Merge pull request #29 from stephen679/tree_tar_stream

*: creating validation manifest from archive and validating a manifest against archive

check.go

@@ -3,14 +3,16 @@ package mtree
 import (
 	"fmt"
 	"os"
-	"path/filepath"
 	"sort"
+	"strings"
 )
 
 // Result of a Check
 type Result struct {
 	// list of any failures in the Check
 	Failures []Failure `json:"failures"`
+	Missing  []Entry
+	Extra    []Entry
 }
 
 // Failure of a particular keyword for a path
@@ -51,7 +53,8 @@ func Check(root string, dh *DirectoryHierarchy, keywords []string) (*Result, err
 				creator.curSet = nil
 			}
 		case RelativeType, FullType:
-			info, err := os.Lstat(e.Path())
+			filename := e.Path()
+			info, err := os.Lstat(filename)
 			if err != nil {
 				return nil, err
 			}
@@ -64,17 +67,31 @@ func Check(root string, dh *DirectoryHierarchy, keywords []string) (*Result, err
 			}
 
 			for _, kv := range kvs {
-				keywordFunc, ok := KeywordFuncs[kv.Keyword()]
+				kw := kv.Keyword()
+				// 'tar_time' keyword evaluation wins against 'time' keyword evaluation
+				if kv.Keyword() == "time" && inSlice("tar_time", keywords) {
+					kw = "tar_time"
+					tartime := fmt.Sprintf("%s.%s", (strings.Split(kv.Value(), ".")[0]), "000000000")
+					kv = KeyVal(KeyVal(kw).ChangeValue(tartime))
+				}
+
+				keywordFunc, ok := KeywordFuncs[kw]
 				if !ok {
 					return nil, fmt.Errorf("Unknown keyword %q for file %q", kv.Keyword(), e.Path())
 				}
 				if keywords != nil && !inSlice(kv.Keyword(), keywords) {
 					continue
 				}
-				curKeyVal, err := keywordFunc(filepath.Join(root, e.Path()), info)
+				fh, err := os.Open(filename)
 				if err != nil {
 					return nil, err
 				}
+				curKeyVal, err := keywordFunc(filename, info, fh)
+				if err != nil {
+					fh.Close()
+					return nil, err
+				}
+				fh.Close()
 				if string(kv) != curKeyVal {
 					failure := Failure{Path: e.Path(), Keyword: kv.Keyword(), Expected: kv.Value(), Got: KeyVal(curKeyVal).Value()}
 					result.Failures = append(result.Failures, failure)
@@ -84,3 +101,99 @@ func Check(root string, dh *DirectoryHierarchy, keywords []string) (*Result, err
 	}
 	return &result, nil
 }
+
+// TarCheck is the tar equivalent of checking a file hierarchy spec against a tar stream to
+// determine if files have been changed.
+func TarCheck(tarDH, dh *DirectoryHierarchy, keywords []string) (*Result, error) {
+	var result Result
+	var err error
+	var tarRoot *Entry
+
+	for _, e := range tarDH.Entries {
+		if e.Name == "." {
+			tarRoot = &e
+			break
+		}
+	}
+	tarRoot.Next = &Entry{
+		Name: "seen",
+		Type: CommentType,
+	}
+	curDir := tarRoot
+	creator := dhCreator{DH: dh}
+	sort.Sort(byPos(creator.DH.Entries))
+
+	var outOfTree bool
+	for i, e := range creator.DH.Entries {
+		switch e.Type {
+		case SpecialType:
+			if e.Name == "/set" {
+				creator.curSet = &creator.DH.Entries[i]
+			} else if e.Name == "/unset" {
+				creator.curSet = nil
+			}
+		case RelativeType, FullType:
+			if outOfTree {
+				return &result, fmt.Errorf("No parent node from %s", e.Path())
+			}
+			// TODO: handle the case where "." is not the first Entry to be found
+			tarEntry := curDir.Descend(e.Name)
+			if tarEntry == nil {
+				result.Missing = append(result.Missing, e)
+				continue
+			}
+
+			tarEntry.Next = &Entry{
+				Type: CommentType,
+				Name: "seen",
+			}
+
+			// expected values from file hierarchy spec
+			var kvs KeyVals
+			if creator.curSet != nil {
+				kvs = MergeSet(creator.curSet.Keywords, e.Keywords)
+			} else {
+				kvs = NewKeyVals(e.Keywords)
+			}
+
+			// actual
+			var tarkvs KeyVals
+			if tarEntry.Set != nil {
+				tarkvs = MergeSet(tarEntry.Set.Keywords, tarEntry.Keywords)
+			} else {
+				tarkvs = NewKeyVals(tarEntry.Keywords)
+			}
+
+			for _, kv := range kvs {
+				if _, ok := KeywordFuncs[kv.Keyword()]; !ok {
+					return nil, fmt.Errorf("Unknown keyword %q for file %q", kv.Keyword(), e.Path())
+				}
+				if keywords != nil && !inSlice(kv.Keyword(), keywords) {
+					continue
+				}
+				if tarkv := tarkvs.Has(kv.Keyword()); tarkv != emptyKV {
+					if string(tarkv) != string(kv) {
+						failure := Failure{Path: tarEntry.Path(), Keyword: kv.Keyword(), Expected: kv.Value(), Got: tarkv.Value()}
+						result.Failures = append(result.Failures, failure)
+					}
+				}
+			}
+			// Step into a directory
+			if tarEntry.Prev != nil {
+				curDir = tarEntry
+			}
+		case DotDotType:
+			if outOfTree {
+				return &result, fmt.Errorf("No parent node.")
+			}
+			curDir = curDir.Ascend()
+			if curDir == nil {
+				outOfTree = true
+			}
+		}
+	}
+	result.Extra = filter(tarRoot, func(e *Entry) bool {
+		return e.Next == nil
+	})
+	return &result, err
+}

check_test.go

@@ -143,6 +143,58 @@ func TestTimeComparison(t *testing.T) {
 	}
 }
 
+func TestTarTime(t *testing.T) {
+	dir, err := ioutil.TempDir("", "test-tar-time.")
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer os.RemoveAll(dir)
+
+	// This is the format of time from FreeBSD
+	spec := `
+/set type=file time=5.454353132
+.               type=dir time=5.123456789
+    file       time=5.911134111
+..
+`
+
+	fh, err := os.Create(filepath.Join(dir, "file"))
+	if err != nil {
+		t.Fatal(err)
+	}
+	// This is what mode we're checking for. Round integer of epoch seconds
+	epoch := time.Unix(5, 0)
+	if err := os.Chtimes(fh.Name(), epoch, epoch); err != nil {
+		t.Fatal(err)
+	}
+	if err := os.Chtimes(dir, epoch, epoch); err != nil {
+		t.Fatal(err)
+	}
+	if err := fh.Close(); err != nil {
+		t.Error(err)
+	}
+
+	dh, err := ParseSpec(bytes.NewBufferString(spec))
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	// make sure "time" keyword works
+	_, err = Check(dir, dh, DefaultKeywords)
+	if err != nil {
+		t.Error(err)
+	}
+
+	// make sure tar_time wins
+	res, err := Check(dir, dh, append(DefaultKeywords, "tar_time"))
+	if err != nil {
+		t.Error(err)
+	}
+	if len(res.Failures) > 0 {
+		t.Fatal(res.Failures)
+	}
+}
+
 func TestIgnoreComments(t *testing.T) {
 	dir, err := ioutil.TempDir("", "test-comments.")
 	if err != nil {

cmd/gomtree/main.go

@@ -5,6 +5,8 @@ import (
 	"encoding/json"
 	"flag"
 	"fmt"
+	"io"
+	"io/ioutil"
 	"log"
 	"os"
 	"strings"
@@ -20,6 +22,7 @@ var (
 	flUseKeywords  = flag.String("k", "", "Use the specified (delimited by comma or space) keywords as the current set of keywords")
 	flListKeywords = flag.Bool("list-keywords", false, "List the keywords available")
 	flResultFormat = flag.String("result-format", "bsd", "output the validation results using the given format (bsd, json, path)")
+	flTar          = flag.String("T", "", "use tar archive to create or validate a directory hierarchy spec")
 )
 
 var formats = map[string]func(*mtree.Result) string{
@@ -91,7 +94,11 @@ func main() {
 			currentKeywords = append([]string{"type"}, currentKeywords...)
 		}
 	} else {
-		currentKeywords = mtree.DefaultKeywords[:]
+		if *flTar != "" {
+			currentKeywords = mtree.DefaultTarKeywords[:]
+		} else {
+			currentKeywords = mtree.DefaultKeywords[:]
+		}
 	}
 	// -K <keywords>
 	if *flAddKeywords != "" {
@@ -123,19 +130,60 @@ func main() {
 		rootPath = *flPath
 	}
 
-	// -c
-	if *flCreate {
-		// create a directory hierarchy
-		dh, err := mtree.Walk(rootPath, nil, currentKeywords)
+	// -T <tar file>
+	var tdh *mtree.DirectoryHierarchy
+	if *flTar != "" {
+		fh, err := os.Open(*flTar)
 		if err != nil {
 			log.Println(err)
 			isErr = true
 			return
 		}
-		dh.WriteTo(os.Stdout)
-	} else if dh != nil {
+		ts := mtree.NewTarStreamer(fh, currentKeywords)
+
+		if _, err := io.Copy(ioutil.Discard, ts); err != nil && err != io.EOF {
+			log.Println(err)
+			isErr = true
+			return
+		}
+		if err := ts.Close(); err != nil {
+			log.Println(err)
+			isErr = true
+			return
+		}
+		defer fh.Close()
+		tdh, err = ts.Hierarchy()
+		if err != nil {
+			log.Println(err)
+			isErr = true
+			return
+		}
+	}
+	// -c
+	if *flCreate {
+		// create a directory hierarchy
+		// with a tar stream
+		if tdh != nil {
+			tdh.WriteTo(os.Stdout)
+		} else {
+			// with a root directory
+			dh, err := mtree.Walk(rootPath, nil, currentKeywords)
+			if err != nil {
+				log.Println(err)
+				isErr = true
+				return
+			}
+			dh.WriteTo(os.Stdout)
+		}
+	} else if tdh != nil || dh != nil {
+		var res *mtree.Result
+		var err error
 		// else this is a validation
-		res, err := mtree.Check(rootPath, dh, currentKeywords)
+		if *flTar != "" {
+			res, err = mtree.TarCheck(tdh, dh, currentKeywords)
+		} else {
+			res, err = mtree.Check(rootPath, dh, currentKeywords)
+		}
 		if err != nil {
 			log.Println(err)
 			isErr = true
@@ -150,11 +198,25 @@ func main() {
 				return
 			}
 		}
-	} else {
-		log.Println("neither validating or creating a manifest. Please provide additional arguments")
-		isErr = true
-		defer os.Exit(1)
-		return
+		if res != nil {
+			if len(res.Extra) > 0 {
+				defer os.Exit(1)
+				for _, extra := range res.Extra {
+					fmt.Printf("%s extra\n", extra.Path())
+				}
+			}
+			if len(res.Missing) > 0 {
+				defer os.Exit(1)
+				for _, missing := range res.Missing {
+					fmt.Printf("%s missing\n", missing.Path())
+				}
+			}
+		} else {
+			log.Println("neither validating or creating a manifest. Please provide additional arguments")
+			isErr = true
+			defer os.Exit(1)
+			return
+		}
 	}
 }
 

creator.go

@@ -0,0 +1,9 @@
+package mtree
+
+// dhCreator is used in when building a DirectoryHierarchy
+type dhCreator struct {
+	DH     *DirectoryHierarchy
+	curSet *Entry
+	curDir *Entry
+	curEnt *Entry
+}

entry.go

@@ -0,0 +1,102 @@
+package mtree
+
+import (
+	"fmt"
+	"path/filepath"
+	"strings"
+)
+
+type byPos []Entry
+
+func (bp byPos) Len() int           { return len(bp) }
+func (bp byPos) Less(i, j int) bool { return bp[i].Pos < bp[j].Pos }
+func (bp byPos) Swap(i, j int)      { bp[i], bp[j] = bp[j], bp[i] }
+
+// Entry is each component of content in the mtree spec file
+type Entry struct {
+	Parent     *Entry   // up
+	Children   []*Entry // down
+	Prev, Next *Entry   // left, right
+	Set        *Entry   // current `/set` for additional keywords
+	Pos        int      // order in the spec
+	Raw        string   // file or directory name
+	Name       string   // file or directory name
+	Keywords   []string // TODO(vbatts) maybe a keyword typed set of values?
+	Type       EntryType
+}
+
+// Descend searches thru an Entry's children to find the Entry associated with
+// `filename`. Directories are stored at the end of an Entry's children so do a
+// traverse backwards. If you descend to a "."
+func (e Entry) Descend(filename string) *Entry {
+	if filename == "." || filename == "" {
+		return &e
+	}
+	numChildren := len(e.Children)
+	for i := range e.Children {
+		c := e.Children[numChildren-1-i]
+		if c.Name == filename {
+			return c
+		}
+	}
+	return nil
+}
+
+// Ascend gets the parent of an Entry. Serves mainly to maintain readability
+// when traversing up and down an Entry tree
+func (e Entry) Ascend() *Entry {
+	return e.Parent
+}
+
+// Path provides the full path of the file, despite RelativeType or FullType
+func (e Entry) Path() string {
+	if e.Parent == nil || e.Type == FullType {
+		return filepath.Clean(e.Name)
+	}
+	return filepath.Clean(filepath.Join(e.Parent.Path(), e.Name))
+}
+
+func (e Entry) String() string {
+	if e.Raw != "" {
+		return e.Raw
+	}
+	if e.Type == BlankType {
+		return ""
+	}
+	if e.Type == DotDotType {
+		return e.Name
+	}
+	if e.Type == SpecialType || e.Type == FullType || inSlice("type=dir", e.Keywords) {
+		return fmt.Sprintf("%s %s", e.Name, strings.Join(e.Keywords, " "))
+	}
+	return fmt.Sprintf("    %s %s", e.Name, strings.Join(e.Keywords, " "))
+}
+
+// EntryType are the formats of lines in an mtree spec file
+type EntryType int
+
+// The types of lines to be found in an mtree spec file
+const (
+	SignatureType EntryType = iota // first line of the file, like `#mtree v2.0`
+	BlankType                      // blank lines are ignored
+	CommentType                    // Lines beginning with `#` are ignored
+	SpecialType                    // line that has `/` prefix issue a "special" command (currently only /set and /unset)
+	RelativeType                   // if the first white-space delimited word does not have a '/' in it. Options/keywords are applied.
+	DotDotType                     // .. - A relative path step. keywords/options are ignored
+	FullType                       // if the first word on the line has a `/` after the first character, it interpretted as a file pathname with options
+)
+
+// String returns the name of the EntryType
+func (et EntryType) String() string {
+	return typeNames[et]
+}
+
+var typeNames = map[EntryType]string{
+	SignatureType: "SignatureType",
+	BlankType:     "BlankType",
+	CommentType:   "CommentType",
+	SpecialType:   "SpecialType",
+	RelativeType:  "RelativeType",
+	DotDotType:    "DotDotType",
+	FullType:      "FullType",
+}

hierarchy.go

@@ -1,11 +1,8 @@
 package mtree
 
 import (
-	"fmt"
 	"io"
-	"path/filepath"
 	"sort"
-	"strings"
 )
 
 // DirectoryHierarchy is the mapped structure for an mtree directory hierarchy
@@ -27,75 +24,3 @@ func (dh DirectoryHierarchy) WriteTo(w io.Writer) (n int64, err error) {
 	}
 	return sum, nil
 }
-
-type byPos []Entry
-
-func (bp byPos) Len() int           { return len(bp) }
-func (bp byPos) Less(i, j int) bool { return bp[i].Pos < bp[j].Pos }
-func (bp byPos) Swap(i, j int)      { bp[i], bp[j] = bp[j], bp[i] }
-
-// Entry is each component of content in the mtree spec file
-type Entry struct {
-	Parent, Child *Entry   // up, down
-	Prev, Next    *Entry   // left, right
-	Set           *Entry   // current `/set` for additional keywords
-	Pos           int      // order in the spec
-	Raw           string   // file or directory name
-	Name          string   // file or directory name
-	Keywords      []string // TODO(vbatts) maybe a keyword typed set of values?
-	Type          EntryType
-}
-
-// Path provides the full path of the file, despite RelativeType or FullType
-func (e Entry) Path() string {
-	if e.Parent == nil || e.Type == FullType {
-		return filepath.Clean(e.Name)
-	}
-	return filepath.Clean(filepath.Join(e.Parent.Path(), e.Name))
-}
-
-func (e Entry) String() string {
-	if e.Raw != "" {
-		return e.Raw
-	}
-	if e.Type == BlankType {
-		return ""
-	}
-	if e.Type == DotDotType {
-		return e.Name
-	}
-	// TODO(vbatts) if type is RelativeType and a keyword of not type=dir
-	if e.Type == SpecialType || e.Type == FullType || inSlice("type=dir", e.Keywords) {
-		return fmt.Sprintf("%s %s", e.Name, strings.Join(e.Keywords, " "))
-	}
-	return fmt.Sprintf("    %s %s", e.Name, strings.Join(e.Keywords, " "))
-}
-
-// EntryType are the formats of lines in an mtree spec file
-type EntryType int
-
-// The types of lines to be found in an mtree spec file
-const (
-	SignatureType EntryType = iota // first line of the file, like `#mtree v2.0`
-	BlankType                      // blank lines are ignored
-	CommentType                    // Lines beginning with `#` are ignored
-	SpecialType                    // line that has `/` prefix issue a "special" command (currently only /set and /unset)
-	RelativeType                   // if the first white-space delimited word does not have a '/' in it. Options/keywords are applied.
-	DotDotType                     // .. - A relative path step. keywords/options are ignored
-	FullType                       // if the first word on the line has a `/` after the first character, it interpretted as a file pathname with options
-)
-
-// String returns the name of the EntryType
-func (et EntryType) String() string {
-	return typeNames[et]
-}
-
-var typeNames = map[EntryType]string{
-	SignatureType: "SignatureType",
-	BlankType:     "BlankType",
-	CommentType:   "CommentType",
-	SpecialType:   "SpecialType",
-	RelativeType:  "RelativeType",
-	DotDotType:    "DotDotType",
-	FullType:      "FullType",
-}

keywords.go

@@ -1,6 +1,7 @@
 package mtree
 
 import (
+	"archive/tar"
 	"crypto/md5"
 	"crypto/sha1"
 	"crypto/sha256"
@@ -17,7 +18,10 @@ import (
 // KeywordFunc is the type of a function called on each file to be included in
 // a DirectoryHierarchy, that will produce the string output of the keyword to
 // be included for the file entry. Otherwise, empty string.
-type KeywordFunc func(path string, info os.FileInfo) (string, error)
+// io.Reader `r` is to the file stream for the file payload. While this
+// function takes an io.Reader, the caller needs to reset it to the beginning
+// for each new KeywordFunc
+type KeywordFunc func(path string, info os.FileInfo, r io.Reader) (string, error)
 
 // KeyVal is a "keyword=value"
 type KeyVal string
@@ -55,6 +59,11 @@ func (kv KeyVal) Value() string {
 	return strings.SplitN(strings.TrimSpace(string(kv)), "=", 2)[1]
 }
 
+// ChangeValue changes the value of a KeyVal
+func (kv KeyVal) ChangeValue(newval string) string {
+	return fmt.Sprintf("%s=%s", kv.Keyword(), newval)
+}
+
 // keywordSelector takes an array of "keyword=value" and filters out that only the set of words
 func keywordSelector(keyval, words []string) []string {
 	retList := []string{}
@@ -125,6 +134,17 @@ var (
 		"nlink",
 		"time",
 	}
+	// DefaultTarKeywords has keywords that should be used when creating a manifest from
+	// an archive. Currently, evaluating the # of hardlinks has not been implemented yet
+	DefaultTarKeywords = []string{
+		"size",
+		"type",
+		"uid",
+		"gid",
+		"mode",
+		"link",
+		"tar_time",
+	}
 	// SetKeywords is the default set of keywords calculated for a `/set` SpecialType
 	SetKeywords = []string{
 		"uid",
@@ -156,6 +176,11 @@ var (
 		"sha512":          hasherKeywordFunc("sha512", sha512.New),             // The SHA512 message digest of the file
 		"sha512digest":    hasherKeywordFunc("sha512digest", sha512.New),       // A synonym for `sha512`
 
+		// This is not an upstreamed keyword, but used to vary from "time", as tar
+		// archives do not store nanosecond precision. So comparing on "time" will
+		// be only seconds level accurate.
+		"tar_time": tartimeKeywordFunc, // The last modification time of the file, from a tar archive mtime
+
 		// This is not an upstreamed keyword, but a needed attribute for file validation.
 		// The pattern for this keyword key is prefixed by "xattr." followed by the extended attribute "namespace.key".
 		// The keyword value is the SHA1 digest of the extended attribute's value.
@@ -165,7 +190,7 @@ var (
 )
 
 var (
-	modeKeywordFunc = func(path string, info os.FileInfo) (string, error) {
+	modeKeywordFunc = func(path string, info os.FileInfo, r io.Reader) (string, error) {
 		permissions := info.Mode().Perm()
 		if os.ModeSetuid&info.Mode() > 0 {
 			permissions |= (1 << 11)
@@ -178,52 +203,49 @@ var (
 		}
 		return fmt.Sprintf("mode=%#o", permissions), nil
 	}
-	sizeKeywordFunc = func(path string, info os.FileInfo) (string, error) {
+	sizeKeywordFunc = func(path string, info os.FileInfo, r io.Reader) (string, error) {
 		return fmt.Sprintf("size=%d", info.Size()), nil
 	}
-	cksumKeywordFunc = func(path string, info os.FileInfo) (string, error) {
+	cksumKeywordFunc = func(path string, info os.FileInfo, r io.Reader) (string, error) {
 		if !info.Mode().IsRegular() {
 			return "", nil
 		}
-
-		fh, err := os.Open(path)
-		if err != nil {
-			return "", err
-		}
-		defer fh.Close()
-		sum, _, err := cksum(fh)
+		sum, _, err := cksum(r)
 		if err != nil {
 			return "", err
 		}
 		return fmt.Sprintf("cksum=%d", sum), nil
 	}
 	hasherKeywordFunc = func(name string, newHash func() hash.Hash) KeywordFunc {
-		return func(path string, info os.FileInfo) (string, error) {
+		return func(path string, info os.FileInfo, r io.Reader) (string, error) {
 			if !info.Mode().IsRegular() {
 				return "", nil
 			}
-
-			fh, err := os.Open(path)
-			if err != nil {
-				return "", err
-			}
-			defer fh.Close()
-
 			h := newHash()
-			if _, err := io.Copy(h, fh); err != nil {
+			if _, err := io.Copy(h, r); err != nil {
 				return "", err
 			}
 			return fmt.Sprintf("%s=%x", name, h.Sum(nil)), nil
 		}
 	}
-	timeKeywordFunc = func(path string, info os.FileInfo) (string, error) {
+	tartimeKeywordFunc = func(path string, info os.FileInfo, r io.Reader) (string, error) {
+		return fmt.Sprintf("tar_time=%d.000000000", info.ModTime().Unix()), nil
+	}
+	timeKeywordFunc = func(path string, info os.FileInfo, r io.Reader) (string, error) {
 		t := info.ModTime().UnixNano()
 		if t == 0 {
 			return "time=0.000000000", nil
 		}
 		return fmt.Sprintf("time=%d.%9.9d", (t / 1e9), (t % (t / 1e9))), nil
 	}
-	linkKeywordFunc = func(path string, info os.FileInfo) (string, error) {
+	linkKeywordFunc = func(path string, info os.FileInfo, r io.Reader) (string, error) {
+		if sys, ok := info.Sys().(*tar.Header); ok {
+			if sys.Linkname != "" {
+				return fmt.Sprintf("link=%s", sys.Linkname), nil
+			}
+			return "", nil
+		}
+
 		if info.Mode()&os.ModeSymlink != 0 {
 			str, err := os.Readlink(path)
 			if err != nil {
@@ -233,7 +255,7 @@ var (
 		}
 		return "", nil
 	}
-	typeKeywordFunc = func(path string, info os.FileInfo) (string, error) {
+	typeKeywordFunc = func(path string, info os.FileInfo, r io.Reader) (string, error) {
 		if info.Mode().IsDir() {
 			return "type=dir", nil
 		}

keywords_linux.go

@@ -3,8 +3,10 @@
 package mtree
 
 import (
+	"archive/tar"
 	"crypto/sha1"
 	"fmt"
+	"io"
 	"os"
 	"os/user"
 	"strings"
@@ -14,7 +16,11 @@ import (
 )
 
 var (
-	unameKeywordFunc = func(path string, info os.FileInfo) (string, error) {
+	unameKeywordFunc = func(path string, info os.FileInfo, r io.Reader) (string, error) {
+		if hdr, ok := info.Sys().(*tar.Header); ok {
+			return fmt.Sprintf("uname=%s", hdr.Uname), nil
+		}
+
 		stat := info.Sys().(*syscall.Stat_t)
 		u, err := user.LookupId(fmt.Sprintf("%d", stat.Uid))
 		if err != nil {
@@ -22,19 +28,40 @@ var (
 		}
 		return fmt.Sprintf("uname=%s", u.Username), nil
 	}
-	uidKeywordFunc = func(path string, info os.FileInfo) (string, error) {
+	uidKeywordFunc = func(path string, info os.FileInfo, r io.Reader) (string, error) {
+		if hdr, ok := info.Sys().(*tar.Header); ok {
+			return fmt.Sprintf("uid=%d", hdr.Uid), nil
+		}
 		stat := info.Sys().(*syscall.Stat_t)
 		return fmt.Sprintf("uid=%d", stat.Uid), nil
 	}
-	gidKeywordFunc = func(path string, info os.FileInfo) (string, error) {
-		stat := info.Sys().(*syscall.Stat_t)
-		return fmt.Sprintf("gid=%d", stat.Gid), nil
+	gidKeywordFunc = func(path string, info os.FileInfo, r io.Reader) (string, error) {
+		if hdr, ok := info.Sys().(*tar.Header); ok {
+			return fmt.Sprintf("gid=%d", hdr.Gid), nil
+		}
+		if stat, ok := info.Sys().(*syscall.Stat_t); ok {
+			return fmt.Sprintf("gid=%d", stat.Gid), nil
+		}
+		return "", nil
 	}
-	nlinkKeywordFunc = func(path string, info os.FileInfo) (string, error) {
-		stat := info.Sys().(*syscall.Stat_t)
-		return fmt.Sprintf("nlink=%d", stat.Nlink), nil
+	nlinkKeywordFunc = func(path string, info os.FileInfo, r io.Reader) (string, error) {
+		if stat, ok := info.Sys().(*syscall.Stat_t); ok {
+			return fmt.Sprintf("nlink=%d", stat.Nlink), nil
+		}
+		return "", nil
 	}
-	xattrKeywordFunc = func(path string, info os.FileInfo) (string, error) {
+	xattrKeywordFunc = func(path string, info os.FileInfo, r io.Reader) (string, error) {
+		if hdr, ok := info.Sys().(*tar.Header); ok {
+			if len(hdr.Xattrs) == 0 {
+				return "", nil
+			}
+			klist := []string{}
+			for k, v := range hdr.Xattrs {
+				klist = append(klist, fmt.Sprintf("xattr.%s=%x", k, sha1.Sum([]byte(v))))
+			}
+			return strings.Join(klist, " "), nil
+		}
+
 		xlist, err := xattr.List(path)
 		if err != nil {
 			return "", err

keywords_unsupported.go

@@ -2,22 +2,36 @@
 
 package mtree
 
-import "os"
+import (
+	"archive/tar"
+	"fmt"
+	"io"
+	"os"
+)
 
 var (
-	unameKeywordFunc = func(path string, info os.FileInfo) (string, error) {
+	unameKeywordFunc = func(path string, info os.FileInfo, r io.Reader) (string, error) {
+		if hdr, ok := info.Sys().(*tar.Header); ok {
+			return fmt.Sprintf("uname=%s", hdr.Uname), nil
+		}
 		return "", nil
 	}
-	uidKeywordFunc = func(path string, info os.FileInfo) (string, error) {
+	uidKeywordFunc = func(path string, info os.FileInfo, r io.Reader) (string, error) {
+		if hdr, ok := info.Sys().(*tar.Header); ok {
+			return fmt.Sprintf("uid=%d", hdr.Uid), nil
+		}
 		return "", nil
 	}
-	gidKeywordFunc = func(path string, info os.FileInfo) (string, error) {
+	gidKeywordFunc = func(path string, info os.FileInfo, r io.Reader) (string, error) {
+		if hdr, ok := info.Sys().(*tar.Header); ok {
+			return fmt.Sprintf("gid=%d", hdr.Gid), nil
+		}
 		return "", nil
 	}
-	nlinkKeywordFunc = func(path string, info os.FileInfo) (string, error) {
+	nlinkKeywordFunc = func(path string, info os.FileInfo, r io.Reader) (string, error) {
 		return "", nil
 	}
-	xattrKeywordFunc = func(path string, info os.FileInfo) (string, error) {
+	xattrKeywordFunc = func(path string, info os.FileInfo, r io.Reader) (string, error) {
 		return "", nil
 	}
 )

parse.go

@@ -81,6 +81,7 @@ func ParseSpec(r io.Reader) (*DirectoryHierarchy, error) {
 				e.Type = RelativeType
 			}
 			e.Keywords = f[1:]
+			// TODO: gather keywords if using tar stream
 			e.Parent = creator.curDir
 			for i := range e.Keywords {
 				kv := KeyVal(e.Keywords[i])

tar.go

@@ -0,0 +1,376 @@
+package mtree
+
+import (
+	"archive/tar"
+	"io"
+	"io/ioutil"
+	"os"
+	"path/filepath"
+	"strings"
+)
+
+// Streamer creates a file hierarchy out of a tar stream
+type Streamer interface {
+	io.ReadCloser
+	Hierarchy() (*DirectoryHierarchy, error)
+}
+
+var tarDefaultSetKeywords = []string{"type=file", "flags=none", "mode=0664"}
+
+// NewTarStreamer streams a tar archive and creates a file hierarchy based off
+// of the tar metadata headers
+func NewTarStreamer(r io.Reader, keywords []string) Streamer {
+	pR, pW := io.Pipe()
+	ts := &tarStream{
+		pipeReader: pR,
+		pipeWriter: pW,
+		creator:    dhCreator{DH: &DirectoryHierarchy{}},
+		teeReader:  io.TeeReader(r, pW),
+		tarReader:  tar.NewReader(pR),
+		keywords:   keywords,
+	}
+	go ts.readHeaders() // I don't like this
+	return ts
+}
+
+type tarStream struct {
+	creator    dhCreator
+	pipeReader *io.PipeReader
+	pipeWriter *io.PipeWriter
+	teeReader  io.Reader
+	tarReader  *tar.Reader
+	keywords   []string
+	err        error
+}
+
+func (ts *tarStream) readHeaders() {
+	// We have to start with the directory we're in, and anything beyond these
+	// items is determined at the time a tar is extracted.
+	rootComment := Entry{
+		Raw:  "# .",
+		Type: CommentType,
+	}
+	root := Entry{
+		Name: ".",
+		Type: RelativeType,
+		Prev: &rootComment,
+		Set: &Entry{
+			Name: "meta-set",
+			Type: SpecialType,
+		},
+	}
+	metadataEntries := signatureEntries("<user specified tar archive>")
+	for _, e := range metadataEntries {
+		e.Pos = len(ts.creator.DH.Entries)
+		ts.creator.DH.Entries = append(ts.creator.DH.Entries, e)
+	}
+	for {
+		hdr, err := ts.tarReader.Next()
+		if err != nil {
+			flatten(&root, ts)
+			ts.pipeReader.CloseWithError(err)
+			return
+		}
+
+		// Because the content of the file may need to be read by several
+		// KeywordFuncs, it needs to be an io.Seeker as well. So, just reading from
+		// ts.tarReader is not enough.
+		tmpFile, err := ioutil.TempFile("", "ts.payload.")
+		if err != nil {
+			ts.pipeReader.CloseWithError(err)
+			return
+		}
+		// for good measure
+		if err := tmpFile.Chmod(0600); err != nil {
+			tmpFile.Close()
+			os.Remove(tmpFile.Name())
+			ts.pipeReader.CloseWithError(err)
+			return
+		}
+		if _, err := io.Copy(tmpFile, ts.tarReader); err != nil {
+			tmpFile.Close()
+			os.Remove(tmpFile.Name())
+			ts.pipeReader.CloseWithError(err)
+			return
+		}
+		defer tmpFile.Close()
+		defer os.Remove(tmpFile.Name())
+
+		// Alright, it's either file or directory
+		e := Entry{
+			Name: filepath.Base(hdr.Name),
+			Type: RelativeType,
+		}
+
+		// now collect keywords on the file
+		for _, keyword := range ts.keywords {
+			if keyword == "time" {
+				keyword = "tar_time"
+			}
+			if keyFunc, ok := KeywordFuncs[keyword]; ok {
+				// We can't extract directories on to disk, so "size" keyword
+				// is irrelevant for now
+				if hdr.FileInfo().IsDir() && keyword == "size" {
+					continue
+				}
+
+				if string(hdr.Typeflag) == string('1') {
+					// TODO: get number of hardlinks for a file
+				}
+				val, err := keyFunc(hdr.Name, hdr.FileInfo(), tmpFile)
+				if err != nil {
+					ts.setErr(err)
+				}
+				// for good measure, check that we actually get a value for a keyword
+				if val != "" {
+					e.Keywords = append(e.Keywords, val)
+				}
+
+				// don't forget to reset the reader
+				if _, err := tmpFile.Seek(0, 0); err != nil {
+					tmpFile.Close()
+					os.Remove(tmpFile.Name())
+					ts.pipeReader.CloseWithError(err)
+					return
+				}
+			}
+		}
+		// collect meta-set keywords for a directory so that we can build the
+		// actual sets in `flatten`
+		if hdr.FileInfo().IsDir() {
+			s := Entry{
+				Name: "meta-set",
+				Type: SpecialType,
+			}
+			for _, setKW := range SetKeywords {
+				if setKW == "time" {
+					setKW = "tar_time"
+				}
+				if keyFunc, ok := KeywordFuncs[setKW]; ok {
+					val, err := keyFunc(hdr.Name, hdr.FileInfo(), tmpFile)
+					if err != nil {
+						ts.setErr(err)
+					}
+					if val != "" {
+						s.Keywords = append(s.Keywords, val)
+					}
+					if _, err := tmpFile.Seek(0, 0); err != nil {
+						tmpFile.Close()
+						os.Remove(tmpFile.Name())
+						ts.pipeReader.CloseWithError(err)
+					}
+				}
+			}
+			if filepath.Dir(filepath.Clean(hdr.Name)) == "." {
+				root.Set = &s
+			} else {
+				e.Set = &s
+			}
+		}
+		populateTree(&root, &e, hdr, ts)
+	}
+}
+
+type relationship int
+
+const (
+	unknownDir relationship = iota
+	sameDir
+	childDir
+	parentDir
+)
+
+// populateTree creates a file tree hierarchy using an Entry's Parent and
+// Children fields. When examining the Entry e to insert in the tree, we
+// determine if the path to that Entry exists yet. If it does, insert it in the
+// appropriate position in the tree. If not, create a path with "placeholder"
+// directories, and then insert the Entry. populateTree does not consider
+// symbolic links yet.
+func populateTree(root, e *Entry, hdr *tar.Header, ts *tarStream) {
+	isDir := hdr.FileInfo().IsDir()
+	wd := filepath.Clean(hdr.Name)
+
+	if !isDir {
+		// If entry is a file, we only want the directory it's in.
+		wd = filepath.Dir(wd)
+	}
+	if filepath.Dir(wd) == "." {
+		if isDir {
+			root.Keywords = e.Keywords
+		} else {
+			root.Children = append([]*Entry{e}, root.Children...)
+			e.Parent = root
+		}
+		return
+	}
+
+	dirNames := strings.Split(wd, "/")
+	parent := root
+	for _, name := range dirNames[1:] {
+		if node := parent.Descend(name); node == nil {
+			// Entry for directory doesn't exist in tree relative to root
+			var newEntry *Entry
+			if isDir {
+				newEntry = e
+			} else {
+				newEntry = &Entry{
+					Name: name,
+					Type: RelativeType,
+				}
+			}
+			newEntry.Parent = parent
+			parent.Children = append(parent.Children, newEntry)
+			parent = newEntry
+		} else {
+			// Entry for directory exists in tree, just keep going
+			parent = node
+		}
+	}
+	if !isDir {
+		parent.Children = append([]*Entry{e}, parent.Children...)
+		e.Parent = parent
+	} else {
+		commentEntry := Entry{
+			Raw:  "# " + e.Path(),
+			Type: CommentType,
+		}
+		e.Prev = &commentEntry
+	}
+}
+
+// After constructing the tree from the tar stream, we want to "flatten" this
+// tree by appending Entry's into ts.creator.DH.Entries in an appropriate
+// manner to simplify writing the output with ts.creator.DH.WriteTo
+// root: the "head" of the sub-tree to flatten
+// ts  : tarStream to keep track of Entry's
+func flatten(root *Entry, ts *tarStream) {
+	if root.Prev != nil {
+		// root.Prev != nil implies root is a directory
+		ts.creator.DH.Entries = append(ts.creator.DH.Entries,
+			Entry{
+				Type: BlankType,
+				Pos:  len(ts.creator.DH.Entries),
+			})
+		root.Prev.Pos = len(ts.creator.DH.Entries)
+		ts.creator.DH.Entries = append(ts.creator.DH.Entries, *root.Prev)
+
+		// Check if we need a new set
+		if ts.creator.curSet == nil {
+			ts.creator.curSet = &Entry{
+				Type:     SpecialType,
+				Name:     "/set",
+				Keywords: keywordSelector(append(tarDefaultSetKeywords, root.Set.Keywords...), ts.keywords),
+				Pos:      len(ts.creator.DH.Entries),
+			}
+			ts.creator.DH.Entries = append(ts.creator.DH.Entries, *ts.creator.curSet)
+		} else {
+			needNewSet := false
+			for _, k := range root.Set.Keywords {
+				if !inSlice(k, ts.creator.curSet.Keywords) {
+					needNewSet = true
+					break
+				}
+			}
+			if needNewSet {
+				ts.creator.curSet = &Entry{
+					Name:     "/set",
+					Type:     SpecialType,
+					Pos:      len(ts.creator.DH.Entries),
+					Keywords: keywordSelector(append(tarDefaultSetKeywords, root.Set.Keywords...), ts.keywords),
+				}
+				ts.creator.DH.Entries = append(ts.creator.DH.Entries, *ts.creator.curSet)
+			}
+		}
+	}
+	root.Set = ts.creator.curSet
+	root.Keywords = setDifference(root.Keywords, ts.creator.curSet.Keywords)
+	root.Pos = len(ts.creator.DH.Entries)
+	ts.creator.DH.Entries = append(ts.creator.DH.Entries, *root)
+
+	for _, c := range root.Children {
+		flatten(c, ts)
+	}
+
+	if root.Prev != nil {
+		// Show a comment when stepping out
+		root.Prev.Pos = len(ts.creator.DH.Entries)
+		ts.creator.DH.Entries = append(ts.creator.DH.Entries, *root.Prev)
+		dotEntry := Entry{
+			Type: DotDotType,
+			Name: "..",
+			Pos:  len(ts.creator.DH.Entries),
+		}
+		ts.creator.DH.Entries = append(ts.creator.DH.Entries, dotEntry)
+	}
+}
+
+// filter takes in a pointer to an Entry, and returns a slice of Entry's that
+// satisfy the predicate p
+func filter(root *Entry, p func(*Entry) bool) []Entry {
+	var validEntrys []Entry
+	if len(root.Children) > 0 || root.Prev != nil {
+		for _, c := range root.Children {
+			// if an Entry is a directory, filter the directory
+			if c.Prev != nil {
+				validEntrys = append(validEntrys, filter(c, p)...)
+			}
+			if p(c) {
+				if c.Prev == nil {
+					// prepend directories
+					validEntrys = append([]Entry{*c}, validEntrys...)
+				} else {
+					validEntrys = append(validEntrys, *c)
+				}
+			}
+		}
+		return validEntrys
+	}
+	return nil
+}
+
+func setDifference(this, that []string) []string {
+	if len(this) == 0 {
+		return that
+	}
+	diff := []string{}
+	for _, kv := range this {
+		if !inSlice(kv, that) {
+			diff = append(diff, kv)
+		}
+	}
+	return diff
+}
+
+func compareDir(curDir, prevDir string) relationship {
+	curDir = filepath.Clean(curDir)
+	prevDir = filepath.Clean(prevDir)
+	if curDir == prevDir {
+		return sameDir
+	}
+	if filepath.Dir(curDir) == prevDir {
+		return childDir
+	}
+	if curDir == filepath.Dir(prevDir) {
+		return parentDir
+	}
+	return unknownDir
+}
+
+func (ts *tarStream) setErr(err error) {
+	ts.err = err
+}
+
+func (ts *tarStream) Read(p []byte) (n int, err error) {
+	return ts.teeReader.Read(p)
+}
+
+func (ts *tarStream) Close() error {
+	return ts.pipeReader.Close()
+}
+
+func (ts *tarStream) Hierarchy() (*DirectoryHierarchy, error) {
+	if ts.err != nil && ts.err != io.EOF {
+		return nil, ts.err
+	}
+	return ts.creator.DH, nil
+}

tar_test.go

@@ -0,0 +1,175 @@
+package mtree
+
+import (
+	"archive/tar"
+	"bytes"
+	"io"
+	"io/ioutil"
+	"os"
+	"testing"
+)
+
+func ExampleStreamer() {
+	fh, err := os.Open("./testdata/test.tar")
+	if err != nil {
+		// handle error ...
+	}
+	str := NewTarStreamer(fh, nil)
+	if err := extractTar("/tmp/dir", str); err != nil {
+		// handle error ...
+	}
+
+	dh, err := str.Hierarchy()
+	if err != nil {
+		// handle error ...
+	}
+
+	res, err := Check("/tmp/dir/", dh, nil)
+	if err != nil {
+		// handle error ...
+	}
+	if len(res.Failures) > 0 {
+		// handle validation issue ...
+	}
+}
+func extractTar(root string, tr io.Reader) error {
+	return nil
+}
+
+func TestTar(t *testing.T) {
+	/*
+		data, err := makeTarStream()
+		if err != nil {
+			t.Fatal(err)
+		}
+		buf := bytes.NewBuffer(data)
+		str := NewTarStreamer(buf, append(DefaultKeywords, "sha1"))
+	*/
+	/*
+		// open empty folder and check size.
+		fh, err := os.Open("./testdata/empty")
+		if err != nil {
+			t.Fatal(err)
+		}
+		log.Println(fh.Stat())
+		fh.Close() */
+	fh, err := os.Open("./testdata/test.tar")
+	if err != nil {
+		t.Fatal(err)
+	}
+	str := NewTarStreamer(fh, append(DefaultKeywords, "sha1"))
+
+	if _, err := io.Copy(ioutil.Discard, str); err != nil && err != io.EOF {
+		t.Fatal(err)
+	}
+	if err := str.Close(); err != nil {
+		t.Fatal(err)
+	}
+	defer fh.Close()
+
+	// get DirectoryHierarcy struct from walking the tar archive
+	tdh, err := str.Hierarchy()
+	if err != nil {
+		t.Fatal(err)
+	}
+	if tdh == nil {
+		t.Fatal("expected a DirectoryHierarchy struct, but got nil")
+	}
+
+	fh, err = os.Create("./testdata/test.mtree")
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer os.Remove("./testdata/test.mtree")
+
+	// put output of tar walk into test.mtree
+	_, err = tdh.WriteTo(fh)
+	if err != nil {
+		t.Fatal(err)
+	}
+	fh.Close()
+
+	// now simulate gomtree -T testdata/test.tar -f testdata/test.mtree
+	fh, err = os.Open("./testdata/test.mtree")
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer fh.Close()
+
+	dh, err := ParseSpec(fh)
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	res, err := TarCheck(tdh, dh, append(DefaultKeywords, "sha1"))
+
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	// print any failures, and then call t.Fatal once all failures/extra/missing
+	// are outputted
+	if res != nil {
+		errors := ""
+		switch {
+		case len(res.Failures) > 0:
+			for _, f := range res.Failures {
+				t.Errorf("%s\n", f)
+			}
+			errors += "Keyword validation errors\n"
+		case len(res.Missing) > 0:
+			for _, m := range res.Missing {
+				t.Errorf("Missing file: %s\n", m.Path())
+			}
+			errors += "Missing files not expected for this test\n"
+		case len(res.Extra) > 0:
+			for _, e := range res.Extra {
+				t.Errorf("Extra file: %s\n", e.Path())
+			}
+			errors += "Extra files not expected for this test\n"
+		}
+		if errors != "" {
+			t.Fatal(errors)
+		}
+	}
+}
+
+// minimal tar archive stream that mimics what is in ./testdata/test.tar
+func makeTarStream() ([]byte, error) {
+	buf := new(bytes.Buffer)
+
+	// Create a new tar archive.
+	tw := tar.NewWriter(buf)
+
+	// Add some files to the archive.
+	var files = []struct {
+		Name, Body string
+		Mode       int64
+		Type       byte
+		Xattrs     map[string]string
+	}{
+		{"x/", "", 0755, '5', nil},
+		{"x/files", "howdy\n", 0644, '0', nil},
+	}
+	for _, file := range files {
+		hdr := &tar.Header{
+			Name:   file.Name,
+			Mode:   file.Mode,
+			Size:   int64(len(file.Body)),
+			Xattrs: file.Xattrs,
+		}
+		if err := tw.WriteHeader(hdr); err != nil {
+			return nil, err
+		}
+		if len(file.Body) > 0 {
+			if _, err := tw.Write([]byte(file.Body)); err != nil {
+				return nil, err
+			}
+		}
+	}
+	// Make sure to check the error on Close.
+	if err := tw.Close(); err != nil {
+		return nil, err
+	}
+	return buf.Bytes(), nil
+}

walk.go

@@ -2,6 +2,7 @@ package mtree
 
 import (
 	"fmt"
+	"io"
 	"os"
 	"os/user"
 	"path/filepath"
@@ -14,13 +15,6 @@ import (
 // returns true, then the path is not included in the spec.
 type ExcludeFunc func(path string, info os.FileInfo) bool
 
-type dhCreator struct {
-	DH     *DirectoryHierarchy
-	curSet *Entry
-	curDir *Entry
-	curEnt *Entry
-}
-
 var defaultSetKeywords = []string{"type=file", "nlink=1", "flags=none", "mode=0664"}
 
 // Walk from root directory and assemble the DirectoryHierarchy. excludes
@@ -76,9 +70,28 @@ func Walk(root string, exlcudes []ExcludeFunc, keywords []string) (*DirectoryHie
 					Keywords: keywordSelector(defaultSetKeywords, keywords),
 				}
 				for _, keyword := range SetKeywords {
-					if str, err := KeywordFuncs[keyword](path, info); err == nil && str != "" {
-						e.Keywords = append(e.Keywords, str)
-					} else if err != nil {
+					err := func() error {
+						var r io.Reader
+						if info.Mode().IsRegular() {
+							fh, err := os.Open(path)
+							if err != nil {
+								return err
+							}
+							defer fh.Close()
+							r = fh
+						}
+						keywordFunc, ok := KeywordFuncs[keyword]
+						if !ok {
+							return fmt.Errorf("Unknown keyword %q for file %q", keyword, path)
+						}
+						if str, err := keywordFunc(path, info, r); err == nil && str != "" {
+							e.Keywords = append(e.Keywords, str)
+						} else if err != nil {
+							return err
+						}
+						return nil
+					}()
+					if err != nil {
 						return err
 					}
 				}
@@ -88,9 +101,30 @@ func Walk(root string, exlcudes []ExcludeFunc, keywords []string) (*DirectoryHie
 				// check the attributes of the /set keywords and re-set if changed
 				klist := []string{}
 				for _, keyword := range SetKeywords {
-					if str, err := KeywordFuncs[keyword](path, info); err == nil && str != "" {
-						klist = append(klist, str)
-					} else if err != nil {
+					err := func() error {
+						var r io.Reader
+						if info.Mode().IsRegular() {
+							fh, err := os.Open(path)
+							if err != nil {
+								return err
+							}
+							defer fh.Close()
+							r = fh
+						}
+						keywordFunc, ok := KeywordFuncs[keyword]
+						if !ok {
+							return fmt.Errorf("Unknown keyword %q for file %q", keyword, path)
+						}
+						str, err := keywordFunc(path, info, r)
+						if err != nil {
+							return err
+						}
+						if str != "" {
+							klist = append(klist, str)
+						}
+						return nil
+					}()
+					if err != nil {
 						return err
 					}
 				}
@@ -122,11 +156,30 @@ func Walk(root string, exlcudes []ExcludeFunc, keywords []string) (*DirectoryHie
 			Parent: creator.curDir,
 		}
 		for _, keyword := range keywords {
-			if str, err := KeywordFuncs[keyword](path, info); err == nil && str != "" {
-				if !inSlice(str, creator.curSet.Keywords) {
+			err := func() error {
+				var r io.Reader
+				if info.Mode().IsRegular() {
+					fh, err := os.Open(path)
+					if err != nil {
+						return err
+					}
+					defer fh.Close()
+					r = fh
+				}
+				keywordFunc, ok := KeywordFuncs[keyword]
+				if !ok {
+					return fmt.Errorf("Unknown keyword %q for file %q", keyword, path)
+				}
+				str, err := keywordFunc(path, info, r)
+				if err != nil {
+					return err
+				}
+				if str != "" && !inSlice(str, creator.curSet.Keywords) {
 					e.Keywords = append(e.Keywords, str)
 				}
-			} else if err != nil {
+				return nil
+			}()
+			if err != nil {
 				return err
 			}
 		}
@@ -245,7 +298,7 @@ func readOrderedDirNames(dirname string) ([]string, error) {
 	return append(names, dirnames...), nil
 }
 
-// signatureEntries is helper function that returns a slice of Entry's
+// signatureEntries is a simple helper function that returns a slice of Entry's
 // that describe the metadata signature about the host. Items like date, user,
 // machine, and tree (which is specified by argument `root`), are considered.
 // These Entry's construct comments in the mtree specification, so if there is