#define FWS_INIT FWS_L
 #endif
 
-static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn);
-static struct fib6_node *fib6_repair_tree(struct net *net, struct fib6_node *fn);
+static struct rt6_info *fib6_find_prefix(struct net *net,
+                                        struct fib6_table *table,
+                                        struct fib6_node *fn);
+static struct fib6_node *fib6_repair_tree(struct net *net,
+                                         struct fib6_table *table,
+                                         struct fib6_node *fn);
 static int fib6_walk(struct net *net, struct fib6_walker *w);
 static int fib6_walk_continue(struct fib6_walker *w);
 
        struct net *net = dev_net(rt->dst.dev);
        struct fib6_node *fn;
 
-       write_lock_bh(&table->tb6_lock);
+       spin_lock_bh(&table->tb6_lock);
        fn = rcu_dereference_protected(rt->rt6i_node,
                        lockdep_is_held(&table->tb6_lock));
        if (fn)
                fn->fn_sernum = fib6_new_sernum(net);
-       write_unlock_bh(&table->tb6_lock);
+       spin_unlock_bh(&table->tb6_lock);
 }
 
 /*
         * Initialize table lock at a single place to give lockdep a key,
         * tables aren't visible prior to being linked to the list.
         */
-       rwlock_init(&tb->tb6_lock);
-
+       spin_lock_init(&tb->tb6_lock);
        h = tb->tb6_id & (FIB6_TABLE_HASHSZ - 1);
 
        /*
        table = kzalloc(sizeof(*table), GFP_ATOMIC);
        if (table) {
                table->tb6_id = id;
-               table->tb6_root.leaf = net->ipv6.ip6_null_entry;
+               rcu_assign_pointer(table->tb6_root.leaf,
+                                  net->ipv6.ip6_null_entry);
                table->tb6_root.fn_flags = RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
                inet_peer_base_init(&table->tb6_peers);
        }
                struct hlist_head *head = &net->ipv6.fib_table_hash[h];
                struct fib6_table *tb;
 
-               hlist_for_each_entry_rcu(tb, head, tb6_hlist) {
-                       read_lock_bh(&tb->tb6_lock);
+               hlist_for_each_entry_rcu(tb, head, tb6_hlist)
                        fib_seq += tb->fib_seq;
-                       read_unlock_bh(&tb->tb6_lock);
-               }
        }
        rcu_read_unlock();
 
 {
        struct rt6_info *rt;
 
-       for (rt = w->leaf; rt; rt = rt->dst.rt6_next)
+       for_each_fib6_walker_rt(w)
                fib6_rt_dump(rt, w->args);
        w->leaf = NULL;
        return 0;
                            struct fib6_walker *w)
 {
        w->root = &tb->tb6_root;
-       read_lock_bh(&tb->tb6_lock);
+       spin_lock_bh(&tb->tb6_lock);
        fib6_walk(net, w);
-       read_unlock_bh(&tb->tb6_lock);
+       spin_unlock_bh(&tb->tb6_lock);
 }
 
 /* Called with rcu_read_lock() */
        int res;
        struct rt6_info *rt;
 
-       for (rt = w->leaf; rt; rt = rt->dst.rt6_next) {
+       for_each_fib6_walker_rt(w) {
                res = rt6_dump_route(rt, w->args);
                if (res < 0) {
                        /* Frame is full, suspend walking */
                w->count = 0;
                w->skip = 0;
 
-               read_lock_bh(&table->tb6_lock);
+               spin_lock_bh(&table->tb6_lock);
                res = fib6_walk(net, w);
-               read_unlock_bh(&table->tb6_lock);
+               spin_unlock_bh(&table->tb6_lock);
                if (res > 0) {
                        cb->args[4] = 1;
                        cb->args[5] = w->root->fn_sernum;
                } else
                        w->skip = 0;
 
-               read_lock_bh(&table->tb6_lock);
+               spin_lock_bh(&table->tb6_lock);
                res = fib6_walk_continue(w);
-               read_unlock_bh(&table->tb6_lock);
+               spin_unlock_bh(&table->tb6_lock);
                if (res <= 0) {
                        fib6_walker_unlink(net, w);
                        cb->args[4] = 0;
  *     node.
  */
 
-static struct fib6_node *fib6_add_1(struct fib6_node *root,
-                                    struct in6_addr *addr, int plen,
-                                    int offset, int allow_create,
-                                    int replace_required,
-                                    struct netlink_ext_ack *extack)
+static struct fib6_node *fib6_add_1(struct fib6_table *table,
+                                   struct fib6_node *root,
+                                   struct in6_addr *addr, int plen,
+                                   int offset, int allow_create,
+                                   int replace_required,
+                                   struct netlink_ext_ack *extack)
 {
        struct fib6_node *fn, *in, *ln;
        struct fib6_node *pn = NULL;
        fn = root;
 
        do {
-               key = (struct rt6key *)((u8 *)fn->leaf + offset);
+               struct rt6_info *leaf = rcu_dereference_protected(fn->leaf,
+                                           lockdep_is_held(&table->tb6_lock));
+               key = (struct rt6key *)((u8 *)leaf + offset);
 
                /*
                 *      Prefix match
                if (plen == fn->fn_bit) {
                        /* clean up an intermediate node */
                        if (!(fn->fn_flags & RTN_RTINFO)) {
-                               rt6_release(fn->leaf);
-                               fn->leaf = NULL;
+                               RCU_INIT_POINTER(fn->leaf, NULL);
+                               rt6_release(leaf);
                        }
 
                        return fn;
                /* Try to walk down on tree. */
                dir = addr_bit_set(addr, fn->fn_bit);
                pn = fn;
-               fn = dir ? fn->right : fn->left;
+               fn = dir ?
+                    rcu_dereference_protected(fn->right,
+                                       lockdep_is_held(&table->tb6_lock)) :
+                    rcu_dereference_protected(fn->left,
+                                       lockdep_is_held(&table->tb6_lock));
        } while (fn);
 
        if (!allow_create) {
        if (!ln)
                return ERR_PTR(-ENOMEM);
        ln->fn_bit = plen;
-
-       ln->parent = pn;
+       RCU_INIT_POINTER(ln->parent, pn);
 
        if (dir)
-               pn->right = ln;
+               rcu_assign_pointer(pn->right, ln);
        else
-               pn->left  = ln;
+               rcu_assign_pointer(pn->left, ln);
 
        return ln;
 
         * and the current
         */
 
-       pn = fn->parent;
+       pn = rcu_dereference_protected(fn->parent,
+                                      lockdep_is_held(&table->tb6_lock));
 
        /* find 1st bit in difference between the 2 addrs.
 
 
                in->fn_bit = bit;
 
-               in->parent = pn;
+               RCU_INIT_POINTER(in->parent, pn);
                in->leaf = fn->leaf;
-               atomic_inc(&in->leaf->rt6i_ref);
+               atomic_inc(&rcu_dereference_protected(in->leaf,
+                               lockdep_is_held(&table->tb6_lock))->rt6i_ref);
 
                /* update parent pointer */
                if (dir)
-                       pn->right = in;
+                       rcu_assign_pointer(pn->right, in);
                else
-                       pn->left  = in;
+                       rcu_assign_pointer(pn->left, in);
 
                ln->fn_bit = plen;
 
-               ln->parent = in;
-               fn->parent = in;
+               RCU_INIT_POINTER(ln->parent, in);
+               rcu_assign_pointer(fn->parent, in);
 
                if (addr_bit_set(addr, bit)) {
-                       in->right = ln;
-                       in->left  = fn;
+                       rcu_assign_pointer(in->right, ln);
+                       rcu_assign_pointer(in->left, fn);
                } else {
-                       in->left  = ln;
-                       in->right = fn;
+                       rcu_assign_pointer(in->left, ln);
+                       rcu_assign_pointer(in->right, fn);
                }
        } else { /* plen <= bit */
 
 
                ln->fn_bit = plen;
 
-               ln->parent = pn;
-
-               if (dir)
-                       pn->right = ln;
-               else
-                       pn->left  = ln;
+               RCU_INIT_POINTER(ln->parent, pn);
 
                if (addr_bit_set(&key->addr, plen))
-                       ln->right = fn;
+                       RCU_INIT_POINTER(ln->right, fn);
                else
-                       ln->left  = fn;
+                       RCU_INIT_POINTER(ln->left, fn);
+
+               rcu_assign_pointer(fn->parent, ln);
 
-               fn->parent = ln;
+               if (dir)
+                       rcu_assign_pointer(pn->right, ln);
+               else
+                       rcu_assign_pointer(pn->left, ln);
        }
        return ln;
 }
 static void fib6_purge_rt(struct rt6_info *rt, struct fib6_node *fn,
                          struct net *net)
 {
+       struct fib6_table *table = rt->rt6i_table;
+
        if (atomic_read(&rt->rt6i_ref) != 1) {
                /* This route is used as dummy address holder in some split
                 * nodes. It is not leaked, but it still holds other resources,
                 * to still alive ones.
                 */
                while (fn) {
-                       if (!(fn->fn_flags & RTN_RTINFO) && fn->leaf == rt) {
-                               fn->leaf = fib6_find_prefix(net, fn);
-                               atomic_inc(&fn->leaf->rt6i_ref);
+                       struct rt6_info *leaf = rcu_dereference_protected(fn->leaf,
+                                           lockdep_is_held(&table->tb6_lock));
+                       struct rt6_info *new_leaf;
+                       if (!(fn->fn_flags & RTN_RTINFO) && leaf == rt) {
+                               new_leaf = fib6_find_prefix(net, table, fn);
+                               atomic_inc(&new_leaf->rt6i_ref);
+                               rcu_assign_pointer(fn->leaf, new_leaf);
                                rt6_release(rt);
                        }
-                       fn = fn->parent;
+                       fn = rcu_dereference_protected(fn->parent,
+                                   lockdep_is_held(&table->tb6_lock));
                }
        }
 }
 static int fib6_add_rt2node(struct fib6_node *fn, struct rt6_info *rt,
                            struct nl_info *info, struct mx6_config *mxc)
 {
+       struct rt6_info *leaf = rcu_dereference_protected(fn->leaf,
+                                   lockdep_is_held(&rt->rt6i_table->tb6_lock));
        struct rt6_info *iter = NULL;
-       struct rt6_info **ins;
-       struct rt6_info **fallback_ins = NULL;
+       struct rt6_info __rcu **ins;
+       struct rt6_info __rcu **fallback_ins = NULL;
        int replace = (info->nlh &&
                       (info->nlh->nlmsg_flags & NLM_F_REPLACE));
        int add = (!info->nlh ||
 
        ins = &fn->leaf;
 
-       for (iter = fn->leaf; iter; iter = iter->dst.rt6_next) {
+       for (iter = leaf; iter;
+            iter = rcu_dereference_protected(iter->dst.rt6_next,
+                               lockdep_is_held(&rt->rt6i_table->tb6_lock))) {
                /*
                 *      Search for duplicates
                 */
        if (fallback_ins && !found) {
                /* No ECMP-able route found, replace first non-ECMP one */
                ins = fallback_ins;
-               iter = *ins;
+               iter = rcu_dereference_protected(*ins,
+                                   lockdep_is_held(&rt->rt6i_table->tb6_lock));
                found++;
        }
 
                struct rt6_info *sibling, *temp_sibling;
 
                /* Find the first route that have the same metric */
-               sibling = fn->leaf;
+               sibling = leaf;
                while (sibling) {
                        if (sibling->rt6i_metric == rt->rt6i_metric &&
                            rt6_qualify_for_ecmp(sibling)) {
                                              &sibling->rt6i_siblings);
                                break;
                        }
-                       sibling = sibling->dst.rt6_next;
+                       sibling = rcu_dereference_protected(sibling->dst.rt6_next,
+                                   lockdep_is_held(&rt->rt6i_table->tb6_lock));
                }
                /* For each sibling in the list, increment the counter of
                 * siblings. BUG() if counters does not match, list of siblings
                if (err)
                        return err;
 
-               rt->dst.rt6_next = iter;
-               *ins = rt;
-               rcu_assign_pointer(rt->rt6i_node, fn);
+               rcu_assign_pointer(rt->dst.rt6_next, iter);
                atomic_inc(&rt->rt6i_ref);
+               rcu_assign_pointer(rt->rt6i_node, fn);
+               rcu_assign_pointer(*ins, rt);
                call_fib6_entry_notifiers(info->nl_net, FIB_EVENT_ENTRY_ADD,
                                          rt);
                if (!info->skip_notify)
                if (err)
                        return err;
 
-               *ins = rt;
+               atomic_inc(&rt->rt6i_ref);
                rcu_assign_pointer(rt->rt6i_node, fn);
                rt->dst.rt6_next = iter->dst.rt6_next;
-               atomic_inc(&rt->rt6i_ref);
+               rcu_assign_pointer(*ins, rt);
                call_fib6_entry_notifiers(info->nl_net, FIB_EVENT_ENTRY_REPLACE,
                                          rt);
                if (!info->skip_notify)
                nsiblings = iter->rt6i_nsiblings;
                iter->rt6i_node = NULL;
                fib6_purge_rt(iter, fn, info->nl_net);
-               if (fn->rr_ptr == iter)
+               if (rcu_access_pointer(fn->rr_ptr) == iter)
                        fn->rr_ptr = NULL;
                rt6_release(iter);
 
                if (nsiblings) {
                        /* Replacing an ECMP route, remove all siblings */
                        ins = &rt->dst.rt6_next;
-                       iter = *ins;
+                       iter = rcu_dereference_protected(*ins,
+                                   lockdep_is_held(&rt->rt6i_table->tb6_lock));
                        while (iter) {
                                if (iter->rt6i_metric > rt->rt6i_metric)
                                        break;
                                        *ins = iter->dst.rt6_next;
                                        iter->rt6i_node = NULL;
                                        fib6_purge_rt(iter, fn, info->nl_net);
-                                       if (fn->rr_ptr == iter)
+                                       if (rcu_access_pointer(fn->rr_ptr) == iter)
                                                fn->rr_ptr = NULL;
                                        rt6_release(iter);
                                        nsiblings--;
                                } else {
                                        ins = &iter->dst.rt6_next;
                                }
-                               iter = *ins;
+                               iter = rcu_dereference_protected(*ins,
+                                       lockdep_is_held(&rt->rt6i_table->tb6_lock));
                        }
                        WARN_ON(nsiblings != 0);
                }
        smp_wmb();
        while (fn) {
                fn->fn_sernum = sernum;
-               fn = fn->parent;
+               fn = rcu_dereference_protected(fn->parent,
+                               lockdep_is_held(&rt->rt6i_table->tb6_lock));
        }
 }
 
  *     Add routing information to the routing tree.
  *     <destination addr>/<source addr>
  *     with source addr info in sub-trees
+ *     Need to own table->tb6_lock
  */
 
 int fib6_add(struct fib6_node *root, struct rt6_info *rt,
             struct nl_info *info, struct mx6_config *mxc,
             struct netlink_ext_ack *extack)
 {
+       struct fib6_table *table = rt->rt6i_table;
        struct fib6_node *fn, *pn = NULL;
        int err = -ENOMEM;
        int allow_create = 1;
        if (!allow_create && !replace_required)
                pr_warn("RTM_NEWROUTE with no NLM_F_CREATE or NLM_F_REPLACE\n");
 
-       fn = fib6_add_1(root, &rt->rt6i_dst.addr, rt->rt6i_dst.plen,
+       fn = fib6_add_1(table, root,
+                       &rt->rt6i_dst.addr, rt->rt6i_dst.plen,
                        offsetof(struct rt6_info, rt6i_dst), allow_create,
                        replace_required, extack);
        if (IS_ERR(fn)) {
        if (rt->rt6i_src.plen) {
                struct fib6_node *sn;
 
-               if (!fn->subtree) {
+               if (!rcu_access_pointer(fn->subtree)) {
                        struct fib6_node *sfn;
 
                        /*
                        if (!sfn)
                                goto failure;
 
-                       sfn->leaf = info->nl_net->ipv6.ip6_null_entry;
                        atomic_inc(&info->nl_net->ipv6.ip6_null_entry->rt6i_ref);
+                       rcu_assign_pointer(sfn->leaf,
+                                          info->nl_net->ipv6.ip6_null_entry);
                        sfn->fn_flags = RTN_ROOT;
 
                        /* Now add the first leaf node to new subtree */
 
-                       sn = fib6_add_1(sfn, &rt->rt6i_src.addr,
+                       sn = fib6_add_1(table, sfn, &rt->rt6i_src.addr,
                                        rt->rt6i_src.plen,
                                        offsetof(struct rt6_info, rt6i_src),
                                        allow_create, replace_required, extack);
                        }
 
                        /* Now link new subtree to main tree */
-                       sfn->parent = fn;
-                       fn->subtree = sfn;
+                       rcu_assign_pointer(sfn->parent, fn);
+                       rcu_assign_pointer(fn->subtree, sfn);
                } else {
-                       sn = fib6_add_1(fn->subtree, &rt->rt6i_src.addr,
+                       sn = fib6_add_1(table, FIB6_SUBTREE(fn), &rt->rt6i_src.addr,
                                        rt->rt6i_src.plen,
                                        offsetof(struct rt6_info, rt6i_src),
                                        allow_create, replace_required, extack);
                        }
                }
 
-               if (!fn->leaf) {
-                       fn->leaf = rt;
+               if (!rcu_access_pointer(fn->leaf)) {
                        atomic_inc(&rt->rt6i_ref);
+                       rcu_assign_pointer(fn->leaf, rt);
                }
                fn = sn;
        }
                 * If fib6_add_1 has cleared the old leaf pointer in the
                 * super-tree leaf node we have to find a new one for it.
                 */
-               if (pn != fn && pn->leaf == rt) {
-                       pn->leaf = NULL;
+               struct rt6_info *pn_leaf = rcu_dereference_protected(pn->leaf,
+                                           lockdep_is_held(&table->tb6_lock));
+               if (pn != fn && pn_leaf == rt) {
+                       pn_leaf = NULL;
+                       RCU_INIT_POINTER(pn->leaf, NULL);
                        atomic_dec(&rt->rt6i_ref);
                }
-               if (pn != fn && !pn->leaf && !(pn->fn_flags & RTN_RTINFO)) {
-                       pn->leaf = fib6_find_prefix(info->nl_net, pn);
+               if (pn != fn && !pn_leaf && !(pn->fn_flags & RTN_RTINFO)) {
+                       pn_leaf = fib6_find_prefix(info->nl_net, table, pn);
 #if RT6_DEBUG >= 2
-                       if (!pn->leaf) {
-                               WARN_ON(pn->leaf == NULL);
-                               pn->leaf = info->nl_net->ipv6.ip6_null_entry;
+                       if (!pn_leaf) {
+                               WARN_ON(!pn_leaf);
+                               pn_leaf = info->nl_net->ipv6.ip6_null_entry;
                        }
 #endif
-                       atomic_inc(&pn->leaf->rt6i_ref);
+                       atomic_inc(&pn_leaf->rt6i_ref);
+                       rcu_assign_pointer(pn->leaf, pn_leaf);
                }
 #endif
                goto failure;
         * fn->leaf.
         */
        if (fn && !(fn->fn_flags & (RTN_RTINFO|RTN_ROOT)))
-               fib6_repair_tree(info->nl_net, fn);
+               fib6_repair_tree(info->nl_net, table, fn);
        /* Always release dst as dst->__refcnt is guaranteed
         * to be taken before entering this function
         */
 
                dir = addr_bit_set(args->addr, fn->fn_bit);
 
-               next = dir ? fn->right : fn->left;
+               next = dir ? rcu_dereference(fn->right) :
+                            rcu_dereference(fn->left);
 
                if (next) {
                        fn = next;
        }
 
        while (fn) {
-               if (FIB6_SUBTREE(fn) || fn->fn_flags & RTN_RTINFO) {
-                       struct rt6_info *leaf = fn->leaf;
+               struct fib6_node *subtree = FIB6_SUBTREE(fn);
+
+               if (subtree || fn->fn_flags & RTN_RTINFO) {
+                       struct rt6_info *leaf = rcu_dereference(fn->leaf);
                        struct rt6key *key;
 
                        if (!leaf)
 
                        if (ipv6_prefix_equal(&key->addr, args->addr, key->plen)) {
 #ifdef CONFIG_IPV6_SUBTREES
-                               if (fn->subtree) {
+                               if (subtree) {
                                        struct fib6_node *sfn;
-                                       sfn = fib6_lookup_1(fn->subtree,
-                                                           args + 1);
+                                       sfn = fib6_lookup_1(subtree, args + 1);
                                        if (!sfn)
                                                goto backtrack;
                                        fn = sfn;
                if (fn->fn_flags & RTN_ROOT)
                        break;
 
-               fn = fn->parent;
+               fn = rcu_dereference(fn->parent);
        }
 
        return NULL;
 }
 
+/* called with rcu_read_lock() held
+ */
 struct fib6_node *fib6_lookup(struct fib6_node *root, const struct in6_addr *daddr,
                              const struct in6_addr *saddr)
 {
        struct fib6_node *fn, *prev = NULL;
 
        for (fn = root; fn ; ) {
-               struct rt6_info *leaf = fn->leaf;
+               struct rt6_info *leaf = rcu_dereference(fn->leaf);
                struct rt6key *key;
 
                /* This node is being deleted */
                 *      We have more bits to go
                 */
                if (addr_bit_set(addr, fn->fn_bit))
-                       fn = fn->right;
+                       fn = rcu_dereference(fn->right);
                else
-                       fn = fn->left;
+                       fn = rcu_dereference(fn->left);
        }
 out:
        if (exact_match)
 
 #ifdef CONFIG_IPV6_SUBTREES
        if (src_len) {
+               struct fib6_node *subtree = FIB6_SUBTREE(fn);
+
                WARN_ON(saddr == NULL);
-               if (fn && fn->subtree)
-                       fn = fib6_locate_1(fn->subtree, saddr, src_len,
+               if (fn && subtree)
+                       fn = fib6_locate_1(subtree, saddr, src_len,
                                           offsetof(struct rt6_info, rt6i_src),
                                           exact_match);
        }
  *
  */
 
-static struct rt6_info *fib6_find_prefix(struct net *net, struct fib6_node *fn)
+static struct rt6_info *fib6_find_prefix(struct net *net,
+                                        struct fib6_table *table,
+                                        struct fib6_node *fn)
 {
+       struct fib6_node *child_left, *child_right;
+
        if (fn->fn_flags & RTN_ROOT)
                return net->ipv6.ip6_null_entry;
 
        while (fn) {
-               if (fn->left)
-                       return fn->left->leaf;
-               if (fn->right)
-                       return fn->right->leaf;
+               child_left = rcu_dereference_protected(fn->left,
+                                   lockdep_is_held(&table->tb6_lock));
+               child_right = rcu_dereference_protected(fn->right,
+                                   lockdep_is_held(&table->tb6_lock));
+               if (child_left)
+                       return rcu_dereference_protected(child_left->leaf,
+                                       lockdep_is_held(&table->tb6_lock));
+               if (child_right)
+                       return rcu_dereference_protected(child_right->leaf,
+                                       lockdep_is_held(&table->tb6_lock));
 
                fn = FIB6_SUBTREE(fn);
        }
 /*
  *     Called to trim the tree of intermediate nodes when possible. "fn"
  *     is the node we want to try and remove.
+ *     Need to own table->tb6_lock
  */
 
 static struct fib6_node *fib6_repair_tree(struct net *net,
-                                          struct fib6_node *fn)
+                                         struct fib6_table *table,
+                                         struct fib6_node *fn)
 {
        int children;
        int nstate;
-       struct fib6_node *child, *pn;
+       struct fib6_node *child;
        struct fib6_walker *w;
        int iter = 0;
 
        for (;;) {
+               struct fib6_node *fn_r = rcu_dereference_protected(fn->right,
+                                           lockdep_is_held(&table->tb6_lock));
+               struct fib6_node *fn_l = rcu_dereference_protected(fn->left,
+                                           lockdep_is_held(&table->tb6_lock));
+               struct fib6_node *pn = rcu_dereference_protected(fn->parent,
+                                           lockdep_is_held(&table->tb6_lock));
+               struct fib6_node *pn_r = rcu_dereference_protected(pn->right,
+                                           lockdep_is_held(&table->tb6_lock));
+               struct fib6_node *pn_l = rcu_dereference_protected(pn->left,
+                                           lockdep_is_held(&table->tb6_lock));
+               struct rt6_info *fn_leaf = rcu_dereference_protected(fn->leaf,
+                                           lockdep_is_held(&table->tb6_lock));
+               struct rt6_info *pn_leaf = rcu_dereference_protected(pn->leaf,
+                                           lockdep_is_held(&table->tb6_lock));
+               struct rt6_info *new_fn_leaf;
+
                RT6_TRACE("fixing tree: plen=%d iter=%d\n", fn->fn_bit, iter);
                iter++;
 
                WARN_ON(fn->fn_flags & RTN_RTINFO);
                WARN_ON(fn->fn_flags & RTN_TL_ROOT);
-               WARN_ON(fn->leaf);
+               WARN_ON(fn_leaf);
 
                children = 0;
                child = NULL;
-               if (fn->right)
-                       child = fn->right, children |= 1;
-               if (fn->left)
-                       child = fn->left, children |= 2;
+               if (fn_r)
+                       child = fn_r, children |= 1;
+               if (fn_l)
+                       child = fn_l, children |= 2;
 
                if (children == 3 || FIB6_SUBTREE(fn)
 #ifdef CONFIG_IPV6_SUBTREES
                    || (children && fn->fn_flags & RTN_ROOT)
 #endif
                    ) {
-                       fn->leaf = fib6_find_prefix(net, fn);
+                       new_fn_leaf = fib6_find_prefix(net, table, fn);
 #if RT6_DEBUG >= 2
-                       if (!fn->leaf) {
-                               WARN_ON(!fn->leaf);
-                               fn->leaf = net->ipv6.ip6_null_entry;
+                       if (!new_fn_leaf) {
+                               WARN_ON(!new_fn_leaf);
+                               new_fn_leaf = net->ipv6.ip6_null_entry;
                        }
 #endif
-                       atomic_inc(&fn->leaf->rt6i_ref);
-                       return fn->parent;
+                       atomic_inc(&new_fn_leaf->rt6i_ref);
+                       rcu_assign_pointer(fn->leaf, new_fn_leaf);
+                       return pn;
                }
 
-               pn = fn->parent;
 #ifdef CONFIG_IPV6_SUBTREES
                if (FIB6_SUBTREE(pn) == fn) {
                        WARN_ON(!(fn->fn_flags & RTN_ROOT));
-                       FIB6_SUBTREE(pn) = NULL;
+                       RCU_INIT_POINTER(pn->subtree, NULL);
                        nstate = FWS_L;
                } else {
                        WARN_ON(fn->fn_flags & RTN_ROOT);
 #endif
-                       if (pn->right == fn)
-                               pn->right = child;
-                       else if (pn->left == fn)
-                               pn->left = child;
+                       if (pn_r == fn)
+                               rcu_assign_pointer(pn->right, child);
+                       else if (pn_l == fn)
+                               rcu_assign_pointer(pn->left, child);
 #if RT6_DEBUG >= 2
                        else
                                WARN_ON(1);
 #endif
                        if (child)
-                               child->parent = pn;
+                               rcu_assign_pointer(child->parent, pn);
                        nstate = FWS_R;
 #ifdef CONFIG_IPV6_SUBTREES
                }
                if (pn->fn_flags & RTN_RTINFO || FIB6_SUBTREE(pn))
                        return pn;
 
-               rt6_release(pn->leaf);
-               pn->leaf = NULL;
+               RCU_INIT_POINTER(pn->leaf, NULL);
+               rt6_release(pn_leaf);
                fn = pn;
        }
 }
 
-static void fib6_del_route(struct fib6_node *fn, struct rt6_info **rtp,
-                          struct nl_info *info)
+static void fib6_del_route(struct fib6_table *table, struct fib6_node *fn,
+                          struct rt6_info __rcu **rtp, struct nl_info *info)
 {
        struct fib6_walker *w;
-       struct rt6_info *rt = *rtp;
+       struct rt6_info *rt = rcu_dereference_protected(*rtp,
+                                   lockdep_is_held(&table->tb6_lock));
        struct net *net = info->nl_net;
 
        RT6_TRACE("fib6_del_route\n");
        rt6_flush_exceptions(rt);
 
        /* Reset round-robin state, if necessary */
-       if (fn->rr_ptr == rt)
+       if (rcu_access_pointer(fn->rr_ptr) == rt)
                fn->rr_ptr = NULL;
 
        /* Remove this entry from other siblings */
        FOR_WALKERS(net, w) {
                if (w->state == FWS_C && w->leaf == rt) {
                        RT6_TRACE("walker %p adjusted by delroute\n", w);
-                       w->leaf = rt->dst.rt6_next;
+                       w->leaf = rcu_dereference_protected(rt->dst.rt6_next,
+                                           lockdep_is_held(&table->tb6_lock));
                        if (!w->leaf)
                                w->state = FWS_U;
                }
        }
        read_unlock(&net->ipv6.fib6_walker_lock);
 
-       rt->dst.rt6_next = NULL;
-
        /* If it was last route, expunge its radix tree node */
-       if (!fn->leaf) {
+       if (!rcu_access_pointer(fn->leaf)) {
                fn->fn_flags &= ~RTN_RTINFO;
                net->ipv6.rt6_stats->fib_route_nodes--;
-               fn = fib6_repair_tree(net, fn);
+               fn = fib6_repair_tree(net, table, fn);
        }
 
        fib6_purge_rt(rt, fn, net);
        rt6_release(rt);
 }
 
+/* Need to own table->tb6_lock */
 int fib6_del(struct rt6_info *rt, struct nl_info *info)
 {
        struct fib6_node *fn = rcu_dereference_protected(rt->rt6i_node,
                                    lockdep_is_held(&rt->rt6i_table->tb6_lock));
+       struct fib6_table *table = rt->rt6i_table;
        struct net *net = info->nl_net;
-       struct rt6_info **rtp;
+       struct rt6_info __rcu **rtp;
+       struct rt6_info __rcu **rtp_next;
 
 #if RT6_DEBUG >= 2
        if (rt->dst.obsolete > 0) {
         *      Walk the leaf entries looking for ourself
         */
 
-       for (rtp = &fn->leaf; *rtp; rtp = &(*rtp)->dst.rt6_next) {
-               if (*rtp == rt) {
-                       fib6_del_route(fn, rtp, info);
+       for (rtp = &fn->leaf; *rtp; rtp = rtp_next) {
+               struct rt6_info *cur = rcu_dereference_protected(*rtp,
+                                       lockdep_is_held(&table->tb6_lock));
+               if (rt == cur) {
+                       fib6_del_route(table, fn, rtp, info);
                        return 0;
                }
+               rtp_next = &cur->dst.rt6_next;
        }
        return -ENOENT;
 }
  *     0   -> walk is complete.
  *     >0  -> walk is incomplete (i.e. suspended)
  *     <0  -> walk is terminated by an error.
+ *
+ *     This function is called with tb6_lock held.
  */
 
 static int fib6_walk_continue(struct fib6_walker *w)
 {
-       struct fib6_node *fn, *pn;
+       struct fib6_node *fn, *pn, *left, *right;
 
        /* w->root should always be table->tb6_root */
        WARN_ON_ONCE(!(w->root->fn_flags & RTN_TL_ROOT));
                        w->state = FWS_L;
 #endif
                case FWS_L:
-                       if (fn->left) {
-                               w->node = fn->left;
+                       left = rcu_dereference_protected(fn->left, 1);
+                       if (left) {
+                               w->node = left;
                                w->state = FWS_INIT;
                                continue;
                        }
                        w->state = FWS_R;
                case FWS_R:
-                       if (fn->right) {
-                               w->node = fn->right;
+                       right = rcu_dereference_protected(fn->right, 1);
+                       if (right) {
+                               w->node = right;
                                w->state = FWS_INIT;
                                continue;
                        }
                        w->state = FWS_C;
-                       w->leaf = fn->leaf;
+                       w->leaf = rcu_dereference_protected(fn->leaf, 1);
                case FWS_C:
                        if (w->leaf && fn->fn_flags & RTN_RTINFO) {
                                int err;
                case FWS_U:
                        if (fn == w->root)
                                return 0;
-                       pn = fn->parent;
+                       pn = rcu_dereference_protected(fn->parent, 1);
+                       left = rcu_dereference_protected(pn->left, 1);
+                       right = rcu_dereference_protected(pn->right, 1);
                        w->node = pn;
 #ifdef CONFIG_IPV6_SUBTREES
                        if (FIB6_SUBTREE(pn) == fn) {
                                continue;
                        }
 #endif
-                       if (pn->left == fn) {
+                       if (left == fn) {
                                w->state = FWS_R;
                                continue;
                        }
-                       if (pn->right == fn) {
+                       if (right == fn) {
                                w->state = FWS_C;
-                               w->leaf = w->node->leaf;
+                               w->leaf = rcu_dereference_protected(w->node->leaf, 1);
                                continue;
                        }
 #if RT6_DEBUG >= 2
                return 0;
        }
 
-       for (rt = w->leaf; rt; rt = rt->dst.rt6_next) {
+       for_each_fib6_walker_rt(w) {
                res = c->func(rt, c->arg);
                if (res < 0) {
                        w->leaf = rt;
        for (h = 0; h < FIB6_TABLE_HASHSZ; h++) {
                head = &net->ipv6.fib_table_hash[h];
                hlist_for_each_entry_rcu(table, head, tb6_hlist) {
-                       write_lock_bh(&table->tb6_lock);
+                       spin_lock_bh(&table->tb6_lock);
                        fib6_clean_tree(net, &table->tb6_root,
                                        func, sernum, arg);
-                       write_unlock_bh(&table->tb6_lock);
+                       spin_unlock_bh(&table->tb6_lock);
                }
        }
        rcu_read_unlock();
                goto out_fib_table_hash;
 
        net->ipv6.fib6_main_tbl->tb6_id = RT6_TABLE_MAIN;
-       net->ipv6.fib6_main_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
+       rcu_assign_pointer(net->ipv6.fib6_main_tbl->tb6_root.leaf,
+                          net->ipv6.ip6_null_entry);
        net->ipv6.fib6_main_tbl->tb6_root.fn_flags =
                RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
        inet_peer_base_init(&net->ipv6.fib6_main_tbl->tb6_peers);
        if (!net->ipv6.fib6_local_tbl)
                goto out_fib6_main_tbl;
        net->ipv6.fib6_local_tbl->tb6_id = RT6_TABLE_LOCAL;
-       net->ipv6.fib6_local_tbl->tb6_root.leaf = net->ipv6.ip6_null_entry;
+       rcu_assign_pointer(net->ipv6.fib6_local_tbl->tb6_root.leaf,
+                          net->ipv6.ip6_null_entry);
        net->ipv6.fib6_local_tbl->tb6_root.fn_flags =
                RTN_ROOT | RTN_TL_ROOT | RTN_RTINFO;
        inet_peer_base_init(&net->ipv6.fib6_local_tbl->tb6_peers);
                return 1;
 
        do {
-               iter->w.leaf = iter->w.leaf->dst.rt6_next;
+               iter->w.leaf = rcu_dereference_protected(
+                               iter->w.leaf->dst.rt6_next,
+                               lockdep_is_held(&iter->tbl->tb6_lock));
                iter->skip--;
                if (!iter->skip && iter->w.leaf)
                        return 1;
        if (!v)
                goto iter_table;
 
-       n = ((struct rt6_info *)v)->dst.rt6_next;
+       n = rcu_dereference(((struct rt6_info *)v)->dst.rt6_next);
        if (n) {
                ++*pos;
                return n;
 
 iter_table:
        ipv6_route_check_sernum(iter);
-       read_lock(&iter->tbl->tb6_lock);
+       spin_lock_bh(&iter->tbl->tb6_lock);
        r = fib6_walk_continue(&iter->w);
-       read_unlock(&iter->tbl->tb6_lock);
+       spin_unlock_bh(&iter->tbl->tb6_lock);
        if (r > 0) {
                if (v)
                        ++*pos;
 
 }
 
 /*
- *     Route lookup. Any table->tb6_lock is implied.
+ *     Route lookup. rcu_read_lock() should be held.
  */
 
 static inline struct rt6_info *rt6_device_match(struct net *net,
        if (!oif && ipv6_addr_any(saddr))
                goto out;
 
-       for (sprt = rt; sprt; sprt = sprt->dst.rt6_next) {
+       for (sprt = rt; sprt; sprt = rcu_dereference(sprt->dst.rt6_next)) {
                struct net_device *dev = sprt->dst.dev;
 
                if (oif) {
 
        match = NULL;
        cont = NULL;
-       for (rt = rr_head; rt; rt = rt->dst.rt6_next) {
+       for (rt = rr_head; rt; rt = rcu_dereference(rt->dst.rt6_next)) {
                if (rt->rt6i_metric != metric) {
                        cont = rt;
                        break;
                match = find_match(rt, oif, strict, &mpri, match, do_rr);
        }
 
-       for (rt = leaf; rt && rt != rr_head; rt = rt->dst.rt6_next) {
+       for (rt = leaf; rt && rt != rr_head;
+            rt = rcu_dereference(rt->dst.rt6_next)) {
                if (rt->rt6i_metric != metric) {
                        cont = rt;
                        break;
        if (match || !cont)
                return match;
 
-       for (rt = cont; rt; rt = rt->dst.rt6_next)
+       for (rt = cont; rt; rt = rcu_dereference(rt->dst.rt6_next))
                match = find_match(rt, oif, strict, &mpri, match, do_rr);
 
        return match;
 static struct rt6_info *rt6_select(struct net *net, struct fib6_node *fn,
                                   int oif, int strict)
 {
-       struct rt6_info *leaf = fn->leaf;
+       struct rt6_info *leaf = rcu_dereference(fn->leaf);
        struct rt6_info *match, *rt0;
        bool do_rr = false;
        int key_plen;
        if (!leaf)
                return net->ipv6.ip6_null_entry;
 
-       rt0 = fn->rr_ptr;
+       rt0 = rcu_dereference(fn->rr_ptr);
        if (!rt0)
-               fn->rr_ptr = rt0 = leaf;
+               rt0 = leaf;
 
        /* Double check to make sure fn is not an intermediate node
         * and fn->leaf does not points to its child's leaf
                             &do_rr);
 
        if (do_rr) {
-               struct rt6_info *next = rt0->dst.rt6_next;
+               struct rt6_info *next = rcu_dereference(rt0->dst.rt6_next);
 
                /* no entries matched; do round-robin */
                if (!next || next->rt6i_metric != rt0->rt6i_metric)
                        next = leaf;
 
-               if (next != rt0)
-                       fn->rr_ptr = next;
+               if (next != rt0) {
+                       spin_lock_bh(&leaf->rt6i_table->tb6_lock);
+                       /* make sure next is not being deleted from the tree */
+                       if (next->rt6i_node)
+                               rcu_assign_pointer(fn->rr_ptr, next);
+                       spin_unlock_bh(&leaf->rt6i_table->tb6_lock);
+               }
        }
 
        return match ? match : net->ipv6.ip6_null_entry;
 static struct fib6_node* fib6_backtrack(struct fib6_node *fn,
                                        struct in6_addr *saddr)
 {
-       struct fib6_node *pn;
+       struct fib6_node *pn, *sn;
        while (1) {
                if (fn->fn_flags & RTN_TL_ROOT)
                        return NULL;
-               pn = fn->parent;
-               if (FIB6_SUBTREE(pn) && FIB6_SUBTREE(pn) != fn)
-                       fn = fib6_lookup(FIB6_SUBTREE(pn), NULL, saddr);
+               pn = rcu_dereference(fn->parent);
+               sn = FIB6_SUBTREE(pn);
+               if (sn && sn != fn)
+                       fn = fib6_lookup(sn, NULL, saddr);
                else
                        fn = pn;
                if (fn->fn_flags & RTN_RTINFO)
        struct rt6_info *rt, *rt_cache;
        struct fib6_node *fn;
 
-       read_lock_bh(&table->tb6_lock);
+       rcu_read_lock();
        fn = fib6_lookup(&table->tb6_root, &fl6->daddr, &fl6->saddr);
 restart:
-       rt = fn->leaf;
-       rt = rt6_device_match(net, rt, &fl6->saddr, fl6->flowi6_oif, flags);
-       if (rt->rt6i_nsiblings && fl6->flowi6_oif == 0)
-               rt = rt6_multipath_select(rt, fl6, fl6->flowi6_oif, flags);
+       rt = rcu_dereference(fn->leaf);
+       if (!rt) {
+               rt = net->ipv6.ip6_null_entry;
+       } else {
+               rt = rt6_device_match(net, rt, &fl6->saddr,
+                                     fl6->flowi6_oif, flags);
+               if (rt->rt6i_nsiblings && fl6->flowi6_oif == 0)
+                       rt = rt6_multipath_select(rt, fl6,
+                                                 fl6->flowi6_oif, flags);
+       }
        if (rt == net->ipv6.ip6_null_entry) {
                fn = fib6_backtrack(fn, &fl6->saddr);
                if (fn)
        if (ip6_hold_safe(net, &rt, true))
                dst_use_noref(&rt->dst, jiffies);
 
-       read_unlock_bh(&table->tb6_lock);
+       rcu_read_unlock();
 
        trace_fib6_table_lookup(net, rt, table->tb6_id, fl6);
 
        struct fib6_table *table;
 
        table = rt->rt6i_table;
-       write_lock_bh(&table->tb6_lock);
+       spin_lock_bh(&table->tb6_lock);
        err = fib6_add(&table->tb6_root, rt, info, mxc, extack);
-       write_unlock_bh(&table->tb6_lock);
+       spin_unlock_bh(&table->tb6_lock);
 
        return err;
 }
        return pcpu_rt;
 }
 
-/* It should be called with read_lock_bh(&tb6_lock) acquired */
+/* It should be called with rcu_read_lock() acquired */
 static struct rt6_info *rt6_get_pcpu_route(struct rt6_info *rt)
 {
        struct rt6_info *pcpu_rt, **p;
        if (net->ipv6.devconf_all->forwarding == 0)
                strict |= RT6_LOOKUP_F_REACHABLE;
 
-       read_lock_bh(&table->tb6_lock);
+       rcu_read_lock();
 
        fn = fib6_lookup(&table->tb6_root, &fl6->daddr, &fl6->saddr);
        saved_fn = fn;
                rt = rt_cache;
 
        if (rt == net->ipv6.ip6_null_entry) {
-               read_unlock_bh(&table->tb6_lock);
+               rcu_read_unlock();
                dst_hold(&rt->dst);
                trace_fib6_table_lookup(net, rt, table->tb6_id, fl6);
                return rt;
                        dst_use_noref(&rt->dst, jiffies);
                        rt6_dst_from_metrics_check(rt);
                }
-               read_unlock_bh(&table->tb6_lock);
+               rcu_read_unlock();
                trace_fib6_table_lookup(net, rt, table->tb6_id, fl6);
                return rt;
        } else if (unlikely((fl6->flowi6_flags & FLOWI_FLAG_KNOWN_NH) &&
                if (ip6_hold_safe(net, &rt, true)) {
                        dst_use_noref(&rt->dst, jiffies);
                } else {
-                       read_unlock_bh(&table->tb6_lock);
+                       rcu_read_unlock();
                        uncached_rt = rt;
                        goto uncached_rt_out;
                }
-               read_unlock_bh(&table->tb6_lock);
+               rcu_read_unlock();
 
                uncached_rt = ip6_rt_cache_alloc(rt, &fl6->daddr, NULL);
                dst_release(&rt->dst);
                pcpu_rt = rt6_get_pcpu_route(rt);
 
                if (pcpu_rt) {
-                       read_unlock_bh(&table->tb6_lock);
+                       rcu_read_unlock();
                } else {
                        /* atomic_inc_not_zero() is needed when using rcu */
                        if (atomic_inc_not_zero(&rt->rt6i_ref)) {
                                 * No dst_hold() on rt is needed because grabbing
                                 * rt->rt6i_ref makes sure rt can't be released.
                                 */
-                               read_unlock_bh(&table->tb6_lock);
+                               rcu_read_unlock();
                                pcpu_rt = rt6_make_pcpu_route(rt);
                                rt6_release(rt);
                        } else {
                                /* rt is already removed from tree */
-                               read_unlock_bh(&table->tb6_lock);
+                               rcu_read_unlock();
                                pcpu_rt = net->ipv6.ip6_null_entry;
                                dst_hold(&pcpu_rt->dst);
                        }
         * routes.
         */
 
-       read_lock_bh(&table->tb6_lock);
+       rcu_read_lock();
        fn = fib6_lookup(&table->tb6_root, &fl6->daddr, &fl6->saddr);
 restart:
-       for (rt = fn->leaf; rt; rt = rt->dst.rt6_next) {
+       for_each_fib6_node_rt_rcu(fn) {
                if (rt6_check_expired(rt))
                        continue;
                if (rt->dst.error)
 out:
        ip6_hold_safe(net, &rt, true);
 
-       read_unlock_bh(&table->tb6_lock);
+       rcu_read_unlock();
 
        trace_fib6_table_lookup(net, rt, table->tb6_id, fl6);
        return rt;
        }
 
        table = rt->rt6i_table;
-       write_lock_bh(&table->tb6_lock);
+       spin_lock_bh(&table->tb6_lock);
        err = fib6_del(rt, info);
-       write_unlock_bh(&table->tb6_lock);
+       spin_unlock_bh(&table->tb6_lock);
 
 out:
        ip6_rt_put(rt);
        if (rt == net->ipv6.ip6_null_entry)
                goto out_put;
        table = rt->rt6i_table;
-       write_lock_bh(&table->tb6_lock);
+       spin_lock_bh(&table->tb6_lock);
 
        if (rt->rt6i_nsiblings && cfg->fc_delete_all_nh) {
                struct rt6_info *sibling, *next_sibling;
 
        err = fib6_del(rt, info);
 out_unlock:
-       write_unlock_bh(&table->tb6_lock);
+       spin_unlock_bh(&table->tb6_lock);
 out_put:
        ip6_rt_put(rt);
 
                return err;
        }
 
-       read_lock_bh(&table->tb6_lock);
+       rcu_read_lock();
 
        fn = fib6_locate(&table->tb6_root,
                         &cfg->fc_dst, cfg->fc_dst_len,
                         !(cfg->fc_flags & RTF_CACHE));
 
        if (fn) {
-               for (rt = fn->leaf; rt; rt = rt->dst.rt6_next) {
+               for_each_fib6_node_rt_rcu(fn) {
                        if (cfg->fc_flags & RTF_CACHE) {
                                rt_cache = rt6_find_cached_rt(rt, &cfg->fc_dst,
                                                              &cfg->fc_src);
                                continue;
                        if (!dst_hold_safe(&rt->dst))
                                break;
-                       read_unlock_bh(&table->tb6_lock);
+                       rcu_read_unlock();
 
                        /* if gateway was specified only delete the one hop */
                        if (cfg->fc_flags & RTF_GATEWAY)
                        return __ip6_del_rt_siblings(rt, cfg);
                }
        }
-       read_unlock_bh(&table->tb6_lock);
+       rcu_read_unlock();
 
        return err;
 }
        if (!table)
                return NULL;
 
-       read_lock_bh(&table->tb6_lock);
+       rcu_read_lock();
        fn = fib6_locate(&table->tb6_root, prefix, prefixlen, NULL, 0, true);
        if (!fn)
                goto out;
 
-       for (rt = fn->leaf; rt; rt = rt->dst.rt6_next) {
+       for_each_fib6_node_rt_rcu(fn) {
                if (rt->dst.dev->ifindex != ifindex)
                        continue;
                if ((rt->rt6i_flags & (RTF_ROUTEINFO|RTF_GATEWAY)) != (RTF_ROUTEINFO|RTF_GATEWAY))
                break;
        }
 out:
-       read_unlock_bh(&table->tb6_lock);
+       rcu_read_unlock();
        return rt;
 }
 
        if (!table)
                return NULL;
 
-       read_lock_bh(&table->tb6_lock);
-       for (rt = table->tb6_root.leaf; rt; rt = rt->dst.rt6_next) {
+       rcu_read_lock();
+       for_each_fib6_node_rt_rcu(&table->tb6_root) {
                if (dev == rt->dst.dev &&
                    ((rt->rt6i_flags & (RTF_ADDRCONF | RTF_DEFAULT)) == (RTF_ADDRCONF | RTF_DEFAULT)) &&
                    ipv6_addr_equal(&rt->rt6i_gateway, addr))
        }
        if (rt)
                ip6_hold_safe(NULL, &rt, false);
-       read_unlock_bh(&table->tb6_lock);
+       rcu_read_unlock();
        return rt;
 }
 
        struct rt6_info *rt;
 
 restart:
-       read_lock_bh(&table->tb6_lock);
-       for (rt = table->tb6_root.leaf; rt; rt = rt->dst.rt6_next) {
+       rcu_read_lock();
+       for_each_fib6_node_rt_rcu(&table->tb6_root) {
                if (rt->rt6i_flags & (RTF_DEFAULT | RTF_ADDRCONF) &&
                    (!rt->rt6i_idev || rt->rt6i_idev->cnf.accept_ra != 2)) {
                        if (dst_hold_safe(&rt->dst)) {
-                               read_unlock_bh(&table->tb6_lock);
+                               rcu_read_unlock();
                                ip6_del_rt(rt);
                        } else {
-                               read_unlock_bh(&table->tb6_lock);
+                               rcu_read_unlock();
                        }
                        goto restart;
                }
        }
-       read_unlock_bh(&table->tb6_lock);
+       rcu_read_unlock();
 
        table->flags &= ~RT6_TABLE_HAS_DFLT_ROUTER;
 }