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);
30 u32 osize = msg->sg.size;
35 struct scatterlist *sge;
39 if (!sk_page_frag_refill(sk, pfrag)) {
44 orig_offset = pfrag->offset;
45 use = min_t(int, len, pfrag->size - orig_offset);
46 if (!sk_wmem_schedule(sk, use)) {
52 sk_msg_iter_var_prev(i);
53 sge = &msg->sg.data[i];
55 if (sk_msg_try_coalesce_ok(msg, elem_first_coalesce) &&
56 sg_page(sge) == pfrag->page &&
57 sge->offset + sge->length == orig_offset) {
60 if (sk_msg_full(msg)) {
65 sge = &msg->sg.data[msg->sg.end];
67 sg_set_page(sge, pfrag->page, use, orig_offset);
68 get_page(pfrag->page);
69 sk_msg_iter_next(msg, end);
72 sk_mem_charge(sk, use);
81 sk_msg_trim(sk, msg, osize);
84 EXPORT_SYMBOL_GPL(sk_msg_alloc);
86 int sk_msg_clone(struct sock *sk, struct sk_msg *dst, struct sk_msg *src,
89 int i = src->sg.start;
90 struct scatterlist *sge = sk_msg_elem(src, i);
91 struct scatterlist *sgd = NULL;
95 if (sge->length > off)
98 sk_msg_iter_var_next(i);
99 if (i == src->sg.end && off)
101 sge = sk_msg_elem(src, i);
105 sge_len = sge->length - off;
110 sgd = sk_msg_elem(dst, dst->sg.end - 1);
113 (sg_page(sge) == sg_page(sgd)) &&
114 (sg_virt(sge) + off == sg_virt(sgd) + sgd->length)) {
115 sgd->length += sge_len;
116 dst->sg.size += sge_len;
117 } else if (!sk_msg_full(dst)) {
118 sge_off = sge->offset + off;
119 sk_msg_page_add(dst, sg_page(sge), sge_len, sge_off);
126 sk_mem_charge(sk, sge_len);
127 sk_msg_iter_var_next(i);
128 if (i == src->sg.end && len)
130 sge = sk_msg_elem(src, i);
135 EXPORT_SYMBOL_GPL(sk_msg_clone);
137 void sk_msg_return_zero(struct sock *sk, struct sk_msg *msg, int bytes)
139 int i = msg->sg.start;
142 struct scatterlist *sge = sk_msg_elem(msg, i);
144 if (bytes < sge->length) {
145 sge->length -= bytes;
146 sge->offset += bytes;
147 sk_mem_uncharge(sk, bytes);
151 sk_mem_uncharge(sk, sge->length);
152 bytes -= sge->length;
155 sk_msg_iter_var_next(i);
156 } while (bytes && i != msg->sg.end);
159 EXPORT_SYMBOL_GPL(sk_msg_return_zero);
161 void sk_msg_return(struct sock *sk, struct sk_msg *msg, int bytes)
163 int i = msg->sg.start;
166 struct scatterlist *sge = &msg->sg.data[i];
167 int uncharge = (bytes < sge->length) ? bytes : sge->length;
169 sk_mem_uncharge(sk, uncharge);
171 sk_msg_iter_var_next(i);
172 } while (i != msg->sg.end);
174 EXPORT_SYMBOL_GPL(sk_msg_return);
176 static int sk_msg_free_elem(struct sock *sk, struct sk_msg *msg, u32 i,
179 struct scatterlist *sge = sk_msg_elem(msg, i);
180 u32 len = sge->length;
182 /* When the skb owns the memory we free it from consume_skb path. */
185 sk_mem_uncharge(sk, len);
186 put_page(sg_page(sge));
188 memset(sge, 0, sizeof(*sge));
192 static int __sk_msg_free(struct sock *sk, struct sk_msg *msg, u32 i,
195 struct scatterlist *sge = sk_msg_elem(msg, i);
198 while (msg->sg.size) {
199 msg->sg.size -= sge->length;
200 freed += sk_msg_free_elem(sk, msg, i, charge);
201 sk_msg_iter_var_next(i);
202 sk_msg_check_to_free(msg, i, msg->sg.size);
203 sge = sk_msg_elem(msg, i);
205 consume_skb(msg->skb);
210 int sk_msg_free_nocharge(struct sock *sk, struct sk_msg *msg)
212 return __sk_msg_free(sk, msg, msg->sg.start, false);
214 EXPORT_SYMBOL_GPL(sk_msg_free_nocharge);
216 int sk_msg_free(struct sock *sk, struct sk_msg *msg)
218 return __sk_msg_free(sk, msg, msg->sg.start, true);
220 EXPORT_SYMBOL_GPL(sk_msg_free);
222 static void __sk_msg_free_partial(struct sock *sk, struct sk_msg *msg,
223 u32 bytes, bool charge)
225 struct scatterlist *sge;
226 u32 i = msg->sg.start;
229 sge = sk_msg_elem(msg, i);
232 if (bytes < sge->length) {
234 sk_mem_uncharge(sk, bytes);
235 sge->length -= bytes;
236 sge->offset += bytes;
237 msg->sg.size -= bytes;
241 msg->sg.size -= sge->length;
242 bytes -= sge->length;
243 sk_msg_free_elem(sk, msg, i, charge);
244 sk_msg_iter_var_next(i);
245 sk_msg_check_to_free(msg, i, bytes);
250 void sk_msg_free_partial(struct sock *sk, struct sk_msg *msg, u32 bytes)
252 __sk_msg_free_partial(sk, msg, bytes, true);
254 EXPORT_SYMBOL_GPL(sk_msg_free_partial);
256 void sk_msg_free_partial_nocharge(struct sock *sk, struct sk_msg *msg,
259 __sk_msg_free_partial(sk, msg, bytes, false);
262 void sk_msg_trim(struct sock *sk, struct sk_msg *msg, int len)
264 int trim = msg->sg.size - len;
272 sk_msg_iter_var_prev(i);
274 while (msg->sg.data[i].length &&
275 trim >= msg->sg.data[i].length) {
276 trim -= msg->sg.data[i].length;
277 sk_msg_free_elem(sk, msg, i, true);
278 sk_msg_iter_var_prev(i);
283 msg->sg.data[i].length -= trim;
284 sk_mem_uncharge(sk, trim);
285 /* Adjust copybreak if it falls into the trimmed part of last buf */
286 if (msg->sg.curr == i && msg->sg.copybreak > msg->sg.data[i].length)
287 msg->sg.copybreak = msg->sg.data[i].length;
289 sk_msg_iter_var_next(i);
292 /* If we trim data a full sg elem before curr pointer update
293 * copybreak and current so that any future copy operations
294 * start at new copy location.
295 * However trimed data that has not yet been used in a copy op
296 * does not require an update.
299 msg->sg.curr = msg->sg.start;
300 msg->sg.copybreak = 0;
301 } else if (sk_msg_iter_dist(msg->sg.start, msg->sg.curr) >=
302 sk_msg_iter_dist(msg->sg.start, msg->sg.end)) {
303 sk_msg_iter_var_prev(i);
305 msg->sg.copybreak = msg->sg.data[i].length;
308 EXPORT_SYMBOL_GPL(sk_msg_trim);
310 int sk_msg_zerocopy_from_iter(struct sock *sk, struct iov_iter *from,
311 struct sk_msg *msg, u32 bytes)
313 int i, maxpages, ret = 0, num_elems = sk_msg_elem_used(msg);
314 const int to_max_pages = MAX_MSG_FRAGS;
315 struct page *pages[MAX_MSG_FRAGS];
316 ssize_t orig, copied, use, offset;
321 maxpages = to_max_pages - num_elems;
327 copied = iov_iter_get_pages(from, pages, bytes, maxpages,
334 iov_iter_advance(from, copied);
336 msg->sg.size += copied;
339 use = min_t(int, copied, PAGE_SIZE - offset);
340 sg_set_page(&msg->sg.data[msg->sg.end],
341 pages[i], use, offset);
342 sg_unmark_end(&msg->sg.data[msg->sg.end]);
343 sk_mem_charge(sk, use);
347 sk_msg_iter_next(msg, end);
351 /* When zerocopy is mixed with sk_msg_*copy* operations we
352 * may have a copybreak set in this case clear and prefer
353 * zerocopy remainder when possible.
355 msg->sg.copybreak = 0;
356 msg->sg.curr = msg->sg.end;
359 /* Revert iov_iter updates, msg will need to use 'trim' later if it
360 * also needs to be cleared.
363 iov_iter_revert(from, msg->sg.size - orig);
366 EXPORT_SYMBOL_GPL(sk_msg_zerocopy_from_iter);
368 int sk_msg_memcopy_from_iter(struct sock *sk, struct iov_iter *from,
369 struct sk_msg *msg, u32 bytes)
371 int ret = -ENOSPC, i = msg->sg.curr;
372 struct scatterlist *sge;
377 sge = sk_msg_elem(msg, i);
378 /* This is possible if a trim operation shrunk the buffer */
379 if (msg->sg.copybreak >= sge->length) {
380 msg->sg.copybreak = 0;
381 sk_msg_iter_var_next(i);
382 if (i == msg->sg.end)
384 sge = sk_msg_elem(msg, i);
387 buf_size = sge->length - msg->sg.copybreak;
388 copy = (buf_size > bytes) ? bytes : buf_size;
389 to = sg_virt(sge) + msg->sg.copybreak;
390 msg->sg.copybreak += copy;
391 if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY)
392 ret = copy_from_iter_nocache(to, copy, from);
394 ret = copy_from_iter(to, copy, from);
402 msg->sg.copybreak = 0;
403 sk_msg_iter_var_next(i);
404 } while (i != msg->sg.end);
409 EXPORT_SYMBOL_GPL(sk_msg_memcopy_from_iter);
411 /* Receive sk_msg from psock->ingress_msg to @msg. */
412 int sk_msg_recvmsg(struct sock *sk, struct sk_psock *psock, struct msghdr *msg,
415 struct iov_iter *iter = &msg->msg_iter;
416 int peek = flags & MSG_PEEK;
417 struct sk_msg *msg_rx;
420 msg_rx = sk_psock_peek_msg(psock);
421 while (copied != len) {
422 struct scatterlist *sge;
424 if (unlikely(!msg_rx))
427 i = msg_rx->sg.start;
432 sge = sk_msg_elem(msg_rx, i);
435 if (copied + copy > len)
437 copy = copy_page_to_iter(page, sge->offset, copy, iter);
439 return copied ? copied : -EFAULT;
446 sk_mem_uncharge(sk, copy);
447 msg_rx->sg.size -= copy;
450 sk_msg_iter_var_next(i);
455 /* Lets not optimize peek case if copy_page_to_iter
456 * didn't copy the entire length lets just break.
458 if (copy != sge->length)
460 sk_msg_iter_var_next(i);
465 } while (i != msg_rx->sg.end);
467 if (unlikely(peek)) {
468 msg_rx = sk_psock_next_msg(psock, msg_rx);
474 msg_rx->sg.start = i;
475 if (!sge->length && msg_rx->sg.start == msg_rx->sg.end) {
476 msg_rx = sk_psock_dequeue_msg(psock);
477 kfree_sk_msg(msg_rx);
479 msg_rx = sk_psock_peek_msg(psock);
484 EXPORT_SYMBOL_GPL(sk_msg_recvmsg);
486 bool sk_msg_is_readable(struct sock *sk)
488 struct sk_psock *psock;
492 psock = sk_psock(sk);
494 empty = list_empty(&psock->ingress_msg);
498 EXPORT_SYMBOL_GPL(sk_msg_is_readable);
500 static struct sk_msg *sk_psock_create_ingress_msg(struct sock *sk,
505 if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf)
508 if (!sk_rmem_schedule(sk, skb, skb->truesize))
511 msg = kzalloc(sizeof(*msg), __GFP_NOWARN | GFP_KERNEL);
519 static int sk_psock_skb_ingress_enqueue(struct sk_buff *skb,
521 struct sk_psock *psock,
527 num_sge = skb_to_sgvec(skb, msg->sg.data, off, len);
529 /* skb linearize may fail with ENOMEM, but lets simply try again
530 * later if this happens. Under memory pressure we don't want to
531 * drop the skb. We need to linearize the skb so that the mapping
532 * in skb_to_sgvec can not error.
534 if (skb_linearize(skb))
537 num_sge = skb_to_sgvec(skb, msg->sg.data, off, len);
538 if (unlikely(num_sge < 0))
544 msg->sg.size = copied;
545 msg->sg.end = num_sge;
548 sk_psock_queue_msg(psock, msg);
549 sk_psock_data_ready(sk, psock);
553 static int sk_psock_skb_ingress_self(struct sk_psock *psock, struct sk_buff *skb,
556 static int sk_psock_skb_ingress(struct sk_psock *psock, struct sk_buff *skb,
559 struct sock *sk = psock->sk;
563 /* If we are receiving on the same sock skb->sk is already assigned,
564 * skip memory accounting and owner transition seeing it already set
567 if (unlikely(skb->sk == sk))
568 return sk_psock_skb_ingress_self(psock, skb, off, len);
569 msg = sk_psock_create_ingress_msg(sk, skb);
573 /* This will transition ownership of the data from the socket where
574 * the BPF program was run initiating the redirect to the socket
575 * we will eventually receive this data on. The data will be released
576 * from skb_consume found in __tcp_bpf_recvmsg() after its been copied
579 skb_set_owner_r(skb, sk);
580 err = sk_psock_skb_ingress_enqueue(skb, off, len, psock, sk, msg);
586 /* Puts an skb on the ingress queue of the socket already assigned to the
587 * skb. In this case we do not need to check memory limits or skb_set_owner_r
588 * because the skb is already accounted for here.
590 static int sk_psock_skb_ingress_self(struct sk_psock *psock, struct sk_buff *skb,
593 struct sk_msg *msg = kzalloc(sizeof(*msg), __GFP_NOWARN | GFP_ATOMIC);
594 struct sock *sk = psock->sk;
600 skb_set_owner_r(skb, sk);
601 err = sk_psock_skb_ingress_enqueue(skb, off, len, psock, sk, msg);
607 static int sk_psock_handle_skb(struct sk_psock *psock, struct sk_buff *skb,
608 u32 off, u32 len, bool ingress)
611 if (!sock_writeable(psock->sk))
613 return skb_send_sock(psock->sk, skb, off, len);
615 return sk_psock_skb_ingress(psock, skb, off, len);
618 static void sk_psock_skb_state(struct sk_psock *psock,
619 struct sk_psock_work_state *state,
623 spin_lock_bh(&psock->ingress_lock);
624 if (sk_psock_test_state(psock, SK_PSOCK_TX_ENABLED)) {
629 sock_drop(psock->sk, skb);
631 spin_unlock_bh(&psock->ingress_lock);
634 static void sk_psock_backlog(struct work_struct *work)
636 struct sk_psock *psock = container_of(work, struct sk_psock, work);
637 struct sk_psock_work_state *state = &psock->work_state;
638 struct sk_buff *skb = NULL;
643 mutex_lock(&psock->work_mutex);
644 if (unlikely(state->skb)) {
645 spin_lock_bh(&psock->ingress_lock);
650 spin_unlock_bh(&psock->ingress_lock);
655 while ((skb = skb_dequeue(&psock->ingress_skb))) {
658 if (skb_bpf_strparser(skb)) {
659 struct strp_msg *stm = strp_msg(skb);
665 ingress = skb_bpf_ingress(skb);
666 skb_bpf_redirect_clear(skb);
669 if (!sock_flag(psock->sk, SOCK_DEAD))
670 ret = sk_psock_handle_skb(psock, skb, off,
673 if (ret == -EAGAIN) {
674 sk_psock_skb_state(psock, state, skb,
678 /* Hard errors break pipe and stop xmit. */
679 sk_psock_report_error(psock, ret ? -ret : EPIPE);
680 sk_psock_clear_state(psock, SK_PSOCK_TX_ENABLED);
681 sock_drop(psock->sk, skb);
692 mutex_unlock(&psock->work_mutex);
695 struct sk_psock *sk_psock_init(struct sock *sk, int node)
697 struct sk_psock *psock;
700 write_lock_bh(&sk->sk_callback_lock);
702 if (sk_is_inet(sk) && inet_csk_has_ulp(sk)) {
703 psock = ERR_PTR(-EINVAL);
707 if (sk->sk_user_data) {
708 psock = ERR_PTR(-EBUSY);
712 psock = kzalloc_node(sizeof(*psock), GFP_ATOMIC | __GFP_NOWARN, node);
714 psock = ERR_PTR(-ENOMEM);
718 prot = READ_ONCE(sk->sk_prot);
720 psock->eval = __SK_NONE;
721 psock->sk_proto = prot;
722 psock->saved_unhash = prot->unhash;
723 psock->saved_close = prot->close;
724 psock->saved_write_space = sk->sk_write_space;
726 INIT_LIST_HEAD(&psock->link);
727 spin_lock_init(&psock->link_lock);
729 INIT_WORK(&psock->work, sk_psock_backlog);
730 mutex_init(&psock->work_mutex);
731 INIT_LIST_HEAD(&psock->ingress_msg);
732 spin_lock_init(&psock->ingress_lock);
733 skb_queue_head_init(&psock->ingress_skb);
735 sk_psock_set_state(psock, SK_PSOCK_TX_ENABLED);
736 refcount_set(&psock->refcnt, 1);
738 rcu_assign_sk_user_data_nocopy(sk, psock);
742 write_unlock_bh(&sk->sk_callback_lock);
745 EXPORT_SYMBOL_GPL(sk_psock_init);
747 struct sk_psock_link *sk_psock_link_pop(struct sk_psock *psock)
749 struct sk_psock_link *link;
751 spin_lock_bh(&psock->link_lock);
752 link = list_first_entry_or_null(&psock->link, struct sk_psock_link,
755 list_del(&link->list);
756 spin_unlock_bh(&psock->link_lock);
760 static void __sk_psock_purge_ingress_msg(struct sk_psock *psock)
762 struct sk_msg *msg, *tmp;
764 list_for_each_entry_safe(msg, tmp, &psock->ingress_msg, list) {
765 list_del(&msg->list);
766 sk_msg_free(psock->sk, msg);
771 static void __sk_psock_zap_ingress(struct sk_psock *psock)
775 while ((skb = skb_dequeue(&psock->ingress_skb)) != NULL) {
776 skb_bpf_redirect_clear(skb);
777 sock_drop(psock->sk, skb);
779 kfree_skb(psock->work_state.skb);
780 /* We null the skb here to ensure that calls to sk_psock_backlog
781 * do not pick up the free'd skb.
783 psock->work_state.skb = NULL;
784 __sk_psock_purge_ingress_msg(psock);
787 static void sk_psock_link_destroy(struct sk_psock *psock)
789 struct sk_psock_link *link, *tmp;
791 list_for_each_entry_safe(link, tmp, &psock->link, list) {
792 list_del(&link->list);
793 sk_psock_free_link(link);
797 void sk_psock_stop(struct sk_psock *psock, bool wait)
799 spin_lock_bh(&psock->ingress_lock);
800 sk_psock_clear_state(psock, SK_PSOCK_TX_ENABLED);
801 sk_psock_cork_free(psock);
802 __sk_psock_zap_ingress(psock);
803 spin_unlock_bh(&psock->ingress_lock);
806 cancel_work_sync(&psock->work);
809 static void sk_psock_done_strp(struct sk_psock *psock);
811 static void sk_psock_destroy(struct work_struct *work)
813 struct sk_psock *psock = container_of(to_rcu_work(work),
814 struct sk_psock, rwork);
815 /* No sk_callback_lock since already detached. */
817 sk_psock_done_strp(psock);
819 cancel_work_sync(&psock->work);
820 mutex_destroy(&psock->work_mutex);
822 psock_progs_drop(&psock->progs);
824 sk_psock_link_destroy(psock);
825 sk_psock_cork_free(psock);
828 sock_put(psock->sk_redir);
833 void sk_psock_drop(struct sock *sk, struct sk_psock *psock)
835 write_lock_bh(&sk->sk_callback_lock);
836 sk_psock_restore_proto(sk, psock);
837 rcu_assign_sk_user_data(sk, NULL);
838 if (psock->progs.stream_parser)
839 sk_psock_stop_strp(sk, psock);
840 else if (psock->progs.stream_verdict || psock->progs.skb_verdict)
841 sk_psock_stop_verdict(sk, psock);
842 write_unlock_bh(&sk->sk_callback_lock);
844 sk_psock_stop(psock, false);
846 INIT_RCU_WORK(&psock->rwork, sk_psock_destroy);
847 queue_rcu_work(system_wq, &psock->rwork);
849 EXPORT_SYMBOL_GPL(sk_psock_drop);
851 static int sk_psock_map_verd(int verdict, bool redir)
855 return redir ? __SK_REDIRECT : __SK_PASS;
864 int sk_psock_msg_verdict(struct sock *sk, struct sk_psock *psock,
867 struct bpf_prog *prog;
871 prog = READ_ONCE(psock->progs.msg_parser);
872 if (unlikely(!prog)) {
877 sk_msg_compute_data_pointers(msg);
879 ret = bpf_prog_run_pin_on_cpu(prog, msg);
880 ret = sk_psock_map_verd(ret, msg->sk_redir);
881 psock->apply_bytes = msg->apply_bytes;
882 if (ret == __SK_REDIRECT) {
884 sock_put(psock->sk_redir);
885 psock->sk_redir = msg->sk_redir;
886 if (!psock->sk_redir) {
890 sock_hold(psock->sk_redir);
896 EXPORT_SYMBOL_GPL(sk_psock_msg_verdict);
898 static int sk_psock_skb_redirect(struct sk_psock *from, struct sk_buff *skb)
900 struct sk_psock *psock_other;
901 struct sock *sk_other;
903 sk_other = skb_bpf_redirect_fetch(skb);
904 /* This error is a buggy BPF program, it returned a redirect
905 * return code, but then didn't set a redirect interface.
907 if (unlikely(!sk_other)) {
908 skb_bpf_redirect_clear(skb);
909 sock_drop(from->sk, skb);
912 psock_other = sk_psock(sk_other);
913 /* This error indicates the socket is being torn down or had another
914 * error that caused the pipe to break. We can't send a packet on
915 * a socket that is in this state so we drop the skb.
917 if (!psock_other || sock_flag(sk_other, SOCK_DEAD)) {
918 skb_bpf_redirect_clear(skb);
919 sock_drop(from->sk, skb);
922 spin_lock_bh(&psock_other->ingress_lock);
923 if (!sk_psock_test_state(psock_other, SK_PSOCK_TX_ENABLED)) {
924 spin_unlock_bh(&psock_other->ingress_lock);
925 skb_bpf_redirect_clear(skb);
926 sock_drop(from->sk, skb);
930 skb_queue_tail(&psock_other->ingress_skb, skb);
931 schedule_work(&psock_other->work);
932 spin_unlock_bh(&psock_other->ingress_lock);
936 static void sk_psock_tls_verdict_apply(struct sk_buff *skb,
937 struct sk_psock *from, int verdict)
941 sk_psock_skb_redirect(from, skb);
950 int sk_psock_tls_strp_read(struct sk_psock *psock, struct sk_buff *skb)
952 struct bpf_prog *prog;
956 prog = READ_ONCE(psock->progs.stream_verdict);
960 skb_bpf_redirect_clear(skb);
961 ret = bpf_prog_run_pin_on_cpu(prog, skb);
962 ret = sk_psock_map_verd(ret, skb_bpf_redirect_fetch(skb));
965 sk_psock_tls_verdict_apply(skb, psock, ret);
969 EXPORT_SYMBOL_GPL(sk_psock_tls_strp_read);
971 static int sk_psock_verdict_apply(struct sk_psock *psock, struct sk_buff *skb,
974 struct sock *sk_other;
981 sk_other = psock->sk;
982 if (sock_flag(sk_other, SOCK_DEAD) ||
983 !sk_psock_test_state(psock, SK_PSOCK_TX_ENABLED)) {
984 skb_bpf_redirect_clear(skb);
988 skb_bpf_set_ingress(skb);
990 /* If the queue is empty then we can submit directly
991 * into the msg queue. If its not empty we have to
992 * queue work otherwise we may get OOO data. Otherwise,
993 * if sk_psock_skb_ingress errors will be handled by
994 * retrying later from workqueue.
996 if (skb_queue_empty(&psock->ingress_skb)) {
999 if (skb_bpf_strparser(skb)) {
1000 struct strp_msg *stm = strp_msg(skb);
1003 len = stm->full_len;
1005 err = sk_psock_skb_ingress_self(psock, skb, off, len);
1008 spin_lock_bh(&psock->ingress_lock);
1009 if (sk_psock_test_state(psock, SK_PSOCK_TX_ENABLED)) {
1010 skb_queue_tail(&psock->ingress_skb, skb);
1011 schedule_work(&psock->work);
1014 spin_unlock_bh(&psock->ingress_lock);
1016 skb_bpf_redirect_clear(skb);
1022 err = sk_psock_skb_redirect(psock, skb);
1027 sock_drop(psock->sk, skb);
1033 static void sk_psock_write_space(struct sock *sk)
1035 struct sk_psock *psock;
1036 void (*write_space)(struct sock *sk) = NULL;
1039 psock = sk_psock(sk);
1040 if (likely(psock)) {
1041 if (sk_psock_test_state(psock, SK_PSOCK_TX_ENABLED))
1042 schedule_work(&psock->work);
1043 write_space = psock->saved_write_space;
1050 #if IS_ENABLED(CONFIG_BPF_STREAM_PARSER)
1051 static void sk_psock_strp_read(struct strparser *strp, struct sk_buff *skb)
1053 struct sk_psock *psock;
1054 struct bpf_prog *prog;
1055 int ret = __SK_DROP;
1060 psock = sk_psock(sk);
1061 if (unlikely(!psock)) {
1065 prog = READ_ONCE(psock->progs.stream_verdict);
1069 skb_bpf_redirect_clear(skb);
1070 ret = bpf_prog_run_pin_on_cpu(prog, skb);
1072 skb_bpf_set_strparser(skb);
1073 ret = sk_psock_map_verd(ret, skb_bpf_redirect_fetch(skb));
1076 sk_psock_verdict_apply(psock, skb, ret);
1081 static int sk_psock_strp_read_done(struct strparser *strp, int err)
1086 static int sk_psock_strp_parse(struct strparser *strp, struct sk_buff *skb)
1088 struct sk_psock *psock = container_of(strp, struct sk_psock, strp);
1089 struct bpf_prog *prog;
1093 prog = READ_ONCE(psock->progs.stream_parser);
1095 skb->sk = psock->sk;
1096 ret = bpf_prog_run_pin_on_cpu(prog, skb);
1103 /* Called with socket lock held. */
1104 static void sk_psock_strp_data_ready(struct sock *sk)
1106 struct sk_psock *psock;
1109 psock = sk_psock(sk);
1110 if (likely(psock)) {
1111 if (tls_sw_has_ctx_rx(sk)) {
1112 psock->saved_data_ready(sk);
1114 write_lock_bh(&sk->sk_callback_lock);
1115 strp_data_ready(&psock->strp);
1116 write_unlock_bh(&sk->sk_callback_lock);
1122 int sk_psock_init_strp(struct sock *sk, struct sk_psock *psock)
1124 static const struct strp_callbacks cb = {
1125 .rcv_msg = sk_psock_strp_read,
1126 .read_sock_done = sk_psock_strp_read_done,
1127 .parse_msg = sk_psock_strp_parse,
1130 return strp_init(&psock->strp, sk, &cb);
1133 void sk_psock_start_strp(struct sock *sk, struct sk_psock *psock)
1135 if (psock->saved_data_ready)
1138 psock->saved_data_ready = sk->sk_data_ready;
1139 sk->sk_data_ready = sk_psock_strp_data_ready;
1140 sk->sk_write_space = sk_psock_write_space;
1143 void sk_psock_stop_strp(struct sock *sk, struct sk_psock *psock)
1145 psock_set_prog(&psock->progs.stream_parser, NULL);
1147 if (!psock->saved_data_ready)
1150 sk->sk_data_ready = psock->saved_data_ready;
1151 psock->saved_data_ready = NULL;
1152 strp_stop(&psock->strp);
1155 static void sk_psock_done_strp(struct sk_psock *psock)
1157 /* Parser has been stopped */
1158 if (psock->progs.stream_parser)
1159 strp_done(&psock->strp);
1162 static void sk_psock_done_strp(struct sk_psock *psock)
1165 #endif /* CONFIG_BPF_STREAM_PARSER */
1167 static int sk_psock_verdict_recv(read_descriptor_t *desc, struct sk_buff *skb,
1168 unsigned int offset, size_t orig_len)
1170 struct sock *sk = (struct sock *)desc->arg.data;
1171 struct sk_psock *psock;
1172 struct bpf_prog *prog;
1173 int ret = __SK_DROP;
1176 /* clone here so sk_eat_skb() in tcp_read_sock does not drop our data */
1177 skb = skb_clone(skb, GFP_ATOMIC);
1179 desc->error = -ENOMEM;
1184 psock = sk_psock(sk);
1185 if (unlikely(!psock)) {
1190 prog = READ_ONCE(psock->progs.stream_verdict);
1192 prog = READ_ONCE(psock->progs.skb_verdict);
1196 skb_bpf_redirect_clear(skb);
1197 ret = bpf_prog_run_pin_on_cpu(prog, skb);
1198 ret = sk_psock_map_verd(ret, skb_bpf_redirect_fetch(skb));
1201 if (sk_psock_verdict_apply(psock, skb, ret) < 0)
1208 static void sk_psock_verdict_data_ready(struct sock *sk)
1210 struct socket *sock = sk->sk_socket;
1211 read_descriptor_t desc;
1213 if (unlikely(!sock || !sock->ops || !sock->ops->read_sock))
1220 sock->ops->read_sock(sk, &desc, sk_psock_verdict_recv);
1223 void sk_psock_start_verdict(struct sock *sk, struct sk_psock *psock)
1225 if (psock->saved_data_ready)
1228 psock->saved_data_ready = sk->sk_data_ready;
1229 sk->sk_data_ready = sk_psock_verdict_data_ready;
1230 sk->sk_write_space = sk_psock_write_space;
1233 void sk_psock_stop_verdict(struct sock *sk, struct sk_psock *psock)
1235 psock_set_prog(&psock->progs.stream_verdict, NULL);
1236 psock_set_prog(&psock->progs.skb_verdict, NULL);
1238 if (!psock->saved_data_ready)
1241 sk->sk_data_ready = psock->saved_data_ready;
1242 psock->saved_data_ready = NULL;