xref: /linux/net/core/sock_map.c (revision b58b13f156c00c2457035b7071eaaac105fe6836)
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 	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 int sock_map_prog_update(struct bpf_map *map, struct bpf_prog *prog,
28 				struct bpf_prog *old, u32 which);
29 static struct sk_psock_progs *sock_map_progs(struct bpf_map *map);
30 
31 static struct bpf_map *sock_map_alloc(union bpf_attr *attr)
32 {
33 	struct bpf_stab *stab;
34 
35 	if (attr->max_entries == 0 ||
36 	    attr->key_size    != 4 ||
37 	    (attr->value_size != sizeof(u32) &&
38 	     attr->value_size != sizeof(u64)) ||
39 	    attr->map_flags & ~SOCK_CREATE_FLAG_MASK)
40 		return ERR_PTR(-EINVAL);
41 
42 	stab = bpf_map_area_alloc(sizeof(*stab), NUMA_NO_NODE);
43 	if (!stab)
44 		return ERR_PTR(-ENOMEM);
45 
46 	bpf_map_init_from_attr(&stab->map, attr);
47 	spin_lock_init(&stab->lock);
48 
49 	stab->sks = bpf_map_area_alloc((u64) stab->map.max_entries *
50 				       sizeof(struct sock *),
51 				       stab->map.numa_node);
52 	if (!stab->sks) {
53 		bpf_map_area_free(stab);
54 		return ERR_PTR(-ENOMEM);
55 	}
56 
57 	return &stab->map;
58 }
59 
60 int sock_map_get_from_fd(const union bpf_attr *attr, struct bpf_prog *prog)
61 {
62 	u32 ufd = attr->target_fd;
63 	struct bpf_map *map;
64 	struct fd f;
65 	int ret;
66 
67 	if (attr->attach_flags || attr->replace_bpf_fd)
68 		return -EINVAL;
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, NULL, attr->attach_type);
75 	fdput(f);
76 	return ret;
77 }
78 
79 int sock_map_prog_detach(const union bpf_attr *attr, enum bpf_prog_type ptype)
80 {
81 	u32 ufd = attr->target_fd;
82 	struct bpf_prog *prog;
83 	struct bpf_map *map;
84 	struct fd f;
85 	int ret;
86 
87 	if (attr->attach_flags || attr->replace_bpf_fd)
88 		return -EINVAL;
89 
90 	f = fdget(ufd);
91 	map = __bpf_map_get(f);
92 	if (IS_ERR(map))
93 		return PTR_ERR(map);
94 
95 	prog = bpf_prog_get(attr->attach_bpf_fd);
96 	if (IS_ERR(prog)) {
97 		ret = PTR_ERR(prog);
98 		goto put_map;
99 	}
100 
101 	if (prog->type != ptype) {
102 		ret = -EINVAL;
103 		goto put_prog;
104 	}
105 
106 	ret = sock_map_prog_update(map, NULL, prog, attr->attach_type);
107 put_prog:
108 	bpf_prog_put(prog);
109 put_map:
110 	fdput(f);
111 	return ret;
112 }
113 
114 static void sock_map_sk_acquire(struct sock *sk)
115 	__acquires(&sk->sk_lock.slock)
116 {
117 	lock_sock(sk);
118 	rcu_read_lock();
119 }
120 
121 static void sock_map_sk_release(struct sock *sk)
122 	__releases(&sk->sk_lock.slock)
123 {
124 	rcu_read_unlock();
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 sk_psock_progs *progs = sock_map_progs(map);
150 
151 			if (psock->saved_data_ready && progs->stream_parser)
152 				strp_stop = true;
153 			if (psock->saved_data_ready && progs->stream_verdict)
154 				verdict_stop = true;
155 			if (psock->saved_data_ready && progs->skb_verdict)
156 				verdict_stop = true;
157 			list_del(&link->list);
158 			sk_psock_free_link(link);
159 		}
160 	}
161 	spin_unlock_bh(&psock->link_lock);
162 	if (strp_stop || verdict_stop) {
163 		write_lock_bh(&sk->sk_callback_lock);
164 		if (strp_stop)
165 			sk_psock_stop_strp(sk, psock);
166 		if (verdict_stop)
167 			sk_psock_stop_verdict(sk, psock);
168 
169 		if (psock->psock_update_sk_prot)
170 			psock->psock_update_sk_prot(sk, psock, false);
171 		write_unlock_bh(&sk->sk_callback_lock);
172 	}
173 }
174 
175 static void sock_map_unref(struct sock *sk, void *link_raw)
176 {
177 	struct sk_psock *psock = sk_psock(sk);
178 
179 	if (likely(psock)) {
180 		sock_map_del_link(sk, psock, link_raw);
181 		sk_psock_put(sk, psock);
182 	}
183 }
184 
185 static int sock_map_init_proto(struct sock *sk, struct sk_psock *psock)
186 {
187 	if (!sk->sk_prot->psock_update_sk_prot)
188 		return -EINVAL;
189 	psock->psock_update_sk_prot = sk->sk_prot->psock_update_sk_prot;
190 	return sk->sk_prot->psock_update_sk_prot(sk, psock, false);
191 }
192 
193 static struct sk_psock *sock_map_psock_get_checked(struct sock *sk)
194 {
195 	struct sk_psock *psock;
196 
197 	rcu_read_lock();
198 	psock = sk_psock(sk);
199 	if (psock) {
200 		if (sk->sk_prot->close != sock_map_close) {
201 			psock = ERR_PTR(-EBUSY);
202 			goto out;
203 		}
204 
205 		if (!refcount_inc_not_zero(&psock->refcnt))
206 			psock = ERR_PTR(-EBUSY);
207 	}
208 out:
209 	rcu_read_unlock();
210 	return psock;
211 }
212 
213 static int sock_map_link(struct bpf_map *map, struct sock *sk)
214 {
215 	struct sk_psock_progs *progs = sock_map_progs(map);
216 	struct bpf_prog *stream_verdict = NULL;
217 	struct bpf_prog *stream_parser = NULL;
218 	struct bpf_prog *skb_verdict = NULL;
219 	struct bpf_prog *msg_parser = NULL;
220 	struct sk_psock *psock;
221 	int ret;
222 
223 	stream_verdict = READ_ONCE(progs->stream_verdict);
224 	if (stream_verdict) {
225 		stream_verdict = bpf_prog_inc_not_zero(stream_verdict);
226 		if (IS_ERR(stream_verdict))
227 			return PTR_ERR(stream_verdict);
228 	}
229 
230 	stream_parser = READ_ONCE(progs->stream_parser);
231 	if (stream_parser) {
232 		stream_parser = bpf_prog_inc_not_zero(stream_parser);
233 		if (IS_ERR(stream_parser)) {
234 			ret = PTR_ERR(stream_parser);
235 			goto out_put_stream_verdict;
236 		}
237 	}
238 
239 	msg_parser = READ_ONCE(progs->msg_parser);
240 	if (msg_parser) {
241 		msg_parser = bpf_prog_inc_not_zero(msg_parser);
242 		if (IS_ERR(msg_parser)) {
243 			ret = PTR_ERR(msg_parser);
244 			goto out_put_stream_parser;
245 		}
246 	}
247 
248 	skb_verdict = READ_ONCE(progs->skb_verdict);
249 	if (skb_verdict) {
250 		skb_verdict = bpf_prog_inc_not_zero(skb_verdict);
251 		if (IS_ERR(skb_verdict)) {
252 			ret = PTR_ERR(skb_verdict);
253 			goto out_put_msg_parser;
254 		}
255 	}
256 
257 	psock = sock_map_psock_get_checked(sk);
258 	if (IS_ERR(psock)) {
259 		ret = PTR_ERR(psock);
260 		goto out_progs;
261 	}
262 
263 	if (psock) {
264 		if ((msg_parser && READ_ONCE(psock->progs.msg_parser)) ||
265 		    (stream_parser  && READ_ONCE(psock->progs.stream_parser)) ||
266 		    (skb_verdict && READ_ONCE(psock->progs.skb_verdict)) ||
267 		    (skb_verdict && READ_ONCE(psock->progs.stream_verdict)) ||
268 		    (stream_verdict && READ_ONCE(psock->progs.skb_verdict)) ||
269 		    (stream_verdict && READ_ONCE(psock->progs.stream_verdict))) {
270 			sk_psock_put(sk, psock);
271 			ret = -EBUSY;
272 			goto out_progs;
273 		}
274 	} else {
275 		psock = sk_psock_init(sk, map->numa_node);
276 		if (IS_ERR(psock)) {
277 			ret = PTR_ERR(psock);
278 			goto out_progs;
279 		}
280 	}
281 
282 	if (msg_parser)
283 		psock_set_prog(&psock->progs.msg_parser, msg_parser);
284 	if (stream_parser)
285 		psock_set_prog(&psock->progs.stream_parser, stream_parser);
286 	if (stream_verdict)
287 		psock_set_prog(&psock->progs.stream_verdict, stream_verdict);
288 	if (skb_verdict)
289 		psock_set_prog(&psock->progs.skb_verdict, skb_verdict);
290 
291 	/* msg_* and stream_* programs references tracked in psock after this
292 	 * point. Reference dec and cleanup will occur through psock destructor
293 	 */
294 	ret = sock_map_init_proto(sk, psock);
295 	if (ret < 0) {
296 		sk_psock_put(sk, psock);
297 		goto out;
298 	}
299 
300 	write_lock_bh(&sk->sk_callback_lock);
301 	if (stream_parser && stream_verdict && !psock->saved_data_ready) {
302 		ret = sk_psock_init_strp(sk, psock);
303 		if (ret) {
304 			write_unlock_bh(&sk->sk_callback_lock);
305 			sk_psock_put(sk, psock);
306 			goto out;
307 		}
308 		sk_psock_start_strp(sk, psock);
309 	} else if (!stream_parser && stream_verdict && !psock->saved_data_ready) {
310 		sk_psock_start_verdict(sk,psock);
311 	} else if (!stream_verdict && skb_verdict && !psock->saved_data_ready) {
312 		sk_psock_start_verdict(sk, psock);
313 	}
314 	write_unlock_bh(&sk->sk_callback_lock);
315 	return 0;
316 out_progs:
317 	if (skb_verdict)
318 		bpf_prog_put(skb_verdict);
319 out_put_msg_parser:
320 	if (msg_parser)
321 		bpf_prog_put(msg_parser);
322 out_put_stream_parser:
323 	if (stream_parser)
324 		bpf_prog_put(stream_parser);
325 out_put_stream_verdict:
326 	if (stream_verdict)
327 		bpf_prog_put(stream_verdict);
328 out:
329 	return ret;
330 }
331 
332 static void sock_map_free(struct bpf_map *map)
333 {
334 	struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
335 	int i;
336 
337 	/* After the sync no updates or deletes will be in-flight so it
338 	 * is safe to walk map and remove entries without risking a race
339 	 * in EEXIST update case.
340 	 */
341 	synchronize_rcu();
342 	for (i = 0; i < stab->map.max_entries; i++) {
343 		struct sock **psk = &stab->sks[i];
344 		struct sock *sk;
345 
346 		sk = xchg(psk, NULL);
347 		if (sk) {
348 			sock_hold(sk);
349 			lock_sock(sk);
350 			rcu_read_lock();
351 			sock_map_unref(sk, psk);
352 			rcu_read_unlock();
353 			release_sock(sk);
354 			sock_put(sk);
355 		}
356 	}
357 
358 	/* wait for psock readers accessing its map link */
359 	synchronize_rcu();
360 
361 	bpf_map_area_free(stab->sks);
362 	bpf_map_area_free(stab);
363 }
364 
365 static void sock_map_release_progs(struct bpf_map *map)
366 {
367 	psock_progs_drop(&container_of(map, struct bpf_stab, map)->progs);
368 }
369 
370 static struct sock *__sock_map_lookup_elem(struct bpf_map *map, u32 key)
371 {
372 	struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
373 
374 	WARN_ON_ONCE(!rcu_read_lock_held());
375 
376 	if (unlikely(key >= map->max_entries))
377 		return NULL;
378 	return READ_ONCE(stab->sks[key]);
379 }
380 
381 static void *sock_map_lookup(struct bpf_map *map, void *key)
382 {
383 	struct sock *sk;
384 
385 	sk = __sock_map_lookup_elem(map, *(u32 *)key);
386 	if (!sk)
387 		return NULL;
388 	if (sk_is_refcounted(sk) && !refcount_inc_not_zero(&sk->sk_refcnt))
389 		return NULL;
390 	return sk;
391 }
392 
393 static void *sock_map_lookup_sys(struct bpf_map *map, void *key)
394 {
395 	struct sock *sk;
396 
397 	if (map->value_size != sizeof(u64))
398 		return ERR_PTR(-ENOSPC);
399 
400 	sk = __sock_map_lookup_elem(map, *(u32 *)key);
401 	if (!sk)
402 		return ERR_PTR(-ENOENT);
403 
404 	__sock_gen_cookie(sk);
405 	return &sk->sk_cookie;
406 }
407 
408 static int __sock_map_delete(struct bpf_stab *stab, struct sock *sk_test,
409 			     struct sock **psk)
410 {
411 	struct sock *sk;
412 	int err = 0;
413 
414 	spin_lock_bh(&stab->lock);
415 	sk = *psk;
416 	if (!sk_test || sk_test == sk)
417 		sk = xchg(psk, NULL);
418 
419 	if (likely(sk))
420 		sock_map_unref(sk, psk);
421 	else
422 		err = -EINVAL;
423 
424 	spin_unlock_bh(&stab->lock);
425 	return err;
426 }
427 
428 static void sock_map_delete_from_link(struct bpf_map *map, struct sock *sk,
429 				      void *link_raw)
430 {
431 	struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
432 
433 	__sock_map_delete(stab, sk, link_raw);
434 }
435 
436 static long sock_map_delete_elem(struct bpf_map *map, void *key)
437 {
438 	struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
439 	u32 i = *(u32 *)key;
440 	struct sock **psk;
441 
442 	if (unlikely(i >= map->max_entries))
443 		return -EINVAL;
444 
445 	psk = &stab->sks[i];
446 	return __sock_map_delete(stab, NULL, psk);
447 }
448 
449 static int sock_map_get_next_key(struct bpf_map *map, void *key, void *next)
450 {
451 	struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
452 	u32 i = key ? *(u32 *)key : U32_MAX;
453 	u32 *key_next = next;
454 
455 	if (i == stab->map.max_entries - 1)
456 		return -ENOENT;
457 	if (i >= stab->map.max_entries)
458 		*key_next = 0;
459 	else
460 		*key_next = i + 1;
461 	return 0;
462 }
463 
464 static int sock_map_update_common(struct bpf_map *map, u32 idx,
465 				  struct sock *sk, u64 flags)
466 {
467 	struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
468 	struct sk_psock_link *link;
469 	struct sk_psock *psock;
470 	struct sock *osk;
471 	int ret;
472 
473 	WARN_ON_ONCE(!rcu_read_lock_held());
474 	if (unlikely(flags > BPF_EXIST))
475 		return -EINVAL;
476 	if (unlikely(idx >= map->max_entries))
477 		return -E2BIG;
478 
479 	link = sk_psock_init_link();
480 	if (!link)
481 		return -ENOMEM;
482 
483 	ret = sock_map_link(map, sk);
484 	if (ret < 0)
485 		goto out_free;
486 
487 	psock = sk_psock(sk);
488 	WARN_ON_ONCE(!psock);
489 
490 	spin_lock_bh(&stab->lock);
491 	osk = stab->sks[idx];
492 	if (osk && flags == BPF_NOEXIST) {
493 		ret = -EEXIST;
494 		goto out_unlock;
495 	} else if (!osk && flags == BPF_EXIST) {
496 		ret = -ENOENT;
497 		goto out_unlock;
498 	}
499 
500 	sock_map_add_link(psock, link, map, &stab->sks[idx]);
501 	stab->sks[idx] = sk;
502 	if (osk)
503 		sock_map_unref(osk, &stab->sks[idx]);
504 	spin_unlock_bh(&stab->lock);
505 	return 0;
506 out_unlock:
507 	spin_unlock_bh(&stab->lock);
508 	if (psock)
509 		sk_psock_put(sk, psock);
510 out_free:
511 	sk_psock_free_link(link);
512 	return ret;
513 }
514 
515 static bool sock_map_op_okay(const struct bpf_sock_ops_kern *ops)
516 {
517 	return ops->op == BPF_SOCK_OPS_PASSIVE_ESTABLISHED_CB ||
518 	       ops->op == BPF_SOCK_OPS_ACTIVE_ESTABLISHED_CB ||
519 	       ops->op == BPF_SOCK_OPS_TCP_LISTEN_CB;
520 }
521 
522 static bool sock_map_redirect_allowed(const struct sock *sk)
523 {
524 	if (sk_is_tcp(sk))
525 		return sk->sk_state != TCP_LISTEN;
526 	else
527 		return sk->sk_state == TCP_ESTABLISHED;
528 }
529 
530 static bool sock_map_sk_is_suitable(const struct sock *sk)
531 {
532 	return !!sk->sk_prot->psock_update_sk_prot;
533 }
534 
535 static bool sock_map_sk_state_allowed(const struct sock *sk)
536 {
537 	if (sk_is_tcp(sk))
538 		return (1 << sk->sk_state) & (TCPF_ESTABLISHED | TCPF_LISTEN);
539 	if (sk_is_stream_unix(sk))
540 		return (1 << sk->sk_state) & TCPF_ESTABLISHED;
541 	return true;
542 }
543 
544 static int sock_hash_update_common(struct bpf_map *map, void *key,
545 				   struct sock *sk, u64 flags);
546 
547 int sock_map_update_elem_sys(struct bpf_map *map, void *key, void *value,
548 			     u64 flags)
549 {
550 	struct socket *sock;
551 	struct sock *sk;
552 	int ret;
553 	u64 ufd;
554 
555 	if (map->value_size == sizeof(u64))
556 		ufd = *(u64 *)value;
557 	else
558 		ufd = *(u32 *)value;
559 	if (ufd > S32_MAX)
560 		return -EINVAL;
561 
562 	sock = sockfd_lookup(ufd, &ret);
563 	if (!sock)
564 		return ret;
565 	sk = sock->sk;
566 	if (!sk) {
567 		ret = -EINVAL;
568 		goto out;
569 	}
570 	if (!sock_map_sk_is_suitable(sk)) {
571 		ret = -EOPNOTSUPP;
572 		goto out;
573 	}
574 
575 	sock_map_sk_acquire(sk);
576 	if (!sock_map_sk_state_allowed(sk))
577 		ret = -EOPNOTSUPP;
578 	else if (map->map_type == BPF_MAP_TYPE_SOCKMAP)
579 		ret = sock_map_update_common(map, *(u32 *)key, sk, flags);
580 	else
581 		ret = sock_hash_update_common(map, key, sk, flags);
582 	sock_map_sk_release(sk);
583 out:
584 	sockfd_put(sock);
585 	return ret;
586 }
587 
588 static long sock_map_update_elem(struct bpf_map *map, void *key,
589 				 void *value, u64 flags)
590 {
591 	struct sock *sk = (struct sock *)value;
592 	int ret;
593 
594 	if (unlikely(!sk || !sk_fullsock(sk)))
595 		return -EINVAL;
596 
597 	if (!sock_map_sk_is_suitable(sk))
598 		return -EOPNOTSUPP;
599 
600 	local_bh_disable();
601 	bh_lock_sock(sk);
602 	if (!sock_map_sk_state_allowed(sk))
603 		ret = -EOPNOTSUPP;
604 	else if (map->map_type == BPF_MAP_TYPE_SOCKMAP)
605 		ret = sock_map_update_common(map, *(u32 *)key, sk, flags);
606 	else
607 		ret = sock_hash_update_common(map, key, sk, flags);
608 	bh_unlock_sock(sk);
609 	local_bh_enable();
610 	return ret;
611 }
612 
613 BPF_CALL_4(bpf_sock_map_update, struct bpf_sock_ops_kern *, sops,
614 	   struct bpf_map *, map, void *, key, u64, flags)
615 {
616 	WARN_ON_ONCE(!rcu_read_lock_held());
617 
618 	if (likely(sock_map_sk_is_suitable(sops->sk) &&
619 		   sock_map_op_okay(sops)))
620 		return sock_map_update_common(map, *(u32 *)key, sops->sk,
621 					      flags);
622 	return -EOPNOTSUPP;
623 }
624 
625 const struct bpf_func_proto bpf_sock_map_update_proto = {
626 	.func		= bpf_sock_map_update,
627 	.gpl_only	= false,
628 	.pkt_access	= true,
629 	.ret_type	= RET_INTEGER,
630 	.arg1_type	= ARG_PTR_TO_CTX,
631 	.arg2_type	= ARG_CONST_MAP_PTR,
632 	.arg3_type	= ARG_PTR_TO_MAP_KEY,
633 	.arg4_type	= ARG_ANYTHING,
634 };
635 
636 BPF_CALL_4(bpf_sk_redirect_map, struct sk_buff *, skb,
637 	   struct bpf_map *, map, u32, key, u64, flags)
638 {
639 	struct sock *sk;
640 
641 	if (unlikely(flags & ~(BPF_F_INGRESS)))
642 		return SK_DROP;
643 
644 	sk = __sock_map_lookup_elem(map, key);
645 	if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
646 		return SK_DROP;
647 
648 	skb_bpf_set_redir(skb, sk, flags & BPF_F_INGRESS);
649 	return SK_PASS;
650 }
651 
652 const struct bpf_func_proto bpf_sk_redirect_map_proto = {
653 	.func           = bpf_sk_redirect_map,
654 	.gpl_only       = false,
655 	.ret_type       = RET_INTEGER,
656 	.arg1_type	= ARG_PTR_TO_CTX,
657 	.arg2_type      = ARG_CONST_MAP_PTR,
658 	.arg3_type      = ARG_ANYTHING,
659 	.arg4_type      = ARG_ANYTHING,
660 };
661 
662 BPF_CALL_4(bpf_msg_redirect_map, struct sk_msg *, msg,
663 	   struct bpf_map *, map, u32, key, u64, flags)
664 {
665 	struct sock *sk;
666 
667 	if (unlikely(flags & ~(BPF_F_INGRESS)))
668 		return SK_DROP;
669 
670 	sk = __sock_map_lookup_elem(map, key);
671 	if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
672 		return SK_DROP;
673 	if (!(flags & BPF_F_INGRESS) && !sk_is_tcp(sk))
674 		return SK_DROP;
675 
676 	msg->flags = flags;
677 	msg->sk_redir = sk;
678 	return SK_PASS;
679 }
680 
681 const struct bpf_func_proto bpf_msg_redirect_map_proto = {
682 	.func           = bpf_msg_redirect_map,
683 	.gpl_only       = false,
684 	.ret_type       = RET_INTEGER,
685 	.arg1_type	= ARG_PTR_TO_CTX,
686 	.arg2_type      = ARG_CONST_MAP_PTR,
687 	.arg3_type      = ARG_ANYTHING,
688 	.arg4_type      = ARG_ANYTHING,
689 };
690 
691 struct sock_map_seq_info {
692 	struct bpf_map *map;
693 	struct sock *sk;
694 	u32 index;
695 };
696 
697 struct bpf_iter__sockmap {
698 	__bpf_md_ptr(struct bpf_iter_meta *, meta);
699 	__bpf_md_ptr(struct bpf_map *, map);
700 	__bpf_md_ptr(void *, key);
701 	__bpf_md_ptr(struct sock *, sk);
702 };
703 
704 DEFINE_BPF_ITER_FUNC(sockmap, struct bpf_iter_meta *meta,
705 		     struct bpf_map *map, void *key,
706 		     struct sock *sk)
707 
708 static void *sock_map_seq_lookup_elem(struct sock_map_seq_info *info)
709 {
710 	if (unlikely(info->index >= info->map->max_entries))
711 		return NULL;
712 
713 	info->sk = __sock_map_lookup_elem(info->map, info->index);
714 
715 	/* can't return sk directly, since that might be NULL */
716 	return info;
717 }
718 
719 static void *sock_map_seq_start(struct seq_file *seq, loff_t *pos)
720 	__acquires(rcu)
721 {
722 	struct sock_map_seq_info *info = seq->private;
723 
724 	if (*pos == 0)
725 		++*pos;
726 
727 	/* pairs with sock_map_seq_stop */
728 	rcu_read_lock();
729 	return sock_map_seq_lookup_elem(info);
730 }
731 
732 static void *sock_map_seq_next(struct seq_file *seq, void *v, loff_t *pos)
733 	__must_hold(rcu)
734 {
735 	struct sock_map_seq_info *info = seq->private;
736 
737 	++*pos;
738 	++info->index;
739 
740 	return sock_map_seq_lookup_elem(info);
741 }
742 
743 static int sock_map_seq_show(struct seq_file *seq, void *v)
744 	__must_hold(rcu)
745 {
746 	struct sock_map_seq_info *info = seq->private;
747 	struct bpf_iter__sockmap ctx = {};
748 	struct bpf_iter_meta meta;
749 	struct bpf_prog *prog;
750 
751 	meta.seq = seq;
752 	prog = bpf_iter_get_info(&meta, !v);
753 	if (!prog)
754 		return 0;
755 
756 	ctx.meta = &meta;
757 	ctx.map = info->map;
758 	if (v) {
759 		ctx.key = &info->index;
760 		ctx.sk = info->sk;
761 	}
762 
763 	return bpf_iter_run_prog(prog, &ctx);
764 }
765 
766 static void sock_map_seq_stop(struct seq_file *seq, void *v)
767 	__releases(rcu)
768 {
769 	if (!v)
770 		(void)sock_map_seq_show(seq, NULL);
771 
772 	/* pairs with sock_map_seq_start */
773 	rcu_read_unlock();
774 }
775 
776 static const struct seq_operations sock_map_seq_ops = {
777 	.start	= sock_map_seq_start,
778 	.next	= sock_map_seq_next,
779 	.stop	= sock_map_seq_stop,
780 	.show	= sock_map_seq_show,
781 };
782 
783 static int sock_map_init_seq_private(void *priv_data,
784 				     struct bpf_iter_aux_info *aux)
785 {
786 	struct sock_map_seq_info *info = priv_data;
787 
788 	bpf_map_inc_with_uref(aux->map);
789 	info->map = aux->map;
790 	return 0;
791 }
792 
793 static void sock_map_fini_seq_private(void *priv_data)
794 {
795 	struct sock_map_seq_info *info = priv_data;
796 
797 	bpf_map_put_with_uref(info->map);
798 }
799 
800 static u64 sock_map_mem_usage(const struct bpf_map *map)
801 {
802 	u64 usage = sizeof(struct bpf_stab);
803 
804 	usage += (u64)map->max_entries * sizeof(struct sock *);
805 	return usage;
806 }
807 
808 static const struct bpf_iter_seq_info sock_map_iter_seq_info = {
809 	.seq_ops		= &sock_map_seq_ops,
810 	.init_seq_private	= sock_map_init_seq_private,
811 	.fini_seq_private	= sock_map_fini_seq_private,
812 	.seq_priv_size		= sizeof(struct sock_map_seq_info),
813 };
814 
815 BTF_ID_LIST_SINGLE(sock_map_btf_ids, struct, bpf_stab)
816 const struct bpf_map_ops sock_map_ops = {
817 	.map_meta_equal		= bpf_map_meta_equal,
818 	.map_alloc		= sock_map_alloc,
819 	.map_free		= sock_map_free,
820 	.map_get_next_key	= sock_map_get_next_key,
821 	.map_lookup_elem_sys_only = sock_map_lookup_sys,
822 	.map_update_elem	= sock_map_update_elem,
823 	.map_delete_elem	= sock_map_delete_elem,
824 	.map_lookup_elem	= sock_map_lookup,
825 	.map_release_uref	= sock_map_release_progs,
826 	.map_check_btf		= map_check_no_btf,
827 	.map_mem_usage		= sock_map_mem_usage,
828 	.map_btf_id		= &sock_map_btf_ids[0],
829 	.iter_seq_info		= &sock_map_iter_seq_info,
830 };
831 
832 struct bpf_shtab_elem {
833 	struct rcu_head rcu;
834 	u32 hash;
835 	struct sock *sk;
836 	struct hlist_node node;
837 	u8 key[];
838 };
839 
840 struct bpf_shtab_bucket {
841 	struct hlist_head head;
842 	spinlock_t lock;
843 };
844 
845 struct bpf_shtab {
846 	struct bpf_map map;
847 	struct bpf_shtab_bucket *buckets;
848 	u32 buckets_num;
849 	u32 elem_size;
850 	struct sk_psock_progs progs;
851 	atomic_t count;
852 };
853 
854 static inline u32 sock_hash_bucket_hash(const void *key, u32 len)
855 {
856 	return jhash(key, len, 0);
857 }
858 
859 static struct bpf_shtab_bucket *sock_hash_select_bucket(struct bpf_shtab *htab,
860 							u32 hash)
861 {
862 	return &htab->buckets[hash & (htab->buckets_num - 1)];
863 }
864 
865 static struct bpf_shtab_elem *
866 sock_hash_lookup_elem_raw(struct hlist_head *head, u32 hash, void *key,
867 			  u32 key_size)
868 {
869 	struct bpf_shtab_elem *elem;
870 
871 	hlist_for_each_entry_rcu(elem, head, node) {
872 		if (elem->hash == hash &&
873 		    !memcmp(&elem->key, key, key_size))
874 			return elem;
875 	}
876 
877 	return NULL;
878 }
879 
880 static struct sock *__sock_hash_lookup_elem(struct bpf_map *map, void *key)
881 {
882 	struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
883 	u32 key_size = map->key_size, hash;
884 	struct bpf_shtab_bucket *bucket;
885 	struct bpf_shtab_elem *elem;
886 
887 	WARN_ON_ONCE(!rcu_read_lock_held());
888 
889 	hash = sock_hash_bucket_hash(key, key_size);
890 	bucket = sock_hash_select_bucket(htab, hash);
891 	elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size);
892 
893 	return elem ? elem->sk : NULL;
894 }
895 
896 static void sock_hash_free_elem(struct bpf_shtab *htab,
897 				struct bpf_shtab_elem *elem)
898 {
899 	atomic_dec(&htab->count);
900 	kfree_rcu(elem, rcu);
901 }
902 
903 static void sock_hash_delete_from_link(struct bpf_map *map, struct sock *sk,
904 				       void *link_raw)
905 {
906 	struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
907 	struct bpf_shtab_elem *elem_probe, *elem = link_raw;
908 	struct bpf_shtab_bucket *bucket;
909 
910 	WARN_ON_ONCE(!rcu_read_lock_held());
911 	bucket = sock_hash_select_bucket(htab, elem->hash);
912 
913 	/* elem may be deleted in parallel from the map, but access here
914 	 * is okay since it's going away only after RCU grace period.
915 	 * However, we need to check whether it's still present.
916 	 */
917 	spin_lock_bh(&bucket->lock);
918 	elem_probe = sock_hash_lookup_elem_raw(&bucket->head, elem->hash,
919 					       elem->key, map->key_size);
920 	if (elem_probe && elem_probe == elem) {
921 		hlist_del_rcu(&elem->node);
922 		sock_map_unref(elem->sk, elem);
923 		sock_hash_free_elem(htab, elem);
924 	}
925 	spin_unlock_bh(&bucket->lock);
926 }
927 
928 static long sock_hash_delete_elem(struct bpf_map *map, void *key)
929 {
930 	struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
931 	u32 hash, key_size = map->key_size;
932 	struct bpf_shtab_bucket *bucket;
933 	struct bpf_shtab_elem *elem;
934 	int ret = -ENOENT;
935 
936 	hash = sock_hash_bucket_hash(key, key_size);
937 	bucket = sock_hash_select_bucket(htab, hash);
938 
939 	spin_lock_bh(&bucket->lock);
940 	elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size);
941 	if (elem) {
942 		hlist_del_rcu(&elem->node);
943 		sock_map_unref(elem->sk, elem);
944 		sock_hash_free_elem(htab, elem);
945 		ret = 0;
946 	}
947 	spin_unlock_bh(&bucket->lock);
948 	return ret;
949 }
950 
951 static struct bpf_shtab_elem *sock_hash_alloc_elem(struct bpf_shtab *htab,
952 						   void *key, u32 key_size,
953 						   u32 hash, struct sock *sk,
954 						   struct bpf_shtab_elem *old)
955 {
956 	struct bpf_shtab_elem *new;
957 
958 	if (atomic_inc_return(&htab->count) > htab->map.max_entries) {
959 		if (!old) {
960 			atomic_dec(&htab->count);
961 			return ERR_PTR(-E2BIG);
962 		}
963 	}
964 
965 	new = bpf_map_kmalloc_node(&htab->map, htab->elem_size,
966 				   GFP_ATOMIC | __GFP_NOWARN,
967 				   htab->map.numa_node);
968 	if (!new) {
969 		atomic_dec(&htab->count);
970 		return ERR_PTR(-ENOMEM);
971 	}
972 	memcpy(new->key, key, key_size);
973 	new->sk = sk;
974 	new->hash = hash;
975 	return new;
976 }
977 
978 static int sock_hash_update_common(struct bpf_map *map, void *key,
979 				   struct sock *sk, u64 flags)
980 {
981 	struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
982 	u32 key_size = map->key_size, hash;
983 	struct bpf_shtab_elem *elem, *elem_new;
984 	struct bpf_shtab_bucket *bucket;
985 	struct sk_psock_link *link;
986 	struct sk_psock *psock;
987 	int ret;
988 
989 	WARN_ON_ONCE(!rcu_read_lock_held());
990 	if (unlikely(flags > BPF_EXIST))
991 		return -EINVAL;
992 
993 	link = sk_psock_init_link();
994 	if (!link)
995 		return -ENOMEM;
996 
997 	ret = sock_map_link(map, sk);
998 	if (ret < 0)
999 		goto out_free;
1000 
1001 	psock = sk_psock(sk);
1002 	WARN_ON_ONCE(!psock);
1003 
1004 	hash = sock_hash_bucket_hash(key, key_size);
1005 	bucket = sock_hash_select_bucket(htab, hash);
1006 
1007 	spin_lock_bh(&bucket->lock);
1008 	elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size);
1009 	if (elem && flags == BPF_NOEXIST) {
1010 		ret = -EEXIST;
1011 		goto out_unlock;
1012 	} else if (!elem && flags == BPF_EXIST) {
1013 		ret = -ENOENT;
1014 		goto out_unlock;
1015 	}
1016 
1017 	elem_new = sock_hash_alloc_elem(htab, key, key_size, hash, sk, elem);
1018 	if (IS_ERR(elem_new)) {
1019 		ret = PTR_ERR(elem_new);
1020 		goto out_unlock;
1021 	}
1022 
1023 	sock_map_add_link(psock, link, map, elem_new);
1024 	/* Add new element to the head of the list, so that
1025 	 * concurrent search will find it before old elem.
1026 	 */
1027 	hlist_add_head_rcu(&elem_new->node, &bucket->head);
1028 	if (elem) {
1029 		hlist_del_rcu(&elem->node);
1030 		sock_map_unref(elem->sk, elem);
1031 		sock_hash_free_elem(htab, elem);
1032 	}
1033 	spin_unlock_bh(&bucket->lock);
1034 	return 0;
1035 out_unlock:
1036 	spin_unlock_bh(&bucket->lock);
1037 	sk_psock_put(sk, psock);
1038 out_free:
1039 	sk_psock_free_link(link);
1040 	return ret;
1041 }
1042 
1043 static int sock_hash_get_next_key(struct bpf_map *map, void *key,
1044 				  void *key_next)
1045 {
1046 	struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
1047 	struct bpf_shtab_elem *elem, *elem_next;
1048 	u32 hash, key_size = map->key_size;
1049 	struct hlist_head *head;
1050 	int i = 0;
1051 
1052 	if (!key)
1053 		goto find_first_elem;
1054 	hash = sock_hash_bucket_hash(key, key_size);
1055 	head = &sock_hash_select_bucket(htab, hash)->head;
1056 	elem = sock_hash_lookup_elem_raw(head, hash, key, key_size);
1057 	if (!elem)
1058 		goto find_first_elem;
1059 
1060 	elem_next = hlist_entry_safe(rcu_dereference(hlist_next_rcu(&elem->node)),
1061 				     struct bpf_shtab_elem, node);
1062 	if (elem_next) {
1063 		memcpy(key_next, elem_next->key, key_size);
1064 		return 0;
1065 	}
1066 
1067 	i = hash & (htab->buckets_num - 1);
1068 	i++;
1069 find_first_elem:
1070 	for (; i < htab->buckets_num; i++) {
1071 		head = &sock_hash_select_bucket(htab, i)->head;
1072 		elem_next = hlist_entry_safe(rcu_dereference(hlist_first_rcu(head)),
1073 					     struct bpf_shtab_elem, node);
1074 		if (elem_next) {
1075 			memcpy(key_next, elem_next->key, key_size);
1076 			return 0;
1077 		}
1078 	}
1079 
1080 	return -ENOENT;
1081 }
1082 
1083 static struct bpf_map *sock_hash_alloc(union bpf_attr *attr)
1084 {
1085 	struct bpf_shtab *htab;
1086 	int i, err;
1087 
1088 	if (attr->max_entries == 0 ||
1089 	    attr->key_size    == 0 ||
1090 	    (attr->value_size != sizeof(u32) &&
1091 	     attr->value_size != sizeof(u64)) ||
1092 	    attr->map_flags & ~SOCK_CREATE_FLAG_MASK)
1093 		return ERR_PTR(-EINVAL);
1094 	if (attr->key_size > MAX_BPF_STACK)
1095 		return ERR_PTR(-E2BIG);
1096 
1097 	htab = bpf_map_area_alloc(sizeof(*htab), NUMA_NO_NODE);
1098 	if (!htab)
1099 		return ERR_PTR(-ENOMEM);
1100 
1101 	bpf_map_init_from_attr(&htab->map, attr);
1102 
1103 	htab->buckets_num = roundup_pow_of_two(htab->map.max_entries);
1104 	htab->elem_size = sizeof(struct bpf_shtab_elem) +
1105 			  round_up(htab->map.key_size, 8);
1106 	if (htab->buckets_num == 0 ||
1107 	    htab->buckets_num > U32_MAX / sizeof(struct bpf_shtab_bucket)) {
1108 		err = -EINVAL;
1109 		goto free_htab;
1110 	}
1111 
1112 	htab->buckets = bpf_map_area_alloc(htab->buckets_num *
1113 					   sizeof(struct bpf_shtab_bucket),
1114 					   htab->map.numa_node);
1115 	if (!htab->buckets) {
1116 		err = -ENOMEM;
1117 		goto free_htab;
1118 	}
1119 
1120 	for (i = 0; i < htab->buckets_num; i++) {
1121 		INIT_HLIST_HEAD(&htab->buckets[i].head);
1122 		spin_lock_init(&htab->buckets[i].lock);
1123 	}
1124 
1125 	return &htab->map;
1126 free_htab:
1127 	bpf_map_area_free(htab);
1128 	return ERR_PTR(err);
1129 }
1130 
1131 static void sock_hash_free(struct bpf_map *map)
1132 {
1133 	struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
1134 	struct bpf_shtab_bucket *bucket;
1135 	struct hlist_head unlink_list;
1136 	struct bpf_shtab_elem *elem;
1137 	struct hlist_node *node;
1138 	int i;
1139 
1140 	/* After the sync no updates or deletes will be in-flight so it
1141 	 * is safe to walk map and remove entries without risking a race
1142 	 * in EEXIST update case.
1143 	 */
1144 	synchronize_rcu();
1145 	for (i = 0; i < htab->buckets_num; i++) {
1146 		bucket = sock_hash_select_bucket(htab, i);
1147 
1148 		/* We are racing with sock_hash_delete_from_link to
1149 		 * enter the spin-lock critical section. Every socket on
1150 		 * the list is still linked to sockhash. Since link
1151 		 * exists, psock exists and holds a ref to socket. That
1152 		 * lets us to grab a socket ref too.
1153 		 */
1154 		spin_lock_bh(&bucket->lock);
1155 		hlist_for_each_entry(elem, &bucket->head, node)
1156 			sock_hold(elem->sk);
1157 		hlist_move_list(&bucket->head, &unlink_list);
1158 		spin_unlock_bh(&bucket->lock);
1159 
1160 		/* Process removed entries out of atomic context to
1161 		 * block for socket lock before deleting the psock's
1162 		 * link to sockhash.
1163 		 */
1164 		hlist_for_each_entry_safe(elem, node, &unlink_list, node) {
1165 			hlist_del(&elem->node);
1166 			lock_sock(elem->sk);
1167 			rcu_read_lock();
1168 			sock_map_unref(elem->sk, elem);
1169 			rcu_read_unlock();
1170 			release_sock(elem->sk);
1171 			sock_put(elem->sk);
1172 			sock_hash_free_elem(htab, elem);
1173 		}
1174 	}
1175 
1176 	/* wait for psock readers accessing its map link */
1177 	synchronize_rcu();
1178 
1179 	bpf_map_area_free(htab->buckets);
1180 	bpf_map_area_free(htab);
1181 }
1182 
1183 static void *sock_hash_lookup_sys(struct bpf_map *map, void *key)
1184 {
1185 	struct sock *sk;
1186 
1187 	if (map->value_size != sizeof(u64))
1188 		return ERR_PTR(-ENOSPC);
1189 
1190 	sk = __sock_hash_lookup_elem(map, key);
1191 	if (!sk)
1192 		return ERR_PTR(-ENOENT);
1193 
1194 	__sock_gen_cookie(sk);
1195 	return &sk->sk_cookie;
1196 }
1197 
1198 static void *sock_hash_lookup(struct bpf_map *map, void *key)
1199 {
1200 	struct sock *sk;
1201 
1202 	sk = __sock_hash_lookup_elem(map, key);
1203 	if (!sk)
1204 		return NULL;
1205 	if (sk_is_refcounted(sk) && !refcount_inc_not_zero(&sk->sk_refcnt))
1206 		return NULL;
1207 	return sk;
1208 }
1209 
1210 static void sock_hash_release_progs(struct bpf_map *map)
1211 {
1212 	psock_progs_drop(&container_of(map, struct bpf_shtab, map)->progs);
1213 }
1214 
1215 BPF_CALL_4(bpf_sock_hash_update, struct bpf_sock_ops_kern *, sops,
1216 	   struct bpf_map *, map, void *, key, u64, flags)
1217 {
1218 	WARN_ON_ONCE(!rcu_read_lock_held());
1219 
1220 	if (likely(sock_map_sk_is_suitable(sops->sk) &&
1221 		   sock_map_op_okay(sops)))
1222 		return sock_hash_update_common(map, key, sops->sk, flags);
1223 	return -EOPNOTSUPP;
1224 }
1225 
1226 const struct bpf_func_proto bpf_sock_hash_update_proto = {
1227 	.func		= bpf_sock_hash_update,
1228 	.gpl_only	= false,
1229 	.pkt_access	= true,
1230 	.ret_type	= RET_INTEGER,
1231 	.arg1_type	= ARG_PTR_TO_CTX,
1232 	.arg2_type	= ARG_CONST_MAP_PTR,
1233 	.arg3_type	= ARG_PTR_TO_MAP_KEY,
1234 	.arg4_type	= ARG_ANYTHING,
1235 };
1236 
1237 BPF_CALL_4(bpf_sk_redirect_hash, struct sk_buff *, skb,
1238 	   struct bpf_map *, map, void *, key, u64, flags)
1239 {
1240 	struct sock *sk;
1241 
1242 	if (unlikely(flags & ~(BPF_F_INGRESS)))
1243 		return SK_DROP;
1244 
1245 	sk = __sock_hash_lookup_elem(map, key);
1246 	if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
1247 		return SK_DROP;
1248 
1249 	skb_bpf_set_redir(skb, sk, flags & BPF_F_INGRESS);
1250 	return SK_PASS;
1251 }
1252 
1253 const struct bpf_func_proto bpf_sk_redirect_hash_proto = {
1254 	.func           = bpf_sk_redirect_hash,
1255 	.gpl_only       = false,
1256 	.ret_type       = RET_INTEGER,
1257 	.arg1_type	= ARG_PTR_TO_CTX,
1258 	.arg2_type      = ARG_CONST_MAP_PTR,
1259 	.arg3_type      = ARG_PTR_TO_MAP_KEY,
1260 	.arg4_type      = ARG_ANYTHING,
1261 };
1262 
1263 BPF_CALL_4(bpf_msg_redirect_hash, struct sk_msg *, msg,
1264 	   struct bpf_map *, map, void *, key, u64, flags)
1265 {
1266 	struct sock *sk;
1267 
1268 	if (unlikely(flags & ~(BPF_F_INGRESS)))
1269 		return SK_DROP;
1270 
1271 	sk = __sock_hash_lookup_elem(map, key);
1272 	if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
1273 		return SK_DROP;
1274 	if (!(flags & BPF_F_INGRESS) && !sk_is_tcp(sk))
1275 		return SK_DROP;
1276 
1277 	msg->flags = flags;
1278 	msg->sk_redir = sk;
1279 	return SK_PASS;
1280 }
1281 
1282 const struct bpf_func_proto bpf_msg_redirect_hash_proto = {
1283 	.func           = bpf_msg_redirect_hash,
1284 	.gpl_only       = false,
1285 	.ret_type       = RET_INTEGER,
1286 	.arg1_type	= ARG_PTR_TO_CTX,
1287 	.arg2_type      = ARG_CONST_MAP_PTR,
1288 	.arg3_type      = ARG_PTR_TO_MAP_KEY,
1289 	.arg4_type      = ARG_ANYTHING,
1290 };
1291 
1292 struct sock_hash_seq_info {
1293 	struct bpf_map *map;
1294 	struct bpf_shtab *htab;
1295 	u32 bucket_id;
1296 };
1297 
1298 static void *sock_hash_seq_find_next(struct sock_hash_seq_info *info,
1299 				     struct bpf_shtab_elem *prev_elem)
1300 {
1301 	const struct bpf_shtab *htab = info->htab;
1302 	struct bpf_shtab_bucket *bucket;
1303 	struct bpf_shtab_elem *elem;
1304 	struct hlist_node *node;
1305 
1306 	/* try to find next elem in the same bucket */
1307 	if (prev_elem) {
1308 		node = rcu_dereference(hlist_next_rcu(&prev_elem->node));
1309 		elem = hlist_entry_safe(node, struct bpf_shtab_elem, node);
1310 		if (elem)
1311 			return elem;
1312 
1313 		/* no more elements, continue in the next bucket */
1314 		info->bucket_id++;
1315 	}
1316 
1317 	for (; info->bucket_id < htab->buckets_num; info->bucket_id++) {
1318 		bucket = &htab->buckets[info->bucket_id];
1319 		node = rcu_dereference(hlist_first_rcu(&bucket->head));
1320 		elem = hlist_entry_safe(node, struct bpf_shtab_elem, node);
1321 		if (elem)
1322 			return elem;
1323 	}
1324 
1325 	return NULL;
1326 }
1327 
1328 static void *sock_hash_seq_start(struct seq_file *seq, loff_t *pos)
1329 	__acquires(rcu)
1330 {
1331 	struct sock_hash_seq_info *info = seq->private;
1332 
1333 	if (*pos == 0)
1334 		++*pos;
1335 
1336 	/* pairs with sock_hash_seq_stop */
1337 	rcu_read_lock();
1338 	return sock_hash_seq_find_next(info, NULL);
1339 }
1340 
1341 static void *sock_hash_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1342 	__must_hold(rcu)
1343 {
1344 	struct sock_hash_seq_info *info = seq->private;
1345 
1346 	++*pos;
1347 	return sock_hash_seq_find_next(info, v);
1348 }
1349 
1350 static int sock_hash_seq_show(struct seq_file *seq, void *v)
1351 	__must_hold(rcu)
1352 {
1353 	struct sock_hash_seq_info *info = seq->private;
1354 	struct bpf_iter__sockmap ctx = {};
1355 	struct bpf_shtab_elem *elem = v;
1356 	struct bpf_iter_meta meta;
1357 	struct bpf_prog *prog;
1358 
1359 	meta.seq = seq;
1360 	prog = bpf_iter_get_info(&meta, !elem);
1361 	if (!prog)
1362 		return 0;
1363 
1364 	ctx.meta = &meta;
1365 	ctx.map = info->map;
1366 	if (elem) {
1367 		ctx.key = elem->key;
1368 		ctx.sk = elem->sk;
1369 	}
1370 
1371 	return bpf_iter_run_prog(prog, &ctx);
1372 }
1373 
1374 static void sock_hash_seq_stop(struct seq_file *seq, void *v)
1375 	__releases(rcu)
1376 {
1377 	if (!v)
1378 		(void)sock_hash_seq_show(seq, NULL);
1379 
1380 	/* pairs with sock_hash_seq_start */
1381 	rcu_read_unlock();
1382 }
1383 
1384 static const struct seq_operations sock_hash_seq_ops = {
1385 	.start	= sock_hash_seq_start,
1386 	.next	= sock_hash_seq_next,
1387 	.stop	= sock_hash_seq_stop,
1388 	.show	= sock_hash_seq_show,
1389 };
1390 
1391 static int sock_hash_init_seq_private(void *priv_data,
1392 				      struct bpf_iter_aux_info *aux)
1393 {
1394 	struct sock_hash_seq_info *info = priv_data;
1395 
1396 	bpf_map_inc_with_uref(aux->map);
1397 	info->map = aux->map;
1398 	info->htab = container_of(aux->map, struct bpf_shtab, map);
1399 	return 0;
1400 }
1401 
1402 static void sock_hash_fini_seq_private(void *priv_data)
1403 {
1404 	struct sock_hash_seq_info *info = priv_data;
1405 
1406 	bpf_map_put_with_uref(info->map);
1407 }
1408 
1409 static u64 sock_hash_mem_usage(const struct bpf_map *map)
1410 {
1411 	struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
1412 	u64 usage = sizeof(*htab);
1413 
1414 	usage += htab->buckets_num * sizeof(struct bpf_shtab_bucket);
1415 	usage += atomic_read(&htab->count) * (u64)htab->elem_size;
1416 	return usage;
1417 }
1418 
1419 static const struct bpf_iter_seq_info sock_hash_iter_seq_info = {
1420 	.seq_ops		= &sock_hash_seq_ops,
1421 	.init_seq_private	= sock_hash_init_seq_private,
1422 	.fini_seq_private	= sock_hash_fini_seq_private,
1423 	.seq_priv_size		= sizeof(struct sock_hash_seq_info),
1424 };
1425 
1426 BTF_ID_LIST_SINGLE(sock_hash_map_btf_ids, struct, bpf_shtab)
1427 const struct bpf_map_ops sock_hash_ops = {
1428 	.map_meta_equal		= bpf_map_meta_equal,
1429 	.map_alloc		= sock_hash_alloc,
1430 	.map_free		= sock_hash_free,
1431 	.map_get_next_key	= sock_hash_get_next_key,
1432 	.map_update_elem	= sock_map_update_elem,
1433 	.map_delete_elem	= sock_hash_delete_elem,
1434 	.map_lookup_elem	= sock_hash_lookup,
1435 	.map_lookup_elem_sys_only = sock_hash_lookup_sys,
1436 	.map_release_uref	= sock_hash_release_progs,
1437 	.map_check_btf		= map_check_no_btf,
1438 	.map_mem_usage		= sock_hash_mem_usage,
1439 	.map_btf_id		= &sock_hash_map_btf_ids[0],
1440 	.iter_seq_info		= &sock_hash_iter_seq_info,
1441 };
1442 
1443 static struct sk_psock_progs *sock_map_progs(struct bpf_map *map)
1444 {
1445 	switch (map->map_type) {
1446 	case BPF_MAP_TYPE_SOCKMAP:
1447 		return &container_of(map, struct bpf_stab, map)->progs;
1448 	case BPF_MAP_TYPE_SOCKHASH:
1449 		return &container_of(map, struct bpf_shtab, map)->progs;
1450 	default:
1451 		break;
1452 	}
1453 
1454 	return NULL;
1455 }
1456 
1457 static int sock_map_prog_lookup(struct bpf_map *map, struct bpf_prog ***pprog,
1458 				u32 which)
1459 {
1460 	struct sk_psock_progs *progs = sock_map_progs(map);
1461 
1462 	if (!progs)
1463 		return -EOPNOTSUPP;
1464 
1465 	switch (which) {
1466 	case BPF_SK_MSG_VERDICT:
1467 		*pprog = &progs->msg_parser;
1468 		break;
1469 #if IS_ENABLED(CONFIG_BPF_STREAM_PARSER)
1470 	case BPF_SK_SKB_STREAM_PARSER:
1471 		*pprog = &progs->stream_parser;
1472 		break;
1473 #endif
1474 	case BPF_SK_SKB_STREAM_VERDICT:
1475 		if (progs->skb_verdict)
1476 			return -EBUSY;
1477 		*pprog = &progs->stream_verdict;
1478 		break;
1479 	case BPF_SK_SKB_VERDICT:
1480 		if (progs->stream_verdict)
1481 			return -EBUSY;
1482 		*pprog = &progs->skb_verdict;
1483 		break;
1484 	default:
1485 		return -EOPNOTSUPP;
1486 	}
1487 
1488 	return 0;
1489 }
1490 
1491 static int sock_map_prog_update(struct bpf_map *map, struct bpf_prog *prog,
1492 				struct bpf_prog *old, u32 which)
1493 {
1494 	struct bpf_prog **pprog;
1495 	int ret;
1496 
1497 	ret = sock_map_prog_lookup(map, &pprog, which);
1498 	if (ret)
1499 		return ret;
1500 
1501 	if (old)
1502 		return psock_replace_prog(pprog, prog, old);
1503 
1504 	psock_set_prog(pprog, prog);
1505 	return 0;
1506 }
1507 
1508 int sock_map_bpf_prog_query(const union bpf_attr *attr,
1509 			    union bpf_attr __user *uattr)
1510 {
1511 	__u32 __user *prog_ids = u64_to_user_ptr(attr->query.prog_ids);
1512 	u32 prog_cnt = 0, flags = 0, ufd = attr->target_fd;
1513 	struct bpf_prog **pprog;
1514 	struct bpf_prog *prog;
1515 	struct bpf_map *map;
1516 	struct fd f;
1517 	u32 id = 0;
1518 	int ret;
1519 
1520 	if (attr->query.query_flags)
1521 		return -EINVAL;
1522 
1523 	f = fdget(ufd);
1524 	map = __bpf_map_get(f);
1525 	if (IS_ERR(map))
1526 		return PTR_ERR(map);
1527 
1528 	rcu_read_lock();
1529 
1530 	ret = sock_map_prog_lookup(map, &pprog, attr->query.attach_type);
1531 	if (ret)
1532 		goto end;
1533 
1534 	prog = *pprog;
1535 	prog_cnt = !prog ? 0 : 1;
1536 
1537 	if (!attr->query.prog_cnt || !prog_ids || !prog_cnt)
1538 		goto end;
1539 
1540 	/* we do not hold the refcnt, the bpf prog may be released
1541 	 * asynchronously and the id would be set to 0.
1542 	 */
1543 	id = data_race(prog->aux->id);
1544 	if (id == 0)
1545 		prog_cnt = 0;
1546 
1547 end:
1548 	rcu_read_unlock();
1549 
1550 	if (copy_to_user(&uattr->query.attach_flags, &flags, sizeof(flags)) ||
1551 	    (id != 0 && copy_to_user(prog_ids, &id, sizeof(u32))) ||
1552 	    copy_to_user(&uattr->query.prog_cnt, &prog_cnt, sizeof(prog_cnt)))
1553 		ret = -EFAULT;
1554 
1555 	fdput(f);
1556 	return ret;
1557 }
1558 
1559 static void sock_map_unlink(struct sock *sk, struct sk_psock_link *link)
1560 {
1561 	switch (link->map->map_type) {
1562 	case BPF_MAP_TYPE_SOCKMAP:
1563 		return sock_map_delete_from_link(link->map, sk,
1564 						 link->link_raw);
1565 	case BPF_MAP_TYPE_SOCKHASH:
1566 		return sock_hash_delete_from_link(link->map, sk,
1567 						  link->link_raw);
1568 	default:
1569 		break;
1570 	}
1571 }
1572 
1573 static void sock_map_remove_links(struct sock *sk, struct sk_psock *psock)
1574 {
1575 	struct sk_psock_link *link;
1576 
1577 	while ((link = sk_psock_link_pop(psock))) {
1578 		sock_map_unlink(sk, link);
1579 		sk_psock_free_link(link);
1580 	}
1581 }
1582 
1583 void sock_map_unhash(struct sock *sk)
1584 {
1585 	void (*saved_unhash)(struct sock *sk);
1586 	struct sk_psock *psock;
1587 
1588 	rcu_read_lock();
1589 	psock = sk_psock(sk);
1590 	if (unlikely(!psock)) {
1591 		rcu_read_unlock();
1592 		saved_unhash = READ_ONCE(sk->sk_prot)->unhash;
1593 	} else {
1594 		saved_unhash = psock->saved_unhash;
1595 		sock_map_remove_links(sk, psock);
1596 		rcu_read_unlock();
1597 	}
1598 	if (WARN_ON_ONCE(saved_unhash == sock_map_unhash))
1599 		return;
1600 	if (saved_unhash)
1601 		saved_unhash(sk);
1602 }
1603 EXPORT_SYMBOL_GPL(sock_map_unhash);
1604 
1605 void sock_map_destroy(struct sock *sk)
1606 {
1607 	void (*saved_destroy)(struct sock *sk);
1608 	struct sk_psock *psock;
1609 
1610 	rcu_read_lock();
1611 	psock = sk_psock_get(sk);
1612 	if (unlikely(!psock)) {
1613 		rcu_read_unlock();
1614 		saved_destroy = READ_ONCE(sk->sk_prot)->destroy;
1615 	} else {
1616 		saved_destroy = psock->saved_destroy;
1617 		sock_map_remove_links(sk, psock);
1618 		rcu_read_unlock();
1619 		sk_psock_stop(psock);
1620 		sk_psock_put(sk, psock);
1621 	}
1622 	if (WARN_ON_ONCE(saved_destroy == sock_map_destroy))
1623 		return;
1624 	if (saved_destroy)
1625 		saved_destroy(sk);
1626 }
1627 EXPORT_SYMBOL_GPL(sock_map_destroy);
1628 
1629 void sock_map_close(struct sock *sk, long timeout)
1630 {
1631 	void (*saved_close)(struct sock *sk, long timeout);
1632 	struct sk_psock *psock;
1633 
1634 	lock_sock(sk);
1635 	rcu_read_lock();
1636 	psock = sk_psock_get(sk);
1637 	if (unlikely(!psock)) {
1638 		rcu_read_unlock();
1639 		release_sock(sk);
1640 		saved_close = READ_ONCE(sk->sk_prot)->close;
1641 	} else {
1642 		saved_close = psock->saved_close;
1643 		sock_map_remove_links(sk, psock);
1644 		rcu_read_unlock();
1645 		sk_psock_stop(psock);
1646 		release_sock(sk);
1647 		cancel_delayed_work_sync(&psock->work);
1648 		sk_psock_put(sk, psock);
1649 	}
1650 
1651 	/* Make sure we do not recurse. This is a bug.
1652 	 * Leak the socket instead of crashing on a stack overflow.
1653 	 */
1654 	if (WARN_ON_ONCE(saved_close == sock_map_close))
1655 		return;
1656 	saved_close(sk, timeout);
1657 }
1658 EXPORT_SYMBOL_GPL(sock_map_close);
1659 
1660 static int sock_map_iter_attach_target(struct bpf_prog *prog,
1661 				       union bpf_iter_link_info *linfo,
1662 				       struct bpf_iter_aux_info *aux)
1663 {
1664 	struct bpf_map *map;
1665 	int err = -EINVAL;
1666 
1667 	if (!linfo->map.map_fd)
1668 		return -EBADF;
1669 
1670 	map = bpf_map_get_with_uref(linfo->map.map_fd);
1671 	if (IS_ERR(map))
1672 		return PTR_ERR(map);
1673 
1674 	if (map->map_type != BPF_MAP_TYPE_SOCKMAP &&
1675 	    map->map_type != BPF_MAP_TYPE_SOCKHASH)
1676 		goto put_map;
1677 
1678 	if (prog->aux->max_rdonly_access > map->key_size) {
1679 		err = -EACCES;
1680 		goto put_map;
1681 	}
1682 
1683 	aux->map = map;
1684 	return 0;
1685 
1686 put_map:
1687 	bpf_map_put_with_uref(map);
1688 	return err;
1689 }
1690 
1691 static void sock_map_iter_detach_target(struct bpf_iter_aux_info *aux)
1692 {
1693 	bpf_map_put_with_uref(aux->map);
1694 }
1695 
1696 static struct bpf_iter_reg sock_map_iter_reg = {
1697 	.target			= "sockmap",
1698 	.attach_target		= sock_map_iter_attach_target,
1699 	.detach_target		= sock_map_iter_detach_target,
1700 	.show_fdinfo		= bpf_iter_map_show_fdinfo,
1701 	.fill_link_info		= bpf_iter_map_fill_link_info,
1702 	.ctx_arg_info_size	= 2,
1703 	.ctx_arg_info		= {
1704 		{ offsetof(struct bpf_iter__sockmap, key),
1705 		  PTR_TO_BUF | PTR_MAYBE_NULL | MEM_RDONLY },
1706 		{ offsetof(struct bpf_iter__sockmap, sk),
1707 		  PTR_TO_BTF_ID_OR_NULL },
1708 	},
1709 };
1710 
1711 static int __init bpf_sockmap_iter_init(void)
1712 {
1713 	sock_map_iter_reg.ctx_arg_info[1].btf_id =
1714 		btf_sock_ids[BTF_SOCK_TYPE_SOCK];
1715 	return bpf_iter_reg_target(&sock_map_iter_reg);
1716 }
1717 late_initcall(bpf_sockmap_iter_init);
1718