Re: [PATCH net v2] net: fix use-after-free when UDP GRO with shared fraglist

[Daniel Borkmann]

On 1/7/21 1:39 AM, Dongseok Yi wrote:
> skbs in fraglist could be shared by a BPF filter loaded at TC. It
> triggers skb_ensure_writable -> pskb_expand_head ->
> skb_clone_fraglist -> skb_get on each skb in the fraglist.
>
> While tcpdump, sk_receive_queue of PF_PACKET has the original fraglist.
> But the same fraglist is queued to PF_INET (or PF_INET6) as the fraglist
> chain made by skb_segment_list.
>
> If the new skb (not fraglist) is queued to one of the sk_receive_queue,
> multiple ptypes can see this. The skb could be released by ptypes and
> it causes use-after-free.
>
> [ 4443.426215] ------------[ cut here ]------------
> [ 4443.426222] refcount_t: underflow; use-after-free.
> [ 4443.426291] WARNING: CPU: 7 PID: 28161 at lib/refcount.c:190
> refcount_dec_and_test_checked+0xa4/0xc8
> [ 4443.426726] pstate: 60400005 (nZCv daif +PAN -UAO)
> [ 4443.426732] pc : refcount_dec_and_test_checked+0xa4/0xc8
> [ 4443.426737] lr : refcount_dec_and_test_checked+0xa0/0xc8
> [ 4443.426808] Call trace:
> [ 4443.426813]  refcount_dec_and_test_checked+0xa4/0xc8
> [ 4443.426823]  skb_release_data+0x144/0x264
> [ 4443.426828]  kfree_skb+0x58/0xc4
> [ 4443.426832]  skb_queue_purge+0x64/0x9c
> [ 4443.426844]  packet_set_ring+0x5f0/0x820
> [ 4443.426849]  packet_setsockopt+0x5a4/0xcd0
> [ 4443.426853]  __sys_setsockopt+0x188/0x278
> [ 4443.426858]  __arm64_sys_setsockopt+0x28/0x38
> [ 4443.426869]  el0_svc_common+0xf0/0x1d0
> [ 4443.426873]  el0_svc_handler+0x74/0x98
> [ 4443.426880]  el0_svc+0x8/0xc
>
> Fixes: 3a1296a38d0c (net: Support GRO/GSO fraglist chaining.)
> Signed-off-by: Dongseok Yi <dseok.yi@xxxxxxxxxxx>
> Acked-by: Willem de Bruijn <willemb@xxxxxxxxxx>
> ---
>   net/core/skbuff.c | 20 +++++++++++++++++++-
>   1 file changed, 19 insertions(+), 1 deletion(-)
>
> v2: Expand the commit message to clarify a BPF filter loaded
>
> diff --git a/net/core/skbuff.c b/net/core/skbuff.c
> index f62cae3..1dcbda8 100644
> --- a/net/core/skbuff.c
> +++ b/net/core/skbuff.c
> @@ -3655,7 +3655,8 @@ struct sk_buff *skb_segment_list(struct sk_buff *skb,
>   	unsigned int delta_truesize = 0;
>   	unsigned int delta_len = 0;
>   	struct sk_buff *tail = NULL;
> -	struct sk_buff *nskb;
> +	struct sk_buff *nskb, *tmp;
> +	int err;
>   
>   	skb_push(skb, -skb_network_offset(skb) + offset);
>   
> @@ -3665,11 +3666,28 @@ struct sk_buff *skb_segment_list(struct sk_buff *skb,
>   		nskb = list_skb;
>   		list_skb = list_skb->next;
>   
> +		err = 0;
> +		if (skb_shared(nskb)) {
> +			tmp = skb_clone(nskb, GFP_ATOMIC);
> +			if (tmp) {
> +				kfree_skb(nskb);

Should use consume_skb() to not trigger skb:kfree_skb tracepoint when looking
for drops in the stack.

> +				nskb = tmp;
> +				err = skb_unclone(nskb, GFP_ATOMIC);

Could you elaborate why you also need to unclone? This looks odd here. tc layer
(independent of BPF) from ingress & egress side generally assumes unshared skb,
so above clone + dropping ref of nskb looks okay to make the main skb struct private
for mangling attributes (e.g. mark) & should suffice. What is the exact purpose of
the additional skb_unclone() in this context?

> +			} else {
> +				err = -ENOMEM;
> +			}
> +		}
> +
>   		if (!tail)
>   			skb->next = nskb;
>   		else
>   			tail->next = nskb;
>   
> +		if (unlikely(err)) {
> +			nskb->next = list_skb;
> +			goto err_linearize;
> +		}
> +
>   		tail = nskb;
>   
>   		delta_len += nskb->len;