Skip to contentSkip to conversation

net: 9p: Fix possible null-pointer dereference in p9_cm_event_handler()

PATCH In review Tuo Li 3 participants 3 messages Aug 10, 2021

Subsystems (3)
lore
Share & tools
lore

Tuo Li

author
#0 lore raw

PATCH 1/1 net: 9p: Fix possible null-pointer dereference in p9_cm_event_handler()

The variable rdma is checked when event->event is equal to RDMA_CM_EVENT_DISCONNECTED:
if (rdma)

This indicates that it can be NULL. If so, a null-pointer dereference will occur when calling complete():
complete(&rdma->cm_done);

To fix this possible null-pointer dereference, calling complete() only when rdma is not NULL.

---
 net/9p/trans_rdma.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
net/9p/trans_rdma.c +2 -1
index af0a8a6cd3fd..fb3435dfd071 100644
@@ -285,7 +285,8 @@ p9_cm_event_handler(struct rdma_cm_id *id, struct rdma_cm_event *event)
 	default:
 		BUG();
 	}
-	complete(&rdma->cm_done);
+	if (rdma)
+		complete(&rdma->cm_done);
 	return 0;
 }
 
Signature · 1 line
2.25.1
#1 lore raw
Quoted 10 lines · Tuo Li

The variable rdma is checked when event->event is equal to RDMA_CM_EVENT_DISCONNECTED:
if (rdma)

This indicates that it can be NULL. If so, a null-pointer dereference will occur when calling complete():
complete(&rdma->cm_done);

To fix this possible null-pointer dereference, calling complete() only when rdma is not NULL.

You need to explain how is it possible and blindly set if () checks. I would say first "if (rdma)" is not needed, but don't know for sure.

Quoted 24 lines
---
 net/9p/trans_rdma.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
net/9p/trans_rdma.c +2 -1
index af0a8a6cd3fd..fb3435dfd071 100644
@@ -285,7 +285,8 @@ p9_cm_event_handler(struct rdma_cm_id *id, struct rdma_cm_event *event)
 	default:
 		BUG();
 	}
-	complete(&rdma->cm_done);
+	if (rdma)
+		complete(&rdma->cm_done);
 	return 0;
 }
 
Signature · 1 line
2.25.1
#2 lore raw

Leon Romanovsky wrote on Tue, Aug 10, 2021 at 04:55:42PM +0300:

Quoted 9 lines · Tuo Li
Quoted 8 lines

The variable rdma is checked when event->event is equal to RDMA_CM_EVENT_DISCONNECTED:
if (rdma)

This indicates that it can be NULL. If so, a null-pointer dereference will occur when calling complete():
complete(&rdma->cm_done);

Quoted 5 lines
Quoted 2 lines

To fix this possible null-pointer dereference, calling complete() only when rdma is not NULL.

You need to explain how is it possible and blindly set if () checks. I would say first "if (rdma)" is not needed, but don't know for sure.

Sounds like static analysis because there's a if (rdma) check in RDMA_CM_EVENT_DISCONNECTED above, so if that needed check then it will bug right afterwards

I'd tend to agree I don't think it's possible client->trans is null there (it's filled right after rdma_create_id which defines the handler, there might be a window where the callback is called before? But as I understand it shouldn't be called until we resolve address and connect then later disconnect)

So, I agree with Leon - unless you have a backtrace of a real bug let's remove the other 'if' if you want to cleanup something for your robot.

Signature · 1 line
Dominique