Re: [PATCH net v2 1/5] ipv6: restrict gc_elasticity to 31

On 10/29/26 7:58 AM, Kuniyuki Iwashima wrote:

On Tue, Oct 27, 2026 at 3:48 AM Jiayuan Chen [off-list ref] wrote:

quoted

Every counted IPv6 dst allocation runs ip6_dst_gc() once the number of
live dst goes over gc_thresh (1024 by default). ip6_dst_gc() decays
ip6_rt_gc_expire by gc_elasticity: the smaller gc_elasticity is, the
smaller ip6_rt_gc_expire gets, i.e. the more aggressive the gc.

ip6_dst_gc() reads ip6_rt_gc_expire into a u32, but gc_elasticity has no
bounds at all, so writing >= 32 shifts a 32-bit value by >= 32 bits, and
a negative value gives a negative shift count. Both are UB:

         val = atomic_read(&net->ipv6.ip6_rt_gc_expire);
         atomic_set(&net->ipv6.ip6_rt_gc_expire, val - (val >> rt_elasticity));

   UBSAN: shift-out-of-bounds in net/ipv6/route.c:3369:53
   shift exponent 32 is too large for 32-bit type 'unsigned int'
   CPU: 2 UID: 0 PID: 615 Comm: ping6 Not tainted 7.3.0-rc7+ #364 PREEMPT
   Call Trace:
    <IRQ>
    dump_stack_lvl+0xc1/0xf0
    dump_stack+0x10/0x20
    __ubsan_handle_shift_out_of_bounds+0x1d1/0x390
    ip6_dst_gc.cold+0x15/0x1a
    dst_alloc+0x4e0/0x620
    ip6_rt_cache_alloc+0x189/0x7e0
    __ip6_rt_update_pmtu+0x68e/0x1160
    ip6_update_pmtu+0x30c/0x3a0
    icmpv6_err+0x1fa/0x330
    icmpv6_notify+0x303/0x770
    icmpv6_rcv+0xc12/0x1670
    ip6_protocol_deliver_rcu+0xb4a/0x1520
    ip6_input_finish+0x19c/0x400
    ip6_input+0xd4/0x240
    ipv6_rcv+0x1fd/0x330
    __netif_receive_skb_one_core+0x118/0x1c0
    __netif_receive_skb+0x21/0x150
    process_backlog+0x37e/0x1200
    __napi_poll.constprop.0+0xa5/0x5e0
    net_rx_action+0x4c0/0xf70
    handle_softirqs+0x204/0x9b0
    __do_softirq+0x10/0x18

Make it a u8 and reject writes outside 0..31 with proc_dou8vec_minmax().

Fixes: 1da177e4c3f4 ("Linux-2.6.12-rc2")
Cc: <redacted>
Reviewed-by: Fernando Fernandez Mancera <redacted>
Reviewed-by: Kuniyuki Iwashima <kuniyu@google.com>
Signed-off-by: Jiayuan Chen <jiayuan.chen@linux.dev>
---
v2: switched to u8 and proc_dou8vec_minmax(), as suggested by Kuniyuki
Iwashima.
---
  include/net/netns/ipv6.h | 2 +-
  net/ipv6/route.c         | 7 +++++--
  2 files changed, 6 insertions(+), 3 deletions(-)

diff --git a/include/net/netns/ipv6.h b/include/net/netns/ipv6.h
index ffda15318be59..aefce37c04a4f 100644
--- a/include/net/netns/ipv6.h
+++ b/include/net/netns/ipv6.h

@@ -25,11 +25,11 @@ struct netns_sysctl_ipv6 {
         int ip6_rt_gc_min_interval;
         int ip6_rt_gc_timeout;
         int ip6_rt_gc_interval;
-       int ip6_rt_gc_elasticity;
         int ip6_rt_mtu_expires;
         int ip6_rt_min_advmss;
         u32 multipath_hash_fields;
         u8 multipath_hash_policy;
+       u8 ip6_rt_gc_elasticity;

         __cacheline_group_begin(sysctl_ipv6_flowlabel);
         u8 flowlabel_consistency;

diff --git a/net/ipv6/route.c b/net/ipv6/route.c
index 94e5ef1c4f83e..abc8b811cd423 100644
--- a/net/ipv6/route.c
+++ b/net/ipv6/route.c

@@ -6579,6 +6579,8 @@ static int ipv6_sysctl_rtcache_flush(const struct ctl_table *ctl, int write,
         return 0;
  }

+static int ip6_rt_gc_elasticity_max = 31;

as you need v3, s/int/u8/

Hi Kuniyuki,

proc_dou8vec_minmax() put extra1/extra2 into proc_uint_conv(), which
reads them as unsigned int, so the max has to stay an int like
tcp_syn_retries_max and friends; a u8 there would read
three bytes past the variable.