On Fri, Dec 21, 2018 at 11:40:36AM +0800, liuyonglong wrote:
> Hi, Dan Carpenter:
>
> On 2018/12/20 17:46, Dan Carpenter wrote:
> > Hello Yonglong Liu,
> >
> > The patch 6adafc356e20: "net: hns: Fix ping failed when use net
> > bridge and send multicast" from Dec 15, 2018, leads to the following
> > static checker warning:
> >
> > drivers/net/ethernet/hisilicon/hns/hns_dsaf_main.c:2822 set_promisc_tcam_enable()
> > error: testing array offset 'port' after use.
> >
> > drivers/net/ethernet/hisilicon/hns/hns_dsaf_main.c
> > 2786
> > 2787 mac_cb = dsaf_dev->mac_cb[port];
> > ^^^^^^^^^^^^
> > The ->mac_cb[] array has 6 elements.
> >
> > 2788 (void)hns_mac_get_inner_port_num(mac_cb, 0, &port_num);
> > 2789 tbl_tcam_ucast.tbl_ucast_out_port = port_num;
> > 2790
> > 2791 /* config uc vague table */
> > 2792 hns_dsaf_tcam_uc_cfg_vague(dsaf_dev, entry_index, &tbl_tcam_data_uc,
> > 2793 &tbl_tcam_mask_uc, &tbl_tcam_ucast);
> > 2794
> > 2795 /* update software entry */
> > 2796 soft_mac_entry = priv->soft_mac_tbl;
> > 2797 soft_mac_entry += entry_index;
> > 2798 soft_mac_entry->index = entry_index;
> > 2799 soft_mac_entry->tcam_key.high.val = mac_key.high.val;
> > 2800 soft_mac_entry->tcam_key.low.val = mac_key.low.val;
> > 2801 /* step back to the START for mc. */
> > 2802 soft_mac_entry = priv->soft_mac_tbl;
> > 2803
> > 2804 /* 2. set promisc multicast vague tcam entry. */
> > 2805 entry_index = hns_dsaf_find_empty_mac_entry_reverse(dsaf_dev);
> > 2806 if (entry_index == DSAF_INVALID_ENTRY_IDX) {
> > 2807 dev_err(dsaf_dev->dev,
> > 2808 "enable mc promisc failed (port:%#x)\n",
> > 2809 port);
> > 2810 return;
> > 2811 }
> > 2812
> > 2813 memset(&mask_entry, 0x0, sizeof(mask_entry));
> > 2814 memset(&mask_key, 0x0, sizeof(mask_key));
> > 2815 memset(&temp_key, 0x0, sizeof(temp_key));
> > 2816 mask_entry.addr[0] = 0x01;
> > 2817 hns_dsaf_set_mac_key(dsaf_dev, &mask_key, mask_entry.in_vlan_id,
> > 2818 port, mask_entry.addr);
> > 2819 tbl_tcam_mcast.tbl_mcast_item_vld = 1;
> > 2820 tbl_tcam_mcast.tbl_mcast_old_en = 0;
> > 2821
> > 2822 if (port < DSAF_SERVICE_NW_NUM) {
> > ^^^^^^^^^^^^^^^^^^^^^^^^^^
> >
> > 2823 mskid = port;
> > 2824 } else if (port >= DSAF_BASE_INNER_PORT_NUM) {
> > ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
> >
> > Here we are testing if "port" is over 127 but we are so toasted if
> > that condition is true.
> >
>
> set_promisc_tcam_enable is only called by hns_dsaf_set_promisc_tcam,
> and hns_dsaf_set_promisc_tcam is only called by hns_mac_set_promisc.
> hns_mac_set_promisc pass "mac_cb->mac_id" to the parameter "port", so the range of "port" is equals to "mac_cb->mac_id".
>
> When hns_mac_init, "mac_cb->mac_id" is limited to smaller than 6, so it's no problem.
If it's 6 then it is one element past the end of the array.
>
> So maybe this case "port >= DSAF_BASE_INNER_PORT_NUM" can never touch, need delete.
>
> I wanna know if u just test this function only, and pass 127 to the parameter "port"?
>
This is from static analysis not testing. Also I looked lower in the
function and just removing the conditions is not enough to fix the bugs.
2821
2822 if (port < DSAF_SERVICE_NW_NUM) {
2823 mskid = port;
2824 } else if (port >= DSAF_BASE_INNER_PORT_NUM) {
2825 mskid = port - DSAF_BASE_INNER_PORT_NUM + DSAF_SERVICE_NW_NUM;
2826 } else {
2827 dev_err(dsaf_dev->dev, "%s,pnum(%d)error,key(%#x:%#x)\n",
2828 dsaf_dev->ae_dev.name, port,
2829 mask_key.high.val, mask_key.low.val);
2830 return;
2831 }
2832
2833 dsaf_set_bit(tbl_tcam_mcast.tbl_mcast_port_msk[mskid / 32],
^^^^^^^^^^
Since mskid is alway 0-5, that means that mskid / 32 is always zero.
2834 mskid % 32, 1);
2835 memcpy(&temp_key, &mask_key, sizeof(mask_key));
2836 hns_dsaf_tcam_mc_cfg_vague(dsaf_dev, entry_index, &tbl_tcam_data_mc,
2837 (struct dsaf_tbl_tcam_data *)(&mask_key),
2838 &tbl_tcam_mcast);
2839
regards,
dan carpenter
