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/bug.h>
39 #include <linux/sched/signal.h>
40 #include <linux/module.h>
41 #include <linux/splice.h>
42 #include <crypto/aead.h>
44 #include <net/strparser.h>
47 struct tls_decrypt_arg {
52 noinline void tls_err_abort(struct sock *sk, int err)
54 WARN_ON_ONCE(err >= 0);
55 /* sk->sk_err should contain a positive error code. */
60 static int __skb_nsg(struct sk_buff *skb, int offset, int len,
61 unsigned int recursion_level)
63 int start = skb_headlen(skb);
64 int i, chunk = start - offset;
65 struct sk_buff *frag_iter;
68 if (unlikely(recursion_level >= 24))
81 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
84 WARN_ON(start > offset + len);
86 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
100 if (unlikely(skb_has_frag_list(skb))) {
101 skb_walk_frags(skb, frag_iter) {
104 WARN_ON(start > offset + len);
106 end = start + frag_iter->len;
107 chunk = end - offset;
111 ret = __skb_nsg(frag_iter, offset - start, chunk,
112 recursion_level + 1);
113 if (unlikely(ret < 0))
128 /* Return the number of scatterlist elements required to completely map the
129 * skb, or -EMSGSIZE if the recursion depth is exceeded.
131 static int skb_nsg(struct sk_buff *skb, int offset, int len)
133 return __skb_nsg(skb, offset, len, 0);
136 static int padding_length(struct tls_prot_info *prot, struct sk_buff *skb)
138 struct strp_msg *rxm = strp_msg(skb);
139 struct tls_msg *tlm = tls_msg(skb);
142 /* Determine zero-padding length */
143 if (prot->version == TLS_1_3_VERSION) {
144 int offset = rxm->full_len - TLS_TAG_SIZE - 1;
145 char content_type = 0;
148 while (content_type == 0) {
149 if (offset < prot->prepend_size)
151 err = skb_copy_bits(skb, rxm->offset + offset,
160 tlm->control = content_type;
165 static void tls_decrypt_done(struct crypto_async_request *req, int err)
167 struct aead_request *aead_req = (struct aead_request *)req;
168 struct scatterlist *sgout = aead_req->dst;
169 struct scatterlist *sgin = aead_req->src;
170 struct tls_sw_context_rx *ctx;
171 struct tls_context *tls_ctx;
172 struct tls_prot_info *prot;
173 struct scatterlist *sg;
177 skb = (struct sk_buff *)req->data;
178 tls_ctx = tls_get_ctx(skb->sk);
179 ctx = tls_sw_ctx_rx(tls_ctx);
180 prot = &tls_ctx->prot_info;
182 /* Propagate if there was an err */
185 TLS_INC_STATS(sock_net(skb->sk),
186 LINUX_MIB_TLSDECRYPTERROR);
187 ctx->async_wait.err = err;
188 tls_err_abort(skb->sk, err);
190 struct strp_msg *rxm = strp_msg(skb);
192 /* No TLS 1.3 support with async crypto */
193 WARN_ON(prot->tail_size);
195 rxm->offset += prot->prepend_size;
196 rxm->full_len -= prot->overhead_size;
199 /* After using skb->sk to propagate sk through crypto async callback
200 * we need to NULL it again.
205 /* Free the destination pages if skb was not decrypted inplace */
207 /* Skip the first S/G entry as it points to AAD */
208 for_each_sg(sg_next(sgout), sg, UINT_MAX, pages) {
211 put_page(sg_page(sg));
217 spin_lock_bh(&ctx->decrypt_compl_lock);
218 if (!atomic_dec_return(&ctx->decrypt_pending))
219 complete(&ctx->async_wait.completion);
220 spin_unlock_bh(&ctx->decrypt_compl_lock);
223 static int tls_do_decryption(struct sock *sk,
225 struct scatterlist *sgin,
226 struct scatterlist *sgout,
229 struct aead_request *aead_req,
230 struct tls_decrypt_arg *darg)
232 struct tls_context *tls_ctx = tls_get_ctx(sk);
233 struct tls_prot_info *prot = &tls_ctx->prot_info;
234 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
237 aead_request_set_tfm(aead_req, ctx->aead_recv);
238 aead_request_set_ad(aead_req, prot->aad_size);
239 aead_request_set_crypt(aead_req, sgin, sgout,
240 data_len + prot->tag_size,
244 /* Using skb->sk to push sk through to crypto async callback
245 * handler. This allows propagating errors up to the socket
246 * if needed. It _must_ be cleared in the async handler
247 * before consume_skb is called. We _know_ skb->sk is NULL
248 * because it is a clone from strparser.
251 aead_request_set_callback(aead_req,
252 CRYPTO_TFM_REQ_MAY_BACKLOG,
253 tls_decrypt_done, skb);
254 atomic_inc(&ctx->decrypt_pending);
256 aead_request_set_callback(aead_req,
257 CRYPTO_TFM_REQ_MAY_BACKLOG,
258 crypto_req_done, &ctx->async_wait);
261 ret = crypto_aead_decrypt(aead_req);
262 if (ret == -EINPROGRESS) {
266 ret = crypto_wait_req(ret, &ctx->async_wait);
271 TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSDECRYPTERROR);
276 static void tls_trim_both_msgs(struct sock *sk, int target_size)
278 struct tls_context *tls_ctx = tls_get_ctx(sk);
279 struct tls_prot_info *prot = &tls_ctx->prot_info;
280 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
281 struct tls_rec *rec = ctx->open_rec;
283 sk_msg_trim(sk, &rec->msg_plaintext, target_size);
285 target_size += prot->overhead_size;
286 sk_msg_trim(sk, &rec->msg_encrypted, target_size);
289 static int tls_alloc_encrypted_msg(struct sock *sk, int len)
291 struct tls_context *tls_ctx = tls_get_ctx(sk);
292 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
293 struct tls_rec *rec = ctx->open_rec;
294 struct sk_msg *msg_en = &rec->msg_encrypted;
296 return sk_msg_alloc(sk, msg_en, len, 0);
299 static int tls_clone_plaintext_msg(struct sock *sk, int required)
301 struct tls_context *tls_ctx = tls_get_ctx(sk);
302 struct tls_prot_info *prot = &tls_ctx->prot_info;
303 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
304 struct tls_rec *rec = ctx->open_rec;
305 struct sk_msg *msg_pl = &rec->msg_plaintext;
306 struct sk_msg *msg_en = &rec->msg_encrypted;
309 /* We add page references worth len bytes from encrypted sg
310 * at the end of plaintext sg. It is guaranteed that msg_en
311 * has enough required room (ensured by caller).
313 len = required - msg_pl->sg.size;
315 /* Skip initial bytes in msg_en's data to be able to use
316 * same offset of both plain and encrypted data.
318 skip = prot->prepend_size + msg_pl->sg.size;
320 return sk_msg_clone(sk, msg_pl, msg_en, skip, len);
323 static struct tls_rec *tls_get_rec(struct sock *sk)
325 struct tls_context *tls_ctx = tls_get_ctx(sk);
326 struct tls_prot_info *prot = &tls_ctx->prot_info;
327 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
328 struct sk_msg *msg_pl, *msg_en;
332 mem_size = sizeof(struct tls_rec) + crypto_aead_reqsize(ctx->aead_send);
334 rec = kzalloc(mem_size, sk->sk_allocation);
338 msg_pl = &rec->msg_plaintext;
339 msg_en = &rec->msg_encrypted;
344 sg_init_table(rec->sg_aead_in, 2);
345 sg_set_buf(&rec->sg_aead_in[0], rec->aad_space, prot->aad_size);
346 sg_unmark_end(&rec->sg_aead_in[1]);
348 sg_init_table(rec->sg_aead_out, 2);
349 sg_set_buf(&rec->sg_aead_out[0], rec->aad_space, prot->aad_size);
350 sg_unmark_end(&rec->sg_aead_out[1]);
355 static void tls_free_rec(struct sock *sk, struct tls_rec *rec)
357 sk_msg_free(sk, &rec->msg_encrypted);
358 sk_msg_free(sk, &rec->msg_plaintext);
362 static void tls_free_open_rec(struct sock *sk)
364 struct tls_context *tls_ctx = tls_get_ctx(sk);
365 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
366 struct tls_rec *rec = ctx->open_rec;
369 tls_free_rec(sk, rec);
370 ctx->open_rec = NULL;
374 int tls_tx_records(struct sock *sk, int flags)
376 struct tls_context *tls_ctx = tls_get_ctx(sk);
377 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
378 struct tls_rec *rec, *tmp;
379 struct sk_msg *msg_en;
380 int tx_flags, rc = 0;
382 if (tls_is_partially_sent_record(tls_ctx)) {
383 rec = list_first_entry(&ctx->tx_list,
384 struct tls_rec, list);
387 tx_flags = rec->tx_flags;
391 rc = tls_push_partial_record(sk, tls_ctx, tx_flags);
395 /* Full record has been transmitted.
396 * Remove the head of tx_list
398 list_del(&rec->list);
399 sk_msg_free(sk, &rec->msg_plaintext);
403 /* Tx all ready records */
404 list_for_each_entry_safe(rec, tmp, &ctx->tx_list, list) {
405 if (READ_ONCE(rec->tx_ready)) {
407 tx_flags = rec->tx_flags;
411 msg_en = &rec->msg_encrypted;
412 rc = tls_push_sg(sk, tls_ctx,
413 &msg_en->sg.data[msg_en->sg.curr],
418 list_del(&rec->list);
419 sk_msg_free(sk, &rec->msg_plaintext);
427 if (rc < 0 && rc != -EAGAIN)
428 tls_err_abort(sk, -EBADMSG);
433 static void tls_encrypt_done(struct crypto_async_request *req, int err)
435 struct aead_request *aead_req = (struct aead_request *)req;
436 struct sock *sk = req->data;
437 struct tls_context *tls_ctx = tls_get_ctx(sk);
438 struct tls_prot_info *prot = &tls_ctx->prot_info;
439 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
440 struct scatterlist *sge;
441 struct sk_msg *msg_en;
446 rec = container_of(aead_req, struct tls_rec, aead_req);
447 msg_en = &rec->msg_encrypted;
449 sge = sk_msg_elem(msg_en, msg_en->sg.curr);
450 sge->offset -= prot->prepend_size;
451 sge->length += prot->prepend_size;
453 /* Check if error is previously set on socket */
454 if (err || sk->sk_err) {
457 /* If err is already set on socket, return the same code */
459 ctx->async_wait.err = -sk->sk_err;
461 ctx->async_wait.err = err;
462 tls_err_abort(sk, err);
467 struct tls_rec *first_rec;
469 /* Mark the record as ready for transmission */
470 smp_store_mb(rec->tx_ready, true);
472 /* If received record is at head of tx_list, schedule tx */
473 first_rec = list_first_entry(&ctx->tx_list,
474 struct tls_rec, list);
475 if (rec == first_rec)
479 spin_lock_bh(&ctx->encrypt_compl_lock);
480 pending = atomic_dec_return(&ctx->encrypt_pending);
482 if (!pending && ctx->async_notify)
483 complete(&ctx->async_wait.completion);
484 spin_unlock_bh(&ctx->encrypt_compl_lock);
489 /* Schedule the transmission */
490 if (!test_and_set_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask))
491 schedule_delayed_work(&ctx->tx_work.work, 1);
494 static int tls_do_encryption(struct sock *sk,
495 struct tls_context *tls_ctx,
496 struct tls_sw_context_tx *ctx,
497 struct aead_request *aead_req,
498 size_t data_len, u32 start)
500 struct tls_prot_info *prot = &tls_ctx->prot_info;
501 struct tls_rec *rec = ctx->open_rec;
502 struct sk_msg *msg_en = &rec->msg_encrypted;
503 struct scatterlist *sge = sk_msg_elem(msg_en, start);
504 int rc, iv_offset = 0;
506 /* For CCM based ciphers, first byte of IV is a constant */
507 switch (prot->cipher_type) {
508 case TLS_CIPHER_AES_CCM_128:
509 rec->iv_data[0] = TLS_AES_CCM_IV_B0_BYTE;
512 case TLS_CIPHER_SM4_CCM:
513 rec->iv_data[0] = TLS_SM4_CCM_IV_B0_BYTE;
518 memcpy(&rec->iv_data[iv_offset], tls_ctx->tx.iv,
519 prot->iv_size + prot->salt_size);
521 xor_iv_with_seq(prot, rec->iv_data + iv_offset, tls_ctx->tx.rec_seq);
523 sge->offset += prot->prepend_size;
524 sge->length -= prot->prepend_size;
526 msg_en->sg.curr = start;
528 aead_request_set_tfm(aead_req, ctx->aead_send);
529 aead_request_set_ad(aead_req, prot->aad_size);
530 aead_request_set_crypt(aead_req, rec->sg_aead_in,
532 data_len, rec->iv_data);
534 aead_request_set_callback(aead_req, CRYPTO_TFM_REQ_MAY_BACKLOG,
535 tls_encrypt_done, sk);
537 /* Add the record in tx_list */
538 list_add_tail((struct list_head *)&rec->list, &ctx->tx_list);
539 atomic_inc(&ctx->encrypt_pending);
541 rc = crypto_aead_encrypt(aead_req);
542 if (!rc || rc != -EINPROGRESS) {
543 atomic_dec(&ctx->encrypt_pending);
544 sge->offset -= prot->prepend_size;
545 sge->length += prot->prepend_size;
549 WRITE_ONCE(rec->tx_ready, true);
550 } else if (rc != -EINPROGRESS) {
551 list_del(&rec->list);
555 /* Unhook the record from context if encryption is not failure */
556 ctx->open_rec = NULL;
557 tls_advance_record_sn(sk, prot, &tls_ctx->tx);
561 static int tls_split_open_record(struct sock *sk, struct tls_rec *from,
562 struct tls_rec **to, struct sk_msg *msg_opl,
563 struct sk_msg *msg_oen, u32 split_point,
564 u32 tx_overhead_size, u32 *orig_end)
566 u32 i, j, bytes = 0, apply = msg_opl->apply_bytes;
567 struct scatterlist *sge, *osge, *nsge;
568 u32 orig_size = msg_opl->sg.size;
569 struct scatterlist tmp = { };
570 struct sk_msg *msg_npl;
574 new = tls_get_rec(sk);
577 ret = sk_msg_alloc(sk, &new->msg_encrypted, msg_opl->sg.size +
578 tx_overhead_size, 0);
580 tls_free_rec(sk, new);
584 *orig_end = msg_opl->sg.end;
585 i = msg_opl->sg.start;
586 sge = sk_msg_elem(msg_opl, i);
587 while (apply && sge->length) {
588 if (sge->length > apply) {
589 u32 len = sge->length - apply;
591 get_page(sg_page(sge));
592 sg_set_page(&tmp, sg_page(sge), len,
593 sge->offset + apply);
598 apply -= sge->length;
599 bytes += sge->length;
602 sk_msg_iter_var_next(i);
603 if (i == msg_opl->sg.end)
605 sge = sk_msg_elem(msg_opl, i);
609 msg_opl->sg.curr = i;
610 msg_opl->sg.copybreak = 0;
611 msg_opl->apply_bytes = 0;
612 msg_opl->sg.size = bytes;
614 msg_npl = &new->msg_plaintext;
615 msg_npl->apply_bytes = apply;
616 msg_npl->sg.size = orig_size - bytes;
618 j = msg_npl->sg.start;
619 nsge = sk_msg_elem(msg_npl, j);
621 memcpy(nsge, &tmp, sizeof(*nsge));
622 sk_msg_iter_var_next(j);
623 nsge = sk_msg_elem(msg_npl, j);
626 osge = sk_msg_elem(msg_opl, i);
627 while (osge->length) {
628 memcpy(nsge, osge, sizeof(*nsge));
630 sk_msg_iter_var_next(i);
631 sk_msg_iter_var_next(j);
634 osge = sk_msg_elem(msg_opl, i);
635 nsge = sk_msg_elem(msg_npl, j);
639 msg_npl->sg.curr = j;
640 msg_npl->sg.copybreak = 0;
646 static void tls_merge_open_record(struct sock *sk, struct tls_rec *to,
647 struct tls_rec *from, u32 orig_end)
649 struct sk_msg *msg_npl = &from->msg_plaintext;
650 struct sk_msg *msg_opl = &to->msg_plaintext;
651 struct scatterlist *osge, *nsge;
655 sk_msg_iter_var_prev(i);
656 j = msg_npl->sg.start;
658 osge = sk_msg_elem(msg_opl, i);
659 nsge = sk_msg_elem(msg_npl, j);
661 if (sg_page(osge) == sg_page(nsge) &&
662 osge->offset + osge->length == nsge->offset) {
663 osge->length += nsge->length;
664 put_page(sg_page(nsge));
667 msg_opl->sg.end = orig_end;
668 msg_opl->sg.curr = orig_end;
669 msg_opl->sg.copybreak = 0;
670 msg_opl->apply_bytes = msg_opl->sg.size + msg_npl->sg.size;
671 msg_opl->sg.size += msg_npl->sg.size;
673 sk_msg_free(sk, &to->msg_encrypted);
674 sk_msg_xfer_full(&to->msg_encrypted, &from->msg_encrypted);
679 static int tls_push_record(struct sock *sk, int flags,
680 unsigned char record_type)
682 struct tls_context *tls_ctx = tls_get_ctx(sk);
683 struct tls_prot_info *prot = &tls_ctx->prot_info;
684 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
685 struct tls_rec *rec = ctx->open_rec, *tmp = NULL;
686 u32 i, split_point, orig_end;
687 struct sk_msg *msg_pl, *msg_en;
688 struct aead_request *req;
695 msg_pl = &rec->msg_plaintext;
696 msg_en = &rec->msg_encrypted;
698 split_point = msg_pl->apply_bytes;
699 split = split_point && split_point < msg_pl->sg.size;
700 if (unlikely((!split &&
702 prot->overhead_size > msg_en->sg.size) ||
705 prot->overhead_size > msg_en->sg.size))) {
707 split_point = msg_en->sg.size;
710 rc = tls_split_open_record(sk, rec, &tmp, msg_pl, msg_en,
711 split_point, prot->overhead_size,
715 /* This can happen if above tls_split_open_record allocates
716 * a single large encryption buffer instead of two smaller
717 * ones. In this case adjust pointers and continue without
720 if (!msg_pl->sg.size) {
721 tls_merge_open_record(sk, rec, tmp, orig_end);
722 msg_pl = &rec->msg_plaintext;
723 msg_en = &rec->msg_encrypted;
726 sk_msg_trim(sk, msg_en, msg_pl->sg.size +
727 prot->overhead_size);
730 rec->tx_flags = flags;
731 req = &rec->aead_req;
734 sk_msg_iter_var_prev(i);
736 rec->content_type = record_type;
737 if (prot->version == TLS_1_3_VERSION) {
738 /* Add content type to end of message. No padding added */
739 sg_set_buf(&rec->sg_content_type, &rec->content_type, 1);
740 sg_mark_end(&rec->sg_content_type);
741 sg_chain(msg_pl->sg.data, msg_pl->sg.end + 1,
742 &rec->sg_content_type);
744 sg_mark_end(sk_msg_elem(msg_pl, i));
747 if (msg_pl->sg.end < msg_pl->sg.start) {
748 sg_chain(&msg_pl->sg.data[msg_pl->sg.start],
749 MAX_SKB_FRAGS - msg_pl->sg.start + 1,
753 i = msg_pl->sg.start;
754 sg_chain(rec->sg_aead_in, 2, &msg_pl->sg.data[i]);
757 sk_msg_iter_var_prev(i);
758 sg_mark_end(sk_msg_elem(msg_en, i));
760 i = msg_en->sg.start;
761 sg_chain(rec->sg_aead_out, 2, &msg_en->sg.data[i]);
763 tls_make_aad(rec->aad_space, msg_pl->sg.size + prot->tail_size,
764 tls_ctx->tx.rec_seq, record_type, prot);
766 tls_fill_prepend(tls_ctx,
767 page_address(sg_page(&msg_en->sg.data[i])) +
768 msg_en->sg.data[i].offset,
769 msg_pl->sg.size + prot->tail_size,
772 tls_ctx->pending_open_record_frags = false;
774 rc = tls_do_encryption(sk, tls_ctx, ctx, req,
775 msg_pl->sg.size + prot->tail_size, i);
777 if (rc != -EINPROGRESS) {
778 tls_err_abort(sk, -EBADMSG);
780 tls_ctx->pending_open_record_frags = true;
781 tls_merge_open_record(sk, rec, tmp, orig_end);
784 ctx->async_capable = 1;
787 msg_pl = &tmp->msg_plaintext;
788 msg_en = &tmp->msg_encrypted;
789 sk_msg_trim(sk, msg_en, msg_pl->sg.size + prot->overhead_size);
790 tls_ctx->pending_open_record_frags = true;
794 return tls_tx_records(sk, flags);
797 static int bpf_exec_tx_verdict(struct sk_msg *msg, struct sock *sk,
798 bool full_record, u8 record_type,
799 ssize_t *copied, int flags)
801 struct tls_context *tls_ctx = tls_get_ctx(sk);
802 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
803 struct sk_msg msg_redir = { };
804 struct sk_psock *psock;
805 struct sock *sk_redir;
811 policy = !(flags & MSG_SENDPAGE_NOPOLICY);
812 psock = sk_psock_get(sk);
813 if (!psock || !policy) {
814 err = tls_push_record(sk, flags, record_type);
815 if (err && sk->sk_err == EBADMSG) {
816 *copied -= sk_msg_free(sk, msg);
817 tls_free_open_rec(sk);
821 sk_psock_put(sk, psock);
825 enospc = sk_msg_full(msg);
826 if (psock->eval == __SK_NONE) {
827 delta = msg->sg.size;
828 psock->eval = sk_psock_msg_verdict(sk, psock, msg);
829 delta -= msg->sg.size;
831 if (msg->cork_bytes && msg->cork_bytes > msg->sg.size &&
832 !enospc && !full_record) {
838 if (msg->apply_bytes && msg->apply_bytes < send)
839 send = msg->apply_bytes;
841 switch (psock->eval) {
843 err = tls_push_record(sk, flags, record_type);
844 if (err && sk->sk_err == EBADMSG) {
845 *copied -= sk_msg_free(sk, msg);
846 tls_free_open_rec(sk);
852 sk_redir = psock->sk_redir;
853 memcpy(&msg_redir, msg, sizeof(*msg));
854 if (msg->apply_bytes < send)
855 msg->apply_bytes = 0;
857 msg->apply_bytes -= send;
858 sk_msg_return_zero(sk, msg, send);
859 msg->sg.size -= send;
861 err = tcp_bpf_sendmsg_redir(sk_redir, &msg_redir, send, flags);
864 *copied -= sk_msg_free_nocharge(sk, &msg_redir);
867 if (msg->sg.size == 0)
868 tls_free_open_rec(sk);
872 sk_msg_free_partial(sk, msg, send);
873 if (msg->apply_bytes < send)
874 msg->apply_bytes = 0;
876 msg->apply_bytes -= send;
877 if (msg->sg.size == 0)
878 tls_free_open_rec(sk);
879 *copied -= (send + delta);
884 bool reset_eval = !ctx->open_rec;
888 msg = &rec->msg_plaintext;
889 if (!msg->apply_bytes)
893 psock->eval = __SK_NONE;
894 if (psock->sk_redir) {
895 sock_put(psock->sk_redir);
896 psock->sk_redir = NULL;
903 sk_psock_put(sk, psock);
907 static int tls_sw_push_pending_record(struct sock *sk, int flags)
909 struct tls_context *tls_ctx = tls_get_ctx(sk);
910 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
911 struct tls_rec *rec = ctx->open_rec;
912 struct sk_msg *msg_pl;
918 msg_pl = &rec->msg_plaintext;
919 copied = msg_pl->sg.size;
923 return bpf_exec_tx_verdict(msg_pl, sk, true, TLS_RECORD_TYPE_DATA,
927 int tls_sw_sendmsg(struct sock *sk, struct msghdr *msg, size_t size)
929 long timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
930 struct tls_context *tls_ctx = tls_get_ctx(sk);
931 struct tls_prot_info *prot = &tls_ctx->prot_info;
932 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
933 bool async_capable = ctx->async_capable;
934 unsigned char record_type = TLS_RECORD_TYPE_DATA;
935 bool is_kvec = iov_iter_is_kvec(&msg->msg_iter);
936 bool eor = !(msg->msg_flags & MSG_MORE);
939 struct sk_msg *msg_pl, *msg_en;
950 if (msg->msg_flags & ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL |
954 mutex_lock(&tls_ctx->tx_lock);
957 if (unlikely(msg->msg_controllen)) {
958 ret = tls_proccess_cmsg(sk, msg, &record_type);
960 if (ret == -EINPROGRESS)
962 else if (ret != -EAGAIN)
967 while (msg_data_left(msg)) {
976 rec = ctx->open_rec = tls_get_rec(sk);
982 msg_pl = &rec->msg_plaintext;
983 msg_en = &rec->msg_encrypted;
985 orig_size = msg_pl->sg.size;
987 try_to_copy = msg_data_left(msg);
988 record_room = TLS_MAX_PAYLOAD_SIZE - msg_pl->sg.size;
989 if (try_to_copy >= record_room) {
990 try_to_copy = record_room;
994 required_size = msg_pl->sg.size + try_to_copy +
997 if (!sk_stream_memory_free(sk))
998 goto wait_for_sndbuf;
1001 ret = tls_alloc_encrypted_msg(sk, required_size);
1004 goto wait_for_memory;
1006 /* Adjust try_to_copy according to the amount that was
1007 * actually allocated. The difference is due
1008 * to max sg elements limit
1010 try_to_copy -= required_size - msg_en->sg.size;
1014 if (!is_kvec && (full_record || eor) && !async_capable) {
1015 u32 first = msg_pl->sg.end;
1017 ret = sk_msg_zerocopy_from_iter(sk, &msg->msg_iter,
1018 msg_pl, try_to_copy);
1020 goto fallback_to_reg_send;
1023 copied += try_to_copy;
1025 sk_msg_sg_copy_set(msg_pl, first);
1026 ret = bpf_exec_tx_verdict(msg_pl, sk, full_record,
1027 record_type, &copied,
1030 if (ret == -EINPROGRESS)
1032 else if (ret == -ENOMEM)
1033 goto wait_for_memory;
1034 else if (ctx->open_rec && ret == -ENOSPC)
1036 else if (ret != -EAGAIN)
1041 copied -= try_to_copy;
1042 sk_msg_sg_copy_clear(msg_pl, first);
1043 iov_iter_revert(&msg->msg_iter,
1044 msg_pl->sg.size - orig_size);
1045 fallback_to_reg_send:
1046 sk_msg_trim(sk, msg_pl, orig_size);
1049 required_size = msg_pl->sg.size + try_to_copy;
1051 ret = tls_clone_plaintext_msg(sk, required_size);
1056 /* Adjust try_to_copy according to the amount that was
1057 * actually allocated. The difference is due
1058 * to max sg elements limit
1060 try_to_copy -= required_size - msg_pl->sg.size;
1062 sk_msg_trim(sk, msg_en,
1063 msg_pl->sg.size + prot->overhead_size);
1067 ret = sk_msg_memcopy_from_iter(sk, &msg->msg_iter,
1068 msg_pl, try_to_copy);
1073 /* Open records defined only if successfully copied, otherwise
1074 * we would trim the sg but not reset the open record frags.
1076 tls_ctx->pending_open_record_frags = true;
1077 copied += try_to_copy;
1078 if (full_record || eor) {
1079 ret = bpf_exec_tx_verdict(msg_pl, sk, full_record,
1080 record_type, &copied,
1083 if (ret == -EINPROGRESS)
1085 else if (ret == -ENOMEM)
1086 goto wait_for_memory;
1087 else if (ret != -EAGAIN) {
1098 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1100 ret = sk_stream_wait_memory(sk, &timeo);
1104 tls_trim_both_msgs(sk, orig_size);
1108 if (ctx->open_rec && msg_en->sg.size < required_size)
1109 goto alloc_encrypted;
1114 } else if (num_zc) {
1115 /* Wait for pending encryptions to get completed */
1116 spin_lock_bh(&ctx->encrypt_compl_lock);
1117 ctx->async_notify = true;
1119 pending = atomic_read(&ctx->encrypt_pending);
1120 spin_unlock_bh(&ctx->encrypt_compl_lock);
1122 crypto_wait_req(-EINPROGRESS, &ctx->async_wait);
1124 reinit_completion(&ctx->async_wait.completion);
1126 /* There can be no concurrent accesses, since we have no
1127 * pending encrypt operations
1129 WRITE_ONCE(ctx->async_notify, false);
1131 if (ctx->async_wait.err) {
1132 ret = ctx->async_wait.err;
1137 /* Transmit if any encryptions have completed */
1138 if (test_and_clear_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask)) {
1139 cancel_delayed_work(&ctx->tx_work.work);
1140 tls_tx_records(sk, msg->msg_flags);
1144 ret = sk_stream_error(sk, msg->msg_flags, ret);
1147 mutex_unlock(&tls_ctx->tx_lock);
1148 return copied > 0 ? copied : ret;
1151 static int tls_sw_do_sendpage(struct sock *sk, struct page *page,
1152 int offset, size_t size, int flags)
1154 long timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT);
1155 struct tls_context *tls_ctx = tls_get_ctx(sk);
1156 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
1157 struct tls_prot_info *prot = &tls_ctx->prot_info;
1158 unsigned char record_type = TLS_RECORD_TYPE_DATA;
1159 struct sk_msg *msg_pl;
1160 struct tls_rec *rec;
1168 eor = !(flags & MSG_SENDPAGE_NOTLAST);
1169 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
1171 /* Call the sk_stream functions to manage the sndbuf mem. */
1173 size_t copy, required_size;
1181 rec = ctx->open_rec;
1183 rec = ctx->open_rec = tls_get_rec(sk);
1189 msg_pl = &rec->msg_plaintext;
1191 full_record = false;
1192 record_room = TLS_MAX_PAYLOAD_SIZE - msg_pl->sg.size;
1194 if (copy >= record_room) {
1199 required_size = msg_pl->sg.size + copy + prot->overhead_size;
1201 if (!sk_stream_memory_free(sk))
1202 goto wait_for_sndbuf;
1204 ret = tls_alloc_encrypted_msg(sk, required_size);
1207 goto wait_for_memory;
1209 /* Adjust copy according to the amount that was
1210 * actually allocated. The difference is due
1211 * to max sg elements limit
1213 copy -= required_size - msg_pl->sg.size;
1217 sk_msg_page_add(msg_pl, page, copy, offset);
1218 sk_mem_charge(sk, copy);
1224 tls_ctx->pending_open_record_frags = true;
1225 if (full_record || eor || sk_msg_full(msg_pl)) {
1226 ret = bpf_exec_tx_verdict(msg_pl, sk, full_record,
1227 record_type, &copied, flags);
1229 if (ret == -EINPROGRESS)
1231 else if (ret == -ENOMEM)
1232 goto wait_for_memory;
1233 else if (ret != -EAGAIN) {
1242 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1244 ret = sk_stream_wait_memory(sk, &timeo);
1247 tls_trim_both_msgs(sk, msg_pl->sg.size);
1256 /* Transmit if any encryptions have completed */
1257 if (test_and_clear_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask)) {
1258 cancel_delayed_work(&ctx->tx_work.work);
1259 tls_tx_records(sk, flags);
1263 ret = sk_stream_error(sk, flags, ret);
1264 return copied > 0 ? copied : ret;
1267 int tls_sw_sendpage_locked(struct sock *sk, struct page *page,
1268 int offset, size_t size, int flags)
1270 if (flags & ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL |
1271 MSG_SENDPAGE_NOTLAST | MSG_SENDPAGE_NOPOLICY |
1272 MSG_NO_SHARED_FRAGS))
1275 return tls_sw_do_sendpage(sk, page, offset, size, flags);
1278 int tls_sw_sendpage(struct sock *sk, struct page *page,
1279 int offset, size_t size, int flags)
1281 struct tls_context *tls_ctx = tls_get_ctx(sk);
1284 if (flags & ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL |
1285 MSG_SENDPAGE_NOTLAST | MSG_SENDPAGE_NOPOLICY))
1288 mutex_lock(&tls_ctx->tx_lock);
1290 ret = tls_sw_do_sendpage(sk, page, offset, size, flags);
1292 mutex_unlock(&tls_ctx->tx_lock);
1296 static struct sk_buff *tls_wait_data(struct sock *sk, struct sk_psock *psock,
1297 bool nonblock, long timeo, int *err)
1299 struct tls_context *tls_ctx = tls_get_ctx(sk);
1300 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1301 struct sk_buff *skb;
1302 DEFINE_WAIT_FUNC(wait, woken_wake_function);
1304 while (!(skb = ctx->recv_pkt) && sk_psock_queue_empty(psock)) {
1306 *err = sock_error(sk);
1310 if (!skb_queue_empty(&sk->sk_receive_queue)) {
1311 __strp_unpause(&ctx->strp);
1313 return ctx->recv_pkt;
1316 if (sk->sk_shutdown & RCV_SHUTDOWN)
1319 if (sock_flag(sk, SOCK_DONE))
1322 if (nonblock || !timeo) {
1327 add_wait_queue(sk_sleep(sk), &wait);
1328 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
1329 sk_wait_event(sk, &timeo,
1330 ctx->recv_pkt != skb ||
1331 !sk_psock_queue_empty(psock),
1333 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
1334 remove_wait_queue(sk_sleep(sk), &wait);
1336 /* Handle signals */
1337 if (signal_pending(current)) {
1338 *err = sock_intr_errno(timeo);
1346 static int tls_setup_from_iter(struct iov_iter *from,
1347 int length, int *pages_used,
1348 struct scatterlist *to,
1351 int rc = 0, i = 0, num_elem = *pages_used, maxpages;
1352 struct page *pages[MAX_SKB_FRAGS];
1353 unsigned int size = 0;
1354 ssize_t copied, use;
1357 while (length > 0) {
1359 maxpages = to_max_pages - num_elem;
1360 if (maxpages == 0) {
1364 copied = iov_iter_get_pages(from, pages,
1372 iov_iter_advance(from, copied);
1377 use = min_t(int, copied, PAGE_SIZE - offset);
1379 sg_set_page(&to[num_elem],
1380 pages[i], use, offset);
1381 sg_unmark_end(&to[num_elem]);
1382 /* We do not uncharge memory from this API */
1391 /* Mark the end in the last sg entry if newly added */
1392 if (num_elem > *pages_used)
1393 sg_mark_end(&to[num_elem - 1]);
1396 iov_iter_revert(from, size);
1397 *pages_used = num_elem;
1402 /* This function decrypts the input skb into either out_iov or in out_sg
1403 * or in skb buffers itself. The input parameter 'zc' indicates if
1404 * zero-copy mode needs to be tried or not. With zero-copy mode, either
1405 * out_iov or out_sg must be non-NULL. In case both out_iov and out_sg are
1406 * NULL, then the decryption happens inside skb buffers itself, i.e.
1407 * zero-copy gets disabled and 'zc' is updated.
1410 static int decrypt_internal(struct sock *sk, struct sk_buff *skb,
1411 struct iov_iter *out_iov,
1412 struct scatterlist *out_sg,
1413 struct tls_decrypt_arg *darg)
1415 struct tls_context *tls_ctx = tls_get_ctx(sk);
1416 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1417 struct tls_prot_info *prot = &tls_ctx->prot_info;
1418 struct strp_msg *rxm = strp_msg(skb);
1419 struct tls_msg *tlm = tls_msg(skb);
1420 int n_sgin, n_sgout, nsg, mem_size, aead_size, err, pages = 0;
1421 struct aead_request *aead_req;
1422 struct sk_buff *unused;
1423 u8 *aad, *iv, *mem = NULL;
1424 struct scatterlist *sgin = NULL;
1425 struct scatterlist *sgout = NULL;
1426 const int data_len = rxm->full_len - prot->overhead_size +
1430 if (darg->zc && (out_iov || out_sg)) {
1433 iov_iter_npages_cap(out_iov, INT_MAX, data_len);
1435 n_sgout = sg_nents(out_sg);
1436 n_sgin = skb_nsg(skb, rxm->offset + prot->prepend_size,
1437 rxm->full_len - prot->prepend_size);
1441 n_sgin = skb_cow_data(skb, 0, &unused);
1447 /* Increment to accommodate AAD */
1448 n_sgin = n_sgin + 1;
1450 nsg = n_sgin + n_sgout;
1452 aead_size = sizeof(*aead_req) + crypto_aead_reqsize(ctx->aead_recv);
1453 mem_size = aead_size + (nsg * sizeof(struct scatterlist));
1454 mem_size = mem_size + prot->aad_size;
1455 mem_size = mem_size + MAX_IV_SIZE;
1457 /* Allocate a single block of memory which contains
1458 * aead_req || sgin[] || sgout[] || aad || iv.
1459 * This order achieves correct alignment for aead_req, sgin, sgout.
1461 mem = kmalloc(mem_size, sk->sk_allocation);
1465 /* Segment the allocated memory */
1466 aead_req = (struct aead_request *)mem;
1467 sgin = (struct scatterlist *)(mem + aead_size);
1468 sgout = sgin + n_sgin;
1469 aad = (u8 *)(sgout + n_sgout);
1470 iv = aad + prot->aad_size;
1472 /* For CCM based ciphers, first byte of nonce+iv is a constant */
1473 switch (prot->cipher_type) {
1474 case TLS_CIPHER_AES_CCM_128:
1475 iv[0] = TLS_AES_CCM_IV_B0_BYTE;
1478 case TLS_CIPHER_SM4_CCM:
1479 iv[0] = TLS_SM4_CCM_IV_B0_BYTE;
1485 if (prot->version == TLS_1_3_VERSION ||
1486 prot->cipher_type == TLS_CIPHER_CHACHA20_POLY1305) {
1487 memcpy(iv + iv_offset, tls_ctx->rx.iv,
1488 prot->iv_size + prot->salt_size);
1490 err = skb_copy_bits(skb, rxm->offset + TLS_HEADER_SIZE,
1491 iv + iv_offset + prot->salt_size,
1497 memcpy(iv + iv_offset, tls_ctx->rx.iv, prot->salt_size);
1499 xor_iv_with_seq(prot, iv + iv_offset, tls_ctx->rx.rec_seq);
1502 tls_make_aad(aad, rxm->full_len - prot->overhead_size +
1504 tls_ctx->rx.rec_seq, tlm->control, prot);
1507 sg_init_table(sgin, n_sgin);
1508 sg_set_buf(&sgin[0], aad, prot->aad_size);
1509 err = skb_to_sgvec(skb, &sgin[1],
1510 rxm->offset + prot->prepend_size,
1511 rxm->full_len - prot->prepend_size);
1519 sg_init_table(sgout, n_sgout);
1520 sg_set_buf(&sgout[0], aad, prot->aad_size);
1522 err = tls_setup_from_iter(out_iov, data_len,
1526 goto fallback_to_reg_recv;
1527 } else if (out_sg) {
1528 memcpy(sgout, out_sg, n_sgout * sizeof(*sgout));
1530 goto fallback_to_reg_recv;
1533 fallback_to_reg_recv:
1539 /* Prepare and submit AEAD request */
1540 err = tls_do_decryption(sk, skb, sgin, sgout, iv,
1541 data_len, aead_req, darg);
1545 /* Release the pages in case iov was mapped to pages */
1546 for (; pages > 0; pages--)
1547 put_page(sg_page(&sgout[pages]));
1553 static int decrypt_skb_update(struct sock *sk, struct sk_buff *skb,
1554 struct iov_iter *dest,
1555 struct tls_decrypt_arg *darg)
1557 struct tls_context *tls_ctx = tls_get_ctx(sk);
1558 struct tls_prot_info *prot = &tls_ctx->prot_info;
1559 struct strp_msg *rxm = strp_msg(skb);
1560 struct tls_msg *tlm = tls_msg(skb);
1563 if (tlm->decrypted) {
1565 darg->async = false;
1569 if (tls_ctx->rx_conf == TLS_HW) {
1570 err = tls_device_decrypted(sk, tls_ctx, skb, rxm);
1576 darg->async = false;
1581 err = decrypt_internal(sk, skb, dest, NULL, darg);
1588 pad = padding_length(prot, skb);
1592 rxm->full_len -= pad;
1593 rxm->offset += prot->prepend_size;
1594 rxm->full_len -= prot->overhead_size;
1597 tls_advance_record_sn(sk, prot, &tls_ctx->rx);
1602 int decrypt_skb(struct sock *sk, struct sk_buff *skb,
1603 struct scatterlist *sgout)
1605 struct tls_decrypt_arg darg = { .zc = true, };
1607 return decrypt_internal(sk, skb, NULL, sgout, &darg);
1610 static int tls_record_content_type(struct msghdr *msg, struct tls_msg *tlm,
1616 *control = tlm->control;
1620 err = put_cmsg(msg, SOL_TLS, TLS_GET_RECORD_TYPE,
1621 sizeof(*control), control);
1622 if (*control != TLS_RECORD_TYPE_DATA) {
1623 if (err || msg->msg_flags & MSG_CTRUNC)
1626 } else if (*control != tlm->control) {
1633 /* This function traverses the rx_list in tls receive context to copies the
1634 * decrypted records into the buffer provided by caller zero copy is not
1635 * true. Further, the records are removed from the rx_list if it is not a peek
1636 * case and the record has been consumed completely.
1638 static int process_rx_list(struct tls_sw_context_rx *ctx,
1646 struct sk_buff *skb = skb_peek(&ctx->rx_list);
1647 struct tls_msg *tlm;
1651 while (skip && skb) {
1652 struct strp_msg *rxm = strp_msg(skb);
1655 err = tls_record_content_type(msg, tlm, control);
1659 if (skip < rxm->full_len)
1662 skip = skip - rxm->full_len;
1663 skb = skb_peek_next(skb, &ctx->rx_list);
1666 while (len && skb) {
1667 struct sk_buff *next_skb;
1668 struct strp_msg *rxm = strp_msg(skb);
1669 int chunk = min_t(unsigned int, rxm->full_len - skip, len);
1673 err = tls_record_content_type(msg, tlm, control);
1677 if (!zc || (rxm->full_len - skip) > len) {
1678 err = skb_copy_datagram_msg(skb, rxm->offset + skip,
1685 copied = copied + chunk;
1687 /* Consume the data from record if it is non-peek case*/
1689 rxm->offset = rxm->offset + chunk;
1690 rxm->full_len = rxm->full_len - chunk;
1692 /* Return if there is unconsumed data in the record */
1693 if (rxm->full_len - skip)
1697 /* The remaining skip-bytes must lie in 1st record in rx_list.
1698 * So from the 2nd record, 'skip' should be 0.
1703 msg->msg_flags |= MSG_EOR;
1705 next_skb = skb_peek_next(skb, &ctx->rx_list);
1708 __skb_unlink(skb, &ctx->rx_list);
1717 return copied ? : err;
1720 int tls_sw_recvmsg(struct sock *sk,
1726 struct tls_context *tls_ctx = tls_get_ctx(sk);
1727 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1728 struct tls_prot_info *prot = &tls_ctx->prot_info;
1729 struct sk_psock *psock;
1730 unsigned char control = 0;
1731 ssize_t decrypted = 0;
1732 struct strp_msg *rxm;
1733 struct tls_msg *tlm;
1734 struct sk_buff *skb;
1737 int target, err = 0;
1739 bool is_kvec = iov_iter_is_kvec(&msg->msg_iter);
1740 bool is_peek = flags & MSG_PEEK;
1741 bool bpf_strp_enabled;
1744 if (unlikely(flags & MSG_ERRQUEUE))
1745 return sock_recv_errqueue(sk, msg, len, SOL_IP, IP_RECVERR);
1747 psock = sk_psock_get(sk);
1749 bpf_strp_enabled = sk_psock_strp_enabled(psock);
1751 /* If crypto failed the connection is broken */
1752 err = ctx->async_wait.err;
1756 /* Process pending decrypted records. It must be non-zero-copy */
1757 err = process_rx_list(ctx, msg, &control, 0, len, false, is_peek);
1765 target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
1767 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1769 zc_capable = !bpf_strp_enabled && !is_kvec && !is_peek &&
1770 prot->version != TLS_1_3_VERSION;
1772 while (len && (decrypted + copied < target || ctx->recv_pkt)) {
1773 struct tls_decrypt_arg darg = {};
1774 int to_decrypt, chunk;
1776 skb = tls_wait_data(sk, psock, flags & MSG_DONTWAIT, timeo, &err);
1779 chunk = sk_msg_recvmsg(sk, psock, msg, len,
1787 rxm = strp_msg(skb);
1790 to_decrypt = rxm->full_len - prot->overhead_size;
1792 if (zc_capable && to_decrypt <= len &&
1793 tlm->control == TLS_RECORD_TYPE_DATA)
1796 /* Do not use async mode if record is non-data */
1797 if (tlm->control == TLS_RECORD_TYPE_DATA && !bpf_strp_enabled)
1798 darg.async = ctx->async_capable;
1802 err = decrypt_skb_update(sk, skb, &msg->msg_iter, &darg);
1804 tls_err_abort(sk, -EBADMSG);
1808 async |= darg.async;
1810 /* If the type of records being processed is not known yet,
1811 * set it to record type just dequeued. If it is already known,
1812 * but does not match the record type just dequeued, go to end.
1813 * We always get record type here since for tls1.2, record type
1814 * is known just after record is dequeued from stream parser.
1815 * For tls1.3, we disable async.
1817 err = tls_record_content_type(msg, tlm, &control);
1821 ctx->recv_pkt = NULL;
1822 __strp_unpause(&ctx->strp);
1823 __skb_queue_tail(&ctx->rx_list, skb);
1826 /* TLS 1.2-only, to_decrypt must be text length */
1827 chunk = min_t(int, to_decrypt, len);
1833 /* TLS 1.3 may have updated the length by more than overhead */
1834 chunk = rxm->full_len;
1837 bool partially_consumed = chunk > len;
1839 if (bpf_strp_enabled) {
1840 /* BPF may try to queue the skb */
1841 __skb_unlink(skb, &ctx->rx_list);
1842 err = sk_psock_tls_strp_read(psock, skb);
1843 if (err != __SK_PASS) {
1844 rxm->offset = rxm->offset + rxm->full_len;
1846 if (err == __SK_DROP)
1850 __skb_queue_tail(&ctx->rx_list, skb);
1853 if (partially_consumed)
1856 err = skb_copy_datagram_msg(skb, rxm->offset,
1864 if (partially_consumed) {
1865 rxm->offset += chunk;
1866 rxm->full_len -= chunk;
1874 __skb_unlink(skb, &ctx->rx_list);
1877 /* Return full control message to userspace before trying
1878 * to parse another message type
1880 msg->msg_flags |= MSG_EOR;
1881 if (control != TLS_RECORD_TYPE_DATA)
1889 /* Wait for all previously submitted records to be decrypted */
1890 spin_lock_bh(&ctx->decrypt_compl_lock);
1891 reinit_completion(&ctx->async_wait.completion);
1892 pending = atomic_read(&ctx->decrypt_pending);
1893 spin_unlock_bh(&ctx->decrypt_compl_lock);
1895 ret = crypto_wait_req(-EINPROGRESS, &ctx->async_wait);
1897 if (err >= 0 || err == -EINPROGRESS)
1904 /* Drain records from the rx_list & copy if required */
1905 if (is_peek || is_kvec)
1906 err = process_rx_list(ctx, msg, &control, copied,
1907 decrypted, false, is_peek);
1909 err = process_rx_list(ctx, msg, &control, 0,
1910 decrypted, true, is_peek);
1911 decrypted = max(err, 0);
1914 copied += decrypted;
1919 sk_psock_put(sk, psock);
1920 return copied ? : err;
1923 ssize_t tls_sw_splice_read(struct socket *sock, loff_t *ppos,
1924 struct pipe_inode_info *pipe,
1925 size_t len, unsigned int flags)
1927 struct tls_context *tls_ctx = tls_get_ctx(sock->sk);
1928 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1929 struct strp_msg *rxm = NULL;
1930 struct sock *sk = sock->sk;
1931 struct tls_msg *tlm;
1932 struct sk_buff *skb;
1941 timeo = sock_rcvtimeo(sk, flags & SPLICE_F_NONBLOCK);
1943 from_queue = !skb_queue_empty(&ctx->rx_list);
1945 skb = __skb_dequeue(&ctx->rx_list);
1947 struct tls_decrypt_arg darg = {};
1949 skb = tls_wait_data(sk, NULL, flags & SPLICE_F_NONBLOCK, timeo,
1952 goto splice_read_end;
1954 err = decrypt_skb_update(sk, skb, NULL, &darg);
1956 tls_err_abort(sk, -EBADMSG);
1957 goto splice_read_end;
1961 rxm = strp_msg(skb);
1964 /* splice does not support reading control messages */
1965 if (tlm->control != TLS_RECORD_TYPE_DATA) {
1967 goto splice_read_end;
1970 chunk = min_t(unsigned int, rxm->full_len, len);
1971 copied = skb_splice_bits(skb, sk, rxm->offset, pipe, chunk, flags);
1973 goto splice_read_end;
1976 ctx->recv_pkt = NULL;
1977 __strp_unpause(&ctx->strp);
1979 if (chunk < rxm->full_len) {
1980 __skb_queue_head(&ctx->rx_list, skb);
1982 rxm->full_len -= len;
1989 return copied ? : err;
1992 bool tls_sw_sock_is_readable(struct sock *sk)
1994 struct tls_context *tls_ctx = tls_get_ctx(sk);
1995 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1996 bool ingress_empty = true;
1997 struct sk_psock *psock;
2000 psock = sk_psock(sk);
2002 ingress_empty = list_empty(&psock->ingress_msg);
2005 return !ingress_empty || ctx->recv_pkt ||
2006 !skb_queue_empty(&ctx->rx_list);
2009 static int tls_read_size(struct strparser *strp, struct sk_buff *skb)
2011 struct tls_context *tls_ctx = tls_get_ctx(strp->sk);
2012 struct tls_prot_info *prot = &tls_ctx->prot_info;
2013 char header[TLS_HEADER_SIZE + MAX_IV_SIZE];
2014 struct strp_msg *rxm = strp_msg(skb);
2015 struct tls_msg *tlm = tls_msg(skb);
2016 size_t cipher_overhead;
2017 size_t data_len = 0;
2020 /* Verify that we have a full TLS header, or wait for more data */
2021 if (rxm->offset + prot->prepend_size > skb->len)
2024 /* Sanity-check size of on-stack buffer. */
2025 if (WARN_ON(prot->prepend_size > sizeof(header))) {
2030 /* Linearize header to local buffer */
2031 ret = skb_copy_bits(skb, rxm->offset, header, prot->prepend_size);
2036 tlm->control = header[0];
2038 data_len = ((header[4] & 0xFF) | (header[3] << 8));
2040 cipher_overhead = prot->tag_size;
2041 if (prot->version != TLS_1_3_VERSION &&
2042 prot->cipher_type != TLS_CIPHER_CHACHA20_POLY1305)
2043 cipher_overhead += prot->iv_size;
2045 if (data_len > TLS_MAX_PAYLOAD_SIZE + cipher_overhead +
2050 if (data_len < cipher_overhead) {
2055 /* Note that both TLS1.3 and TLS1.2 use TLS_1_2 version here */
2056 if (header[1] != TLS_1_2_VERSION_MINOR ||
2057 header[2] != TLS_1_2_VERSION_MAJOR) {
2062 tls_device_rx_resync_new_rec(strp->sk, data_len + TLS_HEADER_SIZE,
2063 TCP_SKB_CB(skb)->seq + rxm->offset);
2064 return data_len + TLS_HEADER_SIZE;
2067 tls_err_abort(strp->sk, ret);
2072 static void tls_queue(struct strparser *strp, struct sk_buff *skb)
2074 struct tls_context *tls_ctx = tls_get_ctx(strp->sk);
2075 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2077 ctx->recv_pkt = skb;
2080 ctx->saved_data_ready(strp->sk);
2083 static void tls_data_ready(struct sock *sk)
2085 struct tls_context *tls_ctx = tls_get_ctx(sk);
2086 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2087 struct sk_psock *psock;
2089 strp_data_ready(&ctx->strp);
2091 psock = sk_psock_get(sk);
2093 if (!list_empty(&psock->ingress_msg))
2094 ctx->saved_data_ready(sk);
2095 sk_psock_put(sk, psock);
2099 void tls_sw_cancel_work_tx(struct tls_context *tls_ctx)
2101 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
2103 set_bit(BIT_TX_CLOSING, &ctx->tx_bitmask);
2104 set_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask);
2105 cancel_delayed_work_sync(&ctx->tx_work.work);
2108 void tls_sw_release_resources_tx(struct sock *sk)
2110 struct tls_context *tls_ctx = tls_get_ctx(sk);
2111 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
2112 struct tls_rec *rec, *tmp;
2115 /* Wait for any pending async encryptions to complete */
2116 spin_lock_bh(&ctx->encrypt_compl_lock);
2117 ctx->async_notify = true;
2118 pending = atomic_read(&ctx->encrypt_pending);
2119 spin_unlock_bh(&ctx->encrypt_compl_lock);
2122 crypto_wait_req(-EINPROGRESS, &ctx->async_wait);
2124 tls_tx_records(sk, -1);
2126 /* Free up un-sent records in tx_list. First, free
2127 * the partially sent record if any at head of tx_list.
2129 if (tls_ctx->partially_sent_record) {
2130 tls_free_partial_record(sk, tls_ctx);
2131 rec = list_first_entry(&ctx->tx_list,
2132 struct tls_rec, list);
2133 list_del(&rec->list);
2134 sk_msg_free(sk, &rec->msg_plaintext);
2138 list_for_each_entry_safe(rec, tmp, &ctx->tx_list, list) {
2139 list_del(&rec->list);
2140 sk_msg_free(sk, &rec->msg_encrypted);
2141 sk_msg_free(sk, &rec->msg_plaintext);
2145 crypto_free_aead(ctx->aead_send);
2146 tls_free_open_rec(sk);
2149 void tls_sw_free_ctx_tx(struct tls_context *tls_ctx)
2151 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
2156 void tls_sw_release_resources_rx(struct sock *sk)
2158 struct tls_context *tls_ctx = tls_get_ctx(sk);
2159 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2161 kfree(tls_ctx->rx.rec_seq);
2162 kfree(tls_ctx->rx.iv);
2164 if (ctx->aead_recv) {
2165 kfree_skb(ctx->recv_pkt);
2166 ctx->recv_pkt = NULL;
2167 __skb_queue_purge(&ctx->rx_list);
2168 crypto_free_aead(ctx->aead_recv);
2169 strp_stop(&ctx->strp);
2170 /* If tls_sw_strparser_arm() was not called (cleanup paths)
2171 * we still want to strp_stop(), but sk->sk_data_ready was
2174 if (ctx->saved_data_ready) {
2175 write_lock_bh(&sk->sk_callback_lock);
2176 sk->sk_data_ready = ctx->saved_data_ready;
2177 write_unlock_bh(&sk->sk_callback_lock);
2182 void tls_sw_strparser_done(struct tls_context *tls_ctx)
2184 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2186 strp_done(&ctx->strp);
2189 void tls_sw_free_ctx_rx(struct tls_context *tls_ctx)
2191 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2196 void tls_sw_free_resources_rx(struct sock *sk)
2198 struct tls_context *tls_ctx = tls_get_ctx(sk);
2200 tls_sw_release_resources_rx(sk);
2201 tls_sw_free_ctx_rx(tls_ctx);
2204 /* The work handler to transmitt the encrypted records in tx_list */
2205 static void tx_work_handler(struct work_struct *work)
2207 struct delayed_work *delayed_work = to_delayed_work(work);
2208 struct tx_work *tx_work = container_of(delayed_work,
2209 struct tx_work, work);
2210 struct sock *sk = tx_work->sk;
2211 struct tls_context *tls_ctx = tls_get_ctx(sk);
2212 struct tls_sw_context_tx *ctx;
2214 if (unlikely(!tls_ctx))
2217 ctx = tls_sw_ctx_tx(tls_ctx);
2218 if (test_bit(BIT_TX_CLOSING, &ctx->tx_bitmask))
2221 if (!test_and_clear_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask))
2223 mutex_lock(&tls_ctx->tx_lock);
2225 tls_tx_records(sk, -1);
2227 mutex_unlock(&tls_ctx->tx_lock);
2230 void tls_sw_write_space(struct sock *sk, struct tls_context *ctx)
2232 struct tls_sw_context_tx *tx_ctx = tls_sw_ctx_tx(ctx);
2234 /* Schedule the transmission if tx list is ready */
2235 if (is_tx_ready(tx_ctx) &&
2236 !test_and_set_bit(BIT_TX_SCHEDULED, &tx_ctx->tx_bitmask))
2237 schedule_delayed_work(&tx_ctx->tx_work.work, 0);
2240 void tls_sw_strparser_arm(struct sock *sk, struct tls_context *tls_ctx)
2242 struct tls_sw_context_rx *rx_ctx = tls_sw_ctx_rx(tls_ctx);
2244 write_lock_bh(&sk->sk_callback_lock);
2245 rx_ctx->saved_data_ready = sk->sk_data_ready;
2246 sk->sk_data_ready = tls_data_ready;
2247 write_unlock_bh(&sk->sk_callback_lock);
2249 strp_check_rcv(&rx_ctx->strp);
2252 int tls_set_sw_offload(struct sock *sk, struct tls_context *ctx, int tx)
2254 struct tls_context *tls_ctx = tls_get_ctx(sk);
2255 struct tls_prot_info *prot = &tls_ctx->prot_info;
2256 struct tls_crypto_info *crypto_info;
2257 struct tls_sw_context_tx *sw_ctx_tx = NULL;
2258 struct tls_sw_context_rx *sw_ctx_rx = NULL;
2259 struct cipher_context *cctx;
2260 struct crypto_aead **aead;
2261 struct strp_callbacks cb;
2262 u16 nonce_size, tag_size, iv_size, rec_seq_size, salt_size;
2263 struct crypto_tfm *tfm;
2264 char *iv, *rec_seq, *key, *salt, *cipher_name;
2274 if (!ctx->priv_ctx_tx) {
2275 sw_ctx_tx = kzalloc(sizeof(*sw_ctx_tx), GFP_KERNEL);
2280 ctx->priv_ctx_tx = sw_ctx_tx;
2283 (struct tls_sw_context_tx *)ctx->priv_ctx_tx;
2286 if (!ctx->priv_ctx_rx) {
2287 sw_ctx_rx = kzalloc(sizeof(*sw_ctx_rx), GFP_KERNEL);
2292 ctx->priv_ctx_rx = sw_ctx_rx;
2295 (struct tls_sw_context_rx *)ctx->priv_ctx_rx;
2300 crypto_init_wait(&sw_ctx_tx->async_wait);
2301 spin_lock_init(&sw_ctx_tx->encrypt_compl_lock);
2302 crypto_info = &ctx->crypto_send.info;
2304 aead = &sw_ctx_tx->aead_send;
2305 INIT_LIST_HEAD(&sw_ctx_tx->tx_list);
2306 INIT_DELAYED_WORK(&sw_ctx_tx->tx_work.work, tx_work_handler);
2307 sw_ctx_tx->tx_work.sk = sk;
2309 crypto_init_wait(&sw_ctx_rx->async_wait);
2310 spin_lock_init(&sw_ctx_rx->decrypt_compl_lock);
2311 crypto_info = &ctx->crypto_recv.info;
2313 skb_queue_head_init(&sw_ctx_rx->rx_list);
2314 aead = &sw_ctx_rx->aead_recv;
2317 switch (crypto_info->cipher_type) {
2318 case TLS_CIPHER_AES_GCM_128: {
2319 struct tls12_crypto_info_aes_gcm_128 *gcm_128_info;
2321 gcm_128_info = (void *)crypto_info;
2322 nonce_size = TLS_CIPHER_AES_GCM_128_IV_SIZE;
2323 tag_size = TLS_CIPHER_AES_GCM_128_TAG_SIZE;
2324 iv_size = TLS_CIPHER_AES_GCM_128_IV_SIZE;
2325 iv = gcm_128_info->iv;
2326 rec_seq_size = TLS_CIPHER_AES_GCM_128_REC_SEQ_SIZE;
2327 rec_seq = gcm_128_info->rec_seq;
2328 keysize = TLS_CIPHER_AES_GCM_128_KEY_SIZE;
2329 key = gcm_128_info->key;
2330 salt = gcm_128_info->salt;
2331 salt_size = TLS_CIPHER_AES_GCM_128_SALT_SIZE;
2332 cipher_name = "gcm(aes)";
2335 case TLS_CIPHER_AES_GCM_256: {
2336 struct tls12_crypto_info_aes_gcm_256 *gcm_256_info;
2338 gcm_256_info = (void *)crypto_info;
2339 nonce_size = TLS_CIPHER_AES_GCM_256_IV_SIZE;
2340 tag_size = TLS_CIPHER_AES_GCM_256_TAG_SIZE;
2341 iv_size = TLS_CIPHER_AES_GCM_256_IV_SIZE;
2342 iv = gcm_256_info->iv;
2343 rec_seq_size = TLS_CIPHER_AES_GCM_256_REC_SEQ_SIZE;
2344 rec_seq = gcm_256_info->rec_seq;
2345 keysize = TLS_CIPHER_AES_GCM_256_KEY_SIZE;
2346 key = gcm_256_info->key;
2347 salt = gcm_256_info->salt;
2348 salt_size = TLS_CIPHER_AES_GCM_256_SALT_SIZE;
2349 cipher_name = "gcm(aes)";
2352 case TLS_CIPHER_AES_CCM_128: {
2353 struct tls12_crypto_info_aes_ccm_128 *ccm_128_info;
2355 ccm_128_info = (void *)crypto_info;
2356 nonce_size = TLS_CIPHER_AES_CCM_128_IV_SIZE;
2357 tag_size = TLS_CIPHER_AES_CCM_128_TAG_SIZE;
2358 iv_size = TLS_CIPHER_AES_CCM_128_IV_SIZE;
2359 iv = ccm_128_info->iv;
2360 rec_seq_size = TLS_CIPHER_AES_CCM_128_REC_SEQ_SIZE;
2361 rec_seq = ccm_128_info->rec_seq;
2362 keysize = TLS_CIPHER_AES_CCM_128_KEY_SIZE;
2363 key = ccm_128_info->key;
2364 salt = ccm_128_info->salt;
2365 salt_size = TLS_CIPHER_AES_CCM_128_SALT_SIZE;
2366 cipher_name = "ccm(aes)";
2369 case TLS_CIPHER_CHACHA20_POLY1305: {
2370 struct tls12_crypto_info_chacha20_poly1305 *chacha20_poly1305_info;
2372 chacha20_poly1305_info = (void *)crypto_info;
2374 tag_size = TLS_CIPHER_CHACHA20_POLY1305_TAG_SIZE;
2375 iv_size = TLS_CIPHER_CHACHA20_POLY1305_IV_SIZE;
2376 iv = chacha20_poly1305_info->iv;
2377 rec_seq_size = TLS_CIPHER_CHACHA20_POLY1305_REC_SEQ_SIZE;
2378 rec_seq = chacha20_poly1305_info->rec_seq;
2379 keysize = TLS_CIPHER_CHACHA20_POLY1305_KEY_SIZE;
2380 key = chacha20_poly1305_info->key;
2381 salt = chacha20_poly1305_info->salt;
2382 salt_size = TLS_CIPHER_CHACHA20_POLY1305_SALT_SIZE;
2383 cipher_name = "rfc7539(chacha20,poly1305)";
2386 case TLS_CIPHER_SM4_GCM: {
2387 struct tls12_crypto_info_sm4_gcm *sm4_gcm_info;
2389 sm4_gcm_info = (void *)crypto_info;
2390 nonce_size = TLS_CIPHER_SM4_GCM_IV_SIZE;
2391 tag_size = TLS_CIPHER_SM4_GCM_TAG_SIZE;
2392 iv_size = TLS_CIPHER_SM4_GCM_IV_SIZE;
2393 iv = sm4_gcm_info->iv;
2394 rec_seq_size = TLS_CIPHER_SM4_GCM_REC_SEQ_SIZE;
2395 rec_seq = sm4_gcm_info->rec_seq;
2396 keysize = TLS_CIPHER_SM4_GCM_KEY_SIZE;
2397 key = sm4_gcm_info->key;
2398 salt = sm4_gcm_info->salt;
2399 salt_size = TLS_CIPHER_SM4_GCM_SALT_SIZE;
2400 cipher_name = "gcm(sm4)";
2403 case TLS_CIPHER_SM4_CCM: {
2404 struct tls12_crypto_info_sm4_ccm *sm4_ccm_info;
2406 sm4_ccm_info = (void *)crypto_info;
2407 nonce_size = TLS_CIPHER_SM4_CCM_IV_SIZE;
2408 tag_size = TLS_CIPHER_SM4_CCM_TAG_SIZE;
2409 iv_size = TLS_CIPHER_SM4_CCM_IV_SIZE;
2410 iv = sm4_ccm_info->iv;
2411 rec_seq_size = TLS_CIPHER_SM4_CCM_REC_SEQ_SIZE;
2412 rec_seq = sm4_ccm_info->rec_seq;
2413 keysize = TLS_CIPHER_SM4_CCM_KEY_SIZE;
2414 key = sm4_ccm_info->key;
2415 salt = sm4_ccm_info->salt;
2416 salt_size = TLS_CIPHER_SM4_CCM_SALT_SIZE;
2417 cipher_name = "ccm(sm4)";
2425 /* Sanity-check the sizes for stack allocations. */
2426 if (iv_size > MAX_IV_SIZE || nonce_size > MAX_IV_SIZE ||
2427 rec_seq_size > TLS_MAX_REC_SEQ_SIZE || tag_size != TLS_TAG_SIZE) {
2432 if (crypto_info->version == TLS_1_3_VERSION) {
2434 prot->aad_size = TLS_HEADER_SIZE;
2435 prot->tail_size = 1;
2437 prot->aad_size = TLS_AAD_SPACE_SIZE;
2438 prot->tail_size = 0;
2441 prot->version = crypto_info->version;
2442 prot->cipher_type = crypto_info->cipher_type;
2443 prot->prepend_size = TLS_HEADER_SIZE + nonce_size;
2444 prot->tag_size = tag_size;
2445 prot->overhead_size = prot->prepend_size +
2446 prot->tag_size + prot->tail_size;
2447 prot->iv_size = iv_size;
2448 prot->salt_size = salt_size;
2449 cctx->iv = kmalloc(iv_size + salt_size, GFP_KERNEL);
2454 /* Note: 128 & 256 bit salt are the same size */
2455 prot->rec_seq_size = rec_seq_size;
2456 memcpy(cctx->iv, salt, salt_size);
2457 memcpy(cctx->iv + salt_size, iv, iv_size);
2458 cctx->rec_seq = kmemdup(rec_seq, rec_seq_size, GFP_KERNEL);
2459 if (!cctx->rec_seq) {
2465 *aead = crypto_alloc_aead(cipher_name, 0, 0);
2466 if (IS_ERR(*aead)) {
2467 rc = PTR_ERR(*aead);
2473 ctx->push_pending_record = tls_sw_push_pending_record;
2475 rc = crypto_aead_setkey(*aead, key, keysize);
2480 rc = crypto_aead_setauthsize(*aead, prot->tag_size);
2485 tfm = crypto_aead_tfm(sw_ctx_rx->aead_recv);
2487 if (crypto_info->version == TLS_1_3_VERSION)
2488 sw_ctx_rx->async_capable = 0;
2490 sw_ctx_rx->async_capable =
2491 !!(tfm->__crt_alg->cra_flags &
2494 /* Set up strparser */
2495 memset(&cb, 0, sizeof(cb));
2496 cb.rcv_msg = tls_queue;
2497 cb.parse_msg = tls_read_size;
2499 strp_init(&sw_ctx_rx->strp, sk, &cb);
2505 crypto_free_aead(*aead);
2508 kfree(cctx->rec_seq);
2509 cctx->rec_seq = NULL;
2515 kfree(ctx->priv_ctx_tx);
2516 ctx->priv_ctx_tx = NULL;
2518 kfree(ctx->priv_ctx_rx);
2519 ctx->priv_ctx_rx = NULL;