Re: [PATCH] xen/netback: handle empty rx queue in xenvif_rx_next_skb()

[Juergen Gross <jgross@xxxxxxxx>]

On 13.07.22 09:59, Jan Beulich wrote:
> On 13.07.2022 09:48, Juergen Gross wrote:
> > xenvif_rx_next_skb() is expecting the rx queue not being empty, but
> > in case the loop in xenvif_rx_action() is doing multiple iterations,
> > the availability of another skb in the rx queue is not being checked.
> >
> > This can lead to crashes:
> >
> > [40072.537261] BUG: unable to handle kernel NULL pointer dereference at 
> > 0000000000000080
> > [40072.537407] IP: xenvif_rx_skb+0x23/0x590 [xen_netback]
> > [40072.537534] PGD 0 P4D 0
> > [40072.537644] Oops: 0000 [#1] SMP NOPTI
> > [40072.537749] CPU: 0 PID: 12505 Comm: v1-c40247-q2-gu Not tainted 
> > 4.12.14-122.121-default #1 SLE12-SP5
> > [40072.537867] Hardware name: HP ProLiant DL580 Gen9/ProLiant DL580 Gen9, BIOS 
> > U17 11/23/2021
> > [40072.537999] task: ffff880433b38100 task.stack: ffffc90043d40000
> > [40072.538112] RIP: e030:xenvif_rx_skb+0x23/0x590 [xen_netback]
> > [40072.538217] RSP: e02b:ffffc90043d43de0 EFLAGS: 00010246
> > [40072.538319] RAX: 0000000000000000 RBX: ffffc90043cd7cd0 RCX: 00000000000000f7
> > [40072.538430] RDX: 0000000000000000 RSI: 0000000000000006 RDI: ffffc90043d43df8
> > [40072.538531] RBP: 000000000000003f R08: 000077ff80000000 R09: 0000000000000008
> > [40072.538644] R10: 0000000000007ff0 R11: 00000000000008f6 R12: ffffc90043ce2708
> > [40072.538745] R13: 0000000000000000 R14: ffffc90043d43ed0 R15: ffff88043ea748c0
> > [40072.538861] FS: 0000000000000000(0000) GS:ffff880484600000(0000) 
> > knlGS:0000000000000000
> > [40072.538988] CS: e033 DS: 0000 ES: 0000 CR0: 0000000080050033
> > [40072.539088] CR2: 0000000000000080 CR3: 0000000407ac8000 CR4: 0000000000040660
> > [40072.539211] Call Trace:
> > [40072.539319] xenvif_rx_action+0x71/0x90 [xen_netback]
> > [40072.539429] xenvif_kthread_guest_rx+0x14a/0x29c [xen_netback]
> >
> > Fix that by stopping the loop in case the rx queue becomes empty.
> >
> > Signed-off-by: Juergen Gross <jgross@xxxxxxxx>
> Reviewed-by: Jan Beulich <jbeulich@xxxxxxxx>
>
> Does this want a Fixes: tag and Cc: to stable@ (not the least since as per
> above the issue was noticed with 4.12.x)?
Hmm, I _think_ the issue was introduced with eb1723a29b9a. Do you agree?

> > --- a/drivers/net/xen-netback/rx.c
> > +++ b/drivers/net/xen-netback/rx.c
> > @@ -495,6 +495,7 @@ void xenvif_rx_action(struct xenvif_queue *queue)
> >         queue->rx_copy.completed = &completed_skbs;
> >   
> >   	while (xenvif_rx_ring_slots_available(queue) &&
> > +              !skb_queue_empty(&queue->rx_queue) &&
> >                work_done < RX_BATCH_SIZE) {
> >                 xenvif_rx_skb(queue);
> >                 work_done++;
> I have to admit that I find the title a little misleading - you don't
> deal with the issue _in_ xenvif_rx_next_skb(); you instead avoid
> entering the function in such a case.
I'm handling the issue to avoid "an empty rx queue in xenvif_rx_next_skb()".

I can rephrase it to "avoid entering xenvif_rx_next_skb() with an empty rx
queue".


Juergen

Attachment: OpenPGP_0xB0DE9DD628BF132F.asc
Description: OpenPGP public key

Attachment: OpenPGP_signature
Description: OpenPGP digital signature