2 * Copyright (c) 2009, 2010, 2011 Nicira Networks.
3 * Distributed under the terms of the GNU GPL version 2.
5 * Significant portions of this file may be copied from parts of the Linux
6 * kernel, by Linus Torvalds and others.
13 #include <linux/gfp.h>
14 #include <linux/slab.h>
16 #include <asm/pgtable.h>
19 * struct tbl_bucket - single bucket within a hash table
20 * @rcu: RCU callback structure
21 * @n_objs: number of objects in @objs[] array
22 * @objs: array of @n_objs pointers to table nodes contained inside objects
24 * The expected number of objects per bucket is 1, but this allows for an
25 * arbitrary number of collisions.
30 struct tbl_node *objs[];
33 static inline int bucket_size(int n_objs)
35 return sizeof(struct tbl_bucket) + sizeof(struct tbl_node *) * n_objs;
38 static struct tbl_bucket *bucket_alloc(int n_objs)
40 return kmalloc(bucket_size(n_objs), GFP_KERNEL);
43 static void free_buckets(struct tbl_bucket __rcu ***l1,
44 unsigned int n_buckets,
45 void (*free_obj)(struct tbl_node *))
49 for (i = 0; i < n_buckets >> TBL_L1_SHIFT; i++) {
50 struct tbl_bucket __rcu **l2 = l1[i];
53 for (j = 0; j < TBL_L2_SIZE; j++) {
54 struct tbl_bucket *bucket = (struct tbl_bucket __force *)l2[j];
60 for (k = 0; k < bucket->n_objs; k++)
61 free_obj(bucket->objs[k]);
65 free_page((unsigned long)l2);
70 static struct tbl_bucket __rcu ***alloc_buckets(unsigned int n_buckets)
72 struct tbl_bucket __rcu ***l1;
75 l1 = kmalloc((n_buckets >> TBL_L1_SHIFT) * sizeof(struct tbl_bucket **),
79 for (i = 0; i < n_buckets >> TBL_L1_SHIFT; i++) {
80 l1[i] = (struct tbl_bucket __rcu **)get_zeroed_page(GFP_KERNEL);
82 free_buckets(l1, i << TBL_L1_SHIFT, NULL);
90 * tbl_create - create and return a new hash table
91 * @n_buckets: number of buckets in the new table
93 * Creates and returns a new hash table, or %NULL if memory cannot be
94 * allocated. @n_buckets must be a power of 2 in the range %TBL_MIN_BUCKETS to
97 struct tbl *tbl_create(unsigned int n_buckets)
101 table = kzalloc(sizeof(*table), GFP_KERNEL);
105 table->n_buckets = n_buckets;
106 table->buckets = alloc_buckets(n_buckets);
119 * tbl_destroy - destroy hash table and optionally the objects it contains
120 * @table: table to destroy
121 * @destructor: function to be called on objects at destruction time
123 * If a destructor is null, then the buckets in @table are destroyed
124 * but not the objects within those buckets. This behavior is useful when a
125 * table is being replaced by a larger or smaller one without destroying the
128 * If a destructor is not null, then it is called on the objects in @table
129 * before destroying the buckets.
131 void tbl_destroy(struct tbl *table, void (*destructor)(struct tbl_node *))
136 free_buckets(table->buckets, table->n_buckets, destructor);
140 static void destroy_table_rcu(struct rcu_head *rcu)
142 struct tbl *table = container_of(rcu, struct tbl, rcu);
143 tbl_destroy(table, table->obj_destructor);
147 * tbl_deferred_destroy - destroy table after a RCU grace period
148 * @table: table to destroy
149 * @destructor: function to be called on objects at destruction time
151 * Calls tbl_destroy() on @table after an RCU grace period. If @destructor is
152 * not null it is called on every element before the table is destroyed. */
153 void tbl_deferred_destroy(struct tbl *table, void (*destructor)(struct tbl_node *))
158 table->obj_destructor = destructor;
159 call_rcu(&table->rcu, destroy_table_rcu);
162 static struct tbl_bucket __rcu **find_bucket(struct tbl *table, u32 hash)
164 unsigned int l1 = (hash & (table->n_buckets - 1)) >> TBL_L1_SHIFT;
165 unsigned int l2 = hash & ((1 << TBL_L2_BITS) - 1);
166 return &table->buckets[l1][l2];
169 static int search_bucket(const struct tbl_bucket *bucket, void *target, u32 hash,
170 int (*cmp)(const struct tbl_node *, void *))
174 for (i = 0; i < bucket->n_objs; i++) {
175 struct tbl_node *obj = bucket->objs[i];
176 if (obj->hash == hash && likely(cmp(obj, target)))
184 * tbl_lookup - searches hash table for a matching object
185 * @table: hash table to search
186 * @target: identifier for the object that is being searched for, will be
187 * provided as an argument to @cmp when making comparisions
188 * @hash: hash of @target
189 * @cmp: comparision function to match objects with the given hash, returns
190 * nonzero if the objects match, zero otherwise
192 * Searches @table for an object identified by @target. Returns the tbl_node
193 * contained in the object if successful, otherwise %NULL.
195 struct tbl_node *tbl_lookup(struct tbl *table, void *target, u32 hash,
196 int (*cmp)(const struct tbl_node *, void *))
198 struct tbl_bucket __rcu **bucketp = find_bucket(table, hash);
199 struct tbl_bucket *bucket = rcu_dereference(*bucketp);
205 index = search_bucket(bucket, target, hash, cmp);
209 return bucket->objs[index];
213 * tbl_foreach - iterate through hash table
214 * @table: table to iterate
215 * @callback: function to call for each entry
216 * @aux: Extra data to pass to @callback
218 * Iterates through all of the objects in @table in hash order, passing each of
219 * them in turn to @callback. If @callback returns nonzero, this terminates
220 * the iteration and tbl_foreach() returns the same value. Returns 0 if
221 * @callback never returns nonzero.
223 * This function does not try to intelligently handle the case where @callback
224 * adds or removes flows in @table.
226 int tbl_foreach(struct tbl *table,
227 int (*callback)(struct tbl_node *, void *aux), void *aux)
229 unsigned int n_l1 = table->n_buckets >> TBL_L1_SHIFT;
232 for (l1_idx = 0; l1_idx < n_l1; l1_idx++) {
233 struct tbl_bucket __rcu **l2 = table->buckets[l1_idx];
236 for (l2_idx = 0; l2_idx < TBL_L2_SIZE; l2_idx++) {
237 struct tbl_bucket *bucket;
240 bucket = rcu_dereference(l2[l2_idx]);
244 for (i = 0; i < bucket->n_objs; i++) {
245 int error = (*callback)(bucket->objs[i], aux);
254 static int insert_table_flow(struct tbl_node *node, void *new_table_)
256 struct tbl *new_table = new_table_;
257 return tbl_insert(new_table, node, node->hash);
261 * tbl_expand - create a hash table with more buckets
262 * @table: table to expand
264 * Creates a new table containing the same objects as @table but with twice
265 * as many buckets. Returns 0 if successful, otherwise a negative error. The
266 * caller should free @table upon success (probably using
267 * tbl_deferred_destroy()).
269 struct tbl *tbl_expand(struct tbl *table)
272 int n_buckets = table->n_buckets * 2;
273 struct tbl *new_table;
275 if (n_buckets >= TBL_MAX_BUCKETS) {
281 new_table = tbl_create(TBL_MIN_BUCKETS);
285 if (tbl_foreach(table, insert_table_flow, new_table))
286 goto error_free_new_table;
290 error_free_new_table:
291 tbl_destroy(new_table, NULL);
297 * tbl_n_buckets - returns the number of buckets
298 * @table: table to examine
300 * Returns the number of buckets currently allocated in @table, useful when
301 * deciding whether to expand.
303 int tbl_n_buckets(struct tbl *table)
305 return table->n_buckets;
308 static void free_bucket_rcu(struct rcu_head *rcu)
310 struct tbl_bucket *bucket = container_of(rcu, struct tbl_bucket, rcu);
315 * tbl_insert - insert object into table
316 * @table: table in which to insert object
317 * @target: tbl_node contained in object to insert
318 * @hash: hash of object to insert
320 * The caller must ensure that no object considered to be identical to @target
321 * already exists in @table. Returns 0 or a negative error (currently just
324 int tbl_insert(struct tbl *table, struct tbl_node *target, u32 hash)
326 struct tbl_bucket __rcu **oldp = find_bucket(table, hash);
327 struct tbl_bucket *old = rcu_dereference(*oldp);
328 unsigned int n = old ? old->n_objs : 0;
329 struct tbl_bucket *new = bucket_alloc(n + 1);
338 memcpy(new->objs, old->objs, n * sizeof(struct tbl_node *));
339 new->objs[n] = target;
341 rcu_assign_pointer(*oldp, new);
343 call_rcu(&old->rcu, free_bucket_rcu);
351 * tbl_remove - remove object from table
352 * @table: table from which to remove object
353 * @target: tbl_node inside of object to remove
355 * The caller must ensure that @target itself is in @table. (It is not
356 * good enough for @table to contain a different object considered identical
359 * Returns 0 or a negative error (currently just -ENOMEM). Yes, it *is*
360 * possible for object deletion to fail due to lack of memory.
362 int tbl_remove(struct tbl *table, struct tbl_node *target)
364 struct tbl_bucket __rcu **oldp = find_bucket(table, target->hash);
365 struct tbl_bucket *old = rcu_dereference(*oldp);
366 unsigned int n = old->n_objs;
367 struct tbl_bucket *new;
372 new = bucket_alloc(n - 1);
377 for (i = 0; i < n; i++) {
378 struct tbl_node *obj = old->objs[i];
380 new->objs[new->n_objs++] = obj;
382 WARN_ON_ONCE(new->n_objs != n - 1);
387 rcu_assign_pointer(*oldp, new);
388 call_rcu(&old->rcu, free_bucket_rcu);
396 * tbl_count - retrieves the number of stored objects
397 * @table: table to count
399 * Returns the number of objects that have been inserted into the hash table.
401 unsigned int tbl_count(struct tbl *table)