Skip to contentSkip to conversation

staging: rtl8723bs: Fix uninitialized variable

PATCH Superseded Wenli Looi 4 participants 15 messages Jun 6, 2021 – Jun 8, 2021 Staging

lore
Share & tools
lore

Wenli Looi

author
#0 lore raw

PATCH 1/1 staging: rtl8723bs: Fix uninitialized variable

Uninitialized struct with invalid pointer causes BUG and prevents access point from working. Access point works once I apply this patch.

https://forum.armbian.com/topic/14727-wifi-ap-kernel-bug-in-kernel-5444/
has more details.

---
 drivers/staging/rtl8723bs/os_dep/ioctl_cfg80211.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/staging/rtl8723bs/os_dep/ioctl_cfg80211.c +1 -1
index 2fb80b6eb..7308e1185 100644
@@ -2384,7 +2384,7 @@ void rtw_cfg80211_indicate_sta_assoc(struct adapter *padapter, u8 *pmgmt_frame,
 	DBG_871X(FUNC_ADPT_FMT"\n", FUNC_ADPT_ARG(padapter));
 
 	{
-		struct station_info sinfo;
+		struct station_info sinfo = {};
 		u8 ie_offset;
 		if (GetFrameSubType(pmgmt_frame) == WIFI_ASSOCREQ)
 			ie_offset = _ASOCREQ_IE_OFFSET_;
Signature · 1 line
2.25.1
#1 lore raw
Quoted 16 lines · Wenli Looi

Uninitialized struct with invalid pointer causes BUG and prevents access point from working. Access point works once I apply this patch.

https://forum.armbian.com/topic/14727-wifi-ap-kernel-bug-in-kernel-5444/
has more details.

---
 drivers/staging/rtl8723bs/os_dep/ioctl_cfg80211.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/staging/rtl8723bs/os_dep/ioctl_cfg80211.c +0 -0
index 2fb80b6eb..7308e1185 100644

drivers/staging/rtl8723bs/os_dep/ioctl_cfg80211.c void rtw_cfg80211_indicate_sta_assoc(struct adapter *padapter, u8 *pmgmt_frame, · line 2387

Quoted 5 lines
drivers/staging/rtl8723bs/os_dep/ioctl_cfg80211.c +1 -1
@@ -2384,7 +2384,7 @@ void rtw_cfg80211_indicate_sta_assoc(struct adapter *padapter, u8 *pmgmt_frame,
 	DBG_871X(FUNC_ADPT_FMT"\n", FUNC_ADPT_ARG(padapter));
 
 	{
-		struct station_info sinfo;
+		struct station_info sinfo = {};

What caused this bug to show up? Did it happen from some other commit?

Are you sure that all of the fields are being cleared properly here, what about any "holes" in the structure?

thanks,
greg k-h

author
#2 lore raw
Quoted 25 lines · Greg Kroah-Hartman
Quoted 16 lines · Wenli Looi

Uninitialized struct with invalid pointer causes BUG and prevents access point from working. Access point works once I apply this patch.

https://forum.armbian.com/topic/14727-wifi-ap-kernel-bug-in-kernel-5444/
has more details.

---
 drivers/staging/rtl8723bs/os_dep/ioctl_cfg80211.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/staging/rtl8723bs/os_dep/ioctl_cfg80211.c +0 -0
index 2fb80b6eb..7308e1185 100644

drivers/staging/rtl8723bs/os_dep/ioctl_cfg80211.c void rtw_cfg80211_indicate_sta_assoc(struct adapter *padapter, u8 *pmgmt_frame, · line 2387

Quoted 5 lines
drivers/staging/rtl8723bs/os_dep/ioctl_cfg80211.c +1 -1
@@ -2384,7 +2384,7 @@ void rtw_cfg80211_indicate_sta_assoc(struct adapter *padapter, u8 *pmgmt_frame,
      DBG_871X(FUNC_ADPT_FMT"\n", FUNC_ADPT_ARG(padapter));
 
      {
-             struct station_info sinfo;
+             struct station_info sinfo = {};

What caused this bug to show up? Did it happen from some other commit?

Quoted 5 lines

Are you sure that all of the fields are being cleared properly here, what about any "holes" in the structure?

thanks,
greg k-h

I believe this bug has been present since the driver was added to
staging: https://github.com/torvalds/linux/commit/554c0a3abf216c991c5ebddcdb2c08689ecd290b

It's probably not necessary to zero the entire struct, only sinfo->pertid, which causes problems with the code here:
https://github.com/torvalds/linux/blob/f5b6eb1e018203913dfefcf6fa988649ad11ad6e/net/wireless/nl80211.c#L5919

You can see the following proposed OpenWrt patch (700-wifi-8723bs-ap-bugfix.patch in
https://github.com/openwrt/openwrt/pull/4053/files) which sets sinfo.pertid = 0; instead of zeroing the struct.

Looking at similar code in a non-staging driver, we can see that the code there zeros the struct using kzalloc():
https://github.com/torvalds/linux/blob/f5b6eb1e018203913dfefcf6fa988649ad11ad6e/drivers/net/wireless/broadcom/brcm80211/brcmfmac/cfg80211.c#L6064

Do you think kzalloc() would be preferable?

Sorry, I'm not familiar with "holes" in the struct.

#3 lore raw
Quoted 10 lines · CEST Wenli Looi

On Sun, Jun 6, 2021 at 12:13 AM Greg Kroah-Hartman

Quoted 7 lines · <[email protected]> wrote:
Quoted 6 lines · Wenli Looi

Uninitialized struct with invalid pointer causes BUG and prevents access point from working. Access point works once I apply this patch.

https://forum.armbian.com/topic/14727-wifi-ap-kernel-bug-in-kernel-5444/
has more details.

Quoted 8 lines
Quoted 8 lines
Quoted 8 lines

---

 drivers/staging/rtl8723bs/os_dep/ioctl_cfg80211.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
(quoted diff) +0 -0

b/drivers/staging/rtl8723bs/os_dep/ioctl_cfg80211.c index 2fb80b6eb..

7308e1185

Quoted 4 lines
Quoted 4 lines

drivers/staging/rtl8723bs/os_dep/ioctl_cfg80211.c void rtw_cfg80211_indicate_sta_assoc(struct

Quoted 4 lines

100644

drivers/staging/rtl8723bs/os_dep/ioctl_cfg80211.c +0 -0
@@ -2384,7 +2384,7 @@ void rtw_cfg80211_indicate_sta_assoc(struct 

adapter *padapter,

Quoted 13 lines
Quoted 13 lines

drivers/staging/rtl8723bs/os_dep/ioctl_cfg80211.c void rtw_cfg80211_indicate_sta_assoc(struct adapter *padapter, u8 *pmgmt_frame, · line 2387

Quoted 7 lines

u8 *pmgmt_frame,> >

      DBG_871X(FUNC_ADPT_FMT"\n", FUNC_ADPT_ARG(padapter));
      
      {
drivers/staging/rtl8723bs/os_dep/ioctl_cfg80211.c +1 -1
-             struct station_info sinfo;
+             struct station_info sinfo = {};

What caused this bug to show up? Did it happen from some other commit?

Are you sure that all of the fields are being cleared properly here, what about any "holes" in the structure?

Quoted 6 lines
Quoted 2 lines

thanks,
greg k-h

I believe this bug has been present since the driver was added to staging:
https://github.com/torvalds/linux/commit/

554c0a3abf216c991c5ebddcdb2c08689ecd290b

Quoted 4 lines

It's probably not necessary to zero the entire struct, only sinfo->pertid, which causes problems with the code here:
https://github.com/torvalds/linux/blob/

f5b6eb1e018203913dfefcf6fa988649ad11ad6e/net/wire

Quoted 10 lines

less/nl80211.c#L5919

You can see the following proposed OpenWrt patch (700-wifi-8723bs-ap-bugfix.patch in
https://github.com/openwrt/openwrt/pull/4053/files) which sets sinfo.pertid = 0; instead of zeroing the struct.

Looking at similar code in a non-staging driver, we can see that the code there zeros the struct using kzalloc():
https://github.com/torvalds/linux/blob/

f5b6eb1e018203913dfefcf6fa988649ad11ad6e/drivers/

Quoted 4 lines

net/wireless/broadcom/brcm80211/brcmfmac/cfg80211.c#L6064

Do you think kzalloc() would be preferable?

You cannot use kzalloc there: 'sinfo' is instantiated automatically on the stack. The example you took had a pointer to the struct.

Fabio

Quoted 2 lines

Sorry, I'm not familiar with "holes" in the struct.

author
#4 lore raw
Quoted 60 lines · Fabio M. De Francesco
Quoted 10 lines · CEST Wenli Looi

On Sun, Jun 6, 2021 at 12:13 AM Greg Kroah-Hartman

Quoted 7 lines · <[email protected]> wrote:
Quoted 6 lines · Wenli Looi

Uninitialized struct with invalid pointer causes BUG and prevents access point from working. Access point works once I apply this patch.

https://forum.armbian.com/topic/14727-wifi-ap-kernel-bug-in-kernel-5444/
has more details.

Quoted 8 lines
Quoted 8 lines
Quoted 8 lines

---

 drivers/staging/rtl8723bs/os_dep/ioctl_cfg80211.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
(quoted diff) +0 -0

b/drivers/staging/rtl8723bs/os_dep/ioctl_cfg80211.c index 2fb80b6eb..

7308e1185

Quoted 4 lines
Quoted 4 lines

drivers/staging/rtl8723bs/os_dep/ioctl_cfg80211.c void rtw_cfg80211_indicate_sta_assoc(struct

Quoted 4 lines

100644

drivers/staging/rtl8723bs/os_dep/ioctl_cfg80211.c +0 -0
@@ -2384,7 +2384,7 @@ void rtw_cfg80211_indicate_sta_assoc(struct

adapter *padapter,

Quoted 13 lines
Quoted 13 lines

drivers/staging/rtl8723bs/os_dep/ioctl_cfg80211.c void rtw_cfg80211_indicate_sta_assoc(struct adapter *padapter, u8 *pmgmt_frame, · line 2387

Quoted 7 lines

u8 *pmgmt_frame,> >

      DBG_871X(FUNC_ADPT_FMT"\n", FUNC_ADPT_ARG(padapter));

      {
drivers/staging/rtl8723bs/os_dep/ioctl_cfg80211.c +1 -1
-             struct station_info sinfo;
+             struct station_info sinfo = {};

What caused this bug to show up? Did it happen from some other commit?

Are you sure that all of the fields are being cleared properly here, what about any "holes" in the structure?

Quoted 6 lines
Quoted 2 lines

thanks,
greg k-h

I believe this bug has been present since the driver was added to staging:
https://github.com/torvalds/linux/commit/

554c0a3abf216c991c5ebddcdb2c08689ecd290b

Quoted 4 lines

It's probably not necessary to zero the entire struct, only sinfo->pertid, which causes problems with the code here:
https://github.com/torvalds/linux/blob/

f5b6eb1e018203913dfefcf6fa988649ad11ad6e/net/wire

Quoted 10 lines

less/nl80211.c#L5919

You can see the following proposed OpenWrt patch (700-wifi-8723bs-ap-bugfix.patch in
https://github.com/openwrt/openwrt/pull/4053/files) which sets sinfo.pertid = 0; instead of zeroing the struct.

Looking at similar code in a non-staging driver, we can see that the code there zeros the struct using kzalloc():
https://github.com/torvalds/linux/blob/

Quoted 7 lines

f5b6eb1e018203913dfefcf6fa988649ad11ad6e/drivers/

Quoted 4 lines

net/wireless/broadcom/brcm80211/brcmfmac/cfg80211.c#L6064

Do you think kzalloc() would be preferable?

You cannot use kzalloc there: 'sinfo' is instantiated automatically on the stack. The example you took had a pointer to the struct.

The stack variable could be replaced with code like:

struct station_info *sinfo;if (!sinfo)
sinfo = kzalloc(sizeof(*sinfo), GFP_KERNEL);
...
cfg80211_new_sta(..., sinfo, ...);
kfree(sinfo);

which is what the linked code basically does. I'm not sure if this is preferred?

Quoted 8 lines

Fabio

Quoted 2 lines

Sorry, I'm not familiar with "holes" in the struct.

#5 lore raw
Quoted 9 lines · CEST Wenli Looi

On Sun, Jun 6, 2021 at 1:00 AM Fabio M. De Francesco

Quoted 6 lines · <[email protected]> wrote:
Quoted 5 lines · CEST Wenli Looi

On Sun, Jun 6, 2021 at 12:13 AM Greg Kroah-Hartman

Quoted 2 lines · <[email protected]> wrote:
Quoted 1 line · Wenli Looi

Uninitialized struct with invalid pointer causes BUG and prevents

access

Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines

point from working. Access point works once I apply this patch.

https://forum.armbian.com/topic/14727-wifi-ap-kernel-bug-in-kernel-5444/
has more details.

Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines

---

 drivers/staging/rtl8723bs/os_dep/ioctl_cfg80211.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
(quoted diff) +0 -0

b/drivers/staging/rtl8723bs/os_dep/ioctl_cfg80211.c index

2fb80b6eb..

Quoted 12 lines
Quoted 12 lines

7308e1185

Quoted 4 lines
Quoted 4 lines

drivers/staging/rtl8723bs/os_dep/ioctl_cfg80211.c void rtw_cfg80211_indicate_sta_assoc(struct

Quoted 4 lines

100644

drivers/staging/rtl8723bs/os_dep/ioctl_cfg80211.c +0 -0
@@ -2384,7 +2384,7 @@ void rtw_cfg80211_indicate_sta_assoc(struct

adapter *padapter,

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

u8 *pmgmt_frame,> >

Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines

drivers/staging/rtl8723bs/os_dep/ioctl_cfg80211.c void rtw_cfg80211_indicate_sta_assoc(struct adapter *padapter, u8 *pmgmt_frame, · line 2387

Quoted 6 lines
      DBG_871X(FUNC_ADPT_FMT"\n", FUNC_ADPT_ARG(padapter));
      
      {
drivers/staging/rtl8723bs/os_dep/ioctl_cfg80211.c +1 -1
-             struct station_info sinfo;
+             struct station_info sinfo = {};

What caused this bug to show up? Did it happen from some other

commit?

Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines

Are you sure that all of the fields are being cleared properly here, what about any "holes" in the structure?

thanks,
greg k-h

[CUT]

Quoted 10 lines
Quoted 5 lines
Quoted 2 lines

Do you think kzalloc() would be preferable?

You cannot use kzalloc there: 'sinfo' is instantiated automatically on the stack. The example you took had a pointer to the struct.

The stack variable could be replaced with code like:

struct station_info *sinfo;if (!sinfo)

Why that "if (!sinfo" before kzalloc?

Quoted 7 lines

sinfo = kzalloc(sizeof(*sinfo), GFP_KERNEL);
...
cfg80211_new_sta(..., sinfo, ...);
kfree(sinfo);

which is what the linked code basically does. I'm not sure if this is

preferred?

Quoted 1 line

I don't know that code, but I think that:

(1) It is generally preferred that big structures should be allocated
dinamically: kernel stack is a scarce resource.
(2) Passing address of variables from the stack to other functions is generally a good source of troubles.

I think you are closer to the proper solution.

Fabio

Quoted 3 lines
Quoted 3 lines
Quoted 3 lines

Sorry, I'm not familiar with "holes" in the struct.

author
#6 lore raw

PATCH v2 1/1 staging: rtl8723bs: Fix uninitialized variable

Uninitialized struct with invalid pointer causes BUG and prevents access point from working. Access point works once I apply this patch.

This problem seems to have been present from the time the driver was added to staging. Most users probably do not use access point so they will not encounter this bug.

https://forum.armbian.com/topic/14727-wifi-ap-kernel-bug-in-kernel-5444/
has more details.

kzalloc() seems to be what other drivers are doing in the same situation of creating struct station_info and calling cfg80211_new_sta. In particular, other drivers like ath6kl and mwifiex will silently return when kzalloc fails, so this seems like the right behavior. (mwifiex returns -ENOMEM from the place kzalloc is called, but if you follow the chain of calls, the return value is ultimately ignored)

Links to same situation in other drivers:
https://github.com/torvalds/linux/blob/f5b6eb1e018203913dfefcf6fa988649ad11ad6e/drivers/net/wireless/ath/ath6kl/main.c#L488
https://github.com/torvalds/linux/blob/f5b6eb1e018203913dfefcf6fa988649ad11ad6e/drivers/net/wireless/marvell/mwifiex/uap_event.c#L120

---

v1 -> v2: Switched from large stack variable to kzalloc

---
 .../staging/rtl8723bs/os_dep/ioctl_cfg80211.c | 28 ++++++++++---------
 1 file changed, 15 insertions(+), 13 deletions(-)
drivers/staging/rtl8723bs/os_dep/ioctl_cfg80211.c +15 -13
index 9a6e47877..3bc498558 100644
@@ -2082,20 +2082,22 @@ static int cfg80211_rtw_flush_pmksa(struct wiphy *wiphy,
 void rtw_cfg80211_indicate_sta_assoc(struct adapter *padapter, u8 *pmgmt_frame, uint frame_len)
 {
 	struct net_device *ndev = padapter->pnetdev;
+	struct station_info *sinfo;
+	u8 ie_offset;
 
-	{
-		struct station_info sinfo;
-		u8 ie_offset;
-		if (GetFrameSubType(pmgmt_frame) == WIFI_ASSOCREQ)
-			ie_offset = _ASOCREQ_IE_OFFSET_;
-		else /*  WIFI_REASSOCREQ */
-			ie_offset = _REASOCREQ_IE_OFFSET_;
-
-		sinfo.filled = 0;
-		sinfo.assoc_req_ies = pmgmt_frame + WLAN_HDR_A3_LEN + ie_offset;
-		sinfo.assoc_req_ies_len = frame_len - WLAN_HDR_A3_LEN - ie_offset;
-		cfg80211_new_sta(ndev, GetAddr2Ptr(pmgmt_frame), &sinfo, GFP_ATOMIC);
-	}
+	sinfo = kzalloc(sizeof(*sinfo), GFP_KERNEL);
+	if (!sinfo)
+		return;
+
+	if (GetFrameSubType(pmgmt_frame) == WIFI_ASSOCREQ)
+		ie_offset = _ASOCREQ_IE_OFFSET_;
+	else /*  WIFI_REASSOCREQ */
+		ie_offset = _REASOCREQ_IE_OFFSET_;
+
+	sinfo->assoc_req_ies = pmgmt_frame + WLAN_HDR_A3_LEN + ie_offset;
+	sinfo->assoc_req_ies_len = frame_len - WLAN_HDR_A3_LEN - ie_offset;
+	cfg80211_new_sta(ndev, GetAddr2Ptr(pmgmt_frame), sinfo, GFP_ATOMIC);
+	kfree(sinfo);
 }
 
 void rtw_cfg80211_indicate_sta_disassoc(struct adapter *padapter, unsigned char *da, unsigned short reason)
Signature · 1 line
2.25.1
#7 lore raw
Quoted 17 lines · Wenli Looi

Uninitialized struct with invalid pointer causes BUG and prevents access point from working. Access point works once I apply this patch.

This problem seems to have been present from the time the driver was added to staging. Most users probably do not use access point so they will not encounter this bug.

https://forum.armbian.com/topic/14727-wifi-ap-kernel-bug-in-kernel-5444/
has more details.

kzalloc() seems to be what other drivers are doing in the same situation of creating struct station_info and calling cfg80211_new_sta. In particular, other drivers like ath6kl and mwifiex will silently return when kzalloc fails, so this seems like the right behavior. (mwifiex returns -ENOMEM from the place kzalloc is called, but if you follow the chain of calls, the return value is ultimately ignored)

Quoted 8 lines

Links to same situation in other drivers:
https://github.com/torvalds/linux/blob/f5b6eb1e018203913dfefcf6fa988649ad11ad6e/drivers/net/wireless/ath/ath6kl/main.c#L488
https://github.com/torvalds/linux/blob/f5b6eb1e018203913dfefcf6fa988649ad11ad6e/drivers/net/wireless/marvell/mwifiex/uap_event.c#L120

---

v1 -> v2: Switched from large stack variable to kzalloc

Nah, v1 was better, it just needs an updated commit message. See my other email for more details.

regards,
dan carpenter

#8 lore raw
Quoted 22 lines · Dan Carpenter
Quoted 21 lines · Wenli Looi

Uninitialized struct with invalid pointer causes BUG and prevents access point from working. Access point works once I apply this patch.

This problem seems to have been present from the time the driver was added to staging. Most users probably do not use access point so they will not encounter this bug.

https://forum.armbian.com/topic/14727-wifi-ap-kernel-bug-in-kernel-5444/
has more details.

kzalloc() seems to be what other drivers are doing in the same situation of creating struct station_info and calling cfg80211_new_sta. In particular, other drivers like ath6kl and mwifiex will silently return when kzalloc fails, so this seems like the right behavior. (mwifiex returns -ENOMEM from the place kzalloc is called, but if you follow the chain of calls, the return value is ultimately ignored)

Links to same situation in other drivers:
https://github.com/torvalds/linux/blob/f5b6eb1e018203913dfefcf6fa988649ad11ad6e/drivers/net/wireless/ath/ath6kl/main.c#L488
https://github.com/torvalds/linux/blob/f5b6eb1e018203913dfefcf6fa988649ad11ad6e/drivers/net/wireless/marvell/mwifiex/uap_event.c#L120

Quoted 8 lines
Quoted 4 lines

---

v1 -> v2: Switched from large stack variable to kzalloc

Nah, v1 was better, it just needs an updated commit message. See my other email for more details.

I read this again and I thought I should provide some more information.

This sinfo struct used to be huge and that's why people used to allocate it if kzalloc() but now it's only 224 bytes so it's okay to put it on the stack.

And the problem was never whether the variable was on the stack vs on the heap so changing that wasn't part of the "a patch should do one thing." If you want to change it to kzalloc you have to do that in a separate patch (don't do that).

And you were reading Greg's questions as saying the patch was wrong, but actually they were just questions.

regards,
dan carpenter

author
#9 lore raw

I will resend the first patch. Thanks for your insightful comments. I was wondering why every other driver seemed to be allocating "struct station_info" using kzalloc()

Quoted 54 lines · Dan Carpenter

[△EXTERNAL]

Quoted 22 lines · Dan Carpenter
Quoted 21 lines · Wenli Looi

Uninitialized struct with invalid pointer causes BUG and prevents access point from working. Access point works once I apply this patch.

This problem seems to have been present from the time the driver was added to staging. Most users probably do not use access point so they will not encounter this bug.

https://forum.armbian.com/topic/14727-wifi-ap-kernel-bug-in-kernel-5444/
has more details.

kzalloc() seems to be what other drivers are doing in the same situation of creating struct station_info and calling cfg80211_new_sta. In particular, other drivers like ath6kl and mwifiex will silently return when kzalloc fails, so this seems like the right behavior. (mwifiex returns -ENOMEM from the place kzalloc is called, but if you follow the chain of calls, the return value is ultimately ignored)

Links to same situation in other drivers:
https://github.com/torvalds/linux/blob/f5b6eb1e018203913dfefcf6fa988649ad11ad6e/drivers/net/wireless/ath/ath6kl/main.c#L488
https://github.com/torvalds/linux/blob/f5b6eb1e018203913dfefcf6fa988649ad11ad6e/drivers/net/wireless/marvell/mwifiex/uap_event.c#L120

Quoted 8 lines
Quoted 4 lines

---

v1 -> v2: Switched from large stack variable to kzalloc

Nah, v1 was better, it just needs an updated commit message. See my other email for more details.

I read this again and I thought I should provide some more information.

This sinfo struct used to be huge and that's why people used to allocate it if kzalloc() but now it's only 224 bytes so it's okay to put it on the stack.

And the problem was never whether the variable was on the stack vs on the heap so changing that wasn't part of the "a patch should do one thing." If you want to change it to kzalloc you have to do that in a separate patch (don't do that).

And you were reading Greg's questions as saying the patch was wrong, but actually they were just questions.

regards,
dan carpenter

#10 lore raw
In reply to Wenli Looi
Quoted 8 lines · Wenli Looi

Uninitialized struct with invalid pointer causes BUG and prevents access point from working. Access point works once I apply this patch.

https://forum.armbian.com/topic/14727-wifi-ap-kernel-bug-in-kernel-5444/
has more details.

---

This patch is correct but the commit message needs to be updated. Your version 2 patch is not correct.

We don't like "follow this link for all the information" type commit messages. Clicking on a link is annoying and links die after five years. The link can be there but the main information needs to be in the commit message. Generally it's good to put the stack trace in the commit so that people can search for it.

As Greg pointed out, you need to add a Fixes tag. So far as I can see it's ->pertid and ->generation which are not initialized and the bugs were introduced in two different commits so you need two Fixes tags.

Fixes 8689c051a201 ("cfg80211: dynamically allocate per-tid stats for station info")
Fixes f5ea9120be2e ("nl80211: add generation number to all dumps")

Adding a Fixes tag will mean the correct people are CC'd in the patch and can review the fix.

Greg asked about struct holes and the answer is "= {}" will zero out struct holes but it's not important in this case. The "= {}" is a GCC extension for zeroing structs and it's not part of the C standard. The struct has a kernel pointer in it so we had better not be shairing it to user space.

Here is a better commit message. Please resend the commit with something like the following.

staging: rtl8723bs: Fix uninitialized variables

The sinfo.pertid and sinfo.generation variables are not initialized and it causes a crash when we use this as a wireless access point.

[  456.873025] ------------[ cut here ]------------
[  456.878198] kernel BUG at mm/slub.c:3968!
[  456.882680] Internal error: Oops - BUG: 0 [#1] PREEMPT SMP ARM

[ snip ]

[  457.271004] Backtrace: 
[  457.273733] [<c02b7ee4>] (kfree) from [<c0e2a470>] (nl80211_send_station+0x954/0xfc4)
[  457.282481]  r9:eccca0c0 r8:e8edfec0 r7:00000000 r6:00000011 r5:e80a9480 r4:e8edfe00
[  457.291132] [<c0e29b1c>] (nl80211_send_station) from [<c0e2b18c>] (cfg80211_new_sta+0x90/0x1cc)
[  457.300850]  r10:e80a9480 r9:e8edfe00 r8:ea678cca r7:00000a20 r6:00000000 r5:ec46d000
[  457.309586]  r4:ec46d9e0
[  457.312433] [<c0e2b0fc>] (cfg80211_new_sta) from [<bf086684>] (rtw_cfg80211_indicate_sta_assoc+0x80/0x9c [r8723bs])
[  457.324095]  r10:00009930 r9:e85b9d80 r8:bf091050 r7:00000000 r6:00000000 r5:0000001c
[  457.332831]  r4:c1606788
[  457.335692] [<bf086604>] (rtw_cfg80211_indicate_sta_assoc [r8723bs]) from [<bf03df38>] (rtw_stassoc_event_callback+0x1c8/0x1d4 [r8723bs])
[  457.349489]  r7:ea678cc0 r6:000000a1 r5:f1225f84 r4:f086b000
[  457.355845] [<bf03dd70>] (rtw_stassoc_event_callback [r8723bs]) from [<bf048e4c>] (mlme_evt_hdl+0x8c/0xb4 [r8723bs])
[  457.367601]  r7:c1604900 r6:f086c4b8 r5:00000000 r4:f086c000
[  457.373959] [<bf048dc0>] (mlme_evt_hdl [r8723bs]) from [<bf03693c>] (rtw_cmd_thread+0x198/0x3d8 [r8723bs])
[  457.384744]  r5:f086e000 r4:f086c000
[  457.388754] [<bf0367a4>] (rtw_cmd_thread [r8723bs]) from [<c014a214>] (kthread+0x170/0x174)
[  457.398083]  r10:ed7a57e8 r9:bf0367a4 r8:f086b000 r7:e8ede000 r6:00000000 r5:e9975200
[  457.406828]  r4:e8369900
[  457.409653] [<c014a0a4>] (kthread) from [<c01010e8>] (ret_from_fork+0x14/0x2c)
[  457.417718] Exception stack(0xe8edffb0 to 0xe8edfff8)
[  457.423356] ffa0:                                     00000000 00000000 00000000 00000000
[  457.432492] ffc0: 00000000 00000000 00000000 00000000 00000000 00000000 00000000 00000000
[  457.441618] ffe0: 00000000 00000000 00000000 00000000 00000013 00000000
[  457.449006]  r10:00000000 r9:00000000 r8:00000000 r7:00000000 r6:00000000 r5:c014a0a4
[  457.457750]  r4:e9975200
[  457.460574] Code: 1a000003 e5953004 e3130001 1a000000 (e7f001f2) 
[  457.467381] ---[ end trace 4acbc8c15e9e6aa7 ]---
Fixes 8689c051a201 ("cfg80211: dynamically allocate per-tid stats for station info")
Fixes f5ea9120be2e ("nl80211: add generation number to all dumps")

Signed-off-by:

regards,
dan carpenter

#11 lore raw
Quoted 5 lines · Dan Carpenter

Greg asked about struct holes and the answer is "= {}" will zero out struct holes but it's not important in this case. The "= {}" is a GCC extension for zeroing structs and it's not part of the C standard. The struct has a kernel pointer in it so we had better not be shairing it to user space.

I thought we proved that "= {}" will _NOT_ zero out holes in structures. Or did we really prove that? I can't remember now, do you?

thanks,

greg k-h

#12 lore raw
Quoted 10 lines · Greg Kroah-Hartman
Quoted 5 lines · Dan Carpenter

Greg asked about struct holes and the answer is "= {}" will zero out struct holes but it's not important in this case. The "= {}" is a GCC extension for zeroing structs and it's not part of the C standard. The struct has a kernel pointer in it so we had better not be shairing it to user space.

I thought we proved that "= {}" will _NOT_ zero out holes in structures. Or did we really prove that? I can't remember now, do you?

Assigning a struct to a struct will not initialize the struct holes.

	struct foo foo = *p;

We worried about = {} and people looked at the C standard. The standard is not clear. But then people said that = {} is a GCC extension and will initialize the struct holes.

The other thing is that in GCC they had intended "= {0};" to work exactly the same as "= {};" and initialize the holes but there was a version which had a bug and didn't. :P

regards,
dan carpenter

author
#13 lore raw

PATCH 1/1 staging: rtl8723bs: Fix uninitialized variables

In reply to Dan Carpenter

The sinfo.pertid and sinfo.generation variables are not initialized and it causes a crash when we use this as a wireless access point.

[  456.873025] ------------[ cut here ]------------
[  456.878198] kernel BUG at mm/slub.c:3968!
[  456.882680] Internal error: Oops - BUG: 0 [#1] PREEMPT SMP ARM

[ snip ]

[  457.271004] Backtrace:
[  457.273733] [<c02b7ee4>] (kfree) from [<c0e2a470>] (nl80211_send_station+0x954/0xfc4)
[  457.282481]  r9:eccca0c0 r8:e8edfec0 r7:00000000 r6:00000011 r5:e80a9480 r4:e8edfe00
[  457.291132] [<c0e29b1c>] (nl80211_send_station) from [<c0e2b18c>] (cfg80211_new_sta+0x90/0x1cc)
[  457.300850]  r10:e80a9480 r9:e8edfe00 r8:ea678cca r7:00000a20 r6:00000000 r5:ec46d000
[  457.309586]  r4:ec46d9e0
[  457.312433] [<c0e2b0fc>] (cfg80211_new_sta) from [<bf086684>] (rtw_cfg80211_indicate_sta_assoc+0x80/0x9c [r8723bs])
[  457.324095]  r10:00009930 r9:e85b9d80 r8:bf091050 r7:00000000 r6:00000000 r5:0000001c
[  457.332831]  r4:c1606788
[  457.335692] [<bf086604>] (rtw_cfg80211_indicate_sta_assoc [r8723bs]) from [<bf03df38>] (rtw_stassoc_event_callback+0x1c8/0x1d4 [r8723bs])
[  457.349489]  r7:ea678cc0 r6:000000a1 r5:f1225f84 r4:f086b000
[  457.355845] [<bf03dd70>] (rtw_stassoc_event_callback [r8723bs]) from [<bf048e4c>] (mlme_evt_hdl+0x8c/0xb4 [r8723bs])
[  457.367601]  r7:c1604900 r6:f086c4b8 r5:00000000 r4:f086c000
[  457.373959] [<bf048dc0>] (mlme_evt_hdl [r8723bs]) from [<bf03693c>] (rtw_cmd_thread+0x198/0x3d8 [r8723bs])
[  457.384744]  r5:f086e000 r4:f086c000
[  457.388754] [<bf0367a4>] (rtw_cmd_thread [r8723bs]) from [<c014a214>] (kthread+0x170/0x174)
[  457.398083]  r10:ed7a57e8 r9:bf0367a4 r8:f086b000 r7:e8ede000 r6:00000000 r5:e9975200
[  457.406828]  r4:e8369900
[  457.409653] [<c014a0a4>] (kthread) from [<c01010e8>] (ret_from_fork+0x14/0x2c)
[  457.417718] Exception stack(0xe8edffb0 to 0xe8edfff8)
[  457.423356] ffa0:                                     00000000 00000000 00000000 00000000
[  457.432492] ffc0: 00000000 00000000 00000000 00000000 00000000 00000000 00000000 00000000
[  457.441618] ffe0: 00000000 00000000 00000000 00000000 00000013 00000000
[  457.449006]  r10:00000000 r9:00000000 r8:00000000 r7:00000000 r6:00000000 r5:c014a0a4
[  457.457750]  r4:e9975200
[  457.460574] Code: 1a000003 e5953004 e3130001 1a000000 (e7f001f2)
[  457.467381] ---[ end trace 4acbc8c15e9e6aa7 ]---
Fixes 8689c051a201 ("cfg80211: dynamically allocate per-tid stats for station info")
Fixes f5ea9120be2e ("nl80211: add generation number to all dumps")
---
 drivers/staging/rtl8723bs/os_dep/ioctl_cfg80211.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/staging/rtl8723bs/os_dep/ioctl_cfg80211.c +1 -1
index 9a6e47877..2b45df79c 100644
@@ -2084,7 +2084,7 @@ void rtw_cfg80211_indicate_sta_assoc(struct adapter *padapter, u8 *pmgmt_frame,
 	struct net_device *ndev = padapter->pnetdev;
 
 	{
-		struct station_info sinfo;
+		struct station_info sinfo = {};
 		u8 ie_offset;
 		if (GetFrameSubType(pmgmt_frame) == WIFI_ASSOCREQ)
 			ie_offset = _ASOCREQ_IE_OFFSET_;
Signature · 1 line
2.25.1
Reviewed-by on v1 1/1
#14 lore raw

Re: [PATCH] staging: rtl8723bs: Fix uninitialized variables

If you run get_maintainer.pl on the patch then adds the developers from the Fixes tags so they can review the patch. I've added them but it's harder for them to review when they can't apply the patch...

Anyway, it looks good. Thanks!

regards,
dan carpenter

Quoted 61 lines · Wenli Looi

The sinfo.pertid and sinfo.generation variables are not initialized and it causes a crash when we use this as a wireless access point.

[  456.873025] ------------[ cut here ]------------
[  456.878198] kernel BUG at mm/slub.c:3968!
[  456.882680] Internal error: Oops - BUG: 0 [#1] PREEMPT SMP ARM

[ snip ]

[  457.271004] Backtrace:
[  457.273733] [<c02b7ee4>] (kfree) from [<c0e2a470>] (nl80211_send_station+0x954/0xfc4)
[  457.282481]  r9:eccca0c0 r8:e8edfec0 r7:00000000 r6:00000011 r5:e80a9480 r4:e8edfe00
[  457.291132] [<c0e29b1c>] (nl80211_send_station) from [<c0e2b18c>] (cfg80211_new_sta+0x90/0x1cc)
[  457.300850]  r10:e80a9480 r9:e8edfe00 r8:ea678cca r7:00000a20 r6:00000000 r5:ec46d000
[  457.309586]  r4:ec46d9e0
[  457.312433] [<c0e2b0fc>] (cfg80211_new_sta) from [<bf086684>] (rtw_cfg80211_indicate_sta_assoc+0x80/0x9c [r8723bs])
[  457.324095]  r10:00009930 r9:e85b9d80 r8:bf091050 r7:00000000 r6:00000000 r5:0000001c
[  457.332831]  r4:c1606788
[  457.335692] [<bf086604>] (rtw_cfg80211_indicate_sta_assoc [r8723bs]) from [<bf03df38>] (rtw_stassoc_event_callback+0x1c8/0x1d4 [r8723bs])
[  457.349489]  r7:ea678cc0 r6:000000a1 r5:f1225f84 r4:f086b000
[  457.355845] [<bf03dd70>] (rtw_stassoc_event_callback [r8723bs]) from [<bf048e4c>] (mlme_evt_hdl+0x8c/0xb4 [r8723bs])
[  457.367601]  r7:c1604900 r6:f086c4b8 r5:00000000 r4:f086c000
[  457.373959] [<bf048dc0>] (mlme_evt_hdl [r8723bs]) from [<bf03693c>] (rtw_cmd_thread+0x198/0x3d8 [r8723bs])
[  457.384744]  r5:f086e000 r4:f086c000
[  457.388754] [<bf0367a4>] (rtw_cmd_thread [r8723bs]) from [<c014a214>] (kthread+0x170/0x174)
[  457.398083]  r10:ed7a57e8 r9:bf0367a4 r8:f086b000 r7:e8ede000 r6:00000000 r5:e9975200
[  457.406828]  r4:e8369900
[  457.409653] [<c014a0a4>] (kthread) from [<c01010e8>] (ret_from_fork+0x14/0x2c)
[  457.417718] Exception stack(0xe8edffb0 to 0xe8edfff8)
[  457.423356] ffa0:                                     00000000 00000000 00000000 00000000
[  457.432492] ffc0: 00000000 00000000 00000000 00000000 00000000 00000000 00000000 00000000
[  457.441618] ffe0: 00000000 00000000 00000000 00000000 00000013 00000000
[  457.449006]  r10:00000000 r9:00000000 r8:00000000 r7:00000000 r6:00000000 r5:c014a0a4
[  457.457750]  r4:e9975200
[  457.460574] Code: 1a000003 e5953004 e3130001 1a000000 (e7f001f2)
[  457.467381] ---[ end trace 4acbc8c15e9e6aa7 ]---
Fixes 8689c051a201 ("cfg80211: dynamically allocate per-tid stats for station info")
Fixes f5ea9120be2e ("nl80211: add generation number to all dumps")
---
 drivers/staging/rtl8723bs/os_dep/ioctl_cfg80211.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/staging/rtl8723bs/os_dep/ioctl_cfg80211.c +1 -1
index 9a6e47877..2b45df79c 100644
@@ -2084,7 +2084,7 @@ void rtw_cfg80211_indicate_sta_assoc(struct adapter *padapter, u8 *pmgmt_frame,
 	struct net_device *ndev = padapter->pnetdev;
 
 	{
-		struct station_info sinfo;
+		struct station_info sinfo = {};
 		u8 ie_offset;
 		if (GetFrameSubType(pmgmt_frame) == WIFI_ASSOCREQ)
 			ie_offset = _ASOCREQ_IE_OFFSET_;
Signature · 1 line
2.25.1