tcp: make trace of reset logic complete
Subsystems (4)
v2 cover tcp: make trace of reset logic complete
From: Jason Xing <[email protected]>
Before this, we miss some cases where the TCP layer could send rst but we cannot trace it. So I decided to complete it :)
v2
1. fix spelling mistakes
Jason Xing (3):
trace: adjust TP_STORE_ADDR_PORTS_SKB() parameters
trace: tcp: fully support trace_tcp_send_reset
tcp: add location into reset trace process
include/trace/events/tcp.h | 68 ++++++++++++++++++++++++++++++-------- net/ipv4/tcp_ipv4.c | 4 +-- net/ipv4/tcp_output.c | 2 +- net/ipv6/tcp_ipv6.c | 3 +- 4 files changed, 60 insertions(+), 17 deletions(-)
Signature · 1 line
2.37.3
PATCH v2 1/3 trace: adjust TP_STORE_ADDR_PORTS_SKB() parameters
From: Jason Xing <[email protected]>
Introducing entry_saddr and entry_daddr parameters in this macro for later use can help us record the reverse 4-tuple by analyzing the 4-tuple of the incoming skb when receiving.
--- include/trace/events/tcp.h | 21 +++++++++++---------- 1 file changed, 11 insertions(+), 10 deletions(-)
include/trace/events/tcp.h +11 -10
@@ -302,15 +302,15 @@ TRACE_EVENT(tcp_probe, __entry->skbaddr, __entry->skaddr) ); -#define TP_STORE_ADDR_PORTS_SKB_V4 (__entry, skb) \ +#define TP_STORE_ADDR_PORTS_SKB_V4 (skb, entry_saddr, entry_daddr) \ do { \ const struct tcphdr *th = (const struct tcphdr *)skb->data; \ - struct sockaddr_in *v4 = (void *)__entry->saddr; \ + struct sockaddr_in *v4 = (void *)entry_saddr; \ \ v4->sin_family = AF_INET; \ v4->sin_port = th->source; \ v4->sin_addr.s_addr = ip_hdr(skb)->saddr; \ - v4 = (void *)__entry->daddr; \ + v4 = (void *)entry_daddr; \ v4->sin_family = AF_INET; \ v4->sin_port = th->dest; \ v4->sin_addr.s_addr = ip_hdr(skb)->daddr; \
@@ -318,29 +318,30 @@ TRACE_EVENT(tcp_probe,#if IS_ENABLED (CONFIG_IPV6 ) -#define TP_STORE_ADDR_PORTS_SKB (__entry, skb) \ +#define TP_STORE_ADDR_PORTS_SKB (skb, entry_saddr, entry_daddr) \ do { \ const struct iphdr *iph = ip_hdr(skb); \ \ if (iph->version == 6) { \ const struct tcphdr *th = (const struct tcphdr *)skb->data; \ - struct sockaddr_in6 *v6 = (void *)__entry->saddr; \ + struct sockaddr_in6 *v6 = (void *)entry_saddr; \ \ v6->sin6_family = AF_INET6; \ v6->sin6_port = th->source; \ v6->sin6_addr = ipv6_hdr(skb)->saddr; \ - v6 = (void *)__entry->daddr; \ + v6 = (void *)entry_daddr; \ v6->sin6_family = AF_INET6; \ v6->sin6_port = th->dest; \ v6->sin6_addr = ipv6_hdr(skb)->daddr; \ } else \ - TP_STORE_ADDR_PORTS_SKB_V4(__entry, skb); \ + TP_STORE_ADDR_PORTS_SKB_V4(skb, entry_saddr, \ + entry_daddr); \ } while (0)#else -#define TP_STORE_ADDR_PORTS_SKB (__entry, skb) \ - TP_STORE_ADDR_PORTS_SKB_V4(__entry, skb) +#define TP_STORE_ADDR_PORTS_SKB (skb, entry_saddr, entry_daddr) \ + TP_STORE_ADDR_PORTS_SKB_V4(skb, entry_saddr, entry_daddr)#endif
@@ -365,7 +366,7 @@ DECLARE_EVENT_CLASS(tcp_event_skb,memset (__entry->saddr,0 ,sizeof (struct sockaddr_in6 ));memset (__entry->daddr,0 ,sizeof (struct sockaddr_in6 )); -TP_STORE_ADDR_PORTS_SKB (__entry, skb); +TP_STORE_ADDR_PORTS_SKB (skb, __entry->saddr, __entry->daddr); ), TP_printk("skbaddr=%p src=%pISpc dest=%pISpc",
Signature · 1 line
2.37.3
PATCH v2 2/3 trace: tcp: fully support trace_tcp_send_reset
In reply to Jason XingFrom: Jason Xing <[email protected]>
Prior to this patch, what we can see by enabling trace_tcp_send is only happening under two circumstances:
1) active rst mode
2) non-active rst mode and based on the full socket
That means the inconsistency occurs if we use tcpdump and trace simultaneously to see how rst happens.
It's necessary that we should take into other cases into considerations, say:
1) time-wait socket
2) no socket
...
By parsing the incoming skb and reversing its 4-tuple can we know the exact 'flow' which might not exist.
Samples after applied this patch:
1. tcp_send_reset: skbaddr=XXX skaddr=XXX src=ip:port dest=ip:port state=TCP_ESTABLISHED
2. tcp_send_reset: skbaddr=000...000 skaddr=XXX src=ip:port dest=ip:port state=UNKNOWN
Note:
1) UNKNOWN means we cannot extract the right information from skb.
2) skbaddr/skaddr could be 0
--- include/trace/events/tcp.h | 39 ++++++++++++++++++++++++++++++++++++-- net/ipv4/tcp_ipv4.c | 4 ++-- net/ipv6/tcp_ipv6.c | 3 ++- 3 files changed, 41 insertions(+), 5 deletions(-)
include/trace/events/tcp.h +37 -2
@@ -107,11 +107,46 @@ DEFINE_EVENT(tcp_event_sk_skb, tcp_retransmit_skb,* skb of trace_tcp_send_reset is the skb that caused RST. In case of * active reset, skb should be NULL */ -DEFINE_EVENT (tcp_event_sk_skb, tcp_send_reset, +TRACE_EVENT (tcp_send_reset, TP_PROTO(const struct sock *sk, const struct sk_buff *skb), -TP_ARGS (sk, skb) +TP_ARGS (sk, skb), + +TP_STRUCT__entry ( + __field(const void *, skbaddr) + __field(const void *, skaddr) + __field(int, state) + __array(__u8, saddr, sizeof(struct sockaddr_in6)) +__array (__u8, daddr,sizeof (struct sockaddr_in6 )) + ), + +TP_fast_assign ( + __entry->skbaddr = skb; + __entry->skaddr = sk; +/* Zero means unknown state. */ + __entry->state = sk ? sk->sk_state : 0; + + memset(__entry->saddr, 0, sizeof(struct sockaddr_in6)); + memset(__entry->daddr, 0, sizeof(struct sockaddr_in6)); + + if (sk && sk_fullsock(sk)) { + const struct inet_sock *inet = inet_sk(sk); + + TP_STORE_ADDR_PORTS(__entry, inet, sk); + } else { +/* +* We should reverse the 4-tuple of skb, so later +* it can print the right flow direction of rst. +*/ +TP_STORE_ADDR_PORTS_SKB (skb, entry->daddr, entry->saddr); + } + ), + +TP_printk ("skbaddr=%p skaddr=%p src=%pISpc dest=%pISpc state=%s" , + __entry->skbaddr, __entry->skaddr, + __entry->saddr, __entry->daddr, + __entry->state ?show_tcp_state_name (__entry->state) :"UNKNOWN" ) );/*
net/ipv4/tcp_ipv4.c +2 -2
@@ -868,10 +868,10 @@ static void tcp_v4_send_reset(const struct sock *sk, struct sk_buff *skb)*/ if (sk) { arg.bound_dev_if = sk->sk_bound_dev_if; -if (sk_fullsock (sk)) -trace_tcp_send_reset (sk, skb); } +trace_tcp_send_reset (sk, skb); +BUILD_BUG_ON (offsetof(struct sock , sk_bound_dev_if) != offsetof(struct inet_timewait_sock , tw_bound_dev_if));
net/ipv6/tcp_ipv6.c +2 -1
@@ -1113,7 +1113,6 @@ static void tcp_v6_send_reset(const struct sock *sk, struct sk_buff *skb)if (sk) { oif = sk->sk_bound_dev_if;if (sk_fullsock (sk)) { -trace_tcp_send_reset (sk, skb);if (inet6_test_bit (REPFLOW , sk)) label =ip6_flowlabel (ipv6h); priority =READ_ONCE (sk->sk_priority);
@@ -1129,6 +1128,8 @@ static void tcp_v6_send_reset(const struct sock *sk, struct sk_buff *skb) label =ip6_flowlabel (ipv6h); } +trace_tcp_send_reset (sk, skb); +tcp_v6_send_response (sk, skb, seq, ack_seq,0 ,0 ,0 , oif,1 ,ipv6_get_dsfield (ipv6h), label, priority, txhash, &key);
Signature · 1 line
2.37.3
PATCH v2 3/3 tcp: add location into reset trace process
In reply to Jason XingFrom: Jason Xing <[email protected]>
In addition to knowing the 4-tuple of the flow which generates RST, the reason why it does so is very important because we have some cases where the RST should be sent and have no clue which one exactly.
Adding location of reset process can help us more, like what trace_kfree_skb does.
--- include/trace/events/tcp.h | 14 ++++++++++---- net/ipv4/tcp_ipv4.c | 2 +- net/ipv4/tcp_output.c | 2 +- net/ipv6/tcp_ipv6.c | 2 +- 4 files changed, 13 insertions(+), 7 deletions(-)
include/trace/events/tcp.h +10 -4
@@ -109,13 +109,17 @@ DEFINE_EVENT(tcp_event_sk_skb, tcp_retransmit_skb,*/ TRACE_EVENT(tcp_send_reset, - TP_PROTO(const struct sock *sk, const struct sk_buff *skb), + TP_PROTO( + const struct sock *sk, + const struct sk_buff *skb, + void *location), - TP_ARGS(sk, skb), + TP_ARGS(sk, skb, location), TP_STRUCT__entry( __field(const void *, skbaddr) __field(const void *, skaddr) + __field(void *, location) __field(int, state) __array(__u8, saddr, sizeof(struct sockaddr_in6)) __array(__u8, daddr, sizeof(struct sockaddr_in6))
@@ -141,12 +145,14 @@ TRACE_EVENT(tcp_send_reset,*/ TP_STORE_ADDR_PORTS_SKB (skb, entry->daddr, entry->saddr); } + __entry->location = location; ), -TP_printk ("skbaddr=%p skaddr=%p src=%pISpc dest=%pISpc state=%s" , +TP_printk ("skbaddr=%p skaddr=%p src=%pISpc dest=%pISpc state=%s location=%pS" , __entry->skbaddr, __entry->skaddr, __entry->saddr, __entry->daddr, - __entry->state ?show_tcp_state_name (__entry->state) :"UNKNOWN" ) + __entry->state ?show_tcp_state_name (__entry->state) :"UNKNOWN" , + __entry->location) );/*
net/ipv4/tcp_ipv4.c +1 -1
@@ -870,7 +870,7 @@ static void tcp_v4_send_reset(const struct sock *sk, struct sk_buff *skb) arg.bound_dev_if = sk->sk_bound_dev_if; } -trace_tcp_send_reset (sk, skb); +trace_tcp_send_reset (sk, skb,__builtin_return_address (0 ));BUILD_BUG_ON (offsetof(struct sock , sk_bound_dev_if) != offsetof(struct inet_timewait_sock , tw_bound_dev_if));
net/ipv4/tcp_output.c +1 -1
@@ -3608,7 +3608,7 @@ void tcp_send_active_reset(struct sock *sk, gfp_t priority)/* skb of trace_tcp_send_reset() keeps the skb that caused RST, * skb here is different to the troublesome skb, so use NULL */ -trace_tcp_send_reset (sk,NULL ); +trace_tcp_send_reset (sk,NULL ,__builtin_return_address (0 )); }/* Send a crossed SYN-ACK during socket establishment.
net/ipv6/tcp_ipv6.c +1 -1
@@ -1128,7 +1128,7 @@ static void tcp_v6_send_reset(const struct sock *sk, struct sk_buff *skb) label =ip6_flowlabel (ipv6h); } -trace_tcp_send_reset (sk, skb); +trace_tcp_send_reset (sk, skb,__builtin_return_address (0 )); tcp_v6_send_response(sk, skb, seq, ack_seq, 0, 0, 0, oif, 1, ipv6_get_dsfield(ipv6h), label, priority, txhash,
Signature · 1 line
2.37.3
Jakub Kicinski
maintainerQuoted 5 lines · On Mon, 25 Mar 2024 14:28:28 +0800 Jason Xing wrote:
Before this, we miss some cases where the TCP layer could send rst but we cannot trace it. So I decided to complete it :)
v2
1. fix spelling mistakes
Not only do you post it before we "officially" open net-next but also ignoring the 24h wait period.
https://www.kernel.org/doc/html/next/process/maintainer-netdev.html#tl-dr
The main goal of the 24h rule is to stop people from bombarding us with new versions for silly reasons.
You show know better than this, it's hardly your first contribution :(
Signature · 1 line
pv-bot: 24h
Quoted 8 lines · Jakub Kicinski
Quoted 5 lines · On Mon, 25 Mar 2024 14:28:28 +0800 Jason Xing wrote:
Before this, we miss some cases where the TCP layer could send rst but we cannot trace it. So I decided to complete it :)
v2
1. fix spelling mistakes
Quoted 7 lines
Not only do you post it before we "officially" open net-next but also ignoring the 24h wait period.
https://www.kernel.org/doc/html/next/process/maintainer-netdev.html#tl-dr
The main goal of the 24h rule is to stop people from bombarding us with new versions for silly reasons.
Sorry, I don't understand what you mean. I definitely know the rules. But the first version of this patch series was sent about two weeks ago (see link [1])
Yesterday, I posted two series to do two kinds of things. They are not the same. Maybe you get me wrong :S
link [1]: https://patchwork.kernel.org/project/netdevbpf/list/?series=834178&state=*
Thanks,
Jason
Quoted 4 lines
You show know better than this, it's hardly your first contribution :( --
pv-bot: 24h
Jakub Kicinski
maintainerQuoted 2 lines · On Tue, 26 Mar 2024 10:13:55 +0800 Jason Xing wrote:
Yesterday, I posted two series to do two kinds of things. They are not the same. Maybe you get me wrong :S
Ah, my bad, sorry about that. I see that they are different now. One is v1 the other v2, both targeting tcp tracing... Easy to miss in the post merge window rush :(
Quoted 6 lines · Jakub Kicinski
Quoted 2 lines · On Tue, 26 Mar 2024 10:13:55 +0800 Jason Xing wrote:
Yesterday, I posted two series to do two kinds of things. They are not the same. Maybe you get me wrong :S
Ah, my bad, sorry about that. I see that they are different now.
That's all right :)
Quoted 2 lines
One is v1 the other v2, both targeting tcp tracing... Easy to miss in the post merge window rush :(
Yes, and thanks for the check :)
Paolo Abeni
maintainerRe: [PATCH net-next v2 3/3] tcp: add location into reset trace process
In reply to Jason XingQuoted 23 lines · Jason Xing
From: Jason Xing <[email protected]>
In addition to knowing the 4-tuple of the flow which generates RST, the reason why it does so is very important because we have some cases where the RST should be sent and have no clue which one exactly.
Adding location of reset process can help us more, like what
trace_kfree_skbdoes.Signed-off-by Jason Xing <[email protected]>--- include/trace/events/tcp.h | 14 ++++++++++---- net/ipv4/tcp_ipv4.c | 2 +- net/ipv4/tcp_output.c | 2 +- net/ipv6/tcp_ipv6.c | 2 +- 4 files changed, 13 insertions(+), 7 deletions(-)include/trace/events/tcp.h +0
-0
include/trace/events/tcp.h DEFINE_EVENT(tcp_event_sk_skb, tcp_retransmit_skb, · line 115
Quoted 8 lines
include/trace/events/tcp.h +4
-1@@ -109,13 +109,17 @@ DEFINE_EVENT(tcp_event_sk_skb, tcp_retransmit_skb,*/ TRACE_EVENT(tcp_send_reset, - TP_PROTO(const struct sock *sk, const struct sk_buff *skb), + TP_PROTO( + const struct sock *sk, + const struct sk_buff *skb, + void *location),
Very minor nit: the above lines should be aligned with the open bracket.
No need to repost just for this, but let's wait for Eric's feedback.
Cheers,
Paolo
Jakub Kicinski
maintainerRe: [PATCH net-next v2 3/3] tcp: add location into reset trace process
Quoted 8 lines · On Tue, 26 Mar 2024 12:08:01 +0100 Paolo Abeni wrote:
include/trace/events/tcp.h DEFINE_EVENT(tcp_event_sk_skb, tcp_retransmit_skb, · line 115
Quoted 5 lines
include/trace/events/tcp.h +4
-1- TP_PROTO(const struct sock *sk, const struct sk_buff *skb), +TP_PROTO ( + const struct sock *sk, + const struct sk_buff *skb, + void *location),Very minor nit: the above lines should be aligned with the open bracket.
Yes, and a very odd way of breaking it up. Empty line after ( but ) not on a separate line.
Quoted 1 line
No need to repost just for this, but let's wait for Eric's feedback.
Erring on the side of caution I'd read this:
https://lore.kernel.org/all/CANn89iKK-qPhQ91Sq8rR_=KDWajnY2=Et2bUjDsgoQK4wxFOHw@mail.gmail.com/
as lukewarm towards tp changes. Please repost if you think otherwise (with the formatting fixed)
Signature · 1 line
pw-bot: cr
Re: [PATCH net-next v2 3/3] tcp: add location into reset trace process
Quoted 8 lines · Jakub Kicinski
Quoted 6 lines · On Tue, 26 Mar 2024 12:08:01 +0100 Paolo Abeni wrote:
Quoted 5 lines
include/trace/events/tcp.h +4
-1- TP_PROTO(const struct sock *sk, const struct sk_buff *skb), +TP_PROTO ( + const struct sock *sk, + const struct sk_buff *skb, + void *location),
Quoted 5 lines
Quoted 2 lines
Very minor nit: the above lines should be aligned with the open bracket.
Yes, and a very odd way of breaking it up. Empty line after ( but ) not on a separate line.
After I blamed the history, maybe I should follow the format like TRACE_EVENT(netfs_read)?
Quoted 7 lines
Quoted 2 lines
No need to repost just for this, but let's wait for Eric's feedback.
Erring on the side of caution I'd read this:
https://lore.kernel.org/all/CANn89iKK-qPhQ91Sq8rR_=KDWajnY2=Et2bUjDsgoQK4wxFOHw@mail.gmail.com/
as lukewarm towards tp changes. Please repost if you think otherwise (with the formatting fixed)
Yes, I will repost it. I'm not introducing a controversial new tracepoint.
This patch is not only about whether we should use 'old-way' tracing but about the tracepoint of this tcp reset that is not complete. Some admins could use bpf to capture RST behaviours through hooking this tracepoint which is not right currently apparently.
Besides, I simply tested the performance between using tracing and bpf to monitor the fast path (like __tcp_transmit_skb()) on my loopback. I saw at least 12% degradation with BPF used. So the advantage of trace is obvious even though nowadays it is considered as an old school method.
Thanks,
Jason
Quoted 2 lines
--
pw-bot: cr