Fix AF_PACKET ABI breakage in 4.2
authorDavid Woodhouse <dwmw2@infradead.org>
Wed, 23 Sep 2015 18:45:08 +0000 (19:45 +0100)
committerDavid S. Miller <davem@davemloft.net>
Wed, 23 Sep 2015 21:33:55 +0000 (14:33 -0700)
Commit 7d82410950aa ("virtio: add explicit big-endian support to memory
accessors") accidentally changed the virtio_net header used by
AF_PACKET with PACKET_VNET_HDR from host-endian to big-endian.

Since virtio_legacy_is_little_endian() is a very long identifier,
define a vio_le macro and use that throughout the code instead of the
hard-coded 'false' for little-endian.

This restores the ABI to match 4.1 and earlier kernels, and makes my
test program work again.

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/packet/af_packet.c

index 7b8e39a223879c3cb0a9d2a0bc6ff003118379f9..aa4b15c3588445ae67568d35bddbeed6a3508cbd 100644 (file)
@@ -230,6 +230,8 @@ struct packet_skb_cb {
        } sa;
 };
 
+#define vio_le() virtio_legacy_is_little_endian()
+
 #define PACKET_SKB_CB(__skb)   ((struct packet_skb_cb *)((__skb)->cb))
 
 #define GET_PBDQC_FROM_RB(x)   ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
@@ -2680,15 +2682,15 @@ static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
                        goto out_unlock;
 
                if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
-                   (__virtio16_to_cpu(false, vnet_hdr.csum_start) +
-                    __virtio16_to_cpu(false, vnet_hdr.csum_offset) + 2 >
-                     __virtio16_to_cpu(false, vnet_hdr.hdr_len)))
-                       vnet_hdr.hdr_len = __cpu_to_virtio16(false,
-                                __virtio16_to_cpu(false, vnet_hdr.csum_start) +
-                               __virtio16_to_cpu(false, vnet_hdr.csum_offset) + 2);
+                   (__virtio16_to_cpu(vio_le(), vnet_hdr.csum_start) +
+                    __virtio16_to_cpu(vio_le(), vnet_hdr.csum_offset) + 2 >
+                     __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len)))
+                       vnet_hdr.hdr_len = __cpu_to_virtio16(vio_le(),
+                                __virtio16_to_cpu(vio_le(), vnet_hdr.csum_start) +
+                               __virtio16_to_cpu(vio_le(), vnet_hdr.csum_offset) + 2);
 
                err = -EINVAL;
-               if (__virtio16_to_cpu(false, vnet_hdr.hdr_len) > len)
+               if (__virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len) > len)
                        goto out_unlock;
 
                if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
@@ -2731,7 +2733,7 @@ static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
        hlen = LL_RESERVED_SPACE(dev);
        tlen = dev->needed_tailroom;
        skb = packet_alloc_skb(sk, hlen + tlen, hlen, len,
-                              __virtio16_to_cpu(false, vnet_hdr.hdr_len),
+                              __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len),
                               msg->msg_flags & MSG_DONTWAIT, &err);
        if (skb == NULL)
                goto out_unlock;
@@ -2778,8 +2780,8 @@ static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
 
        if (po->has_vnet_hdr) {
                if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
-                       u16 s = __virtio16_to_cpu(false, vnet_hdr.csum_start);
-                       u16 o = __virtio16_to_cpu(false, vnet_hdr.csum_offset);
+                       u16 s = __virtio16_to_cpu(vio_le(), vnet_hdr.csum_start);
+                       u16 o = __virtio16_to_cpu(vio_le(), vnet_hdr.csum_offset);
                        if (!skb_partial_csum_set(skb, s, o)) {
                                err = -EINVAL;
                                goto out_free;
@@ -2787,7 +2789,7 @@ static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
                }
 
                skb_shinfo(skb)->gso_size =
-                       __virtio16_to_cpu(false, vnet_hdr.gso_size);
+                       __virtio16_to_cpu(vio_le(), vnet_hdr.gso_size);
                skb_shinfo(skb)->gso_type = gso_type;
 
                /* Header must be checked, and gso_segs computed. */
@@ -3161,9 +3163,9 @@ static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
 
                        /* This is a hint as to how much should be linear. */
                        vnet_hdr.hdr_len =
-                               __cpu_to_virtio16(false, skb_headlen(skb));
+                               __cpu_to_virtio16(vio_le(), skb_headlen(skb));
                        vnet_hdr.gso_size =
-                               __cpu_to_virtio16(false, sinfo->gso_size);
+                               __cpu_to_virtio16(vio_le(), sinfo->gso_size);
                        if (sinfo->gso_type & SKB_GSO_TCPV4)
                                vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
                        else if (sinfo->gso_type & SKB_GSO_TCPV6)
@@ -3181,9 +3183,9 @@ static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
 
                if (skb->ip_summed == CHECKSUM_PARTIAL) {
                        vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
-                       vnet_hdr.csum_start = __cpu_to_virtio16(false,
+                       vnet_hdr.csum_start = __cpu_to_virtio16(vio_le(),
                                          skb_checksum_start_offset(skb));
-                       vnet_hdr.csum_offset = __cpu_to_virtio16(false,
+                       vnet_hdr.csum_offset = __cpu_to_virtio16(vio_le(),
                                                         skb->csum_offset);
                } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
                        vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;