The TSO and GSO segmented packets already benefit from bulking
on their own.

The TSO packets have always taken advantage of the only updating
the tailptr once for a large packet.

The GSO segmented packets have recently taken advantage of
bulking xmit_more API, via merge commit 53fda7f7 ("Merge
branch 'xmit_list'"), specifically via commit 7f2e870f ("net:
Move main gso loop out of dev_hard_start_xmit() into helper.")
allowing qdisc requeue of remaining list.  And via commit
ce93718f ("net: Don't keep around original SKB when we
software segment GSO frames.").

This patch allow further bulking of TSO/GSO packets together,
when dequeueing from the qdisc.

Testing:
 Measuring HoL (Head-of-Line) blocking for TSO and GSO, with
netperf-wrapper. Bulking several TSO show no performance regressions
(requeues were in the area 32 requeues/sec).

Bulking several GSOs does show small regression or very small
improvement (requeues were in the area 8000 requeues/sec).

 Using ixgbe 10Gbit/s with GSO bulking, we can measure some additional
latency. Base-case, which is "normal" GSO bulking, sees varying
high-prio queue delay between 0.38ms to 0.47ms.  Bulking several GSOs
together, result in a stable high-prio queue delay of 0.50ms.

 Using igb at 100Mbit/s with GSO bulking, shows an improvement.
Base-case sees varying high-prio queue delay between 2.23ms to 2.35ms
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Based on DaveM's recent API work on dev_hard_start_xmit(), that allows
sending/processing an entire skb list.

This patch implements qdisc bulk dequeue, by allowing multiple packets
to be dequeued in dequeue_skb().

The optimization principle for this is two fold, (1) to amortize
locking cost and (2) avoid expensive tailptr update for notifying HW.
 (1) Several packets are dequeued while holding the qdisc root_lock,
amortizing locking cost over several packet.  The dequeued SKB list is
processed under the TXQ lock in dev_hard_start_xmit(), thus also
amortizing the cost of the TXQ lock.
 (2) Further more, dev_hard_start_xmit() will utilize the skb->xmit_more
API to delay HW tailptr update, which also reduces the cost per
packet.

One restriction of the new API is that every SKB must belong to the
same TXQ.  This patch takes the easy way out, by restricting bulk
dequeue to qdisc's with the TCQ_F_ONETXQUEUE flag, that specifies the
qdisc only have attached a single TXQ.

Some detail about the flow; dev_hard_start_xmit() will process the skb
list, and transmit packets individually towards the driver (see
xmit_one()).  In case the driver stops midway in the list, the
remaining skb list is returned by dev_hard_start_xmit().  In
sch_direct_xmit() this returned list is requeued by dev_requeue_skb().

To avoid overshooting the HW limits, which results in requeuing, the
patch limits the amount of bytes dequeued, based on the drivers BQL
limits.  In-effect bulking will only happen for BQL enabled drivers.

Small amounts for extra HoL blocking (2x MTU/0.24ms) were
measured at 100Mbit/s, with bulking 8 packets, but the
oscillating nature of the measurement indicate something, like
sched latency might be causing this effect. More comparisons
show, that this oscillation goes away occationally. Thus, we
disregard this artifact completely and remove any "magic" bulking
limit.

For now, as a conservative approach, stop bulking when seeing TSO and
segmented GSO packets.  They already benefit from bulking on their own.
A followup patch add this, to allow easier bisect-ability for finding
regressions.

Jointed work with Hannes, Daniel and Florian.
Signed-off-by: NJesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: NHannes Frederic Sowa <hannes@stressinduktion.org>
Signed-off-by: NDaniel Borkmann <dborkman@redhat.com>
Signed-off-by: NFlorian Westphal <fw@strlen.de>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This patch demonstrates the effect of delaying update of HW tailptr.
(based on earlier patch by Jesper)

burst=1 is the default. It sends one packet with xmit_more=false
burst=2 sends one packet with xmit_more=true and
        2nd copy of the same packet with xmit_more=false
burst=3 sends two copies of the same packet with xmit_more=true and
        3rd copy with xmit_more=false

Performance with ixgbe (usec 30):
burst=1  tx:9.2 Mpps
burst=2  tx:13.5 Mpps
burst=3  tx:14.5 Mpps full 10G line rate
Signed-off-by: NAlexei Starovoitov <ast@plumgrid.com>
Signed-off-by: NEric Dumazet <edumazet@google.com>
Acked-by: NJesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

In preparation for being able to propagate port states to e.g: notifiers
or other kernel parts, do not manipulate the port state directly, but
instead use a helper function which will allow us to do a bit more than
just setting the state.
Signed-off-by: NFlorian Fainelli <f.fainelli@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This fixes the following crash:

[   63.976822] general protection fault: 0000 [#1] PREEMPT SMP DEBUG_PAGEALLOC
[   63.980094] CPU: 1 PID: 15 Comm: ksoftirqd/1 Not tainted 3.17.0-rc6+ #648
[   63.980094] Hardware name: Bochs Bochs, BIOS Bochs 01/01/2011
[   63.980094] task: ffff880117dea690 ti: ffff880117dfc000 task.ti: ffff880117dfc000
[   63.980094] RIP: 0010:[<ffffffff817e6d07>]  [<ffffffff817e6d07>] u32_destroy_key+0x27/0x6d
[   63.980094] RSP: 0018:ffff880117dffcc0  EFLAGS: 00010202
[   63.980094] RAX: ffff880117dea690 RBX: ffff8800d02e0820 RCX: 0000000000000000
[   63.980094] RDX: 0000000000000001 RSI: 0000000000000002 RDI: 6b6b6b6b6b6b6b6b
[   63.980094] RBP: ffff880117dffcd0 R08: 0000000000000000 R09: 0000000000000000
[   63.980094] R10: 00006c0900006ba8 R11: 00006ba100006b9d R12: 0000000000000001
[   63.980094] R13: ffff8800d02e0898 R14: ffffffff817e6d4d R15: ffff880117387a30
[   63.980094] FS:  0000000000000000(0000) GS:ffff88011a800000(0000) knlGS:0000000000000000
[   63.980094] CS:  0010 DS: 0000 ES: 0000 CR0: 000000008005003b
[   63.980094] CR2: 00007f07e6732fed CR3: 000000011665b000 CR4: 00000000000006e0
[   63.980094] Stack:
[   63.980094]  ffff88011a9cd300 ffffffff82051ac0 ffff880117dffce0 ffffffff817e6d68
[   63.980094]  ffff880117dffd70 ffffffff810cb4c7 ffffffff810cb3cd ffff880117dfffd8
[   63.980094]  ffff880117dea690 ffff880117dea690 ffff880117dfffd8 000000000000000a
[   63.980094] Call Trace:
[   63.980094]  [<ffffffff817e6d68>] u32_delete_key_freepf_rcu+0x1b/0x1d
[   63.980094]  [<ffffffff810cb4c7>] rcu_process_callbacks+0x3bb/0x691
[   63.980094]  [<ffffffff810cb3cd>] ? rcu_process_callbacks+0x2c1/0x691
[   63.980094]  [<ffffffff817e6d4d>] ? u32_destroy_key+0x6d/0x6d
[   63.980094]  [<ffffffff810780a4>] __do_softirq+0x142/0x323
[   63.980094]  [<ffffffff810782a8>] run_ksoftirqd+0x23/0x53
[   63.980094]  [<ffffffff81092126>] smpboot_thread_fn+0x203/0x221
[   63.980094]  [<ffffffff81091f23>] ? smpboot_unpark_thread+0x33/0x33
[   63.980094]  [<ffffffff8108e44d>] kthread+0xc9/0xd1
[   63.980094]  [<ffffffff819e00ea>] ? do_wait_for_common+0xf8/0x125
[   63.980094]  [<ffffffff8108e384>] ? __kthread_parkme+0x61/0x61
[   63.980094]  [<ffffffff819e43ec>] ret_from_fork+0x7c/0xb0
[   63.980094]  [<ffffffff8108e384>] ? __kthread_parkme+0x61/0x61

tp could be freed in call_rcu callback too, the order is not guaranteed.

John Fastabend says:

====================
Its worth noting why this is safe. Any running schedulers will either
read the valid class field or it will be zeroed.

All schedulers today when the class is 0 do a lookup using the
same call used by the tcf_exts_bind(). So even if we have a running
classifier hit the null class pointer it will do a lookup and get
to the same result. This is particularly fragile at the moment because
the only way to verify this is to audit the schedulers call sites.
====================

Cc: John Fastabend <john.r.fastabend@intel.com>
Signed-off-by: NCong Wang <xiyou.wangcong@gmail.com>
Acked-by: NJohn Fastabend <john.r.fastabend@intel.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This patch fixes the following crash:

[  166.670795] BUG: unable to handle kernel NULL pointer dereference at           (null)
[  166.674230] IP: [<ffffffff814b739f>] __list_del_entry+0x5c/0x98
[  166.674230] PGD d0ea5067 PUD ce7fc067 PMD 0
[  166.674230] Oops: 0000 [#1] PREEMPT SMP DEBUG_PAGEALLOC
[  166.674230] CPU: 1 PID: 775 Comm: tc Not tainted 3.17.0-rc6+ #642
[  166.674230] Hardware name: Bochs Bochs, BIOS Bochs 01/01/2011
[  166.674230] task: ffff8800d03c4d20 ti: ffff8800cae7c000 task.ti: ffff8800cae7c000
[  166.674230] RIP: 0010:[<ffffffff814b739f>]  [<ffffffff814b739f>] __list_del_entry+0x5c/0x98
[  166.674230] RSP: 0018:ffff8800cae7f7d0  EFLAGS: 00010207
[  166.674230] RAX: 0000000000000000 RBX: ffff8800cba8d700 RCX: ffff8800cba8d700
[  166.674230] RDX: 0000000000000000 RSI: dead000000200200 RDI: ffff8800cba8d700
[  166.674230] RBP: ffff8800cae7f7d0 R08: 0000000000000001 R09: 0000000000000001
[  166.674230] R10: 0000000000000000 R11: 000000000000859a R12: ffffffffffffffe8
[  166.674230] R13: ffff8800cba8c5b8 R14: 0000000000000001 R15: ffff8800cba8d700
[  166.674230] FS:  00007fdb5f04a740(0000) GS:ffff88011a800000(0000) knlGS:0000000000000000
[  166.674230] CS:  0010 DS: 0000 ES: 0000 CR0: 000000008005003b
[  166.674230] CR2: 0000000000000000 CR3: 00000000cf929000 CR4: 00000000000006e0
[  166.674230] Stack:
[  166.674230]  ffff8800cae7f7e8 ffffffff814b73e8 ffff8800cba8d6e8 ffff8800cae7f828
[  166.674230]  ffffffff817caeec 0000000000000046 ffff8800cba8c5b0 ffff8800cba8c5b8
[  166.674230]  0000000000000000 0000000000000001 ffff8800cf8e33e8 ffff8800cae7f848
[  166.674230] Call Trace:
[  166.674230]  [<ffffffff814b73e8>] list_del+0xd/0x2b
[  166.674230]  [<ffffffff817caeec>] tcf_action_destroy+0x4c/0x71
[  166.674230]  [<ffffffff817ca0ce>] tcf_exts_destroy+0x20/0x2d
[  166.674230]  [<ffffffff817ec2b5>] tcindex_delete+0x196/0x1b7

struct list_head can not be simply copied and we should always init it.

Cc: John Fastabend <john.r.fastabend@intel.com>
Signed-off-by: NCong Wang <xiyou.wangcong@gmail.com>
Acked-by: NJohn Fastabend <john.r.fastabend@intel.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Call skb_set_inner_protocol to set inner Ethernet protocol to
protocol being encapsulation by GRE before tunnel_xmit. This is
needed for GSO if UDP encapsulation (fou) is being done.
Signed-off-by: NTom Herbert <therbert@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Call skb_set_inner_ipproto to set inner IP protocol to IPPROTO_IPV4
before tunnel_xmit. This is needed if UDP encapsulation (fou) is
being done.
Signed-off-by: NTom Herbert <therbert@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Call skb_set_inner_ipproto to set inner IP protocol to IPPROTO_IPV6
before tunnel_xmit. This is needed if UDP encapsulation (fou) is
being done.
Signed-off-by: NTom Herbert <therbert@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

skb_udp_segment is the function called from udp4_ufo_fragment to
segment a UDP tunnel packet. This function currently assumes
segmentation is transparent Ethernet bridging (i.e. VXLAN
encapsulation). This patch generalizes the function to
operate on either Ethertype or IP protocol.

The inner_protocol field must be set to the protocol of the inner
header. This can now be either an Ethertype or an IP protocol
(in a union). A new flag in the skbuff indicates which type is
effective. skb_set_inner_protocol and skb_set_inner_ipproto
helper functions were added to set the inner_protocol. These
functions are called from the point where the tunnel encapsulation
is occuring.

When skb_udp_tunnel_segment is called, the function to segment the
inner packet is selected based on the inner IP or Ethertype. In the
case of an IP protocol encapsulation, the function is derived from
inet[6]_offloads. In the case of Ethertype, skb->protocol is
set to the inner_protocol and skb_mac_gso_segment is called. (GRE
currently does this, but it might be possible to lookup the protocol
in offload_base and call the appropriate segmenation function
directly).
Signed-off-by: NTom Herbert <therbert@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Fast clone cloning can actually avoid an atomic_inc(), if we
guarantee prior clone_ref value is 1.

This requires a change kfree_skbmem(), to perform the
atomic_dec_and_test() on clone_ref before setting fclone to
SKB_FCLONE_UNAVAILABLE.
Signed-off-by: NEric Dumazet <edumazet@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

ccid_activate is only called by __init ccid_initialize_builtins in same module.
Signed-off-by: NFabian Frederick <fabf@skynet.be>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

dccp_mib_init is only called by __init dccp_init in same module.
Signed-off-by: NFabian Frederick <fabf@skynet.be>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Lets use a proper structure to clearly document and implement
skb fast clones.

Then, we might experiment more easily alternative layouts.

This patch adds a new skb_fclone_busy() helper, used by tcp and xfrm,
to stop leaking of implementation details.
Signed-off-by: NEric Dumazet <edumazet@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Currently we have two different policies for orphan sockets
that repeatedly stall on zero window ACKs. If a socket gets
a zero window ACK when it is transmitting data, the RTO is
used to probe the window. The socket is aborted after roughly
tcp_orphan_retries() retries (as in tcp_write_timeout()).

But if the socket was idle when it received the zero window ACK,
and later wants to send more data, we use the probe timer to
probe the window. If the receiver always returns zero window ACKs,
icsk_probes keeps getting reset in tcp_ack() and the orphan socket
can stall forever until the system reaches the orphan limit (as
commented in tcp_probe_timer()). This opens up a simple attack
to create lots of hanging orphan sockets to burn the memory
and the CPU, as demonstrated in the recent netdev post "TCP
connection will hang in FIN_WAIT1 after closing if zero window is
advertised." http://www.spinics.net/lists/netdev/msg296539.html

This patch follows the design in RTO-based probe: we abort an orphan
socket stalling on zero window when the probe timer reaches both
the maximum backoff and the maximum RTO. For example, an 100ms RTT
connection will timeout after roughly 153 seconds (0.3 + 0.6 +
.... + 76.8) if the receiver keeps the window shut. If the orphan
socket passes this check, but the system already has too many orphans
(as in tcp_out_of_resources()), we still abort it but we'll also
send an RST packet as the connection may still be active.

In addition, we change TCP_USER_TIMEOUT to cover (life or dead)
sockets stalled on zero-window probes. This changes the semantics
of TCP_USER_TIMEOUT slightly because it previously only applies
when the socket has pending transmission.
Signed-off-by: NYuchung Cheng <ycheng@google.com>
Signed-off-by: NEric Dumazet <edumazet@google.com>
Signed-off-by: NNeal Cardwell <ncardwell@google.com>
Reported-by: NAndrey Dmitrov <andrey.dmitrov@oktetlabs.ru>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

cipso_v4_cache_init is only called by __init cipso_v4_init
Signed-off-by: NFabian Frederick <fabf@skynet.be>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

ip4_frags_ctl_register is only called by __init ipfrag_init
Signed-off-by: NFabian Frederick <fabf@skynet.be>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

tcp_init_mem is only called by __init tcp_init.
Signed-off-by: NFabian Frederick <fabf@skynet.be>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

The dsa_switch_suspend() and dsa_switch_resume() functions are only used
when PM_SLEEP is enabled, so they need #ifdef CONFIG_PM_SLEEP protection
to avoid a compiler warning.
Signed-off-by: NThierry Reding <treding@nvidia.com>
Acked-by: NFlorian Fainelli <f.fainelli@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Proper CHECKSUM_COMPLETE support needs to adjust skb->csum
when we remove one header. Its done using skb_gro_postpull_rcsum()

In the case of IPv4, we know that the adjustment is not really needed,
because the checksum over IPv4 header is 0. Lets add a comment to
ease code comprehension and avoid copy/paste errors.
Signed-off-by: NEric Dumazet <edumazet@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Commit 3243acd3
("ieee802154: add __init to lowpan_frags_sysctl_register")

added __init to lowpan_frags_ns_sysctl_register instead of
lowpan_frags_sysctl_register
Suggested-by: NAlexander Aring <alex.aring@gmail.com>
Signed-off-by: NFabian Frederick <fabf@skynet.be>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

No caller uses the return value, so make this function return void.
Signed-off-by: NLi RongQing <roy.qing.li@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

lowpan_frags_sysctl_register is only called by __init lowpan_net_frag_init
(part of the lowpan module).
Signed-off-by: NFabian Frederick <fabf@skynet.be>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

irlan_open is only called by __init irlan_init in same module.
Signed-off-by: NFabian Frederick <fabf@skynet.be>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Eric reports build failure with
CONFIG_BRIDGE_NETFILTER=n

We insist to build br_nf_core.o unconditionally, but we must only do so
if br_netfilter was enabled, else it fails to build due to
functions being defined to empty stubs (and some structure members
being defined out).

Also, BRIDGE_NETFILTER=y|m makes no sense when BRIDGE=n.

Fixes: 34666d46 (netfilter: bridge: move br_netfilter out of the core)
Reported-by: NEric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: NFlorian Westphal <fw@strlen.de>
Signed-off-by: NPablo Neira Ayuso <pablo@netfilter.org>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Eric Dumazet noticed that all no-nonexthop or no-gateway routes which
are already marked DST_HOST (e.g. input routes routes) will always be
invalidated during sk_dst_check. Thus per-socket dst caching absolutely
had no effect and early demuxing had no effect.

Thus this patch removes rt6i_genid: fn_sernum already gets modified during
add operations, so we only must ensure we mutate fn_sernum during ipv6
address remove operations. This is a fairly cost extensive operations,
but address removal should not happen that often. Also our mtu update
functions do the same and we heard no complains so far. xfrm policy
changes also cause a call into fib6_flush_trees. Also plug a hole in
rt6_info (no cacheline changes).

I verified via tracing that this change has effect.

Cc: Eric Dumazet <eric.dumazet@gmail.com>
Cc: YOSHIFUJI Hideaki <hideaki@yoshifuji.org>
Cc: Vlad Yasevich <vyasevich@gmail.com>
Cc: Nicolas Dichtel <nicolas.dichtel@6wind.com>
Cc: Martin Lau <kafai@fb.com>
Signed-off-by: NHannes Frederic Sowa <hannes@stressinduktion.org>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

After previous patches to simplify qstats the qstats can be
made per cpu with a packed union in Qdisc struct.
Signed-off-by: NJohn Fastabend <john.r.fastabend@intel.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This removes the use of qstats->qlen variable from the classifiers
and makes it an explicit argument to gnet_stats_copy_queue().

The qlen represents the qdisc queue length and is packed into
the qstats at the last moment before passnig to user space. By
handling it explicitely we avoid, in the percpu stats case, having
to figure out which per_cpu variable to put it in.

It would probably be best to remove it from qstats completely
but qstats is a user space ABI and can't be broken. A future
patch could make an internal only qstats structure that would
avoid having to allocate an additional u32 variable on the
Qdisc struct. This would make the qstats struct 128bits instead
of 128+32.
Signed-off-by: NJohn Fastabend <john.r.fastabend@intel.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This adds helpers to manipulate qstats logic and replaces locations
that touch the counters directly. This simplifies future patches
to push qstats onto per cpu counters.
Signed-off-by: NJohn Fastabend <john.r.fastabend@intel.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

In order to run qdisc's without locking statistics and estimators
need to be handled correctly.

To resolve bstats make the statistics per cpu. And because this is
only needed for qdiscs that are running without locks which is not
the case for most qdiscs in the near future only create percpu
stats when qdiscs set the TCQ_F_CPUSTATS flag.

Next because estimators use the bstats to calculate packets per
second and bytes per second the estimator code paths are updated
to use the per cpu statistics.
Signed-off-by: NJohn Fastabend <john.r.fastabend@intel.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Negated expressions and sub-expressions need to have their flags checked for
TCF_EM_INVERT and their result negated accordingly.
Signed-off-by: NIgnacy Gawędzki <ignacy.gawedzki@green-communications.fr>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

In commit 8a29111c ("net: gro: allow to build full sized skb")
I added a regression for linear skb that traditionally force GRO
to use the frag_list fallback.

Erez Shitrit found that at most two segments were aggregated and
the "if (skb_gro_len(p) != pinfo->gso_size)" test was failing.

This is because pinfo at this spot still points to the last skb in the
chain, instead of the first one, where we find the correct gso_size
information.
Signed-off-by: NEric Dumazet <edumazet@google.com>
Fixes: 8a29111c ("net: gro: allow to build full sized skb")
Reported-by: NErez Shitrit <erezsh@mellanox.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Suggested by Stephen. Also drop inline keyword and let compiler decide.

gcc 4.7.3 decides to no longer inline tcp_ecn_check_ce, so split it up.
The actual evaluation is not inlined anymore while the ECN_OK test is.
Suggested-by: NStephen Hemminger <stephen@networkplumber.org>
Signed-off-by: NDaniel Borkmann <dborkman@redhat.com>
Signed-off-by: NFlorian Westphal <fw@strlen.de>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

After Octavian Purdilas tcp ipv4/ipv6 unification work this helper only
has a single callsite.

While at it, convert name to lowercase, suggested by Stephen.
Suggested-by: NStephen Hemminger <stephen@networkplumber.org>
Signed-off-by: NDaniel Borkmann <dborkman@redhat.com>
Signed-off-by: NFlorian Westphal <fw@strlen.de>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This variable i is overwritten to 0 by following code
Signed-off-by: NLi RongQing <roy.qing.li@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

With proliferation of bit fields in sk_buff, __copy_skb_header() became
quite expensive, showing as the most expensive function in a GSO
workload.

__copy_skb_header() performance is also critical for non GSO TCP
operations, as it is used from skb_clone()

This patch carefully moves all the fields that were not copied in a
separate zone : cloned, nohdr, fclone, peeked, head_frag, xmit_more

Then I moved all other fields and all other copied fields in a section
delimited by headers_start[0]/headers_end[0] section so that we
can use a single memcpy() call, inlined by compiler using long
word load/stores.

I also tried to make all copies in the natural orders of sk_buff,
to help hardware prefetching.

I made sure sk_buff size did not change.
Signed-off-by: NEric Dumazet <edumazet@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Given following iptables ruleset:

-P FORWARD DROP
-A FORWARD -m sctp --dport 9 -j ACCEPT
-A FORWARD -p tcp --dport 80 -j ACCEPT
-A FORWARD -p tcp -m conntrack -m state ESTABLISHED,RELATED -j ACCEPT

One would assume that this allows SCTP on port 9 and TCP on port 80.
Unfortunately, if the SCTP conntrack module is not loaded, this allows
*all* SCTP communication, to pass though, i.e. -p sctp -j ACCEPT,
which we think is a security issue.

This is because on the first SCTP packet on port 9, we create a dummy
"generic l4" conntrack entry without any port information (since
conntrack doesn't know how to extract this information).

All subsequent packets that are unknown will then be in established
state since they will fallback to proto_generic and will match the
'generic' entry.

Our originally proposed version [1] completely disabled generic protocol
tracking, but Jozsef suggests to not track protocols for which a more
suitable helper is available, hence we now mitigate the issue for in
tree known ct protocol helpers only, so that at least NAT and direction
information will still be preserved for others.

 [1] http://www.spinics.net/lists/netfilter-devel/msg33430.html

Joint work with Daniel Borkmann.
Signed-off-by: NFlorian Westphal <fw@strlen.de>
Signed-off-by: NDaniel Borkmann <dborkman@redhat.com>
Acked-by: NJozsef Kadlecsik <kadlec@blackhole.kfki.hu>
Signed-off-by: NPablo Neira Ayuso <pablo@netfilter.org>

We want to know in which cases the user explicitly sets the policy
options. In that case, we also want to dump back the info.
Signed-off-by: NArturo Borrero Gonzalez <arturo.borrero.glez@gmail.com>
Signed-off-by: NPablo Neira Ayuso <pablo@netfilter.org>

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: NDaniel Borkmann <dborkman@redhat.com>
Signed-off-by: NFlorian Westphal <fw@strlen.de>
Signed-off-by: NGlenn Judd <glenn.judd@morganstanley.com>
Acked-by: NStephen Hemminger <stephen@networkplumber.org>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

DataCenter TCP (DCTCP) determines cwnd growth based on ECN information
and ACK properties, e.g. ACK that updates window is treated differently
than DUPACK.

Also DCTCP needs information whether ACK was delayed ACK. Furthermore,
DCTCP also implements a CE state machine that keeps track of CE markings
of incoming packets.

Therefore, extend the congestion control framework to provide these
event types, so that DCTCP can be properly implemented as a normal
congestion algorithm module outside of the core stack.

Joint work with Daniel Borkmann and Glenn Judd.
Signed-off-by: NFlorian Westphal <fw@strlen.de>
Signed-off-by: NDaniel Borkmann <dborkman@redhat.com>
Signed-off-by: NGlenn Judd <glenn.judd@morganstanley.com>
Acked-by: NStephen Hemminger <stephen@networkplumber.org>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>