| 1 | |
| 2 | #define _LGPL_SOURCE |
| 3 | #include <stdlib.h> |
| 4 | #include <urcu.h> |
| 5 | #include <arch.h> |
| 6 | #include <arch_atomic.h> |
| 7 | #include <assert.h> |
| 8 | #include <compiler.h> |
| 9 | #include <urcu-defer.h> |
| 10 | #include <errno.h> |
| 11 | #include <urcu-ht.h> |
| 12 | |
| 13 | struct rcu_ht_node; |
| 14 | |
| 15 | struct rcu_ht_node { |
| 16 | struct rcu_ht_node *next; |
| 17 | void *key; |
| 18 | void *data; |
| 19 | }; |
| 20 | |
| 21 | struct rcu_ht { |
| 22 | struct rcu_ht_node *tbl[HASH_SIZE]; |
| 23 | ht_hash_fct hash_fct; |
| 24 | void (*free_fct)(void *data); /* fct to free data */ |
| 25 | }; |
| 26 | |
| 27 | struct rcu_ht *ht_new(ht_hash_fct hash_fct, void (*free_fct)(void *data)) |
| 28 | { |
| 29 | struct rcu_ht *ht; |
| 30 | |
| 31 | ht = calloc(1, sizeof(struct rcu_ht)); |
| 32 | ht->hash_fct = hash_fct; |
| 33 | ht->free_fct = free_fct; |
| 34 | return ht; |
| 35 | } |
| 36 | |
| 37 | /* delete all elements */ |
| 38 | void ht_delete_all(struct rcu_ht *ht) |
| 39 | { |
| 40 | unsigned long i; |
| 41 | struct rcu_ht_node **prev, *node; |
| 42 | |
| 43 | for (i = 0; i < HASH_SIZE; i++) { |
| 44 | rcu_read_lock(); |
| 45 | |
| 46 | prev = &ht->tbl[i]; |
| 47 | node = rcu_dereference(*prev); |
| 48 | /* |
| 49 | * Cut the head, therefore whole bucket will be unused |
| 50 | * after GP. (except for concurrent additions) |
| 51 | */ |
| 52 | rcu_assign_pointer(prev, NULL); |
| 53 | for (;;) { |
| 54 | if (likely(!node)) { |
| 55 | break; |
| 56 | } |
| 57 | prev = &node->next; |
| 58 | /* |
| 59 | * FIXME: calling call_rcu within a read-side C.S. is a |
| 60 | * bug, because it may call synchronize_rcu() if the |
| 61 | * callback queue is full. |
| 62 | */ |
| 63 | if (node->data) |
| 64 | call_rcu(ht->free_fct, node->data); |
| 65 | call_rcu(free, node); |
| 66 | node = rcu_dereference(*prev); |
| 67 | } |
| 68 | rcu_read_unlock(); |
| 69 | } |
| 70 | } |
| 71 | |
| 72 | /* |
| 73 | * Should only be called when no more concurrent readers nor writers can |
| 74 | * possibly access the table. |
| 75 | */ |
| 76 | void ht_destroy(struct rcu_ht *ht) |
| 77 | { |
| 78 | ht_delete_all(ht); |
| 79 | free(ht); |
| 80 | } |
| 81 | |
| 82 | void *ht_lookup(struct rcu_ht *ht, void *key) |
| 83 | { |
| 84 | unsigned long hash; |
| 85 | struct rcu_ht_node *node; |
| 86 | void *ret; |
| 87 | |
| 88 | hash = ht->hash_fct(key); |
| 89 | |
| 90 | rcu_read_lock(); |
| 91 | node = rcu_dereference(ht->tbl[hash]); |
| 92 | for (;;) { |
| 93 | if (likely(!node)) { |
| 94 | ret = NULL; |
| 95 | break; |
| 96 | } |
| 97 | if (node->key == key) { |
| 98 | ret = node->data; |
| 99 | break; |
| 100 | } |
| 101 | node = rcu_dereference(node->next); |
| 102 | } |
| 103 | rcu_read_unlock(); |
| 104 | |
| 105 | return ret; |
| 106 | } |
| 107 | |
| 108 | /* |
| 109 | * Will re-try until either: |
| 110 | * - The key is already there (-EEXIST) |
| 111 | * - We successfully add the key at the head of a table bucket. |
| 112 | */ |
| 113 | int ht_add(struct rcu_ht *ht, void *key, void *data) |
| 114 | { |
| 115 | struct rcu_ht_node *node, *old_head, *new_head; |
| 116 | unsigned long hash; |
| 117 | int ret = 0; |
| 118 | |
| 119 | new_head = calloc(1, sizeof(struct rcu_ht_node)); |
| 120 | new_head->key = key; |
| 121 | new_head->data = data; |
| 122 | hash = ht->hash_fct(key); |
| 123 | |
| 124 | /* here comes the fun and tricky part. |
| 125 | * Add at the beginning with a cmpxchg. |
| 126 | * Hold a read lock between the moment the first element is read |
| 127 | * and the nodes traversal (to find duplicates). This ensures |
| 128 | * the head pointer has not been reclaimed when cmpxchg is done. |
| 129 | * Always adding at the head ensures that we would have to |
| 130 | * re-try if a new item has been added concurrently. So we ensure that |
| 131 | * we never add duplicates. */ |
| 132 | retry: |
| 133 | rcu_read_lock(); |
| 134 | |
| 135 | old_head = node = rcu_dereference(ht->tbl[hash]); |
| 136 | for (;;) { |
| 137 | if (likely(!node)) { |
| 138 | break; |
| 139 | } |
| 140 | if (node->key == key) { |
| 141 | ret = -EEXIST; |
| 142 | goto end; |
| 143 | } |
| 144 | node = rcu_dereference(node->next); |
| 145 | } |
| 146 | if (rcu_cmpxchg_pointer(&ht->tbl[hash], old_head, new_head) != old_head) |
| 147 | goto restart; |
| 148 | end: |
| 149 | rcu_read_unlock(); |
| 150 | |
| 151 | return ret; |
| 152 | |
| 153 | /* restart loop, release and re-take the read lock to be kind to GP */ |
| 154 | restart: |
| 155 | rcu_read_unlock(); |
| 156 | goto retry; |
| 157 | } |
| 158 | |
| 159 | /* |
| 160 | * Restart until we successfully remove the entry, or no entry is left |
| 161 | * ((void *)(unsigned long)-ENOENT). |
| 162 | */ |
| 163 | void *ht_steal(struct rcu_ht *ht, void *key) |
| 164 | { |
| 165 | struct rcu_ht_node **prev, *node; |
| 166 | unsigned long hash; |
| 167 | void *data; |
| 168 | |
| 169 | hash = ht->hash_fct(key); |
| 170 | |
| 171 | retry: |
| 172 | rcu_read_lock(); |
| 173 | |
| 174 | prev = &ht->tbl[hash]; |
| 175 | node = rcu_dereference(*prev); |
| 176 | for (;;) { |
| 177 | if (likely(!node)) { |
| 178 | data = (void *)(unsigned long)-ENOENT; |
| 179 | goto end; |
| 180 | } |
| 181 | if (node->key == key) { |
| 182 | break; |
| 183 | } |
| 184 | prev = &node->next; |
| 185 | node = rcu_dereference(*prev); |
| 186 | } |
| 187 | /* Found it ! pointer to object is in "prev" */ |
| 188 | if (rcu_cmpxchg_pointer(prev, node, node->next) != node) |
| 189 | goto restart; |
| 190 | end: |
| 191 | rcu_read_unlock(); |
| 192 | |
| 193 | data = node->data; |
| 194 | call_rcu(free, node); |
| 195 | |
| 196 | return data; |
| 197 | |
| 198 | /* restart loop, release and re-take the read lock to be kind to GP */ |
| 199 | restart: |
| 200 | rcu_read_unlock(); |
| 201 | goto retry; |
| 202 | } |
| 203 | |
| 204 | int ht_delete(struct rcu_ht *ht, void *key) |
| 205 | { |
| 206 | void *data; |
| 207 | |
| 208 | data = ht_steal(ht, key); |
| 209 | if (data) { |
| 210 | if (ht->free_fct && data) |
| 211 | call_rcu(ht->free_fct, data); |
| 212 | return 0; |
| 213 | } else { |
| 214 | return -ENOENT; |
| 215 | } |
| 216 | } |
| 217 | |
| 218 | unsigned long stupid_hash(void *key) |
| 219 | { |
| 220 | return (unsigned long)key % HASH_SIZE; |
| 221 | } |