In efx_probe_filters, the channel->rps_flow_id is freed in a
efx_for_each_channel marco when success equals to 0.
However, after the following call chain:
efx_probe_filters
|-> ef100_net_open
|-> ef100_net_stop
|-> efx_remove_filters
The channel->rps_flow_id is freed again in the efx_for_each_channel of
efx_remove_filters, triggering a double-free bug.
Fixes: a9dc3d5612ce ("sfc_ef100: RX filter table management and related gubbins")
Signed-off-by: Zhipeng Lu <[email protected]>
---
drivers/net/ethernet/sfc/rx_common.c | 4 +++-
1 file changed, 3 insertions(+), 1 deletion(-)
diff --git a/drivers/net/ethernet/sfc/rx_common.c b/drivers/net/ethernet/sfc/rx_common.c
index d2f35ee15eff..fac227d372db 100644
--- a/drivers/net/ethernet/sfc/rx_common.c
+++ b/drivers/net/ethernet/sfc/rx_common.c
@@ -823,8 +823,10 @@ int efx_probe_filters(struct efx_nic *efx)
}
if (!success) {
- efx_for_each_channel(channel, efx)
+ efx_for_each_channel(channel, efx) {
kfree(channel->rps_flow_id);
+ channel->rps_flow_id = NULL;
+ }
efx->type->filter_table_remove(efx);
rc = -ENOMEM;
goto out_unlock;
--
2.34.1
On Thu, Dec 14, 2023 at 11:22:46PM +0800, Zhipeng Lu wrote:
> In efx_probe_filters, the channel->rps_flow_id is freed in a
> efx_for_each_channel marco when success equals to 0.
> However, after the following call chain:
>
> efx_probe_filters
> |-> ef100_net_open
> |-> ef100_net_stop
> |-> efx_remove_filters
I think the call chain may be a bit more like:
ef100_net_open
|-> efx_probe_filters
|-> ef100_net_stop
|-> efx_remove_filters
>
> The channel->rps_flow_id is freed again in the efx_for_each_channel of
> efx_remove_filters, triggering a double-free bug.
>
> Fixes: a9dc3d5612ce ("sfc_ef100: RX filter table management and related gubbins")
> Signed-off-by: Zhipeng Lu <[email protected]>
The above nit not withstanding, I agree with your reasoning.
And that the problem was introduced in the cited commit.
Reviewed-by: Simon Horman <[email protected]>
...
Hi,
On Thu, 2023-12-14 at 23:22 +0800, Zhipeng Lu wrote:
> In efx_probe_filters, the channel->rps_flow_id is freed in a
> efx_for_each_channel marco when success equals to 0.
> However, after the following call chain:
>
> efx_probe_filters
> |-> ef100_net_open
> |-> ef100_net_stop
> |-> efx_remove_filters
>
> The channel->rps_flow_id is freed again in the efx_for_each_channel of
> efx_remove_filters, triggering a double-free bug.
>
> Fixes: a9dc3d5612ce ("sfc_ef100: RX filter table management and related gubbins")
> Signed-off-by: Zhipeng Lu <[email protected]>
The patch LGTM, but could you please update the commit message as per
Simon's suggestions make it more consistent? You can retain Simon's RB
tag.
Thanks!
Paolo
On 14/12/2023 15:22, Zhipeng Lu wrote:
> In efx_probe_filters, the channel->rps_flow_id is freed in a
> efx_for_each_channel marco when success equals to 0.
> However, after the following call chain:
>
> efx_probe_filters
> |-> ef100_net_open
> |-> ef100_net_stop
> |-> efx_remove_filters
>
> The channel->rps_flow_id is freed again in the efx_for_each_channel of
> efx_remove_filters, triggering a double-free bug.
>
> Fixes: a9dc3d5612ce ("sfc_ef100: RX filter table management and related gubbins")
> Signed-off-by: Zhipeng Lu <[email protected]>
Subject line should probably say [PATCH net] to specify the tree.
Modulo that, and Simon's correction to the commit message,
Reviewed-by: Edward Cree <[email protected]>
> On Thu, Dec 14, 2023 at 11:22:46PM +0800, Zhipeng Lu wrote:
> > In efx_probe_filters, the channel->rps_flow_id is freed in a
> > efx_for_each_channel marco when success equals to 0.
> > However, after the following call chain:
> >
> > efx_probe_filters
> > |-> ef100_net_open
> > |-> ef100_net_stop
> > |-> efx_remove_filters
>
> I think the call chain may be a bit more like:
>
> ef100_net_open
> |-> efx_probe_filters
> |-> ef100_net_stop
> |-> efx_remove_filters
>
> >
> > The channel->rps_flow_id is freed again in the efx_for_each_channel of
> > efx_remove_filters, triggering a double-free bug.
> >
> > Fixes: a9dc3d5612ce ("sfc_ef100: RX filter table management and related gubbins")
> > Signed-off-by: Zhipeng Lu <[email protected]>
>
> The above nit not withstanding, I agree with your reasoning.
> And that the problem was introduced in the cited commit.
>
> Reviewed-by: Simon Horman <[email protected]>
Sorry for the call-chain's problem, I was not familiar with it at that time.
Thanks for Simon's correction. Appreciate!
I'll soon send a v2 patch with the corrected call-chain and RB tags.