With EDT model, SRTT no longer is inflated by pacing delays.
This means that RTO and some other xmit timers might be setup
incorrectly. This is particularly visible with either :
- Very small enforced pacing rates (SO_MAX_PACING_RATE)
- Reduced rto (from the default 200 ms)
This can lead to TCP flows aborts in the worst case,
or spurious retransmits in other cases.
For example, this session gets far more throughput
than the requested 80kbit :
$ netperf -H 127.0.0.2 -l 100 -- -q 10000
MIGRATED TCP STREAM TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 127.0.0.2 () port 0 AF_INET
Recv Send Send
Socket Socket Message Elapsed
Size Size Size Time Throughput
bytes bytes bytes secs. 10^6bits/sec
540000 262144 262144 104.00 2.66
With the fix :
$ netperf -H 127.0.0.2 -l 100 -- -q 10000
MIGRATED TCP STREAM TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 127.0.0.2 () port 0 AF_INET
Recv Send Send
Socket Socket Message Elapsed
Size Size Size Time Throughput
bytes bytes bytes secs. 10^6bits/sec
540000 262144 262144 104.00 0.12
EDT allows for better control of rtx timers, since TCP has
a better idea of the earliest departure time of each skb
in the rtx queue. We only have to eventually add to the
timer the difference of the EDT time with current time.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Andrey reported the following warning triggered while running CRIU tests:
tcp_clean_rtx_queue()
...
last_ackt = tcp_skb_timestamp_us(skb);
WARN_ON_ONCE(last_ackt == 0);
This is caused by 5f6188a800 ("tcp: do not change tcp_wstamp_ns
in tcp_mstamp_refresh"), as we end up having skbs in retransmit queue
with a zero skb->skb_mstamp_ns field.
We could fix this bug in different ways, like making sure
tp->tcp_wstamp_ns is not zero at socket creation, but as Neal pointed
out, we also do not want that pacing status of a repaired socket
could push tp->tcp_wstamp_ns far ahead in the future.
So we prefer changing tcp_write_xmit() to not call tcp_update_skb_after_send()
and instead do what is requested by TCP_REPAIR logic.
Fixes: 5f6188a800 ("tcp: do not change tcp_wstamp_ns in tcp_mstamp_refresh")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: Andrey Vagin <avagin@openvz.org>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When TCP implements its own pacing (when no fq packet scheduler is used),
it is arming high resolution timer after a packet is sent.
But in many cases (like TCP_RR kind of workloads), this high resolution
timer expires before the application attempts to write the following
packet. This overhead also happens when the flow is ACK clocked and
cwnd limited instead of being limited by the pacing rate.
This leads to extra overhead (high number of IRQ)
Now tcp_wstamp_ns is reserved for the pacing timer only
(after commit "tcp: do not change tcp_wstamp_ns in tcp_mstamp_refresh"),
we can setup the timer only when a packet is about to be sent,
and if tcp_wstamp_ns is in the future.
This leads to a ~10% performance increase in TCP_RR workloads.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In commit fefa569a9d ("net_sched: sch_fq: account for schedule/timers
drifts") we added a mitigation for scheduling jitter in fq packet scheduler.
This patch does the same in TCP stack, now it is using EDT model.
Note that this mitigation is valid for both external (fq packet scheduler)
or internal TCP pacing.
This uses the same strategy than the above commit, allowing
a time credit of half the packet currently sent.
Consider following case :
An skb is sent, after an idle period of 300 usec.
The air-time (skb->len/pacing_rate) is 500 usec
Instead of setting the pacing timer to now+500 usec,
it will use now+min(500/2, 300) -> now+250usec
This is like having a token bucket with a depth of half
an skb.
Tested:
tc qdisc replace dev eth0 root pfifo_fast
Before
netperf -P0 -H remote -- -q 1000000000 # 8000Mbit
540000 262144 262144 10.00 7710.43
After :
netperf -P0 -H remote -- -q 1000000000 # 8000 Mbit
540000 262144 262144 10.00 7999.75 # Much closer to 8000Mbit target
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
sk_pacing_rate has beed introduced as a u32 field in 2013,
effectively limiting per flow pacing to 34Gbit.
We believe it is time to allow TCP to pace high speed flows
on 64bit hosts, as we now can reach 100Gbit on one TCP flow.
This patch adds no cost for 32bit kernels.
The tcpi_pacing_rate and tcpi_max_pacing_rate were already
exported as 64bit, so iproute2/ss command require no changes.
Unfortunately the SO_MAX_PACING_RATE socket option will stay
32bit and we will need to add a new option to let applications
control high pacing rates.
State Recv-Q Send-Q Local Address:Port Peer Address:Port
ESTAB 0 1787144 10.246.9.76:49992 10.246.9.77:36741
timer:(on,003ms,0) ino:91863 sk:2 <->
skmem:(r0,rb540000,t66440,tb2363904,f605944,w1822984,o0,bl0,d0)
ts sack bbr wscale:8,8 rto:201 rtt:0.057/0.006 mss:1448
rcvmss:536 advmss:1448
cwnd:138 ssthresh:178 bytes_acked:256699822585 segs_out:177279177
segs_in:3916318 data_segs_out:177279175
bbr:(bw:31276.8Mbps,mrtt:0,pacing_gain:1.25,cwnd_gain:2)
send 28045.5Mbps lastrcv:73333
pacing_rate 38705.0Mbps delivery_rate 22997.6Mbps
busy:73333ms unacked:135 retrans:0/157 rcv_space:14480
notsent:2085120 minrtt:0.013
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In EDT design, I made the mistake of using tcp_wstamp_ns
to store the last tcp_clock_ns() sample and to store the
pacing virtual timer.
This causes major regressions at high speed flows.
Introduce tcp_clock_cache to store last tcp_clock_ns().
This is needed because some arches have slow high-resolution
kernel time service.
tcp_wstamp_ns is only updated when a packet is sent.
Note that we can remove tcp_mstamp in the future since
tcp_mstamp is essentially tcp_clock_cache/1000, so the
apparent socket size increase is temporary.
Fixes: 9799ccb0e9 ("tcp: add tcp_wstamp_ns socket field")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Previously TCP initial receive buffer is ~87KB by default and
the initial receive window is ~29KB (20 MSS). This patch changes
the two numbers to 128KB and ~64KB (rounding down to the multiples
of MSS) respectively. The patch also simplifies the calculations s.t.
the two numbers are directly controlled by sysctl tcp_rmem[1]:
1) Initial receiver buffer budget (sk_rcvbuf): while this should
be configured via sysctl tcp_rmem[1], previously tcp_fixup_rcvbuf()
always override and set a larger size when a new connection
establishes.
2) Initial receive window in SYN: previously it is set to 20
packets if MSS <= 1460. The number 20 was based on the initial
congestion window of 10: the receiver needs twice amount to
avoid being limited by the receive window upon out-of-order
delivery in the first window burst. But since this only
applies if the receiving MSS <= 1460, connection using large MTU
(e.g. to utilize receiver zero-copy) may be limited by the
receive window.
With this patch TCP memory configuration is more straight-forward and
more properly sized to modern high-speed networks by default. Several
popular stacks have been announcing 64KB rwin in SYNs as well.
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: Wei Wang <weiwan@google.com>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Now TCP keeps track of tcp_wstamp_ns, recording the earliest
departure time of next packet, we can remove duplicate code
from tcp_internal_pacing()
This removes one ktime_get_tai_ns() call, and a divide.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
TCP keeps track of tcp_wstamp_ns by itself, meaning sch_fq
no longer has to do it.
Thanks to this model, TCP can get more accurate RTT samples,
since pacing no longer inflates them.
This has the nice effect of removing some delays caused by FQ
quantum mechanism, causing inflated max/P99 latencies.
Also we might relax TCP Small Queue tight limits in the future,
since this new model allow TCP to build bigger batches, since
sch_fq (or a device with earliest departure time offload) ensure
these packets will be delivered on time.
Note that other protocols are not converted (they will probably
never be) so sch_fq has still support for SO_MAX_PACING_RATE
Tested:
Test showing FQ pacing quantum artifact for low-rate flows,
adding unexpected throttles for RPC flows, inflating max and P99 latencies.
The parameters chosen here are to show what happens typically when
a TCP flow has a reduced pacing rate (this can be caused by a reduced
cwin after few losses, or/and rtt above few ms)
MIBS="MIN_LATENCY,MEAN_LATENCY,MAX_LATENCY,P99_LATENCY,STDDEV_LATENCY"
Before :
$ netperf -H 10.246.7.133 -t TCP_RR -Cc -T6,6 -- -q 2000000 -r 100,100 -o $MIBS
MIGRATED TCP REQUEST/RESPONSE TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 10.246.7.133 () port 0 AF_INET : first burst 0 : cpu bind
Minimum Latency Microseconds,Mean Latency Microseconds,Maximum Latency Microseconds,99th Percentile Latency Microseconds,Stddev Latency Microseconds
19,82.78,5279,3825,482.02
After :
$ netperf -H 10.246.7.133 -t TCP_RR -Cc -T6,6 -- -q 2000000 -r 100,100 -o $MIBS
MIGRATED TCP REQUEST/RESPONSE TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 10.246.7.133 () port 0 AF_INET : first burst 0 : cpu bind
Minimum Latency Microseconds,Mean Latency Microseconds,Maximum Latency Microseconds,99th Percentile Latency Microseconds,Stddev Latency Microseconds
20,49.94,128,63,3.18
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Next patch will use tcp_wstamp_ns to feed internal
TCP pacing timer, so switch to CLOCK_TAI to share same base.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Switch internal TCP skb->skb_mstamp to skb->skb_mstamp_ns,
from usec units to nsec units.
Do not clear skb->tstamp before entering IP stacks in TX,
so that qdisc or devices can implement pacing based on the
earliest departure time instead of socket sk->sk_pacing_rate
Packets are fed with tcp_wstamp_ns, and following patch
will update tcp_wstamp_ns when both TCP and sch_fq switch to
the earliest departure time mechanism.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
TCP will soon provide earliest departure time on TX skbs.
It needs to track this in a new variable.
tcp_mstamp_refresh() needs to update this variable, and
became too big to stay an inline.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
There are few places where TCP reads skb->skb_mstamp expecting
a value in usec unit.
skb->tstamp (aka skb->skb_mstamp) will soon store CLOCK_TAI nsec value.
Add tcp_skb_timestamp_us() to provide proper conversion when needed.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Fixes gcc '-Wunused-but-set-variable' warning:
net/ipv4/tcp_output.c: In function 'tcp_collapse_retrans':
net/ipv4/tcp_output.c:2700:6: warning:
variable 'skb_size' set but not used [-Wunused-but-set-variable]
int skb_size, next_skb_size;
^
Signed-off-by: Wei Yongjun <weiyongjun1@huawei.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Introduce a new TCP stat to record the number of bytes retransmitted
(RFC4898 tcpEStatsPerfOctetsRetrans) and expose it in both tcp_info
(TCP_INFO) and opt_stats (SOF_TIMESTAMPING_OPT_STATS).
Signed-off-by: Wei Wang <weiwan@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Acked-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Introduce a new TCP stat to record the number of bytes sent
(RFC4898 tcpEStatsPerfHCDataOctetsOut) and expose it in both tcp_info
(TCP_INFO) and opt_stats (SOF_TIMESTAMPING_OPT_STATS).
Signed-off-by: Wei Wang <weiwan@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Acked-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Currently when a DCTCP receiver delays an ACK and receive a
data packet with a different CE mark from the previous one's, it
sends two immediate ACKs acking previous and latest sequences
respectly (for ECN accounting).
Previously sending the first ACK may mark off the delayed ACK timer
(tcp_event_ack_sent). This may subsequently prevent sending the
second ACK to acknowledge the latest sequence (tcp_ack_snd_check).
The culprit is that tcp_send_ack() assumes it always acknowleges
the latest sequence, which is not true for the first special ACK.
The fix is to not make the assumption in tcp_send_ack and check the
actual ack sequence before cancelling the delayed ACK. Further it's
safer to pass the ack sequence number as a local variable into
tcp_send_ack routine, instead of intercepting tp->rcv_nxt to avoid
future bugs like this.
Reported-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Refactor and create helpers to send the special ACK in DCTCP.
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
After fixing the way DCTCP tracking delayed ACKs, the delayed-ACK
related callbacks are no longer needed
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Acked-by: Lawrence Brakmo <brakmo@fb.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This commit makes BBR use only the MSS (without any headers) to
calculate pacing rates when internal TCP-layer pacing is used.
This is necessary to achieve the correct pacing behavior in this case,
since tcp_internal_pacing() uses only the payload length to calculate
pacing delays.
Signed-off-by: Kevin Yang <yyd@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
S390 bpf_jit.S is removed in net-next and had changes in 'net',
since that code isn't used any more take the removal.
TLS data structures split the TX and RX components in 'net-next',
put the new struct members from the bug fix in 'net' into the RX
part.
The 'net-next' tree had some reworking of how the ERSPAN code works in
the GRE tunneling code, overlapping with a one-line headroom
calculation fix in 'net'.
Overlapping changes in __sock_map_ctx_update_elem(), keep the bits
that read the prog members via READ_ONCE() into local variables
before using them.
Signed-off-by: David S. Miller <davem@davemloft.net>
This counter tracks number of ACK packets that the host has not sent,
thanks to ACK compression.
Sample output :
$ nstat -n;sleep 1;nstat|egrep "IpInReceives|IpOutRequests|TcpInSegs|TcpOutSegs|TcpExtTCPAckCompressed"
IpInReceives 123250 0.0
IpOutRequests 3684 0.0
TcpInSegs 123251 0.0
TcpOutSegs 3684 0.0
TcpExtTCPAckCompressed 119252 0.0
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When TCP receives an out-of-order packet, it immediately sends
a SACK packet, generating network load but also forcing the
receiver to send 1-MSS pathological packets, increasing its
RTX queue length/depth, and thus processing time.
Wifi networks suffer from this aggressive behavior, but generally
speaking, all these SACK packets add fuel to the fire when networks
are under congestion.
This patch adds a high resolution timer and tp->compressed_ack counter.
Instead of sending a SACK, we program this timer with a small delay,
based on RTT and capped to 1 ms :
delay = min ( 5 % of RTT, 1 ms)
If subsequent SACKs need to be sent while the timer has not yet
expired, we simply increment tp->compressed_ack.
When timer expires, a SACK is sent with the latest information.
Whenever an ACK is sent (if data is sent, or if in-order
data is received) timer is canceled.
Note that tcp_sack_new_ofo_skb() is able to force a SACK to be sent
if the sack blocks need to be shuffled, even if the timer has not
expired.
A new SNMP counter is added in the following patch.
Two other patches add sysctls to allow changing the 1,000,000 and 44
values that this commit hard-coded.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Acked-by: Yuchung Cheng <ycheng@google.com>
Acked-by: Toke Høiland-Jørgensen <toke@toke.dk>
Signed-off-by: David S. Miller <davem@davemloft.net>
linux-4.16 got support for softirq based hrtimers.
TCP can switch its pacing hrtimer to this variant, since this
avoids going through a tasklet and some atomic operations.
pacing timer logic looks like other (jiffies based) tcp timers.
v2: use hrtimer_try_to_cancel() in tcp_clear_xmit_timers()
to correctly release reference on socket if needed.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In tcp_select_initial_window(), we only set rcv_wnd to
tcp_default_init_rwnd() if current mss > (1 << wscale). Otherwise,
rcv_wnd is kept at the full receive space of the socket which is a
value way larger than tcp_default_init_rwnd().
With larger initial rcv_wnd value, receive buffer autotuning logic
takes longer to kick in and increase the receive buffer.
In a TCP throughput test where receiver has rmem[2] set to 125MB
(wscale is 11), we see the connection gets recvbuf limited at the
beginning of the connection and gets less throughput overall.
Signed-off-by: Wei Wang <weiwan@google.com>
Acked-by: Eric Dumazet <edumazet@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Acked-by: Yuchung Cheng <ycheng@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
RETPOLINE made calls to tp->af_specific->md5_lookup() quite expensive,
given they have no result.
We can omit the calls for sockets that have no md5 keys.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This is second part of dealing with suboptimal device gso parameters.
In first patch (350c9f484b "tcp_bbr: better deal with suboptimal GSO")
we dealt with devices having low gso_max_segs
Some devices lower gso_max_size from 64KB to 16 KB (r8152 is an example)
In order to probe an optimal cwnd, we want BBR being not sensitive
to whatever GSO constraint a device can have.
This patch removes tso_segs_goal() CC callback in favor of
min_tso_segs() for CC wanting to override sysctl_tcp_min_tso_segs
Next patch will remove bbr->tso_segs_goal since it does not have
to be persistent.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
BBR uses tcp_tso_autosize() in an attempt to probe what would be the
burst sizes and to adjust cwnd in bbr_target_cwnd() with following
gold formula :
/* Allow enough full-sized skbs in flight to utilize end systems. */
cwnd += 3 * bbr->tso_segs_goal;
But GSO can be lacking or be constrained to very small
units (ip link set dev ... gso_max_segs 2)
What we really want is to have enough packets in flight so that both
GSO and GRO are efficient.
So in the case GSO is off or downgraded, we still want to have the same
number of packets in flight as if GSO/TSO was fully operational, so
that GRO can hopefully be working efficiently.
To fix this issue, we make tcp_tso_autosize() unaware of
sk->sk_gso_max_segs
Only tcp_tso_segs() has to enforce the gso_max_segs limit.
Tested:
ethtool -K eth0 tso off gso off
tc qd replace dev eth0 root pfifo_fast
Before patch:
for f in {1..5}; do ./super_netperf 1 -H lpaa24 -- -K bbr; done
691 (ss -temoi shows cwnd is stuck around 6 )
667
651
631
517
After patch :
# for f in {1..5}; do ./super_netperf 1 -H lpaa24 -- -K bbr; done
1733 (ss -temoi shows cwnd is around 386 )
1778
1746
1781
1718
Fixes: 0f8782ea14 ("tcp_bbr: add BBR congestion control")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: Oleksandr Natalenko <oleksandr@natalenko.name>
Acked-by: Neal Cardwell <ncardwell@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Since all skbs in write/rtx queues have CHECKSUM_PARTIAL,
we can remove dead code.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
We no longer have skbs with skb->ip_summed == CHECKSUM_NONE
in TCP write queues.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Avoid SKB coalescing if eor bit is set in one of the relevant
SKBs.
Fixes: c134ecb878 ("tcp: Make use of MSG_EOR in tcp_sendmsg")
Signed-off-by: Ilya Lesokhin <ilyal@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Adds support for calling sock_ops BPF program when there is a
retransmission. Three arguments are used; one for the sequence number,
another for the number of segments retransmitted, and the last one for
the return value of tcp_transmit_skb (0 => success).
Does not include syn-ack retransmissions.
New op: BPF_SOCK_OPS_RETRANS_CB.
Signed-off-by: Lawrence Brakmo <brakmo@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Adds support for passing up to 4 arguments to sock_ops bpf functions. It
reusues the reply union, so the bpf_sock_ops structures are not
increased in size.
Signed-off-by: Lawrence Brakmo <brakmo@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
The two conditions triggering BUG_ON() are somewhat unrelated:
the tcp_skb_pcount() check is meant to catch TSO flaws, the
second one checks sanity of congestion window bookkeeping.
Split them into two separate BUG_ON() assertions on two lines,
so that we know which one actually triggers, when they do.
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch enables tail loss probe in cwnd reduction (CWR) state
to detect potential losses. Prior to this patch, since the sender
uses PRR to determine the cwnd in CWR state, the combination of
CWR+PRR plus tcp_tso_should_defer() could cause unnecessary stalls
upon losses: PRR makes cwnd so gentle that tcp_tso_should_defer()
defers sending wait for more ACKs. The ACKs may not come due to
packet losses.
Disallowing TLP when there is unused cwnd had the primary effect
of disallowing TLP when there is TSO deferral, Nagle deferral,
or we hit the rwin limit. Because basically every application
write() or incoming ACK will cause us to run tcp_write_xmit()
to see if we can send more, and then if we sent something we call
tcp_schedule_loss_probe() to see if we should schedule a TLP. At
that point, there are a few common reasons why some cwnd budget
could still be unused:
(a) rwin limit
(b) nagle check
(c) TSO deferral
(d) TSQ
For (d), after the next packet tx completion the TSQ mechanism
will allow us to send more packets, so we don't really need a
TLP (in practice it shouldn't matter whether we schedule one
or not). But for (a), (b), (c) the sender won't send any more
packets until it gets another ACK. But if the whole flight was
lost, or all the ACKs were lost, then we won't get any more ACKs,
and ideally we should schedule and send a TLP to get more feedback.
In particular for a long time we have wanted some kind of timer for
TSO deferral, and at least this would give us some kind of timer
Reported-by: Steve Ibanez <sibanez@stanford.edu>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Reviewed-by: Nandita Dukkipati <nanditad@google.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Fix the TLP scheduling logic so that when scheduling a TLP probe, we
ensure that the estimated time at which an RTO would fire accounts for
the fact that ACKs indicating forward progress should push back RTO
times.
After the following fix:
df92c8394e ("tcp: fix xmit timer to only be reset if data ACKed/SACKed")
we had an unintentional behavior change in the following kind of
scenario: suppose the RTT variance has been very low recently. Then
suppose we send out a flight of N packets and our RTT is 100ms:
t=0: send a flight of N packets
t=100ms: receive an ACK for N-1 packets
The response before df92c8394e that was:
-> schedule a TLP for now + RTO_interval
The response after df92c8394e is:
-> schedule a TLP for t=0 + RTO_interval
Since RTO_interval = srtt + RTT_variance, this means that we have
scheduled a TLP timer at a point in the future that only accounts for
RTT_variance. If the RTT_variance term is small, this means that the
timer fires soon.
Before df92c8394e this would not happen, because in that code, when
we receive an ACK for a prefix of flight, we did:
1) Near the top of tcp_ack(), switch from TLP timer to RTO
at write_queue_head->paket_tx_time + RTO_interval:
if (icsk->icsk_pending == ICSK_TIME_LOSS_PROBE)
tcp_rearm_rto(sk);
2) In tcp_clean_rtx_queue(), update the RTO to now + RTO_interval:
if (flag & FLAG_ACKED) {
tcp_rearm_rto(sk);
3) In tcp_ack() after tcp_fastretrans_alert() switch from RTO
to TLP at now + RTO_interval:
if (icsk->icsk_pending == ICSK_TIME_RETRANS)
tcp_schedule_loss_probe(sk);
In df92c8394e we removed that 3-phase dance, and instead directly
set the TLP timer once: we set the TLP timer in cases like this to
write_queue_head->packet_tx_time + RTO_interval. So if the RTT
variance is small, then this means that this is setting the TLP timer
to fire quite soon. This means if the ACK for the tail of the flight
takes longer than an RTT to arrive (often due to delayed ACKs), then
the TLP timer fires too quickly.
Fixes: df92c8394e ("tcp: fix xmit timer to only be reset if data ACKed/SACKed")
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Pull networking updates from David Miller:
"Highlights:
1) Maintain the TCP retransmit queue using an rbtree, with 1GB
windows at 100Gb this really has become necessary. From Eric
Dumazet.
2) Multi-program support for cgroup+bpf, from Alexei Starovoitov.
3) Perform broadcast flooding in hardware in mv88e6xxx, from Andrew
Lunn.
4) Add meter action support to openvswitch, from Andy Zhou.
5) Add a data meta pointer for BPF accessible packets, from Daniel
Borkmann.
6) Namespace-ify almost all TCP sysctl knobs, from Eric Dumazet.
7) Turn on Broadcom Tags in b53 driver, from Florian Fainelli.
8) More work to move the RTNL mutex down, from Florian Westphal.
9) Add 'bpftool' utility, to help with bpf program introspection.
From Jakub Kicinski.
10) Add new 'cpumap' type for XDP_REDIRECT action, from Jesper
Dangaard Brouer.
11) Support 'blocks' of transformations in the packet scheduler which
can span multiple network devices, from Jiri Pirko.
12) TC flower offload support in cxgb4, from Kumar Sanghvi.
13) Priority based stream scheduler for SCTP, from Marcelo Ricardo
Leitner.
14) Thunderbolt networking driver, from Amir Levy and Mika Westerberg.
15) Add RED qdisc offloadability, and use it in mlxsw driver. From
Nogah Frankel.
16) eBPF based device controller for cgroup v2, from Roman Gushchin.
17) Add some fundamental tracepoints for TCP, from Song Liu.
18) Remove garbage collection from ipv6 route layer, this is a
significant accomplishment. From Wei Wang.
19) Add multicast route offload support to mlxsw, from Yotam Gigi"
* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-next: (2177 commits)
tcp: highest_sack fix
geneve: fix fill_info when link down
bpf: fix lockdep splat
net: cdc_ncm: GetNtbFormat endian fix
openvswitch: meter: fix NULL pointer dereference in ovs_meter_cmd_reply_start
netem: remove unnecessary 64 bit modulus
netem: use 64 bit divide by rate
tcp: Namespace-ify sysctl_tcp_default_congestion_control
net: Protect iterations over net::fib_notifier_ops in fib_seq_sum()
ipv6: set all.accept_dad to 0 by default
uapi: fix linux/tls.h userspace compilation error
usbnet: ipheth: prevent TX queue timeouts when device not ready
vhost_net: conditionally enable tx polling
uapi: fix linux/rxrpc.h userspace compilation errors
net: stmmac: fix LPI transitioning for dwmac4
atm: horizon: Fix irq release error
net-sysfs: trigger netlink notification on ifalias change via sysfs
openvswitch: Using kfree_rcu() to simplify the code
openvswitch: Make local function ovs_nsh_key_attr_size() static
openvswitch: Fix return value check in ovs_meter_cmd_features()
...
I had many reports that TSQ logic breaks wifi aggregation.
Current logic is to allow up to 1 ms of bytes to be queued into qdisc
and drivers queues.
But Wifi aggregation needs a bigger budget to allow bigger rates to
be discovered by various TCP Congestion Controls algorithms.
This patch adds an extra socket field, allowing wifi drivers to select
another log scale to derive TCP Small Queue credit from current pacing
rate.
Initial value is 10, meaning that this patch does not change current
behavior.
We expect wifi drivers to set this field to smaller values (tests have
been done with values from 6 to 9)
They would have to use following template :
if (skb->sk && skb->sk->sk_pacing_shift != MY_PACING_SHIFT)
skb->sk->sk_pacing_shift = MY_PACING_SHIFT;
Ref: https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1670041
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Johannes Berg <johannes.berg@intel.com>
Cc: Toke Høiland-Jørgensen <toke@toke.dk>
Cc: Kir Kolyshkin <kir@openvz.org>
Acked-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Replace the reordering distance measurement in packet unit with
sequence based approach. Previously it trackes the number of "packets"
toward the forward ACK (i.e. highest sacked sequence)in a state
variable "fackets_out".
Precisely measuring reordering degree on packet distance has not much
benefit, as the degree constantly changes by factors like path, load,
and congestion window. It is also complicated and prone to arcane bugs.
This patch replaces with sequence-based approach that's much simpler.
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Neal Cardwell <ncardwell@google.com>
Reviewed-by: Soheil Hassas Yeganeh <soheil@google.com>
Reviewed-by: Priyaranjan Jha <priyarjha@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
FACK loss detection has been disabled by default and the
successor RACK subsumed FACK and can handle reordering better.
This patch removes FACK to simplify TCP loss recovery.
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Neal Cardwell <ncardwell@google.com>
Reviewed-by: Soheil Hassas Yeganeh <soheil@google.com>
Reviewed-by: Priyaranjan Jha <priyarjha@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Note that when a new netns is created, it inherits its
sysctl_tcp_rmem and sysctl_tcp_wmem from initial netns.
This change is needed so that we can refine TCP rcvbuf autotuning,
to take RTT into consideration.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Wei Wang <weiwan@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
tcp_init_nondata_skb() is fed with freshly allocated skbs.
They already have a cleared csum field, no need to clear it again.
This is based on Neal review on commit 3b11775033 ("tcp: do not mangle
skb->cb[] in tcp_make_synack()"), noticing I did not clear skb->csum.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Files removed in 'net-next' had their license header updated
in 'net'. We take the remove from 'net-next'.
Signed-off-by: David S. Miller <davem@davemloft.net>
While stress testing MTU probing, we had crashes in list_del() that we root-caused
to the fact that tcp_fragment() is unconditionally inserting the freshly allocated
skb into tsorted_sent_queue list.
But this list is supposed to contain skbs that were sent.
This was mostly harmless until MTU probing was enabled.
Fortunately we can use the tcp_queue enum added later (but in same linux version)
for rtx-rb-tree to fix the bug.
Fixes: e2080072ed ("tcp: new list for sent but unacked skbs for RACK recovery")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Yuchung Cheng <ycheng@google.com>
Cc: Neal Cardwell <ncardwell@google.com>
Cc: Soheil Hassas Yeganeh <soheil@google.com>
Cc: Alexei Starovoitov <ast@kernel.org>
Cc: Priyaranjan Jha <priyarjha@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Christoph Paasch sent a patch to address the following issue :
tcp_make_synack() is leaving some TCP private info in skb->cb[],
then send the packet by other means than tcp_transmit_skb()
tcp_transmit_skb() makes sure to clear skb->cb[] to not confuse
IPv4/IPV6 stacks, but we have no such cleanup for SYNACK.
tcp_make_synack() should not use tcp_init_nondata_skb() :
tcp_init_nondata_skb() really should be limited to skbs put in write/rtx
queues (the ones that are only sent via tcp_transmit_skb())
This patch fixes the issue and should even save few cpu cycles ;)
Fixes: 971f10eca1 ("tcp: better TCP_SKB_CB layout to reduce cache line misses")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: Christoph Paasch <cpaasch@apple.com>
Reviewed-by: Christoph Paasch <cpaasch@apple.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Eric Dumazet