Bug 969950

(In reply to Geng Sheng Liu from comment #0)
> Additional info:
> 
> https://access.redhat.com/site/solutions/36876

This seems completely unrelated.

> http://post-office.corp.redhat.com/archives/tech-list/2010-February/msg00468.
> html

Basically I'd say the same as Neil did back in 2010, meaning IPv4 behaves incorrectly here.

Out of curiosity, could you elaborate on use cases of this "feature"?

Correction:

KBase related is as below

https://access.redhat.com/site/solutions/27284

(In reply to Petr Šabata from comment #2)
> (In reply to Geng Sheng Liu from comment #0)
> > Additional info:
> > 
> > https://access.redhat.com/site/solutions/36876
> 
> This seems completely unrelated.
> 
> > http://post-office.corp.redhat.com/archives/tech-list/2010-February/msg00468.
> > html
> 
> Basically I'd say the same as Neil did back in 2010, meaning IPv4 behaves
> incorrectly here.
> 
> Out of curiosity, could you elaborate on use cases of this "feature"?
This feature is useful for stress testing. We used to simulated tens of thousands clients or severs on one blade server in IPv4 scenario, but in IPv6 scenario, we can't do same things again.

Hi,

We found a workaround for this issue:

Steps:
1. add a IPv6 address on loopback interface
2. delete the unreachable route entry
3. add a local route entry

Example:
[root@bee12 ~]# ip -6 addr add fd96:38d8:6f3d:5e::/64 dev lo
[root@bee12 ~]# ip -6 route del unreachable fd96:38d8:6f3d:5e::/64
[root@bee12 ~]# ip -6 route add table local local fd96:38d8:6f3d:5e::/64 dev lo proto  kernel scope host

All IPv6 addresses under that network prefix can be accessed:
[root@bee12 ~]# ping6 fd96:38d8:6f3d:5e::1
PING fd96:38d8:6f3d:5e::1(fd96:38d8:6f3d:5e::1) 56 data bytes
64 bytes from fd96:38d8:6f3d:5e::1: icmp_seq=1 ttl=64 time=0.028 ms
^C
--- fd96:38d8:6f3d:5e::1 ping statistics ---
1 packets transmitted, 1 received, 0% packet loss, time 730ms
rtt min/avg/max/mdev = 0.028/0.028/0.028/0.000 ms
[root@bee12 ~]# ping6 fd96:38d8:6f3d:5e::2
PING fd96:38d8:6f3d:5e::2(fd96:38d8:6f3d:5e::2) 56 data bytes
64 bytes from fd96:38d8:6f3d:5e::2: icmp_seq=1 ttl=64 time=0.024 ms
^C
--- fd96:38d8:6f3d:5e::2 ping statistics ---
1 packets transmitted, 1 received, 0% packet loss, time 453ms
rtt min/avg/max/mdev = 0.024/0.024/0.024/0.000 ms
[root@bee12 ~]# ping6 fd96:38d8:6f3d:5e::fff1
PING fd96:38d8:6f3d:5e::fff1(fd96:38d8:6f3d:5e::fff1) 56 data bytes
64 bytes from fd96:38d8:6f3d:5e::fff1: icmp_seq=1 ttl=64 time=0.027 ms
^C
--- fd96:38d8:6f3d:5e::fff1 ping statistics ---
1 packets transmitted, 1 received, 0% packet loss, time 676ms
rtt min/avg/max/mdev = 0.027/0.027/0.027/0.000 ms

I would like to note that this issue is specific to the loopback interface, as this bug report is used as a reference elsewhere.

It is related to the fact that the IPv6 standards do not define a loopback subnet but a single loopback address and therefore there is no need for it under normal circumstances.

(In reply to Hafid Lin from comment #6)
> Hi,
> 
> We found a workaround for this issue:
> 
> Steps:
> 1. add a IPv6 address on loopback interface
> 2. delete the unreachable route entry
> 3. add a local route entry

You might also want to try skipping steps #1 and #2 and only do #3 if the only thing you need to get it work is the local route entry itself.

> Example:
> [root@bee12 ~]# ip -6 addr add fd96:38d8:6f3d:5e::/64 dev lo
> [root@bee12 ~]# ip -6 route del unreachable fd96:38d8:6f3d:5e::/64
> [root@bee12 ~]# ip -6 route add table local local fd96:38d8:6f3d:5e::/64 dev
> lo proto  kernel scope host

As a sidenote, when I test this on 3.8.11 kernel (3.8.11-200.fc18.x86_64), and perform the three steps above, the ping won't work and reports 'Invalid argument' instead. Tried several times, used different IP ranges but might still be doing something wrong.

increase the kernel parameter below can solve the problem.

"net/core/dst.c"

    if (ops->gc && atomic_read(&ops->entries) > ops->gc_thresh) {
                if (ops->gc(ops))     <-----通过函数指针调用ip6_dst_gc
                        return NULL;

"net/ipv6/route.c"

static int ip6_dst_gc(struct dst_ops *ops)     <---------ipv6的gc函数
{
        unsigned long now = jiffies;
        struct net *net = container_of(ops, struct net, ipv6.ip6_dst_ops);
        int rt_min_interval = net->ipv6.sysctl.ip6_rt_gc_min_interval;
        int rt_max_size = net->ipv6.sysctl.ip6_rt_max_size;       <----------rt max size
        int rt_elasticity = net->ipv6.sysctl.ip6_rt_gc_elasticity;
        int rt_gc_timeout = net->ipv6.sysctl.ip6_rt_gc_timeout;
        unsigned long rt_last_gc = net->ipv6.ip6_rt_last_gc;

        if (time_after(rt_last_gc + rt_min_interval, now) &&
            atomic_read(&ops->entries) <= rt_max_size)
                goto out;

        net->ipv6.ip6_rt_gc_expire++;
        fib6_run_gc(net->ipv6.ip6_rt_gc_expire, net);
        net->ipv6.ip6_rt_last_gc = now;
        if (atomic_read(&ops->entries) < ops->gc_thresh)
                net->ipv6.ip6_rt_gc_expire = rt_gc_timeout>>1;
out:
        net->ipv6.ip6_rt_gc_expire -= net->ipv6.ip6_rt_gc_expire>>rt_elasticity;
        return (atomic_read(&ops->entries) > rt_max_size);        ------->here......
}

"net/ipv6/route.c" 
        net->ipv6.sysctl.flush_delay = 0;
        net->ipv6.sysctl.ip6_rt_max_size = 4096;    ----> default value of rt_max_size is 4096
        net->ipv6.sysctl.ip6_rt_gc_min_interval = HZ / 2;
        net->ipv6.sysctl.ip6_rt_gc_timeout = 60*HZ;


net.ipv6.route.max_size = 116184
(In reply to Pavel Šimerda from comment #7)
> I would like to note that this issue is specific to the loopback interface,
> as this bug report is used as a reference elsewhere.
> 
> It is related to the fact that the IPv6 standards do not define a loopback
> subnet but a single loopback address and therefore there is no need for it
> under normal circumstances.
> 
> (In reply to Hafid Lin from comment #6)
> > Hi,
> > 
> > We found a workaround for this issue:
> > 
> > Steps:
> > 1. add a IPv6 address on loopback interface
> > 2. delete the unreachable route entry
> > 3. add a local route entry
> 
> You might also want to try skipping steps #1 and #2 and only do #3 if the
> only thing you need to get it work is the local route entry itself.
> 
> > Example:
> > [root@bee12 ~]# ip -6 addr add fd96:38d8:6f3d:5e::/64 dev lo
> > [root@bee12 ~]# ip -6 route del unreachable fd96:38d8:6f3d:5e::/64
> > [root@bee12 ~]# ip -6 route add table local local fd96:38d8:6f3d:5e::/64 dev
> > lo proto  kernel scope host
> 
> As a sidenote, when I test this on 3.8.11 kernel (3.8.11-200.fc18.x86_64),
> and perform the three steps above, the ping won't work and reports 'Invalid
> argument' instead. Tried several times, used different IP ranges but might
> still be doing something wrong.

As per Pavel's comment, this is not really an issue and the requested behavior can be achieved by adding the local route.

The proposed change in Comment #8 is for kernel and the customer ticket is already resolved.  I'm closing this as NOTABUG.