Re: [net-next] ipv6: fix routing cache overflow for raw sockets

[Jonathan Maxwell <jmaxwell37@gmail.com>]

On Wed, Dec 21, 2022 at 8:55 AM Jonathan Maxwell <jmaxwell37@gmail.com> wrote:
>
> On Wed, Dec 21, 2022 at 2:10 AM David Ahern <dsahern@kernel.org> wrote:
> >
> > On 12/20/22 5:35 AM, Paolo Abeni wrote:
> > > On Mon, 2022-12-19 at 10:48 +1100, Jon Maxwell wrote:
> > >> Sending Ipv6 packets in a loop via a raw socket triggers an issue where a
> > >> route is cloned by ip6_rt_cache_alloc() for each packet sent. This quickly
> > >> consumes the Ipv6 max_size threshold which defaults to 4096 resulting in
> > >> these warnings:
> > >>
> > >> [1]   99.187805] dst_alloc: 7728 callbacks suppressed
> > >> [2] Route cache is full: consider increasing sysctl net.ipv6.route.max_size.
> > >> .
> > >> .
> > >> [300] Route cache is full: consider increasing sysctl net.ipv6.route.max_size.
> > >
> > > If I read correctly, the maximum number of dst that the raw socket can
> > > use this way is limited by the number of packets it allows via the
> > > sndbuf limit, right?
> > >
> > > Are other FLOWI_FLAG_KNOWN_NH users affected, too? e.g. nf_dup_ipv6,
> > > ipvs, seg6?
> > >
> > > @DavidA: why do we need to create RTF_CACHE clones for KNOWN_NH flows?
> > >
> > > Thanks,
> > >
> > > Paolo
> > >
> >
> > If I recall the details correctly: that sysctl limit was added back when
> > ipv6 routes were managed as dst_entries and there was a desire to allow
> > an admin to limit the memory consumed. At this point in time, IPv6 is
> > more inline with IPv4 - a separate struct for fib entries from dst
> > entries. That "Route cache is full" message is now out of date since
> > this is dst_entries which have a gc mechanism.
> >
> > IPv4 does not limit the number of dst_entries that can be allocated
> > (ip_rt_max_size is the sysctl variable behind the ipv4 version of
> > max_size and it is a no-op). IPv6 can probably do the same here?
> >
>
> diff --git a/net/ipv6/route.c b/net/ipv6/route.c
> index dbc224023977..701aba7feaf5 100644
> --- a/net/ipv6/route.c
> +++ b/net/ipv6/route.c
> @@ -6470,7 +6470,7 @@ static int __net_init ip6_route_net_init(struct net *net)
>  #endif
>
>         net->ipv6.sysctl.flush_delay = 0;
> -       net->ipv6.sysctl.ip6_rt_max_size = 4096;
> +       net->ipv6.sysctl.ip6_rt_max_size = INT_MAX;
>         net->ipv6.sysctl.ip6_rt_gc_min_interval = HZ / 2;
>         net->ipv6.sysctl.ip6_rt_gc_timeout = 60*HZ;
>         net->ipv6.sysctl.ip6_rt_gc_interval = 30*HZ;
>
> The above patch resolved it for the Ipv6 reproducer.
>
> Would that be sufficient?
>

Otherwise if you prefer to make Ipv6 behaviour similar to IPv4.
Rather than upping max_size.

Here is prototype patch that removes the max_size check for Ipv6:

diff --git a/include/net/dst_ops.h b/include/net/dst_ops.h
index 88ff7bb2bb9b..632086b2f644 100644
--- a/include/net/dst_ops.h
+++ b/include/net/dst_ops.h
@@ -16,7 +16,7 @@ struct dst_ops {
        unsigned short          family;
        unsigned int            gc_thresh;

-       int                     (*gc)(struct dst_ops *ops);
+       void                    (*gc)(struct dst_ops *ops);
        struct dst_entry *      (*check)(struct dst_entry *, __u32 cookie);
        unsigned int            (*default_advmss)(const struct dst_entry *);
        unsigned int            (*mtu)(const struct dst_entry *);
diff --git a/net/core/dst.c b/net/core/dst.c
index 497ef9b3fc6a..dcb85267bc4c 100644
--- a/net/core/dst.c
+++ b/net/core/dst.c
@@ -82,12 +82,8 @@ void *dst_alloc(struct dst_ops *ops, struct net_device *dev,

        if (ops->gc &&
            !(flags & DST_NOCOUNT) &&
-           dst_entries_get_fast(ops) > ops->gc_thresh) {
-               if (ops->gc(ops)) {
-                       pr_notice_ratelimited("Route cache is full:
consider increasing sysctl net.ipv6.route.max_size.\n");
-                       return NULL;
-               }
-       }
+           dst_entries_get_fast(ops) > ops->gc_thresh)
+               ops->gc(ops);

        dst = kmem_cache_alloc(ops->kmem_cachep, GFP_ATOMIC);
        if (!dst)
diff --git a/net/ipv6/route.c b/net/ipv6/route.c
index dbc224023977..8db7c5436da4 100644
--- a/net/ipv6/route.c
+++ b/net/ipv6/route.c
@@ -91,7 +91,7 @@ static struct dst_entry *ip6_negative_advice(struct
dst_entry *);
 static void            ip6_dst_destroy(struct dst_entry *);
 static void            ip6_dst_ifdown(struct dst_entry *,
                                       struct net_device *dev, int how);
-static int              ip6_dst_gc(struct dst_ops *ops);
+static void             ip6_dst_gc(struct dst_ops *ops);

 static int             ip6_pkt_discard(struct sk_buff *skb);
 static int             ip6_pkt_discard_out(struct net *net, struct
sock *sk, struct sk_buff *skb);
@@ -3295,32 +3295,21 @@ struct dst_entry *icmp6_dst_alloc(struct
net_device *dev,
        return dst;
 }

-static int ip6_dst_gc(struct dst_ops *ops)
+static void ip6_dst_gc(struct dst_ops *ops)
 {
        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;
        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;
        int entries;

        entries = dst_entries_get_fast(ops);
-       if (entries > rt_max_size)
-               entries = dst_entries_get_slow(ops);
-
-       if (time_after(rt_last_gc + rt_min_interval, jiffies) &&
-           entries <= rt_max_size)
-               goto out;

        net->ipv6.ip6_rt_gc_expire++;
        fib6_run_gc(net->ipv6.ip6_rt_gc_expire, net, true);
        entries = dst_entries_get_slow(ops);
        if (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 entries > rt_max_size;
 }

 static int ip6_nh_lookup_table(struct net *net, struct fib6_config *cfg,

> > I do not believe the suggested flag is the right change.
>
> Regards
>
> Jon