[XFRM] xfrm_policy_destroy: Rename and relative fixes.

Since __xfrm_policy_destroy is used to destory the resources
allocated by xfrm_policy_alloc. So using the name
__xfrm_policy_destroy is not correspond with xfrm_policy_alloc.
Rename it to xfrm_policy_destroy.

And along with some instances that call xfrm_policy_alloc
but not using xfrm_policy_destroy to destroy the resource,
fix them.

Signed-off-by: WANG Cong <xiyou.wangcong@gmail.com>
Acked-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>

[XFRM] Statistics: Add outbound-dropping error.

o Increment PolError counter when flow_cache_lookup() returns
  errored pointer.

o Increment NoStates counter at larval-drop.

Signed-off-by: Masahide NAKAMURA <nakam@linux-ipv6.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NET]: Remove obsolete comment

It seems that ip_build_xmit is no longer used in here and
ip_append_data is used.

Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>

[CCID3]: Kill some bloat

Without a number of CONFIG.*DEBUG:

net/dccp/ccids/ccid3.c:
  ccid3_hc_tx_update_x          | -170
  ccid3_hc_tx_packet_sent       | -175
  ccid3_hc_tx_packet_recv       | -169
  ccid3_hc_tx_no_feedback_timer | -192
  ccid3_hc_tx_send_packet       | -144
 5 functions changed, 850 bytes removed, diff: -850

net/dccp/ccids/ccid3.c:
  ccid3_update_send_interval | +191
 1 function changed, 191 bytes added, diff: +191

net/dccp/ccids/ccid3.o:
 6 functions changed, 191 bytes added, 850 bytes removed, diff: -659

Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>

[XFRM]: Kill some bloat

net/xfrm/xfrm_state.c:
  xfrm_audit_state_delete          | -589
  xfrm_replay_check                | -542
  xfrm_audit_state_icvfail         | -520
  xfrm_audit_state_add             | -589
  xfrm_audit_state_replay_overflow | -523
  xfrm_audit_state_notfound_simple | -509
  xfrm_audit_state_notfound        | -521
 7 functions changed, 3793 bytes removed, diff: -3793

net/xfrm/xfrm_state.c:
  xfrm_audit_helper_pktinfo | +522
  xfrm_audit_helper_sainfo  | +598
 2 functions changed, 1120 bytes added, diff: +1120

net/xfrm/xfrm_state.o:
 9 functions changed, 1120 bytes added, 3793 bytes removed, diff: -2673

Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>

[IPVS]: Kill some bloat

net/ipv4/ipvs/ip_vs_xmit.c:
  ip_vs_icmp_xmit   | -638
  ip_vs_tunnel_xmit | -674
  ip_vs_nat_xmit    | -716
  ip_vs_dr_xmit     | -682
 4 functions changed, 2710 bytes removed, diff: -2710

net/ipv4/ipvs/ip_vs_xmit.c:
  __ip_vs_get_out_rt | +595
 1 function changed, 595 bytes added, diff: +595

net/ipv4/ipvs/ip_vs_xmit.o:
 5 functions changed, 595 bytes added, 2710 bytes removed, diff: -2115

Without some CONFIG.*DEBUGs:

net/ipv4/ipvs/ip_vs_xmit.o:
 5 functions changed, 383 bytes added, 1513 bytes removed, diff: -1130

Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NETFILTER]: Kill some supper dupper bloatry

/me awards the bloatiest-of-all-net/-.c-code award to
nf_conntrack_netlink.c, congratulations to all the authors :-/!

Hall of (unquestionable) fame (measured per inline, top 10 under
net/):
  -4496 ctnetlink_parse_tuple        netfilter/nf_conntrack_netlink.c
  -2165 ctnetlink_dump_tuples        netfilter/nf_conntrack_netlink.c
  -2115 __ip_vs_get_out_rt           ipv4/ipvs/ip_vs_xmit.c
  -1924 xfrm_audit_helper_pktinfo    xfrm/xfrm_state.c
  -1799 ctnetlink_parse_tuple_proto  netfilter/nf_conntrack_netlink.c
  -1268 ctnetlink_parse_tuple_ip     netfilter/nf_conntrack_netlink.c
  -1093 ctnetlink_exp_dump_expect    netfilter/nf_conntrack_netlink.c
  -1060 void ccid3_update_send_interval  dccp/ccids/ccid3.c
  -983  ctnetlink_dump_tuples_proto  netfilter/nf_conntrack_netlink.c
  -827  ctnetlink_exp_dump_tuple     netfilter/nf_conntrack_netlink.c

  (i386 / gcc (GCC) 4.1.2 20070626 (Red Hat 4.1.2-13) /
   allyesconfig except CONFIG_FORCED_INLINING)

...and I left < 200 byte gains as future work item.

After iterative inline removal, I finally have this:

net/netfilter/nf_conntrack_netlink.c:
  ctnetlink_exp_fill_info   | -1104
  ctnetlink_new_expect      | -1572
  ctnetlink_fill_info       | -1303
  ctnetlink_new_conntrack   | -2230
  ctnetlink_get_expect      | -341
  ctnetlink_del_expect      | -352
  ctnetlink_expect_event    | -1110
  ctnetlink_conntrack_event | -1548
  ctnetlink_del_conntrack   | -729
  ctnetlink_get_conntrack   | -728
 10 functions changed, 11017 bytes removed, diff: -11017

net/netfilter/nf_conntrack_netlink.c:
  ctnetlink_parse_tuple     | +419
  dump_nat_seq_adj          | +183
  ctnetlink_dump_counters   | +166
  ctnetlink_dump_tuples     | +261
  ctnetlink_exp_dump_expect | +633
  ctnetlink_change_status   | +460
 6 functions changed, 2122 bytes added, diff: +2122

net/netfilter/nf_conntrack_netlink.o:
 16 functions changed, 2122 bytes added, 11017 bytes removed, diff: -8895

Without a number of CONFIG.*DEBUGs, I got this:
net/netfilter/nf_conntrack_netlink.o:
 16 functions changed, 2122 bytes added, 11029 bytes removed, diff: -8907

Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NETNS]: Should build with CONFIG_SYSCTL=n

Previous NETNS patches broke CONFIG_SYSCTL=n case

Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

[CONNECTOR]: Cleanup struct cn_callback_entry

- 'cb' is a fake struct member. In a previous patch struct cn_callback
was renamed to cn_callback_id, so 'cb' should have been deleted at that
time.

- 'nls' isn't used and is redundant, we can retrieve this data through
cn_callback_entry.pdev->nls.

- 'seq' and 'group' should be u32, as they are declared to be u32 in
other places.

Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

[CONNECTOR]: Cleanup struct cn_queue_dev

Struct member netlink_groups is never used, and I don't see how it can
be useful.

Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

[CONNECTOR]: clean up {,__}cn_rx_skb()

- __cn_rx_skb() does nothing but calls cn_call_callback(), it doesn't
check skb and msg sizes as the comment suggests, but cn_rx_skb() checks
those sizes.

- In cn_rx_skb() Local variable 'len' is not used. 'len' is probably
intended to be passed to skb_pull(), but here skb_pull() is not needed,
instead skb_free() is called.

Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

[CONNECTOR]: add a missing break in cn_netlink_send()

Each entry in the list has a unique id, so just break out of the
loop if the matched id is found.

Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

[ICMP]: Avoid sparse warnings in net/ipv4/icmp.c

   CHECK   net/ipv4/icmp.c
net/ipv4/icmp.c:249:13: warning: context imbalance in 'icmp_xmit_unlock' -
unexpected unlock
net/ipv4/icmp.c:376:13: warning: context imbalance in 'icmp_reply' - different
lock contexts for basic block
net/ipv4/icmp.c:430:6: warning: context imbalance in 'icmp_send' - different
lock contexts for basic block

Solution is to declare both icmp_xmit_lock() and icmp_xmit_unlock() as inline

Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NET]: prot_inuse cleanups and optimizations

1) Cleanups (all functions are prefixed by sock_prot_inuse)

sock_prot_inc_use(prot) -> sock_prot_inuse_add(prot,-1)
sock_prot_dec_use(prot) -> sock_prot_inuse_add(prot,-1)
sock_prot_inuse()       -> sock_prot_inuse_get()

New functions :

sock_prot_inuse_init() and sock_prot_inuse_free() to abstract pcounter use.

2) if CONFIG_PROC_FS=n, we can zap 'inuse' member from "struct proto",
since nobody wants to read the inuse value.

This saves 1372 bytes on i386/SMP and some cpu cycles.

Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

[LIB] pcounter : unline too big functions

Before pushing pcounter to Linus tree, I would like to make some adjustments.

Goal is to reduce kernel text size, by unlining too big functions.

When a pcounter is bound to a statically defined per_cpu variable,
we define two small helpers functions. (No more folding function
using the fat for_each_possible_cpu(cpu) ... )

static DEFINE_PER_CPU(int, NAME##_pcounter_values);
static void NAME##_pcounter_add(struct pcounter *self, int val)
{
       __get_cpu_var(NAME##_pcounter_values) += val;
}
static int NAME##_pcounter_getval(const struct pcounter *self, int cpu)
{
       return per_cpu(NAME##_pcounter_values, cpu);
}

Fast path is therefore unchanged, while folding/alloc/free is now unlined.

This saves 228 bytes on i386

Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NET]: Avoid divides in net/core/gen_estimator.c

We can void divides (as seen with CONFIG_CC_OPTIMIZE_FOR_SIZE=y on x86)
changing ((HZ<<idx)/4) to ((HZ/4) << idx)

Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

[TCP]: Perform setting of common control fields in one place

In case of segments which are purely for control without any
data (SYN/ACK/FIN/RST), many fields are set to common values
in multiple places.

i386 results:

$ gcc --version
gcc (GCC) 4.1.2 20070626 (Red Hat 4.1.2-13)

$ codiff tcp_output.o.old tcp_output.o.new
net/ipv4/tcp_output.c:
  tcp_xmit_probe_skb    |  -48
  tcp_send_ack          |  -56
  tcp_retransmit_skb    |  -79
  tcp_connect           |  -43
  tcp_send_active_reset |  -35
  tcp_make_synack       |  -42
  tcp_send_fin          |  -48
 7 functions changed, 351 bytes removed

net/ipv4/tcp_output.c:
  tcp_init_nondata_skb |  +90
 1 function changed, 90 bytes added

tcp_output.o.mid:
 8 functions changed, 90 bytes added, 351 bytes removed, diff: -261

Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>

[TCP]: Urgent parameter effect can be simplified.

Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>

[TCP]: cleanup tcp_parse_options deep indented switch

Removed case indentation level & combined some nested ifs, mostly
within 80 lines now. This is a leftover from indent patch, it
just had to be done manually to avoid messing it up completely.

Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>

[IPSEC]: Return EOVERFLOW when output sequence number overflows

Previously we made it an error on the output path if the sequence number
overflowed.  However we did not set the err variable accordingly.  This
patch sets err to -EOVERFLOW in that case.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NET]: Add some acquires/releases sparse annotations.

Add __acquires() and __releases() annotations to suppress some sparse
warnings.

example of warnings :

net/ipv4/udp.c:1555:14: warning: context imbalance in 'udp_seq_start' - wrong
count at exit
net/ipv4/udp.c:1571:13: warning: context imbalance in 'udp_seq_stop' -
unexpected unlock

Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

[IPVS]: Remove declaration of unimplemented method and remove unused definition from include/net/ip_vs.h

In include/net/ip_vs.h:
- The ip_vs_secure_tcp_set() method is not implemented anywhere.
- IP_VS_APP_TYPE_FTP is an unused definition.

Signed-off-by: Rami Rosen <ramirose@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

[IPV4]: Remove three declarations of unimplemented methods and correct a typo in include/net/ip.h

These three declarations in include/net/ip.h are not implemented
anywhere:

ip_mc_dropsocket(), ip_mc_dropdevice() and ip_net_unreachable().

Also, correct a comment to be "Functions provided by ip_fragment.c"
(instead of by ip_fragment.o) in consistency with the other comments
in this header.

Signed-off-by: Rami Rosen <ramirose@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

[SHAPER]: The scheduled shaper removal.

This patch contains the scheduled removal of the shaper driver.

Signed-off-by: Adrian Bunk <bunk@kernel.org>
Acked-by: Alan Cox <alan@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

[IPSEC]: Kill duplicate xfrm_policy_flush prototype

For five years we had two xfrm_policy_flush prototypes and every time that
function's signature changed people have been diligently updating both of
them without noticing :)

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>

[PATCH] use SK_MEM_QUANTUM_SHIFT in __sk_mem_reclaim()

Avoid an expensive divide (as done in commit
18030477e70a826b91608aee40a987bbd368fec6 but lost in commit
23821d2653111d20e75472c8c5003df1a55309a8)

Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

[TCP]: Remove unnecessary local variable

Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>

[TCP]: Code duplication removal, added tcp_bound_to_half_wnd()

Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>

[TCP]: cleanup tcp_{in,out}put.c style

These were manually selected from indent's results which as is
are too noisy to be of any use without human reason. In addition,
some extra newlines between function and its comment were removed
too.

Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>

[TCP]: reduce tcp_output's indentation levels a bit

Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>

[TCP]: Remove TCPCB_URG & TCPCB_AT_TAIL as unnecessary

The snd_up check should be enough. I suspect this has been
there to provide a minor optimization in clean_rtx_queue which
used to have a small if (!->sacked) block which could skip
snd_up check among the other work.

Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>

[TCP]: Dropped unnecessary skb/sacked accessing in reneging

SACK reneging can be precalculated to a FLAG in clean_rtx_queue
which has the right skb looked up. This will help a bit in
future because skb->sacked access will be changed eventually,
changing it already won't hurt any.

Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>

[TCP]: Introduce tcp_wnd_end() to reduce line lengths

Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>

[TCP]: Rename update_send_head & include related increment to it

There's very little need to have the packets_out incrementing in
a separate function. Also name the combined function
appropriately.

Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>

[TCP]: Make invariant check complain about invalid sacked_out

Earlier resolution for NewReno's sacked_out should now keep
it small enough for this to become invariant-like check.

Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>

[IPV4]: Remove unused multipath cached routing defintion in net/flow.h

Signed-off-by: Rami Rosen <ramirose@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

[UDP]: Add memory accounting.

Signed-off-by: Takahiro Yasui <tyasui@redhat.com>
Signed-off-by: Hideo Aoki <haoki@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NET] CORE: Introducing new memory accounting interface.

This patch introduces new memory accounting functions for each network
protocol. Most of them are renamed from memory accounting functions
for stream protocols. At the same time, some stream memory accounting
functions are removed since other functions do same thing.

Renaming:
sk_stream_free_skb() -> sk_wmem_free_skb()
__sk_stream_mem_reclaim() -> __sk_mem_reclaim()
sk_stream_mem_reclaim() -> sk_mem_reclaim()
sk_stream_mem_schedule  ->     __sk_mem_schedule()
sk_stream_pages()       -> sk_mem_pages()
sk_stream_rmem_schedule() -> sk_rmem_schedule()
sk_stream_wmem_schedule() -> sk_wmem_schedule()
sk_charge_skb() -> sk_mem_charge()

Removeing
sk_stream_rfree(): consolidates into sock_rfree()
sk_stream_set_owner_r(): consolidates into skb_set_owner_r()
sk_stream_mem_schedule()

The following functions are added.
     sk_has_account(): check if the protocol supports accounting
sk_mem_uncharge(): do the opposite of sk_mem_charge()

In addition, to achieve consolidation, updating sk_wmem_queued is
removed from sk_mem_charge().

Next, to consolidate memory accounting functions, this patch adds
memory accounting calls to network core functions. Moreover, present
memory accounting call is renamed to new accounting call.

Finally we replace present memory accounting calls with new interface
in TCP and SCTP.

Signed-off-by: Takahiro Yasui <tyasui@redhat.com>
Signed-off-by: Hideo Aoki <haoki@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

[IPV6]: Remove useless code from fib6_del_route().

There are useless codes in fib6_del_route(). The following patch has
been tested, every thing looks fine, as usual.

Signed-off-by: Gui Jianfeng <guijianfeng@cn.fujitsu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NEIGH]: Remove unused method from include/net/neighbour.h

Signed-off-by: Rami Rosen <ramirose@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

[IPV4]: Remove unused define in include/net/arp.h (HAVE_ARP_CREATE)

Signed-off-by: Rami Rosen <ramirose@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

[ATM]: [he] fixing compilation when you define USE_RBPS_POOL/USE_RBPL_POOL

Signed-off-by: Jorge Boncompte <jorge@dti2.net>
Signed-off-by: Chas Williams <chas@cmf.nrl.navy.mil>
Signed-off-by: David S. Miller <davem@davemloft.net>

[ATM]: [ambassador] kmalloc + memset conversion to kzalloc

Signed-off-by: Joonwoo Park <joonwpark81@gmail.com>
Signed-off-by: Chas Williams <chas@cmf.nrl.navy.mil>
Signed-off-by: David S. Miller <davem@davemloft.net>

[ATM]: [br2864] whitespace cleanup

Signed-off-by: Chas Williams <chas@cmf.nrl.navy.mil>
Signed-off-by: David S. Miller <davem@davemloft.net>

[ATM]: [br2864] routed support

Signed-off-by: Chas Williams <chas@cmf.nrl.navy.mil>
Signed-off-by: David S. Miller <davem@davemloft.net>

[ATM]: [he] This patch removes the ancient version string.

Signed-off-by: Adrian Bunk <bunk@stusta.de>
Signed-off-by: Chas Williams <chas@cmf.nrl.navy.mil>
Signed-off-by: David S. Miller <davem@davemloft.net>

[ATM]: Convert struct class_device to struct device

Signed-off-by: Kay Sievers <kay.sievers@vrfy.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Signed-off-by: Chas Williams <chas@cmf.nrl.navy.mil>

[ATM]: atm is no longer experimental

Signed-off-by: Chas Williams <chas@cmf.nrl.navy.mil>
Signed-off-by: David S. Miller <davem@davemloft.net>

[IPSEC]: Move all calls to xfrm_audit_state_icvfail to xfrm_input

Let's nip the code duplication in the bud :)

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>

[IPSEC]: Fix transport-mode async resume on intput without netfilter

When netfilter is off the transport-mode async resumption doesn't work
because we don't push back the IP header.  This patch fixes that by
moving most of the code outside of ifdef NETFILTER since the only part
that's not common is the short-circuit in the protocol handler.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>

[IPSEC]: Fix double free on skb on async output

When the output transform returns EINPROGRESS due to async operation we'll
free the skb the straight away as if it were an error.  This patch fixes
that so that the skb is freed when the async operation completes.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>

[LIBERTAS]: Remove last stray user of MAC_FMT.

Reported by Denis V. Lunev

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

[XFRM] Documentaion: Fix error example at XFRMOUTSTATEMODEERROR.

Signed-off-by: Masahide NAKAMURA <nakam@linux-ipv6.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

[TCP]: Remove seq_rtt ptr from clean_rtx_queue args

While checking Gavin's patch I noticed that the returned seq_rtt
is not used by the caller.

Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>

[TCP]: Force TSO splits to MSS boundaries

If snd_wnd - snd_nxt wasn't multiple of MSS, skb was split on
odd boundary by the callers of tcp_window_allows.

We try really hard to avoid unnecessary modulos. Therefore the
old caller side check "if (skb->len < limit)" was too wide as
well because limit is not bound in any way to skb->len and can
cause spurious testing for trimming in the middle of the queue
while we only wanted that to happen at the tail of the queue.
A simple additional caller side check for tcp_write_queue_tail
would likely have resulted 2 x modulos because the limit would
have to be first calculated from window, however, doing that
unnecessary modulo is not mandatory. After a minor change to
the algorithm, simply determine first if the modulo is needed
at all and at that point immediately decide also from which
value it should be calculated from.

This approach also kills some duplicated code.

Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>

[ETH]: Combine format_addr() with print_mac().

print_mac() used many most net drivers and format_addr() used by
net-sysfs.c are very similar and they can be intergrated.

format_addr() is also identically redefined in the qla4xxx iscsi
driver.

Export a new function sysfs_format_mac() to be used by net-sysfs,
qla4xxx and others in the future.  Both print_mac() and
sysfs_format_mac() call _format_mac_addr() to do the formatting.

Changed print_mac() to use unsigned char * to be consistent with
net_device struct's dev_addr.  Added buffer length overrun checking
as suggested by Joe Perches.

Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

[SOCK] Avoid divides in sk_stream_pages() and __sk_stream_mem_reclaim()

sk_forward_alloc being signed, we should take care of divides by
SK_STREAM_MEM_QUANTUM we do in sk_stream_pages() and
__sk_stream_mem_reclaim()

This patchs introduces SK_STREAM_MEM_QUANTUM_SHIFT, defined
as ilog2(SK_STREAM_MEM_QUANTUM), to be able to use right
shifts instead of plain divides.

This should help compiler to choose right shifts instead of
expensive divides (as seen with CONFIG_CC_OPTIMIZE_FOR_SIZE=y on x86)

Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

[XFRM]: Fix outbound statistics.

Signed-off-by: Masahide NAKAMURA <nakam@linux-ipv6.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NETNS]: Modify the neighbour table code so it handles multiple network namespaces

I'm actually surprised at how much was involved.  At first glance it
appears that the neighbour table data structures are already split by
network device so all that should be needed is to modify the user
interface commands to filter the set of neighbours by the network
namespace of their devices.

However a couple things turned up while I was reading through the
code.  The proxy neighbour table allows entries with no network
device, and the neighbour parms are per network device (except for the
defaults) so they now need a per network namespace default.

So I updated the two structures (which surprised me) with their very
own network namespace parameter.  Updated the relevant lookup and
destroy routines with a network namespace parameter and modified the
code that interacts with users to filter out neighbour table entries
for devices of other namespaces.

I'm a little concerned that we can modify and display the global table
configuration and from all network namespaces.  But this appears good
enough for now.

I keep thinking modifying the neighbour table to have per network
namespace instances of each table type would should be cleaner.  The
hash table is already dynamically sized so there are it is not a
limiter.  The default parameter would be straight forward to take care
of.  However when I look at the how the network table is built and
used I still find some assumptions that there is only a single
neighbour table for each type of table in the kernel.  The netlink
operations, neigh_seq_start, the non-core network users that call
neigh_lookup.  So while it might be doable it would require more
refactoring than my current approach of just doing a little extra
filtering in the code.

Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: Daniel Lezcano <dlezcano@fr.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

[XFRM]: Drop packets when replay counter would overflow

According to RFC4303, section 3.3.3 we need to drop outgoing packets which
cause the replay counter to overflow:

   3.3.3.  Sequence Number Generation

   The sender's counter is initialized to 0 when an SA is established.
   The sender increments the sequence number (or ESN) counter for this
   SA and inserts the low-order 32 bits of the value into the Sequence
   Number field.  Thus, the first packet sent using a given SA will
   contain a sequence number of 1.

   If anti-replay is enabled (the default), the sender checks to ensure
   that the counter has not cycled before inserting the new value in the
   Sequence Number field.  In other words, the sender MUST NOT send a
   packet on an SA if doing so would cause the sequence number to cycle.
   An attempt to transmit a packet that would result in sequence number
   overflow is an auditable event.  The audit log entry for this event
   SHOULD include the SPI value, current date/time, Source Address,
   Destination Address, and (in IPv6) the cleartext Flow ID.

Signed-off-by: Paul Moore <paul.moore@hp.com>
Acked-by: James Morris <jmorris@namei.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

[XFRM]: RFC4303 compliant auditing

This patch adds a number of new IPsec audit events to meet the auditing
requirements of RFC4303.  This includes audit hooks for the following events:

 * Could not find a valid SA [sections 2.1, 3.4.2]
   . xfrm_audit_state_notfound()
   . xfrm_audit_state_notfound_simple()

 * Sequence number overflow [section 3.3.3]
   . xfrm_audit_state_replay_overflow()

 * Replayed packet [section 3.4.3]
   . xfrm_audit_state_replay()

 * Integrity check failure [sections 3.4.4.1, 3.4.4.2]
   . xfrm_audit_state_icvfail()

While RFC4304 deals only with ESP most of the changes in this patch apply to
IPsec in general, i.e. both AH and ESP.  The one case, integrity check
failure, where ESP specific code had to be modified the same was done to the
AH code for the sake of consistency.

Signed-off-by: Paul Moore <paul.moore@hp.com>
Acked-by: James Morris <jmorris@namei.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

[TCP]: Avoid two divides in __tcp_grow_window()

tcp_win_from_space() being signed, compiler might emit an integer divide
to compute tcp_win_from_space()/2 .

Using right shifts is OK here and less expensive.

Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

[TCP]: Avoid a divide in tcp_mtu_probing()

tcp_mtu_to_mss() being signed, compiler might emit an integer divide
to compute tcp_mtu_to_mss()/2 .

Using a right shift is OK here and less expensive.

Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

[TCP]: Move mss variable in tcp_mtu_probing()

Down into the only scope where it is used.

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

[SOCK] Avoid integer divides where not necessary in include/net/sock.h

Because sk_wmem_queued, sk_sndbuf are signed, a divide per two
may force compiler to use an integer divide.

We can instead use a right shift.

Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

[TCP]: tcp_write_timeout.c cleanup

Before submiting a patch to change a divide to a right shift, I felt
necessary to create a helper function tcp_mtu_probing() to reduce length of
lines exceeding 100 chars in tcp_write_timeout().

Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

[INET]: Avoid an integer divide in rt_garbage_collect()

Since 'goal' is a signed int, compiler may emit an integer divide
to compute goal/2.

Using a right shift is OK here and less expensive.

Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

[TCP]: Convert several length variable to unsigned.

Several length variables cannot be negative, so convert int to
unsigned int.  This also allows us to do sane shift operations
on those variables.

Signed-off-by: YOSHIFUJI Hideaki <yoshfuji@linux-ipv6.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

net/mac80211/Kconfig: whitespace corrections

Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net/wireless/Kconfig: whitespace corrections

Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

mac80211: don't read ERP information from (re)association response

According to the standard, the field cannot be present, so don't
try to interpret it either.

Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Cc: Daniel Drake <dsd@gentoo.org>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

mac80211: move tx crypto decision

This patch moves the decision making about whether a frame is encrypted
with a certain algorithm up into the TX handlers rather than having it
in the crypto algorithm implementation.

This fixes a problem with the radiotap injection code where injecting
a non-data packet and requesting encryption could end up asking the
driver to encrypt a packet without giving it a key.

Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

mac80211: implement station stats retrieval

This implements the required cfg80211 callback in mac80211
to allow userspace to get station statistics.

Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

cfg80211/nl80211: implement station attribute retrieval

After a station is added to the kernel's structures, userspace
has to be able to retrieve statistics about that station, especially
whether the station was idle and how much bytes were transferred
to and from it. This adds the necessary code to nl80211.

Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

cfg80211/nl80211: station handling

This patch adds station handling to cfg80211/nl80211.

Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

cfg80211/nl80211: add beacon settings

This adds the necessary API to cfg80211/nl80211 to allow
changing beaconing settings.

Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

mac80211: support getting key sequence counters via cfg80211

This implements cfg80211's get_key() to allow retrieving the sequence
counter for a TKIP or CCMP key from userspace. It also cleans up and
documents the associated low-level driver interface.

Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

mac80211: support adding/removing keys via cfg80211

This adds the necessary hooks to mac80211 to allow userspace
to edit keys with cfg80211 (through nl80211.)

Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

cfg80211/nl80211: introduce key handling

This introduces key handling to cfg80211/nl80211. Default
and group keys can be added, changed and removed; sequence
counters for each key can be retrieved.

Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

doc: fix typo in feature-removal-schedule

Signed-off-by: Stefano Brivio <stefano.brivio@polimi.it>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

mac80211: allow easier multicast/broadcast buffering in hardware

There are various decisions influencing the decision whether to buffer
a frame for after the next DTIM beacon. The "do we have stations in PS
mode" condition cannot be tested by the driver so mac80211 has to do
that. To ease driver writing for hardware that can buffer frames until
after the next DTIM beacon, introduce a new txctl flag telling the
driver to buffer a specific frame.

While at it, restructure and comment the code for multicast buffering
and remove spurious "inline" directives.

Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Cc: Michael Buesch <mb@bu3sch.de>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

mac80211: make ieee80211_rx_mgmt_action static

The function is only used locally.

Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

mac80211: clean up eapol handling in TX path

The previous patch left only one user of the ieee80211_is_eapol()
function and that user can be eliminated easily by introducing
a new "frame is EAPOL" flag to handle the frame specially (we
already have this information) instead of doing the (expensive)
ieee80211_is_eapol() all the time.

Also, allow unencrypted frames to be sent when they are injected.

Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

mac80211: clean up eapol frame handling/port control

This cleans up the eapol frame handling and some related code in the
receive and transmit paths. After this patch
 * EAPOL frames addressed to us or the EAPOL group address are
   always accepted regardless of whether they are encrypted or not
 * other frames from a station are dropped if PAE is enabled and
   the station is not authorized
 * unencrypted frames (except the EAPOL frames above) are dropped if
   drop_unencrypted is enabled
 * some superfluous code that eth_type_trans handles anyway is gone
 * port control is done for transmitted packets

Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

rc80211-pid: export tuning parameters through debugfs

This adds all the tunable parameters used by rc80211_pid to debugfs for easy
testing and tuning.

Signed-off-by: Mattias Nissler <mattias.nissler@gmx.de>
Signed-off-by: Stefano Brivio <stefano.brivio@polimi.it>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

rc80211-pid: add debugging

This adds a new debugfs file from which rate control relevant events can be
read one event per line. The output includes the current time, so graphs can be
created showing the rate control parameters. This helps in evaluating and
tuning rate control parameters. While at it, we split headers and code for
better readability.

Signed-off-by: Mattias Nissler <mattias.nissler@gmx.de>
Signed-off-by: Stefano Brivio <stefano.brivio@polimi.it>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

rc80211-pid: add sharpening factor

This patch introduces a PID sharpening factor for faster response after
association and low activity events.

Signed-off-by: Stefano Brivio <stefano.brivio@polimi.it>
Signed-off-by: Mattias Nissler <mattias.nissler@gmx.de>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

rc80211-pid: add rate behaviour learning algorithm

This patch introduces a learning algorithm in order for the PID controller
to learn how to map adjustment values to rates. This is better described in
code comments.

Signed-off-by: Stefano Brivio <stefano.brivio@polimi.it>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

mac80211: make PID rate control algorithm the default

This makes the new PID TX rate control algorithm the default instead of the
rc80211_simple rate control algorithm. The simple algorithm was flawed in
several ways: it wasn't responsive at all and didn't age the information it was
relying on properly. The PID algorithm allows us to tune characteristics such
as responsiveness by adjusting parameters and was found to generally behave
better.

The default algorithm can be overridden to select simple instead. Which
ever algorithm is the default is included as part of the mac80211
module automatically. The other algorithm (simple vs. pid) can
be selected for inclusion as well. If EMBEDDED is selected then
the choice is available to have no default specified and neither
algorithm included in mac80211. The default algorithm can be set
through a modparam.

While at it, mark rc80211-simple as deprecated, and schedule it
for removal.

Signed-off-by: Stefano Brivio <stefano.brivio@polimi.it>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

[TCP] Avoid two divides in tcp_output.c

Because 'free_space' variable in __tcp_select_window() is signed,
expression (free_space / 2) forces compiler to emit an integer divide.

This can be changed to a plain right shift, less expensive.

Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

[XFRM]: Assorted IPsec fixups

This patch fixes a number of small but potentially troublesome things in the
XFRM/IPsec code:

 * Use the 'audit_enabled' variable already in include/linux/audit.h
   Removed the need for extern declarations local to each XFRM audit fuction

 * Convert 'sid' to 'secid' everywhere we can
   The 'sid' name is specific to SELinux, 'secid' is the common naming
   convention used by the kernel when refering to tokenized LSM labels,
   unfortunately we have to leave 'ctx_sid' in 'struct xfrm_sec_ctx' otherwise
   we risk breaking userspace

 * Convert address display to use standard NIP* macros
   Similar to what was recently done with the SPD audit code, this also also
   includes the removal of some unnecessary memcpy() calls

 * Move common code to xfrm_audit_common_stateinfo()
   Code consolidation from the "less is more" book on software development

 * Proper spacing around commas in function arguments
   Minor style tweak since I was already touching the code

Signed-off-by: Paul Moore <paul.moore@hp.com>
Acked-by: James Morris <jmorris@namei.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

[XFRM]: Add packet processing statistics option.

Signed-off-by: Masahide NAKAMURA <nakam@linux-ipv6.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

[XFRM]: Support to increment packet dropping statistics.

Signed-off-by: Masahide NAKAMURA <nakam@linux-ipv6.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

[XFRM]: Define packet dropping statistics.

This statistics is shown factor dropped by transformation
at /proc/net/xfrm_stat for developer.
It is a counter designed from current transformation source code
and defined as linux private MIB.

See Documentation/networking/xfrm_proc.txt for the detail.

Signed-off-by: Masahide NAKAMURA <nakam@linux-ipv6.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

[XFRM] MIPv6: Fix to input RO state correctly.

Disable spin_lock during xfrm_type.input() function.
Follow design as IPsec inbound does.

Signed-off-by: Masahide NAKAMURA <nakam@linux-ipv6.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

[XFRM] IPv6: Fix dst/routing check at transformation.

IPv6 specific thing is wrongly removed from transformation at net-2.6.25.
This patch recovers it with current design.

o Update "path" of xfrm_dst since IPv6 transformation should
  care about routing changes. It is required by MIPv6 and
  off-link destined IPsec.
o Rename nfheader_len which is for non-fragment transformation used by
  MIPv6 to rt6i_nfheader_len as IPv6 name space.

Signed-off-by: Masahide NAKAMURA <nakam@linux-ipv6.org>
Acked-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>

[TCP]: Fix TSO deferring

I'd say that most of what tcp_tso_should_defer had in between
there was dead code because of this.

Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>

[TG3]: Update version to 3.87

This patch updates the version number to 3.87.

Signed-off-by: Matt Carlson <mcarlson@broadcom.com>
Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

[TG3]: Fix supporting flowctrl code

This patch does three things.  It modifies tg3_setup_flow_control() to
use the administrator requested flow control settings if
autonegotiation is turned off.  It slightly modifies the
tg3_setup_fiber_mii_phy() function to account for this new use case.
And finally, it does the same for tg3_setup_copper_phy().

The copper modifications are more than a small multi-line change.  The
new code makes an attempt to avoid a link renegotiation if the link is
active at half duplex and the only difference between the current
advertised settings and requested advertised settings is the
flow control advertisements.

Signed-off-by: Matt Carlson <mcarlson@broadcom.com>
Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

[TG3]: Correct sw autoneg flow control advertisements

This patch modifies the software autoneg code to use the administrator
specified flow control parameters.  Since the autonegotiation code uses
alternative flow control enumerations, the 1000-BaseX utility functions
are used and code was added to convert the definitions to and from the
alternate enumerations.

Signed-off-by: Matt Carlson <mcarlson@broadcom.com>
Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>