This patch adds support for the Cavium ThunderX network controller.
The driver is on the pci bus and thus requires the Thunder PCIe host
controller driver to be enabled.
Signed-off-by: Maciej Czekaj <mjc@semihalf.com>
Signed-off-by: David Daney <david.daney@cavium.com>
Signed-off-by: Sunil Goutham <sgoutham@cavium.com>
Signed-off-by: Ganapatrao Kulkarni <ganapatrao.kulkarni@caviumnetworks.com>
Signed-off-by: Aleksey Makarov <aleksey.makarov@caviumnetworks.com>
Signed-off-by: Tomasz Nowicki <tomasz.nowicki@linaro.org>
Signed-off-by: Robert Richter <rrichter@cavium.com>
Signed-off-by: Kamil Rytarowski <kamil@semihalf.com>
Signed-off-by: Thanneeru Srinivasulu <tsrinivasulu@caviumnetworks.com>
Signed-off-by: Sruthi Vangala <svangala@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

This vendor id will be used by network (vNIC), USB (xHCI),
SATA (AHCI), GPIO, I2C, MMC and maybe other drivers
for ThunderX SoC.
Acked-by: Bjorn Helgaas <bhelgaas@google.com>
Signed-off-by: Sunil Goutham <sgoutham@cavium.com>
Signed-off-by: Aleksey Makarov <aleksey.makarov@caviumnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

While 3b529602 ("test_bpf: add more eBPF jump torture cases")
added the int3 bug test case only for eBPF, which needs exactly 11
passes to converge, here's a version for classic BPF with 11 passes,
and one that would need 70 passes on x86_64 to actually converge for
being successfully JITed. Effectively, all jumps are being optimized
out resulting in a JIT image of just 89 bytes (from originally max
BPF insns), only returning K.

Might be useful as a receipe for folks wanting to craft a test case
when backporting the fix in commit 3f7352bf ("x86: bpf_jit: fix
compilation of large bpf programs") while not having eBPF. The 2nd
one is delegated to the interpreter as the last pass still results
in shrinking, in other words, this one won't be JITed on x86_64.
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@plumgrid.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Edward Cree says:

====================
sfc: add MCDI tracing

This patchset adds support for logging MCDI (Management-Controller-to-
 Driver Interface) interactions between the sfc driver and a bound device,
 to aid in debugging.
Solarflare has a tool to decode the resulting traces and will look to
 open-source this if there is any external interest, but the protocol is
 already detailed in drivers/net/ethernet/sfc/mcdi_pcol.h.
The logging buffer we allocate per MCDI context is a work area for
 constructing each individual message before logging it with netif_info.
The reason the buffer is long-lived is simply to avoid the overhead of
 allocating and freeing it every MCDI call, since MCDIs are already known
 to be serialised for other reasons.

--
v4: remove patch #4, which has already been applied via sshah
v3: add some explanations to cover letter and patch #4
v2: avoid long lines in cover letter; fix multiline comment style
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

As many issues are encountered at probe time, where MCDI logging can't be
 enabled through the sysfs node, this change adds a module parameter
 'mcdi_logging_default', which defaults to false.  When set to true, newly-
 probed functions will have MCDI logging enabled.  The setting can
 subsequently be changed as normal through the sysfs node.
Signed-off-by: Edward Cree <ecree@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

MCDI tracing is enabled per-function with a sysfs file
    /sys/class/net/<NET_DEV>/device/mcdi_logging
Signed-off-by: Edward Cree <ecree@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

MCDI tracing is conditional on CONFIG_SFC_MCDI_LOGGING, which is enabled
 by default.

Each MCDI command will produce a console line like
    sfc dom:bus:dev:fn ifname: MCDI RPC REQ: xxxxxxxx [yyyyyyyy...]
where xxxxxxxx etc. are the raw MCDI payload in 32-bit hex chunks.
The response will then produce a similar line with "RESP" instead of "REQ",
 and containing the MCDI response payload (if any).
Signed-off-by: Edward Cree <ecree@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

We're seeing some softlockups from this function when there
are a lot fdb entries on a vxlan device. Taking the lock for
each bucket instead of the whole table is enough to fix that.
Signed-off-by: Sorin Dumitru <sdumitru@ixiacom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

A long standing problem on busy servers is the tiny available TCP port
range (/proc/sys/net/ipv4/ip_local_port_range) and the default
sequential allocation of source ports in connect() system call.

If a host is having a lot of active TCP sessions, chances are
very high that all ports are in use by at least one flow,
and subsequent bind(0) attempts fail, or have to scan a big portion of
space to find a slot.

In this patch, I changed the starting point in __inet_hash_connect()
so that we try to favor even [1] ports, leaving odd ports for bind()
users.

We still perform a sequential search, so there is no guarantee, but
if connect() targets are very different, end result is we leave
more ports available to bind(), and we spread them all over the range,
lowering time for both connect() and bind() to find a slot.

This strategy only works well if /proc/sys/net/ipv4/ip_local_port_range
is even, ie if start/end values have different parity.

Therefore, default /proc/sys/net/ipv4/ip_local_port_range was changed to
32768 - 60999 (instead of 32768 - 61000)

There is no change on security aspects here, only some poor hashing
schemes could be eventually impacted by this change.

[1] : The odd/even property depends on ip_local_port_range values parity
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Florian Westphal says:

====================
net: force refragmentation for DF reassembed skbs

output path tests:

    if (skb->len > mtu) ip_fragment()

This breaks connectivity in one corner case:
 If the skb was reassembled, but has the DF bit set and ..
 .. its reassembled size is <= outdev mtu ..
 .. we will forward a DF packet larger than what the sender
    transmitted on wire.

If a router later in the path can't forward this packet, it will send an
icmp error in response to an mtu that the original sender never exceeded.

This changes ipv4 defrag/output path to

a) force refragmentation for DF reassembled skbs and
b) set DF bit on all fragments when refragmenting if it was set on original
frags.

tested via:
from scapy.all import *
dip="10.23.42.2"
payload="A"*1400
packet=IP(dst=dip,id=12345,flags='DF')/UDP(sport=42,dport=42)/payload
frags=fragment(packet,fragsize=1200)
for fragment in frags:
    send(fragment)

Without this patch, we generate fragments without df bit set based
on the outgoing device mtu when fragmenting after forwarding, ie.

IP (ttl 64, id 12345, offset 0, flags [+, DF], proto UDP (17), length 1204)
    192.168.7.1.42 > 10.23.42.2.42: UDP, length 1400
IP (ttl 64, id 12345, offset 1184, flags [DF], proto UDP (17), length 244)
    192.168.7.1 > 10.23.42.2: ip-proto-17

on ingress will either turn into

IP (ttl 63, id 12345, offset 0, flags [+], proto UDP (17), length 1396)
    192.168.7.1.42 > 10.23.42.2.42: UDP, length 1400
IP (ttl 63, id 12345, offset 1376, flags [none], proto UDP (17), length 52)

(mtu 1400: We strip df and send larger fragment), or

IP (ttl 63, id 12345, offset 0, flags [DF], proto UDP (17), length 1428)
    192.168.7.1.42 > 10.23.42.2.42: [udp sum ok] UDP, length 1400

if mtu is 1500.  And in this case things break; router with a smaller mtu
will send icmp error, but original sender only sent packets <= 1204 byte.

With patch, we keep intent of such fragments and will emit DF-fragments
that won't exceed 1204 byte in size.

Joint work with Hannes Frederic Sowa.

Changes since v2:
 - split unrelated patches from series
 - rework changelog of patch #2 to better illustrate breakage
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

We currently always send fragments without DF bit set.

Thus, given following setup:

mtu1500 - mtu1500:1400 - mtu1400:1280 - mtu1280
   A           R1              R2         B

Where R1 and R2 run linux with netfilter defragmentation/conntrack
enabled, then if Host A sent a fragmented packet _with_ DF set to B, R1
will respond with icmp too big error if one of these fragments exceeded
1400 bytes.

However, if R1 receives fragment sizes 1200 and 100, it would
forward the reassembled packet without refragmenting, i.e.
R2 will send an icmp error in response to a packet that was never sent,
citing mtu that the original sender never exceeded.

The other minor issue is that a refragmentation on R1 will conceal the
MTU of R2-B since refragmentation does not set DF bit on the fragments.

This modifies ip_fragment so that we track largest fragment size seen
both for DF and non-DF packets, and set frag_max_size to the largest
value.

If the DF fragment size is larger or equal to the non-df one, we will
consider the packet a path mtu probe:
We set DF bit on the reassembled skb and also tag it with a new IPCB flag
to force refragmentation even if skb fits outdev mtu.

We will also set DF bit on each fragment in this case.

Joint work with Hannes Frederic Sowa.
Reported-by: Jesse Gross <jesse@nicira.com>
Signed-off-by: Florian Westphal <fw@strlen.de>
Acked-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

ip_skb_dst_mtu is small inline helper, but its called in several places.

before: 17061      44       0   17105    42d1 net/ipv4/ip_output.o
after:  16805      44       0   16849    41d1 net/ipv4/ip_output.o
Signed-off-by: Florian Westphal <fw@strlen.de>
Acked-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

Florian Fainelli says:

====================
net: phy: phy_interface_is_rgmii helper

As you suggested, here is the helper function to avoid missing some RGMII
interface checks. Had to wait for net to be merged in net-next to avoid
submitting the same patch/commit.

Dan, you might want to rebase your dp83867 submission to use that helper
when you this patchset gets merged into net-next, thanks!
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

Update all open-coded tests for all 4 PHY_INTERFACE_MODE_RGMII* values
to use the newly introduced helper: phy_interface_is_rgmii.
Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

RGMII interfaces come in 4 different flavors that the PHY library needs
to care about: regular RGMII (no delays), RGMII with either RX or TX
delay, and both. In order to avoid errors of checking only for one type
of RGMII interface and miss the 3 others, introduce a convenience
function which tests for all values.
Suggested-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

We used to get this indirectly I supposed, but no longer do.

Either way, an explicit include should have been done in the
first place.

   net/ipv4/fib_trie.c: In function '__node_free_rcu':
>> net/ipv4/fib_trie.c:293:3: error: implicit declaration of function 'vfree' [-Werror=implicit-function-declaration]
      vfree(n);
      ^
   net/ipv4/fib_trie.c: In function 'tnode_alloc':
>> net/ipv4/fib_trie.c:312:3: error: implicit declaration of function 'vzalloc' [-Werror=implicit-function-declaration]
      return vzalloc(size);
      ^
>> net/ipv4/fib_trie.c:312:3: warning: return makes pointer from integer without a cast
   cc1: some warnings being treated as errors
Reported-by: kbuild test robot <fengguang.wu@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

By making sure sk->sk_gso_max_segs minimal value is one,
and sysctl_tcp_min_tso_segs minimal value is one as well,
tcp_tso_autosize() will return a non zero value.

We can then revert 843925f3
("tcp: Do not apply TSO segment limit to non-TSO packets")
and save few cpu cycles in fast path.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Neal Cardwell <ncardwell@google.com>
Cc: Herbert Xu <herbert@gondor.apana.org.au>
Acked-by: Neal Cardwell <ncardwell@google.com>
Acked-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>

If tcp ehash table is constrained to a very small number of buckets
(eg boot parameter thash_entries=128), then we can crash if spinlock
array has more entries.

While we are at it, un-inline inet_ehash_locks_alloc() and make
following changes :

- Budget 2 cache lines per cpu worth of 'spinlocks'
- Try to kmalloc() the array to avoid extra TLB pressure.
  (Most servers at Google allocate 8192 bytes for this hash table)
- Get rid of various #ifdef
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

In commit dd3f9e70
("tipc: add packet sequence number at instant of transmission") we
made a change with the consequence that packets in the link backlog
queue don't contain valid sequence numbers.

However, when we create a link protocol message, we still use the
sequence number of the first packet in the backlog, if there is any,
as "next_sent" indicator in the message. This may entail unnecessary
retransissions or stale packet transmission when there is very low
traffic on the link.

This commit fixes this issue by only using the current value of
tipc_link::snd_nxt as indicator.
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

make C=2 CF=-D__CHECK_ENDIAN__ net/core/utils.o
...
net/core/utils.c:307:72: warning: incorrect type in argument 2 (different base types)
net/core/utils.c:307:72:    expected restricted __wsum [usertype] addend
net/core/utils.c:307:72:    got restricted __be32 [usertype] from
net/core/utils.c:308:34: warning: incorrect type in argument 2 (different base types)
net/core/utils.c:308:34:    expected restricted __wsum [usertype] addend
net/core/utils.c:308:34:    got restricted __be32 [usertype] to
net/core/utils.c:310:70: warning: incorrect type in argument 2 (different base types)
net/core/utils.c:310:70:    expected restricted __wsum [usertype] addend
net/core/utils.c:310:70:    got restricted __be32 [usertype] from
net/core/utils.c:310:77: warning: incorrect type in argument 2 (different base types)
net/core/utils.c:310:77:    expected restricted __wsum [usertype] addend
net/core/utils.c:310:77:    got restricted __be32 [usertype] to
net/core/utils.c:312:72: warning: incorrect type in argument 2 (different base types)
net/core/utils.c:312:72:    expected restricted __wsum [usertype] addend
net/core/utils.c:312:72:    got restricted __be32 [usertype] from
net/core/utils.c:313:35: warning: incorrect type in argument 2 (different base types)
net/core/utils.c:313:35:    expected restricted __wsum [usertype] addend
net/core/utils.c:313:35:    got restricted __be32 [usertype] to

Note we can use csum_replace4() helper

Fixes: 58e3cac5 ("net: optimise inet_proto_csum_replace4()")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

make C=2 CF=-D__CHECK_ENDIAN__ net/core/secure_seq.o
net/core/secure_seq.c:157:50: warning: restricted __be32 degrades to
integer
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Check in fdb_add_entry() if the source port should learn, similar
check is used in br_fdb_update.
Note that new fdb entries which are added manually or
as local ones are still permitted.
This patch has been tested by running traffic via a bridge port and
switching the port's state, also by manually adding/removing entries
from the bridge's fdb.
Signed-off-by: Wilson Kok <wkok@cumulusnetworks.com>
Signed-off-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net/core/pktgen.c:2672:43: warning: incorrect type in assignment (different base types)
net/core/pktgen.c:2672:43:    expected unsigned short [unsigned] [short] [usertype] <noident>
net/core/pktgen.c:2672:43:    got restricted __be16 [usertype] protocol

Let's use proper struct ethhdr instead of hard coding everything.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

ipv6_select_ident() returns a 32bit value in network order.

Fixes: 286c2349 ("ipv6: Clean up ipv6_select_ident() and ip6_fragment()")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: kbuild test robot <fengguang.wu@intel.com>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Richard Cochran says:

====================
cpsw cleanups

While working on an out-of-tree customization, I noticed a few minor
problems in the cpsw code.  This series cleans up the issues I found.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

The function, cpsw_intr_disable, already calls cpdma_ctlr_int_ctrl.  There
is no need to disable the dma interrupts twice.  This patch removes the
extra calls.
Signed-off-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

The function, cpsw_intr_enable, already calls cpdma_ctlr_int_ctrl.  There
is no need to enable the dma interrupts twice.  This patch removes the
extra call.
Signed-off-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

The funtions, cpsw_ale_flush and cpsw_ale_set_ageout, have never been used
since they were first introduced.  This patch removes the dead code.
Signed-off-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Having the breaks too far to the left makes parsing the dense switch/case
block unnecessarily harder.
Signed-off-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Simon Horman says:

====================
rocker: unused parameter and const cleanups

This series provides some minor though verbose cleanup of rocker.

The second patch depends on the first though it could be rebased.

I had previously asked for v2 to be put on hold while some bugs I had found
in the rocker driver were shaken out. That has now happened and the bugs
turned out to be unrelated.  Accordingly I am reposting the series.

* Changes v2 -> v3
  - Rebase and update for new variables and parameters that may be const

* Changes v1 -> v2
  - Found quite a few more variables and parameters to make const
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

Mark parameters and local variables as const where possible.
Signed-off-by: Simon Horman <simon.horman@netronome.com>
Acked-by: Scott Feldman <sfeldma@gmail.com>
Acked-by: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: David S. Miller <davem@davemloft.net>

Remove unused rocker_port parameter from rocker_port_kfree.
Also remove the rocker_port parameter from callers of rocker_port_kfree
where the parameter it is now unused.
Signed-off-by: Simon Horman <simon.horman@netronome.com>
Acked-by: Scott Feldman <sfeldma@gmail.com>
Acked-by: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: David S. Miller <davem@davemloft.net>

API compliance scanning with coccinelle flagged:
./net/irda/timer.c:63:35-37: use of msecs_to_jiffies probably perferable

Converting milliseconds to jiffies by "val * HZ / 1000" technically
is not a clean solution as it does not handle all corner cases correctly.
By changing the conversion to use msecs_to_jiffies(val) conversion is
correct in all cases. Further the () around the arithmetic expression
was dropped.

Patch was compile tested for x86_64_defconfig + CONFIG_IRDA=m

Patch is against 4.1-rc4 (localversion-next is -next-20150522)
Signed-off-by: Nicholas Mc Guire <hofrat@osadl.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

These two uses seem to have had carriage returns removed.
Make these entries like all the others in this file.
Signed-off-by: Joe Perches <joe@perches.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

API compliance scanning with coccinelle flagged:

Converting milliseconds to jiffies by "val * HZ / 1000" is technically
is not a clean solution as it does not handle all corner cases correctly.
By changing the conversion to use msecs_to_jiffies(val) conversion is
correct in all cases.

in the current code:
  mod_timer(&self->rx_defer_timer, jiffies + (10 * HZ / 1000));
for HZ < 100 (e.g. CONFIG_HZ == 64|32 in alpha) this effectively results
in no delay at all.

Patch was compile tested for x86_64_defconfig (implies CONFIG_USB_IRDA=m)

Patch is against 4.1-rc4 (localversion-next is -next-20150522)
Signed-off-by: Nicholas Mc Guire <hofrat@osadl.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

Network managers like netifd (used in OpenWRT for instance) try to
configure interface options after creation but before setting the
interface up.

Unfortunately the sysfs / bridge currently only allows to configure the
hash_max and multicast_router options when the bridge interface is up.
But since br_multicast_init() doesn't start any timers and only sets
default values and initializes timers it should be save to reconfigure
the default values after that, before things actually get active after
the bridge is set up.
Signed-off-by: Linus Lüssing <linus.luessing@c0d3.blue>
Signed-off-by: David S. Miller <davem@davemloft.net>

since commit 6aafeef0 ("netfilter: push reasm skb through instead of
original frag skbs") we will end up sometimes re-fragmenting skbs
that we've reassembled.

ipv6 defrag preserves the original skbs using the skb frag list, i.e. as long
as the skb frag list is preserved there is no problem since we keep
original geometry of fragments intact.

However, in the rare case where the frag list is munged or skb
is linearized, we might send larger fragments than what we originally
received.

A router in the path might then send packet-too-big errors even if
sender never sent fragments exceeding the reported mtu:

mtu 1500 - 1500:1400 - 1400:1280 - 1280
     A         R1         R2        B

1 - A sends to B, fragment size 1400
2 - R2 sends pkttoobig error for 1280
3 - A sends to B, fragment size 1280
4 - R2 sends pkttoobig error for 1280 again because it sees fragments of size 1400.

make sure ip6_fragment always caps MTU at largest packet size seen
when defragmented skb is forwarded.
Acked-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: David S. Miller <davem@davemloft.net>

The variable irq_coalesce is bool type.
So assign the value true instead of 1.
Signed-off-by: Shailendra Verma <shailendra.capricorn@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

The variable separate_tx_rx_irq is bool type so assigning true
instead of 1.
Signed-off-by: Shailendra Verma <shailendra.capricorn@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Martin KaFai Lau says:

====================
ipv6: Only create RTF_CACHE route after encountering pmtu exception

v4 -> v5:
- Patch 1 is new. Clean up the ipv6_select_ident() and ip6_fragment().

- Further simplify the newly added rt6_get_pcpu_route().  If there is a
  'prev' after cmpxchg, return prev instead of the newly created percpu
  clone.

v3 -> v4:
- Patch 8 is new. It keeps track of the DST_NOCACHE routes in a list to handle
  the iface down/unregister event.

- Remove rcu from the newly added rt6i_pcpu variable.  It is not needed
  because it has already been protected by the existing reader/writer lock.

- Thanks to 'Julian Anastasov <ja@ssi.bg>' for testing the FLOWI_FLAG_KNOWN_NH
  patches.

v2 -> v3:
- Patch 5 to 7 are new.  They take care of cases where the daddr in
  skb is not the one used to do the route look-up.  There is also
  related changes to rt6_nexthop() since v2 which is in patch 2/9.
  Thanks to 'Julian Anastasov <ja@ssi.bg>' for pointing it out.

- Fix a few problems in __ip6_rt_update_pmtu(), like setting the expire
  and mtu before inserting to the tree and don't do dst_destroy() after
  tree insertion failure.  Also update the rt6i_pmtu in fib6_add_rt2node().
  Thanks to 'Steffen Klassert <steffen.klassert@secunet.com>' for pointing
  it out.

- Merge ip6_pmtu_rt_cache_alloc() into ip6_rt_cache_alloc().

v1 -> v2:
- Move the /128 route bug fixes to another series (accepted).
- Create a function for checking (rt6i_flags & (RTF_NONEXTHOP | RTF_GATEWAY)).
- Avoid shuffling the skb network_header.  Instead, change the function
  signature to take iph instead of skb.

- Many Thanks to 'Hannes Frederic Sowa <hannes@stressinduktion.org>' on
  reviewing v1 and v2 and giving advice.

--Martin

~~~ start: v1 compose message (with the out-dated parts removed) ~~~

This series is to avoid creating a RTF_CACHE route whenever we are consulting
the fib6 tree with a new destination.  Instead, only create RTF_CACHE route
when we see a pmtu exception.

Out of all ipv6 RTF_CACHE routes that are created, the percentage that has a
different mtu is very small. In one of our end-user facing proxy server,
only 1k out of 80k RTF_CACHE routes have a smaller MTU.  For our DC
traffic, there is no mtu exception.

A large fib6 tree has problems like, 'ip -6 r show' takes a long time.
gc may kick in too often.  Also, when a service has restarted and a lot
of new TCP conn requests come in, it creates pressure on the tree by inserting
a lot of RTF_CACHE in a short time and it currently requires a write lock
to do that.

The first few patches are prep works to remove assumption that the
returned rt is always RTF_CACHE.

The patch 'ipv6: Only create RTF_CACHE routes after encountering pmtu exception'
do the lazy RTF_CACHE route creation.

The following patches added percpu rt to compensate the performance loss after
doing the RTF_CACHE lazy creation.

Here is some numbers of the udpflood test.  The udpflood has been
slightly modified to have a time limit instead of count limit.

A /64 via gateway route is used for the test. Each udpflood uses 10000 dst
addresses.  The dst addresses of different udpflood processes do not overlap
with each other.

1                    16M                          15M
10                   61M                          61M
20                   65M                          62M
40                   88M                          83M

~~~ end: v1 compose message ~~~
====================
Signed-off-by: David S. Miller <davem@davemloft.net>