]> git.proxmox.com Git - mirror_ubuntu-kernels.git/commitdiff
Staging: batman-adv: Rewrite hash using hlist_*
authorSven Eckelmann <sven.eckelmann@gmx.de>
Sun, 21 Nov 2010 23:56:00 +0000 (00:56 +0100)
committerGreg Kroah-Hartman <gregkh@suse.de>
Mon, 29 Nov 2010 19:09:13 +0000 (11:09 -0800)
The hash implementation is a complete implementation of a hash using
buckets as hash entries and overflow buckets attached to them.

The kernel already provides datastructures hlist_head and hlist_node
which can be used to implement an hash using lists as hash buckets. So
it is better to implement heavily used functionality on top of those
instead of providing a full hash implementation.

The rewrite changes the behavior of some functions slightly:
 * hash_add add elements to the front instead of the tail
 * hash_iterate doesn't provide pointer to access bucket->data directly,
   but it can be accessed using hlist_entry

Reported-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Sven Eckelmann <sven.eckelmann@gmx.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
drivers/staging/batman-adv/TODO
drivers/staging/batman-adv/hash.c
drivers/staging/batman-adv/hash.h
drivers/staging/batman-adv/originator.c
drivers/staging/batman-adv/routing.c
drivers/staging/batman-adv/translation-table.c
drivers/staging/batman-adv/vis.c

index 91b5e9cd48d56601d7d1108c85e7f79c384a6d46..ba69ba3ab62d85d35813713ecb915e24cfb66061 100644 (file)
@@ -1,5 +1,3 @@
- * remove own list functionality from hash
- * use hlist_head, hlist_node in hash
  * Request a new review
  * Process the comments from the review
  * Move into mainline proper
index bfe943cac1ab462ff276577c4cc38291aa99417d..8605e2fd0427254e9971ec790b118da10ad76494 100644 (file)
@@ -30,7 +30,7 @@ static void hash_init(struct hashtable_t *hash)
        hash->elements = 0;
 
        for (i = 0 ; i < hash->size; i++)
-               hash->table[i] = NULL;
+               INIT_HLIST_HEAD(&hash->table[i]);
 }
 
 /* free only the hashtable and the hash itself. */
@@ -70,15 +70,13 @@ struct hashtable_t *hash_new(int size)
 void *hash_remove_bucket(struct hashtable_t *hash, struct hash_it_t *hash_it_t)
 {
        void *data_save;
+       struct element_t *bucket;
 
-       data_save = hash_it_t->bucket->data;
+       bucket = hlist_entry(hash_it_t->walk, struct element_t, hlist);
+       data_save = bucket->data;
 
-       if (hash_it_t->prev_bucket != NULL)
-               hash_it_t->prev_bucket->next = hash_it_t->bucket->next;
-       else if (hash_it_t->first_bucket != NULL)
-               (*hash_it_t->first_bucket) = hash_it_t->bucket->next;
-
-       kfree(hash_it_t->bucket);
+       hlist_del(hash_it_t->walk);
+       kfree(bucket);
        hash->elements--;
 
        return data_save;
index a8e4dd1ec6a0b47d631ee4f40612c463ef2ebba7..0b61c6e3e7cf7520c97a82a6f7b648751360ade9 100644 (file)
 #ifndef _NET_BATMAN_ADV_HASH_H_
 #define _NET_BATMAN_ADV_HASH_H_
 
+#include <linux/list.h>
+
 #define HASHIT(name) struct hash_it_t name = { \
-               .index = -1, .bucket = NULL, \
-               .prev_bucket = NULL, \
-               .first_bucket = NULL }
+               .index = 0, .walk = NULL, \
+               .safe = NULL}
 
 /* callback to a compare function.  should
  * compare 2 element datas for their keys,
@@ -41,18 +42,17 @@ typedef void (*hashdata_free_cb)(void *, void *);
 
 struct element_t {
        void *data;             /* pointer to the data */
-       struct element_t *next; /* overflow bucket pointer */
+       struct hlist_node hlist;        /* bucket list pointer */
 };
 
 struct hash_it_t {
-       int index;
-       struct element_t *bucket;
-       struct element_t *prev_bucket;
-       struct element_t **first_bucket;
+       size_t index;
+       struct hlist_node *walk;
+       struct hlist_node *safe;
 };
 
 struct hashtable_t {
-       struct element_t **table;   /* the hashtable itself, with the buckets */
+       struct hlist_head *table;   /* the hashtable itself, with the buckets */
        int elements;               /* number of elements registered */
        int size;                   /* size of hashtable */
 };
@@ -75,19 +75,21 @@ void hash_destroy(struct hashtable_t *hash);
 static inline void hash_delete(struct hashtable_t *hash,
                               hashdata_free_cb free_cb, void *arg)
 {
-       struct element_t *bucket, *last_bucket;
+       struct hlist_head *head;
+       struct hlist_node *walk, *safe;
+       struct element_t *bucket;
        int i;
 
        for (i = 0; i < hash->size; i++) {
-               bucket = hash->table[i];
+               head = &hash->table[i];
 
-               while (bucket != NULL) {
+               hlist_for_each_safe(walk, safe, head) {
+                       bucket = hlist_entry(walk, struct element_t, hlist);
                        if (free_cb != NULL)
                                free_cb(bucket->data, arg);
 
-                       last_bucket = bucket;
-                       bucket = bucket->next;
-                       kfree(last_bucket);
+                       hlist_del(walk);
+                       kfree(bucket);
                }
        }
 
@@ -100,36 +102,30 @@ static inline int hash_add(struct hashtable_t *hash,
                           hashdata_choose_cb choose, void *data)
 {
        int index;
-       struct element_t *bucket, *prev_bucket = NULL;
+       struct hlist_head *head;
+       struct hlist_node *walk, *safe;
+       struct element_t *bucket;
 
        if (!hash)
                return -1;
 
        index = choose(data, hash->size);
-       bucket = hash->table[index];
+       head = &hash->table[index];
 
-       while (bucket != NULL) {
+       hlist_for_each_safe(walk, safe, head) {
+               bucket = hlist_entry(walk, struct element_t, hlist);
                if (compare(bucket->data, data))
                        return -1;
-
-               prev_bucket = bucket;
-               bucket = bucket->next;
        }
 
-       /* found the tail of the list, add new element */
+       /* no duplicate found in list, add new element */
        bucket = kmalloc(sizeof(struct element_t), GFP_ATOMIC);
 
        if (bucket == NULL)
                return -1;
 
        bucket->data = data;
-       bucket->next = NULL;
-
-       /* and link it */
-       if (prev_bucket == NULL)
-               hash->table[index] = bucket;
-       else
-               prev_bucket->next = bucket;
+       hlist_add_head(&bucket->hlist, head);
 
        hash->elements++;
        return 0;
@@ -144,22 +140,16 @@ static inline void *hash_remove(struct hashtable_t *hash,
                                hashdata_choose_cb choose, void *data)
 {
        struct hash_it_t hash_it_t;
+       struct element_t *bucket;
+       struct hlist_head *head;
 
        hash_it_t.index = choose(data, hash->size);
-       hash_it_t.bucket = hash->table[hash_it_t.index];
-       hash_it_t.prev_bucket = NULL;
-
-       while (hash_it_t.bucket != NULL) {
-               if (compare(hash_it_t.bucket->data, data)) {
-                       hash_it_t.first_bucket =
-                               (hash_it_t.bucket ==
-                                hash->table[hash_it_t.index] ?
-                                &hash->table[hash_it_t.index] : NULL);
-                       return hash_remove_bucket(hash, &hash_it_t);
-               }
+       head = &hash->table[hash_it_t.index];
 
-               hash_it_t.prev_bucket = hash_it_t.bucket;
-               hash_it_t.bucket = hash_it_t.bucket->next;
+       hlist_for_each(hash_it_t.walk, head) {
+               bucket = hlist_entry(hash_it_t.walk, struct element_t, hlist);
+               if (compare(bucket->data, data))
+                       return hash_remove_bucket(hash, &hash_it_t);
        }
 
        return NULL;
@@ -172,19 +162,20 @@ static inline void *hash_find(struct hashtable_t *hash,
                              hashdata_choose_cb choose, void *keydata)
 {
        int index;
+       struct hlist_head *head;
+       struct hlist_node *walk;
        struct element_t *bucket;
 
        if (!hash)
                return NULL;
 
        index = choose(keydata , hash->size);
-       bucket = hash->table[index];
+       head = &hash->table[index];
 
-       while (bucket != NULL) {
+       hlist_for_each(walk, head) {
+               bucket = hlist_entry(walk, struct element_t, hlist);
                if (compare(bucket->data, keydata))
                        return bucket->data;
-
-               bucket = bucket->next;
        }
 
        return NULL;
@@ -193,13 +184,14 @@ static inline void *hash_find(struct hashtable_t *hash,
 /* resize the hash, returns the pointer to the new hash or NULL on
  * error. removes the old hash on success */
 static inline struct hashtable_t *hash_resize(struct hashtable_t *hash,
-                                             hashdata_compare_cb compare,
                                              hashdata_choose_cb choose,
                                              int size)
 {
        struct hashtable_t *new_hash;
+       struct hlist_head *head, *new_head;
+       struct hlist_node *walk, *safe;
        struct element_t *bucket;
-       int i;
+       int i, new_index;
 
        /* initialize a new hash with the new size */
        new_hash = hash_new(size);
@@ -209,17 +201,20 @@ static inline struct hashtable_t *hash_resize(struct hashtable_t *hash,
 
        /* copy the elements */
        for (i = 0; i < hash->size; i++) {
-               bucket = hash->table[i];
+               head = &hash->table[i];
+
+               hlist_for_each_safe(walk, safe, head) {
+                       bucket = hlist_entry(walk, struct element_t, hlist);
 
-               while (bucket != NULL) {
-                       hash_add(new_hash, compare, choose, bucket->data);
-                       bucket = bucket->next;
+                       new_index = choose(bucket->data, size);
+                       new_head = &new_hash->table[new_index];
+
+                       hlist_del(walk);
+                       hlist_add_head(walk, new_head);
                }
        }
 
-       /* remove hash and eventual overflow buckets but not the content
-        * itself. */
-       hash_delete(hash, NULL, NULL);
+       hash_destroy(hash);
 
        return new_hash;
 }
@@ -236,63 +231,29 @@ static inline struct hash_it_t *hash_iterate(struct hashtable_t *hash,
        if (!iter)
                return NULL;
 
-       /* sanity checks first (if our bucket got deleted in the last
-        * iteration): */
-       if (iter->bucket != NULL) {
-               if (iter->first_bucket != NULL) {
-                       /* we're on the first element and it got removed after
-                        * the last iteration. */
-                       if ((*iter->first_bucket) != iter->bucket) {
-                               /* there are still other elements in the list */
-                               if ((*iter->first_bucket) != NULL) {
-                                       iter->prev_bucket = NULL;
-                                       iter->bucket = (*iter->first_bucket);
-                                       iter->first_bucket =
-                                               &hash->table[iter->index];
-                                       return iter;
-                               } else {
-                                       iter->bucket = NULL;
-                               }
-                       }
-               } else if (iter->prev_bucket != NULL) {
-                       /*
-                       * we're not on the first element, and the bucket got
-                       * removed after the last iteration.  the last bucket's
-                       * next pointer is not pointing to our actual bucket
-                       * anymore.  select the next.
-                       */
-                       if (iter->prev_bucket->next != iter->bucket)
-                               iter->bucket = iter->prev_bucket;
-               }
-       }
+       iter->walk = iter->safe;
 
-       /* now as we are sane, select the next one if there is some */
-       if (iter->bucket != NULL) {
-               if (iter->bucket->next != NULL) {
-                       iter->prev_bucket = iter->bucket;
-                       iter->bucket = iter->bucket->next;
-                       iter->first_bucket = NULL;
-                       return iter;
-               }
-       }
+       /* we search for the next head with list entries */
+       if (!iter->walk) {
+               while (iter->index < hash->size) {
+                       if (hlist_empty(&hash->table[iter->index]))
+                               iter->index++;
+                       else {
+                               iter->walk = hash->table[iter->index].first;
 
-       /* if not returned yet, we've reached the last one on the index and have
-        * to search forward */
-       iter->index++;
-       /* go through the entries of the hash table */
-       while (iter->index < hash->size) {
-               if ((hash->table[iter->index]) != NULL) {
-                       iter->prev_bucket = NULL;
-                       iter->bucket = hash->table[iter->index];
-                       iter->first_bucket = &hash->table[iter->index];
-                       return iter;
-               } else {
-                       iter->index++;
+                               /* search next time */
+                               ++iter->index;
+                               break;
+                       }
                }
        }
 
-       /* nothing to iterate over anymore */
-       return NULL;
+       /* return iter when we found bucket otherwise null */
+       if (!iter->walk)
+               return NULL;
+
+       iter->safe = iter->walk->next;
+       return iter;
 }
 
 #endif /* _NET_BATMAN_ADV_HASH_H_ */
index 7c1fae7a2fac48e1801ffa7fb323d69fc5537022..a4b7d377d94150771712a201bce7950427ec21aa 100644 (file)
@@ -175,8 +175,7 @@ struct orig_node *get_orig_node(struct bat_priv *bat_priv, uint8_t *addr)
                goto free_bcast_own_sum;
 
        if (bat_priv->orig_hash->elements * 4 > bat_priv->orig_hash->size) {
-               swaphash = hash_resize(bat_priv->orig_hash, compare_orig,
-                                      choose_orig,
+               swaphash = hash_resize(bat_priv->orig_hash, choose_orig,
                                       bat_priv->orig_hash->size * 2);
 
                if (!swaphash)
@@ -272,6 +271,7 @@ static bool purge_orig_node(struct bat_priv *bat_priv,
 static void _purge_orig(struct bat_priv *bat_priv)
 {
        HASHIT(hashit);
+       struct element_t *bucket;
        struct orig_node *orig_node;
        unsigned long flags;
 
@@ -279,7 +279,8 @@ static void _purge_orig(struct bat_priv *bat_priv)
 
        /* for all origins... */
        while (hash_iterate(bat_priv->orig_hash, &hashit)) {
-               orig_node = hashit.bucket->data;
+               bucket = hlist_entry(hashit.walk, struct element_t, hlist);
+               orig_node = bucket->data;
 
                if (purge_orig_node(bat_priv, orig_node)) {
                        hash_remove_bucket(bat_priv->orig_hash, &hashit);
@@ -315,6 +316,7 @@ void purge_orig_ref(struct bat_priv *bat_priv)
 int orig_seq_print_text(struct seq_file *seq, void *offset)
 {
        HASHIT(hashit);
+       struct element_t *bucket;
        struct net_device *net_dev = (struct net_device *)seq->private;
        struct bat_priv *bat_priv = netdev_priv(net_dev);
        struct orig_node *orig_node;
@@ -347,8 +349,8 @@ int orig_seq_print_text(struct seq_file *seq, void *offset)
        spin_lock_irqsave(&bat_priv->orig_hash_lock, flags);
 
        while (hash_iterate(bat_priv->orig_hash, &hashit)) {
-
-               orig_node = hashit.bucket->data;
+               bucket = hlist_entry(hashit.walk, struct element_t, hlist);
+               orig_node = bucket->data;
 
                if (!orig_node->router)
                        continue;
@@ -419,13 +421,15 @@ int orig_hash_add_if(struct batman_if *batman_if, int max_if_num)
        struct orig_node *orig_node;
        unsigned long flags;
        HASHIT(hashit);
+       struct element_t *bucket;
 
        /* resize all orig nodes because orig_node->bcast_own(_sum) depend on
         * if_num */
        spin_lock_irqsave(&bat_priv->orig_hash_lock, flags);
 
        while (hash_iterate(bat_priv->orig_hash, &hashit)) {
-               orig_node = hashit.bucket->data;
+               bucket = hlist_entry(hashit.walk, struct element_t, hlist);
+               orig_node = bucket->data;
 
                if (orig_node_add_if(orig_node, max_if_num) == -1)
                        goto err;
@@ -498,6 +502,7 @@ int orig_hash_del_if(struct batman_if *batman_if, int max_if_num)
        struct orig_node *orig_node;
        unsigned long flags;
        HASHIT(hashit);
+       struct element_t *bucket;
        int ret;
 
        /* resize all orig nodes because orig_node->bcast_own(_sum) depend on
@@ -505,7 +510,8 @@ int orig_hash_del_if(struct batman_if *batman_if, int max_if_num)
        spin_lock_irqsave(&bat_priv->orig_hash_lock, flags);
 
        while (hash_iterate(bat_priv->orig_hash, &hashit)) {
-               orig_node = hashit.bucket->data;
+               bucket = hlist_entry(hashit.walk, struct element_t, hlist);
+               orig_node = bucket->data;
 
                ret = orig_node_del_if(orig_node, max_if_num,
                                       batman_if->if_num);
index 9cbb19594e34379c6c0c1acc4746fce76031dc91..77e5d1411c99883a00e4ef540b8e9f66637a8a7b 100644 (file)
@@ -38,6 +38,7 @@ void slide_own_bcast_window(struct batman_if *batman_if)
 {
        struct bat_priv *bat_priv = netdev_priv(batman_if->soft_iface);
        HASHIT(hashit);
+       struct element_t *bucket;
        struct orig_node *orig_node;
        TYPE_OF_WORD *word;
        unsigned long flags;
@@ -45,7 +46,8 @@ void slide_own_bcast_window(struct batman_if *batman_if)
        spin_lock_irqsave(&bat_priv->orig_hash_lock, flags);
 
        while (hash_iterate(bat_priv->orig_hash, &hashit)) {
-               orig_node = hashit.bucket->data;
+               bucket = hlist_entry(hashit.walk, struct element_t, hlist);
+               orig_node = bucket->data;
                word = &(orig_node->bcast_own[batman_if->if_num * NUM_WORDS]);
 
                bit_get_packet(bat_priv, word, 1, 0);
index 96d59b100df897886fb9c5479c3f79ca7152c565..6639bfb66ab840b7e60f589b9ff8696b3cf0d0c1 100644 (file)
@@ -116,8 +116,7 @@ void hna_local_add(struct net_device *soft_iface, uint8_t *addr)
 
        if (bat_priv->hna_local_hash->elements * 4 >
                                        bat_priv->hna_local_hash->size) {
-               swaphash = hash_resize(bat_priv->hna_local_hash, compare_orig,
-                                      choose_orig,
+               swaphash = hash_resize(bat_priv->hna_local_hash, choose_orig,
                                       bat_priv->hna_local_hash->size * 2);
 
                if (!swaphash)
@@ -146,6 +145,7 @@ int hna_local_fill_buffer(struct bat_priv *bat_priv,
                          unsigned char *buff, int buff_len)
 {
        struct hna_local_entry *hna_local_entry;
+       struct element_t *bucket;
        HASHIT(hashit);
        int i = 0;
        unsigned long flags;
@@ -157,7 +157,8 @@ int hna_local_fill_buffer(struct bat_priv *bat_priv,
                if (buff_len < (i + 1) * ETH_ALEN)
                        break;
 
-               hna_local_entry = hashit.bucket->data;
+               bucket = hlist_entry(hashit.walk, struct element_t, hlist);
+               hna_local_entry = bucket->data;
                memcpy(buff + (i * ETH_ALEN), hna_local_entry->addr, ETH_ALEN);
 
                i++;
@@ -178,6 +179,7 @@ int hna_local_seq_print_text(struct seq_file *seq, void *offset)
        struct hna_local_entry *hna_local_entry;
        HASHIT(hashit);
        HASHIT(hashit_count);
+       struct element_t *bucket;
        unsigned long flags;
        size_t buf_size, pos;
        char *buff;
@@ -208,7 +210,8 @@ int hna_local_seq_print_text(struct seq_file *seq, void *offset)
        pos = 0;
 
        while (hash_iterate(bat_priv->hna_local_hash, &hashit)) {
-               hna_local_entry = hashit.bucket->data;
+               bucket = hlist_entry(hashit.walk, struct element_t, hlist);
+               hna_local_entry = bucket->data;
 
                pos += snprintf(buff + pos, 22, " * %pM\n",
                                hna_local_entry->addr);
@@ -267,13 +270,15 @@ static void hna_local_purge(struct work_struct *work)
                container_of(delayed_work, struct bat_priv, hna_work);
        struct hna_local_entry *hna_local_entry;
        HASHIT(hashit);
+       struct element_t *bucket;
        unsigned long flags;
        unsigned long timeout;
 
        spin_lock_irqsave(&bat_priv->hna_lhash_lock, flags);
 
        while (hash_iterate(bat_priv->hna_local_hash, &hashit)) {
-               hna_local_entry = hashit.bucket->data;
+               bucket = hlist_entry(hashit.walk, struct element_t, hlist);
+               hna_local_entry = bucket->data;
 
                timeout = hna_local_entry->last_seen + LOCAL_HNA_TIMEOUT * HZ;
 
@@ -389,8 +394,7 @@ void hna_global_add_orig(struct bat_priv *bat_priv,
 
        if (bat_priv->hna_global_hash->elements * 4 >
                                        bat_priv->hna_global_hash->size) {
-               swaphash = hash_resize(bat_priv->hna_global_hash, compare_orig,
-                                      choose_orig,
+               swaphash = hash_resize(bat_priv->hna_global_hash, choose_orig,
                                       bat_priv->hna_global_hash->size * 2);
 
                if (!swaphash)
@@ -409,6 +413,7 @@ int hna_global_seq_print_text(struct seq_file *seq, void *offset)
        struct hna_global_entry *hna_global_entry;
        HASHIT(hashit);
        HASHIT(hashit_count);
+       struct element_t *bucket;
        unsigned long flags;
        size_t buf_size, pos;
        char *buff;
@@ -438,7 +443,8 @@ int hna_global_seq_print_text(struct seq_file *seq, void *offset)
        pos = 0;
 
        while (hash_iterate(bat_priv->hna_global_hash, &hashit)) {
-               hna_global_entry = hashit.bucket->data;
+               bucket = hlist_entry(hashit.walk, struct element_t, hlist);
+               hna_global_entry = bucket->data;
 
                pos += snprintf(buff + pos, 44,
                                " * %pM via %pM\n", hna_global_entry->addr,
index dccd296c6ff382fd9c8ed440df9df8048fc6adba..e2031c9aab9da8953ca1b53ca64a22b2dc24c2ed 100644 (file)
@@ -177,6 +177,7 @@ int vis_seq_print_text(struct seq_file *seq, void *offset)
 {
        HASHIT(hashit);
        HASHIT(hashit_count);
+       struct element_t *bucket;
        struct vis_info *info;
        struct vis_packet *packet;
        struct vis_info_entry *entries;
@@ -199,7 +200,9 @@ int vis_seq_print_text(struct seq_file *seq, void *offset)
        /* Estimate length */
        spin_lock_irqsave(&bat_priv->vis_hash_lock, flags);
        while (hash_iterate(bat_priv->vis_hash, &hashit_count)) {
-               info = hashit_count.bucket->data;
+               bucket = hlist_entry(hashit_count.walk, struct element_t,
+                                    hlist);
+               info = bucket->data;
                packet = (struct vis_packet *)info->skb_packet->data;
                entries = (struct vis_info_entry *)
                          ((char *)packet + sizeof(struct vis_packet));
@@ -237,7 +240,8 @@ int vis_seq_print_text(struct seq_file *seq, void *offset)
        buff_pos = 0;
 
        while (hash_iterate(bat_priv->vis_hash, &hashit)) {
-               info = hashit.bucket->data;
+               bucket = hlist_entry(hashit.walk, struct element_t, hlist);
+               info = bucket->data;
                packet = (struct vis_packet *)info->skb_packet->data;
                entries = (struct vis_info_entry *)
                          ((char *)packet + sizeof(struct vis_packet));
@@ -515,6 +519,7 @@ static int find_best_vis_server(struct bat_priv *bat_priv,
                                struct vis_info *info)
 {
        HASHIT(hashit);
+       struct element_t *bucket;
        struct orig_node *orig_node;
        struct vis_packet *packet;
        int best_tq = -1;
@@ -522,7 +527,8 @@ static int find_best_vis_server(struct bat_priv *bat_priv,
        packet = (struct vis_packet *)info->skb_packet->data;
 
        while (hash_iterate(bat_priv->orig_hash, &hashit)) {
-               orig_node = hashit.bucket->data;
+               bucket = hlist_entry(hashit.walk, struct element_t, hlist);
+               orig_node = bucket->data;
                if ((orig_node) && (orig_node->router) &&
                    (orig_node->flags & VIS_SERVER) &&
                    (orig_node->router->tq_avg > best_tq)) {
@@ -551,6 +557,7 @@ static int generate_vis_packet(struct bat_priv *bat_priv)
 {
        HASHIT(hashit_local);
        HASHIT(hashit_global);
+       struct element_t *bucket;
        struct orig_node *orig_node;
        struct vis_info *info = (struct vis_info *)bat_priv->my_vis_info;
        struct vis_packet *packet = (struct vis_packet *)info->skb_packet->data;
@@ -580,7 +587,9 @@ static int generate_vis_packet(struct bat_priv *bat_priv)
        }
 
        while (hash_iterate(bat_priv->orig_hash, &hashit_global)) {
-               orig_node = hashit_global.bucket->data;
+               bucket = hlist_entry(hashit_global.walk, struct element_t,
+                                    hlist);
+               orig_node = bucket->data;
 
                if (!orig_node->router)
                        continue;
@@ -615,7 +624,9 @@ static int generate_vis_packet(struct bat_priv *bat_priv)
 
        spin_lock_irqsave(&bat_priv->hna_lhash_lock, flags);
        while (hash_iterate(bat_priv->hna_local_hash, &hashit_local)) {
-               hna_local_entry = hashit_local.bucket->data;
+               bucket = hlist_entry(hashit_local.walk, struct element_t,
+                                    hlist);
+               hna_local_entry = bucket->data;
                entry = (struct vis_info_entry *)skb_put(info->skb_packet,
                                                         sizeof(*entry));
                memset(entry->src, 0, ETH_ALEN);
@@ -639,10 +650,12 @@ static int generate_vis_packet(struct bat_priv *bat_priv)
 static void purge_vis_packets(struct bat_priv *bat_priv)
 {
        HASHIT(hashit);
+       struct element_t *bucket;
        struct vis_info *info;
 
        while (hash_iterate(bat_priv->vis_hash, &hashit)) {
-               info = hashit.bucket->data;
+               bucket = hlist_entry(hashit.walk, struct element_t, hlist);
+               info = bucket->data;
 
                /* never purge own data. */
                if (info == bat_priv->my_vis_info)
@@ -661,6 +674,7 @@ static void broadcast_vis_packet(struct bat_priv *bat_priv,
                                 struct vis_info *info)
 {
        HASHIT(hashit);
+       struct element_t *bucket;
        struct orig_node *orig_node;
        struct vis_packet *packet;
        struct sk_buff *skb;
@@ -674,7 +688,8 @@ static void broadcast_vis_packet(struct bat_priv *bat_priv,
 
        /* send to all routers in range. */
        while (hash_iterate(bat_priv->orig_hash, &hashit)) {
-               orig_node = hashit.bucket->data;
+               bucket = hlist_entry(hashit.walk, struct element_t, hlist);
+               orig_node = bucket->data;
 
                /* if it's a vis server and reachable, send it. */
                if ((!orig_node) || (!orig_node->router))