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,
  35                                 offsetof(struct sw_flow_key, wildcards));
  36         return &th->buckets[crc & th->bucket_mask];
  37 }
  38
  39 static struct sw_flow *table_hash_lookup(struct sw_table *swt,
  40                                                                                  const struct sw_flow_key *key)
  41 {
  42         struct sw_flow *flow = *find_bucket(swt, key);
  43         return flow && flow_keys_equal(&flow->key, key) ? flow : NULL;
  44 }
  45
  46 static int table_hash_insert(struct sw_table *swt, struct sw_flow *flow)
  47 {
  48         struct sw_table_hash *th = (struct sw_table_hash *) swt;
  49         struct sw_flow **bucket;
  50         int retval;
  51
  52         if (flow->key.wildcards != 0)
  53                 return 0;
  54
  55         bucket = find_bucket(swt, &flow->key);
  56         if (*bucket == NULL) {
  57                 th->n_flows++;
  58                 rcu_assign_pointer(*bucket, flow);
  59                 retval = 1;
  60         } else {
  61                 struct sw_flow *old_flow = *bucket;
  62                 if (flow_keys_equal(&old_flow->key, &flow->key)) {
  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         rcu_assign_pointer(*bucket, NULL);
  77         flow_deferred_free(flow);
  78         return 1;
  79 }
  80
  81 /* Returns number of deleted flows.  We can ignore the priority
  82  * argument, since all exact-match entries are the same (highest)
  83  * priority. */
  84 static int table_hash_delete(struct sw_table *swt,
  85                                                          const struct sw_flow_key *key,
  86                                                          uint16_t priority, int strict)
  87 {
  88         struct sw_table_hash *th = (struct sw_table_hash *) swt;
  89         unsigned int count = 0;
  90
  91         if (key->wildcards == 0) {
  92                 struct sw_flow **bucket = find_bucket(swt, key);
  93                 struct sw_flow *flow = *bucket;
  94                 if (flow && flow_keys_equal(&flow->key, key))
  95                         count = do_delete(bucket, flow);
  96         } else {
  97                 unsigned int i;
  98
  99                 for (i = 0; i <= th->bucket_mask; i++) {
 100                         struct sw_flow **bucket = &th->buckets[i];
 101                         struct sw_flow *flow = *bucket;
 102                         if (flow && flow_del_matches(&flow->key, key, strict))
 103                                 count += do_delete(bucket, flow);
 104                 }
 105         }
 106         th->n_flows -= count;
 107         return count;
 108 }
 109
 110 static int table_hash_timeout(struct datapath *dp, struct sw_table *swt)
 111 {
 112         struct sw_table_hash *th = (struct sw_table_hash *) swt;
 113         unsigned int i;
 114         int count = 0;
 115
 116         mutex_lock(&dp_mutex);
 117         for (i = 0; i <= th->bucket_mask; i++) {
 118                 struct sw_flow **bucket = &th->buckets[i];
 119                 struct sw_flow *flow = *bucket;
 120                 if (flow) {
 121                         int reason = flow_timeout(flow);
 122                         if (reason >= 0) {
 123                                 count += do_delete(bucket, flow);
 124                                 dp_send_flow_expired(dp, flow, reason);
 125                         }
 126                 }
 127         }
 128         th->n_flows -= count;
 129         mutex_unlock(&dp_mutex);
 130
 131         return count;
 132 }
 133
 134 static void table_hash_destroy(struct sw_table *swt)
 135 {
 136         struct sw_table_hash *th = (struct sw_table_hash *) swt;
 137         unsigned int i;
 138         for (i = 0; i <= th->bucket_mask; i++)
 139         if (th->buckets[i])
 140                 flow_free(th->buckets[i]);
 141         kmem_free(th->buckets, (th->bucket_mask + 1) * sizeof *th->buckets);
 142         kfree(th);
 143 }
 144
 145 static int table_hash_iterate(struct sw_table *swt,
 146                               const struct sw_flow_key *key,
 147                               struct sw_table_position *position,
 148                               int (*callback)(struct sw_flow *, void *private),
 149                               void *private)
 150 {
 151         struct sw_table_hash *th = (struct sw_table_hash *) swt;
 152
 153         if (position->private[0] > th->bucket_mask)
 154                 return 0;
 155
 156         if (key->wildcards == 0) {
 157                 struct sw_flow *flow;
 158                 int error;
 159
 160                 flow = table_hash_lookup(swt, key);
 161                 if (!flow)
 162                         return 0;
 163
 164                 error = callback(flow, private);
 165                 if (!error)
 166                         position->private[0] = -1;
 167                 return error;
 168         } else {
 169                 int i;
 170
 171                 for (i = position->private[0]; i <= th->bucket_mask; i++) {
 172                         struct sw_flow *flow = th->buckets[i];
 173                         if (flow && flow_matches_1wild(&flow->key, key)) {
 174                                 int error = callback(flow, private);
 175                                 if (error) {
 176                                         position->private[0] = i;
 177                                         return error;
 178                                 }
 179                         }
 180                 }
 181                 return 0;
 182         }
 183 }
 184 static void table_hash_stats(struct sw_table *swt,
 185                                  struct sw_table_stats *stats)
 186 {
 187         struct sw_table_hash *th = (struct sw_table_hash *) swt;
 188         stats->name = "hash";
 189         stats->n_flows = th->n_flows;
 190         stats->max_flows = th->bucket_mask + 1;
 191         stats->n_matched = swt->n_matched;
 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 = kzalloc(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 && flow_keys_equal(&flow->key, 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         stats->n_matched = swt->n_matched;
 316 }
 317
 318 struct sw_table *table_hash2_create(unsigned int poly0, unsigned int buckets0,
 319                                                                         unsigned int poly1, unsigned int buckets1)
 320
 321 {
 322         struct sw_table_hash2 *t2;
 323         struct sw_table *swt;
 324
 325         t2 = kzalloc(sizeof *t2, GFP_KERNEL);
 326         if (t2 == NULL)
 327                 return NULL;
 328
 329         t2->subtable[0] = table_hash_create(poly0, buckets0);
 330         if (t2->subtable[0] == NULL)
 331                 goto out_free_t2;
 332
 333         t2->subtable[1] = table_hash_create(poly1, buckets1);
 334         if (t2->subtable[1] == NULL)
 335                 goto out_free_subtable0;
 336
 337         swt = &t2->swt;
 338         swt->lookup = table_hash2_lookup;
 339         swt->insert = table_hash2_insert;
 340         swt->delete = table_hash2_delete;
 341         swt->timeout = table_hash2_timeout;
 342         swt->destroy = table_hash2_destroy;
 343         swt->iterate = table_hash2_iterate;
 344         swt->stats = table_hash2_stats;
 345
 346         return swt;
 347
 348 out_free_subtable0:
 349         table_hash_destroy(t2->subtable[0]);
 350 out_free_t2:
 351         kfree(t2);
 352         return NULL;
 353 }
 354
 355 /* From fs/xfs/linux-2.4/kmem.c. */
 356
 357 static void *
 358 kmem_alloc(size_t size)
 359 {
 360         void *ptr;
 361
 362 #ifdef KMALLOC_MAX_SIZE
 363         if (size > KMALLOC_MAX_SIZE)
 364                 return NULL;
 365 #endif
 366         ptr = kmalloc(size, GFP_KERNEL);
 367         if (!ptr) {
 368                 ptr = vmalloc(size);
 369                 if (ptr)
 370                         printk("openflow: used vmalloc for %lu bytes\n",
 371                                         (unsigned long)size);
 372         }
 373         return ptr;
 374 }
 375
 376 static void *
 377 kmem_zalloc(size_t size)
 378 {
 379         void *ptr = kmem_alloc(size);
 380         if (ptr)
 381                 memset(ptr, 0, size);
 382         return ptr;
 383 }
 384
 385 static void
 386 kmem_free(void *ptr, size_t size)
 387 {
 388         if (((unsigned long)ptr < VMALLOC_START) ||
 389                 ((unsigned long)ptr >= VMALLOC_END)) {
 390                 kfree(ptr);
 391         } else {
 392                 vfree(ptr);
 393         }
 394 }