xref: /linux/net/ipv4/tcp_bpf.c (revision 906fd46a65383cd639e5eec72a047efc33045d86)
1 // SPDX-License-Identifier: GPL-2.0
2 /* Copyright (c) 2017 - 2018 Covalent IO, Inc. http://covalent.io */
3 
4 #include <linux/skmsg.h>
5 #include <linux/filter.h>
6 #include <linux/bpf.h>
7 #include <linux/init.h>
8 #include <linux/wait.h>
9 #include <linux/util_macros.h>
10 
11 #include <net/inet_common.h>
12 #include <net/tls.h>
13 
14 void tcp_eat_skb(struct sock *sk, struct sk_buff *skb)
15 {
16 	struct tcp_sock *tcp;
17 	int copied;
18 
19 	if (!skb || !skb->len || !sk_is_tcp(sk))
20 		return;
21 
22 	if (skb_bpf_strparser(skb))
23 		return;
24 
25 	tcp = tcp_sk(sk);
26 	copied = tcp->copied_seq + skb->len;
27 	WRITE_ONCE(tcp->copied_seq, copied);
28 	tcp_rcv_space_adjust(sk);
29 	__tcp_cleanup_rbuf(sk, skb->len);
30 }
31 
32 static int bpf_tcp_ingress(struct sock *sk, struct sk_psock *psock,
33 			   struct sk_msg *msg, u32 apply_bytes, int flags)
34 {
35 	bool apply = apply_bytes;
36 	struct scatterlist *sge;
37 	u32 size, copied = 0;
38 	struct sk_msg *tmp;
39 	int i, ret = 0;
40 
41 	tmp = kzalloc(sizeof(*tmp), __GFP_NOWARN | GFP_KERNEL);
42 	if (unlikely(!tmp))
43 		return -ENOMEM;
44 
45 	lock_sock(sk);
46 	tmp->sg.start = msg->sg.start;
47 	i = msg->sg.start;
48 	do {
49 		sge = sk_msg_elem(msg, i);
50 		size = (apply && apply_bytes < sge->length) ?
51 			apply_bytes : sge->length;
52 		if (!sk_wmem_schedule(sk, size)) {
53 			if (!copied)
54 				ret = -ENOMEM;
55 			break;
56 		}
57 
58 		sk_mem_charge(sk, size);
59 		sk_msg_xfer(tmp, msg, i, size);
60 		copied += size;
61 		if (sge->length)
62 			get_page(sk_msg_page(tmp, i));
63 		sk_msg_iter_var_next(i);
64 		tmp->sg.end = i;
65 		if (apply) {
66 			apply_bytes -= size;
67 			if (!apply_bytes) {
68 				if (sge->length)
69 					sk_msg_iter_var_prev(i);
70 				break;
71 			}
72 		}
73 	} while (i != msg->sg.end);
74 
75 	if (!ret) {
76 		msg->sg.start = i;
77 		sk_psock_queue_msg(psock, tmp);
78 		sk_psock_data_ready(sk, psock);
79 	} else {
80 		sk_msg_free(sk, tmp);
81 		kfree(tmp);
82 	}
83 
84 	release_sock(sk);
85 	return ret;
86 }
87 
88 static int tcp_bpf_push(struct sock *sk, struct sk_msg *msg, u32 apply_bytes,
89 			int flags, bool uncharge)
90 {
91 	struct msghdr msghdr = {};
92 	bool apply = apply_bytes;
93 	struct scatterlist *sge;
94 	struct page *page;
95 	int size, ret = 0;
96 	u32 off;
97 
98 	while (1) {
99 		struct bio_vec bvec;
100 		bool has_tx_ulp;
101 
102 		sge = sk_msg_elem(msg, msg->sg.start);
103 		size = (apply && apply_bytes < sge->length) ?
104 			apply_bytes : sge->length;
105 		off  = sge->offset;
106 		page = sg_page(sge);
107 
108 		tcp_rate_check_app_limited(sk);
109 retry:
110 		msghdr.msg_flags = flags | MSG_SPLICE_PAGES;
111 		has_tx_ulp = tls_sw_has_ctx_tx(sk);
112 		if (has_tx_ulp)
113 			msghdr.msg_flags |= MSG_SENDPAGE_NOPOLICY;
114 
115 		if (size < sge->length && msg->sg.start != msg->sg.end)
116 			msghdr.msg_flags |= MSG_MORE;
117 
118 		bvec_set_page(&bvec, page, size, off);
119 		iov_iter_bvec(&msghdr.msg_iter, ITER_SOURCE, &bvec, 1, size);
120 		ret = tcp_sendmsg_locked(sk, &msghdr, size);
121 		if (ret <= 0)
122 			return ret;
123 
124 		if (apply)
125 			apply_bytes -= ret;
126 		msg->sg.size -= ret;
127 		sge->offset += ret;
128 		sge->length -= ret;
129 		if (uncharge)
130 			sk_mem_uncharge(sk, ret);
131 		if (ret != size) {
132 			size -= ret;
133 			off  += ret;
134 			goto retry;
135 		}
136 		if (!sge->length) {
137 			put_page(page);
138 			sk_msg_iter_next(msg, start);
139 			sg_init_table(sge, 1);
140 			if (msg->sg.start == msg->sg.end)
141 				break;
142 		}
143 		if (apply && !apply_bytes)
144 			break;
145 	}
146 
147 	return 0;
148 }
149 
150 static int tcp_bpf_push_locked(struct sock *sk, struct sk_msg *msg,
151 			       u32 apply_bytes, int flags, bool uncharge)
152 {
153 	int ret;
154 
155 	lock_sock(sk);
156 	ret = tcp_bpf_push(sk, msg, apply_bytes, flags, uncharge);
157 	release_sock(sk);
158 	return ret;
159 }
160 
161 int tcp_bpf_sendmsg_redir(struct sock *sk, bool ingress,
162 			  struct sk_msg *msg, u32 bytes, int flags)
163 {
164 	struct sk_psock *psock = sk_psock_get(sk);
165 	int ret;
166 
167 	if (unlikely(!psock))
168 		return -EPIPE;
169 
170 	ret = ingress ? bpf_tcp_ingress(sk, psock, msg, bytes, flags) :
171 			tcp_bpf_push_locked(sk, msg, bytes, flags, false);
172 	sk_psock_put(sk, psock);
173 	return ret;
174 }
175 EXPORT_SYMBOL_GPL(tcp_bpf_sendmsg_redir);
176 
177 #ifdef CONFIG_BPF_SYSCALL
178 static int tcp_msg_wait_data(struct sock *sk, struct sk_psock *psock,
179 			     long timeo)
180 {
181 	DEFINE_WAIT_FUNC(wait, woken_wake_function);
182 	int ret = 0;
183 
184 	if (sk->sk_shutdown & RCV_SHUTDOWN)
185 		return 1;
186 
187 	if (!timeo)
188 		return ret;
189 
190 	add_wait_queue(sk_sleep(sk), &wait);
191 	sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
192 	ret = sk_wait_event(sk, &timeo,
193 			    !list_empty(&psock->ingress_msg) ||
194 			    !skb_queue_empty_lockless(&sk->sk_receive_queue), &wait);
195 	sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
196 	remove_wait_queue(sk_sleep(sk), &wait);
197 	return ret;
198 }
199 
200 static bool is_next_msg_fin(struct sk_psock *psock)
201 {
202 	struct scatterlist *sge;
203 	struct sk_msg *msg_rx;
204 	int i;
205 
206 	msg_rx = sk_psock_peek_msg(psock);
207 	i = msg_rx->sg.start;
208 	sge = sk_msg_elem(msg_rx, i);
209 	if (!sge->length) {
210 		struct sk_buff *skb = msg_rx->skb;
211 
212 		if (skb && TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN)
213 			return true;
214 	}
215 	return false;
216 }
217 
218 static int tcp_bpf_recvmsg_parser(struct sock *sk,
219 				  struct msghdr *msg,
220 				  size_t len,
221 				  int flags,
222 				  int *addr_len)
223 {
224 	struct tcp_sock *tcp = tcp_sk(sk);
225 	int peek = flags & MSG_PEEK;
226 	u32 seq = tcp->copied_seq;
227 	struct sk_psock *psock;
228 	int copied = 0;
229 
230 	if (unlikely(flags & MSG_ERRQUEUE))
231 		return inet_recv_error(sk, msg, len, addr_len);
232 
233 	if (!len)
234 		return 0;
235 
236 	psock = sk_psock_get(sk);
237 	if (unlikely(!psock))
238 		return tcp_recvmsg(sk, msg, len, flags, addr_len);
239 
240 	lock_sock(sk);
241 
242 	/* We may have received data on the sk_receive_queue pre-accept and
243 	 * then we can not use read_skb in this context because we haven't
244 	 * assigned a sk_socket yet so have no link to the ops. The work-around
245 	 * is to check the sk_receive_queue and in these cases read skbs off
246 	 * queue again. The read_skb hook is not running at this point because
247 	 * of lock_sock so we avoid having multiple runners in read_skb.
248 	 */
249 	if (unlikely(!skb_queue_empty(&sk->sk_receive_queue))) {
250 		tcp_data_ready(sk);
251 		/* This handles the ENOMEM errors if we both receive data
252 		 * pre accept and are already under memory pressure. At least
253 		 * let user know to retry.
254 		 */
255 		if (unlikely(!skb_queue_empty(&sk->sk_receive_queue))) {
256 			copied = -EAGAIN;
257 			goto out;
258 		}
259 	}
260 
261 msg_bytes_ready:
262 	copied = sk_msg_recvmsg(sk, psock, msg, len, flags);
263 	/* The typical case for EFAULT is the socket was gracefully
264 	 * shutdown with a FIN pkt. So check here the other case is
265 	 * some error on copy_page_to_iter which would be unexpected.
266 	 * On fin return correct return code to zero.
267 	 */
268 	if (copied == -EFAULT) {
269 		bool is_fin = is_next_msg_fin(psock);
270 
271 		if (is_fin) {
272 			copied = 0;
273 			seq++;
274 			goto out;
275 		}
276 	}
277 	seq += copied;
278 	if (!copied) {
279 		long timeo;
280 		int data;
281 
282 		if (sock_flag(sk, SOCK_DONE))
283 			goto out;
284 
285 		if (sk->sk_err) {
286 			copied = sock_error(sk);
287 			goto out;
288 		}
289 
290 		if (sk->sk_shutdown & RCV_SHUTDOWN)
291 			goto out;
292 
293 		if (sk->sk_state == TCP_CLOSE) {
294 			copied = -ENOTCONN;
295 			goto out;
296 		}
297 
298 		timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
299 		if (!timeo) {
300 			copied = -EAGAIN;
301 			goto out;
302 		}
303 
304 		if (signal_pending(current)) {
305 			copied = sock_intr_errno(timeo);
306 			goto out;
307 		}
308 
309 		data = tcp_msg_wait_data(sk, psock, timeo);
310 		if (data < 0) {
311 			copied = data;
312 			goto unlock;
313 		}
314 		if (data && !sk_psock_queue_empty(psock))
315 			goto msg_bytes_ready;
316 		copied = -EAGAIN;
317 	}
318 out:
319 	if (!peek)
320 		WRITE_ONCE(tcp->copied_seq, seq);
321 	tcp_rcv_space_adjust(sk);
322 	if (copied > 0)
323 		__tcp_cleanup_rbuf(sk, copied);
324 
325 unlock:
326 	release_sock(sk);
327 	sk_psock_put(sk, psock);
328 	return copied;
329 }
330 
331 static int tcp_bpf_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
332 			   int flags, int *addr_len)
333 {
334 	struct sk_psock *psock;
335 	int copied, ret;
336 
337 	if (unlikely(flags & MSG_ERRQUEUE))
338 		return inet_recv_error(sk, msg, len, addr_len);
339 
340 	if (!len)
341 		return 0;
342 
343 	psock = sk_psock_get(sk);
344 	if (unlikely(!psock))
345 		return tcp_recvmsg(sk, msg, len, flags, addr_len);
346 	if (!skb_queue_empty(&sk->sk_receive_queue) &&
347 	    sk_psock_queue_empty(psock)) {
348 		sk_psock_put(sk, psock);
349 		return tcp_recvmsg(sk, msg, len, flags, addr_len);
350 	}
351 	lock_sock(sk);
352 msg_bytes_ready:
353 	copied = sk_msg_recvmsg(sk, psock, msg, len, flags);
354 	if (!copied) {
355 		long timeo;
356 		int data;
357 
358 		timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
359 		data = tcp_msg_wait_data(sk, psock, timeo);
360 		if (data < 0) {
361 			ret = data;
362 			goto unlock;
363 		}
364 		if (data) {
365 			if (!sk_psock_queue_empty(psock))
366 				goto msg_bytes_ready;
367 			release_sock(sk);
368 			sk_psock_put(sk, psock);
369 			return tcp_recvmsg(sk, msg, len, flags, addr_len);
370 		}
371 		copied = -EAGAIN;
372 	}
373 	ret = copied;
374 
375 unlock:
376 	release_sock(sk);
377 	sk_psock_put(sk, psock);
378 	return ret;
379 }
380 
381 static int tcp_bpf_send_verdict(struct sock *sk, struct sk_psock *psock,
382 				struct sk_msg *msg, int *copied, int flags)
383 {
384 	bool cork = false, enospc = sk_msg_full(msg), redir_ingress;
385 	struct sock *sk_redir;
386 	u32 tosend, origsize, sent, delta = 0;
387 	u32 eval;
388 	int ret;
389 
390 more_data:
391 	if (psock->eval == __SK_NONE) {
392 		/* Track delta in msg size to add/subtract it on SK_DROP from
393 		 * returned to user copied size. This ensures user doesn't
394 		 * get a positive return code with msg_cut_data and SK_DROP
395 		 * verdict.
396 		 */
397 		delta = msg->sg.size;
398 		psock->eval = sk_psock_msg_verdict(sk, psock, msg);
399 		delta -= msg->sg.size;
400 	}
401 
402 	if (msg->cork_bytes &&
403 	    msg->cork_bytes > msg->sg.size && !enospc) {
404 		psock->cork_bytes = msg->cork_bytes - msg->sg.size;
405 		if (!psock->cork) {
406 			psock->cork = kzalloc(sizeof(*psock->cork),
407 					      GFP_ATOMIC | __GFP_NOWARN);
408 			if (!psock->cork)
409 				return -ENOMEM;
410 		}
411 		memcpy(psock->cork, msg, sizeof(*msg));
412 		return 0;
413 	}
414 
415 	tosend = msg->sg.size;
416 	if (psock->apply_bytes && psock->apply_bytes < tosend)
417 		tosend = psock->apply_bytes;
418 	eval = __SK_NONE;
419 
420 	switch (psock->eval) {
421 	case __SK_PASS:
422 		ret = tcp_bpf_push(sk, msg, tosend, flags, true);
423 		if (unlikely(ret)) {
424 			*copied -= sk_msg_free(sk, msg);
425 			break;
426 		}
427 		sk_msg_apply_bytes(psock, tosend);
428 		break;
429 	case __SK_REDIRECT:
430 		redir_ingress = psock->redir_ingress;
431 		sk_redir = psock->sk_redir;
432 		sk_msg_apply_bytes(psock, tosend);
433 		if (!psock->apply_bytes) {
434 			/* Clean up before releasing the sock lock. */
435 			eval = psock->eval;
436 			psock->eval = __SK_NONE;
437 			psock->sk_redir = NULL;
438 		}
439 		if (psock->cork) {
440 			cork = true;
441 			psock->cork = NULL;
442 		}
443 		sk_msg_return(sk, msg, tosend);
444 		release_sock(sk);
445 
446 		origsize = msg->sg.size;
447 		ret = tcp_bpf_sendmsg_redir(sk_redir, redir_ingress,
448 					    msg, tosend, flags);
449 		sent = origsize - msg->sg.size;
450 
451 		if (eval == __SK_REDIRECT)
452 			sock_put(sk_redir);
453 
454 		lock_sock(sk);
455 		if (unlikely(ret < 0)) {
456 			int free = sk_msg_free_nocharge(sk, msg);
457 
458 			if (!cork)
459 				*copied -= free;
460 		}
461 		if (cork) {
462 			sk_msg_free(sk, msg);
463 			kfree(msg);
464 			msg = NULL;
465 			ret = 0;
466 		}
467 		break;
468 	case __SK_DROP:
469 	default:
470 		sk_msg_free_partial(sk, msg, tosend);
471 		sk_msg_apply_bytes(psock, tosend);
472 		*copied -= (tosend + delta);
473 		return -EACCES;
474 	}
475 
476 	if (likely(!ret)) {
477 		if (!psock->apply_bytes) {
478 			psock->eval =  __SK_NONE;
479 			if (psock->sk_redir) {
480 				sock_put(psock->sk_redir);
481 				psock->sk_redir = NULL;
482 			}
483 		}
484 		if (msg &&
485 		    msg->sg.data[msg->sg.start].page_link &&
486 		    msg->sg.data[msg->sg.start].length) {
487 			if (eval == __SK_REDIRECT)
488 				sk_mem_charge(sk, tosend - sent);
489 			goto more_data;
490 		}
491 	}
492 	return ret;
493 }
494 
495 static int tcp_bpf_sendmsg(struct sock *sk, struct msghdr *msg, size_t size)
496 {
497 	struct sk_msg tmp, *msg_tx = NULL;
498 	int copied = 0, err = 0;
499 	struct sk_psock *psock;
500 	long timeo;
501 	int flags;
502 
503 	/* Don't let internal flags through */
504 	flags = (msg->msg_flags & ~MSG_SENDPAGE_DECRYPTED);
505 	flags |= MSG_NO_SHARED_FRAGS;
506 
507 	psock = sk_psock_get(sk);
508 	if (unlikely(!psock))
509 		return tcp_sendmsg(sk, msg, size);
510 
511 	lock_sock(sk);
512 	timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
513 	while (msg_data_left(msg)) {
514 		bool enospc = false;
515 		u32 copy, osize;
516 
517 		if (sk->sk_err) {
518 			err = -sk->sk_err;
519 			goto out_err;
520 		}
521 
522 		copy = msg_data_left(msg);
523 		if (!sk_stream_memory_free(sk))
524 			goto wait_for_sndbuf;
525 		if (psock->cork) {
526 			msg_tx = psock->cork;
527 		} else {
528 			msg_tx = &tmp;
529 			sk_msg_init(msg_tx);
530 		}
531 
532 		osize = msg_tx->sg.size;
533 		err = sk_msg_alloc(sk, msg_tx, msg_tx->sg.size + copy, msg_tx->sg.end - 1);
534 		if (err) {
535 			if (err != -ENOSPC)
536 				goto wait_for_memory;
537 			enospc = true;
538 			copy = msg_tx->sg.size - osize;
539 		}
540 
541 		err = sk_msg_memcopy_from_iter(sk, &msg->msg_iter, msg_tx,
542 					       copy);
543 		if (err < 0) {
544 			sk_msg_trim(sk, msg_tx, osize);
545 			goto out_err;
546 		}
547 
548 		copied += copy;
549 		if (psock->cork_bytes) {
550 			if (size > psock->cork_bytes)
551 				psock->cork_bytes = 0;
552 			else
553 				psock->cork_bytes -= size;
554 			if (psock->cork_bytes && !enospc)
555 				goto out_err;
556 			/* All cork bytes are accounted, rerun the prog. */
557 			psock->eval = __SK_NONE;
558 			psock->cork_bytes = 0;
559 		}
560 
561 		err = tcp_bpf_send_verdict(sk, psock, msg_tx, &copied, flags);
562 		if (unlikely(err < 0))
563 			goto out_err;
564 		continue;
565 wait_for_sndbuf:
566 		set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
567 wait_for_memory:
568 		err = sk_stream_wait_memory(sk, &timeo);
569 		if (err) {
570 			if (msg_tx && msg_tx != psock->cork)
571 				sk_msg_free(sk, msg_tx);
572 			goto out_err;
573 		}
574 	}
575 out_err:
576 	if (err < 0)
577 		err = sk_stream_error(sk, msg->msg_flags, err);
578 	release_sock(sk);
579 	sk_psock_put(sk, psock);
580 	return copied ? copied : err;
581 }
582 
583 enum {
584 	TCP_BPF_IPV4,
585 	TCP_BPF_IPV6,
586 	TCP_BPF_NUM_PROTS,
587 };
588 
589 enum {
590 	TCP_BPF_BASE,
591 	TCP_BPF_TX,
592 	TCP_BPF_RX,
593 	TCP_BPF_TXRX,
594 	TCP_BPF_NUM_CFGS,
595 };
596 
597 static struct proto *tcpv6_prot_saved __read_mostly;
598 static DEFINE_SPINLOCK(tcpv6_prot_lock);
599 static struct proto tcp_bpf_prots[TCP_BPF_NUM_PROTS][TCP_BPF_NUM_CFGS];
600 
601 static void tcp_bpf_rebuild_protos(struct proto prot[TCP_BPF_NUM_CFGS],
602 				   struct proto *base)
603 {
604 	prot[TCP_BPF_BASE]			= *base;
605 	prot[TCP_BPF_BASE].destroy		= sock_map_destroy;
606 	prot[TCP_BPF_BASE].close		= sock_map_close;
607 	prot[TCP_BPF_BASE].recvmsg		= tcp_bpf_recvmsg;
608 	prot[TCP_BPF_BASE].sock_is_readable	= sk_msg_is_readable;
609 
610 	prot[TCP_BPF_TX]			= prot[TCP_BPF_BASE];
611 	prot[TCP_BPF_TX].sendmsg		= tcp_bpf_sendmsg;
612 
613 	prot[TCP_BPF_RX]			= prot[TCP_BPF_BASE];
614 	prot[TCP_BPF_RX].recvmsg		= tcp_bpf_recvmsg_parser;
615 
616 	prot[TCP_BPF_TXRX]			= prot[TCP_BPF_TX];
617 	prot[TCP_BPF_TXRX].recvmsg		= tcp_bpf_recvmsg_parser;
618 }
619 
620 static void tcp_bpf_check_v6_needs_rebuild(struct proto *ops)
621 {
622 	if (unlikely(ops != smp_load_acquire(&tcpv6_prot_saved))) {
623 		spin_lock_bh(&tcpv6_prot_lock);
624 		if (likely(ops != tcpv6_prot_saved)) {
625 			tcp_bpf_rebuild_protos(tcp_bpf_prots[TCP_BPF_IPV6], ops);
626 			smp_store_release(&tcpv6_prot_saved, ops);
627 		}
628 		spin_unlock_bh(&tcpv6_prot_lock);
629 	}
630 }
631 
632 static int __init tcp_bpf_v4_build_proto(void)
633 {
634 	tcp_bpf_rebuild_protos(tcp_bpf_prots[TCP_BPF_IPV4], &tcp_prot);
635 	return 0;
636 }
637 late_initcall(tcp_bpf_v4_build_proto);
638 
639 static int tcp_bpf_assert_proto_ops(struct proto *ops)
640 {
641 	/* In order to avoid retpoline, we make assumptions when we call
642 	 * into ops if e.g. a psock is not present. Make sure they are
643 	 * indeed valid assumptions.
644 	 */
645 	return ops->recvmsg  == tcp_recvmsg &&
646 	       ops->sendmsg  == tcp_sendmsg ? 0 : -ENOTSUPP;
647 }
648 
649 int tcp_bpf_update_proto(struct sock *sk, struct sk_psock *psock, bool restore)
650 {
651 	int family = sk->sk_family == AF_INET6 ? TCP_BPF_IPV6 : TCP_BPF_IPV4;
652 	int config = psock->progs.msg_parser   ? TCP_BPF_TX   : TCP_BPF_BASE;
653 
654 	if (psock->progs.stream_verdict || psock->progs.skb_verdict) {
655 		config = (config == TCP_BPF_TX) ? TCP_BPF_TXRX : TCP_BPF_RX;
656 	}
657 
658 	if (restore) {
659 		if (inet_csk_has_ulp(sk)) {
660 			/* TLS does not have an unhash proto in SW cases,
661 			 * but we need to ensure we stop using the sock_map
662 			 * unhash routine because the associated psock is being
663 			 * removed. So use the original unhash handler.
664 			 */
665 			WRITE_ONCE(sk->sk_prot->unhash, psock->saved_unhash);
666 			tcp_update_ulp(sk, psock->sk_proto, psock->saved_write_space);
667 		} else {
668 			sk->sk_write_space = psock->saved_write_space;
669 			/* Pairs with lockless read in sk_clone_lock() */
670 			sock_replace_proto(sk, psock->sk_proto);
671 		}
672 		return 0;
673 	}
674 
675 	if (sk->sk_family == AF_INET6) {
676 		if (tcp_bpf_assert_proto_ops(psock->sk_proto))
677 			return -EINVAL;
678 
679 		tcp_bpf_check_v6_needs_rebuild(psock->sk_proto);
680 	}
681 
682 	/* Pairs with lockless read in sk_clone_lock() */
683 	sock_replace_proto(sk, &tcp_bpf_prots[family][config]);
684 	return 0;
685 }
686 EXPORT_SYMBOL_GPL(tcp_bpf_update_proto);
687 
688 /* If a child got cloned from a listening socket that had tcp_bpf
689  * protocol callbacks installed, we need to restore the callbacks to
690  * the default ones because the child does not inherit the psock state
691  * that tcp_bpf callbacks expect.
692  */
693 void tcp_bpf_clone(const struct sock *sk, struct sock *newsk)
694 {
695 	struct proto *prot = newsk->sk_prot;
696 
697 	if (is_insidevar(prot, tcp_bpf_prots))
698 		newsk->sk_prot = sk->sk_prot_creator;
699 }
700 #endif /* CONFIG_BPF_SYSCALL */
701