xref: /linux/net/core/sock_map.c (revision 8b40a46966d294bc64bad0feb13d3304fde738f2)
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 	if (irqs_disabled())
415 		return -EOPNOTSUPP; /* locks here are hardirq-unsafe */
416 
417 	spin_lock_bh(&stab->lock);
418 	sk = *psk;
419 	if (!sk_test || sk_test == sk)
420 		sk = xchg(psk, NULL);
421 
422 	if (likely(sk))
423 		sock_map_unref(sk, psk);
424 	else
425 		err = -EINVAL;
426 
427 	spin_unlock_bh(&stab->lock);
428 	return err;
429 }
430 
431 static void sock_map_delete_from_link(struct bpf_map *map, struct sock *sk,
432 				      void *link_raw)
433 {
434 	struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
435 
436 	__sock_map_delete(stab, sk, link_raw);
437 }
438 
439 static long sock_map_delete_elem(struct bpf_map *map, void *key)
440 {
441 	struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
442 	u32 i = *(u32 *)key;
443 	struct sock **psk;
444 
445 	if (unlikely(i >= map->max_entries))
446 		return -EINVAL;
447 
448 	psk = &stab->sks[i];
449 	return __sock_map_delete(stab, NULL, psk);
450 }
451 
452 static int sock_map_get_next_key(struct bpf_map *map, void *key, void *next)
453 {
454 	struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
455 	u32 i = key ? *(u32 *)key : U32_MAX;
456 	u32 *key_next = next;
457 
458 	if (i == stab->map.max_entries - 1)
459 		return -ENOENT;
460 	if (i >= stab->map.max_entries)
461 		*key_next = 0;
462 	else
463 		*key_next = i + 1;
464 	return 0;
465 }
466 
467 static int sock_map_update_common(struct bpf_map *map, u32 idx,
468 				  struct sock *sk, u64 flags)
469 {
470 	struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
471 	struct sk_psock_link *link;
472 	struct sk_psock *psock;
473 	struct sock *osk;
474 	int ret;
475 
476 	WARN_ON_ONCE(!rcu_read_lock_held());
477 	if (unlikely(flags > BPF_EXIST))
478 		return -EINVAL;
479 	if (unlikely(idx >= map->max_entries))
480 		return -E2BIG;
481 
482 	link = sk_psock_init_link();
483 	if (!link)
484 		return -ENOMEM;
485 
486 	ret = sock_map_link(map, sk);
487 	if (ret < 0)
488 		goto out_free;
489 
490 	psock = sk_psock(sk);
491 	WARN_ON_ONCE(!psock);
492 
493 	spin_lock_bh(&stab->lock);
494 	osk = stab->sks[idx];
495 	if (osk && flags == BPF_NOEXIST) {
496 		ret = -EEXIST;
497 		goto out_unlock;
498 	} else if (!osk && flags == BPF_EXIST) {
499 		ret = -ENOENT;
500 		goto out_unlock;
501 	}
502 
503 	sock_map_add_link(psock, link, map, &stab->sks[idx]);
504 	stab->sks[idx] = sk;
505 	if (osk)
506 		sock_map_unref(osk, &stab->sks[idx]);
507 	spin_unlock_bh(&stab->lock);
508 	return 0;
509 out_unlock:
510 	spin_unlock_bh(&stab->lock);
511 	if (psock)
512 		sk_psock_put(sk, psock);
513 out_free:
514 	sk_psock_free_link(link);
515 	return ret;
516 }
517 
518 static bool sock_map_op_okay(const struct bpf_sock_ops_kern *ops)
519 {
520 	return ops->op == BPF_SOCK_OPS_PASSIVE_ESTABLISHED_CB ||
521 	       ops->op == BPF_SOCK_OPS_ACTIVE_ESTABLISHED_CB ||
522 	       ops->op == BPF_SOCK_OPS_TCP_LISTEN_CB;
523 }
524 
525 static bool sock_map_redirect_allowed(const struct sock *sk)
526 {
527 	if (sk_is_tcp(sk))
528 		return sk->sk_state != TCP_LISTEN;
529 	else
530 		return sk->sk_state == TCP_ESTABLISHED;
531 }
532 
533 static bool sock_map_sk_is_suitable(const struct sock *sk)
534 {
535 	return !!sk->sk_prot->psock_update_sk_prot;
536 }
537 
538 static bool sock_map_sk_state_allowed(const struct sock *sk)
539 {
540 	if (sk_is_tcp(sk))
541 		return (1 << sk->sk_state) & (TCPF_ESTABLISHED | TCPF_LISTEN);
542 	if (sk_is_stream_unix(sk))
543 		return (1 << sk->sk_state) & TCPF_ESTABLISHED;
544 	return true;
545 }
546 
547 static int sock_hash_update_common(struct bpf_map *map, void *key,
548 				   struct sock *sk, u64 flags);
549 
550 int sock_map_update_elem_sys(struct bpf_map *map, void *key, void *value,
551 			     u64 flags)
552 {
553 	struct socket *sock;
554 	struct sock *sk;
555 	int ret;
556 	u64 ufd;
557 
558 	if (map->value_size == sizeof(u64))
559 		ufd = *(u64 *)value;
560 	else
561 		ufd = *(u32 *)value;
562 	if (ufd > S32_MAX)
563 		return -EINVAL;
564 
565 	sock = sockfd_lookup(ufd, &ret);
566 	if (!sock)
567 		return ret;
568 	sk = sock->sk;
569 	if (!sk) {
570 		ret = -EINVAL;
571 		goto out;
572 	}
573 	if (!sock_map_sk_is_suitable(sk)) {
574 		ret = -EOPNOTSUPP;
575 		goto out;
576 	}
577 
578 	sock_map_sk_acquire(sk);
579 	if (!sock_map_sk_state_allowed(sk))
580 		ret = -EOPNOTSUPP;
581 	else if (map->map_type == BPF_MAP_TYPE_SOCKMAP)
582 		ret = sock_map_update_common(map, *(u32 *)key, sk, flags);
583 	else
584 		ret = sock_hash_update_common(map, key, sk, flags);
585 	sock_map_sk_release(sk);
586 out:
587 	sockfd_put(sock);
588 	return ret;
589 }
590 
591 static long sock_map_update_elem(struct bpf_map *map, void *key,
592 				 void *value, u64 flags)
593 {
594 	struct sock *sk = (struct sock *)value;
595 	int ret;
596 
597 	if (unlikely(!sk || !sk_fullsock(sk)))
598 		return -EINVAL;
599 
600 	if (!sock_map_sk_is_suitable(sk))
601 		return -EOPNOTSUPP;
602 
603 	local_bh_disable();
604 	bh_lock_sock(sk);
605 	if (!sock_map_sk_state_allowed(sk))
606 		ret = -EOPNOTSUPP;
607 	else if (map->map_type == BPF_MAP_TYPE_SOCKMAP)
608 		ret = sock_map_update_common(map, *(u32 *)key, sk, flags);
609 	else
610 		ret = sock_hash_update_common(map, key, sk, flags);
611 	bh_unlock_sock(sk);
612 	local_bh_enable();
613 	return ret;
614 }
615 
616 BPF_CALL_4(bpf_sock_map_update, struct bpf_sock_ops_kern *, sops,
617 	   struct bpf_map *, map, void *, key, u64, flags)
618 {
619 	WARN_ON_ONCE(!rcu_read_lock_held());
620 
621 	if (likely(sock_map_sk_is_suitable(sops->sk) &&
622 		   sock_map_op_okay(sops)))
623 		return sock_map_update_common(map, *(u32 *)key, sops->sk,
624 					      flags);
625 	return -EOPNOTSUPP;
626 }
627 
628 const struct bpf_func_proto bpf_sock_map_update_proto = {
629 	.func		= bpf_sock_map_update,
630 	.gpl_only	= false,
631 	.pkt_access	= true,
632 	.ret_type	= RET_INTEGER,
633 	.arg1_type	= ARG_PTR_TO_CTX,
634 	.arg2_type	= ARG_CONST_MAP_PTR,
635 	.arg3_type	= ARG_PTR_TO_MAP_KEY,
636 	.arg4_type	= ARG_ANYTHING,
637 };
638 
639 BPF_CALL_4(bpf_sk_redirect_map, struct sk_buff *, skb,
640 	   struct bpf_map *, map, u32, key, u64, flags)
641 {
642 	struct sock *sk;
643 
644 	if (unlikely(flags & ~(BPF_F_INGRESS)))
645 		return SK_DROP;
646 
647 	sk = __sock_map_lookup_elem(map, key);
648 	if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
649 		return SK_DROP;
650 
651 	skb_bpf_set_redir(skb, sk, flags & BPF_F_INGRESS);
652 	return SK_PASS;
653 }
654 
655 const struct bpf_func_proto bpf_sk_redirect_map_proto = {
656 	.func           = bpf_sk_redirect_map,
657 	.gpl_only       = false,
658 	.ret_type       = RET_INTEGER,
659 	.arg1_type	= ARG_PTR_TO_CTX,
660 	.arg2_type      = ARG_CONST_MAP_PTR,
661 	.arg3_type      = ARG_ANYTHING,
662 	.arg4_type      = ARG_ANYTHING,
663 };
664 
665 BPF_CALL_4(bpf_msg_redirect_map, struct sk_msg *, msg,
666 	   struct bpf_map *, map, u32, key, u64, flags)
667 {
668 	struct sock *sk;
669 
670 	if (unlikely(flags & ~(BPF_F_INGRESS)))
671 		return SK_DROP;
672 
673 	sk = __sock_map_lookup_elem(map, key);
674 	if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
675 		return SK_DROP;
676 	if (!(flags & BPF_F_INGRESS) && !sk_is_tcp(sk))
677 		return SK_DROP;
678 
679 	msg->flags = flags;
680 	msg->sk_redir = sk;
681 	return SK_PASS;
682 }
683 
684 const struct bpf_func_proto bpf_msg_redirect_map_proto = {
685 	.func           = bpf_msg_redirect_map,
686 	.gpl_only       = false,
687 	.ret_type       = RET_INTEGER,
688 	.arg1_type	= ARG_PTR_TO_CTX,
689 	.arg2_type      = ARG_CONST_MAP_PTR,
690 	.arg3_type      = ARG_ANYTHING,
691 	.arg4_type      = ARG_ANYTHING,
692 };
693 
694 struct sock_map_seq_info {
695 	struct bpf_map *map;
696 	struct sock *sk;
697 	u32 index;
698 };
699 
700 struct bpf_iter__sockmap {
701 	__bpf_md_ptr(struct bpf_iter_meta *, meta);
702 	__bpf_md_ptr(struct bpf_map *, map);
703 	__bpf_md_ptr(void *, key);
704 	__bpf_md_ptr(struct sock *, sk);
705 };
706 
707 DEFINE_BPF_ITER_FUNC(sockmap, struct bpf_iter_meta *meta,
708 		     struct bpf_map *map, void *key,
709 		     struct sock *sk)
710 
711 static void *sock_map_seq_lookup_elem(struct sock_map_seq_info *info)
712 {
713 	if (unlikely(info->index >= info->map->max_entries))
714 		return NULL;
715 
716 	info->sk = __sock_map_lookup_elem(info->map, info->index);
717 
718 	/* can't return sk directly, since that might be NULL */
719 	return info;
720 }
721 
722 static void *sock_map_seq_start(struct seq_file *seq, loff_t *pos)
723 	__acquires(rcu)
724 {
725 	struct sock_map_seq_info *info = seq->private;
726 
727 	if (*pos == 0)
728 		++*pos;
729 
730 	/* pairs with sock_map_seq_stop */
731 	rcu_read_lock();
732 	return sock_map_seq_lookup_elem(info);
733 }
734 
735 static void *sock_map_seq_next(struct seq_file *seq, void *v, loff_t *pos)
736 	__must_hold(rcu)
737 {
738 	struct sock_map_seq_info *info = seq->private;
739 
740 	++*pos;
741 	++info->index;
742 
743 	return sock_map_seq_lookup_elem(info);
744 }
745 
746 static int sock_map_seq_show(struct seq_file *seq, void *v)
747 	__must_hold(rcu)
748 {
749 	struct sock_map_seq_info *info = seq->private;
750 	struct bpf_iter__sockmap ctx = {};
751 	struct bpf_iter_meta meta;
752 	struct bpf_prog *prog;
753 
754 	meta.seq = seq;
755 	prog = bpf_iter_get_info(&meta, !v);
756 	if (!prog)
757 		return 0;
758 
759 	ctx.meta = &meta;
760 	ctx.map = info->map;
761 	if (v) {
762 		ctx.key = &info->index;
763 		ctx.sk = info->sk;
764 	}
765 
766 	return bpf_iter_run_prog(prog, &ctx);
767 }
768 
769 static void sock_map_seq_stop(struct seq_file *seq, void *v)
770 	__releases(rcu)
771 {
772 	if (!v)
773 		(void)sock_map_seq_show(seq, NULL);
774 
775 	/* pairs with sock_map_seq_start */
776 	rcu_read_unlock();
777 }
778 
779 static const struct seq_operations sock_map_seq_ops = {
780 	.start	= sock_map_seq_start,
781 	.next	= sock_map_seq_next,
782 	.stop	= sock_map_seq_stop,
783 	.show	= sock_map_seq_show,
784 };
785 
786 static int sock_map_init_seq_private(void *priv_data,
787 				     struct bpf_iter_aux_info *aux)
788 {
789 	struct sock_map_seq_info *info = priv_data;
790 
791 	bpf_map_inc_with_uref(aux->map);
792 	info->map = aux->map;
793 	return 0;
794 }
795 
796 static void sock_map_fini_seq_private(void *priv_data)
797 {
798 	struct sock_map_seq_info *info = priv_data;
799 
800 	bpf_map_put_with_uref(info->map);
801 }
802 
803 static u64 sock_map_mem_usage(const struct bpf_map *map)
804 {
805 	u64 usage = sizeof(struct bpf_stab);
806 
807 	usage += (u64)map->max_entries * sizeof(struct sock *);
808 	return usage;
809 }
810 
811 static const struct bpf_iter_seq_info sock_map_iter_seq_info = {
812 	.seq_ops		= &sock_map_seq_ops,
813 	.init_seq_private	= sock_map_init_seq_private,
814 	.fini_seq_private	= sock_map_fini_seq_private,
815 	.seq_priv_size		= sizeof(struct sock_map_seq_info),
816 };
817 
818 BTF_ID_LIST_SINGLE(sock_map_btf_ids, struct, bpf_stab)
819 const struct bpf_map_ops sock_map_ops = {
820 	.map_meta_equal		= bpf_map_meta_equal,
821 	.map_alloc		= sock_map_alloc,
822 	.map_free		= sock_map_free,
823 	.map_get_next_key	= sock_map_get_next_key,
824 	.map_lookup_elem_sys_only = sock_map_lookup_sys,
825 	.map_update_elem	= sock_map_update_elem,
826 	.map_delete_elem	= sock_map_delete_elem,
827 	.map_lookup_elem	= sock_map_lookup,
828 	.map_release_uref	= sock_map_release_progs,
829 	.map_check_btf		= map_check_no_btf,
830 	.map_mem_usage		= sock_map_mem_usage,
831 	.map_btf_id		= &sock_map_btf_ids[0],
832 	.iter_seq_info		= &sock_map_iter_seq_info,
833 };
834 
835 struct bpf_shtab_elem {
836 	struct rcu_head rcu;
837 	u32 hash;
838 	struct sock *sk;
839 	struct hlist_node node;
840 	u8 key[];
841 };
842 
843 struct bpf_shtab_bucket {
844 	struct hlist_head head;
845 	spinlock_t lock;
846 };
847 
848 struct bpf_shtab {
849 	struct bpf_map map;
850 	struct bpf_shtab_bucket *buckets;
851 	u32 buckets_num;
852 	u32 elem_size;
853 	struct sk_psock_progs progs;
854 	atomic_t count;
855 };
856 
857 static inline u32 sock_hash_bucket_hash(const void *key, u32 len)
858 {
859 	return jhash(key, len, 0);
860 }
861 
862 static struct bpf_shtab_bucket *sock_hash_select_bucket(struct bpf_shtab *htab,
863 							u32 hash)
864 {
865 	return &htab->buckets[hash & (htab->buckets_num - 1)];
866 }
867 
868 static struct bpf_shtab_elem *
869 sock_hash_lookup_elem_raw(struct hlist_head *head, u32 hash, void *key,
870 			  u32 key_size)
871 {
872 	struct bpf_shtab_elem *elem;
873 
874 	hlist_for_each_entry_rcu(elem, head, node) {
875 		if (elem->hash == hash &&
876 		    !memcmp(&elem->key, key, key_size))
877 			return elem;
878 	}
879 
880 	return NULL;
881 }
882 
883 static struct sock *__sock_hash_lookup_elem(struct bpf_map *map, void *key)
884 {
885 	struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
886 	u32 key_size = map->key_size, hash;
887 	struct bpf_shtab_bucket *bucket;
888 	struct bpf_shtab_elem *elem;
889 
890 	WARN_ON_ONCE(!rcu_read_lock_held());
891 
892 	hash = sock_hash_bucket_hash(key, key_size);
893 	bucket = sock_hash_select_bucket(htab, hash);
894 	elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size);
895 
896 	return elem ? elem->sk : NULL;
897 }
898 
899 static void sock_hash_free_elem(struct bpf_shtab *htab,
900 				struct bpf_shtab_elem *elem)
901 {
902 	atomic_dec(&htab->count);
903 	kfree_rcu(elem, rcu);
904 }
905 
906 static void sock_hash_delete_from_link(struct bpf_map *map, struct sock *sk,
907 				       void *link_raw)
908 {
909 	struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
910 	struct bpf_shtab_elem *elem_probe, *elem = link_raw;
911 	struct bpf_shtab_bucket *bucket;
912 
913 	WARN_ON_ONCE(!rcu_read_lock_held());
914 	bucket = sock_hash_select_bucket(htab, elem->hash);
915 
916 	/* elem may be deleted in parallel from the map, but access here
917 	 * is okay since it's going away only after RCU grace period.
918 	 * However, we need to check whether it's still present.
919 	 */
920 	spin_lock_bh(&bucket->lock);
921 	elem_probe = sock_hash_lookup_elem_raw(&bucket->head, elem->hash,
922 					       elem->key, map->key_size);
923 	if (elem_probe && elem_probe == elem) {
924 		hlist_del_rcu(&elem->node);
925 		sock_map_unref(elem->sk, elem);
926 		sock_hash_free_elem(htab, elem);
927 	}
928 	spin_unlock_bh(&bucket->lock);
929 }
930 
931 static long sock_hash_delete_elem(struct bpf_map *map, void *key)
932 {
933 	struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
934 	u32 hash, key_size = map->key_size;
935 	struct bpf_shtab_bucket *bucket;
936 	struct bpf_shtab_elem *elem;
937 	int ret = -ENOENT;
938 
939 	if (irqs_disabled())
940 		return -EOPNOTSUPP; /* locks here are hardirq-unsafe */
941 
942 	hash = sock_hash_bucket_hash(key, key_size);
943 	bucket = sock_hash_select_bucket(htab, hash);
944 
945 	spin_lock_bh(&bucket->lock);
946 	elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size);
947 	if (elem) {
948 		hlist_del_rcu(&elem->node);
949 		sock_map_unref(elem->sk, elem);
950 		sock_hash_free_elem(htab, elem);
951 		ret = 0;
952 	}
953 	spin_unlock_bh(&bucket->lock);
954 	return ret;
955 }
956 
957 static struct bpf_shtab_elem *sock_hash_alloc_elem(struct bpf_shtab *htab,
958 						   void *key, u32 key_size,
959 						   u32 hash, struct sock *sk,
960 						   struct bpf_shtab_elem *old)
961 {
962 	struct bpf_shtab_elem *new;
963 
964 	if (atomic_inc_return(&htab->count) > htab->map.max_entries) {
965 		if (!old) {
966 			atomic_dec(&htab->count);
967 			return ERR_PTR(-E2BIG);
968 		}
969 	}
970 
971 	new = bpf_map_kmalloc_node(&htab->map, htab->elem_size,
972 				   GFP_ATOMIC | __GFP_NOWARN,
973 				   htab->map.numa_node);
974 	if (!new) {
975 		atomic_dec(&htab->count);
976 		return ERR_PTR(-ENOMEM);
977 	}
978 	memcpy(new->key, key, key_size);
979 	new->sk = sk;
980 	new->hash = hash;
981 	return new;
982 }
983 
984 static int sock_hash_update_common(struct bpf_map *map, void *key,
985 				   struct sock *sk, u64 flags)
986 {
987 	struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
988 	u32 key_size = map->key_size, hash;
989 	struct bpf_shtab_elem *elem, *elem_new;
990 	struct bpf_shtab_bucket *bucket;
991 	struct sk_psock_link *link;
992 	struct sk_psock *psock;
993 	int ret;
994 
995 	WARN_ON_ONCE(!rcu_read_lock_held());
996 	if (unlikely(flags > BPF_EXIST))
997 		return -EINVAL;
998 
999 	link = sk_psock_init_link();
1000 	if (!link)
1001 		return -ENOMEM;
1002 
1003 	ret = sock_map_link(map, sk);
1004 	if (ret < 0)
1005 		goto out_free;
1006 
1007 	psock = sk_psock(sk);
1008 	WARN_ON_ONCE(!psock);
1009 
1010 	hash = sock_hash_bucket_hash(key, key_size);
1011 	bucket = sock_hash_select_bucket(htab, hash);
1012 
1013 	spin_lock_bh(&bucket->lock);
1014 	elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size);
1015 	if (elem && flags == BPF_NOEXIST) {
1016 		ret = -EEXIST;
1017 		goto out_unlock;
1018 	} else if (!elem && flags == BPF_EXIST) {
1019 		ret = -ENOENT;
1020 		goto out_unlock;
1021 	}
1022 
1023 	elem_new = sock_hash_alloc_elem(htab, key, key_size, hash, sk, elem);
1024 	if (IS_ERR(elem_new)) {
1025 		ret = PTR_ERR(elem_new);
1026 		goto out_unlock;
1027 	}
1028 
1029 	sock_map_add_link(psock, link, map, elem_new);
1030 	/* Add new element to the head of the list, so that
1031 	 * concurrent search will find it before old elem.
1032 	 */
1033 	hlist_add_head_rcu(&elem_new->node, &bucket->head);
1034 	if (elem) {
1035 		hlist_del_rcu(&elem->node);
1036 		sock_map_unref(elem->sk, elem);
1037 		sock_hash_free_elem(htab, elem);
1038 	}
1039 	spin_unlock_bh(&bucket->lock);
1040 	return 0;
1041 out_unlock:
1042 	spin_unlock_bh(&bucket->lock);
1043 	sk_psock_put(sk, psock);
1044 out_free:
1045 	sk_psock_free_link(link);
1046 	return ret;
1047 }
1048 
1049 static int sock_hash_get_next_key(struct bpf_map *map, void *key,
1050 				  void *key_next)
1051 {
1052 	struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
1053 	struct bpf_shtab_elem *elem, *elem_next;
1054 	u32 hash, key_size = map->key_size;
1055 	struct hlist_head *head;
1056 	int i = 0;
1057 
1058 	if (!key)
1059 		goto find_first_elem;
1060 	hash = sock_hash_bucket_hash(key, key_size);
1061 	head = &sock_hash_select_bucket(htab, hash)->head;
1062 	elem = sock_hash_lookup_elem_raw(head, hash, key, key_size);
1063 	if (!elem)
1064 		goto find_first_elem;
1065 
1066 	elem_next = hlist_entry_safe(rcu_dereference(hlist_next_rcu(&elem->node)),
1067 				     struct bpf_shtab_elem, node);
1068 	if (elem_next) {
1069 		memcpy(key_next, elem_next->key, key_size);
1070 		return 0;
1071 	}
1072 
1073 	i = hash & (htab->buckets_num - 1);
1074 	i++;
1075 find_first_elem:
1076 	for (; i < htab->buckets_num; i++) {
1077 		head = &sock_hash_select_bucket(htab, i)->head;
1078 		elem_next = hlist_entry_safe(rcu_dereference(hlist_first_rcu(head)),
1079 					     struct bpf_shtab_elem, node);
1080 		if (elem_next) {
1081 			memcpy(key_next, elem_next->key, key_size);
1082 			return 0;
1083 		}
1084 	}
1085 
1086 	return -ENOENT;
1087 }
1088 
1089 static struct bpf_map *sock_hash_alloc(union bpf_attr *attr)
1090 {
1091 	struct bpf_shtab *htab;
1092 	int i, err;
1093 
1094 	if (attr->max_entries == 0 ||
1095 	    attr->key_size    == 0 ||
1096 	    (attr->value_size != sizeof(u32) &&
1097 	     attr->value_size != sizeof(u64)) ||
1098 	    attr->map_flags & ~SOCK_CREATE_FLAG_MASK)
1099 		return ERR_PTR(-EINVAL);
1100 	if (attr->key_size > MAX_BPF_STACK)
1101 		return ERR_PTR(-E2BIG);
1102 
1103 	htab = bpf_map_area_alloc(sizeof(*htab), NUMA_NO_NODE);
1104 	if (!htab)
1105 		return ERR_PTR(-ENOMEM);
1106 
1107 	bpf_map_init_from_attr(&htab->map, attr);
1108 
1109 	htab->buckets_num = roundup_pow_of_two(htab->map.max_entries);
1110 	htab->elem_size = sizeof(struct bpf_shtab_elem) +
1111 			  round_up(htab->map.key_size, 8);
1112 	if (htab->buckets_num == 0 ||
1113 	    htab->buckets_num > U32_MAX / sizeof(struct bpf_shtab_bucket)) {
1114 		err = -EINVAL;
1115 		goto free_htab;
1116 	}
1117 
1118 	htab->buckets = bpf_map_area_alloc(htab->buckets_num *
1119 					   sizeof(struct bpf_shtab_bucket),
1120 					   htab->map.numa_node);
1121 	if (!htab->buckets) {
1122 		err = -ENOMEM;
1123 		goto free_htab;
1124 	}
1125 
1126 	for (i = 0; i < htab->buckets_num; i++) {
1127 		INIT_HLIST_HEAD(&htab->buckets[i].head);
1128 		spin_lock_init(&htab->buckets[i].lock);
1129 	}
1130 
1131 	return &htab->map;
1132 free_htab:
1133 	bpf_map_area_free(htab);
1134 	return ERR_PTR(err);
1135 }
1136 
1137 static void sock_hash_free(struct bpf_map *map)
1138 {
1139 	struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
1140 	struct bpf_shtab_bucket *bucket;
1141 	struct hlist_head unlink_list;
1142 	struct bpf_shtab_elem *elem;
1143 	struct hlist_node *node;
1144 	int i;
1145 
1146 	/* After the sync no updates or deletes will be in-flight so it
1147 	 * is safe to walk map and remove entries without risking a race
1148 	 * in EEXIST update case.
1149 	 */
1150 	synchronize_rcu();
1151 	for (i = 0; i < htab->buckets_num; i++) {
1152 		bucket = sock_hash_select_bucket(htab, i);
1153 
1154 		/* We are racing with sock_hash_delete_from_link to
1155 		 * enter the spin-lock critical section. Every socket on
1156 		 * the list is still linked to sockhash. Since link
1157 		 * exists, psock exists and holds a ref to socket. That
1158 		 * lets us to grab a socket ref too.
1159 		 */
1160 		spin_lock_bh(&bucket->lock);
1161 		hlist_for_each_entry(elem, &bucket->head, node)
1162 			sock_hold(elem->sk);
1163 		hlist_move_list(&bucket->head, &unlink_list);
1164 		spin_unlock_bh(&bucket->lock);
1165 
1166 		/* Process removed entries out of atomic context to
1167 		 * block for socket lock before deleting the psock's
1168 		 * link to sockhash.
1169 		 */
1170 		hlist_for_each_entry_safe(elem, node, &unlink_list, node) {
1171 			hlist_del(&elem->node);
1172 			lock_sock(elem->sk);
1173 			rcu_read_lock();
1174 			sock_map_unref(elem->sk, elem);
1175 			rcu_read_unlock();
1176 			release_sock(elem->sk);
1177 			sock_put(elem->sk);
1178 			sock_hash_free_elem(htab, elem);
1179 		}
1180 	}
1181 
1182 	/* wait for psock readers accessing its map link */
1183 	synchronize_rcu();
1184 
1185 	bpf_map_area_free(htab->buckets);
1186 	bpf_map_area_free(htab);
1187 }
1188 
1189 static void *sock_hash_lookup_sys(struct bpf_map *map, void *key)
1190 {
1191 	struct sock *sk;
1192 
1193 	if (map->value_size != sizeof(u64))
1194 		return ERR_PTR(-ENOSPC);
1195 
1196 	sk = __sock_hash_lookup_elem(map, key);
1197 	if (!sk)
1198 		return ERR_PTR(-ENOENT);
1199 
1200 	__sock_gen_cookie(sk);
1201 	return &sk->sk_cookie;
1202 }
1203 
1204 static void *sock_hash_lookup(struct bpf_map *map, void *key)
1205 {
1206 	struct sock *sk;
1207 
1208 	sk = __sock_hash_lookup_elem(map, key);
1209 	if (!sk)
1210 		return NULL;
1211 	if (sk_is_refcounted(sk) && !refcount_inc_not_zero(&sk->sk_refcnt))
1212 		return NULL;
1213 	return sk;
1214 }
1215 
1216 static void sock_hash_release_progs(struct bpf_map *map)
1217 {
1218 	psock_progs_drop(&container_of(map, struct bpf_shtab, map)->progs);
1219 }
1220 
1221 BPF_CALL_4(bpf_sock_hash_update, struct bpf_sock_ops_kern *, sops,
1222 	   struct bpf_map *, map, void *, key, u64, flags)
1223 {
1224 	WARN_ON_ONCE(!rcu_read_lock_held());
1225 
1226 	if (likely(sock_map_sk_is_suitable(sops->sk) &&
1227 		   sock_map_op_okay(sops)))
1228 		return sock_hash_update_common(map, key, sops->sk, flags);
1229 	return -EOPNOTSUPP;
1230 }
1231 
1232 const struct bpf_func_proto bpf_sock_hash_update_proto = {
1233 	.func		= bpf_sock_hash_update,
1234 	.gpl_only	= false,
1235 	.pkt_access	= true,
1236 	.ret_type	= RET_INTEGER,
1237 	.arg1_type	= ARG_PTR_TO_CTX,
1238 	.arg2_type	= ARG_CONST_MAP_PTR,
1239 	.arg3_type	= ARG_PTR_TO_MAP_KEY,
1240 	.arg4_type	= ARG_ANYTHING,
1241 };
1242 
1243 BPF_CALL_4(bpf_sk_redirect_hash, struct sk_buff *, skb,
1244 	   struct bpf_map *, map, void *, key, u64, flags)
1245 {
1246 	struct sock *sk;
1247 
1248 	if (unlikely(flags & ~(BPF_F_INGRESS)))
1249 		return SK_DROP;
1250 
1251 	sk = __sock_hash_lookup_elem(map, key);
1252 	if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
1253 		return SK_DROP;
1254 
1255 	skb_bpf_set_redir(skb, sk, flags & BPF_F_INGRESS);
1256 	return SK_PASS;
1257 }
1258 
1259 const struct bpf_func_proto bpf_sk_redirect_hash_proto = {
1260 	.func           = bpf_sk_redirect_hash,
1261 	.gpl_only       = false,
1262 	.ret_type       = RET_INTEGER,
1263 	.arg1_type	= ARG_PTR_TO_CTX,
1264 	.arg2_type      = ARG_CONST_MAP_PTR,
1265 	.arg3_type      = ARG_PTR_TO_MAP_KEY,
1266 	.arg4_type      = ARG_ANYTHING,
1267 };
1268 
1269 BPF_CALL_4(bpf_msg_redirect_hash, struct sk_msg *, msg,
1270 	   struct bpf_map *, map, void *, key, u64, flags)
1271 {
1272 	struct sock *sk;
1273 
1274 	if (unlikely(flags & ~(BPF_F_INGRESS)))
1275 		return SK_DROP;
1276 
1277 	sk = __sock_hash_lookup_elem(map, key);
1278 	if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
1279 		return SK_DROP;
1280 	if (!(flags & BPF_F_INGRESS) && !sk_is_tcp(sk))
1281 		return SK_DROP;
1282 
1283 	msg->flags = flags;
1284 	msg->sk_redir = sk;
1285 	return SK_PASS;
1286 }
1287 
1288 const struct bpf_func_proto bpf_msg_redirect_hash_proto = {
1289 	.func           = bpf_msg_redirect_hash,
1290 	.gpl_only       = false,
1291 	.ret_type       = RET_INTEGER,
1292 	.arg1_type	= ARG_PTR_TO_CTX,
1293 	.arg2_type      = ARG_CONST_MAP_PTR,
1294 	.arg3_type      = ARG_PTR_TO_MAP_KEY,
1295 	.arg4_type      = ARG_ANYTHING,
1296 };
1297 
1298 struct sock_hash_seq_info {
1299 	struct bpf_map *map;
1300 	struct bpf_shtab *htab;
1301 	u32 bucket_id;
1302 };
1303 
1304 static void *sock_hash_seq_find_next(struct sock_hash_seq_info *info,
1305 				     struct bpf_shtab_elem *prev_elem)
1306 {
1307 	const struct bpf_shtab *htab = info->htab;
1308 	struct bpf_shtab_bucket *bucket;
1309 	struct bpf_shtab_elem *elem;
1310 	struct hlist_node *node;
1311 
1312 	/* try to find next elem in the same bucket */
1313 	if (prev_elem) {
1314 		node = rcu_dereference(hlist_next_rcu(&prev_elem->node));
1315 		elem = hlist_entry_safe(node, struct bpf_shtab_elem, node);
1316 		if (elem)
1317 			return elem;
1318 
1319 		/* no more elements, continue in the next bucket */
1320 		info->bucket_id++;
1321 	}
1322 
1323 	for (; info->bucket_id < htab->buckets_num; info->bucket_id++) {
1324 		bucket = &htab->buckets[info->bucket_id];
1325 		node = rcu_dereference(hlist_first_rcu(&bucket->head));
1326 		elem = hlist_entry_safe(node, struct bpf_shtab_elem, node);
1327 		if (elem)
1328 			return elem;
1329 	}
1330 
1331 	return NULL;
1332 }
1333 
1334 static void *sock_hash_seq_start(struct seq_file *seq, loff_t *pos)
1335 	__acquires(rcu)
1336 {
1337 	struct sock_hash_seq_info *info = seq->private;
1338 
1339 	if (*pos == 0)
1340 		++*pos;
1341 
1342 	/* pairs with sock_hash_seq_stop */
1343 	rcu_read_lock();
1344 	return sock_hash_seq_find_next(info, NULL);
1345 }
1346 
1347 static void *sock_hash_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1348 	__must_hold(rcu)
1349 {
1350 	struct sock_hash_seq_info *info = seq->private;
1351 
1352 	++*pos;
1353 	return sock_hash_seq_find_next(info, v);
1354 }
1355 
1356 static int sock_hash_seq_show(struct seq_file *seq, void *v)
1357 	__must_hold(rcu)
1358 {
1359 	struct sock_hash_seq_info *info = seq->private;
1360 	struct bpf_iter__sockmap ctx = {};
1361 	struct bpf_shtab_elem *elem = v;
1362 	struct bpf_iter_meta meta;
1363 	struct bpf_prog *prog;
1364 
1365 	meta.seq = seq;
1366 	prog = bpf_iter_get_info(&meta, !elem);
1367 	if (!prog)
1368 		return 0;
1369 
1370 	ctx.meta = &meta;
1371 	ctx.map = info->map;
1372 	if (elem) {
1373 		ctx.key = elem->key;
1374 		ctx.sk = elem->sk;
1375 	}
1376 
1377 	return bpf_iter_run_prog(prog, &ctx);
1378 }
1379 
1380 static void sock_hash_seq_stop(struct seq_file *seq, void *v)
1381 	__releases(rcu)
1382 {
1383 	if (!v)
1384 		(void)sock_hash_seq_show(seq, NULL);
1385 
1386 	/* pairs with sock_hash_seq_start */
1387 	rcu_read_unlock();
1388 }
1389 
1390 static const struct seq_operations sock_hash_seq_ops = {
1391 	.start	= sock_hash_seq_start,
1392 	.next	= sock_hash_seq_next,
1393 	.stop	= sock_hash_seq_stop,
1394 	.show	= sock_hash_seq_show,
1395 };
1396 
1397 static int sock_hash_init_seq_private(void *priv_data,
1398 				      struct bpf_iter_aux_info *aux)
1399 {
1400 	struct sock_hash_seq_info *info = priv_data;
1401 
1402 	bpf_map_inc_with_uref(aux->map);
1403 	info->map = aux->map;
1404 	info->htab = container_of(aux->map, struct bpf_shtab, map);
1405 	return 0;
1406 }
1407 
1408 static void sock_hash_fini_seq_private(void *priv_data)
1409 {
1410 	struct sock_hash_seq_info *info = priv_data;
1411 
1412 	bpf_map_put_with_uref(info->map);
1413 }
1414 
1415 static u64 sock_hash_mem_usage(const struct bpf_map *map)
1416 {
1417 	struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map);
1418 	u64 usage = sizeof(*htab);
1419 
1420 	usage += htab->buckets_num * sizeof(struct bpf_shtab_bucket);
1421 	usage += atomic_read(&htab->count) * (u64)htab->elem_size;
1422 	return usage;
1423 }
1424 
1425 static const struct bpf_iter_seq_info sock_hash_iter_seq_info = {
1426 	.seq_ops		= &sock_hash_seq_ops,
1427 	.init_seq_private	= sock_hash_init_seq_private,
1428 	.fini_seq_private	= sock_hash_fini_seq_private,
1429 	.seq_priv_size		= sizeof(struct sock_hash_seq_info),
1430 };
1431 
1432 BTF_ID_LIST_SINGLE(sock_hash_map_btf_ids, struct, bpf_shtab)
1433 const struct bpf_map_ops sock_hash_ops = {
1434 	.map_meta_equal		= bpf_map_meta_equal,
1435 	.map_alloc		= sock_hash_alloc,
1436 	.map_free		= sock_hash_free,
1437 	.map_get_next_key	= sock_hash_get_next_key,
1438 	.map_update_elem	= sock_map_update_elem,
1439 	.map_delete_elem	= sock_hash_delete_elem,
1440 	.map_lookup_elem	= sock_hash_lookup,
1441 	.map_lookup_elem_sys_only = sock_hash_lookup_sys,
1442 	.map_release_uref	= sock_hash_release_progs,
1443 	.map_check_btf		= map_check_no_btf,
1444 	.map_mem_usage		= sock_hash_mem_usage,
1445 	.map_btf_id		= &sock_hash_map_btf_ids[0],
1446 	.iter_seq_info		= &sock_hash_iter_seq_info,
1447 };
1448 
1449 static struct sk_psock_progs *sock_map_progs(struct bpf_map *map)
1450 {
1451 	switch (map->map_type) {
1452 	case BPF_MAP_TYPE_SOCKMAP:
1453 		return &container_of(map, struct bpf_stab, map)->progs;
1454 	case BPF_MAP_TYPE_SOCKHASH:
1455 		return &container_of(map, struct bpf_shtab, map)->progs;
1456 	default:
1457 		break;
1458 	}
1459 
1460 	return NULL;
1461 }
1462 
1463 static int sock_map_prog_lookup(struct bpf_map *map, struct bpf_prog ***pprog,
1464 				u32 which)
1465 {
1466 	struct sk_psock_progs *progs = sock_map_progs(map);
1467 
1468 	if (!progs)
1469 		return -EOPNOTSUPP;
1470 
1471 	switch (which) {
1472 	case BPF_SK_MSG_VERDICT:
1473 		*pprog = &progs->msg_parser;
1474 		break;
1475 #if IS_ENABLED(CONFIG_BPF_STREAM_PARSER)
1476 	case BPF_SK_SKB_STREAM_PARSER:
1477 		*pprog = &progs->stream_parser;
1478 		break;
1479 #endif
1480 	case BPF_SK_SKB_STREAM_VERDICT:
1481 		if (progs->skb_verdict)
1482 			return -EBUSY;
1483 		*pprog = &progs->stream_verdict;
1484 		break;
1485 	case BPF_SK_SKB_VERDICT:
1486 		if (progs->stream_verdict)
1487 			return -EBUSY;
1488 		*pprog = &progs->skb_verdict;
1489 		break;
1490 	default:
1491 		return -EOPNOTSUPP;
1492 	}
1493 
1494 	return 0;
1495 }
1496 
1497 static int sock_map_prog_update(struct bpf_map *map, struct bpf_prog *prog,
1498 				struct bpf_prog *old, u32 which)
1499 {
1500 	struct bpf_prog **pprog;
1501 	int ret;
1502 
1503 	ret = sock_map_prog_lookup(map, &pprog, which);
1504 	if (ret)
1505 		return ret;
1506 
1507 	if (old)
1508 		return psock_replace_prog(pprog, prog, old);
1509 
1510 	psock_set_prog(pprog, prog);
1511 	return 0;
1512 }
1513 
1514 int sock_map_bpf_prog_query(const union bpf_attr *attr,
1515 			    union bpf_attr __user *uattr)
1516 {
1517 	__u32 __user *prog_ids = u64_to_user_ptr(attr->query.prog_ids);
1518 	u32 prog_cnt = 0, flags = 0, ufd = attr->target_fd;
1519 	struct bpf_prog **pprog;
1520 	struct bpf_prog *prog;
1521 	struct bpf_map *map;
1522 	struct fd f;
1523 	u32 id = 0;
1524 	int ret;
1525 
1526 	if (attr->query.query_flags)
1527 		return -EINVAL;
1528 
1529 	f = fdget(ufd);
1530 	map = __bpf_map_get(f);
1531 	if (IS_ERR(map))
1532 		return PTR_ERR(map);
1533 
1534 	rcu_read_lock();
1535 
1536 	ret = sock_map_prog_lookup(map, &pprog, attr->query.attach_type);
1537 	if (ret)
1538 		goto end;
1539 
1540 	prog = *pprog;
1541 	prog_cnt = !prog ? 0 : 1;
1542 
1543 	if (!attr->query.prog_cnt || !prog_ids || !prog_cnt)
1544 		goto end;
1545 
1546 	/* we do not hold the refcnt, the bpf prog may be released
1547 	 * asynchronously and the id would be set to 0.
1548 	 */
1549 	id = data_race(prog->aux->id);
1550 	if (id == 0)
1551 		prog_cnt = 0;
1552 
1553 end:
1554 	rcu_read_unlock();
1555 
1556 	if (copy_to_user(&uattr->query.attach_flags, &flags, sizeof(flags)) ||
1557 	    (id != 0 && copy_to_user(prog_ids, &id, sizeof(u32))) ||
1558 	    copy_to_user(&uattr->query.prog_cnt, &prog_cnt, sizeof(prog_cnt)))
1559 		ret = -EFAULT;
1560 
1561 	fdput(f);
1562 	return ret;
1563 }
1564 
1565 static void sock_map_unlink(struct sock *sk, struct sk_psock_link *link)
1566 {
1567 	switch (link->map->map_type) {
1568 	case BPF_MAP_TYPE_SOCKMAP:
1569 		return sock_map_delete_from_link(link->map, sk,
1570 						 link->link_raw);
1571 	case BPF_MAP_TYPE_SOCKHASH:
1572 		return sock_hash_delete_from_link(link->map, sk,
1573 						  link->link_raw);
1574 	default:
1575 		break;
1576 	}
1577 }
1578 
1579 static void sock_map_remove_links(struct sock *sk, struct sk_psock *psock)
1580 {
1581 	struct sk_psock_link *link;
1582 
1583 	while ((link = sk_psock_link_pop(psock))) {
1584 		sock_map_unlink(sk, link);
1585 		sk_psock_free_link(link);
1586 	}
1587 }
1588 
1589 void sock_map_unhash(struct sock *sk)
1590 {
1591 	void (*saved_unhash)(struct sock *sk);
1592 	struct sk_psock *psock;
1593 
1594 	rcu_read_lock();
1595 	psock = sk_psock(sk);
1596 	if (unlikely(!psock)) {
1597 		rcu_read_unlock();
1598 		saved_unhash = READ_ONCE(sk->sk_prot)->unhash;
1599 	} else {
1600 		saved_unhash = psock->saved_unhash;
1601 		sock_map_remove_links(sk, psock);
1602 		rcu_read_unlock();
1603 	}
1604 	if (WARN_ON_ONCE(saved_unhash == sock_map_unhash))
1605 		return;
1606 	if (saved_unhash)
1607 		saved_unhash(sk);
1608 }
1609 EXPORT_SYMBOL_GPL(sock_map_unhash);
1610 
1611 void sock_map_destroy(struct sock *sk)
1612 {
1613 	void (*saved_destroy)(struct sock *sk);
1614 	struct sk_psock *psock;
1615 
1616 	rcu_read_lock();
1617 	psock = sk_psock_get(sk);
1618 	if (unlikely(!psock)) {
1619 		rcu_read_unlock();
1620 		saved_destroy = READ_ONCE(sk->sk_prot)->destroy;
1621 	} else {
1622 		saved_destroy = psock->saved_destroy;
1623 		sock_map_remove_links(sk, psock);
1624 		rcu_read_unlock();
1625 		sk_psock_stop(psock);
1626 		sk_psock_put(sk, psock);
1627 	}
1628 	if (WARN_ON_ONCE(saved_destroy == sock_map_destroy))
1629 		return;
1630 	if (saved_destroy)
1631 		saved_destroy(sk);
1632 }
1633 EXPORT_SYMBOL_GPL(sock_map_destroy);
1634 
1635 void sock_map_close(struct sock *sk, long timeout)
1636 {
1637 	void (*saved_close)(struct sock *sk, long timeout);
1638 	struct sk_psock *psock;
1639 
1640 	lock_sock(sk);
1641 	rcu_read_lock();
1642 	psock = sk_psock_get(sk);
1643 	if (unlikely(!psock)) {
1644 		rcu_read_unlock();
1645 		release_sock(sk);
1646 		saved_close = READ_ONCE(sk->sk_prot)->close;
1647 	} else {
1648 		saved_close = psock->saved_close;
1649 		sock_map_remove_links(sk, psock);
1650 		rcu_read_unlock();
1651 		sk_psock_stop(psock);
1652 		release_sock(sk);
1653 		cancel_delayed_work_sync(&psock->work);
1654 		sk_psock_put(sk, psock);
1655 	}
1656 
1657 	/* Make sure we do not recurse. This is a bug.
1658 	 * Leak the socket instead of crashing on a stack overflow.
1659 	 */
1660 	if (WARN_ON_ONCE(saved_close == sock_map_close))
1661 		return;
1662 	saved_close(sk, timeout);
1663 }
1664 EXPORT_SYMBOL_GPL(sock_map_close);
1665 
1666 static int sock_map_iter_attach_target(struct bpf_prog *prog,
1667 				       union bpf_iter_link_info *linfo,
1668 				       struct bpf_iter_aux_info *aux)
1669 {
1670 	struct bpf_map *map;
1671 	int err = -EINVAL;
1672 
1673 	if (!linfo->map.map_fd)
1674 		return -EBADF;
1675 
1676 	map = bpf_map_get_with_uref(linfo->map.map_fd);
1677 	if (IS_ERR(map))
1678 		return PTR_ERR(map);
1679 
1680 	if (map->map_type != BPF_MAP_TYPE_SOCKMAP &&
1681 	    map->map_type != BPF_MAP_TYPE_SOCKHASH)
1682 		goto put_map;
1683 
1684 	if (prog->aux->max_rdonly_access > map->key_size) {
1685 		err = -EACCES;
1686 		goto put_map;
1687 	}
1688 
1689 	aux->map = map;
1690 	return 0;
1691 
1692 put_map:
1693 	bpf_map_put_with_uref(map);
1694 	return err;
1695 }
1696 
1697 static void sock_map_iter_detach_target(struct bpf_iter_aux_info *aux)
1698 {
1699 	bpf_map_put_with_uref(aux->map);
1700 }
1701 
1702 static struct bpf_iter_reg sock_map_iter_reg = {
1703 	.target			= "sockmap",
1704 	.attach_target		= sock_map_iter_attach_target,
1705 	.detach_target		= sock_map_iter_detach_target,
1706 	.show_fdinfo		= bpf_iter_map_show_fdinfo,
1707 	.fill_link_info		= bpf_iter_map_fill_link_info,
1708 	.ctx_arg_info_size	= 2,
1709 	.ctx_arg_info		= {
1710 		{ offsetof(struct bpf_iter__sockmap, key),
1711 		  PTR_TO_BUF | PTR_MAYBE_NULL | MEM_RDONLY },
1712 		{ offsetof(struct bpf_iter__sockmap, sk),
1713 		  PTR_TO_BTF_ID_OR_NULL },
1714 	},
1715 };
1716 
1717 static int __init bpf_sockmap_iter_init(void)
1718 {
1719 	sock_map_iter_reg.ctx_arg_info[1].btf_id =
1720 		btf_sock_ids[BTF_SOCK_TYPE_SOCK];
1721 	return bpf_iter_reg_target(&sock_map_iter_reg);
1722 }
1723 late_initcall(bpf_sockmap_iter_init);
1724