This is second part of dealing with suboptimal device gso parameters.
In first patch (350c9f48 "tcp_bbr: better deal with suboptimal GSO")
we dealt with devices having low gso_max_segs

Some devices lower gso_max_size from 64KB to 16 KB (r8152 is an example)

In order to probe an optimal cwnd, we want BBR being not sensitive
to whatever GSO constraint a device can have.

This patch removes tso_segs_goal() CC callback in favor of
min_tso_segs() for CC wanting to override sysctl_tcp_min_tso_segs

Next patch will remove bbr->tso_segs_goal since it does not have
to be persistent.
Signed-off-by: NEric Dumazet <edumazet@google.com>
Acked-by: NNeal Cardwell <ncardwell@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

recvmmsg does not call ___sys_recvmsg when sk_err is set.
That is fine for normal reads but, for MSG_ERRQUEUE, recvmmsg
should always call ___sys_recvmsg regardless of sk->sk_err to
be able to clear error queue. Otherwise, users are not able to
drain the error queue using recvmmsg.
Signed-off-by: NSoheil Hassas Yeganeh <soheil@google.com>
Reviewed-by: NEric Dumazet <edumazet@google.com>
Signed-off-by: NWillem de Bruijn <willemb@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Florian Fainelli says:

====================
net: phy: Reduce duplication

This patch series reduces the duplication among 10G PHY drivers that just
essentially stub most functions, but do that while replicating what the existing
generic functions do.

Changes in v3:

- removed unused "reg" variable in teranetics.c
- fixed subject for patch 5 since we actually use gen10g_no_soft_reset()

Changes in v2:

- rename gen10g_soft_reset() to gen10g_no_soft_reset() to better illustrate
  what it does (or does not)
- removed stray comment in marvell10g.c
====================
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

We do the same thing as the generic function: nothing, so utilize it.
Signed-off-by: NFlorian Fainelli <f.fainelli@gmail.com>

cortina_soft_reset() does the same thing as gen10g_soft_reset(), and
cortina_config_aneg() is actually doing what gen10g_config_init() does
for 10G capable PHYs.
Signed-off-by: NFlorian Fainelli <f.fainelli@gmail.com>

Update teranetics_aneg_done() to use genphy_c45_aneg_done() instead of
duplicating that code, and switch to gen10g_* functions where
appropriate instead of maintaining identical copies doing nothing.
Signed-off-by: NFlorian Fainelli <f.fainelli@gmail.com>

In order to remove a fair amount of duplication in the different 10G PHY
drivers, export all gen10g_* functions to be able to make use of those.
While we are at it, rename gen10g_soft_reset() to gen10g_no_soft_reset()
to illustrate what it does.
Signed-off-by: NFlorian Fainelli <f.fainelli@gmail.com>

The driver duplicates what the generic function does, so use the generic
function intead.
Signed-off-by: NFlorian Fainelli <f.fainelli@gmail.com>

Finn Thain says:

====================
Fixes, cleanup and modernization for mac89x0 driver

Changes since v4 of combined patch series:
- Removed redundant and non-portable MACH_IS_MAC tests.
- Added acked-by tags from Geert Uytterhoeven.
- Omitted patches unrelated to mac89x0 driver.
====================
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Adopt the conventional style of debug logging because it is both
shorter and more flexible.
Remove the 'version_printed' flag as the version will be printed
only once anyway (when the module loads).
Signed-off-by: NFinn Thain <fthain@telegraphics.com.au>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Fix log message fragments that no longer produce the desired output
since the behaviour of printk() was changed.
Add missing printk severity levels.
Drop deprecated "out of memory" message as per checkpatch advice.
Signed-off-by: NFinn Thain <fthain@telegraphics.com.au>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Apparently these Dayna cards don't have a pseudoslot declaration ROM
which means they can't be probed like NuBus cards.

Cc: Geert Uytterhoeven <geert@linux-m68k.org>
Signed-off-by: NFinn Thain <fthain@telegraphics.com.au>
Acked-by: NGeert Uytterhoeven <geert@linux-m68k.org>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Signed-off-by: NFinn Thain <fthain@telegraphics.com.au>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

David Ahern says:

====================
selftests: forwarding: misc bug fixes and enhancements

Bug fixes and an enhancement for the recent forwarding tests:
- only check tc version on tc tests
- handle multipath tests failing with 0 packet count
- fix ping command for IPv6 on Debian jessie
- improve summary of multipath tests

v2
- add CHECK_TC to bridge_vlan_aware.sh (Ido)
- dropped patch 2; always check for mz given its use
- fixed commit message for the last patch (Multipath: was dropped)
====================
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Add a better description to the summary for multipath tests. e.g.,

INFO: Running IPv6 multipath tests
TEST: ECMP                                               [PASS]
INFO: Expected ratio 1.00 Measured ratio 1.02
TEST: Weighted MP 2:1                                    [PASS]
INFO: Expected ratio 2.00 Measured ratio 2.02
TEST: Weighted MP 11:45                                  [PASS]
INFO: Expected ratio 4.09 Measured ratio 4.03
Signed-off-by: NDavid Ahern <dsahern@gmail.com>
Reviewed-by: NIdo Schimmel <idosch@mellanox.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

On Debian jessie ping can not handle IPv6 addresses so the command
fails. Use PING6 which is set to ping6.
Signed-off-by: NDavid Ahern <dsahern@gmail.com>
Reviewed-by: NIdo Schimmel <idosch@mellanox.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

If the packet stats have a difference of 0, the test output shows:
INFO: Expected ratio 2.00 Measured ratio
Runtime error (func=(main), adr=9): Divide by zero
(standard_in) 2: syntax error
(standard_in) 1: syntax error
./router_multipath.sh: line 187: test: : integer expression expected
TEST: Multipath                                                     [FAIL]
	Too large discrepancy between expected and measured ratios

Handle the 0 and display a cleaner message:
INFO: Running IPv6 multipath tests
TEST: Multipath                                                     [FAIL]
	Packet difference is 0
Signed-off-by: NDavid Ahern <dsahern@gmail.com>
Reviewed-by: NIdo Schimmel <idosch@mellanox.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Capabilities of tc command are irrelevant for router tests:
    $ ./router.sh
    SKIP: iproute2 too old, missing shared block support

Add a CHECK_TC flag and only check tc capabilities if set. Add flag to
tc_common.sh and have it sourced before lib.sh

Also, if the command lacks some feature the test should exit non-0.
Signed-off-by: NDavid Ahern <dsahern@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Fixes: bfff4862 ("net: fib_rules: support for match on ip_proto, sport and dport")
Reported-by: NEric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: NRoopa Prabhu <roopa@cumulusnetworks.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

According to RFC 4429 (section 3.1), adding new IPv6 addresses as
optimistic addresses is acceptable, as long as the implementation
follows some rules:

   * Optimistic DAD SHOULD only be used when the implementation is aware
        that the address is based on a most likely unique interface
        identifier (such as in [RFC2464]), generated randomly [RFC3041],
        or by a well-distributed hash function [RFC3972] or assigned by
        Dynamic Host Configuration Protocol for IPv6 (DHCPv6) [RFC3315].
        Optimistic DAD SHOULD NOT be used for manually entered
        addresses.

Thus, it seems reasonable to allow userspace to set the optimistic flag
when adding new addresses.

We must not let userspace set NODAD + OPTIMISTIC, since if the kernel is
not performing DAD we would never clear the optimistic flag. We must
also ignore userspace's request to add OPTIMISTIC flag to addresses that
have already completed DAD (addresses that don't have the TENTATIVE
flag, or that have the DADFAILED flag).

Then we also need to clear the OPTIMISTIC flag on permanent addresses
when DAD fails. Otherwise, IFA_F_OPTIMISTIC addresses added by userspace
can still be used after DAD has failed, because in
ipv6_chk_addr_and_flags(), IFA_F_OPTIMISTIC overrides IFA_F_TENTATIVE.

Setting IFA_F_OPTIMISTIC from userspace is conditional on
CONFIG_IPV6_OPTIMISTIC_DAD and the optimistic_dad sysctl.
Signed-off-by: NSabrina Dubroca <sd@queasysnail.net>
Reviewed-by: NStefano Brivio <sbrivio@redhat.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Fix trivial spelling mistake "greater then" -> "greater than".
Signed-off-by: NGal Pressman <galp@mellanox.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

As a part of working on MII time stamping infrastructure, I was trying
to figure out how netdev->phydev gets assigned, and I stumbled across
this.  Ever since the new phylink code came in, the field is assigned
twice.

The function, phylink_connect_phy(), calls

	phy_attach_direct()
	phylink_bringup_phy()

and phy_attach_direct() sets

	dev->phydev = phydev;

but phylink_bringup_phy() then sets the same field again:

	pl->netdev->phydev = phy;

Similarly, the function, phylink_of_phy_connect(), calls

	of_phy_attach()
		phy_attach_direct()
	phylink_bringup_phy()

The removal code is also duplicated:

phylink_disconnect_phy()
	pl->netdev->phydev = NULL;
	phy_disconnect()
		phy_detach()
			phydev->attached_dev->phydev = NULL;

This patch removes the redundant assignments, restricting manipulation
of the netdev.phydev field to phy_attach_direct() and phy_detach().
Signed-off-by: NRichard Cochran <richardcochran@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Ursula Braun says:

====================
net/smc: Link Layer Control enhancements

here is a series of smc patches enabling SMC communication with peers
supporting more than one link per link group.

The first three patches are preparing code cleanups.
====================
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

When the processing of a DELETE LINK message has started,
new connections should not be added to the link group that
is about to terminate.
Signed-off-by: NKarsten Graul <kgraul@linux.vnet.ibm.com>
Signed-off-by: NUrsula Braun <ubraun@linux.vnet.ibm.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Add initial support for the LLC messages ADD LINK and DELETE LINK.
Introduce a link state field. Extend the initial LLC handshake with
ADD LINK processing.
Signed-off-by: NKarsten Graul <kgraul@linux.vnet.ibm.com>
Signed-off-by: NUrsula Braun <ubraun@linux.vnet.ibm.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

SMC does not support eyecatchers in RMB elements,
decline peers requesting this support.
Signed-off-by: NKarsten Graul <kgraul@linux.vnet.ibm.com>
Signed-off-by: NUrsula Braun <ubraun@linux.vnet.ibm.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Process and respond to CONFIRM RKEY and DELETE RKEY messages.
Signed-off-by: NKarsten Graul <kgraul@linux.vnet.ibm.com>
Signed-off-by: NUrsula Braun <ubraun@linux.vnet.ibm.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Add TEST LINK message responses, which also serves as preparation for
support of sockopt TCP_KEEPALIVE.
Signed-off-by: NKarsten Graul <kgraul@linux.vnet.ibm.com>
Signed-off-by: NUrsula Braun <ubraun@linux.vnet.ibm.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

The daddr field holds the destination IPv4 address. The field was set but
never used and can be removed. The addr field was a left-over from an
earlier version of non-blocking connects and can be removed.
The result of the call to kernel_getpeername is not used, the call can be
removed. Non-blocking connects are working, so remove restriction comment.
Signed-off-by: NKarsten Graul <kgraul@linux.vnet.ibm.com>
Signed-off-by: NUrsula Braun <ubraun@linux.vnet.ibm.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

The function smc_netinfo_by_tcpsk() belongs to CLC handling.
Move it to smc_clc.c and rename to smc_clc_netinfo_by_tcpsk.
Signed-off-by: NKarsten Graul <kgraul@linux.vnet.ibm.com>
Signed-off-by: NUrsula Braun <ubraun@linux.vnet.ibm.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Remove structures used internal only from headers.
And remove an extra function parameter.
Signed-off-by: NStefan Raspl <raspl@linux.vnet.ibm.com>
Signed-off-by: NUrsula Braun <ubraun@linux.vnet.ibm.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Yuval Mintz says:

====================
ipmr, ip6mr: Align multicast routing for IPv4 & IPv6

Historically ip6mr was based [cut-n-paste] on ipmr and the two have not
diverged too much. Apparently as ipv4 multicast routing is more common
than its ipv6 brethren modifications since then are mostly one-way,
affecting ipmr while leaving ip6mr unchanged.

This series is meant to re-factor both ipmr and ip6mr into having common
structures [and some functionality], adding 2 new common files -
mroute_base.h and ipmr_base.c.

The series begins by bringing ip6mr up to speed to some of the changes
applied in the past to ipmr [#2, #3].
It is then possible to re-factor a lot of the common structures -
vif devices [#1], mr_table [#4] mfc_cache [#6], and use the common
structures in both ipmr and ip6mr.

The rest of the patches re-factor some choice flows used by both ipmr
and ip6mr and eliminates duplicity.

This series would later allow for easy extension of ipmr offloading
to support ip6mr offloading as well, as almost all structures
related to the offloading would be shared between the two protocols.

Changes from previous versions
------------------------------
v2:
  - #6 Corrected reporting logic when hitting an unresolved cache
  - #7 Addressed kernel doc style [Thanks Nikolay]

RFC -> v1:
  - Corrected support for CONFIG_IP{,V6}_MROUTE_MULTIPLE_TABLES
  - Addressed a couple of kbuild test robot issues
====================
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

The various MFC entries are being held in the same kind of mr_tables
for both ipmr and ip6mr, and their traversal logic is identical.
Also, with the exception of the addresses [and other small tidbits]
the major bulk of the nla setting is identical.

Unite as much of the dumping as possible between the two.
Notice this requires creating an mr_table iterator for each, as the
for-each preprocessor macro can't be used by the common logic.
Signed-off-by: NYuval Mintz <yuvalm@mellanox.com>
Acked-by: NNikolay Aleksandrov <nikolay@cumulusnetworks.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

MFC_NOTIFY exists in ip6mr, probably as some legacy code
[was already removed for ipmr in commit
06bd6c03 ("net: ipmr: remove unused MFC_NOTIFY flag and make the flags enum").
Remove it from ip6mr as well, and move the enum into a common file;
Notice MFC_OFFLOAD is currently only used by ipmr.
Signed-off-by: NYuval Mintz <yuvalm@mellanox.com>
Acked-by: NNikolay Aleksandrov <nikolay@cumulusnetworks.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Same as previously done with the mfc seq, the logic for the vif seq is
refactored to be shared between ipmr and ip6mr.
Signed-off-by: NYuval Mintz <yuvalm@mellanox.com>
Acked-by: NNikolay Aleksandrov <nikolay@cumulusnetworks.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

With the exception of the final dump, ipmr and ip6mr have the exact same
seq logic for traversing a given mr_table. Refactor that code and make
it common.
Signed-off-by: NYuval Mintz <yuvalm@mellanox.com>
Acked-by: NNikolay Aleksandrov <nikolay@cumulusnetworks.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

ipmr and ip6mr utilize the exact same methods for searching the
hashed resolved connections, difference being only in the construction
of the hash comparison key.

In order to unite the flow, introduce an mr_table operation set that
would contain the protocol specific information required for common
flows, in this case - the hash parameters and a comparison key
representing a (*,*) route.
Signed-off-by: NYuval Mintz <yuvalm@mellanox.com>
Acked-by: NNikolay Aleksandrov <nikolay@cumulusnetworks.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

mfc_cache and mfc6_cache are almost identical - the main difference is
in the origin/group addresses and comparison-key. Make a common
structure encapsulating most of the multicast routing logic  - mr_mfc
and convert both ipmr and ip6mr into using it.

For easy conversion [casting, in this case] mr_mfc has to be the first
field inside every multicast routing abstraction utilizing it.
Signed-off-by: NYuval Mintz <yuvalm@mellanox.com>
Acked-by: NNikolay Aleksandrov <nikolay@cumulusnetworks.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Now that both ipmr and ip6mr are using the same mr_table structure,
we can have a common function to allocate & initialize a new instance.
Signed-off-by: NYuval Mintz <yuvalm@mellanox.com>
Acked-by: NNikolay Aleksandrov <nikolay@cumulusnetworks.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Following previous changes to ip6mr, mr_table and mr6_table are
basically the same [up to mr6_table having additional '6' suffixes to
its variable names].
Move the common structure definition into a common header; This
requires renaming all references in ip6mr to variables that had the
distinct suffix.
Signed-off-by: NYuval Mintz <yuvalm@mellanox.com>
Acked-by: NNikolay Aleksandrov <nikolay@cumulusnetworks.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>