The Web Origin Concept
University of California, Berkeley
abarth@eecs.berkeley.eduhttp://www.adambarth.com/
Stanford University
collinj@cs.stanford.eduhttp://www.collinjackson.com/
Google, Inc.
ian@hixie.chhttp://ln.hixie.ch/Working GroupInternet-DraftThis document defines the concept of an "origin," which is used by
web browsers to isolate content retrieved from different parties. The
origin concept is defined by a "same-origin" relation and a serialization
algorithm. This document also defines an HTTP Origin header, which a
user agent can use to describe the security contexts that caused the user
agent to initiate an HTTP request. HTTP servers can use the Origin
header to mitigate against Cross-Site Request Forgery (CSRF)
vulnerabilities.This document defines the concept of an "origin," which is used by
web browsers to isolate content retrieved from different parties. The
origin concept is defined by a "same-origin" relation and a serialization
algorithm. This document also defines an HTTP Origin header, which a
user agent can use to describe the security contexts that caused the user
agent to initiate an HTTP request. HTTP servers can use the Origin
header to mitigate against Cross-Site Request Forgery (CSRF)
vulnerabilities.TODO: Discuss other CSRF defenses.The following algorithm MUST be used to compute the origin of a URI.
Let /uri/ be the URI for which the origin is being
determined.Parse /uri/.If /uri/ does not use a server-based naming authority, or if
parsing /uri/ failed, or if /uri/ is not an absolute URI, then
return an implementation-defined value.Let /scheme/ be the scheme component of /uri/, converted to
lowercase.If the implementation doesn't support the protocol given by
/scheme/, then return an implementation-defined value.If /scheme/ is "file", then the implementation MAY return a
implementation-defined value.Let /host/ be the host component of /uri/.Apply the IDNA ToASCII algorithm [RFC3490] to /host/, with both the
AllowUnassigned and UseSTD3ASCIIRules flags set. Let /host/ be the
result of the ToASCII algorithm.If ToASCII fails to convert one of the components of the string
(e.g. because it is too long or because it contains invalid
characters), then return an implementation-defined value.Let /host/ be the result of converting /host/ to lowercase.If there is no port component of /uri/, then let /port/ be the
default port for the protocol given by /scheme/. Otherwise, let
/port/ be the port component of /uri/.Return the tuple (/scheme/, /host/, /port/).Implementations MAY define other types of origins in addition to
the scheme/host/port tuple type defined above. (For example, user
agents could implement globally unique origins or certificate-based
origins.)Implementations MUST use the following algorithm to test whether
two origins are the "same origin".
Let /A/ be the first origin being compared, and let /B/ be the
second origin being compared.If either /A/ or /B/ is not a scheme/host/port tuple, return an
implementation-defined value.If /A/ and /B/ have scheme components that are not identical,
return false.If /A/ and /B/ have host components that are not identical,
return false.If /A/ and /B/ have port components that are not identical,
return false.Return true.Implementations MUST using the following algorithm to compute the
Unicode serialization of an origin:
If the origin in question is not a scheme/host/port tuple, then
return the string
null
(i.e., the code point sequence U+006E, U+0075, U+006C, U+006C) and
abort these steps.Otherwise, let /result/ be the scheme part of the origin
tuple.Append the string "://" to /result/.Apply the IDNA ToUnicode algorithm [RFC3490] to each component of
the host part of the origin tuple, and append the results of each
component, in the same order, separated by U+002E FULL STOP code
points (".") to /result/.If the port part of the origin tuple gives a port that is
different from the default port for the protocol given by the
scheme part of the origin tuple, then append a U+003A COLON
code points (":") and the given port, in base ten, to /result/.Return /result/.TODO: Check that we handle IPv6 literals correctly.Implementations MUST using the following algorithm to compute the
ASCII serialization of an origin:
If the origin in question is not a scheme/host/port tuple, then
return the string
null
(i.e., the code point sequence U+006E, U+0075, U+006C, U+006C) and
abort these steps.Otherwise, let /result/ be the scheme part of the origin tuple.Append the string "://" to /result/.If the host part of the origin tuple is in the form of a DNS
domain name, apply the IDNA conversion algorithm ([RFC3490] section
4) to it, with both the AllowUnassigned and UseSTD3ASCIIRules flags
set, employ the ToASCII operation, and append the result to
/result/.If ToASCII fails to convert one of the components of the
string, e.g. because it is too long or because it contains invalid
characters, then return the literal string "null" and abort these
steps.If the port part of the origin tuple gives a port that is
different from the default port for the protocol given by the
scheme part of the origin tuple, then append a U+003A COLON
code point (":") and the given port, in base ten, to /result/.Return /result/.Whenever a user agent issues an HTTP request, the user agent MUST
include exactly one HTTP header named "Origin" that conforms to the
following ABNF [RFC5234] grammar:
Whenever a user agent would send a Origin header containing two
consecutive, identical origin serializations, the user agent MUST remove
one such origin serialization from the header.Whenever a user agent issues an HTTP request from a
"privacy-sensitive" context, the user agent MUST send the value "null"
in the Origin header.If /B/ is the Request-URI in the Request-Line of an HTTP request, then
the associated HTTP response is an "HTTP redirect from URI /B/" if the
response contains a 3xx Status Code (all terms RFC2616).Whenever a user agent issues an HTTP request to URI /A/ as a result of
an HTTP redirect from URI /B/, the user agent MUST either:
set the value of the Origin header in the HTTP request to /A/
to "null" (i.e., the code point sequence U+006E, U+0075, U+006C,
U+006C),set the value of the Origin header in the /A/ request to the value
of the Origin header in the /B/ request extended with a space and the
ASCII serialization of the origin of /B/, unless this would result in
the header containing the origin serialization "null" in a
component.Whenever a user agent issues an HTTP request that (1) is *not* the
result of an HTTP redirect and (2) is *not* initiated from a
"privacy-sensitive" context, the user agent SHOULD set the value of the
Origin header to the ASCII serialization of the origin that
initiated the HTTP request.Note: This behavior differs from that of the HTTP Referer header,
which user agents often suppress when an origin with an "https" scheme
issues a request for a URI with an "http" scheme.HTTP Servers MAY use the Origin header to "defend themselves
against CSRF attacks." Such servers are known as "participating
servers" in this section.Let the /origin white list/ of a participating server be a set of
strings selected by the operator of that server.The string "null" MUST NOT be a member of the /origin white list/
for any participating server.
Example: The origin white list for the example.com Web server
could be the strings "http://example.com", "https://example.com",
"http://www.example.com", and "https://www.example.com".A participating server MUST use the following algorithm when
determining whether to modify state in response to an HTTP request:
If the request method is safe (as defined by RFC 2616,
Section 9.1.1, e.g. either "GET" nor "HEAD"), return "MUST NOT
modify state" and abort these steps.If the request does not contain a header named "Origin", return
"MAY modify state" abort these steps.For each request header named "Origin", let the /initiating
origin list/ be the list of origins represented in the header:
If there exists a origin in the /initiating origin list/ is not
a member of the /origin white list/ for this server, return "MUST
NOT modify state" and abort these steps.Return "MAY modify state".Example: A Web server could modify state in response to POST
requests that lack an Origin header (because these requests are sent
by non-supporting user agents) and could modify state in response to
POST requests that have an Origin header of "http://example.com",
"https://example.com", "http://www.example.com", or
"https://www.example.com".This section is not normative.The Origin header improves on the Referer header by respecting the
user's privacy: The Origin header includes only the information
required to identify the principal that initiated the request
(typically the scheme, host, and port of initiating origin). In
particular, the Origin header does not contain the path or query
portions of the URI included in the Referer header that invade privacy
without providing additional security.The Origin header also improves on the Referer header by not
leaking intranet host names to external web sites when a user follows
a hyperlink from an intranet host to an external site because
hyperlinks generate privacy-sensitive requests.This section is not normative.Because a supporting user agent will always include the Origin
header when making HTTP requests, HTTP servers can detect that a
request was initiated by a supporting user agent by observing the
presence of the header. This design prevents a malicious web site from
making a supporting user agent appear to be a non-supporting user agent.
Unlike the Referer header, which is absent when suppressed by the user
agent, the Origin header takes on the value "null" when suppressed by
the user agent.In some legacy user agents, The Origin header can be spoofed for
same-site XMLHttpRequests. Sites that rely only on network
connectivity for authentication should use a DNS rebinding defense,
such as validating the HTTP Host header, in addition to CSRF
protection.TODO: The "Origin" header should be registered.Think about how this interacts with proxies.Think about how this interacts with caches.Think about how this interacts with IPv6.