As part of:
commit c2edacf80e
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: Jay Vosburgh <fubar@us.ibm.com>
Signed-off-by: Jeff 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: Pavel Emelyanov <xemul@openvz.org>
Acked-by: Jay Vosburgh <fubar@us.ibm.com>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
If the call to bond_create_sysfs_entry in bond_create fails, the
proper rollback is to call unregister_netdevice, not free_netdev.
Otherwise - kernel BUG at net/core/dev.c:4057!
Checked with artificial failures injected into bond_create_sysfs_entry.
Pavel's original patch modified by Jay Vosburgh to move code around
for clarity (remove goto-hopping within the unwind block).
Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Jay Vosburgh <fubar@us.ibm.com>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
Use proc_create()/proc_create_data() to make sure that ->proc_fops and ->data
be setup before gluing PDE to main tree.
Signed-off-by: Denis V. Lunev <den@openvz.org>
Cc: Jeff Garzik <jgarzik@pobox.com>
Cc: Alexey Dobriyan <adobriyan@gmail.com>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
For bonding interfaces any attempt to read the sysfs directory contents after
module removal results in an oops. The fix is to release sysfs attributes
for the interfaces upon module unload.
Signed-off-by: Libor Pechacek <lpechacek@suse.cz>
Acked-by: Jay Vosburgh <fubar@us.ibm.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
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>
There are some place, that calculate the ARP header length. These
calculations are correct, but
a) some operate with "magic" constants,
b) enlarge the code length (sometimes at the cost of coding style),
c) are not informative from the first glance.
The proposal is to introduce a helper, that includes all the good
sides of these calculations.
Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
ip_fib_init is kept enabled. It is already namespace-aware.
Signed-off-by: Denis V. Lunev <den@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
The ARP monitor functions currently acquire RTNL when performing
failover operations, but do so incorrectly (out of order). This causes
various warnings from might_sleep.
The ARP monitor isn't supported for any of the bonding modes
that actually require RTNL, so it is safe to not hold RTNL when
failing over in the ARP monitor.
Signed-off-by: Jay Vosburgh <fubar@us.ibm.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
I've seen reports of invalid stats in /proc/net/dev for bonding
interfaces, and found it's a pretty easy problem to reproduce. Since
the current code zeros the bonding stats when a read is requested and a
pointer to that data is returned to the caller we cannot guarantee that
the caller has completely accessed the data before a successive call to
request the stats zeroes the stats again.
This patch creates a new stack variable to keep track of the updated
stats and copies the data from that variable into the bonding stats
structure. This ensures that the value for any of the bonding stats
should not incorrectly return zero for any of the bonding statistics.
This does use more stack space and require an extra memcpy, but it seems
like a fair trade-off for consistently correct bonding statistics.
Signed-off-by: Andy Gospodarek <andy@greyhouse.net>
Signed-off-by: Chris Snook <csnook@redhat.com>
Acked-by: Jay Vosburgh <fubar@us.ibm.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Fix the "are we creating a duplicate" check to not compare
the name if the name is NULL (meaning that the system should select
a name). Bug reported by Benny Amorsen <benny+usenet@amorsen.dk>.
Signed-off-by: Jay Vosburgh <fubar@us.ibm.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch eliminates a problem (reported by lockdep) in the
bond_set_multicast_list function. It first reduces the locking on
bond->lock to a simple read_lock, and second, adds netif_tx locking
around the bonding mc_list manipulations that occur outside of the
set_multicast_list function.
The original problem was related to IPv6 addrconf activity.
Signed-off-by: Jay Vosburgh <fubar@us.ibm.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
My last fix (commit ece95f7fef)
didn't handle one case correctly. This resolves that, and it will now
correctly parse parameters with arbitrary white space, and either text
names or mode values.
Signed-off-by: Jay Vosburgh <fubar@us.ibm.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Needed to propagate it down to the ip_route_output_flow.
Signed-off-by: Denis V. Lunev <den@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Change bond_mii_monitor to not hold any locks when calling rtnl_unlock,
as rtnl_unlock can sleep (when acquring another mutex in netdev_run_todo).
Bug reported by Makito SHIOKAWA <mshiokawa@miraclelinux.com>, who
included a different patch.
Signed-off-by: Jay Vosburgh <fubar@us.ibm.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
Fix the handling of rtnl and the bonding_rwsem to always be acquired
in a consistent order (rtnl, then bonding_rwsem).
The existing code sometimes acquired them in this order, and sometimes
in the opposite order, which opens a window for deadlock between ifenslave
and sysfs.
Signed-off-by: Jay Vosburgh <fubar@us.ibm.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
A recent change to add an additional hash policy modified
bond_parse_parm, but it now does not correctly match parameters passed in
via sysfs.
Rewrote bond_parse_parm to handle (a) parameter matches that
are substrings of one another and (b) user input with whitespace (e.g.,
sysfs input often has a trailing newline).
Signed-off-by: Jay Vosburgh <fubar@us.ibm.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
Add a call to bond_release_all in the bonding netdev event
handler for the master. This releases the slaves for the case of, e.g.,
"echo -bond0 > /sys/class/net/bonding_masters", which otherwise will spin
forever waiting for references to be released.
Signed-off-by: Jay Vosburgh <fubar@us.ibm.com>
Signed-off-by: Jeff Garzik <jeff@garzik.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>
Fixes a race condition in module unload. Without this change,
workqueue events may fire while bonding data structures are partially
freed but before bond_close() is invoked by unregister_netdevice().
Update version to 3.2.3.
Signed-off-by: Jay Vosburgh <fubar@us.ibm.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
Add new hash for balance-xor and 802.3ad modes. Originally
submitted by "Glenn Griffin" <ggriffin.kernel@gmail.com>; modified by
Jay Vosburgh to move setting of hash policy out of line, tweak the
documentation update and add version update to 3.2.2.
Glenn's original comment follows:
Included is a patch for a new xmit_hash_policy for the bonding driver
that selects slaves based on MAC and IP information. This is a middle
ground between what currently exists in the layer2 only policy and the
layer3+4 policy. This policy strives to be fully 802.3ad compliant by
transmitting every packet of any particular flow over the same link.
As documented the layer3+4 policy is not fully compliant for extreme
cases such as ip fragmentation, so this policy is a nice compromise
for environments that require full compliance but desire more than the
layer2 only policy.
Signed-off-by: "Glenn Griffin" <ggriffin.kernel@gmail.com>
Signed-off-by: Jay Vosburgh <fubar@us.ibm.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
From: David Sterba <dsterba@suse.cz>
Use macros for comparing jiffies. Jiffies' wrap caused missed events and hangs.
Module reinsert was needed to make bonding work again.
Signed-off-by: David Sterba <dsterba@suse.cz>
Acked-by: Jay Vosburgh <fubar@us.ibm.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
Fix bond_destroy and bond_free_all to not reference the struct
net_device after calling unregister_netdevice.
Bug and offending change reported by Moni Shoua <monis@voltaire.com>
Signed-off-by: Jay Vosburgh <fubar@us.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The standard validate_addr handler refuses to accept the all zeroes address
as valid. However, it's common historical practice for the bonding
master to be configured up prior to having any slaves, at which time the
master will have a MAC address of all zeroes.
Resolved by setting the dev->validate_addr to NULL. The master still can't
end up with an invalid address, as the set_mac_address function tests
for validity.
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 mii (link state) monitor to acquire correct locks for
failover events. In particular, failovers generally require RTNL at a low
level (when manipulating device MAC addresses, for example) and no other
locks. The high level monitor is responsible for acquiring a known set
of locks, RTNL, the bond->lock for read and the slave_lock for write, and
the low level failover processing can then release appropriate locks as
needed. This patch provides the high level portion.
As it is undesirable to acquire RTNL for every monitor pass (which
may occur as often as every 10 ms), the miimon has been converted to
do conditional locking. A first pass inspects all slaves to determine
if any action is required, and if so, a second pass (after acquring RTNL)
is done to perform any actions (doing a complete rescan, as the situation
may have changed when all locks were released).
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>
Change locking in balance-rr transmit processing to use a free
running counter to determine which slave to transmit on. Instead, a
free-running counter is maintained, and modulo arithmetic used to select
a slave for transmit.
This removes lock operations from the TX path, and eliminates
a deadlock introduced by the conversion to work queues.
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>
Two small fixes to IPoIB support for bonding:
1- copy header_ops from slave to bonding for IPoIB slaves
2- move release and destroy logic to UNREGISTER from GOING_DOWN
notifier to avoid double release
Set bonding to version 3.2.1.
Signed-off-by: Moni Shoua <monis at voltaire.com>
Signed-off-by: Jay Vosburgh <fubar@us.ibm.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
Update the "don't change MAC of slaves" functionality added in
previous changes to be a generic option, rather than something tied to
IB devices, as it's occasionally useful for regular ethernet devices as
well.
Adds "fail_over_mac" option (which is automatically enabled for IB
slaves), applicable only to active-backup mode.
Includes documentation update.
Updates bonding driver version to 3.2.0.
Signed-off-by: Jay Vosburgh <fubar@us.ibm.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
When bonding enslaves non Ethernet devices it takes pointers to functions
in the module that owns the slaves. In this case it becomes unsafe
to keep the bonding master registered after last slave was unenslaved
because we don't know if the pointers are still valid. Destroying the bond when slave_cnt is zero
ensures that these functions be used anymore.
Signed-off-by: Moni Shoua <monis at voltaire.com>
Acked-by: Jay Vosburgh <fubar@us.ibm.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
Delay sending a gratuitous_arp when LINK_STATE_LINKWATCH_PENDING bit
in dev->state field is on. This improves the chances for the arp packet to
be transmitted.
Signed-off-by: Moni Shoua <monis at voltaire.com>
Acked-by: Jay Vosburgh <fubar@us.ibm.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
bonding sometimes uses Ethernet constants (such as MTU and address length) which
are not good when it enslaves non Ethernet devices (such as InfiniBand).
Signed-off-by: Moni Shoua <monis at voltaire.com>
Acked-by: Jay Vosburgh <fubar@us.ibm.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
Allow to enslave devices when the bonding device is not up. Over the discussion
held at the previous post this seemed to be the most clean way to go, where it
is not expected to cause instabilities.
Normally, the bonding driver is UP before any enslavement takes place.
Once a netdevice is UP, the network stack acts to have it join some multicast groups
(eg the all-hosts 224.0.0.1). Now, since ether_setup() have set the bonding device
type to be ARPHRD_ETHER and address len to be ETHER_ALEN, the net core code
computes a wrong multicast link address. This is b/c ip_eth_mc_map() is called
where for multicast joins taking place after the enslavement another ip_xxx_mc_map()
is called (eg ip_ib_mc_map() when the bond type is ARPHRD_INFINIBAND)
Signed-off-by: Moni Shoua <monis at voltaire.com>
Signed-off-by: Or Gerlitz <ogerlitz at voltaire.com>
Acked-by: Jay Vosburgh <fubar@us.ibm.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
This patch allows for enslaving netdevices which do not support
the set_mac_address() function. In that case the bond mac address is the one
of the active slave, where remote peers are notified on the mac address
(neighbour) change by Gratuitous ARP sent by bonding when fail-over occurs
(this is already done by the bonding code).
Signed-off-by: Moni Shoua <monis at voltaire.com>
Signed-off-by: Or Gerlitz <ogerlitz at voltaire.com>
Acked-by: Jay Vosburgh <fubar@us.ibm.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
This patch changes some of the bond netdevice attributes and functions
to be that of the active slave for the case of the enslaved device not being
of ARPHRD_ETHER type. Basically it overrides those setting done by ether_setup(),
which are netdevice **type** dependent and hence might be not appropriate for
devices of other types. It also enforces mutual exclusion on bonding slaves
from dissimilar ether types, as was concluded over the v1 discussion.
IPoIB (see Documentation/infiniband/ipoib.txt) MAC address is made of a 3 bytes
IB QP (Queue Pair) number and 16 bytes IB port GID (Global ID) of the port this
IPoIB device is bounded to. The QP is a resource created by the IB HW and the
GID is an identifier burned into the HCA (i have omitted here some details which
are not important for the bonding RFC).
Signed-off-by: Moni Shoua <monis at voltaire.com>
Signed-off-by: Or Gerlitz <ogerlitz at voltaire.com>
Acked-by: Jay Vosburgh <fubar@us.ibm.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
For the operations
get-tx-csum
get-sg
get-tso
get-ufo
the default ethtool_op_xxx behavior is fine for all drivers, so we
permit op==NULL to imply the default behavior.
This provides a more uniform behavior across all drivers, eliminating
ethtool(8) "ioctl not supported" errors on older drivers that had
not been updated for the latest sub-ioctls.
The ethtool_op_xxx() functions are left exported, in case anyone
wishes to call them directly from a driver-private implementation --
a not-uncommon case. Should an ethtool_op_xxx() helper remain unused
for a while, except by net/core/ethtool.c, we can un-export it at a
later date.
[ Resolved conflicts with set/get value ethtool patch... -DaveM ]
Signed-off-by: Jeff Garzik <jeff@garzik.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
It's been a useless no-op for long enough in 2.6 so I figured it's time to
remove it. The number of people that could object because they're
maintaining unified 2.4 and 2.6 drivers is probably rather small.
[ Handled drivers added by netdev tree and some missed IRDA cases... -DaveM ]
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch makes most of the generic device layer network
namespace safe. This patch makes dev_base_head a
network namespace variable, and then it picks up
a few associated variables. The functions:
dev_getbyhwaddr
dev_getfirsthwbytype
dev_get_by_flags
dev_get_by_name
__dev_get_by_name
dev_get_by_index
__dev_get_by_index
dev_ioctl
dev_ethtool
dev_load
wireless_process_ioctl
were modified to take a network namespace argument, and
deal with it.
vlan_ioctl_set and brioctl_set were modified so their
hooks will receive a network namespace argument.
So basically anthing in the core of the network stack that was
affected to by the change of dev_base was modified to handle
multiple network namespaces. The rest of the network stack was
simply modified to explicitly use &init_net the initial network
namespace. This can be fixed when those components of the network
stack are modified to handle multiple network namespaces.
For now the ifindex generator is left global.
Fundametally ifindex numbers are per namespace, or else
we will have corner case problems with migration when
we get that far.
At the same time there are assumptions in the network stack
that the ifindex of a network device won't change. Making
the ifindex number global seems a good compromise until
the network stack can cope with ifindex changes when
you change namespaces, and the like.
Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Every user of the network device notifiers is either a protocol
stack or a pseudo device. If a protocol stack that does not have
support for multiple network namespaces receives an event for a
device that is not in the initial network namespace it quite possibly
can get confused and do the wrong thing.
To avoid problems until all of the protocol stacks are converted
this patch modifies all netdev event handlers to ignore events on
devices that are not in the initial network namespace.
As the rest of the code is made network namespace aware these
checks can be removed.
Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
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>
This patch makes /proc/net per network namespace. It modifies the global
variables proc_net and proc_net_stat to be per network namespace.
The proc_net file helpers are modified to take a network namespace argument,
and all of their callers are fixed to pass &init_net for that argument.
This ensures that all of the /proc/net files are only visible and
usable in the initial network namespace until the code behind them
has been updated to be handle multiple network namespaces.
Making /proc/net per namespace is necessary as at least some files
in /proc/net depend upon the set of network devices which is per
network namespace, and even more files in /proc/net have contents
that are relevant to a single network namespace.
Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
