This patch implements alternative aggregator selection policies
for 802.3ad.  The existing policy, now termed "stable," selects the active
aggregator by greatest bandwidth, and only reselects a new aggregator
if the active aggregator is entirely disabled (no more ports or all ports
down).

	This patch adds two new policies: bandwidth and count, selecting
the active aggregator by total bandwidth (like the stable policy) or by
the number of ports in the aggregator, respectively.  These two policies
also differ from the stable policy in that they will reselect the active
aggregator when availability-related changes occur in the bond (e.g.,
link state change).

	This permits "gang failover" within 802.3ad, allowing redundant
aggregators along parallel paths to always maintain the "best" aggregator
as the active aggregator (rather than having to wait for the active to
entirely fail).

	This patch also updates the driver version to 3.5.0.
Signed-off-by: NJay Vosburgh <fubar@us.ibm.com>
Signed-off-by: NJeff Garzik <jgarzik@redhat.com>

	The current ALB function that processes incoming ARPs
does not handle traffic for VLANs configured above bonding.  This causes
traffic on those VLANs to all be assigned the same slave.  This patch
corrects that misbehavior by locating the bonding interface nested below
the VLAN interface.

	Bug reported by Sven Anders <anders@anduras.de>, who also
tested an earlier version of this patch and confirmed that it resolved
the problem.
Signed-off-by: NJay Vosburgh <fubar@us.ibm.com>
Signed-off-by: NJeff Garzik <jgarzik@redhat.com>

This patch adds better IPv6 failover support for bonding devices,
especially when in active-backup mode and there are only IPv6 addresses
configured, as reported by Alex Sidorenko.

- Creates a new file, net/drivers/bonding/bond_ipv6.c, for the
   IPv6-specific routines.  Both regular bonds and VLANs over bonds
   are supported.

- Adds a new tunable, num_unsol_na, to limit the number of unsolicited
   IPv6 Neighbor Advertisements that are sent on a failover event.
   Default is 1.

- Creates two new IPv6 neighbor discovery functions:

   ndisc_build_skb()
   ndisc_send_skb()

   These were required to support VLANs since we have to be able to
   add the VLAN id to the skb since ndisc_send_na() and friends
   shouldn't be asked to do this.  These two routines are basically
   __ndisc_send() split into two pieces, in a slightly different order.

- Updates Documentation/networking/bonding.txt and bumps the rev of bond
   support to 3.4.0.

On failover, this new code will generate one packet:

- An unsolicited IPv6 Neighbor Advertisement, which helps the switch
   learn that the address has moved to the new slave.

Testing has shown that sending just the NA results in pretty good
behavior when in active-back mode, I saw no lost ping packets for example.
Signed-off-by: NBrian Haley <brian.haley@hp.com>
Signed-off-by: NJay Vosburgh <fubar@us.ibm.com>
Signed-off-by: NJeff Garzik <jgarzik@redhat.com>

	The only user of the net_device->last_rx field is bonding.
This patch adds a conditional update of last_rx to the bonding special
logic in skb_bond_should_drop, causing last_rx to only be updated when
the ARP monitor is running.

	This frees network device drivers from the necessity of
updating last_rx, which can have cache line thrash issues.
Signed-off-by: NJay Vosburgh <fubar@us.ibm.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Using NIPQUAD() with NIPQUAD_FMT, %d.%d.%d.%d or %u.%u.%u.%u
can be replaced with %pI4
Signed-off-by: NHarvey Harrison <harvey.harrison@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

A panic was discovered with bonding when using mode 5 or 6 and trying to
remove the slaves from the bond after the interface was taken down.
When calling 'ifconfig bond0 down' the following happens:

    bond_close()
        bond_alb_deinitialize()
            tlb_deinitialize()
		kfree(bond_info->tx_hashtbl)
                bond_info->tx_hashtbl = NULL

Unfortunately if there are still slaves in the bond, when removing the
module the following happens:

    bonding_exit()
        bond_free_all()
            bond_release_all()
                bond_alb_deinit_slave()
                    tlb_clear_slave()
                        tx_hash_table = BOND_ALB_INFO(bond).tx_hashtbl
			u32 next_index = tx_hash_table[index].next

As you might guess we panic when trying to access a few entries into the
table that no longer exists.

I experimented with several options (like moving the calls to
tlb_deinitialize somewhere else), but it really makes the most sense to
be part of the bond_close routine.  It also didn't seem logical move
tlb_clear_slave around too much, so the simplest option seems to add a
check in tlb_clear_slave to make sure we haven't already wiped the
tx_hashtbl away before searching for all the non-existent hash-table
entries that used to point to the slave as the output interface.
Signed-off-by: NAndy Gospodarek <andy@greyhouse.net>
Signed-off-by: NJay Vosburgh <fubar@us.ibm.com>
Signed-off-by: NJeff Garzik <jgarzik@redhat.com>

This patch reworks the resource free logic performed at the time
a bonding device is released.  This (a) closes two resource leaks, one
for workqueues and one for multicast lists, and (b) improves commonality
of code between the "destroy one" and "destroy all" paths by performing
final free activity via destructor instead of explicitly (and differently)
in each path.

"Sean E. Millichamp" <sean@bruenor.org> reported the workqueue
leak, and included a different patch.
Signed-off-by: NJay Vosburgh <fubar@us.ibm.com>
Signed-off-by: NJeff Garzik <jgarzik@redhat.com>

During the rework of the mii monitor for:

  commit f0c76d61
  Author: Jay Vosburgh <fubar@us.ibm.com>
  Date:   Wed Jul 2 18:21:58 2008 -0700

    bonding: refactor mii monitor

I left out the increment of the link failure counter.  This
patch corrects that omission.
Signed-off-by: NJay Vosburgh <fubar@us.ibm.com>
Signed-off-by: NJeff Garzik <jgarzik@redhat.com>

As a bonus, removes some unnecessary byteswapping.
Signed-off-by: NHarvey Harrison <harvey.harrison@gmail.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

This converts pretty much everything to print_mac. There were
a few things that had conflicts which I have just dropped for
now, no harm done.

I've built an allyesconfig with this and looked at the files
that weren't built very carefully, but it's a huge patch.
Signed-off-by: NJohannes Berg <johannes@sipsolutions.net>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

My change

    commit e2a6b852
    net: Enable TSO if supported by at least one device

didn't do what was intended because the netdev_compute_features
function was designed for conjunctions.  So what happened was that
it would simply take the TSO status of the last constituent device.

This patch extends it to support both conjunctions and disjunctions
under the new name of netdev_increment_features.

It also adds a new function netdev_fix_features which does the
sanity checking that usually occurs upon registration.  This ensures
that the computation doesn't result in an illegal combination
since this checking is absent when the change is initiated via
ethtool.

The two users of netdev_compute_features have been converted.
Signed-off-by: NHerbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

The following sparse warnings are being generated
because bonding.h is missing definitons for items
declared in bond_main.c but also used in bond_sysfs.h

Also export bond_dev_list as this is also declared
in bond_main but used elsewhere in drivers/net/bonding.

bond_main.c:105:20: warning: symbol 'bonding_defaults' was not declared. Should it be static?
bond_main.c:148:1: warning: symbol 'bond_dev_list' was not declared. Should it be static?
bond_main.c:162:22: warning: symbol 'bond_lacp_tbl' was not declared. Should it be static?
bond_main.c:168:22: warning: symbol 'bond_mode_tbl' was not declared. Should it be static?
bond_main.c:179:22: warning: symbol 'xmit_hashtype_tbl' was not declared. Should it be static?
bond_main.c:186:22: warning: symbol 'arp_validate_tbl' was not declared. Should it be static?
bond_main.c:194:22: warning: symbol 'fail_over_mac_tbl' was not declared. Should it be static?
Signed-off-by: NBen Dooks <ben-linux@fluff.org>
Signed-off-by: NJeff Garzik <jgarzik@redhat.com>

This patch allows reporting the link, checksum, and feature settings
of bonded device by using generic hooks.
Signed-off-by: NStephen Hemminger <shemminger@vyatta.com>
Acked-by: NJay Vosburgh <fubar@us.ibm.com>
Signed-off-by: NJeff Garzik <jgarzik@redhat.com>

__FUNCTION__ is gcc-specific, use __func__
Signed-off-by: NHarvey Harrison <harvey.harrison@gmail.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NJeff Garzik <jgarzik@redhat.com>

IPv6 all-node-multicasts and DAD probes should not be tx-balanced
on ALB/TLB bonds.  The all-node-multicast is an equivalent to IPv4
broadcasts.  DAD probes have to be sent only on the primary so that
we don't get false-positive detections.
Signed-off-by: NJeff Garzik <jgarzik@redhat.com>

Resending since I didn't see any responses from the first try.

Change __constant_htons() to htons() in the bonding driver, it should
only be used for initializers.

-Brian
Signed-off-by: NBrian Haley <brian.haley@hp.com>
Signed-off-by: NJeff Garzik <jgarzik@redhat.com>

It is wrong to destroy a bonding master from a context that uses the sysfs
of that bond. When last IPoIB slave is unenslaved from by writing to a
sysfs file (for bond0 this would be /sys/class/net/bond0/bonding/slaves)
the driver tries to destroy the bond. This is wrong and can lead to a
lockup or a crash.  This fix lets the bonding master stay and relies on
the user to destroy the bonding master if necessary (i.e. before module
ib_ipoib is unloaded)

This patch affects only bonds of IPoIB slaves. Ethernet slaves stay
unaffected.
Signed-off-by: NMoni Shoua <monis@voltaire.com>
Signed-off-by: NJay Vosburgh <fubar@us.ibm.com>
Signed-off-by: NJeff Garzik <jgarzik@redhat.com>

	Refactor mii monitor.  As with the previous ARP monitor refactor,
the motivation for this is to handle locking rationally (in this case,
removing conditional locking) and generally clean up the code.

	This patch breaks up the monolithic mii monitor into two phases:
an inspection phase, followed by an optional commit phase.  The commit phase
is the only portion that requires RTNL or makes changes to state, and is
only called when inspection finds something to change.
Signed-off-by: NJay Vosburgh <fubar@us.ibm.com>
Signed-off-by: NJeff Garzik <jgarzik@redhat.com>

The new address list lock needs to handle the same device layering
issues that the _xmit_lock one does.

This integrates work done by Patrick McHardy.
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

alloc_netdev_mq() now allocates an array of netdev_queue
structures for TX, based upon the queue_count argument.

Furthermore, all accesses to the TX queues are now vectored
through the netdev_get_tx_queue() and netdev_for_each_tx_queue()
interfaces.  This makes it easy to grep the tree for all
things that want to get to a TX queue of a net device.

Problem spots which are not really multiqueue aware yet, and
only work with one queue, can easily be spotted by grepping
for all netdev_get_tx_queue() calls that pass in a zero index.
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Now that we have a specific lock to protect the network
device unicast and multicast lists, remove extraneous
grabs of the TX lock in cases where the code only needs
address list protection.
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Add netif_addr_{lock,unlock}{,_bh}() helpers.

Use them to protect operations that operate on or read
the network device unicast and multicast address lists.

Also use them in cases where the code simply wants to
block calls into the driver's ->set_rx_mode() and
->set_multicast_list() methods.
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

dev_set_promiscuity/allmulti might overflow.
Commit: "netdevice: Fix promiscuity and allmulti overflow" in net-next makes
dev_set_promiscuity/allmulti return error number if overflow happened.

In bond_alb and bond_main, we check all positive increment for promiscuity
and allmulti to get error return.
But there are still two problems left.
1. Some code path has no mechanism to signal errors upstream.
2. If there are multi slaves, it's hard to tell which slaves increment
   promisc/allmulti successfully and which failed.
So I left these problems to be FIXME.
Fortunately, the overflow is very rare case.
Signed-off-by: NWang Chen <wangchen@cn.fujitsu.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Accesses are mostly structured such that when there are multiple TX
queues the code transformations will be a little bit simpler.
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

	Permit bonding to function rationally if max_bonds is set to
zero.  This will load the module, but create no master devices (which can
be created via sysfs).

	Requires some change to bond_create_sysfs; currently, the
netdev sysfs directory is determined from the first bonding device created,
but this is no longer possible.  Instead, an interface from net/core is
created to create and destroy files in net_class.

	Based on a patch submitted by Phil Oester <kernel@linuxaces.com>.
Modified by Jay Vosburgh to fix the sysfs issue mentioned above and to
update the documentation.
Signed-off-by: NPhil Oester <kernel@linuxace.com>
Signed-off-by: NJay Vosburgh <fubar@us.ibm.com>
Signed-off-by: NJeff Garzik <jgarzik@redhat.com>

	Support for sending multiple gratuitous ARPs during failovers
was added by commit:

commit 7893b249
Author: Moni Shoua <monis@voltaire.com>
Date:   Sat May 17 21:10:12 2008 -0700

    bonding: Send more than one gratuitous ARP when slave takes over

	This change modifies that support to remove duplicated code,
add support for ARP monitor (the original only supported miimon), clear
the grat ARP counter in bond_close (lest a later "ifconfig up" immediately
start spewing ARPs), and add documentation for the module parameter.

	Also updated driver version to 3.3.0.
Signed-off-by: NJay Vosburgh <fubar@us.ibm.com>
Signed-off-by: NJeff Garzik <jgarzik@redhat.com>

under active-backup mode and when there's actual new_active slave,
have bond_change_active_slave() call the networking core to deliver
NETDEV_BONDING_FAILOVER event such that the fail-over can be notable
by code outside of the bonding driver such as the RDMA stack and
monitoring tools.

As the correct context of locking appropriate for notifier calls is RTNL
and nothing else, bond->curr_slave_lock and bond->lock are unlocked and
later locked again. This is ensured by the rest of the code to be safe
under backup-mode AND when new_active is not NULL.

Jay Vosburgh modified the original patch for formatting and fixed a
compiler error.
Signed-off-by: NOr Gerlitz <ogerlitz@voltaire.com>
Signed-off-by: NJay Vosburgh <fubar@us.ibm.com>
Signed-off-by: NJeff Garzik <jgarzik@redhat.com>

simplified the code of bond_change_active_slave() such that under
active-backup mode there's one "if (new_active)" test and the rest
of the code only does extra checks on top of it. This removed an
unneeded "if (bond->send_grat_arp > 0)" check and avoid calling
bond_send_gratuitous_arp when there's no active slave.

Jay Vosburgh made minor coding style changes to the orignal patch.
Signed-off-by: NOr Gerlitz <ogerlitz@voltaire.com>
Signed-off-by: NJay Vosburgh <fubar@us.ibm.com>
Signed-off-by: NJeff Garzik <jgarzik@redhat.com>

	Add a "follow" selection for fail_over_mac.  This option
causes the MAC address to move from slave to slave as the active
slave changes.  This is in addition to the existing fail_over_mac option
that causes the bond's MAC address to change during failover.

	This new option is useful for devices that cannot tolerate
multiple ports using the same MAC address simultaneously, either
because it confuses them or incurs a performance penalty (as is the
case with some LPAR-aware multiport devices).  Because the MAC of the
bond itself does not change, the "follow" option is slightly more
reliable during failover and doesn't change the MAC of the bond during
operation.

	This patch requires a previous ARP monitor change to properly
handle RTNL during failovers.
Signed-off-by: NJay Vosburgh <fubar@us.ibm.com>
Signed-off-by: NJeff Garzik <jgarzik@redhat.com>

	Refactor ARP monitor for active-backup mode.  The motivation for
this is to take care of locking issues in a clear manner (particularly to
correctly handle RTNL vs. the bonding locks).  Currently, the a-b ARP
monitor does not hold RTNL at all, but future changes will require RTNL
during ARP monitor failovers.

	Rather than using conditional locking, this patch instead breaks
up the ARP monitor into three discrete steps: inspection, commit changes,
and probe.  The inspection phase marks slaves that require link state
changes.  The commit phase is only called if inspection detects that
changes are needed, and is called with RTNL.  Lastly, the probe phase
issues the ARP probes that the inspection phase uses to determine link
state.
Signed-off-by: NJay Vosburgh <fubar@us.ibm.com>
Signed-off-by: NJeff Garzik <jgarzik@redhat.com>

With IPoIB, reception of gratuitous ARP by neighboring hosts
is essential for a successful change of slaves in case of failure.
Otherwise, they won't learn about the HW address change and need
to wait a long time until the neighboring system gives up and sends
an ARP request to learn the new HW address.  This patch decreases
the chance for a lost of a gratuitous ARP packet by sending it more
than once. The number retries is configurable and can be set with a
module param.
Signed-off-by: NMoni Shoua <monis@voltaire.com>
Acked-by: NJay Vosburgh <fubar@us.ibm.com>
Signed-off-by: NJeff Garzik <jgarzik@redhat.com>

Some places iterate over the checked list right after the check
itself, so even if the list is empty, the list_for_each_xxx
iterator will make everything right by himself.
Signed-off-by: NPavel Emelyanov <xemul@openvz.org>
Acked-by: NJay Vosburgh <fubar@us.ibm.com>
Signed-off-by: NJeff Garzik <jgarzik@redhat.com>

Many places either do not modify the list under the list_for_each_xxx,
or break out of the loop as soon as the first element is removed.

Thus, this _safe iteration just occupies some unneeded .text space
and requires an additional variable.
Signed-off-by: NPavel Emelyanov <xemul@openvz.org>
Acked-by: NJay Vosburgh <fubar@us.ibm.com>
Signed-off-by: NJeff Garzik <jgarzik@redhat.com>

While we're fixing the bond_create, I hope it's OK to polish it
a bit after the fixes.

The third argument is NULL at the first caller and is ignored by
the second one, so remove it.
Signed-off-by: NPavel Emelyanov <xemul@openvz.org>
Acked-by: NJay Vosburgh <fubar@us.ibm.com>
Signed-off-by: NJeff Garzik <jgarzik@redhat.com>

	Remove bond_has_ip and all references to it.  With this change,
the ARP monitor will always send ARP probes if the master is up and has
at least one slave.  If the bond has an IP address, it is used in the
ARP probe; if not, the probes are sent with all zeros in the sender's
IP address (which is consistent with an RFC 2131 4.4.1 duplicate address
probe).

	This is useful for cases when bonding itself is hidden underneath
a layer of virtual devices, e.g., with Xen.

	Change suggested by Tsutomu Fujii <t-fujii@nb.jp.nec.com>, who
included a one-line patch that only affected active-backup mode.
Signed-off-by: NJay Vosburgh <fubar@us.ibm.com>
Signed-off-by: NJeff Garzik <jgarzik@redhat.com>

	Convert bonding to use msecs_to_jiffies instead of doing the
math.  For the ARP monitor, there was an underflow problem that could
result in an infinite loop.  The miimon already had that worked around,
but this is cleaner.

	Originally by Nicolas de Pesloüan <nicolas.2p.debian@free.fr>
Jay Vosburgh corrected a math error in the original; Nicolas' original
commit message is:

When setting arp_interval parameter to a very low value, delta_in_ticks
for next arp might become 0, causing an infinite loop.

See http://bugzilla.kernel.org/show_bug.cgi?id=10680

Same problem for miimon parameter already fixed, but fix might be
enhanced, by using msecs_to_jiffies() function.
Signed-off-by: NNicolas de Pesloüan <nicolas.2p.debian@free.fr>
Signed-off-by: NJay Vosburgh <fubar@us.ibm.com>
Signed-off-by: NJeff Garzik <jgarzik@redhat.com>

If device already exists named bonding_masters, then fail. This is a wierd
corner case only a QA group could love.
Signed-off-by: NStephen Hemminger <shemminger@vyatta.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

	As part of:

commit c2edacf8
Author: Jay Vosburgh <fubar@us.ibm.com>
Date:   Mon Jul 9 10:42:47 2007 -0700

    bonding / ipv6: no addrconf for slaves separately from master

two steps were rearranged in the enslavement process: netdev_set_master
is now before the call to dev_open to open the slave.

	This patch updates the error cases and unwind process at the
end of bond_enslave to match the new order.  Without this patch, it is
possible for the enslavement to fail, but leave the slave with IFF_SLAVE
set in its flags.
Signed-off-by: NJay Vosburgh <fubar@us.ibm.com>
Signed-off-by: NJeff Garzik <jgarzik@redhat.com>

The sysfs layer has an internal protection, that ensures, that
all the process sitting inside ->sore/->show callback exits
before the appropriate entry is unregistered (the calltraces
are rather big, but I can provide them if required).

On the other hand, bonding takes rtnl_lock in
a) the bonding_store_bonds, i.e. in ->store callback,
b) module exit before calling the sysfs unregister routines.

Thus, the classical AB-BA deadlock may occur. To reproduce run
# while :; do modprobe bonding; rmmod bonding; done
and
# while :; do echo '+bond%d' > /sys/class/net/bonding_masters ; done
in parallel.

The fix is to move the bond_destroy_sysfs out of the rtnl_lock,
but _before_ bond_free_all to make sure no bonding devices exist
after module unload.
Signed-off-by: NPavel Emelyanov <xemul@openvz.org>
Acked-by: NJay Vosburgh <fubar@us.ibm.com>
Signed-off-by: NJeff Garzik <jgarzik@redhat.com>

	Fixed an error unwind in bonding_store_bonds that didn't release
the locks it held, and consolidated unwinds into a common block at the
end of the function.  Bug reported by Pavel Emelyanov <xemul@openvz.org>,
who provided a different fix.
Signed-off-by: NJay Vosburgh <fubar@us.ibm.com>
Signed-off-by: NJeff Garzik <jgarzik@redhat.com>