[PPPOE]: memory leak when socket is release()d before PPPIOCGCHAN has been called on it

below you find a patch that fixes a memory leak when a PPPoE socket is
release()d after it has been connect()ed, but before the PPPIOCGCHAN ioctl
ever has been called on it.

This is somewhat of a security problem, too, since PPPoE sockets can be
created by any user, so any user can easily allocate all the machine's
RAM to non-swappable address space and thus DoS the system.

Is there any specific reason for PPPoE sockets being available to any
unprivileged process, BTW? After all, you need a packet socket for the
discovery stage anyway, so it's unlikely that any unprivileged process
will ever need to create a PPPoE socket, no? Allocating all session IDs
for a known AC is a kind of DoS, too, after all - with Juniper ERXes,
this is really easy, actually, since they don't ever assign session ids
above 8000 ...

Signed-off-by: Florian Zumbiehl <florz@florz.de>
Acked-by: Michal Ostrowski <mostrows@earthlink.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

[PPPOE]: race between interface going down and connect()

below you find a patch that (hopefully) fixes a race between an interface
going down and a connect() to a peer on that interface. Before,
connect() would determine that an interface is up, then the interface
could go down and all entries referring to that interface in the
item_hash_table would be marked as ZOMBIEs and their references to
the device would be freed, and after that, connect() would put a new
entry into the hash table referring to the device that meanwhile is
down already - which also would cause unregister_netdevice() to wait
until the socket has been release()d.

This patch does not suffice if we are not allowed to accept connect()s
referring to a device that we already acked a NETDEV_GOING_DOWN for
(that is: all references are only guaranteed to be freed after
NETDEV_DOWN has been acknowledged, not necessarily after the
NETDEV_GOING_DOWN already). And if we are allowed to, we could avoid
looking through the hash table upon NETDEV_GOING_DOWN completely and
only do that once we get the NETDEV_DOWN ...

mostrows:
pppoe_flush_dev is called on NETDEV_GOING_DOWN and NETDEV_DOWN to deal with
this "late connect" issue.  Ideally one would hope to notify users at the
"NETDEV_GOING_DOWN" phase (just to pretend to be nice).  However, it is the
NETDEV_DOWN scan that takes all the responsibility for ensuring nobody is
hanging around at that time.

Signed-off-by: Florian Zumbiehl <florz@florz.de>
Acked-by: Michal Ostrowski <mostrows@earthlink.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

[PPPoE]: miscellaneous smaller cleanups

below is a patch that just removes dead code/initializers without any
effect (first access is an assignment) that I stumbled accross while
reading the source.

Signed-off-by: Florian Zumbiehl <florz@florz.de>
Acked-by: Michal Ostrowski <mostrows@earthlink.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NET] skbuff: skb_store_bits const is backwards

Getting warnings becuase skb_store_bits has skb as constant,
but the function overwrites it. Looks like const was on the
wrong side.

Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

[BRIDGE]: Fix warning in net-2.6.22

The following is leftover from earlier change in net-2.6.22.

Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

[AX25/NETROM/ROSE]: Convert to use modern wait queue API

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

[AF_PACKET]: Add option to return orig_dev to userspace.

Add a packet socket option to allow the orig_dev index to be returned
to userspace when passing traffic through a decapsulated device, such
as the bonding driver.

This is very useful for layer 2 traffic being able to report which
physical device actually received the traffic, instead of having the
encapsulating device hide that information.

The new option is called PACKET_ORIGDEV.

Signed-off-by: Peter P. Waskiewicz Jr. <peter.p.waskiewicz.jr@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

[IPV6] SNMP: Export statistics via netlink without CONFIG_PROC_FS.

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

[IPV4] SNMP: Move some statistic bits to net/ipv4/proc.c.

This also fixes memory leak in error path.

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

[IPV6] SNMP: Move some statistic bits to net/ipv6/proc.c.

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

[IPV6] SNMP: Netlink interface.

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

[INET]: Add IP(V6)_PMTUDISC_RPOBE

Add IP(V6)_PMTUDISC_PROBE value for IP(V6)_MTU_DISCOVER.  This option forces
us not to fragment, but does not make use of the kernel path MTU discovery.
That is, it allows for user-mode MTU probing (or, packetization-layer path
MTU discovery).  This is particularly useful for diagnostic utilities, like
traceroute/tracepath.

Signed-off-by: John Heffner <jheffner@psc.edu>
Signed-off-by: David S. Miller <davem@davemloft.net>

[IPV6]: MTU discovery check in ip6_fragment()

Adds a check in ip6_fragment() mirroring ip_fragment() for packets
that we can't fragment, and sends an ICMP Packet Too Big message
in response.

Signed-off-by: John Heffner <jheffner@psc.edu>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NET_SCHED]: ingress: switch back to using ingress_lock

Switch ingress queueing back to use ingress_lock. qdisc_lock_tree now locks
both the ingress and egress qdiscs on the device. All changes to data that
might be used on both ingress and egress needs to be protected by using
qdisc_lock_tree instead of manually taking dev->queue_lock. Additionally
the qdisc stats_lock needs to be initialized to ingress_lock for ingress
qdiscs.

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

[NET_SCHED]: Eliminate qdisc_tree_lock

Since we're now holding the rtnl during the entire dump operation, we
can remove qdisc_tree_lock, whose only purpose is to protect dump
callbacks from concurrent changes to the qdisc tree.

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

[NETLINK]: don't reinitialize callback mutex

Don't reinitialize the callback mutex the netlink_kernel_create caller
handed in, it is supposed to already be initialized and could already
be held by someone.

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

[RTNETLINK]: Remove unnecessary locking in dump callbacks

Since we're now holding the rtnl during the entire dump operation, we can
remove additional locking for rtnl protected data. This patch does that
for all simple cases (dev_base_lock for dev_base walking, RCU protection
for FIB rule dumping).

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

[RTNETLINK]: Hold rtnl_mutex during netlink dump callbacks

Hold rtnl_mutex during the entire netlink dump operation. This allows
to simplify locking in the dump callbacks, since they can now rely on
that no concurrent changes happen.

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

[NETLINK]: Switch cb_lock spinlock to mutex and allow to override it

Switch cb_lock to mutex and allow netlink kernel users to override it
with a subsystem specific mutex for consistent locking in dump callbacks.
All netlink_dump_start users have been audited not to rely on any
side-effects of the previously used spinlock.

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

[NETFILTER]: ipt_ULOG: add compat conversion functions

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

[NETFILTER]: nfnetlink_log: remove fallback to group 0

Don't fallback to group 0 if no instance can be found for the given group.
This potentially confuses the listener and is not what the user configured.
Also remove the ring buffer spamming that happens when rules are set up
before the logging daemon is started.

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

[NETFILTER]: {eb,ip6,ip}t_LOG: remove remains of LOG target overloading

All LOG targets always use their internal logging function nowadays, so
remove the incorrect error message and handle real errors (!= -EEXIST)
by failing to load.

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

[NETFILTER]: nf_nat: use HW checksumming when possible

When mangling packets forwarded to a HW checksumming capable device,
offload recalculation of the checksum instead of doing it in software.

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

[NETFILTER]: ebt_arp: add gratuitous arp filtering

The attached patch adds gratuitous arp filtering, more precisely: it
allows checking that the IPv4 source address matches the IPv4
destination address inside the ARP header. It also adds a check for the
hardware address type when matching MAC addresses (nothing critical,
just for better consistency).

Signed-off-by: Bart De Schuymer <bdschuym@pandora.be>
Acked-by: Carl-Daniel Hailfinger <c-d.hailfinger.devel.2006@gmx.net>
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NETFILTER]: bridge-nf: filter bridged IPv4/IPv6 encapsulated in pppoe traffic

The attached patch by Michael Milner adds support for using iptables and
ip6tables on bridged traffic encapsulated in ppoe frames, similar to
what's already supported for vlan.

Signed-off-by: Michael Milner <milner@blissisland.ca>
Signed-off-by: Bart De Schuymer <bdschuym@pandora.be>
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

[DCCP]: Complete documentation of dccp_sock

This fills in missing documentation for dccp_sock 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@ghostprotocols.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

[DCCP]: Debug statements for Elapsed Time option

This prints the value of the parsed Elapsed Time when received via a
Timestamp Echo option [RFC 4342, 13.3].

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

[DCCP]: Fix bug in the calculation of very low sending rates

This fixes an error in the calculation of t_ipi when X converges towards
very low sending rates (between 1 and 64 bytes per second).

Although this case may not sound likely, it can be reproduced by connecting,
hitting enter (1 byte sent) and waiting for some time, during which the
nofeedback timer halves the sending rate until finally it reaches the region
1..64 bytes/sec. Computing X is handled correctly (tested separately); but by
dividing X _before_ entering the calculation of t_ipi, X becomes zero as
a result.  This in turn triggers a BUG condition caught in scaled_div().

Fixed by replacing with equivalent statement and explicit typecast for good
measure.

Calculation verified and effect of patch tested - reduced never below 1 byte
per 64 seconds afterwards, i.e. not allowing divide-by-zero.

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

[S390]: Fix build on 31-bit.

Allow s390 to properly override the generic
__div64_32() implementation by:

1) Using obj-y for div64.o in s390's makefile instead
   of lib-y

2) Adding the weak attribute to the generic implementation.

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

[SK_BUFF]: Fix missing offset adjustment in skb_copy_expand

skb_copy_expand changes the headroom, so it needs to adjust the header
offsets by the difference between the old and the new value.

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

[NET]: loopback driver can use loopback_dev integrated net_device_stats

Rusty added a new 'stats' field to struct net_device.

loopback driver can use it instead of declaring another struct
net_device_stats This saves some memory.

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

bridge: check kmem_cache_create() error

This patch checks kmem_cache_create() error and aborts loading module
on failure.

Signed-off-by: Akinobu Mita <akinobu.mita@gmail.com>
Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>

bridge: allow changing hardware address to any valid address

For case of bridging pseudo devices, the get created/destroyed (Xen)
need to allow setting address to any valid value.

Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>

bridge: change when netlink events go to STP

Need to tell STP daemon about more events, like any time a
device is added even when it is down.

Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>

bridge: add support for user mode STP

This patchset based on work by Aji_Srinivas@emc.com provides allows
spanning tree to be controled from userspace.  Like hotplug, it
uses call_usermodehelper when spanning tree is enabled so there
is no visible API change. If call to start usermode STP fails
it falls back to existing kernel STP.

Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>

bridge: add sysfs hook to flush forwarding table

The RSTP daemon needs to be able to flush all dynamic forwarding
entries in the case of topology change.

This is a temporary interface. It will change to a netlink interface
before RSTP daemon is officially released.

Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>

bridge: simpler hash with salt

Instead of hashing the whole Ethernet address, it should be faster
to just use the last 4 bytes. Add a random salt value to the hash
to make it more difficult to construct worst case DoS hash chains.

Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>

bridge: don't route packets while learning

While in the STP learning state, don't route packets; wait until
forwarding delay has expired. The purpose of the forwarding delay
is to detect loops in the network, and if a brouter started up
and started forwarding, it could cause a flood.

Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>

bridge: eliminate call by reference

Change the bridging hook to be simple function with return value
rather than modifying the skb argument. This could generate better
code and is cleaner.

Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>

[NET]: Treat CHECKSUM_PARTIAL as CHECKSUM_UNNECESSARY

When a transmitted packet is looped back directly, CHECKSUM_PARTIAL
maps to the semantics of CHECKSUM_UNNECESSARY.  Therefore we should
treat it as such in the stack.

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

[NETDRV]: Perform missing csum_offset conversions

When csum_offset was introduced we did a conversion from csum to
csum_offset where applicable.  A couple of drivers were missed in
this process.

It was harmless to begin with since the two fields coincided.  Now
that we've made them different with the addition of csum_start, the
missed drivers must be converted or they can't send packets out at
all that require checksum offload.

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

[NET]: Use csum_start offset instead of skb_transport_header

The skb transport pointer is currently used to specify the start
of the checksum region for transmit checksum offload.  Unfortunately,
the same pointer is also used during receive side processing.

This creates a problem when we want to retransmit a received
packet with partial checksums since the skb transport pointer
would be overwritten.

This patch solves this problem by creating a new 16-bit csum_start
offset value to replace the skb transport header for the purpose
of checksums.  This offset is calculated from skb->head so that
it does not have to change when skb->data changes.

No extra space is required since csum_offset itself fits within
a 16-bit word so we can use the other 16 bits for csum_start.

For backwards compatibility, just before we push a packet with
partial checksums off into the device driver, we set the skb
transport header to what it would have been under the old scheme.

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

[XFRM]: beet: fix worst case header_len calculation

esp_init_state doesn't account for the beet pseudo header in the header_len
calculation, which may result in undersized skbs hitting xfrm4_beet_output,
causing unnecessary reallocations in ip_finish_output2.

The skbs should still always have enough room to avoid causing
skb_under_panic in skb_push since we have at least 16 bytes available
from LL_RESERVED_SPACE in xfrm_state_check_space.

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

[XFRM]: Optimize MTU calculation

Replace the probing based MTU estimation, which usually takes 2-3 iterations
to find a fitting value and may underestimate the MTU, by an exact calculation.

Also fix underestimation of the XFRM trailer_len, which causes unnecessary
reallocations.

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

[XFRM]: esp: fix skb_tail_pointer conversion bug

Fix incorrect switch of "trailer" skb by "skb" during skb_tail_pointer
conversion:

-       *(u8*)(trailer->tail - 1) = top_iph->protocol;
+       *(skb_tail_pointer(skb) - 1) = top_iph->protocol;

-       *(u8 *)(trailer->tail - 1) = *skb_network_header(skb);
+       *(skb_tail_pointer(skb) - 1) = *skb_network_header(skb);

Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

[SK_BUFF]: Fix missing offset adjustment in pskb_expand_head

Since we're increasing the headroom, the header offsets need to be
increased by the same amount as well.

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

[IPV6] FIB6RULE: Find source address during looking up route.

When looking up route for destination with rules with
source address restrictions, we may need to find a source
address for the traffic if not given.

Based on patch from Noriaki TAKAMIYA <takamiya@po.ntts.co.jp>.

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

[XFRM]: beet: minor cleanups

Remove unnecessary initialization/variable.

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

[RTNL]: Improve error codes for unsupported operations

The most common trigger of these errors is that the
config option hasn't been enable wich would make the
functionality available. Therefore returning EOPNOTSUPP
gives a better idea on what is going wrong.

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

[NET]: Move generic skbuff stuff from XFRM code to generic code

Move generic skbuff stuff from XFRM code to generic code so that
AF_RXRPC can use it too.

The kdoc comments I've attached to the functions needs to be checked
by whoever wrote them as I had to make some guesses about the workings
of these functions.

Signed-off-By: David Howells <dhowells@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

[CREDITS]: Update Arnaldo entry

Signed-off-by: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>

[SK_BUFF]: Some more conversions to skb_copy_from_linear_data

Signed-off-by: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>

[SK_BUFF]: Introduce skb_copy_to_linear_data{_offset}

To clearly state the intent of copying to linear sk_buffs, _offset being a
overly long variant but interesting for the sake of saving some bytes.

Signed-off-by: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>

[NET]: Fix warnings in 3c523.c and ni52.c

We have to put back the cast to "char *" because these
pointers are volatile.

Reported by Andrew Morton.

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

[NET]: Inline net_device_stats

Network drivers which keep stats allocate their own stats structure
then write a get_stats() function to return them.  It would be nice if
this were done by default.

1) Add a new "stats" field to "struct net_device".
2) Add a new feature field to say "this driver uses the internal one"
3) Have a default "get_stats" which returns NULL if that feature not set.
4) Change callers to check result of get_stats call for NULL, not if
   ->get_stats is set.

This should not break backwards compatibility with older drivers, yet
allow modern drivers to shed some boilerplate code.

Lightly tested: works for a modified lguest network driver.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NET]: random functions can use nsec resolution instead of usec

In order to get more randomness for secure_tcpv6_sequence_number(),
secure_tcp_sequence_number(), secure_dccp_sequence_number() functions,
we can use the high resolution time services, providing nanosec
resolution.

I've also done two kmalloc()/kzalloc() conversions.

Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Acked-by: James Morris <jmorris@namei.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NET] fib_rules: delay route cache flush by ip_rt_min_delay

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

[SK_BUFF]: Introduce skb_copy_from_linear_data{_offset}

To clearly state the intent of copying from linear sk_buffs, _offset being a
overly long variant but interesting for the sake of saving some bytes.

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

[BLUETOOTH]: Introduce skb->data accessor methods for hci_{acl,event,sco}_hdr

For consistency with other skb data accessors, reducing the number of direct
accesses to skb->data.

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

[IPV4]: align inet_protos[] on SMP

As IPPROTO_TCP is 6, it makes sense to make sure inet_protos[] array
is properly cache line aligned to avoid false sharing on SMP.

c0680540 b peer_total
c0680544 b inet_peer_unused_head
c0680560 B inet_protos

On i386 this example, we can see that inet_protos[IPPROTO_TCP] shares
a potentially hot (and modified) cache line.

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

[TCP]: tcp_memory_pressure and tcp_socket are__read_mostly candidates

tcp_memory_pressure and tcp_socket currently share a cache line with tcp_memory_allocated, tcp_sockets_allocated.
(Very hot cache line)
It makes sense to declare these variables as __read_mostly, to avoid false sharing on SMP.

ffffffff8081d9c0 B tcp_orphan_count
ffffffff8081d9c4 B tcp_memory_allocated
ffffffff8081d9c8 B tcp_sockets_allocated
ffffffff8081d9cc B tcp_memory_pressure
ffffffff8081d9d0 b tcp_md5sig_users
ffffffff8081d9d8 b tcp_md5sig_pool
ffffffff8081d9e0 b warntime.31570
ffffffff8081d9e8 b tcp_socket

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

[NET] fib_rules: Flush route cache after rule modifications

The results of FIB rules lookups are cached in the routing cache
except for IPv6 as no such cache exists. So far, it was the
responsibility of the user to flush the cache after modifying any
rules. This lead to many false bug reports due to misunderstanding
of this concept.

This patch automatically flushes the route cache after inserting
or deleting a rule.

Thanks to Muli Ben-Yehuda <muli@il.ibm.com> for catching a bug
in the previous patch.

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

[NET]: inet_ehash_secret should be __read_mostly and set only once

There is a very tiny probability that build_ehash_secret() is called
at the same time by different CPUS.

Also, using __read_mostly is a must for inet_ehash_secret

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

[NET]: Allow forwarding of ip_summed except CHECKSUM_COMPLETE

Right now Xen has a horrible hack that lets it forward packets with
partial checksums.  One of the reasons that CHECKSUM_PARTIAL and
CHECKSUM_COMPLETE were added is so that we can get rid of this hack
(where it creates two extra bits in the skbuff to essentially mirror
ip_summed without being destroyed by the forwarding code).

I had forgotten that I've already gone through all the deivce drivers
last time around to make sure that they're looking at ip_summed ==
CHECKSUM_PARTIAL rather than ip_summed != 0 on transmit.  In any case,
I've now done that again so it should definitely be safe.

Unfortunately nobody has yet added any code to update CHECKSUM_COMPLETE
values on forward so we I'm setting that to CHECKSUM_NONE.  This should
be safe to remove for bridging but I'd like to check that code path
first.

So here is the patch that lets us get rid of the hack by preserving
ip_summed (mostly) on forwarded packets.

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

[IPV4] LVS: Allow to send ICMP unreachable responses when real-servers are removed

this is a small patch by Janusz Krzysztofik to ip_route_output_slow()
that allows VIP-less LVS linux director to generate packets
originating >From VIP if sysctl_ip_nonlocal_bind is set.

In a nutshell, the intention is for an LVS linux director to be able
to send ICMP unreachable responses to end-users when real-servers are
removed.

http://archive.linuxvirtualserver.org/html/lvs-users/2007-01/msg00106.html

Signed-off-by: Simon Horman <horms@verge.net.au>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NET] fib_rules: Add no-operation action

The use of nop rules simplifies the usage of goto rules
and adds more flexibility as they allow targets to remain
while the actual content of the branches can change easly.

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

[NET] fib_rules: Mark rules detached from the device

Rules which match against device names in their selector can
remain while the device itself disappears, in fact the device
doesn't have to present when the rule is added in the first
place. The device name is resolved by trying when the rule is
added and later by listening to NETDEV_REGISTER/UNREGISTER
notifications.

This patch adds the flag FIB_RULE_DEV_DETACHED which is set
towards userspace when a rule contains a device match which
is unresolved at the moment. This eases spotting the reason
why certain rules seem not to function properly.

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

[NET] fib_rules: goto rule action

This patch adds a new rule action FR_ACT_GOTO which allows
to skip a set of rules by jumping to another rule. The rule
to jump to is specified via the FRA_GOTO attribute which
carries a rule preference.

Referring to a rule which doesn't exists is explicitely allowed.
Such goto rules are marked with the flag FIB_RULE_UNRESOLVED
and will act like a rule with a non-matching selector. The rule
will become functional as soon as its target is present.

The goto action enables performance optimizations by reducing
the average number of rules that have to be passed per lookup.

Example:
0:      from all lookup local
40:     not from all to 192.168.23.128 goto 32766
41:     from all fwmark 0xa blackhole
42:     from all fwmark 0xff blackhole
32766:  from all lookup main

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

[WAN] cosa.c: Build fix.

Caused by skb_reset_mac_header() changes, missing semicolon.

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

[TCP] tcp_probe: improvements for net-2.6.22

Change tcp_probe to use ktime (needed to add one export).
Add option to only get events when cwnd changes - from Doug Leith

Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

[TCP]: cubic update for net-2.6.22

The following update received from Injong updates TCP cubic to the latest
version. I am running more complete tests and will have results after 4/1.

According to Injong: the new version improves on its scalability,
fairness and stability.  So in all properties, we confirmed it shows better
performance.

NCSU results (for 2.6.18 and 2.6.20) available:
http://netsrv.csc.ncsu.edu/wiki/index.php/TCP_Testing

This version is described in a new Internet draft for CUBIC.
http://www.ietf.org/internet-drafts/draft-rhee-tcp-cubic-00.txt

Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NET] Move DF check to ip_forward

Do fragmentation check in ip_forward, similar to ipv6 forwarding.

Signed-off-by: John Heffner <jheffner@psc.edu>
Signed-off-by: David S. Miller <davem@davemloft.net>

[INET]: Use jhash + random secret for ehash.

The days are gone when this was not an issue, there are folks out
there with huge bot networks that can be used to attack the
established hash tables on remote systems.

So just like the routing cache and connection tracking
hash, use Jenkins hash with random secret input.

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

[NETLINK]: introduce NLA_BINARY type

This patch introduces a new NLA_BINARY attribute policy type with the
verification of simply checking the maximum length of the payload.

It also fixes a small typo in the example.

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

[SCTP]: Implement SCTP_MAX_BURST socket option.

Signed-off-by: Vlad Yasevich <vladislav.yasevich@hp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

[SCTP]: Implement sac_info field in SCTP_ASSOC_CHANGE notification.

As stated in the sctp socket api draft:

   sac_info: variable

   If the sac_state is SCTP_COMM_LOST and an ABORT chunk was received
   for this association, sac_info[] contains the complete ABORT chunk as
   defined in the SCTP specification RFC2960 [RFC2960] section 3.3.7.

We now save received ABORT chunks into the sac_info field and pass that
to the user.

Signed-off-by: Vlad Yasevich <vladislav.yasevich@hp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

[SCTP]: Honor flags when setting peer address parameters

Parameters only take effect when a corresponding flag bit is set
and a value is specified. This means we need to check the flags
in addition to checking for non-zero value.

Signed-off-by: Vlad Yasevich <vladislav.yasevich@hp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

[SCTP]: Implement SCTP_ADDR_CONFIRMED state for ADDR_CHNAGE event

Signed-off-by: Vlad Yasevich <vladislav.yasevich@hp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

[SCTP]: Implement SCTP_PARTIAL_DELIVERY_POINT option.

This option induces partial delivery to run as soon
as the specified amount of data has been accumulated on
the association.  However, we give preference to fully
reassembled messages over PD messages.  In any case,
window and buffer is freed up.

Signed-off-by: Vlad Yasevich <vladislav.yasevich@.hp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

[SCTP]: Implement SCTP_FRAGMENT_INTERLEAVE socket option

This option was introduced in draft-ietf-tsvwg-sctpsocket-13.  It
prevents head-of-line blocking in the case of one-to-many endpoint.
Applications enabling this option really must enable SCTP_SNDRCV event
so that they would know where the data belongs.  Based on an
earlier patch by Ivan Skytte Jørgensen.

Additionally, this functionality now permits multiple associations
on the same endpoint to enter Partial Delivery.  Applications should
be extra careful, when using this functionality, to track EOR indicators.

Signed-off-by: Vlad Yasevich <vladislav.yasevich@hp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NET_SCHED]: qdisc: remove unnecessary memory barriers

We're holding dev->queue_lock in qdisc_watchdog_schedule and
qdisc_watchdog_cancel, no need for the barriers.

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

[NET_SCHED]: Unline tcf_destroy

Uninline tcf_destroy and add a helper function to destroy an entire filter
chain.

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

[NET_SCHED]: turn PSCHED_GET_TIME into inline function

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

[NET_SCHED]: turn PSCHED_TDIFF_SAFE into inline function

Also rename to psched_tdiff_bounded.

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

[NET_SCHED]: kill PSCHED_TDIFF

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

[NET_SCHED]: kill PSCHED_SET_PASTPERFECT/PSCHED_IS_PASTPERFECT

Use direct assignment and comparison instead.

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

[NET_SCHED]: kill PSCHED_TLESS

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

[NET_SCHED]: kill PSCHED_TADD/PSCHED_TADD2

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

[NET_SCHED]: kill PSCHED_AUDIT_TDIFF

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

[NET_SCHED]: sch_netem: fix off-by-one in send time comparison

netem checks PSCHED_TLESS(cb->time_to_send, now) to find out whether it is
allowed to send a packet, which is equivalent to cb->time_to_send < now.
Use !PSCHED_TLESS(now, cb->time_to_send) instead to properly handle
cb->time_to_send == now.

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

[NETFILTER] nfnetlink: netlink_run_queue() already checks for NLM_F_REQUEST

Patrick has made use of netlink_run_queue() in nfnetlink while my patches
have been waiting for net-2.6.22 to open. So this check for NLM_F_REQUEST
can go as well.

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

[NETFILTER]: nf_conntrack: kill destroy() in struct nf_conntrack for diet

The destructor per conntrack is unnecessary, then this replaces it with
system wide destructor.

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

[NETFILTER]: nf_conntrack: don't use nfct in skb if conntrack is disabled

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

[NETFILTER]: Use setup_timer

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

[NETFILTER]: nfnetlink_log: remove conditional locking

This is gross, have the wrapper function take the lock.

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

[NETFILTER]: nfnetlink_log: micro-optimization: inst->skb != NULL in __nfulnl_send()

No other function calls __nfulnl_send() with inst->skb == NULL than
nfulnl_timer().

Signed-off-by: Michal Miroslaw <mirq-linux@rere.qmqm.pl>
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NETFILTER]: nfnetlink_log: iterator functions need iter_state * only

get_*() don't need access to seq_file - iter_state is enough for them.

Signed-off-by: Michal Miroslaw <mirq-linux@rere.qmqm.pl>
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NETFILTER]: nfnetlink_log: micro-optimization: don't modify destroyed instance

Simple micro-optimization: Don't change any options if the instance is
being destroyed.

Signed-off-by: Michal Miroslaw <mirq-linux@rere.qmqm.pl>
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NETFILTER]: nfnetlink_log: micro-optimization for inst==NULL in nfulnl_recv_config()

Simple micro-optimization: don't call instance_put() on known NULL pointers.

Signed-off-by: Michal Miroslaw <mirq-linux@rere.qmqm.pl>
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

[NETFILTER]: nfnetlink_log: kill duplicate code

Kill some duplicate code in nfulnl_log_packet().

Signed-off-by: Michal Miroslaw <mirq-linux@rere.qmqm.pl>
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>