Hi Oleg,

Sounds good, thanks! I'll put up PRs on GitHub with tests, starting with
the Path trailing slash one since it's the smallest, then the empty Domain
and Expires ones after

Best,
Dickson

On Tue, Nov 17, 2026 02:38 PM, Oleg Kalnichevski <[email protected]> wrote:

> On Tue, 2026-11-17 at 12:05 -0500, Dickson Wu via dev wrote:
> > Hi all,
> >
> > I'm Dickson, a security researcher at SEAL & CertiK. While running
> > the
> > public http-state cookie tests against HttpClient I hit three spots
> > where
> > the RFC 6265 cookie spec behaves differently from the RFC. None of
> > them has
> > security impact as far as I can tell - they're interop issues. I
> > don't have
> > a JIRA account yet, so I'm sending them here first. Happy to open
> > JIRA
> > tickets instead if you'd rather track them there.
> >
> > All three reproduce on 5.7-beta1 from Maven Central and on master
> > b202e12,
> > using RFC6265CookieSpecFactory in STRICT mode.
> >
> > 1. Expires with a mismatched weekday is rejected. RFC 6265 5.1.1 only
> > uses
> > the time, day, month and year tokens, so a valid date whose weekday
> > doesn't
> > match should still be accepted. HttpClient's date parse fails and the
> > whole
> > cookie is dropped (http-state cases 0002, 0003, comma0006,
> > comma0007). For
> > example "a=b; Expires=Mon, 01 Jan 2030 00:00:00 GMT" (that date is a
> > Tuesday).
> > https://www.google.com/url?q=https://github.com/apache/httpc
> omponents-client/blob/b202e125f2144bf4bc136d91895abf18fecf5f
> ba/httpclient5/src/main/java/org/apache/hc/client5/http/
> impl/cookie/BasicExpiresHandler.java%23L84-L89&source=gmail&
> ust=1795021508838000&sa=E
> >
> > 2. A Path ending in a slash matches the path without it. Per 5.1.4,
> > cookie
> > path "/foo/" shouldn't match a request for "/foo", but pathMatch()
> > strips
> > the trailing slash first, so the cookie gets sent (case path0014).
> > Section
> > 8.5 already says Path isn't a security boundary, so this is just
> > conformance.
> > https://www.google.com/url?q=https://github.com/apache/httpc
> omponents-client/blob/b202e125f2144bf4bc136d91895abf18fecf5f
> ba/httpclient5/src/main/java/org/apache/hc/client5/http/
> impl/cookie/BasicPathHandler.java%23L70-L95&source=gmail&
> ust=1795021508838000&sa=E
> >
> > 3. An empty Domain attribute after a valid one rejects the cookie.
> > Per
> > 5.2.3 the empty "Domain=" should be ignored and the earlier Domain=
> > https://www.google.com/url?q=http://example.com&source=gmail
> &ust=1795021508838000&sa=E
> > kept. The attribute map keeps only the last value, and
> > BasicDomainHandler
> > then rejects it as blank (case optional-domain0041).
> > https://www.google.com/url?q=https://github.com/apache/httpc
> omponents-client/blob/b202e125f2144bf4bc136d91895abf18fecf5f
> ba/httpclient5/src/main/java/org/apache/hc/client5/http/
> impl/cookie/RFC6265CookieSpec.java%23L172-L187&source=gmail&
> ust=1795021508838000&sa=E
> > https://www.google.com/url?q=https://github.com/apache/httpc
> omponents-client/blob/b202e125f2144bf4bc136d91895abf18fecf5f
> ba/httpclient5/src/main/java/org/apache/hc/client5/http/
> impl/cookie/BasicDomainHandler.java%23L63-L69&source=gmail&
> ust=1795021508838000&sa=E
> >
> > I left out quoted values since HTTPCLIENT-661 and HTTPCLIENT-1006
> > already
> > cover them.
> >
> > I don't have a patch for these yet. The Expires one probably needs a
> > looser
> > date parser, and the Domain one needs the attributes processed in
> > order, so
> > I wanted to check whether you'd want any of them changed before
> > writing one.
> >
> > Best,
> > Dickson
>
> Hi Dickson
>
> Some of us will likely have to re-read RFC 6265 and verify your
> findings. However you are really welcome to propose fixes for these
> defects as pull requests at GitHub in order to speed up the process. Do
> feel free to start with the simplest one.
>
> Cheers
>
> Oleg
>
>

Reply via email to