net: fix GSO for SG-enabled devices
authorPaolo Abeni <pabeni@redhat.com>
Tue, 19 Jan 2021 16:56:56 +0000 (17:56 +0100)
committerJakub Kicinski <kuba@kernel.org>
Wed, 20 Jan 2021 02:40:40 +0000 (18:40 -0800)
The commit dbd50f238dec ("net: move the hsize check to the else
block in skb_segment") introduced a data corruption for devices
supporting scatter-gather.

The problem boils down to signed/unsigned comparison given
unexpected results: if signed 'hsize' is negative, it will be
considered greater than a positive 'len', which is unsigned.

This commit addresses resorting to the old checks order, so that
'hsize' never has a negative value when compared with 'len'.

v1 -> v2:
 - reorder hsize checks instead of explicit cast (Alex)

Bisected-by: Matthieu Baerts <matthieu.baerts@tessares.net>
Fixes: dbd50f238dec ("net: move the hsize check to the else block in skb_segment")
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Reviewed-by: Xin Long <lucien.xin@gmail.com>
Link: https://lore.kernel.org/r/861947c2d2d087db82af93c21920ce8147d15490.1611074818.git.pabeni@redhat.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
net/core/skbuff.c

index e835193..cf2c4dc 100644 (file)
@@ -3938,10 +3938,10 @@ normal:
                        skb_release_head_state(nskb);
                        __skb_push(nskb, doffset);
                } else {
+                       if (hsize < 0)
+                               hsize = 0;
                        if (hsize > len || !sg)
                                hsize = len;
-                       else if (hsize < 0)
-                               hsize = 0;
 
                        nskb = __alloc_skb(hsize + doffset + headroom,
                                           GFP_ATOMIC, skb_alloc_rx_flag(head_skb),