Re: [PATCH net] virtio/vsock: fix sock refcnt bug on owner set failure

Re: [PATCH net] virtio/vsock: fix sock refcnt bug on owner set failure

[Stefano Garzarella]

On Wed, May 31, 2023 at 07:47:32PM +0000, Bobby Eshleman wrote:
>Previous to setting the owner the socket is found via
>vsock_find_connected_socket(), which returns sk after a call to
>sock_hold().
>
>If setting the owner fails, then sock_put() needs to be called.
>
>Fixes: f9d2b1e146e0 ("virtio/vsock: fix leaks due to missing skb owner")
>Signed-off-by: Bobby Eshleman <bobby.eshleman@bytedance.com>
>---
> net/vmw_vsock/virtio_transport_common.c | 1 +
> 1 file changed, 1 insertion(+)
>
>diff --git a/net/vmw_vsock/virtio_transport_common.c b/net/vmw_vsock/virtio_transport_common.c
>index b769fc258931..f01cd6adc5cb 100644
>--- a/net/vmw_vsock/virtio_transport_common.c
>+++ b/net/vmw_vsock/virtio_transport_common.c
>@@ -1343,6 +1343,7 @@ void virtio_transport_recv_pkt(struct virtio_transport *t,
>
> 	if (!skb_set_owner_sk_safe(skb, sk)) {
> 		WARN_ONCE(1, "receiving vsock socket has sk_refcnt == 0\n");
>+		sock_put(sk);

Did you have any warning, issue here?

IIUC skb_set_owner_sk_safe() can return false only if the ref counter
is 0, so calling a sock_put() on it should have no effect except to
produce a warning.

Thanks,
Stefano

Re: [PATCH net] virtio/vsock: fix sock refcnt bug on owner set failure

[Bobby Eshleman]

On Thu, Jun 01, 2023 at 09:58:47AM +0200, Stefano Garzarella wrote:
> On Wed, May 31, 2023 at 07:47:32PM +0000, Bobby Eshleman wrote:
> > Previous to setting the owner the socket is found via
> > vsock_find_connected_socket(), which returns sk after a call to
> > sock_hold().
> > 
> > If setting the owner fails, then sock_put() needs to be called.
> > 
> > Fixes: f9d2b1e146e0 ("virtio/vsock: fix leaks due to missing skb owner")
> > Signed-off-by: Bobby Eshleman <bobby.eshleman@bytedance.com>
> > ---
> > net/vmw_vsock/virtio_transport_common.c | 1 +
> > 1 file changed, 1 insertion(+)
> > 
> > diff --git a/net/vmw_vsock/virtio_transport_common.c b/net/vmw_vsock/virtio_transport_common.c
> > index b769fc258931..f01cd6adc5cb 100644
> > --- a/net/vmw_vsock/virtio_transport_common.c
> > +++ b/net/vmw_vsock/virtio_transport_common.c
> > @@ -1343,6 +1343,7 @@ void virtio_transport_recv_pkt(struct virtio_transport *t,
> > 
> > 	if (!skb_set_owner_sk_safe(skb, sk)) {
> > 		WARN_ONCE(1, "receiving vsock socket has sk_refcnt == 0\n");
> > +		sock_put(sk);
> 
> Did you have any warning, issue here?
> 
> IIUC skb_set_owner_sk_safe() can return false only if the ref counter
> is 0, so calling a sock_put() on it should have no effect except to
> produce a warning.
> 

Oh yeah, you're totally right. I did not recall how
skb_set_owner_sk_safe() worked internally and thought I'd introduced an
uneven hold/put count with that prior patch when reading through the
code again. I haven't seen any live issue, just misread the code.

Sorry about that, feel free to ignore this patch.

Best,
Bobby

Re: [PATCH net] virtio/vsock: fix sock refcnt bug on owner set failure

[Stefano Garzarella]

On Wed, May 31, 2023 at 03:35:30AM +0000, Bobby Eshleman wrote:
>On Thu, Jun 01, 2023 at 09:58:47AM +0200, Stefano Garzarella wrote:
>> On Wed, May 31, 2023 at 07:47:32PM +0000, Bobby Eshleman wrote:
>> > Previous to setting the owner the socket is found via
>> > vsock_find_connected_socket(), which returns sk after a call to
>> > sock_hold().
>> >
>> > If setting the owner fails, then sock_put() needs to be called.
>> >
>> > Fixes: f9d2b1e146e0 ("virtio/vsock: fix leaks due to missing skb owner")
>> > Signed-off-by: Bobby Eshleman <bobby.eshleman@bytedance.com>
>> > ---
>> > net/vmw_vsock/virtio_transport_common.c | 1 +
>> > 1 file changed, 1 insertion(+)
>> >
>> > diff --git a/net/vmw_vsock/virtio_transport_common.c b/net/vmw_vsock/virtio_transport_common.c
>> > index b769fc258931..f01cd6adc5cb 100644
>> > --- a/net/vmw_vsock/virtio_transport_common.c
>> > +++ b/net/vmw_vsock/virtio_transport_common.c
>> > @@ -1343,6 +1343,7 @@ void virtio_transport_recv_pkt(struct virtio_transport *t,
>> >
>> > 	if (!skb_set_owner_sk_safe(skb, sk)) {
>> > 		WARN_ONCE(1, "receiving vsock socket has sk_refcnt == 0\n");
>> > +		sock_put(sk);
>>
>> Did you have any warning, issue here?
>>
>> IIUC skb_set_owner_sk_safe() can return false only if the ref counter
>> is 0, so calling a sock_put() on it should have no effect except to
>> produce a warning.
>>
>
>Oh yeah, you're totally right. I did not recall how
>skb_set_owner_sk_safe() worked internally and thought I'd introduced an
>uneven hold/put count with that prior patch when reading through the
>code again. I haven't seen any live issue, just misread the code.
>
>Sorry about that, feel free to ignore this patch.

No problem ;-)

Maybe we should add a comment on it.

Thanks,
Stefano