• 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 
10 #include <net/inet_common.h>
11 #include <net/tls.h>
12 
tcp_bpf_stream_read(const struct sock * sk)13 static bool tcp_bpf_stream_read(const struct sock *sk)
14 {
15 	struct sk_psock *psock;
16 	bool empty = true;
17 
18 	rcu_read_lock();
19 	psock = sk_psock(sk);
20 	if (likely(psock))
21 		empty = list_empty(&psock->ingress_msg);
22 	rcu_read_unlock();
23 	return !empty;
24 }
25 
tcp_bpf_wait_data(struct sock * sk,struct sk_psock * psock,int flags,long timeo,int * err)26 static int tcp_bpf_wait_data(struct sock *sk, struct sk_psock *psock,
27 			     int flags, long timeo, int *err)
28 {
29 	DEFINE_WAIT_FUNC(wait, woken_wake_function);
30 	int ret = 0;
31 
32 	if (!timeo)
33 		return ret;
34 
35 	add_wait_queue(sk_sleep(sk), &wait);
36 	sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
37 	ret = sk_wait_event(sk, &timeo,
38 			    !list_empty(&psock->ingress_msg) ||
39 			    !skb_queue_empty(&sk->sk_receive_queue), &wait);
40 	sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
41 	remove_wait_queue(sk_sleep(sk), &wait);
42 	return ret;
43 }
44 
__tcp_bpf_recvmsg(struct sock * sk,struct sk_psock * psock,struct msghdr * msg,int len,int flags)45 int __tcp_bpf_recvmsg(struct sock *sk, struct sk_psock *psock,
46 		      struct msghdr *msg, int len, int flags)
47 {
48 	struct iov_iter *iter = &msg->msg_iter;
49 	int peek = flags & MSG_PEEK;
50 	int i, ret, copied = 0;
51 	struct sk_msg *msg_rx;
52 
53 	msg_rx = list_first_entry_or_null(&psock->ingress_msg,
54 					  struct sk_msg, list);
55 
56 	while (copied != len) {
57 		struct scatterlist *sge;
58 
59 		if (unlikely(!msg_rx))
60 			break;
61 
62 		i = msg_rx->sg.start;
63 		do {
64 			struct page *page;
65 			int copy;
66 
67 			sge = sk_msg_elem(msg_rx, i);
68 			copy = sge->length;
69 			page = sg_page(sge);
70 			if (copied + copy > len)
71 				copy = len - copied;
72 			ret = copy_page_to_iter(page, sge->offset, copy, iter);
73 			if (ret != copy) {
74 				msg_rx->sg.start = i;
75 				return -EFAULT;
76 			}
77 
78 			copied += copy;
79 			if (likely(!peek)) {
80 				sge->offset += copy;
81 				sge->length -= copy;
82 				sk_mem_uncharge(sk, copy);
83 				msg_rx->sg.size -= copy;
84 
85 				if (!sge->length) {
86 					sk_msg_iter_var_next(i);
87 					if (!msg_rx->skb)
88 						put_page(page);
89 				}
90 			} else {
91 				sk_msg_iter_var_next(i);
92 			}
93 
94 			if (copied == len)
95 				break;
96 		} while (i != msg_rx->sg.end);
97 
98 		if (unlikely(peek)) {
99 			msg_rx = list_next_entry(msg_rx, list);
100 			continue;
101 		}
102 
103 		msg_rx->sg.start = i;
104 		if (!sge->length && msg_rx->sg.start == msg_rx->sg.end) {
105 			list_del(&msg_rx->list);
106 			if (msg_rx->skb)
107 				consume_skb(msg_rx->skb);
108 			kfree(msg_rx);
109 		}
110 		msg_rx = list_first_entry_or_null(&psock->ingress_msg,
111 						  struct sk_msg, list);
112 	}
113 
114 	return copied;
115 }
116 EXPORT_SYMBOL_GPL(__tcp_bpf_recvmsg);
117 
tcp_bpf_recvmsg(struct sock * sk,struct msghdr * msg,size_t len,int nonblock,int flags,int * addr_len)118 int tcp_bpf_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
119 		    int nonblock, int flags, int *addr_len)
120 {
121 	struct sk_psock *psock;
122 	int copied, ret;
123 
124 	psock = sk_psock_get(sk);
125 	if (unlikely(!psock))
126 		return tcp_recvmsg(sk, msg, len, nonblock, flags, addr_len);
127 	if (unlikely(flags & MSG_ERRQUEUE))
128 		return inet_recv_error(sk, msg, len, addr_len);
129 	if (!skb_queue_empty(&sk->sk_receive_queue) &&
130 	    sk_psock_queue_empty(psock))
131 		return tcp_recvmsg(sk, msg, len, nonblock, flags, addr_len);
132 	lock_sock(sk);
133 msg_bytes_ready:
134 	copied = __tcp_bpf_recvmsg(sk, psock, msg, len, flags);
135 	if (!copied) {
136 		int data, err = 0;
137 		long timeo;
138 
139 		timeo = sock_rcvtimeo(sk, nonblock);
140 		data = tcp_bpf_wait_data(sk, psock, flags, timeo, &err);
141 		if (data) {
142 			if (!sk_psock_queue_empty(psock))
143 				goto msg_bytes_ready;
144 			release_sock(sk);
145 			sk_psock_put(sk, psock);
146 			return tcp_recvmsg(sk, msg, len, nonblock, flags, addr_len);
147 		}
148 		if (err) {
149 			ret = err;
150 			goto out;
151 		}
152 		copied = -EAGAIN;
153 	}
154 	ret = copied;
155 out:
156 	release_sock(sk);
157 	sk_psock_put(sk, psock);
158 	return ret;
159 }
160 
bpf_tcp_ingress(struct sock * sk,struct sk_psock * psock,struct sk_msg * msg,u32 apply_bytes,int flags)161 static int bpf_tcp_ingress(struct sock *sk, struct sk_psock *psock,
162 			   struct sk_msg *msg, u32 apply_bytes, int flags)
163 {
164 	bool apply = apply_bytes;
165 	struct scatterlist *sge;
166 	u32 size, copied = 0;
167 	struct sk_msg *tmp;
168 	int i, ret = 0;
169 
170 	tmp = kzalloc(sizeof(*tmp), __GFP_NOWARN | GFP_KERNEL);
171 	if (unlikely(!tmp))
172 		return -ENOMEM;
173 
174 	lock_sock(sk);
175 	tmp->sg.start = msg->sg.start;
176 	i = msg->sg.start;
177 	do {
178 		sge = sk_msg_elem(msg, i);
179 		size = (apply && apply_bytes < sge->length) ?
180 			apply_bytes : sge->length;
181 		if (!sk_wmem_schedule(sk, size)) {
182 			if (!copied)
183 				ret = -ENOMEM;
184 			break;
185 		}
186 
187 		sk_mem_charge(sk, size);
188 		sk_msg_xfer(tmp, msg, i, size);
189 		copied += size;
190 		if (sge->length)
191 			get_page(sk_msg_page(tmp, i));
192 		sk_msg_iter_var_next(i);
193 		tmp->sg.end = i;
194 		if (apply) {
195 			apply_bytes -= size;
196 			if (!apply_bytes)
197 				break;
198 		}
199 	} while (i != msg->sg.end);
200 
201 	if (!ret) {
202 		msg->sg.start = i;
203 		msg->sg.size -= apply_bytes;
204 		sk_psock_queue_msg(psock, tmp);
205 		sk_psock_data_ready(sk, psock);
206 	} else {
207 		sk_msg_free(sk, tmp);
208 		kfree(tmp);
209 	}
210 
211 	release_sock(sk);
212 	return ret;
213 }
214 
tcp_bpf_push(struct sock * sk,struct sk_msg * msg,u32 apply_bytes,int flags,bool uncharge)215 static int tcp_bpf_push(struct sock *sk, struct sk_msg *msg, u32 apply_bytes,
216 			int flags, bool uncharge)
217 {
218 	bool apply = apply_bytes;
219 	struct scatterlist *sge;
220 	struct page *page;
221 	int size, ret = 0;
222 	u32 off;
223 
224 	while (1) {
225 		bool has_tx_ulp;
226 
227 		sge = sk_msg_elem(msg, msg->sg.start);
228 		size = (apply && apply_bytes < sge->length) ?
229 			apply_bytes : sge->length;
230 		off  = sge->offset;
231 		page = sg_page(sge);
232 
233 		tcp_rate_check_app_limited(sk);
234 retry:
235 		has_tx_ulp = tls_sw_has_ctx_tx(sk);
236 		if (has_tx_ulp) {
237 			flags |= MSG_SENDPAGE_NOPOLICY;
238 			ret = kernel_sendpage_locked(sk,
239 						     page, off, size, flags);
240 		} else {
241 			ret = do_tcp_sendpages(sk, page, off, size, flags);
242 		}
243 
244 		if (ret <= 0)
245 			return ret;
246 		if (apply)
247 			apply_bytes -= ret;
248 		msg->sg.size -= ret;
249 		sge->offset += ret;
250 		sge->length -= ret;
251 		if (uncharge)
252 			sk_mem_uncharge(sk, ret);
253 		if (ret != size) {
254 			size -= ret;
255 			off  += ret;
256 			goto retry;
257 		}
258 		if (!sge->length) {
259 			put_page(page);
260 			sk_msg_iter_next(msg, start);
261 			sg_init_table(sge, 1);
262 			if (msg->sg.start == msg->sg.end)
263 				break;
264 		}
265 		if (apply && !apply_bytes)
266 			break;
267 	}
268 
269 	return 0;
270 }
271 
tcp_bpf_push_locked(struct sock * sk,struct sk_msg * msg,u32 apply_bytes,int flags,bool uncharge)272 static int tcp_bpf_push_locked(struct sock *sk, struct sk_msg *msg,
273 			       u32 apply_bytes, int flags, bool uncharge)
274 {
275 	int ret;
276 
277 	lock_sock(sk);
278 	ret = tcp_bpf_push(sk, msg, apply_bytes, flags, uncharge);
279 	release_sock(sk);
280 	return ret;
281 }
282 
tcp_bpf_sendmsg_redir(struct sock * sk,struct sk_msg * msg,u32 bytes,int flags)283 int tcp_bpf_sendmsg_redir(struct sock *sk, struct sk_msg *msg,
284 			  u32 bytes, int flags)
285 {
286 	bool ingress = sk_msg_to_ingress(msg);
287 	struct sk_psock *psock = sk_psock_get(sk);
288 	int ret;
289 
290 	if (unlikely(!psock)) {
291 		sk_msg_free(sk, msg);
292 		return 0;
293 	}
294 	ret = ingress ? bpf_tcp_ingress(sk, psock, msg, bytes, flags) :
295 			tcp_bpf_push_locked(sk, msg, bytes, flags, false);
296 	sk_psock_put(sk, psock);
297 	return ret;
298 }
299 EXPORT_SYMBOL_GPL(tcp_bpf_sendmsg_redir);
300 
tcp_bpf_send_verdict(struct sock * sk,struct sk_psock * psock,struct sk_msg * msg,int * copied,int flags)301 static int tcp_bpf_send_verdict(struct sock *sk, struct sk_psock *psock,
302 				struct sk_msg *msg, int *copied, int flags)
303 {
304 	bool cork = false, enospc = sk_msg_full(msg);
305 	struct sock *sk_redir;
306 	u32 tosend, delta = 0;
307 	int ret;
308 
309 more_data:
310 	if (psock->eval == __SK_NONE) {
311 		/* Track delta in msg size to add/subtract it on SK_DROP from
312 		 * returned to user copied size. This ensures user doesn't
313 		 * get a positive return code with msg_cut_data and SK_DROP
314 		 * verdict.
315 		 */
316 		delta = msg->sg.size;
317 		psock->eval = sk_psock_msg_verdict(sk, psock, msg);
318 		delta -= msg->sg.size;
319 	}
320 
321 	if (msg->cork_bytes &&
322 	    msg->cork_bytes > msg->sg.size && !enospc) {
323 		psock->cork_bytes = msg->cork_bytes - msg->sg.size;
324 		if (!psock->cork) {
325 			psock->cork = kzalloc(sizeof(*psock->cork),
326 					      GFP_ATOMIC | __GFP_NOWARN);
327 			if (!psock->cork)
328 				return -ENOMEM;
329 		}
330 		memcpy(psock->cork, msg, sizeof(*msg));
331 		return 0;
332 	}
333 
334 	tosend = msg->sg.size;
335 	if (psock->apply_bytes && psock->apply_bytes < tosend)
336 		tosend = psock->apply_bytes;
337 
338 	switch (psock->eval) {
339 	case __SK_PASS:
340 		ret = tcp_bpf_push(sk, msg, tosend, flags, true);
341 		if (unlikely(ret)) {
342 			*copied -= sk_msg_free(sk, msg);
343 			break;
344 		}
345 		sk_msg_apply_bytes(psock, tosend);
346 		break;
347 	case __SK_REDIRECT:
348 		sk_redir = psock->sk_redir;
349 		sk_msg_apply_bytes(psock, tosend);
350 		if (psock->cork) {
351 			cork = true;
352 			psock->cork = NULL;
353 		}
354 		sk_msg_return(sk, msg, tosend);
355 		release_sock(sk);
356 		ret = tcp_bpf_sendmsg_redir(sk_redir, msg, tosend, flags);
357 		lock_sock(sk);
358 		if (unlikely(ret < 0)) {
359 			int free = sk_msg_free_nocharge(sk, msg);
360 
361 			if (!cork)
362 				*copied -= free;
363 		}
364 		if (cork) {
365 			sk_msg_free(sk, msg);
366 			kfree(msg);
367 			msg = NULL;
368 			ret = 0;
369 		}
370 		break;
371 	case __SK_DROP:
372 	default:
373 		sk_msg_free_partial(sk, msg, tosend);
374 		sk_msg_apply_bytes(psock, tosend);
375 		*copied -= (tosend + delta);
376 		return -EACCES;
377 	}
378 
379 	if (likely(!ret)) {
380 		if (!psock->apply_bytes) {
381 			psock->eval =  __SK_NONE;
382 			if (psock->sk_redir) {
383 				sock_put(psock->sk_redir);
384 				psock->sk_redir = NULL;
385 			}
386 		}
387 		if (msg &&
388 		    msg->sg.data[msg->sg.start].page_link &&
389 		    msg->sg.data[msg->sg.start].length)
390 			goto more_data;
391 	}
392 	return ret;
393 }
394 
tcp_bpf_sendmsg(struct sock * sk,struct msghdr * msg,size_t size)395 static int tcp_bpf_sendmsg(struct sock *sk, struct msghdr *msg, size_t size)
396 {
397 	struct sk_msg tmp, *msg_tx = NULL;
398 	int copied = 0, err = 0;
399 	struct sk_psock *psock;
400 	long timeo;
401 	int flags;
402 
403 	/* Don't let internal do_tcp_sendpages() flags through */
404 	flags = (msg->msg_flags & ~MSG_SENDPAGE_DECRYPTED);
405 	flags |= MSG_NO_SHARED_FRAGS;
406 
407 	psock = sk_psock_get(sk);
408 	if (unlikely(!psock))
409 		return tcp_sendmsg(sk, msg, size);
410 
411 	lock_sock(sk);
412 	timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
413 	while (msg_data_left(msg)) {
414 		bool enospc = false;
415 		u32 copy, osize;
416 
417 		if (sk->sk_err) {
418 			err = -sk->sk_err;
419 			goto out_err;
420 		}
421 
422 		copy = msg_data_left(msg);
423 		if (!sk_stream_memory_free(sk))
424 			goto wait_for_sndbuf;
425 		if (psock->cork) {
426 			msg_tx = psock->cork;
427 		} else {
428 			msg_tx = &tmp;
429 			sk_msg_init(msg_tx);
430 		}
431 
432 		osize = msg_tx->sg.size;
433 		err = sk_msg_alloc(sk, msg_tx, msg_tx->sg.size + copy, msg_tx->sg.end - 1);
434 		if (err) {
435 			if (err != -ENOSPC)
436 				goto wait_for_memory;
437 			enospc = true;
438 			copy = msg_tx->sg.size - osize;
439 		}
440 
441 		err = sk_msg_memcopy_from_iter(sk, &msg->msg_iter, msg_tx,
442 					       copy);
443 		if (err < 0) {
444 			sk_msg_trim(sk, msg_tx, osize);
445 			goto out_err;
446 		}
447 
448 		copied += copy;
449 		if (psock->cork_bytes) {
450 			if (size > psock->cork_bytes)
451 				psock->cork_bytes = 0;
452 			else
453 				psock->cork_bytes -= size;
454 			if (psock->cork_bytes && !enospc)
455 				goto out_err;
456 			/* All cork bytes are accounted, rerun the prog. */
457 			psock->eval = __SK_NONE;
458 			psock->cork_bytes = 0;
459 		}
460 
461 		err = tcp_bpf_send_verdict(sk, psock, msg_tx, &copied, flags);
462 		if (unlikely(err < 0))
463 			goto out_err;
464 		continue;
465 wait_for_sndbuf:
466 		set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
467 wait_for_memory:
468 		err = sk_stream_wait_memory(sk, &timeo);
469 		if (err) {
470 			if (msg_tx && msg_tx != psock->cork)
471 				sk_msg_free(sk, msg_tx);
472 			goto out_err;
473 		}
474 	}
475 out_err:
476 	if (err < 0)
477 		err = sk_stream_error(sk, msg->msg_flags, err);
478 	release_sock(sk);
479 	sk_psock_put(sk, psock);
480 	return copied ? copied : err;
481 }
482 
tcp_bpf_sendpage(struct sock * sk,struct page * page,int offset,size_t size,int flags)483 static int tcp_bpf_sendpage(struct sock *sk, struct page *page, int offset,
484 			    size_t size, int flags)
485 {
486 	struct sk_msg tmp, *msg = NULL;
487 	int err = 0, copied = 0;
488 	struct sk_psock *psock;
489 	bool enospc = false;
490 
491 	psock = sk_psock_get(sk);
492 	if (unlikely(!psock))
493 		return tcp_sendpage(sk, page, offset, size, flags);
494 
495 	lock_sock(sk);
496 	if (psock->cork) {
497 		msg = psock->cork;
498 	} else {
499 		msg = &tmp;
500 		sk_msg_init(msg);
501 	}
502 
503 	/* Catch case where ring is full and sendpage is stalled. */
504 	if (unlikely(sk_msg_full(msg)))
505 		goto out_err;
506 
507 	sk_msg_page_add(msg, page, size, offset);
508 	sk_mem_charge(sk, size);
509 	copied = size;
510 	if (sk_msg_full(msg))
511 		enospc = true;
512 	if (psock->cork_bytes) {
513 		if (size > psock->cork_bytes)
514 			psock->cork_bytes = 0;
515 		else
516 			psock->cork_bytes -= size;
517 		if (psock->cork_bytes && !enospc)
518 			goto out_err;
519 		/* All cork bytes are accounted, rerun the prog. */
520 		psock->eval = __SK_NONE;
521 		psock->cork_bytes = 0;
522 	}
523 
524 	err = tcp_bpf_send_verdict(sk, psock, msg, &copied, flags);
525 out_err:
526 	release_sock(sk);
527 	sk_psock_put(sk, psock);
528 	return copied ? copied : err;
529 }
530 
tcp_bpf_remove(struct sock * sk,struct sk_psock * psock)531 static void tcp_bpf_remove(struct sock *sk, struct sk_psock *psock)
532 {
533 	struct sk_psock_link *link;
534 
535 	while ((link = sk_psock_link_pop(psock))) {
536 		sk_psock_unlink(sk, link);
537 		sk_psock_free_link(link);
538 	}
539 }
540 
tcp_bpf_unhash(struct sock * sk)541 static void tcp_bpf_unhash(struct sock *sk)
542 {
543 	void (*saved_unhash)(struct sock *sk);
544 	struct sk_psock *psock;
545 
546 	rcu_read_lock();
547 	psock = sk_psock(sk);
548 	if (unlikely(!psock)) {
549 		rcu_read_unlock();
550 		if (sk->sk_prot->unhash)
551 			sk->sk_prot->unhash(sk);
552 		return;
553 	}
554 
555 	saved_unhash = psock->saved_unhash;
556 	tcp_bpf_remove(sk, psock);
557 	rcu_read_unlock();
558 	saved_unhash(sk);
559 }
560 
tcp_bpf_close(struct sock * sk,long timeout)561 static void tcp_bpf_close(struct sock *sk, long timeout)
562 {
563 	void (*saved_close)(struct sock *sk, long timeout);
564 	struct sk_psock *psock;
565 
566 	lock_sock(sk);
567 	rcu_read_lock();
568 	psock = sk_psock(sk);
569 	if (unlikely(!psock)) {
570 		rcu_read_unlock();
571 		release_sock(sk);
572 		return sk->sk_prot->close(sk, timeout);
573 	}
574 
575 	saved_close = psock->saved_close;
576 	tcp_bpf_remove(sk, psock);
577 	rcu_read_unlock();
578 	release_sock(sk);
579 	saved_close(sk, timeout);
580 }
581 
582 enum {
583 	TCP_BPF_IPV4,
584 	TCP_BPF_IPV6,
585 	TCP_BPF_NUM_PROTS,
586 };
587 
588 enum {
589 	TCP_BPF_BASE,
590 	TCP_BPF_TX,
591 	TCP_BPF_NUM_CFGS,
592 };
593 
594 static struct proto *tcpv6_prot_saved __read_mostly;
595 static DEFINE_SPINLOCK(tcpv6_prot_lock);
596 static struct proto tcp_bpf_prots[TCP_BPF_NUM_PROTS][TCP_BPF_NUM_CFGS];
597 
tcp_bpf_rebuild_protos(struct proto prot[TCP_BPF_NUM_CFGS],struct proto * base)598 static void tcp_bpf_rebuild_protos(struct proto prot[TCP_BPF_NUM_CFGS],
599 				   struct proto *base)
600 {
601 	prot[TCP_BPF_BASE]			= *base;
602 	prot[TCP_BPF_BASE].unhash		= tcp_bpf_unhash;
603 	prot[TCP_BPF_BASE].close		= tcp_bpf_close;
604 	prot[TCP_BPF_BASE].recvmsg		= tcp_bpf_recvmsg;
605 	prot[TCP_BPF_BASE].stream_memory_read	= tcp_bpf_stream_read;
606 
607 	prot[TCP_BPF_TX]			= prot[TCP_BPF_BASE];
608 	prot[TCP_BPF_TX].sendmsg		= tcp_bpf_sendmsg;
609 	prot[TCP_BPF_TX].sendpage		= tcp_bpf_sendpage;
610 }
611 
tcp_bpf_check_v6_needs_rebuild(struct sock * sk,struct proto * ops)612 static void tcp_bpf_check_v6_needs_rebuild(struct sock *sk, struct proto *ops)
613 {
614 	if (sk->sk_family == AF_INET6 &&
615 	    unlikely(ops != smp_load_acquire(&tcpv6_prot_saved))) {
616 		spin_lock_bh(&tcpv6_prot_lock);
617 		if (likely(ops != tcpv6_prot_saved)) {
618 			tcp_bpf_rebuild_protos(tcp_bpf_prots[TCP_BPF_IPV6], ops);
619 			smp_store_release(&tcpv6_prot_saved, ops);
620 		}
621 		spin_unlock_bh(&tcpv6_prot_lock);
622 	}
623 }
624 
tcp_bpf_v4_build_proto(void)625 static int __init tcp_bpf_v4_build_proto(void)
626 {
627 	tcp_bpf_rebuild_protos(tcp_bpf_prots[TCP_BPF_IPV4], &tcp_prot);
628 	return 0;
629 }
630 core_initcall(tcp_bpf_v4_build_proto);
631 
tcp_bpf_update_sk_prot(struct sock * sk,struct sk_psock * psock)632 static void tcp_bpf_update_sk_prot(struct sock *sk, struct sk_psock *psock)
633 {
634 	int family = sk->sk_family == AF_INET6 ? TCP_BPF_IPV6 : TCP_BPF_IPV4;
635 	int config = psock->progs.msg_parser   ? TCP_BPF_TX   : TCP_BPF_BASE;
636 
637 	sk_psock_update_proto(sk, psock, &tcp_bpf_prots[family][config]);
638 }
639 
tcp_bpf_reinit_sk_prot(struct sock * sk,struct sk_psock * psock)640 static void tcp_bpf_reinit_sk_prot(struct sock *sk, struct sk_psock *psock)
641 {
642 	int family = sk->sk_family == AF_INET6 ? TCP_BPF_IPV6 : TCP_BPF_IPV4;
643 	int config = psock->progs.msg_parser   ? TCP_BPF_TX   : TCP_BPF_BASE;
644 
645 	/* Reinit occurs when program types change e.g. TCP_BPF_TX is removed
646 	 * or added requiring sk_prot hook updates. We keep original saved
647 	 * hooks in this case.
648 	 */
649 	sk->sk_prot = &tcp_bpf_prots[family][config];
650 }
651 
tcp_bpf_assert_proto_ops(struct proto * ops)652 static int tcp_bpf_assert_proto_ops(struct proto *ops)
653 {
654 	/* In order to avoid retpoline, we make assumptions when we call
655 	 * into ops if e.g. a psock is not present. Make sure they are
656 	 * indeed valid assumptions.
657 	 */
658 	return ops->recvmsg  == tcp_recvmsg &&
659 	       ops->sendmsg  == tcp_sendmsg &&
660 	       ops->sendpage == tcp_sendpage ? 0 : -ENOTSUPP;
661 }
662 
tcp_bpf_reinit(struct sock * sk)663 void tcp_bpf_reinit(struct sock *sk)
664 {
665 	struct sk_psock *psock;
666 
667 	sock_owned_by_me(sk);
668 
669 	rcu_read_lock();
670 	psock = sk_psock(sk);
671 	tcp_bpf_reinit_sk_prot(sk, psock);
672 	rcu_read_unlock();
673 }
674 
tcp_bpf_init(struct sock * sk)675 int tcp_bpf_init(struct sock *sk)
676 {
677 	struct proto *ops = READ_ONCE(sk->sk_prot);
678 	struct sk_psock *psock;
679 
680 	sock_owned_by_me(sk);
681 
682 	rcu_read_lock();
683 	psock = sk_psock(sk);
684 	if (unlikely(!psock || psock->sk_proto ||
685 		     tcp_bpf_assert_proto_ops(ops))) {
686 		rcu_read_unlock();
687 		return -EINVAL;
688 	}
689 	tcp_bpf_check_v6_needs_rebuild(sk, ops);
690 	tcp_bpf_update_sk_prot(sk, psock);
691 	rcu_read_unlock();
692 	return 0;
693 }
694