xref: /openbmc/linux/net/core/sock_map.c (revision 31e67366) 
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/btf_ids.h>
6 #include <linux/filter.h>
7 #include <linux/errno.h>
8 #include <linux/file.h>
9 #include <linux/net.h>
10 #include <linux/workqueue.h>
11 #include <linux/skmsg.h>
12 #include <linux/list.h>
13 #include <linux/jhash.h>
14 #include <linux/sock_diag.h>
15 #include <net/udp.h>
16 
17 struct bpf_stab {
18 	struct bpf_map map;
19 	struct sock **sks;
20 	struct sk_psock_progs progs;
21 	raw_spinlock_t lock;
22 };
23 
24 #define SOCK_CREATE_FLAG_MASK				\
25 	(BPF_F_NUMA_NODE | BPF_F_RDONLY | BPF_F_WRONLY)
26 
27 static struct bpf_map *sock_map_alloc(union bpf_attr *attr)
28 {
29 	struct bpf_stab *stab;
30 
31 	if (!capable(CAP_NET_ADMIN))
32 		return ERR_PTR(-EPERM);
33 	if (attr->max_entries == 0 ||
34 	    attr->key_size    != 4 ||
35 	    (attr->value_size != sizeof(u32) &&
36 	     attr->value_size != sizeof(u64)) ||
37 	    attr->map_flags & ~SOCK_CREATE_FLAG_MASK)
38 		return ERR_PTR(-EINVAL);
39 
40 	stab = kzalloc(sizeof(*stab), GFP_USER | __GFP_ACCOUNT);
41 	if (!stab)
42 		return ERR_PTR(-ENOMEM);
43 
44 	bpf_map_init_from_attr(&stab->map, attr);
45 	raw_spin_lock_init(&stab->lock);
46 
47 	stab->sks = bpf_map_area_alloc(stab->map.max_entries *
48 				       sizeof(struct sock *),
49 				       stab->map.numa_node);
50 	if (!stab->sks) {
51 		kfree(stab);
52 		return ERR_PTR(-ENOMEM);
53 	}
54 
55 	return &stab->map;
56 }
57 
58 int sock_map_get_from_fd(const union bpf_attr *attr, struct bpf_prog *prog)
59 {
60 	u32 ufd = attr->target_fd;
61 	struct bpf_map *map;
62 	struct fd f;
63 	int ret;
64 
65 	if (attr->attach_flags || attr->replace_bpf_fd)
66 		return -EINVAL;
67 
68 	f = fdget(ufd);
69 	map = __bpf_map_get(f);
70 	if (IS_ERR(map))
71 		return PTR_ERR(map);
72 	ret = sock_map_prog_update(map, prog, NULL, attr->attach_type);
73 	fdput(f);
74 	return ret;
75 }
76 
77 int sock_map_prog_detach(const union bpf_attr *attr, enum bpf_prog_type ptype)
78 {
79 	u32 ufd = attr->target_fd;
80 	struct bpf_prog *prog;
81 	struct bpf_map *map;
82 	struct fd f;
83 	int ret;
84 
85 	if (attr->attach_flags || attr->replace_bpf_fd)
86 		return -EINVAL;
87 
88 	f = fdget(ufd);
89 	map = __bpf_map_get(f);
90 	if (IS_ERR(map))
91 		return PTR_ERR(map);
92 
93 	prog = bpf_prog_get(attr->attach_bpf_fd);
94 	if (IS_ERR(prog)) {
95 		ret = PTR_ERR(prog);
96 		goto put_map;
97 	}
98 
99 	if (prog->type != ptype) {
100 		ret = -EINVAL;
101 		goto put_prog;
102 	}
103 
104 	ret = sock_map_prog_update(map, NULL, prog, attr->attach_type);
105 put_prog:
106 	bpf_prog_put(prog);
107 put_map:
108 	fdput(f);
109 	return ret;
110 }
111 
112 static void sock_map_sk_acquire(struct sock *sk)
113 	__acquires(&sk->sk_lock.slock)
114 {
115 	lock_sock(sk);
116 	preempt_disable();
117 	rcu_read_lock();
118 }
119 
120 static void sock_map_sk_release(struct sock *sk)
121 	__releases(&sk->sk_lock.slock)
122 {
123 	rcu_read_unlock();
124 	preempt_enable();
125 	release_sock(sk);
126 }
127 
128 static void sock_map_add_link(struct sk_psock *psock,
129 			      struct sk_psock_link *link,
130 			      struct bpf_map *map, void *link_raw)
131 {
132 	link->link_raw = link_raw;
133 	link->map = map;
134 	spin_lock_bh(&psock->link_lock);
135 	list_add_tail(&link->list, &psock->link);
136 	spin_unlock_bh(&psock->link_lock);
137 }
138 
139 static void sock_map_del_link(struct sock *sk,
140 			      struct sk_psock *psock, void *link_raw)
141 {
142 	bool strp_stop = false, verdict_stop = false;
143 	struct sk_psock_link *link, *tmp;
144 
145 	spin_lock_bh(&psock->link_lock);
146 	list_for_each_entry_safe(link, tmp, &psock->link, list) {
147 		if (link->link_raw == link_raw) {
148 			struct bpf_map *map = link->map;
149 			struct bpf_stab *stab = container_of(map, struct bpf_stab,
150 							     map);
151 			if (psock->parser.enabled && stab->progs.skb_parser)
152 				strp_stop = true;
153 			if (psock->parser.enabled && stab->progs.skb_verdict)
154 				verdict_stop = true;
155 			list_del(&link->list);
156 			sk_psock_free_link(link);
157 		}
158 	}
159 	spin_unlock_bh(&psock->link_lock);
160 	if (strp_stop || verdict_stop) {
161 		write_lock_bh(&sk->sk_callback_lock);
162 		if (strp_stop)
163 			sk_psock_stop_strp(sk, psock);
164 		else
165 			sk_psock_stop_verdict(sk, psock);
166 		write_unlock_bh(&sk->sk_callback_lock);
167 	}
168 }
169 
170 static void sock_map_unref(struct sock *sk, void *link_raw)
171 {
172 	struct sk_psock *psock = sk_psock(sk);
173 
174 	if (likely(psock)) {
175 		sock_map_del_link(sk, psock, link_raw);
176 		sk_psock_put(sk, psock);
177 	}
178 }
179 
180 static int sock_map_init_proto(struct sock *sk, struct sk_psock *psock)
181 {
182 	struct proto *prot;
183 
184 	switch (sk->sk_type) {
185 	case SOCK_STREAM:
186 		prot = tcp_bpf_get_proto(sk, psock);
187 		break;
188 
189 	case SOCK_DGRAM:
190 		prot = udp_bpf_get_proto(sk, psock);
191 		break;
192 
193 	default:
194 		return -EINVAL;
195 	}
196 
197 	if (IS_ERR(prot))
198 		return PTR_ERR(prot);
199 
200 	sk_psock_update_proto(sk, psock, prot);
201 	return 0;
202 }
203 
204 static struct sk_psock *sock_map_psock_get_checked(struct sock *sk)
205 {
206 	struct sk_psock *psock;
207 
208 	rcu_read_lock();
209 	psock = sk_psock(sk);
210 	if (psock) {
211 		if (sk->sk_prot->close != sock_map_close) {
212 			psock = ERR_PTR(-EBUSY);
213 			goto out;
214 		}
215 
216 		if (!refcount_inc_not_zero(&psock->refcnt))
217 			psock = ERR_PTR(-EBUSY);
218 	}
219 out:
220 	rcu_read_unlock();
221 	return psock;
222 }
223 
224 static int sock_map_link(struct bpf_map *map, struct sk_psock_progs *progs,
225 			 struct sock *sk)
226 {
227 	struct bpf_prog *msg_parser, *skb_parser, *skb_verdict;
228 	struct sk_psock *psock;
229 	int ret;
230 
231 	skb_verdict = READ_ONCE(progs->skb_verdict);
232 	if (skb_verdict) {
233 		skb_verdict = bpf_prog_inc_not_zero(skb_verdict);
234 		if (IS_ERR(skb_verdict))
235 			return PTR_ERR(skb_verdict);
236 	}
237 
238 	skb_parser = READ_ONCE(progs->skb_parser);
239 	if (skb_parser) {
240 		skb_parser = bpf_prog_inc_not_zero(skb_parser);
241 		if (IS_ERR(skb_parser)) {
242 			ret = PTR_ERR(skb_parser);
243 			goto out_put_skb_verdict;
244 		}
245 	}
246 
247 	msg_parser = READ_ONCE(progs->msg_parser);
248 	if (msg_parser) {
249 		msg_parser = bpf_prog_inc_not_zero(msg_parser);
250 		if (IS_ERR(msg_parser)) {
251 			ret = PTR_ERR(msg_parser);
252 			goto out_put_skb_parser;
253 		}
254 	}
255 
256 	psock = sock_map_psock_get_checked(sk);
257 	if (IS_ERR(psock)) {
258 		ret = PTR_ERR(psock);
259 		goto out_progs;
260 	}
261 
262 	if (psock) {
263 		if ((msg_parser && READ_ONCE(psock->progs.msg_parser)) ||
264 		    (skb_parser  && READ_ONCE(psock->progs.skb_parser)) ||
265 		    (skb_verdict && READ_ONCE(psock->progs.skb_verdict))) {
266 			sk_psock_put(sk, psock);
267 			ret = -EBUSY;
268 			goto out_progs;
269 		}
270 	} else {
271 		psock = sk_psock_init(sk, map->numa_node);
272 		if (IS_ERR(psock)) {
273 			ret = PTR_ERR(psock);
274 			goto out_progs;
275 		}
276 	}
277 
278 	if (msg_parser)
279 		psock_set_prog(&psock->progs.msg_parser, msg_parser);
280 
281 	ret = sock_map_init_proto(sk, psock);
282 	if (ret < 0)
283 		goto out_drop;
284 
285 	write_lock_bh(&sk->sk_callback_lock);
286 	if (skb_parser && skb_verdict && !psock->parser.enabled) {
287 		ret = sk_psock_init_strp(sk, psock);
288 		if (ret)
289 			goto out_unlock_drop;
290 		psock_set_prog(&psock->progs.skb_verdict, skb_verdict);
291 		psock_set_prog(&psock->progs.skb_parser, skb_parser);
292 		sk_psock_start_strp(sk, psock);
293 	} else if (!skb_parser && skb_verdict && !psock->parser.enabled) {
294 		psock_set_prog(&psock->progs.skb_verdict, skb_verdict);
295 		sk_psock_start_verdict(sk,psock);
296 	}
297 	write_unlock_bh(&sk->sk_callback_lock);
298 	return 0;
299 out_unlock_drop:
300 	write_unlock_bh(&sk->sk_callback_lock);
301 out_drop:
302 	sk_psock_put(sk, psock);
303 out_progs:
304 	if (msg_parser)
305 		bpf_prog_put(msg_parser);
306 out_put_skb_parser:
307 	if (skb_parser)
308 		bpf_prog_put(skb_parser);
309 out_put_skb_verdict:
310 	if (skb_verdict)
311 		bpf_prog_put(skb_verdict);
312 	return ret;
313 }
314 
315 static int sock_map_link_no_progs(struct bpf_map *map, struct sock *sk)
316 {
317 	struct sk_psock *psock;
318 	int ret;
319 
320 	psock = sock_map_psock_get_checked(sk);
321 	if (IS_ERR(psock))
322 		return PTR_ERR(psock);
323 
324 	if (!psock) {
325 		psock = sk_psock_init(sk, map->numa_node);
326 		if (IS_ERR(psock))
327 			return PTR_ERR(psock);
328 	}
329 
330 	ret = sock_map_init_proto(sk, psock);
331 	if (ret < 0)
332 		sk_psock_put(sk, psock);
333 	return ret;
334 }
335 
336 static void sock_map_free(struct bpf_map *map)
337 {
338 	struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
339 	int i;
340 
341 	/* After the sync no updates or deletes will be in-flight so it
342 	 * is safe to walk map and remove entries without risking a race
343 	 * in EEXIST update case.
344 	 */
345 	synchronize_rcu();
346 	for (i = 0; i < stab->map.max_entries; i++) {
347 		struct sock **psk = &stab->sks[i];
348 		struct sock *sk;
349 
350 		sk = xchg(psk, NULL);
351 		if (sk) {
352 			lock_sock(sk);
353 			rcu_read_lock();
354 			sock_map_unref(sk, psk);
355 			rcu_read_unlock();
356 			release_sock(sk);
357 		}
358 	}
359 
360 	/* wait for psock readers accessing its map link */
361 	synchronize_rcu();
362 
363 	bpf_map_area_free(stab->sks);
364 	kfree(stab);
365 }
366 
367 static void sock_map_release_progs(struct bpf_map *map)
368 {
369 	psock_progs_drop(&container_of(map, struct bpf_stab, map)->progs);
370 }
371 
372 static struct sock *__sock_map_lookup_elem(struct bpf_map *map, u32 key)
373 {
374 	struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
375 
376 	WARN_ON_ONCE(!rcu_read_lock_held());
377 
378 	if (unlikely(key >= map->max_entries))
379 		return NULL;
380 	return READ_ONCE(stab->sks[key]);
381 }
382 
383 static void *sock_map_lookup(struct bpf_map *map, void *key)
384 {
385 	struct sock *sk;
386 
387 	sk = __sock_map_lookup_elem(map, *(u32 *)key);
388 	if (!sk)
389 		return NULL;
390 	if (sk_is_refcounted(sk) && !refcount_inc_not_zero(&sk->sk_refcnt))
391 		return NULL;
392 	return sk;
393 }
394 
395 static void *sock_map_lookup_sys(struct bpf_map *map, void *key)
396 {
397 	struct sock *sk;
398 
399 	if (map->value_size != sizeof(u64))
400 		return ERR_PTR(-ENOSPC);
401 
402 	sk = __sock_map_lookup_elem(map, *(u32 *)key);
403 	if (!sk)
404 		return ERR_PTR(-ENOENT);
405 
406 	__sock_gen_cookie(sk);
407 	return &sk->sk_cookie;
408 }
409 
410 static int __sock_map_delete(struct bpf_stab *stab, struct sock *sk_test,
411 			     struct sock **psk)
412 {
413 	struct sock *sk;
414 	int err = 0;
415 
416 	raw_spin_lock_bh(&stab->lock);
417 	sk = *psk;
418 	if (!sk_test || sk_test == sk)
419 		sk = xchg(psk, NULL);
420 
421 	if (likely(sk))
422 		sock_map_unref(sk, psk);
423 	else
424 		err = -EINVAL;
425 
426 	raw_spin_unlock_bh(&stab->lock);
427 	return err;
428 }
429 
430 static void sock_map_delete_from_link(struct bpf_map *map, struct sock *sk,
431 				      void *link_raw)
432 {
433 	struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
434 
435 	__sock_map_delete(stab, sk, link_raw);
436 }
437 
438 static int sock_map_delete_elem(struct bpf_map *map, void *key)
439 {
440 	struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
441 	u32 i = *(u32 *)key;
442 	struct sock **psk;
443 
444 	if (unlikely(i >= map->max_entries))
445 		return -EINVAL;
446 
447 	psk = &stab->sks[i];
448 	return __sock_map_delete(stab, NULL, psk);
449 }
450 
451 static int sock_map_get_next_key(struct bpf_map *map, void *key, void *next)
452 {
453 	struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
454 	u32 i = key ? *(u32 *)key : U32_MAX;
455 	u32 *key_next = next;
456 
457 	if (i == stab->map.max_entries - 1)
458 		return -ENOENT;
459 	if (i >= stab->map.max_entries)
460 		*key_next = 0;
461 	else
462 		*key_next = i + 1;
463 	return 0;
464 }
465 
466 static bool sock_map_redirect_allowed(const struct sock *sk);
467 
468 static int sock_map_update_common(struct bpf_map *map, u32 idx,
469 				  struct sock *sk, u64 flags)
470 {
471 	struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
472 	struct sk_psock_link *link;
473 	struct sk_psock *psock;
474 	struct sock *osk;
475 	int ret;
476 
477 	WARN_ON_ONCE(!rcu_read_lock_held());
478 	if (unlikely(flags > BPF_EXIST))
479 		return -EINVAL;
480 	if (unlikely(idx >= map->max_entries))
481 		return -E2BIG;
482 
483 	link = sk_psock_init_link();
484 	if (!link)
485 		return -ENOMEM;
486 
487 	/* Only sockets we can redirect into/from in BPF need to hold
488 	 * refs to parser/verdict progs and have their sk_data_ready
489 	 * and sk_write_space callbacks overridden.
490 	 */
491 	if (sock_map_redirect_allowed(sk))
492 		ret = sock_map_link(map, &stab->progs, sk);
493 	else
494 		ret = sock_map_link_no_progs(map, sk);
495 	if (ret < 0)
496 		goto out_free;
497 
498 	psock = sk_psock(sk);
499 	WARN_ON_ONCE(!psock);
500 
501 	raw_spin_lock_bh(&stab->lock);
502 	osk = stab->sks[idx];
503 	if (osk && flags == BPF_NOEXIST) {
504 		ret = -EEXIST;
505 		goto out_unlock;
506 	} else if (!osk && flags == BPF_EXIST) {
507 		ret = -ENOENT;
508 		goto out_unlock;
509 	}
510 
511 	sock_map_add_link(psock, link, map, &stab->sks[idx]);
512 	stab->sks[idx] = sk;
513 	if (osk)
514 		sock_map_unref(osk, &stab->sks[idx]);
515 	raw_spin_unlock_bh(&stab->lock);
516 	return 0;
517 out_unlock:
518 	raw_spin_unlock_bh(&stab->lock);
519 	if (psock)
520 		sk_psock_put(sk, psock);
521 out_free:
522 	sk_psock_free_link(link);
523 	return ret;
524 }
525 
526 static bool sock_map_op_okay(const struct bpf_sock_ops_kern *ops)
527 {
528 	return ops->op == BPF_SOCK_OPS_PASSIVE_ESTABLISHED_CB ||
529 	       ops->op == BPF_SOCK_OPS_ACTIVE_ESTABLISHED_CB ||
530 	       ops->op == BPF_SOCK_OPS_TCP_LISTEN_CB;
531 }
532 
533 static bool sk_is_tcp(const struct sock *sk)
534 {
535 	return sk->sk_type == SOCK_STREAM &&
536 	       sk->sk_protocol == IPPROTO_TCP;
537 }
538 
539 static bool sk_is_udp(const struct sock *sk)
540 {
541 	return sk->sk_type == SOCK_DGRAM &&
542 	       sk->sk_protocol == IPPROTO_UDP;
543 }
544 
545 static bool sock_map_redirect_allowed(const struct sock *sk)
546 {
547 	return sk_is_tcp(sk) && sk->sk_state != TCP_LISTEN;
548 }
549 
550 static bool sock_map_sk_is_suitable(const struct sock *sk)
551 {
552 	return sk_is_tcp(sk) || sk_is_udp(sk);
553 }
554 
555 static bool sock_map_sk_state_allowed(const struct sock *sk)
556 {
557 	if (sk_is_tcp(sk))
558 		return (1 << sk->sk_state) & (TCPF_ESTABLISHED | TCPF_LISTEN);
559 	else if (sk_is_udp(sk))
560 		return sk_hashed(sk);
561 
562 	return false;
563 }
564 
565 static int sock_hash_update_common(struct bpf_map *map, void *key,
566 				   struct sock *sk, u64 flags);
567 
568 int sock_map_update_elem_sys(struct bpf_map *map, void *key, void *value,
569 			     u64 flags)
570 {
571 	struct socket *sock;
572 	struct sock *sk;
573 	int ret;
574 	u64 ufd;
575 
576 	if (map->value_size == sizeof(u64))
577 		ufd = *(u64 *)value;
578 	else
579 		ufd = *(u32 *)value;
580 	if (ufd > S32_MAX)
581 		return -EINVAL;
582 
583 	sock = sockfd_lookup(ufd, &ret);
584 	if (!sock)
585 		return ret;
586 	sk = sock->sk;
587 	if (!sk) {
588 		ret = -EINVAL;
589 		goto out;
590 	}
591 	if (!sock_map_sk_is_suitable(sk)) {
592 		ret = -EOPNOTSUPP;
593 		goto out;
594 	}
595 
596 	sock_map_sk_acquire(sk);
597 	if (!sock_map_sk_state_allowed(sk))
598 		ret = -EOPNOTSUPP;
599 	else if (map->map_type == BPF_MAP_TYPE_SOCKMAP)
600 		ret = sock_map_update_common(map, *(u32 *)key, sk, flags);
601 	else
602 		ret = sock_hash_update_common(map, key, sk, flags);
603 	sock_map_sk_release(sk);
604 out:
605 	sockfd_put(sock);
606 	return ret;
607 }
608 
609 static int sock_map_update_elem(struct bpf_map *map, void *key,
610 				void *value, u64 flags)
611 {
612 	struct sock *sk = (struct sock *)value;
613 	int ret;
614 
615 	if (unlikely(!sk || !sk_fullsock(sk)))
616 		return -EINVAL;
617 
618 	if (!sock_map_sk_is_suitable(sk))
619 		return -EOPNOTSUPP;
620 
621 	local_bh_disable();
622 	bh_lock_sock(sk);
623 	if (!sock_map_sk_state_allowed(sk))
624 		ret = -EOPNOTSUPP;
625 	else if (map->map_type == BPF_MAP_TYPE_SOCKMAP)
626 		ret = sock_map_update_common(map, *(u32 *)key, sk, flags);
627 	else
628 		ret = sock_hash_update_common(map, key, sk, flags);
629 	bh_unlock_sock(sk);
630 	local_bh_enable();
631 	return ret;
632 }
633 
634 BPF_CALL_4(bpf_sock_map_update, struct bpf_sock_ops_kern *, sops,
635 	   struct bpf_map *, map, void *, key, u64, flags)
636 {
637 	WARN_ON_ONCE(!rcu_read_lock_held());
638 
639 	if (likely(sock_map_sk_is_suitable(sops->sk) &&
640 		   sock_map_op_okay(sops)))
641 		return sock_map_update_common(map, *(u32 *)key, sops->sk,
642 					      flags);
643 	return -EOPNOTSUPP;
644 }
645 
646 const struct bpf_func_proto bpf_sock_map_update_proto = {
647 	.func		= bpf_sock_map_update,
648 	.gpl_only	= false,
649 	.pkt_access	= true,
650 	.ret_type	= RET_INTEGER,
651 	.arg1_type	= ARG_PTR_TO_CTX,
652 	.arg2_type	= ARG_CONST_MAP_PTR,
653 	.arg3_type	= ARG_PTR_TO_MAP_KEY,
654 	.arg4_type	= ARG_ANYTHING,
655 };
656 
657 BPF_CALL_4(bpf_sk_redirect_map, struct sk_buff *, skb,
658 	   struct bpf_map *, map, u32, key, u64, flags)
659 {
660 	struct tcp_skb_cb *tcb = TCP_SKB_CB(skb);
661 	struct sock *sk;
662 
663 	if (unlikely(flags & ~(BPF_F_INGRESS)))
664 		return SK_DROP;
665 
666 	sk = __sock_map_lookup_elem(map, key);
667 	if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
668 		return SK_DROP;
669 
670 	tcb->bpf.flags = flags;
671 	tcb->bpf.sk_redir = sk;
672 	return SK_PASS;
673 }
674 
675 const struct bpf_func_proto bpf_sk_redirect_map_proto = {
676 	.func           = bpf_sk_redirect_map,
677 	.gpl_only       = false,
678 	.ret_type       = RET_INTEGER,
679 	.arg1_type	= ARG_PTR_TO_CTX,
680 	.arg2_type      = ARG_CONST_MAP_PTR,
681 	.arg3_type      = ARG_ANYTHING,
682 	.arg4_type      = ARG_ANYTHING,
683 };
684 
685 BPF_CALL_4(bpf_msg_redirect_map, struct sk_msg *, msg,
686 	   struct bpf_map *, map, u32, key, u64, flags)
687 {
688 	struct sock *sk;
689 
690 	if (unlikely(flags & ~(BPF_F_INGRESS)))
691 		return SK_DROP;
692 
693 	sk = __sock_map_lookup_elem(map, key);
694 	if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
695 		return SK_DROP;
696 
697 	msg->flags = flags;
698 	msg->sk_redir = sk;
699 	return SK_PASS;
700 }
701 
702 const struct bpf_func_proto bpf_msg_redirect_map_proto = {
703 	.func           = bpf_msg_redirect_map,
704 	.gpl_only       = false,
705 	.ret_type       = RET_INTEGER,
706 	.arg1_type	= ARG_PTR_TO_CTX,
707 	.arg2_type      = ARG_CONST_MAP_PTR,
708 	.arg3_type      = ARG_ANYTHING,
709 	.arg4_type      = ARG_ANYTHING,
710 };
711 
712 struct sock_map_seq_info {
713 	struct bpf_map *map;
714 	struct sock *sk;
715 	u32 index;
716 };
717 
718 struct bpf_iter__sockmap {
719 	__bpf_md_ptr(struct bpf_iter_meta *, meta);
720 	__bpf_md_ptr(struct bpf_map *, map);
721 	__bpf_md_ptr(void *, key);
722 	__bpf_md_ptr(struct sock *, sk);
723 };
724 
725 DEFINE_BPF_ITER_FUNC(sockmap, struct bpf_iter_meta *meta,
726 		     struct bpf_map *map, void *key,
727 		     struct sock *sk)
728 
729 static void *sock_map_seq_lookup_elem(struct sock_map_seq_info *info)
730 {
731 	if (unlikely(info->index >= info->map->max_entries))
732 		return NULL;
733 
734 	info->sk = __sock_map_lookup_elem(info->map, info->index);
735 
736 	/* can't return sk directly, since that might be NULL */
737 	return info;
738 }
739 
740 static void *sock_map_seq_start(struct seq_file *seq, loff_t *pos)
741 	__acquires(rcu)
742 {
743 	struct sock_map_seq_info *info = seq->private;
744 
745 	if (*pos == 0)
746 		++*pos;
747 
748 	/* pairs with sock_map_seq_stop */
749 	rcu_read_lock();
750 	return sock_map_seq_lookup_elem(info);
751 }
752 
753 static void *sock_map_seq_next(struct seq_file *seq, void *v, loff_t *pos)
754 	__must_hold(rcu)
755 {
756 	struct sock_map_seq_info *info = seq->private;
757 
758 	++*pos;
759 	++info->index;
760 
761 	return sock_map_seq_lookup_elem(info);
762 }
763 
764 static int sock_map_seq_show(struct seq_file *seq, void *v)
765 	__must_hold(rcu)
766 {
767 	struct sock_map_seq_info *info = seq->private;
768 	struct bpf_iter__sockmap ctx = {};
769 	struct bpf_iter_meta meta;
770 	struct bpf_prog *prog;
771 
772 	meta.seq = seq;
773 	prog = bpf_iter_get_info(&meta, !v);
774 	if (!prog)
775 		return 0;
776 
777 	ctx.meta = &meta;
778 	ctx.map = info->map;
779 	if (v) {
780 		ctx.key = &info->index;
781 		ctx.sk = info->sk;
782 	}
783 
784 	return bpf_iter_run_prog(prog, &ctx);
785 }
786 
787 static void sock_map_seq_stop(struct seq_file *seq, void *v)
788 	__releases(rcu)
789 {
790 	if (!v)
791 		(void)sock_map_seq_show(seq, NULL);
792 
793 	/* pairs with sock_map_seq_start */
794 	rcu_read_unlock();
795 }
796 
797 static const struct seq_operations sock_map_seq_ops = {
798 	.start	= sock_map_seq_start,
799 	.next	= sock_map_seq_next,
800 	.stop	= sock_map_seq_stop,
801 	.show	= sock_map_seq_show,
802 };
803 
804 static int sock_map_init_seq_private(void *priv_data,
805 				     struct bpf_iter_aux_info *aux)
806 {
807 	struct sock_map_seq_info *info = priv_data;
808 
809 	info->map = aux->map;
810 	return 0;
811 }
812 
813 static const struct bpf_iter_seq_info sock_map_iter_seq_info = {
814 	.seq_ops		= &sock_map_seq_ops,
815 	.init_seq_private	= sock_map_init_seq_private,
816 	.seq_priv_size		= sizeof(struct sock_map_seq_info),
817 };
818 
819 static int sock_map_btf_id;
820 const struct bpf_map_ops sock_map_ops = {
821 	.map_meta_equal		= bpf_map_meta_equal,
822 	.map_alloc		= sock_map_alloc,
823 	.map_free		= sock_map_free,
824 	.map_get_next_key	= sock_map_get_next_key,
825 	.map_lookup_elem_sys_only = sock_map_lookup_sys,
826 	.map_update_elem	= sock_map_update_elem,
827 	.map_delete_elem	= sock_map_delete_elem,
828 	.map_lookup_elem	= sock_map_lookup,
829 	.map_release_uref	= sock_map_release_progs,
830 	.map_check_btf		= map_check_no_btf,
831 	.map_btf_name		= "bpf_stab",
832 	.map_btf_id		= &sock_map_btf_id,
833 	.iter_seq_info		= &sock_map_iter_seq_info,
834 };
835 
836 struct bpf_shtab_elem {
837 	struct rcu_head rcu;
838 	u32 hash;
839 	struct sock *sk;
840 	struct hlist_node node;
841 	u8 key[];
842 };
843 
844 struct bpf_shtab_bucket {
845 	struct hlist_head head;
846 	raw_spinlock_t lock;
847 };
848 
849 struct bpf_shtab {
850 	struct bpf_map map;
851 	struct bpf_shtab_bucket *buckets;
852 	u32 buckets_num;
853 	u32 elem_size;
854 	struct sk_psock_progs progs;
855 	atomic_t count;
856 };
857 
858 static inline u32 sock_hash_bucket_hash(const void *key, u32 len)
859 {
860 	return jhash(key, len, 0);
861 }
862 
863 static struct bpf_shtab_bucket *sock_hash_select_bucket(struct bpf_shtab *htab,
864 							u32 hash)
865 {
866 	return &htab->buckets[hash & (htab->buckets_num - 1)];
867 }
868 
869 static struct bpf_shtab_elem *
870 sock_hash_lookup_elem_raw(struct hlist_head *head, u32 hash, void *key,
871 			  u32 key_size)
872 {
873 	struct bpf_shtab_elem *elem;
874 
875 	hlist_for_each_entry_rcu(elem, head, node) {
876 		if (elem->hash == hash &&
877 		    !memcmp(&elem->key, key, key_size))
878 			return elem;
879 	}
880 
881 	return NULL;
882 }
883 
884 static struct sock *__sock_hash_lookup_elem(struct bpf_map *map, void *key)
885 {
886 	struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
887 	u32 key_size = map->key_size, hash;
888 	struct bpf_shtab_bucket *bucket;
889 	struct bpf_shtab_elem *elem;
890 
891 	WARN_ON_ONCE(!rcu_read_lock_held());
892 
893 	hash = sock_hash_bucket_hash(key, key_size);
894 	bucket = sock_hash_select_bucket(htab, hash);
895 	elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size);
896 
897 	return elem ? elem->sk : NULL;
898 }
899 
900 static void sock_hash_free_elem(struct bpf_shtab *htab,
901 				struct bpf_shtab_elem *elem)
902 {
903 	atomic_dec(&htab->count);
904 	kfree_rcu(elem, rcu);
905 }
906 
907 static void sock_hash_delete_from_link(struct bpf_map *map, struct sock *sk,
908 				       void *link_raw)
909 {
910 	struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
911 	struct bpf_shtab_elem *elem_probe, *elem = link_raw;
912 	struct bpf_shtab_bucket *bucket;
913 
914 	WARN_ON_ONCE(!rcu_read_lock_held());
915 	bucket = sock_hash_select_bucket(htab, elem->hash);
916 
917 	/* elem may be deleted in parallel from the map, but access here
918 	 * is okay since it's going away only after RCU grace period.
919 	 * However, we need to check whether it's still present.
920 	 */
921 	raw_spin_lock_bh(&bucket->lock);
922 	elem_probe = sock_hash_lookup_elem_raw(&bucket->head, elem->hash,
923 					       elem->key, map->key_size);
924 	if (elem_probe && elem_probe == elem) {
925 		hlist_del_rcu(&elem->node);
926 		sock_map_unref(elem->sk, elem);
927 		sock_hash_free_elem(htab, elem);
928 	}
929 	raw_spin_unlock_bh(&bucket->lock);
930 }
931 
932 static int sock_hash_delete_elem(struct bpf_map *map, void *key)
933 {
934 	struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
935 	u32 hash, key_size = map->key_size;
936 	struct bpf_shtab_bucket *bucket;
937 	struct bpf_shtab_elem *elem;
938 	int ret = -ENOENT;
939 
940 	hash = sock_hash_bucket_hash(key, key_size);
941 	bucket = sock_hash_select_bucket(htab, hash);
942 
943 	raw_spin_lock_bh(&bucket->lock);
944 	elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size);
945 	if (elem) {
946 		hlist_del_rcu(&elem->node);
947 		sock_map_unref(elem->sk, elem);
948 		sock_hash_free_elem(htab, elem);
949 		ret = 0;
950 	}
951 	raw_spin_unlock_bh(&bucket->lock);
952 	return ret;
953 }
954 
955 static struct bpf_shtab_elem *sock_hash_alloc_elem(struct bpf_shtab *htab,
956 						   void *key, u32 key_size,
957 						   u32 hash, struct sock *sk,
958 						   struct bpf_shtab_elem *old)
959 {
960 	struct bpf_shtab_elem *new;
961 
962 	if (atomic_inc_return(&htab->count) > htab->map.max_entries) {
963 		if (!old) {
964 			atomic_dec(&htab->count);
965 			return ERR_PTR(-E2BIG);
966 		}
967 	}
968 
969 	new = bpf_map_kmalloc_node(&htab->map, htab->elem_size,
970 				   GFP_ATOMIC | __GFP_NOWARN,
971 				   htab->map.numa_node);
972 	if (!new) {
973 		atomic_dec(&htab->count);
974 		return ERR_PTR(-ENOMEM);
975 	}
976 	memcpy(new->key, key, key_size);
977 	new->sk = sk;
978 	new->hash = hash;
979 	return new;
980 }
981 
982 static int sock_hash_update_common(struct bpf_map *map, void *key,
983 				   struct sock *sk, u64 flags)
984 {
985 	struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
986 	u32 key_size = map->key_size, hash;
987 	struct bpf_shtab_elem *elem, *elem_new;
988 	struct bpf_shtab_bucket *bucket;
989 	struct sk_psock_link *link;
990 	struct sk_psock *psock;
991 	int ret;
992 
993 	WARN_ON_ONCE(!rcu_read_lock_held());
994 	if (unlikely(flags > BPF_EXIST))
995 		return -EINVAL;
996 
997 	link = sk_psock_init_link();
998 	if (!link)
999 		return -ENOMEM;
1000 
1001 	/* Only sockets we can redirect into/from in BPF need to hold
1002 	 * refs to parser/verdict progs and have their sk_data_ready
1003 	 * and sk_write_space callbacks overridden.
1004 	 */
1005 	if (sock_map_redirect_allowed(sk))
1006 		ret = sock_map_link(map, &htab->progs, sk);
1007 	else
1008 		ret = sock_map_link_no_progs(map, sk);
1009 	if (ret < 0)
1010 		goto out_free;
1011 
1012 	psock = sk_psock(sk);
1013 	WARN_ON_ONCE(!psock);
1014 
1015 	hash = sock_hash_bucket_hash(key, key_size);
1016 	bucket = sock_hash_select_bucket(htab, hash);
1017 
1018 	raw_spin_lock_bh(&bucket->lock);
1019 	elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size);
1020 	if (elem && flags == BPF_NOEXIST) {
1021 		ret = -EEXIST;
1022 		goto out_unlock;
1023 	} else if (!elem && flags == BPF_EXIST) {
1024 		ret = -ENOENT;
1025 		goto out_unlock;
1026 	}
1027 
1028 	elem_new = sock_hash_alloc_elem(htab, key, key_size, hash, sk, elem);
1029 	if (IS_ERR(elem_new)) {
1030 		ret = PTR_ERR(elem_new);
1031 		goto out_unlock;
1032 	}
1033 
1034 	sock_map_add_link(psock, link, map, elem_new);
1035 	/* Add new element to the head of the list, so that
1036 	 * concurrent search will find it before old elem.
1037 	 */
1038 	hlist_add_head_rcu(&elem_new->node, &bucket->head);
1039 	if (elem) {
1040 		hlist_del_rcu(&elem->node);
1041 		sock_map_unref(elem->sk, elem);
1042 		sock_hash_free_elem(htab, elem);
1043 	}
1044 	raw_spin_unlock_bh(&bucket->lock);
1045 	return 0;
1046 out_unlock:
1047 	raw_spin_unlock_bh(&bucket->lock);
1048 	sk_psock_put(sk, psock);
1049 out_free:
1050 	sk_psock_free_link(link);
1051 	return ret;
1052 }
1053 
1054 static int sock_hash_get_next_key(struct bpf_map *map, void *key,
1055 				  void *key_next)
1056 {
1057 	struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
1058 	struct bpf_shtab_elem *elem, *elem_next;
1059 	u32 hash, key_size = map->key_size;
1060 	struct hlist_head *head;
1061 	int i = 0;
1062 
1063 	if (!key)
1064 		goto find_first_elem;
1065 	hash = sock_hash_bucket_hash(key, key_size);
1066 	head = &sock_hash_select_bucket(htab, hash)->head;
1067 	elem = sock_hash_lookup_elem_raw(head, hash, key, key_size);
1068 	if (!elem)
1069 		goto find_first_elem;
1070 
1071 	elem_next = hlist_entry_safe(rcu_dereference(hlist_next_rcu(&elem->node)),
1072 				     struct bpf_shtab_elem, node);
1073 	if (elem_next) {
1074 		memcpy(key_next, elem_next->key, key_size);
1075 		return 0;
1076 	}
1077 
1078 	i = hash & (htab->buckets_num - 1);
1079 	i++;
1080 find_first_elem:
1081 	for (; i < htab->buckets_num; i++) {
1082 		head = &sock_hash_select_bucket(htab, i)->head;
1083 		elem_next = hlist_entry_safe(rcu_dereference(hlist_first_rcu(head)),
1084 					     struct bpf_shtab_elem, node);
1085 		if (elem_next) {
1086 			memcpy(key_next, elem_next->key, key_size);
1087 			return 0;
1088 		}
1089 	}
1090 
1091 	return -ENOENT;
1092 }
1093 
1094 static struct bpf_map *sock_hash_alloc(union bpf_attr *attr)
1095 {
1096 	struct bpf_shtab *htab;
1097 	int i, err;
1098 
1099 	if (!capable(CAP_NET_ADMIN))
1100 		return ERR_PTR(-EPERM);
1101 	if (attr->max_entries == 0 ||
1102 	    attr->key_size    == 0 ||
1103 	    (attr->value_size != sizeof(u32) &&
1104 	     attr->value_size != sizeof(u64)) ||
1105 	    attr->map_flags & ~SOCK_CREATE_FLAG_MASK)
1106 		return ERR_PTR(-EINVAL);
1107 	if (attr->key_size > MAX_BPF_STACK)
1108 		return ERR_PTR(-E2BIG);
1109 
1110 	htab = kzalloc(sizeof(*htab), GFP_USER | __GFP_ACCOUNT);
1111 	if (!htab)
1112 		return ERR_PTR(-ENOMEM);
1113 
1114 	bpf_map_init_from_attr(&htab->map, attr);
1115 
1116 	htab->buckets_num = roundup_pow_of_two(htab->map.max_entries);
1117 	htab->elem_size = sizeof(struct bpf_shtab_elem) +
1118 			  round_up(htab->map.key_size, 8);
1119 	if (htab->buckets_num == 0 ||
1120 	    htab->buckets_num > U32_MAX / sizeof(struct bpf_shtab_bucket)) {
1121 		err = -EINVAL;
1122 		goto free_htab;
1123 	}
1124 
1125 	htab->buckets = bpf_map_area_alloc(htab->buckets_num *
1126 					   sizeof(struct bpf_shtab_bucket),
1127 					   htab->map.numa_node);
1128 	if (!htab->buckets) {
1129 		err = -ENOMEM;
1130 		goto free_htab;
1131 	}
1132 
1133 	for (i = 0; i < htab->buckets_num; i++) {
1134 		INIT_HLIST_HEAD(&htab->buckets[i].head);
1135 		raw_spin_lock_init(&htab->buckets[i].lock);
1136 	}
1137 
1138 	return &htab->map;
1139 free_htab:
1140 	kfree(htab);
1141 	return ERR_PTR(err);
1142 }
1143 
1144 static void sock_hash_free(struct bpf_map *map)
1145 {
1146 	struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
1147 	struct bpf_shtab_bucket *bucket;
1148 	struct hlist_head unlink_list;
1149 	struct bpf_shtab_elem *elem;
1150 	struct hlist_node *node;
1151 	int i;
1152 
1153 	/* After the sync no updates or deletes will be in-flight so it
1154 	 * is safe to walk map and remove entries without risking a race
1155 	 * in EEXIST update case.
1156 	 */
1157 	synchronize_rcu();
1158 	for (i = 0; i < htab->buckets_num; i++) {
1159 		bucket = sock_hash_select_bucket(htab, i);
1160 
1161 		/* We are racing with sock_hash_delete_from_link to
1162 		 * enter the spin-lock critical section. Every socket on
1163 		 * the list is still linked to sockhash. Since link
1164 		 * exists, psock exists and holds a ref to socket. That
1165 		 * lets us to grab a socket ref too.
1166 		 */
1167 		raw_spin_lock_bh(&bucket->lock);
1168 		hlist_for_each_entry(elem, &bucket->head, node)
1169 			sock_hold(elem->sk);
1170 		hlist_move_list(&bucket->head, &unlink_list);
1171 		raw_spin_unlock_bh(&bucket->lock);
1172 
1173 		/* Process removed entries out of atomic context to
1174 		 * block for socket lock before deleting the psock's
1175 		 * link to sockhash.
1176 		 */
1177 		hlist_for_each_entry_safe(elem, node, &unlink_list, node) {
1178 			hlist_del(&elem->node);
1179 			lock_sock(elem->sk);
1180 			rcu_read_lock();
1181 			sock_map_unref(elem->sk, elem);
1182 			rcu_read_unlock();
1183 			release_sock(elem->sk);
1184 			sock_put(elem->sk);
1185 			sock_hash_free_elem(htab, elem);
1186 		}
1187 	}
1188 
1189 	/* wait for psock readers accessing its map link */
1190 	synchronize_rcu();
1191 
1192 	bpf_map_area_free(htab->buckets);
1193 	kfree(htab);
1194 }
1195 
1196 static void *sock_hash_lookup_sys(struct bpf_map *map, void *key)
1197 {
1198 	struct sock *sk;
1199 
1200 	if (map->value_size != sizeof(u64))
1201 		return ERR_PTR(-ENOSPC);
1202 
1203 	sk = __sock_hash_lookup_elem(map, key);
1204 	if (!sk)
1205 		return ERR_PTR(-ENOENT);
1206 
1207 	__sock_gen_cookie(sk);
1208 	return &sk->sk_cookie;
1209 }
1210 
1211 static void *sock_hash_lookup(struct bpf_map *map, void *key)
1212 {
1213 	struct sock *sk;
1214 
1215 	sk = __sock_hash_lookup_elem(map, key);
1216 	if (!sk)
1217 		return NULL;
1218 	if (sk_is_refcounted(sk) && !refcount_inc_not_zero(&sk->sk_refcnt))
1219 		return NULL;
1220 	return sk;
1221 }
1222 
1223 static void sock_hash_release_progs(struct bpf_map *map)
1224 {
1225 	psock_progs_drop(&container_of(map, struct bpf_shtab, map)->progs);
1226 }
1227 
1228 BPF_CALL_4(bpf_sock_hash_update, struct bpf_sock_ops_kern *, sops,
1229 	   struct bpf_map *, map, void *, key, u64, flags)
1230 {
1231 	WARN_ON_ONCE(!rcu_read_lock_held());
1232 
1233 	if (likely(sock_map_sk_is_suitable(sops->sk) &&
1234 		   sock_map_op_okay(sops)))
1235 		return sock_hash_update_common(map, key, sops->sk, flags);
1236 	return -EOPNOTSUPP;
1237 }
1238 
1239 const struct bpf_func_proto bpf_sock_hash_update_proto = {
1240 	.func		= bpf_sock_hash_update,
1241 	.gpl_only	= false,
1242 	.pkt_access	= true,
1243 	.ret_type	= RET_INTEGER,
1244 	.arg1_type	= ARG_PTR_TO_CTX,
1245 	.arg2_type	= ARG_CONST_MAP_PTR,
1246 	.arg3_type	= ARG_PTR_TO_MAP_KEY,
1247 	.arg4_type	= ARG_ANYTHING,
1248 };
1249 
1250 BPF_CALL_4(bpf_sk_redirect_hash, struct sk_buff *, skb,
1251 	   struct bpf_map *, map, void *, key, u64, flags)
1252 {
1253 	struct tcp_skb_cb *tcb = TCP_SKB_CB(skb);
1254 	struct sock *sk;
1255 
1256 	if (unlikely(flags & ~(BPF_F_INGRESS)))
1257 		return SK_DROP;
1258 
1259 	sk = __sock_hash_lookup_elem(map, key);
1260 	if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
1261 		return SK_DROP;
1262 
1263 	tcb->bpf.flags = flags;
1264 	tcb->bpf.sk_redir = sk;
1265 	return SK_PASS;
1266 }
1267 
1268 const struct bpf_func_proto bpf_sk_redirect_hash_proto = {
1269 	.func           = bpf_sk_redirect_hash,
1270 	.gpl_only       = false,
1271 	.ret_type       = RET_INTEGER,
1272 	.arg1_type	= ARG_PTR_TO_CTX,
1273 	.arg2_type      = ARG_CONST_MAP_PTR,
1274 	.arg3_type      = ARG_PTR_TO_MAP_KEY,
1275 	.arg4_type      = ARG_ANYTHING,
1276 };
1277 
1278 BPF_CALL_4(bpf_msg_redirect_hash, struct sk_msg *, msg,
1279 	   struct bpf_map *, map, void *, key, u64, flags)
1280 {
1281 	struct sock *sk;
1282 
1283 	if (unlikely(flags & ~(BPF_F_INGRESS)))
1284 		return SK_DROP;
1285 
1286 	sk = __sock_hash_lookup_elem(map, key);
1287 	if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
1288 		return SK_DROP;
1289 
1290 	msg->flags = flags;
1291 	msg->sk_redir = sk;
1292 	return SK_PASS;
1293 }
1294 
1295 const struct bpf_func_proto bpf_msg_redirect_hash_proto = {
1296 	.func           = bpf_msg_redirect_hash,
1297 	.gpl_only       = false,
1298 	.ret_type       = RET_INTEGER,
1299 	.arg1_type	= ARG_PTR_TO_CTX,
1300 	.arg2_type      = ARG_CONST_MAP_PTR,
1301 	.arg3_type      = ARG_PTR_TO_MAP_KEY,
1302 	.arg4_type      = ARG_ANYTHING,
1303 };
1304 
1305 struct sock_hash_seq_info {
1306 	struct bpf_map *map;
1307 	struct bpf_shtab *htab;
1308 	u32 bucket_id;
1309 };
1310 
1311 static void *sock_hash_seq_find_next(struct sock_hash_seq_info *info,
1312 				     struct bpf_shtab_elem *prev_elem)
1313 {
1314 	const struct bpf_shtab *htab = info->htab;
1315 	struct bpf_shtab_bucket *bucket;
1316 	struct bpf_shtab_elem *elem;
1317 	struct hlist_node *node;
1318 
1319 	/* try to find next elem in the same bucket */
1320 	if (prev_elem) {
1321 		node = rcu_dereference(hlist_next_rcu(&prev_elem->node));
1322 		elem = hlist_entry_safe(node, struct bpf_shtab_elem, node);
1323 		if (elem)
1324 			return elem;
1325 
1326 		/* no more elements, continue in the next bucket */
1327 		info->bucket_id++;
1328 	}
1329 
1330 	for (; info->bucket_id < htab->buckets_num; info->bucket_id++) {
1331 		bucket = &htab->buckets[info->bucket_id];
1332 		node = rcu_dereference(hlist_first_rcu(&bucket->head));
1333 		elem = hlist_entry_safe(node, struct bpf_shtab_elem, node);
1334 		if (elem)
1335 			return elem;
1336 	}
1337 
1338 	return NULL;
1339 }
1340 
1341 static void *sock_hash_seq_start(struct seq_file *seq, loff_t *pos)
1342 	__acquires(rcu)
1343 {
1344 	struct sock_hash_seq_info *info = seq->private;
1345 
1346 	if (*pos == 0)
1347 		++*pos;
1348 
1349 	/* pairs with sock_hash_seq_stop */
1350 	rcu_read_lock();
1351 	return sock_hash_seq_find_next(info, NULL);
1352 }
1353 
1354 static void *sock_hash_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1355 	__must_hold(rcu)
1356 {
1357 	struct sock_hash_seq_info *info = seq->private;
1358 
1359 	++*pos;
1360 	return sock_hash_seq_find_next(info, v);
1361 }
1362 
1363 static int sock_hash_seq_show(struct seq_file *seq, void *v)
1364 	__must_hold(rcu)
1365 {
1366 	struct sock_hash_seq_info *info = seq->private;
1367 	struct bpf_iter__sockmap ctx = {};
1368 	struct bpf_shtab_elem *elem = v;
1369 	struct bpf_iter_meta meta;
1370 	struct bpf_prog *prog;
1371 
1372 	meta.seq = seq;
1373 	prog = bpf_iter_get_info(&meta, !elem);
1374 	if (!prog)
1375 		return 0;
1376 
1377 	ctx.meta = &meta;
1378 	ctx.map = info->map;
1379 	if (elem) {
1380 		ctx.key = elem->key;
1381 		ctx.sk = elem->sk;
1382 	}
1383 
1384 	return bpf_iter_run_prog(prog, &ctx);
1385 }
1386 
1387 static void sock_hash_seq_stop(struct seq_file *seq, void *v)
1388 	__releases(rcu)
1389 {
1390 	if (!v)
1391 		(void)sock_hash_seq_show(seq, NULL);
1392 
1393 	/* pairs with sock_hash_seq_start */
1394 	rcu_read_unlock();
1395 }
1396 
1397 static const struct seq_operations sock_hash_seq_ops = {
1398 	.start	= sock_hash_seq_start,
1399 	.next	= sock_hash_seq_next,
1400 	.stop	= sock_hash_seq_stop,
1401 	.show	= sock_hash_seq_show,
1402 };
1403 
1404 static int sock_hash_init_seq_private(void *priv_data,
1405 				     struct bpf_iter_aux_info *aux)
1406 {
1407 	struct sock_hash_seq_info *info = priv_data;
1408 
1409 	info->map = aux->map;
1410 	info->htab = container_of(aux->map, struct bpf_shtab, map);
1411 	return 0;
1412 }
1413 
1414 static const struct bpf_iter_seq_info sock_hash_iter_seq_info = {
1415 	.seq_ops		= &sock_hash_seq_ops,
1416 	.init_seq_private	= sock_hash_init_seq_private,
1417 	.seq_priv_size		= sizeof(struct sock_hash_seq_info),
1418 };
1419 
1420 static int sock_hash_map_btf_id;
1421 const struct bpf_map_ops sock_hash_ops = {
1422 	.map_meta_equal		= bpf_map_meta_equal,
1423 	.map_alloc		= sock_hash_alloc,
1424 	.map_free		= sock_hash_free,
1425 	.map_get_next_key	= sock_hash_get_next_key,
1426 	.map_update_elem	= sock_map_update_elem,
1427 	.map_delete_elem	= sock_hash_delete_elem,
1428 	.map_lookup_elem	= sock_hash_lookup,
1429 	.map_lookup_elem_sys_only = sock_hash_lookup_sys,
1430 	.map_release_uref	= sock_hash_release_progs,
1431 	.map_check_btf		= map_check_no_btf,
1432 	.map_btf_name		= "bpf_shtab",
1433 	.map_btf_id		= &sock_hash_map_btf_id,
1434 	.iter_seq_info		= &sock_hash_iter_seq_info,
1435 };
1436 
1437 static struct sk_psock_progs *sock_map_progs(struct bpf_map *map)
1438 {
1439 	switch (map->map_type) {
1440 	case BPF_MAP_TYPE_SOCKMAP:
1441 		return &container_of(map, struct bpf_stab, map)->progs;
1442 	case BPF_MAP_TYPE_SOCKHASH:
1443 		return &container_of(map, struct bpf_shtab, map)->progs;
1444 	default:
1445 		break;
1446 	}
1447 
1448 	return NULL;
1449 }
1450 
1451 int sock_map_prog_update(struct bpf_map *map, struct bpf_prog *prog,
1452 			 struct bpf_prog *old, u32 which)
1453 {
1454 	struct sk_psock_progs *progs = sock_map_progs(map);
1455 	struct bpf_prog **pprog;
1456 
1457 	if (!progs)
1458 		return -EOPNOTSUPP;
1459 
1460 	switch (which) {
1461 	case BPF_SK_MSG_VERDICT:
1462 		pprog = &progs->msg_parser;
1463 		break;
1464 	case BPF_SK_SKB_STREAM_PARSER:
1465 		pprog = &progs->skb_parser;
1466 		break;
1467 	case BPF_SK_SKB_STREAM_VERDICT:
1468 		pprog = &progs->skb_verdict;
1469 		break;
1470 	default:
1471 		return -EOPNOTSUPP;
1472 	}
1473 
1474 	if (old)
1475 		return psock_replace_prog(pprog, prog, old);
1476 
1477 	psock_set_prog(pprog, prog);
1478 	return 0;
1479 }
1480 
1481 static void sock_map_unlink(struct sock *sk, struct sk_psock_link *link)
1482 {
1483 	switch (link->map->map_type) {
1484 	case BPF_MAP_TYPE_SOCKMAP:
1485 		return sock_map_delete_from_link(link->map, sk,
1486 						 link->link_raw);
1487 	case BPF_MAP_TYPE_SOCKHASH:
1488 		return sock_hash_delete_from_link(link->map, sk,
1489 						  link->link_raw);
1490 	default:
1491 		break;
1492 	}
1493 }
1494 
1495 static void sock_map_remove_links(struct sock *sk, struct sk_psock *psock)
1496 {
1497 	struct sk_psock_link *link;
1498 
1499 	while ((link = sk_psock_link_pop(psock))) {
1500 		sock_map_unlink(sk, link);
1501 		sk_psock_free_link(link);
1502 	}
1503 }
1504 
1505 void sock_map_unhash(struct sock *sk)
1506 {
1507 	void (*saved_unhash)(struct sock *sk);
1508 	struct sk_psock *psock;
1509 
1510 	rcu_read_lock();
1511 	psock = sk_psock(sk);
1512 	if (unlikely(!psock)) {
1513 		rcu_read_unlock();
1514 		if (sk->sk_prot->unhash)
1515 			sk->sk_prot->unhash(sk);
1516 		return;
1517 	}
1518 
1519 	saved_unhash = psock->saved_unhash;
1520 	sock_map_remove_links(sk, psock);
1521 	rcu_read_unlock();
1522 	saved_unhash(sk);
1523 }
1524 
1525 void sock_map_close(struct sock *sk, long timeout)
1526 {
1527 	void (*saved_close)(struct sock *sk, long timeout);
1528 	struct sk_psock *psock;
1529 
1530 	lock_sock(sk);
1531 	rcu_read_lock();
1532 	psock = sk_psock(sk);
1533 	if (unlikely(!psock)) {
1534 		rcu_read_unlock();
1535 		release_sock(sk);
1536 		return sk->sk_prot->close(sk, timeout);
1537 	}
1538 
1539 	saved_close = psock->saved_close;
1540 	sock_map_remove_links(sk, psock);
1541 	rcu_read_unlock();
1542 	release_sock(sk);
1543 	saved_close(sk, timeout);
1544 }
1545 
1546 static int sock_map_iter_attach_target(struct bpf_prog *prog,
1547 				       union bpf_iter_link_info *linfo,
1548 				       struct bpf_iter_aux_info *aux)
1549 {
1550 	struct bpf_map *map;
1551 	int err = -EINVAL;
1552 
1553 	if (!linfo->map.map_fd)
1554 		return -EBADF;
1555 
1556 	map = bpf_map_get_with_uref(linfo->map.map_fd);
1557 	if (IS_ERR(map))
1558 		return PTR_ERR(map);
1559 
1560 	if (map->map_type != BPF_MAP_TYPE_SOCKMAP &&
1561 	    map->map_type != BPF_MAP_TYPE_SOCKHASH)
1562 		goto put_map;
1563 
1564 	if (prog->aux->max_rdonly_access > map->key_size) {
1565 		err = -EACCES;
1566 		goto put_map;
1567 	}
1568 
1569 	aux->map = map;
1570 	return 0;
1571 
1572 put_map:
1573 	bpf_map_put_with_uref(map);
1574 	return err;
1575 }
1576 
1577 static void sock_map_iter_detach_target(struct bpf_iter_aux_info *aux)
1578 {
1579 	bpf_map_put_with_uref(aux->map);
1580 }
1581 
1582 static struct bpf_iter_reg sock_map_iter_reg = {
1583 	.target			= "sockmap",
1584 	.attach_target		= sock_map_iter_attach_target,
1585 	.detach_target		= sock_map_iter_detach_target,
1586 	.show_fdinfo		= bpf_iter_map_show_fdinfo,
1587 	.fill_link_info		= bpf_iter_map_fill_link_info,
1588 	.ctx_arg_info_size	= 2,
1589 	.ctx_arg_info		= {
1590 		{ offsetof(struct bpf_iter__sockmap, key),
1591 		  PTR_TO_RDONLY_BUF_OR_NULL },
1592 		{ offsetof(struct bpf_iter__sockmap, sk),
1593 		  PTR_TO_BTF_ID_OR_NULL },
1594 	},
1595 };
1596 
1597 static int __init bpf_sockmap_iter_init(void)
1598 {
1599 	sock_map_iter_reg.ctx_arg_info[1].btf_id =
1600 		btf_sock_ids[BTF_SOCK_TYPE_SOCK];
1601 	return bpf_iter_reg_target(&sock_map_iter_reg);
1602 }
1603 late_initcall(bpf_sockmap_iter_init);
1604