As it noted in the comment before the br_handle_frame_finish
function, this function should be called under rcu_read_lock.
The problem callgraph:
br_dev_xmit -> br_nf_pre_routing_finish_bridge_slow ->
-> br_handle_frame_finish -> br_port_get_rcu -> rcu_dereference
And in this case there is no read-lock section.
I have put lock/unlock in br_nf_pre_routing_finish_bridge_slow,
as the only function that calls it(for now) - is br_dev_xmit.
Found by Linux Driver Verification project (linuxtesting.org).
Signed-off-by: Denis Efremov <[email protected]>
---
include/linux/netfilter_bridge.h | 6 +++++-
1 files changed, 5 insertions(+), 1 deletions(-)
diff --git a/include/linux/netfilter_bridge.h b/include/linux/netfilter_bridge.h
index 31d2844..ceb048e 100644
--- a/include/linux/netfilter_bridge.h
+++ b/include/linux/netfilter_bridge.h
@@ -79,6 +79,7 @@ extern int br_handle_frame_finish(struct sk_buff *skb);
/* Only used in br_device.c */
static inline int br_nf_pre_routing_finish_bridge_slow(struct sk_buff *skb)
{
+ int res;
struct nf_bridge_info *nf_bridge = skb->nf_bridge;
skb_pull(skb, ETH_HLEN);
@@ -86,7 +87,10 @@ static inline int br_nf_pre_routing_finish_bridge_slow(struct sk_buff *skb)
skb_copy_to_linear_data_offset(skb, -(ETH_HLEN-ETH_ALEN),
skb->nf_bridge->data, ETH_HLEN-ETH_ALEN);
skb->dev = nf_bridge->physindev;
- return br_handle_frame_finish(skb);
+ rcu_read_lock();
+ res = br_handle_frame_finish(skb);
+ rcu_read_unlock();
+ return res;
}
/* This is called by the IP fragmenting code and it ensures there is
--
1.7.7
On Tue, 14 Aug 2012 02:47:42 +0400
Denis Efremov <[email protected]> wrote:
> As it noted in the comment before the br_handle_frame_finish
> function, this function should be called under rcu_read_lock.
>
> The problem callgraph:
> br_dev_xmit -> br_nf_pre_routing_finish_bridge_slow ->
> -> br_handle_frame_finish -> br_port_get_rcu -> rcu_dereference
>
> And in this case there is no read-lock section.
> I have put lock/unlock in br_nf_pre_routing_finish_bridge_slow,
> as the only function that calls it(for now) - is br_dev_xmit.
>
> Found by Linux Driver Verification project (linuxtesting.org).
>
> Signed-off-by: Denis Efremov <[email protected]>
> ---
> include/linux/netfilter_bridge.h | 6 +++++-
> 1 files changed, 5 insertions(+), 1 deletions(-)
>
> diff --git a/include/linux/netfilter_bridge.h b/include/linux/netfilter_bridge.h
> index 31d2844..ceb048e 100644
> --- a/include/linux/netfilter_bridge.h
> +++ b/include/linux/netfilter_bridge.h
> @@ -79,6 +79,7 @@ extern int br_handle_frame_finish(struct sk_buff *skb);
> /* Only used in br_device.c */
> static inline int br_nf_pre_routing_finish_bridge_slow(struct sk_buff *skb)
> {
> + int res;
> struct nf_bridge_info *nf_bridge = skb->nf_bridge;
>
> skb_pull(skb, ETH_HLEN);
> @@ -86,7 +87,10 @@ static inline int br_nf_pre_routing_finish_bridge_slow(struct sk_buff *skb)
> skb_copy_to_linear_data_offset(skb, -(ETH_HLEN-ETH_ALEN),
> skb->nf_bridge->data, ETH_HLEN-ETH_ALEN);
> skb->dev = nf_bridge->physindev;
> - return br_handle_frame_finish(skb);
> + rcu_read_lock();
> + res = br_handle_frame_finish(skb);
> + rcu_read_unlock();
> + return res;
> }
>
> /* This is called by the IP fragmenting code and it ensures there is
Why not just move the rcu_read_lock() in br_dev_xmit earlier?
On 14.08.2012 3:36, Stephen Hemminger wrote:
> Why not just move the rcu_read_lock() in br_dev_xmit earlier?
I don't know the semantics of this code, so it's difficult for me to
choose the appropriate place. I thought that the callers
of br_nf_pre_routing_finish_bridge_slow function are limited to
br_dev_xmit and we can modify header without problems.
Anyway, I will send another version of the patch.
As it noted in the comment before the br_handle_frame_finish
function, this function should be called under rcu_read_lock.
The problem callgraph:
br_dev_xmit -> br_nf_pre_routing_finish_bridge_slow ->
-> br_handle_frame_finish -> br_port_get_rcu -> rcu_dereference
And in this case there is no read-lock section.
Found by Linux Driver Verification project (linuxtesting.org).
Signed-off-by: Denis Efremov <[email protected]>
---
net/bridge/br_device.c | 2 ++
1 files changed, 2 insertions(+), 0 deletions(-)
diff --git a/net/bridge/br_device.c b/net/bridge/br_device.c
index 3334845..f9ae07f 100644
--- a/net/bridge/br_device.c
+++ b/net/bridge/br_device.c
@@ -33,7 +33,9 @@ netdev_tx_t br_dev_xmit(struct sk_buff *skb, struct net_device *dev)
#ifdef CONFIG_BRIDGE_NETFILTER
if (skb->nf_bridge && (skb->nf_bridge->mask & BRNF_BRIDGED_DNAT)) {
+ rcu_read_lock();
br_nf_pre_routing_finish_bridge_slow(skb);
+ rcu_read_unlock();
return NETDEV_TX_OK;
}
#endif
--
1.7.7
Alternative solution for problem found by Linux Driver Verification
project (linuxtesting.org).
As it noted in the comment before the br_handle_frame_finish
function, this function should be called under rcu_read_lock.
The problem callgraph:
br_dev_xmit -> br_nf_pre_routing_finish_bridge_slow ->
-> br_handle_frame_finish -> br_port_get_rcu -> rcu_dereference
And in this case there is no read-lock section.
Reported-by: Denis Efremov <[email protected]>
Signed-off-by: Stephen Hemminger <[email protected]>
---
net/bridge/br_device.c | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/net/bridge/br_device.c b/net/bridge/br_device.c
index 3334845..8a3b50f 100644
--- a/net/bridge/br_device.c
+++ b/net/bridge/br_device.c
@@ -31,9 +31,11 @@ netdev_tx_t br_dev_xmit(struct sk_buff *skb, struct net_device *dev)
struct net_bridge_mdb_entry *mdst;
struct br_cpu_netstats *brstats = this_cpu_ptr(br->stats);
+ rcu_read_lock();
#ifdef CONFIG_BRIDGE_NETFILTER
if (skb->nf_bridge && (skb->nf_bridge->mask & BRNF_BRIDGED_DNAT)) {
br_nf_pre_routing_finish_bridge_slow(skb);
+ rcu_read_unlock();
return NETDEV_TX_OK;
}
#endif
@@ -48,7 +50,6 @@ netdev_tx_t br_dev_xmit(struct sk_buff *skb, struct net_device *dev)
skb_reset_mac_header(skb);
skb_pull(skb, ETH_HLEN);
- rcu_read_lock();
if (is_broadcast_ether_addr(dest))
br_flood_deliver(br, skb);
else if (is_multicast_ether_addr(dest)) {
--
1.7.10.4
From: Stephen Hemminger <[email protected]>
Date: Tue, 14 Aug 2012 08:19:33 -0700
> Alternative solution for problem found by Linux Driver Verification
> project (linuxtesting.org).
>
> As it noted in the comment before the br_handle_frame_finish
> function, this function should be called under rcu_read_lock.
>
> The problem callgraph:
> br_dev_xmit -> br_nf_pre_routing_finish_bridge_slow ->
> -> br_handle_frame_finish -> br_port_get_rcu -> rcu_dereference
>
> And in this case there is no read-lock section.
>
> Reported-by: Denis Efremov <[email protected]>
> Signed-off-by: Stephen Hemminger <[email protected]>
Applied, thanks Stephen.
Please CC: this to netdev next time. It's in a grey area whether I or
the netfilter folks should apply this one, and here I'm deciding to
take care of it.