[NETFILTER]: ebtables: add --snap-arp option

The attached patch adds --snat-arp support, which makes it possible to
change the source mac address in both the mac header and the arp header
with one rule.

Signed-off-by: Bart De Schuymer <bdschuym@pandora.be>
Signed-off-by: Patrick McHardy <kaber@trash.net>

[NETFILTER]: x_tables: add NFLOG target

Add new NFLOG target to allow use of nfnetlink_log for both IPv4 and IPv6.
Currently we have two (unsupported by userspace) hacks in the LOG and ULOG
targets to optionally call to the nflog API. They lack a few features,
namely the IPv4 and IPv6 LOG targets can not specify a number of arguments
related to nfnetlink_log, while the ULOG target is only available for IPv4.
Remove those hacks and add a clean way to use nfnetlink_log.

Signed-off-by: Patrick McHardy <kaber@trash.net>

[NETFILTER]: x_tables: add port of hashlimit match for IPv4 and IPv6

Signed-off-by: Patrick McHardy <kaber@trash.net>

[NETFILTER]: nfnetlink_log: remove useless prefix length limitation

There is no reason for limiting netlink attributes in size.

Signed-off-by: Patrick McHardy <kaber@trash.net>

[NETFILTER]: nfnetlink_queue: allow changing queue length through netlink

Signed-off-by: Eric Leblond <eric@inl.fr>
Signed-off-by: Patrick McHardy <kaber@trash.net>

[NETFILTER]: ctnetlink: rework conntrack fields dumping logic on events

               |   NEW   | UPDATE  | DESTROY |
     ----------------------------------------|
     tuples    |    Y    |    Y    |    Y    |
     status    |    Y    |    Y    |    N    |
     timeout   |    Y    |    Y    |    N    |
     protoinfo |    S    |    S    |    N    |
     helper    |    S    |    S    |    N    |
     mark      |    S    |    S    |    N    |
     counters  |    F    |    F    |    Y    |

 Leyend:
         Y: yes
         N: no
         S: iif the field is set
 F: iif overflow

This patch also replace IPCT_HELPINFO by IPCT_HELPER since we want to
track the helper assignation process, not the changes in the private
information held by the helper.

Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Signed-off-by: Patrick McHardy <kaber@trash.net>

[NETFILTER]: ctnetlink: check for status attribute existence on conntrack creation

Check that status flags are available in the netlink message received
to create a new conntrack.

Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Signed-off-by: Patrick McHardy <kaber@trash.net>

[NETFILTER]: sip conntrack: better NAT handling

The NAT handling of the SIP helper has a few problems:

- Request headers are only mangled in the reply direction, From/To headers
  not at all, which can lead to authentication failures with DNAT in case
  the authentication domain is the IP address

- Contact headers in responses are only mangled for REGISTER responses

- Headers may be mangled even though they contain addresses not
  participating in the connection, like alternative addresses

- Packets are droppen when domain names are used where the helper expects
  IP addresses

This patch takes a different approach, instead of fixed rules what field
to mangle to what content, it adds symetric mapping of From/To/Via/Contact
headers, which allows to deal properly with echoed addresses in responses
and foreign addresses not belonging to the connection.

Signed-off-by: Patrick McHardy <kaber@trash.net>

[NETFILTER]: sip conntrack: make header shortcuts optional

Not every header has a shortcut, so make them optional instead
of searching for the same string twice.

Signed-off-by: Patrick McHardy <kaber@trash.net>

[NETFILTER]: sip conntrack: do case insensitive SIP header search

SIP headers are generally case-insensitive, only SDP headers are
case sensitive.

Signed-off-by: Patrick McHardy <kaber@trash.net>

[NETFILTER]: sip conntrack: minor cleanup

- Use enum for header field enumeration
- Use numerical value instead of pointer to header info structure to
  identify headers, unexport ct_sip_hdrs
- group SIP and SDP entries in header info structure
- remove double forward declaration of ct_sip_get_info

Signed-off-by: Patrick McHardy <kaber@trash.net>

[NETFILTER]: ip_conntrack: fix NAT helper unload races

The NAT helpr hooks are protected by RCU, but all of the
conntrack helpers test and use the global pointers instead
of copying them first using rcu_dereference()

Also replace synchronize_net() by synchronize_rcu() for clarity
since sychronizing only with packet receive processing is
insufficient to prevent races.

Signed-off-by: Patrick McHardy <kaber@trash.net>

[NETFILTER]: conntrack: add '_get' to {ip, nf}_conntrack_expect_find

We usually uses 'xxx_find_get' for function which increments
reference count.

Signed-off-by: Yasuyuki Kozakai <yasuyuki.kozakai@toshiba.co.jp>
Signed-off-by: Patrick McHardy <kaber@trash.net>

[NETFILTER]: nf_conntrack: /proc compatibility with old connection tracking

This patch adds /proc/net/ip_conntrack, /proc/net/ip_conntrack_expect and
/proc/net/stat/ip_conntrack files to keep old programs using them working.

The /proc/net/ip_conntrack and /proc/net/ip_conntrack_expect files show only
IPv4 entries, the /proc/net/stat/ip_conntrack shows global statistics.

Signed-off-by: Patrick McHardy <kaber@trash.net>

[NETFILTER]: nf_conntrack: sysctl compatibility with old connection tracking

This patch adds an option to keep the connection tracking sysctls visible
under their old names.

Signed-off-by: Patrick McHardy <kaber@trash.net>

[NETFILTER]: nf_conntrack: move conntrack protocol sysctls to individual modules

Signed-off-by: Patrick McHardy <kaber@trash.net>

[NETFILTER]: nf_conntrack: automatic sysctl registation for conntrack protocols

Add helper functions for sysctl registration with optional instantiating
of common path elements (like net/netfilter) and use it for support for
automatic registation of conntrack protocol sysctls.

Signed-off-by: Patrick McHardy <kaber@trash.net>

[NETFILTER]: nf_conntrack: move extern declaration to header files

Using extern in a C file is a bad idea because the compiler can't
catch type errors.

Signed-off-by: Patrick McHardy <kaber@trash.net>

[NETFILTER]: nf_conntrack_ftp: fix missing helper mask initilization

Signed-off-by: Patrick McHardy <kaber@trash.net>

[NETFILTER]: nf_conntrack: reduce timer updates in __nf_ct_refresh_acct()

Only update the conntrack timer if there's been at least HZ jiffies since
the last update. Reduces the number of del_timer/add_timer cycles from one
per packet to one per connection per second (plus once for each state change
of a connection)

Should handle timer wraparounds and connection timeout changes.

Signed-off-by: Martin Josefsson <gandalf@wlug.westbo.se>
Signed-off-by: Patrick McHardy <kaber@trash.net>

[NETFILTER]: nf_conntrack: remove unused struct list_head from protocols

Remove unused struct list_head from struct nf_conntrack_l3proto and
nf_conntrack_l4proto as all protocols are kept in arrays, not linked
lists.

Signed-off-by: Martin Josefsson <gandalf@wlug.westbo.se>
Signed-off-by: Patrick McHardy <kaber@trash.net>

[NETFILTER]: nf_conntrack: minor __nf_ct_refresh_acct() whitespace cleanup

Minor whitespace cleanup.

Signed-off-by: Martin Josefsson <gandalf@wlug.westbo.se>
Signed-off-by: Patrick McHardy <kaber@trash.net>

[NETFILTER]: nf_conntrack: remove ASSERT_{READ,WRITE}_LOCK

Remove the usage of ASSERT_READ_LOCK/ASSERT_WRITE_LOCK in nf_conntrack,
it didn't do anything, it was just an empty define and it uglified the code.

Signed-off-by: Martin Josefsson <gandalf@wlug.westbo.se>
Signed-off-by: Patrick McHardy <kaber@trash.net>

[NETFILTER]: nf_conntrack: more sanity checks in protocol registration/unregistration

Add some more sanity checks when registering/unregistering l3/l4 protocols.

Signed-off-by: Martin Josefsson <gandalf@wlug.westbo.se>
Signed-off-by: Patrick McHardy <kaber@trash.net>

[NETFILTER]: nf_conntrack: rename struct nf_conntrack_protocol

Rename 'struct nf_conntrack_protocol' to 'struct nf_conntrack_l4proto' in
order to help distinguish it from 'struct nf_conntrack_l3proto'. It gets
rather confusing with 'nf_conntrack_protocol'.

Signed-off-by: Martin Josefsson <gandalf@wlug.westbo.se>
Signed-off-by: Patrick McHardy <kaber@trash.net>

[NETFILTER]: More __read_mostly annotations

Place rarely written variables in the read-mostly section by using
__read_mostly

Signed-off-by: Martin Josefsson <gandalf@wlug.westbo.se>
Signed-off-by: Patrick McHardy <kaber@trash.net>

[NETFILTER]: nf_conntrack: split out protocol handling

This patch splits out L3/L4 protocol handling into its own file
nf_conntrack_proto.c

Signed-off-by: Martin Josefsson <gandalf@wlug.westbo.se>
Signed-off-by: Patrick McHardy <kaber@trash.net>

[NETFILTER]: nf_conntrack: split out the event cache

This patch splits out the event cache into its own file
nf_conntrack_ecache.c

Signed-off-by: Martin Josefsson <gandalf@wlug.westbo.se>
Signed-off-by: Patrick McHardy <kaber@trash.net>

[NETFILTER]: nf_conntrack: split out helper handling

This patch splits out handling of helpers into its own file
nf_conntrack_helper.c

Signed-off-by: Martin Josefsson <gandalf@wlug.westbo.se>
Signed-off-by: Patrick McHardy <kaber@trash.net>

[NETFILTER]: nf_conntrack: split out expectation handling

This patch splits out expectation handling into its own file
nf_conntrack_expect.c

Signed-off-by: Martin Josefsson <gandalf@wlug.westbo.se>
Signed-off-by: Patrick McHardy <kaber@trash.net>

[TCP] Vegas: Increase default alpha to 2 and beta to 4.

This helps Vegas cope better with delayed ACKs, see
analysis at:

http://www.cs.caltech.edu/%7Eweixl/technical/ns2linux/known_linux/index.html#vegas

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

[DCCP]: Use `unsigned' for packet lengths

This patch implements a suggestion by Ian McDonald and

 1) Avoids tests against negative packet lengths by using unsigned int
    for packet payload lengths in the CCID send_packet()/packet_sent() routines

 2) As a consequence, it removes an now unnecessary test with regard to `len > 0'
    in ccid3_hc_tx_packet_sent: that condition is always true, since
      * negative packet lengths are avoided
      * ccid3_hc_tx_send_packet flags an error whenever the payload length is 0.
        As a consequence, ccid3_hc_tx_packet_sent is never called as all errors
        returned by ccid_hc_tx_send_packet are caught in dccp_write_xmit

 3) Removes the third argument of ccid_hc_tx_send_packet (the `len' parameter),
    since it is currently always set to skb->len. The code is updated with regard
    to this parameter change.

Signed-off-by: Gerrit Renker <gerrit@erg.abdn.ac.uk>
Signed-off-by: Ian McDonald <ian.mcdonald@jandi.co.nz>
Signed-off-by: Arnaldo Carvalho de Melo <acme@mandriva.com>

[DCCP] ccid3: Larger initial windows

This implements the larger-initial-windows feature for CCID 3, as described in
section 5 of RFC 4342. When the first feedback packet arrives, the sender can
send up to 2..4 packets per RTT, instead of just one.

The patch further
 * reduces the number of timestamping calls by passing the timestamp value
   (which is computed in one of the calling functions anyway) as argument

 * renames one constant with a very long name into one which is shorter and
   resembles the one in RFC 3448 (t_mbi)

 * simplifies some of the min_t/max_t cases where both `x', `y' have the same
   type

Commiter note: renamed TFRC_t_mbi to TFRC_T_MBI, to follow Linux coding style.

Signed-off-by: Gerrit Renker <gerrit@erg.abdn.ac.uk>
Acked-by: Ian McDonald <ian.mcdonald@jandi.co.nz>
Signed-off-by: Arnaldo Carvalho de Melo <acme@mandriva.com>

[DCCP]: Make {set,get}sockopt(DCCP_SOCKOPT_PACKET_SIZE) return 0

To reflect the fact that this now is of no effect, not making apps
stop working, just be warned in the system log.

Signed-off-by: Arnaldo Carvalho de Melo <acme@mandriva.com>

[DCCP]: Tidy up unused structures

This removes and cleans up unused variables and structures which have become
unnecessary following the introduction of the EWMA patch to automatically track
the CCID 3 receiver/sender packet sizes `s'.

It deprecates the PACKET_SIZE socket option by returning an error code and
printing a deprecation warning if an application tries to read or write this
socket option.

Signed-off-by: Gerrit Renker <gerrit@erg.abdn.ac.uk>
Signed-off-by: Arnaldo Carvalho de Melo <acme@mandriva.com>

[DCCP] ccid3: Track RX/TX packet size `s' using moving-average

Problem:

[DCCP] ccid3: Set NoFeedback Timeout according to RFC 3448

This corrects the setting of the nofeedback timer with regard to RFC
3448 - previously it was not set to max(4*R, 2*s/X) as specified. Using
the maximum of 1 second as upper bound (as it was done before) can have
detrimental effects, especially if R is small.

Signed-off-by: Ian McDonald <ian.mcdonald@jandi.co.nz>
Signed-off-by: Gerrit Renker <gerrit@erg.abdn.ac.uk>
Signed-off-by: Arnaldo Carvalho de Melo <acme@mandriva.com>

[DCCP]: Remove allocation of sysctl numbers

This is in response to a request sent earlier by Eric W. Biederman
and replaces all sysctl numbers for net.dccp.default with CTL_UNNUMBERED.

It has been tested to compile and to work.

Commiter note: I've removed the use of CTL_UNNUMBERED, not setting .ctl_name
               sets it to 0, that is the what CTL_UNNUMBERED is, reason is
               to avoid unneeded source code cluttering.

Signed-off-by: Gerrit Renker <gerrit@erg.abdn.ac.uk>
Signed-off-by: Ian McDonald <ian.mcdonald@jandi.co.nz>
Signed-off-by: Arnaldo Carvalho de Melo <acme@mandriva.com>

[INET]: Change protocol field in struct inet_protosw to u16

[acme@newtoy net-2.6.20]$ pahole /tmp/tcp_ipv6.o inet_protosw
/* /pub/scm/linux/kernel/git/acme/net-2.6.20/include/net/protocol.h:69 */
struct inet_protosw {
        struct list_head           list;                 /*     0     8 */
        short unsigned int         type;                 /*     8     2 */

        /* XXX 2 bytes hole, try to pack */

        int                        protocol;             /*    12     4 */
        struct proto *             prot;                 /*    16     4 */
        const struct proto_ops  *  ops;                  /*    20     4 */
        int                        capability;           /*    24     4 */
        char                       no_check;             /*    28     1 */
        unsigned char              flags;                /*    29     1 */
}; /* size: 32, sum members: 28, holes: 1, sum holes: 2, padding: 2 */

So that we can kill that hole, protocol can only go all the way to 255 (RAW).

Signed-off-by: Arnaldo Carvalho de Melo <acme@mandriva.com>

[TCP]: Renove the __ prefix on the struct tcp_sock members

As this struct is not userland visible at all.

Signed-off-by: Arnaldo Carvalho de Melo <acme@mandriva.com>

[TCP]: Change tcp_header_len member in tcp_sock to u16

With this we eliminate the last hole in struct tcp_sock.

End result:

[acme@newtoy net-2.6.20]$ codiff -sV /tmp/tcp.o.before net/ipv4/tcp.o
/pub/scm/linux/kernel/git/acme/net-2.6.20/net/ipv4/tcp.c:
  struct tcp_sock |   -4
    tcp_header_len;
     from: int                   /*  1000(0)     4(0) */
     to:   u16                   /*  1000(0)     2(0) */
 1 struct changed
[acme@newtoy net-2.6.20]$

Now sizeof(tcp_sock) is just...

[acme@newtoy net-2.6.20]$ pahole --sizes ../OUTPUT/qemu/net-2.6.20/net/ipv4/tcp.o | grep -w tcp_sock
struct tcp_sock: 1500 0

1500 bytes ;-)

Signed-off-by: Arnaldo Carvalho de Melo <acme@mandriva.com>

[DCCP] ccid3: Consolidate handling of t_RTO

This patch
 * removes setting t_RTO in ccid3_hc_tx_init (per [RFC 3448, 4.2], t_RTO is
   undefined until feedback has been received);

 * makes some trivial changes (updates of comments);

 * performs a small optimisation by exploiting that the feedback timeout
   uses the value of t_ipi. The way it is done is safe, because the timeouts
   appear after the changes to t_ipi, ensuring that up-to-date values are used;

 * in ccid3_hc_tx_packet_recv, moves the t_rto statement closer to the calculation
   of the next_tmout. This makes the code clearer to read and is also safe, since
   t_rto is not updated until the next call of ccid3_hc_tx_packet_recv, and is not
   read by the functions called via ccid_wait_for_ccid();

 * removes a `max' statement in sk_reset_timer, this is not needed since the timeout
   value is always greater than 1E6 microseconds.

 * adds `XXX'es to highlight that currently the nofeedback timer is set
   in a non-standard way

Signed-off-by: Gerrit Renker <gerrit@erg.abdn.ac.uk>
Acked-by: Ian McDonald <ian.mcdonald@jandi.co.nz>
Signed-off-by: Arnaldo Carvalho de Melo <acme@mandriva.com>

[DCCP] ccid3: Consistently update t_nom, t_ipi, t_delta

This patch:

 * consolidates updating of parameters (t_nom, t_ipi, t_delta) which
   need to be updated at the same time, since they are inter-dependent

 * removes two inline functions which are no longer needed as a result of
   the above consolidation

 * resolves a FIXME regarding the re-calculation of t_ipi within the nofeedback
   timer, in the state where no feedback has previously been received

 * ties updating these parameters to updating the sending rate X, exploiting
   that all three parameters in turn depend on X; and using a small optimisation
   which can reduce the number of required instructions: only update the three
   parameters when X really changes

Signed-off-by: Gerrit Renker <gerrit@erg.abdn.ac.uk>
Acked-by: Ian McDonald <ian.mcdonald@jandi.co.nz>
Signed-off-by: Arnaldo Carvalho de Melo <acme@mandriva.com>

[DCCP] ccid3: Consolidate timer resets

This patch concerns updating the value of the nofeedback timer when no feedback
has been received so far.

Since in this case the value of R is still undefined according to [RFC 3448,
4.2], we can not perform step (3) of [RFC 3448, 4.3].  A clarification is
provided in [RFC 4342, sec. 5], which states that in these cases the nofeedback
timer (still) expires "after two seconds".

Many thanks to Ian McDonald for pointing this out and providing the
clarification.

The patch
  * implements [RFC 4342, sec. 5] with regard to the above case
  * consolidates handling timer restart by
- adding an appropriate jump label and
- initialising the timeout value

Acked-by: Ian McDonald <ian.mcdonald@jandi.co.nz>
Signed-off-by: Gerrit Renker <gerrit@erg.abdn.ac.uk>
Signed-off-by: Arnaldo Carvalho de Melo <acme@mandriva.com>

[XFRM]: Convert a few __u8 to proper u8

Caught by the EyeBalls(tm) of Thomas Graf

Signed-off-by: Jamal Hadi Salim <hadi@cyberus.ca>
Signed-off-by: David S. Miller <davem@davemloft.net>

[XFRM]: Make flush notifier prettier when subpolicy used

Might as well make flush notifier prettier when subpolicy used

Signed-off-by: Jamal Hadi Salim <hadi@cyberus.ca>
Signed-off-by: David S. Miller <davem@davemloft.net>

[XFRM]: Pack struct xfrm_policy

[acme@newtoy net-2.6.20]$ pahole net/ipv4/tcp.o xfrm_policy
/* /pub/scm/linux/kernel/git/acme/net-2.6.20/include/linux/security.h:67 */
struct xfrm_policy {
        struct xfrm_policy *       next;                 /*     0     4 */
        struct hlist_node          bydst;                /*     4     8 */
        struct hlist_node          byidx;                /*    12     8 */
        rwlock_t                   lock;                 /*    20    36 */
        atomic_t                   refcnt;               /*    56     4 */
        struct timer_list          timer;                /*    60    24 */
        u8                         type;                 /*    84     1 */

        /* XXX 3 bytes hole, try to pack */

        u32                        priority;             /*    88     4 */
        u32                        index;                /*    92     4 */
        struct xfrm_selector       selector;             /*    96    56 */
        struct xfrm_lifetime_cfg   lft;                  /*   152    64 */
        struct xfrm_lifetime_cur   curlft;               /*   216    32 */
        struct dst_entry *         bundles;              /*   248     4 */
        __u16                      family;               /*   252     2 */
        __u8                       action;               /*   254     1 */
        __u8                       flags;                /*   255     1 */
        __u8                       dead;                 /*   256     1 */
        __u8                       xfrm_nr;              /*   257     1 */

        /* XXX 2 bytes hole, try to pack */

        struct xfrm_sec_ctx *      security;             /*   260     4 */
        struct xfrm_tmpl           xfrm_vec[6];          /*   264   360 */
}; /* size: 624, sum members: 619, holes: 2, sum holes: 5 */

So lets have just one hole instead of two, by moving 'type' to just before 'action',
end result:

[acme@newtoy net-2.6.20]$ codiff -s /tmp/tcp.o.before net/ipv4/tcp.o
/pub/scm/linux/kernel/git/acme/net-2.6.20/net/ipv4/tcp.c:
  struct xfrm_policy |   -4
 1 struct changed
[acme@newtoy net-2.6.20]$

[acme@newtoy net-2.6.20]$ pahole -c 64 net/ipv4/tcp.o xfrm_policy
/* /pub/scm/linux/kernel/git/acme/net-2.6.20/include/linux/security.h:67 */
struct xfrm_policy {
        struct xfrm_policy *       next;                 /*     0     4 */
        struct hlist_node          bydst;                /*     4     8 */
        struct hlist_node          byidx;                /*    12     8 */
        rwlock_t                   lock;                 /*    20    36 */
        atomic_t                   refcnt;               /*    56     4 */
        struct timer_list          timer;                /*    60    24 */
        u32                        priority;             /*    84     4 */
        u32                        index;                /*    88     4 */
        struct xfrm_selector       selector;             /*    92    56 */
        struct xfrm_lifetime_cfg   lft;                  /*   148    64 */
        struct xfrm_lifetime_cur   curlft;               /*   212    32 */
        struct dst_entry *         bundles;              /*   244     4 */
        u16                        family;               /*   248     2 */
        u8                         type;                 /*   250     1 */
        u8                         action;               /*   251     1 */
        u8                         flags;                /*   252     1 */
        u8                         dead;                 /*   253     1 */
        u8                         xfrm_nr;              /*   254     1 */

        /* XXX 1 byte hole, try to pack */

        struct xfrm_sec_ctx *      security;             /*   256     4 */
        struct xfrm_tmpl           xfrm_vec[6];          /*   260   360 */
}; /* size: 620, sum members: 619, holes: 1, sum holes: 1 */

Are there any fugly data dependencies here? None that I know.

In the process changed the removed the __ prefixed types, that are just for
userspace visible headers.

Signed-off-by: Arnaldo Carvalho de Melo <acme@mandriva.com>

[NET]: Pack struct hh_cache

[acme@newtoy net-2.6.20]$ pahole net/ipv4/tcp.o hh_cache
/* /pub/scm/linux/kernel/git/acme/net-2.6.20/include/linux/netdevice.h:190 */
struct hh_cache {
        struct hh_cache *          hh_next;              /*     0     4 */
        atomic_t                   hh_refcnt;            /*     4     4 */
        __be16                     hh_type;              /*     8     2 */

        /* XXX 2 bytes hole, try to pack */

        int                        hh_len;               /*    12     4 */
        int                        (*hh_output)();       /*    16     4 */
        rwlock_t                   hh_lock;              /*    20    36 */
        long unsigned int          hh_data[24];          /*    56    96 */
}; /* size: 152, sum members: 150, holes: 1, sum holes: 2 */

[acme@newtoy net-2.6.20]$ find net -name "*.[ch]" | xargs grep 'hh_len.\+=' | sort -u
net/atm/br2684.c:               hh->hh_len = PADLEN + ETH_HLEN;
net/ethernet/eth.c:     hh->hh_len = ETH_HLEN;
net/ipv4/ipconfig.c:    int hh_len = LL_RESERVED_SPACE(dev);
net/ipv4/ip_output.c:   hh_len = LL_RESERVED_SPACE(rt->u.dst.dev);
net/ipv4/ip_output.c:   int hh_len = LL_RESERVED_SPACE(dev);
net/ipv4/netfilter.c:   hh_len = (*pskb)->dst->dev->hard_header_len;
net/ipv4/raw.c: hh_len = LL_RESERVED_SPACE(rt->u.dst.dev);
net/ipv6/ip6_output.c:  hh_len = LL_RESERVED_SPACE(rt->u.dst.dev);
net/ipv6/netfilter/ip6t_REJECT.c:       hh_len = (dst->dev->hard_header_len + 15)&~15;
net/ipv6/raw.c: hh_len = LL_RESERVED_SPACE(rt->u.dst.dev);
[acme@newtoy net-2.6.20]$

[acme@newtoy net-2.6.20]$ find include -name "*.h" | xargs grep 'define ETH_HLEN'
include/linux/if_ether.h:#define ETH_HLEN       14              /* Total octets in header.       */

        (((dev)->hard_header_len&~(HH_DATA_MOD - 1)) + HH_DATA_MOD)

[acme@newtoy net-2.6.20]$ pahole net/ipv4/tcp.o net_device | grep hard_header_len
        short unsigned int         hard_header_len;      /*   106     2 */
[acme@newtoy net-2.6.20]$

So I think we're safe in turning hh_len an u16, end result:

[acme@newtoy net-2.6.20]$ codiff -sV /tmp/tcp.o.before net/ipv4/tcp.o
/pub/scm/linux/kernel/git/acme/net-2.6.20/net/ipv4/tcp.c:
  struct hh_cache |   -4
    hh_len;
     from: int                   /*    12(0)     4(0) */
     to:   u16                   /*    10(0)     2(0) */
 1 struct changed
[acme@newtoy net-2.6.20]$

Signed-off-by: Arnaldo Carvalho de Melo <acme@mandriva.com>

[INET_CONNECTION_SOCK]: Pack struct inet_connection_sock_af_ops

We have a hole in:

[acme@newtoy net-2.6.20]$ pahole net/ipv6/tcp_ipv6.o inet_connection_sock_af_ops
/* /pub/scm/linux/kernel/git/acme/net-2.6.20/include/net/inet_connection_sock.h:38 */
struct inet_connection_sock_af_ops {
        int                        (*queue_xmit)();      /*     0     4 */
        void                       (*send_check)();      /*     4     4 */
        int                        (*rebuild_header)();  /*     8     4 */
        int                        (*conn_request)();    /*    12     4 */
        struct sock *              (*syn_recv_sock)();   /*    16     4 */
        int                        (*remember_stamp)();  /*    20     4 */
        __u16                      net_header_len;       /*    24     2 */

        /* XXX 2 bytes hole, try to pack */

        int                        (*setsockopt)();      /*    28     4 */
        int                        (*getsockopt)();      /*    32     4 */
        int                        (*compat_setsockopt)(); /*    36     4 */
        int                        (*compat_getsockopt)(); /*    40     4 */
        void                       (*addr2sockaddr)();   /*    44     4 */
        int                        sockaddr_len;         /*    48     4 */
}; /* size: 52, sum members: 50, holes: 1, sum holes: 2 */

But we don't need sockaddr_len to be an int:

[acme@newtoy net-2.6.20]$ find net -name "*.[ch]" | xargs grep '\.sockaddr_len.\+=' | sort -u
net/dccp/ipv4.c:        .sockaddr_len      = sizeof(struct sockaddr_in),
net/dccp/ipv6.c:        .sockaddr_len      = sizeof(struct sockaddr_in6),
net/ipv4/tcp_ipv4.c:    .sockaddr_len      = sizeof(struct sockaddr_in),
net/ipv6/tcp_ipv6.c:    .sockaddr_len      = sizeof(struct sockaddr_in6),
net/sctp/ipv6.c:        .sockaddr_len      = sizeof(struct sockaddr_in6),
net/sctp/protocol.c:    .sockaddr_len      = sizeof(struct sockaddr_in),

[acme@newtoy net-2.6.20]$ pahole --sizes net/ipv6/tcp_ipv6.o | grep sockaddr_in
struct sockaddr_in: 16 0
struct sockaddr_in6: 28 0
[acme@newtoy net-2.6.20]$

So I turned sockaddr_len a 'u16', and now:

[acme@newtoy net-2.6.20]$ pahole net/ipv6/tcp_ipv6.o inet_connection_sock_af_ops
/* /pub/scm/linux/kernel/git/acme/net-2.6.20/include/net/inet_connection_sock.h:38 */
struct inet_connection_sock_af_ops {
        int            (*queue_xmit)();        /*     0   4 */
        void           (*send_check)();        /*     4   4 */
        int            (*rebuild_header)();    /*     8   4 */
        int            (*conn_request)();      /*    12   4 */
        struct sock *  (*syn_recv_sock)();     /*    16   4 */
        int            (*remember_stamp)();    /*    20   4 */
        u16            net_header_len;         /*    24   2 */
        u16            sockaddr_len;           /*    26   2 */
        int            (*setsockopt)();        /*    28   4 */
        int            (*getsockopt)();        /*    32   4 */
        int            (*compat_setsockopt)(); /*    36   4 */
        int            (*compat_getsockopt)(); /*    40   4 */
        void           (*addr2sockaddr)();     /*    44   4 */
}; /* size: 48 */

So we've saved 4 bytes:

[acme@newtoy net-2.6.20]$ codiff -sV /tmp/tcp_ipv6.o.before net/ipv6/tcp_ipv6.o
/pub/scm/linux/kernel/git/acme/net-2.6.20/net/ipv6/tcp_ipv6.c:
  struct inet_connection_sock_af_ops |   -4
    net_header_len;
     from: __u16                 /*    24(0)     2(0) */
     to:   u16                   /*    24(0)     2(0) */
    sockaddr_len;
     from: int                   /*    48(0)     4(0) */
     to:   u16                   /*    26(0)     2(0) */
 1 struct changed
[acme@newtoy net-2.6.20]$

Signed-off-by: Arnaldo Carvalho de Melo <acme@mandriva.com>

[UDP(-Lite)]: consolidate v4 and v6 get|setsockopt code

This patch consolidates set/getsockopt code between UDP(-Lite) v4 and 6. The
justification is that UDP(-Lite) is a transport-layer protocol and therefore
the socket option code (at least in theory) should be AF-independent.

Furthermore, there is the following code reduplication:
 * do_udp{,v6}_getsockopt is 100% identical between v4 and v6
 * do_udp{,v6}_setsockopt is identical up to the following differerence
--v4 in contrast to v4 additionally allows the experimental encapsulation
          types  UDP_ENCAP_ESPINUDP and UDP_ENCAP_ESPINUDP_NON_IKE
--the remainder is identical between v4 and v6
   I believe that this difference is of little relevance.

The advantages in not duplicating twice almost completely identical code.

The patch further simplifies the interface of udp{,v6}_push_pending_frames,
since for the second argument (struct udp_sock *up) it always holds that
up = udp_sk(sk); where sk is the first function argument.

Signed-off-by: Gerrit Renker <gerrit@erg.abdn.ac.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>

[RTNETLINK]: Add rtnl_put_cacheinfo() to unify some code

IPv4, IPv6, and DECNet all use struct rta_cacheinfo in a similiar
way, therefore rtnl_put_cacheinfo() is added to reuse code.

Signed-off-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NETLINK]: Remove unused dst_pid field in netlink_skb_parms

The destination PID is passed directly to netlink_unicast()
respectively netlink_multicast().

Signed-off-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NET]: Add documentation for TFRC structures

This adds documentation for the TFRC structure fields.

Signed-off-by: Gerrit Renker <gerrit@erg.abdn.ac.uk>
Signed-off-by: Ian McDonald <ian.mcdonald@jandi.co.nz>
Signed-off-by: Arnaldo Carvalho de Melo <acme@mandriva.com>

[DCCP] ccid3: Resolve small FIXME

This considers the  case - ACK received while no packet has been sent
so far. Resolved by printing a (rate-limited) warning message.

Further removes an unnecessary BUG_ON in ccid3_hc_tx_packet_recv,
received feedback on a terminating connection is simply ignored.

Signed-off-by: Gerrit Renker <gerrit@erg.abdn.ac.uk>
Signed-off-by: Ian McDonald <ian.mcdonald@jandi.co.nz>
Signed-off-by: Arnaldo Carvalho de Melo <acme@mandriva.com>

[DCCP] ccid3: Remove redundant statements in ccid3_hc_tx_packet_sent

This patch removes a switch statement which is redundant since,
 * nothing is done in states TFRC_SSTATE_NO_SENT/TFRC_SSTATE_NO_FBACK
 * it is impossible that the function is called in the state TFRC_SSTATE_TERM, since
       --the function is called, in dccp_write_xmit, after ccid3_hc_tx_send_packet
       --if ccid3_hc_tx_send_packet is called in state TFRC_SSTATE_TERM, it returns
         -EINVAL, which means that ccid3_hc_tx_packet_sent will not be called
 (compare dccp_write_xmit)
       --> therefore, this case is logically impossible
 * the remaining state is TFRC_SSTATE_FBACK which conditionally updates t_ipi, t_nom,
   and t_delta. This is a no-op, since
       --t_ipi only changes when feedback is received
       --however, when feedback arrives via ccid3_hc_tx_packet_recv, there is an identical
         code block which performs the same set of operations
       --performing the same set of operations again in ccid3_hc_tx_packet_sent therefore
         does not change anything, since between the time of receiving the last feedback
 (and therefore update of t_ipi, t_nom, and t_delta), the value of t_ipi has not
 changed
       --since t_ipi has not changed, the values of t_delta and t_nom also do not change,
         they depend fully on t_ipi

Signed-off-by: Gerrit Renker <gerrit@erg.abdn.ac.uk>
Acked-by: Ian McDonald <ian.mcdonald@jandi.co.nz>
Signed-off-by: Arnaldo Carvalho de Melo <acme@mandriva.com>

[DCCP] ccid3: Avoid congestion control on zero-sized data packets

This resolves an `XXX' in ccid3_hc_tx_send_packet().

The function is only called on Data and DataAck packets and returns a negative
result on zero-sized messages. This is a reasonable policy since CCID 3 is a
congestion-control module and congestion control on zero-sized Data(Ack)
packets is in a way pathological.

The patch uses a more suitable error code for this case, it returns the Posix.1
code `EBADMSG' ("Not a data message") instead of `ENOTCONN'.

As a result of ignoring zero-sized packets, a the condition for a warning
"First packet is data" in ccid3_hc_tx_packet_sent is always satisfied; this
message has been removed since it will always be printed.

Signed-off-by: Gerrit Renker <gerrit@erg.abdn.ac.uk>
Signed-off-by: Ian McDonald <ian.mcdonald@jandi.co.nz>
Signed-off-by: Arnaldo Carvalho de Melo <acme@mandriva.com>

[DCCP] ccid3: Simplify control flow of ccid3_hc_tx_send_packet

This makes some logically equivalent simplifications, by replacing
rc - values plus goto's with direct return statements.

Signed-off-by: Gerrit Renker <gerrit@erg.abdn.ac.uk>
Signed-off-by: Ian McDonald <ian.mcdonald@jandi.co.nz>
Signed-off-by: Arnaldo Carvalho de Melo <acme@mandriva.com>

[DCCP] ccid3: Fix calculation of t_ipi time of scheduled transmission

Problem:

[DCCP] ccid3: Simplify control flow in the calculation of t_ipi

This patch performs a simplifying (performance) optimisation:

 In each call of the inline function ccid3_calc_new_t_ipi(), the state is
 tested against TFRC_SSTATE_NO_FBACK. This is expensive when the function
 is called very often. A simpler solution, implemented by this patch, is
 to adapt the control flow.

Background:

[DCCP] ccid3: Fix bug in calculation of first t_nom and first t_ipi

Problem:

[DCCP] ccid2: Allow window to grow larger

Now that we can stuff bigger ack vectors into options.

Signed-off-by: Andrea Bittau <a.bittau@cs.ucl.ac.uk>
Signed-off-by: Arnaldo Carvalho de Melo <acme@mandriva.com>

[DCCP] ackvec: Split long ack vectors across multiple options

Ack vectors grow proportional to the window size.  If an ack vector does not fit
into a single option, it must be spread across multiple options.  This patch
will allow for windows to grow larger.

Committer note: Simplified the patch a bit, original algorithm kept.

Signed-off-by: Andrea Bittau <a.bittau@cs.ucl.ac.uk>
Signed-off-by: Arnaldo Carvalho de Melo <acme@mandriva.com>

[DCCP] ackvec: infrastructure for sending more than one ackvec per packet

Commiter note:

This was split from Andrea's original patch, in the process I changed the type
of the ackvec index fields to u16 instead of to int and haven't folded
dccp_ackvec_parse with dccp_ackvec_check_rcv_ackno.

Next patch will actually do the insertion of more than one ackvec per packet,
using, initially, up to a max of 2 ackvecs as per Andrea's original patch, then
I'll work on support for larger ackvecs, be it using a sysctl or using
setsockopt.

Signed-off-by: Andrea Bittau <a.bittau@cs.ucl.ac.uk>
Signed-off-by: Arnaldo Carvalho de Melo <acme@mandriva.com>

[DCCP] ackvec: Remove unused dccpav_ack_ptr field from dccp_ackvec

Commiter note: original patch was splitted.

Signed-off-by: Andrea Bittau <a.bittau@cs.ucl.ac.uk>
Signed-off-by: Arnaldo Carvalho de Melo <acme@mandriva.com>

[DECNET] address: Convert to new netlink interface

Extends the netlink interface to support the __le16 type and
converts address addition, deletion and, dumping to use the
new netlink interface.

Fixes multiple occasions of possible illegal memory references
due to not validated netlink attributes.

Signed-off-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>

[DECNET] address: Rename rtmsg_ifa() to dn_ifaddr_notify()

The name rtmsg_ifa is heavly overused and confusing.

Signed-off-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>

[DECNET] address: Calculate accurate message size for netlink notifications

Signed-off-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>

[IPV6]: Improve IPv6 tunnel error reporting

Log an error if the remote tunnel endpoint is unable to handle
tunneled packets.

Signed-off-by: Ville Nuorvala <vnuorval@tcs.hut.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>

[IPV6]: Don't allocate memory for Tunnel Encapsulation Limit Option

Signed-off-by: Ville Nuorvala <vnuorval@tcs.hut.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>

[IPV6]: Allow link-local tunnel endpoints

Allow link-local tunnel endpoints if the underlying link is defined.

Signed-off-by: Ville Nuorvala <vnuorval@tcs.hut.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>

[IPV6]: Do mandatory IPv6 tunnel endpoint checks in realtime

Doing the mandatory tunnel endpoint checks when the tunnel is set up
isn't enough as interfaces can go up or down and addresses can be
added or deleted after this. The checks need to be done realtime when
the tunnel is processing a packet.

Signed-off-by: Ville Nuorvala <vnuorval@tcs.hut.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>

[IPV6]: Fix SIOCCHGTUNNEL bug in IPv6 tunnels

A logic bug in tunnel lookup could result in duplicate tunnels when
changing an existing device.

Signed-off-by: Ville Nuorvala <vnuorval@tcs.hut.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>

[GENETLINK] ctrl: Avoid empty CTRL_ATTR_OPS attribute when dumping

Based on Jamal's patch but compiled and even tested. :-)

Signed-off-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>

[XFRM]: Use kmemdup where appropriate

Signed-off-by: Arnaldo Carvalho de Melo <acme@mandriva.com>

[TIPC]: Use kzalloc where appropriate

Signed-off-by: Arnaldo Carvalho de Melo <acme@mandriva.com>

[SUNRPC]: Use k{mem,str}dup where applicable

Signed-off-by: Arnaldo Carvalho de Melo <acme@mandriva.com>

[SCTP]: Use kzalloc where appropriate

Signed-off-by: Arnaldo Carvalho de Melo <acme@mandriva.com>

[SCHED]: Use kmemdup & kzalloc where appropriate

Signed-off-by: Arnaldo Carvalho de Melo <acme@mandriva.com>

[IRDA]: Use kmemdup where applicable

Signed-off-by: Arnaldo Carvalho de Melo <acme@mandriva.com>

[IPVS]: Use kmemdup where appropriate

Signed-off-by: Arnaldo Carvalho de Melo <acme@mandriva.com>

[DECNET]: Use kzalloc where applicable

Signed-off-by: Arnaldo Carvalho de Melo <acme@mandriva.com>

[NET] neighbour: Use kmemdup where applicable

Signed-off-by: Arnaldo Carvalho de Melo <acme@mandriva.com>

[ATM]: Use kmemdup & kzalloc where applicable

Signed-off-by: Arnaldo Carvalho de Melo <acme@mandriva.com>

[CIPSO]: Missing annotation in cipso_ipv4 update.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NETPOLL]: Another udp checksum mangling.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NET]: net/sched annotations.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NET]: Split skb->csum

... into anonymous union of __wsum and __u32 (csum and csum_offset resp.)

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NETFILTER]: ip_nat_snmp_basic annotations.

... and switch the damn checksum update to something saner

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NET]: Fix assorted misannotations (from md5 and udplite merges).

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>

[SCTP]: Even more trivial sctp annotations.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>

[SCTP]: SCTP_CMD_PROCESS_CTSN annotations.

argument passed as __be32

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>

[SCTP]: More trivial sctp annotations.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>

[SCTP]: Netfilter sctp annotations.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>

[SCTP]: Annotate ->supported_addrs().

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>

[SCTP]: sctp_association ->peer.i is a host-endian analog of sctp_inthdr.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>

[SCTP]: Get rid of the last remnants of sin_port flipping.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>

[SCTP]: Annotate ->inaddr_any().

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>

[SCTP]: flip_to_{h,n}() are not needed anymore.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>

[SCTP]: Don't bother setting sin_port in ->from_sk().

... the only caller will overwrite immediately

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>

[SCTP]: ->a_h is gone now.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>