On 8/5/21 2:55 PM, Julian Wiedmann wrote:

On 02.08.21 11:52, Vasily Averin wrote:

quoted

Unlike skb_realloc_headroom, new helper skb_expand_head
does not allocate a new skb if possible.

Signed-off-by: Vasily Averin <redacted>
---
 drivers/net/vrf.c | 21 +++++++--------------
 1 file changed, 7 insertions(+), 14 deletions(-)

[...]

quoted

 	/* Be paranoid, rather than too clever. */
 	if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
-		struct sk_buff *skb2;
-
-		skb2 = skb_realloc_headroom(skb, LL_RESERVED_SPACE(dev));
-		if (!skb2) {
-			ret = -ENOMEM;
-			goto err;
+		skb = skb_expand_head(skb, hh_len);
+		if (!skb) {
+			skb->dev->stats.tx_errors++;
+			return -ENOMEM;

Hello Vasily,

FYI, Coverity complains that we check skb != NULL here but then
still dereference skb->dev:


*** CID 1506214:  Null pointer dereferences  (FORWARD_NULL)
/drivers/net/vrf.c: 867 in vrf_finish_output()
861     	nf_reset_ct(skb);
862     
863     	/* Be paranoid, rather than too clever. */
864     	if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
865     		skb = skb_expand_head(skb, hh_len);
866     		if (!skb) {

quoted

quoted

quoted

    CID 1506214:  Null pointer dereferences  (FORWARD_NULL)
    Dereferencing null pointer "skb".

867     			skb->dev->stats.tx_errors++;
868     			return -ENOMEM;

My fault, I missed it.

Thank you,
	Vasily Averin