Skip to contentSkip to conversation

net/mlx5e: TC, fix a NULL pointer write in the peer flow error path

PATCH New Zhao Li 2 participants 2 messages Nov 18 – Nov 19

Subsystems (5)
lore
Share & tools
lore

Zhao Li

author
#0 lore raw

PATCH 1/1 net/mlx5e: TC, fix a NULL pointer write in the peer flow error path

The peer flow add path hands the duplicated flow over to the origin flow before it looks up the peer entry for the peer vhca_id. When that entry is missing, the function returns -ENODEV, and the caller then deletes the duplicated flow with list_del() on its peer node, a list head that was never initialised and never linked. list_del() on such a node writes through the NULL page.

Look the peer entry up before the duplicated flow is created and handed over, so that a missing entry is reported without leaving a duplicated flow behind. The entry cannot change in between: the peer loop in mlx5e_add_fdb_flow() holds the devcom component's read lock, and peer_flows entries are only stored and erased by the pair and unpair events, under its write lock.

Found by code review and not reproduced; compile-tested only.

AI assistance was used.

Fixes 9d608a40c2d4 ("net/mlx5e: TC, track peer flows in a vhca_id xarray")
---
 drivers/net/ethernet/mellanox/mlx5/core/en_tc.c | 7 ++++---
 1 file changed, 4 insertions(+), 3 deletions(-)
drivers/net/ethernet/mellanox/mlx5/core/en_tc.c +4 -3
index 708b7987ea4c..d41969117d30 100644
@@ -4630,6 +4630,10 @@ static int mlx5e_tc_add_fdb_peer_flow(struct flow_cls_offload *f,
 	else
 		in_mdev = priv->mdev;
 
+	dup_peer_flows = xa_load(&esw->offloads.peer_flows, peer_vhca_id);
+	if (!dup_peer_flows)
+		return -ENODEV;
+
 	flow_flags |= BIT(MLX5E_TC_FLOW_FLAG_PEER);
 	parse_attr = flow->attr->parse_attr;
 	peer_flow = __mlx5e_add_fdb_flow(peer_priv, f, flow_flags,
@@ -4643,9 +4647,6 @@ static int mlx5e_tc_add_fdb_peer_flow(struct flow_cls_offload *f,
 	peer_flow->peer_orig = flow;
 	list_add_tail(&peer_flow->peer_flows, &flow->peer_flows);
 	flow_flag_set(flow, DUP);
-	dup_peer_flows = xa_load(&esw->offloads.peer_flows, peer_vhca_id);
-	if (!dup_peer_flows)
-		return -ENODEV;
 	mutex_lock(&esw->offloads.peer_mutex);
 	list_add_tail(&peer_flow->peer, dup_peer_flows);
 	mutex_unlock(&esw->offloads.peer_mutex);

base-commit: 6b4b543fa18757297ac756a6fe6e64f506b7844a

Signature · 1 line
2.56.1
#1 lore raw
Thank you for your contribution! Sashiko AI review found no new issues in this patch, but noticed 1 related pre-existing issue it does not introduce: - [Critical] net/mlx5e: missing lock in…

Thank you for your contribution! Sashiko AI review found no new issues in this patch, but noticed 1 related pre-existing issue it does not introduce:
- [Critical] net/mlx5e: missing lock in mlx5e_tc_clean_fdb_peer_flows()
[1]
--

commit ee17d0ca384445e08f72eadef94d22d26656d903
Author: Zhao Li <[email protected]>

net/mlx5e: TC, fix a NULL pointer write in the peer flow error path

This commit moves the peer_flows xarray lookup before __mlx5e_add_fdb_flow() in mlx5e_tc_add_fdb_peer_flow() so a missing peer entry returns -ENODEV before creating and linking the duplicated flow.

Quoted 16 lines
drivers/net/ethernet/mellanox/mlx5/core/en_tc.c +4 -0
index 708b7987ea4cd..d41969117d303 100644
@@ -4630,6 +4630,10 @@ static int mlx5e_tc_add_fdb_peer_flow(struct flow_cls_offload *f,
 	else
 		in_mdev = priv->mdev;
 
+	dup_peer_flows = xa_load(&esw->offloads.peer_flows, peer_vhca_id);
+	if (!dup_peer_flows)
+		return -ENODEV;
+
 	flow_flags |= BIT(MLX5E_TC_FLOW_FLAG_PEER);
 	parse_attr = flow->attr->parse_attr;
 	peer_flow = __mlx5e_add_fdb_flow(peer_priv, f, flow_flags,

drivers/net/ethernet/mellanox/mlx5/core/en_tc.c static int mlx5e_tc_add_fdb_peer_flow(struct flow_cls_offload *f, · line 4652

Quoted 9 lines
drivers/net/ethernet/mellanox/mlx5/core/en_tc.c +0 -3
@@ -4643,9 +4647,6 @@ static int mlx5e_tc_add_fdb_peer_flow(struct flow_cls_offload *f,
 	peer_flow->peer_orig = flow;
 	list_add_tail(&peer_flow->peer_flows, &flow->peer_flows);
 	flow_flag_set(flow, DUP);
-	dup_peer_flows = xa_load(&esw->offloads.peer_flows, peer_vhca_id);
-	if (!dup_peer_flows)
-		return -ENODEV;
 	mutex_lock(&esw->offloads.peer_mutex);
 	list_add_tail(&peer_flow->peer, dup_peer_flows);
 	mutex_unlock(&esw->offloads.peer_mutex);

[Severity: Critical]
This problem wasn't introduced by this patch, but can mlx5e_tc_clean_fdb_peer_flows() corrupt the dup_peer_flows list or cause a use-after-free when racing with concurrent flow additions or deletions [1]?

mlx5e_tc_add_fdb_peer_flow() and mlx5e_tc_del_fdb_peer_flow() hold esw->offloads.peer_mutex while modifying dup_peer_flows, whereas mlx5e_tc_clean_fdb_peer_flows() walks the same list without taking peer_mutex:

drivers/net/ethernet/mellanox/mlx5/core/en_tc.c:mlx5e_tc_clean_fdb_peer_flows() {
	xa_for_each(&esw->offloads.peer_flows, index, dup_peer_flows)
		list_for_each_entry_safe(peer_flow, tmp_peer_flow,
					 dup_peer_flows, peer)
			mlx5e_tc_del_fdb_peer_flow(peer_flow);
}

list_for_each_entry_safe() only tolerates removals by the current iteration; a concurrent mlx5e_tc_add_fdb_peer_flow() or mlx5e_tc_del_fdb_peer_flow() modifying dup_peer_flows can invalidate tmp_peer_flow or corrupt the list pointers mid-traversal.

[1] https://sashiko.dev/#/bug/linux-184eb70a-9503-4852-b7bd-b8b8cdd3b641

Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=1