textparse: Add fuzzing and fix bug caught

See https://github.com/cznic/golex/issues/11 for info on the bug

Signed-off-by: Goutham Veeramachaneni <cs14btech11014@iith.ac.in>

pkg/textparse/lex.l

@@ -23,22 +23,22 @@ import (
         "github.com/prometheus/prometheus/pkg/value"
 )
 
+const (
+    lstateInit = iota
+    lstateName
+    lstateValue
+    lstateTimestamp
+    lstateLabels
+    lstateLName
+    lstateLValue
+    lstateLValueIn
+)
 
 // Lex is called by the parser generated by "go tool yacc" to obtain each
 // token. The method is opened before the matching rules block and closed at
 // the end of the file.
 func (l *lexer) Lex() int {
-    const (
+    l.state = lstateInit
-        lstateInit = iota
-        lstateName
-        lstateValue
-        lstateTimestamp
-        lstateLabels
-        lstateLName
-        lstateLValue
-        lstateLValueIn
-    )
-    s := lstateInit
     
     if l.i >= len(l.b) {
         return eof
@@ -60,7 +60,7 @@ M     [a-zA-Z_:]
 
 %yyc c
 %yyn c = l.next()
-%yyt s
+%yyt l.state
 
 
 %%
@@ -69,29 +69,29 @@ M     [a-zA-Z_:]
 #[^\r\n]*\n                           l.mstart = l.i
 [\r\n \t]+                            l.mstart = l.i
 
-{S}({M}|{D})*                         s = lstateName
+{S}({M}|{D})*                         l.state = lstateName
                                       l.offsets = append(l.offsets, l.i)
                                       l.mend = l.i
 
-<lstateName>([ \t]*)\{                 s = lstateLabels
+<lstateName>([ \t]*)\{                l.state = lstateLabels
 
-<lstateName>[ \t]+                    s = lstateValue
+<lstateName>[ \t]+                    l.state = lstateValue
                                       l.vstart = l.i
 
 
 <lstateLabels>[ \t]+
-<lstateLabels>,?\}                    s = lstateValue
+<lstateLabels>,?\}                    l.state = lstateValue
                                       l.mend = l.i
-<lstateLabels>(,?[ \t]*)              s = lstateLName
+<lstateLabels>(,?[ \t]*)              l.state = lstateLName
                                       l.offsets = append(l.offsets, l.i)
 
 <lstateLName>{S}({L}|{D})*            l.offsets = append(l.offsets, l.i)
-<lstateLName>[ \t]*=                  s = lstateLValue
+<lstateLName>[ \t]*=                  l.state = lstateLValue
 
 <lstateLValue>[ \t]+
-<lstateLValue>\"                      s = lstateLValueIn
+<lstateLValue>\"                      l.state = lstateLValueIn
                                       l.offsets = append(l.offsets, l.i)
-<lstateLValueIn>(\\.|[^\\"]|\0)*\"    s = lstateLabels
+<lstateLValueIn>(\\.|[^\\"])*\"       l.state = lstateLabels
                                       if !utf8.Valid(l.b[l.offsets[len(l.offsets)-1]:l.i-1]) {
                                         l.err = fmt.Errorf("Invalid UTF-8 label value.")
                                         return -1
@@ -100,7 +100,7 @@ M     [a-zA-Z_:]
 
 <lstateValue>[ \t]+                   l.vstart = l.i
 <lstateValue>(NaN)                    l.val = math.Float64frombits(value.NormalNaN)
-                                      s = lstateTimestamp
+                                      l.state = lstateTimestamp
 
 <lstateValue>[^\n \t\r]+              // We don't parse strictly correct floats as the conversion
                                       // repeats the effort anyway.
@@ -108,7 +108,7 @@ M     [a-zA-Z_:]
                                       if l.err != nil {
                                           return -1
                                       }
-                                      s = lstateTimestamp
+                                      l.state = lstateTimestamp
 
 <lstateTimestamp>[ \t]+               l.tstart = l.i         
 <lstateTimestamp>{D}+                 ts, err := strconv.ParseInt(yoloString(l.b[l.tstart:l.i]), 10, 64)

pkg/textparse/lex.l.go

@@ -24,21 +24,22 @@ import (
 	"github.com/prometheus/prometheus/pkg/value"
 )
 
+const (
+	lstateInit = iota
+	lstateName
+	lstateValue
+	lstateTimestamp
+	lstateLabels
+	lstateLName
+	lstateLValue
+	lstateLValueIn
+)
+
 // Lex is called by the parser generated by "go tool yacc" to obtain each
 // token. The method is opened before the matching rules block and closed at
 // the end of the file.
 func (l *lexer) Lex() int {
-	const (
+	l.state = lstateInit
-		lstateInit = iota
-		lstateName
-		lstateValue
-		lstateTimestamp
-		lstateLabels
-		lstateLName
-		lstateLValue
-		lstateLValueIn
-	)
-	s := lstateInit
 
 	if l.i >= len(l.b) {
 		return eof
@@ -51,7 +52,7 @@ func (l *lexer) Lex() int {
 
 yystate0:
 
-	switch yyt := s; yyt {
+	switch yyt := l.state; yyt {
 	default:
 		panic(fmt.Errorf(`invalid start condition %d`, yyt))
 	case 0: // start condition: INITIAL
@@ -386,22 +387,26 @@ yystate33:
 yystart33:
 	switch {
 	default:
-		goto yystate34 // c >= '\x00' && c <= '!' || c >= '#' && c <= '[' || c >= ']' && c <= 'ÿ'
+		goto yyabort
 	case c == '"':
 		goto yystate35
 	case c == '\\':
 		goto yystate36
+	case c >= '\x01' && c <= '!' || c >= '#' && c <= '[' || c >= ']' && c <= 'ÿ':
+		goto yystate34
 	}
 
 yystate34:
 	c = l.next()
 	switch {
 	default:
-		goto yystate34 // c >= '\x00' && c <= '!' || c >= '#' && c <= '[' || c >= ']' && c <= 'ÿ'
+		goto yyabort
 	case c == '"':
 		goto yystate35
 	case c == '\\':
 		goto yystate36
+	case c >= '\x01' && c <= '!' || c >= '#' && c <= '[' || c >= ']' && c <= 'ÿ':
+		goto yystate34
 	}
 
 yystate35:
@@ -433,19 +438,19 @@ yyrule3: // [\r\n \t]+
 	}
 yyrule4: // {S}({M}|{D})*
 	{
-		s = lstateName
+		l.state = lstateName
 		l.offsets = append(l.offsets, l.i)
 		l.mend = l.i
 		goto yystate0
 	}
 yyrule5: // ([ \t]*)\{
 	{
-		s = lstateLabels
+		l.state = lstateLabels
 		goto yystate0
 	}
 yyrule6: // [ \t]+
 	{
-		s = lstateValue
+		l.state = lstateValue
 		l.vstart = l.i
 		goto yystate0
 	}
@@ -454,13 +459,13 @@ yyrule7: // [ \t]+
 	goto yystate0
 yyrule8: // ,?\}
 	{
-		s = lstateValue
+		l.state = lstateValue
 		l.mend = l.i
 		goto yystate0
 	}
 yyrule9: // (,?[ \t]*)
 	{
-		s = lstateLName
+		l.state = lstateLName
 		l.offsets = append(l.offsets, l.i)
 		goto yystate0
 	}
@@ -471,7 +476,7 @@ yyrule10: // {S}({L}|{D})*
 	}
 yyrule11: // [ \t]*=
 	{
-		s = lstateLValue
+		l.state = lstateLValue
 		goto yystate0
 	}
 yyrule12: // [ \t]+
@@ -479,13 +484,13 @@ yyrule12: // [ \t]+
 	goto yystate0
 yyrule13: // \"
 	{
-		s = lstateLValueIn
+		l.state = lstateLValueIn
 		l.offsets = append(l.offsets, l.i)
 		goto yystate0
 	}
-yyrule14: // (\\.|[^\\"]|\0)*\"
+yyrule14: // (\\.|[^\\"])*\"
 	{
-		s = lstateLabels
+		l.state = lstateLabels
 		if !utf8.Valid(l.b[l.offsets[len(l.offsets)-1] : l.i-1]) {
 			l.err = fmt.Errorf("Invalid UTF-8 label value.")
 			return -1
@@ -501,7 +506,7 @@ yyrule15: // [ \t]+
 yyrule16: // (NaN)
 	{
 		l.val = math.Float64frombits(value.NormalNaN)
-		s = lstateTimestamp
+		l.state = lstateTimestamp
 		goto yystate0
 	}
 yyrule17: // [^\n \t\r]+
@@ -512,7 +517,7 @@ yyrule17: // [^\n \t\r]+
 		if l.err != nil {
 			return -1
 		}
-		s = lstateTimestamp
+		l.state = lstateTimestamp
 		goto yystate0
 	}
 yyrule18: // [ \t]+

pkg/textparse/parse.go

@@ -38,6 +38,8 @@ type lexer struct {
 	offsets      []int
 	mstart, mend int
 	nextMstart   int
+
+	state int
 }
 
 const eof = 0
@@ -49,6 +51,11 @@ func (l *lexer) next() byte {
 		return eof
 	}
 	c := l.b[l.i]
+
+	// Consume null byte when encountered in label-value.
+	if c == eof && (l.state == lstateLValueIn || l.state == lstateLValue) {
+		return l.next()
+	}
 	return c
 }
 

pkg/textparse/parse_test.go

@@ -163,6 +163,10 @@ func TestParseErrors(t *testing.T) {
 			input: "a true\n",
 			err:   "strconv.ParseFloat: parsing \"true\": invalid syntax",
 		},
+		{
+			input: "something_weird{problem=\"",
+			err:   "no token found",
+		},
 	}
 
 	for _, c := range cases {
@@ -174,6 +178,60 @@ func TestParseErrors(t *testing.T) {
 	}
 }
 
+func TestNullByteHandling(t *testing.T) {
+	cases := []struct {
+		input string
+		err   string
+	}{
+		{
+			input: "null_byte_metric{a=\"abc\x00\"} 1",
+			err:   "",
+		},
+		{
+			input: "a{b=\"\x00ss\"} 1\n",
+			err:   "",
+		},
+		{
+			input: "a{b=\"\x00\"} 1\n",
+			err:   "",
+		},
+		{
+			input: "a{b=\"\x00\"} 1\n",
+			err:   "",
+		},
+		{
+			input: "a{b=\x00\"ssss\"} 1\n",
+			err:   "no token found",
+		},
+		{
+			input: "a{b=\"\x00",
+			err:   "no token found",
+		},
+		{
+			input: "a{b\x00=\"hiih\"}	1",
+			err: "no token found",
+		},
+		{
+			input: "a\x00{b=\"ddd\"} 1",
+			err:   "no token found",
+		},
+	}
+
+	for _, c := range cases {
+		p := New([]byte(c.input))
+		for p.Next() {
+		}
+
+		if c.err == "" {
+			require.NoError(t, p.Err())
+			continue
+		}
+
+		require.Error(t, p.Err())
+		require.Equal(t, c.err, p.Err().Error())
+	}
+}
+
 const (
 	testdataSampleCount = 410
 )

promql/fuzz-data/ParseMetric/corpus/982cbe5ad899f03c630b1a21876a206707ea3dc9

@@ -0,0 +1 @@
+o { quantile = "1.0", a = "b" } 8.3835e-05

promql/fuzz-data/ParseMetric/corpus/exposition_formats_5

@@ -0,0 +1 @@
+go_gc_duration_seconds{ quantile="0.9", a="b"} 8.3835e-05

promql/fuzz-data/ParseMetric/corpus/exposition_formats_6

@@ -0,0 +1 @@
+go_gc_duration_seconds{ quantile="1.0", a="b" } 8.3835e-05

promql/fuzz-data/ParseMetric/corpus/exposition_formats_7

@@ -0,0 +1 @@
+go_gc_duration_seconds { quantile = "1.0", a = "b" } 8.3835e-05

promql/fuzz.go

@@ -16,6 +16,8 @@
 
 package promql
 
+import "github.com/prometheus/prometheus/pkg/textparse"
+
 // PromQL parser fuzzing instrumentation for use with
 // https://github.com/dvyukov/go-fuzz.
 //
@@ -48,8 +50,11 @@ const (
 // Note that his is not the parser for the text-based exposition-format; that
 // lives in github.com/prometheus/client_golang/text.
 func FuzzParseMetric(in []byte) int {
-	_, err := ParseMetric(string(in))
+	p := textparse.New(in)
-	if err == nil {
+	for p.Next() {
+	}
+
+	if p.Err() == nil {
 		return fuzzInteresting
 	}