In zerocopy_fill_skb_from_iter(), if two copy operations are performed and the first one succeeds while the second one fails, it returns a failure but the count in iterator has already been decremented due to the first successful copy. This ultimately affects the local variable rest_len in virtio_transport_send_pkt_info(), causing the remaining count in rest_len to be greater than the actual iterator count. As a result, packet sending operations continue even when the iterator count is zero, which further leads to skb->len being 0 and triggers the warning reported by syzbot [1].
Therefore, if the zerocopy operation fails, we should revert the iterator to its original state. [1] 'send_pkt()' returns 0, but 4096 expected WARNING: net/vmw_vsock/virtio_transport_common.c:430 at virtio_transport_send_pkt_info+0xd1e/0xef0 net/vmw_vsock/virtio_transport_common.c:428, CPU#1: syz.0.17/5986 Call Trace: virtio_transport_stream_enqueue net/vmw_vsock/virtio_transport_common.c:1113 [inline] virtio_transport_seqpacket_enqueue+0x143/0x1c0 net/vmw_vsock/virtio_transport_common.c:841 vsock_connectible_sendmsg+0xabf/0x1040 net/vmw_vsock/af_vsock.c:2158 sock_sendmsg_nosec net/socket.c:727 [inline] __sock_sendmsg+0x21c/0x270 net/socket.c:746 Reported-by: [email protected] Closes: https://syzkaller.appspot.com/bug?extid=28e5f3d207b14bae122a Tested-by: [email protected] Signed-off-by: Edward Adam Davis <[email protected]> --- net/core/datagram.c | 6 ++++++ 1 file changed, 6 insertions(+) diff --git a/net/core/datagram.c b/net/core/datagram.c index c285c6465923..da10465cd8a4 100644 --- a/net/core/datagram.c +++ b/net/core/datagram.c @@ -748,10 +748,13 @@ int __zerocopy_sg_from_iter(struct msghdr *msg, struct sock *sk, size_t length, struct net_devmem_dmabuf_binding *binding) { + struct iov_iter_state state; unsigned long orig_size = skb->truesize; unsigned long truesize; int ret; + iov_iter_save_state(from, &state); + if (msg && msg->msg_ubuf && msg->sg_from_iter) ret = msg->sg_from_iter(skb, from, length); else if (binding) @@ -759,6 +762,9 @@ int __zerocopy_sg_from_iter(struct msghdr *msg, struct sock *sk, else ret = zerocopy_fill_skb_from_iter(skb, from, length); + if (ret) + iov_iter_restore(from, &state); + truesize = skb->truesize - orig_size; if (sk && sk->sk_type == SOCK_STREAM) { sk_wmem_queued_add(sk, truesize); -- 2.43.0
