Thread (2 messages) 2 messages, 2 authors, 2025-07-31

Re: [PATCH v2] sfc: handle NULL returned by xdp_convert_buff_to_frame()

From: Jesper Dangaard Brouer <hawk@kernel.org>
Date: 2025-07-31 09:23:30
Also in: bpf


On 26/07/2025 21.56, Chenyuan Yang wrote:
The xdp_convert_buff_to_frame() function can return NULL when there is
insufficient headroom in the buffer to store the xdp_frame structure
or when the driver didn't reserve enough tailroom for skb_shared_info.

Currently, the sfc driver does not check for this NULL return value
in the XDP_TX case within efx_do_xdp(). While the efx_xdp_tx_buffers()
function has some defensive checks, passing a NULL xdpf can still lead
to undefined behavior when the function tries to access xdpf->len and
xdpf->data.
The XDP_TX case is only for driver local frames/packets. And Edward says
the sfc driver reserves both enough headroom and tailroom.
Fix by adding a proper NULL check in the XDP_TX case, following the
suggestions of the developers.
In [V1] reply I question if this is possible
  - [v1] 
https://lore.kernel.org/all/a13646af-78f7-4ba7-9767-41d598222b1d@kernel.org/ (local)

Hmm... have you actually tested that XDP/BPF can adjust headroom so much
that xdp_convert_buff_to_frame() function fails?

I really doubt this possible for BPF-progs to violate this.

The XDP BPF-prog can only adjust the headroom via the helpers
bpf_xdp_adjust_head() and bpf_xdp_adjust_meta().  These helpers reserve
room for sizeof(struct xdp_frame).

The tailroom can be adjusted via helper bpf_xdp_adjust_tail() and it
also reserve room for sizeof(struct skb_shared_info) such that BPF-progs
cannot get access to this area. See define for xdp_data_hard_end.

quoted hunk ↗ jump to hunk
Fixes: 1b698fa5d8ef ("xdp: Rename convert_to_xdp_frame in xdp_convert_buff_to_frame")
Signed-off-by: Chenyuan Yang <redacted>
Cc: Kunwu Chan <redacted>
Cc: Edward Cree <redacted>
---
  drivers/net/ethernet/sfc/rx.c | 12 +++++++++---
  1 file changed, 9 insertions(+), 3 deletions(-)
diff --git a/drivers/net/ethernet/sfc/rx.c b/drivers/net/ethernet/sfc/rx.c
index ffca82207e47..b56457c23f66 100644
--- a/drivers/net/ethernet/sfc/rx.c
+++ b/drivers/net/ethernet/sfc/rx.c
@@ -308,14 +308,20 @@ static bool efx_do_xdp(struct efx_nic *efx, struct efx_channel *channel,
  	case XDP_TX:
  		/* Buffer ownership passes to tx on success. */
  		xdpf = xdp_convert_buff_to_frame(&xdp);
-		err = efx_xdp_tx_buffers(efx, 1, &xdpf, true);
+		if (unlikely(!xdpf))
+			err = -ENOBUFS;
+		else
+			err = efx_xdp_tx_buffers(efx, 1, &xdpf, true);
+
  		if (unlikely(err != 1)) {
  			efx_free_rx_buffers(rx_queue, rx_buf, 1);
  			if (net_ratelimit())
  				netif_err(efx, rx_err, efx->net_dev,
-					  "XDP TX failed (%d)\n", err);
+					  "XDP TX failed (%d)%s\n", err,
+					  err == -ENOBUFS ? " [frame conversion]" : "");
  			channel->n_rx_xdp_bad_drops++;
-			trace_xdp_exception(efx->net_dev, xdp_prog, xdp_act);
+			if (err != -ENOBUFS)
+				trace_xdp_exception(efx->net_dev, xdp_prog, xdp_act);
  		} else {
  			channel->n_rx_xdp_tx++;
  		}
  
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help