With the introduction of TCP Small Queues, TSO auto sizing, and TCP
pacing, we can implement Automatic Corking in the kernel, to help
applications doing small write()/sendmsg() to TCP sockets.
Idea is to change tcp_push() to check if the current skb payload is
under skb optimal size (a multiple of MSS bytes)
If under 'size_goal', and at least one packet is still in Qdisc or
NIC TX queues, set the TCP Small Queue Throttled bit, so that the push
will be delayed up to TX completion time.
This delay might allow the application to coalesce more bytes
in the skb in following write()/sendmsg()/sendfile() system calls.
The exact duration of the delay is depending on the dynamics
of the system, and might be zero if no packet for this flow
is actually held in Qdisc or NIC TX ring.
Using FQ/pacing is a way to increase the probability of
autocorking being triggered.
Add a new sysctl (/proc/sys/net/ipv4/tcp_autocorking) to control
this feature and default it to 1 (enabled)
Add a new SNMP counter : nstat -a | grep TcpExtTCPAutoCorking
This counter is incremented every time we detected skb was under used
and its flush was deferred.
Tested:
Interesting effects when using line buffered commands under ssh.
Excellent performance results in term of cpu usage and total throughput.
lpq83:~# echo 1 >/proc/sys/net/ipv4/tcp_autocorking
lpq83:~# perf stat ./super_netperf 4 -t TCP_STREAM -H lpq84 -- -m 128
9410.39
Performance counter stats for './super_netperf 4 -t TCP_STREAM -H lpq84 -- -m 128':
35209.439626 task-clock # 2.901 CPUs utilized
2,294 context-switches # 0.065 K/sec
101 CPU-migrations # 0.003 K/sec
4,079 page-faults # 0.116 K/sec
97,923,241,298 cycles # 2.781 GHz [83.31%]
51,832,908,236 stalled-cycles-frontend # 52.93% frontend cycles idle [83.30%]
25,697,986,603 stalled-cycles-backend # 26.24% backend cycles idle [66.70%]
102,225,978,536 instructions # 1.04 insns per cycle
# 0.51 stalled cycles per insn [83.38%]
18,657,696,819 branches # 529.906 M/sec [83.29%]
91,679,646 branch-misses # 0.49% of all branches [83.40%]
12.136204899 seconds time elapsed
lpq83:~# echo 0 >/proc/sys/net/ipv4/tcp_autocorking
lpq83:~# perf stat ./super_netperf 4 -t TCP_STREAM -H lpq84 -- -m 128
6624.89
Performance counter stats for './super_netperf 4 -t TCP_STREAM -H lpq84 -- -m 128':
40045.864494 task-clock # 3.301 CPUs utilized
171 context-switches # 0.004 K/sec
53 CPU-migrations # 0.001 K/sec
4,080 page-faults # 0.102 K/sec
111,340,458,645 cycles # 2.780 GHz [83.34%]
61,778,039,277 stalled-cycles-frontend # 55.49% frontend cycles idle [83.31%]
29,295,522,759 stalled-cycles-backend # 26.31% backend cycles idle [66.67%]
108,654,349,355 instructions # 0.98 insns per cycle
# 0.57 stalled cycles per insn [83.34%]
19,552,170,748 branches # 488.244 M/sec [83.34%]
157,875,417 branch-misses # 0.81% of all branches [83.34%]
12.130267788 seconds time elapsed
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Slow start now increases cwnd by 1 if an ACK acknowledges some packets,
regardless the number of packets. Consequently slow start performance
is highly dependent on the degree of the stretch ACKs caused by
receiver or network ACK compression mechanisms (e.g., delayed-ACK,
GRO, etc). But slow start algorithm is to send twice the amount of
packets of packets left so it should process a stretch ACK of degree
N as if N ACKs of degree 1, then exits when cwnd exceeds ssthresh. A
follow up patch will use the remainder of the N (if greater than 1)
to adjust cwnd in the congestion avoidance phase.
In addition this patch retires the experimental limited slow start
(LSS) feature. LSS has multiple drawbacks but questionable benefit. The
fractional cwnd increase in LSS requires a loop in slow start even
though it's rarely used. Configuring such an increase step via a global
sysctl on different BDPS seems hard. Finally and most importantly the
slow start overshoot concern is now better covered by the Hybrid slow
start (hystart) enabled by default.
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The code that is implemented is per memory cgroup not per netns, and
having per netns bits is just confusing. Remove the per netns bits to
make it easier to see what is really going on.
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Changed key initialization of tcp_fastopen cookies to net_get_random_once.
If the user sets a custom key net_get_random_once must be called at
least once to ensure we don't overwrite the user provided key when the
first cookie is generated later on.
Cc: Yuchung Cheng <ycheng@google.com>
Cc: Eric Dumazet <edumazet@google.com>
Cc: "David S. Miller" <davem@davemloft.net>
Signed-off-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch splits the secret key for syncookies for ipv4 and ipv6 and
initializes them with net_get_random_once. This change was the reason I
did this series. I think the initialization of the syncookie_secret is
way to early.
Cc: Florian Westphal <fw@strlen.de>
Cc: Eric Dumazet <edumazet@google.com>
Cc: "David S. Miller" <davem@davemloft.net>
Signed-off-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Rename tcp_tso_segment() to tcp_gso_segment(), to better reflect
what is going on, and ease grep games.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In commit 634fb979e8 ("inet: includes a sock_common in request_sock")
I forgot that the two ports in sock_common do not have same byte order :
skc_dport is __be16 (network order), but skc_num is __u16 (host order)
So sparse complains because ir_loc_port (mapped into skc_num) is
considered as __u16 while it should be __be16
Let rename ir_loc_port to ireq->ir_num (analogy with inet->inet_num),
and perform appropriate htons/ntohs conversions.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: Wu Fengguang <fengguang.wu@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
TCP listener refactoring, part 5 :
We want to be able to insert request sockets (SYN_RECV) into main
ehash table instead of the per listener hash table to allow RCU
lookups and remove listener lock contention.
This patch includes the needed struct sock_common in front
of struct request_sock
This means there is no more inet6_request_sock IPv6 specific
structure.
Following inet_request_sock fields were renamed as they became
macros to reference fields from struct sock_common.
Prefix ir_ was chosen to avoid name collisions.
loc_port -> ir_loc_port
loc_addr -> ir_loc_addr
rmt_addr -> ir_rmt_addr
rmt_port -> ir_rmt_port
iif -> ir_iif
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
TCP listener refactoring, part 3 :
Our goal is to hash SYN_RECV sockets into main ehash for fast lookup,
and parallel SYN processing.
Current inet_ehash_bucket contains two chains, one for ESTABLISH (and
friend states) sockets, another for TIME_WAIT sockets only.
As the hash table is sized to get at most one socket per bucket, it
makes little sense to have separate twchain, as it makes the lookup
slightly more complicated, and doubles hash table memory usage.
If we make sure all socket types have the lookup keys at the same
offsets, we can use a generic and faster lookup. It turns out TIME_WAIT
and ESTABLISHED sockets already have common lookup fields for IPv4.
[ INET_TW_MATCH() is no longer needed ]
I'll provide a follow-up to factorize IPv6 lookup as well, to remove
INET6_TW_MATCH()
This way, SYN_RECV pseudo sockets will be supported the same.
A new sock_gen_put() helper is added, doing either a sock_put() or
inet_twsk_put() [ and will support SYN_RECV later ].
Note this helper should only be called in real slow path, when rcu
lookup found a socket that was moved to another identity (freed/reused
immediately), but could eventually be used in other contexts, like
sock_edemux()
Before patch :
dmesg | grep "TCP established"
TCP established hash table entries: 524288 (order: 11, 8388608 bytes)
After patch :
TCP established hash table entries: 524288 (order: 10, 4194304 bytes)
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
We currently accept cookies that were created less than 4 minutes ago
(ie, cookies with counter delta 0-3). Combined with the 8 mss table
values, this yields 32 possible values (out of 2**32) that will be valid.
Reducing the lifetime to < 2 minutes halves the guessing chance while
still providing a large enough period.
While at it, get rid of jiffies value -- they overflow too quickly on
32 bit platforms.
getnstimeofday is used to create a counter that increments every 64s.
perf shows getnstimeofday cost is negible compared to sha_transform;
normal tcp initial sequence number generation uses getnstimeofday, too.
Reported-by: Jakob Lell <jakob@jakoblell.com>
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
There are a mix of function prototypes with and without extern
in the kernel sources. Standardize on not using extern for
function prototypes.
Function prototypes don't need to be written with extern.
extern is assumed by the compiler. Its use is as unnecessary as
using auto to declare automatic/local variables in a block.
Signed-off-by: Joe Perches <joe@perches.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
tcp_rcv_established() returns only one value namely 0. We change the return
value to void (as suggested by David Miller).
After commit 0c24604b (tcp: implement RFC 5961 4.2), we no longer send RSTs in
response to SYNs. We can remove the check and processing on the return value of
tcp_rcv_established().
We also fix jtcp_rcv_established() in tcp_probe.c to match that of
tcp_rcv_established().
Signed-off-by: Vijay Subramanian <subramanian.vijay@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
After hearing many people over past years complaining against TSO being
bursty or even buggy, we are proud to present automatic sizing of TSO
packets.
One part of the problem is that tcp_tso_should_defer() uses an heuristic
relying on upcoming ACKS instead of a timer, but more generally, having
big TSO packets makes little sense for low rates, as it tends to create
micro bursts on the network, and general consensus is to reduce the
buffering amount.
This patch introduces a per socket sk_pacing_rate, that approximates
the current sending rate, and allows us to size the TSO packets so
that we try to send one packet every ms.
This field could be set by other transports.
Patch has no impact for high speed flows, where having large TSO packets
makes sense to reach line rate.
For other flows, this helps better packet scheduling and ACK clocking.
This patch increases performance of TCP flows in lossy environments.
A new sysctl (tcp_min_tso_segs) is added, to specify the
minimal size of a TSO packet (default being 2).
A follow-up patch will provide a new packet scheduler (FQ), using
sk_pacing_rate as an input to perform optional per flow pacing.
This explains why we chose to set sk_pacing_rate to twice the current
rate, allowing 'slow start' ramp up.
sk_pacing_rate = 2 * cwnd * mss / srtt
v2: Neal Cardwell reported a suspect deferring of last two segments on
initial write of 10 MSS, I had to change tcp_tso_should_defer() to take
into account tp->xmit_size_goal_segs
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Neal Cardwell <ncardwell@google.com>
Cc: Yuchung Cheng <ycheng@google.com>
Cc: Van Jacobson <vanj@google.com>
Cc: Tom Herbert <therbert@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>
Extract the local TCP stack independant parts of tcp_v6_init_sequence()
and cookie_v6_check() and export them for use by the upcoming IPv6 SYNPROXY
target.
Signed-off-by: Patrick McHardy <kaber@trash.net>
Acked-by: David S. Miller <davem@davemloft.net>
Tested-by: Martin Topholm <mph@one.com>
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Extract the local TCP stack independant parts of tcp_v4_init_sequence()
and cookie_v4_check() and export them for use by the upcoming SYNPROXY
target.
Signed-off-by: Patrick McHardy <kaber@trash.net>
Acked-by: David S. Miller <davem@davemloft.net>
Tested-by: Martin Topholm <mph@one.com>
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Encrypt the cookie with both server and client IPv4 addresses,
such that multi-homed server will grant different cookies
based on both the source and destination IPs. No client change
is needed since cookie is opaque to the client.
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Remove declaration, 4 defines and confusing comment that are no longer used
since 1a2c6181c4 ("tcp: Remove TCPCT").
Signed-off-by: Dmitry Popov <dp@highloadlab.com>
Acked-by: Christoph Paasch <christoph.paasch@uclouvain.be>
Signed-off-by: David S. Miller <davem@davemloft.net>
Idea of this patch is to add optional limitation of number of
unsent bytes in TCP sockets, to reduce usage of kernel memory.
TCP receiver might announce a big window, and TCP sender autotuning
might allow a large amount of bytes in write queue, but this has little
performance impact if a large part of this buffering is wasted :
Write queue needs to be large only to deal with large BDP, not
necessarily to cope with scheduling delays (incoming ACKS make room
for the application to queue more bytes)
For most workloads, using a value of 128 KB or less is OK to give
applications enough time to react to POLLOUT events in time
(or being awaken in a blocking sendmsg())
This patch adds two ways to set the limit :
1) Per socket option TCP_NOTSENT_LOWAT
2) A sysctl (/proc/sys/net/ipv4/tcp_notsent_lowat) for sockets
not using TCP_NOTSENT_LOWAT socket option (or setting a zero value)
Default value being UINT_MAX (0xFFFFFFFF), meaning this has no effect.
This changes poll()/select()/epoll() to report POLLOUT
only if number of unsent bytes is below tp->nosent_lowat
Note this might increase number of sendmsg()/sendfile() calls
when using non blocking sockets,
and increase number of context switches for blocking sockets.
Note this is not related to SO_SNDLOWAT (as SO_SNDLOWAT is
defined as :
Specify the minimum number of bytes in the buffer until
the socket layer will pass the data to the protocol)
Tested:
netperf sessions, and watching /proc/net/protocols "memory" column for TCP
With 200 concurrent netperf -t TCP_STREAM sessions, amount of kernel memory
used by TCP buffers shrinks by ~55 % (20567 pages instead of 45458)
lpq83:~# echo -1 >/proc/sys/net/ipv4/tcp_notsent_lowat
lpq83:~# (super_netperf 200 -t TCP_STREAM -H remote -l 90 &); sleep 60 ; grep TCP /proc/net/protocols
TCPv6 1880 2 45458 no 208 yes ipv6 y y y y y y y y y y y y y n y y y y y
TCP 1696 508 45458 no 208 yes kernel y y y y y y y y y y y y y n y y y y y
lpq83:~# echo 131072 >/proc/sys/net/ipv4/tcp_notsent_lowat
lpq83:~# (super_netperf 200 -t TCP_STREAM -H remote -l 90 &); sleep 60 ; grep TCP /proc/net/protocols
TCPv6 1880 2 20567 no 208 yes ipv6 y y y y y y y y y y y y y n y y y y y
TCP 1696 508 20567 no 208 yes kernel y y y y y y y y y y y y y n y y y y y
Using 128KB has no bad effect on the throughput or cpu usage
of a single flow, although there is an increase of context switches.
A bonus is that we hold socket lock for a shorter amount
of time and should improve latencies of ACK processing.
lpq83:~# echo -1 >/proc/sys/net/ipv4/tcp_notsent_lowat
lpq83:~# perf stat -e context-switches ./netperf -H 7.7.7.84 -t omni -l 20 -c -i10,3
OMNI Send TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 7.7.7.84 () port 0 AF_INET : +/-2.500% @ 99% conf.
Local Remote Local Elapsed Throughput Throughput Local Local Remote Remote Local Remote Service
Send Socket Recv Socket Send Time Units CPU CPU CPU CPU Service Service Demand
Size Size Size (sec) Util Util Util Util Demand Demand Units
Final Final % Method % Method
1651584 6291456 16384 20.00 17447.90 10^6bits/s 3.13 S -1.00 U 0.353 -1.000 usec/KB
Performance counter stats for './netperf -H 7.7.7.84 -t omni -l 20 -c -i10,3':
412,514 context-switches
200.034645535 seconds time elapsed
lpq83:~# echo 131072 >/proc/sys/net/ipv4/tcp_notsent_lowat
lpq83:~# perf stat -e context-switches ./netperf -H 7.7.7.84 -t omni -l 20 -c -i10,3
OMNI Send TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 7.7.7.84 () port 0 AF_INET : +/-2.500% @ 99% conf.
Local Remote Local Elapsed Throughput Throughput Local Local Remote Remote Local Remote Service
Send Socket Recv Socket Send Time Units CPU CPU CPU CPU Service Service Demand
Size Size Size (sec) Util Util Util Util Demand Demand Units
Final Final % Method % Method
1593240 6291456 16384 20.00 17321.16 10^6bits/s 3.35 S -1.00 U 0.381 -1.000 usec/KB
Performance counter stats for './netperf -H 7.7.7.84 -t omni -l 20 -c -i10,3':
2,675,818 context-switches
200.029651391 seconds time elapsed
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Neal Cardwell <ncardwell@google.com>
Cc: Yuchung Cheng <ycheng@google.com>
Acked-By: Yuchung Cheng <ycheng@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Prefer packet timings to TS-ecr for RTT measurements when both
sources are available. That's because broken middle-boxes and remote
peer can return packets with corrupted TS ECR fields. Similarly most
congestion controls that require RTT signals favor timing-based
sources as well. Also check for bad TS ECR values to avoid RTT
blow-ups. It has happened on production Web servers.
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The first patch consolidates SYNACK and other RTT measurement to use a
central function tcp_ack_update_rtt(). A (small) bonus is now SYNACK
RTT measurement happens after PAWS check, potentially reducing the
impact of RTO seeding on bad TCP timestamps values.
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
struct tcp_fastopen_context has a field named tfm, which is a pointer
to a crypto_cipher structure.
It currently has a __rcu annotation, which is not needed at all.
tcp_fastopen_ctx is the pointer fetched by rcu_dereference(), but once
we have a pointer to current tcp_fastopen_context, we do not use/need
rcu_dereference() to access tfm.
This fixes a lot of sparse errors like the following :
net/ipv4/tcp_fastopen.c:21:31: warning: incorrect type in argument 1 (different address spaces)
net/ipv4/tcp_fastopen.c:21:31: expected struct crypto_cipher *tfm
net/ipv4/tcp_fastopen.c:21:31: got struct crypto_cipher [noderef] <asn:4>*tfm
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Jerry Chu <hkchu@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Linux sends new unset data during disorder and recovery state if all
(suspected) lost packets have been retransmitted ( RFC5681, section
3.2 step 1 & 2, RFC3517 section 4, NexSeg() Rule 2). One requirement
is to keep the receive window about twice the estimated sender's
congestion window (tcp_rcv_space_adjust()), assuming the fast
retransmits repair the losses in the next round trip.
But currently it's not the case on the first round trip in either
normal or Fast Open connection, beucase the initial receive window
is identical to (expected) sender's initial congestion window. The
fix is to double it.
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Acked-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Would be good to make things explicit and move those functions to
a new file called tcp_offload.c, thus make this similar to tcpv6_offload.c.
While moving all related functions into tcp_offload.c, we can also
make some of them static, since they are only used there. Also, add
an explicit registration function.
Suggested-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: Daniel Borkmann <dborkman@redhat.com>
Acked-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
TCP md5 code uses per cpu variables but protects access to them with
a shared spinlock, which is a contention point.
[ tcp_md5sig_pool_lock is locked twice per incoming packet ]
Makes things much simpler, by allocating crypto structures once, first
time a socket needs md5 keys, and not deallocating them as they are
really small.
Next step would be to allow crypto allocations being done in a NUMA
aware way.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
tcp_timeout_skb() was intended to trigger fast recovery on timeout,
unfortunately in reality it often causes spurious retransmission
storms during fast recovery. The particular sign is a fast retransmit
over the highest sacked sequence (SND.FACK).
Currently the RTO timer re-arming (as in RFC6298) offers a nice cushion
to avoid spurious timeout: when SND.UNA advances the sender re-arms
RTO and extends the timeout by icsk_rto. The sender does not offset
the time elapsed since the packet at SND.UNA was sent.
But if the next (DUP)ACK arrives later than ~RTTVAR and triggers
tcp_fastretrans_alert(), then tcp_timeout_skb() will mark any packet
sent before the icsk_rto interval lost, including one that's above the
highest sacked sequence. Most likely a large part of scorebard will be
marked.
If most packets are not lost then the subsequent DUPACKs with new SACK
blocks will cause the sender to continue to retransmit packets beyond
SND.FACK spuriously. Even if only one packet is lost the sender may
falsely retransmit almost the entire window.
The situation becomes common in the world of bufferbloat: the RTT
continues to grow as the queue builds up but RTTVAR remains small and
close to the minimum 200ms. If a data packet is lost and the DUPACK
triggered by the next data packet is slightly delayed, then a spurious
retransmission storm forms.
As the original comment on tcp_timeout_skb() suggests: the usefulness
of this feature is questionable. It also wastes cycles walking the
sack scoreboard and is actually harmful because of false recovery.
It's time to remove this.
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Acked-by: Eric Dumazet <edumazet@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Acked-by: Nandita Dukkipati <nanditad@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
I noticed that TSQ (TCP Small queues) was less effective when TSO is
turned off, and GSO is on. If BQL is not enabled, TSQ has then no
effect.
It turns out the GSO engine frees the original gso_skb at the time the
fragments are generated and queued to the NIC.
We should instead call the tcp_wfree() destructor for the last fragment,
to keep the flow control as intended in TSQ. This effectively limits
the number of queued packets on qdisc + NIC layers.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Tom Herbert <therbert@google.com>
Cc: Yuchung Cheng <ycheng@google.com>
Cc: Nandita Dukkipati <nanditad@google.com>
Cc: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Remove a declaration left over from the TCPCT-ectomy. This sysctl is
no longer referenced anywhere since 1a2c6181c4 ("tcp: Remove TCPCT").
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The patch series refactor the F-RTO feature (RFC4138/5682).
This is to simplify the loss recovery processing. Existing F-RTO
was developed during the experimental stage (RFC4138) and has
many experimental features. It takes a separate code path from
the traditional timeout processing by overloading CA_Disorder
instead of using CA_Loss state. This complicates CA_Disorder state
handling because it's also used for handling dubious ACKs and undos.
While the algorithm in the RFC does not change the congestion control,
the implementation intercepts congestion control in various places
(e.g., frto_cwnd in tcp_ack()).
The new code implements newer F-RTO RFC5682 using CA_Loss processing
path. F-RTO becomes a small extension in the timeout processing
and interfaces with congestion control and Eifel undo modules.
It lets congestion control (module) determines how many to send
independently. F-RTO only chooses what to send in order to detect
spurious retranmission. If timeout is found spurious it invokes
existing Eifel undo algorithms like DSACK or TCP timestamp based
detection.
The first patch removes all F-RTO code except the sysctl_tcp_frto is
left for the new implementation. Since CA_EVENT_FRTO is removed, TCP
westwood now computes ssthresh on regular timeout CA_EVENT_LOSS event.
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Acked-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
TCPCT uses option-number 253, reserved for experimental use and should
not be used in production environments.
Further, TCPCT does not fully implement RFC 6013.
As a nice side-effect, removing TCPCT increases TCP's performance for
very short flows:
Doing an apache-benchmark with -c 100 -n 100000, sending HTTP-requests
for files of 1KB size.
before this patch:
average (among 7 runs) of 20845.5 Requests/Second
after:
average (among 7 runs) of 21403.6 Requests/Second
Signed-off-by: Christoph Paasch <christoph.paasch@uclouvain.be>
Signed-off-by: David S. Miller <davem@davemloft.net>
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>
TCP prequeue mechanism purpose is to let incoming packets
being processed by the thread currently blocked in tcp_recvmsg(),
instead of behalf of the softirq handler, to better adapt flow
control on receiver host capacity to schedule the consumer.
But in typical request/answer workloads, we send request, then
block to receive the answer. And before the actual answer, TCP
stack receives the ACK packets acknowledging the request.
Processing pure ACK on behalf of the thread blocked in tcp_recvmsg()
is a waste of resources, as thread has to immediately sleep again
because it got no payload.
This patch avoids the extra context switches and scheduler overhead.
Before patch :
a:~# echo 0 >/proc/sys/net/ipv4/tcp_low_latency
a:~# perf stat ./super_netperf 300 -t TCP_RR -l 10 -H 7.7.7.84 -- -r 8k,8k
231676
Performance counter stats for './super_netperf 300 -t TCP_RR -l 10 -H 7.7.7.84 -- -r 8k,8k':
116251.501765 task-clock # 11.369 CPUs utilized
5,025,463 context-switches # 0.043 M/sec
1,074,511 CPU-migrations # 0.009 M/sec
216,923 page-faults # 0.002 M/sec
311,636,972,396 cycles # 2.681 GHz
260,507,138,069 stalled-cycles-frontend # 83.59% frontend cycles idle
155,590,092,840 stalled-cycles-backend # 49.93% backend cycles idle
100,101,255,411 instructions # 0.32 insns per cycle
# 2.60 stalled cycles per insn
16,535,930,999 branches # 142.243 M/sec
646,483,591 branch-misses # 3.91% of all branches
10.225482774 seconds time elapsed
After patch :
a:~# echo 0 >/proc/sys/net/ipv4/tcp_low_latency
a:~# perf stat ./super_netperf 300 -t TCP_RR -l 10 -H 7.7.7.84 -- -r 8k,8k
233297
Performance counter stats for './super_netperf 300 -t TCP_RR -l 10 -H 7.7.7.84 -- -r 8k,8k':
91084.870855 task-clock # 8.887 CPUs utilized
2,485,916 context-switches # 0.027 M/sec
815,520 CPU-migrations # 0.009 M/sec
216,932 page-faults # 0.002 M/sec
245,195,022,629 cycles # 2.692 GHz
202,635,777,041 stalled-cycles-frontend # 82.64% frontend cycles idle
124,280,372,407 stalled-cycles-backend # 50.69% backend cycles idle
83,457,289,618 instructions # 0.34 insns per cycle
# 2.43 stalled cycles per insn
13,431,472,361 branches # 147.461 M/sec
504,470,665 branch-misses # 3.76% of all branches
10.249594448 seconds time elapsed
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Neal Cardwell <ncardwell@google.com>
Cc: Tom Herbert <therbert@google.com>
Cc: Yuchung Cheng <ycheng@google.com>
Cc: Andi Kleen <ak@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
TCP Appropriate Byte Count was added by me, but later disabled.
There is no point in maintaining it since it is a potential source
of bugs and Linux already implements other better window protection
heuristics.
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
As per suggestion from Eric Dumazet this patch makes tcp_ecn sysctl
namespace aware. The reason behind this patch is to ease the testing
of ecn problems on the internet and allows applications to tune their
own use of ecn.
Cc: Eric Dumazet <eric.dumazet@gmail.com>
Cc: David Miller <davem@davemloft.net>
Cc: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Acked-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Pull networking changes from David Miller:
1) Allow to dump, monitor, and change the bridge multicast database
using netlink. From Cong Wang.
2) RFC 5961 TCP blind data injection attack mitigation, from Eric
Dumazet.
3) Networking user namespace support from Eric W. Biederman.
4) tuntap/virtio-net multiqueue support by Jason Wang.
5) Support for checksum offload of encapsulated packets (basically,
tunneled traffic can still be checksummed by HW). From Joseph
Gasparakis.
6) Allow BPF filter access to VLAN tags, from Eric Dumazet and
Daniel Borkmann.
7) Bridge port parameters over netlink and BPDU blocking support
from Stephen Hemminger.
8) Improve data access patterns during inet socket demux by rearranging
socket layout, from Eric Dumazet.
9) TIPC protocol updates and cleanups from Ying Xue, Paul Gortmaker, and
Jon Maloy.
10) Update TCP socket hash sizing to be more in line with current day
realities. The existing heurstics were choosen a decade ago.
From Eric Dumazet.
11) Fix races, queue bloat, and excessive wakeups in ATM and
associated drivers, from Krzysztof Mazur and David Woodhouse.
12) Support DOVE (Distributed Overlay Virtual Ethernet) extensions
in VXLAN driver, from David Stevens.
13) Add "oops_only" mode to netconsole, from Amerigo Wang.
14) Support set and query of VEB/VEPA bridge mode via PF_BRIDGE, also
allow DCB netlink to work on namespaces other than the initial
namespace. From John Fastabend.
15) Support PTP in the Tigon3 driver, from Matt Carlson.
16) tun/vhost zero copy fixes and improvements, plus turn it on
by default, from Michael S. Tsirkin.
17) Support per-association statistics in SCTP, from Michele
Baldessari.
And many, many, driver updates, cleanups, and improvements. Too
numerous to mention individually.
* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-next: (1722 commits)
net/mlx4_en: Add support for destination MAC in steering rules
net/mlx4_en: Use generic etherdevice.h functions.
net: ethtool: Add destination MAC address to flow steering API
bridge: add support of adding and deleting mdb entries
bridge: notify mdb changes via netlink
ndisc: Unexport ndisc_{build,send}_skb().
uapi: add missing netconf.h to export list
pkt_sched: avoid requeues if possible
solos-pci: fix double-free of TX skb in DMA mode
bnx2: Fix accidental reversions.
bna: Driver Version Updated to 3.1.2.1
bna: Firmware update
bna: Add RX State
bna: Rx Page Based Allocation
bna: TX Intr Coalescing Fix
bna: Tx and Rx Optimizations
bna: Code Cleanup and Enhancements
ath9k: check pdata variable before dereferencing it
ath5k: RX timestamp is reported at end of frame
ath9k_htc: RX timestamp is reported at end of frame
...
If SYN-ACK partially acks SYN-data, the client retransmits the
remaining data by tcp_retransmit_skb(). This increments lost recovery
state variables like tp->retrans_out in Open state. If loss recovery
happens before the retransmission is acked, it triggers the WARN_ON
check in tcp_fastretrans_alert(). For example: the client sends
SYN-data, gets SYN-ACK acking only ISN, retransmits data, sends
another 4 data packets and get 3 dupacks.
Since the retransmission is not caused by network drop it should not
update the recovery state variables. Further the server may return a
smaller MSS than the cached MSS used for SYN-data, so the retranmission
needs a loop. Otherwise some data will not be retransmitted until timeout
or other loss recovery events.
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When taking SYNACK RTT samples for servers using TCP Fast Open, fix
the code to ensure that we only call tcp_valid_rtt_meas() after we
receive the ACK that completes the 3-way handshake.
Previously we were always taking an RTT sample in
tcp_v4_syn_recv_sock(). However, for TCP Fast Open connections
tcp_v4_conn_req_fastopen() calls tcp_v4_syn_recv_sock() at the time we
receive the SYN. So for TFO we must wait until tcp_rcv_state_process()
to take the RTT sample.
To fix this, we wait until after TFO calls tcp_v4_syn_recv_sock()
before we set the snt_synack timestamp, since tcp_synack_rtt_meas()
already ensures that we only take a SYNACK RTT sample if snt_synack is
non-zero. To be careful, we only take a snt_synack timestamp when
a SYNACK transmit or retransmit succeeds.
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In preparation for adding another spot where we compute the SYNACK
RTT, extract this code so that it can be shared.
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Use proportional rate reduction (PRR) algorithm to reduce cwnd in CWR state,
in addition to Recovery state. Retire the current rate-halving in CWR.
When losses are detected via ACKs in CWR state, the sender enters Recovery
state but the cwnd reduction continues and does not restart.
Rename and refactor cwnd reduction functions since both CWR and Recovery
use the same algorithm:
tcp_init_cwnd_reduction() is new and initiates reduction state variables.
tcp_cwnd_reduction() is previously tcp_update_cwnd_in_recovery().
tcp_ends_cwnd_reduction() is previously tcp_complete_cwr().
The rate halving functions and logic such as tcp_cwnd_down(), tcp_min_cwnd(),
and the cwnd moderation inside tcp_enter_cwr() are removed. The unused
parameter, flag, in tcp_cwnd_reduction() is also removed.
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch builds on top of the previous patch to add the support
for TFO listeners. This includes -
1. allocating, properly initializing, and managing the per listener
fastopen_queue structure when TFO is enabled
2. changes to the inet_csk_accept code to support TFO. E.g., the
request_sock can no longer be freed upon accept(), not until 3WHS
finishes
3. allowing a TCP_SYN_RECV socket to properly poll() and sendmsg()
if it's a TFO socket
4. properly closing a TFO listener, and a TFO socket before 3WHS
finishes
5. supporting TCP_FASTOPEN socket option
6. modifying tcp_check_req() to use to check a TFO socket as well
as request_sock
7. supporting TCP's TFO cookie option
8. adding a new SYN-ACK retransmit handler to use the timer directly
off the TFO socket rather than the listener socket. Note that TFO
server side will not retransmit anything other than SYN-ACK until
the 3WHS is completed.
The patch also contains an important function
"reqsk_fastopen_remove()" to manage the somewhat complex relation
between a listener, its request_sock, and the corresponding child
socket. See the comment above the function for the detail.
Signed-off-by: H.K. Jerry Chu <hkchu@google.com>
Cc: Yuchung Cheng <ycheng@google.com>
Cc: Neal Cardwell <ncardwell@google.com>
Cc: Eric Dumazet <edumazet@google.com>
Cc: Tom Herbert <therbert@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch adds all the necessary data structure and support
functions to implement TFO server side. It also documents a number
of flags for the sysctl_tcp_fastopen knob, and adds a few Linux
extension MIBs.
In addition, it includes the following:
1. a new TCP_FASTOPEN socket option an application must call to
supply a max backlog allowed in order to enable TFO on its listener.
2. A number of key data structures:
"fastopen_rsk" in tcp_sock - for a big socket to access its
request_sock for retransmission and ack processing purpose. It is
non-NULL iff 3WHS not completed.
"fastopenq" in request_sock_queue - points to a per Fast Open
listener data structure "fastopen_queue" to keep track of qlen (# of
outstanding Fast Open requests) and max_qlen, among other things.
"listener" in tcp_request_sock - to point to the original listener
for book-keeping purpose, i.e., to maintain qlen against max_qlen
as part of defense against IP spoofing attack.
3. various data structure and functions, many in tcp_fastopen.c, to
support server side Fast Open cookie operations, including
/proc/sys/net/ipv4/tcp_fastopen_key to allow manual rekeying.
Signed-off-by: H.K. Jerry Chu <hkchu@google.com>
Cc: Yuchung Cheng <ycheng@google.com>
Cc: Neal Cardwell <ncardwell@google.com>
Cc: Eric Dumazet <edumazet@google.com>
Cc: Tom Herbert <therbert@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Commit 9ad7c049 ("tcp: RFC2988bis + taking RTT sample from 3WHS for
the passive open side") changed the initRTO from 3secs to 1sec in
accordance to RFC6298 (former RFC2988bis). This reduced the time till
the last SYN retransmission packet gets sent from 93secs to 31secs.
RFC1122 is stating that the retransmission should be done for at least 3
minutes, but this seems to be quite high.
"However, the values of R1 and R2 may be different for SYN
and data segments. In particular, R2 for a SYN segment MUST
be set large enough to provide retransmission of the segment
for at least 3 minutes. The application can close the
connection (i.e., give up on the open attempt) sooner, of
course."
This patch increases the value of TCP_SYN_RETRIES to the value of 6,
providing a retransmission window of 63secs.
The comments for SYN and SYNACK retries have also been updated to
describe the current settings. The same goes for the documentation file
"Documentation/networking/ip-sysctl.txt".
Signed-off-by: Alexander Bergmann <alex@linlab.net>
Acked-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This is an initial merge in of Eric Biederman's work to start adding
user namespace support to the networking.
Signed-off-by: David S. Miller <davem@davemloft.net>
Modern TCP stack highly depends on tcp_write_timer() having a small
latency, but current implementation doesn't exactly meet the
expectations.
When a timer fires but finds the socket is owned by the user, it rearms
itself for an additional delay hoping next run will be more
successful.
tcp_write_timer() for example uses a 50ms delay for next try, and it
defeats many attempts to get predictable TCP behavior in term of
latencies.
Use the recently introduced tcp_release_cb(), so that the user owning
the socket will call various handlers right before socket release.
This will permit us to post a followup patch to address the
tcp_tso_should_defer() syndrome (some deferred packets have to wait
RTO timer to be transmitted, while cwnd should allow us to send them
sooner)
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Tom Herbert <therbert@google.com>
Cc: Yuchung Cheng <ycheng@google.com>
Cc: Neal Cardwell <ncardwell@google.com>
Cc: Nandita Dukkipati <nanditad@google.com>
Cc: H.K. Jerry Chu <hkchu@google.com>
Cc: John Heffner <johnwheffner@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In trusted networks, e.g., intranet, data-center, the client does not
need to use Fast Open cookie to mitigate DoS attacks. In cookie-less
mode, sendmsg() with MSG_FASTOPEN flag will send SYN-data regardless
of cookie availability.
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Acked-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
On paths with firewalls dropping SYN with data or experimental TCP options,
Fast Open connections will have experience SYN timeout and bad performance.
The solution is to track such incidents in the cookie cache and disables
Fast Open temporarily.
Since only the original SYN includes data and/or Fast Open option, the
SYN-ACK has some tell-tale sign (tcp_rcv_fastopen_synack()) to detect
such drops. If a path has recurring Fast Open SYN drops, Fast Open is
disabled for 2^(recurring_losses) minutes starting from four minutes up to
roughly one and half day. sendmsg with MSG_FASTOPEN flag will succeed but
it behaves as connect() then write().
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Acked-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
sendmsg() (or sendto()) with MSG_FASTOPEN is a combo of connect(2)
and write(2). The application should replace connect() with it to
send data in the opening SYN packet.
For blocking socket, sendmsg() blocks until all the data are buffered
locally and the handshake is completed like connect() call. It
returns similar errno like connect() if the TCP handshake fails.
For non-blocking socket, it returns the number of bytes queued (and
transmitted in the SYN-data packet) if cookie is available. If cookie
is not available, it transmits a data-less SYN packet with Fast Open
cookie request option and returns -EINPROGRESS like connect().
Using MSG_FASTOPEN on connecting or connected socket will result in
simlar errno like repeating connect() calls. Therefore the application
should only use this flag on new sockets.
The buffer size of sendmsg() is independent of the MSS of the connection.
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Acked-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
