core, nfqueue, openvswitch: Orphan frags in skb_zerocopy and handle errors
authorZoltan Kiss <zoltan.kiss@citrix.com>
Wed, 26 Mar 2014 22:37:45 +0000 (22:37 +0000)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Thu, 31 Jul 2014 19:53:52 +0000 (12:53 -0700)
commit 36d5fe6a000790f56039afe26834265db0a3ad4c upstream.

skb_zerocopy can copy elements of the frags array between skbs, but it doesn't
orphan them. Also, it doesn't handle errors, so this patch takes care of that
as well, and modify the callers accordingly. skb_tx_error() is also added to
the callers so they will signal the failed delivery towards the creator of the
skb.

Signed-off-by: Zoltan Kiss <zoltan.kiss@citrix.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
[bwh: Backported to 3.13: skb_zerocopy() is new in 3.14, but was moved from a
 static function in nfnetlink_queue.  We need to patch that and its caller, but
 not openvswitch.]
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
net/netfilter/nfnetlink_queue_core.c

index 5352b2d..2b8199f 100644 (file)
@@ -227,22 +227,23 @@ nfqnl_flush(struct nfqnl_instance *queue, nfqnl_cmpfn cmpfn, unsigned long data)
        spin_unlock_bh(&queue->lock);
 }
 
-static void
+static int
 nfqnl_zcopy(struct sk_buff *to, const struct sk_buff *from, int len, int hlen)
 {
        int i, j = 0;
        int plen = 0; /* length of skb->head fragment */
+       int ret;
        struct page *page;
        unsigned int offset;
 
        /* dont bother with small payloads */
-       if (len <= skb_tailroom(to)) {
-               skb_copy_bits(from, 0, skb_put(to, len), len);
-               return;
-       }
+       if (len <= skb_tailroom(to))
+               return skb_copy_bits(from, 0, skb_put(to, len), len);
 
        if (hlen) {
-               skb_copy_bits(from, 0, skb_put(to, hlen), hlen);
+               ret = skb_copy_bits(from, 0, skb_put(to, hlen), hlen);
+               if (unlikely(ret))
+                       return ret;
                len -= hlen;
        } else {
                plen = min_t(int, skb_headlen(from), len);
@@ -260,6 +261,11 @@ nfqnl_zcopy(struct sk_buff *to, const struct sk_buff *from, int len, int hlen)
        to->len += len + plen;
        to->data_len += len + plen;
 
+       if (unlikely(skb_orphan_frags(from, GFP_ATOMIC))) {
+               skb_tx_error(from);
+               return -ENOMEM;
+       }
+
        for (i = 0; i < skb_shinfo(from)->nr_frags; i++) {
                if (!len)
                        break;
@@ -270,6 +276,8 @@ nfqnl_zcopy(struct sk_buff *to, const struct sk_buff *from, int len, int hlen)
                j++;
        }
        skb_shinfo(to)->nr_frags = j;
+
+       return 0;
 }
 
 static int nfqnl_put_packet_info(struct sk_buff *nlskb, struct sk_buff *packet)
@@ -355,13 +363,16 @@ nfqnl_build_packet_message(struct nfqnl_instance *queue,
 
        skb = nfnetlink_alloc_skb(&init_net, size, queue->peer_portid,
                                  GFP_ATOMIC);
-       if (!skb)
+       if (!skb) {
+               skb_tx_error(entskb);
                return NULL;
+       }
 
        nlh = nlmsg_put(skb, 0, 0,
                        NFNL_SUBSYS_QUEUE << 8 | NFQNL_MSG_PACKET,
                        sizeof(struct nfgenmsg), 0);
        if (!nlh) {
+               skb_tx_error(entskb);
                kfree_skb(skb);
                return NULL;
        }
@@ -481,13 +492,15 @@ nfqnl_build_packet_message(struct nfqnl_instance *queue,
                nla->nla_type = NFQA_PAYLOAD;
                nla->nla_len = nla_attr_size(data_len);
 
-               nfqnl_zcopy(skb, entskb, data_len, hlen);
+               if (nfqnl_zcopy(skb, entskb, data_len, hlen))
+                       goto nla_put_failure;
        }
 
        nlh->nlmsg_len = skb->len;
        return skb;
 
 nla_put_failure:
+       skb_tx_error(entskb);
        kfree_skb(skb);
        net_err_ratelimited("nf_queue: error creating packet message\n");
        return NULL;