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/skbuff.h>
6 #include <linux/scatterlist.h>
11 static bool sk_msg_try_coalesce_ok(struct sk_msg *msg, int elem_first_coalesce)
13 if (msg->sg.end > msg->sg.start &&
14 elem_first_coalesce < msg->sg.end)
17 if (msg->sg.end < msg->sg.start &&
18 (elem_first_coalesce > msg->sg.start ||
19 elem_first_coalesce < msg->sg.end))
25 int sk_msg_alloc(struct sock *sk, struct sk_msg *msg, int len,
26 int elem_first_coalesce)
28 struct page_frag *pfrag = sk_page_frag(sk);
33 struct scatterlist *sge;
37 if (!sk_page_frag_refill(sk, pfrag))
40 orig_offset = pfrag->offset;
41 use = min_t(int, len, pfrag->size - orig_offset);
42 if (!sk_wmem_schedule(sk, use))
46 sk_msg_iter_var_prev(i);
47 sge = &msg->sg.data[i];
49 if (sk_msg_try_coalesce_ok(msg, elem_first_coalesce) &&
50 sg_page(sge) == pfrag->page &&
51 sge->offset + sge->length == orig_offset) {
54 if (sk_msg_full(msg)) {
59 sge = &msg->sg.data[msg->sg.end];
61 sg_set_page(sge, pfrag->page, use, orig_offset);
62 get_page(pfrag->page);
63 sk_msg_iter_next(msg, end);
66 sk_mem_charge(sk, use);
74 EXPORT_SYMBOL_GPL(sk_msg_alloc);
76 int sk_msg_clone(struct sock *sk, struct sk_msg *dst, struct sk_msg *src,
79 int i = src->sg.start;
80 struct scatterlist *sge = sk_msg_elem(src, i);
87 if (sge->length > off)
90 sk_msg_iter_var_next(i);
91 if (i == src->sg.end && off)
93 sge = sk_msg_elem(src, i);
97 sge_len = sge->length - off;
98 sge_off = sge->offset + off;
103 sk_msg_page_add(dst, sg_page(sge), sge_len, sge_off);
104 sk_mem_charge(sk, sge_len);
105 sk_msg_iter_var_next(i);
106 if (i == src->sg.end && len)
108 sge = sk_msg_elem(src, i);
113 EXPORT_SYMBOL_GPL(sk_msg_clone);
115 void sk_msg_return_zero(struct sock *sk, struct sk_msg *msg, int bytes)
117 int i = msg->sg.start;
120 struct scatterlist *sge = sk_msg_elem(msg, i);
122 if (bytes < sge->length) {
123 sge->length -= bytes;
124 sge->offset += bytes;
125 sk_mem_uncharge(sk, bytes);
129 sk_mem_uncharge(sk, sge->length);
130 bytes -= sge->length;
133 sk_msg_iter_var_next(i);
134 } while (bytes && i != msg->sg.end);
137 EXPORT_SYMBOL_GPL(sk_msg_return_zero);
139 void sk_msg_return(struct sock *sk, struct sk_msg *msg, int bytes)
141 int i = msg->sg.start;
144 struct scatterlist *sge = &msg->sg.data[i];
145 int uncharge = (bytes < sge->length) ? bytes : sge->length;
147 sk_mem_uncharge(sk, uncharge);
149 sk_msg_iter_var_next(i);
150 } while (i != msg->sg.end);
152 EXPORT_SYMBOL_GPL(sk_msg_return);
154 static int sk_msg_free_elem(struct sock *sk, struct sk_msg *msg, u32 i,
157 struct scatterlist *sge = sk_msg_elem(msg, i);
158 u32 len = sge->length;
161 sk_mem_uncharge(sk, len);
163 put_page(sg_page(sge));
164 memset(sge, 0, sizeof(*sge));
168 static int __sk_msg_free(struct sock *sk, struct sk_msg *msg, u32 i,
171 struct scatterlist *sge = sk_msg_elem(msg, i);
174 while (msg->sg.size) {
175 msg->sg.size -= sge->length;
176 freed += sk_msg_free_elem(sk, msg, i, charge);
177 sk_msg_iter_var_next(i);
178 sk_msg_check_to_free(msg, i, msg->sg.size);
179 sge = sk_msg_elem(msg, i);
182 consume_skb(msg->skb);
187 int sk_msg_free_nocharge(struct sock *sk, struct sk_msg *msg)
189 return __sk_msg_free(sk, msg, msg->sg.start, false);
191 EXPORT_SYMBOL_GPL(sk_msg_free_nocharge);
193 int sk_msg_free(struct sock *sk, struct sk_msg *msg)
195 return __sk_msg_free(sk, msg, msg->sg.start, true);
197 EXPORT_SYMBOL_GPL(sk_msg_free);
199 static void __sk_msg_free_partial(struct sock *sk, struct sk_msg *msg,
200 u32 bytes, bool charge)
202 struct scatterlist *sge;
203 u32 i = msg->sg.start;
206 sge = sk_msg_elem(msg, i);
209 if (bytes < sge->length) {
211 sk_mem_uncharge(sk, bytes);
212 sge->length -= bytes;
213 sge->offset += bytes;
214 msg->sg.size -= bytes;
218 msg->sg.size -= sge->length;
219 bytes -= sge->length;
220 sk_msg_free_elem(sk, msg, i, charge);
221 sk_msg_iter_var_next(i);
222 sk_msg_check_to_free(msg, i, bytes);
227 void sk_msg_free_partial(struct sock *sk, struct sk_msg *msg, u32 bytes)
229 __sk_msg_free_partial(sk, msg, bytes, true);
231 EXPORT_SYMBOL_GPL(sk_msg_free_partial);
233 void sk_msg_free_partial_nocharge(struct sock *sk, struct sk_msg *msg,
236 __sk_msg_free_partial(sk, msg, bytes, false);
239 void sk_msg_trim(struct sock *sk, struct sk_msg *msg, int len)
241 int trim = msg->sg.size - len;
249 sk_msg_iter_var_prev(i);
251 while (msg->sg.data[i].length &&
252 trim >= msg->sg.data[i].length) {
253 trim -= msg->sg.data[i].length;
254 sk_msg_free_elem(sk, msg, i, true);
255 sk_msg_iter_var_prev(i);
260 msg->sg.data[i].length -= trim;
261 sk_mem_uncharge(sk, trim);
263 /* If we trim data before curr pointer update copybreak and current
264 * so that any future copy operations start at new copy location.
265 * However trimed data that has not yet been used in a copy op
266 * does not require an update.
268 if (msg->sg.curr >= i) {
270 msg->sg.copybreak = msg->sg.data[i].length;
272 sk_msg_iter_var_next(i);
275 EXPORT_SYMBOL_GPL(sk_msg_trim);
277 int sk_msg_zerocopy_from_iter(struct sock *sk, struct iov_iter *from,
278 struct sk_msg *msg, u32 bytes)
280 int i, maxpages, ret = 0, num_elems = sk_msg_elem_used(msg);
281 const int to_max_pages = MAX_MSG_FRAGS;
282 struct page *pages[MAX_MSG_FRAGS];
283 ssize_t orig, copied, use, offset;
288 maxpages = to_max_pages - num_elems;
294 copied = iov_iter_get_pages(from, pages, bytes, maxpages,
301 iov_iter_advance(from, copied);
303 msg->sg.size += copied;
306 use = min_t(int, copied, PAGE_SIZE - offset);
307 sg_set_page(&msg->sg.data[msg->sg.end],
308 pages[i], use, offset);
309 sg_unmark_end(&msg->sg.data[msg->sg.end]);
310 sk_mem_charge(sk, use);
314 sk_msg_iter_next(msg, end);
318 /* When zerocopy is mixed with sk_msg_*copy* operations we
319 * may have a copybreak set in this case clear and prefer
320 * zerocopy remainder when possible.
322 msg->sg.copybreak = 0;
323 msg->sg.curr = msg->sg.end;
326 /* Revert iov_iter updates, msg will need to use 'trim' later if it
327 * also needs to be cleared.
330 iov_iter_revert(from, msg->sg.size - orig);
333 EXPORT_SYMBOL_GPL(sk_msg_zerocopy_from_iter);
335 int sk_msg_memcopy_from_iter(struct sock *sk, struct iov_iter *from,
336 struct sk_msg *msg, u32 bytes)
338 int ret = -ENOSPC, i = msg->sg.curr;
339 struct scatterlist *sge;
344 sge = sk_msg_elem(msg, i);
345 /* This is possible if a trim operation shrunk the buffer */
346 if (msg->sg.copybreak >= sge->length) {
347 msg->sg.copybreak = 0;
348 sk_msg_iter_var_next(i);
349 if (i == msg->sg.end)
351 sge = sk_msg_elem(msg, i);
354 buf_size = sge->length - msg->sg.copybreak;
355 copy = (buf_size > bytes) ? bytes : buf_size;
356 to = sg_virt(sge) + msg->sg.copybreak;
357 msg->sg.copybreak += copy;
358 if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY)
359 ret = copy_from_iter_nocache(to, copy, from);
361 ret = copy_from_iter(to, copy, from);
369 msg->sg.copybreak = 0;
370 sk_msg_iter_var_next(i);
371 } while (i != msg->sg.end);
376 EXPORT_SYMBOL_GPL(sk_msg_memcopy_from_iter);
378 static int sk_psock_skb_ingress(struct sk_psock *psock, struct sk_buff *skb)
380 struct sock *sk = psock->sk;
381 int copied = 0, num_sge;
384 msg = kzalloc(sizeof(*msg), __GFP_NOWARN | GFP_ATOMIC);
387 if (!sk_rmem_schedule(sk, skb, skb->len)) {
393 num_sge = skb_to_sgvec(skb, msg->sg.data, 0, skb->len);
394 if (unlikely(num_sge < 0)) {
399 sk_mem_charge(sk, skb->len);
402 msg->sg.end = num_sge == MAX_MSG_FRAGS ? 0 : num_sge;
405 sk_psock_queue_msg(psock, msg);
406 sk->sk_data_ready(sk);
410 static int sk_psock_handle_skb(struct sk_psock *psock, struct sk_buff *skb,
411 u32 off, u32 len, bool ingress)
414 return sk_psock_skb_ingress(psock, skb);
416 return skb_send_sock_locked(psock->sk, skb, off, len);
419 static void sk_psock_backlog(struct work_struct *work)
421 struct sk_psock *psock = container_of(work, struct sk_psock, work);
422 struct sk_psock_work_state *state = &psock->work_state;
428 /* Lock sock to avoid losing sk_socket during loop. */
429 lock_sock(psock->sk);
438 while ((skb = skb_dequeue(&psock->ingress_skb))) {
442 ingress = tcp_skb_bpf_ingress(skb);
445 if (likely(psock->sk->sk_socket))
446 ret = sk_psock_handle_skb(psock, skb, off,
449 if (ret == -EAGAIN) {
455 /* Hard errors break pipe and stop xmit. */
456 sk_psock_report_error(psock, ret ? -ret : EPIPE);
457 sk_psock_clear_state(psock, SK_PSOCK_TX_ENABLED);
469 release_sock(psock->sk);
472 struct sk_psock *sk_psock_init(struct sock *sk, int node)
474 struct sk_psock *psock = kzalloc_node(sizeof(*psock),
475 GFP_ATOMIC | __GFP_NOWARN,
481 psock->eval = __SK_NONE;
483 INIT_LIST_HEAD(&psock->link);
484 spin_lock_init(&psock->link_lock);
486 INIT_WORK(&psock->work, sk_psock_backlog);
487 INIT_LIST_HEAD(&psock->ingress_msg);
488 skb_queue_head_init(&psock->ingress_skb);
490 sk_psock_set_state(psock, SK_PSOCK_TX_ENABLED);
491 refcount_set(&psock->refcnt, 1);
493 rcu_assign_sk_user_data(sk, psock);
498 EXPORT_SYMBOL_GPL(sk_psock_init);
500 struct sk_psock_link *sk_psock_link_pop(struct sk_psock *psock)
502 struct sk_psock_link *link;
504 spin_lock_bh(&psock->link_lock);
505 link = list_first_entry_or_null(&psock->link, struct sk_psock_link,
508 list_del(&link->list);
509 spin_unlock_bh(&psock->link_lock);
513 void __sk_psock_purge_ingress_msg(struct sk_psock *psock)
515 struct sk_msg *msg, *tmp;
517 list_for_each_entry_safe(msg, tmp, &psock->ingress_msg, list) {
518 list_del(&msg->list);
519 sk_msg_free(psock->sk, msg);
524 static void sk_psock_zap_ingress(struct sk_psock *psock)
526 __skb_queue_purge(&psock->ingress_skb);
527 __sk_psock_purge_ingress_msg(psock);
530 static void sk_psock_link_destroy(struct sk_psock *psock)
532 struct sk_psock_link *link, *tmp;
534 list_for_each_entry_safe(link, tmp, &psock->link, list) {
535 list_del(&link->list);
536 sk_psock_free_link(link);
540 static void sk_psock_destroy_deferred(struct work_struct *gc)
542 struct sk_psock *psock = container_of(gc, struct sk_psock, gc);
544 /* No sk_callback_lock since already detached. */
545 if (psock->parser.enabled)
546 strp_done(&psock->parser.strp);
548 cancel_work_sync(&psock->work);
550 psock_progs_drop(&psock->progs);
552 sk_psock_link_destroy(psock);
553 sk_psock_cork_free(psock);
554 sk_psock_zap_ingress(psock);
557 sock_put(psock->sk_redir);
562 void sk_psock_destroy(struct rcu_head *rcu)
564 struct sk_psock *psock = container_of(rcu, struct sk_psock, rcu);
566 INIT_WORK(&psock->gc, sk_psock_destroy_deferred);
567 schedule_work(&psock->gc);
569 EXPORT_SYMBOL_GPL(sk_psock_destroy);
571 void sk_psock_drop(struct sock *sk, struct sk_psock *psock)
573 rcu_assign_sk_user_data(sk, NULL);
574 sk_psock_cork_free(psock);
575 sk_psock_restore_proto(sk, psock);
577 write_lock_bh(&sk->sk_callback_lock);
578 if (psock->progs.skb_parser)
579 sk_psock_stop_strp(sk, psock);
580 write_unlock_bh(&sk->sk_callback_lock);
581 sk_psock_clear_state(psock, SK_PSOCK_TX_ENABLED);
583 call_rcu_sched(&psock->rcu, sk_psock_destroy);
585 EXPORT_SYMBOL_GPL(sk_psock_drop);
587 static int sk_psock_map_verd(int verdict, bool redir)
591 return redir ? __SK_REDIRECT : __SK_PASS;
600 int sk_psock_msg_verdict(struct sock *sk, struct sk_psock *psock,
603 struct bpf_prog *prog;
608 prog = READ_ONCE(psock->progs.msg_parser);
609 if (unlikely(!prog)) {
614 sk_msg_compute_data_pointers(msg);
616 ret = BPF_PROG_RUN(prog, msg);
617 ret = sk_psock_map_verd(ret, msg->sk_redir);
618 psock->apply_bytes = msg->apply_bytes;
619 if (ret == __SK_REDIRECT) {
621 sock_put(psock->sk_redir);
622 psock->sk_redir = msg->sk_redir;
623 if (!psock->sk_redir) {
627 sock_hold(psock->sk_redir);
634 EXPORT_SYMBOL_GPL(sk_psock_msg_verdict);
636 static int sk_psock_bpf_run(struct sk_psock *psock, struct bpf_prog *prog,
642 bpf_compute_data_end_sk_skb(skb);
644 ret = BPF_PROG_RUN(prog, skb);
646 /* strparser clones the skb before handing it to a upper layer,
647 * meaning skb_orphan has been called. We NULL sk on the way out
648 * to ensure we don't trigger a BUG_ON() in skb/sk operations
649 * later and because we are not charging the memory of this skb
656 static struct sk_psock *sk_psock_from_strp(struct strparser *strp)
658 struct sk_psock_parser *parser;
660 parser = container_of(strp, struct sk_psock_parser, strp);
661 return container_of(parser, struct sk_psock, parser);
664 static void sk_psock_verdict_apply(struct sk_psock *psock,
665 struct sk_buff *skb, int verdict)
667 struct sk_psock *psock_other;
668 struct sock *sk_other;
673 sk_other = tcp_skb_bpf_redirect_fetch(skb);
674 if (unlikely(!sk_other))
676 psock_other = sk_psock(sk_other);
677 if (!psock_other || sock_flag(sk_other, SOCK_DEAD) ||
678 !sk_psock_test_state(psock_other, SK_PSOCK_TX_ENABLED))
680 ingress = tcp_skb_bpf_ingress(skb);
681 if ((!ingress && sock_writeable(sk_other)) ||
683 atomic_read(&sk_other->sk_rmem_alloc) <=
684 sk_other->sk_rcvbuf)) {
686 skb_set_owner_w(skb, sk_other);
687 skb_queue_tail(&psock_other->ingress_skb, skb);
688 schedule_work(&psock_other->work);
700 static void sk_psock_strp_read(struct strparser *strp, struct sk_buff *skb)
702 struct sk_psock *psock = sk_psock_from_strp(strp);
703 struct bpf_prog *prog;
707 prog = READ_ONCE(psock->progs.skb_verdict);
710 tcp_skb_bpf_redirect_clear(skb);
711 ret = sk_psock_bpf_run(psock, prog, skb);
712 ret = sk_psock_map_verd(ret, tcp_skb_bpf_redirect_fetch(skb));
715 sk_psock_verdict_apply(psock, skb, ret);
718 static int sk_psock_strp_read_done(struct strparser *strp, int err)
723 static int sk_psock_strp_parse(struct strparser *strp, struct sk_buff *skb)
725 struct sk_psock *psock = sk_psock_from_strp(strp);
726 struct bpf_prog *prog;
730 prog = READ_ONCE(psock->progs.skb_parser);
732 ret = sk_psock_bpf_run(psock, prog, skb);
737 /* Called with socket lock held. */
738 static void sk_psock_data_ready(struct sock *sk)
740 struct sk_psock *psock;
743 psock = sk_psock(sk);
745 write_lock_bh(&sk->sk_callback_lock);
746 strp_data_ready(&psock->parser.strp);
747 write_unlock_bh(&sk->sk_callback_lock);
752 static void sk_psock_write_space(struct sock *sk)
754 struct sk_psock *psock;
755 void (*write_space)(struct sock *sk);
758 psock = sk_psock(sk);
759 if (likely(psock && sk_psock_test_state(psock, SK_PSOCK_TX_ENABLED)))
760 schedule_work(&psock->work);
761 write_space = psock->saved_write_space;
766 int sk_psock_init_strp(struct sock *sk, struct sk_psock *psock)
768 static const struct strp_callbacks cb = {
769 .rcv_msg = sk_psock_strp_read,
770 .read_sock_done = sk_psock_strp_read_done,
771 .parse_msg = sk_psock_strp_parse,
774 psock->parser.enabled = false;
775 return strp_init(&psock->parser.strp, sk, &cb);
778 void sk_psock_start_strp(struct sock *sk, struct sk_psock *psock)
780 struct sk_psock_parser *parser = &psock->parser;
785 parser->saved_data_ready = sk->sk_data_ready;
786 sk->sk_data_ready = sk_psock_data_ready;
787 sk->sk_write_space = sk_psock_write_space;
788 parser->enabled = true;
791 void sk_psock_stop_strp(struct sock *sk, struct sk_psock *psock)
793 struct sk_psock_parser *parser = &psock->parser;
795 if (!parser->enabled)
798 sk->sk_data_ready = parser->saved_data_ready;
799 parser->saved_data_ready = NULL;
800 strp_stop(&parser->strp);
801 parser->enabled = false;