LVS
lvs-devel
Google
 
Web LinuxVirtualServer.org

Re: [PATCH V3 0/8] ipvs: IPv6 fragment handling for IPVS

To: Julian Anastasov <ja@xxxxxx>
Subject: Re: [PATCH V3 0/8] ipvs: IPv6 fragment handling for IPVS
Cc: Hans Schillstrom <hans@xxxxxxxxxxxxxxx>, Hans Schillstrom <hans.schillstrom@xxxxxxxxxxxx>, netdev@xxxxxxxxxxxxxxx, Patrick McHardy <kaber@xxxxxxxxx>, Pablo Neira Ayuso <pablo@xxxxxxxxxxxxx>, lvs-devel@xxxxxxxxxxxxxxx, Thomas Graf <tgraf@xxxxxxx>, Wensong Zhang <wensong@xxxxxxxxxxxx>, netfilter-devel@xxxxxxxxxxxxxxx, Simon Horman <horms@xxxxxxxxxxxx>
From: Jesper Dangaard Brouer <brouer@xxxxxxxxxx>
Date: Tue, 25 Sep 2012 15:11:14 +0200
On Thu, 2012-09-13 at 01:57 +0300, Julian Anastasov wrote:
>       Hello,
> 
> On Tue, 11 Sep 2012, Jesper Dangaard Brouer wrote:
> 
> > The following patchset implement IPv6 fragment handling for IPVS.
> > 
> > This work is based upon patches from Hans Schillstrom.  I have taken
> > over the patchset, in close agreement with Hans, because he don't have
> > (gotten allocated) time to complete his work.
> > 
> > I have cleaned up the patchset significantly, and split the patchset
> > up into eight patches.
> > 
> > The first 4 patches, are ready to be merged
> > 
> >  Patch01: Trivial changes, use compressed IPv6 address in output
> >  Patch02: IPv6 extend ICMPv6 handling for future types
> >  Patch03: Use config macro IS_ENABLED()
> >  Patch04: Fix bug in IPVS IPv6 NAT mangling of ports inside ICMPv6 packets
> > 
> > The next 4 patches, I consider V3 of the patches I have submitted
> > earlier, where I have incorporated all of Julian's feedback.  I have
> > also tried to make the patches easier to review, by reorganizing the
> > changes, to be more strictly split (exthdr vs. fragment handling).
> > 
> > I have also removed the API changes, and moved those to patch07.  This
> > is done, (1) to make it easier to review the patches, and (2) to allow
> > easier integration of Patricks idea and my RFC patch of caching exthdr
> > info in skb->cb[].  Thus, we can get these patches applied (and later
> > go back and apply the caching scheme easier).
> > 
> >  Patch05: Fix faulty IPv6 extension header handling in IPVS
> >  Patch06: Complete IPv6 fragment handling for IPVS
> >  Patch07: IPVS API change to avoid rescan of IPv6 exthdr
> >  Patch08: IPVS SIP fragment handling
> > 
> > The SIP frag handling have been split into its own patch, as I have
> > not been able to test this part my self.
> > 
> > This patchset is based upon:
> >   Pablo's nf-next tree:  git://1984.lsi.us.es/nf-next
> >   On top of commit 0edd94887d19ad73539477395c17ea0d6898947a
> > 
> > ---
> > 
> > Jesper Dangaard Brouer (8):
> >       ipvs: SIP fragment handling
> >       ipvs: API change to avoid rescan of IPv6 exthdr
> >       ipvs: Complete IPv6 fragment handling for IPVS
> >       ipvs: Fix faulty IPv6 extension header handling in IPVS
> >       ipvs: Fix bug in IPv6 NAT mangling of ports inside ICMPv6 packets
> >       ipvs: Use config macro IS_ENABLED()
> >       ipvs: IPv6 extend ICMPv6 handling for future types
> >       ipvs: Trivial changes, use compressed IPv6 address in output
> 
>       Some comments:
> 
> - About patch 4: ip_vs_icmp_xmit_v6 already calls skb_make_writable
> before ip_vs_nat_icmp_v6, that is why we provide 'offset'.

I see, that call path is correct, BUT I was talking about another call
path of ip_vs_nat_icmp_v6(), via handle_response_icmp() (which also
calls skb_make_writable).  That call path is triggered, if the
real-server, have shutdown its service and send back an ICMPv6 packet.

Hmm, testing it again, I cannot trigger this issue.  Perhaps I was
confusing my self and were using my test script that added IPv6 exthdrs
to the packet.  Adding print statements to the code, also show the
correct offset now.
I'm dropping this patch-4, and I'll adjust/fix patch-5 ("ipvs: fix
faulty IPv6 extension header handling in IPVS") accordingly.  And I'll
double check patch-5, that exthdr have been accounted for (in the offset
used by skb_make_writable() before calling ip_vs_nat_icmp_v6()).



> - May be we can provide the offset of ICMPv6 header
> from ip_vs_in_icmp_v6 to ip_vs_icmp_xmit_v6 as additional
> argument (icmp_offset) and then to ip_vs_nat_icmp_v6. By this
> way we can avoid the two ipv6_find_hdr calls if we also provide
> the iph argument from ip_vs_icmp_xmit_v6 to ip_vs_nat_icmp_v6,
> its ->len points to the ports. ip_vs_in_icmp_v6 provides
> also protocol in this ciph, so may be we have everything.

Is this comment for the API patch-7 ("ipvs: API change to avoid rescan
of IPv6 exthdr") ?
The API patch is going to save 19 calls to ipv6_find_hdr ().


> - in ip_vs_in_icmp_v6 there must be 'offs_ciph = ciph.len;'
> just before this line:
> 
> if (IPPROTO_TCP == ciph.protocol || IPPROTO_UDP == ciph.protocol ||
> 

It would be a lot easier for me, if you commented directly on the
patches.

I can see that 'offs_ciph = ciph.len;' is set earlier in this patch, but
that value is primarily used by IP_VS_DBG_PKT.  And offs_ciph, needs to
be updated, again, with the value of ciph.len after the call to
ipv6_find_hdr().  So, yes you are right ;-)

I'll rename offs_ciph to "writable" to emphasize what we are using this
value for.

>       The idea is that we linearize for writing the inner
> IP header and optionally the 2 ports. That is why old
> logic was 'offset += 2 * sizeof(__u16);'

The port logic was kept.  But I'll make it more clear whats happening,
and keep the "+=" coding style.


> - initially, ip_vs_fill_iph_skb fills iphdr->flags from
> current fragment, later ip_vs_out_icmp_v6 uses the same
> ipvsh when calling ipv6_find_hdr. Should we initialize
> ipvsh->flags to 0 before calling ipv6_find_hdr because
> it is I/O argument?

As we don't use the flag, after this point, we can just give
ipv6_find_hdr() a NULL value instead.

But I must give you, that it's a little confusing the way we reuse the
ipvsh variable (in ip_vs_out_icmp_v6()).  Think, this needs to be
rewritten to use a separate variable, like in ip_vs_in_icmp_v6().


> - in patch 5: in ip_vs_nat_icmp_v6 skb_make_writable can
> move data to other addresses on linearization. Any pointers
> like 'ciph' should be recalculated based on offsets. But
> it does not matter because we should not call skb_make_writable
> here.

Yes, as mentioned earlier, I'll fix up patch-5 and remove the
skb_make_writable call.


>       I also see that we should not send ICMP
> errors (FRAG NEEDED/TOO BIG) in response to large
> ICMP error packets but it is not related to your changes,
> it needs separate change to all transmitters.

Yes, its unrelated, lets fix that in another patchset.

I'll hopefully soon have a patchset ready with these changes/updates...

-- 
Best regards,
  Jesper Dangaard Brouer
  MSc.CS, Sr. Network Kernel Developer at Red Hat
  Author of http://www.iptv-analyzer.org
  LinkedIn: http://www.linkedin.com/in/brouer


--
To unsubscribe from this list: send the line "unsubscribe lvs-devel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html

<Prev in Thread] Current Thread [Next in Thread>