htmLawed documentation

1.1  Example uses

(to top)

  *  Filtering of text submitted as comments on blogs to allow only certain HTML elements

  *  Making RSS newsfeed items standard-compliant: often one uses an excerpt from an HTML document for the content, and with unbalanced tags, non-numerical entities, etc., such excerpts may not be XML-compliant

  *  Beautifying or pretty-printing HTML code

  *  Text processing for stricter XML standard-compliance: e.g., to have lowercased x in hexadecimal numeric entities becomes necessary if an HTML document with MathML content needs to be served as application/xml

  *  Scraping text from web-pages

  *  Transforming an HTML element to another

1.2  Features

(to top)

  Key: * security feature, ^ standard compliance, ~ requires setting right options

  htmLawed:

  *  makes input more secure and standard-compliant for HTML as well as generic XML documents  ^
  *  supports markup for HTML 5, custom elements, and microdata, ARIA, Ruby, custom attributes, etc.  ^
  *  can beautify or compact HTML  ~
  *  works with input of almost any character encoding and does not affect it
  *  has good tolerance for ill-written HTML

  *  can enforce restricted use of elements  *~
  *  ensures proper closure of empty elements like img  ^
  *  transforms deprecated elements like font  ^~
  *  can permit HTML comments and CDATA sections  ^~
  *  can permit all elements, including script, object and form  ~

  *  can restrict attributes by element  ^~
  *  removes invalid attributes  ^
  *  lower-cases element and attribute names  ^
  *  provides required attributes, like alt for image  ^
  *  transforms deprecated attributes  ^~
  *  ensures attributes are declared only once  ^
  *  permits custom, non-standard attributes as well as custom rules for standard attributes  ~

  *  declares value for empty (minimized or boolean) attributes like checked  ^
  *  checks for potentially dangerous attribute values  *~
  *  ensures unique id attribute values  ^~
  *  double-quotes attribute values  ^
  *  lower-cases standard attribute values like password  ^

  *  can restrict URL protocol/scheme by attribute  *~
  *  can disable dynamic expressions in style values  *~

  *  neutralizes invalid named character entities  ^
  *  converts hexadecimal numeric entities to decimal ones, or vice versa  ^~
  *  converts named entities to numeric ones for generic XML use  ^~

  *  removes null characters  *
  *  neutralizes potentially dangerous proprietary Netscape Javascript entities  *
  *  replaces potentially dangerous soft-hyphen character in URL-accepting attribute values with spaces  *

  *  removes common invalid characters not allowed in HTML or XML  ^
  *  replaces characters from Microsoft applications like Word that are discouraged in HTML or XML  ^~
  *  neutralize entities for characters invalid or discouraged in HTML or XML  ^
  *  appropriately neutralize <, &, ", and > characters  ^*

  *  understands improperly spaced tag content (e.g., spread over more than a line) and properly spaces them
  *  attempts to balance tags for well-formedness  ^~
  *  understands when omitable closing tags like </p> are missing  ^~
  *  attempts to permit only validly nested tags  ^~
  *  can either remove or neutralize bad content ^~
  *  attempts to rectify common errors of plain-text misplacement (e.g., directly inside blockquote) ^~

  *  has optional anti-spam measures such as addition of rel="nofollow" and link-disabling  ~
  *  optionally makes relative URLs absolute, and vice versa  ~

  *  optionally marks & to identify the entities for &, < and > introduced by it  ~

  *  allows deployment of powerful hook functions to inject HTML, consolidate style attributes to class, finely check attribute values, etc.  ~

1.3  History

(to top)

  htmLawed was created in 2007 for use with LabWiki, a wiki software developed at PHP Labware, as a suitable software could not be found. Existing PHP software like Kses and HTMLPurifier were deemed inadequate, slow, resource-intensive, or dependent on an extension or external application like HTML Tidy. The core logic of htmLawed, that of identifying HTML elements and attributes, was based on the Kses (version 0.2.2) HTML filter software of Ulf Harnhammar (it can still be used with code that uses Kses; see section 2.6.). Support for HTML version 5 was added in May 2013 in a beta and in February 2017 in a production release.

  See section 4.3 for a detailed log of changes in htmLawed over the years, and section 4.10 for acknowledgements.

1.4  License & copyright

(to top)

  htmLawed is free and open-source software, copyrighted by Santosh Patnaik, MD, PhD, and dual-licensed with LGPL version 3, and GPL version 2 (or later) licenses.

1.5  Terms used here

(to top)

  In this document, only HTML body-level elements are considered. htmLawed does not have support for head-level elements, body, and the frame-level elements, frameset, frame and noframes, and these elements are ignored here.

  *  administrator - or admin; person setting up the code that utilizes htmLawed; also, user
  *  attributes - name-value pairs like href="http://x.com" in opening tags
  *  author - see writer
  *  character - atomic unit of text; internally represented by a numeric code-point as specified by the encoding or charset in use
  *  entity - markup like &gt; and &#160; used to refer to a character
  *  element - HTML element like a and img
  *  element content -  content between the opening and closing tags of an element, like click of the <a href="x">click</a> element
  *  HTML - implies XHTML unless specified otherwise
  *  HTML body - content in the body container of an HTML document
  *  input - text given to htmLawed to process
  *  legal – standard-compliant; also, valid
  *  processing - involves filtering, correction, etc., of input
  *  safe - absence or reduction of certain characters and HTML elements and attributes in HTML of text that can otherwise potentially, and circumstantially, expose text readers to security vulnerabilities like cross-site scripting attacks (XSS)
  *  scheme - a URL protocol like http and ftp
  *  specification - detailed description including rules that define HTML
  *  standard – widely accepted specification
  *  style property - terms like border and height for which declarations are made in values for the style attribute of elements
  *  tag - markers like <a href="x"> and </a> delineating element content; the opening tag can contain attributes
  *  tag content - consists of tag markers < and >, element names like div, and possibly attributes
  *  user - administrator
  *  valid - see legal
  *  writer - end-user like a blog commenter providing the input that is to be processed; also, author
  *  XHTML - XML-compliant HTML; parsing rules for XHTML are more strict than for regular HTML

1.6  Availability

(to top)

  htmLawed can be downloaded for free at its website. Besides the htmLawed.php file, the download has the htmLawed documentation (this document) in plain text and HTML formats, a script for testing, and a text file for test-cases. htmLawed can be installed with Composer, and is also available as a PHP class (OOP code) – see the website. Official htmLawed releases are also put up on Sourceforge.

2.1  Simple

(to top)

  The input text to be processed, $text, is passed as an argument of type string; htmLawed() returns the processed string:

    $processed = htmLawed($text);

  With the htmLawed class (section 1.6), usage is:

    $processed = htmLawed::hl($text);

  Notes: (1) If input is from a $_GET or $_POST value, and magic quotes are enabled on the PHP setup, run stripslashes() on the input before passing to htmLawed. (2) htmLawed does not have support for head-level elements, body, and the frame-level elements, frameset, frame and noframes.

  By default, htmLawed will process the text allowing all valid HTML elements/tags and commonly used URL schemes and CSS style properties. It will allow Javascript code, CDATA sections and HTML comments, balance tags, and ensure proper nesting of elements. Such actions can be configured using two other optional arguments -- $config and $spec:

    $processed = htmLawed($text, $config, $spec);

  The $config and $spec arguments are detailed below. Some examples are shown in section 2.9. For maximum protection against XSS and other security vulnerabilities, consider using the safe parameter; see section 3.6.

2.2  Configuring htmLawed using the $config argument

(to top)

  $config instructs htmLawed on how to tackle certain tasks. When $config is not specified, or not set as an array (e.g., $config = 1), htmLawed will take default actions. One or many of the task-action or parameter-value pairs can be specified in $config as array key-value pairs. If a parameter is not specified, htmLawed will use the default value for it, indicated further below. In PHP code, parameter values that are integers should not be quoted and should be used as numeric types (unless meant as string/text). Thus, for instance:

    $config = array('comment'=>0, 'cdata'=>1, 'elements'=>'a, b, strong');
    $processed = htmLawed($text, $config);

  Below are the various parameters that can be specified in $config.

  Key: * default, ^ different from htmLawed versions below 1.2, ~ different default when valid_xhtml is set to 1 (see section 3.5), " different default when safe is set to 1 (see section 3.6)

  abs_url
  Make URLs absolute or relative; $config["base_url"] needs to be set; see section 3.4.4

  -1 - make relative
  0 - no action  *
  1 - make absolute

  and_mark
  Mark & characters in the original input; see section 3.2

  anti_link_spam
  Anti-link-spam measure; see section 3.4.7

  0 - no measure taken  *
  array("regex1", "regex2") - will ensure a rel attribute with nofollow in its value in case the href attribute value matches the regular expression pattern regex1, and/or will remove href if its value matches the regular expression pattern regex2. E.g., array("/./", "/://\W*(?!(abc\.com|xyz\.org))/"); see section 3.4.7 for more.

  anti_mail_spam
  Anti-mail-spam measure; see section 3.4.7

  0 - no measure taken  *
  word - @ in mail address in href attribute value is replaced with specified word

  any_custom_element
  Permit any custom element; regardless of this setting, specific custom elements can be denied or permitted through $config["elements"]; see section 3.3.6

  0 - no
  1 - yes  *

  balance
  Balance tags for well-formedness and proper nesting; see section 3.3.3

  0 - no
  1 - yes  *

  base_url
  Base URL value that needs to be set if $config["abs_url"] is not 0; see section 3.4.4

  cdata
  Handling of CDATA sections; see section 3.3.1

  0 - don't consider CDATA sections as markup and proceed as if plain text  "
  1 - remove
  2 - allow, but neutralize any <, >, and & inside by converting them to named entities
  3 - allow  *

  clean_ms_char
  Replace discouraged characters introduced by Microsoft Word, etc.; see section 3.1

  0 - no  *
  1 - yes
  2 - yes, but replace special single & double quotes with ordinary ones

  comment
  Handling of HTML comments; see section 3.3.1

  0 - don't consider comments as markup and proceed as if plain text  "
  1 - remove
  2 - allow, but neutralize any <, >, and & inside by converting to named entities
  3 - allow  *

  css_expression
  Allow dynamic CSS expression by not removing the expression from CSS property values in style attributes; see section 3.4.8

  0 - remove  *
  1 - allow

  deny_attribute
  Denied HTML attributes; see section 3.4

  0 - none  *
  string - dictated by values in string
  on* - on* event attributes like onfocus not allowed  "

  direct_nest_list
  Allow direct nesting of a list within another without requiring it to be a list item; see section 3.3.4

  0 - no  *
  1 - yes

  elements
  Allowed HTML elements; see section 3.3

  all - *^
  * -acronym -big -center -dir -font -isindex -s -strike -tt -  ~^
  applet, audio, canvas, dialog, embed, iframe, object, script, and video elements not allowed -  "^

  hexdec_entity
  Allow hexadecimal numeric entities and do not convert to the more widely accepted decimal ones, or convert decimal to hexadecimal ones; see section 3.2

  0 - no
  1 - yes  *
  2 - convert decimal to hexadecimal ones

  hook
  Name of an optional hook function to alter the input string, $config or $spec before htmLawed enters the main phase of its work; see section 3.7

  0 - no hook function  *
  name - name is name of the hook function

  hook_tag
  Name of an optional hook function to alter tag content finalized by htmLawed; see section 3.4.9

  0 - no hook function  *
  name - name is name of the hook function

  keep_bad
  Neutralize bad tags by converting their < and > characters to entities, or remove them; see section 3.3.3

  0 - remove
  1 - neutralize both tags and element content
  2 - remove tags but neutralize element content
  3 and 4 - like 1 and 2 but remove if text (pcdata) is invalid in parent element
  5 and 6 * -  like 3 and 4 but line-breaks, tabs and spaces are left

  lc_std_val
  For XHTML compliance, predefined, standard attribute values, like get for the method attribute of form, must be lowercased; see section 3.4.5

  0 - no
  1 - yes  *

  make_tag_strict
  Transform or remove these deprecated HTML elements, even if they are allowed by the admin: acronym, applet, big, center, dir, font, isindex, s, strike, tt; see section 3.3.2

  0 - no
  1 - yes, but leave applet and isindex that currently cannot be transformed  *^
  2 - yes, removing applet and isindex elements and their contents (nested elements remain)  ~^

  named_entity
  Allow non-universal named HTML entities, or convert to numeric ones; see section 3.2

  0 - convert
  1 - allow  *

  no_deprecated_attr
  Allow deprecated attributes or transform them; see section 3.4.6

  0 - allow
  1 - transform, but name attributes for a and map are retained  *
  2 - transform

  parent
  Name of the parent element, possibly imagined, that will hold the input; see section 3.3

  safe
  Magic parameter to make input the most secure against vulnerabilities like XSS without needing to specify other relevant $config parameters; see section 3.6

  0 - no  *
  1 - will auto-adjust other relevant $config parameters (indicated by " in this list)  ^

  schemes
  Array of attribute-specific, comma-separated, lower-cased list of schemes (protocols) allowed in attributes accepting URLs (or ! to deny any URL); * covers all unspecified attributes; see section 3.4.3

  href: aim, app, feed, file, ftp, gopher, http, https, javascript, irc, mailto, news, nntp, sftp, ssh, tel, telnet, ws, wss; *:data, file, http, https, javascript, ws, wss  *^
  href: aim, feed, file, ftp, gopher, http, https, irc, mailto, news, nntp, sftp, ssh, tel, telnet, ws, wss; style: !; *:file, http, https, ws, wss  "

  show_setting
  Name of a PHP variable to assign the finalized $config and $spec values; see section 3.8

  style_pass
  Ignore style attribute values, letting them through without any alteration

  0 - no *
  1 - htmLawed will let through any style value; see section 3.4.8

  tidy
  Beautify or compact HTML code; see section 3.3.5

  -1 - compact
  0 - no  *
  1 or string - beautify (custom format specified by string)

  unique_ids
  id attribute value checks; see section 3.4.2

  0 - no
  1 - remove duplicate and/or invalid ones  *
  word - remove invalid ones and replace duplicate ones with new and unique ones based on the word; the admin-specified word cannot contain a space character

  valid_xhtml
  Magic parameter to make input the most valid XHTML without needing to specify other relevant $config parameters; see section 3.5

  0 - no  *
  1 - will auto-adjust other relevant $config parameters (indicated by ~ in this list)

  xml:lang
  Auto-add xml:lang attribute; see section 3.4.1

  0 - no  *
  1 - add if lang attribute is present
  2 - add if lang attribute is present, and remove lang  ~

2.3  Extra HTML specifications using the $spec parameter

(to top)

  The $spec argument of htmLawed can be used to disallow an otherwise legal attribute for an element, or to restrict the attribute's values. This can also be helpful as a security measure (e.g., in certain versions of browsers, certain values can cause buffer overflows and denial of service attacks), or in enforcing admin policies. $spec is specified as a string of text containing one or more rules, with multiple rules separated from each other by a semi-colon (;). E.g.,

    $spec = 'i=-*; td, tr=style, id, -*; a=id(match="/[a-z][a-z\d.:\-`"]*/i"/minval=2), href(maxlen=100/minlen=34); img=-width,-alt';
    $processed = htmLawed($text, $config, $spec);

  Or,

    $processed = htmLawed($text, $config, 'i=-*; td, tr=style, id, -*; a=id(match="/[a-z][a-z\d.:\-`"]*/i"/minval=2), href(maxlen=100/minlen=34); img=-width,-alt');

  A rule begins with an HTML element name(s) (rule-element), for which the rule applies, followed by an equal-to (=) sign. A rule-element may represent multiple elements if comma (,)-separated element names are used. E.g., th,td,tr=.

  Rest of the rule consists of comma-separated HTML attribute names, which can be the wildcard references *, aria*, data*, and on* for the sets of all standard, Aria, data-*, and event (on*) attributes, respectively. A minus (-) character before an attribute means that the attribute is not permitted inside the rule-element. E.g., -width. To deny all attributes, -* can be used. All Aria, data-*, and event (on*) attributes can similarly be denined using aria*, data*, and on*, respectively.

  Following shows examples of rule excerpts with rule-element a and the attributes that are being permitted:

  *  a= - all
  *  a=id - all
  *  a=href, title, -id, -onclick - all except id and onclick
  *  a=*, id, -id - all except id
  *  a=-* - none
  *  a=-*, href, title - none except href and title
  *  a=-*, -id, href, title - none except href and title
  *  a=-on*, -id, href, onclick, title - all except id and on* other than onclick

  Rules regarding attribute values are optionally specified inside round brackets after attribute names – which cannot be wildcard references like * or data* – in solidus (/)-separated parameter = value pairs. E.g., title(maxlen=30/minlen=5). None or one or more of the following parameters may be specified:

  *  oneof - one or more choices separated by | that the value should match; if only one choice is provided, then the value must match that choice; matching is case-sensitive

  *  noneof - one or more choices separated by | that the value should not match; matching is case-sensitive

  *  maxlen and minlen - upper and lower limits for the number of characters in the attribute value; specified in numbers

  *  maxval and minval - upper and lower limits for the numerical value specified in the attribute value; specified in numbers

  *  match and nomatch - pattern that the attribute value should or should not match; specified as PHP/PCRE-compatible regular expressions with delimiters and possibly modifiers (e.g., to specify case-sensitivity for matching)

  *  default - a value to force on the attribute if the value provided by the writer does not fit any of the specified parameters

  If default is not set and the attribute value does not satisfy any of the specified parameters, then the attribute is removed. The default value can also be used to force all attribute declarations to take the same value (by getting the values declared illegal by setting, e.g., maxlen to -1).

  Examples with input <input title="WIDTH" value="10em" /><input title="length" value="5" class="ic1 ic2" /> are shown below.

  Rule: input=title(maxlen=60/minlen=6), value
  Output: <input value="10em" /><input title="length" value="5" class="ic1 ic2" />

  Rule: input=title(), value(maxval=8/default=6)
  Output: <input title="WIDTH" value="6" /><input title="length" value="5" class="ic1 ic2" />

  Rule: input=title(nomatch=%w.d%i), value(match=%em%/default=6em)
  Output: <input value="10em" /><input title="length" value="6em" class="ic1 ic2" />

  Rule: input=class(noneof=ic2|ic3/oneof=ic1|ic4), title(oneof=height|depth/default=depth), value(noneof=5|6)
  Output: <input title="depth" value="10em" /><input title="depth" class="ic1" />

  Special characters: The characters ;, ,, /, (, ), |, ~ and space have special meanings in the rules. Words in the rules that use such characters, or the characters themselves, should be escaped by enclosing in pairs of double-quotes ("). A back-tick (`) can be used to escape a literal ". An example rule illustrating this is input=value(maxlen=30/match="/^\w/"/default="your `"ID`"").

  Attributes that accept multiple values: If an attribute is accesskey, class, itemtype or rel, or archive in case of object element, which can have multiple, space-separated values, or archive in case of object element and srcset, which can have multiple, comma-separated values, htmLawed will parse the attribute value for such multiple values and will individually test each of them. The parsing is performed after any URL assessment of the attribute values (section 3.4.3).

  Note: To deny an attribute for all elements for which it is legal, $config["deny_attribute"] (see section 3.4) can be used instead of $spec. Also, attributes can be allowed element-specifically through $spec while being denied globally through $config["deny_attribute"]. The hook_tag parameter (section 3.4.9) can also be possibly used to implement a functionality like that achieved using $spec functionality.

  Note: Attributes permitted through $spec are permitted regardless of any denial through $config. An attribute for which $spec indicates both permission and denial will be permitted. E.g., onclick with $spec value of a = *, -onclick, onclick, a = -on*, onclick or a = on*, -onclick will be permitted inside a.

  Note: Attributes' specifications for an element may be (inadvertently) set through multiple rules. In case of conflict, the attribute specification in the first rule will get precedence.

  $spec can also be used to permit custom or non-standard attributes. Thus, the following value of $spec will permit the custom uses of the standard rel attribute in input (not permitted as per standards) and of a non-standard attribute, vFlag, in img.

    $spec = 'img=vFlag; input=rel'

  The attribute names must begin with an alphabet and cannot have space, equal-to (=) and solidus (/) characters.

2.4  Performance time & memory usage

(to top)

  The time and memory consumed during text processing by htmLawed depends on its configuration, the size of the input, and the amount, nestedness and well-formedness of the HTML markup within the input. In particular, tag balancing and beautification each can increase the processing time by about a quarter.

  The htmLawed demo can be used to evaluate the performance and effects of different types of input and $config.

2.5  Some security risks to keep in mind

(to top)

  When setting the parameters/arguments (like those to allow certain HTML elements) for use with htmLawed, one should bear in mind that the setting may let through potentially dangerous HTML code which is meant to steal user-data, deface a website, render a page non-functional, etc. Unless end-users, either people or software, supplying the content are completely trusted, security issues arising from the degree of HTML usage permitted through htmLawed's setting should be considered. For example, following increase security risks:

  *  Allowing script, applet, embed, iframe, canvas, audio, video, dialog or object elements, or certain of their attributes like allowscriptaccess

  *  Allowing HTML comments (some Internet Explorer versions are vulnerable with, e.g., <!--[if gte IE 4]><script>alert("xss");</script><![endif]-->

  *  Allowing dynamic CSS expressions (some Internet Explorer versions are vulnerable)

  *  Allowing the style attribute

  To remove unsecure HTML, code-developers using htmLawed must set $config appropriately. E.g., $config["elements"] = "* -script" to deny the script element (section 3.3), $config["safe"] = 1 to auto-configure ceratin htmLawed parameters for maximizing security (section 3.6), etc.

  Permitting the *style* attribute brings in risks of click-jacking, phishing, web-page overlays, etc., even when the safe parameter is enabled (see section 3.6). Except for URLs and a few other things like CSS dynamic expressions, htmLawed currently does not check every CSS style property. It does provide ways for the code-developer implementing htmLawed to do such checks through htmLawed's $spec argument, and through the hook_tag parameter (see section 3.4.8 for more). Disallowing style completely and relying on CSS classes and stylesheet files is recommended.

  htmLawed does not check or correct the character encoding of the input it receives. In conjunction with permissive circumstances, such as when the character encoding is left undefined through HTTP headers or HTML meta tags, this can allow for an exploit (like Google's UTF-7/XSS vulnerability of the past).

  Ocassionally, though very rarely, the default settings with which htmLawed runs may change between different versions of htmLawed. Admins should keep this in mind when upgrading htmLawed. Important changes in htmLawed's default behavior in new releases of the software are noted in section 4.5 on upgrades.

2.6  Use with kses() code

(to top)

  The Kses PHP script for HTML filtering is used by many applications (like WordPress, as in year 2012). It is possible to have such applications use htmLawed instead, since it is compatible with code that calls the kses() function declared in the Kses file (usually named kses.php). E.g., application code like this will continue to work after replacing Kses with htmLawed:

    $comment_filtered = kses($comment_input, array('a'=>array(), 'b'=>array(), 'i'=>array()));

  If the application uses a Kses file that has the kses() function declared, then, to have the application use htmLawed instead of Kses, rename htmLawed.php (to kses.php, e.g.) and replace the Kses file (or just replace the code in the Kses file with the htmLawed code). If the kses() function in the Kses file had been renamed by the application developer (e.g., in WordPress, it is named wp_kses()), then appropriately rename the kses() function in the htmLawed code. Then, add the following code (which was a part of htmLawed prior to version 1.2):

    // kses compatibility
    function kses($t, $h, $p=array('http', 'https', 'ftp', 'news', 'nntp', 'telnet', 'gopher', 'mailto')){
     foreach($h as $k=>$v){
      $h[$k]['n']['*'] = 1;
     }
     $C['cdata'] = $C['comment'] = $C['make_tag_strict'] = $C['no_deprecated_attr'] = $C['unique_ids'] = 0;
     $C['keep_bad'] = 1;
     $C['elements'] = count($h) ? strtolower(implode(',', array_keys($h))) : '-*';
     $C['hook'] = 'kses_hook';
     $C['schemes'] = '*:'. implode(',', $p);
     return htmLawed($t, $C, $h);
     }

    function kses_hook($t, &$C, &$S){
     return $t;
    }

  If the Kses file used by the application has been significantly altered by the application developers, then one may need a different approach. E.g., with WordPress (as in the year 2012), it is best to copy the htmLawed code, along with the above-mentioned additions, to wp_includes/kses.php, rename the newly added function kses() to wp_kses(), and delete the code for the original wp_kses() function.

  If the Kses code has a non-empty hook function (e.g., wp_kses_hook() in case of WordPress), then the code for htmLawed's kses_hook() function should be appropriately edited. However, the requirement of the hook function should be re-evaluated considering that htmLawed has extra capabilities. With WordPress, the hook function is an essential one. The following code is suggested for the htmLawed kses_hook() in case of WordPress:

    // kses compatibility
    function kses_hook($string, &$cf, &$spec){
     $allowed_html = $spec;
     $allowed_protocols = array();
     foreach($cf['schemes'] as $v){
      foreach($v as $k2=>$v2){
       if(!in_array($k2, $allowed_protocols)){
        $allowed_protocols[] = $k2;
       }
      }
     }
     return wp_kses_hook($string, $allowed_html, $allowed_protocols);
    }

2.7  Tolerance for ill-written HTML

(to top)

  htmLawed can work with ill-written HTML code in the input. However, HTML that is too ill-written may not be read as HTML, and may therefore get identified as mere plain text. Following statements indicate the degree of looseness that htmLawed can work with, and can be provided in instructions to writers:

  *  Tags must be flanked by < and > with no > inside -- any needed > should be put in as &gt;. It is possible for tag content (element name and attributes) to be spread over many lines instead of being on one. A space may be present between the tag content and >, like <div > and <img / >, but not after the <.

  *  Element and attribute names need not be lower-cased.

  *  Attribute string of elements may be liberally spaced with tabs, line-breaks, etc.

  *  Attribute values may be single- and not double-quoted.

  *  Left-padding of numeric entities (like, &#0160;, &x07ff;) with 0 is okay as long as the number of characters between between the & and the ; does not exceed 8. All entities must end with ; though.

  *  Named character entities must be properly cased. Thus, &Lt; or &TILDE; will not be recognized as entities and will be neutralized.

  *  HTML comments should not be inside element tags (they can be between tags), and should begin with <!-- and end with -->. Characters like <, >, and & may be allowed inside depending on $config, but any --> inside should be put in as --&gt;. Any -- inside will be automatically converted to -, and a space will be added before the --> comment-closing marker  unless $config["comments"] is set to 4 (section 3.3.1).

  *  CDATA sections should not be inside element tags, and can be in element content only if plain text is allowed for that element. They should begin with <[CDATA[ and end with ]]>. Characters like <, >, and & may be allowed inside depending on $config, but any ]]> inside should be put in as ]]&gt;.

  *  For attribute values, character entities &lt;, &gt; and &amp; should be used instead of characters < and >, and & (when & is not part of a character entity). This applies even for Javascript code in values of attributes like onclick.

  *  Characters <, >, & and " that are part of actual Javascript, etc., code in script elements should be used as such and not be put in as entities like &gt;. Otherwise, though the HTML will be valid, the code may fail to work. Further, if such characters have to be used, then they should be put inside CDATA sections.

  *  Simple instructions like "an opening tag cannot be present between two closing tags" and "nested elements should be closed in the reverse order of how they were opened" can help authors write balanced HTML. If tags are imbalanced, htmLawed will try to balance them, but in the process, depending on $config["keep_bad"], some code/text may be lost.

  *  Input authors should be notified of admin-specified allowed elements, attributes, configuration values (like conversion of named entities to numeric ones), etc.

  *  With $config["unique_ids"] not 0 and the id attribute being permitted, writers should carefully avoid using duplicate or invalid id values as even though htmLawed will correct/remove the values, the final output may not be the one desired. E.g., when <a id="home"></a><input id="home" /><label for="home"></label> is processed into
<a id="home"></a><input id="prefix_home" /><label for="home"></label>.

  *  Even if intended HTML is lost from an ill-written input, the processed output will be more secure and standard-compliant.

  *  For URLs, unless $config["scheme"] is appropriately set, writers should avoid using escape characters or entities in schemes. E.g., htt&#112; (which many browsers will read as the harmless http) may be considered bad by htmLawed.

  *  htmLawed will attempt to put plain text present directly inside blockquote, form, map and noscript elements (illegal as per the specifications) inside auto-generated div elements during tag balancing (section 3.3.3).

2.8  Limitations & work-arounds

(to top)

  htmLawed's main objective is to make the input text more standard-compliant, secure for readers, and free of HTML elements and attributes considered undesirable by the administrator. Some of its current limitations, regardless of this objective, are noted below along with possible work-arounds.

  It should be borne in mind that no browser application is 100% standard-compliant, standard specifications continue to evolve, and many browsers accept commonly used non-standard HTML. Regarding security, note that unsafe HTML code is not legally invalid per se.

  *  htmLawed might not strictly adhere to current HTML standards as standard specification for HTML by WHATWG is continuously evolving, and there is laxity among HTML interpreters (browsers) regarding standards. Admins can configure htmLawed to be more strict about standard compliance.

  *  In general, htmLawed processes input to generate output that is most likely to be standard-compatible in most users' browsers. Thus, for example, it does not enforce the required value of 0 on border attribute of img (an HTML version 5 specification).

  *  htmLawed is meant for input that goes into the body of HTML documents. HTML's head-level elements are not supported, nor are the frame-specific elements frameset, frame and noframes. However, content of the latter elements can be individually filtered through htmLawed.

  *  It cannot handle input that has non-HTML code like SVG and MathML. One way around is to break the input into pieces and passing only those without non-HTML code to htmLawed. Another is described in section 3.9. A third way may be to some how take advantage of the $config["and_mark"] parameter (see section 3.2).

  *  By default, htmLawed won't check many attribute values for standard compliance. E.g., width="20m" with the dimension in non-standard m is let through. Implementing universal and strict attribute value checks can make htmLawed slow and resource-intensive. Admins should look at the hook_tag parameter (section 3.4.9) or $spec to enforce finer checks on attribute values.

  *  By default, htmLawed considers all ARIA, data-*, event, and microdata attributes as global attributes and permits them in all elements. This is not strictly standard-compliant. E.g., the itemtype microdata attribute is permitted only in elements that also have the itemscope attribute. Admins can configure htmLawed to be more strict about this (section 2.3).

  *  The attributes, whether deprecated (which can be transformed by htmLawed) or not, that it supports are largely those that are in the specifications. Only a few of the proprietary attributes are supported. However, $spec can be used to allow custom attributes (section 2.3).

  *  Except for contained URLs and dynamic expressions (also optional), htmLawed does not check CSS style property values. Admins should look at using the hook_tag parameter (section 3.4.9) or $spec for finer checks. Perhaps the best option is to disallow style but allow class attributes with the right oneof or match values for class, and have the various class style properties in .css CSS stylesheet files.

  *  htmLawed does not parse emoticons, decode BBcode, or wikify, auto-converting text to proper HTML. Similarly, it won't convert line-breaks to br elements. Such functions are beyond its purview. Admins should use other code to pre- or post-process the input for such purposes.

  *  htmLawed cannot be used to have links force-opened in new windows (by auto-adding appropriate target and onclick attributes to a). Admins should look at Javascript-based DOM-modifying solutions for this. Admins may also be able to use a custom hook function to enforce such checks (hook_tag parameter; see section 3.4.9).

  *  Nesting-based checks are not possible. E.g., one cannot disallow p elements specifically inside td while permitting it elsewhere. Admins may be able to use a custom hook function to enforce such checks (hook_tag parameter; see section 3.4.9).

  *  Except for optionally converting absolute or relative URLs to the other type, htmLawed will not alter URLs (e.g., to change the value of query strings or to convert http to https. Having absolute URLs may be a standard-requirement, e.g., when HTML is embedded in email messages, whereas altering URLs for other purposes is beyond htmLawed's goals. Admins may be able to use a custom hook function to enforce such checks (hook_tag parameter; see section 3.4.9).

  *  Pairs of opening and closing tags that do not enclose any content (like <em></em>) are not removed. This may be against the standard specification for certain elements (e.g., table). However, presence of such standard-incompliant code will not break the display or layout of content. Admins can also use simple regex-based code to filter out such code.

  *  htmLawed does not check for certain element orderings described in the standard specifications (e.g., in a table, tbody is allowed before tfoot). Admins may be able to use a custom hook function to enforce such checks (hook_tag parameter; see section 3.4.9).

  *  htmLawed does not check the number of nested elements. E.g., it will allow two caption elements in a table element, illegal as per standard specifications. Admins may be able to use a custom hook function to enforce such checks (hook_tag parameter; see section 3.4.9).

  *  There are multiple ways to interpret ill-written HTML. E.g., in <small><small>text</small>, is it that the second closing tag for small is missing or is it that the second opening tag for small was put in by mistake? htmLawed corrects the HTML in the string assuming the former, while the user may have intended the string for the latter. This is an issue that is impossible to address perfectly.

  *  htmLawed might convert certain entities to actual characters and remove backslashes and CSS comment-markers (/*) in style attribute values in order to detect malicious HTML like crafted, Internet Explorer browser-specific dynamic expressions like &#101;xpression.... If this is too harsh, admins can allow CSS expressions through htmLawed core but then use a custom function through the hook_tag parameter (section 3.4.9) to more specifically identify CSS expressions in the style attribute values. Also, using $config["style_pass"], it is possible to have htmLawed pass style attribute values without even looking at them (section 3.4.8).

  *  htmLawed does not correct certain possible attribute-based security vulnerabilities (e.g., <a href="http://x%22+style=%22background-image:xss">x</a>). These arise when browsers mis-identify markup in escaped text, defeating the very purpose of escaping text (a bad browser will read the given example as <a href="http://x" style="background-image:xss">x</a>).

  *  Because of inadequate Unicode support in PHP, htmLawed does not remove the high value HTML-invalid characters with multi-byte code-points. Such characters however are extremely unlikely to be in the input. (see section 3.1).

  *  htmLawed does not check or correct the character encoding of the input it receives. In conjunction with permitting circumstances such as when the character encoding is left undefined through HTTP headers or HTML meta tags, this can permit an exploit (like Google's UTF-7/XSS vulnerability of the past). Also, htmLawed can mangle input text if it is not well-formed in terms of character encoding. Administrators can consider using code available elsewhere to check well-formedness of input text characters to correct any defect.

  *  htmLawed is expected to work with input texts in ASCII standard-compatible single-byte encodings such as national variants of ASCII (like ISO-646-DE/German of the ISO 646 standard), extended ASCII variants (like ISO 8859-10/Turkish of the ISO 8859/ISO Latin standard), ISO 8859-based Windows variants (like Windows 1252), EBCDIC, Shift JIS (Japanese), GB-Roman (Chinese), and KS-Roman (Korean). It should also properly handle texts with variable-byte encodings like UTF-7 (Unicode) and UTF-8 (Unicode). However, htmLawed may mangle input texts with double-byte encodings like UTF-16 (Unicode), JIS X 0208:1997 (Japanese) and K SX 1001:1992 (Korean), or the UTF-32 (Unicode) quadruple-byte encoding. If an input text has such an encoding, administrators can use PHP's iconv functions, or some other mean, to convert text to UTF-8 before passing it to htmLawed.

  *  Like any script using PHP's PCRE regex functions, PHP setup-specific low PCRE limit values can cause htmLawed to at least partially fail with very long input texts.

2.9  Examples of usage

(to top)

  Safest, allowing only safe HTML markup --

    $config = array('safe'=>1);
    $out = htmLawed($in, $config);

  Simplest, allowing all valid HTML markup including Javascript --

    $out = htmLawed($in);

  Allowing all valid HTML markup but restricting URL schemes in src attribute values to http and https --

    $config = array('schemes'=>'*:*; src:http, https');
    $out = htmLawed($in, $config);

  Allowing only safe HTML and the elements a, em, and strong --

    $config = array('safe'=>1, 'elements'=>'a, em, strong');
    $out = htmLawed($in, $config);

  Not allowing elements script and object --

    $config = array('elements'=>'* -script -object');
    $out = htmLawed($in, $config);

  Not allowing attributes id and style --

    $config = array('deny_attribute'=>'id, style');
    $out = htmLawed($in, $config);

  Permitting only attributes title and href --

    $config = array('deny_attribute'=>'* -title -href');
    $out = htmLawed($in, $config);

  Remove bad/disallowed tags altogether instead of converting them to entities --

    $config = array('keep_bad'=>0);
    $out = htmLawed($in, $config);

  Allowing attribute title only in a and not allowing attributes id, style, or scriptable on* attributes like onclick --

    $config = array('deny_attribute'=>'title, id, style, on*');
    $spec = 'a=title';
    $out = htmLawed($in, $config, $spec);

  Allowing a custom attribute, vFlag, in img and permitting custom use of the standard attribute, rel, in input --

    $spec = 'img=vFlag; input=rel';
    $out = htmLawed($in, $config, $spec);

  Some case-studies are presented below.

  1. A blog administrator wants to allow only a, em, strike, strong and u in comments, but needs strike and u transformed to span for better XHTML 1-strict compliance, and, he wants the a links to point only to http or https resources:

    $processed = htmLawed($in, array('elements'=>'a, em, strike, strong, u', 'make_tag_strict'=>1, 'safe'=>1, 'schemes'=>'*:http, https'), 'a=href');

  2. An author uses a custom-made web application to load content on his website. He is the only one using that application and the content he generates has all types of HTML, including scripts. The web application uses htmLawed primarily as a tool to correct errors that creep in while writing HTML and to take care of the occasional bad characters in copy-paste text introduced by Microsoft Office. The web application provides a preview before submitted input is added to the content. For the previewing process, htmLawed is set up as follows:

    $processed = htmLawed($in, array('css_expression'=>1, 'keep_bad'=>1, 'make_tag_strict'=>1, 'schemes'=>'*:*', 'valid_xhtml'=>1));

  For the final submission process, keep_bad is set to 6. A value of 1 for the preview process allows the author to note and correct any HTML mistake without losing any of the typed text.

  3. A data-miner is scraping information in a specific table of similar web-pages and is collating the data rows, and uses htmLawed to reduce unnecessary markup and white-spaces:

    $processed = htmLawed($in, array('elements'=>'tr, td', 'tidy'=>-1), 'tr, td =');

3.1  Invalid/dangerous characters

(to top)

  Valid characters (more correctly, their code-points) in HTML or XML are, hexadecimally, 9, a, d, 20 to d7ff, and e000 to 10ffff, except fffe and ffff (decimally, 9, 10, 13, 32 to 55295, and 57344 to 1114111, except 65534 and 65535). htmLawed removes the invalid characters 0 to 8, b, c, and e to 1f.

  Because of PHP's poor native support for multi-byte characters, htmLawed cannot check for the remaining invalid code-points. However, for various reasons, it is very unlikely for any of those characters to be in the input.

  Characters that are discouraged (see section 5.1) but not invalid are not removed by htmLawed.

  It (function hl_tag()) also replaces the potentially dangerous (in some Mozilla [Firefox] and Opera browsers) soft-hyphen character (code-point, hexadecimally, ad, or decimally, 173) in attribute values with spaces. Where required, the characters <, >, &, and " are converted to entities.

  With $config["clean_ms_char"] set as 1 or 2, many of the discouraged characters (decimal code-points 127 to 159 except 133) that many Microsoft applications incorrectly use (as per the Windows 1252 [Cp-1252] or a similar encoding system), and the character for decimal code-point 133, are converted to appropriate decimal numerical entities (or removed for a few cases)-- see appendix in section 5.4. This can help avoid some display issues arising from copying-pasting of content.

  With $config["clean_ms_char"] set as 2, characters for the hexadecimal code-points 82, 91, and 92 (for special single-quotes), and 84, 93, and 94 (for special double-quotes) are converted to ordinary single and double quotes respectively and not to entities.

  The character values are replaced with entities/characters and not character values referred to by the entities/characters to keep this task independent of the character-encoding of input text.

  The $config["clean_ms_char"] parameter should not be used if authors do not copy-paste Microsoft-created text, or if the input text is not believed to use the Windows 1252 (Cp-1252) or a similar encoding like Cp-1251 (otherwise, for example when UTF-8 encoding is in use, Japanese or Korean characters can get mangled). Further, the input form and the web-pages displaying it or its content should have the character encoding appropriately marked-up.

3.2  Character references/entities

(to top)

  Valid character entities take the form &*; where * is #x followed by a hexadecimal number (hexadecimal numeric entity; like &#xA0; for non-breaking space), or alphanumeric like gt (external or named entity; like &nbsp; for non-breaking space), or # followed by a number (decimal numeric entity; like &#160; for non-breaking space). Character entities referring to the soft-hyphen character (the &shy; or \xad character; hexadecimal code-point ad [decimal 173]) in URL-accepting attribute values are always replaced with spaces; soft-hyphens in attribute values introduce vulnerabilities in some older versions of the Opera and Mozilla [Firefox] browsers.

  htmLawed (function hl_entity()):

  *  Neutralizes entities with multiple leading zeroes or missing semi-colons (potentially dangerous)

  *  Lowercases the X (for XML-compliance) and A-F of hexadecimal numeric entities

  *  Neutralizes entities referring to characters that are HTML-invalid (see section 3.1)

  *  Neutralizes entities referring to characters that are HTML-discouraged (code-points, hexadecimally, 7f to 84, 86 to 9f, and fdd0 to fddf, or decimally, 127 to 132, 134 to 159, and 64991 to 64976). Entities referring to the remaining discouraged characters (see section 5.1 for a full list) are let through.

  *  Neutralizes named entities that are not in the HTML5 specification

  *  Optionally converts valid HTML-specific named entities except &gt;, &lt;, &quot;, and &amp; to decimal numeric ones (hexadecimal if $config["hexdec_entity"] is 2) for generic XML-compliance. For this, $config["named_entity"] should be 1.

  *  Optionally converts hexadecimal numeric entities to the more widely supported decimal ones. For this, $config["hexdec_entity"] should be 0.

  *  Optionally converts decimal numeric entities to the hexadecimal ones. For this, $config["hexdec_entity"] should be 2.

  Neutralization refers to the entitification of & to &amp;.

  Note: htmLawed does not convert entities to the actual characters represented by them; one can pass the htmLawed output through PHP's html_entity_decode function for that.

  Note: If $config["and_mark"] is set, and set to a value other than 0, then the & characters in the original input are replaced with the control character for the hexadecimal code-point 6 (\x06; & characters introduced by htmLawed, e.g., after converting < to &lt;, are not affected). This allows one to distinguish, say, an &gt; introduced by htmLawed and an &gt; put in by the input writer, and can be helpful in further processing of the htmLawed-processed text (e.g., to identify the character sequence o(><)o to generate an emoticon image). When this feature is active, admins should ensure that the htmLawed output is not directly used in web pages or XML documents as the presence of the \x06 can break documents. Before use in such documents, and preferably before any storage, any remaining \x06 should be changed back to &, e.g., with:

    $final = str_replace("\x06", '&', $prelim);

  Also, see section 3.9.

3.3  HTML elements

(to top)

  htmLawed can be configured to allow only certain HTML elements (tags) in the input. Disallowed elements (just tag-content, and not element-content), based on $config["keep_bad"], are either neutralized (converted to plain text by entitification of < and >) or removed.

  E.g., with only em permitted:

  Input:

      <em>My</em> website is <a href="http://a.com>a.com</a>.

  Output, with $config["keep_bad"] = 0:

      <em>My</em> website is a.com.

  Output, with $config["keep_bad"] not 0:

      <em>My</em> website is &lt;a href=""&gt;a.com&lt;/a&gt;.

  See section 3.3.3 for differences between the various non-zero $config["keep_bad"] values.

  htmLawed by default permits these 122 HTML elements:

    a, abbr, acronym, address, applet, area, article, aside, audio, b, bdi, bdo, big, blockquote, br, button, canvas, caption, center, cite, code, col, colgroup, command, data, datalist, dd, del, details, dfn, dialog, dir, div, dl, dt, em, embed, fieldset, figcaption, figure, font, footer, form, h1, h2, h3, h4, h5, h6, header, hgroup, hr, i, iframe, img, input, ins, isindex, kbd, keygen, label, legend, li, link, main, map, mark, menu, meta, meter, nav, noscript, object, ol, optgroup, option, output, p, param, picture, pre, progress, q, rb, rbc, rp, rt, rtc, ruby, s, samp, script, section, select, slot, small, source, span, strike, strong, style, sub, summary, sup, table, tbody, td, template, textarea, tfoot, th, thead, time, tr, track, tt, u, ul, var, video, wbr

  htmLawed also supports use of custom HTML elements, but this support can be turned off when $config is appropriately set (i.e., in default configuration, such elements are permitted); see section 3.3.6.

  Elements math and svg are not supported. They and their content will get filtered unless a strategy like in section 3.9 is used.

  Elements like acronym, applet, basefont, bgsound, big, blink, center, command, dir, font, hgroup, image, keygen, marquee, menuitem, nobr, noembed, rb, rtc, shadow, spacer, strike, tt, and xmp are currently obsolete/deprecated. Some of them, like acronym and keygen, are supported in htmLawed (see above list). Tag transformation is possible for improving compliance with HTML standards -- most, but not all, of the obsolete/deprecated elements are converted to valid  ones; see section 3.3.2.

  These 16 htmLawed-supported elements are empty elements that have an opening tag with possible content but no element content (thus, no closing tag): area, br, col, command, embed, hr, img, input, isindex, keygen, link, meta, param, source, track, and wbr.

  As per standards, closing tags are optional for these elements under certain conditions: caption, colgroup, dd, dt, li, optgroup, option, p, rp, rt, tbody, td, tfoot, th, thead, and tr. By default, htmLawed will add a missing closing tag for such elements, unless balancing (section 3.3.3) is turned off.

  With $config["safe"] = 1, the default set of htmLawed-supported elements will exclude applet, audio, canvas, dialog, embed, iframe, object, script and video; see section 3.6.

  When $config["elements"], which specifies allowed elements, is properly defined, and neither empty nor set to 0 or *, the default set is not used. To have elements added to or removed from the default set, a +/- notation is used. E.g., *-script-object implies that only script and object are disallowed, whereas *+noembed means that noembed is also allowed. For an element with a hyphen in name, use round brackets around the name; e.g., (my-custom-element). Elements can also be specified as comma separated names. E.g., a, b, i means only a, b and i are permitted. In this notation, *, + and - have no significance and can actually cause a mis-reading.

  Some more examples of $config["elements"] values indicating permitted elements (note that empty spaces are liberally allowed for clarity):

  *  a, blockquote, code, em, strong -- only a, blockquote, code, em, and strong
  *  *-script -- all excluding script
  *  * -acronym -big -center -dir -font -isindex -s -strike -tt -- only non-obsolete/deprecated elements of HTML5
  *  *+noembed-script -- all including noembed excluding script
  *  *+noembed+(my-custom-element) -- all including noembed and my-custom-element

  Some mis-usages (and the resulting permitted elements) that can be avoided:

  *  -* -- none; instead of htmLawed, one might just use, e.g., the htmlspecialchars() PHP function
  *  *, -script -- all except script; admin probably meant *-script
  *  -*, a, em, strong -- all; admin probably meant a, em, strong
  *  * -- all; admin need not have set elements
  *  *-form+form -- all; a + will always over-ride any -
  *  *, noembed -- only noembed; admin probably meant *+noembed
  *  a, +b, i -- only a and i; admin probably meant a, b, i

  Basically, when using the +/- notation, commas (,) should not be used, and vice versa, and * should be used with the former but not the latter.

  Note: Even if an element that is not in the default set is allowed through $config["elements"], like noembed in the last example, it will eventually be removed during tag balancing unless such balancing is turned off ($config["balance"] set to 0). Currently, the only way around this, which actually is simple, is to edit htmLawed's PHP code which define various arrays in the function hl_balance() to accommodate the element and its nesting properties.

  A possible second way to specify allowed elements is to set $config["parent"] to an element name that supposedly will hold the input, and to set $config["balance"] to 1. During tag balancing (see section 3.3.3), all elements that cannot legally nest inside the parent element will be removed. The parent element is auto-reset to div if $config["parent"] is empty, body, or an element not in htmLawed's default set of 122 elements.

---

HTM version of htmLawed_README.txt generated on 05 Aug, 2023 using rTxt2htm from PHP Labware