- From: Mike O'Neill <michael.oneill@baycloud.com>
- Date: Tue, 5 Jan 2016 11:29:36 -0000
- To: "'Nottingham, Mark'" <mnotting@akamai.com>, "'Chris Palmer'" <palmer@google.com>
- Cc: "'Jonathan Kingston'" <jonathan@jooped.co.uk>, "'Brad Hill'" <hillbrad@gmail.com>, "'Joel Weinberger'" <jww@chromium.org>, "'Richard Barnes'" <rbarnes@mozilla.com>, "'Patrick Toomey'" <patrick.toomey@github.com>, "'WebAppSec WG'" <public-webappsec@w3.org>
I don’t know if this has already been talked about, but maybe long headers like CSP can be could be put in a well-known resource. It would cost another roundtrip but save bandwidth in the end because the resource would be cached. The CSP header would only need to contain a hash of the resource to confirm -----Original Message----- From: Nottingham, Mark [mailto:mnotting@akamai.com] Sent: 05 January 2016 01:59 To: Chris Palmer <palmer@google.com> Cc: Jonathan Kingston <jonathan@jooped.co.uk>; Brad Hill <hillbrad@gmail.com>; Joel Weinberger <jww@chromium.org>; Richard Barnes <rbarnes@mozilla.com>; Patrick Toomey <patrick.toomey@github.com>; WebAppSec WG <public-webappsec@w3.org> Subject: Re: [CSP] "sri" source expression to enforce SRI Catching up after holidays -- I've been wanting to talk about this. In HTTP/2, the default of SETTINGS_HEADER_TABLE_SIZE is 4k. >From what I've seen, Chrome and Firefox both stick with the default. While 4k of header compression context can help performance considerably, it's important to understand that HPACK's compression scheme is coarse-grained, so when the encoder is faced with a large header, it has to choose between putting it into the dynamic table -- thereby denying use of that space to other headers -- or repeatedly putting it out onto the wire. For example, Twitter's response headers already get close to this limit, mostly thanks to CSP: https://redbot.org/?id=w5yLyD Their server has to choose between putting that ~3K CSP header into the dynamic table, leaving them only about 1k to play with for other headers per connection, or leave it out, and send it verbatim on EVERY response. They'll get small benefit from static Huffman coding (which reduces the numbers above a bit), but that's it. If a single header value exceeds SETTINGS_HEADER_TABLE_SIZE, it can't be encoded by reference, and the sender has no choice but to emit it on every message. Things get even nastier if there are several large versions of CSP on a single connection. Clients could start advertising a larger SETTINGS_HEADER_TABLE_SIZE, but that means a larger state commitment (both client-side and server-side, where it can hurt a lot more, offers more DoS exposure, etc.). Given that we're already seeing popular sites brush up against this, PLEASE don't assume that HTTP/2 == free compression, and that we can continue to merrily add headers. Also - when a header is both large and monolithic like CSP (i.e., it doesn't allow multiple values to be combined into a comma-separated value), it makes it much harder to optimise for compression, because of HPACK's granularity (again). I realise that there are security motivations behind this for CSP, but I wonder if the cost is justified (because once somebody can append headers, there's a lot of other damage they can do). Cheers, > On 23 Dec 2015, at 1:38 pm, Chris Palmer <palmer@google.com> wrote: > > HTTP/2 should do a lot to address header bloat, just as it addresses other performance problems. > > And, as usual, import content_layer_heaviest from stdarg. :) > > On Tue, Dec 22, 2015 at 6:13 PM, Jonathan Kingston <jonathan@jooped.co.uk> wrote: > Perhaps the bloat is something that actually needs to be addressed? Creating many headers doesn't really solve the bloat issue. > I agree that it doesn't need to be the core CSP spec especially as we have UI Security separate etc. > But yes when we discussed this last certainly one directive isn't flexible enough for example when SRI expands to images having all assets on the page requiring SRI would probably be too inflexible. > > On Tue, Dec 22, 2015 at 7:40 PM Brad Hill <hillbrad@gmail.com> wrote: > I'm open to either possibility. In the past we've talked about things like fallback policy (e.g. if CDN content from untrusted host X fails the hash check, try to load from a trusted canonical https source, host Y) that would be tricky to shoehorn into the CSP directive parsing logic, and policy combination is another area where it is good not to overcomplicate CSP. > > On Tue, Dec 22, 2015 at 11:29 AM Joel Weinberger <jww@chromium.org> wrote: > FWIW, I think either approach is fine. I know that, in general, we've been concerned about CSP bloat, so for that reason alone it might be worth moving it to its own header. But I don't really care at all either way. > > On Tue, Dec 22, 2015 at 2:28 PM Richard Barnes <rbarnes@mozilla.com> wrote: > I'm not sure I agree with that, Brad :) CSP is where we place restrictions on loading things, and "must have SRI" is a restriction on loading things. > > On Tue, Dec 22, 2015 at 2:26 PM, Brad Hill <hillbrad@gmail.com> wrote: > Yeah, we'd discussed a SRI policy header / meta tag to express a number of things like this, it just got dropped from v1 to get it out the door. Not sure shoehorning it into CSP is the right choice, especially since the reporting mechanism is already being factored out into its own, reusable, feature. Might be simpler to define a standalone header. > > On Tue, Dec 22, 2015 at 11:24 AM Richard Barnes <rbarnes@mozilla.com> wrote: > Some sort of "must-sri" directive is something we had considered inside Mozilla for some of our properties, so this does seem like a productive thing to look at. I don't have any personal biases about how exactly to express it. > > > On Tue, Dec 22, 2015 at 12:07 PM, Patrick Toomey <patrick.toomey@github.com> wrote: > Yeah, a separate directive probably makes sense. I was originally thinking it fit into the "locations that are safe" pattern since we are stating that a location is only safe if it has a known hash (using SRI) from that location. But, I realize that is a stretch. And, you have a good point about being able to put other SRI related things in if we have a separate directive. So, yeah, that is probably the cleaner way to go. Thanks for opening the tracking issue. > On Tue, Dec 22, 2015 at 9:32 AM Joel Weinberger <jww@chromium.org> wrote: > That's a good point about SRI in general; it's hard to know if you've forgotten to SRI anything. I'm not sure source-expression is the right place to put it in CSP, though, as that's meant to be "locations that are safe," and that's not exactly what you're requesting. It probably makes sense to have an 'sri-options' directive, though, since we'll probably want SRI 'report-only' eventually anyway. > > I've filed this as a feature request in GitHub, too: https://github.com/w3c/webappsec-subresource-integrity/issues/23 > --Joel > > > On Tue, Dec 22, 2015 at 2:50 AM Patrick Toomey <patrick.toomey@github.com> wrote: > We recently deployed subresource integrity across GitHub.com: https://github.com/blog/2058-github-implements-subresource-integrity. However, a few days after deployment we determined that one of our JS scripts did not have an "integrity" attribute assigned to it. It was our intent to add the integrity attribute to all subresources on GitHub.com. We statically vendor in all CSS/JS and use Sprockets (SRI support was added in https://github.com/sstephenson/sprockets/pull/645) to package these assets for production deployments. There happened to be one JS file that had not been vendored, and hence was not being packaged by Sprockets. This violated two of our goals: > > * Not allowing any dynamically sourced JS (we vendor everything to ensure what is in version control is what is used in production) > * Enforcing SRI on all supported subresources on GitHub.com > > Reflecting back on this situation, it would have been nice to have support in CSP for a source expression such as "sri"/"sri-only"/"sri-naming-things-is-hard" to ensure SRI is being used everywhere. In the above scenario, the related JS would have failed to load and we would have identified both of the issues listed above in testing. > > > -- Mark Nottingham mnot@akamai.com https://www.mnot.net/
Received on Tuesday, 5 January 2016 11:30:29 UTC