One point of contention in high network loads is the dst_release() performed
when a transmited skb is freed. This is because NIC tx completion calls
dev_kree_skb() long after original call to dev_queue_xmit(skb).
CPU cache is cold and the atomic op in dst_release() stalls. On SMP, this is
quite visible if one CPU is 100% handling softirqs for a network device,
since dst_clone() is done by other cpus, involving cache line ping pongs.
It seems right place to release dst is in dev_hard_start_xmit(), for most
devices but ones that are virtual, and some exceptions.
David Miller suggested to define a new device flag, set in alloc_netdev_mq()
(so that most devices set it at init time), and carefuly unset in devices
which dont want a NULL skb->dst in their ndo_start_xmit().
List of devices that must clear this flag is :
- loopback device, because it calls netif_rx() and quoting Patrick :
"ip_route_input() doesn't accept loopback addresses, so loopback packets
already need to have a dst_entry attached."
- appletalk/ipddp.c : needs skb->dst in its xmit function
- And all devices that call again dev_queue_xmit() from their xmit function
(as some classifiers need skb->dst) : bonding, vlan, macvlan, eql, ifb, hdlc_fr
Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
struct net_device trans_start field is a hot spot on SMP and high performance
devices, particularly multi queues ones, because every transmitter dirties
it. Is main use is tx watchdog and bonding alive checks.
But as most devices dont use NETIF_F_LLTX, we have to lock
a netdev_queue before calling their ndo_start_xmit(). So it makes
sense to move trans_start from net_device to netdev_queue. Its update
will occur on a already present (and in exclusive state) cache line, for
free.
We can do this transition smoothly. An old driver continue to
update dev->trans_start, while an updated one updates txq->trans_start.
Further patches could also put tx_bytes/tx_packets counters in
netdev_queue to avoid dirtying dev->stats (vlan device comes to mind)
Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
If module initialisation failed (e.g. because the bonding sysfs entry
cannot be created), kernel panics:
IP: [<ffffffff8024910a>] destroy_workqueue+0x2d/0x146
Call Trace:
[<ffffffff808268c4>] bond_destructor+0x28/0x78
[<ffffffff80b64471>] netdev_run_todo+0x231/0x25a
[<ffffffff80b6dbcd>] rtnl_unlock+0x9/0xb
[<ffffffff81567907>] bonding_init+0x83e/0x84a
Remove the calls to bond_work_cancel_all() and destroy_workqueue();
both are also called/scheduled via bond_free_all().
bond_destroy_sysfs is unecessary because the sysfs entry has
not been created in the error case.
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Jay Vosburgh <fubar@us.ibm.com>
Signed-off-by: Jiri Pirko <jpirko@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Remove CONFIG_PROC_FS ifdefs from the code by adding void functions.
Signed-off-by: Jiri Pirko <jpirko@redhat.com>
drivers/net/bonding/bond_main.c | 30 ++++++++++++++++++++----------
1 files changed, 20 insertions(+), 10 deletions(-)
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch makes the cleanup in bond_create nicer :) Also now the forgotten
free_netdev is called.
Signed-off-by: Jiri Pirko <jpirko@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
bond_slave_info_query() should keep a read lock while accessing slave info,
or risk accessing stale data and corruption.
Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Signed-off-by: Jay Vosburgh <fubar@us.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Pointed out by Sean E. Millichamp.
Quote from Documentation/networking/bonding.txt:
"Note that when a bonding interface has no active links, the
driver will immediately reuse the first link that goes up, even if the
updelay parameter has been specified (the updelay is ignored in this
case). If there are slave interfaces waiting for the updelay timeout
to expire, the interface that first went into that state will be
immediately reused. This reduces down time of the network if the
value of updelay has been overestimated, and since this occurs only in
cases with no connectivity, there is no additional penalty for
ignoring the updelay."
This patch actually changes the behaviour in this way.
Signed-off-by: Jiri Pirko <jpirko@redhat.com>
drivers/net/bonding/bond_main.c | 8 ++++++++
1 files changed, 8 insertions(+), 0 deletions(-)
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch only changes the order of interfaces to use for checking slave link
status in bond_check_dev_link() to priorize ethtool interface. Should safe some
troubles as ethtool seems to be more supported.
Jirka
Signed-off-by: Jiri Pirko <jpirko@redhat.com>
drivers/net/bonding/bond_main.c | 26 ++++++++++++--------------
1 files changed, 12 insertions(+), 14 deletions(-)
Signed-off-by: David S. Miller <davem@davemloft.net>
Fix a zero address hole bug in the bonding arp_ip_target list
that was causing the bond to ignore ARP replies (bugz 13006).
Instead of just setting the array entry to zero, we now
copy any additional entries down one slot, putting the
zero entry at the end. With this change we can now have
all the loops that walk the array stop when they hit a zero
since there will be no addresses after it.
Changes are based in part on code fragment provided in kernel:
bugzilla 13006:
http://bugzilla.kernel.org/show_bug.cgi?id=13006
by Steve Howard <steve@astutenetworks.com>
Signed-off-by: Brian Haley <brian.haley@hp.com>
Signed-off-by: Jay Vosburgh <fubar@us.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Setting ->owner as done currently (pde->owner = THIS_MODULE) is racy
as correctly noted at bug #12454. Someone can lookup entry with NULL
->owner, thus not pinning enything, and release it later resulting
in module refcount underflow.
We can keep ->owner and supply it at registration time like ->proc_fops
and ->data.
But this leaves ->owner as easy-manipulative field (just one C assignment)
and somebody will forget to unpin previous/pin current module when
switching ->owner. ->proc_fops is declared as "const" which should give
some thoughts.
->read_proc/->write_proc were just fixed to not require ->owner for
protection.
rmmod'ed directories will be empty and return "." and ".." -- no harm.
And directories with tricky enough readdir and lookup shouldn't be modular.
We definitely don't want such modular code.
Removing ->owner will also make PDE smaller.
So, let's nuke it.
Kudos to Jeff Layton for reminding about this, let's say, oversight.
http://bugzilla.kernel.org/show_bug.cgi?id=12454
Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
I've hit an issue on my system when I've been using RealTek RTL8139D cards in
bonding interface in mode balancing-alb. When I enslave a card, the current
active slave (bond->curr_active_slave) is not set and the link is therefore
not functional.
----
# cat /proc/net/bonding/bond0
Ethernet Channel Bonding Driver: v3.5.0 (November 4, 2008)
Bonding Mode: adaptive load balancing
Primary Slave: None
Currently Active Slave: None
MII Status: up
MII Polling Interval (ms): 100
Up Delay (ms): 0
Down Delay (ms): 0
Slave Interface: eth1
MII Status: up
Link Failure Count: 0
Permanent HW addr: 00:1f:1f:01:2f:22
----
The thing that gets it right is when I unplug the cable and then I put it back
into the NIC. Then the current active slave is set to eth1 and link is working
just fine. Here is dmesg log with bonding DEBUG messages turned on:
----
ADDRCONF(NETDEV_UP): bond0: link is not ready
event_dev: bond0, event: 1
IFF_MASTER
event_dev: bond0, event: 8
IFF_MASTER
bond_ioctl: master=bond0, cmd=35216
slave_dev=cac5d800:
slave_dev->name=eth1:
eth1: ! NETIF_F_VLAN_CHALLENGED
event_dev: eth1, event: 8
eth1: link up, 100Mbps, full-duplex, lpa 0xC5E1
event_dev: eth1, event: 1
event_dev: eth1, event: 8
IFF_SLAVE
Initial state of slave_dev is BOND_LINK_UP
bonding: bond0: enslaving eth1 as an active interface with an up link.
ADDRCONF(NETDEV_CHANGE): bond0: link becomes ready
event_dev: bond0, event: 4
IFF_MASTER
bond0: no IPv6 routers present
<<<<cable unplug>>>>
eth1: link down
event_dev: eth1, event: 4
IFF_SLAVE
bonding: bond0: link status definitely down for interface eth1, disabling it
event_dev: bond0, event: 4
IFF_MASTER
<<<<cable plug>>>>
eth1: link up, 100Mbps, full-duplex, lpa 0xC5E1
event_dev: eth1, event: 4
IFF_SLAVE
bonding: bond0: link status definitely up for interface eth1.
bonding: bond0: making interface eth1 the new active one.
event_dev: eth1, event: 8
IFF_SLAVE
event_dev: eth1, event: 8
IFF_SLAVE
bonding: bond0: first active interface up!
event_dev: bond0, event: 4
IFF_MASTER
----
The current active slave is set by calling bond_select_active_slave() function
from bond_miimon_commit() function when the slave (eth1) link goes to state up.
I also tested this on other machine with Broadcom NetXtreme II BCM5708
1000Base-T NIC and there all works fine. The thing is that this adapter is down
and goes up after few seconds after it is enslaved.
This patch calls bond_select_active_slave() in bond_enslave() function for modes
alb and tlb and makes sure that the current active slave is set up properly even
when the slave state is already up. Tested on both systems, works fine.
Notice: The same problem can maybe also occrur in mode 8023AD but I'm unable to
test that.
Signed-off-by: Jiri Pirko <jpirko@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch corrects an omission from the following commit:
commit f0c76d6177
Author: Jay Vosburgh <fubar@us.ibm.com>
Date: Wed Jul 2 18:21:58 2008 -0700
bonding: refactor mii monitor
The un-refactored code checked the link speed and duplex of
every slave on every pass; the refactored code did not do so.
The 802.3ad and balance-alb/tlb modes utilize the speed and
duplex information, and require it to be kept up to date. This patch
adds a notifier check to perform the appropriate updating when the slave
device speed changes.
Signed-off-by: Jay Vosburgh <fubar@us.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Impact: Rename function scope variable.
Fix this sparse warning:
drivers/net/bonding/bond_main.c:4704:13: warning: symbol 'mode' shadows an earlier one
drivers/net/bonding/bond_main.c:95:13: originally declared here
Signed-off-by: Hannes Eder <hannes@hanneseder.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
Use the correct pointer in debug message.
Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Fix this sparse warnings:
drivers/net/bonding/bond_main.c:104:20: warning: symbol 'bonding_defaults' was not declared. Should it be static?
drivers/net/bonding/bond_main.c:204:22: warning: symbol 'ad_select_tbl' was not declared. Should it be static?
drivers/net/bonding/bond_sysfs.c:60:21: warning: symbol 'bonding_rwsem' was not declared. Should it be static?
Signed-off-by: Hannes Eder <hannes@hanneseder.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
bond_parse_parm() parses a parameter table for a particular value and
is therefore not modifying the table at all. Therefore make the 2nd
argument const, thus allowing to make the tables const later.
Signed-off-by: Holger Eitzenberger <holger@eitzenberger.org>
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>
Use a small array in bond_mode_name() for the names, thus saving some
space:
before
text data bss dec hex filename
57736 9372 344 67452 1077c drivers/net/bonding/bonding.ko
after
text data bss dec hex filename
57441 9372 344 67157 10655 drivers/net/bonding/bonding.ko
Signed-off-by: Holger Eitzenberger <holger@eitzenberger.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Introduce and use bond_is_lb(), it is usefull to shorten the repetitive
check for either ALB or TLB mode.
Signed-off-by: Holger Eitzenberger <holger@eitzenberger.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Simply replace netdev->priv with netdev_priv().
Signed-off-by: Wang Chen <wangchen@cn.fujitsu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch moves neigh_setup and hard_start_xmit into the network device ops
structure. For bisection, fix all the previously converted drivers as well.
Bonding driver took the biggest hit on this.
Added a prefetch of the hard_start_xmit in the fast path to try and reduce
any impact this would have.
Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
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>
In order for the network device ops get_stats call to be immutable, the handling
of the default internal network device stats block has to be changed. Add a new
helper function which replaces the old use of internal_get_stats.
Note: change return code to make it clear that the caller should not
go changing the returned statistics.
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>
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: Jay Vosburgh <fubar@us.ibm.com>
Signed-off-by: Jeff 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: Brian Haley <brian.haley@hp.com>
Signed-off-by: Jay Vosburgh <fubar@us.ibm.com>
Signed-off-by: Jeff 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: Jay Vosburgh <fubar@us.ibm.com>
Signed-off-by: David 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: Harvey Harrison <harvey.harrison@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
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: Jay Vosburgh <fubar@us.ibm.com>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
During the rework of the mii monitor for:
commit f0c76d6177
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: Jay Vosburgh <fubar@us.ibm.com>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
As a bonus, removes some unnecessary byteswapping.
Signed-off-by: Harvey Harrison <harvey.harrison@gmail.com>
Signed-off-by: David 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: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
My change
commit e2a6b85247
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: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch allows reporting the link, checksum, and feature settings
of bonded device by using generic hooks.
Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
Acked-by: Jay Vosburgh <fubar@us.ibm.com>
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>
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: Jay Vosburgh <fubar@us.ibm.com>
Signed-off-by: Jeff 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: David 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: David 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: David 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: David S. Miller <davem@davemloft.net>
