[PATCH] Bluetooth: Fix potential memory leak

[PATCH] Bluetooth: Fix potential memory leak

[Jaganath Kanakkassery]

If command is added to req then it should be freed in case if
hdev is down or HCI_ADVERTISING flag is not set.

Signed-off-by: Jaganath Kanakkassery <jaganathx.kanakkassery@intel.com>
---
 net/bluetooth/mgmt.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/net/bluetooth/mgmt.c b/net/bluetooth/mgmt.c
index 1fba2a0..58045ee 100644
--- a/net/bluetooth/mgmt.c
+++ b/net/bluetooth/mgmt.c
@@ -6383,6 +6383,7 @@ static int remove_advertising(struct sock *sk, struct hci_dev *hdev,
 	if (skb_queue_empty(&req.cmd_q) ||
 	    !hdev_is_powered(hdev) ||
 	    hci_dev_test_flag(hdev, HCI_ADVERTISING)) {
+		skb_queue_purge(&req.cmd_q);
 		rp.instance = cp->instance;
 		err = mgmt_cmd_complete(sk, hdev->id,
 					MGMT_OP_REMOVE_ADVERTISING,
-- 
2.7.4

Re: [PATCH] Bluetooth: Fix potential memory leak

[Marcel Holtmann]

Hi Jaganath,

> If command is added to req then it should be freed in case if
> hdev is down or HCI_ADVERTISING flag is not set.
> 
> Signed-off-by: Jaganath Kanakkassery <jaganathx.kanakkassery@intel.com>
> ---
> net/bluetooth/mgmt.c | 1 +
> 1 file changed, 1 insertion(+)
> 
> diff --git a/net/bluetooth/mgmt.c b/net/bluetooth/mgmt.c
> index 1fba2a0..58045ee 100644
> --- a/net/bluetooth/mgmt.c
> +++ b/net/bluetooth/mgmt.c
> @@ -6383,6 +6383,7 @@ static int remove_advertising(struct sock *sk, struct hci_dev *hdev,
> 	if (skb_queue_empty(&req.cmd_q) ||
> 	    !hdev_is_powered(hdev) ||
> 	    hci_dev_test_flag(hdev, HCI_ADVERTISING)) {
> +		skb_queue_purge(&req.cmd_q);
> 		rp.instance = cp->instance;
> 		err = mgmt_cmd_complete(sk, hdev->id,
> 					MGMT_OP_REMOVE_ADVERTISING,

this does not look right to me. It most likely has side affects. The fix must be differently if there is a memory leak.

Regards

Marcel

Re: [PATCH] Bluetooth: Fix potential memory leak

[Johan Hedberg]

Hi Marcel,

On Tue, Oct 24, 2017, Marcel Holtmann wrote:
> > index 1fba2a0..58045ee 100644
> > --- a/net/bluetooth/mgmt.c
> > +++ b/net/bluetooth/mgmt.c
> > @@ -6383,6 +6383,7 @@ static int remove_advertising(struct sock *sk, struct hci_dev *hdev,
> > 	if (skb_queue_empty(&req.cmd_q) ||
> > 	    !hdev_is_powered(hdev) ||
> > 	    hci_dev_test_flag(hdev, HCI_ADVERTISING)) {
> > +		skb_queue_purge(&req.cmd_q);
> > 		rp.instance = cp->instance;
> > 		err = mgmt_cmd_complete(sk, hdev->id,
> > 					MGMT_OP_REMOVE_ADVERTISING,
> 
> this does not look right to me. It most likely has side affects. The
> fix must be differently if there is a memory leak.

Actually, it looks like the right fix to me. We don't have a separate
helper to clean up a hci_request that never got used, and
skb_queue_purge is what other places in the code seem to be doing in
this kind of cases.

Johan

Re: [PATCH] Bluetooth: Fix potential memory leak

[Marcel Holtmann]

Hi Johan,

>>> index 1fba2a0..58045ee 100644
>>> --- a/net/bluetooth/mgmt.c
>>> +++ b/net/bluetooth/mgmt.c
>>> @@ -6383,6 +6383,7 @@ static int remove_advertising(struct sock *sk, struct hci_dev *hdev,
>>> 	if (skb_queue_empty(&req.cmd_q) ||
>>> 	    !hdev_is_powered(hdev) ||
>>> 	    hci_dev_test_flag(hdev, HCI_ADVERTISING)) {
>>> +		skb_queue_purge(&req.cmd_q);
>>> 		rp.instance = cp->instance;
>>> 		err = mgmt_cmd_complete(sk, hdev->id,
>>> 					MGMT_OP_REMOVE_ADVERTISING,
>> 
>> this does not look right to me. It most likely has side affects. The
>> fix must be differently if there is a memory leak.
> 
> Actually, it looks like the right fix to me. We don't have a separate
> helper to clean up a hci_request that never got used, and
> skb_queue_purge is what other places in the code seem to be doing in
> this kind of cases.

then we need a helper to do this. This is pretty bad style if we hook into the queue directly with a skb_queue_purge. Since that cmd_q should be a hidden detail.

Regards

Marcel

Re: [PATCH] Bluetooth: Fix potential memory leak

[Jaganath K]

Hi Marcel, Johan,

On Tue, Oct 24, 2017 at 6:13 PM, Marcel Holtmann <marcel@holtmann.org> wrote:
> Hi Johan,
>
>>>> index 1fba2a0..58045ee 100644
>>>> --- a/net/bluetooth/mgmt.c
>>>> +++ b/net/bluetooth/mgmt.c
>>>> @@ -6383,6 +6383,7 @@ static int remove_advertising(struct sock *sk, struct hci_dev *hdev,
>>>>     if (skb_queue_empty(&req.cmd_q) ||
>>>>         !hdev_is_powered(hdev) ||
>>>>         hci_dev_test_flag(hdev, HCI_ADVERTISING)) {
>>>> +           skb_queue_purge(&req.cmd_q);
>>>>             rp.instance = cp->instance;
>>>>             err = mgmt_cmd_complete(sk, hdev->id,
>>>>                                     MGMT_OP_REMOVE_ADVERTISING,
>>>
>>> this does not look right to me. It most likely has side affects. The
>>> fix must be differently if there is a memory leak.
>>
>> Actually, it looks like the right fix to me. We don't have a separate
>> helper to clean up a hci_request that never got used, and
>> skb_queue_purge is what other places in the code seem to be doing in
>> this kind of cases.
>
> then we need a helper to do this. This is pretty bad style if we hook into the queue directly with a skb_queue_purge. Since that cmd_q should be a hidden detail.

I have raised v1 with the suggested changes.

Thanks,
Jaganath