xref: /openbmc/linux/net/core/sock_map.c (revision 4c5a116a)
1 // SPDX-License-Identifier: GPL-2.0
2 /* Copyright (c) 2017 - 2018 Covalent IO, Inc. http://covalent.io */
3 
4 #include <linux/bpf.h>
5 #include <linux/filter.h>
6 #include <linux/errno.h>
7 #include <linux/file.h>
8 #include <linux/net.h>
9 #include <linux/workqueue.h>
10 #include <linux/skmsg.h>
11 #include <linux/list.h>
12 #include <linux/jhash.h>
13 #include <linux/sock_diag.h>
14 #include <net/udp.h>
15 
16 struct bpf_stab {
17 	struct bpf_map map;
18 	struct sock **sks;
19 	struct sk_psock_progs progs;
20 	raw_spinlock_t lock;
21 };
22 
23 #define SOCK_CREATE_FLAG_MASK				\
24 	(BPF_F_NUMA_NODE | BPF_F_RDONLY | BPF_F_WRONLY)
25 
26 static struct bpf_map *sock_map_alloc(union bpf_attr *attr)
27 {
28 	struct bpf_stab *stab;
29 	u64 cost;
30 	int err;
31 
32 	if (!capable(CAP_NET_ADMIN))
33 		return ERR_PTR(-EPERM);
34 	if (attr->max_entries == 0 ||
35 	    attr->key_size    != 4 ||
36 	    (attr->value_size != sizeof(u32) &&
37 	     attr->value_size != sizeof(u64)) ||
38 	    attr->map_flags & ~SOCK_CREATE_FLAG_MASK)
39 		return ERR_PTR(-EINVAL);
40 
41 	stab = kzalloc(sizeof(*stab), GFP_USER);
42 	if (!stab)
43 		return ERR_PTR(-ENOMEM);
44 
45 	bpf_map_init_from_attr(&stab->map, attr);
46 	raw_spin_lock_init(&stab->lock);
47 
48 	/* Make sure page count doesn't overflow. */
49 	cost = (u64) stab->map.max_entries * sizeof(struct sock *);
50 	err = bpf_map_charge_init(&stab->map.memory, cost);
51 	if (err)
52 		goto free_stab;
53 
54 	stab->sks = bpf_map_area_alloc(stab->map.max_entries *
55 				       sizeof(struct sock *),
56 				       stab->map.numa_node);
57 	if (stab->sks)
58 		return &stab->map;
59 	err = -ENOMEM;
60 	bpf_map_charge_finish(&stab->map.memory);
61 free_stab:
62 	kfree(stab);
63 	return ERR_PTR(err);
64 }
65 
66 int sock_map_get_from_fd(const union bpf_attr *attr, struct bpf_prog *prog)
67 {
68 	u32 ufd = attr->target_fd;
69 	struct bpf_map *map;
70 	struct fd f;
71 	int ret;
72 
73 	if (attr->attach_flags || attr->replace_bpf_fd)
74 		return -EINVAL;
75 
76 	f = fdget(ufd);
77 	map = __bpf_map_get(f);
78 	if (IS_ERR(map))
79 		return PTR_ERR(map);
80 	ret = sock_map_prog_update(map, prog, NULL, attr->attach_type);
81 	fdput(f);
82 	return ret;
83 }
84 
85 int sock_map_prog_detach(const union bpf_attr *attr, enum bpf_prog_type ptype)
86 {
87 	u32 ufd = attr->target_fd;
88 	struct bpf_prog *prog;
89 	struct bpf_map *map;
90 	struct fd f;
91 	int ret;
92 
93 	if (attr->attach_flags || attr->replace_bpf_fd)
94 		return -EINVAL;
95 
96 	f = fdget(ufd);
97 	map = __bpf_map_get(f);
98 	if (IS_ERR(map))
99 		return PTR_ERR(map);
100 
101 	prog = bpf_prog_get(attr->attach_bpf_fd);
102 	if (IS_ERR(prog)) {
103 		ret = PTR_ERR(prog);
104 		goto put_map;
105 	}
106 
107 	if (prog->type != ptype) {
108 		ret = -EINVAL;
109 		goto put_prog;
110 	}
111 
112 	ret = sock_map_prog_update(map, NULL, prog, attr->attach_type);
113 put_prog:
114 	bpf_prog_put(prog);
115 put_map:
116 	fdput(f);
117 	return ret;
118 }
119 
120 static void sock_map_sk_acquire(struct sock *sk)
121 	__acquires(&sk->sk_lock.slock)
122 {
123 	lock_sock(sk);
124 	preempt_disable();
125 	rcu_read_lock();
126 }
127 
128 static void sock_map_sk_release(struct sock *sk)
129 	__releases(&sk->sk_lock.slock)
130 {
131 	rcu_read_unlock();
132 	preempt_enable();
133 	release_sock(sk);
134 }
135 
136 static void sock_map_add_link(struct sk_psock *psock,
137 			      struct sk_psock_link *link,
138 			      struct bpf_map *map, void *link_raw)
139 {
140 	link->link_raw = link_raw;
141 	link->map = map;
142 	spin_lock_bh(&psock->link_lock);
143 	list_add_tail(&link->list, &psock->link);
144 	spin_unlock_bh(&psock->link_lock);
145 }
146 
147 static void sock_map_del_link(struct sock *sk,
148 			      struct sk_psock *psock, void *link_raw)
149 {
150 	struct sk_psock_link *link, *tmp;
151 	bool strp_stop = false;
152 
153 	spin_lock_bh(&psock->link_lock);
154 	list_for_each_entry_safe(link, tmp, &psock->link, list) {
155 		if (link->link_raw == link_raw) {
156 			struct bpf_map *map = link->map;
157 			struct bpf_stab *stab = container_of(map, struct bpf_stab,
158 							     map);
159 			if (psock->parser.enabled && stab->progs.skb_parser)
160 				strp_stop = true;
161 			list_del(&link->list);
162 			sk_psock_free_link(link);
163 		}
164 	}
165 	spin_unlock_bh(&psock->link_lock);
166 	if (strp_stop) {
167 		write_lock_bh(&sk->sk_callback_lock);
168 		sk_psock_stop_strp(sk, psock);
169 		write_unlock_bh(&sk->sk_callback_lock);
170 	}
171 }
172 
173 static void sock_map_unref(struct sock *sk, void *link_raw)
174 {
175 	struct sk_psock *psock = sk_psock(sk);
176 
177 	if (likely(psock)) {
178 		sock_map_del_link(sk, psock, link_raw);
179 		sk_psock_put(sk, psock);
180 	}
181 }
182 
183 static int sock_map_init_proto(struct sock *sk, struct sk_psock *psock)
184 {
185 	struct proto *prot;
186 
187 	sock_owned_by_me(sk);
188 
189 	switch (sk->sk_type) {
190 	case SOCK_STREAM:
191 		prot = tcp_bpf_get_proto(sk, psock);
192 		break;
193 
194 	case SOCK_DGRAM:
195 		prot = udp_bpf_get_proto(sk, psock);
196 		break;
197 
198 	default:
199 		return -EINVAL;
200 	}
201 
202 	if (IS_ERR(prot))
203 		return PTR_ERR(prot);
204 
205 	sk_psock_update_proto(sk, psock, prot);
206 	return 0;
207 }
208 
209 static struct sk_psock *sock_map_psock_get_checked(struct sock *sk)
210 {
211 	struct sk_psock *psock;
212 
213 	rcu_read_lock();
214 	psock = sk_psock(sk);
215 	if (psock) {
216 		if (sk->sk_prot->close != sock_map_close) {
217 			psock = ERR_PTR(-EBUSY);
218 			goto out;
219 		}
220 
221 		if (!refcount_inc_not_zero(&psock->refcnt))
222 			psock = ERR_PTR(-EBUSY);
223 	}
224 out:
225 	rcu_read_unlock();
226 	return psock;
227 }
228 
229 static int sock_map_link(struct bpf_map *map, struct sk_psock_progs *progs,
230 			 struct sock *sk)
231 {
232 	struct bpf_prog *msg_parser, *skb_parser, *skb_verdict;
233 	struct sk_psock *psock;
234 	bool skb_progs;
235 	int ret;
236 
237 	skb_verdict = READ_ONCE(progs->skb_verdict);
238 	skb_parser = READ_ONCE(progs->skb_parser);
239 	skb_progs = skb_parser && skb_verdict;
240 	if (skb_progs) {
241 		skb_verdict = bpf_prog_inc_not_zero(skb_verdict);
242 		if (IS_ERR(skb_verdict))
243 			return PTR_ERR(skb_verdict);
244 		skb_parser = bpf_prog_inc_not_zero(skb_parser);
245 		if (IS_ERR(skb_parser)) {
246 			bpf_prog_put(skb_verdict);
247 			return PTR_ERR(skb_parser);
248 		}
249 	}
250 
251 	msg_parser = READ_ONCE(progs->msg_parser);
252 	if (msg_parser) {
253 		msg_parser = bpf_prog_inc_not_zero(msg_parser);
254 		if (IS_ERR(msg_parser)) {
255 			ret = PTR_ERR(msg_parser);
256 			goto out;
257 		}
258 	}
259 
260 	psock = sock_map_psock_get_checked(sk);
261 	if (IS_ERR(psock)) {
262 		ret = PTR_ERR(psock);
263 		goto out_progs;
264 	}
265 
266 	if (psock) {
267 		if ((msg_parser && READ_ONCE(psock->progs.msg_parser)) ||
268 		    (skb_progs  && READ_ONCE(psock->progs.skb_parser))) {
269 			sk_psock_put(sk, psock);
270 			ret = -EBUSY;
271 			goto out_progs;
272 		}
273 	} else {
274 		psock = sk_psock_init(sk, map->numa_node);
275 		if (!psock) {
276 			ret = -ENOMEM;
277 			goto out_progs;
278 		}
279 	}
280 
281 	if (msg_parser)
282 		psock_set_prog(&psock->progs.msg_parser, msg_parser);
283 
284 	ret = sock_map_init_proto(sk, psock);
285 	if (ret < 0)
286 		goto out_drop;
287 
288 	write_lock_bh(&sk->sk_callback_lock);
289 	if (skb_progs && !psock->parser.enabled) {
290 		ret = sk_psock_init_strp(sk, psock);
291 		if (ret) {
292 			write_unlock_bh(&sk->sk_callback_lock);
293 			goto out_drop;
294 		}
295 		psock_set_prog(&psock->progs.skb_verdict, skb_verdict);
296 		psock_set_prog(&psock->progs.skb_parser, skb_parser);
297 		sk_psock_start_strp(sk, psock);
298 	}
299 	write_unlock_bh(&sk->sk_callback_lock);
300 	return 0;
301 out_drop:
302 	sk_psock_put(sk, psock);
303 out_progs:
304 	if (msg_parser)
305 		bpf_prog_put(msg_parser);
306 out:
307 	if (skb_progs) {
308 		bpf_prog_put(skb_verdict);
309 		bpf_prog_put(skb_parser);
310 	}
311 	return ret;
312 }
313 
314 static int sock_map_link_no_progs(struct bpf_map *map, struct sock *sk)
315 {
316 	struct sk_psock *psock;
317 	int ret;
318 
319 	psock = sock_map_psock_get_checked(sk);
320 	if (IS_ERR(psock))
321 		return PTR_ERR(psock);
322 
323 	if (!psock) {
324 		psock = sk_psock_init(sk, map->numa_node);
325 		if (!psock)
326 			return -ENOMEM;
327 	}
328 
329 	ret = sock_map_init_proto(sk, psock);
330 	if (ret < 0)
331 		sk_psock_put(sk, psock);
332 	return ret;
333 }
334 
335 static void sock_map_free(struct bpf_map *map)
336 {
337 	struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
338 	int i;
339 
340 	/* After the sync no updates or deletes will be in-flight so it
341 	 * is safe to walk map and remove entries without risking a race
342 	 * in EEXIST update case.
343 	 */
344 	synchronize_rcu();
345 	for (i = 0; i < stab->map.max_entries; i++) {
346 		struct sock **psk = &stab->sks[i];
347 		struct sock *sk;
348 
349 		sk = xchg(psk, NULL);
350 		if (sk) {
351 			lock_sock(sk);
352 			rcu_read_lock();
353 			sock_map_unref(sk, psk);
354 			rcu_read_unlock();
355 			release_sock(sk);
356 		}
357 	}
358 
359 	/* wait for psock readers accessing its map link */
360 	synchronize_rcu();
361 
362 	bpf_map_area_free(stab->sks);
363 	kfree(stab);
364 }
365 
366 static void sock_map_release_progs(struct bpf_map *map)
367 {
368 	psock_progs_drop(&container_of(map, struct bpf_stab, map)->progs);
369 }
370 
371 static struct sock *__sock_map_lookup_elem(struct bpf_map *map, u32 key)
372 {
373 	struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
374 
375 	WARN_ON_ONCE(!rcu_read_lock_held());
376 
377 	if (unlikely(key >= map->max_entries))
378 		return NULL;
379 	return READ_ONCE(stab->sks[key]);
380 }
381 
382 static void *sock_map_lookup(struct bpf_map *map, void *key)
383 {
384 	struct sock *sk;
385 
386 	sk = __sock_map_lookup_elem(map, *(u32 *)key);
387 	if (!sk || !sk_fullsock(sk))
388 		return NULL;
389 	if (sk_is_refcounted(sk) && !refcount_inc_not_zero(&sk->sk_refcnt))
390 		return NULL;
391 	return sk;
392 }
393 
394 static void *sock_map_lookup_sys(struct bpf_map *map, void *key)
395 {
396 	struct sock *sk;
397 
398 	if (map->value_size != sizeof(u64))
399 		return ERR_PTR(-ENOSPC);
400 
401 	sk = __sock_map_lookup_elem(map, *(u32 *)key);
402 	if (!sk)
403 		return ERR_PTR(-ENOENT);
404 
405 	sock_gen_cookie(sk);
406 	return &sk->sk_cookie;
407 }
408 
409 static int __sock_map_delete(struct bpf_stab *stab, struct sock *sk_test,
410 			     struct sock **psk)
411 {
412 	struct sock *sk;
413 	int err = 0;
414 
415 	raw_spin_lock_bh(&stab->lock);
416 	sk = *psk;
417 	if (!sk_test || sk_test == sk)
418 		sk = xchg(psk, NULL);
419 
420 	if (likely(sk))
421 		sock_map_unref(sk, psk);
422 	else
423 		err = -EINVAL;
424 
425 	raw_spin_unlock_bh(&stab->lock);
426 	return err;
427 }
428 
429 static void sock_map_delete_from_link(struct bpf_map *map, struct sock *sk,
430 				      void *link_raw)
431 {
432 	struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
433 
434 	__sock_map_delete(stab, sk, link_raw);
435 }
436 
437 static int sock_map_delete_elem(struct bpf_map *map, void *key)
438 {
439 	struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
440 	u32 i = *(u32 *)key;
441 	struct sock **psk;
442 
443 	if (unlikely(i >= map->max_entries))
444 		return -EINVAL;
445 
446 	psk = &stab->sks[i];
447 	return __sock_map_delete(stab, NULL, psk);
448 }
449 
450 static int sock_map_get_next_key(struct bpf_map *map, void *key, void *next)
451 {
452 	struct bpf_stab *stab = container_of(map, struct bpf_stab, map);
453 	u32 i = key ? *(u32 *)key : U32_MAX;
454 	u32 *key_next = next;
455 
456 	if (i == stab->map.max_entries - 1)
457 		return -ENOENT;
458 	if (i >= stab->map.max_entries)
459 		*key_next = 0;
460 	else
461 		*key_next = i + 1;
462 	return 0;
463 }
464 
465 static bool sock_map_redirect_allowed(const struct sock *sk);
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 	if (inet_csk_has_ulp(sk))
482 		return -EINVAL;
483 
484 	link = sk_psock_init_link();
485 	if (!link)
486 		return -ENOMEM;
487 
488 	/* Only sockets we can redirect into/from in BPF need to hold
489 	 * refs to parser/verdict progs and have their sk_data_ready
490 	 * and sk_write_space callbacks overridden.
491 	 */
492 	if (sock_map_redirect_allowed(sk))
493 		ret = sock_map_link(map, &stab->progs, sk);
494 	else
495 		ret = sock_map_link_no_progs(map, sk);
496 	if (ret < 0)
497 		goto out_free;
498 
499 	psock = sk_psock(sk);
500 	WARN_ON_ONCE(!psock);
501 
502 	raw_spin_lock_bh(&stab->lock);
503 	osk = stab->sks[idx];
504 	if (osk && flags == BPF_NOEXIST) {
505 		ret = -EEXIST;
506 		goto out_unlock;
507 	} else if (!osk && flags == BPF_EXIST) {
508 		ret = -ENOENT;
509 		goto out_unlock;
510 	}
511 
512 	sock_map_add_link(psock, link, map, &stab->sks[idx]);
513 	stab->sks[idx] = sk;
514 	if (osk)
515 		sock_map_unref(osk, &stab->sks[idx]);
516 	raw_spin_unlock_bh(&stab->lock);
517 	return 0;
518 out_unlock:
519 	raw_spin_unlock_bh(&stab->lock);
520 	if (psock)
521 		sk_psock_put(sk, psock);
522 out_free:
523 	sk_psock_free_link(link);
524 	return ret;
525 }
526 
527 static bool sock_map_op_okay(const struct bpf_sock_ops_kern *ops)
528 {
529 	return ops->op == BPF_SOCK_OPS_PASSIVE_ESTABLISHED_CB ||
530 	       ops->op == BPF_SOCK_OPS_ACTIVE_ESTABLISHED_CB ||
531 	       ops->op == BPF_SOCK_OPS_TCP_LISTEN_CB;
532 }
533 
534 static bool sk_is_tcp(const struct sock *sk)
535 {
536 	return sk->sk_type == SOCK_STREAM &&
537 	       sk->sk_protocol == IPPROTO_TCP;
538 }
539 
540 static bool sk_is_udp(const struct sock *sk)
541 {
542 	return sk->sk_type == SOCK_DGRAM &&
543 	       sk->sk_protocol == IPPROTO_UDP;
544 }
545 
546 static bool sock_map_redirect_allowed(const struct sock *sk)
547 {
548 	return sk_is_tcp(sk) && sk->sk_state != TCP_LISTEN;
549 }
550 
551 static bool sock_map_sk_is_suitable(const struct sock *sk)
552 {
553 	return sk_is_tcp(sk) || sk_is_udp(sk);
554 }
555 
556 static bool sock_map_sk_state_allowed(const struct sock *sk)
557 {
558 	if (sk_is_tcp(sk))
559 		return (1 << sk->sk_state) & (TCPF_ESTABLISHED | TCPF_LISTEN);
560 	else if (sk_is_udp(sk))
561 		return sk_hashed(sk);
562 
563 	return false;
564 }
565 
566 static int sock_map_update_elem(struct bpf_map *map, void *key,
567 				void *value, u64 flags)
568 {
569 	u32 idx = *(u32 *)key;
570 	struct socket *sock;
571 	struct sock *sk;
572 	int ret;
573 	u64 ufd;
574 
575 	if (map->value_size == sizeof(u64))
576 		ufd = *(u64 *)value;
577 	else
578 		ufd = *(u32 *)value;
579 	if (ufd > S32_MAX)
580 		return -EINVAL;
581 
582 	sock = sockfd_lookup(ufd, &ret);
583 	if (!sock)
584 		return ret;
585 	sk = sock->sk;
586 	if (!sk) {
587 		ret = -EINVAL;
588 		goto out;
589 	}
590 	if (!sock_map_sk_is_suitable(sk)) {
591 		ret = -EOPNOTSUPP;
592 		goto out;
593 	}
594 
595 	sock_map_sk_acquire(sk);
596 	if (!sock_map_sk_state_allowed(sk))
597 		ret = -EOPNOTSUPP;
598 	else
599 		ret = sock_map_update_common(map, idx, sk, flags);
600 	sock_map_sk_release(sk);
601 out:
602 	fput(sock->file);
603 	return ret;
604 }
605 
606 BPF_CALL_4(bpf_sock_map_update, struct bpf_sock_ops_kern *, sops,
607 	   struct bpf_map *, map, void *, key, u64, flags)
608 {
609 	WARN_ON_ONCE(!rcu_read_lock_held());
610 
611 	if (likely(sock_map_sk_is_suitable(sops->sk) &&
612 		   sock_map_op_okay(sops)))
613 		return sock_map_update_common(map, *(u32 *)key, sops->sk,
614 					      flags);
615 	return -EOPNOTSUPP;
616 }
617 
618 const struct bpf_func_proto bpf_sock_map_update_proto = {
619 	.func		= bpf_sock_map_update,
620 	.gpl_only	= false,
621 	.pkt_access	= true,
622 	.ret_type	= RET_INTEGER,
623 	.arg1_type	= ARG_PTR_TO_CTX,
624 	.arg2_type	= ARG_CONST_MAP_PTR,
625 	.arg3_type	= ARG_PTR_TO_MAP_KEY,
626 	.arg4_type	= ARG_ANYTHING,
627 };
628 
629 BPF_CALL_4(bpf_sk_redirect_map, struct sk_buff *, skb,
630 	   struct bpf_map *, map, u32, key, u64, flags)
631 {
632 	struct tcp_skb_cb *tcb = TCP_SKB_CB(skb);
633 	struct sock *sk;
634 
635 	if (unlikely(flags & ~(BPF_F_INGRESS)))
636 		return SK_DROP;
637 
638 	sk = __sock_map_lookup_elem(map, key);
639 	if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
640 		return SK_DROP;
641 
642 	tcb->bpf.flags = flags;
643 	tcb->bpf.sk_redir = sk;
644 	return SK_PASS;
645 }
646 
647 const struct bpf_func_proto bpf_sk_redirect_map_proto = {
648 	.func           = bpf_sk_redirect_map,
649 	.gpl_only       = false,
650 	.ret_type       = RET_INTEGER,
651 	.arg1_type	= ARG_PTR_TO_CTX,
652 	.arg2_type      = ARG_CONST_MAP_PTR,
653 	.arg3_type      = ARG_ANYTHING,
654 	.arg4_type      = ARG_ANYTHING,
655 };
656 
657 BPF_CALL_4(bpf_msg_redirect_map, struct sk_msg *, msg,
658 	   struct bpf_map *, map, u32, key, u64, flags)
659 {
660 	struct sock *sk;
661 
662 	if (unlikely(flags & ~(BPF_F_INGRESS)))
663 		return SK_DROP;
664 
665 	sk = __sock_map_lookup_elem(map, key);
666 	if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
667 		return SK_DROP;
668 
669 	msg->flags = flags;
670 	msg->sk_redir = sk;
671 	return SK_PASS;
672 }
673 
674 const struct bpf_func_proto bpf_msg_redirect_map_proto = {
675 	.func           = bpf_msg_redirect_map,
676 	.gpl_only       = false,
677 	.ret_type       = RET_INTEGER,
678 	.arg1_type	= ARG_PTR_TO_CTX,
679 	.arg2_type      = ARG_CONST_MAP_PTR,
680 	.arg3_type      = ARG_ANYTHING,
681 	.arg4_type      = ARG_ANYTHING,
682 };
683 
684 const struct bpf_map_ops sock_map_ops = {
685 	.map_alloc		= sock_map_alloc,
686 	.map_free		= sock_map_free,
687 	.map_get_next_key	= sock_map_get_next_key,
688 	.map_lookup_elem_sys_only = sock_map_lookup_sys,
689 	.map_update_elem	= sock_map_update_elem,
690 	.map_delete_elem	= sock_map_delete_elem,
691 	.map_lookup_elem	= sock_map_lookup,
692 	.map_release_uref	= sock_map_release_progs,
693 	.map_check_btf		= map_check_no_btf,
694 };
695 
696 struct bpf_htab_elem {
697 	struct rcu_head rcu;
698 	u32 hash;
699 	struct sock *sk;
700 	struct hlist_node node;
701 	u8 key[];
702 };
703 
704 struct bpf_htab_bucket {
705 	struct hlist_head head;
706 	raw_spinlock_t lock;
707 };
708 
709 struct bpf_htab {
710 	struct bpf_map map;
711 	struct bpf_htab_bucket *buckets;
712 	u32 buckets_num;
713 	u32 elem_size;
714 	struct sk_psock_progs progs;
715 	atomic_t count;
716 };
717 
718 static inline u32 sock_hash_bucket_hash(const void *key, u32 len)
719 {
720 	return jhash(key, len, 0);
721 }
722 
723 static struct bpf_htab_bucket *sock_hash_select_bucket(struct bpf_htab *htab,
724 						       u32 hash)
725 {
726 	return &htab->buckets[hash & (htab->buckets_num - 1)];
727 }
728 
729 static struct bpf_htab_elem *
730 sock_hash_lookup_elem_raw(struct hlist_head *head, u32 hash, void *key,
731 			  u32 key_size)
732 {
733 	struct bpf_htab_elem *elem;
734 
735 	hlist_for_each_entry_rcu(elem, head, node) {
736 		if (elem->hash == hash &&
737 		    !memcmp(&elem->key, key, key_size))
738 			return elem;
739 	}
740 
741 	return NULL;
742 }
743 
744 static struct sock *__sock_hash_lookup_elem(struct bpf_map *map, void *key)
745 {
746 	struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
747 	u32 key_size = map->key_size, hash;
748 	struct bpf_htab_bucket *bucket;
749 	struct bpf_htab_elem *elem;
750 
751 	WARN_ON_ONCE(!rcu_read_lock_held());
752 
753 	hash = sock_hash_bucket_hash(key, key_size);
754 	bucket = sock_hash_select_bucket(htab, hash);
755 	elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size);
756 
757 	return elem ? elem->sk : NULL;
758 }
759 
760 static void sock_hash_free_elem(struct bpf_htab *htab,
761 				struct bpf_htab_elem *elem)
762 {
763 	atomic_dec(&htab->count);
764 	kfree_rcu(elem, rcu);
765 }
766 
767 static void sock_hash_delete_from_link(struct bpf_map *map, struct sock *sk,
768 				       void *link_raw)
769 {
770 	struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
771 	struct bpf_htab_elem *elem_probe, *elem = link_raw;
772 	struct bpf_htab_bucket *bucket;
773 
774 	WARN_ON_ONCE(!rcu_read_lock_held());
775 	bucket = sock_hash_select_bucket(htab, elem->hash);
776 
777 	/* elem may be deleted in parallel from the map, but access here
778 	 * is okay since it's going away only after RCU grace period.
779 	 * However, we need to check whether it's still present.
780 	 */
781 	raw_spin_lock_bh(&bucket->lock);
782 	elem_probe = sock_hash_lookup_elem_raw(&bucket->head, elem->hash,
783 					       elem->key, map->key_size);
784 	if (elem_probe && elem_probe == elem) {
785 		hlist_del_rcu(&elem->node);
786 		sock_map_unref(elem->sk, elem);
787 		sock_hash_free_elem(htab, elem);
788 	}
789 	raw_spin_unlock_bh(&bucket->lock);
790 }
791 
792 static int sock_hash_delete_elem(struct bpf_map *map, void *key)
793 {
794 	struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
795 	u32 hash, key_size = map->key_size;
796 	struct bpf_htab_bucket *bucket;
797 	struct bpf_htab_elem *elem;
798 	int ret = -ENOENT;
799 
800 	hash = sock_hash_bucket_hash(key, key_size);
801 	bucket = sock_hash_select_bucket(htab, hash);
802 
803 	raw_spin_lock_bh(&bucket->lock);
804 	elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size);
805 	if (elem) {
806 		hlist_del_rcu(&elem->node);
807 		sock_map_unref(elem->sk, elem);
808 		sock_hash_free_elem(htab, elem);
809 		ret = 0;
810 	}
811 	raw_spin_unlock_bh(&bucket->lock);
812 	return ret;
813 }
814 
815 static struct bpf_htab_elem *sock_hash_alloc_elem(struct bpf_htab *htab,
816 						  void *key, u32 key_size,
817 						  u32 hash, struct sock *sk,
818 						  struct bpf_htab_elem *old)
819 {
820 	struct bpf_htab_elem *new;
821 
822 	if (atomic_inc_return(&htab->count) > htab->map.max_entries) {
823 		if (!old) {
824 			atomic_dec(&htab->count);
825 			return ERR_PTR(-E2BIG);
826 		}
827 	}
828 
829 	new = kmalloc_node(htab->elem_size, GFP_ATOMIC | __GFP_NOWARN,
830 			   htab->map.numa_node);
831 	if (!new) {
832 		atomic_dec(&htab->count);
833 		return ERR_PTR(-ENOMEM);
834 	}
835 	memcpy(new->key, key, key_size);
836 	new->sk = sk;
837 	new->hash = hash;
838 	return new;
839 }
840 
841 static int sock_hash_update_common(struct bpf_map *map, void *key,
842 				   struct sock *sk, u64 flags)
843 {
844 	struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
845 	u32 key_size = map->key_size, hash;
846 	struct bpf_htab_elem *elem, *elem_new;
847 	struct bpf_htab_bucket *bucket;
848 	struct sk_psock_link *link;
849 	struct sk_psock *psock;
850 	int ret;
851 
852 	WARN_ON_ONCE(!rcu_read_lock_held());
853 	if (unlikely(flags > BPF_EXIST))
854 		return -EINVAL;
855 	if (inet_csk_has_ulp(sk))
856 		return -EINVAL;
857 
858 	link = sk_psock_init_link();
859 	if (!link)
860 		return -ENOMEM;
861 
862 	/* Only sockets we can redirect into/from in BPF need to hold
863 	 * refs to parser/verdict progs and have their sk_data_ready
864 	 * and sk_write_space callbacks overridden.
865 	 */
866 	if (sock_map_redirect_allowed(sk))
867 		ret = sock_map_link(map, &htab->progs, sk);
868 	else
869 		ret = sock_map_link_no_progs(map, sk);
870 	if (ret < 0)
871 		goto out_free;
872 
873 	psock = sk_psock(sk);
874 	WARN_ON_ONCE(!psock);
875 
876 	hash = sock_hash_bucket_hash(key, key_size);
877 	bucket = sock_hash_select_bucket(htab, hash);
878 
879 	raw_spin_lock_bh(&bucket->lock);
880 	elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size);
881 	if (elem && flags == BPF_NOEXIST) {
882 		ret = -EEXIST;
883 		goto out_unlock;
884 	} else if (!elem && flags == BPF_EXIST) {
885 		ret = -ENOENT;
886 		goto out_unlock;
887 	}
888 
889 	elem_new = sock_hash_alloc_elem(htab, key, key_size, hash, sk, elem);
890 	if (IS_ERR(elem_new)) {
891 		ret = PTR_ERR(elem_new);
892 		goto out_unlock;
893 	}
894 
895 	sock_map_add_link(psock, link, map, elem_new);
896 	/* Add new element to the head of the list, so that
897 	 * concurrent search will find it before old elem.
898 	 */
899 	hlist_add_head_rcu(&elem_new->node, &bucket->head);
900 	if (elem) {
901 		hlist_del_rcu(&elem->node);
902 		sock_map_unref(elem->sk, elem);
903 		sock_hash_free_elem(htab, elem);
904 	}
905 	raw_spin_unlock_bh(&bucket->lock);
906 	return 0;
907 out_unlock:
908 	raw_spin_unlock_bh(&bucket->lock);
909 	sk_psock_put(sk, psock);
910 out_free:
911 	sk_psock_free_link(link);
912 	return ret;
913 }
914 
915 static int sock_hash_update_elem(struct bpf_map *map, void *key,
916 				 void *value, u64 flags)
917 {
918 	struct socket *sock;
919 	struct sock *sk;
920 	int ret;
921 	u64 ufd;
922 
923 	if (map->value_size == sizeof(u64))
924 		ufd = *(u64 *)value;
925 	else
926 		ufd = *(u32 *)value;
927 	if (ufd > S32_MAX)
928 		return -EINVAL;
929 
930 	sock = sockfd_lookup(ufd, &ret);
931 	if (!sock)
932 		return ret;
933 	sk = sock->sk;
934 	if (!sk) {
935 		ret = -EINVAL;
936 		goto out;
937 	}
938 	if (!sock_map_sk_is_suitable(sk)) {
939 		ret = -EOPNOTSUPP;
940 		goto out;
941 	}
942 
943 	sock_map_sk_acquire(sk);
944 	if (!sock_map_sk_state_allowed(sk))
945 		ret = -EOPNOTSUPP;
946 	else
947 		ret = sock_hash_update_common(map, key, sk, flags);
948 	sock_map_sk_release(sk);
949 out:
950 	fput(sock->file);
951 	return ret;
952 }
953 
954 static int sock_hash_get_next_key(struct bpf_map *map, void *key,
955 				  void *key_next)
956 {
957 	struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
958 	struct bpf_htab_elem *elem, *elem_next;
959 	u32 hash, key_size = map->key_size;
960 	struct hlist_head *head;
961 	int i = 0;
962 
963 	if (!key)
964 		goto find_first_elem;
965 	hash = sock_hash_bucket_hash(key, key_size);
966 	head = &sock_hash_select_bucket(htab, hash)->head;
967 	elem = sock_hash_lookup_elem_raw(head, hash, key, key_size);
968 	if (!elem)
969 		goto find_first_elem;
970 
971 	elem_next = hlist_entry_safe(rcu_dereference_raw(hlist_next_rcu(&elem->node)),
972 				     struct bpf_htab_elem, node);
973 	if (elem_next) {
974 		memcpy(key_next, elem_next->key, key_size);
975 		return 0;
976 	}
977 
978 	i = hash & (htab->buckets_num - 1);
979 	i++;
980 find_first_elem:
981 	for (; i < htab->buckets_num; i++) {
982 		head = &sock_hash_select_bucket(htab, i)->head;
983 		elem_next = hlist_entry_safe(rcu_dereference_raw(hlist_first_rcu(head)),
984 					     struct bpf_htab_elem, node);
985 		if (elem_next) {
986 			memcpy(key_next, elem_next->key, key_size);
987 			return 0;
988 		}
989 	}
990 
991 	return -ENOENT;
992 }
993 
994 static struct bpf_map *sock_hash_alloc(union bpf_attr *attr)
995 {
996 	struct bpf_htab *htab;
997 	int i, err;
998 	u64 cost;
999 
1000 	if (!capable(CAP_NET_ADMIN))
1001 		return ERR_PTR(-EPERM);
1002 	if (attr->max_entries == 0 ||
1003 	    attr->key_size    == 0 ||
1004 	    (attr->value_size != sizeof(u32) &&
1005 	     attr->value_size != sizeof(u64)) ||
1006 	    attr->map_flags & ~SOCK_CREATE_FLAG_MASK)
1007 		return ERR_PTR(-EINVAL);
1008 	if (attr->key_size > MAX_BPF_STACK)
1009 		return ERR_PTR(-E2BIG);
1010 
1011 	htab = kzalloc(sizeof(*htab), GFP_USER);
1012 	if (!htab)
1013 		return ERR_PTR(-ENOMEM);
1014 
1015 	bpf_map_init_from_attr(&htab->map, attr);
1016 
1017 	htab->buckets_num = roundup_pow_of_two(htab->map.max_entries);
1018 	htab->elem_size = sizeof(struct bpf_htab_elem) +
1019 			  round_up(htab->map.key_size, 8);
1020 	if (htab->buckets_num == 0 ||
1021 	    htab->buckets_num > U32_MAX / sizeof(struct bpf_htab_bucket)) {
1022 		err = -EINVAL;
1023 		goto free_htab;
1024 	}
1025 
1026 	cost = (u64) htab->buckets_num * sizeof(struct bpf_htab_bucket) +
1027 	       (u64) htab->elem_size * htab->map.max_entries;
1028 	if (cost >= U32_MAX - PAGE_SIZE) {
1029 		err = -EINVAL;
1030 		goto free_htab;
1031 	}
1032 	err = bpf_map_charge_init(&htab->map.memory, cost);
1033 	if (err)
1034 		goto free_htab;
1035 
1036 	htab->buckets = bpf_map_area_alloc(htab->buckets_num *
1037 					   sizeof(struct bpf_htab_bucket),
1038 					   htab->map.numa_node);
1039 	if (!htab->buckets) {
1040 		bpf_map_charge_finish(&htab->map.memory);
1041 		err = -ENOMEM;
1042 		goto free_htab;
1043 	}
1044 
1045 	for (i = 0; i < htab->buckets_num; i++) {
1046 		INIT_HLIST_HEAD(&htab->buckets[i].head);
1047 		raw_spin_lock_init(&htab->buckets[i].lock);
1048 	}
1049 
1050 	return &htab->map;
1051 free_htab:
1052 	kfree(htab);
1053 	return ERR_PTR(err);
1054 }
1055 
1056 static void sock_hash_free(struct bpf_map *map)
1057 {
1058 	struct bpf_htab *htab = container_of(map, struct bpf_htab, map);
1059 	struct bpf_htab_bucket *bucket;
1060 	struct hlist_head unlink_list;
1061 	struct bpf_htab_elem *elem;
1062 	struct hlist_node *node;
1063 	int i;
1064 
1065 	/* After the sync no updates or deletes will be in-flight so it
1066 	 * is safe to walk map and remove entries without risking a race
1067 	 * in EEXIST update case.
1068 	 */
1069 	synchronize_rcu();
1070 	for (i = 0; i < htab->buckets_num; i++) {
1071 		bucket = sock_hash_select_bucket(htab, i);
1072 
1073 		/* We are racing with sock_hash_delete_from_link to
1074 		 * enter the spin-lock critical section. Every socket on
1075 		 * the list is still linked to sockhash. Since link
1076 		 * exists, psock exists and holds a ref to socket. That
1077 		 * lets us to grab a socket ref too.
1078 		 */
1079 		raw_spin_lock_bh(&bucket->lock);
1080 		hlist_for_each_entry(elem, &bucket->head, node)
1081 			sock_hold(elem->sk);
1082 		hlist_move_list(&bucket->head, &unlink_list);
1083 		raw_spin_unlock_bh(&bucket->lock);
1084 
1085 		/* Process removed entries out of atomic context to
1086 		 * block for socket lock before deleting the psock's
1087 		 * link to sockhash.
1088 		 */
1089 		hlist_for_each_entry_safe(elem, node, &unlink_list, node) {
1090 			hlist_del(&elem->node);
1091 			lock_sock(elem->sk);
1092 			rcu_read_lock();
1093 			sock_map_unref(elem->sk, elem);
1094 			rcu_read_unlock();
1095 			release_sock(elem->sk);
1096 			sock_put(elem->sk);
1097 			sock_hash_free_elem(htab, elem);
1098 		}
1099 	}
1100 
1101 	/* wait for psock readers accessing its map link */
1102 	synchronize_rcu();
1103 
1104 	bpf_map_area_free(htab->buckets);
1105 	kfree(htab);
1106 }
1107 
1108 static void *sock_hash_lookup_sys(struct bpf_map *map, void *key)
1109 {
1110 	struct sock *sk;
1111 
1112 	if (map->value_size != sizeof(u64))
1113 		return ERR_PTR(-ENOSPC);
1114 
1115 	sk = __sock_hash_lookup_elem(map, key);
1116 	if (!sk)
1117 		return ERR_PTR(-ENOENT);
1118 
1119 	sock_gen_cookie(sk);
1120 	return &sk->sk_cookie;
1121 }
1122 
1123 static void *sock_hash_lookup(struct bpf_map *map, void *key)
1124 {
1125 	struct sock *sk;
1126 
1127 	sk = __sock_hash_lookup_elem(map, key);
1128 	if (!sk || !sk_fullsock(sk))
1129 		return NULL;
1130 	if (sk_is_refcounted(sk) && !refcount_inc_not_zero(&sk->sk_refcnt))
1131 		return NULL;
1132 	return sk;
1133 }
1134 
1135 static void sock_hash_release_progs(struct bpf_map *map)
1136 {
1137 	psock_progs_drop(&container_of(map, struct bpf_htab, map)->progs);
1138 }
1139 
1140 BPF_CALL_4(bpf_sock_hash_update, struct bpf_sock_ops_kern *, sops,
1141 	   struct bpf_map *, map, void *, key, u64, flags)
1142 {
1143 	WARN_ON_ONCE(!rcu_read_lock_held());
1144 
1145 	if (likely(sock_map_sk_is_suitable(sops->sk) &&
1146 		   sock_map_op_okay(sops)))
1147 		return sock_hash_update_common(map, key, sops->sk, flags);
1148 	return -EOPNOTSUPP;
1149 }
1150 
1151 const struct bpf_func_proto bpf_sock_hash_update_proto = {
1152 	.func		= bpf_sock_hash_update,
1153 	.gpl_only	= false,
1154 	.pkt_access	= true,
1155 	.ret_type	= RET_INTEGER,
1156 	.arg1_type	= ARG_PTR_TO_CTX,
1157 	.arg2_type	= ARG_CONST_MAP_PTR,
1158 	.arg3_type	= ARG_PTR_TO_MAP_KEY,
1159 	.arg4_type	= ARG_ANYTHING,
1160 };
1161 
1162 BPF_CALL_4(bpf_sk_redirect_hash, struct sk_buff *, skb,
1163 	   struct bpf_map *, map, void *, key, u64, flags)
1164 {
1165 	struct tcp_skb_cb *tcb = TCP_SKB_CB(skb);
1166 	struct sock *sk;
1167 
1168 	if (unlikely(flags & ~(BPF_F_INGRESS)))
1169 		return SK_DROP;
1170 
1171 	sk = __sock_hash_lookup_elem(map, key);
1172 	if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
1173 		return SK_DROP;
1174 
1175 	tcb->bpf.flags = flags;
1176 	tcb->bpf.sk_redir = sk;
1177 	return SK_PASS;
1178 }
1179 
1180 const struct bpf_func_proto bpf_sk_redirect_hash_proto = {
1181 	.func           = bpf_sk_redirect_hash,
1182 	.gpl_only       = false,
1183 	.ret_type       = RET_INTEGER,
1184 	.arg1_type	= ARG_PTR_TO_CTX,
1185 	.arg2_type      = ARG_CONST_MAP_PTR,
1186 	.arg3_type      = ARG_PTR_TO_MAP_KEY,
1187 	.arg4_type      = ARG_ANYTHING,
1188 };
1189 
1190 BPF_CALL_4(bpf_msg_redirect_hash, struct sk_msg *, msg,
1191 	   struct bpf_map *, map, void *, key, u64, flags)
1192 {
1193 	struct sock *sk;
1194 
1195 	if (unlikely(flags & ~(BPF_F_INGRESS)))
1196 		return SK_DROP;
1197 
1198 	sk = __sock_hash_lookup_elem(map, key);
1199 	if (unlikely(!sk || !sock_map_redirect_allowed(sk)))
1200 		return SK_DROP;
1201 
1202 	msg->flags = flags;
1203 	msg->sk_redir = sk;
1204 	return SK_PASS;
1205 }
1206 
1207 const struct bpf_func_proto bpf_msg_redirect_hash_proto = {
1208 	.func           = bpf_msg_redirect_hash,
1209 	.gpl_only       = false,
1210 	.ret_type       = RET_INTEGER,
1211 	.arg1_type	= ARG_PTR_TO_CTX,
1212 	.arg2_type      = ARG_CONST_MAP_PTR,
1213 	.arg3_type      = ARG_PTR_TO_MAP_KEY,
1214 	.arg4_type      = ARG_ANYTHING,
1215 };
1216 
1217 const struct bpf_map_ops sock_hash_ops = {
1218 	.map_alloc		= sock_hash_alloc,
1219 	.map_free		= sock_hash_free,
1220 	.map_get_next_key	= sock_hash_get_next_key,
1221 	.map_update_elem	= sock_hash_update_elem,
1222 	.map_delete_elem	= sock_hash_delete_elem,
1223 	.map_lookup_elem	= sock_hash_lookup,
1224 	.map_lookup_elem_sys_only = sock_hash_lookup_sys,
1225 	.map_release_uref	= sock_hash_release_progs,
1226 	.map_check_btf		= map_check_no_btf,
1227 };
1228 
1229 static struct sk_psock_progs *sock_map_progs(struct bpf_map *map)
1230 {
1231 	switch (map->map_type) {
1232 	case BPF_MAP_TYPE_SOCKMAP:
1233 		return &container_of(map, struct bpf_stab, map)->progs;
1234 	case BPF_MAP_TYPE_SOCKHASH:
1235 		return &container_of(map, struct bpf_htab, map)->progs;
1236 	default:
1237 		break;
1238 	}
1239 
1240 	return NULL;
1241 }
1242 
1243 int sock_map_prog_update(struct bpf_map *map, struct bpf_prog *prog,
1244 			 struct bpf_prog *old, u32 which)
1245 {
1246 	struct sk_psock_progs *progs = sock_map_progs(map);
1247 	struct bpf_prog **pprog;
1248 
1249 	if (!progs)
1250 		return -EOPNOTSUPP;
1251 
1252 	switch (which) {
1253 	case BPF_SK_MSG_VERDICT:
1254 		pprog = &progs->msg_parser;
1255 		break;
1256 	case BPF_SK_SKB_STREAM_PARSER:
1257 		pprog = &progs->skb_parser;
1258 		break;
1259 	case BPF_SK_SKB_STREAM_VERDICT:
1260 		pprog = &progs->skb_verdict;
1261 		break;
1262 	default:
1263 		return -EOPNOTSUPP;
1264 	}
1265 
1266 	if (old)
1267 		return psock_replace_prog(pprog, prog, old);
1268 
1269 	psock_set_prog(pprog, prog);
1270 	return 0;
1271 }
1272 
1273 static void sock_map_unlink(struct sock *sk, struct sk_psock_link *link)
1274 {
1275 	switch (link->map->map_type) {
1276 	case BPF_MAP_TYPE_SOCKMAP:
1277 		return sock_map_delete_from_link(link->map, sk,
1278 						 link->link_raw);
1279 	case BPF_MAP_TYPE_SOCKHASH:
1280 		return sock_hash_delete_from_link(link->map, sk,
1281 						  link->link_raw);
1282 	default:
1283 		break;
1284 	}
1285 }
1286 
1287 static void sock_map_remove_links(struct sock *sk, struct sk_psock *psock)
1288 {
1289 	struct sk_psock_link *link;
1290 
1291 	while ((link = sk_psock_link_pop(psock))) {
1292 		sock_map_unlink(sk, link);
1293 		sk_psock_free_link(link);
1294 	}
1295 }
1296 
1297 void sock_map_unhash(struct sock *sk)
1298 {
1299 	void (*saved_unhash)(struct sock *sk);
1300 	struct sk_psock *psock;
1301 
1302 	rcu_read_lock();
1303 	psock = sk_psock(sk);
1304 	if (unlikely(!psock)) {
1305 		rcu_read_unlock();
1306 		if (sk->sk_prot->unhash)
1307 			sk->sk_prot->unhash(sk);
1308 		return;
1309 	}
1310 
1311 	saved_unhash = psock->saved_unhash;
1312 	sock_map_remove_links(sk, psock);
1313 	rcu_read_unlock();
1314 	saved_unhash(sk);
1315 }
1316 
1317 void sock_map_close(struct sock *sk, long timeout)
1318 {
1319 	void (*saved_close)(struct sock *sk, long timeout);
1320 	struct sk_psock *psock;
1321 
1322 	lock_sock(sk);
1323 	rcu_read_lock();
1324 	psock = sk_psock(sk);
1325 	if (unlikely(!psock)) {
1326 		rcu_read_unlock();
1327 		release_sock(sk);
1328 		return sk->sk_prot->close(sk, timeout);
1329 	}
1330 
1331 	saved_close = psock->saved_close;
1332 	sock_map_remove_links(sk, psock);
1333 	rcu_read_unlock();
1334 	release_sock(sk);
1335 	saved_close(sk, timeout);
1336 }
1337