It was recently brought to my attention that 802.3ad mode bonds would no
longer form when using some network hardware after a driver update.
After snooping around I realized that the particular hardware was using
page-based skbs and found that skb->data did not contain a valid LACPDU
as it was not stored there. That explained the inability to form an
802.3ad-based bond. For balance-alb mode bonds this was also an issue
as ARPs would not be properly processed.
This patch fixes the issue in my tests and should be applied to 2.6.36
and as far back as anyone cares to add it to stable.
Thanks to Alexander Duyck <alexander.h.duyck@intel.com> and Jesse
Brandeburg <jesse.brandeburg@intel.com> for the suggestions on this one.
Signed-off-by: Andy Gospodarek <andy@greyhouse.net>
CC: Alexander Duyck <alexander.h.duyck@intel.com>
CC: Jesse Brandeburg <jesse.brandeburg@intel.com>
CC: stable@kerne.org
Signed-off-by: Jay Vosburgh <fubar@us.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
After:
commit 6146b1a4da
Author: Jay Vosburgh <fubar@us.ibm.com>
Date: Tue Nov 4 17:51:15 2008 -0800
bonding: Fix ALB mode to balance traffic on VLANs
the dev field in the RLB ARP packet handler was set to NULL to wildcard
and accommodate balancing VLANs on top of bonds.
This has the side-effect of the packet handler being called against
other, non RLB-enabled bonds, and a kernel oops results when it tries to
dereference rx_hashtbl in rlb_update_entry_from_arp(), which won't be
set for those bonds, e.g. active-backup.
With the __netif_receive_skb() changes from:
commit 1f3c8804ac
Author: Andy Gospodarek <andy@greyhouse.net>
Date: Mon Dec 14 10:48:58 2009 +0000
bonding: allow arp_ip_targets on separate vlans to use arp validation
frames received on VLANs correctly make their way to the bond's handler,
so we no longer need to wildcard the device.
The oops can be reproduced by:
modprobe bonding
echo active-backup > /sys/class/net/bond0/bonding/mode
echo 100 > /sys/class/net/bond0/bonding/miimon
ifconfig bond0 xxx.xxx.xxx.xxx netmask xxx.xxx.xxx.xxx
echo +eth0 > /sys/class/net/bond0/bonding/slaves
echo +eth1 > /sys/class/net/bond0/bonding/slaves
echo +bond1 > /sys/class/net/bonding_masters
echo balance-alb > /sys/class/net/bond1/bonding/mode
echo 100 > /sys/class/net/bond1/bonding/miimon
ifconfig bond1 xxx.xxx.xxx.xxx netmask xxx.xxx.xxx.xxx
echo +eth2 > /sys/class/net/bond1/bonding/slaves
echo +eth3 > /sys/class/net/bond1/bonding/slaves
Pass some traffic on bond0. Boom.
[ Tested, behaves as advertised. I do not believe a test of the bonding
mode is necessary, as there is no race between the packet handler and
the bonding mode changing (the mode can only change when the device is
closed). Also updated the log message to include the reproduction and
full commit ids. -J ]
Signed-off-by: Greg Edwards <greg.edwards@hp.com>
Signed-off-by: Jay Vosburgh <fubar@us.ibm.com>
Acked-by: Andy Gospodarek <andy@greyhouse.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
After commit ad1afb0039
("vlan_dev: VLAN 0 should be treated as "no vlan tag" (802.1p packet)")
it is now regular practice for a VLAN "add vid" for VLAN 0 to
arrive prior to any VLAN registration or creation of a vlan_group.
This patch updates the bonding code that tests for the presence
of VLANs configured above bonding. The new logic tests for bond->vlgrp
to determine if a registration has occured, instead of testing that
bonding's internal vlan_list is empty.
The old code would panic when vlan_list was not empty, but
vlgrp was still NULL (because only an "add vid" for VLAN 0 had occured).
Bonding still adds VLAN 0 to its internal list so that 802.1p
frames are handled correctly on transmit when non-VLAN accelerated
slaves are members of the bond. The test against bond->vlan_list
remains in bond_dev_queue_xmit for this reason.
Modification to the bond->vlgrp now occurs under lock (in
addition to RTNL), because not all inspections of it occur under RTNL.
Additionally, because 8021q will never issue a "kill vid" for
VLAN 0, there is now logic in bond_uninit to release any remaining
entries from vlan_list.
Signed-off-by: Jay Vosburgh <fubar@us.ibm.com>
Cc: Pedro Garcia <pedro.netdev@dondevamos.com>
Cc: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
When two systems using bonding devices in adaptive load
balancing (ALB) communicates with each other, an endless
ping-pong of ARP replies starts between these two systems.
What happens? In the ALB mode, bonding driver keeps track
of each client connected in a hash table, so it can do the
receive load balancing (RLB). This hash table is updated
when an ARP reply is received, then it scans for the client
entry, updates its MAC address and flag it to be announced
later. Therefore, two seconds later, the alb monitor runs
and send for each updated client entry two ARP replies
updating this specific client. The same process happens on
the receiving system, causing the endless ping-pong of arp
replies.
See more information including the relevant functions below:
System 1 System 2
bond0 bond0
ping <system2>
ARP request --------->
<--------- ARP reply
+->rlb_arp_recv <---------------------+ <--- loop begins
| rlb_update_entry_from_arp |
| client_info->ntt = 1; |
| bond_info->rx_ntt = 1; |
| |
| <communication succeed> |
| |
| bond_alb_monitor |
| rlb_update_rx_clients |
| rlb_update_client |
| arp_create(ARPOP_REPLY) |
| send ARP reply --------------> V
| send ARP reply -------------->
| rlb_arp_recv
| rlb_update_entry_from_arp
| client_info->ntt = 1;
| bond_info->rx_ntt = 1;
| < snipped, same as in system 1>
+------- <-------------- send ARP reply
<-------------- send ARP reply
Besides the unneeded networking traffic, this loop breaks
a cluster because a backup system can't take over the IP
address. There is always one system sending an ARP reply
poisoning the network.
This patch fixes the problem adding a check for the MAC
address before updating it. Thus, if the MAC address didn't
change, there is no need to update neither to announce it later.
Signed-off-by: Flavio Leitner <fleitner@redhat.com>
Signed-off-by: Jay Vosburgh <fubar@us.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In the worst case, when the first loop breaks an the end of the slave list,
the slave list is iterated through twice. This patch reduces this
function only to one loop. Also makes it simpler.
Signed-off-by: Jiri Pirko <jpirko@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
Remove DRV_NAME from pr_<level>s
Consolidate long format strings
Remove some extra tab indents
Remove some unnecessary ()s from pr_<level>s arguments
Align pr_<level> arguments
Signed-off-by: Joe Perches <joe@perches.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
fix some typos and punctuation in comments
Signed-off-by: Thadeu Lima de Souza Cascardo <cascardo@holoscopio.com>
Signed-off-by: Jiri Kosina <jkosina@suse.cz>
We can speedup ether addresses compares using compare_ether_addr_64bits()
instead of memcmp(). We make sure all operands are at least 8 bytes long and
16bits aligned (or better, long word aligned if possible)
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Reviewed-by: Jiri Pirko <jpirko@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
I did not introduce new lines over 80 chars. I even eliminated some of
them.
Signed-off-by: Jiri Pirko <jpirko@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch converts the remaining occurences of raw return values to their
symbolic counterparts in ndo_start_xmit() functions that were missed by the
previous automatic conversion.
Additionally code that assumed the symbolic value of NETDEV_TX_OK to be zero
is changed to explicitly use NETDEV_TX_OK.
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
Fix locking issue in alb MAC address management; removed
incorrect locking and replaced with correct locking. This bug was
introduced in commit 059fe7a578
("bonding: Convert locks to _bh, rework alb locking for new locking")
Bug reported by Paul Smith <paul@mad-scientist.net>, who also
tested the fix.
Signed-off-by: Jay Vosburgh <fubar@us.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Remove debug printk I accidently left in as part of commit:
commit 6146b1a4da
Author: Jay Vosburgh <fubar@us.ibm.com>
Date: Tue Nov 4 17:51:15 2008 -0800
bonding: Fix ALB mode to balance traffic on VLANs
Reported by Duncan Gibb <duncan.gibb@siriusit.co.uk>
Signed-off-by: Jay Vosburgh <fubar@us.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Base versions handle constant folding now.
Signed-off-by: Harvey Harrison <harvey.harrison@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Use pr_debug() instead of own macros.
Signed-off-by: Holger Eitzenberger <holger@eitzenberger.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Convert to net_device_ops table.
Note: for some operations move error checking into generic networking
layer (rather than looking at pointers in bonding).
A couple of gratituous style cleanups to get rid of extra {}
Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
We have some reasons to kill netdev->priv:
1. netdev->priv is equal to netdev_priv().
2. netdev_priv() wraps the calculation of netdev->priv's offset, obviously
netdev_priv() is more flexible than netdev->priv.
But we cann't kill netdev->priv, because so many drivers reference to it
directly.
This patch is a safe convert for netdev->priv to netdev_priv(netdev).
Since all of the netdev->priv is only for read.
But it is too big to be sent in one mail.
I split it to 4 parts and make every part smaller than 100,000 bytes,
which is max size allowed by vger.
Signed-off-by: Wang Chen <wangchen@cn.fujitsu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
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: Jay Vosburgh <fubar@us.ibm.com>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
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: Andy Gospodarek <andy@greyhouse.net>
Signed-off-by: Jay Vosburgh <fubar@us.ibm.com>
Signed-off-by: Jeff 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: Jeff 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: Brian Haley <brian.haley@hp.com>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
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: Wang Chen <wangchen@cn.fujitsu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Fix two compiler warnings that are new with recent versions of gcc
(apparently 4.2 and up). One is fixed by refactoring; this change was
supplied by Stephen Hemminger. The other was fixed by labelling the
variable as uninitialized_var() after confirming via inspection that it
cannot actually be used uninitialized.
Signed-off-by: Jay Vosburgh <fubar@us.ibm.com>
Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
Introduce per-net_device inlines: dev_net(), dev_net_set().
Without CONFIG_NET_NS, no namespace other than &init_net exists.
Let's explicitly define them to help compiler optimizations.
Signed-off-by: YOSHIFUJI Hideaki <yoshfuji@linux-ipv6.org>
alb_fasten_mac_swap (actually rlb_teach_disabled_mac_on_primary)
requries RTNL and no other locks. This could cause dev_set_promiscuity
and/or dev_set_mac_address to be called with improper locking.
Changed callers to hold only RTNL during calls to alb_fasten_mac_swap
or functions calling it. Updated header comments in affected functions to
reflect proper reality of locking requirements.
Signed-off-by: Jay Vosburgh <fubar@us.ibm.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
Move an ASSERT_RTNL down to where we should hold only RTNL;
the existing check produces spurious warnings because we hold additional
locks at _bh, tripping a debug warning in spin_lock_mutex().
Signed-off-by: Jay Vosburgh <fubar@us.ibm.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
Update ALB mode monitor to hold correct locks (RTNL and nothing
else) when calling dev_set_promiscuity.
Signed-off-by: Andy Gospodarek <andy@greyhouse.net>
Signed-off-by: Jay Vosburgh <fubar@us.ibm.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
Convert more lock acquisitions to _bh flavor to avoid deadlock
with workqueue activity and add acquisition of RTNL in appropriate places.
Affects ALB mode, as well as core bonding functions and sysfs.
Signed-off-by: Andy Gospodarek <andy@greyhouse.net>
Signed-off-by: Jay Vosburgh <fubar@us.ibm.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
Convert locking-related activity to new & improved system.
Convert some lock acquisitions to _bh and rework parts of ALB mode, both
to avoid deadlocks with workqueue activity.
Signed-off-by: Andy Gospodarek <andy@greyhouse.net>
Signed-off-by: Jay Vosburgh <fubar@us.ibm.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
Convert bonding timers to workqueues. This converts the various
monitor functions to run in periodic work queues instead of timers. This
patch introduces the framework and convers the calls, but does not resolve
various locking issues, and does not stand alone.
Signed-off-by: Andy Gospodarek <andy@greyhouse.net>
Signed-off-by: Jay Vosburgh <fubar@us.ibm.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
This patch modifies every packet receive function
registered with dev_add_pack() to drop packets if they
are not from the initial network namespace.
This should ensure that the various network stacks do
not receive packets in a anything but the initial network
namespace until the code has been converted and is ready
for them.
Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Renaming skb->h to skb->transport_header, skb->nh to skb->network_header and
skb->mac to skb->mac_header, to match the names of the associated helpers
(skb[_[re]set]_{transport,network,mac}_header).
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Now the skb->nh union has just one member, .raw, i.e. it is just like the
skb->mac union, strange, no? I'm just leaving it like that till the transport
layer is done with, when we'll rename skb->mac.raw to skb->mac_header (or
->mac_header_offset?), ditto for ->{h,nh}.
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
For the places where we need a pointer to the network header, it is still legal
to touch skb->nh.raw directly if just adding to, subtracting from or setting it
to another layer header.
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
For consistency with all the other skb->nh.raw accessors.
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
For the common, open coded 'skb->mac.raw = skb->data' operation, so that we can
later turn skb->mac.raw into a offset, reducing the size of struct sk_buff in
64bit land while possibly keeping it as a pointer on 32bit.
This one touches just the most simple case, next will handle the slightly more
"complex" cases.
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Replace kmalloc() + memset() pairs with the appropriate kzalloc() calls in
the bonding driver.
Signed-off-by: Joe Jin <lkmaillist@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
In bond_alb_monitor the bond->curr_slave_lock write lock is taken
and then dev_set_promiscuity maybe called which can take some time,
depending on the network HW. If a network IRQ for this card come in
the softirq handler maybe try to deliver more packets which end up in
a request to the read lock of bond->curr_slave_lock -> deadlock.
This issue was found by a test lab during network stress tests, this patch
disable the softirq handler for this case and solved the issue.
Signed-off-by: Karsten Keil <kkeil@suse.de>
Acked-by: Jay Vosburgh <fubar@us.ibm.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
I believe I see the race Michael refers to (tlb_choose_channel
may set head, which tlb_init_slave clears), although I was not able to
reproduce it. I have updated his patch for the current netdev-2.6.git
tree and added a version update. His original comment follows:
Our systems have been crashing during testing of PCI HotPlug
support in the various networking components. We've faulted in
the bonding driver due to a bug in bond_alb.c:tlb_clear_slave()
In that routine, the last modification to the TLB hash table is
made without protection of the lock, allowing a race that can lead
tlb_choose_channel() to select an invalid table element.
-J
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
Minor spelling and whitespace corrections.
Signed-off-by: Mitch Williams <mitch.a.williams@intel.com>
Acked-by: Jay Vosburgh <fubar@us.ibm.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Andy Gospodarek