From: Eric Dumazet <edumazet@google.com>
To: Eric Dumazet <edumazet@google.com>,
"David S. Miller" <davem@davemloft.net>,
David Ahern <dsahern@kernel.org>,
Jakub Kicinski <kuba@kernel.org>, Paolo Abeni <pabeni@redhat.com>,
Neal Cardwell <ncardwell@google.com>,
netdev@vger.kernel.org, Jason Xing <kerneljasonxing@gmail.com>,
zhangweiping <zhangweiping@didiglobal.com>,
tiozhang <tiozhang@didiglobal.com>,
linux-kernel@vger.kernel.org, bpf@vger.kernel.org
Subject: Re: [PATCH net] tcp: introduce a compack timer handler in sack compression
Date: Tue, 30 May 2023 09:12:41 +0200 [thread overview]
Message-ID: <CANn89iJw3Ehoj7GfYnc6Xv5N2wULqNuP3zNBwQx97i-YJD5avg@mail.gmail.com> (raw)
In-Reply-To: <20230529113804.GA20300@didi-ThinkCentre-M920t-N000>
On Mon, May 29, 2023 at 1:38 PM fuyuanli <fuyuanli@didiglobal.com> wrote:
>
> We've got some issues when sending a compressed ack is deferred to
> release phrase due to the socket owned by another user:
> 1. a compressed ack would not be sent because of lack of ICSK_ACK_TIMER
> flag.
Are you sure ? Just add it then, your patch will be a one-liner
instead of a complex one adding
more code in a fast path.
> 2. the tp->compressed_ack counter should be decremented by 1.
> 3. we cannot pass timeout check and reset the delack timer in
> tcp_delack_timer_handler().
> 4. we are not supposed to increment the LINUX_MIB_DELAYEDACKS counter.
> ...
>
> The reason why it could happen is that we previously reuse the delayed
> ack logic when handling the sack compression. With this patch applied,
> the sack compression logic would go into the same function
> (tcp_compack_timer_handler()) whether we defer sending ack or not.
> Therefore, those two issued could be easily solved.
>
> Here are more details in the old logic:
> When sack compression is triggered in the tcp_compressed_ack_kick(),
> if the sock is owned by user, it will set TCP_DELACK_TIMER_DEFERRED and
> then defer to the release cb phrase. Later once user releases the sock,
> tcp_delack_timer_handler() should send a ack as expected, which, however,
> cannot happen due to lack of ICSK_ACK_TIMER flag. Therefore, the receiver
> would not sent an ack until the sender's retransmission timeout. It
> definitely increases unnecessary latency.
>
> This issue happens rarely in the production environment. I used kprobe
> to hook some key functions like tcp_compressed_ack_kick, tcp_release_cb,
> tcp_delack_timer_handler and then found that when tcp_delack_timer_handler
> was called, value of icsk_ack.pending was 1, which means we only had
> flag ICSK_ACK_SCHED set, not including ICSK_ACK_TIMER. It was against
> our expectations.
>
> In conclusion, we chose to separate the sack compression from delayed
> ack logic to solve issues only happening when the process is deferred.
>
> Fixes: 5d9f4262b7ea ("tcp: add SACK compression")
> Signed-off-by: fuyuanli <fuyuanli@didiglobal.com>
> Signed-off-by: Jason Xing <kerneljasonxing@gmail.com>
> ---
> include/linux/tcp.h | 2 ++
> include/net/tcp.h | 1 +
> net/ipv4/tcp_output.c | 4 ++++
> net/ipv4/tcp_timer.c | 28 +++++++++++++++++++---------
> 4 files changed, 26 insertions(+), 9 deletions(-)
>
> diff --git a/include/linux/tcp.h b/include/linux/tcp.h
> index b4c08ac86983..cd15a9972c48 100644
> --- a/include/linux/tcp.h
> +++ b/include/linux/tcp.h
> @@ -461,6 +461,7 @@ enum tsq_enum {
> TCP_MTU_REDUCED_DEFERRED, /* tcp_v{4|6}_err() could not call
> * tcp_v{4|6}_mtu_reduced()
> */
> + TCP_COMPACK_TIMER_DEFERRED, /* tcp_compressed_ack_kick() found socket was owned */
> };
>
> enum tsq_flags {
> @@ -470,6 +471,7 @@ enum tsq_flags {
> TCPF_WRITE_TIMER_DEFERRED = (1UL << TCP_WRITE_TIMER_DEFERRED),
> TCPF_DELACK_TIMER_DEFERRED = (1UL << TCP_DELACK_TIMER_DEFERRED),
> TCPF_MTU_REDUCED_DEFERRED = (1UL << TCP_MTU_REDUCED_DEFERRED),
> + TCPF_COMPACK_TIMER_DEFERRED = (1UL << TCP_DELACK_TIMER_DEFERRED),
> };
>
> #define tcp_sk(ptr) container_of_const(ptr, struct tcp_sock, inet_conn.icsk_inet.sk)
> diff --git a/include/net/tcp.h b/include/net/tcp.h
> index 18a038d16434..e310d7bf400c 100644
> --- a/include/net/tcp.h
> +++ b/include/net/tcp.h
> @@ -342,6 +342,7 @@ void tcp_release_cb(struct sock *sk);
> void tcp_wfree(struct sk_buff *skb);
> void tcp_write_timer_handler(struct sock *sk);
> void tcp_delack_timer_handler(struct sock *sk);
> +void tcp_compack_timer_handler(struct sock *sk);
> int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg);
> int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb);
> void tcp_rcv_established(struct sock *sk, struct sk_buff *skb);
> diff --git a/net/ipv4/tcp_output.c b/net/ipv4/tcp_output.c
> index cfe128b81a01..1703caab6632 100644
> --- a/net/ipv4/tcp_output.c
> +++ b/net/ipv4/tcp_output.c
> @@ -1110,6 +1110,10 @@ void tcp_release_cb(struct sock *sk)
> tcp_delack_timer_handler(sk);
> __sock_put(sk);
> }
> + if (flags & TCPF_COMPACK_TIMER_DEFERRED) {
> + tcp_compack_timer_handler(sk);
> + __sock_put(sk);
> + }
Please do not add another test in the fast path.
Just make sure tcp_delack_timer_handler() handles the case (this
certainly was my intent)
> if (flags & TCPF_MTU_REDUCED_DEFERRED) {
> inet_csk(sk)->icsk_af_ops->mtu_reduced(sk);
> __sock_put(sk);
> diff --git a/net/ipv4/tcp_timer.c b/net/ipv4/tcp_timer.c
> index b839c2f91292..069f6442069b 100644
> --- a/net/ipv4/tcp_timer.c
> +++ b/net/ipv4/tcp_timer.c
> @@ -318,6 +318,23 @@ void tcp_delack_timer_handler(struct sock *sk)
> }
> }
>
> +/* Called with BH disabled */
> +void tcp_compack_timer_handler(struct sock *sk)
> +{
> + struct tcp_sock *tp = tcp_sk(sk);
> +
> + if (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN)))
> + return;
> +
> + if (tp->compressed_ack) {
> + /* Since we have to send one ack finally,
> + * subtract one from tp->compressed_ack to keep
> + * LINUX_MIB_TCPACKCOMPRESSED accurate.
> + */
> + tp->compressed_ack--;
> + tcp_send_ack(sk);
> + }
> +}
>
> /**
> * tcp_delack_timer() - The TCP delayed ACK timeout handler
> @@ -757,16 +774,9 @@ static enum hrtimer_restart tcp_compressed_ack_kick(struct hrtimer *timer)
>
> bh_lock_sock(sk);
> if (!sock_owned_by_user(sk)) {
> - if (tp->compressed_ack) {
> - /* Since we have to send one ack finally,
> - * subtract one from tp->compressed_ack to keep
> - * LINUX_MIB_TCPACKCOMPRESSED accurate.
> - */
> - tp->compressed_ack--;
> - tcp_send_ack(sk);
> - }
> + tcp_compack_timer_handler(sk);
> } else {
> - if (!test_and_set_bit(TCP_DELACK_TIMER_DEFERRED,
See, I was clearly intending to let tcp_delack_timer_handler() deal with this.
> + if (!test_and_set_bit(TCP_COMPACK_TIMER_DEFERRED,
> &sk->sk_tsq_flags))
> sock_hold(sk);
> }
> --
> 2.17.1
>
next prev parent reply other threads:[~2023-05-30 7:12 UTC|newest]
Thread overview: 8+ messages / expand[flat|nested] mbox.gz Atom feed top
2023-05-29 11:38 [PATCH net] tcp: introduce a compack timer handler in sack compression fuyuanli
2023-05-30 2:15 ` Jason Xing
2023-05-30 7:12 ` Eric Dumazet [this message]
2023-05-30 7:45 ` Jason Xing
2023-05-30 7:49 ` Jason Xing
2023-05-30 10:00 ` Eric Dumazet
2023-05-30 11:10 ` Jason Xing
2023-05-30 12:52 ` Jason Xing
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=CANn89iJw3Ehoj7GfYnc6Xv5N2wULqNuP3zNBwQx97i-YJD5avg@mail.gmail.com \
--to=edumazet@google.com \
--cc=bpf@vger.kernel.org \
--cc=davem@davemloft.net \
--cc=dsahern@kernel.org \
--cc=kerneljasonxing@gmail.com \
--cc=kuba@kernel.org \
--cc=linux-kernel@vger.kernel.org \
--cc=ncardwell@google.com \
--cc=netdev@vger.kernel.org \
--cc=pabeni@redhat.com \
--cc=tiozhang@didiglobal.com \
--cc=zhangweiping@didiglobal.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).