Implement the peer RxRPC endpoint cache.  Only the primary cache is used.
This is indexed on the following details:

  - Network transport family - currently only AF_INET.
  - Network transport type - currently only UDP.
  - Peer network transport address.

We use the RCU read lock to handle non-creating lookups so that we can do
the look up from bottom half context in the sk_error_report handler.
Creating lookups are done under spinlock rather than mutex as they might be
set up due to an external stimulus if the local endpoint is a server.

Captured network error messages (ICMP) are handled with respect to this
struct and MTU size and RTT are cached here.
Signed-off-by: David Howells <dhowells@redhat.com>

Rename ar-peer.c to peer-object.c for consistency (to be) with the other
new object cache management files.  It will be modified in a later commit.
Signed-off-by: David Howells <dhowells@redhat.com>

Rename ar-local.c to local-event.c.  What's left in the file is the local
endpoint event/packet handling code after the object management has been
separated out.
Signed-off-by: David Howells <dhowells@redhat.com>

Add a proc file to list local rxrpc endpoints using the object cache
facility to do much of the work.
Signed-off-by: David Howells <dhowells@redhat.com>

Implement the local RxRPC endpoint cache.  Only the primary cache is used.
This is indexed on the following details:

  - Local network transport family - currently only AF_INET.
  - Local network transport type - currently only UDP.
  - Local network transport address.

The hash isn't very big since we don't expect to have many local endpoints
hanging around - RxRPC sockets opened with a 0 service ID (ie. client-only
sockets) share local endpoints if they have matching local network
addresses (typically all zeros).

We use a mutex to handle lookups and don't provide RCU-only lookups since
we only expect write access to this cache to be done from process context
when opening a socket.  The local endpoint object is pointed to by the
transport socket's sk_user_data for the life of the transport socket so
that it's fast to access by the transport socket sk_data_ready and
sk_error_report callbacks.

Further, the transport socket is shut down before we clear the sk_user_data
pointer, so that we can be sure that the transport socket's callbacks won't
be invoked once the RCU destruction is scheduled.

The local endpoint retains the transport socket that we use to send and
capture packets and capture network error messages (ICMP).  The socket is
opened when an endpoint is looked up - if it doesn't already exist.

Note that to make this work, we have to get rid of rxrpc_local_lock as that
causes a potential deadlock between a softirq looking in an object cache
whilst holding that lock vs objcache_clear() taking the cache lock and then
an interrupt.

However, since the socket is locked by the caller of the rxrpc_data_ready()
function and given that we don't clear sk_user_data until after we've shut
down the socket, we are guaranteed that the local endpoint struct is pinned
until rxrpc_data_ready() returns - so we don't need to lock the local
endpoint struct there.

The other places we've taken the lock where we read the usage count and
then increment it if not zero can be replaced by atomic_inc_not_zero()
(hidden inside rxrpc_get_local_maybe()).
Signed-off-by: David Howells <dhowells@redhat.com>

Separate local endpoint object handling out into its own file preparatory
to overhauling it to use the new object cache.  The original file is then
be used exclusively for the local endpoint packet and event handling.
Signed-off-by: David Howells <dhowells@redhat.com>

Use the object cache primary hash to provide lists of RxRPC objects through
/proc/net/ for all caches where desired.  Each user of the cache just needs
to provide a show function in its objcache struct and register the proc
file with objcache_seq_fops as its file operations.
Signed-off-by: David Howells <dhowells@redhat.com>

Add a common object cache implementation for RxRPC.  This will be used to
cache objects of various types (calls, connections, local and remote
endpoint records).  Each object that would be cached must contain an
obj_node struct for the cache to use.  The object's usage count and link
pointers are here, plus other internal metadata.

Each object cache consists of a primary hash to which all objects of that
type must be added and a secondary hash to which objects may also be added
and removed a single time.  Objects are automatically removed from both
hashes when they expire.

Objects start off life with a usage count of 2 - one for the cache and one
for the caller.  When an object's usage count is reduced to 1, it sits in
the cache until its expiry time is reached, at which point the cache
attempts to reduce the count to 0 and, if successful, clean it up.  An
object with a usage count of 1 in the cache can be looked up and have its
usage count increased, thereby stopping the expiry process.

Objects are looked up, unlinked and destroyed under RCU-safe conditions.

A garbage collector cycles through all the hash buckets in the primary hash
and compares the expiry times of the usage-count-1 objects to the current
time, removing any that have expired.  This is kicked by a single timer for
the whole cache rather than having a timer per object.
Signed-off-by: David Howells <dhowells@redhat.com>

Until now, we have kept a pre-allocated protocol message header
aggregated into struct tipc_link. Apart from adding unnecessary
footprint to the link instances, this requires extra code both to
initialize and re-initialize it.

We now remove this sub-optimization. This change also makes it
possible to clean up the function tipc_build_proto_msg() and remove
a couple of small functions that were accessing the mentioned header.
In particular, we can replace all occurrences of the local function
call link_own_addr(link) with the generic tipc_own_addr(net).
Acked-by: Ying Xue <ying.xue@windriver.com>
Signed-off-by: Jon Maloy <jon.maloy@ericsson.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Make sure the user has provided a scope for multicast and link local
addresses used locally by a UDP bearer.
Signed-off-by: Richard Alpe <richard.alpe@ericsson.com>
Acked-by: Jon Maloy <jon.maloy@ericsson.com>
Reviewed-by: Erik Hugne <erik.hugne@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

The netlink policy for TIPC_NLA_UDP_LOCAL and TIPC_NLA_UDP_REMOTE
is of type binary with a defined length. This causes the policy
framework to threat the defined length as maximum length.

There is however no protection against a user sending a smaller
amount of data. Prior to this patch this wasn't handled which could
result in a partially incomplete sockaddr_storage struct containing
uninitialized data.

In this patch we use nla_memcpy() when copying the user data. This
ensures a potential gap at the end is cleared out properly.

This was found by Julia with Coccinelle tool.
Reported-by: Daniel Borkmann <daniel@iogearbox.net>
Reported-by: Julia Lawall <julia.lawall@lip6.fr>
Signed-off-by: Richard Alpe <richard.alpe@ericsson.com>
Acked-by: Jon Maloy <jon.maloy@ericsson.com>
Reviewed-by: Erik Hugne <erik.hugne@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Make sure we have a link before checking if it has been reset or not.

Prior to this patch tipc_link_is_reset() could be called with a non
existing link, resulting in a null pointer dereference.
Signed-off-by: Richard Alpe <richard.alpe@ericsson.com>
Acked-by: Jon Maloy <jon.maloy@ericsson.com>
Reviewed-by: Erik Hugne <erik.hugne@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Prior to this patch enabling a IPv4 UDP bearer caused a null pointer
dereference in iptunnel_xmit_stats(), when it tried to dereference the
net device from the skb. To resolve this we now point the skb device
to the net device resolved from the routing table.

Fixes: 039f5062 (ip_tunnel: Move stats update to iptunnel_xmit())
Signed-off-by: Richard Alpe <richard.alpe@ericsson.com>
Acked-by: Jon Maloy <jon.maloy@ericsson.com>
Reviewed-by: Erik Hugne <erik.hugne@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

is actually "active high". Thanks for Troy Kisky for pointing
that out.

Since the patch is in linux-next, this patch is incremental and doesn't
replace the original patch.
Signed-off-by: Bernhard Walle <bernhard@bwalle.de>
Acked-by: Fugang Duan <fugang.duan@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

drivers/net/ethernet/stmicro/stmmac/stmmac_platform.c:115:15-21: ERROR: application of sizeof to pointer

 sizeof when applied to a pointer typed expression gives the size of
 the pointer

Generated by: scripts/coccinelle/misc/noderef.cocci

CC: Giuseppe Cavallaro <peppe.cavallaro@st.com>
Signed-off-by: Fengguang Wu <fengguang.wu@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Acked-by: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Zhang Shengju says:

====================
use reset to set header pointers

This patch series replace set function with reset when offset is zero.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

Since offset is zero, it's not necessary to use set function. Reset
function is straightforward, and will remove the unnecessary add
operation in set function.
Signed-off-by: Zhang Shengju <zhangshengju@cmss.chinamobile.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Since offset is zero, it's not necessary to use set function. Reset
function is straightforward, and will remove the unnecessary add
operation in set function.
Signed-off-by: Zhang Shengju <zhangshengju@cmss.chinamobile.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Since offset is zero, it's not necessary to use set function. Reset
function is straightforward, and will remove the unnecessary add
operation in set function.
Signed-off-by: Zhang Shengju <zhangshengju@cmss.chinamobile.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Since offset is zero, it's not necessary to use set function. Reset
function is straightforward, and will remove the unnecessary add operation
in set function.
Signed-off-by: Zhang Shengju <zhangshengju@cmss.chinamobile.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Merge tag 'rxrpc-rewrite-20160304' of git://git.kernel.org/pub/scm/linux/kernel/git/dhowells/linux-fs

David Howells says:

====================
RxRPC: Rewrite part 1

Here's the first set of patches from my RxRPC rewrite, aimed at net-next.
These do some clean ups and bug fixes.  Most of the changes are small, but
there are a couple of bigger changes:

 (*) Convert call flag numbers and event numbers into enums.  Then rename
     the event numbers to all have _EV_ in their name to stop confusion.
     Fix one instance of an event bit being used instead of a flag bit.

 (*) A copy of the Rx protocol header is kept in the sk_buff private data.
     Keep this in host byte order rather than network byte order as it
     makes more sense.  A number of other fields then get converted into
     host byte order too.

     Conversion between host and network byte order is then done at the
     packet reception/generation stage.

This is based on net-next/master
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

Nicolas Dichtel says:

====================
uapi: consolidate DIV_ROUND_UP definition

The inital goal was to consolidate ethtool.h uapi header. But I took the
opportunity to remove all duplicate definitions of DIV_ROUND_UP.

v3: add patch #2 and #3

v2: split the patch
    define DIV_ROUND_UP in uapi
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

INT_MAX needs limits.h in userland.
When ethtool.h is included by a userland app, we got the following error:

.../usr/include/linux/ethtool.h: In function 'ethtool_validate_speed':
.../usr/include/linux/ethtool.h:1471:18: error: 'INT_MAX' undeclared (first use in this function)
  return speed <= INT_MAX || speed == SPEED_UNKNOWN
                  ^

Fixes: e02564ee ("ethtool: make validate_speed accept all speeds between 0 and INT_MAX")
CC: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Signed-off-by: Nicolas Dichtel <nicolas.dichtel@6wind.com>
Acked-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Let's use __KERNEL_DIV_ROUND_UP, which is defined in uapi/linux/kernel.h.
Signed-off-by: Nicolas Dichtel <nicolas.dichtel@6wind.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

let's use the common definition to avoid the following warning during the
compilation:

drivers/scsi/cxgbi/cxgb4i/cxgb4i.c:161:0: warning: "DIV_ROUND_UP" redefined
 #define DIV_ROUND_UP(n, d) (((n) + (d) - 1) / (d))
 ^
In file included from include/linux/list.h:8:0,
                 from include/linux/module.h:9,
                 from drivers/scsi/cxgbi/cxgb4i/cxgb4i.c:16:
include/linux/kernel.h:67:0: note: this is the location of the previous definition
 #define DIV_ROUND_UP __KERNEL_DIV_ROUND_UP
 ^
Signed-off-by: Nicolas Dichtel <nicolas.dichtel@6wind.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

DIV_ROUND_UP is defined in linux/kernel.h only for the kernel.
When ethtool.h is included by a userland app, we got the following error:

include/linux/ethtool.h:1218:8: error: variably modified 'queue_mask' at file scope
  __u32 queue_mask[DIV_ROUND_UP(MAX_NUM_QUEUE, 32)];
        ^

Let's add a common definition in uapi and use it everywhere.

Fixes: ac2c7ad0 ("net/ethtool: introduce a new ioctl for per queue setting")
CC: Kan Liang <kan.liang@intel.com>
Suggested-by: Ben Hutchings <ben@decadent.org.uk>
Signed-off-by: Nicolas Dichtel <nicolas.dichtel@6wind.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Rolf is no longer in his previous role at Netronome and as far as I know no
longer working on the NFP driver. Thus it does not seem appropriate for him
to be a co-maintainer anymore.
Reviewed-by: Dinan Gunawardena <dinan.gunawardena@netronome.com>
Signed-off-by: Simon Horman <simon.horman@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

In the ICMP message processing code, don't try to map ICMP codes to UNIX
error codes as the caller (IPv4/IPv6) already did that for us (ee_errno).
Signed-off-by: David Howells <dhowells@redhat.com>

Clear the unused part of a sockaddr_rxrpc structs so that memcmp() can be
used to compare them.
Signed-off-by: David Howells <dhowells@redhat.com>

Forced casts are needed to avoid sparse warning when directly comparing
be32 values.
Signed-off-by: David Howells <dhowells@redhat.com>

The version number rxkad places in the response should be network byte
order.

Whilst we're at it, rearrange the code to be more readable.
Signed-off-by: David Howells <dhowells@redhat.com>

Use ACCESS_ONCE() when accessing the other-end pointer into a circular
buffer as it's possible the other-end pointer might change whilst we're
doing this, and if we access it twice, we might get some weird things
happening.
Signed-off-by: David Howells <dhowells@redhat.com>

Remove some excess whitespace, insert some missing spaces and adjust a
couple of comments.
Signed-off-by: David Howells <dhowells@redhat.com>

Currently, received RxRPC packets outside the range 1-13 are rejected.
There are, however, holes in the range that should also be rejected - plus
at least one type we don't yet support - so reject these also.
Signed-off-by: David Howells <dhowells@redhat.com>

The upper bound of the defined range for rx_mtu is being set in the same
member as the lower bound (extra1) rather than the correct place (extra2).
I'm not entirely sure why this compiles.
Signed-off-by: David Howells <dhowells@redhat.com>

Fix the protocol family set in the proto_ops for rxrpc to be PF_RXRPC not
PF_UNIX.
Signed-off-by: David Howells <dhowells@redhat.com>

Currently, a copy of the Rx packet header is copied into the the sk_buff
private data so that we can advance the pointer into the buffer,
potentially discarding the original.  At the moment, this copy is held in
network byte order, but this means we're doing a lot of unnecessary
translations.

The reasons it was done this way are that we need the values in network
byte order occasionally and we can use the copy, slightly modified, as part
of an iov array when sending an ack or an abort packet.

However, it seems more reasonable on review that it would be better kept in
host byte order and that we make up a new header when we want to send
another packet.

To this end, rename the original header struct to rxrpc_wire_header (with
BE fields) and institute a variant called rxrpc_host_header that has host
order fields.  Change the struct in the sk_buff private data into an
rxrpc_host_header and translate the values when filling it in.

This further allows us to keep values kept in various structures in host
byte order rather than network byte order and allows removal of some fields
that are byteswapped duplicates.
Signed-off-by: David Howells <dhowells@redhat.com>

Rename call event names to begin RXRPC_CALL_EV_ to distinguish them from the
flags.
Signed-off-by: David Howells <dhowells@redhat.com>

Convert call flag and event numbers into enums and move their definitions
outside of the struct.

Also move the call state enum outside of the struct and add an extra
element to count the number of states.
Signed-off-by: David Howells <dhowells@redhat.com>