[PATCH] scsi: target: cxgbit: check skb dequeue result

[PATCH] scsi: target: cxgbit: check skb dequeue result

[Denis Plotnikov]

On a couple of abort packet paths skb dequeuing may end up with
returning NULL, which, in turn, may end up with further null
pointer dereference.

Fix it by checking the return value of skb dequeuing.

Found by Linux Verification Center(linuxtesting.org) with SVACE.

Fixes: 9730ffcb8957 (cxgbit: add files for cxgbit.ko)
Signed-off-by: Denis Plotnikov <den-plotnikov@yandex-team.ru>
---
 drivers/target/iscsi/cxgbit/cxgbit_cm.c | 10 +++++++---
 1 file changed, 7 insertions(+), 3 deletions(-)

diff --git a/drivers/target/iscsi/cxgbit/cxgbit_cm.c b/drivers/target/iscsi/cxgbit/cxgbit_cm.c
index 518ded214e74..d43fd761c20a 100644
--- a/drivers/target/iscsi/cxgbit/cxgbit_cm.c
+++ b/drivers/target/iscsi/cxgbit/cxgbit_cm.c
@@ -669,6 +669,9 @@ static int cxgbit_send_abort_req(struct cxgbit_sock *csk)
 		cxgbit_send_tx_flowc_wr(csk);
 
 	skb = __skb_dequeue(&csk->skbq);
+	if (!skb)
+		return 0;
+
 	cxgb_mk_abort_req(skb, len, csk->tid, csk->txq_idx,
 			  csk->com.cdev, cxgbit_abort_arp_failure);
 
@@ -1769,9 +1772,10 @@ static void cxgbit_abort_req_rss(struct cxgbit_sock *csk, struct sk_buff *skb)
 		cxgbit_send_tx_flowc_wr(csk);
 
 	rpl_skb = __skb_dequeue(&csk->skbq);
-
-	cxgb_mk_abort_rpl(rpl_skb, len, csk->tid, csk->txq_idx);
-	cxgbit_ofld_send(csk->com.cdev, rpl_skb);
+	if (!rpl_skb) {
+		cxgb_mk_abort_rpl(rpl_skb, len, csk->tid, csk->txq_idx);
+		cxgbit_ofld_send(csk->com.cdev, rpl_skb);
+	}
 
 	if (wakeup_thread) {
 		cxgbit_queue_rx_skb(csk, skb);
-- 
2.25.1

Re: [PATCH] scsi: target: cxgbit: check skb dequeue result

[Bodo Stroesser]

On 06.04.23 17:04, Denis Plotnikov wrote:
> On a couple of abort packet paths skb dequeuing may end up with
> returning NULL, which, in turn, may end up with further null
> pointer dereference.
> 
> Fix it by checking the return value of skb dequeuing.
> 
> Found by Linux Verification Center(linuxtesting.org) with SVACE.
> 
> Fixes: 9730ffcb8957 (cxgbit: add files for cxgbit.ko)
> Signed-off-by: Denis Plotnikov <den-plotnikov@yandex-team.ru>
> ---
>   drivers/target/iscsi/cxgbit/cxgbit_cm.c | 10 +++++++---
>   1 file changed, 7 insertions(+), 3 deletions(-)
> 
> diff --git a/drivers/target/iscsi/cxgbit/cxgbit_cm.c b/drivers/target/iscsi/cxgbit/cxgbit_cm.c
> index 518ded214e74..d43fd761c20a 100644
> --- a/drivers/target/iscsi/cxgbit/cxgbit_cm.c
> +++ b/drivers/target/iscsi/cxgbit/cxgbit_cm.c
> @@ -669,6 +669,9 @@ static int cxgbit_send_abort_req(struct cxgbit_sock *csk)
>   		cxgbit_send_tx_flowc_wr(csk);
>   
>   	skb = __skb_dequeue(&csk->skbq);
> +	if (!skb)
> +		return 0;
> +
>   	cxgb_mk_abort_req(skb, len, csk->tid, csk->txq_idx,
>   			  csk->com.cdev, cxgbit_abort_arp_failure);
>   
> @@ -1769,9 +1772,10 @@ static void cxgbit_abort_req_rss(struct cxgbit_sock *csk, struct sk_buff *skb)
>   		cxgbit_send_tx_flowc_wr(csk);
>   
>   	rpl_skb = __skb_dequeue(&csk->skbq);
> -
> -	cxgb_mk_abort_rpl(rpl_skb, len, csk->tid, csk->txq_idx);
> -	cxgbit_ofld_send(csk->com.cdev, rpl_skb);
> +	if (!rpl_skb) {

Honestly I have no clue about cxgbit, but to avoid null pointer
dereference, shouldn't it be "if (rpl_skb) {"?

Bodo


> +		cxgb_mk_abort_rpl(rpl_skb, len, csk->tid, csk->txq_idx);
> +		cxgbit_ofld_send(csk->com.cdev, rpl_skb);
> +	}
>   
>   	if (wakeup_thread) {
>   		cxgbit_queue_rx_skb(csk, skb);

Re: [PATCH] scsi: target: cxgbit: check skb dequeue result

[Denis Plotnikov]

On 07.04.2023 22:16, Bodo Stroesser wrote:
> On 06.04.23 17:04, Denis Plotnikov wrote:
>> On a couple of abort packet paths skb dequeuing may end up with
>> returning NULL, which, in turn, may end up with further null
>> pointer dereference.
>>
>> Fix it by checking the return value of skb dequeuing.
>>
>> Found by Linux Verification Center(linuxtesting.org) with SVACE.
>>
>> Fixes: 9730ffcb8957 (cxgbit: add files for cxgbit.ko)
>> Signed-off-by: Denis Plotnikov <den-plotnikov@yandex-team.ru>
>> ---
>>   drivers/target/iscsi/cxgbit/cxgbit_cm.c | 10 +++++++---
>>   1 file changed, 7 insertions(+), 3 deletions(-)
>>
>> diff --git a/drivers/target/iscsi/cxgbit/cxgbit_cm.c 
>> b/drivers/target/iscsi/cxgbit/cxgbit_cm.c
>> index 518ded214e74..d43fd761c20a 100644
>> --- a/drivers/target/iscsi/cxgbit/cxgbit_cm.c
>> +++ b/drivers/target/iscsi/cxgbit/cxgbit_cm.c
>> @@ -669,6 +669,9 @@ static int cxgbit_send_abort_req(struct 
>> cxgbit_sock *csk)
>>           cxgbit_send_tx_flowc_wr(csk);
>>         skb = __skb_dequeue(&csk->skbq);
>> +    if (!skb)
>> +        return 0;
>> +
>>       cxgb_mk_abort_req(skb, len, csk->tid, csk->txq_idx,
>>                 csk->com.cdev, cxgbit_abort_arp_failure);
>>   @@ -1769,9 +1772,10 @@ static void cxgbit_abort_req_rss(struct 
>> cxgbit_sock *csk, struct sk_buff *skb)
>>           cxgbit_send_tx_flowc_wr(csk);
>>         rpl_skb = __skb_dequeue(&csk->skbq);
>> -
>> -    cxgb_mk_abort_rpl(rpl_skb, len, csk->tid, csk->txq_idx);
>> -    cxgbit_ofld_send(csk->com.cdev, rpl_skb);
>> +    if (!rpl_skb) {
>
> Honestly I have no clue about cxgbit, but to avoid null pointer
> dereference, shouldn't it be "if (rpl_skb) {"?

Yes it should. Thanks for finding that. I'll resend the patch.

Frankly, I have no clue about cxgbit as well, so may be there is a 
better way to cope with the null pointer issue here.

Denis

>
> Bodo
>
>
>> +        cxgb_mk_abort_rpl(rpl_skb, len, csk->tid, csk->txq_idx);
>> +        cxgbit_ofld_send(csk->com.cdev, rpl_skb);
>> +    }
>>         if (wakeup_thread) {
>>           cxgbit_queue_rx_skb(csk, skb);

RE: [PATCH] scsi: target: cxgbit: check skb dequeue result

[Varun Prakash]

>On a couple of abort packet paths skb dequeuing may end up with
>returning NULL, which, in turn, may end up with further null
>pointer dereference.
>
>Fix it by checking the return value of skb dequeuing.

skbs in csk->skbq are preallocated for critical events in cxgbit_alloc_csk_skb(),
so there is no need to check for NULL.

Re: [PATCH] scsi: target: cxgbit: check skb dequeue result

[Denis Plotnikov]

On 10.04.2023 14:45, Varun Prakash wrote:
>> On a couple of abort packet paths skb dequeuing may end up with
>> returning NULL, which, in turn, may end up with further null
>> pointer dereference.
>>
>> Fix it by checking the return value of skb dequeuing.
> skbs in csk->skbq are preallocated for critical events in cxgbit_alloc_csk_skb(),
> so there is no need to check for NULL.
Ok, thanks for the reply