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