Skip to content

Commit 4c4fde2

Browse files
davem330davem330
committed
Merge branch 'tcp-tracepoints'
Song Liu says: ==================== net: add a set of tracepoints to tcp stack Changes from v1: Fix build error (with ipv6 as ko) by adding EXPORT_TRACEPOINT_SYMBOL_GPL for trace_tcp_send_reset. These patches add the following tracepoints to tcp stack. tcp_send_reset tcp_receive_reset tcp_destroy_sock tcp_set_state These tracepoints can be used to track TCP state changes. Such state changes include but are not limited to: connection establish, connection termination, tx and rx of RST, various retransmits. Currently, we use the following kprobes to trace these events: int kprobe__tcp_validate_incoming int kprobe__tcp_send_active_reset int kprobe__tcp_v4_send_reset int kprobe__tcp_v6_send_reset int kprobe__tcp_v4_destroy_sock int kprobe__tcp_set_state int kprobe__tcp_retransmit_skb These tracepoints will help us simplify this work. ==================== Signed-off-by: David S. Miller <davem@davemloft.net>
2 parents bded442 + e8fce23 commit 4c4fde2

File tree

7 files changed

+206
-7
lines changed

7 files changed

+206
-7
lines changed

‎include/trace/events/tcp.h

Lines changed: 177 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -9,15 +9,37 @@
99
#include <linux/tracepoint.h>
1010
#include <net/ipv6.h>
1111

12-
TRACE_EVENT(tcp_retransmit_skb,
12+
#define tcp_state_name(state) { state, #state }
13+
#define show_tcp_state_name(val) \
14+
__print_symbolic(val, \
15+
tcp_state_name(TCP_ESTABLISHED), \
16+
tcp_state_name(TCP_SYN_SENT), \
17+
tcp_state_name(TCP_SYN_RECV), \
18+
tcp_state_name(TCP_FIN_WAIT1), \
19+
tcp_state_name(TCP_FIN_WAIT2), \
20+
tcp_state_name(TCP_TIME_WAIT), \
21+
tcp_state_name(TCP_CLOSE), \
22+
tcp_state_name(TCP_CLOSE_WAIT), \
23+
tcp_state_name(TCP_LAST_ACK), \
24+
tcp_state_name(TCP_LISTEN), \
25+
tcp_state_name(TCP_CLOSING), \
26+
tcp_state_name(TCP_NEW_SYN_RECV))
1327

14-
TP_PROTO(struct sock *sk, struct sk_buff *skb),
28+
/*
29+
* tcp event with arguments sk and skb
30+
*
31+
* Note: this class requires a valid sk pointer; while skb pointer could
32+
* be NULL.
33+
*/
34+
DECLARE_EVENT_CLASS(tcp_event_sk_skb,
35+
36+
TP_PROTO(const struct sock *sk, const struct sk_buff *skb),
1537

1638
TP_ARGS(sk, skb),
1739

1840
TP_STRUCT__entry(
19-
__field(void *, skbaddr)
20-
__field(void *, skaddr)
41+
__field(const void *, skbaddr)
42+
__field(const void *, skaddr)
2143
__field(__u16, sport)
2244
__field(__u16, dport)
2345
__array(__u8, saddr, 4)
@@ -64,6 +86,157 @@ TRACE_EVENT(tcp_retransmit_skb,
6486
__entry->saddr_v6, __entry->daddr_v6)
6587
);
6688

89+
DEFINE_EVENT(tcp_event_sk_skb, tcp_retransmit_skb,
90+
91+
TP_PROTO(const struct sock *sk, const struct sk_buff *skb),
92+
93+
TP_ARGS(sk, skb)
94+
);
95+
96+
/*
97+
* skb of trace_tcp_send_reset is the skb that caused RST. In case of
98+
* active reset, skb should be NULL
99+
*/
100+
DEFINE_EVENT(tcp_event_sk_skb, tcp_send_reset,
101+
102+
TP_PROTO(const struct sock *sk, const struct sk_buff *skb),
103+
104+
TP_ARGS(sk, skb)
105+
);
106+
107+
/*
108+
* tcp event with arguments sk
109+
*
110+
* Note: this class requires a valid sk pointer.
111+
*/
112+
DECLARE_EVENT_CLASS(tcp_event_sk,
113+
114+
TP_PROTO(const struct sock *sk),
115+
116+
TP_ARGS(sk),
117+
118+
TP_STRUCT__entry(
119+
__field(const void *, skaddr)
120+
__field(__u16, sport)
121+
__field(__u16, dport)
122+
__array(__u8, saddr, 4)
123+
__array(__u8, daddr, 4)
124+
__array(__u8, saddr_v6, 16)
125+
__array(__u8, daddr_v6, 16)
126+
),
127+
128+
TP_fast_assign(
129+
struct inet_sock *inet = inet_sk(sk);
130+
struct in6_addr *pin6;
131+
__be32 *p32;
132+
133+
__entry->skaddr = sk;
134+
135+
__entry->sport = ntohs(inet->inet_sport);
136+
__entry->dport = ntohs(inet->inet_dport);
137+
138+
p32 = (__be32 *) __entry->saddr;
139+
*p32 = inet->inet_saddr;
140+
141+
p32 = (__be32 *) __entry->daddr;
142+
*p32 = inet->inet_daddr;
143+
144+
#if IS_ENABLED(CONFIG_IPV6)
145+
if (sk->sk_family == AF_INET6) {
146+
pin6 = (struct in6_addr *)__entry->saddr_v6;
147+
*pin6 = sk->sk_v6_rcv_saddr;
148+
pin6 = (struct in6_addr *)__entry->daddr_v6;
149+
*pin6 = sk->sk_v6_daddr;
150+
} else
151+
#endif
152+
{
153+
pin6 = (struct in6_addr *)__entry->saddr_v6;
154+
ipv6_addr_set_v4mapped(inet->inet_saddr, pin6);
155+
pin6 = (struct in6_addr *)__entry->daddr_v6;
156+
ipv6_addr_set_v4mapped(inet->inet_daddr, pin6);
157+
}
158+
),
159+
160+
TP_printk("sport=%hu dport=%hu saddr=%pI4 daddr=%pI4 saddrv6=%pI6c daddrv6=%pI6c",
161+
__entry->sport, __entry->dport,
162+
__entry->saddr, __entry->daddr,
163+
__entry->saddr_v6, __entry->daddr_v6)
164+
);
165+
166+
DEFINE_EVENT(tcp_event_sk, tcp_receive_reset,
167+
168+
TP_PROTO(const struct sock *sk),
169+
170+
TP_ARGS(sk)
171+
);
172+
173+
DEFINE_EVENT(tcp_event_sk, tcp_destroy_sock,
174+
175+
TP_PROTO(const struct sock *sk),
176+
177+
TP_ARGS(sk)
178+
);
179+
180+
TRACE_EVENT(tcp_set_state,
181+
182+
TP_PROTO(const struct sock *sk, const int oldstate, const int newstate),
183+
184+
TP_ARGS(sk, oldstate, newstate),
185+
186+
TP_STRUCT__entry(
187+
__field(const void *, skaddr)
188+
__field(int, oldstate)
189+
__field(int, newstate)
190+
__field(__u16, sport)
191+
__field(__u16, dport)
192+
__array(__u8, saddr, 4)
193+
__array(__u8, daddr, 4)
194+
__array(__u8, saddr_v6, 16)
195+
__array(__u8, daddr_v6, 16)
196+
),
197+
198+
TP_fast_assign(
199+
struct inet_sock *inet = inet_sk(sk);
200+
struct in6_addr *pin6;
201+
__be32 *p32;
202+
203+
__entry->skaddr = sk;
204+
__entry->oldstate = oldstate;
205+
__entry->newstate = newstate;
206+
207+
__entry->sport = ntohs(inet->inet_sport);
208+
__entry->dport = ntohs(inet->inet_dport);
209+
210+
p32 = (__be32 *) __entry->saddr;
211+
*p32 = inet->inet_saddr;
212+
213+
p32 = (__be32 *) __entry->daddr;
214+
*p32 = inet->inet_daddr;
215+
216+
#if IS_ENABLED(CONFIG_IPV6)
217+
if (sk->sk_family == AF_INET6) {
218+
pin6 = (struct in6_addr *)__entry->saddr_v6;
219+
*pin6 = sk->sk_v6_rcv_saddr;
220+
pin6 = (struct in6_addr *)__entry->daddr_v6;
221+
*pin6 = sk->sk_v6_daddr;
222+
} else
223+
#endif
224+
{
225+
pin6 = (struct in6_addr *)__entry->saddr_v6;
226+
ipv6_addr_set_v4mapped(inet->inet_saddr, pin6);
227+
pin6 = (struct in6_addr *)__entry->daddr_v6;
228+
ipv6_addr_set_v4mapped(inet->inet_daddr, pin6);
229+
}
230+
),
231+
232+
TP_printk("sport=%hu dport=%hu saddr=%pI4 daddr=%pI4 saddrv6=%pI6c daddrv6=%pI6c oldstate=%s newstate=%s",
233+
__entry->sport, __entry->dport,
234+
__entry->saddr, __entry->daddr,
235+
__entry->saddr_v6, __entry->daddr_v6,
236+
show_tcp_state_name(__entry->oldstate),
237+
show_tcp_state_name(__entry->newstate))
238+
);
239+
67240
#endif /* _TRACE_TCP_H */
68241

69242
/* This part must be outside protection */

‎net/core/net-traces.c

Lines changed: 2 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -49,3 +49,5 @@ EXPORT_TRACEPOINT_SYMBOL_GPL(br_fdb_update);
4949
EXPORT_TRACEPOINT_SYMBOL_GPL(kfree_skb);
5050

5151
EXPORT_TRACEPOINT_SYMBOL_GPL(napi_poll);
52+
53+
EXPORT_TRACEPOINT_SYMBOL_GPL(tcp_send_reset);

‎net/ipv4/tcp.c

Lines changed: 4 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -282,6 +282,8 @@
282282
#include <asm/ioctls.h>
283283
#include <net/busy_poll.h>
284284

285+
#include <trace/events/tcp.h>
286+
285287
int sysctl_tcp_min_tso_segs __read_mostly = 2;
286288

287289
int sysctl_tcp_autocorking __read_mostly = 1;
@@ -2040,6 +2042,8 @@ void tcp_set_state(struct sock *sk, int state)
20402042
{
20412043
int oldstate = sk->sk_state;
20422044

2045+
trace_tcp_set_state(sk, oldstate, state);
2046+
20432047
switch (state) {
20442048
case TCP_ESTABLISHED:
20452049
if (oldstate != TCP_ESTABLISHED)

‎net/ipv4/tcp_input.c

Lines changed: 3 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -75,6 +75,7 @@
7575
#include <linux/ipsec.h>
7676
#include <asm/unaligned.h>
7777
#include <linux/errqueue.h>
78+
#include <trace/events/tcp.h>
7879

7980
int sysctl_tcp_fack __read_mostly;
8081
int sysctl_tcp_max_reordering __read_mostly = 300;
@@ -4010,6 +4011,8 @@ static inline bool tcp_sequence(const struct tcp_sock *tp, u32 seq, u32 end_seq)
40104011
/* When we get a reset we do this. */
40114012
void tcp_reset(struct sock *sk)
40124013
{
4014+
trace_tcp_receive_reset(sk);
4015+
40134016
/* We want the right error as BSD sees it (and indeed as we do). */
40144017
switch (sk->sk_state) {
40154018
case TCP_SYN_SENT:

‎net/ipv4/tcp_ipv4.c

Lines changed: 7 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -85,6 +85,8 @@
8585
#include <crypto/hash.h>
8686
#include <linux/scatterlist.h>
8787

88+
#include <trace/events/tcp.h>
89+
8890
#ifdef CONFIG_TCP_MD5SIG
8991
static int tcp_v4_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
9092
__be32 daddr, __be32 saddr, const struct tcphdr *th);
@@ -701,8 +703,10 @@ static void tcp_v4_send_reset(const struct sock *sk, struct sk_buff *skb)
701703
* routing might fail in this case. No choice here, if we choose to force
702704
* input interface, we will misroute in case of asymmetric route.
703705
*/
704-
if (sk)
706+
if (sk) {
705707
arg.bound_dev_if = sk->sk_bound_dev_if;
708+
trace_tcp_send_reset(sk, skb);
709+
}
706710

707711
BUILD_BUG_ON(offsetof(struct sock, sk_bound_dev_if) !=
708712
offsetof(struct inet_timewait_sock, tw_bound_dev_if));
@@ -1865,6 +1869,8 @@ void tcp_v4_destroy_sock(struct sock *sk)
18651869
{
18661870
struct tcp_sock *tp = tcp_sk(sk);
18671871

1872+
trace_tcp_destroy_sock(sk);
1873+
18681874
tcp_clear_xmit_timers(sk);
18691875

18701876
tcp_cleanup_congestion_control(sk);

‎net/ipv4/tcp_output.c

Lines changed: 5 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -3084,6 +3084,11 @@ void tcp_send_active_reset(struct sock *sk, gfp_t priority)
30843084
/* Send it off. */
30853085
if (tcp_transmit_skb(sk, skb, 0, priority))
30863086
NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPABORTFAILED);
3087+
3088+
/* skb of trace_tcp_send_reset() keeps the skb that caused RST,
3089+
* skb here is different to the troublesome skb, so use NULL
3090+
*/
3091+
trace_tcp_send_reset(sk, NULL);
30873092
}
30883093

30893094
/* Send a crossed SYN-ACK during socket establishment.

‎net/ipv6/tcp_ipv6.c

Lines changed: 8 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -69,6 +69,8 @@
6969
#include <crypto/hash.h>
7070
#include <linux/scatterlist.h>
7171

72+
#include <trace/events/tcp.h>
73+
7274
static void tcp_v6_send_reset(const struct sock *sk, struct sk_buff *skb);
7375
static void tcp_v6_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
7476
struct request_sock *req);
@@ -890,7 +892,7 @@ static void tcp_v6_send_reset(const struct sock *sk, struct sk_buff *skb)
890892
int genhash;
891893
struct sock *sk1 = NULL;
892894
#endif
893-
int oif;
895+
int oif = 0;
894896

895897
if (th->rst)
896898
return;
@@ -939,7 +941,11 @@ static void tcp_v6_send_reset(const struct sock *sk, struct sk_buff *skb)
939941
ack_seq = ntohl(th->seq) + th->syn + th->fin + skb->len -
940942
(th->doff << 2);
941943

942-
oif = sk ? sk->sk_bound_dev_if : 0;
944+
if (sk) {
945+
oif = sk->sk_bound_dev_if;
946+
trace_tcp_send_reset(sk, skb);
947+
}
948+
943949
tcp_v6_send_response(sk, skb, seq, ack_seq, 0, 0, 0, oif, key, 1, 0, 0);
944950

945951
#ifdef CONFIG_TCP_MD5SIG

0 commit comments

Comments
 (0)