An easy safelist-based HTML-sanitizing tool.
.. image:: https://github.com/mozilla/bleach/workflows/Test/badge.svg :target: https://github.com/mozilla/bleach/actions?query=workflow%3ATest
.. image:: https://github.com/mozilla/bleach/workflows/Lint/badge.svg :target: https://github.com/mozilla/bleach/actions?query=workflow%3ALint
.. image:: https://badge.fury.io/py/bleach.svg :target: http://badge.fury.io/py/bleach
NOTE: 2023-01-23: Bleach is deprecated. See issue:
<https://github.com/mozilla/bleach/issues/698>
__
Bleach is an allowed-list-based HTML sanitizing library that escapes or strips markup and attributes.
Bleach can also linkify text safely, applying filters that Django's urlize
filter cannot, and optionally setting rel
attributes, even on links already
in the text.
Bleach is intended for sanitizing text from untrusted sources. If you find yourself jumping through hoops to allow your site administrators to do lots of things, you're probably outside the use cases. Either trust those users, or don't.
Because it relies on html5lib_, Bleach is as good as modern browsers at dealing with weird, quirky HTML fragments. And any of Bleach's methods will fix unbalanced or mis-nested tags.
The version on GitHub_ is the most up-to-date and contains the latest bug
fixes. You can find full documentation on ReadTheDocs
_.
:Code: https://github.com/mozilla/bleach :Documentation: https://bleach.readthedocs.io/ :Issue tracker: https://github.com/mozilla/bleach/issues :License: Apache License v2; see LICENSE file
For regular bugs, please report them in our issue tracker <https://github.com/mozilla/bleach/issues>
_.
If you believe that you've found a security vulnerability, please file a secure bug report in our bug tracker <https://bugzilla.mozilla.org/enter_bug.cgi?assigned_to=nobody%40mozilla.org&product=Webtools&component=Bleach-security&groups=webtools-security>
_
or send an email to security AT mozilla DOT org.
For more information on security-related bug disclosure and the PGP key to use
for sending encrypted mail or to verify responses received from that address,
please read our wiki page at
<https://www.mozilla.org/en-US/security/#For_Developers>
_.
Bleach is a security-focused library.
We have a responsible security vulnerability reporting process. Please use that if you're reporting a security issue.
Security issues are fixed in private. After we land such a fix, we'll do a release.
For every release, we mark security issues we've fixed in the CHANGES
in
the Security issues section. We include any relevant CVE links.
Bleach is available on PyPI_, so you can install it with pip
::
$ pip install bleach
.. warning::
Before doing any upgrades, read through Bleach Changes <https://bleach.readthedocs.io/en/latest/changes.html>
_ for backwards
incompatible changes, newer versions, etc.
Bleach follows semver 2
_ versioning. Vendored libraries will not
be changed in patch releases.
The simplest way to use Bleach is:
.. code-block:: python
>>> import bleach
>>> bleach.clean('an <script>evil()</script> example')
u'an <script>evil()</script> example'
>>> bleach.linkify('an http://example.com url')
u'an <a href="http://example.com" rel="nofollow">http://example.com</a> url'
This project and repository is governed by Mozilla's code of conduct and
etiquette guidelines. For more details please see the CODE_OF_CONDUCT.md </CODE_OF_CONDUCT.md>
_
.. _html5lib: https://github.com/html5lib/html5lib-python .. _GitHub: https://github.com/mozilla/bleach .. _ReadTheDocs: https://bleach.readthedocs.io/ .. _PyPI: https://pypi.org/project/bleach/ .. _semver 2: https://semver.org/
Backwards incompatible changes
Security fixes
None
Bug fixes
Backwards incompatible changes
bleach.clean
, bleach.sanitizer.Cleaner
,
bleach.html5lib_shim.BleachHTMLParser
: the tags
and protocols
arguments were changed from lists to sets.
Old pre-6.0.0:
.. code-block:: python
bleach.clean( "some text", tags=["a", "p", "img"], # ^ ^ list protocols=["http", "https"], # ^ ^ list )
New 6.0.0 and later:
.. code-block:: python
bleach.clean( "some text", tags={"a", "p", "img"}, # ^ ^ set protocols={"http", "https"}, # ^ ^ set )
bleach.linkify
, bleach.linkifier.Linker
: the skip_tags
and
recognized_tags
arguments were changed from lists to sets.
Old pre-6.0.0:
.. code-block:: python
bleach.linkify( "some text", skip_tags=["pre"], # ^ ^ list )
linker = Linker( skip_tags=["pre"], # ^ ^ list recognized_tags=html5lib_shim.HTML_TAGS + ["custom-element"], # ^ ^ ^ list # | # | list concatenation )
New 6.0.0 and later:
.. code-block:: python
bleach.linkify( "some text", skip_tags={"pre"}, # ^ ^ set )
linker = Linker( skip_tags={"pre"}, # ^ ^ set recognized_tags=html5lib_shim.HTML_TAGS | {"custom-element"}, # ^ ^ ^ set # | # | union operator )
bleach.sanitizer.BleachSanitizerFilter
: strip_allowed_elements
is now
strip_allowed_tags
. We now use "tags" everywhere rather than a mishmash
of "tags" in some places and "elements" in others.
Security fixes
None
Bug fixes
Add support for Python 3.11. (#675)
Fix API weirness in BleachSanitizerFilter
. (#649)
We're using "tags" instead of "elements" everywhere--no more weird overloading of "elements" anymore.
Also, it no longer calls the superclass constructor.
Add warning when css_sanitizer
isn't set, but the style
attribute is allowed. (#676)
Fix linkify handling of character entities. (#501)
Rework dev dependencies to use requirements-dev.txt
and
requirements-flake8.txt
instead of extras.
Fix project infrastructure to be tox-based so it's easier to have CI run the same things we're running in development and with flake8 in an isolated environment.
Update action versions in CI.
Switch to f-strings where possible. Make tests parametrized to be easier to read/maintain.
Security fixes
None
Bug fixes
Add missing comma to tinycss2 require. Thank you, @shadchin!
Add url parse tests based on wpt url tests. (#688)
Support scheme-less urls if "https" is in allow list. (#662)
Handle escaping <
in edge cases where it doesn't start a tag. (#544)
Fix reference warnings in docs. (#660)
Correctly urlencode email address parts. Thank you, @larseggert! (#659)
Backwards incompatible changes
clean
and linkify
now preserve the order of HTML attributes. Thank
you, @askoretskly! (#566)
Drop support for Python 3.6. Thank you, @hugovk! (#629)
CSS sanitization in style tags is completely different now. If you're using
Bleach clean
to sanitize css in style tags, you'll need to update your
code and you'll need to install the css
extras::
pip install 'bleach[css]'
See the documentation on sanitizing CSS for how to do it <https://bleach.readthedocs.io/en/latest/clean.html#sanitizing-css>
_. (#633)
Security fixes
None
Bug fixes
Rework dev dependencies. We no longer have
requirements-dev.in
/requirements-dev.txt
. Instead, we're using
dev
extras.
See development docs <https://bleach.readthedocs.io/en/latest/dev.html>
_
for more details. (#620)
Add newline when dropping block-level tags. Thank you, @jvanasco! (#369)
Features
Security fixes
None
Bug fixes
Backwards incompatible changes
Security fixes
None
Features
Security fixes
None
Features
Bug fixes
Backwards incompatible changes
Security fixes
Features
None
Bug fixes
None
Security fixes
None
Features
None
Bug fixes
Security fixes
None
Features
Bug fixes
Security fixes
None
Features
None
Bug fixes
Security fixes
None
Features
None
Bug fixes
html5lib
dependency to version 1.1.0. Thank you Sam Sneddon.Security fixes
None
Features
None
Bug fixes
setuptools
dependency with packaging
. Thank you Benjamin Peterson.Security fixes
bleach.clean
behavior parsing style attributes could result in a
regular expression denial of service (ReDoS).
Calls to bleach.clean
with an allowed tag with an allowed
style
attribute were vulnerable to ReDoS. For example,
bleach.clean(..., attributes={'a': ['style']})
.
This issue was confirmed in Bleach versions v3.1.3, v3.1.2, v3.1.1, v3.1.0, v3.0.0, v2.1.4, and v2.1.3. Earlier versions used a similar regular expression and should be considered vulnerable too.
Anyone using Bleach <=v3.1.3 is encouraged to upgrade.
https://bugzilla.mozilla.org/show_bug.cgi?id=1623633
Backwards incompatible changes
Features
None
Bug fixes
None
Security fixes
None
Backwards incompatible changes
Drop support for Python 3.4. Thank you, @hugovk!
Drop deprecated setup.py test
support. Thank you, @jdufresne! (#507)
Features
Add support for Python 3.8. Thank you, @jdufresne!
Add support for PyPy 7. Thank you, @hugovk!
Add pypy3 testing to tox and travis. Thank you, @jdufresne!
Bug fixes
Add relative link to code of conduct. (#442)
Fix typo: curren -> current in tests/test_clean.py Thank you, timgates42! (#504)
Fix handling of non-ascii style attributes. Thank you, @sekineh! (#426)
Simplify tox configuration. Thank you, @jdufresne!
Make documentation reproducible. Thank you, @lamby!
Fix typos in code comments. Thank you, @zborboa-g!
Fix exception value testing. Thank you, @mastizada!
Fix parser-tags NoneType exception. Thank you, @bope!
Improve TLD support in linkify. Thank you, @pc-coholic!
Security fixes
bleach.clean
behavior parsing embedded MathML and SVG content
with RCDATA tags did not match browser behavior and could result in
a mutation XSS.
Calls to bleach.clean
with strip=False
and math
or
svg
tags and one or more of the RCDATA tags script
,
noscript
, style
, noframes
, iframe
, noembed
, or
xmp
in the allowed tags whitelist were vulnerable to a mutation
XSS.
This security issue was confirmed in Bleach version v3.1.1. Earlier versions are likely affected too.
Anyone using Bleach <=v3.1.1 is encouraged to upgrade.
https://bugzilla.mozilla.org/show_bug.cgi?id=1621692
Backwards incompatible changes
None
Features
None
Bug fixes
None
Security fixes
bleach.clean
behavior parsing noscript
tags did not match
browser behavior.
Calls to bleach.clean
allowing noscript
and one or more of
the raw text tags (title
, textarea
, script
, style
,
noembed
, noframes
, iframe
, and xmp
) were vulnerable
to a mutation XSS.
This security issue was confirmed in Bleach versions v2.1.4, v3.0.2, and v3.1.0. Earlier versions are probably affected too.
Anyone using Bleach <=v3.1.0 is highly encouraged to upgrade.
https://bugzilla.mozilla.org/show_bug.cgi?id=1615315
Backwards incompatible changes
None
Features
None
Bug fixes
None
Security fixes
None
Backwards incompatible changes
None
Features
recognized_tags
argument to the linkify Linker
class. This
fixes issues when linkifying on its own and having some tags get escaped.
It defaults to a list of HTML5 tags. Thank you, Chad Birch! (#409)Bug fixes
Add six>=1.9
to requirements. Thank you, Dave Shawley (#416)
Fix cases where attribute names could have invalid characters in them. (#419)
Fix problems with LinkifyFilter
not being able to match links
across &
. (#422)
Fix InputStreamWithMemory
when the BleachHTMLParser
is
parsing meta
tags. (#431)
Fix doctests. (#357)
Security fixes
None
Backwards incompatible changes
None
Features
None
Bug fixes
Characters
tokens after sanitizing them. This fixes issues in the
LinkifyFilter
where it was only linkifying parts of urls. (#374)Security fixes
None
Backwards incompatible changes
None
Features
Bug fixes
list
object has no attribute lower
in clean
. (#398)abbr
getting escaped in linkify
. (#400)Security fixes
None
Backwards incompatible changes
A bunch of functions were moved from one module to another.
These were moved from bleach.sanitizer
to bleach.html5lib_shim
:
convert_entity
convert_entities
match_entity
next_possible_entity
BleachHTMLSerializer
BleachHTMLTokenizer
BleachHTMLParser
These functions and classes weren't documented and aren't part of the public API, but people read code and might be using them so we're considering it an incompatible API change.
If you're using them, you'll need to update your code.
Features
Bleach no longer depends on html5lib. html5lib==1.0.1 is now vendored into Bleach. You can remove it from your requirements file if none of your other requirements require html5lib.
This means Bleach will now work fine with other libraries that depend on html5lib regardless of what version of html5lib they require. (#386)
Bug fixes
Fixed tags getting added when using clean or linkify. This was a long-standing regression from the Bleach 2.0 rewrite. (#280, #392)
Fixed <isindex>
getting replaced with a string. Now it gets escaped or
stripped depending on whether it's in the allowed tags or not. (#279)
Security fixes
None
Backwards incompatible changes
Features
None
Bug fixes
Security fixes
Attributes that have URI values weren't properly sanitized if the values contained character entities. Using character entities, it was possible to construct a URI value with a scheme that was not allowed that would slide through unsanitized.
This security issue was introduced in Bleach 2.1. Anyone using Bleach 2.1 is highly encouraged to upgrade.
https://bugzilla.mozilla.org/show_bug.cgi?id=1442745
Backwards incompatible changes
None
Features
None
Bug fixes
Security fixes
None
Backwards incompatible changes
None
Features
None
Bug fixes
Support html5lib-python 1.0.1. (#337)
Add deprecation warning for supporting html5lib-python < 1.0.
Switch to semver.
Security fixes
None
Backwards incompatible changes
None
Features
None
Bug fixes
setup.py
opening files when LANG=
. (#324)Security fixes
Convert control characters (backspace particularly) to "?" preventing malicious copy-and-paste situations. (#298)
See <https://github.com/mozilla/bleach/issues/298>
_ for more details.
This affects all previous versions of Bleach. Check the comments on that issue for ways to alleviate the issue if you can't upgrade to Bleach 2.1.
Backwards incompatible changes
Redid versioning. bleach.VERSION
is no longer available. Use the string
version at bleach.__version__
and parse it with
pkg_resources.parse_version
. (#307)
clean, linkify: linkify and clean should only accept text types; thank you, Janusz! (#292)
clean, linkify: accept only unicode or utf-8-encoded str (#176)
Features
Bug fixes
bleach.clean()
no longer unescapes entities including ones that are missing
a ;
at the end which can happen in urls and other places. (#143)
linkify: fix http links inside of mailto links; thank you, sedrubal! (#300)
clarify security policy in docs (#303)
fix dependency specification for html5lib 1.0b8, 1.0b9, and 1.0b10; thank you, Zoltán! (#268)
add Bleach vs. html5lib comparison to README; thank you, Stu Cox! (#278)
fix KeyError exceptions on tags without href attr; thank you, Alex Defsen! (#273)
add test website and scripts to test bleach.clean()
output in browser;
thank you, Greg Guthe!
Security fixes
Backwards incompatible changes
Removed support for Python 2.6. (#206)
Removed support for Python 3.2. (#224)
Bleach no longer supports html5lib < 0.99999999 (8 9s).
This version is a rewrite to use the new sanitizing API since the old one was dropped in html5lib 0.99999999 (8 9s).
If you're using 0.9999999 (7 9s) upgrade to 0.99999999 (8 9s) or higher.
If you're using 1.0b8 (equivalent to 0.9999999 (7 9s)), upgrade to 1.0b9 (equivalent to 0.99999999 (8 9s)) or higher.
bleach.clean
and friends were rewritten
clean
was reimplemented as an html5lib filter and happens at a different
step in the HTML parsing -> traversing -> serializing process. Because of
that, there are some differences in clean's output as compared with previous
versions.
Amongst other things, this version will add end tags even if the tag in question is to be escaped.
bleach.clean
and friends attribute callables now take three arguments:
tag, attribute name and attribute value. Previously they only took attribute
name and attribute value.
All attribute callables will need to be updated.
bleach.linkify
was rewritten
linkify
was reimplemented as an html5lib Filter. As such, it no longer
accepts a tokenizer
argument.
The callback functions for adjusting link attributes now takes a namespaced attribute.
Previously you'd do something like this::
def check_protocol(attrs, is_new):
if not attrs.get('href', '').startswith('http:', 'https:')):
return None
return attrs
Now it's more like this::
def check_protocol(attrs, is_new):
if not attrs.get((None, u'href'), u'').startswith(('http:', 'https:')):
# ^^^^^^^^^^^^^^^
return None
return attrs
Further, you need to make sure you're always using unicode values. If you don't then html5lib will raise an assertion error that the value is not unicode.
All linkify filters will need to be updated.
bleach.linkify
and friends had a skip_pre
argument--that's been
replaced with a more general skip_tags
argument.
Before, you might do::
bleach.linkify(some_text, skip_pre=True)
The equivalent with Bleach 2.0 is::
bleach.linkify(some_text, skip_tags=['pre'])
You can skip other tags, too, like style
or script
or other places
where you don't want linkification happening.
All uses of linkify that use skip_pre
will need to be updated.
Changes
Supports Python 3.6.
Supports html5lib >= 0.99999999 (8 9s).
There's a bleach.sanitizer.Cleaner
class that you can instantiate with your
favorite clean settings for easy reuse.
There's a bleach.linkifier.Linker
class that you can instantiate with your
favorite linkify settings for easy reuse.
There's a bleach.linkifier.LinkifyFilter
which is an htm5lib filter that
you can pass as a filter to bleach.sanitizer.Cleaner
allowing you to clean
and linkify in one pass.
bleach.clean
and friends can now take a callable as an attributes arg value.
Tons of bug fixes.
Cleaned up tests.
Documentation fixes.
Security fixes
Backwards incompatible changes
clean: The list of ALLOWED_PROTOCOLS
now defaults to http, https and
mailto.
Previously it was a long list of protocols something like ed2k, ftp, http, https, irc, mailto, news, gopher, nntp, telnet, webcal, xmpp, callto, feed, urn, aim, rsync, tag, ssh, sftp, rtsp, afs, data. (#149)
Changes
clean: Added protocols
to arguments list to let you override the list of
allowed protocols. Thank you, Andreas Malecki! (#149)
linkify: Fix a bug involving periods at the end of an email address. Thank you, Lorenz Schori! (#219)
linkify: Fix linkification of non-ascii ports. Thank you Alexandre, Macabies! (#207)
linkify: Fix linkify inappropriately removing node tails when dropping nodes. (#132)
Fixed a test that failed periodically. (#161)
Switched from nose to py.test. (#204)
Add test matrix for all supported Python and html5lib versions. (#230)
Limit to html5lib >=0.999,!=0.9999,!=0.99999,<0.99999999
because 0.9999
and 0.99999 are busted.
Add support for python setup.py test
. (#97)
Security fixes
Changes
>=0.999,<0.99999999
because of impending change to
sanitizer api. #195Changes
linkify: Fix hang in linkify with parse_email=True
. (#124)
linkify: Fix crash in linkify when removing a link that is a first-child. (#136)
Updated TLDs.
linkify: Don't remove exterior brackets when linkifying. (#146)
Changes
Consistent order of attributes in output.
Python 3.4 support.
Changes
linkify: Update linkify to use etree type Treewalker instead of simpletree.
Updated html5lib to version >=0.999
.
Update all code to be compatible with Python 3 and 2 using six.
Switch to Apache License.
clean()
no longer considers feed:
an acceptable protocol due to
inconsistencies in browser behavior.linkify()
has changed considerably. Many keyword arguments have been
replaced with a single callbacks list. Please see the documentation for more
information.
Bleach will no longer consider unacceptable protocols when linkifying.
linkify()
now takes a tokenizer argument that allows it to skip
sanitization.
delinkify()
is gone.
Removed exception handling from _render
. clean()
and linkify()
may
now throw.
linkify()
correctly ignores case for protocols and domain names.
linkify()
correctly handles markup within an tag.
Fix hang in style attribute sanitizer. (#61)
Allow /
in style attribute values.
linkify()
now understands port numbers. (#38)
Documented character encoding behavior. (#41)
Add an optional target argument to linkify()
.
Add delinkify()
method. (#45)
Support subdomain whitelist for delinkify()
. (#47, #48)
Switch to SemVer git tags.
Make linkify()
smarter about trailing punctuation. (#30)
Pass exc_info
to logger during rendering issues.
Add wildcard key for attributes. (#19)
Make linkify()
use the HTMLSanitizer
tokenizer. (#36)
Fix URLs wrapped in parentheses. (#23)
Make linkify()
UTF-8 safe. (#33)
linkify()
works with 3rd level domains. (#24)
clean()
supports vendor prefixes in style values. (#31, #32)
Fix linkify()
email escaping.
linkify()
supports email addresses.
clean()
supports callables in attributes filter.
linkify()
doesn't drop trailing slashes. (#21)linkify()
won't linkify 'libgl.so.1'. (#22)