Skip to contentSkip to conversation

udp_bpf: fix use-after-free in udp_bpf_recvmsg()

PATCH In review Deepanshu Kartikey 3 participants 4 messages Apr 2

Subsystems (4)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 udp_bpf: fix use-after-free in udp_bpf_recvmsg()

udp_bpf_recvmsg() calls sk_msg_recvmsg() without holding lock_sock(), unlike tcp_bpf_recvmsg() which properly acquires lock_sock() before calling __sk_msg_recvmsg(). This allows concurrent tasks to race inside sk_msg_recvmsg() on the same psock ingress queue, where one task can free msg_rx via kfree_sk_msg() while another task is still reading it via sk_msg_elem(), causing a slab-use-after-free.

Fix this by adding lock_sock()/release_sock() around the sk_msg_recvmsg() path in udp_bpf_recvmsg(), consistent with tcp_bpf_recvmsg(). Also make udp_msg_wait_data() release lock_sock() before sleeping and reacquire it after waking, so it can be called with the socket lock held, consistent with how tcp_msg_wait_data() uses sk_wait_event() which does the same internally.

Note: syzbot testing shows a separate pre-existing warning:
sk->sk_forward_alloc

  WARNING: net/ipv4/af_inet.c:162 inet_sock_destruct

This warning triggers from the idle CPU path (pv_native_safe_halt) and is unrelated to this patch. It appears to be a pre-existing memory accounting issue in the UDP sockmap path that requires separate investigation.

Reported-by [email protected]
Fixes 1f5be6b3b063 ("udp: Implement udp_bpf_recvmsg() for sockmap")
---
 net/ipv4/udp_bpf.c | 5 +++++
 1 file changed, 5 insertions(+)
net/ipv4/udp_bpf.c +5 -0
index 9f33b07b1481..f924b255cee6 100644
@@ -50,7 +50,9 @@ static int udp_msg_wait_data(struct sock *sk, struct sk_psock *psock,
 	sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
 	ret = udp_msg_has_data(sk, psock);
 	if (!ret) {
+		release_sock(sk);
 		wait_woken(&wait, TASK_INTERRUPTIBLE, timeo);
+		lock_sock(sk);
 		ret = udp_msg_has_data(sk, psock);
 	}
 	sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
@@ -79,6 +81,7 @@ static int udp_bpf_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
 		goto out;
 	}
 
+	lock_sock(sk);
 msg_bytes_ready:
 	copied = sk_msg_recvmsg(sk, psock, msg, len, flags);
 	if (!copied) {
@@ -90,12 +93,14 @@ static int udp_bpf_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
 		if (data) {
 			if (psock_has_data(psock))
 				goto msg_bytes_ready;
+			release_sock(sk);
 			ret = sk_udp_recvmsg(sk, msg, len, flags);
 			goto out;
 		}
 		copied = -EAGAIN;
 	}
 	ret = copied;
+	release_sock(sk);
 out:
 	sk_psock_put(sk, psock);
 	return ret;
Signature · 1 line
2.43.0
maintainer
#1 lore raw
Quoted 58 lines · Deepanshu Kartikey

udp_bpf_recvmsg() calls sk_msg_recvmsg() without holding lock_sock(), unlike tcp_bpf_recvmsg() which properly acquires lock_sock() before calling __sk_msg_recvmsg(). This allows concurrent tasks to race inside sk_msg_recvmsg() on the same psock ingress queue, where one task can free msg_rx via kfree_sk_msg() while another task is still reading it via sk_msg_elem(), causing a slab-use-after-free.

Fix this by adding lock_sock()/release_sock() around the sk_msg_recvmsg() path in udp_bpf_recvmsg(), consistent with tcp_bpf_recvmsg(). Also make udp_msg_wait_data() release lock_sock() before sleeping and reacquire it after waking, so it can be called with the socket lock held, consistent with how tcp_msg_wait_data() uses sk_wait_event() which does the same internally.

Note: syzbot testing shows a separate pre-existing warning:
sk->sk_forward_alloc

   WARNING: net/ipv4/af_inet.c:162 inet_sock_destruct

This warning triggers from the idle CPU path (pv_native_safe_halt) and is unrelated to this patch. It appears to be a pre-existing memory accounting issue in the UDP sockmap path that requires separate investigation.

Reported-by [email protected]
Fixes 1f5be6b3b063 ("udp: Implement udp_bpf_recvmsg() for sockmap")
---
  net/ipv4/udp_bpf.c | 5 +++++
  1 file changed, 5 insertions(+)
net/ipv4/udp_bpf.c +4 -0
index 9f33b07b1481..f924b255cee6 100644
@@ -50,7 +50,9 @@ static int udp_msg_wait_data(struct sock *sk, struct sk_psock *psock,
  	sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
  	ret = udp_msg_has_data(sk, psock);
  	if (!ret) {
+		release_sock(sk);
  		wait_woken(&wait, TASK_INTERRUPTIBLE, timeo);
+		lock_sock(sk);
  		ret = udp_msg_has_data(sk, psock);
  	}
  	sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
@@ -79,6 +81,7 @@ static int udp_bpf_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
  		goto out;
  	}
 
+	lock_sock(sk);
  msg_bytes_ready:
  	copied = sk_msg_recvmsg(sk, psock, msg, len, flags);
  	if (!copied) {
@@ -90,12 +93,14 @@ static int udp_bpf_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
  		if (data) {
  			if (psock_has_data(psock))
  				goto msg_bytes_ready;
+			release_sock(sk);

net/ipv4/udp_bpf.c static int udp_bpf_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, · line 106

Quoted 10 lines
net/ipv4/udp_bpf.c +1 -0
  			ret = sk_udp_recvmsg(sk, msg, len, flags);
  			goto out;
  		}
  		copied = -EAGAIN;
  	}
  	ret = copied;
+	release_sock(sk);
  out:
  	sk_psock_put(sk, psock);
  	return ret;

Kuniyuki  is already working on this. Please see the existing discussion.

https://lore.kernel.org/bpf/[email protected]/

#3 lore raw
In reply to Jiayuan Chen
Quoted 75 lines · Jiayuan Chen
Quoted 58 lines · Deepanshu Kartikey

udp_bpf_recvmsg() calls sk_msg_recvmsg() without holding lock_sock(), unlike tcp_bpf_recvmsg() which properly acquires lock_sock() before calling __sk_msg_recvmsg(). This allows concurrent tasks to race inside sk_msg_recvmsg() on the same psock ingress queue, where one task can free msg_rx via kfree_sk_msg() while another task is still reading it via sk_msg_elem(), causing a slab-use-after-free.

Fix this by adding lock_sock()/release_sock() around the sk_msg_recvmsg() path in udp_bpf_recvmsg(), consistent with tcp_bpf_recvmsg(). Also make udp_msg_wait_data() release lock_sock() before sleeping and reacquire it after waking, so it can be called with the socket lock held, consistent with how tcp_msg_wait_data() uses sk_wait_event() which does the same internally.

Note: syzbot testing shows a separate pre-existing warning:
sk->sk_forward_alloc

   WARNING: net/ipv4/af_inet.c:162 inet_sock_destruct

This warning triggers from the idle CPU path (pv_native_safe_halt) and is unrelated to this patch. It appears to be a pre-existing memory accounting issue in the UDP sockmap path that requires separate investigation.

Reported-by [email protected]
Fixes 1f5be6b3b063 ("udp: Implement udp_bpf_recvmsg() for sockmap")
---
  net/ipv4/udp_bpf.c | 5 +++++
  1 file changed, 5 insertions(+)
net/ipv4/udp_bpf.c +4 -0
index 9f33b07b1481..f924b255cee6 100644
@@ -50,7 +50,9 @@ static int udp_msg_wait_data(struct sock *sk, struct sk_psock *psock,
      sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
      ret = udp_msg_has_data(sk, psock);
      if (!ret) {
+             release_sock(sk);
              wait_woken(&wait, TASK_INTERRUPTIBLE, timeo);
+             lock_sock(sk);
              ret = udp_msg_has_data(sk, psock);
      }
      sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
@@ -79,6 +81,7 @@ static int udp_bpf_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
              goto out;
      }
 
+     lock_sock(sk);
  msg_bytes_ready:
      copied = sk_msg_recvmsg(sk, psock, msg, len, flags);
      if (!copied) {
@@ -90,12 +93,14 @@ static int udp_bpf_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
              if (data) {
                      if (psock_has_data(psock))
                              goto msg_bytes_ready;
+                     release_sock(sk);

net/ipv4/udp_bpf.c static int udp_bpf_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, · line 106

Quoted 10 lines
net/ipv4/udp_bpf.c +1 -0
                      ret = sk_udp_recvmsg(sk, msg, len, flags);
                      goto out;
              }
              copied = -EAGAIN;
      }
      ret = copied;
+     release_sock(sk);
  out:
      sk_psock_put(sk, psock);
      return ret;

Kuniyuki is already working on this. Please see the existing discussion.

Quoted 1 line

https://lore.kernel.org/bpf/[email protected]/

Oh I almost forgot about this.. will respin shortly. Thanks.