Tuning coalescing parameters on NIC can be really hard.

Servers can handle both bulk and RPC like traffic, with conflicting
goals : bulk flows want as big GRO packets as possible, RPC want minimal
latencies.

To reach big GRO packets on 10Gbe NIC, one can use :

ethtool -C eth0 rx-usecs 4 rx-frames 44

But this penalizes rpc sessions, with an increase of latencies, up to
50% in some cases, as NICs generally do not force an interrupt when
a packet with TCP Push flag is received.

Some NICs do not have an absolute timer, only a timer rearmed for every
incoming packet.

This patch uses a different strategy : Let GRO stack decides what do do,
based on traffic pattern.

Packets with Push flag wont be delayed.
Packets without Push flag might be held in GRO engine, if we keep
receiving data.

This new mechanism is off by default, and shall be enabled by setting
/sys/class/net/ethX/gro_flush_timeout to a value in nanosecond.

To fully enable this mechanism, drivers should use napi_complete_done()
instead of napi_complete().

Tested:
 Ran 200 netperf TCP_STREAM from A to B (10Gbe mlx4 link, 8 RX queues)

Without this feature, we send back about 305,000 ACK per second.

GRO aggregation ratio is low (811/305 = 2.65 segments per GRO packet)

Setting a timer of 2000 nsec is enough to increase GRO packet sizes
and reduce number of ACK packets. (811/19.2 = 42)

Receiver performs less calls to upper stacks, less wakes up.
This also reduces cpu usage on the sender, as it receives less ACK
packets.

Note that reducing number of wakes up increases cpu efficiency, but can
decrease QPS, as applications wont have the chance to warmup cpu caches
doing a partial read of RPC requests/answers if they fit in one skb.

B:~# sar -n DEV 1 10 | grep eth0 | tail -1
Average:         eth0 811269.80 305732.30 1199462.57  19705.72      0.00
0.00      0.50

B:~# echo 2000 >/sys/class/net/eth0/gro_flush_timeout

B:~# sar -n DEV 1 10 | grep eth0 | tail -1
Average:         eth0 811577.30  19230.80 1199916.51   1239.80      0.00
0.00      0.50
Signed-off-by: NEric Dumazet <edumazet@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Allow datapath to recognize and extract MPLS labels into flow keys
and execute actions which push, pop, and set labels on packets.

Based heavily on work by Leo Alterman, Ravi K, Isaku Yamahata and Joe Stringer.

Cc: Ravi K <rkerur@gmail.com>
Cc: Leo Alterman <lalterman@nicira.com>
Cc: Isaku Yamahata <yamahata@valinux.co.jp>
Cc: Joe Stringer <joe@wand.net.nz>
Signed-off-by: NSimon Horman <horms@verge.net.au>
Signed-off-by: NJesse Gross <jesse@nicira.com>
Signed-off-by: NPravin B Shelar <pshelar@nicira.com>

net_rx_action() can mask irqs a single time to transfert sd->poll_list
into a private list, for a very short duration.

Then, napi_complete() can avoid masking irqs again,
and net_rx_action() only needs to mask irq again in slow path.

This patch removes 2 couples of irq mask/unmask per typical NAPI run,
more if multiple napi were triggered.

Note this also allows to give control back to caller (do_softirq())
more often, so that other softirq handlers can be called a bit earlier,
or ksoftirqd can be wakeup earlier under pressure.

This was developed while testing an alternative to RX interrupt
mitigation to reduce latencies while keeping or improving GRO
aggregation on fast NIC.

Idea is to test napi->gro_list at the end of a napi->poll() and
reschedule one NAPI poll, but after servicing a full round of
softirqs (timers, TX, rcu, ...). This will be allowed only if softirq
is currently serviced by idle task or ksoftirqd, and resched not needed.
Signed-off-by: NEric Dumazet <edumazet@google.com>
Cc: Willem de Bruijn <willemb@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

napi_schedule() can be called from any context and has to mask hard
irqs.

Add a variant that can only be called from hard interrupts handlers
or when irqs are already masked.

Many NIC drivers can use it from their hard IRQ handler instead of
generic variant.
Signed-off-by: NEric Dumazet <edumazet@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Do not reuse skb if it was pfmemalloc tainted, otherwise
future frame might be dropped anyway.
Signed-off-by: NEric Dumazet <edumazet@google.com>
Signed-off-by: NRoman Gushchin <klamm@yandex-team.ru>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Add ndo_gso_check which a device can define to indicate whether is
is capable of doing GSO on a packet. This funciton would be called from
the stack to determine whether software GSO is needed to be done. A
driver should populate this function if it advertises GSO types for
which there are combinations that it wouldn't be able to handle. For
instance a device that performs UDP tunneling might only implement
support for transparent Ethernet bridging type of inner packets
or might have limitations on lengths of inner headers.
Signed-off-by: NTom Herbert <therbert@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Testing xmit_more support with netperf and connected UDP sockets,
I found strange dst refcount false sharing.

Current handling of IFF_XMIT_DST_RELEASE is not optimal.

Dropping dst in validate_xmit_skb() is certainly too late in case
packet was queued by cpu X but dequeued by cpu Y

The logical point to take care of drop/force is in __dev_queue_xmit()
before even taking qdisc lock.

As Julian Anastasov pointed out, need for skb_dst() might come from some
packet schedulers or classifiers.

This patch adds new helper to cleanly express needs of various drivers
or qdiscs/classifiers.

Drivers that need skb_dst() in their ndo_start_xmit() should call
following helper in their setup instead of the prior :

	dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
->
	netif_keep_dst(dev);

Instead of using a single bit, we use two bits, one being
eventually rebuilt in bonding/team drivers.

The other one, is permanent and blocks IFF_XMIT_DST_RELEASE being
rebuilt in bonding/team. Eventually, we could add something
smarter later.
Signed-off-by: NEric Dumazet <edumazet@google.com>
Cc: Julian Anastasov <ja@ssi.bg>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Marking this as static allows compiler to inline it.
Signed-off-by: NEric Dumazet <edumazet@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Some TSO engines might have a too heavy setup cost, that impacts
performance on hosts sending small bursts (2 MSS per packet).

This patch adds a device gso_min_segs, allowing drivers to set
a minimum segment size for TSO packets, according to the NIC
performance.

Tested on a mlx4 NIC, this allows to get a ~110% increase of
throughput when sending 2 MSS per packet.
Signed-off-by: NEric Dumazet <edumazet@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Its unfortunate we have to walk again skb list to find the tail
after segmentation, even if data is probably hot in cpu caches.

skb_segment() can store the tail of the list into segs->prev,
and validate_xmit_skb_list() can immediately get the tail.
Signed-off-by: NEric Dumazet <edumazet@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Validation of skb can be pretty expensive :

GSO segmentation and/or checksum computations.

We can do this without holding qdisc lock, so that other cpus
can queue additional packets.

Trick is that requeued packets were already validated, so we carry
a boolean so that sch_direct_xmit() can validate a fresh skb list,
or directly use an old one.

Tested on 40Gb NIC (8 TX queues) and 200 concurrent flows, 48 threads
host.

Turning TSO on or off had no effect on throughput, only few more cpu
cycles. Lock contention on qdisc lock disappeared.

Same if disabling TX checksum offload.
Signed-off-by: NEric Dumazet <edumazet@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Per commit "77873803 net_dma: mark broken" net_dma is no longer used
and there is no plan to fix it.

This is the mechanical removal of bits in CONFIG_NET_DMA ifdef guards.
Reverting the remainder of the net_dma induced changes is deferred to
subsequent patches.

Marked for stable due to Roman's report of a memory leak in
dma_pin_iovec_pages():

    https://lkml.org/lkml/2014/9/3/177

Cc: Dave Jiang <dave.jiang@intel.com>
Cc: Vinod Koul <vinod.koul@intel.com>
Cc: David Whipple <whipple@securedatainnovations.ch>
Cc: Alexander Duyck <alexander.h.duyck@intel.com>
Cc: <stable@vger.kernel.org>
Reported-by: NRoman Gushchin <klamm@yandex-team.ru>
Acked-by: NDavid S. Miller <davem@davemloft.net>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

No caller or macro uses the return value so make all
the functions return void.
Signed-off-by: NJoe Perches <joe@perches.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

The send_check logic was only interesting in cases of TCP offload and
UDP UFO where the checksum needed to be initialized to the pseudo
header checksum. Now we've moved that logic into the related
gso_segment functions so gso_send_check is no longer needed.
Signed-off-by: NTom Herbert <therbert@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Commit ce93718f ("net: Don't keep
around original SKB when we software segment GSO frames") frees the
original skb after software GSO even for dodgy gso skbs. This breaks
the stream throughput from untrusted sources, since only header
checking was done during software GSO instead of a true
segmentation. This patch fixes this by freeing the original gso skb
only when it was really segmented by software.

Fixes ce93718f ("net: Don't keep
around original SKB when we software segment GSO frames.")

Cc: David S. Miller <davem@davemloft.net>
Cc: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: NJason Wang <jasowang@redhat.com>
Acked-by: NEric Dumazet <edumazet@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

__netdev_adjacent_dev_insert may add adjust device of different net
namespace, without proper check it leads to emergence of broken
sysfs links from/to devices in another namespace.
Fix: rewrite netdev_adjacent_is_neigh_list macro as a function,
     move net_eq check into netdev_adjacent_is_neigh_list.
     (thanks David)
     related to: 4c75431aSigned-off-by: NAlexander Fomichev <git.user@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

These code is now protected by rtnl lock, rcu read lock
is useless now.
Signed-off-by: NCong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Add __rcu notation to qdisc handling by doing this we can make
smatch output more legible. And anyways some of the cases should
be using rcu_dereference() see qdisc_all_tx_empty(),
qdisc_tx_chainging(), and so on.

Also *wake_queue() API is commonly called from driver timer routines
without rcu lock or rtnl lock. So I added rcu_read_lock() blocks
around netif_wake_subqueue and netif_tx_wake_queue.
Signed-off-by: NJohn Fastabend <john.r.fastabend@intel.com>
Acked-by: NEric Dumazet <edumazet@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Add __rcu notation to qdisc handling by doing this we can make
smatch output more legible. And anyways some of the cases should
be using rcu_dereference() see qdisc_all_tx_empty(),
qdisc_tx_chainging(), and so on.

Also *wake_queue() API is commonly called from driver timer routines
without rcu lock or rtnl lock. So I added rcu_read_lock() blocks
around netif_wake_subqueue and netif_tx_wake_queue.
Signed-off-by: NJohn Fastabend <john.r.fastabend@intel.com>
Acked-by: NEric Dumazet <edumazet@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

In commit 50cbe9ab ("net: Validate xmit SKBs right when we
pull them out of the qdisc") the validation code was moved out of
dev_hard_start_xmit and into dequeue_skb.

However this overlooked the fact that we do not always enqueue
the skb onto a qdisc. First situation is if qdisc have flag
TCQ_F_CAN_BYPASS and qdisc is empty.  Second situation is if
there is no qdisc on the device, which is a common case for
software devices.

Originally spotted and inital patch by Alexander Duyck.
As a result Alex was seeing issues trying to connect to a
vhost_net interface after commit 50cbe9ab was applied.

Added a call to validate_xmit_skb() in __dev_xmit_skb(), in the
code path for qdiscs with TCQ_F_CAN_BYPASS flag, and in
__dev_queue_xmit() when no qdisc.

Also handle the error situation where dev_hard_start_xmit() could
return a skb list, and does not return dev_xmit_complete(rc) and
falls through to the kfree_skb(), in that situation it should
call kfree_skb_list().

Fixes:  50cbe9ab ("net: Validate xmit SKBs right when we pull them out of the qdisc")
Signed-off-by: NAlexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: NJesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This flag indicates that an invalid checksum was detected in the
packet. __skb_mark_checksum_bad helper function was added to set this.

Checksums can be marked bad from a driver or the GRO path (the latter
is implemented in this patch). csum_bad is checked in
__skb_checksum_validate_complete (i.e. calling that when ip_summed ==
CHECKSUM_NONE).

csum_bad works in conjunction with ip_summed value. In the case that
ip_summed is CHECKSUM_NONE and csum_bad is set, this implies that the
first (or next) checksum encountered in the packet is bad. When
ip_summed is CHECKSUM_UNNECESSARY, the first checksum after the last
one validated is bad. For example, if ip_summed == CHECKSUM_UNNECESSARY,
csum_level == 1, and csum_bad is set-- then the third checksum in the
packet is bad. In the normal path, the packet will be dropped when
processing the protocol layer of the bad checksum:
__skb_decr_checksum_unnecessary called twice for the good checksums
changing ip_summed to CHECKSUM_NONE so that
__skb_checksum_validate_complete is called to validate the third
checksum and that will fail since csum_bad is set.
Signed-off-by: NTom Herbert <therbert@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Now fundamentally we can process lists of SKBs as cheaply
as single packets.
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Just maintain the list properly by returning the head of the remaining
SKB list from dev_hard_start_xmit().
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Signed-off-by: NDavid S. Miller <davem@davemloft.net>

dev_hard_start_xmit() does two things, it first validates and
canonicalizes the SKB, then it actually sends it.

Make a set of helper functions for doing the first part.
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Signed-off-by: NDavid S. Miller <davem@davemloft.net>

For now it will always be false.
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

There is a slight policy change happening here as well.

The previous code would drop the entire rest of the GSO skb if any of
them got, for example, a congestion notification.

That makes no sense, anything NET_XMIT_MASK and below is something
like congestion or policing.  And in the congestion case it doesn't
even mean the packet was actually dropped.

Just continue until dev_xmit_complete() evaluates to false.
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Hopefully making the code a bit easier to read and digest.
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

That way we don't have to audit every call site to make sure it is
doing this properly.
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Allow GRO path to "consume" checksums provided in CHECKSUM_UNNECESSARY
and to report new checksums verfied for use in fallback to normal
path.

Change GRO checksum path to track csum_level using a csum_cnt field
in NAPI_GRO_CB. On GRO initialization, if ip_summed is
CHECKSUM_UNNECESSARY set NAPI_GRO_CB(skb)->csum_cnt to
skb->csum_level + 1. For each checksum verified, decrement
NAPI_GRO_CB(skb)->csum_cnt while its greater than zero. If a checksum
is verfied and NAPI_GRO_CB(skb)->csum_cnt == 0, we have verified a
deeper checksum than originally indicated in skbuf so increment
csum_level (or initialize to CHECKSUM_UNNECESSARY if ip_summed is
CHECKSUM_NONE or CHECKSUM_COMPLETE).
Signed-off-by: NTom Herbert <therbert@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Replace uses of get_cpu_var for address calculation through this_cpu_ptr.

Cc: netdev@vger.kernel.org
Cc: Eric Dumazet <edumazet@google.com>
Acked-by: NDavid S. Miller <davem@davemloft.net>
Signed-off-by: NChristoph Lameter <cl@linux.com>
Signed-off-by: NTejun Heo <tj@kernel.org>

This is follow-up to

  da08143b ("vlan: more careful checksum features handling")

which introduced more careful feature intersection in vlan code,
taking into account that HW_CSUM should be considered superset
of IP_CSUM/IPV6_CSUM. The same is needed in netif_skb_features()
in order to avoid offloading mismatch warning when vlan is
created on top of a bond consisting of slaves supporting IP/IPv6
checksumming but not vlan Tx offloading.
Signed-off-by: NMichal Kubecek <mkubecek@suse.cz>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Code manipulating sysfs symlinks on adjacent net_devices(s)
currently doesn't take into account that devices potentially
belong to different namespaces.

This patch trying to fix an issue as follows:
- check for net_ns before creating / deleting symlink.
  for now only netdev_adjacent_rename_links and
  __netdev_adjacent_dev_remove are affected, afaics
  __netdev_adjacent_dev_insert implies both net_devs
  belong to the same namespace.
- Drop all existing symlinks to / from all adj_devs before
  switching namespace and recreate them just after.
Signed-off-by: NAlexander Y. Fomichev <git.user@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Add skb_gro_checksum_validate, skb_gro_checksum_validate_zero_check,
and skb_gro_checksum_simple_validate, and __skb_gro_checksum_complete.
These are the cognates of the normal checksum functions but are used
in the gro_receive path and operate on GRO related fields in sk_buffs.
Signed-off-by: NTom Herbert <therbert@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Replace open codings of (((u64) <x> * <y>) >> 32) with reciprocal_scale().
Signed-off-by: NDaniel Borkmann <dborkman@redhat.com>
Cc: Hannes Frederic Sowa <hannes@stressinduktion.org>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Currently the functionality to untag traffic on input resides
as part of the vlan module and is build only when VLAN support
is enabled in the kernel.  When VLAN is disabled, the function
vlan_untag() turns into a stub and doesn't really untag the
packets.  This seems to create an interesting interaction
between VMs supporting checksum offloading and some network drivers.

There are some drivers that do not allow the user to change
tx-vlan-offload feature of the driver.  These drivers also seem
to assume that any VLAN-tagged traffic they transmit will
have the vlan information in the vlan_tci and not in the vlan
header already in the skb.  When transmitting skbs that already
have tagged data with partial checksum set, the checksum doesn't
appear to be updated correctly by the card thus resulting in a
failure to establish TCP connections.

The following is a packet trace taken on the receiver where a
sender is a VM with a VLAN configued.  The host VM is running on
doest not have VLAN support and the outging interface on the
host is tg3:
10:12:43.503055 52:54:00:ae:42:3f > 28:d2:44:7d:c2:de, ethertype 802.1Q
(0x8100), length 78: vlan 100, p 0, ethertype IPv4, (tos 0x0, ttl 64, id 27243,
offset 0, flags [DF], proto TCP (6), length 60)
    10.0.100.1.58545 > 10.0.100.10.ircu-2: Flags [S], cksum 0xdc39 (incorrect
-> 0x48d9), seq 1069378582, win 29200, options [mss 1460,sackOK,TS val
4294837885 ecr 0,nop,wscale 7], length 0
10:12:44.505556 52:54:00:ae:42:3f > 28:d2:44:7d:c2:de, ethertype 802.1Q
(0x8100), length 78: vlan 100, p 0, ethertype IPv4, (tos 0x0, ttl 64, id 27244,
offset 0, flags [DF], proto TCP (6), length 60)
    10.0.100.1.58545 > 10.0.100.10.ircu-2: Flags [S], cksum 0xdc39 (incorrect
-> 0x44ee), seq 1069378582, win 29200, options [mss 1460,sackOK,TS val
4294838888 ecr 0,nop,wscale 7], length 0

This connection finally times out.

I've only access to the TG3 hardware in this configuration thus have
only tested this with TG3 driver.  There are a lot of other drivers
that do not permit user changes to vlan acceleration features, and
I don't know if they all suffere from a similar issue.

The patch attempt to fix this another way.  It moves the vlan header
stipping code out of the vlan module and always builds it into the
kernel network core.  This way, even if vlan is not supported on
a virtualizatoin host, the virtual machines running on top of such
host will still work with VLANs enabled.

CC: Patrick McHardy <kaber@trash.net>
CC: Nithin Nayak Sujir <nsujir@broadcom.com>
CC: Michael Chan <mchan@broadcom.com>
CC: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: NVladislav Yasevich <vyasevic@redhat.com>
Acked-by: NJiri Pirko <jiri@resnulli.us>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Kernel transmit latency is often incurred in the packet scheduler.
Introduce a new timestamp on transmission just before entering the
scheduler. When data travels through multiple devices (bonding,
tunneling, ...) each device will export an individual timestamp.
Signed-off-by: NWillem de Bruijn <willemb@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

No need for the unlikely(), WARN_ON() and BUG_ON() internally use
unlikely() on the condition.
Signed-off-by: NThomas Graf <tgraf@suug.ch>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>