Re: [PATCH net] bonding: alb: fix UAF in rlb_arp_recv during bond up/down

[Nikolay Aleksandrov <nikolay@nvidia.com>]

On Fri, Feb 13, 2026 at 09:54:24AM +0000, Hangbin Liu wrote:
> The ALB RX path may access rx_hashtbl concurrently with bond
> teardown. During rapid bond up/down cycles, rlb_deinitialize()
> frees rx_hashtbl while RX handlers are still running, leading
> to a use-after-free detected by KASAN.
> 
> [  214.174138] Oops: general protection fault, probably for non-canonical address 0xdffffc000000001d: 0000 [#1] SMP KASAN PTI
> [  214.186478] KASAN: null-ptr-deref in range [0x00000000000000e8-0x00000000000000ef]
> [  214.194933] CPU: 30 UID: 0 PID: 2375 Comm: ping Kdump: loaded Not tainted 6.19.0-rc8+ #2 PREEMPT(voluntary)
> [  214.205907] Hardware name: Dell Inc. PowerEdge R730/0WCJNT, BIOS 2.14.0 01/14/2022
> [  214.214357] RIP: 0010:rlb_arp_recv+0x505/0xab0 [bonding]
> [  214.220320] Code: 0f 85 2b 05 00 00 48 b8 00 00 00 00 00 fc ff df 40 0f b6 ed 48 c1 e5 06 49 03 ad 78 01 00 00 48 8d 7d 28 48 89 fa 48 c1 ea 03 <0f> b6
>  04 02 84 c0 74 06 0f 8e 12 05 00 00 80 7d 28 00 0f 84 8c 00
> [  214.241280] RSP: 0018:ffffc900073d8870 EFLAGS: 00010206
> [  214.247116] RAX: dffffc0000000000 RBX: ffff888168556822 RCX: ffff88816855681e
> [  214.255082] RDX: 000000000000001d RSI: dffffc0000000000 RDI: 00000000000000e8
> [  214.263048] RBP: 00000000000000c0 R08: 0000000000000002 R09: ffffed11192021c8
> [  214.271013] R10: ffff8888c9010e43 R11: 0000000000000001 R12: 1ffff92000e7b119
> [  214.278978] R13: ffff8888c9010e00 R14: ffff888168556822 R15: ffff888168556810
> [  214.286943] FS:  00007f85d2d9cb80(0000) GS:ffff88886ccb3000(0000) knlGS:0000000000000000
> [  214.295966] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
> [  214.302380] CR2: 00007f0d047b5e34 CR3: 00000008a1c2e002 CR4: 00000000001726f0
> [  214.310347] Call Trace:
> [  214.313070]  <IRQ>
> [  214.315318]  ? __pfx_rlb_arp_recv+0x10/0x10 [bonding]                                                                                                  [  214.320975]  bond_handle_frame+0x166/0xb60 [bonding]
> [  214.326537]  ? __pfx_bond_handle_frame+0x10/0x10 [bonding]
> [  214.332680]  __netif_receive_skb_core.constprop.0+0x576/0x2710
> [  214.339199]  ? __pfx_arp_process+0x10/0x10
> [  214.343775]  ? sched_balance_find_src_group+0x98/0x630
> [  214.349513]  ? __pfx___netif_receive_skb_core.constprop.0+0x10/0x10
> [  214.356513]  ? arp_rcv+0x307/0x690
> [  214.360311]  ? __pfx_arp_rcv+0x10/0x10
> [  214.364499]  ? __lock_acquire+0x58c/0xbd0
> [  214.368975]  __netif_receive_skb_one_core+0xae/0x1b0
> [  214.374518]  ? __pfx___netif_receive_skb_one_core+0x10/0x10
> [  214.380743]  ? lock_acquire+0x10b/0x140
> [  214.385026]  process_backlog+0x3f1/0x13a0
> [  214.389502]  ? process_backlog+0x3aa/0x13a0
> [  214.394174]  __napi_poll.constprop.0+0x9f/0x370
> [  214.399233]  net_rx_action+0x8c1/0xe60
> [  214.403423]  ? __pfx_net_rx_action+0x10/0x10
> [  214.408193]  ? lock_acquire.part.0+0xbd/0x260
> [  214.413058]  ? sched_clock_cpu+0x6c/0x540
> [  214.417540]  ? mark_held_locks+0x40/0x70
> [  214.421920]  handle_softirqs+0x1fd/0x860
> [  214.426302]  ? __pfx_handle_softirqs+0x10/0x10
> [  214.431264]  ? __neigh_event_send+0x2d6/0xf50
> [  214.436131]  do_softirq+0xb1/0xf0
> [  214.439830]  </IRQ>
> 
> The issue is reproducible by looping ip link set bond0 up/down,
> where rlb_arp_recv() can race with rlb_deinitialize() and dereference
> a freed rx_hashtbl entry.
> 
> Fix this by setting recv_probe to NULL before calling
> bond_alb_deinitialize(), and then calling synchronize_net() to wait
> for any concurrent RX processing to finish.
> 
> This ensures that no RX handler can access rx_hashtbl after it is
> freed, while preserving the existing locking semantics.
> 
> Reported-by: Liang Li <liali@redhat.com>
> Fixes: e53665c6eaa6 ("bonding: delete migrated IP addresses from the rlb hash table")
> Signed-off-by: Hangbin Liu <liuhangbin@gmail.com>
> ---
>  drivers/net/bonding/bond_main.c | 6 +++++-
>  1 file changed, 5 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/net/bonding/bond_main.c b/drivers/net/bonding/bond_main.c
> index 47f13d86cb7e..8e1057a2a061 100644
> --- a/drivers/net/bonding/bond_main.c
> +++ b/drivers/net/bonding/bond_main.c
> @@ -4314,9 +4314,13 @@ static int bond_close(struct net_device *bond_dev)
>  
>  	bond_work_cancel_all(bond);
>  	bond->send_peer_notif = 0;
> +	bond->recv_probe = NULL;
> +
> +	/* Wait for any in-flight RX handlers */
> +	synchronize_net();
> +
>  	if (bond_is_lb(bond))
>  		bond_alb_deinitialize(bond);
> -	bond->recv_probe = NULL;
>  
>  	if (BOND_MODE(bond) == BOND_MODE_8023AD &&
>  	    bond->params.broadcast_neighbor)
> -- 
> 2.50.1
> 

Ouch, good catch! By the way I have a (very old) patch that properly
converts recv_probe to RCU, currently it's missing any annotations on
the write side. I think I'll revive it when net-next opens up. :)

Reviewed-by: Nikolay Aleksandrov <nikolay@nvidia.com>

Cheers,
 Nik