Fix external render, make iframe render work (#35727, #35730) (#35731)

Backport #35727 and #35730

---------

Signed-off-by: wxiaoguang <wxiaoguang@gmail.com>

custom/conf/app.example.ini

@@ -2540,7 +2540,19 @@ LEVEL = Info
 ;; * sanitized: Sanitize the content and render it inside current page, default to only allow a few HTML tags and attributes. Customized sanitizer rules can be defined in [markup.sanitizer.*] .
 ;; * no-sanitizer: Disable the sanitizer and render the content inside current page. It's **insecure** and may lead to XSS attack if the content contains malicious code.
 ;; * iframe: Render the content in a separate standalone page and embed it into current page by iframe. The iframe is in sandbox mode with same-origin disabled, and the JS code are safely isolated from parent page.
-;RENDER_CONTENT_MODE=sanitized
+;RENDER_CONTENT_MODE = sanitized
+;; The sandbox applied to the iframe and Content-Security-Policy header when RENDER_CONTENT_MODE is `iframe`.
+;; It defaults to a safe set of "allow-*" restrictions (space separated).
+;; You can also set it by your requirements or use "disabled" to disable the sandbox completely.
+;; When set it, make sure there is no security risk:
+;; * PDF-only content: generally safe to use "disabled", and it needs to be "disabled" because PDF only renders with no sandbox.
+;; * HTML content with JS: if the "RENDER_COMMAND" can guarantee there is no XSS, then it is safe, otherwise, you need to fine tune the "allow-*" restrictions.
+;RENDER_CONTENT_SANDBOX =
+;; Whether post-process the rendered HTML content, including:
+;; resolve relative links and image sources, recognizing issue/commit references, escaping invisible characters,
+;; mentioning users, rendering permlink code blocks, replacing emoji shorthands, etc.
+;; By default, this is true when RENDER_CONTENT_MODE is `sanitized`, otherwise false.
+;NEED_POST_PROCESS = false
 
 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;

modules/httplib/serve.go

@@ -126,6 +126,7 @@ func setServeHeadersByFile(r *http.Request, w http.ResponseWriter, mineBuf []byt
 		// no sandbox attribute for pdf as it breaks rendering in at least safari. this
 		// should generally be safe as scripts inside PDF can not escape the PDF document
 		// see https://bugs.chromium.org/p/chromium/issues/detail?id=413851 for more discussion
+		// HINT: PDF-RENDER-SANDBOX: PDF won't render in sandboxed context
 		w.Header().Set("Content-Security-Policy", "default-src 'none'; style-src 'unsafe-inline'")
 	}
 

modules/markup/external/external.go

@@ -15,6 +15,8 @@ import (
 	"code.gitea.io/gitea/modules/markup"
 	"code.gitea.io/gitea/modules/process"
 	"code.gitea.io/gitea/modules/setting"
+
+	"github.com/kballard/go-shellquote"
 )
 
 // RegisterRenderers registers all supported third part renderers according settings
@@ -56,14 +58,11 @@ func (p *Renderer) SanitizerRules() []setting.MarkupSanitizerRule {
 	return p.MarkupSanitizerRules
 }
 
-// SanitizerDisabled disabled sanitize if return true
+func (p *Renderer) GetExternalRendererOptions() (ret markup.ExternalRendererOptions) {
-func (p *Renderer) SanitizerDisabled() bool {
+	ret.SanitizerDisabled = p.RenderContentMode == setting.RenderContentModeNoSanitizer || p.RenderContentMode == setting.RenderContentModeIframe
-	return p.RenderContentMode == setting.RenderContentModeNoSanitizer || p.RenderContentMode == setting.RenderContentModeIframe
+	ret.DisplayInIframe = p.RenderContentMode == setting.RenderContentModeIframe
-}
+	ret.ContentSandbox = p.RenderContentSandbox
-
+	return ret
-// DisplayInIFrame represents whether render the content with an iframe
-func (p *Renderer) DisplayInIFrame() bool {
-	return p.RenderContentMode == setting.RenderContentModeIframe
 }
 
 func envMark(envName string) string {
@@ -81,7 +80,10 @@ func (p *Renderer) Render(ctx *markup.RenderContext, input io.Reader, output io.
 		envMark("GITEA_PREFIX_SRC"), baseLinkSrc,
 		envMark("GITEA_PREFIX_RAW"), baseLinkRaw,
 	).Replace(p.Command)
-	commands := strings.Fields(command)
+	commands, err := shellquote.Split(command)
+	if err != nil || len(commands) == 0 {
+		return fmt.Errorf("%s invalid command %q: %w", p.Name(), p.Command, err)
+	}
 	args := commands[1:]
 
 	if p.IsInputFile {

modules/markup/internal/finalprocessor.go

@@ -5,11 +5,13 @@ package internal
 
 import (
 	"bytes"
+	"html/template"
 	"io"
 )
 
 type finalProcessor struct {
 	renderInternal *RenderInternal
+	extraHeadHTML  template.HTML
 
 	output io.Writer
 	buf    bytes.Buffer
@@ -25,6 +27,32 @@ func (p *finalProcessor) Close() error {
 	// because "postProcess" already does so. In the future we could optimize the code to process data on the fly.
 	buf := p.buf.Bytes()
 	buf = bytes.ReplaceAll(buf, []byte(` data-attr-class="`+p.renderInternal.secureIDPrefix), []byte(` class="`))
-	_, err := p.output.Write(buf)
+
+	tmp := bytes.TrimSpace(buf)
+	isLikelyHTML := len(tmp) != 0 && tmp[0] == '<' && tmp[len(tmp)-1] == '>' && bytes.Index(tmp, []byte(`</`)) > 0
+	if !isLikelyHTML {
+		// not HTML, write back directly
+		_, err := p.output.Write(buf)
+		return err
+	}
+
+	// add our extra head HTML into output
+	headBytes := []byte("<head>")
+	posHead := bytes.Index(buf, headBytes)
+	var part1, part2 []byte
+	if posHead >= 0 {
+		part1, part2 = buf[:posHead+len(headBytes)], buf[posHead+len(headBytes):]
+	} else {
+		part1, part2 = nil, buf
+	}
+	if len(part1) > 0 {
+		if _, err := p.output.Write(part1); err != nil {
+			return err
+		}
+	}
+	if _, err := io.WriteString(p.output, string(p.extraHeadHTML)); err != nil {
+		return err
+	}
+	_, err := p.output.Write(part2)
 	return err
 }

modules/markup/internal/internal_test.go

@@ -12,7 +12,7 @@ import (
 	"github.com/stretchr/testify/assert"
 )
 
-func TestRenderInternal(t *testing.T) {
+func TestRenderInternalAttrs(t *testing.T) {
 	cases := []struct {
 		input, protected, recovered string
 	}{
@@ -30,7 +30,7 @@ func TestRenderInternal(t *testing.T) {
 	for _, c := range cases {
 		var r RenderInternal
 		out := &bytes.Buffer{}
-		in := r.init("sec", out)
+		in := r.init("sec", out, "")
 		protected := r.ProtectSafeAttrs(template.HTML(c.input))
 		assert.EqualValues(t, c.protected, protected)
 		_, _ = io.WriteString(in, string(protected))
@@ -41,7 +41,7 @@ func TestRenderInternal(t *testing.T) {
 	var r1, r2 RenderInternal
 	protected := r1.ProtectSafeAttrs(`<div class="test"></div>`)
 	assert.EqualValues(t, `<div class="test"></div>`, protected, "non-initialized RenderInternal should not protect any attributes")
-	_ = r1.init("sec", nil)
+	_ = r1.init("sec", nil, "")
 	protected = r1.ProtectSafeAttrs(`<div class="test"></div>`)
 	assert.EqualValues(t, `<div data-attr-class="sec:test"></div>`, protected)
 	assert.Equal(t, "data-attr-class", r1.SafeAttr("class"))
@@ -54,8 +54,37 @@ func TestRenderInternal(t *testing.T) {
 	assert.Empty(t, recovered)
 
 	out2 := &bytes.Buffer{}
-	in2 := r2.init("sec-other", out2)
+	in2 := r2.init("sec-other", out2, "")
 	_, _ = io.WriteString(in2, string(protected))
 	_ = in2.Close()
 	assert.Equal(t, `<div data-attr-class="sec:test"></div>`, out2.String(), "different secureID should not recover the value")
 }
+
+func TestRenderInternalExtraHead(t *testing.T) {
+	t.Run("HeadExists", func(t *testing.T) {
+		out := &bytes.Buffer{}
+		var r RenderInternal
+		in := r.init("sec", out, `<MY-TAG>`)
+		_, _ = io.WriteString(in, `<head>any</head>`)
+		_ = in.Close()
+		assert.Equal(t, `<head><MY-TAG>any</head>`, out.String())
+	})
+
+	t.Run("HeadNotExists", func(t *testing.T) {
+		out := &bytes.Buffer{}
+		var r RenderInternal
+		in := r.init("sec", out, `<MY-TAG>`)
+		_, _ = io.WriteString(in, `<div></div>`)
+		_ = in.Close()
+		assert.Equal(t, `<MY-TAG><div></div>`, out.String())
+	})
+
+	t.Run("NotHTML", func(t *testing.T) {
+		out := &bytes.Buffer{}
+		var r RenderInternal
+		in := r.init("sec", out, `<MY-TAG>`)
+		_, _ = io.WriteString(in, `<any>`)
+		_ = in.Close()
+		assert.Equal(t, `<any>`, out.String())
+	})
+}

modules/markup/internal/renderinternal.go

@@ -29,19 +29,19 @@ type RenderInternal struct {
 	secureIDPrefix string
 }
 
-func (r *RenderInternal) Init(output io.Writer) io.WriteCloser {
+func (r *RenderInternal) Init(output io.Writer, extraHeadHTML template.HTML) io.WriteCloser {
 	buf := make([]byte, 12)
 	_, err := rand.Read(buf)
 	if err != nil {
 		panic("unable to generate secure id")
 	}
-	return r.init(base64.URLEncoding.EncodeToString(buf), output)
+	return r.init(base64.URLEncoding.EncodeToString(buf), output, extraHeadHTML)
 }
 
-func (r *RenderInternal) init(secID string, output io.Writer) io.WriteCloser {
+func (r *RenderInternal) init(secID string, output io.Writer, extraHeadHTML template.HTML) io.WriteCloser {
 	r.secureID = secID
 	r.secureIDPrefix = r.secureID + ":"
-	return &finalProcessor{renderInternal: r, output: output}
+	return &finalProcessor{renderInternal: r, output: output, extraHeadHTML: extraHeadHTML}
 }
 
 func (r *RenderInternal) RecoverProtectedValue(v string) (string, bool) {

modules/markup/render.go

@@ -6,12 +6,14 @@ package markup
 import (
 	"context"
 	"fmt"
+	"html/template"
 	"io"
 	"net/url"
 	"strconv"
 	"strings"
 	"time"
 
+	"code.gitea.io/gitea/modules/htmlutil"
 	"code.gitea.io/gitea/modules/markup/internal"
 	"code.gitea.io/gitea/modules/setting"
 	"code.gitea.io/gitea/modules/util"
@@ -120,31 +122,38 @@ func (ctx *RenderContext) WithHelper(helper RenderHelper) *RenderContext {
 	return ctx
 }
 
-// Render renders markup file to HTML with all specific handling stuff.
+// FindRendererByContext finds renderer by RenderContext
-func Render(ctx *RenderContext, input io.Reader, output io.Writer) error {
+// TODO: it should be merged with other similar functions like GetRendererByFileName, DetectMarkupTypeByFileName, etc
+func FindRendererByContext(ctx *RenderContext) (Renderer, error) {
 	if ctx.RenderOptions.MarkupType == "" && ctx.RenderOptions.RelativePath != "" {
 		ctx.RenderOptions.MarkupType = DetectMarkupTypeByFileName(ctx.RenderOptions.RelativePath)
 		if ctx.RenderOptions.MarkupType == "" {
-			return util.NewInvalidArgumentErrorf("unsupported file to render: %q", ctx.RenderOptions.RelativePath)
+			return nil, util.NewInvalidArgumentErrorf("unsupported file to render: %q", ctx.RenderOptions.RelativePath)
 		}
 	}
 
 	renderer := renderers[ctx.RenderOptions.MarkupType]
 	if renderer == nil {
-		return util.NewInvalidArgumentErrorf("unsupported markup type: %q", ctx.RenderOptions.MarkupType)
+		return nil, util.NewNotExistErrorf("unsupported markup type: %q", ctx.RenderOptions.MarkupType)
 	}
 
-	if ctx.RenderOptions.RelativePath != "" {
+	return renderer, nil
-		if externalRender, ok := renderer.(ExternalRenderer); ok && externalRender.DisplayInIFrame() {
+}
-			if !ctx.RenderOptions.InStandalonePage {
+
-				// for an external "DisplayInIFrame" render, it could only output its content in a standalone page
+func RendererNeedPostProcess(renderer Renderer) bool {
-				// otherwise, a <iframe> should be outputted to embed the external rendered page
+	if r, ok := renderer.(PostProcessRenderer); ok && r.NeedPostProcess() {
-				return renderIFrame(ctx, output)
+		return true
-			}
-		}
 	}
+	return false
+}
 
-	return render(ctx, renderer, input, output)
+// Render renders markup file to HTML with all specific handling stuff.
+func Render(ctx *RenderContext, input io.Reader, output io.Writer) error {
+	renderer, err := FindRendererByContext(ctx)
+	if err != nil {
+		return err
+	}
+	return RenderWithRenderer(ctx, renderer, input, output)
 }
 
 // RenderString renders Markup string to HTML with all specific handling stuff and return string
@@ -156,24 +165,20 @@ func RenderString(ctx *RenderContext, content string) (string, error) {
 	return buf.String(), nil
 }
 
-func renderIFrame(ctx *RenderContext, output io.Writer) error {
+func renderIFrame(ctx *RenderContext, sandbox string, output io.Writer) error {
-	// set height="0" ahead, otherwise the scrollHeight would be max(150, realHeight)
+	src := fmt.Sprintf("%s/%s/%s/render/%s/%s", setting.AppSubURL,
-	// at the moment, only "allow-scripts" is allowed for sandbox mode.
-	// "allow-same-origin" should never be used, it leads to XSS attack, and it makes the JS in iframe can access parent window's config and CSRF token
-	// TODO: when using dark theme, if the rendered content doesn't have proper style, the default text color is black, which is not easy to read
-	_, err := io.WriteString(output, fmt.Sprintf(`
-<iframe src="%s/%s/%s/render/%s/%s"
-name="giteaExternalRender"
-onload="this.height=giteaExternalRender.document.documentElement.scrollHeight"
-width="100%%" height="0" scrolling="no" frameborder="0" style="overflow: hidden"
-sandbox="allow-scripts"
-></iframe>`,
-		setting.AppSubURL,
 		url.PathEscape(ctx.RenderOptions.Metas["user"]),
 		url.PathEscape(ctx.RenderOptions.Metas["repo"]),
-		ctx.RenderOptions.Metas["RefTypeNameSubURL"],
+		util.PathEscapeSegments(ctx.RenderOptions.Metas["RefTypeNameSubURL"]),
-		url.PathEscape(ctx.RenderOptions.RelativePath),
+		util.PathEscapeSegments(ctx.RenderOptions.RelativePath),
-	))
+	)
+
+	var sandboxAttrValue template.HTML
+	if sandbox != "" {
+		sandboxAttrValue = htmlutil.HTMLFormat(`sandbox="%s"`, sandbox)
+	}
+	iframe := htmlutil.HTMLFormat(`<iframe data-src="%s" class="external-render-iframe" %s></iframe>`, src, sandboxAttrValue)
+	_, err := io.WriteString(output, string(iframe))
 	return err
 }
 
@@ -185,13 +190,34 @@ func pipes() (io.ReadCloser, io.WriteCloser, func()) {
 	}
 }
 
-func render(ctx *RenderContext, renderer Renderer, input io.Reader, output io.Writer) error {
+func getExternalRendererOptions(renderer Renderer) (ret ExternalRendererOptions, _ bool) {
+	if externalRender, ok := renderer.(ExternalRenderer); ok {
+		return externalRender.GetExternalRendererOptions(), true
+	}
+	return ret, false
+}
+
+func RenderWithRenderer(ctx *RenderContext, renderer Renderer, input io.Reader, output io.Writer) error {
+	var extraHeadHTML template.HTML
+	if extOpts, ok := getExternalRendererOptions(renderer); ok && extOpts.DisplayInIframe {
+		if !ctx.RenderOptions.InStandalonePage {
+			// for an external "DisplayInIFrame" render, it could only output its content in a standalone page
+			// otherwise, a <iframe> should be outputted to embed the external rendered page
+			return renderIFrame(ctx, extOpts.ContentSandbox, output)
+		}
+		// else: this is a standalone page, fallthrough to the real rendering, and add extra JS/CSS
+		extraStyleHref := setting.AppSubURL + "/assets/css/external-render-iframe.css"
+		extraScriptSrc := setting.AppSubURL + "/assets/js/external-render-iframe.js"
+		// "<script>" must go before "<link>", to make Golang's http.DetectContentType() can still recognize the content as "text/html"
+		extraHeadHTML = htmlutil.HTMLFormat(`<script src="%s"></script><link rel="stylesheet" href="%s">`, extraScriptSrc, extraStyleHref)
+	}
+
 	ctx.usedByRender = true
 	if ctx.RenderHelper != nil {
 		defer ctx.RenderHelper.CleanUp()
 	}
 
-	finalProcessor := ctx.RenderInternal.Init(output)
+	finalProcessor := ctx.RenderInternal.Init(output, extraHeadHTML)
 	defer finalProcessor.Close()
 
 	// input -> (pw1=pr1) -> renderer -> (pw2=pr2) -> SanitizeReader -> finalProcessor -> output
@@ -202,7 +228,7 @@ func render(ctx *RenderContext, renderer Renderer, input io.Reader, output io.Wr
 	eg, _ := errgroup.WithContext(ctx)
 	var pw2 io.WriteCloser = util.NopCloser{Writer: finalProcessor}
 
-	if r, ok := renderer.(ExternalRenderer); !ok || !r.SanitizerDisabled() {
+	if r, ok := renderer.(ExternalRenderer); !ok || !r.GetExternalRendererOptions().SanitizerDisabled {
 		var pr2 io.ReadCloser
 		var close2 func()
 		pr2, pw2, close2 = pipes()
@@ -214,7 +240,7 @@ func render(ctx *RenderContext, renderer Renderer, input io.Reader, output io.Wr
 	}
 
 	eg.Go(func() (err error) {
-		if r, ok := renderer.(PostProcessRenderer); ok && r.NeedPostProcess() {
+		if RendererNeedPostProcess(renderer) {
 			err = PostProcessDefault(ctx, pr1, pw2)
 		} else {
 			_, err = io.Copy(pw2, pr1)

modules/markup/renderer.go

@@ -25,13 +25,15 @@ type PostProcessRenderer interface {
 	NeedPostProcess() bool
 }
 
+type ExternalRendererOptions struct {
+	SanitizerDisabled bool
+	DisplayInIframe   bool
+	ContentSandbox    string
+}
+
 // ExternalRenderer defines an interface for external renderers
 type ExternalRenderer interface {
-	// SanitizerDisabled disabled sanitize if return true
+	GetExternalRendererOptions() ExternalRendererOptions
-	SanitizerDisabled() bool
-
-	// DisplayInIFrame represents whether render the content with an iframe
-	DisplayInIFrame() bool
 }
 
 // RendererContentDetector detects if the content can be rendered

modules/setting/markup.go

@@ -63,6 +63,7 @@ type MarkupRenderer struct {
 	NeedPostProcess      bool
 	MarkupSanitizerRules []MarkupSanitizerRule
 	RenderContentMode    string
+	RenderContentSandbox string
 }
 
 // MarkupSanitizerRule defines the policy for whitelisting attributes on
@@ -253,13 +254,24 @@ func newMarkupRenderer(name string, sec ConfigSection) {
 		renderContentMode = RenderContentModeSanitized
 	}
 
+	// ATTENTION! at the moment, only a safe set like "allow-scripts" are allowed for sandbox mode.
+	// "allow-same-origin" should never be used, it leads to XSS attack, and it makes the JS in iframe can access parent window's config and CSRF token
+	renderContentSandbox := sec.Key("RENDER_CONTENT_SANDBOX").MustString("allow-scripts allow-popups")
+	if renderContentSandbox == "disabled" {
+		renderContentSandbox = ""
+	}
+
 	ExternalMarkupRenderers = append(ExternalMarkupRenderers, &MarkupRenderer{
-		Enabled:           sec.Key("ENABLED").MustBool(false),
+		Enabled:        sec.Key("ENABLED").MustBool(false),
-		MarkupName:        name,
+		MarkupName:     name,
-		FileExtensions:    exts,
+		FileExtensions: exts,
-		Command:           command,
+		Command:        command,
-		IsInputFile:       sec.Key("IS_INPUT_FILE").MustBool(false),
+		IsInputFile:    sec.Key("IS_INPUT_FILE").MustBool(false),
-		NeedPostProcess:   sec.Key("NEED_POSTPROCESS").MustBool(true),
+
-		RenderContentMode: renderContentMode,
+		RenderContentMode:    renderContentMode,
+		RenderContentSandbox: renderContentSandbox,
+
+		// if no sanitizer is needed, no post process is needed
+		NeedPostProcess: sec.Key("NEED_POST_PROCESS").MustBool(renderContentMode == RenderContentModeSanitized),
 	})
 }

routers/web/repo/render.go

@@ -4,18 +4,13 @@
 package repo
 
 import (
-	"bytes"
-	"io"
 	"net/http"
 	"path"
 
 	"code.gitea.io/gitea/models/renderhelper"
-	"code.gitea.io/gitea/modules/charset"
 	"code.gitea.io/gitea/modules/git"
 	"code.gitea.io/gitea/modules/log"
 	"code.gitea.io/gitea/modules/markup"
-	"code.gitea.io/gitea/modules/typesniffer"
-	"code.gitea.io/gitea/modules/util"
 	"code.gitea.io/gitea/services/context"
 )
 
@@ -44,22 +39,8 @@ func RenderFile(ctx *context.Context) {
 	}
 	defer dataRc.Close()
 
-	buf := make([]byte, 1024)
-	n, _ := util.ReadAtMost(dataRc, buf)
-	buf = buf[:n]
-
-	st := typesniffer.DetectContentType(buf)
-	isTextFile := st.IsText()
-
-	rd := charset.ToUTF8WithFallbackReader(io.MultiReader(bytes.NewReader(buf), dataRc), charset.ConvertOpts{})
-	ctx.Resp.Header().Add("Content-Security-Policy", "frame-src 'self'; sandbox allow-scripts")
-
 	if markupType := markup.DetectMarkupTypeByFileName(blob.Name()); markupType == "" {
-		if isTextFile {
+		http.Error(ctx.Resp, "Unsupported file type render", http.StatusBadRequest)
-			_, _ = io.Copy(ctx.Resp, rd)
-		} else {
-			http.Error(ctx.Resp, "Unsupported file type render", http.StatusInternalServerError)
-		}
 		return
 	}
 
@@ -68,7 +49,29 @@ func RenderFile(ctx *context.Context) {
 		CurrentTreePath: path.Dir(ctx.Repo.TreePath),
 	}).WithRelativePath(ctx.Repo.TreePath).WithInStandalonePage(true)
 
-	err = markup.Render(rctx, rd, ctx.Resp)
+	renderer, err := markup.FindRendererByContext(rctx)
+	if err != nil {
+		http.Error(ctx.Resp, "Unable to find renderer", http.StatusBadRequest)
+		return
+	}
+
+	extRenderer, ok := renderer.(markup.ExternalRenderer)
+	if !ok {
+		http.Error(ctx.Resp, "Unable to get external renderer", http.StatusBadRequest)
+		return
+	}
+
+	// To render PDF in iframe, the sandbox must NOT be used (iframe & CSP header).
+	// Chrome blocks the PDF rendering when sandboxed, even if all "allow-*" are set.
+	// HINT: PDF-RENDER-SANDBOX: PDF won't render in sandboxed context
+	extRendererOpts := extRenderer.GetExternalRendererOptions()
+	if extRendererOpts.ContentSandbox != "" {
+		ctx.Resp.Header().Add("Content-Security-Policy", "frame-src 'self'; sandbox "+extRendererOpts.ContentSandbox)
+	} else {
+		ctx.Resp.Header().Add("Content-Security-Policy", "frame-src 'self'")
+	}
+
+	err = markup.RenderWithRenderer(rctx, renderer, dataRc, ctx.Resp)
 	if err != nil {
 		log.Error("Failed to render file %q: %v", ctx.Repo.TreePath, err)
 		http.Error(ctx.Resp, "Failed to render file", http.StatusInternalServerError)

routers/web/repo/view.go

@@ -151,17 +151,28 @@ func loadLatestCommitData(ctx *context.Context, latestCommit *git.Commit) bool {
 }
 
 func markupRender(ctx *context.Context, renderCtx *markup.RenderContext, input io.Reader) (escaped *charset.EscapeStatus, output template.HTML, err error) {
+	renderer, err := markup.FindRendererByContext(renderCtx)
+	if err != nil {
+		return nil, "", err
+	}
+
 	markupRd, markupWr := io.Pipe()
 	defer markupWr.Close()
+
 	done := make(chan struct{})
 	go func() {
 		sb := &strings.Builder{}
-		// We allow NBSP here this is rendered
+		if markup.RendererNeedPostProcess(renderer) {
-		escaped, _ = charset.EscapeControlReader(markupRd, sb, ctx.Locale, charset.RuneNBSP)
+			escaped, _ = charset.EscapeControlReader(markupRd, sb, ctx.Locale, charset.RuneNBSP) // We allow NBSP here this is rendered
+		} else {
+			escaped = &charset.EscapeStatus{}
+			_, _ = io.Copy(sb, markupRd)
+		}
 		output = template.HTML(sb.String())
 		close(done)
 	}()
-	err = markup.Render(renderCtx, input, markupWr)
+
+	err = markup.RenderWithRenderer(renderCtx, renderer, input, markupWr)
 	_ = markupWr.CloseWithError(err)
 	<-done
 	return escaped, output, err

tests/integration/markup_external_test.go

@@ -4,18 +4,22 @@
 package integration
 
 import (
-	"bytes"
-	"io"
 	"net/http"
+	"net/url"
 	"strings"
 	"testing"
 
+	repo_model "code.gitea.io/gitea/models/repo"
+	"code.gitea.io/gitea/models/unittest"
+	user_model "code.gitea.io/gitea/models/user"
 	"code.gitea.io/gitea/modules/markup"
 	"code.gitea.io/gitea/modules/markup/external"
 	"code.gitea.io/gitea/modules/setting"
+	"code.gitea.io/gitea/modules/test"
 	"code.gitea.io/gitea/tests"
 
 	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
 )
 
 func TestExternalMarkupRenderer(t *testing.T) {
@@ -25,36 +29,79 @@ func TestExternalMarkupRenderer(t *testing.T) {
 		return
 	}
 
-	req := NewRequest(t, "GET", "/user30/renderer/src/branch/master/README.html")
+	onGiteaRun(t, func(t *testing.T, _ *url.URL) {
-	resp := MakeRequest(t, req, http.StatusOK)
+		user2 := unittest.AssertExistsAndLoadBean(t, &user_model.User{ID: 2})
-	assert.Equal(t, "text/html; charset=utf-8", resp.Header().Get("Content-Type"))
+		repo1 := unittest.AssertExistsAndLoadBean(t, &repo_model.Repository{ID: 1})
-
+		_, err := createFile(user2, repo1, "file.no-sanitizer", "master", `any content`)
-	bs, err := io.ReadAll(resp.Body)
+		require.NoError(t, err)
-	assert.NoError(t, err)
+
-
+		t.Run("RenderNoSanitizer", func(t *testing.T) {
-	doc := NewHTMLParser(t, bytes.NewBuffer(bs))
+			req := NewRequest(t, "GET", "/user2/repo1/src/branch/master/file.no-sanitizer")
-	div := doc.Find("div.file-view")
+			resp := MakeRequest(t, req, http.StatusOK)
-	data, err := div.Html()
+			doc := NewHTMLParser(t, resp.Body)
-	assert.NoError(t, err)
+			div := doc.Find("div.file-view")
-	assert.Equal(t, "<div>\n\ttest external renderer\n</div>", strings.TrimSpace(data))
+			data, err := div.Html()
-
+			assert.NoError(t, err)
-	r := markup.GetRendererByFileName("a.html").(*external.Renderer)
+			assert.Equal(t, `<script>window.alert("hi")</script>`, strings.TrimSpace(data))
-	r.RenderContentMode = setting.RenderContentModeIframe
+		})
-
+	})
-	req = NewRequest(t, "GET", "/user30/renderer/src/branch/master/README.html")
+
-	resp = MakeRequest(t, req, http.StatusOK)
+	t.Run("RenderContentDirectly", func(t *testing.T) {
-	assert.Equal(t, "text/html; charset=utf-8", resp.Header().Get("Content-Type"))
+		req := NewRequest(t, "GET", "/user30/renderer/src/branch/master/README.html")
-	bs, err = io.ReadAll(resp.Body)
+		resp := MakeRequest(t, req, http.StatusOK)
-	assert.NoError(t, err)
+		assert.Equal(t, "text/html; charset=utf-8", resp.Header().Get("Content-Type"))
-	doc = NewHTMLParser(t, bytes.NewBuffer(bs))
+
-	iframe := doc.Find("iframe")
+		doc := NewHTMLParser(t, resp.Body)
-	assert.Equal(t, "/user30/renderer/render/branch/master/README.html", iframe.AttrOr("src", ""))
+		div := doc.Find("div.file-view")
-
+		data, err := div.Html()
-	req = NewRequest(t, "GET", "/user30/renderer/render/branch/master/README.html")
+		assert.NoError(t, err)
-	resp = MakeRequest(t, req, http.StatusOK)
+		assert.Equal(t, "<div>\n\ttest external renderer\n</div>", strings.TrimSpace(data))
-	assert.Equal(t, "text/html; charset=utf-8", resp.Header().Get("Content-Type"))
+	})
-	bs, err = io.ReadAll(resp.Body)
+
-	assert.NoError(t, err)
+	// above tested "no-sanitizer" mode, then we test iframe mode below
-	assert.Equal(t, "frame-src 'self'; sandbox allow-scripts", resp.Header().Get("Content-Security-Policy"))
+	r := markup.GetRendererByFileName("any-file.html").(*external.Renderer)
-	assert.Equal(t, "<div>\n\ttest external renderer\n</div>\n", string(bs))
+	defer test.MockVariableValue(&r.RenderContentMode, setting.RenderContentModeIframe)()
+	r = markup.GetRendererByFileName("any-file.no-sanitizer").(*external.Renderer)
+	defer test.MockVariableValue(&r.RenderContentMode, setting.RenderContentModeIframe)()
+
+	t.Run("RenderContentInIFrame", func(t *testing.T) {
+		t.Run("DefaultSandbox", func(t *testing.T) {
+			req := NewRequest(t, "GET", "/user30/renderer/src/branch/master/README.html")
+
+			t.Run("ParentPage", func(t *testing.T) {
+				respParent := MakeRequest(t, req, http.StatusOK)
+				assert.Equal(t, "text/html; charset=utf-8", respParent.Header().Get("Content-Type"))
+
+				iframe := NewHTMLParser(t, respParent.Body).Find("iframe.external-render-iframe")
+				assert.Empty(t, iframe.AttrOr("src", "")) // src should be empty, "data-src" is used instead
+
+				// default sandbox on parent page
+				assert.Equal(t, "allow-scripts allow-popups", iframe.AttrOr("sandbox", ""))
+				assert.Equal(t, "/user30/renderer/render/branch/master/README.html", iframe.AttrOr("data-src", ""))
+			})
+			t.Run("SubPage", func(t *testing.T) {
+				req = NewRequest(t, "GET", "/user30/renderer/render/branch/master/README.html")
+				respSub := MakeRequest(t, req, http.StatusOK)
+				assert.Equal(t, "text/html; charset=utf-8", respSub.Header().Get("Content-Type"))
+
+				// default sandbox in sub page response
+				assert.Equal(t, "frame-src 'self'; sandbox allow-scripts allow-popups", respSub.Header().Get("Content-Security-Policy"))
+				assert.Equal(t, "<script src=\"/assets/js/external-render-iframe.js\"></script><link rel=\"stylesheet\" href=\"/assets/css/external-render-iframe.css\"><div>\n\ttest external renderer\n</div>\n", respSub.Body.String())
+			})
+		})
+
+		t.Run("NoSanitizerNoSandbox", func(t *testing.T) {
+			req := NewRequest(t, "GET", "/user2/repo1/src/branch/master/file.no-sanitizer")
+			respParent := MakeRequest(t, req, http.StatusOK)
+			iframe := NewHTMLParser(t, respParent.Body).Find("iframe.external-render-iframe")
+			assert.Equal(t, "/user2/repo1/render/branch/master/file.no-sanitizer", iframe.AttrOr("data-src", ""))
+
+			req = NewRequest(t, "GET", "/user2/repo1/render/branch/master/file.no-sanitizer")
+			respSub := MakeRequest(t, req, http.StatusOK)
+
+			// no sandbox (disabled by RENDER_CONTENT_SANDBOX)
+			assert.Empty(t, iframe.AttrOr("sandbox", ""))
+			assert.Equal(t, "frame-src 'self'", respSub.Header().Get("Content-Security-Policy"))
+		})
+	})
 }

tests/sqlite.ini.tmpl

@@ -114,9 +114,19 @@ ENABLED = true
 [markup.html]
 ENABLED = true
 FILE_EXTENSIONS = .html
-RENDER_COMMAND = `go run build/test-echo.go`
+RENDER_COMMAND = go run build/test-echo.go
-IS_INPUT_FILE = false
+;RENDER_COMMAND = cat
-RENDER_CONTENT_MODE=sanitized
+;IS_INPUT_FILE = true
+RENDER_CONTENT_MODE = sanitized
+
+[markup.no-sanitizer]
+ENABLED = true
+FILE_EXTENSIONS = .no-sanitizer
+RENDER_COMMAND = echo '<script>window.alert("hi")</script>'
+; This test case is reused, at first it is used to test "no-sanitizer" (sandbox doesn't take effect here)
+; Then it will be updated and used to test "iframe + sandbox-disabled"
+RENDER_CONTENT_MODE = no-sanitizer
+RENDER_CONTENT_SANDBOX = disabled
 
 [actions]
 ENABLED = true

web_src/css/markup/content.css

@@ -545,6 +545,11 @@ In markup content, we always use bottom margin for all elements */
   margin: 0 0.25em;
 }
 
+.external-render-iframe {
+  width: 100%;
+  height: max(300px, 80vh);
+}
+
 .markup-content-iframe {
   display: block;
   border: none;

web_src/css/standalone/external-render-iframe.css

@@ -0,0 +1 @@
+/* dummy */

web_src/js/markup/content.ts

@@ -4,6 +4,7 @@ import {initMarkupCodeCopy} from './codecopy.ts';
 import {initMarkupRenderAsciicast} from './asciicast.ts';
 import {initMarkupTasklist} from './tasklist.ts';
 import {registerGlobalSelectorFunc} from '../modules/observer.ts';
+import {initMarkupRenderIframe} from './render-iframe.ts';
 
 // code that runs for all markup content
 export function initMarkupContent(): void {
@@ -13,5 +14,6 @@ export function initMarkupContent(): void {
     initMarkupCodeMermaid(el);
     initMarkupCodeMath(el);
     initMarkupRenderAsciicast(el);
+    initMarkupRenderIframe(el);
   });
 }

web_src/js/markup/render-iframe.ts

@@ -0,0 +1,32 @@
+import {generateElemId, queryElemChildren} from '../utils/dom.ts';
+import {isDarkTheme} from '../utils.ts';
+
+export async function loadRenderIframeContent(iframe: HTMLIFrameElement) {
+  const iframeSrcUrl = iframe.getAttribute('data-src');
+  if (!iframe.id) iframe.id = generateElemId('gitea-iframe-');
+
+  window.addEventListener('message', (e) => {
+    if (!e.data?.giteaIframeCmd || e.data?.giteaIframeId !== iframe.id) return;
+    const cmd = e.data.giteaIframeCmd;
+    if (cmd === 'resize') {
+      iframe.style.height = `${e.data.iframeHeight}px`;
+    } else if (cmd === 'open-link') {
+      if (e.data.anchorTarget === '_blank') {
+        window.open(e.data.openLink, '_blank');
+      } else {
+        window.location.href = e.data.openLink;
+      }
+    } else {
+      throw new Error(`Unknown gitea iframe cmd: ${cmd}`);
+    }
+  });
+
+  const u = new URL(iframeSrcUrl, window.location.origin);
+  u.searchParams.set('gitea-is-dark-theme', String(isDarkTheme()));
+  u.searchParams.set('gitea-iframe-id', iframe.id);
+  iframe.src = u.href;
+}
+
+export function initMarkupRenderIframe(el: HTMLElement) {
+  queryElemChildren(el, 'iframe.external-render-iframe', loadRenderIframeContent);
+}

web_src/js/standalone/external-render-iframe.ts

@@ -0,0 +1,43 @@
+/* To manually test:
+
+[markup.in-iframe]
+ENABLED = true
+FILE_EXTENSIONS = .in-iframe
+RENDER_CONTENT_MODE = iframe
+RENDER_COMMAND = `echo '<div style="width: 100%; height: 2000px; border: 10px solid red; box-sizing: border-box;"><a href="/">a link</a> <a target="_blank" href="//gitea.com">external link</a></div>'`
+
+;RENDER_COMMAND = cat /path/to/file.pdf
+;RENDER_CONTENT_SANDBOX = disabled
+
+*/
+
+function mainExternalRenderIframe() {
+  const u = new URL(window.location.href);
+  const iframeId = u.searchParams.get('gitea-iframe-id');
+
+  // iframe is in different origin, so we need to use postMessage to communicate
+  const postIframeMsg = (cmd: string, data: Record<string, any> = {}) => {
+    window.parent.postMessage({giteaIframeCmd: cmd, giteaIframeId: iframeId, ...data}, '*');
+  };
+
+  const updateIframeHeight = () => postIframeMsg('resize', {iframeHeight: document.documentElement.scrollHeight});
+  updateIframeHeight();
+  window.addEventListener('DOMContentLoaded', updateIframeHeight);
+  // the easiest way to handle dynamic content changes and easy to debug, can be fine-tuned in the future
+  setInterval(updateIframeHeight, 1000);
+
+  //  no way to open an absolute link with CSP frame-src, it also needs some tricks like "postMessage" or "copy the link to clipboard"
+  const openIframeLink = (link: string, target: string) => postIframeMsg('open-link', {openLink: link, anchorTarget: target});
+  document.addEventListener('click', (e) => {
+    const el = e.target as HTMLAnchorElement;
+    if (el.nodeName !== 'A') return;
+    const href = el.getAttribute('href') || '';
+    // safe links: "./any", "../any", "/any", "//host/any", "http://host/any", "https://host/any"
+    if (href.startsWith('.') || href.startsWith('/') || href.startsWith('http://') || href.startsWith('https://')) {
+      e.preventDefault();
+      openIframeLink(href, el.getAttribute('target'));
+    }
+  });
+}
+
+mainExternalRenderIframe();

webpack.config.ts

@@ -75,6 +75,10 @@ export default {
       fileURLToPath(new URL('web_src/js/standalone/swagger.ts', import.meta.url)),
       fileURLToPath(new URL('web_src/css/standalone/swagger.css', import.meta.url)),
     ],
+    'external-render-iframe': [
+      fileURLToPath(new URL('web_src/js/standalone/external-render-iframe.ts', import.meta.url)),
+      fileURLToPath(new URL('web_src/css/standalone/external-render-iframe.css', import.meta.url)),
+    ],
     'eventsource.sharedworker': [
       fileURLToPath(new URL('web_src/js/features/eventsource.sharedworker.ts', import.meta.url)),
     ],