archive/tar: make output deterministic

Replaces PID in PaxHeaders with 0.  Sorts PAX header keys before writing
them to the archive.

Fixes golang#12358

Change-Id: If239f89c85f1c9d9895a253fb06a47ad44960124
Reviewed-on: https://go-review.googlesource.com/13975
Reviewed-by: Russ Cox <[email protected]>
Reviewed-by: Joe Tsai <[email protected]>

src/archive/tar/writer.go

@@ -12,8 +12,8 @@ import (
 	"errors"
 	"fmt"
 	"io"
-	"os"
 	"path"
+	"sort"
 	"strconv"
 	"strings"
 	"time"
@@ -288,11 +288,11 @@ func (tw *Writer) writePAXHeader(hdr *Header, paxHeaders map[string]string) erro
 	// succeed, and seems harmless enough.
 	ext.ModTime = hdr.ModTime
 	// The spec asks that we namespace our pseudo files
-	// with the current pid.
-	pid := os.Getpid()
+	// with the current pid.  However, this results in differing outputs
+	// for identical inputs.  As such, the constant 0 is now used instead.
+	// golang.org/issue/12358
 	dir, file := path.Split(hdr.Name)
-	fullName := path.Join(dir,
-		fmt.Sprintf("PaxHeaders.%d", pid), file)
+	fullName := path.Join(dir, "PaxHeaders.0", file)
 
 	ascii := toASCII(fullName)
 	if len(ascii) > 100 {
@@ -302,8 +302,15 @@ func (tw *Writer) writePAXHeader(hdr *Header, paxHeaders map[string]string) erro
 	// Construct the body
 	var buf bytes.Buffer
 
-	for k, v := range paxHeaders {
-		fmt.Fprint(&buf, paxHeader(k+"="+v))
+	// Keys are sorted before writing to body to allow deterministic output.
+	var keys []string
+	for k := range paxHeaders {
+		keys = append(keys, k)
+	}
+	sort.Strings(keys)
+
+	for _, k := range keys {
+		fmt.Fprint(&buf, paxHeader(k+"="+paxHeaders[k]))
 	}
 
 	ext.Size = int64(len(buf.Bytes()))

src/archive/tar/writer_test.go

@@ -11,6 +11,7 @@ import (
 	"io/ioutil"
 	"os"
 	"reflect"
+	"sort"
 	"strings"
 	"testing"
 	"testing/iotest"
@@ -291,7 +292,7 @@ func TestPax(t *testing.T) {
 		t.Fatal(err)
 	}
 	// Simple test to make sure PAX extensions are in effect
-	if !bytes.Contains(buf.Bytes(), []byte("PaxHeaders.")) {
+	if !bytes.Contains(buf.Bytes(), []byte("PaxHeaders.0")) {
 		t.Fatal("Expected at least one PAX header to be written.")
 	}
 	// Test that we can get a long name back out of the archive.
@@ -330,7 +331,7 @@ func TestPaxSymlink(t *testing.T) {
 		t.Fatal(err)
 	}
 	// Simple test to make sure PAX extensions are in effect
-	if !bytes.Contains(buf.Bytes(), []byte("PaxHeaders.")) {
+	if !bytes.Contains(buf.Bytes(), []byte("PaxHeaders.0")) {
 		t.Fatal("Expected at least one PAX header to be written.")
 	}
 	// Test that we can get a long name back out of the archive.
@@ -380,7 +381,7 @@ func TestPaxNonAscii(t *testing.T) {
 		t.Fatal(err)
 	}
 	// Simple test to make sure PAX extensions are in effect
-	if !bytes.Contains(buf.Bytes(), []byte("PaxHeaders.")) {
+	if !bytes.Contains(buf.Bytes(), []byte("PaxHeaders.0")) {
 		t.Fatal("Expected at least one PAX header to be written.")
 	}
 	// Test that we can get a long name back out of the archive.
@@ -439,6 +440,52 @@ func TestPaxXattrs(t *testing.T) {
 	}
 }
 
+func TestPaxHeadersSorted(t *testing.T) {
+	fileinfo, err := os.Stat("testdata/small.txt")
+	if err != nil {
+		t.Fatal(err)
+	}
+	hdr, err := FileInfoHeader(fileinfo, "")
+	if err != nil {
+		t.Fatalf("os.Stat: %v", err)
+	}
+	contents := strings.Repeat(" ", int(hdr.Size))
+
+	hdr.Xattrs = map[string]string{
+		"foo": "foo",
+		"bar": "bar",
+		"baz": "baz",
+		"qux": "qux",
+	}
+
+	var buf bytes.Buffer
+	writer := NewWriter(&buf)
+	if err := writer.WriteHeader(hdr); err != nil {
+		t.Fatal(err)
+	}
+	if _, err = writer.Write([]byte(contents)); err != nil {
+		t.Fatal(err)
+	}
+	if err := writer.Close(); err != nil {
+		t.Fatal(err)
+	}
+	// Simple test to make sure PAX extensions are in effect
+	if !bytes.Contains(buf.Bytes(), []byte("PaxHeaders.0")) {
+		t.Fatal("Expected at least one PAX header to be written.")
+	}
+
+	// xattr bar should always appear before others
+	indices := []int{
+		bytes.Index(buf.Bytes(), []byte("bar=bar")),
+		bytes.Index(buf.Bytes(), []byte("baz=baz")),
+		bytes.Index(buf.Bytes(), []byte("foo=foo")),
+		bytes.Index(buf.Bytes(), []byte("qux=qux")),
+	}
+	if !sort.IntsAreSorted(indices) {
+		t.Fatal("PAX headers are not sorted")
+	}
+}
+
 func TestPAXHeader(t *testing.T) {
 	medName := strings.Repeat("CD", 50)
 	longName := strings.Repeat("AB", 100)