[PATCH v2 net-next 0/2] bonding: fix locking in bond_ab_arp_probe

[PATCH v2 net-next 0/2] bonding: fix locking in bond_ab_arp_probe

[Veaceslav Falico]

Hi,

After the latest patches, on every call of bond_ab_arp_probe() without an
active slave I see the following warning:

[    7.912314] RTNL: assertion failed at net/core/dev.c (4494)
...
[    7.922495]  [<ffffffff817acc6f>] dump_stack+0x51/0x72
[    7.923714]  [<ffffffff8168795e>] netdev_master_upper_dev_get+0x6e/0x70
[    7.924940]  [<ffffffff816a2a66>] rtnl_link_fill+0x116/0x260
[    7.926143]  [<ffffffff817acc6f>] ? dump_stack+0x51/0x72
[    7.927333]  [<ffffffff816a350c>] rtnl_fill_ifinfo+0x95c/0xb90
[    7.928529]  [<ffffffff8167af2b>] ? __kmalloc_reserve+0x3b/0xa0
[    7.929681]  [<ffffffff8167bfcf>] ? __alloc_skb+0x9f/0x1e0
[    7.930827]  [<ffffffff816a3b64>] rtmsg_ifinfo+0x84/0x100
[    7.931960]  [<ffffffffa00bca07>] bond_ab_arp_probe+0x1a7/0x370 [bonding]
[    7.933133]  [<ffffffffa00bcd78>] bond_activebackup_arp_mon+0x1a8/0x2f0 [bonding]
...

It happens because in bond_ab_arp_probe() we change the flags of a slave
without holding the RTNL lock.

To fix this - remove the useless curr_active_lock, RCUify it and lock RTNL
while changing the slave's flags. Also, remove bond_ab_arp_probe() from
under any locks in bond_ab_arp_mon().

CC: Jay Vosburgh <fubar@us.ibm.com>
CC: Andy Gospodarek <andy@greyhouse.net>
CC: netdev@vger.kernel.org
Signed-off-by: Veaceslav Falico <vfalico@redhat.com>

---
 drivers/net/bonding/bond_main.c | 48 +++++++++++++++++++++++------------------
 1 file changed, 27 insertions(+), 21 deletions(-)

[PATCH v2 net-next 1/2] bonding: RCUify bond_ab_arp_probe

[Veaceslav Falico]

Currently bond_ab_arp_probe() is always called under rcu_read_lock(),
however to work with curr_active_slave we're still holding the
curr_slave_lock.

To remove that curr_slave_lock - rcu_dereference the bond's
curr_active_slave and use it further - so that we're sure the slave won't
go away, and we don't care if it will change in the meanwhile.

CC: Jay Vosburgh <fubar@us.ibm.com>
CC: Andy Gospodarek <andy@greyhouse.net>
Signed-off-by: Veaceslav Falico <vfalico@redhat.com>
---
 drivers/net/bonding/bond_main.c | 16 ++++++----------
 1 file changed, 6 insertions(+), 10 deletions(-)

diff --git a/drivers/net/bonding/bond_main.c b/drivers/net/bonding/bond_main.c
index f9e0c8b..22d8b69 100644
--- a/drivers/net/bonding/bond_main.c
+++ b/drivers/net/bonding/bond_main.c
@@ -2605,25 +2605,21 @@ do_failover:
 static void bond_ab_arp_probe(struct bonding *bond)
 {
 	struct slave *slave, *before = NULL, *new_slave = NULL,
-		     *curr_arp_slave = rcu_dereference(bond->current_arp_slave);
+		     *curr_arp_slave = rcu_dereference(bond->current_arp_slave),
+		     *curr_active_slave = rcu_dereference(bond->curr_active_slave);
 	struct list_head *iter;
 	bool found = false;
 
-	read_lock(&bond->curr_slave_lock);
-
-	if (curr_arp_slave && bond->curr_active_slave)
+	if (curr_arp_slave && curr_active_slave)
 		pr_info("PROBE: c_arp %s && cas %s BAD\n",
 			curr_arp_slave->dev->name,
-			bond->curr_active_slave->dev->name);
+			curr_active_slave->dev->name);
 
-	if (bond->curr_active_slave) {
-		bond_arp_send_all(bond, bond->curr_active_slave);
-		read_unlock(&bond->curr_slave_lock);
+	if (curr_active_slave) {
+		bond_arp_send_all(bond, curr_active_slave);
 		return;
 	}
 
-	read_unlock(&bond->curr_slave_lock);
-
 	/* if we don't have a curr_active_slave, search for the next available
 	 * backup slave from the current_arp_slave and make it the candidate
 	 * for becoming the curr_active_slave
-- 
1.8.4

[PATCH v2 net-next 2/2] bonding: restructure locking of bond_ab_arp_probe()

[Veaceslav Falico]

Currently we're calling it from under RCU context, however we're using some
functions that require rtnl to be held.

Fix this by restructuring the locking - don't call it under any locks,
aquire rcu_read_lock() if we're sending _only_ (i.e. we have the active
slave present), and use rtnl locking otherwise - if we need to modify
(in)active flags of a slave.

CC: Jay Vosburgh <fubar@us.ibm.com>
CC: Andy Gospodarek <andy@greyhouse.net>
Signed-off-by: Veaceslav Falico <vfalico@redhat.com>
---
 drivers/net/bonding/bond_main.c | 38 ++++++++++++++++++++++++--------------
 1 file changed, 24 insertions(+), 14 deletions(-)

diff --git a/drivers/net/bonding/bond_main.c b/drivers/net/bonding/bond_main.c
index 22d8b69..f879e9e 100644
--- a/drivers/net/bonding/bond_main.c
+++ b/drivers/net/bonding/bond_main.c
@@ -2605,11 +2605,14 @@ do_failover:
 static void bond_ab_arp_probe(struct bonding *bond)
 {
 	struct slave *slave, *before = NULL, *new_slave = NULL,
-		     *curr_arp_slave = rcu_dereference(bond->current_arp_slave),
-		     *curr_active_slave = rcu_dereference(bond->curr_active_slave);
+		     *curr_arp_slave, *curr_active_slave;
 	struct list_head *iter;
 	bool found = false;
 
+	rcu_read_lock();
+	curr_arp_slave = rcu_dereference(bond->current_arp_slave);
+	curr_active_slave = rcu_dereference(bond->curr_active_slave);
+
 	if (curr_arp_slave && curr_active_slave)
 		pr_info("PROBE: c_arp %s && cas %s BAD\n",
 			curr_arp_slave->dev->name,
@@ -2617,23 +2620,31 @@ static void bond_ab_arp_probe(struct bonding *bond)
 
 	if (curr_active_slave) {
 		bond_arp_send_all(bond, curr_active_slave);
+		rcu_read_unlock();
 		return;
 	}
+	rcu_read_unlock();
 
 	/* if we don't have a curr_active_slave, search for the next available
 	 * backup slave from the current_arp_slave and make it the candidate
 	 * for becoming the curr_active_slave
 	 */
 
+	rtnl_lock();
+	/* curr_arp_slave might have gone away */
+	curr_arp_slave = rcu_dereference(bond->current_arp_slave);
+
 	if (!curr_arp_slave) {
-		curr_arp_slave = bond_first_slave_rcu(bond);
-		if (!curr_arp_slave)
+		curr_arp_slave = bond_first_slave(bond);
+		if (!curr_arp_slave) {
+			rtnl_unlock();
 			return;
+		}
 	}
 
 	bond_set_slave_inactive_flags(curr_arp_slave);
 
-	bond_for_each_slave_rcu(bond, slave, iter) {
+	bond_for_each_slave(bond, slave, iter) {
 		if (!found && !before && IS_UP(slave->dev))
 			before = slave;
 
@@ -2663,21 +2674,24 @@ static void bond_ab_arp_probe(struct bonding *bond)
 	if (!new_slave && before)
 		new_slave = before;
 
-	if (!new_slave)
+	if (!new_slave) {
+		rtnl_unlock();
 		return;
+	}
 
 	new_slave->link = BOND_LINK_BACK;
 	bond_set_slave_active_flags(new_slave);
 	bond_arp_send_all(bond, new_slave);
 	new_slave->jiffies = jiffies;
 	rcu_assign_pointer(bond->current_arp_slave, new_slave);
+	rtnl_unlock();
 }
 
 static void bond_activebackup_arp_mon(struct work_struct *work)
 {
 	struct bonding *bond = container_of(work, struct bonding,
 					    arp_work.work);
-	bool should_notify_peers = false;
+	bool should_notify_peers = false, should_commit = false;
 	int delta_in_ticks;
 
 	delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
@@ -2686,12 +2700,11 @@ static void bond_activebackup_arp_mon(struct work_struct *work)
 		goto re_arm;
 
 	rcu_read_lock();
-
 	should_notify_peers = bond_should_notify_peers(bond);
+	should_commit = bond_ab_arp_inspect(bond);
+	rcu_read_unlock();
 
-	if (bond_ab_arp_inspect(bond)) {
-		rcu_read_unlock();
-
+	if (should_commit) {
 		/* Race avoidance with bond_close flush of workqueue */
 		if (!rtnl_trylock()) {
 			delta_in_ticks = 1;
@@ -2700,13 +2713,10 @@ static void bond_activebackup_arp_mon(struct work_struct *work)
 		}
 
 		bond_ab_arp_commit(bond);
-
 		rtnl_unlock();
-		rcu_read_lock();
 	}
 
 	bond_ab_arp_probe(bond);
-	rcu_read_unlock();
 
 re_arm:
 	if (bond->params.arp_interval)
-- 
1.8.4

Re: [PATCH v2 net-next 2/2] bonding: restructure locking of bond_ab_arp_probe()

[Jay Vosburgh]

Veaceslav Falico <vfalico@redhat.com> wrote:

>Currently we're calling it from under RCU context, however we're using some
>functions that require rtnl to be held.
>
>Fix this by restructuring the locking - don't call it under any locks,
>aquire rcu_read_lock() if we're sending _only_ (i.e. we have the active
>slave present), and use rtnl locking otherwise - if we need to modify
>(in)active flags of a slave.
>
>CC: Jay Vosburgh <fubar@us.ibm.com>
>CC: Andy Gospodarek <andy@greyhouse.net>
>Signed-off-by: Veaceslav Falico <vfalico@redhat.com>
>---
> drivers/net/bonding/bond_main.c | 38 ++++++++++++++++++++++++--------------
> 1 file changed, 24 insertions(+), 14 deletions(-)
>
>diff --git a/drivers/net/bonding/bond_main.c b/drivers/net/bonding/bond_main.c
>index 22d8b69..f879e9e 100644
>--- a/drivers/net/bonding/bond_main.c
>+++ b/drivers/net/bonding/bond_main.c
>@@ -2605,11 +2605,14 @@ do_failover:
> static void bond_ab_arp_probe(struct bonding *bond)
> {
> 	struct slave *slave, *before = NULL, *new_slave = NULL,
>-		     *curr_arp_slave = rcu_dereference(bond->current_arp_slave),
>-		     *curr_active_slave = rcu_dereference(bond->curr_active_slave);
>+		     *curr_arp_slave, *curr_active_slave;
> 	struct list_head *iter;
> 	bool found = false;
>
>+	rcu_read_lock();
>+	curr_arp_slave = rcu_dereference(bond->current_arp_slave);
>+	curr_active_slave = rcu_dereference(bond->curr_active_slave);
>+
> 	if (curr_arp_slave && curr_active_slave)
> 		pr_info("PROBE: c_arp %s && cas %s BAD\n",
> 			curr_arp_slave->dev->name,
>@@ -2617,23 +2620,31 @@ static void bond_ab_arp_probe(struct bonding *bond)
>
> 	if (curr_active_slave) {
> 		bond_arp_send_all(bond, curr_active_slave);
>+		rcu_read_unlock();
> 		return;
> 	}
>+	rcu_read_unlock();
>
> 	/* if we don't have a curr_active_slave, search for the next available
> 	 * backup slave from the current_arp_slave and make it the candidate
> 	 * for becoming the curr_active_slave
> 	 */
>
>+	rtnl_lock();

	I don't believe we can unconditionally acquire RTNL here, as it
may deadlock with bond_close's cancel_delayed_work_sync() calls (which
occur under RTNL).

	I think I'd make this a rtnl_trylock, and if that fails, have
the bond_ab_arp_probe function return non-zero as an indication for
activebackup_arp_mon to change delta_in_ticks to 1.  Changing the delta
is just an optimization; without it, things will still work, but it will
take longer to run the curr_arp_slave probe cycle.

	-J

>+	/* curr_arp_slave might have gone away */
>+	curr_arp_slave = rcu_dereference(bond->current_arp_slave);
>+
> 	if (!curr_arp_slave) {
>-		curr_arp_slave = bond_first_slave_rcu(bond);
>-		if (!curr_arp_slave)
>+		curr_arp_slave = bond_first_slave(bond);
>+		if (!curr_arp_slave) {
>+			rtnl_unlock();
> 			return;
>+		}
> 	}
>
> 	bond_set_slave_inactive_flags(curr_arp_slave);
>
>-	bond_for_each_slave_rcu(bond, slave, iter) {
>+	bond_for_each_slave(bond, slave, iter) {
> 		if (!found && !before && IS_UP(slave->dev))
> 			before = slave;
>
>@@ -2663,21 +2674,24 @@ static void bond_ab_arp_probe(struct bonding *bond)
> 	if (!new_slave && before)
> 		new_slave = before;
>
>-	if (!new_slave)
>+	if (!new_slave) {
>+		rtnl_unlock();
> 		return;
>+	}
>
> 	new_slave->link = BOND_LINK_BACK;
> 	bond_set_slave_active_flags(new_slave);
> 	bond_arp_send_all(bond, new_slave);
> 	new_slave->jiffies = jiffies;
> 	rcu_assign_pointer(bond->current_arp_slave, new_slave);
>+	rtnl_unlock();
> }
>
> static void bond_activebackup_arp_mon(struct work_struct *work)
> {
> 	struct bonding *bond = container_of(work, struct bonding,
> 					    arp_work.work);
>-	bool should_notify_peers = false;
>+	bool should_notify_peers = false, should_commit = false;
> 	int delta_in_ticks;
>
> 	delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
>@@ -2686,12 +2700,11 @@ static void bond_activebackup_arp_mon(struct work_struct *work)
> 		goto re_arm;
>
> 	rcu_read_lock();
>-
> 	should_notify_peers = bond_should_notify_peers(bond);
>+	should_commit = bond_ab_arp_inspect(bond);
>+	rcu_read_unlock();
>
>-	if (bond_ab_arp_inspect(bond)) {
>-		rcu_read_unlock();
>-
>+	if (should_commit) {
> 		/* Race avoidance with bond_close flush of workqueue */
> 		if (!rtnl_trylock()) {
> 			delta_in_ticks = 1;
>@@ -2700,13 +2713,10 @@ static void bond_activebackup_arp_mon(struct work_struct *work)
> 		}
>
> 		bond_ab_arp_commit(bond);
>-
> 		rtnl_unlock();
>-		rcu_read_lock();
> 	}
>
> 	bond_ab_arp_probe(bond);
>-	rcu_read_unlock();
>
> re_arm:
> 	if (bond->params.arp_interval)
>-- 
>1.8.4
>

---
	-Jay Vosburgh, IBM Linux Technology Center, fubar@us.ibm.com

Re: [PATCH v2 net-next 2/2] bonding: restructure locking of bond_ab_arp_probe()

[Veaceslav Falico]

On Thu, Jan 23, 2014 at 11:25:38AM -0800, Jay Vosburgh wrote:
>Veaceslav Falico <vfalico@redhat.com> wrote:
>
>>Currently we're calling it from under RCU context, however we're using some
>>functions that require rtnl to be held.
>>
>>Fix this by restructuring the locking - don't call it under any locks,
>>aquire rcu_read_lock() if we're sending _only_ (i.e. we have the active
>>slave present), and use rtnl locking otherwise - if we need to modify
>>(in)active flags of a slave.
>>
>>CC: Jay Vosburgh <fubar@us.ibm.com>
>>CC: Andy Gospodarek <andy@greyhouse.net>
>>Signed-off-by: Veaceslav Falico <vfalico@redhat.com>
>>---
>> drivers/net/bonding/bond_main.c | 38 ++++++++++++++++++++++++--------------
>> 1 file changed, 24 insertions(+), 14 deletions(-)
>>
>>diff --git a/drivers/net/bonding/bond_main.c b/drivers/net/bonding/bond_main.c
>>index 22d8b69..f879e9e 100644
>>--- a/drivers/net/bonding/bond_main.c
>>+++ b/drivers/net/bonding/bond_main.c
>>@@ -2605,11 +2605,14 @@ do_failover:
>> static void bond_ab_arp_probe(struct bonding *bond)
>> {
>> 	struct slave *slave, *before = NULL, *new_slave = NULL,
>>-		     *curr_arp_slave = rcu_dereference(bond->current_arp_slave),
>>-		     *curr_active_slave = rcu_dereference(bond->curr_active_slave);
>>+		     *curr_arp_slave, *curr_active_slave;
>> 	struct list_head *iter;
>> 	bool found = false;
>>
>>+	rcu_read_lock();
>>+	curr_arp_slave = rcu_dereference(bond->current_arp_slave);
>>+	curr_active_slave = rcu_dereference(bond->curr_active_slave);
>>+
>> 	if (curr_arp_slave && curr_active_slave)
>> 		pr_info("PROBE: c_arp %s && cas %s BAD\n",
>> 			curr_arp_slave->dev->name,
>>@@ -2617,23 +2620,31 @@ static void bond_ab_arp_probe(struct bonding *bond)
>>
>> 	if (curr_active_slave) {
>> 		bond_arp_send_all(bond, curr_active_slave);
>>+		rcu_read_unlock();
>> 		return;
>> 	}
>>+	rcu_read_unlock();
>>
>> 	/* if we don't have a curr_active_slave, search for the next available
>> 	 * backup slave from the current_arp_slave and make it the candidate
>> 	 * for becoming the curr_active_slave
>> 	 */
>>
>>+	rtnl_lock();
>
>	I don't believe we can unconditionally acquire RTNL here, as it
>may deadlock with bond_close's cancel_delayed_work_sync() calls (which
>occur under RTNL).
>
>	I think I'd make this a rtnl_trylock, and if that fails, have
>the bond_ab_arp_probe function return non-zero as an indication for
>activebackup_arp_mon to change delta_in_ticks to 1.  Changing the delta
>is just an optimization; without it, things will still work, but it will
>take longer to run the curr_arp_slave probe cycle.

True, missed this race. Will re-send once I'll fix it.

Thank you!

>
>	-J
>
>>+	/* curr_arp_slave might have gone away */
>>+	curr_arp_slave = rcu_dereference(bond->current_arp_slave);
>>+
>> 	if (!curr_arp_slave) {
>>-		curr_arp_slave = bond_first_slave_rcu(bond);
>>-		if (!curr_arp_slave)
>>+		curr_arp_slave = bond_first_slave(bond);
>>+		if (!curr_arp_slave) {
>>+			rtnl_unlock();
>> 			return;
>>+		}
>> 	}
>>
>> 	bond_set_slave_inactive_flags(curr_arp_slave);
>>
>>-	bond_for_each_slave_rcu(bond, slave, iter) {
>>+	bond_for_each_slave(bond, slave, iter) {
>> 		if (!found && !before && IS_UP(slave->dev))
>> 			before = slave;
>>
>>@@ -2663,21 +2674,24 @@ static void bond_ab_arp_probe(struct bonding *bond)
>> 	if (!new_slave && before)
>> 		new_slave = before;
>>
>>-	if (!new_slave)
>>+	if (!new_slave) {
>>+		rtnl_unlock();
>> 		return;
>>+	}
>>
>> 	new_slave->link = BOND_LINK_BACK;
>> 	bond_set_slave_active_flags(new_slave);
>> 	bond_arp_send_all(bond, new_slave);
>> 	new_slave->jiffies = jiffies;
>> 	rcu_assign_pointer(bond->current_arp_slave, new_slave);
>>+	rtnl_unlock();
>> }
>>
>> static void bond_activebackup_arp_mon(struct work_struct *work)
>> {
>> 	struct bonding *bond = container_of(work, struct bonding,
>> 					    arp_work.work);
>>-	bool should_notify_peers = false;
>>+	bool should_notify_peers = false, should_commit = false;
>> 	int delta_in_ticks;
>>
>> 	delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
>>@@ -2686,12 +2700,11 @@ static void bond_activebackup_arp_mon(struct work_struct *work)
>> 		goto re_arm;
>>
>> 	rcu_read_lock();
>>-
>> 	should_notify_peers = bond_should_notify_peers(bond);
>>+	should_commit = bond_ab_arp_inspect(bond);
>>+	rcu_read_unlock();
>>
>>-	if (bond_ab_arp_inspect(bond)) {
>>-		rcu_read_unlock();
>>-
>>+	if (should_commit) {
>> 		/* Race avoidance with bond_close flush of workqueue */
>> 		if (!rtnl_trylock()) {
>> 			delta_in_ticks = 1;
>>@@ -2700,13 +2713,10 @@ static void bond_activebackup_arp_mon(struct work_struct *work)
>> 		}
>>
>> 		bond_ab_arp_commit(bond);
>>-
>> 		rtnl_unlock();
>>-		rcu_read_lock();
>> 	}
>>
>> 	bond_ab_arp_probe(bond);
>>-	rcu_read_unlock();
>>
>> re_arm:
>> 	if (bond->params.arp_interval)
>>--
>>1.8.4
>>
>
>---
>	-Jay Vosburgh, IBM Linux Technology Center, fubar@us.ibm.com
>