xref: /openbmc/linux/kernel/bpf/hashtab.c (revision a06c488d)
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);
400