[NETFILTER]: nf_conntrack: make print_conntrack function optional for l4protos

Allows to remove five empty implementations.

Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NETFILTER]: nf_conntrack: remove print_conntrack function from l3protos

Its unused and unlikely to ever be used.

Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NETFILTER]: nf_conntrack: clean up a few header files

- Remove declarations of non-existing variables and functions
- Move helper init/cleanup function declarations to nf_conntrack_helper.h
- Remove unneeded __nf_conntrack_attach declaration and make it static

Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NETFILTER]: kill nf_sysctl.c

Since there now is generic support for shared sysctl paths, the only
remains are the net/netfilter and net/ipv4/netfilter paths. Move them
to net/netfilter/core.c and net/ipv4/netfilter.c and kill nf_sysctl.c.

Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NETFILTER]: nf_conntrack_sctp: remove timeout indirection

Instead of keeping pointers to the timeout values in a table, simply
put the timeout values in the table directly.

Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NETFILTER]: nf_conntrack_sctp: replace magic value by symbolic constant

Use SCTP_CHUNK_FLAG_T instead of 0x1.

Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NETFILTER]: nf_conntrack_sctp: remove unused ttag field from conntrack data

Spotted by Pablo Neira Ayuso <pablo@netfilter.org>.

Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NETFILTER]: nf_conntrack_sctp: don't take sctp_lock once per chunk

Don't take and release the lock once per SCTP chunk, simply hold it
the entire time while iterating through the chunks.

Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NETFILTER]: nf_conntrack_sctp: rename "newconntrack" variable

The name is misleading, it holds the new connection state, so rename it
to "newstate". Also rename "oldsctpstate" to "oldstate" for consistency.

Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NETFILTER]: nf_conntrack_sctp: consolidate sctp_packet() error paths

Consolidate error paths and use proper symbolic return value instead
of magic values.

Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NETFILTER]: nf_conntrack_sctp: reduce line length further

Eliminate a few lines over 80 characters by using a local variable to
hold the conntrack direction instead of using CTINFO2DIR everywhere.

Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NETFILTER]: nf_conntrack_sctp: reduce line length

Reduce the length of some overly long lines by renaming all
"conntrack" variables to "ct".

Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NETFILTER]: nf_conntrack_sctp: use proper types for bitops

Use unsigned long instead of char for the bitmap and removed lots
of casts.

Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NETFILTER]: nf_conntrack_sctp: basic cleanups

Reindent switch cases properly, get rid of weird constructs like "!(x == y)",
put logical operations on the end of the line instead of the next line, get
rid of superfluous braces.

Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NETFILTER]: nf_conntrack_tcp: remove timeout indirection

Instead of keeping pointers to the timeout values in a table, simply
put the timeout values in the table directly.

Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NETFILTER]: nf_conntrack_{tcp,sctp}: shrink state table

The TCP and SCTP conntrack state transition tables only holds
small numbers, but gcc uses 4 byte per entry for the enum. Switching
to an u8 reduces the size from 480 to 120 bytes for TCP and from
576 to 144 bytes for SCTP.

Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NETFILTER]: nf_conntrack_{tcp,sctp}: mark state table const

Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NETFILTER]: ipt_REJECT: properly handle IP options

The current TCP RST construction reuses the old packet and can't
deal with IP options as a consequence of that. Construct the
RST from scratch instead.

Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NETFILTER]: {ip,ip6}_tables: remove some inlines

This patch removes inlines except those which are used
by packet matching code and thus are performance-critical.

Before:

$ size */*/*/ip*tables*.o
   text    data     bss     dec     hex filename
   6402     500      16    6918    1b06 net/ipv4/netfilter/ip_tables.o
   7130     500      16    7646    1dde net/ipv6/netfilter/ip6_tables.o

After:

$ size */*/*/ip*tables*.o
   text    data     bss     dec     hex filename
   6307     500      16    6823    1aa7 net/ipv4/netfilter/ip_tables.o
   7010     500      16    7526    1d66 net/ipv6/netfilter/ip6_tables.o

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NETFILTER]: Update feature-removal-schedule.txt

With all the newly introduced features, there is a lot to remove
later on after a compatibility grace period of 2 years.

Signed-off-by: Jan Engelhardt <jengelh@computergmbh.de>
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NETFILTER]: xt_iprange match, revision 1

Adds IPv6 support to xt_iprange, making it possible to match on IPv6
address ranges with ip6tables.

Signed-off-by: Jan Engelhardt <jengelh@computergmbh.de>
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NETFILTER]: Rename ipt_iprange to xt_iprange

This patch moves ipt_iprange to xt_iprange, in preparation for adding
IPv6 support to xt_iprange.

Signed-off-by: Jan Engelhardt <jengelh@computergmbh.de>
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NETFILTER]: Update modules' descriptions

Updates the MODULE_DESCRIPTION() tags for all Netfilter modules,
actually describing what the module does and not just
"netfilter XYZ target".

Signed-off-by: Jan Engelhardt <jengelh@computergmbh.de>
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NETFILTER]: xt_policy: use the new union nf_inet_addr

Signed-off-by: Jan Engelhardt <jengelh@computergmbh.de>
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NETFILTER]: xt_pkttype: IPv6 multicast address recognition

Signed-off-by: Jan Engelhart <jengelh@computergmbh.de>
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NETFILTER]: xt_pkttype: Add explicit check for IPv4

In the PACKET_LOOPBACK case, the skb data was always interpreted as
IPv4, but that is not valid for IPv6, obviously. Fix this by adding an
extra condition to check for AF_INET.

Signed-off-by: Jan Engelhardt <jengelh@computergmbh.de>
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NETFILTER]: xt_mark match, revision 1

Introduces the xt_mark match revision 1. It uses fixed types,
eventually obsoleting revision 0 some day (uses nonfixed types).

Signed-off-by: Jan Engelhardt <jengelh@computergmbh.de>
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NETFILTER]: xt_conntrack match, revision 1

Introduces the xt_conntrack match revision 1. It uses fixed types, the
new nf_inet_addr and comes with IPv6 support, thereby completely
superseding xt_state.

Signed-off-by: Jan Engelhardt <jengelh@computergmbh.de>
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NETFILTER]: Extend nf_inet_addr with in{,6}_addr

Extend union nf_inet_addr with struct in_addr and in6_addr. Useful
because a lot of in-kernel IPv4 and IPv6 functions use
in_addr/in6_addr.

Signed-off-by: Jan Engelhardt <jengelh@computergmbh.de>
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NETFILTER]: xt_connmark match, revision 1

Introduces the xt_connmark match revision 1. It uses fixed types,
eventually obsoleting revision 0 some day (uses nonfixed types).
(Unfixed types like "unsigned long" do not play well with mixed
user-/kernelspace "bitness", e.g. 32/64, as is common on SPARC64,
and need extra compat code.)

Signed-off-by: Jan Engelhardt <jengelh@computergmbh.de>
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NETFILTER]: xt_MARK target, revision 2

Introduces the xt_MARK target revision 2. It uses fixed types, and
also uses the more expressive XOR logic.

Signed-off-by: Jan Engelhardt <jengelh@computergmbh.de>
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NETFILTER]: xt_CONNMARK target, revision 1

Introduces the xt_CONNMARK target revision 1. It uses fixed types, and
also uses the more expressive XOR logic. Futhermore, it allows to
selectively pick bits from both the ctmark and the nfmark in the SAVE
and RESTORE operations.

Signed-off-by: Jan Engelhardt <jengelh@computergmbh.de>
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NETFILTER]: Annotate start of kernel fields in NF headers

Signed-off-by: Jan Engelhardt <jengelh@computergmbh.de>
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NETFILTER]: xt_TOS: Properly set the TOS field

Fix incorrect mask value passed to ipv4_change_dsfield/ipv6_change_dsfield.

Signed-off-by: Jan Engelhardt <jengelh@computergmbh.de>
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NETFILTER]: xt_TOS: Change semantic of mask value

This patch changes the behavior of xt_TOS v1 so that the mask value
the user supplies means "zero out these bits" rather than "keep these
bits". This is more easy on the user, as (I would assume) people keep
more bits than zeroing, so, an example:

Action:     Set bit 0x01.
     before (&): iptables -j TOS --set-tos 0x01/0xFE
     after (&~): iptables -j TOS --set-tos 0x01/0x01

This is not too "tragic" with xt_TOS, but where larger fields are used
(e.g. proposed xt_MARK v2), `--set-xmar 0x01/0x01` vs. `--set-xmark
0x01/0xFFFFFFFE` really makes a difference. Other target(!) modules,
such as xt_TPROXY also use &~ rather than &, so let's get to a common
ground.

(Since xt_TOS has not yet left the development tree en direction to
mainline, the semantic can be changed as proposed without breaking
iptables.)

Signed-off-by: Jan Engelhardt <jengelh@computergmbh.de>
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NETFILTER]: remove ipt_TOS.c

Commit 88c85d81f74f92371745158aebc5cbf490412002 forgot to remove the
old ipt_TOS file (whose code has been merged into xt_DSCP). Remove
it now.

Signed-off-by: Jan Engelhardt <jengelh@computergmbh.de>
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NETFILTER]: Remove some EXPERIMENTAL dependencies

Most of the netfilter modules are not considered experimental anymore,
the only ones I want to keep marked as EXPERIMENTAL are:

- TCPOPTSTRIP target, which is brand new.

- SANE helper, which is quite new.

- CLUSTERIP target, which I believe hasn't had much testing despite
  being in the kernel for quite a long time.

- SCTP match and conntrack protocol, which are a mess and need to
  be reviewed and cleaned up before I would trust them.

Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NETFILTER]: Hide a few more options under NETFILTER_ADVANCED

Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

[IPV4]: fib hash|trie initialization

Initialization of the slab cache's should be done when IP is
initialized to make sure of available memory, and that code can be
marked __init.

Signed-off-by: Stephen Hemminger <stephen.hemminger@vyatta.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

[IPV4] fib_trie: size and statistics

Show number of entries in trie, the size field was being set but never used,
but it only counted leaves, not all entries. Refactor the two cases in
fib_triestat_seq_show into a single routine.

Note: the stat structure was being malloc'd but the stack usage isn't so
high (288 bytes) that it is worth the additional complexity.

Signed-off-by: Stephen Hemminger <stephen.hemminger@vyatta.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

[FIB]: Avoid using static variables without proper locking

fib_trie_seq_show() uses two helper functions, rtn_scope() and
rtn_type() that can write to static storage without locking.

Just pass to them a temporary buffer to avoid potential corruption
(probably not triggerable but still...)

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

[NETNS]: Process inet_confirm_addr in the correct namespace.

inet_confirm_addr can be called with NULL in_dev from arp_ignore iff
scope is RT_SCOPE_LINK.

Lets always pass the device and check for RT_SCOPE_LINK scope inside
inet_confirm_addr. This let us take network namespace from in_device a
need for an additional argument.

Signed-off-by: Denis V. Lunev <den@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

[IPV4]: Remove extra argument from arp_ignore.

arp_ignore has two arguments: dev & in_dev. dev is used for
inet_confirm_addr calling only.

inet_confirm_addr, in turn, either gets in_dev from the device passed
or iterates over all network devices if the device passed is NULL. It
seems logical to directly pass in_dev into inet_confirm_addr.

Signed-off-by: Denis V. Lunev <den@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

[ARP]: neigh_parms_put(destroy) are essentially local to core/neighbour.c.

Make them static.

[ Moved the inline before, instead of after, call sites. -DaveM ]

Signed-off-by: Denis V. Lunev <den@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

[ARP]: Remove forward declaration of neigh_changeaddr.

No need for this. It is declared in the neighbour.h

Signed-off-by: Denis V. Lunev <den@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

[ARP]: Remove overkill checks from neigh_param_alloc.

Valid network device is always passed into neigh_param_alloc, so
remove extra checking for dev == NULL. Additionally, cleanup bogus
netns assignment.

Signed-off-by: Denis V. Lunev <den@openvz.org>
Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

[IPV4]: fib_rules_unregister is essentially void.

fib_rules_unregister is called only after successful register and the
return code is never checked.

Signed-off-by: Denis V. Lunev <den@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NETNS]: Make arp code network namespace consistent.

Some calls in the arp.c have network namespace as an argument. Getting
init_net inside these functions is simply inconsistent. Fix this.

Signed-off-by: Denis V. Lunev <den@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

[ARP]: Move inet_addr_type call after simple error checks in arp_contructor.

The neighbour entry will be destroyed in the case of error, so it is
pointless to perform constly routing table lookup in this case.

Signed-off-by: Denis V. Lunev <den@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NETNS][RAW]: Create the /proc/net/raw(6) in each namespace.

To do so, just register the proper subsystem and create files in
->init callbacks.

No other special per-namespace handling for raw sockets is required.

Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NETNS][RAW]: Eliminate explicit init_net references.

Happily, in all the rest places (->bind callbacks only), that require the
struct net, we have a socket, so get the net from it.

Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NETNS][RAW]: Make /proc/net/raw(6) show per-namespace socket list.

Pull the struct net pointer up to the showing functions
to filter the sockets depending on their namespaces.

Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NETNS][RAW]: Make ipv[46] raw sockets lookup namespaces aware.

This requires just to pass the appropriate struct net pointer
into __raw_v[46]_lookup and skip sockets that do not belong
to a needed namespace.

The proper net is get from skb->dev in all the cases.

Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

[FIB]: full_children & empty_children should be uint, not ushort

If declared as unsigned short, these fields can overflow, and whole
trie logic is broken. I could not make the machine crash, but some
tnode can never be freed.

Note for 64 bit arches : By reordering t_key and parent in [node,
leaf, tnode] structures, we can use 32 bits hole after t_key so that
sizeof(struct tnode) doesnt change after this patch.

Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Signed-off-by: Robert Olsson <robert.olsson@its.uu.se>
Signed-off-by: David S. Miller <davem@davemloft.net>

[AX25]: sparse cleanups

net/ax25/ax25_route.c:251:13: warning: context imbalance in
'ax25_rt_seq_start' - wrong count at exit
net/ax25/ax25_route.c:276:13: warning: context imbalance in 'ax25_rt_seq_stop'
- unexpected unlock
net/ax25/ax25_std_timer.c:65:25: warning: expensive signed divide
net/ax25/ax25_uid.c:46:1: warning: symbol 'ax25_uid_list' was not declared.
Should it be static?
net/ax25/ax25_uid.c:146:13: warning: context imbalance in 'ax25_uid_seq_start'
- wrong count at exit
net/ax25/ax25_uid.c:169:13: warning: context imbalance in 'ax25_uid_seq_stop'
- unexpected unlock
net/ax25/af_ax25.c:573:28: warning: expensive signed divide
net/ax25/af_ax25.c:1865:13: warning: context imbalance in 'ax25_info_start' -
wrong count at exit
net/ax25/af_ax25.c:1888:13: warning: context imbalance in 'ax25_info_stop' -
unexpected unlock
net/ax25/ax25_ds_timer.c:133:25: warning: expensive signed divide

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

[X25]: Avoid divides and sparse warnings

   CHECK   net/x25/af_x25.c
net/x25/af_x25.c:117:46: warning: expensive signed divide
   CHECK   net/x25/x25_facilities.c
net/x25/x25_facilities.c:209:30: warning: expensive signed divide
   CHECK   net/x25/x25_in.c
net/x25/x25_in.c:250:26: warning: expensive signed divide
   CHECK   net/x25/x25_proc.c
net/x25/x25_proc.c:48:11: warning: context imbalance in 'x25_seq_route_start'
- wrong count at exit
net/x25/x25_proc.c:72:13: warning: context imbalance in 'x25_seq_route_stop' -
unexpected unlock
net/x25/x25_proc.c:112:11: warning: context imbalance in
'x25_seq_socket_start' - wrong count at exit
net/x25/x25_proc.c:129:13: warning: context imbalance in 'x25_seq_socket_stop'
- unexpected unlock
net/x25/x25_proc.c:190:11: warning: context imbalance in
'x25_seq_forward_start' - wrong count at exit
net/x25/x25_proc.c:215:13: warning: context imbalance in
'x25_seq_forward_stop' - unexpected unlock
   CHECK   net/x25/x25_subr.c
net/x25/x25_subr.c:362:57: warning: expensive signed divide

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

[IPV4] fib_trie: removes a memset() call in tnode_new()

tnode_alloc() already clears allocated memory, using kcalloc() or
alloc_pages(GFP_KERNEL|__GFP_ZERO, ...)

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

[IPV4] FIB: Include nexthop device indexes in fib_info hashfn.

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

[XFRM]: Fix struct xfrm_algo code formatting.

Realign struct members.

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

[XFRM]: alg_key_len should be unsigned to avoid integer divides

alg_key_len is currently defined as 'signed int'. This unfortunatly
leads to integer divides in several paths.

Converting it to unsigned is safe and saves 208 bytes of text on i386.

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

[PKT_SCHED] HTB: htb_classid is dead static inline

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

[NET] core/utils.c: digit2bin is dead static inline

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

[FIB]: Reduce text size of net/ipv4/fib_trie.o

In struct tnode, we use two fields of 5 bits for 'pos' and 'bits'.
Switching to plain 'unsigned char' (8 bits) take the same space
because of compiler alignments, and reduce text size by 435 bytes
on i386.

On i386 :
$ size net/ipv4/fib_trie.o.before_patch net/ipv4/fib_trie.o
    text    data     bss     dec     hex filename
   13714       4      64   13782    35d6 net/ipv4/fib_trie.o.before
   13279       4      64   13347    3423 net/ipv4/fib_trie.o

Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Acked-by: Stephen Hemminger <stephen.hemminger@vyatta.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NETFILTER] xt_policy.c: kill some bloat

net/netfilter/xt_policy.c:
  policy_mt | -906
 1 function changed, 906 bytes removed, diff: -906

net/netfilter/xt_policy.c:
  match_xfrm_state | +427
 1 function changed, 427 bytes added, diff: +427

net/netfilter/xt_policy.o:
 2 functions changed, 427 bytes added, 906 bytes removed, diff: -479

Alternatively, this could be done by combining identical
parts of the match_policy_in/out()

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

[IPV4] fib_trie: Fix sparse warnings.

Make FIB TRIE go through sparse checker without warnings.

Signed-off-by: Stephen Hemminger <stephen.hemminger@vyatta.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

[IPV4] fib_trie: Add statistics.

The FIB TRIE code has a bunch of statistics, but the code is hidden
behind an ifdef that was never implemented. Since it was dead code, it
was broken as well.

This patch fixes that by making it a config option.

Signed-off-by: Stephen Hemminger <stephen.hemminger@vyatta.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

[IPV4] FIB: printk related cleanups

printk related cleanups:
 * Get rid of unused printk wrappers.
 * Make bug checks into KERN_WARNING because KERN_DEBUG gets ignored
 * Turn one cryptic old message into something real
 * Make sure all messages have KERN_XXX

Signed-off-by: Stephen Hemminger <stephen.hemminger@vyatta.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

[IPV4] fib_trie: fib_insert_node cleanup

The only error from fib_insert_node is if memory allocation fails, so
instead of passing by reference, just use the convention of returning
NULL.

Signed-off-by: Stephen Hemminger <stephen.hemminger@vyatta.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

[IPV4] fib_trie: Use %u for unsigned printfs.

Use %u instead of %d when printing unsigned values.

Signed-off-by: Stephen Hemminger <stephen.hemminger@vyatta.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

[IPV4] fib_trie: Get rid of unused revision element.

The revision element must of been part of an earlier design, because
currently it is set but never used.

Signed-off-by: Stephen Hemminger <stephen.hemminger@vyatta.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

[IPV4] fib_trie: Get rid of trie_init().

trie_init is worthless it is just zeroing stuff that is already zero!
Move the memset() down to make it obvious.

Signed-off-by: Stephen Hemminger <stephen.hemminger@vyatta.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

[PKTGEN]: uninline getCurUs

net/core/pktgen.c:
  pktgen_stop_device   |  -50
  pktgen_run           | -105
  pktgen_if_show       |  -37
  pktgen_thread_worker | -702
 4 functions changed, 894 bytes removed, diff: -894

net/core/pktgen.c:
  getCurUs |  +36
 1 function changed, 36 bytes added, diff: +36

net/core/pktgen.o:
 5 functions changed, 36 bytes added, 894 bytes removed, diff: -858

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

[PKTGEN]: Kill dead static inlines

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

[NETLINK] af_netlink: kill some bloat

net/netlink/af_netlink.c:
  netlink_realloc_groups        |  -46
  netlink_insert                |  -49
  netlink_autobind              |  -94
  netlink_clear_multicast_users |  -48
  netlink_bind                  |  -55
  netlink_setsockopt            |  -54
  netlink_release               |  -86
  netlink_kernel_create         |  -47
  netlink_change_ngroups        |  -56
 9 functions changed, 535 bytes removed, diff: -535

net/netlink/af_netlink.c:
  netlink_table_ungrab |  +53
 1 function changed, 53 bytes added, diff: +53

net/netlink/af_netlink.o:
 10 functions changed, 53 bytes added, 535 bytes removed, diff: -482

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

[IPV6] route: kill some bloat

net/ipv6/route.c:
  ip6_pkt_prohibit_out | -130
  ip6_pkt_discard      | -261
  ip6_pkt_discard_out  | -130
  ip6_pkt_prohibit     | -261
 4 functions changed, 782 bytes removed, diff: -782

net/ipv6/route.c:
  ip6_pkt_drop | +300
 1 function changed, 300 bytes added, diff: +300

net/ipv6/route.o:
 5 functions changed, 300 bytes added, 782 bytes removed, diff: -482

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

[XFRM] xfrm_policy: kill some bloat

net/xfrm/xfrm_policy.c:
  xfrm_audit_policy_delete | -692
  xfrm_audit_policy_add    | -692
 2 functions changed, 1384 bytes removed, diff: -1384

net/xfrm/xfrm_policy.c:
  xfrm_audit_common_policyinfo | +704
 1 function changed, 704 bytes added, diff: +704

net/xfrm/xfrm_policy.o:
 3 functions changed, 704 bytes added, 1384 bytes removed, diff: -680

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

[TCP]: Uninline tcp_is_cwnd_limited

net/ipv4/tcp_cong.c:
  tcp_reno_cong_avoid |  -65
 1 function changed, 65 bytes removed, diff: -65

net/ipv4/arp.c:
  arp_ignore |   -5
 1 function changed, 5 bytes removed, diff: -5

net/ipv4/tcp_bic.c:
  bictcp_cong_avoid |  -57
 1 function changed, 57 bytes removed, diff: -57

net/ipv4/tcp_cubic.c:
  bictcp_cong_avoid |  -61
 1 function changed, 61 bytes removed, diff: -61

net/ipv4/tcp_highspeed.c:
  hstcp_cong_avoid |  -63
 1 function changed, 63 bytes removed, diff: -63

net/ipv4/tcp_hybla.c:
  hybla_cong_avoid |  -85
 1 function changed, 85 bytes removed, diff: -85

net/ipv4/tcp_htcp.c:
  htcp_cong_avoid |  -57
 1 function changed, 57 bytes removed, diff: -57

net/ipv4/tcp_veno.c:
  tcp_veno_cong_avoid |  -52
 1 function changed, 52 bytes removed, diff: -52

net/ipv4/tcp_scalable.c:
  tcp_scalable_cong_avoid |  -61
 1 function changed, 61 bytes removed, diff: -61

net/ipv4/tcp_yeah.c:
  tcp_yeah_cong_avoid |  -75
 1 function changed, 75 bytes removed, diff: -75

net/ipv4/tcp_illinois.c:
  tcp_illinois_cong_avoid |  -54
 1 function changed, 54 bytes removed, diff: -54

net/dccp/ccids/ccid3.c:
  ccid3_update_send_interval |   -7
  ccid3_hc_tx_packet_recv    |   +7
 2 functions changed, 7 bytes added, 7 bytes removed, diff: +0

net/ipv4/tcp_cong.c:
  tcp_is_cwnd_limited |  +88
 1 function changed, 88 bytes added, diff: +88

built-in.o:
 14 functions changed, 95 bytes added, 642 bytes removed, diff: -547

...Again some gcc artifacts visible as well.

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

[TCP]: Uninline tcp_set_state

net/ipv4/tcp.c:
  tcp_close_state | -226
  tcp_done        | -145
  tcp_close       | -564
  tcp_disconnect  | -141
 4 functions changed, 1076 bytes removed, diff: -1076

net/ipv4/tcp_input.c:
  tcp_fin               |  -86
  tcp_rcv_state_process | -164
 2 functions changed, 250 bytes removed, diff: -250

net/ipv4/tcp_ipv4.c:
  tcp_v4_connect | -209
 1 function changed, 209 bytes removed, diff: -209

net/ipv4/arp.c:
  arp_ignore |   +5
 1 function changed, 5 bytes added, diff: +5

net/ipv6/tcp_ipv6.c:
  tcp_v6_connect | -158
 1 function changed, 158 bytes removed, diff: -158

net/sunrpc/xprtsock.c:
  xs_sendpages |   -2
 1 function changed, 2 bytes removed, diff: -2

net/dccp/ccids/ccid3.c:
  ccid3_update_send_interval |   +7
 1 function changed, 7 bytes added, diff: +7

net/ipv4/tcp.c:
  tcp_set_state | +238
 1 function changed, 238 bytes added, diff: +238

built-in.o:
 12 functions changed, 250 bytes added, 1695 bytes removed, diff: -1445

I've no explanation why some unrelated changes seem to occur
consistently as well (arp_ignore, ccid3_update_send_interval;
I checked the arp_ignore asm and it seems to be due to some
reordered of operation order causing some extra opcodes to be
generated). Still, the benefits are pretty obvious from the
codiff's results.

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

[NETNS][IPV6]: inet6_addr - make ipv6_chk_home_addr namespace aware

Looks if the address is belonging to the network namespace, otherwise
discard the address for the check.

Signed-off-by: Daniel Lezcano <dlezcano@fr.ibm.com>
Signed-off-by: Benjamin Thery <benjamin.thery@bull.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NETNS][IPV6]: inet6_addr - ipv6_get_ifaddr namespace aware

The inet6_addr_lst is browsed taking into account the network
namespace specified as parameter. If an address does not belong
to the specified namespace, it is ignored.

Signed-off-by: Daniel Lezcano <dlezcano@fr.ibm.com>
Signed-off-by: Benjamin Thery <benjamin.thery@bull.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NETNS][IPV6]: inet6_addr - ipv6_chk_same_addr namespace aware

This patch makes ipv6_chk_same_addr function to be aware of the
network namespace. The addresses not belonging to the network
namespace are discarded.

Signed-off-by: Daniel Lezcano <dlezcano@fr.ibm.com>
Signed-off-by: Benjamin Thery <benjamin.thery@bull.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NETNS][IPV6]: inet6_addr - check ipv6 address per namespace

When a new address is added, we must check if the new address does not
already exists.  This patch makes this check to be aware of a network
namespace, so the check will look if the address already exists for
the specified network namespace. While the addresses are browsed, the
addresses which do not belong to the namespace are discarded.

Signed-off-by: Daniel Lezcano <dlezcano@fr.ibm.com>
Signed-off-by: Benjamin Thery <benjamin.thery@bull.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NETNS][IPV6]: inet6_addr - isolate inet6 addresses from proc file

Make /proc/net/if_inet6 show only inet6 addresses belonging to the
namespace.

Signed-off-by: Daniel Lezcano <dlezcano@fr.ibm.com>
Signed-off-by: Benjamin Thery <benjamin.thery@bull.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NEIGH]: Add a comment describing what a NUD stands for.

When I studied the neighbor code I puzzled over what the NUD can mean
for quite a long time.

Finally I asked Alexey and he said that this was smth like "neighbor
unreachability detection".

Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

[TCP]: Do not purge sk_forward_alloc entirely in tcp_delack_timer().

Otherwise we beat heavily on the global tcp_memory atomics
when all of the sockets in the system are slowly sending
perioding packet clumps.

Noticed and suggested by Eric Dumazet.

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

[NETNS]: Use the per-net ipv6_devconf(_all) in sysctl handlers

Actually the net->ipv6.devconf_all can be used in a few places,
but to keep the /proc/sys/net/ipv6/conf/ sysctls work consistently
in the namespace we should use the per-net devconf_all in the
sysctl "forwarding" handler.

Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NETNS]: Use the per-net ipv6_devconf_dflt

All its users are in net/ipv6/addrconf.c's sysctl handlers.
Since they already have the struct net to get from, the
per-net ipv6_devconf_dflt can already be used.

Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NETNS]: Create ipv6 devconf-s for namespaces

This is the core. Declare and register the pernet subsys for
addrconf. The init callback the will create the devconf-s.

The init_net will reuse the existing statically declared confs,
so that accessing them from inside the ipv6 code will still
work.

The register_pernet_subsys() is moved above the ipv6_add_dev()
call for loopback, because this function will need the
net->devconf_dflt pointer to be already set.

Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NETNS]: Make the ctl-tables per-namespace

This includes passing the net to __addrconf_sysctl_register
and saving this on the ctl_table->extra2 to be used in
handlers (those, needing it).

Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NETNS]: Make the __addrconf_sysctl_register return an error

This error code will be needed to abort the namespace
creation if needed.

Probably, this is to be checked when a new device is
created (currently it is ignored).

Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NETNS]: Clean out the ipv6-related sysctls creation/destruction

The addrconf sysctls and neigh sysctls are registered and
unregistered always in pairs, so they can be joined into
one (well, two) functions, that accept the struct inet6_dev
and do all the job.

This also get rids of unneeded ifdefs inside the code.

Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NEIGH]: Make /proc/net/arp opening consistent with seq_net_open semantics

seq_open_net requires that first field of the seq->private data to be
struct seq_net_private. In reality this is a single pointer to a
struct net for now. The patch makes code consistent.

Signed-off-by: Denis V. Lunev <den@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

[ATM]: Simplify /proc/net/atm/arp opening

The iterator state->ns.neigh_sub_iter initialization is moved from
arp_seq_open to clip_seq_start for convinience. This should not be a
problem as the iterator will be used only after the seq_start
callback.

Signed-off-by: Denis V. Lunev <den@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

[ATM]: Oops reading net/atm/arp

cat /proc/net/atm/arp causes the NULL pointer dereference in the
get_proc_net+0xc/0x3a. This happens as proc_get_net believes that the
parent proc dir entry contains struct net.

Fix this assumption for "net/atm" case.

The problem is introduced by the commit c0097b07abf5f92ab135d024dd41bd2aada1512f
from Eric W. Biederman/Daniel Lezcano.

Signed-off-by: Denis V. Lunev <den@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NETNS]: Enable routing configuration in non-initial namespace.

I.e. remove the net != &init_net checks from the places, that now can
handle other-than-init net namespace.

Acked-by: Benjamin Thery <benjamin.thery@bull.net>
Acked-by: Daniel Lezcano <dlezcano@fr.ibm.com>
Signed-off-by: Denis V. Lunev <den@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NETNS]: Replace init_net with the correct context in fib_frontend.c

Acked-by: Benjamin Thery <benjamin.thery@bull.net>
Acked-by: Daniel Lezcano <dlezcano@fr.ibm.com>
Signed-off-by: Denis V. Lunev <den@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NETNS]: Pass namespace through ip_rt_ioctl.

... up to rtentry_to_fib_config

Acked-by: Benjamin Thery <benjamin.thery@bull.net>
Acked-by: Daniel Lezcano <dlezcano@fr.ibm.com>
Signed-off-by: Denis V. Lunev <den@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NETNS]: Correctly fill fib_config data.

Acked-by: Benjamin Thery <benjamin.thery@bull.net>
Acked-by: Daniel Lezcano <dlezcano@fr.ibm.com>
Signed-off-by: Denis V. Lunev <den@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NETNS]: Provide correct namespace for fibnl netlink socket.

This patch makes the netlink socket to be per namespace. That allows
to have each namespace its own socket for routing queries.

Acked-by: Benjamin Thery <benjamin.thery@bull.net>
Acked-by: Daniel Lezcano <dlezcano@fr.ibm.com>
Signed-off-by: Denis V. Lunev <den@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NETNS]: Place fib tables into netns.

The preparatory work has been done. All we need is to substitute
fib_table_hash with net->ipv4.fib_table_hash. Netns context is
available when required.

Acked-by: Benjamin Thery <benjamin.thery@bull.net>
Acked-by: Daniel Lezcano <dlezcano@fr.ibm.com>
Signed-off-by: Denis V. Lunev <den@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>