The local and the global translation-tables are now lock free and rcu
protected.
Signed-off-by: NAntonio Quartulli <ordex@autistici.org>
Acked-by: NSimon Wunderlich <siwu@hrz.tu-chemnitz.de>
Signed-off-by: NSven Eckelmann <sven@narfation.org>

With the current client announcement implementation, in case of roaming,
an update is triggered on the new AP serving the client. At that point
the new information is spread around by means of the OGM broadcasting
mechanism. Until this operations is not executed, no node is able to
correctly route traffic towards the client. This obviously causes packet
drops and introduces a delay in the time needed by the client to recover
its connections.

A new packet type called ROAMING_ADVERTISEMENT is added to account this
issue.

This message is sent in case of roaming from the new AP serving the
client to the old one and will contain the client MAC address. In this
way an out-of-OGM update is immediately committed, so that the old node
can update its global translation table. Traffic reaching this node will
then be redirected to the correct destination utilising the fresher
information. Thus reducing the packet drops and the connection recovery
delay.
Signed-off-by: NAntonio Quartulli <ordex@autistici.org>
Signed-off-by: NSven Eckelmann <sven@narfation.org>

The client announcement mechanism informs every mesh node in the network
of any connected non-mesh client, in order to find the path towards that
client from any given point in the mesh.

The old implementation was based on the simple idea of appending a data
buffer to each OGM containing all the client MAC addresses the node is
serving. All other nodes can populate their global translation tables
(table which links client MAC addresses to node addresses) using this
MAC address buffer and linking it to the node's address contained in the
OGM. A node that wants to contact a client has to lookup the node the
client is connected to and its address in the global translation table.

It is easy to understand that this implementation suffers from several
issues:
 - big overhead (each and every OGM contains the entire list of
   connected clients)
 - high latencies for client route updates due to long OGM trip time and
   OGM losses

The new implementation addresses these issues by appending client
changes (new client joined or a client left) to the OGM instead of
filling it with all the client addresses each time. In this way nodes
can modify their global tables by means of "updates", thus reducing the
overhead within the OGMs.

To keep the entire network in sync each node maintains a translation
table version number (ttvn) and a translation table checksum. These
values are spread with the OGM to allow all the network participants to
determine whether or not they need to update their translation table
information.

When a translation table lookup is performed in order to send a packet
to a client attached to another node, the destination's ttvn is added to
the payload packet. Forwarding nodes can compare the packet's ttvn with
their destination's ttvn (this node could have a fresher information
than the source) and re-route the packet if necessary. This greatly
reduces the packet loss of clients roaming from one AP to the next.
Signed-off-by: NAntonio Quartulli <ordex@autistici.org>
Signed-off-by: NMarek Lindner <lindner_marek@yahoo.de>
Signed-off-by: NSven Eckelmann <sven@narfation.org>

The amount of duplicated code in the receive and routing code can be
reduced when all headers provide the packet type, version and ttl in the
same first bytes.
Signed-off-by: NAntonio Quartulli <ordex@autistici.org>
Signed-off-by: NSven Eckelmann <sven@narfation.org>

char was used in different places to store information without really
using the characteristics of that data type or by ignoring the fact that
char has not a well defined signedness.
Signed-off-by: NSven Eckelmann <sven@narfation.org>

count_real_packets() in batman-adv assumes char is signed, and returns -1
through it:

net/batman-adv/routing.c: In function 'receive_bat_packet':
net/batman-adv/routing.c:739: warning: comparison is always false due to limited range of data type

Use int instead.
Signed-off-by: NDavid Howells <dhowells@redhat.com>
[sven@narfation.org: Rebase on top of current version]
Signed-off-by: NSven Eckelmann <sven@narfation.org>

interface_tx is not used outside of soft-interface.c and thus doesn't
need to be declared inside soft-interface.h
Signed-off-by: NSven Eckelmann <sven@narfation.org>

compare_orig is only used in context of orig_node which is managed
inside originator.c. It is not necessary to keep that function inside
the header originator.h.
Signed-off-by: NSven Eckelmann <sven@narfation.org>

The definition NO_FLAGS was introduced to make the code more
readable and shall be used to initialize flag fields.
Signed-off-by: NMarek Lindner <lindner_marek@yahoo.de>
Signed-off-by: NSven Eckelmann <sven@narfation.org>

CodingStyle "Chapter 12: Macros, Enums and RTL" recommends to use enums
for several related constants. Internal states can be used without
defining the actual value, but all values which are visible to the
outside must be defined as before. Normal values are assigned as usual
and flags are defined by shifts of a bit.
Signed-off-by: NSven Eckelmann <sven@narfation.org>

CodingStyle "Chapter 12: Macros, Enums and RTL" highly recommends to use
functions instead of macros were possible. This ensures type safety and
prevents shadowing of other variables.
Signed-off-by: NSven Eckelmann <sven@narfation.org>

strict_strtoul as used in parse_gw_bandwidth is defined for unsigned
long and strict_strtol should be used instead for long.
Signed-off-by: NSven Eckelmann <sven@narfation.org>

gw_node_delete is defined with "void" as return type, but still tries to
return a value. The called function gw_node_delete is also return as
void and thus doesn't provide a value for us.
Signed-off-by: NSven Eckelmann <sven@narfation.org>

When a link is saturated (re)broadcasts of OGMs are delayed. Under heavy
load this delay may exceed the orig interval which leads to OGMs being
dropped (the code would only accept an OGM rebroadcast if it arrived
before the next OGM was broadcasted). With this patch batman-adv will
also accept delayed OGMs in order to avoid a bogus influence on the
routing metric.
Signed-off-by: NDaniele Furlan <daniele.furlan@gmail.com>
Signed-off-by: NSven Eckelmann <sven@narfation.org>

The debug output of update_route has tests for "route deleted" and "route
added". All other situations are handled as "route changed". This is not
true because neigh_node and curr_router could be both NULL.

The function is not called in this situation, but the code might be
interpreted wrong when reading it without this test.
Signed-off-by: NSven Eckelmann <sven@narfation.org>

This comment has been wrongly put after the variable it refers to and was also bad indented
Signed-off-by: NAntonio Quartulli <ordex@autistici.org>
Acked-by: NSven Eckelmann <sven@narfation.org>
Signed-off-by: NSven Eckelmann <sven@narfation.org>

Instead of comparing mac addresses with the broadcast address by means
of compare_eth(), the is_broadcast_ether_addr() kernel function has to be
used.
Signed-off-by: NAntonio Quartulli <ordex@autistici.org>
Acked-by: NSven Eckelmann <sven@narfation.org>
Signed-off-by: NSven Eckelmann <sven@narfation.org>

seq_before and seq_after depend on the fact that both sequence numbers
have the same type and thus the same bitwidth. We can ensure that by
compile time checking using a compare between the pointer to the
temporary buffers which were created using the typeof of both
parameters. For example gcc would create a warning like
"warning: comparison of distinct pointer types lacks a cast".
Signed-off-by: NSven Eckelmann <sven@narfation.org>

smallest_signed_int(), seq_before() and seq_after() are very useful
functions that help to handle comparisons between sequence numbers.
However they were only defined in vis.c. With this patch every
batman-adv function will be able to use them.
Signed-off-by: NAntonio Quartulli <ordex@autistici.org>
Signed-off-by: NSven Eckelmann <sven@narfation.org>

Usually rcu_dereference isn't necessary in situations were the
RCU-protected data structure cannot change, but sparse and lockdep still
need a similar functionality for analysis. rcu_dereference_protected
implements the reduced version which should be used to support the
dynamic and static analysis.
Signed-off-by: NSven Eckelmann <sven@narfation.org>

Documentation/CodingStyle recommends to use the form

	p = kmalloc(sizeof(*p), ...);

to calculate the size of a struct and not the version where the struct
name is spelled out to prevent bugs when the type of p changes. This
also seems appropriate for manipulation of buffers when they are
directly associated with p.
Signed-off-by: NSven Eckelmann <sven@narfation.org>

Casting from pointer like 'struct orig_node*' to 'struct orig_node *'
doesn't provide any additional functionality and can be savely removed.
Signed-off-by: NSven Eckelmann <sven@narfation.org>

It is not necessary to cast a void* to the pointer type when we just
store it and don't want to do pointer arithmetic before the actual
assignment.
Signed-off-by: NSven Eckelmann <sven@narfation.org>

It is not save to provide memory for an int and then cast the pointer to
it to long*. It is better to standardize the up and down gateway
bandwith representation to simple ints and only use long inside
conversation routines.
Signed-off-by: NSven Eckelmann <sven@narfation.org>

batman-adv uses pointers which are marked as const and should not
violate that type qualifier by passing it to functions which force a
cast to the non-const version.
Signed-off-by: NSven Eckelmann <sven@narfation.org>

The size of void is currently set by gcc to 1, but is not well defined
in general. Therefore it is more advisable to cast it to char* before
doing pointer arithmetic.
Signed-off-by: NSven Eckelmann <sven@narfation.org>

Signed-off-by: NSven Eckelmann <sven@narfation.org>

Signed-off-by: NSven Eckelmann <sven@narfation.org>

Signed-off-by: NSven Eckelmann <sven@narfation.org>

Signed-off-by: NSven Eckelmann <sven@narfation.org>

Every time that find_router() is invoked, if_status has to be compared with
IF_ACTIVE. Moving this comparison inside find_router() will avoid to write it
each time.
Signed-off-by: NAntonio Quartulli <ordex@autistici.org>
Signed-off-by: NSven Eckelmann <sven@narfation.org>

Initialize xdr_stream and xdr_buf using an array of page pointers
and length of buffer.
Signed-off-by: NBenny Halevy <bhalevy@panasas.com>

TI-RPC introduces the capability of performing RPC over AF_LOCAL
sockets.  It uses this mainly for registering and unregistering
local RPC services securely with the local rpcbind, but we could
also conceivably use it as a generic upcall mechanism.

This patch provides a client-side only implementation for the moment.
We might also consider a server-side implementation to provide
AF_LOCAL access to NLM (for statd downcalls, and such like).

Autobinding is not supported on kernel AF_LOCAL transports at this
time.  Kernel ULPs must specify the pathname of the remote endpoint
when an AF_LOCAL transport is created.  rpcbind supports registering
services available via AF_LOCAL, so the kernel could handle it with
some adjustment to ->rpcbind and ->set_port.  But we don't need this
feature for doing upcalls via well-known named sockets.

This has not been tested with ULPs that move a substantial amount of
data.  Thus, I can't attest to how robust the write_space and
congestion management logic is.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NTrond Myklebust <Trond.Myklebust@netapp.com>

Clean up.  The documenting comment at the top of net/sunrpc/clnt.c is
out of date.  We adopted BSD's RTO estimation mechanism years ago.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NTrond Myklebust <Trond.Myklebust@netapp.com>

As libtirpc does in user space, have our registration API try using an
AF_LOCAL transport first when registering and unregistering.

This means we don't chew up privileged ports, and our registration is
bound to an "owner" (the effective uid of the process on the sending
end of the transport).  Only that "owner" may unregister the service.

The kernel could probe rpcbind via an rpcbind query to determine
whether rpcbind has an AF_LOCAL service. For simplicity, we use the
same technique that libtirpc uses: simply fail over to network
loopback if creating an AF_LOCAL transport to the well-known rpcbind
service socket fails.

This means we open-code the pathname of the rpcbind socket in the
kernel.  For now we have to do that anyway because the kernel's
RPC over AF_LOCAL implementation does not support autobind.  That may
be undesirable in the long term.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NTrond Myklebust <Trond.Myklebust@netapp.com>

Clean up.  Preferred style is not to use curly braces around
switch cases.  I'm about to add another case that needs a third
type cast.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NTrond Myklebust <Trond.Myklebust@netapp.com>

Clean up: Use a more generic name for xs_encode_tcp_fragment_header();
it's appropriate to use for all stream transport types.  We're about
to add new stream transport.

Also, move it to a place where it is more easily shared amongst the
various send_request methods.  And finally, replace the "htonl" macro
invocation with its modern equivalent.
Signed-off-by: NChuck Lever <chuck.lever@oracle.com>
Signed-off-by: NTrond Myklebust <Trond.Myklebust@netapp.com>

The TCP connection state code depends on the state_change() callback
being called when the SYN_SENT state is set. However the networking layer
doesn't actually call us back in that case.
Signed-off-by: NTrond Myklebust <Trond.Myklebust@netapp.com>
Cc: stable@kernel.org

Ingo Molnar noticed that we have this unnecessary ratelimit.h
dependency in linux/net.h, which hid compilation problems from
people doing builds only with CONFIG_NET enabled.

Move this stuff out to a seperate net/net_ratelimit.h file and
include that in the only two places where this thing is needed.
Signed-off-by: NDavid S. Miller <davem@davemloft.net>
Acked-by: NIngo Molnar <mingo@elte.hu>

Several crashes in cleanup_once() were reported in recent kernels.

Commit d6cc1d64 (inetpeer: various changes) added a race in
unlink_from_unused().

One way to avoid taking unused_peers.lock before doing the list_empty()
test is to catch 0->1 refcnt transitions, using full barrier atomic
operations variants (atomic_cmpxchg() and atomic_inc_return()) instead
of previous atomic_inc() and atomic_add_unless() variants.

We then call unlink_from_unused() only for the owner of the 0->1
transition.

Add a new atomic_add_unless_return() static helper

With help from Arun Sharma.

Refs: https://bugzilla.kernel.org/show_bug.cgi?id=32772Reported-by: NArun Sharma <asharma@fb.com>
Reported-by: NMaximilian Engelhardt <maxi@daemonizer.de>
Reported-by: NYann Dupont <Yann.Dupont@univ-nantes.fr>
Reported-by: NDenys Fedoryshchenko <denys@visp.net.lb>
Signed-off-by: NEric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>