ixgbe: fix error handling in TC cls_u32 offload routines
authorSridhar Samudrala <sridhar.samudrala@intel.com>
Mon, 7 Mar 2016 17:41:47 +0000 (09:41 -0800)
committerJeff Kirsher <jeffrey.t.kirsher@intel.com>
Wed, 30 Mar 2016 06:05:51 +0000 (23:05 -0700)
Check for handle ids when adding/deleting hash nodes OR adding/deleting
filter entries and limit them to max number of links or header nodes
supported(IXGBE_MAX_LINK_HANDLE).

Start from bit 0 when setting hash table bit-map.(adapter->tables)

Signed-off-by: Sridhar Samudrala <sridhar.samudrala@intel.com>
Acked-by: John Fastabend <john.r.fastabend@intel.com>
Tested-by: Andrew Bowers <andrewx.bowers@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
drivers/net/ethernet/intel/ixgbe/ixgbe_main.c

index f5736a3b252ef7f2269eda18f12c17deb2d41b40..ca9c543417847e4a805dea08efc1ca84a78cae59 100644 (file)
@@ -8192,10 +8192,17 @@ int ixgbe_setup_tc(struct net_device *dev, u8 tc)
 static int ixgbe_delete_clsu32(struct ixgbe_adapter *adapter,
                               struct tc_cls_u32_offload *cls)
 {
+       u32 uhtid = TC_U32_USERHTID(cls->knode.handle);
+       u32 loc;
        int err;
 
+       if ((uhtid != 0x800) && (uhtid >= IXGBE_MAX_LINK_HANDLE))
+               return -EINVAL;
+
+       loc = cls->knode.handle & 0xfffff;
+
        spin_lock(&adapter->fdir_perfect_lock);
-       err = ixgbe_update_ethtool_fdir_entry(adapter, NULL, cls->knode.handle);
+       err = ixgbe_update_ethtool_fdir_entry(adapter, NULL, loc);
        spin_unlock(&adapter->fdir_perfect_lock);
        return err;
 }
@@ -8204,20 +8211,30 @@ static int ixgbe_configure_clsu32_add_hnode(struct ixgbe_adapter *adapter,
                                            __be16 protocol,
                                            struct tc_cls_u32_offload *cls)
 {
+       u32 uhtid = TC_U32_USERHTID(cls->hnode.handle);
+
+       if (uhtid >= IXGBE_MAX_LINK_HANDLE)
+               return -EINVAL;
+
        /* This ixgbe devices do not support hash tables at the moment
         * so abort when given hash tables.
         */
        if (cls->hnode.divisor > 0)
                return -EINVAL;
 
-       set_bit(TC_U32_USERHTID(cls->hnode.handle), &adapter->tables);
+       set_bit(uhtid - 1, &adapter->tables);
        return 0;
 }
 
 static int ixgbe_configure_clsu32_del_hnode(struct ixgbe_adapter *adapter,
                                            struct tc_cls_u32_offload *cls)
 {
-       clear_bit(TC_U32_USERHTID(cls->hnode.handle), &adapter->tables);
+       u32 uhtid = TC_U32_USERHTID(cls->hnode.handle);
+
+       if (uhtid >= IXGBE_MAX_LINK_HANDLE)
+               return -EINVAL;
+
+       clear_bit(uhtid - 1, &adapter->tables);
        return 0;
 }
 
@@ -8235,27 +8252,29 @@ static int ixgbe_configure_clsu32(struct ixgbe_adapter *adapter,
 #endif
        int i, err = 0;
        u8 queue;
-       u32 handle;
+       u32 uhtid, link_uhtid;
 
        memset(&mask, 0, sizeof(union ixgbe_atr_input));
-       handle = cls->knode.handle;
+       uhtid = TC_U32_USERHTID(cls->knode.handle);
+       link_uhtid = TC_U32_USERHTID(cls->knode.link_handle);
 
-       /* At the moment cls_u32 jumps to transport layer and skips past
+       /* At the moment cls_u32 jumps to network layer and skips past
         * L2 headers. The canonical method to match L2 frames is to use
         * negative values. However this is error prone at best but really
         * just broken because there is no way to "know" what sort of hdr
-        * is in front of the transport layer. Fix cls_u32 to support L2
+        * is in front of the network layer. Fix cls_u32 to support L2
         * headers when needed.
         */
        if (protocol != htons(ETH_P_IP))
                return -EINVAL;
 
-       if (cls->knode.link_handle ||
-           cls->knode.link_handle >= IXGBE_MAX_LINK_HANDLE) {
+       if (link_uhtid) {
                struct ixgbe_nexthdr *nexthdr = ixgbe_ipv4_jumps;
-               u32 uhtid = TC_U32_USERHTID(cls->knode.link_handle);
 
-               if (!test_bit(uhtid, &adapter->tables))
+               if (link_uhtid >= IXGBE_MAX_LINK_HANDLE)
+                       return -EINVAL;
+
+               if (!test_bit(link_uhtid - 1, &adapter->tables))
                        return -EINVAL;
 
                for (i = 0; nexthdr[i].jump; i++) {
@@ -8271,10 +8290,7 @@ static int ixgbe_configure_clsu32(struct ixgbe_adapter *adapter,
                            nexthdr->mask != cls->knode.sel->keys[0].mask)
                                return -EINVAL;
 
-                       if (uhtid >= IXGBE_MAX_LINK_HANDLE)
-                               return -EINVAL;
-
-                       adapter->jump_tables[uhtid] = nexthdr->jump;
+                       adapter->jump_tables[link_uhtid] = nexthdr->jump;
                }
                return 0;
        }
@@ -8291,13 +8307,13 @@ static int ixgbe_configure_clsu32(struct ixgbe_adapter *adapter,
         * To add support for new nodes update ixgbe_model.h parse structures
         * this function _should_ be generic try not to hardcode values here.
         */
-       if (TC_U32_USERHTID(handle) == 0x800) {
+       if (uhtid == 0x800) {
                field_ptr = adapter->jump_tables[0];
        } else {
-               if (TC_U32_USERHTID(handle) >= ARRAY_SIZE(adapter->jump_tables))
+               if (uhtid >= IXGBE_MAX_LINK_HANDLE)
                        return -EINVAL;
 
-               field_ptr = adapter->jump_tables[TC_U32_USERHTID(handle)];
+               field_ptr = adapter->jump_tables[uhtid];
        }
 
        if (!field_ptr)