    81  	}
    82  	fi, err := f.Stat()
    83  	if err != nil {
    84  		f.Close()
    85  		return nil, err
    86  	}
    87  	r := new(ReadCloser)
    88  	if err = r.init(f, fi.Size()); err != nil && err != ErrInsecurePath {
    89  		f.Close()
    90  		return nil, err
    91  	}
    92  	r.f = f
    93  	return r, err
    94  }
    95  
    96  // NewReader returns a new [Reader] reading from r, which is assumed to
    97  // have the given size in bytes.
    98  //
    99  // If any file inside the archive uses a non-local name
   100  // (as defined by [filepath.IsLocal]) or a name containing backslashes
   101  // and the GODEBUG environment variable contains `zipinsecurepath=0`,
   102  // NewReader returns the reader with an [ErrInsecurePath] error.
   103  // A future version of Go may introduce this behavior by default.
   104  // Programs that want to accept non-local names can ignore
   105  // the [ErrInsecurePath] error and use the returned reader.
   106  func NewReader(r io.ReaderAt, size int64) (*Reader, error) {
   107  	if size < 0 {
   108  		return nil, errors.New("zip: size cannot be negative")
   109  	}
   110  	zr := new(Reader)
   111  	var err error
   112  	if err = zr.init(r, size); err != nil && err != ErrInsecurePath {
   113  		return nil, err
   114  	}
   115  	return zr, err
   116  }
   117  
   118  func (r *Reader) init(rdr io.ReaderAt, size int64) error {
   119  	end, baseOffset, err := readDirectoryEnd(rdr, size)
   120  	if err != nil {
   121  		return err
   122  	}
   123  	r.r = rdr
   124  	r.baseOffset = baseOffset
   125  	// Since the number of directory records is not validated, it is not
   126  	// safe to preallocate r.File without first checking that the specified
   127  	// number of files is reasonable, since a malformed archive may
   128  	// indicate it contains up to 1 << 128 - 1 files. Since each file has a
   129  	// header which will be _at least_ 30 bytes we can safely preallocate
   130  	// if (data size / 30) >= end.directoryRecords.
   131  	if end.directorySize < uint64(size) && (uint64(size)-end.directorySize)/30 >= end.directoryRecords {
   132  		r.File = make([]*File, 0, end.directoryRecords)
   133  	}
   134  	r.Comment = end.comment
   135  	rs := io.NewSectionReader(rdr, 0, size)
   136  	if _, err = rs.Seek(r.baseOffset+int64(end.directoryOffset), io.SeekStart); err != nil {
   137  		return err
   138  	}
   139  	buf := bufio.NewReader(rs)
   140  
   141  	// The count of files inside a zip is truncated to fit in a uint16.
   142  	// Gloss over this by reading headers until we encounter
   143  	// a bad one, and then only report an ErrFormat or UnexpectedEOF if
   144  	// the file count modulo 65536 is incorrect.
   145  	for {
   146  		f := &File{zip: r, zipr: rdr}
   147  		err = readDirectoryHeader(f, buf)
   148  		if err == ErrFormat || err == io.ErrUnexpectedEOF {
   149  			break
   150  		}
   151  		if err != nil {
   152  			return err
   153  		}
   154  		f.headerOffset += r.baseOffset
   155  		r.File = append(r.File, f)
   156  	}
   157  	if uint16(len(r.File)) != uint16(end.directoryRecords) { // only compare 16 bits here
   158  		// Return the readDirectoryHeader error if we read
   159  		// the wrong number of directory entries.
   160  		return err
   161  	}
   162  	if zipinsecurepath.Value() == "0" {
   163  		for _, f := range r.File {
   164  			if f.Name == "" {
   165  				// Zip permits an empty file name field.
   166  				continue
   167  			}
   168  			// The zip specification states that names must use forward slashes,
   169  			// so consider any backslashes in the name insecure.
   170  			if !filepath.IsLocal(f.Name) || strings.Contains(f.Name, `\`) {
   171  				zipinsecurepath.IncNonDefault()
   172  				return ErrInsecurePath
   173  			}
   174  		}
   175  	}
   176  	return nil
   177  }
   178  
   179  // RegisterDecompressor registers or overrides a custom decompressor for a
   180  // specific method ID. If a decompressor for a given method is not found,
   181  // [Reader] will default to looking up the decompressor at the package level.
   182  func (r *Reader) RegisterDecompressor(method uint16, dcomp Decompressor) {
   183  	if r.decompressors == nil {
   184  		r.decompressors = make(map[uint16]Decompressor)
   185  	}
   186  	r.decompressors[method] = dcomp
   187  }
   188  
   189  func (r *Reader) decompressor(method uint16) Decompressor {
   190  	dcomp := r.decompressors[method]
   191  	if dcomp == nil {
   192  		dcomp = decompressor(method)
   193  	}
   194  	return dcomp
   195  }
   196  
   197  // Close closes the Zip file, rendering it unusable for I/O.
   198  func (rc *ReadCloser) Close() error {
   199  	return rc.f.Close()
   200  }
   201  
   202  // DataOffset returns the offset of the file's possibly-compressed
   203  // data, relative to the beginning of the zip file.
   204  //
   205  // Most callers should instead use [File.Open], which transparently
   206  // decompresses data and verifies checksums.
   207  func (f *File) DataOffset() (offset int64, err error) {
   208  	bodyOffset, err := f.findBodyOffset()
   209  	if err != nil {
   210  		return
   211  	}
   212  	return f.headerOffset + bodyOffset, nil
   213  }
   214  
   215  // Open returns a [ReadCloser] that provides access to the [File]'s contents.
   216  // Multiple files may be read concurrently.
   217  func (f *File) Open() (io.ReadCloser, error) {
   218  	bodyOffset, err := f.findBodyOffset()
   219  	if err != nil {
   220  		return nil, err
   221  	}
   222  	if strings.HasSuffix(f.Name, "/") {
   223  		// The ZIP specification (APPNOTE.TXT) specifies that directories, which
   224  		// are technically zero-byte files, must not have any associated file
   225  		// data. We previously tried failing here if f.CompressedSize64 != 0,
   226  		// but it turns out that a number of implementations (namely, the Java
   227  		// jar tool) don't properly set the storage method on directories
   228  		// resulting in a file with compressed size > 0 but uncompressed size ==
   229  		// 0. We still want to fail when a directory has associated uncompressed
   230  		// data, but we are tolerant of cases where the uncompressed size is
   231  		// zero but compressed size is not.
   232  		if f.UncompressedSize64 != 0 {
   233  			return &dirReader{ErrFormat}, nil
   234  		} else {
   235  			return &dirReader{io.EOF}, nil
   236  		}
   237  	}
   238  	size := int64(f.CompressedSize64)
   239  	r := io.NewSectionReader(f.zipr, f.headerOffset+bodyOffset, size)
   240  	dcomp := f.zip.decompressor(f.Method)
   241  	if dcomp == nil {
   242  		return nil, ErrAlgorithm
   243  	}
   244  	var rc io.ReadCloser = dcomp(r)
   245  	var desr io.Reader
   246  	if f.hasDataDescriptor() {
   247  		desr = io.NewSectionReader(f.zipr, f.headerOffset+bodyOffset+size, dataDescriptorLen)
   248  	}
   249  	rc = &checksumReader{
   250  		rc:   rc,
   251  		hash: crc32.NewIEEE(),
   252  		f:    f,
   253  		desr: desr,
   254  	}
   255  	return rc, nil
   256  }
   257  
   258  // OpenRaw returns a [Reader] that provides access to the [File]'s contents without
   259  // decompression.
   260  func (f *File) OpenRaw() (io.Reader, error) {
   261  	bodyOffset, err := f.findBodyOffset()
   262  	if err != nil {
   263  		return nil, err
   264  	}
   265  	r := io.NewSectionReader(f.zipr, f.headerOffset+bodyOffset, int64(f.CompressedSize64))
   266  	return r, nil
   267  }
   268  
   269  type dirReader struct {
   270  	err error
   271  }
   272  
   273  func (r *dirReader) Read([]byte) (int, error) {
   274  	return 0, r.err
   275  }
   276  
   277  func (r *dirReader) Close() error {
   278  	return nil
   279  }
   280  
   281  type checksumReader struct {
   282  	rc    io.ReadCloser
   283  	hash  hash.Hash32
   284  	nread uint64 // number of bytes read so far
   285  	f     *File
   286  	desr  io.Reader // if non-nil, where to read the data descriptor
   287  	err   error     // sticky error
   288  }
   289  
   290  func (r *checksumReader) Stat() (fs.FileInfo, error) {
   291  	return headerFileInfo{&r.f.FileHeader}, nil
   292  }
   293  
   294  func (r *checksumReader) Read(b []byte) (n int, err error) {
   295  	if r.err != nil {
   296  		return 0, r.err
   297  	}
   298  	n, err = r.rc.Read(b)
   299  	r.hash.Write(b[:n])
   300  	r.nread += uint64(n)
   301  	if r.nread > r.f.UncompressedSize64 {
   302  		return 0, ErrFormat
   303  	}
   304  	if err == nil {
   305  		return
   306  	}
   307  	if err == io.EOF {
   308  		if r.nread != r.f.UncompressedSize64 {
   309  			return 0, io.ErrUnexpectedEOF
   310  		}
   311  		if r.desr != nil {
   312  			if err1 := readDataDescriptor(r.desr, r.f); err1 != nil {
   313  				if err1 == io.EOF {
   314  					err = io.ErrUnexpectedEOF
   315  				} else {
   316  					err = err1
   317  				}
   318  			} else if r.hash.Sum32() != r.f.CRC32 {
   319  				err = ErrChecksum
   320  			}
   321  		} else {
   322  			// If there's not a data descriptor, we still compare
   323  			// the CRC32 of what we've read against the file header
   324  			// or TOC's CRC32, if it seems like it was set.
   325  			if r.f.CRC32 != 0 && r.hash.Sum32() != r.f.CRC32 {
   326  				err = ErrChecksum
   327  			}
   328  		}
   329  	}
   330  	r.err = err
   331  	return
   332  }
   333  
   334  func (r *checksumReader) Close() error { return r.rc.Close() }
   335  
   336  // findBodyOffset does the minimum work to verify the file has a header
   337  // and returns the file body offset.
   338  func (f *File) findBodyOffset() (int64, error) {
   339  	var buf [fileHeaderLen]byte
   340  	if _, err := f.zipr.ReadAt(buf[:], f.headerOffset); err != nil {
   341  		return 0, err
   342  	}
   343  	b := readBuf(buf[:])
   344  	if sig := b.uint32(); sig != fileHeaderSignature {
   345  		return 0, ErrFormat
   346  	}
   347  	b = b[22:] // skip over most of the header
   348  	filenameLen := int(b.uint16())
   349  	extraLen := int(b.uint16())
   350  	return int64(fileHeaderLen + filenameLen + extraLen), nil
   351  }
   352  
   353  // readDirectoryHeader attempts to read a directory header from r.
   354  // It returns io.ErrUnexpectedEOF if it cannot read a complete header,
   355  // and ErrFormat if it doesn't find a valid header signature.
   356  func readDirectoryHeader(f *File, r io.Reader) error {
   357  	var buf [directoryHeaderLen]byte
   358  	if _, err := io.ReadFull(r, buf[:]); err != nil {
   359  		return err
   360  	}
   361  	b := readBuf(buf[:])
   362  	if sig := b.uint32(); sig != directoryHeaderSignature {
   363  		return ErrFormat
   364  	}
   365  	f.CreatorVersion = b.uint16()
   366  	f.ReaderVersion = b.uint16()
   367  	f.Flags = b.uint16()
   368  	f.Method = b.uint16()
   369  	f.ModifiedTime = b.uint16()
   370  	f.ModifiedDate = b.uint16()
   371  	f.CRC32 = b.uint32()
   372  	f.CompressedSize = b.uint32()
   373  	f.UncompressedSize = b.uint32()
   374  	f.CompressedSize64 = uint64(f.CompressedSize)
   375  	f.UncompressedSize64 = uint64(f.UncompressedSize)
   376  	filenameLen := int(b.uint16())
   377  	extraLen := int(b.uint16())
   378  	commentLen := int(b.uint16())
   379  	b = b[4:] // skipped start disk number and internal attributes (2x uint16)
   380  	f.ExternalAttrs = b.uint32()
   381  	f.headerOffset = int64(b.uint32())
   382  	d := make([]byte, filenameLen+extraLen+commentLen)
   383  	if _, err := io.ReadFull(r, d); err != nil {
   384  		return err
   385  	}
   386  	f.Name = string(d[:filenameLen])
   387  	f.Extra = d[filenameLen : filenameLen+extraLen]
   388  	f.Comment = string(d[filenameLen+extraLen:])
   389  
   390  	// Determine the character encoding.
   391  	utf8Valid1, utf8Require1 := detectUTF8(f.Name)
   392  	utf8Valid2, utf8Require2 := detectUTF8(f.Comment)
   393  	switch {
   394  	case !utf8Valid1 || !utf8Valid2:
   395  		// Name and Comment definitely not UTF-8.
   396  		f.NonUTF8 = true
   397  	case !utf8Require1 && !utf8Require2:
   398  		// Name and Comment use only single-byte runes that overlap with UTF-8.
   399  		f.NonUTF8 = false
   400  	default:
   401  		// Might be UTF-8, might be some other encoding; preserve existing flag.
   402  		// Some ZIP writers use UTF-8 encoding without setting the UTF-8 flag.
   403  		// Since it is impossible to always distinguish valid UTF-8 from some
   404  		// other encoding (e.g., GBK or Shift-JIS), we trust the flag.
   405  		f.NonUTF8 = f.Flags&0x800 == 0
   406  	}
   407  
   408  	needUSize := f.UncompressedSize == ^uint32(0)
   409  	needCSize := f.CompressedSize == ^uint32(0)
   410  	needHeaderOffset := f.headerOffset == int64(^uint32(0))
   411  
   412  	// Best effort to find what we need.
   413  	// Other zip authors might not even follow the basic format,
   414  	// and we'll just ignore the Extra content in that case.
   415  	var modified time.Time
   416  parseExtras:
   417  	for extra := readBuf(f.Extra); len(extra) >= 4; { // need at least tag and size
   418  		fieldTag := extra.uint16()
   419  		fieldSize := int(extra.uint16())
   420  		if len(extra) < fieldSize {
   421  			break
   422  		}
   423  		fieldBuf := extra.sub(fieldSize)
   424  
   425  		switch fieldTag {
   426  		case zip64ExtraID:
   427  			f.zip64 = true
   428  
   429  			// update directory values from the zip64 extra block.
   430  			// They should only be consulted if the sizes read earlier
   431  			// are maxed out.
   432  			// See golang.org/issue/13367.
   433  			if needUSize {
   434  				needUSize = false
   435  				if len(fieldBuf) < 8 {
   436  					return ErrFormat
   437  				}
   438  				f.UncompressedSize64 = fieldBuf.uint64()
   439  			}
   440  			if needCSize {
   441  				needCSize = false
   442  				if len(fieldBuf) < 8 {
   443  					return ErrFormat
   444  				}
   445  				f.CompressedSize64 = fieldBuf.uint64()
   446  			}
   447  			if needHeaderOffset {
   448  				needHeaderOffset = false
   449  				if len(fieldBuf) < 8 {
   450  					return ErrFormat
   451  				}
   452  				f.headerOffset = int64(fieldBuf.uint64())
   453  			}
   454  		case ntfsExtraID:
   455  			if len(fieldBuf) < 4 {
   456  				continue parseExtras
   457  			}
   458  			fieldBuf.uint32()        // reserved (ignored)
   459  			for len(fieldBuf) >= 4 { // need at least tag and size
   460  				attrTag := fieldBuf.uint16()
   461  				attrSize := int(fieldBuf.uint16())
   462  				if len(fieldBuf) < attrSize {
   463  					continue parseExtras
   464  				}
   465  				attrBuf := fieldBuf.sub(attrSize)
   466  				if attrTag != 1 || attrSize != 24 {
   467  					continue // Ignore irrelevant attributes
   468  				}
   469  
   470  				const ticksPerSecond = 1e7    // Windows timestamp resolution
   471  				ts := int64(attrBuf.uint64()) // ModTime since Windows epoch
   472  				secs := ts / ticksPerSecond
   473  				nsecs := (1e9 / ticksPerSecond) * (ts % ticksPerSecond)
   474  				epoch := time.Date(1601, time.January, 1, 0, 0, 0, 0, time.UTC)
   475  				modified = time.Unix(epoch.Unix()+secs, nsecs)
   476  			}
   477  		case unixExtraID, infoZipUnixExtraID:
   478  			if len(fieldBuf) < 8 {
   479  				continue parseExtras
   480  			}
   481  			fieldBuf.uint32()              // AcTime (ignored)
   482  			ts := int64(fieldBuf.uint32()) // ModTime since Unix epoch
   483  			modified = time.Unix(ts, 0)
   484  		case extTimeExtraID:
   485  			if len(fieldBuf) < 5 || fieldBuf.uint8()&1 == 0 {
   486  				continue parseExtras
   487  			}
   488  			ts := int64(fieldBuf.uint32()) // ModTime since Unix epoch
   489  			modified = time.Unix(ts, 0)
   490  		}
   491  	}
   492  
   493  	msdosModified := msDosTimeToTime(f.ModifiedDate, f.ModifiedTime)
   494  	f.Modified = msdosModified
   495  	if !modified.IsZero() {
   496  		f.Modified = modified.UTC()
   497  
   498  		// If legacy MS-DOS timestamps are set, we can use the delta between
   499  		// the legacy and extended versions to estimate timezone offset.
   500  		//
   501  		// A non-UTC timezone is always used (even if offset is zero).
   502  		// Thus, FileHeader.Modified.Location() == time.UTC is useful for
   503  		// determining whether extended timestamps are present.
   504  		// This is necessary for users that need to do additional time
   505  		// calculations when dealing with legacy ZIP formats.
   506  		if f.ModifiedTime != 0 || f.ModifiedDate != 0 {
   507  			f.Modified = modified.In(timeZone(msdosModified.Sub(modified)))
   508  		}
   509  	}
   510  
   511  	// Assume that uncompressed size 2³²-1 could plausibly happen in
   512  	// an old zip32 file that was sharding inputs into the largest chunks
   513  	// possible (or is just malicious; search the web for 42.zip).
   514  	// If needUSize is true still, it means we didn't see a zip64 extension.
   515  	// As long as the compressed size is not also 2³²-1 (implausible)
   516  	// and the header is not also 2³²-1 (equally implausible),
   517  	// accept the uncompressed size 2³²-1 as valid.
   518  	// If nothing else, this keeps archive/zip working with 42.zip.
   519  	_ = needUSize
   520  
   521  	if needCSize || needHeaderOffset {
   522  		return ErrFormat
   523  	}
   524  
   525  	return nil
   526  }
   527  
   528  func readDataDescriptor(r io.Reader, f *File) error {
   529  	var buf [dataDescriptorLen]byte
   530  	// The spec says: "Although not originally assigned a
   531  	// signature, the value 0x08074b50 has commonly been adopted
   532  	// as a signature value for the data descriptor record.
   533  	// Implementers should be aware that ZIP files may be
   534  	// encountered with or without this signature marking data
   535  	// descriptors and should account for either case when reading
   536  	// ZIP files to ensure compatibility."
   537  	//
   538  	// dataDescriptorLen includes the size of the signature but
   539  	// first read just those 4 bytes to see if it exists.
   540  	if _, err := io.ReadFull(r, buf[:4]); err != nil {
   541  		return err
   542  	}
   543  	off := 0
   544  	maybeSig := readBuf(buf[:4])
   545  	if maybeSig.uint32() != dataDescriptorSignature {
   546  		// No data descriptor signature. Keep these four
   547  		// bytes.
   548  		off += 4
   549  	}
   550  	if _, err := io.ReadFull(r, buf[off:12]); err != nil {
   551  		return err
   552  	}
   553  	b := readBuf(buf[:12])
   554  	if b.uint32() != f.CRC32 {
   555  		return ErrChecksum
   556  	}
   557  
   558  	// The two sizes that follow here can be either 32 bits or 64 bits
   559  	// but the spec is not very clear on this and different
   560  	// interpretations has been made causing incompatibilities. We
   561  	// already have the sizes from the central directory so we can
   562  	// just ignore these.
   563  
   564  	return nil
   565  }
   566  
   567  func readDirectoryEnd(r io.ReaderAt, size int64) (dir *directoryEnd, baseOffset int64, err error) {
   568  	// look for directoryEndSignature in the last 1k, then in the last 65k
   569  	var buf []byte
   570  	var directoryEndOffset int64
   571  	for i, bLen := range []int64{1024, 65 * 1024} {
   572  		if bLen > size {
   573  			bLen = size
   574  		}
   575  		buf = make([]byte, int(bLen))
   576  		if _, err := r.ReadAt(buf, size-bLen); err != nil && err != io.EOF {
   577  			return nil, 0, err
   578  		}
   579  		if p := findSignatureInBlock(buf); p >= 0 {
   580  			buf = buf[p:]
   581  			directoryEndOffset = size - bLen + int64(p)
   582  			break
   583  		}
   584  		if i == 1 || bLen == size {
   585  			return nil, 0, ErrFormat
   586  		}
   587  	}
   588  
   589  	// read header into struct
   590  	b := readBuf(buf[4:]) // skip signature
   591  	d := &directoryEnd{
   592  		diskNbr:            uint32(b.uint16()),
   593  		dirDiskNbr:         uint32(b.uint16()),
   594  		dirRecordsThisDisk: uint64(b.uint16()),
   595  		directoryRecords:   uint64(b.uint16()),
   596  		directorySize:      uint64(b.uint32()),
   597  		directoryOffset:    uint64(b.uint32()),
   598  		commentLen:         b.uint16(),
   599  	}
   600  	l := int(d.commentLen)
   601  	if l > len(b) {
   602  		return nil, 0, errors.New("zip: invalid comment length")
   603  	}
   604  	d.comment = string(b[:l])
   605  
   606  	// These values mean that the file can be a zip64 file
   607  	if d.directoryRecords == 0xffff || d.directorySize == 0xffff || d.directoryOffset == 0xffffffff {
   608  		p, err := findDirectory64End(r, directoryEndOffset)
   609  		if err == nil && p >= 0 {
   610  			directoryEndOffset = p
   611  			err = readDirectory64End(r, p, d)
   612  		}
   613  		if err != nil {
   614  			return nil, 0, err
   615  		}
   616  	}
   617  
   618  	maxInt64 := uint64(1<<63 - 1)
   619  	if d.directorySize > maxInt64 || d.directoryOffset > maxInt64 {
   620  		return nil, 0, ErrFormat
   621  	}
   622  
   623  	baseOffset = directoryEndOffset - int64(d.directorySize) - int64(d.directoryOffset)
   624  
   625  	// Make sure directoryOffset points to somewhere in our file.
   626  	if o := baseOffset + int64(d.directoryOffset); o < 0 || o >= size {
   627  		return nil, 0, ErrFormat
   628  	}
   629  
   630  	// If the directory end data tells us to use a non-zero baseOffset,
   631  	// but we would find a valid directory entry if we assume that the
   632  	// baseOffset is 0, then just use a baseOffset of 0.
   633  	// We've seen files in which the directory end data gives us
   634  	// an incorrect baseOffset.
   635  	if baseOffset > 0 {
   636  		off := int64(d.directoryOffset)
   637  		rs := io.NewSectionReader(r, off, size-off)
   638  		if readDirectoryHeader(&File{}, rs) == nil {
   639  			baseOffset = 0
   640  		}
   641  	}
   642  
   643  	return d, baseOffset, nil
   644  }
   645  
   646  // findDirectory64End tries to read the zip64 locator just before the
   647  // directory end and returns the offset of the zip64 directory end if
   648  // found.
   649  func findDirectory64End(r io.ReaderAt, directoryEndOffset int64) (int64, error) {
   650  	locOffset := directoryEndOffset - directory64LocLen
   651  	if locOffset < 0 {
   652  		return -1, nil // no need to look for a header outside the file
   653  	}
   654  	buf := make([]byte, directory64LocLen)
   655  	if _, err := r.ReadAt(buf, locOffset); err != nil {
   656  		return -1, err
   657  	}
   658  	b := readBuf(buf)
   659  	if sig := b.uint32(); sig != directory64LocSignature {
   660  		return -1, nil
   661  	}
   662  	if b.uint32() != 0 { // number of the disk with the start of the zip64 end of central directory
   663  		return -1, nil // the file is not a valid zip64-file
   664  	}
   665  	p := b.uint64()      // relative offset of the zip64 end of central directory record
   666  	if b.uint32() != 1 { // total number of disks
   667  		return -1, nil // the file is not a valid zip64-file
   668  	}
   669  	return int64(p), nil
   670  }
   671  
   672  // readDirectory64End reads the zip64 directory end and updates the
   673  // directory end with the zip64 directory end values.
   674  func readDirectory64End(r io.ReaderAt, offset int64, d *directoryEnd) (err error) {
   675  	buf := make([]byte, directory64EndLen)
   676  	if _, err := r.ReadAt(buf, offset); err != nil {
   677  		return err
   678  	}
   679  
   680  	b := readBuf(buf)
   681  	if sig := b.uint32(); sig != directory64EndSignature {
   682  		return ErrFormat
   683  	}
   684  
   685  	b = b[12:]                        // skip dir size, version and version needed (uint64 + 2x uint16)
   686  	d.diskNbr = b.uint32()            // number of this disk
   687  	d.dirDiskNbr = b.uint32()         // number of the disk with the start of the central directory
   688  	d.dirRecordsThisDisk = b.uint64() // total number of entries in the central directory on this disk
   689  	d.directoryRecords = b.uint64()   // total number of entries in the central directory
   690  	d.directorySize = b.uint64()      // size of the central directory
   691  	d.directoryOffset = b.uint64()    // offset of start of central directory with respect to the starting disk number
   692  
   693  	return nil
   694  }
   695  
   696  func findSignatureInBlock(b []byte) int {
   697  	for i := len(b) - directoryEndLen; i >= 0; i-- {
   698  		// defined from directoryEndSignature in struct.go
   699  		if b[i] == 'P' && b[i+1] == 'K' && b[i+2] == 0x05 && b[i+3] == 0x06 {
   700  			// n is length of comment
   701  			n := int(b[i+directoryEndLen-2]) | int(b[i+directoryEndLen-1])<<8
   702  			if n+directoryEndLen+i > len(b) {
   703  				// Truncated comment.
   704  				// Some parsers (such as Info-ZIP) ignore the truncated comment
   705  				// rather than treating it as a hard error.
   706  				return -1
   707  			}
   708  			return i
   709  		}
   710  	}
   711  	return -1
   712  }
   713  
   714  type readBuf []byte
   715  
   716  func (b *readBuf) uint8() uint8 {
   717  	v := (*b)[0]
   718  	*b = (*b)[1:]
   719  	return v
   720  }
   721  
   722  func (b *readBuf) uint16() uint16 {
   723  	v := binary.LittleEndian.Uint16(*b)
   724  	*b = (*b)[2:]
   725  	return v
   726  }
   727  
   728  func (b *readBuf) uint32() uint32 {
   729  	v := binary.LittleEndian.Uint32(*b)
   730  	*b = (*b)[4:]
   731  	return v
   732  }
   733  
   734  func (b *readBuf) uint64() uint64 {
   735  	v := binary.LittleEndian.Uint64(*b)
   736  	*b = (*b)[8:]
   737  	return v
   738  }
   739  
   740  func (b *readBuf) sub(n int) readBuf {
   741  	b2 := (*b)[:n]
   742  	*b = (*b)[n:]
   743  	return b2
   744  }
   745  
   746  // A fileListEntry is a File and its ename.
   747  // If file == nil, the fileListEntry describes a directory without metadata.
   748  type fileListEntry struct {
   749  	name  string
   750  	file  *File
   751  	isDir bool
   752  	isDup bool
   753  }
   754  
   755  type fileInfoDirEntry interface {
   756  	fs.FileInfo
   757  	fs.DirEntry
   758  }
   759  
   760  func (f *fileListEntry) stat() (fileInfoDirEntry, error) {
   761  	if f.isDup {
   762  		return nil, errors.New(f.name + ": duplicate entries in zip file")
   763  	}
   764  	if !f.isDir {
   765  		return headerFileInfo{&f.file.FileHeader}, nil
   766  	}
   767  	return f, nil
   768  }
   769  
   770  // Only used for directories.
   771  func (f *fileListEntry) Name() string      { _, elem, _ := split(f.name); return elem }
   772  func (f *fileListEntry) Size() int64       { return 0 }
   773  func (f *fileListEntry) Mode() fs.FileMode { return fs.ModeDir | 0555 }
   774  func (f *fileListEntry) Type() fs.FileMode { return fs.ModeDir }
   775  func (f *fileListEntry) IsDir() bool       { return true }
   776  func (f *fileListEntry) Sys() any          { return nil }
   777  
   778  func (f *fileListEntry) ModTime() time.Time {
   779  	if f.file == nil {
   780  		return time.Time{}
   781  	}
   782  	return f.file.FileHeader.Modified.UTC()
   783  }
   784  
   785  func (f *fileListEntry) Info() (fs.FileInfo, error) { return f, nil }
   786  
   787  func (f *fileListEntry) String() string {
   788  	return fs.FormatDirEntry(f)
   789  }
   790  
   791  // toValidName coerces name to be a valid name for fs.FS.Open.
   792  func toValidName(name string) string {
   793  	name = strings.ReplaceAll(name, `\`, `/`)
   794  	p := path.Clean(name)
   795  
   796  	p = strings.TrimPrefix(p, "/")
   797  
   798  	for strings.HasPrefix(p, "../") {
   799  		p = p[len("../"):]
   800  	}
   801  
   802  	return p
   803  }
   804  
   805  func (r *Reader) initFileList() {
   806  	r.fileListOnce.Do(func() {
   807  		// files and knownDirs map from a file/directory name
   808  		// to an index into the r.fileList entry that we are
   809  		// building. They are used to mark duplicate entries.
   810  		files := make(map[string]int)
   811  		knownDirs := make(map[string]int)
   812  
   813  		// dirs[name] is true if name is known to be a directory,
   814  		// because it appears as a prefix in a path.
   815  		dirs := make(map[string]bool)
   816  
   817  		for _, file := range r.File {
   818  			isDir := len(file.Name) > 0 && file.Name[len(file.Name)-1] == '/'
   819  			name := toValidName(file.Name)
   820  			if name == "" {
   821  				continue
   822  			}
   823  
   824  			if idx, ok := files[name]; ok {
   825  				r.fileList[idx].isDup = true
   826  				continue
   827  			}
   828  			if idx, ok := knownDirs[name]; ok {
   829  				r.fileList[idx].isDup = true
   830  				continue
   831  			}
   832  
   833  			for dir := path.Dir(name); dir != "."; dir = path.Dir(dir) {
   834  				dirs[dir] = true
   835  			}
   836  
   837  			idx := len(r.fileList)
   838  			entry := fileListEntry{
   839  				name:  name,
   840  				file:  file,
   841  				isDir: isDir,
   842  			}
   843  			r.fileList = append(r.fileList, entry)
   844  			if isDir {
   845  				knownDirs[name] = idx
   846  			} else {
   847  				files[name] = idx
   848  			}
   849  		}
   850  		for dir := range dirs {
   851  			if _, ok := knownDirs[dir]; !ok {
   852  				if idx, ok := files[dir]; ok {
   853  					r.fileList[idx].isDup = true
   854  				} else {
   855  					entry := fileListEntry{
   856  						name:  dir,
   857  						file:  nil,
   858  						isDir: true,
   859  					}
   860  					r.fileList = append(r.fileList, entry)
   861  				}
   862  			}
   863  		}
   864  
   865  		slices.SortFunc(r.fileList, func(a, b fileListEntry) int {
   866  			return fileEntryCompare(a.name, b.name)
   867  		})
   868  	})
   869  }
   870  
   871  func fileEntryCompare(x, y string) int {
   872  	xdir, xelem, _ := split(x)
   873  	ydir, yelem, _ := split(y)
   874  	if xdir != ydir {
   875  		return strings.Compare(xdir, ydir)
   876  	}
   877  	return strings.Compare(xelem, yelem)
   878  }
   879  
   880  // Open opens the named file in the ZIP archive,
   881  // using the semantics of fs.FS.Open:
   882  // paths are always slash separated, with no
   883  // leading / or ../ elements.
   884  func (r *Reader) Open(name string) (fs.File, error) {
   885  	r.initFileList()
   886  
   887  	if !fs.ValidPath(name) {
   888  		return nil, &fs.PathError{Op: "open", Path: name, Err: fs.ErrInvalid}
   889  	}
   890  	e := r.openLookup(name)
   891  	if e == nil {
   892  		return nil, &fs.PathError{Op: "open", Path: name, Err: fs.ErrNotExist}
   893  	}
   894  	if e.isDir {
   895  		return &openDir{e, r.openReadDir(name), 0}, nil
   896  	}
   897  	rc, err := e.file.Open()
   898  	if err != nil {
   899  		return nil, err
   900  	}
   901  	return rc.(fs.File), nil
   902  }
   903  
   904  func split(name string) (dir, elem string, isDir bool) {
   905  	name, isDir = strings.CutSuffix(name, "/")
   906  	i := strings.LastIndexByte(name, '/')
   907  	if i < 0 {
   908  		return ".", name, isDir
   909  	}
   910  	return name[:i], name[i+1:], isDir
   911  }
   912  
   913  var dotFile = &fileListEntry{name: "./", isDir: true}
   914  
   915  func (r *Reader) openLookup(name string) *fileListEntry {
   916  	if name == "." {
   917  		return dotFile
   918  	}
   919  
   920  	dir, elem, _ := split(name)
   921  	files := r.fileList
   922  	i, _ := slices.BinarySearchFunc(files, dir, func(a fileListEntry, dir string) (ret int) {
   923  		idir, ielem, _ := split(a.name)
   924  		if dir != idir {
   925  			return strings.Compare(idir, dir)
   926  		}
   927  		return strings.Compare(ielem, elem)
   928  	})
   929  	if i < len(files) {
   930  		fname := files[i].name
   931  		if fname == name || len(fname) == len(name)+1 && fname[len(name)] == '/' && fname[:len(name)] == name {
   932  			return &files[i]
   933  		}
   934  	}
   935  	return nil
   936  }
   937  
   938  func (r *Reader) openReadDir(dir string) []fileListEntry {
   939  	files := r.fileList
   940  	i, _ := slices.BinarySearchFunc(files, dir, func(a fileListEntry, dir string) int {
   941  		idir, _, _ := split(a.name)
   942  		if dir != idir {
   943  			return strings.Compare(idir, dir)
   944  		}
   945  		// find the first entry with dir
   946  		return +1
   947  	})
   948  	j, _ := slices.BinarySearchFunc(files, dir, func(a fileListEntry, dir string) int {
   949  		jdir, _, _ := split(a.name)
   950  		if dir != jdir {
   951  			return strings.Compare(jdir, dir)
   952  		}
   953  		// find the last entry with dir
   954  		return -1
   955  	})
   956  	return files[i:j]
   957  }
   958  
   959  type openDir struct {
   960  	e      *fileListEntry
   961  	files  []fileListEntry
   962  	offset int
   963  }
   964  
   965  func (d *openDir) Close() error               { return nil }
   966  func (d *openDir) Stat() (fs.FileInfo, error) { return d.e.stat() }
   967  
   968  func (d *openDir) Read([]byte) (int, error) {
   969  	return 0, &fs.PathError{Op: "read", Path: d.e.name, Err: errors.New("is a directory")}
   970  }
   971  
   972  func (d *openDir) ReadDir(count int) ([]fs.DirEntry, error) {
   973  	n := len(d.files) - d.offset
   974  	if count > 0 && n > count {
   975  		n = count
   976  	}
   977  	if n == 0 {
   978  		if count <= 0 {
   979  			return nil, nil
   980  		}
   981  		return nil, io.EOF
   982  	}
   983  	list := make([]fs.DirEntry, n)
   984  	for i := range list {
   985  		s, err := d.files[d.offset+i].stat()
   986  		if err != nil {
   987  			return nil, err
   988  		}
   989  		list[i] = s
   990  	}
   991  	d.offset += n
   992  	return list, nil
   993  }
   994  

View as plain text