Re: [PATCH v2 net-next] packet: add a generic drop reason for receive

[Eric Dumazet]

On Fri, Dec 1, 2023 at 6:43 PM Yan Zhai <yan@xxxxxxxxxxxxxx> wrote:
>
> Commit da37845fdce2 ("packet: uses kfree_skb() for errors.") switches
> from consume_skb to kfree_skb to improve error handling. However, this
> could bring a lot of noises when we monitor real packet drops in
> kfree_skb[1], because in tpacket_rcv or packet_rcv only packet clones
> can be freed, not actual packets.
>
> Adding a generic drop reason to allow distinguish these "clone drops".
>
> [1]: https://lore.kernel.org/netdev/CABWYdi00L+O30Q=Zah28QwZ_5RU-xcxLFUK2Zj08A8MrLk9jzg@xxxxxxxxxxxxxx/
> Fixes: da37845fdce2 ("packet: uses kfree_skb() for errors.")
> Suggested-by: Eric Dumazet <edumazet@xxxxxxxxxx>
> Signed-off-by: Yan Zhai <yan@xxxxxxxxxxxxxx>
> ---
>  include/net/dropreason-core.h |  6 ++++++
>  net/packet/af_packet.c        | 22 +++++++++++++---------
>  2 files changed, 19 insertions(+), 9 deletions(-)
>
> diff --git a/include/net/dropreason-core.h b/include/net/dropreason-core.h
> index 3c70ad53a49c..278e4c7d465c 100644
> --- a/include/net/dropreason-core.h
> +++ b/include/net/dropreason-core.h
> @@ -86,6 +86,7 @@
>         FN(IPV6_NDISC_NS_OTHERHOST)     \
>         FN(QUEUE_PURGE)                 \
>         FN(TC_ERROR)                    \
> +       FN(PACKET_SOCK_ERROR)           \
>         FNe(MAX)
>
>  /**
> @@ -378,6 +379,11 @@ enum skb_drop_reason {
>         SKB_DROP_REASON_QUEUE_PURGE,
>         /** @SKB_DROP_REASON_TC_ERROR: generic internal tc error. */
>         SKB_DROP_REASON_TC_ERROR,
> +       /**
> +        * @SKB_DROP_REASON_PACKET_SOCK_ERROR: generic packet socket errors
> +        * after its filter matches an incoming packet.
> +        */
> +       SKB_DROP_REASON_PACKET_SOCK_ERROR,
>         /**
>          * @SKB_DROP_REASON_MAX: the maximum of core drop reasons, which
>          * shouldn't be used as a real 'reason' - only for tracing code gen
> diff --git a/net/packet/af_packet.c b/net/packet/af_packet.c
> index a84e00b5904b..0a7c05d8fe9f 100644
> --- a/net/packet/af_packet.c
> +++ b/net/packet/af_packet.c
> @@ -2128,6 +2128,7 @@ static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
>         int skb_len = skb->len;
>         unsigned int snaplen, res;
>         bool is_drop_n_account = false;

Why keeping is_drop_n_account  then ?


> +       enum skb_drop_reason drop_reason = SKB_CONSUMED;
>
>         if (skb->pkt_type == PACKET_LOOPBACK)
>                 goto drop;
> @@ -2161,6 +2162,10 @@ static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
>         res = run_filter(skb, sk, snaplen);
>         if (!res)
>                 goto drop_n_restore;
> +
> +       /* skb will only be "consumed" not "dropped" before this */
> +       drop_reason = SKB_DROP_REASON_PACKET_SOCK_ERROR;
> +
>         if (snaplen > res)
>                 snaplen = res;
>
> @@ -2227,10 +2232,7 @@ static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
>                 skb->len = skb_len;
>         }
>  drop:
> -       if (!is_drop_n_account)

Because after your patch it will be set but never read.

> -               consume_skb(skb);
> -       else
> -               kfree_skb(skb);
> +       kfree_skb_reason(skb, drop_reason);
>         return 0;
>  }
>
> @@ -2253,6 +2255,7 @@ static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
>         bool is_drop_n_account = false;

Same remark here.

>         unsigned int slot_id = 0;
>         int vnet_hdr_sz = 0;
> +       enum skb_drop_reason drop_reason = SKB_CONSUMED;
>
>         /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
>          * We may add members to them until current aligned size without forcing
> @@ -2355,6 +2358,10 @@ static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
>                         vnet_hdr_sz = 0;
>                 }
>         }
> +
> +       /* skb will only be "consumed" not "dropped" before this */
> +       drop_reason = SKB_DROP_REASON_PACKET_SOCK_ERROR;
> +
>         spin_lock(&sk->sk_receive_queue.lock);
>         h.raw = packet_current_rx_frame(po, skb,
>                                         TP_STATUS_KERNEL, (macoff+snaplen));
> @@ -2498,10 +2505,7 @@ static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
>                 skb->len = skb_len;
>         }
>  drop:
> -       if (!is_drop_n_account)
> -               consume_skb(skb);
> -       else
> -               kfree_skb(skb);
> +       kfree_skb_reason(skb, drop_reason);
>         return 0;
>
>  drop_n_account:
> @@ -2510,7 +2514,7 @@ static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
>         is_drop_n_account = true;
>
>         sk->sk_data_ready(sk);
> -       kfree_skb(copy_skb);
> +       kfree_skb_reason(copy_skb, drop_reason);
>         goto drop_n_restore;
>  }
>
> --
> 2.30.2
>