snd_nxt must be updated synchronously with sk_send_head.  Otherwise
tp->packets_out may be updated incorrectly, what may bring a kernel panic.

Here is a kernel panic from my host.
[  103.043194] BUG: unable to handle kernel NULL pointer dereference at 0000000000000048
[  103.044025] IP: [<ffffffff815aaaaf>] tcp_rearm_rto+0xcf/0x150
...
[  146.301158] Call Trace:
[  146.301158]  [<ffffffff815ab7f0>] tcp_ack+0xcc0/0x12c0

Before this panic a tcp socket was restored. This socket had sent and
unsent data in the write queue. Sent data was restored in repair mode,
then the socket was switched from reapair mode and unsent data was
restored. After that the socket was switched back into repair mode.

In that moment we had a socket where write queue looks like this:
snd_una    snd_nxt   write_seq
   |_________|________|
             |
	  sk_send_head

After a second switching from repair mode the state of socket was
changed:

snd_una          snd_nxt, write_seq
   |_________ ________|
             |
	  sk_send_head

This state is inconsistent, because snd_nxt and sk_send_head are not
synchronized.

Bellow you can find a call trace, how packets_out can be incremented
twice for one skb, if snd_nxt and sk_send_head are not synchronized.
In this case packets_out will be always positive, even when
sk_write_queue is empty.

tcp_write_wakeup
	skb = tcp_send_head(sk);
	tcp_fragment
		if (!before(tp->snd_nxt, TCP_SKB_CB(buff)->end_seq))
			tcp_adjust_pcount(sk, skb, diff);
	tcp_event_new_data_sent
		tp->packets_out += tcp_skb_pcount(skb);

I think update of snd_nxt isn't required, when a socket is switched from
repair mode.  Because it's initialized in tcp_connect_init. Then when a
write queue is restored, snd_nxt is incremented in tcp_event_new_data_sent,
so it's always is in consistent state.

I have checked, that the bug is not reproduced with this patch and
all tests about restoring tcp connections work fine.

Cc: Pavel Emelyanov <xemul@parallels.com>
Cc: Eric Dumazet <edumazet@google.com>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
Cc: James Morris <jmorris@namei.org>
Cc: Hideaki YOSHIFUJI <yoshfuji@linux-ipv6.org>
Cc: Patrick McHardy <kaber@trash.net>
Signed-off-by: NAndrey Vagin <avagin@openvz.org>
Acked-by: NPavel Emelyanov <xemul@parallels.com>
Acked-by: NEric Dumazet <edumazet@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

After searching rt by the vti tunnel dst/src parameter,
if this rt has neither attached to any transformation
nor the transformation is not tunnel oriented, this rt
should be released back to ip layer.

otherwise causing dst memory leakage.
Signed-off-by: NFan Du <fan.du@windriver.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

The parameter is just 'group', not 'groups', fix the documentation typo.
Signed-off-by: NJohannes Berg <johannes.berg@intel.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

A plain read() on a socket does set msg->msg_name to NULL. So check for
NULL pointer first.
Signed-off-by: NHannes Frederic Sowa <hannes@stressinduktion.org>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Commit 6d0bfe22
	net: ipv6: Add IPv6 support to the ping socket

introduced a change in the cleanup logic of inet6_init and
has a bug in that ipv6_packet_cleanup() may not be called.
Fix the cleanup ordering.

CC: Hannes Frederic Sowa <hannes@stressinduktion.org>
CC: Lorenzo Colitti <lorenzo@google.com>
CC: Fabio Estevam <fabio.estevam@freescale.com>
Signed-off-by: NVlad Yasevich <vyasevich@gmail.com>
Acked-by: NHannes Frederic Sowa <hannes@stressinduktion.org>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Sparse pointed out that the new flags variable I had added
shadowed an existing one, rename the new one to avoid that,
making the code clearer.
Signed-off-by: NJohannes Berg <johannes.berg@intel.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Only update *addr_len when we actually fill in sockaddr, otherwise we
can return uninitialized memory from the stack to the caller in the
recvfrom, recvmmsg and recvmsg syscalls. Drop the the (addr_len == NULL)
checks because we only get called with a valid addr_len pointer either
from sock_common_recvmsg or inet_recvmsg.

If a blocking read waits on a socket which is concurrently shut down we
now return zero and set msg_msgnamelen to 0.
Reported-by: Nmpb <mpb.mail@gmail.com>
Suggested-by: NEric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: NHannes Frederic Sowa <hannes@stressinduktion.org>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

When CONFIG_SYSCTL=n the following build warning happens:

net/ipv6/ndisc.c:1730:1: warning: label 'out' defined but not used [-Wunused-label]

The 'out' label is only used when CONFIG_SYSCTL=y, so move it inside the
'ifdef CONFIG_SYSCTL' block.
Signed-off-by: NFabio Estevam <fabio.estevam@freescale.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

For performance reasons, sch_fq tried hard to not setup timers for every
sent packet, using a quantum based heuristic : A delay is setup only if
the flow exhausted its credit.

Problem is that application limited flows can refill their credit
for every queued packet, and they can evade pacing.

This problem can also be triggered when TCP flows use small MSS values,
as TSO auto sizing builds packets that are smaller than the default fq
quantum (3028 bytes)

This patch adds a 40 ms delay to guard flow credit refill.

Fixes: afe4fd06 ("pkt_sched: fq: Fair Queue packet scheduler")
Signed-off-by: NEric Dumazet <edumazet@google.com>
Cc: Maciej Żenczykowski <maze@google.com>
Cc: Willem de Bruijn <willemb@google.com>
Cc: Yuchung Cheng <ycheng@google.com>
Cc: Neal Cardwell <ncardwell@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Commit 7eec4174 ("pkt_sched: fq: fix non TCP flows pacing")
obsoleted TCA_FQ_FLOW_DEFAULT_RATE without notice for the users.

Suggested by David Miller
Signed-off-by: NEric Dumazet <edumazet@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Now that the ops assignment is just two variables rather than a
long list iteration etc., there's no reason to separately export
__genl_register_family() and __genl_register_family_with_ops().

Unify the two functions into __genl_register_family() and make
genl_register_family_with_ops() call it after assigning the ops.
Signed-off-by: NJohannes Berg <johannes.berg@intel.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

A macvlan device has always LRO disabled so that calling
dev_disable_lro() on it does nothing. If we need to disable LRO
e.g. because

  - the macvlan device is inserted into a bridge
  - IPv6 forwarding is enabled for it
  - it is in a different namespace than lowerdev and IPv4
    forwarding is enabled in it

we need to disable LRO on its underlying device instead (as we
do for 802.1q VLAN devices).

v2: use newly introduced netif_is_macvlan()
Signed-off-by: NMichal Kubecek <mkubecek@suse.cz>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

If priority/traffic class field in IPv6 header is set (seen when
using ssh), the uncompression sets the TC and Flow fields incorrectly.

Example:

This is IPv6 header of a sent packet. Note the priority/TC (=1) in
the first byte.

00000000: 61 00 00 00 00 2c 06 40 fe 80 00 00 00 00 00 00
00000010: 02 02 72 ff fe c6 42 10 fe 80 00 00 00 00 00 00
00000020: 02 1e ab ff fe 4c 52 57

This gets compressed like this in the sending side

00000000: 72 31 04 06 02 1e ab ff fe 4c 52 57 ec c2 00 16
00000010: aa 2d fe 92 86 4e be c6 ....

In the receiving end, the packet gets uncompressed to this
IPv6 header

00000000: 60 06 06 02 00 2a 1e 40 fe 80 00 00 00 00 00 00
00000010: 02 02 72 ff fe c6 42 10 fe 80 00 00 00 00 00 00
00000020: ab ff fe 4c 52 57 ec c2

First four bytes are set incorrectly and we have also lost
two bytes from destination address.

The fix is to switch the case values in switch statement
when checking the TC field.
Signed-off-by: NJukka Rissanen <jukka.rissanen@linux.intel.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This fixes the following Smatch warning:
net/tipc/link.c:2364 tipc_link_recv_fragment()
    warn: variable dereferenced before check '*head' (see line 2361)

A null pointer might be passed to skb_try_coalesce if
a malicious sender injects orphan fragments on a link.
Signed-off-by: NErik Hugne <erik.hugne@ericsson.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

ip4_datagram_connect() being called from process context,
it should use IP_INC_STATS() instead of IP_INC_STATS_BH()
otherwise we can deadlock on 32bit arches, or get corruptions of
SNMP counters.

Fixes: 584bdf8c ("[IPV4]: Fix "ipOutNoRoutes" counter error for TCP and UDP")
Signed-off-by: NEric Dumazet <edumazet@google.com>
Reported-by: NDave Jones <davej@redhat.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

If 'hsr_get_node_data()' returns error, going directly to 'fail' label
doesn't free the memory pointed by 'skb_out'.
Signed-off-by: NGeyslan G. Bem <geyslan@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Initial sch_fq implementation copied code from pfifo_fast to classify
a packet as a high prio packet.

This clashes with setups using PRIO with say 7 bands, as one of the
band could be incorrectly (mis)classified by FQ.

Packets would be queued in the 'internal' queue, and no pacing ever
happen for this special queue.

Fixes: afe4fd06 ("pkt_sched: fq: Fair Queue packet scheduler")
Signed-off-by: NMaciej Żenczykowski <maze@google.com>
Signed-off-by: NEric Dumazet <edumazet@google.com>
Cc: Stephen Hemminger <stephen@networkplumber.org>
Cc: Willem de Bruijn <willemb@google.com>
Cc: Yuchung Cheng <ycheng@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Now that genl_ops are no longer modified in place when
registering, they can be made const. This patch was done
mostly with spatch:

@@
identifier ops;
@@
+const
 struct genl_ops ops[] = {
 ...
 };

(except the struct thing in net/openvswitch/datapath.c)
Signed-off-by: NJohannes Berg <johannes.berg@intel.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Allow making the ops array const by not modifying the ops
flags on registration but rather only when ops are sent
out in the family information.

No users are updated yet except for the pre_doit/post_doit
calls in wireless (the only ones that exist now.)
Signed-off-by: NJohannes Berg <johannes.berg@intel.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Instead of using a linked list, use an array. This reduces
the data size needed by the users of genetlink, for example
in wireless (net/wireless/nl80211.c) on 64-bit it frees up
over 1K of data space.

Remove the attempted sending of CTRL_CMD_NEWOPS ctrl event
since genl_ctrl_event(CTRL_CMD_NEWOPS, ...) only returns
-EINVAL anyway, therefore no such event could ever be sent.
Signed-off-by: NJohannes Berg <johannes.berg@intel.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

genl_register_ops() is still needed for internal registration,
but is no longer available to users of the API.
Signed-off-by: NJohannes Berg <johannes.berg@intel.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This simplifies the code since there's no longer a need to
have error handling in the registration.

Unfortunately it means more extern function declarations are
needed, but the overall goal would seem to justify this.

Due to the removal of duplication in the netlink policies,
this reduces the size of wimax by almost 1k.
Signed-off-by: NJohannes Berg <johannes.berg@intel.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This simplifies the code since there's no longer a need to
have error handling in the registration.

Unfortunately it means more extern function declarations are
needed, but the overall goal would seem to justify this.

While at it, also fix the registration error path - if the
family registration failed then it shouldn't be unregistered.
Signed-off-by: NJohannes Berg <johannes.berg@intel.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This simplifies the code since there's no longer a
need to have error handling in the registration.
Signed-off-by: NJohannes Berg <johannes.berg@intel.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Bug has been introduced by commit bb814094 ("ip6tnl: allow to use rtnl ops
on fb tunnel").

When ip6_tunnel.ko is unloaded, FB device is delete by rtnl_link_unregister()
and then we try to use the pointer in ip6_tnl_destroy_tunnels().

Let's add an handler for dellink, which will never remove the FB tunnel. With
this patch it will no more be possible to remove it via 'ip link del ip6tnl0',
but it's safer.

The same fix was already proposed by Willem de Bruijn <willemb@google.com> for
sit interfaces.

CC: Willem de Bruijn <willemb@google.com>
Reported-by: NSteven Rostedt <rostedt@goodmis.org>
Signed-off-by: NNicolas Dichtel <nicolas.dichtel@6wind.com>
Acked-by: NWillem de Bruijn <willemb@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

addrconf_add_linklocal() already adds the link local route, so there is no
reason to add it before calling this function.
Signed-off-by: NNicolas Dichtel <nicolas.dichtel@6wind.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

When a link local address was added to a sit interface, the corresponding route
was not configured. This breaks routing protocols that use the link local
address, like OSPFv3.

To ease the code reading, I remove sit_route_add(), which only adds v4 mapped
routes, and add this kind of route directly in sit_add_v4_addrs(). Thus link
local and v4 mapped routes are configured in the same place.
Reported-by: NLi Hongjun <hongjun.li@6wind.com>
Signed-off-by: NNicolas Dichtel <nicolas.dichtel@6wind.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

When the local IPv4 endpoint is wilcard (0.0.0.0), the prefix length is
correctly set, ie 64 if the address is a link local one or 96 if the address is
a v4 mapped one.
But when the local endpoint is specified, the prefix length is set to 128 for
both kind of address. This patch fix this wrong prefix length.
Signed-off-by: NNicolas Dichtel <nicolas.dichtel@6wind.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Bug: The fallback device is created in sit_init_net and assumed to be
freed in sit_exit_net. First, it is dereferenced in that function, in
sit_destroy_tunnels:

        struct net *net = dev_net(sitn->fb_tunnel_dev);

Prior to this, rtnl_unlink_register has removed all devices that match
rtnl_link_ops == sit_link_ops.

Commit 205983c4 added the line

+       sitn->fb_tunnel_dev->rtnl_link_ops = &sit_link_ops;

which cases the fallback device to match here and be freed before it
is last dereferenced.

Fix: This commit adds an explicit .delllink callback to sit_link_ops
that skips deallocation at rtnl_unlink_register for the fallback
device. This mechanism is comparable to the one in ip_tunnel.

It also modifies sit_destroy_tunnels and its only caller sit_exit_net
to avoid the offending dereference in the first place. That double
lookup is more complicated than required.

Test: The bug is only triggered when CONFIG_NET_NS is enabled. It
causes a GPF only when CONFIG_DEBUG_SLAB is enabled. Verified that
this bug exists at the mentioned commit, at davem-net HEAD and at
3.11.y HEAD. Verified that it went away after applying this patch.

Fixes: 205983c4 ("sit: allow to use rtnl ops on fb tunnel")
Signed-off-by: NWillem de Bruijn <willemb@google.com>
Acked-by: NNicolas Dichtel <nicolas.dichtel@6wind.com>
Acked-by: NEric Dumazet <edumazet@google.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

When a transport recovers due to the new coming sack, SCTP should
iterate all of its transport_list to locate the __two__ most recently used
transport and set to active_path and retran_path respectively. The exising
code does not find the two properly - In case of the following list:

[most-recent] -> [2nd-most-recent] -> ...

Both active_path and retran_path would be set to the 1st element.

The bug happens when:
1) multi-homing
2) failure/partial_failure transport recovers
Both active_path and retran_path would be set to the same most-recent one, in
other words, retran_path would not take its role - an end user might not even
notice this issue.
Signed-off-by: NChang Xiangzhong <changxiangzhong@gmail.com>
Acked-by: NVlad Yasevich <vyasevich@gmail.com>
Acked-by: NNeil Horman <nhorman@tuxdriver.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

We had some reports of crashes using TCP fastopen, and Dave Jones
gave a nice stack trace pointing to the error.

Issue is that tcp_get_metrics() should not be called with a NULL dst

Fixes: 1fe4c481 ("net-tcp: Fast Open client - cookie cache")
Signed-off-by: NEric Dumazet <edumazet@google.com>
Reported-by: NDave Jones <davej@redhat.com>
Cc: Yuchung Cheng <ycheng@google.com>
Acked-by: NYuchung Cheng <ycheng@google.com>
Tested-by: NDave Jones <davej@fedoraproject.org>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

After commit c9eeec26 ("tcp: TSQ can use a dynamic limit"), several
users reported throughput regressions, notably on mvneta and wifi
adapters.

802.11 AMPDU requires a fair amount of queueing to be effective.

This patch partially reverts the change done in tcp_write_xmit()
so that the minimal amount is sysctl_tcp_limit_output_bytes.

It also remove the use of this sysctl while building skb stored
in write queue, as TSO autosizing does the right thing anyway.

Users with well behaving NICS and correct qdisc (like sch_fq),
can then lower the default sysctl_tcp_limit_output_bytes value from
128KB to 8KB.

This new usage of sysctl_tcp_limit_output_bytes permits each driver
authors to check how their driver performs when/if the value is set
to a minimum of 4KB.

Normally, line rate for a single TCP flow should be possible,
but some drivers rely on timers to perform TX completion and
too long TX completion delays prevent reaching full throughput.

Fixes: c9eeec26 ("tcp: TSQ can use a dynamic limit")
Signed-off-by: NEric Dumazet <edumazet@google.com>
Reported-by: NSujith Manoharan <sujith@msujith.org>
Reported-by: NArnaud Ebalard <arno@natisbad.org>
Tested-by: NSujith Manoharan <sujith@msujith.org>
Cc: Felix Fietkau <nbd@openwrt.org>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

We currently don't call br_vlan_flush() when deleting a bridge, which
leads to memory leak if br->vlan_info is allocated.

Steps to reproduce:
  while :
  do
    brctl addbr br0
    bridge vlan add dev br0 vid 10 self
    brctl delbr br0
  done
We can observe the cache size of corresponding slab entry
(as kmalloc-2048 in SLUB) is increased.

kmemleak output:
unreferenced object 0xffff8800b68a7000 (size 2048):
  comm "bridge", pid 2086, jiffies 4295774704 (age 47.656s)
  hex dump (first 32 bytes):
    00 00 00 00 00 00 00 00 00 48 9b 36 00 88 ff ff  .........H.6....
    00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00  ................
  backtrace:
    [<ffffffff815eb6ae>] kmemleak_alloc+0x4e/0xb0
    [<ffffffff8116a1ca>] kmem_cache_alloc_trace+0xca/0x220
    [<ffffffffa03eddd6>] br_vlan_add+0x66/0xe0 [bridge]
    [<ffffffffa03e543c>] br_setlink+0x2dc/0x340 [bridge]
    [<ffffffff8150e481>] rtnl_bridge_setlink+0x101/0x200
    [<ffffffff8150d9d9>] rtnetlink_rcv_msg+0x99/0x260
    [<ffffffff81528679>] netlink_rcv_skb+0xa9/0xc0
    [<ffffffff8150d938>] rtnetlink_rcv+0x28/0x30
    [<ffffffff81527ccd>] netlink_unicast+0xdd/0x190
    [<ffffffff8152807f>] netlink_sendmsg+0x2ff/0x740
    [<ffffffff814e8368>] sock_sendmsg+0x88/0xc0
    [<ffffffff814e8ac8>] ___sys_sendmsg.part.14+0x298/0x2b0
    [<ffffffff814e91de>] __sys_sendmsg+0x4e/0x90
    [<ffffffff814e922e>] SyS_sendmsg+0xe/0x10
    [<ffffffff81601669>] system_call_fastpath+0x16/0x1b
    [<ffffffffffffffff>] 0xffffffffffffffff
Signed-off-by: NToshiaki Makita <makita.toshiaki@lab.ntt.co.jp>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

We should call vlan_vid_del for all vids at nbp_vlan_flush to prevent
vid_info->refcount from being leaked when detaching a bridge port.
Signed-off-by: NToshiaki Makita <makita.toshiaki@lab.ntt.co.jp>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

We should use wrapper functions vlan_vid_[add/del] instead of
ndo_vlan_rx_[add/kill]_vid. Otherwise, we might be not able to communicate
using vlan interface in a certain situation.

Example of problematic case:
  vconfig add eth0 10
  brctl addif br0 eth0
  bridge vlan add dev eth0 vid 10
  bridge vlan del dev eth0 vid 10
  brctl delif br0 eth0
In this case, we cannot communicate via eth0.10 because vlan 10 is
filtered by NIC that has the vlan filtering feature.
Signed-off-by: NToshiaki Makita <makita.toshiaki@lab.ntt.co.jp>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

commit 06a23fe3
("core/dev: set pkt_type after eth_type_trans() in dev_forward_skb()")
and refactoring 64261f23
("dev: move skb_scrub_packet() after eth_type_trans()")

are forcing pkt_type to be PACKET_HOST when skb traverses veth.

which means that ip forwarding will kick in inside netns
even if skb->eth->h_dest != dev->dev_addr

Fix order of eth_type_trans() and skb_scrub_packet() in dev_forward_skb()
and in ip_tunnel_rcv()

Fixes: 06a23fe3 ("core/dev: set pkt_type after eth_type_trans() in dev_forward_skb()")
CC: Isaku Yamahata <yamahatanetdev@gmail.com>
CC: Maciej Zenczykowski <zenczykowski@gmail.com>
CC: Nicolas Dichtel <nicolas.dichtel@6wind.com>
Signed-off-by: NAlexei Starovoitov <ast@plumgrid.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Fixes a suspicious rcu derference warning.

Cc: Florent Fourcot <florent.fourcot@enst-bretagne.fr>
Signed-off-by: NHannes Frederic Sowa <hannes@stressinduktion.org>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This is to avoid very silly Kconfig dependencies for modules
using this routine.
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Pushing original fragments through causes several problems. For example
for matching, frags may not be matched correctly. Take following
example:

<example>
On HOSTA do:
ip6tables -I INPUT -p icmpv6 -j DROP
ip6tables -I INPUT -p icmpv6 -m icmp6 --icmpv6-type 128 -j ACCEPT

and on HOSTB you do:
ping6 HOSTA -s2000    (MTU is 1500)

Incoming echo requests will be filtered out on HOSTA. This issue does
not occur with smaller packets than MTU (where fragmentation does not happen)
</example>

As was discussed previously, the only correct solution seems to be to use
reassembled skb instead of separete frags. Doing this has positive side
effects in reducing sk_buff by one pointer (nfct_reasm) and also the reams
dances in ipvs and conntrack can be removed.

Future plan is to remove net/ipv6/netfilter/nf_conntrack_reasm.c
entirely and use code in net/ipv6/reassembly.c instead.
Signed-off-by: NJiri Pirko <jiri@resnulli.us>
Acked-by: NJulian Anastasov <ja@ssi.bg>
Signed-off-by: NMarcelo Ricardo Leitner <mleitner@redhat.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

If reassembled packet would fit into outdev MTU, it is not fragmented
according the original frag size and it is send as single big packet.

The second case is if skb is gso. In that case fragmentation does not happen
according to the original frag size.

This patch fixes these.
Signed-off-by: NJiri Pirko <jiri@resnulli.us>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>