2024-02-10 12:56:54

by Rémi Denis-Courmont

Subject: [PATCH 2/2] phonet/pep: fix racy skb_queue_empty() use

From: Rémi Denis-Courmont <[email protected]>

The receive queues are protected by their respective spin-lock, not
the socket lock. This could lead to skb_peek() unexpectedly
returning NULL or a pointer to an already dequeued socket buffer.

Signed-off-by: Rémi Denis-Courmont <[email protected]>
---
net/phonet/pep.c | 41 ++++++++++++++++++++++++++++++++---------
1 file changed, 32 insertions(+), 9 deletions(-)

diff --git a/net/phonet/pep.c b/net/phonet/pep.c
index faba31f2eff2..3dd5f52bc1b5 100644
--- a/net/phonet/pep.c
+++ b/net/phonet/pep.c
@@ -917,6 +917,37 @@ static int pep_sock_enable(struct sock *sk, struct sockaddr *addr, int len)
return 0;
}

+static unsigned int pep_first_packet_length(struct sock *sk)
+{
+ struct pep_sock *pn = pep_sk(sk);
+ struct sk_buff_head *q;
+ struct sk_buff *skb;
+ unsigned int len = 0;
+ bool found = false;
+
+ if (sock_flag(sk, SOCK_URGINLINE)) {
+ q = &pn->ctrlreq_queue;
+ spin_lock_bh(&q->lock);
+ skb = skb_peek(q);
+ if (skb) {
+ len = skb->len;
+ found = true;
+ }
+ spin_unlock_bh(&q->lock);
+ }
+
+ if (likely(!found)) {
+ q = &sk->sk_receive_queue;
+ spin_lock_bh(&q->lock);
+ skb = skb_peek(q);
+ if (skb)
+ len = skb->len;
+ spin_unlock_bh(&q->lock);
+ }
+
+ return len;
+}
+
static int pep_ioctl(struct sock *sk, int cmd, int *karg)
{
struct pep_sock *pn = pep_sk(sk);
@@ -929,15 +960,7 @@ static int pep_ioctl(struct sock *sk, int cmd, int *karg)
break;
}

- lock_sock(sk);
- if (sock_flag(sk, SOCK_URGINLINE) &&
- !skb_queue_empty(&pn->ctrlreq_queue))
- *karg = skb_peek(&pn->ctrlreq_queue)->len;
- else if (!skb_queue_empty(&sk->sk_receive_queue))
- *karg = skb_peek(&sk->sk_receive_queue)->len;
- else
- *karg = 0;
- release_sock(sk);
+ *karg = pep_first_packet_length(sk);
ret = 0;
break;

--
2.43.0

2024-02-13 12:19:41

by Paolo Abeni

Subject: Re: [PATCH 2/2] phonet/pep: fix racy skb_queue_empty() use

On Sat, 2024-02-10 at 14:50 +0200, Rémi Denis-Courmont wrote:
> From: Rémi Denis-Courmont <[email protected]>
>
> The receive queues are protected by their respective spin-lock, not
> the socket lock. This could lead to skb_peek() unexpectedly
> returning NULL or a pointer to an already dequeued socket buffer.
>
> Signed-off-by: Rémi Denis-Courmont <[email protected]>

Please provide a suitable fixes tag here.

Also include the target tree into the subj prefix when you post the v2.

Thank,

Paolo