2005-04-17 06:20:36 +08:00
|
|
|
/*
|
|
|
|
* Copyright(c) 1999 - 2004 Intel Corporation. All rights reserved.
|
|
|
|
*
|
|
|
|
* This program is free software; you can redistribute it and/or modify it
|
|
|
|
* under the terms of the GNU General Public License as published by the Free
|
|
|
|
* Software Foundation; either version 2 of the License, or (at your option)
|
|
|
|
* any later version.
|
|
|
|
*
|
|
|
|
* This program is distributed in the hope that it will be useful, but WITHOUT
|
|
|
|
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
|
|
|
|
* FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
|
|
|
|
* more details.
|
|
|
|
*
|
|
|
|
* You should have received a copy of the GNU General Public License along with
|
|
|
|
* this program; if not, write to the Free Software Foundation, Inc., 59
|
|
|
|
* Temple Place - Suite 330, Boston, MA 02111-1307, USA.
|
|
|
|
*
|
|
|
|
* The full GNU General Public License is included in this distribution in the
|
|
|
|
* file called LICENSE.
|
|
|
|
*
|
|
|
|
*/
|
|
|
|
|
|
|
|
#include <linux/skbuff.h>
|
|
|
|
#include <linux/if_ether.h>
|
|
|
|
#include <linux/netdevice.h>
|
|
|
|
#include <linux/spinlock.h>
|
|
|
|
#include <linux/ethtool.h>
|
2008-11-05 09:51:16 +08:00
|
|
|
#include <linux/etherdevice.h>
|
2005-04-17 06:20:36 +08:00
|
|
|
#include <linux/if_bonding.h>
|
|
|
|
#include <linux/pkt_sched.h>
|
2007-09-18 02:53:39 +08:00
|
|
|
#include <net/net_namespace.h>
|
2014-11-11 02:27:49 +08:00
|
|
|
#include <net/bonding.h>
|
|
|
|
#include <net/bond_3ad.h>
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2014-01-08 23:46:46 +08:00
|
|
|
/* General definitions */
|
2005-04-17 06:20:36 +08:00
|
|
|
#define AD_SHORT_TIMEOUT 1
|
|
|
|
#define AD_LONG_TIMEOUT 0
|
|
|
|
#define AD_STANDBY 0x2
|
|
|
|
#define AD_MAX_TX_IN_SECOND 3
|
|
|
|
#define AD_COLLECTOR_MAX_DELAY 0
|
|
|
|
|
2014-01-08 23:46:46 +08:00
|
|
|
/* Timer definitions (43.4.4 in the 802.3ad standard) */
|
2005-04-17 06:20:36 +08:00
|
|
|
#define AD_FAST_PERIODIC_TIME 1
|
|
|
|
#define AD_SLOW_PERIODIC_TIME 30
|
|
|
|
#define AD_SHORT_TIMEOUT_TIME (3*AD_FAST_PERIODIC_TIME)
|
|
|
|
#define AD_LONG_TIMEOUT_TIME (3*AD_SLOW_PERIODIC_TIME)
|
|
|
|
#define AD_CHURN_DETECTION_TIME 60
|
|
|
|
#define AD_AGGREGATE_WAIT_TIME 2
|
|
|
|
|
2014-01-08 23:46:46 +08:00
|
|
|
/* Port state definitions (43.4.2.2 in the 802.3ad standard) */
|
2005-04-17 06:20:36 +08:00
|
|
|
#define AD_STATE_LACP_ACTIVITY 0x1
|
|
|
|
#define AD_STATE_LACP_TIMEOUT 0x2
|
|
|
|
#define AD_STATE_AGGREGATION 0x4
|
|
|
|
#define AD_STATE_SYNCHRONIZATION 0x8
|
|
|
|
#define AD_STATE_COLLECTING 0x10
|
|
|
|
#define AD_STATE_DISTRIBUTING 0x20
|
|
|
|
#define AD_STATE_DEFAULTED 0x40
|
|
|
|
#define AD_STATE_EXPIRED 0x80
|
|
|
|
|
2014-01-08 23:46:46 +08:00
|
|
|
/* Port Variables definitions used by the State Machines (43.4.7 in the
|
|
|
|
* 802.3ad standard)
|
|
|
|
*/
|
2005-04-17 06:20:36 +08:00
|
|
|
#define AD_PORT_BEGIN 0x1
|
|
|
|
#define AD_PORT_LACP_ENABLED 0x2
|
|
|
|
#define AD_PORT_ACTOR_CHURN 0x4
|
|
|
|
#define AD_PORT_PARTNER_CHURN 0x8
|
|
|
|
#define AD_PORT_READY 0x10
|
|
|
|
#define AD_PORT_READY_N 0x20
|
|
|
|
#define AD_PORT_MATCHED 0x40
|
|
|
|
#define AD_PORT_STANDBY 0x80
|
|
|
|
#define AD_PORT_SELECTED 0x100
|
|
|
|
#define AD_PORT_MOVED 0x200
|
2015-03-31 05:30:40 +08:00
|
|
|
#define AD_PORT_CHURNED (AD_PORT_ACTOR_CHURN | AD_PORT_PARTNER_CHURN)
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2014-01-08 23:46:46 +08:00
|
|
|
/* Port Key definitions
|
|
|
|
* key is determined according to the link speed, duplex and
|
|
|
|
* user key (which is yet not supported)
|
2015-05-09 15:01:57 +08:00
|
|
|
* --------------------------------------------------------------
|
|
|
|
* Port key | User key (10 bits) | Speed (5 bits) | Duplex|
|
|
|
|
* --------------------------------------------------------------
|
|
|
|
* |15 6|5 1|0
|
2014-01-08 23:46:46 +08:00
|
|
|
*/
|
2014-11-19 16:48:58 +08:00
|
|
|
#define AD_DUPLEX_KEY_MASKS 0x1
|
|
|
|
#define AD_SPEED_KEY_MASKS 0x3E
|
|
|
|
#define AD_USER_KEY_MASKS 0xFFC0
|
|
|
|
|
|
|
|
enum ad_link_speed_type {
|
|
|
|
AD_LINK_SPEED_1MBPS = 1,
|
|
|
|
AD_LINK_SPEED_10MBPS,
|
|
|
|
AD_LINK_SPEED_100MBPS,
|
|
|
|
AD_LINK_SPEED_1000MBPS,
|
2014-11-19 16:48:59 +08:00
|
|
|
AD_LINK_SPEED_2500MBPS,
|
|
|
|
AD_LINK_SPEED_10000MBPS,
|
|
|
|
AD_LINK_SPEED_20000MBPS,
|
|
|
|
AD_LINK_SPEED_40000MBPS,
|
2015-12-03 19:12:05 +08:00
|
|
|
AD_LINK_SPEED_56000MBPS,
|
|
|
|
AD_LINK_SPEED_100000MBPS,
|
2014-11-19 16:48:58 +08:00
|
|
|
};
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2014-01-02 09:12:54 +08:00
|
|
|
/* compare MAC addresses */
|
|
|
|
#define MAC_ADDRESS_EQUAL(A, B) \
|
|
|
|
ether_addr_equal_64bits((const u8 *)A, (const u8 *)B)
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2016-06-30 22:13:41 +08:00
|
|
|
static const u8 null_mac_addr[ETH_ALEN + 2] __long_aligned = {
|
|
|
|
0, 0, 0, 0, 0, 0
|
|
|
|
};
|
2005-04-17 06:20:36 +08:00
|
|
|
static u16 ad_ticks_per_sec;
|
|
|
|
static const int ad_delta_in_ticks = (AD_TIMER_INTERVAL * HZ) / 1000;
|
|
|
|
|
2016-06-30 22:13:41 +08:00
|
|
|
static const u8 lacpdu_mcast_addr[ETH_ALEN + 2] __long_aligned =
|
|
|
|
MULTICAST_LACPDU_ADDR;
|
2008-12-27 05:40:48 +08:00
|
|
|
|
2014-01-08 23:46:46 +08:00
|
|
|
/* ================= main 802.3ad protocol functions ================== */
|
2005-04-17 06:20:36 +08:00
|
|
|
static int ad_lacpdu_send(struct port *port);
|
2007-10-19 14:41:04 +08:00
|
|
|
static int ad_marker_send(struct port *port, struct bond_marker *marker);
|
2014-10-05 08:45:01 +08:00
|
|
|
static void ad_mux_machine(struct port *port, bool *update_slave_arr);
|
2005-04-17 06:20:36 +08:00
|
|
|
static void ad_rx_machine(struct lacpdu *lacpdu, struct port *port);
|
|
|
|
static void ad_tx_machine(struct port *port);
|
|
|
|
static void ad_periodic_machine(struct port *port);
|
2014-10-05 08:45:01 +08:00
|
|
|
static void ad_port_selection_logic(struct port *port, bool *update_slave_arr);
|
|
|
|
static void ad_agg_selection_logic(struct aggregator *aggregator,
|
|
|
|
bool *update_slave_arr);
|
2005-04-17 06:20:36 +08:00
|
|
|
static void ad_clear_agg(struct aggregator *aggregator);
|
|
|
|
static void ad_initialize_agg(struct aggregator *aggregator);
|
|
|
|
static void ad_initialize_port(struct port *port, int lacp_fast);
|
2014-10-05 08:45:01 +08:00
|
|
|
static void ad_enable_collecting_distributing(struct port *port,
|
|
|
|
bool *update_slave_arr);
|
|
|
|
static void ad_disable_collecting_distributing(struct port *port,
|
|
|
|
bool *update_slave_arr);
|
2014-01-08 23:46:46 +08:00
|
|
|
static void ad_marker_info_received(struct bond_marker *marker_info,
|
|
|
|
struct port *port);
|
|
|
|
static void ad_marker_response_received(struct bond_marker *marker,
|
|
|
|
struct port *port);
|
2015-11-01 03:45:06 +08:00
|
|
|
static void ad_update_actor_keys(struct port *port, bool reset);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
|
2014-01-08 23:46:46 +08:00
|
|
|
/* ================= api to bonding and kernel code ================== */
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
/**
|
|
|
|
* __get_bond_by_port - get the port's bonding struct
|
|
|
|
* @port: the port we're looking at
|
|
|
|
*
|
|
|
|
* Return @port's bonding struct, or %NULL if it can't be found.
|
|
|
|
*/
|
|
|
|
static inline struct bonding *__get_bond_by_port(struct port *port)
|
|
|
|
{
|
2010-10-17 04:19:59 +08:00
|
|
|
if (port->slave == NULL)
|
2005-04-17 06:20:36 +08:00
|
|
|
return NULL;
|
|
|
|
|
|
|
|
return bond_get_bond_by_slave(port->slave);
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* __get_first_agg - get the first aggregator in the bond
|
|
|
|
* @bond: the bond we're looking at
|
|
|
|
*
|
|
|
|
* Return the aggregator of the first slave in @bond, or %NULL if it can't be
|
|
|
|
* found.
|
2014-01-10 18:59:44 +08:00
|
|
|
* The caller must hold RCU or RTNL lock.
|
2005-04-17 06:20:36 +08:00
|
|
|
*/
|
|
|
|
static inline struct aggregator *__get_first_agg(struct port *port)
|
|
|
|
{
|
|
|
|
struct bonding *bond = __get_bond_by_port(port);
|
2013-08-01 22:54:47 +08:00
|
|
|
struct slave *first_slave;
|
2014-01-10 18:59:44 +08:00
|
|
|
struct aggregator *agg;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
bonding: add RCU for bond_3ad_state_machine_handler()
The bond_3ad_state_machine_handler() use the bond lock to protect
the bond slave list and slave port together, but it is not enough,
the bond slave list was link and unlink in RTNL, not bond lock,
so I add RCU to protect the slave list from leaving.
The bond lock is still used here, because when the slave has been
removed from the list by the time the state machine runs, it appears
to be possible for both function to manupulate the same aggregator->lag_ports
by finding the aggregator via two different ports that are both members of
that aggregator (i.e., port A of the agg is being unbound, and port B
of the agg is runing its state machine).
If I remove the bond lock, there are nothing to mutex changes
to aggregator->lag_ports between bond_3ad_state_machine_handler and
bond_3ad_unbind_slave, So the bond lock is the simplest way to protect
aggregator->lag_ports.
There was a lot of function need RCU protect, I have two choice
to make the function in RCU-safe, (1) create new similar functions
and make the bond slave list in RCU. (2) modify the existed functions
and make them in read-side critical section, because the RCU
read-side critical sections may be nested.
I choose (2) because it is no need to create more similar functions.
The nots in the function is still too old, clean up the nots.
Suggested-by: Nikolay Aleksandrov <nikolay@redhat.com>
Suggested-by: Jay Vosburgh <fubar@us.ibm.com>
Suggested-by: Veaceslav Falico <vfalico@redhat.com>
Signed-off-by: Ding Tianhong <dingtianhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-12-13 10:20:12 +08:00
|
|
|
/* If there's no bond for this port, or bond has no slaves */
|
2013-08-01 22:54:47 +08:00
|
|
|
if (bond == NULL)
|
2005-04-17 06:20:36 +08:00
|
|
|
return NULL;
|
2014-01-08 23:46:46 +08:00
|
|
|
|
bonding: add RCU for bond_3ad_state_machine_handler()
The bond_3ad_state_machine_handler() use the bond lock to protect
the bond slave list and slave port together, but it is not enough,
the bond slave list was link and unlink in RTNL, not bond lock,
so I add RCU to protect the slave list from leaving.
The bond lock is still used here, because when the slave has been
removed from the list by the time the state machine runs, it appears
to be possible for both function to manupulate the same aggregator->lag_ports
by finding the aggregator via two different ports that are both members of
that aggregator (i.e., port A of the agg is being unbound, and port B
of the agg is runing its state machine).
If I remove the bond lock, there are nothing to mutex changes
to aggregator->lag_ports between bond_3ad_state_machine_handler and
bond_3ad_unbind_slave, So the bond lock is the simplest way to protect
aggregator->lag_ports.
There was a lot of function need RCU protect, I have two choice
to make the function in RCU-safe, (1) create new similar functions
and make the bond slave list in RCU. (2) modify the existed functions
and make them in read-side critical section, because the RCU
read-side critical sections may be nested.
I choose (2) because it is no need to create more similar functions.
The nots in the function is still too old, clean up the nots.
Suggested-by: Nikolay Aleksandrov <nikolay@redhat.com>
Suggested-by: Jay Vosburgh <fubar@us.ibm.com>
Suggested-by: Veaceslav Falico <vfalico@redhat.com>
Signed-off-by: Ding Tianhong <dingtianhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-12-13 10:20:12 +08:00
|
|
|
rcu_read_lock();
|
|
|
|
first_slave = bond_first_slave_rcu(bond);
|
2014-05-12 15:08:43 +08:00
|
|
|
agg = first_slave ? &(SLAVE_AD_INFO(first_slave)->aggregator) : NULL;
|
bonding: add RCU for bond_3ad_state_machine_handler()
The bond_3ad_state_machine_handler() use the bond lock to protect
the bond slave list and slave port together, but it is not enough,
the bond slave list was link and unlink in RTNL, not bond lock,
so I add RCU to protect the slave list from leaving.
The bond lock is still used here, because when the slave has been
removed from the list by the time the state machine runs, it appears
to be possible for both function to manupulate the same aggregator->lag_ports
by finding the aggregator via two different ports that are both members of
that aggregator (i.e., port A of the agg is being unbound, and port B
of the agg is runing its state machine).
If I remove the bond lock, there are nothing to mutex changes
to aggregator->lag_ports between bond_3ad_state_machine_handler and
bond_3ad_unbind_slave, So the bond lock is the simplest way to protect
aggregator->lag_ports.
There was a lot of function need RCU protect, I have two choice
to make the function in RCU-safe, (1) create new similar functions
and make the bond slave list in RCU. (2) modify the existed functions
and make them in read-side critical section, because the RCU
read-side critical sections may be nested.
I choose (2) because it is no need to create more similar functions.
The nots in the function is still too old, clean up the nots.
Suggested-by: Nikolay Aleksandrov <nikolay@redhat.com>
Suggested-by: Jay Vosburgh <fubar@us.ibm.com>
Suggested-by: Veaceslav Falico <vfalico@redhat.com>
Signed-off-by: Ding Tianhong <dingtianhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-12-13 10:20:12 +08:00
|
|
|
rcu_read_unlock();
|
2014-01-08 23:46:46 +08:00
|
|
|
|
2014-01-10 18:59:44 +08:00
|
|
|
return agg;
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
|
2014-01-08 23:46:46 +08:00
|
|
|
/**
|
|
|
|
* __agg_has_partner - see if we have a partner
|
|
|
|
* @agg: the agregator we're looking at
|
2008-11-05 09:51:16 +08:00
|
|
|
*
|
|
|
|
* Return nonzero if aggregator has a partner (denoted by a non-zero ether
|
2014-01-08 23:46:46 +08:00
|
|
|
* address for the partner). Return 0 if not.
|
2008-11-05 09:51:16 +08:00
|
|
|
*/
|
|
|
|
static inline int __agg_has_partner(struct aggregator *agg)
|
|
|
|
{
|
|
|
|
return !is_zero_ether_addr(agg->partner_system.mac_addr_value);
|
|
|
|
}
|
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
/**
|
|
|
|
* __disable_port - disable the port's slave
|
|
|
|
* @port: the port we're looking at
|
|
|
|
*/
|
|
|
|
static inline void __disable_port(struct port *port)
|
|
|
|
{
|
bonding: Fix RTNL: assertion failed at net/core/rtnetlink.c for 802.3ad mode
The problem was introduced by the commit 1d3ee88ae0d
(bonding: add netlink attributes to slave link dev).
The bond_set_active_slave() and bond_set_backup_slave()
will use rtmsg_ifinfo to send slave's states, so these
two functions should be called in RTNL.
In 802.3ad mode, acquiring RTNL for the __enable_port and
__disable_port cases is difficult, as those calls generally
already hold the state machine lock, and cannot unconditionally
call rtnl_lock because either they already hold RTNL (for calls
via bond_3ad_unbind_slave) or due to the potential for deadlock
with bond_3ad_adapter_speed_changed, bond_3ad_adapter_duplex_changed,
bond_3ad_link_change, or bond_3ad_update_lacp_rate. All four of
those are called with RTNL held, and acquire the state machine lock
second. The calling contexts for __enable_port and __disable_port
already hold the state machine lock, and may or may not need RTNL.
According to the Jay's opinion, I don't think it is a problem that
the slave don't send notify message synchronously when the status
changed, normally the state machine is running every 100 ms, send
the notify message at the end of the state machine if the slave's
state changed should be better.
I fix the problem through these steps:
1). add a new function bond_set_slave_state() which could change
the slave's state and call rtmsg_ifinfo() according to the input
parameters called notify.
2). Add a new slave parameter which called should_notify, if the slave's state
changed and don't notify yet, the parameter will be set to 1, and then if
the slave's state changed again, the param will be set to 0, it indicate that
the slave's state has been restored, no need to notify any one.
3). the __enable_port and __disable_port should not call rtmsg_ifinfo
in the state machine lock, any change in the state of slave could
set a flag in the slave, it will indicated that an rtmsg_ifinfo
should be called at the end of the state machine.
Cc: Jay Vosburgh <fubar@us.ibm.com>
Cc: Veaceslav Falico <vfalico@redhat.com>
Cc: Andy Gospodarek <andy@greyhouse.net>
Signed-off-by: Ding Tianhong <dingtianhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2014-02-26 11:05:22 +08:00
|
|
|
bond_set_slave_inactive_flags(port->slave, BOND_SLAVE_NOTIFY_LATER);
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* __enable_port - enable the port's slave, if it's up
|
|
|
|
* @port: the port we're looking at
|
|
|
|
*/
|
|
|
|
static inline void __enable_port(struct port *port)
|
|
|
|
{
|
|
|
|
struct slave *slave = port->slave;
|
|
|
|
|
2014-05-16 03:39:57 +08:00
|
|
|
if ((slave->link == BOND_LINK_UP) && bond_slave_is_up(slave))
|
bonding: Fix RTNL: assertion failed at net/core/rtnetlink.c for 802.3ad mode
The problem was introduced by the commit 1d3ee88ae0d
(bonding: add netlink attributes to slave link dev).
The bond_set_active_slave() and bond_set_backup_slave()
will use rtmsg_ifinfo to send slave's states, so these
two functions should be called in RTNL.
In 802.3ad mode, acquiring RTNL for the __enable_port and
__disable_port cases is difficult, as those calls generally
already hold the state machine lock, and cannot unconditionally
call rtnl_lock because either they already hold RTNL (for calls
via bond_3ad_unbind_slave) or due to the potential for deadlock
with bond_3ad_adapter_speed_changed, bond_3ad_adapter_duplex_changed,
bond_3ad_link_change, or bond_3ad_update_lacp_rate. All four of
those are called with RTNL held, and acquire the state machine lock
second. The calling contexts for __enable_port and __disable_port
already hold the state machine lock, and may or may not need RTNL.
According to the Jay's opinion, I don't think it is a problem that
the slave don't send notify message synchronously when the status
changed, normally the state machine is running every 100 ms, send
the notify message at the end of the state machine if the slave's
state changed should be better.
I fix the problem through these steps:
1). add a new function bond_set_slave_state() which could change
the slave's state and call rtmsg_ifinfo() according to the input
parameters called notify.
2). Add a new slave parameter which called should_notify, if the slave's state
changed and don't notify yet, the parameter will be set to 1, and then if
the slave's state changed again, the param will be set to 0, it indicate that
the slave's state has been restored, no need to notify any one.
3). the __enable_port and __disable_port should not call rtmsg_ifinfo
in the state machine lock, any change in the state of slave could
set a flag in the slave, it will indicated that an rtmsg_ifinfo
should be called at the end of the state machine.
Cc: Jay Vosburgh <fubar@us.ibm.com>
Cc: Veaceslav Falico <vfalico@redhat.com>
Cc: Andy Gospodarek <andy@greyhouse.net>
Signed-off-by: Ding Tianhong <dingtianhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2014-02-26 11:05:22 +08:00
|
|
|
bond_set_slave_active_flags(slave, BOND_SLAVE_NOTIFY_LATER);
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* __port_is_enabled - check if the port's slave is in active state
|
|
|
|
* @port: the port we're looking at
|
|
|
|
*/
|
|
|
|
static inline int __port_is_enabled(struct port *port)
|
|
|
|
{
|
2011-03-12 11:14:37 +08:00
|
|
|
return bond_is_active_slave(port->slave);
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* __get_agg_selection_mode - get the aggregator selection mode
|
|
|
|
* @port: the port we're looking at
|
|
|
|
*
|
2008-11-05 09:51:16 +08:00
|
|
|
* Get the aggregator selection mode. Can be %STABLE, %BANDWIDTH or %COUNT.
|
2005-04-17 06:20:36 +08:00
|
|
|
*/
|
|
|
|
static inline u32 __get_agg_selection_mode(struct port *port)
|
|
|
|
{
|
|
|
|
struct bonding *bond = __get_bond_by_port(port);
|
|
|
|
|
2010-10-17 04:19:59 +08:00
|
|
|
if (bond == NULL)
|
2008-11-05 09:51:16 +08:00
|
|
|
return BOND_AD_STABLE;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2011-06-09 05:19:03 +08:00
|
|
|
return bond->params.ad_select;
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* __check_agg_selection_timer - check if the selection timer has expired
|
|
|
|
* @port: the port we're looking at
|
|
|
|
*/
|
|
|
|
static inline int __check_agg_selection_timer(struct port *port)
|
|
|
|
{
|
|
|
|
struct bonding *bond = __get_bond_by_port(port);
|
|
|
|
|
2010-10-17 04:19:59 +08:00
|
|
|
if (bond == NULL)
|
2005-04-17 06:20:36 +08:00
|
|
|
return 0;
|
|
|
|
|
|
|
|
return BOND_AD_INFO(bond).agg_select_timer ? 1 : 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* __get_link_speed - get a port's speed
|
|
|
|
* @port: the port we're looking at
|
|
|
|
*
|
2014-11-19 16:48:58 +08:00
|
|
|
* Return @port's speed in 802.3ad enum format. i.e. one of:
|
2005-04-17 06:20:36 +08:00
|
|
|
* 0,
|
2014-11-19 16:48:58 +08:00
|
|
|
* %AD_LINK_SPEED_10MBPS,
|
|
|
|
* %AD_LINK_SPEED_100MBPS,
|
|
|
|
* %AD_LINK_SPEED_1000MBPS,
|
2014-11-19 16:48:59 +08:00
|
|
|
* %AD_LINK_SPEED_2500MBPS,
|
2014-11-19 16:48:58 +08:00
|
|
|
* %AD_LINK_SPEED_10000MBPS
|
2014-11-19 16:48:59 +08:00
|
|
|
* %AD_LINK_SPEED_20000MBPS
|
|
|
|
* %AD_LINK_SPEED_40000MBPS
|
|
|
|
* %AD_LINK_SPEED_56000MBPS
|
2015-12-03 19:12:05 +08:00
|
|
|
* %AD_LINK_SPEED_100000MBPS
|
2005-04-17 06:20:36 +08:00
|
|
|
*/
|
|
|
|
static u16 __get_link_speed(struct port *port)
|
|
|
|
{
|
|
|
|
struct slave *slave = port->slave;
|
|
|
|
u16 speed;
|
|
|
|
|
2014-01-08 23:46:46 +08:00
|
|
|
/* this if covers only a special case: when the configuration starts
|
|
|
|
* with link down, it sets the speed to 0.
|
|
|
|
* This is done in spite of the fact that the e100 driver reports 0
|
|
|
|
* to be compatible with MVT in the future.
|
|
|
|
*/
|
2010-10-17 04:19:59 +08:00
|
|
|
if (slave->link != BOND_LINK_UP)
|
2010-10-17 04:19:58 +08:00
|
|
|
speed = 0;
|
2010-10-17 04:19:59 +08:00
|
|
|
else {
|
2005-04-17 06:20:36 +08:00
|
|
|
switch (slave->speed) {
|
|
|
|
case SPEED_10:
|
2014-11-19 16:48:58 +08:00
|
|
|
speed = AD_LINK_SPEED_10MBPS;
|
2005-04-17 06:20:36 +08:00
|
|
|
break;
|
|
|
|
|
|
|
|
case SPEED_100:
|
2014-11-19 16:48:58 +08:00
|
|
|
speed = AD_LINK_SPEED_100MBPS;
|
2005-04-17 06:20:36 +08:00
|
|
|
break;
|
|
|
|
|
|
|
|
case SPEED_1000:
|
2014-11-19 16:48:58 +08:00
|
|
|
speed = AD_LINK_SPEED_1000MBPS;
|
2005-04-17 06:20:36 +08:00
|
|
|
break;
|
|
|
|
|
2014-11-19 16:48:59 +08:00
|
|
|
case SPEED_2500:
|
|
|
|
speed = AD_LINK_SPEED_2500MBPS;
|
|
|
|
break;
|
|
|
|
|
2006-09-23 12:52:15 +08:00
|
|
|
case SPEED_10000:
|
2014-11-19 16:48:58 +08:00
|
|
|
speed = AD_LINK_SPEED_10000MBPS;
|
2006-09-23 12:52:15 +08:00
|
|
|
break;
|
|
|
|
|
2014-11-19 16:48:59 +08:00
|
|
|
case SPEED_20000:
|
|
|
|
speed = AD_LINK_SPEED_20000MBPS;
|
|
|
|
break;
|
|
|
|
|
|
|
|
case SPEED_40000:
|
|
|
|
speed = AD_LINK_SPEED_40000MBPS;
|
|
|
|
break;
|
|
|
|
|
|
|
|
case SPEED_56000:
|
|
|
|
speed = AD_LINK_SPEED_56000MBPS;
|
|
|
|
break;
|
|
|
|
|
2015-12-03 19:12:05 +08:00
|
|
|
case SPEED_100000:
|
|
|
|
speed = AD_LINK_SPEED_100000MBPS;
|
|
|
|
break;
|
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
default:
|
2014-01-08 23:46:46 +08:00
|
|
|
/* unknown speed value from ethtool. shouldn't happen */
|
|
|
|
speed = 0;
|
2005-04-17 06:20:36 +08:00
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2014-07-16 01:36:00 +08:00
|
|
|
netdev_dbg(slave->bond->dev, "Port %d Received link speed %d update from adapter\n",
|
|
|
|
port->actor_port_number, speed);
|
2005-04-17 06:20:36 +08:00
|
|
|
return speed;
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* __get_duplex - get a port's duplex
|
|
|
|
* @port: the port we're looking at
|
|
|
|
*
|
|
|
|
* Return @port's duplex in 802.3ad bitmask format. i.e.:
|
|
|
|
* 0x01 if in full duplex
|
|
|
|
* 0x00 otherwise
|
|
|
|
*/
|
|
|
|
static u8 __get_duplex(struct port *port)
|
|
|
|
{
|
|
|
|
struct slave *slave = port->slave;
|
2015-11-01 03:45:00 +08:00
|
|
|
u8 retval = 0x0;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2014-01-08 23:46:46 +08:00
|
|
|
/* handling a special case: when the configuration starts with
|
|
|
|
* link down, it sets the duplex to 0.
|
|
|
|
*/
|
2015-11-01 03:45:00 +08:00
|
|
|
if (slave->link == BOND_LINK_UP) {
|
2005-04-17 06:20:36 +08:00
|
|
|
switch (slave->duplex) {
|
|
|
|
case DUPLEX_FULL:
|
2010-10-17 04:19:58 +08:00
|
|
|
retval = 0x1;
|
2014-07-16 01:36:00 +08:00
|
|
|
netdev_dbg(slave->bond->dev, "Port %d Received status full duplex update from adapter\n",
|
|
|
|
port->actor_port_number);
|
2005-04-17 06:20:36 +08:00
|
|
|
break;
|
|
|
|
case DUPLEX_HALF:
|
|
|
|
default:
|
2010-10-17 04:19:58 +08:00
|
|
|
retval = 0x0;
|
2014-07-16 01:36:00 +08:00
|
|
|
netdev_dbg(slave->bond->dev, "Port %d Received status NOT full duplex update from adapter\n",
|
|
|
|
port->actor_port_number);
|
2005-04-17 06:20:36 +08:00
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
return retval;
|
|
|
|
}
|
|
|
|
|
2016-02-03 20:17:01 +08:00
|
|
|
static void __ad_actor_update_port(struct port *port)
|
|
|
|
{
|
|
|
|
const struct bonding *bond = bond_get_bond_by_slave(port->slave);
|
|
|
|
|
|
|
|
port->actor_system = BOND_AD_INFO(bond).system.sys_mac_addr;
|
|
|
|
port->actor_system_priority = BOND_AD_INFO(bond).system.sys_priority;
|
|
|
|
}
|
|
|
|
|
2014-01-08 23:46:46 +08:00
|
|
|
/* Conversions */
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
/**
|
|
|
|
* __ad_timer_to_ticks - convert a given timer type to AD module ticks
|
|
|
|
* @timer_type: which timer to operate
|
|
|
|
* @par: timer parameter. see below
|
|
|
|
*
|
|
|
|
* If @timer_type is %current_while_timer, @par indicates long/short timer.
|
|
|
|
* If @timer_type is %periodic_timer, @par is one of %FAST_PERIODIC_TIME,
|
2014-01-08 23:46:46 +08:00
|
|
|
* %SLOW_PERIODIC_TIME.
|
2005-04-17 06:20:36 +08:00
|
|
|
*/
|
|
|
|
static u16 __ad_timer_to_ticks(u16 timer_type, u16 par)
|
|
|
|
{
|
2010-10-17 04:19:58 +08:00
|
|
|
u16 retval = 0; /* to silence the compiler */
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
switch (timer_type) {
|
2014-01-08 23:46:46 +08:00
|
|
|
case AD_CURRENT_WHILE_TIMER: /* for rx machine usage */
|
2010-10-17 04:19:59 +08:00
|
|
|
if (par)
|
2014-01-08 23:46:46 +08:00
|
|
|
retval = (AD_SHORT_TIMEOUT_TIME*ad_ticks_per_sec);
|
2010-10-17 04:19:59 +08:00
|
|
|
else
|
2014-01-08 23:46:46 +08:00
|
|
|
retval = (AD_LONG_TIMEOUT_TIME*ad_ticks_per_sec);
|
2005-04-17 06:20:36 +08:00
|
|
|
break;
|
2014-01-08 23:46:46 +08:00
|
|
|
case AD_ACTOR_CHURN_TIMER: /* for local churn machine */
|
2005-04-17 06:20:36 +08:00
|
|
|
retval = (AD_CHURN_DETECTION_TIME*ad_ticks_per_sec);
|
|
|
|
break;
|
2014-01-08 23:46:46 +08:00
|
|
|
case AD_PERIODIC_TIMER: /* for periodic machine */
|
|
|
|
retval = (par*ad_ticks_per_sec); /* long timeout */
|
2005-04-17 06:20:36 +08:00
|
|
|
break;
|
2014-01-08 23:46:46 +08:00
|
|
|
case AD_PARTNER_CHURN_TIMER: /* for remote churn machine */
|
2005-04-17 06:20:36 +08:00
|
|
|
retval = (AD_CHURN_DETECTION_TIME*ad_ticks_per_sec);
|
|
|
|
break;
|
2014-01-08 23:46:46 +08:00
|
|
|
case AD_WAIT_WHILE_TIMER: /* for selection machine */
|
2005-04-17 06:20:36 +08:00
|
|
|
retval = (AD_AGGREGATE_WAIT_TIME*ad_ticks_per_sec);
|
|
|
|
break;
|
|
|
|
}
|
2014-01-08 23:46:46 +08:00
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
return retval;
|
|
|
|
}
|
|
|
|
|
|
|
|
|
2014-01-08 23:46:46 +08:00
|
|
|
/* ================= ad_rx_machine helper functions ================== */
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2009-11-13 21:13:01 +08:00
|
|
|
/**
|
|
|
|
* __choose_matched - update a port's matched variable from a received lacpdu
|
|
|
|
* @lacpdu: the lacpdu we've received
|
|
|
|
* @port: the port we're looking at
|
|
|
|
*
|
|
|
|
* Update the value of the matched variable, using parameter values from a
|
|
|
|
* newly received lacpdu. Parameter values for the partner carried in the
|
|
|
|
* received PDU are compared with the corresponding operational parameter
|
|
|
|
* values for the actor. Matched is set to TRUE if all of these parameters
|
|
|
|
* match and the PDU parameter partner_state.aggregation has the same value as
|
|
|
|
* actor_oper_port_state.aggregation and lacp will actively maintain the link
|
|
|
|
* in the aggregation. Matched is also set to TRUE if the value of
|
|
|
|
* actor_state.aggregation in the received PDU is set to FALSE, i.e., indicates
|
|
|
|
* an individual link and lacp will actively maintain the link. Otherwise,
|
|
|
|
* matched is set to FALSE. LACP is considered to be actively maintaining the
|
|
|
|
* link if either the PDU's actor_state.lacp_activity variable is TRUE or both
|
|
|
|
* the actor's actor_oper_port_state.lacp_activity and the PDU's
|
|
|
|
* partner_state.lacp_activity variables are TRUE.
|
|
|
|
*
|
|
|
|
* Note: the AD_PORT_MATCHED "variable" is not specified by 802.3ad; it is
|
|
|
|
* used here to implement the language from 802.3ad 43.4.9 that requires
|
|
|
|
* recordPDU to "match" the LACPDU parameters to the stored values.
|
|
|
|
*/
|
|
|
|
static void __choose_matched(struct lacpdu *lacpdu, struct port *port)
|
|
|
|
{
|
2014-01-02 09:12:54 +08:00
|
|
|
/* check if all parameters are alike
|
|
|
|
* or this is individual link(aggregation == FALSE)
|
|
|
|
* then update the state machine Matched variable.
|
|
|
|
*/
|
2009-11-13 21:13:01 +08:00
|
|
|
if (((ntohs(lacpdu->partner_port) == port->actor_port_number) &&
|
|
|
|
(ntohs(lacpdu->partner_port_priority) == port->actor_port_priority) &&
|
2014-01-02 09:12:54 +08:00
|
|
|
MAC_ADDRESS_EQUAL(&(lacpdu->partner_system), &(port->actor_system)) &&
|
2009-11-13 21:13:01 +08:00
|
|
|
(ntohs(lacpdu->partner_system_priority) == port->actor_system_priority) &&
|
|
|
|
(ntohs(lacpdu->partner_key) == port->actor_oper_port_key) &&
|
|
|
|
((lacpdu->partner_state & AD_STATE_AGGREGATION) == (port->actor_oper_port_state & AD_STATE_AGGREGATION))) ||
|
|
|
|
((lacpdu->actor_state & AD_STATE_AGGREGATION) == 0)
|
|
|
|
) {
|
|
|
|
port->sm_vars |= AD_PORT_MATCHED;
|
|
|
|
} else {
|
|
|
|
port->sm_vars &= ~AD_PORT_MATCHED;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
/**
|
|
|
|
* __record_pdu - record parameters from a received lacpdu
|
|
|
|
* @lacpdu: the lacpdu we've received
|
|
|
|
* @port: the port we're looking at
|
|
|
|
*
|
|
|
|
* Record the parameter values for the Actor carried in a received lacpdu as
|
|
|
|
* the current partner operational parameter values and sets
|
|
|
|
* actor_oper_port_state.defaulted to FALSE.
|
|
|
|
*/
|
|
|
|
static void __record_pdu(struct lacpdu *lacpdu, struct port *port)
|
|
|
|
{
|
|
|
|
if (lacpdu && port) {
|
2008-12-18 11:08:14 +08:00
|
|
|
struct port_params *partner = &port->partner_oper;
|
|
|
|
|
2009-11-13 21:13:01 +08:00
|
|
|
__choose_matched(lacpdu, port);
|
2014-01-08 23:46:46 +08:00
|
|
|
/* record the new parameter values for the partner
|
|
|
|
* operational
|
|
|
|
*/
|
2008-12-18 11:08:14 +08:00
|
|
|
partner->port_number = ntohs(lacpdu->actor_port);
|
|
|
|
partner->port_priority = ntohs(lacpdu->actor_port_priority);
|
|
|
|
partner->system = lacpdu->actor_system;
|
|
|
|
partner->system_priority = ntohs(lacpdu->actor_system_priority);
|
|
|
|
partner->key = ntohs(lacpdu->actor_key);
|
|
|
|
partner->port_state = lacpdu->actor_state;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2014-01-08 23:46:46 +08:00
|
|
|
/* set actor_oper_port_state.defaulted to FALSE */
|
2005-04-17 06:20:36 +08:00
|
|
|
port->actor_oper_port_state &= ~AD_STATE_DEFAULTED;
|
|
|
|
|
2014-01-08 23:46:46 +08:00
|
|
|
/* set the partner sync. to on if the partner is sync,
|
|
|
|
* and the port is matched
|
|
|
|
*/
|
2015-01-26 14:16:59 +08:00
|
|
|
if ((port->sm_vars & AD_PORT_MATCHED) &&
|
|
|
|
(lacpdu->actor_state & AD_STATE_SYNCHRONIZATION)) {
|
2008-12-18 11:08:14 +08:00
|
|
|
partner->port_state |= AD_STATE_SYNCHRONIZATION;
|
2015-01-26 14:16:59 +08:00
|
|
|
pr_debug("%s partner sync=1\n", port->slave->dev->name);
|
|
|
|
} else {
|
2008-12-18 11:08:14 +08:00
|
|
|
partner->port_state &= ~AD_STATE_SYNCHRONIZATION;
|
2015-01-26 14:16:59 +08:00
|
|
|
pr_debug("%s partner sync=0\n", port->slave->dev->name);
|
|
|
|
}
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* __record_default - record default parameters
|
|
|
|
* @port: the port we're looking at
|
|
|
|
*
|
|
|
|
* This function records the default parameter values for the partner carried
|
|
|
|
* in the Partner Admin parameters as the current partner operational parameter
|
|
|
|
* values and sets actor_oper_port_state.defaulted to TRUE.
|
|
|
|
*/
|
|
|
|
static void __record_default(struct port *port)
|
|
|
|
{
|
|
|
|
if (port) {
|
2014-01-08 23:46:46 +08:00
|
|
|
/* record the partner admin parameters */
|
2008-12-18 11:08:46 +08:00
|
|
|
memcpy(&port->partner_oper, &port->partner_admin,
|
|
|
|
sizeof(struct port_params));
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2014-01-08 23:46:46 +08:00
|
|
|
/* set actor_oper_port_state.defaulted to true */
|
2005-04-17 06:20:36 +08:00
|
|
|
port->actor_oper_port_state |= AD_STATE_DEFAULTED;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* __update_selected - update a port's Selected variable from a received lacpdu
|
|
|
|
* @lacpdu: the lacpdu we've received
|
|
|
|
* @port: the port we're looking at
|
|
|
|
*
|
|
|
|
* Update the value of the selected variable, using parameter values from a
|
|
|
|
* newly received lacpdu. The parameter values for the Actor carried in the
|
|
|
|
* received PDU are compared with the corresponding operational parameter
|
|
|
|
* values for the ports partner. If one or more of the comparisons shows that
|
|
|
|
* the value(s) received in the PDU differ from the current operational values,
|
|
|
|
* then selected is set to FALSE and actor_oper_port_state.synchronization is
|
|
|
|
* set to out_of_sync. Otherwise, selected remains unchanged.
|
|
|
|
*/
|
|
|
|
static void __update_selected(struct lacpdu *lacpdu, struct port *port)
|
|
|
|
{
|
|
|
|
if (lacpdu && port) {
|
2008-12-18 11:13:07 +08:00
|
|
|
const struct port_params *partner = &port->partner_oper;
|
|
|
|
|
2014-01-02 09:12:54 +08:00
|
|
|
/* check if any parameter is different then
|
|
|
|
* update the state machine selected variable.
|
|
|
|
*/
|
2009-12-03 15:58:21 +08:00
|
|
|
if (ntohs(lacpdu->actor_port) != partner->port_number ||
|
|
|
|
ntohs(lacpdu->actor_port_priority) != partner->port_priority ||
|
2014-01-02 09:12:54 +08:00
|
|
|
!MAC_ADDRESS_EQUAL(&lacpdu->actor_system, &partner->system) ||
|
2009-12-03 15:58:21 +08:00
|
|
|
ntohs(lacpdu->actor_system_priority) != partner->system_priority ||
|
|
|
|
ntohs(lacpdu->actor_key) != partner->key ||
|
|
|
|
(lacpdu->actor_state & AD_STATE_AGGREGATION) != (partner->port_state & AD_STATE_AGGREGATION)) {
|
2005-04-17 06:20:36 +08:00
|
|
|
port->sm_vars &= ~AD_PORT_SELECTED;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* __update_default_selected - update a port's Selected variable from Partner
|
|
|
|
* @port: the port we're looking at
|
|
|
|
*
|
|
|
|
* This function updates the value of the selected variable, using the partner
|
|
|
|
* administrative parameter values. The administrative values are compared with
|
|
|
|
* the corresponding operational parameter values for the partner. If one or
|
|
|
|
* more of the comparisons shows that the administrative value(s) differ from
|
|
|
|
* the current operational values, then Selected is set to FALSE and
|
|
|
|
* actor_oper_port_state.synchronization is set to OUT_OF_SYNC. Otherwise,
|
|
|
|
* Selected remains unchanged.
|
|
|
|
*/
|
|
|
|
static void __update_default_selected(struct port *port)
|
|
|
|
{
|
|
|
|
if (port) {
|
2008-12-18 11:13:27 +08:00
|
|
|
const struct port_params *admin = &port->partner_admin;
|
|
|
|
const struct port_params *oper = &port->partner_oper;
|
|
|
|
|
2014-01-02 09:12:54 +08:00
|
|
|
/* check if any parameter is different then
|
|
|
|
* update the state machine selected variable.
|
|
|
|
*/
|
2009-12-03 15:58:21 +08:00
|
|
|
if (admin->port_number != oper->port_number ||
|
|
|
|
admin->port_priority != oper->port_priority ||
|
2014-01-02 09:12:54 +08:00
|
|
|
!MAC_ADDRESS_EQUAL(&admin->system, &oper->system) ||
|
2009-12-03 15:58:21 +08:00
|
|
|
admin->system_priority != oper->system_priority ||
|
|
|
|
admin->key != oper->key ||
|
|
|
|
(admin->port_state & AD_STATE_AGGREGATION)
|
2008-12-18 11:13:27 +08:00
|
|
|
!= (oper->port_state & AD_STATE_AGGREGATION)) {
|
2005-04-17 06:20:36 +08:00
|
|
|
port->sm_vars &= ~AD_PORT_SELECTED;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* __update_ntt - update a port's ntt variable from a received lacpdu
|
|
|
|
* @lacpdu: the lacpdu we've received
|
|
|
|
* @port: the port we're looking at
|
|
|
|
*
|
|
|
|
* Updates the value of the ntt variable, using parameter values from a newly
|
|
|
|
* received lacpdu. The parameter values for the partner carried in the
|
|
|
|
* received PDU are compared with the corresponding operational parameter
|
|
|
|
* values for the Actor. If one or more of the comparisons shows that the
|
|
|
|
* value(s) received in the PDU differ from the current operational values,
|
|
|
|
* then ntt is set to TRUE. Otherwise, ntt remains unchanged.
|
|
|
|
*/
|
|
|
|
static void __update_ntt(struct lacpdu *lacpdu, struct port *port)
|
|
|
|
{
|
2014-01-02 09:12:54 +08:00
|
|
|
/* validate lacpdu and port */
|
2005-04-17 06:20:36 +08:00
|
|
|
if (lacpdu && port) {
|
2014-01-02 09:12:54 +08:00
|
|
|
/* check if any parameter is different then
|
|
|
|
* update the port->ntt.
|
|
|
|
*/
|
2006-09-23 12:55:32 +08:00
|
|
|
if ((ntohs(lacpdu->partner_port) != port->actor_port_number) ||
|
|
|
|
(ntohs(lacpdu->partner_port_priority) != port->actor_port_priority) ||
|
2014-01-02 09:12:54 +08:00
|
|
|
!MAC_ADDRESS_EQUAL(&(lacpdu->partner_system), &(port->actor_system)) ||
|
2006-09-23 12:55:32 +08:00
|
|
|
(ntohs(lacpdu->partner_system_priority) != port->actor_system_priority) ||
|
|
|
|
(ntohs(lacpdu->partner_key) != port->actor_oper_port_key) ||
|
2005-04-17 06:20:36 +08:00
|
|
|
((lacpdu->partner_state & AD_STATE_LACP_ACTIVITY) != (port->actor_oper_port_state & AD_STATE_LACP_ACTIVITY)) ||
|
|
|
|
((lacpdu->partner_state & AD_STATE_LACP_TIMEOUT) != (port->actor_oper_port_state & AD_STATE_LACP_TIMEOUT)) ||
|
|
|
|
((lacpdu->partner_state & AD_STATE_SYNCHRONIZATION) != (port->actor_oper_port_state & AD_STATE_SYNCHRONIZATION)) ||
|
|
|
|
((lacpdu->partner_state & AD_STATE_AGGREGATION) != (port->actor_oper_port_state & AD_STATE_AGGREGATION))
|
|
|
|
) {
|
2008-12-27 03:18:15 +08:00
|
|
|
port->ntt = true;
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* __agg_ports_are_ready - check if all ports in an aggregator are ready
|
|
|
|
* @aggregator: the aggregator we're looking at
|
|
|
|
*
|
|
|
|
*/
|
|
|
|
static int __agg_ports_are_ready(struct aggregator *aggregator)
|
|
|
|
{
|
|
|
|
struct port *port;
|
|
|
|
int retval = 1;
|
|
|
|
|
|
|
|
if (aggregator) {
|
2014-01-08 23:46:46 +08:00
|
|
|
/* scan all ports in this aggregator to verfy if they are
|
|
|
|
* all ready.
|
|
|
|
*/
|
2010-10-17 04:19:58 +08:00
|
|
|
for (port = aggregator->lag_ports;
|
|
|
|
port;
|
|
|
|
port = port->next_port_in_aggregator) {
|
2005-04-17 06:20:36 +08:00
|
|
|
if (!(port->sm_vars & AD_PORT_READY_N)) {
|
|
|
|
retval = 0;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
return retval;
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* __set_agg_ports_ready - set value of Ready bit in all ports of an aggregator
|
|
|
|
* @aggregator: the aggregator we're looking at
|
|
|
|
* @val: Should the ports' ready bit be set on or off
|
|
|
|
*
|
|
|
|
*/
|
|
|
|
static void __set_agg_ports_ready(struct aggregator *aggregator, int val)
|
|
|
|
{
|
|
|
|
struct port *port;
|
|
|
|
|
2010-10-17 04:19:58 +08:00
|
|
|
for (port = aggregator->lag_ports; port;
|
|
|
|
port = port->next_port_in_aggregator) {
|
2010-10-17 04:19:59 +08:00
|
|
|
if (val)
|
2005-04-17 06:20:36 +08:00
|
|
|
port->sm_vars |= AD_PORT_READY;
|
2010-10-17 04:19:59 +08:00
|
|
|
else
|
2005-04-17 06:20:36 +08:00
|
|
|
port->sm_vars &= ~AD_PORT_READY;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
bonding: fix 802.3ad aggregator reselection
Since commit 7bb11dc9f59d ("bonding: unify all places where
actor-oper key needs to be updated."), the logic in bonding to handle
selection between multiple aggregators has not functioned.
This affects only configurations wherein the bonding slaves
connect to two discrete aggregators (e.g., two independent switches, each
with LACP enabled), thus creating two separate aggregation groups within a
single bond.
The cause is a change in 7bb11dc9f59d to no longer set
AD_PORT_BEGIN on a port after a link state change, which would cause the
port to be reselected for attachment to an aggregator as if were newly
added to the bond. We cannot restore the prior behavior, as it
contradicts IEEE 802.1AX 5.4.12, which requires ports that "become
inoperable" (lose carrier, setting port_enabled=false as per 802.1AX
5.4.7) to remain selected (i.e., assigned to the aggregator). As the port
now remains selected, the aggregator selection logic is not invoked.
A side effect of this change is that aggregators in bonding will
now contain ports that are link down. The aggregator selection logic
does not currently handle this situation correctly, causing incorrect
aggregator selection.
This patch makes two changes to repair the aggregator selection
logic in bonding to function as documented and within the confines of the
standard:
First, the aggregator selection and related logic now utilizes the
number of active ports per aggregator, not the number of selected ports
(as some selected ports may be down). The ad_select "bandwidth" and
"count" options only consider ports that are link up.
Second, on any carrier state change of any slave, the aggregator
selection logic is explicitly called to insure the correct aggregator is
active.
Reported-by: Veli-Matti Lintu <veli-matti.lintu@opinsys.fi>
Fixes: 7bb11dc9f59d ("bonding: unify all places where actor-oper key needs to be updated.")
Signed-off-by: Jay Vosburgh <jay.vosburgh@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2016-06-24 05:20:51 +08:00
|
|
|
static int __agg_active_ports(struct aggregator *agg)
|
|
|
|
{
|
|
|
|
struct port *port;
|
|
|
|
int active = 0;
|
|
|
|
|
|
|
|
for (port = agg->lag_ports; port;
|
|
|
|
port = port->next_port_in_aggregator) {
|
|
|
|
if (port->is_enabled)
|
|
|
|
active++;
|
|
|
|
}
|
|
|
|
|
|
|
|
return active;
|
|
|
|
}
|
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
/**
|
|
|
|
* __get_agg_bandwidth - get the total bandwidth of an aggregator
|
|
|
|
* @aggregator: the aggregator we're looking at
|
|
|
|
*
|
|
|
|
*/
|
|
|
|
static u32 __get_agg_bandwidth(struct aggregator *aggregator)
|
|
|
|
{
|
bonding: fix 802.3ad aggregator reselection
Since commit 7bb11dc9f59d ("bonding: unify all places where
actor-oper key needs to be updated."), the logic in bonding to handle
selection between multiple aggregators has not functioned.
This affects only configurations wherein the bonding slaves
connect to two discrete aggregators (e.g., two independent switches, each
with LACP enabled), thus creating two separate aggregation groups within a
single bond.
The cause is a change in 7bb11dc9f59d to no longer set
AD_PORT_BEGIN on a port after a link state change, which would cause the
port to be reselected for attachment to an aggregator as if were newly
added to the bond. We cannot restore the prior behavior, as it
contradicts IEEE 802.1AX 5.4.12, which requires ports that "become
inoperable" (lose carrier, setting port_enabled=false as per 802.1AX
5.4.7) to remain selected (i.e., assigned to the aggregator). As the port
now remains selected, the aggregator selection logic is not invoked.
A side effect of this change is that aggregators in bonding will
now contain ports that are link down. The aggregator selection logic
does not currently handle this situation correctly, causing incorrect
aggregator selection.
This patch makes two changes to repair the aggregator selection
logic in bonding to function as documented and within the confines of the
standard:
First, the aggregator selection and related logic now utilizes the
number of active ports per aggregator, not the number of selected ports
(as some selected ports may be down). The ad_select "bandwidth" and
"count" options only consider ports that are link up.
Second, on any carrier state change of any slave, the aggregator
selection logic is explicitly called to insure the correct aggregator is
active.
Reported-by: Veli-Matti Lintu <veli-matti.lintu@opinsys.fi>
Fixes: 7bb11dc9f59d ("bonding: unify all places where actor-oper key needs to be updated.")
Signed-off-by: Jay Vosburgh <jay.vosburgh@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2016-06-24 05:20:51 +08:00
|
|
|
int nports = __agg_active_ports(aggregator);
|
2010-10-17 04:19:58 +08:00
|
|
|
u32 bandwidth = 0;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
bonding: fix 802.3ad aggregator reselection
Since commit 7bb11dc9f59d ("bonding: unify all places where
actor-oper key needs to be updated."), the logic in bonding to handle
selection between multiple aggregators has not functioned.
This affects only configurations wherein the bonding slaves
connect to two discrete aggregators (e.g., two independent switches, each
with LACP enabled), thus creating two separate aggregation groups within a
single bond.
The cause is a change in 7bb11dc9f59d to no longer set
AD_PORT_BEGIN on a port after a link state change, which would cause the
port to be reselected for attachment to an aggregator as if were newly
added to the bond. We cannot restore the prior behavior, as it
contradicts IEEE 802.1AX 5.4.12, which requires ports that "become
inoperable" (lose carrier, setting port_enabled=false as per 802.1AX
5.4.7) to remain selected (i.e., assigned to the aggregator). As the port
now remains selected, the aggregator selection logic is not invoked.
A side effect of this change is that aggregators in bonding will
now contain ports that are link down. The aggregator selection logic
does not currently handle this situation correctly, causing incorrect
aggregator selection.
This patch makes two changes to repair the aggregator selection
logic in bonding to function as documented and within the confines of the
standard:
First, the aggregator selection and related logic now utilizes the
number of active ports per aggregator, not the number of selected ports
(as some selected ports may be down). The ad_select "bandwidth" and
"count" options only consider ports that are link up.
Second, on any carrier state change of any slave, the aggregator
selection logic is explicitly called to insure the correct aggregator is
active.
Reported-by: Veli-Matti Lintu <veli-matti.lintu@opinsys.fi>
Fixes: 7bb11dc9f59d ("bonding: unify all places where actor-oper key needs to be updated.")
Signed-off-by: Jay Vosburgh <jay.vosburgh@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2016-06-24 05:20:51 +08:00
|
|
|
if (nports) {
|
2011-04-13 23:22:30 +08:00
|
|
|
switch (__get_link_speed(aggregator->lag_ports)) {
|
2014-11-19 16:48:58 +08:00
|
|
|
case AD_LINK_SPEED_1MBPS:
|
bonding: fix 802.3ad aggregator reselection
Since commit 7bb11dc9f59d ("bonding: unify all places where
actor-oper key needs to be updated."), the logic in bonding to handle
selection between multiple aggregators has not functioned.
This affects only configurations wherein the bonding slaves
connect to two discrete aggregators (e.g., two independent switches, each
with LACP enabled), thus creating two separate aggregation groups within a
single bond.
The cause is a change in 7bb11dc9f59d to no longer set
AD_PORT_BEGIN on a port after a link state change, which would cause the
port to be reselected for attachment to an aggregator as if were newly
added to the bond. We cannot restore the prior behavior, as it
contradicts IEEE 802.1AX 5.4.12, which requires ports that "become
inoperable" (lose carrier, setting port_enabled=false as per 802.1AX
5.4.7) to remain selected (i.e., assigned to the aggregator). As the port
now remains selected, the aggregator selection logic is not invoked.
A side effect of this change is that aggregators in bonding will
now contain ports that are link down. The aggregator selection logic
does not currently handle this situation correctly, causing incorrect
aggregator selection.
This patch makes two changes to repair the aggregator selection
logic in bonding to function as documented and within the confines of the
standard:
First, the aggregator selection and related logic now utilizes the
number of active ports per aggregator, not the number of selected ports
(as some selected ports may be down). The ad_select "bandwidth" and
"count" options only consider ports that are link up.
Second, on any carrier state change of any slave, the aggregator
selection logic is explicitly called to insure the correct aggregator is
active.
Reported-by: Veli-Matti Lintu <veli-matti.lintu@opinsys.fi>
Fixes: 7bb11dc9f59d ("bonding: unify all places where actor-oper key needs to be updated.")
Signed-off-by: Jay Vosburgh <jay.vosburgh@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2016-06-24 05:20:51 +08:00
|
|
|
bandwidth = nports;
|
2005-04-17 06:20:36 +08:00
|
|
|
break;
|
2014-11-19 16:48:58 +08:00
|
|
|
case AD_LINK_SPEED_10MBPS:
|
bonding: fix 802.3ad aggregator reselection
Since commit 7bb11dc9f59d ("bonding: unify all places where
actor-oper key needs to be updated."), the logic in bonding to handle
selection between multiple aggregators has not functioned.
This affects only configurations wherein the bonding slaves
connect to two discrete aggregators (e.g., two independent switches, each
with LACP enabled), thus creating two separate aggregation groups within a
single bond.
The cause is a change in 7bb11dc9f59d to no longer set
AD_PORT_BEGIN on a port after a link state change, which would cause the
port to be reselected for attachment to an aggregator as if were newly
added to the bond. We cannot restore the prior behavior, as it
contradicts IEEE 802.1AX 5.4.12, which requires ports that "become
inoperable" (lose carrier, setting port_enabled=false as per 802.1AX
5.4.7) to remain selected (i.e., assigned to the aggregator). As the port
now remains selected, the aggregator selection logic is not invoked.
A side effect of this change is that aggregators in bonding will
now contain ports that are link down. The aggregator selection logic
does not currently handle this situation correctly, causing incorrect
aggregator selection.
This patch makes two changes to repair the aggregator selection
logic in bonding to function as documented and within the confines of the
standard:
First, the aggregator selection and related logic now utilizes the
number of active ports per aggregator, not the number of selected ports
(as some selected ports may be down). The ad_select "bandwidth" and
"count" options only consider ports that are link up.
Second, on any carrier state change of any slave, the aggregator
selection logic is explicitly called to insure the correct aggregator is
active.
Reported-by: Veli-Matti Lintu <veli-matti.lintu@opinsys.fi>
Fixes: 7bb11dc9f59d ("bonding: unify all places where actor-oper key needs to be updated.")
Signed-off-by: Jay Vosburgh <jay.vosburgh@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2016-06-24 05:20:51 +08:00
|
|
|
bandwidth = nports * 10;
|
2005-04-17 06:20:36 +08:00
|
|
|
break;
|
2014-11-19 16:48:58 +08:00
|
|
|
case AD_LINK_SPEED_100MBPS:
|
bonding: fix 802.3ad aggregator reselection
Since commit 7bb11dc9f59d ("bonding: unify all places where
actor-oper key needs to be updated."), the logic in bonding to handle
selection between multiple aggregators has not functioned.
This affects only configurations wherein the bonding slaves
connect to two discrete aggregators (e.g., two independent switches, each
with LACP enabled), thus creating two separate aggregation groups within a
single bond.
The cause is a change in 7bb11dc9f59d to no longer set
AD_PORT_BEGIN on a port after a link state change, which would cause the
port to be reselected for attachment to an aggregator as if were newly
added to the bond. We cannot restore the prior behavior, as it
contradicts IEEE 802.1AX 5.4.12, which requires ports that "become
inoperable" (lose carrier, setting port_enabled=false as per 802.1AX
5.4.7) to remain selected (i.e., assigned to the aggregator). As the port
now remains selected, the aggregator selection logic is not invoked.
A side effect of this change is that aggregators in bonding will
now contain ports that are link down. The aggregator selection logic
does not currently handle this situation correctly, causing incorrect
aggregator selection.
This patch makes two changes to repair the aggregator selection
logic in bonding to function as documented and within the confines of the
standard:
First, the aggregator selection and related logic now utilizes the
number of active ports per aggregator, not the number of selected ports
(as some selected ports may be down). The ad_select "bandwidth" and
"count" options only consider ports that are link up.
Second, on any carrier state change of any slave, the aggregator
selection logic is explicitly called to insure the correct aggregator is
active.
Reported-by: Veli-Matti Lintu <veli-matti.lintu@opinsys.fi>
Fixes: 7bb11dc9f59d ("bonding: unify all places where actor-oper key needs to be updated.")
Signed-off-by: Jay Vosburgh <jay.vosburgh@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2016-06-24 05:20:51 +08:00
|
|
|
bandwidth = nports * 100;
|
2005-04-17 06:20:36 +08:00
|
|
|
break;
|
2014-11-19 16:48:58 +08:00
|
|
|
case AD_LINK_SPEED_1000MBPS:
|
bonding: fix 802.3ad aggregator reselection
Since commit 7bb11dc9f59d ("bonding: unify all places where
actor-oper key needs to be updated."), the logic in bonding to handle
selection between multiple aggregators has not functioned.
This affects only configurations wherein the bonding slaves
connect to two discrete aggregators (e.g., two independent switches, each
with LACP enabled), thus creating two separate aggregation groups within a
single bond.
The cause is a change in 7bb11dc9f59d to no longer set
AD_PORT_BEGIN on a port after a link state change, which would cause the
port to be reselected for attachment to an aggregator as if were newly
added to the bond. We cannot restore the prior behavior, as it
contradicts IEEE 802.1AX 5.4.12, which requires ports that "become
inoperable" (lose carrier, setting port_enabled=false as per 802.1AX
5.4.7) to remain selected (i.e., assigned to the aggregator). As the port
now remains selected, the aggregator selection logic is not invoked.
A side effect of this change is that aggregators in bonding will
now contain ports that are link down. The aggregator selection logic
does not currently handle this situation correctly, causing incorrect
aggregator selection.
This patch makes two changes to repair the aggregator selection
logic in bonding to function as documented and within the confines of the
standard:
First, the aggregator selection and related logic now utilizes the
number of active ports per aggregator, not the number of selected ports
(as some selected ports may be down). The ad_select "bandwidth" and
"count" options only consider ports that are link up.
Second, on any carrier state change of any slave, the aggregator
selection logic is explicitly called to insure the correct aggregator is
active.
Reported-by: Veli-Matti Lintu <veli-matti.lintu@opinsys.fi>
Fixes: 7bb11dc9f59d ("bonding: unify all places where actor-oper key needs to be updated.")
Signed-off-by: Jay Vosburgh <jay.vosburgh@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2016-06-24 05:20:51 +08:00
|
|
|
bandwidth = nports * 1000;
|
2005-04-17 06:20:36 +08:00
|
|
|
break;
|
2014-11-19 16:48:59 +08:00
|
|
|
case AD_LINK_SPEED_2500MBPS:
|
bonding: fix 802.3ad aggregator reselection
Since commit 7bb11dc9f59d ("bonding: unify all places where
actor-oper key needs to be updated."), the logic in bonding to handle
selection between multiple aggregators has not functioned.
This affects only configurations wherein the bonding slaves
connect to two discrete aggregators (e.g., two independent switches, each
with LACP enabled), thus creating two separate aggregation groups within a
single bond.
The cause is a change in 7bb11dc9f59d to no longer set
AD_PORT_BEGIN on a port after a link state change, which would cause the
port to be reselected for attachment to an aggregator as if were newly
added to the bond. We cannot restore the prior behavior, as it
contradicts IEEE 802.1AX 5.4.12, which requires ports that "become
inoperable" (lose carrier, setting port_enabled=false as per 802.1AX
5.4.7) to remain selected (i.e., assigned to the aggregator). As the port
now remains selected, the aggregator selection logic is not invoked.
A side effect of this change is that aggregators in bonding will
now contain ports that are link down. The aggregator selection logic
does not currently handle this situation correctly, causing incorrect
aggregator selection.
This patch makes two changes to repair the aggregator selection
logic in bonding to function as documented and within the confines of the
standard:
First, the aggregator selection and related logic now utilizes the
number of active ports per aggregator, not the number of selected ports
(as some selected ports may be down). The ad_select "bandwidth" and
"count" options only consider ports that are link up.
Second, on any carrier state change of any slave, the aggregator
selection logic is explicitly called to insure the correct aggregator is
active.
Reported-by: Veli-Matti Lintu <veli-matti.lintu@opinsys.fi>
Fixes: 7bb11dc9f59d ("bonding: unify all places where actor-oper key needs to be updated.")
Signed-off-by: Jay Vosburgh <jay.vosburgh@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2016-06-24 05:20:51 +08:00
|
|
|
bandwidth = nports * 2500;
|
2014-11-19 16:48:59 +08:00
|
|
|
break;
|
2014-11-19 16:48:58 +08:00
|
|
|
case AD_LINK_SPEED_10000MBPS:
|
bonding: fix 802.3ad aggregator reselection
Since commit 7bb11dc9f59d ("bonding: unify all places where
actor-oper key needs to be updated."), the logic in bonding to handle
selection between multiple aggregators has not functioned.
This affects only configurations wherein the bonding slaves
connect to two discrete aggregators (e.g., two independent switches, each
with LACP enabled), thus creating two separate aggregation groups within a
single bond.
The cause is a change in 7bb11dc9f59d to no longer set
AD_PORT_BEGIN on a port after a link state change, which would cause the
port to be reselected for attachment to an aggregator as if were newly
added to the bond. We cannot restore the prior behavior, as it
contradicts IEEE 802.1AX 5.4.12, which requires ports that "become
inoperable" (lose carrier, setting port_enabled=false as per 802.1AX
5.4.7) to remain selected (i.e., assigned to the aggregator). As the port
now remains selected, the aggregator selection logic is not invoked.
A side effect of this change is that aggregators in bonding will
now contain ports that are link down. The aggregator selection logic
does not currently handle this situation correctly, causing incorrect
aggregator selection.
This patch makes two changes to repair the aggregator selection
logic in bonding to function as documented and within the confines of the
standard:
First, the aggregator selection and related logic now utilizes the
number of active ports per aggregator, not the number of selected ports
(as some selected ports may be down). The ad_select "bandwidth" and
"count" options only consider ports that are link up.
Second, on any carrier state change of any slave, the aggregator
selection logic is explicitly called to insure the correct aggregator is
active.
Reported-by: Veli-Matti Lintu <veli-matti.lintu@opinsys.fi>
Fixes: 7bb11dc9f59d ("bonding: unify all places where actor-oper key needs to be updated.")
Signed-off-by: Jay Vosburgh <jay.vosburgh@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2016-06-24 05:20:51 +08:00
|
|
|
bandwidth = nports * 10000;
|
2006-09-23 12:52:15 +08:00
|
|
|
break;
|
2014-11-19 16:48:59 +08:00
|
|
|
case AD_LINK_SPEED_20000MBPS:
|
bonding: fix 802.3ad aggregator reselection
Since commit 7bb11dc9f59d ("bonding: unify all places where
actor-oper key needs to be updated."), the logic in bonding to handle
selection between multiple aggregators has not functioned.
This affects only configurations wherein the bonding slaves
connect to two discrete aggregators (e.g., two independent switches, each
with LACP enabled), thus creating two separate aggregation groups within a
single bond.
The cause is a change in 7bb11dc9f59d to no longer set
AD_PORT_BEGIN on a port after a link state change, which would cause the
port to be reselected for attachment to an aggregator as if were newly
added to the bond. We cannot restore the prior behavior, as it
contradicts IEEE 802.1AX 5.4.12, which requires ports that "become
inoperable" (lose carrier, setting port_enabled=false as per 802.1AX
5.4.7) to remain selected (i.e., assigned to the aggregator). As the port
now remains selected, the aggregator selection logic is not invoked.
A side effect of this change is that aggregators in bonding will
now contain ports that are link down. The aggregator selection logic
does not currently handle this situation correctly, causing incorrect
aggregator selection.
This patch makes two changes to repair the aggregator selection
logic in bonding to function as documented and within the confines of the
standard:
First, the aggregator selection and related logic now utilizes the
number of active ports per aggregator, not the number of selected ports
(as some selected ports may be down). The ad_select "bandwidth" and
"count" options only consider ports that are link up.
Second, on any carrier state change of any slave, the aggregator
selection logic is explicitly called to insure the correct aggregator is
active.
Reported-by: Veli-Matti Lintu <veli-matti.lintu@opinsys.fi>
Fixes: 7bb11dc9f59d ("bonding: unify all places where actor-oper key needs to be updated.")
Signed-off-by: Jay Vosburgh <jay.vosburgh@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2016-06-24 05:20:51 +08:00
|
|
|
bandwidth = nports * 20000;
|
2014-11-19 16:48:59 +08:00
|
|
|
break;
|
|
|
|
case AD_LINK_SPEED_40000MBPS:
|
bonding: fix 802.3ad aggregator reselection
Since commit 7bb11dc9f59d ("bonding: unify all places where
actor-oper key needs to be updated."), the logic in bonding to handle
selection between multiple aggregators has not functioned.
This affects only configurations wherein the bonding slaves
connect to two discrete aggregators (e.g., two independent switches, each
with LACP enabled), thus creating two separate aggregation groups within a
single bond.
The cause is a change in 7bb11dc9f59d to no longer set
AD_PORT_BEGIN on a port after a link state change, which would cause the
port to be reselected for attachment to an aggregator as if were newly
added to the bond. We cannot restore the prior behavior, as it
contradicts IEEE 802.1AX 5.4.12, which requires ports that "become
inoperable" (lose carrier, setting port_enabled=false as per 802.1AX
5.4.7) to remain selected (i.e., assigned to the aggregator). As the port
now remains selected, the aggregator selection logic is not invoked.
A side effect of this change is that aggregators in bonding will
now contain ports that are link down. The aggregator selection logic
does not currently handle this situation correctly, causing incorrect
aggregator selection.
This patch makes two changes to repair the aggregator selection
logic in bonding to function as documented and within the confines of the
standard:
First, the aggregator selection and related logic now utilizes the
number of active ports per aggregator, not the number of selected ports
(as some selected ports may be down). The ad_select "bandwidth" and
"count" options only consider ports that are link up.
Second, on any carrier state change of any slave, the aggregator
selection logic is explicitly called to insure the correct aggregator is
active.
Reported-by: Veli-Matti Lintu <veli-matti.lintu@opinsys.fi>
Fixes: 7bb11dc9f59d ("bonding: unify all places where actor-oper key needs to be updated.")
Signed-off-by: Jay Vosburgh <jay.vosburgh@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2016-06-24 05:20:51 +08:00
|
|
|
bandwidth = nports * 40000;
|
2014-11-19 16:48:59 +08:00
|
|
|
break;
|
|
|
|
case AD_LINK_SPEED_56000MBPS:
|
bonding: fix 802.3ad aggregator reselection
Since commit 7bb11dc9f59d ("bonding: unify all places where
actor-oper key needs to be updated."), the logic in bonding to handle
selection between multiple aggregators has not functioned.
This affects only configurations wherein the bonding slaves
connect to two discrete aggregators (e.g., two independent switches, each
with LACP enabled), thus creating two separate aggregation groups within a
single bond.
The cause is a change in 7bb11dc9f59d to no longer set
AD_PORT_BEGIN on a port after a link state change, which would cause the
port to be reselected for attachment to an aggregator as if were newly
added to the bond. We cannot restore the prior behavior, as it
contradicts IEEE 802.1AX 5.4.12, which requires ports that "become
inoperable" (lose carrier, setting port_enabled=false as per 802.1AX
5.4.7) to remain selected (i.e., assigned to the aggregator). As the port
now remains selected, the aggregator selection logic is not invoked.
A side effect of this change is that aggregators in bonding will
now contain ports that are link down. The aggregator selection logic
does not currently handle this situation correctly, causing incorrect
aggregator selection.
This patch makes two changes to repair the aggregator selection
logic in bonding to function as documented and within the confines of the
standard:
First, the aggregator selection and related logic now utilizes the
number of active ports per aggregator, not the number of selected ports
(as some selected ports may be down). The ad_select "bandwidth" and
"count" options only consider ports that are link up.
Second, on any carrier state change of any slave, the aggregator
selection logic is explicitly called to insure the correct aggregator is
active.
Reported-by: Veli-Matti Lintu <veli-matti.lintu@opinsys.fi>
Fixes: 7bb11dc9f59d ("bonding: unify all places where actor-oper key needs to be updated.")
Signed-off-by: Jay Vosburgh <jay.vosburgh@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2016-06-24 05:20:51 +08:00
|
|
|
bandwidth = nports * 56000;
|
2014-11-19 16:48:59 +08:00
|
|
|
break;
|
2015-12-03 19:12:05 +08:00
|
|
|
case AD_LINK_SPEED_100000MBPS:
|
bonding: fix 802.3ad aggregator reselection
Since commit 7bb11dc9f59d ("bonding: unify all places where
actor-oper key needs to be updated."), the logic in bonding to handle
selection between multiple aggregators has not functioned.
This affects only configurations wherein the bonding slaves
connect to two discrete aggregators (e.g., two independent switches, each
with LACP enabled), thus creating two separate aggregation groups within a
single bond.
The cause is a change in 7bb11dc9f59d to no longer set
AD_PORT_BEGIN on a port after a link state change, which would cause the
port to be reselected for attachment to an aggregator as if were newly
added to the bond. We cannot restore the prior behavior, as it
contradicts IEEE 802.1AX 5.4.12, which requires ports that "become
inoperable" (lose carrier, setting port_enabled=false as per 802.1AX
5.4.7) to remain selected (i.e., assigned to the aggregator). As the port
now remains selected, the aggregator selection logic is not invoked.
A side effect of this change is that aggregators in bonding will
now contain ports that are link down. The aggregator selection logic
does not currently handle this situation correctly, causing incorrect
aggregator selection.
This patch makes two changes to repair the aggregator selection
logic in bonding to function as documented and within the confines of the
standard:
First, the aggregator selection and related logic now utilizes the
number of active ports per aggregator, not the number of selected ports
(as some selected ports may be down). The ad_select "bandwidth" and
"count" options only consider ports that are link up.
Second, on any carrier state change of any slave, the aggregator
selection logic is explicitly called to insure the correct aggregator is
active.
Reported-by: Veli-Matti Lintu <veli-matti.lintu@opinsys.fi>
Fixes: 7bb11dc9f59d ("bonding: unify all places where actor-oper key needs to be updated.")
Signed-off-by: Jay Vosburgh <jay.vosburgh@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2016-06-24 05:20:51 +08:00
|
|
|
bandwidth = nports * 100000;
|
2015-12-03 19:12:05 +08:00
|
|
|
break;
|
2005-04-17 06:20:36 +08:00
|
|
|
default:
|
2014-01-08 23:46:46 +08:00
|
|
|
bandwidth = 0; /* to silence the compiler */
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
return bandwidth;
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* __get_active_agg - get the current active aggregator
|
|
|
|
* @aggregator: the aggregator we're looking at
|
2014-01-10 18:59:45 +08:00
|
|
|
*
|
|
|
|
* Caller must hold RCU lock.
|
2005-04-17 06:20:36 +08:00
|
|
|
*/
|
|
|
|
static struct aggregator *__get_active_agg(struct aggregator *aggregator)
|
|
|
|
{
|
2013-09-27 22:12:00 +08:00
|
|
|
struct bonding *bond = aggregator->slave->bond;
|
|
|
|
struct list_head *iter;
|
|
|
|
struct slave *slave;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
bonding: add RCU for bond_3ad_state_machine_handler()
The bond_3ad_state_machine_handler() use the bond lock to protect
the bond slave list and slave port together, but it is not enough,
the bond slave list was link and unlink in RTNL, not bond lock,
so I add RCU to protect the slave list from leaving.
The bond lock is still used here, because when the slave has been
removed from the list by the time the state machine runs, it appears
to be possible for both function to manupulate the same aggregator->lag_ports
by finding the aggregator via two different ports that are both members of
that aggregator (i.e., port A of the agg is being unbound, and port B
of the agg is runing its state machine).
If I remove the bond lock, there are nothing to mutex changes
to aggregator->lag_ports between bond_3ad_state_machine_handler and
bond_3ad_unbind_slave, So the bond lock is the simplest way to protect
aggregator->lag_ports.
There was a lot of function need RCU protect, I have two choice
to make the function in RCU-safe, (1) create new similar functions
and make the bond slave list in RCU. (2) modify the existed functions
and make them in read-side critical section, because the RCU
read-side critical sections may be nested.
I choose (2) because it is no need to create more similar functions.
The nots in the function is still too old, clean up the nots.
Suggested-by: Nikolay Aleksandrov <nikolay@redhat.com>
Suggested-by: Jay Vosburgh <fubar@us.ibm.com>
Suggested-by: Veaceslav Falico <vfalico@redhat.com>
Signed-off-by: Ding Tianhong <dingtianhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-12-13 10:20:12 +08:00
|
|
|
bond_for_each_slave_rcu(bond, slave, iter)
|
2014-05-12 15:08:43 +08:00
|
|
|
if (SLAVE_AD_INFO(slave)->aggregator.is_active)
|
|
|
|
return &(SLAVE_AD_INFO(slave)->aggregator);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2013-09-27 22:12:00 +08:00
|
|
|
return NULL;
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* __update_lacpdu_from_port - update a port's lacpdu fields
|
|
|
|
* @port: the port we're looking at
|
|
|
|
*/
|
|
|
|
static inline void __update_lacpdu_from_port(struct port *port)
|
|
|
|
{
|
|
|
|
struct lacpdu *lacpdu = &port->lacpdu;
|
2008-12-18 11:13:53 +08:00
|
|
|
const struct port_params *partner = &port->partner_oper;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2014-01-08 23:46:46 +08:00
|
|
|
/* update current actual Actor parameters
|
|
|
|
* lacpdu->subtype initialized
|
2005-04-17 06:20:36 +08:00
|
|
|
* lacpdu->version_number initialized
|
|
|
|
* lacpdu->tlv_type_actor_info initialized
|
|
|
|
* lacpdu->actor_information_length initialized
|
|
|
|
*/
|
|
|
|
|
2007-08-23 08:06:58 +08:00
|
|
|
lacpdu->actor_system_priority = htons(port->actor_system_priority);
|
2005-04-17 06:20:36 +08:00
|
|
|
lacpdu->actor_system = port->actor_system;
|
2007-08-23 08:06:58 +08:00
|
|
|
lacpdu->actor_key = htons(port->actor_oper_port_key);
|
|
|
|
lacpdu->actor_port_priority = htons(port->actor_port_priority);
|
|
|
|
lacpdu->actor_port = htons(port->actor_port_number);
|
2005-04-17 06:20:36 +08:00
|
|
|
lacpdu->actor_state = port->actor_oper_port_state;
|
2015-01-26 14:16:59 +08:00
|
|
|
pr_debug("update lacpdu: %s, actor port state %x\n",
|
|
|
|
port->slave->dev->name, port->actor_oper_port_state);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
/* lacpdu->reserved_3_1 initialized
|
|
|
|
* lacpdu->tlv_type_partner_info initialized
|
|
|
|
* lacpdu->partner_information_length initialized
|
|
|
|
*/
|
|
|
|
|
2008-12-18 11:13:53 +08:00
|
|
|
lacpdu->partner_system_priority = htons(partner->system_priority);
|
|
|
|
lacpdu->partner_system = partner->system;
|
|
|
|
lacpdu->partner_key = htons(partner->key);
|
|
|
|
lacpdu->partner_port_priority = htons(partner->port_priority);
|
|
|
|
lacpdu->partner_port = htons(partner->port_number);
|
|
|
|
lacpdu->partner_state = partner->port_state;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
/* lacpdu->reserved_3_2 initialized
|
|
|
|
* lacpdu->tlv_type_collector_info initialized
|
|
|
|
* lacpdu->collector_information_length initialized
|
|
|
|
* collector_max_delay initialized
|
|
|
|
* reserved_12[12] initialized
|
|
|
|
* tlv_type_terminator initialized
|
|
|
|
* terminator_length initialized
|
|
|
|
* reserved_50[50] initialized
|
|
|
|
*/
|
|
|
|
}
|
|
|
|
|
2014-01-08 23:46:46 +08:00
|
|
|
/* ================= main 802.3ad protocol code ========================= */
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
/**
|
|
|
|
* ad_lacpdu_send - send out a lacpdu packet on a given port
|
|
|
|
* @port: the port we're looking at
|
|
|
|
*
|
|
|
|
* Returns: 0 on success
|
|
|
|
* < 0 on error
|
|
|
|
*/
|
|
|
|
static int ad_lacpdu_send(struct port *port)
|
|
|
|
{
|
|
|
|
struct slave *slave = port->slave;
|
|
|
|
struct sk_buff *skb;
|
|
|
|
struct lacpdu_header *lacpdu_header;
|
|
|
|
int length = sizeof(struct lacpdu_header);
|
|
|
|
|
|
|
|
skb = dev_alloc_skb(length);
|
2010-10-17 04:19:59 +08:00
|
|
|
if (!skb)
|
2005-04-17 06:20:36 +08:00
|
|
|
return -ENOMEM;
|
|
|
|
|
|
|
|
skb->dev = slave->dev;
|
2007-03-20 06:30:44 +08:00
|
|
|
skb_reset_mac_header(skb);
|
2007-04-11 12:21:55 +08:00
|
|
|
skb->network_header = skb->mac_header + ETH_HLEN;
|
2005-04-17 06:20:36 +08:00
|
|
|
skb->protocol = PKT_TYPE_LACPDU;
|
|
|
|
skb->priority = TC_PRIO_CONTROL;
|
|
|
|
|
|
|
|
lacpdu_header = (struct lacpdu_header *)skb_put(skb, length);
|
|
|
|
|
2014-02-16 08:02:17 +08:00
|
|
|
ether_addr_copy(lacpdu_header->hdr.h_dest, lacpdu_mcast_addr);
|
tree-wide: fix comment/printk typos
"gadget", "through", "command", "maintain", "maintain", "controller", "address",
"between", "initiali[zs]e", "instead", "function", "select", "already",
"equal", "access", "management", "hierarchy", "registration", "interest",
"relative", "memory", "offset", "already",
Signed-off-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de>
Signed-off-by: Jiri Kosina <jkosina@suse.cz>
2010-11-02 03:38:34 +08:00
|
|
|
/* Note: source address is set to be the member's PERMANENT address,
|
2014-01-08 23:46:46 +08:00
|
|
|
* because we use it to identify loopback lacpdus in receive.
|
|
|
|
*/
|
2014-02-16 08:02:17 +08:00
|
|
|
ether_addr_copy(lacpdu_header->hdr.h_source, slave->perm_hwaddr);
|
2008-12-27 05:41:53 +08:00
|
|
|
lacpdu_header->hdr.h_proto = PKT_TYPE_LACPDU;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2014-01-08 23:46:46 +08:00
|
|
|
lacpdu_header->lacpdu = port->lacpdu;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
dev_queue_xmit(skb);
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* ad_marker_send - send marker information/response on a given port
|
|
|
|
* @port: the port we're looking at
|
|
|
|
* @marker: marker data to send
|
|
|
|
*
|
|
|
|
* Returns: 0 on success
|
|
|
|
* < 0 on error
|
|
|
|
*/
|
2007-10-19 14:41:04 +08:00
|
|
|
static int ad_marker_send(struct port *port, struct bond_marker *marker)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
|
|
|
struct slave *slave = port->slave;
|
|
|
|
struct sk_buff *skb;
|
2007-10-19 14:41:04 +08:00
|
|
|
struct bond_marker_header *marker_header;
|
|
|
|
int length = sizeof(struct bond_marker_header);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
skb = dev_alloc_skb(length + 16);
|
2010-10-17 04:19:59 +08:00
|
|
|
if (!skb)
|
2005-04-17 06:20:36 +08:00
|
|
|
return -ENOMEM;
|
|
|
|
|
|
|
|
skb_reserve(skb, 16);
|
|
|
|
|
|
|
|
skb->dev = slave->dev;
|
2007-03-20 06:30:44 +08:00
|
|
|
skb_reset_mac_header(skb);
|
2007-04-11 12:21:55 +08:00
|
|
|
skb->network_header = skb->mac_header + ETH_HLEN;
|
2005-04-17 06:20:36 +08:00
|
|
|
skb->protocol = PKT_TYPE_LACPDU;
|
|
|
|
|
2007-10-19 14:41:04 +08:00
|
|
|
marker_header = (struct bond_marker_header *)skb_put(skb, length);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2014-02-16 08:02:17 +08:00
|
|
|
ether_addr_copy(marker_header->hdr.h_dest, lacpdu_mcast_addr);
|
tree-wide: fix comment/printk typos
"gadget", "through", "command", "maintain", "maintain", "controller", "address",
"between", "initiali[zs]e", "instead", "function", "select", "already",
"equal", "access", "management", "hierarchy", "registration", "interest",
"relative", "memory", "offset", "already",
Signed-off-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de>
Signed-off-by: Jiri Kosina <jkosina@suse.cz>
2010-11-02 03:38:34 +08:00
|
|
|
/* Note: source address is set to be the member's PERMANENT address,
|
2014-01-08 23:46:46 +08:00
|
|
|
* because we use it to identify loopback MARKERs in receive.
|
|
|
|
*/
|
2014-02-16 08:02:17 +08:00
|
|
|
ether_addr_copy(marker_header->hdr.h_source, slave->perm_hwaddr);
|
2008-12-27 05:41:53 +08:00
|
|
|
marker_header->hdr.h_proto = PKT_TYPE_LACPDU;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2014-01-08 23:46:46 +08:00
|
|
|
marker_header->marker = *marker;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
dev_queue_xmit(skb);
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* ad_mux_machine - handle a port's mux state machine
|
|
|
|
* @port: the port we're looking at
|
2014-10-05 08:45:01 +08:00
|
|
|
* @update_slave_arr: Does slave array need update?
|
2005-04-17 06:20:36 +08:00
|
|
|
*/
|
2014-10-05 08:45:01 +08:00
|
|
|
static void ad_mux_machine(struct port *port, bool *update_slave_arr)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
|
|
|
mux_states_t last_state;
|
|
|
|
|
2014-01-08 23:46:46 +08:00
|
|
|
/* keep current State Machine state to compare later if it was
|
|
|
|
* changed
|
|
|
|
*/
|
2005-04-17 06:20:36 +08:00
|
|
|
last_state = port->sm_mux_state;
|
|
|
|
|
|
|
|
if (port->sm_vars & AD_PORT_BEGIN) {
|
2014-01-08 23:46:46 +08:00
|
|
|
port->sm_mux_state = AD_MUX_DETACHED;
|
2005-04-17 06:20:36 +08:00
|
|
|
} else {
|
|
|
|
switch (port->sm_mux_state) {
|
|
|
|
case AD_MUX_DETACHED:
|
2010-10-17 04:19:59 +08:00
|
|
|
if ((port->sm_vars & AD_PORT_SELECTED)
|
|
|
|
|| (port->sm_vars & AD_PORT_STANDBY))
|
|
|
|
/* if SELECTED or STANDBY */
|
2014-01-08 23:46:46 +08:00
|
|
|
port->sm_mux_state = AD_MUX_WAITING;
|
2005-04-17 06:20:36 +08:00
|
|
|
break;
|
|
|
|
case AD_MUX_WAITING:
|
2014-01-08 23:46:46 +08:00
|
|
|
/* if SELECTED == FALSE return to DETACH state */
|
|
|
|
if (!(port->sm_vars & AD_PORT_SELECTED)) {
|
2005-04-17 06:20:36 +08:00
|
|
|
port->sm_vars &= ~AD_PORT_READY_N;
|
2014-01-08 23:46:46 +08:00
|
|
|
/* in order to withhold the Selection Logic to
|
|
|
|
* check all ports READY_N value every callback
|
|
|
|
* cycle to update ready variable, we check
|
|
|
|
* READY_N and update READY here
|
|
|
|
*/
|
2005-04-17 06:20:36 +08:00
|
|
|
__set_agg_ports_ready(port->aggregator, __agg_ports_are_ready(port->aggregator));
|
2014-01-08 23:46:46 +08:00
|
|
|
port->sm_mux_state = AD_MUX_DETACHED;
|
2005-04-17 06:20:36 +08:00
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
2014-01-08 23:46:46 +08:00
|
|
|
/* check if the wait_while_timer expired */
|
2010-10-17 04:19:59 +08:00
|
|
|
if (port->sm_mux_timer_counter
|
|
|
|
&& !(--port->sm_mux_timer_counter))
|
2005-04-17 06:20:36 +08:00
|
|
|
port->sm_vars |= AD_PORT_READY_N;
|
|
|
|
|
2014-01-08 23:46:46 +08:00
|
|
|
/* in order to withhold the selection logic to check
|
|
|
|
* all ports READY_N value every callback cycle to
|
|
|
|
* update ready variable, we check READY_N and update
|
|
|
|
* READY here
|
|
|
|
*/
|
2005-04-17 06:20:36 +08:00
|
|
|
__set_agg_ports_ready(port->aggregator, __agg_ports_are_ready(port->aggregator));
|
|
|
|
|
2014-01-08 23:46:46 +08:00
|
|
|
/* if the wait_while_timer expired, and the port is
|
|
|
|
* in READY state, move to ATTACHED state
|
|
|
|
*/
|
2010-10-17 04:19:59 +08:00
|
|
|
if ((port->sm_vars & AD_PORT_READY)
|
|
|
|
&& !port->sm_mux_timer_counter)
|
2014-01-08 23:46:46 +08:00
|
|
|
port->sm_mux_state = AD_MUX_ATTACHED;
|
2005-04-17 06:20:36 +08:00
|
|
|
break;
|
|
|
|
case AD_MUX_ATTACHED:
|
2014-01-08 23:46:46 +08:00
|
|
|
/* check also if agg_select_timer expired (so the
|
|
|
|
* edable port will take place only after this timer)
|
|
|
|
*/
|
|
|
|
if ((port->sm_vars & AD_PORT_SELECTED) &&
|
|
|
|
(port->partner_oper.port_state & AD_STATE_SYNCHRONIZATION) &&
|
|
|
|
!__check_agg_selection_timer(port)) {
|
2015-01-26 14:16:59 +08:00
|
|
|
if (port->aggregator->is_active)
|
|
|
|
port->sm_mux_state =
|
|
|
|
AD_MUX_COLLECTING_DISTRIBUTING;
|
2014-01-08 23:46:46 +08:00
|
|
|
} else if (!(port->sm_vars & AD_PORT_SELECTED) ||
|
|
|
|
(port->sm_vars & AD_PORT_STANDBY)) {
|
|
|
|
/* if UNSELECTED or STANDBY */
|
2005-04-17 06:20:36 +08:00
|
|
|
port->sm_vars &= ~AD_PORT_READY_N;
|
2014-01-08 23:46:46 +08:00
|
|
|
/* in order to withhold the selection logic to
|
|
|
|
* check all ports READY_N value every callback
|
|
|
|
* cycle to update ready variable, we check
|
|
|
|
* READY_N and update READY here
|
|
|
|
*/
|
2005-04-17 06:20:36 +08:00
|
|
|
__set_agg_ports_ready(port->aggregator, __agg_ports_are_ready(port->aggregator));
|
2014-01-08 23:46:46 +08:00
|
|
|
port->sm_mux_state = AD_MUX_DETACHED;
|
2015-01-26 14:16:59 +08:00
|
|
|
} else if (port->aggregator->is_active) {
|
|
|
|
port->actor_oper_port_state |=
|
|
|
|
AD_STATE_SYNCHRONIZATION;
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
break;
|
|
|
|
case AD_MUX_COLLECTING_DISTRIBUTING:
|
2014-01-08 23:46:46 +08:00
|
|
|
if (!(port->sm_vars & AD_PORT_SELECTED) ||
|
|
|
|
(port->sm_vars & AD_PORT_STANDBY) ||
|
2015-01-26 14:16:59 +08:00
|
|
|
!(port->partner_oper.port_state & AD_STATE_SYNCHRONIZATION) ||
|
|
|
|
!(port->actor_oper_port_state & AD_STATE_SYNCHRONIZATION)) {
|
2014-01-08 23:46:46 +08:00
|
|
|
port->sm_mux_state = AD_MUX_ATTACHED;
|
2005-04-17 06:20:36 +08:00
|
|
|
} else {
|
2014-01-08 23:46:46 +08:00
|
|
|
/* if port state hasn't changed make
|
|
|
|
* sure that a collecting distributing
|
|
|
|
* port in an active aggregator is enabled
|
|
|
|
*/
|
2005-04-17 06:20:36 +08:00
|
|
|
if (port->aggregator &&
|
|
|
|
port->aggregator->is_active &&
|
|
|
|
!__port_is_enabled(port)) {
|
|
|
|
|
|
|
|
__enable_port(port);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
break;
|
2014-01-08 23:46:46 +08:00
|
|
|
default:
|
2005-04-17 06:20:36 +08:00
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2014-01-08 23:46:46 +08:00
|
|
|
/* check if the state machine was changed */
|
2005-04-17 06:20:36 +08:00
|
|
|
if (port->sm_mux_state != last_state) {
|
2015-01-26 14:16:59 +08:00
|
|
|
pr_debug("Mux Machine: Port=%d (%s), Last State=%d, Curr State=%d\n",
|
|
|
|
port->actor_port_number,
|
|
|
|
port->slave->dev->name,
|
|
|
|
last_state,
|
2009-12-14 12:06:07 +08:00
|
|
|
port->sm_mux_state);
|
2005-04-17 06:20:36 +08:00
|
|
|
switch (port->sm_mux_state) {
|
|
|
|
case AD_MUX_DETACHED:
|
|
|
|
port->actor_oper_port_state &= ~AD_STATE_SYNCHRONIZATION;
|
2014-10-05 08:45:01 +08:00
|
|
|
ad_disable_collecting_distributing(port,
|
|
|
|
update_slave_arr);
|
2005-04-17 06:20:36 +08:00
|
|
|
port->actor_oper_port_state &= ~AD_STATE_COLLECTING;
|
|
|
|
port->actor_oper_port_state &= ~AD_STATE_DISTRIBUTING;
|
2008-12-27 03:18:15 +08:00
|
|
|
port->ntt = true;
|
2005-04-17 06:20:36 +08:00
|
|
|
break;
|
|
|
|
case AD_MUX_WAITING:
|
|
|
|
port->sm_mux_timer_counter = __ad_timer_to_ticks(AD_WAIT_WHILE_TIMER, 0);
|
|
|
|
break;
|
|
|
|
case AD_MUX_ATTACHED:
|
2015-01-26 14:16:59 +08:00
|
|
|
if (port->aggregator->is_active)
|
|
|
|
port->actor_oper_port_state |=
|
|
|
|
AD_STATE_SYNCHRONIZATION;
|
|
|
|
else
|
|
|
|
port->actor_oper_port_state &=
|
|
|
|
~AD_STATE_SYNCHRONIZATION;
|
2005-04-17 06:20:36 +08:00
|
|
|
port->actor_oper_port_state &= ~AD_STATE_COLLECTING;
|
|
|
|
port->actor_oper_port_state &= ~AD_STATE_DISTRIBUTING;
|
2014-10-05 08:45:01 +08:00
|
|
|
ad_disable_collecting_distributing(port,
|
|
|
|
update_slave_arr);
|
2008-12-27 03:18:15 +08:00
|
|
|
port->ntt = true;
|
2005-04-17 06:20:36 +08:00
|
|
|
break;
|
|
|
|
case AD_MUX_COLLECTING_DISTRIBUTING:
|
|
|
|
port->actor_oper_port_state |= AD_STATE_COLLECTING;
|
|
|
|
port->actor_oper_port_state |= AD_STATE_DISTRIBUTING;
|
2015-01-26 14:16:59 +08:00
|
|
|
port->actor_oper_port_state |= AD_STATE_SYNCHRONIZATION;
|
2014-10-05 08:45:01 +08:00
|
|
|
ad_enable_collecting_distributing(port,
|
|
|
|
update_slave_arr);
|
2008-12-27 03:18:15 +08:00
|
|
|
port->ntt = true;
|
2005-04-17 06:20:36 +08:00
|
|
|
break;
|
2014-01-08 23:46:46 +08:00
|
|
|
default:
|
2005-04-17 06:20:36 +08:00
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* ad_rx_machine - handle a port's rx State Machine
|
|
|
|
* @lacpdu: the lacpdu we've received
|
|
|
|
* @port: the port we're looking at
|
|
|
|
*
|
|
|
|
* If lacpdu arrived, stop previous timer (if exists) and set the next state as
|
|
|
|
* CURRENT. If timer expired set the state machine in the proper state.
|
|
|
|
* In other cases, this function checks if we need to switch to other state.
|
|
|
|
*/
|
|
|
|
static void ad_rx_machine(struct lacpdu *lacpdu, struct port *port)
|
|
|
|
{
|
|
|
|
rx_states_t last_state;
|
|
|
|
|
2014-01-08 23:46:46 +08:00
|
|
|
/* keep current State Machine state to compare later if it was
|
|
|
|
* changed
|
|
|
|
*/
|
2005-04-17 06:20:36 +08:00
|
|
|
last_state = port->sm_rx_state;
|
|
|
|
|
2014-01-08 23:46:46 +08:00
|
|
|
/* check if state machine should change state */
|
|
|
|
|
|
|
|
/* first, check if port was reinitialized */
|
2015-02-24 09:50:11 +08:00
|
|
|
if (port->sm_vars & AD_PORT_BEGIN) {
|
2010-10-17 04:19:59 +08:00
|
|
|
port->sm_rx_state = AD_RX_INITIALIZE;
|
2015-03-31 05:30:40 +08:00
|
|
|
port->sm_vars |= AD_PORT_CHURNED;
|
2014-01-08 23:46:46 +08:00
|
|
|
/* check if port is not enabled */
|
2015-02-24 09:50:11 +08:00
|
|
|
} else if (!(port->sm_vars & AD_PORT_BEGIN)
|
2010-10-17 04:19:59 +08:00
|
|
|
&& !port->is_enabled && !(port->sm_vars & AD_PORT_MOVED))
|
|
|
|
port->sm_rx_state = AD_RX_PORT_DISABLED;
|
2014-01-08 23:46:46 +08:00
|
|
|
/* check if new lacpdu arrived */
|
|
|
|
else if (lacpdu && ((port->sm_rx_state == AD_RX_EXPIRED) ||
|
|
|
|
(port->sm_rx_state == AD_RX_DEFAULTED) ||
|
|
|
|
(port->sm_rx_state == AD_RX_CURRENT))) {
|
2015-02-24 09:50:11 +08:00
|
|
|
if (port->sm_rx_state != AD_RX_CURRENT)
|
2015-03-31 05:30:40 +08:00
|
|
|
port->sm_vars |= AD_PORT_CHURNED;
|
2014-01-08 23:46:46 +08:00
|
|
|
port->sm_rx_timer_counter = 0;
|
2005-04-17 06:20:36 +08:00
|
|
|
port->sm_rx_state = AD_RX_CURRENT;
|
|
|
|
} else {
|
2014-01-08 23:46:46 +08:00
|
|
|
/* if timer is on, and if it is expired */
|
|
|
|
if (port->sm_rx_timer_counter &&
|
|
|
|
!(--port->sm_rx_timer_counter)) {
|
2005-04-17 06:20:36 +08:00
|
|
|
switch (port->sm_rx_state) {
|
|
|
|
case AD_RX_EXPIRED:
|
2014-01-08 23:46:46 +08:00
|
|
|
port->sm_rx_state = AD_RX_DEFAULTED;
|
2005-04-17 06:20:36 +08:00
|
|
|
break;
|
|
|
|
case AD_RX_CURRENT:
|
2014-01-08 23:46:46 +08:00
|
|
|
port->sm_rx_state = AD_RX_EXPIRED;
|
2005-04-17 06:20:36 +08:00
|
|
|
break;
|
2014-01-08 23:46:46 +08:00
|
|
|
default:
|
2005-04-17 06:20:36 +08:00
|
|
|
break;
|
|
|
|
}
|
|
|
|
} else {
|
2014-01-08 23:46:46 +08:00
|
|
|
/* if no lacpdu arrived and no timer is on */
|
2005-04-17 06:20:36 +08:00
|
|
|
switch (port->sm_rx_state) {
|
|
|
|
case AD_RX_PORT_DISABLED:
|
2010-10-17 04:19:59 +08:00
|
|
|
if (port->sm_vars & AD_PORT_MOVED)
|
2014-01-08 23:46:46 +08:00
|
|
|
port->sm_rx_state = AD_RX_INITIALIZE;
|
2010-10-17 04:19:59 +08:00
|
|
|
else if (port->is_enabled
|
|
|
|
&& (port->sm_vars
|
|
|
|
& AD_PORT_LACP_ENABLED))
|
2014-01-08 23:46:46 +08:00
|
|
|
port->sm_rx_state = AD_RX_EXPIRED;
|
2010-10-17 04:19:59 +08:00
|
|
|
else if (port->is_enabled
|
|
|
|
&& ((port->sm_vars
|
|
|
|
& AD_PORT_LACP_ENABLED) == 0))
|
2014-01-08 23:46:46 +08:00
|
|
|
port->sm_rx_state = AD_RX_LACP_DISABLED;
|
2005-04-17 06:20:36 +08:00
|
|
|
break;
|
2014-01-08 23:46:46 +08:00
|
|
|
default:
|
2005-04-17 06:20:36 +08:00
|
|
|
break;
|
|
|
|
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2014-01-08 23:46:46 +08:00
|
|
|
/* check if the State machine was changed or new lacpdu arrived */
|
2005-04-17 06:20:36 +08:00
|
|
|
if ((port->sm_rx_state != last_state) || (lacpdu)) {
|
2015-01-26 14:16:59 +08:00
|
|
|
pr_debug("Rx Machine: Port=%d (%s), Last State=%d, Curr State=%d\n",
|
|
|
|
port->actor_port_number,
|
|
|
|
port->slave->dev->name,
|
|
|
|
last_state,
|
2009-12-14 12:06:07 +08:00
|
|
|
port->sm_rx_state);
|
2005-04-17 06:20:36 +08:00
|
|
|
switch (port->sm_rx_state) {
|
|
|
|
case AD_RX_INITIALIZE:
|
2014-11-19 16:48:58 +08:00
|
|
|
if (!(port->actor_oper_port_key & AD_DUPLEX_KEY_MASKS))
|
2005-04-17 06:20:36 +08:00
|
|
|
port->sm_vars &= ~AD_PORT_LACP_ENABLED;
|
2010-10-17 04:19:59 +08:00
|
|
|
else
|
2005-04-17 06:20:36 +08:00
|
|
|
port->sm_vars |= AD_PORT_LACP_ENABLED;
|
|
|
|
port->sm_vars &= ~AD_PORT_SELECTED;
|
|
|
|
__record_default(port);
|
|
|
|
port->actor_oper_port_state &= ~AD_STATE_EXPIRED;
|
|
|
|
port->sm_vars &= ~AD_PORT_MOVED;
|
2014-01-08 23:46:46 +08:00
|
|
|
port->sm_rx_state = AD_RX_PORT_DISABLED;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2014-01-08 23:46:46 +08:00
|
|
|
/* Fall Through */
|
2005-04-17 06:20:36 +08:00
|
|
|
case AD_RX_PORT_DISABLED:
|
|
|
|
port->sm_vars &= ~AD_PORT_MATCHED;
|
|
|
|
break;
|
|
|
|
case AD_RX_LACP_DISABLED:
|
|
|
|
port->sm_vars &= ~AD_PORT_SELECTED;
|
|
|
|
__record_default(port);
|
2008-12-18 11:07:38 +08:00
|
|
|
port->partner_oper.port_state &= ~AD_STATE_AGGREGATION;
|
2005-04-17 06:20:36 +08:00
|
|
|
port->sm_vars |= AD_PORT_MATCHED;
|
|
|
|
port->actor_oper_port_state &= ~AD_STATE_EXPIRED;
|
|
|
|
break;
|
|
|
|
case AD_RX_EXPIRED:
|
2014-01-08 23:46:46 +08:00
|
|
|
/* Reset of the Synchronization flag (Standard 43.4.12)
|
|
|
|
* This reset cause to disable this port in the
|
|
|
|
* COLLECTING_DISTRIBUTING state of the mux machine in
|
|
|
|
* case of EXPIRED even if LINK_DOWN didn't arrive for
|
|
|
|
* the port.
|
|
|
|
*/
|
2008-12-18 11:07:38 +08:00
|
|
|
port->partner_oper.port_state &= ~AD_STATE_SYNCHRONIZATION;
|
2005-04-17 06:20:36 +08:00
|
|
|
port->sm_vars &= ~AD_PORT_MATCHED;
|
2015-02-24 09:50:11 +08:00
|
|
|
port->partner_oper.port_state |= AD_STATE_LACP_TIMEOUT;
|
2014-01-08 23:46:46 +08:00
|
|
|
port->partner_oper.port_state |= AD_STATE_LACP_ACTIVITY;
|
2005-04-17 06:20:36 +08:00
|
|
|
port->sm_rx_timer_counter = __ad_timer_to_ticks(AD_CURRENT_WHILE_TIMER, (u16)(AD_SHORT_TIMEOUT));
|
|
|
|
port->actor_oper_port_state |= AD_STATE_EXPIRED;
|
2015-03-31 05:30:40 +08:00
|
|
|
port->sm_vars |= AD_PORT_CHURNED;
|
2005-04-17 06:20:36 +08:00
|
|
|
break;
|
|
|
|
case AD_RX_DEFAULTED:
|
|
|
|
__update_default_selected(port);
|
|
|
|
__record_default(port);
|
|
|
|
port->sm_vars |= AD_PORT_MATCHED;
|
|
|
|
port->actor_oper_port_state &= ~AD_STATE_EXPIRED;
|
|
|
|
break;
|
|
|
|
case AD_RX_CURRENT:
|
2014-01-02 09:12:54 +08:00
|
|
|
/* detect loopback situation */
|
2014-01-08 23:46:46 +08:00
|
|
|
if (MAC_ADDRESS_EQUAL(&(lacpdu->actor_system),
|
|
|
|
&(port->actor_system))) {
|
2014-07-16 01:36:00 +08:00
|
|
|
netdev_err(port->slave->bond->dev, "An illegal loopback occurred on adapter (%s)\n"
|
2014-02-16 08:01:45 +08:00
|
|
|
"Check the configuration to verify that all adapters are connected to 802.3ad compliant switch ports\n",
|
2014-01-08 23:46:46 +08:00
|
|
|
port->slave->dev->name);
|
2005-04-17 06:20:36 +08:00
|
|
|
return;
|
|
|
|
}
|
|
|
|
__update_selected(lacpdu, port);
|
|
|
|
__update_ntt(lacpdu, port);
|
|
|
|
__record_pdu(lacpdu, port);
|
|
|
|
port->sm_rx_timer_counter = __ad_timer_to_ticks(AD_CURRENT_WHILE_TIMER, (u16)(port->actor_oper_port_state & AD_STATE_LACP_TIMEOUT));
|
|
|
|
port->actor_oper_port_state &= ~AD_STATE_EXPIRED;
|
|
|
|
break;
|
2014-01-08 23:46:46 +08:00
|
|
|
default:
|
2005-04-17 06:20:36 +08:00
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2015-02-24 09:50:11 +08:00
|
|
|
/**
|
|
|
|
* ad_churn_machine - handle port churn's state machine
|
|
|
|
* @port: the port we're looking at
|
|
|
|
*
|
|
|
|
*/
|
|
|
|
static void ad_churn_machine(struct port *port)
|
|
|
|
{
|
2015-03-31 05:30:40 +08:00
|
|
|
if (port->sm_vars & AD_PORT_CHURNED) {
|
|
|
|
port->sm_vars &= ~AD_PORT_CHURNED;
|
2015-02-24 09:50:11 +08:00
|
|
|
port->sm_churn_actor_state = AD_CHURN_MONITOR;
|
|
|
|
port->sm_churn_partner_state = AD_CHURN_MONITOR;
|
|
|
|
port->sm_churn_actor_timer_counter =
|
|
|
|
__ad_timer_to_ticks(AD_ACTOR_CHURN_TIMER, 0);
|
|
|
|
port->sm_churn_partner_timer_counter =
|
|
|
|
__ad_timer_to_ticks(AD_PARTNER_CHURN_TIMER, 0);
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
if (port->sm_churn_actor_timer_counter &&
|
|
|
|
!(--port->sm_churn_actor_timer_counter) &&
|
|
|
|
port->sm_churn_actor_state == AD_CHURN_MONITOR) {
|
|
|
|
if (port->actor_oper_port_state & AD_STATE_SYNCHRONIZATION) {
|
|
|
|
port->sm_churn_actor_state = AD_NO_CHURN;
|
|
|
|
} else {
|
|
|
|
port->churn_actor_count++;
|
|
|
|
port->sm_churn_actor_state = AD_CHURN;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
if (port->sm_churn_partner_timer_counter &&
|
|
|
|
!(--port->sm_churn_partner_timer_counter) &&
|
|
|
|
port->sm_churn_partner_state == AD_CHURN_MONITOR) {
|
|
|
|
if (port->partner_oper.port_state & AD_STATE_SYNCHRONIZATION) {
|
|
|
|
port->sm_churn_partner_state = AD_NO_CHURN;
|
|
|
|
} else {
|
|
|
|
port->churn_partner_count++;
|
|
|
|
port->sm_churn_partner_state = AD_CHURN;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
/**
|
|
|
|
* ad_tx_machine - handle a port's tx state machine
|
|
|
|
* @port: the port we're looking at
|
|
|
|
*/
|
|
|
|
static void ad_tx_machine(struct port *port)
|
|
|
|
{
|
2014-01-08 23:46:46 +08:00
|
|
|
/* check if tx timer expired, to verify that we do not send more than
|
|
|
|
* 3 packets per second
|
|
|
|
*/
|
2005-04-17 06:20:36 +08:00
|
|
|
if (port->sm_tx_timer_counter && !(--port->sm_tx_timer_counter)) {
|
2014-01-08 23:46:46 +08:00
|
|
|
/* check if there is something to send */
|
2005-04-17 06:20:36 +08:00
|
|
|
if (port->ntt && (port->sm_vars & AD_PORT_LACP_ENABLED)) {
|
|
|
|
__update_lacpdu_from_port(port);
|
2008-12-27 03:18:15 +08:00
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
if (ad_lacpdu_send(port) >= 0) {
|
2009-12-14 12:06:07 +08:00
|
|
|
pr_debug("Sent LACPDU on port %d\n",
|
|
|
|
port->actor_port_number);
|
2008-12-27 03:18:15 +08:00
|
|
|
|
2014-01-08 23:46:46 +08:00
|
|
|
/* mark ntt as false, so it will not be sent
|
|
|
|
* again until demanded
|
|
|
|
*/
|
2008-12-27 03:18:15 +08:00
|
|
|
port->ntt = false;
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
}
|
2014-01-08 23:46:46 +08:00
|
|
|
/* restart tx timer(to verify that we will not exceed
|
|
|
|
* AD_MAX_TX_IN_SECOND
|
|
|
|
*/
|
|
|
|
port->sm_tx_timer_counter = ad_ticks_per_sec/AD_MAX_TX_IN_SECOND;
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* ad_periodic_machine - handle a port's periodic state machine
|
|
|
|
* @port: the port we're looking at
|
|
|
|
*
|
|
|
|
* Turn ntt flag on priodically to perform periodic transmission of lacpdu's.
|
|
|
|
*/
|
|
|
|
static void ad_periodic_machine(struct port *port)
|
|
|
|
{
|
|
|
|
periodic_states_t last_state;
|
|
|
|
|
2014-01-08 23:46:46 +08:00
|
|
|
/* keep current state machine state to compare later if it was changed */
|
2005-04-17 06:20:36 +08:00
|
|
|
last_state = port->sm_periodic_state;
|
|
|
|
|
2014-01-08 23:46:46 +08:00
|
|
|
/* check if port was reinitialized */
|
2005-04-17 06:20:36 +08:00
|
|
|
if (((port->sm_vars & AD_PORT_BEGIN) || !(port->sm_vars & AD_PORT_LACP_ENABLED) || !port->is_enabled) ||
|
2008-12-18 11:07:38 +08:00
|
|
|
(!(port->actor_oper_port_state & AD_STATE_LACP_ACTIVITY) && !(port->partner_oper.port_state & AD_STATE_LACP_ACTIVITY))
|
2005-04-17 06:20:36 +08:00
|
|
|
) {
|
2014-01-08 23:46:46 +08:00
|
|
|
port->sm_periodic_state = AD_NO_PERIODIC;
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
2014-01-08 23:46:46 +08:00
|
|
|
/* check if state machine should change state */
|
2005-04-17 06:20:36 +08:00
|
|
|
else if (port->sm_periodic_timer_counter) {
|
2014-01-08 23:46:46 +08:00
|
|
|
/* check if periodic state machine expired */
|
2005-04-17 06:20:36 +08:00
|
|
|
if (!(--port->sm_periodic_timer_counter)) {
|
2014-01-08 23:46:46 +08:00
|
|
|
/* if expired then do tx */
|
|
|
|
port->sm_periodic_state = AD_PERIODIC_TX;
|
2005-04-17 06:20:36 +08:00
|
|
|
} else {
|
2014-01-08 23:46:46 +08:00
|
|
|
/* If not expired, check if there is some new timeout
|
|
|
|
* parameter from the partner state
|
|
|
|
*/
|
2005-04-17 06:20:36 +08:00
|
|
|
switch (port->sm_periodic_state) {
|
|
|
|
case AD_FAST_PERIODIC:
|
2010-10-17 04:19:59 +08:00
|
|
|
if (!(port->partner_oper.port_state
|
|
|
|
& AD_STATE_LACP_TIMEOUT))
|
2014-01-08 23:46:46 +08:00
|
|
|
port->sm_periodic_state = AD_SLOW_PERIODIC;
|
2005-04-17 06:20:36 +08:00
|
|
|
break;
|
|
|
|
case AD_SLOW_PERIODIC:
|
2008-12-18 11:07:38 +08:00
|
|
|
if ((port->partner_oper.port_state & AD_STATE_LACP_TIMEOUT)) {
|
2005-04-17 06:20:36 +08:00
|
|
|
port->sm_periodic_timer_counter = 0;
|
2014-01-08 23:46:46 +08:00
|
|
|
port->sm_periodic_state = AD_PERIODIC_TX;
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
break;
|
2014-01-08 23:46:46 +08:00
|
|
|
default:
|
2005-04-17 06:20:36 +08:00
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
switch (port->sm_periodic_state) {
|
|
|
|
case AD_NO_PERIODIC:
|
2014-01-08 23:46:46 +08:00
|
|
|
port->sm_periodic_state = AD_FAST_PERIODIC;
|
2005-04-17 06:20:36 +08:00
|
|
|
break;
|
|
|
|
case AD_PERIODIC_TX:
|
2014-01-08 23:46:46 +08:00
|
|
|
if (!(port->partner_oper.port_state &
|
|
|
|
AD_STATE_LACP_TIMEOUT))
|
|
|
|
port->sm_periodic_state = AD_SLOW_PERIODIC;
|
2010-10-17 04:19:59 +08:00
|
|
|
else
|
2014-01-08 23:46:46 +08:00
|
|
|
port->sm_periodic_state = AD_FAST_PERIODIC;
|
2005-04-17 06:20:36 +08:00
|
|
|
break;
|
2014-01-08 23:46:46 +08:00
|
|
|
default:
|
2005-04-17 06:20:36 +08:00
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2014-01-08 23:46:46 +08:00
|
|
|
/* check if the state machine was changed */
|
2005-04-17 06:20:36 +08:00
|
|
|
if (port->sm_periodic_state != last_state) {
|
2009-12-14 12:06:07 +08:00
|
|
|
pr_debug("Periodic Machine: Port=%d, Last State=%d, Curr State=%d\n",
|
|
|
|
port->actor_port_number, last_state,
|
|
|
|
port->sm_periodic_state);
|
2005-04-17 06:20:36 +08:00
|
|
|
switch (port->sm_periodic_state) {
|
|
|
|
case AD_NO_PERIODIC:
|
2014-01-08 23:46:46 +08:00
|
|
|
port->sm_periodic_timer_counter = 0;
|
2005-04-17 06:20:36 +08:00
|
|
|
break;
|
|
|
|
case AD_FAST_PERIODIC:
|
2014-01-08 23:46:46 +08:00
|
|
|
/* decrement 1 tick we lost in the PERIODIC_TX cycle */
|
|
|
|
port->sm_periodic_timer_counter = __ad_timer_to_ticks(AD_PERIODIC_TIMER, (u16)(AD_FAST_PERIODIC_TIME))-1;
|
2005-04-17 06:20:36 +08:00
|
|
|
break;
|
|
|
|
case AD_SLOW_PERIODIC:
|
2014-01-08 23:46:46 +08:00
|
|
|
/* decrement 1 tick we lost in the PERIODIC_TX cycle */
|
|
|
|
port->sm_periodic_timer_counter = __ad_timer_to_ticks(AD_PERIODIC_TIMER, (u16)(AD_SLOW_PERIODIC_TIME))-1;
|
2005-04-17 06:20:36 +08:00
|
|
|
break;
|
|
|
|
case AD_PERIODIC_TX:
|
2008-12-27 03:18:15 +08:00
|
|
|
port->ntt = true;
|
2005-04-17 06:20:36 +08:00
|
|
|
break;
|
2014-01-08 23:46:46 +08:00
|
|
|
default:
|
2005-04-17 06:20:36 +08:00
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* ad_port_selection_logic - select aggregation groups
|
|
|
|
* @port: the port we're looking at
|
2014-10-05 08:45:01 +08:00
|
|
|
* @update_slave_arr: Does slave array need update?
|
2005-04-17 06:20:36 +08:00
|
|
|
*
|
|
|
|
* Select aggregation groups, and assign each port for it's aggregetor. The
|
|
|
|
* selection logic is called in the inititalization (after all the handshkes),
|
|
|
|
* and after every lacpdu receive (if selected is off).
|
|
|
|
*/
|
2014-10-05 08:45:01 +08:00
|
|
|
static void ad_port_selection_logic(struct port *port, bool *update_slave_arr)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
|
|
|
struct aggregator *aggregator, *free_aggregator = NULL, *temp_aggregator;
|
|
|
|
struct port *last_port = NULL, *curr_port;
|
2013-09-27 22:11:59 +08:00
|
|
|
struct list_head *iter;
|
|
|
|
struct bonding *bond;
|
|
|
|
struct slave *slave;
|
2005-04-17 06:20:36 +08:00
|
|
|
int found = 0;
|
|
|
|
|
2014-01-08 23:46:46 +08:00
|
|
|
/* if the port is already Selected, do nothing */
|
2010-10-17 04:19:59 +08:00
|
|
|
if (port->sm_vars & AD_PORT_SELECTED)
|
2005-04-17 06:20:36 +08:00
|
|
|
return;
|
|
|
|
|
2013-09-27 22:11:59 +08:00
|
|
|
bond = __get_bond_by_port(port);
|
|
|
|
|
2014-01-08 23:46:46 +08:00
|
|
|
/* if the port is connected to other aggregator, detach it */
|
2005-04-17 06:20:36 +08:00
|
|
|
if (port->aggregator) {
|
2014-01-08 23:46:46 +08:00
|
|
|
/* detach the port from its former aggregator */
|
2010-10-17 04:19:58 +08:00
|
|
|
temp_aggregator = port->aggregator;
|
|
|
|
for (curr_port = temp_aggregator->lag_ports; curr_port;
|
|
|
|
last_port = curr_port,
|
2014-01-08 23:46:46 +08:00
|
|
|
curr_port = curr_port->next_port_in_aggregator) {
|
2005-04-17 06:20:36 +08:00
|
|
|
if (curr_port == port) {
|
|
|
|
temp_aggregator->num_of_ports--;
|
2014-01-08 23:46:46 +08:00
|
|
|
/* if it is the first port attached to the
|
|
|
|
* aggregator
|
|
|
|
*/
|
|
|
|
if (!last_port) {
|
2010-10-17 04:19:58 +08:00
|
|
|
temp_aggregator->lag_ports =
|
|
|
|
port->next_port_in_aggregator;
|
2014-01-08 23:46:46 +08:00
|
|
|
} else {
|
|
|
|
/* not the first port attached to the
|
|
|
|
* aggregator
|
|
|
|
*/
|
2010-10-17 04:19:58 +08:00
|
|
|
last_port->next_port_in_aggregator =
|
|
|
|
port->next_port_in_aggregator;
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
|
2014-01-08 23:46:46 +08:00
|
|
|
/* clear the port's relations to this
|
|
|
|
* aggregator
|
|
|
|
*/
|
2005-04-17 06:20:36 +08:00
|
|
|
port->aggregator = NULL;
|
2010-10-17 04:19:58 +08:00
|
|
|
port->next_port_in_aggregator = NULL;
|
|
|
|
port->actor_port_aggregator_identifier = 0;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2014-07-16 01:36:00 +08:00
|
|
|
netdev_dbg(bond->dev, "Port %d left LAG %d\n",
|
|
|
|
port->actor_port_number,
|
|
|
|
temp_aggregator->aggregator_identifier);
|
2014-01-08 23:46:46 +08:00
|
|
|
/* if the aggregator is empty, clear its
|
|
|
|
* parameters, and set it ready to be attached
|
|
|
|
*/
|
2010-10-17 04:19:59 +08:00
|
|
|
if (!temp_aggregator->lag_ports)
|
2005-04-17 06:20:36 +08:00
|
|
|
ad_clear_agg(temp_aggregator);
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
2014-01-08 23:46:46 +08:00
|
|
|
if (!curr_port) {
|
|
|
|
/* meaning: the port was related to an aggregator
|
|
|
|
* but was not on the aggregator port list
|
|
|
|
*/
|
2014-07-16 01:36:00 +08:00
|
|
|
net_warn_ratelimited("%s: Warning: Port %d (on %s) was related to aggregator %d but was not on its port list\n",
|
|
|
|
port->slave->bond->dev->name,
|
|
|
|
port->actor_port_number,
|
|
|
|
port->slave->dev->name,
|
|
|
|
port->aggregator->aggregator_identifier);
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
}
|
2014-01-08 23:46:46 +08:00
|
|
|
/* search on all aggregators for a suitable aggregator for this port */
|
2013-09-27 22:11:59 +08:00
|
|
|
bond_for_each_slave(bond, slave, iter) {
|
2014-05-12 15:08:43 +08:00
|
|
|
aggregator = &(SLAVE_AD_INFO(slave)->aggregator);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2014-01-08 23:46:46 +08:00
|
|
|
/* keep a free aggregator for later use(if needed) */
|
2005-04-17 06:20:36 +08:00
|
|
|
if (!aggregator->lag_ports) {
|
2010-10-17 04:19:59 +08:00
|
|
|
if (!free_aggregator)
|
2010-10-17 04:19:58 +08:00
|
|
|
free_aggregator = aggregator;
|
2005-04-17 06:20:36 +08:00
|
|
|
continue;
|
|
|
|
}
|
2014-01-02 09:12:54 +08:00
|
|
|
/* check if current aggregator suits us */
|
|
|
|
if (((aggregator->actor_oper_aggregator_key == port->actor_oper_port_key) && /* if all parameters match AND */
|
|
|
|
MAC_ADDRESS_EQUAL(&(aggregator->partner_system), &(port->partner_oper.system)) &&
|
2008-12-18 11:07:38 +08:00
|
|
|
(aggregator->partner_system_priority == port->partner_oper.system_priority) &&
|
|
|
|
(aggregator->partner_oper_aggregator_key == port->partner_oper.key)
|
2005-04-17 06:20:36 +08:00
|
|
|
) &&
|
2014-01-02 09:12:54 +08:00
|
|
|
((!MAC_ADDRESS_EQUAL(&(port->partner_oper.system), &(null_mac_addr)) && /* partner answers */
|
|
|
|
!aggregator->is_individual) /* but is not individual OR */
|
2005-04-17 06:20:36 +08:00
|
|
|
)
|
|
|
|
) {
|
2014-01-02 09:12:54 +08:00
|
|
|
/* attach to the founded aggregator */
|
2005-04-17 06:20:36 +08:00
|
|
|
port->aggregator = aggregator;
|
2010-10-17 04:19:58 +08:00
|
|
|
port->actor_port_aggregator_identifier =
|
|
|
|
port->aggregator->aggregator_identifier;
|
|
|
|
port->next_port_in_aggregator = aggregator->lag_ports;
|
2005-04-17 06:20:36 +08:00
|
|
|
port->aggregator->num_of_ports++;
|
2010-10-17 04:19:58 +08:00
|
|
|
aggregator->lag_ports = port;
|
2014-07-16 01:36:00 +08:00
|
|
|
netdev_dbg(bond->dev, "Port %d joined LAG %d(existing LAG)\n",
|
|
|
|
port->actor_port_number,
|
|
|
|
port->aggregator->aggregator_identifier);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2014-01-08 23:46:46 +08:00
|
|
|
/* mark this port as selected */
|
2005-04-17 06:20:36 +08:00
|
|
|
port->sm_vars |= AD_PORT_SELECTED;
|
|
|
|
found = 1;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2014-01-08 23:46:46 +08:00
|
|
|
/* the port couldn't find an aggregator - attach it to a new
|
|
|
|
* aggregator
|
|
|
|
*/
|
2005-04-17 06:20:36 +08:00
|
|
|
if (!found) {
|
|
|
|
if (free_aggregator) {
|
2014-01-08 23:46:46 +08:00
|
|
|
/* assign port a new aggregator */
|
2005-04-17 06:20:36 +08:00
|
|
|
port->aggregator = free_aggregator;
|
2010-10-17 04:19:58 +08:00
|
|
|
port->actor_port_aggregator_identifier =
|
|
|
|
port->aggregator->aggregator_identifier;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2014-01-08 23:46:46 +08:00
|
|
|
/* update the new aggregator's parameters
|
|
|
|
* if port was responsed from the end-user
|
|
|
|
*/
|
2014-11-19 16:48:58 +08:00
|
|
|
if (port->actor_oper_port_key & AD_DUPLEX_KEY_MASKS)
|
2010-10-17 04:19:59 +08:00
|
|
|
/* if port is full duplex */
|
2008-12-27 05:27:21 +08:00
|
|
|
port->aggregator->is_individual = false;
|
2010-10-17 04:19:59 +08:00
|
|
|
else
|
2008-12-27 05:27:21 +08:00
|
|
|
port->aggregator->is_individual = true;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2015-04-08 07:16:11 +08:00
|
|
|
port->aggregator->actor_admin_aggregator_key =
|
|
|
|
port->actor_admin_port_key;
|
|
|
|
port->aggregator->actor_oper_aggregator_key =
|
|
|
|
port->actor_oper_port_key;
|
2010-10-17 04:19:58 +08:00
|
|
|
port->aggregator->partner_system =
|
|
|
|
port->partner_oper.system;
|
|
|
|
port->aggregator->partner_system_priority =
|
|
|
|
port->partner_oper.system_priority;
|
2008-12-18 11:07:38 +08:00
|
|
|
port->aggregator->partner_oper_aggregator_key = port->partner_oper.key;
|
2005-04-17 06:20:36 +08:00
|
|
|
port->aggregator->receive_state = 1;
|
|
|
|
port->aggregator->transmit_state = 1;
|
|
|
|
port->aggregator->lag_ports = port;
|
|
|
|
port->aggregator->num_of_ports++;
|
|
|
|
|
2014-01-08 23:46:46 +08:00
|
|
|
/* mark this port as selected */
|
2005-04-17 06:20:36 +08:00
|
|
|
port->sm_vars |= AD_PORT_SELECTED;
|
|
|
|
|
2014-07-16 01:36:00 +08:00
|
|
|
netdev_dbg(bond->dev, "Port %d joined LAG %d(new LAG)\n",
|
|
|
|
port->actor_port_number,
|
|
|
|
port->aggregator->aggregator_identifier);
|
2005-04-17 06:20:36 +08:00
|
|
|
} else {
|
2014-07-16 01:36:00 +08:00
|
|
|
netdev_err(bond->dev, "Port %d (on %s) did not find a suitable aggregator\n",
|
2005-04-17 06:20:36 +08:00
|
|
|
port->actor_port_number, port->slave->dev->name);
|
|
|
|
}
|
|
|
|
}
|
2014-01-08 23:46:46 +08:00
|
|
|
/* if all aggregator's ports are READY_N == TRUE, set ready=TRUE
|
|
|
|
* in all aggregator's ports, else set ready=FALSE in all
|
|
|
|
* aggregator's ports
|
|
|
|
*/
|
|
|
|
__set_agg_ports_ready(port->aggregator,
|
|
|
|
__agg_ports_are_ready(port->aggregator));
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2008-11-05 09:51:16 +08:00
|
|
|
aggregator = __get_first_agg(port);
|
2014-10-05 08:45:01 +08:00
|
|
|
ad_agg_selection_logic(aggregator, update_slave_arr);
|
2015-01-26 14:16:59 +08:00
|
|
|
|
|
|
|
if (!port->aggregator->is_active)
|
|
|
|
port->actor_oper_port_state &= ~AD_STATE_SYNCHRONIZATION;
|
2008-11-05 09:51:16 +08:00
|
|
|
}
|
|
|
|
|
2014-01-08 23:46:46 +08:00
|
|
|
/* Decide if "agg" is a better choice for the new active aggregator that
|
2008-11-05 09:51:16 +08:00
|
|
|
* the current best, according to the ad_select policy.
|
|
|
|
*/
|
|
|
|
static struct aggregator *ad_agg_selection_test(struct aggregator *best,
|
|
|
|
struct aggregator *curr)
|
|
|
|
{
|
2014-01-08 23:46:46 +08:00
|
|
|
/* 0. If no best, select current.
|
2008-11-05 09:51:16 +08:00
|
|
|
*
|
|
|
|
* 1. If the current agg is not individual, and the best is
|
|
|
|
* individual, select current.
|
|
|
|
*
|
|
|
|
* 2. If current agg is individual and the best is not, keep best.
|
|
|
|
*
|
|
|
|
* 3. Therefore, current and best are both individual or both not
|
|
|
|
* individual, so:
|
|
|
|
*
|
|
|
|
* 3a. If current agg partner replied, and best agg partner did not,
|
|
|
|
* select current.
|
|
|
|
*
|
|
|
|
* 3b. If current agg partner did not reply and best agg partner
|
|
|
|
* did reply, keep best.
|
|
|
|
*
|
|
|
|
* 4. Therefore, current and best both have partner replies or
|
|
|
|
* both do not, so perform selection policy:
|
|
|
|
*
|
|
|
|
* BOND_AD_COUNT: Select by count of ports. If count is equal,
|
|
|
|
* select by bandwidth.
|
|
|
|
*
|
|
|
|
* BOND_AD_STABLE, BOND_AD_BANDWIDTH: Select by bandwidth.
|
|
|
|
*/
|
|
|
|
if (!best)
|
|
|
|
return curr;
|
|
|
|
|
|
|
|
if (!curr->is_individual && best->is_individual)
|
|
|
|
return curr;
|
|
|
|
|
|
|
|
if (curr->is_individual && !best->is_individual)
|
|
|
|
return best;
|
|
|
|
|
|
|
|
if (__agg_has_partner(curr) && !__agg_has_partner(best))
|
|
|
|
return curr;
|
|
|
|
|
|
|
|
if (!__agg_has_partner(curr) && __agg_has_partner(best))
|
|
|
|
return best;
|
|
|
|
|
|
|
|
switch (__get_agg_selection_mode(curr->lag_ports)) {
|
|
|
|
case BOND_AD_COUNT:
|
bonding: fix 802.3ad aggregator reselection
Since commit 7bb11dc9f59d ("bonding: unify all places where
actor-oper key needs to be updated."), the logic in bonding to handle
selection between multiple aggregators has not functioned.
This affects only configurations wherein the bonding slaves
connect to two discrete aggregators (e.g., two independent switches, each
with LACP enabled), thus creating two separate aggregation groups within a
single bond.
The cause is a change in 7bb11dc9f59d to no longer set
AD_PORT_BEGIN on a port after a link state change, which would cause the
port to be reselected for attachment to an aggregator as if were newly
added to the bond. We cannot restore the prior behavior, as it
contradicts IEEE 802.1AX 5.4.12, which requires ports that "become
inoperable" (lose carrier, setting port_enabled=false as per 802.1AX
5.4.7) to remain selected (i.e., assigned to the aggregator). As the port
now remains selected, the aggregator selection logic is not invoked.
A side effect of this change is that aggregators in bonding will
now contain ports that are link down. The aggregator selection logic
does not currently handle this situation correctly, causing incorrect
aggregator selection.
This patch makes two changes to repair the aggregator selection
logic in bonding to function as documented and within the confines of the
standard:
First, the aggregator selection and related logic now utilizes the
number of active ports per aggregator, not the number of selected ports
(as some selected ports may be down). The ad_select "bandwidth" and
"count" options only consider ports that are link up.
Second, on any carrier state change of any slave, the aggregator
selection logic is explicitly called to insure the correct aggregator is
active.
Reported-by: Veli-Matti Lintu <veli-matti.lintu@opinsys.fi>
Fixes: 7bb11dc9f59d ("bonding: unify all places where actor-oper key needs to be updated.")
Signed-off-by: Jay Vosburgh <jay.vosburgh@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2016-06-24 05:20:51 +08:00
|
|
|
if (__agg_active_ports(curr) > __agg_active_ports(best))
|
2008-11-05 09:51:16 +08:00
|
|
|
return curr;
|
|
|
|
|
bonding: fix 802.3ad aggregator reselection
Since commit 7bb11dc9f59d ("bonding: unify all places where
actor-oper key needs to be updated."), the logic in bonding to handle
selection between multiple aggregators has not functioned.
This affects only configurations wherein the bonding slaves
connect to two discrete aggregators (e.g., two independent switches, each
with LACP enabled), thus creating two separate aggregation groups within a
single bond.
The cause is a change in 7bb11dc9f59d to no longer set
AD_PORT_BEGIN on a port after a link state change, which would cause the
port to be reselected for attachment to an aggregator as if were newly
added to the bond. We cannot restore the prior behavior, as it
contradicts IEEE 802.1AX 5.4.12, which requires ports that "become
inoperable" (lose carrier, setting port_enabled=false as per 802.1AX
5.4.7) to remain selected (i.e., assigned to the aggregator). As the port
now remains selected, the aggregator selection logic is not invoked.
A side effect of this change is that aggregators in bonding will
now contain ports that are link down. The aggregator selection logic
does not currently handle this situation correctly, causing incorrect
aggregator selection.
This patch makes two changes to repair the aggregator selection
logic in bonding to function as documented and within the confines of the
standard:
First, the aggregator selection and related logic now utilizes the
number of active ports per aggregator, not the number of selected ports
(as some selected ports may be down). The ad_select "bandwidth" and
"count" options only consider ports that are link up.
Second, on any carrier state change of any slave, the aggregator
selection logic is explicitly called to insure the correct aggregator is
active.
Reported-by: Veli-Matti Lintu <veli-matti.lintu@opinsys.fi>
Fixes: 7bb11dc9f59d ("bonding: unify all places where actor-oper key needs to be updated.")
Signed-off-by: Jay Vosburgh <jay.vosburgh@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2016-06-24 05:20:51 +08:00
|
|
|
if (__agg_active_ports(curr) < __agg_active_ports(best))
|
2008-11-05 09:51:16 +08:00
|
|
|
return best;
|
|
|
|
|
|
|
|
/*FALLTHROUGH*/
|
|
|
|
case BOND_AD_STABLE:
|
|
|
|
case BOND_AD_BANDWIDTH:
|
|
|
|
if (__get_agg_bandwidth(curr) > __get_agg_bandwidth(best))
|
|
|
|
return curr;
|
|
|
|
|
|
|
|
break;
|
|
|
|
|
|
|
|
default:
|
2014-07-16 01:36:00 +08:00
|
|
|
net_warn_ratelimited("%s: Impossible agg select mode %d\n",
|
|
|
|
curr->slave->bond->dev->name,
|
|
|
|
__get_agg_selection_mode(curr->lag_ports));
|
2008-11-05 09:51:16 +08:00
|
|
|
break;
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
2008-11-05 09:51:16 +08:00
|
|
|
|
|
|
|
return best;
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
|
2009-05-15 16:44:32 +08:00
|
|
|
static int agg_device_up(const struct aggregator *agg)
|
|
|
|
{
|
2011-04-19 10:09:55 +08:00
|
|
|
struct port *port = agg->lag_ports;
|
2014-01-08 23:46:46 +08:00
|
|
|
|
2011-04-19 10:09:55 +08:00
|
|
|
if (!port)
|
|
|
|
return 0;
|
2014-01-08 23:46:46 +08:00
|
|
|
|
bonding: fix 802.3ad aggregator reselection
Since commit 7bb11dc9f59d ("bonding: unify all places where
actor-oper key needs to be updated."), the logic in bonding to handle
selection between multiple aggregators has not functioned.
This affects only configurations wherein the bonding slaves
connect to two discrete aggregators (e.g., two independent switches, each
with LACP enabled), thus creating two separate aggregation groups within a
single bond.
The cause is a change in 7bb11dc9f59d to no longer set
AD_PORT_BEGIN on a port after a link state change, which would cause the
port to be reselected for attachment to an aggregator as if were newly
added to the bond. We cannot restore the prior behavior, as it
contradicts IEEE 802.1AX 5.4.12, which requires ports that "become
inoperable" (lose carrier, setting port_enabled=false as per 802.1AX
5.4.7) to remain selected (i.e., assigned to the aggregator). As the port
now remains selected, the aggregator selection logic is not invoked.
A side effect of this change is that aggregators in bonding will
now contain ports that are link down. The aggregator selection logic
does not currently handle this situation correctly, causing incorrect
aggregator selection.
This patch makes two changes to repair the aggregator selection
logic in bonding to function as documented and within the confines of the
standard:
First, the aggregator selection and related logic now utilizes the
number of active ports per aggregator, not the number of selected ports
(as some selected ports may be down). The ad_select "bandwidth" and
"count" options only consider ports that are link up.
Second, on any carrier state change of any slave, the aggregator
selection logic is explicitly called to insure the correct aggregator is
active.
Reported-by: Veli-Matti Lintu <veli-matti.lintu@opinsys.fi>
Fixes: 7bb11dc9f59d ("bonding: unify all places where actor-oper key needs to be updated.")
Signed-off-by: Jay Vosburgh <jay.vosburgh@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2016-06-24 05:20:51 +08:00
|
|
|
for (port = agg->lag_ports; port;
|
|
|
|
port = port->next_port_in_aggregator) {
|
|
|
|
if (netif_running(port->slave->dev) &&
|
|
|
|
netif_carrier_ok(port->slave->dev))
|
|
|
|
return 1;
|
|
|
|
}
|
|
|
|
|
|
|
|
return 0;
|
2009-05-15 16:44:32 +08:00
|
|
|
}
|
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
/**
|
|
|
|
* ad_agg_selection_logic - select an aggregation group for a team
|
|
|
|
* @aggregator: the aggregator we're looking at
|
2014-10-05 08:45:01 +08:00
|
|
|
* @update_slave_arr: Does slave array need update?
|
2005-04-17 06:20:36 +08:00
|
|
|
*
|
|
|
|
* It is assumed that only one aggregator may be selected for a team.
|
2008-11-05 09:51:16 +08:00
|
|
|
*
|
|
|
|
* The logic of this function is to select the aggregator according to
|
|
|
|
* the ad_select policy:
|
|
|
|
*
|
|
|
|
* BOND_AD_STABLE: select the aggregator with the most ports attached to
|
|
|
|
* it, and to reselect the active aggregator only if the previous
|
|
|
|
* aggregator has no more ports related to it.
|
|
|
|
*
|
|
|
|
* BOND_AD_BANDWIDTH: select the aggregator with the highest total
|
|
|
|
* bandwidth, and reselect whenever a link state change takes place or the
|
|
|
|
* set of slaves in the bond changes.
|
|
|
|
*
|
|
|
|
* BOND_AD_COUNT: select the aggregator with largest number of ports
|
|
|
|
* (slaves), and reselect whenever a link state change takes place or the
|
|
|
|
* set of slaves in the bond changes.
|
2005-04-17 06:20:36 +08:00
|
|
|
*
|
|
|
|
* FIXME: this function MUST be called with the first agg in the bond, or
|
|
|
|
* __get_active_agg() won't work correctly. This function should be better
|
|
|
|
* called with the bond itself, and retrieve the first agg from it.
|
|
|
|
*/
|
2014-10-05 08:45:01 +08:00
|
|
|
static void ad_agg_selection_logic(struct aggregator *agg,
|
|
|
|
bool *update_slave_arr)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
2008-11-05 09:51:16 +08:00
|
|
|
struct aggregator *best, *active, *origin;
|
2013-09-27 22:12:01 +08:00
|
|
|
struct bonding *bond = agg->slave->bond;
|
|
|
|
struct list_head *iter;
|
|
|
|
struct slave *slave;
|
2005-04-17 06:20:36 +08:00
|
|
|
struct port *port;
|
|
|
|
|
2014-01-10 18:59:45 +08:00
|
|
|
rcu_read_lock();
|
2008-11-05 09:51:16 +08:00
|
|
|
origin = agg;
|
|
|
|
active = __get_active_agg(agg);
|
2009-05-15 16:44:32 +08:00
|
|
|
best = (active && agg_device_up(active)) ? active : NULL;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
bonding: add RCU for bond_3ad_state_machine_handler()
The bond_3ad_state_machine_handler() use the bond lock to protect
the bond slave list and slave port together, but it is not enough,
the bond slave list was link and unlink in RTNL, not bond lock,
so I add RCU to protect the slave list from leaving.
The bond lock is still used here, because when the slave has been
removed from the list by the time the state machine runs, it appears
to be possible for both function to manupulate the same aggregator->lag_ports
by finding the aggregator via two different ports that are both members of
that aggregator (i.e., port A of the agg is being unbound, and port B
of the agg is runing its state machine).
If I remove the bond lock, there are nothing to mutex changes
to aggregator->lag_ports between bond_3ad_state_machine_handler and
bond_3ad_unbind_slave, So the bond lock is the simplest way to protect
aggregator->lag_ports.
There was a lot of function need RCU protect, I have two choice
to make the function in RCU-safe, (1) create new similar functions
and make the bond slave list in RCU. (2) modify the existed functions
and make them in read-side critical section, because the RCU
read-side critical sections may be nested.
I choose (2) because it is no need to create more similar functions.
The nots in the function is still too old, clean up the nots.
Suggested-by: Nikolay Aleksandrov <nikolay@redhat.com>
Suggested-by: Jay Vosburgh <fubar@us.ibm.com>
Suggested-by: Veaceslav Falico <vfalico@redhat.com>
Signed-off-by: Ding Tianhong <dingtianhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-12-13 10:20:12 +08:00
|
|
|
bond_for_each_slave_rcu(bond, slave, iter) {
|
2014-05-12 15:08:43 +08:00
|
|
|
agg = &(SLAVE_AD_INFO(slave)->aggregator);
|
2013-09-27 22:12:01 +08:00
|
|
|
|
2008-11-05 09:51:16 +08:00
|
|
|
agg->is_active = 0;
|
|
|
|
|
bonding: fix 802.3ad aggregator reselection
Since commit 7bb11dc9f59d ("bonding: unify all places where
actor-oper key needs to be updated."), the logic in bonding to handle
selection between multiple aggregators has not functioned.
This affects only configurations wherein the bonding slaves
connect to two discrete aggregators (e.g., two independent switches, each
with LACP enabled), thus creating two separate aggregation groups within a
single bond.
The cause is a change in 7bb11dc9f59d to no longer set
AD_PORT_BEGIN on a port after a link state change, which would cause the
port to be reselected for attachment to an aggregator as if were newly
added to the bond. We cannot restore the prior behavior, as it
contradicts IEEE 802.1AX 5.4.12, which requires ports that "become
inoperable" (lose carrier, setting port_enabled=false as per 802.1AX
5.4.7) to remain selected (i.e., assigned to the aggregator). As the port
now remains selected, the aggregator selection logic is not invoked.
A side effect of this change is that aggregators in bonding will
now contain ports that are link down. The aggregator selection logic
does not currently handle this situation correctly, causing incorrect
aggregator selection.
This patch makes two changes to repair the aggregator selection
logic in bonding to function as documented and within the confines of the
standard:
First, the aggregator selection and related logic now utilizes the
number of active ports per aggregator, not the number of selected ports
(as some selected ports may be down). The ad_select "bandwidth" and
"count" options only consider ports that are link up.
Second, on any carrier state change of any slave, the aggregator
selection logic is explicitly called to insure the correct aggregator is
active.
Reported-by: Veli-Matti Lintu <veli-matti.lintu@opinsys.fi>
Fixes: 7bb11dc9f59d ("bonding: unify all places where actor-oper key needs to be updated.")
Signed-off-by: Jay Vosburgh <jay.vosburgh@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2016-06-24 05:20:51 +08:00
|
|
|
if (__agg_active_ports(agg) && agg_device_up(agg))
|
2008-11-05 09:51:16 +08:00
|
|
|
best = ad_agg_selection_test(best, agg);
|
2013-09-27 22:12:01 +08:00
|
|
|
}
|
2008-11-05 09:51:16 +08:00
|
|
|
|
|
|
|
if (best &&
|
|
|
|
__get_agg_selection_mode(best->lag_ports) == BOND_AD_STABLE) {
|
2014-01-08 23:46:46 +08:00
|
|
|
/* For the STABLE policy, don't replace the old active
|
2008-11-05 09:51:16 +08:00
|
|
|
* aggregator if it's still active (it has an answering
|
|
|
|
* partner) or if both the best and active don't have an
|
|
|
|
* answering partner.
|
|
|
|
*/
|
|
|
|
if (active && active->lag_ports &&
|
bonding: fix 802.3ad aggregator reselection
Since commit 7bb11dc9f59d ("bonding: unify all places where
actor-oper key needs to be updated."), the logic in bonding to handle
selection between multiple aggregators has not functioned.
This affects only configurations wherein the bonding slaves
connect to two discrete aggregators (e.g., two independent switches, each
with LACP enabled), thus creating two separate aggregation groups within a
single bond.
The cause is a change in 7bb11dc9f59d to no longer set
AD_PORT_BEGIN on a port after a link state change, which would cause the
port to be reselected for attachment to an aggregator as if were newly
added to the bond. We cannot restore the prior behavior, as it
contradicts IEEE 802.1AX 5.4.12, which requires ports that "become
inoperable" (lose carrier, setting port_enabled=false as per 802.1AX
5.4.7) to remain selected (i.e., assigned to the aggregator). As the port
now remains selected, the aggregator selection logic is not invoked.
A side effect of this change is that aggregators in bonding will
now contain ports that are link down. The aggregator selection logic
does not currently handle this situation correctly, causing incorrect
aggregator selection.
This patch makes two changes to repair the aggregator selection
logic in bonding to function as documented and within the confines of the
standard:
First, the aggregator selection and related logic now utilizes the
number of active ports per aggregator, not the number of selected ports
(as some selected ports may be down). The ad_select "bandwidth" and
"count" options only consider ports that are link up.
Second, on any carrier state change of any slave, the aggregator
selection logic is explicitly called to insure the correct aggregator is
active.
Reported-by: Veli-Matti Lintu <veli-matti.lintu@opinsys.fi>
Fixes: 7bb11dc9f59d ("bonding: unify all places where actor-oper key needs to be updated.")
Signed-off-by: Jay Vosburgh <jay.vosburgh@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2016-06-24 05:20:51 +08:00
|
|
|
__agg_active_ports(active) &&
|
2008-11-05 09:51:16 +08:00
|
|
|
(__agg_has_partner(active) ||
|
2014-01-08 23:46:46 +08:00
|
|
|
(!__agg_has_partner(active) &&
|
|
|
|
!__agg_has_partner(best)))) {
|
2008-11-05 09:51:16 +08:00
|
|
|
if (!(!active->actor_oper_aggregator_key &&
|
|
|
|
best->actor_oper_aggregator_key)) {
|
|
|
|
best = NULL;
|
|
|
|
active->is_active = 1;
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
}
|
2008-11-05 09:51:16 +08:00
|
|
|
}
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2008-11-05 09:51:16 +08:00
|
|
|
if (best && (best == active)) {
|
|
|
|
best = NULL;
|
|
|
|
active->is_active = 1;
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
|
bonding: add RCU for bond_3ad_state_machine_handler()
The bond_3ad_state_machine_handler() use the bond lock to protect
the bond slave list and slave port together, but it is not enough,
the bond slave list was link and unlink in RTNL, not bond lock,
so I add RCU to protect the slave list from leaving.
The bond lock is still used here, because when the slave has been
removed from the list by the time the state machine runs, it appears
to be possible for both function to manupulate the same aggregator->lag_ports
by finding the aggregator via two different ports that are both members of
that aggregator (i.e., port A of the agg is being unbound, and port B
of the agg is runing its state machine).
If I remove the bond lock, there are nothing to mutex changes
to aggregator->lag_ports between bond_3ad_state_machine_handler and
bond_3ad_unbind_slave, So the bond lock is the simplest way to protect
aggregator->lag_ports.
There was a lot of function need RCU protect, I have two choice
to make the function in RCU-safe, (1) create new similar functions
and make the bond slave list in RCU. (2) modify the existed functions
and make them in read-side critical section, because the RCU
read-side critical sections may be nested.
I choose (2) because it is no need to create more similar functions.
The nots in the function is still too old, clean up the nots.
Suggested-by: Nikolay Aleksandrov <nikolay@redhat.com>
Suggested-by: Jay Vosburgh <fubar@us.ibm.com>
Suggested-by: Veaceslav Falico <vfalico@redhat.com>
Signed-off-by: Ding Tianhong <dingtianhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-12-13 10:20:12 +08:00
|
|
|
/* if there is new best aggregator, activate it */
|
2008-11-05 09:51:16 +08:00
|
|
|
if (best) {
|
2014-07-16 01:36:00 +08:00
|
|
|
netdev_dbg(bond->dev, "best Agg=%d; P=%d; a k=%d; p k=%d; Ind=%d; Act=%d\n",
|
|
|
|
best->aggregator_identifier, best->num_of_ports,
|
|
|
|
best->actor_oper_aggregator_key,
|
|
|
|
best->partner_oper_aggregator_key,
|
|
|
|
best->is_individual, best->is_active);
|
|
|
|
netdev_dbg(bond->dev, "best ports %p slave %p %s\n",
|
|
|
|
best->lag_ports, best->slave,
|
|
|
|
best->slave ? best->slave->dev->name : "NULL");
|
2008-11-05 09:51:16 +08:00
|
|
|
|
bonding: add RCU for bond_3ad_state_machine_handler()
The bond_3ad_state_machine_handler() use the bond lock to protect
the bond slave list and slave port together, but it is not enough,
the bond slave list was link and unlink in RTNL, not bond lock,
so I add RCU to protect the slave list from leaving.
The bond lock is still used here, because when the slave has been
removed from the list by the time the state machine runs, it appears
to be possible for both function to manupulate the same aggregator->lag_ports
by finding the aggregator via two different ports that are both members of
that aggregator (i.e., port A of the agg is being unbound, and port B
of the agg is runing its state machine).
If I remove the bond lock, there are nothing to mutex changes
to aggregator->lag_ports between bond_3ad_state_machine_handler and
bond_3ad_unbind_slave, So the bond lock is the simplest way to protect
aggregator->lag_ports.
There was a lot of function need RCU protect, I have two choice
to make the function in RCU-safe, (1) create new similar functions
and make the bond slave list in RCU. (2) modify the existed functions
and make them in read-side critical section, because the RCU
read-side critical sections may be nested.
I choose (2) because it is no need to create more similar functions.
The nots in the function is still too old, clean up the nots.
Suggested-by: Nikolay Aleksandrov <nikolay@redhat.com>
Suggested-by: Jay Vosburgh <fubar@us.ibm.com>
Suggested-by: Veaceslav Falico <vfalico@redhat.com>
Signed-off-by: Ding Tianhong <dingtianhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-12-13 10:20:12 +08:00
|
|
|
bond_for_each_slave_rcu(bond, slave, iter) {
|
2014-05-12 15:08:43 +08:00
|
|
|
agg = &(SLAVE_AD_INFO(slave)->aggregator);
|
2008-11-05 09:51:16 +08:00
|
|
|
|
2014-07-16 01:36:00 +08:00
|
|
|
netdev_dbg(bond->dev, "Agg=%d; P=%d; a k=%d; p k=%d; Ind=%d; Act=%d\n",
|
|
|
|
agg->aggregator_identifier, agg->num_of_ports,
|
|
|
|
agg->actor_oper_aggregator_key,
|
|
|
|
agg->partner_oper_aggregator_key,
|
|
|
|
agg->is_individual, agg->is_active);
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
|
bonding: add RCU for bond_3ad_state_machine_handler()
The bond_3ad_state_machine_handler() use the bond lock to protect
the bond slave list and slave port together, but it is not enough,
the bond slave list was link and unlink in RTNL, not bond lock,
so I add RCU to protect the slave list from leaving.
The bond lock is still used here, because when the slave has been
removed from the list by the time the state machine runs, it appears
to be possible for both function to manupulate the same aggregator->lag_ports
by finding the aggregator via two different ports that are both members of
that aggregator (i.e., port A of the agg is being unbound, and port B
of the agg is runing its state machine).
If I remove the bond lock, there are nothing to mutex changes
to aggregator->lag_ports between bond_3ad_state_machine_handler and
bond_3ad_unbind_slave, So the bond lock is the simplest way to protect
aggregator->lag_ports.
There was a lot of function need RCU protect, I have two choice
to make the function in RCU-safe, (1) create new similar functions
and make the bond slave list in RCU. (2) modify the existed functions
and make them in read-side critical section, because the RCU
read-side critical sections may be nested.
I choose (2) because it is no need to create more similar functions.
The nots in the function is still too old, clean up the nots.
Suggested-by: Nikolay Aleksandrov <nikolay@redhat.com>
Suggested-by: Jay Vosburgh <fubar@us.ibm.com>
Suggested-by: Veaceslav Falico <vfalico@redhat.com>
Signed-off-by: Ding Tianhong <dingtianhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-12-13 10:20:12 +08:00
|
|
|
/* check if any partner replys */
|
2008-11-05 09:51:16 +08:00
|
|
|
if (best->is_individual) {
|
2014-07-16 01:36:00 +08:00
|
|
|
net_warn_ratelimited("%s: Warning: No 802.3ad response from the link partner for any adapters in the bond\n",
|
|
|
|
best->slave ?
|
|
|
|
best->slave->bond->dev->name : "NULL");
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
|
2008-11-05 09:51:16 +08:00
|
|
|
best->is_active = 1;
|
2014-07-16 01:36:00 +08:00
|
|
|
netdev_dbg(bond->dev, "LAG %d chosen as the active LAG\n",
|
|
|
|
best->aggregator_identifier);
|
|
|
|
netdev_dbg(bond->dev, "Agg=%d; P=%d; a k=%d; p k=%d; Ind=%d; Act=%d\n",
|
|
|
|
best->aggregator_identifier, best->num_of_ports,
|
|
|
|
best->actor_oper_aggregator_key,
|
|
|
|
best->partner_oper_aggregator_key,
|
|
|
|
best->is_individual, best->is_active);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2014-01-08 23:46:46 +08:00
|
|
|
/* disable the ports that were related to the former
|
|
|
|
* active_aggregator
|
|
|
|
*/
|
2008-11-05 09:51:16 +08:00
|
|
|
if (active) {
|
|
|
|
for (port = active->lag_ports; port;
|
|
|
|
port = port->next_port_in_aggregator) {
|
2005-04-17 06:20:36 +08:00
|
|
|
__disable_port(port);
|
|
|
|
}
|
|
|
|
}
|
2014-10-05 08:45:01 +08:00
|
|
|
/* Slave array needs update. */
|
|
|
|
*update_slave_arr = true;
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
|
2014-01-08 23:46:46 +08:00
|
|
|
/* if the selected aggregator is of join individuals
|
2008-11-05 09:51:16 +08:00
|
|
|
* (partner_system is NULL), enable their ports
|
|
|
|
*/
|
|
|
|
active = __get_active_agg(origin);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2008-11-05 09:51:16 +08:00
|
|
|
if (active) {
|
|
|
|
if (!__agg_has_partner(active)) {
|
|
|
|
for (port = active->lag_ports; port;
|
|
|
|
port = port->next_port_in_aggregator) {
|
2005-04-17 06:20:36 +08:00
|
|
|
__enable_port(port);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
2008-11-05 09:51:16 +08:00
|
|
|
|
bonding: add RCU for bond_3ad_state_machine_handler()
The bond_3ad_state_machine_handler() use the bond lock to protect
the bond slave list and slave port together, but it is not enough,
the bond slave list was link and unlink in RTNL, not bond lock,
so I add RCU to protect the slave list from leaving.
The bond lock is still used here, because when the slave has been
removed from the list by the time the state machine runs, it appears
to be possible for both function to manupulate the same aggregator->lag_ports
by finding the aggregator via two different ports that are both members of
that aggregator (i.e., port A of the agg is being unbound, and port B
of the agg is runing its state machine).
If I remove the bond lock, there are nothing to mutex changes
to aggregator->lag_ports between bond_3ad_state_machine_handler and
bond_3ad_unbind_slave, So the bond lock is the simplest way to protect
aggregator->lag_ports.
There was a lot of function need RCU protect, I have two choice
to make the function in RCU-safe, (1) create new similar functions
and make the bond slave list in RCU. (2) modify the existed functions
and make them in read-side critical section, because the RCU
read-side critical sections may be nested.
I choose (2) because it is no need to create more similar functions.
The nots in the function is still too old, clean up the nots.
Suggested-by: Nikolay Aleksandrov <nikolay@redhat.com>
Suggested-by: Jay Vosburgh <fubar@us.ibm.com>
Suggested-by: Veaceslav Falico <vfalico@redhat.com>
Signed-off-by: Ding Tianhong <dingtianhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-12-13 10:20:12 +08:00
|
|
|
rcu_read_unlock();
|
|
|
|
|
2013-09-27 22:12:01 +08:00
|
|
|
bond_3ad_set_carrier(bond);
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* ad_clear_agg - clear a given aggregator's parameters
|
|
|
|
* @aggregator: the aggregator we're looking at
|
|
|
|
*/
|
|
|
|
static void ad_clear_agg(struct aggregator *aggregator)
|
|
|
|
{
|
|
|
|
if (aggregator) {
|
2008-12-27 05:27:21 +08:00
|
|
|
aggregator->is_individual = false;
|
2005-04-17 06:20:36 +08:00
|
|
|
aggregator->actor_admin_aggregator_key = 0;
|
|
|
|
aggregator->actor_oper_aggregator_key = 0;
|
2016-06-30 22:13:41 +08:00
|
|
|
eth_zero_addr(aggregator->partner_system.mac_addr_value);
|
2005-04-17 06:20:36 +08:00
|
|
|
aggregator->partner_system_priority = 0;
|
|
|
|
aggregator->partner_oper_aggregator_key = 0;
|
|
|
|
aggregator->receive_state = 0;
|
|
|
|
aggregator->transmit_state = 0;
|
|
|
|
aggregator->lag_ports = NULL;
|
|
|
|
aggregator->is_active = 0;
|
|
|
|
aggregator->num_of_ports = 0;
|
2009-12-14 12:06:07 +08:00
|
|
|
pr_debug("LAG %d was cleared\n",
|
|
|
|
aggregator->aggregator_identifier);
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* ad_initialize_agg - initialize a given aggregator's parameters
|
|
|
|
* @aggregator: the aggregator we're looking at
|
|
|
|
*/
|
|
|
|
static void ad_initialize_agg(struct aggregator *aggregator)
|
|
|
|
{
|
|
|
|
if (aggregator) {
|
|
|
|
ad_clear_agg(aggregator);
|
|
|
|
|
2016-06-30 22:13:41 +08:00
|
|
|
eth_zero_addr(aggregator->aggregator_mac_address.mac_addr_value);
|
2005-04-17 06:20:36 +08:00
|
|
|
aggregator->aggregator_identifier = 0;
|
|
|
|
aggregator->slave = NULL;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* ad_initialize_port - initialize a given port's parameters
|
|
|
|
* @aggregator: the aggregator we're looking at
|
|
|
|
* @lacp_fast: boolean. whether fast periodic should be used
|
|
|
|
*/
|
|
|
|
static void ad_initialize_port(struct port *port, int lacp_fast)
|
|
|
|
{
|
2008-12-18 11:12:07 +08:00
|
|
|
static const struct port_params tmpl = {
|
|
|
|
.system_priority = 0xffff,
|
|
|
|
.key = 1,
|
|
|
|
.port_number = 1,
|
|
|
|
.port_priority = 0xff,
|
|
|
|
.port_state = 1,
|
|
|
|
};
|
2008-12-27 05:28:33 +08:00
|
|
|
static const struct lacpdu lacpdu = {
|
|
|
|
.subtype = 0x01,
|
|
|
|
.version_number = 0x01,
|
|
|
|
.tlv_type_actor_info = 0x01,
|
|
|
|
.actor_information_length = 0x14,
|
|
|
|
.tlv_type_partner_info = 0x02,
|
|
|
|
.partner_information_length = 0x14,
|
|
|
|
.tlv_type_collector_info = 0x03,
|
|
|
|
.collector_information_length = 0x10,
|
|
|
|
.collector_max_delay = htons(AD_COLLECTOR_MAX_DELAY),
|
|
|
|
};
|
2008-12-18 11:12:07 +08:00
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
if (port) {
|
|
|
|
port->actor_port_priority = 0xff;
|
|
|
|
port->actor_port_aggregator_identifier = 0;
|
2008-12-27 03:18:15 +08:00
|
|
|
port->ntt = false;
|
2014-01-08 23:46:46 +08:00
|
|
|
port->actor_admin_port_state = AD_STATE_AGGREGATION |
|
|
|
|
AD_STATE_LACP_ACTIVITY;
|
|
|
|
port->actor_oper_port_state = AD_STATE_AGGREGATION |
|
|
|
|
AD_STATE_LACP_ACTIVITY;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2010-10-17 04:19:59 +08:00
|
|
|
if (lacp_fast)
|
2005-04-17 06:20:36 +08:00
|
|
|
port->actor_oper_port_state |= AD_STATE_LACP_TIMEOUT;
|
|
|
|
|
2008-12-18 11:12:07 +08:00
|
|
|
memcpy(&port->partner_admin, &tmpl, sizeof(tmpl));
|
|
|
|
memcpy(&port->partner_oper, &tmpl, sizeof(tmpl));
|
|
|
|
|
2008-12-27 05:26:54 +08:00
|
|
|
port->is_enabled = true;
|
2014-01-08 23:46:46 +08:00
|
|
|
/* private parameters */
|
2015-03-28 13:34:31 +08:00
|
|
|
port->sm_vars = AD_PORT_BEGIN | AD_PORT_LACP_ENABLED;
|
2005-04-17 06:20:36 +08:00
|
|
|
port->sm_rx_state = 0;
|
|
|
|
port->sm_rx_timer_counter = 0;
|
|
|
|
port->sm_periodic_state = 0;
|
|
|
|
port->sm_periodic_timer_counter = 0;
|
|
|
|
port->sm_mux_state = 0;
|
|
|
|
port->sm_mux_timer_counter = 0;
|
|
|
|
port->sm_tx_state = 0;
|
|
|
|
port->aggregator = NULL;
|
|
|
|
port->next_port_in_aggregator = NULL;
|
|
|
|
port->transaction_id = 0;
|
|
|
|
|
2015-02-24 09:50:11 +08:00
|
|
|
port->sm_churn_actor_timer_counter = 0;
|
|
|
|
port->sm_churn_actor_state = 0;
|
|
|
|
port->churn_actor_count = 0;
|
|
|
|
port->sm_churn_partner_timer_counter = 0;
|
|
|
|
port->sm_churn_partner_state = 0;
|
|
|
|
port->churn_partner_count = 0;
|
|
|
|
|
2008-12-27 05:28:33 +08:00
|
|
|
memcpy(&port->lacpdu, &lacpdu, sizeof(lacpdu));
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* ad_enable_collecting_distributing - enable a port's transmit/receive
|
|
|
|
* @port: the port we're looking at
|
2014-10-05 08:45:01 +08:00
|
|
|
* @update_slave_arr: Does slave array need update?
|
2005-04-17 06:20:36 +08:00
|
|
|
*
|
|
|
|
* Enable @port if it's in an active aggregator
|
|
|
|
*/
|
2014-10-05 08:45:01 +08:00
|
|
|
static void ad_enable_collecting_distributing(struct port *port,
|
|
|
|
bool *update_slave_arr)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
|
|
|
if (port->aggregator->is_active) {
|
2009-12-14 12:06:07 +08:00
|
|
|
pr_debug("Enabling port %d(LAG %d)\n",
|
|
|
|
port->actor_port_number,
|
|
|
|
port->aggregator->aggregator_identifier);
|
2005-04-17 06:20:36 +08:00
|
|
|
__enable_port(port);
|
2014-10-05 08:45:01 +08:00
|
|
|
/* Slave array needs update */
|
|
|
|
*update_slave_arr = true;
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* ad_disable_collecting_distributing - disable a port's transmit/receive
|
|
|
|
* @port: the port we're looking at
|
2014-10-05 08:45:01 +08:00
|
|
|
* @update_slave_arr: Does slave array need update?
|
2005-04-17 06:20:36 +08:00
|
|
|
*/
|
2014-10-05 08:45:01 +08:00
|
|
|
static void ad_disable_collecting_distributing(struct port *port,
|
|
|
|
bool *update_slave_arr)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
2014-01-08 23:46:46 +08:00
|
|
|
if (port->aggregator &&
|
|
|
|
!MAC_ADDRESS_EQUAL(&(port->aggregator->partner_system),
|
|
|
|
&(null_mac_addr))) {
|
2009-12-14 12:06:07 +08:00
|
|
|
pr_debug("Disabling port %d(LAG %d)\n",
|
|
|
|
port->actor_port_number,
|
|
|
|
port->aggregator->aggregator_identifier);
|
2005-04-17 06:20:36 +08:00
|
|
|
__disable_port(port);
|
2014-10-05 08:45:01 +08:00
|
|
|
/* Slave array needs an update */
|
|
|
|
*update_slave_arr = true;
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* ad_marker_info_received - handle receive of a Marker information frame
|
|
|
|
* @marker_info: Marker info received
|
|
|
|
* @port: the port we're looking at
|
|
|
|
*/
|
2007-10-19 14:41:04 +08:00
|
|
|
static void ad_marker_info_received(struct bond_marker *marker_info,
|
|
|
|
struct port *port)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
2007-10-19 14:41:04 +08:00
|
|
|
struct bond_marker marker;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2014-01-08 23:46:46 +08:00
|
|
|
/* copy the received marker data to the response marker */
|
2007-10-19 14:41:04 +08:00
|
|
|
memcpy(&marker, marker_info, sizeof(struct bond_marker));
|
2014-01-08 23:46:46 +08:00
|
|
|
/* change the marker subtype to marker response */
|
2010-10-17 04:19:58 +08:00
|
|
|
marker.tlv_type = AD_MARKER_RESPONSE_SUBTYPE;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2014-01-08 23:46:46 +08:00
|
|
|
/* send the marker response */
|
2005-04-17 06:20:36 +08:00
|
|
|
if (ad_marker_send(port, &marker) >= 0) {
|
2009-12-14 12:06:07 +08:00
|
|
|
pr_debug("Sent Marker Response on port %d\n",
|
|
|
|
port->actor_port_number);
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* ad_marker_response_received - handle receive of a marker response frame
|
|
|
|
* @marker: marker PDU received
|
|
|
|
* @port: the port we're looking at
|
|
|
|
*
|
|
|
|
* This function does nothing since we decided not to implement send and handle
|
|
|
|
* response for marker PDU's, in this stage, but only to respond to marker
|
|
|
|
* information.
|
|
|
|
*/
|
2007-10-19 14:41:04 +08:00
|
|
|
static void ad_marker_response_received(struct bond_marker *marker,
|
2014-01-08 23:46:46 +08:00
|
|
|
struct port *port)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
2014-01-08 23:46:46 +08:00
|
|
|
/* DO NOTHING, SINCE WE DECIDED NOT TO IMPLEMENT THIS FEATURE FOR NOW */
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
|
2014-01-08 23:46:46 +08:00
|
|
|
/* ========= AD exported functions to the main bonding code ========= */
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2014-01-08 23:46:46 +08:00
|
|
|
/* Check aggregators status in team every T seconds */
|
2005-04-17 06:20:36 +08:00
|
|
|
#define AD_AGGREGATOR_SELECTION_TIMER 8
|
|
|
|
|
2014-01-08 23:46:46 +08:00
|
|
|
/**
|
|
|
|
* bond_3ad_initiate_agg_selection - initate aggregator selection
|
|
|
|
* @bond: bonding struct
|
2008-11-05 09:51:16 +08:00
|
|
|
*
|
|
|
|
* Set the aggregation selection timer, to initiate an agg selection in
|
|
|
|
* the very near future. Called during first initialization, and during
|
|
|
|
* any down to up transitions of the bond.
|
|
|
|
*/
|
|
|
|
void bond_3ad_initiate_agg_selection(struct bonding *bond, int timeout)
|
|
|
|
{
|
|
|
|
BOND_AD_INFO(bond).agg_select_timer = timeout;
|
|
|
|
}
|
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
/**
|
|
|
|
* bond_3ad_initialize - initialize a bond's 802.3ad parameters and structures
|
|
|
|
* @bond: bonding struct to work on
|
|
|
|
* @tick_resolution: tick duration (millisecond resolution)
|
|
|
|
*
|
|
|
|
* Can be called only after the mac address of the bond is set.
|
|
|
|
*/
|
2011-06-09 05:19:02 +08:00
|
|
|
void bond_3ad_initialize(struct bonding *bond, u16 tick_resolution)
|
2009-05-12 07:37:15 +08:00
|
|
|
{
|
2014-01-02 09:12:54 +08:00
|
|
|
/* check that the bond is not initialized yet */
|
|
|
|
if (!MAC_ADDRESS_EQUAL(&(BOND_AD_INFO(bond).system.sys_mac_addr),
|
2009-05-12 07:37:15 +08:00
|
|
|
bond->dev->dev_addr)) {
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2014-02-15 01:13:50 +08:00
|
|
|
BOND_AD_INFO(bond).aggregator_identifier = 0;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2015-05-09 15:01:55 +08:00
|
|
|
BOND_AD_INFO(bond).system.sys_priority =
|
|
|
|
bond->params.ad_actor_sys_prio;
|
2015-05-09 15:01:56 +08:00
|
|
|
if (is_zero_ether_addr(bond->params.ad_actor_system))
|
|
|
|
BOND_AD_INFO(bond).system.sys_mac_addr =
|
|
|
|
*((struct mac_addr *)bond->dev->dev_addr);
|
|
|
|
else
|
|
|
|
BOND_AD_INFO(bond).system.sys_mac_addr =
|
|
|
|
*((struct mac_addr *)bond->params.ad_actor_system);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2014-01-08 23:46:46 +08:00
|
|
|
/* initialize how many times this module is called in one
|
|
|
|
* second (should be about every 100ms)
|
|
|
|
*/
|
2005-04-17 06:20:36 +08:00
|
|
|
ad_ticks_per_sec = tick_resolution;
|
|
|
|
|
2008-11-05 09:51:16 +08:00
|
|
|
bond_3ad_initiate_agg_selection(bond,
|
|
|
|
AD_AGGREGATOR_SELECTION_TIMER *
|
|
|
|
ad_ticks_per_sec);
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* bond_3ad_bind_slave - initialize a slave's port
|
|
|
|
* @slave: slave struct to work on
|
|
|
|
*
|
|
|
|
* Returns: 0 on success
|
|
|
|
* < 0 on error
|
|
|
|
*/
|
2014-01-02 09:13:12 +08:00
|
|
|
void bond_3ad_bind_slave(struct slave *slave)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
|
|
|
struct bonding *bond = bond_get_bond_by_slave(slave);
|
|
|
|
struct port *port;
|
|
|
|
struct aggregator *aggregator;
|
|
|
|
|
2014-01-02 09:13:12 +08:00
|
|
|
/* check that the slave has not been initialized yet. */
|
2014-05-12 15:08:43 +08:00
|
|
|
if (SLAVE_AD_INFO(slave)->port.slave != slave) {
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2014-01-02 09:13:12 +08:00
|
|
|
/* port initialization */
|
2014-05-12 15:08:43 +08:00
|
|
|
port = &(SLAVE_AD_INFO(slave)->port);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2011-06-13 12:30:10 +08:00
|
|
|
ad_initialize_port(port, bond->params.lacp_fast);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
port->slave = slave;
|
2014-05-12 15:08:43 +08:00
|
|
|
port->actor_port_number = SLAVE_AD_INFO(slave)->id;
|
2015-05-09 15:01:57 +08:00
|
|
|
/* key is determined according to the link speed, duplex and
|
|
|
|
* user key
|
2014-01-02 09:13:12 +08:00
|
|
|
*/
|
2015-05-09 15:01:57 +08:00
|
|
|
port->actor_admin_port_key = bond->params.ad_user_port_key << 6;
|
2015-11-01 03:45:06 +08:00
|
|
|
ad_update_actor_keys(port, false);
|
2014-01-02 09:13:12 +08:00
|
|
|
/* actor system is the bond's system */
|
2016-02-03 20:17:01 +08:00
|
|
|
__ad_actor_update_port(port);
|
2014-01-08 23:46:46 +08:00
|
|
|
/* tx timer(to verify that no more than MAX_TX_IN_SECOND
|
|
|
|
* lacpdu's are sent in one second)
|
|
|
|
*/
|
2005-04-17 06:20:36 +08:00
|
|
|
port->sm_tx_timer_counter = ad_ticks_per_sec/AD_MAX_TX_IN_SECOND;
|
|
|
|
|
|
|
|
__disable_port(port);
|
|
|
|
|
2014-01-02 09:13:12 +08:00
|
|
|
/* aggregator initialization */
|
2014-05-12 15:08:43 +08:00
|
|
|
aggregator = &(SLAVE_AD_INFO(slave)->aggregator);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
ad_initialize_agg(aggregator);
|
|
|
|
|
|
|
|
aggregator->aggregator_mac_address = *((struct mac_addr *)bond->dev->dev_addr);
|
2014-02-15 01:13:50 +08:00
|
|
|
aggregator->aggregator_identifier = ++BOND_AD_INFO(bond).aggregator_identifier;
|
2005-04-17 06:20:36 +08:00
|
|
|
aggregator->slave = slave;
|
|
|
|
aggregator->is_active = 0;
|
|
|
|
aggregator->num_of_ports = 0;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* bond_3ad_unbind_slave - deinitialize a slave's port
|
|
|
|
* @slave: slave struct to work on
|
|
|
|
*
|
|
|
|
* Search for the aggregator that is related to this port, remove the
|
|
|
|
* aggregator and assign another aggregator for other port related to it
|
|
|
|
* (if any), and remove the port.
|
|
|
|
*/
|
|
|
|
void bond_3ad_unbind_slave(struct slave *slave)
|
|
|
|
{
|
|
|
|
struct port *port, *prev_port, *temp_port;
|
|
|
|
struct aggregator *aggregator, *new_aggregator, *temp_aggregator;
|
|
|
|
int select_new_active_agg = 0;
|
2013-09-27 22:12:02 +08:00
|
|
|
struct bonding *bond = slave->bond;
|
|
|
|
struct slave *slave_iter;
|
|
|
|
struct list_head *iter;
|
2014-10-05 08:45:01 +08:00
|
|
|
bool dummy_slave_update; /* Ignore this value as caller updates array */
|
2009-10-23 12:09:24 +08:00
|
|
|
|
2014-09-12 04:49:27 +08:00
|
|
|
/* Sync against bond_3ad_state_machine_handler() */
|
|
|
|
spin_lock_bh(&bond->mode_lock);
|
2014-05-12 15:08:43 +08:00
|
|
|
aggregator = &(SLAVE_AD_INFO(slave)->aggregator);
|
|
|
|
port = &(SLAVE_AD_INFO(slave)->port);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2014-01-08 23:46:46 +08:00
|
|
|
/* if slave is null, the whole port is not initialized */
|
2005-04-17 06:20:36 +08:00
|
|
|
if (!port->slave) {
|
2014-07-16 01:36:00 +08:00
|
|
|
netdev_warn(bond->dev, "Trying to unbind an uninitialized port on %s\n",
|
|
|
|
slave->dev->name);
|
2014-09-12 04:49:27 +08:00
|
|
|
goto out;
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
|
2014-07-16 01:36:00 +08:00
|
|
|
netdev_dbg(bond->dev, "Unbinding Link Aggregation Group %d\n",
|
|
|
|
aggregator->aggregator_identifier);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
/* Tell the partner that this port is not suitable for aggregation */
|
|
|
|
port->actor_oper_port_state &= ~AD_STATE_AGGREGATION;
|
|
|
|
__update_lacpdu_from_port(port);
|
|
|
|
ad_lacpdu_send(port);
|
|
|
|
|
2014-01-08 23:46:46 +08:00
|
|
|
/* check if this aggregator is occupied */
|
2005-04-17 06:20:36 +08:00
|
|
|
if (aggregator->lag_ports) {
|
2014-01-08 23:46:46 +08:00
|
|
|
/* check if there are other ports related to this aggregator
|
|
|
|
* except the port related to this slave(thats ensure us that
|
|
|
|
* there is a reason to search for new aggregator, and that we
|
|
|
|
* will find one
|
|
|
|
*/
|
|
|
|
if ((aggregator->lag_ports != port) ||
|
|
|
|
(aggregator->lag_ports->next_port_in_aggregator)) {
|
|
|
|
/* find new aggregator for the related port(s) */
|
2013-09-27 22:12:02 +08:00
|
|
|
bond_for_each_slave(bond, slave_iter, iter) {
|
2014-05-12 15:08:43 +08:00
|
|
|
new_aggregator = &(SLAVE_AD_INFO(slave_iter)->aggregator);
|
2014-01-08 23:46:46 +08:00
|
|
|
/* if the new aggregator is empty, or it is
|
|
|
|
* connected to our port only
|
|
|
|
*/
|
|
|
|
if (!new_aggregator->lag_ports ||
|
|
|
|
((new_aggregator->lag_ports == port) &&
|
|
|
|
!new_aggregator->lag_ports->next_port_in_aggregator))
|
2005-04-17 06:20:36 +08:00
|
|
|
break;
|
|
|
|
}
|
2013-09-27 22:12:02 +08:00
|
|
|
if (!slave_iter)
|
|
|
|
new_aggregator = NULL;
|
2014-01-08 23:46:46 +08:00
|
|
|
|
|
|
|
/* if new aggregator found, copy the aggregator's
|
|
|
|
* parameters and connect the related lag_ports to the
|
|
|
|
* new aggregator
|
|
|
|
*/
|
2005-04-17 06:20:36 +08:00
|
|
|
if ((new_aggregator) && ((!new_aggregator->lag_ports) || ((new_aggregator->lag_ports == port) && !new_aggregator->lag_ports->next_port_in_aggregator))) {
|
2014-07-16 01:36:00 +08:00
|
|
|
netdev_dbg(bond->dev, "Some port(s) related to LAG %d - replacing with LAG %d\n",
|
|
|
|
aggregator->aggregator_identifier,
|
|
|
|
new_aggregator->aggregator_identifier);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2014-01-08 23:46:46 +08:00
|
|
|
if ((new_aggregator->lag_ports == port) &&
|
|
|
|
new_aggregator->is_active) {
|
2014-07-16 01:36:00 +08:00
|
|
|
netdev_info(bond->dev, "Removing an active aggregator\n");
|
2005-04-17 06:20:36 +08:00
|
|
|
select_new_active_agg = 1;
|
|
|
|
}
|
|
|
|
|
|
|
|
new_aggregator->is_individual = aggregator->is_individual;
|
|
|
|
new_aggregator->actor_admin_aggregator_key = aggregator->actor_admin_aggregator_key;
|
|
|
|
new_aggregator->actor_oper_aggregator_key = aggregator->actor_oper_aggregator_key;
|
|
|
|
new_aggregator->partner_system = aggregator->partner_system;
|
|
|
|
new_aggregator->partner_system_priority = aggregator->partner_system_priority;
|
|
|
|
new_aggregator->partner_oper_aggregator_key = aggregator->partner_oper_aggregator_key;
|
|
|
|
new_aggregator->receive_state = aggregator->receive_state;
|
|
|
|
new_aggregator->transmit_state = aggregator->transmit_state;
|
|
|
|
new_aggregator->lag_ports = aggregator->lag_ports;
|
|
|
|
new_aggregator->is_active = aggregator->is_active;
|
|
|
|
new_aggregator->num_of_ports = aggregator->num_of_ports;
|
|
|
|
|
2014-01-08 23:46:46 +08:00
|
|
|
/* update the information that is written on
|
|
|
|
* the ports about the aggregator
|
|
|
|
*/
|
2010-10-17 04:19:58 +08:00
|
|
|
for (temp_port = aggregator->lag_ports; temp_port;
|
|
|
|
temp_port = temp_port->next_port_in_aggregator) {
|
|
|
|
temp_port->aggregator = new_aggregator;
|
2005-04-17 06:20:36 +08:00
|
|
|
temp_port->actor_port_aggregator_identifier = new_aggregator->aggregator_identifier;
|
|
|
|
}
|
|
|
|
|
|
|
|
ad_clear_agg(aggregator);
|
2009-10-23 12:09:24 +08:00
|
|
|
|
2010-10-17 04:19:59 +08:00
|
|
|
if (select_new_active_agg)
|
2014-10-05 08:45:01 +08:00
|
|
|
ad_agg_selection_logic(__get_first_agg(port),
|
|
|
|
&dummy_slave_update);
|
2005-04-17 06:20:36 +08:00
|
|
|
} else {
|
2014-07-16 01:36:00 +08:00
|
|
|
netdev_warn(bond->dev, "unbinding aggregator, and could not find a new aggregator for its ports\n");
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
2014-01-08 23:46:46 +08:00
|
|
|
} else {
|
|
|
|
/* in case that the only port related to this
|
|
|
|
* aggregator is the one we want to remove
|
|
|
|
*/
|
2005-04-17 06:20:36 +08:00
|
|
|
select_new_active_agg = aggregator->is_active;
|
|
|
|
ad_clear_agg(aggregator);
|
|
|
|
if (select_new_active_agg) {
|
2014-07-16 01:36:00 +08:00
|
|
|
netdev_info(bond->dev, "Removing an active aggregator\n");
|
2014-01-08 23:46:46 +08:00
|
|
|
/* select new active aggregator */
|
2013-09-27 21:10:58 +08:00
|
|
|
temp_aggregator = __get_first_agg(port);
|
|
|
|
if (temp_aggregator)
|
2014-10-05 08:45:01 +08:00
|
|
|
ad_agg_selection_logic(temp_aggregator,
|
|
|
|
&dummy_slave_update);
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2014-07-16 01:36:00 +08:00
|
|
|
netdev_dbg(bond->dev, "Unbinding port %d\n", port->actor_port_number);
|
2014-01-08 23:46:46 +08:00
|
|
|
|
|
|
|
/* find the aggregator that this port is connected to */
|
2013-09-27 22:12:02 +08:00
|
|
|
bond_for_each_slave(bond, slave_iter, iter) {
|
2014-05-12 15:08:43 +08:00
|
|
|
temp_aggregator = &(SLAVE_AD_INFO(slave_iter)->aggregator);
|
2005-04-17 06:20:36 +08:00
|
|
|
prev_port = NULL;
|
2014-01-08 23:46:46 +08:00
|
|
|
/* search the port in the aggregator's related ports */
|
2010-10-17 04:19:58 +08:00
|
|
|
for (temp_port = temp_aggregator->lag_ports; temp_port;
|
|
|
|
prev_port = temp_port,
|
2014-01-08 23:46:46 +08:00
|
|
|
temp_port = temp_port->next_port_in_aggregator) {
|
|
|
|
if (temp_port == port) {
|
|
|
|
/* the aggregator found - detach the port from
|
|
|
|
* this aggregator
|
|
|
|
*/
|
2010-10-17 04:19:59 +08:00
|
|
|
if (prev_port)
|
2005-04-17 06:20:36 +08:00
|
|
|
prev_port->next_port_in_aggregator = temp_port->next_port_in_aggregator;
|
2010-10-17 04:19:59 +08:00
|
|
|
else
|
2005-04-17 06:20:36 +08:00
|
|
|
temp_aggregator->lag_ports = temp_port->next_port_in_aggregator;
|
|
|
|
temp_aggregator->num_of_ports--;
|
bonding: fix 802.3ad aggregator reselection
Since commit 7bb11dc9f59d ("bonding: unify all places where
actor-oper key needs to be updated."), the logic in bonding to handle
selection between multiple aggregators has not functioned.
This affects only configurations wherein the bonding slaves
connect to two discrete aggregators (e.g., two independent switches, each
with LACP enabled), thus creating two separate aggregation groups within a
single bond.
The cause is a change in 7bb11dc9f59d to no longer set
AD_PORT_BEGIN on a port after a link state change, which would cause the
port to be reselected for attachment to an aggregator as if were newly
added to the bond. We cannot restore the prior behavior, as it
contradicts IEEE 802.1AX 5.4.12, which requires ports that "become
inoperable" (lose carrier, setting port_enabled=false as per 802.1AX
5.4.7) to remain selected (i.e., assigned to the aggregator). As the port
now remains selected, the aggregator selection logic is not invoked.
A side effect of this change is that aggregators in bonding will
now contain ports that are link down. The aggregator selection logic
does not currently handle this situation correctly, causing incorrect
aggregator selection.
This patch makes two changes to repair the aggregator selection
logic in bonding to function as documented and within the confines of the
standard:
First, the aggregator selection and related logic now utilizes the
number of active ports per aggregator, not the number of selected ports
(as some selected ports may be down). The ad_select "bandwidth" and
"count" options only consider ports that are link up.
Second, on any carrier state change of any slave, the aggregator
selection logic is explicitly called to insure the correct aggregator is
active.
Reported-by: Veli-Matti Lintu <veli-matti.lintu@opinsys.fi>
Fixes: 7bb11dc9f59d ("bonding: unify all places where actor-oper key needs to be updated.")
Signed-off-by: Jay Vosburgh <jay.vosburgh@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2016-06-24 05:20:51 +08:00
|
|
|
if (__agg_active_ports(temp_aggregator) == 0) {
|
2005-04-17 06:20:36 +08:00
|
|
|
select_new_active_agg = temp_aggregator->is_active;
|
|
|
|
ad_clear_agg(temp_aggregator);
|
|
|
|
if (select_new_active_agg) {
|
2014-07-16 01:36:00 +08:00
|
|
|
netdev_info(bond->dev, "Removing an active aggregator\n");
|
2014-01-08 23:46:46 +08:00
|
|
|
/* select new active aggregator */
|
2014-10-05 08:45:01 +08:00
|
|
|
ad_agg_selection_logic(__get_first_agg(port),
|
|
|
|
&dummy_slave_update);
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
2010-10-17 04:19:58 +08:00
|
|
|
port->slave = NULL;
|
2014-09-12 04:49:27 +08:00
|
|
|
|
|
|
|
out:
|
|
|
|
spin_unlock_bh(&bond->mode_lock);
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
|
2016-02-03 20:17:01 +08:00
|
|
|
/**
|
|
|
|
* bond_3ad_update_ad_actor_settings - reflect change of actor settings to ports
|
|
|
|
* @bond: bonding struct to work on
|
|
|
|
*
|
|
|
|
* If an ad_actor setting gets changed we need to update the individual port
|
|
|
|
* settings so the bond device will use the new values when it gets upped.
|
|
|
|
*/
|
|
|
|
void bond_3ad_update_ad_actor_settings(struct bonding *bond)
|
|
|
|
{
|
|
|
|
struct list_head *iter;
|
|
|
|
struct slave *slave;
|
|
|
|
|
|
|
|
ASSERT_RTNL();
|
|
|
|
|
|
|
|
BOND_AD_INFO(bond).system.sys_priority = bond->params.ad_actor_sys_prio;
|
|
|
|
if (is_zero_ether_addr(bond->params.ad_actor_system))
|
|
|
|
BOND_AD_INFO(bond).system.sys_mac_addr =
|
|
|
|
*((struct mac_addr *)bond->dev->dev_addr);
|
|
|
|
else
|
|
|
|
BOND_AD_INFO(bond).system.sys_mac_addr =
|
|
|
|
*((struct mac_addr *)bond->params.ad_actor_system);
|
|
|
|
|
|
|
|
spin_lock_bh(&bond->mode_lock);
|
2016-02-05 00:42:28 +08:00
|
|
|
bond_for_each_slave(bond, slave, iter) {
|
|
|
|
struct port *port = &(SLAVE_AD_INFO(slave))->port;
|
|
|
|
|
|
|
|
__ad_actor_update_port(port);
|
|
|
|
port->ntt = true;
|
|
|
|
}
|
2016-02-03 20:17:01 +08:00
|
|
|
spin_unlock_bh(&bond->mode_lock);
|
|
|
|
}
|
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
/**
|
|
|
|
* bond_3ad_state_machine_handler - handle state machines timeout
|
|
|
|
* @bond: bonding struct to work on
|
|
|
|
*
|
|
|
|
* The state machine handling concept in this module is to check every tick
|
|
|
|
* which state machine should operate any function. The execution order is
|
|
|
|
* round robin, so when we have an interaction between state machines, the
|
|
|
|
* reply of one to each other might be delayed until next tick.
|
|
|
|
*
|
|
|
|
* This function also complete the initialization when the agg_select_timer
|
|
|
|
* times out, and it selects an aggregator for the ports that are yet not
|
|
|
|
* related to any aggregator, and selects the active aggregator for a bond.
|
|
|
|
*/
|
2007-10-18 08:37:45 +08:00
|
|
|
void bond_3ad_state_machine_handler(struct work_struct *work)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
2007-10-18 08:37:45 +08:00
|
|
|
struct bonding *bond = container_of(work, struct bonding,
|
|
|
|
ad_work.work);
|
2005-04-17 06:20:36 +08:00
|
|
|
struct aggregator *aggregator;
|
2013-09-27 22:11:57 +08:00
|
|
|
struct list_head *iter;
|
|
|
|
struct slave *slave;
|
|
|
|
struct port *port;
|
bonding: Fix RTNL: assertion failed at net/core/rtnetlink.c for 802.3ad mode
The problem was introduced by the commit 1d3ee88ae0d
(bonding: add netlink attributes to slave link dev).
The bond_set_active_slave() and bond_set_backup_slave()
will use rtmsg_ifinfo to send slave's states, so these
two functions should be called in RTNL.
In 802.3ad mode, acquiring RTNL for the __enable_port and
__disable_port cases is difficult, as those calls generally
already hold the state machine lock, and cannot unconditionally
call rtnl_lock because either they already hold RTNL (for calls
via bond_3ad_unbind_slave) or due to the potential for deadlock
with bond_3ad_adapter_speed_changed, bond_3ad_adapter_duplex_changed,
bond_3ad_link_change, or bond_3ad_update_lacp_rate. All four of
those are called with RTNL held, and acquire the state machine lock
second. The calling contexts for __enable_port and __disable_port
already hold the state machine lock, and may or may not need RTNL.
According to the Jay's opinion, I don't think it is a problem that
the slave don't send notify message synchronously when the status
changed, normally the state machine is running every 100 ms, send
the notify message at the end of the state machine if the slave's
state changed should be better.
I fix the problem through these steps:
1). add a new function bond_set_slave_state() which could change
the slave's state and call rtmsg_ifinfo() according to the input
parameters called notify.
2). Add a new slave parameter which called should_notify, if the slave's state
changed and don't notify yet, the parameter will be set to 1, and then if
the slave's state changed again, the param will be set to 0, it indicate that
the slave's state has been restored, no need to notify any one.
3). the __enable_port and __disable_port should not call rtmsg_ifinfo
in the state machine lock, any change in the state of slave could
set a flag in the slave, it will indicated that an rtmsg_ifinfo
should be called at the end of the state machine.
Cc: Jay Vosburgh <fubar@us.ibm.com>
Cc: Veaceslav Falico <vfalico@redhat.com>
Cc: Andy Gospodarek <andy@greyhouse.net>
Signed-off-by: Ding Tianhong <dingtianhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2014-02-26 11:05:22 +08:00
|
|
|
bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER;
|
2014-10-05 08:45:01 +08:00
|
|
|
bool update_slave_arr = false;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2014-09-12 04:49:27 +08:00
|
|
|
/* Lock to protect data accessed by all (e.g., port->sm_vars) and
|
|
|
|
* against running with bond_3ad_unbind_slave. ad_rx_machine may run
|
|
|
|
* concurrently due to incoming LACPDU as well.
|
|
|
|
*/
|
2014-09-12 04:49:25 +08:00
|
|
|
spin_lock_bh(&bond->mode_lock);
|
bonding: add RCU for bond_3ad_state_machine_handler()
The bond_3ad_state_machine_handler() use the bond lock to protect
the bond slave list and slave port together, but it is not enough,
the bond slave list was link and unlink in RTNL, not bond lock,
so I add RCU to protect the slave list from leaving.
The bond lock is still used here, because when the slave has been
removed from the list by the time the state machine runs, it appears
to be possible for both function to manupulate the same aggregator->lag_ports
by finding the aggregator via two different ports that are both members of
that aggregator (i.e., port A of the agg is being unbound, and port B
of the agg is runing its state machine).
If I remove the bond lock, there are nothing to mutex changes
to aggregator->lag_ports between bond_3ad_state_machine_handler and
bond_3ad_unbind_slave, So the bond lock is the simplest way to protect
aggregator->lag_ports.
There was a lot of function need RCU protect, I have two choice
to make the function in RCU-safe, (1) create new similar functions
and make the bond slave list in RCU. (2) modify the existed functions
and make them in read-side critical section, because the RCU
read-side critical sections may be nested.
I choose (2) because it is no need to create more similar functions.
The nots in the function is still too old, clean up the nots.
Suggested-by: Nikolay Aleksandrov <nikolay@redhat.com>
Suggested-by: Jay Vosburgh <fubar@us.ibm.com>
Suggested-by: Veaceslav Falico <vfalico@redhat.com>
Signed-off-by: Ding Tianhong <dingtianhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-12-13 10:20:12 +08:00
|
|
|
rcu_read_lock();
|
2013-10-28 12:11:22 +08:00
|
|
|
|
bonding: add RCU for bond_3ad_state_machine_handler()
The bond_3ad_state_machine_handler() use the bond lock to protect
the bond slave list and slave port together, but it is not enough,
the bond slave list was link and unlink in RTNL, not bond lock,
so I add RCU to protect the slave list from leaving.
The bond lock is still used here, because when the slave has been
removed from the list by the time the state machine runs, it appears
to be possible for both function to manupulate the same aggregator->lag_ports
by finding the aggregator via two different ports that are both members of
that aggregator (i.e., port A of the agg is being unbound, and port B
of the agg is runing its state machine).
If I remove the bond lock, there are nothing to mutex changes
to aggregator->lag_ports between bond_3ad_state_machine_handler and
bond_3ad_unbind_slave, So the bond lock is the simplest way to protect
aggregator->lag_ports.
There was a lot of function need RCU protect, I have two choice
to make the function in RCU-safe, (1) create new similar functions
and make the bond slave list in RCU. (2) modify the existed functions
and make them in read-side critical section, because the RCU
read-side critical sections may be nested.
I choose (2) because it is no need to create more similar functions.
The nots in the function is still too old, clean up the nots.
Suggested-by: Nikolay Aleksandrov <nikolay@redhat.com>
Suggested-by: Jay Vosburgh <fubar@us.ibm.com>
Suggested-by: Veaceslav Falico <vfalico@redhat.com>
Signed-off-by: Ding Tianhong <dingtianhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-12-13 10:20:12 +08:00
|
|
|
/* check if there are any slaves */
|
2013-09-25 15:20:21 +08:00
|
|
|
if (!bond_has_slaves(bond))
|
2005-04-17 06:20:36 +08:00
|
|
|
goto re_arm;
|
|
|
|
|
bonding: add RCU for bond_3ad_state_machine_handler()
The bond_3ad_state_machine_handler() use the bond lock to protect
the bond slave list and slave port together, but it is not enough,
the bond slave list was link and unlink in RTNL, not bond lock,
so I add RCU to protect the slave list from leaving.
The bond lock is still used here, because when the slave has been
removed from the list by the time the state machine runs, it appears
to be possible for both function to manupulate the same aggregator->lag_ports
by finding the aggregator via two different ports that are both members of
that aggregator (i.e., port A of the agg is being unbound, and port B
of the agg is runing its state machine).
If I remove the bond lock, there are nothing to mutex changes
to aggregator->lag_ports between bond_3ad_state_machine_handler and
bond_3ad_unbind_slave, So the bond lock is the simplest way to protect
aggregator->lag_ports.
There was a lot of function need RCU protect, I have two choice
to make the function in RCU-safe, (1) create new similar functions
and make the bond slave list in RCU. (2) modify the existed functions
and make them in read-side critical section, because the RCU
read-side critical sections may be nested.
I choose (2) because it is no need to create more similar functions.
The nots in the function is still too old, clean up the nots.
Suggested-by: Nikolay Aleksandrov <nikolay@redhat.com>
Suggested-by: Jay Vosburgh <fubar@us.ibm.com>
Suggested-by: Veaceslav Falico <vfalico@redhat.com>
Signed-off-by: Ding Tianhong <dingtianhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-12-13 10:20:12 +08:00
|
|
|
/* check if agg_select_timer timer after initialize is timed out */
|
2014-01-08 23:46:46 +08:00
|
|
|
if (BOND_AD_INFO(bond).agg_select_timer &&
|
|
|
|
!(--BOND_AD_INFO(bond).agg_select_timer)) {
|
bonding: add RCU for bond_3ad_state_machine_handler()
The bond_3ad_state_machine_handler() use the bond lock to protect
the bond slave list and slave port together, but it is not enough,
the bond slave list was link and unlink in RTNL, not bond lock,
so I add RCU to protect the slave list from leaving.
The bond lock is still used here, because when the slave has been
removed from the list by the time the state machine runs, it appears
to be possible for both function to manupulate the same aggregator->lag_ports
by finding the aggregator via two different ports that are both members of
that aggregator (i.e., port A of the agg is being unbound, and port B
of the agg is runing its state machine).
If I remove the bond lock, there are nothing to mutex changes
to aggregator->lag_ports between bond_3ad_state_machine_handler and
bond_3ad_unbind_slave, So the bond lock is the simplest way to protect
aggregator->lag_ports.
There was a lot of function need RCU protect, I have two choice
to make the function in RCU-safe, (1) create new similar functions
and make the bond slave list in RCU. (2) modify the existed functions
and make them in read-side critical section, because the RCU
read-side critical sections may be nested.
I choose (2) because it is no need to create more similar functions.
The nots in the function is still too old, clean up the nots.
Suggested-by: Nikolay Aleksandrov <nikolay@redhat.com>
Suggested-by: Jay Vosburgh <fubar@us.ibm.com>
Suggested-by: Veaceslav Falico <vfalico@redhat.com>
Signed-off-by: Ding Tianhong <dingtianhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-12-13 10:20:12 +08:00
|
|
|
slave = bond_first_slave_rcu(bond);
|
2014-05-12 15:08:43 +08:00
|
|
|
port = slave ? &(SLAVE_AD_INFO(slave)->port) : NULL;
|
2013-09-27 22:11:58 +08:00
|
|
|
|
bonding: add RCU for bond_3ad_state_machine_handler()
The bond_3ad_state_machine_handler() use the bond lock to protect
the bond slave list and slave port together, but it is not enough,
the bond slave list was link and unlink in RTNL, not bond lock,
so I add RCU to protect the slave list from leaving.
The bond lock is still used here, because when the slave has been
removed from the list by the time the state machine runs, it appears
to be possible for both function to manupulate the same aggregator->lag_ports
by finding the aggregator via two different ports that are both members of
that aggregator (i.e., port A of the agg is being unbound, and port B
of the agg is runing its state machine).
If I remove the bond lock, there are nothing to mutex changes
to aggregator->lag_ports between bond_3ad_state_machine_handler and
bond_3ad_unbind_slave, So the bond lock is the simplest way to protect
aggregator->lag_ports.
There was a lot of function need RCU protect, I have two choice
to make the function in RCU-safe, (1) create new similar functions
and make the bond slave list in RCU. (2) modify the existed functions
and make them in read-side critical section, because the RCU
read-side critical sections may be nested.
I choose (2) because it is no need to create more similar functions.
The nots in the function is still too old, clean up the nots.
Suggested-by: Nikolay Aleksandrov <nikolay@redhat.com>
Suggested-by: Jay Vosburgh <fubar@us.ibm.com>
Suggested-by: Veaceslav Falico <vfalico@redhat.com>
Signed-off-by: Ding Tianhong <dingtianhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-12-13 10:20:12 +08:00
|
|
|
/* select the active aggregator for the bond */
|
2013-09-27 22:11:58 +08:00
|
|
|
if (port) {
|
2005-04-17 06:20:36 +08:00
|
|
|
if (!port->slave) {
|
2014-07-16 01:36:00 +08:00
|
|
|
net_warn_ratelimited("%s: Warning: bond's first port is uninitialized\n",
|
|
|
|
bond->dev->name);
|
2005-04-17 06:20:36 +08:00
|
|
|
goto re_arm;
|
|
|
|
}
|
|
|
|
|
|
|
|
aggregator = __get_first_agg(port);
|
2014-10-05 08:45:01 +08:00
|
|
|
ad_agg_selection_logic(aggregator, &update_slave_arr);
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
2008-07-03 09:21:58 +08:00
|
|
|
bond_3ad_set_carrier(bond);
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
|
bonding: add RCU for bond_3ad_state_machine_handler()
The bond_3ad_state_machine_handler() use the bond lock to protect
the bond slave list and slave port together, but it is not enough,
the bond slave list was link and unlink in RTNL, not bond lock,
so I add RCU to protect the slave list from leaving.
The bond lock is still used here, because when the slave has been
removed from the list by the time the state machine runs, it appears
to be possible for both function to manupulate the same aggregator->lag_ports
by finding the aggregator via two different ports that are both members of
that aggregator (i.e., port A of the agg is being unbound, and port B
of the agg is runing its state machine).
If I remove the bond lock, there are nothing to mutex changes
to aggregator->lag_ports between bond_3ad_state_machine_handler and
bond_3ad_unbind_slave, So the bond lock is the simplest way to protect
aggregator->lag_ports.
There was a lot of function need RCU protect, I have two choice
to make the function in RCU-safe, (1) create new similar functions
and make the bond slave list in RCU. (2) modify the existed functions
and make them in read-side critical section, because the RCU
read-side critical sections may be nested.
I choose (2) because it is no need to create more similar functions.
The nots in the function is still too old, clean up the nots.
Suggested-by: Nikolay Aleksandrov <nikolay@redhat.com>
Suggested-by: Jay Vosburgh <fubar@us.ibm.com>
Suggested-by: Veaceslav Falico <vfalico@redhat.com>
Signed-off-by: Ding Tianhong <dingtianhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-12-13 10:20:12 +08:00
|
|
|
/* for each port run the state machines */
|
|
|
|
bond_for_each_slave_rcu(bond, slave, iter) {
|
2014-05-12 15:08:43 +08:00
|
|
|
port = &(SLAVE_AD_INFO(slave)->port);
|
2005-04-17 06:20:36 +08:00
|
|
|
if (!port->slave) {
|
2014-07-16 01:36:00 +08:00
|
|
|
net_warn_ratelimited("%s: Warning: Found an uninitialized port\n",
|
2014-03-17 00:55:03 +08:00
|
|
|
bond->dev->name);
|
2005-04-17 06:20:36 +08:00
|
|
|
goto re_arm;
|
|
|
|
}
|
|
|
|
|
|
|
|
ad_rx_machine(NULL, port);
|
|
|
|
ad_periodic_machine(port);
|
2014-10-05 08:45:01 +08:00
|
|
|
ad_port_selection_logic(port, &update_slave_arr);
|
|
|
|
ad_mux_machine(port, &update_slave_arr);
|
2005-04-17 06:20:36 +08:00
|
|
|
ad_tx_machine(port);
|
2015-02-24 09:50:11 +08:00
|
|
|
ad_churn_machine(port);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
bonding: add RCU for bond_3ad_state_machine_handler()
The bond_3ad_state_machine_handler() use the bond lock to protect
the bond slave list and slave port together, but it is not enough,
the bond slave list was link and unlink in RTNL, not bond lock,
so I add RCU to protect the slave list from leaving.
The bond lock is still used here, because when the slave has been
removed from the list by the time the state machine runs, it appears
to be possible for both function to manupulate the same aggregator->lag_ports
by finding the aggregator via two different ports that are both members of
that aggregator (i.e., port A of the agg is being unbound, and port B
of the agg is runing its state machine).
If I remove the bond lock, there are nothing to mutex changes
to aggregator->lag_ports between bond_3ad_state_machine_handler and
bond_3ad_unbind_slave, So the bond lock is the simplest way to protect
aggregator->lag_ports.
There was a lot of function need RCU protect, I have two choice
to make the function in RCU-safe, (1) create new similar functions
and make the bond slave list in RCU. (2) modify the existed functions
and make them in read-side critical section, because the RCU
read-side critical sections may be nested.
I choose (2) because it is no need to create more similar functions.
The nots in the function is still too old, clean up the nots.
Suggested-by: Nikolay Aleksandrov <nikolay@redhat.com>
Suggested-by: Jay Vosburgh <fubar@us.ibm.com>
Suggested-by: Veaceslav Falico <vfalico@redhat.com>
Signed-off-by: Ding Tianhong <dingtianhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-12-13 10:20:12 +08:00
|
|
|
/* turn off the BEGIN bit, since we already handled it */
|
2010-10-17 04:19:59 +08:00
|
|
|
if (port->sm_vars & AD_PORT_BEGIN)
|
2005-04-17 06:20:36 +08:00
|
|
|
port->sm_vars &= ~AD_PORT_BEGIN;
|
|
|
|
}
|
|
|
|
|
|
|
|
re_arm:
|
bonding: Fix RTNL: assertion failed at net/core/rtnetlink.c for 802.3ad mode
The problem was introduced by the commit 1d3ee88ae0d
(bonding: add netlink attributes to slave link dev).
The bond_set_active_slave() and bond_set_backup_slave()
will use rtmsg_ifinfo to send slave's states, so these
two functions should be called in RTNL.
In 802.3ad mode, acquiring RTNL for the __enable_port and
__disable_port cases is difficult, as those calls generally
already hold the state machine lock, and cannot unconditionally
call rtnl_lock because either they already hold RTNL (for calls
via bond_3ad_unbind_slave) or due to the potential for deadlock
with bond_3ad_adapter_speed_changed, bond_3ad_adapter_duplex_changed,
bond_3ad_link_change, or bond_3ad_update_lacp_rate. All four of
those are called with RTNL held, and acquire the state machine lock
second. The calling contexts for __enable_port and __disable_port
already hold the state machine lock, and may or may not need RTNL.
According to the Jay's opinion, I don't think it is a problem that
the slave don't send notify message synchronously when the status
changed, normally the state machine is running every 100 ms, send
the notify message at the end of the state machine if the slave's
state changed should be better.
I fix the problem through these steps:
1). add a new function bond_set_slave_state() which could change
the slave's state and call rtmsg_ifinfo() according to the input
parameters called notify.
2). Add a new slave parameter which called should_notify, if the slave's state
changed and don't notify yet, the parameter will be set to 1, and then if
the slave's state changed again, the param will be set to 0, it indicate that
the slave's state has been restored, no need to notify any one.
3). the __enable_port and __disable_port should not call rtmsg_ifinfo
in the state machine lock, any change in the state of slave could
set a flag in the slave, it will indicated that an rtmsg_ifinfo
should be called at the end of the state machine.
Cc: Jay Vosburgh <fubar@us.ibm.com>
Cc: Veaceslav Falico <vfalico@redhat.com>
Cc: Andy Gospodarek <andy@greyhouse.net>
Signed-off-by: Ding Tianhong <dingtianhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2014-02-26 11:05:22 +08:00
|
|
|
bond_for_each_slave_rcu(bond, slave, iter) {
|
|
|
|
if (slave->should_notify) {
|
|
|
|
should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
bonding: add RCU for bond_3ad_state_machine_handler()
The bond_3ad_state_machine_handler() use the bond lock to protect
the bond slave list and slave port together, but it is not enough,
the bond slave list was link and unlink in RTNL, not bond lock,
so I add RCU to protect the slave list from leaving.
The bond lock is still used here, because when the slave has been
removed from the list by the time the state machine runs, it appears
to be possible for both function to manupulate the same aggregator->lag_ports
by finding the aggregator via two different ports that are both members of
that aggregator (i.e., port A of the agg is being unbound, and port B
of the agg is runing its state machine).
If I remove the bond lock, there are nothing to mutex changes
to aggregator->lag_ports between bond_3ad_state_machine_handler and
bond_3ad_unbind_slave, So the bond lock is the simplest way to protect
aggregator->lag_ports.
There was a lot of function need RCU protect, I have two choice
to make the function in RCU-safe, (1) create new similar functions
and make the bond slave list in RCU. (2) modify the existed functions
and make them in read-side critical section, because the RCU
read-side critical sections may be nested.
I choose (2) because it is no need to create more similar functions.
The nots in the function is still too old, clean up the nots.
Suggested-by: Nikolay Aleksandrov <nikolay@redhat.com>
Suggested-by: Jay Vosburgh <fubar@us.ibm.com>
Suggested-by: Veaceslav Falico <vfalico@redhat.com>
Signed-off-by: Ding Tianhong <dingtianhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-12-13 10:20:12 +08:00
|
|
|
rcu_read_unlock();
|
2014-09-12 04:49:25 +08:00
|
|
|
spin_unlock_bh(&bond->mode_lock);
|
bonding: Fix RTNL: assertion failed at net/core/rtnetlink.c for 802.3ad mode
The problem was introduced by the commit 1d3ee88ae0d
(bonding: add netlink attributes to slave link dev).
The bond_set_active_slave() and bond_set_backup_slave()
will use rtmsg_ifinfo to send slave's states, so these
two functions should be called in RTNL.
In 802.3ad mode, acquiring RTNL for the __enable_port and
__disable_port cases is difficult, as those calls generally
already hold the state machine lock, and cannot unconditionally
call rtnl_lock because either they already hold RTNL (for calls
via bond_3ad_unbind_slave) or due to the potential for deadlock
with bond_3ad_adapter_speed_changed, bond_3ad_adapter_duplex_changed,
bond_3ad_link_change, or bond_3ad_update_lacp_rate. All four of
those are called with RTNL held, and acquire the state machine lock
second. The calling contexts for __enable_port and __disable_port
already hold the state machine lock, and may or may not need RTNL.
According to the Jay's opinion, I don't think it is a problem that
the slave don't send notify message synchronously when the status
changed, normally the state machine is running every 100 ms, send
the notify message at the end of the state machine if the slave's
state changed should be better.
I fix the problem through these steps:
1). add a new function bond_set_slave_state() which could change
the slave's state and call rtmsg_ifinfo() according to the input
parameters called notify.
2). Add a new slave parameter which called should_notify, if the slave's state
changed and don't notify yet, the parameter will be set to 1, and then if
the slave's state changed again, the param will be set to 0, it indicate that
the slave's state has been restored, no need to notify any one.
3). the __enable_port and __disable_port should not call rtmsg_ifinfo
in the state machine lock, any change in the state of slave could
set a flag in the slave, it will indicated that an rtmsg_ifinfo
should be called at the end of the state machine.
Cc: Jay Vosburgh <fubar@us.ibm.com>
Cc: Veaceslav Falico <vfalico@redhat.com>
Cc: Andy Gospodarek <andy@greyhouse.net>
Signed-off-by: Ding Tianhong <dingtianhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2014-02-26 11:05:22 +08:00
|
|
|
|
2014-10-05 08:45:01 +08:00
|
|
|
if (update_slave_arr)
|
|
|
|
bond_slave_arr_work_rearm(bond, 0);
|
|
|
|
|
bonding: Fix RTNL: assertion failed at net/core/rtnetlink.c for 802.3ad mode
The problem was introduced by the commit 1d3ee88ae0d
(bonding: add netlink attributes to slave link dev).
The bond_set_active_slave() and bond_set_backup_slave()
will use rtmsg_ifinfo to send slave's states, so these
two functions should be called in RTNL.
In 802.3ad mode, acquiring RTNL for the __enable_port and
__disable_port cases is difficult, as those calls generally
already hold the state machine lock, and cannot unconditionally
call rtnl_lock because either they already hold RTNL (for calls
via bond_3ad_unbind_slave) or due to the potential for deadlock
with bond_3ad_adapter_speed_changed, bond_3ad_adapter_duplex_changed,
bond_3ad_link_change, or bond_3ad_update_lacp_rate. All four of
those are called with RTNL held, and acquire the state machine lock
second. The calling contexts for __enable_port and __disable_port
already hold the state machine lock, and may or may not need RTNL.
According to the Jay's opinion, I don't think it is a problem that
the slave don't send notify message synchronously when the status
changed, normally the state machine is running every 100 ms, send
the notify message at the end of the state machine if the slave's
state changed should be better.
I fix the problem through these steps:
1). add a new function bond_set_slave_state() which could change
the slave's state and call rtmsg_ifinfo() according to the input
parameters called notify.
2). Add a new slave parameter which called should_notify, if the slave's state
changed and don't notify yet, the parameter will be set to 1, and then if
the slave's state changed again, the param will be set to 0, it indicate that
the slave's state has been restored, no need to notify any one.
3). the __enable_port and __disable_port should not call rtmsg_ifinfo
in the state machine lock, any change in the state of slave could
set a flag in the slave, it will indicated that an rtmsg_ifinfo
should be called at the end of the state machine.
Cc: Jay Vosburgh <fubar@us.ibm.com>
Cc: Veaceslav Falico <vfalico@redhat.com>
Cc: Andy Gospodarek <andy@greyhouse.net>
Signed-off-by: Ding Tianhong <dingtianhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2014-02-26 11:05:22 +08:00
|
|
|
if (should_notify_rtnl && rtnl_trylock()) {
|
2014-02-26 11:05:23 +08:00
|
|
|
bond_slave_state_notify(bond);
|
bonding: Fix RTNL: assertion failed at net/core/rtnetlink.c for 802.3ad mode
The problem was introduced by the commit 1d3ee88ae0d
(bonding: add netlink attributes to slave link dev).
The bond_set_active_slave() and bond_set_backup_slave()
will use rtmsg_ifinfo to send slave's states, so these
two functions should be called in RTNL.
In 802.3ad mode, acquiring RTNL for the __enable_port and
__disable_port cases is difficult, as those calls generally
already hold the state machine lock, and cannot unconditionally
call rtnl_lock because either they already hold RTNL (for calls
via bond_3ad_unbind_slave) or due to the potential for deadlock
with bond_3ad_adapter_speed_changed, bond_3ad_adapter_duplex_changed,
bond_3ad_link_change, or bond_3ad_update_lacp_rate. All four of
those are called with RTNL held, and acquire the state machine lock
second. The calling contexts for __enable_port and __disable_port
already hold the state machine lock, and may or may not need RTNL.
According to the Jay's opinion, I don't think it is a problem that
the slave don't send notify message synchronously when the status
changed, normally the state machine is running every 100 ms, send
the notify message at the end of the state machine if the slave's
state changed should be better.
I fix the problem through these steps:
1). add a new function bond_set_slave_state() which could change
the slave's state and call rtmsg_ifinfo() according to the input
parameters called notify.
2). Add a new slave parameter which called should_notify, if the slave's state
changed and don't notify yet, the parameter will be set to 1, and then if
the slave's state changed again, the param will be set to 0, it indicate that
the slave's state has been restored, no need to notify any one.
3). the __enable_port and __disable_port should not call rtmsg_ifinfo
in the state machine lock, any change in the state of slave could
set a flag in the slave, it will indicated that an rtmsg_ifinfo
should be called at the end of the state machine.
Cc: Jay Vosburgh <fubar@us.ibm.com>
Cc: Veaceslav Falico <vfalico@redhat.com>
Cc: Andy Gospodarek <andy@greyhouse.net>
Signed-off-by: Ding Tianhong <dingtianhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2014-02-26 11:05:22 +08:00
|
|
|
rtnl_unlock();
|
|
|
|
}
|
bonding: add RCU for bond_3ad_state_machine_handler()
The bond_3ad_state_machine_handler() use the bond lock to protect
the bond slave list and slave port together, but it is not enough,
the bond slave list was link and unlink in RTNL, not bond lock,
so I add RCU to protect the slave list from leaving.
The bond lock is still used here, because when the slave has been
removed from the list by the time the state machine runs, it appears
to be possible for both function to manupulate the same aggregator->lag_ports
by finding the aggregator via two different ports that are both members of
that aggregator (i.e., port A of the agg is being unbound, and port B
of the agg is runing its state machine).
If I remove the bond lock, there are nothing to mutex changes
to aggregator->lag_ports between bond_3ad_state_machine_handler and
bond_3ad_unbind_slave, So the bond lock is the simplest way to protect
aggregator->lag_ports.
There was a lot of function need RCU protect, I have two choice
to make the function in RCU-safe, (1) create new similar functions
and make the bond slave list in RCU. (2) modify the existed functions
and make them in read-side critical section, because the RCU
read-side critical sections may be nested.
I choose (2) because it is no need to create more similar functions.
The nots in the function is still too old, clean up the nots.
Suggested-by: Nikolay Aleksandrov <nikolay@redhat.com>
Suggested-by: Jay Vosburgh <fubar@us.ibm.com>
Suggested-by: Veaceslav Falico <vfalico@redhat.com>
Signed-off-by: Ding Tianhong <dingtianhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-12-13 10:20:12 +08:00
|
|
|
queue_delayed_work(bond->wq, &bond->ad_work, ad_delta_in_ticks);
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* bond_3ad_rx_indication - handle a received frame
|
|
|
|
* @lacpdu: received lacpdu
|
|
|
|
* @slave: slave struct to work on
|
|
|
|
* @length: length of the data received
|
|
|
|
*
|
|
|
|
* It is assumed that frames that were sent on this NIC don't returned as new
|
|
|
|
* received frames (loopback). Since only the payload is given to this
|
|
|
|
* function, it check for loopback.
|
|
|
|
*/
|
2014-01-08 23:46:46 +08:00
|
|
|
static int bond_3ad_rx_indication(struct lacpdu *lacpdu, struct slave *slave,
|
|
|
|
u16 length)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
|
|
|
struct port *port;
|
2012-05-09 09:01:40 +08:00
|
|
|
int ret = RX_HANDLER_ANOTHER;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
if (length >= sizeof(struct lacpdu)) {
|
|
|
|
|
2014-05-12 15:08:43 +08:00
|
|
|
port = &(SLAVE_AD_INFO(slave)->port);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
if (!port->slave) {
|
2014-07-16 01:36:00 +08:00
|
|
|
net_warn_ratelimited("%s: Warning: port of slave %s is uninitialized\n",
|
|
|
|
slave->dev->name, slave->bond->dev->name);
|
2012-05-09 09:01:40 +08:00
|
|
|
return ret;
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
switch (lacpdu->subtype) {
|
|
|
|
case AD_TYPE_LACPDU:
|
2012-05-09 09:01:40 +08:00
|
|
|
ret = RX_HANDLER_CONSUMED;
|
2015-01-26 14:17:00 +08:00
|
|
|
netdev_dbg(slave->bond->dev,
|
|
|
|
"Received LACPDU on port %d slave %s\n",
|
|
|
|
port->actor_port_number,
|
|
|
|
slave->dev->name);
|
2011-03-04 06:09:11 +08:00
|
|
|
/* Protect against concurrent state machines */
|
2014-09-12 04:49:27 +08:00
|
|
|
spin_lock(&slave->bond->mode_lock);
|
2005-04-17 06:20:36 +08:00
|
|
|
ad_rx_machine(lacpdu, port);
|
2014-09-12 04:49:27 +08:00
|
|
|
spin_unlock(&slave->bond->mode_lock);
|
2005-04-17 06:20:36 +08:00
|
|
|
break;
|
|
|
|
|
|
|
|
case AD_TYPE_MARKER:
|
2012-05-09 09:01:40 +08:00
|
|
|
ret = RX_HANDLER_CONSUMED;
|
2014-01-08 23:46:46 +08:00
|
|
|
/* No need to convert fields to Little Endian since we
|
|
|
|
* don't use the marker's fields.
|
|
|
|
*/
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2007-10-19 14:41:04 +08:00
|
|
|
switch (((struct bond_marker *)lacpdu)->tlv_type) {
|
2005-04-17 06:20:36 +08:00
|
|
|
case AD_MARKER_INFORMATION_SUBTYPE:
|
2014-07-16 01:36:00 +08:00
|
|
|
netdev_dbg(slave->bond->dev, "Received Marker Information on port %d\n",
|
|
|
|
port->actor_port_number);
|
2007-10-19 14:41:04 +08:00
|
|
|
ad_marker_info_received((struct bond_marker *)lacpdu, port);
|
2005-04-17 06:20:36 +08:00
|
|
|
break;
|
|
|
|
|
|
|
|
case AD_MARKER_RESPONSE_SUBTYPE:
|
2014-07-16 01:36:00 +08:00
|
|
|
netdev_dbg(slave->bond->dev, "Received Marker Response on port %d\n",
|
|
|
|
port->actor_port_number);
|
2007-10-19 14:41:04 +08:00
|
|
|
ad_marker_response_received((struct bond_marker *)lacpdu, port);
|
2005-04-17 06:20:36 +08:00
|
|
|
break;
|
|
|
|
|
|
|
|
default:
|
2014-07-16 01:36:00 +08:00
|
|
|
netdev_dbg(slave->bond->dev, "Received an unknown Marker subtype on slot %d\n",
|
|
|
|
port->actor_port_number);
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
2012-05-09 09:01:40 +08:00
|
|
|
return ret;
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
|
2015-11-01 03:45:06 +08:00
|
|
|
/**
|
|
|
|
* ad_update_actor_keys - Update the oper / admin keys for a port based on
|
|
|
|
* its current speed and duplex settings.
|
|
|
|
*
|
|
|
|
* @port: the port we'are looking at
|
|
|
|
* @reset: Boolean to just reset the speed and the duplex part of the key
|
|
|
|
*
|
|
|
|
* The logic to change the oper / admin keys is:
|
|
|
|
* (a) A full duplex port can participate in LACP with partner.
|
|
|
|
* (b) When the speed is changed, LACP need to be reinitiated.
|
|
|
|
*/
|
|
|
|
static void ad_update_actor_keys(struct port *port, bool reset)
|
|
|
|
{
|
|
|
|
u8 duplex = 0;
|
|
|
|
u16 ospeed = 0, speed = 0;
|
|
|
|
u16 old_oper_key = port->actor_oper_port_key;
|
|
|
|
|
|
|
|
port->actor_admin_port_key &= ~(AD_SPEED_KEY_MASKS|AD_DUPLEX_KEY_MASKS);
|
|
|
|
if (!reset) {
|
|
|
|
speed = __get_link_speed(port);
|
|
|
|
ospeed = (old_oper_key & AD_SPEED_KEY_MASKS) >> 1;
|
|
|
|
duplex = __get_duplex(port);
|
|
|
|
port->actor_admin_port_key |= (speed << 1) | duplex;
|
|
|
|
}
|
|
|
|
port->actor_oper_port_key = port->actor_admin_port_key;
|
|
|
|
|
|
|
|
if (old_oper_key != port->actor_oper_port_key) {
|
|
|
|
/* Only 'duplex' port participates in LACP */
|
|
|
|
if (duplex)
|
|
|
|
port->sm_vars |= AD_PORT_LACP_ENABLED;
|
|
|
|
else
|
|
|
|
port->sm_vars &= ~AD_PORT_LACP_ENABLED;
|
|
|
|
|
|
|
|
if (!reset) {
|
|
|
|
if (!speed) {
|
|
|
|
netdev_err(port->slave->dev,
|
|
|
|
"speed changed to 0 for port %s",
|
|
|
|
port->slave->dev->name);
|
|
|
|
} else if (duplex && ospeed != speed) {
|
|
|
|
/* Speed change restarts LACP state-machine */
|
|
|
|
port->sm_vars |= AD_PORT_BEGIN;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
/**
|
2015-11-01 03:45:11 +08:00
|
|
|
* bond_3ad_adapter_speed_duplex_changed - handle a slave's speed / duplex
|
|
|
|
* change indication
|
2005-04-17 06:20:36 +08:00
|
|
|
*
|
|
|
|
* @slave: slave struct to work on
|
|
|
|
*
|
|
|
|
* Handle reselection of aggregator (if needed) for this port.
|
|
|
|
*/
|
2015-11-01 03:45:11 +08:00
|
|
|
void bond_3ad_adapter_speed_duplex_changed(struct slave *slave)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
|
|
|
struct port *port;
|
|
|
|
|
2014-05-12 15:08:43 +08:00
|
|
|
port = &(SLAVE_AD_INFO(slave)->port);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2013-12-13 17:29:24 +08:00
|
|
|
/* if slave is null, the whole port is not initialized */
|
2005-04-17 06:20:36 +08:00
|
|
|
if (!port->slave) {
|
2015-11-01 03:45:11 +08:00
|
|
|
netdev_warn(slave->bond->dev,
|
|
|
|
"speed/duplex changed for uninitialized port %s\n",
|
2014-07-16 01:36:00 +08:00
|
|
|
slave->dev->name);
|
2005-04-17 06:20:36 +08:00
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
2014-09-12 04:49:27 +08:00
|
|
|
spin_lock_bh(&slave->bond->mode_lock);
|
2015-11-01 03:45:06 +08:00
|
|
|
ad_update_actor_keys(port, false);
|
2015-11-01 03:45:11 +08:00
|
|
|
netdev_dbg(slave->bond->dev, "Port %d slave %s changed speed/duplex\n",
|
2015-01-26 14:17:00 +08:00
|
|
|
port->actor_port_number, slave->dev->name);
|
2014-09-12 04:49:27 +08:00
|
|
|
spin_unlock_bh(&slave->bond->mode_lock);
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* bond_3ad_handle_link_change - handle a slave's link status change indication
|
|
|
|
* @slave: slave struct to work on
|
|
|
|
* @status: whether the link is now up or down
|
|
|
|
*
|
|
|
|
* Handle reselection of aggregator (if needed) for this port.
|
|
|
|
*/
|
|
|
|
void bond_3ad_handle_link_change(struct slave *slave, char link)
|
|
|
|
{
|
bonding: fix 802.3ad aggregator reselection
Since commit 7bb11dc9f59d ("bonding: unify all places where
actor-oper key needs to be updated."), the logic in bonding to handle
selection between multiple aggregators has not functioned.
This affects only configurations wherein the bonding slaves
connect to two discrete aggregators (e.g., two independent switches, each
with LACP enabled), thus creating two separate aggregation groups within a
single bond.
The cause is a change in 7bb11dc9f59d to no longer set
AD_PORT_BEGIN on a port after a link state change, which would cause the
port to be reselected for attachment to an aggregator as if were newly
added to the bond. We cannot restore the prior behavior, as it
contradicts IEEE 802.1AX 5.4.12, which requires ports that "become
inoperable" (lose carrier, setting port_enabled=false as per 802.1AX
5.4.7) to remain selected (i.e., assigned to the aggregator). As the port
now remains selected, the aggregator selection logic is not invoked.
A side effect of this change is that aggregators in bonding will
now contain ports that are link down. The aggregator selection logic
does not currently handle this situation correctly, causing incorrect
aggregator selection.
This patch makes two changes to repair the aggregator selection
logic in bonding to function as documented and within the confines of the
standard:
First, the aggregator selection and related logic now utilizes the
number of active ports per aggregator, not the number of selected ports
(as some selected ports may be down). The ad_select "bandwidth" and
"count" options only consider ports that are link up.
Second, on any carrier state change of any slave, the aggregator
selection logic is explicitly called to insure the correct aggregator is
active.
Reported-by: Veli-Matti Lintu <veli-matti.lintu@opinsys.fi>
Fixes: 7bb11dc9f59d ("bonding: unify all places where actor-oper key needs to be updated.")
Signed-off-by: Jay Vosburgh <jay.vosburgh@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2016-06-24 05:20:51 +08:00
|
|
|
struct aggregator *agg;
|
2005-04-17 06:20:36 +08:00
|
|
|
struct port *port;
|
bonding: fix 802.3ad aggregator reselection
Since commit 7bb11dc9f59d ("bonding: unify all places where
actor-oper key needs to be updated."), the logic in bonding to handle
selection between multiple aggregators has not functioned.
This affects only configurations wherein the bonding slaves
connect to two discrete aggregators (e.g., two independent switches, each
with LACP enabled), thus creating two separate aggregation groups within a
single bond.
The cause is a change in 7bb11dc9f59d to no longer set
AD_PORT_BEGIN on a port after a link state change, which would cause the
port to be reselected for attachment to an aggregator as if were newly
added to the bond. We cannot restore the prior behavior, as it
contradicts IEEE 802.1AX 5.4.12, which requires ports that "become
inoperable" (lose carrier, setting port_enabled=false as per 802.1AX
5.4.7) to remain selected (i.e., assigned to the aggregator). As the port
now remains selected, the aggregator selection logic is not invoked.
A side effect of this change is that aggregators in bonding will
now contain ports that are link down. The aggregator selection logic
does not currently handle this situation correctly, causing incorrect
aggregator selection.
This patch makes two changes to repair the aggregator selection
logic in bonding to function as documented and within the confines of the
standard:
First, the aggregator selection and related logic now utilizes the
number of active ports per aggregator, not the number of selected ports
(as some selected ports may be down). The ad_select "bandwidth" and
"count" options only consider ports that are link up.
Second, on any carrier state change of any slave, the aggregator
selection logic is explicitly called to insure the correct aggregator is
active.
Reported-by: Veli-Matti Lintu <veli-matti.lintu@opinsys.fi>
Fixes: 7bb11dc9f59d ("bonding: unify all places where actor-oper key needs to be updated.")
Signed-off-by: Jay Vosburgh <jay.vosburgh@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2016-06-24 05:20:51 +08:00
|
|
|
bool dummy;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2014-05-12 15:08:43 +08:00
|
|
|
port = &(SLAVE_AD_INFO(slave)->port);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2013-12-13 17:29:29 +08:00
|
|
|
/* if slave is null, the whole port is not initialized */
|
2005-04-17 06:20:36 +08:00
|
|
|
if (!port->slave) {
|
2014-07-16 01:36:00 +08:00
|
|
|
netdev_warn(slave->bond->dev, "link status changed for uninitialized port on %s\n",
|
|
|
|
slave->dev->name);
|
2005-04-17 06:20:36 +08:00
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
2014-09-12 04:49:27 +08:00
|
|
|
spin_lock_bh(&slave->bond->mode_lock);
|
2013-12-13 17:29:29 +08:00
|
|
|
/* on link down we are zeroing duplex and speed since
|
|
|
|
* some of the adaptors(ce1000.lan) report full duplex/speed
|
|
|
|
* instead of N/A(duplex) / 0(speed).
|
|
|
|
*
|
|
|
|
* on link up we are forcing recheck on the duplex and speed since
|
|
|
|
* some of he adaptors(ce1000.lan) report.
|
|
|
|
*/
|
2005-04-17 06:20:36 +08:00
|
|
|
if (link == BOND_LINK_UP) {
|
2008-12-27 05:26:54 +08:00
|
|
|
port->is_enabled = true;
|
2015-11-01 03:45:06 +08:00
|
|
|
ad_update_actor_keys(port, false);
|
2005-04-17 06:20:36 +08:00
|
|
|
} else {
|
|
|
|
/* link has failed */
|
2008-12-27 05:26:54 +08:00
|
|
|
port->is_enabled = false;
|
2015-11-01 03:45:06 +08:00
|
|
|
ad_update_actor_keys(port, true);
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
bonding: fix 802.3ad aggregator reselection
Since commit 7bb11dc9f59d ("bonding: unify all places where
actor-oper key needs to be updated."), the logic in bonding to handle
selection between multiple aggregators has not functioned.
This affects only configurations wherein the bonding slaves
connect to two discrete aggregators (e.g., two independent switches, each
with LACP enabled), thus creating two separate aggregation groups within a
single bond.
The cause is a change in 7bb11dc9f59d to no longer set
AD_PORT_BEGIN on a port after a link state change, which would cause the
port to be reselected for attachment to an aggregator as if were newly
added to the bond. We cannot restore the prior behavior, as it
contradicts IEEE 802.1AX 5.4.12, which requires ports that "become
inoperable" (lose carrier, setting port_enabled=false as per 802.1AX
5.4.7) to remain selected (i.e., assigned to the aggregator). As the port
now remains selected, the aggregator selection logic is not invoked.
A side effect of this change is that aggregators in bonding will
now contain ports that are link down. The aggregator selection logic
does not currently handle this situation correctly, causing incorrect
aggregator selection.
This patch makes two changes to repair the aggregator selection
logic in bonding to function as documented and within the confines of the
standard:
First, the aggregator selection and related logic now utilizes the
number of active ports per aggregator, not the number of selected ports
(as some selected ports may be down). The ad_select "bandwidth" and
"count" options only consider ports that are link up.
Second, on any carrier state change of any slave, the aggregator
selection logic is explicitly called to insure the correct aggregator is
active.
Reported-by: Veli-Matti Lintu <veli-matti.lintu@opinsys.fi>
Fixes: 7bb11dc9f59d ("bonding: unify all places where actor-oper key needs to be updated.")
Signed-off-by: Jay Vosburgh <jay.vosburgh@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2016-06-24 05:20:51 +08:00
|
|
|
agg = __get_first_agg(port);
|
|
|
|
ad_agg_selection_logic(agg, &dummy);
|
|
|
|
|
2014-07-16 01:36:00 +08:00
|
|
|
netdev_dbg(slave->bond->dev, "Port %d changed link status to %s\n",
|
|
|
|
port->actor_port_number,
|
|
|
|
link == BOND_LINK_UP ? "UP" : "DOWN");
|
2013-12-13 17:29:29 +08:00
|
|
|
|
2014-09-12 04:49:27 +08:00
|
|
|
spin_unlock_bh(&slave->bond->mode_lock);
|
2014-10-05 08:45:01 +08:00
|
|
|
|
|
|
|
/* RTNL is held and mode_lock is released so it's safe
|
|
|
|
* to update slave_array here.
|
|
|
|
*/
|
|
|
|
bond_update_slave_arr(slave->bond, NULL);
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
|
2014-01-08 23:46:46 +08:00
|
|
|
/**
|
|
|
|
* bond_3ad_set_carrier - set link state for bonding master
|
|
|
|
* @bond - bonding structure
|
|
|
|
*
|
|
|
|
* if we have an active aggregator, we're up, if not, we're down.
|
|
|
|
* Presumes that we cannot have an active aggregator if there are
|
|
|
|
* no slaves with link up.
|
2006-03-28 05:27:43 +08:00
|
|
|
*
|
2007-06-14 13:11:34 +08:00
|
|
|
* This behavior complies with IEEE 802.3 section 43.3.9.
|
|
|
|
*
|
2006-03-28 05:27:43 +08:00
|
|
|
* Called by bond_set_carrier(). Return zero if carrier state does not
|
|
|
|
* change, nonzero if it does.
|
|
|
|
*/
|
|
|
|
int bond_3ad_set_carrier(struct bonding *bond)
|
|
|
|
{
|
2011-06-22 17:54:39 +08:00
|
|
|
struct aggregator *active;
|
2013-08-01 22:54:47 +08:00
|
|
|
struct slave *first_slave;
|
2014-01-10 18:59:43 +08:00
|
|
|
int ret = 1;
|
2011-06-22 17:54:39 +08:00
|
|
|
|
bonding: add RCU for bond_3ad_state_machine_handler()
The bond_3ad_state_machine_handler() use the bond lock to protect
the bond slave list and slave port together, but it is not enough,
the bond slave list was link and unlink in RTNL, not bond lock,
so I add RCU to protect the slave list from leaving.
The bond lock is still used here, because when the slave has been
removed from the list by the time the state machine runs, it appears
to be possible for both function to manupulate the same aggregator->lag_ports
by finding the aggregator via two different ports that are both members of
that aggregator (i.e., port A of the agg is being unbound, and port B
of the agg is runing its state machine).
If I remove the bond lock, there are nothing to mutex changes
to aggregator->lag_ports between bond_3ad_state_machine_handler and
bond_3ad_unbind_slave, So the bond lock is the simplest way to protect
aggregator->lag_ports.
There was a lot of function need RCU protect, I have two choice
to make the function in RCU-safe, (1) create new similar functions
and make the bond slave list in RCU. (2) modify the existed functions
and make them in read-side critical section, because the RCU
read-side critical sections may be nested.
I choose (2) because it is no need to create more similar functions.
The nots in the function is still too old, clean up the nots.
Suggested-by: Nikolay Aleksandrov <nikolay@redhat.com>
Suggested-by: Jay Vosburgh <fubar@us.ibm.com>
Suggested-by: Veaceslav Falico <vfalico@redhat.com>
Signed-off-by: Ding Tianhong <dingtianhong@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2013-12-13 10:20:12 +08:00
|
|
|
rcu_read_lock();
|
|
|
|
first_slave = bond_first_slave_rcu(bond);
|
2014-01-10 18:59:43 +08:00
|
|
|
if (!first_slave) {
|
|
|
|
ret = 0;
|
|
|
|
goto out;
|
|
|
|
}
|
2014-05-12 15:08:43 +08:00
|
|
|
active = __get_active_agg(&(SLAVE_AD_INFO(first_slave)->aggregator));
|
2011-06-22 17:54:39 +08:00
|
|
|
if (active) {
|
|
|
|
/* are enough slaves available to consider link up? */
|
bonding: fix 802.3ad aggregator reselection
Since commit 7bb11dc9f59d ("bonding: unify all places where
actor-oper key needs to be updated."), the logic in bonding to handle
selection between multiple aggregators has not functioned.
This affects only configurations wherein the bonding slaves
connect to two discrete aggregators (e.g., two independent switches, each
with LACP enabled), thus creating two separate aggregation groups within a
single bond.
The cause is a change in 7bb11dc9f59d to no longer set
AD_PORT_BEGIN on a port after a link state change, which would cause the
port to be reselected for attachment to an aggregator as if were newly
added to the bond. We cannot restore the prior behavior, as it
contradicts IEEE 802.1AX 5.4.12, which requires ports that "become
inoperable" (lose carrier, setting port_enabled=false as per 802.1AX
5.4.7) to remain selected (i.e., assigned to the aggregator). As the port
now remains selected, the aggregator selection logic is not invoked.
A side effect of this change is that aggregators in bonding will
now contain ports that are link down. The aggregator selection logic
does not currently handle this situation correctly, causing incorrect
aggregator selection.
This patch makes two changes to repair the aggregator selection
logic in bonding to function as documented and within the confines of the
standard:
First, the aggregator selection and related logic now utilizes the
number of active ports per aggregator, not the number of selected ports
(as some selected ports may be down). The ad_select "bandwidth" and
"count" options only consider ports that are link up.
Second, on any carrier state change of any slave, the aggregator
selection logic is explicitly called to insure the correct aggregator is
active.
Reported-by: Veli-Matti Lintu <veli-matti.lintu@opinsys.fi>
Fixes: 7bb11dc9f59d ("bonding: unify all places where actor-oper key needs to be updated.")
Signed-off-by: Jay Vosburgh <jay.vosburgh@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2016-06-24 05:20:51 +08:00
|
|
|
if (__agg_active_ports(active) < bond->params.min_links) {
|
2011-06-22 17:54:39 +08:00
|
|
|
if (netif_carrier_ok(bond->dev)) {
|
|
|
|
netif_carrier_off(bond->dev);
|
2014-01-10 18:59:43 +08:00
|
|
|
goto out;
|
2011-06-22 17:54:39 +08:00
|
|
|
}
|
|
|
|
} else if (!netif_carrier_ok(bond->dev)) {
|
2006-03-28 05:27:43 +08:00
|
|
|
netif_carrier_on(bond->dev);
|
2014-01-10 18:59:43 +08:00
|
|
|
goto out;
|
2006-03-28 05:27:43 +08:00
|
|
|
}
|
2014-01-10 18:59:43 +08:00
|
|
|
} else if (netif_carrier_ok(bond->dev)) {
|
2006-03-28 05:27:43 +08:00
|
|
|
netif_carrier_off(bond->dev);
|
|
|
|
}
|
2014-01-10 18:59:43 +08:00
|
|
|
out:
|
|
|
|
rcu_read_unlock();
|
|
|
|
return ret;
|
2006-03-28 05:27:43 +08:00
|
|
|
}
|
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
/**
|
2013-05-18 09:18:31 +08:00
|
|
|
* __bond_3ad_get_active_agg_info - get information of the active aggregator
|
2005-04-17 06:20:36 +08:00
|
|
|
* @bond: bonding struct to work on
|
|
|
|
* @ad_info: ad_info struct to fill with the bond's info
|
|
|
|
*
|
|
|
|
* Returns: 0 on success
|
|
|
|
* < 0 on error
|
|
|
|
*/
|
2013-05-18 09:18:31 +08:00
|
|
|
int __bond_3ad_get_active_agg_info(struct bonding *bond,
|
|
|
|
struct ad_info *ad_info)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
|
|
|
struct aggregator *aggregator = NULL;
|
2013-09-27 22:11:57 +08:00
|
|
|
struct list_head *iter;
|
|
|
|
struct slave *slave;
|
2005-04-17 06:20:36 +08:00
|
|
|
struct port *port;
|
|
|
|
|
2013-10-15 16:28:35 +08:00
|
|
|
bond_for_each_slave_rcu(bond, slave, iter) {
|
2014-05-12 15:08:43 +08:00
|
|
|
port = &(SLAVE_AD_INFO(slave)->port);
|
2005-04-17 06:20:36 +08:00
|
|
|
if (port->aggregator && port->aggregator->is_active) {
|
|
|
|
aggregator = port->aggregator;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2014-02-19 01:42:47 +08:00
|
|
|
if (!aggregator)
|
|
|
|
return -1;
|
|
|
|
|
|
|
|
ad_info->aggregator_id = aggregator->aggregator_identifier;
|
|
|
|
ad_info->ports = aggregator->num_of_ports;
|
|
|
|
ad_info->actor_key = aggregator->actor_oper_aggregator_key;
|
|
|
|
ad_info->partner_key = aggregator->partner_oper_aggregator_key;
|
|
|
|
ether_addr_copy(ad_info->partner_system,
|
|
|
|
aggregator->partner_system.mac_addr_value);
|
|
|
|
return 0;
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
|
2013-05-18 09:18:31 +08:00
|
|
|
int bond_3ad_get_active_agg_info(struct bonding *bond, struct ad_info *ad_info)
|
|
|
|
{
|
|
|
|
int ret;
|
|
|
|
|
2013-10-15 16:28:35 +08:00
|
|
|
rcu_read_lock();
|
2013-05-18 09:18:31 +08:00
|
|
|
ret = __bond_3ad_get_active_agg_info(bond, ad_info);
|
2013-10-15 16:28:35 +08:00
|
|
|
rcu_read_unlock();
|
2013-05-18 09:18:31 +08:00
|
|
|
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2012-06-12 03:23:07 +08:00
|
|
|
int bond_3ad_lacpdu_recv(const struct sk_buff *skb, struct bonding *bond,
|
|
|
|
struct slave *slave)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
2012-06-12 03:23:07 +08:00
|
|
|
struct lacpdu *lacpdu, _lacpdu;
|
|
|
|
|
2011-04-19 11:48:16 +08:00
|
|
|
if (skb->protocol != PKT_TYPE_LACPDU)
|
2014-09-12 04:49:22 +08:00
|
|
|
return RX_HANDLER_ANOTHER;
|
2011-01-20 17:02:31 +08:00
|
|
|
|
2015-02-24 09:50:10 +08:00
|
|
|
if (!MAC_ADDRESS_EQUAL(eth_hdr(skb)->h_dest, lacpdu_mcast_addr))
|
|
|
|
return RX_HANDLER_ANOTHER;
|
|
|
|
|
2012-06-12 03:23:07 +08:00
|
|
|
lacpdu = skb_header_pointer(skb, 0, sizeof(_lacpdu), &_lacpdu);
|
|
|
|
if (!lacpdu)
|
2014-09-12 04:49:22 +08:00
|
|
|
return RX_HANDLER_ANOTHER;
|
2010-09-10 19:43:20 +08:00
|
|
|
|
2014-09-12 04:49:22 +08:00
|
|
|
return bond_3ad_rx_indication(lacpdu, slave, skb->len);
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
2011-06-09 05:19:01 +08:00
|
|
|
|
2014-01-08 23:46:46 +08:00
|
|
|
/**
|
|
|
|
* bond_3ad_update_lacp_rate - change the lacp rate
|
|
|
|
* @bond - bonding struct
|
|
|
|
*
|
2011-06-09 05:19:01 +08:00
|
|
|
* When modify lacp_rate parameter via sysfs,
|
|
|
|
* update actor_oper_port_state of each port.
|
|
|
|
*
|
2014-09-12 04:49:27 +08:00
|
|
|
* Hold bond->mode_lock,
|
2011-06-09 05:19:01 +08:00
|
|
|
* so we can modify port->actor_oper_port_state,
|
|
|
|
* no matter bond is up or down.
|
|
|
|
*/
|
|
|
|
void bond_3ad_update_lacp_rate(struct bonding *bond)
|
|
|
|
{
|
|
|
|
struct port *port = NULL;
|
2013-09-25 15:20:14 +08:00
|
|
|
struct list_head *iter;
|
2013-09-02 19:51:40 +08:00
|
|
|
struct slave *slave;
|
2011-06-09 05:19:01 +08:00
|
|
|
int lacp_fast;
|
|
|
|
|
|
|
|
lacp_fast = bond->params.lacp_fast;
|
2014-09-12 04:49:27 +08:00
|
|
|
spin_lock_bh(&bond->mode_lock);
|
2013-09-25 15:20:14 +08:00
|
|
|
bond_for_each_slave(bond, slave, iter) {
|
2014-05-12 15:08:43 +08:00
|
|
|
port = &(SLAVE_AD_INFO(slave)->port);
|
2011-06-09 05:19:01 +08:00
|
|
|
if (lacp_fast)
|
|
|
|
port->actor_oper_port_state |= AD_STATE_LACP_TIMEOUT;
|
|
|
|
else
|
|
|
|
port->actor_oper_port_state &= ~AD_STATE_LACP_TIMEOUT;
|
|
|
|
}
|
2014-09-12 04:49:27 +08:00
|
|
|
spin_unlock_bh(&bond->mode_lock);
|
2011-06-09 05:19:01 +08:00
|
|
|
}
|