[PATCH net] tcp: tcp_get_info() should read tcp_time_stamp later

[PATCH net] tcp: tcp_get_info() should read tcp_time_stamp later

[Eric Dumazet]

From: Eric Dumazet <edumazet@google.com>

Commit b369e7fd41f7 ("tcp: make TCP_INFO more consistent") moved
lock_sock_fast() earlier in tcp_get_info()

This has the minor effect that jiffies value being sampled at the
beginning of tcp_get_info() is more likely to be off by one, and we
report big tcpi_last_data_sent values (like 0xFFFFFFFF).

Since we lock the socket, fetching tcp_time_stamp right before
doing the jiffies_to_msecs() calls is enough to remove these
wrong values.

Signed-off-by: Eric Dumazet <edumazet@google.com>
---
 net/ipv4/tcp.c |    3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/net/ipv4/tcp.c b/net/ipv4/tcp.c
index cf4555581282c608f920254078264e36e18584c6..1e319a525d51b0b603a5ccc5143381c752b9f2c7 100644
--- a/net/ipv4/tcp.c
+++ b/net/ipv4/tcp.c
@@ -2770,7 +2770,7 @@ void tcp_get_info(struct sock *sk, struct tcp_info *info)
 {
 	const struct tcp_sock *tp = tcp_sk(sk); /* iff sk_type == SOCK_STREAM */
 	const struct inet_connection_sock *icsk = inet_csk(sk);
-	u32 now = tcp_time_stamp, intv;
+	u32 now, intv;
 	u64 rate64;
 	bool slow;
 	u32 rate;
@@ -2839,6 +2839,7 @@ void tcp_get_info(struct sock *sk, struct tcp_info *info)
 	info->tcpi_retrans = tp->retrans_out;
 	info->tcpi_fackets = tp->fackets_out;
 
+	now = tcp_time_stamp;
 	info->tcpi_last_data_sent = jiffies_to_msecs(now - tp->lsndtime);
 	info->tcpi_last_data_recv = jiffies_to_msecs(now - icsk->icsk_ack.lrcvtime);
 	info->tcpi_last_ack_recv = jiffies_to_msecs(now - tp->rcv_tstamp);

Re: [PATCH net] tcp: tcp_get_info() should read tcp_time_stamp later

[David Miller]

From: Eric Dumazet <eric.dumazet@gmail.com>
Date: Thu, 16 Mar 2017 15:43:19 -0700

> From: Eric Dumazet <edumazet@google.com>
> 
> Commit b369e7fd41f7 ("tcp: make TCP_INFO more consistent") moved
> lock_sock_fast() earlier in tcp_get_info()
> 
> This has the minor effect that jiffies value being sampled at the
> beginning of tcp_get_info() is more likely to be off by one, and we
> report big tcpi_last_data_sent values (like 0xFFFFFFFF).
> 
> Since we lock the socket, fetching tcp_time_stamp right before
> doing the jiffies_to_msecs() calls is enough to remove these
> wrong values.
> 
> Signed-off-by: Eric Dumazet <edumazet@google.com>

Applied, thanks Eric.