| Message ID | 20220215112812.2093852-2-imagedong@tencent.com (mailing list archive) |
|---|---|
| State | Not Applicable |
| Delegated to: | Netdev Maintainers |
| Headers | show |
| Series | net: add skb drop reasons for TCP, IP, dev and neigh | expand |
| Context | Check | Description |
|---|---|---|
| netdev/tree_selection | success | Clearly marked for net-next |
| netdev/apply | fail | Patch does not apply to net-next |
On Tue, Feb 15, 2022 at 3:30 AM <menglong8.dong@gmail.com> wrote: > > From: Menglong Dong <imagedong@tencent.com> > > For TCP protocol, tcp_drop() is used to free the skb when it needs > to be dropped. To make use of kfree_skb_reason() and collect drop > reasons, introduce the function tcp_drop_reason(). > > tcp_drop_reason() will finally call kfree_skb_reason() and pass the > drop reason to 'kfree_skb' tracepoint. > > PS: __kfree_skb() was used in tcp_drop(), I'm not sure if it's ok > to replace it with kfree_skb_reason(). > > Signed-off-by: Menglong Dong <imagedong@tencent.com> > --- > net/ipv4/tcp_input.c | 13 +++++++++++-- > 1 file changed, 11 insertions(+), 2 deletions(-) > > diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c > index af94a6d22a9d..e3811afd1756 100644 > --- a/net/ipv4/tcp_input.c > +++ b/net/ipv4/tcp_input.c > @@ -4684,10 +4684,19 @@ static bool tcp_ooo_try_coalesce(struct sock *sk, > return res; > } > > -static void tcp_drop(struct sock *sk, struct sk_buff *skb) > +static void tcp_drop_reason(struct sock *sk, struct sk_buff *skb, > + enum skb_drop_reason reason) > { > sk_drops_add(sk, skb); > - __kfree_skb(skb); > + /* why __kfree_skb() used here before, other than kfree_skb()? > + * confusing...... Do not add comments like that if you do not know the difference... __kfree_skb() is used by TCP stack because it owns skb in receive queues, and avoids touching skb->users because it must be one already. (We made sure not using skb_get() in TCP) It seems fine to use kfree_skb() in tcp_drop(), it is hardly fast path, and the added cost is pure noise.
On 2/15/22 10:34 AM, Eric Dumazet wrote: >> diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c >> index af94a6d22a9d..e3811afd1756 100644 >> --- a/net/ipv4/tcp_input.c >> +++ b/net/ipv4/tcp_input.c >> @@ -4684,10 +4684,19 @@ static bool tcp_ooo_try_coalesce(struct sock *sk, >> return res; >> } >> >> -static void tcp_drop(struct sock *sk, struct sk_buff *skb) >> +static void tcp_drop_reason(struct sock *sk, struct sk_buff *skb, >> + enum skb_drop_reason reason) >> { >> sk_drops_add(sk, skb); >> - __kfree_skb(skb); >> + /* why __kfree_skb() used here before, other than kfree_skb()? >> + * confusing...... > > Do not add comments like that if you do not know the difference... > > __kfree_skb() is used by TCP stack because it owns skb in receive > queues, and avoids touching skb->users > because it must be one already. and it bypasses kfree_skb tracepoint which seems by design.
On Wed, Feb 16, 2022 at 1:34 AM Eric Dumazet <edumazet@google.com> wrote: > > On Tue, Feb 15, 2022 at 3:30 AM <menglong8.dong@gmail.com> wrote: > > > > From: Menglong Dong <imagedong@tencent.com> > > > > For TCP protocol, tcp_drop() is used to free the skb when it needs > > to be dropped. To make use of kfree_skb_reason() and collect drop > > reasons, introduce the function tcp_drop_reason(). > > > > tcp_drop_reason() will finally call kfree_skb_reason() and pass the > > drop reason to 'kfree_skb' tracepoint. > > > > PS: __kfree_skb() was used in tcp_drop(), I'm not sure if it's ok > > to replace it with kfree_skb_reason(). > > > > Signed-off-by: Menglong Dong <imagedong@tencent.com> > > --- > > net/ipv4/tcp_input.c | 13 +++++++++++-- > > 1 file changed, 11 insertions(+), 2 deletions(-) > > > > diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c > > index af94a6d22a9d..e3811afd1756 100644 > > --- a/net/ipv4/tcp_input.c > > +++ b/net/ipv4/tcp_input.c > > @@ -4684,10 +4684,19 @@ static bool tcp_ooo_try_coalesce(struct sock *sk, > > return res; > > } > > > > -static void tcp_drop(struct sock *sk, struct sk_buff *skb) > > +static void tcp_drop_reason(struct sock *sk, struct sk_buff *skb, > > + enum skb_drop_reason reason) > > { > > sk_drops_add(sk, skb); > > - __kfree_skb(skb); > > + /* why __kfree_skb() used here before, other than kfree_skb()? > > + * confusing...... > > Do not add comments like that if you do not know the difference... > > __kfree_skb() is used by TCP stack because it owns skb in receive > queues, and avoids touching skb->users > because it must be one already. > > (We made sure not using skb_get() in TCP) > > It seems fine to use kfree_skb() in tcp_drop(), it is hardly fast > path, and the added cost is pure noise. I understand why __kfree_skb() was used now, and it seems this commit is ok (with the comments removed of course). I'll keep it still. Thanks! Menglong Dong
On Wed, Feb 16, 2022 at 2:47 AM David Ahern <dsahern@kernel.org> wrote: > > On 2/15/22 10:34 AM, Eric Dumazet wrote: > >> diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c > >> index af94a6d22a9d..e3811afd1756 100644 > >> --- a/net/ipv4/tcp_input.c > >> +++ b/net/ipv4/tcp_input.c > >> @@ -4684,10 +4684,19 @@ static bool tcp_ooo_try_coalesce(struct sock *sk, > >> return res; > >> } > >> > >> -static void tcp_drop(struct sock *sk, struct sk_buff *skb) > >> +static void tcp_drop_reason(struct sock *sk, struct sk_buff *skb, > >> + enum skb_drop_reason reason) > >> { > >> sk_drops_add(sk, skb); > >> - __kfree_skb(skb); > >> + /* why __kfree_skb() used here before, other than kfree_skb()? > >> + * confusing...... > > > > Do not add comments like that if you do not know the difference... > > > > __kfree_skb() is used by TCP stack because it owns skb in receive > > queues, and avoids touching skb->users > > because it must be one already. > > and it bypasses kfree_skb tracepoint which seems by design. Do you mean it shouldn't be traced here? According to my understanding, __kfree_skb() was used in the beginning as skb->users aren't touched by TCP. Later, tcp_drop() was introduced to record drop count to the socket. Considering the skb is indeed dropped and no other event is triggered, is it ok to trigger the kfree_skb tracepoint? Thanks! Menglong Dong
diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c index af94a6d22a9d..e3811afd1756 100644 --- a/net/ipv4/tcp_input.c +++ b/net/ipv4/tcp_input.c @@ -4684,10 +4684,19 @@ static bool tcp_ooo_try_coalesce(struct sock *sk, return res; } -static void tcp_drop(struct sock *sk, struct sk_buff *skb) +static void tcp_drop_reason(struct sock *sk, struct sk_buff *skb, + enum skb_drop_reason reason) { sk_drops_add(sk, skb); - __kfree_skb(skb); + /* why __kfree_skb() used here before, other than kfree_skb()? + * confusing...... + */ + kfree_skb_reason(skb, reason); +} + +static inline void tcp_drop(struct sock *sk, struct sk_buff *skb) +{ + tcp_drop_reason(sk, skb, SKB_DROP_REASON_NOT_SPECIFIED); } /* This one checks to see if we can put data from the