Linux Ethernet Bridging - Jan 2022 - [Bridge] [PATCH net] net: bridge: vlan: Fix dumping with ifindex

On 25/01/2022 08:19, Benjamin Poirier wrote:
> Specifying ifindex in a RTM_GETVLAN dump leads to an infinite repetition
> of the same entries. netlink_dump() normally calls the dump function
> repeatedly until it returns 0 which br_vlan_rtm_dump() never does in
> that case.
> 
> Fixes: 8dcea187088b ("net: bridge: vlan: add rtm definitions and dump
support")
> Signed-off-by: Benjamin Poirier <bpoirier at nvidia.com>
> ---
>  net/bridge/br_vlan.c | 6 ++++--
>  1 file changed, 4 insertions(+), 2 deletions(-)
> 
[snip]
> 
> diff --git a/net/bridge/br_vlan.c b/net/bridge/br_vlan.c
> index 84ba456a78cc..2e606f2b9a4d 100644
> --- a/net/bridge/br_vlan.c
> +++ b/net/bridge/br_vlan.c
> @@ -2013,7 +2013,7 @@ static int br_vlan_rtm_dump(struct sk_buff *skb,
struct netlink_callback *cb)
>  		dump_flags = nla_get_u32(dtb[BRIDGE_VLANDB_DUMP_FLAGS]);
>  
>  	rcu_read_lock();
> -	if (bvm->ifindex) {
> +	if (bvm->ifindex && !s_idx) {
>  		dev = dev_get_by_index_rcu(net, bvm->ifindex);
>  		if (!dev) {
>  			err = -ENODEV;
> @@ -2022,7 +2022,9 @@ static int br_vlan_rtm_dump(struct sk_buff *skb,
struct netlink_callback *cb)
>  		err = br_vlan_dump_dev(dev, skb, cb, dump_flags);
>  		if (err && err != -EMSGSIZE)
>  			goto out_err;
> -	} else {
> +		else if (!err)
> +			idx++;
> +	} else if (!bvm->ifindex) {
>  		for_each_netdev_rcu(net, dev) {
>  			if (idx < s_idx)
>  				goto skip;
Acked-by: Nikolay Aleksandrov <nikolay at nvidia.com>

On 25/01/2022 10:24, Nikolay Aleksandrov wrote:
> On 25/01/2022 08:19, Benjamin Poirier wrote:
>> Specifying ifindex in a RTM_GETVLAN dump leads to an infinite
repetition
>> of the same entries. netlink_dump() normally calls the dump function
>> repeatedly until it returns 0 which br_vlan_rtm_dump() never does in
>> that case.
>>
>> Fixes: 8dcea187088b ("net: bridge: vlan: add rtm definitions and
dump support")
>> Signed-off-by: Benjamin Poirier <bpoirier at nvidia.com>
>> ---
>>  net/bridge/br_vlan.c | 6 ++++--
>>  1 file changed, 4 insertions(+), 2 deletions(-)
>>
> [snip]
>>
>> diff --git a/net/bridge/br_vlan.c b/net/bridge/br_vlan.c
>> index 84ba456a78cc..2e606f2b9a4d 100644
>> --- a/net/bridge/br_vlan.c
>> +++ b/net/bridge/br_vlan.c
>> @@ -2013,7 +2013,7 @@ static int br_vlan_rtm_dump(struct sk_buff *skb,
struct netlink_callback *cb)
>>  		dump_flags = nla_get_u32(dtb[BRIDGE_VLANDB_DUMP_FLAGS]);
>>  
>>  	rcu_read_lock();
>> -	if (bvm->ifindex) {
>> +	if (bvm->ifindex && !s_idx) {
>>  		dev = dev_get_by_index_rcu(net, bvm->ifindex);
>>  		if (!dev) {
>>  			err = -ENODEV;
>> @@ -2022,7 +2022,9 @@ static int br_vlan_rtm_dump(struct sk_buff *skb,
struct netlink_callback *cb)
>>  		err = br_vlan_dump_dev(dev, skb, cb, dump_flags);
>>  		if (err && err != -EMSGSIZE)
>>  			goto out_err;
>> -	} else {
>> +		else if (!err)
>> +			idx++;
>> +	} else if (!bvm->ifindex) {
>>  		for_each_netdev_rcu(net, dev) {
>>  			if (idx < s_idx)
>>  				goto skip;
> 
> Acked-by: Nikolay Aleksandrov <nikolay at nvidia.com>
Actually I'd prefer an alternative that would encapsulate handling the
single
device dump in its block, avoid all the "else if"s and is simpler
(untested):

diff --git a/net/bridge/br_vlan.c b/net/bridge/br_vlan.c
index 84ba456a78cc..43201260e37b 100644
--- a/net/bridge/br_vlan.c
+++ b/net/bridge/br_vlan.c
@@ -2020,7 +2020,8 @@ static int br_vlan_rtm_dump(struct sk_buff *skb, struct
netlink_callback *cb)
                        goto out_err;
                }
                err = br_vlan_dump_dev(dev, skb, cb, dump_flags);
-               if (err && err != -EMSGSIZE)
+               /* if the dump completed without an error we return 0 here */
+               if (err != -EMSGSIZE)
                        goto out_err;
        } else {
                for_each_netdev_rcu(net, dev) {