* [PATCH net-next] tcp: use TCP_DEFAULT_INIT_RCVWND in tcp_fixup_rcvbuf()
@ 2011-10-20 19:16 Eric Dumazet
2011-10-20 19:50 ` David Miller
0 siblings, 1 reply; 6+ messages in thread
From: Eric Dumazet @ 2011-10-20 19:16 UTC (permalink / raw)
To: David Miller; +Cc: netdev
Since commit 356f039822b (TCP: increase default initial receive
window.), we allow sender to send 10 (TCP_DEFAULT_INIT_RCVWND) segments.
Change tcp_fixup_rcvbuf() to reflect this change, even if no real change
is expected, since sysctl_tcp_rmem[1] = 87380 and this value
is bigger than tcp_fixup_rcvbuf() computed rcvmem (~23720)
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
---
net/ipv4/tcp_input.c | 16 +++++++---------
1 file changed, 7 insertions(+), 9 deletions(-)
diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c
index 1e848b2..5a29ecc 100644
--- a/net/ipv4/tcp_input.c
+++ b/net/ipv4/tcp_input.c
@@ -345,17 +345,15 @@ static void tcp_grow_window(struct sock *sk, struct sk_buff *skb)
static void tcp_fixup_rcvbuf(struct sock *sk)
{
- struct tcp_sock *tp = tcp_sk(sk);
- int rcvmem = SKB_TRUESIZE(tp->advmss + MAX_TCP_HEADER);
+ const struct tcp_sock *tp = tcp_sk(sk);
+ unsigned int mss = min_t(unsigned int, tp->advmss, 1460);
+ int rcvmem = SKB_TRUESIZE(mss + MAX_TCP_HEADER);
- /* Try to select rcvbuf so that 4 mss-sized segments
- * will fit to window and corresponding skbs will fit to our rcvbuf.
- * (was 3; 4 is minimum to allow fast retransmit to work.)
- */
- while (tcp_win_from_space(rcvmem) < tp->advmss)
+ while (tcp_win_from_space(rcvmem) < mss)
rcvmem += 128;
- if (sk->sk_rcvbuf < 4 * rcvmem)
- sk->sk_rcvbuf = min(4 * rcvmem, sysctl_tcp_rmem[2]);
+ rcvmem *= TCP_DEFAULT_INIT_RCVWND;
+ if (sk->sk_rcvbuf < rcvmem)
+ sk->sk_rcvbuf = min(rcvmem, sysctl_tcp_rmem[2]);
}
/* 4. Try to fixup all. It is made immediately after connection enters
^ permalink raw reply related [flat|nested] 6+ messages in thread* Re: [PATCH net-next] tcp: use TCP_DEFAULT_INIT_RCVWND in tcp_fixup_rcvbuf()
2011-10-20 19:16 [PATCH net-next] tcp: use TCP_DEFAULT_INIT_RCVWND in tcp_fixup_rcvbuf() Eric Dumazet
@ 2011-10-20 19:50 ` David Miller
2011-10-20 20:02 ` Eric Dumazet
0 siblings, 1 reply; 6+ messages in thread
From: David Miller @ 2011-10-20 19:50 UTC (permalink / raw)
To: eric.dumazet; +Cc: netdev
From: Eric Dumazet <eric.dumazet@gmail.com>
Date: Thu, 20 Oct 2011 21:16:26 +0200
> Since commit 356f039822b (TCP: increase default initial receive
> window.), we allow sender to send 10 (TCP_DEFAULT_INIT_RCVWND) segments.
>
> Change tcp_fixup_rcvbuf() to reflect this change, even if no real change
> is expected, since sysctl_tcp_rmem[1] = 87380 and this value
> is bigger than tcp_fixup_rcvbuf() computed rcvmem (~23720)
>
> Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
...
> + unsigned int mss = min_t(unsigned int, tp->advmss, 1460);
I don't understand where this calculation comes from, and even if it
should be obvious it isn't to me and deserves a mention in the commit
message at a minimum.
^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: [PATCH net-next] tcp: use TCP_DEFAULT_INIT_RCVWND in tcp_fixup_rcvbuf()
2011-10-20 19:50 ` David Miller
@ 2011-10-20 20:02 ` Eric Dumazet
2011-10-20 20:13 ` David Miller
0 siblings, 1 reply; 6+ messages in thread
From: Eric Dumazet @ 2011-10-20 20:02 UTC (permalink / raw)
To: David Miller; +Cc: netdev
Le jeudi 20 octobre 2011 à 15:50 -0400, David Miller a écrit :
> From: Eric Dumazet <eric.dumazet@gmail.com>
> Date: Thu, 20 Oct 2011 21:16:26 +0200
>
> > Since commit 356f039822b (TCP: increase default initial receive
> > window.), we allow sender to send 10 (TCP_DEFAULT_INIT_RCVWND) segments.
> >
> > Change tcp_fixup_rcvbuf() to reflect this change, even if no real change
> > is expected, since sysctl_tcp_rmem[1] = 87380 and this value
> > is bigger than tcp_fixup_rcvbuf() computed rcvmem (~23720)
> >
> > Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
> ...
> > + unsigned int mss = min_t(unsigned int, tp->advmss, 1460);
>
> I don't understand where this calculation comes from, and even if it
> should be obvious it isn't to me and deserves a mention in the commit
> message at a minimum.
This is the calculation done in commit 356f039822b as well.
The window is 10*MSS, but no more than 14600
On loopback, this matters, because we could end with rcvmem=219680
^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: [PATCH net-next] tcp: use TCP_DEFAULT_INIT_RCVWND in tcp_fixup_rcvbuf()
2011-10-20 20:02 ` Eric Dumazet
@ 2011-10-20 20:13 ` David Miller
2011-10-20 20:41 ` [PATCH v2 " Eric Dumazet
0 siblings, 1 reply; 6+ messages in thread
From: David Miller @ 2011-10-20 20:13 UTC (permalink / raw)
To: eric.dumazet; +Cc: netdev
From: Eric Dumazet <eric.dumazet@gmail.com>
Date: Thu, 20 Oct 2011 22:02:34 +0200
> Le jeudi 20 octobre 2011 à 15:50 -0400, David Miller a écrit :
>> From: Eric Dumazet <eric.dumazet@gmail.com>
>> Date: Thu, 20 Oct 2011 21:16:26 +0200
>>
>> > Since commit 356f039822b (TCP: increase default initial receive
>> > window.), we allow sender to send 10 (TCP_DEFAULT_INIT_RCVWND) segments.
>> >
>> > Change tcp_fixup_rcvbuf() to reflect this change, even if no real change
>> > is expected, since sysctl_tcp_rmem[1] = 87380 and this value
>> > is bigger than tcp_fixup_rcvbuf() computed rcvmem (~23720)
>> >
>> > Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
>> ...
>> > + unsigned int mss = min_t(unsigned int, tp->advmss, 1460);
>>
>> I don't understand where this calculation comes from, and even if it
>> should be obvious it isn't to me and deserves a mention in the commit
>> message at a minimum.
>
> This is the calculation done in commit 356f039822b as well.
>
> The window is 10*MSS, but no more than 14600
>
> On loopback, this matters, because we could end with rcvmem=219680
Thanks, please help weak brains like mine by adding this to the commit message.
:-)
^ permalink raw reply [flat|nested] 6+ messages in thread
* [PATCH v2 net-next] tcp: use TCP_DEFAULT_INIT_RCVWND in tcp_fixup_rcvbuf()
2011-10-20 20:13 ` David Miller
@ 2011-10-20 20:41 ` Eric Dumazet
2011-10-20 20:54 ` David Miller
0 siblings, 1 reply; 6+ messages in thread
From: Eric Dumazet @ 2011-10-20 20:41 UTC (permalink / raw)
To: David Miller; +Cc: netdev
Since commit 356f039822b (TCP: increase default initial receive
window.), we allow sender to send 10 (TCP_DEFAULT_INIT_RCVWND) segments.
Change tcp_fixup_rcvbuf() to reflect this change, even if no real change
is expected, since sysctl_tcp_rmem[1] = 87380 and this value
is bigger than tcp_fixup_rcvbuf() computed rcvmem (~23720)
Note: Since commit 356f039822b limited default window to maximum of
10*1460 and 2*MSS, we use same heuristic in this patch.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
---
net/ipv4/tcp_input.c | 23 +++++++++++++++--------
1 file changed, 15 insertions(+), 8 deletions(-)
diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c
index 1e848b2..e8e6d49 100644
--- a/net/ipv4/tcp_input.c
+++ b/net/ipv4/tcp_input.c
@@ -345,17 +345,24 @@ static void tcp_grow_window(struct sock *sk, struct sk_buff *skb)
static void tcp_fixup_rcvbuf(struct sock *sk)
{
- struct tcp_sock *tp = tcp_sk(sk);
- int rcvmem = SKB_TRUESIZE(tp->advmss + MAX_TCP_HEADER);
+ u32 mss = tcp_sk(sk)->advmss;
+ u32 icwnd = TCP_DEFAULT_INIT_RCVWND;
+ int rcvmem;
- /* Try to select rcvbuf so that 4 mss-sized segments
- * will fit to window and corresponding skbs will fit to our rcvbuf.
- * (was 3; 4 is minimum to allow fast retransmit to work.)
+ /* Limit to 10 segments if mss <= 1460,
+ * or 14600/mss segments, with a minimum of two segments.
*/
- while (tcp_win_from_space(rcvmem) < tp->advmss)
+ if (mss > 1460)
+ icwnd = max_t(u32, (1460 * TCP_DEFAULT_INIT_RCVWND) / mss, 2);
+
+ rcvmem = SKB_TRUESIZE(mss + MAX_TCP_HEADER);
+ while (tcp_win_from_space(rcvmem) < mss)
rcvmem += 128;
- if (sk->sk_rcvbuf < 4 * rcvmem)
- sk->sk_rcvbuf = min(4 * rcvmem, sysctl_tcp_rmem[2]);
+
+ rcvmem *= icwnd;
+
+ if (sk->sk_rcvbuf < rcvmem)
+ sk->sk_rcvbuf = min(rcvmem, sysctl_tcp_rmem[2]);
}
/* 4. Try to fixup all. It is made immediately after connection enters
^ permalink raw reply related [flat|nested] 6+ messages in thread* Re: [PATCH v2 net-next] tcp: use TCP_DEFAULT_INIT_RCVWND in tcp_fixup_rcvbuf()
2011-10-20 20:41 ` [PATCH v2 " Eric Dumazet
@ 2011-10-20 20:54 ` David Miller
0 siblings, 0 replies; 6+ messages in thread
From: David Miller @ 2011-10-20 20:54 UTC (permalink / raw)
To: eric.dumazet; +Cc: netdev
From: Eric Dumazet <eric.dumazet@gmail.com>
Date: Thu, 20 Oct 2011 22:41:21 +0200
> Since commit 356f039822b (TCP: increase default initial receive
> window.), we allow sender to send 10 (TCP_DEFAULT_INIT_RCVWND) segments.
>
> Change tcp_fixup_rcvbuf() to reflect this change, even if no real change
> is expected, since sysctl_tcp_rmem[1] = 87380 and this value
> is bigger than tcp_fixup_rcvbuf() computed rcvmem (~23720)
>
> Note: Since commit 356f039822b limited default window to maximum of
> 10*1460 and 2*MSS, we use same heuristic in this patch.
>
> Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Applied, thanks a lot Eric.
^ permalink raw reply [flat|nested] 6+ messages in thread
end of thread, other threads:[~2011-10-20 20:54 UTC | newest]
Thread overview: 6+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2011-10-20 19:16 [PATCH net-next] tcp: use TCP_DEFAULT_INIT_RCVWND in tcp_fixup_rcvbuf() Eric Dumazet
2011-10-20 19:50 ` David Miller
2011-10-20 20:02 ` Eric Dumazet
2011-10-20 20:13 ` David Miller
2011-10-20 20:41 ` [PATCH v2 " Eric Dumazet
2011-10-20 20:54 ` David Miller
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).