> On May 2, 2017, 12:34 a.m., Benjamin Mahler wrote:
> > 3rdparty/libprocess/src/process.cpp
> > Lines 480-492 (patched)
> > <https://reviews.apache.org/r/58224/diff/2/?file=1693806#file1693806line480>
> >
> > I'm wondering if we can eliminate the need for this via global flag access and
peer caching in the Socket implementation, see other comments.
Fixed the global flag, filed MESOS-7452 for the peer address caching.
> On May 2, 2017, 12:34 a.m., Benjamin Mahler wrote:
> > 3rdparty/libprocess/src/process.cpp
> > Line 942 (original), 965 (patched)
> > <https://reviews.apache.org/r/58224/diff/2/?file=1693806#file1693806line979>
> >
> > Per our offline discussion, could we acheive the elimination of `getpeername`-per-read
by having the `Socket` perform this optimization for a connected socket? That would also avoid
the need to carry the peer around.
I filed [MESOS-7452](https://issues.apache.org/jira/browse/MESOS-7452) to implement the `Socket`
peer address caching.
- James
-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/58224/#review173536
-----------------------------------------------------------
On May 3, 2017, 10:35 p.m., James Peach wrote:
>
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/58224/
> -----------------------------------------------------------
>
> (Updated May 3, 2017, 10:35 p.m.)
>
>
> Review request for mesos and Benjamin Mahler.
>
>
> Bugs: MESOS-7401
> https://issues.apache.org/jira/browse/MESOS-7401
>
>
> Repository: mesos
>
>
> Description
> -------
>
> In general, libprocess is unable to validate that a peer
> is a legitimate owner of the UPID it claims in a libprocess
> message. This change adds a check that the IP address in the
> UPID matches the peer address. This makes spoofing the UPID
> harder (eg. to send authenticated messages), but also breaks
> some legitimate configurations, particularly on multihomed
> hosts.
>
>
> Diffs
> -----
>
> 3rdparty/libprocess/src/process.cpp f5b666f894215cb1861c244c94b382e0739bc5c9
>
>
> Diff: https://reviews.apache.org/r/58224/diff/5/
>
>
> Testing
> -------
>
> make check (Fedora 25). Light manual testing.
>
> With LIBPROCESS_pin_peer_address=true, all Mesos tests pass except ``ExamplesTest.DiskFullFramework``,
however enabling this will definitely break some libprocess APIs (though not in the way that
Mesos uses them) and legitimate multi-homed configurations. Note that setting LIBPROCESS_ip=127.0.0.1
makes you multihomed for this purpose, which is why ``ExamplesTest.DiskFullFramework`` breaks.
>
>
> Thanks,
>
> James Peach
>
>
|