1 // SPDX-License-Identifier: GPL-2.0
2 /* Copyright (c) 2017 - 2018 Covalent IO, Inc. http://covalent.io */
4 #include <linux/skmsg.h>
5 #include <linux/filter.h>
7 #include <linux/init.h>
8 #include <linux/wait.h>
10 #include <net/inet_common.h>
13 static bool tcp_bpf_stream_read(const struct sock *sk)
15 struct sk_psock *psock;
21 empty = list_empty(&psock->ingress_msg);
26 static int tcp_bpf_wait_data(struct sock *sk, struct sk_psock *psock,
27 int flags, long timeo, int *err)
29 DEFINE_WAIT_FUNC(wait, woken_wake_function);
32 add_wait_queue(sk_sleep(sk), &wait);
33 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
34 ret = sk_wait_event(sk, &timeo,
35 !list_empty(&psock->ingress_msg) ||
36 !skb_queue_empty(&sk->sk_receive_queue), &wait);
37 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
38 remove_wait_queue(sk_sleep(sk), &wait);
42 int __tcp_bpf_recvmsg(struct sock *sk, struct sk_psock *psock,
43 struct msghdr *msg, int len, int flags)
45 struct iov_iter *iter = &msg->msg_iter;
46 int peek = flags & MSG_PEEK;
47 int i, ret, copied = 0;
48 struct sk_msg *msg_rx;
50 msg_rx = list_first_entry_or_null(&psock->ingress_msg,
53 while (copied != len) {
54 struct scatterlist *sge;
56 if (unlikely(!msg_rx))
64 sge = sk_msg_elem(msg_rx, i);
67 if (copied + copy > len)
69 ret = copy_page_to_iter(page, sge->offset, copy, iter);
79 sk_mem_uncharge(sk, copy);
80 msg_rx->sg.size -= copy;
83 sk_msg_iter_var_next(i);
88 sk_msg_iter_var_next(i);
93 } while (i != msg_rx->sg.end);
96 msg_rx = list_next_entry(msg_rx, list);
100 msg_rx->sg.start = i;
101 if (!sge->length && msg_rx->sg.start == msg_rx->sg.end) {
102 list_del(&msg_rx->list);
104 consume_skb(msg_rx->skb);
107 msg_rx = list_first_entry_or_null(&psock->ingress_msg,
108 struct sk_msg, list);
113 EXPORT_SYMBOL_GPL(__tcp_bpf_recvmsg);
115 int tcp_bpf_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
116 int nonblock, int flags, int *addr_len)
118 struct sk_psock *psock;
121 if (unlikely(flags & MSG_ERRQUEUE))
122 return inet_recv_error(sk, msg, len, addr_len);
123 if (!skb_queue_empty(&sk->sk_receive_queue))
124 return tcp_recvmsg(sk, msg, len, nonblock, flags, addr_len);
126 psock = sk_psock_get(sk);
127 if (unlikely(!psock))
128 return tcp_recvmsg(sk, msg, len, nonblock, flags, addr_len);
131 copied = __tcp_bpf_recvmsg(sk, psock, msg, len, flags);
136 timeo = sock_rcvtimeo(sk, nonblock);
137 data = tcp_bpf_wait_data(sk, psock, flags, timeo, &err);
139 if (skb_queue_empty(&sk->sk_receive_queue))
140 goto msg_bytes_ready;
142 sk_psock_put(sk, psock);
143 return tcp_recvmsg(sk, msg, len, nonblock, flags, addr_len);
154 sk_psock_put(sk, psock);
158 static int bpf_tcp_ingress(struct sock *sk, struct sk_psock *psock,
159 struct sk_msg *msg, u32 apply_bytes, int flags)
161 bool apply = apply_bytes;
162 struct scatterlist *sge;
163 u32 size, copied = 0;
167 tmp = kzalloc(sizeof(*tmp), __GFP_NOWARN | GFP_KERNEL);
172 tmp->sg.start = msg->sg.start;
175 sge = sk_msg_elem(msg, i);
176 size = (apply && apply_bytes < sge->length) ?
177 apply_bytes : sge->length;
178 if (!sk_wmem_schedule(sk, size)) {
184 sk_mem_charge(sk, size);
185 sk_msg_xfer(tmp, msg, i, size);
188 get_page(sk_msg_page(tmp, i));
189 sk_msg_iter_var_next(i);
196 } while (i != msg->sg.end);
200 msg->sg.size -= apply_bytes;
201 sk_psock_queue_msg(psock, tmp);
202 sk_psock_data_ready(sk, psock);
204 sk_msg_free(sk, tmp);
212 static int tcp_bpf_push(struct sock *sk, struct sk_msg *msg, u32 apply_bytes,
213 int flags, bool uncharge)
215 bool apply = apply_bytes;
216 struct scatterlist *sge;
224 sge = sk_msg_elem(msg, msg->sg.start);
225 size = (apply && apply_bytes < sge->length) ?
226 apply_bytes : sge->length;
230 tcp_rate_check_app_limited(sk);
232 has_tx_ulp = tls_sw_has_ctx_tx(sk);
234 flags |= MSG_SENDPAGE_NOPOLICY;
235 ret = kernel_sendpage_locked(sk,
236 page, off, size, flags);
238 ret = do_tcp_sendpages(sk, page, off, size, flags);
249 sk_mem_uncharge(sk, ret);
257 sk_msg_iter_next(msg, start);
258 sg_init_table(sge, 1);
259 if (msg->sg.start == msg->sg.end)
262 if (apply && !apply_bytes)
269 static int tcp_bpf_push_locked(struct sock *sk, struct sk_msg *msg,
270 u32 apply_bytes, int flags, bool uncharge)
275 ret = tcp_bpf_push(sk, msg, apply_bytes, flags, uncharge);
280 int tcp_bpf_sendmsg_redir(struct sock *sk, struct sk_msg *msg,
281 u32 bytes, int flags)
283 bool ingress = sk_msg_to_ingress(msg);
284 struct sk_psock *psock = sk_psock_get(sk);
287 if (unlikely(!psock)) {
288 sk_msg_free(sk, msg);
291 ret = ingress ? bpf_tcp_ingress(sk, psock, msg, bytes, flags) :
292 tcp_bpf_push_locked(sk, msg, bytes, flags, false);
293 sk_psock_put(sk, psock);
296 EXPORT_SYMBOL_GPL(tcp_bpf_sendmsg_redir);
298 static int tcp_bpf_send_verdict(struct sock *sk, struct sk_psock *psock,
299 struct sk_msg *msg, int *copied, int flags)
301 bool cork = false, enospc = msg->sg.start == msg->sg.end;
302 struct sock *sk_redir;
303 u32 tosend, delta = 0;
307 if (psock->eval == __SK_NONE) {
308 /* Track delta in msg size to add/subtract it on SK_DROP from
309 * returned to user copied size. This ensures user doesn't
310 * get a positive return code with msg_cut_data and SK_DROP
313 delta = msg->sg.size;
314 psock->eval = sk_psock_msg_verdict(sk, psock, msg);
315 if (msg->sg.size < delta)
316 delta -= msg->sg.size;
321 if (msg->cork_bytes &&
322 msg->cork_bytes > msg->sg.size && !enospc) {
323 psock->cork_bytes = msg->cork_bytes - msg->sg.size;
325 psock->cork = kzalloc(sizeof(*psock->cork),
326 GFP_ATOMIC | __GFP_NOWARN);
330 memcpy(psock->cork, msg, sizeof(*msg));
334 tosend = msg->sg.size;
335 if (psock->apply_bytes && psock->apply_bytes < tosend)
336 tosend = psock->apply_bytes;
338 switch (psock->eval) {
340 ret = tcp_bpf_push(sk, msg, tosend, flags, true);
342 *copied -= sk_msg_free(sk, msg);
345 sk_msg_apply_bytes(psock, tosend);
348 sk_redir = psock->sk_redir;
349 sk_msg_apply_bytes(psock, tosend);
354 sk_msg_return(sk, msg, tosend);
356 ret = tcp_bpf_sendmsg_redir(sk_redir, msg, tosend, flags);
358 if (unlikely(ret < 0)) {
359 int free = sk_msg_free_nocharge(sk, msg);
365 sk_msg_free(sk, msg);
373 sk_msg_free_partial(sk, msg, tosend);
374 sk_msg_apply_bytes(psock, tosend);
375 *copied -= (tosend + delta);
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;
388 msg->sg.data[msg->sg.start].page_link &&
389 msg->sg.data[msg->sg.start].length)
395 static int tcp_bpf_sendmsg(struct sock *sk, struct msghdr *msg, size_t size)
397 struct sk_msg tmp, *msg_tx = NULL;
398 int flags = msg->msg_flags | MSG_NO_SHARED_FRAGS;
399 int copied = 0, err = 0;
400 struct sk_psock *psock;
403 psock = sk_psock_get(sk);
404 if (unlikely(!psock))
405 return tcp_sendmsg(sk, msg, size);
408 timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
409 while (msg_data_left(msg)) {
418 copy = msg_data_left(msg);
419 if (!sk_stream_memory_free(sk))
420 goto wait_for_sndbuf;
422 msg_tx = psock->cork;
428 osize = msg_tx->sg.size;
429 err = sk_msg_alloc(sk, msg_tx, msg_tx->sg.size + copy, msg_tx->sg.end - 1);
432 goto wait_for_memory;
434 copy = msg_tx->sg.size - osize;
437 err = sk_msg_memcopy_from_iter(sk, &msg->msg_iter, msg_tx,
440 sk_msg_trim(sk, msg_tx, osize);
445 if (psock->cork_bytes) {
446 if (size > psock->cork_bytes)
447 psock->cork_bytes = 0;
449 psock->cork_bytes -= size;
450 if (psock->cork_bytes && !enospc)
452 /* All cork bytes are accounted, rerun the prog. */
453 psock->eval = __SK_NONE;
454 psock->cork_bytes = 0;
457 err = tcp_bpf_send_verdict(sk, psock, msg_tx, &copied, flags);
458 if (unlikely(err < 0))
462 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
464 err = sk_stream_wait_memory(sk, &timeo);
466 if (msg_tx && msg_tx != psock->cork)
467 sk_msg_free(sk, msg_tx);
473 err = sk_stream_error(sk, msg->msg_flags, err);
475 sk_psock_put(sk, psock);
476 return copied ? copied : err;
479 static int tcp_bpf_sendpage(struct sock *sk, struct page *page, int offset,
480 size_t size, int flags)
482 struct sk_msg tmp, *msg = NULL;
483 int err = 0, copied = 0;
484 struct sk_psock *psock;
487 psock = sk_psock_get(sk);
488 if (unlikely(!psock))
489 return tcp_sendpage(sk, page, offset, size, flags);
499 /* Catch case where ring is full and sendpage is stalled. */
500 if (unlikely(sk_msg_full(msg)))
503 sk_msg_page_add(msg, page, size, offset);
504 sk_mem_charge(sk, size);
506 if (sk_msg_full(msg))
508 if (psock->cork_bytes) {
509 if (size > psock->cork_bytes)
510 psock->cork_bytes = 0;
512 psock->cork_bytes -= size;
513 if (psock->cork_bytes && !enospc)
515 /* All cork bytes are accounted, rerun the prog. */
516 psock->eval = __SK_NONE;
517 psock->cork_bytes = 0;
520 err = tcp_bpf_send_verdict(sk, psock, msg, &copied, flags);
523 sk_psock_put(sk, psock);
524 return copied ? copied : err;
527 static void tcp_bpf_remove(struct sock *sk, struct sk_psock *psock)
529 struct sk_psock_link *link;
531 sk_psock_cork_free(psock);
532 __sk_psock_purge_ingress_msg(psock);
533 while ((link = sk_psock_link_pop(psock))) {
534 sk_psock_unlink(sk, link);
535 sk_psock_free_link(link);
539 static void tcp_bpf_unhash(struct sock *sk)
541 void (*saved_unhash)(struct sock *sk);
542 struct sk_psock *psock;
545 psock = sk_psock(sk);
546 if (unlikely(!psock)) {
548 if (sk->sk_prot->unhash)
549 sk->sk_prot->unhash(sk);
553 saved_unhash = psock->saved_unhash;
554 tcp_bpf_remove(sk, psock);
559 static void tcp_bpf_close(struct sock *sk, long timeout)
561 void (*saved_close)(struct sock *sk, long timeout);
562 struct sk_psock *psock;
566 psock = sk_psock(sk);
567 if (unlikely(!psock)) {
570 return sk->sk_prot->close(sk, timeout);
573 saved_close = psock->saved_close;
574 tcp_bpf_remove(sk, psock);
577 saved_close(sk, timeout);
592 static struct proto *tcpv6_prot_saved __read_mostly;
593 static DEFINE_SPINLOCK(tcpv6_prot_lock);
594 static struct proto tcp_bpf_prots[TCP_BPF_NUM_PROTS][TCP_BPF_NUM_CFGS];
596 static void tcp_bpf_rebuild_protos(struct proto prot[TCP_BPF_NUM_CFGS],
599 prot[TCP_BPF_BASE] = *base;
600 prot[TCP_BPF_BASE].unhash = tcp_bpf_unhash;
601 prot[TCP_BPF_BASE].close = tcp_bpf_close;
602 prot[TCP_BPF_BASE].recvmsg = tcp_bpf_recvmsg;
603 prot[TCP_BPF_BASE].stream_memory_read = tcp_bpf_stream_read;
605 prot[TCP_BPF_TX] = prot[TCP_BPF_BASE];
606 prot[TCP_BPF_TX].sendmsg = tcp_bpf_sendmsg;
607 prot[TCP_BPF_TX].sendpage = tcp_bpf_sendpage;
610 static void tcp_bpf_check_v6_needs_rebuild(struct sock *sk, struct proto *ops)
612 if (sk->sk_family == AF_INET6 &&
613 unlikely(ops != smp_load_acquire(&tcpv6_prot_saved))) {
614 spin_lock_bh(&tcpv6_prot_lock);
615 if (likely(ops != tcpv6_prot_saved)) {
616 tcp_bpf_rebuild_protos(tcp_bpf_prots[TCP_BPF_IPV6], ops);
617 smp_store_release(&tcpv6_prot_saved, ops);
619 spin_unlock_bh(&tcpv6_prot_lock);
623 static int __init tcp_bpf_v4_build_proto(void)
625 tcp_bpf_rebuild_protos(tcp_bpf_prots[TCP_BPF_IPV4], &tcp_prot);
628 core_initcall(tcp_bpf_v4_build_proto);
630 static void tcp_bpf_update_sk_prot(struct sock *sk, struct sk_psock *psock)
632 int family = sk->sk_family == AF_INET6 ? TCP_BPF_IPV6 : TCP_BPF_IPV4;
633 int config = psock->progs.msg_parser ? TCP_BPF_TX : TCP_BPF_BASE;
635 sk_psock_update_proto(sk, psock, &tcp_bpf_prots[family][config]);
638 static void tcp_bpf_reinit_sk_prot(struct sock *sk, struct sk_psock *psock)
640 int family = sk->sk_family == AF_INET6 ? TCP_BPF_IPV6 : TCP_BPF_IPV4;
641 int config = psock->progs.msg_parser ? TCP_BPF_TX : TCP_BPF_BASE;
643 /* Reinit occurs when program types change e.g. TCP_BPF_TX is removed
644 * or added requiring sk_prot hook updates. We keep original saved
645 * hooks in this case.
647 sk->sk_prot = &tcp_bpf_prots[family][config];
650 static int tcp_bpf_assert_proto_ops(struct proto *ops)
652 /* In order to avoid retpoline, we make assumptions when we call
653 * into ops if e.g. a psock is not present. Make sure they are
654 * indeed valid assumptions.
656 return ops->recvmsg == tcp_recvmsg &&
657 ops->sendmsg == tcp_sendmsg &&
658 ops->sendpage == tcp_sendpage ? 0 : -ENOTSUPP;
661 void tcp_bpf_reinit(struct sock *sk)
663 struct sk_psock *psock;
665 sock_owned_by_me(sk);
668 psock = sk_psock(sk);
669 tcp_bpf_reinit_sk_prot(sk, psock);
673 int tcp_bpf_init(struct sock *sk)
675 struct proto *ops = READ_ONCE(sk->sk_prot);
676 struct sk_psock *psock;
678 sock_owned_by_me(sk);
681 psock = sk_psock(sk);
682 if (unlikely(!psock || psock->sk_proto ||
683 tcp_bpf_assert_proto_ops(ops))) {
687 tcp_bpf_check_v6_needs_rebuild(sk, ops);
688 tcp_bpf_update_sk_prot(sk, psock);