2 * Copyright (c) 2016-2017, Mellanox Technologies. All rights reserved.
3 * Copyright (c) 2016-2017, Dave Watson <davejwatson@fb.com>. All rights reserved.
4 * Copyright (c) 2016-2017, Lance Chao <lancerchao@fb.com>. All rights reserved.
5 * Copyright (c) 2016, Fridolin Pokorny <fridolin.pokorny@gmail.com>. All rights reserved.
6 * Copyright (c) 2016, Nikos Mavrogiannopoulos <nmav@gnutls.org>. All rights reserved.
7 * Copyright (c) 2018, Covalent IO, Inc. http://covalent.io
9 * This software is available to you under a choice of one of two
10 * licenses. You may choose to be licensed under the terms of the GNU
11 * General Public License (GPL) Version 2, available from the file
12 * COPYING in the main directory of this source tree, or the
13 * OpenIB.org BSD license below:
15 * Redistribution and use in source and binary forms, with or
16 * without modification, are permitted provided that the following
19 * - Redistributions of source code must retain the above
20 * copyright notice, this list of conditions and the following
23 * - Redistributions in binary form must reproduce the above
24 * copyright notice, this list of conditions and the following
25 * disclaimer in the documentation and/or other materials
26 * provided with the distribution.
28 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
29 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
30 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
31 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
32 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
33 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
34 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
38 #include <linux/sched/signal.h>
39 #include <linux/module.h>
40 #include <linux/splice.h>
41 #include <crypto/aead.h>
43 #include <net/strparser.h>
46 static int __skb_nsg(struct sk_buff *skb, int offset, int len,
47 unsigned int recursion_level)
49 int start = skb_headlen(skb);
50 int i, chunk = start - offset;
51 struct sk_buff *frag_iter;
54 if (unlikely(recursion_level >= 24))
67 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
70 WARN_ON(start > offset + len);
72 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
86 if (unlikely(skb_has_frag_list(skb))) {
87 skb_walk_frags(skb, frag_iter) {
90 WARN_ON(start > offset + len);
92 end = start + frag_iter->len;
97 ret = __skb_nsg(frag_iter, offset - start, chunk,
99 if (unlikely(ret < 0))
114 /* Return the number of scatterlist elements required to completely map the
115 * skb, or -EMSGSIZE if the recursion depth is exceeded.
117 static int skb_nsg(struct sk_buff *skb, int offset, int len)
119 return __skb_nsg(skb, offset, len, 0);
122 static int padding_length(struct tls_sw_context_rx *ctx,
123 struct tls_prot_info *prot, struct sk_buff *skb)
125 struct strp_msg *rxm = strp_msg(skb);
128 /* Determine zero-padding length */
129 if (prot->version == TLS_1_3_VERSION) {
130 char content_type = 0;
134 while (content_type == 0) {
135 if (back > rxm->full_len - prot->prepend_size)
137 err = skb_copy_bits(skb,
138 rxm->offset + rxm->full_len - back,
147 ctx->control = content_type;
152 static void tls_decrypt_done(struct crypto_async_request *req, int err)
154 struct aead_request *aead_req = (struct aead_request *)req;
155 struct scatterlist *sgout = aead_req->dst;
156 struct scatterlist *sgin = aead_req->src;
157 struct tls_sw_context_rx *ctx;
158 struct tls_context *tls_ctx;
159 struct tls_prot_info *prot;
160 struct scatterlist *sg;
165 skb = (struct sk_buff *)req->data;
166 tls_ctx = tls_get_ctx(skb->sk);
167 ctx = tls_sw_ctx_rx(tls_ctx);
168 prot = &tls_ctx->prot_info;
170 /* Propagate if there was an err */
173 TLS_INC_STATS(sock_net(skb->sk),
174 LINUX_MIB_TLSDECRYPTERROR);
175 ctx->async_wait.err = err;
176 tls_err_abort(skb->sk, err);
178 struct strp_msg *rxm = strp_msg(skb);
181 pad = padding_length(ctx, prot, skb);
183 ctx->async_wait.err = pad;
184 tls_err_abort(skb->sk, pad);
186 rxm->full_len -= pad;
187 rxm->offset += prot->prepend_size;
188 rxm->full_len -= prot->overhead_size;
192 /* After using skb->sk to propagate sk through crypto async callback
193 * we need to NULL it again.
198 /* Free the destination pages if skb was not decrypted inplace */
200 /* Skip the first S/G entry as it points to AAD */
201 for_each_sg(sg_next(sgout), sg, UINT_MAX, pages) {
204 put_page(sg_page(sg));
210 spin_lock_bh(&ctx->decrypt_compl_lock);
211 pending = atomic_dec_return(&ctx->decrypt_pending);
213 if (!pending && ctx->async_notify)
214 complete(&ctx->async_wait.completion);
215 spin_unlock_bh(&ctx->decrypt_compl_lock);
218 static int tls_do_decryption(struct sock *sk,
220 struct scatterlist *sgin,
221 struct scatterlist *sgout,
224 struct aead_request *aead_req,
227 struct tls_context *tls_ctx = tls_get_ctx(sk);
228 struct tls_prot_info *prot = &tls_ctx->prot_info;
229 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
232 aead_request_set_tfm(aead_req, ctx->aead_recv);
233 aead_request_set_ad(aead_req, prot->aad_size);
234 aead_request_set_crypt(aead_req, sgin, sgout,
235 data_len + prot->tag_size,
239 /* Using skb->sk to push sk through to crypto async callback
240 * handler. This allows propagating errors up to the socket
241 * if needed. It _must_ be cleared in the async handler
242 * before consume_skb is called. We _know_ skb->sk is NULL
243 * because it is a clone from strparser.
246 aead_request_set_callback(aead_req,
247 CRYPTO_TFM_REQ_MAY_BACKLOG,
248 tls_decrypt_done, skb);
249 atomic_inc(&ctx->decrypt_pending);
251 aead_request_set_callback(aead_req,
252 CRYPTO_TFM_REQ_MAY_BACKLOG,
253 crypto_req_done, &ctx->async_wait);
256 ret = crypto_aead_decrypt(aead_req);
257 if (ret == -EINPROGRESS) {
261 ret = crypto_wait_req(ret, &ctx->async_wait);
265 atomic_dec(&ctx->decrypt_pending);
270 static void tls_trim_both_msgs(struct sock *sk, int target_size)
272 struct tls_context *tls_ctx = tls_get_ctx(sk);
273 struct tls_prot_info *prot = &tls_ctx->prot_info;
274 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
275 struct tls_rec *rec = ctx->open_rec;
277 sk_msg_trim(sk, &rec->msg_plaintext, target_size);
279 target_size += prot->overhead_size;
280 sk_msg_trim(sk, &rec->msg_encrypted, target_size);
283 static int tls_alloc_encrypted_msg(struct sock *sk, int len)
285 struct tls_context *tls_ctx = tls_get_ctx(sk);
286 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
287 struct tls_rec *rec = ctx->open_rec;
288 struct sk_msg *msg_en = &rec->msg_encrypted;
290 return sk_msg_alloc(sk, msg_en, len, 0);
293 static int tls_clone_plaintext_msg(struct sock *sk, int required)
295 struct tls_context *tls_ctx = tls_get_ctx(sk);
296 struct tls_prot_info *prot = &tls_ctx->prot_info;
297 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
298 struct tls_rec *rec = ctx->open_rec;
299 struct sk_msg *msg_pl = &rec->msg_plaintext;
300 struct sk_msg *msg_en = &rec->msg_encrypted;
303 /* We add page references worth len bytes from encrypted sg
304 * at the end of plaintext sg. It is guaranteed that msg_en
305 * has enough required room (ensured by caller).
307 len = required - msg_pl->sg.size;
309 /* Skip initial bytes in msg_en's data to be able to use
310 * same offset of both plain and encrypted data.
312 skip = prot->prepend_size + msg_pl->sg.size;
314 return sk_msg_clone(sk, msg_pl, msg_en, skip, len);
317 static struct tls_rec *tls_get_rec(struct sock *sk)
319 struct tls_context *tls_ctx = tls_get_ctx(sk);
320 struct tls_prot_info *prot = &tls_ctx->prot_info;
321 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
322 struct sk_msg *msg_pl, *msg_en;
326 mem_size = sizeof(struct tls_rec) + crypto_aead_reqsize(ctx->aead_send);
328 rec = kzalloc(mem_size, sk->sk_allocation);
332 msg_pl = &rec->msg_plaintext;
333 msg_en = &rec->msg_encrypted;
338 sg_init_table(rec->sg_aead_in, 2);
339 sg_set_buf(&rec->sg_aead_in[0], rec->aad_space, prot->aad_size);
340 sg_unmark_end(&rec->sg_aead_in[1]);
342 sg_init_table(rec->sg_aead_out, 2);
343 sg_set_buf(&rec->sg_aead_out[0], rec->aad_space, prot->aad_size);
344 sg_unmark_end(&rec->sg_aead_out[1]);
349 static void tls_free_rec(struct sock *sk, struct tls_rec *rec)
351 sk_msg_free(sk, &rec->msg_encrypted);
352 sk_msg_free(sk, &rec->msg_plaintext);
356 static void tls_free_open_rec(struct sock *sk)
358 struct tls_context *tls_ctx = tls_get_ctx(sk);
359 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
360 struct tls_rec *rec = ctx->open_rec;
363 tls_free_rec(sk, rec);
364 ctx->open_rec = NULL;
368 int tls_tx_records(struct sock *sk, int flags)
370 struct tls_context *tls_ctx = tls_get_ctx(sk);
371 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
372 struct tls_rec *rec, *tmp;
373 struct sk_msg *msg_en;
374 int tx_flags, rc = 0;
376 if (tls_is_partially_sent_record(tls_ctx)) {
377 rec = list_first_entry(&ctx->tx_list,
378 struct tls_rec, list);
381 tx_flags = rec->tx_flags;
385 rc = tls_push_partial_record(sk, tls_ctx, tx_flags);
389 /* Full record has been transmitted.
390 * Remove the head of tx_list
392 list_del(&rec->list);
393 sk_msg_free(sk, &rec->msg_plaintext);
397 /* Tx all ready records */
398 list_for_each_entry_safe(rec, tmp, &ctx->tx_list, list) {
399 if (READ_ONCE(rec->tx_ready)) {
401 tx_flags = rec->tx_flags;
405 msg_en = &rec->msg_encrypted;
406 rc = tls_push_sg(sk, tls_ctx,
407 &msg_en->sg.data[msg_en->sg.curr],
412 list_del(&rec->list);
413 sk_msg_free(sk, &rec->msg_plaintext);
421 if (rc < 0 && rc != -EAGAIN)
422 tls_err_abort(sk, EBADMSG);
427 static void tls_encrypt_done(struct crypto_async_request *req, int err)
429 struct aead_request *aead_req = (struct aead_request *)req;
430 struct sock *sk = req->data;
431 struct tls_context *tls_ctx = tls_get_ctx(sk);
432 struct tls_prot_info *prot = &tls_ctx->prot_info;
433 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
434 struct scatterlist *sge;
435 struct sk_msg *msg_en;
440 rec = container_of(aead_req, struct tls_rec, aead_req);
441 msg_en = &rec->msg_encrypted;
443 sge = sk_msg_elem(msg_en, msg_en->sg.curr);
444 sge->offset -= prot->prepend_size;
445 sge->length += prot->prepend_size;
447 /* Check if error is previously set on socket */
448 if (err || sk->sk_err) {
451 /* If err is already set on socket, return the same code */
453 ctx->async_wait.err = sk->sk_err;
455 ctx->async_wait.err = err;
456 tls_err_abort(sk, err);
461 struct tls_rec *first_rec;
463 /* Mark the record as ready for transmission */
464 smp_store_mb(rec->tx_ready, true);
466 /* If received record is at head of tx_list, schedule tx */
467 first_rec = list_first_entry(&ctx->tx_list,
468 struct tls_rec, list);
469 if (rec == first_rec)
473 spin_lock_bh(&ctx->encrypt_compl_lock);
474 pending = atomic_dec_return(&ctx->encrypt_pending);
476 if (!pending && ctx->async_notify)
477 complete(&ctx->async_wait.completion);
478 spin_unlock_bh(&ctx->encrypt_compl_lock);
483 /* Schedule the transmission */
484 if (!test_and_set_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask))
485 schedule_delayed_work(&ctx->tx_work.work, 1);
488 static int tls_do_encryption(struct sock *sk,
489 struct tls_context *tls_ctx,
490 struct tls_sw_context_tx *ctx,
491 struct aead_request *aead_req,
492 size_t data_len, u32 start)
494 struct tls_prot_info *prot = &tls_ctx->prot_info;
495 struct tls_rec *rec = ctx->open_rec;
496 struct sk_msg *msg_en = &rec->msg_encrypted;
497 struct scatterlist *sge = sk_msg_elem(msg_en, start);
498 int rc, iv_offset = 0;
500 /* For CCM based ciphers, first byte of IV is a constant */
501 if (prot->cipher_type == TLS_CIPHER_AES_CCM_128) {
502 rec->iv_data[0] = TLS_AES_CCM_IV_B0_BYTE;
506 memcpy(&rec->iv_data[iv_offset], tls_ctx->tx.iv,
507 prot->iv_size + prot->salt_size);
509 xor_iv_with_seq(prot, rec->iv_data, tls_ctx->tx.rec_seq);
511 sge->offset += prot->prepend_size;
512 sge->length -= prot->prepend_size;
514 msg_en->sg.curr = start;
516 aead_request_set_tfm(aead_req, ctx->aead_send);
517 aead_request_set_ad(aead_req, prot->aad_size);
518 aead_request_set_crypt(aead_req, rec->sg_aead_in,
520 data_len, rec->iv_data);
522 aead_request_set_callback(aead_req, CRYPTO_TFM_REQ_MAY_BACKLOG,
523 tls_encrypt_done, sk);
525 /* Add the record in tx_list */
526 list_add_tail((struct list_head *)&rec->list, &ctx->tx_list);
527 atomic_inc(&ctx->encrypt_pending);
529 rc = crypto_aead_encrypt(aead_req);
530 if (!rc || rc != -EINPROGRESS) {
531 atomic_dec(&ctx->encrypt_pending);
532 sge->offset -= prot->prepend_size;
533 sge->length += prot->prepend_size;
537 WRITE_ONCE(rec->tx_ready, true);
538 } else if (rc != -EINPROGRESS) {
539 list_del(&rec->list);
543 /* Unhook the record from context if encryption is not failure */
544 ctx->open_rec = NULL;
545 tls_advance_record_sn(sk, prot, &tls_ctx->tx);
549 static int tls_split_open_record(struct sock *sk, struct tls_rec *from,
550 struct tls_rec **to, struct sk_msg *msg_opl,
551 struct sk_msg *msg_oen, u32 split_point,
552 u32 tx_overhead_size, u32 *orig_end)
554 u32 i, j, bytes = 0, apply = msg_opl->apply_bytes;
555 struct scatterlist *sge, *osge, *nsge;
556 u32 orig_size = msg_opl->sg.size;
557 struct scatterlist tmp = { };
558 struct sk_msg *msg_npl;
562 new = tls_get_rec(sk);
565 ret = sk_msg_alloc(sk, &new->msg_encrypted, msg_opl->sg.size +
566 tx_overhead_size, 0);
568 tls_free_rec(sk, new);
572 *orig_end = msg_opl->sg.end;
573 i = msg_opl->sg.start;
574 sge = sk_msg_elem(msg_opl, i);
575 while (apply && sge->length) {
576 if (sge->length > apply) {
577 u32 len = sge->length - apply;
579 get_page(sg_page(sge));
580 sg_set_page(&tmp, sg_page(sge), len,
581 sge->offset + apply);
586 apply -= sge->length;
587 bytes += sge->length;
590 sk_msg_iter_var_next(i);
591 if (i == msg_opl->sg.end)
593 sge = sk_msg_elem(msg_opl, i);
597 msg_opl->sg.curr = i;
598 msg_opl->sg.copybreak = 0;
599 msg_opl->apply_bytes = 0;
600 msg_opl->sg.size = bytes;
602 msg_npl = &new->msg_plaintext;
603 msg_npl->apply_bytes = apply;
604 msg_npl->sg.size = orig_size - bytes;
606 j = msg_npl->sg.start;
607 nsge = sk_msg_elem(msg_npl, j);
609 memcpy(nsge, &tmp, sizeof(*nsge));
610 sk_msg_iter_var_next(j);
611 nsge = sk_msg_elem(msg_npl, j);
614 osge = sk_msg_elem(msg_opl, i);
615 while (osge->length) {
616 memcpy(nsge, osge, sizeof(*nsge));
618 sk_msg_iter_var_next(i);
619 sk_msg_iter_var_next(j);
622 osge = sk_msg_elem(msg_opl, i);
623 nsge = sk_msg_elem(msg_npl, j);
627 msg_npl->sg.curr = j;
628 msg_npl->sg.copybreak = 0;
634 static void tls_merge_open_record(struct sock *sk, struct tls_rec *to,
635 struct tls_rec *from, u32 orig_end)
637 struct sk_msg *msg_npl = &from->msg_plaintext;
638 struct sk_msg *msg_opl = &to->msg_plaintext;
639 struct scatterlist *osge, *nsge;
643 sk_msg_iter_var_prev(i);
644 j = msg_npl->sg.start;
646 osge = sk_msg_elem(msg_opl, i);
647 nsge = sk_msg_elem(msg_npl, j);
649 if (sg_page(osge) == sg_page(nsge) &&
650 osge->offset + osge->length == nsge->offset) {
651 osge->length += nsge->length;
652 put_page(sg_page(nsge));
655 msg_opl->sg.end = orig_end;
656 msg_opl->sg.curr = orig_end;
657 msg_opl->sg.copybreak = 0;
658 msg_opl->apply_bytes = msg_opl->sg.size + msg_npl->sg.size;
659 msg_opl->sg.size += msg_npl->sg.size;
661 sk_msg_free(sk, &to->msg_encrypted);
662 sk_msg_xfer_full(&to->msg_encrypted, &from->msg_encrypted);
667 static int tls_push_record(struct sock *sk, int flags,
668 unsigned char record_type)
670 struct tls_context *tls_ctx = tls_get_ctx(sk);
671 struct tls_prot_info *prot = &tls_ctx->prot_info;
672 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
673 struct tls_rec *rec = ctx->open_rec, *tmp = NULL;
674 u32 i, split_point, orig_end;
675 struct sk_msg *msg_pl, *msg_en;
676 struct aead_request *req;
683 msg_pl = &rec->msg_plaintext;
684 msg_en = &rec->msg_encrypted;
686 split_point = msg_pl->apply_bytes;
687 split = split_point && split_point < msg_pl->sg.size;
688 if (unlikely((!split &&
690 prot->overhead_size > msg_en->sg.size) ||
693 prot->overhead_size > msg_en->sg.size))) {
695 split_point = msg_en->sg.size;
698 rc = tls_split_open_record(sk, rec, &tmp, msg_pl, msg_en,
699 split_point, prot->overhead_size,
703 /* This can happen if above tls_split_open_record allocates
704 * a single large encryption buffer instead of two smaller
705 * ones. In this case adjust pointers and continue without
708 if (!msg_pl->sg.size) {
709 tls_merge_open_record(sk, rec, tmp, orig_end);
710 msg_pl = &rec->msg_plaintext;
711 msg_en = &rec->msg_encrypted;
714 sk_msg_trim(sk, msg_en, msg_pl->sg.size +
715 prot->overhead_size);
718 rec->tx_flags = flags;
719 req = &rec->aead_req;
722 sk_msg_iter_var_prev(i);
724 rec->content_type = record_type;
725 if (prot->version == TLS_1_3_VERSION) {
726 /* Add content type to end of message. No padding added */
727 sg_set_buf(&rec->sg_content_type, &rec->content_type, 1);
728 sg_mark_end(&rec->sg_content_type);
729 sg_chain(msg_pl->sg.data, msg_pl->sg.end + 1,
730 &rec->sg_content_type);
732 sg_mark_end(sk_msg_elem(msg_pl, i));
735 if (msg_pl->sg.end < msg_pl->sg.start) {
736 sg_chain(&msg_pl->sg.data[msg_pl->sg.start],
737 MAX_SKB_FRAGS - msg_pl->sg.start + 1,
741 i = msg_pl->sg.start;
742 sg_chain(rec->sg_aead_in, 2, &msg_pl->sg.data[i]);
745 sk_msg_iter_var_prev(i);
746 sg_mark_end(sk_msg_elem(msg_en, i));
748 i = msg_en->sg.start;
749 sg_chain(rec->sg_aead_out, 2, &msg_en->sg.data[i]);
751 tls_make_aad(rec->aad_space, msg_pl->sg.size + prot->tail_size,
752 tls_ctx->tx.rec_seq, record_type, prot);
754 tls_fill_prepend(tls_ctx,
755 page_address(sg_page(&msg_en->sg.data[i])) +
756 msg_en->sg.data[i].offset,
757 msg_pl->sg.size + prot->tail_size,
760 tls_ctx->pending_open_record_frags = false;
762 rc = tls_do_encryption(sk, tls_ctx, ctx, req,
763 msg_pl->sg.size + prot->tail_size, i);
765 if (rc != -EINPROGRESS) {
766 tls_err_abort(sk, EBADMSG);
768 tls_ctx->pending_open_record_frags = true;
769 tls_merge_open_record(sk, rec, tmp, orig_end);
772 ctx->async_capable = 1;
775 msg_pl = &tmp->msg_plaintext;
776 msg_en = &tmp->msg_encrypted;
777 sk_msg_trim(sk, msg_en, msg_pl->sg.size + prot->overhead_size);
778 tls_ctx->pending_open_record_frags = true;
782 return tls_tx_records(sk, flags);
785 static int bpf_exec_tx_verdict(struct sk_msg *msg, struct sock *sk,
786 bool full_record, u8 record_type,
787 ssize_t *copied, int flags)
789 struct tls_context *tls_ctx = tls_get_ctx(sk);
790 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
791 struct sk_msg msg_redir = { };
792 struct sk_psock *psock;
793 struct sock *sk_redir;
799 policy = !(flags & MSG_SENDPAGE_NOPOLICY);
800 psock = sk_psock_get(sk);
801 if (!psock || !policy) {
802 err = tls_push_record(sk, flags, record_type);
803 if (err && sk->sk_err == EBADMSG) {
804 *copied -= sk_msg_free(sk, msg);
805 tls_free_open_rec(sk);
809 sk_psock_put(sk, psock);
813 enospc = sk_msg_full(msg);
814 if (psock->eval == __SK_NONE) {
815 delta = msg->sg.size;
816 psock->eval = sk_psock_msg_verdict(sk, psock, msg);
817 delta -= msg->sg.size;
819 if (msg->cork_bytes && msg->cork_bytes > msg->sg.size &&
820 !enospc && !full_record) {
826 if (msg->apply_bytes && msg->apply_bytes < send)
827 send = msg->apply_bytes;
829 switch (psock->eval) {
831 err = tls_push_record(sk, flags, record_type);
832 if (err && sk->sk_err == EBADMSG) {
833 *copied -= sk_msg_free(sk, msg);
834 tls_free_open_rec(sk);
840 sk_redir = psock->sk_redir;
841 memcpy(&msg_redir, msg, sizeof(*msg));
842 if (msg->apply_bytes < send)
843 msg->apply_bytes = 0;
845 msg->apply_bytes -= send;
846 sk_msg_return_zero(sk, msg, send);
847 msg->sg.size -= send;
849 err = tcp_bpf_sendmsg_redir(sk_redir, &msg_redir, send, flags);
852 *copied -= sk_msg_free_nocharge(sk, &msg_redir);
855 if (msg->sg.size == 0)
856 tls_free_open_rec(sk);
860 sk_msg_free_partial(sk, msg, send);
861 if (msg->apply_bytes < send)
862 msg->apply_bytes = 0;
864 msg->apply_bytes -= send;
865 if (msg->sg.size == 0)
866 tls_free_open_rec(sk);
867 *copied -= (send + delta);
872 bool reset_eval = !ctx->open_rec;
876 msg = &rec->msg_plaintext;
877 if (!msg->apply_bytes)
881 psock->eval = __SK_NONE;
882 if (psock->sk_redir) {
883 sock_put(psock->sk_redir);
884 psock->sk_redir = NULL;
891 sk_psock_put(sk, psock);
895 static int tls_sw_push_pending_record(struct sock *sk, int flags)
897 struct tls_context *tls_ctx = tls_get_ctx(sk);
898 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
899 struct tls_rec *rec = ctx->open_rec;
900 struct sk_msg *msg_pl;
906 msg_pl = &rec->msg_plaintext;
907 copied = msg_pl->sg.size;
911 return bpf_exec_tx_verdict(msg_pl, sk, true, TLS_RECORD_TYPE_DATA,
915 int tls_sw_sendmsg(struct sock *sk, struct msghdr *msg, size_t size)
917 long timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
918 struct tls_context *tls_ctx = tls_get_ctx(sk);
919 struct tls_prot_info *prot = &tls_ctx->prot_info;
920 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
921 bool async_capable = ctx->async_capable;
922 unsigned char record_type = TLS_RECORD_TYPE_DATA;
923 bool is_kvec = iov_iter_is_kvec(&msg->msg_iter);
924 bool eor = !(msg->msg_flags & MSG_MORE);
927 struct sk_msg *msg_pl, *msg_en;
938 if (msg->msg_flags & ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL |
942 mutex_lock(&tls_ctx->tx_lock);
945 if (unlikely(msg->msg_controllen)) {
946 ret = tls_proccess_cmsg(sk, msg, &record_type);
948 if (ret == -EINPROGRESS)
950 else if (ret != -EAGAIN)
955 while (msg_data_left(msg)) {
964 rec = ctx->open_rec = tls_get_rec(sk);
970 msg_pl = &rec->msg_plaintext;
971 msg_en = &rec->msg_encrypted;
973 orig_size = msg_pl->sg.size;
975 try_to_copy = msg_data_left(msg);
976 record_room = TLS_MAX_PAYLOAD_SIZE - msg_pl->sg.size;
977 if (try_to_copy >= record_room) {
978 try_to_copy = record_room;
982 required_size = msg_pl->sg.size + try_to_copy +
985 if (!sk_stream_memory_free(sk))
986 goto wait_for_sndbuf;
989 ret = tls_alloc_encrypted_msg(sk, required_size);
992 goto wait_for_memory;
994 /* Adjust try_to_copy according to the amount that was
995 * actually allocated. The difference is due
996 * to max sg elements limit
998 try_to_copy -= required_size - msg_en->sg.size;
1002 if (!is_kvec && (full_record || eor) && !async_capable) {
1003 u32 first = msg_pl->sg.end;
1005 ret = sk_msg_zerocopy_from_iter(sk, &msg->msg_iter,
1006 msg_pl, try_to_copy);
1008 goto fallback_to_reg_send;
1011 copied += try_to_copy;
1013 sk_msg_sg_copy_set(msg_pl, first);
1014 ret = bpf_exec_tx_verdict(msg_pl, sk, full_record,
1015 record_type, &copied,
1018 if (ret == -EINPROGRESS)
1020 else if (ret == -ENOMEM)
1021 goto wait_for_memory;
1022 else if (ctx->open_rec && ret == -ENOSPC)
1024 else if (ret != -EAGAIN)
1029 copied -= try_to_copy;
1030 sk_msg_sg_copy_clear(msg_pl, first);
1031 iov_iter_revert(&msg->msg_iter,
1032 msg_pl->sg.size - orig_size);
1033 fallback_to_reg_send:
1034 sk_msg_trim(sk, msg_pl, orig_size);
1037 required_size = msg_pl->sg.size + try_to_copy;
1039 ret = tls_clone_plaintext_msg(sk, required_size);
1044 /* Adjust try_to_copy according to the amount that was
1045 * actually allocated. The difference is due
1046 * to max sg elements limit
1048 try_to_copy -= required_size - msg_pl->sg.size;
1050 sk_msg_trim(sk, msg_en,
1051 msg_pl->sg.size + prot->overhead_size);
1055 ret = sk_msg_memcopy_from_iter(sk, &msg->msg_iter,
1056 msg_pl, try_to_copy);
1061 /* Open records defined only if successfully copied, otherwise
1062 * we would trim the sg but not reset the open record frags.
1064 tls_ctx->pending_open_record_frags = true;
1065 copied += try_to_copy;
1066 if (full_record || eor) {
1067 ret = bpf_exec_tx_verdict(msg_pl, sk, full_record,
1068 record_type, &copied,
1071 if (ret == -EINPROGRESS)
1073 else if (ret == -ENOMEM)
1074 goto wait_for_memory;
1075 else if (ret != -EAGAIN) {
1086 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1088 ret = sk_stream_wait_memory(sk, &timeo);
1092 tls_trim_both_msgs(sk, orig_size);
1096 if (ctx->open_rec && msg_en->sg.size < required_size)
1097 goto alloc_encrypted;
1102 } else if (num_zc) {
1103 /* Wait for pending encryptions to get completed */
1104 spin_lock_bh(&ctx->encrypt_compl_lock);
1105 ctx->async_notify = true;
1107 pending = atomic_read(&ctx->encrypt_pending);
1108 spin_unlock_bh(&ctx->encrypt_compl_lock);
1110 crypto_wait_req(-EINPROGRESS, &ctx->async_wait);
1112 reinit_completion(&ctx->async_wait.completion);
1114 /* There can be no concurrent accesses, since we have no
1115 * pending encrypt operations
1117 WRITE_ONCE(ctx->async_notify, false);
1119 if (ctx->async_wait.err) {
1120 ret = ctx->async_wait.err;
1125 /* Transmit if any encryptions have completed */
1126 if (test_and_clear_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask)) {
1127 cancel_delayed_work(&ctx->tx_work.work);
1128 tls_tx_records(sk, msg->msg_flags);
1132 ret = sk_stream_error(sk, msg->msg_flags, ret);
1135 mutex_unlock(&tls_ctx->tx_lock);
1136 return copied > 0 ? copied : ret;
1139 static int tls_sw_do_sendpage(struct sock *sk, struct page *page,
1140 int offset, size_t size, int flags)
1142 long timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT);
1143 struct tls_context *tls_ctx = tls_get_ctx(sk);
1144 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
1145 struct tls_prot_info *prot = &tls_ctx->prot_info;
1146 unsigned char record_type = TLS_RECORD_TYPE_DATA;
1147 struct sk_msg *msg_pl;
1148 struct tls_rec *rec;
1156 eor = !(flags & MSG_SENDPAGE_NOTLAST);
1157 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
1159 /* Call the sk_stream functions to manage the sndbuf mem. */
1161 size_t copy, required_size;
1169 rec = ctx->open_rec;
1171 rec = ctx->open_rec = tls_get_rec(sk);
1177 msg_pl = &rec->msg_plaintext;
1179 full_record = false;
1180 record_room = TLS_MAX_PAYLOAD_SIZE - msg_pl->sg.size;
1182 if (copy >= record_room) {
1187 required_size = msg_pl->sg.size + copy + prot->overhead_size;
1189 if (!sk_stream_memory_free(sk))
1190 goto wait_for_sndbuf;
1192 ret = tls_alloc_encrypted_msg(sk, required_size);
1195 goto wait_for_memory;
1197 /* Adjust copy according to the amount that was
1198 * actually allocated. The difference is due
1199 * to max sg elements limit
1201 copy -= required_size - msg_pl->sg.size;
1205 sk_msg_page_add(msg_pl, page, copy, offset);
1206 sk_mem_charge(sk, copy);
1212 tls_ctx->pending_open_record_frags = true;
1213 if (full_record || eor || sk_msg_full(msg_pl)) {
1214 ret = bpf_exec_tx_verdict(msg_pl, sk, full_record,
1215 record_type, &copied, flags);
1217 if (ret == -EINPROGRESS)
1219 else if (ret == -ENOMEM)
1220 goto wait_for_memory;
1221 else if (ret != -EAGAIN) {
1230 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1232 ret = sk_stream_wait_memory(sk, &timeo);
1235 tls_trim_both_msgs(sk, msg_pl->sg.size);
1244 /* Transmit if any encryptions have completed */
1245 if (test_and_clear_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask)) {
1246 cancel_delayed_work(&ctx->tx_work.work);
1247 tls_tx_records(sk, flags);
1251 ret = sk_stream_error(sk, flags, ret);
1252 return copied > 0 ? copied : ret;
1255 int tls_sw_sendpage_locked(struct sock *sk, struct page *page,
1256 int offset, size_t size, int flags)
1258 if (flags & ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL |
1259 MSG_SENDPAGE_NOTLAST | MSG_SENDPAGE_NOPOLICY |
1260 MSG_NO_SHARED_FRAGS))
1263 return tls_sw_do_sendpage(sk, page, offset, size, flags);
1266 int tls_sw_sendpage(struct sock *sk, struct page *page,
1267 int offset, size_t size, int flags)
1269 struct tls_context *tls_ctx = tls_get_ctx(sk);
1272 if (flags & ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL |
1273 MSG_SENDPAGE_NOTLAST | MSG_SENDPAGE_NOPOLICY))
1276 mutex_lock(&tls_ctx->tx_lock);
1278 ret = tls_sw_do_sendpage(sk, page, offset, size, flags);
1280 mutex_unlock(&tls_ctx->tx_lock);
1284 static struct sk_buff *tls_wait_data(struct sock *sk, struct sk_psock *psock,
1285 bool nonblock, long timeo, int *err)
1287 struct tls_context *tls_ctx = tls_get_ctx(sk);
1288 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1289 struct sk_buff *skb;
1290 DEFINE_WAIT_FUNC(wait, woken_wake_function);
1292 while (!(skb = ctx->recv_pkt) && sk_psock_queue_empty(psock)) {
1294 *err = sock_error(sk);
1298 if (!skb_queue_empty(&sk->sk_receive_queue)) {
1299 __strp_unpause(&ctx->strp);
1301 return ctx->recv_pkt;
1304 if (sk->sk_shutdown & RCV_SHUTDOWN)
1307 if (sock_flag(sk, SOCK_DONE))
1310 if (nonblock || !timeo) {
1315 add_wait_queue(sk_sleep(sk), &wait);
1316 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
1317 sk_wait_event(sk, &timeo,
1318 ctx->recv_pkt != skb ||
1319 !sk_psock_queue_empty(psock),
1321 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
1322 remove_wait_queue(sk_sleep(sk), &wait);
1324 /* Handle signals */
1325 if (signal_pending(current)) {
1326 *err = sock_intr_errno(timeo);
1334 static int tls_setup_from_iter(struct sock *sk, struct iov_iter *from,
1335 int length, int *pages_used,
1336 unsigned int *size_used,
1337 struct scatterlist *to,
1340 int rc = 0, i = 0, num_elem = *pages_used, maxpages;
1341 struct page *pages[MAX_SKB_FRAGS];
1342 unsigned int size = *size_used;
1343 ssize_t copied, use;
1346 while (length > 0) {
1348 maxpages = to_max_pages - num_elem;
1349 if (maxpages == 0) {
1353 copied = iov_iter_get_pages(from, pages,
1361 iov_iter_advance(from, copied);
1366 use = min_t(int, copied, PAGE_SIZE - offset);
1368 sg_set_page(&to[num_elem],
1369 pages[i], use, offset);
1370 sg_unmark_end(&to[num_elem]);
1371 /* We do not uncharge memory from this API */
1380 /* Mark the end in the last sg entry if newly added */
1381 if (num_elem > *pages_used)
1382 sg_mark_end(&to[num_elem - 1]);
1385 iov_iter_revert(from, size - *size_used);
1387 *pages_used = num_elem;
1392 /* This function decrypts the input skb into either out_iov or in out_sg
1393 * or in skb buffers itself. The input parameter 'zc' indicates if
1394 * zero-copy mode needs to be tried or not. With zero-copy mode, either
1395 * out_iov or out_sg must be non-NULL. In case both out_iov and out_sg are
1396 * NULL, then the decryption happens inside skb buffers itself, i.e.
1397 * zero-copy gets disabled and 'zc' is updated.
1400 static int decrypt_internal(struct sock *sk, struct sk_buff *skb,
1401 struct iov_iter *out_iov,
1402 struct scatterlist *out_sg,
1403 int *chunk, bool *zc, bool async)
1405 struct tls_context *tls_ctx = tls_get_ctx(sk);
1406 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1407 struct tls_prot_info *prot = &tls_ctx->prot_info;
1408 struct strp_msg *rxm = strp_msg(skb);
1409 int n_sgin, n_sgout, nsg, mem_size, aead_size, err, pages = 0;
1410 struct aead_request *aead_req;
1411 struct sk_buff *unused;
1412 u8 *aad, *iv, *mem = NULL;
1413 struct scatterlist *sgin = NULL;
1414 struct scatterlist *sgout = NULL;
1415 const int data_len = rxm->full_len - prot->overhead_size +
1419 if (*zc && (out_iov || out_sg)) {
1421 n_sgout = iov_iter_npages(out_iov, INT_MAX) + 1;
1423 n_sgout = sg_nents(out_sg);
1424 n_sgin = skb_nsg(skb, rxm->offset + prot->prepend_size,
1425 rxm->full_len - prot->prepend_size);
1429 n_sgin = skb_cow_data(skb, 0, &unused);
1435 /* Increment to accommodate AAD */
1436 n_sgin = n_sgin + 1;
1438 nsg = n_sgin + n_sgout;
1440 aead_size = sizeof(*aead_req) + crypto_aead_reqsize(ctx->aead_recv);
1441 mem_size = aead_size + (nsg * sizeof(struct scatterlist));
1442 mem_size = mem_size + prot->aad_size;
1443 mem_size = mem_size + crypto_aead_ivsize(ctx->aead_recv);
1445 /* Allocate a single block of memory which contains
1446 * aead_req || sgin[] || sgout[] || aad || iv.
1447 * This order achieves correct alignment for aead_req, sgin, sgout.
1449 mem = kmalloc(mem_size, sk->sk_allocation);
1453 /* Segment the allocated memory */
1454 aead_req = (struct aead_request *)mem;
1455 sgin = (struct scatterlist *)(mem + aead_size);
1456 sgout = sgin + n_sgin;
1457 aad = (u8 *)(sgout + n_sgout);
1458 iv = aad + prot->aad_size;
1460 /* For CCM based ciphers, first byte of nonce+iv is always '2' */
1461 if (prot->cipher_type == TLS_CIPHER_AES_CCM_128) {
1467 err = skb_copy_bits(skb, rxm->offset + TLS_HEADER_SIZE,
1468 iv + iv_offset + prot->salt_size,
1474 if (prot->version == TLS_1_3_VERSION ||
1475 prot->cipher_type == TLS_CIPHER_CHACHA20_POLY1305)
1476 memcpy(iv + iv_offset, tls_ctx->rx.iv,
1477 crypto_aead_ivsize(ctx->aead_recv));
1479 memcpy(iv + iv_offset, tls_ctx->rx.iv, prot->salt_size);
1481 xor_iv_with_seq(prot, iv, tls_ctx->rx.rec_seq);
1484 tls_make_aad(aad, rxm->full_len - prot->overhead_size +
1486 tls_ctx->rx.rec_seq, ctx->control, prot);
1489 sg_init_table(sgin, n_sgin);
1490 sg_set_buf(&sgin[0], aad, prot->aad_size);
1491 err = skb_to_sgvec(skb, &sgin[1],
1492 rxm->offset + prot->prepend_size,
1493 rxm->full_len - prot->prepend_size);
1501 sg_init_table(sgout, n_sgout);
1502 sg_set_buf(&sgout[0], aad, prot->aad_size);
1505 err = tls_setup_from_iter(sk, out_iov, data_len,
1506 &pages, chunk, &sgout[1],
1509 goto fallback_to_reg_recv;
1510 } else if (out_sg) {
1511 memcpy(sgout, out_sg, n_sgout * sizeof(*sgout));
1513 goto fallback_to_reg_recv;
1516 fallback_to_reg_recv:
1523 /* Prepare and submit AEAD request */
1524 err = tls_do_decryption(sk, skb, sgin, sgout, iv,
1525 data_len, aead_req, async);
1526 if (err == -EINPROGRESS)
1529 /* Release the pages in case iov was mapped to pages */
1530 for (; pages > 0; pages--)
1531 put_page(sg_page(&sgout[pages]));
1537 static int decrypt_skb_update(struct sock *sk, struct sk_buff *skb,
1538 struct iov_iter *dest, int *chunk, bool *zc,
1541 struct tls_context *tls_ctx = tls_get_ctx(sk);
1542 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1543 struct tls_prot_info *prot = &tls_ctx->prot_info;
1544 struct strp_msg *rxm = strp_msg(skb);
1547 if (!ctx->decrypted) {
1548 if (tls_ctx->rx_conf == TLS_HW) {
1549 err = tls_device_decrypted(sk, tls_ctx, skb, rxm);
1554 /* Still not decrypted after tls_device */
1555 if (!ctx->decrypted) {
1556 err = decrypt_internal(sk, skb, dest, NULL, chunk, zc,
1559 if (err == -EINPROGRESS)
1560 tls_advance_record_sn(sk, prot,
1562 else if (err == -EBADMSG)
1563 TLS_INC_STATS(sock_net(sk),
1564 LINUX_MIB_TLSDECRYPTERROR);
1571 pad = padding_length(ctx, prot, skb);
1575 rxm->full_len -= pad;
1576 rxm->offset += prot->prepend_size;
1577 rxm->full_len -= prot->overhead_size;
1578 tls_advance_record_sn(sk, prot, &tls_ctx->rx);
1580 ctx->saved_data_ready(sk);
1588 int decrypt_skb(struct sock *sk, struct sk_buff *skb,
1589 struct scatterlist *sgout)
1594 return decrypt_internal(sk, skb, NULL, sgout, &chunk, &zc, false);
1597 static bool tls_sw_advance_skb(struct sock *sk, struct sk_buff *skb,
1600 struct tls_context *tls_ctx = tls_get_ctx(sk);
1601 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1604 struct strp_msg *rxm = strp_msg(skb);
1606 if (len < rxm->full_len) {
1608 rxm->full_len -= len;
1614 /* Finished with message */
1615 ctx->recv_pkt = NULL;
1616 __strp_unpause(&ctx->strp);
1621 /* This function traverses the rx_list in tls receive context to copies the
1622 * decrypted records into the buffer provided by caller zero copy is not
1623 * true. Further, the records are removed from the rx_list if it is not a peek
1624 * case and the record has been consumed completely.
1626 static int process_rx_list(struct tls_sw_context_rx *ctx,
1635 struct sk_buff *skb = skb_peek(&ctx->rx_list);
1638 struct tls_msg *tlm;
1641 /* Set the record type in 'control' if caller didn't pass it */
1644 ctrl = tlm->control;
1647 while (skip && skb) {
1648 struct strp_msg *rxm = strp_msg(skb);
1651 /* Cannot process a record of different type */
1652 if (ctrl != tlm->control)
1655 if (skip < rxm->full_len)
1658 skip = skip - rxm->full_len;
1659 skb = skb_peek_next(skb, &ctx->rx_list);
1662 while (len && skb) {
1663 struct sk_buff *next_skb;
1664 struct strp_msg *rxm = strp_msg(skb);
1665 int chunk = min_t(unsigned int, rxm->full_len - skip, len);
1669 /* Cannot process a record of different type */
1670 if (ctrl != tlm->control)
1673 /* Set record type if not already done. For a non-data record,
1674 * do not proceed if record type could not be copied.
1677 int cerr = put_cmsg(msg, SOL_TLS, TLS_GET_RECORD_TYPE,
1678 sizeof(ctrl), &ctrl);
1680 if (ctrl != TLS_RECORD_TYPE_DATA) {
1681 if (cerr || msg->msg_flags & MSG_CTRUNC)
1688 if (!zc || (rxm->full_len - skip) > len) {
1689 int err = skb_copy_datagram_msg(skb, rxm->offset + skip,
1696 copied = copied + chunk;
1698 /* Consume the data from record if it is non-peek case*/
1700 rxm->offset = rxm->offset + chunk;
1701 rxm->full_len = rxm->full_len - chunk;
1703 /* Return if there is unconsumed data in the record */
1704 if (rxm->full_len - skip)
1708 /* The remaining skip-bytes must lie in 1st record in rx_list.
1709 * So from the 2nd record, 'skip' should be 0.
1714 msg->msg_flags |= MSG_EOR;
1716 next_skb = skb_peek_next(skb, &ctx->rx_list);
1719 skb_unlink(skb, &ctx->rx_list);
1730 int tls_sw_recvmsg(struct sock *sk,
1737 struct tls_context *tls_ctx = tls_get_ctx(sk);
1738 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1739 struct tls_prot_info *prot = &tls_ctx->prot_info;
1740 struct sk_psock *psock;
1741 unsigned char control = 0;
1742 ssize_t decrypted = 0;
1743 struct strp_msg *rxm;
1744 struct tls_msg *tlm;
1745 struct sk_buff *skb;
1748 int target, err = 0;
1750 bool is_kvec = iov_iter_is_kvec(&msg->msg_iter);
1751 bool is_peek = flags & MSG_PEEK;
1752 bool bpf_strp_enabled;
1758 if (unlikely(flags & MSG_ERRQUEUE))
1759 return sock_recv_errqueue(sk, msg, len, SOL_IP, IP_RECVERR);
1761 psock = sk_psock_get(sk);
1763 bpf_strp_enabled = sk_psock_strp_enabled(psock);
1765 /* Process pending decrypted records. It must be non-zero-copy */
1766 err = process_rx_list(ctx, msg, &control, &cmsg, 0, len, false,
1769 tls_err_abort(sk, err);
1778 target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
1780 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1782 while (len && (decrypted + copied < target || ctx->recv_pkt)) {
1783 bool retain_skb = false;
1790 skb = tls_wait_data(sk, psock, flags & MSG_DONTWAIT, timeo, &err);
1793 int ret = sk_msg_recvmsg(sk, psock, msg, len,
1805 if (prot->version == TLS_1_3_VERSION)
1808 tlm->control = ctx->control;
1811 rxm = strp_msg(skb);
1813 to_decrypt = rxm->full_len - prot->overhead_size;
1815 if (to_decrypt <= len && !is_kvec && !is_peek &&
1816 ctx->control == TLS_RECORD_TYPE_DATA &&
1817 prot->version != TLS_1_3_VERSION &&
1821 /* Do not use async mode if record is non-data */
1822 if (ctx->control == TLS_RECORD_TYPE_DATA && !bpf_strp_enabled)
1823 async_capable = ctx->async_capable;
1825 async_capable = false;
1827 err = decrypt_skb_update(sk, skb, &msg->msg_iter,
1828 &chunk, &zc, async_capable);
1829 if (err < 0 && err != -EINPROGRESS) {
1830 tls_err_abort(sk, EBADMSG);
1834 if (err == -EINPROGRESS) {
1837 } else if (prot->version == TLS_1_3_VERSION) {
1838 tlm->control = ctx->control;
1841 /* If the type of records being processed is not known yet,
1842 * set it to record type just dequeued. If it is already known,
1843 * but does not match the record type just dequeued, go to end.
1844 * We always get record type here since for tls1.2, record type
1845 * is known just after record is dequeued from stream parser.
1846 * For tls1.3, we disable async.
1850 control = tlm->control;
1851 else if (control != tlm->control)
1857 cerr = put_cmsg(msg, SOL_TLS, TLS_GET_RECORD_TYPE,
1858 sizeof(control), &control);
1860 if (control != TLS_RECORD_TYPE_DATA) {
1861 if (cerr || msg->msg_flags & MSG_CTRUNC) {
1869 goto pick_next_record;
1872 if (bpf_strp_enabled) {
1873 err = sk_psock_tls_strp_read(psock, skb);
1874 if (err != __SK_PASS) {
1875 rxm->offset = rxm->offset + rxm->full_len;
1877 if (err == __SK_DROP)
1879 ctx->recv_pkt = NULL;
1880 __strp_unpause(&ctx->strp);
1885 if (rxm->full_len > len) {
1889 chunk = rxm->full_len;
1892 err = skb_copy_datagram_msg(skb, rxm->offset,
1898 rxm->offset = rxm->offset + chunk;
1899 rxm->full_len = rxm->full_len - chunk;
1910 /* For async or peek case, queue the current skb */
1911 if (async || is_peek || retain_skb) {
1912 skb_queue_tail(&ctx->rx_list, skb);
1916 if (tls_sw_advance_skb(sk, skb, chunk)) {
1917 /* Return full control message to
1918 * userspace before trying to parse
1919 * another message type
1921 msg->msg_flags |= MSG_EOR;
1922 if (control != TLS_RECORD_TYPE_DATA)
1931 /* Wait for all previously submitted records to be decrypted */
1932 spin_lock_bh(&ctx->decrypt_compl_lock);
1933 ctx->async_notify = true;
1934 pending = atomic_read(&ctx->decrypt_pending);
1935 spin_unlock_bh(&ctx->decrypt_compl_lock);
1937 err = crypto_wait_req(-EINPROGRESS, &ctx->async_wait);
1939 /* one of async decrypt failed */
1940 tls_err_abort(sk, err);
1946 reinit_completion(&ctx->async_wait.completion);
1949 /* There can be no concurrent accesses, since we have no
1950 * pending decrypt operations
1952 WRITE_ONCE(ctx->async_notify, false);
1954 /* Drain records from the rx_list & copy if required */
1955 if (is_peek || is_kvec)
1956 err = process_rx_list(ctx, msg, &control, &cmsg, copied,
1957 decrypted, false, is_peek);
1959 err = process_rx_list(ctx, msg, &control, &cmsg, 0,
1960 decrypted, true, is_peek);
1962 tls_err_abort(sk, err);
1968 copied += decrypted;
1973 sk_psock_put(sk, psock);
1974 return copied ? : err;
1977 ssize_t tls_sw_splice_read(struct socket *sock, loff_t *ppos,
1978 struct pipe_inode_info *pipe,
1979 size_t len, unsigned int flags)
1981 struct tls_context *tls_ctx = tls_get_ctx(sock->sk);
1982 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1983 struct strp_msg *rxm = NULL;
1984 struct sock *sk = sock->sk;
1985 struct sk_buff *skb;
1994 timeo = sock_rcvtimeo(sk, flags & SPLICE_F_NONBLOCK);
1996 skb = tls_wait_data(sk, NULL, flags & SPLICE_F_NONBLOCK, timeo, &err);
1998 goto splice_read_end;
2000 if (!ctx->decrypted) {
2001 err = decrypt_skb_update(sk, skb, NULL, &chunk, &zc, false);
2003 /* splice does not support reading control messages */
2004 if (ctx->control != TLS_RECORD_TYPE_DATA) {
2006 goto splice_read_end;
2010 tls_err_abort(sk, EBADMSG);
2011 goto splice_read_end;
2015 rxm = strp_msg(skb);
2017 chunk = min_t(unsigned int, rxm->full_len, len);
2018 copied = skb_splice_bits(skb, sk, rxm->offset, pipe, chunk, flags);
2020 goto splice_read_end;
2022 tls_sw_advance_skb(sk, skb, copied);
2026 return copied ? : err;
2029 bool tls_sw_sock_is_readable(struct sock *sk)
2031 struct tls_context *tls_ctx = tls_get_ctx(sk);
2032 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2033 bool ingress_empty = true;
2034 struct sk_psock *psock;
2037 psock = sk_psock(sk);
2039 ingress_empty = list_empty(&psock->ingress_msg);
2042 return !ingress_empty || ctx->recv_pkt ||
2043 !skb_queue_empty(&ctx->rx_list);
2046 static int tls_read_size(struct strparser *strp, struct sk_buff *skb)
2048 struct tls_context *tls_ctx = tls_get_ctx(strp->sk);
2049 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2050 struct tls_prot_info *prot = &tls_ctx->prot_info;
2051 char header[TLS_HEADER_SIZE + MAX_IV_SIZE];
2052 struct strp_msg *rxm = strp_msg(skb);
2053 size_t cipher_overhead;
2054 size_t data_len = 0;
2057 /* Verify that we have a full TLS header, or wait for more data */
2058 if (rxm->offset + prot->prepend_size > skb->len)
2061 /* Sanity-check size of on-stack buffer. */
2062 if (WARN_ON(prot->prepend_size > sizeof(header))) {
2067 /* Linearize header to local buffer */
2068 ret = skb_copy_bits(skb, rxm->offset, header, prot->prepend_size);
2073 ctx->control = header[0];
2075 data_len = ((header[4] & 0xFF) | (header[3] << 8));
2077 cipher_overhead = prot->tag_size;
2078 if (prot->version != TLS_1_3_VERSION &&
2079 prot->cipher_type != TLS_CIPHER_CHACHA20_POLY1305)
2080 cipher_overhead += prot->iv_size;
2082 if (data_len > TLS_MAX_PAYLOAD_SIZE + cipher_overhead +
2087 if (data_len < cipher_overhead) {
2092 /* Note that both TLS1.3 and TLS1.2 use TLS_1_2 version here */
2093 if (header[1] != TLS_1_2_VERSION_MINOR ||
2094 header[2] != TLS_1_2_VERSION_MAJOR) {
2099 tls_device_rx_resync_new_rec(strp->sk, data_len + TLS_HEADER_SIZE,
2100 TCP_SKB_CB(skb)->seq + rxm->offset);
2101 return data_len + TLS_HEADER_SIZE;
2104 tls_err_abort(strp->sk, ret);
2109 static void tls_queue(struct strparser *strp, struct sk_buff *skb)
2111 struct tls_context *tls_ctx = tls_get_ctx(strp->sk);
2112 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2116 ctx->recv_pkt = skb;
2119 ctx->saved_data_ready(strp->sk);
2122 static void tls_data_ready(struct sock *sk)
2124 struct tls_context *tls_ctx = tls_get_ctx(sk);
2125 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2126 struct sk_psock *psock;
2128 strp_data_ready(&ctx->strp);
2130 psock = sk_psock_get(sk);
2132 if (!list_empty(&psock->ingress_msg))
2133 ctx->saved_data_ready(sk);
2134 sk_psock_put(sk, psock);
2138 void tls_sw_cancel_work_tx(struct tls_context *tls_ctx)
2140 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
2142 set_bit(BIT_TX_CLOSING, &ctx->tx_bitmask);
2143 set_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask);
2144 cancel_delayed_work_sync(&ctx->tx_work.work);
2147 void tls_sw_release_resources_tx(struct sock *sk)
2149 struct tls_context *tls_ctx = tls_get_ctx(sk);
2150 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
2151 struct tls_rec *rec, *tmp;
2154 /* Wait for any pending async encryptions to complete */
2155 spin_lock_bh(&ctx->encrypt_compl_lock);
2156 ctx->async_notify = true;
2157 pending = atomic_read(&ctx->encrypt_pending);
2158 spin_unlock_bh(&ctx->encrypt_compl_lock);
2161 crypto_wait_req(-EINPROGRESS, &ctx->async_wait);
2163 tls_tx_records(sk, -1);
2165 /* Free up un-sent records in tx_list. First, free
2166 * the partially sent record if any at head of tx_list.
2168 if (tls_ctx->partially_sent_record) {
2169 tls_free_partial_record(sk, tls_ctx);
2170 rec = list_first_entry(&ctx->tx_list,
2171 struct tls_rec, list);
2172 list_del(&rec->list);
2173 sk_msg_free(sk, &rec->msg_plaintext);
2177 list_for_each_entry_safe(rec, tmp, &ctx->tx_list, list) {
2178 list_del(&rec->list);
2179 sk_msg_free(sk, &rec->msg_encrypted);
2180 sk_msg_free(sk, &rec->msg_plaintext);
2184 crypto_free_aead(ctx->aead_send);
2185 tls_free_open_rec(sk);
2188 void tls_sw_free_ctx_tx(struct tls_context *tls_ctx)
2190 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
2195 void tls_sw_release_resources_rx(struct sock *sk)
2197 struct tls_context *tls_ctx = tls_get_ctx(sk);
2198 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2200 kfree(tls_ctx->rx.rec_seq);
2201 kfree(tls_ctx->rx.iv);
2203 if (ctx->aead_recv) {
2204 kfree_skb(ctx->recv_pkt);
2205 ctx->recv_pkt = NULL;
2206 skb_queue_purge(&ctx->rx_list);
2207 crypto_free_aead(ctx->aead_recv);
2208 strp_stop(&ctx->strp);
2209 /* If tls_sw_strparser_arm() was not called (cleanup paths)
2210 * we still want to strp_stop(), but sk->sk_data_ready was
2213 if (ctx->saved_data_ready) {
2214 write_lock_bh(&sk->sk_callback_lock);
2215 sk->sk_data_ready = ctx->saved_data_ready;
2216 write_unlock_bh(&sk->sk_callback_lock);
2221 void tls_sw_strparser_done(struct tls_context *tls_ctx)
2223 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2225 strp_done(&ctx->strp);
2228 void tls_sw_free_ctx_rx(struct tls_context *tls_ctx)
2230 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2235 void tls_sw_free_resources_rx(struct sock *sk)
2237 struct tls_context *tls_ctx = tls_get_ctx(sk);
2239 tls_sw_release_resources_rx(sk);
2240 tls_sw_free_ctx_rx(tls_ctx);
2243 /* The work handler to transmitt the encrypted records in tx_list */
2244 static void tx_work_handler(struct work_struct *work)
2246 struct delayed_work *delayed_work = to_delayed_work(work);
2247 struct tx_work *tx_work = container_of(delayed_work,
2248 struct tx_work, work);
2249 struct sock *sk = tx_work->sk;
2250 struct tls_context *tls_ctx = tls_get_ctx(sk);
2251 struct tls_sw_context_tx *ctx;
2253 if (unlikely(!tls_ctx))
2256 ctx = tls_sw_ctx_tx(tls_ctx);
2257 if (test_bit(BIT_TX_CLOSING, &ctx->tx_bitmask))
2260 if (!test_and_clear_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask))
2262 mutex_lock(&tls_ctx->tx_lock);
2264 tls_tx_records(sk, -1);
2266 mutex_unlock(&tls_ctx->tx_lock);
2269 void tls_sw_write_space(struct sock *sk, struct tls_context *ctx)
2271 struct tls_sw_context_tx *tx_ctx = tls_sw_ctx_tx(ctx);
2273 /* Schedule the transmission if tx list is ready */
2274 if (is_tx_ready(tx_ctx) &&
2275 !test_and_set_bit(BIT_TX_SCHEDULED, &tx_ctx->tx_bitmask))
2276 schedule_delayed_work(&tx_ctx->tx_work.work, 0);
2279 void tls_sw_strparser_arm(struct sock *sk, struct tls_context *tls_ctx)
2281 struct tls_sw_context_rx *rx_ctx = tls_sw_ctx_rx(tls_ctx);
2283 write_lock_bh(&sk->sk_callback_lock);
2284 rx_ctx->saved_data_ready = sk->sk_data_ready;
2285 sk->sk_data_ready = tls_data_ready;
2286 write_unlock_bh(&sk->sk_callback_lock);
2288 strp_check_rcv(&rx_ctx->strp);
2291 int tls_set_sw_offload(struct sock *sk, struct tls_context *ctx, int tx)
2293 struct tls_context *tls_ctx = tls_get_ctx(sk);
2294 struct tls_prot_info *prot = &tls_ctx->prot_info;
2295 struct tls_crypto_info *crypto_info;
2296 struct tls12_crypto_info_aes_gcm_128 *gcm_128_info;
2297 struct tls12_crypto_info_aes_gcm_256 *gcm_256_info;
2298 struct tls12_crypto_info_aes_ccm_128 *ccm_128_info;
2299 struct tls12_crypto_info_chacha20_poly1305 *chacha20_poly1305_info;
2300 struct tls_sw_context_tx *sw_ctx_tx = NULL;
2301 struct tls_sw_context_rx *sw_ctx_rx = NULL;
2302 struct cipher_context *cctx;
2303 struct crypto_aead **aead;
2304 struct strp_callbacks cb;
2305 u16 nonce_size, tag_size, iv_size, rec_seq_size, salt_size;
2306 struct crypto_tfm *tfm;
2307 char *iv, *rec_seq, *key, *salt, *cipher_name;
2317 if (!ctx->priv_ctx_tx) {
2318 sw_ctx_tx = kzalloc(sizeof(*sw_ctx_tx), GFP_KERNEL);
2323 ctx->priv_ctx_tx = sw_ctx_tx;
2326 (struct tls_sw_context_tx *)ctx->priv_ctx_tx;
2329 if (!ctx->priv_ctx_rx) {
2330 sw_ctx_rx = kzalloc(sizeof(*sw_ctx_rx), GFP_KERNEL);
2335 ctx->priv_ctx_rx = sw_ctx_rx;
2338 (struct tls_sw_context_rx *)ctx->priv_ctx_rx;
2343 crypto_init_wait(&sw_ctx_tx->async_wait);
2344 spin_lock_init(&sw_ctx_tx->encrypt_compl_lock);
2345 crypto_info = &ctx->crypto_send.info;
2347 aead = &sw_ctx_tx->aead_send;
2348 INIT_LIST_HEAD(&sw_ctx_tx->tx_list);
2349 INIT_DELAYED_WORK(&sw_ctx_tx->tx_work.work, tx_work_handler);
2350 sw_ctx_tx->tx_work.sk = sk;
2352 crypto_init_wait(&sw_ctx_rx->async_wait);
2353 spin_lock_init(&sw_ctx_rx->decrypt_compl_lock);
2354 crypto_info = &ctx->crypto_recv.info;
2356 skb_queue_head_init(&sw_ctx_rx->rx_list);
2357 aead = &sw_ctx_rx->aead_recv;
2360 switch (crypto_info->cipher_type) {
2361 case TLS_CIPHER_AES_GCM_128: {
2362 nonce_size = TLS_CIPHER_AES_GCM_128_IV_SIZE;
2363 tag_size = TLS_CIPHER_AES_GCM_128_TAG_SIZE;
2364 iv_size = TLS_CIPHER_AES_GCM_128_IV_SIZE;
2365 iv = ((struct tls12_crypto_info_aes_gcm_128 *)crypto_info)->iv;
2366 rec_seq_size = TLS_CIPHER_AES_GCM_128_REC_SEQ_SIZE;
2368 ((struct tls12_crypto_info_aes_gcm_128 *)crypto_info)->rec_seq;
2370 (struct tls12_crypto_info_aes_gcm_128 *)crypto_info;
2371 keysize = TLS_CIPHER_AES_GCM_128_KEY_SIZE;
2372 key = gcm_128_info->key;
2373 salt = gcm_128_info->salt;
2374 salt_size = TLS_CIPHER_AES_GCM_128_SALT_SIZE;
2375 cipher_name = "gcm(aes)";
2378 case TLS_CIPHER_AES_GCM_256: {
2379 nonce_size = TLS_CIPHER_AES_GCM_256_IV_SIZE;
2380 tag_size = TLS_CIPHER_AES_GCM_256_TAG_SIZE;
2381 iv_size = TLS_CIPHER_AES_GCM_256_IV_SIZE;
2382 iv = ((struct tls12_crypto_info_aes_gcm_256 *)crypto_info)->iv;
2383 rec_seq_size = TLS_CIPHER_AES_GCM_256_REC_SEQ_SIZE;
2385 ((struct tls12_crypto_info_aes_gcm_256 *)crypto_info)->rec_seq;
2387 (struct tls12_crypto_info_aes_gcm_256 *)crypto_info;
2388 keysize = TLS_CIPHER_AES_GCM_256_KEY_SIZE;
2389 key = gcm_256_info->key;
2390 salt = gcm_256_info->salt;
2391 salt_size = TLS_CIPHER_AES_GCM_256_SALT_SIZE;
2392 cipher_name = "gcm(aes)";
2395 case TLS_CIPHER_AES_CCM_128: {
2396 nonce_size = TLS_CIPHER_AES_CCM_128_IV_SIZE;
2397 tag_size = TLS_CIPHER_AES_CCM_128_TAG_SIZE;
2398 iv_size = TLS_CIPHER_AES_CCM_128_IV_SIZE;
2399 iv = ((struct tls12_crypto_info_aes_ccm_128 *)crypto_info)->iv;
2400 rec_seq_size = TLS_CIPHER_AES_CCM_128_REC_SEQ_SIZE;
2402 ((struct tls12_crypto_info_aes_ccm_128 *)crypto_info)->rec_seq;
2404 (struct tls12_crypto_info_aes_ccm_128 *)crypto_info;
2405 keysize = TLS_CIPHER_AES_CCM_128_KEY_SIZE;
2406 key = ccm_128_info->key;
2407 salt = ccm_128_info->salt;
2408 salt_size = TLS_CIPHER_AES_CCM_128_SALT_SIZE;
2409 cipher_name = "ccm(aes)";
2412 case TLS_CIPHER_CHACHA20_POLY1305: {
2413 chacha20_poly1305_info = (void *)crypto_info;
2415 tag_size = TLS_CIPHER_CHACHA20_POLY1305_TAG_SIZE;
2416 iv_size = TLS_CIPHER_CHACHA20_POLY1305_IV_SIZE;
2417 iv = chacha20_poly1305_info->iv;
2418 rec_seq_size = TLS_CIPHER_CHACHA20_POLY1305_REC_SEQ_SIZE;
2419 rec_seq = chacha20_poly1305_info->rec_seq;
2420 keysize = TLS_CIPHER_CHACHA20_POLY1305_KEY_SIZE;
2421 key = chacha20_poly1305_info->key;
2422 salt = chacha20_poly1305_info->salt;
2423 salt_size = TLS_CIPHER_CHACHA20_POLY1305_SALT_SIZE;
2424 cipher_name = "rfc7539(chacha20,poly1305)";
2432 /* Sanity-check the sizes for stack allocations. */
2433 if (iv_size > MAX_IV_SIZE || nonce_size > MAX_IV_SIZE ||
2434 rec_seq_size > TLS_MAX_REC_SEQ_SIZE) {
2439 if (crypto_info->version == TLS_1_3_VERSION) {
2441 prot->aad_size = TLS_HEADER_SIZE;
2442 prot->tail_size = 1;
2444 prot->aad_size = TLS_AAD_SPACE_SIZE;
2445 prot->tail_size = 0;
2448 prot->version = crypto_info->version;
2449 prot->cipher_type = crypto_info->cipher_type;
2450 prot->prepend_size = TLS_HEADER_SIZE + nonce_size;
2451 prot->tag_size = tag_size;
2452 prot->overhead_size = prot->prepend_size +
2453 prot->tag_size + prot->tail_size;
2454 prot->iv_size = iv_size;
2455 prot->salt_size = salt_size;
2456 cctx->iv = kmalloc(iv_size + salt_size, GFP_KERNEL);
2461 /* Note: 128 & 256 bit salt are the same size */
2462 prot->rec_seq_size = rec_seq_size;
2463 memcpy(cctx->iv, salt, salt_size);
2464 memcpy(cctx->iv + salt_size, iv, iv_size);
2465 cctx->rec_seq = kmemdup(rec_seq, rec_seq_size, GFP_KERNEL);
2466 if (!cctx->rec_seq) {
2472 *aead = crypto_alloc_aead(cipher_name, 0, 0);
2473 if (IS_ERR(*aead)) {
2474 rc = PTR_ERR(*aead);
2480 ctx->push_pending_record = tls_sw_push_pending_record;
2482 rc = crypto_aead_setkey(*aead, key, keysize);
2487 rc = crypto_aead_setauthsize(*aead, prot->tag_size);
2492 tfm = crypto_aead_tfm(sw_ctx_rx->aead_recv);
2494 if (crypto_info->version == TLS_1_3_VERSION)
2495 sw_ctx_rx->async_capable = 0;
2497 sw_ctx_rx->async_capable =
2498 !!(tfm->__crt_alg->cra_flags &
2501 /* Set up strparser */
2502 memset(&cb, 0, sizeof(cb));
2503 cb.rcv_msg = tls_queue;
2504 cb.parse_msg = tls_read_size;
2506 strp_init(&sw_ctx_rx->strp, sk, &cb);
2512 crypto_free_aead(*aead);
2515 kfree(cctx->rec_seq);
2516 cctx->rec_seq = NULL;
2522 kfree(ctx->priv_ctx_tx);
2523 ctx->priv_ctx_tx = NULL;
2525 kfree(ctx->priv_ctx_rx);
2526 ctx->priv_ctx_rx = NULL;