xref: /linux/net/ipv4/tcp_bpf.c (revision fbf46565c67c626849c7ce2a326972d3008d2a91)
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 
13 static int bpf_tcp_ingress(struct sock *sk, struct sk_psock *psock,
14 			   struct sk_msg *msg, u32 apply_bytes, int flags)
15 {
16 	bool apply = apply_bytes;
17 	struct scatterlist *sge;
18 	u32 size, copied = 0;
19 	struct sk_msg *tmp;
20 	int i, ret = 0;
21 
22 	tmp = kzalloc(sizeof(*tmp), __GFP_NOWARN | GFP_KERNEL);
23 	if (unlikely(!tmp))
24 		return -ENOMEM;
25 
26 	lock_sock(sk);
27 	tmp->sg.start = msg->sg.start;
28 	i = msg->sg.start;
29 	do {
30 		sge = sk_msg_elem(msg, i);
31 		size = (apply && apply_bytes < sge->length) ?
32 			apply_bytes : sge->length;
33 		if (!sk_wmem_schedule(sk, size)) {
34 			if (!copied)
35 				ret = -ENOMEM;
36 			break;
37 		}
38 
39 		sk_mem_charge(sk, size);
40 		sk_msg_xfer(tmp, msg, i, size);
41 		copied += size;
42 		if (sge->length)
43 			get_page(sk_msg_page(tmp, i));
44 		sk_msg_iter_var_next(i);
45 		tmp->sg.end = i;
46 		if (apply) {
47 			apply_bytes -= size;
48 			if (!apply_bytes) {
49 				if (sge->length)
50 					sk_msg_iter_var_prev(i);
51 				break;
52 			}
53 		}
54 	} while (i != msg->sg.end);
55 
56 	if (!ret) {
57 		msg->sg.start = i;
58 		sk_psock_queue_msg(psock, tmp);
59 		sk_psock_data_ready(sk, psock);
60 	} else {
61 		sk_msg_free(sk, tmp);
62 		kfree(tmp);
63 	}
64 
65 	release_sock(sk);
66 	return ret;
67 }
68 
69 static int tcp_bpf_push(struct sock *sk, struct sk_msg *msg, u32 apply_bytes,
70 			int flags, bool uncharge)
71 {
72 	bool apply = apply_bytes;
73 	struct scatterlist *sge;
74 	struct page *page;
75 	int size, ret = 0;
76 	u32 off;
77 
78 	while (1) {
79 		bool has_tx_ulp;
80 
81 		sge = sk_msg_elem(msg, msg->sg.start);
82 		size = (apply && apply_bytes < sge->length) ?
83 			apply_bytes : sge->length;
84 		off  = sge->offset;
85 		page = sg_page(sge);
86 
87 		tcp_rate_check_app_limited(sk);
88 retry:
89 		has_tx_ulp = tls_sw_has_ctx_tx(sk);
90 		if (has_tx_ulp) {
91 			flags |= MSG_SENDPAGE_NOPOLICY;
92 			ret = kernel_sendpage_locked(sk,
93 						     page, off, size, flags);
94 		} else {
95 			ret = do_tcp_sendpages(sk, page, off, size, flags);
96 		}
97 
98 		if (ret <= 0)
99 			return ret;
100 		if (apply)
101 			apply_bytes -= ret;
102 		msg->sg.size -= ret;
103 		sge->offset += ret;
104 		sge->length -= ret;
105 		if (uncharge)
106 			sk_mem_uncharge(sk, ret);
107 		if (ret != size) {
108 			size -= ret;
109 			off  += ret;
110 			goto retry;
111 		}
112 		if (!sge->length) {
113 			put_page(page);
114 			sk_msg_iter_next(msg, start);
115 			sg_init_table(sge, 1);
116 			if (msg->sg.start == msg->sg.end)
117 				break;
118 		}
119 		if (apply && !apply_bytes)
120 			break;
121 	}
122 
123 	return 0;
124 }
125 
126 static int tcp_bpf_push_locked(struct sock *sk, struct sk_msg *msg,
127 			       u32 apply_bytes, int flags, bool uncharge)
128 {
129 	int ret;
130 
131 	lock_sock(sk);
132 	ret = tcp_bpf_push(sk, msg, apply_bytes, flags, uncharge);
133 	release_sock(sk);
134 	return ret;
135 }
136 
137 int tcp_bpf_sendmsg_redir(struct sock *sk, bool ingress,
138 			  struct sk_msg *msg, u32 bytes, int flags)
139 {
140 	struct sk_psock *psock = sk_psock_get(sk);
141 	int ret;
142 
143 	if (unlikely(!psock))
144 		return -EPIPE;
145 
146 	ret = ingress ? bpf_tcp_ingress(sk, psock, msg, bytes, flags) :
147 			tcp_bpf_push_locked(sk, msg, bytes, flags, false);
148 	sk_psock_put(sk, psock);
149 	return ret;
150 }
151 EXPORT_SYMBOL_GPL(tcp_bpf_sendmsg_redir);
152 
153 #ifdef CONFIG_BPF_SYSCALL
154 static int tcp_msg_wait_data(struct sock *sk, struct sk_psock *psock,
155 			     long timeo)
156 {
157 	DEFINE_WAIT_FUNC(wait, woken_wake_function);
158 	int ret = 0;
159 
160 	if (sk->sk_shutdown & RCV_SHUTDOWN)
161 		return 1;
162 
163 	if (!timeo)
164 		return ret;
165 
166 	add_wait_queue(sk_sleep(sk), &wait);
167 	sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
168 	ret = sk_wait_event(sk, &timeo,
169 			    !list_empty(&psock->ingress_msg) ||
170 			    !skb_queue_empty(&sk->sk_receive_queue), &wait);
171 	sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
172 	remove_wait_queue(sk_sleep(sk), &wait);
173 	return ret;
174 }
175 
176 static int tcp_bpf_recvmsg_parser(struct sock *sk,
177 				  struct msghdr *msg,
178 				  size_t len,
179 				  int flags,
180 				  int *addr_len)
181 {
182 	struct sk_psock *psock;
183 	int copied;
184 
185 	if (unlikely(flags & MSG_ERRQUEUE))
186 		return inet_recv_error(sk, msg, len, addr_len);
187 
188 	psock = sk_psock_get(sk);
189 	if (unlikely(!psock))
190 		return tcp_recvmsg(sk, msg, len, flags, addr_len);
191 
192 	lock_sock(sk);
193 msg_bytes_ready:
194 	copied = sk_msg_recvmsg(sk, psock, msg, len, flags);
195 	if (!copied) {
196 		long timeo;
197 		int data;
198 
199 		if (sock_flag(sk, SOCK_DONE))
200 			goto out;
201 
202 		if (sk->sk_err) {
203 			copied = sock_error(sk);
204 			goto out;
205 		}
206 
207 		if (sk->sk_shutdown & RCV_SHUTDOWN)
208 			goto out;
209 
210 		if (sk->sk_state == TCP_CLOSE) {
211 			copied = -ENOTCONN;
212 			goto out;
213 		}
214 
215 		timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
216 		if (!timeo) {
217 			copied = -EAGAIN;
218 			goto out;
219 		}
220 
221 		if (signal_pending(current)) {
222 			copied = sock_intr_errno(timeo);
223 			goto out;
224 		}
225 
226 		data = tcp_msg_wait_data(sk, psock, timeo);
227 		if (data && !sk_psock_queue_empty(psock))
228 			goto msg_bytes_ready;
229 		copied = -EAGAIN;
230 	}
231 out:
232 	release_sock(sk);
233 	sk_psock_put(sk, psock);
234 	return copied;
235 }
236 
237 static int tcp_bpf_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
238 			   int flags, int *addr_len)
239 {
240 	struct sk_psock *psock;
241 	int copied, ret;
242 
243 	if (unlikely(flags & MSG_ERRQUEUE))
244 		return inet_recv_error(sk, msg, len, addr_len);
245 
246 	psock = sk_psock_get(sk);
247 	if (unlikely(!psock))
248 		return tcp_recvmsg(sk, msg, len, flags, addr_len);
249 	if (!skb_queue_empty(&sk->sk_receive_queue) &&
250 	    sk_psock_queue_empty(psock)) {
251 		sk_psock_put(sk, psock);
252 		return tcp_recvmsg(sk, msg, len, flags, addr_len);
253 	}
254 	lock_sock(sk);
255 msg_bytes_ready:
256 	copied = sk_msg_recvmsg(sk, psock, msg, len, flags);
257 	if (!copied) {
258 		long timeo;
259 		int data;
260 
261 		timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
262 		data = tcp_msg_wait_data(sk, psock, timeo);
263 		if (data) {
264 			if (!sk_psock_queue_empty(psock))
265 				goto msg_bytes_ready;
266 			release_sock(sk);
267 			sk_psock_put(sk, psock);
268 			return tcp_recvmsg(sk, msg, len, flags, addr_len);
269 		}
270 		copied = -EAGAIN;
271 	}
272 	ret = copied;
273 	release_sock(sk);
274 	sk_psock_put(sk, psock);
275 	return ret;
276 }
277 
278 static int tcp_bpf_send_verdict(struct sock *sk, struct sk_psock *psock,
279 				struct sk_msg *msg, int *copied, int flags)
280 {
281 	bool cork = false, enospc = sk_msg_full(msg), redir_ingress;
282 	struct sock *sk_redir;
283 	u32 tosend, origsize, sent, delta = 0;
284 	u32 eval;
285 	int ret;
286 
287 more_data:
288 	if (psock->eval == __SK_NONE) {
289 		/* Track delta in msg size to add/subtract it on SK_DROP from
290 		 * returned to user copied size. This ensures user doesn't
291 		 * get a positive return code with msg_cut_data and SK_DROP
292 		 * verdict.
293 		 */
294 		delta = msg->sg.size;
295 		psock->eval = sk_psock_msg_verdict(sk, psock, msg);
296 		delta -= msg->sg.size;
297 	}
298 
299 	if (msg->cork_bytes &&
300 	    msg->cork_bytes > msg->sg.size && !enospc) {
301 		psock->cork_bytes = msg->cork_bytes - msg->sg.size;
302 		if (!psock->cork) {
303 			psock->cork = kzalloc(sizeof(*psock->cork),
304 					      GFP_ATOMIC | __GFP_NOWARN);
305 			if (!psock->cork)
306 				return -ENOMEM;
307 		}
308 		memcpy(psock->cork, msg, sizeof(*msg));
309 		return 0;
310 	}
311 
312 	tosend = msg->sg.size;
313 	if (psock->apply_bytes && psock->apply_bytes < tosend)
314 		tosend = psock->apply_bytes;
315 	eval = __SK_NONE;
316 
317 	switch (psock->eval) {
318 	case __SK_PASS:
319 		ret = tcp_bpf_push(sk, msg, tosend, flags, true);
320 		if (unlikely(ret)) {
321 			*copied -= sk_msg_free(sk, msg);
322 			break;
323 		}
324 		sk_msg_apply_bytes(psock, tosend);
325 		break;
326 	case __SK_REDIRECT:
327 		redir_ingress = psock->redir_ingress;
328 		sk_redir = psock->sk_redir;
329 		sk_msg_apply_bytes(psock, tosend);
330 		if (!psock->apply_bytes) {
331 			/* Clean up before releasing the sock lock. */
332 			eval = psock->eval;
333 			psock->eval = __SK_NONE;
334 			psock->sk_redir = NULL;
335 		}
336 		if (psock->cork) {
337 			cork = true;
338 			psock->cork = NULL;
339 		}
340 		sk_msg_return(sk, msg, tosend);
341 		release_sock(sk);
342 
343 		origsize = msg->sg.size;
344 		ret = tcp_bpf_sendmsg_redir(sk_redir, redir_ingress,
345 					    msg, tosend, flags);
346 		sent = origsize - msg->sg.size;
347 
348 		if (eval == __SK_REDIRECT)
349 			sock_put(sk_redir);
350 
351 		lock_sock(sk);
352 		if (unlikely(ret < 0)) {
353 			int free = sk_msg_free_nocharge(sk, msg);
354 
355 			if (!cork)
356 				*copied -= free;
357 		}
358 		if (cork) {
359 			sk_msg_free(sk, msg);
360 			kfree(msg);
361 			msg = NULL;
362 			ret = 0;
363 		}
364 		break;
365 	case __SK_DROP:
366 	default:
367 		sk_msg_free_partial(sk, msg, tosend);
368 		sk_msg_apply_bytes(psock, tosend);
369 		*copied -= (tosend + delta);
370 		return -EACCES;
371 	}
372 
373 	if (likely(!ret)) {
374 		if (!psock->apply_bytes) {
375 			psock->eval =  __SK_NONE;
376 			if (psock->sk_redir) {
377 				sock_put(psock->sk_redir);
378 				psock->sk_redir = NULL;
379 			}
380 		}
381 		if (msg &&
382 		    msg->sg.data[msg->sg.start].page_link &&
383 		    msg->sg.data[msg->sg.start].length) {
384 			if (eval == __SK_REDIRECT)
385 				sk_mem_charge(sk, tosend - sent);
386 			goto more_data;
387 		}
388 	}
389 	return ret;
390 }
391 
392 static int tcp_bpf_sendmsg(struct sock *sk, struct msghdr *msg, size_t size)
393 {
394 	struct sk_msg tmp, *msg_tx = NULL;
395 	int copied = 0, err = 0;
396 	struct sk_psock *psock;
397 	long timeo;
398 	int flags;
399 
400 	/* Don't let internal do_tcp_sendpages() flags through */
401 	flags = (msg->msg_flags & ~MSG_SENDPAGE_DECRYPTED);
402 	flags |= MSG_NO_SHARED_FRAGS;
403 
404 	psock = sk_psock_get(sk);
405 	if (unlikely(!psock))
406 		return tcp_sendmsg(sk, msg, size);
407 
408 	lock_sock(sk);
409 	timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
410 	while (msg_data_left(msg)) {
411 		bool enospc = false;
412 		u32 copy, osize;
413 
414 		if (sk->sk_err) {
415 			err = -sk->sk_err;
416 			goto out_err;
417 		}
418 
419 		copy = msg_data_left(msg);
420 		if (!sk_stream_memory_free(sk))
421 			goto wait_for_sndbuf;
422 		if (psock->cork) {
423 			msg_tx = psock->cork;
424 		} else {
425 			msg_tx = &tmp;
426 			sk_msg_init(msg_tx);
427 		}
428 
429 		osize = msg_tx->sg.size;
430 		err = sk_msg_alloc(sk, msg_tx, msg_tx->sg.size + copy, msg_tx->sg.end - 1);
431 		if (err) {
432 			if (err != -ENOSPC)
433 				goto wait_for_memory;
434 			enospc = true;
435 			copy = msg_tx->sg.size - osize;
436 		}
437 
438 		err = sk_msg_memcopy_from_iter(sk, &msg->msg_iter, msg_tx,
439 					       copy);
440 		if (err < 0) {
441 			sk_msg_trim(sk, msg_tx, osize);
442 			goto out_err;
443 		}
444 
445 		copied += copy;
446 		if (psock->cork_bytes) {
447 			if (size > psock->cork_bytes)
448 				psock->cork_bytes = 0;
449 			else
450 				psock->cork_bytes -= size;
451 			if (psock->cork_bytes && !enospc)
452 				goto out_err;
453 			/* All cork bytes are accounted, rerun the prog. */
454 			psock->eval = __SK_NONE;
455 			psock->cork_bytes = 0;
456 		}
457 
458 		err = tcp_bpf_send_verdict(sk, psock, msg_tx, &copied, flags);
459 		if (unlikely(err < 0))
460 			goto out_err;
461 		continue;
462 wait_for_sndbuf:
463 		set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
464 wait_for_memory:
465 		err = sk_stream_wait_memory(sk, &timeo);
466 		if (err) {
467 			if (msg_tx && msg_tx != psock->cork)
468 				sk_msg_free(sk, msg_tx);
469 			goto out_err;
470 		}
471 	}
472 out_err:
473 	if (err < 0)
474 		err = sk_stream_error(sk, msg->msg_flags, err);
475 	release_sock(sk);
476 	sk_psock_put(sk, psock);
477 	return copied ? copied : err;
478 }
479 
480 static int tcp_bpf_sendpage(struct sock *sk, struct page *page, int offset,
481 			    size_t size, int flags)
482 {
483 	struct sk_msg tmp, *msg = NULL;
484 	int err = 0, copied = 0;
485 	struct sk_psock *psock;
486 	bool enospc = false;
487 
488 	psock = sk_psock_get(sk);
489 	if (unlikely(!psock))
490 		return tcp_sendpage(sk, page, offset, size, flags);
491 
492 	lock_sock(sk);
493 	if (psock->cork) {
494 		msg = psock->cork;
495 	} else {
496 		msg = &tmp;
497 		sk_msg_init(msg);
498 	}
499 
500 	/* Catch case where ring is full and sendpage is stalled. */
501 	if (unlikely(sk_msg_full(msg)))
502 		goto out_err;
503 
504 	sk_msg_page_add(msg, page, size, offset);
505 	sk_mem_charge(sk, size);
506 	copied = size;
507 	if (sk_msg_full(msg))
508 		enospc = true;
509 	if (psock->cork_bytes) {
510 		if (size > psock->cork_bytes)
511 			psock->cork_bytes = 0;
512 		else
513 			psock->cork_bytes -= size;
514 		if (psock->cork_bytes && !enospc)
515 			goto out_err;
516 		/* All cork bytes are accounted, rerun the prog. */
517 		psock->eval = __SK_NONE;
518 		psock->cork_bytes = 0;
519 	}
520 
521 	err = tcp_bpf_send_verdict(sk, psock, msg, &copied, flags);
522 out_err:
523 	release_sock(sk);
524 	sk_psock_put(sk, psock);
525 	return copied ? copied : err;
526 }
527 
528 enum {
529 	TCP_BPF_IPV4,
530 	TCP_BPF_IPV6,
531 	TCP_BPF_NUM_PROTS,
532 };
533 
534 enum {
535 	TCP_BPF_BASE,
536 	TCP_BPF_TX,
537 	TCP_BPF_RX,
538 	TCP_BPF_TXRX,
539 	TCP_BPF_NUM_CFGS,
540 };
541 
542 static struct proto *tcpv6_prot_saved __read_mostly;
543 static DEFINE_SPINLOCK(tcpv6_prot_lock);
544 static struct proto tcp_bpf_prots[TCP_BPF_NUM_PROTS][TCP_BPF_NUM_CFGS];
545 
546 static void tcp_bpf_rebuild_protos(struct proto prot[TCP_BPF_NUM_CFGS],
547 				   struct proto *base)
548 {
549 	prot[TCP_BPF_BASE]			= *base;
550 	prot[TCP_BPF_BASE].destroy		= sock_map_destroy;
551 	prot[TCP_BPF_BASE].close		= sock_map_close;
552 	prot[TCP_BPF_BASE].recvmsg		= tcp_bpf_recvmsg;
553 	prot[TCP_BPF_BASE].sock_is_readable	= sk_msg_is_readable;
554 
555 	prot[TCP_BPF_TX]			= prot[TCP_BPF_BASE];
556 	prot[TCP_BPF_TX].sendmsg		= tcp_bpf_sendmsg;
557 	prot[TCP_BPF_TX].sendpage		= tcp_bpf_sendpage;
558 
559 	prot[TCP_BPF_RX]			= prot[TCP_BPF_BASE];
560 	prot[TCP_BPF_RX].recvmsg		= tcp_bpf_recvmsg_parser;
561 
562 	prot[TCP_BPF_TXRX]			= prot[TCP_BPF_TX];
563 	prot[TCP_BPF_TXRX].recvmsg		= tcp_bpf_recvmsg_parser;
564 }
565 
566 static void tcp_bpf_check_v6_needs_rebuild(struct proto *ops)
567 {
568 	if (unlikely(ops != smp_load_acquire(&tcpv6_prot_saved))) {
569 		spin_lock_bh(&tcpv6_prot_lock);
570 		if (likely(ops != tcpv6_prot_saved)) {
571 			tcp_bpf_rebuild_protos(tcp_bpf_prots[TCP_BPF_IPV6], ops);
572 			smp_store_release(&tcpv6_prot_saved, ops);
573 		}
574 		spin_unlock_bh(&tcpv6_prot_lock);
575 	}
576 }
577 
578 static int __init tcp_bpf_v4_build_proto(void)
579 {
580 	tcp_bpf_rebuild_protos(tcp_bpf_prots[TCP_BPF_IPV4], &tcp_prot);
581 	return 0;
582 }
583 late_initcall(tcp_bpf_v4_build_proto);
584 
585 static int tcp_bpf_assert_proto_ops(struct proto *ops)
586 {
587 	/* In order to avoid retpoline, we make assumptions when we call
588 	 * into ops if e.g. a psock is not present. Make sure they are
589 	 * indeed valid assumptions.
590 	 */
591 	return ops->recvmsg  == tcp_recvmsg &&
592 	       ops->sendmsg  == tcp_sendmsg &&
593 	       ops->sendpage == tcp_sendpage ? 0 : -ENOTSUPP;
594 }
595 
596 int tcp_bpf_update_proto(struct sock *sk, struct sk_psock *psock, bool restore)
597 {
598 	int family = sk->sk_family == AF_INET6 ? TCP_BPF_IPV6 : TCP_BPF_IPV4;
599 	int config = psock->progs.msg_parser   ? TCP_BPF_TX   : TCP_BPF_BASE;
600 
601 	if (psock->progs.stream_verdict || psock->progs.skb_verdict) {
602 		config = (config == TCP_BPF_TX) ? TCP_BPF_TXRX : TCP_BPF_RX;
603 	}
604 
605 	if (restore) {
606 		if (inet_csk_has_ulp(sk)) {
607 			/* TLS does not have an unhash proto in SW cases,
608 			 * but we need to ensure we stop using the sock_map
609 			 * unhash routine because the associated psock is being
610 			 * removed. So use the original unhash handler.
611 			 */
612 			WRITE_ONCE(sk->sk_prot->unhash, psock->saved_unhash);
613 			tcp_update_ulp(sk, psock->sk_proto, psock->saved_write_space);
614 		} else {
615 			sk->sk_write_space = psock->saved_write_space;
616 			/* Pairs with lockless read in sk_clone_lock() */
617 			sock_replace_proto(sk, psock->sk_proto);
618 		}
619 		return 0;
620 	}
621 
622 	if (sk->sk_family == AF_INET6) {
623 		if (tcp_bpf_assert_proto_ops(psock->sk_proto))
624 			return -EINVAL;
625 
626 		tcp_bpf_check_v6_needs_rebuild(psock->sk_proto);
627 	}
628 
629 	/* Pairs with lockless read in sk_clone_lock() */
630 	sock_replace_proto(sk, &tcp_bpf_prots[family][config]);
631 	return 0;
632 }
633 EXPORT_SYMBOL_GPL(tcp_bpf_update_proto);
634 
635 /* If a child got cloned from a listening socket that had tcp_bpf
636  * protocol callbacks installed, we need to restore the callbacks to
637  * the default ones because the child does not inherit the psock state
638  * that tcp_bpf callbacks expect.
639  */
640 void tcp_bpf_clone(const struct sock *sk, struct sock *newsk)
641 {
642 	int family = sk->sk_family == AF_INET6 ? TCP_BPF_IPV6 : TCP_BPF_IPV4;
643 	struct proto *prot = newsk->sk_prot;
644 
645 	if (prot == &tcp_bpf_prots[family][TCP_BPF_BASE])
646 		newsk->sk_prot = sk->sk_prot_creator;
647 }
648 #endif /* CONFIG_BPF_SYSCALL */
649