bonding: prepare to and add arp_all_targets

Hi,

This patchset does some cleanup, fixes a few flaws and adds a new parameter to control whether the slave should stay up when any of arp_ip_targets are up or all of them are up (i.e. if the slave should go down if even one target from arp_ip_targets fail).

The patches 1-5 basically prepare the bonding code to work properly with this new parameter, however it's not depending on them.

All comments are really welcome.

v3->v4:

Fix "[6/6] bonding: add helper function bond_get_targets_ip(targets, ip)", per Nikolay's advice and kill that lonely rtnl_unlock(). As Jay suggested, rephrase the documentation part to be more clear, don't fail setting arp_all_targets if arp_validate is not set - it has no effect anyway but can be easier to set up. Jay also spotted that there's no need to print warning if the last arp_target is removed and arp_validate is on, so change it to be paired with arp_interval warning - if the arp_interval is on and we delete the last arp_target - print a warning.

Fix "[4/6] bonding: don't trust arp requests unless active slave really works", per Sergei's suggestion - split the lines to make checkpatch.pl happy.

v2->v3:

Fix "[6/6] bonding: add helper function bond_get_targets_ip(targets, ip)", per Nikolay's advice, to use _bh spinlock, remove useless rtnl_lock() and use jiffies for new arp_ip_target last arp, instead of slave_last_rx(). As Michal suggested, initialize the target_last_arp_rx[] the same as the default last_arp_rx, to avoid flapping. Also, don't fail while removing the last arp_ip_target, and instead just print a warning - otherwise it might break existing scripts.

v1->v2:

Drop "[1/6] bonding: verify if bond has ip only once on arp validate", per Jay's advice, it's really useless as a standalone patch (though the change itself will appear in 2/6).

Fix "[2/6] bonding: add helper function bond_get_targets_ip(targets, ip)", per Nikolay's advice, to verify if source ip != 0.0.0.0, otherwise we might update 'null' arp_ip_targets' last_rx. Also, address style.

Drop "[5/6] bonding: don't swap arp's ips on validation for backup slave", per Jay's advice, though it does address cyclic failover issue - it disables a good optimization. Instead, update the documentation and fix the cyclic loop issue in the following patch.

Add "bonding: don't trust arp requests unless active slave really works", this patch fixes the cyclic loop that occurs when we can't reach arp_ip_target and fool ourselves with our own generated arp requests. It's done by checking if we currently have a current active slave and if this slave received at least one arp reply after its 'nomination' to active.

Add "bonding: doc: some details on backup slave arp validation", this patch describes a bit more in detail how the backup slave arp validation works.

Fix "[6/6] bonding: add an option to fail when any of arp_ip_target is inaccessible", per Nikolay's advice, to correctly handle adding/removing hosts in arp_ip_target - we need to shift/initialize all slave's target_last_arp_rx. Also, don't fail module loading on arp_all_targets misconfiguration, just disable it, and some minor style fixes.

Thanks!

Veaceslav Falico (6):
bonding: add helper function bond_get_targets_ip(targets, ip)
bonding: don't add duplicate targets to arp_ip_target
bonding: don't validate arp if we don't have to
bonding: don't trust arp requests unless active slave really works
bonding: doc: some details on backup slave arp validation
bonding: add an option to fail when any of arp_ip_target is inaccessible

 Documentation/networking/bonding.txt |   23 ++++++
 drivers/net/bonding/bond_main.c      |   70 +++++++++++++++----
 drivers/net/bonding/bond_sysfs.c     |  125 ++++++++++++++++++++++++----------
 drivers/net/bonding/bonding.h        |   46 ++++++++++++-
 4 files changed, 212 insertions(+), 52 deletions(-)