Re: [PATCH v3] net: use sock_gen_put() when sk_state is TCP_TIME_WAIT

[Jacob Keller <jacob.e.keller@intel.com>]

On 4/28/2025 6:59 PM, Shiming Cheng wrote:
> From: Jibin Zhang <jibin.zhang@mediatek.com>
> 
> It is possible for a pointer of type struct inet_timewait_sock to be
> returned from the functions __inet_lookup_established() and
> __inet6_lookup_established(). This can cause a crash when the
> returned pointer is of type struct inet_timewait_sock and
> sock_put() is called on it. The following is a crash call stack that
> shows sk->sk_wmem_alloc being accessed in sk_free() during the call to
> sock_put() on a struct inet_timewait_sock pointer. To avoid this issue,
> use sock_gen_put() instead of sock_put() when sk->sk_state
> is TCP_TIME_WAIT.
> 
> mrdump.ko        ipanic() + 120
> vmlinux          notifier_call_chain(nr_to_call=-1, nr_calls=0) + 132
> vmlinux          atomic_notifier_call_chain(val=0) + 56
> vmlinux          panic() + 344
> vmlinux          add_taint() + 164
> vmlinux          end_report() + 136
> vmlinux          kasan_report(size=0) + 236
> vmlinux          report_tag_fault() + 16
> vmlinux          do_tag_recovery() + 16
> vmlinux          __do_kernel_fault() + 88
> vmlinux          do_bad_area() + 28
> vmlinux          do_tag_check_fault() + 60
> vmlinux          do_mem_abort() + 80
> vmlinux          el1_abort() + 56
> vmlinux          el1h_64_sync_handler() + 124
> vmlinux        > 0xFFFFFFC080011294()
> vmlinux          __lse_atomic_fetch_add_release(v=0xF2FFFF82A896087C)
> vmlinux          __lse_atomic_fetch_sub_release(v=0xF2FFFF82A896087C)
> vmlinux          arch_atomic_fetch_sub_release(i=1, v=0xF2FFFF82A896087C)
> + 8
> vmlinux          raw_atomic_fetch_sub_release(i=1, v=0xF2FFFF82A896087C)
> + 8
> vmlinux          atomic_fetch_sub_release(i=1, v=0xF2FFFF82A896087C) + 8
> vmlinux          __refcount_sub_and_test(i=1, r=0xF2FFFF82A896087C,
> oldp=0) + 8
> vmlinux          __refcount_dec_and_test(r=0xF2FFFF82A896087C, oldp=0) + 8
> vmlinux          refcount_dec_and_test(r=0xF2FFFF82A896087C) + 8
> vmlinux          sk_free(sk=0xF2FFFF82A8960700) + 28
> vmlinux          sock_put() + 48
> vmlinux          tcp6_check_fraglist_gro() + 236
> vmlinux          tcp6_gro_receive() + 624
> vmlinux          ipv6_gro_receive() + 912
> vmlinux          dev_gro_receive() + 1116
> vmlinux          napi_gro_receive() + 196
> ccmni.ko         ccmni_rx_callback() + 208
> ccmni.ko         ccmni_queue_recv_skb() + 388
> ccci_dpmaif.ko   dpmaif_rxq_push_thread() + 1088
> vmlinux          kthread() + 268
> vmlinux          0xFFFFFFC08001F30C()
> 
> Fixes: c9d1d23e5239 ("net: add heuristic for enabling TCP fraglist GRO")
> Signed-off-by: Jibin Zhang <jibin.zhang@mediatek.com>
> Signed-off-by: Shiming Cheng <shiming.cheng@mediatek.com>
> ---

The code change looks good to me.

Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>

One note is that it helps reviewers if you could include a terse summary
of the changes between versions. This is marked as v3, but there's no
mention of what changed.

>  net/ipv4/tcp_offload.c   | 2 +-
>  net/ipv6/tcpv6_offload.c | 2 +-
>  2 files changed, 2 insertions(+), 2 deletions(-)
> 
> diff --git a/net/ipv4/tcp_offload.c b/net/ipv4/tcp_offload.c
> index 2308665b51c5..f55026b597ff 100644
> --- a/net/ipv4/tcp_offload.c
> +++ b/net/ipv4/tcp_offload.c
> @@ -432,7 +432,7 @@ static void tcp4_check_fraglist_gro(struct list_head *head, struct sk_buff *skb,
>  				       iif, sdif);
>  	NAPI_GRO_CB(skb)->is_flist = !sk;
>  	if (sk)
> -		sock_put(sk);
> +		sock_gen_put(sk);
>  }
>  
>  INDIRECT_CALLABLE_SCOPE
> diff --git a/net/ipv6/tcpv6_offload.c b/net/ipv6/tcpv6_offload.c
> index a45bf17cb2a1..b1f284e0c15a 100644
> --- a/net/ipv6/tcpv6_offload.c
> +++ b/net/ipv6/tcpv6_offload.c
> @@ -42,7 +42,7 @@ static void tcp6_check_fraglist_gro(struct list_head *head, struct sk_buff *skb,
>  					iif, sdif);
>  	NAPI_GRO_CB(skb)->is_flist = !sk;
>  	if (sk)
> -		sock_put(sk);
> +		sock_gen_put(sk);
>  #endif /* IS_ENABLED(CONFIG_IPV6) */
>  }
>