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