Re: [PATCH next 00/84] ipvs: Stop guessing the network namespace

To: "Eric W. Biederman" <ebiederm@xxxxxxxxxxxx>
Subject: Re: [PATCH next 00/84] ipvs: Stop guessing the network namespace
Cc: Pablo Neira Ayuso <pablo@xxxxxxxxxxxxx>, David Miller <davem@xxxxxxxxxxxxx>, Simon Horman <horms@xxxxxxxxxxxx>, netfilter-devel@xxxxxxxxxxxxxxx, netdev@xxxxxxxxxxxxxxx, Nicolas Dichtel <nicolas.dichtel@xxxxxxxxx>, lvs-devel@xxxxxxxxxxxxxxx
From: Julian Anastasov <ja@xxxxxx>
Date: Sun, 20 Sep 2015 17:38:32 +0300 (EEST)

On Sun, 20 Sep 2015, Eric W. Biederman wrote:

> I am gradually working my way through the netfilter stack passing struct
> down into the netfilter hooks and from the netfilter hooks and from
> there down into the functions that actually care.  This removes the need
> for netfilter functions to guess how to figure out how to compute which
> network namespace they are in and instead provides a simple and reliable
> method to do so.
> The cleanups stand on their own but this is part of a larger effort
> to have routes with an output device that is not in the current network
> namespace.
> The IPVS code has been a bit more of a challenge than most.  Just
> passing struct net through to where it is needed did not feel clean
> to me.  The practical issue is that the ipvs code in most places
> actually wants struct netns_ipvs and not struct net.
> So as part of this process I have turned the relationship between struct
> net and the structs netns_ipvs, ip_vs_conn_param, ip_vs_conn, and
> ip_vs_service inside out.  I have modified the ipvs functions to take a
> struct netns_ipvs not a struct net.  The net is code with fewer
> conversions from one type of structure to another.  I did wind up adding
> a struct netns_ipvs parameter to quite a few functions that did not have
> it before so I could pass the structure down from the netfilter hooks to
> where it is actually needed to avoid guessing.
> I have broken up the work in a bunch of small patches so there is at
> least a chance and reviewing that each step I took is correct.  The
> series compiles at each step so bisecting it should not be a problem
> if something weird comes up.
> The first two changes in this series are actually bug fixes.  The first
> is a compile fix for a bug in sctp that came in, in the last round of
> ipvs changes merged into nf-next.  The second fixes an older bug where
> in pathological circumstances the wrong network namespace could be used
> when a proc file is written to.
> The rest of the patchset is a bunch of boring changes getting pushing
> struct netns_ipvs (and by extension ipvs->net) where it needs to be.
> Either by replacing struct net pointers or adding new struct netns_ipvs
> pointers.  With a handful of other minor cleanups (like removing skb_net).

        I reviewed the patchset. Nice work, thanks!
Here are some comments:

01/84 ipvs: Hoist computation of ipvs earlier in sctp_conn_schedule

        Simon had a fix for this problem, not sure what happened,
may be it was lost...

04/84 ipvs: Store ipvs not net in struct ip_vs_conn

        Lost '!' here:

@@ -359,7 +359,7 @@ struct ip_vs_conn *ip_vs_ct_in_get(const struct
ip_vs_conn_param *p)
        hlist_for_each_entry_rcu(cp, &ip_vs_conn_tab[hash], c_list) {
                if (unlikely(p->pe_data && p->pe->ct_match)) {
-                       if (!ip_vs_conn_net_eq(cp, p->net))
+                       if (net_eq(cp->ipvs->net, p->net))

        Problem is then propagated to patch 05/84:

-                       if (net_eq(cp->ipvs->net, p->net))
+                       if (cp->ipvs == p->ipvs)

26/84 ipvs: Pass ipvs not net to __ip_vs_get_servie_entries

        Missing 'c' in Subject

55/84 ipvs: Pass ipvs not net to register_ip_vs_app and 

        Empty line after ipvs declaration:

 void __net_exit ip_vs_app_net_cleanup(struct net *net)
-       unregister_ip_vs_app(net, NULL /* all */);
+       struct netns_ipvs *ipvs = net_ipvs(net);
+       unregister_ip_vs_app(ipvs, NULL /* all */);

        here too:

 static void __ip_vs_ftp_exit(struct net *net)
-       unregister_ip_vs_app(net, &ip_vs_ftp);
+       struct netns_ipvs *ipvs = net_ipvs(net);
+       if (!ipvs)

61/84 ipvs: Pass ipvs into .conn_in_get and ip_vs_conn_in_get_proto


62/84 ipvs: Pass ipvs into conn_out_get


        Also, scripts/ --strict /tmp/*.patch gives
me warnings, sometimes for inherited syntax...


Julian Anastasov <ja@xxxxxx>
To unsubscribe from this list: send the line "unsubscribe lvs-devel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at

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