2005-04-17 05:20:36 +07:00
|
|
|
/*
|
|
|
|
* INET An implementation of the TCP/IP protocol suite for the LINUX
|
|
|
|
* operating system. INET is implemented using the BSD Socket
|
|
|
|
* interface as the means of communication with the user level.
|
|
|
|
*
|
|
|
|
* Implementation of the Transmission Control Protocol(TCP).
|
|
|
|
*
|
2005-05-06 06:16:16 +07:00
|
|
|
* Authors: Ross Biro
|
2005-04-17 05:20:36 +07:00
|
|
|
* Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
|
|
|
|
* Mark Evans, <evansmp@uhura.aston.ac.uk>
|
|
|
|
* Corey Minyard <wf-rch!minyard@relay.EU.net>
|
|
|
|
* Florian La Roche, <flla@stud.uni-sb.de>
|
|
|
|
* Charles Hedrick, <hedrick@klinzhai.rutgers.edu>
|
|
|
|
* Linus Torvalds, <torvalds@cs.helsinki.fi>
|
|
|
|
* Alan Cox, <gw4pts@gw4pts.ampr.org>
|
|
|
|
* Matthew Dillon, <dillon@apollo.west.oic.com>
|
|
|
|
* Arnt Gulbrandsen, <agulbra@nvg.unit.no>
|
|
|
|
* Jorge Cwik, <jorge@laser.satlink.net>
|
|
|
|
*/
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Changes: Pedro Roque : Retransmit queue handled by TCP.
|
|
|
|
* : Fragmentation on mtu decrease
|
|
|
|
* : Segment collapse on retransmit
|
|
|
|
* : AF independence
|
|
|
|
*
|
|
|
|
* Linus Torvalds : send_delayed_ack
|
|
|
|
* David S. Miller : Charge memory using the right skb
|
|
|
|
* during syn/ack processing.
|
|
|
|
* David S. Miller : Output engine completely rewritten.
|
|
|
|
* Andrea Arcangeli: SYNACK carry ts_recent in tsecr.
|
|
|
|
* Cacophonix Gaul : draft-minshall-nagle-01
|
|
|
|
* J Hadi Salim : ECN support
|
|
|
|
*
|
|
|
|
*/
|
|
|
|
|
2012-05-15 21:11:54 +07:00
|
|
|
#define pr_fmt(fmt) "TCP: " fmt
|
|
|
|
|
2005-04-17 05:20:36 +07:00
|
|
|
#include <net/tcp.h>
|
|
|
|
|
|
|
|
#include <linux/compiler.h>
|
include cleanup: Update gfp.h and slab.h includes to prepare for breaking implicit slab.h inclusion from percpu.h
percpu.h is included by sched.h and module.h and thus ends up being
included when building most .c files. percpu.h includes slab.h which
in turn includes gfp.h making everything defined by the two files
universally available and complicating inclusion dependencies.
percpu.h -> slab.h dependency is about to be removed. Prepare for
this change by updating users of gfp and slab facilities include those
headers directly instead of assuming availability. As this conversion
needs to touch large number of source files, the following script is
used as the basis of conversion.
http://userweb.kernel.org/~tj/misc/slabh-sweep.py
The script does the followings.
* Scan files for gfp and slab usages and update includes such that
only the necessary includes are there. ie. if only gfp is used,
gfp.h, if slab is used, slab.h.
* When the script inserts a new include, it looks at the include
blocks and try to put the new include such that its order conforms
to its surrounding. It's put in the include block which contains
core kernel includes, in the same order that the rest are ordered -
alphabetical, Christmas tree, rev-Xmas-tree or at the end if there
doesn't seem to be any matching order.
* If the script can't find a place to put a new include (mostly
because the file doesn't have fitting include block), it prints out
an error message indicating which .h file needs to be added to the
file.
The conversion was done in the following steps.
1. The initial automatic conversion of all .c files updated slightly
over 4000 files, deleting around 700 includes and adding ~480 gfp.h
and ~3000 slab.h inclusions. The script emitted errors for ~400
files.
2. Each error was manually checked. Some didn't need the inclusion,
some needed manual addition while adding it to implementation .h or
embedding .c file was more appropriate for others. This step added
inclusions to around 150 files.
3. The script was run again and the output was compared to the edits
from #2 to make sure no file was left behind.
4. Several build tests were done and a couple of problems were fixed.
e.g. lib/decompress_*.c used malloc/free() wrappers around slab
APIs requiring slab.h to be added manually.
5. The script was run on all .h files but without automatically
editing them as sprinkling gfp.h and slab.h inclusions around .h
files could easily lead to inclusion dependency hell. Most gfp.h
inclusion directives were ignored as stuff from gfp.h was usually
wildly available and often used in preprocessor macros. Each
slab.h inclusion directive was examined and added manually as
necessary.
6. percpu.h was updated not to include slab.h.
7. Build test were done on the following configurations and failures
were fixed. CONFIG_GCOV_KERNEL was turned off for all tests (as my
distributed build env didn't work with gcov compiles) and a few
more options had to be turned off depending on archs to make things
build (like ipr on powerpc/64 which failed due to missing writeq).
* x86 and x86_64 UP and SMP allmodconfig and a custom test config.
* powerpc and powerpc64 SMP allmodconfig
* sparc and sparc64 SMP allmodconfig
* ia64 SMP allmodconfig
* s390 SMP allmodconfig
* alpha SMP allmodconfig
* um on x86_64 SMP allmodconfig
8. percpu.h modifications were reverted so that it could be applied as
a separate patch and serve as bisection point.
Given the fact that I had only a couple of failures from tests on step
6, I'm fairly confident about the coverage of this conversion patch.
If there is a breakage, it's likely to be something in one of the arch
headers which should be easily discoverable easily on most builds of
the specific arch.
Signed-off-by: Tejun Heo <tj@kernel.org>
Guess-its-ok-by: Christoph Lameter <cl@linux-foundation.org>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Lee Schermerhorn <Lee.Schermerhorn@hp.com>
2010-03-24 15:04:11 +07:00
|
|
|
#include <linux/gfp.h>
|
2005-04-17 05:20:36 +07:00
|
|
|
#include <linux/module.h>
|
|
|
|
|
|
|
|
/* People can turn this off for buggy TCP's found in printers etc. */
|
2006-09-23 04:15:41 +07:00
|
|
|
int sysctl_tcp_retrans_collapse __read_mostly = 1;
|
2005-04-17 05:20:36 +07:00
|
|
|
|
2008-11-03 15:27:11 +07:00
|
|
|
/* People can turn this on to work with those rare, broken TCPs that
|
2006-03-21 13:40:29 +07:00
|
|
|
* interpret the window field as a signed quantity.
|
|
|
|
*/
|
2006-09-23 04:15:41 +07:00
|
|
|
int sysctl_tcp_workaround_signed_windows __read_mostly = 0;
|
2006-03-21 13:40:29 +07:00
|
|
|
|
2015-06-03 17:10:42 +07:00
|
|
|
/* Default TSQ limit of four TSO segments */
|
|
|
|
int sysctl_tcp_limit_output_bytes __read_mostly = 262144;
|
tcp: TCP Small Queues
This introduce TSQ (TCP Small Queues)
TSQ goal is to reduce number of TCP packets in xmit queues (qdisc &
device queues), to reduce RTT and cwnd bias, part of the bufferbloat
problem.
sk->sk_wmem_alloc not allowed to grow above a given limit,
allowing no more than ~128KB [1] per tcp socket in qdisc/dev layers at a
given time.
TSO packets are sized/capped to half the limit, so that we have two
TSO packets in flight, allowing better bandwidth use.
As a side effect, setting the limit to 40000 automatically reduces the
standard gso max limit (65536) to 40000/2 : It can help to reduce
latencies of high prio packets, having smaller TSO packets.
This means we divert sock_wfree() to a tcp_wfree() handler, to
queue/send following frames when skb_orphan() [2] is called for the
already queued skbs.
Results on my dev machines (tg3/ixgbe nics) are really impressive,
using standard pfifo_fast, and with or without TSO/GSO.
Without reduction of nominal bandwidth, we have reduction of buffering
per bulk sender :
< 1ms on Gbit (instead of 50ms with TSO)
< 8ms on 100Mbit (instead of 132 ms)
I no longer have 4 MBytes backlogged in qdisc by a single netperf
session, and both side socket autotuning no longer use 4 Mbytes.
As skb destructor cannot restart xmit itself ( as qdisc lock might be
taken at this point ), we delegate the work to a tasklet. We use one
tasklest per cpu for performance reasons.
If tasklet finds a socket owned by the user, it sets TSQ_OWNED flag.
This flag is tested in a new protocol method called from release_sock(),
to eventually send new segments.
[1] New /proc/sys/net/ipv4/tcp_limit_output_bytes tunable
[2] skb_orphan() is usually called at TX completion time,
but some drivers call it in their start_xmit() handler.
These drivers should at least use BQL, or else a single TCP
session can still fill the whole NIC TX ring, since TSQ will
have no effect.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Dave Taht <dave.taht@bufferbloat.net>
Cc: Tom Herbert <therbert@google.com>
Cc: Matt Mathis <mattmathis@google.com>
Cc: Yuchung Cheng <ycheng@google.com>
Cc: Nandita Dukkipati <nanditad@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2012-07-11 12:50:31 +07:00
|
|
|
|
2005-04-17 05:20:36 +07:00
|
|
|
/* This limits the percentage of the congestion window which we
|
|
|
|
* will allow a single TSO frame to consume. Building TSO frames
|
|
|
|
* which are too large can cause TCP streams to be bursty.
|
|
|
|
*/
|
2006-09-23 04:15:41 +07:00
|
|
|
int sysctl_tcp_tso_win_divisor __read_mostly = 3;
|
2005-04-17 05:20:36 +07:00
|
|
|
|
2006-06-14 12:33:04 +07:00
|
|
|
/* By default, RFC2861 behavior. */
|
2006-09-23 04:15:41 +07:00
|
|
|
int sysctl_tcp_slow_start_after_idle __read_mostly = 1;
|
2006-06-14 12:33:04 +07:00
|
|
|
|
tcp: TCP Small Queues
This introduce TSQ (TCP Small Queues)
TSQ goal is to reduce number of TCP packets in xmit queues (qdisc &
device queues), to reduce RTT and cwnd bias, part of the bufferbloat
problem.
sk->sk_wmem_alloc not allowed to grow above a given limit,
allowing no more than ~128KB [1] per tcp socket in qdisc/dev layers at a
given time.
TSO packets are sized/capped to half the limit, so that we have two
TSO packets in flight, allowing better bandwidth use.
As a side effect, setting the limit to 40000 automatically reduces the
standard gso max limit (65536) to 40000/2 : It can help to reduce
latencies of high prio packets, having smaller TSO packets.
This means we divert sock_wfree() to a tcp_wfree() handler, to
queue/send following frames when skb_orphan() [2] is called for the
already queued skbs.
Results on my dev machines (tg3/ixgbe nics) are really impressive,
using standard pfifo_fast, and with or without TSO/GSO.
Without reduction of nominal bandwidth, we have reduction of buffering
per bulk sender :
< 1ms on Gbit (instead of 50ms with TSO)
< 8ms on 100Mbit (instead of 132 ms)
I no longer have 4 MBytes backlogged in qdisc by a single netperf
session, and both side socket autotuning no longer use 4 Mbytes.
As skb destructor cannot restart xmit itself ( as qdisc lock might be
taken at this point ), we delegate the work to a tasklet. We use one
tasklest per cpu for performance reasons.
If tasklet finds a socket owned by the user, it sets TSQ_OWNED flag.
This flag is tested in a new protocol method called from release_sock(),
to eventually send new segments.
[1] New /proc/sys/net/ipv4/tcp_limit_output_bytes tunable
[2] skb_orphan() is usually called at TX completion time,
but some drivers call it in their start_xmit() handler.
These drivers should at least use BQL, or else a single TCP
session can still fill the whole NIC TX ring, since TSQ will
have no effect.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Dave Taht <dave.taht@bufferbloat.net>
Cc: Tom Herbert <therbert@google.com>
Cc: Matt Mathis <mattmathis@google.com>
Cc: Yuchung Cheng <ycheng@google.com>
Cc: Nandita Dukkipati <nanditad@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2012-07-11 12:50:31 +07:00
|
|
|
static bool tcp_write_xmit(struct sock *sk, unsigned int mss_now, int nonagle,
|
|
|
|
int push_one, gfp_t gfp);
|
2009-12-03 01:14:19 +07:00
|
|
|
|
2009-07-22 06:00:40 +07:00
|
|
|
/* Account for new data that has been sent to the network. */
|
2011-10-21 16:22:42 +07:00
|
|
|
static void tcp_event_new_data_sent(struct sock *sk, const struct sk_buff *skb)
|
2005-04-17 05:20:36 +07:00
|
|
|
{
|
tcp: Tail loss probe (TLP)
This patch series implement the Tail loss probe (TLP) algorithm described
in http://tools.ietf.org/html/draft-dukkipati-tcpm-tcp-loss-probe-01. The
first patch implements the basic algorithm.
TLP's goal is to reduce tail latency of short transactions. It achieves
this by converting retransmission timeouts (RTOs) occuring due
to tail losses (losses at end of transactions) into fast recovery.
TLP transmits one packet in two round-trips when a connection is in
Open state and isn't receiving any ACKs. The transmitted packet, aka
loss probe, can be either new or a retransmission. When there is tail
loss, the ACK from a loss probe triggers FACK/early-retransmit based
fast recovery, thus avoiding a costly RTO. In the absence of loss,
there is no change in the connection state.
PTO stands for probe timeout. It is a timer event indicating
that an ACK is overdue and triggers a loss probe packet. The PTO value
is set to max(2*SRTT, 10ms) and is adjusted to account for delayed
ACK timer when there is only one oustanding packet.
TLP Algorithm
On transmission of new data in Open state:
-> packets_out > 1: schedule PTO in max(2*SRTT, 10ms).
-> packets_out == 1: schedule PTO in max(2*RTT, 1.5*RTT + 200ms)
-> PTO = min(PTO, RTO)
Conditions for scheduling PTO:
-> Connection is in Open state.
-> Connection is either cwnd limited or no new data to send.
-> Number of probes per tail loss episode is limited to one.
-> Connection is SACK enabled.
When PTO fires:
new_segment_exists:
-> transmit new segment.
-> packets_out++. cwnd remains same.
no_new_packet:
-> retransmit the last segment.
Its ACK triggers FACK or early retransmit based recovery.
ACK path:
-> rearm RTO at start of ACK processing.
-> reschedule PTO if need be.
In addition, the patch includes a small variation to the Early Retransmit
(ER) algorithm, such that ER and TLP together can in principle recover any
N-degree of tail loss through fast recovery. TLP is controlled by the same
sysctl as ER, tcp_early_retrans sysctl.
tcp_early_retrans==0; disables TLP and ER.
==1; enables RFC5827 ER.
==2; delayed ER.
==3; TLP and delayed ER. [DEFAULT]
==4; TLP only.
The TLP patch series have been extensively tested on Google Web servers.
It is most effective for short Web trasactions, where it reduced RTOs by 15%
and improved HTTP response time (average by 6%, 99th percentile by 10%).
The transmitted probes account for <0.5% of the overall transmissions.
Signed-off-by: Nandita Dukkipati <nanditad@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Acked-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-03-11 17:00:43 +07:00
|
|
|
struct inet_connection_sock *icsk = inet_csk(sk);
|
[TCP]: Sed magic converts func(sk, tp, ...) -> func(sk, ...)
This is (mostly) automated change using magic:
sed -e '/struct sock \*sk/ N' -e '/struct sock \*sk/ N'
-e '/struct sock \*sk/ N' -e '/struct sock \*sk/ N'
-e 's|struct sock \*sk,[\n\t ]*struct tcp_sock \*tp\([^{]*\n{\n\)|
struct sock \*sk\1\tstruct tcp_sock *tp = tcp_sk(sk);\n|g'
-e 's|struct sock \*sk, struct tcp_sock \*tp|
struct sock \*sk|g' -e 's|sk, tp\([^-]\)|sk\1|g'
Fixed four unused variable (tp) warnings that were introduced.
In addition, manually added newlines after local variables and
tweaked function arguments positioning.
$ gcc --version
gcc (GCC) 4.1.1 20060525 (Red Hat 4.1.1-1)
...
$ codiff -fV built-in.o.old built-in.o.new
net/ipv4/route.c:
rt_cache_flush | +14
1 function changed, 14 bytes added
net/ipv4/tcp.c:
tcp_setsockopt | -5
tcp_sendpage | -25
tcp_sendmsg | -16
3 functions changed, 46 bytes removed
net/ipv4/tcp_input.c:
tcp_try_undo_recovery | +3
tcp_try_undo_dsack | +2
tcp_mark_head_lost | -12
tcp_ack | -15
tcp_event_data_recv | -32
tcp_rcv_state_process | -10
tcp_rcv_established | +1
7 functions changed, 6 bytes added, 69 bytes removed, diff: -63
net/ipv4/tcp_output.c:
update_send_head | -9
tcp_transmit_skb | +19
tcp_cwnd_validate | +1
tcp_write_wakeup | -17
__tcp_push_pending_frames | -25
tcp_push_one | -8
tcp_send_fin | -4
7 functions changed, 20 bytes added, 63 bytes removed, diff: -43
built-in.o.new:
18 functions changed, 40 bytes added, 178 bytes removed, diff: -138
Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>
2007-04-21 12:18:02 +07:00
|
|
|
struct tcp_sock *tp = tcp_sk(sk);
|
2007-12-31 19:43:57 +07:00
|
|
|
unsigned int prior_packets = tp->packets_out;
|
[TCP]: Sed magic converts func(sk, tp, ...) -> func(sk, ...)
This is (mostly) automated change using magic:
sed -e '/struct sock \*sk/ N' -e '/struct sock \*sk/ N'
-e '/struct sock \*sk/ N' -e '/struct sock \*sk/ N'
-e 's|struct sock \*sk,[\n\t ]*struct tcp_sock \*tp\([^{]*\n{\n\)|
struct sock \*sk\1\tstruct tcp_sock *tp = tcp_sk(sk);\n|g'
-e 's|struct sock \*sk, struct tcp_sock \*tp|
struct sock \*sk|g' -e 's|sk, tp\([^-]\)|sk\1|g'
Fixed four unused variable (tp) warnings that were introduced.
In addition, manually added newlines after local variables and
tweaked function arguments positioning.
$ gcc --version
gcc (GCC) 4.1.1 20060525 (Red Hat 4.1.1-1)
...
$ codiff -fV built-in.o.old built-in.o.new
net/ipv4/route.c:
rt_cache_flush | +14
1 function changed, 14 bytes added
net/ipv4/tcp.c:
tcp_setsockopt | -5
tcp_sendpage | -25
tcp_sendmsg | -16
3 functions changed, 46 bytes removed
net/ipv4/tcp_input.c:
tcp_try_undo_recovery | +3
tcp_try_undo_dsack | +2
tcp_mark_head_lost | -12
tcp_ack | -15
tcp_event_data_recv | -32
tcp_rcv_state_process | -10
tcp_rcv_established | +1
7 functions changed, 6 bytes added, 69 bytes removed, diff: -63
net/ipv4/tcp_output.c:
update_send_head | -9
tcp_transmit_skb | +19
tcp_cwnd_validate | +1
tcp_write_wakeup | -17
__tcp_push_pending_frames | -25
tcp_push_one | -8
tcp_send_fin | -4
7 functions changed, 20 bytes added, 63 bytes removed, diff: -43
built-in.o.new:
18 functions changed, 40 bytes added, 178 bytes removed, diff: -138
Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>
2007-04-21 12:18:02 +07:00
|
|
|
|
2007-03-08 03:12:44 +07:00
|
|
|
tcp_advance_send_head(sk, skb);
|
2005-04-17 05:20:36 +07:00
|
|
|
tp->snd_nxt = TCP_SKB_CB(skb)->end_seq;
|
2007-11-26 19:17:38 +07:00
|
|
|
|
2007-12-31 19:43:57 +07:00
|
|
|
tp->packets_out += tcp_skb_pcount(skb);
|
tcp: Tail loss probe (TLP)
This patch series implement the Tail loss probe (TLP) algorithm described
in http://tools.ietf.org/html/draft-dukkipati-tcpm-tcp-loss-probe-01. The
first patch implements the basic algorithm.
TLP's goal is to reduce tail latency of short transactions. It achieves
this by converting retransmission timeouts (RTOs) occuring due
to tail losses (losses at end of transactions) into fast recovery.
TLP transmits one packet in two round-trips when a connection is in
Open state and isn't receiving any ACKs. The transmitted packet, aka
loss probe, can be either new or a retransmission. When there is tail
loss, the ACK from a loss probe triggers FACK/early-retransmit based
fast recovery, thus avoiding a costly RTO. In the absence of loss,
there is no change in the connection state.
PTO stands for probe timeout. It is a timer event indicating
that an ACK is overdue and triggers a loss probe packet. The PTO value
is set to max(2*SRTT, 10ms) and is adjusted to account for delayed
ACK timer when there is only one oustanding packet.
TLP Algorithm
On transmission of new data in Open state:
-> packets_out > 1: schedule PTO in max(2*SRTT, 10ms).
-> packets_out == 1: schedule PTO in max(2*RTT, 1.5*RTT + 200ms)
-> PTO = min(PTO, RTO)
Conditions for scheduling PTO:
-> Connection is in Open state.
-> Connection is either cwnd limited or no new data to send.
-> Number of probes per tail loss episode is limited to one.
-> Connection is SACK enabled.
When PTO fires:
new_segment_exists:
-> transmit new segment.
-> packets_out++. cwnd remains same.
no_new_packet:
-> retransmit the last segment.
Its ACK triggers FACK or early retransmit based recovery.
ACK path:
-> rearm RTO at start of ACK processing.
-> reschedule PTO if need be.
In addition, the patch includes a small variation to the Early Retransmit
(ER) algorithm, such that ER and TLP together can in principle recover any
N-degree of tail loss through fast recovery. TLP is controlled by the same
sysctl as ER, tcp_early_retrans sysctl.
tcp_early_retrans==0; disables TLP and ER.
==1; enables RFC5827 ER.
==2; delayed ER.
==3; TLP and delayed ER. [DEFAULT]
==4; TLP only.
The TLP patch series have been extensively tested on Google Web servers.
It is most effective for short Web trasactions, where it reduced RTOs by 15%
and improved HTTP response time (average by 6%, 99th percentile by 10%).
The transmitted probes account for <0.5% of the overall transmissions.
Signed-off-by: Nandita Dukkipati <nanditad@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Acked-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-03-11 17:00:43 +07:00
|
|
|
if (!prior_packets || icsk->icsk_pending == ICSK_TIME_EARLY_RETRANS ||
|
2013-04-29 15:39:56 +07:00
|
|
|
icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) {
|
2012-05-02 20:30:04 +07:00
|
|
|
tcp_rearm_rto(sk);
|
2013-04-29 15:39:56 +07:00
|
|
|
}
|
2014-03-04 03:31:36 +07:00
|
|
|
|
2014-03-07 03:03:17 +07:00
|
|
|
NET_ADD_STATS(sock_net(sk), LINUX_MIB_TCPORIGDATASENT,
|
|
|
|
tcp_skb_pcount(skb));
|
2005-04-17 05:20:36 +07:00
|
|
|
}
|
|
|
|
|
|
|
|
/* SND.NXT, if window was not shrunk.
|
|
|
|
* If window has been shrunk, what should we make? It is not clear at all.
|
|
|
|
* Using SND.UNA we will fail to open window, SND.NXT is out of window. :-(
|
|
|
|
* Anything in between SND.UNA...SND.UNA+SND.WND also can be already
|
|
|
|
* invalid. OK, let's make this for now:
|
|
|
|
*/
|
2011-10-21 16:22:42 +07:00
|
|
|
static inline __u32 tcp_acceptable_seq(const struct sock *sk)
|
2005-04-17 05:20:36 +07:00
|
|
|
{
|
2011-10-21 16:22:42 +07:00
|
|
|
const struct tcp_sock *tp = tcp_sk(sk);
|
[TCP]: Sed magic converts func(sk, tp, ...) -> func(sk, ...)
This is (mostly) automated change using magic:
sed -e '/struct sock \*sk/ N' -e '/struct sock \*sk/ N'
-e '/struct sock \*sk/ N' -e '/struct sock \*sk/ N'
-e 's|struct sock \*sk,[\n\t ]*struct tcp_sock \*tp\([^{]*\n{\n\)|
struct sock \*sk\1\tstruct tcp_sock *tp = tcp_sk(sk);\n|g'
-e 's|struct sock \*sk, struct tcp_sock \*tp|
struct sock \*sk|g' -e 's|sk, tp\([^-]\)|sk\1|g'
Fixed four unused variable (tp) warnings that were introduced.
In addition, manually added newlines after local variables and
tweaked function arguments positioning.
$ gcc --version
gcc (GCC) 4.1.1 20060525 (Red Hat 4.1.1-1)
...
$ codiff -fV built-in.o.old built-in.o.new
net/ipv4/route.c:
rt_cache_flush | +14
1 function changed, 14 bytes added
net/ipv4/tcp.c:
tcp_setsockopt | -5
tcp_sendpage | -25
tcp_sendmsg | -16
3 functions changed, 46 bytes removed
net/ipv4/tcp_input.c:
tcp_try_undo_recovery | +3
tcp_try_undo_dsack | +2
tcp_mark_head_lost | -12
tcp_ack | -15
tcp_event_data_recv | -32
tcp_rcv_state_process | -10
tcp_rcv_established | +1
7 functions changed, 6 bytes added, 69 bytes removed, diff: -63
net/ipv4/tcp_output.c:
update_send_head | -9
tcp_transmit_skb | +19
tcp_cwnd_validate | +1
tcp_write_wakeup | -17
__tcp_push_pending_frames | -25
tcp_push_one | -8
tcp_send_fin | -4
7 functions changed, 20 bytes added, 63 bytes removed, diff: -43
built-in.o.new:
18 functions changed, 40 bytes added, 178 bytes removed, diff: -138
Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>
2007-04-21 12:18:02 +07:00
|
|
|
|
2007-12-31 19:48:41 +07:00
|
|
|
if (!before(tcp_wnd_end(tp), tp->snd_nxt))
|
2005-04-17 05:20:36 +07:00
|
|
|
return tp->snd_nxt;
|
|
|
|
else
|
2007-12-31 19:48:41 +07:00
|
|
|
return tcp_wnd_end(tp);
|
2005-04-17 05:20:36 +07:00
|
|
|
}
|
|
|
|
|
|
|
|
/* Calculate mss to advertise in SYN segment.
|
|
|
|
* RFC1122, RFC1063, draft-ietf-tcpimpl-pmtud-01 state that:
|
|
|
|
*
|
|
|
|
* 1. It is independent of path mtu.
|
|
|
|
* 2. Ideally, it is maximal possible segment size i.e. 65535-40.
|
|
|
|
* 3. For IPv4 it is reasonable to calculate it from maximal MTU of
|
|
|
|
* attached devices, because some buggy hosts are confused by
|
|
|
|
* large MSS.
|
|
|
|
* 4. We do not make 3, we advertise MSS, calculated from first
|
|
|
|
* hop device mtu, but allow to raise it to ip_rt_min_advmss.
|
|
|
|
* This may be overridden via information stored in routing table.
|
|
|
|
* 5. Value 65535 for MSS is valid in IPv6 and means "as large as possible,
|
|
|
|
* probably even Jumbo".
|
|
|
|
*/
|
|
|
|
static __u16 tcp_advertise_mss(struct sock *sk)
|
|
|
|
{
|
|
|
|
struct tcp_sock *tp = tcp_sk(sk);
|
2011-10-21 16:22:42 +07:00
|
|
|
const struct dst_entry *dst = __sk_dst_get(sk);
|
2005-04-17 05:20:36 +07:00
|
|
|
int mss = tp->advmss;
|
|
|
|
|
2010-12-14 03:52:14 +07:00
|
|
|
if (dst) {
|
|
|
|
unsigned int metric = dst_metric_advmss(dst);
|
|
|
|
|
|
|
|
if (metric < mss) {
|
|
|
|
mss = metric;
|
|
|
|
tp->advmss = mss;
|
|
|
|
}
|
2005-04-17 05:20:36 +07:00
|
|
|
}
|
|
|
|
|
|
|
|
return (__u16)mss;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* RFC2861. Reset CWND after idle period longer RTO to "restart window".
|
tcp: fix slow start after idle vs TSO/GSO
slow start after idle might reduce cwnd, but we perform this
after first packet was cooked and sent.
With TSO/GSO, it means that we might send a full TSO packet
even if cwnd should have been reduced to IW10.
Moving the SSAI check in skb_entail() makes sense, because
we slightly reduce number of times this check is done,
especially for large send() and TCP Small queue callbacks from
softirq context.
As Neal pointed out, we also need to perform the check
if/when receive window opens.
Tested:
Following packetdrill test demonstrates the problem
// Test of slow start after idle
`sysctl -q net.ipv4.tcp_slow_start_after_idle=1`
0.000 socket(..., SOCK_STREAM, IPPROTO_TCP) = 3
+0 setsockopt(3, SOL_SOCKET, SO_REUSEADDR, [1], 4) = 0
+0 bind(3, ..., ...) = 0
+0 listen(3, 1) = 0
+0 < S 0:0(0) win 65535 <mss 1000,sackOK,nop,nop,nop,wscale 7>
+0 > S. 0:0(0) ack 1 <mss 1460,nop,nop,sackOK,nop,wscale 6>
+.100 < . 1:1(0) ack 1 win 511
+0 accept(3, ..., ...) = 4
+0 setsockopt(4, SOL_SOCKET, SO_SNDBUF, [200000], 4) = 0
+0 write(4, ..., 26000) = 26000
+0 > . 1:5001(5000) ack 1
+0 > . 5001:10001(5000) ack 1
+0 %{ assert tcpi_snd_cwnd == 10 }%
+.100 < . 1:1(0) ack 10001 win 511
+0 %{ assert tcpi_snd_cwnd == 20, tcpi_snd_cwnd }%
+0 > . 10001:20001(10000) ack 1
+0 > P. 20001:26001(6000) ack 1
+.100 < . 1:1(0) ack 26001 win 511
+0 %{ assert tcpi_snd_cwnd == 36, tcpi_snd_cwnd }%
+4 write(4, ..., 20000) = 20000
// If slow start after idle works properly, we should send 5 MSS here (cwnd/2)
+0 > . 26001:31001(5000) ack 1
+0 %{ assert tcpi_snd_cwnd == 10, tcpi_snd_cwnd }%
+0 > . 31001:36001(5000) ack 1
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Neal Cardwell <ncardwell@google.com>
Cc: Yuchung Cheng <ycheng@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2015-08-22 02:30:00 +07:00
|
|
|
* This is the first part of cwnd validation mechanism.
|
|
|
|
*/
|
|
|
|
void tcp_cwnd_restart(struct sock *sk, s32 delta)
|
2005-04-17 05:20:36 +07:00
|
|
|
{
|
2005-08-10 10:10:42 +07:00
|
|
|
struct tcp_sock *tp = tcp_sk(sk);
|
tcp: fix slow start after idle vs TSO/GSO
slow start after idle might reduce cwnd, but we perform this
after first packet was cooked and sent.
With TSO/GSO, it means that we might send a full TSO packet
even if cwnd should have been reduced to IW10.
Moving the SSAI check in skb_entail() makes sense, because
we slightly reduce number of times this check is done,
especially for large send() and TCP Small queue callbacks from
softirq context.
As Neal pointed out, we also need to perform the check
if/when receive window opens.
Tested:
Following packetdrill test demonstrates the problem
// Test of slow start after idle
`sysctl -q net.ipv4.tcp_slow_start_after_idle=1`
0.000 socket(..., SOCK_STREAM, IPPROTO_TCP) = 3
+0 setsockopt(3, SOL_SOCKET, SO_REUSEADDR, [1], 4) = 0
+0 bind(3, ..., ...) = 0
+0 listen(3, 1) = 0
+0 < S 0:0(0) win 65535 <mss 1000,sackOK,nop,nop,nop,wscale 7>
+0 > S. 0:0(0) ack 1 <mss 1460,nop,nop,sackOK,nop,wscale 6>
+.100 < . 1:1(0) ack 1 win 511
+0 accept(3, ..., ...) = 4
+0 setsockopt(4, SOL_SOCKET, SO_SNDBUF, [200000], 4) = 0
+0 write(4, ..., 26000) = 26000
+0 > . 1:5001(5000) ack 1
+0 > . 5001:10001(5000) ack 1
+0 %{ assert tcpi_snd_cwnd == 10 }%
+.100 < . 1:1(0) ack 10001 win 511
+0 %{ assert tcpi_snd_cwnd == 20, tcpi_snd_cwnd }%
+0 > . 10001:20001(10000) ack 1
+0 > P. 20001:26001(6000) ack 1
+.100 < . 1:1(0) ack 26001 win 511
+0 %{ assert tcpi_snd_cwnd == 36, tcpi_snd_cwnd }%
+4 write(4, ..., 20000) = 20000
// If slow start after idle works properly, we should send 5 MSS here (cwnd/2)
+0 > . 26001:31001(5000) ack 1
+0 %{ assert tcpi_snd_cwnd == 10, tcpi_snd_cwnd }%
+0 > . 31001:36001(5000) ack 1
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Neal Cardwell <ncardwell@google.com>
Cc: Yuchung Cheng <ycheng@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2015-08-22 02:30:00 +07:00
|
|
|
u32 restart_cwnd = tcp_init_cwnd(tp, __sk_dst_get(sk));
|
2005-04-17 05:20:36 +07:00
|
|
|
u32 cwnd = tp->snd_cwnd;
|
|
|
|
|
2005-08-10 14:03:31 +07:00
|
|
|
tcp_ca_event(sk, CA_EVENT_CWND_RESTART);
|
2005-04-17 05:20:36 +07:00
|
|
|
|
2005-08-10 14:03:31 +07:00
|
|
|
tp->snd_ssthresh = tcp_current_ssthresh(sk);
|
2005-04-17 05:20:36 +07:00
|
|
|
restart_cwnd = min(restart_cwnd, cwnd);
|
|
|
|
|
2005-08-10 10:10:42 +07:00
|
|
|
while ((delta -= inet_csk(sk)->icsk_rto) > 0 && cwnd > restart_cwnd)
|
2005-04-17 05:20:36 +07:00
|
|
|
cwnd >>= 1;
|
|
|
|
tp->snd_cwnd = max(cwnd, restart_cwnd);
|
|
|
|
tp->snd_cwnd_stamp = tcp_time_stamp;
|
|
|
|
tp->snd_cwnd_used = 0;
|
|
|
|
}
|
|
|
|
|
2009-07-22 06:00:40 +07:00
|
|
|
/* Congestion state accounting after a packet has been sent. */
|
2006-01-04 07:03:49 +07:00
|
|
|
static void tcp_event_data_sent(struct tcp_sock *tp,
|
2011-10-21 16:22:42 +07:00
|
|
|
struct sock *sk)
|
2005-04-17 05:20:36 +07:00
|
|
|
{
|
2005-08-10 10:10:42 +07:00
|
|
|
struct inet_connection_sock *icsk = inet_csk(sk);
|
|
|
|
const u32 now = tcp_time_stamp;
|
2005-04-17 05:20:36 +07:00
|
|
|
|
2015-09-10 11:54:37 +07:00
|
|
|
if (tcp_packets_in_flight(tp) == 0)
|
|
|
|
tcp_ca_event(sk, CA_EVENT_TX_START);
|
|
|
|
|
2005-04-17 05:20:36 +07:00
|
|
|
tp->lsndtime = now;
|
|
|
|
|
|
|
|
/* If it is a reply for ato after last received
|
|
|
|
* packet, enter pingpong mode.
|
|
|
|
*/
|
2015-07-08 07:12:28 +07:00
|
|
|
if ((u32)(now - icsk->icsk_ack.lrcvtime) < icsk->icsk_ack.ato)
|
|
|
|
icsk->icsk_ack.pingpong = 1;
|
2005-04-17 05:20:36 +07:00
|
|
|
}
|
|
|
|
|
2009-07-22 06:00:40 +07:00
|
|
|
/* Account for an ACK we sent. */
|
2006-01-04 07:03:49 +07:00
|
|
|
static inline void tcp_event_ack_sent(struct sock *sk, unsigned int pkts)
|
2005-04-17 05:20:36 +07:00
|
|
|
{
|
2005-08-10 10:10:42 +07:00
|
|
|
tcp_dec_quickack_mode(sk, pkts);
|
|
|
|
inet_csk_clear_xmit_timer(sk, ICSK_TIME_DACK);
|
2005-04-17 05:20:36 +07:00
|
|
|
}
|
|
|
|
|
2013-06-12 05:35:32 +07:00
|
|
|
|
|
|
|
u32 tcp_default_init_rwnd(u32 mss)
|
|
|
|
{
|
|
|
|
/* Initial receive window should be twice of TCP_INIT_CWND to
|
2013-06-18 20:00:31 +07:00
|
|
|
* enable proper sending of new unsent data during fast recovery
|
2013-06-12 05:35:32 +07:00
|
|
|
* (RFC 3517, Section 4, NextSeg() rule (2)). Further place a
|
|
|
|
* limit when mss is larger than 1460.
|
|
|
|
*/
|
|
|
|
u32 init_rwnd = TCP_INIT_CWND * 2;
|
|
|
|
|
|
|
|
if (mss > 1460)
|
|
|
|
init_rwnd = max((1460 * init_rwnd) / mss, 2U);
|
|
|
|
return init_rwnd;
|
|
|
|
}
|
|
|
|
|
2005-04-17 05:20:36 +07:00
|
|
|
/* Determine a window scaling and initial window to offer.
|
|
|
|
* Based on the assumption that the given amount of space
|
|
|
|
* will be offered. Store the results in the tp structure.
|
|
|
|
* NOTE: for smooth operation initial space offering should
|
|
|
|
* be a multiple of mss if possible. We assume here that mss >= 1.
|
|
|
|
* This MUST be enforced by all callers.
|
|
|
|
*/
|
|
|
|
void tcp_select_initial_window(int __space, __u32 mss,
|
|
|
|
__u32 *rcv_wnd, __u32 *window_clamp,
|
2009-12-15 18:15:28 +07:00
|
|
|
int wscale_ok, __u8 *rcv_wscale,
|
|
|
|
__u32 init_rcv_wnd)
|
2005-04-17 05:20:36 +07:00
|
|
|
{
|
|
|
|
unsigned int space = (__space < 0 ? 0 : __space);
|
|
|
|
|
|
|
|
/* If no clamp set the clamp to the max possible scaled window */
|
|
|
|
if (*window_clamp == 0)
|
|
|
|
(*window_clamp) = (65535 << 14);
|
|
|
|
space = min(*window_clamp, space);
|
|
|
|
|
|
|
|
/* Quantize space offering to a multiple of mss if possible. */
|
|
|
|
if (space > mss)
|
|
|
|
space = (space / mss) * mss;
|
|
|
|
|
|
|
|
/* NOTE: offering an initial window larger than 32767
|
2006-03-21 13:40:29 +07:00
|
|
|
* will break some buggy TCP stacks. If the admin tells us
|
|
|
|
* it is likely we could be speaking with such a buggy stack
|
|
|
|
* we will truncate our initial window offering to 32K-1
|
|
|
|
* unless the remote has sent us a window scaling option,
|
|
|
|
* which we interpret as a sign the remote TCP is not
|
|
|
|
* misinterpreting the window field as a signed quantity.
|
2005-04-17 05:20:36 +07:00
|
|
|
*/
|
2006-03-21 13:40:29 +07:00
|
|
|
if (sysctl_tcp_workaround_signed_windows)
|
|
|
|
(*rcv_wnd) = min(space, MAX_TCP_WINDOW);
|
|
|
|
else
|
|
|
|
(*rcv_wnd) = space;
|
|
|
|
|
2005-04-17 05:20:36 +07:00
|
|
|
(*rcv_wscale) = 0;
|
|
|
|
if (wscale_ok) {
|
|
|
|
/* Set window scaling on max possible window
|
2007-02-09 21:24:47 +07:00
|
|
|
* See RFC1323 for an explanation of the limit to 14
|
2005-04-17 05:20:36 +07:00
|
|
|
*/
|
|
|
|
space = max_t(u32, sysctl_tcp_rmem[2], sysctl_rmem_max);
|
2006-08-22 14:06:11 +07:00
|
|
|
space = min_t(u32, space, *window_clamp);
|
2005-04-17 05:20:36 +07:00
|
|
|
while (space > 65535 && (*rcv_wscale) < 14) {
|
|
|
|
space >>= 1;
|
|
|
|
(*rcv_wscale)++;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2008-01-01 05:57:14 +07:00
|
|
|
if (mss > (1 << *rcv_wscale)) {
|
2013-06-12 05:35:32 +07:00
|
|
|
if (!init_rcv_wnd) /* Use default unless specified otherwise */
|
|
|
|
init_rcv_wnd = tcp_default_init_rwnd(mss);
|
|
|
|
*rcv_wnd = min(*rcv_wnd, init_rcv_wnd * mss);
|
2005-04-17 05:20:36 +07:00
|
|
|
}
|
|
|
|
|
|
|
|
/* Set the clamp no higher than max representable value */
|
|
|
|
(*window_clamp) = min(65535U << (*rcv_wscale), *window_clamp);
|
|
|
|
}
|
2010-07-10 04:22:10 +07:00
|
|
|
EXPORT_SYMBOL(tcp_select_initial_window);
|
2005-04-17 05:20:36 +07:00
|
|
|
|
|
|
|
/* Chose a new window to advertise, update state in tcp_sock for the
|
|
|
|
* socket, and return result with RFC1323 scaling applied. The return
|
|
|
|
* value can be stuffed directly into th->window for an outgoing
|
|
|
|
* frame.
|
|
|
|
*/
|
2006-01-04 07:03:49 +07:00
|
|
|
static u16 tcp_select_window(struct sock *sk)
|
2005-04-17 05:20:36 +07:00
|
|
|
{
|
|
|
|
struct tcp_sock *tp = tcp_sk(sk);
|
2014-02-25 20:34:32 +07:00
|
|
|
u32 old_win = tp->rcv_wnd;
|
2005-04-17 05:20:36 +07:00
|
|
|
u32 cur_win = tcp_receive_window(tp);
|
|
|
|
u32 new_win = __tcp_select_window(sk);
|
|
|
|
|
|
|
|
/* Never shrink the offered window */
|
2007-03-09 11:45:19 +07:00
|
|
|
if (new_win < cur_win) {
|
2005-04-17 05:20:36 +07:00
|
|
|
/* Danger Will Robinson!
|
|
|
|
* Don't update rcv_wup/rcv_wnd here or else
|
|
|
|
* we will not be able to advertise a zero
|
|
|
|
* window in time. --DaveM
|
|
|
|
*
|
|
|
|
* Relax Will Robinson.
|
|
|
|
*/
|
2014-02-25 20:34:32 +07:00
|
|
|
if (new_win == 0)
|
|
|
|
NET_INC_STATS(sock_net(sk),
|
|
|
|
LINUX_MIB_TCPWANTZEROWINDOWADV);
|
[TCP]: Fix shrinking windows with window scaling
When selecting a new window, tcp_select_window() tries not to shrink
the offered window by using the maximum of the remaining offered window
size and the newly calculated window size. The newly calculated window
size is always a multiple of the window scaling factor, the remaining
window size however might not be since it depends on rcv_wup/rcv_nxt.
This means we're effectively shrinking the window when scaling it down.
The dump below shows the problem (scaling factor 2^7):
- Window size of 557 (71296) is advertised, up to 3111907257:
IP 172.2.2.3.33000 > 172.2.2.2.33000: . ack 3111835961 win 557 <...>
- New window size of 514 (65792) is advertised, up to 3111907217, 40 bytes
below the last end:
IP 172.2.2.3.33000 > 172.2.2.2.33000: . 3113575668:3113577116(1448) ack 3111841425 win 514 <...>
The number 40 results from downscaling the remaining window:
3111907257 - 3111841425 = 65832
65832 / 2^7 = 514
65832 % 2^7 = 40
If the sender uses up the entire window before it is shrunk, this can have
chaotic effects on the connection. When sending ACKs, tcp_acceptable_seq()
will notice that the window has been shrunk since tcp_wnd_end() is before
tp->snd_nxt, which makes it choose tcp_wnd_end() as sequence number.
This will fail the receivers checks in tcp_sequence() however since it
is before it's tp->rcv_wup, making it respond with a dupack.
If both sides are in this condition, this leads to a constant flood of
ACKs until the connection times out.
Make sure the window is never shrunk by aligning the remaining window to
the window scaling factor.
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
2008-03-21 06:11:27 +07:00
|
|
|
new_win = ALIGN(cur_win, 1 << tp->rx_opt.rcv_wscale);
|
2005-04-17 05:20:36 +07:00
|
|
|
}
|
|
|
|
tp->rcv_wnd = new_win;
|
|
|
|
tp->rcv_wup = tp->rcv_nxt;
|
|
|
|
|
|
|
|
/* Make sure we do not exceed the maximum possible
|
|
|
|
* scaled window.
|
|
|
|
*/
|
2006-03-21 13:40:29 +07:00
|
|
|
if (!tp->rx_opt.rcv_wscale && sysctl_tcp_workaround_signed_windows)
|
2005-04-17 05:20:36 +07:00
|
|
|
new_win = min(new_win, MAX_TCP_WINDOW);
|
|
|
|
else
|
|
|
|
new_win = min(new_win, (65535U << tp->rx_opt.rcv_wscale));
|
|
|
|
|
|
|
|
/* RFC1323 scaling applied */
|
|
|
|
new_win >>= tp->rx_opt.rcv_wscale;
|
|
|
|
|
|
|
|
/* If we advertise zero window, disable fast path. */
|
2014-02-25 20:34:32 +07:00
|
|
|
if (new_win == 0) {
|
2005-04-17 05:20:36 +07:00
|
|
|
tp->pred_flags = 0;
|
2014-02-25 20:34:32 +07:00
|
|
|
if (old_win)
|
|
|
|
NET_INC_STATS(sock_net(sk),
|
|
|
|
LINUX_MIB_TCPTOZEROWINDOWADV);
|
|
|
|
} else if (old_win == 0) {
|
|
|
|
NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPFROMZEROWINDOWADV);
|
|
|
|
}
|
2005-04-17 05:20:36 +07:00
|
|
|
|
|
|
|
return new_win;
|
|
|
|
}
|
|
|
|
|
2009-07-22 06:00:40 +07:00
|
|
|
/* Packet ECN state for a SYN-ACK */
|
2014-09-29 18:08:30 +07:00
|
|
|
static void tcp_ecn_send_synack(struct sock *sk, struct sk_buff *skb)
|
2007-05-27 16:04:16 +07:00
|
|
|
{
|
2014-09-27 03:37:33 +07:00
|
|
|
const struct tcp_sock *tp = tcp_sk(sk);
|
|
|
|
|
2011-09-28 00:25:05 +07:00
|
|
|
TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_CWR;
|
2008-01-01 05:57:14 +07:00
|
|
|
if (!(tp->ecn_flags & TCP_ECN_OK))
|
2011-09-28 00:25:05 +07:00
|
|
|
TCP_SKB_CB(skb)->tcp_flags &= ~TCPHDR_ECE;
|
2014-09-27 03:37:33 +07:00
|
|
|
else if (tcp_ca_needs_ecn(sk))
|
|
|
|
INET_ECN_xmit(sk);
|
2007-05-27 16:04:16 +07:00
|
|
|
}
|
|
|
|
|
2009-07-22 06:00:40 +07:00
|
|
|
/* Packet ECN state for a SYN. */
|
2014-09-29 18:08:30 +07:00
|
|
|
static void tcp_ecn_send_syn(struct sock *sk, struct sk_buff *skb)
|
2007-05-27 16:04:16 +07:00
|
|
|
{
|
|
|
|
struct tcp_sock *tp = tcp_sk(sk);
|
net: allow setting ecn via routing table
This patch allows to set ECN on a per-route basis in case the sysctl
tcp_ecn is not set to 1. In other words, when ECN is set for specific
routes, it provides a tcp_ecn=1 behaviour for that route while the rest
of the stack acts according to the global settings.
One can use 'ip route change dev $dev $net features ecn' to toggle this.
Having a more fine-grained per-route setting can be beneficial for various
reasons, for example, 1) within data centers, or 2) local ISPs may deploy
ECN support for their own video/streaming services [1], etc.
There was a recent measurement study/paper [2] which scanned the Alexa's
publicly available top million websites list from a vantage point in US,
Europe and Asia:
Half of the Alexa list will now happily use ECN (tcp_ecn=2, most likely
blamed to commit 255cac91c3 ("tcp: extend ECN sysctl to allow server-side
only ECN") ;)); the break in connectivity on-path was found is about
1 in 10,000 cases. Timeouts rather than receiving back RSTs were much
more common in the negotiation phase (and mostly seen in the Alexa
middle band, ranks around 50k-150k): from 12-thousand hosts on which
there _may_ be ECN-linked connection failures, only 79 failed with RST
when _not_ failing with RST when ECN is not requested.
It's unclear though, how much equipment in the wild actually marks CE
when buffers start to fill up.
We thought about a fallback to non-ECN for retransmitted SYNs as another
global option (which could perhaps one day be made default), but as Eric
points out, there's much more work needed to detect broken middleboxes.
Two examples Eric mentioned are buggy firewalls that accept only a single
SYN per flow, and middleboxes that successfully let an ECN flow establish,
but later mark CE for all packets (so cwnd converges to 1).
[1] http://www.ietf.org/proceedings/89/slides/slides-89-tsvarea-1.pdf, p.15
[2] http://ecn.ethz.ch/
Joint work with Daniel Borkmann.
Reference: http://thread.gmane.org/gmane.linux.network/335797
Suggested-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Acked-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Daniel Borkmann <dborkman@redhat.com>
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
2014-11-03 23:35:03 +07:00
|
|
|
bool use_ecn = sock_net(sk)->ipv4.sysctl_tcp_ecn == 1 ||
|
|
|
|
tcp_ca_needs_ecn(sk);
|
|
|
|
|
|
|
|
if (!use_ecn) {
|
|
|
|
const struct dst_entry *dst = __sk_dst_get(sk);
|
|
|
|
|
|
|
|
if (dst && dst_feature(dst, RTAX_FEATURE_ECN))
|
|
|
|
use_ecn = true;
|
|
|
|
}
|
2007-05-27 16:04:16 +07:00
|
|
|
|
|
|
|
tp->ecn_flags = 0;
|
net: allow setting ecn via routing table
This patch allows to set ECN on a per-route basis in case the sysctl
tcp_ecn is not set to 1. In other words, when ECN is set for specific
routes, it provides a tcp_ecn=1 behaviour for that route while the rest
of the stack acts according to the global settings.
One can use 'ip route change dev $dev $net features ecn' to toggle this.
Having a more fine-grained per-route setting can be beneficial for various
reasons, for example, 1) within data centers, or 2) local ISPs may deploy
ECN support for their own video/streaming services [1], etc.
There was a recent measurement study/paper [2] which scanned the Alexa's
publicly available top million websites list from a vantage point in US,
Europe and Asia:
Half of the Alexa list will now happily use ECN (tcp_ecn=2, most likely
blamed to commit 255cac91c3 ("tcp: extend ECN sysctl to allow server-side
only ECN") ;)); the break in connectivity on-path was found is about
1 in 10,000 cases. Timeouts rather than receiving back RSTs were much
more common in the negotiation phase (and mostly seen in the Alexa
middle band, ranks around 50k-150k): from 12-thousand hosts on which
there _may_ be ECN-linked connection failures, only 79 failed with RST
when _not_ failing with RST when ECN is not requested.
It's unclear though, how much equipment in the wild actually marks CE
when buffers start to fill up.
We thought about a fallback to non-ECN for retransmitted SYNs as another
global option (which could perhaps one day be made default), but as Eric
points out, there's much more work needed to detect broken middleboxes.
Two examples Eric mentioned are buggy firewalls that accept only a single
SYN per flow, and middleboxes that successfully let an ECN flow establish,
but later mark CE for all packets (so cwnd converges to 1).
[1] http://www.ietf.org/proceedings/89/slides/slides-89-tsvarea-1.pdf, p.15
[2] http://ecn.ethz.ch/
Joint work with Daniel Borkmann.
Reference: http://thread.gmane.org/gmane.linux.network/335797
Suggested-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Acked-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Daniel Borkmann <dborkman@redhat.com>
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
2014-11-03 23:35:03 +07:00
|
|
|
|
|
|
|
if (use_ecn) {
|
2011-09-28 00:25:05 +07:00
|
|
|
TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_ECE | TCPHDR_CWR;
|
2007-05-27 16:04:16 +07:00
|
|
|
tp->ecn_flags = TCP_ECN_OK;
|
2014-09-27 03:37:33 +07:00
|
|
|
if (tcp_ca_needs_ecn(sk))
|
|
|
|
INET_ECN_xmit(sk);
|
2007-05-27 16:04:16 +07:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
tcp: add rfc3168, section 6.1.1.1. fallback
This work as a follow-up of commit f7b3bec6f516 ("net: allow setting ecn
via routing table") and adds RFC3168 section 6.1.1.1. fallback for outgoing
ECN connections. In other words, this work adds a retry with a non-ECN
setup SYN packet, as suggested from the RFC on the first timeout:
[...] A host that receives no reply to an ECN-setup SYN within the
normal SYN retransmission timeout interval MAY resend the SYN and
any subsequent SYN retransmissions with CWR and ECE cleared. [...]
Schematic client-side view when assuming the server is in tcp_ecn=2 mode,
that is, Linux default since 2009 via commit 255cac91c3c9 ("tcp: extend
ECN sysctl to allow server-side only ECN"):
1) Normal ECN-capable path:
SYN ECE CWR ----->
<----- SYN ACK ECE
ACK ----->
2) Path with broken middlebox, when client has fallback:
SYN ECE CWR ----X crappy middlebox drops packet
(timeout, rtx)
SYN ----->
<----- SYN ACK
ACK ----->
In case we would not have the fallback implemented, the middlebox drop
point would basically end up as:
SYN ECE CWR ----X crappy middlebox drops packet
(timeout, rtx)
SYN ECE CWR ----X crappy middlebox drops packet
(timeout, rtx)
SYN ECE CWR ----X crappy middlebox drops packet
(timeout, rtx)
In any case, it's rather a smaller percentage of sites where there would
occur such additional setup latency: it was found in end of 2014 that ~56%
of IPv4 and 65% of IPv6 servers of Alexa 1 million list would negotiate
ECN (aka tcp_ecn=2 default), 0.42% of these webservers will fail to connect
when trying to negotiate with ECN (tcp_ecn=1) due to timeouts, which the
fallback would mitigate with a slight latency trade-off. Recent related
paper on this topic:
Brian Trammell, Mirja Kühlewind, Damiano Boppart, Iain Learmonth,
Gorry Fairhurst, and Richard Scheffenegger:
"Enabling Internet-Wide Deployment of Explicit Congestion Notification."
Proc. PAM 2015, New York.
http://ecn.ethz.ch/ecn-pam15.pdf
Thus, when net.ipv4.tcp_ecn=1 is being set, the patch will perform RFC3168,
section 6.1.1.1. fallback on timeout. For users explicitly not wanting this
which can be in DC use case, we add a net.ipv4.tcp_ecn_fallback knob that
allows for disabling the fallback.
tp->ecn_flags are not being cleared in tcp_ecn_clear_syn() on output, but
rather we let tcp_ecn_rcv_synack() take that over on input path in case a
SYN ACK ECE was delayed. Thus a spurious SYN retransmission will not prevent
ECN being negotiated eventually in that case.
Reference: https://www.ietf.org/proceedings/92/slides/slides-92-iccrg-1.pdf
Reference: https://www.ietf.org/proceedings/89/slides/slides-89-tsvarea-1.pdf
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Mirja Kühlewind <mirja.kuehlewind@tik.ee.ethz.ch>
Signed-off-by: Brian Trammell <trammell@tik.ee.ethz.ch>
Cc: Eric Dumazet <edumazet@google.com>
Cc: Dave That <dave.taht@gmail.com>
Acked-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2015-05-20 02:04:22 +07:00
|
|
|
static void tcp_ecn_clear_syn(struct sock *sk, struct sk_buff *skb)
|
|
|
|
{
|
|
|
|
if (sock_net(sk)->ipv4.sysctl_tcp_ecn_fallback)
|
|
|
|
/* tp->ecn_flags are cleared at a later point in time when
|
|
|
|
* SYN ACK is ultimatively being received.
|
|
|
|
*/
|
|
|
|
TCP_SKB_CB(skb)->tcp_flags &= ~(TCPHDR_ECE | TCPHDR_CWR);
|
|
|
|
}
|
|
|
|
|
2014-09-29 18:08:30 +07:00
|
|
|
static void
|
2015-09-25 21:39:18 +07:00
|
|
|
tcp_ecn_make_synack(const struct request_sock *req, struct tcphdr *th)
|
2007-05-27 16:04:16 +07:00
|
|
|
{
|
2015-09-25 21:39:18 +07:00
|
|
|
if (inet_rsk(req)->ecn_ok)
|
2007-05-27 16:04:16 +07:00
|
|
|
th->ece = 1;
|
|
|
|
}
|
|
|
|
|
2009-07-22 06:00:40 +07:00
|
|
|
/* Set up ECN state for a packet on a ESTABLISHED socket that is about to
|
|
|
|
* be sent.
|
|
|
|
*/
|
2014-09-29 18:08:30 +07:00
|
|
|
static void tcp_ecn_send(struct sock *sk, struct sk_buff *skb,
|
2007-05-27 16:04:16 +07:00
|
|
|
int tcp_header_len)
|
|
|
|
{
|
|
|
|
struct tcp_sock *tp = tcp_sk(sk);
|
|
|
|
|
|
|
|
if (tp->ecn_flags & TCP_ECN_OK) {
|
|
|
|
/* Not-retransmitted data segment: set ECT and inject CWR. */
|
|
|
|
if (skb->len != tcp_header_len &&
|
|
|
|
!before(TCP_SKB_CB(skb)->seq, tp->snd_nxt)) {
|
|
|
|
INET_ECN_xmit(sk);
|
2008-01-01 05:57:14 +07:00
|
|
|
if (tp->ecn_flags & TCP_ECN_QUEUE_CWR) {
|
2007-05-27 16:04:16 +07:00
|
|
|
tp->ecn_flags &= ~TCP_ECN_QUEUE_CWR;
|
|
|
|
tcp_hdr(skb)->cwr = 1;
|
|
|
|
skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
|
|
|
|
}
|
2014-09-27 03:37:33 +07:00
|
|
|
} else if (!tcp_ca_needs_ecn(sk)) {
|
2007-05-27 16:04:16 +07:00
|
|
|
/* ACK or retransmitted segment: clear ECT|CE */
|
|
|
|
INET_ECN_dontxmit(sk);
|
|
|
|
}
|
|
|
|
if (tp->ecn_flags & TCP_ECN_DEMAND_CWR)
|
|
|
|
tcp_hdr(skb)->ece = 1;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2008-01-04 11:39:01 +07:00
|
|
|
/* Constructs common control bits of non-data skb. If SYN/FIN is present,
|
|
|
|
* auto increment end seqno.
|
|
|
|
*/
|
|
|
|
static void tcp_init_nondata_skb(struct sk_buff *skb, u32 seq, u8 flags)
|
|
|
|
{
|
2010-04-09 01:32:30 +07:00
|
|
|
skb->ip_summed = CHECKSUM_PARTIAL;
|
2008-01-04 11:39:01 +07:00
|
|
|
skb->csum = 0;
|
|
|
|
|
2011-09-28 00:25:05 +07:00
|
|
|
TCP_SKB_CB(skb)->tcp_flags = flags;
|
2008-01-04 11:39:01 +07:00
|
|
|
TCP_SKB_CB(skb)->sacked = 0;
|
|
|
|
|
2014-09-24 18:11:22 +07:00
|
|
|
tcp_skb_pcount_set(skb, 1);
|
2008-01-04 11:39:01 +07:00
|
|
|
|
|
|
|
TCP_SKB_CB(skb)->seq = seq;
|
2010-06-12 21:01:43 +07:00
|
|
|
if (flags & (TCPHDR_SYN | TCPHDR_FIN))
|
2008-01-04 11:39:01 +07:00
|
|
|
seq++;
|
|
|
|
TCP_SKB_CB(skb)->end_seq = seq;
|
|
|
|
}
|
|
|
|
|
2012-05-17 06:15:34 +07:00
|
|
|
static inline bool tcp_urg_mode(const struct tcp_sock *tp)
|
2008-10-08 04:43:06 +07:00
|
|
|
{
|
|
|
|
return tp->snd_una != tp->snd_up;
|
|
|
|
}
|
|
|
|
|
2008-07-19 14:04:31 +07:00
|
|
|
#define OPTION_SACK_ADVERTISE (1 << 0)
|
|
|
|
#define OPTION_TS (1 << 1)
|
|
|
|
#define OPTION_MD5 (1 << 2)
|
2009-10-01 13:41:59 +07:00
|
|
|
#define OPTION_WSCALE (1 << 3)
|
2012-07-19 13:43:05 +07:00
|
|
|
#define OPTION_FAST_OPEN_COOKIE (1 << 8)
|
2008-07-19 14:04:31 +07:00
|
|
|
|
|
|
|
struct tcp_out_options {
|
2012-07-19 13:43:05 +07:00
|
|
|
u16 options; /* bit field of OPTION_* */
|
|
|
|
u16 mss; /* 0 to disable */
|
2008-07-19 14:04:31 +07:00
|
|
|
u8 ws; /* window scale, 0 to disable */
|
|
|
|
u8 num_sack_blocks; /* number of SACK blocks to include */
|
2009-12-03 01:23:05 +07:00
|
|
|
u8 hash_size; /* bytes in hash_location */
|
|
|
|
__u8 *hash_location; /* temporary pointer, overloaded */
|
2012-07-19 13:43:05 +07:00
|
|
|
__u32 tsval, tsecr; /* need to include OPTION_TS */
|
|
|
|
struct tcp_fastopen_cookie *fastopen_cookie; /* Fast open cookie */
|
2008-07-19 14:04:31 +07:00
|
|
|
};
|
|
|
|
|
2009-07-22 06:00:40 +07:00
|
|
|
/* Write previously computed TCP options to the packet.
|
|
|
|
*
|
|
|
|
* Beware: Something in the Internet is very sensitive to the ordering of
|
2008-10-24 04:06:35 +07:00
|
|
|
* TCP options, we learned this through the hard way, so be careful here.
|
|
|
|
* Luckily we can at least blame others for their non-compliance but from
|
2013-12-09 03:15:44 +07:00
|
|
|
* inter-operability perspective it seems that we're somewhat stuck with
|
2008-10-24 04:06:35 +07:00
|
|
|
* the ordering which we have been using if we want to keep working with
|
|
|
|
* those broken things (not that it currently hurts anybody as there isn't
|
|
|
|
* particular reason why the ordering would need to be changed).
|
|
|
|
*
|
|
|
|
* At least SACK_PERM as the first option is known to lead to a disaster
|
|
|
|
* (but it may well be that other scenarios fail similarly).
|
|
|
|
*/
|
2008-07-19 14:04:31 +07:00
|
|
|
static void tcp_options_write(__be32 *ptr, struct tcp_sock *tp,
|
2009-12-03 01:23:05 +07:00
|
|
|
struct tcp_out_options *opts)
|
|
|
|
{
|
2012-07-19 13:43:05 +07:00
|
|
|
u16 options = opts->options; /* mungable copy */
|
2009-12-03 01:23:05 +07:00
|
|
|
|
|
|
|
if (unlikely(OPTION_MD5 & options)) {
|
2013-03-17 15:23:34 +07:00
|
|
|
*ptr++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
|
|
|
|
(TCPOPT_MD5SIG << 8) | TCPOLEN_MD5SIG);
|
2009-12-03 01:23:05 +07:00
|
|
|
/* overload cookie hash location */
|
|
|
|
opts->hash_location = (__u8 *)ptr;
|
2008-07-19 14:04:31 +07:00
|
|
|
ptr += 4;
|
2006-01-04 07:03:49 +07:00
|
|
|
}
|
2008-07-19 14:04:31 +07:00
|
|
|
|
2008-10-24 04:06:35 +07:00
|
|
|
if (unlikely(opts->mss)) {
|
|
|
|
*ptr++ = htonl((TCPOPT_MSS << 24) |
|
|
|
|
(TCPOLEN_MSS << 16) |
|
|
|
|
opts->mss);
|
|
|
|
}
|
|
|
|
|
2009-12-03 01:23:05 +07:00
|
|
|
if (likely(OPTION_TS & options)) {
|
|
|
|
if (unlikely(OPTION_SACK_ADVERTISE & options)) {
|
2008-07-19 14:04:31 +07:00
|
|
|
*ptr++ = htonl((TCPOPT_SACK_PERM << 24) |
|
|
|
|
(TCPOLEN_SACK_PERM << 16) |
|
|
|
|
(TCPOPT_TIMESTAMP << 8) |
|
|
|
|
TCPOLEN_TIMESTAMP);
|
2009-12-03 01:23:05 +07:00
|
|
|
options &= ~OPTION_SACK_ADVERTISE;
|
2008-07-19 14:04:31 +07:00
|
|
|
} else {
|
|
|
|
*ptr++ = htonl((TCPOPT_NOP << 24) |
|
|
|
|
(TCPOPT_NOP << 16) |
|
|
|
|
(TCPOPT_TIMESTAMP << 8) |
|
|
|
|
TCPOLEN_TIMESTAMP);
|
|
|
|
}
|
|
|
|
*ptr++ = htonl(opts->tsval);
|
|
|
|
*ptr++ = htonl(opts->tsecr);
|
|
|
|
}
|
|
|
|
|
2009-12-03 01:23:05 +07:00
|
|
|
if (unlikely(OPTION_SACK_ADVERTISE & options)) {
|
2008-07-19 14:04:31 +07:00
|
|
|
*ptr++ = htonl((TCPOPT_NOP << 24) |
|
|
|
|
(TCPOPT_NOP << 16) |
|
|
|
|
(TCPOPT_SACK_PERM << 8) |
|
|
|
|
TCPOLEN_SACK_PERM);
|
|
|
|
}
|
|
|
|
|
2009-12-03 01:23:05 +07:00
|
|
|
if (unlikely(OPTION_WSCALE & options)) {
|
2008-07-19 14:04:31 +07:00
|
|
|
*ptr++ = htonl((TCPOPT_NOP << 24) |
|
|
|
|
(TCPOPT_WINDOW << 16) |
|
|
|
|
(TCPOLEN_WINDOW << 8) |
|
|
|
|
opts->ws);
|
|
|
|
}
|
|
|
|
|
|
|
|
if (unlikely(opts->num_sack_blocks)) {
|
|
|
|
struct tcp_sack_block *sp = tp->rx_opt.dsack ?
|
|
|
|
tp->duplicate_sack : tp->selective_acks;
|
2006-01-04 07:03:49 +07:00
|
|
|
int this_sack;
|
|
|
|
|
|
|
|
*ptr++ = htonl((TCPOPT_NOP << 24) |
|
|
|
|
(TCPOPT_NOP << 16) |
|
|
|
|
(TCPOPT_SACK << 8) |
|
2008-07-19 14:04:31 +07:00
|
|
|
(TCPOLEN_SACK_BASE + (opts->num_sack_blocks *
|
2006-01-04 07:03:49 +07:00
|
|
|
TCPOLEN_SACK_PERBLOCK)));
|
2007-03-09 11:45:19 +07:00
|
|
|
|
2008-07-19 14:04:31 +07:00
|
|
|
for (this_sack = 0; this_sack < opts->num_sack_blocks;
|
|
|
|
++this_sack) {
|
2006-01-04 07:03:49 +07:00
|
|
|
*ptr++ = htonl(sp[this_sack].start_seq);
|
|
|
|
*ptr++ = htonl(sp[this_sack].end_seq);
|
|
|
|
}
|
2007-03-09 11:45:19 +07:00
|
|
|
|
2009-03-14 21:23:01 +07:00
|
|
|
tp->rx_opt.dsack = 0;
|
2006-01-04 07:03:49 +07:00
|
|
|
}
|
2012-07-19 13:43:05 +07:00
|
|
|
|
|
|
|
if (unlikely(OPTION_FAST_OPEN_COOKIE & options)) {
|
|
|
|
struct tcp_fastopen_cookie *foc = opts->fastopen_cookie;
|
2015-04-07 04:37:26 +07:00
|
|
|
u8 *p = (u8 *)ptr;
|
|
|
|
u32 len; /* Fast Open option length */
|
|
|
|
|
|
|
|
if (foc->exp) {
|
|
|
|
len = TCPOLEN_EXP_FASTOPEN_BASE + foc->len;
|
|
|
|
*ptr = htonl((TCPOPT_EXP << 24) | (len << 16) |
|
|
|
|
TCPOPT_FASTOPEN_MAGIC);
|
|
|
|
p += TCPOLEN_EXP_FASTOPEN_BASE;
|
|
|
|
} else {
|
|
|
|
len = TCPOLEN_FASTOPEN_BASE + foc->len;
|
|
|
|
*p++ = TCPOPT_FASTOPEN;
|
|
|
|
*p++ = len;
|
|
|
|
}
|
2012-07-19 13:43:05 +07:00
|
|
|
|
2015-04-07 04:37:26 +07:00
|
|
|
memcpy(p, foc->val, foc->len);
|
|
|
|
if ((len & 3) == 2) {
|
|
|
|
p[foc->len] = TCPOPT_NOP;
|
|
|
|
p[foc->len + 1] = TCPOPT_NOP;
|
2012-07-19 13:43:05 +07:00
|
|
|
}
|
2015-04-07 04:37:26 +07:00
|
|
|
ptr += (len + 3) >> 2;
|
2012-07-19 13:43:05 +07:00
|
|
|
}
|
2008-07-19 14:04:31 +07:00
|
|
|
}
|
|
|
|
|
2009-07-22 06:00:40 +07:00
|
|
|
/* Compute TCP options for SYN packets. This is not the final
|
|
|
|
* network wire format yet.
|
|
|
|
*/
|
2012-04-15 12:58:06 +07:00
|
|
|
static unsigned int tcp_syn_options(struct sock *sk, struct sk_buff *skb,
|
2008-07-19 14:04:31 +07:00
|
|
|
struct tcp_out_options *opts,
|
2011-10-21 16:22:42 +07:00
|
|
|
struct tcp_md5sig_key **md5)
|
|
|
|
{
|
2008-07-19 14:04:31 +07:00
|
|
|
struct tcp_sock *tp = tcp_sk(sk);
|
2012-04-15 12:58:06 +07:00
|
|
|
unsigned int remaining = MAX_TCP_OPTION_SPACE;
|
2012-07-19 13:43:07 +07:00
|
|
|
struct tcp_fastopen_request *fastopen = tp->fastopen_req;
|
2008-07-19 14:04:31 +07:00
|
|
|
|
2006-11-15 10:07:45 +07:00
|
|
|
#ifdef CONFIG_TCP_MD5SIG
|
2008-07-19 14:04:31 +07:00
|
|
|
*md5 = tp->af_specific->md5_lookup(sk, sk);
|
|
|
|
if (*md5) {
|
|
|
|
opts->options |= OPTION_MD5;
|
2009-12-03 01:23:05 +07:00
|
|
|
remaining -= TCPOLEN_MD5SIG_ALIGNED;
|
2006-11-15 10:07:45 +07:00
|
|
|
}
|
2008-07-19 14:04:31 +07:00
|
|
|
#else
|
|
|
|
*md5 = NULL;
|
2006-11-15 10:07:45 +07:00
|
|
|
#endif
|
2008-07-19 14:04:31 +07:00
|
|
|
|
|
|
|
/* We always get an MSS option. The option bytes which will be seen in
|
|
|
|
* normal data packets should timestamps be used, must be in the MSS
|
|
|
|
* advertised. But we subtract them from tp->mss_cache so that
|
|
|
|
* calculations in tcp_sendmsg are simpler etc. So account for this
|
|
|
|
* fact here if necessary. If we don't do this correctly, as a
|
|
|
|
* receiver we won't recognize data packets as being full sized when we
|
|
|
|
* should, and thus we won't abide by the delayed ACK rules correctly.
|
|
|
|
* SACKs don't matter, we never delay an ACK when we have any of those
|
|
|
|
* going out. */
|
|
|
|
opts->mss = tcp_advertise_mss(sk);
|
2009-12-03 01:23:05 +07:00
|
|
|
remaining -= TCPOLEN_MSS_ALIGNED;
|
2008-07-19 14:04:31 +07:00
|
|
|
|
2015-04-03 15:17:26 +07:00
|
|
|
if (likely(sysctl_tcp_timestamps && !*md5)) {
|
2008-07-19 14:04:31 +07:00
|
|
|
opts->options |= OPTION_TS;
|
2014-09-06 05:33:33 +07:00
|
|
|
opts->tsval = tcp_skb_timestamp(skb) + tp->tsoffset;
|
2008-07-19 14:04:31 +07:00
|
|
|
opts->tsecr = tp->rx_opt.ts_recent;
|
2009-12-03 01:23:05 +07:00
|
|
|
remaining -= TCPOLEN_TSTAMP_ALIGNED;
|
2008-07-19 14:04:31 +07:00
|
|
|
}
|
tcp: Revert per-route SACK/DSACK/TIMESTAMP changes.
It creates a regression, triggering badness for SYN_RECV
sockets, for example:
[19148.022102] Badness at net/ipv4/inet_connection_sock.c:293
[19148.022570] NIP: c02a0914 LR: c02a0904 CTR: 00000000
[19148.023035] REGS: eeecbd30 TRAP: 0700 Not tainted (2.6.32)
[19148.023496] MSR: 00029032 <EE,ME,CE,IR,DR> CR: 24002442 XER: 00000000
[19148.024012] TASK = eee9a820[1756] 'privoxy' THREAD: eeeca000
This is likely caused by the change in the 'estab' parameter
passed to tcp_parse_options() when invoked by the functions
in net/ipv4/tcp_minisocks.c
But even if that is fixed, the ->conn_request() changes made in
this patch series is fundamentally wrong. They try to use the
listening socket's 'dst' to probe the route settings. The
listening socket doesn't even have a route, and you can't
get the right route (the child request one) until much later
after we setup all of the state, and it must be done by hand.
This stuff really isn't ready, so the best thing to do is a
full revert. This reverts the following commits:
f55017a93f1a74d50244b1254b9a2bd7ac9bbf7d
022c3f7d82f0f1c68018696f2f027b87b9bb45c2
1aba721eba1d84a2defce45b950272cee1e6c72a
cda42ebd67ee5fdf09d7057b5a4584d36fe8a335
345cda2fd695534be5a4494f1b59da9daed33663
dc343475ed062e13fc260acccaab91d7d80fd5b2
05eaade2782fb0c90d3034fd7a7d5a16266182bb
6a2a2d6bf8581216e08be15fcb563cfd6c430e1e
Signed-off-by: David S. Miller <davem@davemloft.net>
2009-12-16 11:56:42 +07:00
|
|
|
if (likely(sysctl_tcp_window_scaling)) {
|
2008-07-19 14:04:31 +07:00
|
|
|
opts->ws = tp->rx_opt.rcv_wscale;
|
2009-10-01 13:41:59 +07:00
|
|
|
opts->options |= OPTION_WSCALE;
|
2009-12-03 01:23:05 +07:00
|
|
|
remaining -= TCPOLEN_WSCALE_ALIGNED;
|
2008-07-19 14:04:31 +07:00
|
|
|
}
|
tcp: Revert per-route SACK/DSACK/TIMESTAMP changes.
It creates a regression, triggering badness for SYN_RECV
sockets, for example:
[19148.022102] Badness at net/ipv4/inet_connection_sock.c:293
[19148.022570] NIP: c02a0914 LR: c02a0904 CTR: 00000000
[19148.023035] REGS: eeecbd30 TRAP: 0700 Not tainted (2.6.32)
[19148.023496] MSR: 00029032 <EE,ME,CE,IR,DR> CR: 24002442 XER: 00000000
[19148.024012] TASK = eee9a820[1756] 'privoxy' THREAD: eeeca000
This is likely caused by the change in the 'estab' parameter
passed to tcp_parse_options() when invoked by the functions
in net/ipv4/tcp_minisocks.c
But even if that is fixed, the ->conn_request() changes made in
this patch series is fundamentally wrong. They try to use the
listening socket's 'dst' to probe the route settings. The
listening socket doesn't even have a route, and you can't
get the right route (the child request one) until much later
after we setup all of the state, and it must be done by hand.
This stuff really isn't ready, so the best thing to do is a
full revert. This reverts the following commits:
f55017a93f1a74d50244b1254b9a2bd7ac9bbf7d
022c3f7d82f0f1c68018696f2f027b87b9bb45c2
1aba721eba1d84a2defce45b950272cee1e6c72a
cda42ebd67ee5fdf09d7057b5a4584d36fe8a335
345cda2fd695534be5a4494f1b59da9daed33663
dc343475ed062e13fc260acccaab91d7d80fd5b2
05eaade2782fb0c90d3034fd7a7d5a16266182bb
6a2a2d6bf8581216e08be15fcb563cfd6c430e1e
Signed-off-by: David S. Miller <davem@davemloft.net>
2009-12-16 11:56:42 +07:00
|
|
|
if (likely(sysctl_tcp_sack)) {
|
2008-07-19 14:04:31 +07:00
|
|
|
opts->options |= OPTION_SACK_ADVERTISE;
|
2008-07-22 08:45:34 +07:00
|
|
|
if (unlikely(!(OPTION_TS & opts->options)))
|
2009-12-03 01:23:05 +07:00
|
|
|
remaining -= TCPOLEN_SACKPERM_ALIGNED;
|
2008-07-19 14:04:31 +07:00
|
|
|
}
|
|
|
|
|
2012-07-19 13:43:07 +07:00
|
|
|
if (fastopen && fastopen->cookie.len >= 0) {
|
2015-04-07 04:37:27 +07:00
|
|
|
u32 need = fastopen->cookie.len;
|
|
|
|
|
|
|
|
need += fastopen->cookie.exp ? TCPOLEN_EXP_FASTOPEN_BASE :
|
|
|
|
TCPOLEN_FASTOPEN_BASE;
|
2012-07-19 13:43:07 +07:00
|
|
|
need = (need + 3) & ~3U; /* Align to 32 bits */
|
|
|
|
if (remaining >= need) {
|
|
|
|
opts->options |= OPTION_FAST_OPEN_COOKIE;
|
|
|
|
opts->fastopen_cookie = &fastopen->cookie;
|
|
|
|
remaining -= need;
|
|
|
|
tp->syn_fastopen = 1;
|
2015-04-07 04:37:27 +07:00
|
|
|
tp->syn_fastopen_exp = fastopen->cookie.exp ? 1 : 0;
|
2012-07-19 13:43:07 +07:00
|
|
|
}
|
|
|
|
}
|
2009-12-03 01:23:05 +07:00
|
|
|
|
|
|
|
return MAX_TCP_OPTION_SPACE - remaining;
|
2006-01-04 07:03:49 +07:00
|
|
|
}
|
|
|
|
|
2009-07-22 06:00:40 +07:00
|
|
|
/* Set up TCP options for SYN-ACKs. */
|
2015-09-25 21:39:17 +07:00
|
|
|
static unsigned int tcp_synack_options(struct request_sock *req,
|
|
|
|
unsigned int mss, struct sk_buff *skb,
|
|
|
|
struct tcp_out_options *opts,
|
|
|
|
const struct tcp_md5sig_key *md5,
|
|
|
|
struct tcp_fastopen_cookie *foc)
|
2009-12-03 01:25:27 +07:00
|
|
|
{
|
2008-07-19 14:04:31 +07:00
|
|
|
struct inet_request_sock *ireq = inet_rsk(req);
|
2012-04-15 12:58:06 +07:00
|
|
|
unsigned int remaining = MAX_TCP_OPTION_SPACE;
|
2008-07-19 14:04:31 +07:00
|
|
|
|
2006-11-15 10:07:45 +07:00
|
|
|
#ifdef CONFIG_TCP_MD5SIG
|
2015-03-25 05:58:52 +07:00
|
|
|
if (md5) {
|
2008-07-19 14:04:31 +07:00
|
|
|
opts->options |= OPTION_MD5;
|
2009-12-03 01:25:27 +07:00
|
|
|
remaining -= TCPOLEN_MD5SIG_ALIGNED;
|
|
|
|
|
|
|
|
/* We can't fit any SACK blocks in a packet with MD5 + TS
|
|
|
|
* options. There was discussion about disabling SACK
|
|
|
|
* rather than TS in order to fit in better with old,
|
|
|
|
* buggy kernels, but that was deemed to be unnecessary.
|
|
|
|
*/
|
2010-05-18 12:35:36 +07:00
|
|
|
ireq->tstamp_ok &= !ireq->sack_ok;
|
2006-11-15 10:07:45 +07:00
|
|
|
}
|
|
|
|
#endif
|
2008-07-19 14:04:31 +07:00
|
|
|
|
2009-12-03 01:25:27 +07:00
|
|
|
/* We always send an MSS option. */
|
2008-07-19 14:04:31 +07:00
|
|
|
opts->mss = mss;
|
2009-12-03 01:25:27 +07:00
|
|
|
remaining -= TCPOLEN_MSS_ALIGNED;
|
2008-07-19 14:04:31 +07:00
|
|
|
|
|
|
|
if (likely(ireq->wscale_ok)) {
|
|
|
|
opts->ws = ireq->rcv_wscale;
|
2009-10-01 13:41:59 +07:00
|
|
|
opts->options |= OPTION_WSCALE;
|
2009-12-03 01:25:27 +07:00
|
|
|
remaining -= TCPOLEN_WSCALE_ALIGNED;
|
2008-07-19 14:04:31 +07:00
|
|
|
}
|
2010-05-18 12:35:36 +07:00
|
|
|
if (likely(ireq->tstamp_ok)) {
|
2008-07-19 14:04:31 +07:00
|
|
|
opts->options |= OPTION_TS;
|
2014-09-06 05:33:33 +07:00
|
|
|
opts->tsval = tcp_skb_timestamp(skb);
|
2008-07-19 14:04:31 +07:00
|
|
|
opts->tsecr = req->ts_recent;
|
2009-12-03 01:25:27 +07:00
|
|
|
remaining -= TCPOLEN_TSTAMP_ALIGNED;
|
2008-07-19 14:04:31 +07:00
|
|
|
}
|
|
|
|
if (likely(ireq->sack_ok)) {
|
|
|
|
opts->options |= OPTION_SACK_ADVERTISE;
|
2010-05-18 12:35:36 +07:00
|
|
|
if (unlikely(!ireq->tstamp_ok))
|
2009-12-03 01:25:27 +07:00
|
|
|
remaining -= TCPOLEN_SACKPERM_ALIGNED;
|
2008-07-19 14:04:31 +07:00
|
|
|
}
|
2015-04-07 04:37:26 +07:00
|
|
|
if (foc != NULL && foc->len >= 0) {
|
|
|
|
u32 need = foc->len;
|
|
|
|
|
|
|
|
need += foc->exp ? TCPOLEN_EXP_FASTOPEN_BASE :
|
|
|
|
TCPOLEN_FASTOPEN_BASE;
|
2012-08-31 19:29:12 +07:00
|
|
|
need = (need + 3) & ~3U; /* Align to 32 bits */
|
|
|
|
if (remaining >= need) {
|
|
|
|
opts->options |= OPTION_FAST_OPEN_COOKIE;
|
|
|
|
opts->fastopen_cookie = foc;
|
|
|
|
remaining -= need;
|
|
|
|
}
|
|
|
|
}
|
2013-03-17 15:23:34 +07:00
|
|
|
|
2009-12-03 01:25:27 +07:00
|
|
|
return MAX_TCP_OPTION_SPACE - remaining;
|
2008-07-19 14:04:31 +07:00
|
|
|
}
|
|
|
|
|
2009-07-22 06:00:40 +07:00
|
|
|
/* Compute TCP options for ESTABLISHED sockets. This is not the
|
|
|
|
* final wire format yet.
|
|
|
|
*/
|
2012-04-15 12:58:06 +07:00
|
|
|
static unsigned int tcp_established_options(struct sock *sk, struct sk_buff *skb,
|
2008-07-19 14:04:31 +07:00
|
|
|
struct tcp_out_options *opts,
|
2011-10-21 16:22:42 +07:00
|
|
|
struct tcp_md5sig_key **md5)
|
|
|
|
{
|
2008-07-19 14:04:31 +07:00
|
|
|
struct tcp_sock *tp = tcp_sk(sk);
|
2012-04-15 12:58:06 +07:00
|
|
|
unsigned int size = 0;
|
2009-02-28 11:44:38 +07:00
|
|
|
unsigned int eff_sacks;
|
2008-07-19 14:04:31 +07:00
|
|
|
|
2013-10-01 03:29:11 +07:00
|
|
|
opts->options = 0;
|
|
|
|
|
2008-07-19 14:04:31 +07:00
|
|
|
#ifdef CONFIG_TCP_MD5SIG
|
|
|
|
*md5 = tp->af_specific->md5_lookup(sk, sk);
|
|
|
|
if (unlikely(*md5)) {
|
|
|
|
opts->options |= OPTION_MD5;
|
|
|
|
size += TCPOLEN_MD5SIG_ALIGNED;
|
|
|
|
}
|
|
|
|
#else
|
|
|
|
*md5 = NULL;
|
|
|
|
#endif
|
|
|
|
|
|
|
|
if (likely(tp->rx_opt.tstamp_ok)) {
|
|
|
|
opts->options |= OPTION_TS;
|
2014-09-06 05:33:33 +07:00
|
|
|
opts->tsval = skb ? tcp_skb_timestamp(skb) + tp->tsoffset : 0;
|
2008-07-19 14:04:31 +07:00
|
|
|
opts->tsecr = tp->rx_opt.ts_recent;
|
|
|
|
size += TCPOLEN_TSTAMP_ALIGNED;
|
|
|
|
}
|
|
|
|
|
2009-02-28 11:44:38 +07:00
|
|
|
eff_sacks = tp->rx_opt.num_sacks + tp->rx_opt.dsack;
|
|
|
|
if (unlikely(eff_sacks)) {
|
2012-04-15 12:58:06 +07:00
|
|
|
const unsigned int remaining = MAX_TCP_OPTION_SPACE - size;
|
2008-07-19 14:04:31 +07:00
|
|
|
opts->num_sack_blocks =
|
2012-04-15 12:58:06 +07:00
|
|
|
min_t(unsigned int, eff_sacks,
|
2008-07-19 14:04:31 +07:00
|
|
|
(remaining - TCPOLEN_SACK_BASE_ALIGNED) /
|
|
|
|
TCPOLEN_SACK_PERBLOCK);
|
|
|
|
size += TCPOLEN_SACK_BASE_ALIGNED +
|
|
|
|
opts->num_sack_blocks * TCPOLEN_SACK_PERBLOCK;
|
|
|
|
}
|
|
|
|
|
|
|
|
return size;
|
2006-01-04 07:03:49 +07:00
|
|
|
}
|
2005-04-17 05:20:36 +07:00
|
|
|
|
tcp: TCP Small Queues
This introduce TSQ (TCP Small Queues)
TSQ goal is to reduce number of TCP packets in xmit queues (qdisc &
device queues), to reduce RTT and cwnd bias, part of the bufferbloat
problem.
sk->sk_wmem_alloc not allowed to grow above a given limit,
allowing no more than ~128KB [1] per tcp socket in qdisc/dev layers at a
given time.
TSO packets are sized/capped to half the limit, so that we have two
TSO packets in flight, allowing better bandwidth use.
As a side effect, setting the limit to 40000 automatically reduces the
standard gso max limit (65536) to 40000/2 : It can help to reduce
latencies of high prio packets, having smaller TSO packets.
This means we divert sock_wfree() to a tcp_wfree() handler, to
queue/send following frames when skb_orphan() [2] is called for the
already queued skbs.
Results on my dev machines (tg3/ixgbe nics) are really impressive,
using standard pfifo_fast, and with or without TSO/GSO.
Without reduction of nominal bandwidth, we have reduction of buffering
per bulk sender :
< 1ms on Gbit (instead of 50ms with TSO)
< 8ms on 100Mbit (instead of 132 ms)
I no longer have 4 MBytes backlogged in qdisc by a single netperf
session, and both side socket autotuning no longer use 4 Mbytes.
As skb destructor cannot restart xmit itself ( as qdisc lock might be
taken at this point ), we delegate the work to a tasklet. We use one
tasklest per cpu for performance reasons.
If tasklet finds a socket owned by the user, it sets TSQ_OWNED flag.
This flag is tested in a new protocol method called from release_sock(),
to eventually send new segments.
[1] New /proc/sys/net/ipv4/tcp_limit_output_bytes tunable
[2] skb_orphan() is usually called at TX completion time,
but some drivers call it in their start_xmit() handler.
These drivers should at least use BQL, or else a single TCP
session can still fill the whole NIC TX ring, since TSQ will
have no effect.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Dave Taht <dave.taht@bufferbloat.net>
Cc: Tom Herbert <therbert@google.com>
Cc: Matt Mathis <mattmathis@google.com>
Cc: Yuchung Cheng <ycheng@google.com>
Cc: Nandita Dukkipati <nanditad@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2012-07-11 12:50:31 +07:00
|
|
|
|
|
|
|
/* TCP SMALL QUEUES (TSQ)
|
|
|
|
*
|
|
|
|
* TSQ goal is to keep small amount of skbs per tcp flow in tx queues (qdisc+dev)
|
|
|
|
* to reduce RTT and bufferbloat.
|
|
|
|
* We do this using a special skb destructor (tcp_wfree).
|
|
|
|
*
|
|
|
|
* Its important tcp_wfree() can be replaced by sock_wfree() in the event skb
|
|
|
|
* needs to be reallocated in a driver.
|
2013-12-09 03:15:44 +07:00
|
|
|
* The invariant being skb->truesize subtracted from sk->sk_wmem_alloc
|
tcp: TCP Small Queues
This introduce TSQ (TCP Small Queues)
TSQ goal is to reduce number of TCP packets in xmit queues (qdisc &
device queues), to reduce RTT and cwnd bias, part of the bufferbloat
problem.
sk->sk_wmem_alloc not allowed to grow above a given limit,
allowing no more than ~128KB [1] per tcp socket in qdisc/dev layers at a
given time.
TSO packets are sized/capped to half the limit, so that we have two
TSO packets in flight, allowing better bandwidth use.
As a side effect, setting the limit to 40000 automatically reduces the
standard gso max limit (65536) to 40000/2 : It can help to reduce
latencies of high prio packets, having smaller TSO packets.
This means we divert sock_wfree() to a tcp_wfree() handler, to
queue/send following frames when skb_orphan() [2] is called for the
already queued skbs.
Results on my dev machines (tg3/ixgbe nics) are really impressive,
using standard pfifo_fast, and with or without TSO/GSO.
Without reduction of nominal bandwidth, we have reduction of buffering
per bulk sender :
< 1ms on Gbit (instead of 50ms with TSO)
< 8ms on 100Mbit (instead of 132 ms)
I no longer have 4 MBytes backlogged in qdisc by a single netperf
session, and both side socket autotuning no longer use 4 Mbytes.
As skb destructor cannot restart xmit itself ( as qdisc lock might be
taken at this point ), we delegate the work to a tasklet. We use one
tasklest per cpu for performance reasons.
If tasklet finds a socket owned by the user, it sets TSQ_OWNED flag.
This flag is tested in a new protocol method called from release_sock(),
to eventually send new segments.
[1] New /proc/sys/net/ipv4/tcp_limit_output_bytes tunable
[2] skb_orphan() is usually called at TX completion time,
but some drivers call it in their start_xmit() handler.
These drivers should at least use BQL, or else a single TCP
session can still fill the whole NIC TX ring, since TSQ will
have no effect.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Dave Taht <dave.taht@bufferbloat.net>
Cc: Tom Herbert <therbert@google.com>
Cc: Matt Mathis <mattmathis@google.com>
Cc: Yuchung Cheng <ycheng@google.com>
Cc: Nandita Dukkipati <nanditad@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2012-07-11 12:50:31 +07:00
|
|
|
*
|
|
|
|
* Since transmit from skb destructor is forbidden, we use a tasklet
|
|
|
|
* to process all sockets that eventually need to send more skbs.
|
|
|
|
* We use one tasklet per cpu, with its own queue of sockets.
|
|
|
|
*/
|
|
|
|
struct tsq_tasklet {
|
|
|
|
struct tasklet_struct tasklet;
|
|
|
|
struct list_head head; /* queue of tcp sockets */
|
|
|
|
};
|
|
|
|
static DEFINE_PER_CPU(struct tsq_tasklet, tsq_tasklet);
|
|
|
|
|
2012-07-20 12:45:50 +07:00
|
|
|
static void tcp_tsq_handler(struct sock *sk)
|
|
|
|
{
|
|
|
|
if ((1 << sk->sk_state) &
|
|
|
|
(TCPF_ESTABLISHED | TCPF_FIN_WAIT1 | TCPF_CLOSING |
|
|
|
|
TCPF_CLOSE_WAIT | TCPF_LAST_ACK))
|
2014-02-10 09:40:11 +07:00
|
|
|
tcp_write_xmit(sk, tcp_current_mss(sk), tcp_sk(sk)->nonagle,
|
|
|
|
0, GFP_ATOMIC);
|
2012-07-20 12:45:50 +07:00
|
|
|
}
|
tcp: TCP Small Queues
This introduce TSQ (TCP Small Queues)
TSQ goal is to reduce number of TCP packets in xmit queues (qdisc &
device queues), to reduce RTT and cwnd bias, part of the bufferbloat
problem.
sk->sk_wmem_alloc not allowed to grow above a given limit,
allowing no more than ~128KB [1] per tcp socket in qdisc/dev layers at a
given time.
TSO packets are sized/capped to half the limit, so that we have two
TSO packets in flight, allowing better bandwidth use.
As a side effect, setting the limit to 40000 automatically reduces the
standard gso max limit (65536) to 40000/2 : It can help to reduce
latencies of high prio packets, having smaller TSO packets.
This means we divert sock_wfree() to a tcp_wfree() handler, to
queue/send following frames when skb_orphan() [2] is called for the
already queued skbs.
Results on my dev machines (tg3/ixgbe nics) are really impressive,
using standard pfifo_fast, and with or without TSO/GSO.
Without reduction of nominal bandwidth, we have reduction of buffering
per bulk sender :
< 1ms on Gbit (instead of 50ms with TSO)
< 8ms on 100Mbit (instead of 132 ms)
I no longer have 4 MBytes backlogged in qdisc by a single netperf
session, and both side socket autotuning no longer use 4 Mbytes.
As skb destructor cannot restart xmit itself ( as qdisc lock might be
taken at this point ), we delegate the work to a tasklet. We use one
tasklest per cpu for performance reasons.
If tasklet finds a socket owned by the user, it sets TSQ_OWNED flag.
This flag is tested in a new protocol method called from release_sock(),
to eventually send new segments.
[1] New /proc/sys/net/ipv4/tcp_limit_output_bytes tunable
[2] skb_orphan() is usually called at TX completion time,
but some drivers call it in their start_xmit() handler.
These drivers should at least use BQL, or else a single TCP
session can still fill the whole NIC TX ring, since TSQ will
have no effect.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Dave Taht <dave.taht@bufferbloat.net>
Cc: Tom Herbert <therbert@google.com>
Cc: Matt Mathis <mattmathis@google.com>
Cc: Yuchung Cheng <ycheng@google.com>
Cc: Nandita Dukkipati <nanditad@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2012-07-11 12:50:31 +07:00
|
|
|
/*
|
2013-12-09 03:15:44 +07:00
|
|
|
* One tasklet per cpu tries to send more skbs.
|
tcp: TCP Small Queues
This introduce TSQ (TCP Small Queues)
TSQ goal is to reduce number of TCP packets in xmit queues (qdisc &
device queues), to reduce RTT and cwnd bias, part of the bufferbloat
problem.
sk->sk_wmem_alloc not allowed to grow above a given limit,
allowing no more than ~128KB [1] per tcp socket in qdisc/dev layers at a
given time.
TSO packets are sized/capped to half the limit, so that we have two
TSO packets in flight, allowing better bandwidth use.
As a side effect, setting the limit to 40000 automatically reduces the
standard gso max limit (65536) to 40000/2 : It can help to reduce
latencies of high prio packets, having smaller TSO packets.
This means we divert sock_wfree() to a tcp_wfree() handler, to
queue/send following frames when skb_orphan() [2] is called for the
already queued skbs.
Results on my dev machines (tg3/ixgbe nics) are really impressive,
using standard pfifo_fast, and with or without TSO/GSO.
Without reduction of nominal bandwidth, we have reduction of buffering
per bulk sender :
< 1ms on Gbit (instead of 50ms with TSO)
< 8ms on 100Mbit (instead of 132 ms)
I no longer have 4 MBytes backlogged in qdisc by a single netperf
session, and both side socket autotuning no longer use 4 Mbytes.
As skb destructor cannot restart xmit itself ( as qdisc lock might be
taken at this point ), we delegate the work to a tasklet. We use one
tasklest per cpu for performance reasons.
If tasklet finds a socket owned by the user, it sets TSQ_OWNED flag.
This flag is tested in a new protocol method called from release_sock(),
to eventually send new segments.
[1] New /proc/sys/net/ipv4/tcp_limit_output_bytes tunable
[2] skb_orphan() is usually called at TX completion time,
but some drivers call it in their start_xmit() handler.
These drivers should at least use BQL, or else a single TCP
session can still fill the whole NIC TX ring, since TSQ will
have no effect.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Dave Taht <dave.taht@bufferbloat.net>
Cc: Tom Herbert <therbert@google.com>
Cc: Matt Mathis <mattmathis@google.com>
Cc: Yuchung Cheng <ycheng@google.com>
Cc: Nandita Dukkipati <nanditad@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2012-07-11 12:50:31 +07:00
|
|
|
* We run in tasklet context but need to disable irqs when
|
2013-12-09 03:15:44 +07:00
|
|
|
* transferring tsq->head because tcp_wfree() might
|
tcp: TCP Small Queues
This introduce TSQ (TCP Small Queues)
TSQ goal is to reduce number of TCP packets in xmit queues (qdisc &
device queues), to reduce RTT and cwnd bias, part of the bufferbloat
problem.
sk->sk_wmem_alloc not allowed to grow above a given limit,
allowing no more than ~128KB [1] per tcp socket in qdisc/dev layers at a
given time.
TSO packets are sized/capped to half the limit, so that we have two
TSO packets in flight, allowing better bandwidth use.
As a side effect, setting the limit to 40000 automatically reduces the
standard gso max limit (65536) to 40000/2 : It can help to reduce
latencies of high prio packets, having smaller TSO packets.
This means we divert sock_wfree() to a tcp_wfree() handler, to
queue/send following frames when skb_orphan() [2] is called for the
already queued skbs.
Results on my dev machines (tg3/ixgbe nics) are really impressive,
using standard pfifo_fast, and with or without TSO/GSO.
Without reduction of nominal bandwidth, we have reduction of buffering
per bulk sender :
< 1ms on Gbit (instead of 50ms with TSO)
< 8ms on 100Mbit (instead of 132 ms)
I no longer have 4 MBytes backlogged in qdisc by a single netperf
session, and both side socket autotuning no longer use 4 Mbytes.
As skb destructor cannot restart xmit itself ( as qdisc lock might be
taken at this point ), we delegate the work to a tasklet. We use one
tasklest per cpu for performance reasons.
If tasklet finds a socket owned by the user, it sets TSQ_OWNED flag.
This flag is tested in a new protocol method called from release_sock(),
to eventually send new segments.
[1] New /proc/sys/net/ipv4/tcp_limit_output_bytes tunable
[2] skb_orphan() is usually called at TX completion time,
but some drivers call it in their start_xmit() handler.
These drivers should at least use BQL, or else a single TCP
session can still fill the whole NIC TX ring, since TSQ will
have no effect.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Dave Taht <dave.taht@bufferbloat.net>
Cc: Tom Herbert <therbert@google.com>
Cc: Matt Mathis <mattmathis@google.com>
Cc: Yuchung Cheng <ycheng@google.com>
Cc: Nandita Dukkipati <nanditad@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2012-07-11 12:50:31 +07:00
|
|
|
* interrupt us (non NAPI drivers)
|
|
|
|
*/
|
|
|
|
static void tcp_tasklet_func(unsigned long data)
|
|
|
|
{
|
|
|
|
struct tsq_tasklet *tsq = (struct tsq_tasklet *)data;
|
|
|
|
LIST_HEAD(list);
|
|
|
|
unsigned long flags;
|
|
|
|
struct list_head *q, *n;
|
|
|
|
struct tcp_sock *tp;
|
|
|
|
struct sock *sk;
|
|
|
|
|
|
|
|
local_irq_save(flags);
|
|
|
|
list_splice_init(&tsq->head, &list);
|
|
|
|
local_irq_restore(flags);
|
|
|
|
|
|
|
|
list_for_each_safe(q, n, &list) {
|
|
|
|
tp = list_entry(q, struct tcp_sock, tsq_node);
|
|
|
|
list_del(&tp->tsq_node);
|
|
|
|
|
|
|
|
sk = (struct sock *)tp;
|
|
|
|
bh_lock_sock(sk);
|
|
|
|
|
|
|
|
if (!sock_owned_by_user(sk)) {
|
2012-07-20 12:45:50 +07:00
|
|
|
tcp_tsq_handler(sk);
|
tcp: TCP Small Queues
This introduce TSQ (TCP Small Queues)
TSQ goal is to reduce number of TCP packets in xmit queues (qdisc &
device queues), to reduce RTT and cwnd bias, part of the bufferbloat
problem.
sk->sk_wmem_alloc not allowed to grow above a given limit,
allowing no more than ~128KB [1] per tcp socket in qdisc/dev layers at a
given time.
TSO packets are sized/capped to half the limit, so that we have two
TSO packets in flight, allowing better bandwidth use.
As a side effect, setting the limit to 40000 automatically reduces the
standard gso max limit (65536) to 40000/2 : It can help to reduce
latencies of high prio packets, having smaller TSO packets.
This means we divert sock_wfree() to a tcp_wfree() handler, to
queue/send following frames when skb_orphan() [2] is called for the
already queued skbs.
Results on my dev machines (tg3/ixgbe nics) are really impressive,
using standard pfifo_fast, and with or without TSO/GSO.
Without reduction of nominal bandwidth, we have reduction of buffering
per bulk sender :
< 1ms on Gbit (instead of 50ms with TSO)
< 8ms on 100Mbit (instead of 132 ms)
I no longer have 4 MBytes backlogged in qdisc by a single netperf
session, and both side socket autotuning no longer use 4 Mbytes.
As skb destructor cannot restart xmit itself ( as qdisc lock might be
taken at this point ), we delegate the work to a tasklet. We use one
tasklest per cpu for performance reasons.
If tasklet finds a socket owned by the user, it sets TSQ_OWNED flag.
This flag is tested in a new protocol method called from release_sock(),
to eventually send new segments.
[1] New /proc/sys/net/ipv4/tcp_limit_output_bytes tunable
[2] skb_orphan() is usually called at TX completion time,
but some drivers call it in their start_xmit() handler.
These drivers should at least use BQL, or else a single TCP
session can still fill the whole NIC TX ring, since TSQ will
have no effect.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Dave Taht <dave.taht@bufferbloat.net>
Cc: Tom Herbert <therbert@google.com>
Cc: Matt Mathis <mattmathis@google.com>
Cc: Yuchung Cheng <ycheng@google.com>
Cc: Nandita Dukkipati <nanditad@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2012-07-11 12:50:31 +07:00
|
|
|
} else {
|
|
|
|
/* defer the work to tcp_release_cb() */
|
2012-07-20 12:45:50 +07:00
|
|
|
set_bit(TCP_TSQ_DEFERRED, &tp->tsq_flags);
|
tcp: TCP Small Queues
This introduce TSQ (TCP Small Queues)
TSQ goal is to reduce number of TCP packets in xmit queues (qdisc &
device queues), to reduce RTT and cwnd bias, part of the bufferbloat
problem.
sk->sk_wmem_alloc not allowed to grow above a given limit,
allowing no more than ~128KB [1] per tcp socket in qdisc/dev layers at a
given time.
TSO packets are sized/capped to half the limit, so that we have two
TSO packets in flight, allowing better bandwidth use.
As a side effect, setting the limit to 40000 automatically reduces the
standard gso max limit (65536) to 40000/2 : It can help to reduce
latencies of high prio packets, having smaller TSO packets.
This means we divert sock_wfree() to a tcp_wfree() handler, to
queue/send following frames when skb_orphan() [2] is called for the
already queued skbs.
Results on my dev machines (tg3/ixgbe nics) are really impressive,
using standard pfifo_fast, and with or without TSO/GSO.
Without reduction of nominal bandwidth, we have reduction of buffering
per bulk sender :
< 1ms on Gbit (instead of 50ms with TSO)
< 8ms on 100Mbit (instead of 132 ms)
I no longer have 4 MBytes backlogged in qdisc by a single netperf
session, and both side socket autotuning no longer use 4 Mbytes.
As skb destructor cannot restart xmit itself ( as qdisc lock might be
taken at this point ), we delegate the work to a tasklet. We use one
tasklest per cpu for performance reasons.
If tasklet finds a socket owned by the user, it sets TSQ_OWNED flag.
This flag is tested in a new protocol method called from release_sock(),
to eventually send new segments.
[1] New /proc/sys/net/ipv4/tcp_limit_output_bytes tunable
[2] skb_orphan() is usually called at TX completion time,
but some drivers call it in their start_xmit() handler.
These drivers should at least use BQL, or else a single TCP
session can still fill the whole NIC TX ring, since TSQ will
have no effect.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Dave Taht <dave.taht@bufferbloat.net>
Cc: Tom Herbert <therbert@google.com>
Cc: Matt Mathis <mattmathis@google.com>
Cc: Yuchung Cheng <ycheng@google.com>
Cc: Nandita Dukkipati <nanditad@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2012-07-11 12:50:31 +07:00
|
|
|
}
|
|
|
|
bh_unlock_sock(sk);
|
|
|
|
|
|
|
|
clear_bit(TSQ_QUEUED, &tp->tsq_flags);
|
|
|
|
sk_free(sk);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2012-07-20 12:45:50 +07:00
|
|
|
#define TCP_DEFERRED_ALL ((1UL << TCP_TSQ_DEFERRED) | \
|
|
|
|
(1UL << TCP_WRITE_TIMER_DEFERRED) | \
|
2012-07-23 14:48:52 +07:00
|
|
|
(1UL << TCP_DELACK_TIMER_DEFERRED) | \
|
|
|
|
(1UL << TCP_MTU_REDUCED_DEFERRED))
|
tcp: TCP Small Queues
This introduce TSQ (TCP Small Queues)
TSQ goal is to reduce number of TCP packets in xmit queues (qdisc &
device queues), to reduce RTT and cwnd bias, part of the bufferbloat
problem.
sk->sk_wmem_alloc not allowed to grow above a given limit,
allowing no more than ~128KB [1] per tcp socket in qdisc/dev layers at a
given time.
TSO packets are sized/capped to half the limit, so that we have two
TSO packets in flight, allowing better bandwidth use.
As a side effect, setting the limit to 40000 automatically reduces the
standard gso max limit (65536) to 40000/2 : It can help to reduce
latencies of high prio packets, having smaller TSO packets.
This means we divert sock_wfree() to a tcp_wfree() handler, to
queue/send following frames when skb_orphan() [2] is called for the
already queued skbs.
Results on my dev machines (tg3/ixgbe nics) are really impressive,
using standard pfifo_fast, and with or without TSO/GSO.
Without reduction of nominal bandwidth, we have reduction of buffering
per bulk sender :
< 1ms on Gbit (instead of 50ms with TSO)
< 8ms on 100Mbit (instead of 132 ms)
I no longer have 4 MBytes backlogged in qdisc by a single netperf
session, and both side socket autotuning no longer use 4 Mbytes.
As skb destructor cannot restart xmit itself ( as qdisc lock might be
taken at this point ), we delegate the work to a tasklet. We use one
tasklest per cpu for performance reasons.
If tasklet finds a socket owned by the user, it sets TSQ_OWNED flag.
This flag is tested in a new protocol method called from release_sock(),
to eventually send new segments.
[1] New /proc/sys/net/ipv4/tcp_limit_output_bytes tunable
[2] skb_orphan() is usually called at TX completion time,
but some drivers call it in their start_xmit() handler.
These drivers should at least use BQL, or else a single TCP
session can still fill the whole NIC TX ring, since TSQ will
have no effect.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Dave Taht <dave.taht@bufferbloat.net>
Cc: Tom Herbert <therbert@google.com>
Cc: Matt Mathis <mattmathis@google.com>
Cc: Yuchung Cheng <ycheng@google.com>
Cc: Nandita Dukkipati <nanditad@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2012-07-11 12:50:31 +07:00
|
|
|
/**
|
|
|
|
* tcp_release_cb - tcp release_sock() callback
|
|
|
|
* @sk: socket
|
|
|
|
*
|
|
|
|
* called from release_sock() to perform protocol dependent
|
|
|
|
* actions before socket release.
|
|
|
|
*/
|
|
|
|
void tcp_release_cb(struct sock *sk)
|
|
|
|
{
|
|
|
|
struct tcp_sock *tp = tcp_sk(sk);
|
2012-07-20 12:45:50 +07:00
|
|
|
unsigned long flags, nflags;
|
tcp: TCP Small Queues
This introduce TSQ (TCP Small Queues)
TSQ goal is to reduce number of TCP packets in xmit queues (qdisc &
device queues), to reduce RTT and cwnd bias, part of the bufferbloat
problem.
sk->sk_wmem_alloc not allowed to grow above a given limit,
allowing no more than ~128KB [1] per tcp socket in qdisc/dev layers at a
given time.
TSO packets are sized/capped to half the limit, so that we have two
TSO packets in flight, allowing better bandwidth use.
As a side effect, setting the limit to 40000 automatically reduces the
standard gso max limit (65536) to 40000/2 : It can help to reduce
latencies of high prio packets, having smaller TSO packets.
This means we divert sock_wfree() to a tcp_wfree() handler, to
queue/send following frames when skb_orphan() [2] is called for the
already queued skbs.
Results on my dev machines (tg3/ixgbe nics) are really impressive,
using standard pfifo_fast, and with or without TSO/GSO.
Without reduction of nominal bandwidth, we have reduction of buffering
per bulk sender :
< 1ms on Gbit (instead of 50ms with TSO)
< 8ms on 100Mbit (instead of 132 ms)
I no longer have 4 MBytes backlogged in qdisc by a single netperf
session, and both side socket autotuning no longer use 4 Mbytes.
As skb destructor cannot restart xmit itself ( as qdisc lock might be
taken at this point ), we delegate the work to a tasklet. We use one
tasklest per cpu for performance reasons.
If tasklet finds a socket owned by the user, it sets TSQ_OWNED flag.
This flag is tested in a new protocol method called from release_sock(),
to eventually send new segments.
[1] New /proc/sys/net/ipv4/tcp_limit_output_bytes tunable
[2] skb_orphan() is usually called at TX completion time,
but some drivers call it in their start_xmit() handler.
These drivers should at least use BQL, or else a single TCP
session can still fill the whole NIC TX ring, since TSQ will
have no effect.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Dave Taht <dave.taht@bufferbloat.net>
Cc: Tom Herbert <therbert@google.com>
Cc: Matt Mathis <mattmathis@google.com>
Cc: Yuchung Cheng <ycheng@google.com>
Cc: Nandita Dukkipati <nanditad@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2012-07-11 12:50:31 +07:00
|
|
|
|
2012-07-20 12:45:50 +07:00
|
|
|
/* perform an atomic operation only if at least one flag is set */
|
|
|
|
do {
|
|
|
|
flags = tp->tsq_flags;
|
|
|
|
if (!(flags & TCP_DEFERRED_ALL))
|
|
|
|
return;
|
|
|
|
nflags = flags & ~TCP_DEFERRED_ALL;
|
|
|
|
} while (cmpxchg(&tp->tsq_flags, flags, nflags) != flags);
|
|
|
|
|
|
|
|
if (flags & (1UL << TCP_TSQ_DEFERRED))
|
|
|
|
tcp_tsq_handler(sk);
|
|
|
|
|
tcp: tcp_release_cb() should release socket ownership
Lars Persson reported following deadlock :
-000 |M:0x0:0x802B6AF8(asm) <-- arch_spin_lock
-001 |tcp_v4_rcv(skb = 0x8BD527A0) <-- sk = 0x8BE6B2A0
-002 |ip_local_deliver_finish(skb = 0x8BD527A0)
-003 |__netif_receive_skb_core(skb = 0x8BD527A0, ?)
-004 |netif_receive_skb(skb = 0x8BD527A0)
-005 |elk_poll(napi = 0x8C770500, budget = 64)
-006 |net_rx_action(?)
-007 |__do_softirq()
-008 |do_softirq()
-009 |local_bh_enable()
-010 |tcp_rcv_established(sk = 0x8BE6B2A0, skb = 0x87D3A9E0, th = 0x814EBE14, ?)
-011 |tcp_v4_do_rcv(sk = 0x8BE6B2A0, skb = 0x87D3A9E0)
-012 |tcp_delack_timer_handler(sk = 0x8BE6B2A0)
-013 |tcp_release_cb(sk = 0x8BE6B2A0)
-014 |release_sock(sk = 0x8BE6B2A0)
-015 |tcp_sendmsg(?, sk = 0x8BE6B2A0, ?, ?)
-016 |sock_sendmsg(sock = 0x8518C4C0, msg = 0x87D8DAA8, size = 4096)
-017 |kernel_sendmsg(?, ?, ?, ?, size = 4096)
-018 |smb_send_kvec()
-019 |smb_send_rqst(server = 0x87C4D400, rqst = 0x87D8DBA0)
-020 |cifs_call_async()
-021 |cifs_async_writev(wdata = 0x87FD6580)
-022 |cifs_writepages(mapping = 0x852096E4, wbc = 0x87D8DC88)
-023 |__writeback_single_inode(inode = 0x852095D0, wbc = 0x87D8DC88)
-024 |writeback_sb_inodes(sb = 0x87D6D800, wb = 0x87E4A9C0, work = 0x87D8DD88)
-025 |__writeback_inodes_wb(wb = 0x87E4A9C0, work = 0x87D8DD88)
-026 |wb_writeback(wb = 0x87E4A9C0, work = 0x87D8DD88)
-027 |wb_do_writeback(wb = 0x87E4A9C0, force_wait = 0)
-028 |bdi_writeback_workfn(work = 0x87E4A9CC)
-029 |process_one_work(worker = 0x8B045880, work = 0x87E4A9CC)
-030 |worker_thread(__worker = 0x8B045880)
-031 |kthread(_create = 0x87CADD90)
-032 |ret_from_kernel_thread(asm)
Bug occurs because __tcp_checksum_complete_user() enables BH, assuming
it is running from softirq context.
Lars trace involved a NIC without RX checksum support but other points
are problematic as well, like the prequeue stuff.
Problem is triggered by a timer, that found socket being owned by user.
tcp_release_cb() should call tcp_write_timer_handler() or
tcp_delack_timer_handler() in the appropriate context :
BH disabled and socket lock held, but 'owned' field cleared,
as if they were running from timer handlers.
Fixes: 6f458dfb4092 ("tcp: improve latencies of timer triggered events")
Reported-by: Lars Persson <lars.persson@axis.com>
Tested-by: Lars Persson <lars.persson@axis.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2014-03-10 23:50:11 +07:00
|
|
|
/* Here begins the tricky part :
|
|
|
|
* We are called from release_sock() with :
|
|
|
|
* 1) BH disabled
|
|
|
|
* 2) sk_lock.slock spinlock held
|
|
|
|
* 3) socket owned by us (sk->sk_lock.owned == 1)
|
|
|
|
*
|
|
|
|
* But following code is meant to be called from BH handlers,
|
|
|
|
* so we should keep BH disabled, but early release socket ownership
|
|
|
|
*/
|
|
|
|
sock_release_ownership(sk);
|
|
|
|
|
tcp: fix possible socket refcount problem
Commit 6f458dfb40 (tcp: improve latencies of timer triggered events)
added bug leading to following trace :
[ 2866.131281] IPv4: Attempt to release TCP socket in state 1 ffff880019ec0000
[ 2866.131726]
[ 2866.132188] =========================
[ 2866.132281] [ BUG: held lock freed! ]
[ 2866.132281] 3.6.0-rc1+ #622 Not tainted
[ 2866.132281] -------------------------
[ 2866.132281] kworker/0:1/652 is freeing memory ffff880019ec0000-ffff880019ec0a1f, with a lock still held there!
[ 2866.132281] (sk_lock-AF_INET-RPC){+.+...}, at: [<ffffffff81903619>] tcp_sendmsg+0x29/0xcc6
[ 2866.132281] 4 locks held by kworker/0:1/652:
[ 2866.132281] #0: (rpciod){.+.+.+}, at: [<ffffffff81083567>] process_one_work+0x1de/0x47f
[ 2866.132281] #1: ((&task->u.tk_work)){+.+.+.}, at: [<ffffffff81083567>] process_one_work+0x1de/0x47f
[ 2866.132281] #2: (sk_lock-AF_INET-RPC){+.+...}, at: [<ffffffff81903619>] tcp_sendmsg+0x29/0xcc6
[ 2866.132281] #3: (&icsk->icsk_retransmit_timer){+.-...}, at: [<ffffffff81078017>] run_timer_softirq+0x1ad/0x35f
[ 2866.132281]
[ 2866.132281] stack backtrace:
[ 2866.132281] Pid: 652, comm: kworker/0:1 Not tainted 3.6.0-rc1+ #622
[ 2866.132281] Call Trace:
[ 2866.132281] <IRQ> [<ffffffff810bc527>] debug_check_no_locks_freed+0x112/0x159
[ 2866.132281] [<ffffffff818a0839>] ? __sk_free+0xfd/0x114
[ 2866.132281] [<ffffffff811549fa>] kmem_cache_free+0x6b/0x13a
[ 2866.132281] [<ffffffff818a0839>] __sk_free+0xfd/0x114
[ 2866.132281] [<ffffffff818a08c0>] sk_free+0x1c/0x1e
[ 2866.132281] [<ffffffff81911e1c>] tcp_write_timer+0x51/0x56
[ 2866.132281] [<ffffffff81078082>] run_timer_softirq+0x218/0x35f
[ 2866.132281] [<ffffffff81078017>] ? run_timer_softirq+0x1ad/0x35f
[ 2866.132281] [<ffffffff810f5831>] ? rb_commit+0x58/0x85
[ 2866.132281] [<ffffffff81911dcb>] ? tcp_write_timer_handler+0x148/0x148
[ 2866.132281] [<ffffffff81070bd6>] __do_softirq+0xcb/0x1f9
[ 2866.132281] [<ffffffff81a0a00c>] ? _raw_spin_unlock+0x29/0x2e
[ 2866.132281] [<ffffffff81a1227c>] call_softirq+0x1c/0x30
[ 2866.132281] [<ffffffff81039f38>] do_softirq+0x4a/0xa6
[ 2866.132281] [<ffffffff81070f2b>] irq_exit+0x51/0xad
[ 2866.132281] [<ffffffff81a129cd>] do_IRQ+0x9d/0xb4
[ 2866.132281] [<ffffffff81a0a3ef>] common_interrupt+0x6f/0x6f
[ 2866.132281] <EOI> [<ffffffff8109d006>] ? sched_clock_cpu+0x58/0xd1
[ 2866.132281] [<ffffffff81a0a172>] ? _raw_spin_unlock_irqrestore+0x4c/0x56
[ 2866.132281] [<ffffffff81078692>] mod_timer+0x178/0x1a9
[ 2866.132281] [<ffffffff818a00aa>] sk_reset_timer+0x19/0x26
[ 2866.132281] [<ffffffff8190b2cc>] tcp_rearm_rto+0x99/0xa4
[ 2866.132281] [<ffffffff8190dfba>] tcp_event_new_data_sent+0x6e/0x70
[ 2866.132281] [<ffffffff8190f7ea>] tcp_write_xmit+0x7de/0x8e4
[ 2866.132281] [<ffffffff818a565d>] ? __alloc_skb+0xa0/0x1a1
[ 2866.132281] [<ffffffff8190f952>] __tcp_push_pending_frames+0x2e/0x8a
[ 2866.132281] [<ffffffff81904122>] tcp_sendmsg+0xb32/0xcc6
[ 2866.132281] [<ffffffff819229c2>] inet_sendmsg+0xaa/0xd5
[ 2866.132281] [<ffffffff81922918>] ? inet_autobind+0x5f/0x5f
[ 2866.132281] [<ffffffff810ee7f1>] ? trace_clock_local+0x9/0xb
[ 2866.132281] [<ffffffff8189adab>] sock_sendmsg+0xa3/0xc4
[ 2866.132281] [<ffffffff810f5de6>] ? rb_reserve_next_event+0x26f/0x2d5
[ 2866.132281] [<ffffffff8103e6a9>] ? native_sched_clock+0x29/0x6f
[ 2866.132281] [<ffffffff8103e6f8>] ? sched_clock+0x9/0xd
[ 2866.132281] [<ffffffff810ee7f1>] ? trace_clock_local+0x9/0xb
[ 2866.132281] [<ffffffff8189ae03>] kernel_sendmsg+0x37/0x43
[ 2866.132281] [<ffffffff8199ce49>] xs_send_kvec+0x77/0x80
[ 2866.132281] [<ffffffff8199cec1>] xs_sendpages+0x6f/0x1a0
[ 2866.132281] [<ffffffff8107826d>] ? try_to_del_timer_sync+0x55/0x61
[ 2866.132281] [<ffffffff8199d0d2>] xs_tcp_send_request+0x55/0xf1
[ 2866.132281] [<ffffffff8199bb90>] xprt_transmit+0x89/0x1db
[ 2866.132281] [<ffffffff81999bcd>] ? call_connect+0x3c/0x3c
[ 2866.132281] [<ffffffff81999d92>] call_transmit+0x1c5/0x20e
[ 2866.132281] [<ffffffff819a0d55>] __rpc_execute+0x6f/0x225
[ 2866.132281] [<ffffffff81999bcd>] ? call_connect+0x3c/0x3c
[ 2866.132281] [<ffffffff819a0f33>] rpc_async_schedule+0x28/0x34
[ 2866.132281] [<ffffffff810835d6>] process_one_work+0x24d/0x47f
[ 2866.132281] [<ffffffff81083567>] ? process_one_work+0x1de/0x47f
[ 2866.132281] [<ffffffff819a0f0b>] ? __rpc_execute+0x225/0x225
[ 2866.132281] [<ffffffff81083a6d>] worker_thread+0x236/0x317
[ 2866.132281] [<ffffffff81083837>] ? process_scheduled_works+0x2f/0x2f
[ 2866.132281] [<ffffffff8108b7b8>] kthread+0x9a/0xa2
[ 2866.132281] [<ffffffff81a12184>] kernel_thread_helper+0x4/0x10
[ 2866.132281] [<ffffffff81a0a4b0>] ? retint_restore_args+0x13/0x13
[ 2866.132281] [<ffffffff8108b71e>] ? __init_kthread_worker+0x5a/0x5a
[ 2866.132281] [<ffffffff81a12180>] ? gs_change+0x13/0x13
[ 2866.308506] IPv4: Attempt to release TCP socket in state 1 ffff880019ec0000
[ 2866.309689] =============================================================================
[ 2866.310254] BUG TCP (Not tainted): Object already free
[ 2866.310254] -----------------------------------------------------------------------------
[ 2866.310254]
The bug comes from the fact that timer set in sk_reset_timer() can run
before we actually do the sock_hold(). socket refcount reaches zero and
we free the socket too soon.
timer handler is not allowed to reduce socket refcnt if socket is owned
by the user, or we need to change sk_reset_timer() implementation.
We should take a reference on the socket in case TCP_DELACK_TIMER_DEFERRED
or TCP_DELACK_TIMER_DEFERRED bit are set in tsq_flags
Also fix a typo in tcp_delack_timer(), where TCP_WRITE_TIMER_DEFERRED
was used instead of TCP_DELACK_TIMER_DEFERRED.
For consistency, use same socket refcount change for TCP_MTU_REDUCED_DEFERRED,
even if not fired from a timer.
Reported-by: Fengguang Wu <fengguang.wu@intel.com>
Tested-by: Fengguang Wu <fengguang.wu@intel.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2012-08-20 07:22:46 +07:00
|
|
|
if (flags & (1UL << TCP_WRITE_TIMER_DEFERRED)) {
|
2012-07-20 12:45:50 +07:00
|
|
|
tcp_write_timer_handler(sk);
|
tcp: fix possible socket refcount problem
Commit 6f458dfb40 (tcp: improve latencies of timer triggered events)
added bug leading to following trace :
[ 2866.131281] IPv4: Attempt to release TCP socket in state 1 ffff880019ec0000
[ 2866.131726]
[ 2866.132188] =========================
[ 2866.132281] [ BUG: held lock freed! ]
[ 2866.132281] 3.6.0-rc1+ #622 Not tainted
[ 2866.132281] -------------------------
[ 2866.132281] kworker/0:1/652 is freeing memory ffff880019ec0000-ffff880019ec0a1f, with a lock still held there!
[ 2866.132281] (sk_lock-AF_INET-RPC){+.+...}, at: [<ffffffff81903619>] tcp_sendmsg+0x29/0xcc6
[ 2866.132281] 4 locks held by kworker/0:1/652:
[ 2866.132281] #0: (rpciod){.+.+.+}, at: [<ffffffff81083567>] process_one_work+0x1de/0x47f
[ 2866.132281] #1: ((&task->u.tk_work)){+.+.+.}, at: [<ffffffff81083567>] process_one_work+0x1de/0x47f
[ 2866.132281] #2: (sk_lock-AF_INET-RPC){+.+...}, at: [<ffffffff81903619>] tcp_sendmsg+0x29/0xcc6
[ 2866.132281] #3: (&icsk->icsk_retransmit_timer){+.-...}, at: [<ffffffff81078017>] run_timer_softirq+0x1ad/0x35f
[ 2866.132281]
[ 2866.132281] stack backtrace:
[ 2866.132281] Pid: 652, comm: kworker/0:1 Not tainted 3.6.0-rc1+ #622
[ 2866.132281] Call Trace:
[ 2866.132281] <IRQ> [<ffffffff810bc527>] debug_check_no_locks_freed+0x112/0x159
[ 2866.132281] [<ffffffff818a0839>] ? __sk_free+0xfd/0x114
[ 2866.132281] [<ffffffff811549fa>] kmem_cache_free+0x6b/0x13a
[ 2866.132281] [<ffffffff818a0839>] __sk_free+0xfd/0x114
[ 2866.132281] [<ffffffff818a08c0>] sk_free+0x1c/0x1e
[ 2866.132281] [<ffffffff81911e1c>] tcp_write_timer+0x51/0x56
[ 2866.132281] [<ffffffff81078082>] run_timer_softirq+0x218/0x35f
[ 2866.132281] [<ffffffff81078017>] ? run_timer_softirq+0x1ad/0x35f
[ 2866.132281] [<ffffffff810f5831>] ? rb_commit+0x58/0x85
[ 2866.132281] [<ffffffff81911dcb>] ? tcp_write_timer_handler+0x148/0x148
[ 2866.132281] [<ffffffff81070bd6>] __do_softirq+0xcb/0x1f9
[ 2866.132281] [<ffffffff81a0a00c>] ? _raw_spin_unlock+0x29/0x2e
[ 2866.132281] [<ffffffff81a1227c>] call_softirq+0x1c/0x30
[ 2866.132281] [<ffffffff81039f38>] do_softirq+0x4a/0xa6
[ 2866.132281] [<ffffffff81070f2b>] irq_exit+0x51/0xad
[ 2866.132281] [<ffffffff81a129cd>] do_IRQ+0x9d/0xb4
[ 2866.132281] [<ffffffff81a0a3ef>] common_interrupt+0x6f/0x6f
[ 2866.132281] <EOI> [<ffffffff8109d006>] ? sched_clock_cpu+0x58/0xd1
[ 2866.132281] [<ffffffff81a0a172>] ? _raw_spin_unlock_irqrestore+0x4c/0x56
[ 2866.132281] [<ffffffff81078692>] mod_timer+0x178/0x1a9
[ 2866.132281] [<ffffffff818a00aa>] sk_reset_timer+0x19/0x26
[ 2866.132281] [<ffffffff8190b2cc>] tcp_rearm_rto+0x99/0xa4
[ 2866.132281] [<ffffffff8190dfba>] tcp_event_new_data_sent+0x6e/0x70
[ 2866.132281] [<ffffffff8190f7ea>] tcp_write_xmit+0x7de/0x8e4
[ 2866.132281] [<ffffffff818a565d>] ? __alloc_skb+0xa0/0x1a1
[ 2866.132281] [<ffffffff8190f952>] __tcp_push_pending_frames+0x2e/0x8a
[ 2866.132281] [<ffffffff81904122>] tcp_sendmsg+0xb32/0xcc6
[ 2866.132281] [<ffffffff819229c2>] inet_sendmsg+0xaa/0xd5
[ 2866.132281] [<ffffffff81922918>] ? inet_autobind+0x5f/0x5f
[ 2866.132281] [<ffffffff810ee7f1>] ? trace_clock_local+0x9/0xb
[ 2866.132281] [<ffffffff8189adab>] sock_sendmsg+0xa3/0xc4
[ 2866.132281] [<ffffffff810f5de6>] ? rb_reserve_next_event+0x26f/0x2d5
[ 2866.132281] [<ffffffff8103e6a9>] ? native_sched_clock+0x29/0x6f
[ 2866.132281] [<ffffffff8103e6f8>] ? sched_clock+0x9/0xd
[ 2866.132281] [<ffffffff810ee7f1>] ? trace_clock_local+0x9/0xb
[ 2866.132281] [<ffffffff8189ae03>] kernel_sendmsg+0x37/0x43
[ 2866.132281] [<ffffffff8199ce49>] xs_send_kvec+0x77/0x80
[ 2866.132281] [<ffffffff8199cec1>] xs_sendpages+0x6f/0x1a0
[ 2866.132281] [<ffffffff8107826d>] ? try_to_del_timer_sync+0x55/0x61
[ 2866.132281] [<ffffffff8199d0d2>] xs_tcp_send_request+0x55/0xf1
[ 2866.132281] [<ffffffff8199bb90>] xprt_transmit+0x89/0x1db
[ 2866.132281] [<ffffffff81999bcd>] ? call_connect+0x3c/0x3c
[ 2866.132281] [<ffffffff81999d92>] call_transmit+0x1c5/0x20e
[ 2866.132281] [<ffffffff819a0d55>] __rpc_execute+0x6f/0x225
[ 2866.132281] [<ffffffff81999bcd>] ? call_connect+0x3c/0x3c
[ 2866.132281] [<ffffffff819a0f33>] rpc_async_schedule+0x28/0x34
[ 2866.132281] [<ffffffff810835d6>] process_one_work+0x24d/0x47f
[ 2866.132281] [<ffffffff81083567>] ? process_one_work+0x1de/0x47f
[ 2866.132281] [<ffffffff819a0f0b>] ? __rpc_execute+0x225/0x225
[ 2866.132281] [<ffffffff81083a6d>] worker_thread+0x236/0x317
[ 2866.132281] [<ffffffff81083837>] ? process_scheduled_works+0x2f/0x2f
[ 2866.132281] [<ffffffff8108b7b8>] kthread+0x9a/0xa2
[ 2866.132281] [<ffffffff81a12184>] kernel_thread_helper+0x4/0x10
[ 2866.132281] [<ffffffff81a0a4b0>] ? retint_restore_args+0x13/0x13
[ 2866.132281] [<ffffffff8108b71e>] ? __init_kthread_worker+0x5a/0x5a
[ 2866.132281] [<ffffffff81a12180>] ? gs_change+0x13/0x13
[ 2866.308506] IPv4: Attempt to release TCP socket in state 1 ffff880019ec0000
[ 2866.309689] =============================================================================
[ 2866.310254] BUG TCP (Not tainted): Object already free
[ 2866.310254] -----------------------------------------------------------------------------
[ 2866.310254]
The bug comes from the fact that timer set in sk_reset_timer() can run
before we actually do the sock_hold(). socket refcount reaches zero and
we free the socket too soon.
timer handler is not allowed to reduce socket refcnt if socket is owned
by the user, or we need to change sk_reset_timer() implementation.
We should take a reference on the socket in case TCP_DELACK_TIMER_DEFERRED
or TCP_DELACK_TIMER_DEFERRED bit are set in tsq_flags
Also fix a typo in tcp_delack_timer(), where TCP_WRITE_TIMER_DEFERRED
was used instead of TCP_DELACK_TIMER_DEFERRED.
For consistency, use same socket refcount change for TCP_MTU_REDUCED_DEFERRED,
even if not fired from a timer.
Reported-by: Fengguang Wu <fengguang.wu@intel.com>
Tested-by: Fengguang Wu <fengguang.wu@intel.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2012-08-20 07:22:46 +07:00
|
|
|
__sock_put(sk);
|
|
|
|
}
|
|
|
|
if (flags & (1UL << TCP_DELACK_TIMER_DEFERRED)) {
|
2012-07-20 12:45:50 +07:00
|
|
|
tcp_delack_timer_handler(sk);
|
tcp: fix possible socket refcount problem
Commit 6f458dfb40 (tcp: improve latencies of timer triggered events)
added bug leading to following trace :
[ 2866.131281] IPv4: Attempt to release TCP socket in state 1 ffff880019ec0000
[ 2866.131726]
[ 2866.132188] =========================
[ 2866.132281] [ BUG: held lock freed! ]
[ 2866.132281] 3.6.0-rc1+ #622 Not tainted
[ 2866.132281] -------------------------
[ 2866.132281] kworker/0:1/652 is freeing memory ffff880019ec0000-ffff880019ec0a1f, with a lock still held there!
[ 2866.132281] (sk_lock-AF_INET-RPC){+.+...}, at: [<ffffffff81903619>] tcp_sendmsg+0x29/0xcc6
[ 2866.132281] 4 locks held by kworker/0:1/652:
[ 2866.132281] #0: (rpciod){.+.+.+}, at: [<ffffffff81083567>] process_one_work+0x1de/0x47f
[ 2866.132281] #1: ((&task->u.tk_work)){+.+.+.}, at: [<ffffffff81083567>] process_one_work+0x1de/0x47f
[ 2866.132281] #2: (sk_lock-AF_INET-RPC){+.+...}, at: [<ffffffff81903619>] tcp_sendmsg+0x29/0xcc6
[ 2866.132281] #3: (&icsk->icsk_retransmit_timer){+.-...}, at: [<ffffffff81078017>] run_timer_softirq+0x1ad/0x35f
[ 2866.132281]
[ 2866.132281] stack backtrace:
[ 2866.132281] Pid: 652, comm: kworker/0:1 Not tainted 3.6.0-rc1+ #622
[ 2866.132281] Call Trace:
[ 2866.132281] <IRQ> [<ffffffff810bc527>] debug_check_no_locks_freed+0x112/0x159
[ 2866.132281] [<ffffffff818a0839>] ? __sk_free+0xfd/0x114
[ 2866.132281] [<ffffffff811549fa>] kmem_cache_free+0x6b/0x13a
[ 2866.132281] [<ffffffff818a0839>] __sk_free+0xfd/0x114
[ 2866.132281] [<ffffffff818a08c0>] sk_free+0x1c/0x1e
[ 2866.132281] [<ffffffff81911e1c>] tcp_write_timer+0x51/0x56
[ 2866.132281] [<ffffffff81078082>] run_timer_softirq+0x218/0x35f
[ 2866.132281] [<ffffffff81078017>] ? run_timer_softirq+0x1ad/0x35f
[ 2866.132281] [<ffffffff810f5831>] ? rb_commit+0x58/0x85
[ 2866.132281] [<ffffffff81911dcb>] ? tcp_write_timer_handler+0x148/0x148
[ 2866.132281] [<ffffffff81070bd6>] __do_softirq+0xcb/0x1f9
[ 2866.132281] [<ffffffff81a0a00c>] ? _raw_spin_unlock+0x29/0x2e
[ 2866.132281] [<ffffffff81a1227c>] call_softirq+0x1c/0x30
[ 2866.132281] [<ffffffff81039f38>] do_softirq+0x4a/0xa6
[ 2866.132281] [<ffffffff81070f2b>] irq_exit+0x51/0xad
[ 2866.132281] [<ffffffff81a129cd>] do_IRQ+0x9d/0xb4
[ 2866.132281] [<ffffffff81a0a3ef>] common_interrupt+0x6f/0x6f
[ 2866.132281] <EOI> [<ffffffff8109d006>] ? sched_clock_cpu+0x58/0xd1
[ 2866.132281] [<ffffffff81a0a172>] ? _raw_spin_unlock_irqrestore+0x4c/0x56
[ 2866.132281] [<ffffffff81078692>] mod_timer+0x178/0x1a9
[ 2866.132281] [<ffffffff818a00aa>] sk_reset_timer+0x19/0x26
[ 2866.132281] [<ffffffff8190b2cc>] tcp_rearm_rto+0x99/0xa4
[ 2866.132281] [<ffffffff8190dfba>] tcp_event_new_data_sent+0x6e/0x70
[ 2866.132281] [<ffffffff8190f7ea>] tcp_write_xmit+0x7de/0x8e4
[ 2866.132281] [<ffffffff818a565d>] ? __alloc_skb+0xa0/0x1a1
[ 2866.132281] [<ffffffff8190f952>] __tcp_push_pending_frames+0x2e/0x8a
[ 2866.132281] [<ffffffff81904122>] tcp_sendmsg+0xb32/0xcc6
[ 2866.132281] [<ffffffff819229c2>] inet_sendmsg+0xaa/0xd5
[ 2866.132281] [<ffffffff81922918>] ? inet_autobind+0x5f/0x5f
[ 2866.132281] [<ffffffff810ee7f1>] ? trace_clock_local+0x9/0xb
[ 2866.132281] [<ffffffff8189adab>] sock_sendmsg+0xa3/0xc4
[ 2866.132281] [<ffffffff810f5de6>] ? rb_reserve_next_event+0x26f/0x2d5
[ 2866.132281] [<ffffffff8103e6a9>] ? native_sched_clock+0x29/0x6f
[ 2866.132281] [<ffffffff8103e6f8>] ? sched_clock+0x9/0xd
[ 2866.132281] [<ffffffff810ee7f1>] ? trace_clock_local+0x9/0xb
[ 2866.132281] [<ffffffff8189ae03>] kernel_sendmsg+0x37/0x43
[ 2866.132281] [<ffffffff8199ce49>] xs_send_kvec+0x77/0x80
[ 2866.132281] [<ffffffff8199cec1>] xs_sendpages+0x6f/0x1a0
[ 2866.132281] [<ffffffff8107826d>] ? try_to_del_timer_sync+0x55/0x61
[ 2866.132281] [<ffffffff8199d0d2>] xs_tcp_send_request+0x55/0xf1
[ 2866.132281] [<ffffffff8199bb90>] xprt_transmit+0x89/0x1db
[ 2866.132281] [<ffffffff81999bcd>] ? call_connect+0x3c/0x3c
[ 2866.132281] [<ffffffff81999d92>] call_transmit+0x1c5/0x20e
[ 2866.132281] [<ffffffff819a0d55>] __rpc_execute+0x6f/0x225
[ 2866.132281] [<ffffffff81999bcd>] ? call_connect+0x3c/0x3c
[ 2866.132281] [<ffffffff819a0f33>] rpc_async_schedule+0x28/0x34
[ 2866.132281] [<ffffffff810835d6>] process_one_work+0x24d/0x47f
[ 2866.132281] [<ffffffff81083567>] ? process_one_work+0x1de/0x47f
[ 2866.132281] [<ffffffff819a0f0b>] ? __rpc_execute+0x225/0x225
[ 2866.132281] [<ffffffff81083a6d>] worker_thread+0x236/0x317
[ 2866.132281] [<ffffffff81083837>] ? process_scheduled_works+0x2f/0x2f
[ 2866.132281] [<ffffffff8108b7b8>] kthread+0x9a/0xa2
[ 2866.132281] [<ffffffff81a12184>] kernel_thread_helper+0x4/0x10
[ 2866.132281] [<ffffffff81a0a4b0>] ? retint_restore_args+0x13/0x13
[ 2866.132281] [<ffffffff8108b71e>] ? __init_kthread_worker+0x5a/0x5a
[ 2866.132281] [<ffffffff81a12180>] ? gs_change+0x13/0x13
[ 2866.308506] IPv4: Attempt to release TCP socket in state 1 ffff880019ec0000
[ 2866.309689] =============================================================================
[ 2866.310254] BUG TCP (Not tainted): Object already free
[ 2866.310254] -----------------------------------------------------------------------------
[ 2866.310254]
The bug comes from the fact that timer set in sk_reset_timer() can run
before we actually do the sock_hold(). socket refcount reaches zero and
we free the socket too soon.
timer handler is not allowed to reduce socket refcnt if socket is owned
by the user, or we need to change sk_reset_timer() implementation.
We should take a reference on the socket in case TCP_DELACK_TIMER_DEFERRED
or TCP_DELACK_TIMER_DEFERRED bit are set in tsq_flags
Also fix a typo in tcp_delack_timer(), where TCP_WRITE_TIMER_DEFERRED
was used instead of TCP_DELACK_TIMER_DEFERRED.
For consistency, use same socket refcount change for TCP_MTU_REDUCED_DEFERRED,
even if not fired from a timer.
Reported-by: Fengguang Wu <fengguang.wu@intel.com>
Tested-by: Fengguang Wu <fengguang.wu@intel.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2012-08-20 07:22:46 +07:00
|
|
|
__sock_put(sk);
|
|
|
|
}
|
|
|
|
if (flags & (1UL << TCP_MTU_REDUCED_DEFERRED)) {
|
2014-08-14 23:40:05 +07:00
|
|
|
inet_csk(sk)->icsk_af_ops->mtu_reduced(sk);
|
tcp: fix possible socket refcount problem
Commit 6f458dfb40 (tcp: improve latencies of timer triggered events)
added bug leading to following trace :
[ 2866.131281] IPv4: Attempt to release TCP socket in state 1 ffff880019ec0000
[ 2866.131726]
[ 2866.132188] =========================
[ 2866.132281] [ BUG: held lock freed! ]
[ 2866.132281] 3.6.0-rc1+ #622 Not tainted
[ 2866.132281] -------------------------
[ 2866.132281] kworker/0:1/652 is freeing memory ffff880019ec0000-ffff880019ec0a1f, with a lock still held there!
[ 2866.132281] (sk_lock-AF_INET-RPC){+.+...}, at: [<ffffffff81903619>] tcp_sendmsg+0x29/0xcc6
[ 2866.132281] 4 locks held by kworker/0:1/652:
[ 2866.132281] #0: (rpciod){.+.+.+}, at: [<ffffffff81083567>] process_one_work+0x1de/0x47f
[ 2866.132281] #1: ((&task->u.tk_work)){+.+.+.}, at: [<ffffffff81083567>] process_one_work+0x1de/0x47f
[ 2866.132281] #2: (sk_lock-AF_INET-RPC){+.+...}, at: [<ffffffff81903619>] tcp_sendmsg+0x29/0xcc6
[ 2866.132281] #3: (&icsk->icsk_retransmit_timer){+.-...}, at: [<ffffffff81078017>] run_timer_softirq+0x1ad/0x35f
[ 2866.132281]
[ 2866.132281] stack backtrace:
[ 2866.132281] Pid: 652, comm: kworker/0:1 Not tainted 3.6.0-rc1+ #622
[ 2866.132281] Call Trace:
[ 2866.132281] <IRQ> [<ffffffff810bc527>] debug_check_no_locks_freed+0x112/0x159
[ 2866.132281] [<ffffffff818a0839>] ? __sk_free+0xfd/0x114
[ 2866.132281] [<ffffffff811549fa>] kmem_cache_free+0x6b/0x13a
[ 2866.132281] [<ffffffff818a0839>] __sk_free+0xfd/0x114
[ 2866.132281] [<ffffffff818a08c0>] sk_free+0x1c/0x1e
[ 2866.132281] [<ffffffff81911e1c>] tcp_write_timer+0x51/0x56
[ 2866.132281] [<ffffffff81078082>] run_timer_softirq+0x218/0x35f
[ 2866.132281] [<ffffffff81078017>] ? run_timer_softirq+0x1ad/0x35f
[ 2866.132281] [<ffffffff810f5831>] ? rb_commit+0x58/0x85
[ 2866.132281] [<ffffffff81911dcb>] ? tcp_write_timer_handler+0x148/0x148
[ 2866.132281] [<ffffffff81070bd6>] __do_softirq+0xcb/0x1f9
[ 2866.132281] [<ffffffff81a0a00c>] ? _raw_spin_unlock+0x29/0x2e
[ 2866.132281] [<ffffffff81a1227c>] call_softirq+0x1c/0x30
[ 2866.132281] [<ffffffff81039f38>] do_softirq+0x4a/0xa6
[ 2866.132281] [<ffffffff81070f2b>] irq_exit+0x51/0xad
[ 2866.132281] [<ffffffff81a129cd>] do_IRQ+0x9d/0xb4
[ 2866.132281] [<ffffffff81a0a3ef>] common_interrupt+0x6f/0x6f
[ 2866.132281] <EOI> [<ffffffff8109d006>] ? sched_clock_cpu+0x58/0xd1
[ 2866.132281] [<ffffffff81a0a172>] ? _raw_spin_unlock_irqrestore+0x4c/0x56
[ 2866.132281] [<ffffffff81078692>] mod_timer+0x178/0x1a9
[ 2866.132281] [<ffffffff818a00aa>] sk_reset_timer+0x19/0x26
[ 2866.132281] [<ffffffff8190b2cc>] tcp_rearm_rto+0x99/0xa4
[ 2866.132281] [<ffffffff8190dfba>] tcp_event_new_data_sent+0x6e/0x70
[ 2866.132281] [<ffffffff8190f7ea>] tcp_write_xmit+0x7de/0x8e4
[ 2866.132281] [<ffffffff818a565d>] ? __alloc_skb+0xa0/0x1a1
[ 2866.132281] [<ffffffff8190f952>] __tcp_push_pending_frames+0x2e/0x8a
[ 2866.132281] [<ffffffff81904122>] tcp_sendmsg+0xb32/0xcc6
[ 2866.132281] [<ffffffff819229c2>] inet_sendmsg+0xaa/0xd5
[ 2866.132281] [<ffffffff81922918>] ? inet_autobind+0x5f/0x5f
[ 2866.132281] [<ffffffff810ee7f1>] ? trace_clock_local+0x9/0xb
[ 2866.132281] [<ffffffff8189adab>] sock_sendmsg+0xa3/0xc4
[ 2866.132281] [<ffffffff810f5de6>] ? rb_reserve_next_event+0x26f/0x2d5
[ 2866.132281] [<ffffffff8103e6a9>] ? native_sched_clock+0x29/0x6f
[ 2866.132281] [<ffffffff8103e6f8>] ? sched_clock+0x9/0xd
[ 2866.132281] [<ffffffff810ee7f1>] ? trace_clock_local+0x9/0xb
[ 2866.132281] [<ffffffff8189ae03>] kernel_sendmsg+0x37/0x43
[ 2866.132281] [<ffffffff8199ce49>] xs_send_kvec+0x77/0x80
[ 2866.132281] [<ffffffff8199cec1>] xs_sendpages+0x6f/0x1a0
[ 2866.132281] [<ffffffff8107826d>] ? try_to_del_timer_sync+0x55/0x61
[ 2866.132281] [<ffffffff8199d0d2>] xs_tcp_send_request+0x55/0xf1
[ 2866.132281] [<ffffffff8199bb90>] xprt_transmit+0x89/0x1db
[ 2866.132281] [<ffffffff81999bcd>] ? call_connect+0x3c/0x3c
[ 2866.132281] [<ffffffff81999d92>] call_transmit+0x1c5/0x20e
[ 2866.132281] [<ffffffff819a0d55>] __rpc_execute+0x6f/0x225
[ 2866.132281] [<ffffffff81999bcd>] ? call_connect+0x3c/0x3c
[ 2866.132281] [<ffffffff819a0f33>] rpc_async_schedule+0x28/0x34
[ 2866.132281] [<ffffffff810835d6>] process_one_work+0x24d/0x47f
[ 2866.132281] [<ffffffff81083567>] ? process_one_work+0x1de/0x47f
[ 2866.132281] [<ffffffff819a0f0b>] ? __rpc_execute+0x225/0x225
[ 2866.132281] [<ffffffff81083a6d>] worker_thread+0x236/0x317
[ 2866.132281] [<ffffffff81083837>] ? process_scheduled_works+0x2f/0x2f
[ 2866.132281] [<ffffffff8108b7b8>] kthread+0x9a/0xa2
[ 2866.132281] [<ffffffff81a12184>] kernel_thread_helper+0x4/0x10
[ 2866.132281] [<ffffffff81a0a4b0>] ? retint_restore_args+0x13/0x13
[ 2866.132281] [<ffffffff8108b71e>] ? __init_kthread_worker+0x5a/0x5a
[ 2866.132281] [<ffffffff81a12180>] ? gs_change+0x13/0x13
[ 2866.308506] IPv4: Attempt to release TCP socket in state 1 ffff880019ec0000
[ 2866.309689] =============================================================================
[ 2866.310254] BUG TCP (Not tainted): Object already free
[ 2866.310254] -----------------------------------------------------------------------------
[ 2866.310254]
The bug comes from the fact that timer set in sk_reset_timer() can run
before we actually do the sock_hold(). socket refcount reaches zero and
we free the socket too soon.
timer handler is not allowed to reduce socket refcnt if socket is owned
by the user, or we need to change sk_reset_timer() implementation.
We should take a reference on the socket in case TCP_DELACK_TIMER_DEFERRED
or TCP_DELACK_TIMER_DEFERRED bit are set in tsq_flags
Also fix a typo in tcp_delack_timer(), where TCP_WRITE_TIMER_DEFERRED
was used instead of TCP_DELACK_TIMER_DEFERRED.
For consistency, use same socket refcount change for TCP_MTU_REDUCED_DEFERRED,
even if not fired from a timer.
Reported-by: Fengguang Wu <fengguang.wu@intel.com>
Tested-by: Fengguang Wu <fengguang.wu@intel.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2012-08-20 07:22:46 +07:00
|
|
|
__sock_put(sk);
|
|
|
|
}
|
tcp: TCP Small Queues
This introduce TSQ (TCP Small Queues)
TSQ goal is to reduce number of TCP packets in xmit queues (qdisc &
device queues), to reduce RTT and cwnd bias, part of the bufferbloat
problem.
sk->sk_wmem_alloc not allowed to grow above a given limit,
allowing no more than ~128KB [1] per tcp socket in qdisc/dev layers at a
given time.
TSO packets are sized/capped to half the limit, so that we have two
TSO packets in flight, allowing better bandwidth use.
As a side effect, setting the limit to 40000 automatically reduces the
standard gso max limit (65536) to 40000/2 : It can help to reduce
latencies of high prio packets, having smaller TSO packets.
This means we divert sock_wfree() to a tcp_wfree() handler, to
queue/send following frames when skb_orphan() [2] is called for the
already queued skbs.
Results on my dev machines (tg3/ixgbe nics) are really impressive,
using standard pfifo_fast, and with or without TSO/GSO.
Without reduction of nominal bandwidth, we have reduction of buffering
per bulk sender :
< 1ms on Gbit (instead of 50ms with TSO)
< 8ms on 100Mbit (instead of 132 ms)
I no longer have 4 MBytes backlogged in qdisc by a single netperf
session, and both side socket autotuning no longer use 4 Mbytes.
As skb destructor cannot restart xmit itself ( as qdisc lock might be
taken at this point ), we delegate the work to a tasklet. We use one
tasklest per cpu for performance reasons.
If tasklet finds a socket owned by the user, it sets TSQ_OWNED flag.
This flag is tested in a new protocol method called from release_sock(),
to eventually send new segments.
[1] New /proc/sys/net/ipv4/tcp_limit_output_bytes tunable
[2] skb_orphan() is usually called at TX completion time,
but some drivers call it in their start_xmit() handler.
These drivers should at least use BQL, or else a single TCP
session can still fill the whole NIC TX ring, since TSQ will
have no effect.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Dave Taht <dave.taht@bufferbloat.net>
Cc: Tom Herbert <therbert@google.com>
Cc: Matt Mathis <mattmathis@google.com>
Cc: Yuchung Cheng <ycheng@google.com>
Cc: Nandita Dukkipati <nanditad@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2012-07-11 12:50:31 +07:00
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(tcp_release_cb);
|
|
|
|
|
|
|
|
void __init tcp_tasklet_init(void)
|
|
|
|
{
|
|
|
|
int i;
|
|
|
|
|
|
|
|
for_each_possible_cpu(i) {
|
|
|
|
struct tsq_tasklet *tsq = &per_cpu(tsq_tasklet, i);
|
|
|
|
|
|
|
|
INIT_LIST_HEAD(&tsq->head);
|
|
|
|
tasklet_init(&tsq->tasklet,
|
|
|
|
tcp_tasklet_func,
|
|
|
|
(unsigned long)tsq);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Write buffer destructor automatically called from kfree_skb.
|
2013-12-09 03:15:44 +07:00
|
|
|
* We can't xmit new skbs from this context, as we might already
|
tcp: TCP Small Queues
This introduce TSQ (TCP Small Queues)
TSQ goal is to reduce number of TCP packets in xmit queues (qdisc &
device queues), to reduce RTT and cwnd bias, part of the bufferbloat
problem.
sk->sk_wmem_alloc not allowed to grow above a given limit,
allowing no more than ~128KB [1] per tcp socket in qdisc/dev layers at a
given time.
TSO packets are sized/capped to half the limit, so that we have two
TSO packets in flight, allowing better bandwidth use.
As a side effect, setting the limit to 40000 automatically reduces the
standard gso max limit (65536) to 40000/2 : It can help to reduce
latencies of high prio packets, having smaller TSO packets.
This means we divert sock_wfree() to a tcp_wfree() handler, to
queue/send following frames when skb_orphan() [2] is called for the
already queued skbs.
Results on my dev machines (tg3/ixgbe nics) are really impressive,
using standard pfifo_fast, and with or without TSO/GSO.
Without reduction of nominal bandwidth, we have reduction of buffering
per bulk sender :
< 1ms on Gbit (instead of 50ms with TSO)
< 8ms on 100Mbit (instead of 132 ms)
I no longer have 4 MBytes backlogged in qdisc by a single netperf
session, and both side socket autotuning no longer use 4 Mbytes.
As skb destructor cannot restart xmit itself ( as qdisc lock might be
taken at this point ), we delegate the work to a tasklet. We use one
tasklest per cpu for performance reasons.
If tasklet finds a socket owned by the user, it sets TSQ_OWNED flag.
This flag is tested in a new protocol method called from release_sock(),
to eventually send new segments.
[1] New /proc/sys/net/ipv4/tcp_limit_output_bytes tunable
[2] skb_orphan() is usually called at TX completion time,
but some drivers call it in their start_xmit() handler.
These drivers should at least use BQL, or else a single TCP
session can still fill the whole NIC TX ring, since TSQ will
have no effect.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Dave Taht <dave.taht@bufferbloat.net>
Cc: Tom Herbert <therbert@google.com>
Cc: Matt Mathis <mattmathis@google.com>
Cc: Yuchung Cheng <ycheng@google.com>
Cc: Nandita Dukkipati <nanditad@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2012-07-11 12:50:31 +07:00
|
|
|
* hold qdisc lock.
|
|
|
|
*/
|
2013-04-12 18:31:52 +07:00
|
|
|
void tcp_wfree(struct sk_buff *skb)
|
tcp: TCP Small Queues
This introduce TSQ (TCP Small Queues)
TSQ goal is to reduce number of TCP packets in xmit queues (qdisc &
device queues), to reduce RTT and cwnd bias, part of the bufferbloat
problem.
sk->sk_wmem_alloc not allowed to grow above a given limit,
allowing no more than ~128KB [1] per tcp socket in qdisc/dev layers at a
given time.
TSO packets are sized/capped to half the limit, so that we have two
TSO packets in flight, allowing better bandwidth use.
As a side effect, setting the limit to 40000 automatically reduces the
standard gso max limit (65536) to 40000/2 : It can help to reduce
latencies of high prio packets, having smaller TSO packets.
This means we divert sock_wfree() to a tcp_wfree() handler, to
queue/send following frames when skb_orphan() [2] is called for the
already queued skbs.
Results on my dev machines (tg3/ixgbe nics) are really impressive,
using standard pfifo_fast, and with or without TSO/GSO.
Without reduction of nominal bandwidth, we have reduction of buffering
per bulk sender :
< 1ms on Gbit (instead of 50ms with TSO)
< 8ms on 100Mbit (instead of 132 ms)
I no longer have 4 MBytes backlogged in qdisc by a single netperf
session, and both side socket autotuning no longer use 4 Mbytes.
As skb destructor cannot restart xmit itself ( as qdisc lock might be
taken at this point ), we delegate the work to a tasklet. We use one
tasklest per cpu for performance reasons.
If tasklet finds a socket owned by the user, it sets TSQ_OWNED flag.
This flag is tested in a new protocol method called from release_sock(),
to eventually send new segments.
[1] New /proc/sys/net/ipv4/tcp_limit_output_bytes tunable
[2] skb_orphan() is usually called at TX completion time,
but some drivers call it in their start_xmit() handler.
These drivers should at least use BQL, or else a single TCP
session can still fill the whole NIC TX ring, since TSQ will
have no effect.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Dave Taht <dave.taht@bufferbloat.net>
Cc: Tom Herbert <therbert@google.com>
Cc: Matt Mathis <mattmathis@google.com>
Cc: Yuchung Cheng <ycheng@google.com>
Cc: Nandita Dukkipati <nanditad@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2012-07-11 12:50:31 +07:00
|
|
|
{
|
|
|
|
struct sock *sk = skb->sk;
|
|
|
|
struct tcp_sock *tp = tcp_sk(sk);
|
2014-10-13 20:27:47 +07:00
|
|
|
int wmem;
|
|
|
|
|
|
|
|
/* Keep one reference on sk_wmem_alloc.
|
|
|
|
* Will be released by sk_free() from here or tcp_tasklet_func()
|
|
|
|
*/
|
|
|
|
wmem = atomic_sub_return(skb->truesize - 1, &sk->sk_wmem_alloc);
|
|
|
|
|
|
|
|
/* If this softirq is serviced by ksoftirqd, we are likely under stress.
|
|
|
|
* Wait until our queues (qdisc + devices) are drained.
|
|
|
|
* This gives :
|
|
|
|
* - less callbacks to tcp_write_xmit(), reducing stress (batches)
|
|
|
|
* - chance for incoming ACK (processed by another cpu maybe)
|
|
|
|
* to migrate this flow (skb->ooo_okay will be eventually set)
|
|
|
|
*/
|
|
|
|
if (wmem >= SKB_TRUESIZE(1) && this_cpu_ksoftirqd() == current)
|
|
|
|
goto out;
|
tcp: TCP Small Queues
This introduce TSQ (TCP Small Queues)
TSQ goal is to reduce number of TCP packets in xmit queues (qdisc &
device queues), to reduce RTT and cwnd bias, part of the bufferbloat
problem.
sk->sk_wmem_alloc not allowed to grow above a given limit,
allowing no more than ~128KB [1] per tcp socket in qdisc/dev layers at a
given time.
TSO packets are sized/capped to half the limit, so that we have two
TSO packets in flight, allowing better bandwidth use.
As a side effect, setting the limit to 40000 automatically reduces the
standard gso max limit (65536) to 40000/2 : It can help to reduce
latencies of high prio packets, having smaller TSO packets.
This means we divert sock_wfree() to a tcp_wfree() handler, to
queue/send following frames when skb_orphan() [2] is called for the
already queued skbs.
Results on my dev machines (tg3/ixgbe nics) are really impressive,
using standard pfifo_fast, and with or without TSO/GSO.
Without reduction of nominal bandwidth, we have reduction of buffering
per bulk sender :
< 1ms on Gbit (instead of 50ms with TSO)
< 8ms on 100Mbit (instead of 132 ms)
I no longer have 4 MBytes backlogged in qdisc by a single netperf
session, and both side socket autotuning no longer use 4 Mbytes.
As skb destructor cannot restart xmit itself ( as qdisc lock might be
taken at this point ), we delegate the work to a tasklet. We use one
tasklest per cpu for performance reasons.
If tasklet finds a socket owned by the user, it sets TSQ_OWNED flag.
This flag is tested in a new protocol method called from release_sock(),
to eventually send new segments.
[1] New /proc/sys/net/ipv4/tcp_limit_output_bytes tunable
[2] skb_orphan() is usually called at TX completion time,
but some drivers call it in their start_xmit() handler.
These drivers should at least use BQL, or else a single TCP
session can still fill the whole NIC TX ring, since TSQ will
have no effect.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Dave Taht <dave.taht@bufferbloat.net>
Cc: Tom Herbert <therbert@google.com>
Cc: Matt Mathis <mattmathis@google.com>
Cc: Yuchung Cheng <ycheng@google.com>
Cc: Nandita Dukkipati <nanditad@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2012-07-11 12:50:31 +07:00
|
|
|
|
|
|
|
if (test_and_clear_bit(TSQ_THROTTLED, &tp->tsq_flags) &&
|
|
|
|
!test_and_set_bit(TSQ_QUEUED, &tp->tsq_flags)) {
|
|
|
|
unsigned long flags;
|
|
|
|
struct tsq_tasklet *tsq;
|
|
|
|
|
|
|
|
/* queue this socket to tasklet queue */
|
|
|
|
local_irq_save(flags);
|
2014-08-18 00:30:35 +07:00
|
|
|
tsq = this_cpu_ptr(&tsq_tasklet);
|
tcp: TCP Small Queues
This introduce TSQ (TCP Small Queues)
TSQ goal is to reduce number of TCP packets in xmit queues (qdisc &
device queues), to reduce RTT and cwnd bias, part of the bufferbloat
problem.
sk->sk_wmem_alloc not allowed to grow above a given limit,
allowing no more than ~128KB [1] per tcp socket in qdisc/dev layers at a
given time.
TSO packets are sized/capped to half the limit, so that we have two
TSO packets in flight, allowing better bandwidth use.
As a side effect, setting the limit to 40000 automatically reduces the
standard gso max limit (65536) to 40000/2 : It can help to reduce
latencies of high prio packets, having smaller TSO packets.
This means we divert sock_wfree() to a tcp_wfree() handler, to
queue/send following frames when skb_orphan() [2] is called for the
already queued skbs.
Results on my dev machines (tg3/ixgbe nics) are really impressive,
using standard pfifo_fast, and with or without TSO/GSO.
Without reduction of nominal bandwidth, we have reduction of buffering
per bulk sender :
< 1ms on Gbit (instead of 50ms with TSO)
< 8ms on 100Mbit (instead of 132 ms)
I no longer have 4 MBytes backlogged in qdisc by a single netperf
session, and both side socket autotuning no longer use 4 Mbytes.
As skb destructor cannot restart xmit itself ( as qdisc lock might be
taken at this point ), we delegate the work to a tasklet. We use one
tasklest per cpu for performance reasons.
If tasklet finds a socket owned by the user, it sets TSQ_OWNED flag.
This flag is tested in a new protocol method called from release_sock(),
to eventually send new segments.
[1] New /proc/sys/net/ipv4/tcp_limit_output_bytes tunable
[2] skb_orphan() is usually called at TX completion time,
but some drivers call it in their start_xmit() handler.
These drivers should at least use BQL, or else a single TCP
session can still fill the whole NIC TX ring, since TSQ will
have no effect.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Dave Taht <dave.taht@bufferbloat.net>
Cc: Tom Herbert <therbert@google.com>
Cc: Matt Mathis <mattmathis@google.com>
Cc: Yuchung Cheng <ycheng@google.com>
Cc: Nandita Dukkipati <nanditad@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2012-07-11 12:50:31 +07:00
|
|
|
list_add(&tp->tsq_node, &tsq->head);
|
|
|
|
tasklet_schedule(&tsq->tasklet);
|
|
|
|
local_irq_restore(flags);
|
2014-10-13 20:27:47 +07:00
|
|
|
return;
|
tcp: TCP Small Queues
This introduce TSQ (TCP Small Queues)
TSQ goal is to reduce number of TCP packets in xmit queues (qdisc &
device queues), to reduce RTT and cwnd bias, part of the bufferbloat
problem.
sk->sk_wmem_alloc not allowed to grow above a given limit,
allowing no more than ~128KB [1] per tcp socket in qdisc/dev layers at a
given time.
TSO packets are sized/capped to half the limit, so that we have two
TSO packets in flight, allowing better bandwidth use.
As a side effect, setting the limit to 40000 automatically reduces the
standard gso max limit (65536) to 40000/2 : It can help to reduce
latencies of high prio packets, having smaller TSO packets.
This means we divert sock_wfree() to a tcp_wfree() handler, to
queue/send following frames when skb_orphan() [2] is called for the
already queued skbs.
Results on my dev machines (tg3/ixgbe nics) are really impressive,
using standard pfifo_fast, and with or without TSO/GSO.
Without reduction of nominal bandwidth, we have reduction of buffering
per bulk sender :
< 1ms on Gbit (instead of 50ms with TSO)
< 8ms on 100Mbit (instead of 132 ms)
I no longer have 4 MBytes backlogged in qdisc by a single netperf
session, and both side socket autotuning no longer use 4 Mbytes.
As skb destructor cannot restart xmit itself ( as qdisc lock might be
taken at this point ), we delegate the work to a tasklet. We use one
tasklest per cpu for performance reasons.
If tasklet finds a socket owned by the user, it sets TSQ_OWNED flag.
This flag is tested in a new protocol method called from release_sock(),
to eventually send new segments.
[1] New /proc/sys/net/ipv4/tcp_limit_output_bytes tunable
[2] skb_orphan() is usually called at TX completion time,
but some drivers call it in their start_xmit() handler.
These drivers should at least use BQL, or else a single TCP
session can still fill the whole NIC TX ring, since TSQ will
have no effect.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Dave Taht <dave.taht@bufferbloat.net>
Cc: Tom Herbert <therbert@google.com>
Cc: Matt Mathis <mattmathis@google.com>
Cc: Yuchung Cheng <ycheng@google.com>
Cc: Nandita Dukkipati <nanditad@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2012-07-11 12:50:31 +07:00
|
|
|
}
|
2014-10-13 20:27:47 +07:00
|
|
|
out:
|
|
|
|
sk_free(sk);
|
tcp: TCP Small Queues
This introduce TSQ (TCP Small Queues)
TSQ goal is to reduce number of TCP packets in xmit queues (qdisc &
device queues), to reduce RTT and cwnd bias, part of the bufferbloat
problem.
sk->sk_wmem_alloc not allowed to grow above a given limit,
allowing no more than ~128KB [1] per tcp socket in qdisc/dev layers at a
given time.
TSO packets are sized/capped to half the limit, so that we have two
TSO packets in flight, allowing better bandwidth use.
As a side effect, setting the limit to 40000 automatically reduces the
standard gso max limit (65536) to 40000/2 : It can help to reduce
latencies of high prio packets, having smaller TSO packets.
This means we divert sock_wfree() to a tcp_wfree() handler, to
queue/send following frames when skb_orphan() [2] is called for the
already queued skbs.
Results on my dev machines (tg3/ixgbe nics) are really impressive,
using standard pfifo_fast, and with or without TSO/GSO.
Without reduction of nominal bandwidth, we have reduction of buffering
per bulk sender :
< 1ms on Gbit (instead of 50ms with TSO)
< 8ms on 100Mbit (instead of 132 ms)
I no longer have 4 MBytes backlogged in qdisc by a single netperf
session, and both side socket autotuning no longer use 4 Mbytes.
As skb destructor cannot restart xmit itself ( as qdisc lock might be
taken at this point ), we delegate the work to a tasklet. We use one
tasklest per cpu for performance reasons.
If tasklet finds a socket owned by the user, it sets TSQ_OWNED flag.
This flag is tested in a new protocol method called from release_sock(),
to eventually send new segments.
[1] New /proc/sys/net/ipv4/tcp_limit_output_bytes tunable
[2] skb_orphan() is usually called at TX completion time,
but some drivers call it in their start_xmit() handler.
These drivers should at least use BQL, or else a single TCP
session can still fill the whole NIC TX ring, since TSQ will
have no effect.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Dave Taht <dave.taht@bufferbloat.net>
Cc: Tom Herbert <therbert@google.com>
Cc: Matt Mathis <mattmathis@google.com>
Cc: Yuchung Cheng <ycheng@google.com>
Cc: Nandita Dukkipati <nanditad@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2012-07-11 12:50:31 +07:00
|
|
|
}
|
|
|
|
|
2005-04-17 05:20:36 +07:00
|
|
|
/* This routine actually transmits TCP packets queued in by
|
|
|
|
* tcp_do_sendmsg(). This is used by both the initial
|
|
|
|
* transmission and possible later retransmissions.
|
|
|
|
* All SKB's seen here are completely headerless. It is our
|
|
|
|
* job to build the TCP header, and pass the packet down to
|
|
|
|
* IP so it can do the same plus pass the packet off to the
|
|
|
|
* device.
|
|
|
|
*
|
|
|
|
* We are working here with either a clone of the original
|
|
|
|
* SKB, or a fresh unique copy made by the retransmit engine.
|
|
|
|
*/
|
2008-01-01 05:57:14 +07:00
|
|
|
static int tcp_transmit_skb(struct sock *sk, struct sk_buff *skb, int clone_it,
|
|
|
|
gfp_t gfp_mask)
|
2005-04-17 05:20:36 +07:00
|
|
|
{
|
2005-12-07 07:24:52 +07:00
|
|
|
const struct inet_connection_sock *icsk = inet_csk(sk);
|
|
|
|
struct inet_sock *inet;
|
|
|
|
struct tcp_sock *tp;
|
|
|
|
struct tcp_skb_cb *tcb;
|
2008-07-19 14:04:31 +07:00
|
|
|
struct tcp_out_options opts;
|
2012-04-15 12:58:06 +07:00
|
|
|
unsigned int tcp_options_size, tcp_header_size;
|
2006-11-15 10:07:45 +07:00
|
|
|
struct tcp_md5sig_key *md5;
|
2005-12-07 07:24:52 +07:00
|
|
|
struct tcphdr *th;
|
|
|
|
int err;
|
|
|
|
|
|
|
|
BUG_ON(!skb || !tcp_skb_pcount(skb));
|
|
|
|
|
2013-10-10 22:43:00 +07:00
|
|
|
if (clone_it) {
|
2014-02-27 05:02:48 +07:00
|
|
|
skb_mstamp_get(&skb->skb_mstamp);
|
2013-10-10 22:43:00 +07:00
|
|
|
|
2005-12-07 07:24:52 +07:00
|
|
|
if (unlikely(skb_cloned(skb)))
|
|
|
|
skb = pskb_copy(skb, gfp_mask);
|
|
|
|
else
|
|
|
|
skb = skb_clone(skb, gfp_mask);
|
|
|
|
if (unlikely(!skb))
|
|
|
|
return -ENOBUFS;
|
|
|
|
}
|
2005-04-17 05:20:36 +07:00
|
|
|
|
2005-12-07 07:24:52 +07:00
|
|
|
inet = inet_sk(sk);
|
|
|
|
tp = tcp_sk(sk);
|
|
|
|
tcb = TCP_SKB_CB(skb);
|
2008-07-19 14:04:31 +07:00
|
|
|
memset(&opts, 0, sizeof(opts));
|
2005-04-17 05:20:36 +07:00
|
|
|
|
2011-09-28 00:25:05 +07:00
|
|
|
if (unlikely(tcb->tcp_flags & TCPHDR_SYN))
|
2008-07-19 14:04:31 +07:00
|
|
|
tcp_options_size = tcp_syn_options(sk, skb, &opts, &md5);
|
|
|
|
else
|
|
|
|
tcp_options_size = tcp_established_options(sk, skb, &opts,
|
|
|
|
&md5);
|
|
|
|
tcp_header_size = tcp_options_size + sizeof(struct tcphdr);
|
2007-02-09 21:24:47 +07:00
|
|
|
|
2013-05-23 14:44:20 +07:00
|
|
|
/* if no packet is in qdisc/device queue, then allow XPS to select
|
2014-10-11 08:06:35 +07:00
|
|
|
* another queue. We can be called from tcp_tsq_handler()
|
|
|
|
* which holds one reference to sk_wmem_alloc.
|
|
|
|
*
|
|
|
|
* TODO: Ideally, in-flight pure ACK packets should not matter here.
|
|
|
|
* One way to get this would be to set skb->truesize = 2 on them.
|
2013-05-23 14:44:20 +07:00
|
|
|
*/
|
2014-10-11 08:06:35 +07:00
|
|
|
skb->ooo_okay = sk_wmem_alloc_get(sk) < SKB_TRUESIZE(1);
|
2005-12-07 07:24:52 +07:00
|
|
|
|
2007-04-11 11:04:22 +07:00
|
|
|
skb_push(skb, tcp_header_size);
|
|
|
|
skb_reset_transport_header(skb);
|
tcp: TCP Small Queues
This introduce TSQ (TCP Small Queues)
TSQ goal is to reduce number of TCP packets in xmit queues (qdisc &
device queues), to reduce RTT and cwnd bias, part of the bufferbloat
problem.
sk->sk_wmem_alloc not allowed to grow above a given limit,
allowing no more than ~128KB [1] per tcp socket in qdisc/dev layers at a
given time.
TSO packets are sized/capped to half the limit, so that we have two
TSO packets in flight, allowing better bandwidth use.
As a side effect, setting the limit to 40000 automatically reduces the
standard gso max limit (65536) to 40000/2 : It can help to reduce
latencies of high prio packets, having smaller TSO packets.
This means we divert sock_wfree() to a tcp_wfree() handler, to
queue/send following frames when skb_orphan() [2] is called for the
already queued skbs.
Results on my dev machines (tg3/ixgbe nics) are really impressive,
using standard pfifo_fast, and with or without TSO/GSO.
Without reduction of nominal bandwidth, we have reduction of buffering
per bulk sender :
< 1ms on Gbit (instead of 50ms with TSO)
< 8ms on 100Mbit (instead of 132 ms)
I no longer have 4 MBytes backlogged in qdisc by a single netperf
session, and both side socket autotuning no longer use 4 Mbytes.
As skb destructor cannot restart xmit itself ( as qdisc lock might be
taken at this point ), we delegate the work to a tasklet. We use one
tasklest per cpu for performance reasons.
If tasklet finds a socket owned by the user, it sets TSQ_OWNED flag.
This flag is tested in a new protocol method called from release_sock(),
to eventually send new segments.
[1] New /proc/sys/net/ipv4/tcp_limit_output_bytes tunable
[2] skb_orphan() is usually called at TX completion time,
but some drivers call it in their start_xmit() handler.
These drivers should at least use BQL, or else a single TCP
session can still fill the whole NIC TX ring, since TSQ will
have no effect.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Dave Taht <dave.taht@bufferbloat.net>
Cc: Tom Herbert <therbert@google.com>
Cc: Matt Mathis <mattmathis@google.com>
Cc: Yuchung Cheng <ycheng@google.com>
Cc: Nandita Dukkipati <nanditad@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2012-07-11 12:50:31 +07:00
|
|
|
|
|
|
|
skb_orphan(skb);
|
|
|
|
skb->sk = sk;
|
tcp: do not pace pure ack packets
When we added pacing to TCP, we decided to let sch_fq take care
of actual pacing.
All TCP had to do was to compute sk->pacing_rate using simple formula:
sk->pacing_rate = 2 * cwnd * mss / rtt
It works well for senders (bulk flows), but not very well for receivers
or even RPC :
cwnd on the receiver can be less than 10, rtt can be around 100ms, so we
can end up pacing ACK packets, slowing down the sender.
Really, only the sender should pace, according to its own logic.
Instead of adding a new bit in skb, or call yet another flow
dissection, we tweak skb->truesize to a small value (2), and
we instruct sch_fq to use new helper and not pace pure ack.
Note this also helps TCP small queue, as ack packets present
in qdisc/NIC do not prevent sending a data packet (RPC workload)
This helps to reduce tx completion overhead, ack packets can use regular
sock_wfree() instead of tcp_wfree() which is a bit more expensive.
This has no impact in the case packets are sent to loopback interface,
as we do not coalesce ack packets (were we would detect skb->truesize
lie)
In case netem (with a delay) is used, skb_orphan_partial() also sets
skb->truesize to 1.
This patch is a combination of two patches we used for about one year at
Google.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2015-02-04 09:31:53 +07:00
|
|
|
skb->destructor = skb_is_tcp_pure_ack(skb) ? sock_wfree : tcp_wfree;
|
2014-07-02 11:32:17 +07:00
|
|
|
skb_set_hash_from_sk(skb, sk);
|
tcp: TCP Small Queues
This introduce TSQ (TCP Small Queues)
TSQ goal is to reduce number of TCP packets in xmit queues (qdisc &
device queues), to reduce RTT and cwnd bias, part of the bufferbloat
problem.
sk->sk_wmem_alloc not allowed to grow above a given limit,
allowing no more than ~128KB [1] per tcp socket in qdisc/dev layers at a
given time.
TSO packets are sized/capped to half the limit, so that we have two
TSO packets in flight, allowing better bandwidth use.
As a side effect, setting the limit to 40000 automatically reduces the
standard gso max limit (65536) to 40000/2 : It can help to reduce
latencies of high prio packets, having smaller TSO packets.
This means we divert sock_wfree() to a tcp_wfree() handler, to
queue/send following frames when skb_orphan() [2] is called for the
already queued skbs.
Results on my dev machines (tg3/ixgbe nics) are really impressive,
using standard pfifo_fast, and with or without TSO/GSO.
Without reduction of nominal bandwidth, we have reduction of buffering
per bulk sender :
< 1ms on Gbit (instead of 50ms with TSO)
< 8ms on 100Mbit (instead of 132 ms)
I no longer have 4 MBytes backlogged in qdisc by a single netperf
session, and both side socket autotuning no longer use 4 Mbytes.
As skb destructor cannot restart xmit itself ( as qdisc lock might be
taken at this point ), we delegate the work to a tasklet. We use one
tasklest per cpu for performance reasons.
If tasklet finds a socket owned by the user, it sets TSQ_OWNED flag.
This flag is tested in a new protocol method called from release_sock(),
to eventually send new segments.
[1] New /proc/sys/net/ipv4/tcp_limit_output_bytes tunable
[2] skb_orphan() is usually called at TX completion time,
but some drivers call it in their start_xmit() handler.
These drivers should at least use BQL, or else a single TCP
session can still fill the whole NIC TX ring, since TSQ will
have no effect.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Dave Taht <dave.taht@bufferbloat.net>
Cc: Tom Herbert <therbert@google.com>
Cc: Matt Mathis <mattmathis@google.com>
Cc: Yuchung Cheng <ycheng@google.com>
Cc: Nandita Dukkipati <nanditad@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2012-07-11 12:50:31 +07:00
|
|
|
atomic_add(skb->truesize, &sk->sk_wmem_alloc);
|
2005-12-07 07:24:52 +07:00
|
|
|
|
|
|
|
/* Build TCP header and checksum it. */
|
2007-04-11 11:04:22 +07:00
|
|
|
th = tcp_hdr(skb);
|
2009-10-15 13:30:45 +07:00
|
|
|
th->source = inet->inet_sport;
|
|
|
|
th->dest = inet->inet_dport;
|
2005-12-07 07:24:52 +07:00
|
|
|
th->seq = htonl(tcb->seq);
|
|
|
|
th->ack_seq = htonl(tp->rcv_nxt);
|
2006-09-28 08:38:52 +07:00
|
|
|
*(((__be16 *)th) + 6) = htons(((tcp_header_size >> 2) << 12) |
|
2011-09-28 00:25:05 +07:00
|
|
|
tcb->tcp_flags);
|
2005-12-07 07:24:52 +07:00
|
|
|
|
2011-09-28 00:25:05 +07:00
|
|
|
if (unlikely(tcb->tcp_flags & TCPHDR_SYN)) {
|
2005-12-07 07:24:52 +07:00
|
|
|
/* RFC1323: The window in SYN & SYN/ACK segments
|
|
|
|
* is never scaled.
|
|
|
|
*/
|
2007-02-14 03:42:11 +07:00
|
|
|
th->window = htons(min(tp->rcv_wnd, 65535U));
|
2005-12-07 07:24:52 +07:00
|
|
|
} else {
|
|
|
|
th->window = htons(tcp_select_window(sk));
|
|
|
|
}
|
|
|
|
th->check = 0;
|
|
|
|
th->urg_ptr = 0;
|
2005-04-17 05:20:36 +07:00
|
|
|
|
2008-10-08 04:43:06 +07:00
|
|
|
/* The urg_mode check is necessary during a below snd_una win probe */
|
2009-02-22 14:52:29 +07:00
|
|
|
if (unlikely(tcp_urg_mode(tp) && before(tcb->seq, tp->snd_up))) {
|
|
|
|
if (before(tp->snd_up, tcb->seq + 0x10000)) {
|
|
|
|
th->urg_ptr = htons(tp->snd_up - tcb->seq);
|
|
|
|
th->urg = 1;
|
|
|
|
} else if (after(tcb->seq + 0xFFFF, tp->snd_nxt)) {
|
2010-04-21 09:06:52 +07:00
|
|
|
th->urg_ptr = htons(0xFFFF);
|
2009-02-22 14:52:29 +07:00
|
|
|
th->urg = 1;
|
|
|
|
}
|
2005-12-07 07:24:52 +07:00
|
|
|
}
|
2005-04-17 05:20:36 +07:00
|
|
|
|
2009-12-03 01:23:05 +07:00
|
|
|
tcp_options_write((__be32 *)(th + 1), tp, &opts);
|
2015-06-11 23:15:16 +07:00
|
|
|
skb_shinfo(skb)->gso_type = sk->sk_gso_type;
|
2011-09-28 00:25:05 +07:00
|
|
|
if (likely((tcb->tcp_flags & TCPHDR_SYN) == 0))
|
2014-09-29 18:08:30 +07:00
|
|
|
tcp_ecn_send(sk, skb, tcp_header_size);
|
2005-04-17 05:20:36 +07:00
|
|
|
|
2006-11-15 10:07:45 +07:00
|
|
|
#ifdef CONFIG_TCP_MD5SIG
|
|
|
|
/* Calculate the MD5 hash, as we have all we need now */
|
|
|
|
if (md5) {
|
2010-05-16 14:36:33 +07:00
|
|
|
sk_nocaps_add(sk, NETIF_F_GSO_MASK);
|
2009-12-03 01:23:05 +07:00
|
|
|
tp->af_specific->calc_md5_hash(opts.hash_location,
|
2015-03-25 05:58:55 +07:00
|
|
|
md5, sk, skb);
|
2006-11-15 10:07:45 +07:00
|
|
|
}
|
|
|
|
#endif
|
|
|
|
|
2010-04-11 09:15:55 +07:00
|
|
|
icsk->icsk_af_ops->send_check(sk, skb);
|
2005-04-17 05:20:36 +07:00
|
|
|
|
2011-09-28 00:25:05 +07:00
|
|
|
if (likely(tcb->tcp_flags & TCPHDR_ACK))
|
2005-12-07 07:24:52 +07:00
|
|
|
tcp_event_ack_sent(sk, tcp_skb_pcount(skb));
|
2005-04-17 05:20:36 +07:00
|
|
|
|
2016-03-15 00:52:15 +07:00
|
|
|
if (skb->len != tcp_header_size) {
|
2011-10-21 16:22:42 +07:00
|
|
|
tcp_event_data_sent(tp, sk);
|
2016-03-15 00:52:15 +07:00
|
|
|
tp->data_segs_out += tcp_skb_pcount(skb);
|
|
|
|
}
|
2005-04-17 05:20:36 +07:00
|
|
|
|
2006-08-08 11:04:15 +07:00
|
|
|
if (after(tcb->end_seq, tp->snd_nxt) || tcb->seq == tcb->end_seq)
|
2010-04-22 14:00:24 +07:00
|
|
|
TCP_ADD_STATS(sock_net(sk), TCP_MIB_OUTSEGS,
|
|
|
|
tcp_skb_pcount(skb));
|
2005-04-17 05:20:36 +07:00
|
|
|
|
2015-05-21 06:35:41 +07:00
|
|
|
tp->segs_out += tcp_skb_pcount(skb);
|
2015-06-11 23:15:18 +07:00
|
|
|
/* OK, its time to fill skb_shinfo(skb)->gso_{segs|size} */
|
2014-09-24 18:11:22 +07:00
|
|
|
skb_shinfo(skb)->gso_segs = tcp_skb_pcount(skb);
|
2015-06-11 23:15:18 +07:00
|
|
|
skb_shinfo(skb)->gso_size = tcp_skb_mss(skb);
|
2014-09-24 18:11:22 +07:00
|
|
|
|
2014-09-06 05:33:33 +07:00
|
|
|
/* Our usage of tstamp should remain private */
|
|
|
|
skb->tstamp.tv64 = 0;
|
2014-09-27 23:50:57 +07:00
|
|
|
|
|
|
|
/* Cleanup our debris for IP stacks */
|
|
|
|
memset(skb->cb, 0, max(sizeof(struct inet_skb_parm),
|
|
|
|
sizeof(struct inet6_skb_parm)));
|
|
|
|
|
2014-04-15 23:58:34 +07:00
|
|
|
err = icsk->icsk_af_ops->queue_xmit(sk, skb, &inet->cork.fl);
|
2014-09-06 05:33:33 +07:00
|
|
|
|
2006-04-29 05:26:50 +07:00
|
|
|
if (likely(err <= 0))
|
2005-12-07 07:24:52 +07:00
|
|
|
return err;
|
|
|
|
|
2014-07-14 21:58:32 +07:00
|
|
|
tcp_enter_cwr(sk);
|
2005-12-07 07:24:52 +07:00
|
|
|
|
2006-11-14 20:21:36 +07:00
|
|
|
return net_xmit_eval(err);
|
2005-04-17 05:20:36 +07:00
|
|
|
}
|
|
|
|
|
2009-07-22 06:00:40 +07:00
|
|
|
/* This routine just queues the buffer for sending.
|
2005-04-17 05:20:36 +07:00
|
|
|
*
|
|
|
|
* NOTE: probe0 timer is not checked, do not forget tcp_push_pending_frames,
|
|
|
|
* otherwise socket can stall.
|
|
|
|
*/
|
|
|
|
static void tcp_queue_skb(struct sock *sk, struct sk_buff *skb)
|
|
|
|
{
|
|
|
|
struct tcp_sock *tp = tcp_sk(sk);
|
|
|
|
|
|
|
|
/* Advance write_seq and place onto the write_queue. */
|
|
|
|
tp->write_seq = TCP_SKB_CB(skb)->end_seq;
|
2014-09-23 06:29:32 +07:00
|
|
|
__skb_header_release(skb);
|
2007-03-08 03:12:44 +07:00
|
|
|
tcp_add_write_queue_tail(sk, skb);
|
2007-12-31 15:11:19 +07:00
|
|
|
sk->sk_wmem_queued += skb->truesize;
|
|
|
|
sk_mem_charge(sk, skb->truesize);
|
2005-04-17 05:20:36 +07:00
|
|
|
}
|
|
|
|
|
2009-07-22 06:00:40 +07:00
|
|
|
/* Initialize TSO segments for a packet. */
|
2015-06-11 23:15:17 +07:00
|
|
|
static void tcp_set_skb_tso_segs(struct sk_buff *skb, unsigned int mss_now)
|
2005-07-06 05:18:03 +07:00
|
|
|
{
|
2013-10-16 02:24:54 +07:00
|
|
|
if (skb->len <= mss_now || skb->ip_summed == CHECKSUM_NONE) {
|
2005-07-06 05:18:03 +07:00
|
|
|
/* Avoid the costly divide in the normal
|
|
|
|
* non-TSO case.
|
|
|
|
*/
|
2014-09-24 18:11:22 +07:00
|
|
|
tcp_skb_pcount_set(skb, 1);
|
2015-06-11 23:15:18 +07:00
|
|
|
TCP_SKB_CB(skb)->tcp_gso_size = 0;
|
2005-07-06 05:18:03 +07:00
|
|
|
} else {
|
2014-09-24 18:11:22 +07:00
|
|
|
tcp_skb_pcount_set(skb, DIV_ROUND_UP(skb->len, mss_now));
|
2015-06-11 23:15:18 +07:00
|
|
|
TCP_SKB_CB(skb)->tcp_gso_size = mss_now;
|
2005-04-17 05:20:36 +07:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2007-10-09 15:24:15 +07:00
|
|
|
/* When a modification to fackets out becomes necessary, we need to check
|
[TCP]: Rewrite SACK block processing & sack_recv_cache use
Key points of this patch are:
- In case new SACK information is advance only type, no skb
processing below previously discovered highest point is done
- Optimize cases below highest point too since there's no need
to always go up to highest point (which is very likely still
present in that SACK), this is not entirely true though
because I'm dropping the fastpath_skb_hint which could
previously optimize those cases even better. Whether that's
significant, I'm not too sure.
Currently it will provide skipping by walking. Combined with
RB-tree, all skipping would become fast too regardless of window
size (can be done incrementally later).
Previously a number of cases in TCP SACK processing fails to
take advantage of costly stored information in sack_recv_cache,
most importantly, expected events such as cumulative ACK and new
hole ACKs. Processing on such ACKs result in rather long walks
building up latencies (which easily gets nasty when window is
huge). Those latencies are often completely unnecessary
compared with the amount of _new_ information received, usually
for cumulative ACK there's no new information at all, yet TCP
walks whole queue unnecessary potentially taking a number of
costly cache misses on the way, etc.!
Since the inclusion of highest_sack, there's a lot information
that is very likely redundant (SACK fastpath hint stuff,
fackets_out, highest_sack), though there's no ultimate guarantee
that they'll remain the same whole the time (in all unearthly
scenarios). Take advantage of this knowledge here and drop
fastpath hint and use direct access to highest SACKed skb as
a replacement.
Effectively "special cased" fastpath is dropped. This change
adds some complexity to introduce better coveraged "fastpath",
though the added complexity should make TCP behave more cache
friendly.
The current ACK's SACK blocks are compared against each cached
block individially and only ranges that are new are then scanned
by the high constant walk. For other parts of write queue, even
when in previously known part of the SACK blocks, a faster skip
function is used (if necessary at all). In addition, whenever
possible, TCP fast-forwards to highest_sack skb that was made
available by an earlier patch. In typical case, no other things
but this fast-forward and mandatory markings after that occur
making the access pattern quite similar to the former fastpath
"special case".
DSACKs are special case that must always be walked.
The local to recv_sack_cache copying could be more intelligent
w.r.t DSACKs which are likely to be there only once but that
is left to a separate patch.
Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>
2007-11-16 10:50:37 +07:00
|
|
|
* skb is counted to fackets_out or not.
|
2007-10-09 15:24:15 +07:00
|
|
|
*/
|
2011-10-21 16:22:42 +07:00
|
|
|
static void tcp_adjust_fackets_out(struct sock *sk, const struct sk_buff *skb,
|
2007-10-09 15:24:15 +07:00
|
|
|
int decr)
|
|
|
|
{
|
2007-11-16 10:41:46 +07:00
|
|
|
struct tcp_sock *tp = tcp_sk(sk);
|
|
|
|
|
2007-10-02 05:27:19 +07:00
|
|
|
if (!tp->sacked_out || tcp_is_reno(tp))
|
2007-10-09 15:24:15 +07:00
|
|
|
return;
|
|
|
|
|
2007-12-02 05:48:06 +07:00
|
|
|
if (after(tcp_highest_sack_seq(tp), TCP_SKB_CB(skb)->seq))
|
2007-10-09 15:24:15 +07:00
|
|
|
tp->fackets_out -= decr;
|
|
|
|
}
|
|
|
|
|
2009-04-02 06:15:17 +07:00
|
|
|
/* Pcount in the middle of the write queue got changed, we need to do various
|
|
|
|
* tweaks to fix counters
|
|
|
|
*/
|
2011-10-21 16:22:42 +07:00
|
|
|
static void tcp_adjust_pcount(struct sock *sk, const struct sk_buff *skb, int decr)
|
2009-04-02 06:15:17 +07:00
|
|
|
{
|
|
|
|
struct tcp_sock *tp = tcp_sk(sk);
|
|
|
|
|
|
|
|
tp->packets_out -= decr;
|
|
|
|
|
|
|
|
if (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_ACKED)
|
|
|
|
tp->sacked_out -= decr;
|
|
|
|
if (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_RETRANS)
|
|
|
|
tp->retrans_out -= decr;
|
|
|
|
if (TCP_SKB_CB(skb)->sacked & TCPCB_LOST)
|
|
|
|
tp->lost_out -= decr;
|
|
|
|
|
|
|
|
/* Reno case is special. Sigh... */
|
|
|
|
if (tcp_is_reno(tp) && decr > 0)
|
|
|
|
tp->sacked_out -= min_t(u32, tp->sacked_out, decr);
|
|
|
|
|
|
|
|
tcp_adjust_fackets_out(sk, skb, decr);
|
|
|
|
|
|
|
|
if (tp->lost_skb_hint &&
|
|
|
|
before(TCP_SKB_CB(skb)->seq, TCP_SKB_CB(tp->lost_skb_hint)->seq) &&
|
2009-04-18 12:48:48 +07:00
|
|
|
(tcp_is_fack(tp) || (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_ACKED)))
|
2009-04-02 06:15:17 +07:00
|
|
|
tp->lost_cnt_hint -= decr;
|
|
|
|
|
|
|
|
tcp_verify_left_out(tp);
|
|
|
|
}
|
|
|
|
|
2014-08-13 02:08:12 +07:00
|
|
|
static void tcp_fragment_tstamp(struct sk_buff *skb, struct sk_buff *skb2)
|
|
|
|
{
|
|
|
|
struct skb_shared_info *shinfo = skb_shinfo(skb);
|
|
|
|
|
|
|
|
if (unlikely(shinfo->tx_flags & SKBTX_ANY_TSTAMP) &&
|
|
|
|
!before(shinfo->tskey, TCP_SKB_CB(skb2)->seq)) {
|
|
|
|
struct skb_shared_info *shinfo2 = skb_shinfo(skb2);
|
|
|
|
u8 tsflags = shinfo->tx_flags & SKBTX_ANY_TSTAMP;
|
|
|
|
|
|
|
|
shinfo->tx_flags &= ~tsflags;
|
|
|
|
shinfo2->tx_flags |= tsflags;
|
|
|
|
swap(shinfo->tskey, shinfo2->tskey);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2005-04-17 05:20:36 +07:00
|
|
|
/* Function to create two new TCP segments. Shrinks the given segment
|
|
|
|
* to the specified size and appends a new segment with the rest of the
|
2007-02-09 21:24:47 +07:00
|
|
|
* packet to the list. This won't be called frequently, I hope.
|
2005-04-17 05:20:36 +07:00
|
|
|
* Remember, these are still headerless SKBs at this point.
|
|
|
|
*/
|
2008-01-01 05:57:14 +07:00
|
|
|
int tcp_fragment(struct sock *sk, struct sk_buff *skb, u32 len,
|
2014-06-06 21:32:37 +07:00
|
|
|
unsigned int mss_now, gfp_t gfp)
|
2005-04-17 05:20:36 +07:00
|
|
|
{
|
|
|
|
struct tcp_sock *tp = tcp_sk(sk);
|
|
|
|
struct sk_buff *buff;
|
2005-09-02 12:47:01 +07:00
|
|
|
int nsize, old_factor;
|
2006-04-20 11:35:00 +07:00
|
|
|
int nlen;
|
2009-02-28 11:44:42 +07:00
|
|
|
u8 flags;
|
2005-04-17 05:20:36 +07:00
|
|
|
|
2011-04-02 11:47:41 +07:00
|
|
|
if (WARN_ON(len > skb->len))
|
|
|
|
return -EINVAL;
|
2005-11-11 08:14:59 +07:00
|
|
|
|
2005-04-17 05:20:36 +07:00
|
|
|
nsize = skb_headlen(skb) - len;
|
|
|
|
if (nsize < 0)
|
|
|
|
nsize = 0;
|
|
|
|
|
2014-06-06 21:32:37 +07:00
|
|
|
if (skb_unclone(skb, gfp))
|
2005-04-17 05:20:36 +07:00
|
|
|
return -ENOMEM;
|
|
|
|
|
|
|
|
/* Get a new skb... force flag on. */
|
2015-05-20 03:26:55 +07:00
|
|
|
buff = sk_stream_alloc_skb(sk, nsize, gfp, true);
|
2015-04-03 15:17:26 +07:00
|
|
|
if (!buff)
|
2005-04-17 05:20:36 +07:00
|
|
|
return -ENOMEM; /* We'll just try again later. */
|
2006-04-19 03:24:14 +07:00
|
|
|
|
2007-12-31 15:11:19 +07:00
|
|
|
sk->sk_wmem_queued += buff->truesize;
|
|
|
|
sk_mem_charge(sk, buff->truesize);
|
2006-04-20 11:35:00 +07:00
|
|
|
nlen = skb->len - len - nsize;
|
|
|
|
buff->truesize += nlen;
|
|
|
|
skb->truesize -= nlen;
|
2005-04-17 05:20:36 +07:00
|
|
|
|
|
|
|
/* Correct the sequence numbers. */
|
|
|
|
TCP_SKB_CB(buff)->seq = TCP_SKB_CB(skb)->seq + len;
|
|
|
|
TCP_SKB_CB(buff)->end_seq = TCP_SKB_CB(skb)->end_seq;
|
|
|
|
TCP_SKB_CB(skb)->end_seq = TCP_SKB_CB(buff)->seq;
|
|
|
|
|
|
|
|
/* PSH and FIN should only be set in the second packet. */
|
2011-09-28 00:25:05 +07:00
|
|
|
flags = TCP_SKB_CB(skb)->tcp_flags;
|
|
|
|
TCP_SKB_CB(skb)->tcp_flags = flags & ~(TCPHDR_FIN | TCPHDR_PSH);
|
|
|
|
TCP_SKB_CB(buff)->tcp_flags = flags;
|
2005-09-20 08:18:38 +07:00
|
|
|
TCP_SKB_CB(buff)->sacked = TCP_SKB_CB(skb)->sacked;
|
2005-04-17 05:20:36 +07:00
|
|
|
|
2006-08-30 06:44:56 +07:00
|
|
|
if (!skb_shinfo(skb)->nr_frags && skb->ip_summed != CHECKSUM_PARTIAL) {
|
2005-04-17 05:20:36 +07:00
|
|
|
/* Copy and checksum data tail into the new buffer. */
|
2008-01-01 05:57:14 +07:00
|
|
|
buff->csum = csum_partial_copy_nocheck(skb->data + len,
|
|
|
|
skb_put(buff, nsize),
|
2005-04-17 05:20:36 +07:00
|
|
|
nsize, 0);
|
|
|
|
|
|
|
|
skb_trim(skb, len);
|
|
|
|
|
|
|
|
skb->csum = csum_block_sub(skb->csum, buff->csum, len);
|
|
|
|
} else {
|
2006-08-30 06:44:56 +07:00
|
|
|
skb->ip_summed = CHECKSUM_PARTIAL;
|
2005-04-17 05:20:36 +07:00
|
|
|
skb_split(skb, buff, len);
|
|
|
|
}
|
|
|
|
|
|
|
|
buff->ip_summed = skb->ip_summed;
|
|
|
|
|
2005-08-15 07:24:31 +07:00
|
|
|
buff->tstamp = skb->tstamp;
|
2014-08-13 02:08:12 +07:00
|
|
|
tcp_fragment_tstamp(skb, buff);
|
2005-04-17 05:20:36 +07:00
|
|
|
|
2005-09-02 12:47:01 +07:00
|
|
|
old_factor = tcp_skb_pcount(skb);
|
|
|
|
|
2005-04-17 05:20:36 +07:00
|
|
|
/* Fix up tso_factor for both original and new SKB. */
|
2015-06-11 23:15:17 +07:00
|
|
|
tcp_set_skb_tso_segs(skb, mss_now);
|
|
|
|
tcp_set_skb_tso_segs(buff, mss_now);
|
2005-04-17 05:20:36 +07:00
|
|
|
|
2005-09-02 12:47:01 +07:00
|
|
|
/* If this packet has been sent out already, we must
|
|
|
|
* adjust the various packet counters.
|
|
|
|
*/
|
2005-09-09 05:10:52 +07:00
|
|
|
if (!before(tp->snd_nxt, TCP_SKB_CB(buff)->end_seq)) {
|
2005-09-02 12:47:01 +07:00
|
|
|
int diff = old_factor - tcp_skb_pcount(skb) -
|
|
|
|
tcp_skb_pcount(buff);
|
2005-04-17 05:20:36 +07:00
|
|
|
|
2009-04-02 06:15:17 +07:00
|
|
|
if (diff)
|
|
|
|
tcp_adjust_pcount(sk, skb, diff);
|
2005-04-17 05:20:36 +07:00
|
|
|
}
|
|
|
|
|
|
|
|
/* Link BUFF into the send queue. */
|
2014-09-23 06:29:32 +07:00
|
|
|
__skb_header_release(buff);
|
2007-03-08 03:12:44 +07:00
|
|
|
tcp_insert_write_queue_after(skb, buff, sk);
|
2005-04-17 05:20:36 +07:00
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* This is similar to __pskb_pull_head() (it will go to core/skbuff.c
|
|
|
|
* eventually). The difference is that pulled data not copied, but
|
|
|
|
* immediately discarded.
|
|
|
|
*/
|
2006-06-06 05:03:37 +07:00
|
|
|
static void __pskb_trim_head(struct sk_buff *skb, int len)
|
2005-04-17 05:20:36 +07:00
|
|
|
{
|
2013-12-06 13:31:30 +07:00
|
|
|
struct skb_shared_info *shinfo;
|
2005-04-17 05:20:36 +07:00
|
|
|
int i, k, eat;
|
|
|
|
|
2011-12-04 15:51:08 +07:00
|
|
|
eat = min_t(int, len, skb_headlen(skb));
|
|
|
|
if (eat) {
|
|
|
|
__skb_pull(skb, eat);
|
|
|
|
len -= eat;
|
|
|
|
if (!len)
|
|
|
|
return;
|
|
|
|
}
|
2005-04-17 05:20:36 +07:00
|
|
|
eat = len;
|
|
|
|
k = 0;
|
2013-12-06 13:31:30 +07:00
|
|
|
shinfo = skb_shinfo(skb);
|
|
|
|
for (i = 0; i < shinfo->nr_frags; i++) {
|
|
|
|
int size = skb_frag_size(&shinfo->frags[i]);
|
2011-10-19 04:00:24 +07:00
|
|
|
|
|
|
|
if (size <= eat) {
|
2011-08-23 06:44:59 +07:00
|
|
|
skb_frag_unref(skb, i);
|
2011-10-19 04:00:24 +07:00
|
|
|
eat -= size;
|
2005-04-17 05:20:36 +07:00
|
|
|
} else {
|
2013-12-06 13:31:30 +07:00
|
|
|
shinfo->frags[k] = shinfo->frags[i];
|
2005-04-17 05:20:36 +07:00
|
|
|
if (eat) {
|
2013-12-06 13:31:30 +07:00
|
|
|
shinfo->frags[k].page_offset += eat;
|
|
|
|
skb_frag_size_sub(&shinfo->frags[k], eat);
|
2005-04-17 05:20:36 +07:00
|
|
|
eat = 0;
|
|
|
|
}
|
|
|
|
k++;
|
|
|
|
}
|
|
|
|
}
|
2013-12-06 13:31:30 +07:00
|
|
|
shinfo->nr_frags = k;
|
2005-04-17 05:20:36 +07:00
|
|
|
|
2007-04-20 10:29:13 +07:00
|
|
|
skb_reset_tail_pointer(skb);
|
2005-04-17 05:20:36 +07:00
|
|
|
skb->data_len -= len;
|
|
|
|
skb->len = skb->data_len;
|
|
|
|
}
|
|
|
|
|
2009-07-22 06:00:40 +07:00
|
|
|
/* Remove acked data from a packet in the transmit queue. */
|
2005-04-17 05:20:36 +07:00
|
|
|
int tcp_trim_head(struct sock *sk, struct sk_buff *skb, u32 len)
|
|
|
|
{
|
2013-02-14 16:44:49 +07:00
|
|
|
if (skb_unclone(skb, GFP_ATOMIC))
|
2005-04-17 05:20:36 +07:00
|
|
|
return -ENOMEM;
|
|
|
|
|
2011-12-04 15:51:08 +07:00
|
|
|
__pskb_trim_head(skb, len);
|
2005-04-17 05:20:36 +07:00
|
|
|
|
|
|
|
TCP_SKB_CB(skb)->seq += len;
|
2006-08-30 06:44:56 +07:00
|
|
|
skb->ip_summed = CHECKSUM_PARTIAL;
|
2005-04-17 05:20:36 +07:00
|
|
|
|
|
|
|
skb->truesize -= len;
|
|
|
|
sk->sk_wmem_queued -= len;
|
2007-12-31 15:11:19 +07:00
|
|
|
sk_mem_uncharge(sk, len);
|
2005-04-17 05:20:36 +07:00
|
|
|
sock_set_flag(sk, SOCK_QUEUE_SHRUNK);
|
|
|
|
|
2012-01-29 00:29:46 +07:00
|
|
|
/* Any change of skb->len requires recalculation of tso factor. */
|
2005-04-17 05:20:36 +07:00
|
|
|
if (tcp_skb_pcount(skb) > 1)
|
2015-06-11 23:15:17 +07:00
|
|
|
tcp_set_skb_tso_segs(skb, tcp_skb_mss(skb));
|
2005-04-17 05:20:36 +07:00
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2013-02-22 15:59:06 +07:00
|
|
|
/* Calculate MSS not accounting any TCP options. */
|
|
|
|
static inline int __tcp_mtu_to_mss(struct sock *sk, int pmtu)
|
2006-03-21 08:53:41 +07:00
|
|
|
{
|
2011-10-21 16:22:42 +07:00
|
|
|
const struct tcp_sock *tp = tcp_sk(sk);
|
|
|
|
const struct inet_connection_sock *icsk = inet_csk(sk);
|
2006-03-21 08:53:41 +07:00
|
|
|
int mss_now;
|
|
|
|
|
|
|
|
/* Calculate base mss without TCP options:
|
|
|
|
It is MMS_S - sizeof(tcphdr) of rfc1122
|
|
|
|
*/
|
|
|
|
mss_now = pmtu - icsk->icsk_af_ops->net_header_len - sizeof(struct tcphdr);
|
|
|
|
|
ipv6: RTAX_FEATURE_ALLFRAG causes inefficient TCP segment sizing
Quoting Tore Anderson from :
https://bugzilla.kernel.org/show_bug.cgi?id=42572
When RTAX_FEATURE_ALLFRAG is set on a route, the effective TCP segment
size does not take into account the size of the IPv6 Fragmentation
header that needs to be included in outbound packets, causing every
transmitted TCP segment to be fragmented across two IPv6 packets, the
latter of which will only contain 8 bytes of actual payload.
RTAX_FEATURE_ALLFRAG is typically set on a route in response to
receving a ICMPv6 Packet Too Big message indicating a Path MTU of less
than 1280 bytes. 1280 bytes is the minimum IPv6 MTU, however ICMPv6
PTBs with MTU < 1280 are still valid, in particular when an IPv6
packet is sent to an IPv4 destination through a stateless translator.
Any ICMPv4 Need To Fragment packets originated from the IPv4 part of
the path will be translated to ICMPv6 PTB which may then indicate an
MTU of less than 1280.
The Linux kernel refuses to reduce the effective MTU to anything below
1280 bytes, instead it sets it to exactly 1280 bytes, and
RTAX_FEATURE_ALLFRAG is also set. However, the TCP segment size appears
to be set to 1240 bytes (1280 Path MTU - 40 bytes of IPv6 header),
instead of 1232 (additionally taking into account the 8 bytes required
by the IPv6 Fragmentation extension header).
This in turn results in rather inefficient transmission, as every
transmitted TCP segment now is split in two fragments containing
1232+8 bytes of payload.
After this patch, all the outgoing packets that includes a
Fragmentation header all are "atomic" or "non-fragmented" fragments,
i.e., they both have Offset=0 and More Fragments=0.
With help from David S. Miller
Reported-by: Tore Anderson <tore@fud.no>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Maciej Żenczykowski <maze@google.com>
Cc: Tom Herbert <therbert@google.com>
Tested-by: Tore Anderson <tore@fud.no>
Signed-off-by: David S. Miller <davem@davemloft.net>
2012-04-24 14:37:38 +07:00
|
|
|
/* IPv6 adds a frag_hdr in case RTAX_FEATURE_ALLFRAG is set */
|
|
|
|
if (icsk->icsk_af_ops->net_frag_header_len) {
|
|
|
|
const struct dst_entry *dst = __sk_dst_get(sk);
|
|
|
|
|
|
|
|
if (dst && dst_allfrag(dst))
|
|
|
|
mss_now -= icsk->icsk_af_ops->net_frag_header_len;
|
|
|
|
}
|
|
|
|
|
2006-03-21 08:53:41 +07:00
|
|
|
/* Clamp it (mss_clamp does not include tcp options) */
|
|
|
|
if (mss_now > tp->rx_opt.mss_clamp)
|
|
|
|
mss_now = tp->rx_opt.mss_clamp;
|
|
|
|
|
|
|
|
/* Now subtract optional transport overhead */
|
|
|
|
mss_now -= icsk->icsk_ext_hdr_len;
|
|
|
|
|
|
|
|
/* Then reserve room for full set of TCP options and 8 bytes of data */
|
|
|
|
if (mss_now < 48)
|
|
|
|
mss_now = 48;
|
|
|
|
return mss_now;
|
|
|
|
}
|
|
|
|
|
2013-02-22 15:59:06 +07:00
|
|
|
/* Calculate MSS. Not accounting for SACKs here. */
|
|
|
|
int tcp_mtu_to_mss(struct sock *sk, int pmtu)
|
|
|
|
{
|
|
|
|
/* Subtract TCP options size, not including SACKs */
|
|
|
|
return __tcp_mtu_to_mss(sk, pmtu) -
|
|
|
|
(tcp_sk(sk)->tcp_header_len - sizeof(struct tcphdr));
|
|
|
|
}
|
|
|
|
|
2006-03-21 08:53:41 +07:00
|
|
|
/* Inverse of above */
|
ipv6: RTAX_FEATURE_ALLFRAG causes inefficient TCP segment sizing
Quoting Tore Anderson from :
https://bugzilla.kernel.org/show_bug.cgi?id=42572
When RTAX_FEATURE_ALLFRAG is set on a route, the effective TCP segment
size does not take into account the size of the IPv6 Fragmentation
header that needs to be included in outbound packets, causing every
transmitted TCP segment to be fragmented across two IPv6 packets, the
latter of which will only contain 8 bytes of actual payload.
RTAX_FEATURE_ALLFRAG is typically set on a route in response to
receving a ICMPv6 Packet Too Big message indicating a Path MTU of less
than 1280 bytes. 1280 bytes is the minimum IPv6 MTU, however ICMPv6
PTBs with MTU < 1280 are still valid, in particular when an IPv6
packet is sent to an IPv4 destination through a stateless translator.
Any ICMPv4 Need To Fragment packets originated from the IPv4 part of
the path will be translated to ICMPv6 PTB which may then indicate an
MTU of less than 1280.
The Linux kernel refuses to reduce the effective MTU to anything below
1280 bytes, instead it sets it to exactly 1280 bytes, and
RTAX_FEATURE_ALLFRAG is also set. However, the TCP segment size appears
to be set to 1240 bytes (1280 Path MTU - 40 bytes of IPv6 header),
instead of 1232 (additionally taking into account the 8 bytes required
by the IPv6 Fragmentation extension header).
This in turn results in rather inefficient transmission, as every
transmitted TCP segment now is split in two fragments containing
1232+8 bytes of payload.
After this patch, all the outgoing packets that includes a
Fragmentation header all are "atomic" or "non-fragmented" fragments,
i.e., they both have Offset=0 and More Fragments=0.
With help from David S. Miller
Reported-by: Tore Anderson <tore@fud.no>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Maciej Żenczykowski <maze@google.com>
Cc: Tom Herbert <therbert@google.com>
Tested-by: Tore Anderson <tore@fud.no>
Signed-off-by: David S. Miller <davem@davemloft.net>
2012-04-24 14:37:38 +07:00
|
|
|
int tcp_mss_to_mtu(struct sock *sk, int mss)
|
2006-03-21 08:53:41 +07:00
|
|
|
{
|
2011-10-21 16:22:42 +07:00
|
|
|
const struct tcp_sock *tp = tcp_sk(sk);
|
|
|
|
const struct inet_connection_sock *icsk = inet_csk(sk);
|
2006-03-21 08:53:41 +07:00
|
|
|
int mtu;
|
|
|
|
|
|
|
|
mtu = mss +
|
|
|
|
tp->tcp_header_len +
|
|
|
|
icsk->icsk_ext_hdr_len +
|
|
|
|
icsk->icsk_af_ops->net_header_len;
|
|
|
|
|
ipv6: RTAX_FEATURE_ALLFRAG causes inefficient TCP segment sizing
Quoting Tore Anderson from :
https://bugzilla.kernel.org/show_bug.cgi?id=42572
When RTAX_FEATURE_ALLFRAG is set on a route, the effective TCP segment
size does not take into account the size of the IPv6 Fragmentation
header that needs to be included in outbound packets, causing every
transmitted TCP segment to be fragmented across two IPv6 packets, the
latter of which will only contain 8 bytes of actual payload.
RTAX_FEATURE_ALLFRAG is typically set on a route in response to
receving a ICMPv6 Packet Too Big message indicating a Path MTU of less
than 1280 bytes. 1280 bytes is the minimum IPv6 MTU, however ICMPv6
PTBs with MTU < 1280 are still valid, in particular when an IPv6
packet is sent to an IPv4 destination through a stateless translator.
Any ICMPv4 Need To Fragment packets originated from the IPv4 part of
the path will be translated to ICMPv6 PTB which may then indicate an
MTU of less than 1280.
The Linux kernel refuses to reduce the effective MTU to anything below
1280 bytes, instead it sets it to exactly 1280 bytes, and
RTAX_FEATURE_ALLFRAG is also set. However, the TCP segment size appears
to be set to 1240 bytes (1280 Path MTU - 40 bytes of IPv6 header),
instead of 1232 (additionally taking into account the 8 bytes required
by the IPv6 Fragmentation extension header).
This in turn results in rather inefficient transmission, as every
transmitted TCP segment now is split in two fragments containing
1232+8 bytes of payload.
After this patch, all the outgoing packets that includes a
Fragmentation header all are "atomic" or "non-fragmented" fragments,
i.e., they both have Offset=0 and More Fragments=0.
With help from David S. Miller
Reported-by: Tore Anderson <tore@fud.no>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Maciej Żenczykowski <maze@google.com>
Cc: Tom Herbert <therbert@google.com>
Tested-by: Tore Anderson <tore@fud.no>
Signed-off-by: David S. Miller <davem@davemloft.net>
2012-04-24 14:37:38 +07:00
|
|
|
/* IPv6 adds a frag_hdr in case RTAX_FEATURE_ALLFRAG is set */
|
|
|
|
if (icsk->icsk_af_ops->net_frag_header_len) {
|
|
|
|
const struct dst_entry *dst = __sk_dst_get(sk);
|
|
|
|
|
|
|
|
if (dst && dst_allfrag(dst))
|
|
|
|
mtu += icsk->icsk_af_ops->net_frag_header_len;
|
|
|
|
}
|
2006-03-21 08:53:41 +07:00
|
|
|
return mtu;
|
|
|
|
}
|
|
|
|
|
2009-07-22 06:00:40 +07:00
|
|
|
/* MTU probing init per socket */
|
2006-03-21 08:53:41 +07:00
|
|
|
void tcp_mtup_init(struct sock *sk)
|
|
|
|
{
|
|
|
|
struct tcp_sock *tp = tcp_sk(sk);
|
|
|
|
struct inet_connection_sock *icsk = inet_csk(sk);
|
2015-02-10 08:53:16 +07:00
|
|
|
struct net *net = sock_net(sk);
|
2006-03-21 08:53:41 +07:00
|
|
|
|
2015-02-10 08:53:16 +07:00
|
|
|
icsk->icsk_mtup.enabled = net->ipv4.sysctl_tcp_mtu_probing > 1;
|
2006-03-21 08:53:41 +07:00
|
|
|
icsk->icsk_mtup.search_high = tp->rx_opt.mss_clamp + sizeof(struct tcphdr) +
|
2007-02-09 21:24:47 +07:00
|
|
|
icsk->icsk_af_ops->net_header_len;
|
2015-02-10 08:53:16 +07:00
|
|
|
icsk->icsk_mtup.search_low = tcp_mss_to_mtu(sk, net->ipv4.sysctl_tcp_base_mss);
|
2006-03-21 08:53:41 +07:00
|
|
|
icsk->icsk_mtup.probe_size = 0;
|
2015-03-06 10:18:24 +07:00
|
|
|
if (icsk->icsk_mtup.enabled)
|
|
|
|
icsk->icsk_mtup.probe_timestamp = tcp_time_stamp;
|
2006-03-21 08:53:41 +07:00
|
|
|
}
|
2010-07-10 04:22:10 +07:00
|
|
|
EXPORT_SYMBOL(tcp_mtup_init);
|
2006-03-21 08:53:41 +07:00
|
|
|
|
2005-04-17 05:20:36 +07:00
|
|
|
/* This function synchronize snd mss to current pmtu/exthdr set.
|
|
|
|
|
|
|
|
tp->rx_opt.user_mss is mss set by user by TCP_MAXSEG. It does NOT counts
|
|
|
|
for TCP options, but includes only bare TCP header.
|
|
|
|
|
|
|
|
tp->rx_opt.mss_clamp is mss negotiated at connection setup.
|
2005-11-11 08:13:47 +07:00
|
|
|
It is minimum of user_mss and mss received with SYN.
|
2005-04-17 05:20:36 +07:00
|
|
|
It also does not include TCP options.
|
|
|
|
|
2005-12-14 14:26:10 +07:00
|
|
|
inet_csk(sk)->icsk_pmtu_cookie is last pmtu, seen by this function.
|
2005-04-17 05:20:36 +07:00
|
|
|
|
|
|
|
tp->mss_cache is current effective sending mss, including
|
|
|
|
all tcp options except for SACKs. It is evaluated,
|
|
|
|
taking into account current pmtu, but never exceeds
|
|
|
|
tp->rx_opt.mss_clamp.
|
|
|
|
|
|
|
|
NOTE1. rfc1122 clearly states that advertised MSS
|
|
|
|
DOES NOT include either tcp or ip options.
|
|
|
|
|
2005-12-14 14:26:10 +07:00
|
|
|
NOTE2. inet_csk(sk)->icsk_pmtu_cookie and tp->mss_cache
|
|
|
|
are READ ONLY outside this function. --ANK (980731)
|
2005-04-17 05:20:36 +07:00
|
|
|
*/
|
|
|
|
unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu)
|
|
|
|
{
|
|
|
|
struct tcp_sock *tp = tcp_sk(sk);
|
2005-12-14 14:26:10 +07:00
|
|
|
struct inet_connection_sock *icsk = inet_csk(sk);
|
2006-03-21 08:53:41 +07:00
|
|
|
int mss_now;
|
2005-04-17 05:20:36 +07:00
|
|
|
|
2006-03-21 08:53:41 +07:00
|
|
|
if (icsk->icsk_mtup.search_high > pmtu)
|
|
|
|
icsk->icsk_mtup.search_high = pmtu;
|
2005-04-17 05:20:36 +07:00
|
|
|
|
2006-03-21 08:53:41 +07:00
|
|
|
mss_now = tcp_mtu_to_mss(sk, pmtu);
|
2008-01-01 05:57:40 +07:00
|
|
|
mss_now = tcp_bound_to_half_wnd(tp, mss_now);
|
2005-04-17 05:20:36 +07:00
|
|
|
|
|
|
|
/* And store cached results */
|
2005-12-14 14:26:10 +07:00
|
|
|
icsk->icsk_pmtu_cookie = pmtu;
|
2006-03-21 08:53:41 +07:00
|
|
|
if (icsk->icsk_mtup.enabled)
|
|
|
|
mss_now = min(mss_now, tcp_mtu_to_mss(sk, icsk->icsk_mtup.search_low));
|
2005-07-06 05:24:38 +07:00
|
|
|
tp->mss_cache = mss_now;
|
2005-04-17 05:20:36 +07:00
|
|
|
|
|
|
|
return mss_now;
|
|
|
|
}
|
2010-07-10 04:22:10 +07:00
|
|
|
EXPORT_SYMBOL(tcp_sync_mss);
|
2005-04-17 05:20:36 +07:00
|
|
|
|
|
|
|
/* Compute the current effective MSS, taking SACKs and IP options,
|
|
|
|
* and even PMTU discovery events into account.
|
|
|
|
*/
|
2009-03-14 21:23:05 +07:00
|
|
|
unsigned int tcp_current_mss(struct sock *sk)
|
2005-04-17 05:20:36 +07:00
|
|
|
{
|
2011-10-21 16:22:42 +07:00
|
|
|
const struct tcp_sock *tp = tcp_sk(sk);
|
|
|
|
const struct dst_entry *dst = __sk_dst_get(sk);
|
2005-07-06 05:24:38 +07:00
|
|
|
u32 mss_now;
|
2012-04-15 12:58:06 +07:00
|
|
|
unsigned int header_len;
|
2008-07-19 14:04:31 +07:00
|
|
|
struct tcp_out_options opts;
|
|
|
|
struct tcp_md5sig_key *md5;
|
2005-07-06 05:24:38 +07:00
|
|
|
|
|
|
|
mss_now = tp->mss_cache;
|
|
|
|
|
2005-04-17 05:20:36 +07:00
|
|
|
if (dst) {
|
|
|
|
u32 mtu = dst_mtu(dst);
|
2005-12-14 14:26:10 +07:00
|
|
|
if (mtu != inet_csk(sk)->icsk_pmtu_cookie)
|
2005-04-17 05:20:36 +07:00
|
|
|
mss_now = tcp_sync_mss(sk, mtu);
|
|
|
|
}
|
|
|
|
|
2008-07-19 14:04:31 +07:00
|
|
|
header_len = tcp_established_options(sk, NULL, &opts, &md5) +
|
|
|
|
sizeof(struct tcphdr);
|
|
|
|
/* The mss_cache is sized based on tp->tcp_header_len, which assumes
|
|
|
|
* some common options. If this is an odd packet (because we have SACK
|
|
|
|
* blocks etc) then our calculated header_len will be different, and
|
|
|
|
* we have to adjust mss_now correspondingly */
|
|
|
|
if (header_len != tp->tcp_header_len) {
|
|
|
|
int delta = (int) header_len - tp->tcp_header_len;
|
|
|
|
mss_now -= delta;
|
|
|
|
}
|
2006-11-15 10:07:45 +07:00
|
|
|
|
2005-04-17 05:20:36 +07:00
|
|
|
return mss_now;
|
|
|
|
}
|
|
|
|
|
2014-04-18 11:27:46 +07:00
|
|
|
/* RFC2861, slow part. Adjust cwnd, after it was not full during one rto.
|
|
|
|
* As additional protections, we do not touch cwnd in retransmission phases,
|
|
|
|
* and if application hit its sndbuf limit recently.
|
|
|
|
*/
|
|
|
|
static void tcp_cwnd_application_limited(struct sock *sk)
|
|
|
|
{
|
|
|
|
struct tcp_sock *tp = tcp_sk(sk);
|
|
|
|
|
|
|
|
if (inet_csk(sk)->icsk_ca_state == TCP_CA_Open &&
|
|
|
|
sk->sk_socket && !test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
|
|
|
|
/* Limited by application or receiver window. */
|
|
|
|
u32 init_win = tcp_init_cwnd(tp, __sk_dst_get(sk));
|
|
|
|
u32 win_used = max(tp->snd_cwnd_used, init_win);
|
|
|
|
if (win_used < tp->snd_cwnd) {
|
|
|
|
tp->snd_ssthresh = tcp_current_ssthresh(sk);
|
|
|
|
tp->snd_cwnd = (tp->snd_cwnd + win_used) >> 1;
|
|
|
|
}
|
|
|
|
tp->snd_cwnd_used = 0;
|
|
|
|
}
|
|
|
|
tp->snd_cwnd_stamp = tcp_time_stamp;
|
|
|
|
}
|
|
|
|
|
2014-05-22 21:41:08 +07:00
|
|
|
static void tcp_cwnd_validate(struct sock *sk, bool is_cwnd_limited)
|
2005-07-06 05:18:51 +07:00
|
|
|
{
|
[TCP]: Sed magic converts func(sk, tp, ...) -> func(sk, ...)
This is (mostly) automated change using magic:
sed -e '/struct sock \*sk/ N' -e '/struct sock \*sk/ N'
-e '/struct sock \*sk/ N' -e '/struct sock \*sk/ N'
-e 's|struct sock \*sk,[\n\t ]*struct tcp_sock \*tp\([^{]*\n{\n\)|
struct sock \*sk\1\tstruct tcp_sock *tp = tcp_sk(sk);\n|g'
-e 's|struct sock \*sk, struct tcp_sock \*tp|
struct sock \*sk|g' -e 's|sk, tp\([^-]\)|sk\1|g'
Fixed four unused variable (tp) warnings that were introduced.
In addition, manually added newlines after local variables and
tweaked function arguments positioning.
$ gcc --version
gcc (GCC) 4.1.1 20060525 (Red Hat 4.1.1-1)
...
$ codiff -fV built-in.o.old built-in.o.new
net/ipv4/route.c:
rt_cache_flush | +14
1 function changed, 14 bytes added
net/ipv4/tcp.c:
tcp_setsockopt | -5
tcp_sendpage | -25
tcp_sendmsg | -16
3 functions changed, 46 bytes removed
net/ipv4/tcp_input.c:
tcp_try_undo_recovery | +3
tcp_try_undo_dsack | +2
tcp_mark_head_lost | -12
tcp_ack | -15
tcp_event_data_recv | -32
tcp_rcv_state_process | -10
tcp_rcv_established | +1
7 functions changed, 6 bytes added, 69 bytes removed, diff: -63
net/ipv4/tcp_output.c:
update_send_head | -9
tcp_transmit_skb | +19
tcp_cwnd_validate | +1
tcp_write_wakeup | -17
__tcp_push_pending_frames | -25
tcp_push_one | -8
tcp_send_fin | -4
7 functions changed, 20 bytes added, 63 bytes removed, diff: -43
built-in.o.new:
18 functions changed, 40 bytes added, 178 bytes removed, diff: -138
Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>
2007-04-21 12:18:02 +07:00
|
|
|
struct tcp_sock *tp = tcp_sk(sk);
|
2005-07-06 05:18:51 +07:00
|
|
|
|
2014-05-22 21:41:08 +07:00
|
|
|
/* Track the maximum number of outstanding packets in each
|
|
|
|
* window, and remember whether we were cwnd-limited then.
|
|
|
|
*/
|
|
|
|
if (!before(tp->snd_una, tp->max_packets_seq) ||
|
|
|
|
tp->packets_out > tp->max_packets_out) {
|
|
|
|
tp->max_packets_out = tp->packets_out;
|
|
|
|
tp->max_packets_seq = tp->snd_nxt;
|
|
|
|
tp->is_cwnd_limited = is_cwnd_limited;
|
|
|
|
}
|
tcp: fix cwnd limited checking to improve congestion control
Yuchung discovered tcp_is_cwnd_limited() was returning false in
slow start phase even if the application filled the socket write queue.
All congestion modules take into account tcp_is_cwnd_limited()
before increasing cwnd, so this behavior limits slow start from
probing the bandwidth at full speed.
The problem is that even if write queue is full (aka we are _not_
application limited), cwnd can be under utilized if TSO should auto
defer or TCP Small queues decided to hold packets.
So the in_flight can be kept to smaller value, and we can get to the
point tcp_is_cwnd_limited() returns false.
With TCP Small Queues and FQ/pacing, this issue is more visible.
We fix this by having tcp_cwnd_validate(), which is supposed to track
such things, take into account unsent_segs, the number of segs that we
are not sending at the moment due to TSO or TSQ, but intend to send
real soon. Then when we are cwnd-limited, remember this fact while we
are processing the window of ACKs that comes back.
For example, suppose we have a brand new connection with cwnd=10; we
are in slow start, and we send a flight of 9 packets. By the time we
have received ACKs for all 9 packets we want our cwnd to be 18.
We implement this by setting tp->lsnd_pending to 9, and
considering ourselves to be cwnd-limited while cwnd is less than
twice tp->lsnd_pending (2*9 -> 18).
This makes tcp_is_cwnd_limited() more understandable, by removing
the GSO/TSO kludge, that tried to work around the issue.
Note the in_flight parameter can be removed in a followup cleanup
patch.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2014-05-01 01:58:13 +07:00
|
|
|
|
2014-05-03 11:18:05 +07:00
|
|
|
if (tcp_is_cwnd_limited(sk)) {
|
2005-07-06 05:18:51 +07:00
|
|
|
/* Network is feed fully. */
|
|
|
|
tp->snd_cwnd_used = 0;
|
|
|
|
tp->snd_cwnd_stamp = tcp_time_stamp;
|
|
|
|
} else {
|
|
|
|
/* Network starves. */
|
|
|
|
if (tp->packets_out > tp->snd_cwnd_used)
|
|
|
|
tp->snd_cwnd_used = tp->packets_out;
|
|
|
|
|
2007-04-10 03:23:14 +07:00
|
|
|
if (sysctl_tcp_slow_start_after_idle &&
|
|
|
|
(s32)(tcp_time_stamp - tp->snd_cwnd_stamp) >= inet_csk(sk)->icsk_rto)
|
2005-07-06 05:18:51 +07:00
|
|
|
tcp_cwnd_application_limited(sk);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
tcp: refine TSO splits
While investigating performance problems on small RPC workloads,
I noticed linux TCP stack was always splitting the last TSO skb
into two parts (skbs). One being a multiple of MSS, and a small one
with the Push flag. This split is done even if TCP_NODELAY is set,
or if no small packet is in flight.
Example with request/response of 4K/4K
IP A > B: . ack 68432 win 2783 <nop,nop,timestamp 6524593 6525001>
IP A > B: . 65537:68433(2896) ack 69632 win 2783 <nop,nop,timestamp 6524593 6525001>
IP A > B: P 68433:69633(1200) ack 69632 win 2783 <nop,nop,timestamp 6524593 6525001>
IP B > A: . ack 68433 win 2768 <nop,nop,timestamp 6525001 6524593>
IP B > A: . 69632:72528(2896) ack 69633 win 2768 <nop,nop,timestamp 6525001 6524593>
IP B > A: P 72528:73728(1200) ack 69633 win 2768 <nop,nop,timestamp 6525001 6524593>
IP A > B: . ack 72528 win 2783 <nop,nop,timestamp 6524593 6525001>
IP A > B: . 69633:72529(2896) ack 73728 win 2783 <nop,nop,timestamp 6524593 6525001>
IP A > B: P 72529:73729(1200) ack 73728 win 2783 <nop,nop,timestamp 6524593 6525001>
We can avoid this split by including the Nagle tests at the right place.
Note : If some NIC had trouble sending TSO packets with a partial
last segment, we would have hit the problem in GRO/forwarding workload already.
tcp_minshall_update() is moved to tcp_output.c and is updated as we might
feed a TSO packet with a partial last segment.
This patch tremendously improves performance, as the traffic now looks
like :
IP A > B: . ack 98304 win 2783 <nop,nop,timestamp 6834277 6834685>
IP A > B: P 94209:98305(4096) ack 98304 win 2783 <nop,nop,timestamp 6834277 6834685>
IP B > A: . ack 98305 win 2768 <nop,nop,timestamp 6834686 6834277>
IP B > A: P 98304:102400(4096) ack 98305 win 2768 <nop,nop,timestamp 6834686 6834277>
IP A > B: . ack 102400 win 2783 <nop,nop,timestamp 6834279 6834686>
IP A > B: P 98305:102401(4096) ack 102400 win 2783 <nop,nop,timestamp 6834279 6834686>
IP B > A: . ack 102401 win 2768 <nop,nop,timestamp 6834687 6834279>
IP B > A: P 102400:106496(4096) ack 102401 win 2768 <nop,nop,timestamp 6834687 6834279>
IP A > B: . ack 106496 win 2783 <nop,nop,timestamp 6834280 6834687>
IP A > B: P 102401:106497(4096) ack 106496 win 2783 <nop,nop,timestamp 6834280 6834687>
IP B > A: . ack 106497 win 2768 <nop,nop,timestamp 6834688 6834280>
IP B > A: P 106496:110592(4096) ack 106497 win 2768 <nop,nop,timestamp 6834688 6834280>
Before :
lpq83:~# nstat >/dev/null;perf stat ./super_netperf 200 -t TCP_RR -H lpq84 -l 20 -- -r 4K,4K
280774
Performance counter stats for './super_netperf 200 -t TCP_RR -H lpq84 -l 20 -- -r 4K,4K':
205719.049006 task-clock # 9.278 CPUs utilized
8,449,968 context-switches # 0.041 M/sec
1,935,997 CPU-migrations # 0.009 M/sec
160,541 page-faults # 0.780 K/sec
548,478,722,290 cycles # 2.666 GHz [83.20%]
455,240,670,857 stalled-cycles-frontend # 83.00% frontend cycles idle [83.48%]
272,881,454,275 stalled-cycles-backend # 49.75% backend cycles idle [66.73%]
166,091,460,030 instructions # 0.30 insns per cycle
# 2.74 stalled cycles per insn [83.39%]
29,150,229,399 branches # 141.699 M/sec [83.30%]
1,943,814,026 branch-misses # 6.67% of all branches [83.32%]
22.173517844 seconds time elapsed
lpq83:~# nstat | egrep "IpOutRequests|IpExtOutOctets"
IpOutRequests 16851063 0.0
IpExtOutOctets 23878580777 0.0
After patch :
lpq83:~# nstat >/dev/null;perf stat ./super_netperf 200 -t TCP_RR -H lpq84 -l 20 -- -r 4K,4K
280877
Performance counter stats for './super_netperf 200 -t TCP_RR -H lpq84 -l 20 -- -r 4K,4K':
107496.071918 task-clock # 4.847 CPUs utilized
5,635,458 context-switches # 0.052 M/sec
1,374,707 CPU-migrations # 0.013 M/sec
160,920 page-faults # 0.001 M/sec
281,500,010,924 cycles # 2.619 GHz [83.28%]
228,865,069,307 stalled-cycles-frontend # 81.30% frontend cycles idle [83.38%]
142,462,742,658 stalled-cycles-backend # 50.61% backend cycles idle [66.81%]
95,227,712,566 instructions # 0.34 insns per cycle
# 2.40 stalled cycles per insn [83.43%]
16,209,868,171 branches # 150.795 M/sec [83.20%]
874,252,952 branch-misses # 5.39% of all branches [83.37%]
22.175821286 seconds time elapsed
lpq83:~# nstat | egrep "IpOutRequests|IpExtOutOctets"
IpOutRequests 11239428 0.0
IpExtOutOctets 23595191035 0.0
Indeed, the occupancy of tx skbs (IpExtOutOctets/IpOutRequests) is higher :
2099 instead of 1417, thus helping GRO to be more efficient when using FQ packet
scheduler.
Many thanks to Neal for review and ideas.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Yuchung Cheng <ycheng@google.com>
Cc: Neal Cardwell <ncardwell@google.com>
Cc: Nandita Dukkipati <nanditad@google.com>
Cc: Van Jacobson <vanj@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Tested-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-12-14 04:51:23 +07:00
|
|
|
/* Minshall's variant of the Nagle send check. */
|
|
|
|
static bool tcp_minshall_check(const struct tcp_sock *tp)
|
|
|
|
{
|
|
|
|
return after(tp->snd_sml, tp->snd_una) &&
|
|
|
|
!after(tp->snd_sml, tp->snd_nxt);
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Update snd_sml if this skb is under mss
|
|
|
|
* Note that a TSO packet might end with a sub-mss segment
|
|
|
|
* The test is really :
|
|
|
|
* if ((skb->len % mss) != 0)
|
|
|
|
* tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
|
|
|
|
* But we can avoid doing the divide again given we already have
|
|
|
|
* skb_pcount = skb->len / mss_now
|
2007-12-25 12:33:45 +07:00
|
|
|
*/
|
tcp: refine TSO splits
While investigating performance problems on small RPC workloads,
I noticed linux TCP stack was always splitting the last TSO skb
into two parts (skbs). One being a multiple of MSS, and a small one
with the Push flag. This split is done even if TCP_NODELAY is set,
or if no small packet is in flight.
Example with request/response of 4K/4K
IP A > B: . ack 68432 win 2783 <nop,nop,timestamp 6524593 6525001>
IP A > B: . 65537:68433(2896) ack 69632 win 2783 <nop,nop,timestamp 6524593 6525001>
IP A > B: P 68433:69633(1200) ack 69632 win 2783 <nop,nop,timestamp 6524593 6525001>
IP B > A: . ack 68433 win 2768 <nop,nop,timestamp 6525001 6524593>
IP B > A: . 69632:72528(2896) ack 69633 win 2768 <nop,nop,timestamp 6525001 6524593>
IP B > A: P 72528:73728(1200) ack 69633 win 2768 <nop,nop,timestamp 6525001 6524593>
IP A > B: . ack 72528 win 2783 <nop,nop,timestamp 6524593 6525001>
IP A > B: . 69633:72529(2896) ack 73728 win 2783 <nop,nop,timestamp 6524593 6525001>
IP A > B: P 72529:73729(1200) ack 73728 win 2783 <nop,nop,timestamp 6524593 6525001>
We can avoid this split by including the Nagle tests at the right place.
Note : If some NIC had trouble sending TSO packets with a partial
last segment, we would have hit the problem in GRO/forwarding workload already.
tcp_minshall_update() is moved to tcp_output.c and is updated as we might
feed a TSO packet with a partial last segment.
This patch tremendously improves performance, as the traffic now looks
like :
IP A > B: . ack 98304 win 2783 <nop,nop,timestamp 6834277 6834685>
IP A > B: P 94209:98305(4096) ack 98304 win 2783 <nop,nop,timestamp 6834277 6834685>
IP B > A: . ack 98305 win 2768 <nop,nop,timestamp 6834686 6834277>
IP B > A: P 98304:102400(4096) ack 98305 win 2768 <nop,nop,timestamp 6834686 6834277>
IP A > B: . ack 102400 win 2783 <nop,nop,timestamp 6834279 6834686>
IP A > B: P 98305:102401(4096) ack 102400 win 2783 <nop,nop,timestamp 6834279 6834686>
IP B > A: . ack 102401 win 2768 <nop,nop,timestamp 6834687 6834279>
IP B > A: P 102400:106496(4096) ack 102401 win 2768 <nop,nop,timestamp 6834687 6834279>
IP A > B: . ack 106496 win 2783 <nop,nop,timestamp 6834280 6834687>
IP A > B: P 102401:106497(4096) ack 106496 win 2783 <nop,nop,timestamp 6834280 6834687>
IP B > A: . ack 106497 win 2768 <nop,nop,timestamp 6834688 6834280>
IP B > A: P 106496:110592(4096) ack 106497 win 2768 <nop,nop,timestamp 6834688 6834280>
Before :
lpq83:~# nstat >/dev/null;perf stat ./super_netperf 200 -t TCP_RR -H lpq84 -l 20 -- -r 4K,4K
280774
Performance counter stats for './super_netperf 200 -t TCP_RR -H lpq84 -l 20 -- -r 4K,4K':
205719.049006 task-clock # 9.278 CPUs utilized
8,449,968 context-switches # 0.041 M/sec
1,935,997 CPU-migrations # 0.009 M/sec
160,541 page-faults # 0.780 K/sec
548,478,722,290 cycles # 2.666 GHz [83.20%]
455,240,670,857 stalled-cycles-frontend # 83.00% frontend cycles idle [83.48%]
272,881,454,275 stalled-cycles-backend # 49.75% backend cycles idle [66.73%]
166,091,460,030 instructions # 0.30 insns per cycle
# 2.74 stalled cycles per insn [83.39%]
29,150,229,399 branches # 141.699 M/sec [83.30%]
1,943,814,026 branch-misses # 6.67% of all branches [83.32%]
22.173517844 seconds time elapsed
lpq83:~# nstat | egrep "IpOutRequests|IpExtOutOctets"
IpOutRequests 16851063 0.0
IpExtOutOctets 23878580777 0.0
After patch :
lpq83:~# nstat >/dev/null;perf stat ./super_netperf 200 -t TCP_RR -H lpq84 -l 20 -- -r 4K,4K
280877
Performance counter stats for './super_netperf 200 -t TCP_RR -H lpq84 -l 20 -- -r 4K,4K':
107496.071918 task-clock # 4.847 CPUs utilized
5,635,458 context-switches # 0.052 M/sec
1,374,707 CPU-migrations # 0.013 M/sec
160,920 page-faults # 0.001 M/sec
281,500,010,924 cycles # 2.619 GHz [83.28%]
228,865,069,307 stalled-cycles-frontend # 81.30% frontend cycles idle [83.38%]
142,462,742,658 stalled-cycles-backend # 50.61% backend cycles idle [66.81%]
95,227,712,566 instructions # 0.34 insns per cycle
# 2.40 stalled cycles per insn [83.43%]
16,209,868,171 branches # 150.795 M/sec [83.20%]
874,252,952 branch-misses # 5.39% of all branches [83.37%]
22.175821286 seconds time elapsed
lpq83:~# nstat | egrep "IpOutRequests|IpExtOutOctets"
IpOutRequests 11239428 0.0
IpExtOutOctets 23595191035 0.0
Indeed, the occupancy of tx skbs (IpExtOutOctets/IpOutRequests) is higher :
2099 instead of 1417, thus helping GRO to be more efficient when using FQ packet
scheduler.
Many thanks to Neal for review and ideas.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Yuchung Cheng <ycheng@google.com>
Cc: Neal Cardwell <ncardwell@google.com>
Cc: Nandita Dukkipati <nanditad@google.com>
Cc: Van Jacobson <vanj@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Tested-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-12-14 04:51:23 +07:00
|
|
|
static void tcp_minshall_update(struct tcp_sock *tp, unsigned int mss_now,
|
|
|
|
const struct sk_buff *skb)
|
|
|
|
{
|
|
|
|
if (skb->len < tcp_skb_pcount(skb) * mss_now)
|
|
|
|
tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Return false, if packet can be sent now without violation Nagle's rules:
|
|
|
|
* 1. It is full sized. (provided by caller in %partial bool)
|
|
|
|
* 2. Or it contains FIN. (already checked by caller)
|
|
|
|
* 3. Or TCP_CORK is not set, and TCP_NODELAY is set.
|
|
|
|
* 4. Or TCP_CORK is not set, and all sent packets are ACKed.
|
|
|
|
* With Minshall's modification: all sent small packets are ACKed.
|
|
|
|
*/
|
|
|
|
static bool tcp_nagle_check(bool partial, const struct tcp_sock *tp,
|
2014-03-24 13:49:34 +07:00
|
|
|
int nonagle)
|
tcp: refine TSO splits
While investigating performance problems on small RPC workloads,
I noticed linux TCP stack was always splitting the last TSO skb
into two parts (skbs). One being a multiple of MSS, and a small one
with the Push flag. This split is done even if TCP_NODELAY is set,
or if no small packet is in flight.
Example with request/response of 4K/4K
IP A > B: . ack 68432 win 2783 <nop,nop,timestamp 6524593 6525001>
IP A > B: . 65537:68433(2896) ack 69632 win 2783 <nop,nop,timestamp 6524593 6525001>
IP A > B: P 68433:69633(1200) ack 69632 win 2783 <nop,nop,timestamp 6524593 6525001>
IP B > A: . ack 68433 win 2768 <nop,nop,timestamp 6525001 6524593>
IP B > A: . 69632:72528(2896) ack 69633 win 2768 <nop,nop,timestamp 6525001 6524593>
IP B > A: P 72528:73728(1200) ack 69633 win 2768 <nop,nop,timestamp 6525001 6524593>
IP A > B: . ack 72528 win 2783 <nop,nop,timestamp 6524593 6525001>
IP A > B: . 69633:72529(2896) ack 73728 win 2783 <nop,nop,timestamp 6524593 6525001>
IP A > B: P 72529:73729(1200) ack 73728 win 2783 <nop,nop,timestamp 6524593 6525001>
We can avoid this split by including the Nagle tests at the right place.
Note : If some NIC had trouble sending TSO packets with a partial
last segment, we would have hit the problem in GRO/forwarding workload already.
tcp_minshall_update() is moved to tcp_output.c and is updated as we might
feed a TSO packet with a partial last segment.
This patch tremendously improves performance, as the traffic now looks
like :
IP A > B: . ack 98304 win 2783 <nop,nop,timestamp 6834277 6834685>
IP A > B: P 94209:98305(4096) ack 98304 win 2783 <nop,nop,timestamp 6834277 6834685>
IP B > A: . ack 98305 win 2768 <nop,nop,timestamp 6834686 6834277>
IP B > A: P 98304:102400(4096) ack 98305 win 2768 <nop,nop,timestamp 6834686 6834277>
IP A > B: . ack 102400 win 2783 <nop,nop,timestamp 6834279 6834686>
IP A > B: P 98305:102401(4096) ack 102400 win 2783 <nop,nop,timestamp 6834279 6834686>
IP B > A: . ack 102401 win 2768 <nop,nop,timestamp 6834687 6834279>
IP B > A: P 102400:106496(4096) ack 102401 win 2768 <nop,nop,timestamp 6834687 6834279>
IP A > B: . ack 106496 win 2783 <nop,nop,timestamp 6834280 6834687>
IP A > B: P 102401:106497(4096) ack 106496 win 2783 <nop,nop,timestamp 6834280 6834687>
IP B > A: . ack 106497 win 2768 <nop,nop,timestamp 6834688 6834280>
IP B > A: P 106496:110592(4096) ack 106497 win 2768 <nop,nop,timestamp 6834688 6834280>
Before :
lpq83:~# nstat >/dev/null;perf stat ./super_netperf 200 -t TCP_RR -H lpq84 -l 20 -- -r 4K,4K
280774
Performance counter stats for './super_netperf 200 -t TCP_RR -H lpq84 -l 20 -- -r 4K,4K':
205719.049006 task-clock # 9.278 CPUs utilized
8,449,968 context-switches # 0.041 M/sec
1,935,997 CPU-migrations # 0.009 M/sec
160,541 page-faults # 0.780 K/sec
548,478,722,290 cycles # 2.666 GHz [83.20%]
455,240,670,857 stalled-cycles-frontend # 83.00% frontend cycles idle [83.48%]
272,881,454,275 stalled-cycles-backend # 49.75% backend cycles idle [66.73%]
166,091,460,030 instructions # 0.30 insns per cycle
# 2.74 stalled cycles per insn [83.39%]
29,150,229,399 branches # 141.699 M/sec [83.30%]
1,943,814,026 branch-misses # 6.67% of all branches [83.32%]
22.173517844 seconds time elapsed
lpq83:~# nstat | egrep "IpOutRequests|IpExtOutOctets"
IpOutRequests 16851063 0.0
IpExtOutOctets 23878580777 0.0
After patch :
lpq83:~# nstat >/dev/null;perf stat ./super_netperf 200 -t TCP_RR -H lpq84 -l 20 -- -r 4K,4K
280877
Performance counter stats for './super_netperf 200 -t TCP_RR -H lpq84 -l 20 -- -r 4K,4K':
107496.071918 task-clock # 4.847 CPUs utilized
5,635,458 context-switches # 0.052 M/sec
1,374,707 CPU-migrations # 0.013 M/sec
160,920 page-faults # 0.001 M/sec
281,500,010,924 cycles # 2.619 GHz [83.28%]
228,865,069,307 stalled-cycles-frontend # 81.30% frontend cycles idle [83.38%]
142,462,742,658 stalled-cycles-backend # 50.61% backend cycles idle [66.81%]
95,227,712,566 instructions # 0.34 insns per cycle
# 2.40 stalled cycles per insn [83.43%]
16,209,868,171 branches # 150.795 M/sec [83.20%]
874,252,952 branch-misses # 5.39% of all branches [83.37%]
22.175821286 seconds time elapsed
lpq83:~# nstat | egrep "IpOutRequests|IpExtOutOctets"
IpOutRequests 11239428 0.0
IpExtOutOctets 23595191035 0.0
Indeed, the occupancy of tx skbs (IpExtOutOctets/IpOutRequests) is higher :
2099 instead of 1417, thus helping GRO to be more efficient when using FQ packet
scheduler.
Many thanks to Neal for review and ideas.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Yuchung Cheng <ycheng@google.com>
Cc: Neal Cardwell <ncardwell@google.com>
Cc: Nandita Dukkipati <nanditad@google.com>
Cc: Van Jacobson <vanj@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Tested-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-12-14 04:51:23 +07:00
|
|
|
{
|
|
|
|
return partial &&
|
|
|
|
((nonagle & TCP_NAGLE_CORK) ||
|
|
|
|
(!nonagle && tp->packets_out && tcp_minshall_check(tp)));
|
|
|
|
}
|
tcp: refine TSO autosizing
Commit 95bd09eb2750 ("tcp: TSO packets automatic sizing") tried to
control TSO size, but did this at the wrong place (sendmsg() time)
At sendmsg() time, we might have a pessimistic view of flow rate,
and we end up building very small skbs (with 2 MSS per skb).
This is bad because :
- It sends small TSO packets even in Slow Start where rate quickly
increases.
- It tends to make socket write queue very big, increasing tcp_ack()
processing time, but also increasing memory needs, not necessarily
accounted for, as fast clones overhead is currently ignored.
- Lower GRO efficiency and more ACK packets.
Servers with a lot of small lived connections suffer from this.
Lets instead fill skbs as much as possible (64KB of payload), but split
them at xmit time, when we have a precise idea of the flow rate.
skb split is actually quite efficient.
Patch looks bigger than necessary, because TCP Small Queue decision now
has to take place after the eventual split.
As Neal suggested, introduce a new tcp_tso_autosize() helper, so that
tcp_tso_should_defer() can be synchronized on same goal.
Rename tp->xmit_size_goal_segs to tp->gso_segs, as this variable
contains number of mss that we can put in GSO packet, and is not
related to the autosizing goal anymore.
Tested:
40 ms rtt link
nstat >/dev/null
netperf -H remote -l -2000000 -- -s 1000000
nstat | egrep "IpInReceives|IpOutRequests|TcpOutSegs|IpExtOutOctets"
Before patch :
Recv Send Send
Socket Socket Message Elapsed
Size Size Size Time Throughput
bytes bytes bytes secs. 10^6bits/s
87380 2000000 2000000 0.36 44.22
IpInReceives 600 0.0
IpOutRequests 599 0.0
TcpOutSegs 1397 0.0
IpExtOutOctets 2033249 0.0
After patch :
Recv Send Send
Socket Socket Message Elapsed
Size Size Size Time Throughput
bytes bytes bytes secs. 10^6bits/sec
87380 2000000 2000000 0.36 44.27
IpInReceives 221 0.0
IpOutRequests 232 0.0
TcpOutSegs 1397 0.0
IpExtOutOctets 2013953 0.0
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Acked-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2014-12-08 03:22:18 +07:00
|
|
|
|
|
|
|
/* Return how many segs we'd like on a TSO packet,
|
|
|
|
* to send one TSO packet per ms
|
|
|
|
*/
|
|
|
|
static u32 tcp_tso_autosize(const struct sock *sk, unsigned int mss_now)
|
|
|
|
{
|
|
|
|
u32 bytes, segs;
|
|
|
|
|
|
|
|
bytes = min(sk->sk_pacing_rate >> 10,
|
|
|
|
sk->sk_gso_max_size - 1 - MAX_TCP_HEADER);
|
|
|
|
|
|
|
|
/* Goal is to send at least one packet per ms,
|
|
|
|
* not one big TSO packet every 100 ms.
|
|
|
|
* This preserves ACK clocking and is consistent
|
|
|
|
* with tcp_tso_should_defer() heuristic.
|
|
|
|
*/
|
|
|
|
segs = max_t(u32, bytes / mss_now, sysctl_tcp_min_tso_segs);
|
|
|
|
|
|
|
|
return min_t(u32, segs, sk->sk_gso_max_segs);
|
|
|
|
}
|
|
|
|
|
tcp: refine TSO splits
While investigating performance problems on small RPC workloads,
I noticed linux TCP stack was always splitting the last TSO skb
into two parts (skbs). One being a multiple of MSS, and a small one
with the Push flag. This split is done even if TCP_NODELAY is set,
or if no small packet is in flight.
Example with request/response of 4K/4K
IP A > B: . ack 68432 win 2783 <nop,nop,timestamp 6524593 6525001>
IP A > B: . 65537:68433(2896) ack 69632 win 2783 <nop,nop,timestamp 6524593 6525001>
IP A > B: P 68433:69633(1200) ack 69632 win 2783 <nop,nop,timestamp 6524593 6525001>
IP B > A: . ack 68433 win 2768 <nop,nop,timestamp 6525001 6524593>
IP B > A: . 69632:72528(2896) ack 69633 win 2768 <nop,nop,timestamp 6525001 6524593>
IP B > A: P 72528:73728(1200) ack 69633 win 2768 <nop,nop,timestamp 6525001 6524593>
IP A > B: . ack 72528 win 2783 <nop,nop,timestamp 6524593 6525001>
IP A > B: . 69633:72529(2896) ack 73728 win 2783 <nop,nop,timestamp 6524593 6525001>
IP A > B: P 72529:73729(1200) ack 73728 win 2783 <nop,nop,timestamp 6524593 6525001>
We can avoid this split by including the Nagle tests at the right place.
Note : If some NIC had trouble sending TSO packets with a partial
last segment, we would have hit the problem in GRO/forwarding workload already.
tcp_minshall_update() is moved to tcp_output.c and is updated as we might
feed a TSO packet with a partial last segment.
This patch tremendously improves performance, as the traffic now looks
like :
IP A > B: . ack 98304 win 2783 <nop,nop,timestamp 6834277 6834685>
IP A > B: P 94209:98305(4096) ack 98304 win 2783 <nop,nop,timestamp 6834277 6834685>
IP B > A: . ack 98305 win 2768 <nop,nop,timestamp 6834686 6834277>
IP B > A: P 98304:102400(4096) ack 98305 win 2768 <nop,nop,timestamp 6834686 6834277>
IP A > B: . ack 102400 win 2783 <nop,nop,timestamp 6834279 6834686>
IP A > B: P 98305:102401(4096) ack 102400 win 2783 <nop,nop,timestamp 6834279 6834686>
IP B > A: . ack 102401 win 2768 <nop,nop,timestamp 6834687 6834279>
IP B > A: P 102400:106496(4096) ack 102401 win 2768 <nop,nop,timestamp 6834687 6834279>
IP A > B: . ack 106496 win 2783 <nop,nop,timestamp 6834280 6834687>
IP A > B: P 102401:106497(4096) ack 106496 win 2783 <nop,nop,timestamp 6834280 6834687>
IP B > A: . ack 106497 win 2768 <nop,nop,timestamp 6834688 6834280>
IP B > A: P 106496:110592(4096) ack 106497 win 2768 <nop,nop,timestamp 6834688 6834280>
Before :
lpq83:~# nstat >/dev/null;perf stat ./super_netperf 200 -t TCP_RR -H lpq84 -l 20 -- -r 4K,4K
280774
Performance counter stats for './super_netperf 200 -t TCP_RR -H lpq84 -l 20 -- -r 4K,4K':
205719.049006 task-clock # 9.278 CPUs utilized
8,449,968 context-switches # 0.041 M/sec
1,935,997 CPU-migrations # 0.009 M/sec
160,541 page-faults # 0.780 K/sec
548,478,722,290 cycles # 2.666 GHz [83.20%]
455,240,670,857 stalled-cycles-frontend # 83.00% frontend cycles idle [83.48%]
272,881,454,275 stalled-cycles-backend # 49.75% backend cycles idle [66.73%]
166,091,460,030 instructions # 0.30 insns per cycle
# 2.74 stalled cycles per insn [83.39%]
29,150,229,399 branches # 141.699 M/sec [83.30%]
1,943,814,026 branch-misses # 6.67% of all branches [83.32%]
22.173517844 seconds time elapsed
lpq83:~# nstat | egrep "IpOutRequests|IpExtOutOctets"
IpOutRequests 16851063 0.0
IpExtOutOctets 23878580777 0.0
After patch :
lpq83:~# nstat >/dev/null;perf stat ./super_netperf 200 -t TCP_RR -H lpq84 -l 20 -- -r 4K,4K
280877
Performance counter stats for './super_netperf 200 -t TCP_RR -H lpq84 -l 20 -- -r 4K,4K':
107496.071918 task-clock # 4.847 CPUs utilized
5,635,458 context-switches # 0.052 M/sec
1,374,707 CPU-migrations # 0.013 M/sec
160,920 page-faults # 0.001 M/sec
281,500,010,924 cycles # 2.619 GHz [83.28%]
228,865,069,307 stalled-cycles-frontend # 81.30% frontend cycles idle [83.38%]
142,462,742,658 stalled-cycles-backend # 50.61% backend cycles idle [66.81%]
95,227,712,566 instructions # 0.34 insns per cycle
# 2.40 stalled cycles per insn [83.43%]
16,209,868,171 branches # 150.795 M/sec [83.20%]
874,252,952 branch-misses # 5.39% of all branches [83.37%]
22.175821286 seconds time elapsed
lpq83:~# nstat | egrep "IpOutRequests|IpExtOutOctets"
IpOutRequests 11239428 0.0
IpExtOutOctets 23595191035 0.0
Indeed, the occupancy of tx skbs (IpExtOutOctets/IpOutRequests) is higher :
2099 instead of 1417, thus helping GRO to be more efficient when using FQ packet
scheduler.
Many thanks to Neal for review and ideas.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Yuchung Cheng <ycheng@google.com>
Cc: Neal Cardwell <ncardwell@google.com>
Cc: Nandita Dukkipati <nanditad@google.com>
Cc: Van Jacobson <vanj@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Tested-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-12-14 04:51:23 +07:00
|
|
|
/* Returns the portion of skb which can be sent right away */
|
|
|
|
static unsigned int tcp_mss_split_point(const struct sock *sk,
|
|
|
|
const struct sk_buff *skb,
|
|
|
|
unsigned int mss_now,
|
|
|
|
unsigned int max_segs,
|
|
|
|
int nonagle)
|
2005-07-06 05:24:38 +07:00
|
|
|
{
|
2011-10-21 16:22:42 +07:00
|
|
|
const struct tcp_sock *tp = tcp_sk(sk);
|
tcp: refine TSO splits
While investigating performance problems on small RPC workloads,
I noticed linux TCP stack was always splitting the last TSO skb
into two parts (skbs). One being a multiple of MSS, and a small one
with the Push flag. This split is done even if TCP_NODELAY is set,
or if no small packet is in flight.
Example with request/response of 4K/4K
IP A > B: . ack 68432 win 2783 <nop,nop,timestamp 6524593 6525001>
IP A > B: . 65537:68433(2896) ack 69632 win 2783 <nop,nop,timestamp 6524593 6525001>
IP A > B: P 68433:69633(1200) ack 69632 win 2783 <nop,nop,timestamp 6524593 6525001>
IP B > A: . ack 68433 win 2768 <nop,nop,timestamp 6525001 6524593>
IP B > A: . 69632:72528(2896) ack 69633 win 2768 <nop,nop,timestamp 6525001 6524593>
IP B > A: P 72528:73728(1200) ack 69633 win 2768 <nop,nop,timestamp 6525001 6524593>
IP A > B: . ack 72528 win 2783 <nop,nop,timestamp 6524593 6525001>
IP A > B: . 69633:72529(2896) ack 73728 win 2783 <nop,nop,timestamp 6524593 6525001>
IP A > B: P 72529:73729(1200) ack 73728 win 2783 <nop,nop,timestamp 6524593 6525001>
We can avoid this split by including the Nagle tests at the right place.
Note : If some NIC had trouble sending TSO packets with a partial
last segment, we would have hit the problem in GRO/forwarding workload already.
tcp_minshall_update() is moved to tcp_output.c and is updated as we might
feed a TSO packet with a partial last segment.
This patch tremendously improves performance, as the traffic now looks
like :
IP A > B: . ack 98304 win 2783 <nop,nop,timestamp 6834277 6834685>
IP A > B: P 94209:98305(4096) ack 98304 win 2783 <nop,nop,timestamp 6834277 6834685>
IP B > A: . ack 98305 win 2768 <nop,nop,timestamp 6834686 6834277>
IP B > A: P 98304:102400(4096) ack 98305 win 2768 <nop,nop,timestamp 6834686 6834277>
IP A > B: . ack 102400 win 2783 <nop,nop,timestamp 6834279 6834686>
IP A > B: P 98305:102401(4096) ack 102400 win 2783 <nop,nop,timestamp 6834279 6834686>
IP B > A: . ack 102401 win 2768 <nop,nop,timestamp 6834687 6834279>
IP B > A: P 102400:106496(4096) ack 102401 win 2768 <nop,nop,timestamp 6834687 6834279>
IP A > B: . ack 106496 win 2783 <nop,nop,timestamp 6834280 6834687>
IP A > B: P 102401:106497(4096) ack 106496 win 2783 <nop,nop,timestamp 6834280 6834687>
IP B > A: . ack 106497 win 2768 <nop,nop,timestamp 6834688 6834280>
IP B > A: P 106496:110592(4096) ack 106497 win 2768 <nop,nop,timestamp 6834688 6834280>
Before :
lpq83:~# nstat >/dev/null;perf stat ./super_netperf 200 -t TCP_RR -H lpq84 -l 20 -- -r 4K,4K
280774
Performance counter stats for './super_netperf 200 -t TCP_RR -H lpq84 -l 20 -- -r 4K,4K':
205719.049006 task-clock # 9.278 CPUs utilized
8,449,968 context-switches # 0.041 M/sec
1,935,997 CPU-migrations # 0.009 M/sec
160,541 page-faults # 0.780 K/sec
548,478,722,290 cycles # 2.666 GHz [83.20%]
455,240,670,857 stalled-cycles-frontend # 83.00% frontend cycles idle [83.48%]
272,881,454,275 stalled-cycles-backend # 49.75% backend cycles idle [66.73%]
166,091,460,030 instructions # 0.30 insns per cycle
# 2.74 stalled cycles per insn [83.39%]
29,150,229,399 branches # 141.699 M/sec [83.30%]
1,943,814,026 branch-misses # 6.67% of all branches [83.32%]
22.173517844 seconds time elapsed
lpq83:~# nstat | egrep "IpOutRequests|IpExtOutOctets"
IpOutRequests 16851063 0.0
IpExtOutOctets 23878580777 0.0
After patch :
lpq83:~# nstat >/dev/null;perf stat ./super_netperf 200 -t TCP_RR -H lpq84 -l 20 -- -r 4K,4K
280877
Performance counter stats for './super_netperf 200 -t TCP_RR -H lpq84 -l 20 -- -r 4K,4K':
107496.071918 task-clock # 4.847 CPUs utilized
5,635,458 context-switches # 0.052 M/sec
1,374,707 CPU-migrations # 0.013 M/sec
160,920 page-faults # 0.001 M/sec
281,500,010,924 cycles # 2.619 GHz [83.28%]
228,865,069,307 stalled-cycles-frontend # 81.30% frontend cycles idle [83.38%]
142,462,742,658 stalled-cycles-backend # 50.61% backend cycles idle [66.81%]
95,227,712,566 instructions # 0.34 insns per cycle
# 2.40 stalled cycles per insn [83.43%]
16,209,868,171 branches # 150.795 M/sec [83.20%]
874,252,952 branch-misses # 5.39% of all branches [83.37%]
22.175821286 seconds time elapsed
lpq83:~# nstat | egrep "IpOutRequests|IpExtOutOctets"
IpOutRequests 11239428 0.0
IpExtOutOctets 23595191035 0.0
Indeed, the occupancy of tx skbs (IpExtOutOctets/IpOutRequests) is higher :
2099 instead of 1417, thus helping GRO to be more efficient when using FQ packet
scheduler.
Many thanks to Neal for review and ideas.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Yuchung Cheng <ycheng@google.com>
Cc: Neal Cardwell <ncardwell@google.com>
Cc: Nandita Dukkipati <nanditad@google.com>
Cc: Van Jacobson <vanj@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Tested-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-12-14 04:51:23 +07:00
|
|
|
u32 partial, needed, window, max_len;
|
2005-07-06 05:24:38 +07:00
|
|
|
|
2007-12-31 19:48:41 +07:00
|
|
|
window = tcp_wnd_end(tp) - TCP_SKB_CB(skb)->seq;
|
2012-07-30 23:11:42 +07:00
|
|
|
max_len = mss_now * max_segs;
|
2007-12-25 12:33:45 +07:00
|
|
|
|
2012-07-30 23:11:42 +07:00
|
|
|
if (likely(max_len <= window && skb != tcp_write_queue_tail(sk)))
|
|
|
|
return max_len;
|
2007-12-25 12:33:45 +07:00
|
|
|
|
2008-03-12 07:55:27 +07:00
|
|
|
needed = min(skb->len, window);
|
|
|
|
|
2012-07-30 23:11:42 +07:00
|
|
|
if (max_len <= needed)
|
|
|
|
return max_len;
|
2007-12-25 12:33:45 +07:00
|
|
|
|
tcp: refine TSO splits
While investigating performance problems on small RPC workloads,
I noticed linux TCP stack was always splitting the last TSO skb
into two parts (skbs). One being a multiple of MSS, and a small one
with the Push flag. This split is done even if TCP_NODELAY is set,
or if no small packet is in flight.
Example with request/response of 4K/4K
IP A > B: . ack 68432 win 2783 <nop,nop,timestamp 6524593 6525001>
IP A > B: . 65537:68433(2896) ack 69632 win 2783 <nop,nop,timestamp 6524593 6525001>
IP A > B: P 68433:69633(1200) ack 69632 win 2783 <nop,nop,timestamp 6524593 6525001>
IP B > A: . ack 68433 win 2768 <nop,nop,timestamp 6525001 6524593>
IP B > A: . 69632:72528(2896) ack 69633 win 2768 <nop,nop,timestamp 6525001 6524593>
IP B > A: P 72528:73728(1200) ack 69633 win 2768 <nop,nop,timestamp 6525001 6524593>
IP A > B: . ack 72528 win 2783 <nop,nop,timestamp 6524593 6525001>
IP A > B: . 69633:72529(2896) ack 73728 win 2783 <nop,nop,timestamp 6524593 6525001>
IP A > B: P 72529:73729(1200) ack 73728 win 2783 <nop,nop,timestamp 6524593 6525001>
We can avoid this split by including the Nagle tests at the right place.
Note : If some NIC had trouble sending TSO packets with a partial
last segment, we would have hit the problem in GRO/forwarding workload already.
tcp_minshall_update() is moved to tcp_output.c and is updated as we might
feed a TSO packet with a partial last segment.
This patch tremendously improves performance, as the traffic now looks
like :
IP A > B: . ack 98304 win 2783 <nop,nop,timestamp 6834277 6834685>
IP A > B: P 94209:98305(4096) ack 98304 win 2783 <nop,nop,timestamp 6834277 6834685>
IP B > A: . ack 98305 win 2768 <nop,nop,timestamp 6834686 6834277>
IP B > A: P 98304:102400(4096) ack 98305 win 2768 <nop,nop,timestamp 6834686 6834277>
IP A > B: . ack 102400 win 2783 <nop,nop,timestamp 6834279 6834686>
IP A > B: P 98305:102401(4096) ack 102400 win 2783 <nop,nop,timestamp 6834279 6834686>
IP B > A: . ack 102401 win 2768 <nop,nop,timestamp 6834687 6834279>
IP B > A: P 102400:106496(4096) ack 102401 win 2768 <nop,nop,timestamp 6834687 6834279>
IP A > B: . ack 106496 win 2783 <nop,nop,timestamp 6834280 6834687>
IP A > B: P 102401:106497(4096) ack 106496 win 2783 <nop,nop,timestamp 6834280 6834687>
IP B > A: . ack 106497 win 2768 <nop,nop,timestamp 6834688 6834280>
IP B > A: P 106496:110592(4096) ack 106497 win 2768 <nop,nop,timestamp 6834688 6834280>
Before :
lpq83:~# nstat >/dev/null;perf stat ./super_netperf 200 -t TCP_RR -H lpq84 -l 20 -- -r 4K,4K
280774
Performance counter stats for './super_netperf 200 -t TCP_RR -H lpq84 -l 20 -- -r 4K,4K':
205719.049006 task-clock # 9.278 CPUs utilized
8,449,968 context-switches # 0.041 M/sec
1,935,997 CPU-migrations # 0.009 M/sec
160,541 page-faults # 0.780 K/sec
548,478,722,290 cycles # 2.666 GHz [83.20%]
455,240,670,857 stalled-cycles-frontend # 83.00% frontend cycles idle [83.48%]
272,881,454,275 stalled-cycles-backend # 49.75% backend cycles idle [66.73%]
166,091,460,030 instructions # 0.30 insns per cycle
# 2.74 stalled cycles per insn [83.39%]
29,150,229,399 branches # 141.699 M/sec [83.30%]
1,943,814,026 branch-misses # 6.67% of all branches [83.32%]
22.173517844 seconds time elapsed
lpq83:~# nstat | egrep "IpOutRequests|IpExtOutOctets"
IpOutRequests 16851063 0.0
IpExtOutOctets 23878580777 0.0
After patch :
lpq83:~# nstat >/dev/null;perf stat ./super_netperf 200 -t TCP_RR -H lpq84 -l 20 -- -r 4K,4K
280877
Performance counter stats for './super_netperf 200 -t TCP_RR -H lpq84 -l 20 -- -r 4K,4K':
107496.071918 task-clock # 4.847 CPUs utilized
5,635,458 context-switches # 0.052 M/sec
1,374,707 CPU-migrations # 0.013 M/sec
160,920 page-faults # 0.001 M/sec
281,500,010,924 cycles # 2.619 GHz [83.28%]
228,865,069,307 stalled-cycles-frontend # 81.30% frontend cycles idle [83.38%]
142,462,742,658 stalled-cycles-backend # 50.61% backend cycles idle [66.81%]
95,227,712,566 instructions # 0.34 insns per cycle
# 2.40 stalled cycles per insn [83.43%]
16,209,868,171 branches # 150.795 M/sec [83.20%]
874,252,952 branch-misses # 5.39% of all branches [83.37%]
22.175821286 seconds time elapsed
lpq83:~# nstat | egrep "IpOutRequests|IpExtOutOctets"
IpOutRequests 11239428 0.0
IpExtOutOctets 23595191035 0.0
Indeed, the occupancy of tx skbs (IpExtOutOctets/IpOutRequests) is higher :
2099 instead of 1417, thus helping GRO to be more efficient when using FQ packet
scheduler.
Many thanks to Neal for review and ideas.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Yuchung Cheng <ycheng@google.com>
Cc: Neal Cardwell <ncardwell@google.com>
Cc: Nandita Dukkipati <nanditad@google.com>
Cc: Van Jacobson <vanj@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Tested-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-12-14 04:51:23 +07:00
|
|
|
partial = needed % mss_now;
|
|
|
|
/* If last segment is not a full MSS, check if Nagle rules allow us
|
|
|
|
* to include this last segment in this skb.
|
|
|
|
* Otherwise, we'll split the skb at last MSS boundary
|
|
|
|
*/
|
2014-03-24 13:49:34 +07:00
|
|
|
if (tcp_nagle_check(partial != 0, tp, nonagle))
|
tcp: refine TSO splits
While investigating performance problems on small RPC workloads,
I noticed linux TCP stack was always splitting the last TSO skb
into two parts (skbs). One being a multiple of MSS, and a small one
with the Push flag. This split is done even if TCP_NODELAY is set,
or if no small packet is in flight.
Example with request/response of 4K/4K
IP A > B: . ack 68432 win 2783 <nop,nop,timestamp 6524593 6525001>
IP A > B: . 65537:68433(2896) ack 69632 win 2783 <nop,nop,timestamp 6524593 6525001>
IP A > B: P 68433:69633(1200) ack 69632 win 2783 <nop,nop,timestamp 6524593 6525001>
IP B > A: . ack 68433 win 2768 <nop,nop,timestamp 6525001 6524593>
IP B > A: . 69632:72528(2896) ack 69633 win 2768 <nop,nop,timestamp 6525001 6524593>
IP B > A: P 72528:73728(1200) ack 69633 win 2768 <nop,nop,timestamp 6525001 6524593>
IP A > B: . ack 72528 win 2783 <nop,nop,timestamp 6524593 6525001>
IP A > B: . 69633:72529(2896) ack 73728 win 2783 <nop,nop,timestamp 6524593 6525001>
IP A > B: P 72529:73729(1200) ack 73728 win 2783 <nop,nop,timestamp 6524593 6525001>
We can avoid this split by including the Nagle tests at the right place.
Note : If some NIC had trouble sending TSO packets with a partial
last segment, we would have hit the problem in GRO/forwarding workload already.
tcp_minshall_update() is moved to tcp_output.c and is updated as we might
feed a TSO packet with a partial last segment.
This patch tremendously improves performance, as the traffic now looks
like :
IP A > B: . ack 98304 win 2783 <nop,nop,timestamp 6834277 6834685>
IP A > B: P 94209:98305(4096) ack 98304 win 2783 <nop,nop,timestamp 6834277 6834685>
IP B > A: . ack 98305 win 2768 <nop,nop,timestamp 6834686 6834277>
IP B > A: P 98304:102400(4096) ack 98305 win 2768 <nop,nop,timestamp 6834686 6834277>
IP A > B: . ack 102400 win 2783 <nop,nop,timestamp 6834279 6834686>
IP A > B: P 98305:102401(4096) ack 102400 win 2783 <nop,nop,timestamp 6834279 6834686>
IP B > A: . ack 102401 win 2768 <nop,nop,timestamp 6834687 6834279>
IP B > A: P 102400:106496(4096) ack 102401 win 2768 <nop,nop,timestamp 6834687 6834279>
IP A > B: . ack 106496 win 2783 <nop,nop,timestamp 6834280 6834687>
IP A > B: P 102401:106497(4096) ack 106496 win 2783 <nop,nop,timestamp 6834280 6834687>
IP B > A: . ack 106497 win 2768 <nop,nop,timestamp 6834688 6834280>
IP B > A: P 106496:110592(4096) ack 106497 win 2768 <nop,nop,timestamp 6834688 6834280>
Before :
lpq83:~# nstat >/dev/null;perf stat ./super_netperf 200 -t TCP_RR -H lpq84 -l 20 -- -r 4K,4K
280774
Performance counter stats for './super_netperf 200 -t TCP_RR -H lpq84 -l 20 -- -r 4K,4K':
205719.049006 task-clock # 9.278 CPUs utilized
8,449,968 context-switches # 0.041 M/sec
1,935,997 CPU-migrations # 0.009 M/sec
160,541 page-faults # 0.780 K/sec
548,478,722,290 cycles # 2.666 GHz [83.20%]
455,240,670,857 stalled-cycles-frontend # 83.00% frontend cycles idle [83.48%]
272,881,454,275 stalled-cycles-backend # 49.75% backend cycles idle [66.73%]
166,091,460,030 instructions # 0.30 insns per cycle
# 2.74 stalled cycles per insn [83.39%]
29,150,229,399 branches # 141.699 M/sec [83.30%]
1,943,814,026 branch-misses # 6.67% of all branches [83.32%]
22.173517844 seconds time elapsed
lpq83:~# nstat | egrep "IpOutRequests|IpExtOutOctets"
IpOutRequests 16851063 0.0
IpExtOutOctets 23878580777 0.0
After patch :
lpq83:~# nstat >/dev/null;perf stat ./super_netperf 200 -t TCP_RR -H lpq84 -l 20 -- -r 4K,4K
280877
Performance counter stats for './super_netperf 200 -t TCP_RR -H lpq84 -l 20 -- -r 4K,4K':
107496.071918 task-clock # 4.847 CPUs utilized
5,635,458 context-switches # 0.052 M/sec
1,374,707 CPU-migrations # 0.013 M/sec
160,920 page-faults # 0.001 M/sec
281,500,010,924 cycles # 2.619 GHz [83.28%]
228,865,069,307 stalled-cycles-frontend # 81.30% frontend cycles idle [83.38%]
142,462,742,658 stalled-cycles-backend # 50.61% backend cycles idle [66.81%]
95,227,712,566 instructions # 0.34 insns per cycle
# 2.40 stalled cycles per insn [83.43%]
16,209,868,171 branches # 150.795 M/sec [83.20%]
874,252,952 branch-misses # 5.39% of all branches [83.37%]
22.175821286 seconds time elapsed
lpq83:~# nstat | egrep "IpOutRequests|IpExtOutOctets"
IpOutRequests 11239428 0.0
IpExtOutOctets 23595191035 0.0
Indeed, the occupancy of tx skbs (IpExtOutOctets/IpOutRequests) is higher :
2099 instead of 1417, thus helping GRO to be more efficient when using FQ packet
scheduler.
Many thanks to Neal for review and ideas.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Yuchung Cheng <ycheng@google.com>
Cc: Neal Cardwell <ncardwell@google.com>
Cc: Nandita Dukkipati <nanditad@google.com>
Cc: Van Jacobson <vanj@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Tested-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-12-14 04:51:23 +07:00
|
|
|
return needed - partial;
|
|
|
|
|
|
|
|
return needed;
|
2005-07-06 05:24:38 +07:00
|
|
|
}
|
|
|
|
|
|
|
|
/* Can at least one segment of SKB be sent right now, according to the
|
|
|
|
* congestion window rules? If so, return how many segments are allowed.
|
|
|
|
*/
|
2011-10-21 16:22:42 +07:00
|
|
|
static inline unsigned int tcp_cwnd_test(const struct tcp_sock *tp,
|
|
|
|
const struct sk_buff *skb)
|
2005-07-06 05:24:38 +07:00
|
|
|
{
|
tcp: limit GSO packets to half cwnd
In DC world, GSO packets initially cooked by tcp_sendmsg() are usually
big, as sk_pacing_rate is high.
When network is congested, cwnd can be smaller than the GSO packets
found in socket write queue. tcp_write_xmit() splits GSO packets
using the available cwnd, and we end up sending a single GSO packet,
consuming all available cwnd.
With GRO aggregation on the receiver, we might handle a single GRO
packet, sending back a single ACK.
1) This single ACK might be lost
TLP or RTO are forced to attempt a retransmit.
2) This ACK releases a full cwnd, sender sends another big GSO packet,
in a ping pong mode.
This behavior does not fill the pipes in the best way, because of
scheduling artifacts.
Make sure we always have at least two GSO packets in flight.
This allows us to safely increase GRO efficiency without risking
spurious retransmits.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2014-11-14 00:45:22 +07:00
|
|
|
u32 in_flight, cwnd, halfcwnd;
|
2005-07-06 05:24:38 +07:00
|
|
|
|
|
|
|
/* Don't be strict about the congestion window for the final FIN. */
|
2011-09-28 00:25:05 +07:00
|
|
|
if ((TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN) &&
|
|
|
|
tcp_skb_pcount(skb) == 1)
|
2005-07-06 05:24:38 +07:00
|
|
|
return 1;
|
|
|
|
|
|
|
|
in_flight = tcp_packets_in_flight(tp);
|
|
|
|
cwnd = tp->snd_cwnd;
|
tcp: limit GSO packets to half cwnd
In DC world, GSO packets initially cooked by tcp_sendmsg() are usually
big, as sk_pacing_rate is high.
When network is congested, cwnd can be smaller than the GSO packets
found in socket write queue. tcp_write_xmit() splits GSO packets
using the available cwnd, and we end up sending a single GSO packet,
consuming all available cwnd.
With GRO aggregation on the receiver, we might handle a single GRO
packet, sending back a single ACK.
1) This single ACK might be lost
TLP or RTO are forced to attempt a retransmit.
2) This ACK releases a full cwnd, sender sends another big GSO packet,
in a ping pong mode.
This behavior does not fill the pipes in the best way, because of
scheduling artifacts.
Make sure we always have at least two GSO packets in flight.
This allows us to safely increase GRO efficiency without risking
spurious retransmits.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2014-11-14 00:45:22 +07:00
|
|
|
if (in_flight >= cwnd)
|
|
|
|
return 0;
|
2005-07-06 05:24:38 +07:00
|
|
|
|
tcp: limit GSO packets to half cwnd
In DC world, GSO packets initially cooked by tcp_sendmsg() are usually
big, as sk_pacing_rate is high.
When network is congested, cwnd can be smaller than the GSO packets
found in socket write queue. tcp_write_xmit() splits GSO packets
using the available cwnd, and we end up sending a single GSO packet,
consuming all available cwnd.
With GRO aggregation on the receiver, we might handle a single GRO
packet, sending back a single ACK.
1) This single ACK might be lost
TLP or RTO are forced to attempt a retransmit.
2) This ACK releases a full cwnd, sender sends another big GSO packet,
in a ping pong mode.
This behavior does not fill the pipes in the best way, because of
scheduling artifacts.
Make sure we always have at least two GSO packets in flight.
This allows us to safely increase GRO efficiency without risking
spurious retransmits.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2014-11-14 00:45:22 +07:00
|
|
|
/* For better scheduling, ensure we have at least
|
|
|
|
* 2 GSO packets in flight.
|
|
|
|
*/
|
|
|
|
halfcwnd = max(cwnd >> 1, 1U);
|
|
|
|
return min(halfcwnd, cwnd - in_flight);
|
2005-07-06 05:24:38 +07:00
|
|
|
}
|
|
|
|
|
tree-wide: fix comment/printk typos
"gadget", "through", "command", "maintain", "maintain", "controller", "address",
"between", "initiali[zs]e", "instead", "function", "select", "already",
"equal", "access", "management", "hierarchy", "registration", "interest",
"relative", "memory", "offset", "already",
Signed-off-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de>
Signed-off-by: Jiri Kosina <jkosina@suse.cz>
2010-11-02 02:38:34 +07:00
|
|
|
/* Initialize TSO state of a skb.
|
2009-07-22 06:00:40 +07:00
|
|
|
* This must be invoked the first time we consider transmitting
|
2005-07-06 05:24:38 +07:00
|
|
|
* SKB onto the wire.
|
|
|
|
*/
|
2015-06-11 23:15:17 +07:00
|
|
|
static int tcp_init_tso_segs(struct sk_buff *skb, unsigned int mss_now)
|
2005-07-06 05:24:38 +07:00
|
|
|
{
|
|
|
|
int tso_segs = tcp_skb_pcount(skb);
|
|
|
|
|
2008-12-04 12:24:48 +07:00
|
|
|
if (!tso_segs || (tso_segs > 1 && tcp_skb_mss(skb) != mss_now)) {
|
2015-06-11 23:15:17 +07:00
|
|
|
tcp_set_skb_tso_segs(skb, mss_now);
|
2005-07-06 05:24:38 +07:00
|
|
|
tso_segs = tcp_skb_pcount(skb);
|
|
|
|
}
|
|
|
|
return tso_segs;
|
|
|
|
}
|
|
|
|
|
|
|
|
|
2012-05-17 06:15:34 +07:00
|
|
|
/* Return true if the Nagle test allows this packet to be
|
2005-07-06 05:24:38 +07:00
|
|
|
* sent now.
|
|
|
|
*/
|
2012-05-17 06:15:34 +07:00
|
|
|
static inline bool tcp_nagle_test(const struct tcp_sock *tp, const struct sk_buff *skb,
|
|
|
|
unsigned int cur_mss, int nonagle)
|
2005-07-06 05:24:38 +07:00
|
|
|
{
|
|
|
|
/* Nagle rule does not apply to frames, which sit in the middle of the
|
|
|
|
* write_queue (they have no chances to get new data).
|
|
|
|
*
|
|
|
|
* This is implemented in the callers, where they modify the 'nonagle'
|
|
|
|
* argument based upon the location of SKB in the send queue.
|
|
|
|
*/
|
|
|
|
if (nonagle & TCP_NAGLE_PUSH)
|
2012-05-17 06:15:34 +07:00
|
|
|
return true;
|
2005-07-06 05:24:38 +07:00
|
|
|
|
2013-03-20 20:32:58 +07:00
|
|
|
/* Don't use the nagle rule for urgent data (or for the final FIN). */
|
|
|
|
if (tcp_urg_mode(tp) || (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN))
|
2012-05-17 06:15:34 +07:00
|
|
|
return true;
|
2005-07-06 05:24:38 +07:00
|
|
|
|
2014-03-24 13:49:34 +07:00
|
|
|
if (!tcp_nagle_check(skb->len < cur_mss, tp, nonagle))
|
2012-05-17 06:15:34 +07:00
|
|
|
return true;
|
2005-07-06 05:24:38 +07:00
|
|
|
|
2012-05-17 06:15:34 +07:00
|
|
|
return false;
|
2005-07-06 05:24:38 +07:00
|
|
|
}
|
|
|
|
|
|
|
|
/* Does at least the first segment of SKB fit into the send window? */
|
2012-05-17 06:15:34 +07:00
|
|
|
static bool tcp_snd_wnd_test(const struct tcp_sock *tp,
|
|
|
|
const struct sk_buff *skb,
|
|
|
|
unsigned int cur_mss)
|
2005-07-06 05:24:38 +07:00
|
|
|
{
|
|
|
|
u32 end_seq = TCP_SKB_CB(skb)->end_seq;
|
|
|
|
|
|
|
|
if (skb->len > cur_mss)
|
|
|
|
end_seq = TCP_SKB_CB(skb)->seq + cur_mss;
|
|
|
|
|
2007-12-31 19:48:41 +07:00
|
|
|
return !after(end_seq, tcp_wnd_end(tp));
|
2005-07-06 05:24:38 +07:00
|
|
|
}
|
|
|
|
|
2007-03-08 03:12:44 +07:00
|
|
|
/* This checks if the data bearing packet SKB (usually tcp_send_head(sk))
|
2005-07-06 05:24:38 +07:00
|
|
|
* should be put on the wire right now. If so, it returns the number of
|
|
|
|
* packets allowed by the congestion window.
|
|
|
|
*/
|
2011-10-21 16:22:42 +07:00
|
|
|
static unsigned int tcp_snd_test(const struct sock *sk, struct sk_buff *skb,
|
2005-07-06 05:24:38 +07:00
|
|
|
unsigned int cur_mss, int nonagle)
|
|
|
|
{
|
2011-10-21 16:22:42 +07:00
|
|
|
const struct tcp_sock *tp = tcp_sk(sk);
|
2005-07-06 05:24:38 +07:00
|
|
|
unsigned int cwnd_quota;
|
|
|
|
|
2015-06-11 23:15:17 +07:00
|
|
|
tcp_init_tso_segs(skb, cur_mss);
|
2005-07-06 05:24:38 +07:00
|
|
|
|
|
|
|
if (!tcp_nagle_test(tp, skb, cur_mss, nonagle))
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
cwnd_quota = tcp_cwnd_test(tp, skb);
|
2008-01-01 05:57:14 +07:00
|
|
|
if (cwnd_quota && !tcp_snd_wnd_test(tp, skb, cur_mss))
|
2005-07-06 05:24:38 +07:00
|
|
|
cwnd_quota = 0;
|
|
|
|
|
|
|
|
return cwnd_quota;
|
|
|
|
}
|
|
|
|
|
2009-07-22 06:00:40 +07:00
|
|
|
/* Test if sending is allowed right now. */
|
2012-05-17 06:15:34 +07:00
|
|
|
bool tcp_may_send_now(struct sock *sk)
|
2005-07-06 05:24:38 +07:00
|
|
|
{
|
2011-10-21 16:22:42 +07:00
|
|
|
const struct tcp_sock *tp = tcp_sk(sk);
|
2007-03-08 03:12:44 +07:00
|
|
|
struct sk_buff *skb = tcp_send_head(sk);
|
2005-07-06 05:24:38 +07:00
|
|
|
|
2010-09-23 03:43:57 +07:00
|
|
|
return skb &&
|
2009-03-14 21:23:05 +07:00
|
|
|
tcp_snd_test(sk, skb, tcp_current_mss(sk),
|
2005-07-06 05:24:38 +07:00
|
|
|
(tcp_skb_is_last(sk, skb) ?
|
2010-09-23 03:43:57 +07:00
|
|
|
tp->nonagle : TCP_NAGLE_PUSH));
|
2005-07-06 05:24:38 +07:00
|
|
|
}
|
|
|
|
|
|
|
|
/* Trim TSO SKB to LEN bytes, put the remaining data into a new packet
|
|
|
|
* which is put after SKB on the list. It is very much like
|
|
|
|
* tcp_fragment() except that it may make several kinds of assumptions
|
|
|
|
* in order to speed up the splitting operation. In particular, we
|
|
|
|
* know that all the data is in scatter-gather pages, and that the
|
|
|
|
* packet has never been sent out before (and thus is not cloned).
|
|
|
|
*/
|
2008-01-01 05:57:14 +07:00
|
|
|
static int tso_fragment(struct sock *sk, struct sk_buff *skb, unsigned int len,
|
2010-06-24 08:00:22 +07:00
|
|
|
unsigned int mss_now, gfp_t gfp)
|
2005-07-06 05:24:38 +07:00
|
|
|
{
|
|
|
|
struct sk_buff *buff;
|
|
|
|
int nlen = skb->len - len;
|
2009-02-28 11:44:42 +07:00
|
|
|
u8 flags;
|
2005-07-06 05:24:38 +07:00
|
|
|
|
|
|
|
/* All of a TSO frame must be composed of paged data. */
|
2005-08-17 10:43:40 +07:00
|
|
|
if (skb->len != skb->data_len)
|
2014-06-06 21:32:37 +07:00
|
|
|
return tcp_fragment(sk, skb, len, mss_now, gfp);
|
2005-07-06 05:24:38 +07:00
|
|
|
|
2015-05-20 03:26:55 +07:00
|
|
|
buff = sk_stream_alloc_skb(sk, 0, gfp, true);
|
2015-04-03 15:17:26 +07:00
|
|
|
if (unlikely(!buff))
|
2005-07-06 05:24:38 +07:00
|
|
|
return -ENOMEM;
|
|
|
|
|
2007-12-31 15:11:19 +07:00
|
|
|
sk->sk_wmem_queued += buff->truesize;
|
|
|
|
sk_mem_charge(sk, buff->truesize);
|
2006-04-20 11:35:00 +07:00
|
|
|
buff->truesize += nlen;
|
2005-07-06 05:24:38 +07:00
|
|
|
skb->truesize -= nlen;
|
|
|
|
|
|
|
|
/* Correct the sequence numbers. */
|
|
|
|
TCP_SKB_CB(buff)->seq = TCP_SKB_CB(skb)->seq + len;
|
|
|
|
TCP_SKB_CB(buff)->end_seq = TCP_SKB_CB(skb)->end_seq;
|
|
|
|
TCP_SKB_CB(skb)->end_seq = TCP_SKB_CB(buff)->seq;
|
|
|
|
|
|
|
|
/* PSH and FIN should only be set in the second packet. */
|
2011-09-28 00:25:05 +07:00
|
|
|
flags = TCP_SKB_CB(skb)->tcp_flags;
|
|
|
|
TCP_SKB_CB(skb)->tcp_flags = flags & ~(TCPHDR_FIN | TCPHDR_PSH);
|
|
|
|
TCP_SKB_CB(buff)->tcp_flags = flags;
|
2005-07-06 05:24:38 +07:00
|
|
|
|
|
|
|
/* This packet was never sent out yet, so no SACK bits. */
|
|
|
|
TCP_SKB_CB(buff)->sacked = 0;
|
|
|
|
|
2006-08-30 06:44:56 +07:00
|
|
|
buff->ip_summed = skb->ip_summed = CHECKSUM_PARTIAL;
|
2005-07-06 05:24:38 +07:00
|
|
|
skb_split(skb, buff, len);
|
2014-08-13 02:08:12 +07:00
|
|
|
tcp_fragment_tstamp(skb, buff);
|
2005-07-06 05:24:38 +07:00
|
|
|
|
|
|
|
/* Fix up tso_factor for both original and new SKB. */
|
2015-06-11 23:15:17 +07:00
|
|
|
tcp_set_skb_tso_segs(skb, mss_now);
|
|
|
|
tcp_set_skb_tso_segs(buff, mss_now);
|
2005-07-06 05:24:38 +07:00
|
|
|
|
|
|
|
/* Link BUFF into the send queue. */
|
2014-09-23 06:29:32 +07:00
|
|
|
__skb_header_release(buff);
|
2007-03-08 03:12:44 +07:00
|
|
|
tcp_insert_write_queue_after(skb, buff, sk);
|
2005-07-06 05:24:38 +07:00
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Try to defer sending, if possible, in order to minimize the amount
|
|
|
|
* of TSO splitting we do. View it as a kind of TSO Nagle test.
|
|
|
|
*
|
|
|
|
* This algorithm is from John Heffner.
|
|
|
|
*/
|
2014-05-22 21:41:08 +07:00
|
|
|
static bool tcp_tso_should_defer(struct sock *sk, struct sk_buff *skb,
|
tcp: refine TSO autosizing
Commit 95bd09eb2750 ("tcp: TSO packets automatic sizing") tried to
control TSO size, but did this at the wrong place (sendmsg() time)
At sendmsg() time, we might have a pessimistic view of flow rate,
and we end up building very small skbs (with 2 MSS per skb).
This is bad because :
- It sends small TSO packets even in Slow Start where rate quickly
increases.
- It tends to make socket write queue very big, increasing tcp_ack()
processing time, but also increasing memory needs, not necessarily
accounted for, as fast clones overhead is currently ignored.
- Lower GRO efficiency and more ACK packets.
Servers with a lot of small lived connections suffer from this.
Lets instead fill skbs as much as possible (64KB of payload), but split
them at xmit time, when we have a precise idea of the flow rate.
skb split is actually quite efficient.
Patch looks bigger than necessary, because TCP Small Queue decision now
has to take place after the eventual split.
As Neal suggested, introduce a new tcp_tso_autosize() helper, so that
tcp_tso_should_defer() can be synchronized on same goal.
Rename tp->xmit_size_goal_segs to tp->gso_segs, as this variable
contains number of mss that we can put in GSO packet, and is not
related to the autosizing goal anymore.
Tested:
40 ms rtt link
nstat >/dev/null
netperf -H remote -l -2000000 -- -s 1000000
nstat | egrep "IpInReceives|IpOutRequests|TcpOutSegs|IpExtOutOctets"
Before patch :
Recv Send Send
Socket Socket Message Elapsed
Size Size Size Time Throughput
bytes bytes bytes secs. 10^6bits/s
87380 2000000 2000000 0.36 44.22
IpInReceives 600 0.0
IpOutRequests 599 0.0
TcpOutSegs 1397 0.0
IpExtOutOctets 2033249 0.0
After patch :
Recv Send Send
Socket Socket Message Elapsed
Size Size Size Time Throughput
bytes bytes bytes secs. 10^6bits/sec
87380 2000000 2000000 0.36 44.27
IpInReceives 221 0.0
IpOutRequests 232 0.0
TcpOutSegs 1397 0.0
IpExtOutOctets 2013953 0.0
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Acked-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2014-12-08 03:22:18 +07:00
|
|
|
bool *is_cwnd_limited, u32 max_segs)
|
2005-07-06 05:24:38 +07:00
|
|
|
{
|
2005-08-10 14:03:31 +07:00
|
|
|
const struct inet_connection_sock *icsk = inet_csk(sk);
|
tcp: tso: restore IW10 after TSO autosizing
With sysctl_tcp_min_tso_segs being 4, it is very possible
that tcp_tso_should_defer() decides not sending last 2 MSS
of initial window of 10 packets. This also applies if
autosizing decides to send X MSS per GSO packet, and cwnd
is not a multiple of X.
This patch implements an heuristic based on age of first
skb in write queue : If it was sent very recently (less than half srtt),
we can predict that no ACK packet will come in less than half rtt,
so deferring might cause an under utilization of our window.
This is visible on initial send (IW10) on web servers,
but more generally on some RPC, as the last part of the message
might need an extra RTT to get delivered.
Tested:
Ran following packetdrill test
// A simple server-side test that sends exactly an initial window (IW10)
// worth of packets.
`sysctl -e -q net.ipv4.tcp_min_tso_segs=4`
0.000 socket(..., SOCK_STREAM, IPPROTO_TCP) = 3
+0 setsockopt(3, SOL_SOCKET, SO_REUSEADDR, [1], 4) = 0
+0 bind(3, ..., ...) = 0
+0 listen(3, 1) = 0
+.1 < S 0:0(0) win 32792 <mss 1460,sackOK,nop,nop,nop,wscale 7>
+0 > S. 0:0(0) ack 1 <mss 1460,nop,nop,sackOK,nop,wscale 6>
+.1 < . 1:1(0) ack 1 win 257
+0 accept(3, ..., ...) = 4
+0 write(4, ..., 14600) = 14600
+0 > . 1:5841(5840) ack 1 win 457
+0 > . 5841:11681(5840) ack 1 win 457
// Following packet should be sent right now.
+0 > P. 11681:14601(2920) ack 1 win 457
+.1 < . 1:1(0) ack 14601 win 257
+0 close(4) = 0
+0 > F. 14601:14601(0) ack 1
+.1 < F. 1:1(0) ack 14602 win 257
+0 > . 14602:14602(0) ack 2
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2015-02-27 05:10:19 +07:00
|
|
|
u32 age, send_win, cong_win, limit, in_flight;
|
|
|
|
struct tcp_sock *tp = tcp_sk(sk);
|
|
|
|
struct skb_mstamp now;
|
|
|
|
struct sk_buff *head;
|
2010-12-07 19:03:55 +07:00
|
|
|
int win_divisor;
|
2005-07-06 05:24:38 +07:00
|
|
|
|
2011-09-28 00:25:05 +07:00
|
|
|
if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN)
|
2006-10-19 10:36:48 +07:00
|
|
|
goto send_now;
|
2005-07-06 05:24:38 +07:00
|
|
|
|
2015-07-26 14:45:24 +07:00
|
|
|
if (icsk->icsk_ca_state >= TCP_CA_Recovery)
|
2006-10-19 10:36:48 +07:00
|
|
|
goto send_now;
|
|
|
|
|
2015-02-27 05:10:18 +07:00
|
|
|
/* Avoid bursty behavior by allowing defer
|
|
|
|
* only if the last write was recent.
|
|
|
|
*/
|
|
|
|
if ((s32)(tcp_time_stamp - tp->lsndtime) > 0)
|
2006-10-19 10:36:48 +07:00
|
|
|
goto send_now;
|
2005-07-06 05:43:58 +07:00
|
|
|
|
2005-07-06 05:24:38 +07:00
|
|
|
in_flight = tcp_packets_in_flight(tp);
|
|
|
|
|
2008-01-01 05:57:14 +07:00
|
|
|
BUG_ON(tcp_skb_pcount(skb) <= 1 || (tp->snd_cwnd <= in_flight));
|
2005-07-06 05:24:38 +07:00
|
|
|
|
2007-12-31 19:48:41 +07:00
|
|
|
send_win = tcp_wnd_end(tp) - TCP_SKB_CB(skb)->seq;
|
2005-07-06 05:24:38 +07:00
|
|
|
|
|
|
|
/* From in_flight test above, we know that cwnd > in_flight. */
|
|
|
|
cong_win = (tp->snd_cwnd - in_flight) * tp->mss_cache;
|
|
|
|
|
|
|
|
limit = min(send_win, cong_win);
|
|
|
|
|
2006-03-12 09:51:49 +07:00
|
|
|
/* If a full-sized TSO skb can be sent, do it. */
|
tcp: refine TSO autosizing
Commit 95bd09eb2750 ("tcp: TSO packets automatic sizing") tried to
control TSO size, but did this at the wrong place (sendmsg() time)
At sendmsg() time, we might have a pessimistic view of flow rate,
and we end up building very small skbs (with 2 MSS per skb).
This is bad because :
- It sends small TSO packets even in Slow Start where rate quickly
increases.
- It tends to make socket write queue very big, increasing tcp_ack()
processing time, but also increasing memory needs, not necessarily
accounted for, as fast clones overhead is currently ignored.
- Lower GRO efficiency and more ACK packets.
Servers with a lot of small lived connections suffer from this.
Lets instead fill skbs as much as possible (64KB of payload), but split
them at xmit time, when we have a precise idea of the flow rate.
skb split is actually quite efficient.
Patch looks bigger than necessary, because TCP Small Queue decision now
has to take place after the eventual split.
As Neal suggested, introduce a new tcp_tso_autosize() helper, so that
tcp_tso_should_defer() can be synchronized on same goal.
Rename tp->xmit_size_goal_segs to tp->gso_segs, as this variable
contains number of mss that we can put in GSO packet, and is not
related to the autosizing goal anymore.
Tested:
40 ms rtt link
nstat >/dev/null
netperf -H remote -l -2000000 -- -s 1000000
nstat | egrep "IpInReceives|IpOutRequests|TcpOutSegs|IpExtOutOctets"
Before patch :
Recv Send Send
Socket Socket Message Elapsed
Size Size Size Time Throughput
bytes bytes bytes secs. 10^6bits/s
87380 2000000 2000000 0.36 44.22
IpInReceives 600 0.0
IpOutRequests 599 0.0
TcpOutSegs 1397 0.0
IpExtOutOctets 2033249 0.0
After patch :
Recv Send Send
Socket Socket Message Elapsed
Size Size Size Time Throughput
bytes bytes bytes secs. 10^6bits/sec
87380 2000000 2000000 0.36 44.27
IpInReceives 221 0.0
IpOutRequests 232 0.0
TcpOutSegs 1397 0.0
IpExtOutOctets 2013953 0.0
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Acked-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2014-12-08 03:22:18 +07:00
|
|
|
if (limit >= max_segs * tp->mss_cache)
|
2006-10-19 10:36:48 +07:00
|
|
|
goto send_now;
|
2006-03-12 09:51:49 +07:00
|
|
|
|
2009-02-28 11:44:29 +07:00
|
|
|
/* Middle in queue won't get any more data, full sendable already? */
|
|
|
|
if ((skb != tcp_write_queue_tail(sk)) && (limit >= skb->len))
|
|
|
|
goto send_now;
|
|
|
|
|
2010-12-07 19:03:55 +07:00
|
|
|
win_divisor = ACCESS_ONCE(sysctl_tcp_tso_win_divisor);
|
|
|
|
if (win_divisor) {
|
2005-07-06 05:24:38 +07:00
|
|
|
u32 chunk = min(tp->snd_wnd, tp->snd_cwnd * tp->mss_cache);
|
|
|
|
|
|
|
|
/* If at least some fraction of a window is available,
|
|
|
|
* just use it.
|
|
|
|
*/
|
2010-12-07 19:03:55 +07:00
|
|
|
chunk /= win_divisor;
|
2005-07-06 05:24:38 +07:00
|
|
|
if (limit >= chunk)
|
2006-10-19 10:36:48 +07:00
|
|
|
goto send_now;
|
2005-07-06 05:24:38 +07:00
|
|
|
} else {
|
|
|
|
/* Different approach, try not to defer past a single
|
|
|
|
* ACK. Receiver should ACK every other full sized
|
|
|
|
* frame, so if we have space for more than 3 frames
|
|
|
|
* then send now.
|
|
|
|
*/
|
2011-11-22 00:15:14 +07:00
|
|
|
if (limit > tcp_max_tso_deferred_mss(tp) * tp->mss_cache)
|
2006-10-19 10:36:48 +07:00
|
|
|
goto send_now;
|
2005-07-06 05:24:38 +07:00
|
|
|
}
|
|
|
|
|
tcp: tso: restore IW10 after TSO autosizing
With sysctl_tcp_min_tso_segs being 4, it is very possible
that tcp_tso_should_defer() decides not sending last 2 MSS
of initial window of 10 packets. This also applies if
autosizing decides to send X MSS per GSO packet, and cwnd
is not a multiple of X.
This patch implements an heuristic based on age of first
skb in write queue : If it was sent very recently (less than half srtt),
we can predict that no ACK packet will come in less than half rtt,
so deferring might cause an under utilization of our window.
This is visible on initial send (IW10) on web servers,
but more generally on some RPC, as the last part of the message
might need an extra RTT to get delivered.
Tested:
Ran following packetdrill test
// A simple server-side test that sends exactly an initial window (IW10)
// worth of packets.
`sysctl -e -q net.ipv4.tcp_min_tso_segs=4`
0.000 socket(..., SOCK_STREAM, IPPROTO_TCP) = 3
+0 setsockopt(3, SOL_SOCKET, SO_REUSEADDR, [1], 4) = 0
+0 bind(3, ..., ...) = 0
+0 listen(3, 1) = 0
+.1 < S 0:0(0) win 32792 <mss 1460,sackOK,nop,nop,nop,wscale 7>
+0 > S. 0:0(0) ack 1 <mss 1460,nop,nop,sackOK,nop,wscale 6>
+.1 < . 1:1(0) ack 1 win 257
+0 accept(3, ..., ...) = 4
+0 write(4, ..., 14600) = 14600
+0 > . 1:5841(5840) ack 1 win 457
+0 > . 5841:11681(5840) ack 1 win 457
// Following packet should be sent right now.
+0 > P. 11681:14601(2920) ack 1 win 457
+.1 < . 1:1(0) ack 14601 win 257
+0 close(4) = 0
+0 > F. 14601:14601(0) ack 1
+.1 < F. 1:1(0) ack 14602 win 257
+0 > . 14602:14602(0) ack 2
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2015-02-27 05:10:19 +07:00
|
|
|
head = tcp_write_queue_head(sk);
|
|
|
|
skb_mstamp_get(&now);
|
|
|
|
age = skb_mstamp_us_delta(&now, &head->skb_mstamp);
|
|
|
|
/* If next ACK is likely to come too late (half srtt), do not defer */
|
|
|
|
if (age < (tp->srtt_us >> 4))
|
|
|
|
goto send_now;
|
|
|
|
|
2015-02-27 05:10:18 +07:00
|
|
|
/* Ok, it looks like it is advisable to defer. */
|
2006-10-19 10:36:48 +07:00
|
|
|
|
2015-09-23 23:49:53 +07:00
|
|
|
if (cong_win < send_win && cong_win <= skb->len)
|
2014-05-22 21:41:08 +07:00
|
|
|
*is_cwnd_limited = true;
|
|
|
|
|
2012-05-17 06:15:34 +07:00
|
|
|
return true;
|
2006-10-19 10:36:48 +07:00
|
|
|
|
|
|
|
send_now:
|
2012-05-17 06:15:34 +07:00
|
|
|
return false;
|
2005-07-06 05:24:38 +07:00
|
|
|
}
|
|
|
|
|
2015-03-06 10:18:24 +07:00
|
|
|
static inline void tcp_mtu_check_reprobe(struct sock *sk)
|
|
|
|
{
|
|
|
|
struct inet_connection_sock *icsk = inet_csk(sk);
|
|
|
|
struct tcp_sock *tp = tcp_sk(sk);
|
|
|
|
struct net *net = sock_net(sk);
|
|
|
|
u32 interval;
|
|
|
|
s32 delta;
|
|
|
|
|
|
|
|
interval = net->ipv4.sysctl_tcp_probe_interval;
|
|
|
|
delta = tcp_time_stamp - icsk->icsk_mtup.probe_timestamp;
|
|
|
|
if (unlikely(delta >= interval * HZ)) {
|
|
|
|
int mss = tcp_current_mss(sk);
|
|
|
|
|
|
|
|
/* Update current search range */
|
|
|
|
icsk->icsk_mtup.probe_size = 0;
|
|
|
|
icsk->icsk_mtup.search_high = tp->rx_opt.mss_clamp +
|
|
|
|
sizeof(struct tcphdr) +
|
|
|
|
icsk->icsk_af_ops->net_header_len;
|
|
|
|
icsk->icsk_mtup.search_low = tcp_mss_to_mtu(sk, mss);
|
|
|
|
|
|
|
|
/* Update probe time stamp */
|
|
|
|
icsk->icsk_mtup.probe_timestamp = tcp_time_stamp;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2006-03-21 08:53:41 +07:00
|
|
|
/* Create a new MTU probe if we are ready.
|
2009-07-22 06:00:40 +07:00
|
|
|
* MTU probe is regularly attempting to increase the path MTU by
|
|
|
|
* deliberately sending larger packets. This discovers routing
|
|
|
|
* changes resulting in larger path MTUs.
|
|
|
|
*
|
2006-03-21 08:53:41 +07:00
|
|
|
* Returns 0 if we should wait to probe (no cwnd available),
|
|
|
|
* 1 if a probe was sent,
|
2008-01-01 05:57:14 +07:00
|
|
|
* -1 otherwise
|
|
|
|
*/
|
2006-03-21 08:53:41 +07:00
|
|
|
static int tcp_mtu_probe(struct sock *sk)
|
|
|
|
{
|
|
|
|
struct tcp_sock *tp = tcp_sk(sk);
|
|
|
|
struct inet_connection_sock *icsk = inet_csk(sk);
|
|
|
|
struct sk_buff *skb, *nskb, *next;
|
2015-03-06 10:18:23 +07:00
|
|
|
struct net *net = sock_net(sk);
|
2006-03-21 08:53:41 +07:00
|
|
|
int len;
|
|
|
|
int probe_size;
|
2007-11-23 18:08:16 +07:00
|
|
|
int size_needed;
|
2006-03-21 08:53:41 +07:00
|
|
|
int copy;
|
|
|
|
int mss_now;
|
2015-03-06 10:18:23 +07:00
|
|
|
int interval;
|
2006-03-21 08:53:41 +07:00
|
|
|
|
|
|
|
/* Not currently probing/verifying,
|
|
|
|
* not in recovery,
|
|
|
|
* have enough cwnd, and
|
|
|
|
* not SACKing (the variable headers throw things off) */
|
|
|
|
if (!icsk->icsk_mtup.enabled ||
|
|
|
|
icsk->icsk_mtup.probe_size ||
|
|
|
|
inet_csk(sk)->icsk_ca_state != TCP_CA_Open ||
|
|
|
|
tp->snd_cwnd < 11 ||
|
2009-02-28 11:44:38 +07:00
|
|
|
tp->rx_opt.num_sacks || tp->rx_opt.dsack)
|
2006-03-21 08:53:41 +07:00
|
|
|
return -1;
|
|
|
|
|
2015-03-06 10:18:23 +07:00
|
|
|
/* Use binary search for probe_size between tcp_mss_base,
|
|
|
|
* and current mss_clamp. if (search_high - search_low)
|
|
|
|
* smaller than a threshold, backoff from probing.
|
|
|
|
*/
|
2009-03-14 21:23:05 +07:00
|
|
|
mss_now = tcp_current_mss(sk);
|
2015-03-06 10:18:23 +07:00
|
|
|
probe_size = tcp_mtu_to_mss(sk, (icsk->icsk_mtup.search_high +
|
|
|
|
icsk->icsk_mtup.search_low) >> 1);
|
2007-11-23 18:08:16 +07:00
|
|
|
size_needed = probe_size + (tp->reordering + 1) * tp->mss_cache;
|
2015-03-06 10:18:23 +07:00
|
|
|
interval = icsk->icsk_mtup.search_high - icsk->icsk_mtup.search_low;
|
2015-03-06 10:18:24 +07:00
|
|
|
/* When misfortune happens, we are reprobing actively,
|
|
|
|
* and then reprobe timer has expired. We stick with current
|
|
|
|
* probing process by not resetting search range to its orignal.
|
|
|
|
*/
|
2015-03-06 10:18:23 +07:00
|
|
|
if (probe_size > tcp_mtu_to_mss(sk, icsk->icsk_mtup.search_high) ||
|
2015-03-06 10:18:24 +07:00
|
|
|
interval < net->ipv4.sysctl_tcp_probe_threshold) {
|
|
|
|
/* Check whether enough time has elaplased for
|
|
|
|
* another round of probing.
|
|
|
|
*/
|
|
|
|
tcp_mtu_check_reprobe(sk);
|
2006-03-21 08:53:41 +07:00
|
|
|
return -1;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Have enough data in the send queue to probe? */
|
2007-11-23 18:10:56 +07:00
|
|
|
if (tp->write_seq - tp->snd_nxt < size_needed)
|
2006-03-21 08:53:41 +07:00
|
|
|
return -1;
|
|
|
|
|
2007-11-23 18:08:16 +07:00
|
|
|
if (tp->snd_wnd < size_needed)
|
|
|
|
return -1;
|
2007-12-31 19:48:41 +07:00
|
|
|
if (after(tp->snd_nxt + size_needed, tcp_wnd_end(tp)))
|
2007-11-23 18:08:16 +07:00
|
|
|
return 0;
|
2006-03-21 08:53:41 +07:00
|
|
|
|
2007-12-02 05:48:01 +07:00
|
|
|
/* Do we need to wait to drain cwnd? With none in flight, don't stall */
|
|
|
|
if (tcp_packets_in_flight(tp) + 2 > tp->snd_cwnd) {
|
|
|
|
if (!tcp_packets_in_flight(tp))
|
2006-03-21 08:53:41 +07:00
|
|
|
return -1;
|
|
|
|
else
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* We're allowed to probe. Build it now. */
|
2015-05-20 03:26:55 +07:00
|
|
|
nskb = sk_stream_alloc_skb(sk, probe_size, GFP_ATOMIC, false);
|
2015-04-03 15:17:26 +07:00
|
|
|
if (!nskb)
|
2006-03-21 08:53:41 +07:00
|
|
|
return -1;
|
2007-12-31 15:11:19 +07:00
|
|
|
sk->sk_wmem_queued += nskb->truesize;
|
|
|
|
sk_mem_charge(sk, nskb->truesize);
|
2006-03-21 08:53:41 +07:00
|
|
|
|
2007-03-08 03:12:44 +07:00
|
|
|
skb = tcp_send_head(sk);
|
2006-03-21 08:53:41 +07:00
|
|
|
|
|
|
|
TCP_SKB_CB(nskb)->seq = TCP_SKB_CB(skb)->seq;
|
|
|
|
TCP_SKB_CB(nskb)->end_seq = TCP_SKB_CB(skb)->seq + probe_size;
|
2011-09-28 00:25:05 +07:00
|
|
|
TCP_SKB_CB(nskb)->tcp_flags = TCPHDR_ACK;
|
2006-03-21 08:53:41 +07:00
|
|
|
TCP_SKB_CB(nskb)->sacked = 0;
|
|
|
|
nskb->csum = 0;
|
2006-08-30 06:44:56 +07:00
|
|
|
nskb->ip_summed = skb->ip_summed;
|
2006-03-21 08:53:41 +07:00
|
|
|
|
2007-12-02 05:48:00 +07:00
|
|
|
tcp_insert_write_queue_before(nskb, skb, sk);
|
|
|
|
|
2006-03-21 08:53:41 +07:00
|
|
|
len = 0;
|
2007-12-02 05:48:02 +07:00
|
|
|
tcp_for_write_queue_from_safe(skb, next, sk) {
|
2006-03-21 08:53:41 +07:00
|
|
|
copy = min_t(int, skb->len, probe_size - len);
|
|
|
|
if (nskb->ip_summed)
|
|
|
|
skb_copy_bits(skb, 0, skb_put(nskb, copy), copy);
|
|
|
|
else
|
|
|
|
nskb->csum = skb_copy_and_csum_bits(skb, 0,
|
2008-01-01 05:57:14 +07:00
|
|
|
skb_put(nskb, copy),
|
|
|
|
copy, nskb->csum);
|
2006-03-21 08:53:41 +07:00
|
|
|
|
|
|
|
if (skb->len <= copy) {
|
|
|
|
/* We've eaten all the data from this skb.
|
|
|
|
* Throw it away. */
|
2011-09-28 00:25:05 +07:00
|
|
|
TCP_SKB_CB(nskb)->tcp_flags |= TCP_SKB_CB(skb)->tcp_flags;
|
2007-03-08 03:12:44 +07:00
|
|
|
tcp_unlink_write_queue(skb, sk);
|
2007-12-31 15:11:19 +07:00
|
|
|
sk_wmem_free_skb(sk, skb);
|
2006-03-21 08:53:41 +07:00
|
|
|
} else {
|
2011-09-28 00:25:05 +07:00
|
|
|
TCP_SKB_CB(nskb)->tcp_flags |= TCP_SKB_CB(skb)->tcp_flags &
|
2010-06-12 21:01:43 +07:00
|
|
|
~(TCPHDR_FIN|TCPHDR_PSH);
|
2006-03-21 08:53:41 +07:00
|
|
|
if (!skb_shinfo(skb)->nr_frags) {
|
|
|
|
skb_pull(skb, copy);
|
2006-08-30 06:44:56 +07:00
|
|
|
if (skb->ip_summed != CHECKSUM_PARTIAL)
|
2008-01-01 05:57:14 +07:00
|
|
|
skb->csum = csum_partial(skb->data,
|
|
|
|
skb->len, 0);
|
2006-03-21 08:53:41 +07:00
|
|
|
} else {
|
|
|
|
__pskb_trim_head(skb, copy);
|
2015-06-11 23:15:17 +07:00
|
|
|
tcp_set_skb_tso_segs(skb, mss_now);
|
2006-03-21 08:53:41 +07:00
|
|
|
}
|
|
|
|
TCP_SKB_CB(skb)->seq += copy;
|
|
|
|
}
|
|
|
|
|
|
|
|
len += copy;
|
2007-12-02 05:48:02 +07:00
|
|
|
|
|
|
|
if (len >= probe_size)
|
|
|
|
break;
|
2006-03-21 08:53:41 +07:00
|
|
|
}
|
2015-06-11 23:15:17 +07:00
|
|
|
tcp_init_tso_segs(nskb, nskb->len);
|
2006-03-21 08:53:41 +07:00
|
|
|
|
|
|
|
/* We're ready to send. If this fails, the probe will
|
2014-09-06 05:33:33 +07:00
|
|
|
* be resegmented into mss-sized pieces by tcp_write_xmit().
|
|
|
|
*/
|
2006-03-21 08:53:41 +07:00
|
|
|
if (!tcp_transmit_skb(sk, nskb, 1, GFP_ATOMIC)) {
|
|
|
|
/* Decrement cwnd here because we are sending
|
2008-01-01 05:57:14 +07:00
|
|
|
* effectively two packets. */
|
2006-03-21 08:53:41 +07:00
|
|
|
tp->snd_cwnd--;
|
2007-12-31 19:43:57 +07:00
|
|
|
tcp_event_new_data_sent(sk, nskb);
|
2006-03-21 08:53:41 +07:00
|
|
|
|
|
|
|
icsk->icsk_mtup.probe_size = tcp_mss_to_mtu(sk, nskb->len);
|
2006-03-21 12:32:58 +07:00
|
|
|
tp->mtu_probe.probe_seq_start = TCP_SKB_CB(nskb)->seq;
|
|
|
|
tp->mtu_probe.probe_seq_end = TCP_SKB_CB(nskb)->end_seq;
|
2006-03-21 08:53:41 +07:00
|
|
|
|
|
|
|
return 1;
|
|
|
|
}
|
|
|
|
|
|
|
|
return -1;
|
|
|
|
}
|
|
|
|
|
2005-04-17 05:20:36 +07:00
|
|
|
/* This routine writes packets to the network. It advances the
|
|
|
|
* send_head. This happens as incoming acks open up the remote
|
|
|
|
* window for us.
|
|
|
|
*
|
2008-12-04 12:24:48 +07:00
|
|
|
* LARGESEND note: !tcp_urg_mode is overkill, only frames between
|
|
|
|
* snd_up-64k-mss .. snd_up cannot be large. However, taking into
|
|
|
|
* account rare use of URG, this is not a big flaw.
|
|
|
|
*
|
tcp: Tail loss probe (TLP)
This patch series implement the Tail loss probe (TLP) algorithm described
in http://tools.ietf.org/html/draft-dukkipati-tcpm-tcp-loss-probe-01. The
first patch implements the basic algorithm.
TLP's goal is to reduce tail latency of short transactions. It achieves
this by converting retransmission timeouts (RTOs) occuring due
to tail losses (losses at end of transactions) into fast recovery.
TLP transmits one packet in two round-trips when a connection is in
Open state and isn't receiving any ACKs. The transmitted packet, aka
loss probe, can be either new or a retransmission. When there is tail
loss, the ACK from a loss probe triggers FACK/early-retransmit based
fast recovery, thus avoiding a costly RTO. In the absence of loss,
there is no change in the connection state.
PTO stands for probe timeout. It is a timer event indicating
that an ACK is overdue and triggers a loss probe packet. The PTO value
is set to max(2*SRTT, 10ms) and is adjusted to account for delayed
ACK timer when there is only one oustanding packet.
TLP Algorithm
On transmission of new data in Open state:
-> packets_out > 1: schedule PTO in max(2*SRTT, 10ms).
-> packets_out == 1: schedule PTO in max(2*RTT, 1.5*RTT + 200ms)
-> PTO = min(PTO, RTO)
Conditions for scheduling PTO:
-> Connection is in Open state.
-> Connection is either cwnd limited or no new data to send.
-> Number of probes per tail loss episode is limited to one.
-> Connection is SACK enabled.
When PTO fires:
new_segment_exists:
-> transmit new segment.
-> packets_out++. cwnd remains same.
no_new_packet:
-> retransmit the last segment.
Its ACK triggers FACK or early retransmit based recovery.
ACK path:
-> rearm RTO at start of ACK processing.
-> reschedule PTO if need be.
In addition, the patch includes a small variation to the Early Retransmit
(ER) algorithm, such that ER and TLP together can in principle recover any
N-degree of tail loss through fast recovery. TLP is controlled by the same
sysctl as ER, tcp_early_retrans sysctl.
tcp_early_retrans==0; disables TLP and ER.
==1; enables RFC5827 ER.
==2; delayed ER.
==3; TLP and delayed ER. [DEFAULT]
==4; TLP only.
The TLP patch series have been extensively tested on Google Web servers.
It is most effective for short Web trasactions, where it reduced RTOs by 15%
and improved HTTP response time (average by 6%, 99th percentile by 10%).
The transmitted probes account for <0.5% of the overall transmissions.
Signed-off-by: Nandita Dukkipati <nanditad@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Acked-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-03-11 17:00:43 +07:00
|
|
|
* Send at most one packet when push_one > 0. Temporarily ignore
|
|
|
|
* cwnd limit to force at most one packet out when push_one == 2.
|
|
|
|
|
2012-05-17 06:15:34 +07:00
|
|
|
* Returns true, if no segments are in flight and we have queued segments,
|
|
|
|
* but cannot send anything now because of SWS or another problem.
|
2005-04-17 05:20:36 +07:00
|
|
|
*/
|
2012-05-17 06:15:34 +07:00
|
|
|
static bool tcp_write_xmit(struct sock *sk, unsigned int mss_now, int nonagle,
|
|
|
|
int push_one, gfp_t gfp)
|
2005-04-17 05:20:36 +07:00
|
|
|
{
|
|
|
|
struct tcp_sock *tp = tcp_sk(sk);
|
2005-07-06 05:19:06 +07:00
|
|
|
struct sk_buff *skb;
|
2005-07-06 05:24:38 +07:00
|
|
|
unsigned int tso_segs, sent_pkts;
|
|
|
|
int cwnd_quota;
|
2006-03-21 08:53:41 +07:00
|
|
|
int result;
|
2014-05-22 21:41:08 +07:00
|
|
|
bool is_cwnd_limited = false;
|
tcp: refine TSO autosizing
Commit 95bd09eb2750 ("tcp: TSO packets automatic sizing") tried to
control TSO size, but did this at the wrong place (sendmsg() time)
At sendmsg() time, we might have a pessimistic view of flow rate,
and we end up building very small skbs (with 2 MSS per skb).
This is bad because :
- It sends small TSO packets even in Slow Start where rate quickly
increases.
- It tends to make socket write queue very big, increasing tcp_ack()
processing time, but also increasing memory needs, not necessarily
accounted for, as fast clones overhead is currently ignored.
- Lower GRO efficiency and more ACK packets.
Servers with a lot of small lived connections suffer from this.
Lets instead fill skbs as much as possible (64KB of payload), but split
them at xmit time, when we have a precise idea of the flow rate.
skb split is actually quite efficient.
Patch looks bigger than necessary, because TCP Small Queue decision now
has to take place after the eventual split.
As Neal suggested, introduce a new tcp_tso_autosize() helper, so that
tcp_tso_should_defer() can be synchronized on same goal.
Rename tp->xmit_size_goal_segs to tp->gso_segs, as this variable
contains number of mss that we can put in GSO packet, and is not
related to the autosizing goal anymore.
Tested:
40 ms rtt link
nstat >/dev/null
netperf -H remote -l -2000000 -- -s 1000000
nstat | egrep "IpInReceives|IpOutRequests|TcpOutSegs|IpExtOutOctets"
Before patch :
Recv Send Send
Socket Socket Message Elapsed
Size Size Size Time Throughput
bytes bytes bytes secs. 10^6bits/s
87380 2000000 2000000 0.36 44.22
IpInReceives 600 0.0
IpOutRequests 599 0.0
TcpOutSegs 1397 0.0
IpExtOutOctets 2033249 0.0
After patch :
Recv Send Send
Socket Socket Message Elapsed
Size Size Size Time Throughput
bytes bytes bytes secs. 10^6bits/sec
87380 2000000 2000000 0.36 44.27
IpInReceives 221 0.0
IpOutRequests 232 0.0
TcpOutSegs 1397 0.0
IpExtOutOctets 2013953 0.0
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Acked-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2014-12-08 03:22:18 +07:00
|
|
|
u32 max_segs;
|
2005-04-17 05:20:36 +07:00
|
|
|
|
2005-07-06 05:19:06 +07:00
|
|
|
sent_pkts = 0;
|
2006-03-21 08:53:41 +07:00
|
|
|
|
2008-12-06 13:48:55 +07:00
|
|
|
if (!push_one) {
|
|
|
|
/* Do MTU probing. */
|
|
|
|
result = tcp_mtu_probe(sk);
|
|
|
|
if (!result) {
|
2012-05-17 06:15:34 +07:00
|
|
|
return false;
|
2008-12-06 13:48:55 +07:00
|
|
|
} else if (result > 0) {
|
|
|
|
sent_pkts = 1;
|
|
|
|
}
|
2006-03-21 08:53:41 +07:00
|
|
|
}
|
|
|
|
|
tcp: refine TSO autosizing
Commit 95bd09eb2750 ("tcp: TSO packets automatic sizing") tried to
control TSO size, but did this at the wrong place (sendmsg() time)
At sendmsg() time, we might have a pessimistic view of flow rate,
and we end up building very small skbs (with 2 MSS per skb).
This is bad because :
- It sends small TSO packets even in Slow Start where rate quickly
increases.
- It tends to make socket write queue very big, increasing tcp_ack()
processing time, but also increasing memory needs, not necessarily
accounted for, as fast clones overhead is currently ignored.
- Lower GRO efficiency and more ACK packets.
Servers with a lot of small lived connections suffer from this.
Lets instead fill skbs as much as possible (64KB of payload), but split
them at xmit time, when we have a precise idea of the flow rate.
skb split is actually quite efficient.
Patch looks bigger than necessary, because TCP Small Queue decision now
has to take place after the eventual split.
As Neal suggested, introduce a new tcp_tso_autosize() helper, so that
tcp_tso_should_defer() can be synchronized on same goal.
Rename tp->xmit_size_goal_segs to tp->gso_segs, as this variable
contains number of mss that we can put in GSO packet, and is not
related to the autosizing goal anymore.
Tested:
40 ms rtt link
nstat >/dev/null
netperf -H remote -l -2000000 -- -s 1000000
nstat | egrep "IpInReceives|IpOutRequests|TcpOutSegs|IpExtOutOctets"
Before patch :
Recv Send Send
Socket Socket Message Elapsed
Size Size Size Time Throughput
bytes bytes bytes secs. 10^6bits/s
87380 2000000 2000000 0.36 44.22
IpInReceives 600 0.0
IpOutRequests 599 0.0
TcpOutSegs 1397 0.0
IpExtOutOctets 2033249 0.0
After patch :
Recv Send Send
Socket Socket Message Elapsed
Size Size Size Time Throughput
bytes bytes bytes secs. 10^6bits/sec
87380 2000000 2000000 0.36 44.27
IpInReceives 221 0.0
IpOutRequests 232 0.0
TcpOutSegs 1397 0.0
IpExtOutOctets 2013953 0.0
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Acked-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2014-12-08 03:22:18 +07:00
|
|
|
max_segs = tcp_tso_autosize(sk, mss_now);
|
2007-03-08 03:12:44 +07:00
|
|
|
while ((skb = tcp_send_head(sk))) {
|
2005-08-17 10:43:40 +07:00
|
|
|
unsigned int limit;
|
|
|
|
|
2015-06-11 23:15:17 +07:00
|
|
|
tso_segs = tcp_init_tso_segs(skb, mss_now);
|
2005-07-06 05:24:38 +07:00
|
|
|
BUG_ON(!tso_segs);
|
2005-07-06 05:20:09 +07:00
|
|
|
|
2014-08-13 19:03:10 +07:00
|
|
|
if (unlikely(tp->repair) && tp->repair_queue == TCP_SEND_QUEUE) {
|
2014-09-06 05:33:33 +07:00
|
|
|
/* "skb_mstamp" is used as a start point for the retransmit timer */
|
|
|
|
skb_mstamp_get(&skb->skb_mstamp);
|
2012-11-15 11:03:17 +07:00
|
|
|
goto repair; /* Skip network transmission */
|
2014-08-13 19:03:10 +07:00
|
|
|
}
|
2012-11-15 11:03:17 +07:00
|
|
|
|
2005-08-05 09:52:02 +07:00
|
|
|
cwnd_quota = tcp_cwnd_test(tp, skb);
|
tcp: Tail loss probe (TLP)
This patch series implement the Tail loss probe (TLP) algorithm described
in http://tools.ietf.org/html/draft-dukkipati-tcpm-tcp-loss-probe-01. The
first patch implements the basic algorithm.
TLP's goal is to reduce tail latency of short transactions. It achieves
this by converting retransmission timeouts (RTOs) occuring due
to tail losses (losses at end of transactions) into fast recovery.
TLP transmits one packet in two round-trips when a connection is in
Open state and isn't receiving any ACKs. The transmitted packet, aka
loss probe, can be either new or a retransmission. When there is tail
loss, the ACK from a loss probe triggers FACK/early-retransmit based
fast recovery, thus avoiding a costly RTO. In the absence of loss,
there is no change in the connection state.
PTO stands for probe timeout. It is a timer event indicating
that an ACK is overdue and triggers a loss probe packet. The PTO value
is set to max(2*SRTT, 10ms) and is adjusted to account for delayed
ACK timer when there is only one oustanding packet.
TLP Algorithm
On transmission of new data in Open state:
-> packets_out > 1: schedule PTO in max(2*SRTT, 10ms).
-> packets_out == 1: schedule PTO in max(2*RTT, 1.5*RTT + 200ms)
-> PTO = min(PTO, RTO)
Conditions for scheduling PTO:
-> Connection is in Open state.
-> Connection is either cwnd limited or no new data to send.
-> Number of probes per tail loss episode is limited to one.
-> Connection is SACK enabled.
When PTO fires:
new_segment_exists:
-> transmit new segment.
-> packets_out++. cwnd remains same.
no_new_packet:
-> retransmit the last segment.
Its ACK triggers FACK or early retransmit based recovery.
ACK path:
-> rearm RTO at start of ACK processing.
-> reschedule PTO if need be.
In addition, the patch includes a small variation to the Early Retransmit
(ER) algorithm, such that ER and TLP together can in principle recover any
N-degree of tail loss through fast recovery. TLP is controlled by the same
sysctl as ER, tcp_early_retrans sysctl.
tcp_early_retrans==0; disables TLP and ER.
==1; enables RFC5827 ER.
==2; delayed ER.
==3; TLP and delayed ER. [DEFAULT]
==4; TLP only.
The TLP patch series have been extensively tested on Google Web servers.
It is most effective for short Web trasactions, where it reduced RTOs by 15%
and improved HTTP response time (average by 6%, 99th percentile by 10%).
The transmitted probes account for <0.5% of the overall transmissions.
Signed-off-by: Nandita Dukkipati <nanditad@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Acked-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-03-11 17:00:43 +07:00
|
|
|
if (!cwnd_quota) {
|
|
|
|
if (push_one == 2)
|
|
|
|
/* Force out a loss probe pkt. */
|
|
|
|
cwnd_quota = 1;
|
|
|
|
else
|
|
|
|
break;
|
|
|
|
}
|
2005-08-05 09:52:02 +07:00
|
|
|
|
|
|
|
if (unlikely(!tcp_snd_wnd_test(tp, skb, mss_now)))
|
|
|
|
break;
|
|
|
|
|
2015-05-26 22:55:28 +07:00
|
|
|
if (tso_segs == 1) {
|
2005-07-06 05:24:38 +07:00
|
|
|
if (unlikely(!tcp_nagle_test(tp, skb, mss_now,
|
|
|
|
(tcp_skb_is_last(sk, skb) ?
|
|
|
|
nonagle : TCP_NAGLE_PUSH))))
|
|
|
|
break;
|
|
|
|
} else {
|
2014-05-22 21:41:08 +07:00
|
|
|
if (!push_one &&
|
tcp: refine TSO autosizing
Commit 95bd09eb2750 ("tcp: TSO packets automatic sizing") tried to
control TSO size, but did this at the wrong place (sendmsg() time)
At sendmsg() time, we might have a pessimistic view of flow rate,
and we end up building very small skbs (with 2 MSS per skb).
This is bad because :
- It sends small TSO packets even in Slow Start where rate quickly
increases.
- It tends to make socket write queue very big, increasing tcp_ack()
processing time, but also increasing memory needs, not necessarily
accounted for, as fast clones overhead is currently ignored.
- Lower GRO efficiency and more ACK packets.
Servers with a lot of small lived connections suffer from this.
Lets instead fill skbs as much as possible (64KB of payload), but split
them at xmit time, when we have a precise idea of the flow rate.
skb split is actually quite efficient.
Patch looks bigger than necessary, because TCP Small Queue decision now
has to take place after the eventual split.
As Neal suggested, introduce a new tcp_tso_autosize() helper, so that
tcp_tso_should_defer() can be synchronized on same goal.
Rename tp->xmit_size_goal_segs to tp->gso_segs, as this variable
contains number of mss that we can put in GSO packet, and is not
related to the autosizing goal anymore.
Tested:
40 ms rtt link
nstat >/dev/null
netperf -H remote -l -2000000 -- -s 1000000
nstat | egrep "IpInReceives|IpOutRequests|TcpOutSegs|IpExtOutOctets"
Before patch :
Recv Send Send
Socket Socket Message Elapsed
Size Size Size Time Throughput
bytes bytes bytes secs. 10^6bits/s
87380 2000000 2000000 0.36 44.22
IpInReceives 600 0.0
IpOutRequests 599 0.0
TcpOutSegs 1397 0.0
IpExtOutOctets 2033249 0.0
After patch :
Recv Send Send
Socket Socket Message Elapsed
Size Size Size Time Throughput
bytes bytes bytes secs. 10^6bits/sec
87380 2000000 2000000 0.36 44.27
IpInReceives 221 0.0
IpOutRequests 232 0.0
TcpOutSegs 1397 0.0
IpExtOutOctets 2013953 0.0
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Acked-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2014-12-08 03:22:18 +07:00
|
|
|
tcp_tso_should_defer(sk, skb, &is_cwnd_limited,
|
|
|
|
max_segs))
|
2005-07-06 05:24:38 +07:00
|
|
|
break;
|
|
|
|
}
|
2005-07-06 05:20:09 +07:00
|
|
|
|
tcp: refine TSO autosizing
Commit 95bd09eb2750 ("tcp: TSO packets automatic sizing") tried to
control TSO size, but did this at the wrong place (sendmsg() time)
At sendmsg() time, we might have a pessimistic view of flow rate,
and we end up building very small skbs (with 2 MSS per skb).
This is bad because :
- It sends small TSO packets even in Slow Start where rate quickly
increases.
- It tends to make socket write queue very big, increasing tcp_ack()
processing time, but also increasing memory needs, not necessarily
accounted for, as fast clones overhead is currently ignored.
- Lower GRO efficiency and more ACK packets.
Servers with a lot of small lived connections suffer from this.
Lets instead fill skbs as much as possible (64KB of payload), but split
them at xmit time, when we have a precise idea of the flow rate.
skb split is actually quite efficient.
Patch looks bigger than necessary, because TCP Small Queue decision now
has to take place after the eventual split.
As Neal suggested, introduce a new tcp_tso_autosize() helper, so that
tcp_tso_should_defer() can be synchronized on same goal.
Rename tp->xmit_size_goal_segs to tp->gso_segs, as this variable
contains number of mss that we can put in GSO packet, and is not
related to the autosizing goal anymore.
Tested:
40 ms rtt link
nstat >/dev/null
netperf -H remote -l -2000000 -- -s 1000000
nstat | egrep "IpInReceives|IpOutRequests|TcpOutSegs|IpExtOutOctets"
Before patch :
Recv Send Send
Socket Socket Message Elapsed
Size Size Size Time Throughput
bytes bytes bytes secs. 10^6bits/s
87380 2000000 2000000 0.36 44.22
IpInReceives 600 0.0
IpOutRequests 599 0.0
TcpOutSegs 1397 0.0
IpExtOutOctets 2033249 0.0
After patch :
Recv Send Send
Socket Socket Message Elapsed
Size Size Size Time Throughput
bytes bytes bytes secs. 10^6bits/sec
87380 2000000 2000000 0.36 44.27
IpInReceives 221 0.0
IpOutRequests 232 0.0
TcpOutSegs 1397 0.0
IpExtOutOctets 2013953 0.0
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Acked-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2014-12-08 03:22:18 +07:00
|
|
|
limit = mss_now;
|
2015-05-26 22:55:28 +07:00
|
|
|
if (tso_segs > 1 && !tcp_urg_mode(tp))
|
tcp: refine TSO autosizing
Commit 95bd09eb2750 ("tcp: TSO packets automatic sizing") tried to
control TSO size, but did this at the wrong place (sendmsg() time)
At sendmsg() time, we might have a pessimistic view of flow rate,
and we end up building very small skbs (with 2 MSS per skb).
This is bad because :
- It sends small TSO packets even in Slow Start where rate quickly
increases.
- It tends to make socket write queue very big, increasing tcp_ack()
processing time, but also increasing memory needs, not necessarily
accounted for, as fast clones overhead is currently ignored.
- Lower GRO efficiency and more ACK packets.
Servers with a lot of small lived connections suffer from this.
Lets instead fill skbs as much as possible (64KB of payload), but split
them at xmit time, when we have a precise idea of the flow rate.
skb split is actually quite efficient.
Patch looks bigger than necessary, because TCP Small Queue decision now
has to take place after the eventual split.
As Neal suggested, introduce a new tcp_tso_autosize() helper, so that
tcp_tso_should_defer() can be synchronized on same goal.
Rename tp->xmit_size_goal_segs to tp->gso_segs, as this variable
contains number of mss that we can put in GSO packet, and is not
related to the autosizing goal anymore.
Tested:
40 ms rtt link
nstat >/dev/null
netperf -H remote -l -2000000 -- -s 1000000
nstat | egrep "IpInReceives|IpOutRequests|TcpOutSegs|IpExtOutOctets"
Before patch :
Recv Send Send
Socket Socket Message Elapsed
Size Size Size Time Throughput
bytes bytes bytes secs. 10^6bits/s
87380 2000000 2000000 0.36 44.22
IpInReceives 600 0.0
IpOutRequests 599 0.0
TcpOutSegs 1397 0.0
IpExtOutOctets 2033249 0.0
After patch :
Recv Send Send
Socket Socket Message Elapsed
Size Size Size Time Throughput
bytes bytes bytes secs. 10^6bits/sec
87380 2000000 2000000 0.36 44.27
IpInReceives 221 0.0
IpOutRequests 232 0.0
TcpOutSegs 1397 0.0
IpExtOutOctets 2013953 0.0
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Acked-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2014-12-08 03:22:18 +07:00
|
|
|
limit = tcp_mss_split_point(sk, skb, mss_now,
|
|
|
|
min_t(unsigned int,
|
|
|
|
cwnd_quota,
|
|
|
|
max_segs),
|
|
|
|
nonagle);
|
|
|
|
|
|
|
|
if (skb->len > limit &&
|
|
|
|
unlikely(tso_fragment(sk, skb, limit, mss_now, gfp)))
|
|
|
|
break;
|
|
|
|
|
tcp: TSQ can use a dynamic limit
When TCP Small Queues was added, we used a sysctl to limit amount of
packets queues on Qdisc/device queues for a given TCP flow.
Problem is this limit is either too big for low rates, or too small
for high rates.
Now TCP stack has rate estimation in sk->sk_pacing_rate, and TSO
auto sizing, it can better control number of packets in Qdisc/device
queues.
New limit is two packets or at least 1 to 2 ms worth of packets.
Low rates flows benefit from this patch by having even smaller
number of packets in queues, allowing for faster recovery,
better RTT estimations.
High rates flows benefit from this patch by allowing more than 2 packets
in flight as we had reports this was a limiting factor to reach line
rate. [ In particular if TX completion is delayed because of coalescing
parameters ]
Example for a single flow on 10Gbp link controlled by FQ/pacing
14 packets in flight instead of 2
$ tc -s -d qd
qdisc fq 8001: dev eth0 root refcnt 32 limit 10000p flow_limit 100p
buckets 1024 quantum 3028 initial_quantum 15140
Sent 1168459366606 bytes 771822841 pkt (dropped 0, overlimits 0
requeues 6822476)
rate 9346Mbit 771713pps backlog 953820b 14p requeues 6822476
2047 flow, 2046 inactive, 1 throttled, delay 15673 ns
2372 gc, 0 highprio, 0 retrans, 9739249 throttled, 0 flows_plimit
Note that sk_pacing_rate is currently set to twice the actual rate, but
this might be refined in the future when a flow is in congestion
avoidance.
Additional change : skb->destructor should be set to tcp_wfree().
A future patch (for linux 3.13+) might remove tcp_limit_output_bytes
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Wei Liu <wei.liu2@citrix.com>
Cc: Cong Wang <xiyou.wangcong@gmail.com>
Cc: Yuchung Cheng <ycheng@google.com>
Cc: Neal Cardwell <ncardwell@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-09-27 17:28:54 +07:00
|
|
|
/* TCP Small Queues :
|
|
|
|
* Control number of packets in qdisc/devices to two packets / or ~1 ms.
|
|
|
|
* This allows for :
|
|
|
|
* - better RTT estimation and ACK scheduling
|
|
|
|
* - faster recovery
|
|
|
|
* - high rates
|
2013-11-13 21:32:54 +07:00
|
|
|
* Alas, some drivers / subsystems require a fair amount
|
|
|
|
* of queued bytes to ensure line rate.
|
|
|
|
* One example is wifi aggregation (802.11 AMPDU)
|
tcp: TCP Small Queues
This introduce TSQ (TCP Small Queues)
TSQ goal is to reduce number of TCP packets in xmit queues (qdisc &
device queues), to reduce RTT and cwnd bias, part of the bufferbloat
problem.
sk->sk_wmem_alloc not allowed to grow above a given limit,
allowing no more than ~128KB [1] per tcp socket in qdisc/dev layers at a
given time.
TSO packets are sized/capped to half the limit, so that we have two
TSO packets in flight, allowing better bandwidth use.
As a side effect, setting the limit to 40000 automatically reduces the
standard gso max limit (65536) to 40000/2 : It can help to reduce
latencies of high prio packets, having smaller TSO packets.
This means we divert sock_wfree() to a tcp_wfree() handler, to
queue/send following frames when skb_orphan() [2] is called for the
already queued skbs.
Results on my dev machines (tg3/ixgbe nics) are really impressive,
using standard pfifo_fast, and with or without TSO/GSO.
Without reduction of nominal bandwidth, we have reduction of buffering
per bulk sender :
< 1ms on Gbit (instead of 50ms with TSO)
< 8ms on 100Mbit (instead of 132 ms)
I no longer have 4 MBytes backlogged in qdisc by a single netperf
session, and both side socket autotuning no longer use 4 Mbytes.
As skb destructor cannot restart xmit itself ( as qdisc lock might be
taken at this point ), we delegate the work to a tasklet. We use one
tasklest per cpu for performance reasons.
If tasklet finds a socket owned by the user, it sets TSQ_OWNED flag.
This flag is tested in a new protocol method called from release_sock(),
to eventually send new segments.
[1] New /proc/sys/net/ipv4/tcp_limit_output_bytes tunable
[2] skb_orphan() is usually called at TX completion time,
but some drivers call it in their start_xmit() handler.
These drivers should at least use BQL, or else a single TCP
session can still fill the whole NIC TX ring, since TSQ will
have no effect.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Dave Taht <dave.taht@bufferbloat.net>
Cc: Tom Herbert <therbert@google.com>
Cc: Matt Mathis <mattmathis@google.com>
Cc: Yuchung Cheng <ycheng@google.com>
Cc: Nandita Dukkipati <nanditad@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2012-07-11 12:50:31 +07:00
|
|
|
*/
|
tcp: refine TSO autosizing
Commit 95bd09eb2750 ("tcp: TSO packets automatic sizing") tried to
control TSO size, but did this at the wrong place (sendmsg() time)
At sendmsg() time, we might have a pessimistic view of flow rate,
and we end up building very small skbs (with 2 MSS per skb).
This is bad because :
- It sends small TSO packets even in Slow Start where rate quickly
increases.
- It tends to make socket write queue very big, increasing tcp_ack()
processing time, but also increasing memory needs, not necessarily
accounted for, as fast clones overhead is currently ignored.
- Lower GRO efficiency and more ACK packets.
Servers with a lot of small lived connections suffer from this.
Lets instead fill skbs as much as possible (64KB of payload), but split
them at xmit time, when we have a precise idea of the flow rate.
skb split is actually quite efficient.
Patch looks bigger than necessary, because TCP Small Queue decision now
has to take place after the eventual split.
As Neal suggested, introduce a new tcp_tso_autosize() helper, so that
tcp_tso_should_defer() can be synchronized on same goal.
Rename tp->xmit_size_goal_segs to tp->gso_segs, as this variable
contains number of mss that we can put in GSO packet, and is not
related to the autosizing goal anymore.
Tested:
40 ms rtt link
nstat >/dev/null
netperf -H remote -l -2000000 -- -s 1000000
nstat | egrep "IpInReceives|IpOutRequests|TcpOutSegs|IpExtOutOctets"
Before patch :
Recv Send Send
Socket Socket Message Elapsed
Size Size Size Time Throughput
bytes bytes bytes secs. 10^6bits/s
87380 2000000 2000000 0.36 44.22
IpInReceives 600 0.0
IpOutRequests 599 0.0
TcpOutSegs 1397 0.0
IpExtOutOctets 2033249 0.0
After patch :
Recv Send Send
Socket Socket Message Elapsed
Size Size Size Time Throughput
bytes bytes bytes secs. 10^6bits/sec
87380 2000000 2000000 0.36 44.27
IpInReceives 221 0.0
IpOutRequests 232 0.0
TcpOutSegs 1397 0.0
IpExtOutOctets 2013953 0.0
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Acked-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2014-12-08 03:22:18 +07:00
|
|
|
limit = max(2 * skb->truesize, sk->sk_pacing_rate >> 10);
|
|
|
|
limit = min_t(u32, limit, sysctl_tcp_limit_output_bytes);
|
tcp: TSQ can use a dynamic limit
When TCP Small Queues was added, we used a sysctl to limit amount of
packets queues on Qdisc/device queues for a given TCP flow.
Problem is this limit is either too big for low rates, or too small
for high rates.
Now TCP stack has rate estimation in sk->sk_pacing_rate, and TSO
auto sizing, it can better control number of packets in Qdisc/device
queues.
New limit is two packets or at least 1 to 2 ms worth of packets.
Low rates flows benefit from this patch by having even smaller
number of packets in queues, allowing for faster recovery,
better RTT estimations.
High rates flows benefit from this patch by allowing more than 2 packets
in flight as we had reports this was a limiting factor to reach line
rate. [ In particular if TX completion is delayed because of coalescing
parameters ]
Example for a single flow on 10Gbp link controlled by FQ/pacing
14 packets in flight instead of 2
$ tc -s -d qd
qdisc fq 8001: dev eth0 root refcnt 32 limit 10000p flow_limit 100p
buckets 1024 quantum 3028 initial_quantum 15140
Sent 1168459366606 bytes 771822841 pkt (dropped 0, overlimits 0
requeues 6822476)
rate 9346Mbit 771713pps backlog 953820b 14p requeues 6822476
2047 flow, 2046 inactive, 1 throttled, delay 15673 ns
2372 gc, 0 highprio, 0 retrans, 9739249 throttled, 0 flows_plimit
Note that sk_pacing_rate is currently set to twice the actual rate, but
this might be refined in the future when a flow is in congestion
avoidance.
Additional change : skb->destructor should be set to tcp_wfree().
A future patch (for linux 3.13+) might remove tcp_limit_output_bytes
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Wei Liu <wei.liu2@citrix.com>
Cc: Cong Wang <xiyou.wangcong@gmail.com>
Cc: Yuchung Cheng <ycheng@google.com>
Cc: Neal Cardwell <ncardwell@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-09-27 17:28:54 +07:00
|
|
|
|
|
|
|
if (atomic_read(&sk->sk_wmem_alloc) > limit) {
|
tcp: TCP Small Queues
This introduce TSQ (TCP Small Queues)
TSQ goal is to reduce number of TCP packets in xmit queues (qdisc &
device queues), to reduce RTT and cwnd bias, part of the bufferbloat
problem.
sk->sk_wmem_alloc not allowed to grow above a given limit,
allowing no more than ~128KB [1] per tcp socket in qdisc/dev layers at a
given time.
TSO packets are sized/capped to half the limit, so that we have two
TSO packets in flight, allowing better bandwidth use.
As a side effect, setting the limit to 40000 automatically reduces the
standard gso max limit (65536) to 40000/2 : It can help to reduce
latencies of high prio packets, having smaller TSO packets.
This means we divert sock_wfree() to a tcp_wfree() handler, to
queue/send following frames when skb_orphan() [2] is called for the
already queued skbs.
Results on my dev machines (tg3/ixgbe nics) are really impressive,
using standard pfifo_fast, and with or without TSO/GSO.
Without reduction of nominal bandwidth, we have reduction of buffering
per bulk sender :
< 1ms on Gbit (instead of 50ms with TSO)
< 8ms on 100Mbit (instead of 132 ms)
I no longer have 4 MBytes backlogged in qdisc by a single netperf
session, and both side socket autotuning no longer use 4 Mbytes.
As skb destructor cannot restart xmit itself ( as qdisc lock might be
taken at this point ), we delegate the work to a tasklet. We use one
tasklest per cpu for performance reasons.
If tasklet finds a socket owned by the user, it sets TSQ_OWNED flag.
This flag is tested in a new protocol method called from release_sock(),
to eventually send new segments.
[1] New /proc/sys/net/ipv4/tcp_limit_output_bytes tunable
[2] skb_orphan() is usually called at TX completion time,
but some drivers call it in their start_xmit() handler.
These drivers should at least use BQL, or else a single TCP
session can still fill the whole NIC TX ring, since TSQ will
have no effect.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Dave Taht <dave.taht@bufferbloat.net>
Cc: Tom Herbert <therbert@google.com>
Cc: Matt Mathis <mattmathis@google.com>
Cc: Yuchung Cheng <ycheng@google.com>
Cc: Nandita Dukkipati <nanditad@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2012-07-11 12:50:31 +07:00
|
|
|
set_bit(TSQ_THROTTLED, &tp->tsq_flags);
|
2014-02-10 09:40:11 +07:00
|
|
|
/* It is possible TX completion already happened
|
|
|
|
* before we set TSQ_THROTTLED, so we must
|
|
|
|
* test again the condition.
|
|
|
|
*/
|
2014-03-18 00:06:10 +07:00
|
|
|
smp_mb__after_atomic();
|
2014-02-10 09:40:11 +07:00
|
|
|
if (atomic_read(&sk->sk_wmem_alloc) > limit)
|
|
|
|
break;
|
tcp: TCP Small Queues
This introduce TSQ (TCP Small Queues)
TSQ goal is to reduce number of TCP packets in xmit queues (qdisc &
device queues), to reduce RTT and cwnd bias, part of the bufferbloat
problem.
sk->sk_wmem_alloc not allowed to grow above a given limit,
allowing no more than ~128KB [1] per tcp socket in qdisc/dev layers at a
given time.
TSO packets are sized/capped to half the limit, so that we have two
TSO packets in flight, allowing better bandwidth use.
As a side effect, setting the limit to 40000 automatically reduces the
standard gso max limit (65536) to 40000/2 : It can help to reduce
latencies of high prio packets, having smaller TSO packets.
This means we divert sock_wfree() to a tcp_wfree() handler, to
queue/send following frames when skb_orphan() [2] is called for the
already queued skbs.
Results on my dev machines (tg3/ixgbe nics) are really impressive,
using standard pfifo_fast, and with or without TSO/GSO.
Without reduction of nominal bandwidth, we have reduction of buffering
per bulk sender :
< 1ms on Gbit (instead of 50ms with TSO)
< 8ms on 100Mbit (instead of 132 ms)
I no longer have 4 MBytes backlogged in qdisc by a single netperf
session, and both side socket autotuning no longer use 4 Mbytes.
As skb destructor cannot restart xmit itself ( as qdisc lock might be
taken at this point ), we delegate the work to a tasklet. We use one
tasklest per cpu for performance reasons.
If tasklet finds a socket owned by the user, it sets TSQ_OWNED flag.
This flag is tested in a new protocol method called from release_sock(),
to eventually send new segments.
[1] New /proc/sys/net/ipv4/tcp_limit_output_bytes tunable
[2] skb_orphan() is usually called at TX completion time,
but some drivers call it in their start_xmit() handler.
These drivers should at least use BQL, or else a single TCP
session can still fill the whole NIC TX ring, since TSQ will
have no effect.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Dave Taht <dave.taht@bufferbloat.net>
Cc: Tom Herbert <therbert@google.com>
Cc: Matt Mathis <mattmathis@google.com>
Cc: Yuchung Cheng <ycheng@google.com>
Cc: Nandita Dukkipati <nanditad@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2012-07-11 12:50:31 +07:00
|
|
|
}
|
tcp: TSQ can use a dynamic limit
When TCP Small Queues was added, we used a sysctl to limit amount of
packets queues on Qdisc/device queues for a given TCP flow.
Problem is this limit is either too big for low rates, or too small
for high rates.
Now TCP stack has rate estimation in sk->sk_pacing_rate, and TSO
auto sizing, it can better control number of packets in Qdisc/device
queues.
New limit is two packets or at least 1 to 2 ms worth of packets.
Low rates flows benefit from this patch by having even smaller
number of packets in queues, allowing for faster recovery,
better RTT estimations.
High rates flows benefit from this patch by allowing more than 2 packets
in flight as we had reports this was a limiting factor to reach line
rate. [ In particular if TX completion is delayed because of coalescing
parameters ]
Example for a single flow on 10Gbp link controlled by FQ/pacing
14 packets in flight instead of 2
$ tc -s -d qd
qdisc fq 8001: dev eth0 root refcnt 32 limit 10000p flow_limit 100p
buckets 1024 quantum 3028 initial_quantum 15140
Sent 1168459366606 bytes 771822841 pkt (dropped 0, overlimits 0
requeues 6822476)
rate 9346Mbit 771713pps backlog 953820b 14p requeues 6822476
2047 flow, 2046 inactive, 1 throttled, delay 15673 ns
2372 gc, 0 highprio, 0 retrans, 9739249 throttled, 0 flows_plimit
Note that sk_pacing_rate is currently set to twice the actual rate, but
this might be refined in the future when a flow is in congestion
avoidance.
Additional change : skb->destructor should be set to tcp_wfree().
A future patch (for linux 3.13+) might remove tcp_limit_output_bytes
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Wei Liu <wei.liu2@citrix.com>
Cc: Cong Wang <xiyou.wangcong@gmail.com>
Cc: Yuchung Cheng <ycheng@google.com>
Cc: Neal Cardwell <ncardwell@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-09-27 17:28:54 +07:00
|
|
|
|
2008-12-06 13:48:55 +07:00
|
|
|
if (unlikely(tcp_transmit_skb(sk, skb, 1, gfp)))
|
2005-07-06 05:19:06 +07:00
|
|
|
break;
|
2005-04-17 05:20:36 +07:00
|
|
|
|
2012-11-15 11:03:17 +07:00
|
|
|
repair:
|
2005-07-06 05:19:06 +07:00
|
|
|
/* Advance the send_head. This one is sent out.
|
|
|
|
* This call will increment packets_out.
|
|
|
|
*/
|
2007-12-31 19:43:57 +07:00
|
|
|
tcp_event_new_data_sent(sk, skb);
|
2005-04-17 05:20:36 +07:00
|
|
|
|
2005-07-06 05:19:06 +07:00
|
|
|
tcp_minshall_update(tp, mss_now, skb);
|
2011-08-22 03:21:57 +07:00
|
|
|
sent_pkts += tcp_skb_pcount(skb);
|
2008-12-06 13:48:55 +07:00
|
|
|
|
|
|
|
if (push_one)
|
|
|
|
break;
|
2005-07-06 05:19:06 +07:00
|
|
|
}
|
2005-04-17 05:20:36 +07:00
|
|
|
|
2005-07-06 05:20:09 +07:00
|
|
|
if (likely(sent_pkts)) {
|
2012-09-03 00:38:04 +07:00
|
|
|
if (tcp_in_cwnd_reduction(sk))
|
|
|
|
tp->prr_out += sent_pkts;
|
tcp: Tail loss probe (TLP)
This patch series implement the Tail loss probe (TLP) algorithm described
in http://tools.ietf.org/html/draft-dukkipati-tcpm-tcp-loss-probe-01. The
first patch implements the basic algorithm.
TLP's goal is to reduce tail latency of short transactions. It achieves
this by converting retransmission timeouts (RTOs) occuring due
to tail losses (losses at end of transactions) into fast recovery.
TLP transmits one packet in two round-trips when a connection is in
Open state and isn't receiving any ACKs. The transmitted packet, aka
loss probe, can be either new or a retransmission. When there is tail
loss, the ACK from a loss probe triggers FACK/early-retransmit based
fast recovery, thus avoiding a costly RTO. In the absence of loss,
there is no change in the connection state.
PTO stands for probe timeout. It is a timer event indicating
that an ACK is overdue and triggers a loss probe packet. The PTO value
is set to max(2*SRTT, 10ms) and is adjusted to account for delayed
ACK timer when there is only one oustanding packet.
TLP Algorithm
On transmission of new data in Open state:
-> packets_out > 1: schedule PTO in max(2*SRTT, 10ms).
-> packets_out == 1: schedule PTO in max(2*RTT, 1.5*RTT + 200ms)
-> PTO = min(PTO, RTO)
Conditions for scheduling PTO:
-> Connection is in Open state.
-> Connection is either cwnd limited or no new data to send.
-> Number of probes per tail loss episode is limited to one.
-> Connection is SACK enabled.
When PTO fires:
new_segment_exists:
-> transmit new segment.
-> packets_out++. cwnd remains same.
no_new_packet:
-> retransmit the last segment.
Its ACK triggers FACK or early retransmit based recovery.
ACK path:
-> rearm RTO at start of ACK processing.
-> reschedule PTO if need be.
In addition, the patch includes a small variation to the Early Retransmit
(ER) algorithm, such that ER and TLP together can in principle recover any
N-degree of tail loss through fast recovery. TLP is controlled by the same
sysctl as ER, tcp_early_retrans sysctl.
tcp_early_retrans==0; disables TLP and ER.
==1; enables RFC5827 ER.
==2; delayed ER.
==3; TLP and delayed ER. [DEFAULT]
==4; TLP only.
The TLP patch series have been extensively tested on Google Web servers.
It is most effective for short Web trasactions, where it reduced RTOs by 15%
and improved HTTP response time (average by 6%, 99th percentile by 10%).
The transmitted probes account for <0.5% of the overall transmissions.
Signed-off-by: Nandita Dukkipati <nanditad@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Acked-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-03-11 17:00:43 +07:00
|
|
|
|
|
|
|
/* Send one loss probe per tail loss episode. */
|
|
|
|
if (push_one != 2)
|
|
|
|
tcp_schedule_loss_probe(sk);
|
2015-09-23 23:49:53 +07:00
|
|
|
is_cwnd_limited |= (tcp_packets_in_flight(tp) >= tp->snd_cwnd);
|
2014-05-22 21:41:08 +07:00
|
|
|
tcp_cwnd_validate(sk, is_cwnd_limited);
|
2012-05-17 06:15:34 +07:00
|
|
|
return false;
|
2005-04-17 05:20:36 +07:00
|
|
|
}
|
2015-08-13 01:18:19 +07:00
|
|
|
return !tp->packets_out && tcp_send_head(sk);
|
tcp: Tail loss probe (TLP)
This patch series implement the Tail loss probe (TLP) algorithm described
in http://tools.ietf.org/html/draft-dukkipati-tcpm-tcp-loss-probe-01. The
first patch implements the basic algorithm.
TLP's goal is to reduce tail latency of short transactions. It achieves
this by converting retransmission timeouts (RTOs) occuring due
to tail losses (losses at end of transactions) into fast recovery.
TLP transmits one packet in two round-trips when a connection is in
Open state and isn't receiving any ACKs. The transmitted packet, aka
loss probe, can be either new or a retransmission. When there is tail
loss, the ACK from a loss probe triggers FACK/early-retransmit based
fast recovery, thus avoiding a costly RTO. In the absence of loss,
there is no change in the connection state.
PTO stands for probe timeout. It is a timer event indicating
that an ACK is overdue and triggers a loss probe packet. The PTO value
is set to max(2*SRTT, 10ms) and is adjusted to account for delayed
ACK timer when there is only one oustanding packet.
TLP Algorithm
On transmission of new data in Open state:
-> packets_out > 1: schedule PTO in max(2*SRTT, 10ms).
-> packets_out == 1: schedule PTO in max(2*RTT, 1.5*RTT + 200ms)
-> PTO = min(PTO, RTO)
Conditions for scheduling PTO:
-> Connection is in Open state.
-> Connection is either cwnd limited or no new data to send.
-> Number of probes per tail loss episode is limited to one.
-> Connection is SACK enabled.
When PTO fires:
new_segment_exists:
-> transmit new segment.
-> packets_out++. cwnd remains same.
no_new_packet:
-> retransmit the last segment.
Its ACK triggers FACK or early retransmit based recovery.
ACK path:
-> rearm RTO at start of ACK processing.
-> reschedule PTO if need be.
In addition, the patch includes a small variation to the Early Retransmit
(ER) algorithm, such that ER and TLP together can in principle recover any
N-degree of tail loss through fast recovery. TLP is controlled by the same
sysctl as ER, tcp_early_retrans sysctl.
tcp_early_retrans==0; disables TLP and ER.
==1; enables RFC5827 ER.
==2; delayed ER.
==3; TLP and delayed ER. [DEFAULT]
==4; TLP only.
The TLP patch series have been extensively tested on Google Web servers.
It is most effective for short Web trasactions, where it reduced RTOs by 15%
and improved HTTP response time (average by 6%, 99th percentile by 10%).
The transmitted probes account for <0.5% of the overall transmissions.
Signed-off-by: Nandita Dukkipati <nanditad@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Acked-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-03-11 17:00:43 +07:00
|
|
|
}
|
|
|
|
|
|
|
|
bool tcp_schedule_loss_probe(struct sock *sk)
|
|
|
|
{
|
|
|
|
struct inet_connection_sock *icsk = inet_csk(sk);
|
|
|
|
struct tcp_sock *tp = tcp_sk(sk);
|
|
|
|
u32 timeout, tlp_time_stamp, rto_time_stamp;
|
2014-02-27 05:02:48 +07:00
|
|
|
u32 rtt = usecs_to_jiffies(tp->srtt_us >> 3);
|
tcp: Tail loss probe (TLP)
This patch series implement the Tail loss probe (TLP) algorithm described
in http://tools.ietf.org/html/draft-dukkipati-tcpm-tcp-loss-probe-01. The
first patch implements the basic algorithm.
TLP's goal is to reduce tail latency of short transactions. It achieves
this by converting retransmission timeouts (RTOs) occuring due
to tail losses (losses at end of transactions) into fast recovery.
TLP transmits one packet in two round-trips when a connection is in
Open state and isn't receiving any ACKs. The transmitted packet, aka
loss probe, can be either new or a retransmission. When there is tail
loss, the ACK from a loss probe triggers FACK/early-retransmit based
fast recovery, thus avoiding a costly RTO. In the absence of loss,
there is no change in the connection state.
PTO stands for probe timeout. It is a timer event indicating
that an ACK is overdue and triggers a loss probe packet. The PTO value
is set to max(2*SRTT, 10ms) and is adjusted to account for delayed
ACK timer when there is only one oustanding packet.
TLP Algorithm
On transmission of new data in Open state:
-> packets_out > 1: schedule PTO in max(2*SRTT, 10ms).
-> packets_out == 1: schedule PTO in max(2*RTT, 1.5*RTT + 200ms)
-> PTO = min(PTO, RTO)
Conditions for scheduling PTO:
-> Connection is in Open state.
-> Connection is either cwnd limited or no new data to send.
-> Number of probes per tail loss episode is limited to one.
-> Connection is SACK enabled.
When PTO fires:
new_segment_exists:
-> transmit new segment.
-> packets_out++. cwnd remains same.
no_new_packet:
-> retransmit the last segment.
Its ACK triggers FACK or early retransmit based recovery.
ACK path:
-> rearm RTO at start of ACK processing.
-> reschedule PTO if need be.
In addition, the patch includes a small variation to the Early Retransmit
(ER) algorithm, such that ER and TLP together can in principle recover any
N-degree of tail loss through fast recovery. TLP is controlled by the same
sysctl as ER, tcp_early_retrans sysctl.
tcp_early_retrans==0; disables TLP and ER.
==1; enables RFC5827 ER.
==2; delayed ER.
==3; TLP and delayed ER. [DEFAULT]
==4; TLP only.
The TLP patch series have been extensively tested on Google Web servers.
It is most effective for short Web trasactions, where it reduced RTOs by 15%
and improved HTTP response time (average by 6%, 99th percentile by 10%).
The transmitted probes account for <0.5% of the overall transmissions.
Signed-off-by: Nandita Dukkipati <nanditad@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Acked-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-03-11 17:00:43 +07:00
|
|
|
|
|
|
|
if (WARN_ON(icsk->icsk_pending == ICSK_TIME_EARLY_RETRANS))
|
|
|
|
return false;
|
|
|
|
/* No consecutive loss probes. */
|
|
|
|
if (WARN_ON(icsk->icsk_pending == ICSK_TIME_LOSS_PROBE)) {
|
|
|
|
tcp_rearm_rto(sk);
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
/* Don't do any loss probe on a Fast Open connection before 3WHS
|
|
|
|
* finishes.
|
|
|
|
*/
|
2015-09-19 01:40:33 +07:00
|
|
|
if (tp->fastopen_rsk)
|
tcp: Tail loss probe (TLP)
This patch series implement the Tail loss probe (TLP) algorithm described
in http://tools.ietf.org/html/draft-dukkipati-tcpm-tcp-loss-probe-01. The
first patch implements the basic algorithm.
TLP's goal is to reduce tail latency of short transactions. It achieves
this by converting retransmission timeouts (RTOs) occuring due
to tail losses (losses at end of transactions) into fast recovery.
TLP transmits one packet in two round-trips when a connection is in
Open state and isn't receiving any ACKs. The transmitted packet, aka
loss probe, can be either new or a retransmission. When there is tail
loss, the ACK from a loss probe triggers FACK/early-retransmit based
fast recovery, thus avoiding a costly RTO. In the absence of loss,
there is no change in the connection state.
PTO stands for probe timeout. It is a timer event indicating
that an ACK is overdue and triggers a loss probe packet. The PTO value
is set to max(2*SRTT, 10ms) and is adjusted to account for delayed
ACK timer when there is only one oustanding packet.
TLP Algorithm
On transmission of new data in Open state:
-> packets_out > 1: schedule PTO in max(2*SRTT, 10ms).
-> packets_out == 1: schedule PTO in max(2*RTT, 1.5*RTT + 200ms)
-> PTO = min(PTO, RTO)
Conditions for scheduling PTO:
-> Connection is in Open state.
-> Connection is either cwnd limited or no new data to send.
-> Number of probes per tail loss episode is limited to one.
-> Connection is SACK enabled.
When PTO fires:
new_segment_exists:
-> transmit new segment.
-> packets_out++. cwnd remains same.
no_new_packet:
-> retransmit the last segment.
Its ACK triggers FACK or early retransmit based recovery.
ACK path:
-> rearm RTO at start of ACK processing.
-> reschedule PTO if need be.
In addition, the patch includes a small variation to the Early Retransmit
(ER) algorithm, such that ER and TLP together can in principle recover any
N-degree of tail loss through fast recovery. TLP is controlled by the same
sysctl as ER, tcp_early_retrans sysctl.
tcp_early_retrans==0; disables TLP and ER.
==1; enables RFC5827 ER.
==2; delayed ER.
==3; TLP and delayed ER. [DEFAULT]
==4; TLP only.
The TLP patch series have been extensively tested on Google Web servers.
It is most effective for short Web trasactions, where it reduced RTOs by 15%
and improved HTTP response time (average by 6%, 99th percentile by 10%).
The transmitted probes account for <0.5% of the overall transmissions.
Signed-off-by: Nandita Dukkipati <nanditad@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Acked-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-03-11 17:00:43 +07:00
|
|
|
return false;
|
|
|
|
|
|
|
|
/* TLP is only scheduled when next timer event is RTO. */
|
|
|
|
if (icsk->icsk_pending != ICSK_TIME_RETRANS)
|
|
|
|
return false;
|
|
|
|
|
|
|
|
/* Schedule a loss probe in 2*RTT for SACK capable connections
|
|
|
|
* in Open state, that are either limited by cwnd or application.
|
|
|
|
*/
|
2015-09-19 01:40:33 +07:00
|
|
|
if (sysctl_tcp_early_retrans < 3 || !tp->packets_out ||
|
tcp: Tail loss probe (TLP)
This patch series implement the Tail loss probe (TLP) algorithm described
in http://tools.ietf.org/html/draft-dukkipati-tcpm-tcp-loss-probe-01. The
first patch implements the basic algorithm.
TLP's goal is to reduce tail latency of short transactions. It achieves
this by converting retransmission timeouts (RTOs) occuring due
to tail losses (losses at end of transactions) into fast recovery.
TLP transmits one packet in two round-trips when a connection is in
Open state and isn't receiving any ACKs. The transmitted packet, aka
loss probe, can be either new or a retransmission. When there is tail
loss, the ACK from a loss probe triggers FACK/early-retransmit based
fast recovery, thus avoiding a costly RTO. In the absence of loss,
there is no change in the connection state.
PTO stands for probe timeout. It is a timer event indicating
that an ACK is overdue and triggers a loss probe packet. The PTO value
is set to max(2*SRTT, 10ms) and is adjusted to account for delayed
ACK timer when there is only one oustanding packet.
TLP Algorithm
On transmission of new data in Open state:
-> packets_out > 1: schedule PTO in max(2*SRTT, 10ms).
-> packets_out == 1: schedule PTO in max(2*RTT, 1.5*RTT + 200ms)
-> PTO = min(PTO, RTO)
Conditions for scheduling PTO:
-> Connection is in Open state.
-> Connection is either cwnd limited or no new data to send.
-> Number of probes per tail loss episode is limited to one.
-> Connection is SACK enabled.
When PTO fires:
new_segment_exists:
-> transmit new segment.
-> packets_out++. cwnd remains same.
no_new_packet:
-> retransmit the last segment.
Its ACK triggers FACK or early retransmit based recovery.
ACK path:
-> rearm RTO at start of ACK processing.
-> reschedule PTO if need be.
In addition, the patch includes a small variation to the Early Retransmit
(ER) algorithm, such that ER and TLP together can in principle recover any
N-degree of tail loss through fast recovery. TLP is controlled by the same
sysctl as ER, tcp_early_retrans sysctl.
tcp_early_retrans==0; disables TLP and ER.
==1; enables RFC5827 ER.
==2; delayed ER.
==3; TLP and delayed ER. [DEFAULT]
==4; TLP only.
The TLP patch series have been extensively tested on Google Web servers.
It is most effective for short Web trasactions, where it reduced RTOs by 15%
and improved HTTP response time (average by 6%, 99th percentile by 10%).
The transmitted probes account for <0.5% of the overall transmissions.
Signed-off-by: Nandita Dukkipati <nanditad@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Acked-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-03-11 17:00:43 +07:00
|
|
|
!tcp_is_sack(tp) || inet_csk(sk)->icsk_ca_state != TCP_CA_Open)
|
|
|
|
return false;
|
|
|
|
|
|
|
|
if ((tp->snd_cwnd > tcp_packets_in_flight(tp)) &&
|
|
|
|
tcp_send_head(sk))
|
|
|
|
return false;
|
|
|
|
|
|
|
|
/* Probe timeout is at least 1.5*rtt + TCP_DELACK_MAX to account
|
2015-09-19 01:40:33 +07:00
|
|
|
* for delayed ack when there's one outstanding packet. If no RTT
|
|
|
|
* sample is available then probe after TCP_TIMEOUT_INIT.
|
tcp: Tail loss probe (TLP)
This patch series implement the Tail loss probe (TLP) algorithm described
in http://tools.ietf.org/html/draft-dukkipati-tcpm-tcp-loss-probe-01. The
first patch implements the basic algorithm.
TLP's goal is to reduce tail latency of short transactions. It achieves
this by converting retransmission timeouts (RTOs) occuring due
to tail losses (losses at end of transactions) into fast recovery.
TLP transmits one packet in two round-trips when a connection is in
Open state and isn't receiving any ACKs. The transmitted packet, aka
loss probe, can be either new or a retransmission. When there is tail
loss, the ACK from a loss probe triggers FACK/early-retransmit based
fast recovery, thus avoiding a costly RTO. In the absence of loss,
there is no change in the connection state.
PTO stands for probe timeout. It is a timer event indicating
that an ACK is overdue and triggers a loss probe packet. The PTO value
is set to max(2*SRTT, 10ms) and is adjusted to account for delayed
ACK timer when there is only one oustanding packet.
TLP Algorithm
On transmission of new data in Open state:
-> packets_out > 1: schedule PTO in max(2*SRTT, 10ms).
-> packets_out == 1: schedule PTO in max(2*RTT, 1.5*RTT + 200ms)
-> PTO = min(PTO, RTO)
Conditions for scheduling PTO:
-> Connection is in Open state.
-> Connection is either cwnd limited or no new data to send.
-> Number of probes per tail loss episode is limited to one.
-> Connection is SACK enabled.
When PTO fires:
new_segment_exists:
-> transmit new segment.
-> packets_out++. cwnd remains same.
no_new_packet:
-> retransmit the last segment.
Its ACK triggers FACK or early retransmit based recovery.
ACK path:
-> rearm RTO at start of ACK processing.
-> reschedule PTO if need be.
In addition, the patch includes a small variation to the Early Retransmit
(ER) algorithm, such that ER and TLP together can in principle recover any
N-degree of tail loss through fast recovery. TLP is controlled by the same
sysctl as ER, tcp_early_retrans sysctl.
tcp_early_retrans==0; disables TLP and ER.
==1; enables RFC5827 ER.
==2; delayed ER.
==3; TLP and delayed ER. [DEFAULT]
==4; TLP only.
The TLP patch series have been extensively tested on Google Web servers.
It is most effective for short Web trasactions, where it reduced RTOs by 15%
and improved HTTP response time (average by 6%, 99th percentile by 10%).
The transmitted probes account for <0.5% of the overall transmissions.
Signed-off-by: Nandita Dukkipati <nanditad@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Acked-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-03-11 17:00:43 +07:00
|
|
|
*/
|
2015-09-19 01:40:33 +07:00
|
|
|
timeout = rtt << 1 ? : TCP_TIMEOUT_INIT;
|
tcp: Tail loss probe (TLP)
This patch series implement the Tail loss probe (TLP) algorithm described
in http://tools.ietf.org/html/draft-dukkipati-tcpm-tcp-loss-probe-01. The
first patch implements the basic algorithm.
TLP's goal is to reduce tail latency of short transactions. It achieves
this by converting retransmission timeouts (RTOs) occuring due
to tail losses (losses at end of transactions) into fast recovery.
TLP transmits one packet in two round-trips when a connection is in
Open state and isn't receiving any ACKs. The transmitted packet, aka
loss probe, can be either new or a retransmission. When there is tail
loss, the ACK from a loss probe triggers FACK/early-retransmit based
fast recovery, thus avoiding a costly RTO. In the absence of loss,
there is no change in the connection state.
PTO stands for probe timeout. It is a timer event indicating
that an ACK is overdue and triggers a loss probe packet. The PTO value
is set to max(2*SRTT, 10ms) and is adjusted to account for delayed
ACK timer when there is only one oustanding packet.
TLP Algorithm
On transmission of new data in Open state:
-> packets_out > 1: schedule PTO in max(2*SRTT, 10ms).
-> packets_out == 1: schedule PTO in max(2*RTT, 1.5*RTT + 200ms)
-> PTO = min(PTO, RTO)
Conditions for scheduling PTO:
-> Connection is in Open state.
-> Connection is either cwnd limited or no new data to send.
-> Number of probes per tail loss episode is limited to one.
-> Connection is SACK enabled.
When PTO fires:
new_segment_exists:
-> transmit new segment.
-> packets_out++. cwnd remains same.
no_new_packet:
-> retransmit the last segment.
Its ACK triggers FACK or early retransmit based recovery.
ACK path:
-> rearm RTO at start of ACK processing.
-> reschedule PTO if need be.
In addition, the patch includes a small variation to the Early Retransmit
(ER) algorithm, such that ER and TLP together can in principle recover any
N-degree of tail loss through fast recovery. TLP is controlled by the same
sysctl as ER, tcp_early_retrans sysctl.
tcp_early_retrans==0; disables TLP and ER.
==1; enables RFC5827 ER.
==2; delayed ER.
==3; TLP and delayed ER. [DEFAULT]
==4; TLP only.
The TLP patch series have been extensively tested on Google Web servers.
It is most effective for short Web trasactions, where it reduced RTOs by 15%
and improved HTTP response time (average by 6%, 99th percentile by 10%).
The transmitted probes account for <0.5% of the overall transmissions.
Signed-off-by: Nandita Dukkipati <nanditad@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Acked-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-03-11 17:00:43 +07:00
|
|
|
if (tp->packets_out == 1)
|
|
|
|
timeout = max_t(u32, timeout,
|
|
|
|
(rtt + (rtt >> 1) + TCP_DELACK_MAX));
|
|
|
|
timeout = max_t(u32, timeout, msecs_to_jiffies(10));
|
|
|
|
|
|
|
|
/* If RTO is shorter, just schedule TLP in its place. */
|
|
|
|
tlp_time_stamp = tcp_time_stamp + timeout;
|
|
|
|
rto_time_stamp = (u32)inet_csk(sk)->icsk_timeout;
|
|
|
|
if ((s32)(tlp_time_stamp - rto_time_stamp) > 0) {
|
|
|
|
s32 delta = rto_time_stamp - tcp_time_stamp;
|
|
|
|
if (delta > 0)
|
|
|
|
timeout = delta;
|
|
|
|
}
|
|
|
|
|
|
|
|
inet_csk_reset_xmit_timer(sk, ICSK_TIME_LOSS_PROBE, timeout,
|
|
|
|
TCP_RTO_MAX);
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
|
2014-04-21 07:58:17 +07:00
|
|
|
/* Thanks to skb fast clones, we can detect if a prior transmit of
|
|
|
|
* a packet is still in a qdisc or driver queue.
|
|
|
|
* In this case, there is very little point doing a retransmit !
|
|
|
|
* Note: This is called from BH context only.
|
|
|
|
*/
|
|
|
|
static bool skb_still_in_host_queue(const struct sock *sk,
|
|
|
|
const struct sk_buff *skb)
|
|
|
|
{
|
2014-10-31 00:32:34 +07:00
|
|
|
if (unlikely(skb_fclone_busy(sk, skb))) {
|
2014-04-21 07:58:17 +07:00
|
|
|
NET_INC_STATS_BH(sock_net(sk),
|
|
|
|
LINUX_MIB_TCPSPURIOUS_RTX_HOSTQUEUES);
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
2015-08-13 01:18:19 +07:00
|
|
|
/* When probe timeout (PTO) fires, try send a new segment if possible, else
|
tcp: Tail loss probe (TLP)
This patch series implement the Tail loss probe (TLP) algorithm described
in http://tools.ietf.org/html/draft-dukkipati-tcpm-tcp-loss-probe-01. The
first patch implements the basic algorithm.
TLP's goal is to reduce tail latency of short transactions. It achieves
this by converting retransmission timeouts (RTOs) occuring due
to tail losses (losses at end of transactions) into fast recovery.
TLP transmits one packet in two round-trips when a connection is in
Open state and isn't receiving any ACKs. The transmitted packet, aka
loss probe, can be either new or a retransmission. When there is tail
loss, the ACK from a loss probe triggers FACK/early-retransmit based
fast recovery, thus avoiding a costly RTO. In the absence of loss,
there is no change in the connection state.
PTO stands for probe timeout. It is a timer event indicating
that an ACK is overdue and triggers a loss probe packet. The PTO value
is set to max(2*SRTT, 10ms) and is adjusted to account for delayed
ACK timer when there is only one oustanding packet.
TLP Algorithm
On transmission of new data in Open state:
-> packets_out > 1: schedule PTO in max(2*SRTT, 10ms).
-> packets_out == 1: schedule PTO in max(2*RTT, 1.5*RTT + 200ms)
-> PTO = min(PTO, RTO)
Conditions for scheduling PTO:
-> Connection is in Open state.
-> Connection is either cwnd limited or no new data to send.
-> Number of probes per tail loss episode is limited to one.
-> Connection is SACK enabled.
When PTO fires:
new_segment_exists:
-> transmit new segment.
-> packets_out++. cwnd remains same.
no_new_packet:
-> retransmit the last segment.
Its ACK triggers FACK or early retransmit based recovery.
ACK path:
-> rearm RTO at start of ACK processing.
-> reschedule PTO if need be.
In addition, the patch includes a small variation to the Early Retransmit
(ER) algorithm, such that ER and TLP together can in principle recover any
N-degree of tail loss through fast recovery. TLP is controlled by the same
sysctl as ER, tcp_early_retrans sysctl.
tcp_early_retrans==0; disables TLP and ER.
==1; enables RFC5827 ER.
==2; delayed ER.
==3; TLP and delayed ER. [DEFAULT]
==4; TLP only.
The TLP patch series have been extensively tested on Google Web servers.
It is most effective for short Web trasactions, where it reduced RTOs by 15%
and improved HTTP response time (average by 6%, 99th percentile by 10%).
The transmitted probes account for <0.5% of the overall transmissions.
Signed-off-by: Nandita Dukkipati <nanditad@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Acked-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-03-11 17:00:43 +07:00
|
|
|
* retransmit the last segment.
|
|
|
|
*/
|
|
|
|
void tcp_send_loss_probe(struct sock *sk)
|
|
|
|
{
|
2013-03-11 17:00:44 +07:00
|
|
|
struct tcp_sock *tp = tcp_sk(sk);
|
tcp: Tail loss probe (TLP)
This patch series implement the Tail loss probe (TLP) algorithm described
in http://tools.ietf.org/html/draft-dukkipati-tcpm-tcp-loss-probe-01. The
first patch implements the basic algorithm.
TLP's goal is to reduce tail latency of short transactions. It achieves
this by converting retransmission timeouts (RTOs) occuring due
to tail losses (losses at end of transactions) into fast recovery.
TLP transmits one packet in two round-trips when a connection is in
Open state and isn't receiving any ACKs. The transmitted packet, aka
loss probe, can be either new or a retransmission. When there is tail
loss, the ACK from a loss probe triggers FACK/early-retransmit based
fast recovery, thus avoiding a costly RTO. In the absence of loss,
there is no change in the connection state.
PTO stands for probe timeout. It is a timer event indicating
that an ACK is overdue and triggers a loss probe packet. The PTO value
is set to max(2*SRTT, 10ms) and is adjusted to account for delayed
ACK timer when there is only one oustanding packet.
TLP Algorithm
On transmission of new data in Open state:
-> packets_out > 1: schedule PTO in max(2*SRTT, 10ms).
-> packets_out == 1: schedule PTO in max(2*RTT, 1.5*RTT + 200ms)
-> PTO = min(PTO, RTO)
Conditions for scheduling PTO:
-> Connection is in Open state.
-> Connection is either cwnd limited or no new data to send.
-> Number of probes per tail loss episode is limited to one.
-> Connection is SACK enabled.
When PTO fires:
new_segment_exists:
-> transmit new segment.
-> packets_out++. cwnd remains same.
no_new_packet:
-> retransmit the last segment.
Its ACK triggers FACK or early retransmit based recovery.
ACK path:
-> rearm RTO at start of ACK processing.
-> reschedule PTO if need be.
In addition, the patch includes a small variation to the Early Retransmit
(ER) algorithm, such that ER and TLP together can in principle recover any
N-degree of tail loss through fast recovery. TLP is controlled by the same
sysctl as ER, tcp_early_retrans sysctl.
tcp_early_retrans==0; disables TLP and ER.
==1; enables RFC5827 ER.
==2; delayed ER.
==3; TLP and delayed ER. [DEFAULT]
==4; TLP only.
The TLP patch series have been extensively tested on Google Web servers.
It is most effective for short Web trasactions, where it reduced RTOs by 15%
and improved HTTP response time (average by 6%, 99th percentile by 10%).
The transmitted probes account for <0.5% of the overall transmissions.
Signed-off-by: Nandita Dukkipati <nanditad@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Acked-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-03-11 17:00:43 +07:00
|
|
|
struct sk_buff *skb;
|
|
|
|
int pcount;
|
|
|
|
int mss = tcp_current_mss(sk);
|
|
|
|
|
2015-08-13 01:18:19 +07:00
|
|
|
skb = tcp_send_head(sk);
|
|
|
|
if (skb) {
|
|
|
|
if (tcp_snd_wnd_test(tp, skb, mss)) {
|
|
|
|
pcount = tp->packets_out;
|
|
|
|
tcp_write_xmit(sk, mss, TCP_NAGLE_OFF, 2, GFP_ATOMIC);
|
|
|
|
if (tp->packets_out > pcount)
|
|
|
|
goto probe_sent;
|
|
|
|
goto rearm_timer;
|
|
|
|
}
|
|
|
|
skb = tcp_write_queue_prev(sk, skb);
|
|
|
|
} else {
|
|
|
|
skb = tcp_write_queue_tail(sk);
|
tcp: Tail loss probe (TLP)
This patch series implement the Tail loss probe (TLP) algorithm described
in http://tools.ietf.org/html/draft-dukkipati-tcpm-tcp-loss-probe-01. The
first patch implements the basic algorithm.
TLP's goal is to reduce tail latency of short transactions. It achieves
this by converting retransmission timeouts (RTOs) occuring due
to tail losses (losses at end of transactions) into fast recovery.
TLP transmits one packet in two round-trips when a connection is in
Open state and isn't receiving any ACKs. The transmitted packet, aka
loss probe, can be either new or a retransmission. When there is tail
loss, the ACK from a loss probe triggers FACK/early-retransmit based
fast recovery, thus avoiding a costly RTO. In the absence of loss,
there is no change in the connection state.
PTO stands for probe timeout. It is a timer event indicating
that an ACK is overdue and triggers a loss probe packet. The PTO value
is set to max(2*SRTT, 10ms) and is adjusted to account for delayed
ACK timer when there is only one oustanding packet.
TLP Algorithm
On transmission of new data in Open state:
-> packets_out > 1: schedule PTO in max(2*SRTT, 10ms).
-> packets_out == 1: schedule PTO in max(2*RTT, 1.5*RTT + 200ms)
-> PTO = min(PTO, RTO)
Conditions for scheduling PTO:
-> Connection is in Open state.
-> Connection is either cwnd limited or no new data to send.
-> Number of probes per tail loss episode is limited to one.
-> Connection is SACK enabled.
When PTO fires:
new_segment_exists:
-> transmit new segment.
-> packets_out++. cwnd remains same.
no_new_packet:
-> retransmit the last segment.
Its ACK triggers FACK or early retransmit based recovery.
ACK path:
-> rearm RTO at start of ACK processing.
-> reschedule PTO if need be.
In addition, the patch includes a small variation to the Early Retransmit
(ER) algorithm, such that ER and TLP together can in principle recover any
N-degree of tail loss through fast recovery. TLP is controlled by the same
sysctl as ER, tcp_early_retrans sysctl.
tcp_early_retrans==0; disables TLP and ER.
==1; enables RFC5827 ER.
==2; delayed ER.
==3; TLP and delayed ER. [DEFAULT]
==4; TLP only.
The TLP patch series have been extensively tested on Google Web servers.
It is most effective for short Web trasactions, where it reduced RTOs by 15%
and improved HTTP response time (average by 6%, 99th percentile by 10%).
The transmitted probes account for <0.5% of the overall transmissions.
Signed-off-by: Nandita Dukkipati <nanditad@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Acked-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-03-11 17:00:43 +07:00
|
|
|
}
|
|
|
|
|
2013-03-11 17:00:44 +07:00
|
|
|
/* At most one outstanding TLP retransmission. */
|
|
|
|
if (tp->tlp_high_seq)
|
|
|
|
goto rearm_timer;
|
|
|
|
|
tcp: Tail loss probe (TLP)
This patch series implement the Tail loss probe (TLP) algorithm described
in http://tools.ietf.org/html/draft-dukkipati-tcpm-tcp-loss-probe-01. The
first patch implements the basic algorithm.
TLP's goal is to reduce tail latency of short transactions. It achieves
this by converting retransmission timeouts (RTOs) occuring due
to tail losses (losses at end of transactions) into fast recovery.
TLP transmits one packet in two round-trips when a connection is in
Open state and isn't receiving any ACKs. The transmitted packet, aka
loss probe, can be either new or a retransmission. When there is tail
loss, the ACK from a loss probe triggers FACK/early-retransmit based
fast recovery, thus avoiding a costly RTO. In the absence of loss,
there is no change in the connection state.
PTO stands for probe timeout. It is a timer event indicating
that an ACK is overdue and triggers a loss probe packet. The PTO value
is set to max(2*SRTT, 10ms) and is adjusted to account for delayed
ACK timer when there is only one oustanding packet.
TLP Algorithm
On transmission of new data in Open state:
-> packets_out > 1: schedule PTO in max(2*SRTT, 10ms).
-> packets_out == 1: schedule PTO in max(2*RTT, 1.5*RTT + 200ms)
-> PTO = min(PTO, RTO)
Conditions for scheduling PTO:
-> Connection is in Open state.
-> Connection is either cwnd limited or no new data to send.
-> Number of probes per tail loss episode is limited to one.
-> Connection is SACK enabled.
When PTO fires:
new_segment_exists:
-> transmit new segment.
-> packets_out++. cwnd remains same.
no_new_packet:
-> retransmit the last segment.
Its ACK triggers FACK or early retransmit based recovery.
ACK path:
-> rearm RTO at start of ACK processing.
-> reschedule PTO if need be.
In addition, the patch includes a small variation to the Early Retransmit
(ER) algorithm, such that ER and TLP together can in principle recover any
N-degree of tail loss through fast recovery. TLP is controlled by the same
sysctl as ER, tcp_early_retrans sysctl.
tcp_early_retrans==0; disables TLP and ER.
==1; enables RFC5827 ER.
==2; delayed ER.
==3; TLP and delayed ER. [DEFAULT]
==4; TLP only.
The TLP patch series have been extensively tested on Google Web servers.
It is most effective for short Web trasactions, where it reduced RTOs by 15%
and improved HTTP response time (average by 6%, 99th percentile by 10%).
The transmitted probes account for <0.5% of the overall transmissions.
Signed-off-by: Nandita Dukkipati <nanditad@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Acked-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-03-11 17:00:43 +07:00
|
|
|
/* Retransmit last segment. */
|
|
|
|
if (WARN_ON(!skb))
|
|
|
|
goto rearm_timer;
|
|
|
|
|
2014-04-21 07:58:17 +07:00
|
|
|
if (skb_still_in_host_queue(sk, skb))
|
|
|
|
goto rearm_timer;
|
|
|
|
|
tcp: Tail loss probe (TLP)
This patch series implement the Tail loss probe (TLP) algorithm described
in http://tools.ietf.org/html/draft-dukkipati-tcpm-tcp-loss-probe-01. The
first patch implements the basic algorithm.
TLP's goal is to reduce tail latency of short transactions. It achieves
this by converting retransmission timeouts (RTOs) occuring due
to tail losses (losses at end of transactions) into fast recovery.
TLP transmits one packet in two round-trips when a connection is in
Open state and isn't receiving any ACKs. The transmitted packet, aka
loss probe, can be either new or a retransmission. When there is tail
loss, the ACK from a loss probe triggers FACK/early-retransmit based
fast recovery, thus avoiding a costly RTO. In the absence of loss,
there is no change in the connection state.
PTO stands for probe timeout. It is a timer event indicating
that an ACK is overdue and triggers a loss probe packet. The PTO value
is set to max(2*SRTT, 10ms) and is adjusted to account for delayed
ACK timer when there is only one oustanding packet.
TLP Algorithm
On transmission of new data in Open state:
-> packets_out > 1: schedule PTO in max(2*SRTT, 10ms).
-> packets_out == 1: schedule PTO in max(2*RTT, 1.5*RTT + 200ms)
-> PTO = min(PTO, RTO)
Conditions for scheduling PTO:
-> Connection is in Open state.
-> Connection is either cwnd limited or no new data to send.
-> Number of probes per tail loss episode is limited to one.
-> Connection is SACK enabled.
When PTO fires:
new_segment_exists:
-> transmit new segment.
-> packets_out++. cwnd remains same.
no_new_packet:
-> retransmit the last segment.
Its ACK triggers FACK or early retransmit based recovery.
ACK path:
-> rearm RTO at start of ACK processing.
-> reschedule PTO if need be.
In addition, the patch includes a small variation to the Early Retransmit
(ER) algorithm, such that ER and TLP together can in principle recover any
N-degree of tail loss through fast recovery. TLP is controlled by the same
sysctl as ER, tcp_early_retrans sysctl.
tcp_early_retrans==0; disables TLP and ER.
==1; enables RFC5827 ER.
==2; delayed ER.
==3; TLP and delayed ER. [DEFAULT]
==4; TLP only.
The TLP patch series have been extensively tested on Google Web servers.
It is most effective for short Web trasactions, where it reduced RTOs by 15%
and improved HTTP response time (average by 6%, 99th percentile by 10%).
The transmitted probes account for <0.5% of the overall transmissions.
Signed-off-by: Nandita Dukkipati <nanditad@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Acked-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-03-11 17:00:43 +07:00
|
|
|
pcount = tcp_skb_pcount(skb);
|
|
|
|
if (WARN_ON(!pcount))
|
|
|
|
goto rearm_timer;
|
|
|
|
|
|
|
|
if ((pcount > 1) && (skb->len > (pcount - 1) * mss)) {
|
2014-06-06 21:32:37 +07:00
|
|
|
if (unlikely(tcp_fragment(sk, skb, (pcount - 1) * mss, mss,
|
|
|
|
GFP_ATOMIC)))
|
tcp: Tail loss probe (TLP)
This patch series implement the Tail loss probe (TLP) algorithm described
in http://tools.ietf.org/html/draft-dukkipati-tcpm-tcp-loss-probe-01. The
first patch implements the basic algorithm.
TLP's goal is to reduce tail latency of short transactions. It achieves
this by converting retransmission timeouts (RTOs) occuring due
to tail losses (losses at end of transactions) into fast recovery.
TLP transmits one packet in two round-trips when a connection is in
Open state and isn't receiving any ACKs. The transmitted packet, aka
loss probe, can be either new or a retransmission. When there is tail
loss, the ACK from a loss probe triggers FACK/early-retransmit based
fast recovery, thus avoiding a costly RTO. In the absence of loss,
there is no change in the connection state.
PTO stands for probe timeout. It is a timer event indicating
that an ACK is overdue and triggers a loss probe packet. The PTO value
is set to max(2*SRTT, 10ms) and is adjusted to account for delayed
ACK timer when there is only one oustanding packet.
TLP Algorithm
On transmission of new data in Open state:
-> packets_out > 1: schedule PTO in max(2*SRTT, 10ms).
-> packets_out == 1: schedule PTO in max(2*RTT, 1.5*RTT + 200ms)
-> PTO = min(PTO, RTO)
Conditions for scheduling PTO:
-> Connection is in Open state.
-> Connection is either cwnd limited or no new data to send.
-> Number of probes per tail loss episode is limited to one.
-> Connection is SACK enabled.
When PTO fires:
new_segment_exists:
-> transmit new segment.
-> packets_out++. cwnd remains same.
no_new_packet:
-> retransmit the last segment.
Its ACK triggers FACK or early retransmit based recovery.
ACK path:
-> rearm RTO at start of ACK processing.
-> reschedule PTO if need be.
In addition, the patch includes a small variation to the Early Retransmit
(ER) algorithm, such that ER and TLP together can in principle recover any
N-degree of tail loss through fast recovery. TLP is controlled by the same
sysctl as ER, tcp_early_retrans sysctl.
tcp_early_retrans==0; disables TLP and ER.
==1; enables RFC5827 ER.
==2; delayed ER.
==3; TLP and delayed ER. [DEFAULT]
==4; TLP only.
The TLP patch series have been extensively tested on Google Web servers.
It is most effective for short Web trasactions, where it reduced RTOs by 15%
and improved HTTP response time (average by 6%, 99th percentile by 10%).
The transmitted probes account for <0.5% of the overall transmissions.
Signed-off-by: Nandita Dukkipati <nanditad@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Acked-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-03-11 17:00:43 +07:00
|
|
|
goto rearm_timer;
|
2015-08-13 01:18:19 +07:00
|
|
|
skb = tcp_write_queue_next(sk, skb);
|
tcp: Tail loss probe (TLP)
This patch series implement the Tail loss probe (TLP) algorithm described
in http://tools.ietf.org/html/draft-dukkipati-tcpm-tcp-loss-probe-01. The
first patch implements the basic algorithm.
TLP's goal is to reduce tail latency of short transactions. It achieves
this by converting retransmission timeouts (RTOs) occuring due
to tail losses (losses at end of transactions) into fast recovery.
TLP transmits one packet in two round-trips when a connection is in
Open state and isn't receiving any ACKs. The transmitted packet, aka
loss probe, can be either new or a retransmission. When there is tail
loss, the ACK from a loss probe triggers FACK/early-retransmit based
fast recovery, thus avoiding a costly RTO. In the absence of loss,
there is no change in the connection state.
PTO stands for probe timeout. It is a timer event indicating
that an ACK is overdue and triggers a loss probe packet. The PTO value
is set to max(2*SRTT, 10ms) and is adjusted to account for delayed
ACK timer when there is only one oustanding packet.
TLP Algorithm
On transmission of new data in Open state:
-> packets_out > 1: schedule PTO in max(2*SRTT, 10ms).
-> packets_out == 1: schedule PTO in max(2*RTT, 1.5*RTT + 200ms)
-> PTO = min(PTO, RTO)
Conditions for scheduling PTO:
-> Connection is in Open state.
-> Connection is either cwnd limited or no new data to send.
-> Number of probes per tail loss episode is limited to one.
-> Connection is SACK enabled.
When PTO fires:
new_segment_exists:
-> transmit new segment.
-> packets_out++. cwnd remains same.
no_new_packet:
-> retransmit the last segment.
Its ACK triggers FACK or early retransmit based recovery.
ACK path:
-> rearm RTO at start of ACK processing.
-> reschedule PTO if need be.
In addition, the patch includes a small variation to the Early Retransmit
(ER) algorithm, such that ER and TLP together can in principle recover any
N-degree of tail loss through fast recovery. TLP is controlled by the same
sysctl as ER, tcp_early_retrans sysctl.
tcp_early_retrans==0; disables TLP and ER.
==1; enables RFC5827 ER.
==2; delayed ER.
==3; TLP and delayed ER. [DEFAULT]
==4; TLP only.
The TLP patch series have been extensively tested on Google Web servers.
It is most effective for short Web trasactions, where it reduced RTOs by 15%
and improved HTTP response time (average by 6%, 99th percentile by 10%).
The transmitted probes account for <0.5% of the overall transmissions.
Signed-off-by: Nandita Dukkipati <nanditad@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Acked-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-03-11 17:00:43 +07:00
|
|
|
}
|
|
|
|
|
|
|
|
if (WARN_ON(!skb || !tcp_skb_pcount(skb)))
|
|
|
|
goto rearm_timer;
|
|
|
|
|
2015-08-13 01:18:19 +07:00
|
|
|
if (__tcp_retransmit_skb(sk, skb))
|
|
|
|
goto rearm_timer;
|
tcp: Tail loss probe (TLP)
This patch series implement the Tail loss probe (TLP) algorithm described
in http://tools.ietf.org/html/draft-dukkipati-tcpm-tcp-loss-probe-01. The
first patch implements the basic algorithm.
TLP's goal is to reduce tail latency of short transactions. It achieves
this by converting retransmission timeouts (RTOs) occuring due
to tail losses (losses at end of transactions) into fast recovery.
TLP transmits one packet in two round-trips when a connection is in
Open state and isn't receiving any ACKs. The transmitted packet, aka
loss probe, can be either new or a retransmission. When there is tail
loss, the ACK from a loss probe triggers FACK/early-retransmit based
fast recovery, thus avoiding a costly RTO. In the absence of loss,
there is no change in the connection state.
PTO stands for probe timeout. It is a timer event indicating
that an ACK is overdue and triggers a loss probe packet. The PTO value
is set to max(2*SRTT, 10ms) and is adjusted to account for delayed
ACK timer when there is only one oustanding packet.
TLP Algorithm
On transmission of new data in Open state:
-> packets_out > 1: schedule PTO in max(2*SRTT, 10ms).
-> packets_out == 1: schedule PTO in max(2*RTT, 1.5*RTT + 200ms)
-> PTO = min(PTO, RTO)
Conditions for scheduling PTO:
-> Connection is in Open state.
-> Connection is either cwnd limited or no new data to send.
-> Number of probes per tail loss episode is limited to one.
-> Connection is SACK enabled.
When PTO fires:
new_segment_exists:
-> transmit new segment.
-> packets_out++. cwnd remains same.
no_new_packet:
-> retransmit the last segment.
Its ACK triggers FACK or early retransmit based recovery.
ACK path:
-> rearm RTO at start of ACK processing.
-> reschedule PTO if need be.
In addition, the patch includes a small variation to the Early Retransmit
(ER) algorithm, such that ER and TLP together can in principle recover any
N-degree of tail loss through fast recovery. TLP is controlled by the same
sysctl as ER, tcp_early_retrans sysctl.
tcp_early_retrans==0; disables TLP and ER.
==1; enables RFC5827 ER.
==2; delayed ER.
==3; TLP and delayed ER. [DEFAULT]
==4; TLP only.
The TLP patch series have been extensively tested on Google Web servers.
It is most effective for short Web trasactions, where it reduced RTOs by 15%
and improved HTTP response time (average by 6%, 99th percentile by 10%).
The transmitted probes account for <0.5% of the overall transmissions.
Signed-off-by: Nandita Dukkipati <nanditad@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Acked-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-03-11 17:00:43 +07:00
|
|
|
|
2013-03-11 17:00:44 +07:00
|
|
|
/* Record snd_nxt for loss detection. */
|
2015-08-13 01:18:19 +07:00
|
|
|
tp->tlp_high_seq = tp->snd_nxt;
|
2013-03-11 17:00:44 +07:00
|
|
|
|
2015-08-13 01:18:19 +07:00
|
|
|
probe_sent:
|
|
|
|
NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPLOSSPROBES);
|
|
|
|
/* Reset s.t. tcp_rearm_rto will restart timer from now */
|
|
|
|
inet_csk(sk)->icsk_pending = 0;
|
tcp: Tail loss probe (TLP)
This patch series implement the Tail loss probe (TLP) algorithm described
in http://tools.ietf.org/html/draft-dukkipati-tcpm-tcp-loss-probe-01. The
first patch implements the basic algorithm.
TLP's goal is to reduce tail latency of short transactions. It achieves
this by converting retransmission timeouts (RTOs) occuring due
to tail losses (losses at end of transactions) into fast recovery.
TLP transmits one packet in two round-trips when a connection is in
Open state and isn't receiving any ACKs. The transmitted packet, aka
loss probe, can be either new or a retransmission. When there is tail
loss, the ACK from a loss probe triggers FACK/early-retransmit based
fast recovery, thus avoiding a costly RTO. In the absence of loss,
there is no change in the connection state.
PTO stands for probe timeout. It is a timer event indicating
that an ACK is overdue and triggers a loss probe packet. The PTO value
is set to max(2*SRTT, 10ms) and is adjusted to account for delayed
ACK timer when there is only one oustanding packet.
TLP Algorithm
On transmission of new data in Open state:
-> packets_out > 1: schedule PTO in max(2*SRTT, 10ms).
-> packets_out == 1: schedule PTO in max(2*RTT, 1.5*RTT + 200ms)
-> PTO = min(PTO, RTO)
Conditions for scheduling PTO:
-> Connection is in Open state.
-> Connection is either cwnd limited or no new data to send.
-> Number of probes per tail loss episode is limited to one.
-> Connection is SACK enabled.
When PTO fires:
new_segment_exists:
-> transmit new segment.
-> packets_out++. cwnd remains same.
no_new_packet:
-> retransmit the last segment.
Its ACK triggers FACK or early retransmit based recovery.
ACK path:
-> rearm RTO at start of ACK processing.
-> reschedule PTO if need be.
In addition, the patch includes a small variation to the Early Retransmit
(ER) algorithm, such that ER and TLP together can in principle recover any
N-degree of tail loss through fast recovery. TLP is controlled by the same
sysctl as ER, tcp_early_retrans sysctl.
tcp_early_retrans==0; disables TLP and ER.
==1; enables RFC5827 ER.
==2; delayed ER.
==3; TLP and delayed ER. [DEFAULT]
==4; TLP only.
The TLP patch series have been extensively tested on Google Web servers.
It is most effective for short Web trasactions, where it reduced RTOs by 15%
and improved HTTP response time (average by 6%, 99th percentile by 10%).
The transmitted probes account for <0.5% of the overall transmissions.
Signed-off-by: Nandita Dukkipati <nanditad@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Acked-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-03-11 17:00:43 +07:00
|
|
|
rearm_timer:
|
2015-08-13 01:18:18 +07:00
|
|
|
tcp_rearm_rto(sk);
|
2005-04-17 05:20:36 +07:00
|
|
|
}
|
|
|
|
|
2005-07-06 05:18:51 +07:00
|
|
|
/* Push out any pending frames which were held back due to
|
|
|
|
* TCP_CORK or attempt at coalescing tiny packets.
|
|
|
|
* The socket must be locked by the caller.
|
|
|
|
*/
|
[TCP]: Sed magic converts func(sk, tp, ...) -> func(sk, ...)
This is (mostly) automated change using magic:
sed -e '/struct sock \*sk/ N' -e '/struct sock \*sk/ N'
-e '/struct sock \*sk/ N' -e '/struct sock \*sk/ N'
-e 's|struct sock \*sk,[\n\t ]*struct tcp_sock \*tp\([^{]*\n{\n\)|
struct sock \*sk\1\tstruct tcp_sock *tp = tcp_sk(sk);\n|g'
-e 's|struct sock \*sk, struct tcp_sock \*tp|
struct sock \*sk|g' -e 's|sk, tp\([^-]\)|sk\1|g'
Fixed four unused variable (tp) warnings that were introduced.
In addition, manually added newlines after local variables and
tweaked function arguments positioning.
$ gcc --version
gcc (GCC) 4.1.1 20060525 (Red Hat 4.1.1-1)
...
$ codiff -fV built-in.o.old built-in.o.new
net/ipv4/route.c:
rt_cache_flush | +14
1 function changed, 14 bytes added
net/ipv4/tcp.c:
tcp_setsockopt | -5
tcp_sendpage | -25
tcp_sendmsg | -16
3 functions changed, 46 bytes removed
net/ipv4/tcp_input.c:
tcp_try_undo_recovery | +3
tcp_try_undo_dsack | +2
tcp_mark_head_lost | -12
tcp_ack | -15
tcp_event_data_recv | -32
tcp_rcv_state_process | -10
tcp_rcv_established | +1
7 functions changed, 6 bytes added, 69 bytes removed, diff: -63
net/ipv4/tcp_output.c:
update_send_head | -9
tcp_transmit_skb | +19
tcp_cwnd_validate | +1
tcp_write_wakeup | -17
__tcp_push_pending_frames | -25
tcp_push_one | -8
tcp_send_fin | -4
7 functions changed, 20 bytes added, 63 bytes removed, diff: -43
built-in.o.new:
18 functions changed, 40 bytes added, 178 bytes removed, diff: -138
Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>
2007-04-21 12:18:02 +07:00
|
|
|
void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
|
|
|
|
int nonagle)
|
2005-07-06 05:18:51 +07:00
|
|
|
{
|
2008-12-06 13:43:56 +07:00
|
|
|
/* If we are closed, the bytes will have to remain here.
|
|
|
|
* In time closedown will finish, we empty the write queue and
|
|
|
|
* all will be happy.
|
|
|
|
*/
|
|
|
|
if (unlikely(sk->sk_state == TCP_CLOSE))
|
|
|
|
return;
|
|
|
|
|
2012-08-01 06:44:14 +07:00
|
|
|
if (tcp_write_xmit(sk, cur_mss, nonagle, 0,
|
2015-11-30 23:57:28 +07:00
|
|
|
sk_gfp_mask(sk, GFP_ATOMIC)))
|
2008-12-06 13:43:56 +07:00
|
|
|
tcp_check_probe_timer(sk);
|
2005-07-06 05:18:51 +07:00
|
|
|
}
|
|
|
|
|
2005-07-06 05:24:38 +07:00
|
|
|
/* Send _single_ skb sitting at the send head. This function requires
|
|
|
|
* true push pending frames to setup probe timer etc.
|
|
|
|
*/
|
|
|
|
void tcp_push_one(struct sock *sk, unsigned int mss_now)
|
|
|
|
{
|
2007-03-08 03:12:44 +07:00
|
|
|
struct sk_buff *skb = tcp_send_head(sk);
|
2005-07-06 05:24:38 +07:00
|
|
|
|
|
|
|
BUG_ON(!skb || skb->len < mss_now);
|
|
|
|
|
2008-12-06 13:48:55 +07:00
|
|
|
tcp_write_xmit(sk, mss_now, TCP_NAGLE_PUSH, 1, sk->sk_allocation);
|
2005-07-06 05:24:38 +07:00
|
|
|
}
|
|
|
|
|
2005-04-17 05:20:36 +07:00
|
|
|
/* This function returns the amount that we can raise the
|
|
|
|
* usable window based on the following constraints
|
2007-02-09 21:24:47 +07:00
|
|
|
*
|
2005-04-17 05:20:36 +07:00
|
|
|
* 1. The window can never be shrunk once it is offered (RFC 793)
|
|
|
|
* 2. We limit memory per socket
|
|
|
|
*
|
|
|
|
* RFC 1122:
|
|
|
|
* "the suggested [SWS] avoidance algorithm for the receiver is to keep
|
|
|
|
* RECV.NEXT + RCV.WIN fixed until:
|
|
|
|
* RCV.BUFF - RCV.USER - RCV.WINDOW >= min(1/2 RCV.BUFF, MSS)"
|
|
|
|
*
|
|
|
|
* i.e. don't raise the right edge of the window until you can raise
|
|
|
|
* it at least MSS bytes.
|
|
|
|
*
|
|
|
|
* Unfortunately, the recommended algorithm breaks header prediction,
|
|
|
|
* since header prediction assumes th->window stays fixed.
|
|
|
|
*
|
|
|
|
* Strictly speaking, keeping th->window fixed violates the receiver
|
|
|
|
* side SWS prevention criteria. The problem is that under this rule
|
|
|
|
* a stream of single byte packets will cause the right side of the
|
|
|
|
* window to always advance by a single byte.
|
2007-02-09 21:24:47 +07:00
|
|
|
*
|
2005-04-17 05:20:36 +07:00
|
|
|
* Of course, if the sender implements sender side SWS prevention
|
|
|
|
* then this will not be a problem.
|
2007-02-09 21:24:47 +07:00
|
|
|
*
|
2005-04-17 05:20:36 +07:00
|
|
|
* BSD seems to make the following compromise:
|
2007-02-09 21:24:47 +07:00
|
|
|
*
|
2005-04-17 05:20:36 +07:00
|
|
|
* If the free space is less than the 1/4 of the maximum
|
|
|
|
* space available and the free space is less than 1/2 mss,
|
|
|
|
* then set the window to 0.
|
|
|
|
* [ Actually, bsd uses MSS and 1/4 of maximal _window_ ]
|
|
|
|
* Otherwise, just prevent the window from shrinking
|
|
|
|
* and from being larger than the largest representable value.
|
|
|
|
*
|
|
|
|
* This prevents incremental opening of the window in the regime
|
|
|
|
* where TCP is limited by the speed of the reader side taking
|
|
|
|
* data out of the TCP receive queue. It does nothing about
|
|
|
|
* those cases where the window is constrained on the sender side
|
|
|
|
* because the pipeline is full.
|
|
|
|
*
|
|
|
|
* BSD also seems to "accidentally" limit itself to windows that are a
|
|
|
|
* multiple of MSS, at least until the free space gets quite small.
|
|
|
|
* This would appear to be a side effect of the mbuf implementation.
|
|
|
|
* Combining these two algorithms results in the observed behavior
|
|
|
|
* of having a fixed window size at almost all times.
|
|
|
|
*
|
|
|
|
* Below we obtain similar behavior by forcing the offered window to
|
|
|
|
* a multiple of the mss when it is feasible to do so.
|
|
|
|
*
|
|
|
|
* Note, we don't "adjust" for TIMESTAMP or SACK option bytes.
|
|
|
|
* Regular options like TIMESTAMP are taken into account.
|
|
|
|
*/
|
|
|
|
u32 __tcp_select_window(struct sock *sk)
|
|
|
|
{
|
2005-08-10 10:10:42 +07:00
|
|
|
struct inet_connection_sock *icsk = inet_csk(sk);
|
2005-04-17 05:20:36 +07:00
|
|
|
struct tcp_sock *tp = tcp_sk(sk);
|
2005-11-11 08:13:47 +07:00
|
|
|
/* MSS for the peer's data. Previous versions used mss_clamp
|
2005-04-17 05:20:36 +07:00
|
|
|
* here. I don't know if the value based on our guesses
|
|
|
|
* of peer's MSS is better for the performance. It's more correct
|
|
|
|
* but may be worse for the performance because of rcv_mss
|
|
|
|
* fluctuations. --SAW 1998/11/1
|
|
|
|
*/
|
2005-08-10 10:10:42 +07:00
|
|
|
int mss = icsk->icsk_ack.rcv_mss;
|
2005-04-17 05:20:36 +07:00
|
|
|
int free_space = tcp_space(sk);
|
tcp: use zero-window when free_space is low
Currently the kernel tries to announce a zero window when free_space
is below the current receiver mss estimate.
When a sender is transmitting small packets and reader consumes data
slowly (or not at all), receiver might be unable to shrink the receive
win because
a) we cannot withdraw already-commited receive window, and,
b) we have to round the current rwin up to a multiple of the wscale
factor, else we would shrink the current window.
This causes the receive buffer to fill up until the rmem limit is hit.
When this happens, we start dropping packets.
Moreover, tcp_clamp_window may continue to grow sk_rcvbuf towards rmem[2]
even if socket is not being read from.
As we cannot avoid the "current_win is rounded up to multiple of mss"
issue [we would violate a) above] at least try to prevent the receive buf
growth towards tcp_rmem[2] limit by attempting to move to zero-window
announcement when free_space becomes less than 1/16 of the current
allowed receive buffer maximum. If tcp_rmem[2] is large, this will
increase our chances to get a zero-window announcement out in time.
Reproducer:
On server:
$ nc -l -p 12345
<suspend it: CTRL-Z>
Client:
#!/usr/bin/env python
import socket
import time
sock = socket.socket()
sock.setsockopt(socket.IPPROTO_TCP, socket.TCP_NODELAY, 1)
sock.connect(("192.168.4.1", 12345));
while True:
sock.send('A' * 23)
time.sleep(0.005)
socket buffer on server-side will grow until tcp_rmem[2] is hit,
at which point the client rexmits data until -EDTIMEOUT:
tcp_data_queue invokes tcp_try_rmem_schedule which will call
tcp_prune_queue which calls tcp_clamp_window(). And that function will
grow sk->sk_rcvbuf up until it eventually hits tcp_rmem[2].
Thanks to Eric Dumazet for running regression tests.
Cc: Neal Cardwell <ncardwell@google.com>
Cc: Yuchung Cheng <ycheng@google.com>
Acked-by: Eric Dumazet <edumazet@google.com>
Tested-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
2014-02-19 18:51:10 +07:00
|
|
|
int allowed_space = tcp_full_space(sk);
|
|
|
|
int full_space = min_t(int, tp->window_clamp, allowed_space);
|
2005-04-17 05:20:36 +07:00
|
|
|
int window;
|
|
|
|
|
|
|
|
if (mss > full_space)
|
2007-02-09 21:24:47 +07:00
|
|
|
mss = full_space;
|
2005-04-17 05:20:36 +07:00
|
|
|
|
2007-12-21 12:48:32 +07:00
|
|
|
if (free_space < (full_space >> 1)) {
|
2005-08-10 10:10:42 +07:00
|
|
|
icsk->icsk_ack.quick = 0;
|
2005-04-17 05:20:36 +07:00
|
|
|
|
2015-05-16 02:39:27 +07:00
|
|
|
if (tcp_under_memory_pressure(sk))
|
2008-01-01 05:57:14 +07:00
|
|
|
tp->rcv_ssthresh = min(tp->rcv_ssthresh,
|
|
|
|
4U * tp->advmss);
|
2005-04-17 05:20:36 +07:00
|
|
|
|
tcp: use zero-window when free_space is low
Currently the kernel tries to announce a zero window when free_space
is below the current receiver mss estimate.
When a sender is transmitting small packets and reader consumes data
slowly (or not at all), receiver might be unable to shrink the receive
win because
a) we cannot withdraw already-commited receive window, and,
b) we have to round the current rwin up to a multiple of the wscale
factor, else we would shrink the current window.
This causes the receive buffer to fill up until the rmem limit is hit.
When this happens, we start dropping packets.
Moreover, tcp_clamp_window may continue to grow sk_rcvbuf towards rmem[2]
even if socket is not being read from.
As we cannot avoid the "current_win is rounded up to multiple of mss"
issue [we would violate a) above] at least try to prevent the receive buf
growth towards tcp_rmem[2] limit by attempting to move to zero-window
announcement when free_space becomes less than 1/16 of the current
allowed receive buffer maximum. If tcp_rmem[2] is large, this will
increase our chances to get a zero-window announcement out in time.
Reproducer:
On server:
$ nc -l -p 12345
<suspend it: CTRL-Z>
Client:
#!/usr/bin/env python
import socket
import time
sock = socket.socket()
sock.setsockopt(socket.IPPROTO_TCP, socket.TCP_NODELAY, 1)
sock.connect(("192.168.4.1", 12345));
while True:
sock.send('A' * 23)
time.sleep(0.005)
socket buffer on server-side will grow until tcp_rmem[2] is hit,
at which point the client rexmits data until -EDTIMEOUT:
tcp_data_queue invokes tcp_try_rmem_schedule which will call
tcp_prune_queue which calls tcp_clamp_window(). And that function will
grow sk->sk_rcvbuf up until it eventually hits tcp_rmem[2].
Thanks to Eric Dumazet for running regression tests.
Cc: Neal Cardwell <ncardwell@google.com>
Cc: Yuchung Cheng <ycheng@google.com>
Acked-by: Eric Dumazet <edumazet@google.com>
Tested-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
2014-02-19 18:51:10 +07:00
|
|
|
/* free_space might become our new window, make sure we don't
|
|
|
|
* increase it due to wscale.
|
|
|
|
*/
|
|
|
|
free_space = round_down(free_space, 1 << tp->rx_opt.rcv_wscale);
|
|
|
|
|
|
|
|
/* if free space is less than mss estimate, or is below 1/16th
|
|
|
|
* of the maximum allowed, try to move to zero-window, else
|
|
|
|
* tcp_clamp_window() will grow rcv buf up to tcp_rmem[2], and
|
|
|
|
* new incoming data is dropped due to memory limits.
|
|
|
|
* With large window, mss test triggers way too late in order
|
|
|
|
* to announce zero window in time before rmem limit kicks in.
|
|
|
|
*/
|
|
|
|
if (free_space < (allowed_space >> 4) || free_space < mss)
|
2005-04-17 05:20:36 +07:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (free_space > tp->rcv_ssthresh)
|
|
|
|
free_space = tp->rcv_ssthresh;
|
|
|
|
|
|
|
|
/* Don't do rounding if we are using window scaling, since the
|
|
|
|
* scaled window will not line up with the MSS boundary anyway.
|
|
|
|
*/
|
|
|
|
window = tp->rcv_wnd;
|
|
|
|
if (tp->rx_opt.rcv_wscale) {
|
|
|
|
window = free_space;
|
|
|
|
|
|
|
|
/* Advertise enough space so that it won't get scaled away.
|
|
|
|
* Import case: prevent zero window announcement if
|
|
|
|
* 1<<rcv_wscale > mss.
|
|
|
|
*/
|
|
|
|
if (((window >> tp->rx_opt.rcv_wscale) << tp->rx_opt.rcv_wscale) != window)
|
|
|
|
window = (((window >> tp->rx_opt.rcv_wscale) + 1)
|
|
|
|
<< tp->rx_opt.rcv_wscale);
|
|
|
|
} else {
|
|
|
|
/* Get the largest window that is a nice multiple of mss.
|
|
|
|
* Window clamp already applied above.
|
|
|
|
* If our current window offering is within 1 mss of the
|
|
|
|
* free space we just keep it. This prevents the divide
|
|
|
|
* and multiply from happening most of the time.
|
|
|
|
* We also don't do any window rounding when the free space
|
|
|
|
* is too small.
|
|
|
|
*/
|
|
|
|
if (window <= free_space - mss || window > free_space)
|
2008-01-01 05:57:14 +07:00
|
|
|
window = (free_space / mss) * mss;
|
2007-04-03 03:56:32 +07:00
|
|
|
else if (mss == full_space &&
|
2007-12-21 12:48:32 +07:00
|
|
|
free_space > window + (full_space >> 1))
|
2007-04-03 03:56:32 +07:00
|
|
|
window = free_space;
|
2005-04-17 05:20:36 +07:00
|
|
|
}
|
|
|
|
|
|
|
|
return window;
|
|
|
|
}
|
|
|
|
|
2008-11-25 12:03:43 +07:00
|
|
|
/* Collapses two adjacent SKB's during retransmission. */
|
|
|
|
static void tcp_collapse_retrans(struct sock *sk, struct sk_buff *skb)
|
2005-04-17 05:20:36 +07:00
|
|
|
{
|
|
|
|
struct tcp_sock *tp = tcp_sk(sk);
|
2007-03-08 03:12:44 +07:00
|
|
|
struct sk_buff *next_skb = tcp_write_queue_next(sk, skb);
|
2007-12-31 19:51:11 +07:00
|
|
|
int skb_size, next_skb_size;
|
2005-04-17 05:20:36 +07:00
|
|
|
|
2007-12-31 19:51:11 +07:00
|
|
|
skb_size = skb->len;
|
|
|
|
next_skb_size = next_skb->len;
|
2005-04-17 05:20:36 +07:00
|
|
|
|
2007-12-31 19:51:11 +07:00
|
|
|
BUG_ON(tcp_skb_pcount(skb) != 1 || tcp_skb_pcount(next_skb) != 1);
|
2007-09-26 12:44:14 +07:00
|
|
|
|
2007-12-31 19:51:11 +07:00
|
|
|
tcp_highest_sack_combine(sk, next_skb, skb);
|
2005-04-17 05:20:36 +07:00
|
|
|
|
2007-12-31 19:51:11 +07:00
|
|
|
tcp_unlink_write_queue(next_skb, sk);
|
2005-04-17 05:20:36 +07:00
|
|
|
|
2007-12-31 19:51:11 +07:00
|
|
|
skb_copy_from_linear_data(next_skb, skb_put(skb, next_skb_size),
|
|
|
|
next_skb_size);
|
2005-04-17 05:20:36 +07:00
|
|
|
|
2007-12-31 19:51:11 +07:00
|
|
|
if (next_skb->ip_summed == CHECKSUM_PARTIAL)
|
|
|
|
skb->ip_summed = CHECKSUM_PARTIAL;
|
2005-04-17 05:20:36 +07:00
|
|
|
|
2007-12-31 19:51:11 +07:00
|
|
|
if (skb->ip_summed != CHECKSUM_PARTIAL)
|
|
|
|
skb->csum = csum_block_add(skb->csum, next_skb->csum, skb_size);
|
2005-04-17 05:20:36 +07:00
|
|
|
|
2007-12-31 19:51:11 +07:00
|
|
|
/* Update sequence range on original skb. */
|
|
|
|
TCP_SKB_CB(skb)->end_seq = TCP_SKB_CB(next_skb)->end_seq;
|
2005-04-17 05:20:36 +07:00
|
|
|
|
2009-02-28 11:44:35 +07:00
|
|
|
/* Merge over control information. This moves PSH/FIN etc. over */
|
2011-09-28 00:25:05 +07:00
|
|
|
TCP_SKB_CB(skb)->tcp_flags |= TCP_SKB_CB(next_skb)->tcp_flags;
|
2007-12-31 19:51:11 +07:00
|
|
|
|
|
|
|
/* All done, get rid of second SKB and account for it so
|
|
|
|
* packet counting does not break.
|
|
|
|
*/
|
|
|
|
TCP_SKB_CB(skb)->sacked |= TCP_SKB_CB(next_skb)->sacked & TCPCB_EVER_RETRANS;
|
|
|
|
|
|
|
|
/* changed transmit queue under us so clear hints */
|
2008-09-21 11:25:15 +07:00
|
|
|
tcp_clear_retrans_hints_partial(tp);
|
|
|
|
if (next_skb == tp->retransmit_skb_hint)
|
|
|
|
tp->retransmit_skb_hint = skb;
|
2007-12-31 19:51:11 +07:00
|
|
|
|
2009-04-02 06:15:17 +07:00
|
|
|
tcp_adjust_pcount(sk, next_skb, tcp_skb_pcount(next_skb));
|
|
|
|
|
2007-12-31 19:51:11 +07:00
|
|
|
sk_wmem_free_skb(sk, next_skb);
|
2005-04-17 05:20:36 +07:00
|
|
|
}
|
|
|
|
|
2009-07-22 06:00:40 +07:00
|
|
|
/* Check if coalescing SKBs is legal. */
|
2012-05-17 06:15:34 +07:00
|
|
|
static bool tcp_can_collapse(const struct sock *sk, const struct sk_buff *skb)
|
2008-11-25 12:03:43 +07:00
|
|
|
{
|
|
|
|
if (tcp_skb_pcount(skb) > 1)
|
2012-05-17 06:15:34 +07:00
|
|
|
return false;
|
2008-11-25 12:03:43 +07:00
|
|
|
/* TODO: SACK collapsing could be used to remove this condition */
|
|
|
|
if (skb_shinfo(skb)->nr_frags != 0)
|
2012-05-17 06:15:34 +07:00
|
|
|
return false;
|
2008-11-25 12:03:43 +07:00
|
|
|
if (skb_cloned(skb))
|
2012-05-17 06:15:34 +07:00
|
|
|
return false;
|
2008-11-25 12:03:43 +07:00
|
|
|
if (skb == tcp_send_head(sk))
|
2012-05-17 06:15:34 +07:00
|
|
|
return false;
|
2008-11-25 12:03:43 +07:00
|
|
|
/* Some heurestics for collapsing over SACK'd could be invented */
|
|
|
|
if (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_ACKED)
|
2012-05-17 06:15:34 +07:00
|
|
|
return false;
|
2008-11-25 12:03:43 +07:00
|
|
|
|
2012-05-17 06:15:34 +07:00
|
|
|
return true;
|
2008-11-25 12:03:43 +07:00
|
|
|
}
|
|
|
|
|
2009-07-22 06:00:40 +07:00
|
|
|
/* Collapse packets in the retransmit queue to make to create
|
|
|
|
* less packets on the wire. This is only done on retransmission.
|
|
|
|
*/
|
2008-11-25 12:03:43 +07:00
|
|
|
static void tcp_retrans_try_collapse(struct sock *sk, struct sk_buff *to,
|
|
|
|
int space)
|
|
|
|
{
|
|
|
|
struct tcp_sock *tp = tcp_sk(sk);
|
|
|
|
struct sk_buff *skb = to, *tmp;
|
2012-05-17 06:15:34 +07:00
|
|
|
bool first = true;
|
2008-11-25 12:03:43 +07:00
|
|
|
|
|
|
|
if (!sysctl_tcp_retrans_collapse)
|
|
|
|
return;
|
2011-09-28 00:25:05 +07:00
|
|
|
if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)
|
2008-11-25 12:03:43 +07:00
|
|
|
return;
|
|
|
|
|
|
|
|
tcp_for_write_queue_from_safe(skb, tmp, sk) {
|
|
|
|
if (!tcp_can_collapse(sk, skb))
|
|
|
|
break;
|
|
|
|
|
|
|
|
space -= skb->len;
|
|
|
|
|
|
|
|
if (first) {
|
2012-05-17 06:15:34 +07:00
|
|
|
first = false;
|
2008-11-25 12:03:43 +07:00
|
|
|
continue;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (space < 0)
|
|
|
|
break;
|
|
|
|
/* Punt if not enough space exists in the first SKB for
|
|
|
|
* the data in the second
|
|
|
|
*/
|
2012-04-11 03:30:48 +07:00
|
|
|
if (skb->len > skb_availroom(to))
|
2008-11-25 12:03:43 +07:00
|
|
|
break;
|
|
|
|
|
|
|
|
if (after(TCP_SKB_CB(skb)->end_seq, tcp_wnd_end(tp)))
|
|
|
|
break;
|
|
|
|
|
|
|
|
tcp_collapse_retrans(sk, to);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2005-04-17 05:20:36 +07:00
|
|
|
/* This retransmits one SKB. Policy decisions and retransmit queue
|
|
|
|
* state updates are done by the caller. Returns non-zero if an
|
|
|
|
* error occurred which prevented the send.
|
|
|
|
*/
|
2012-12-06 15:45:32 +07:00
|
|
|
int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb)
|
2005-04-17 05:20:36 +07:00
|
|
|
{
|
|
|
|
struct tcp_sock *tp = tcp_sk(sk);
|
2006-03-21 08:53:41 +07:00
|
|
|
struct inet_connection_sock *icsk = inet_csk(sk);
|
2008-05-22 06:42:20 +07:00
|
|
|
unsigned int cur_mss;
|
2014-03-01 07:42:26 +07:00
|
|
|
int err;
|
2005-04-17 05:20:36 +07:00
|
|
|
|
2006-03-21 08:53:41 +07:00
|
|
|
/* Inconslusive MTU probe */
|
|
|
|
if (icsk->icsk_mtup.probe_size) {
|
|
|
|
icsk->icsk_mtup.probe_size = 0;
|
|
|
|
}
|
|
|
|
|
2005-04-17 05:20:36 +07:00
|
|
|
/* Do not sent more than we queued. 1/4 is reserved for possible
|
2005-11-11 08:13:47 +07:00
|
|
|
* copying overhead: fragmentation, tunneling, mangling etc.
|
2005-04-17 05:20:36 +07:00
|
|
|
*/
|
|
|
|
if (atomic_read(&sk->sk_wmem_alloc) >
|
|
|
|
min(sk->sk_wmem_queued + (sk->sk_wmem_queued >> 2), sk->sk_sndbuf))
|
|
|
|
return -EAGAIN;
|
|
|
|
|
2014-04-21 07:58:17 +07:00
|
|
|
if (skb_still_in_host_queue(sk, skb))
|
|
|
|
return -EBUSY;
|
|
|
|
|
2005-04-17 05:20:36 +07:00
|
|
|
if (before(TCP_SKB_CB(skb)->seq, tp->snd_una)) {
|
|
|
|
if (before(TCP_SKB_CB(skb)->end_seq, tp->snd_una))
|
|
|
|
BUG();
|
|
|
|
if (tcp_trim_head(sk, skb, tp->snd_una - TCP_SKB_CB(skb)->seq))
|
|
|
|
return -ENOMEM;
|
|
|
|
}
|
|
|
|
|
2008-05-22 06:42:20 +07:00
|
|
|
if (inet_csk(sk)->icsk_af_ops->rebuild_header(sk))
|
|
|
|
return -EHOSTUNREACH; /* Routing failure or similar. */
|
|
|
|
|
2009-03-14 21:23:05 +07:00
|
|
|
cur_mss = tcp_current_mss(sk);
|
2008-05-22 06:42:20 +07:00
|
|
|
|
2005-04-17 05:20:36 +07:00
|
|
|
/* If receiver has shrunk his window, and skb is out of
|
|
|
|
* new window, do not retransmit it. The exception is the
|
|
|
|
* case, when window is shrunk to zero. In this case
|
|
|
|
* our retransmit serves as a zero window probe.
|
|
|
|
*/
|
2009-11-24 01:41:23 +07:00
|
|
|
if (!before(TCP_SKB_CB(skb)->seq, tcp_wnd_end(tp)) &&
|
|
|
|
TCP_SKB_CB(skb)->seq != tp->snd_una)
|
2005-04-17 05:20:36 +07:00
|
|
|
return -EAGAIN;
|
|
|
|
|
|
|
|
if (skb->len > cur_mss) {
|
2014-06-06 21:32:37 +07:00
|
|
|
if (tcp_fragment(sk, skb, cur_mss, cur_mss, GFP_ATOMIC))
|
2005-04-17 05:20:36 +07:00
|
|
|
return -ENOMEM; /* We'll try again later. */
|
2009-02-28 11:44:31 +07:00
|
|
|
} else {
|
2009-04-02 06:18:20 +07:00
|
|
|
int oldpcount = tcp_skb_pcount(skb);
|
|
|
|
|
|
|
|
if (unlikely(oldpcount > 1)) {
|
2013-10-16 01:54:30 +07:00
|
|
|
if (skb_unclone(skb, GFP_ATOMIC))
|
|
|
|
return -ENOMEM;
|
2015-06-11 23:15:17 +07:00
|
|
|
tcp_init_tso_segs(skb, cur_mss);
|
2009-04-02 06:18:20 +07:00
|
|
|
tcp_adjust_pcount(sk, skb, oldpcount - tcp_skb_pcount(skb));
|
|
|
|
}
|
2005-04-17 05:20:36 +07:00
|
|
|
}
|
|
|
|
|
tcp: add rfc3168, section 6.1.1.1. fallback
This work as a follow-up of commit f7b3bec6f516 ("net: allow setting ecn
via routing table") and adds RFC3168 section 6.1.1.1. fallback for outgoing
ECN connections. In other words, this work adds a retry with a non-ECN
setup SYN packet, as suggested from the RFC on the first timeout:
[...] A host that receives no reply to an ECN-setup SYN within the
normal SYN retransmission timeout interval MAY resend the SYN and
any subsequent SYN retransmissions with CWR and ECE cleared. [...]
Schematic client-side view when assuming the server is in tcp_ecn=2 mode,
that is, Linux default since 2009 via commit 255cac91c3c9 ("tcp: extend
ECN sysctl to allow server-side only ECN"):
1) Normal ECN-capable path:
SYN ECE CWR ----->
<----- SYN ACK ECE
ACK ----->
2) Path with broken middlebox, when client has fallback:
SYN ECE CWR ----X crappy middlebox drops packet
(timeout, rtx)
SYN ----->
<----- SYN ACK
ACK ----->
In case we would not have the fallback implemented, the middlebox drop
point would basically end up as:
SYN ECE CWR ----X crappy middlebox drops packet
(timeout, rtx)
SYN ECE CWR ----X crappy middlebox drops packet
(timeout, rtx)
SYN ECE CWR ----X crappy middlebox drops packet
(timeout, rtx)
In any case, it's rather a smaller percentage of sites where there would
occur such additional setup latency: it was found in end of 2014 that ~56%
of IPv4 and 65% of IPv6 servers of Alexa 1 million list would negotiate
ECN (aka tcp_ecn=2 default), 0.42% of these webservers will fail to connect
when trying to negotiate with ECN (tcp_ecn=1) due to timeouts, which the
fallback would mitigate with a slight latency trade-off. Recent related
paper on this topic:
Brian Trammell, Mirja Kühlewind, Damiano Boppart, Iain Learmonth,
Gorry Fairhurst, and Richard Scheffenegger:
"Enabling Internet-Wide Deployment of Explicit Congestion Notification."
Proc. PAM 2015, New York.
http://ecn.ethz.ch/ecn-pam15.pdf
Thus, when net.ipv4.tcp_ecn=1 is being set, the patch will perform RFC3168,
section 6.1.1.1. fallback on timeout. For users explicitly not wanting this
which can be in DC use case, we add a net.ipv4.tcp_ecn_fallback knob that
allows for disabling the fallback.
tp->ecn_flags are not being cleared in tcp_ecn_clear_syn() on output, but
rather we let tcp_ecn_rcv_synack() take that over on input path in case a
SYN ACK ECE was delayed. Thus a spurious SYN retransmission will not prevent
ECN being negotiated eventually in that case.
Reference: https://www.ietf.org/proceedings/92/slides/slides-92-iccrg-1.pdf
Reference: https://www.ietf.org/proceedings/89/slides/slides-89-tsvarea-1.pdf
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Mirja Kühlewind <mirja.kuehlewind@tik.ee.ethz.ch>
Signed-off-by: Brian Trammell <trammell@tik.ee.ethz.ch>
Cc: Eric Dumazet <edumazet@google.com>
Cc: Dave That <dave.taht@gmail.com>
Acked-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2015-05-20 02:04:22 +07:00
|
|
|
/* RFC3168, section 6.1.1.1. ECN fallback */
|
|
|
|
if ((TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN_ECN) == TCPHDR_SYN_ECN)
|
|
|
|
tcp_ecn_clear_syn(sk, skb);
|
|
|
|
|
2008-11-25 12:03:43 +07:00
|
|
|
tcp_retrans_try_collapse(sk, skb, cur_mss);
|
2005-04-17 05:20:36 +07:00
|
|
|
|
|
|
|
/* Make a copy, if the first transmission SKB clone we made
|
|
|
|
* is still in somebody's hands, else make a clone.
|
|
|
|
*/
|
|
|
|
|
2013-04-11 17:57:18 +07:00
|
|
|
/* make sure skb->data is aligned on arches that require it
|
|
|
|
* and check if ack-trimming & collapsing extended the headroom
|
|
|
|
* beyond what csum_start can cover.
|
|
|
|
*/
|
|
|
|
if (unlikely((NET_IP_ALIGN && ((unsigned long)skb->data & 3)) ||
|
|
|
|
skb_headroom(skb) >= 0xFFFF)) {
|
2011-12-04 04:39:53 +07:00
|
|
|
struct sk_buff *nskb = __pskb_copy(skb, MAX_TCP_HEADER,
|
|
|
|
GFP_ATOMIC);
|
2014-03-01 07:42:26 +07:00
|
|
|
err = nskb ? tcp_transmit_skb(sk, nskb, 0, GFP_ATOMIC) :
|
|
|
|
-ENOBUFS;
|
2011-12-04 04:39:53 +07:00
|
|
|
} else {
|
2014-03-01 07:42:26 +07:00
|
|
|
err = tcp_transmit_skb(sk, skb, 1, GFP_ATOMIC);
|
2011-12-04 04:39:53 +07:00
|
|
|
}
|
2014-03-01 07:42:26 +07:00
|
|
|
|
2014-04-29 12:00:29 +07:00
|
|
|
if (likely(!err)) {
|
2014-03-01 07:42:26 +07:00
|
|
|
TCP_SKB_CB(skb)->sacked |= TCPCB_EVER_RETRANS;
|
2014-04-29 12:00:29 +07:00
|
|
|
/* Update global TCP statistics. */
|
|
|
|
TCP_INC_STATS(sock_net(sk), TCP_MIB_RETRANSSEGS);
|
|
|
|
if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)
|
|
|
|
NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPSYNRETRANS);
|
|
|
|
tp->total_retrans++;
|
|
|
|
}
|
2014-03-01 07:42:26 +07:00
|
|
|
return err;
|
2012-12-06 15:45:32 +07:00
|
|
|
}
|
|
|
|
|
|
|
|
int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb)
|
|
|
|
{
|
|
|
|
struct tcp_sock *tp = tcp_sk(sk);
|
|
|
|
int err = __tcp_retransmit_skb(sk, skb);
|
2005-04-17 05:20:36 +07:00
|
|
|
|
|
|
|
if (err == 0) {
|
|
|
|
#if FASTRETRANS_DEBUG > 0
|
2008-01-01 05:57:14 +07:00
|
|
|
if (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_RETRANS) {
|
2012-05-14 04:56:26 +07:00
|
|
|
net_dbg_ratelimited("retrans_out leaked\n");
|
2005-04-17 05:20:36 +07:00
|
|
|
}
|
|
|
|
#endif
|
|
|
|
TCP_SKB_CB(skb)->sacked |= TCPCB_RETRANS;
|
|
|
|
tp->retrans_out += tcp_skb_pcount(skb);
|
|
|
|
|
|
|
|
/* Save stamp of the first retransmit. */
|
|
|
|
if (!tp->retrans_stamp)
|
2014-09-06 05:33:33 +07:00
|
|
|
tp->retrans_stamp = tcp_skb_timestamp(skb);
|
2005-04-17 05:20:36 +07:00
|
|
|
|
2014-04-21 07:58:17 +07:00
|
|
|
} else if (err != -EBUSY) {
|
2013-07-13 01:33:04 +07:00
|
|
|
NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPRETRANSFAIL);
|
2005-04-17 05:20:36 +07:00
|
|
|
}
|
2014-07-03 02:07:16 +07:00
|
|
|
|
|
|
|
if (tp->undo_retrans < 0)
|
|
|
|
tp->undo_retrans = 0;
|
|
|
|
tp->undo_retrans += tcp_skb_pcount(skb);
|
2005-04-17 05:20:36 +07:00
|
|
|
return err;
|
|
|
|
}
|
|
|
|
|
2009-07-22 06:00:40 +07:00
|
|
|
/* Check if we forward retransmits are possible in the current
|
|
|
|
* window/congestion state.
|
|
|
|
*/
|
2012-05-17 06:15:34 +07:00
|
|
|
static bool tcp_can_forward_retransmit(struct sock *sk)
|
2008-09-21 11:21:54 +07:00
|
|
|
{
|
|
|
|
const struct inet_connection_sock *icsk = inet_csk(sk);
|
2011-10-21 16:22:42 +07:00
|
|
|
const struct tcp_sock *tp = tcp_sk(sk);
|
2008-09-21 11:21:54 +07:00
|
|
|
|
|
|
|
/* Forward retransmissions are possible only during Recovery. */
|
|
|
|
if (icsk->icsk_ca_state != TCP_CA_Recovery)
|
2012-05-17 06:15:34 +07:00
|
|
|
return false;
|
2008-09-21 11:21:54 +07:00
|
|
|
|
|
|
|
/* No forward retransmissions in Reno are possible. */
|
|
|
|
if (tcp_is_reno(tp))
|
2012-05-17 06:15:34 +07:00
|
|
|
return false;
|
2008-09-21 11:21:54 +07:00
|
|
|
|
|
|
|
/* Yeah, we have to make difficult choice between forward transmission
|
|
|
|
* and retransmission... Both ways have their merits...
|
|
|
|
*
|
|
|
|
* For now we do not retransmit anything, while we have some new
|
|
|
|
* segments to send. In the other cases, follow rule 3 for
|
|
|
|
* NextSeg() specified in RFC3517.
|
|
|
|
*/
|
|
|
|
|
|
|
|
if (tcp_may_send_now(sk))
|
2012-05-17 06:15:34 +07:00
|
|
|
return false;
|
2008-09-21 11:21:54 +07:00
|
|
|
|
2012-05-17 06:15:34 +07:00
|
|
|
return true;
|
2008-09-21 11:21:54 +07:00
|
|
|
}
|
|
|
|
|
2005-04-17 05:20:36 +07:00
|
|
|
/* This gets called after a retransmit timeout, and the initially
|
|
|
|
* retransmitted data is acknowledged. It tries to continue
|
|
|
|
* resending the rest of the retransmit queue, until either
|
|
|
|
* we've sent it all or the congestion window limit is reached.
|
|
|
|
* If doing SACK, the first ACK which comes back for a timeout
|
|
|
|
* based retransmit packet might feed us FACK information again.
|
|
|
|
* If so, we use it to avoid unnecessarily retransmissions.
|
|
|
|
*/
|
|
|
|
void tcp_xmit_retransmit_queue(struct sock *sk)
|
|
|
|
{
|
2005-08-10 14:03:31 +07:00
|
|
|
const struct inet_connection_sock *icsk = inet_csk(sk);
|
2005-04-17 05:20:36 +07:00
|
|
|
struct tcp_sock *tp = tcp_sk(sk);
|
|
|
|
struct sk_buff *skb;
|
2008-09-21 11:24:21 +07:00
|
|
|
struct sk_buff *hole = NULL;
|
2008-09-21 11:26:22 +07:00
|
|
|
u32 last_lost;
|
2008-09-21 11:22:59 +07:00
|
|
|
int mib_idx;
|
2008-09-21 11:24:21 +07:00
|
|
|
int fwd_rexmitting = 0;
|
2005-11-11 08:14:59 +07:00
|
|
|
|
2010-07-19 08:16:18 +07:00
|
|
|
if (!tp->packets_out)
|
|
|
|
return;
|
|
|
|
|
2008-09-21 11:23:49 +07:00
|
|
|
if (!tp->lost_out)
|
|
|
|
tp->retransmit_high = tp->snd_una;
|
|
|
|
|
2008-09-21 11:26:22 +07:00
|
|
|
if (tp->retransmit_skb_hint) {
|
2005-11-11 08:14:59 +07:00
|
|
|
skb = tp->retransmit_skb_hint;
|
2008-09-21 11:26:22 +07:00
|
|
|
last_lost = TCP_SKB_CB(skb)->end_seq;
|
|
|
|
if (after(last_lost, tp->retransmit_high))
|
|
|
|
last_lost = tp->retransmit_high;
|
|
|
|
} else {
|
2007-03-08 03:12:44 +07:00
|
|
|
skb = tcp_write_queue_head(sk);
|
2008-09-21 11:26:22 +07:00
|
|
|
last_lost = tp->snd_una;
|
|
|
|
}
|
2005-04-17 05:20:36 +07:00
|
|
|
|
2008-09-21 11:23:49 +07:00
|
|
|
tcp_for_write_queue_from(skb, sk) {
|
|
|
|
__u8 sacked = TCP_SKB_CB(skb)->sacked;
|
2005-04-17 05:20:36 +07:00
|
|
|
|
2008-09-21 11:23:49 +07:00
|
|
|
if (skb == tcp_send_head(sk))
|
|
|
|
break;
|
|
|
|
/* we could do better than to assign each time */
|
2015-04-03 15:17:26 +07:00
|
|
|
if (!hole)
|
2008-09-21 11:24:21 +07:00
|
|
|
tp->retransmit_skb_hint = skb;
|
2008-09-21 11:23:49 +07:00
|
|
|
|
|
|
|
/* Assume this retransmit will generate
|
|
|
|
* only one packet for congestion window
|
|
|
|
* calculation purposes. This works because
|
|
|
|
* tcp_retransmit_skb() will chop up the
|
|
|
|
* packet to be MSS sized and all the
|
|
|
|
* packet counting works out.
|
|
|
|
*/
|
|
|
|
if (tcp_packets_in_flight(tp) >= tp->snd_cwnd)
|
|
|
|
return;
|
2005-04-17 05:20:36 +07:00
|
|
|
|
2008-09-21 11:24:21 +07:00
|
|
|
if (fwd_rexmitting) {
|
|
|
|
begin_fwd:
|
|
|
|
if (!before(TCP_SKB_CB(skb)->seq, tcp_highest_sack_seq(tp)))
|
|
|
|
break;
|
|
|
|
mib_idx = LINUX_MIB_TCPFORWARDRETRANS;
|
2005-11-11 08:14:59 +07:00
|
|
|
|
2008-09-21 11:24:21 +07:00
|
|
|
} else if (!before(TCP_SKB_CB(skb)->seq, tp->retransmit_high)) {
|
2008-09-21 11:26:22 +07:00
|
|
|
tp->retransmit_high = last_lost;
|
2008-09-21 11:24:21 +07:00
|
|
|
if (!tcp_can_forward_retransmit(sk))
|
|
|
|
break;
|
|
|
|
/* Backtrack if necessary to non-L'ed skb */
|
2015-04-03 15:17:27 +07:00
|
|
|
if (hole) {
|
2008-09-21 11:24:21 +07:00
|
|
|
skb = hole;
|
|
|
|
hole = NULL;
|
|
|
|
}
|
|
|
|
fwd_rexmitting = 1;
|
|
|
|
goto begin_fwd;
|
2005-04-17 05:20:36 +07:00
|
|
|
|
2008-09-21 11:24:21 +07:00
|
|
|
} else if (!(sacked & TCPCB_LOST)) {
|
2015-04-03 15:17:26 +07:00
|
|
|
if (!hole && !(sacked & (TCPCB_SACKED_RETRANS|TCPCB_SACKED_ACKED)))
|
2008-09-21 11:24:21 +07:00
|
|
|
hole = skb;
|
|
|
|
continue;
|
2005-04-17 05:20:36 +07:00
|
|
|
|
2008-09-21 11:24:21 +07:00
|
|
|
} else {
|
2008-09-21 11:26:22 +07:00
|
|
|
last_lost = TCP_SKB_CB(skb)->end_seq;
|
2008-09-21 11:24:21 +07:00
|
|
|
if (icsk->icsk_ca_state != TCP_CA_Loss)
|
|
|
|
mib_idx = LINUX_MIB_TCPFASTRETRANS;
|
|
|
|
else
|
|
|
|
mib_idx = LINUX_MIB_TCPSLOWSTARTRETRANS;
|
|
|
|
}
|
2005-04-17 05:20:36 +07:00
|
|
|
|
2008-09-21 11:24:21 +07:00
|
|
|
if (sacked & (TCPCB_SACKED_ACKED|TCPCB_SACKED_RETRANS))
|
2005-04-17 05:20:36 +07:00
|
|
|
continue;
|
|
|
|
|
2013-07-13 01:33:04 +07:00
|
|
|
if (tcp_retransmit_skb(sk, skb))
|
2008-09-21 11:24:21 +07:00
|
|
|
return;
|
2013-07-13 01:33:04 +07:00
|
|
|
|
2008-09-21 11:24:21 +07:00
|
|
|
NET_INC_STATS_BH(sock_net(sk), mib_idx);
|
2005-04-17 05:20:36 +07:00
|
|
|
|
2012-09-03 00:38:04 +07:00
|
|
|
if (tcp_in_cwnd_reduction(sk))
|
2011-08-22 03:21:57 +07:00
|
|
|
tp->prr_out += tcp_skb_pcount(skb);
|
|
|
|
|
2007-03-08 03:12:44 +07:00
|
|
|
if (skb == tcp_write_queue_head(sk))
|
2005-08-10 10:11:08 +07:00
|
|
|
inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
|
|
|
|
inet_csk(sk)->icsk_rto,
|
|
|
|
TCP_RTO_MAX);
|
2005-04-17 05:20:36 +07:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2015-04-22 08:32:24 +07:00
|
|
|
/* We allow to exceed memory limits for FIN packets to expedite
|
|
|
|
* connection tear down and (memory) recovery.
|
2015-04-24 00:42:39 +07:00
|
|
|
* Otherwise tcp_send_fin() could be tempted to either delay FIN
|
|
|
|
* or even be forced to close flow without any FIN.
|
2015-05-16 02:39:26 +07:00
|
|
|
* In general, we want to allow one skb per socket to avoid hangs
|
|
|
|
* with edge trigger epoll()
|
2015-04-22 08:32:24 +07:00
|
|
|
*/
|
2015-05-16 02:39:26 +07:00
|
|
|
void sk_forced_mem_schedule(struct sock *sk, int size)
|
2015-04-22 08:32:24 +07:00
|
|
|
{
|
net: tcp_memcontrol: sanitize tcp memory accounting callbacks
There won't be a tcp control soft limit, so integrating the memcg code
into the global skmem limiting scheme complicates things unnecessarily.
Replace this with simple and clear charge and uncharge calls--hidden
behind a jump label--to account skb memory.
Note that this is not purely aesthetic: as a result of shoehorning the
per-memcg code into the same memory accounting functions that handle the
global level, the old code would compare the per-memcg consumption
against the smaller of the per-memcg limit and the global limit. This
allowed the total consumption of multiple sockets to exceed the global
limit, as long as the individual sockets stayed within bounds. After
this change, the code will always compare the per-memcg consumption to
the per-memcg limit, and the global consumption to the global limit, and
thus close this loophole.
Without a soft limit, the per-memcg memory pressure state in sockets is
generally questionable. However, we did it until now, so we continue to
enter it when the hard limit is hit, and packets are dropped, to let
other sockets in the cgroup know that they shouldn't grow their transmit
windows, either. However, keep it simple in the new callback model and
leave memory pressure lazily when the next packet is accepted (as
opposed to doing it synchroneously when packets are processed). When
packets are dropped, network performance will already be in the toilet,
so that should be a reasonable trade-off.
As described above, consumption is now checked on the per-memcg level
and the global level separately. Likewise, memory pressure states are
maintained on both the per-memcg level and the global level, and a
socket is considered under pressure when either level asserts as much.
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Reviewed-by: Vladimir Davydov <vdavydov@virtuozzo.com>
Acked-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-01-15 06:21:14 +07:00
|
|
|
int amt;
|
2015-04-22 08:32:24 +07:00
|
|
|
|
|
|
|
if (size <= sk->sk_forward_alloc)
|
|
|
|
return;
|
|
|
|
amt = sk_mem_pages(size);
|
|
|
|
sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
|
net: tcp_memcontrol: sanitize tcp memory accounting callbacks
There won't be a tcp control soft limit, so integrating the memcg code
into the global skmem limiting scheme complicates things unnecessarily.
Replace this with simple and clear charge and uncharge calls--hidden
behind a jump label--to account skb memory.
Note that this is not purely aesthetic: as a result of shoehorning the
per-memcg code into the same memory accounting functions that handle the
global level, the old code would compare the per-memcg consumption
against the smaller of the per-memcg limit and the global limit. This
allowed the total consumption of multiple sockets to exceed the global
limit, as long as the individual sockets stayed within bounds. After
this change, the code will always compare the per-memcg consumption to
the per-memcg limit, and the global consumption to the global limit, and
thus close this loophole.
Without a soft limit, the per-memcg memory pressure state in sockets is
generally questionable. However, we did it until now, so we continue to
enter it when the hard limit is hit, and packets are dropped, to let
other sockets in the cgroup know that they shouldn't grow their transmit
windows, either. However, keep it simple in the new callback model and
leave memory pressure lazily when the next packet is accepted (as
opposed to doing it synchroneously when packets are processed). When
packets are dropped, network performance will already be in the toilet,
so that should be a reasonable trade-off.
As described above, consumption is now checked on the per-memcg level
and the global level separately. Likewise, memory pressure states are
maintained on both the per-memcg level and the global level, and a
socket is considered under pressure when either level asserts as much.
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Reviewed-by: Vladimir Davydov <vdavydov@virtuozzo.com>
Acked-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-01-15 06:21:14 +07:00
|
|
|
sk_memory_allocated_add(sk, amt);
|
|
|
|
|
2016-01-15 06:21:17 +07:00
|
|
|
if (mem_cgroup_sockets_enabled && sk->sk_memcg)
|
|
|
|
mem_cgroup_charge_skmem(sk->sk_memcg, amt);
|
2015-04-22 08:32:24 +07:00
|
|
|
}
|
|
|
|
|
2015-04-24 00:42:39 +07:00
|
|
|
/* Send a FIN. The caller locks the socket for us.
|
|
|
|
* We should try to send a FIN packet really hard, but eventually give up.
|
2005-04-17 05:20:36 +07:00
|
|
|
*/
|
|
|
|
void tcp_send_fin(struct sock *sk)
|
|
|
|
{
|
2015-04-24 00:42:39 +07:00
|
|
|
struct sk_buff *skb, *tskb = tcp_write_queue_tail(sk);
|
2007-02-09 21:24:47 +07:00
|
|
|
struct tcp_sock *tp = tcp_sk(sk);
|
|
|
|
|
2015-04-24 00:42:39 +07:00
|
|
|
/* Optimization, tack on the FIN if we have one skb in write queue and
|
|
|
|
* this skb was not yet sent, or we are under memory pressure.
|
|
|
|
* Note: in the latter case, FIN packet will be sent after a timeout,
|
|
|
|
* as TCP stack thinks it has already been transmitted.
|
2005-04-17 05:20:36 +07:00
|
|
|
*/
|
2015-05-16 02:39:27 +07:00
|
|
|
if (tskb && (tcp_send_head(sk) || tcp_under_memory_pressure(sk))) {
|
2015-04-24 00:42:39 +07:00
|
|
|
coalesce:
|
|
|
|
TCP_SKB_CB(tskb)->tcp_flags |= TCPHDR_FIN;
|
|
|
|
TCP_SKB_CB(tskb)->end_seq++;
|
2005-04-17 05:20:36 +07:00
|
|
|
tp->write_seq++;
|
2015-04-24 00:42:39 +07:00
|
|
|
if (!tcp_send_head(sk)) {
|
|
|
|
/* This means tskb was already sent.
|
|
|
|
* Pretend we included the FIN on previous transmit.
|
|
|
|
* We need to set tp->snd_nxt to the value it would have
|
|
|
|
* if FIN had been sent. This is because retransmit path
|
|
|
|
* does not change tp->snd_nxt.
|
|
|
|
*/
|
|
|
|
tp->snd_nxt++;
|
|
|
|
return;
|
|
|
|
}
|
2005-04-17 05:20:36 +07:00
|
|
|
} else {
|
2015-04-24 00:42:39 +07:00
|
|
|
skb = alloc_skb_fclone(MAX_TCP_HEADER, sk->sk_allocation);
|
|
|
|
if (unlikely(!skb)) {
|
|
|
|
if (tskb)
|
|
|
|
goto coalesce;
|
|
|
|
return;
|
2005-04-17 05:20:36 +07:00
|
|
|
}
|
2015-04-22 08:32:24 +07:00
|
|
|
skb_reserve(skb, MAX_TCP_HEADER);
|
2015-05-16 02:39:26 +07:00
|
|
|
sk_forced_mem_schedule(sk, skb->truesize);
|
2005-04-17 05:20:36 +07:00
|
|
|
/* FIN eats a sequence byte, write_seq advanced by tcp_queue_skb(). */
|
2008-01-04 11:39:01 +07:00
|
|
|
tcp_init_nondata_skb(skb, tp->write_seq,
|
2010-06-12 21:01:43 +07:00
|
|
|
TCPHDR_ACK | TCPHDR_FIN);
|
2005-04-17 05:20:36 +07:00
|
|
|
tcp_queue_skb(sk, skb);
|
|
|
|
}
|
2015-04-24 00:42:39 +07:00
|
|
|
__tcp_push_pending_frames(sk, tcp_current_mss(sk), TCP_NAGLE_OFF);
|
2005-04-17 05:20:36 +07:00
|
|
|
}
|
|
|
|
|
|
|
|
/* We get here when a process closes a file descriptor (either due to
|
|
|
|
* an explicit close() or as a byproduct of exit()'ing) and there
|
|
|
|
* was unread data in the receive queue. This behavior is recommended
|
2007-04-29 11:21:46 +07:00
|
|
|
* by RFC 2525, section 2.17. -DaveM
|
2005-04-17 05:20:36 +07:00
|
|
|
*/
|
2005-10-07 13:46:04 +07:00
|
|
|
void tcp_send_active_reset(struct sock *sk, gfp_t priority)
|
2005-04-17 05:20:36 +07:00
|
|
|
{
|
|
|
|
struct sk_buff *skb;
|
|
|
|
|
|
|
|
/* NOTE: No TCP options attached and we never retransmit this. */
|
|
|
|
skb = alloc_skb(MAX_TCP_HEADER, priority);
|
|
|
|
if (!skb) {
|
2008-07-17 10:30:14 +07:00
|
|
|
NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPABORTFAILED);
|
2005-04-17 05:20:36 +07:00
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Reserve space for headers and prepare control bits. */
|
|
|
|
skb_reserve(skb, MAX_TCP_HEADER);
|
2008-01-04 11:39:01 +07:00
|
|
|
tcp_init_nondata_skb(skb, tcp_acceptable_seq(sk),
|
2010-06-12 21:01:43 +07:00
|
|
|
TCPHDR_ACK | TCPHDR_RST);
|
tcp: add proper TS val into RST packets
RST packets sent on behalf of TCP connections with TS option (RFC 7323
TCP timestamps) have incorrect TS val (set to 0), but correct TS ecr.
A > B: Flags [S], seq 0, win 65535, options [mss 1000,nop,nop,TS val 100
ecr 0], length 0
B > A: Flags [S.], seq 2444755794, ack 1, win 28960, options [mss
1460,nop,nop,TS val 7264344 ecr 100], length 0
A > B: Flags [.], ack 1, win 65535, options [nop,nop,TS val 110 ecr
7264344], length 0
B > A: Flags [R.], seq 1, ack 1, win 28960, options [nop,nop,TS val 0
ecr 110], length 0
We need to call skb_mstamp_get() to get proper TS val,
derived from skb->skb_mstamp
Note that RFC 1323 was advocating to not send TS option in RST segment,
but RFC 7323 recommends the opposite :
Once TSopt has been successfully negotiated, that is both <SYN> and
<SYN,ACK> contain TSopt, the TSopt MUST be sent in every non-<RST>
segment for the duration of the connection, and SHOULD be sent in an
<RST> segment (see Section 5.2 for details)
Note this RFC recommends to send TS val = 0, but we believe it is
premature : We do not know if all TCP stacks are properly
handling the receive side :
When an <RST> segment is
received, it MUST NOT be subjected to the PAWS check by verifying an
acceptable value in SEG.TSval, and information from the Timestamps
option MUST NOT be used to update connection state information.
SEG.TSecr MAY be used to provide stricter <RST> acceptance checks.
In 5 years, if/when all TCP stack are RFC 7323 ready, we might consider
to decide to send TS val = 0, if it buys something.
Fixes: 7faee5c0d514 ("tcp: remove TCP_SKB_CB(skb)->when")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2015-09-24 04:00:21 +07:00
|
|
|
skb_mstamp_get(&skb->skb_mstamp);
|
2005-04-17 05:20:36 +07:00
|
|
|
/* Send it off. */
|
2005-12-07 07:24:52 +07:00
|
|
|
if (tcp_transmit_skb(sk, skb, 0, priority))
|
2008-07-17 10:30:14 +07:00
|
|
|
NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPABORTFAILED);
|
2008-06-05 05:19:35 +07:00
|
|
|
|
2008-07-17 10:22:04 +07:00
|
|
|
TCP_INC_STATS(sock_net(sk), TCP_MIB_OUTRSTS);
|
2005-04-17 05:20:36 +07:00
|
|
|
}
|
|
|
|
|
2009-07-22 06:00:40 +07:00
|
|
|
/* Send a crossed SYN-ACK during socket establishment.
|
|
|
|
* WARNING: This routine must only be called when we have already sent
|
2005-04-17 05:20:36 +07:00
|
|
|
* a SYN packet that crossed the incoming SYN that caused this routine
|
|
|
|
* to get called. If this assumption fails then the initial rcv_wnd
|
|
|
|
* and rcv_wscale values will not be correct.
|
|
|
|
*/
|
|
|
|
int tcp_send_synack(struct sock *sk)
|
|
|
|
{
|
2008-01-01 05:57:14 +07:00
|
|
|
struct sk_buff *skb;
|
2005-04-17 05:20:36 +07:00
|
|
|
|
2007-03-08 03:12:44 +07:00
|
|
|
skb = tcp_write_queue_head(sk);
|
2015-04-03 15:17:26 +07:00
|
|
|
if (!skb || !(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_SYN)) {
|
2012-05-15 21:11:54 +07:00
|
|
|
pr_debug("%s: wrong queue state\n", __func__);
|
2005-04-17 05:20:36 +07:00
|
|
|
return -EFAULT;
|
|
|
|
}
|
2011-09-28 00:25:05 +07:00
|
|
|
if (!(TCP_SKB_CB(skb)->tcp_flags & TCPHDR_ACK)) {
|
2005-04-17 05:20:36 +07:00
|
|
|
if (skb_cloned(skb)) {
|
|
|
|
struct sk_buff *nskb = skb_copy(skb, GFP_ATOMIC);
|
2015-04-03 15:17:26 +07:00
|
|
|
if (!nskb)
|
2005-04-17 05:20:36 +07:00
|
|
|
return -ENOMEM;
|
2007-03-08 03:12:44 +07:00
|
|
|
tcp_unlink_write_queue(skb, sk);
|
2014-09-23 06:29:32 +07:00
|
|
|
__skb_header_release(nskb);
|
2007-03-08 03:12:44 +07:00
|
|
|
__tcp_add_write_queue_head(sk, nskb);
|
2007-12-31 15:11:19 +07:00
|
|
|
sk_wmem_free_skb(sk, skb);
|
|
|
|
sk->sk_wmem_queued += nskb->truesize;
|
|
|
|
sk_mem_charge(sk, nskb->truesize);
|
2005-04-17 05:20:36 +07:00
|
|
|
skb = nskb;
|
|
|
|
}
|
|
|
|
|
2011-09-28 00:25:05 +07:00
|
|
|
TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_ACK;
|
2014-09-29 18:08:30 +07:00
|
|
|
tcp_ecn_send_synack(sk, skb);
|
2005-04-17 05:20:36 +07:00
|
|
|
}
|
2005-12-07 07:24:52 +07:00
|
|
|
return tcp_transmit_skb(sk, skb, 1, GFP_ATOMIC);
|
2005-04-17 05:20:36 +07:00
|
|
|
}
|
|
|
|
|
2012-06-04 03:33:21 +07:00
|
|
|
/**
|
|
|
|
* tcp_make_synack - Prepare a SYN-ACK.
|
|
|
|
* sk: listener socket
|
|
|
|
* dst: dst entry attached to the SYNACK
|
|
|
|
* req: request_sock pointer
|
|
|
|
*
|
|
|
|
* Allocate one skb and build a SYNACK packet.
|
|
|
|
* @dst is consumed : Caller should not use it again.
|
|
|
|
*/
|
2015-09-25 21:39:19 +07:00
|
|
|
struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst,
|
2009-12-03 01:07:39 +07:00
|
|
|
struct request_sock *req,
|
2015-10-03 01:43:35 +07:00
|
|
|
struct tcp_fastopen_cookie *foc,
|
|
|
|
bool attach_req)
|
2005-04-17 05:20:36 +07:00
|
|
|
{
|
[NET] Generalise TCP's struct open_request minisock infrastructure
Kept this first changeset minimal, without changing existing names to
ease peer review.
Basicaly tcp_openreq_alloc now receives the or_calltable, that in turn
has two new members:
->slab, that replaces tcp_openreq_cachep
->obj_size, to inform the size of the openreq descendant for
a specific protocol
The protocol specific fields in struct open_request were moved to a
class hierarchy, with the things that are common to all connection
oriented PF_INET protocols in struct inet_request_sock, the TCP ones
in tcp_request_sock, that is an inet_request_sock, that is an
open_request.
I.e. this uses the same approach used for the struct sock class
hierarchy, with sk_prot indicating if the protocol wants to use the
open_request infrastructure by filling in sk_prot->rsk_prot with an
or_calltable.
Results? Performance is improved and TCP v4 now uses only 64 bytes per
open request minisock, down from 96 without this patch :-)
Next changeset will rename some of the structs, fields and functions
mentioned above, struct or_calltable is way unclear, better name it
struct request_sock_ops, s/struct open_request/struct request_sock/g,
etc.
Signed-off-by: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
2005-06-19 12:46:52 +07:00
|
|
|
struct inet_request_sock *ireq = inet_rsk(req);
|
2015-09-25 21:39:19 +07:00
|
|
|
const struct tcp_sock *tp = tcp_sk(sk);
|
2015-03-25 05:58:52 +07:00
|
|
|
struct tcp_md5sig_key *md5 = NULL;
|
2015-09-25 21:39:19 +07:00
|
|
|
struct tcp_out_options opts;
|
|
|
|
struct sk_buff *skb;
|
2009-12-03 01:23:05 +07:00
|
|
|
int tcp_header_size;
|
2015-09-25 21:39:19 +07:00
|
|
|
struct tcphdr *th;
|
|
|
|
u16 user_mss;
|
2008-09-21 14:21:51 +07:00
|
|
|
int mss;
|
2005-04-17 05:20:36 +07:00
|
|
|
|
2015-10-03 01:43:35 +07:00
|
|
|
skb = alloc_skb(MAX_TCP_HEADER, GFP_ATOMIC);
|
2012-06-04 03:33:21 +07:00
|
|
|
if (unlikely(!skb)) {
|
|
|
|
dst_release(dst);
|
2005-04-17 05:20:36 +07:00
|
|
|
return NULL;
|
2012-06-04 03:33:21 +07:00
|
|
|
}
|
2005-04-17 05:20:36 +07:00
|
|
|
/* Reserve space for headers. */
|
|
|
|
skb_reserve(skb, MAX_TCP_HEADER);
|
|
|
|
|
2015-10-03 01:43:35 +07:00
|
|
|
if (attach_req) {
|
2015-11-02 06:36:55 +07:00
|
|
|
skb_set_owner_w(skb, req_to_sk(req));
|
2015-10-03 01:43:35 +07:00
|
|
|
} else {
|
|
|
|
/* sk is a const pointer, because we want to express multiple
|
|
|
|
* cpu might call us concurrently.
|
|
|
|
* sk->sk_wmem_alloc in an atomic, we can promote to rw.
|
|
|
|
*/
|
|
|
|
skb_set_owner_w(skb, (struct sock *)sk);
|
|
|
|
}
|
2012-06-04 03:33:21 +07:00
|
|
|
skb_dst_set(skb, dst);
|
2005-04-17 05:20:36 +07:00
|
|
|
|
2010-12-14 03:52:14 +07:00
|
|
|
mss = dst_metric_advmss(dst);
|
2015-09-25 21:39:19 +07:00
|
|
|
user_mss = READ_ONCE(tp->rx_opt.user_mss);
|
|
|
|
if (user_mss && user_mss < mss)
|
|
|
|
mss = user_mss;
|
2008-09-21 14:21:51 +07:00
|
|
|
|
2008-07-19 14:04:31 +07:00
|
|
|
memset(&opts, 0, sizeof(opts));
|
2008-10-27 13:10:12 +07:00
|
|
|
#ifdef CONFIG_SYN_COOKIES
|
|
|
|
if (unlikely(req->cookie_ts))
|
2014-09-06 05:33:33 +07:00
|
|
|
skb->skb_mstamp.stamp_jiffies = cookie_init_timestamp(req);
|
2008-10-27 13:10:12 +07:00
|
|
|
else
|
|
|
|
#endif
|
2014-09-06 05:33:33 +07:00
|
|
|
skb_mstamp_get(&skb->skb_mstamp);
|
2015-03-25 05:58:52 +07:00
|
|
|
|
|
|
|
#ifdef CONFIG_TCP_MD5SIG
|
|
|
|
rcu_read_lock();
|
2015-03-25 05:58:56 +07:00
|
|
|
md5 = tcp_rsk(req)->af_specific->req_md5_lookup(sk, req_to_sk(req));
|
2015-03-25 05:58:52 +07:00
|
|
|
#endif
|
2015-09-16 05:24:20 +07:00
|
|
|
skb_set_hash(skb, tcp_rsk(req)->txhash, PKT_HASH_TYPE_L4);
|
2015-09-25 21:39:17 +07:00
|
|
|
tcp_header_size = tcp_synack_options(req, mss, skb, &opts, md5, foc) +
|
|
|
|
sizeof(*th);
|
2006-11-15 10:07:45 +07:00
|
|
|
|
2007-04-11 11:04:22 +07:00
|
|
|
skb_push(skb, tcp_header_size);
|
|
|
|
skb_reset_transport_header(skb);
|
2005-04-17 05:20:36 +07:00
|
|
|
|
2007-04-11 11:04:22 +07:00
|
|
|
th = tcp_hdr(skb);
|
2005-04-17 05:20:36 +07:00
|
|
|
memset(th, 0, sizeof(struct tcphdr));
|
|
|
|
th->syn = 1;
|
|
|
|
th->ack = 1;
|
2015-09-25 21:39:18 +07:00
|
|
|
tcp_ecn_make_synack(req, th);
|
2013-10-10 14:04:37 +07:00
|
|
|
th->source = htons(ireq->ir_num);
|
2013-10-10 05:21:29 +07:00
|
|
|
th->dest = ireq->ir_rmt_port;
|
2008-01-04 11:39:01 +07:00
|
|
|
/* Setting of flags are superfluous here for callers (and ECE is
|
|
|
|
* not even correctly set)
|
|
|
|
*/
|
|
|
|
tcp_init_nondata_skb(skb, tcp_rsk(req)->snt_isn,
|
2010-06-12 21:01:43 +07:00
|
|
|
TCPHDR_SYN | TCPHDR_ACK);
|
2009-12-03 01:25:27 +07:00
|
|
|
|
2005-04-17 05:20:36 +07:00
|
|
|
th->seq = htonl(TCP_SKB_CB(skb)->seq);
|
2012-08-31 19:29:12 +07:00
|
|
|
/* XXX data is queued and acked as is. No buffer/window check */
|
|
|
|
th->ack_seq = htonl(tcp_rsk(req)->rcv_nxt);
|
2005-04-17 05:20:36 +07:00
|
|
|
|
|
|
|
/* RFC1323: The window in SYN & SYN/ACK segments is never scaled. */
|
2015-10-09 09:33:23 +07:00
|
|
|
th->window = htons(min(req->rsk_rcv_wnd, 65535U));
|
2015-09-25 21:39:19 +07:00
|
|
|
tcp_options_write((__be32 *)(th + 1), NULL, &opts);
|
2005-04-17 05:20:36 +07:00
|
|
|
th->doff = (tcp_header_size >> 2);
|
2014-03-26 23:57:19 +07:00
|
|
|
TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_OUTSEGS);
|
2006-11-15 10:07:45 +07:00
|
|
|
|
|
|
|
#ifdef CONFIG_TCP_MD5SIG
|
|
|
|
/* Okay, we have all we need - do the md5 hash if needed */
|
2015-03-25 05:58:52 +07:00
|
|
|
if (md5)
|
2009-12-03 01:23:05 +07:00
|
|
|
tcp_rsk(req)->af_specific->calc_md5_hash(opts.hash_location,
|
2015-03-25 05:58:55 +07:00
|
|
|
md5, req_to_sk(req), skb);
|
2015-03-25 05:58:52 +07:00
|
|
|
rcu_read_unlock();
|
2006-11-15 10:07:45 +07:00
|
|
|
#endif
|
|
|
|
|
tcp: tcp_make_synack() should clear skb->tstamp
I noticed tcpdump was giving funky timestamps for locally
generated SYNACK messages on loopback interface.
11:42:46.938990 IP 127.0.0.1.48245 > 127.0.0.2.23850: S
945476042:945476042(0) win 43690 <mss 65495,nop,nop,sackOK,nop,wscale 7>
20:28:58.502209 IP 127.0.0.2.23850 > 127.0.0.1.48245: S
3160535375:3160535375(0) ack 945476043 win 43690 <mss
65495,nop,nop,sackOK,nop,wscale 7>
This is because we need to clear skb->tstamp before
entering lower stack, otherwise net_timestamp_check()
does not set skb->tstamp.
Fixes: 7faee5c0d514 ("tcp: remove TCP_SKB_CB(skb)->when")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2015-04-10 03:31:56 +07:00
|
|
|
/* Do not fool tcpdump (if any), clean our debris */
|
|
|
|
skb->tstamp.tv64 = 0;
|
2005-04-17 05:20:36 +07:00
|
|
|
return skb;
|
|
|
|
}
|
2010-07-10 04:22:10 +07:00
|
|
|
EXPORT_SYMBOL(tcp_make_synack);
|
2005-04-17 05:20:36 +07:00
|
|
|
|
net: tcp: add per route congestion control
This work adds the possibility to define a per route/destination
congestion control algorithm. Generally, this opens up the possibility
for a machine with different links to enforce specific congestion
control algorithms with optimal strategies for each of them based
on their network characteristics, even transparently for a single
application listening on all links.
For our specific use case, this additionally facilitates deployment
of DCTCP, for example, applications can easily serve internal
traffic/dsts in DCTCP and external one with CUBIC. Other scenarios
would also allow for utilizing e.g. long living, low priority
background flows for certain destinations/routes while still being
able for normal traffic to utilize the default congestion control
algorithm. We also thought about a per netns setting (where different
defaults are possible), but given its actually a link specific
property, we argue that a per route/destination setting is the most
natural and flexible.
The administrator can utilize this through ip-route(8) by appending
"congctl [lock] <name>", where <name> denotes the name of a
congestion control algorithm and the optional lock parameter allows
to enforce the given algorithm so that applications in user space
would not be allowed to overwrite that algorithm for that destination.
The dst metric lookups are being done when a dst entry is already
available in order to avoid a costly lookup and still before the
algorithms are being initialized, thus overhead is very low when the
feature is not being used. While the client side would need to drop
the current reference on the module, on server side this can actually
even be avoided as we just got a flat-copied socket clone.
Joint work with Florian Westphal.
Suggested-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Daniel Borkmann <dborkman@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2015-01-06 05:57:48 +07:00
|
|
|
static void tcp_ca_dst_init(struct sock *sk, const struct dst_entry *dst)
|
|
|
|
{
|
|
|
|
struct inet_connection_sock *icsk = inet_csk(sk);
|
|
|
|
const struct tcp_congestion_ops *ca;
|
|
|
|
u32 ca_key = dst_metric(dst, RTAX_CC_ALGO);
|
|
|
|
|
|
|
|
if (ca_key == TCP_CA_UNSPEC)
|
|
|
|
return;
|
|
|
|
|
|
|
|
rcu_read_lock();
|
|
|
|
ca = tcp_ca_find_key(ca_key);
|
|
|
|
if (likely(ca && try_module_get(ca->owner))) {
|
|
|
|
module_put(icsk->icsk_ca_ops->owner);
|
|
|
|
icsk->icsk_ca_dst_locked = tcp_ca_dst_locked(dst);
|
|
|
|
icsk->icsk_ca_ops = ca;
|
|
|
|
}
|
|
|
|
rcu_read_unlock();
|
|
|
|
}
|
|
|
|
|
2009-07-22 06:00:40 +07:00
|
|
|
/* Do all connect socket setups that can be done AF independent. */
|
2013-12-30 02:39:51 +07:00
|
|
|
static void tcp_connect_init(struct sock *sk)
|
2005-04-17 05:20:36 +07:00
|
|
|
{
|
2011-10-21 16:22:42 +07:00
|
|
|
const struct dst_entry *dst = __sk_dst_get(sk);
|
2005-04-17 05:20:36 +07:00
|
|
|
struct tcp_sock *tp = tcp_sk(sk);
|
|
|
|
__u8 rcv_wscale;
|
|
|
|
|
|
|
|
/* We'll fix this up when we get a response from the other end.
|
|
|
|
* See tcp_input.c:tcp_rcv_state_process case TCP_SYN_SENT.
|
|
|
|
*/
|
|
|
|
tp->tcp_header_len = sizeof(struct tcphdr) +
|
tcp: Revert per-route SACK/DSACK/TIMESTAMP changes.
It creates a regression, triggering badness for SYN_RECV
sockets, for example:
[19148.022102] Badness at net/ipv4/inet_connection_sock.c:293
[19148.022570] NIP: c02a0914 LR: c02a0904 CTR: 00000000
[19148.023035] REGS: eeecbd30 TRAP: 0700 Not tainted (2.6.32)
[19148.023496] MSR: 00029032 <EE,ME,CE,IR,DR> CR: 24002442 XER: 00000000
[19148.024012] TASK = eee9a820[1756] 'privoxy' THREAD: eeeca000
This is likely caused by the change in the 'estab' parameter
passed to tcp_parse_options() when invoked by the functions
in net/ipv4/tcp_minisocks.c
But even if that is fixed, the ->conn_request() changes made in
this patch series is fundamentally wrong. They try to use the
listening socket's 'dst' to probe the route settings. The
listening socket doesn't even have a route, and you can't
get the right route (the child request one) until much later
after we setup all of the state, and it must be done by hand.
This stuff really isn't ready, so the best thing to do is a
full revert. This reverts the following commits:
f55017a93f1a74d50244b1254b9a2bd7ac9bbf7d
022c3f7d82f0f1c68018696f2f027b87b9bb45c2
1aba721eba1d84a2defce45b950272cee1e6c72a
cda42ebd67ee5fdf09d7057b5a4584d36fe8a335
345cda2fd695534be5a4494f1b59da9daed33663
dc343475ed062e13fc260acccaab91d7d80fd5b2
05eaade2782fb0c90d3034fd7a7d5a16266182bb
6a2a2d6bf8581216e08be15fcb563cfd6c430e1e
Signed-off-by: David S. Miller <davem@davemloft.net>
2009-12-16 11:56:42 +07:00
|
|
|
(sysctl_tcp_timestamps ? TCPOLEN_TSTAMP_ALIGNED : 0);
|
2005-04-17 05:20:36 +07:00
|
|
|
|
2006-11-15 10:07:45 +07:00
|
|
|
#ifdef CONFIG_TCP_MD5SIG
|
2015-04-03 15:17:27 +07:00
|
|
|
if (tp->af_specific->md5_lookup(sk, sk))
|
2006-11-15 10:07:45 +07:00
|
|
|
tp->tcp_header_len += TCPOLEN_MD5SIG_ALIGNED;
|
|
|
|
#endif
|
|
|
|
|
2005-04-17 05:20:36 +07:00
|
|
|
/* If user gave his TCP_MAXSEG, record it to clamp */
|
|
|
|
if (tp->rx_opt.user_mss)
|
|
|
|
tp->rx_opt.mss_clamp = tp->rx_opt.user_mss;
|
|
|
|
tp->max_window = 0;
|
2006-03-21 08:53:41 +07:00
|
|
|
tcp_mtup_init(sk);
|
2005-04-17 05:20:36 +07:00
|
|
|
tcp_sync_mss(sk, dst_mtu(dst));
|
|
|
|
|
net: tcp: add per route congestion control
This work adds the possibility to define a per route/destination
congestion control algorithm. Generally, this opens up the possibility
for a machine with different links to enforce specific congestion
control algorithms with optimal strategies for each of them based
on their network characteristics, even transparently for a single
application listening on all links.
For our specific use case, this additionally facilitates deployment
of DCTCP, for example, applications can easily serve internal
traffic/dsts in DCTCP and external one with CUBIC. Other scenarios
would also allow for utilizing e.g. long living, low priority
background flows for certain destinations/routes while still being
able for normal traffic to utilize the default congestion control
algorithm. We also thought about a per netns setting (where different
defaults are possible), but given its actually a link specific
property, we argue that a per route/destination setting is the most
natural and flexible.
The administrator can utilize this through ip-route(8) by appending
"congctl [lock] <name>", where <name> denotes the name of a
congestion control algorithm and the optional lock parameter allows
to enforce the given algorithm so that applications in user space
would not be allowed to overwrite that algorithm for that destination.
The dst metric lookups are being done when a dst entry is already
available in order to avoid a costly lookup and still before the
algorithms are being initialized, thus overhead is very low when the
feature is not being used. While the client side would need to drop
the current reference on the module, on server side this can actually
even be avoided as we just got a flat-copied socket clone.
Joint work with Florian Westphal.
Suggested-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Daniel Borkmann <dborkman@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2015-01-06 05:57:48 +07:00
|
|
|
tcp_ca_dst_init(sk, dst);
|
|
|
|
|
2005-04-17 05:20:36 +07:00
|
|
|
if (!tp->window_clamp)
|
|
|
|
tp->window_clamp = dst_metric(dst, RTAX_WINDOW);
|
2010-12-14 03:52:14 +07:00
|
|
|
tp->advmss = dst_metric_advmss(dst);
|
2008-09-21 14:21:51 +07:00
|
|
|
if (tp->rx_opt.user_mss && tp->rx_opt.user_mss < tp->advmss)
|
|
|
|
tp->advmss = tp->rx_opt.user_mss;
|
|
|
|
|
2005-04-17 05:20:36 +07:00
|
|
|
tcp_initialize_rcv_mss(sk);
|
|
|
|
|
tcp: allow effective reduction of TCP's rcv-buffer via setsockopt
Via setsockopt it is possible to reduce the socket RX buffer
(SO_RCVBUF). TCP method to select the initial window and window scaling
option in tcp_select_initial_window() currently misbehaves and do not
consider a reduced RX socket buffer via setsockopt.
Even though the server's RX buffer is reduced via setsockopt() to 256
byte (Initial Window 384 byte => 256 * 2 - (256 * 2 / 4)) the window
scale option is still 7:
192.168.1.38.40676 > 78.47.222.210.5001: Flags [S], seq 2577214362, win 5840, options [mss 1460,sackOK,TS val 338417 ecr 0,nop,wscale 0], length 0
78.47.222.210.5001 > 192.168.1.38.40676: Flags [S.], seq 1570631029, ack 2577214363, win 384, options [mss 1452,sackOK,TS val 2435248895 ecr 338417,nop,wscale 7], length 0
192.168.1.38.40676 > 78.47.222.210.5001: Flags [.], ack 1, win 5840, options [nop,nop,TS val 338421 ecr 2435248895], length 0
Within tcp_select_initial_window() the original space argument - a
representation of the rx buffer size - is expanded during
tcp_select_initial_window(). Only sysctl_tcp_rmem[2], sysctl_rmem_max
and window_clamp are considered to calculate the initial window.
This patch adjust the window_clamp argument if the user explicitly
reduce the receive buffer.
Signed-off-by: Hagen Paul Pfeifer <hagen@jauu.net>
Cc: David S. Miller <davem@davemloft.net>
Cc: Patrick McHardy <kaber@trash.net>
Cc: Eric Dumazet <eric.dumazet@gmail.com>
Cc: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>
2010-08-19 13:33:05 +07:00
|
|
|
/* limit the window selection if the user enforce a smaller rx buffer */
|
|
|
|
if (sk->sk_userlocks & SOCK_RCVBUF_LOCK &&
|
|
|
|
(tp->window_clamp > tcp_full_space(sk) || tp->window_clamp == 0))
|
|
|
|
tp->window_clamp = tcp_full_space(sk);
|
|
|
|
|
2005-04-17 05:20:36 +07:00
|
|
|
tcp_select_initial_window(tcp_full_space(sk),
|
|
|
|
tp->advmss - (tp->rx_opt.ts_recent_stamp ? tp->tcp_header_len - sizeof(struct tcphdr) : 0),
|
|
|
|
&tp->rcv_wnd,
|
|
|
|
&tp->window_clamp,
|
tcp: Revert per-route SACK/DSACK/TIMESTAMP changes.
It creates a regression, triggering badness for SYN_RECV
sockets, for example:
[19148.022102] Badness at net/ipv4/inet_connection_sock.c:293
[19148.022570] NIP: c02a0914 LR: c02a0904 CTR: 00000000
[19148.023035] REGS: eeecbd30 TRAP: 0700 Not tainted (2.6.32)
[19148.023496] MSR: 00029032 <EE,ME,CE,IR,DR> CR: 24002442 XER: 00000000
[19148.024012] TASK = eee9a820[1756] 'privoxy' THREAD: eeeca000
This is likely caused by the change in the 'estab' parameter
passed to tcp_parse_options() when invoked by the functions
in net/ipv4/tcp_minisocks.c
But even if that is fixed, the ->conn_request() changes made in
this patch series is fundamentally wrong. They try to use the
listening socket's 'dst' to probe the route settings. The
listening socket doesn't even have a route, and you can't
get the right route (the child request one) until much later
after we setup all of the state, and it must be done by hand.
This stuff really isn't ready, so the best thing to do is a
full revert. This reverts the following commits:
f55017a93f1a74d50244b1254b9a2bd7ac9bbf7d
022c3f7d82f0f1c68018696f2f027b87b9bb45c2
1aba721eba1d84a2defce45b950272cee1e6c72a
cda42ebd67ee5fdf09d7057b5a4584d36fe8a335
345cda2fd695534be5a4494f1b59da9daed33663
dc343475ed062e13fc260acccaab91d7d80fd5b2
05eaade2782fb0c90d3034fd7a7d5a16266182bb
6a2a2d6bf8581216e08be15fcb563cfd6c430e1e
Signed-off-by: David S. Miller <davem@davemloft.net>
2009-12-16 11:56:42 +07:00
|
|
|
sysctl_tcp_window_scaling,
|
2009-12-15 18:15:28 +07:00
|
|
|
&rcv_wscale,
|
|
|
|
dst_metric(dst, RTAX_INITRWND));
|
2005-04-17 05:20:36 +07:00
|
|
|
|
|
|
|
tp->rx_opt.rcv_wscale = rcv_wscale;
|
|
|
|
tp->rcv_ssthresh = tp->rcv_wnd;
|
|
|
|
|
|
|
|
sk->sk_err = 0;
|
|
|
|
sock_reset_flag(sk, SOCK_DONE);
|
|
|
|
tp->snd_wnd = 0;
|
2009-03-03 13:42:02 +07:00
|
|
|
tcp_init_wl(tp, 0);
|
2005-04-17 05:20:36 +07:00
|
|
|
tp->snd_una = tp->write_seq;
|
|
|
|
tp->snd_sml = tp->write_seq;
|
2008-10-08 04:43:06 +07:00
|
|
|
tp->snd_up = tp->write_seq;
|
2012-04-19 10:40:01 +07:00
|
|
|
tp->snd_nxt = tp->write_seq;
|
2012-04-19 10:40:39 +07:00
|
|
|
|
|
|
|
if (likely(!tp->repair))
|
|
|
|
tp->rcv_nxt = 0;
|
2013-08-27 15:20:40 +07:00
|
|
|
else
|
|
|
|
tp->rcv_tstamp = tcp_time_stamp;
|
2012-04-19 10:40:39 +07:00
|
|
|
tp->rcv_wup = tp->rcv_nxt;
|
|
|
|
tp->copied_seq = tp->rcv_nxt;
|
2005-04-17 05:20:36 +07:00
|
|
|
|
2005-08-10 10:10:42 +07:00
|
|
|
inet_csk(sk)->icsk_rto = TCP_TIMEOUT_INIT;
|
|
|
|
inet_csk(sk)->icsk_retransmits = 0;
|
2005-04-17 05:20:36 +07:00
|
|
|
tcp_clear_retrans(tp);
|
|
|
|
}
|
|
|
|
|
2012-07-19 13:43:07 +07:00
|
|
|
static void tcp_connect_queue_skb(struct sock *sk, struct sk_buff *skb)
|
|
|
|
{
|
|
|
|
struct tcp_sock *tp = tcp_sk(sk);
|
|
|
|
struct tcp_skb_cb *tcb = TCP_SKB_CB(skb);
|
|
|
|
|
|
|
|
tcb->end_seq += skb->len;
|
2014-09-23 06:29:32 +07:00
|
|
|
__skb_header_release(skb);
|
2012-07-19 13:43:07 +07:00
|
|
|
__tcp_add_write_queue_tail(sk, skb);
|
|
|
|
sk->sk_wmem_queued += skb->truesize;
|
|
|
|
sk_mem_charge(sk, skb->truesize);
|
|
|
|
tp->write_seq = tcb->end_seq;
|
|
|
|
tp->packets_out += tcp_skb_pcount(skb);
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Build and send a SYN with data and (cached) Fast Open cookie. However,
|
|
|
|
* queue a data-only packet after the regular SYN, such that regular SYNs
|
|
|
|
* are retransmitted on timeouts. Also if the remote SYN-ACK acknowledges
|
|
|
|
* only the SYN sequence, the data are retransmitted in the first ACK.
|
|
|
|
* If cookie is not cached or other error occurs, falls back to send a
|
|
|
|
* regular SYN with Fast Open cookie request option.
|
|
|
|
*/
|
|
|
|
static int tcp_send_syn_data(struct sock *sk, struct sk_buff *syn)
|
|
|
|
{
|
|
|
|
struct tcp_sock *tp = tcp_sk(sk);
|
|
|
|
struct tcp_fastopen_request *fo = tp->fastopen_req;
|
2015-12-17 04:53:10 +07:00
|
|
|
int syn_loss = 0, space, err = 0;
|
2012-07-19 13:43:10 +07:00
|
|
|
unsigned long last_syn_loss = 0;
|
2014-11-18 14:06:20 +07:00
|
|
|
struct sk_buff *syn_data;
|
2012-07-19 13:43:10 +07:00
|
|
|
|
2012-07-19 13:43:11 +07:00
|
|
|
tp->rx_opt.mss_clamp = tp->advmss; /* If MSS is not cached */
|
2012-07-19 13:43:10 +07:00
|
|
|
tcp_fastopen_cache_get(sk, &tp->rx_opt.mss_clamp, &fo->cookie,
|
|
|
|
&syn_loss, &last_syn_loss);
|
|
|
|
/* Recurring FO SYN losses: revert to regular handshake temporarily */
|
|
|
|
if (syn_loss > 1 &&
|
|
|
|
time_before(jiffies, last_syn_loss + (60*HZ << syn_loss))) {
|
|
|
|
fo->cookie.len = -1;
|
|
|
|
goto fallback;
|
|
|
|
}
|
2012-07-19 13:43:07 +07:00
|
|
|
|
2012-07-19 13:43:11 +07:00
|
|
|
if (sysctl_tcp_fastopen & TFO_CLIENT_NO_COOKIE)
|
|
|
|
fo->cookie.len = -1;
|
|
|
|
else if (fo->cookie.len <= 0)
|
2012-07-19 13:43:07 +07:00
|
|
|
goto fallback;
|
|
|
|
|
|
|
|
/* MSS for SYN-data is based on cached MSS and bounded by PMTU and
|
|
|
|
* user-MSS. Reserve maximum option space for middleboxes that add
|
|
|
|
* private TCP options. The cost is reduced data space in SYN :(
|
|
|
|
*/
|
|
|
|
if (tp->rx_opt.user_mss && tp->rx_opt.user_mss < tp->rx_opt.mss_clamp)
|
|
|
|
tp->rx_opt.mss_clamp = tp->rx_opt.user_mss;
|
2013-02-22 15:59:06 +07:00
|
|
|
space = __tcp_mtu_to_mss(sk, inet_csk(sk)->icsk_pmtu_cookie) -
|
2012-07-19 13:43:07 +07:00
|
|
|
MAX_TCP_OPTION_SPACE;
|
|
|
|
|
2014-02-21 01:09:18 +07:00
|
|
|
space = min_t(size_t, space, fo->size);
|
|
|
|
|
|
|
|
/* limit to order-0 allocations */
|
|
|
|
space = min_t(size_t, space, SKB_MAX_HEAD(MAX_TCP_HEADER));
|
|
|
|
|
2015-05-20 03:26:55 +07:00
|
|
|
syn_data = sk_stream_alloc_skb(sk, space, sk->sk_allocation, false);
|
2014-11-18 14:06:20 +07:00
|
|
|
if (!syn_data)
|
2012-07-19 13:43:07 +07:00
|
|
|
goto fallback;
|
2014-11-18 14:06:20 +07:00
|
|
|
syn_data->ip_summed = CHECKSUM_PARTIAL;
|
|
|
|
memcpy(syn_data->cb, syn->cb, sizeof(syn->cb));
|
2015-12-17 04:53:10 +07:00
|
|
|
if (space) {
|
|
|
|
int copied = copy_from_iter(skb_put(syn_data, space), space,
|
|
|
|
&fo->data->msg_iter);
|
|
|
|
if (unlikely(!copied)) {
|
|
|
|
kfree_skb(syn_data);
|
|
|
|
goto fallback;
|
|
|
|
}
|
|
|
|
if (copied != space) {
|
|
|
|
skb_trim(syn_data, copied);
|
|
|
|
space = copied;
|
|
|
|
}
|
2014-11-29 01:40:20 +07:00
|
|
|
}
|
2014-11-18 14:06:20 +07:00
|
|
|
/* No more data pending in inet_wait_for_connect() */
|
|
|
|
if (space == fo->size)
|
|
|
|
fo->data = NULL;
|
|
|
|
fo->copied = space;
|
2012-07-19 13:43:07 +07:00
|
|
|
|
2014-11-18 14:06:20 +07:00
|
|
|
tcp_connect_queue_skb(sk, syn_data);
|
2012-07-19 13:43:07 +07:00
|
|
|
|
2014-11-18 14:06:20 +07:00
|
|
|
err = tcp_transmit_skb(sk, syn_data, 1, sk->sk_allocation);
|
2012-07-19 13:43:07 +07:00
|
|
|
|
2014-11-18 14:06:20 +07:00
|
|
|
syn->skb_mstamp = syn_data->skb_mstamp;
|
2014-03-10 07:36:02 +07:00
|
|
|
|
2014-11-18 14:06:20 +07:00
|
|
|
/* Now full SYN+DATA was cloned and sent (or not),
|
|
|
|
* remove the SYN from the original skb (syn_data)
|
|
|
|
* we keep in write queue in case of a retransmit, as we
|
|
|
|
* also have the SYN packet (with no data) in the same queue.
|
|
|
|
*/
|
|
|
|
TCP_SKB_CB(syn_data)->seq++;
|
|
|
|
TCP_SKB_CB(syn_data)->tcp_flags = TCPHDR_ACK | TCPHDR_PSH;
|
|
|
|
if (!err) {
|
2012-07-19 13:43:11 +07:00
|
|
|
tp->syn_data = (fo->copied > 0);
|
2014-03-04 03:31:36 +07:00
|
|
|
NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPORIGDATASENT);
|
2012-07-19 13:43:07 +07:00
|
|
|
goto done;
|
|
|
|
}
|
|
|
|
|
|
|
|
fallback:
|
|
|
|
/* Send a regular SYN with Fast Open cookie request option */
|
|
|
|
if (fo->cookie.len > 0)
|
|
|
|
fo->cookie.len = 0;
|
|
|
|
err = tcp_transmit_skb(sk, syn, 1, sk->sk_allocation);
|
|
|
|
if (err)
|
|
|
|
tp->syn_fastopen = 0;
|
|
|
|
done:
|
|
|
|
fo->cookie.len = -1; /* Exclude Fast Open option for SYN retries */
|
|
|
|
return err;
|
|
|
|
}
|
|
|
|
|
2009-07-22 06:00:40 +07:00
|
|
|
/* Build a SYN and send it off. */
|
2005-04-17 05:20:36 +07:00
|
|
|
int tcp_connect(struct sock *sk)
|
|
|
|
{
|
|
|
|
struct tcp_sock *tp = tcp_sk(sk);
|
|
|
|
struct sk_buff *buff;
|
2010-11-16 18:52:49 +07:00
|
|
|
int err;
|
2005-04-17 05:20:36 +07:00
|
|
|
|
|
|
|
tcp_connect_init(sk);
|
|
|
|
|
2012-11-22 08:13:58 +07:00
|
|
|
if (unlikely(tp->repair)) {
|
|
|
|
tcp_finish_connect(sk, NULL);
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2015-05-20 03:26:55 +07:00
|
|
|
buff = sk_stream_alloc_skb(sk, 0, sk->sk_allocation, true);
|
2014-11-18 14:06:20 +07:00
|
|
|
if (unlikely(!buff))
|
2005-04-17 05:20:36 +07:00
|
|
|
return -ENOBUFS;
|
|
|
|
|
2010-06-12 21:01:43 +07:00
|
|
|
tcp_init_nondata_skb(buff, tp->write_seq++, TCPHDR_SYN);
|
2014-09-06 05:33:33 +07:00
|
|
|
tp->retrans_stamp = tcp_time_stamp;
|
2012-07-19 13:43:07 +07:00
|
|
|
tcp_connect_queue_skb(sk, buff);
|
2014-09-29 18:08:30 +07:00
|
|
|
tcp_ecn_send_syn(sk, buff);
|
2005-04-17 05:20:36 +07:00
|
|
|
|
2012-07-19 13:43:07 +07:00
|
|
|
/* Send off SYN; include data in Fast Open. */
|
|
|
|
err = tp->fastopen_req ? tcp_send_syn_data(sk, buff) :
|
|
|
|
tcp_transmit_skb(sk, buff, 1, sk->sk_allocation);
|
2010-11-16 18:52:49 +07:00
|
|
|
if (err == -ECONNREFUSED)
|
|
|
|
return err;
|
2006-08-08 11:04:15 +07:00
|
|
|
|
|
|
|
/* We change tp->snd_nxt after the tcp_transmit_skb() call
|
|
|
|
* in order to make this packet get counted in tcpOutSegs.
|
|
|
|
*/
|
|
|
|
tp->snd_nxt = tp->write_seq;
|
|
|
|
tp->pushed_seq = tp->write_seq;
|
2008-07-17 10:22:04 +07:00
|
|
|
TCP_INC_STATS(sock_net(sk), TCP_MIB_ACTIVEOPENS);
|
2005-04-17 05:20:36 +07:00
|
|
|
|
|
|
|
/* Timer for repeating the SYN until an answer. */
|
2005-08-10 10:11:08 +07:00
|
|
|
inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
|
|
|
|
inet_csk(sk)->icsk_rto, TCP_RTO_MAX);
|
2005-04-17 05:20:36 +07:00
|
|
|
return 0;
|
|
|
|
}
|
2010-07-10 04:22:10 +07:00
|
|
|
EXPORT_SYMBOL(tcp_connect);
|
2005-04-17 05:20:36 +07:00
|
|
|
|
|
|
|
/* Send out a delayed ack, the caller does the policy checking
|
|
|
|
* to see if we should even be here. See tcp_input.c:tcp_ack_snd_check()
|
|
|
|
* for details.
|
|
|
|
*/
|
|
|
|
void tcp_send_delayed_ack(struct sock *sk)
|
|
|
|
{
|
2005-08-10 10:10:42 +07:00
|
|
|
struct inet_connection_sock *icsk = inet_csk(sk);
|
|
|
|
int ato = icsk->icsk_ack.ato;
|
2005-04-17 05:20:36 +07:00
|
|
|
unsigned long timeout;
|
|
|
|
|
2014-09-27 03:37:35 +07:00
|
|
|
tcp_ca_event(sk, CA_EVENT_DELAYED_ACK);
|
|
|
|
|
2005-04-17 05:20:36 +07:00
|
|
|
if (ato > TCP_DELACK_MIN) {
|
2005-08-10 10:10:42 +07:00
|
|
|
const struct tcp_sock *tp = tcp_sk(sk);
|
2008-01-01 05:57:14 +07:00
|
|
|
int max_ato = HZ / 2;
|
2005-04-17 05:20:36 +07:00
|
|
|
|
2008-01-01 05:57:14 +07:00
|
|
|
if (icsk->icsk_ack.pingpong ||
|
|
|
|
(icsk->icsk_ack.pending & ICSK_ACK_PUSHED))
|
2005-04-17 05:20:36 +07:00
|
|
|
max_ato = TCP_DELACK_MAX;
|
|
|
|
|
|
|
|
/* Slow path, intersegment interval is "high". */
|
|
|
|
|
|
|
|
/* If some rtt estimate is known, use it to bound delayed ack.
|
2005-08-10 10:10:42 +07:00
|
|
|
* Do not use inet_csk(sk)->icsk_rto here, use results of rtt measurements
|
2005-04-17 05:20:36 +07:00
|
|
|
* directly.
|
|
|
|
*/
|
2014-02-27 05:02:48 +07:00
|
|
|
if (tp->srtt_us) {
|
|
|
|
int rtt = max_t(int, usecs_to_jiffies(tp->srtt_us >> 3),
|
|
|
|
TCP_DELACK_MIN);
|
2005-04-17 05:20:36 +07:00
|
|
|
|
|
|
|
if (rtt < max_ato)
|
|
|
|
max_ato = rtt;
|
|
|
|
}
|
|
|
|
|
|
|
|
ato = min(ato, max_ato);
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Stay within the limit we were given */
|
|
|
|
timeout = jiffies + ato;
|
|
|
|
|
|
|
|
/* Use new timeout only if there wasn't a older one earlier. */
|
2005-08-10 10:10:42 +07:00
|
|
|
if (icsk->icsk_ack.pending & ICSK_ACK_TIMER) {
|
2005-04-17 05:20:36 +07:00
|
|
|
/* If delack timer was blocked or is about to expire,
|
|
|
|
* send ACK now.
|
|
|
|
*/
|
2005-08-10 10:10:42 +07:00
|
|
|
if (icsk->icsk_ack.blocked ||
|
|
|
|
time_before_eq(icsk->icsk_ack.timeout, jiffies + (ato >> 2))) {
|
2005-04-17 05:20:36 +07:00
|
|
|
tcp_send_ack(sk);
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
2005-08-10 10:10:42 +07:00
|
|
|
if (!time_before(timeout, icsk->icsk_ack.timeout))
|
|
|
|
timeout = icsk->icsk_ack.timeout;
|
2005-04-17 05:20:36 +07:00
|
|
|
}
|
2005-08-10 10:10:42 +07:00
|
|
|
icsk->icsk_ack.pending |= ICSK_ACK_SCHED | ICSK_ACK_TIMER;
|
|
|
|
icsk->icsk_ack.timeout = timeout;
|
|
|
|
sk_reset_timer(sk, &icsk->icsk_delack_timer, timeout);
|
2005-04-17 05:20:36 +07:00
|
|
|
}
|
|
|
|
|
|
|
|
/* This routine sends an ack and also updates the window. */
|
|
|
|
void tcp_send_ack(struct sock *sk)
|
|
|
|
{
|
2007-12-31 19:51:11 +07:00
|
|
|
struct sk_buff *buff;
|
2005-04-17 05:20:36 +07:00
|
|
|
|
2007-12-31 19:51:11 +07:00
|
|
|
/* If we have been reset, we may not send again. */
|
|
|
|
if (sk->sk_state == TCP_CLOSE)
|
|
|
|
return;
|
2005-04-17 05:20:36 +07:00
|
|
|
|
2014-09-27 03:37:35 +07:00
|
|
|
tcp_ca_event(sk, CA_EVENT_NON_DELAYED_ACK);
|
|
|
|
|
2007-12-31 19:51:11 +07:00
|
|
|
/* We are not putting this on the write queue, so
|
|
|
|
* tcp_transmit_skb() will set the ownership to this
|
|
|
|
* sock.
|
|
|
|
*/
|
2015-11-30 23:57:28 +07:00
|
|
|
buff = alloc_skb(MAX_TCP_HEADER,
|
|
|
|
sk_gfp_mask(sk, GFP_ATOMIC | __GFP_NOWARN));
|
|
|
|
if (unlikely(!buff)) {
|
2007-12-31 19:51:11 +07:00
|
|
|
inet_csk_schedule_ack(sk);
|
|
|
|
inet_csk(sk)->icsk_ack.ato = TCP_ATO_MIN;
|
|
|
|
inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
|
|
|
|
TCP_DELACK_MAX, TCP_RTO_MAX);
|
|
|
|
return;
|
2005-04-17 05:20:36 +07:00
|
|
|
}
|
2007-12-31 19:51:11 +07:00
|
|
|
|
|
|
|
/* Reserve space for headers and prepare control bits. */
|
|
|
|
skb_reserve(buff, MAX_TCP_HEADER);
|
2010-06-12 21:01:43 +07:00
|
|
|
tcp_init_nondata_skb(buff, tcp_acceptable_seq(sk), TCPHDR_ACK);
|
2007-12-31 19:51:11 +07:00
|
|
|
|
tcp: do not pace pure ack packets
When we added pacing to TCP, we decided to let sch_fq take care
of actual pacing.
All TCP had to do was to compute sk->pacing_rate using simple formula:
sk->pacing_rate = 2 * cwnd * mss / rtt
It works well for senders (bulk flows), but not very well for receivers
or even RPC :
cwnd on the receiver can be less than 10, rtt can be around 100ms, so we
can end up pacing ACK packets, slowing down the sender.
Really, only the sender should pace, according to its own logic.
Instead of adding a new bit in skb, or call yet another flow
dissection, we tweak skb->truesize to a small value (2), and
we instruct sch_fq to use new helper and not pace pure ack.
Note this also helps TCP small queue, as ack packets present
in qdisc/NIC do not prevent sending a data packet (RPC workload)
This helps to reduce tx completion overhead, ack packets can use regular
sock_wfree() instead of tcp_wfree() which is a bit more expensive.
This has no impact in the case packets are sent to loopback interface,
as we do not coalesce ack packets (were we would detect skb->truesize
lie)
In case netem (with a delay) is used, skb_orphan_partial() also sets
skb->truesize to 1.
This patch is a combination of two patches we used for about one year at
Google.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2015-02-04 09:31:53 +07:00
|
|
|
/* We do not want pure acks influencing TCP Small Queues or fq/pacing
|
|
|
|
* too much.
|
|
|
|
* SKB_TRUESIZE(max(1 .. 66, MAX_TCP_HEADER)) is unfortunately ~784
|
|
|
|
* We also avoid tcp_wfree() overhead (cache line miss accessing
|
|
|
|
* tp->tsq_flags) by using regular sock_wfree()
|
|
|
|
*/
|
|
|
|
skb_set_tcp_pure_ack(buff);
|
|
|
|
|
2007-12-31 19:51:11 +07:00
|
|
|
/* Send it off, this clears delayed acks for us. */
|
2014-09-06 05:33:33 +07:00
|
|
|
skb_mstamp_get(&buff->skb_mstamp);
|
2015-11-30 23:57:28 +07:00
|
|
|
tcp_transmit_skb(sk, buff, 0, (__force gfp_t)0);
|
2005-04-17 05:20:36 +07:00
|
|
|
}
|
net: tcp: add DCTCP congestion control algorithm
This work adds the DataCenter TCP (DCTCP) congestion control
algorithm [1], which has been first published at SIGCOMM 2010 [2],
resp. follow-up analysis at SIGMETRICS 2011 [3] (and also, more
recently as an informational IETF draft available at [4]).
DCTCP is an enhancement to the TCP congestion control algorithm for
data center networks. Typical data center workloads are i.e.
i) partition/aggregate (queries; bursty, delay sensitive), ii) short
messages e.g. 50KB-1MB (for coordination and control state; delay
sensitive), and iii) large flows e.g. 1MB-100MB (data update;
throughput sensitive). DCTCP has therefore been designed for such
environments to provide/achieve the following three requirements:
* High burst tolerance (incast due to partition/aggregate)
* Low latency (short flows, queries)
* High throughput (continuous data updates, large file
transfers) with commodity, shallow buffered switches
The basic idea of its design consists of two fundamentals: i) on the
switch side, packets are being marked when its internal queue
length > threshold K (K is chosen so that a large enough headroom
for marked traffic is still available in the switch queue); ii) the
sender/host side maintains a moving average of the fraction of marked
packets, so each RTT, F is being updated as follows:
F := X / Y, where X is # of marked ACKs, Y is total # of ACKs
alpha := (1 - g) * alpha + g * F, where g is a smoothing constant
The resulting alpha (iow: probability that switch queue is congested)
is then being used in order to adaptively decrease the congestion
window W:
W := (1 - (alpha / 2)) * W
The means for receiving marked packets resp. marking them on switch
side in DCTCP is the use of ECN.
RFC3168 describes a mechanism for using Explicit Congestion Notification
from the switch for early detection of congestion, rather than waiting
for segment loss to occur.
However, this method only detects the presence of congestion, not
the *extent*. In the presence of mild congestion, it reduces the TCP
congestion window too aggressively and unnecessarily affects the
throughput of long flows [4].
DCTCP, as mentioned, enhances Explicit Congestion Notification (ECN)
processing to estimate the fraction of bytes that encounter congestion,
rather than simply detecting that some congestion has occurred. DCTCP
then scales the TCP congestion window based on this estimate [4],
thus it can derive multibit feedback from the information present in
the single-bit sequence of marks in its control law. And thus act in
*proportion* to the extent of congestion, not its *presence*.
Switches therefore set the Congestion Experienced (CE) codepoint in
packets when internal queue lengths exceed threshold K. Resulting,
DCTCP delivers the same or better throughput than normal TCP, while
using 90% less buffer space.
It was found in [2] that DCTCP enables the applications to handle 10x
the current background traffic, without impacting foreground traffic.
Moreover, a 10x increase in foreground traffic did not cause any
timeouts, and thus largely eliminates TCP incast collapse problems.
The algorithm itself has already seen deployments in large production
data centers since then.
We did a long-term stress-test and analysis in a data center, short
summary of our TCP incast tests with iperf compared to cubic:
This test measured DCTCP throughput and latency and compared it with
CUBIC throughput and latency for an incast scenario. In this test, 19
senders sent at maximum rate to a single receiver. The receiver simply
ran iperf -s.
The senders ran iperf -c <receiver> -t 30. All senders started
simultaneously (using local clocks synchronized by ntp).
This test was repeated multiple times. Below shows the results from a
single test. Other tests are similar. (DCTCP results were extremely
consistent, CUBIC results show some variance induced by the TCP timeouts
that CUBIC encountered.)
For this test, we report statistics on the number of TCP timeouts,
flow throughput, and traffic latency.
1) Timeouts (total over all flows, and per flow summaries):
CUBIC DCTCP
Total 3227 25
Mean 169.842 1.316
Median 183 1
Max 207 5
Min 123 0
Stddev 28.991 1.600
Timeout data is taken by measuring the net change in netstat -s
"other TCP timeouts" reported. As a result, the timeout measurements
above are not restricted to the test traffic, and we believe that it
is likely that all of the "DCTCP timeouts" are actually timeouts for
non-test traffic. We report them nevertheless. CUBIC will also include
some non-test timeouts, but they are drawfed by bona fide test traffic
timeouts for CUBIC. Clearly DCTCP does an excellent job of preventing
TCP timeouts. DCTCP reduces timeouts by at least two orders of
magnitude and may well have eliminated them in this scenario.
2) Throughput (per flow in Mbps):
CUBIC DCTCP
Mean 521.684 521.895
Median 464 523
Max 776 527
Min 403 519
Stddev 105.891 2.601
Fairness 0.962 0.999
Throughput data was simply the average throughput for each flow
reported by iperf. By avoiding TCP timeouts, DCTCP is able to
achieve much better per-flow results. In CUBIC, many flows
experience TCP timeouts which makes flow throughput unpredictable and
unfair. DCTCP, on the other hand, provides very clean predictable
throughput without incurring TCP timeouts. Thus, the standard deviation
of CUBIC throughput is dramatically higher than the standard deviation
of DCTCP throughput.
Mean throughput is nearly identical because even though cubic flows
suffer TCP timeouts, other flows will step in and fill the unused
bandwidth. Note that this test is something of a best case scenario
for incast under CUBIC: it allows other flows to fill in for flows
experiencing a timeout. Under situations where the receiver is issuing
requests and then waiting for all flows to complete, flows cannot fill
in for timed out flows and throughput will drop dramatically.
3) Latency (in ms):
CUBIC DCTCP
Mean 4.0088 0.04219
Median 4.055 0.0395
Max 4.2 0.085
Min 3.32 0.028
Stddev 0.1666 0.01064
Latency for each protocol was computed by running "ping -i 0.2
<receiver>" from a single sender to the receiver during the incast
test. For DCTCP, "ping -Q 0x6 -i 0.2 <receiver>" was used to ensure
that traffic traversed the DCTCP queue and was not dropped when the
queue size was greater than the marking threshold. The summary
statistics above are over all ping metrics measured between the single
sender, receiver pair.
The latency results for this test show a dramatic difference between
CUBIC and DCTCP. CUBIC intentionally overflows the switch buffer
which incurs the maximum queue latency (more buffer memory will lead
to high latency.) DCTCP, on the other hand, deliberately attempts to
keep queue occupancy low. The result is a two orders of magnitude
reduction of latency with DCTCP - even with a switch with relatively
little RAM. Switches with larger amounts of RAM will incur increasing
amounts of latency for CUBIC, but not for DCTCP.
4) Convergence and stability test:
This test measured the time that DCTCP took to fairly redistribute
bandwidth when a new flow commences. It also measured DCTCP's ability
to remain stable at a fair bandwidth distribution. DCTCP is compared
with CUBIC for this test.
At the commencement of this test, a single flow is sending at maximum
rate (near 10 Gbps) to a single receiver. One second after that first
flow commences, a new flow from a distinct server begins sending to
the same receiver as the first flow. After the second flow has sent
data for 10 seconds, the second flow is terminated. The first flow
sends for an additional second. Ideally, the bandwidth would be evenly
shared as soon as the second flow starts, and recover as soon as it
stops.
The results of this test are shown below. Note that the flow bandwidth
for the two flows was measured near the same time, but not
simultaneously.
DCTCP performs nearly perfectly within the measurement limitations
of this test: bandwidth is quickly distributed fairly between the two
flows, remains stable throughout the duration of the test, and
recovers quickly. CUBIC, in contrast, is slow to divide the bandwidth
fairly, and has trouble remaining stable.
CUBIC DCTCP
Seconds Flow 1 Flow 2 Seconds Flow 1 Flow 2
0 9.93 0 0 9.92 0
0.5 9.87 0 0.5 9.86 0
1 8.73 2.25 1 6.46 4.88
1.5 7.29 2.8 1.5 4.9 4.99
2 6.96 3.1 2 4.92 4.94
2.5 6.67 3.34 2.5 4.93 5
3 6.39 3.57 3 4.92 4.99
3.5 6.24 3.75 3.5 4.94 4.74
4 6 3.94 4 5.34 4.71
4.5 5.88 4.09 4.5 4.99 4.97
5 5.27 4.98 5 4.83 5.01
5.5 4.93 5.04 5.5 4.89 4.99
6 4.9 4.99 6 4.92 5.04
6.5 4.93 5.1 6.5 4.91 4.97
7 4.28 5.8 7 4.97 4.97
7.5 4.62 4.91 7.5 4.99 4.82
8 5.05 4.45 8 5.16 4.76
8.5 5.93 4.09 8.5 4.94 4.98
9 5.73 4.2 9 4.92 5.02
9.5 5.62 4.32 9.5 4.87 5.03
10 6.12 3.2 10 4.91 5.01
10.5 6.91 3.11 10.5 4.87 5.04
11 8.48 0 11 8.49 4.94
11.5 9.87 0 11.5 9.9 0
SYN/ACK ECT test:
This test demonstrates the importance of ECT on SYN and SYN-ACK packets
by measuring the connection probability in the presence of competing
flows for a DCTCP connection attempt *without* ECT in the SYN packet.
The test was repeated five times for each number of competing flows.
Competing Flows 1 | 2 | 4 | 8 | 16
------------------------------
Mean Connection Probability 1 | 0.67 | 0.45 | 0.28 | 0
Median Connection Probability 1 | 0.65 | 0.45 | 0.25 | 0
As the number of competing flows moves beyond 1, the connection
probability drops rapidly.
Enabling DCTCP with this patch requires the following steps:
DCTCP must be running both on the sender and receiver side in your
data center, i.e.:
sysctl -w net.ipv4.tcp_congestion_control=dctcp
Also, ECN functionality must be enabled on all switches in your
data center for DCTCP to work. The default ECN marking threshold (K)
heuristic on the switch for DCTCP is e.g., 20 packets (30KB) at
1Gbps, and 65 packets (~100KB) at 10Gbps (K > 1/7 * C * RTT, [4]).
In above tests, for each switch port, traffic was segregated into two
queues. For any packet with a DSCP of 0x01 - or equivalently a TOS of
0x04 - the packet was placed into the DCTCP queue. All other packets
were placed into the default drop-tail queue. For the DCTCP queue,
RED/ECN marking was enabled, here, with a marking threshold of 75 KB.
More details however, we refer you to the paper [2] under section 3).
There are no code changes required to applications running in user
space. DCTCP has been implemented in full *isolation* of the rest of
the TCP code as its own congestion control module, so that it can run
without a need to expose code to the core of the TCP stack, and thus
nothing changes for non-DCTCP users.
Changes in the CA framework code are minimal, and DCTCP algorithm
operates on mechanisms that are already available in most Silicon.
The gain (dctcp_shift_g) is currently a fixed constant (1/16) from
the paper, but we leave the option that it can be chosen carefully
to a different value by the user.
In case DCTCP is being used and ECN support on peer site is off,
DCTCP falls back after 3WHS to operate in normal TCP Reno mode.
ss {-4,-6} -t -i diag interface:
... dctcp wscale:7,7 rto:203 rtt:2.349/0.026 mss:1448 cwnd:2054
ssthresh:1102 ce_state 0 alpha 15 ab_ecn 0 ab_tot 735584
send 10129.2Mbps pacing_rate 20254.1Mbps unacked:1822 retrans:0/15
reordering:101 rcv_space:29200
... dctcp-reno wscale:7,7 rto:201 rtt:0.711/1.327 ato:40 mss:1448
cwnd:10 ssthresh:1102 fallback_mode send 162.9Mbps pacing_rate
325.5Mbps rcv_rtt:1.5 rcv_space:29200
More information about DCTCP can be found in [1-4].
[1] http://simula.stanford.edu/~alizade/Site/DCTCP.html
[2] http://simula.stanford.edu/~alizade/Site/DCTCP_files/dctcp-final.pdf
[3] http://simula.stanford.edu/~alizade/Site/DCTCP_files/dctcp_analysis-full.pdf
[4] http://tools.ietf.org/html/draft-bensley-tcpm-dctcp-00
Joint work with Florian Westphal and Glenn Judd.
Signed-off-by: Daniel Borkmann <dborkman@redhat.com>
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Glenn Judd <glenn.judd@morganstanley.com>
Acked-by: Stephen Hemminger <stephen@networkplumber.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
2014-09-27 03:37:36 +07:00
|
|
|
EXPORT_SYMBOL_GPL(tcp_send_ack);
|
2005-04-17 05:20:36 +07:00
|
|
|
|
|
|
|
/* This routine sends a packet with an out of date sequence
|
|
|
|
* number. It assumes the other end will try to ack it.
|
|
|
|
*
|
|
|
|
* Question: what should we make while urgent mode?
|
|
|
|
* 4.4BSD forces sending single byte of data. We cannot send
|
|
|
|
* out of window data, because we have SND.NXT==SND.MAX...
|
|
|
|
*
|
|
|
|
* Current solution: to send TWO zero-length segments in urgent mode:
|
|
|
|
* one is with SEG.SEQ=SND.UNA to deliver urgent pointer, another is
|
|
|
|
* out-of-date with SND.UNA-1 to probe window.
|
|
|
|
*/
|
2015-05-07 04:26:25 +07:00
|
|
|
static int tcp_xmit_probe_skb(struct sock *sk, int urgent, int mib)
|
2005-04-17 05:20:36 +07:00
|
|
|
{
|
|
|
|
struct tcp_sock *tp = tcp_sk(sk);
|
|
|
|
struct sk_buff *skb;
|
|
|
|
|
|
|
|
/* We don't queue it, tcp_transmit_skb() sets ownership. */
|
2015-11-30 23:57:28 +07:00
|
|
|
skb = alloc_skb(MAX_TCP_HEADER,
|
|
|
|
sk_gfp_mask(sk, GFP_ATOMIC | __GFP_NOWARN));
|
2015-04-03 15:17:26 +07:00
|
|
|
if (!skb)
|
2005-04-17 05:20:36 +07:00
|
|
|
return -1;
|
|
|
|
|
|
|
|
/* Reserve space for headers and set control bits. */
|
|
|
|
skb_reserve(skb, MAX_TCP_HEADER);
|
|
|
|
/* Use a previous sequence. This should cause the other
|
|
|
|
* end to send an ack. Don't queue or clone SKB, just
|
|
|
|
* send it.
|
|
|
|
*/
|
2010-06-12 21:01:43 +07:00
|
|
|
tcp_init_nondata_skb(skb, tp->snd_una - !urgent, TCPHDR_ACK);
|
2014-09-06 05:33:33 +07:00
|
|
|
skb_mstamp_get(&skb->skb_mstamp);
|
2015-10-20 03:51:34 +07:00
|
|
|
NET_INC_STATS(sock_net(sk), mib);
|
2015-11-30 23:57:28 +07:00
|
|
|
return tcp_transmit_skb(sk, skb, 0, (__force gfp_t)0);
|
2005-04-17 05:20:36 +07:00
|
|
|
}
|
|
|
|
|
2012-04-19 10:40:39 +07:00
|
|
|
void tcp_send_window_probe(struct sock *sk)
|
|
|
|
{
|
|
|
|
if (sk->sk_state == TCP_ESTABLISHED) {
|
|
|
|
tcp_sk(sk)->snd_wl1 = tcp_sk(sk)->rcv_nxt - 1;
|
2015-05-07 04:26:25 +07:00
|
|
|
tcp_xmit_probe_skb(sk, 0, LINUX_MIB_TCPWINPROBE);
|
2012-04-19 10:40:39 +07:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2009-07-22 06:00:40 +07:00
|
|
|
/* Initiate keepalive or window probe from timer. */
|
2015-05-07 04:26:25 +07:00
|
|
|
int tcp_write_wakeup(struct sock *sk, int mib)
|
2005-04-17 05:20:36 +07:00
|
|
|
{
|
2007-12-31 19:51:11 +07:00
|
|
|
struct tcp_sock *tp = tcp_sk(sk);
|
|
|
|
struct sk_buff *skb;
|
2005-04-17 05:20:36 +07:00
|
|
|
|
2007-12-31 19:51:11 +07:00
|
|
|
if (sk->sk_state == TCP_CLOSE)
|
|
|
|
return -1;
|
|
|
|
|
2015-04-03 15:17:27 +07:00
|
|
|
skb = tcp_send_head(sk);
|
|
|
|
if (skb && before(TCP_SKB_CB(skb)->seq, tcp_wnd_end(tp))) {
|
2007-12-31 19:51:11 +07:00
|
|
|
int err;
|
2009-03-14 21:23:05 +07:00
|
|
|
unsigned int mss = tcp_current_mss(sk);
|
2007-12-31 19:51:11 +07:00
|
|
|
unsigned int seg_size = tcp_wnd_end(tp) - TCP_SKB_CB(skb)->seq;
|
|
|
|
|
|
|
|
if (before(tp->pushed_seq, TCP_SKB_CB(skb)->end_seq))
|
|
|
|
tp->pushed_seq = TCP_SKB_CB(skb)->end_seq;
|
|
|
|
|
|
|
|
/* We are probing the opening of a window
|
|
|
|
* but the window size is != 0
|
|
|
|
* must have been a result SWS avoidance ( sender )
|
|
|
|
*/
|
|
|
|
if (seg_size < TCP_SKB_CB(skb)->end_seq - TCP_SKB_CB(skb)->seq ||
|
|
|
|
skb->len > mss) {
|
|
|
|
seg_size = min(seg_size, mss);
|
2011-09-28 00:25:05 +07:00
|
|
|
TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_PSH;
|
2014-06-06 21:32:37 +07:00
|
|
|
if (tcp_fragment(sk, skb, seg_size, mss, GFP_ATOMIC))
|
2007-12-31 19:51:11 +07:00
|
|
|
return -1;
|
|
|
|
} else if (!tcp_skb_pcount(skb))
|
2015-06-11 23:15:17 +07:00
|
|
|
tcp_set_skb_tso_segs(skb, mss);
|
2007-12-31 19:51:11 +07:00
|
|
|
|
2011-09-28 00:25:05 +07:00
|
|
|
TCP_SKB_CB(skb)->tcp_flags |= TCPHDR_PSH;
|
2007-12-31 19:51:11 +07:00
|
|
|
err = tcp_transmit_skb(sk, skb, 1, GFP_ATOMIC);
|
|
|
|
if (!err)
|
|
|
|
tcp_event_new_data_sent(sk, skb);
|
|
|
|
return err;
|
|
|
|
} else {
|
2008-10-08 04:43:06 +07:00
|
|
|
if (between(tp->snd_up, tp->snd_una + 1, tp->snd_una + 0xFFFF))
|
2015-05-07 04:26:25 +07:00
|
|
|
tcp_xmit_probe_skb(sk, 1, mib);
|
|
|
|
return tcp_xmit_probe_skb(sk, 0, mib);
|
2005-04-17 05:20:36 +07:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/* A window probe timeout has occurred. If window is not closed send
|
|
|
|
* a partial packet else a zero probe.
|
|
|
|
*/
|
|
|
|
void tcp_send_probe0(struct sock *sk)
|
|
|
|
{
|
2005-08-10 10:10:42 +07:00
|
|
|
struct inet_connection_sock *icsk = inet_csk(sk);
|
2005-04-17 05:20:36 +07:00
|
|
|
struct tcp_sock *tp = tcp_sk(sk);
|
2016-02-03 14:46:54 +07:00
|
|
|
struct net *net = sock_net(sk);
|
2014-09-23 03:19:44 +07:00
|
|
|
unsigned long probe_max;
|
2005-04-17 05:20:36 +07:00
|
|
|
int err;
|
|
|
|
|
2015-05-07 04:26:25 +07:00
|
|
|
err = tcp_write_wakeup(sk, LINUX_MIB_TCPWINPROBE);
|
2005-04-17 05:20:36 +07:00
|
|
|
|
2007-03-08 03:12:44 +07:00
|
|
|
if (tp->packets_out || !tcp_send_head(sk)) {
|
2005-04-17 05:20:36 +07:00
|
|
|
/* Cancel probe timer, if it is not required. */
|
2005-08-10 14:03:31 +07:00
|
|
|
icsk->icsk_probes_out = 0;
|
2005-08-10 10:10:42 +07:00
|
|
|
icsk->icsk_backoff = 0;
|
2005-04-17 05:20:36 +07:00
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (err <= 0) {
|
2016-02-03 14:46:54 +07:00
|
|
|
if (icsk->icsk_backoff < net->ipv4.sysctl_tcp_retries2)
|
2005-08-10 10:10:42 +07:00
|
|
|
icsk->icsk_backoff++;
|
2005-08-10 14:03:31 +07:00
|
|
|
icsk->icsk_probes_out++;
|
2014-09-23 03:19:44 +07:00
|
|
|
probe_max = TCP_RTO_MAX;
|
2005-04-17 05:20:36 +07:00
|
|
|
} else {
|
|
|
|
/* If packet was not sent due to local congestion,
|
2005-08-10 14:03:31 +07:00
|
|
|
* do not backoff and do not remember icsk_probes_out.
|
2005-04-17 05:20:36 +07:00
|
|
|
* Let local senders to fight for local resources.
|
|
|
|
*
|
|
|
|
* Use accumulated backoff yet.
|
|
|
|
*/
|
2005-08-10 14:03:31 +07:00
|
|
|
if (!icsk->icsk_probes_out)
|
|
|
|
icsk->icsk_probes_out = 1;
|
2014-09-23 03:19:44 +07:00
|
|
|
probe_max = TCP_RESOURCE_PROBE_INTERVAL;
|
2005-04-17 05:20:36 +07:00
|
|
|
}
|
2014-09-23 03:19:44 +07:00
|
|
|
inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
|
tcp: adjust window probe timers to safer values
With the advent of small rto timers in datacenter TCP,
(ip route ... rto_min x), the following can happen :
1) Qdisc is full, transmit fails.
TCP sets a timer based on icsk_rto to retry the transmit, without
exponential backoff.
With low icsk_rto, and lot of sockets, all cpus are servicing timer
interrupts like crazy.
Intent of the code was to retry with a timer between 200 (TCP_RTO_MIN)
and 500ms (TCP_RESOURCE_PROBE_INTERVAL)
2) Receivers can send zero windows if they don't drain their receive queue.
TCP sends zero window probes, based on icsk_rto current value, with
exponential backoff.
With /proc/sys/net/ipv4/tcp_retries2 being 15 (or even smaller in
some cases), sender can abort in less than one or two minutes !
If receiver stops the sender, it obviously doesn't care of very tight
rto. Probability of dropping the ACK reopening the window is not
worth the risk.
Lets change the base timer to be at least 200ms (TCP_RTO_MIN) for these
events (but not normal RTO based retransmits)
A followup patch adds a new SNMP counter, as it would have helped a lot
diagnosing this issue.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2015-05-07 04:26:24 +07:00
|
|
|
tcp_probe0_when(sk, probe_max),
|
2014-09-23 03:19:44 +07:00
|
|
|
TCP_RTO_MAX);
|
2005-04-17 05:20:36 +07:00
|
|
|
}
|
2014-06-25 21:09:59 +07:00
|
|
|
|
2015-09-25 21:39:23 +07:00
|
|
|
int tcp_rtx_synack(const struct sock *sk, struct request_sock *req)
|
2014-06-25 21:09:59 +07:00
|
|
|
{
|
|
|
|
const struct tcp_request_sock_ops *af_ops = tcp_rsk(req)->af_specific;
|
|
|
|
struct flowi fl;
|
|
|
|
int res;
|
|
|
|
|
2015-09-16 05:24:20 +07:00
|
|
|
tcp_rsk(req)->txhash = net_tx_rndhash();
|
2015-10-17 03:00:01 +07:00
|
|
|
res = af_ops->send_synack(sk, NULL, &fl, req, NULL, true);
|
2014-06-25 21:09:59 +07:00
|
|
|
if (!res) {
|
|
|
|
TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_RETRANSSEGS);
|
|
|
|
NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPSYNRETRANS);
|
|
|
|
}
|
|
|
|
return res;
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(tcp_rtx_synack);
|