Re: [PATCH net v4 2/2] iavf: Fix out-of-bounds when setting channels on remove

[Ding Hui]

On 2023/5/4 15:57, Leon Romanovsky wrote:
> On Wed, May 03, 2023 at 12:22:00PM -0700, Chittim, Madhu wrote:
> > On 5/3/2023 9:29 AM, Leon Romanovsky wrote:
> > > On Wed, May 03, 2023 at 10:00:49PM +0800, Ding Hui wrote:
> > > > On 2023/5/3 4:24 下午, Leon Romanovsky wrote:
> > > > > On Wed, May 03, 2023 at 11:15:41AM +0800, Ding Hui wrote:
> > > >
> > > > > > If we detected removing is in processing, we can avoid unnecessary
> > > > > > waiting and return error faster.
> > > > > >
> > > > > > On the other hand in timeout handling, we should keep the original
> > > > > > num_active_queues and reset num_req_queues to 0.
> > > > > >
> > > > > > Fixes: 4e5e6b5d9d13 ("iavf: Fix return of set the new channel count")
> > > > > > Signed-off-by: Ding Hui <dinghui@xxxxxxxxxxxxxx>
> > > > > > Cc: Donglin Peng <pengdonglin@xxxxxxxxxxxxxx>
> > > > > > Cc: Huang Cun <huangcun@xxxxxxxxxxxxxx>
> > > > > > Reviewed-by: Simon Horman <simon.horman@xxxxxxxxxxxx>
> > > > > > Reviewed-by: Michal Kubiak <michal.kubiak@xxxxxxxxx>
> > > > > > ---
> > > > > > v3 to v4:
> > > > > >      - nothing changed
> > > > > >
> > > > > > v2 to v3:
> > > > > >      - fix review tag
> > > > > >
> > > > > > v1 to v2:
> > > > > >      - add reproduction script
> > > > > >
> > > > > > ---
> > > > > >     drivers/net/ethernet/intel/iavf/iavf_ethtool.c | 4 +++-
> > > > > >     1 file changed, 3 insertions(+), 1 deletion(-)
> > > > > >
> > > > > > diff --git a/drivers/net/ethernet/intel/iavf/iavf_ethtool.c b/drivers/net/ethernet/intel/iavf/iavf_ethtool.c
> > > > > > index 6f171d1d85b7..d8a3c0cfedd0 100644
> > > > > > --- a/drivers/net/ethernet/intel/iavf/iavf_ethtool.c
> > > > > > +++ b/drivers/net/ethernet/intel/iavf/iavf_ethtool.c
> > > > > > @@ -1857,13 +1857,15 @@ static int iavf_set_channels(struct net_device *netdev,
> > > > > >     	/* wait for the reset is done */
> > > > > >     	for (i = 0; i < IAVF_RESET_WAIT_COMPLETE_COUNT; i++) {
> > > > > >     		msleep(IAVF_RESET_WAIT_MS);
> > > > > > +		if (test_bit(__IAVF_IN_REMOVE_TASK, &adapter->crit_section))
> > > > > > +			return -EOPNOTSUPP;
> > > > >
> > > > > This makes no sense without locking as change to __IAVF_IN_REMOVE_TASK
> > > > > can happen any time.
> > > >
> > > > The state doesn't need to be that precise here, it is optimized only for
> > > > the fast path. During the lifecycle of the adapter, the __IAVF_IN_REMOVE_TASK
> > > > state will only be set and not cleared.
> > > >
> > > > If we didn't detect the "removing" state, we also can fallback to timeout
> > > > handling.
> > > >
> > > > So I don't think the locking is necessary here, what do the maintainers
> > > > at Intel think?
> > >
> > > I'm not Intel maintainer, but your change, explanation and the following
> > > line from your commit message aren't really aligned.
> > >
> > > [ 3510.400799] ==================================================================
> > > [ 3510.400820] BUG: KASAN: slab-out-of-bounds in iavf_free_all_tx_resources+0x156/0x160 [iavf]
> >
> > __IAVF_IN_REMOVE_TASK is being set only in iavf_remove() and the above
> > change is ok in terms of coming out of setting channels early enough while
> > remove is in progress.
>
> It is not, __IAVF_IN_REMOVE_TASK, set bit can be changed any time during
> iavf_set_channels() and if it is not, I would expect test_bit(..) placed
> at the beginning of iavf_set_channels() or even earlier.

Since we have a little dispute on __IAVF_IN_REMOVE_TASK, I'll remove the
test_bit() in v5, and remove Reviewed-by: tags of 2/2 to review again.

--
Thanks,
- Ding Hui