Re: [patch net-next] net: sched: move block device tracking into tcf_block_get/put_ext()

[Jiri Pirko <jiri@resnulli.us>]

Wed, Jan 10, 2024 at 01:09:55PM CET, idosch@idosch.org wrote:
>On Thu, Jan 04, 2024 at 01:58:44PM +0100, Jiri Pirko wrote:
>> diff --git a/net/sched/cls_api.c b/net/sched/cls_api.c
>> index adf5de1ff773..253b26f2eddd 100644
>> --- a/net/sched/cls_api.c
>> +++ b/net/sched/cls_api.c
>> @@ -1428,6 +1428,7 @@ int tcf_block_get_ext(struct tcf_block **p_block, struct Qdisc *q,
>>  		      struct tcf_block_ext_info *ei,
>>  		      struct netlink_ext_ack *extack)
>>  {
>> +	struct net_device *dev = qdisc_dev(q);
>>  	struct net *net = qdisc_net(q);
>>  	struct tcf_block *block = NULL;
>>  	int err;
>> @@ -1461,9 +1462,18 @@ int tcf_block_get_ext(struct tcf_block **p_block, struct Qdisc *q,
>>  	if (err)
>>  		goto err_block_offload_bind;
>>  
>> +	if (tcf_block_shared(block)) {
>> +		err = xa_insert(&block->ports, dev->ifindex, dev, GFP_KERNEL);
>> +		if (err) {
>> +			NL_SET_ERR_MSG(extack, "block dev insert failed");
>> +			goto err_dev_insert;
>> +		}
>> +	}
>
>While this patch fixes the original issue, it creates another one:
>
># ip link add name swp1 type dummy
># tc qdisc replace dev swp1 root handle 10: prio bands 8 priomap 7 6 5 4 3 2 1
># tc qdisc add dev swp1 parent 10:8 handle 108: red limit 1000000 min 200000 max 200001 probability 1.0 avpkt 8000 burst 38 qevent early_drop block 10
>RED: set bandwidth to 10Mbit
># tc qdisc add dev swp1 parent 10:7 handle 107: red limit 1000000 min 500000 max 500001 probability 1.0 avpkt 8000 burst 63 qevent early_drop block 10
>RED: set bandwidth to 10Mbit
>Error: block dev insert failed.

Feel free to text following patch. Since I believe it is not possible
to send fixes to net-next now, I will send it once net-next merges
into net.

net: sched: track device in tcf_block_get/put_ext() only for clsact binder types

Clsact/ingress qdisc is not the only one using shared block,
red is also using it. The device tracking was originally introduced
by commit 913b47d3424e ("net/sched: Introduce tc block netdev
tracking infra") for clsact/ingress only. Commit 94e2557d086a ("net:
sched: move block device tracking into tcf_block_get/put_ext()")
mistakenly enabled that for red as well.

Fix that by adding a check for the binder type being clsact when adding
device to the block->ports xarray.

Reported-by: Ido Schimmel <idosch@idosch.org>
Closes: https://lore.kernel.org/all/ZZ6JE0odnu1lLPtu@shredder/
Fixes: 94e2557d086a ("net: sched: move block device tracking into tcf_block_get/put_ext()")
Signed-off-by: Jiri Pirko <jiri@nvidia.com>
---
 net/sched/cls_api.c | 12 ++++++++++--
 1 file changed, 10 insertions(+), 2 deletions(-)

diff --git a/net/sched/cls_api.c b/net/sched/cls_api.c
index e3236a3169c3..92a12e3d0fe6 100644
--- a/net/sched/cls_api.c
+++ b/net/sched/cls_api.c
@@ -1424,6 +1424,14 @@ static void tcf_block_owner_del(struct tcf_block *block,
 	WARN_ON(1);
 }
 
+static bool tcf_block_tracks_dev(struct tcf_block *block,
+				 struct tcf_block_ext_info *ei)
+{
+	return tcf_block_shared(block) &&
+	       (ei->binder_type == FLOW_BLOCK_BINDER_TYPE_CLSACT_INGRESS ||
+		ei->binder_type == FLOW_BLOCK_BINDER_TYPE_CLSACT_EGRESS);
+}
+
 int tcf_block_get_ext(struct tcf_block **p_block, struct Qdisc *q,
 		      struct tcf_block_ext_info *ei,
 		      struct netlink_ext_ack *extack)
@@ -1462,7 +1470,7 @@ int tcf_block_get_ext(struct tcf_block **p_block, struct Qdisc *q,
 	if (err)
 		goto err_block_offload_bind;
 
-	if (tcf_block_shared(block)) {
+	if (tcf_block_tracks_dev(block, ei)) {
 		err = xa_insert(&block->ports, dev->ifindex, dev, GFP_KERNEL);
 		if (err) {
 			NL_SET_ERR_MSG(extack, "block dev insert failed");
@@ -1516,7 +1524,7 @@ void tcf_block_put_ext(struct tcf_block *block, struct Qdisc *q,
 
 	if (!block)
 		return;
-	if (tcf_block_shared(block))
+	if (tcf_block_tracks_dev(block, ei))
 		xa_erase(&block->ports, dev->ifindex);
 	tcf_chain0_head_change_cb_del(block, ei);
 	tcf_block_owner_del(block, q, ei->binder_type);
-- 
2.43.0