Merge pull request #48 from kolyshkin/go111

rebase to go-1.11's archive/tar

.travis.yml

@@ -1,11 +1,8 @@
 language: go
 go:
   - tip
-  - 1.x
+  - 1.11.x
-  - 1.8.x
+  - 1.10.x
-  - 1.7.x
-  - 1.6.x
-  - 1.5.x
 
 # let us have pretty, fast Docker-based Travis workers!
 sudo: false

README.md

@@ -67,7 +67,7 @@ Do not break the API of stdlib `archive/tar` in our fork (ideally find an upstre
 
 ## Std Version
 
-The version of golang stdlib `archive/tar` is from go1.6
+The version of golang stdlib `archive/tar` is from go1.11
 It is minimally extended to expose the raw bytes of the TAR, rather than just the marshalled headers and file stream.
 
 
@@ -135,4 +135,3 @@ bytes-per-file rate for the storage implications.
 ## License
 
 See [LICENSE](LICENSE)
-

archive/tar/common.go

@@ -3,70 +3,528 @@
 // license that can be found in the LICENSE file.
 
 // Package tar implements access to tar archives.
-// It aims to cover most of the variations, including those produced
-// by GNU and BSD tars.
 //
-// References:
+// Tape archives (tar) are a file format for storing a sequence of files that
-//   http://www.freebsd.org/cgi/man.cgi?query=tar&sektion=5
+// can be read and written in a streaming manner.
-//   http://www.gnu.org/software/tar/manual/html_node/Standard.html
+// This package aims to cover most variations of the format,
-//   http://pubs.opengroup.org/onlinepubs/9699919799/utilities/pax.html
+// including those produced by GNU and BSD tar tools.
 package tar
 
 import (
-	"bytes"
 	"errors"
 	"fmt"
+	"math"
 	"os"
 	"path"
+	"reflect"
+	"strconv"
+	"strings"
 	"time"
 )
 
-const (
+// BUG: Use of the Uid and Gid fields in Header could overflow on 32-bit
-	blockSize = 512
+// architectures. If a large value is encountered when decoding, the result
+// stored in Header will be the truncated version.
 
-	// Types
+var (
-	TypeReg           = '0'    // regular file
+	ErrHeader          = errors.New("archive/tar: invalid tar header")
-	TypeRegA          = '\x00' // regular file
+	ErrWriteTooLong    = errors.New("archive/tar: write too long")
-	TypeLink          = '1'    // hard link
+	ErrFieldTooLong    = errors.New("archive/tar: header field too long")
-	TypeSymlink       = '2'    // symbolic link
+	ErrWriteAfterClose = errors.New("archive/tar: write after close")
-	TypeChar          = '3'    // character device node
+	errMissData        = errors.New("archive/tar: sparse file references non-existent data")
-	TypeBlock         = '4'    // block device node
+	errUnrefData       = errors.New("archive/tar: sparse file contains unreferenced data")
-	TypeDir           = '5'    // directory
+	errWriteHole       = errors.New("archive/tar: write non-NUL byte in sparse hole")
-	TypeFifo          = '6'    // fifo node
-	TypeCont          = '7'    // reserved
-	TypeXHeader       = 'x'    // extended header
-	TypeXGlobalHeader = 'g'    // global extended header
-	TypeGNULongName   = 'L'    // Next file has a long name
-	TypeGNULongLink   = 'K'    // Next file symlinks to a file w/ a long name
-	TypeGNUSparse     = 'S'    // sparse file
 )
 
+type headerError []string
+
+func (he headerError) Error() string {
+	const prefix = "archive/tar: cannot encode header"
+	var ss []string
+	for _, s := range he {
+		if s != "" {
+			ss = append(ss, s)
+		}
+	}
+	if len(ss) == 0 {
+		return prefix
+	}
+	return fmt.Sprintf("%s: %v", prefix, strings.Join(ss, "; and "))
+}
+
+// Type flags for Header.Typeflag.
+const (
+	// Type '0' indicates a regular file.
+	TypeReg  = '0'
+	TypeRegA = '\x00' // Deprecated: Use TypeReg instead.
+
+	// Type '1' to '6' are header-only flags and may not have a data body.
+	TypeLink    = '1' // Hard link
+	TypeSymlink = '2' // Symbolic link
+	TypeChar    = '3' // Character device node
+	TypeBlock   = '4' // Block device node
+	TypeDir     = '5' // Directory
+	TypeFifo    = '6' // FIFO node
+
+	// Type '7' is reserved.
+	TypeCont = '7'
+
+	// Type 'x' is used by the PAX format to store key-value records that
+	// are only relevant to the next file.
+	// This package transparently handles these types.
+	TypeXHeader = 'x'
+
+	// Type 'g' is used by the PAX format to store key-value records that
+	// are relevant to all subsequent files.
+	// This package only supports parsing and composing such headers,
+	// but does not currently support persisting the global state across files.
+	TypeXGlobalHeader = 'g'
+
+	// Type 'S' indicates a sparse file in the GNU format.
+	TypeGNUSparse = 'S'
+
+	// Types 'L' and 'K' are used by the GNU format for a meta file
+	// used to store the path or link name for the next file.
+	// This package transparently handles these types.
+	TypeGNULongName = 'L'
+	TypeGNULongLink = 'K'
+)
+
+// Keywords for PAX extended header records.
+const (
+	paxNone     = "" // Indicates that no PAX key is suitable
+	paxPath     = "path"
+	paxLinkpath = "linkpath"
+	paxSize     = "size"
+	paxUid      = "uid"
+	paxGid      = "gid"
+	paxUname    = "uname"
+	paxGname    = "gname"
+	paxMtime    = "mtime"
+	paxAtime    = "atime"
+	paxCtime    = "ctime"   // Removed from later revision of PAX spec, but was valid
+	paxCharset  = "charset" // Currently unused
+	paxComment  = "comment" // Currently unused
+
+	paxSchilyXattr = "SCHILY.xattr."
+
+	// Keywords for GNU sparse files in a PAX extended header.
+	paxGNUSparse          = "GNU.sparse."
+	paxGNUSparseNumBlocks = "GNU.sparse.numblocks"
+	paxGNUSparseOffset    = "GNU.sparse.offset"
+	paxGNUSparseNumBytes  = "GNU.sparse.numbytes"
+	paxGNUSparseMap       = "GNU.sparse.map"
+	paxGNUSparseName      = "GNU.sparse.name"
+	paxGNUSparseMajor     = "GNU.sparse.major"
+	paxGNUSparseMinor     = "GNU.sparse.minor"
+	paxGNUSparseSize      = "GNU.sparse.size"
+	paxGNUSparseRealSize  = "GNU.sparse.realsize"
+)
+
+// basicKeys is a set of the PAX keys for which we have built-in support.
+// This does not contain "charset" or "comment", which are both PAX-specific,
+// so adding them as first-class features of Header is unlikely.
+// Users can use the PAXRecords field to set it themselves.
+var basicKeys = map[string]bool{
+	paxPath: true, paxLinkpath: true, paxSize: true, paxUid: true, paxGid: true,
+	paxUname: true, paxGname: true, paxMtime: true, paxAtime: true, paxCtime: true,
+}
+
 // A Header represents a single header in a tar archive.
 // Some fields may not be populated.
+//
+// For forward compatibility, users that retrieve a Header from Reader.Next,
+// mutate it in some ways, and then pass it back to Writer.WriteHeader
+// should do so by creating a new Header and copying the fields
+// that they are interested in preserving.
 type Header struct {
-	Name       string    // name of header file entry
+	// Typeflag is the type of header entry.
-	Mode       int64     // permission and mode bits
+	// The zero value is automatically promoted to either TypeReg or TypeDir
-	Uid        int       // user id of owner
+	// depending on the presence of a trailing slash in Name.
-	Gid        int       // group id of owner
+	Typeflag byte
-	Size       int64     // length in bytes
+
-	ModTime    time.Time // modified time
+	Name     string // Name of file entry
-	Typeflag   byte      // type of header entry
+	Linkname string // Target name of link (valid for TypeLink or TypeSymlink)
-	Linkname   string    // target name of link
+
-	Uname      string    // user name of owner
+	Size  int64  // Logical file size in bytes
-	Gname      string    // group name of owner
+	Mode  int64  // Permission and mode bits
-	Devmajor   int64     // major number of character or block device
+	Uid   int    // User ID of owner
-	Devminor   int64     // minor number of character or block device
+	Gid   int    // Group ID of owner
-	AccessTime time.Time // access time
+	Uname string // User name of owner
-	ChangeTime time.Time // status change time
+	Gname string // Group name of owner
-	Xattrs     map[string]string
+
+	// If the Format is unspecified, then Writer.WriteHeader rounds ModTime
+	// to the nearest second and ignores the AccessTime and ChangeTime fields.
+	//
+	// To use AccessTime or ChangeTime, specify the Format as PAX or GNU.
+	// To use sub-second resolution, specify the Format as PAX.
+	ModTime    time.Time // Modification time
+	AccessTime time.Time // Access time (requires either PAX or GNU support)
+	ChangeTime time.Time // Change time (requires either PAX or GNU support)
+
+	Devmajor int64 // Major device number (valid for TypeChar or TypeBlock)
+	Devminor int64 // Minor device number (valid for TypeChar or TypeBlock)
+
+	// Xattrs stores extended attributes as PAX records under the
+	// "SCHILY.xattr." namespace.
+	//
+	// The following are semantically equivalent:
+	//  h.Xattrs[key] = value
+	//  h.PAXRecords["SCHILY.xattr."+key] = value
+	//
+	// When Writer.WriteHeader is called, the contents of Xattrs will take
+	// precedence over those in PAXRecords.
+	//
+	// Deprecated: Use PAXRecords instead.
+	Xattrs map[string]string
+
+	// PAXRecords is a map of PAX extended header records.
+	//
+	// User-defined records should have keys of the following form:
+	//	VENDOR.keyword
+	// Where VENDOR is some namespace in all uppercase, and keyword may
+	// not contain the '=' character (e.g., "GOLANG.pkg.version").
+	// The key and value should be non-empty UTF-8 strings.
+	//
+	// When Writer.WriteHeader is called, PAX records derived from the
+	// other fields in Header take precedence over PAXRecords.
+	PAXRecords map[string]string
+
+	// Format specifies the format of the tar header.
+	//
+	// This is set by Reader.Next as a best-effort guess at the format.
+	// Since the Reader liberally reads some non-compliant files,
+	// it is possible for this to be FormatUnknown.
+	//
+	// If the format is unspecified when Writer.WriteHeader is called,
+	// then it uses the first format (in the order of USTAR, PAX, GNU)
+	// capable of encoding this Header (see Format).
+	Format Format
 }
 
-// File name constants from the tar spec.
+// sparseEntry represents a Length-sized fragment at Offset in the file.
-const (
+type sparseEntry struct{ Offset, Length int64 }
-	fileNameSize       = 100 // Maximum number of bytes in a standard tar name.
+
-	fileNamePrefixSize = 155 // Maximum number of ustar extension bytes.
+func (s sparseEntry) endOffset() int64 { return s.Offset + s.Length }
+
+// A sparse file can be represented as either a sparseDatas or a sparseHoles.
+// As long as the total size is known, they are equivalent and one can be
+// converted to the other form and back. The various tar formats with sparse
+// file support represent sparse files in the sparseDatas form. That is, they
+// specify the fragments in the file that has data, and treat everything else as
+// having zero bytes. As such, the encoding and decoding logic in this package
+// deals with sparseDatas.
+//
+// However, the external API uses sparseHoles instead of sparseDatas because the
+// zero value of sparseHoles logically represents a normal file (i.e., there are
+// no holes in it). On the other hand, the zero value of sparseDatas implies
+// that the file has no data in it, which is rather odd.
+//
+// As an example, if the underlying raw file contains the 10-byte data:
+//	var compactFile = "abcdefgh"
+//
+// And the sparse map has the following entries:
+//	var spd sparseDatas = []sparseEntry{
+//		{Offset: 2,  Length: 5},  // Data fragment for 2..6
+//		{Offset: 18, Length: 3},  // Data fragment for 18..20
+//	}
+//	var sph sparseHoles = []sparseEntry{
+//		{Offset: 0,  Length: 2},  // Hole fragment for 0..1
+//		{Offset: 7,  Length: 11}, // Hole fragment for 7..17
+//		{Offset: 21, Length: 4},  // Hole fragment for 21..24
+//	}
+//
+// Then the content of the resulting sparse file with a Header.Size of 25 is:
+//	var sparseFile = "\x00"*2 + "abcde" + "\x00"*11 + "fgh" + "\x00"*4
+type (
+	sparseDatas []sparseEntry
+	sparseHoles []sparseEntry
 )
 
+// validateSparseEntries reports whether sp is a valid sparse map.
+// It does not matter whether sp represents data fragments or hole fragments.
+func validateSparseEntries(sp []sparseEntry, size int64) bool {
+	// Validate all sparse entries. These are the same checks as performed by
+	// the BSD tar utility.
+	if size < 0 {
+		return false
+	}
+	var pre sparseEntry
+	for _, cur := range sp {
+		switch {
+		case cur.Offset < 0 || cur.Length < 0:
+			return false // Negative values are never okay
+		case cur.Offset > math.MaxInt64-cur.Length:
+			return false // Integer overflow with large length
+		case cur.endOffset() > size:
+			return false // Region extends beyond the actual size
+		case pre.endOffset() > cur.Offset:
+			return false // Regions cannot overlap and must be in order
+		}
+		pre = cur
+	}
+	return true
+}
+
+// alignSparseEntries mutates src and returns dst where each fragment's
+// starting offset is aligned up to the nearest block edge, and each
+// ending offset is aligned down to the nearest block edge.
+//
+// Even though the Go tar Reader and the BSD tar utility can handle entries
+// with arbitrary offsets and lengths, the GNU tar utility can only handle
+// offsets and lengths that are multiples of blockSize.
+func alignSparseEntries(src []sparseEntry, size int64) []sparseEntry {
+	dst := src[:0]
+	for _, s := range src {
+		pos, end := s.Offset, s.endOffset()
+		pos += blockPadding(+pos) // Round-up to nearest blockSize
+		if end != size {
+			end -= blockPadding(-end) // Round-down to nearest blockSize
+		}
+		if pos < end {
+			dst = append(dst, sparseEntry{Offset: pos, Length: end - pos})
+		}
+	}
+	return dst
+}
+
+// invertSparseEntries converts a sparse map from one form to the other.
+// If the input is sparseHoles, then it will output sparseDatas and vice-versa.
+// The input must have been already validated.
+//
+// This function mutates src and returns a normalized map where:
+//	* adjacent fragments are coalesced together
+//	* only the last fragment may be empty
+//	* the endOffset of the last fragment is the total size
+func invertSparseEntries(src []sparseEntry, size int64) []sparseEntry {
+	dst := src[:0]
+	var pre sparseEntry
+	for _, cur := range src {
+		if cur.Length == 0 {
+			continue // Skip empty fragments
+		}
+		pre.Length = cur.Offset - pre.Offset
+		if pre.Length > 0 {
+			dst = append(dst, pre) // Only add non-empty fragments
+		}
+		pre.Offset = cur.endOffset()
+	}
+	pre.Length = size - pre.Offset // Possibly the only empty fragment
+	return append(dst, pre)
+}
+
+// fileState tracks the number of logical (includes sparse holes) and physical
+// (actual in tar archive) bytes remaining for the current file.
+//
+// Invariant: LogicalRemaining >= PhysicalRemaining
+type fileState interface {
+	LogicalRemaining() int64
+	PhysicalRemaining() int64
+}
+
+// allowedFormats determines which formats can be used.
+// The value returned is the logical OR of multiple possible formats.
+// If the value is FormatUnknown, then the input Header cannot be encoded
+// and an error is returned explaining why.
+//
+// As a by-product of checking the fields, this function returns paxHdrs, which
+// contain all fields that could not be directly encoded.
+// A value receiver ensures that this method does not mutate the source Header.
+func (h Header) allowedFormats() (format Format, paxHdrs map[string]string, err error) {
+	format = FormatUSTAR | FormatPAX | FormatGNU
+	paxHdrs = make(map[string]string)
+
+	var whyNoUSTAR, whyNoPAX, whyNoGNU string
+	var preferPAX bool // Prefer PAX over USTAR
+	verifyString := func(s string, size int, name, paxKey string) {
+		// NUL-terminator is optional for path and linkpath.
+		// Technically, it is required for uname and gname,
+		// but neither GNU nor BSD tar checks for it.
+		tooLong := len(s) > size
+		allowLongGNU := paxKey == paxPath || paxKey == paxLinkpath
+		if hasNUL(s) || (tooLong && !allowLongGNU) {
+			whyNoGNU = fmt.Sprintf("GNU cannot encode %s=%q", name, s)
+			format.mustNotBe(FormatGNU)
+		}
+		if !isASCII(s) || tooLong {
+			canSplitUSTAR := paxKey == paxPath
+			if _, _, ok := splitUSTARPath(s); !canSplitUSTAR || !ok {
+				whyNoUSTAR = fmt.Sprintf("USTAR cannot encode %s=%q", name, s)
+				format.mustNotBe(FormatUSTAR)
+			}
+			if paxKey == paxNone {
+				whyNoPAX = fmt.Sprintf("PAX cannot encode %s=%q", name, s)
+				format.mustNotBe(FormatPAX)
+			} else {
+				paxHdrs[paxKey] = s
+			}
+		}
+		if v, ok := h.PAXRecords[paxKey]; ok && v == s {
+			paxHdrs[paxKey] = v
+		}
+	}
+	verifyNumeric := func(n int64, size int, name, paxKey string) {
+		if !fitsInBase256(size, n) {
+			whyNoGNU = fmt.Sprintf("GNU cannot encode %s=%d", name, n)
+			format.mustNotBe(FormatGNU)
+		}
+		if !fitsInOctal(size, n) {
+			whyNoUSTAR = fmt.Sprintf("USTAR cannot encode %s=%d", name, n)
+			format.mustNotBe(FormatUSTAR)
+			if paxKey == paxNone {
+				whyNoPAX = fmt.Sprintf("PAX cannot encode %s=%d", name, n)
+				format.mustNotBe(FormatPAX)
+			} else {
+				paxHdrs[paxKey] = strconv.FormatInt(n, 10)
+			}
+		}
+		if v, ok := h.PAXRecords[paxKey]; ok && v == strconv.FormatInt(n, 10) {
+			paxHdrs[paxKey] = v
+		}
+	}
+	verifyTime := func(ts time.Time, size int, name, paxKey string) {
+		if ts.IsZero() {
+			return // Always okay
+		}
+		if !fitsInBase256(size, ts.Unix()) {
+			whyNoGNU = fmt.Sprintf("GNU cannot encode %s=%v", name, ts)
+			format.mustNotBe(FormatGNU)
+		}
+		isMtime := paxKey == paxMtime
+		fitsOctal := fitsInOctal(size, ts.Unix())
+		if (isMtime && !fitsOctal) || !isMtime {
+			whyNoUSTAR = fmt.Sprintf("USTAR cannot encode %s=%v", name, ts)
+			format.mustNotBe(FormatUSTAR)
+		}
+		needsNano := ts.Nanosecond() != 0
+		if !isMtime || !fitsOctal || needsNano {
+			preferPAX = true // USTAR may truncate sub-second measurements
+			if paxKey == paxNone {
+				whyNoPAX = fmt.Sprintf("PAX cannot encode %s=%v", name, ts)
+				format.mustNotBe(FormatPAX)
+			} else {
+				paxHdrs[paxKey] = formatPAXTime(ts)
+			}
+		}
+		if v, ok := h.PAXRecords[paxKey]; ok && v == formatPAXTime(ts) {
+			paxHdrs[paxKey] = v
+		}
+	}
+
+	// Check basic fields.
+	var blk block
+	v7 := blk.V7()
+	ustar := blk.USTAR()
+	gnu := blk.GNU()
+	verifyString(h.Name, len(v7.Name()), "Name", paxPath)
+	verifyString(h.Linkname, len(v7.LinkName()), "Linkname", paxLinkpath)
+	verifyString(h.Uname, len(ustar.UserName()), "Uname", paxUname)
+	verifyString(h.Gname, len(ustar.GroupName()), "Gname", paxGname)
+	verifyNumeric(h.Mode, len(v7.Mode()), "Mode", paxNone)
+	verifyNumeric(int64(h.Uid), len(v7.UID()), "Uid", paxUid)
+	verifyNumeric(int64(h.Gid), len(v7.GID()), "Gid", paxGid)
+	verifyNumeric(h.Size, len(v7.Size()), "Size", paxSize)
+	verifyNumeric(h.Devmajor, len(ustar.DevMajor()), "Devmajor", paxNone)
+	verifyNumeric(h.Devminor, len(ustar.DevMinor()), "Devminor", paxNone)
+	verifyTime(h.ModTime, len(v7.ModTime()), "ModTime", paxMtime)
+	verifyTime(h.AccessTime, len(gnu.AccessTime()), "AccessTime", paxAtime)
+	verifyTime(h.ChangeTime, len(gnu.ChangeTime()), "ChangeTime", paxCtime)
+
+	// Check for header-only types.
+	var whyOnlyPAX, whyOnlyGNU string
+	switch h.Typeflag {
+	case TypeReg, TypeChar, TypeBlock, TypeFifo, TypeGNUSparse:
+		// Exclude TypeLink and TypeSymlink, since they may reference directories.
+		if strings.HasSuffix(h.Name, "/") {
+			return FormatUnknown, nil, headerError{"filename may not have trailing slash"}
+		}
+	case TypeXHeader, TypeGNULongName, TypeGNULongLink:
+		return FormatUnknown, nil, headerError{"cannot manually encode TypeXHeader, TypeGNULongName, or TypeGNULongLink headers"}
+	case TypeXGlobalHeader:
+		h2 := Header{Name: h.Name, Typeflag: h.Typeflag, Xattrs: h.Xattrs, PAXRecords: h.PAXRecords, Format: h.Format}
+		if !reflect.DeepEqual(h, h2) {
+			return FormatUnknown, nil, headerError{"only PAXRecords should be set for TypeXGlobalHeader"}
+		}
+		whyOnlyPAX = "only PAX supports TypeXGlobalHeader"
+		format.mayOnlyBe(FormatPAX)
+	}
+	if !isHeaderOnlyType(h.Typeflag) && h.Size < 0 {
+		return FormatUnknown, nil, headerError{"negative size on header-only type"}
+	}
+
+	// Check PAX records.
+	if len(h.Xattrs) > 0 {
+		for k, v := range h.Xattrs {
+			paxHdrs[paxSchilyXattr+k] = v
+		}
+		whyOnlyPAX = "only PAX supports Xattrs"
+		format.mayOnlyBe(FormatPAX)
+	}
+	if len(h.PAXRecords) > 0 {
+		for k, v := range h.PAXRecords {
+			switch _, exists := paxHdrs[k]; {
+			case exists:
+				continue // Do not overwrite existing records
+			case h.Typeflag == TypeXGlobalHeader:
+				paxHdrs[k] = v // Copy all records
+			case !basicKeys[k] && !strings.HasPrefix(k, paxGNUSparse):
+				paxHdrs[k] = v // Ignore local records that may conflict
+			}
+		}
+		whyOnlyPAX = "only PAX supports PAXRecords"
+		format.mayOnlyBe(FormatPAX)
+	}
+	for k, v := range paxHdrs {
+		if !validPAXRecord(k, v) {
+			return FormatUnknown, nil, headerError{fmt.Sprintf("invalid PAX record: %q", k+" = "+v)}
+		}
+	}
+
+	// TODO(dsnet): Re-enable this when adding sparse support.
+	// See https://golang.org/issue/22735
+	/*
+		// Check sparse files.
+		if len(h.SparseHoles) > 0 || h.Typeflag == TypeGNUSparse {
+			if isHeaderOnlyType(h.Typeflag) {
+				return FormatUnknown, nil, headerError{"header-only type cannot be sparse"}
+			}
+			if !validateSparseEntries(h.SparseHoles, h.Size) {
+				return FormatUnknown, nil, headerError{"invalid sparse holes"}
+			}
+			if h.Typeflag == TypeGNUSparse {
+				whyOnlyGNU = "only GNU supports TypeGNUSparse"
+				format.mayOnlyBe(FormatGNU)
+			} else {
+				whyNoGNU = "GNU supports sparse files only with TypeGNUSparse"
+				format.mustNotBe(FormatGNU)
+			}
+			whyNoUSTAR = "USTAR does not support sparse files"
+			format.mustNotBe(FormatUSTAR)
+		}
+	*/
+
+	// Check desired format.
+	if wantFormat := h.Format; wantFormat != FormatUnknown {
+		if wantFormat.has(FormatPAX) && !preferPAX {
+			wantFormat.mayBe(FormatUSTAR) // PAX implies USTAR allowed too
+		}
+		format.mayOnlyBe(wantFormat) // Set union of formats allowed and format wanted
+	}
+	if format == FormatUnknown {
+		switch h.Format {
+		case FormatUSTAR:
+			err = headerError{"Format specifies USTAR", whyNoUSTAR, whyOnlyPAX, whyOnlyGNU}
+		case FormatPAX:
+			err = headerError{"Format specifies PAX", whyNoPAX, whyOnlyGNU}
+		case FormatGNU:
+			err = headerError{"Format specifies GNU", whyNoGNU, whyOnlyPAX}
+		default:
+			err = headerError{whyNoUSTAR, whyNoPAX, whyNoGNU, whyOnlyPAX, whyOnlyGNU}
+		}
+	}
+	return format, paxHdrs, err
+}
+
 // FileInfo returns an os.FileInfo for the Header.
 func (h *Header) FileInfo() os.FileInfo {
 	return headerFileInfo{h}
@@ -97,63 +555,43 @@ func (fi headerFileInfo) Mode() (mode os.FileMode) {
 
 	// Set setuid, setgid and sticky bits.
 	if fi.h.Mode&c_ISUID != 0 {
-		// setuid
 		mode |= os.ModeSetuid
 	}
 	if fi.h.Mode&c_ISGID != 0 {
-		// setgid
 		mode |= os.ModeSetgid
 	}
 	if fi.h.Mode&c_ISVTX != 0 {
-		// sticky
 		mode |= os.ModeSticky
 	}
 
-	// Set file mode bits.
+	// Set file mode bits; clear perm, setuid, setgid, and sticky bits.
-	// clear perm, setuid, setgid and sticky bits.
+	switch m := os.FileMode(fi.h.Mode) &^ 07777; m {
-	m := os.FileMode(fi.h.Mode) &^ 07777
+	case c_ISDIR:
-	if m == c_ISDIR {
-		// directory
 		mode |= os.ModeDir
-	}
+	case c_ISFIFO:
-	if m == c_ISFIFO {
-		// named pipe (FIFO)
 		mode |= os.ModeNamedPipe
-	}
+	case c_ISLNK:
-	if m == c_ISLNK {
-		// symbolic link
 		mode |= os.ModeSymlink
-	}
+	case c_ISBLK:
-	if m == c_ISBLK {
-		// device file
 		mode |= os.ModeDevice
-	}
+	case c_ISCHR:
-	if m == c_ISCHR {
-		// Unix character device
 		mode |= os.ModeDevice
 		mode |= os.ModeCharDevice
-	}
+	case c_ISSOCK:
-	if m == c_ISSOCK {
-		// Unix domain socket
 		mode |= os.ModeSocket
 	}
 
 	switch fi.h.Typeflag {
 	case TypeSymlink:
-		// symbolic link
 		mode |= os.ModeSymlink
 	case TypeChar:
-		// character device node
 		mode |= os.ModeDevice
 		mode |= os.ModeCharDevice
 	case TypeBlock:
-		// block device node
 		mode |= os.ModeDevice
 	case TypeDir:
-		// directory
 		mode |= os.ModeDir
 	case TypeFifo:
-		// fifo node
 		mode |= os.ModeNamedPipe
 	}
 
@@ -163,11 +601,15 @@ func (fi headerFileInfo) Mode() (mode os.FileMode) {
 // sysStat, if non-nil, populates h from system-dependent fields of fi.
 var sysStat func(fi os.FileInfo, h *Header) error
 
-// Mode constants from the tar spec.
 const (
-	c_ISUID  = 04000   // Set uid
+	// Mode constants from the USTAR spec:
-	c_ISGID  = 02000   // Set gid
+	// See http://pubs.opengroup.org/onlinepubs/9699919799/utilities/pax.html#tag_20_92_13_06
-	c_ISVTX  = 01000   // Save text (sticky bit)
+	c_ISUID = 04000 // Set uid
+	c_ISGID = 02000 // Set gid
+	c_ISVTX = 01000 // Save text (sticky bit)
+
+	// Common Unix mode constants; these are not defined in any common tar standard.
+	// Header.FileInfo understands these, but FileInfoHeader will never produce these.
 	c_ISDIR  = 040000  // Directory
 	c_ISFIFO = 010000  // FIFO
 	c_ISREG  = 0100000 // Regular file
@@ -177,33 +619,16 @@ const (
 	c_ISSOCK = 0140000 // Socket
 )
 
-// Keywords for the PAX Extended Header
-const (
-	paxAtime    = "atime"
-	paxCharset  = "charset"
-	paxComment  = "comment"
-	paxCtime    = "ctime" // please note that ctime is not a valid pax header.
-	paxGid      = "gid"
-	paxGname    = "gname"
-	paxLinkpath = "linkpath"
-	paxMtime    = "mtime"
-	paxPath     = "path"
-	paxSize     = "size"
-	paxUid      = "uid"
-	paxUname    = "uname"
-	paxXattr    = "SCHILY.xattr."
-	paxNone     = ""
-)
-
 // FileInfoHeader creates a partially-populated Header from fi.
 // If fi describes a symlink, FileInfoHeader records link as the link target.
 // If fi describes a directory, a slash is appended to the name.
-// Because os.FileInfo's Name method returns only the base name of
+//
-// the file it describes, it may be necessary to modify the Name field
+// Since os.FileInfo's Name method only returns the base name of
-// of the returned header to provide the full path name of the file.
+// the file it describes, it may be necessary to modify Header.Name
+// to provide the full path name of the file.
 func FileInfoHeader(fi os.FileInfo, link string) (*Header, error) {
 	if fi == nil {
-		return nil, errors.New("tar: FileInfo is nil")
+		return nil, errors.New("archive/tar: FileInfo is nil")
 	}
 	fm := fi.Mode()
 	h := &Header{
@@ -213,30 +638,24 @@ func FileInfoHeader(fi os.FileInfo, link string) (*Header, error) {
 	}
 	switch {
 	case fm.IsRegular():
-		h.Mode |= c_ISREG
 		h.Typeflag = TypeReg
 		h.Size = fi.Size()
 	case fi.IsDir():
 		h.Typeflag = TypeDir
-		h.Mode |= c_ISDIR
 		h.Name += "/"
 	case fm&os.ModeSymlink != 0:
 		h.Typeflag = TypeSymlink
-		h.Mode |= c_ISLNK
 		h.Linkname = link
 	case fm&os.ModeDevice != 0:
 		if fm&os.ModeCharDevice != 0 {
-			h.Mode |= c_ISCHR
 			h.Typeflag = TypeChar
 		} else {
-			h.Mode |= c_ISBLK
 			h.Typeflag = TypeBlock
 		}
 	case fm&os.ModeNamedPipe != 0:
 		h.Typeflag = TypeFifo
-		h.Mode |= c_ISFIFO
 	case fm&os.ModeSocket != 0:
-		h.Mode |= c_ISSOCK
+		return nil, fmt.Errorf("archive/tar: sockets not supported")
 	default:
 		return nil, fmt.Errorf("archive/tar: unknown file mode %v", fm)
 	}
@@ -272,6 +691,12 @@ func FileInfoHeader(fi os.FileInfo, link string) (*Header, error) {
 			h.Size = 0
 			h.Linkname = sys.Linkname
 		}
+		if sys.PAXRecords != nil {
+			h.PAXRecords = make(map[string]string)
+			for k, v := range sys.PAXRecords {
+				h.PAXRecords[k] = v
+			}
+		}
 	}
 	if sysStat != nil {
 		return h, sysStat(fi, h)
@@ -279,55 +704,6 @@ func FileInfoHeader(fi os.FileInfo, link string) (*Header, error) {
 	return h, nil
 }
 
-var zeroBlock = make([]byte, blockSize)
-
-// POSIX specifies a sum of the unsigned byte values, but the Sun tar uses signed byte values.
-// We compute and return both.
-func checksum(header []byte) (unsigned int64, signed int64) {
-	for i := 0; i < len(header); i++ {
-		if i == 148 {
-			// The chksum field (header[148:156]) is special: it should be treated as space bytes.
-			unsigned += ' ' * 8
-			signed += ' ' * 8
-			i += 7
-			continue
-		}
-		unsigned += int64(header[i])
-		signed += int64(int8(header[i]))
-	}
-	return
-}
-
-type slicer []byte
-
-func (sp *slicer) next(n int) (b []byte) {
-	s := *sp
-	b, *sp = s[0:n], s[n:]
-	return
-}
-
-func isASCII(s string) bool {
-	for _, c := range s {
-		if c >= 0x80 {
-			return false
-		}
-	}
-	return true
-}
-
-func toASCII(s string) string {
-	if isASCII(s) {
-		return s
-	}
-	var buf bytes.Buffer
-	for _, c := range s {
-		if c < 0x80 {
-			buf.WriteByte(byte(c))
-		}
-	}
-	return buf.String()
-}
-
 // isHeaderOnlyType checks if the given type flag is of the type that has no
 // data section even if a size is specified.
 func isHeaderOnlyType(flag byte) bool {
@@ -338,3 +714,10 @@ func isHeaderOnlyType(flag byte) bool {
 		return false
 	}
 }
+
+func min(a, b int64) int64 {
+	if a < b {
+		return a
+	}
+	return b
+}

archive/tar/example_test.go

@@ -13,14 +13,10 @@ import (
 	"os"
 )
 
-func Example() {
+func Example_minimal() {
-	// Create a buffer to write our archive to.
+	// Create and add some files to the archive.
-	buf := new(bytes.Buffer)
+	var buf bytes.Buffer
-
+	tw := tar.NewWriter(&buf)
-	// Create a new tar archive.
-	tw := tar.NewWriter(buf)
-
-	// Add some files to the archive.
 	var files = []struct {
 		Name, Body string
 	}{
@@ -35,34 +31,29 @@ func Example() {
 			Size: int64(len(file.Body)),
 		}
 		if err := tw.WriteHeader(hdr); err != nil {
-			log.Fatalln(err)
+			log.Fatal(err)
 		}
 		if _, err := tw.Write([]byte(file.Body)); err != nil {
-			log.Fatalln(err)
+			log.Fatal(err)
 		}
 	}
-	// Make sure to check the error on Close.
 	if err := tw.Close(); err != nil {
-		log.Fatalln(err)
+		log.Fatal(err)
 	}
 
-	// Open the tar archive for reading.
+	// Open and iterate through the files in the archive.
-	r := bytes.NewReader(buf.Bytes())
+	tr := tar.NewReader(&buf)
-	tr := tar.NewReader(r)
-
-	// Iterate through the files in the archive.
 	for {
 		hdr, err := tr.Next()
 		if err == io.EOF {
-			// end of tar archive
+			break // End of archive
-			break
 		}
 		if err != nil {
-			log.Fatalln(err)
+			log.Fatal(err)
 		}
 		fmt.Printf("Contents of %s:\n", hdr.Name)
 		if _, err := io.Copy(os.Stdout, tr); err != nil {
-			log.Fatalln(err)
+			log.Fatal(err)
 		}
 		fmt.Println()
 	}

archive/tar/format.go

@@ -0,0 +1,303 @@
+// Copyright 2016 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package tar
+
+import "strings"
+
+// Format represents the tar archive format.
+//
+// The original tar format was introduced in Unix V7.
+// Since then, there have been multiple competing formats attempting to
+// standardize or extend the V7 format to overcome its limitations.
+// The most common formats are the USTAR, PAX, and GNU formats,
+// each with their own advantages and limitations.
+//
+// The following table captures the capabilities of each format:
+//
+//	                  |  USTAR |       PAX |       GNU
+//	------------------+--------+-----------+----------
+//	Name              |   256B | unlimited | unlimited
+//	Linkname          |   100B | unlimited | unlimited
+//	Size              | uint33 | unlimited |    uint89
+//	Mode              | uint21 |    uint21 |    uint57
+//	Uid/Gid           | uint21 | unlimited |    uint57
+//	Uname/Gname       |    32B | unlimited |       32B
+//	ModTime           | uint33 | unlimited |     int89
+//	AccessTime        |    n/a | unlimited |     int89
+//	ChangeTime        |    n/a | unlimited |     int89
+//	Devmajor/Devminor | uint21 |    uint21 |    uint57
+//	------------------+--------+-----------+----------
+//	string encoding   |  ASCII |     UTF-8 |    binary
+//	sub-second times  |     no |       yes |        no
+//	sparse files      |     no |       yes |       yes
+//
+// The table's upper portion shows the Header fields, where each format reports
+// the maximum number of bytes allowed for each string field and
+// the integer type used to store each numeric field
+// (where timestamps are stored as the number of seconds since the Unix epoch).
+//
+// The table's lower portion shows specialized features of each format,
+// such as supported string encodings, support for sub-second timestamps,
+// or support for sparse files.
+//
+// The Writer currently provides no support for sparse files.
+type Format int
+
+// Constants to identify various tar formats.
+const (
+	// Deliberately hide the meaning of constants from public API.
+	_ Format = (1 << iota) / 4 // Sequence of 0, 0, 1, 2, 4, 8, etc...
+
+	// FormatUnknown indicates that the format is unknown.
+	FormatUnknown
+
+	// The format of the original Unix V7 tar tool prior to standardization.
+	formatV7
+
+	// FormatUSTAR represents the USTAR header format defined in POSIX.1-1988.
+	//
+	// While this format is compatible with most tar readers,
+	// the format has several limitations making it unsuitable for some usages.
+	// Most notably, it cannot support sparse files, files larger than 8GiB,
+	// filenames larger than 256 characters, and non-ASCII filenames.
+	//
+	// Reference:
+	//	http://pubs.opengroup.org/onlinepubs/9699919799/utilities/pax.html#tag_20_92_13_06
+	FormatUSTAR
+
+	// FormatPAX represents the PAX header format defined in POSIX.1-2001.
+	//
+	// PAX extends USTAR by writing a special file with Typeflag TypeXHeader
+	// preceding the original header. This file contains a set of key-value
+	// records, which are used to overcome USTAR's shortcomings, in addition to
+	// providing the ability to have sub-second resolution for timestamps.
+	//
+	// Some newer formats add their own extensions to PAX by defining their
+	// own keys and assigning certain semantic meaning to the associated values.
+	// For example, sparse file support in PAX is implemented using keys
+	// defined by the GNU manual (e.g., "GNU.sparse.map").
+	//
+	// Reference:
+	//	http://pubs.opengroup.org/onlinepubs/009695399/utilities/pax.html
+	FormatPAX
+
+	// FormatGNU represents the GNU header format.
+	//
+	// The GNU header format is older than the USTAR and PAX standards and
+	// is not compatible with them. The GNU format supports
+	// arbitrary file sizes, filenames of arbitrary encoding and length,
+	// sparse files, and other features.
+	//
+	// It is recommended that PAX be chosen over GNU unless the target
+	// application can only parse GNU formatted archives.
+	//
+	// Reference:
+	//	https://www.gnu.org/software/tar/manual/html_node/Standard.html
+	FormatGNU
+
+	// Schily's tar format, which is incompatible with USTAR.
+	// This does not cover STAR extensions to the PAX format; these fall under
+	// the PAX format.
+	formatSTAR
+
+	formatMax
+)
+
+func (f Format) has(f2 Format) bool   { return f&f2 != 0 }
+func (f *Format) mayBe(f2 Format)     { *f |= f2 }
+func (f *Format) mayOnlyBe(f2 Format) { *f &= f2 }
+func (f *Format) mustNotBe(f2 Format) { *f &^= f2 }
+
+var formatNames = map[Format]string{
+	formatV7: "V7", FormatUSTAR: "USTAR", FormatPAX: "PAX", FormatGNU: "GNU", formatSTAR: "STAR",
+}
+
+func (f Format) String() string {
+	var ss []string
+	for f2 := Format(1); f2 < formatMax; f2 <<= 1 {
+		if f.has(f2) {
+			ss = append(ss, formatNames[f2])
+		}
+	}
+	switch len(ss) {
+	case 0:
+		return "<unknown>"
+	case 1:
+		return ss[0]
+	default:
+		return "(" + strings.Join(ss, " | ") + ")"
+	}
+}
+
+// Magics used to identify various formats.
+const (
+	magicGNU, versionGNU     = "ustar ", " \x00"
+	magicUSTAR, versionUSTAR = "ustar\x00", "00"
+	trailerSTAR              = "tar\x00"
+)
+
+// Size constants from various tar specifications.
+const (
+	blockSize  = 512 // Size of each block in a tar stream
+	nameSize   = 100 // Max length of the name field in USTAR format
+	prefixSize = 155 // Max length of the prefix field in USTAR format
+)
+
+// blockPadding computes the number of bytes needed to pad offset up to the
+// nearest block edge where 0 <= n < blockSize.
+func blockPadding(offset int64) (n int64) {
+	return -offset & (blockSize - 1)
+}
+
+var zeroBlock block
+
+type block [blockSize]byte
+
+// Convert block to any number of formats.
+func (b *block) V7() *headerV7       { return (*headerV7)(b) }
+func (b *block) GNU() *headerGNU     { return (*headerGNU)(b) }
+func (b *block) STAR() *headerSTAR   { return (*headerSTAR)(b) }
+func (b *block) USTAR() *headerUSTAR { return (*headerUSTAR)(b) }
+func (b *block) Sparse() sparseArray { return (sparseArray)(b[:]) }
+
+// GetFormat checks that the block is a valid tar header based on the checksum.
+// It then attempts to guess the specific format based on magic values.
+// If the checksum fails, then FormatUnknown is returned.
+func (b *block) GetFormat() Format {
+	// Verify checksum.
+	var p parser
+	value := p.parseOctal(b.V7().Chksum())
+	chksum1, chksum2 := b.ComputeChecksum()
+	if p.err != nil || (value != chksum1 && value != chksum2) {
+		return FormatUnknown
+	}
+
+	// Guess the magic values.
+	magic := string(b.USTAR().Magic())
+	version := string(b.USTAR().Version())
+	trailer := string(b.STAR().Trailer())
+	switch {
+	case magic == magicUSTAR && trailer == trailerSTAR:
+		return formatSTAR
+	case magic == magicUSTAR:
+		return FormatUSTAR | FormatPAX
+	case magic == magicGNU && version == versionGNU:
+		return FormatGNU
+	default:
+		return formatV7
+	}
+}
+
+// SetFormat writes the magic values necessary for specified format
+// and then updates the checksum accordingly.
+func (b *block) SetFormat(format Format) {
+	// Set the magic values.
+	switch {
+	case format.has(formatV7):
+		// Do nothing.
+	case format.has(FormatGNU):
+		copy(b.GNU().Magic(), magicGNU)
+		copy(b.GNU().Version(), versionGNU)
+	case format.has(formatSTAR):
+		copy(b.STAR().Magic(), magicUSTAR)
+		copy(b.STAR().Version(), versionUSTAR)
+		copy(b.STAR().Trailer(), trailerSTAR)
+	case format.has(FormatUSTAR | FormatPAX):
+		copy(b.USTAR().Magic(), magicUSTAR)
+		copy(b.USTAR().Version(), versionUSTAR)
+	default:
+		panic("invalid format")
+	}
+
+	// Update checksum.
+	// This field is special in that it is terminated by a NULL then space.
+	var f formatter
+	field := b.V7().Chksum()
+	chksum, _ := b.ComputeChecksum() // Possible values are 256..128776
+	f.formatOctal(field[:7], chksum) // Never fails since 128776 < 262143
+	field[7] = ' '
+}
+
+// ComputeChecksum computes the checksum for the header block.
+// POSIX specifies a sum of the unsigned byte values, but the Sun tar used
+// signed byte values.
+// We compute and return both.
+func (b *block) ComputeChecksum() (unsigned, signed int64) {
+	for i, c := range b {
+		if 148 <= i && i < 156 {
+			c = ' ' // Treat the checksum field itself as all spaces.
+		}
+		unsigned += int64(c)
+		signed += int64(int8(c))
+	}
+	return unsigned, signed
+}
+
+// Reset clears the block with all zeros.
+func (b *block) Reset() {
+	*b = block{}
+}
+
+type headerV7 [blockSize]byte
+
+func (h *headerV7) Name() []byte     { return h[000:][:100] }
+func (h *headerV7) Mode() []byte     { return h[100:][:8] }
+func (h *headerV7) UID() []byte      { return h[108:][:8] }
+func (h *headerV7) GID() []byte      { return h[116:][:8] }
+func (h *headerV7) Size() []byte     { return h[124:][:12] }
+func (h *headerV7) ModTime() []byte  { return h[136:][:12] }
+func (h *headerV7) Chksum() []byte   { return h[148:][:8] }
+func (h *headerV7) TypeFlag() []byte { return h[156:][:1] }
+func (h *headerV7) LinkName() []byte { return h[157:][:100] }
+
+type headerGNU [blockSize]byte
+
+func (h *headerGNU) V7() *headerV7       { return (*headerV7)(h) }
+func (h *headerGNU) Magic() []byte       { return h[257:][:6] }
+func (h *headerGNU) Version() []byte     { return h[263:][:2] }
+func (h *headerGNU) UserName() []byte    { return h[265:][:32] }
+func (h *headerGNU) GroupName() []byte   { return h[297:][:32] }
+func (h *headerGNU) DevMajor() []byte    { return h[329:][:8] }
+func (h *headerGNU) DevMinor() []byte    { return h[337:][:8] }
+func (h *headerGNU) AccessTime() []byte  { return h[345:][:12] }
+func (h *headerGNU) ChangeTime() []byte  { return h[357:][:12] }
+func (h *headerGNU) Sparse() sparseArray { return (sparseArray)(h[386:][:24*4+1]) }
+func (h *headerGNU) RealSize() []byte    { return h[483:][:12] }
+
+type headerSTAR [blockSize]byte
+
+func (h *headerSTAR) V7() *headerV7      { return (*headerV7)(h) }
+func (h *headerSTAR) Magic() []byte      { return h[257:][:6] }
+func (h *headerSTAR) Version() []byte    { return h[263:][:2] }
+func (h *headerSTAR) UserName() []byte   { return h[265:][:32] }
+func (h *headerSTAR) GroupName() []byte  { return h[297:][:32] }
+func (h *headerSTAR) DevMajor() []byte   { return h[329:][:8] }
+func (h *headerSTAR) DevMinor() []byte   { return h[337:][:8] }
+func (h *headerSTAR) Prefix() []byte     { return h[345:][:131] }
+func (h *headerSTAR) AccessTime() []byte { return h[476:][:12] }
+func (h *headerSTAR) ChangeTime() []byte { return h[488:][:12] }
+func (h *headerSTAR) Trailer() []byte    { return h[508:][:4] }
+
+type headerUSTAR [blockSize]byte
+
+func (h *headerUSTAR) V7() *headerV7     { return (*headerV7)(h) }
+func (h *headerUSTAR) Magic() []byte     { return h[257:][:6] }
+func (h *headerUSTAR) Version() []byte   { return h[263:][:2] }
+func (h *headerUSTAR) UserName() []byte  { return h[265:][:32] }
+func (h *headerUSTAR) GroupName() []byte { return h[297:][:32] }
+func (h *headerUSTAR) DevMajor() []byte  { return h[329:][:8] }
+func (h *headerUSTAR) DevMinor() []byte  { return h[337:][:8] }
+func (h *headerUSTAR) Prefix() []byte    { return h[345:][:155] }
+
+type sparseArray []byte
+
+func (s sparseArray) Entry(i int) sparseElem { return (sparseElem)(s[i*24:]) }
+func (s sparseArray) IsExtended() []byte     { return s[24*s.MaxEntries():][:1] }
+func (s sparseArray) MaxEntries() int        { return len(s) / 24 }
+
+type sparseElem []byte
+
+func (s sparseElem) Offset() []byte { return s[00:][:12] }
+func (s sparseElem) Length() []byte { return s[12:][:12] }

archive/tar/stat_unix.go

@@ -8,6 +8,10 @@ package tar
 
 import (
 	"os"
+	"os/user"
+	"runtime"
+	"strconv"
+	"sync"
 	"syscall"
 )
 
@@ -15,6 +19,10 @@ func init() {
 	sysStat = statUnix
 }
 
+// userMap and groupMap caches UID and GID lookups for performance reasons.
+// The downside is that renaming uname or gname by the OS never takes effect.
+var userMap, groupMap sync.Map // map[int]string
+
 func statUnix(fi os.FileInfo, h *Header) error {
 	sys, ok := fi.Sys().(*syscall.Stat_t)
 	if !ok {
@@ -22,11 +30,67 @@ func statUnix(fi os.FileInfo, h *Header) error {
 	}
 	h.Uid = int(sys.Uid)
 	h.Gid = int(sys.Gid)
-	// TODO(bradfitz): populate username & group.  os/user
+
-	// doesn't cache LookupId lookups, and lacks group
+	// Best effort at populating Uname and Gname.
-	// lookup functions.
+	// The os/user functions may fail for any number of reasons
+	// (not implemented on that platform, cgo not enabled, etc).
+	if u, ok := userMap.Load(h.Uid); ok {
+		h.Uname = u.(string)
+	} else if u, err := user.LookupId(strconv.Itoa(h.Uid)); err == nil {
+		h.Uname = u.Username
+		userMap.Store(h.Uid, h.Uname)
+	}
+	if g, ok := groupMap.Load(h.Gid); ok {
+		h.Gname = g.(string)
+	} else if g, err := user.LookupGroupId(strconv.Itoa(h.Gid)); err == nil {
+		h.Gname = g.Name
+		groupMap.Store(h.Gid, h.Gname)
+	}
+
 	h.AccessTime = statAtime(sys)
 	h.ChangeTime = statCtime(sys)
-	// TODO(bradfitz): major/minor device numbers?
+
+	// Best effort at populating Devmajor and Devminor.
+	if h.Typeflag == TypeChar || h.Typeflag == TypeBlock {
+		dev := uint64(sys.Rdev) // May be int32 or uint32
+		switch runtime.GOOS {
+		case "linux":
+			// Copied from golang.org/x/sys/unix/dev_linux.go.
+			major := uint32((dev & 0x00000000000fff00) >> 8)
+			major |= uint32((dev & 0xfffff00000000000) >> 32)
+			minor := uint32((dev & 0x00000000000000ff) >> 0)
+			minor |= uint32((dev & 0x00000ffffff00000) >> 12)
+			h.Devmajor, h.Devminor = int64(major), int64(minor)
+		case "darwin":
+			// Copied from golang.org/x/sys/unix/dev_darwin.go.
+			major := uint32((dev >> 24) & 0xff)
+			minor := uint32(dev & 0xffffff)
+			h.Devmajor, h.Devminor = int64(major), int64(minor)
+		case "dragonfly":
+			// Copied from golang.org/x/sys/unix/dev_dragonfly.go.
+			major := uint32((dev >> 8) & 0xff)
+			minor := uint32(dev & 0xffff00ff)
+			h.Devmajor, h.Devminor = int64(major), int64(minor)
+		case "freebsd":
+			// Copied from golang.org/x/sys/unix/dev_freebsd.go.
+			major := uint32((dev >> 8) & 0xff)
+			minor := uint32(dev & 0xffff00ff)
+			h.Devmajor, h.Devminor = int64(major), int64(minor)
+		case "netbsd":
+			// Copied from golang.org/x/sys/unix/dev_netbsd.go.
+			major := uint32((dev & 0x000fff00) >> 8)
+			minor := uint32((dev & 0x000000ff) >> 0)
+			minor |= uint32((dev & 0xfff00000) >> 12)
+			h.Devmajor, h.Devminor = int64(major), int64(minor)
+		case "openbsd":
+			// Copied from golang.org/x/sys/unix/dev_openbsd.go.
+			major := uint32((dev & 0x0000ff00) >> 8)
+			minor := uint32((dev & 0x000000ff) >> 0)
+			minor |= uint32((dev & 0xffff0000) >> 8)
+			h.Devmajor, h.Devminor = int64(major), int64(minor)
+		default:
+			// TODO: Implement solaris (see https://golang.org/issue/8106)
+		}
+	}
 	return nil
 }

archive/tar/strconv.go

@@ -0,0 +1,326 @@
+// Copyright 2016 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package tar
+
+import (
+	"bytes"
+	"fmt"
+	"strconv"
+	"strings"
+	"time"
+)
+
+// hasNUL reports whether the NUL character exists within s.
+func hasNUL(s string) bool {
+	return strings.IndexByte(s, 0) >= 0
+}
+
+// isASCII reports whether the input is an ASCII C-style string.
+func isASCII(s string) bool {
+	for _, c := range s {
+		if c >= 0x80 || c == 0x00 {
+			return false
+		}
+	}
+	return true
+}
+
+// toASCII converts the input to an ASCII C-style string.
+// This a best effort conversion, so invalid characters are dropped.
+func toASCII(s string) string {
+	if isASCII(s) {
+		return s
+	}
+	b := make([]byte, 0, len(s))
+	for _, c := range s {
+		if c < 0x80 && c != 0x00 {
+			b = append(b, byte(c))
+		}
+	}
+	return string(b)
+}
+
+type parser struct {
+	err error // Last error seen
+}
+
+type formatter struct {
+	err error // Last error seen
+}
+
+// parseString parses bytes as a NUL-terminated C-style string.
+// If a NUL byte is not found then the whole slice is returned as a string.
+func (*parser) parseString(b []byte) string {
+	if i := bytes.IndexByte(b, 0); i >= 0 {
+		return string(b[:i])
+	}
+	return string(b)
+}
+
+// formatString copies s into b, NUL-terminating if possible.
+func (f *formatter) formatString(b []byte, s string) {
+	if len(s) > len(b) {
+		f.err = ErrFieldTooLong
+	}
+	copy(b, s)
+	if len(s) < len(b) {
+		b[len(s)] = 0
+	}
+
+	// Some buggy readers treat regular files with a trailing slash
+	// in the V7 path field as a directory even though the full path
+	// recorded elsewhere (e.g., via PAX record) contains no trailing slash.
+	if len(s) > len(b) && b[len(b)-1] == '/' {
+		n := len(strings.TrimRight(s[:len(b)], "/"))
+		b[n] = 0 // Replace trailing slash with NUL terminator
+	}
+}
+
+// fitsInBase256 reports whether x can be encoded into n bytes using base-256
+// encoding. Unlike octal encoding, base-256 encoding does not require that the
+// string ends with a NUL character. Thus, all n bytes are available for output.
+//
+// If operating in binary mode, this assumes strict GNU binary mode; which means
+// that the first byte can only be either 0x80 or 0xff. Thus, the first byte is
+// equivalent to the sign bit in two's complement form.
+func fitsInBase256(n int, x int64) bool {
+	binBits := uint(n-1) * 8
+	return n >= 9 || (x >= -1<<binBits && x < 1<<binBits)
+}
+
+// parseNumeric parses the input as being encoded in either base-256 or octal.
+// This function may return negative numbers.
+// If parsing fails or an integer overflow occurs, err will be set.
+func (p *parser) parseNumeric(b []byte) int64 {
+	// Check for base-256 (binary) format first.
+	// If the first bit is set, then all following bits constitute a two's
+	// complement encoded number in big-endian byte order.
+	if len(b) > 0 && b[0]&0x80 != 0 {
+		// Handling negative numbers relies on the following identity:
+		//	-a-1 == ^a
+		//
+		// If the number is negative, we use an inversion mask to invert the
+		// data bytes and treat the value as an unsigned number.
+		var inv byte // 0x00 if positive or zero, 0xff if negative
+		if b[0]&0x40 != 0 {
+			inv = 0xff
+		}
+
+		var x uint64
+		for i, c := range b {
+			c ^= inv // Inverts c only if inv is 0xff, otherwise does nothing
+			if i == 0 {
+				c &= 0x7f // Ignore signal bit in first byte
+			}
+			if (x >> 56) > 0 {
+				p.err = ErrHeader // Integer overflow
+				return 0
+			}
+			x = x<<8 | uint64(c)
+		}
+		if (x >> 63) > 0 {
+			p.err = ErrHeader // Integer overflow
+			return 0
+		}
+		if inv == 0xff {
+			return ^int64(x)
+		}
+		return int64(x)
+	}
+
+	// Normal case is base-8 (octal) format.
+	return p.parseOctal(b)
+}
+
+// formatNumeric encodes x into b using base-8 (octal) encoding if possible.
+// Otherwise it will attempt to use base-256 (binary) encoding.
+func (f *formatter) formatNumeric(b []byte, x int64) {
+	if fitsInOctal(len(b), x) {
+		f.formatOctal(b, x)
+		return
+	}
+
+	if fitsInBase256(len(b), x) {
+		for i := len(b) - 1; i >= 0; i-- {
+			b[i] = byte(x)
+			x >>= 8
+		}
+		b[0] |= 0x80 // Highest bit indicates binary format
+		return
+	}
+
+	f.formatOctal(b, 0) // Last resort, just write zero
+	f.err = ErrFieldTooLong
+}
+
+func (p *parser) parseOctal(b []byte) int64 {
+	// Because unused fields are filled with NULs, we need
+	// to skip leading NULs. Fields may also be padded with
+	// spaces or NULs.
+	// So we remove leading and trailing NULs and spaces to
+	// be sure.
+	b = bytes.Trim(b, " \x00")
+
+	if len(b) == 0 {
+		return 0
+	}
+	x, perr := strconv.ParseUint(p.parseString(b), 8, 64)
+	if perr != nil {
+		p.err = ErrHeader
+	}
+	return int64(x)
+}
+
+func (f *formatter) formatOctal(b []byte, x int64) {
+	if !fitsInOctal(len(b), x) {
+		x = 0 // Last resort, just write zero
+		f.err = ErrFieldTooLong
+	}
+
+	s := strconv.FormatInt(x, 8)
+	// Add leading zeros, but leave room for a NUL.
+	if n := len(b) - len(s) - 1; n > 0 {
+		s = strings.Repeat("0", n) + s
+	}
+	f.formatString(b, s)
+}
+
+// fitsInOctal reports whether the integer x fits in a field n-bytes long
+// using octal encoding with the appropriate NUL terminator.
+func fitsInOctal(n int, x int64) bool {
+	octBits := uint(n-1) * 3
+	return x >= 0 && (n >= 22 || x < 1<<octBits)
+}
+
+// parsePAXTime takes a string of the form %d.%d as described in the PAX
+// specification. Note that this implementation allows for negative timestamps,
+// which is allowed for by the PAX specification, but not always portable.
+func parsePAXTime(s string) (time.Time, error) {
+	const maxNanoSecondDigits = 9
+
+	// Split string into seconds and sub-seconds parts.
+	ss, sn := s, ""
+	if pos := strings.IndexByte(s, '.'); pos >= 0 {
+		ss, sn = s[:pos], s[pos+1:]
+	}
+
+	// Parse the seconds.
+	secs, err := strconv.ParseInt(ss, 10, 64)
+	if err != nil {
+		return time.Time{}, ErrHeader
+	}
+	if len(sn) == 0 {
+		return time.Unix(secs, 0), nil // No sub-second values
+	}
+
+	// Parse the nanoseconds.
+	if strings.Trim(sn, "0123456789") != "" {
+		return time.Time{}, ErrHeader
+	}
+	if len(sn) < maxNanoSecondDigits {
+		sn += strings.Repeat("0", maxNanoSecondDigits-len(sn)) // Right pad
+	} else {
+		sn = sn[:maxNanoSecondDigits] // Right truncate
+	}
+	nsecs, _ := strconv.ParseInt(sn, 10, 64) // Must succeed
+	if len(ss) > 0 && ss[0] == '-' {
+		return time.Unix(secs, -1*nsecs), nil // Negative correction
+	}
+	return time.Unix(secs, nsecs), nil
+}
+
+// formatPAXTime converts ts into a time of the form %d.%d as described in the
+// PAX specification. This function is capable of negative timestamps.
+func formatPAXTime(ts time.Time) (s string) {
+	secs, nsecs := ts.Unix(), ts.Nanosecond()
+	if nsecs == 0 {
+		return strconv.FormatInt(secs, 10)
+	}
+
+	// If seconds is negative, then perform correction.
+	sign := ""
+	if secs < 0 {
+		sign = "-"             // Remember sign
+		secs = -(secs + 1)     // Add a second to secs
+		nsecs = -(nsecs - 1E9) // Take that second away from nsecs
+	}
+	return strings.TrimRight(fmt.Sprintf("%s%d.%09d", sign, secs, nsecs), "0")
+}
+
+// parsePAXRecord parses the input PAX record string into a key-value pair.
+// If parsing is successful, it will slice off the currently read record and
+// return the remainder as r.
+func parsePAXRecord(s string) (k, v, r string, err error) {
+	// The size field ends at the first space.
+	sp := strings.IndexByte(s, ' ')
+	if sp == -1 {
+		return "", "", s, ErrHeader
+	}
+
+	// Parse the first token as a decimal integer.
+	n, perr := strconv.ParseInt(s[:sp], 10, 0) // Intentionally parse as native int
+	if perr != nil || n < 5 || int64(len(s)) < n {
+		return "", "", s, ErrHeader
+	}
+
+	// Extract everything between the space and the final newline.
+	rec, nl, rem := s[sp+1:n-1], s[n-1:n], s[n:]
+	if nl != "\n" {
+		return "", "", s, ErrHeader
+	}
+
+	// The first equals separates the key from the value.
+	eq := strings.IndexByte(rec, '=')
+	if eq == -1 {
+		return "", "", s, ErrHeader
+	}
+	k, v = rec[:eq], rec[eq+1:]
+
+	if !validPAXRecord(k, v) {
+		return "", "", s, ErrHeader
+	}
+	return k, v, rem, nil
+}
+
+// formatPAXRecord formats a single PAX record, prefixing it with the
+// appropriate length.
+func formatPAXRecord(k, v string) (string, error) {
+	if !validPAXRecord(k, v) {
+		return "", ErrHeader
+	}
+
+	const padding = 3 // Extra padding for ' ', '=', and '\n'
+	size := len(k) + len(v) + padding
+	size += len(strconv.Itoa(size))
+	record := strconv.Itoa(size) + " " + k + "=" + v + "\n"
+
+	// Final adjustment if adding size field increased the record size.
+	if len(record) != size {
+		size = len(record)
+		record = strconv.Itoa(size) + " " + k + "=" + v + "\n"
+	}
+	return record, nil
+}
+
+// validPAXRecord reports whether the key-value pair is valid where each
+// record is formatted as:
+//	"%d %s=%s\n" % (size, key, value)
+//
+// Keys and values should be UTF-8, but the number of bad writers out there
+// forces us to be a more liberal.
+// Thus, we only reject all keys with NUL, and only reject NULs in values
+// for the PAX version of the USTAR string fields.
+// The key must not contain an '=' character.
+func validPAXRecord(k, v string) bool {
+	if k == "" || strings.IndexByte(k, '=') >= 0 {
+		return false
+	}
+	switch k {
+	case paxPath, paxLinkpath, paxUname, paxGname:
+		return !hasNUL(v)
+	default:
+		return !hasNUL(k)
+	}
+}

archive/tar/strconv_test.go

@@ -0,0 +1,434 @@
+// Copyright 2016 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package tar
+
+import (
+	"math"
+	"strings"
+	"testing"
+	"time"
+)
+
+func TestFitsInBase256(t *testing.T) {
+	vectors := []struct {
+		in    int64
+		width int
+		ok    bool
+	}{
+		{+1, 8, true},
+		{0, 8, true},
+		{-1, 8, true},
+		{1 << 56, 8, false},
+		{(1 << 56) - 1, 8, true},
+		{-1 << 56, 8, true},
+		{(-1 << 56) - 1, 8, false},
+		{121654, 8, true},
+		{-9849849, 8, true},
+		{math.MaxInt64, 9, true},
+		{0, 9, true},
+		{math.MinInt64, 9, true},
+		{math.MaxInt64, 12, true},
+		{0, 12, true},
+		{math.MinInt64, 12, true},
+	}
+
+	for _, v := range vectors {
+		ok := fitsInBase256(v.width, v.in)
+		if ok != v.ok {
+			t.Errorf("fitsInBase256(%d, %d): got %v, want %v", v.in, v.width, ok, v.ok)
+		}
+	}
+}
+
+func TestParseNumeric(t *testing.T) {
+	vectors := []struct {
+		in   string
+		want int64
+		ok   bool
+	}{
+		// Test base-256 (binary) encoded values.
+		{"", 0, true},
+		{"\x80", 0, true},
+		{"\x80\x00", 0, true},
+		{"\x80\x00\x00", 0, true},
+		{"\xbf", (1 << 6) - 1, true},
+		{"\xbf\xff", (1 << 14) - 1, true},
+		{"\xbf\xff\xff", (1 << 22) - 1, true},
+		{"\xff", -1, true},
+		{"\xff\xff", -1, true},
+		{"\xff\xff\xff", -1, true},
+		{"\xc0", -1 * (1 << 6), true},
+		{"\xc0\x00", -1 * (1 << 14), true},
+		{"\xc0\x00\x00", -1 * (1 << 22), true},
+		{"\x87\x76\xa2\x22\xeb\x8a\x72\x61", 537795476381659745, true},
+		{"\x80\x00\x00\x00\x07\x76\xa2\x22\xeb\x8a\x72\x61", 537795476381659745, true},
+		{"\xf7\x76\xa2\x22\xeb\x8a\x72\x61", -615126028225187231, true},
+		{"\xff\xff\xff\xff\xf7\x76\xa2\x22\xeb\x8a\x72\x61", -615126028225187231, true},
+		{"\x80\x7f\xff\xff\xff\xff\xff\xff\xff", math.MaxInt64, true},
+		{"\x80\x80\x00\x00\x00\x00\x00\x00\x00", 0, false},
+		{"\xff\x80\x00\x00\x00\x00\x00\x00\x00", math.MinInt64, true},
+		{"\xff\x7f\xff\xff\xff\xff\xff\xff\xff", 0, false},
+		{"\xf5\xec\xd1\xc7\x7e\x5f\x26\x48\x81\x9f\x8f\x9b", 0, false},
+
+		// Test base-8 (octal) encoded values.
+		{"0000000\x00", 0, true},
+		{" \x0000000\x00", 0, true},
+		{" \x0000003\x00", 3, true},
+		{"00000000227\x00", 0227, true},
+		{"032033\x00 ", 032033, true},
+		{"320330\x00 ", 0320330, true},
+		{"0000660\x00 ", 0660, true},
+		{"\x00 0000660\x00 ", 0660, true},
+		{"0123456789abcdef", 0, false},
+		{"0123456789\x00abcdef", 0, false},
+		{"01234567\x0089abcdef", 342391, true},
+		{"0123\x7e\x5f\x264123", 0, false},
+	}
+
+	for _, v := range vectors {
+		var p parser
+		got := p.parseNumeric([]byte(v.in))
+		ok := (p.err == nil)
+		if ok != v.ok {
+			if v.ok {
+				t.Errorf("parseNumeric(%q): got parsing failure, want success", v.in)
+			} else {
+				t.Errorf("parseNumeric(%q): got parsing success, want failure", v.in)
+			}
+		}
+		if ok && got != v.want {
+			t.Errorf("parseNumeric(%q): got %d, want %d", v.in, got, v.want)
+		}
+	}
+}
+
+func TestFormatNumeric(t *testing.T) {
+	vectors := []struct {
+		in   int64
+		want string
+		ok   bool
+	}{
+		// Test base-8 (octal) encoded values.
+		{0, "0\x00", true},
+		{7, "7\x00", true},
+		{8, "\x80\x08", true},
+		{077, "77\x00", true},
+		{0100, "\x80\x00\x40", true},
+		{0, "0000000\x00", true},
+		{0123, "0000123\x00", true},
+		{07654321, "7654321\x00", true},
+		{07777777, "7777777\x00", true},
+		{010000000, "\x80\x00\x00\x00\x00\x20\x00\x00", true},
+		{0, "00000000000\x00", true},
+		{000001234567, "00001234567\x00", true},
+		{076543210321, "76543210321\x00", true},
+		{012345670123, "12345670123\x00", true},
+		{077777777777, "77777777777\x00", true},
+		{0100000000000, "\x80\x00\x00\x00\x00\x00\x00\x02\x00\x00\x00\x00", true},
+		{math.MaxInt64, "777777777777777777777\x00", true},
+
+		// Test base-256 (binary) encoded values.
+		{-1, "\xff", true},
+		{-1, "\xff\xff", true},
+		{-1, "\xff\xff\xff", true},
+		{(1 << 0), "0", false},
+		{(1 << 8) - 1, "\x80\xff", true},
+		{(1 << 8), "0\x00", false},
+		{(1 << 16) - 1, "\x80\xff\xff", true},
+		{(1 << 16), "00\x00", false},
+		{-1 * (1 << 0), "\xff", true},
+		{-1*(1<<0) - 1, "0", false},
+		{-1 * (1 << 8), "\xff\x00", true},
+		{-1*(1<<8) - 1, "0\x00", false},
+		{-1 * (1 << 16), "\xff\x00\x00", true},
+		{-1*(1<<16) - 1, "00\x00", false},
+		{537795476381659745, "0000000\x00", false},
+		{537795476381659745, "\x80\x00\x00\x00\x07\x76\xa2\x22\xeb\x8a\x72\x61", true},
+		{-615126028225187231, "0000000\x00", false},
+		{-615126028225187231, "\xff\xff\xff\xff\xf7\x76\xa2\x22\xeb\x8a\x72\x61", true},
+		{math.MaxInt64, "0000000\x00", false},
+		{math.MaxInt64, "\x80\x00\x00\x00\x7f\xff\xff\xff\xff\xff\xff\xff", true},
+		{math.MinInt64, "0000000\x00", false},
+		{math.MinInt64, "\xff\xff\xff\xff\x80\x00\x00\x00\x00\x00\x00\x00", true},
+		{math.MaxInt64, "\x80\x7f\xff\xff\xff\xff\xff\xff\xff", true},
+		{math.MinInt64, "\xff\x80\x00\x00\x00\x00\x00\x00\x00", true},
+	}
+
+	for _, v := range vectors {
+		var f formatter
+		got := make([]byte, len(v.want))
+		f.formatNumeric(got, v.in)
+		ok := (f.err == nil)
+		if ok != v.ok {
+			if v.ok {
+				t.Errorf("formatNumeric(%d): got formatting failure, want success", v.in)
+			} else {
+				t.Errorf("formatNumeric(%d): got formatting success, want failure", v.in)
+			}
+		}
+		if string(got) != v.want {
+			t.Errorf("formatNumeric(%d): got %q, want %q", v.in, got, v.want)
+		}
+	}
+}
+
+func TestFitsInOctal(t *testing.T) {
+	vectors := []struct {
+		input int64
+		width int
+		ok    bool
+	}{
+		{-1, 1, false},
+		{-1, 2, false},
+		{-1, 3, false},
+		{0, 1, true},
+		{0 + 1, 1, false},
+		{0, 2, true},
+		{07, 2, true},
+		{07 + 1, 2, false},
+		{0, 4, true},
+		{0777, 4, true},
+		{0777 + 1, 4, false},
+		{0, 8, true},
+		{07777777, 8, true},
+		{07777777 + 1, 8, false},
+		{0, 12, true},
+		{077777777777, 12, true},
+		{077777777777 + 1, 12, false},
+		{math.MaxInt64, 22, true},
+		{012345670123, 12, true},
+		{01564164, 12, true},
+		{-012345670123, 12, false},
+		{-01564164, 12, false},
+		{-1564164, 30, false},
+	}
+
+	for _, v := range vectors {
+		ok := fitsInOctal(v.width, v.input)
+		if ok != v.ok {
+			t.Errorf("checkOctal(%d, %d): got %v, want %v", v.input, v.width, ok, v.ok)
+		}
+	}
+}
+
+func TestParsePAXTime(t *testing.T) {
+	vectors := []struct {
+		in   string
+		want time.Time
+		ok   bool
+	}{
+		{"1350244992.023960108", time.Unix(1350244992, 23960108), true},
+		{"1350244992.02396010", time.Unix(1350244992, 23960100), true},
+		{"1350244992.0239601089", time.Unix(1350244992, 23960108), true},
+		{"1350244992.3", time.Unix(1350244992, 300000000), true},
+		{"1350244992", time.Unix(1350244992, 0), true},
+		{"-1.000000001", time.Unix(-1, -1e0+0e0), true},
+		{"-1.000001", time.Unix(-1, -1e3+0e0), true},
+		{"-1.001000", time.Unix(-1, -1e6+0e0), true},
+		{"-1", time.Unix(-1, -0e0+0e0), true},
+		{"-1.999000", time.Unix(-1, -1e9+1e6), true},
+		{"-1.999999", time.Unix(-1, -1e9+1e3), true},
+		{"-1.999999999", time.Unix(-1, -1e9+1e0), true},
+		{"0.000000001", time.Unix(0, 1e0+0e0), true},
+		{"0.000001", time.Unix(0, 1e3+0e0), true},
+		{"0.001000", time.Unix(0, 1e6+0e0), true},
+		{"0", time.Unix(0, 0e0), true},
+		{"0.999000", time.Unix(0, 1e9-1e6), true},
+		{"0.999999", time.Unix(0, 1e9-1e3), true},
+		{"0.999999999", time.Unix(0, 1e9-1e0), true},
+		{"1.000000001", time.Unix(+1, +1e0-0e0), true},
+		{"1.000001", time.Unix(+1, +1e3-0e0), true},
+		{"1.001000", time.Unix(+1, +1e6-0e0), true},
+		{"1", time.Unix(+1, +0e0-0e0), true},
+		{"1.999000", time.Unix(+1, +1e9-1e6), true},
+		{"1.999999", time.Unix(+1, +1e9-1e3), true},
+		{"1.999999999", time.Unix(+1, +1e9-1e0), true},
+		{"-1350244992.023960108", time.Unix(-1350244992, -23960108), true},
+		{"-1350244992.02396010", time.Unix(-1350244992, -23960100), true},
+		{"-1350244992.0239601089", time.Unix(-1350244992, -23960108), true},
+		{"-1350244992.3", time.Unix(-1350244992, -300000000), true},
+		{"-1350244992", time.Unix(-1350244992, 0), true},
+		{"", time.Time{}, false},
+		{"0", time.Unix(0, 0), true},
+		{"1.", time.Unix(1, 0), true},
+		{"0.0", time.Unix(0, 0), true},
+		{".5", time.Time{}, false},
+		{"-1.3", time.Unix(-1, -3e8), true},
+		{"-1.0", time.Unix(-1, -0e0), true},
+		{"-0.0", time.Unix(-0, -0e0), true},
+		{"-0.1", time.Unix(-0, -1e8), true},
+		{"-0.01", time.Unix(-0, -1e7), true},
+		{"-0.99", time.Unix(-0, -99e7), true},
+		{"-0.98", time.Unix(-0, -98e7), true},
+		{"-1.1", time.Unix(-1, -1e8), true},
+		{"-1.01", time.Unix(-1, -1e7), true},
+		{"-2.99", time.Unix(-2, -99e7), true},
+		{"-5.98", time.Unix(-5, -98e7), true},
+		{"-", time.Time{}, false},
+		{"+", time.Time{}, false},
+		{"-1.-1", time.Time{}, false},
+		{"99999999999999999999999999999999999999999999999", time.Time{}, false},
+		{"0.123456789abcdef", time.Time{}, false},
+		{"foo", time.Time{}, false},
+		{"\x00", time.Time{}, false},
+		{"𝟵𝟴𝟳𝟲𝟱.𝟰𝟯𝟮𝟭𝟬", time.Time{}, false}, // Unicode numbers (U+1D7EC to U+1D7F5)
+		{"98765﹒43210", time.Time{}, false}, // Unicode period (U+FE52)
+	}
+
+	for _, v := range vectors {
+		ts, err := parsePAXTime(v.in)
+		ok := (err == nil)
+		if v.ok != ok {
+			if v.ok {
+				t.Errorf("parsePAXTime(%q): got parsing failure, want success", v.in)
+			} else {
+				t.Errorf("parsePAXTime(%q): got parsing success, want failure", v.in)
+			}
+		}
+		if ok && !ts.Equal(v.want) {
+			t.Errorf("parsePAXTime(%q): got (%ds %dns), want (%ds %dns)",
+				v.in, ts.Unix(), ts.Nanosecond(), v.want.Unix(), v.want.Nanosecond())
+		}
+	}
+}
+
+func TestFormatPAXTime(t *testing.T) {
+	vectors := []struct {
+		sec, nsec int64
+		want      string
+	}{
+		{1350244992, 0, "1350244992"},
+		{1350244992, 300000000, "1350244992.3"},
+		{1350244992, 23960100, "1350244992.0239601"},
+		{1350244992, 23960108, "1350244992.023960108"},
+		{+1, +1E9 - 1E0, "1.999999999"},
+		{+1, +1E9 - 1E3, "1.999999"},
+		{+1, +1E9 - 1E6, "1.999"},
+		{+1, +0E0 - 0E0, "1"},
+		{+1, +1E6 - 0E0, "1.001"},
+		{+1, +1E3 - 0E0, "1.000001"},
+		{+1, +1E0 - 0E0, "1.000000001"},
+		{0, 1E9 - 1E0, "0.999999999"},
+		{0, 1E9 - 1E3, "0.999999"},
+		{0, 1E9 - 1E6, "0.999"},
+		{0, 0E0, "0"},
+		{0, 1E6 + 0E0, "0.001"},
+		{0, 1E3 + 0E0, "0.000001"},
+		{0, 1E0 + 0E0, "0.000000001"},
+		{-1, -1E9 + 1E0, "-1.999999999"},
+		{-1, -1E9 + 1E3, "-1.999999"},
+		{-1, -1E9 + 1E6, "-1.999"},
+		{-1, -0E0 + 0E0, "-1"},
+		{-1, -1E6 + 0E0, "-1.001"},
+		{-1, -1E3 + 0E0, "-1.000001"},
+		{-1, -1E0 + 0E0, "-1.000000001"},
+		{-1350244992, 0, "-1350244992"},
+		{-1350244992, -300000000, "-1350244992.3"},
+		{-1350244992, -23960100, "-1350244992.0239601"},
+		{-1350244992, -23960108, "-1350244992.023960108"},
+	}
+
+	for _, v := range vectors {
+		got := formatPAXTime(time.Unix(v.sec, v.nsec))
+		if got != v.want {
+			t.Errorf("formatPAXTime(%ds, %dns): got %q, want %q",
+				v.sec, v.nsec, got, v.want)
+		}
+	}
+}
+
+func TestParsePAXRecord(t *testing.T) {
+	medName := strings.Repeat("CD", 50)
+	longName := strings.Repeat("AB", 100)
+
+	vectors := []struct {
+		in      string
+		wantRes string
+		wantKey string
+		wantVal string
+		ok      bool
+	}{
+		{"6 k=v\n\n", "\n", "k", "v", true},
+		{"19 path=/etc/hosts\n", "", "path", "/etc/hosts", true},
+		{"210 path=" + longName + "\nabc", "abc", "path", longName, true},
+		{"110 path=" + medName + "\n", "", "path", medName, true},
+		{"9 foo=ba\n", "", "foo", "ba", true},
+		{"11 foo=bar\n\x00", "\x00", "foo", "bar", true},
+		{"18 foo=b=\nar=\n==\x00\n", "", "foo", "b=\nar=\n==\x00", true},
+		{"27 foo=hello9 foo=ba\nworld\n", "", "foo", "hello9 foo=ba\nworld", true},
+		{"27 ☺☻☹=日a本b語ç\nmeow mix", "meow mix", "☺☻☹", "日a本b語ç", true},
+		{"17 \x00hello=\x00world\n", "17 \x00hello=\x00world\n", "", "", false},
+		{"1 k=1\n", "1 k=1\n", "", "", false},
+		{"6 k~1\n", "6 k~1\n", "", "", false},
+		{"6_k=1\n", "6_k=1\n", "", "", false},
+		{"6 k=1 ", "6 k=1 ", "", "", false},
+		{"632 k=1\n", "632 k=1\n", "", "", false},
+		{"16 longkeyname=hahaha\n", "16 longkeyname=hahaha\n", "", "", false},
+		{"3 somelongkey=\n", "3 somelongkey=\n", "", "", false},
+		{"50 tooshort=\n", "50 tooshort=\n", "", "", false},
+	}
+
+	for _, v := range vectors {
+		key, val, res, err := parsePAXRecord(v.in)
+		ok := (err == nil)
+		if ok != v.ok {
+			if v.ok {
+				t.Errorf("parsePAXRecord(%q): got parsing failure, want success", v.in)
+			} else {
+				t.Errorf("parsePAXRecord(%q): got parsing success, want failure", v.in)
+			}
+		}
+		if v.ok && (key != v.wantKey || val != v.wantVal) {
+			t.Errorf("parsePAXRecord(%q): got (%q: %q), want (%q: %q)",
+				v.in, key, val, v.wantKey, v.wantVal)
+		}
+		if res != v.wantRes {
+			t.Errorf("parsePAXRecord(%q): got residual %q, want residual %q",
+				v.in, res, v.wantRes)
+		}
+	}
+}
+
+func TestFormatPAXRecord(t *testing.T) {
+	medName := strings.Repeat("CD", 50)
+	longName := strings.Repeat("AB", 100)
+
+	vectors := []struct {
+		inKey string
+		inVal string
+		want  string
+		ok    bool
+	}{
+		{"k", "v", "6 k=v\n", true},
+		{"path", "/etc/hosts", "19 path=/etc/hosts\n", true},
+		{"path", longName, "210 path=" + longName + "\n", true},
+		{"path", medName, "110 path=" + medName + "\n", true},
+		{"foo", "ba", "9 foo=ba\n", true},
+		{"foo", "bar", "11 foo=bar\n", true},
+		{"foo", "b=\nar=\n==\x00", "18 foo=b=\nar=\n==\x00\n", true},
+		{"foo", "hello9 foo=ba\nworld", "27 foo=hello9 foo=ba\nworld\n", true},
+		{"☺☻☹", "日a本b語ç", "27 ☺☻☹=日a本b語ç\n", true},
+		{"xhello", "\x00world", "17 xhello=\x00world\n", true},
+		{"path", "null\x00", "", false},
+		{"null\x00", "value", "", false},
+		{paxSchilyXattr + "key", "null\x00", "26 SCHILY.xattr.key=null\x00\n", true},
+	}
+
+	for _, v := range vectors {
+		got, err := formatPAXRecord(v.inKey, v.inVal)
+		ok := (err == nil)
+		if ok != v.ok {
+			if v.ok {
+				t.Errorf("formatPAXRecord(%q, %q): got format failure, want success", v.inKey, v.inVal)
+			} else {
+				t.Errorf("formatPAXRecord(%q, %q): got format success, want failure", v.inKey, v.inVal)
+			}
+		}
+		if got != v.want {
+			t.Errorf("formatPAXRecord(%q, %q): got %q, want %q",
+				v.inKey, v.inVal, got, v.want)
+		}
+	}
+}

archive/tar/tar_test.go

@@ -6,15 +6,208 @@ package tar
 
 import (
 	"bytes"
+	"errors"
+	"fmt"
+	"io"
 	"io/ioutil"
+	"math"
 	"os"
 	"path"
+	"path/filepath"
 	"reflect"
+	"runtime"
 	"strings"
 	"testing"
 	"time"
 )
 
+type testError struct{ error }
+
+type fileOps []interface{} // []T where T is (string | int64)
+
+// testFile is an io.ReadWriteSeeker where the IO operations performed
+// on it must match the list of operations in ops.
+type testFile struct {
+	ops fileOps
+	pos int64
+}
+
+func (f *testFile) Read(b []byte) (int, error) {
+	if len(b) == 0 {
+		return 0, nil
+	}
+	if len(f.ops) == 0 {
+		return 0, io.EOF
+	}
+	s, ok := f.ops[0].(string)
+	if !ok {
+		return 0, errors.New("unexpected Read operation")
+	}
+
+	n := copy(b, s)
+	if len(s) > n {
+		f.ops[0] = s[n:]
+	} else {
+		f.ops = f.ops[1:]
+	}
+	f.pos += int64(len(b))
+	return n, nil
+}
+
+func (f *testFile) Write(b []byte) (int, error) {
+	if len(b) == 0 {
+		return 0, nil
+	}
+	if len(f.ops) == 0 {
+		return 0, errors.New("unexpected Write operation")
+	}
+	s, ok := f.ops[0].(string)
+	if !ok {
+		return 0, errors.New("unexpected Write operation")
+	}
+
+	if !strings.HasPrefix(s, string(b)) {
+		return 0, testError{fmt.Errorf("got Write(%q), want Write(%q)", b, s)}
+	}
+	if len(s) > len(b) {
+		f.ops[0] = s[len(b):]
+	} else {
+		f.ops = f.ops[1:]
+	}
+	f.pos += int64(len(b))
+	return len(b), nil
+}
+
+func (f *testFile) Seek(pos int64, whence int) (int64, error) {
+	if pos == 0 && whence == io.SeekCurrent {
+		return f.pos, nil
+	}
+	if len(f.ops) == 0 {
+		return 0, errors.New("unexpected Seek operation")
+	}
+	s, ok := f.ops[0].(int64)
+	if !ok {
+		return 0, errors.New("unexpected Seek operation")
+	}
+
+	if s != pos || whence != io.SeekCurrent {
+		return 0, testError{fmt.Errorf("got Seek(%d, %d), want Seek(%d, %d)", pos, whence, s, io.SeekCurrent)}
+	}
+	f.pos += s
+	f.ops = f.ops[1:]
+	return f.pos, nil
+}
+
+func equalSparseEntries(x, y []sparseEntry) bool {
+	return (len(x) == 0 && len(y) == 0) || reflect.DeepEqual(x, y)
+}
+
+func TestSparseEntries(t *testing.T) {
+	vectors := []struct {
+		in   []sparseEntry
+		size int64
+
+		wantValid    bool          // Result of validateSparseEntries
+		wantAligned  []sparseEntry // Result of alignSparseEntries
+		wantInverted []sparseEntry // Result of invertSparseEntries
+	}{{
+		in: []sparseEntry{}, size: 0,
+		wantValid:    true,
+		wantInverted: []sparseEntry{{0, 0}},
+	}, {
+		in: []sparseEntry{}, size: 5000,
+		wantValid:    true,
+		wantInverted: []sparseEntry{{0, 5000}},
+	}, {
+		in: []sparseEntry{{0, 5000}}, size: 5000,
+		wantValid:    true,
+		wantAligned:  []sparseEntry{{0, 5000}},
+		wantInverted: []sparseEntry{{5000, 0}},
+	}, {
+		in: []sparseEntry{{1000, 4000}}, size: 5000,
+		wantValid:    true,
+		wantAligned:  []sparseEntry{{1024, 3976}},
+		wantInverted: []sparseEntry{{0, 1000}, {5000, 0}},
+	}, {
+		in: []sparseEntry{{0, 3000}}, size: 5000,
+		wantValid:    true,
+		wantAligned:  []sparseEntry{{0, 2560}},
+		wantInverted: []sparseEntry{{3000, 2000}},
+	}, {
+		in: []sparseEntry{{3000, 2000}}, size: 5000,
+		wantValid:    true,
+		wantAligned:  []sparseEntry{{3072, 1928}},
+		wantInverted: []sparseEntry{{0, 3000}, {5000, 0}},
+	}, {
+		in: []sparseEntry{{2000, 2000}}, size: 5000,
+		wantValid:    true,
+		wantAligned:  []sparseEntry{{2048, 1536}},
+		wantInverted: []sparseEntry{{0, 2000}, {4000, 1000}},
+	}, {
+		in: []sparseEntry{{0, 2000}, {8000, 2000}}, size: 10000,
+		wantValid:    true,
+		wantAligned:  []sparseEntry{{0, 1536}, {8192, 1808}},
+		wantInverted: []sparseEntry{{2000, 6000}, {10000, 0}},
+	}, {
+		in: []sparseEntry{{0, 2000}, {2000, 2000}, {4000, 0}, {4000, 3000}, {7000, 1000}, {8000, 0}, {8000, 2000}}, size: 10000,
+		wantValid:    true,
+		wantAligned:  []sparseEntry{{0, 1536}, {2048, 1536}, {4096, 2560}, {7168, 512}, {8192, 1808}},
+		wantInverted: []sparseEntry{{10000, 0}},
+	}, {
+		in: []sparseEntry{{0, 0}, {1000, 0}, {2000, 0}, {3000, 0}, {4000, 0}, {5000, 0}}, size: 5000,
+		wantValid:    true,
+		wantInverted: []sparseEntry{{0, 5000}},
+	}, {
+		in: []sparseEntry{{1, 0}}, size: 0,
+		wantValid: false,
+	}, {
+		in: []sparseEntry{{-1, 0}}, size: 100,
+		wantValid: false,
+	}, {
+		in: []sparseEntry{{0, -1}}, size: 100,
+		wantValid: false,
+	}, {
+		in: []sparseEntry{{0, 0}}, size: -100,
+		wantValid: false,
+	}, {
+		in: []sparseEntry{{math.MaxInt64, 3}, {6, -5}}, size: 35,
+		wantValid: false,
+	}, {
+		in: []sparseEntry{{1, 3}, {6, -5}}, size: 35,
+		wantValid: false,
+	}, {
+		in: []sparseEntry{{math.MaxInt64, math.MaxInt64}}, size: math.MaxInt64,
+		wantValid: false,
+	}, {
+		in: []sparseEntry{{3, 3}}, size: 5,
+		wantValid: false,
+	}, {
+		in: []sparseEntry{{2, 0}, {1, 0}, {0, 0}}, size: 3,
+		wantValid: false,
+	}, {
+		in: []sparseEntry{{1, 3}, {2, 2}}, size: 10,
+		wantValid: false,
+	}}
+
+	for i, v := range vectors {
+		gotValid := validateSparseEntries(v.in, v.size)
+		if gotValid != v.wantValid {
+			t.Errorf("test %d, validateSparseEntries() = %v, want %v", i, gotValid, v.wantValid)
+		}
+		if !v.wantValid {
+			continue
+		}
+		gotAligned := alignSparseEntries(append([]sparseEntry{}, v.in...), v.size)
+		if !equalSparseEntries(gotAligned, v.wantAligned) {
+			t.Errorf("test %d, alignSparseEntries():\ngot  %v\nwant %v", i, gotAligned, v.wantAligned)
+		}
+		gotInverted := invertSparseEntries(append([]sparseEntry{}, v.in...), v.size)
+		if !equalSparseEntries(gotInverted, v.wantInverted) {
+			t.Errorf("test %d, inverseSparseEntries():\ngot  %v\nwant %v", i, gotInverted, v.wantInverted)
+		}
+	}
+}
+
 func TestFileInfoHeader(t *testing.T) {
 	fi, err := os.Stat("testdata/small.txt")
 	if err != nil {
@@ -27,7 +220,7 @@ func TestFileInfoHeader(t *testing.T) {
 	if g, e := h.Name, "small.txt"; g != e {
 		t.Errorf("Name = %q; want %q", g, e)
 	}
-	if g, e := h.Mode, int64(fi.Mode().Perm())|c_ISREG; g != e {
+	if g, e := h.Mode, int64(fi.Mode().Perm()); g != e {
 		t.Errorf("Mode = %#o; want %#o", g, e)
 	}
 	if g, e := h.Size, int64(5); g != e {
@@ -55,7 +248,7 @@ func TestFileInfoHeaderDir(t *testing.T) {
 		t.Errorf("Name = %q; want %q", g, e)
 	}
 	// Ignoring c_ISGID for golang.org/issue/4867
-	if g, e := h.Mode&^c_ISGID, int64(fi.Mode().Perm())|c_ISDIR; g != e {
+	if g, e := h.Mode&^c_ISGID, int64(fi.Mode().Perm()); g != e {
 		t.Errorf("Mode = %#o; want %#o", g, e)
 	}
 	if g, e := h.Size, int64(0); g != e {
@@ -67,38 +260,55 @@ func TestFileInfoHeaderDir(t *testing.T) {
 }
 
 func TestFileInfoHeaderSymlink(t *testing.T) {
-	h, err := FileInfoHeader(symlink{}, "some-target")
+	switch runtime.GOOS {
+	case "android", "nacl", "plan9", "windows":
+		t.Skip("symlinks not supported")
+	}
+	tmpdir, err := ioutil.TempDir("", "TestFileInfoHeaderSymlink")
 	if err != nil {
 		t.Fatal(err)
 	}
-	if g, e := h.Name, "some-symlink"; g != e {
+	defer os.RemoveAll(tmpdir)
+
+	link := filepath.Join(tmpdir, "link")
+	target := tmpdir
+	err = os.Symlink(target, link)
+	if err != nil {
+		t.Fatal(err)
+	}
+	fi, err := os.Lstat(link)
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	h, err := FileInfoHeader(fi, target)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if g, e := h.Name, fi.Name(); g != e {
 		t.Errorf("Name = %q; want %q", g, e)
 	}
-	if g, e := h.Linkname, "some-target"; g != e {
+	if g, e := h.Linkname, target; g != e {
 		t.Errorf("Linkname = %q; want %q", g, e)
 	}
+	if g, e := h.Typeflag, byte(TypeSymlink); g != e {
+		t.Errorf("Typeflag = %v; want %v", g, e)
+	}
 }
 
-type symlink struct{}
-
-func (symlink) Name() string       { return "some-symlink" }
-func (symlink) Size() int64        { return 0 }
-func (symlink) Mode() os.FileMode  { return os.ModeSymlink }
-func (symlink) ModTime() time.Time { return time.Time{} }
-func (symlink) IsDir() bool        { return false }
-func (symlink) Sys() interface{}   { return nil }
-
 func TestRoundTrip(t *testing.T) {
 	data := []byte("some file contents")
 
 	var b bytes.Buffer
 	tw := NewWriter(&b)
 	hdr := &Header{
-		Name: "file.txt",
+		Name:       "file.txt",
-		Uid:  1 << 21, // too big for 8 octal digits
+		Uid:        1 << 21, // Too big for 8 octal digits
-		Size: int64(len(data)),
+		Size:       int64(len(data)),
-		// https://github.com/golang/go/commit/0e3355903d2ebcf5ee9e76096f51ac9a116a9dbb#diff-d7bf2a98d7b57b6ff754ca406f1b7581R105
+		ModTime:    time.Now().Round(time.Second),
-		ModTime: time.Now().AddDate(0, 0, 0).Round(1 * time.Second),
+		PAXRecords: map[string]string{"uid": "2097152"},
+		Format:     FormatPAX,
+		Typeflag:   TypeReg,
 	}
 	if err := tw.WriteHeader(hdr); err != nil {
 		t.Fatalf("tw.WriteHeader: %v", err)
@@ -134,191 +344,514 @@ type headerRoundTripTest struct {
 }
 
 func TestHeaderRoundTrip(t *testing.T) {
-	golden := []headerRoundTripTest{
+	vectors := []headerRoundTripTest{{
 		// regular file.
-		{
+		h: &Header{
-			h: &Header{
+			Name:     "test.txt",
-				Name:     "test.txt",
+			Mode:     0644,
-				Mode:     0644 | c_ISREG,
+			Size:     12,
-				Size:     12,
+			ModTime:  time.Unix(1360600916, 0),
-				ModTime:  time.Unix(1360600916, 0),
+			Typeflag: TypeReg,
-				Typeflag: TypeReg,
-			},
-			fm: 0644,
 		},
+		fm: 0644,
+	}, {
 		// symbolic link.
-		{
+		h: &Header{
-			h: &Header{
+			Name:     "link.txt",
-				Name:     "link.txt",
+			Mode:     0777,
-				Mode:     0777 | c_ISLNK,
+			Size:     0,
-				Size:     0,
+			ModTime:  time.Unix(1360600852, 0),
-				ModTime:  time.Unix(1360600852, 0),
+			Typeflag: TypeSymlink,
-				Typeflag: TypeSymlink,
-			},
-			fm: 0777 | os.ModeSymlink,
 		},
+		fm: 0777 | os.ModeSymlink,
+	}, {
 		// character device node.
-		{
+		h: &Header{
-			h: &Header{
+			Name:     "dev/null",
-				Name:     "dev/null",
+			Mode:     0666,
-				Mode:     0666 | c_ISCHR,
+			Size:     0,
-				Size:     0,
+			ModTime:  time.Unix(1360578951, 0),
-				ModTime:  time.Unix(1360578951, 0),
+			Typeflag: TypeChar,
-				Typeflag: TypeChar,
-			},
-			fm: 0666 | os.ModeDevice | os.ModeCharDevice,
 		},
+		fm: 0666 | os.ModeDevice | os.ModeCharDevice,
+	}, {
 		// block device node.
-		{
+		h: &Header{
-			h: &Header{
+			Name:     "dev/sda",
-				Name:     "dev/sda",
+			Mode:     0660,
-				Mode:     0660 | c_ISBLK,
+			Size:     0,
-				Size:     0,
+			ModTime:  time.Unix(1360578954, 0),
-				ModTime:  time.Unix(1360578954, 0),
+			Typeflag: TypeBlock,
-				Typeflag: TypeBlock,
-			},
-			fm: 0660 | os.ModeDevice,
 		},
+		fm: 0660 | os.ModeDevice,
+	}, {
 		// directory.
-		{
+		h: &Header{
-			h: &Header{
+			Name:     "dir/",
-				Name:     "dir/",
+			Mode:     0755,
-				Mode:     0755 | c_ISDIR,
+			Size:     0,
-				Size:     0,
+			ModTime:  time.Unix(1360601116, 0),
-				ModTime:  time.Unix(1360601116, 0),
+			Typeflag: TypeDir,
-				Typeflag: TypeDir,
-			},
-			fm: 0755 | os.ModeDir,
 		},
+		fm: 0755 | os.ModeDir,
+	}, {
 		// fifo node.
-		{
+		h: &Header{
-			h: &Header{
+			Name:     "dev/initctl",
-				Name:     "dev/initctl",
+			Mode:     0600,
-				Mode:     0600 | c_ISFIFO,
+			Size:     0,
-				Size:     0,
+			ModTime:  time.Unix(1360578949, 0),
-				ModTime:  time.Unix(1360578949, 0),
+			Typeflag: TypeFifo,
-				Typeflag: TypeFifo,
-			},
-			fm: 0600 | os.ModeNamedPipe,
 		},
+		fm: 0600 | os.ModeNamedPipe,
+	}, {
 		// setuid.
-		{
+		h: &Header{
-			h: &Header{
+			Name:     "bin/su",
-				Name:     "bin/su",
+			Mode:     0755 | c_ISUID,
-				Mode:     0755 | c_ISREG | c_ISUID,
+			Size:     23232,
-				Size:     23232,
+			ModTime:  time.Unix(1355405093, 0),
-				ModTime:  time.Unix(1355405093, 0),
+			Typeflag: TypeReg,
-				Typeflag: TypeReg,
-			},
-			fm: 0755 | os.ModeSetuid,
 		},
+		fm: 0755 | os.ModeSetuid,
+	}, {
 		// setguid.
-		{
+		h: &Header{
-			h: &Header{
+			Name:     "group.txt",
-				Name:     "group.txt",
+			Mode:     0750 | c_ISGID,
-				Mode:     0750 | c_ISREG | c_ISGID,
+			Size:     0,
-				Size:     0,
+			ModTime:  time.Unix(1360602346, 0),
-				ModTime:  time.Unix(1360602346, 0),
+			Typeflag: TypeReg,
-				Typeflag: TypeReg,
-			},
-			fm: 0750 | os.ModeSetgid,
 		},
+		fm: 0750 | os.ModeSetgid,
+	}, {
 		// sticky.
-		{
+		h: &Header{
-			h: &Header{
+			Name:     "sticky.txt",
-				Name:     "sticky.txt",
+			Mode:     0600 | c_ISVTX,
-				Mode:     0600 | c_ISREG | c_ISVTX,
+			Size:     7,
-				Size:     7,
+			ModTime:  time.Unix(1360602540, 0),
-				ModTime:  time.Unix(1360602540, 0),
+			Typeflag: TypeReg,
-				Typeflag: TypeReg,
-			},
-			fm: 0600 | os.ModeSticky,
 		},
+		fm: 0600 | os.ModeSticky,
+	}, {
 		// hard link.
-		{
+		h: &Header{
-			h: &Header{
+			Name:     "hard.txt",
-				Name:     "hard.txt",
+			Mode:     0644,
-				Mode:     0644 | c_ISREG,
+			Size:     0,
-				Size:     0,
+			Linkname: "file.txt",
-				Linkname: "file.txt",
+			ModTime:  time.Unix(1360600916, 0),
-				ModTime:  time.Unix(1360600916, 0),
+			Typeflag: TypeLink,
-				Typeflag: TypeLink,
-			},
-			fm: 0644,
 		},
+		fm: 0644,
+	}, {
 		// More information.
-		{
+		h: &Header{
-			h: &Header{
+			Name:     "info.txt",
-				Name:     "info.txt",
+			Mode:     0600,
-				Mode:     0600 | c_ISREG,
+			Size:     0,
-				Size:     0,
+			Uid:      1000,
-				Uid:      1000,
+			Gid:      1000,
-				Gid:      1000,
+			ModTime:  time.Unix(1360602540, 0),
-				ModTime:  time.Unix(1360602540, 0),
+			Uname:    "slartibartfast",
-				Uname:    "slartibartfast",
+			Gname:    "users",
-				Gname:    "users",
+			Typeflag: TypeReg,
-				Typeflag: TypeReg,
-			},
-			fm: 0600,
 		},
-	}
+		fm: 0600,
+	}}
 
-	for i, g := range golden {
+	for i, v := range vectors {
-		fi := g.h.FileInfo()
+		fi := v.h.FileInfo()
 		h2, err := FileInfoHeader(fi, "")
 		if err != nil {
 			t.Error(err)
 			continue
 		}
 		if strings.Contains(fi.Name(), "/") {
-			t.Errorf("FileInfo of %q contains slash: %q", g.h.Name, fi.Name())
+			t.Errorf("FileInfo of %q contains slash: %q", v.h.Name, fi.Name())
 		}
-		name := path.Base(g.h.Name)
+		name := path.Base(v.h.Name)
 		if fi.IsDir() {
 			name += "/"
 		}
 		if got, want := h2.Name, name; got != want {
 			t.Errorf("i=%d: Name: got %v, want %v", i, got, want)
 		}
-		if got, want := h2.Size, g.h.Size; got != want {
+		if got, want := h2.Size, v.h.Size; got != want {
 			t.Errorf("i=%d: Size: got %v, want %v", i, got, want)
 		}
-		if got, want := h2.Uid, g.h.Uid; got != want {
+		if got, want := h2.Uid, v.h.Uid; got != want {
 			t.Errorf("i=%d: Uid: got %d, want %d", i, got, want)
 		}
-		if got, want := h2.Gid, g.h.Gid; got != want {
+		if got, want := h2.Gid, v.h.Gid; got != want {
 			t.Errorf("i=%d: Gid: got %d, want %d", i, got, want)
 		}
-		if got, want := h2.Uname, g.h.Uname; got != want {
+		if got, want := h2.Uname, v.h.Uname; got != want {
 			t.Errorf("i=%d: Uname: got %q, want %q", i, got, want)
 		}
-		if got, want := h2.Gname, g.h.Gname; got != want {
+		if got, want := h2.Gname, v.h.Gname; got != want {
 			t.Errorf("i=%d: Gname: got %q, want %q", i, got, want)
 		}
-		if got, want := h2.Linkname, g.h.Linkname; got != want {
+		if got, want := h2.Linkname, v.h.Linkname; got != want {
 			t.Errorf("i=%d: Linkname: got %v, want %v", i, got, want)
 		}
-		if got, want := h2.Typeflag, g.h.Typeflag; got != want {
+		if got, want := h2.Typeflag, v.h.Typeflag; got != want {
-			t.Logf("%#v %#v", g.h, fi.Sys())
+			t.Logf("%#v %#v", v.h, fi.Sys())
 			t.Errorf("i=%d: Typeflag: got %q, want %q", i, got, want)
 		}
-		if got, want := h2.Mode, g.h.Mode; got != want {
+		if got, want := h2.Mode, v.h.Mode; got != want {
 			t.Errorf("i=%d: Mode: got %o, want %o", i, got, want)
 		}
-		if got, want := fi.Mode(), g.fm; got != want {
+		if got, want := fi.Mode(), v.fm; got != want {
 			t.Errorf("i=%d: fi.Mode: got %o, want %o", i, got, want)
 		}
-		if got, want := h2.AccessTime, g.h.AccessTime; got != want {
+		if got, want := h2.AccessTime, v.h.AccessTime; got != want {
 			t.Errorf("i=%d: AccessTime: got %v, want %v", i, got, want)
 		}
-		if got, want := h2.ChangeTime, g.h.ChangeTime; got != want {
+		if got, want := h2.ChangeTime, v.h.ChangeTime; got != want {
 			t.Errorf("i=%d: ChangeTime: got %v, want %v", i, got, want)
 		}
-		if got, want := h2.ModTime, g.h.ModTime; got != want {
+		if got, want := h2.ModTime, v.h.ModTime; got != want {
 			t.Errorf("i=%d: ModTime: got %v, want %v", i, got, want)
 		}
-		if sysh, ok := fi.Sys().(*Header); !ok || sysh != g.h {
+		if sysh, ok := fi.Sys().(*Header); !ok || sysh != v.h {
 			t.Errorf("i=%d: Sys didn't return original *Header", i)
 		}
 	}
 }
+
+func TestHeaderAllowedFormats(t *testing.T) {
+	vectors := []struct {
+		header  *Header           // Input header
+		paxHdrs map[string]string // Expected PAX headers that may be needed
+		formats Format            // Expected formats that can encode the header
+	}{{
+		header:  &Header{},
+		formats: FormatUSTAR | FormatPAX | FormatGNU,
+	}, {
+		header:  &Header{Size: 077777777777},
+		formats: FormatUSTAR | FormatPAX | FormatGNU,
+	}, {
+		header:  &Header{Size: 077777777777, Format: FormatUSTAR},
+		formats: FormatUSTAR,
+	}, {
+		header:  &Header{Size: 077777777777, Format: FormatPAX},
+		formats: FormatUSTAR | FormatPAX,
+	}, {
+		header:  &Header{Size: 077777777777, Format: FormatGNU},
+		formats: FormatGNU,
+	}, {
+		header:  &Header{Size: 077777777777 + 1},
+		paxHdrs: map[string]string{paxSize: "8589934592"},
+		formats: FormatPAX | FormatGNU,
+	}, {
+		header:  &Header{Size: 077777777777 + 1, Format: FormatPAX},
+		paxHdrs: map[string]string{paxSize: "8589934592"},
+		formats: FormatPAX,
+	}, {
+		header:  &Header{Size: 077777777777 + 1, Format: FormatGNU},
+		paxHdrs: map[string]string{paxSize: "8589934592"},
+		formats: FormatGNU,
+	}, {
+		header:  &Header{Mode: 07777777},
+		formats: FormatUSTAR | FormatPAX | FormatGNU,
+	}, {
+		header:  &Header{Mode: 07777777 + 1},
+		formats: FormatGNU,
+	}, {
+		header:  &Header{Devmajor: -123},
+		formats: FormatGNU,
+	}, {
+		header:  &Header{Devmajor: 1<<56 - 1},
+		formats: FormatGNU,
+	}, {
+		header:  &Header{Devmajor: 1 << 56},
+		formats: FormatUnknown,
+	}, {
+		header:  &Header{Devmajor: -1 << 56},
+		formats: FormatGNU,
+	}, {
+		header:  &Header{Devmajor: -1<<56 - 1},
+		formats: FormatUnknown,
+	}, {
+		header:  &Header{Name: "用戶名", Devmajor: -1 << 56},
+		formats: FormatGNU,
+	}, {
+		header:  &Header{Size: math.MaxInt64},
+		paxHdrs: map[string]string{paxSize: "9223372036854775807"},
+		formats: FormatPAX | FormatGNU,
+	}, {
+		header:  &Header{Size: math.MinInt64},
+		paxHdrs: map[string]string{paxSize: "-9223372036854775808"},
+		formats: FormatUnknown,
+	}, {
+		header:  &Header{Uname: "0123456789abcdef0123456789abcdef"},
+		formats: FormatUSTAR | FormatPAX | FormatGNU,
+	}, {
+		header:  &Header{Uname: "0123456789abcdef0123456789abcdefx"},
+		paxHdrs: map[string]string{paxUname: "0123456789abcdef0123456789abcdefx"},
+		formats: FormatPAX,
+	}, {
+		header:  &Header{Name: "foobar"},
+		formats: FormatUSTAR | FormatPAX | FormatGNU,
+	}, {
+		header:  &Header{Name: strings.Repeat("a", nameSize)},
+		formats: FormatUSTAR | FormatPAX | FormatGNU,
+	}, {
+		header:  &Header{Name: strings.Repeat("a", nameSize+1)},
+		paxHdrs: map[string]string{paxPath: strings.Repeat("a", nameSize+1)},
+		formats: FormatPAX | FormatGNU,
+	}, {
+		header:  &Header{Linkname: "用戶名"},
+		paxHdrs: map[string]string{paxLinkpath: "用戶名"},
+		formats: FormatPAX | FormatGNU,
+	}, {
+		header:  &Header{Linkname: strings.Repeat("用戶名\x00", nameSize)},
+		paxHdrs: map[string]string{paxLinkpath: strings.Repeat("用戶名\x00", nameSize)},
+		formats: FormatUnknown,
+	}, {
+		header:  &Header{Linkname: "\x00hello"},
+		paxHdrs: map[string]string{paxLinkpath: "\x00hello"},
+		formats: FormatUnknown,
+	}, {
+		header:  &Header{Uid: 07777777},
+		formats: FormatUSTAR | FormatPAX | FormatGNU,
+	}, {
+		header:  &Header{Uid: 07777777 + 1},
+		paxHdrs: map[string]string{paxUid: "2097152"},
+		formats: FormatPAX | FormatGNU,
+	}, {
+		header:  &Header{Xattrs: nil},
+		formats: FormatUSTAR | FormatPAX | FormatGNU,
+	}, {
+		header:  &Header{Xattrs: map[string]string{"foo": "bar"}},
+		paxHdrs: map[string]string{paxSchilyXattr + "foo": "bar"},
+		formats: FormatPAX,
+	}, {
+		header:  &Header{Xattrs: map[string]string{"foo": "bar"}, Format: FormatGNU},
+		paxHdrs: map[string]string{paxSchilyXattr + "foo": "bar"},
+		formats: FormatUnknown,
+	}, {
+		header:  &Header{Xattrs: map[string]string{"用戶名": "\x00hello"}},
+		paxHdrs: map[string]string{paxSchilyXattr + "用戶名": "\x00hello"},
+		formats: FormatPAX,
+	}, {
+		header:  &Header{Xattrs: map[string]string{"foo=bar": "baz"}},
+		formats: FormatUnknown,
+	}, {
+		header:  &Header{Xattrs: map[string]string{"foo": ""}},
+		paxHdrs: map[string]string{paxSchilyXattr + "foo": ""},
+		formats: FormatPAX,
+	}, {
+		header:  &Header{ModTime: time.Unix(0, 0)},
+		formats: FormatUSTAR | FormatPAX | FormatGNU,
+	}, {
+		header:  &Header{ModTime: time.Unix(077777777777, 0)},
+		formats: FormatUSTAR | FormatPAX | FormatGNU,
+	}, {
+		header:  &Header{ModTime: time.Unix(077777777777+1, 0)},
+		paxHdrs: map[string]string{paxMtime: "8589934592"},
+		formats: FormatPAX | FormatGNU,
+	}, {
+		header:  &Header{ModTime: time.Unix(math.MaxInt64, 0)},
+		paxHdrs: map[string]string{paxMtime: "9223372036854775807"},
+		formats: FormatPAX | FormatGNU,
+	}, {
+		header:  &Header{ModTime: time.Unix(math.MaxInt64, 0), Format: FormatUSTAR},
+		paxHdrs: map[string]string{paxMtime: "9223372036854775807"},
+		formats: FormatUnknown,
+	}, {
+		header:  &Header{ModTime: time.Unix(-1, 0)},
+		paxHdrs: map[string]string{paxMtime: "-1"},
+		formats: FormatPAX | FormatGNU,
+	}, {
+		header:  &Header{ModTime: time.Unix(1, 500)},
+		paxHdrs: map[string]string{paxMtime: "1.0000005"},
+		formats: FormatUSTAR | FormatPAX | FormatGNU,
+	}, {
+		header:  &Header{ModTime: time.Unix(1, 0)},
+		formats: FormatUSTAR | FormatPAX | FormatGNU,
+	}, {
+		header:  &Header{ModTime: time.Unix(1, 0), Format: FormatPAX},
+		formats: FormatUSTAR | FormatPAX,
+	}, {
+		header:  &Header{ModTime: time.Unix(1, 500), Format: FormatUSTAR},
+		paxHdrs: map[string]string{paxMtime: "1.0000005"},
+		formats: FormatUSTAR,
+	}, {
+		header:  &Header{ModTime: time.Unix(1, 500), Format: FormatPAX},
+		paxHdrs: map[string]string{paxMtime: "1.0000005"},
+		formats: FormatPAX,
+	}, {
+		header:  &Header{ModTime: time.Unix(1, 500), Format: FormatGNU},
+		paxHdrs: map[string]string{paxMtime: "1.0000005"},
+		formats: FormatGNU,
+	}, {
+		header:  &Header{ModTime: time.Unix(-1, 500)},
+		paxHdrs: map[string]string{paxMtime: "-0.9999995"},
+		formats: FormatPAX | FormatGNU,
+	}, {
+		header:  &Header{ModTime: time.Unix(-1, 500), Format: FormatGNU},
+		paxHdrs: map[string]string{paxMtime: "-0.9999995"},
+		formats: FormatGNU,
+	}, {
+		header:  &Header{AccessTime: time.Unix(0, 0)},
+		paxHdrs: map[string]string{paxAtime: "0"},
+		formats: FormatPAX | FormatGNU,
+	}, {
+		header:  &Header{AccessTime: time.Unix(0, 0), Format: FormatUSTAR},
+		paxHdrs: map[string]string{paxAtime: "0"},
+		formats: FormatUnknown,
+	}, {
+		header:  &Header{AccessTime: time.Unix(0, 0), Format: FormatPAX},
+		paxHdrs: map[string]string{paxAtime: "0"},
+		formats: FormatPAX,
+	}, {
+		header:  &Header{AccessTime: time.Unix(0, 0), Format: FormatGNU},
+		paxHdrs: map[string]string{paxAtime: "0"},
+		formats: FormatGNU,
+	}, {
+		header:  &Header{AccessTime: time.Unix(-123, 0)},
+		paxHdrs: map[string]string{paxAtime: "-123"},
+		formats: FormatPAX | FormatGNU,
+	}, {
+		header:  &Header{AccessTime: time.Unix(-123, 0), Format: FormatPAX},
+		paxHdrs: map[string]string{paxAtime: "-123"},
+		formats: FormatPAX,
+	}, {
+		header:  &Header{ChangeTime: time.Unix(123, 456)},
+		paxHdrs: map[string]string{paxCtime: "123.000000456"},
+		formats: FormatPAX | FormatGNU,
+	}, {
+		header:  &Header{ChangeTime: time.Unix(123, 456), Format: FormatUSTAR},
+		paxHdrs: map[string]string{paxCtime: "123.000000456"},
+		formats: FormatUnknown,
+	}, {
+		header:  &Header{ChangeTime: time.Unix(123, 456), Format: FormatGNU},
+		paxHdrs: map[string]string{paxCtime: "123.000000456"},
+		formats: FormatGNU,
+	}, {
+		header:  &Header{ChangeTime: time.Unix(123, 456), Format: FormatPAX},
+		paxHdrs: map[string]string{paxCtime: "123.000000456"},
+		formats: FormatPAX,
+	}, {
+		header:  &Header{Name: "foo/", Typeflag: TypeDir},
+		formats: FormatUSTAR | FormatPAX | FormatGNU,
+	}, {
+		header:  &Header{Name: "foo/", Typeflag: TypeReg},
+		formats: FormatUnknown,
+	}, {
+		header:  &Header{Name: "foo/", Typeflag: TypeSymlink},
+		formats: FormatUSTAR | FormatPAX | FormatGNU,
+	}}
+
+	for i, v := range vectors {
+		formats, paxHdrs, err := v.header.allowedFormats()
+		if formats != v.formats {
+			t.Errorf("test %d, allowedFormats(): got %v, want %v", i, formats, v.formats)
+		}
+		if formats&FormatPAX > 0 && !reflect.DeepEqual(paxHdrs, v.paxHdrs) && !(len(paxHdrs) == 0 && len(v.paxHdrs) == 0) {
+			t.Errorf("test %d, allowedFormats():\ngot  %v\nwant %s", i, paxHdrs, v.paxHdrs)
+		}
+		if (formats != FormatUnknown) && (err != nil) {
+			t.Errorf("test %d, unexpected error: %v", i, err)
+		}
+		if (formats == FormatUnknown) && (err == nil) {
+			t.Errorf("test %d, got nil-error, want non-nil error", i)
+		}
+	}
+}
+
+func Benchmark(b *testing.B) {
+	type file struct {
+		hdr  *Header
+		body []byte
+	}
+
+	vectors := []struct {
+		label string
+		files []file
+	}{{
+		"USTAR",
+		[]file{{
+			&Header{Name: "bar", Mode: 0640, Size: int64(3)},
+			[]byte("foo"),
+		}, {
+			&Header{Name: "world", Mode: 0640, Size: int64(5)},
+			[]byte("hello"),
+		}},
+	}, {
+		"GNU",
+		[]file{{
+			&Header{Name: "bar", Mode: 0640, Size: int64(3), Devmajor: -1},
+			[]byte("foo"),
+		}, {
+			&Header{Name: "world", Mode: 0640, Size: int64(5), Devmajor: -1},
+			[]byte("hello"),
+		}},
+	}, {
+		"PAX",
+		[]file{{
+			&Header{Name: "bar", Mode: 0640, Size: int64(3), Xattrs: map[string]string{"foo": "bar"}},
+			[]byte("foo"),
+		}, {
+			&Header{Name: "world", Mode: 0640, Size: int64(5), Xattrs: map[string]string{"foo": "bar"}},
+			[]byte("hello"),
+		}},
+	}}
+
+	b.Run("Writer", func(b *testing.B) {
+		for _, v := range vectors {
+			b.Run(v.label, func(b *testing.B) {
+				b.ReportAllocs()
+				for i := 0; i < b.N; i++ {
+					// Writing to ioutil.Discard because we want to
+					// test purely the writer code and not bring in disk performance into this.
+					tw := NewWriter(ioutil.Discard)
+					for _, file := range v.files {
+						if err := tw.WriteHeader(file.hdr); err != nil {
+							b.Errorf("unexpected WriteHeader error: %v", err)
+						}
+						if _, err := tw.Write(file.body); err != nil {
+							b.Errorf("unexpected Write error: %v", err)
+						}
+					}
+					if err := tw.Close(); err != nil {
+						b.Errorf("unexpected Close error: %v", err)
+					}
+				}
+			})
+		}
+	})
+
+	b.Run("Reader", func(b *testing.B) {
+		for _, v := range vectors {
+			var buf bytes.Buffer
+			var r bytes.Reader
+
+			// Write the archive to a byte buffer.
+			tw := NewWriter(&buf)
+			for _, file := range v.files {
+				tw.WriteHeader(file.hdr)
+				tw.Write(file.body)
+			}
+			tw.Close()
+			b.Run(v.label, func(b *testing.B) {
+				b.ReportAllocs()
+				// Read from the byte buffer.
+				for i := 0; i < b.N; i++ {
+					r.Reset(buf.Bytes())
+					tr := NewReader(&r)
+					if _, err := tr.Next(); err != nil {
+						b.Errorf("unexpected Next error: %v", err)
+					}
+					if _, err := io.Copy(ioutil.Discard, tr); err != nil {
+						b.Errorf("unexpected Copy error : %v", err)
+					}
+				}
+			})
+		}
+	})
+
+}

archive/tar/writer.go

@@ -4,295 +4,410 @@
 
 package tar
 
-// TODO(dsymonds):
-// - catch more errors (no first header, etc.)
-
 import (
-	"bytes"
-	"errors"
 	"fmt"
 	"io"
 	"path"
 	"sort"
-	"strconv"
 	"strings"
 	"time"
 )
 
-var (
+// Writer provides sequential writing of a tar archive.
-	ErrWriteTooLong    = errors.New("archive/tar: write too long")
+// Write.WriteHeader begins a new file with the provided Header,
-	ErrFieldTooLong    = errors.New("archive/tar: header field too long")
+// and then Writer can be treated as an io.Writer to supply that file's data.
-	ErrWriteAfterClose = errors.New("archive/tar: write after close")
-	errInvalidHeader   = errors.New("archive/tar: header field too long or contains invalid values")
-)
-
-// A Writer provides sequential writing of a tar archive in POSIX.1 format.
-// A tar archive consists of a sequence of files.
-// Call WriteHeader to begin a new file, and then call Write to supply that file's data,
-// writing at most hdr.Size bytes in total.
 type Writer struct {
-	w          io.Writer
+	w    io.Writer
-	err        error
+	pad  int64      // Amount of padding to write after current file entry
-	nb         int64 // number of unwritten bytes for current file entry
+	curr fileWriter // Writer for current file entry
-	pad        int64 // amount of padding to write after current file entry
+	hdr  Header     // Shallow copy of Header that is safe for mutations
-	closed     bool
+	blk  block      // Buffer to use as temporary local storage
-	usedBinary bool            // whether the binary numeric field extension was used
-	preferPax  bool            // use pax header instead of binary numeric header
-	hdrBuff    [blockSize]byte // buffer to use in writeHeader when writing a regular header
-	paxHdrBuff [blockSize]byte // buffer to use in writeHeader when writing a pax header
-}
 
-type formatter struct {
+	// err is a persistent error.
-	err error // Last error seen
+	// It is only the responsibility of every exported method of Writer to
+	// ensure that this error is sticky.
+	err error
 }
 
 // NewWriter creates a new Writer writing to w.
-func NewWriter(w io.Writer) *Writer { return &Writer{w: w} }
+func NewWriter(w io.Writer) *Writer {
-
+	return &Writer{w: w, curr: &regFileWriter{w, 0}}
-// Flush finishes writing the current file (optional).
-func (tw *Writer) Flush() error {
-	if tw.nb > 0 {
-		tw.err = fmt.Errorf("archive/tar: missed writing %d bytes", tw.nb)
-		return tw.err
-	}
-
-	n := tw.nb + tw.pad
-	for n > 0 && tw.err == nil {
-		nr := n
-		if nr > blockSize {
-			nr = blockSize
-		}
-		var nw int
-		nw, tw.err = tw.w.Write(zeroBlock[0:nr])
-		n -= int64(nw)
-	}
-	tw.nb = 0
-	tw.pad = 0
-	return tw.err
 }
 
-// Write s into b, terminating it with a NUL if there is room.
+type fileWriter interface {
-func (f *formatter) formatString(b []byte, s string) {
+	io.Writer
-	if len(s) > len(b) {
+	fileState
-		f.err = ErrFieldTooLong
+
-		return
+	ReadFrom(io.Reader) (int64, error)
-	}
-	ascii := toASCII(s)
-	copy(b, ascii)
-	if len(ascii) < len(b) {
-		b[len(ascii)] = 0
-	}
 }
 
-// Encode x as an octal ASCII string and write it into b with leading zeros.
+// Flush finishes writing the current file's block padding.
-func (f *formatter) formatOctal(b []byte, x int64) {
+// The current file must be fully written before Flush can be called.
-	s := strconv.FormatInt(x, 8)
-	// leading zeros, but leave room for a NUL.
-	for len(s)+1 < len(b) {
-		s = "0" + s
-	}
-	f.formatString(b, s)
-}
-
-// fitsInBase256 reports whether x can be encoded into n bytes using base-256
-// encoding. Unlike octal encoding, base-256 encoding does not require that the
-// string ends with a NUL character. Thus, all n bytes are available for output.
 //
-// If operating in binary mode, this assumes strict GNU binary mode; which means
+// This is unnecessary as the next call to WriteHeader or Close
-// that the first byte can only be either 0x80 or 0xff. Thus, the first byte is
+// will implicitly flush out the file's padding.
-// equivalent to the sign bit in two's complement form.
+func (tw *Writer) Flush() error {
-func fitsInBase256(n int, x int64) bool {
-	var binBits = uint(n-1) * 8
-	return n >= 9 || (x >= -1<<binBits && x < 1<<binBits)
-}
-
-// Write x into b, as binary (GNUtar/star extension).
-func (f *formatter) formatNumeric(b []byte, x int64) {
-	if fitsInBase256(len(b), x) {
-		for i := len(b) - 1; i >= 0; i-- {
-			b[i] = byte(x)
-			x >>= 8
-		}
-		b[0] |= 0x80 // Highest bit indicates binary format
-		return
-	}
-
-	f.formatOctal(b, 0) // Last resort, just write zero
-	f.err = ErrFieldTooLong
-}
-
-var (
-	minTime = time.Unix(0, 0)
-	// There is room for 11 octal digits (33 bits) of mtime.
-	maxTime = minTime.Add((1<<33 - 1) * time.Second)
-)
-
-// WriteHeader writes hdr and prepares to accept the file's contents.
-// WriteHeader calls Flush if it is not the first header.
-// Calling after a Close will return ErrWriteAfterClose.
-func (tw *Writer) WriteHeader(hdr *Header) error {
-	return tw.writeHeader(hdr, true)
-}
-
-// WriteHeader writes hdr and prepares to accept the file's contents.
-// WriteHeader calls Flush if it is not the first header.
-// Calling after a Close will return ErrWriteAfterClose.
-// As this method is called internally by writePax header to allow it to
-// suppress writing the pax header.
-func (tw *Writer) writeHeader(hdr *Header, allowPax bool) error {
-	if tw.closed {
-		return ErrWriteAfterClose
-	}
-	if tw.err == nil {
-		tw.Flush()
-	}
 	if tw.err != nil {
 		return tw.err
 	}
-
+	if nb := tw.curr.LogicalRemaining(); nb > 0 {
-	// a map to hold pax header records, if any are needed
+		return fmt.Errorf("archive/tar: missed writing %d bytes", nb)
-	paxHeaders := make(map[string]string)
-
-	// TODO(shanemhansen): we might want to use PAX headers for
-	// subsecond time resolution, but for now let's just capture
-	// too long fields or non ascii characters
-
-	var f formatter
-	var header []byte
-
-	// We need to select which scratch buffer to use carefully,
-	// since this method is called recursively to write PAX headers.
-	// If allowPax is true, this is the non-recursive call, and we will use hdrBuff.
-	// If allowPax is false, we are being called by writePAXHeader, and hdrBuff is
-	// already being used by the non-recursive call, so we must use paxHdrBuff.
-	header = tw.hdrBuff[:]
-	if !allowPax {
-		header = tw.paxHdrBuff[:]
 	}
-	copy(header, zeroBlock)
+	if _, tw.err = tw.w.Write(zeroBlock[:tw.pad]); tw.err != nil {
-	s := slicer(header)
-
-	// Wrappers around formatter that automatically sets paxHeaders if the
-	// argument extends beyond the capacity of the input byte slice.
-	var formatString = func(b []byte, s string, paxKeyword string) {
-		needsPaxHeader := paxKeyword != paxNone && len(s) > len(b) || !isASCII(s)
-		if needsPaxHeader {
-			paxHeaders[paxKeyword] = s
-			return
-		}
-		f.formatString(b, s)
-	}
-	var formatNumeric = func(b []byte, x int64, paxKeyword string) {
-		// Try octal first.
-		s := strconv.FormatInt(x, 8)
-		if len(s) < len(b) {
-			f.formatOctal(b, x)
-			return
-		}
-
-		// If it is too long for octal, and PAX is preferred, use a PAX header.
-		if paxKeyword != paxNone && tw.preferPax {
-			f.formatOctal(b, 0)
-			s := strconv.FormatInt(x, 10)
-			paxHeaders[paxKeyword] = s
-			return
-		}
-
-		tw.usedBinary = true
-		f.formatNumeric(b, x)
-	}
-
-	// keep a reference to the filename to allow to overwrite it later if we detect that we can use ustar longnames instead of pax
-	pathHeaderBytes := s.next(fileNameSize)
-
-	formatString(pathHeaderBytes, hdr.Name, paxPath)
-
-	// Handle out of range ModTime carefully.
-	var modTime int64
-	if !hdr.ModTime.Before(minTime) && !hdr.ModTime.After(maxTime) {
-		modTime = hdr.ModTime.Unix()
-	}
-
-	f.formatOctal(s.next(8), hdr.Mode)               // 100:108
-	formatNumeric(s.next(8), int64(hdr.Uid), paxUid) // 108:116
-	formatNumeric(s.next(8), int64(hdr.Gid), paxGid) // 116:124
-	formatNumeric(s.next(12), hdr.Size, paxSize)     // 124:136
-	formatNumeric(s.next(12), modTime, paxNone)      // 136:148 --- consider using pax for finer granularity
-	s.next(8)                                        // chksum (148:156)
-	s.next(1)[0] = hdr.Typeflag                      // 156:157
-
-	formatString(s.next(100), hdr.Linkname, paxLinkpath)
-
-	copy(s.next(8), []byte("ustar\x0000"))          // 257:265
-	formatString(s.next(32), hdr.Uname, paxUname)   // 265:297
-	formatString(s.next(32), hdr.Gname, paxGname)   // 297:329
-	formatNumeric(s.next(8), hdr.Devmajor, paxNone) // 329:337
-	formatNumeric(s.next(8), hdr.Devminor, paxNone) // 337:345
-
-	// keep a reference to the prefix to allow to overwrite it later if we detect that we can use ustar longnames instead of pax
-	prefixHeaderBytes := s.next(155)
-	formatString(prefixHeaderBytes, "", paxNone) // 345:500  prefix
-
-	// Use the GNU magic instead of POSIX magic if we used any GNU extensions.
-	if tw.usedBinary {
-		copy(header[257:265], []byte("ustar  \x00"))
-	}
-
-	_, paxPathUsed := paxHeaders[paxPath]
-	// try to use a ustar header when only the name is too long
-	if !tw.preferPax && len(paxHeaders) == 1 && paxPathUsed {
-		prefix, suffix, ok := splitUSTARPath(hdr.Name)
-		if ok {
-			// Since we can encode in USTAR format, disable PAX header.
-			delete(paxHeaders, paxPath)
-
-			// Update the path fields
-			formatString(pathHeaderBytes, suffix, paxNone)
-			formatString(prefixHeaderBytes, prefix, paxNone)
-		}
-	}
-
-	// The chksum field is terminated by a NUL and a space.
-	// This is different from the other octal fields.
-	chksum, _ := checksum(header)
-	f.formatOctal(header[148:155], chksum) // Never fails
-	header[155] = ' '
-
-	// Check if there were any formatting errors.
-	if f.err != nil {
-		tw.err = f.err
 		return tw.err
 	}
+	tw.pad = 0
+	return nil
+}
 
-	if allowPax {
+// WriteHeader writes hdr and prepares to accept the file's contents.
-		for k, v := range hdr.Xattrs {
+// The Header.Size determines how many bytes can be written for the next file.
-			paxHeaders[paxXattr+k] = v
+// If the current file is not fully written, then this returns an error.
+// This implicitly flushes any padding necessary before writing the header.
+func (tw *Writer) WriteHeader(hdr *Header) error {
+	if err := tw.Flush(); err != nil {
+		return err
+	}
+	tw.hdr = *hdr // Shallow copy of Header
+
+	// Avoid usage of the legacy TypeRegA flag, and automatically promote
+	// it to use TypeReg or TypeDir.
+	if tw.hdr.Typeflag == TypeRegA {
+		if strings.HasSuffix(tw.hdr.Name, "/") {
+			tw.hdr.Typeflag = TypeDir
+		} else {
+			tw.hdr.Typeflag = TypeReg
 		}
 	}
 
-	if len(paxHeaders) > 0 {
+	// Round ModTime and ignore AccessTime and ChangeTime unless
-		if !allowPax {
+	// the format is explicitly chosen.
-			return errInvalidHeader
+	// This ensures nominal usage of WriteHeader (without specifying the format)
+	// does not always result in the PAX format being chosen, which
+	// causes a 1KiB increase to every header.
+	if tw.hdr.Format == FormatUnknown {
+		tw.hdr.ModTime = tw.hdr.ModTime.Round(time.Second)
+		tw.hdr.AccessTime = time.Time{}
+		tw.hdr.ChangeTime = time.Time{}
+	}
+
+	allowedFormats, paxHdrs, err := tw.hdr.allowedFormats()
+	switch {
+	case allowedFormats.has(FormatUSTAR):
+		tw.err = tw.writeUSTARHeader(&tw.hdr)
+		return tw.err
+	case allowedFormats.has(FormatPAX):
+		tw.err = tw.writePAXHeader(&tw.hdr, paxHdrs)
+		return tw.err
+	case allowedFormats.has(FormatGNU):
+		tw.err = tw.writeGNUHeader(&tw.hdr)
+		return tw.err
+	default:
+		return err // Non-fatal error
+	}
+}
+
+func (tw *Writer) writeUSTARHeader(hdr *Header) error {
+	// Check if we can use USTAR prefix/suffix splitting.
+	var namePrefix string
+	if prefix, suffix, ok := splitUSTARPath(hdr.Name); ok {
+		namePrefix, hdr.Name = prefix, suffix
+	}
+
+	// Pack the main header.
+	var f formatter
+	blk := tw.templateV7Plus(hdr, f.formatString, f.formatOctal)
+	f.formatString(blk.USTAR().Prefix(), namePrefix)
+	blk.SetFormat(FormatUSTAR)
+	if f.err != nil {
+		return f.err // Should never happen since header is validated
+	}
+	return tw.writeRawHeader(blk, hdr.Size, hdr.Typeflag)
+}
+
+func (tw *Writer) writePAXHeader(hdr *Header, paxHdrs map[string]string) error {
+	realName, realSize := hdr.Name, hdr.Size
+
+	// TODO(dsnet): Re-enable this when adding sparse support.
+	// See https://golang.org/issue/22735
+	/*
+		// Handle sparse files.
+		var spd sparseDatas
+		var spb []byte
+		if len(hdr.SparseHoles) > 0 {
+			sph := append([]sparseEntry{}, hdr.SparseHoles...) // Copy sparse map
+			sph = alignSparseEntries(sph, hdr.Size)
+			spd = invertSparseEntries(sph, hdr.Size)
+
+			// Format the sparse map.
+			hdr.Size = 0 // Replace with encoded size
+			spb = append(strconv.AppendInt(spb, int64(len(spd)), 10), '\n')
+			for _, s := range spd {
+				hdr.Size += s.Length
+				spb = append(strconv.AppendInt(spb, s.Offset, 10), '\n')
+				spb = append(strconv.AppendInt(spb, s.Length, 10), '\n')
+			}
+			pad := blockPadding(int64(len(spb)))
+			spb = append(spb, zeroBlock[:pad]...)
+			hdr.Size += int64(len(spb)) // Accounts for encoded sparse map
+
+			// Add and modify appropriate PAX records.
+			dir, file := path.Split(realName)
+			hdr.Name = path.Join(dir, "GNUSparseFile.0", file)
+			paxHdrs[paxGNUSparseMajor] = "1"
+			paxHdrs[paxGNUSparseMinor] = "0"
+			paxHdrs[paxGNUSparseName] = realName
+			paxHdrs[paxGNUSparseRealSize] = strconv.FormatInt(realSize, 10)
+			paxHdrs[paxSize] = strconv.FormatInt(hdr.Size, 10)
+			delete(paxHdrs, paxPath) // Recorded by paxGNUSparseName
 		}
-		if err := tw.writePAXHeader(hdr, paxHeaders); err != nil {
+	*/
+	_ = realSize
+
+	// Write PAX records to the output.
+	isGlobal := hdr.Typeflag == TypeXGlobalHeader
+	if len(paxHdrs) > 0 || isGlobal {
+		// Sort keys for deterministic ordering.
+		var keys []string
+		for k := range paxHdrs {
+			keys = append(keys, k)
+		}
+		sort.Strings(keys)
+
+		// Write each record to a buffer.
+		var buf strings.Builder
+		for _, k := range keys {
+			rec, err := formatPAXRecord(k, paxHdrs[k])
+			if err != nil {
+				return err
+			}
+			buf.WriteString(rec)
+		}
+
+		// Write the extended header file.
+		var name string
+		var flag byte
+		if isGlobal {
+			name = realName
+			if name == "" {
+				name = "GlobalHead.0.0"
+			}
+			flag = TypeXGlobalHeader
+		} else {
+			dir, file := path.Split(realName)
+			name = path.Join(dir, "PaxHeaders.0", file)
+			flag = TypeXHeader
+		}
+		data := buf.String()
+		if err := tw.writeRawFile(name, data, flag, FormatPAX); err != nil || isGlobal {
+			return err // Global headers return here
+		}
+	}
+
+	// Pack the main header.
+	var f formatter // Ignore errors since they are expected
+	fmtStr := func(b []byte, s string) { f.formatString(b, toASCII(s)) }
+	blk := tw.templateV7Plus(hdr, fmtStr, f.formatOctal)
+	blk.SetFormat(FormatPAX)
+	if err := tw.writeRawHeader(blk, hdr.Size, hdr.Typeflag); err != nil {
+		return err
+	}
+
+	// TODO(dsnet): Re-enable this when adding sparse support.
+	// See https://golang.org/issue/22735
+	/*
+		// Write the sparse map and setup the sparse writer if necessary.
+		if len(spd) > 0 {
+			// Use tw.curr since the sparse map is accounted for in hdr.Size.
+			if _, err := tw.curr.Write(spb); err != nil {
+				return err
+			}
+			tw.curr = &sparseFileWriter{tw.curr, spd, 0}
+		}
+	*/
+	return nil
+}
+
+func (tw *Writer) writeGNUHeader(hdr *Header) error {
+	// Use long-link files if Name or Linkname exceeds the field size.
+	const longName = "././@LongLink"
+	if len(hdr.Name) > nameSize {
+		data := hdr.Name + "\x00"
+		if err := tw.writeRawFile(longName, data, TypeGNULongName, FormatGNU); err != nil {
+			return err
+		}
+	}
+	if len(hdr.Linkname) > nameSize {
+		data := hdr.Linkname + "\x00"
+		if err := tw.writeRawFile(longName, data, TypeGNULongLink, FormatGNU); err != nil {
 			return err
 		}
 	}
-	tw.nb = int64(hdr.Size)
-	tw.pad = (blockSize - (tw.nb % blockSize)) % blockSize
 
-	_, tw.err = tw.w.Write(header)
+	// Pack the main header.
-	return tw.err
+	var f formatter // Ignore errors since they are expected
+	var spd sparseDatas
+	var spb []byte
+	blk := tw.templateV7Plus(hdr, f.formatString, f.formatNumeric)
+	if !hdr.AccessTime.IsZero() {
+		f.formatNumeric(blk.GNU().AccessTime(), hdr.AccessTime.Unix())
+	}
+	if !hdr.ChangeTime.IsZero() {
+		f.formatNumeric(blk.GNU().ChangeTime(), hdr.ChangeTime.Unix())
+	}
+	// TODO(dsnet): Re-enable this when adding sparse support.
+	// See https://golang.org/issue/22735
+	/*
+		if hdr.Typeflag == TypeGNUSparse {
+			sph := append([]sparseEntry{}, hdr.SparseHoles...) // Copy sparse map
+			sph = alignSparseEntries(sph, hdr.Size)
+			spd = invertSparseEntries(sph, hdr.Size)
+
+			// Format the sparse map.
+			formatSPD := func(sp sparseDatas, sa sparseArray) sparseDatas {
+				for i := 0; len(sp) > 0 && i < sa.MaxEntries(); i++ {
+					f.formatNumeric(sa.Entry(i).Offset(), sp[0].Offset)
+					f.formatNumeric(sa.Entry(i).Length(), sp[0].Length)
+					sp = sp[1:]
+				}
+				if len(sp) > 0 {
+					sa.IsExtended()[0] = 1
+				}
+				return sp
+			}
+			sp2 := formatSPD(spd, blk.GNU().Sparse())
+			for len(sp2) > 0 {
+				var spHdr block
+				sp2 = formatSPD(sp2, spHdr.Sparse())
+				spb = append(spb, spHdr[:]...)
+			}
+
+			// Update size fields in the header block.
+			realSize := hdr.Size
+			hdr.Size = 0 // Encoded size; does not account for encoded sparse map
+			for _, s := range spd {
+				hdr.Size += s.Length
+			}
+			copy(blk.V7().Size(), zeroBlock[:]) // Reset field
+			f.formatNumeric(blk.V7().Size(), hdr.Size)
+			f.formatNumeric(blk.GNU().RealSize(), realSize)
+		}
+	*/
+	blk.SetFormat(FormatGNU)
+	if err := tw.writeRawHeader(blk, hdr.Size, hdr.Typeflag); err != nil {
+		return err
+	}
+
+	// Write the extended sparse map and setup the sparse writer if necessary.
+	if len(spd) > 0 {
+		// Use tw.w since the sparse map is not accounted for in hdr.Size.
+		if _, err := tw.w.Write(spb); err != nil {
+			return err
+		}
+		tw.curr = &sparseFileWriter{tw.curr, spd, 0}
+	}
+	return nil
+}
+
+type (
+	stringFormatter func([]byte, string)
+	numberFormatter func([]byte, int64)
+)
+
+// templateV7Plus fills out the V7 fields of a block using values from hdr.
+// It also fills out fields (uname, gname, devmajor, devminor) that are
+// shared in the USTAR, PAX, and GNU formats using the provided formatters.
+//
+// The block returned is only valid until the next call to
+// templateV7Plus or writeRawFile.
+func (tw *Writer) templateV7Plus(hdr *Header, fmtStr stringFormatter, fmtNum numberFormatter) *block {
+	tw.blk.Reset()
+
+	modTime := hdr.ModTime
+	if modTime.IsZero() {
+		modTime = time.Unix(0, 0)
+	}
+
+	v7 := tw.blk.V7()
+	v7.TypeFlag()[0] = hdr.Typeflag
+	fmtStr(v7.Name(), hdr.Name)
+	fmtStr(v7.LinkName(), hdr.Linkname)
+	fmtNum(v7.Mode(), hdr.Mode)
+	fmtNum(v7.UID(), int64(hdr.Uid))
+	fmtNum(v7.GID(), int64(hdr.Gid))
+	fmtNum(v7.Size(), hdr.Size)
+	fmtNum(v7.ModTime(), modTime.Unix())
+
+	ustar := tw.blk.USTAR()
+	fmtStr(ustar.UserName(), hdr.Uname)
+	fmtStr(ustar.GroupName(), hdr.Gname)
+	fmtNum(ustar.DevMajor(), hdr.Devmajor)
+	fmtNum(ustar.DevMinor(), hdr.Devminor)
+
+	return &tw.blk
+}
+
+// writeRawFile writes a minimal file with the given name and flag type.
+// It uses format to encode the header format and will write data as the body.
+// It uses default values for all of the other fields (as BSD and GNU tar does).
+func (tw *Writer) writeRawFile(name, data string, flag byte, format Format) error {
+	tw.blk.Reset()
+
+	// Best effort for the filename.
+	name = toASCII(name)
+	if len(name) > nameSize {
+		name = name[:nameSize]
+	}
+	name = strings.TrimRight(name, "/")
+
+	var f formatter
+	v7 := tw.blk.V7()
+	v7.TypeFlag()[0] = flag
+	f.formatString(v7.Name(), name)
+	f.formatOctal(v7.Mode(), 0)
+	f.formatOctal(v7.UID(), 0)
+	f.formatOctal(v7.GID(), 0)
+	f.formatOctal(v7.Size(), int64(len(data))) // Must be < 8GiB
+	f.formatOctal(v7.ModTime(), 0)
+	tw.blk.SetFormat(format)
+	if f.err != nil {
+		return f.err // Only occurs if size condition is violated
+	}
+
+	// Write the header and data.
+	if err := tw.writeRawHeader(&tw.blk, int64(len(data)), flag); err != nil {
+		return err
+	}
+	_, err := io.WriteString(tw, data)
+	return err
+}
+
+// writeRawHeader writes the value of blk, regardless of its value.
+// It sets up the Writer such that it can accept a file of the given size.
+// If the flag is a special header-only flag, then the size is treated as zero.
+func (tw *Writer) writeRawHeader(blk *block, size int64, flag byte) error {
+	if err := tw.Flush(); err != nil {
+		return err
+	}
+	if _, err := tw.w.Write(blk[:]); err != nil {
+		return err
+	}
+	if isHeaderOnlyType(flag) {
+		size = 0
+	}
+	tw.curr = &regFileWriter{tw.w, size}
+	tw.pad = blockPadding(size)
+	return nil
 }
 
 // splitUSTARPath splits a path according to USTAR prefix and suffix rules.
 // If the path is not splittable, then it will return ("", "", false).
 func splitUSTARPath(name string) (prefix, suffix string, ok bool) {
 	length := len(name)
-	if length <= fileNameSize || !isASCII(name) {
+	if length <= nameSize || !isASCII(name) {
 		return "", "", false
-	} else if length > fileNamePrefixSize+1 {
+	} else if length > prefixSize+1 {
-		length = fileNamePrefixSize + 1
+		length = prefixSize + 1
 	} else if name[length-1] == '/' {
 		length--
 	}
@@ -300,117 +415,239 @@ func splitUSTARPath(name string) (prefix, suffix string, ok bool) {
 	i := strings.LastIndex(name[:length], "/")
 	nlen := len(name) - i - 1 // nlen is length of suffix
 	plen := i                 // plen is length of prefix
-	if i <= 0 || nlen > fileNameSize || nlen == 0 || plen > fileNamePrefixSize {
+	if i <= 0 || nlen > nameSize || nlen == 0 || plen > prefixSize {
 		return "", "", false
 	}
 	return name[:i], name[i+1:], true
 }
 
-// writePaxHeader writes an extended pax header to the
+// Write writes to the current file in the tar archive.
-// archive.
-func (tw *Writer) writePAXHeader(hdr *Header, paxHeaders map[string]string) error {
-	// Prepare extended header
-	ext := new(Header)
-	ext.Typeflag = TypeXHeader
-	// Setting ModTime is required for reader parsing to
-	// succeed, and seems harmless enough.
-	ext.ModTime = hdr.ModTime
-	// The spec asks that we namespace our pseudo files
-	// with the current pid.  However, this results in differing outputs
-	// for identical inputs.  As such, the constant 0 is now used instead.
-	// golang.org/issue/12358
-	dir, file := path.Split(hdr.Name)
-	fullName := path.Join(dir, "PaxHeaders.0", file)
-
-	ascii := toASCII(fullName)
-	if len(ascii) > 100 {
-		ascii = ascii[:100]
-	}
-	ext.Name = ascii
-	// Construct the body
-	var buf bytes.Buffer
-
-	// Keys are sorted before writing to body to allow deterministic output.
-	var keys []string
-	for k := range paxHeaders {
-		keys = append(keys, k)
-	}
-	sort.Strings(keys)
-
-	for _, k := range keys {
-		fmt.Fprint(&buf, formatPAXRecord(k, paxHeaders[k]))
-	}
-
-	ext.Size = int64(len(buf.Bytes()))
-	if err := tw.writeHeader(ext, false); err != nil {
-		return err
-	}
-	if _, err := tw.Write(buf.Bytes()); err != nil {
-		return err
-	}
-	if err := tw.Flush(); err != nil {
-		return err
-	}
-	return nil
-}
-
-// formatPAXRecord formats a single PAX record, prefixing it with the
-// appropriate length.
-func formatPAXRecord(k, v string) string {
-	const padding = 3 // Extra padding for ' ', '=', and '\n'
-	size := len(k) + len(v) + padding
-	size += len(strconv.Itoa(size))
-	record := fmt.Sprintf("%d %s=%s\n", size, k, v)
-
-	// Final adjustment if adding size field increased the record size.
-	if len(record) != size {
-		size = len(record)
-		record = fmt.Sprintf("%d %s=%s\n", size, k, v)
-	}
-	return record
-}
-
-// Write writes to the current entry in the tar archive.
 // Write returns the error ErrWriteTooLong if more than
-// hdr.Size bytes are written after WriteHeader.
+// Header.Size bytes are written after WriteHeader.
-func (tw *Writer) Write(b []byte) (n int, err error) {
+//
-	if tw.closed {
+// Calling Write on special types like TypeLink, TypeSymlink, TypeChar,
-		err = ErrWriteAfterClose
+// TypeBlock, TypeDir, and TypeFifo returns (0, ErrWriteTooLong) regardless
-		return
+// of what the Header.Size claims.
+func (tw *Writer) Write(b []byte) (int, error) {
+	if tw.err != nil {
+		return 0, tw.err
 	}
-	overwrite := false
+	n, err := tw.curr.Write(b)
-	if int64(len(b)) > tw.nb {
+	if err != nil && err != ErrWriteTooLong {
-		b = b[0:tw.nb]
+		tw.err = err
-		overwrite = true
 	}
-	n, err = tw.w.Write(b)
+	return n, err
-	tw.nb -= int64(n)
-	if err == nil && overwrite {
-		err = ErrWriteTooLong
-		return
-	}
-	tw.err = err
-	return
 }
 
-// Close closes the tar archive, flushing any unwritten
+// readFrom populates the content of the current file by reading from r.
-// data to the underlying writer.
+// The bytes read must match the number of remaining bytes in the current file.
-func (tw *Writer) Close() error {
+//
-	if tw.err != nil || tw.closed {
+// If the current file is sparse and r is an io.ReadSeeker,
-		return tw.err
+// then readFrom uses Seek to skip past holes defined in Header.SparseHoles,
+// assuming that skipped regions are all NULs.
+// This always reads the last byte to ensure r is the right size.
+//
+// TODO(dsnet): Re-export this when adding sparse file support.
+// See https://golang.org/issue/22735
+func (tw *Writer) readFrom(r io.Reader) (int64, error) {
+	if tw.err != nil {
+		return 0, tw.err
+	}
+	n, err := tw.curr.ReadFrom(r)
+	if err != nil && err != ErrWriteTooLong {
+		tw.err = err
+	}
+	return n, err
+}
+
+// Close closes the tar archive by flushing the padding, and writing the footer.
+// If the current file (from a prior call to WriteHeader) is not fully written,
+// then this returns an error.
+func (tw *Writer) Close() error {
+	if tw.err == ErrWriteAfterClose {
+		return nil
 	}
-	tw.Flush()
-	tw.closed = true
 	if tw.err != nil {
 		return tw.err
 	}
 
-	// trailer: two zero blocks
+	// Trailer: two zero blocks.
-	for i := 0; i < 2; i++ {
+	err := tw.Flush()
-		_, tw.err = tw.w.Write(zeroBlock)
+	for i := 0; i < 2 && err == nil; i++ {
-		if tw.err != nil {
+		_, err = tw.w.Write(zeroBlock[:])
-			break
+	}
+
+	// Ensure all future actions are invalid.
+	tw.err = ErrWriteAfterClose
+	return err // Report IO errors
+}
+
+// regFileWriter is a fileWriter for writing data to a regular file entry.
+type regFileWriter struct {
+	w  io.Writer // Underlying Writer
+	nb int64     // Number of remaining bytes to write
+}
+
+func (fw *regFileWriter) Write(b []byte) (n int, err error) {
+	overwrite := int64(len(b)) > fw.nb
+	if overwrite {
+		b = b[:fw.nb]
+	}
+	if len(b) > 0 {
+		n, err = fw.w.Write(b)
+		fw.nb -= int64(n)
+	}
+	switch {
+	case err != nil:
+		return n, err
+	case overwrite:
+		return n, ErrWriteTooLong
+	default:
+		return n, nil
+	}
+}
+
+func (fw *regFileWriter) ReadFrom(r io.Reader) (int64, error) {
+	return io.Copy(struct{ io.Writer }{fw}, r)
+}
+
+func (fw regFileWriter) LogicalRemaining() int64 {
+	return fw.nb
+}
+func (fw regFileWriter) PhysicalRemaining() int64 {
+	return fw.nb
+}
+
+// sparseFileWriter is a fileWriter for writing data to a sparse file entry.
+type sparseFileWriter struct {
+	fw  fileWriter  // Underlying fileWriter
+	sp  sparseDatas // Normalized list of data fragments
+	pos int64       // Current position in sparse file
+}
+
+func (sw *sparseFileWriter) Write(b []byte) (n int, err error) {
+	overwrite := int64(len(b)) > sw.LogicalRemaining()
+	if overwrite {
+		b = b[:sw.LogicalRemaining()]
+	}
+
+	b0 := b
+	endPos := sw.pos + int64(len(b))
+	for endPos > sw.pos && err == nil {
+		var nf int // Bytes written in fragment
+		dataStart, dataEnd := sw.sp[0].Offset, sw.sp[0].endOffset()
+		if sw.pos < dataStart { // In a hole fragment
+			bf := b[:min(int64(len(b)), dataStart-sw.pos)]
+			nf, err = zeroWriter{}.Write(bf)
+		} else { // In a data fragment
+			bf := b[:min(int64(len(b)), dataEnd-sw.pos)]
+			nf, err = sw.fw.Write(bf)
+		}
+		b = b[nf:]
+		sw.pos += int64(nf)
+		if sw.pos >= dataEnd && len(sw.sp) > 1 {
+			sw.sp = sw.sp[1:] // Ensure last fragment always remains
 		}
 	}
-	return tw.err
+
+	n = len(b0) - len(b)
+	switch {
+	case err == ErrWriteTooLong:
+		return n, errMissData // Not possible; implies bug in validation logic
+	case err != nil:
+		return n, err
+	case sw.LogicalRemaining() == 0 && sw.PhysicalRemaining() > 0:
+		return n, errUnrefData // Not possible; implies bug in validation logic
+	case overwrite:
+		return n, ErrWriteTooLong
+	default:
+		return n, nil
+	}
+}
+
+func (sw *sparseFileWriter) ReadFrom(r io.Reader) (n int64, err error) {
+	rs, ok := r.(io.ReadSeeker)
+	if ok {
+		if _, err := rs.Seek(0, io.SeekCurrent); err != nil {
+			ok = false // Not all io.Seeker can really seek
+		}
+	}
+	if !ok {
+		return io.Copy(struct{ io.Writer }{sw}, r)
+	}
+
+	var readLastByte bool
+	pos0 := sw.pos
+	for sw.LogicalRemaining() > 0 && !readLastByte && err == nil {
+		var nf int64 // Size of fragment
+		dataStart, dataEnd := sw.sp[0].Offset, sw.sp[0].endOffset()
+		if sw.pos < dataStart { // In a hole fragment
+			nf = dataStart - sw.pos
+			if sw.PhysicalRemaining() == 0 {
+				readLastByte = true
+				nf--
+			}
+			_, err = rs.Seek(nf, io.SeekCurrent)
+		} else { // In a data fragment
+			nf = dataEnd - sw.pos
+			nf, err = io.CopyN(sw.fw, rs, nf)
+		}
+		sw.pos += nf
+		if sw.pos >= dataEnd && len(sw.sp) > 1 {
+			sw.sp = sw.sp[1:] // Ensure last fragment always remains
+		}
+	}
+
+	// If the last fragment is a hole, then seek to 1-byte before EOF, and
+	// read a single byte to ensure the file is the right size.
+	if readLastByte && err == nil {
+		_, err = mustReadFull(rs, []byte{0})
+		sw.pos++
+	}
+
+	n = sw.pos - pos0
+	switch {
+	case err == io.EOF:
+		return n, io.ErrUnexpectedEOF
+	case err == ErrWriteTooLong:
+		return n, errMissData // Not possible; implies bug in validation logic
+	case err != nil:
+		return n, err
+	case sw.LogicalRemaining() == 0 && sw.PhysicalRemaining() > 0:
+		return n, errUnrefData // Not possible; implies bug in validation logic
+	default:
+		return n, ensureEOF(rs)
+	}
+}
+
+func (sw sparseFileWriter) LogicalRemaining() int64 {
+	return sw.sp[len(sw.sp)-1].endOffset() - sw.pos
+}
+func (sw sparseFileWriter) PhysicalRemaining() int64 {
+	return sw.fw.PhysicalRemaining()
+}
+
+// zeroWriter may only be written with NULs, otherwise it returns errWriteHole.
+type zeroWriter struct{}
+
+func (zeroWriter) Write(b []byte) (int, error) {
+	for i, c := range b {
+		if c != 0 {
+			return i, errWriteHole
+		}
+	}
+	return len(b), nil
+}
+
+// ensureEOF checks whether r is at EOF, reporting ErrWriteTooLong if not so.
+func ensureEOF(r io.Reader) error {
+	n, err := tryReadFull(r, []byte{0})
+	switch {
+	case n > 0:
+		return ErrWriteTooLong
+	case err == io.EOF:
+		return nil
+	default:
+		return err
+	}
 }