聊聊directory traversal attack

序

本文主要研究一下directory traversal attack及其防范

directory traversal attack

又称Path Traversal attack，即目录遍历攻击，旨在访问web服务器根目录外的文件/目录。通过是通过url或变量里头传递"../"来进行目录遍历。

通过url

比如

http://some_site.com.br/../../../../some dir/some file 

或者

http://some_site.com.br/../../../../etc/shadow 

通过变量名

通常是在文件下载接口中，比如

http://some_site.com.br/get-files?file=/etc/passwd 

或者

http://some_site.com.br/get-files?file=../../../../some dir/some file 

防范

针对url

spring security提供了DefaultHttpFirewall来进行处理，是为了防止一些web框架没有遵循servlet规范而进行的防范。
spring-security-web-4.2.3.RELEASE-sources.jar!/org/springframework/security/web/firewall/DefaultHttpFirewall.java

/**
 * Default implementation which wraps requests in order to provide consistent
 * values of the {@code servletPath} and {@code pathInfo}, which do not contain
 * path parameters (as defined in
 * <a href="http://www.ietf.org/rfc/rfc2396.txt">RFC 2396</a>). Different
 * servlet containers interpret the servlet spec differently as to how path
 * parameters are treated and it is possible they might be added in order to
 * bypass particular security constraints. When using this implementation, they
 * will be removed for all requests as the request passes through the security
 * filter chain. Note that this means that any segments in the decoded path
 * which contain a semi-colon, will have the part following the semi-colon
 * removed for request matching. Your application should not contain any valid
 * paths which contain semi-colons.
 * <p>
 * If any un-normalized paths are found (containing directory-traversal
 * character sequences), the request will be rejected immediately. Most
 * containers normalize the paths before performing the servlet-mapping, but
 * again this is not guaranteed by the servlet spec.
 *
 * @author Luke Taylor
 */
public class DefaultHttpFirewall implements HttpFirewall {
    private boolean allowUrlEncodedSlash;

    @Override
    public FirewalledRequest getFirewalledRequest(HttpServletRequest request) throws RequestRejectedException {
        FirewalledRequest fwr = new RequestWrapper(request);

        if (!isNormalized(fwr.getServletPath()) || !isNormalized(fwr.getPathInfo())) {
            throw new RequestRejectedException("Un-normalized paths are not supported: " + fwr.getServletPath()
                    + (fwr.getPathInfo() != null ? fwr.getPathInfo() : ""));
        }

        String requestURI = fwr.getRequestURI();
        if (containsInvalidUrlEncodedSlash(requestURI)) {
            throw new RequestRejectedException("The requestURI cannot contain encoded slash. Got " + requestURI);
        }

        return fwr;
    }

    @Override
    public HttpServletResponse getFirewalledResponse(HttpServletResponse response) {
        return new FirewalledResponse(response);
    }

    /**
     * <p>
     * Sets if the application should allow a URL encoded slash character.
     * </p>
     * <p>
     * If true (default is false), a URL encoded slash will be allowed in the
     * URL. Allowing encoded slashes can cause security vulnerabilities in some
     * situations depending on how the container constructs the
     * HttpServletRequest.
     * </p>
     *
     * @param allowUrlEncodedSlash
     *            the new value (default false)
     */
    public void setAllowUrlEncodedSlash(boolean allowUrlEncodedSlash) {
        this.allowUrlEncodedSlash = allowUrlEncodedSlash;
    }

    private boolean containsInvalidUrlEncodedSlash(String uri) {
        if (this.allowUrlEncodedSlash || uri == null) {
            return false;
        }

        if (uri.contains("%2f") || uri.contains("%2F")) {
            return true;
        }

        return false;
    }

    /**
     * Checks whether a path is normalized (doesn't contain path traversal
     * sequences like "./", "/../" or "/.")
     *
     * @param path
     *            the path to test
     * @return true if the path doesn't contain any path-traversal character
     *         sequences.
     */
    private boolean isNormalized(String path) {
        if (path == null) {
            return true;
        }

        for (int j = path.length(); j > 0;) {
            int i = path.lastIndexOf('/', j - 1);
            int gap = j - i;

            if (gap == 2 && path.charAt(i + 1) == '.') {
                // ".", "/./" or "/."
                return false;
            } else if (gap == 3 && path.charAt(i + 1) == '.' && path.charAt(i + 2) == '.') {
                return false;
            }

            j = i;
        }

        return true;
    }

}

这里会对url进行判断

通过变量

这种框架没有内置进行判断，需要自己在开发应用服务的时候额外关注。这里谈谈几种防范方法。

• 对变量名进行过滤

final Pattern INVALID_PATH_PATTERN = Pattern.compile("(\\.\\.\\/|\\.\\.\\\\)");
if(INVALID_PATH_PATTERN.matcher(path).find()){
    throw new BadRequestException("invalid path");
}

• 利用absolutePath与canonicalPath

absolutePath不会处理../之类的，而canonicalPath会翻译../，判断两者是否相等即可判断是否有../

        if (!file.getAbsolutePath().equals(file.getCanonicalPath())) {
            throw new BadRequestException("invalid path");
        }

小结

在编写文件下载服务的时候，需要特别关注directory traversal attack。通常url层面的web框架会帮你防范，但是变量层面的，则需要自己开发额外注意。

doc

• Path Traversal
• Directory traversal attack