[PATCH net] bridge: use _bh spinlock variant for br_fdb_update to avoid lockup

[PATCH net] bridge: use _bh spinlock variant for br_fdb_update to avoid lockup

[Nikolay Aleksandrov]

From: Wilson Kok <wkok@cumulusnetworks.com>

br_fdb_update() can be called in process context in the following way:
br_fdb_add() -> __br_fdb_add() -> br_fdb_update() (if NTF_USE flag is set)
so we need to use spin_lock_bh because there are softirq users of the
hash_lock. One easy way to reproduce this is to modify the bridge utility
to set NTF_USE, enable stp and then set maxageing to a low value so
br_fdb_cleanup() is called frequently and then just add new entries in
a loop. This happens because br_fdb_cleanup() is called from timer/softirq
context. These locks were _bh before commit f8ae737deea1
("[BRIDGE]: forwarding remove unneeded preempt and bh diasables")
and at the time that commit was correct because br_fdb_update() couldn't be
called from process context, but that changed after commit:
292d1398983f ("bridge: add NTF_USE support")

Signed-off-by: Wilson Kok <wkok@cumulusnetworks.com>
Signed-off-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Fixes: 292d1398983f ("bridge: add NTF_USE support")
---
Nik: Something that just occurred - we can disable softirqs around the call
     to br_fdb_update() in br_fdb_add() so we can keep the optimization.
     What do you think ?

 net/bridge/br_fdb.c |    4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/net/bridge/br_fdb.c b/net/bridge/br_fdb.c
index 7896cf143045..523d329a5bd1 100644
--- a/net/bridge/br_fdb.c
+++ b/net/bridge/br_fdb.c
@@ -569,7 +569,7 @@ void br_fdb_update(struct net_bridge *br, struct net_bridge_port *source,
 				fdb_notify(br, fdb, RTM_NEWNEIGH);
 		}
 	} else {
-		spin_lock(&br->hash_lock);
+		spin_lock_bh(&br->hash_lock);
 		if (likely(!fdb_find(head, addr, vid))) {
 			fdb = fdb_create(head, source, addr, vid);
 			if (fdb) {
@@ -581,7 +581,7 @@ void br_fdb_update(struct net_bridge *br, struct net_bridge_port *source,
 		/* else  we lose race and someone else inserts
 		 * it first, don't bother updating
 		 */
-		spin_unlock(&br->hash_lock);
+		spin_unlock_bh(&br->hash_lock);
 	}
 }
 
-- 
1.7.10.4

[PATCH net v2] bridge: disable softirqs around br_fdb_update to avoid lockup

[Nikolay Aleksandrov]

br_fdb_update() can be called in process context in the following way:
br_fdb_add() -> __br_fdb_add() -> br_fdb_update() (if NTF_USE flag is set)
so we need to disable softirqs because there are softirq users of the
hash_lock. One easy way to reproduce this is to modify the bridge utility
to set NTF_USE, enable stp and then set maxageing to a low value so
br_fdb_cleanup() is called frequently and then just add new entries in
a loop. This happens because br_fdb_cleanup() is called from timer/softirq
context. The spin locks in br_fdb_update were _bh before commit f8ae737deea1
("[BRIDGE]: forwarding remove unneeded preempt and bh diasables")
and at the time that commit was correct because br_fdb_update() couldn't be
called from process context, but that changed after commit:
292d1398983f ("bridge: add NTF_USE support")
Using local_bh_disable/enable around br_fdb_update() allows us to keep
using the spin_lock/unlock in br_fdb_update for the fast-path.

Signed-off-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Fixes: 292d1398983f ("bridge: add NTF_USE support")
---
v2: new patch, uses local_bh_disable/enable instead of _bh variant
    of spinlocks to keep the fast-path optimization. Please drop
    "bridge: use _bh spinlock variant for br_fdb_update to avoid lockup"
    in favor of this one.

 net/bridge/br_fdb.c |    2 ++
 1 file changed, 2 insertions(+)

diff --git a/net/bridge/br_fdb.c b/net/bridge/br_fdb.c
index e0670d7054f9..659fb96672e4 100644
--- a/net/bridge/br_fdb.c
+++ b/net/bridge/br_fdb.c
@@ -796,9 +796,11 @@ static int __br_fdb_add(struct ndmsg *ndm, struct net_bridge_port *p,
 	int err = 0;
 
 	if (ndm->ndm_flags & NTF_USE) {
+		local_bh_disable();
 		rcu_read_lock();
 		br_fdb_update(p->br, p, addr, vid, true);
 		rcu_read_unlock();
+		local_bh_enable();
 	} else {
 		spin_lock_bh(&p->br->hash_lock);
 		err = fdb_add_entry(p, addr, ndm->ndm_state,
-- 
1.7.10.4

Re: [PATCH net v2] bridge: disable softirqs around br_fdb_update to avoid lockup

[David Miller]

From: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Date: Sat,  6 Jun 2015 06:49:00 -0700

> br_fdb_update() can be called in process context in the following way:
> br_fdb_add() -> __br_fdb_add() -> br_fdb_update() (if NTF_USE flag is set)
> so we need to disable softirqs because there are softirq users of the
> hash_lock. One easy way to reproduce this is to modify the bridge utility
> to set NTF_USE, enable stp and then set maxageing to a low value so
> br_fdb_cleanup() is called frequently and then just add new entries in
> a loop. This happens because br_fdb_cleanup() is called from timer/softirq
> context. The spin locks in br_fdb_update were _bh before commit f8ae737deea1
> ("[BRIDGE]: forwarding remove unneeded preempt and bh diasables")
> and at the time that commit was correct because br_fdb_update() couldn't be
> called from process context, but that changed after commit:
> 292d1398983f ("bridge: add NTF_USE support")
> Using local_bh_disable/enable around br_fdb_update() allows us to keep
> using the spin_lock/unlock in br_fdb_update for the fast-path.
> 
> Signed-off-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
> Fixes: 292d1398983f ("bridge: add NTF_USE support")
> ---
> v2: new patch, uses local_bh_disable/enable instead of _bh variant
>     of spinlocks to keep the fast-path optimization. Please drop
>     "bridge: use _bh spinlock variant for br_fdb_update to avoid lockup"
>     in favor of this one.

Ok, I accidently applied v1, so I reverted that and applied and queued
this one instead for -stable.

Thanks.

Re: [PATCH net] bridge: use _bh spinlock variant for br_fdb_update to avoid lockup

[David Miller]

From: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Date: Fri,  5 Jun 2015 00:52:57 -0700

> From: Wilson Kok <wkok@cumulusnetworks.com>
> 
> br_fdb_update() can be called in process context in the following way:
> br_fdb_add() -> __br_fdb_add() -> br_fdb_update() (if NTF_USE flag is set)
> so we need to use spin_lock_bh because there are softirq users of the
> hash_lock. One easy way to reproduce this is to modify the bridge utility
> to set NTF_USE, enable stp and then set maxageing to a low value so
> br_fdb_cleanup() is called frequently and then just add new entries in
> a loop. This happens because br_fdb_cleanup() is called from timer/softirq
> context. These locks were _bh before commit f8ae737deea1
> ("[BRIDGE]: forwarding remove unneeded preempt and bh diasables")
> and at the time that commit was correct because br_fdb_update() couldn't be
> called from process context, but that changed after commit:
> 292d1398983f ("bridge: add NTF_USE support")
> 
> Signed-off-by: Wilson Kok <wkok@cumulusnetworks.com>
> Signed-off-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
> Fixes: 292d1398983f ("bridge: add NTF_USE support")

Applied and queued up for -stable, thanks Nik.