mwifiex: potential integer underflow in mwifiex_ret_wmm_get_status()
Subsystems (2)
PATCH 1/1 mwifiex: potential integer underflow in mwifiex_ret_wmm_get_status()
If "resp_len" gets set to negative then it counts as a high positive value.
---
I spotted this reviewing the int => bool changes, but I don't have the hardware and can't test it.
drivers/net/wireless/mwifiex/wmm.c +2 -0
@@ -772,6 +772,8 @@ int mwifiex_ret_wmm_get_status(struct mwifiex_private *priv,break ; } +if (resp_len < tlv_len +sizeof (tlv_hdr->header)) +break ; curr += (tlv_len +sizeof (tlv_hdr->header)); resp_len -= (tlv_len +sizeof (tlv_hdr->header)); }
Hi Dan,
Quoted 6 lines
If "
resp_len" gets set to negative then it counts as a high positive value.Signed-off-by Dan Carpenter <[email protected]>---
I spotted this reviewing the int => bool changes, but I don't have the hardware and can't test it.
Thanks for spotting this potential integer underflow problem.
I think we can change the 'resp_len' variable type to a signed integer to fix this issue.
Thanks,
Bing
Quoted 16 lines
drivers/net/wireless/mwifiex/wmm.c +2
-0@@ -772,6 +772,8 @@ int mwifiex_ret_wmm_get_status(structmwifiex_private *priv, break; } +if (resp_len < tlv_len +sizeof (tlv_hdr->header)) +break ; curr += (tlv_len +sizeof (tlv_hdr->header)); resp_len -= (tlv_len +sizeof (tlv_hdr->header));}
Quoted 9 lines · Bing Zhao
Hi Dan,
Quoted 6 lines
If "
resp_len" gets set to negative then it counts as a high positive value.Signed-off-by Dan Carpenter <[email protected]>---
I spotted this reviewing the int => bool changes, but I don't have the hardware and can't test it.
Quoted 4 lines
Thanks for spotting this potential integer underflow problem.
I think we can change the '
resp_len' variable type to a signed integer to fix this issue.
No, that doesn't work because the comparison against sizeof() get's promoted to size_t. In other words, negative values still count as large positive values.
regards,
dan carpenter
Hi Dan,
Thanks for your comments.
Quoted 6 lines
Quoted 2 lines
I think we can change the '
resp_len' variable type to a signed integer to fix this issue.No, that doesn't work because the comparison against
sizeof()get's promoted tosize_t. In other words, negative values still count as large positive values.
You are right. The negative value counts as a large positive number while comparing against sizeof().
I can add a "const int hdr_size" variable to store the value of sizeof(...) and compare resp_len to hdr_size. The "sizeof(...)" has been used multiple times in this function, so I think it's worth adding a variable for it.
drivers/net/wireless/mwifiex/wmm.c +5 -5
@@ -707,10 +707,10 @@ int mwifiex_ret_wmm_get_status(struct mwifiex_private *priv, const struct host_cmd_ds_command *resp) {u8 *curr = (u8 *) &resp->params.get_wmm_status; -uint16_t resp_len =le16_to_cpu (resp->size), tlv_len; +int resp_len =le16_to_cpu (resp->size), tlv_len;int valid = true; -struct mwifiex_ie_types_data *tlv_hdr; +const int hdr_size =sizeof (tlv_hdr->header);struct mwifiex_ie_types_wmm_queue_status *tlv_wmm_qstatus;struct ieee_types_wmm_parameter *wmm_param_ie =NULL ;struct mwifiex_wmm_ac_status *ac_status;
@@ -718,7 +718,7 @@ int mwifiex_ret_wmm_get_status(struct mwifiex_private *priv,dev_dbg (priv->adapter->dev,"info: WMM: WMM_GET_STATUS cmdresp received: %d\n" , resp_len); -while ((resp_len >=sizeof (tlv_hdr->header)) && valid) { +while (resp_len >= hdr_size && valid) { tlv_hdr = (struct mwifiex_ie_types_data *) curr; tlv_len =le16_to_cpu (tlv_hdr->header.len);
@@ -772,8 +772,8 @@ int mwifiex_ret_wmm_get_status(struct mwifiex_private *priv,break ; } - curr += (tlv_len +sizeof (tlv_hdr->header)); - resp_len -= (tlv_len +sizeof (tlv_hdr->header)); + curr += tlv_len + hdr_size; + resp_len -= tlv_len + hdr_size; }mwifiex_wmm_setup_queue_priorities (priv, wmm_param_ie);
Thanks,
Bing
Hi Dan,
Quoted 8 lines
Quoted 3 lines
No, that doesn't work because the comparison against
sizeof()get's promoted tosize_t. In other words, negative values still count as large positive values.You are right. The negative value counts as a large positive number while comparing against
sizeof(). I can add a "const inthdr_size" variable to store the value of sizeof(...) and compareresp_lentohdr_size. The "sizeof(...)" has been used multiple times in this function, so I think it's worth adding a variable for it.
Well, there is another problem here.
We might have accessed invalid memory while handling switch case in last iteration, because we didn't count tlv_len in "while(...)" condition check.
To fix this, Dan's change makes sense. But we should check the length before accessing the buffer.
@@ -722,6 +722,9 @@ int mwifiex_ret_wmm_get_status(struct mwifiex_private *priv,
tlv_hdr = (struct mwifiex_ie_types_data *) curr; tlv_len = le16_to_cpu(tlv_hdr->header.len);
+ if (resp_len < tlv_len + sizeof(tlv_hdr->header))
+ break;
+
switch (le16_to_cpu(tlv_hdr->header.type)) { case TLV_TYPE_WMMQSTATUS:
tlv_wmm_qstatus
Thanks,
Bing
There are several other similar functions as well. For example, the while loops in mwifiex_ret_tx_rate_cfg() and mwifiex_get_power_level().
regards,
dan carpenter
Hi Dan,
Quoted 2 lines
There are several other similar functions as well. For example, the while loops in
mwifiex_ret_tx_rate_cfg()andmwifiex_get_power_level().
Thanks for pointing it out. We will have them fixed.
Regards,
Bing
Quoted 9 lines · Bing Zhao
Hi Dan,
Quoted 2 lines
There are several other similar functions as well. For example, the while loops in
mwifiex_ret_tx_rate_cfg()andmwifiex_get_power_level().Thanks for pointing it out. We will have them fixed.
Regards,
Bing
OK, I'm confused by this thread...I'm dropping it.
Dan & Bing,
Please repost as a series of patches that I don't have to piece together from a series of replies...
John
Signature · 2 lines
John W. Linville Someday the world will need a hero, and you [email protected] might be all we have. Be ready.
Hi John,
Quoted 1 line
OK, I'm confused by this thread...I'm dropping it.
Yes, please drop it.
Quoted 5 lines
Dan & Bing,
Please repost as a series of patches that I don't have to piece together from a series of replies...
I'm working on some more patches that fix the similar issue. I will repost Dan's patch along with these patches in a series.
Thanks,
Bing