1# bluemonday [](https://godoc.org/github.com/microcosm-cc/bluemonday) [](https://sourcegraph.com/github.com/microcosm-cc/bluemonday?badge)
2
3bluemonday is a HTML sanitizer implemented in Go. It is fast and highly configurable.
4
5bluemonday takes untrusted user generated content as an input, and will return HTML that has been sanitised against an allowlist of approved HTML elements and attributes so that you can safely include the content in your web page.
6
7If you accept user generated content, and your server uses Go, you **need** bluemonday.
8
9The default policy for user generated content (`bluemonday.UGCPolicy().Sanitize()`) turns this:
10```html
11Hello <STYLE>.XSS{background-image:url("javascript:alert('XSS')");}</STYLE><A CLASS=XSS></A>World
12```
13
14Into a harmless:
15```html
16Hello World
17```
18
19And it turns this:
20```html
21<a href="javascript:alert('XSS1')" onmouseover="alert('XSS2')">XSS<a>
22```
23
24Into this:
25```html
26XSS
27```
28
29Whilst still allowing this:
30```html
31<a href="http://www.google.com/">
32 <img src="https://ssl.gstatic.com/accounts/ui/logo_2x.png"/>
33</a>
34```
35
36To pass through mostly unaltered (it gained a rel="nofollow" which is a good thing for user generated content):
37```html
38<a href="http://www.google.com/" rel="nofollow">
39 <img src="https://ssl.gstatic.com/accounts/ui/logo_2x.png"/>
40</a>
41```
42
43It protects sites from [XSS](http://en.wikipedia.org/wiki/Cross-site_scripting) attacks. There are many [vectors for an XSS attack](https://www.owasp.org/index.php/XSS_Filter_Evasion_Cheat_Sheet) and the best way to mitigate the risk is to sanitize user input against a known safe list of HTML elements and attributes.
44
45You should **always** run bluemonday **after** any other processing.
46
47If you use [blackfriday](https://github.com/russross/blackfriday) or [Pandoc](http://johnmacfarlane.net/pandoc/) then bluemonday should be run after these steps. This ensures that no insecure HTML is introduced later in your process.
48
49bluemonday is heavily inspired by both the [OWASP Java HTML Sanitizer](https://code.google.com/p/owasp-java-html-sanitizer/) and the [HTML Purifier](http://htmlpurifier.org/).
50
51## Technical Summary
52
53Allowlist based, you need to either build a policy describing the HTML elements and attributes to permit (and the `regexp` patterns of attributes), or use one of the supplied policies representing good defaults.
54
55The policy containing the allowlist is applied using a fast non-validating, forward only, token-based parser implemented in the [Go net/html library](https://godoc.org/golang.org/x/net/html) by the core Go team.
56
57We expect to be supplied with well-formatted HTML (closing elements for every applicable open element, nested correctly) and so we do not focus on repairing badly nested or incomplete HTML. We focus on simply ensuring that whatever elements do exist are described in the policy allowlist and that attributes and links are safe for use on your web page. [GIGO](http://en.wikipedia.org/wiki/Garbage_in,_garbage_out) does apply and if you feed it bad HTML bluemonday is not tasked with figuring out how to make it good again.
58
59## Is it production ready?
60
61*Yes*
62
63We are using bluemonday in production having migrated from the widely used and heavily field tested OWASP Java HTML Sanitizer.
64
65We are passing our extensive test suite (including AntiSamy tests as well as tests for any issues raised). Check for any [unresolved issues](https://github.com/microcosm-cc/bluemonday/issues?page=1&state=open) to see whether anything may be a blocker for you.
66
67We invite pull requests and issues to help us ensure we are offering comprehensive protection against various attacks via user generated content.
68
69## Usage
70
71Install using `go get github.com/microcosm-cc/bluemonday`
72
73Then call it:
74```go
75package main
76
77import (
78 "fmt"
79
80 "github.com/microcosm-cc/bluemonday"
81)
82
83func main() {
84 // Do this once for each unique policy, and use the policy for the life of the program
85 // Policy creation/editing is not safe to use in multiple goroutines
86 p := bluemonday.UGCPolicy()
87
88 // The policy can then be used to sanitize lots of input and it is safe to use the policy in multiple goroutines
89 html := p.Sanitize(
90 `<a onblur="alert(secret)" href="http://www.google.com">Google</a>`,
91 )
92
93 // Output:
94 // <a href="http://www.google.com" rel="nofollow">Google</a>
95 fmt.Println(html)
96}
97```
98
99We offer three ways to call Sanitize:
100```go
101p.Sanitize(string) string
102p.SanitizeBytes([]byte) []byte
103p.SanitizeReader(io.Reader) bytes.Buffer
104```
105
106If you are obsessed about performance, `p.SanitizeReader(r).Bytes()` will return a `[]byte` without performing any unnecessary casting of the inputs or outputs. Though the difference is so negligible you should never need to care.
107
108You can build your own policies:
109```go
110package main
111
112import (
113 "fmt"
114
115 "github.com/microcosm-cc/bluemonday"
116)
117
118func main() {
119 p := bluemonday.NewPolicy()
120
121 // Require URLs to be parseable by net/url.Parse and either:
122 // mailto: http:// or https://
123 p.AllowStandardURLs()
124
125 // We only allow <p> and <a href="">
126 p.AllowAttrs("href").OnElements("a")
127 p.AllowElements("p")
128
129 html := p.Sanitize(
130 `<a onblur="alert(secret)" href="http://www.google.com">Google</a>`,
131 )
132
133 // Output:
134 // <a href="http://www.google.com">Google</a>
135 fmt.Println(html)
136}
137```
138
139We ship two default policies:
140
1411. `bluemonday.StrictPolicy()` which can be thought of as equivalent to stripping all HTML elements and their attributes as it has nothing on its allowlist. An example usage scenario would be blog post titles where HTML tags are not expected at all and if they are then the elements *and* the content of the elements should be stripped. This is a *very* strict policy.
1422. `bluemonday.UGCPolicy()` which allows a broad selection of HTML elements and attributes that are safe for user generated content. Note that this policy does *not* allow iframes, object, embed, styles, script, etc. An example usage scenario would be blog post bodies where a variety of formatting is expected along with the potential for TABLEs and IMGs.
143
144## Policy Building
145
146The essence of building a policy is to determine which HTML elements and attributes are considered safe for your scenario. OWASP provide an [XSS prevention cheat sheet](https://www.owasp.org/index.php/XSS_(Cross_Site_Scripting)_Prevention_Cheat_Sheet) to help explain the risks, but essentially:
147
1481. Avoid anything other than the standard HTML elements
1491. Avoid `script`, `style`, `iframe`, `object`, `embed`, `base` elements that allow code to be executed by the client or third party content to be included that can execute code
1501. Avoid anything other than plain HTML attributes with values matched to a regexp
151
152Basically, you should be able to describe what HTML is fine for your scenario. If you do not have confidence that you can describe your policy please consider using one of the shipped policies such as `bluemonday.UGCPolicy()`.
153
154To create a new policy:
155```go
156p := bluemonday.NewPolicy()
157```
158
159To add elements to a policy either add just the elements:
160```go
161p.AllowElements("b", "strong")
162```
163
164Or using a regex:
165
166_Note: if an element is added by name as shown above, any matching regex will be ignored_
167
168It is also recommended to ensure multiple patterns don't overlap as order of execution is not guaranteed and can result in some rules being missed.
169```go
170p.AllowElementsMatching(regex.MustCompile(`^my-element-`))
171```
172
173Or add elements as a virtue of adding an attribute:
174```go
175// Note the recommended pattern, see the recommendation on using .Matching() below
176p.AllowAttrs("nowrap").OnElements("td", "th")
177```
178
179Again, this also supports a regex pattern match alternative:
180```go
181p.AllowAttrs("nowrap").OnElementsMatching(regex.MustCompile(`^my-element-`))
182```
183
184Attributes can either be added to all elements:
185```go
186p.AllowAttrs("dir").Matching(regexp.MustCompile("(?i)rtl|ltr")).Globally()
187```
188
189Or attributes can be added to specific elements:
190```go
191// Not the recommended pattern, see the recommendation on using .Matching() below
192p.AllowAttrs("value").OnElements("li")
193```
194
195It is **always** recommended that an attribute be made to match a pattern. XSS in HTML attributes is very easy otherwise:
196```go
197// \p{L} matches unicode letters, \p{N} matches unicode numbers
198p.AllowAttrs("title").Matching(regexp.MustCompile(`[\p{L}\p{N}\s\-_',:\[\]!\./\\\(\)&]*`)).Globally()
199```
200
201You can stop at any time and call .Sanitize():
202```go
203// string htmlIn passed in from a HTTP POST
204htmlOut := p.Sanitize(htmlIn)
205```
206
207And you can take any existing policy and extend it:
208```go
209p := bluemonday.UGCPolicy()
210p.AllowElements("fieldset", "select", "option")
211```
212
213### Inline CSS
214
215Although it's possible to handle inline CSS using `AllowAttrs` with a `Matching` rule, writing a single monolithic regular expression to safely process all inline CSS which you wish to allow is not a trivial task. Instead of attempting to do so, you can allow the `style` attribute on whichever element(s) you desire and use style policies to control and sanitize inline styles.
216
217It is strongly recommended that you use `Matching` (with a suitable regular expression)
218`MatchingEnum`, or `MatchingHandler` to ensure each style matches your needs,
219but default handlers are supplied for most widely used styles.
220
221Similar to attributes, you can allow specific CSS properties to be set inline:
222```go
223p.AllowAttrs("style").OnElements("span", "p")
224// Allow the 'color' property with valid RGB(A) hex values only (on any element allowed a 'style' attribute)
225p.AllowStyles("color").Matching(regexp.MustCompile("(?i)^#([0-9a-f]{3,4}|[0-9a-f]{6}|[0-9a-f]{8})$")).Globally()
226```
227
228Additionally, you can allow a CSS property to be set only to an allowed value:
229```go
230p.AllowAttrs("style").OnElements("span", "p")
231// Allow the 'text-decoration' property to be set to 'underline', 'line-through' or 'none'
232// on 'span' elements only
233p.AllowStyles("text-decoration").MatchingEnum("underline", "line-through", "none").OnElements("span")
234```
235
236Or you can specify elements based on a regex pattern match:
237```go
238p.AllowAttrs("style").OnElementsMatching(regex.MustCompile(`^my-element-`))
239// Allow the 'text-decoration' property to be set to 'underline', 'line-through' or 'none'
240// on 'span' elements only
241p.AllowStyles("text-decoration").MatchingEnum("underline", "line-through", "none").OnElementsMatching(regex.MustCompile(`^my-element-`))
242```
243
244If you need more specific checking, you can create a handler that takes in a string and returns a bool to
245validate the values for a given property. The string parameter has been
246converted to lowercase and unicode code points have been converted.
247```go
248myHandler := func(value string) bool{
249 // Validate your input here
250 return true
251}
252p.AllowAttrs("style").OnElements("span", "p")
253// Allow the 'color' property with values validated by the handler (on any element allowed a 'style' attribute)
254p.AllowStyles("color").MatchingHandler(myHandler).Globally()
255```
256
257### Links
258
259Links are difficult beasts to sanitise safely and also one of the biggest attack vectors for malicious content.
260
261It is possible to do this:
262```go
263p.AllowAttrs("href").Matching(regexp.MustCompile(`(?i)mailto|https?`)).OnElements("a")
264```
265
266But that will not protect you as the regular expression is insufficient in this case to have prevented a malformed value doing something unexpected.
267
268We provide some additional global options for safely working with links.
269
270`RequireParseableURLs` will ensure that URLs are parseable by Go's `net/url` package:
271```go
272p.RequireParseableURLs(true)
273```
274
275If you have enabled parseable URLs then the following option will `AllowRelativeURLs`. By default this is disabled (bluemonday is an allowlist tool... you need to explicitly tell us to permit things) and when disabled it will prevent all local and scheme relative URLs (i.e. `href="localpage.html"`, `href="../home.html"` and even `href="//www.google.com"` are relative):
276```go
277p.AllowRelativeURLs(true)
278```
279
280If you have enabled parseable URLs then you can allow the schemes (commonly called protocol when thinking of `http` and `https`) that are permitted. Bear in mind that allowing relative URLs in the above option will allow for a blank scheme:
281```go
282p.AllowURLSchemes("mailto", "http", "https")
283```
284
285Regardless of whether you have enabled parseable URLs, you can force all URLs to have a rel="nofollow" attribute. This will be added if it does not exist, but only when the `href` is valid:
286```go
287// This applies to "a" "area" "link" elements that have a "href" attribute
288p.RequireNoFollowOnLinks(true)
289```
290
291Similarly, you can force all URLs to have "noreferrer" in their rel attribute.
292```go
293// This applies to "a" "area" "link" elements that have a "href" attribute
294p.RequireNoReferrerOnLinks(true)
295```
296
297
298We provide a convenience method that applies all of the above, but you will still need to allow the linkable elements for the URL rules to be applied to:
299```go
300p.AllowStandardURLs()
301p.AllowAttrs("cite").OnElements("blockquote", "q")
302p.AllowAttrs("href").OnElements("a", "area")
303p.AllowAttrs("src").OnElements("img")
304```
305
306An additional complexity regarding links is the data URI as defined in [RFC2397](http://tools.ietf.org/html/rfc2397). The data URI allows for images to be served inline using this format:
307
308```html
309<img src="data:image/webp;base64,UklGRh4AAABXRUJQVlA4TBEAAAAvAAAAAAfQ//73v/+BiOh/AAA=">
310```
311
312We have provided a helper to verify the mimetype followed by base64 content of data URIs links:
313
314```go
315p.AllowDataURIImages()
316```
317
318That helper will enable GIF, JPEG, PNG and WEBP images.
319
320It should be noted that there is a potential [security](http://palizine.plynt.com/issues/2010Oct/bypass-xss-filters/) [risk](https://capec.mitre.org/data/definitions/244.html) with the use of data URI links. You should only enable data URI links if you already trust the content.
321
322We also have some features to help deal with user generated content:
323```go
324p.AddTargetBlankToFullyQualifiedLinks(true)
325```
326
327This will ensure that anchor `<a href="" />` links that are fully qualified (the href destination includes a host name) will get `target="_blank"` added to them.
328
329Additionally any link that has `target="_blank"` after the policy has been applied will also have the `rel` attribute adjusted to add `noopener`. This means a link may start like `<a href="//host/path"/>` and will end up as `<a href="//host/path" rel="noopener" target="_blank">`. It is important to note that the addition of `noopener` is a security feature and not an issue. There is an unfortunate feature to browsers that a browser window opened as a result of `target="_blank"` can still control the opener (your web page) and this protects against that. The background to this can be found here: [https://dev.to/ben/the-targetblank-vulnerability-by-example](https://dev.to/ben/the-targetblank-vulnerability-by-example)
330
331### Policy Building Helpers
332
333We also bundle some helpers to simplify policy building:
334```go
335
336// Permits the "dir", "id", "lang", "title" attributes globally
337p.AllowStandardAttributes()
338
339// Permits the "img" element and its standard attributes
340p.AllowImages()
341
342// Permits ordered and unordered lists, and also definition lists
343p.AllowLists()
344
345// Permits HTML tables and all applicable elements and non-styling attributes
346p.AllowTables()
347```
348
349### Invalid Instructions
350
351The following are invalid:
352```go
353// This does not say where the attributes are allowed, you need to add
354// .Globally() or .OnElements(...)
355// This will be ignored without error.
356p.AllowAttrs("value")
357
358// This does not say where the attributes are allowed, you need to add
359// .Globally() or .OnElements(...)
360// This will be ignored without error.
361p.AllowAttrs(
362 "type",
363).Matching(
364 regexp.MustCompile("(?i)^(circle|disc|square|a|A|i|I|1)$"),
365)
366```
367
368Both examples exhibit the same issue, they declare attributes but do not then specify whether they are allowed globally or only on specific elements (and which elements). Attributes belong to one or more elements, and the policy needs to declare this.
369
370## Limitations
371
372We are not yet including any tools to help allow and sanitize CSS. Which means that unless you wish to do the heavy lifting in a single regular expression (inadvisable), **you should not allow the "style" attribute anywhere**.
373
374In the same theme, both `<script>` and `<style>` are considered harmful. These elements (and their content) will not be rendered by default, and require you to explicitly set `p.AllowUnsafe(true)`. You should be aware that allowing these elements defeats the purpose of using a HTML sanitizer as you would be explicitly allowing either JavaScript (and any plainly written XSS) and CSS (which can modify a DOM to insert JS), and additionally but limitations in this library mean it is not aware of whether HTML is validly structured and that can allow these elements to bypass some of the safety mechanisms built into the [WhatWG HTML parser standard](https://html.spec.whatwg.org/multipage/parsing.html#parsing-main-inselect).
375
376It is not the job of bluemonday to fix your bad HTML, it is merely the job of bluemonday to prevent malicious HTML getting through. If you have mismatched HTML elements, or non-conforming nesting of elements, those will remain. But if you have well-structured HTML bluemonday will not break it.
377
378## TODO
379
380* Investigate whether devs want to blacklist elements and attributes. This would allow devs to take an existing policy (such as the `bluemonday.UGCPolicy()` ) that encapsulates 90% of what they're looking for but does more than they need, and to remove the extra things they do not want to make it 100% what they want
381* Investigate whether devs want a validating HTML mode, in which the HTML elements are not just transformed into a balanced tree (every start tag has a closing tag at the correct depth) but also that elements and character data appear only in their allowed context (i.e. that a `table` element isn't a descendent of a `caption`, that `colgroup`, `thead`, `tbody`, `tfoot` and `tr` are permitted, and that character data is not permitted)
382
383## Long term goals
384
3851. Open the code to adversarial peer review similar to the [Attack Review Ground Rules](https://code.google.com/p/owasp-java-html-sanitizer/wiki/AttackReviewGroundRules)
3861. Raise funds and pay for an external security review