Re: [PATCH v3 net-next] packet: add a generic drop reason for receive

[Yan Zhai]

On Sat, Dec 2, 2023 at 8:07 AM Willem de Bruijn
<willemdebruijn.kernel@xxxxxxxxx> wrote:
>
> Yan Zhai 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>
> > ---
> > v2->v3: removed an unused variable
> > v1->v2: fixups suggested by Eric Dumazet
> > v2: https://lore.kernel.org/netdev/ZWobMUp22oTpP3FW@debian.debian/
> > v1: https://lore.kernel.org/netdev/ZU3EZKQ3dyLE6T8z@debian.debian/
> > ---
> >  include/net/dropreason-core.h |  6 ++++++
> >  net/packet/af_packet.c        | 26 +++++++++++++-------------
> >  2 files changed, 19 insertions(+), 13 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..933fdfaacc44 100644
> > --- a/net/packet/af_packet.c
> > +++ b/net/packet/af_packet.c
> > @@ -2127,7 +2127,7 @@ static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
> >       u8 *skb_head = skb->data;
> >       int skb_len = skb->len;
> >       unsigned int snaplen, res;
> > -     bool is_drop_n_account = false;
> > +     enum skb_drop_reason drop_reason = SKB_CONSUMED;
>
> Reverse xmas tree
>
oh I didn't know we have requirements on variable ordering. Will pay
attention in future.

> >
> >       if (skb->pkt_type == PACKET_LOOPBACK)
> >               goto drop;
> > @@ -2161,6 +2161,10 @@ 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;
> > +
>
> This can be set in drop_n_account, rather than the common path.
>
> Same in tpacket_rcv.

Sure, let me shoot a v4 to move it.