Re: [PATCH net] net: vertexcom: mse102x: Fix possible double free of TX skb

From: Stefan Wahren
Date: Tue Oct 29 2024 - 19:08:44 EST


Am 29.10.24 um 23:01 schrieb Jakub Kicinski:
On Tue, 29 Oct 2024 22:15:15 +0100 Stefan Wahren wrote:
Isn't it easier to change this function to free the copy rather than
the original? That way the original will remain valid for the callers.
You mean something like this?

diff --git a/drivers/net/ethernet/vertexcom/mse102x.c
b/drivers/net/ethernet/vertexcom/mse102x.c
index a04d4073def9..2c37957478fb 100644
--- a/drivers/net/ethernet/vertexcom/mse102x.c
+++ b/drivers/net/ethernet/vertexcom/mse102x.c
@@ -222,7 +222,7 @@ static int mse102x_tx_frame_spi(struct mse102x_net
*mse, struct sk_buff *txp,
     struct mse102x_net_spi *mses = to_mse102x_spi(mse);
     struct spi_transfer *xfer = &mses->spi_xfer;
     struct spi_message *msg = &mses->spi_msg;
-    struct sk_buff *tskb;
+    struct sk_buff *tskb = NULL;
     int ret;

     netif_dbg(mse, tx_queued, mse->ndev, "%s: skb %p, %d@%p\n",
@@ -235,7 +235,6 @@ static int mse102x_tx_frame_spi(struct mse102x_net
*mse, struct sk_buff *txp,
         if (!tskb)
             return -ENOMEM;

-        dev_kfree_skb(txp);
         txp = tskb;
     }

@@ -257,6 +256,8 @@ static int mse102x_tx_frame_spi(struct mse102x_net
*mse, struct sk_buff *txp,
         mse->stats.xfer_err++;
     }

+    dev_kfree_skb(tskb);
+
     return ret;
 }
Exactly, I think it would work and it feels simpler.
I didn't test it yet, i need access to evaluation board before. But this
change will behave differently regarding stats of tx_bytes [1]. The
first version will include the padding, while the second does not.

[1] -
https://elixir.bootlin.com/linux/v6.12-rc5/source/drivers/net/ethernet/vertexcom/mse102x.c#L445