svn commit: r280971 - in head: contrib/ipfilter/tools share/man/man4 sys/contrib/ipfilter/netinet sys/netinet sys/netipsec sys/netpfil/pf

[Gleb Smirnoff]

On Thu, Apr 02, 2015 at 03:37:51PM +0200, Mateusz Guzik wrote:
M> On Thu, Apr 02, 2015 at 03:35:22PM +0300, Gleb Smirnoff wrote:
M> > On Wed, Apr 01, 2015 at 05:07:56PM -0600, Ian Lepore wrote:
M> > I> > Author: glebius
M> > I> > Date: Wed Apr  1 22:26:39 2015
M> > I> > New Revision: 280971
M> > I> > URL: https://svnweb.freebsd.org/changeset/base/280971
M> > I> > 
M> > I> > Log:
M> > I> >   o Use new function ip_fillid() in all places throughout the kernel,
M> > I> >     where we want to create a new IP datagram.
M> > I> >   o Add support for RFC6864, which allows to set IP ID for atomic IP
M> > I> >     datagrams to any value, to improve performance. The behaviour is
M> > I> >     controlled by net.inet.ip.rfc6864 sysctl knob, which is enabled by
M> > I> >     default.
M> > I> >   o In case if we generate IP ID, use counter(9) to improve performance.
M> > I> >   o Gather all code related to IP ID into ip_id.c.
M> > I> >   
M> > I> >   Differential Revision:		https://reviews.freebsd.org/D2177
M> > I> >   Reviewed by:			adrian, cy, rpaulo
M> > I> >   Tested by:			Emeric POUPON <emeric.poupon stormshield.eu>
M> > I> >   Sponsored by:			Netflix
M> > I> >   Sponsored by:			Nginx, Inc.
M> > I> >   Relnotes:			yes
M> > I> > 
M> > I> [...]
M> > I> > +void
M> > I> > +ip_fillid(struct ip *ip)
M> > I> > +{
M> > I> > +
M> > I> > +	/*
M> > I> > +	 * Per RFC6864 Section 4
M> > I> > +	 *
M> > I> > +	 * o  Atomic datagrams: (DF==1) && (MF==0) && (frag_offset==0)
M> > I> > +	 * o  Non-atomic datagrams: (DF==0) || (MF==1) || (frag_offset>0)
M> > I> > +	 */
M> > I> > +	if (V_ip_rfc6864 && (ip->ip_off & htons(IP_DF)) == htons(IP_DF))
M> > I> > +		ip->ip_id = 0;
M> > I> > +	else if (V_ip_do_randomid)
M> > I> > +		ip->ip_id = ip_randomid();
M> > I> > +	else {
M> > I> > +		counter_u64_add(V_ip_id, 1);
M> > I> > +		ip->ip_id = htons((*(uint64_t *)zpcpu_get(V_ip_id)) & 0xffff);
M> > I> > +	}
M> > I> > +}
M> > I> > +
M> > I> 
M> > I> This is completely bogus.  It's a big opacity violation (it relies on
M> > I> what should be opaque private internal implementation details of
M> > I> counter(9)).  The fact that the counter api doesn't provide a function
M> > I> for retrieving one cpu's counter value should be a big clue there -- the
M> > I> fact that you know the internals doesn't make it okay to reach behind
M> > I> the counter and grab a value like that.  It may not even be safe to do
M> > I> so on any given architecture; it certainly isn't safe on arm, and that
M> > I> line of code above will work only by accident because you're throwing
M> > I> way all but 16 bits.
M> > 
M> > I though about providing that API, but since it isn't safe in general,
M> > I decided to not do that.
M> > 
M> > I> But even more importantly, this WILL result in multiple threads using
M> > I> the same value at the same time...
M> > I>  
M> > I>  - Thread A on CPU 1 and thread B on CPU 2 both begin executing here at
M> > I> the same time, and both get through counter_u64_add().
M> > I>  - Thread A keeps running and uses CPU 1's new value, call it 27.
M> > I>  - Thread B gets prempted between counter_u64_add() and zpcpu_get().
M> > I> When it resumes it's now on CPU 1, so it retrieves value 27 as well.
M> > 
M> > This was already discussed in this thread:
M> > 
M> > https://lists.freebsd.org/pipermail/svn-src-head/2015-March/069864.html
M> > 
M> 
M> For this particular use-case you never care what CPU you are executing
M> on, you only want to obtain a unique number.
M> 
M> per-cpu counters can serve this purpose no problem, just provide an
M> operation which guarantees to return the new value of the counter it
M> incremented. Should be easily achieved with e.g. just pinning curthread
M> to the cpu it executes on for the duration of inc + fetch.

I'd ask to pay attention to this particular email:

https://lists.freebsd.org/pipermail/svn-src-head/2015-March/069966.html

Just to justify probabilities, risks and countermeasures.

For those, who don't believe in theory and prefers practice:

https://lists.freebsd.org/pipermail/svn-src-head/2015-March/070091.html

Note that Emeric was the one who observed collisions for the ip_id++
code, that we used before.

-- 
Totus tuus, Glebius.