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 #include <asm/ioctls.h>
14
tcp_eat_skb(struct sock * sk,struct sk_buff * skb)15 void tcp_eat_skb(struct sock *sk, struct sk_buff *skb)
16 {
17 struct tcp_sock *tcp;
18 int copied;
19
20 if (!skb || !skb->len || !sk_is_tcp(sk))
21 return;
22
23 if (skb_bpf_strparser(skb))
24 return;
25
26 tcp = tcp_sk(sk);
27 copied = tcp->copied_seq + skb->len;
28 WRITE_ONCE(tcp->copied_seq, copied);
29 tcp_rcv_space_adjust(sk);
30 __tcp_cleanup_rbuf(sk, skb->len);
31 }
32
bpf_tcp_ingress(struct sock * sk,struct sk_psock * psock,struct sk_msg * msg,u32 apply_bytes)33 static int bpf_tcp_ingress(struct sock *sk, struct sk_psock *psock,
34 struct sk_msg *msg, u32 apply_bytes)
35 {
36 bool apply = apply_bytes;
37 struct scatterlist *sge;
38 u32 size, copied = 0;
39 struct sk_msg *tmp;
40 int i, ret = 0;
41
42 tmp = kzalloc_obj(*tmp, __GFP_NOWARN | GFP_KERNEL);
43 if (unlikely(!tmp))
44 return -ENOMEM;
45
46 lock_sock(sk);
47 tmp->sg.start = msg->sg.start;
48 i = msg->sg.start;
49 do {
50 sge = sk_msg_elem(msg, i);
51 size = (apply && apply_bytes < sge->length) ?
52 apply_bytes : sge->length;
53 if (!__sk_rmem_schedule(sk, size, false)) {
54 if (!copied)
55 ret = -ENOMEM;
56 break;
57 }
58
59 sk_mem_charge(sk, size);
60 atomic_add(size, &sk->sk_rmem_alloc);
61 sk_msg_xfer(tmp, msg, i, size);
62 copied += size;
63 if (sge->length)
64 get_page(sk_msg_page(tmp, i));
65 sk_msg_iter_var_next(i);
66 tmp->sg.end = i;
67 if (apply) {
68 apply_bytes -= size;
69 if (!apply_bytes) {
70 if (sge->length)
71 sk_msg_iter_var_prev(i);
72 break;
73 }
74 }
75 } while (i != msg->sg.end);
76
77 if (!ret) {
78 msg->sg.start = i;
79 if (!sk_psock_queue_msg(psock, tmp))
80 atomic_sub(copied, &sk->sk_rmem_alloc);
81 sk_psock_data_ready(sk, psock);
82 } else {
83 sk_msg_free(sk, tmp);
84 kfree(tmp);
85 }
86
87 release_sock(sk);
88 return ret;
89 }
90
tcp_bpf_push(struct sock * sk,struct sk_msg * msg,u32 apply_bytes,int flags,bool uncharge)91 static int tcp_bpf_push(struct sock *sk, struct sk_msg *msg, u32 apply_bytes,
92 int flags, bool uncharge)
93 {
94 struct msghdr msghdr = {};
95 bool apply = apply_bytes;
96 struct scatterlist *sge;
97 struct page *page;
98 int size, ret = 0;
99 u32 off;
100
101 while (1) {
102 struct bio_vec bvec;
103 bool has_tx_ulp;
104
105 sge = sk_msg_elem(msg, msg->sg.start);
106 size = (apply && apply_bytes < sge->length) ?
107 apply_bytes : sge->length;
108 off = sge->offset;
109 page = sg_page(sge);
110
111 tcp_rate_check_app_limited(sk);
112 retry:
113 msghdr.msg_flags = flags | MSG_SPLICE_PAGES;
114 has_tx_ulp = tls_sw_has_ctx_tx(sk);
115 if (has_tx_ulp)
116 msghdr.msg_flags |= MSG_SENDPAGE_NOPOLICY;
117
118 if (size < sge->length && msg->sg.start != msg->sg.end)
119 msghdr.msg_flags |= MSG_MORE;
120
121 bvec_set_page(&bvec, page, size, off);
122 iov_iter_bvec(&msghdr.msg_iter, ITER_SOURCE, &bvec, 1, size);
123 ret = tcp_sendmsg_locked(sk, &msghdr, size);
124 if (ret <= 0)
125 return ret;
126
127 if (apply)
128 apply_bytes -= ret;
129 msg->sg.size -= ret;
130 sge->offset += ret;
131 sge->length -= ret;
132 if (uncharge)
133 sk_mem_uncharge(sk, ret);
134 if (ret != size) {
135 size -= ret;
136 off += ret;
137 goto retry;
138 }
139 if (!sge->length) {
140 put_page(page);
141 sk_msg_iter_next(msg, start);
142 sg_init_table(sge, 1);
143 if (msg->sg.start == msg->sg.end)
144 break;
145 }
146 if (apply && !apply_bytes)
147 break;
148 }
149
150 return 0;
151 }
152
tcp_bpf_push_locked(struct sock * sk,struct sk_msg * msg,u32 apply_bytes,int flags,bool uncharge)153 static int tcp_bpf_push_locked(struct sock *sk, struct sk_msg *msg,
154 u32 apply_bytes, int flags, bool uncharge)
155 {
156 int ret;
157
158 lock_sock(sk);
159 ret = tcp_bpf_push(sk, msg, apply_bytes, flags, uncharge);
160 release_sock(sk);
161 return ret;
162 }
163
tcp_bpf_sendmsg_redir(struct sock * sk,bool ingress,struct sk_msg * msg,u32 bytes,int flags)164 int tcp_bpf_sendmsg_redir(struct sock *sk, bool ingress,
165 struct sk_msg *msg, u32 bytes, int flags)
166 {
167 struct sk_psock *psock = sk_psock_get(sk);
168 int ret;
169
170 if (unlikely(!psock))
171 return -EPIPE;
172
173 ret = ingress ? bpf_tcp_ingress(sk, psock, msg, bytes) :
174 tcp_bpf_push_locked(sk, msg, bytes, flags, false);
175 sk_psock_put(sk, psock);
176 return ret;
177 }
178 EXPORT_SYMBOL_GPL(tcp_bpf_sendmsg_redir);
179
180 #ifdef CONFIG_BPF_SYSCALL
tcp_msg_wait_data(struct sock * sk,struct sk_psock * psock,long timeo)181 static int tcp_msg_wait_data(struct sock *sk, struct sk_psock *psock,
182 long timeo)
183 {
184 DEFINE_WAIT_FUNC(wait, woken_wake_function);
185 int ret = 0;
186
187 if (sk->sk_shutdown & RCV_SHUTDOWN)
188 return 1;
189
190 if (!timeo)
191 return ret;
192
193 add_wait_queue(sk_sleep(sk), &wait);
194 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
195 ret = sk_wait_event(sk, &timeo,
196 !list_empty(&psock->ingress_msg) ||
197 !skb_queue_empty_lockless(&sk->sk_receive_queue), &wait);
198 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
199 remove_wait_queue(sk_sleep(sk), &wait);
200 return ret;
201 }
202
is_next_msg_fin(struct sk_psock * psock)203 static bool is_next_msg_fin(struct sk_psock *psock)
204 {
205 struct scatterlist *sge;
206 struct sk_msg *msg_rx;
207 int i;
208
209 msg_rx = sk_psock_peek_msg(psock);
210 i = msg_rx->sg.start;
211 sge = sk_msg_elem(msg_rx, i);
212 if (!sge->length) {
213 struct sk_buff *skb = msg_rx->skb;
214
215 if (skb && TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN)
216 return true;
217 }
218 return false;
219 }
220
tcp_bpf_recvmsg_parser(struct sock * sk,struct msghdr * msg,size_t len,int flags,int * addr_len)221 static int tcp_bpf_recvmsg_parser(struct sock *sk,
222 struct msghdr *msg,
223 size_t len,
224 int flags,
225 int *addr_len)
226 {
227 int peek = flags & MSG_PEEK;
228 struct sk_psock *psock;
229 struct tcp_sock *tcp;
230 int copied_from_self = 0;
231 int copied = 0;
232 u32 seq;
233
234 if (unlikely(flags & MSG_ERRQUEUE))
235 return inet_recv_error(sk, msg, len, addr_len);
236
237 if (!len)
238 return 0;
239
240 psock = sk_psock_get(sk);
241 if (unlikely(!psock))
242 return tcp_recvmsg(sk, msg, len, flags, addr_len);
243
244 lock_sock(sk);
245 tcp = tcp_sk(sk);
246 seq = tcp->copied_seq;
247 /* We may have received data on the sk_receive_queue pre-accept and
248 * then we can not use read_skb in this context because we haven't
249 * assigned a sk_socket yet so have no link to the ops. The work-around
250 * is to check the sk_receive_queue and in these cases read skbs off
251 * queue again. The read_skb hook is not running at this point because
252 * of lock_sock so we avoid having multiple runners in read_skb.
253 */
254 if (unlikely(!skb_queue_empty(&sk->sk_receive_queue))) {
255 tcp_data_ready(sk);
256 /* This handles the ENOMEM errors if we both receive data
257 * pre accept and are already under memory pressure. At least
258 * let user know to retry.
259 */
260 if (unlikely(!skb_queue_empty(&sk->sk_receive_queue))) {
261 copied = -EAGAIN;
262 goto out;
263 }
264 }
265
266 msg_bytes_ready:
267 copied = __sk_msg_recvmsg(sk, psock, msg, len, flags, &copied_from_self);
268 /* The typical case for EFAULT is the socket was gracefully
269 * shutdown with a FIN pkt. So check here the other case is
270 * some error on copy_page_to_iter which would be unexpected.
271 * On fin return correct return code to zero.
272 */
273 if (copied == -EFAULT) {
274 bool is_fin = is_next_msg_fin(psock);
275
276 if (is_fin) {
277 copied = 0;
278 seq++;
279 goto out;
280 }
281 }
282 seq += copied_from_self;
283 if (!copied) {
284 long timeo;
285 int data;
286
287 if (sock_flag(sk, SOCK_DONE))
288 goto out;
289
290 if (sk->sk_err) {
291 copied = sock_error(sk);
292 goto out;
293 }
294
295 if (sk->sk_shutdown & RCV_SHUTDOWN)
296 goto out;
297
298 if (sk->sk_state == TCP_CLOSE) {
299 copied = -ENOTCONN;
300 goto out;
301 }
302
303 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
304 if (!timeo) {
305 copied = -EAGAIN;
306 goto out;
307 }
308
309 if (signal_pending(current)) {
310 copied = sock_intr_errno(timeo);
311 goto out;
312 }
313
314 data = tcp_msg_wait_data(sk, psock, timeo);
315 if (data < 0) {
316 copied = data;
317 goto unlock;
318 }
319 if (data && !sk_psock_queue_empty(psock))
320 goto msg_bytes_ready;
321 copied = -EAGAIN;
322 }
323 out:
324 if (!peek)
325 WRITE_ONCE(tcp->copied_seq, seq);
326 tcp_rcv_space_adjust(sk);
327 if (copied > 0)
328 __tcp_cleanup_rbuf(sk, copied);
329
330 unlock:
331 release_sock(sk);
332 sk_psock_put(sk, psock);
333 return copied;
334 }
335
tcp_bpf_ioctl(struct sock * sk,int cmd,int * karg)336 static int tcp_bpf_ioctl(struct sock *sk, int cmd, int *karg)
337 {
338 bool slow;
339
340 if (cmd != SIOCINQ)
341 return tcp_ioctl(sk, cmd, karg);
342
343 /* works similar as tcp_ioctl */
344 if (sk->sk_state == TCP_LISTEN)
345 return -EINVAL;
346
347 slow = lock_sock_fast(sk);
348 *karg = sk_psock_msg_inq(sk);
349 unlock_sock_fast(sk, slow);
350
351 return 0;
352 }
353
tcp_bpf_recvmsg(struct sock * sk,struct msghdr * msg,size_t len,int flags,int * addr_len)354 static int tcp_bpf_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
355 int flags, int *addr_len)
356 {
357 struct sk_psock *psock;
358 int copied, ret;
359
360 if (unlikely(flags & MSG_ERRQUEUE))
361 return inet_recv_error(sk, msg, len, addr_len);
362
363 if (!len)
364 return 0;
365
366 psock = sk_psock_get(sk);
367 if (unlikely(!psock))
368 return tcp_recvmsg(sk, msg, len, flags, addr_len);
369 if (!skb_queue_empty(&sk->sk_receive_queue) &&
370 sk_psock_queue_empty(psock)) {
371 sk_psock_put(sk, psock);
372 return tcp_recvmsg(sk, msg, len, flags, addr_len);
373 }
374 lock_sock(sk);
375 msg_bytes_ready:
376 copied = sk_msg_recvmsg(sk, psock, msg, len, flags);
377 if (!copied) {
378 long timeo;
379 int data;
380
381 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
382 data = tcp_msg_wait_data(sk, psock, timeo);
383 if (data < 0) {
384 ret = data;
385 goto unlock;
386 }
387 if (data) {
388 if (!sk_psock_queue_empty(psock))
389 goto msg_bytes_ready;
390 release_sock(sk);
391 sk_psock_put(sk, psock);
392 return tcp_recvmsg(sk, msg, len, flags, addr_len);
393 }
394 copied = -EAGAIN;
395 }
396 ret = copied;
397
398 unlock:
399 release_sock(sk);
400 sk_psock_put(sk, psock);
401 return ret;
402 }
403
tcp_bpf_send_verdict(struct sock * sk,struct sk_psock * psock,struct sk_msg * msg,int * copied,int flags)404 static int tcp_bpf_send_verdict(struct sock *sk, struct sk_psock *psock,
405 struct sk_msg *msg, int *copied, int flags)
406 {
407 bool cork = false, enospc = sk_msg_full(msg), redir_ingress;
408 struct sock *sk_redir;
409 u32 tosend, origsize, sent, delta = 0;
410 u32 eval;
411 int ret;
412
413 more_data:
414 if (psock->eval == __SK_NONE) {
415 /* Track delta in msg size to add/subtract it on SK_DROP from
416 * returned to user copied size. This ensures user doesn't
417 * get a positive return code with msg_cut_data and SK_DROP
418 * verdict.
419 */
420 delta = msg->sg.size;
421 psock->eval = sk_psock_msg_verdict(sk, psock, msg);
422 delta -= msg->sg.size;
423 }
424
425 if (msg->cork_bytes &&
426 msg->cork_bytes > msg->sg.size && !enospc) {
427 psock->cork_bytes = msg->cork_bytes - msg->sg.size;
428 if (!psock->cork) {
429 psock->cork = kzalloc_obj(*psock->cork,
430 GFP_ATOMIC | __GFP_NOWARN);
431 if (!psock->cork) {
432 sk_msg_free(sk, msg);
433 *copied = 0;
434 return -ENOMEM;
435 }
436 }
437 memcpy(psock->cork, msg, sizeof(*msg));
438 return 0;
439 }
440
441 tosend = msg->sg.size;
442 if (psock->apply_bytes && psock->apply_bytes < tosend)
443 tosend = psock->apply_bytes;
444 eval = __SK_NONE;
445
446 switch (psock->eval) {
447 case __SK_PASS:
448 ret = tcp_bpf_push(sk, msg, tosend, flags, true);
449 if (unlikely(ret)) {
450 *copied -= sk_msg_free(sk, msg);
451 break;
452 }
453 sk_msg_apply_bytes(psock, tosend);
454 break;
455 case __SK_REDIRECT:
456 redir_ingress = psock->redir_ingress;
457 sk_redir = psock->sk_redir;
458 sk_msg_apply_bytes(psock, tosend);
459 if (!psock->apply_bytes) {
460 /* Clean up before releasing the sock lock. */
461 eval = psock->eval;
462 psock->eval = __SK_NONE;
463 psock->sk_redir = NULL;
464 }
465 if (psock->cork) {
466 cork = true;
467 psock->cork = NULL;
468 }
469 release_sock(sk);
470
471 origsize = msg->sg.size;
472 ret = tcp_bpf_sendmsg_redir(sk_redir, redir_ingress,
473 msg, tosend, flags);
474 sent = origsize - msg->sg.size;
475
476 if (eval == __SK_REDIRECT)
477 sock_put(sk_redir);
478
479 lock_sock(sk);
480 sk_mem_uncharge(sk, sent);
481 if (unlikely(ret < 0)) {
482 int free = sk_msg_free(sk, msg);
483
484 if (!cork)
485 *copied -= free;
486 }
487 if (cork) {
488 sk_msg_free(sk, msg);
489 kfree(msg);
490 msg = NULL;
491 ret = 0;
492 }
493 break;
494 case __SK_DROP:
495 default:
496 sk_msg_free(sk, msg);
497 sk_msg_apply_bytes(psock, tosend);
498 *copied -= (tosend + delta);
499 return -EACCES;
500 }
501
502 if (likely(!ret)) {
503 if (!psock->apply_bytes) {
504 psock->eval = __SK_NONE;
505 if (psock->sk_redir) {
506 sock_put(psock->sk_redir);
507 psock->sk_redir = NULL;
508 }
509 }
510 if (msg &&
511 msg->sg.data[msg->sg.start].page_link &&
512 msg->sg.data[msg->sg.start].length)
513 goto more_data;
514 }
515 return ret;
516 }
517
tcp_bpf_sendmsg(struct sock * sk,struct msghdr * msg,size_t size)518 static int tcp_bpf_sendmsg(struct sock *sk, struct msghdr *msg, size_t size)
519 {
520 struct sk_msg tmp, *msg_tx = NULL;
521 int copied = 0, err = 0, ret = 0;
522 struct sk_psock *psock;
523 long timeo;
524 int flags;
525
526 /* Don't let internal flags through */
527 flags = (msg->msg_flags & ~MSG_SENDPAGE_DECRYPTED);
528 flags |= MSG_NO_SHARED_FRAGS;
529
530 psock = sk_psock_get(sk);
531 if (unlikely(!psock))
532 return tcp_sendmsg(sk, msg, size);
533
534 lock_sock(sk);
535 timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
536 while (msg_data_left(msg)) {
537 bool enospc = false;
538 u32 copy, osize;
539
540 if (sk->sk_err) {
541 err = -sk->sk_err;
542 goto out_err;
543 }
544
545 copy = msg_data_left(msg);
546 if (!sk_stream_memory_free(sk))
547 goto wait_for_sndbuf;
548 if (psock->cork) {
549 msg_tx = psock->cork;
550 } else {
551 msg_tx = &tmp;
552 sk_msg_init(msg_tx);
553 }
554
555 osize = msg_tx->sg.size;
556 err = sk_msg_alloc(sk, msg_tx, msg_tx->sg.size + copy, msg_tx->sg.end - 1);
557 if (err) {
558 if (err != -ENOSPC)
559 goto wait_for_memory;
560 enospc = true;
561 copy = msg_tx->sg.size - osize;
562 }
563
564 ret = sk_msg_memcopy_from_iter(sk, &msg->msg_iter, msg_tx,
565 copy);
566 if (ret < 0) {
567 sk_msg_trim(sk, msg_tx, osize);
568 goto out_err;
569 }
570
571 copied += ret;
572 if (psock->cork_bytes) {
573 if (size > psock->cork_bytes)
574 psock->cork_bytes = 0;
575 else
576 psock->cork_bytes -= size;
577 if (psock->cork_bytes && !enospc)
578 goto out_err;
579 /* All cork bytes are accounted, rerun the prog. */
580 psock->eval = __SK_NONE;
581 psock->cork_bytes = 0;
582 }
583
584 err = tcp_bpf_send_verdict(sk, psock, msg_tx, &copied, flags);
585 if (unlikely(err < 0))
586 goto out_err;
587 continue;
588 wait_for_sndbuf:
589 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
590 wait_for_memory:
591 err = sk_stream_wait_memory(sk, &timeo);
592 if (err) {
593 if (msg_tx && msg_tx != psock->cork)
594 sk_msg_free(sk, msg_tx);
595 goto out_err;
596 }
597 }
598 out_err:
599 if (err < 0)
600 err = sk_stream_error(sk, msg->msg_flags, err);
601 release_sock(sk);
602 sk_psock_put(sk, psock);
603 return copied > 0 ? copied : err;
604 }
605
606 enum {
607 TCP_BPF_IPV4,
608 TCP_BPF_IPV6,
609 TCP_BPF_NUM_PROTS,
610 };
611
612 enum {
613 TCP_BPF_BASE,
614 TCP_BPF_TX,
615 TCP_BPF_RX,
616 TCP_BPF_TXRX,
617 TCP_BPF_NUM_CFGS,
618 };
619
620 static struct proto *tcpv6_prot_saved __read_mostly;
621 static DEFINE_SPINLOCK(tcpv6_prot_lock);
622 static struct proto tcp_bpf_prots[TCP_BPF_NUM_PROTS][TCP_BPF_NUM_CFGS];
623
tcp_bpf_rebuild_protos(struct proto prot[TCP_BPF_NUM_CFGS],struct proto * base)624 static void tcp_bpf_rebuild_protos(struct proto prot[TCP_BPF_NUM_CFGS],
625 struct proto *base)
626 {
627 prot[TCP_BPF_BASE] = *base;
628 prot[TCP_BPF_BASE].destroy = sock_map_destroy;
629 prot[TCP_BPF_BASE].close = sock_map_close;
630 prot[TCP_BPF_BASE].recvmsg = tcp_bpf_recvmsg;
631 prot[TCP_BPF_BASE].sock_is_readable = sk_msg_is_readable;
632 prot[TCP_BPF_BASE].ioctl = tcp_bpf_ioctl;
633
634 prot[TCP_BPF_TX] = prot[TCP_BPF_BASE];
635 prot[TCP_BPF_TX].sendmsg = tcp_bpf_sendmsg;
636
637 prot[TCP_BPF_RX] = prot[TCP_BPF_BASE];
638 prot[TCP_BPF_RX].recvmsg = tcp_bpf_recvmsg_parser;
639
640 prot[TCP_BPF_TXRX] = prot[TCP_BPF_TX];
641 prot[TCP_BPF_TXRX].recvmsg = tcp_bpf_recvmsg_parser;
642 }
643
tcp_bpf_check_v6_needs_rebuild(struct proto * ops)644 static void tcp_bpf_check_v6_needs_rebuild(struct proto *ops)
645 {
646 if (unlikely(ops != smp_load_acquire(&tcpv6_prot_saved))) {
647 spin_lock_bh(&tcpv6_prot_lock);
648 if (likely(ops != tcpv6_prot_saved)) {
649 tcp_bpf_rebuild_protos(tcp_bpf_prots[TCP_BPF_IPV6], ops);
650 smp_store_release(&tcpv6_prot_saved, ops);
651 }
652 spin_unlock_bh(&tcpv6_prot_lock);
653 }
654 }
655
tcp_bpf_v4_build_proto(void)656 static int __init tcp_bpf_v4_build_proto(void)
657 {
658 tcp_bpf_rebuild_protos(tcp_bpf_prots[TCP_BPF_IPV4], &tcp_prot);
659 return 0;
660 }
661 late_initcall(tcp_bpf_v4_build_proto);
662
tcp_bpf_assert_proto_ops(struct proto * ops)663 static int tcp_bpf_assert_proto_ops(struct proto *ops)
664 {
665 /* In order to avoid retpoline, we make assumptions when we call
666 * into ops if e.g. a psock is not present. Make sure they are
667 * indeed valid assumptions.
668 */
669 return ops->recvmsg == tcp_recvmsg &&
670 ops->sendmsg == tcp_sendmsg ? 0 : -ENOTSUPP;
671 }
672
673 #if IS_ENABLED(CONFIG_BPF_STREAM_PARSER)
tcp_bpf_strp_read_sock(struct strparser * strp,read_descriptor_t * desc,sk_read_actor_t recv_actor)674 int tcp_bpf_strp_read_sock(struct strparser *strp, read_descriptor_t *desc,
675 sk_read_actor_t recv_actor)
676 {
677 struct sock *sk = strp->sk;
678 struct sk_psock *psock;
679 struct tcp_sock *tp;
680 int copied = 0;
681
682 tp = tcp_sk(sk);
683 rcu_read_lock();
684 psock = sk_psock(sk);
685 if (WARN_ON_ONCE(!psock)) {
686 desc->error = -EINVAL;
687 goto out;
688 }
689
690 psock->ingress_bytes = 0;
691 copied = tcp_read_sock_noack(sk, desc, recv_actor, true,
692 &psock->copied_seq);
693 if (copied < 0)
694 goto out;
695 /* recv_actor may redirect skb to another socket (SK_REDIRECT) or
696 * just put skb into ingress queue of current socket (SK_PASS).
697 * For SK_REDIRECT, we need to ack the frame immediately but for
698 * SK_PASS, we want to delay the ack until tcp_bpf_recvmsg_parser().
699 */
700 tp->copied_seq = psock->copied_seq - psock->ingress_bytes;
701 tcp_rcv_space_adjust(sk);
702 __tcp_cleanup_rbuf(sk, copied - psock->ingress_bytes);
703 out:
704 rcu_read_unlock();
705 return copied;
706 }
707 #endif /* CONFIG_BPF_STREAM_PARSER */
708
tcp_bpf_update_proto(struct sock * sk,struct sk_psock * psock,bool restore)709 int tcp_bpf_update_proto(struct sock *sk, struct sk_psock *psock, bool restore)
710 {
711 int family = sk->sk_family == AF_INET6 ? TCP_BPF_IPV6 : TCP_BPF_IPV4;
712 int config = psock->progs.msg_parser ? TCP_BPF_TX : TCP_BPF_BASE;
713
714 if (psock->progs.stream_verdict || psock->progs.skb_verdict) {
715 config = (config == TCP_BPF_TX) ? TCP_BPF_TXRX : TCP_BPF_RX;
716 }
717
718 if (restore) {
719 if (inet_csk_has_ulp(sk)) {
720 /* TLS does not have an unhash proto in SW cases,
721 * but we need to ensure we stop using the sock_map
722 * unhash routine because the associated psock is being
723 * removed. So use the original unhash handler.
724 */
725 WRITE_ONCE(sk->sk_prot->unhash, psock->saved_unhash);
726 tcp_update_ulp(sk, psock->sk_proto, psock->saved_write_space);
727 } else {
728 WRITE_ONCE(sk->sk_write_space, psock->saved_write_space);
729 /* Pairs with lockless read in sk_clone_lock() */
730 sock_replace_proto(sk, psock->sk_proto);
731 }
732 return 0;
733 }
734
735 if (sk->sk_family == AF_INET6) {
736 if (tcp_bpf_assert_proto_ops(psock->sk_proto))
737 return -EINVAL;
738
739 tcp_bpf_check_v6_needs_rebuild(psock->sk_proto);
740 }
741
742 /* Pairs with lockless read in sk_clone_lock() */
743 sock_replace_proto(sk, &tcp_bpf_prots[family][config]);
744 return 0;
745 }
746 EXPORT_SYMBOL_GPL(tcp_bpf_update_proto);
747
748 /* If a child got cloned from a listening socket that had tcp_bpf
749 * protocol callbacks installed, we need to restore the callbacks to
750 * the default ones because the child does not inherit the psock state
751 * that tcp_bpf callbacks expect.
752 */
tcp_bpf_clone(const struct sock * sk,struct sock * newsk)753 void tcp_bpf_clone(const struct sock *sk, struct sock *newsk)
754 {
755 struct proto *prot = newsk->sk_prot;
756
757 if (is_insidevar(prot, tcp_bpf_prots))
758 newsk->sk_prot = sk->sk_prot_creator;
759 }
760 #endif /* CONFIG_BPF_SYSCALL */
761