Browse code

archive/tar: vendor again

Signed-off-by: Kir Kolyshkin <kolyshkin@gmail.com>
(cherry picked from commit 7b30fd0e1d8bc77f0556181c82f85d046b058f27)
Signed-off-by: Sebastiaan van Stijn <github@gone.nl>

Kir Kolyshkin authored on 2018/04/11 07:46:21
Showing 12 changed files
... ...
@@ -41,6 +41,7 @@ RUN apt-get update && apt-get install -y --no-install-recommends \
41 41
 # IMPORTANT: If the version of Go is updated, the Windows to Linux CI machines
42 42
 #            will need updating, to avoid errors. Ping #docker-maintainers on IRC
43 43
 #            with a heads-up.
44
+# IMPORTANT: When updating this please note that stdlib archive/tar pkg is vendored
44 45
 ENV GO_VERSION 1.10.1
45 46
 RUN curl -fsSL "https://golang.org/dl/go${GO_VERSION}.linux-amd64.tar.gz" \
46 47
 	| tar -xzC /usr/local
... ...
@@ -150,3 +150,9 @@ github.com/Nvveen/Gotty a8b993ba6abdb0e0c12b0125c603323a71c7790c https://github.
150 150
 github.com/docker/go-metrics d466d4f6fd960e01820085bd7e1a24426ee7ef18
151 151
 
152 152
 github.com/opencontainers/selinux b29023b86e4a69d1b46b7e7b4e2b6fda03f0b9cd
153
+
154
+
155
+# archive/tar (for Go 1.10, see https://github.com/golang/go/issues/24787)
156
+# mkdir -p ./vendor/archive
157
+# git clone -b go-1.10 --depth=1 git@github.com:kolyshkin/go-tar.git ./vendor/archive/tar
158
+# vndr # to clean up test files
153 159
new file mode 100644
... ...
@@ -0,0 +1,27 @@
0
+Copyright (c) 2009 The Go Authors. All rights reserved.
1
+
2
+Redistribution and use in source and binary forms, with or without
3
+modification, are permitted provided that the following conditions are
4
+met:
5
+
6
+   * Redistributions of source code must retain the above copyright
7
+notice, this list of conditions and the following disclaimer.
8
+   * Redistributions in binary form must reproduce the above
9
+copyright notice, this list of conditions and the following disclaimer
10
+in the documentation and/or other materials provided with the
11
+distribution.
12
+   * Neither the name of Google Inc. nor the names of its
13
+contributors may be used to endorse or promote products derived from
14
+this software without specific prior written permission.
15
+
16
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
17
+"AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
18
+LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
19
+A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
20
+OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
21
+SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
22
+LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
23
+DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
24
+THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
25
+(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
26
+OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
0 27
new file mode 100644
... ...
@@ -0,0 +1,27 @@
0
+This is a fork of Go 1.10 `archive/tar` package from the official
1
+[repo](https://github.com/golang/go/tree/release-branch.go1.10/src/archive/tar),
2
+with a partial [revert](https://github.com/kolyshkin/go-tar/commit/d651d6e45972363e9bb62b8e9d876df440b31628)
3
+of upstream [commit 0564e304a6ea](https://github.com/golang/go/commit/0564e304a6ea394a42929060c588469dbd6f32af).
4
+It is suggested as a replacement to the original package included with Go 1.10
5
+in case you want to build a static Linux/glibc binary that works, and
6
+can't afford to use `CGO_ENABLED=0`.
7
+
8
+## Details
9
+
10
+Using Go 1.10 [archive/tar](https://golang.org/pkg/archive/tar/) from a static binary
11
+compiled with glibc on Linux can result in a panic upon calling
12
+[`tar.FileInfoHeader()`](https://golang.org/pkg/archive/tar/#FileInfoHeader).
13
+This is a major regression in Go 1.10, filed as
14
+[Go issue #24787](https://github.com/golang/go/issues/24787).
15
+
16
+The above issue is caused by an unfortunate combination of:
17
+1. glibc way of dynamic loading of nss libraries even for a static build;
18
+2. Go `os/user` package hard-coded reliance on libc to resolve user/group IDs to names (unless CGO is disabled).
19
+
20
+While glibc can probably not be fixed and is not considered a bug per se,
21
+the `os/user` issue is documented (see [Go issue #23265](https://github.com/golang/go/issues/23265))
22
+and already fixed by [Go commit 62f0127d81](https://github.com/golang/go/commit/62f0127d8104d8266d9a3fb5a87e2f09ec8b6f5b).
23
+The fix is expected to make its way to Go 1.11, and requires `osusergo` build tag
24
+to be used for a static build.
25
+
26
+This repository serves as a temporary workaround until the above fix is available.
0 27
new file mode 100644
... ...
@@ -0,0 +1,720 @@
0
+// Copyright 2009 The Go Authors. All rights reserved.
1
+// Use of this source code is governed by a BSD-style
2
+// license that can be found in the LICENSE file.
3
+
4
+// Package tar implements access to tar archives.
5
+//
6
+// Tape archives (tar) are a file format for storing a sequence of files that
7
+// can be read and written in a streaming manner.
8
+// This package aims to cover most variations of the format,
9
+// including those produced by GNU and BSD tar tools.
10
+package tar
11
+
12
+import (
13
+	"errors"
14
+	"fmt"
15
+	"math"
16
+	"os"
17
+	"path"
18
+	"reflect"
19
+	"strconv"
20
+	"strings"
21
+	"time"
22
+)
23
+
24
+// BUG: Use of the Uid and Gid fields in Header could overflow on 32-bit
25
+// architectures. If a large value is encountered when decoding, the result
26
+// stored in Header will be the truncated version.
27
+
28
+var (
29
+	ErrHeader          = errors.New("archive/tar: invalid tar header")
30
+	ErrWriteTooLong    = errors.New("archive/tar: write too long")
31
+	ErrFieldTooLong    = errors.New("archive/tar: header field too long")
32
+	ErrWriteAfterClose = errors.New("archive/tar: write after close")
33
+	errMissData        = errors.New("archive/tar: sparse file references non-existent data")
34
+	errUnrefData       = errors.New("archive/tar: sparse file contains unreferenced data")
35
+	errWriteHole       = errors.New("archive/tar: write non-NUL byte in sparse hole")
36
+)
37
+
38
+type headerError []string
39
+
40
+func (he headerError) Error() string {
41
+	const prefix = "archive/tar: cannot encode header"
42
+	var ss []string
43
+	for _, s := range he {
44
+		if s != "" {
45
+			ss = append(ss, s)
46
+		}
47
+	}
48
+	if len(ss) == 0 {
49
+		return prefix
50
+	}
51
+	return fmt.Sprintf("%s: %v", prefix, strings.Join(ss, "; and "))
52
+}
53
+
54
+// Type flags for Header.Typeflag.
55
+const (
56
+	// Type '0' indicates a regular file.
57
+	TypeReg  = '0'
58
+	TypeRegA = '\x00' // For legacy support; use TypeReg instead
59
+
60
+	// Type '1' to '6' are header-only flags and may not have a data body.
61
+	TypeLink    = '1' // Hard link
62
+	TypeSymlink = '2' // Symbolic link
63
+	TypeChar    = '3' // Character device node
64
+	TypeBlock   = '4' // Block device node
65
+	TypeDir     = '5' // Directory
66
+	TypeFifo    = '6' // FIFO node
67
+
68
+	// Type '7' is reserved.
69
+	TypeCont = '7'
70
+
71
+	// Type 'x' is used by the PAX format to store key-value records that
72
+	// are only relevant to the next file.
73
+	// This package transparently handles these types.
74
+	TypeXHeader = 'x'
75
+
76
+	// Type 'g' is used by the PAX format to store key-value records that
77
+	// are relevant to all subsequent files.
78
+	// This package only supports parsing and composing such headers,
79
+	// but does not currently support persisting the global state across files.
80
+	TypeXGlobalHeader = 'g'
81
+
82
+	// Type 'S' indicates a sparse file in the GNU format.
83
+	TypeGNUSparse = 'S'
84
+
85
+	// Types 'L' and 'K' are used by the GNU format for a meta file
86
+	// used to store the path or link name for the next file.
87
+	// This package transparently handles these types.
88
+	TypeGNULongName = 'L'
89
+	TypeGNULongLink = 'K'
90
+)
91
+
92
+// Keywords for PAX extended header records.
93
+const (
94
+	paxNone     = "" // Indicates that no PAX key is suitable
95
+	paxPath     = "path"
96
+	paxLinkpath = "linkpath"
97
+	paxSize     = "size"
98
+	paxUid      = "uid"
99
+	paxGid      = "gid"
100
+	paxUname    = "uname"
101
+	paxGname    = "gname"
102
+	paxMtime    = "mtime"
103
+	paxAtime    = "atime"
104
+	paxCtime    = "ctime"   // Removed from later revision of PAX spec, but was valid
105
+	paxCharset  = "charset" // Currently unused
106
+	paxComment  = "comment" // Currently unused
107
+
108
+	paxSchilyXattr = "SCHILY.xattr."
109
+
110
+	// Keywords for GNU sparse files in a PAX extended header.
111
+	paxGNUSparse          = "GNU.sparse."
112
+	paxGNUSparseNumBlocks = "GNU.sparse.numblocks"
113
+	paxGNUSparseOffset    = "GNU.sparse.offset"
114
+	paxGNUSparseNumBytes  = "GNU.sparse.numbytes"
115
+	paxGNUSparseMap       = "GNU.sparse.map"
116
+	paxGNUSparseName      = "GNU.sparse.name"
117
+	paxGNUSparseMajor     = "GNU.sparse.major"
118
+	paxGNUSparseMinor     = "GNU.sparse.minor"
119
+	paxGNUSparseSize      = "GNU.sparse.size"
120
+	paxGNUSparseRealSize  = "GNU.sparse.realsize"
121
+)
122
+
123
+// basicKeys is a set of the PAX keys for which we have built-in support.
124
+// This does not contain "charset" or "comment", which are both PAX-specific,
125
+// so adding them as first-class features of Header is unlikely.
126
+// Users can use the PAXRecords field to set it themselves.
127
+var basicKeys = map[string]bool{
128
+	paxPath: true, paxLinkpath: true, paxSize: true, paxUid: true, paxGid: true,
129
+	paxUname: true, paxGname: true, paxMtime: true, paxAtime: true, paxCtime: true,
130
+}
131
+
132
+// A Header represents a single header in a tar archive.
133
+// Some fields may not be populated.
134
+//
135
+// For forward compatibility, users that retrieve a Header from Reader.Next,
136
+// mutate it in some ways, and then pass it back to Writer.WriteHeader
137
+// should do so by creating a new Header and copying the fields
138
+// that they are interested in preserving.
139
+type Header struct {
140
+	Typeflag byte // Type of header entry (should be TypeReg for most files)
141
+
142
+	Name     string // Name of file entry
143
+	Linkname string // Target name of link (valid for TypeLink or TypeSymlink)
144
+
145
+	Size  int64  // Logical file size in bytes
146
+	Mode  int64  // Permission and mode bits
147
+	Uid   int    // User ID of owner
148
+	Gid   int    // Group ID of owner
149
+	Uname string // User name of owner
150
+	Gname string // Group name of owner
151
+
152
+	// If the Format is unspecified, then Writer.WriteHeader rounds ModTime
153
+	// to the nearest second and ignores the AccessTime and ChangeTime fields.
154
+	//
155
+	// To use AccessTime or ChangeTime, specify the Format as PAX or GNU.
156
+	// To use sub-second resolution, specify the Format as PAX.
157
+	ModTime    time.Time // Modification time
158
+	AccessTime time.Time // Access time (requires either PAX or GNU support)
159
+	ChangeTime time.Time // Change time (requires either PAX or GNU support)
160
+
161
+	Devmajor int64 // Major device number (valid for TypeChar or TypeBlock)
162
+	Devminor int64 // Minor device number (valid for TypeChar or TypeBlock)
163
+
164
+	// Xattrs stores extended attributes as PAX records under the
165
+	// "SCHILY.xattr." namespace.
166
+	//
167
+	// The following are semantically equivalent:
168
+	//  h.Xattrs[key] = value
169
+	//  h.PAXRecords["SCHILY.xattr."+key] = value
170
+	//
171
+	// When Writer.WriteHeader is called, the contents of Xattrs will take
172
+	// precedence over those in PAXRecords.
173
+	//
174
+	// Deprecated: Use PAXRecords instead.
175
+	Xattrs map[string]string
176
+
177
+	// PAXRecords is a map of PAX extended header records.
178
+	//
179
+	// User-defined records should have keys of the following form:
180
+	//	VENDOR.keyword
181
+	// Where VENDOR is some namespace in all uppercase, and keyword may
182
+	// not contain the '=' character (e.g., "GOLANG.pkg.version").
183
+	// The key and value should be non-empty UTF-8 strings.
184
+	//
185
+	// When Writer.WriteHeader is called, PAX records derived from the
186
+	// the other fields in Header take precedence over PAXRecords.
187
+	PAXRecords map[string]string
188
+
189
+	// Format specifies the format of the tar header.
190
+	//
191
+	// This is set by Reader.Next as a best-effort guess at the format.
192
+	// Since the Reader liberally reads some non-compliant files,
193
+	// it is possible for this to be FormatUnknown.
194
+	//
195
+	// If the format is unspecified when Writer.WriteHeader is called,
196
+	// then it uses the first format (in the order of USTAR, PAX, GNU)
197
+	// capable of encoding this Header (see Format).
198
+	Format Format
199
+}
200
+
201
+// sparseEntry represents a Length-sized fragment at Offset in the file.
202
+type sparseEntry struct{ Offset, Length int64 }
203
+
204
+func (s sparseEntry) endOffset() int64 { return s.Offset + s.Length }
205
+
206
+// A sparse file can be represented as either a sparseDatas or a sparseHoles.
207
+// As long as the total size is known, they are equivalent and one can be
208
+// converted to the other form and back. The various tar formats with sparse
209
+// file support represent sparse files in the sparseDatas form. That is, they
210
+// specify the fragments in the file that has data, and treat everything else as
211
+// having zero bytes. As such, the encoding and decoding logic in this package
212
+// deals with sparseDatas.
213
+//
214
+// However, the external API uses sparseHoles instead of sparseDatas because the
215
+// zero value of sparseHoles logically represents a normal file (i.e., there are
216
+// no holes in it). On the other hand, the zero value of sparseDatas implies
217
+// that the file has no data in it, which is rather odd.
218
+//
219
+// As an example, if the underlying raw file contains the 10-byte data:
220
+//	var compactFile = "abcdefgh"
221
+//
222
+// And the sparse map has the following entries:
223
+//	var spd sparseDatas = []sparseEntry{
224
+//		{Offset: 2,  Length: 5},  // Data fragment for 2..6
225
+//		{Offset: 18, Length: 3},  // Data fragment for 18..20
226
+//	}
227
+//	var sph sparseHoles = []sparseEntry{
228
+//		{Offset: 0,  Length: 2},  // Hole fragment for 0..1
229
+//		{Offset: 7,  Length: 11}, // Hole fragment for 7..17
230
+//		{Offset: 21, Length: 4},  // Hole fragment for 21..24
231
+//	}
232
+//
233
+// Then the content of the resulting sparse file with a Header.Size of 25 is:
234
+//	var sparseFile = "\x00"*2 + "abcde" + "\x00"*11 + "fgh" + "\x00"*4
235
+type (
236
+	sparseDatas []sparseEntry
237
+	sparseHoles []sparseEntry
238
+)
239
+
240
+// validateSparseEntries reports whether sp is a valid sparse map.
241
+// It does not matter whether sp represents data fragments or hole fragments.
242
+func validateSparseEntries(sp []sparseEntry, size int64) bool {
243
+	// Validate all sparse entries. These are the same checks as performed by
244
+	// the BSD tar utility.
245
+	if size < 0 {
246
+		return false
247
+	}
248
+	var pre sparseEntry
249
+	for _, cur := range sp {
250
+		switch {
251
+		case cur.Offset < 0 || cur.Length < 0:
252
+			return false // Negative values are never okay
253
+		case cur.Offset > math.MaxInt64-cur.Length:
254
+			return false // Integer overflow with large length
255
+		case cur.endOffset() > size:
256
+			return false // Region extends beyond the actual size
257
+		case pre.endOffset() > cur.Offset:
258
+			return false // Regions cannot overlap and must be in order
259
+		}
260
+		pre = cur
261
+	}
262
+	return true
263
+}
264
+
265
+// alignSparseEntries mutates src and returns dst where each fragment's
266
+// starting offset is aligned up to the nearest block edge, and each
267
+// ending offset is aligned down to the nearest block edge.
268
+//
269
+// Even though the Go tar Reader and the BSD tar utility can handle entries
270
+// with arbitrary offsets and lengths, the GNU tar utility can only handle
271
+// offsets and lengths that are multiples of blockSize.
272
+func alignSparseEntries(src []sparseEntry, size int64) []sparseEntry {
273
+	dst := src[:0]
274
+	for _, s := range src {
275
+		pos, end := s.Offset, s.endOffset()
276
+		pos += blockPadding(+pos) // Round-up to nearest blockSize
277
+		if end != size {
278
+			end -= blockPadding(-end) // Round-down to nearest blockSize
279
+		}
280
+		if pos < end {
281
+			dst = append(dst, sparseEntry{Offset: pos, Length: end - pos})
282
+		}
283
+	}
284
+	return dst
285
+}
286
+
287
+// invertSparseEntries converts a sparse map from one form to the other.
288
+// If the input is sparseHoles, then it will output sparseDatas and vice-versa.
289
+// The input must have been already validated.
290
+//
291
+// This function mutates src and returns a normalized map where:
292
+//	* adjacent fragments are coalesced together
293
+//	* only the last fragment may be empty
294
+//	* the endOffset of the last fragment is the total size
295
+func invertSparseEntries(src []sparseEntry, size int64) []sparseEntry {
296
+	dst := src[:0]
297
+	var pre sparseEntry
298
+	for _, cur := range src {
299
+		if cur.Length == 0 {
300
+			continue // Skip empty fragments
301
+		}
302
+		pre.Length = cur.Offset - pre.Offset
303
+		if pre.Length > 0 {
304
+			dst = append(dst, pre) // Only add non-empty fragments
305
+		}
306
+		pre.Offset = cur.endOffset()
307
+	}
308
+	pre.Length = size - pre.Offset // Possibly the only empty fragment
309
+	return append(dst, pre)
310
+}
311
+
312
+// fileState tracks the number of logical (includes sparse holes) and physical
313
+// (actual in tar archive) bytes remaining for the current file.
314
+//
315
+// Invariant: LogicalRemaining >= PhysicalRemaining
316
+type fileState interface {
317
+	LogicalRemaining() int64
318
+	PhysicalRemaining() int64
319
+}
320
+
321
+// allowedFormats determines which formats can be used.
322
+// The value returned is the logical OR of multiple possible formats.
323
+// If the value is FormatUnknown, then the input Header cannot be encoded
324
+// and an error is returned explaining why.
325
+//
326
+// As a by-product of checking the fields, this function returns paxHdrs, which
327
+// contain all fields that could not be directly encoded.
328
+// A value receiver ensures that this method does not mutate the source Header.
329
+func (h Header) allowedFormats() (format Format, paxHdrs map[string]string, err error) {
330
+	format = FormatUSTAR | FormatPAX | FormatGNU
331
+	paxHdrs = make(map[string]string)
332
+
333
+	var whyNoUSTAR, whyNoPAX, whyNoGNU string
334
+	var preferPAX bool // Prefer PAX over USTAR
335
+	verifyString := func(s string, size int, name, paxKey string) {
336
+		// NUL-terminator is optional for path and linkpath.
337
+		// Technically, it is required for uname and gname,
338
+		// but neither GNU nor BSD tar checks for it.
339
+		tooLong := len(s) > size
340
+		allowLongGNU := paxKey == paxPath || paxKey == paxLinkpath
341
+		if hasNUL(s) || (tooLong && !allowLongGNU) {
342
+			whyNoGNU = fmt.Sprintf("GNU cannot encode %s=%q", name, s)
343
+			format.mustNotBe(FormatGNU)
344
+		}
345
+		if !isASCII(s) || tooLong {
346
+			canSplitUSTAR := paxKey == paxPath
347
+			if _, _, ok := splitUSTARPath(s); !canSplitUSTAR || !ok {
348
+				whyNoUSTAR = fmt.Sprintf("USTAR cannot encode %s=%q", name, s)
349
+				format.mustNotBe(FormatUSTAR)
350
+			}
351
+			if paxKey == paxNone {
352
+				whyNoPAX = fmt.Sprintf("PAX cannot encode %s=%q", name, s)
353
+				format.mustNotBe(FormatPAX)
354
+			} else {
355
+				paxHdrs[paxKey] = s
356
+			}
357
+		}
358
+		if v, ok := h.PAXRecords[paxKey]; ok && v == s {
359
+			paxHdrs[paxKey] = v
360
+		}
361
+	}
362
+	verifyNumeric := func(n int64, size int, name, paxKey string) {
363
+		if !fitsInBase256(size, n) {
364
+			whyNoGNU = fmt.Sprintf("GNU cannot encode %s=%d", name, n)
365
+			format.mustNotBe(FormatGNU)
366
+		}
367
+		if !fitsInOctal(size, n) {
368
+			whyNoUSTAR = fmt.Sprintf("USTAR cannot encode %s=%d", name, n)
369
+			format.mustNotBe(FormatUSTAR)
370
+			if paxKey == paxNone {
371
+				whyNoPAX = fmt.Sprintf("PAX cannot encode %s=%d", name, n)
372
+				format.mustNotBe(FormatPAX)
373
+			} else {
374
+				paxHdrs[paxKey] = strconv.FormatInt(n, 10)
375
+			}
376
+		}
377
+		if v, ok := h.PAXRecords[paxKey]; ok && v == strconv.FormatInt(n, 10) {
378
+			paxHdrs[paxKey] = v
379
+		}
380
+	}
381
+	verifyTime := func(ts time.Time, size int, name, paxKey string) {
382
+		if ts.IsZero() {
383
+			return // Always okay
384
+		}
385
+		if !fitsInBase256(size, ts.Unix()) {
386
+			whyNoGNU = fmt.Sprintf("GNU cannot encode %s=%v", name, ts)
387
+			format.mustNotBe(FormatGNU)
388
+		}
389
+		isMtime := paxKey == paxMtime
390
+		fitsOctal := fitsInOctal(size, ts.Unix())
391
+		if (isMtime && !fitsOctal) || !isMtime {
392
+			whyNoUSTAR = fmt.Sprintf("USTAR cannot encode %s=%v", name, ts)
393
+			format.mustNotBe(FormatUSTAR)
394
+		}
395
+		needsNano := ts.Nanosecond() != 0
396
+		if !isMtime || !fitsOctal || needsNano {
397
+			preferPAX = true // USTAR may truncate sub-second measurements
398
+			if paxKey == paxNone {
399
+				whyNoPAX = fmt.Sprintf("PAX cannot encode %s=%v", name, ts)
400
+				format.mustNotBe(FormatPAX)
401
+			} else {
402
+				paxHdrs[paxKey] = formatPAXTime(ts)
403
+			}
404
+		}
405
+		if v, ok := h.PAXRecords[paxKey]; ok && v == formatPAXTime(ts) {
406
+			paxHdrs[paxKey] = v
407
+		}
408
+	}
409
+
410
+	// Check basic fields.
411
+	var blk block
412
+	v7 := blk.V7()
413
+	ustar := blk.USTAR()
414
+	gnu := blk.GNU()
415
+	verifyString(h.Name, len(v7.Name()), "Name", paxPath)
416
+	verifyString(h.Linkname, len(v7.LinkName()), "Linkname", paxLinkpath)
417
+	verifyString(h.Uname, len(ustar.UserName()), "Uname", paxUname)
418
+	verifyString(h.Gname, len(ustar.GroupName()), "Gname", paxGname)
419
+	verifyNumeric(h.Mode, len(v7.Mode()), "Mode", paxNone)
420
+	verifyNumeric(int64(h.Uid), len(v7.UID()), "Uid", paxUid)
421
+	verifyNumeric(int64(h.Gid), len(v7.GID()), "Gid", paxGid)
422
+	verifyNumeric(h.Size, len(v7.Size()), "Size", paxSize)
423
+	verifyNumeric(h.Devmajor, len(ustar.DevMajor()), "Devmajor", paxNone)
424
+	verifyNumeric(h.Devminor, len(ustar.DevMinor()), "Devminor", paxNone)
425
+	verifyTime(h.ModTime, len(v7.ModTime()), "ModTime", paxMtime)
426
+	verifyTime(h.AccessTime, len(gnu.AccessTime()), "AccessTime", paxAtime)
427
+	verifyTime(h.ChangeTime, len(gnu.ChangeTime()), "ChangeTime", paxCtime)
428
+
429
+	// Check for header-only types.
430
+	var whyOnlyPAX, whyOnlyGNU string
431
+	switch h.Typeflag {
432
+	case TypeReg, TypeChar, TypeBlock, TypeFifo, TypeGNUSparse:
433
+		// Exclude TypeLink and TypeSymlink, since they may reference directories.
434
+		if strings.HasSuffix(h.Name, "/") {
435
+			return FormatUnknown, nil, headerError{"filename may not have trailing slash"}
436
+		}
437
+	case TypeXHeader, TypeGNULongName, TypeGNULongLink:
438
+		return FormatUnknown, nil, headerError{"cannot manually encode TypeXHeader, TypeGNULongName, or TypeGNULongLink headers"}
439
+	case TypeXGlobalHeader:
440
+		h2 := Header{Name: h.Name, Typeflag: h.Typeflag, Xattrs: h.Xattrs, PAXRecords: h.PAXRecords, Format: h.Format}
441
+		if !reflect.DeepEqual(h, h2) {
442
+			return FormatUnknown, nil, headerError{"only PAXRecords should be set for TypeXGlobalHeader"}
443
+		}
444
+		whyOnlyPAX = "only PAX supports TypeXGlobalHeader"
445
+		format.mayOnlyBe(FormatPAX)
446
+	}
447
+	if !isHeaderOnlyType(h.Typeflag) && h.Size < 0 {
448
+		return FormatUnknown, nil, headerError{"negative size on header-only type"}
449
+	}
450
+
451
+	// Check PAX records.
452
+	if len(h.Xattrs) > 0 {
453
+		for k, v := range h.Xattrs {
454
+			paxHdrs[paxSchilyXattr+k] = v
455
+		}
456
+		whyOnlyPAX = "only PAX supports Xattrs"
457
+		format.mayOnlyBe(FormatPAX)
458
+	}
459
+	if len(h.PAXRecords) > 0 {
460
+		for k, v := range h.PAXRecords {
461
+			switch _, exists := paxHdrs[k]; {
462
+			case exists:
463
+				continue // Do not overwrite existing records
464
+			case h.Typeflag == TypeXGlobalHeader:
465
+				paxHdrs[k] = v // Copy all records
466
+			case !basicKeys[k] && !strings.HasPrefix(k, paxGNUSparse):
467
+				paxHdrs[k] = v // Ignore local records that may conflict
468
+			}
469
+		}
470
+		whyOnlyPAX = "only PAX supports PAXRecords"
471
+		format.mayOnlyBe(FormatPAX)
472
+	}
473
+	for k, v := range paxHdrs {
474
+		if !validPAXRecord(k, v) {
475
+			return FormatUnknown, nil, headerError{fmt.Sprintf("invalid PAX record: %q", k+" = "+v)}
476
+		}
477
+	}
478
+
479
+	// TODO(dsnet): Re-enable this when adding sparse support.
480
+	// See https://golang.org/issue/22735
481
+	/*
482
+		// Check sparse files.
483
+		if len(h.SparseHoles) > 0 || h.Typeflag == TypeGNUSparse {
484
+			if isHeaderOnlyType(h.Typeflag) {
485
+				return FormatUnknown, nil, headerError{"header-only type cannot be sparse"}
486
+			}
487
+			if !validateSparseEntries(h.SparseHoles, h.Size) {
488
+				return FormatUnknown, nil, headerError{"invalid sparse holes"}
489
+			}
490
+			if h.Typeflag == TypeGNUSparse {
491
+				whyOnlyGNU = "only GNU supports TypeGNUSparse"
492
+				format.mayOnlyBe(FormatGNU)
493
+			} else {
494
+				whyNoGNU = "GNU supports sparse files only with TypeGNUSparse"
495
+				format.mustNotBe(FormatGNU)
496
+			}
497
+			whyNoUSTAR = "USTAR does not support sparse files"
498
+			format.mustNotBe(FormatUSTAR)
499
+		}
500
+	*/
501
+
502
+	// Check desired format.
503
+	if wantFormat := h.Format; wantFormat != FormatUnknown {
504
+		if wantFormat.has(FormatPAX) && !preferPAX {
505
+			wantFormat.mayBe(FormatUSTAR) // PAX implies USTAR allowed too
506
+		}
507
+		format.mayOnlyBe(wantFormat) // Set union of formats allowed and format wanted
508
+	}
509
+	if format == FormatUnknown {
510
+		switch h.Format {
511
+		case FormatUSTAR:
512
+			err = headerError{"Format specifies USTAR", whyNoUSTAR, whyOnlyPAX, whyOnlyGNU}
513
+		case FormatPAX:
514
+			err = headerError{"Format specifies PAX", whyNoPAX, whyOnlyGNU}
515
+		case FormatGNU:
516
+			err = headerError{"Format specifies GNU", whyNoGNU, whyOnlyPAX}
517
+		default:
518
+			err = headerError{whyNoUSTAR, whyNoPAX, whyNoGNU, whyOnlyPAX, whyOnlyGNU}
519
+		}
520
+	}
521
+	return format, paxHdrs, err
522
+}
523
+
524
+// FileInfo returns an os.FileInfo for the Header.
525
+func (h *Header) FileInfo() os.FileInfo {
526
+	return headerFileInfo{h}
527
+}
528
+
529
+// headerFileInfo implements os.FileInfo.
530
+type headerFileInfo struct {
531
+	h *Header
532
+}
533
+
534
+func (fi headerFileInfo) Size() int64        { return fi.h.Size }
535
+func (fi headerFileInfo) IsDir() bool        { return fi.Mode().IsDir() }
536
+func (fi headerFileInfo) ModTime() time.Time { return fi.h.ModTime }
537
+func (fi headerFileInfo) Sys() interface{}   { return fi.h }
538
+
539
+// Name returns the base name of the file.
540
+func (fi headerFileInfo) Name() string {
541
+	if fi.IsDir() {
542
+		return path.Base(path.Clean(fi.h.Name))
543
+	}
544
+	return path.Base(fi.h.Name)
545
+}
546
+
547
+// Mode returns the permission and mode bits for the headerFileInfo.
548
+func (fi headerFileInfo) Mode() (mode os.FileMode) {
549
+	// Set file permission bits.
550
+	mode = os.FileMode(fi.h.Mode).Perm()
551
+
552
+	// Set setuid, setgid and sticky bits.
553
+	if fi.h.Mode&c_ISUID != 0 {
554
+		mode |= os.ModeSetuid
555
+	}
556
+	if fi.h.Mode&c_ISGID != 0 {
557
+		mode |= os.ModeSetgid
558
+	}
559
+	if fi.h.Mode&c_ISVTX != 0 {
560
+		mode |= os.ModeSticky
561
+	}
562
+
563
+	// Set file mode bits; clear perm, setuid, setgid, and sticky bits.
564
+	switch m := os.FileMode(fi.h.Mode) &^ 07777; m {
565
+	case c_ISDIR:
566
+		mode |= os.ModeDir
567
+	case c_ISFIFO:
568
+		mode |= os.ModeNamedPipe
569
+	case c_ISLNK:
570
+		mode |= os.ModeSymlink
571
+	case c_ISBLK:
572
+		mode |= os.ModeDevice
573
+	case c_ISCHR:
574
+		mode |= os.ModeDevice
575
+		mode |= os.ModeCharDevice
576
+	case c_ISSOCK:
577
+		mode |= os.ModeSocket
578
+	}
579
+
580
+	switch fi.h.Typeflag {
581
+	case TypeSymlink:
582
+		mode |= os.ModeSymlink
583
+	case TypeChar:
584
+		mode |= os.ModeDevice
585
+		mode |= os.ModeCharDevice
586
+	case TypeBlock:
587
+		mode |= os.ModeDevice
588
+	case TypeDir:
589
+		mode |= os.ModeDir
590
+	case TypeFifo:
591
+		mode |= os.ModeNamedPipe
592
+	}
593
+
594
+	return mode
595
+}
596
+
597
+// sysStat, if non-nil, populates h from system-dependent fields of fi.
598
+var sysStat func(fi os.FileInfo, h *Header) error
599
+
600
+const (
601
+	// Mode constants from the USTAR spec:
602
+	// See http://pubs.opengroup.org/onlinepubs/9699919799/utilities/pax.html#tag_20_92_13_06
603
+	c_ISUID = 04000 // Set uid
604
+	c_ISGID = 02000 // Set gid
605
+	c_ISVTX = 01000 // Save text (sticky bit)
606
+
607
+	// Common Unix mode constants; these are not defined in any common tar standard.
608
+	// Header.FileInfo understands these, but FileInfoHeader will never produce these.
609
+	c_ISDIR  = 040000  // Directory
610
+	c_ISFIFO = 010000  // FIFO
611
+	c_ISREG  = 0100000 // Regular file
612
+	c_ISLNK  = 0120000 // Symbolic link
613
+	c_ISBLK  = 060000  // Block special file
614
+	c_ISCHR  = 020000  // Character special file
615
+	c_ISSOCK = 0140000 // Socket
616
+)
617
+
618
+// FileInfoHeader creates a partially-populated Header from fi.
619
+// If fi describes a symlink, FileInfoHeader records link as the link target.
620
+// If fi describes a directory, a slash is appended to the name.
621
+//
622
+// Since os.FileInfo's Name method only returns the base name of
623
+// the file it describes, it may be necessary to modify Header.Name
624
+// to provide the full path name of the file.
625
+func FileInfoHeader(fi os.FileInfo, link string) (*Header, error) {
626
+	if fi == nil {
627
+		return nil, errors.New("archive/tar: FileInfo is nil")
628
+	}
629
+	fm := fi.Mode()
630
+	h := &Header{
631
+		Name:    fi.Name(),
632
+		ModTime: fi.ModTime(),
633
+		Mode:    int64(fm.Perm()), // or'd with c_IS* constants later
634
+	}
635
+	switch {
636
+	case fm.IsRegular():
637
+		h.Typeflag = TypeReg
638
+		h.Size = fi.Size()
639
+	case fi.IsDir():
640
+		h.Typeflag = TypeDir
641
+		h.Name += "/"
642
+	case fm&os.ModeSymlink != 0:
643
+		h.Typeflag = TypeSymlink
644
+		h.Linkname = link
645
+	case fm&os.ModeDevice != 0:
646
+		if fm&os.ModeCharDevice != 0 {
647
+			h.Typeflag = TypeChar
648
+		} else {
649
+			h.Typeflag = TypeBlock
650
+		}
651
+	case fm&os.ModeNamedPipe != 0:
652
+		h.Typeflag = TypeFifo
653
+	case fm&os.ModeSocket != 0:
654
+		return nil, fmt.Errorf("archive/tar: sockets not supported")
655
+	default:
656
+		return nil, fmt.Errorf("archive/tar: unknown file mode %v", fm)
657
+	}
658
+	if fm&os.ModeSetuid != 0 {
659
+		h.Mode |= c_ISUID
660
+	}
661
+	if fm&os.ModeSetgid != 0 {
662
+		h.Mode |= c_ISGID
663
+	}
664
+	if fm&os.ModeSticky != 0 {
665
+		h.Mode |= c_ISVTX
666
+	}
667
+	// If possible, populate additional fields from OS-specific
668
+	// FileInfo fields.
669
+	if sys, ok := fi.Sys().(*Header); ok {
670
+		// This FileInfo came from a Header (not the OS). Use the
671
+		// original Header to populate all remaining fields.
672
+		h.Uid = sys.Uid
673
+		h.Gid = sys.Gid
674
+		h.Uname = sys.Uname
675
+		h.Gname = sys.Gname
676
+		h.AccessTime = sys.AccessTime
677
+		h.ChangeTime = sys.ChangeTime
678
+		if sys.Xattrs != nil {
679
+			h.Xattrs = make(map[string]string)
680
+			for k, v := range sys.Xattrs {
681
+				h.Xattrs[k] = v
682
+			}
683
+		}
684
+		if sys.Typeflag == TypeLink {
685
+			// hard link
686
+			h.Typeflag = TypeLink
687
+			h.Size = 0
688
+			h.Linkname = sys.Linkname
689
+		}
690
+		if sys.PAXRecords != nil {
691
+			h.PAXRecords = make(map[string]string)
692
+			for k, v := range sys.PAXRecords {
693
+				h.PAXRecords[k] = v
694
+			}
695
+		}
696
+	}
697
+	if sysStat != nil {
698
+		return h, sysStat(fi, h)
699
+	}
700
+	return h, nil
701
+}
702
+
703
+// isHeaderOnlyType checks if the given type flag is of the type that has no
704
+// data section even if a size is specified.
705
+func isHeaderOnlyType(flag byte) bool {
706
+	switch flag {
707
+	case TypeLink, TypeSymlink, TypeChar, TypeBlock, TypeDir, TypeFifo:
708
+		return true
709
+	default:
710
+		return false
711
+	}
712
+}
713
+
714
+func min(a, b int64) int64 {
715
+	if a < b {
716
+		return a
717
+	}
718
+	return b
719
+}
0 720
new file mode 100644
... ...
@@ -0,0 +1,303 @@
0
+// Copyright 2016 The Go Authors. All rights reserved.
1
+// Use of this source code is governed by a BSD-style
2
+// license that can be found in the LICENSE file.
3
+
4
+package tar
5
+
6
+import "strings"
7
+
8
+// Format represents the tar archive format.
9
+//
10
+// The original tar format was introduced in Unix V7.
11
+// Since then, there have been multiple competing formats attempting to
12
+// standardize or extend the V7 format to overcome its limitations.
13
+// The most common formats are the USTAR, PAX, and GNU formats,
14
+// each with their own advantages and limitations.
15
+//
16
+// The following table captures the capabilities of each format:
17
+//
18
+//	                  |  USTAR |       PAX |       GNU
19
+//	------------------+--------+-----------+----------
20
+//	Name              |   256B | unlimited | unlimited
21
+//	Linkname          |   100B | unlimited | unlimited
22
+//	Size              | uint33 | unlimited |    uint89
23
+//	Mode              | uint21 |    uint21 |    uint57
24
+//	Uid/Gid           | uint21 | unlimited |    uint57
25
+//	Uname/Gname       |    32B | unlimited |       32B
26
+//	ModTime           | uint33 | unlimited |     int89
27
+//	AccessTime        |    n/a | unlimited |     int89
28
+//	ChangeTime        |    n/a | unlimited |     int89
29
+//	Devmajor/Devminor | uint21 |    uint21 |    uint57
30
+//	------------------+--------+-----------+----------
31
+//	string encoding   |  ASCII |     UTF-8 |    binary
32
+//	sub-second times  |     no |       yes |        no
33
+//	sparse files      |     no |       yes |       yes
34
+//
35
+// The table's upper portion shows the Header fields, where each format reports
36
+// the maximum number of bytes allowed for each string field and
37
+// the integer type used to store each numeric field
38
+// (where timestamps are stored as the number of seconds since the Unix epoch).
39
+//
40
+// The table's lower portion shows specialized features of each format,
41
+// such as supported string encodings, support for sub-second timestamps,
42
+// or support for sparse files.
43
+//
44
+// The Writer currently provides no support for sparse files.
45
+type Format int
46
+
47
+// Constants to identify various tar formats.
48
+const (
49
+	// Deliberately hide the meaning of constants from public API.
50
+	_ Format = (1 << iota) / 4 // Sequence of 0, 0, 1, 2, 4, 8, etc...
51
+
52
+	// FormatUnknown indicates that the format is unknown.
53
+	FormatUnknown
54
+
55
+	// The format of the original Unix V7 tar tool prior to standardization.
56
+	formatV7
57
+
58
+	// FormatUSTAR represents the USTAR header format defined in POSIX.1-1988.
59
+	//
60
+	// While this format is compatible with most tar readers,
61
+	// the format has several limitations making it unsuitable for some usages.
62
+	// Most notably, it cannot support sparse files, files larger than 8GiB,
63
+	// filenames larger than 256 characters, and non-ASCII filenames.
64
+	//
65
+	// Reference:
66
+	//	http://pubs.opengroup.org/onlinepubs/9699919799/utilities/pax.html#tag_20_92_13_06
67
+	FormatUSTAR
68
+
69
+	// FormatPAX represents the PAX header format defined in POSIX.1-2001.
70
+	//
71
+	// PAX extends USTAR by writing a special file with Typeflag TypeXHeader
72
+	// preceding the original header. This file contains a set of key-value
73
+	// records, which are used to overcome USTAR's shortcomings, in addition to
74
+	// providing the ability to have sub-second resolution for timestamps.
75
+	//
76
+	// Some newer formats add their own extensions to PAX by defining their
77
+	// own keys and assigning certain semantic meaning to the associated values.
78
+	// For example, sparse file support in PAX is implemented using keys
79
+	// defined by the GNU manual (e.g., "GNU.sparse.map").
80
+	//
81
+	// Reference:
82
+	//	http://pubs.opengroup.org/onlinepubs/009695399/utilities/pax.html
83
+	FormatPAX
84
+
85
+	// FormatGNU represents the GNU header format.
86
+	//
87
+	// The GNU header format is older than the USTAR and PAX standards and
88
+	// is not compatible with them. The GNU format supports
89
+	// arbitrary file sizes, filenames of arbitrary encoding and length,
90
+	// sparse files, and other features.
91
+	//
92
+	// It is recommended that PAX be chosen over GNU unless the target
93
+	// application can only parse GNU formatted archives.
94
+	//
95
+	// Reference:
96
+	//	http://www.gnu.org/software/tar/manual/html_node/Standard.html
97
+	FormatGNU
98
+
99
+	// Schily's tar format, which is incompatible with USTAR.
100
+	// This does not cover STAR extensions to the PAX format; these fall under
101
+	// the PAX format.
102
+	formatSTAR
103
+
104
+	formatMax
105
+)
106
+
107
+func (f Format) has(f2 Format) bool   { return f&f2 != 0 }
108
+func (f *Format) mayBe(f2 Format)     { *f |= f2 }
109
+func (f *Format) mayOnlyBe(f2 Format) { *f &= f2 }
110
+func (f *Format) mustNotBe(f2 Format) { *f &^= f2 }
111
+
112
+var formatNames = map[Format]string{
113
+	formatV7: "V7", FormatUSTAR: "USTAR", FormatPAX: "PAX", FormatGNU: "GNU", formatSTAR: "STAR",
114
+}
115
+
116
+func (f Format) String() string {
117
+	var ss []string
118
+	for f2 := Format(1); f2 < formatMax; f2 <<= 1 {
119
+		if f.has(f2) {
120
+			ss = append(ss, formatNames[f2])
121
+		}
122
+	}
123
+	switch len(ss) {
124
+	case 0:
125
+		return "<unknown>"
126
+	case 1:
127
+		return ss[0]
128
+	default:
129
+		return "(" + strings.Join(ss, " | ") + ")"
130
+	}
131
+}
132
+
133
+// Magics used to identify various formats.
134
+const (
135
+	magicGNU, versionGNU     = "ustar ", " \x00"
136
+	magicUSTAR, versionUSTAR = "ustar\x00", "00"
137
+	trailerSTAR              = "tar\x00"
138
+)
139
+
140
+// Size constants from various tar specifications.
141
+const (
142
+	blockSize  = 512 // Size of each block in a tar stream
143
+	nameSize   = 100 // Max length of the name field in USTAR format
144
+	prefixSize = 155 // Max length of the prefix field in USTAR format
145
+)
146
+
147
+// blockPadding computes the number of bytes needed to pad offset up to the
148
+// nearest block edge where 0 <= n < blockSize.
149
+func blockPadding(offset int64) (n int64) {
150
+	return -offset & (blockSize - 1)
151
+}
152
+
153
+var zeroBlock block
154
+
155
+type block [blockSize]byte
156
+
157
+// Convert block to any number of formats.
158
+func (b *block) V7() *headerV7       { return (*headerV7)(b) }
159
+func (b *block) GNU() *headerGNU     { return (*headerGNU)(b) }
160
+func (b *block) STAR() *headerSTAR   { return (*headerSTAR)(b) }
161
+func (b *block) USTAR() *headerUSTAR { return (*headerUSTAR)(b) }
162
+func (b *block) Sparse() sparseArray { return (sparseArray)(b[:]) }
163
+
164
+// GetFormat checks that the block is a valid tar header based on the checksum.
165
+// It then attempts to guess the specific format based on magic values.
166
+// If the checksum fails, then FormatUnknown is returned.
167
+func (b *block) GetFormat() Format {
168
+	// Verify checksum.
169
+	var p parser
170
+	value := p.parseOctal(b.V7().Chksum())
171
+	chksum1, chksum2 := b.ComputeChecksum()
172
+	if p.err != nil || (value != chksum1 && value != chksum2) {
173
+		return FormatUnknown
174
+	}
175
+
176
+	// Guess the magic values.
177
+	magic := string(b.USTAR().Magic())
178
+	version := string(b.USTAR().Version())
179
+	trailer := string(b.STAR().Trailer())
180
+	switch {
181
+	case magic == magicUSTAR && trailer == trailerSTAR:
182
+		return formatSTAR
183
+	case magic == magicUSTAR:
184
+		return FormatUSTAR | FormatPAX
185
+	case magic == magicGNU && version == versionGNU:
186
+		return FormatGNU
187
+	default:
188
+		return formatV7
189
+	}
190
+}
191
+
192
+// SetFormat writes the magic values necessary for specified format
193
+// and then updates the checksum accordingly.
194
+func (b *block) SetFormat(format Format) {
195
+	// Set the magic values.
196
+	switch {
197
+	case format.has(formatV7):
198
+		// Do nothing.
199
+	case format.has(FormatGNU):
200
+		copy(b.GNU().Magic(), magicGNU)
201
+		copy(b.GNU().Version(), versionGNU)
202
+	case format.has(formatSTAR):
203
+		copy(b.STAR().Magic(), magicUSTAR)
204
+		copy(b.STAR().Version(), versionUSTAR)
205
+		copy(b.STAR().Trailer(), trailerSTAR)
206
+	case format.has(FormatUSTAR | FormatPAX):
207
+		copy(b.USTAR().Magic(), magicUSTAR)
208
+		copy(b.USTAR().Version(), versionUSTAR)
209
+	default:
210
+		panic("invalid format")
211
+	}
212
+
213
+	// Update checksum.
214
+	// This field is special in that it is terminated by a NULL then space.
215
+	var f formatter
216
+	field := b.V7().Chksum()
217
+	chksum, _ := b.ComputeChecksum() // Possible values are 256..128776
218
+	f.formatOctal(field[:7], chksum) // Never fails since 128776 < 262143
219
+	field[7] = ' '
220
+}
221
+
222
+// ComputeChecksum computes the checksum for the header block.
223
+// POSIX specifies a sum of the unsigned byte values, but the Sun tar used
224
+// signed byte values.
225
+// We compute and return both.
226
+func (b *block) ComputeChecksum() (unsigned, signed int64) {
227
+	for i, c := range b {
228
+		if 148 <= i && i < 156 {
229
+			c = ' ' // Treat the checksum field itself as all spaces.
230
+		}
231
+		unsigned += int64(c)
232
+		signed += int64(int8(c))
233
+	}
234
+	return unsigned, signed
235
+}
236
+
237
+// Reset clears the block with all zeros.
238
+func (b *block) Reset() {
239
+	*b = block{}
240
+}
241
+
242
+type headerV7 [blockSize]byte
243
+
244
+func (h *headerV7) Name() []byte     { return h[000:][:100] }
245
+func (h *headerV7) Mode() []byte     { return h[100:][:8] }
246
+func (h *headerV7) UID() []byte      { return h[108:][:8] }
247
+func (h *headerV7) GID() []byte      { return h[116:][:8] }
248
+func (h *headerV7) Size() []byte     { return h[124:][:12] }
249
+func (h *headerV7) ModTime() []byte  { return h[136:][:12] }
250
+func (h *headerV7) Chksum() []byte   { return h[148:][:8] }
251
+func (h *headerV7) TypeFlag() []byte { return h[156:][:1] }
252
+func (h *headerV7) LinkName() []byte { return h[157:][:100] }
253
+
254
+type headerGNU [blockSize]byte
255
+
256
+func (h *headerGNU) V7() *headerV7       { return (*headerV7)(h) }
257
+func (h *headerGNU) Magic() []byte       { return h[257:][:6] }
258
+func (h *headerGNU) Version() []byte     { return h[263:][:2] }
259
+func (h *headerGNU) UserName() []byte    { return h[265:][:32] }
260
+func (h *headerGNU) GroupName() []byte   { return h[297:][:32] }
261
+func (h *headerGNU) DevMajor() []byte    { return h[329:][:8] }
262
+func (h *headerGNU) DevMinor() []byte    { return h[337:][:8] }
263
+func (h *headerGNU) AccessTime() []byte  { return h[345:][:12] }
264
+func (h *headerGNU) ChangeTime() []byte  { return h[357:][:12] }
265
+func (h *headerGNU) Sparse() sparseArray { return (sparseArray)(h[386:][:24*4+1]) }
266
+func (h *headerGNU) RealSize() []byte    { return h[483:][:12] }
267
+
268
+type headerSTAR [blockSize]byte
269
+
270
+func (h *headerSTAR) V7() *headerV7      { return (*headerV7)(h) }
271
+func (h *headerSTAR) Magic() []byte      { return h[257:][:6] }
272
+func (h *headerSTAR) Version() []byte    { return h[263:][:2] }
273
+func (h *headerSTAR) UserName() []byte   { return h[265:][:32] }
274
+func (h *headerSTAR) GroupName() []byte  { return h[297:][:32] }
275
+func (h *headerSTAR) DevMajor() []byte   { return h[329:][:8] }
276
+func (h *headerSTAR) DevMinor() []byte   { return h[337:][:8] }
277
+func (h *headerSTAR) Prefix() []byte     { return h[345:][:131] }
278
+func (h *headerSTAR) AccessTime() []byte { return h[476:][:12] }
279
+func (h *headerSTAR) ChangeTime() []byte { return h[488:][:12] }
280
+func (h *headerSTAR) Trailer() []byte    { return h[508:][:4] }
281
+
282
+type headerUSTAR [blockSize]byte
283
+
284
+func (h *headerUSTAR) V7() *headerV7     { return (*headerV7)(h) }
285
+func (h *headerUSTAR) Magic() []byte     { return h[257:][:6] }
286
+func (h *headerUSTAR) Version() []byte   { return h[263:][:2] }
287
+func (h *headerUSTAR) UserName() []byte  { return h[265:][:32] }
288
+func (h *headerUSTAR) GroupName() []byte { return h[297:][:32] }
289
+func (h *headerUSTAR) DevMajor() []byte  { return h[329:][:8] }
290
+func (h *headerUSTAR) DevMinor() []byte  { return h[337:][:8] }
291
+func (h *headerUSTAR) Prefix() []byte    { return h[345:][:155] }
292
+
293
+type sparseArray []byte
294
+
295
+func (s sparseArray) Entry(i int) sparseElem { return (sparseElem)(s[i*24:]) }
296
+func (s sparseArray) IsExtended() []byte     { return s[24*s.MaxEntries():][:1] }
297
+func (s sparseArray) MaxEntries() int        { return len(s) / 24 }
298
+
299
+type sparseElem []byte
300
+
301
+func (s sparseElem) Offset() []byte { return s[00:][:12] }
302
+func (s sparseElem) Length() []byte { return s[12:][:12] }
0 303
new file mode 100644
... ...
@@ -0,0 +1,855 @@
0
+// Copyright 2009 The Go Authors. All rights reserved.
1
+// Use of this source code is governed by a BSD-style
2
+// license that can be found in the LICENSE file.
3
+
4
+package tar
5
+
6
+import (
7
+	"bytes"
8
+	"io"
9
+	"io/ioutil"
10
+	"strconv"
11
+	"strings"
12
+	"time"
13
+)
14
+
15
+// Reader provides sequential access to the contents of a tar archive.
16
+// Reader.Next advances to the next file in the archive (including the first),
17
+// and then Reader can be treated as an io.Reader to access the file's data.
18
+type Reader struct {
19
+	r    io.Reader
20
+	pad  int64      // Amount of padding (ignored) after current file entry
21
+	curr fileReader // Reader for current file entry
22
+	blk  block      // Buffer to use as temporary local storage
23
+
24
+	// err is a persistent error.
25
+	// It is only the responsibility of every exported method of Reader to
26
+	// ensure that this error is sticky.
27
+	err error
28
+}
29
+
30
+type fileReader interface {
31
+	io.Reader
32
+	fileState
33
+
34
+	WriteTo(io.Writer) (int64, error)
35
+}
36
+
37
+// NewReader creates a new Reader reading from r.
38
+func NewReader(r io.Reader) *Reader {
39
+	return &Reader{r: r, curr: &regFileReader{r, 0}}
40
+}
41
+
42
+// Next advances to the next entry in the tar archive.
43
+// The Header.Size determines how many bytes can be read for the next file.
44
+// Any remaining data in the current file is automatically discarded.
45
+//
46
+// io.EOF is returned at the end of the input.
47
+func (tr *Reader) Next() (*Header, error) {
48
+	if tr.err != nil {
49
+		return nil, tr.err
50
+	}
51
+	hdr, err := tr.next()
52
+	tr.err = err
53
+	return hdr, err
54
+}
55
+
56
+func (tr *Reader) next() (*Header, error) {
57
+	var paxHdrs map[string]string
58
+	var gnuLongName, gnuLongLink string
59
+
60
+	// Externally, Next iterates through the tar archive as if it is a series of
61
+	// files. Internally, the tar format often uses fake "files" to add meta
62
+	// data that describes the next file. These meta data "files" should not
63
+	// normally be visible to the outside. As such, this loop iterates through
64
+	// one or more "header files" until it finds a "normal file".
65
+	format := FormatUSTAR | FormatPAX | FormatGNU
66
+loop:
67
+	for {
68
+		// Discard the remainder of the file and any padding.
69
+		if err := discard(tr.r, tr.curr.PhysicalRemaining()); err != nil {
70
+			return nil, err
71
+		}
72
+		if _, err := tryReadFull(tr.r, tr.blk[:tr.pad]); err != nil {
73
+			return nil, err
74
+		}
75
+		tr.pad = 0
76
+
77
+		hdr, rawHdr, err := tr.readHeader()
78
+		if err != nil {
79
+			return nil, err
80
+		}
81
+		if err := tr.handleRegularFile(hdr); err != nil {
82
+			return nil, err
83
+		}
84
+		format.mayOnlyBe(hdr.Format)
85
+
86
+		// Check for PAX/GNU special headers and files.
87
+		switch hdr.Typeflag {
88
+		case TypeXHeader, TypeXGlobalHeader:
89
+			format.mayOnlyBe(FormatPAX)
90
+			paxHdrs, err = parsePAX(tr)
91
+			if err != nil {
92
+				return nil, err
93
+			}
94
+			if hdr.Typeflag == TypeXGlobalHeader {
95
+				mergePAX(hdr, paxHdrs)
96
+				return &Header{
97
+					Name:       hdr.Name,
98
+					Typeflag:   hdr.Typeflag,
99
+					Xattrs:     hdr.Xattrs,
100
+					PAXRecords: hdr.PAXRecords,
101
+					Format:     format,
102
+				}, nil
103
+			}
104
+			continue loop // This is a meta header affecting the next header
105
+		case TypeGNULongName, TypeGNULongLink:
106
+			format.mayOnlyBe(FormatGNU)
107
+			realname, err := ioutil.ReadAll(tr)
108
+			if err != nil {
109
+				return nil, err
110
+			}
111
+
112
+			var p parser
113
+			switch hdr.Typeflag {
114
+			case TypeGNULongName:
115
+				gnuLongName = p.parseString(realname)
116
+			case TypeGNULongLink:
117
+				gnuLongLink = p.parseString(realname)
118
+			}
119
+			continue loop // This is a meta header affecting the next header
120
+		default:
121
+			// The old GNU sparse format is handled here since it is technically
122
+			// just a regular file with additional attributes.
123
+
124
+			if err := mergePAX(hdr, paxHdrs); err != nil {
125
+				return nil, err
126
+			}
127
+			if gnuLongName != "" {
128
+				hdr.Name = gnuLongName
129
+			}
130
+			if gnuLongLink != "" {
131
+				hdr.Linkname = gnuLongLink
132
+			}
133
+			if hdr.Typeflag == TypeRegA && strings.HasSuffix(hdr.Name, "/") {
134
+				hdr.Typeflag = TypeDir // Legacy archives use trailing slash for directories
135
+			}
136
+
137
+			// The extended headers may have updated the size.
138
+			// Thus, setup the regFileReader again after merging PAX headers.
139
+			if err := tr.handleRegularFile(hdr); err != nil {
140
+				return nil, err
141
+			}
142
+
143
+			// Sparse formats rely on being able to read from the logical data
144
+			// section; there must be a preceding call to handleRegularFile.
145
+			if err := tr.handleSparseFile(hdr, rawHdr); err != nil {
146
+				return nil, err
147
+			}
148
+
149
+			// Set the final guess at the format.
150
+			if format.has(FormatUSTAR) && format.has(FormatPAX) {
151
+				format.mayOnlyBe(FormatUSTAR)
152
+			}
153
+			hdr.Format = format
154
+			return hdr, nil // This is a file, so stop
155
+		}
156
+	}
157
+}
158
+
159
+// handleRegularFile sets up the current file reader and padding such that it
160
+// can only read the following logical data section. It will properly handle
161
+// special headers that contain no data section.
162
+func (tr *Reader) handleRegularFile(hdr *Header) error {
163
+	nb := hdr.Size
164
+	if isHeaderOnlyType(hdr.Typeflag) {
165
+		nb = 0
166
+	}
167
+	if nb < 0 {
168
+		return ErrHeader
169
+	}
170
+
171
+	tr.pad = blockPadding(nb)
172
+	tr.curr = &regFileReader{r: tr.r, nb: nb}
173
+	return nil
174
+}
175
+
176
+// handleSparseFile checks if the current file is a sparse format of any type
177
+// and sets the curr reader appropriately.
178
+func (tr *Reader) handleSparseFile(hdr *Header, rawHdr *block) error {
179
+	var spd sparseDatas
180
+	var err error
181
+	if hdr.Typeflag == TypeGNUSparse {
182
+		spd, err = tr.readOldGNUSparseMap(hdr, rawHdr)
183
+	} else {
184
+		spd, err = tr.readGNUSparsePAXHeaders(hdr)
185
+	}
186
+
187
+	// If sp is non-nil, then this is a sparse file.
188
+	// Note that it is possible for len(sp) == 0.
189
+	if err == nil && spd != nil {
190
+		if isHeaderOnlyType(hdr.Typeflag) || !validateSparseEntries(spd, hdr.Size) {
191
+			return ErrHeader
192
+		}
193
+		sph := invertSparseEntries(spd, hdr.Size)
194
+		tr.curr = &sparseFileReader{tr.curr, sph, 0}
195
+	}
196
+	return err
197
+}
198
+
199
+// readGNUSparsePAXHeaders checks the PAX headers for GNU sparse headers.
200
+// If they are found, then this function reads the sparse map and returns it.
201
+// This assumes that 0.0 headers have already been converted to 0.1 headers
202
+// by the the PAX header parsing logic.
203
+func (tr *Reader) readGNUSparsePAXHeaders(hdr *Header) (sparseDatas, error) {
204
+	// Identify the version of GNU headers.
205
+	var is1x0 bool
206
+	major, minor := hdr.PAXRecords[paxGNUSparseMajor], hdr.PAXRecords[paxGNUSparseMinor]
207
+	switch {
208
+	case major == "0" && (minor == "0" || minor == "1"):
209
+		is1x0 = false
210
+	case major == "1" && minor == "0":
211
+		is1x0 = true
212
+	case major != "" || minor != "":
213
+		return nil, nil // Unknown GNU sparse PAX version
214
+	case hdr.PAXRecords[paxGNUSparseMap] != "":
215
+		is1x0 = false // 0.0 and 0.1 did not have explicit version records, so guess
216
+	default:
217
+		return nil, nil // Not a PAX format GNU sparse file.
218
+	}
219
+	hdr.Format.mayOnlyBe(FormatPAX)
220
+
221
+	// Update hdr from GNU sparse PAX headers.
222
+	if name := hdr.PAXRecords[paxGNUSparseName]; name != "" {
223
+		hdr.Name = name
224
+	}
225
+	size := hdr.PAXRecords[paxGNUSparseSize]
226
+	if size == "" {
227
+		size = hdr.PAXRecords[paxGNUSparseRealSize]
228
+	}
229
+	if size != "" {
230
+		n, err := strconv.ParseInt(size, 10, 64)
231
+		if err != nil {
232
+			return nil, ErrHeader
233
+		}
234
+		hdr.Size = n
235
+	}
236
+
237
+	// Read the sparse map according to the appropriate format.
238
+	if is1x0 {
239
+		return readGNUSparseMap1x0(tr.curr)
240
+	}
241
+	return readGNUSparseMap0x1(hdr.PAXRecords)
242
+}
243
+
244
+// mergePAX merges paxHdrs into hdr for all relevant fields of Header.
245
+func mergePAX(hdr *Header, paxHdrs map[string]string) (err error) {
246
+	for k, v := range paxHdrs {
247
+		if v == "" {
248
+			continue // Keep the original USTAR value
249
+		}
250
+		var id64 int64
251
+		switch k {
252
+		case paxPath:
253
+			hdr.Name = v
254
+		case paxLinkpath:
255
+			hdr.Linkname = v
256
+		case paxUname:
257
+			hdr.Uname = v
258
+		case paxGname:
259
+			hdr.Gname = v
260
+		case paxUid:
261
+			id64, err = strconv.ParseInt(v, 10, 64)
262
+			hdr.Uid = int(id64) // Integer overflow possible
263
+		case paxGid:
264
+			id64, err = strconv.ParseInt(v, 10, 64)
265
+			hdr.Gid = int(id64) // Integer overflow possible
266
+		case paxAtime:
267
+			hdr.AccessTime, err = parsePAXTime(v)
268
+		case paxMtime:
269
+			hdr.ModTime, err = parsePAXTime(v)
270
+		case paxCtime:
271
+			hdr.ChangeTime, err = parsePAXTime(v)
272
+		case paxSize:
273
+			hdr.Size, err = strconv.ParseInt(v, 10, 64)
274
+		default:
275
+			if strings.HasPrefix(k, paxSchilyXattr) {
276
+				if hdr.Xattrs == nil {
277
+					hdr.Xattrs = make(map[string]string)
278
+				}
279
+				hdr.Xattrs[k[len(paxSchilyXattr):]] = v
280
+			}
281
+		}
282
+		if err != nil {
283
+			return ErrHeader
284
+		}
285
+	}
286
+	hdr.PAXRecords = paxHdrs
287
+	return nil
288
+}
289
+
290
+// parsePAX parses PAX headers.
291
+// If an extended header (type 'x') is invalid, ErrHeader is returned
292
+func parsePAX(r io.Reader) (map[string]string, error) {
293
+	buf, err := ioutil.ReadAll(r)
294
+	if err != nil {
295
+		return nil, err
296
+	}
297
+	sbuf := string(buf)
298
+
299
+	// For GNU PAX sparse format 0.0 support.
300
+	// This function transforms the sparse format 0.0 headers into format 0.1
301
+	// headers since 0.0 headers were not PAX compliant.
302
+	var sparseMap []string
303
+
304
+	paxHdrs := make(map[string]string)
305
+	for len(sbuf) > 0 {
306
+		key, value, residual, err := parsePAXRecord(sbuf)
307
+		if err != nil {
308
+			return nil, ErrHeader
309
+		}
310
+		sbuf = residual
311
+
312
+		switch key {
313
+		case paxGNUSparseOffset, paxGNUSparseNumBytes:
314
+			// Validate sparse header order and value.
315
+			if (len(sparseMap)%2 == 0 && key != paxGNUSparseOffset) ||
316
+				(len(sparseMap)%2 == 1 && key != paxGNUSparseNumBytes) ||
317
+				strings.Contains(value, ",") {
318
+				return nil, ErrHeader
319
+			}
320
+			sparseMap = append(sparseMap, value)
321
+		default:
322
+			paxHdrs[key] = value
323
+		}
324
+	}
325
+	if len(sparseMap) > 0 {
326
+		paxHdrs[paxGNUSparseMap] = strings.Join(sparseMap, ",")
327
+	}
328
+	return paxHdrs, nil
329
+}
330
+
331
+// readHeader reads the next block header and assumes that the underlying reader
332
+// is already aligned to a block boundary. It returns the raw block of the
333
+// header in case further processing is required.
334
+//
335
+// The err will be set to io.EOF only when one of the following occurs:
336
+//	* Exactly 0 bytes are read and EOF is hit.
337
+//	* Exactly 1 block of zeros is read and EOF is hit.
338
+//	* At least 2 blocks of zeros are read.
339
+func (tr *Reader) readHeader() (*Header, *block, error) {
340
+	// Two blocks of zero bytes marks the end of the archive.
341
+	if _, err := io.ReadFull(tr.r, tr.blk[:]); err != nil {
342
+		return nil, nil, err // EOF is okay here; exactly 0 bytes read
343
+	}
344
+	if bytes.Equal(tr.blk[:], zeroBlock[:]) {
345
+		if _, err := io.ReadFull(tr.r, tr.blk[:]); err != nil {
346
+			return nil, nil, err // EOF is okay here; exactly 1 block of zeros read
347
+		}
348
+		if bytes.Equal(tr.blk[:], zeroBlock[:]) {
349
+			return nil, nil, io.EOF // normal EOF; exactly 2 block of zeros read
350
+		}
351
+		return nil, nil, ErrHeader // Zero block and then non-zero block
352
+	}
353
+
354
+	// Verify the header matches a known format.
355
+	format := tr.blk.GetFormat()
356
+	if format == FormatUnknown {
357
+		return nil, nil, ErrHeader
358
+	}
359
+
360
+	var p parser
361
+	hdr := new(Header)
362
+
363
+	// Unpack the V7 header.
364
+	v7 := tr.blk.V7()
365
+	hdr.Typeflag = v7.TypeFlag()[0]
366
+	hdr.Name = p.parseString(v7.Name())
367
+	hdr.Linkname = p.parseString(v7.LinkName())
368
+	hdr.Size = p.parseNumeric(v7.Size())
369
+	hdr.Mode = p.parseNumeric(v7.Mode())
370
+	hdr.Uid = int(p.parseNumeric(v7.UID()))
371
+	hdr.Gid = int(p.parseNumeric(v7.GID()))
372
+	hdr.ModTime = time.Unix(p.parseNumeric(v7.ModTime()), 0)
373
+
374
+	// Unpack format specific fields.
375
+	if format > formatV7 {
376
+		ustar := tr.blk.USTAR()
377
+		hdr.Uname = p.parseString(ustar.UserName())
378
+		hdr.Gname = p.parseString(ustar.GroupName())
379
+		hdr.Devmajor = p.parseNumeric(ustar.DevMajor())
380
+		hdr.Devminor = p.parseNumeric(ustar.DevMinor())
381
+
382
+		var prefix string
383
+		switch {
384
+		case format.has(FormatUSTAR | FormatPAX):
385
+			hdr.Format = format
386
+			ustar := tr.blk.USTAR()
387
+			prefix = p.parseString(ustar.Prefix())
388
+
389
+			// For Format detection, check if block is properly formatted since
390
+			// the parser is more liberal than what USTAR actually permits.
391
+			notASCII := func(r rune) bool { return r >= 0x80 }
392
+			if bytes.IndexFunc(tr.blk[:], notASCII) >= 0 {
393
+				hdr.Format = FormatUnknown // Non-ASCII characters in block.
394
+			}
395
+			nul := func(b []byte) bool { return int(b[len(b)-1]) == 0 }
396
+			if !(nul(v7.Size()) && nul(v7.Mode()) && nul(v7.UID()) && nul(v7.GID()) &&
397
+				nul(v7.ModTime()) && nul(ustar.DevMajor()) && nul(ustar.DevMinor())) {
398
+				hdr.Format = FormatUnknown // Numeric fields must end in NUL
399
+			}
400
+		case format.has(formatSTAR):
401
+			star := tr.blk.STAR()
402
+			prefix = p.parseString(star.Prefix())
403
+			hdr.AccessTime = time.Unix(p.parseNumeric(star.AccessTime()), 0)
404
+			hdr.ChangeTime = time.Unix(p.parseNumeric(star.ChangeTime()), 0)
405
+		case format.has(FormatGNU):
406
+			hdr.Format = format
407
+			var p2 parser
408
+			gnu := tr.blk.GNU()
409
+			if b := gnu.AccessTime(); b[0] != 0 {
410
+				hdr.AccessTime = time.Unix(p2.parseNumeric(b), 0)
411
+			}
412
+			if b := gnu.ChangeTime(); b[0] != 0 {
413
+				hdr.ChangeTime = time.Unix(p2.parseNumeric(b), 0)
414
+			}
415
+
416
+			// Prior to Go1.8, the Writer had a bug where it would output
417
+			// an invalid tar file in certain rare situations because the logic
418
+			// incorrectly believed that the old GNU format had a prefix field.
419
+			// This is wrong and leads to an output file that mangles the
420
+			// atime and ctime fields, which are often left unused.
421
+			//
422
+			// In order to continue reading tar files created by former, buggy
423
+			// versions of Go, we skeptically parse the atime and ctime fields.
424
+			// If we are unable to parse them and the prefix field looks like
425
+			// an ASCII string, then we fallback on the pre-Go1.8 behavior
426
+			// of treating these fields as the USTAR prefix field.
427
+			//
428
+			// Note that this will not use the fallback logic for all possible
429
+			// files generated by a pre-Go1.8 toolchain. If the generated file
430
+			// happened to have a prefix field that parses as valid
431
+			// atime and ctime fields (e.g., when they are valid octal strings),
432
+			// then it is impossible to distinguish between an valid GNU file
433
+			// and an invalid pre-Go1.8 file.
434
+			//
435
+			// See https://golang.org/issues/12594
436
+			// See https://golang.org/issues/21005
437
+			if p2.err != nil {
438
+				hdr.AccessTime, hdr.ChangeTime = time.Time{}, time.Time{}
439
+				ustar := tr.blk.USTAR()
440
+				if s := p.parseString(ustar.Prefix()); isASCII(s) {
441
+					prefix = s
442
+				}
443
+				hdr.Format = FormatUnknown // Buggy file is not GNU
444
+			}
445
+		}
446
+		if len(prefix) > 0 {
447
+			hdr.Name = prefix + "/" + hdr.Name
448
+		}
449
+	}
450
+	return hdr, &tr.blk, p.err
451
+}
452
+
453
+// readOldGNUSparseMap reads the sparse map from the old GNU sparse format.
454
+// The sparse map is stored in the tar header if it's small enough.
455
+// If it's larger than four entries, then one or more extension headers are used
456
+// to store the rest of the sparse map.
457
+//
458
+// The Header.Size does not reflect the size of any extended headers used.
459
+// Thus, this function will read from the raw io.Reader to fetch extra headers.
460
+// This method mutates blk in the process.
461
+func (tr *Reader) readOldGNUSparseMap(hdr *Header, blk *block) (sparseDatas, error) {
462
+	// Make sure that the input format is GNU.
463
+	// Unfortunately, the STAR format also has a sparse header format that uses
464
+	// the same type flag but has a completely different layout.
465
+	if blk.GetFormat() != FormatGNU {
466
+		return nil, ErrHeader
467
+	}
468
+	hdr.Format.mayOnlyBe(FormatGNU)
469
+
470
+	var p parser
471
+	hdr.Size = p.parseNumeric(blk.GNU().RealSize())
472
+	if p.err != nil {
473
+		return nil, p.err
474
+	}
475
+	s := blk.GNU().Sparse()
476
+	spd := make(sparseDatas, 0, s.MaxEntries())
477
+	for {
478
+		for i := 0; i < s.MaxEntries(); i++ {
479
+			// This termination condition is identical to GNU and BSD tar.
480
+			if s.Entry(i).Offset()[0] == 0x00 {
481
+				break // Don't return, need to process extended headers (even if empty)
482
+			}
483
+			offset := p.parseNumeric(s.Entry(i).Offset())
484
+			length := p.parseNumeric(s.Entry(i).Length())
485
+			if p.err != nil {
486
+				return nil, p.err
487
+			}
488
+			spd = append(spd, sparseEntry{Offset: offset, Length: length})
489
+		}
490
+
491
+		if s.IsExtended()[0] > 0 {
492
+			// There are more entries. Read an extension header and parse its entries.
493
+			if _, err := mustReadFull(tr.r, blk[:]); err != nil {
494
+				return nil, err
495
+			}
496
+			s = blk.Sparse()
497
+			continue
498
+		}
499
+		return spd, nil // Done
500
+	}
501
+}
502
+
503
+// readGNUSparseMap1x0 reads the sparse map as stored in GNU's PAX sparse format
504
+// version 1.0. The format of the sparse map consists of a series of
505
+// newline-terminated numeric fields. The first field is the number of entries
506
+// and is always present. Following this are the entries, consisting of two
507
+// fields (offset, length). This function must stop reading at the end
508
+// boundary of the block containing the last newline.
509
+//
510
+// Note that the GNU manual says that numeric values should be encoded in octal
511
+// format. However, the GNU tar utility itself outputs these values in decimal.
512
+// As such, this library treats values as being encoded in decimal.
513
+func readGNUSparseMap1x0(r io.Reader) (sparseDatas, error) {
514
+	var (
515
+		cntNewline int64
516
+		buf        bytes.Buffer
517
+		blk        block
518
+	)
519
+
520
+	// feedTokens copies data in blocks from r into buf until there are
521
+	// at least cnt newlines in buf. It will not read more blocks than needed.
522
+	feedTokens := func(n int64) error {
523
+		for cntNewline < n {
524
+			if _, err := mustReadFull(r, blk[:]); err != nil {
525
+				return err
526
+			}
527
+			buf.Write(blk[:])
528
+			for _, c := range blk {
529
+				if c == '\n' {
530
+					cntNewline++
531
+				}
532
+			}
533
+		}
534
+		return nil
535
+	}
536
+
537
+	// nextToken gets the next token delimited by a newline. This assumes that
538
+	// at least one newline exists in the buffer.
539
+	nextToken := func() string {
540
+		cntNewline--
541
+		tok, _ := buf.ReadString('\n')
542
+		return strings.TrimRight(tok, "\n")
543
+	}
544
+
545
+	// Parse for the number of entries.
546
+	// Use integer overflow resistant math to check this.
547
+	if err := feedTokens(1); err != nil {
548
+		return nil, err
549
+	}
550
+	numEntries, err := strconv.ParseInt(nextToken(), 10, 0) // Intentionally parse as native int
551
+	if err != nil || numEntries < 0 || int(2*numEntries) < int(numEntries) {
552
+		return nil, ErrHeader
553
+	}
554
+
555
+	// Parse for all member entries.
556
+	// numEntries is trusted after this since a potential attacker must have
557
+	// committed resources proportional to what this library used.
558
+	if err := feedTokens(2 * numEntries); err != nil {
559
+		return nil, err
560
+	}
561
+	spd := make(sparseDatas, 0, numEntries)
562
+	for i := int64(0); i < numEntries; i++ {
563
+		offset, err1 := strconv.ParseInt(nextToken(), 10, 64)
564
+		length, err2 := strconv.ParseInt(nextToken(), 10, 64)
565
+		if err1 != nil || err2 != nil {
566
+			return nil, ErrHeader
567
+		}
568
+		spd = append(spd, sparseEntry{Offset: offset, Length: length})
569
+	}
570
+	return spd, nil
571
+}
572
+
573
+// readGNUSparseMap0x1 reads the sparse map as stored in GNU's PAX sparse format
574
+// version 0.1. The sparse map is stored in the PAX headers.
575
+func readGNUSparseMap0x1(paxHdrs map[string]string) (sparseDatas, error) {
576
+	// Get number of entries.
577
+	// Use integer overflow resistant math to check this.
578
+	numEntriesStr := paxHdrs[paxGNUSparseNumBlocks]
579
+	numEntries, err := strconv.ParseInt(numEntriesStr, 10, 0) // Intentionally parse as native int
580
+	if err != nil || numEntries < 0 || int(2*numEntries) < int(numEntries) {
581
+		return nil, ErrHeader
582
+	}
583
+
584
+	// There should be two numbers in sparseMap for each entry.
585
+	sparseMap := strings.Split(paxHdrs[paxGNUSparseMap], ",")
586
+	if len(sparseMap) == 1 && sparseMap[0] == "" {
587
+		sparseMap = sparseMap[:0]
588
+	}
589
+	if int64(len(sparseMap)) != 2*numEntries {
590
+		return nil, ErrHeader
591
+	}
592
+
593
+	// Loop through the entries in the sparse map.
594
+	// numEntries is trusted now.
595
+	spd := make(sparseDatas, 0, numEntries)
596
+	for len(sparseMap) >= 2 {
597
+		offset, err1 := strconv.ParseInt(sparseMap[0], 10, 64)
598
+		length, err2 := strconv.ParseInt(sparseMap[1], 10, 64)
599
+		if err1 != nil || err2 != nil {
600
+			return nil, ErrHeader
601
+		}
602
+		spd = append(spd, sparseEntry{Offset: offset, Length: length})
603
+		sparseMap = sparseMap[2:]
604
+	}
605
+	return spd, nil
606
+}
607
+
608
+// Read reads from the current file in the tar archive.
609
+// It returns (0, io.EOF) when it reaches the end of that file,
610
+// until Next is called to advance to the next file.
611
+//
612
+// If the current file is sparse, then the regions marked as a hole
613
+// are read back as NUL-bytes.
614
+//
615
+// Calling Read on special types like TypeLink, TypeSymlink, TypeChar,
616
+// TypeBlock, TypeDir, and TypeFifo returns (0, io.EOF) regardless of what
617
+// the Header.Size claims.
618
+func (tr *Reader) Read(b []byte) (int, error) {
619
+	if tr.err != nil {
620
+		return 0, tr.err
621
+	}
622
+	n, err := tr.curr.Read(b)
623
+	if err != nil && err != io.EOF {
624
+		tr.err = err
625
+	}
626
+	return n, err
627
+}
628
+
629
+// writeTo writes the content of the current file to w.
630
+// The bytes written matches the number of remaining bytes in the current file.
631
+//
632
+// If the current file is sparse and w is an io.WriteSeeker,
633
+// then writeTo uses Seek to skip past holes defined in Header.SparseHoles,
634
+// assuming that skipped regions are filled with NULs.
635
+// This always writes the last byte to ensure w is the right size.
636
+//
637
+// TODO(dsnet): Re-export this when adding sparse file support.
638
+// See https://golang.org/issue/22735
639
+func (tr *Reader) writeTo(w io.Writer) (int64, error) {
640
+	if tr.err != nil {
641
+		return 0, tr.err
642
+	}
643
+	n, err := tr.curr.WriteTo(w)
644
+	if err != nil {
645
+		tr.err = err
646
+	}
647
+	return n, err
648
+}
649
+
650
+// regFileReader is a fileReader for reading data from a regular file entry.
651
+type regFileReader struct {
652
+	r  io.Reader // Underlying Reader
653
+	nb int64     // Number of remaining bytes to read
654
+}
655
+
656
+func (fr *regFileReader) Read(b []byte) (n int, err error) {
657
+	if int64(len(b)) > fr.nb {
658
+		b = b[:fr.nb]
659
+	}
660
+	if len(b) > 0 {
661
+		n, err = fr.r.Read(b)
662
+		fr.nb -= int64(n)
663
+	}
664
+	switch {
665
+	case err == io.EOF && fr.nb > 0:
666
+		return n, io.ErrUnexpectedEOF
667
+	case err == nil && fr.nb == 0:
668
+		return n, io.EOF
669
+	default:
670
+		return n, err
671
+	}
672
+}
673
+
674
+func (fr *regFileReader) WriteTo(w io.Writer) (int64, error) {
675
+	return io.Copy(w, struct{ io.Reader }{fr})
676
+}
677
+
678
+func (fr regFileReader) LogicalRemaining() int64 {
679
+	return fr.nb
680
+}
681
+
682
+func (fr regFileReader) PhysicalRemaining() int64 {
683
+	return fr.nb
684
+}
685
+
686
+// sparseFileReader is a fileReader for reading data from a sparse file entry.
687
+type sparseFileReader struct {
688
+	fr  fileReader  // Underlying fileReader
689
+	sp  sparseHoles // Normalized list of sparse holes
690
+	pos int64       // Current position in sparse file
691
+}
692
+
693
+func (sr *sparseFileReader) Read(b []byte) (n int, err error) {
694
+	finished := int64(len(b)) >= sr.LogicalRemaining()
695
+	if finished {
696
+		b = b[:sr.LogicalRemaining()]
697
+	}
698
+
699
+	b0 := b
700
+	endPos := sr.pos + int64(len(b))
701
+	for endPos > sr.pos && err == nil {
702
+		var nf int // Bytes read in fragment
703
+		holeStart, holeEnd := sr.sp[0].Offset, sr.sp[0].endOffset()
704
+		if sr.pos < holeStart { // In a data fragment
705
+			bf := b[:min(int64(len(b)), holeStart-sr.pos)]
706
+			nf, err = tryReadFull(sr.fr, bf)
707
+		} else { // In a hole fragment
708
+			bf := b[:min(int64(len(b)), holeEnd-sr.pos)]
709
+			nf, err = tryReadFull(zeroReader{}, bf)
710
+		}
711
+		b = b[nf:]
712
+		sr.pos += int64(nf)
713
+		if sr.pos >= holeEnd && len(sr.sp) > 1 {
714
+			sr.sp = sr.sp[1:] // Ensure last fragment always remains
715
+		}
716
+	}
717
+
718
+	n = len(b0) - len(b)
719
+	switch {
720
+	case err == io.EOF:
721
+		return n, errMissData // Less data in dense file than sparse file
722
+	case err != nil:
723
+		return n, err
724
+	case sr.LogicalRemaining() == 0 && sr.PhysicalRemaining() > 0:
725
+		return n, errUnrefData // More data in dense file than sparse file
726
+	case finished:
727
+		return n, io.EOF
728
+	default:
729
+		return n, nil
730
+	}
731
+}
732
+
733
+func (sr *sparseFileReader) WriteTo(w io.Writer) (n int64, err error) {
734
+	ws, ok := w.(io.WriteSeeker)
735
+	if ok {
736
+		if _, err := ws.Seek(0, io.SeekCurrent); err != nil {
737
+			ok = false // Not all io.Seeker can really seek
738
+		}
739
+	}
740
+	if !ok {
741
+		return io.Copy(w, struct{ io.Reader }{sr})
742
+	}
743
+
744
+	var writeLastByte bool
745
+	pos0 := sr.pos
746
+	for sr.LogicalRemaining() > 0 && !writeLastByte && err == nil {
747
+		var nf int64 // Size of fragment
748
+		holeStart, holeEnd := sr.sp[0].Offset, sr.sp[0].endOffset()
749
+		if sr.pos < holeStart { // In a data fragment
750
+			nf = holeStart - sr.pos
751
+			nf, err = io.CopyN(ws, sr.fr, nf)
752
+		} else { // In a hole fragment
753
+			nf = holeEnd - sr.pos
754
+			if sr.PhysicalRemaining() == 0 {
755
+				writeLastByte = true
756
+				nf--
757
+			}
758
+			_, err = ws.Seek(nf, io.SeekCurrent)
759
+		}
760
+		sr.pos += nf
761
+		if sr.pos >= holeEnd && len(sr.sp) > 1 {
762
+			sr.sp = sr.sp[1:] // Ensure last fragment always remains
763
+		}
764
+	}
765
+
766
+	// If the last fragment is a hole, then seek to 1-byte before EOF, and
767
+	// write a single byte to ensure the file is the right size.
768
+	if writeLastByte && err == nil {
769
+		_, err = ws.Write([]byte{0})
770
+		sr.pos++
771
+	}
772
+
773
+	n = sr.pos - pos0
774
+	switch {
775
+	case err == io.EOF:
776
+		return n, errMissData // Less data in dense file than sparse file
777
+	case err != nil:
778
+		return n, err
779
+	case sr.LogicalRemaining() == 0 && sr.PhysicalRemaining() > 0:
780
+		return n, errUnrefData // More data in dense file than sparse file
781
+	default:
782
+		return n, nil
783
+	}
784
+}
785
+
786
+func (sr sparseFileReader) LogicalRemaining() int64 {
787
+	return sr.sp[len(sr.sp)-1].endOffset() - sr.pos
788
+}
789
+func (sr sparseFileReader) PhysicalRemaining() int64 {
790
+	return sr.fr.PhysicalRemaining()
791
+}
792
+
793
+type zeroReader struct{}
794
+
795
+func (zeroReader) Read(b []byte) (int, error) {
796
+	for i := range b {
797
+		b[i] = 0
798
+	}
799
+	return len(b), nil
800
+}
801
+
802
+// mustReadFull is like io.ReadFull except it returns
803
+// io.ErrUnexpectedEOF when io.EOF is hit before len(b) bytes are read.
804
+func mustReadFull(r io.Reader, b []byte) (int, error) {
805
+	n, err := tryReadFull(r, b)
806
+	if err == io.EOF {
807
+		err = io.ErrUnexpectedEOF
808
+	}
809
+	return n, err
810
+}
811
+
812
+// tryReadFull is like io.ReadFull except it returns
813
+// io.EOF when it is hit before len(b) bytes are read.
814
+func tryReadFull(r io.Reader, b []byte) (n int, err error) {
815
+	for len(b) > n && err == nil {
816
+		var nn int
817
+		nn, err = r.Read(b[n:])
818
+		n += nn
819
+	}
820
+	if len(b) == n && err == io.EOF {
821
+		err = nil
822
+	}
823
+	return n, err
824
+}
825
+
826
+// discard skips n bytes in r, reporting an error if unable to do so.
827
+func discard(r io.Reader, n int64) error {
828
+	// If possible, Seek to the last byte before the end of the data section.
829
+	// Do this because Seek is often lazy about reporting errors; this will mask
830
+	// the fact that the stream may be truncated. We can rely on the
831
+	// io.CopyN done shortly afterwards to trigger any IO errors.
832
+	var seekSkipped int64 // Number of bytes skipped via Seek
833
+	if sr, ok := r.(io.Seeker); ok && n > 1 {
834
+		// Not all io.Seeker can actually Seek. For example, os.Stdin implements
835
+		// io.Seeker, but calling Seek always returns an error and performs
836
+		// no action. Thus, we try an innocent seek to the current position
837
+		// to see if Seek is really supported.
838
+		pos1, err := sr.Seek(0, io.SeekCurrent)
839
+		if pos1 >= 0 && err == nil {
840
+			// Seek seems supported, so perform the real Seek.
841
+			pos2, err := sr.Seek(n-1, io.SeekCurrent)
842
+			if pos2 < 0 || err != nil {
843
+				return err
844
+			}
845
+			seekSkipped = pos2 - pos1
846
+		}
847
+	}
848
+
849
+	copySkipped, err := io.CopyN(ioutil.Discard, r, n-seekSkipped)
850
+	if err == io.EOF && seekSkipped+copySkipped < n {
851
+		err = io.ErrUnexpectedEOF
852
+	}
853
+	return err
854
+}
0 855
new file mode 100644
... ...
@@ -0,0 +1,20 @@
0
+// Copyright 2012 The Go Authors. All rights reserved.
1
+// Use of this source code is governed by a BSD-style
2
+// license that can be found in the LICENSE file.
3
+
4
+// +build linux dragonfly openbsd solaris
5
+
6
+package tar
7
+
8
+import (
9
+	"syscall"
10
+	"time"
11
+)
12
+
13
+func statAtime(st *syscall.Stat_t) time.Time {
14
+	return time.Unix(st.Atim.Unix())
15
+}
16
+
17
+func statCtime(st *syscall.Stat_t) time.Time {
18
+	return time.Unix(st.Ctim.Unix())
19
+}
0 20
new file mode 100644
... ...
@@ -0,0 +1,20 @@
0
+// Copyright 2012 The Go Authors. All rights reserved.
1
+// Use of this source code is governed by a BSD-style
2
+// license that can be found in the LICENSE file.
3
+
4
+// +build darwin freebsd netbsd
5
+
6
+package tar
7
+
8
+import (
9
+	"syscall"
10
+	"time"
11
+)
12
+
13
+func statAtime(st *syscall.Stat_t) time.Time {
14
+	return time.Unix(st.Atimespec.Unix())
15
+}
16
+
17
+func statCtime(st *syscall.Stat_t) time.Time {
18
+	return time.Unix(st.Ctimespec.Unix())
19
+}
0 20
new file mode 100644
... ...
@@ -0,0 +1,76 @@
0
+// Copyright 2012 The Go Authors. All rights reserved.
1
+// Use of this source code is governed by a BSD-style
2
+// license that can be found in the LICENSE file.
3
+
4
+// +build linux darwin dragonfly freebsd openbsd netbsd solaris
5
+
6
+package tar
7
+
8
+import (
9
+	"os"
10
+	"runtime"
11
+	"syscall"
12
+)
13
+
14
+func init() {
15
+	sysStat = statUnix
16
+}
17
+
18
+func statUnix(fi os.FileInfo, h *Header) error {
19
+	sys, ok := fi.Sys().(*syscall.Stat_t)
20
+	if !ok {
21
+		return nil
22
+	}
23
+	h.Uid = int(sys.Uid)
24
+	h.Gid = int(sys.Gid)
25
+
26
+	// TODO(bradfitz): populate username & group.  os/user
27
+	// doesn't cache LookupId lookups, and lacks group
28
+	// lookup functions.
29
+	h.AccessTime = statAtime(sys)
30
+	h.ChangeTime = statCtime(sys)
31
+
32
+	// Best effort at populating Devmajor and Devminor.
33
+	if h.Typeflag == TypeChar || h.Typeflag == TypeBlock {
34
+		dev := uint64(sys.Rdev) // May be int32 or uint32
35
+		switch runtime.GOOS {
36
+		case "linux":
37
+			// Copied from golang.org/x/sys/unix/dev_linux.go.
38
+			major := uint32((dev & 0x00000000000fff00) >> 8)
39
+			major |= uint32((dev & 0xfffff00000000000) >> 32)
40
+			minor := uint32((dev & 0x00000000000000ff) >> 0)
41
+			minor |= uint32((dev & 0x00000ffffff00000) >> 12)
42
+			h.Devmajor, h.Devminor = int64(major), int64(minor)
43
+		case "darwin":
44
+			// Copied from golang.org/x/sys/unix/dev_darwin.go.
45
+			major := uint32((dev >> 24) & 0xff)
46
+			minor := uint32(dev & 0xffffff)
47
+			h.Devmajor, h.Devminor = int64(major), int64(minor)
48
+		case "dragonfly":
49
+			// Copied from golang.org/x/sys/unix/dev_dragonfly.go.
50
+			major := uint32((dev >> 8) & 0xff)
51
+			minor := uint32(dev & 0xffff00ff)
52
+			h.Devmajor, h.Devminor = int64(major), int64(minor)
53
+		case "freebsd":
54
+			// Copied from golang.org/x/sys/unix/dev_freebsd.go.
55
+			major := uint32((dev >> 8) & 0xff)
56
+			minor := uint32(dev & 0xffff00ff)
57
+			h.Devmajor, h.Devminor = int64(major), int64(minor)
58
+		case "netbsd":
59
+			// Copied from golang.org/x/sys/unix/dev_netbsd.go.
60
+			major := uint32((dev & 0x000fff00) >> 8)
61
+			minor := uint32((dev & 0x000000ff) >> 0)
62
+			minor |= uint32((dev & 0xfff00000) >> 12)
63
+			h.Devmajor, h.Devminor = int64(major), int64(minor)
64
+		case "openbsd":
65
+			// Copied from golang.org/x/sys/unix/dev_openbsd.go.
66
+			major := uint32((dev & 0x0000ff00) >> 8)
67
+			minor := uint32((dev & 0x000000ff) >> 0)
68
+			minor |= uint32((dev & 0xffff0000) >> 8)
69
+			h.Devmajor, h.Devminor = int64(major), int64(minor)
70
+		default:
71
+			// TODO: Implement solaris (see https://golang.org/issue/8106)
72
+		}
73
+	}
74
+	return nil
75
+}
0 76
new file mode 100644
... ...
@@ -0,0 +1,326 @@
0
+// Copyright 2016 The Go Authors. All rights reserved.
1
+// Use of this source code is governed by a BSD-style
2
+// license that can be found in the LICENSE file.
3
+
4
+package tar
5
+
6
+import (
7
+	"bytes"
8
+	"fmt"
9
+	"strconv"
10
+	"strings"
11
+	"time"
12
+)
13
+
14
+// hasNUL reports whether the NUL character exists within s.
15
+func hasNUL(s string) bool {
16
+	return strings.IndexByte(s, 0) >= 0
17
+}
18
+
19
+// isASCII reports whether the input is an ASCII C-style string.
20
+func isASCII(s string) bool {
21
+	for _, c := range s {
22
+		if c >= 0x80 || c == 0x00 {
23
+			return false
24
+		}
25
+	}
26
+	return true
27
+}
28
+
29
+// toASCII converts the input to an ASCII C-style string.
30
+// This a best effort conversion, so invalid characters are dropped.
31
+func toASCII(s string) string {
32
+	if isASCII(s) {
33
+		return s
34
+	}
35
+	b := make([]byte, 0, len(s))
36
+	for _, c := range s {
37
+		if c < 0x80 && c != 0x00 {
38
+			b = append(b, byte(c))
39
+		}
40
+	}
41
+	return string(b)
42
+}
43
+
44
+type parser struct {
45
+	err error // Last error seen
46
+}
47
+
48
+type formatter struct {
49
+	err error // Last error seen
50
+}
51
+
52
+// parseString parses bytes as a NUL-terminated C-style string.
53
+// If a NUL byte is not found then the whole slice is returned as a string.
54
+func (*parser) parseString(b []byte) string {
55
+	if i := bytes.IndexByte(b, 0); i >= 0 {
56
+		return string(b[:i])
57
+	}
58
+	return string(b)
59
+}
60
+
61
+// formatString copies s into b, NUL-terminating if possible.
62
+func (f *formatter) formatString(b []byte, s string) {
63
+	if len(s) > len(b) {
64
+		f.err = ErrFieldTooLong
65
+	}
66
+	copy(b, s)
67
+	if len(s) < len(b) {
68
+		b[len(s)] = 0
69
+	}
70
+
71
+	// Some buggy readers treat regular files with a trailing slash
72
+	// in the V7 path field as a directory even though the full path
73
+	// recorded elsewhere (e.g., via PAX record) contains no trailing slash.
74
+	if len(s) > len(b) && b[len(b)-1] == '/' {
75
+		n := len(strings.TrimRight(s[:len(b)], "/"))
76
+		b[n] = 0 // Replace trailing slash with NUL terminator
77
+	}
78
+}
79
+
80
+// fitsInBase256 reports whether x can be encoded into n bytes using base-256
81
+// encoding. Unlike octal encoding, base-256 encoding does not require that the
82
+// string ends with a NUL character. Thus, all n bytes are available for output.
83
+//
84
+// If operating in binary mode, this assumes strict GNU binary mode; which means
85
+// that the first byte can only be either 0x80 or 0xff. Thus, the first byte is
86
+// equivalent to the sign bit in two's complement form.
87
+func fitsInBase256(n int, x int64) bool {
88
+	binBits := uint(n-1) * 8
89
+	return n >= 9 || (x >= -1<<binBits && x < 1<<binBits)
90
+}
91
+
92
+// parseNumeric parses the input as being encoded in either base-256 or octal.
93
+// This function may return negative numbers.
94
+// If parsing fails or an integer overflow occurs, err will be set.
95
+func (p *parser) parseNumeric(b []byte) int64 {
96
+	// Check for base-256 (binary) format first.
97
+	// If the first bit is set, then all following bits constitute a two's
98
+	// complement encoded number in big-endian byte order.
99
+	if len(b) > 0 && b[0]&0x80 != 0 {
100
+		// Handling negative numbers relies on the following identity:
101
+		//	-a-1 == ^a
102
+		//
103
+		// If the number is negative, we use an inversion mask to invert the
104
+		// data bytes and treat the value as an unsigned number.
105
+		var inv byte // 0x00 if positive or zero, 0xff if negative
106
+		if b[0]&0x40 != 0 {
107
+			inv = 0xff
108
+		}
109
+
110
+		var x uint64
111
+		for i, c := range b {
112
+			c ^= inv // Inverts c only if inv is 0xff, otherwise does nothing
113
+			if i == 0 {
114
+				c &= 0x7f // Ignore signal bit in first byte
115
+			}
116
+			if (x >> 56) > 0 {
117
+				p.err = ErrHeader // Integer overflow
118
+				return 0
119
+			}
120
+			x = x<<8 | uint64(c)
121
+		}
122
+		if (x >> 63) > 0 {
123
+			p.err = ErrHeader // Integer overflow
124
+			return 0
125
+		}
126
+		if inv == 0xff {
127
+			return ^int64(x)
128
+		}
129
+		return int64(x)
130
+	}
131
+
132
+	// Normal case is base-8 (octal) format.
133
+	return p.parseOctal(b)
134
+}
135
+
136
+// formatNumeric encodes x into b using base-8 (octal) encoding if possible.
137
+// Otherwise it will attempt to use base-256 (binary) encoding.
138
+func (f *formatter) formatNumeric(b []byte, x int64) {
139
+	if fitsInOctal(len(b), x) {
140
+		f.formatOctal(b, x)
141
+		return
142
+	}
143
+
144
+	if fitsInBase256(len(b), x) {
145
+		for i := len(b) - 1; i >= 0; i-- {
146
+			b[i] = byte(x)
147
+			x >>= 8
148
+		}
149
+		b[0] |= 0x80 // Highest bit indicates binary format
150
+		return
151
+	}
152
+
153
+	f.formatOctal(b, 0) // Last resort, just write zero
154
+	f.err = ErrFieldTooLong
155
+}
156
+
157
+func (p *parser) parseOctal(b []byte) int64 {
158
+	// Because unused fields are filled with NULs, we need
159
+	// to skip leading NULs. Fields may also be padded with
160
+	// spaces or NULs.
161
+	// So we remove leading and trailing NULs and spaces to
162
+	// be sure.
163
+	b = bytes.Trim(b, " \x00")
164
+
165
+	if len(b) == 0 {
166
+		return 0
167
+	}
168
+	x, perr := strconv.ParseUint(p.parseString(b), 8, 64)
169
+	if perr != nil {
170
+		p.err = ErrHeader
171
+	}
172
+	return int64(x)
173
+}
174
+
175
+func (f *formatter) formatOctal(b []byte, x int64) {
176
+	if !fitsInOctal(len(b), x) {
177
+		x = 0 // Last resort, just write zero
178
+		f.err = ErrFieldTooLong
179
+	}
180
+
181
+	s := strconv.FormatInt(x, 8)
182
+	// Add leading zeros, but leave room for a NUL.
183
+	if n := len(b) - len(s) - 1; n > 0 {
184
+		s = strings.Repeat("0", n) + s
185
+	}
186
+	f.formatString(b, s)
187
+}
188
+
189
+// fitsInOctal reports whether the integer x fits in a field n-bytes long
190
+// using octal encoding with the appropriate NUL terminator.
191
+func fitsInOctal(n int, x int64) bool {
192
+	octBits := uint(n-1) * 3
193
+	return x >= 0 && (n >= 22 || x < 1<<octBits)
194
+}
195
+
196
+// parsePAXTime takes a string of the form %d.%d as described in the PAX
197
+// specification. Note that this implementation allows for negative timestamps,
198
+// which is allowed for by the PAX specification, but not always portable.
199
+func parsePAXTime(s string) (time.Time, error) {
200
+	const maxNanoSecondDigits = 9
201
+
202
+	// Split string into seconds and sub-seconds parts.
203
+	ss, sn := s, ""
204
+	if pos := strings.IndexByte(s, '.'); pos >= 0 {
205
+		ss, sn = s[:pos], s[pos+1:]
206
+	}
207
+
208
+	// Parse the seconds.
209
+	secs, err := strconv.ParseInt(ss, 10, 64)
210
+	if err != nil {
211
+		return time.Time{}, ErrHeader
212
+	}
213
+	if len(sn) == 0 {
214
+		return time.Unix(secs, 0), nil // No sub-second values
215
+	}
216
+
217
+	// Parse the nanoseconds.
218
+	if strings.Trim(sn, "0123456789") != "" {
219
+		return time.Time{}, ErrHeader
220
+	}
221
+	if len(sn) < maxNanoSecondDigits {
222
+		sn += strings.Repeat("0", maxNanoSecondDigits-len(sn)) // Right pad
223
+	} else {
224
+		sn = sn[:maxNanoSecondDigits] // Right truncate
225
+	}
226
+	nsecs, _ := strconv.ParseInt(sn, 10, 64) // Must succeed
227
+	if len(ss) > 0 && ss[0] == '-' {
228
+		return time.Unix(secs, -1*nsecs), nil // Negative correction
229
+	}
230
+	return time.Unix(secs, nsecs), nil
231
+}
232
+
233
+// formatPAXTime converts ts into a time of the form %d.%d as described in the
234
+// PAX specification. This function is capable of negative timestamps.
235
+func formatPAXTime(ts time.Time) (s string) {
236
+	secs, nsecs := ts.Unix(), ts.Nanosecond()
237
+	if nsecs == 0 {
238
+		return strconv.FormatInt(secs, 10)
239
+	}
240
+
241
+	// If seconds is negative, then perform correction.
242
+	sign := ""
243
+	if secs < 0 {
244
+		sign = "-"             // Remember sign
245
+		secs = -(secs + 1)     // Add a second to secs
246
+		nsecs = -(nsecs - 1E9) // Take that second away from nsecs
247
+	}
248
+	return strings.TrimRight(fmt.Sprintf("%s%d.%09d", sign, secs, nsecs), "0")
249
+}
250
+
251
+// parsePAXRecord parses the input PAX record string into a key-value pair.
252
+// If parsing is successful, it will slice off the currently read record and
253
+// return the remainder as r.
254
+func parsePAXRecord(s string) (k, v, r string, err error) {
255
+	// The size field ends at the first space.
256
+	sp := strings.IndexByte(s, ' ')
257
+	if sp == -1 {
258
+		return "", "", s, ErrHeader
259
+	}
260
+
261
+	// Parse the first token as a decimal integer.
262
+	n, perr := strconv.ParseInt(s[:sp], 10, 0) // Intentionally parse as native int
263
+	if perr != nil || n < 5 || int64(len(s)) < n {
264
+		return "", "", s, ErrHeader
265
+	}
266
+
267
+	// Extract everything between the space and the final newline.
268
+	rec, nl, rem := s[sp+1:n-1], s[n-1:n], s[n:]
269
+	if nl != "\n" {
270
+		return "", "", s, ErrHeader
271
+	}
272
+
273
+	// The first equals separates the key from the value.
274
+	eq := strings.IndexByte(rec, '=')
275
+	if eq == -1 {
276
+		return "", "", s, ErrHeader
277
+	}
278
+	k, v = rec[:eq], rec[eq+1:]
279
+
280
+	if !validPAXRecord(k, v) {
281
+		return "", "", s, ErrHeader
282
+	}
283
+	return k, v, rem, nil
284
+}
285
+
286
+// formatPAXRecord formats a single PAX record, prefixing it with the
287
+// appropriate length.
288
+func formatPAXRecord(k, v string) (string, error) {
289
+	if !validPAXRecord(k, v) {
290
+		return "", ErrHeader
291
+	}
292
+
293
+	const padding = 3 // Extra padding for ' ', '=', and '\n'
294
+	size := len(k) + len(v) + padding
295
+	size += len(strconv.Itoa(size))
296
+	record := strconv.Itoa(size) + " " + k + "=" + v + "\n"
297
+
298
+	// Final adjustment if adding size field increased the record size.
299
+	if len(record) != size {
300
+		size = len(record)
301
+		record = strconv.Itoa(size) + " " + k + "=" + v + "\n"
302
+	}
303
+	return record, nil
304
+}
305
+
306
+// validPAXRecord reports whether the key-value pair is valid where each
307
+// record is formatted as:
308
+//	"%d %s=%s\n" % (size, key, value)
309
+//
310
+// Keys and values should be UTF-8, but the number of bad writers out there
311
+// forces us to be a more liberal.
312
+// Thus, we only reject all keys with NUL, and only reject NULs in values
313
+// for the PAX version of the USTAR string fields.
314
+// The key must not contain an '=' character.
315
+func validPAXRecord(k, v string) bool {
316
+	if k == "" || strings.IndexByte(k, '=') >= 0 {
317
+		return false
318
+	}
319
+	switch k {
320
+	case paxPath, paxLinkpath, paxUname, paxGname:
321
+		return !hasNUL(v)
322
+	default:
323
+		return !hasNUL(k)
324
+	}
325
+}
0 326
new file mode 100644
... ...
@@ -0,0 +1,644 @@
0
+// Copyright 2009 The Go Authors. All rights reserved.
1
+// Use of this source code is governed by a BSD-style
2
+// license that can be found in the LICENSE file.
3
+
4
+package tar
5
+
6
+import (
7
+	"bytes"
8
+	"fmt"
9
+	"io"
10
+	"path"
11
+	"sort"
12
+	"strings"
13
+	"time"
14
+)
15
+
16
+// Writer provides sequential writing of a tar archive.
17
+// Write.WriteHeader begins a new file with the provided Header,
18
+// and then Writer can be treated as an io.Writer to supply that file's data.
19
+type Writer struct {
20
+	w    io.Writer
21
+	pad  int64      // Amount of padding to write after current file entry
22
+	curr fileWriter // Writer for current file entry
23
+	hdr  Header     // Shallow copy of Header that is safe for mutations
24
+	blk  block      // Buffer to use as temporary local storage
25
+
26
+	// err is a persistent error.
27
+	// It is only the responsibility of every exported method of Writer to
28
+	// ensure that this error is sticky.
29
+	err error
30
+}
31
+
32
+// NewWriter creates a new Writer writing to w.
33
+func NewWriter(w io.Writer) *Writer {
34
+	return &Writer{w: w, curr: &regFileWriter{w, 0}}
35
+}
36
+
37
+type fileWriter interface {
38
+	io.Writer
39
+	fileState
40
+
41
+	ReadFrom(io.Reader) (int64, error)
42
+}
43
+
44
+// Flush finishes writing the current file's block padding.
45
+// The current file must be fully written before Flush can be called.
46
+//
47
+// This is unnecessary as the next call to WriteHeader or Close
48
+// will implicitly flush out the file's padding.
49
+func (tw *Writer) Flush() error {
50
+	if tw.err != nil {
51
+		return tw.err
52
+	}
53
+	if nb := tw.curr.LogicalRemaining(); nb > 0 {
54
+		return fmt.Errorf("archive/tar: missed writing %d bytes", nb)
55
+	}
56
+	if _, tw.err = tw.w.Write(zeroBlock[:tw.pad]); tw.err != nil {
57
+		return tw.err
58
+	}
59
+	tw.pad = 0
60
+	return nil
61
+}
62
+
63
+// WriteHeader writes hdr and prepares to accept the file's contents.
64
+// The Header.Size determines how many bytes can be written for the next file.
65
+// If the current file is not fully written, then this returns an error.
66
+// This implicitly flushes any padding necessary before writing the header.
67
+func (tw *Writer) WriteHeader(hdr *Header) error {
68
+	if err := tw.Flush(); err != nil {
69
+		return err
70
+	}
71
+	tw.hdr = *hdr // Shallow copy of Header
72
+
73
+	// Round ModTime and ignore AccessTime and ChangeTime unless
74
+	// the format is explicitly chosen.
75
+	// This ensures nominal usage of WriteHeader (without specifying the format)
76
+	// does not always result in the PAX format being chosen, which
77
+	// causes a 1KiB increase to every header.
78
+	if tw.hdr.Format == FormatUnknown {
79
+		tw.hdr.ModTime = tw.hdr.ModTime.Round(time.Second)
80
+		tw.hdr.AccessTime = time.Time{}
81
+		tw.hdr.ChangeTime = time.Time{}
82
+	}
83
+
84
+	allowedFormats, paxHdrs, err := tw.hdr.allowedFormats()
85
+	switch {
86
+	case allowedFormats.has(FormatUSTAR):
87
+		tw.err = tw.writeUSTARHeader(&tw.hdr)
88
+		return tw.err
89
+	case allowedFormats.has(FormatPAX):
90
+		tw.err = tw.writePAXHeader(&tw.hdr, paxHdrs)
91
+		return tw.err
92
+	case allowedFormats.has(FormatGNU):
93
+		tw.err = tw.writeGNUHeader(&tw.hdr)
94
+		return tw.err
95
+	default:
96
+		return err // Non-fatal error
97
+	}
98
+}
99
+
100
+func (tw *Writer) writeUSTARHeader(hdr *Header) error {
101
+	// Check if we can use USTAR prefix/suffix splitting.
102
+	var namePrefix string
103
+	if prefix, suffix, ok := splitUSTARPath(hdr.Name); ok {
104
+		namePrefix, hdr.Name = prefix, suffix
105
+	}
106
+
107
+	// Pack the main header.
108
+	var f formatter
109
+	blk := tw.templateV7Plus(hdr, f.formatString, f.formatOctal)
110
+	f.formatString(blk.USTAR().Prefix(), namePrefix)
111
+	blk.SetFormat(FormatUSTAR)
112
+	if f.err != nil {
113
+		return f.err // Should never happen since header is validated
114
+	}
115
+	return tw.writeRawHeader(blk, hdr.Size, hdr.Typeflag)
116
+}
117
+
118
+func (tw *Writer) writePAXHeader(hdr *Header, paxHdrs map[string]string) error {
119
+	realName, realSize := hdr.Name, hdr.Size
120
+
121
+	// TODO(dsnet): Re-enable this when adding sparse support.
122
+	// See https://golang.org/issue/22735
123
+	/*
124
+		// Handle sparse files.
125
+		var spd sparseDatas
126
+		var spb []byte
127
+		if len(hdr.SparseHoles) > 0 {
128
+			sph := append([]sparseEntry{}, hdr.SparseHoles...) // Copy sparse map
129
+			sph = alignSparseEntries(sph, hdr.Size)
130
+			spd = invertSparseEntries(sph, hdr.Size)
131
+
132
+			// Format the sparse map.
133
+			hdr.Size = 0 // Replace with encoded size
134
+			spb = append(strconv.AppendInt(spb, int64(len(spd)), 10), '\n')
135
+			for _, s := range spd {
136
+				hdr.Size += s.Length
137
+				spb = append(strconv.AppendInt(spb, s.Offset, 10), '\n')
138
+				spb = append(strconv.AppendInt(spb, s.Length, 10), '\n')
139
+			}
140
+			pad := blockPadding(int64(len(spb)))
141
+			spb = append(spb, zeroBlock[:pad]...)
142
+			hdr.Size += int64(len(spb)) // Accounts for encoded sparse map
143
+
144
+			// Add and modify appropriate PAX records.
145
+			dir, file := path.Split(realName)
146
+			hdr.Name = path.Join(dir, "GNUSparseFile.0", file)
147
+			paxHdrs[paxGNUSparseMajor] = "1"
148
+			paxHdrs[paxGNUSparseMinor] = "0"
149
+			paxHdrs[paxGNUSparseName] = realName
150
+			paxHdrs[paxGNUSparseRealSize] = strconv.FormatInt(realSize, 10)
151
+			paxHdrs[paxSize] = strconv.FormatInt(hdr.Size, 10)
152
+			delete(paxHdrs, paxPath) // Recorded by paxGNUSparseName
153
+		}
154
+	*/
155
+	_ = realSize
156
+
157
+	// Write PAX records to the output.
158
+	isGlobal := hdr.Typeflag == TypeXGlobalHeader
159
+	if len(paxHdrs) > 0 || isGlobal {
160
+		// Sort keys for deterministic ordering.
161
+		var keys []string
162
+		for k := range paxHdrs {
163
+			keys = append(keys, k)
164
+		}
165
+		sort.Strings(keys)
166
+
167
+		// Write each record to a buffer.
168
+		var buf bytes.Buffer
169
+		for _, k := range keys {
170
+			rec, err := formatPAXRecord(k, paxHdrs[k])
171
+			if err != nil {
172
+				return err
173
+			}
174
+			buf.WriteString(rec)
175
+		}
176
+
177
+		// Write the extended header file.
178
+		var name string
179
+		var flag byte
180
+		if isGlobal {
181
+			name = realName
182
+			if name == "" {
183
+				name = "GlobalHead.0.0"
184
+			}
185
+			flag = TypeXGlobalHeader
186
+		} else {
187
+			dir, file := path.Split(realName)
188
+			name = path.Join(dir, "PaxHeaders.0", file)
189
+			flag = TypeXHeader
190
+		}
191
+		data := buf.String()
192
+		if err := tw.writeRawFile(name, data, flag, FormatPAX); err != nil || isGlobal {
193
+			return err // Global headers return here
194
+		}
195
+	}
196
+
197
+	// Pack the main header.
198
+	var f formatter // Ignore errors since they are expected
199
+	fmtStr := func(b []byte, s string) { f.formatString(b, toASCII(s)) }
200
+	blk := tw.templateV7Plus(hdr, fmtStr, f.formatOctal)
201
+	blk.SetFormat(FormatPAX)
202
+	if err := tw.writeRawHeader(blk, hdr.Size, hdr.Typeflag); err != nil {
203
+		return err
204
+	}
205
+
206
+	// TODO(dsnet): Re-enable this when adding sparse support.
207
+	// See https://golang.org/issue/22735
208
+	/*
209
+		// Write the sparse map and setup the sparse writer if necessary.
210
+		if len(spd) > 0 {
211
+			// Use tw.curr since the sparse map is accounted for in hdr.Size.
212
+			if _, err := tw.curr.Write(spb); err != nil {
213
+				return err
214
+			}
215
+			tw.curr = &sparseFileWriter{tw.curr, spd, 0}
216
+		}
217
+	*/
218
+	return nil
219
+}
220
+
221
+func (tw *Writer) writeGNUHeader(hdr *Header) error {
222
+	// Use long-link files if Name or Linkname exceeds the field size.
223
+	const longName = "././@LongLink"
224
+	if len(hdr.Name) > nameSize {
225
+		data := hdr.Name + "\x00"
226
+		if err := tw.writeRawFile(longName, data, TypeGNULongName, FormatGNU); err != nil {
227
+			return err
228
+		}
229
+	}
230
+	if len(hdr.Linkname) > nameSize {
231
+		data := hdr.Linkname + "\x00"
232
+		if err := tw.writeRawFile(longName, data, TypeGNULongLink, FormatGNU); err != nil {
233
+			return err
234
+		}
235
+	}
236
+
237
+	// Pack the main header.
238
+	var f formatter // Ignore errors since they are expected
239
+	var spd sparseDatas
240
+	var spb []byte
241
+	blk := tw.templateV7Plus(hdr, f.formatString, f.formatNumeric)
242
+	if !hdr.AccessTime.IsZero() {
243
+		f.formatNumeric(blk.GNU().AccessTime(), hdr.AccessTime.Unix())
244
+	}
245
+	if !hdr.ChangeTime.IsZero() {
246
+		f.formatNumeric(blk.GNU().ChangeTime(), hdr.ChangeTime.Unix())
247
+	}
248
+	// TODO(dsnet): Re-enable this when adding sparse support.
249
+	// See https://golang.org/issue/22735
250
+	/*
251
+		if hdr.Typeflag == TypeGNUSparse {
252
+			sph := append([]sparseEntry{}, hdr.SparseHoles...) // Copy sparse map
253
+			sph = alignSparseEntries(sph, hdr.Size)
254
+			spd = invertSparseEntries(sph, hdr.Size)
255
+
256
+			// Format the sparse map.
257
+			formatSPD := func(sp sparseDatas, sa sparseArray) sparseDatas {
258
+				for i := 0; len(sp) > 0 && i < sa.MaxEntries(); i++ {
259
+					f.formatNumeric(sa.Entry(i).Offset(), sp[0].Offset)
260
+					f.formatNumeric(sa.Entry(i).Length(), sp[0].Length)
261
+					sp = sp[1:]
262
+				}
263
+				if len(sp) > 0 {
264
+					sa.IsExtended()[0] = 1
265
+				}
266
+				return sp
267
+			}
268
+			sp2 := formatSPD(spd, blk.GNU().Sparse())
269
+			for len(sp2) > 0 {
270
+				var spHdr block
271
+				sp2 = formatSPD(sp2, spHdr.Sparse())
272
+				spb = append(spb, spHdr[:]...)
273
+			}
274
+
275
+			// Update size fields in the header block.
276
+			realSize := hdr.Size
277
+			hdr.Size = 0 // Encoded size; does not account for encoded sparse map
278
+			for _, s := range spd {
279
+				hdr.Size += s.Length
280
+			}
281
+			copy(blk.V7().Size(), zeroBlock[:]) // Reset field
282
+			f.formatNumeric(blk.V7().Size(), hdr.Size)
283
+			f.formatNumeric(blk.GNU().RealSize(), realSize)
284
+		}
285
+	*/
286
+	blk.SetFormat(FormatGNU)
287
+	if err := tw.writeRawHeader(blk, hdr.Size, hdr.Typeflag); err != nil {
288
+		return err
289
+	}
290
+
291
+	// Write the extended sparse map and setup the sparse writer if necessary.
292
+	if len(spd) > 0 {
293
+		// Use tw.w since the sparse map is not accounted for in hdr.Size.
294
+		if _, err := tw.w.Write(spb); err != nil {
295
+			return err
296
+		}
297
+		tw.curr = &sparseFileWriter{tw.curr, spd, 0}
298
+	}
299
+	return nil
300
+}
301
+
302
+type (
303
+	stringFormatter func([]byte, string)
304
+	numberFormatter func([]byte, int64)
305
+)
306
+
307
+// templateV7Plus fills out the V7 fields of a block using values from hdr.
308
+// It also fills out fields (uname, gname, devmajor, devminor) that are
309
+// shared in the USTAR, PAX, and GNU formats using the provided formatters.
310
+//
311
+// The block returned is only valid until the next call to
312
+// templateV7Plus or writeRawFile.
313
+func (tw *Writer) templateV7Plus(hdr *Header, fmtStr stringFormatter, fmtNum numberFormatter) *block {
314
+	tw.blk.Reset()
315
+
316
+	modTime := hdr.ModTime
317
+	if modTime.IsZero() {
318
+		modTime = time.Unix(0, 0)
319
+	}
320
+
321
+	v7 := tw.blk.V7()
322
+	v7.TypeFlag()[0] = hdr.Typeflag
323
+	fmtStr(v7.Name(), hdr.Name)
324
+	fmtStr(v7.LinkName(), hdr.Linkname)
325
+	fmtNum(v7.Mode(), hdr.Mode)
326
+	fmtNum(v7.UID(), int64(hdr.Uid))
327
+	fmtNum(v7.GID(), int64(hdr.Gid))
328
+	fmtNum(v7.Size(), hdr.Size)
329
+	fmtNum(v7.ModTime(), modTime.Unix())
330
+
331
+	ustar := tw.blk.USTAR()
332
+	fmtStr(ustar.UserName(), hdr.Uname)
333
+	fmtStr(ustar.GroupName(), hdr.Gname)
334
+	fmtNum(ustar.DevMajor(), hdr.Devmajor)
335
+	fmtNum(ustar.DevMinor(), hdr.Devminor)
336
+
337
+	return &tw.blk
338
+}
339
+
340
+// writeRawFile writes a minimal file with the given name and flag type.
341
+// It uses format to encode the header format and will write data as the body.
342
+// It uses default values for all of the other fields (as BSD and GNU tar does).
343
+func (tw *Writer) writeRawFile(name, data string, flag byte, format Format) error {
344
+	tw.blk.Reset()
345
+
346
+	// Best effort for the filename.
347
+	name = toASCII(name)
348
+	if len(name) > nameSize {
349
+		name = name[:nameSize]
350
+	}
351
+	name = strings.TrimRight(name, "/")
352
+
353
+	var f formatter
354
+	v7 := tw.blk.V7()
355
+	v7.TypeFlag()[0] = flag
356
+	f.formatString(v7.Name(), name)
357
+	f.formatOctal(v7.Mode(), 0)
358
+	f.formatOctal(v7.UID(), 0)
359
+	f.formatOctal(v7.GID(), 0)
360
+	f.formatOctal(v7.Size(), int64(len(data))) // Must be < 8GiB
361
+	f.formatOctal(v7.ModTime(), 0)
362
+	tw.blk.SetFormat(format)
363
+	if f.err != nil {
364
+		return f.err // Only occurs if size condition is violated
365
+	}
366
+
367
+	// Write the header and data.
368
+	if err := tw.writeRawHeader(&tw.blk, int64(len(data)), flag); err != nil {
369
+		return err
370
+	}
371
+	_, err := io.WriteString(tw, data)
372
+	return err
373
+}
374
+
375
+// writeRawHeader writes the value of blk, regardless of its value.
376
+// It sets up the Writer such that it can accept a file of the given size.
377
+// If the flag is a special header-only flag, then the size is treated as zero.
378
+func (tw *Writer) writeRawHeader(blk *block, size int64, flag byte) error {
379
+	if err := tw.Flush(); err != nil {
380
+		return err
381
+	}
382
+	if _, err := tw.w.Write(blk[:]); err != nil {
383
+		return err
384
+	}
385
+	if isHeaderOnlyType(flag) {
386
+		size = 0
387
+	}
388
+	tw.curr = &regFileWriter{tw.w, size}
389
+	tw.pad = blockPadding(size)
390
+	return nil
391
+}
392
+
393
+// splitUSTARPath splits a path according to USTAR prefix and suffix rules.
394
+// If the path is not splittable, then it will return ("", "", false).
395
+func splitUSTARPath(name string) (prefix, suffix string, ok bool) {
396
+	length := len(name)
397
+	if length <= nameSize || !isASCII(name) {
398
+		return "", "", false
399
+	} else if length > prefixSize+1 {
400
+		length = prefixSize + 1
401
+	} else if name[length-1] == '/' {
402
+		length--
403
+	}
404
+
405
+	i := strings.LastIndex(name[:length], "/")
406
+	nlen := len(name) - i - 1 // nlen is length of suffix
407
+	plen := i                 // plen is length of prefix
408
+	if i <= 0 || nlen > nameSize || nlen == 0 || plen > prefixSize {
409
+		return "", "", false
410
+	}
411
+	return name[:i], name[i+1:], true
412
+}
413
+
414
+// Write writes to the current file in the tar archive.
415
+// Write returns the error ErrWriteTooLong if more than
416
+// Header.Size bytes are written after WriteHeader.
417
+//
418
+// Calling Write on special types like TypeLink, TypeSymlink, TypeChar,
419
+// TypeBlock, TypeDir, and TypeFifo returns (0, ErrWriteTooLong) regardless
420
+// of what the Header.Size claims.
421
+func (tw *Writer) Write(b []byte) (int, error) {
422
+	if tw.err != nil {
423
+		return 0, tw.err
424
+	}
425
+	n, err := tw.curr.Write(b)
426
+	if err != nil && err != ErrWriteTooLong {
427
+		tw.err = err
428
+	}
429
+	return n, err
430
+}
431
+
432
+// readFrom populates the content of the current file by reading from r.
433
+// The bytes read must match the number of remaining bytes in the current file.
434
+//
435
+// If the current file is sparse and r is an io.ReadSeeker,
436
+// then readFrom uses Seek to skip past holes defined in Header.SparseHoles,
437
+// assuming that skipped regions are all NULs.
438
+// This always reads the last byte to ensure r is the right size.
439
+//
440
+// TODO(dsnet): Re-export this when adding sparse file support.
441
+// See https://golang.org/issue/22735
442
+func (tw *Writer) readFrom(r io.Reader) (int64, error) {
443
+	if tw.err != nil {
444
+		return 0, tw.err
445
+	}
446
+	n, err := tw.curr.ReadFrom(r)
447
+	if err != nil && err != ErrWriteTooLong {
448
+		tw.err = err
449
+	}
450
+	return n, err
451
+}
452
+
453
+// Close closes the tar archive by flushing the padding, and writing the footer.
454
+// If the current file (from a prior call to WriteHeader) is not fully written,
455
+// then this returns an error.
456
+func (tw *Writer) Close() error {
457
+	if tw.err == ErrWriteAfterClose {
458
+		return nil
459
+	}
460
+	if tw.err != nil {
461
+		return tw.err
462
+	}
463
+
464
+	// Trailer: two zero blocks.
465
+	err := tw.Flush()
466
+	for i := 0; i < 2 && err == nil; i++ {
467
+		_, err = tw.w.Write(zeroBlock[:])
468
+	}
469
+
470
+	// Ensure all future actions are invalid.
471
+	tw.err = ErrWriteAfterClose
472
+	return err // Report IO errors
473
+}
474
+
475
+// regFileWriter is a fileWriter for writing data to a regular file entry.
476
+type regFileWriter struct {
477
+	w  io.Writer // Underlying Writer
478
+	nb int64     // Number of remaining bytes to write
479
+}
480
+
481
+func (fw *regFileWriter) Write(b []byte) (n int, err error) {
482
+	overwrite := int64(len(b)) > fw.nb
483
+	if overwrite {
484
+		b = b[:fw.nb]
485
+	}
486
+	if len(b) > 0 {
487
+		n, err = fw.w.Write(b)
488
+		fw.nb -= int64(n)
489
+	}
490
+	switch {
491
+	case err != nil:
492
+		return n, err
493
+	case overwrite:
494
+		return n, ErrWriteTooLong
495
+	default:
496
+		return n, nil
497
+	}
498
+}
499
+
500
+func (fw *regFileWriter) ReadFrom(r io.Reader) (int64, error) {
501
+	return io.Copy(struct{ io.Writer }{fw}, r)
502
+}
503
+
504
+func (fw regFileWriter) LogicalRemaining() int64 {
505
+	return fw.nb
506
+}
507
+func (fw regFileWriter) PhysicalRemaining() int64 {
508
+	return fw.nb
509
+}
510
+
511
+// sparseFileWriter is a fileWriter for writing data to a sparse file entry.
512
+type sparseFileWriter struct {
513
+	fw  fileWriter  // Underlying fileWriter
514
+	sp  sparseDatas // Normalized list of data fragments
515
+	pos int64       // Current position in sparse file
516
+}
517
+
518
+func (sw *sparseFileWriter) Write(b []byte) (n int, err error) {
519
+	overwrite := int64(len(b)) > sw.LogicalRemaining()
520
+	if overwrite {
521
+		b = b[:sw.LogicalRemaining()]
522
+	}
523
+
524
+	b0 := b
525
+	endPos := sw.pos + int64(len(b))
526
+	for endPos > sw.pos && err == nil {
527
+		var nf int // Bytes written in fragment
528
+		dataStart, dataEnd := sw.sp[0].Offset, sw.sp[0].endOffset()
529
+		if sw.pos < dataStart { // In a hole fragment
530
+			bf := b[:min(int64(len(b)), dataStart-sw.pos)]
531
+			nf, err = zeroWriter{}.Write(bf)
532
+		} else { // In a data fragment
533
+			bf := b[:min(int64(len(b)), dataEnd-sw.pos)]
534
+			nf, err = sw.fw.Write(bf)
535
+		}
536
+		b = b[nf:]
537
+		sw.pos += int64(nf)
538
+		if sw.pos >= dataEnd && len(sw.sp) > 1 {
539
+			sw.sp = sw.sp[1:] // Ensure last fragment always remains
540
+		}
541
+	}
542
+
543
+	n = len(b0) - len(b)
544
+	switch {
545
+	case err == ErrWriteTooLong:
546
+		return n, errMissData // Not possible; implies bug in validation logic
547
+	case err != nil:
548
+		return n, err
549
+	case sw.LogicalRemaining() == 0 && sw.PhysicalRemaining() > 0:
550
+		return n, errUnrefData // Not possible; implies bug in validation logic
551
+	case overwrite:
552
+		return n, ErrWriteTooLong
553
+	default:
554
+		return n, nil
555
+	}
556
+}
557
+
558
+func (sw *sparseFileWriter) ReadFrom(r io.Reader) (n int64, err error) {
559
+	rs, ok := r.(io.ReadSeeker)
560
+	if ok {
561
+		if _, err := rs.Seek(0, io.SeekCurrent); err != nil {
562
+			ok = false // Not all io.Seeker can really seek
563
+		}
564
+	}
565
+	if !ok {
566
+		return io.Copy(struct{ io.Writer }{sw}, r)
567
+	}
568
+
569
+	var readLastByte bool
570
+	pos0 := sw.pos
571
+	for sw.LogicalRemaining() > 0 && !readLastByte && err == nil {
572
+		var nf int64 // Size of fragment
573
+		dataStart, dataEnd := sw.sp[0].Offset, sw.sp[0].endOffset()
574
+		if sw.pos < dataStart { // In a hole fragment
575
+			nf = dataStart - sw.pos
576
+			if sw.PhysicalRemaining() == 0 {
577
+				readLastByte = true
578
+				nf--
579
+			}
580
+			_, err = rs.Seek(nf, io.SeekCurrent)
581
+		} else { // In a data fragment
582
+			nf = dataEnd - sw.pos
583
+			nf, err = io.CopyN(sw.fw, rs, nf)
584
+		}
585
+		sw.pos += nf
586
+		if sw.pos >= dataEnd && len(sw.sp) > 1 {
587
+			sw.sp = sw.sp[1:] // Ensure last fragment always remains
588
+		}
589
+	}
590
+
591
+	// If the last fragment is a hole, then seek to 1-byte before EOF, and
592
+	// read a single byte to ensure the file is the right size.
593
+	if readLastByte && err == nil {
594
+		_, err = mustReadFull(rs, []byte{0})
595
+		sw.pos++
596
+	}
597
+
598
+	n = sw.pos - pos0
599
+	switch {
600
+	case err == io.EOF:
601
+		return n, io.ErrUnexpectedEOF
602
+	case err == ErrWriteTooLong:
603
+		return n, errMissData // Not possible; implies bug in validation logic
604
+	case err != nil:
605
+		return n, err
606
+	case sw.LogicalRemaining() == 0 && sw.PhysicalRemaining() > 0:
607
+		return n, errUnrefData // Not possible; implies bug in validation logic
608
+	default:
609
+		return n, ensureEOF(rs)
610
+	}
611
+}
612
+
613
+func (sw sparseFileWriter) LogicalRemaining() int64 {
614
+	return sw.sp[len(sw.sp)-1].endOffset() - sw.pos
615
+}
616
+func (sw sparseFileWriter) PhysicalRemaining() int64 {
617
+	return sw.fw.PhysicalRemaining()
618
+}
619
+
620
+// zeroWriter may only be written with NULs, otherwise it returns errWriteHole.
621
+type zeroWriter struct{}
622
+
623
+func (zeroWriter) Write(b []byte) (int, error) {
624
+	for i, c := range b {
625
+		if c != 0 {
626
+			return i, errWriteHole
627
+		}
628
+	}
629
+	return len(b), nil
630
+}
631
+
632
+// ensureEOF checks whether r is at EOF, reporting ErrWriteTooLong if not so.
633
+func ensureEOF(r io.Reader) error {
634
+	n, err := tryReadFull(r, []byte{0})
635
+	switch {
636
+	case n > 0:
637
+		return ErrWriteTooLong
638
+	case err == io.EOF:
639
+		return nil
640
+	default:
641
+		return err
642
+	}
643
+}