brcm80211: use endian annotation for pmk related structure
Hello Arend van Spriel,
The patch 40c8e95af02d: "brcm80211: use endian annotation for pmk related structure" from Oct 12, 2011, leads to the following static checker warning:
drivers/net/wireless/brcm80211/brcmfmac/cfg80211.c:2965 brcmf_cfg80211_set_pmksa() warn: can 'pmkid_len' be negative?
drivers/net/wireless/brcm80211/brcmfmac/cfg80211.c
2950 brcmf_cfg80211_set_pmksa(struct wiphy *wiphy, struct net_device *ndev,
2951 struct cfg80211_pmksa *pmksa)
2952 {
2953 struct brcmf_cfg80211_info *cfg = wiphy_to_cfg(wiphy); 2954 struct brcmf_if *ifp = netdev_priv(ndev); 2955 struct pmkid_list *pmkids = &cfg->pmk_list->pmkids; 2956 s32 err = 0; 2957 int i; 2958 int pmkid_len;
2959
2960 brcmf_dbg(TRACE, "Enter\n");
2961 if (!check_vif_up(ifp->vif))
2962 return -EIO;
2963
2964 pmkid_len = le32_to_cpu(pmkids->npmkid);
The thinking behind this check is that endian data is less trust worthy than cpu data. But probably this comes from the hardware so it's fine?
Anyway, let's assume pmkid_len = -1.
2965 for (i = 0; i < pmkid_len; i++)
2966 if (!memcmp(pmksa->bssid, pmkids->pmkid[i].BSSID, ETH_ALEN))
2967 break;
We don't enter this loop.
2968 if (i < WL_NUM_PMKIDS_MAX) {Zero is less than WL_NUM_PMKIDS_MAX.
2969 memcpy(pmkids->pmkid[i].BSSID, pmksa->bssid, ETH_ALEN);
2970 memcpy(pmkids->pmkid[i].PMKID, pmksa->pmkid, WLAN_PMKID_LEN);
2971 if (i == pmkid_len) {
^^^^^^^^^^^^^^This is false.
2972 pmkid_len++;
2973 pmkids->npmkid = cpu_to_le32(pmkid_len);
2974 }
2975 } else
2976 err = -EINVAL;
2977
2978 brcmf_dbg(CONN, "set_pmksa,IW_PMKSA_ADD - PMKID: %pM =\n",
2979 pmkids->pmkid[pmkid_len].BSSID);
^^^^^^^^^^^^^^^^^^^^^^^Underflow.
2980 for (i = 0; i < WLAN_PMKID_LEN; i++)
2981 brcmf_dbg(CONN, "%02x\n", pmkids->pmkid[pmkid_len].PMKID[i]);
^^^^^^^^^^^^^^^^^^^^^^^^Underflow.
2982 2983 err = brcmf_update_pmklist(ndev, cfg->pmk_list, err); 2984 2985 brcmf_dbg(TRACE, "Exit\n"); 2986 return err; 2987 }
regards,
dan carpenter
Quoted 61 lines · Dan Carpenter
Hello Arend van Spriel,
The patch
40c8e95af02d: "brcm80211: use endian annotation for pmk related structure" from Oct 12, 2011, leads to the following static checker warning:drivers/net/wireless/brcm80211/brcmfmac/cfg80211.c:2965 brcmf_cfg80211_set_pmksa() warn: can 'pmkid_len' be negative?
drivers/net/wireless/brcm80211/brcmfmac/cfg80211.c
2950 brcmf_cfg80211_set_pmksa(struct wiphy *wiphy, structnet_device*ndev,
2951 structcfg80211_pmksa*pmksa)2952 { 2953 struct brcmf_cfg80211_info *cfg = wiphy_to_cfg(wiphy); 2954 struct brcmf_if *ifp = netdev_priv(ndev); 2955 struct pmkid_list *pmkids =&cfg->pmk_list->pmkids; 2956 s32 err = 0; 2957 int i; 2958 int pmkid_len; 2959 2960 brcmf_dbg(TRACE, "Enter\n"); 2961 if (!check_vif_up(ifp->vif)) 2962 return -EIO; 2963 2964 pmkid_len = le32_to_cpu(pmkids->npmkid);The thinking behind this check is that endian data is less trust worthy than cpu data. But probably this comes from the hardware so it's fine?
Anyway, let's assume
pmkid_len= -1.2965 for (i = 0; i< pmkid_len; i++) 2966 if (!memcmp(pmksa->bssid, pmkids->pmkid[i].BSSID, ETH_ALEN)) 2967 break;We don't enter this loop.
2968 if (i< WL_NUM_PMKIDS_MAX) {Zero is less than
WL_NUM_PMKIDS_MAX.2969 memcpy(pmkids->pmkid[i].BSSID, pmksa->bssid, ETH_ALEN); 2970 memcpy(pmkids->pmkid[i].PMKID, pmksa->pmkid, WLAN_PMKID_LEN); 2971 if (i == pmkid_len) { ^^^^^^^^^^^^^^This is false.
2972 pmkid_len++; 2973 pmkids->npmkid = cpu_to_le32(pmkid_len); 2974 } 2975 } else 2976 err = -EINVAL; 2977 2978 brcmf_dbg(CONN, "set_pmksa,IW_PMKSA_ADD - PMKID: %pM =\n", 2979 pmkids->pmkid[pmkid_len].BSSID); ^^^^^^^^^^^^^^^^^^^^^^^Underflow.
2980 for (i = 0; i< WLAN_PMKID_LEN; i++) 2981 brcmf_dbg(CONN, "%02x\n", pmkids->pmkid[pmkid_len].PMKID[i]); ^^^^^^^^^^^^^^^^^^^^^^^^
Quoted 8 lines
Underflow.
2982 2983 err = brcmf_update_pmklist(ndev, cfg->pmk_list, err); 2984 2985 brcmf_dbg(TRACE, "Exit\n"); 2986 return err; 2987 }
Hi Dan,
Understood. Not sure what the motivation is to mistrust endian more. Simply because there could be conversion errors? Anyway, the main question is whether pmkid_len is always between 0 and WLAN_PMKID_LEN. As far as I know it is. We could 1) add additional checks here, 2) make pmkid_len of u32 type, or 3) just mention the (sure) assumption in a comment. I would prefer option 2) or 3).
Regards,
Arend
Quoted 1 line · Arend van Spriel
Understood. Not sure what the motivation is to mistrust endian more.
Endian data tends to come from suspicious places such as disk images, usb devices, and networks.
Quoted 5 lines
Simply because there could be conversion errors? Anyway, the main question is whether
pmkid_lenis always between 0 andWLAN_PMKID_LEN. As far as I know it is. We could 1) add additional checks here, 2) makepmkid_lenof u32 type, or 3) just mention the (sure) assumption in a comment. I would prefer option 2) or 3).
I would prefer 2. Static checker warnings are a pain.
regards,
dan carpenter
Quoted 6 lines · Dan Carpenter
Quoted 1 line · Arend van Spriel
Understood. Not sure what the motivation is to mistrust endian more.
Endian data tends to come from suspicious places such as disk images, usb devices, and networks.
Quoted 7 lines
Quoted 5 lines
Simply because there could be conversion errors? Anyway, the main question is whether
pmkid_lenis always between 0 andWLAN_PMKID_LEN. As far as I know it is. We could 1) add additional checks here, 2) makepmkid_lenof u32 type, or 3) just mention the (sure) assumption in a comment. I would prefer option 2) or 3).I would prefer 2. Static checker warnings are a pain.
Now who has been tinkering on static checkers ;-) Anyway, option 2) it is. Who will do the patch? :-p
Regards,
Arend
Quoted 13 lines · Arend van Spriel
Quoted 12 lines · Dan Carpenter
Quoted 1 line · Arend van Spriel
Understood. Not sure what the motivation is to mistrust endian more.
Endian data tends to come from suspicious places such as disk images, usb devices, and networks.
Quoted 5 lines
Simply because there could be conversion errors? Anyway, the main question is whether
pmkid_lenis always between 0 andWLAN_PMKID_LEN. As far as I know it is. We could 1) add additional checks here, 2) makepmkid_lenof u32 type, or 3) just mention the (sure) assumption in a comment. I would prefer option 2) or 3).
Quoted 4 lines
Quoted 1 line
I would prefer 2. Static checker warnings are a pain.
Now who has been tinkering on static checkers ;-) Anyway, option 2) it is. Who will do the patch? :-p
Actually, in the end making it unsigned doesn't make sense because we don't check for upper bound either.
Let's just forget about it.
regards,
dan carpenter