Skip to content
  1. Dec 06, 2018
  3. Dec 05, 2018
    • Eric Dumazet's avatar
      tcp: reduce POLLOUT events caused by TCP_NOTSENT_LOWAT · a74f0fa0
      Eric Dumazet authored 
      

TCP_NOTSENT_LOWAT socket option or sysctl was added in linux-3.12
as a step to enable bigger tcp sndbuf limits.

It works reasonably well, but the following happens :

Once the limit is reached, TCP stack generates
an [E]POLLOUT event for every incoming ACK packet.

This causes a high number of context switches.

This patch implements the strategy David Miller added
in sock_def_write_space() :

 - If TCP socket has a notsent_lowat constraint of X bytes,
   allow sendmsg() to fill up to X bytes, but send [E]POLLOUT
   only if number of notsent bytes is below X/2

This considerably reduces TCP_NOTSENT_LOWAT overhead,
while allowing to keep the pipe full.

Tested:
 100 ms RTT netem testbed between A and B, 100 concurrent TCP_STREAM

A:/# cat /proc/sys/net/ipv4/tcp_wmem
4096	262144	64000000
A:/# super_netperf 100 -H B -l 1000 -- -K bbr &

A:/# grep TCP /proc/net/sockstat
TCP: inuse 203 orphan 0 tw 19 alloc 414 mem 1364904 # This is about 54 MB of memory per flow :/

A:/# vmstat 5 5
procs -----------memory---------- ---swap-- -----io---- -system-- ------cpu-----
 r  b   swpd   free   buff  cache   si   so    bi    bo   in   cs us sy id wa st
 0  0      0 256220672  13532 694976    0    0    10     0   28   14  0  1 99  0  0
 2  0      0 256320016  13532 698480    0    0   512     0 715901 5927  0 10 90  0  0
 0  0      0 256197232  13532 700992    0    0   735    13 771161 5849  0 11 89  0  0
 1  0      0 256233824  13532 703320    0    0   512    23 719650 6635  0 11 89  0  0
 2  0      0 256226880  13532 705780    0    0   642     4 775650 6009  0 12 88  0  0

A:/# echo 2097152 >/proc/sys/net/ipv4/tcp_notsent_lowat

A:/# grep TCP /proc/net/sockstat
TCP: inuse 203 orphan 0 tw 19 alloc 414 mem 86411 # 3.5 MB per flow

A:/# vmstat 5 5  # check that context switches have not inflated too much.
procs -----------memory---------- ---swap-- -----io---- -system-- ------cpu-----
 r  b   swpd   free   buff  cache   si   so    bi    bo   in   cs us sy id wa st
 2  0      0 260386512  13592 662148    0    0    10     0   17   14  0  1 99  0  0
 0  0      0 260519680  13592 604184    0    0   512    13 726843 12424  0 10 90  0  0
 1  1      0 260435424  13592 598360    0    0   512    25 764645 12925  0 10 90  0  0
 1  0      0 260855392  13592 578380    0    0   512     7 722943 13624  0 11 88  0  0
 1  0      0 260445008  13592 601176    0    0   614    34 772288 14317  0 10 90  0  0

Signed-off-by: default avatarEric Dumazet <edumazet@google.com>
Acked-by: default avatarSoheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      a74f0fa0
    • David S. Miller's avatar
      Merge branch 'act_tunnel_key-support-key-less-tunnels' · 4dc88ce6
      David S. Miller authored 
      

Or Gerlitz says:

====================
net/sched: act_tunnel_key: support key-less tunnels

This short series from Adi Nissim allows to support key-less tunnels
by the tc tunnel key actions, which is needed for some GRE use-cases.

changes from V0:
 - addresses build warning spotted by kbuild, make sure to always init
   to zero the tunnel key
====================

Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      4dc88ce6
    • Adi Nissim's avatar
      net/sched: act_tunnel_key: Don't dump dst port if it wasn't set · 1c25324c
      Adi Nissim authored 
      

It's possible to set a tunnel without a destination port. However,
on dump(), a zero dst port is returned to user space even if it was not
set, fix that.

Note that so far it wasn't required, b/c key less tunnels were not
supported and the UDP tunnels do require destination port.

Signed-off-by: default avatarAdi Nissim <adin@mellanox.com>
Reviewed-by: default avatarOz Shlomo <ozsh@mellanox.com>
Acked-by: default avatarJiri Pirko <jiri@mellanox.com>
Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      1c25324c
    • Adi Nissim's avatar
      net/sched: act_tunnel_key: Allow key-less tunnels · 80ef0f22
      Adi Nissim authored 
      

Allow setting a tunnel without a tunnel key. This is required for
tunneling protocols, such as GRE, that define the key as an optional
field.

Signed-off-by: default avatarAdi Nissim <adin@mellanox.com>
Acked-by: default avatarOr Gerlitz <ogerlitz@mellanox.com>
Reviewed-by: default avatarOz Shlomo <ozsh@mellanox.com>
Acked-by: default avatarJiri Pirko <jiri@mellanox.com>
Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      80ef0f22
    • Colin Ian King's avatar
      qed: fix spelling mistake "Dispalying" -> "Displaying" · d1ecf8a6
      Colin Ian King authored 
      

There is a spelling mistake in a DP_NOTICE message, fix it.

Signed-off-by: default avatarColin Ian King <colin.king@canonical.com>
Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      d1ecf8a6
    • David S. Miller's avatar
      Merge branch 'mlxsw-Add-one-armed-router-support' · 55827458
      David S. Miller authored 
      

Ido Schimmel says:

====================
mlxsw: Add one-armed router support

Up until now, when a packet was routed by the ASIC through the same
router interface (RIF) from which it ingressed from, the ASIC passed the
sole copy of the packet to the kernel. This allowed the kernel to route
the packet and also potentially generate an ICMP redirect.

There are scenarios (e.g., "one-armed router") where packets are
intentionally routed this way and are therefore not deemed as
exceptions. In such scenarios the current method of trapping packets to
the CPU is problematic, as it results in major packet loss.

This patchset solves the problem by having the ASIC forward the packet,
but also send a copy to the CPU, which gives the kernel the opportunity
to generate required exceptions.

To prevent the kernel from forwarding such packets again, the driver
marks them with 'offload_l3_fwd_mark', which causes the kernel to
consume them in ip{,6}_forward_finish().

Patch #1 renames 'offload_mr_fwd_mark' to 'offload_l3_fwd_mark'. When
set, the field indicates that a packet was already forwarded in L3
(unicast / multicast) by a capable device.

Patch #2 teaches the kernel to consume unicast packets that have
'offload_l3_fwd_mark' set.

Patch #3 changes mlxsw to mirror loopbacked (iRIF == eRIF) packets,
instead of trapping them.

Patch #4 adds a test case for above mentioned scenario.
====================

Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      55827458
    • Ido Schimmel's avatar
      selftests: mlxsw: Add one-armed router test · b6f153d3
      Ido Schimmel authored 
      

Construct a "one-armed router" topology and test that packets are
forwarded by the ASIC and that a copy of the packet is sent to the
kernel, which does not forward the packet again.

Signed-off-by: default avatarIdo Schimmel <idosch@mellanox.com>
Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      b6f153d3
    • Ido Schimmel's avatar
      mlxsw: spectrum: Mirror loopbacked packets instead of trapping them · 2f4f4494
      Ido Schimmel authored 
      

When the ASIC detects that a unicast packet is routed through the same
router interface (RIF) from which it ingressed (iRIF == eRIF), it raises
a trap called loopback error (LBERROR).

Thus far, this trap was configured to send a sole copy of the packet to
the CPU so that ICMP redirect packets could be potentially generated by
the kernel.

This is problematic as the CPU cannot forward packets at 3.2Tb/s and
there are scenarios (e.g., "one-armed router") where iRIF == eRIF is not
an exception.

Solve this by changing the trap to send a copy of the packet to the CPU.
To prevent the kernel from forwarding the packet again, it is marked
with 'offload_l3_fwd_mark'.

The trap is configured in a trap group of its own with a dedicated
policer in order not to prevent packets trapped by other traps from
reaching the CPU.

Signed-off-by: default avatarIdo Schimmel <idosch@mellanox.com>
Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      2f4f4494
    • Ido Schimmel's avatar
      net: Do not route unicast IP packets twice · f839a6c9
      Ido Schimmel authored 
      

Packets marked with 'offload_l3_fwd_mark' were already forwarded by a
capable device and should not be forwarded again by the kernel.
Therefore, have the kernel consume them.

The check is performed in ip{,6}_forward_finish() in order to allow the
kernel to process such packets in ip{,6}_forward() and generate required
exceptions. For example, ICMP redirects.

Signed-off-by: default avatarIdo Schimmel <idosch@mellanox.com>
Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      f839a6c9
    • Ido Schimmel's avatar
      skbuff: Rename 'offload_mr_fwd_mark' to 'offload_l3_fwd_mark' · 875e8939
      Ido Schimmel authored 
      Commit abf4bb6b

 ("skbuff: Add the offload_mr_fwd_mark field") added
the 'offload_mr_fwd_mark' field to indicate that a packet has already
undergone L3 multicast routing by a capable device. The field is used to
prevent the kernel from forwarding a packet through a netdev through
which the device has already forwarded the packet.

Currently, no unicast packet is routed by both the device and the
kernel, but this is about to change by subsequent patches and we need to
be able to mark such packets, so that they will no be forwarded twice.

Instead of adding yet another field to 'struct sk_buff', we can just
rename 'offload_mr_fwd_mark' to 'offload_l3_fwd_mark', as a packet
either has a multicast or a unicast destination IP.

While at it, add a comment about both 'offload_fwd_mark' and
'offload_l3_fwd_mark'.

Signed-off-by: default avatarIdo Schimmel <idosch@mellanox.com>
Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
      875e8939
  5. Dec 04, 2018