Don't use atomic_t for tracking table flow counts.
[openvswitch] / datapath / table-hash.c
1 /*
2  * Distributed under the terms of the GNU GPL version 2.
3  * Copyright (c) 2007, 2008 The Board of Trustees of The Leland 
4  * Stanford Junior University
5  */
6
7 #include "table.h"
8 #include "crc32.h"
9 #include "flow.h"
10 #include "datapath.h"
11
12 #include <linux/slab.h>
13 #include <linux/vmalloc.h>
14 #include <linux/mm.h>
15 #include <linux/highmem.h>
16 #include <asm/pgtable.h>
17
18 static void *kmem_alloc(size_t);
19 static void *kmem_zalloc(size_t);
20 static void kmem_free(void *, size_t);
21
22 struct sw_table_hash {
23         struct sw_table swt;
24         struct crc32 crc32;
25         unsigned int n_flows;
26         unsigned int bucket_mask; /* Number of buckets minus 1. */
27         struct sw_flow **buckets;
28 };
29
30 static struct sw_flow **find_bucket(struct sw_table *swt,
31                                                                         const struct sw_flow_key *key)
32 {
33         struct sw_table_hash *th = (struct sw_table_hash *) swt;
34         unsigned int crc = crc32_calculate(&th->crc32, key, sizeof *key);
35         return &th->buckets[crc & th->bucket_mask];
36 }
37
38 static struct sw_flow *table_hash_lookup(struct sw_table *swt,
39                                                                                  const struct sw_flow_key *key)
40 {
41         struct sw_flow *flow = *find_bucket(swt, key);
42         return flow && !memcmp(&flow->key, key, sizeof *key) ? flow : NULL;
43 }
44
45 static int table_hash_insert(struct sw_table *swt, struct sw_flow *flow)
46 {
47         struct sw_table_hash *th = (struct sw_table_hash *) swt;
48         struct sw_flow **bucket;
49         int retval;
50
51         if (flow->key.wildcards != 0)
52                 return 0;
53
54         bucket = find_bucket(swt, &flow->key);
55         if (*bucket == NULL) {
56                 th->n_flows++;
57                 rcu_assign_pointer(*bucket, flow);
58                 retval = 1;
59         } else {
60                 struct sw_flow *old_flow = *bucket;
61                 if (!memcmp(&old_flow->key, &flow->key, sizeof flow->key)
62                                         && flow_del(old_flow)) {
63                         rcu_assign_pointer(*bucket, flow);
64                         flow_deferred_free(old_flow);
65                         retval = 1;
66                 } else {
67                         retval = 0;
68                 }
69         }
70         return retval;
71 }
72
73 /* Caller must update n_flows. */
74 static int do_delete(struct sw_flow **bucket, struct sw_flow *flow)
75 {
76         if (flow_del(flow)) {
77                 rcu_assign_pointer(*bucket, NULL);
78                 flow_deferred_free(flow);
79                 return 1;
80         }
81         return 0;
82 }
83
84 /* Returns number of deleted flows.  We can ignore the priority
85  * argument, since all exact-match entries are the same (highest)
86  * priority. */
87 static int table_hash_delete(struct sw_table *swt,
88                                                          const struct sw_flow_key *key, 
89                                                          uint16_t priority, int strict)
90 {
91         struct sw_table_hash *th = (struct sw_table_hash *) swt;
92         unsigned int count = 0;
93
94         if (key->wildcards == 0) {
95                 struct sw_flow **bucket = find_bucket(swt, key);
96                 struct sw_flow *flow = *bucket;
97                 if (flow && !memcmp(&flow->key, key, sizeof *key))
98                         count = do_delete(bucket, flow);
99         } else {
100                 unsigned int i;
101
102                 for (i = 0; i <= th->bucket_mask; i++) {
103                         struct sw_flow **bucket = &th->buckets[i];
104                         struct sw_flow *flow = *bucket;
105                         if (flow && flow_del_matches(&flow->key, key, strict))
106                                 count += do_delete(bucket, flow);
107                 }
108         }
109         th->n_flows -= count;
110         return count;
111 }
112
113 static int table_hash_timeout(struct datapath *dp, struct sw_table *swt)
114 {
115         struct sw_table_hash *th = (struct sw_table_hash *) swt;
116         unsigned int i;
117         int count = 0;
118
119         mutex_lock(&dp_mutex);
120         for (i = 0; i <= th->bucket_mask; i++) {
121                 struct sw_flow **bucket = &th->buckets[i];
122                 struct sw_flow *flow = *bucket;
123                 if (flow && flow_timeout(flow)) {
124                         count += do_delete(bucket, flow); 
125                         if (dp->flags & OFPC_SEND_FLOW_EXP)
126                                 dp_send_flow_expired(dp, flow);
127                 }
128         }
129         th->n_flows -= count;
130         mutex_unlock(&dp_mutex);
131
132         return count;
133 }
134
135 static void table_hash_destroy(struct sw_table *swt)
136 {
137         struct sw_table_hash *th = (struct sw_table_hash *) swt;
138         unsigned int i;
139         for (i = 0; i <= th->bucket_mask; i++)
140         if (th->buckets[i])
141                 flow_free(th->buckets[i]);
142         kmem_free(th->buckets, (th->bucket_mask + 1) * sizeof *th->buckets);
143         kfree(th);
144 }
145
146 static int table_hash_iterate(struct sw_table *swt,
147                               const struct sw_flow_key *key,
148                               struct sw_table_position *position,
149                               int (*callback)(struct sw_flow *, void *private),
150                               void *private) 
151 {
152         struct sw_table_hash *th = (struct sw_table_hash *) swt;
153
154         if (position->private[0] > th->bucket_mask)
155                 return 0;
156
157         if (key->wildcards == 0) {
158                 struct sw_flow *flow;
159                 int error;
160
161                 flow = table_hash_lookup(swt, key);
162                 if (!flow)
163                         return 0;
164
165                 error = callback(flow, private);
166                 if (!error)
167                         position->private[0] = -1;
168                 return error;
169         } else {
170                 int i;
171
172                 for (i = position->private[0]; i <= th->bucket_mask; i++) {
173                         struct sw_flow *flow = th->buckets[i];
174                         if (flow && flow_matches(key, &flow->key)) {
175                                 int error = callback(flow, private);
176                                 if (error) {
177                                         position->private[0] = i;
178                                         return error;
179                                 }
180                         }
181                 }
182                 return 0;
183         }
184 }
185 static void table_hash_stats(struct sw_table *swt,
186                                  struct sw_table_stats *stats) 
187 {
188         struct sw_table_hash *th = (struct sw_table_hash *) swt;
189         stats->name = "hash";
190         stats->n_flows = th->n_flows;
191         stats->max_flows = th->bucket_mask + 1;
192 }
193
194 struct sw_table *table_hash_create(unsigned int polynomial,
195                         unsigned int n_buckets)
196 {
197         struct sw_table_hash *th;
198         struct sw_table *swt;
199
200         th = kmalloc(sizeof *th, GFP_KERNEL);
201         if (th == NULL)
202                 return NULL;
203
204         BUG_ON(n_buckets & (n_buckets - 1));
205         th->buckets = kmem_zalloc(n_buckets * sizeof *th->buckets);
206         if (th->buckets == NULL) {
207                 printk("failed to allocate %u buckets\n", n_buckets);
208                 kfree(th);
209                 return NULL;
210         }
211         th->bucket_mask = n_buckets - 1;
212
213         swt = &th->swt;
214         swt->lookup = table_hash_lookup;
215         swt->insert = table_hash_insert;
216         swt->delete = table_hash_delete;
217         swt->timeout = table_hash_timeout;
218         swt->destroy = table_hash_destroy;
219         swt->iterate = table_hash_iterate;
220         swt->stats = table_hash_stats;
221
222         crc32_init(&th->crc32, polynomial);
223         th->n_flows = 0;
224
225         return swt;
226 }
227
228 /* Double-hashing table. */
229
230 struct sw_table_hash2 {
231         struct sw_table swt;
232         struct sw_table *subtable[2];
233 };
234
235 static struct sw_flow *table_hash2_lookup(struct sw_table *swt,
236                                                                                   const struct sw_flow_key *key)
237 {
238         struct sw_table_hash2 *t2 = (struct sw_table_hash2 *) swt;
239         int i;
240         
241         for (i = 0; i < 2; i++) {
242                 struct sw_flow *flow = *find_bucket(t2->subtable[i], key);
243                 if (flow && !memcmp(&flow->key, key, sizeof *key))
244                         return flow;
245         }
246         return NULL;
247 }
248
249 static int table_hash2_insert(struct sw_table *swt, struct sw_flow *flow)
250 {
251         struct sw_table_hash2 *t2 = (struct sw_table_hash2 *) swt;
252
253         if (table_hash_insert(t2->subtable[0], flow))
254                 return 1;
255         return table_hash_insert(t2->subtable[1], flow);
256 }
257
258 static int table_hash2_delete(struct sw_table *swt,
259                                                           const struct sw_flow_key *key, 
260                                                           uint16_t priority, int strict)
261 {
262         struct sw_table_hash2 *t2 = (struct sw_table_hash2 *) swt;
263         return (table_hash_delete(t2->subtable[0], key, priority, strict)
264                         + table_hash_delete(t2->subtable[1], key, priority, strict));
265 }
266
267 static int table_hash2_timeout(struct datapath *dp, struct sw_table *swt)
268 {
269         struct sw_table_hash2 *t2 = (struct sw_table_hash2 *) swt;
270         return (table_hash_timeout(dp, t2->subtable[0])
271                         + table_hash_timeout(dp, t2->subtable[1]));
272 }
273
274 static void table_hash2_destroy(struct sw_table *swt)
275 {
276         struct sw_table_hash2 *t2 = (struct sw_table_hash2 *) swt;
277         table_hash_destroy(t2->subtable[0]);
278         table_hash_destroy(t2->subtable[1]);
279         kfree(t2);
280 }
281
282 static int table_hash2_iterate(struct sw_table *swt,
283                                const struct sw_flow_key *key,
284                                struct sw_table_position *position,
285                                int (*callback)(struct sw_flow *, void *),
286                                void *private)
287 {
288         struct sw_table_hash2 *t2 = (struct sw_table_hash2 *) swt;
289         int i;
290
291         for (i = position->private[1]; i < 2; i++) {
292                 int error = table_hash_iterate(t2->subtable[i], key, position,
293                                                callback, private);
294                 if (error) {
295                         return error;
296                 }
297                 position->private[0] = 0;
298                 position->private[1]++;
299         }
300         return 0;
301 }
302
303 static void table_hash2_stats(struct sw_table *swt,
304                                  struct sw_table_stats *stats)
305 {
306         struct sw_table_hash2 *t2 = (struct sw_table_hash2 *) swt;
307         struct sw_table_stats substats[2];
308         int i;
309
310         for (i = 0; i < 2; i++)
311                 table_hash_stats(t2->subtable[i], &substats[i]);
312         stats->name = "hash2";
313         stats->n_flows = substats[0].n_flows + substats[1].n_flows;
314         stats->max_flows = substats[0].max_flows + substats[1].max_flows;
315 }
316
317 struct sw_table *table_hash2_create(unsigned int poly0, unsigned int buckets0,
318                                                                         unsigned int poly1, unsigned int buckets1)
319
320 {
321         struct sw_table_hash2 *t2;
322         struct sw_table *swt;
323
324         t2 = kmalloc(sizeof *t2, GFP_KERNEL);
325         if (t2 == NULL)
326                 return NULL;
327
328         t2->subtable[0] = table_hash_create(poly0, buckets0);
329         if (t2->subtable[0] == NULL)
330                 goto out_free_t2;
331
332         t2->subtable[1] = table_hash_create(poly1, buckets1);
333         if (t2->subtable[1] == NULL)
334                 goto out_free_subtable0;
335
336         swt = &t2->swt;
337         swt->lookup = table_hash2_lookup;
338         swt->insert = table_hash2_insert;
339         swt->delete = table_hash2_delete;
340         swt->timeout = table_hash2_timeout;
341         swt->destroy = table_hash2_destroy;
342         swt->iterate = table_hash2_iterate;
343         swt->stats = table_hash2_stats;
344
345         return swt;
346
347 out_free_subtable0:
348         table_hash_destroy(t2->subtable[0]);
349 out_free_t2:
350         kfree(t2);
351         return NULL;
352 }
353
354 /* From fs/xfs/linux-2.4/kmem.c. */
355
356 static void *
357 kmem_alloc(size_t size)
358 {
359         void *ptr;
360
361 #ifdef KMALLOC_MAX_SIZE
362         if (size > KMALLOC_MAX_SIZE)
363                 return NULL;
364 #endif
365         ptr = kmalloc(size, GFP_KERNEL);
366         if (!ptr) {
367                 ptr = vmalloc(size);
368                 if (ptr)
369                         printk("openflow: used vmalloc for %lu bytes\n", 
370                                         (unsigned long)size);
371         }
372         return ptr;
373 }
374
375 static void *
376 kmem_zalloc(size_t size)
377 {
378         void *ptr = kmem_alloc(size);
379         if (ptr)
380                 memset(ptr, 0, size);
381         return ptr;
382 }
383
384 static void
385 kmem_free(void *ptr, size_t size)
386 {
387         if (((unsigned long)ptr < VMALLOC_START) ||
388                 ((unsigned long)ptr >= VMALLOC_END)) {
389                 kfree(ptr);
390         } else {
391                 vfree(ptr);
392         }
393 }