|
Jiri Slaby |
b11903 |
From: Jay Vosburgh <jay.vosburgh@canonical.com>
|
|
Jiri Slaby |
b11903 |
Date: Tue, 7 Nov 2017 19:50:07 +0900
|
|
Jiri Slaby |
b11903 |
Subject: bonding: fix slave stuck in BOND_LINK_FAIL state
|
|
Jiri Slaby |
b11903 |
Git-commit: 055db6957e4735b16cd2fa94a5bbfb754c9b8023
|
|
Jiri Slaby |
b11903 |
Patch-mainline: 4.14
|
|
Jiri Slaby |
b11903 |
References: networking-stable-19_11_10
|
|
Jiri Slaby |
b11903 |
|
|
Jiri Slaby |
b11903 |
The bonding miimon logic has a flaw, in that a failure of the
|
|
Jiri Slaby |
b11903 |
rtnl_trylock can cause a slave to become permanently stuck in
|
|
Jiri Slaby |
b11903 |
BOND_LINK_FAIL state.
|
|
Jiri Slaby |
b11903 |
|
|
Jiri Slaby |
b11903 |
The sequence of events to cause this is as follows:
|
|
Jiri Slaby |
b11903 |
|
|
Jiri Slaby |
b11903 |
1) bond_miimon_inspect finds that a slave's link is down, and so
|
|
Jiri Slaby |
b11903 |
calls bond_propose_link_state, setting slave->new_link_state to
|
|
Jiri Slaby |
b11903 |
BOND_LINK_FAIL, then sets slave->new_link to BOND_LINK_DOWN and returns
|
|
Jiri Slaby |
b11903 |
non-zero.
|
|
Jiri Slaby |
b11903 |
|
|
Jiri Slaby |
b11903 |
2) In bond_mii_monitor, the rtnl_trylock fails, and the timer is
|
|
Jiri Slaby |
b11903 |
rescheduled. No change is committed.
|
|
Jiri Slaby |
b11903 |
|
|
Jiri Slaby |
b11903 |
3) bond_miimon_inspect is called again, but this time the slave
|
|
Jiri Slaby |
b11903 |
from step 1 has recovered. slave->new_link is reset to NOCHANGE, and, as
|
|
Jiri Slaby |
b11903 |
slave->link was never changed, the switch enters the BOND_LINK_UP case,
|
|
Jiri Slaby |
b11903 |
and does nothing. The pending BOND_LINK_FAIL state from step 1 remains
|
|
Jiri Slaby |
b11903 |
pending, as new_link_state is not reset.
|
|
Jiri Slaby |
b11903 |
|
|
Jiri Slaby |
b11903 |
4) The state from step 3 persists until another slave changes link
|
|
Jiri Slaby |
b11903 |
state and causes bond_miimon_inspect to return non-zero. At this point,
|
|
Jiri Slaby |
b11903 |
the BOND_LINK_FAIL state change on the slave from steps 1-3 is committed,
|
|
Jiri Slaby |
b11903 |
and the slave will remain stuck in BOND_LINK_FAIL state even though it
|
|
Jiri Slaby |
b11903 |
is actually link up.
|
|
Jiri Slaby |
b11903 |
|
|
Jiri Slaby |
b11903 |
The remedy for this is to initialize new_link_state on each entry
|
|
Jiri Slaby |
b11903 |
to bond_miimon_inspect, as is already done with new_link.
|
|
Jiri Slaby |
b11903 |
|
|
Jiri Slaby |
b11903 |
Fixes: fb9eb899a6dc ("bonding: handle link transition from FAIL to UP correctly")
|
|
Jiri Slaby |
b11903 |
Reported-by: Alex Sidorenko <alexandre.sidorenko@hpe.com>
|
|
Jiri Slaby |
b11903 |
Reviewed-by: Jarod Wilson <jarod@redhat.com>
|
|
Jiri Slaby |
b11903 |
Signed-off-by: Jay Vosburgh <jay.vosburgh@canonical.com>
|
|
Jiri Slaby |
b11903 |
Acked-by: Mahesh Bandewar <maheshb@google.com>
|
|
Jiri Slaby |
b11903 |
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Jiri Slaby |
b11903 |
Signed-off-by: Jiri Slaby <jslaby@suse.cz>
|
|
Jiri Slaby |
b11903 |
---
|
|
Jiri Slaby |
b11903 |
drivers/net/bonding/bond_main.c | 1 +
|
|
Jiri Slaby |
b11903 |
1 file changed, 1 insertion(+)
|
|
Jiri Slaby |
b11903 |
|
|
Jiri Slaby |
b11903 |
--- a/drivers/net/bonding/bond_main.c
|
|
Jiri Slaby |
b11903 |
+++ b/drivers/net/bonding/bond_main.c
|
|
Jiri Slaby |
b11903 |
@@ -2039,6 +2039,7 @@ static int bond_miimon_inspect(struct bo
|
|
Jiri Slaby |
b11903 |
|
|
Jiri Slaby |
b11903 |
bond_for_each_slave_rcu(bond, slave, iter) {
|
|
Jiri Slaby |
b11903 |
slave->new_link = BOND_LINK_NOCHANGE;
|
|
Jiri Slaby |
b11903 |
+ slave->link_new_state = slave->link;
|
|
Jiri Slaby |
b11903 |
|
|
Jiri Slaby |
b11903 |
link_state = bond_check_dev_link(bond, slave->dev, 0);
|
|
Jiri Slaby |
b11903 |
|