net: aquantia: memory corruption on jumbo frames
authorFriedemann Gerold <f.gerold@b-c-s.de>
Sat, 15 Sep 2018 15:03:39 +0000 (18:03 +0300)
committerDavid S. Miller <davem@davemloft.net>
Mon, 24 Sep 2018 05:25:25 +0000 (22:25 -0700)
This patch fixes skb_shared area, which will be corrupted
upon reception of 4K jumbo packets.

Originally build_skb usage purpose was to reuse page for skb to eliminate
needs of extra fragments. But that logic does not take into account that
skb_shared_info should be reserved at the end of skb data area.

In case packet data consumes all the page (4K), skb_shinfo location
overflows the page. As a consequence, __build_skb zeroed shinfo data above
the allocated page, corrupting next page.

The issue is rarely seen in real life because jumbo are normally larger
than 4K and that causes another code path to trigger.
But it 100% reproducible with simple scapy packet, like:

    sendp(IP(dst="192.168.100.3") / TCP(dport=443) \
          / Raw(RandString(size=(4096-40))), iface="enp1s0")

Fixes: 018423e90bee ("net: ethernet: aquantia: Add ring support code")
Reported-by: Friedemann Gerold <f.gerold@b-c-s.de>
Reported-by: Michael Rauch <michael@rauch.be>
Signed-off-by: Friedemann Gerold <f.gerold@b-c-s.de>
Tested-by: Nikita Danilov <nikita.danilov@aquantia.com>
Signed-off-by: Igor Russkikh <igor.russkikh@aquantia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
drivers/net/ethernet/aquantia/atlantic/aq_ring.c

index b5f1f62e8e253785436fa7cd9119a8467edf4fd4..d1e1a0ba86150208e73ffdbce006e8d1a822302c 100644 (file)
@@ -225,9 +225,10 @@ int aq_ring_rx_clean(struct aq_ring_s *self,
                }
 
                /* for single fragment packets use build_skb() */
-               if (buff->is_eop) {
+               if (buff->is_eop &&
+                   buff->len <= AQ_CFG_RX_FRAME_MAX - AQ_SKB_ALIGN) {
                        skb = build_skb(page_address(buff->page),
-                                       buff->len + AQ_SKB_ALIGN);
+                                       AQ_CFG_RX_FRAME_MAX);
                        if (unlikely(!skb)) {
                                err = -ENOMEM;
                                goto err_exit;
@@ -247,18 +248,21 @@ int aq_ring_rx_clean(struct aq_ring_s *self,
                                        buff->len - ETH_HLEN,
                                        SKB_TRUESIZE(buff->len - ETH_HLEN));
 
-                       for (i = 1U, next_ = buff->next,
-                            buff_ = &self->buff_ring[next_]; true;
-                            next_ = buff_->next,
-                            buff_ = &self->buff_ring[next_], ++i) {
-                               skb_add_rx_frag(skb, i, buff_->page, 0,
-                                               buff_->len,
-                                               SKB_TRUESIZE(buff->len -
-                                               ETH_HLEN));
-                               buff_->is_cleaned = 1;
-
-                               if (buff_->is_eop)
-                                       break;
+                       if (!buff->is_eop) {
+                               for (i = 1U, next_ = buff->next,
+                                    buff_ = &self->buff_ring[next_];
+                                    true; next_ = buff_->next,
+                                    buff_ = &self->buff_ring[next_], ++i) {
+                                       skb_add_rx_frag(skb, i,
+                                                       buff_->page, 0,
+                                                       buff_->len,
+                                                       SKB_TRUESIZE(buff->len -
+                                                       ETH_HLEN));
+                                       buff_->is_cleaned = 1;
+
+                                       if (buff_->is_eop)
+                                               break;
+                               }
                        }
                }