kernel/bpf/hashtab.c

   1 /* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com
   2  *
   3  * This program is free software; you can redistribute it and/or
   4  * modify it under the terms of version 2 of the GNU General Public
   5  * License as published by the Free Software Foundation.
   6  *
   7  * This program is distributed in the hope that it will be useful, but
   8  * WITHOUT ANY WARRANTY; without even the implied warranty of
   9  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  10  * General Public License for more details.
  11  */
  12 #include <linux/bpf.h>
  13 #include <linux/jhash.h>
  14 #include <linux/filter.h>
  15 #include <linux/vmalloc.h>
  16
  17 struct bucket {
  18         struct hlist_head head;
  19         raw_spinlock_t lock;
  20 };
  21
  22 struct bpf_htab {
  23         struct bpf_map map;
  24         struct bucket *buckets;
  25         atomic_t count; /* number of elements in this hashtable */
  26         u32 n_buckets;  /* number of hash buckets */
  27         u32 elem_size;  /* size of each element in bytes */
  28 };
  29
  30 /* each htab element is struct htab_elem + key + value */
  31 struct htab_elem {
  32         struct hlist_node hash_node;
  33         struct rcu_head rcu;
  34         u32 hash;
  35         char key[0] __aligned(8);
  36 };
  37
  38 /* Called from syscall */
  39 static struct bpf_map *htab_map_alloc(union bpf_attr *attr)
  40 {
  41         struct bpf_htab *htab;
  42         int err, i;
  43
  44         htab = kzalloc(sizeof(*htab), GFP_USER);
  45         if (!htab)
  46                 return ERR_PTR(-ENOMEM);
  47
  48         /* mandatory map attributes */
  49         htab->map.key_size = attr->key_size;
  50         htab->map.value_size = attr->value_size;
  51         htab->map.max_entries = attr->max_entries;
  52
  53         /* check sanity of attributes.
  54          * value_size == 0 may be allowed in the future to use map as a set
  55          */
  56         err = -EINVAL;
  57         if (htab->map.max_entries == 0 || htab->map.key_size == 0 ||
  58             htab->map.value_size == 0)
  59                 goto free_htab;
  60
  61         /* hash table size must be power of 2 */
  62         htab->n_buckets = roundup_pow_of_two(htab->map.max_entries);
  63
  64         err = -E2BIG;
  65         if (htab->map.key_size > MAX_BPF_STACK)
  66                 /* eBPF programs initialize keys on stack, so they cannot be
  67                  * larger than max stack size
  68                  */
  69                 goto free_htab;
  70
  71         if (htab->map.value_size >= (1 << (KMALLOC_SHIFT_MAX - 1)) -
  72             MAX_BPF_STACK - sizeof(struct htab_elem))
  73                 /* if value_size is bigger, the user space won't be able to
  74                  * access the elements via bpf syscall. This check also makes
  75                  * sure that the elem_size doesn't overflow and it's
  76                  * kmalloc-able later in htab_map_update_elem()
  77                  */
  78                 goto free_htab;
  79
  80         htab->elem_size = sizeof(struct htab_elem) +
  81                           round_up(htab->map.key_size, 8) +
  82                           htab->map.value_size;
  83
  84         /* prevent zero size kmalloc and check for u32 overflow */
  85         if (htab->n_buckets == 0 ||
  86             htab->n_buckets > U32_MAX / sizeof(struct bucket))
  87                 goto free_htab;
  88
  89         if ((u64) htab->n_buckets * sizeof(struct bucket) +
  90             (u64) htab->elem_size * htab->map.max_entries >=
  91             U32_MAX - PAGE_SIZE)
  92                 /* make sure page count doesn't overflow */
  93                 goto free_htab;
  94
  95         htab->map.pages = round_up(htab->n_buckets * sizeof(struct bucket) +
  96                                    htab->elem_size * htab->map.max_entries,
  97                                    PAGE_SIZE) >> PAGE_SHIFT;
  98
  99         err = -ENOMEM;
 100         htab->buckets = kmalloc_array(htab->n_buckets, sizeof(struct bucket),
 101                                       GFP_USER | __GFP_NOWARN);
 102
 103         if (!htab->buckets) {
 104                 htab->buckets = vmalloc(htab->n_buckets * sizeof(struct bucket));
 105                 if (!htab->buckets)
 106                         goto free_htab;
 107         }
 108
 109         for (i = 0; i < htab->n_buckets; i++) {
 110                 INIT_HLIST_HEAD(&htab->buckets[i].head);
 111                 raw_spin_lock_init(&htab->buckets[i].lock);
 112         }
 113
 114         atomic_set(&htab->count, 0);
 115
 116         return &htab->map;
 117
 118 free_htab:
 119         kfree(htab);
 120         return ERR_PTR(err);
 121 }
 122
 123 static inline u32 htab_map_hash(const void *key, u32 key_len)
 124 {
 125         return jhash(key, key_len, 0);
 126 }
 127
 128 static inline struct bucket *__select_bucket(struct bpf_htab *htab, u32 hash)
 129 {
 130         return &htab->buckets[hash & (htab->n_buckets - 1)];
 131 }
 132
 133 static inline struct hlist_head *select_bucket(struct bpf_htab *htab, u32 hash)
 134 {
 135         return &__select_bucket(htab, hash)->head;
 136 }
 137
 138 static struct htab_elem *lookup_elem_raw(struct hlist_head *head, u32 hash,
 139                                          void *key, u32 key_size)
 140 {
 141         struct htab_elem *l;
 142
 143         hlist_for_each_entry_rcu(l, head, hash_node)
 144                 if (l->hash == hash && !memcmp(&l->key, key, key_size))
 145                         return l;
 146
 147         return NULL;
 148 }
 149
 150 /* Called from syscall or from eBPF program */
 151 static void *htab_map_lookup_elem(struct bpf_map *map, void *key)
 152 {
 153         struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
 154         struct hlist_head *head;
 155         struct htab_elem *l;
 156         u32 hash, key_size;
 157
 158         /* Must be called with rcu_read_lock. */
 159         WARN_ON_ONCE(!rcu_read_lock_held());
 160
 161         key_size = map->key_size;
 162
 163         hash = htab_map_hash(key, key_size);
 164
 165         head = select_bucket(htab, hash);
 166
 167         l = lookup_elem_raw(head, hash, key, key_size);
 168
 169         if (l)
 170                 return l->key + round_up(map->key_size, 8);
 171
 172         return NULL;
 173 }
 174
 175 /* Called from syscall */
 176 static int htab_map_get_next_key(struct bpf_map *map, void *key, void *next_key)
 177 {
 178         struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
 179         struct hlist_head *head;
 180         struct htab_elem *l, *next_l;
 181         u32 hash, key_size;
 182         int i;
 183
 184         WARN_ON_ONCE(!rcu_read_lock_held());
 185
 186         key_size = map->key_size;
 187
 188         hash = htab_map_hash(key, key_size);
 189
 190         head = select_bucket(htab, hash);
 191
 192         /* lookup the key */
 193         l = lookup_elem_raw(head, hash, key, key_size);
 194
 195         if (!l) {
 196                 i = 0;
 197                 goto find_first_elem;
 198         }
 199
 200         /* key was found, get next key in the same bucket */
 201         next_l = hlist_entry_safe(rcu_dereference_raw(hlist_next_rcu(&l->hash_node)),
 202                                   struct htab_elem, hash_node);
 203
 204         if (next_l) {
 205                 /* if next elem in this hash list is non-zero, just return it */
 206                 memcpy(next_key, next_l->key, key_size);
 207                 return 0;
 208         }
 209
 210         /* no more elements in this hash list, go to the next bucket */
 211         i = hash & (htab->n_buckets - 1);
 212         i++;
 213
 214 find_first_elem:
 215         /* iterate over buckets */
 216         for (; i < htab->n_buckets; i++) {
 217                 head = select_bucket(htab, i);
 218
 219                 /* pick first element in the bucket */
 220                 next_l = hlist_entry_safe(rcu_dereference_raw(hlist_first_rcu(head)),
 221                                           struct htab_elem, hash_node);
 222                 if (next_l) {
 223                         /* if it's not empty, just return it */
 224                         memcpy(next_key, next_l->key, key_size);
 225                         return 0;
 226                 }
 227         }
 228
 229         /* itereated over all buckets and all elements */
 230         return -ENOENT;
 231 }
 232
 233 /* Called from syscall or from eBPF program */
 234 static int htab_map_update_elem(struct bpf_map *map, void *key, void *value,
 235                                 u64 map_flags)
 236 {
 237         struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
 238         struct htab_elem *l_new, *l_old;
 239         struct hlist_head *head;
 240         struct bucket *b;
 241         unsigned long flags;
 242         u32 key_size;
 243         int ret;
 244
 245         if (map_flags > BPF_EXIST)
 246                 /* unknown flags */
 247                 return -EINVAL;
 248
 249         WARN_ON_ONCE(!rcu_read_lock_held());
 250
 251         /* allocate new element outside of lock */
 252         l_new = kmalloc(htab->elem_size, GFP_ATOMIC | __GFP_NOWARN);
 253         if (!l_new)
 254                 return -ENOMEM;
 255
 256         key_size = map->key_size;
 257
 258         memcpy(l_new->key, key, key_size);
 259         memcpy(l_new->key + round_up(key_size, 8), value, map->value_size);
 260
 261         l_new->hash = htab_map_hash(l_new->key, key_size);
 262         b = __select_bucket(htab, l_new->hash);
 263         head = &b->head;
 264
 265         /* bpf_map_update_elem() can be called in_irq() */
 266         raw_spin_lock_irqsave(&b->lock, flags);
 267
 268         l_old = lookup_elem_raw(head, l_new->hash, key, key_size);
 269
 270         if (!l_old && unlikely(atomic_read(&htab->count) >= map->max_entries)) {
 271                 /* if elem with this 'key' doesn't exist and we've reached
 272                  * max_entries limit, fail insertion of new elem
 273                  */
 274                 ret = -E2BIG;
 275                 goto err;
 276         }
 277
 278         if (l_old && map_flags == BPF_NOEXIST) {
 279                 /* elem already exists */
 280                 ret = -EEXIST;
 281                 goto err;
 282         }
 283
 284         if (!l_old && map_flags == BPF_EXIST) {
 285                 /* elem doesn't exist, cannot update it */
 286                 ret = -ENOENT;
 287                 goto err;
 288         }
 289
 290         /* add new element to the head of the list, so that concurrent
 291          * search will find it before old elem
 292          */
 293         hlist_add_head_rcu(&l_new->hash_node, head);
 294         if (l_old) {
 295                 hlist_del_rcu(&l_old->hash_node);
 296                 kfree_rcu(l_old, rcu);
 297         } else {
 298                 atomic_inc(&htab->count);
 299         }
 300         raw_spin_unlock_irqrestore(&b->lock, flags);
 301
 302         return 0;
 303 err:
 304         raw_spin_unlock_irqrestore(&b->lock, flags);
 305         kfree(l_new);
 306         return ret;
 307 }
 308
 309 /* Called from syscall or from eBPF program */
 310 static int htab_map_delete_elem(struct bpf_map *map, void *key)
 311 {
 312         struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
 313         struct hlist_head *head;
 314         struct bucket *b;
 315         struct htab_elem *l;
 316         unsigned long flags;
 317         u32 hash, key_size;
 318         int ret = -ENOENT;
 319
 320         WARN_ON_ONCE(!rcu_read_lock_held());
 321
 322         key_size = map->key_size;
 323
 324         hash = htab_map_hash(key, key_size);
 325         b = __select_bucket(htab, hash);
 326         head = &b->head;
 327
 328         raw_spin_lock_irqsave(&b->lock, flags);
 329
 330         l = lookup_elem_raw(head, hash, key, key_size);
 331
 332         if (l) {
 333                 hlist_del_rcu(&l->hash_node);
 334                 atomic_dec(&htab->count);
 335                 kfree_rcu(l, rcu);
 336                 ret = 0;
 337         }
 338
 339         raw_spin_unlock_irqrestore(&b->lock, flags);
 340         return ret;
 341 }
 342
 343 static void delete_all_elements(struct bpf_htab *htab)
 344 {
 345         int i;
 346
 347         for (i = 0; i < htab->n_buckets; i++) {
 348                 struct hlist_head *head = select_bucket(htab, i);
 349                 struct hlist_node *n;
 350                 struct htab_elem *l;
 351
 352                 hlist_for_each_entry_safe(l, n, head, hash_node) {
 353                         hlist_del_rcu(&l->hash_node);
 354                         atomic_dec(&htab->count);
 355                         kfree(l);
 356                 }
 357         }
 358 }
 359
 360 /* Called when map->refcnt goes to zero, either from workqueue or from syscall */
 361 static void htab_map_free(struct bpf_map *map)
 362 {
 363         struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
 364
 365         /* at this point bpf_prog->aux->refcnt == 0 and this map->refcnt == 0,
 366          * so the programs (can be more than one that used this map) were
 367          * disconnected from events. Wait for outstanding critical sections in
 368          * these programs to complete
 369          */
 370         synchronize_rcu();
 371
 372         /* some of kfree_rcu() callbacks for elements of this map may not have
 373          * executed. It's ok. Proceed to free residual elements and map itself
 374          */
 375         delete_all_elements(htab);
 376         kvfree(htab->buckets);
 377         kfree(htab);
 378 }
 379
 380 static const struct bpf_map_ops htab_ops = {
 381         .map_alloc = htab_map_alloc,
 382         .map_free = htab_map_free,
 383         .map_get_next_key = htab_map_get_next_key,
 384         .map_lookup_elem = htab_map_lookup_elem,
 385         .map_update_elem = htab_map_update_elem,
 386         .map_delete_elem = htab_map_delete_elem,
 387 };
 388
 389 static struct bpf_map_type_list htab_type __read_mostly = {
 390         .ops = &htab_ops,
 391         .type = BPF_MAP_TYPE_HASH,
 392 };
 393
 394 static int __init register_htab_map(void)
 395 {
 396         bpf_register_map_type(&htab_type);
 397         return 0;
 398 }
 399 late_initcall(register_htab_map);