Fix several render issues (#14986)

* Fix an issue with panics related to attributes
* Wrap goldmark render in a recovery function
* Reduce memory use in render emoji
* Use a pipe for rendering goldmark - still needs more work and a limiter

Signed-off-by: Andrew Thornton <art27@cantab.net>
Co-authored-by: Lauris BH <lauris@nix.lv>

modules/emoji/emoji.go

@@ -30,6 +30,9 @@ var (
 	// aliasMap provides a map of the alias to its emoji data.
 	aliasMap map[string]int
 
+	// emptyReplacer is the string replacer for emoji codes.
+	emptyReplacer *strings.Replacer
+
 	// codeReplacer is the string replacer for emoji codes.
 	codeReplacer *strings.Replacer
 
@@ -49,6 +52,7 @@ func loadMap() {
 
 		// process emoji codes and aliases
 		codePairs := make([]string, 0)
+		emptyPairs := make([]string, 0)
 		aliasPairs := make([]string, 0)
 
 		// sort from largest to small so we match combined emoji first
@@ -64,6 +68,7 @@ func loadMap() {
 			// setup codes
 			codeMap[e.Emoji] = i
 			codePairs = append(codePairs, e.Emoji, ":"+e.Aliases[0]+":")
+			emptyPairs = append(emptyPairs, e.Emoji, e.Emoji)
 
 			// setup aliases
 			for _, a := range e.Aliases {
@@ -77,6 +82,7 @@ func loadMap() {
 		}
 
 		// create replacers
+		emptyReplacer = strings.NewReplacer(emptyPairs...)
 		codeReplacer = strings.NewReplacer(codePairs...)
 		aliasReplacer = strings.NewReplacer(aliasPairs...)
 	})
@@ -127,38 +133,53 @@ func ReplaceAliases(s string) string {
 	return aliasReplacer.Replace(s)
 }
 
-// FindEmojiSubmatchIndex returns index pair of longest emoji in a string
-func FindEmojiSubmatchIndex(s string) []int {
-	loadMap()
-	found := make(map[int]int)
-	keys := make([]int, 0)
+type rememberSecondWriteWriter struct {
+	pos        int
+	idx        int
+	end        int
+	writecount int
+}
 
-	//see if there are any emoji in string before looking for position of specific ones
-	//no performance difference when there is a match but 10x faster when there are not
-	if s == ReplaceCodes(s) {
-		return nil
+func (n *rememberSecondWriteWriter) Write(p []byte) (int, error) {
+	n.writecount++
+	if n.writecount == 2 {
+		n.idx = n.pos
+		n.end = n.pos + len(p)
 	}
+	n.pos += len(p)
+	return len(p), nil
+}
 
-	// get index of first emoji occurrence while also checking for longest combination
-	for j := range GemojiData {
-		i := strings.Index(s, GemojiData[j].Emoji)
-		if i != -1 {
-			if _, ok := found[i]; !ok {
-				if len(keys) == 0 || i < keys[0] {
-					found[i] = j
-					keys = []int{i}
-				}
-				if i == 0 {
-					break
-				}
-			}
-		}
+func (n *rememberSecondWriteWriter) WriteString(s string) (int, error) {
+	n.writecount++
+	if n.writecount == 2 {
+		n.idx = n.pos
+		n.end = n.pos + len(s)
 	}
+	n.pos += len(s)
+	return len(s), nil
+}
 
-	if len(keys) > 0 {
-		index := keys[0]
-		return []int{index, index + len(GemojiData[found[index]].Emoji)}
+// FindEmojiSubmatchIndex returns index pair of longest emoji in a string
+func FindEmojiSubmatchIndex(s string) []int {
+	loadMap()
+	secondWriteWriter := rememberSecondWriteWriter{}
+
+	// A faster and clean implementation would copy the trie tree formation in strings.NewReplacer but
+	// we can be lazy here.
+	//
+	// The implementation of strings.Replacer.WriteString is such that the first index of the emoji
+	// submatch is simply the second thing that is written to WriteString in the writer.
+	//
+	// Therefore we can simply take the index of the second write as our first emoji
+	//
+	// FIXME: just copy the trie implementation from strings.NewReplacer
+	_, _ = emptyReplacer.WriteString(&secondWriteWriter, s)
+
+	// if we wrote less than twice then we never "replaced"
+	if secondWriteWriter.writecount < 2 {
+		return nil
 	}
 
-	return nil
+	return []int{secondWriteWriter.idx, secondWriteWriter.end}
 }

modules/emoji/emoji_test.go

@@ -8,6 +8,8 @@ package emoji
 import (
 	"reflect"
 	"testing"
+
+	"github.com/stretchr/testify/assert"
 )
 
 func TestDumpInfo(t *testing.T) {
@@ -65,3 +67,34 @@ func TestReplacers(t *testing.T) {
 		}
 	}
 }
+
+func TestFindEmojiSubmatchIndex(t *testing.T) {
+	type testcase struct {
+		teststring string
+		expected   []int
+	}
+
+	testcases := []testcase{
+		{
+			"\U0001f44d",
+			[]int{0, len("\U0001f44d")},
+		},
+		{
+			"\U0001f44d +1 \U0001f44d \U0001f37a",
+			[]int{0, 4},
+		},
+		{
+			" \U0001f44d",
+			[]int{1, 1 + len("\U0001f44d")},
+		},
+		{
+			string([]byte{'\u0001'}) + "\U0001f44d",
+			[]int{1, 1 + len("\U0001f44d")},
+		},
+	}
+
+	for _, kase := range testcases {
+		actual := FindEmojiSubmatchIndex(kase.teststring)
+		assert.Equal(t, kase.expected, actual)
+	}
+}

modules/markup/html.go

@@ -313,41 +313,27 @@ func RenderEmoji(
 	return ctx.postProcess(rawHTML)
 }
 
+var tagCleaner = regexp.MustCompile(`<((?:/?\w+/\w+)|(?:/[\w ]+/)|(/?[hH][tT][mM][lL][ />]))`)
+var nulCleaner = strings.NewReplacer("\000", "")
+
 func (ctx *postProcessCtx) postProcess(rawHTML []byte) ([]byte, error) {
 	if ctx.procs == nil {
 		ctx.procs = defaultProcessors
 	}
 
 	// give a generous extra 50 bytes
-	res := make([]byte, 0, len(rawHTML)+50)
-
+	res := bytes.NewBuffer(make([]byte, 0, len(rawHTML)+50))
 	// prepend "<html><body>"
-	res = append(res, "<html><body>"...)
+	_, _ = res.WriteString("<html><body>")
 
 	// Strip out nuls - they're always invalid
-	start := bytes.IndexByte(rawHTML, '\000')
-	if start >= 0 {
-		res = append(res, rawHTML[:start]...)
-		start++
-		for start < len(rawHTML) {
-			end := bytes.IndexByte(rawHTML[start:], '\000')
-			if end < 0 {
-				res = append(res, rawHTML[start:]...)
-				break
-			} else if end > 0 {
-				res = append(res, rawHTML[start:start+end]...)
-			}
-			start += end + 1
-		}
-	} else {
-		res = append(res, rawHTML...)
-	}
+	_, _ = nulCleaner.WriteString(res, string(tagCleaner.ReplaceAll(rawHTML, []byte("&lt;$1"))))
 
 	// close the tags
-	res = append(res, "</body></html>"...)
+	_, _ = res.WriteString("</body></html>")
 
 	// parse the HTML
-	nodes, err := html.ParseFragment(bytes.NewReader(res), nil)
+	nodes, err := html.ParseFragment(res, nil)
 	if err != nil {
 		return nil, &postProcessError{"invalid HTML", err}
 	}
@@ -384,17 +370,17 @@ func (ctx *postProcessCtx) postProcess(rawHTML []byte) ([]byte, error) {
 	// Create buffer in which the data will be placed again. We know that the
 	// length will be at least that of res; to spare a few alloc+copy, we
 	// reuse res, resetting its length to 0.
-	buf := bytes.NewBuffer(res[:0])
+	res.Reset()
 	// Render everything to buf.
 	for _, node := range nodes {
-		err = html.Render(buf, node)
+		err = html.Render(res, node)
 		if err != nil {
 			return nil, &postProcessError{"error rendering processed HTML", err}
 		}
 	}
 
 	// Everything done successfully, return parsed data.
-	return buf.Bytes(), nil
+	return res.Bytes(), nil
 }
 
 func (ctx *postProcessCtx) visitNode(node *html.Node, visitText bool) {

modules/markup/markdown/goldmark.go

@@ -77,6 +77,12 @@ func (g *ASTTransformer) Transform(node *ast.Document, reader text.Reader, pc pa
 					header.ID = util.BytesToReadOnlyString(id.([]byte))
 				}
 				toc = append(toc, header)
+			} else {
+				for _, attr := range v.Attributes() {
+					if _, ok := attr.Value.([]byte); !ok {
+						v.SetAttribute(attr.Name, []byte(fmt.Sprintf("%v", attr.Value)))
+					}
+				}
 			}
 		case *ast.Image:
 			// Images need two things:

modules/markup/markdown/markdown.go

@@ -6,7 +6,8 @@
 package markdown
 
 import (
-	"bytes"
+	"fmt"
+	"io"
 	"strings"
 	"sync"
 
@@ -18,7 +19,7 @@ import (
 
 	chromahtml "github.com/alecthomas/chroma/formatters/html"
 	"github.com/yuin/goldmark"
-	"github.com/yuin/goldmark-highlighting"
+	highlighting "github.com/yuin/goldmark-highlighting"
 	meta "github.com/yuin/goldmark-meta"
 	"github.com/yuin/goldmark/extension"
 	"github.com/yuin/goldmark/parser"
@@ -34,6 +35,44 @@ var urlPrefixKey = parser.NewContextKey()
 var isWikiKey = parser.NewContextKey()
 var renderMetasKey = parser.NewContextKey()
 
+type closesWithError interface {
+	io.WriteCloser
+	CloseWithError(err error) error
+}
+
+type limitWriter struct {
+	w     closesWithError
+	sum   int64
+	limit int64
+}
+
+// Write implements the standard Write interface:
+func (l *limitWriter) Write(data []byte) (int, error) {
+	leftToWrite := l.limit - l.sum
+	if leftToWrite < int64(len(data)) {
+		n, err := l.w.Write(data[:leftToWrite])
+		l.sum += int64(n)
+		if err != nil {
+			return n, err
+		}
+		_ = l.w.Close()
+		return n, fmt.Errorf("Rendered content too large - truncating render")
+	}
+	n, err := l.w.Write(data)
+	l.sum += int64(n)
+	return n, err
+}
+
+// Close closes the writer
+func (l *limitWriter) Close() error {
+	return l.w.Close()
+}
+
+// CloseWithError closes the writer
+func (l *limitWriter) CloseWithError(err error) error {
+	return l.w.CloseWithError(err)
+}
+
 // NewGiteaParseContext creates a parser.Context with the gitea context set
 func NewGiteaParseContext(urlPrefix string, metas map[string]string, isWiki bool) parser.Context {
 	pc := parser.NewContext(parser.WithIDs(newPrefixedIDs()))
@@ -43,8 +82,8 @@ func NewGiteaParseContext(urlPrefix string, metas map[string]string, isWiki bool
 	return pc
 }
 
-// render renders Markdown to HTML without handling special links.
-func render(body []byte, urlPrefix string, metas map[string]string, wikiMarkdown bool) []byte {
+// actualRender renders Markdown to HTML without handling special links.
+func actualRender(body []byte, urlPrefix string, metas map[string]string, wikiMarkdown bool) []byte {
 	once.Do(func() {
 		converter = goldmark.New(
 			goldmark.WithExtensions(extension.Table,
@@ -119,12 +158,57 @@ func render(body []byte, urlPrefix string, metas map[string]string, wikiMarkdown
 
 	})
 
-	pc := NewGiteaParseContext(urlPrefix, metas, wikiMarkdown)
-	var buf bytes.Buffer
-	if err := converter.Convert(giteautil.NormalizeEOL(body), &buf, parser.WithContext(pc)); err != nil {
-		log.Error("Unable to render: %v", err)
+	rd, wr := io.Pipe()
+	defer func() {
+		_ = rd.Close()
+		_ = wr.Close()
+	}()
+
+	lw := &limitWriter{
+		w:     wr,
+		limit: setting.UI.MaxDisplayFileSize * 3,
 	}
-	return markup.SanitizeReader(&buf).Bytes()
+
+	// FIXME: should we include a timeout that closes the pipe to abort the parser and sanitizer if it takes too long?
+	go func() {
+		defer func() {
+			err := recover()
+			if err == nil {
+				return
+			}
+
+			log.Warn("Unable to render markdown due to panic in goldmark: %v", err)
+			if log.IsDebug() {
+				log.Debug("Panic in markdown: %v\n%s", err, string(log.Stack(2)))
+			}
+			_ = lw.CloseWithError(fmt.Errorf("%v", err))
+		}()
+
+		pc := NewGiteaParseContext(urlPrefix, metas, wikiMarkdown)
+		if err := converter.Convert(giteautil.NormalizeEOL(body), lw, parser.WithContext(pc)); err != nil {
+			log.Error("Unable to render: %v", err)
+			_ = lw.CloseWithError(err)
+			return
+		}
+		_ = lw.Close()
+	}()
+	return markup.SanitizeReader(rd).Bytes()
+}
+
+func render(body []byte, urlPrefix string, metas map[string]string, wikiMarkdown bool) (ret []byte) {
+	defer func() {
+		err := recover()
+		if err == nil {
+			return
+		}
+
+		log.Warn("Unable to render markdown due to panic in goldmark - will return sanitized raw bytes")
+		if log.IsDebug() {
+			log.Debug("Panic in markdown: %v\n%s", err, string(log.Stack(2)))
+		}
+		ret = markup.SanitizeBytes(body)
+	}()
+	return actualRender(body, urlPrefix, metas, wikiMarkdown)
 }
 
 var (