Re: [PATCH net-next 8/9] net: tcp: add skb drop reasons to tcp tw code path

From: Menglong Dong
Date: Mon May 16 2022 - 00:55:17 EST


On Mon, May 16, 2022 at 12:22 PM Eric Dumazet <edumazet@xxxxxxxxxx> wrote:
>
> On Sun, May 15, 2022 at 8:46 PM <menglong8.dong@xxxxxxxxx> wrote:
> >
> > From: Menglong Dong <imagedong@xxxxxxxxxxx>
> >
> > In order to get the reasons of skb drops, add a function argument of
> > type 'enum skb_drop_reason *reason' to tcp_timewait_state_process().
> >
> > In the origin code, all packets to time-wait socket are treated as
> > dropping with kfree_skb(), which can make users confused. Therefore,
> > we use consume_skb() for the skbs that are 'good'. We can check the
> > value of 'reason' to decide use kfree_skb() or consume_skb().
> >
> > The new reason 'TIMEWAIT' is added for the case that the skb is dropped
> > as the socket in time-wait state.
> >
> > Signed-off-by: Menglong Dong <imagedong@xxxxxxxxxxx>
> > ---
> > include/linux/skbuff.h | 5 +++++
> > include/net/tcp.h | 7 ++++---
> > net/ipv4/tcp_ipv4.c | 11 +++++++++--
> > net/ipv4/tcp_minisocks.c | 24 ++++++++++++++++++++----
> > net/ipv6/tcp_ipv6.c | 10 ++++++++--
> > 5 files changed, 46 insertions(+), 11 deletions(-)
> >
> > diff --git a/include/linux/skbuff.h b/include/linux/skbuff.h
> > index 4578bbab5a3e..8d18fc5a5af6 100644
> > --- a/include/linux/skbuff.h
> > +++ b/include/linux/skbuff.h
> > @@ -560,6 +560,10 @@ struct sk_buff;
> > * SKB_DROP_REASON_TCP_REQQFULLDROP
> > * request queue of the listen socket is full, corresponding to
> > * LINUX_MIB_TCPREQQFULLDROP
> > + *
> > + * SKB_DROP_REASON_TIMEWAIT
> > + * socket is in time-wait state and all packet that received will
> > + * be treated as 'drop', except a good 'SYN' packet
> > */
> > #define __DEFINE_SKB_DROP_REASON(FN) \
> > FN(NOT_SPECIFIED) \
> > @@ -631,6 +635,7 @@ struct sk_buff;
> > FN(TCP_ABORTONDATA) \
> > FN(LISTENOVERFLOWS) \
> > FN(TCP_REQQFULLDROP) \
> > + FN(TIMEWAIT) \
> > FN(MAX)
> >
> > /* The reason of skb drop, which is used in kfree_skb_reason().
> > diff --git a/include/net/tcp.h b/include/net/tcp.h
> > index 082dd0627e2e..88217b8d95ac 100644
> > --- a/include/net/tcp.h
> > +++ b/include/net/tcp.h
> > @@ -380,9 +380,10 @@ enum tcp_tw_status {
> > };
> >
> >
> > -enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
> > - struct sk_buff *skb,
> > - const struct tcphdr *th);
> > +enum tcp_tw_status
> > +tcp_timewait_state_process(struct inet_timewait_sock *tw, struct sk_buff *skb,
> > + const struct tcphdr *th,
> > + enum skb_drop_reason *reason);
> > struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
> > struct request_sock *req, bool fastopen,
> > bool *lost_race);
> > diff --git a/net/ipv4/tcp_ipv4.c b/net/ipv4/tcp_ipv4.c
> > index 708f92b03f42..9174ee162633 100644
> > --- a/net/ipv4/tcp_ipv4.c
> > +++ b/net/ipv4/tcp_ipv4.c
> > @@ -2134,7 +2134,8 @@ int tcp_v4_rcv(struct sk_buff *skb)
> > inet_twsk_put(inet_twsk(sk));
> > goto csum_error;
> > }
> > - switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
> > + switch (tcp_timewait_state_process(inet_twsk(sk), skb, th,
> > + &drop_reason)) {
> > case TCP_TW_SYN: {
> > struct sock *sk2 = inet_lookup_listener(dev_net(skb->dev),
> > &tcp_hashinfo, skb,
> > @@ -2150,12 +2151,18 @@ int tcp_v4_rcv(struct sk_buff *skb)
> > refcounted = false;
> > goto process;
> > }
> > + /* TCP_FLAGS or NO_SOCKET? */
> > + SKB_DR_SET(drop_reason, TCP_FLAGS);
> > }
> > /* to ACK */
> > fallthrough;
> > case TCP_TW_ACK:
> > tcp_v4_timewait_ack(sk, skb);
> > - break;
> > + refcounted = false;
> > + if (drop_reason)
> > + goto discard_it;
> > + else
> > + goto put_and_return;
> > case TCP_TW_RST:
> > tcp_v4_send_reset(sk, skb);
> > inet_twsk_deschedule_put(inet_twsk(sk));
> > diff --git a/net/ipv4/tcp_minisocks.c b/net/ipv4/tcp_minisocks.c
> > index 1a21018f6f64..329724118b7f 100644
> > --- a/net/ipv4/tcp_minisocks.c
> > +++ b/net/ipv4/tcp_minisocks.c
> > @@ -83,13 +83,15 @@ tcp_timewait_check_oow_rate_limit(struct inet_timewait_sock *tw,
> > */
> > enum tcp_tw_status
> > tcp_timewait_state_process(struct inet_timewait_sock *tw, struct sk_buff *skb,
> > - const struct tcphdr *th)
> > + const struct tcphdr *th,
> > + enum skb_drop_reason *reason)
> > {
> > struct tcp_options_received tmp_opt;
> > struct tcp_timewait_sock *tcptw = tcp_twsk((struct sock *)tw);
> > bool paws_reject = false;
> >
> > tmp_opt.saw_tstamp = 0;
> > + *reason = SKB_DROP_REASON_NOT_SPECIFIED;
> > if (th->doff > (sizeof(*th) >> 2) && tcptw->tw_ts_recent_stamp) {
> > tcp_parse_options(twsk_net(tw), skb, &tmp_opt, 0, NULL);
> >
> > @@ -113,11 +115,16 @@ tcp_timewait_state_process(struct inet_timewait_sock *tw, struct sk_buff *skb,
> > return tcp_timewait_check_oow_rate_limit(
> > tw, skb, LINUX_MIB_TCPACKSKIPPEDFINWAIT2);
> >
> > - if (th->rst)
> > + if (th->rst) {
> > + SKB_DR_SET(*reason, TCP_RESET);
> > goto kill;
> > + }
> >
> > - if (th->syn && !before(TCP_SKB_CB(skb)->seq, tcptw->tw_rcv_nxt))
> > + if (th->syn && !before(TCP_SKB_CB(skb)->seq,
> > + tcptw->tw_rcv_nxt)) {
> > + SKB_DR_SET(*reason, TCP_FLAGS);
> > return TCP_TW_RST;
> > + }
> >
> > /* Dup ACK? */
> > if (!th->ack ||
> > @@ -143,6 +150,9 @@ tcp_timewait_state_process(struct inet_timewait_sock *tw, struct sk_buff *skb,
> > }
> >
> > inet_twsk_reschedule(tw, TCP_TIMEWAIT_LEN);
> > +
> > + /* skb should be free normally on this case. */
> > + *reason = SKB_NOT_DROPPED_YET;
> > return TCP_TW_ACK;
> > }
> >
> > @@ -174,6 +184,7 @@ tcp_timewait_state_process(struct inet_timewait_sock *tw, struct sk_buff *skb,
> > * protocol bug yet.
> > */
> > if (twsk_net(tw)->ipv4.sysctl_tcp_rfc1337 == 0) {
> > + SKB_DR_SET(*reason, TCP_RESET);
> > kill:
> > inet_twsk_deschedule_put(tw);
> > return TCP_TW_SUCCESS;
> > @@ -216,11 +227,14 @@ tcp_timewait_state_process(struct inet_timewait_sock *tw, struct sk_buff *skb,
> > if (isn == 0)
> > isn++;
> > TCP_SKB_CB(skb)->tcp_tw_isn = isn;
> > + *reason = SKB_NOT_DROPPED_YET;
> > return TCP_TW_SYN;
> > }
> >
> > - if (paws_reject)
> > + if (paws_reject) {
> > + SKB_DR_SET(*reason, TCP_RFC7323_PAWS);
> > __NET_INC_STATS(twsk_net(tw), LINUX_MIB_PAWSESTABREJECTED);
> > + }
> >
> > if (!th->rst) {
> > /* In this case we must reset the TIMEWAIT timer.
> > @@ -232,9 +246,11 @@ tcp_timewait_state_process(struct inet_timewait_sock *tw, struct sk_buff *skb,
> > if (paws_reject || th->ack)
> > inet_twsk_reschedule(tw, TCP_TIMEWAIT_LEN);
> >
> > + SKB_DR_OR(*reason, TIMEWAIT);
> > return tcp_timewait_check_oow_rate_limit(
> > tw, skb, LINUX_MIB_TCPACKSKIPPEDTIMEWAIT);
> > }
> > + SKB_DR_SET(*reason, TCP_RESET);
> > inet_twsk_put(tw);
> > return TCP_TW_SUCCESS;
> > }
> > diff --git a/net/ipv6/tcp_ipv6.c b/net/ipv6/tcp_ipv6.c
> > index 27c51991bd54..5c777006de3d 100644
> > --- a/net/ipv6/tcp_ipv6.c
> > +++ b/net/ipv6/tcp_ipv6.c
> > @@ -1795,7 +1795,8 @@ INDIRECT_CALLABLE_SCOPE int tcp_v6_rcv(struct sk_buff *skb)
> > goto csum_error;
> > }
> >
> > - switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
> > + switch (tcp_timewait_state_process(inet_twsk(sk), skb, th,
> > + &drop_reason)) {
> > case TCP_TW_SYN:
> > {
> > struct sock *sk2;
> > @@ -1815,12 +1816,17 @@ INDIRECT_CALLABLE_SCOPE int tcp_v6_rcv(struct sk_buff *skb)
> > refcounted = false;
> > goto process;
> > }
> > + SKB_DR_SET(drop_reason, TCP_FLAGS);
> > }
> > /* to ACK */
> > fallthrough;
> > case TCP_TW_ACK:
> > tcp_v6_timewait_ack(sk, skb);
> > - break;
> > + refcounted = false;
> > + if (drop_reason)
> > + goto discard_it;
> > + else
> > + goto put_and_return;
>
> My brain exploded.
> I guarantee you that whoever is going to look at this code in one year
> will be completely lost.
> Adding so much complexity is crazy.
>

Yeah, I'm almost lost while adding this code.

> About: refcounted = false ;
>
> This is not going to be used id "goto discard_it;" is taken.
>
> If the "goto put_and_return;" is taken, this boils down to:
>
> return ret ? -1 : 0;
>
> Are you sure ret is even initialized at this point ?
>

About this part, it seems it's my mistake. The correct code
should be:

case TCP_TW_ACK:
tcp_v6_timewait_ack(sk, skb);
- break;
+ if (drop_reason) {
+ goto discard_it;
+ } else {
+ consume_skb(skb); /* free skb here */
+ return 0;
+ }

As you said, it's a little complex.

> Why are we doing this ? We were doing "return 0;"
>
> Also, where is skb freed ?
>
> Are you calling tcp_v6_timewait_ack(sk, skb) after skb has been freed ???
>
>



>
>
> > case TCP_TW_RST:
> > tcp_v6_send_reset(sk, skb);
> > inet_twsk_deschedule_put(inet_twsk(sk));
> > --
> > 2.36.1
> >