[NETFILTER]: x_tables: pass registered match/target data to match/target functions
[linux-2.6.git] / net / ipv4 / fib_trie.c
index 914a4c2..e320b32 100644 (file)
  *             modify it under the terms of the GNU General Public License
  *             as published by the Free Software Foundation; either version
  *             2 of the License, or (at your option) any later version.
+ *
+ * Substantial contributions to this work comes from:
+ *
+ *             David S. Miller, <davem@davemloft.net>
+ *             Stephen Hemminger <shemminger@osdl.org>
+ *             Paul E. McKenney <paulmck@us.ibm.com>
+ *             Patrick McHardy <kaber@trash.net>
  */
 
-#define VERSION "0.325"
+#define VERSION "0.404"
 
 #include <linux/config.h>
 #include <asm/uaccess.h>
 #include <linux/errno.h>
 #include <linux/in.h>
 #include <linux/inet.h>
+#include <linux/inetdevice.h>
 #include <linux/netdevice.h>
 #include <linux/if_arp.h>
 #include <linux/proc_fs.h>
+#include <linux/rcupdate.h>
 #include <linux/skbuff.h>
 #include <linux/netlink.h>
 #include <linux/init.h>
 #undef CONFIG_IP_FIB_TRIE_STATS
 #define MAX_CHILDS 16384
 
-#define EXTRACT(p, n, str) ((str)<<(p)>>(32-(n)))
 #define KEYLENGTH (8*sizeof(t_key))
 #define MASK_PFX(k, l) (((l)==0)?0:(k >> (KEYLENGTH-l)) << (KEYLENGTH-l))
 #define TKEY_GET_MASK(offset, bits) (((bits)==0)?0:((t_key)(-1) << (KEYLENGTH - bits) >> offset))
 
-static DEFINE_RWLOCK(fib_lock);
-
 typedef unsigned int t_key;
 
 #define T_TNODE 0
 #define T_LEAF  1
 #define NODE_TYPE_MASK 0x1UL
 #define NODE_PARENT(node) \
-       ((struct tnode *)((node)->parent & ~NODE_TYPE_MASK))
-#define NODE_SET_PARENT(node, ptr) \
-       ((node)->parent = (((unsigned long)(ptr)) | \
-                     ((node)->parent & NODE_TYPE_MASK)))
-#define NODE_INIT_PARENT(node, type) \
-       ((node)->parent = (type))
-#define NODE_TYPE(node) \
-       ((node)->parent & NODE_TYPE_MASK)
+       ((struct tnode *)rcu_dereference(((node)->parent & ~NODE_TYPE_MASK)))
+
+#define NODE_TYPE(node) ((node)->parent & NODE_TYPE_MASK)
+
+#define NODE_SET_PARENT(node, ptr)             \
+       rcu_assign_pointer((node)->parent,      \
+                          ((unsigned long)(ptr)) | NODE_TYPE(node))
 
 #define IS_TNODE(n) (!(n->parent & T_LEAF))
 #define IS_LEAF(n) (n->parent & T_LEAF)
@@ -111,10 +116,12 @@ struct leaf {
        t_key key;
        unsigned long parent;
        struct hlist_head list;
+       struct rcu_head rcu;
 };
 
 struct leaf_info {
        struct hlist_node hlist;
+       struct rcu_head rcu;
        int plen;
        struct list_head falh;
 };
@@ -126,6 +133,7 @@ struct tnode {
        unsigned short bits:5;          /* 2log(KEYLENGTH) bits needed */
        unsigned short full_children;   /* KEYLENGTH bits needed */
        unsigned short empty_children;  /* KEYLENGTH bits needed */
+       struct rcu_head rcu;
        struct node *child[0];
 };
 
@@ -158,37 +166,27 @@ struct trie {
        unsigned int revision;
 };
 
-static int trie_debug = 0;
-
-#define DBG(x...) do { if (trie_debug) printk(x); } while (0)
-
-static int tnode_full(struct tnode *tn, struct node *n);
 static void put_child(struct trie *t, struct tnode *tn, int i, struct node *n);
 static void tnode_put_child_reorg(struct tnode *tn, int i, struct node *n, int wasfull);
-static int tnode_child_length(struct tnode *tn);
 static struct node *resize(struct trie *t, struct tnode *tn);
 static struct tnode *inflate(struct trie *t, struct tnode *tn);
 static struct tnode *halve(struct trie *t, struct tnode *tn);
 static void tnode_free(struct tnode *tn);
-static void trie_dump_seq(struct seq_file *seq, struct trie *t);
-extern struct fib_alias *fib_find_alias(struct list_head *fah, u8 tos, u32 prio);
-extern int fib_detect_death(struct fib_info *fi, int order,
-                           struct fib_info **last_resort, int *last_idx, int *dflt);
 
-extern void rtmsg_fib(int event, u32 key, struct fib_alias *fa, int z, int tb_id,
-                     struct nlmsghdr *n, struct netlink_skb_parms *req);
-
-static kmem_cache_t *fn_alias_kmem;
+static kmem_cache_t *fn_alias_kmem __read_mostly;
 static struct trie *trie_local = NULL, *trie_main = NULL;
 
+
+/* rcu_read_lock needs to be hold by caller from readside */
+
 static inline struct node *tnode_get_child(struct tnode *tn, int i)
 {
        BUG_ON(i >= 1 << tn->bits);
 
-       return tn->child[i];
+       return rcu_dereference(tn->child[i]);
 }
 
-static inline int tnode_child_length(struct tnode *tn)
+static inline int tnode_child_length(const struct tnode *tn)
 {
        return 1 << tn->bits;
 }
@@ -226,14 +224,6 @@ static inline int tkey_mismatch(t_key a, int offset, t_key b)
        return i;
 }
 
-/* Candidate for fib_semantics */
-
-static void fn_free_alias(struct fib_alias *fa)
-{
-       fib_release_info(fa->fa_info);
-       kmem_cache_free(fn_alias_kmem, fa);
-}
-
 /*
   To understand this stuff, an understanding of keys and all their bits is 
   necessary. Every node in the trie has a key associated with it, but not 
@@ -242,7 +232,7 @@ static void fn_free_alias(struct fib_alias *fa)
   Consider a node 'n' and its parent 'tp'.
 
   If n is a leaf, every bit in its key is significant. Its presence is 
-  necessitaded by path compression, since during a tree traversal (when 
+  necessitated by path compression, since during a tree traversal (when 
   searching for a leaf - unless we are doing an insertion) we will completely 
   ignore all skipped bits we encounter. Thus we need to verify, at the end of 
   a potentially successful search, that we have indeed been walking the 
@@ -297,63 +287,67 @@ static void fn_free_alias(struct fib_alias *fa)
 
 */
 
-static void check_tnode(struct tnode *tn)
+static inline void check_tnode(const struct tnode *tn)
 {
-       if (tn && tn->pos+tn->bits > 32) {
-               printk("TNODE ERROR tn=%p, pos=%d, bits=%d\n", tn, tn->pos, tn->bits);
-       }
+       WARN_ON(tn && tn->pos+tn->bits > 32);
 }
 
 static int halve_threshold = 25;
 static int inflate_threshold = 50;
+static int halve_threshold_root = 15;
+static int inflate_threshold_root = 25; 
 
-static struct leaf *leaf_new(void)
+
+static void __alias_free_mem(struct rcu_head *head)
 {
-       struct leaf *l = kmalloc(sizeof(struct leaf),  GFP_KERNEL);
-       if (l) {
-               NODE_INIT_PARENT(l, T_LEAF);
-               INIT_HLIST_HEAD(&l->list);
-       }
-       return l;
+       struct fib_alias *fa = container_of(head, struct fib_alias, rcu);
+       kmem_cache_free(fn_alias_kmem, fa);
 }
 
-static struct leaf_info *leaf_info_new(int plen)
+static inline void alias_free_mem_rcu(struct fib_alias *fa)
 {
-       struct leaf_info *li = kmalloc(sizeof(struct leaf_info),  GFP_KERNEL);
-
-       if (!li)
-               return NULL;
+       call_rcu(&fa->rcu, __alias_free_mem);
+}
 
-       li->plen = plen;
-       INIT_LIST_HEAD(&li->falh);
+static void __leaf_free_rcu(struct rcu_head *head)
+{
+       kfree(container_of(head, struct leaf, rcu));
+}
 
-       return li;
+static inline void free_leaf(struct leaf *leaf)
+{
+       call_rcu(&leaf->rcu, __leaf_free_rcu);
 }
 
-static inline void free_leaf(struct leaf *l)
+static void __leaf_info_free_rcu(struct rcu_head *head)
 {
-       kfree(l);
+       kfree(container_of(head, struct leaf_info, rcu));
 }
 
-static inline void free_leaf_info(struct leaf_info *li)
+static inline void free_leaf_info(struct leaf_info *leaf)
 {
-       kfree(li);
+       call_rcu(&leaf->rcu, __leaf_info_free_rcu);
 }
 
 static struct tnode *tnode_alloc(unsigned int size)
 {
-       if (size <= PAGE_SIZE) {
-               return kmalloc(size, GFP_KERNEL);
-       } else {
-               return (struct tnode *)
-                       __get_free_pages(GFP_KERNEL, get_order(size));
-       }
+       struct page *pages;
+
+       if (size <= PAGE_SIZE)
+               return kcalloc(size, 1, GFP_KERNEL);
+
+       pages = alloc_pages(GFP_KERNEL|__GFP_ZERO, get_order(size));
+       if (!pages)
+               return NULL;
+
+       return page_address(pages);
 }
 
-static void __tnode_free(struct tnode *tn)
+static void __tnode_free_rcu(struct rcu_head *head)
 {
+       struct tnode *tn = container_of(head, struct tnode, rcu);
        unsigned int size = sizeof(struct tnode) +
-                               (1 << tn->bits) * sizeof(struct node *);
+               (1 << tn->bits) * sizeof(struct node *);
 
        if (size <= PAGE_SIZE)
                kfree(tn);
@@ -361,6 +355,31 @@ static void __tnode_free(struct tnode *tn)
                free_pages((unsigned long)tn, get_order(size));
 }
 
+static inline void tnode_free(struct tnode *tn)
+{
+       call_rcu(&tn->rcu, __tnode_free_rcu);
+}
+
+static struct leaf *leaf_new(void)
+{
+       struct leaf *l = kmalloc(sizeof(struct leaf),  GFP_KERNEL);
+       if (l) {
+               l->parent = T_LEAF;
+               INIT_HLIST_HEAD(&l->list);
+       }
+       return l;
+}
+
+static struct leaf_info *leaf_info_new(int plen)
+{
+       struct leaf_info *li = kmalloc(sizeof(struct leaf_info),  GFP_KERNEL);
+       if (li) {
+               li->plen = plen;
+               INIT_LIST_HEAD(&li->falh);
+       }
+       return li;
+}
+
 static struct tnode* tnode_new(t_key key, int pos, int bits)
 {
        int nchildren = 1<<bits;
@@ -369,7 +388,7 @@ static struct tnode* tnode_new(t_key key, int pos, int bits)
 
        if (tn) {
                memset(tn, 0, sz);
-               NODE_INIT_PARENT(tn, T_TNODE);
+               tn->parent = T_TNODE;
                tn->pos = pos;
                tn->bits = bits;
                tn->key = key;
@@ -377,30 +396,17 @@ static struct tnode* tnode_new(t_key key, int pos, int bits)
                tn->empty_children = 1<<bits;
        }
 
-       DBG("AT %p s=%u %u\n", tn, (unsigned int) sizeof(struct tnode),
-              (unsigned int) (sizeof(struct node) * 1<<bits));
+       pr_debug("AT %p s=%u %u\n", tn, (unsigned int) sizeof(struct tnode),
+                (unsigned int) (sizeof(struct node) * 1<<bits));
        return tn;
 }
 
-static void tnode_free(struct tnode *tn)
-{
-       BUG_ON(!tn);
-
-       if (IS_LEAF(tn)) {
-               free_leaf((struct leaf *)tn);
-               DBG("FL %p \n", tn);
-       } else {
-               __tnode_free(tn);
-               DBG("FT %p \n", tn);
-       }
-}
-
 /*
  * Check whether a tnode 'n' is "full", i.e. it is an internal node
  * and no bits are skipped. See discussion in dyntree paper p. 6
  */
 
-static inline int tnode_full(struct tnode *tn, struct node *n)
+static inline int tnode_full(const struct tnode *tn, const struct node *n)
 {
        if (n == NULL || IS_LEAF(n))
                return 0;
@@ -420,15 +426,11 @@ static inline void put_child(struct trie *t, struct tnode *tn, int i, struct nod
 
 static void tnode_put_child_reorg(struct tnode *tn, int i, struct node *n, int wasfull)
 {
-       struct node *chi;
+       struct node *chi = tn->child[i];
        int isfull;
 
-       if (i >= 1<<tn->bits) {
-               printk("bits=%d, i=%d\n", tn->bits, i);
-               BUG();
-       }
-       write_lock_bh(&fib_lock);
-       chi = tn->child[i];
+       BUG_ON(i >= 1<<tn->bits);
+
 
        /* update emptyChildren */
        if (n == NULL && chi != NULL)
@@ -449,8 +451,7 @@ static void tnode_put_child_reorg(struct tnode *tn, int i, struct node *n, int w
        if (n)
                NODE_SET_PARENT(n, tn);
 
-       tn->child[i] = n;
-       write_unlock_bh(&fib_lock);
+       rcu_assign_pointer(tn->child[i], n);
 }
 
 static struct node *resize(struct trie *t, struct tnode *tn)
@@ -458,12 +459,14 @@ static struct node *resize(struct trie *t, struct tnode *tn)
        int i;
        int err = 0;
        struct tnode *old_tn;
+       int inflate_threshold_use;
+       int halve_threshold_use;
 
        if (!tn)
                return NULL;
 
-       DBG("In tnode_resize %p inflate_threshold=%d threshold=%d\n",
-             tn, inflate_threshold, halve_threshold);
+       pr_debug("In tnode_resize %p inflate_threshold=%d threshold=%d\n",
+                tn, inflate_threshold, halve_threshold);
 
        /* No children */
        if (tn->empty_children == tnode_child_length(tn)) {
@@ -475,17 +478,12 @@ static struct node *resize(struct trie *t, struct tnode *tn)
                for (i = 0; i < tnode_child_length(tn); i++) {
                        struct node *n;
 
-                       write_lock_bh(&fib_lock);
                        n = tn->child[i];
-                       if (!n) {
-                               write_unlock_bh(&fib_lock);
+                       if (!n)
                                continue;
-                       }
 
                        /* compress one level */
-                       NODE_INIT_PARENT(n, NODE_TYPE(n));
-
-                       write_unlock_bh(&fib_lock);
+                       NODE_SET_PARENT(n, NULL);
                        tnode_free(tn);
                        return n;
                }
@@ -555,10 +553,17 @@ static struct node *resize(struct trie *t, struct tnode *tn)
 
        check_tnode(tn);
 
+       /* Keep root node larger  */
+
+       if(!tn->parent)
+               inflate_threshold_use = inflate_threshold_root;
+       else 
+               inflate_threshold_use = inflate_threshold;
+
        err = 0;
        while ((tn->full_children > 0 &&
               50 * (tn->full_children + tnode_child_length(tn) - tn->empty_children) >=
-                               inflate_threshold * tnode_child_length(tn))) {
+                               inflate_threshold_use * tnode_child_length(tn))) {
 
                old_tn = tn;
                tn = inflate(t, tn);
@@ -578,10 +583,18 @@ static struct node *resize(struct trie *t, struct tnode *tn)
         * node is above threshold.
         */
 
+
+       /* Keep root node larger  */
+
+       if(!tn->parent)
+               halve_threshold_use = halve_threshold_root;
+       else 
+               halve_threshold_use = halve_threshold;
+
        err = 0;
        while (tn->bits > 1 &&
               100 * (tnode_child_length(tn) - tn->empty_children) <
-              halve_threshold * tnode_child_length(tn)) {
+              halve_threshold_use * tnode_child_length(tn)) {
 
                old_tn = tn;
                tn = halve(t, tn);
@@ -596,24 +609,17 @@ static struct node *resize(struct trie *t, struct tnode *tn)
 
 
        /* Only one child remains */
-
        if (tn->empty_children == tnode_child_length(tn) - 1)
                for (i = 0; i < tnode_child_length(tn); i++) {
                        struct node *n;
 
-                       write_lock_bh(&fib_lock);
-
                        n = tn->child[i];
-                       if (!n) {
-                               write_unlock_bh(&fib_lock);
+                       if (!n)
                                continue;
-                       }
 
                        /* compress one level */
 
-                       NODE_INIT_PARENT(n, NODE_TYPE(n));
-
-                       write_unlock_bh(&fib_lock);
+                       NODE_SET_PARENT(n, NULL);
                        tnode_free(tn);
                        return n;
                }
@@ -628,11 +634,11 @@ static struct tnode *inflate(struct trie *t, struct tnode *tn)
        int olen = tnode_child_length(tn);
        int i;
 
-       DBG("In inflate\n");
+       pr_debug("In inflate\n");
 
        tn = tnode_new(oldtnode->key, oldtnode->pos, oldtnode->bits + 1);
 
-       if (!tn) 
+       if (!tn)
                return ERR_PTR(-ENOMEM);
 
        /*
@@ -752,12 +758,12 @@ nomem:
                int size = tnode_child_length(tn);
                int j;
 
-               for(j = 0; j < size; j++)
+               for (j = 0; j < size; j++)
                        if (tn->child[j])
                                tnode_free((struct tnode *)tn->child[j]);
 
                tnode_free(tn);
-       
+
                return ERR_PTR(-ENOMEM);
        }
 }
@@ -769,7 +775,7 @@ static struct tnode *halve(struct trie *t, struct tnode *tn)
        int i;
        int olen = tnode_child_length(tn);
 
-       DBG("In halve\n");
+       pr_debug("In halve\n");
 
        tn = tnode_new(oldtnode->key, oldtnode->pos, oldtnode->bits - 1);
 
@@ -788,14 +794,14 @@ static struct tnode *halve(struct trie *t, struct tnode *tn)
                right = tnode_get_child(oldtnode, i+1);
 
                /* Two nonempty children */
-               if (left && right)  {
+               if (left && right) {
                        struct tnode *newn;
-  
+
                        newn = tnode_new(left->key, tn->pos + tn->bits, 1);
-  
-                       if (!newn) 
+
+                       if (!newn)
                                goto nomem;
-  
+
                        put_child(t, tn, i/2, (struct node *)newn);
                }
 
@@ -813,7 +819,7 @@ static struct tnode *halve(struct trie *t, struct tnode *tn)
                                continue;
                        put_child(t, tn, i/2, right);
                        continue;
-               } 
+               }
 
                if (right == NULL) {
                        put_child(t, tn, i/2, left);
@@ -823,9 +829,6 @@ static struct tnode *halve(struct trie *t, struct tnode *tn)
                /* Two nonempty children */
                newBinNode = (struct tnode *) tnode_get_child(tn, i/2);
                put_child(t, tn, i/2, NULL);
-
-               BUG_ON(!newBinNode);
-
                put_child(t, newBinNode, 0, left);
                put_child(t, newBinNode, 1, right);
                put_child(t, tn, i/2, resize(t, newBinNode));
@@ -837,12 +840,12 @@ nomem:
                int size = tnode_child_length(tn);
                int j;
 
-               for(j = 0; j < size; j++)
+               for (j = 0; j < size; j++)
                        if (tn->child[j])
                                tnode_free((struct tnode *)tn->child[j]);
 
                tnode_free(tn);
-       
+
                return ERR_PTR(-ENOMEM);
        }
 }
@@ -853,19 +856,23 @@ static void trie_init(struct trie *t)
                return;
 
        t->size = 0;
-       t->trie = NULL;
+       rcu_assign_pointer(t->trie, NULL);
        t->revision = 0;
 #ifdef CONFIG_IP_FIB_TRIE_STATS
        memset(&t->stats, 0, sizeof(struct trie_use_stats));
 #endif
 }
 
-static struct leaf_info *find_leaf_info(struct hlist_head *head, int plen)
+/* readside must use rcu_read_lock currently dump routines
+ via get_fa_head and dump */
+
+static struct leaf_info *find_leaf_info(struct leaf *l, int plen)
 {
+       struct hlist_head *head = &l->list;
        struct hlist_node *node;
        struct leaf_info *li;
 
-       hlist_for_each_entry(li, node, head, hlist)
+       hlist_for_each_entry_rcu(li, node, head, hlist)
                if (li->plen == plen)
                        return li;
 
@@ -874,7 +881,7 @@ static struct leaf_info *find_leaf_info(struct hlist_head *head, int plen)
 
 static inline struct list_head * get_fa_head(struct leaf *l, int plen)
 {
-       struct leaf_info *li = find_leaf_info(&l->list, plen);
+       struct leaf_info *li = find_leaf_info(l, plen);
 
        if (!li)
                return NULL;
@@ -884,28 +891,27 @@ static inline struct list_head * get_fa_head(struct leaf *l, int plen)
 
 static void insert_leaf_info(struct hlist_head *head, struct leaf_info *new)
 {
-       struct leaf_info *li = NULL, *last = NULL;
-       struct hlist_node *node;
+        struct leaf_info *li = NULL, *last = NULL;
+        struct hlist_node *node;
 
-       write_lock_bh(&fib_lock);
+        if (hlist_empty(head)) {
+                hlist_add_head_rcu(&new->hlist, head);
+        } else {
+                hlist_for_each_entry(li, node, head, hlist) {
+                        if (new->plen > li->plen)
+                                break;
 
-       if (hlist_empty(head)) {
-               hlist_add_head(&new->hlist, head);
-       } else {
-               hlist_for_each_entry(li, node, head, hlist) {
-                       if (new->plen > li->plen)
-                               break;
-
-                       last = li;
-               }
-               if (last)
-                       hlist_add_after(&last->hlist, &new->hlist);
-               else
-                       hlist_add_before(&new->hlist, &li->hlist);
-       }
-       write_unlock_bh(&fib_lock);
+                        last = li;
+                }
+                if (last)
+                        hlist_add_after_rcu(&last->hlist, &new->hlist);
+                else
+                        hlist_add_before_rcu(&new->hlist, &li->hlist);
+        }
 }
 
+/* rcu_read_lock needs to be hold by caller from readside */
+
 static struct leaf *
 fib_find_node(struct trie *t, u32 key)
 {
@@ -914,7 +920,7 @@ fib_find_node(struct trie *t, u32 key)
        struct node *n;
 
        pos = 0;
-       n = t->trie;
+       n = rcu_dereference(t->trie);
 
        while (n != NULL &&  NODE_TYPE(n) == T_TNODE) {
                tn = (struct tnode *) n;
@@ -937,29 +943,13 @@ fib_find_node(struct trie *t, u32 key)
 
 static struct node *trie_rebalance(struct trie *t, struct tnode *tn)
 {
-       int i;
        int wasfull;
        t_key cindex, key;
        struct tnode *tp = NULL;
 
-       BUG_ON(!tn);
-
        key = tn->key;
-       i = 0;
 
        while (tn != NULL && NODE_PARENT(tn) != NULL) {
-               if (i > 10) {
-                       printk("Rebalance tn=%p \n", tn);
-                       if (tn)
-                               printk("tn->parent=%p \n", NODE_PARENT(tn));
-
-                       printk("Rebalance tp=%p \n", tp);
-                       if (tp)
-                               printk("tp->parent=%p \n", NODE_PARENT(tp));
-               }
-
-               BUG_ON(i > 12); /* Why is this a bug? -ojn */
-               i++;
 
                tp = NODE_PARENT(tn);
                cindex = tkey_extract_bits(key, tp->pos, tp->bits);
@@ -979,6 +969,8 @@ static struct node *trie_rebalance(struct trie *t, struct tnode *tn)
        return (struct node*) tn;
 }
 
+/* only used from updater-side */
+
 static  struct list_head *
 fib_insert_node(struct trie *t, int *err, u32 key, int plen)
 {
@@ -1022,10 +1014,7 @@ fib_insert_node(struct trie *t, int *err, u32 key, int plen)
                        pos = tn->pos + tn->bits;
                        n = tnode_get_child(tn, tkey_extract_bits(key, tn->pos, tn->bits));
 
-                       if (n && NODE_PARENT(n) != tn) {
-                               printk("BUG tn=%p, n->parent=%p\n", tn, NODE_PARENT(n));
-                               BUG();
-                       }
+                       BUG_ON(n && NODE_PARENT(n) != tn);
                } else
                        break;
        }
@@ -1079,8 +1068,6 @@ fib_insert_node(struct trie *t, int *err, u32 key, int plen)
 
                NODE_SET_PARENT(l, tp);
 
-               BUG_ON(!tp);
-
                cindex = tkey_extract_bits(key, tp->pos, tp->bits);
                put_child(t, (struct tnode *)tp, cindex, (struct node *)l);
        } else {
@@ -1120,17 +1107,18 @@ fib_insert_node(struct trie *t, int *err, u32 key, int plen)
                        cindex = tkey_extract_bits(key, tp->pos, tp->bits);
                        put_child(t, (struct tnode *)tp, cindex, (struct node *)tn);
                } else {
-                       t->trie = (struct node*) tn; /* First tnode */
+                       rcu_assign_pointer(t->trie, (struct node *)tn); /* First tnode */
                        tp = tn;
                }
        }
 
        if (tp && tp->pos + tp->bits > 32)
-               printk("ERROR tp=%p pos=%d, bits=%d, key=%0x plen=%d\n",
+               printk(KERN_WARNING "fib_trie tp=%p pos=%d, bits=%d, key=%0x plen=%d\n",
                       tp, tp->pos, tp->bits, key, plen);
 
        /* Rebalance the trie */
-       t->trie = trie_rebalance(t, tp);
+
+       rcu_assign_pointer(t->trie, trie_rebalance(t, tp));
 done:
        t->revision++;
 err:
@@ -1161,7 +1149,7 @@ fn_trie_insert(struct fib_table *tb, struct rtmsg *r, struct kern_rta *rta,
 
        key = ntohl(key);
 
-       DBG("Insert table=%d %08x/%d\n", tb->tb_id, key, plen);
+       pr_debug("Insert table=%d %08x/%d\n", tb->tb_id, key, plen);
 
        mask = ntohl(inet_make_mask(plen));
 
@@ -1205,16 +1193,21 @@ fn_trie_insert(struct fib_table *tb, struct rtmsg *r, struct kern_rta *rta,
                        struct fib_info *fi_drop;
                        u8 state;
 
-                       write_lock_bh(&fib_lock);
+                       err = -ENOBUFS;
+                       new_fa = kmem_cache_alloc(fn_alias_kmem, SLAB_KERNEL);
+                       if (new_fa == NULL)
+                               goto out;
 
                        fi_drop = fa->fa_info;
-                       fa->fa_info = fi;
-                       fa->fa_type = type;
-                       fa->fa_scope = r->rtm_scope;
+                       new_fa->fa_tos = fa->fa_tos;
+                       new_fa->fa_info = fi;
+                       new_fa->fa_type = type;
+                       new_fa->fa_scope = r->rtm_scope;
                        state = fa->fa_state;
-                       fa->fa_state &= ~FA_S_ACCESSED;
+                       new_fa->fa_state &= ~FA_S_ACCESSED;
 
-                       write_unlock_bh(&fib_lock);
+                       list_replace_rcu(&fa->fa_list, &new_fa->fa_list);
+                       alias_free_mem_rcu(fa);
 
                        fib_release_info(fi_drop);
                        if (state & FA_S_ACCESSED)
@@ -1266,11 +1259,8 @@ fn_trie_insert(struct fib_table *tb, struct rtmsg *r, struct kern_rta *rta,
                        goto out_free_new_fa;
        }
 
-       write_lock_bh(&fib_lock);
-
-       list_add_tail(&new_fa->fa_list, (fa ? &fa->fa_list : fa_head));
-
-       write_unlock_bh(&fib_lock);
+       list_add_tail_rcu(&new_fa->fa_list,
+                         (fa ? &fa->fa_list : fa_head));
 
        rt_cache_flush(-1);
        rtmsg_fib(RTM_NEWROUTE, htonl(key), new_fa, plen, tb->tb_id, nlhdr, req);
@@ -1285,7 +1275,10 @@ err:
        return err;
 }
 
-static inline int check_leaf(struct trie *t, struct leaf *l,  t_key key, int *plen, const struct flowi *flp,
+
+/* should be called with rcu_read_lock */
+static inline int check_leaf(struct trie *t, struct leaf *l,
+                            t_key key, int *plen, const struct flowi *flp,
                             struct fib_result *res)
 {
        int err, i;
@@ -1294,7 +1287,7 @@ static inline int check_leaf(struct trie *t, struct leaf *l,  t_key key, int *pl
        struct hlist_head *hhead = &l->list;
        struct hlist_node *node;
 
-       hlist_for_each_entry(li, node, hhead, hlist) {
+       hlist_for_each_entry_rcu(li, node, hhead, hlist) {
                i = li->plen;
                mask = ntohl(inet_make_mask(i));
                if (l->key != (key & mask))
@@ -1330,10 +1323,9 @@ fn_trie_lookup(struct fib_table *tb, const struct flowi *flp, struct fib_result
        t_key node_prefix, key_prefix, pref_mismatch;
        int mp;
 
-       n = t->trie;
-
-       read_lock(&fib_lock);
+       rcu_read_lock();
 
+       n = rcu_dereference(t->trie);
        if (!n)
                goto failed;
 
@@ -1503,10 +1495,11 @@ backtrace:
 failed:
        ret = 1;
 found:
-       read_unlock(&fib_lock);
+       rcu_read_unlock();
        return ret;
 }
 
+/* only called from updater side */
 static int trie_leaf_remove(struct trie *t, t_key key)
 {
        t_key cindex;
@@ -1514,7 +1507,7 @@ static int trie_leaf_remove(struct trie *t, t_key key)
        struct node *n = t->trie;
        struct leaf *l;
 
-       DBG("entering trie_leaf_remove(%p)\n", n);
+       pr_debug("entering trie_leaf_remove(%p)\n", n);
 
        /* Note that in the case skipped bits, those bits are *not* checked!
         * When we finish this, we will have NULL or a T_LEAF, and the
@@ -1526,10 +1519,7 @@ static int trie_leaf_remove(struct trie *t, t_key key)
                check_tnode(tn);
                n = tnode_get_child(tn ,tkey_extract_bits(key, tn->pos, tn->bits));
 
-               if (n && NODE_PARENT(n) != tn) {
-                       printk("BUG tn=%p, n->parent=%p\n", tn, NODE_PARENT(n));
-                       BUG();
-               }
+               BUG_ON(n && NODE_PARENT(n) != tn);
        }
        l = (struct leaf *) n;
 
@@ -1544,15 +1534,17 @@ static int trie_leaf_remove(struct trie *t, t_key key)
        t->revision++;
        t->size--;
 
+       preempt_disable();
        tp = NODE_PARENT(n);
        tnode_free((struct tnode *) n);
 
        if (tp) {
                cindex = tkey_extract_bits(key, tp->pos, tp->bits);
                put_child(t, (struct tnode *)tp, cindex, NULL);
-               t->trie = trie_rebalance(t, tp);
+               rcu_assign_pointer(t->trie, trie_rebalance(t, tp));
        } else
-               t->trie = NULL;
+               rcu_assign_pointer(t->trie, NULL);
+       preempt_enable();
 
        return 1;
 }
@@ -1568,7 +1560,6 @@ fn_trie_delete(struct fib_table *tb, struct rtmsg *r, struct kern_rta *rta,
        struct fib_alias *fa, *fa_to_delete;
        struct list_head *fa_head;
        struct leaf *l;
-       int kill_li = 0;
        struct leaf_info *li;
 
 
@@ -1597,10 +1588,11 @@ fn_trie_delete(struct fib_table *tb, struct rtmsg *r, struct kern_rta *rta,
        if (!fa)
                return -ESRCH;
 
-       DBG("Deleting %08x/%d tos=%d t=%p\n", key, plen, tos, t);
+       pr_debug("Deleting %08x/%d tos=%d t=%p\n", key, plen, tos, t);
 
        fa_to_delete = NULL;
        fa_head = fa->fa_list.prev;
+
        list_for_each_entry(fa, fa_head, fa_list) {
                struct fib_info *fi = fa->fa_info;
 
@@ -1626,20 +1618,14 @@ fn_trie_delete(struct fib_table *tb, struct rtmsg *r, struct kern_rta *rta,
        rtmsg_fib(RTM_DELROUTE, htonl(key), fa, plen, tb->tb_id, nlhdr, req);
 
        l = fib_find_node(t, key);
-       li = find_leaf_info(&l->list, plen);
+       li = find_leaf_info(l, plen);
 
-       write_lock_bh(&fib_lock);
-
-       list_del(&fa->fa_list);
+       list_del_rcu(&fa->fa_list);
 
        if (list_empty(fa_head)) {
-               hlist_del(&li->hlist);
-               kill_li = 1;
-       }
-       write_unlock_bh(&fib_lock);
-
-       if (kill_li)
+               hlist_del_rcu(&li->hlist);
                free_leaf_info(li);
+       }
 
        if (hlist_empty(&l->list))
                trie_leaf_remove(t, key);
@@ -1647,7 +1633,8 @@ fn_trie_delete(struct fib_table *tb, struct rtmsg *r, struct kern_rta *rta,
        if (fa->fa_state & FA_S_ACCESSED)
                rt_cache_flush(-1);
 
-       fn_free_alias(fa);
+       fib_release_info(fa->fa_info);
+       alias_free_mem_rcu(fa);
        return 0;
 }
 
@@ -1659,12 +1646,10 @@ static int trie_flush_list(struct trie *t, struct list_head *head)
        list_for_each_entry_safe(fa, fa_node, head, fa_list) {
                struct fib_info *fi = fa->fa_info;
 
-               if (fi && (fi->fib_flags&RTNH_F_DEAD)) {
-                       write_lock_bh(&fib_lock);
-                       list_del(&fa->fa_list);
-                       write_unlock_bh(&fib_lock);
-
-                       fn_free_alias(fa);
+               if (fi && (fi->fib_flags & RTNH_F_DEAD)) {
+                       list_del_rcu(&fa->fa_list);
+                       fib_release_info(fa->fa_info);
+                       alias_free_mem_rcu(fa);
                        found++;
                }
        }
@@ -1682,30 +1667,30 @@ static int trie_flush_leaf(struct trie *t, struct leaf *l)
                found += trie_flush_list(t, &li->falh);
 
                if (list_empty(&li->falh)) {
-                       write_lock_bh(&fib_lock);
-                       hlist_del(&li->hlist);
-                       write_unlock_bh(&fib_lock);
-
+                       hlist_del_rcu(&li->hlist);
                        free_leaf_info(li);
                }
        }
        return found;
 }
 
+/* rcu_read_lock needs to be hold by caller from readside */
+
 static struct leaf *nextleaf(struct trie *t, struct leaf *thisleaf)
 {
        struct node *c = (struct node *) thisleaf;
        struct tnode *p;
        int idx;
+       struct node *trie = rcu_dereference(t->trie);
 
        if (c == NULL) {
-               if (t->trie == NULL)
+               if (trie == NULL)
                        return NULL;
 
-               if (IS_LEAF(t->trie))          /* trie w. just a leaf */
-                       return (struct leaf *) t->trie;
+               if (IS_LEAF(trie))          /* trie w. just a leaf */
+                       return (struct leaf *) trie;
 
-               p = (struct tnode*) t->trie;  /* Start */
+               p = (struct tnode*) trie;  /* Start */
        } else
                p = (struct tnode *) NODE_PARENT(c);
 
@@ -1720,23 +1705,26 @@ static struct leaf *nextleaf(struct trie *t, struct leaf *thisleaf)
 
                last = 1 << p->bits;
                for (idx = pos; idx < last ; idx++) {
-                       if (!p->child[idx])
+                       c = rcu_dereference(p->child[idx]);
+
+                       if (!c)
                                continue;
 
                        /* Decend if tnode */
-                       while (IS_TNODE(p->child[idx])) {
-                               p = (struct tnode*) p->child[idx];
-                               idx = 0;
+                       while (IS_TNODE(c)) {
+                               p = (struct tnode *) c;
+                               idx = 0;
 
                                /* Rightmost non-NULL branch */
                                if (p && IS_TNODE(p))
-                                       while (p->child[idx] == NULL && idx < (1 << p->bits)) idx++;
+                                       while (!(c = rcu_dereference(p->child[idx]))
+                                              && idx < (1<<p->bits)) idx++;
 
                                /* Done with this tnode? */
-                               if (idx >= (1 << p->bits) || p->child[idx] == NULL)
+                               if (idx >= (1 << p->bits) || !c)
                                        goto up;
                        }
-                       return (struct leaf*) p->child[idx];
+                       return (struct leaf *) c;
                }
 up:
                /* No more children go up one step  */
@@ -1765,7 +1753,7 @@ static int fn_trie_flush(struct fib_table *tb)
        if (ll && hlist_empty(&ll->list))
                trie_leaf_remove(t, ll->key);
 
-       DBG("trie_flush found=%d\n", found);
+       pr_debug("trie_flush found=%d\n", found);
        return found;
 }
 
@@ -1786,7 +1774,7 @@ fn_trie_select_default(struct fib_table *tb, const struct flowi *flp, struct fib
        last_resort = NULL;
        order = -1;
 
-       read_lock(&fib_lock);
+       rcu_read_lock();
 
        l = fib_find_node(t, 0);
        if (!l)
@@ -1799,7 +1787,7 @@ fn_trie_select_default(struct fib_table *tb, const struct flowi *flp, struct fib
        if (list_empty(fa_head))
                goto out;
 
-       list_for_each_entry(fa, fa_head, fa_list) {
+       list_for_each_entry_rcu(fa, fa_head, fa_list) {
                struct fib_info *next_fi = fa->fa_info;
 
                if (fa->fa_scope != res->scope ||
@@ -1850,7 +1838,7 @@ fn_trie_select_default(struct fib_table *tb, const struct flowi *flp, struct fib
        }
        trie_last_dflt = last_idx;
  out:;
-       read_unlock(&fib_lock);
+       rcu_read_unlock();
 }
 
 static int fn_trie_dump_fa(t_key key, int plen, struct list_head *fah, struct fib_table *tb,
@@ -1864,21 +1852,14 @@ static int fn_trie_dump_fa(t_key key, int plen, struct list_head *fah, struct fi
        s_i = cb->args[3];
        i = 0;
 
-       list_for_each_entry(fa, fah, fa_list) {
+       /* rcu_read_lock is hold by caller */
+
+       list_for_each_entry_rcu(fa, fah, fa_list) {
                if (i < s_i) {
                        i++;
                        continue;
                }
-               if (fa->fa_info->fib_nh == NULL) {
-                       printk("Trie error _fib_nh=NULL in fa[%d] k=%08x plen=%d\n", i, key, plen);
-                       i++;
-                       continue;
-               }
-               if (fa->fa_info == NULL) {
-                       printk("Trie error fa_info=NULL in fa[%d] k=%08x plen=%d\n", i, key, plen);
-                       i++;
-                       continue;
-               }
+               BUG_ON(!fa->fa_info);
 
                if (fib_dump_info(skb, NETLINK_CB(cb->skb).pid,
                                  cb->nlh->nlmsg_seq,
@@ -1939,7 +1920,7 @@ static int fn_trie_dump(struct fib_table *tb, struct sk_buff *skb, struct netlin
 
        s_m = cb->args[1];
 
-       read_lock(&fib_lock);
+       rcu_read_lock();
        for (m = 0; m <= 32; m++) {
                if (m < s_m)
                        continue;
@@ -1952,11 +1933,11 @@ static int fn_trie_dump(struct fib_table *tb, struct sk_buff *skb, struct netlin
                        goto out;
                }
        }
-       read_unlock(&fib_lock);
+       rcu_read_unlock();
        cb->args[1] = m;
        return skb->len;
 out:
-       read_unlock(&fib_lock);
+       rcu_read_unlock();
        return -1;
 }
 
@@ -2001,554 +1982,534 @@ struct fib_table * __init fib_hash_init(int id)
                trie_main = t;
 
        if (id == RT_TABLE_LOCAL)
-               printk("IPv4 FIB: Using LC-trie version %s\n", VERSION);
+               printk(KERN_INFO "IPv4 FIB: Using LC-trie version %s\n", VERSION);
 
        return tb;
 }
 
-/* Trie dump functions */
-
-static void putspace_seq(struct seq_file *seq, int n)
-{
-       while (n--)
-               seq_printf(seq, " ");
-}
-
-static void printbin_seq(struct seq_file *seq, unsigned int v, int bits)
-{
-       while (bits--)
-               seq_printf(seq, "%s", (v & (1<<bits))?"1":"0");
-}
+#ifdef CONFIG_PROC_FS
+/* Depth first Trie walk iterator */
+struct fib_trie_iter {
+       struct tnode *tnode;
+       struct trie *trie;
+       unsigned index;
+       unsigned depth;
+};
 
-static void printnode_seq(struct seq_file *seq, int indent, struct node *n,
-                  int pend, int cindex, int bits)
+static struct node *fib_trie_get_next(struct fib_trie_iter *iter)
 {
-       putspace_seq(seq, indent);
-       if (IS_LEAF(n))
-               seq_printf(seq, "|");
-       else
-               seq_printf(seq, "+");
-       if (bits) {
-               seq_printf(seq, "%d/", cindex);
-               printbin_seq(seq, cindex, bits);
-               seq_printf(seq, ": ");
-       } else
-               seq_printf(seq, "<root>: ");
-       seq_printf(seq, "%s:%p ", IS_LEAF(n)?"Leaf":"Internal node", n);
+       struct tnode *tn = iter->tnode;
+       unsigned cindex = iter->index;
+       struct tnode *p;
 
-       if (IS_LEAF(n)) {
-               struct leaf *l = (struct leaf *)n;
-               struct fib_alias *fa;
-               int i;
+       pr_debug("get_next iter={node=%p index=%d depth=%d}\n",
+                iter->tnode, iter->index, iter->depth);
+rescan:
+       while (cindex < (1<<tn->bits)) {
+               struct node *n = tnode_get_child(tn, cindex);
 
-               seq_printf(seq, "key=%d.%d.%d.%d\n",
-                          n->key >> 24, (n->key >> 16) % 256, (n->key >> 8) % 256, n->key % 256);
-
-               for (i = 32; i >= 0; i--)
-                       if (find_leaf_info(&l->list, i)) {
-                               struct list_head *fa_head = get_fa_head(l, i);
-
-                               if (!fa_head)
-                                       continue;
-
-                               if (list_empty(fa_head))
-                                       continue;
-
-                               putspace_seq(seq, indent+2);
-                               seq_printf(seq, "{/%d...dumping}\n", i);
-
-                               list_for_each_entry(fa, fa_head, fa_list) {
-                                       putspace_seq(seq, indent+2);
-                                       if (fa->fa_info == NULL) {
-                                               seq_printf(seq, "Error fa_info=NULL\n");
-                                               continue;
-                                       }
-                                       if (fa->fa_info->fib_nh == NULL) {
-                                               seq_printf(seq, "Error _fib_nh=NULL\n");
-                                               continue;
-                                       }
-
-                                       seq_printf(seq, "{type=%d scope=%d TOS=%d}\n",
-                                             fa->fa_type,
-                                             fa->fa_scope,
-                                             fa->fa_tos);
-                               }
+               if (n) {
+                       if (IS_LEAF(n)) {
+                               iter->tnode = tn;
+                               iter->index = cindex + 1;
+                       } else {
+                               /* push down one level */
+                               iter->tnode = (struct tnode *) n;
+                               iter->index = 0;
+                               ++iter->depth;
                        }
-       } else {
-               struct tnode *tn = (struct tnode *)n;
-               int plen = ((struct tnode *)n)->pos;
-               t_key prf = MASK_PFX(n->key, plen);
-
-               seq_printf(seq, "key=%d.%d.%d.%d/%d\n",
-                          prf >> 24, (prf >> 16) % 256, (prf >> 8) % 256, prf % 256, plen);
-
-               putspace_seq(seq, indent); seq_printf(seq, "|    ");
-               seq_printf(seq, "{key prefix=%08x/", tn->key & TKEY_GET_MASK(0, tn->pos));
-               printbin_seq(seq, tkey_extract_bits(tn->key, 0, tn->pos), tn->pos);
-               seq_printf(seq, "}\n");
-               putspace_seq(seq, indent); seq_printf(seq, "|    ");
-               seq_printf(seq, "{pos=%d", tn->pos);
-               seq_printf(seq, " (skip=%d bits)", tn->pos - pend);
-               seq_printf(seq, " bits=%d (%u children)}\n", tn->bits, (1 << tn->bits));
-               putspace_seq(seq, indent); seq_printf(seq, "|    ");
-               seq_printf(seq, "{empty=%d full=%d}\n", tn->empty_children, tn->full_children);
-       }
-}
-
-static void trie_dump_seq(struct seq_file *seq, struct trie *t)
-{
-       struct node *n = t->trie;
-       int cindex = 0;
-       int indent = 1;
-       int pend = 0;
-       int depth = 0;
-       struct tnode *tn;
-
-       read_lock(&fib_lock);
-
-       seq_printf(seq, "------ trie_dump of t=%p ------\n", t);
-
-       if (!n) {
-               seq_printf(seq, "------ trie is empty\n");
+                       return n;
+               }
 
-               read_unlock(&fib_lock);
-               return;
+               ++cindex;
        }
 
-       printnode_seq(seq, indent, n, pend, cindex, 0);
-
-       if (!IS_TNODE(n)) {
-               read_unlock(&fib_lock);
-               return;
+       /* Current node exhausted, pop back up */
+       p = NODE_PARENT(tn);
+       if (p) {
+               cindex = tkey_extract_bits(tn->key, p->pos, p->bits)+1;
+               tn = p;
+               --iter->depth;
+               goto rescan;
        }
 
-       tn = (struct tnode *)n;
-       pend = tn->pos+tn->bits;
-       putspace_seq(seq, indent); seq_printf(seq, "\\--\n");
-       indent += 3;
-       depth++;
-
-       while (tn && cindex < (1 << tn->bits)) {
-               if (tn->child[cindex]) {
-                       /* Got a child */
+       /* got root? */
+       return NULL;
+}
 
-                       printnode_seq(seq, indent, tn->child[cindex], pend, cindex, tn->bits);
-                       if (IS_LEAF(tn->child[cindex])) {
-                               cindex++;
-                       } else {
-                               /*
-                                * New tnode. Decend one level
-                                */
-
-                               depth++;
-                               tn = (struct tnode *)tn->child[cindex];
-                               pend = tn->pos + tn->bits;
-                               putspace_seq(seq, indent); seq_printf(seq, "\\--\n");
-                               indent += 3;
-                               cindex = 0;
-                       }
-               } else
-                       cindex++;
+static struct node *fib_trie_get_first(struct fib_trie_iter *iter,
+                                      struct trie *t)
+{
+       struct node *n = rcu_dereference(t->trie);
 
-               /*
-                * Test if we are done
-                */
+       if (n && IS_TNODE(n)) {
+               iter->tnode = (struct tnode *) n;
+               iter->trie = t;
+               iter->index = 0;
+               iter->depth = 1;
+               return n;
+       }
+       return NULL;
+}
 
-               while (cindex >= (1 << tn->bits)) {
-                       /*
-                        * Move upwards and test for root
-                        * pop off all traversed  nodes
-                        */
+static void trie_collect_stats(struct trie *t, struct trie_stat *s)
+{
+       struct node *n;
+       struct fib_trie_iter iter;
 
-                       if (NODE_PARENT(tn) == NULL) {
-                               tn = NULL;
-                               break;
-                       }
+       memset(s, 0, sizeof(*s));
 
-                       cindex = tkey_extract_bits(tn->key, NODE_PARENT(tn)->pos, NODE_PARENT(tn)->bits);
-                       cindex++;
-                       tn = NODE_PARENT(tn);
-                       pend = tn->pos + tn->bits;
-                       indent -= 3;
-                       depth--;
+       rcu_read_lock();
+       for (n = fib_trie_get_first(&iter, t); n;
+            n = fib_trie_get_next(&iter)) {
+               if (IS_LEAF(n)) {
+                       s->leaves++;
+                       s->totdepth += iter.depth;
+                       if (iter.depth > s->maxdepth)
+                               s->maxdepth = iter.depth;
+               } else {
+                       const struct tnode *tn = (const struct tnode *) n;
+                       int i;
+
+                       s->tnodes++;
+                       s->nodesizes[tn->bits]++;
+                       for (i = 0; i < (1<<tn->bits); i++)
+                               if (!tn->child[i])
+                                       s->nullpointers++;
                }
        }
-
-       read_unlock(&fib_lock);
+       rcu_read_unlock();
 }
 
-static struct trie_stat *trie_stat_new(void)
+/*
+ *     This outputs /proc/net/fib_triestats
+ */
+static void trie_show_stats(struct seq_file *seq, struct trie_stat *stat)
 {
-       struct trie_stat *s;
-       int i;
+       unsigned i, max, pointers, bytes, avdepth;
 
-       s = kmalloc(sizeof(struct trie_stat), GFP_KERNEL);
-       if (!s)
-               return NULL;
+       if (stat->leaves)
+               avdepth = stat->totdepth*100 / stat->leaves;
+       else
+               avdepth = 0;
 
-       s->totdepth = 0;
-       s->maxdepth = 0;
-       s->tnodes = 0;
-       s->leaves = 0;
-       s->nullpointers = 0;
+       seq_printf(seq, "\tAver depth:     %d.%02d\n", avdepth / 100, avdepth % 100 );
+       seq_printf(seq, "\tMax depth:      %u\n", stat->maxdepth);
 
-       for (i = 0; i < MAX_CHILDS; i++)
-               s->nodesizes[i] = 0;
+       seq_printf(seq, "\tLeaves:         %u\n", stat->leaves);
 
-       return s;
-}
+       bytes = sizeof(struct leaf) * stat->leaves;
+       seq_printf(seq, "\tInternal nodes: %d\n\t", stat->tnodes);
+       bytes += sizeof(struct tnode) * stat->tnodes;
 
-static struct trie_stat *trie_collect_stats(struct trie *t)
-{
-       struct node *n = t->trie;
-       struct trie_stat *s = trie_stat_new();
-       int cindex = 0;
-       int pend = 0;
-       int depth = 0;
+       max = MAX_CHILDS-1;
+       while (max >= 0 && stat->nodesizes[max] == 0)
+               max--;
 
-       if (!s)
-               return NULL;
-       if (!n)
-               return s;
+       pointers = 0;
+       for (i = 1; i <= max; i++)
+               if (stat->nodesizes[i] != 0) {
+                       seq_printf(seq, "  %d: %d",  i, stat->nodesizes[i]);
+                       pointers += (1<<i) * stat->nodesizes[i];
+               }
+       seq_putc(seq, '\n');
+       seq_printf(seq, "\tPointers: %d\n", pointers);
 
-       read_lock(&fib_lock);
+       bytes += sizeof(struct node *) * pointers;
+       seq_printf(seq, "Null ptrs: %d\n", stat->nullpointers);
+       seq_printf(seq, "Total size: %d  kB\n", (bytes + 1023) / 1024);
 
-       if (IS_TNODE(n)) {
-               struct tnode *tn = (struct tnode *)n;
-               pend = tn->pos+tn->bits;
-               s->nodesizes[tn->bits]++;
-               depth++;
-
-               while (tn && cindex < (1 << tn->bits)) {
-                       if (tn->child[cindex]) {
-                               /* Got a child */
-
-                               if (IS_LEAF(tn->child[cindex])) {
-                                       cindex++;
-
-                                       /* stats */
-                                       if (depth > s->maxdepth)
-                                               s->maxdepth = depth;
-                                       s->totdepth += depth;
-                                       s->leaves++;
-                               } else {
-                                       /*
-                                        * New tnode. Decend one level
-                                        */
-
-                                       s->tnodes++;
-                                       s->nodesizes[tn->bits]++;
-                                       depth++;
-
-                                       n = tn->child[cindex];
-                                       tn = (struct tnode *)n;
-                                       pend = tn->pos+tn->bits;
-
-                                       cindex = 0;
-                               }
-                       } else {
-                               cindex++;
-                               s->nullpointers++;
-                       }
+#ifdef CONFIG_IP_FIB_TRIE_STATS
+       seq_printf(seq, "Counters:\n---------\n");
+       seq_printf(seq,"gets = %d\n", t->stats.gets);
+       seq_printf(seq,"backtracks = %d\n", t->stats.backtrack);
+       seq_printf(seq,"semantic match passed = %d\n", t->stats.semantic_match_passed);
+       seq_printf(seq,"semantic match miss = %d\n", t->stats.semantic_match_miss);
+       seq_printf(seq,"null node hit= %d\n", t->stats.null_node_hit);
+       seq_printf(seq,"skipped node resize = %d\n", t->stats.resize_node_skipped);
+#ifdef CLEAR_STATS
+       memset(&(t->stats), 0, sizeof(t->stats));
+#endif
+#endif /*  CONFIG_IP_FIB_TRIE_STATS */
+}
 
-                       /*
-                        * Test if we are done
-                        */
+static int fib_triestat_seq_show(struct seq_file *seq, void *v)
+{
+       struct trie_stat *stat;
 
-                       while (cindex >= (1 << tn->bits)) {
-                               /*
-                                * Move upwards and test for root
-                                * pop off all traversed  nodes
-                                */
+       stat = kmalloc(sizeof(*stat), GFP_KERNEL);
+       if (!stat)
+               return -ENOMEM;
 
-                               if (NODE_PARENT(tn) == NULL) {
-                                       tn = NULL;
-                                       n = NULL;
-                                       break;
-                               }
+       seq_printf(seq, "Basic info: size of leaf: %Zd bytes, size of tnode: %Zd bytes.\n",
+                  sizeof(struct leaf), sizeof(struct tnode));
 
-                               cindex = tkey_extract_bits(tn->key, NODE_PARENT(tn)->pos, NODE_PARENT(tn)->bits);
-                               tn = NODE_PARENT(tn);
-                               cindex++;
-                               n = (struct node *)tn;
-                               pend = tn->pos+tn->bits;
-                               depth--;
-                       }
-               }
+       if (trie_local) {
+               seq_printf(seq, "Local:\n");
+               trie_collect_stats(trie_local, stat);
+               trie_show_stats(seq, stat);
        }
 
-       read_unlock(&fib_lock);
-       return s;
-}
+       if (trie_main) {
+               seq_printf(seq, "Main:\n");
+               trie_collect_stats(trie_main, stat);
+               trie_show_stats(seq, stat);
+       }
+       kfree(stat);
 
-#ifdef CONFIG_PROC_FS
+       return 0;
+}
 
-static struct fib_alias *fib_triestat_get_first(struct seq_file *seq)
+static int fib_triestat_seq_open(struct inode *inode, struct file *file)
 {
-       return NULL;
+       return single_open(file, fib_triestat_seq_show, NULL);
 }
 
-static struct fib_alias *fib_triestat_get_next(struct seq_file *seq)
+static struct file_operations fib_triestat_fops = {
+       .owner  = THIS_MODULE,
+       .open   = fib_triestat_seq_open,
+       .read   = seq_read,
+       .llseek = seq_lseek,
+       .release = single_release,
+};
+
+static struct node *fib_trie_get_idx(struct fib_trie_iter *iter,
+                                     loff_t pos)
 {
+       loff_t idx = 0;
+       struct node *n;
+
+       for (n = fib_trie_get_first(iter, trie_local);
+            n; ++idx, n = fib_trie_get_next(iter)) {
+               if (pos == idx)
+                       return n;
+       }
+
+       for (n = fib_trie_get_first(iter, trie_main);
+            n; ++idx, n = fib_trie_get_next(iter)) {
+               if (pos == idx)
+                       return n;
+       }
        return NULL;
 }
 
-static void *fib_triestat_seq_start(struct seq_file *seq, loff_t *pos)
+static void *fib_trie_seq_start(struct seq_file *seq, loff_t *pos)
 {
-       if (!ip_fib_main_table)
-               return NULL;
-
-       if (*pos)
-               return fib_triestat_get_next(seq);
-       else
+       rcu_read_lock();
+       if (*pos == 0)
                return SEQ_START_TOKEN;
+       return fib_trie_get_idx(seq->private, *pos - 1);
 }
 
-static void *fib_triestat_seq_next(struct seq_file *seq, void *v, loff_t *pos)
+static void *fib_trie_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 {
+       struct fib_trie_iter *iter = seq->private;
+       void *l = v;
+
        ++*pos;
        if (v == SEQ_START_TOKEN)
-               return fib_triestat_get_first(seq);
-       else
-               return fib_triestat_get_next(seq);
-}
+               return fib_trie_get_idx(iter, 0);
 
-static void fib_triestat_seq_stop(struct seq_file *seq, void *v)
-{
+       v = fib_trie_get_next(iter);
+       BUG_ON(v == l);
+       if (v)
+               return v;
 
-}
+       /* continue scan in next trie */
+       if (iter->trie == trie_local)
+               return fib_trie_get_first(iter, trie_main);
 
-/*
- *     This outputs /proc/net/fib_triestats
- *
- *     It always works in backward compatibility mode.
- *     The format of the file is not supposed to be changed.
- */
+       return NULL;
+}
 
-static void collect_and_show(struct trie *t, struct seq_file *seq)
+static void fib_trie_seq_stop(struct seq_file *seq, void *v)
 {
-       int bytes = 0; /* How many bytes are used, a ref is 4 bytes */
-       int i, max, pointers;
-       struct trie_stat *stat;
-       int avdepth;
-
-       stat = trie_collect_stats(t);
-
-       bytes = 0;
-       seq_printf(seq, "trie=%p\n", t);
-
-       if (stat) {
-               if (stat->leaves)
-                       avdepth = stat->totdepth*100 / stat->leaves;
-               else
-                       avdepth = 0;
-               seq_printf(seq, "Aver depth: %d.%02d\n", avdepth / 100, avdepth % 100);
-               seq_printf(seq, "Max depth: %4d\n", stat->maxdepth);
+       rcu_read_unlock();
+}
 
-               seq_printf(seq, "Leaves: %d\n", stat->leaves);
-               bytes += sizeof(struct leaf) * stat->leaves;
-               seq_printf(seq, "Internal nodes: %d\n", stat->tnodes);
-               bytes += sizeof(struct tnode) * stat->tnodes;
+static void seq_indent(struct seq_file *seq, int n)
+{
+       while (n-- > 0) seq_puts(seq, "   ");
+}
 
-               max = MAX_CHILDS-1;
+static inline const char *rtn_scope(enum rt_scope_t s)
+{
+       static char buf[32];
 
-               while (max >= 0 && stat->nodesizes[max] == 0)
-                       max--;
-               pointers = 0;
+       switch(s) {
+       case RT_SCOPE_UNIVERSE: return "universe";
+       case RT_SCOPE_SITE:     return "site";
+       case RT_SCOPE_LINK:     return "link";
+       case RT_SCOPE_HOST:     return "host";
+       case RT_SCOPE_NOWHERE:  return "nowhere";
+       default:
+               snprintf(buf, sizeof(buf), "scope=%d", s);
+               return buf;
+       }
+}
 
-               for (i = 1; i <= max; i++)
-                       if (stat->nodesizes[i] != 0) {
-                               seq_printf(seq, "  %d: %d",  i, stat->nodesizes[i]);
-                               pointers += (1<<i) * stat->nodesizes[i];
-                       }
-               seq_printf(seq, "\n");
-               seq_printf(seq, "Pointers: %d\n", pointers);
-               bytes += sizeof(struct node *) * pointers;
-               seq_printf(seq, "Null ptrs: %d\n", stat->nullpointers);
-               seq_printf(seq, "Total size: %d  kB\n", bytes / 1024);
+static const char *rtn_type_names[__RTN_MAX] = {
+       [RTN_UNSPEC] = "UNSPEC",
+       [RTN_UNICAST] = "UNICAST",
+       [RTN_LOCAL] = "LOCAL",
+       [RTN_BROADCAST] = "BROADCAST",
+       [RTN_ANYCAST] = "ANYCAST",
+       [RTN_MULTICAST] = "MULTICAST",
+       [RTN_BLACKHOLE] = "BLACKHOLE",
+       [RTN_UNREACHABLE] = "UNREACHABLE",
+       [RTN_PROHIBIT] = "PROHIBIT",
+       [RTN_THROW] = "THROW",
+       [RTN_NAT] = "NAT",
+       [RTN_XRESOLVE] = "XRESOLVE",
+};
 
-               kfree(stat);
-       }
+static inline const char *rtn_type(unsigned t)
+{
+       static char buf[32];
 
-#ifdef CONFIG_IP_FIB_TRIE_STATS
-       seq_printf(seq, "Counters:\n---------\n");
-       seq_printf(seq,"gets = %d\n", t->stats.gets);
-       seq_printf(seq,"backtracks = %d\n", t->stats.backtrack);
-       seq_printf(seq,"semantic match passed = %d\n", t->stats.semantic_match_passed);
-       seq_printf(seq,"semantic match miss = %d\n", t->stats.semantic_match_miss);
-       seq_printf(seq,"null node hit= %d\n", t->stats.null_node_hit);
-       seq_printf(seq,"skipped node resize = %d\n", t->stats.resize_node_skipped);
-#ifdef CLEAR_STATS
-       memset(&(t->stats), 0, sizeof(t->stats));
-#endif
-#endif /*  CONFIG_IP_FIB_TRIE_STATS */
+       if (t < __RTN_MAX && rtn_type_names[t])
+               return rtn_type_names[t];
+       snprintf(buf, sizeof(buf), "type %d", t);
+       return buf;
 }
 
-static int fib_triestat_seq_show(struct seq_file *seq, void *v)
+/* Pretty print the trie */
+static int fib_trie_seq_show(struct seq_file *seq, void *v)
 {
-       char bf[128];
+       const struct fib_trie_iter *iter = seq->private;
+       struct node *n = v;
 
-       if (v == SEQ_START_TOKEN) {
-               seq_printf(seq, "Basic info: size of leaf: %Zd bytes, size of tnode: %Zd bytes.\n",
-                          sizeof(struct leaf), sizeof(struct tnode));
-               if (trie_local)
-                       collect_and_show(trie_local, seq);
+       if (v == SEQ_START_TOKEN)
+               return 0;
 
-               if (trie_main)
-                       collect_and_show(trie_main, seq);
-       } else {
-               snprintf(bf, sizeof(bf), "*\t%08X\t%08X", 200, 400);
+       if (IS_TNODE(n)) {
+               struct tnode *tn = (struct tnode *) n;
+               t_key prf = ntohl(MASK_PFX(tn->key, tn->pos));
 
-               seq_printf(seq, "%-127s\n", bf);
+               if (!NODE_PARENT(n)) {
+                       if (iter->trie == trie_local)
+                               seq_puts(seq, "<local>:\n");
+                       else
+                               seq_puts(seq, "<main>:\n");
+               } 
+               seq_indent(seq, iter->depth-1);
+               seq_printf(seq, "  +-- %d.%d.%d.%d/%d %d %d %d\n",
+                          NIPQUAD(prf), tn->pos, tn->bits, tn->full_children, 
+                          tn->empty_children);
+               
+       } else {
+               struct leaf *l = (struct leaf *) n;
+               int i;
+               u32 val = ntohl(l->key);
+
+               seq_indent(seq, iter->depth);
+               seq_printf(seq, "  |-- %d.%d.%d.%d\n", NIPQUAD(val));
+               for (i = 32; i >= 0; i--) {
+                       struct leaf_info *li = find_leaf_info(l, i);
+                       if (li) {
+                               struct fib_alias *fa;
+                               list_for_each_entry_rcu(fa, &li->falh, fa_list) {
+                                       seq_indent(seq, iter->depth+1);
+                                       seq_printf(seq, "  /%d %s %s", i,
+                                                  rtn_scope(fa->fa_scope),
+                                                  rtn_type(fa->fa_type));
+                                       if (fa->fa_tos)
+                                               seq_printf(seq, "tos =%d\n",
+                                                          fa->fa_tos);
+                                       seq_putc(seq, '\n');
+                               }
+                       }
+               }
        }
+
        return 0;
 }
 
-static struct seq_operations fib_triestat_seq_ops = {
-       .start = fib_triestat_seq_start,
-       .next  = fib_triestat_seq_next,
-       .stop  = fib_triestat_seq_stop,
-       .show  = fib_triestat_seq_show,
+static struct seq_operations fib_trie_seq_ops = {
+       .start  = fib_trie_seq_start,
+       .next   = fib_trie_seq_next,
+       .stop   = fib_trie_seq_stop,
+       .show   = fib_trie_seq_show,
 };
 
-static int fib_triestat_seq_open(struct inode *inode, struct file *file)
+static int fib_trie_seq_open(struct inode *inode, struct file *file)
 {
        struct seq_file *seq;
        int rc = -ENOMEM;
+       struct fib_trie_iter *s = kmalloc(sizeof(*s), GFP_KERNEL);
+
+       if (!s)
+               goto out;
 
-       rc = seq_open(file, &fib_triestat_seq_ops);
+       rc = seq_open(file, &fib_trie_seq_ops);
        if (rc)
                goto out_kfree;
 
-       seq = file->private_data;
+       seq          = file->private_data;
+       seq->private = s;
+       memset(s, 0, sizeof(*s));
 out:
        return rc;
 out_kfree:
+       kfree(s);
        goto out;
 }
 
-static struct file_operations fib_triestat_seq_fops = {
-       .owner  = THIS_MODULE,
-       .open   = fib_triestat_seq_open,
-       .read   = seq_read,
-       .llseek = seq_lseek,
+static struct file_operations fib_trie_fops = {
+       .owner  = THIS_MODULE,
+       .open   = fib_trie_seq_open,
+       .read   = seq_read,
+       .llseek = seq_lseek,
        .release = seq_release_private,
 };
 
-int __init fib_stat_proc_init(void)
+static unsigned fib_flag_trans(int type, u32 mask, const struct fib_info *fi)
 {
-       if (!proc_net_fops_create("fib_triestat", S_IRUGO, &fib_triestat_seq_fops))
-               return -ENOMEM;
-       return 0;
-}
-
-void __init fib_stat_proc_exit(void)
-{
-       proc_net_remove("fib_triestat");
-}
+       static unsigned type2flags[RTN_MAX + 1] = {
+               [7] = RTF_REJECT, [8] = RTF_REJECT,
+       };
+       unsigned flags = type2flags[type];
 
-static struct fib_alias *fib_trie_get_first(struct seq_file *seq)
-{
-       return NULL;
+       if (fi && fi->fib_nh->nh_gw)
+               flags |= RTF_GATEWAY;
+       if (mask == 0xFFFFFFFF)
+               flags |= RTF_HOST;
+       flags |= RTF_UP;
+       return flags;
 }
 
-static struct fib_alias *fib_trie_get_next(struct seq_file *seq)
+/*
+ *     This outputs /proc/net/route.
+ *     The format of the file is not supposed to be changed
+ *     and needs to be same as fib_hash output to avoid breaking
+ *     legacy utilities
+ */
+static int fib_route_seq_show(struct seq_file *seq, void *v)
 {
-       return NULL;
-}
+       const struct fib_trie_iter *iter = seq->private;
+       struct leaf *l = v;
+       int i;
+       char bf[128];
 
-static void *fib_trie_seq_start(struct seq_file *seq, loff_t *pos)
-{
-       if (!ip_fib_main_table)
-               return NULL;
+       if (v == SEQ_START_TOKEN) {
+               seq_printf(seq, "%-127s\n", "Iface\tDestination\tGateway "
+                          "\tFlags\tRefCnt\tUse\tMetric\tMask\t\tMTU"
+                          "\tWindow\tIRTT");
+               return 0;
+       }
 
-       if (*pos)
-               return fib_trie_get_next(seq);
-       else
-               return SEQ_START_TOKEN;
-}
+       if (iter->trie == trie_local)
+               return 0;
+       if (IS_TNODE(l))
+               return 0;
 
-static void *fib_trie_seq_next(struct seq_file *seq, void *v, loff_t *pos)
-{
-       ++*pos;
-       if (v == SEQ_START_TOKEN)
-               return fib_trie_get_first(seq);
-       else
-               return fib_trie_get_next(seq);
+       for (i=32; i>=0; i--) {
+               struct leaf_info *li = find_leaf_info(l, i);
+               struct fib_alias *fa;
+               u32 mask, prefix;
 
-}
+               if (!li)
+                       continue;
 
-static void fib_trie_seq_stop(struct seq_file *seq, void *v)
-{
-}
+               mask = inet_make_mask(li->plen);
+               prefix = htonl(l->key);
 
-/*
- *     This outputs /proc/net/fib_trie.
- *
- *     It always works in backward compatibility mode.
- *     The format of the file is not supposed to be changed.
- */
+               list_for_each_entry_rcu(fa, &li->falh, fa_list) {
+                       const struct fib_info *fi = fa->fa_info;
+                       unsigned flags = fib_flag_trans(fa->fa_type, mask, fi);
 
-static int fib_trie_seq_show(struct seq_file *seq, void *v)
-{
-       char bf[128];
+                       if (fa->fa_type == RTN_BROADCAST
+                           || fa->fa_type == RTN_MULTICAST)
+                               continue;
 
-       if (v == SEQ_START_TOKEN) {
-               if (trie_local)
-                       trie_dump_seq(seq, trie_local);
+                       if (fi)
+                               snprintf(bf, sizeof(bf),
+                                        "%s\t%08X\t%08X\t%04X\t%d\t%u\t%d\t%08X\t%d\t%u\t%u",
+                                        fi->fib_dev ? fi->fib_dev->name : "*",
+                                        prefix,
+                                        fi->fib_nh->nh_gw, flags, 0, 0,
+                                        fi->fib_priority,
+                                        mask,
+                                        (fi->fib_advmss ? fi->fib_advmss + 40 : 0),
+                                        fi->fib_window,
+                                        fi->fib_rtt >> 3);
+                       else
+                               snprintf(bf, sizeof(bf),
+                                        "*\t%08X\t%08X\t%04X\t%d\t%u\t%d\t%08X\t%d\t%u\t%u",
+                                        prefix, 0, flags, 0, 0, 0,
+                                        mask, 0, 0, 0);
 
-               if (trie_main)
-                       trie_dump_seq(seq, trie_main);
-       } else {
-               snprintf(bf, sizeof(bf),
-                        "*\t%08X\t%08X", 200, 400);
-               seq_printf(seq, "%-127s\n", bf);
+                       seq_printf(seq, "%-127s\n", bf);
+               }
        }
 
        return 0;
 }
 
-static struct seq_operations fib_trie_seq_ops = {
-       .start = fib_trie_seq_start,
-       .next  = fib_trie_seq_next,
-       .stop  = fib_trie_seq_stop,
-       .show  = fib_trie_seq_show,
+static struct seq_operations fib_route_seq_ops = {
+       .start  = fib_trie_seq_start,
+       .next   = fib_trie_seq_next,
+       .stop   = fib_trie_seq_stop,
+       .show   = fib_route_seq_show,
 };
 
-static int fib_trie_seq_open(struct inode *inode, struct file *file)
+static int fib_route_seq_open(struct inode *inode, struct file *file)
 {
        struct seq_file *seq;
        int rc = -ENOMEM;
+       struct fib_trie_iter *s = kmalloc(sizeof(*s), GFP_KERNEL);
 
-       rc = seq_open(file, &fib_trie_seq_ops);
+       if (!s)
+               goto out;
+
+       rc = seq_open(file, &fib_route_seq_ops);
        if (rc)
                goto out_kfree;
 
-       seq = file->private_data;
+       seq          = file->private_data;
+       seq->private = s;
+       memset(s, 0, sizeof(*s));
 out:
        return rc;
 out_kfree:
+       kfree(s);
        goto out;
 }
 
-static struct file_operations fib_trie_seq_fops = {
-       .owner  = THIS_MODULE,
-       .open   = fib_trie_seq_open,
-       .read   = seq_read,
-       .llseek = seq_lseek,
-       .release= seq_release_private,
+static struct file_operations fib_route_fops = {
+       .owner  = THIS_MODULE,
+       .open   = fib_route_seq_open,
+       .read   = seq_read,
+       .llseek = seq_lseek,
+       .release = seq_release_private,
 };
 
 int __init fib_proc_init(void)
 {
-       if (!proc_net_fops_create("fib_trie", S_IRUGO, &fib_trie_seq_fops))
-               return -ENOMEM;
+       if (!proc_net_fops_create("fib_trie", S_IRUGO, &fib_trie_fops))
+               goto out1;
+
+       if (!proc_net_fops_create("fib_triestat", S_IRUGO, &fib_triestat_fops))
+               goto out2;
+
+       if (!proc_net_fops_create("route", S_IRUGO, &fib_route_fops))
+               goto out3;
+
        return 0;
+
+out3:
+       proc_net_remove("fib_triestat");
+out2:
+       proc_net_remove("fib_trie");
+out1:
+       return -ENOMEM;
 }
 
 void __init fib_proc_exit(void)
 {
        proc_net_remove("fib_trie");
+       proc_net_remove("fib_triestat");
+       proc_net_remove("route");
 }
 
 #endif /* CONFIG_PROC_FS */