nuclei/pkg/protocols/headless/engine/engine.go

package engine

import (
	"fmt"
	"net/http"
	"os"
	"strings"
	"sync"

	"github.com/go-rod/rod"
	"github.com/go-rod/rod/lib/launcher"
	"github.com/go-rod/rod/lib/launcher/flags"
	"github.com/pkg/errors"

	"github.com/projectdiscovery/gologger"
	"github.com/projectdiscovery/nuclei/v3/pkg/types"
	fileutil "github.com/projectdiscovery/utils/file"
	osutils "github.com/projectdiscovery/utils/os"
)

// Browser is a browser structure for nuclei headless module
type Browser struct {
	customAgent    string
	defaultHeaders map[string]string
	tempDir        string
	engine         *rod.Browser
	options        *types.Options
	launcher       *launcher.Launcher
	proxy          *headlessProxy

	// use getHTTPClient to get the http client
	httpClient     *http.Client
	httpClientOnce *sync.Once
}

// New creates a new nuclei headless browser module
func New(options *types.Options) (*Browser, error) {
	var launcherURL, dataStore string
	var proxyServer *headlessProxy
	var err error

	var chromeLauncher *launcher.Launcher

	if options.CDPEndpoint == "" {
		dataStore, err = os.MkdirTemp("", "nuclei-*")
		if err != nil {
			return nil, errors.Wrap(err, "could not create temporary directory")
		}

		proxyServer, err = newHeadlessProxy(options)
		if err != nil {
			return nil, err
		}

		noSandbox := MustDisableSandbox()
		chromeLauncher, err = newChromeLauncher(options, dataStore, proxyServer.Addr(), noSandbox)
		if err != nil {
			_ = proxyServer.Close()
			return nil, err
		}

		launcherURL, err = chromeLauncher.Launch()
		if err != nil && !noSandbox && !chromeLauncher.Has(flags.NoSandbox) && isSandboxLaunchError(err) {
			logger := options.Logger
			if logger == nil {
				logger = gologger.DefaultLogger
			}
			logger.Warning().Msgf("Could not start the browser with its sandbox, starting it without sandbox. To keep the sandbox, allow unprivileged user namespaces (for example sysctl kernel.unprivileged_userns_clone=1 or kernel.apparmor_restrict_unprivileged_userns=0) or in docker use a seccomp profile that allows user namespaces (for example --security-opt seccomp=unconfined): %s", sandboxLine(err.Error()))
			chromeLauncher.Kill()
			chromeLauncher, err = newChromeLauncher(options, dataStore, proxyServer.Addr(), true)
			if err == nil {
				launcherURL, err = chromeLauncher.Launch()
			}
		}
		if err != nil {
			_ = proxyServer.Close()
			return nil, err
		}
	} else {
		launcherURL = options.CDPEndpoint
	}

	browser := rod.New().ControlURL(launcherURL)
	if browserErr := browser.Connect(); browserErr != nil {
		if proxyServer != nil {
			_ = proxyServer.Close()
		}
		return nil, browserErr
	}
	defaultHeaders := make(map[string]string)
	customAgent := ""
	for _, option := range options.CustomHeaders {
		parts := strings.SplitN(option, ":", 2)
		if len(parts) != 2 {
			continue
		}
		if strings.EqualFold(parts[0], "User-Agent") {
			customAgent = parts[1]
		} else {
			k := strings.TrimSpace(parts[0])
			v := strings.TrimSpace(parts[1])
			if k == "" || v == "" {
				continue
			}
			defaultHeaders[k] = v
		}
	}

	engine := &Browser{
		tempDir:        dataStore,
		customAgent:    customAgent,
		defaultHeaders: defaultHeaders,
		engine:         browser,
		options:        options,
		httpClientOnce: &sync.Once{},
		launcher:       chromeLauncher,
		proxy:          proxyServer,
	}
	return engine, nil
}

// MustDisableSandbox reports whether the browser must run without its sandbox.
// Chrome refuses to start sandboxed as root on Linux, so only that case needs
// "--no-sandbox". Other users keep the sandbox; New falls back to no sandbox
// with a warning when the system can not provide one.
// https://github.com/chromium/chromium/blob/c4d3c31083a2e1481253ff2d24298a1dfe19c754/chrome/test/chromedriver/client/chromedriver.py#L209
func MustDisableSandbox() bool {
	return mustDisableSandbox(osutils.IsLinux(), os.Geteuid())
}

func mustDisableSandbox(linux bool, euid int) bool {
	return linux && euid == 0
}

// newChromeLauncher returns the launcher for a local browser. User headless
// options are applied last, so "-headless-options no-sandbox" still works.
func newChromeLauncher(options *types.Options, dataStore, proxyAddr string, noSandbox bool) (*launcher.Launcher, error) {
	chromeLauncher := launcher.New().
		Leakless(false).
		Set("disable-crash-reporter").
		Set("disable-gpu").
		Set("disable-notifications").
		Set("hide-scrollbars").
		Set("ignore-certificate-errors").
		Set("ignore-ssl-errors").
		Set("incognito").
		Set("mute-audio").
		Set("window-size", fmt.Sprintf("%d,%d", 1080, 1920)).
		Delete("use-mock-keychain").
		UserDataDir(dataStore).
		// rod adds "--no-sandbox" by default inside containers, so the
		// sandbox is always set explicitly.
		NoSandbox(noSandbox)

	executablePath, err := os.Executable()
	if err != nil {
		return nil, err
	}

	// if musl is used, most likely we are on alpine linux which is not supported by go-rod, so we fallback to default chrome
	useMusl, _ := fileutil.UseMusl(executablePath)
	if options.UseInstalledChrome || useMusl {
		if chromePath, hasChrome := launcher.LookPath(); hasChrome {
			chromeLauncher.Bin(chromePath)
		} else {
			return nil, errors.New("the chrome browser is not installed")
		}
	}

	chromeLauncher = chromeLauncher.
		Headless(!options.ShowBrowser).
		Proxy(proxyAddr).
		Set("proxy-bypass-list", "<-loopback>")

	for k, v := range options.ParseHeadlessOptionalArguments() {
		chromeLauncher.Set(flags.Flag(k), v)
	}
	return chromeLauncher, nil
}

// isSandboxLaunchError reports whether the browser exited because its
// sandbox could not be set up, for example "No usable sandbox!".
func isSandboxLaunchError(err error) bool {
	return err != nil && strings.Contains(strings.ToLower(err.Error()), "sandbox")
}

// sandboxLine returns the browser output line about the sandbox.
func sandboxLine(value string) string {
	for _, line := range strings.Split(value, "\n") {
		if strings.Contains(strings.ToLower(line), "sandbox") {
			return strings.TrimSpace(line)
		}
	}
	return strings.TrimSpace(value)
}

// SetUserAgent sets custom user agent to the browser
func (b *Browser) SetUserAgent(customUserAgent string) {
	b.customAgent = customUserAgent
}

// UserAgent fetch the currently set custom user agent
func (b *Browser) UserAgent() string {
	return b.customAgent
}

// applyDefaultHeaders setsheaders passed via cli -H flag
func (b *Browser) applyDefaultHeaders(p *rod.Page) error {
	pairs := make([]string, 0, len(b.defaultHeaders)*2+2)

	hasAcceptLanguage := false
	for k := range b.defaultHeaders {
		if strings.EqualFold(k, "Accept-Language") {
			hasAcceptLanguage = true
			break
		}
	}
	if !hasAcceptLanguage {
		pairs = append(pairs, "Accept-Language", "en, en-GB, en-us;")
	}
	for k, v := range b.defaultHeaders {
		pairs = append(pairs, k, v)
	}
	if len(pairs) == 0 {
		return nil
	}
	_, err := p.SetExtraHeaders(pairs)
	return err
}

func (b *Browser) getHTTPClient() (*http.Client, error) {
	var err error
	b.httpClientOnce.Do(func() {
		b.httpClient, err = newHttpClient(b.options)
	})
	return b.httpClient, err
}

// Close closes the browser engine
//
// When connected over CDP, it does NOT close the browsers.
func (b *Browser) Close() {
	if b.proxy != nil {
		_ = b.proxy.Close()
	}
	if b.options.CDPEndpoint != "" {
		return
	}

	_ = b.engine.Close()
	b.launcher.Kill()
	_ = os.RemoveAll(b.tempDir)
}