The IPV6_UNICAST_IF feature is the IPv6 compliment to IP_UNICAST_IF.
Signed-off-by: Erich E. Hoover <ehoover@mines.edu>
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

The IP_UNICAST_IF feature is needed by the Wine project.  This patch
implements the feature by setting the outgoing interface in a similar
fashion to that of IP_MULTICAST_IF.  A separate option is needed to
handle this feature since the existing options do not provide all of
the characteristics required by IP_UNICAST_IF, a summary is provided
below.

SO_BINDTODEVICE:
* SO_BINDTODEVICE requires administrative privileges, IP_UNICAST_IF
does not.  From reading some old mailing list articles my
understanding is that SO_BINDTODEVICE requires administrative
privileges because it can override the administrator's routing
settings.
* The SO_BINDTODEVICE option restricts both outbound and inbound
traffic, IP_UNICAST_IF only impacts outbound traffic.

IP_PKTINFO:
* Since IP_PKTINFO and IP_UNICAST_IF are independent options,
implementing IP_UNICAST_IF with IP_PKTINFO will likely break some
applications.
* Implementing IP_UNICAST_IF on top of IP_PKTINFO significantly
complicates the Wine codebase and reduces the socket performance
(doing this requires a lot of extra communication between the
"server" and "user" layers).

bind():
* bind() does not work on broadcast packets, IP_UNICAST_IF is
specifically intended to work with broadcast packets.
* Like SO_BINDTODEVICE, bind() restricts both outbound and inbound
traffic.
Signed-off-by: Erich E. Hoover <ehoover@mines.edu>
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Shlomo Pongratz reported GRO L2 header check was suited for Ethernet
only, and failed on IB/ipoib traffic.

He provided a patch faking a zeroed header to let GRO aggregates frames.

Roland Dreier, Herbert Xu, and others suggested we change GRO L2 header
check to be more generic, ie not assuming L2 header is 14 bytes, but
taking into account hard_header_len.

__napi_gro_receive() has special handling for the common case (Ethernet)
to avoid a memcmp() call and use an inline optimized function instead.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Reported-by: Shlomo Pongratz <shlomop@mellanox.com>
Cc: Roland Dreier <roland@kernel.org>
Cc: Or Gerlitz <ogerlitz@mellanox.com>
Cc: Herbert Xu <herbert@gondor.apana.org.au>
Tested-by: Sean Hefty <sean.hefty@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Commit a0417fa3 ("net: Make qdisc_skb_cb upper size bound
explicit.") made it possible for a netdev driver to use skb->cb
between its header_ops.create method and its .ndo_start_xmit
method.  Use this in ipoib_hard_header() to stash away the LL address
(GID + QPN), instead of the "ipoib_pseudoheader" hack.  This allows
IPoIB to stop lying about its hard_header_len, which will let us fix
the L2 check for GRO.
Signed-off-by: Roland Dreier <roland@purestorage.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Correct spelling "regiser" to "register" in
drivers/net/ethernet/stmicro/stmmac/stmmac_pci.c
Signed-off-by: Masanari Iida <standby24x7@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Refs:
1. pmac32_defconfig
http://kisskb.ellerman.id.au/kisskb/buildresult/5583746/
2. ppc6xx_defconfig
http://kisskb.ellerman.id.au/kisskb/buildresult/5584116/

Confirmed any such occurances from all failed defconfigs &
in net-next sources with
grep -nrs "netdev_alloc_skb" drivers/net/ethernet/ | grep -v ","
Reported-by: Geert Uytterhoeven <geert@linux-m68k.org>
Signed-off-by: Pradeep A Dalvi <netdev@pradeepdalvi.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Reported-by: Geert Uytterhoeven <geert@linux-m68k.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

"priv" is initialized twice.  I kept the second one, because it is next
to the check for NULL.
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

No need to maintain a parallel net_device_stats structure in
sh_eth_private, since we have a generic one in netdev

Fix two dma_map_single() incorrect parameters, passing skb->tail instead
of skb->data. Seems that there is no corresponding dmap_unmap_single()
calls for the moment in this driver.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Cc: Yoshihiro Shimoda <yoshihiro.shimoda.uh@renesas.com>
Tested-by: Yoshihiro Shimoda <yoshihiro.shimoda.uh@renesas.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

commit 21a4e469 (netdev: ethernet dev_alloc_skb to netdev_alloc_skb)
should have used "ndev" instead of "dev".

This causes the following build errors:

drivers/net/ethernet/freescale/fec.c: In function 'fec_enet_rx':
drivers/net/ethernet/freescale/fec.c:714: error: 'dev' undeclared (first use in this function)
drivers/net/ethernet/freescale/fec.c:714: error: (Each undeclared identifier is reported only once
drivers/net/ethernet/freescale/fec.c:714: error: for each function it appears in.)
drivers/net/ethernet/freescale/fec.c: In function 'fec_enet_alloc_buffers':
drivers/net/ethernet/freescale/fec.c:1213: error: 'dev' undeclared (first use in this function)

Fix it, so that fec driver can be built again.
Signed-off-by: Fabio Estevam <fabio.estevam@freescale.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

The qdisc supports two operations - plug and unplug. When the
qdisc receives a plug command via netlink request, packets arriving
henceforth are buffered until a corresponding unplug command is received.
Depending on the type of unplug command, the queue can be unplugged
indefinitely or selectively.

This qdisc can be used to implement output buffering, an essential
functionality required for consistent recovery in checkpoint based
fault-tolerance systems. Output buffering enables speculative execution
by allowing generated network traffic to be rolled back. It is used to
provide network protection for Xen Guests in the Remus high availability
project, available as part of Xen.

This module is generic enough to be used by any other system that wishes
to add speculative execution and output buffering to its applications.

This module was originally available in the linux 2.6.32 PV-OPS tree,
used as dom0 for Xen.

For more information, please refer to http://nss.cs.ubc.ca/remus/
and http://wiki.xensource.com/xenwiki/Remus

Changes in V3:
  * Removed debug output (printk) on queue overflow
  * Added TCQ_PLUG_RELEASE_INDEFINITE - that allows the user to
    use this qdisc, for simple plug/unplug operations.
  * Use of packet counts instead of pointers to keep track of
    the buffers in the queue.
Signed-off-by: Shriram Rajagopalan <rshriram@cs.ubc.ca>
Signed-off-by: Brendan Cully <brendan@cs.ubc.ca>
[author of the code in the linux 2.6.32 pvops tree]
Signed-off-by: David S. Miller <davem@davemloft.net>

Cleanup of some whitespace and indentation of a single code block.
Signed-off-by: Bruce Allan <bruce.w.allan@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>

Signed-off-by: Bruce Allan <bruce.w.allan@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>

WARNING: min() should probably be min_t(unsigned int, 4, skb->data_len)
Signed-off-by: Bruce Allan <bruce.w.allan@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>

Signed-off-by: Bruce Allan <bruce.w.allan@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>

Use the existing hw pointer.
Signed-off-by: Bruce Allan <bruce.w.allan@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>

Combine two switch statements into one, convert a nebulous pointer to one
that is a bit more in keeping with the rest of the driver code and cleanup
some coding style.  No change in functionality, just cosmetic changes.
Signed-off-by: Bruce Allan <bruce.w.allan@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>

Combine two switch statements into one, convert a nebulous pointer to one
that is a bit more in keeping with the rest of the driver code and remove
some dead code (there are no 80003es2lan devices with fiber).  No change in
functionality, just cosmetic changes.
Signed-off-by: Bruce Allan <bruce.w.allan@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>

The majority of the e1000e code checks most function return values using a
test like 'if (ret_val)' or 'if (!ret_val)' but there are a few instances
of 'if (ret_val == 0)'.  This patch converts the latter to the former for
consistency.
Signed-off-by: Bruce Allan <bruce.w.allan@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>

warning: missing initializer
Signed-off-by: Bruce Allan <bruce.w.allan@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>

When TX hang occurs e1000_dump prints TX ring, RX ring and Device registers.
Signed-off-by: Tushar Dave <tushar.n.dave@intel.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>

The existing ITR code is broken and confusing, with lots of similarly-named
variables that do different things. Additionally, after the driver carefully
determines the optimal interrupt rate for the adapter, it then
ignores it and always writes a fixed, suboptimal value.

This patch refactors that code to make variable names more descriptive of
what they actually do, and then actually writes the calculated result to
the hardware.

Preliminary testing shows that netperf TCP_STREAM tests goes from ~918Mbps
to ~940Mbps, and TCP_RR goes from ~2k transactions/sec up to > 8k.
Signed-off-by: Mitch Williams <mitch.a.williams@intel.com>
Tested-by: Robert E Garrett <robertX.e.garrett@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>

Modifies message rejection logic so that TIPC doesn't attempt to
send a FIN message to the rejecting port if it is known in advance
that there is no such message because the rejecting port doesn't exist.
Signed-off-by: Allan Stephens <allan.stephens@windriver.com>
Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com>

Removes code that alters the publication key of a name table entry
that is being forcibly purged from TIPC's name table after contact
with the publishing node has been lost.

Current TIPC ensures that all defunct names are purged before
re-establishing contact with a failed node.  There used to be a risk
that the publication might be accidentally deleted because it might be
re-added to the name table before the purge operation was completed.
But now there is no longer a need to ensure that the new key is different
than the old one.
Signed-off-by: Allan Stephens <allan.stephens@windriver.com>
Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com>

Modifies broadcast link so that an incoming fragmented message is not
lost if reassembly cannot begin because there currently is no buffer
big enough to hold the entire reassembled message. The broadcast link
now ignores the first fragment completely, which causes the sending node
to retransmit the first fragment so that reassembly can be re-attempted.

Previously, the sender would have had no reason to retransmit the 1st
fragment, so we would never have a chance to re-try the allocation.

To do this cleanly without duplicaton, a new bclink_accept_pkt()
function is introduced.
Signed-off-by: Allan Stephens <allan.stephens@windriver.com>
Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com>

Modifies unicast link endpoint logic so an incoming fragmented message
is not lost if reassembly cannot begin because there is no buffer big
enough to hold the entire reassembled message. The link endpoint now
ignores the first fragment completely, which causes the sending node to
retransmit the first fragment so that reassembly can be re-attempted.

Previously, the sender would have had no reason to retransmit the 1st
fragment, so we would never have a chance to re-try the allocation.
Signed-off-by: Allan Stephens <allan.stephens@windriver.com>

Eliminates support for the broadcast tag field, which is no longer
used by broadcast link NACK messages.
Signed-off-by: Allan Stephens <allan.stephens@windriver.com>
Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com>

Completely redesigns broadcast link ACK and NACK mechanisms to prevent
spurious retransmit requests in dual LAN networks, and to prevent the
broadcast link from stalling due to the failure of a receiving node to
acknowledge receiving a broadcast message or request its retransmission.

Note: These changes only impact the timing of when ACK and NACK messages
are sent, and not the basic broadcast link protocol itself, so inter-
operability with nodes using the "classic" algorithms is maintained.

The revised algorithms are as follows:

1) An explicit ACK message is still sent after receiving 16 in-sequence
messages, and implicit ACK information continues to be carried in other
unicast link message headers (including link state messages).  However,
the timing of explicit ACKs is now based on the receiving node's absolute
network address rather than its relative network address to ensure that
the failure of another node does not delay the ACK beyond its 16 message
target.

2) A NACK message is now typically sent only when a message gap persists
for two consecutive incoming link state messages; this ensures that a
suspected gap is not confirmed until both LANs in a dual LAN network have
had an opportunity to deliver the message, thereby preventing spurious NACKs.
A NACK message can also be generated by the arrival of a single link state
message, if the deferred queue is so big that the current message gap
cannot be the result of "normal" mis-ordering due to the use of dual LANs
(or one LAN using a bonded interface). Since link state messages typically
arrive at different nodes at different times the problem of multiple nodes
issuing identical NACKs simultaneously is inherently avoided.

3) Nodes continue to "peek" at NACK messages sent by other nodes. If
another node requests retransmission of a message gap suspected (but not
yet confirmed) by the peeking node, the peeking node forgets about the
gap and does not generate a duplicate retransmit request. (If the peeking
node subsequently fails to receive the lost message, later link state
messages will cause it to rediscover and confirm the gap and send another
NACK.)

4) Message gap "equality" is now determined by the start of the gap only.
This is sufficient to deal with the most common cases of message loss,
and eliminates the need for complex end of gap computations.

5) A peeking node no longer tries to determine whether it should send a
complementary NACK, since the most common cases of message loss don't
require it to be sent. Consequently, the node no longer examines the
"broadcast tag" field of a NACK message when peeking.
Signed-off-by: Allan Stephens <allan.stephens@windriver.com>
Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com>

Ensures that all attempts to update broadcast link statistics are done
only while holding the lock that protects the link's main data structures,
to prevent interference by simultaneous updates caused by messages
arriving on other interfaces.
Signed-off-by: Allan Stephens <allan.stephens@windriver.com>
Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com>

Modifies broadcast link so that it increments the "received duplicate
message" count if an incoming message cannot be added to the deferred
message queue because it is already present in the queue. (The aligns
broadcast link behavior with that of TIPC's unicast links.)
Signed-off-by: Allan Stephens <allan.stephens@windriver.com>
Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com>

Fixes a pair of problems in broadcast link message reception code
relating to the reclamation of the node lock after consuming an
in-sequence message.

1) Now retests to see if the sending node is still up after reclaiming
   the node lock, and bails out if it is non-operational.

2) Now manipulates the node's deferred message queue only after
   reclaiming the node lock, rather than using queue head pointer
   information that was cached previously.
Signed-off-by: Allan Stephens <allan.stephens@windriver.com>
Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com>

Ensures that any attempt to send a NACK message over TIPC's broadcast
link has exclusive access to the link's main data structures, to prevent
interference with a simultaneous attempt to send other broadcast link
traffic (such as application-generated multicast messages).
Signed-off-by: Allan Stephens <allan.stephens@windriver.com>
Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com>

Corrects a problem in which a link endpoint that activates as the
result of receiving a RESET/STATE sequence of link protocol messages
fails to properly record the broadcast link status information about
the node to which it is now communicating with. (The problem does
not occur with the more common RESET/ACTIVATE sequence of messages.)
The fix ensures that the broadcast link status info is updated after
the RESET message resets the link endpoint, rather than before, thereby
preventing new information from being overwritten by the reset operation.
Signed-off-by: Allan Stephens <allan.stephens@windriver.com>
Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com>

Fix a bug that can prevent TIPC from sending broadcast messages to a node
if contact with the node is lost and then regained. The problem occurs if
the broadcast link first clears the flag indicating the node is part of the
link's distribution set (when it loses contact with the node), and later
fails to restore the flag (when contact is regained); restoration fails
if contact with the node is regained by implicit unicast link activation
triggered by the arrival of a data message, rather than explicitly by the
arrival of a link activation message.

The broadcast link now uses separate fields to track whether a node is
theoretically capable of receiving broadcast messages versus whether it is
actually part of the link's distribution set. The former member is updated
by the receipt of link protocol messages, which can occur at any time; the
latter member is updated only when contact with the node is gained or lost.
This change also permits the simplification of several conditional
expressions since the broadcast link's "supported" field can now only be
set if there are working links to the associated node.
Signed-off-by: Allan Stephens <allan.stephens@windriver.com>
Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com>

Ensure that sequence number information about incoming broadcast link
messages is initialized only by the activation of the first link to a
given cluster node.  Previously, a race condition allowed reset and/or
activation messages for a second link to re-initialize this sequence
number information with obsolete values. This could trigger TIPC to
request the retransmission of previously acknowledged broadcast link
messages from that node, resulting in broadcast link processing becoming
stalled if the node had already released one or more of those messages
and was unable to perform the required retransmission.

Thanks to Laser <gotolaser@gmail.com> for identifying this problem
and assisting in the development of this fix.
Signed-off-by: Allan Stephens <allan.stephens@windriver.com>
Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com>

Ensures that a link endpoint discards any previously deferred link
protocol message whenever it attempts to send a new one.

Previously, it was possible for a link protocol message that was unsent
due to congestion to be transmitted after newer protocol messages had
been sent. The stale link protocol message might then cause the receiving
link endpoint to malfunction because of its outdated conent.

Thanks to Osamu Kaminuma [okaminum@avaya.com] for diagnosing the problem
and contributing a prototype patch.
Signed-off-by: Allan Stephens <allan.stephens@windriver.com>
Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com>

Re-code the algorithm for inserting an out-of-sequence message into
a unicast or broadcast link's deferred message queue.  It remains
functionally equivalent but should be easier to understand/maintain.
Signed-off-by: Allan Stephens <allan.stephens@windriver.com>
Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com>

Just like skb->cb[], so that qdisc_skb_cb can be encapsulated inside
of other data structures.

This is intended to be used by IPoIB so that it can remember
addressing information stored at hard_header_ops->create() time that
it can fetch when the packet gets to the transmit routine.
Signed-off-by: David S. Miller <davem@davemloft.net>