-
Notifications
You must be signed in to change notification settings - Fork 337
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Define Cross-Origin-Resource-Policy response header #733
Conversation
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Thanks! I have a few quick thoughts, only one real issue.
fetch.bs
Outdated
<p>Its <a for=header>value</a> <a>ABNF</a>: | ||
|
||
<pre> | ||
Cross-Origin-Resource-Policy = %x73.61.6D.65 / %x73.61.6D.65.2D.73.69.74.65 ; "same" / "same-site"; case-sensitive</pre> |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Nit: The last semicolon should be a comma (similar to the origin-or-null
definition).
fetch.bs
Outdated
<var>request</var>'s <a for=request>current url</a>'s <a for=url>origin</a>, then return | ||
<b>allowed</b>. | ||
|
||
<p class=note>A cross-origin response redirecting to a same or same-site resource with the |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Nit: This is a little unclear. Is the redirect target same-site with the request's origin, or with the cross-origin response?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
The former was the intent, but yeah, that needs to be clarified, if we indeed want this behavior.
fetch.bs
Outdated
|
||
<li><p><var>request</var>'s <a for=request>current url</a>'s <a for=url>host</a> | ||
<a>is a registrable domain suffix of or is equal to</a> <var>request</var>'s | ||
<a for=request>origin</a>'s <a for=origin>host</a> |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I think this isn't broad enough. The bidirectional is a registrable domain suffix of or is equal to
comparison would return false for a request whose origin's host is subdomain.subdomain.example.com
and whose current url is notsubdomain.example.com
.
Instead, I think we need to calculate the relevant registrable domain (perhaps by pointing to the (not terribly clear) algorithm on https://publicsuffix.org/list/) for both the request's origin and the request's current URL, and compare those.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
(And, actually, looking at that algorithm again, I'm not sure it does what it needs to do for document.domain
either... It doesn't calculate the registrable domain at all, but the public suffix. I'll file a separate bug.)
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Time to revisit whatwg/url#72? I suspect having "registrable domain" and "same-site" as a primitive available will be quite useful going forward.
We also need to think to what extent we want to do scheme/port-comparisons as well.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Yeah. Ok, assuming I can dig myself out from under everything else, would you prefer those concepts be defined in HTML, Fetch, or URL?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Note: this is now tracked by whatwg/url#391.
fetch.bs
Outdated
|
||
<li><p>Let <var>policy</var> be the <a>combined value</a> with | ||
`<a http-header><code>Cross-Origin-Resource-Policy</code></a>` and <var>response</var>'s | ||
<a for=response>header list</a>. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
It might be worth adding a note here about the meaning of same, same
or same, same-site
, as I believe this means that both will be ignored.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Yeah, that's the intent. Perhaps we should split on ,
though given that it seems likely we'll add support for multiple explicit origins later.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I don't think it matters in this patch, but I agree that that's probably what we'd want to do in a future patch (e.g. define this as a structured header list whose members are strings that we parse as origins or something.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Hmm, wouldn't that be incompatible with same
and same-site
as tokens? And origins cannot be used as tokens, at least not unless httpwg/http-extensions#629 is fixed somehow.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I don't think there's any requirement that all the items in a list are the same type. @mnot can confirm, but my understanding is that we could have same
and same-site
as "labels" and some other type ("string", I suppose) for origins.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
PR sounds good to me, in particular:
- When happens the CORP check. It is handling all responses even intermediate ones
- CORP does apply to redirections but not to the final response-after-redirection if it is again same origin
The one thing that might need further discussion is whether to apply CORP on preflight responses.
It seems this PR implies that CORP should be checked. WebKit does not enforce it and it makes sense to me to keep it that way.
This could for instance ease deployment strategies such as "stick that CORP: same-site header on any response"
FWIW, the WebKit patch aligning with this PR should land shortly.
I will upstream the tests we have so far to WPT.
They should be further completed by header value parsing, preflight and challenge response dedicated tests.
fetch.bs
Outdated
@@ -3736,6 +3794,9 @@ Range Requests</cite>. [[HTTP-RANGE]] However, this is not widely supported by b | |||
</ol> | |||
</ol> | |||
|
|||
<li><p>If the <a>cross-origin resource policy check</a> with <var>request</var> and |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I pushed a commit that should address the review feedback. It builds upon whatwg/url#391 (nearly done) and #743 (needs review). @mnot can you confirm that structured headers will allow mixing of identifiers and strings? |
FYI: I rebased and renamed |
Dependencies have been merged, plan of action is at #687 (comment). TL;DR: this will merge next week barring any objections. |
This header makes it easier for sites to block unwanted "no-cors" cross-origin requests. Tests: * web-platform-tests/wpt#11171 * web-platform-tests/wpt#11427 * web-platform-tests/wpt#11428 Follow-up: #760. Fixes #687.
<var>request</var>'s <a for=request>current url</a>'s <a for=url>host</a> | ||
<li><var>request</var>'s <a for=request>origin</a>'s <a for=url>scheme</a> is | ||
"<code>https</code>" or <var>response</var>'s <a for=response>HTTPS state</a> is | ||
"<code>none</code>" |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
@mikewest is this too cute and should I be checking the scheme of request's current url instead or is this fine/preferred?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I feel like I had to think about what this means more than I'd like to. An explicit comparison seems simpler. Or maybe just skipping the check around? Is something like this equivalent?
If response's HTTPS state is
modern
, request's origin's scheme ishttps
.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
That doesn't seem equivalent as it doesn't clearly evaluate to a boolean.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
LGTM. This patch is a pretty reasonable definition of the feature, thanks for putting it together, @annevk!
That said, I think there's enough subtlety here that it would be worth writing up a separate document that walked through the details, expectations, and rationales behind the header's design. The underlying proposal went through a few mutations, and it feels like we're doing it a disservice by only recording the algorithm, and none of the ancillary justification. If I pointed a developer to this section of Fetch, I suspect they'd have no idea what problem it attempts to solve, or why it's relevant to them at all.
Anyway, LGTM to land the definition as something browser vendors can rely on while implementing, but I hope we can collectively find time to fill in some of the details at greater length.
<var>request</var>'s <a for=request>current url</a>'s <a for=url>host</a> | ||
<li><var>request</var>'s <a for=request>origin</a>'s <a for=url>scheme</a> is | ||
"<code>https</code>" or <var>response</var>'s <a for=response>HTTPS state</a> is | ||
"<code>none</code>" |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I feel like I had to think about what this means more than I'd like to. An explicit comparison seems simpler. Or maybe just skipping the check around? Is something like this equivalent?
If response's HTTPS state is
modern
, request's origin's scheme ishttps
.
fetch.bs
Outdated
<p class="note no-backref">A cross-origin response redirecting to a response that is | ||
<a>same origin</a> or <a>same site</a> with the initial request and has a | ||
`<a http-header><code>Cross-Origin-Resource-Policy</code></a>` header specified, does not affect | ||
anything. I.e., <var>request</var>'s <a for=request>tainted origin flag</a> is not checked. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Nit: I'd rephrase this a little bit. Perhaps something like "The cross-origin resource policy check does not consider a request's redirect chain when processing a given response's Cross-Origin-Resource-Policy
header, but only the relationship between the requesting origin and the response which asserts the header."?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Rewrote, but didn't use "redirect chain" as that's not really defined.
I filed #767 on adding advice. |
…stonly Automatic update from web-platform-testsFetch: Cross-Origin-Resource-Policy tests For whatwg/fetch#733. WebKit export of https://bugs.webkit.org/show_bug.cgi?id=185840. -- wpt-commits: 53f7340307c1c0fa4ab96e79d88c69a7870030f4 wpt-pr: 11171
…in-Resource-Policy, a=testonly Automatic update from web-platform-testsFetch: basic syntax tests for Cross-Origin-Resource-Policy Supplements #11171. For whatwg/fetch#733. -- wpt-commits: b7373b42eeac24ff6cb3ed494ffbf09e781287da wpt-pr: 11427
…: same-site's scheme restriction, a=testonly Automatic update from web-platform-testsFetch: test Cross-Origin-Resource-Policy: same-site's scheme restriction Supplements #11171. For whatwg/fetch#733. -- wpt-commits: 7f0a106f3d5e9d3e7f70ba52aae896a3fffc2cc6 wpt-pr: 11428
…stonly Automatic update from web-platform-testsFetch: Cross-Origin-Resource-Policy tests For whatwg/fetch#733. WebKit export of https://bugs.webkit.org/show_bug.cgi?id=185840. -- wpt-commits: 53f7340307c1c0fa4ab96e79d88c69a7870030f4 wpt-pr: 11171 UltraBlame original commit: 2d3d375085ab145c32c6a3de5589739a3a5ceaf0
…in-Resource-Policy, a=testonly Automatic update from web-platform-testsFetch: basic syntax tests for Cross-Origin-Resource-Policy Supplements #11171. For whatwg/fetch#733. -- wpt-commits: b7373b42eeac24ff6cb3ed494ffbf09e781287da wpt-pr: 11427 UltraBlame original commit: 0d5c1298c115f6ccc0b1e19e7494b1dd72ae7f18
…: same-site's scheme restriction, a=testonly Automatic update from web-platform-testsFetch: test Cross-Origin-Resource-Policy: same-site's scheme restriction Supplements #11171. For whatwg/fetch#733. -- wpt-commits: 7f0a106f3d5e9d3e7f70ba52aae896a3fffc2cc6 wpt-pr: 11428 UltraBlame original commit: 2088a3b878a40a0f734fa8e04ade1417b951e1d4
…stonly Automatic update from web-platform-testsFetch: Cross-Origin-Resource-Policy tests For whatwg/fetch#733. WebKit export of https://bugs.webkit.org/show_bug.cgi?id=185840. -- wpt-commits: 53f7340307c1c0fa4ab96e79d88c69a7870030f4 wpt-pr: 11171 UltraBlame original commit: 2d3d375085ab145c32c6a3de5589739a3a5ceaf0
…in-Resource-Policy, a=testonly Automatic update from web-platform-testsFetch: basic syntax tests for Cross-Origin-Resource-Policy Supplements #11171. For whatwg/fetch#733. -- wpt-commits: b7373b42eeac24ff6cb3ed494ffbf09e781287da wpt-pr: 11427 UltraBlame original commit: 0d5c1298c115f6ccc0b1e19e7494b1dd72ae7f18
…: same-site's scheme restriction, a=testonly Automatic update from web-platform-testsFetch: test Cross-Origin-Resource-Policy: same-site's scheme restriction Supplements #11171. For whatwg/fetch#733. -- wpt-commits: 7f0a106f3d5e9d3e7f70ba52aae896a3fffc2cc6 wpt-pr: 11428 UltraBlame original commit: 2088a3b878a40a0f734fa8e04ade1417b951e1d4
…stonly Automatic update from web-platform-testsFetch: Cross-Origin-Resource-Policy tests For whatwg/fetch#733. WebKit export of https://bugs.webkit.org/show_bug.cgi?id=185840. -- wpt-commits: 53f7340307c1c0fa4ab96e79d88c69a7870030f4 wpt-pr: 11171 UltraBlame original commit: 2d3d375085ab145c32c6a3de5589739a3a5ceaf0
…in-Resource-Policy, a=testonly Automatic update from web-platform-testsFetch: basic syntax tests for Cross-Origin-Resource-Policy Supplements #11171. For whatwg/fetch#733. -- wpt-commits: b7373b42eeac24ff6cb3ed494ffbf09e781287da wpt-pr: 11427 UltraBlame original commit: 0d5c1298c115f6ccc0b1e19e7494b1dd72ae7f18
…: same-site's scheme restriction, a=testonly Automatic update from web-platform-testsFetch: test Cross-Origin-Resource-Policy: same-site's scheme restriction Supplements #11171. For whatwg/fetch#733. -- wpt-commits: 7f0a106f3d5e9d3e7f70ba52aae896a3fffc2cc6 wpt-pr: 11428 UltraBlame original commit: 2088a3b878a40a0f734fa8e04ade1417b951e1d4
This header makes it easier for sites to block unwanted "no-cors" cross-origin requests.
Tests: ...
Fixes #687.
Preview | Diff