xref: /openbmc/linux/net/core/sock_map.c (revision f9834f18)
1 // SPDX-License-Identifier: GPL-2.0
2 /* Copyright (c) 2017 - 2018 Covalent IO, Inc. http://covalent.io */
3 
4 #include <linux/bpf.h>
5 #include <linux/filter.h>
6 #include <linux/errno.h>
7 #include <linux/file.h>
8 #include <linux/net.h>
9 #include <linux/workqueue.h>
10 #include <linux/skmsg.h>
11 #include <linux/list.h>
12 #include <linux/jhash.h>
13 
14 struct bpf_stab {
15 	struct bpf_map map;
16 	struct sock **sks;
17 	struct sk_psock_progs progs;
18 	raw_spinlock_t lock;
19 };
20 
21 #define SOCK_CREATE_FLAG_MASK				\
22 	(BPF_F_NUMA_NODE | BPF_F_RDONLY | BPF_F_WRONLY)
23 
24 static struct bpf_map *sock_map_alloc(union bpf_attr *attr)
25 {
26 	struct bpf_stab *stab;
27 	u64 cost;
28 	int err;
29 
30 	if (!capable(CAP_NET_ADMIN))
31 		return ERR_PTR(-EPERM);
32 	if (attr->max_entries == 0 ||
33 	    attr->key_size    != 4 ||
34 	    attr->value_size  != 4 ||
35 	    attr->map_flags & ~SOCK_CREATE_FLAG_MASK)
36 		return ERR_PTR(-EINVAL);
37 
38 	stab = kzalloc(sizeof(*stab), GFP_USER);
39 	if (!stab)
40 		return ERR_PTR(-ENOMEM);
41 
42 	bpf_map_init_from_attr(&stab->map, attr);
43 	raw_spin_lock_init(&stab->lock);
44 
45 	/* Make sure page count doesn't overflow. */
46 	cost = (u64) stab->map.max_entries * sizeof(struct sock *);
47 	err = bpf_map_charge_init(&stab->map.memory, cost);
48 	if (err)
49 		goto free_stab;
50 
51 	stab->sks = bpf_map_area_alloc(stab->map.max_entries *
52 				       sizeof(struct sock *),
53 				       stab->map.numa_node);
54 	if (stab->sks)
55 		return &stab->map;
56 	err = -ENOMEM;
57 	bpf_map_charge_finish(&stab->map.memory);
58 free_stab:
59 	kfree(stab);
60 	return ERR_PTR(err);
61 }
62 
63 int sock_map_get_from_fd(const union bpf_attr *attr, struct bpf_prog *prog)
64 {
65 	u32 ufd = attr->target_fd;
66 	struct bpf_map *map;
67 	struct fd f;
68 	int ret;
69 
70 	f = fdget(ufd);
71 	map = __bpf_map_get(f);
72 	if (IS_ERR(map))
73 		return PTR_ERR(map);
74 	ret = sock_map_prog_update(map, prog, attr->attach_type);
75 	fdput(f);
76 	return ret;
77 }
78 
79 static void sock_map_sk_acquire(struct sock *sk)
80 	__acquires(&sk->sk_lock.slock)
81 {
82 	lock_sock(sk);
83 	preempt_disable();
84 	rcu_read_lock();
85 }
86 
87 static void sock_map_sk_release(struct sock *sk)
88 	__releases(&sk->sk_lock.slock)
89 {
90 	rcu_read_unlock();
91 	preempt_enable();
92 	release_sock(sk);
93 }
94 
95 static void sock_map_add_link(struct sk_psock *psock,
96 			      struct sk_psock_link *link,
97 			      struct bpf_map *map, void *link_raw)
98 {
99 	link->link_raw = link_raw;
100 	link->map = map;
101 	spin_lock_bh(&psock->link_lock);
102 	list_add_tail(&link->list, &psock->link);
103 	spin_unlock_bh(&psock->link_lock);
104 }
105 
106 static void sock_map_del_link(struct sock *sk,
107 			      struct sk_psock *psock, void *link_raw)
108 {
109 	struct sk_psock_link *link, *tmp;
110 	bool strp_stop = false;
111 
112 	spin_lock_bh(&psock->link_lock);
113 	list_for_each_entry_safe(link, tmp, &psock->link, list) {
114 		if (link->link_raw == link_raw) {
115 			struct bpf_map *map = link->map;
116 			struct bpf_stab *stab = container_of(map, struct bpf_stab,
117 							     map);
118 			if (psock->parser.enabled && stab->progs.skb_parser)
119 				strp_stop = true;
120 			list_del(&link->list);
121 			sk_psock_free_link(link);
122 		}
123 	}
124 	spin_unlock_bh(&psock->link_lock);
125 	if (strp_stop) {
126 		write_lock_bh(&sk->sk_callback_lock);
127 		sk_psock_stop_strp(sk, psock);
128 		write_unlock_bh(&sk->sk_callback_lock);
129 	}
130 }
131 
132 static void sock_map_unref(struct sock *sk, void *link_raw)
133 {
134 	struct sk_psock *psock = sk_psock(sk);
135 
136 	if (likely(psock)) {
137 		sock_map_del_link(sk, psock, link_raw);
138 		sk_psock_put(sk, psock);
139 	}
140 }
141 
142 static int sock_map_link(struct bpf_map *map, struct sk_psock_progs *progs,
143 			 struct sock *sk)
144 {
145 	struct bpf_prog *msg_parser, *skb_parser, *skb_verdict;
146 	bool skb_progs, sk_psock_is_new = false;
147 	struct sk_psock *psock;
148 	int ret;
149 
150 	skb_verdict = READ_ONCE(progs->skb_verdict);
151 	skb_parser = READ_ONCE(progs->skb_parser);
152 	skb_progs = skb_parser && skb_verdict;
153 	if (skb_progs) {
154 		skb_verdict = bpf_prog_inc_not_zero(skb_verdict);
155 		if (IS_ERR(skb_verdict))
156 			return PTR_ERR(skb_verdict);
157 		skb_parser = bpf_prog_inc_not_zero(skb_parser);
158 		if (IS_ERR(skb_parser)) {
159 			bpf_prog_put(skb_verdict);
160 			return PTR_ERR(skb_parser);
161 		}
162 	}
163 
164 	msg_parser = READ_ONCE(progs->msg_parser);
165 	if (msg_parser) {
166 		msg_parser = bpf_prog_inc_not_zero(msg_parser);
167 		if (IS_ERR(msg_parser)) {
168 			ret = PTR_ERR(msg_parser);
169 			goto out;
170 		}
171 	}
172 
173 	psock = sk_psock_get_checked(sk);
174 	if (IS_ERR(psock)) {
175 		ret = PTR_ERR(psock);
176 		goto out_progs;
177 	}
178 
179 	if (psock) {
180 		if ((msg_parser && READ_ONCE(psock->progs.msg_parser)) ||
181 		    (skb_progs  && READ_ONCE(psock->progs.skb_parser))) {
182 			sk_psock_put(sk, psock);
183 			ret = -EBUSY;
184 			goto out_progs;
185 		}
186 	} else {
187 		psock = sk_psock_init(sk, map->numa_node);
188 		if (!psock) {
189 			ret = -ENOMEM;
190 			goto out_progs;
191 		}
192 		sk_psock_is_new = true;
193 	}
194 
195 	if (msg_parser)
196 		psock_set_prog(&psock->progs.msg_parser, msg_parser);
197 	if (sk_psock_is_new) {
198 		ret = tcp_bpf_init(sk);
199 		if (ret < 0)
200 			goto out_drop;
201 	} else {
202 		tcp_bpf_reinit(sk);
203 	}
204 
205 	write_lock_bh(&sk->sk_callback_lock);
206 	if (skb_progs && !psock->parser.enabled) {
207 		ret = sk_psock_init_strp(sk, psock);
208 		if (ret) {
209 			write_unlock_bh(&sk->sk_callback_lock);
210 			goto out_drop;
211 		}
212 		psock_set_prog(&psock->progs.skb_verdict, skb_verdict);
213 		psock_set_prog(&psock->progs.skb_parser, skb_parser);
214 		sk_psock_start_strp(sk, psock);
215 	}
216 	write_unlock_bh(&sk->sk_callback_lock);
217 	return 0;
218 out_drop:
219 	sk_psock_put(sk, psock);
220 out_progs:
221 	if (msg_parser)
222 		bpf_prog_put(msg_parser);
223 out:
224 	if (skb_progs) {
225 		bpf_prog_put(skb_verdict);
226 		bpf_prog_put(skb_parser);
227 	}
228 	return ret;
229 }
230 
231 static void sock_map_free(struct bpf_map *map)
232 {
233 	struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
234 	int i;
235 
236 	synchronize_rcu();
237 	raw_spin_lock_bh(&stab->lock);
238 	for (i = 0; i < stab->map.max_entries; i++) {
239 		struct sock **psk = &stab->sks[i];
240 		struct sock *sk;
241 
242 		sk = xchg(psk, NULL);
243 		if (sk) {
244 			lock_sock(sk);
245 			rcu_read_lock();
246 			sock_map_unref(sk, psk);
247 			rcu_read_unlock();
248 			release_sock(sk);
249 		}
250 	}
251 	raw_spin_unlock_bh(&stab->lock);
252 
253 	/* wait for psock readers accessing its map link */
254 	synchronize_rcu();
255 
256 	bpf_map_area_free(stab->sks);
257 	kfree(stab);
258 }
259 
260 static void sock_map_release_progs(struct bpf_map *map)
261 {
262 	psock_progs_drop(&container_of(map, struct bpf_stab, map)->progs);
263 }
264 
265 static struct sock *__sock_map_lookup_elem(struct bpf_map *map, u32 key)
266 {
267 	struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
268 
269 	WARN_ON_ONCE(!rcu_read_lock_held());
270 
271 	if (unlikely(key >= map->max_entries))
272 		return NULL;
273 	return READ_ONCE(stab->sks[key]);
274 }
275 
276 static void *sock_map_lookup(struct bpf_map *map, void *key)
277 {
278 	return ERR_PTR(-EOPNOTSUPP);
279 }
280 
281 static int __sock_map_delete(struct bpf_stab *stab, struct sock *sk_test,
282 			     struct sock **psk)
283 {
284 	struct sock *sk;
285 	int err = 0;
286 
287 	raw_spin_lock_bh(&stab->lock);
288 	sk = *psk;
289 	if (!sk_test || sk_test == sk)
290 		sk = xchg(psk, NULL);
291 
292 	if (likely(sk))
293 		sock_map_unref(sk, psk);
294 	else
295 		err = -EINVAL;
296 
297 	raw_spin_unlock_bh(&stab->lock);
298 	return err;
299 }
300 
301 static void sock_map_delete_from_link(struct bpf_map *map, struct sock *sk,
302 				      void *link_raw)
303 {
304 	struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
305 
306 	__sock_map_delete(stab, sk, link_raw);
307 }
308 
309 static int sock_map_delete_elem(struct bpf_map *map, void *key)
310 {
311 	struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
312 	u32 i = *(u32 *)key;
313 	struct sock **psk;
314 
315 	if (unlikely(i >= map->max_entries))
316 		return -EINVAL;
317 
318 	psk = &stab->sks[i];
319 	return __sock_map_delete(stab, NULL, psk);
320 }
321 
322 static int sock_map_get_next_key(struct bpf_map *map, void *key, void *next)
323 {
324 	struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
325 	u32 i = key ? *(u32 *)key : U32_MAX;
326 	u32 *key_next = next;
327 
328 	if (i == stab->map.max_entries - 1)
329 		return -ENOENT;
330 	if (i >= stab->map.max_entries)
331 		*key_next = 0;
332 	else
333 		*key_next = i + 1;
334 	return 0;
335 }
336 
337 static int sock_map_update_common(struct bpf_map *map, u32 idx,
338 				  struct sock *sk, u64 flags)
339 {
340 	struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
341 	struct inet_connection_sock *icsk = inet_csk(sk);
342 	struct sk_psock_link *link;
343 	struct sk_psock *psock;
344 	struct sock *osk;
345 	int ret;
346 
347 	WARN_ON_ONCE(!rcu_read_lock_held());
348 	if (unlikely(flags > BPF_EXIST))
349 		return -EINVAL;
350 	if (unlikely(idx >= map->max_entries))
351 		return -E2BIG;
352 	if (unlikely(rcu_access_pointer(icsk->icsk_ulp_data)))
353 		return -EINVAL;
354 
355 	link = sk_psock_init_link();
356 	if (!link)
357 		return -ENOMEM;
358 
359 	ret = sock_map_link(map, &stab->progs, sk);
360 	if (ret < 0)
361 		goto out_free;
362 
363 	psock = sk_psock(sk);
364 	WARN_ON_ONCE(!psock);
365 
366 	raw_spin_lock_bh(&stab->lock);
367 	osk = stab->sks[idx];
368 	if (osk && flags == BPF_NOEXIST) {
369 		ret = -EEXIST;
370 		goto out_unlock;
371 	} else if (!osk && flags == BPF_EXIST) {
372 		ret = -ENOENT;
373 		goto out_unlock;
374 	}
375 
376 	sock_map_add_link(psock, link, map, &stab->sks[idx]);
377 	stab->sks[idx] = sk;
378 	if (osk)
379 		sock_map_unref(osk, &stab->sks[idx]);
380 	raw_spin_unlock_bh(&stab->lock);
381 	return 0;
382 out_unlock:
383 	raw_spin_unlock_bh(&stab->lock);
384 	if (psock)
385 		sk_psock_put(sk, psock);
386 out_free:
387 	sk_psock_free_link(link);
388 	return ret;
389 }
390 
391 static bool sock_map_op_okay(const struct bpf_sock_ops_kern *ops)
392 {
393 	return ops->op == BPF_SOCK_OPS_PASSIVE_ESTABLISHED_CB ||
394 	       ops->op == BPF_SOCK_OPS_ACTIVE_ESTABLISHED_CB;
395 }
396 
397 static bool sock_map_sk_is_suitable(const struct sock *sk)
398 {
399 	return sk->sk_type == SOCK_STREAM &&
400 	       sk->sk_protocol == IPPROTO_TCP;
401 }
402 
403 static int sock_map_update_elem(struct bpf_map *map, void *key,
404 				void *value, u64 flags)
405 {
406 	u32 ufd = *(u32 *)value;
407 	u32 idx = *(u32 *)key;
408 	struct socket *sock;
409 	struct sock *sk;
410 	int ret;
411 
412 	sock = sockfd_lookup(ufd, &ret);
413 	if (!sock)
414 		return ret;
415 	sk = sock->sk;
416 	if (!sk) {
417 		ret = -EINVAL;
418 		goto out;
419 	}
420 	if (!sock_map_sk_is_suitable(sk)) {
421 		ret = -EOPNOTSUPP;
422 		goto out;
423 	}
424 
425 	sock_map_sk_acquire(sk);
426 	if (sk->sk_state != TCP_ESTABLISHED)
427 		ret = -EOPNOTSUPP;
428 	else
429 		ret = sock_map_update_common(map, idx, sk, flags);
430 	sock_map_sk_release(sk);
431 out:
432 	fput(sock->file);
433 	return ret;
434 }
435 
436 BPF_CALL_4(bpf_sock_map_update, struct bpf_sock_ops_kern *, sops,
437 	   struct bpf_map *, map, void *, key, u64, flags)
438 {
439 	WARN_ON_ONCE(!rcu_read_lock_held());
440 
441 	if (likely(sock_map_sk_is_suitable(sops->sk) &&
442 		   sock_map_op_okay(sops)))
443 		return sock_map_update_common(map, *(u32 *)key, sops->sk,
444 					      flags);
445 	return -EOPNOTSUPP;
446 }
447 
448 const struct bpf_func_proto bpf_sock_map_update_proto = {
449 	.func		= bpf_sock_map_update,
450 	.gpl_only	= false,
451 	.pkt_access	= true,
452 	.ret_type	= RET_INTEGER,
453 	.arg1_type	= ARG_PTR_TO_CTX,
454 	.arg2_type	= ARG_CONST_MAP_PTR,
455 	.arg3_type	= ARG_PTR_TO_MAP_KEY,
456 	.arg4_type	= ARG_ANYTHING,
457 };
458 
459 BPF_CALL_4(bpf_sk_redirect_map, struct sk_buff *, skb,
460 	   struct bpf_map *, map, u32, key, u64, flags)
461 {
462 	struct tcp_skb_cb *tcb = TCP_SKB_CB(skb);
463 
464 	if (unlikely(flags & ~(BPF_F_INGRESS)))
465 		return SK_DROP;
466 	tcb->bpf.flags = flags;
467 	tcb->bpf.sk_redir = __sock_map_lookup_elem(map, key);
468 	if (!tcb->bpf.sk_redir)
469 		return SK_DROP;
470 	return SK_PASS;
471 }
472 
473 const struct bpf_func_proto bpf_sk_redirect_map_proto = {
474 	.func           = bpf_sk_redirect_map,
475 	.gpl_only       = false,
476 	.ret_type       = RET_INTEGER,
477 	.arg1_type	= ARG_PTR_TO_CTX,
478 	.arg2_type      = ARG_CONST_MAP_PTR,
479 	.arg3_type      = ARG_ANYTHING,
480 	.arg4_type      = ARG_ANYTHING,
481 };
482 
483 BPF_CALL_4(bpf_msg_redirect_map, struct sk_msg *, msg,
484 	   struct bpf_map *, map, u32, key, u64, flags)
485 {
486 	if (unlikely(flags & ~(BPF_F_INGRESS)))
487 		return SK_DROP;
488 	msg->flags = flags;
489 	msg->sk_redir = __sock_map_lookup_elem(map, key);
490 	if (!msg->sk_redir)
491 		return SK_DROP;
492 	return SK_PASS;
493 }
494 
495 const struct bpf_func_proto bpf_msg_redirect_map_proto = {
496 	.func           = bpf_msg_redirect_map,
497 	.gpl_only       = false,
498 	.ret_type       = RET_INTEGER,
499 	.arg1_type	= ARG_PTR_TO_CTX,
500 	.arg2_type      = ARG_CONST_MAP_PTR,
501 	.arg3_type      = ARG_ANYTHING,
502 	.arg4_type      = ARG_ANYTHING,
503 };
504 
505 const struct bpf_map_ops sock_map_ops = {
506 	.map_alloc		= sock_map_alloc,
507 	.map_free		= sock_map_free,
508 	.map_get_next_key	= sock_map_get_next_key,
509 	.map_update_elem	= sock_map_update_elem,
510 	.map_delete_elem	= sock_map_delete_elem,
511 	.map_lookup_elem	= sock_map_lookup,
512 	.map_release_uref	= sock_map_release_progs,
513 	.map_check_btf		= map_check_no_btf,
514 };
515 
516 struct bpf_htab_elem {
517 	struct rcu_head rcu;
518 	u32 hash;
519 	struct sock *sk;
520 	struct hlist_node node;
521 	u8 key[0];
522 };
523 
524 struct bpf_htab_bucket {
525 	struct hlist_head head;
526 	raw_spinlock_t lock;
527 };
528 
529 struct bpf_htab {
530 	struct bpf_map map;
531 	struct bpf_htab_bucket *buckets;
532 	u32 buckets_num;
533 	u32 elem_size;
534 	struct sk_psock_progs progs;
535 	atomic_t count;
536 };
537 
538 static inline u32 sock_hash_bucket_hash(const void *key, u32 len)
539 {
540 	return jhash(key, len, 0);
541 }
542 
543 static struct bpf_htab_bucket *sock_hash_select_bucket(struct bpf_htab *htab,
544 						       u32 hash)
545 {
546 	return &htab->buckets[hash & (htab->buckets_num - 1)];
547 }
548 
549 static struct bpf_htab_elem *
550 sock_hash_lookup_elem_raw(struct hlist_head *head, u32 hash, void *key,
551 			  u32 key_size)
552 {
553 	struct bpf_htab_elem *elem;
554 
555 	hlist_for_each_entry_rcu(elem, head, node) {
556 		if (elem->hash == hash &&
557 		    !memcmp(&elem->key, key, key_size))
558 			return elem;
559 	}
560 
561 	return NULL;
562 }
563 
564 static struct sock *__sock_hash_lookup_elem(struct bpf_map *map, void *key)
565 {
566 	struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
567 	u32 key_size = map->key_size, hash;
568 	struct bpf_htab_bucket *bucket;
569 	struct bpf_htab_elem *elem;
570 
571 	WARN_ON_ONCE(!rcu_read_lock_held());
572 
573 	hash = sock_hash_bucket_hash(key, key_size);
574 	bucket = sock_hash_select_bucket(htab, hash);
575 	elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size);
576 
577 	return elem ? elem->sk : NULL;
578 }
579 
580 static void sock_hash_free_elem(struct bpf_htab *htab,
581 				struct bpf_htab_elem *elem)
582 {
583 	atomic_dec(&htab->count);
584 	kfree_rcu(elem, rcu);
585 }
586 
587 static void sock_hash_delete_from_link(struct bpf_map *map, struct sock *sk,
588 				       void *link_raw)
589 {
590 	struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
591 	struct bpf_htab_elem *elem_probe, *elem = link_raw;
592 	struct bpf_htab_bucket *bucket;
593 
594 	WARN_ON_ONCE(!rcu_read_lock_held());
595 	bucket = sock_hash_select_bucket(htab, elem->hash);
596 
597 	/* elem may be deleted in parallel from the map, but access here
598 	 * is okay since it's going away only after RCU grace period.
599 	 * However, we need to check whether it's still present.
600 	 */
601 	raw_spin_lock_bh(&bucket->lock);
602 	elem_probe = sock_hash_lookup_elem_raw(&bucket->head, elem->hash,
603 					       elem->key, map->key_size);
604 	if (elem_probe && elem_probe == elem) {
605 		hlist_del_rcu(&elem->node);
606 		sock_map_unref(elem->sk, elem);
607 		sock_hash_free_elem(htab, elem);
608 	}
609 	raw_spin_unlock_bh(&bucket->lock);
610 }
611 
612 static int sock_hash_delete_elem(struct bpf_map *map, void *key)
613 {
614 	struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
615 	u32 hash, key_size = map->key_size;
616 	struct bpf_htab_bucket *bucket;
617 	struct bpf_htab_elem *elem;
618 	int ret = -ENOENT;
619 
620 	hash = sock_hash_bucket_hash(key, key_size);
621 	bucket = sock_hash_select_bucket(htab, hash);
622 
623 	raw_spin_lock_bh(&bucket->lock);
624 	elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size);
625 	if (elem) {
626 		hlist_del_rcu(&elem->node);
627 		sock_map_unref(elem->sk, elem);
628 		sock_hash_free_elem(htab, elem);
629 		ret = 0;
630 	}
631 	raw_spin_unlock_bh(&bucket->lock);
632 	return ret;
633 }
634 
635 static struct bpf_htab_elem *sock_hash_alloc_elem(struct bpf_htab *htab,
636 						  void *key, u32 key_size,
637 						  u32 hash, struct sock *sk,
638 						  struct bpf_htab_elem *old)
639 {
640 	struct bpf_htab_elem *new;
641 
642 	if (atomic_inc_return(&htab->count) > htab->map.max_entries) {
643 		if (!old) {
644 			atomic_dec(&htab->count);
645 			return ERR_PTR(-E2BIG);
646 		}
647 	}
648 
649 	new = kmalloc_node(htab->elem_size, GFP_ATOMIC | __GFP_NOWARN,
650 			   htab->map.numa_node);
651 	if (!new) {
652 		atomic_dec(&htab->count);
653 		return ERR_PTR(-ENOMEM);
654 	}
655 	memcpy(new->key, key, key_size);
656 	new->sk = sk;
657 	new->hash = hash;
658 	return new;
659 }
660 
661 static int sock_hash_update_common(struct bpf_map *map, void *key,
662 				   struct sock *sk, u64 flags)
663 {
664 	struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
665 	struct inet_connection_sock *icsk = inet_csk(sk);
666 	u32 key_size = map->key_size, hash;
667 	struct bpf_htab_elem *elem, *elem_new;
668 	struct bpf_htab_bucket *bucket;
669 	struct sk_psock_link *link;
670 	struct sk_psock *psock;
671 	int ret;
672 
673 	WARN_ON_ONCE(!rcu_read_lock_held());
674 	if (unlikely(flags > BPF_EXIST))
675 		return -EINVAL;
676 	if (unlikely(icsk->icsk_ulp_data))
677 		return -EINVAL;
678 
679 	link = sk_psock_init_link();
680 	if (!link)
681 		return -ENOMEM;
682 
683 	ret = sock_map_link(map, &htab->progs, sk);
684 	if (ret < 0)
685 		goto out_free;
686 
687 	psock = sk_psock(sk);
688 	WARN_ON_ONCE(!psock);
689 
690 	hash = sock_hash_bucket_hash(key, key_size);
691 	bucket = sock_hash_select_bucket(htab, hash);
692 
693 	raw_spin_lock_bh(&bucket->lock);
694 	elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size);
695 	if (elem && flags == BPF_NOEXIST) {
696 		ret = -EEXIST;
697 		goto out_unlock;
698 	} else if (!elem && flags == BPF_EXIST) {
699 		ret = -ENOENT;
700 		goto out_unlock;
701 	}
702 
703 	elem_new = sock_hash_alloc_elem(htab, key, key_size, hash, sk, elem);
704 	if (IS_ERR(elem_new)) {
705 		ret = PTR_ERR(elem_new);
706 		goto out_unlock;
707 	}
708 
709 	sock_map_add_link(psock, link, map, elem_new);
710 	/* Add new element to the head of the list, so that
711 	 * concurrent search will find it before old elem.
712 	 */
713 	hlist_add_head_rcu(&elem_new->node, &bucket->head);
714 	if (elem) {
715 		hlist_del_rcu(&elem->node);
716 		sock_map_unref(elem->sk, elem);
717 		sock_hash_free_elem(htab, elem);
718 	}
719 	raw_spin_unlock_bh(&bucket->lock);
720 	return 0;
721 out_unlock:
722 	raw_spin_unlock_bh(&bucket->lock);
723 	sk_psock_put(sk, psock);
724 out_free:
725 	sk_psock_free_link(link);
726 	return ret;
727 }
728 
729 static int sock_hash_update_elem(struct bpf_map *map, void *key,
730 				 void *value, u64 flags)
731 {
732 	u32 ufd = *(u32 *)value;
733 	struct socket *sock;
734 	struct sock *sk;
735 	int ret;
736 
737 	sock = sockfd_lookup(ufd, &ret);
738 	if (!sock)
739 		return ret;
740 	sk = sock->sk;
741 	if (!sk) {
742 		ret = -EINVAL;
743 		goto out;
744 	}
745 	if (!sock_map_sk_is_suitable(sk)) {
746 		ret = -EOPNOTSUPP;
747 		goto out;
748 	}
749 
750 	sock_map_sk_acquire(sk);
751 	if (sk->sk_state != TCP_ESTABLISHED)
752 		ret = -EOPNOTSUPP;
753 	else
754 		ret = sock_hash_update_common(map, key, sk, flags);
755 	sock_map_sk_release(sk);
756 out:
757 	fput(sock->file);
758 	return ret;
759 }
760 
761 static int sock_hash_get_next_key(struct bpf_map *map, void *key,
762 				  void *key_next)
763 {
764 	struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
765 	struct bpf_htab_elem *elem, *elem_next;
766 	u32 hash, key_size = map->key_size;
767 	struct hlist_head *head;
768 	int i = 0;
769 
770 	if (!key)
771 		goto find_first_elem;
772 	hash = sock_hash_bucket_hash(key, key_size);
773 	head = &sock_hash_select_bucket(htab, hash)->head;
774 	elem = sock_hash_lookup_elem_raw(head, hash, key, key_size);
775 	if (!elem)
776 		goto find_first_elem;
777 
778 	elem_next = hlist_entry_safe(rcu_dereference_raw(hlist_next_rcu(&elem->node)),
779 				     struct bpf_htab_elem, node);
780 	if (elem_next) {
781 		memcpy(key_next, elem_next->key, key_size);
782 		return 0;
783 	}
784 
785 	i = hash & (htab->buckets_num - 1);
786 	i++;
787 find_first_elem:
788 	for (; i < htab->buckets_num; i++) {
789 		head = &sock_hash_select_bucket(htab, i)->head;
790 		elem_next = hlist_entry_safe(rcu_dereference_raw(hlist_first_rcu(head)),
791 					     struct bpf_htab_elem, node);
792 		if (elem_next) {
793 			memcpy(key_next, elem_next->key, key_size);
794 			return 0;
795 		}
796 	}
797 
798 	return -ENOENT;
799 }
800 
801 static struct bpf_map *sock_hash_alloc(union bpf_attr *attr)
802 {
803 	struct bpf_htab *htab;
804 	int i, err;
805 	u64 cost;
806 
807 	if (!capable(CAP_NET_ADMIN))
808 		return ERR_PTR(-EPERM);
809 	if (attr->max_entries == 0 ||
810 	    attr->key_size    == 0 ||
811 	    attr->value_size  != 4 ||
812 	    attr->map_flags & ~SOCK_CREATE_FLAG_MASK)
813 		return ERR_PTR(-EINVAL);
814 	if (attr->key_size > MAX_BPF_STACK)
815 		return ERR_PTR(-E2BIG);
816 
817 	htab = kzalloc(sizeof(*htab), GFP_USER);
818 	if (!htab)
819 		return ERR_PTR(-ENOMEM);
820 
821 	bpf_map_init_from_attr(&htab->map, attr);
822 
823 	htab->buckets_num = roundup_pow_of_two(htab->map.max_entries);
824 	htab->elem_size = sizeof(struct bpf_htab_elem) +
825 			  round_up(htab->map.key_size, 8);
826 	if (htab->buckets_num == 0 ||
827 	    htab->buckets_num > U32_MAX / sizeof(struct bpf_htab_bucket)) {
828 		err = -EINVAL;
829 		goto free_htab;
830 	}
831 
832 	cost = (u64) htab->buckets_num * sizeof(struct bpf_htab_bucket) +
833 	       (u64) htab->elem_size * htab->map.max_entries;
834 	if (cost >= U32_MAX - PAGE_SIZE) {
835 		err = -EINVAL;
836 		goto free_htab;
837 	}
838 
839 	htab->buckets = bpf_map_area_alloc(htab->buckets_num *
840 					   sizeof(struct bpf_htab_bucket),
841 					   htab->map.numa_node);
842 	if (!htab->buckets) {
843 		err = -ENOMEM;
844 		goto free_htab;
845 	}
846 
847 	for (i = 0; i < htab->buckets_num; i++) {
848 		INIT_HLIST_HEAD(&htab->buckets[i].head);
849 		raw_spin_lock_init(&htab->buckets[i].lock);
850 	}
851 
852 	return &htab->map;
853 free_htab:
854 	kfree(htab);
855 	return ERR_PTR(err);
856 }
857 
858 static void sock_hash_free(struct bpf_map *map)
859 {
860 	struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
861 	struct bpf_htab_bucket *bucket;
862 	struct bpf_htab_elem *elem;
863 	struct hlist_node *node;
864 	int i;
865 
866 	synchronize_rcu();
867 	for (i = 0; i < htab->buckets_num; i++) {
868 		bucket = sock_hash_select_bucket(htab, i);
869 		raw_spin_lock_bh(&bucket->lock);
870 		hlist_for_each_entry_safe(elem, node, &bucket->head, node) {
871 			hlist_del_rcu(&elem->node);
872 			lock_sock(elem->sk);
873 			rcu_read_lock();
874 			sock_map_unref(elem->sk, elem);
875 			rcu_read_unlock();
876 			release_sock(elem->sk);
877 		}
878 		raw_spin_unlock_bh(&bucket->lock);
879 	}
880 
881 	/* wait for psock readers accessing its map link */
882 	synchronize_rcu();
883 
884 	bpf_map_area_free(htab->buckets);
885 	kfree(htab);
886 }
887 
888 static void sock_hash_release_progs(struct bpf_map *map)
889 {
890 	psock_progs_drop(&container_of(map, struct bpf_htab, map)->progs);
891 }
892 
893 BPF_CALL_4(bpf_sock_hash_update, struct bpf_sock_ops_kern *, sops,
894 	   struct bpf_map *, map, void *, key, u64, flags)
895 {
896 	WARN_ON_ONCE(!rcu_read_lock_held());
897 
898 	if (likely(sock_map_sk_is_suitable(sops->sk) &&
899 		   sock_map_op_okay(sops)))
900 		return sock_hash_update_common(map, key, sops->sk, flags);
901 	return -EOPNOTSUPP;
902 }
903 
904 const struct bpf_func_proto bpf_sock_hash_update_proto = {
905 	.func		= bpf_sock_hash_update,
906 	.gpl_only	= false,
907 	.pkt_access	= true,
908 	.ret_type	= RET_INTEGER,
909 	.arg1_type	= ARG_PTR_TO_CTX,
910 	.arg2_type	= ARG_CONST_MAP_PTR,
911 	.arg3_type	= ARG_PTR_TO_MAP_KEY,
912 	.arg4_type	= ARG_ANYTHING,
913 };
914 
915 BPF_CALL_4(bpf_sk_redirect_hash, struct sk_buff *, skb,
916 	   struct bpf_map *, map, void *, key, u64, flags)
917 {
918 	struct tcp_skb_cb *tcb = TCP_SKB_CB(skb);
919 
920 	if (unlikely(flags & ~(BPF_F_INGRESS)))
921 		return SK_DROP;
922 	tcb->bpf.flags = flags;
923 	tcb->bpf.sk_redir = __sock_hash_lookup_elem(map, key);
924 	if (!tcb->bpf.sk_redir)
925 		return SK_DROP;
926 	return SK_PASS;
927 }
928 
929 const struct bpf_func_proto bpf_sk_redirect_hash_proto = {
930 	.func           = bpf_sk_redirect_hash,
931 	.gpl_only       = false,
932 	.ret_type       = RET_INTEGER,
933 	.arg1_type	= ARG_PTR_TO_CTX,
934 	.arg2_type      = ARG_CONST_MAP_PTR,
935 	.arg3_type      = ARG_PTR_TO_MAP_KEY,
936 	.arg4_type      = ARG_ANYTHING,
937 };
938 
939 BPF_CALL_4(bpf_msg_redirect_hash, struct sk_msg *, msg,
940 	   struct bpf_map *, map, void *, key, u64, flags)
941 {
942 	if (unlikely(flags & ~(BPF_F_INGRESS)))
943 		return SK_DROP;
944 	msg->flags = flags;
945 	msg->sk_redir = __sock_hash_lookup_elem(map, key);
946 	if (!msg->sk_redir)
947 		return SK_DROP;
948 	return SK_PASS;
949 }
950 
951 const struct bpf_func_proto bpf_msg_redirect_hash_proto = {
952 	.func           = bpf_msg_redirect_hash,
953 	.gpl_only       = false,
954 	.ret_type       = RET_INTEGER,
955 	.arg1_type	= ARG_PTR_TO_CTX,
956 	.arg2_type      = ARG_CONST_MAP_PTR,
957 	.arg3_type      = ARG_PTR_TO_MAP_KEY,
958 	.arg4_type      = ARG_ANYTHING,
959 };
960 
961 const struct bpf_map_ops sock_hash_ops = {
962 	.map_alloc		= sock_hash_alloc,
963 	.map_free		= sock_hash_free,
964 	.map_get_next_key	= sock_hash_get_next_key,
965 	.map_update_elem	= sock_hash_update_elem,
966 	.map_delete_elem	= sock_hash_delete_elem,
967 	.map_lookup_elem	= sock_map_lookup,
968 	.map_release_uref	= sock_hash_release_progs,
969 	.map_check_btf		= map_check_no_btf,
970 };
971 
972 static struct sk_psock_progs *sock_map_progs(struct bpf_map *map)
973 {
974 	switch (map->map_type) {
975 	case BPF_MAP_TYPE_SOCKMAP:
976 		return &container_of(map, struct bpf_stab, map)->progs;
977 	case BPF_MAP_TYPE_SOCKHASH:
978 		return &container_of(map, struct bpf_htab, map)->progs;
979 	default:
980 		break;
981 	}
982 
983 	return NULL;
984 }
985 
986 int sock_map_prog_update(struct bpf_map *map, struct bpf_prog *prog,
987 			 u32 which)
988 {
989 	struct sk_psock_progs *progs = sock_map_progs(map);
990 
991 	if (!progs)
992 		return -EOPNOTSUPP;
993 
994 	switch (which) {
995 	case BPF_SK_MSG_VERDICT:
996 		psock_set_prog(&progs->msg_parser, prog);
997 		break;
998 	case BPF_SK_SKB_STREAM_PARSER:
999 		psock_set_prog(&progs->skb_parser, prog);
1000 		break;
1001 	case BPF_SK_SKB_STREAM_VERDICT:
1002 		psock_set_prog(&progs->skb_verdict, prog);
1003 		break;
1004 	default:
1005 		return -EOPNOTSUPP;
1006 	}
1007 
1008 	return 0;
1009 }
1010 
1011 void sk_psock_unlink(struct sock *sk, struct sk_psock_link *link)
1012 {
1013 	switch (link->map->map_type) {
1014 	case BPF_MAP_TYPE_SOCKMAP:
1015 		return sock_map_delete_from_link(link->map, sk,
1016 						 link->link_raw);
1017 	case BPF_MAP_TYPE_SOCKHASH:
1018 		return sock_hash_delete_from_link(link->map, sk,
1019 						  link->link_raw);
1020 	default:
1021 		break;
1022 	}
1023 }
1024