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>
12 static bool sk_msg_try_coalesce_ok(struct sk_msg *msg, int elem_first_coalesce)
14 if (msg->sg.end > msg->sg.start &&
15 elem_first_coalesce < msg->sg.end)
18 if (msg->sg.end < msg->sg.start &&
19 (elem_first_coalesce > msg->sg.start ||
20 elem_first_coalesce < msg->sg.end))
26 int sk_msg_alloc(struct sock *sk, struct sk_msg *msg, int len,
27 int elem_first_coalesce)
29 struct page_frag *pfrag = sk_page_frag(sk);
34 struct scatterlist *sge;
38 if (!sk_page_frag_refill(sk, pfrag))
41 orig_offset = pfrag->offset;
42 use = min_t(int, len, pfrag->size - orig_offset);
43 if (!sk_wmem_schedule(sk, use))
47 sk_msg_iter_var_prev(i);
48 sge = &msg->sg.data[i];
50 if (sk_msg_try_coalesce_ok(msg, elem_first_coalesce) &&
51 sg_page(sge) == pfrag->page &&
52 sge->offset + sge->length == orig_offset) {
55 if (sk_msg_full(msg)) {
60 sge = &msg->sg.data[msg->sg.end];
62 sg_set_page(sge, pfrag->page, use, orig_offset);
63 get_page(pfrag->page);
64 sk_msg_iter_next(msg, end);
67 sk_mem_charge(sk, use);
75 EXPORT_SYMBOL_GPL(sk_msg_alloc);
77 int sk_msg_clone(struct sock *sk, struct sk_msg *dst, struct sk_msg *src,
80 int i = src->sg.start;
81 struct scatterlist *sge = sk_msg_elem(src, i);
82 struct scatterlist *sgd = NULL;
86 if (sge->length > off)
89 sk_msg_iter_var_next(i);
90 if (i == src->sg.end && off)
92 sge = sk_msg_elem(src, i);
96 sge_len = sge->length - off;
101 sgd = sk_msg_elem(dst, dst->sg.end - 1);
104 (sg_page(sge) == sg_page(sgd)) &&
105 (sg_virt(sge) + off == sg_virt(sgd) + sgd->length)) {
106 sgd->length += sge_len;
107 dst->sg.size += sge_len;
108 } else if (!sk_msg_full(dst)) {
109 sge_off = sge->offset + off;
110 sk_msg_page_add(dst, sg_page(sge), sge_len, sge_off);
117 sk_mem_charge(sk, sge_len);
118 sk_msg_iter_var_next(i);
119 if (i == src->sg.end && len)
121 sge = sk_msg_elem(src, i);
126 EXPORT_SYMBOL_GPL(sk_msg_clone);
128 void sk_msg_return_zero(struct sock *sk, struct sk_msg *msg, int bytes)
130 int i = msg->sg.start;
133 struct scatterlist *sge = sk_msg_elem(msg, i);
135 if (bytes < sge->length) {
136 sge->length -= bytes;
137 sge->offset += bytes;
138 sk_mem_uncharge(sk, bytes);
142 sk_mem_uncharge(sk, sge->length);
143 bytes -= sge->length;
146 sk_msg_iter_var_next(i);
147 } while (bytes && i != msg->sg.end);
150 EXPORT_SYMBOL_GPL(sk_msg_return_zero);
152 void sk_msg_return(struct sock *sk, struct sk_msg *msg, int bytes)
154 int i = msg->sg.start;
157 struct scatterlist *sge = &msg->sg.data[i];
158 int uncharge = (bytes < sge->length) ? bytes : sge->length;
160 sk_mem_uncharge(sk, uncharge);
162 sk_msg_iter_var_next(i);
163 } while (i != msg->sg.end);
165 EXPORT_SYMBOL_GPL(sk_msg_return);
167 static int sk_msg_free_elem(struct sock *sk, struct sk_msg *msg, u32 i,
170 struct scatterlist *sge = sk_msg_elem(msg, i);
171 u32 len = sge->length;
173 /* When the skb owns the memory we free it from consume_skb path. */
176 sk_mem_uncharge(sk, len);
177 put_page(sg_page(sge));
179 memset(sge, 0, sizeof(*sge));
183 static int __sk_msg_free(struct sock *sk, struct sk_msg *msg, u32 i,
186 struct scatterlist *sge = sk_msg_elem(msg, i);
189 while (msg->sg.size) {
190 msg->sg.size -= sge->length;
191 freed += sk_msg_free_elem(sk, msg, i, charge);
192 sk_msg_iter_var_next(i);
193 sk_msg_check_to_free(msg, i, msg->sg.size);
194 sge = sk_msg_elem(msg, i);
196 consume_skb(msg->skb);
201 int sk_msg_free_nocharge(struct sock *sk, struct sk_msg *msg)
203 return __sk_msg_free(sk, msg, msg->sg.start, false);
205 EXPORT_SYMBOL_GPL(sk_msg_free_nocharge);
207 int sk_msg_free(struct sock *sk, struct sk_msg *msg)
209 return __sk_msg_free(sk, msg, msg->sg.start, true);
211 EXPORT_SYMBOL_GPL(sk_msg_free);
213 static void __sk_msg_free_partial(struct sock *sk, struct sk_msg *msg,
214 u32 bytes, bool charge)
216 struct scatterlist *sge;
217 u32 i = msg->sg.start;
220 sge = sk_msg_elem(msg, i);
223 if (bytes < sge->length) {
225 sk_mem_uncharge(sk, bytes);
226 sge->length -= bytes;
227 sge->offset += bytes;
228 msg->sg.size -= bytes;
232 msg->sg.size -= sge->length;
233 bytes -= sge->length;
234 sk_msg_free_elem(sk, msg, i, charge);
235 sk_msg_iter_var_next(i);
236 sk_msg_check_to_free(msg, i, bytes);
241 void sk_msg_free_partial(struct sock *sk, struct sk_msg *msg, u32 bytes)
243 __sk_msg_free_partial(sk, msg, bytes, true);
245 EXPORT_SYMBOL_GPL(sk_msg_free_partial);
247 void sk_msg_free_partial_nocharge(struct sock *sk, struct sk_msg *msg,
250 __sk_msg_free_partial(sk, msg, bytes, false);
253 void sk_msg_trim(struct sock *sk, struct sk_msg *msg, int len)
255 int trim = msg->sg.size - len;
263 sk_msg_iter_var_prev(i);
265 while (msg->sg.data[i].length &&
266 trim >= msg->sg.data[i].length) {
267 trim -= msg->sg.data[i].length;
268 sk_msg_free_elem(sk, msg, i, true);
269 sk_msg_iter_var_prev(i);
274 msg->sg.data[i].length -= trim;
275 sk_mem_uncharge(sk, trim);
276 /* Adjust copybreak if it falls into the trimmed part of last buf */
277 if (msg->sg.curr == i && msg->sg.copybreak > msg->sg.data[i].length)
278 msg->sg.copybreak = msg->sg.data[i].length;
280 sk_msg_iter_var_next(i);
283 /* If we trim data a full sg elem before curr pointer update
284 * copybreak and current so that any future copy operations
285 * start at new copy location.
286 * However trimed data that has not yet been used in a copy op
287 * does not require an update.
290 msg->sg.curr = msg->sg.start;
291 msg->sg.copybreak = 0;
292 } else if (sk_msg_iter_dist(msg->sg.start, msg->sg.curr) >=
293 sk_msg_iter_dist(msg->sg.start, msg->sg.end)) {
294 sk_msg_iter_var_prev(i);
296 msg->sg.copybreak = msg->sg.data[i].length;
299 EXPORT_SYMBOL_GPL(sk_msg_trim);
301 int sk_msg_zerocopy_from_iter(struct sock *sk, struct iov_iter *from,
302 struct sk_msg *msg, u32 bytes)
304 int i, maxpages, ret = 0, num_elems = sk_msg_elem_used(msg);
305 const int to_max_pages = MAX_MSG_FRAGS;
306 struct page *pages[MAX_MSG_FRAGS];
307 ssize_t orig, copied, use, offset;
312 maxpages = to_max_pages - num_elems;
318 copied = iov_iter_get_pages(from, pages, bytes, maxpages,
325 iov_iter_advance(from, copied);
327 msg->sg.size += copied;
330 use = min_t(int, copied, PAGE_SIZE - offset);
331 sg_set_page(&msg->sg.data[msg->sg.end],
332 pages[i], use, offset);
333 sg_unmark_end(&msg->sg.data[msg->sg.end]);
334 sk_mem_charge(sk, use);
338 sk_msg_iter_next(msg, end);
342 /* When zerocopy is mixed with sk_msg_*copy* operations we
343 * may have a copybreak set in this case clear and prefer
344 * zerocopy remainder when possible.
346 msg->sg.copybreak = 0;
347 msg->sg.curr = msg->sg.end;
350 /* Revert iov_iter updates, msg will need to use 'trim' later if it
351 * also needs to be cleared.
354 iov_iter_revert(from, msg->sg.size - orig);
357 EXPORT_SYMBOL_GPL(sk_msg_zerocopy_from_iter);
359 int sk_msg_memcopy_from_iter(struct sock *sk, struct iov_iter *from,
360 struct sk_msg *msg, u32 bytes)
362 int ret = -ENOSPC, i = msg->sg.curr;
363 struct scatterlist *sge;
368 sge = sk_msg_elem(msg, i);
369 /* This is possible if a trim operation shrunk the buffer */
370 if (msg->sg.copybreak >= sge->length) {
371 msg->sg.copybreak = 0;
372 sk_msg_iter_var_next(i);
373 if (i == msg->sg.end)
375 sge = sk_msg_elem(msg, i);
378 buf_size = sge->length - msg->sg.copybreak;
379 copy = (buf_size > bytes) ? bytes : buf_size;
380 to = sg_virt(sge) + msg->sg.copybreak;
381 msg->sg.copybreak += copy;
382 if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY)
383 ret = copy_from_iter_nocache(to, copy, from);
385 ret = copy_from_iter(to, copy, from);
393 msg->sg.copybreak = 0;
394 sk_msg_iter_var_next(i);
395 } while (i != msg->sg.end);
400 EXPORT_SYMBOL_GPL(sk_msg_memcopy_from_iter);
402 /* Receive sk_msg from psock->ingress_msg to @msg. */
403 int sk_msg_recvmsg(struct sock *sk, struct sk_psock *psock, struct msghdr *msg,
406 struct iov_iter *iter = &msg->msg_iter;
407 int peek = flags & MSG_PEEK;
408 struct sk_msg *msg_rx;
411 msg_rx = sk_psock_peek_msg(psock);
412 while (copied != len) {
413 struct scatterlist *sge;
415 if (unlikely(!msg_rx))
418 i = msg_rx->sg.start;
423 sge = sk_msg_elem(msg_rx, i);
426 if (copied + copy > len)
428 copy = copy_page_to_iter(page, sge->offset, copy, iter);
430 return copied ? copied : -EFAULT;
437 sk_mem_uncharge(sk, copy);
438 msg_rx->sg.size -= copy;
441 sk_msg_iter_var_next(i);
446 /* Lets not optimize peek case if copy_page_to_iter
447 * didn't copy the entire length lets just break.
449 if (copy != sge->length)
451 sk_msg_iter_var_next(i);
456 } while (i != msg_rx->sg.end);
458 if (unlikely(peek)) {
459 msg_rx = sk_psock_next_msg(psock, msg_rx);
465 msg_rx->sg.start = i;
466 if (!sge->length && msg_rx->sg.start == msg_rx->sg.end) {
467 msg_rx = sk_psock_dequeue_msg(psock);
468 kfree_sk_msg(msg_rx);
470 msg_rx = sk_psock_peek_msg(psock);
475 EXPORT_SYMBOL_GPL(sk_msg_recvmsg);
477 static struct sk_msg *sk_psock_create_ingress_msg(struct sock *sk,
482 if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf)
485 if (!sk_rmem_schedule(sk, skb, skb->truesize))
488 msg = kzalloc(sizeof(*msg), __GFP_NOWARN | GFP_KERNEL);
496 static int sk_psock_skb_ingress_enqueue(struct sk_buff *skb,
497 struct sk_psock *psock,
503 /* skb linearize may fail with ENOMEM, but lets simply try again
504 * later if this happens. Under memory pressure we don't want to
505 * drop the skb. We need to linearize the skb so that the mapping
506 * in skb_to_sgvec can not error.
508 if (skb_linearize(skb))
510 num_sge = skb_to_sgvec(skb, msg->sg.data, 0, skb->len);
511 if (unlikely(num_sge < 0)) {
518 msg->sg.size = copied;
519 msg->sg.end = num_sge;
522 sk_psock_queue_msg(psock, msg);
523 sk_psock_data_ready(sk, psock);
527 static int sk_psock_skb_ingress_self(struct sk_psock *psock, struct sk_buff *skb);
529 static int sk_psock_skb_ingress(struct sk_psock *psock, struct sk_buff *skb)
531 struct sock *sk = psock->sk;
534 /* If we are receiving on the same sock skb->sk is already assigned,
535 * skip memory accounting and owner transition seeing it already set
538 if (unlikely(skb->sk == sk))
539 return sk_psock_skb_ingress_self(psock, skb);
540 msg = sk_psock_create_ingress_msg(sk, skb);
544 /* This will transition ownership of the data from the socket where
545 * the BPF program was run initiating the redirect to the socket
546 * we will eventually receive this data on. The data will be released
547 * from skb_consume found in __tcp_bpf_recvmsg() after its been copied
550 skb_set_owner_r(skb, sk);
551 return sk_psock_skb_ingress_enqueue(skb, psock, sk, msg);
554 /* Puts an skb on the ingress queue of the socket already assigned to the
555 * skb. In this case we do not need to check memory limits or skb_set_owner_r
556 * because the skb is already accounted for here.
558 static int sk_psock_skb_ingress_self(struct sk_psock *psock, struct sk_buff *skb)
560 struct sk_msg *msg = kzalloc(sizeof(*msg), __GFP_NOWARN | GFP_ATOMIC);
561 struct sock *sk = psock->sk;
566 skb_set_owner_r(skb, sk);
567 return sk_psock_skb_ingress_enqueue(skb, psock, sk, msg);
570 static int sk_psock_handle_skb(struct sk_psock *psock, struct sk_buff *skb,
571 u32 off, u32 len, bool ingress)
574 if (!sock_writeable(psock->sk))
576 return skb_send_sock(psock->sk, skb, off, len);
578 return sk_psock_skb_ingress(psock, skb);
581 static void sock_drop(struct sock *sk, struct sk_buff *skb)
583 sk_drops_add(sk, skb);
587 static void sk_psock_backlog(struct work_struct *work)
589 struct sk_psock *psock = container_of(work, struct sk_psock, work);
590 struct sk_psock_work_state *state = &psock->work_state;
596 mutex_lock(&psock->work_mutex);
605 while ((skb = skb_dequeue(&psock->ingress_skb))) {
609 ingress = skb_bpf_ingress(skb);
610 skb_bpf_redirect_clear(skb);
613 if (!sock_flag(psock->sk, SOCK_DEAD))
614 ret = sk_psock_handle_skb(psock, skb, off,
617 if (ret == -EAGAIN) {
623 /* Hard errors break pipe and stop xmit. */
624 sk_psock_report_error(psock, ret ? -ret : EPIPE);
625 sk_psock_clear_state(psock, SK_PSOCK_TX_ENABLED);
626 sock_drop(psock->sk, skb);
637 mutex_unlock(&psock->work_mutex);
640 struct sk_psock *sk_psock_init(struct sock *sk, int node)
642 struct sk_psock *psock;
645 write_lock_bh(&sk->sk_callback_lock);
647 if (sk->sk_user_data) {
648 psock = ERR_PTR(-EBUSY);
652 psock = kzalloc_node(sizeof(*psock), GFP_ATOMIC | __GFP_NOWARN, node);
654 psock = ERR_PTR(-ENOMEM);
658 prot = READ_ONCE(sk->sk_prot);
660 psock->eval = __SK_NONE;
661 psock->sk_proto = prot;
662 psock->saved_unhash = prot->unhash;
663 psock->saved_close = prot->close;
664 psock->saved_write_space = sk->sk_write_space;
666 INIT_LIST_HEAD(&psock->link);
667 spin_lock_init(&psock->link_lock);
669 INIT_WORK(&psock->work, sk_psock_backlog);
670 mutex_init(&psock->work_mutex);
671 INIT_LIST_HEAD(&psock->ingress_msg);
672 spin_lock_init(&psock->ingress_lock);
673 skb_queue_head_init(&psock->ingress_skb);
675 sk_psock_set_state(psock, SK_PSOCK_TX_ENABLED);
676 refcount_set(&psock->refcnt, 1);
678 rcu_assign_sk_user_data_nocopy(sk, psock);
682 write_unlock_bh(&sk->sk_callback_lock);
685 EXPORT_SYMBOL_GPL(sk_psock_init);
687 struct sk_psock_link *sk_psock_link_pop(struct sk_psock *psock)
689 struct sk_psock_link *link;
691 spin_lock_bh(&psock->link_lock);
692 link = list_first_entry_or_null(&psock->link, struct sk_psock_link,
695 list_del(&link->list);
696 spin_unlock_bh(&psock->link_lock);
700 static void __sk_psock_purge_ingress_msg(struct sk_psock *psock)
702 struct sk_msg *msg, *tmp;
704 list_for_each_entry_safe(msg, tmp, &psock->ingress_msg, list) {
705 list_del(&msg->list);
706 sk_msg_free(psock->sk, msg);
711 static void __sk_psock_zap_ingress(struct sk_psock *psock)
715 while ((skb = skb_dequeue(&psock->ingress_skb)) != NULL) {
716 skb_bpf_redirect_clear(skb);
717 sock_drop(psock->sk, skb);
719 __sk_psock_purge_ingress_msg(psock);
722 static void sk_psock_link_destroy(struct sk_psock *psock)
724 struct sk_psock_link *link, *tmp;
726 list_for_each_entry_safe(link, tmp, &psock->link, list) {
727 list_del(&link->list);
728 sk_psock_free_link(link);
732 void sk_psock_stop(struct sk_psock *psock, bool wait)
734 spin_lock_bh(&psock->ingress_lock);
735 sk_psock_clear_state(psock, SK_PSOCK_TX_ENABLED);
736 sk_psock_cork_free(psock);
737 __sk_psock_zap_ingress(psock);
738 spin_unlock_bh(&psock->ingress_lock);
741 cancel_work_sync(&psock->work);
744 static void sk_psock_done_strp(struct sk_psock *psock);
746 static void sk_psock_destroy(struct work_struct *work)
748 struct sk_psock *psock = container_of(to_rcu_work(work),
749 struct sk_psock, rwork);
750 /* No sk_callback_lock since already detached. */
752 sk_psock_done_strp(psock);
754 cancel_work_sync(&psock->work);
755 mutex_destroy(&psock->work_mutex);
757 psock_progs_drop(&psock->progs);
759 sk_psock_link_destroy(psock);
760 sk_psock_cork_free(psock);
763 sock_put(psock->sk_redir);
768 void sk_psock_drop(struct sock *sk, struct sk_psock *psock)
770 sk_psock_stop(psock, false);
772 write_lock_bh(&sk->sk_callback_lock);
773 sk_psock_restore_proto(sk, psock);
774 rcu_assign_sk_user_data(sk, NULL);
775 if (psock->progs.stream_parser)
776 sk_psock_stop_strp(sk, psock);
777 else if (psock->progs.stream_verdict || psock->progs.skb_verdict)
778 sk_psock_stop_verdict(sk, psock);
779 write_unlock_bh(&sk->sk_callback_lock);
781 INIT_RCU_WORK(&psock->rwork, sk_psock_destroy);
782 queue_rcu_work(system_wq, &psock->rwork);
784 EXPORT_SYMBOL_GPL(sk_psock_drop);
786 static int sk_psock_map_verd(int verdict, bool redir)
790 return redir ? __SK_REDIRECT : __SK_PASS;
799 int sk_psock_msg_verdict(struct sock *sk, struct sk_psock *psock,
802 struct bpf_prog *prog;
806 prog = READ_ONCE(psock->progs.msg_parser);
807 if (unlikely(!prog)) {
812 sk_msg_compute_data_pointers(msg);
814 ret = bpf_prog_run_pin_on_cpu(prog, msg);
815 ret = sk_psock_map_verd(ret, msg->sk_redir);
816 psock->apply_bytes = msg->apply_bytes;
817 if (ret == __SK_REDIRECT) {
819 sock_put(psock->sk_redir);
820 psock->sk_redir = msg->sk_redir;
821 if (!psock->sk_redir) {
825 sock_hold(psock->sk_redir);
831 EXPORT_SYMBOL_GPL(sk_psock_msg_verdict);
833 static int sk_psock_skb_redirect(struct sk_psock *from, struct sk_buff *skb)
835 struct sk_psock *psock_other;
836 struct sock *sk_other;
838 sk_other = skb_bpf_redirect_fetch(skb);
839 /* This error is a buggy BPF program, it returned a redirect
840 * return code, but then didn't set a redirect interface.
842 if (unlikely(!sk_other)) {
843 sock_drop(from->sk, skb);
846 psock_other = sk_psock(sk_other);
847 /* This error indicates the socket is being torn down or had another
848 * error that caused the pipe to break. We can't send a packet on
849 * a socket that is in this state so we drop the skb.
851 if (!psock_other || sock_flag(sk_other, SOCK_DEAD)) {
852 skb_bpf_redirect_clear(skb);
853 sock_drop(from->sk, skb);
856 spin_lock_bh(&psock_other->ingress_lock);
857 if (!sk_psock_test_state(psock_other, SK_PSOCK_TX_ENABLED)) {
858 spin_unlock_bh(&psock_other->ingress_lock);
859 skb_bpf_redirect_clear(skb);
860 sock_drop(from->sk, skb);
864 skb_queue_tail(&psock_other->ingress_skb, skb);
865 schedule_work(&psock_other->work);
866 spin_unlock_bh(&psock_other->ingress_lock);
870 static void sk_psock_tls_verdict_apply(struct sk_buff *skb,
871 struct sk_psock *from, int verdict)
875 sk_psock_skb_redirect(from, skb);
884 int sk_psock_tls_strp_read(struct sk_psock *psock, struct sk_buff *skb)
886 struct bpf_prog *prog;
890 prog = READ_ONCE(psock->progs.stream_verdict);
894 skb_bpf_redirect_clear(skb);
895 ret = bpf_prog_run_pin_on_cpu(prog, skb);
896 ret = sk_psock_map_verd(ret, skb_bpf_redirect_fetch(skb));
899 sk_psock_tls_verdict_apply(skb, psock, ret);
903 EXPORT_SYMBOL_GPL(sk_psock_tls_strp_read);
905 static int sk_psock_verdict_apply(struct sk_psock *psock, struct sk_buff *skb,
908 struct sock *sk_other;
914 sk_other = psock->sk;
915 if (sock_flag(sk_other, SOCK_DEAD) ||
916 !sk_psock_test_state(psock, SK_PSOCK_TX_ENABLED)) {
920 skb_bpf_set_ingress(skb);
922 /* If the queue is empty then we can submit directly
923 * into the msg queue. If its not empty we have to
924 * queue work otherwise we may get OOO data. Otherwise,
925 * if sk_psock_skb_ingress errors will be handled by
926 * retrying later from workqueue.
928 if (skb_queue_empty(&psock->ingress_skb)) {
929 err = sk_psock_skb_ingress_self(psock, skb);
932 spin_lock_bh(&psock->ingress_lock);
933 if (sk_psock_test_state(psock, SK_PSOCK_TX_ENABLED)) {
934 skb_queue_tail(&psock->ingress_skb, skb);
935 schedule_work(&psock->work);
938 spin_unlock_bh(&psock->ingress_lock);
940 skb_bpf_redirect_clear(skb);
946 err = sk_psock_skb_redirect(psock, skb);
951 sock_drop(psock->sk, skb);
957 static void sk_psock_write_space(struct sock *sk)
959 struct sk_psock *psock;
960 void (*write_space)(struct sock *sk) = NULL;
963 psock = sk_psock(sk);
965 if (sk_psock_test_state(psock, SK_PSOCK_TX_ENABLED))
966 schedule_work(&psock->work);
967 write_space = psock->saved_write_space;
974 #if IS_ENABLED(CONFIG_BPF_STREAM_PARSER)
975 static void sk_psock_strp_read(struct strparser *strp, struct sk_buff *skb)
977 struct sk_psock *psock;
978 struct bpf_prog *prog;
984 psock = sk_psock(sk);
985 if (unlikely(!psock)) {
989 prog = READ_ONCE(psock->progs.stream_verdict);
993 skb_bpf_redirect_clear(skb);
994 ret = bpf_prog_run_pin_on_cpu(prog, skb);
995 ret = sk_psock_map_verd(ret, skb_bpf_redirect_fetch(skb));
998 sk_psock_verdict_apply(psock, skb, ret);
1003 static int sk_psock_strp_read_done(struct strparser *strp, int err)
1008 static int sk_psock_strp_parse(struct strparser *strp, struct sk_buff *skb)
1010 struct sk_psock *psock = container_of(strp, struct sk_psock, strp);
1011 struct bpf_prog *prog;
1015 prog = READ_ONCE(psock->progs.stream_parser);
1017 skb->sk = psock->sk;
1018 ret = bpf_prog_run_pin_on_cpu(prog, skb);
1025 /* Called with socket lock held. */
1026 static void sk_psock_strp_data_ready(struct sock *sk)
1028 struct sk_psock *psock;
1031 psock = sk_psock(sk);
1032 if (likely(psock)) {
1033 if (tls_sw_has_ctx_rx(sk)) {
1034 psock->saved_data_ready(sk);
1036 write_lock_bh(&sk->sk_callback_lock);
1037 strp_data_ready(&psock->strp);
1038 write_unlock_bh(&sk->sk_callback_lock);
1044 int sk_psock_init_strp(struct sock *sk, struct sk_psock *psock)
1046 static const struct strp_callbacks cb = {
1047 .rcv_msg = sk_psock_strp_read,
1048 .read_sock_done = sk_psock_strp_read_done,
1049 .parse_msg = sk_psock_strp_parse,
1052 return strp_init(&psock->strp, sk, &cb);
1055 void sk_psock_start_strp(struct sock *sk, struct sk_psock *psock)
1057 if (psock->saved_data_ready)
1060 psock->saved_data_ready = sk->sk_data_ready;
1061 sk->sk_data_ready = sk_psock_strp_data_ready;
1062 sk->sk_write_space = sk_psock_write_space;
1065 void sk_psock_stop_strp(struct sock *sk, struct sk_psock *psock)
1067 if (!psock->saved_data_ready)
1070 sk->sk_data_ready = psock->saved_data_ready;
1071 psock->saved_data_ready = NULL;
1072 strp_stop(&psock->strp);
1075 static void sk_psock_done_strp(struct sk_psock *psock)
1077 /* Parser has been stopped */
1078 if (psock->progs.stream_parser)
1079 strp_done(&psock->strp);
1082 static void sk_psock_done_strp(struct sk_psock *psock)
1085 #endif /* CONFIG_BPF_STREAM_PARSER */
1087 static int sk_psock_verdict_recv(read_descriptor_t *desc, struct sk_buff *skb,
1088 unsigned int offset, size_t orig_len)
1090 struct sock *sk = (struct sock *)desc->arg.data;
1091 struct sk_psock *psock;
1092 struct bpf_prog *prog;
1093 int ret = __SK_DROP;
1096 /* clone here so sk_eat_skb() in tcp_read_sock does not drop our data */
1097 skb = skb_clone(skb, GFP_ATOMIC);
1099 desc->error = -ENOMEM;
1104 psock = sk_psock(sk);
1105 if (unlikely(!psock)) {
1110 prog = READ_ONCE(psock->progs.stream_verdict);
1112 prog = READ_ONCE(psock->progs.skb_verdict);
1116 skb_bpf_redirect_clear(skb);
1117 ret = bpf_prog_run_pin_on_cpu(prog, skb);
1118 ret = sk_psock_map_verd(ret, skb_bpf_redirect_fetch(skb));
1121 if (sk_psock_verdict_apply(psock, skb, ret) < 0)
1128 static void sk_psock_verdict_data_ready(struct sock *sk)
1130 struct socket *sock = sk->sk_socket;
1131 read_descriptor_t desc;
1133 if (unlikely(!sock || !sock->ops || !sock->ops->read_sock))
1140 sock->ops->read_sock(sk, &desc, sk_psock_verdict_recv);
1143 void sk_psock_start_verdict(struct sock *sk, struct sk_psock *psock)
1145 if (psock->saved_data_ready)
1148 psock->saved_data_ready = sk->sk_data_ready;
1149 sk->sk_data_ready = sk_psock_verdict_data_ready;
1150 sk->sk_write_space = sk_psock_write_space;
1153 void sk_psock_stop_verdict(struct sock *sk, struct sk_psock *psock)
1155 if (!psock->saved_data_ready)
1158 sk->sk_data_ready = psock->saved_data_ready;
1159 psock->saved_data_ready = NULL;