We currently use a per socket order-0 page cache for tcp_sendmsg()
operations.

This page is used to build fragments for skbs.

Its done to increase probability of coalescing small write() into
single segments in skbs still in write queue (not yet sent)

But it wastes a lot of memory for applications handling many mostly
idle sockets, since each socket holds one page in sk->sk_sndmsg_page

Its also quite inefficient to build TSO 64KB packets, because we need
about 16 pages per skb on arches where PAGE_SIZE = 4096, so we hit
page allocator more than wanted.

This patch adds a per task frag allocator and uses bigger pages,
if available. An automatic fallback is done in case of memory pressure.

(up to 32768 bytes per frag, thats order-3 pages on x86)

This increases TCP stream performance by 20% on loopback device,
but also benefits on other network devices, since 8x less frags are
mapped on transmit and unmapped on tx completion. Alexander Duyck
mentioned a probable performance win on systems with IOMMU enabled.

Its possible some SG enabled hardware cant cope with bigger fragments,
but their ndo_start_xmit() should already handle this, splitting a
fragment in sub fragments, since some arches have PAGE_SIZE=65536

Successfully tested on various ethernet devices.
(ixgbe, igb, bnx2x, tg3, mellanox mlx4)
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Ben Hutchings <bhutchings@solarflare.com>
Cc: Vijay Subramanian <subramanian.vijay@gmail.com>
Cc: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Vijay Subramanian <subramanian.vijay@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

This is primarily to address transmission timeout occurrences, when
multiple H/W Tx queues are being used concurrently. Because in
the priority scheduling mode the controller does not service the
Tx queues equally (but in ascending index order), Tx timeouts are
being triggered rightaway for a basic test with multiple simultaneous
connections like:
iperf -c <server_ip> -n 100M -P 8

resulting in kernel trace:
NETDEV WATCHDOG: eth1 (fsl-gianfar): transmit queue <X> timed out
------------[ cut here ]------------
WARNING: at net/sched/sch_generic.c:255
...
and controller reset during intense traffic, and possibly further
complications.

This patch changes the default H/W Tx scheduling setting (TXSCHED)
for multi-queue devices, from priority scheduling mode to a weighted
round robin mode with equal weights for all H/W Tx queues, and
addresses the issue above.
Signed-off-by: Claudiu Manoil <claudiu.manoil@freescale.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

loopback current mtu of 16436 bytes allows no more than 3 MSS TCP
segments per frame, or 48 Kbytes. Changing mtu to 64K allows TCP
stack to build large frames and significantly reduces stack overhead.

Performance boost on bulk TCP transferts can be up to 30 %, partly
because we now have one ACK message for two 64KB segments, and a lower
probability of hitting /proc/sys/net/ipv4/tcp_reordering default limit.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

All callers provide a non-NULL scm argument.
Signed-off-by: David S. Miller <davem@davemloft.net>

This patch changes the definition of bnx2x_tests_str_arr from static char
pointer to static const char bi-directional array. Also the
bnx2x_get_strings function is simplified.
Reported-by: Joe Perches <joe@perches.com>
Reported-by: David Laight <David.Laight@ACULAB.COM>
Signed-off-by: Merav Sicron <meravs@broadcom.com>
Signed-off-by: Eilon Greenstein <eilong@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Convert a nonnegative error return code to a negative one, as returned
elsewhere in the function.

A simplified version of the semantic match that finds this problem is as
follows: (http://coccinelle.lip6.fr/)

// <smpl>
(
if@p1 (\(ret < 0\|ret != 0\))
 { ... return ret; }
|
ret@p1 = 0
)
... when != ret = e1
    when != &ret
*if(...)
{
  ... when != ret = e2
      when forall
 return ret;
}
// </smpl>
Signed-off-by: Peter Senna Tschudin <peter.senna@gmail.com>
Signed-off-by: Ursula Braun <ursula.braun@de.ibm.com>
Signed-off-by: Frank Blaschka <frank.blaschka@de.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

removes unnecessary semicolon

Found by Coccinelle: http://coccinelle.lip6.fr/Signed-off-by: Peter Senna Tschudin <peter.senna@gmail.com>
Signed-off-by: Ursula Braun <ursula.braun@de.ibm.com>
Signed-off-by: Frank Blaschka <frank.blaschka@de.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

ipaddr has been allocated in function qeth_l3_add_vipa() but
does not free before leaving from the error handling cases. The
same problem also exists in function qeth_l3_add_rxip().

spatch with a semantic match is used to found this problem.
(http://coccinelle.lip6.fr/)
Signed-off-by: Wei Yongjun <yongjun_wei@trendmicro.com.cn>
Signed-off-by: Frank Blaschka <frank.blaschka@de.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Make sure that all ccws used for writing are initialized with
zeros - especially since the last ccw contains a TIC for which
the unused fields have to be zeros.
Signed-off-by: Sebastian Ott <sebott@linux.vnet.ibm.com>
Signed-off-by: Frank Blaschka <frank.blaschka@de.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Cleanup the qeth_get_channel_path_desc function and rename it
to qeth_update_from_chp_desc. No functional change.
Signed-off-by: Sebastian Ott <sebott@linux.vnet.ibm.com>
Signed-off-by: Frank Blaschka <frank.blaschka@de.ibm.com>
Acked-by: Ursula Braun <ursula.braun@de.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Pablo Neira Ayuso says:

====================
This patchset contains updates for your net-next tree, they are:

* Mostly fixes for the recently pushed IPv6 NAT support:

- Fix crash while removing nf_nat modules from Patrick McHardy.
- Fix unbalanced rcu_read_unlock from Ulrich Weber.
- Merge NETMAP and REDIRECT into one single xt_target module, from
  Jan Engelhardt.
- Fix Kconfig for IPv6 NAT, which allows inconsistent configurations,
  from myself.

* Updates for ipset, all of the from Jozsef Kadlecsik:

- Add the new "nomatch" option to obtain reverse set matching.
- Support for /0 CIDR in hash:net,iface set type.
- One non-critical fix for a rare crash due to pass really
  wrong configuration parameters.
- Coding style cleanups.
- Sparse fixes.
- Add set revision supported via modinfo.i

* One extension for the xt_time match, to support matching during
  the transition between two days with one single rule, from
  Florian Westphal.

* Fix maximum packet length supported by nfnetlink_queue and add
  NFQA_CAP_LEN attribute, from myself.

You can notice that this batch contains a couple of fixes that may
go to 3.6-rc but I don't consider them critical to push them:

* The ipset fix for the /0 cidr case, which is triggered with one
  inconsistent command line invocation of ipset.

* The nfnetlink_queue maximum packet length supported since it requires
  the new NFQA_CAP_LEN attribute to provide a full workaround for the
  described problem.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

This patch adds the NFQA_CAP_LEN attribute that allows us to know
what is the real packet size from user-space (even if we decided
to retrieve just a few bytes from the packet instead of all of it).

Security software that inspects packets should always check for
this new attribute to make sure that it is inspecting the entire
packet.

This also helps to provide a workaround for the problem described
in: http://marc.info/?l=netfilter-devel&m=134519473212536&w=2

Original idea from Florian Westphal.
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>

The packets that we send via NFQUEUE are encapsulated in the NFQA_PAYLOAD
attribute. The length of the packet in userspace is obtained via
attr->nla_len field. This field contains the size of the Netlink
attribute header plus the packet length.

If the maximum packet length is specified, ie. 65535 bytes, and
packets in the range of (65531,65535] are sent to userspace, the
attr->nla_len overflows and it reports bogus lengths to the
application.

To fix this, this patch limits the maximum packet length to 65531
bytes. If larger packet length is specified, the packet that we
send to user-space is truncated to 65531 bytes.

To support 65535 bytes packets, we have to revisit the idea of
the 32-bits Netlink attribute length.
Reported-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>

This patch allows the FTP helper to pickup the sequence tracking from
the first packet seen. This is useful to fix the breakage of the first
FTP command after the failover while using conntrackd to synchronize
states.

The seq_aft_nl_num field in struct nf_ct_ftp_info has been shrinked to
16-bits (enough for what it does), so we can use the remaining 16-bits
to store the flags while using the same size for the private FTP helper
data.
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>

Currently, if you want to do something like:
"match Monday, starting 23:00, for two hours"
You need two rules, one for Mon 23:00 to 0:00 and one for Tue 0:00-1:00.

The rule: --weekdays Mo --timestart 23:00  --timestop 01:00

looks correct, but it will first match on monday from midnight to 1 a.m.
and then again for another hour from 23:00 onwards.

This permits userspace to explicitly ignore the day transition and
match for a single, continuous time period instead.
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>

With recent kernel changes we can now return errors on a failure to setup a
VLAN filter.  This patch takes advantage of that opportunity so that we can
return either an EIO error in the case of a mailbox failure, or an EACCESS
error in the case of being denied access to the VLAN filter table by the
PF.
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Robert Garrett <robertx.e.garrett@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>

This change fixes the ixgbevf driver so that it can correctly drop a frame
should it receive a jumbo frame.
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>

While fixing up a patch from Alex Duyck to use q_vectors in ring containers
to update the ITR I bungled it and missed actually updating the counters
in the ring container q_vectors.  This patch fixes my mistake and makes
interrupt moderation actually work.
Signed-off-by: Greg Rose <gregory.v.rose@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>

Remove 'rx_ring' parameter as it is not used in ixgbevf_receive_skb
Signed-off-by: Narendra K <narendra_k@dell.com>
Acked-by: Greg Rose <gregory.v.rose@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>

The counter is not valid unless the controller is running in IOV mode.
Signed-off-by: Greg Rose <gregory.v.rose@intel.com>
Tested-by: Phil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>

The VF driver was not designed to correctly handle a message timeout.  As
a result it is possible for one bad message to invalidate all messages
following it until the part is reset.  Instead we should copy the example
in igbvf of how to handle a mailbox event and message timeout.
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Phil Schmitt <phillip.j.schmitt@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>

Suggested by Jan Engelhardt.
Signed-off-by: David S. Miller <davem@davemloft.net>

When recording the number of SYNACK retransmits for servers using TCP
Fast Open, fix the code to ensure that we copy over the retransmit
count from the request_sock after we receive the ACK that completes
the 3-way handshake.

The story here is similar to that of SYNACK RTT
measurements. Previously we were always doing this 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 copy the final SYNACK retransmit
count in tcp_rcv_state_process().

Note that copying over the SYNACK retransmit count will give us the
correct count since, as is mentioned in a comment in
tcp_retransmit_timer(), before we receive an ACK for our SYN-ACK a TFO
passive connection does not retransmit anything else (e.g., data or
FIN segments).
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

We're now using isa_virt_to_bus(), and there really
isn't a generic and consistent test for whether a
platform provides this interface or not.

This driver is also for an x86-only device.
Signed-off-by: David S. Miller <davem@davemloft.net>

Exceptions can now be matched and we can branch according to the
possible cases:

a. match in the set if the element is not flagged as "nomatch"
b. match in the set if the element is flagged with "nomatch"
c. no match

i.e.

iptables ... -m set --match-set ... -j ...
iptables ... -m set --match-set ... --nomatch-entries -j ...
...
Signed-off-by: Jozsef Kadlecsik <kadlec@blackhole.kfki.hu>

Signed-off-by: Jozsef Kadlecsik <kadlec@blackhole.kfki.hu>

Signed-off-by: Jozsef Kadlecsik <kadlec@blackhole.kfki.hu>

Now it is possible to setup a single hash:net,iface type of set and
a single ip6?tables match which covers all egress/ingress filtering.
Signed-off-by: Jozsef Kadlecsik <kadlec@blackhole.kfki.hu>

Signed-off-by: Jozsef Kadlecsik <kadlec@blackhole.kfki.hu>

A TCP Fast Open (TFO) passive connection must call both
tcp_check_req() and tcp_validate_incoming() for all incoming ACKs that
are attempting to complete the 3WHS.

This is needed to parallel all the action that happens for a non-TFO
connection, where for an ACK that is attempting to complete the 3WHS
we call both tcp_check_req() and tcp_validate_incoming().

For example, upon receiving the ACK that completes the 3WHS, we need
to call tcp_fast_parse_options() and update ts_recent based on the
incoming timestamp value in the ACK.

One symptom of the problem with the previous code was that for passive
TFO connections using TCP timestamps, the outgoing TS ecr values
ignored the incoming TS val value on the ACK that completed the 3WHS.
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Previously, when using TCP Fast Open a server would return from
tcp_check_req() before updating snt_synack based on TCP timestamp echo
replies and whether or not we've retransmitted the SYNACK. The result
was that (a) for TFO connections using timestamps we used an incorrect
baseline SYNACK send time (tcp_time_stamp of SYNACK send instead of
rcv_tsecr), and (b) for TFO connections that do not have TCP
timestamps but retransmit the SYNACK we took a SYNACK RTT sample when
we should not take a sample.

This fix merely moves the snt_synack update logic a bit earlier in the
function, so that connections using TCP Fast Open will properly do
these updates when the ACK for the SYNACK arrives.

Moving this snt_synack update logic means that with TCP_DEFER_ACCEPT
enabled we do a few instructions of wasted work on each bare ACK, but
that seems OK.
Signed-off-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>

There has been some confusion among PHC driver authors about the
intended purpose of the clock_name attribute. This patch expands the
documation in order to clarify how the clock_name field should be
understood.
Signed-off-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

PTP Hardware Clock devices appear as class devices in sysfs. This patch
changes the registration API to use the parent device, clarifying the
clock's relationship to the underlying device.
Signed-off-by: Richard Cochran <richardcochran@gmail.com>
Acked-by: Ben Hutchings <bhutchings@solarflare.com>
Acked-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

If the timex.mode field indicates a query, then we provide the value of
the current frequency adjustment.

[ Get rid of extraneous empty lines -DaveM ]
Signed-off-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

This patch adds a field to the representation of a PTP hardware clock in
order to remember the frequency adjustment value dialed by the user.

Adding this field will let us answer queries in the manner of adjtimex
in a follow on patch.
Signed-off-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Jeff Kirsher says:

====================
This series contains updates to igb only.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

One-shot mode uses the TCS bit of the status register to discern
whether a transmission was successful or not. On a failed
transmission, the frame is not echoed back.
Signed-off-by: Andreas Larsson <andreas@gaisler.com>
Acked-by: Wolfgang Grandegger <wg@grandegger.com>
Acked-by: Oliver Hartkopp <socketcan@hartkopp.net>
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>

This change is meant to improve performance on systems that do not require
the DMA unmap calls.  On those systems we do not need to make use of the
unmap address for Tx or the unmap length so we can drop both thereby
reducing the size of the Tx buffer info structure.

In addition I have changed the logic to check for unmap length instead of
unmap address when checking to see if a buffer needs to be unmapped from
DMA use.  The reasons for this change is that on some platforms it is
possible to receive a valid DMA address of 0 from an IOMMU.
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>