Date:   Mon, 29 Aug 2022 17:56:06 +0800
From:   Hangbin Liu <liuhangbin@...il.com>
To:     Jay Vosburgh <jay.vosburgh@...onical.com>
Cc:     netdev@...r.kernel.org, Veaceslav Falico <vfalico@...il.com>,
        Andy Gospodarek <andy@...yhouse.net>,
        "David S . Miller" <davem@...emloft.net>,
        Jakub Kicinski <kuba@...nel.org>,
        Jonathan Toppins <jtoppins@...hat.com>,
        Paolo Abeni <pabeni@...hat.com>,
        David Ahern <dsahern@...il.com>, LiLiang <liali@...hat.com>
Subject: Re: [PATCH net] bonding: fix lladdr finding and confirmation

On Sun, Aug 28, 2022 at 04:20:43PM -0700, Jay Vosburgh wrote:
> Hangbin Liu <liuhangbin@...il.com> wrote:
> 
> >There are 3 issues when setting lladdr as bonding IPv6 target
> >
> >1. On the send side. When ns_ip6_target was set, the ipv6_dev_get_saddr()
> >   will be called to get available src addr and send IPv6 neighbor solicit
> >   message.
> >
> >   If the target is global address, ipv6_dev_get_saddr() will get any
> >   available src address. But if the target is link local address,
> >   ipv6_dev_get_saddr() will only get available address from out interace,
> 
> 	Should this be "our interface"?

Ah, yes.

> >
> >2. On the receive side. The slave was set down before enslave to bond.
> >   This makes slaves remove mcast address 33:33:00:00:00:01( The IPv6
> >   maddr ff02::1 is kept even when the interface down). When bond set
> >   slave up, the ipv6_mc_up() was not called due to commit c2edacf80e15
> >   ("bonding / ipv6: no addrconf for slaves separately from master").
> >   This makes the slave interface never add the all node mcast address
> >   33:33:00:00:00:01. So there is no way to accept unsolicited NA with
> >   dest ff02::1.
> >
> >   Fix this by adding all node mcast address 33:33:00:00:00:01 back when
> >   the slave interface up.
> >
> >3. On the validating side. The NA message with all-nodes multicast dest
> >   address should also be valid.
> >
> >   Also rename bond_validate_ns() to bond_validate_na().
> 
> 	I'm not exactly sure which change matches which of the three
> above fixes; should this be three separate patches?

The 1st case(send side) is fixed in function bond_ns_send_all().
The 2nd case(receive side) is fixed in addrconf_notify().
The 3rd case(validating side) is fixed in bond_validate_ns/na()

> 
> >Reported-by: LiLiang <liali@...hat.com>
> >Fixes: 5e1eeef69c0f ("bonding: NS target should accept link local address")
> 
> 	Is this fixes tag correct for all the fixes?  Number 2 cites a
> different commit (c2edacf80e15). 

Before we support link local target for bonding. Commit (c2edacf80e15) works
as bond device could up and add the all node multicast correctly.

After we adding the link local target for bonding. The bond could not up
and not able to add node multicast address. So I think the fixes tag should
not be commit (c2edacf80e15).

> Again, should these be three separate patches?

I thought these 3 parts are all to fix lladdr target. So I put them together.
If you think it's easier to review. I can separate the patches of course.

> 
> >@@ -3246,14 +3256,14 @@ static int bond_na_rcv(const struct sk_buff *skb, struct bonding *bond,
> > 	 * see bond_arp_rcv().
> > 	 */
> > 	if (bond_is_active_slave(slave))
> >-		bond_validate_ns(bond, slave, saddr, daddr);
> >+		bond_validate_na(bond, slave, saddr, daddr);
> > 	else if (curr_active_slave &&
> > 		 time_after(slave_last_rx(bond, curr_active_slave),
> > 			    curr_active_slave->last_link_up))
> >-		bond_validate_ns(bond, slave, saddr, daddr);
> >+		bond_validate_na(bond, slave, saddr, daddr);
> > 	else if (curr_arp_slave &&
> > 		 bond_time_in_interval(bond, slave_last_tx(curr_arp_slave), 1))
> >-		bond_validate_ns(bond, slave, saddr, daddr);
> >+		bond_validate_na(bond, slave, saddr, daddr);
> 
> 	Is this logic correct?  If I'm not mistaken, there are two
> receive cases:
> 
> 	1- We receive a reply (Neighbor Advertisement) to our request
> (Neighbor Solicitation).
> 
> 	2- We receive a copy of our request (NS), which passed through
> the switch and was received by another interface of the bond.

No, we don't have this case for IPv6 because I did a check in

static int bond_na_rcv(const struct sk_buff *skb, struct bonding *bond,
                       struct slave *slave)
{
	[...]

        if (skb->pkt_type == PACKET_OTHERHOST ||
            skb->pkt_type == PACKET_LOOPBACK ||
            hdr->icmp6_type != NDISC_NEIGHBOUR_ADVERTISEMENT)
                goto out;

Here we will ignore none NA messages.

Thanks
Hangbin
> 
> 	For the ARP monitor implementation, in the second case, the
> source and target IP addresses are swapped for the validation.
> 
> 	Is such a swap necessary for the NS/NA monitor implementation?
> I would expect this to be in the second block of the if (inside the
> "else if (curr_active_slave &&" block).
> 
> 	-J
> 
> > out:
> > 	return RX_HANDLER_ANOTHER;
> >diff --git a/net/ipv6/addrconf.c b/net/ipv6/addrconf.c
> >index e15f64f22fa8..77750b6327e7 100644
> >--- a/net/ipv6/addrconf.c
> >+++ b/net/ipv6/addrconf.c
> >@@ -3557,11 +3557,14 @@ static int addrconf_notify(struct notifier_block *this, unsigned long event,
> > 		fallthrough;
> > 	case NETDEV_UP:
> > 	case NETDEV_CHANGE:
> >-		if (dev->flags & IFF_SLAVE)
> >+		if (idev && idev->cnf.disable_ipv6)
> > 			break;
> > 
> >-		if (idev && idev->cnf.disable_ipv6)
> >+		if (dev->flags & IFF_SLAVE) {
> >+			if (event == NETDEV_UP && !IS_ERR_OR_NULL(idev))
> >+				ipv6_mc_up(idev);
> > 			break;
> >+		}
> > 
> > 		if (event == NETDEV_UP) {
> > 			/* restore routes for permanent addresses */
> >-- 
> >2.37.1
> >
> 
> ---
> 	-Jay Vosburgh, jay.vosburgh@...onical.com

Powered by blists - more mailing lists