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 <crypto/aead.h>
42 #include <net/strparser.h>
45 static int __skb_nsg(struct sk_buff *skb, int offset, int len,
46 unsigned int recursion_level)
48 int start = skb_headlen(skb);
49 int i, chunk = start - offset;
50 struct sk_buff *frag_iter;
53 if (unlikely(recursion_level >= 24))
66 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
69 WARN_ON(start > offset + len);
71 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
85 if (unlikely(skb_has_frag_list(skb))) {
86 skb_walk_frags(skb, frag_iter) {
89 WARN_ON(start > offset + len);
91 end = start + frag_iter->len;
96 ret = __skb_nsg(frag_iter, offset - start, chunk,
98 if (unlikely(ret < 0))
113 /* Return the number of scatterlist elements required to completely map the
114 * skb, or -EMSGSIZE if the recursion depth is exceeded.
116 static int skb_nsg(struct sk_buff *skb, int offset, int len)
118 return __skb_nsg(skb, offset, len, 0);
121 static int padding_length(struct tls_sw_context_rx *ctx,
122 struct tls_prot_info *prot, struct sk_buff *skb)
124 struct strp_msg *rxm = strp_msg(skb);
127 /* Determine zero-padding length */
128 if (prot->version == TLS_1_3_VERSION) {
129 char content_type = 0;
133 while (content_type == 0) {
134 if (back > rxm->full_len - prot->prepend_size)
136 err = skb_copy_bits(skb,
137 rxm->offset + rxm->full_len - back,
146 ctx->control = content_type;
151 static void tls_decrypt_done(struct crypto_async_request *req, int err)
153 struct aead_request *aead_req = (struct aead_request *)req;
154 struct scatterlist *sgout = aead_req->dst;
155 struct scatterlist *sgin = aead_req->src;
156 struct tls_sw_context_rx *ctx;
157 struct tls_context *tls_ctx;
158 struct tls_prot_info *prot;
159 struct scatterlist *sg;
164 skb = (struct sk_buff *)req->data;
165 tls_ctx = tls_get_ctx(skb->sk);
166 ctx = tls_sw_ctx_rx(tls_ctx);
167 prot = &tls_ctx->prot_info;
169 /* Propagate if there was an err */
172 TLS_INC_STATS(sock_net(skb->sk),
173 LINUX_MIB_TLSDECRYPTERROR);
174 ctx->async_wait.err = err;
175 tls_err_abort(skb->sk, err);
177 struct strp_msg *rxm = strp_msg(skb);
180 pad = padding_length(ctx, prot, skb);
182 ctx->async_wait.err = pad;
183 tls_err_abort(skb->sk, pad);
185 rxm->full_len -= pad;
186 rxm->offset += prot->prepend_size;
187 rxm->full_len -= prot->overhead_size;
191 /* After using skb->sk to propagate sk through crypto async callback
192 * we need to NULL it again.
197 /* Free the destination pages if skb was not decrypted inplace */
199 /* Skip the first S/G entry as it points to AAD */
200 for_each_sg(sg_next(sgout), sg, UINT_MAX, pages) {
203 put_page(sg_page(sg));
209 pending = atomic_dec_return(&ctx->decrypt_pending);
211 if (!pending && READ_ONCE(ctx->async_notify))
212 complete(&ctx->async_wait.completion);
215 static int tls_do_decryption(struct sock *sk,
217 struct scatterlist *sgin,
218 struct scatterlist *sgout,
221 struct aead_request *aead_req,
224 struct tls_context *tls_ctx = tls_get_ctx(sk);
225 struct tls_prot_info *prot = &tls_ctx->prot_info;
226 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
229 aead_request_set_tfm(aead_req, ctx->aead_recv);
230 aead_request_set_ad(aead_req, prot->aad_size);
231 aead_request_set_crypt(aead_req, sgin, sgout,
232 data_len + prot->tag_size,
236 /* Using skb->sk to push sk through to crypto async callback
237 * handler. This allows propagating errors up to the socket
238 * if needed. It _must_ be cleared in the async handler
239 * before consume_skb is called. We _know_ skb->sk is NULL
240 * because it is a clone from strparser.
243 aead_request_set_callback(aead_req,
244 CRYPTO_TFM_REQ_MAY_BACKLOG,
245 tls_decrypt_done, skb);
246 atomic_inc(&ctx->decrypt_pending);
248 aead_request_set_callback(aead_req,
249 CRYPTO_TFM_REQ_MAY_BACKLOG,
250 crypto_req_done, &ctx->async_wait);
253 ret = crypto_aead_decrypt(aead_req);
254 if (ret == -EINPROGRESS) {
258 ret = crypto_wait_req(ret, &ctx->async_wait);
259 } else if (ret == -EBADMSG) {
260 TLS_INC_STATS(sock_net(sk), LINUX_MIB_TLSDECRYPTERROR);
264 atomic_dec(&ctx->decrypt_pending);
269 static void tls_trim_both_msgs(struct sock *sk, int target_size)
271 struct tls_context *tls_ctx = tls_get_ctx(sk);
272 struct tls_prot_info *prot = &tls_ctx->prot_info;
273 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
274 struct tls_rec *rec = ctx->open_rec;
276 sk_msg_trim(sk, &rec->msg_plaintext, target_size);
278 target_size += prot->overhead_size;
279 sk_msg_trim(sk, &rec->msg_encrypted, target_size);
282 static int tls_alloc_encrypted_msg(struct sock *sk, int len)
284 struct tls_context *tls_ctx = tls_get_ctx(sk);
285 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
286 struct tls_rec *rec = ctx->open_rec;
287 struct sk_msg *msg_en = &rec->msg_encrypted;
289 return sk_msg_alloc(sk, msg_en, len, 0);
292 static int tls_clone_plaintext_msg(struct sock *sk, int required)
294 struct tls_context *tls_ctx = tls_get_ctx(sk);
295 struct tls_prot_info *prot = &tls_ctx->prot_info;
296 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
297 struct tls_rec *rec = ctx->open_rec;
298 struct sk_msg *msg_pl = &rec->msg_plaintext;
299 struct sk_msg *msg_en = &rec->msg_encrypted;
302 /* We add page references worth len bytes from encrypted sg
303 * at the end of plaintext sg. It is guaranteed that msg_en
304 * has enough required room (ensured by caller).
306 len = required - msg_pl->sg.size;
308 /* Skip initial bytes in msg_en's data to be able to use
309 * same offset of both plain and encrypted data.
311 skip = prot->prepend_size + msg_pl->sg.size;
313 return sk_msg_clone(sk, msg_pl, msg_en, skip, len);
316 static struct tls_rec *tls_get_rec(struct sock *sk)
318 struct tls_context *tls_ctx = tls_get_ctx(sk);
319 struct tls_prot_info *prot = &tls_ctx->prot_info;
320 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
321 struct sk_msg *msg_pl, *msg_en;
325 mem_size = sizeof(struct tls_rec) + crypto_aead_reqsize(ctx->aead_send);
327 rec = kzalloc(mem_size, sk->sk_allocation);
331 msg_pl = &rec->msg_plaintext;
332 msg_en = &rec->msg_encrypted;
337 sg_init_table(rec->sg_aead_in, 2);
338 sg_set_buf(&rec->sg_aead_in[0], rec->aad_space, prot->aad_size);
339 sg_unmark_end(&rec->sg_aead_in[1]);
341 sg_init_table(rec->sg_aead_out, 2);
342 sg_set_buf(&rec->sg_aead_out[0], rec->aad_space, prot->aad_size);
343 sg_unmark_end(&rec->sg_aead_out[1]);
348 static void tls_free_rec(struct sock *sk, struct tls_rec *rec)
350 sk_msg_free(sk, &rec->msg_encrypted);
351 sk_msg_free(sk, &rec->msg_plaintext);
355 static void tls_free_open_rec(struct sock *sk)
357 struct tls_context *tls_ctx = tls_get_ctx(sk);
358 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
359 struct tls_rec *rec = ctx->open_rec;
362 tls_free_rec(sk, rec);
363 ctx->open_rec = NULL;
367 int tls_tx_records(struct sock *sk, int flags)
369 struct tls_context *tls_ctx = tls_get_ctx(sk);
370 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
371 struct tls_rec *rec, *tmp;
372 struct sk_msg *msg_en;
373 int tx_flags, rc = 0;
375 if (tls_is_partially_sent_record(tls_ctx)) {
376 rec = list_first_entry(&ctx->tx_list,
377 struct tls_rec, list);
380 tx_flags = rec->tx_flags;
384 rc = tls_push_partial_record(sk, tls_ctx, tx_flags);
388 /* Full record has been transmitted.
389 * Remove the head of tx_list
391 list_del(&rec->list);
392 sk_msg_free(sk, &rec->msg_plaintext);
396 /* Tx all ready records */
397 list_for_each_entry_safe(rec, tmp, &ctx->tx_list, list) {
398 if (READ_ONCE(rec->tx_ready)) {
400 tx_flags = rec->tx_flags;
404 msg_en = &rec->msg_encrypted;
405 rc = tls_push_sg(sk, tls_ctx,
406 &msg_en->sg.data[msg_en->sg.curr],
411 list_del(&rec->list);
412 sk_msg_free(sk, &rec->msg_plaintext);
420 if (rc < 0 && rc != -EAGAIN)
421 tls_err_abort(sk, EBADMSG);
426 static void tls_encrypt_done(struct crypto_async_request *req, int err)
428 struct aead_request *aead_req = (struct aead_request *)req;
429 struct sock *sk = req->data;
430 struct tls_context *tls_ctx = tls_get_ctx(sk);
431 struct tls_prot_info *prot = &tls_ctx->prot_info;
432 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
433 struct scatterlist *sge;
434 struct sk_msg *msg_en;
439 rec = container_of(aead_req, struct tls_rec, aead_req);
440 msg_en = &rec->msg_encrypted;
442 sge = sk_msg_elem(msg_en, msg_en->sg.curr);
443 sge->offset -= prot->prepend_size;
444 sge->length += prot->prepend_size;
446 /* Check if error is previously set on socket */
447 if (err || sk->sk_err) {
450 /* If err is already set on socket, return the same code */
452 ctx->async_wait.err = sk->sk_err;
454 ctx->async_wait.err = err;
455 tls_err_abort(sk, err);
460 struct tls_rec *first_rec;
462 /* Mark the record as ready for transmission */
463 smp_store_mb(rec->tx_ready, true);
465 /* If received record is at head of tx_list, schedule tx */
466 first_rec = list_first_entry(&ctx->tx_list,
467 struct tls_rec, list);
468 if (rec == first_rec)
472 pending = atomic_dec_return(&ctx->encrypt_pending);
474 if (!pending && READ_ONCE(ctx->async_notify))
475 complete(&ctx->async_wait.completion);
480 /* Schedule the transmission */
481 if (!test_and_set_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask))
482 schedule_delayed_work(&ctx->tx_work.work, 1);
485 static int tls_do_encryption(struct sock *sk,
486 struct tls_context *tls_ctx,
487 struct tls_sw_context_tx *ctx,
488 struct aead_request *aead_req,
489 size_t data_len, u32 start)
491 struct tls_prot_info *prot = &tls_ctx->prot_info;
492 struct tls_rec *rec = ctx->open_rec;
493 struct sk_msg *msg_en = &rec->msg_encrypted;
494 struct scatterlist *sge = sk_msg_elem(msg_en, start);
495 int rc, iv_offset = 0;
497 /* For CCM based ciphers, first byte of IV is a constant */
498 if (prot->cipher_type == TLS_CIPHER_AES_CCM_128) {
499 rec->iv_data[0] = TLS_AES_CCM_IV_B0_BYTE;
503 memcpy(&rec->iv_data[iv_offset], tls_ctx->tx.iv,
504 prot->iv_size + prot->salt_size);
506 xor_iv_with_seq(prot->version, rec->iv_data, tls_ctx->tx.rec_seq);
508 sge->offset += prot->prepend_size;
509 sge->length -= prot->prepend_size;
511 msg_en->sg.curr = start;
513 aead_request_set_tfm(aead_req, ctx->aead_send);
514 aead_request_set_ad(aead_req, prot->aad_size);
515 aead_request_set_crypt(aead_req, rec->sg_aead_in,
517 data_len, rec->iv_data);
519 aead_request_set_callback(aead_req, CRYPTO_TFM_REQ_MAY_BACKLOG,
520 tls_encrypt_done, sk);
522 /* Add the record in tx_list */
523 list_add_tail((struct list_head *)&rec->list, &ctx->tx_list);
524 atomic_inc(&ctx->encrypt_pending);
526 rc = crypto_aead_encrypt(aead_req);
527 if (!rc || rc != -EINPROGRESS) {
528 atomic_dec(&ctx->encrypt_pending);
529 sge->offset -= prot->prepend_size;
530 sge->length += prot->prepend_size;
534 WRITE_ONCE(rec->tx_ready, true);
535 } else if (rc != -EINPROGRESS) {
536 list_del(&rec->list);
540 /* Unhook the record from context if encryption is not failure */
541 ctx->open_rec = NULL;
542 tls_advance_record_sn(sk, prot, &tls_ctx->tx);
546 static int tls_split_open_record(struct sock *sk, struct tls_rec *from,
547 struct tls_rec **to, struct sk_msg *msg_opl,
548 struct sk_msg *msg_oen, u32 split_point,
549 u32 tx_overhead_size, u32 *orig_end)
551 u32 i, j, bytes = 0, apply = msg_opl->apply_bytes;
552 struct scatterlist *sge, *osge, *nsge;
553 u32 orig_size = msg_opl->sg.size;
554 struct scatterlist tmp = { };
555 struct sk_msg *msg_npl;
559 new = tls_get_rec(sk);
562 ret = sk_msg_alloc(sk, &new->msg_encrypted, msg_opl->sg.size +
563 tx_overhead_size, 0);
565 tls_free_rec(sk, new);
569 *orig_end = msg_opl->sg.end;
570 i = msg_opl->sg.start;
571 sge = sk_msg_elem(msg_opl, i);
572 while (apply && sge->length) {
573 if (sge->length > apply) {
574 u32 len = sge->length - apply;
576 get_page(sg_page(sge));
577 sg_set_page(&tmp, sg_page(sge), len,
578 sge->offset + apply);
583 apply -= sge->length;
584 bytes += sge->length;
587 sk_msg_iter_var_next(i);
588 if (i == msg_opl->sg.end)
590 sge = sk_msg_elem(msg_opl, i);
594 msg_opl->sg.curr = i;
595 msg_opl->sg.copybreak = 0;
596 msg_opl->apply_bytes = 0;
597 msg_opl->sg.size = bytes;
599 msg_npl = &new->msg_plaintext;
600 msg_npl->apply_bytes = apply;
601 msg_npl->sg.size = orig_size - bytes;
603 j = msg_npl->sg.start;
604 nsge = sk_msg_elem(msg_npl, j);
606 memcpy(nsge, &tmp, sizeof(*nsge));
607 sk_msg_iter_var_next(j);
608 nsge = sk_msg_elem(msg_npl, j);
611 osge = sk_msg_elem(msg_opl, i);
612 while (osge->length) {
613 memcpy(nsge, osge, sizeof(*nsge));
615 sk_msg_iter_var_next(i);
616 sk_msg_iter_var_next(j);
619 osge = sk_msg_elem(msg_opl, i);
620 nsge = sk_msg_elem(msg_npl, j);
624 msg_npl->sg.curr = j;
625 msg_npl->sg.copybreak = 0;
631 static void tls_merge_open_record(struct sock *sk, struct tls_rec *to,
632 struct tls_rec *from, u32 orig_end)
634 struct sk_msg *msg_npl = &from->msg_plaintext;
635 struct sk_msg *msg_opl = &to->msg_plaintext;
636 struct scatterlist *osge, *nsge;
640 sk_msg_iter_var_prev(i);
641 j = msg_npl->sg.start;
643 osge = sk_msg_elem(msg_opl, i);
644 nsge = sk_msg_elem(msg_npl, j);
646 if (sg_page(osge) == sg_page(nsge) &&
647 osge->offset + osge->length == nsge->offset) {
648 osge->length += nsge->length;
649 put_page(sg_page(nsge));
652 msg_opl->sg.end = orig_end;
653 msg_opl->sg.curr = orig_end;
654 msg_opl->sg.copybreak = 0;
655 msg_opl->apply_bytes = msg_opl->sg.size + msg_npl->sg.size;
656 msg_opl->sg.size += msg_npl->sg.size;
658 sk_msg_free(sk, &to->msg_encrypted);
659 sk_msg_xfer_full(&to->msg_encrypted, &from->msg_encrypted);
664 static int tls_push_record(struct sock *sk, int flags,
665 unsigned char record_type)
667 struct tls_context *tls_ctx = tls_get_ctx(sk);
668 struct tls_prot_info *prot = &tls_ctx->prot_info;
669 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
670 struct tls_rec *rec = ctx->open_rec, *tmp = NULL;
671 u32 i, split_point, uninitialized_var(orig_end);
672 struct sk_msg *msg_pl, *msg_en;
673 struct aead_request *req;
680 msg_pl = &rec->msg_plaintext;
681 msg_en = &rec->msg_encrypted;
683 split_point = msg_pl->apply_bytes;
684 split = split_point && split_point < msg_pl->sg.size;
686 rc = tls_split_open_record(sk, rec, &tmp, msg_pl, msg_en,
687 split_point, prot->overhead_size,
691 sk_msg_trim(sk, msg_en, msg_pl->sg.size +
692 prot->overhead_size);
695 rec->tx_flags = flags;
696 req = &rec->aead_req;
699 sk_msg_iter_var_prev(i);
701 rec->content_type = record_type;
702 if (prot->version == TLS_1_3_VERSION) {
703 /* Add content type to end of message. No padding added */
704 sg_set_buf(&rec->sg_content_type, &rec->content_type, 1);
705 sg_mark_end(&rec->sg_content_type);
706 sg_chain(msg_pl->sg.data, msg_pl->sg.end + 1,
707 &rec->sg_content_type);
709 sg_mark_end(sk_msg_elem(msg_pl, i));
712 i = msg_pl->sg.start;
713 sg_chain(rec->sg_aead_in, 2, rec->inplace_crypto ?
714 &msg_en->sg.data[i] : &msg_pl->sg.data[i]);
717 sk_msg_iter_var_prev(i);
718 sg_mark_end(sk_msg_elem(msg_en, i));
720 i = msg_en->sg.start;
721 sg_chain(rec->sg_aead_out, 2, &msg_en->sg.data[i]);
723 tls_make_aad(rec->aad_space, msg_pl->sg.size + prot->tail_size,
724 tls_ctx->tx.rec_seq, prot->rec_seq_size,
725 record_type, prot->version);
727 tls_fill_prepend(tls_ctx,
728 page_address(sg_page(&msg_en->sg.data[i])) +
729 msg_en->sg.data[i].offset,
730 msg_pl->sg.size + prot->tail_size,
731 record_type, prot->version);
733 tls_ctx->pending_open_record_frags = false;
735 rc = tls_do_encryption(sk, tls_ctx, ctx, req,
736 msg_pl->sg.size + prot->tail_size, i);
738 if (rc != -EINPROGRESS) {
739 tls_err_abort(sk, EBADMSG);
741 tls_ctx->pending_open_record_frags = true;
742 tls_merge_open_record(sk, rec, tmp, orig_end);
745 ctx->async_capable = 1;
748 msg_pl = &tmp->msg_plaintext;
749 msg_en = &tmp->msg_encrypted;
750 sk_msg_trim(sk, msg_en, msg_pl->sg.size + prot->overhead_size);
751 tls_ctx->pending_open_record_frags = true;
755 return tls_tx_records(sk, flags);
758 static int bpf_exec_tx_verdict(struct sk_msg *msg, struct sock *sk,
759 bool full_record, u8 record_type,
760 size_t *copied, int flags)
762 struct tls_context *tls_ctx = tls_get_ctx(sk);
763 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
764 struct sk_msg msg_redir = { };
765 struct sk_psock *psock;
766 struct sock *sk_redir;
772 policy = !(flags & MSG_SENDPAGE_NOPOLICY);
773 psock = sk_psock_get(sk);
774 if (!psock || !policy) {
775 err = tls_push_record(sk, flags, record_type);
777 *copied -= sk_msg_free(sk, msg);
778 tls_free_open_rec(sk);
783 enospc = sk_msg_full(msg);
784 if (psock->eval == __SK_NONE) {
785 delta = msg->sg.size;
786 psock->eval = sk_psock_msg_verdict(sk, psock, msg);
787 if (delta < msg->sg.size)
788 delta -= msg->sg.size;
792 if (msg->cork_bytes && msg->cork_bytes > msg->sg.size &&
793 !enospc && !full_record) {
799 if (msg->apply_bytes && msg->apply_bytes < send)
800 send = msg->apply_bytes;
802 switch (psock->eval) {
804 err = tls_push_record(sk, flags, record_type);
806 *copied -= sk_msg_free(sk, msg);
807 tls_free_open_rec(sk);
812 sk_redir = psock->sk_redir;
813 memcpy(&msg_redir, msg, sizeof(*msg));
814 if (msg->apply_bytes < send)
815 msg->apply_bytes = 0;
817 msg->apply_bytes -= send;
818 sk_msg_return_zero(sk, msg, send);
819 msg->sg.size -= send;
821 err = tcp_bpf_sendmsg_redir(sk_redir, &msg_redir, send, flags);
824 *copied -= sk_msg_free_nocharge(sk, &msg_redir);
827 if (msg->sg.size == 0)
828 tls_free_open_rec(sk);
832 sk_msg_free_partial(sk, msg, send);
833 if (msg->apply_bytes < send)
834 msg->apply_bytes = 0;
836 msg->apply_bytes -= send;
837 if (msg->sg.size == 0)
838 tls_free_open_rec(sk);
839 *copied -= (send + delta);
844 bool reset_eval = !ctx->open_rec;
848 msg = &rec->msg_plaintext;
849 if (!msg->apply_bytes)
853 psock->eval = __SK_NONE;
854 if (psock->sk_redir) {
855 sock_put(psock->sk_redir);
856 psock->sk_redir = NULL;
863 sk_psock_put(sk, psock);
867 static int tls_sw_push_pending_record(struct sock *sk, int flags)
869 struct tls_context *tls_ctx = tls_get_ctx(sk);
870 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
871 struct tls_rec *rec = ctx->open_rec;
872 struct sk_msg *msg_pl;
878 msg_pl = &rec->msg_plaintext;
879 copied = msg_pl->sg.size;
883 return bpf_exec_tx_verdict(msg_pl, sk, true, TLS_RECORD_TYPE_DATA,
887 int tls_sw_sendmsg(struct sock *sk, struct msghdr *msg, size_t size)
889 long timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
890 struct tls_context *tls_ctx = tls_get_ctx(sk);
891 struct tls_prot_info *prot = &tls_ctx->prot_info;
892 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
893 bool async_capable = ctx->async_capable;
894 unsigned char record_type = TLS_RECORD_TYPE_DATA;
895 bool is_kvec = iov_iter_is_kvec(&msg->msg_iter);
896 bool eor = !(msg->msg_flags & MSG_MORE);
897 size_t try_to_copy, copied = 0;
898 struct sk_msg *msg_pl, *msg_en;
908 if (msg->msg_flags & ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL))
911 mutex_lock(&tls_ctx->tx_lock);
914 if (unlikely(msg->msg_controllen)) {
915 ret = tls_proccess_cmsg(sk, msg, &record_type);
917 if (ret == -EINPROGRESS)
919 else if (ret != -EAGAIN)
924 while (msg_data_left(msg)) {
933 rec = ctx->open_rec = tls_get_rec(sk);
939 msg_pl = &rec->msg_plaintext;
940 msg_en = &rec->msg_encrypted;
942 orig_size = msg_pl->sg.size;
944 try_to_copy = msg_data_left(msg);
945 record_room = TLS_MAX_PAYLOAD_SIZE - msg_pl->sg.size;
946 if (try_to_copy >= record_room) {
947 try_to_copy = record_room;
951 required_size = msg_pl->sg.size + try_to_copy +
954 if (!sk_stream_memory_free(sk))
955 goto wait_for_sndbuf;
958 ret = tls_alloc_encrypted_msg(sk, required_size);
961 goto wait_for_memory;
963 /* Adjust try_to_copy according to the amount that was
964 * actually allocated. The difference is due
965 * to max sg elements limit
967 try_to_copy -= required_size - msg_en->sg.size;
971 if (!is_kvec && (full_record || eor) && !async_capable) {
972 u32 first = msg_pl->sg.end;
974 ret = sk_msg_zerocopy_from_iter(sk, &msg->msg_iter,
975 msg_pl, try_to_copy);
977 goto fallback_to_reg_send;
979 rec->inplace_crypto = 0;
982 copied += try_to_copy;
984 sk_msg_sg_copy_set(msg_pl, first);
985 ret = bpf_exec_tx_verdict(msg_pl, sk, full_record,
986 record_type, &copied,
989 if (ret == -EINPROGRESS)
991 else if (ret == -ENOMEM)
992 goto wait_for_memory;
993 else if (ctx->open_rec && ret == -ENOSPC)
995 else if (ret != -EAGAIN)
1000 copied -= try_to_copy;
1001 sk_msg_sg_copy_clear(msg_pl, first);
1002 iov_iter_revert(&msg->msg_iter,
1003 msg_pl->sg.size - orig_size);
1004 fallback_to_reg_send:
1005 sk_msg_trim(sk, msg_pl, orig_size);
1008 required_size = msg_pl->sg.size + try_to_copy;
1010 ret = tls_clone_plaintext_msg(sk, required_size);
1015 /* Adjust try_to_copy according to the amount that was
1016 * actually allocated. The difference is due
1017 * to max sg elements limit
1019 try_to_copy -= required_size - msg_pl->sg.size;
1021 sk_msg_trim(sk, msg_en,
1022 msg_pl->sg.size + prot->overhead_size);
1026 ret = sk_msg_memcopy_from_iter(sk, &msg->msg_iter,
1027 msg_pl, try_to_copy);
1032 /* Open records defined only if successfully copied, otherwise
1033 * we would trim the sg but not reset the open record frags.
1035 tls_ctx->pending_open_record_frags = true;
1036 copied += try_to_copy;
1037 if (full_record || eor) {
1038 ret = bpf_exec_tx_verdict(msg_pl, sk, full_record,
1039 record_type, &copied,
1042 if (ret == -EINPROGRESS)
1044 else if (ret == -ENOMEM)
1045 goto wait_for_memory;
1046 else if (ret != -EAGAIN) {
1057 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1059 ret = sk_stream_wait_memory(sk, &timeo);
1063 tls_trim_both_msgs(sk, orig_size);
1067 if (ctx->open_rec && msg_en->sg.size < required_size)
1068 goto alloc_encrypted;
1073 } else if (num_zc) {
1074 /* Wait for pending encryptions to get completed */
1075 smp_store_mb(ctx->async_notify, true);
1077 if (atomic_read(&ctx->encrypt_pending))
1078 crypto_wait_req(-EINPROGRESS, &ctx->async_wait);
1080 reinit_completion(&ctx->async_wait.completion);
1082 WRITE_ONCE(ctx->async_notify, false);
1084 if (ctx->async_wait.err) {
1085 ret = ctx->async_wait.err;
1090 /* Transmit if any encryptions have completed */
1091 if (test_and_clear_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask)) {
1092 cancel_delayed_work(&ctx->tx_work.work);
1093 tls_tx_records(sk, msg->msg_flags);
1097 ret = sk_stream_error(sk, msg->msg_flags, ret);
1100 mutex_unlock(&tls_ctx->tx_lock);
1101 return copied ? copied : ret;
1104 static int tls_sw_do_sendpage(struct sock *sk, struct page *page,
1105 int offset, size_t size, int flags)
1107 long timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT);
1108 struct tls_context *tls_ctx = tls_get_ctx(sk);
1109 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
1110 struct tls_prot_info *prot = &tls_ctx->prot_info;
1111 unsigned char record_type = TLS_RECORD_TYPE_DATA;
1112 struct sk_msg *msg_pl;
1113 struct tls_rec *rec;
1121 eor = !(flags & (MSG_MORE | MSG_SENDPAGE_NOTLAST));
1122 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
1124 /* Call the sk_stream functions to manage the sndbuf mem. */
1126 size_t copy, required_size;
1134 rec = ctx->open_rec;
1136 rec = ctx->open_rec = tls_get_rec(sk);
1142 msg_pl = &rec->msg_plaintext;
1144 full_record = false;
1145 record_room = TLS_MAX_PAYLOAD_SIZE - msg_pl->sg.size;
1147 if (copy >= record_room) {
1152 required_size = msg_pl->sg.size + copy + prot->overhead_size;
1154 if (!sk_stream_memory_free(sk))
1155 goto wait_for_sndbuf;
1157 ret = tls_alloc_encrypted_msg(sk, required_size);
1160 goto wait_for_memory;
1162 /* Adjust copy according to the amount that was
1163 * actually allocated. The difference is due
1164 * to max sg elements limit
1166 copy -= required_size - msg_pl->sg.size;
1170 sk_msg_page_add(msg_pl, page, copy, offset);
1171 sk_mem_charge(sk, copy);
1177 tls_ctx->pending_open_record_frags = true;
1178 if (full_record || eor || sk_msg_full(msg_pl)) {
1179 rec->inplace_crypto = 0;
1180 ret = bpf_exec_tx_verdict(msg_pl, sk, full_record,
1181 record_type, &copied, flags);
1183 if (ret == -EINPROGRESS)
1185 else if (ret == -ENOMEM)
1186 goto wait_for_memory;
1187 else if (ret != -EAGAIN) {
1196 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1198 ret = sk_stream_wait_memory(sk, &timeo);
1201 tls_trim_both_msgs(sk, msg_pl->sg.size);
1210 /* Transmit if any encryptions have completed */
1211 if (test_and_clear_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask)) {
1212 cancel_delayed_work(&ctx->tx_work.work);
1213 tls_tx_records(sk, flags);
1217 ret = sk_stream_error(sk, flags, ret);
1218 return copied ? copied : ret;
1221 int tls_sw_sendpage_locked(struct sock *sk, struct page *page,
1222 int offset, size_t size, int flags)
1224 if (flags & ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL |
1225 MSG_SENDPAGE_NOTLAST | MSG_SENDPAGE_NOPOLICY |
1226 MSG_NO_SHARED_FRAGS))
1229 return tls_sw_do_sendpage(sk, page, offset, size, flags);
1232 int tls_sw_sendpage(struct sock *sk, struct page *page,
1233 int offset, size_t size, int flags)
1235 struct tls_context *tls_ctx = tls_get_ctx(sk);
1238 if (flags & ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL |
1239 MSG_SENDPAGE_NOTLAST | MSG_SENDPAGE_NOPOLICY))
1242 mutex_lock(&tls_ctx->tx_lock);
1244 ret = tls_sw_do_sendpage(sk, page, offset, size, flags);
1246 mutex_unlock(&tls_ctx->tx_lock);
1250 static struct sk_buff *tls_wait_data(struct sock *sk, struct sk_psock *psock,
1251 int flags, long timeo, int *err)
1253 struct tls_context *tls_ctx = tls_get_ctx(sk);
1254 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1255 struct sk_buff *skb;
1256 DEFINE_WAIT_FUNC(wait, woken_wake_function);
1258 while (!(skb = ctx->recv_pkt) && sk_psock_queue_empty(psock)) {
1260 *err = sock_error(sk);
1264 if (sk->sk_shutdown & RCV_SHUTDOWN)
1267 if (sock_flag(sk, SOCK_DONE))
1270 if ((flags & MSG_DONTWAIT) || !timeo) {
1275 add_wait_queue(sk_sleep(sk), &wait);
1276 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
1277 sk_wait_event(sk, &timeo,
1278 ctx->recv_pkt != skb ||
1279 !sk_psock_queue_empty(psock),
1281 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
1282 remove_wait_queue(sk_sleep(sk), &wait);
1284 /* Handle signals */
1285 if (signal_pending(current)) {
1286 *err = sock_intr_errno(timeo);
1294 static int tls_setup_from_iter(struct sock *sk, struct iov_iter *from,
1295 int length, int *pages_used,
1296 unsigned int *size_used,
1297 struct scatterlist *to,
1300 int rc = 0, i = 0, num_elem = *pages_used, maxpages;
1301 struct page *pages[MAX_SKB_FRAGS];
1302 unsigned int size = *size_used;
1303 ssize_t copied, use;
1306 while (length > 0) {
1308 maxpages = to_max_pages - num_elem;
1309 if (maxpages == 0) {
1313 copied = iov_iter_get_pages(from, pages,
1321 iov_iter_advance(from, copied);
1326 use = min_t(int, copied, PAGE_SIZE - offset);
1328 sg_set_page(&to[num_elem],
1329 pages[i], use, offset);
1330 sg_unmark_end(&to[num_elem]);
1331 /* We do not uncharge memory from this API */
1340 /* Mark the end in the last sg entry if newly added */
1341 if (num_elem > *pages_used)
1342 sg_mark_end(&to[num_elem - 1]);
1345 iov_iter_revert(from, size - *size_used);
1347 *pages_used = num_elem;
1352 /* This function decrypts the input skb into either out_iov or in out_sg
1353 * or in skb buffers itself. The input parameter 'zc' indicates if
1354 * zero-copy mode needs to be tried or not. With zero-copy mode, either
1355 * out_iov or out_sg must be non-NULL. In case both out_iov and out_sg are
1356 * NULL, then the decryption happens inside skb buffers itself, i.e.
1357 * zero-copy gets disabled and 'zc' is updated.
1360 static int decrypt_internal(struct sock *sk, struct sk_buff *skb,
1361 struct iov_iter *out_iov,
1362 struct scatterlist *out_sg,
1363 int *chunk, bool *zc, bool async)
1365 struct tls_context *tls_ctx = tls_get_ctx(sk);
1366 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1367 struct tls_prot_info *prot = &tls_ctx->prot_info;
1368 struct strp_msg *rxm = strp_msg(skb);
1369 int n_sgin, n_sgout, nsg, mem_size, aead_size, err, pages = 0;
1370 struct aead_request *aead_req;
1371 struct sk_buff *unused;
1372 u8 *aad, *iv, *mem = NULL;
1373 struct scatterlist *sgin = NULL;
1374 struct scatterlist *sgout = NULL;
1375 const int data_len = rxm->full_len - prot->overhead_size +
1379 if (*zc && (out_iov || out_sg)) {
1381 n_sgout = iov_iter_npages(out_iov, INT_MAX) + 1;
1383 n_sgout = sg_nents(out_sg);
1384 n_sgin = skb_nsg(skb, rxm->offset + prot->prepend_size,
1385 rxm->full_len - prot->prepend_size);
1389 n_sgin = skb_cow_data(skb, 0, &unused);
1395 /* Increment to accommodate AAD */
1396 n_sgin = n_sgin + 1;
1398 nsg = n_sgin + n_sgout;
1400 aead_size = sizeof(*aead_req) + crypto_aead_reqsize(ctx->aead_recv);
1401 mem_size = aead_size + (nsg * sizeof(struct scatterlist));
1402 mem_size = mem_size + prot->aad_size;
1403 mem_size = mem_size + crypto_aead_ivsize(ctx->aead_recv);
1405 /* Allocate a single block of memory which contains
1406 * aead_req || sgin[] || sgout[] || aad || iv.
1407 * This order achieves correct alignment for aead_req, sgin, sgout.
1409 mem = kmalloc(mem_size, sk->sk_allocation);
1413 /* Segment the allocated memory */
1414 aead_req = (struct aead_request *)mem;
1415 sgin = (struct scatterlist *)(mem + aead_size);
1416 sgout = sgin + n_sgin;
1417 aad = (u8 *)(sgout + n_sgout);
1418 iv = aad + prot->aad_size;
1420 /* For CCM based ciphers, first byte of nonce+iv is always '2' */
1421 if (prot->cipher_type == TLS_CIPHER_AES_CCM_128) {
1427 err = skb_copy_bits(skb, rxm->offset + TLS_HEADER_SIZE,
1428 iv + iv_offset + prot->salt_size,
1434 if (prot->version == TLS_1_3_VERSION)
1435 memcpy(iv + iv_offset, tls_ctx->rx.iv,
1436 crypto_aead_ivsize(ctx->aead_recv));
1438 memcpy(iv + iv_offset, tls_ctx->rx.iv, prot->salt_size);
1440 xor_iv_with_seq(prot->version, iv, tls_ctx->rx.rec_seq);
1443 tls_make_aad(aad, rxm->full_len - prot->overhead_size +
1445 tls_ctx->rx.rec_seq, prot->rec_seq_size,
1446 ctx->control, prot->version);
1449 sg_init_table(sgin, n_sgin);
1450 sg_set_buf(&sgin[0], aad, prot->aad_size);
1451 err = skb_to_sgvec(skb, &sgin[1],
1452 rxm->offset + prot->prepend_size,
1453 rxm->full_len - prot->prepend_size);
1461 sg_init_table(sgout, n_sgout);
1462 sg_set_buf(&sgout[0], aad, prot->aad_size);
1465 err = tls_setup_from_iter(sk, out_iov, data_len,
1466 &pages, chunk, &sgout[1],
1469 goto fallback_to_reg_recv;
1470 } else if (out_sg) {
1471 memcpy(sgout, out_sg, n_sgout * sizeof(*sgout));
1473 goto fallback_to_reg_recv;
1476 fallback_to_reg_recv:
1483 /* Prepare and submit AEAD request */
1484 err = tls_do_decryption(sk, skb, sgin, sgout, iv,
1485 data_len, aead_req, async);
1486 if (err == -EINPROGRESS)
1489 /* Release the pages in case iov was mapped to pages */
1490 for (; pages > 0; pages--)
1491 put_page(sg_page(&sgout[pages]));
1497 static int decrypt_skb_update(struct sock *sk, struct sk_buff *skb,
1498 struct iov_iter *dest, int *chunk, bool *zc,
1501 struct tls_context *tls_ctx = tls_get_ctx(sk);
1502 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1503 struct tls_prot_info *prot = &tls_ctx->prot_info;
1504 struct strp_msg *rxm = strp_msg(skb);
1507 if (!ctx->decrypted) {
1508 if (tls_ctx->rx_conf == TLS_HW) {
1509 err = tls_device_decrypted(sk, tls_ctx, skb, rxm);
1514 /* Still not decrypted after tls_device */
1515 if (!ctx->decrypted) {
1516 err = decrypt_internal(sk, skb, dest, NULL, chunk, zc,
1519 if (err == -EINPROGRESS)
1520 tls_advance_record_sn(sk, prot,
1529 pad = padding_length(ctx, prot, skb);
1533 rxm->full_len -= pad;
1534 rxm->offset += prot->prepend_size;
1535 rxm->full_len -= prot->overhead_size;
1536 tls_advance_record_sn(sk, prot, &tls_ctx->rx);
1538 ctx->saved_data_ready(sk);
1546 int decrypt_skb(struct sock *sk, struct sk_buff *skb,
1547 struct scatterlist *sgout)
1552 return decrypt_internal(sk, skb, NULL, sgout, &chunk, &zc, false);
1555 static bool tls_sw_advance_skb(struct sock *sk, struct sk_buff *skb,
1558 struct tls_context *tls_ctx = tls_get_ctx(sk);
1559 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1562 struct strp_msg *rxm = strp_msg(skb);
1564 if (len < rxm->full_len) {
1566 rxm->full_len -= len;
1572 /* Finished with message */
1573 ctx->recv_pkt = NULL;
1574 __strp_unpause(&ctx->strp);
1579 /* This function traverses the rx_list in tls receive context to copies the
1580 * decrypted records into the buffer provided by caller zero copy is not
1581 * true. Further, the records are removed from the rx_list if it is not a peek
1582 * case and the record has been consumed completely.
1584 static int process_rx_list(struct tls_sw_context_rx *ctx,
1593 struct sk_buff *skb = skb_peek(&ctx->rx_list);
1596 struct tls_msg *tlm;
1599 /* Set the record type in 'control' if caller didn't pass it */
1602 ctrl = tlm->control;
1605 while (skip && skb) {
1606 struct strp_msg *rxm = strp_msg(skb);
1609 /* Cannot process a record of different type */
1610 if (ctrl != tlm->control)
1613 if (skip < rxm->full_len)
1616 skip = skip - rxm->full_len;
1617 skb = skb_peek_next(skb, &ctx->rx_list);
1620 while (len && skb) {
1621 struct sk_buff *next_skb;
1622 struct strp_msg *rxm = strp_msg(skb);
1623 int chunk = min_t(unsigned int, rxm->full_len - skip, len);
1627 /* Cannot process a record of different type */
1628 if (ctrl != tlm->control)
1631 /* Set record type if not already done. For a non-data record,
1632 * do not proceed if record type could not be copied.
1635 int cerr = put_cmsg(msg, SOL_TLS, TLS_GET_RECORD_TYPE,
1636 sizeof(ctrl), &ctrl);
1638 if (ctrl != TLS_RECORD_TYPE_DATA) {
1639 if (cerr || msg->msg_flags & MSG_CTRUNC)
1646 if (!zc || (rxm->full_len - skip) > len) {
1647 int err = skb_copy_datagram_msg(skb, rxm->offset + skip,
1654 copied = copied + chunk;
1656 /* Consume the data from record if it is non-peek case*/
1658 rxm->offset = rxm->offset + chunk;
1659 rxm->full_len = rxm->full_len - chunk;
1661 /* Return if there is unconsumed data in the record */
1662 if (rxm->full_len - skip)
1666 /* The remaining skip-bytes must lie in 1st record in rx_list.
1667 * So from the 2nd record, 'skip' should be 0.
1672 msg->msg_flags |= MSG_EOR;
1674 next_skb = skb_peek_next(skb, &ctx->rx_list);
1677 skb_unlink(skb, &ctx->rx_list);
1688 int tls_sw_recvmsg(struct sock *sk,
1695 struct tls_context *tls_ctx = tls_get_ctx(sk);
1696 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1697 struct tls_prot_info *prot = &tls_ctx->prot_info;
1698 struct sk_psock *psock;
1699 unsigned char control = 0;
1700 ssize_t decrypted = 0;
1701 struct strp_msg *rxm;
1702 struct tls_msg *tlm;
1703 struct sk_buff *skb;
1706 int target, err = 0;
1708 bool is_kvec = iov_iter_is_kvec(&msg->msg_iter);
1709 bool is_peek = flags & MSG_PEEK;
1714 if (unlikely(flags & MSG_ERRQUEUE))
1715 return sock_recv_errqueue(sk, msg, len, SOL_IP, IP_RECVERR);
1717 psock = sk_psock_get(sk);
1720 /* Process pending decrypted records. It must be non-zero-copy */
1721 err = process_rx_list(ctx, msg, &control, &cmsg, 0, len, false,
1724 tls_err_abort(sk, err);
1733 target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
1735 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1737 while (len && (decrypted + copied < target || ctx->recv_pkt)) {
1738 bool retain_skb = false;
1745 skb = tls_wait_data(sk, psock, flags, timeo, &err);
1748 int ret = __tcp_bpf_recvmsg(sk, psock,
1760 if (prot->version == TLS_1_3_VERSION)
1763 tlm->control = ctx->control;
1766 rxm = strp_msg(skb);
1768 to_decrypt = rxm->full_len - prot->overhead_size;
1770 if (to_decrypt <= len && !is_kvec && !is_peek &&
1771 ctx->control == TLS_RECORD_TYPE_DATA &&
1772 prot->version != TLS_1_3_VERSION)
1775 /* Do not use async mode if record is non-data */
1776 if (ctx->control == TLS_RECORD_TYPE_DATA)
1777 async_capable = ctx->async_capable;
1779 async_capable = false;
1781 err = decrypt_skb_update(sk, skb, &msg->msg_iter,
1782 &chunk, &zc, async_capable);
1783 if (err < 0 && err != -EINPROGRESS) {
1784 tls_err_abort(sk, EBADMSG);
1788 if (err == -EINPROGRESS) {
1791 } else if (prot->version == TLS_1_3_VERSION) {
1792 tlm->control = ctx->control;
1795 /* If the type of records being processed is not known yet,
1796 * set it to record type just dequeued. If it is already known,
1797 * but does not match the record type just dequeued, go to end.
1798 * We always get record type here since for tls1.2, record type
1799 * is known just after record is dequeued from stream parser.
1800 * For tls1.3, we disable async.
1804 control = tlm->control;
1805 else if (control != tlm->control)
1811 cerr = put_cmsg(msg, SOL_TLS, TLS_GET_RECORD_TYPE,
1812 sizeof(control), &control);
1814 if (control != TLS_RECORD_TYPE_DATA) {
1815 if (cerr || msg->msg_flags & MSG_CTRUNC) {
1823 goto pick_next_record;
1826 if (rxm->full_len > len) {
1830 chunk = rxm->full_len;
1833 err = skb_copy_datagram_msg(skb, rxm->offset,
1839 rxm->offset = rxm->offset + chunk;
1840 rxm->full_len = rxm->full_len - chunk;
1851 /* For async or peek case, queue the current skb */
1852 if (async || is_peek || retain_skb) {
1853 skb_queue_tail(&ctx->rx_list, skb);
1857 if (tls_sw_advance_skb(sk, skb, chunk)) {
1858 /* Return full control message to
1859 * userspace before trying to parse
1860 * another message type
1862 msg->msg_flags |= MSG_EOR;
1863 if (ctx->control != TLS_RECORD_TYPE_DATA)
1872 /* Wait for all previously submitted records to be decrypted */
1873 smp_store_mb(ctx->async_notify, true);
1874 if (atomic_read(&ctx->decrypt_pending)) {
1875 err = crypto_wait_req(-EINPROGRESS, &ctx->async_wait);
1877 /* one of async decrypt failed */
1878 tls_err_abort(sk, err);
1884 reinit_completion(&ctx->async_wait.completion);
1886 WRITE_ONCE(ctx->async_notify, false);
1888 /* Drain records from the rx_list & copy if required */
1889 if (is_peek || is_kvec)
1890 err = process_rx_list(ctx, msg, &control, &cmsg, copied,
1891 decrypted, false, is_peek);
1893 err = process_rx_list(ctx, msg, &control, &cmsg, 0,
1894 decrypted, true, is_peek);
1896 tls_err_abort(sk, err);
1902 copied += decrypted;
1907 sk_psock_put(sk, psock);
1908 return copied ? : err;
1911 ssize_t tls_sw_splice_read(struct socket *sock, loff_t *ppos,
1912 struct pipe_inode_info *pipe,
1913 size_t len, unsigned int flags)
1915 struct tls_context *tls_ctx = tls_get_ctx(sock->sk);
1916 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1917 struct strp_msg *rxm = NULL;
1918 struct sock *sk = sock->sk;
1919 struct sk_buff *skb;
1928 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1930 skb = tls_wait_data(sk, NULL, flags, timeo, &err);
1932 goto splice_read_end;
1934 if (!ctx->decrypted) {
1935 err = decrypt_skb_update(sk, skb, NULL, &chunk, &zc, false);
1937 /* splice does not support reading control messages */
1938 if (ctx->control != TLS_RECORD_TYPE_DATA) {
1940 goto splice_read_end;
1944 tls_err_abort(sk, EBADMSG);
1945 goto splice_read_end;
1949 rxm = strp_msg(skb);
1951 chunk = min_t(unsigned int, rxm->full_len, len);
1952 copied = skb_splice_bits(skb, sk, rxm->offset, pipe, chunk, flags);
1954 goto splice_read_end;
1956 if (likely(!(flags & MSG_PEEK)))
1957 tls_sw_advance_skb(sk, skb, copied);
1961 return copied ? : err;
1964 bool tls_sw_stream_read(const struct sock *sk)
1966 struct tls_context *tls_ctx = tls_get_ctx(sk);
1967 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1968 bool ingress_empty = true;
1969 struct sk_psock *psock;
1972 psock = sk_psock(sk);
1974 ingress_empty = list_empty(&psock->ingress_msg);
1977 return !ingress_empty || ctx->recv_pkt ||
1978 !skb_queue_empty(&ctx->rx_list);
1981 static int tls_read_size(struct strparser *strp, struct sk_buff *skb)
1983 struct tls_context *tls_ctx = tls_get_ctx(strp->sk);
1984 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1985 struct tls_prot_info *prot = &tls_ctx->prot_info;
1986 char header[TLS_HEADER_SIZE + MAX_IV_SIZE];
1987 struct strp_msg *rxm = strp_msg(skb);
1988 size_t cipher_overhead;
1989 size_t data_len = 0;
1992 /* Verify that we have a full TLS header, or wait for more data */
1993 if (rxm->offset + prot->prepend_size > skb->len)
1996 /* Sanity-check size of on-stack buffer. */
1997 if (WARN_ON(prot->prepend_size > sizeof(header))) {
2002 /* Linearize header to local buffer */
2003 ret = skb_copy_bits(skb, rxm->offset, header, prot->prepend_size);
2008 ctx->control = header[0];
2010 data_len = ((header[4] & 0xFF) | (header[3] << 8));
2012 cipher_overhead = prot->tag_size;
2013 if (prot->version != TLS_1_3_VERSION)
2014 cipher_overhead += prot->iv_size;
2016 if (data_len > TLS_MAX_PAYLOAD_SIZE + cipher_overhead +
2021 if (data_len < cipher_overhead) {
2026 /* Note that both TLS1.3 and TLS1.2 use TLS_1_2 version here */
2027 if (header[1] != TLS_1_2_VERSION_MINOR ||
2028 header[2] != TLS_1_2_VERSION_MAJOR) {
2033 tls_device_rx_resync_new_rec(strp->sk, data_len + TLS_HEADER_SIZE,
2034 TCP_SKB_CB(skb)->seq + rxm->offset);
2035 return data_len + TLS_HEADER_SIZE;
2038 tls_err_abort(strp->sk, ret);
2043 static void tls_queue(struct strparser *strp, struct sk_buff *skb)
2045 struct tls_context *tls_ctx = tls_get_ctx(strp->sk);
2046 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2050 ctx->recv_pkt = skb;
2053 ctx->saved_data_ready(strp->sk);
2056 static void tls_data_ready(struct sock *sk)
2058 struct tls_context *tls_ctx = tls_get_ctx(sk);
2059 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2060 struct sk_psock *psock;
2062 strp_data_ready(&ctx->strp);
2064 psock = sk_psock_get(sk);
2065 if (psock && !list_empty(&psock->ingress_msg)) {
2066 ctx->saved_data_ready(sk);
2067 sk_psock_put(sk, psock);
2071 void tls_sw_cancel_work_tx(struct tls_context *tls_ctx)
2073 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
2075 set_bit(BIT_TX_CLOSING, &ctx->tx_bitmask);
2076 set_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask);
2077 cancel_delayed_work_sync(&ctx->tx_work.work);
2080 void tls_sw_release_resources_tx(struct sock *sk)
2082 struct tls_context *tls_ctx = tls_get_ctx(sk);
2083 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
2084 struct tls_rec *rec, *tmp;
2086 /* Wait for any pending async encryptions to complete */
2087 smp_store_mb(ctx->async_notify, true);
2088 if (atomic_read(&ctx->encrypt_pending))
2089 crypto_wait_req(-EINPROGRESS, &ctx->async_wait);
2091 tls_tx_records(sk, -1);
2093 /* Free up un-sent records in tx_list. First, free
2094 * the partially sent record if any at head of tx_list.
2096 if (tls_free_partial_record(sk, tls_ctx)) {
2097 rec = list_first_entry(&ctx->tx_list,
2098 struct tls_rec, list);
2099 list_del(&rec->list);
2100 sk_msg_free(sk, &rec->msg_plaintext);
2104 list_for_each_entry_safe(rec, tmp, &ctx->tx_list, list) {
2105 list_del(&rec->list);
2106 sk_msg_free(sk, &rec->msg_encrypted);
2107 sk_msg_free(sk, &rec->msg_plaintext);
2111 crypto_free_aead(ctx->aead_send);
2112 tls_free_open_rec(sk);
2115 void tls_sw_free_ctx_tx(struct tls_context *tls_ctx)
2117 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
2122 void tls_sw_release_resources_rx(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);
2127 kfree(tls_ctx->rx.rec_seq);
2128 kfree(tls_ctx->rx.iv);
2130 if (ctx->aead_recv) {
2131 kfree_skb(ctx->recv_pkt);
2132 ctx->recv_pkt = NULL;
2133 skb_queue_purge(&ctx->rx_list);
2134 crypto_free_aead(ctx->aead_recv);
2135 strp_stop(&ctx->strp);
2136 /* If tls_sw_strparser_arm() was not called (cleanup paths)
2137 * we still want to strp_stop(), but sk->sk_data_ready was
2140 if (ctx->saved_data_ready) {
2141 write_lock_bh(&sk->sk_callback_lock);
2142 sk->sk_data_ready = ctx->saved_data_ready;
2143 write_unlock_bh(&sk->sk_callback_lock);
2148 void tls_sw_strparser_done(struct tls_context *tls_ctx)
2150 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2152 strp_done(&ctx->strp);
2155 void tls_sw_free_ctx_rx(struct tls_context *tls_ctx)
2157 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2162 void tls_sw_free_resources_rx(struct sock *sk)
2164 struct tls_context *tls_ctx = tls_get_ctx(sk);
2166 tls_sw_release_resources_rx(sk);
2167 tls_sw_free_ctx_rx(tls_ctx);
2170 /* The work handler to transmitt the encrypted records in tx_list */
2171 static void tx_work_handler(struct work_struct *work)
2173 struct delayed_work *delayed_work = to_delayed_work(work);
2174 struct tx_work *tx_work = container_of(delayed_work,
2175 struct tx_work, work);
2176 struct sock *sk = tx_work->sk;
2177 struct tls_context *tls_ctx = tls_get_ctx(sk);
2178 struct tls_sw_context_tx *ctx;
2180 if (unlikely(!tls_ctx))
2183 ctx = tls_sw_ctx_tx(tls_ctx);
2184 if (test_bit(BIT_TX_CLOSING, &ctx->tx_bitmask))
2187 if (!test_and_clear_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask))
2189 mutex_lock(&tls_ctx->tx_lock);
2191 tls_tx_records(sk, -1);
2193 mutex_unlock(&tls_ctx->tx_lock);
2196 void tls_sw_write_space(struct sock *sk, struct tls_context *ctx)
2198 struct tls_sw_context_tx *tx_ctx = tls_sw_ctx_tx(ctx);
2200 /* Schedule the transmission if tx list is ready */
2201 if (is_tx_ready(tx_ctx) &&
2202 !test_and_set_bit(BIT_TX_SCHEDULED, &tx_ctx->tx_bitmask))
2203 schedule_delayed_work(&tx_ctx->tx_work.work, 0);
2206 void tls_sw_strparser_arm(struct sock *sk, struct tls_context *tls_ctx)
2208 struct tls_sw_context_rx *rx_ctx = tls_sw_ctx_rx(tls_ctx);
2210 write_lock_bh(&sk->sk_callback_lock);
2211 rx_ctx->saved_data_ready = sk->sk_data_ready;
2212 sk->sk_data_ready = tls_data_ready;
2213 write_unlock_bh(&sk->sk_callback_lock);
2215 strp_check_rcv(&rx_ctx->strp);
2218 int tls_set_sw_offload(struct sock *sk, struct tls_context *ctx, int tx)
2220 struct tls_context *tls_ctx = tls_get_ctx(sk);
2221 struct tls_prot_info *prot = &tls_ctx->prot_info;
2222 struct tls_crypto_info *crypto_info;
2223 struct tls12_crypto_info_aes_gcm_128 *gcm_128_info;
2224 struct tls12_crypto_info_aes_gcm_256 *gcm_256_info;
2225 struct tls12_crypto_info_aes_ccm_128 *ccm_128_info;
2226 struct tls_sw_context_tx *sw_ctx_tx = NULL;
2227 struct tls_sw_context_rx *sw_ctx_rx = NULL;
2228 struct cipher_context *cctx;
2229 struct crypto_aead **aead;
2230 struct strp_callbacks cb;
2231 u16 nonce_size, tag_size, iv_size, rec_seq_size, salt_size;
2232 struct crypto_tfm *tfm;
2233 char *iv, *rec_seq, *key, *salt, *cipher_name;
2243 if (!ctx->priv_ctx_tx) {
2244 sw_ctx_tx = kzalloc(sizeof(*sw_ctx_tx), GFP_KERNEL);
2249 ctx->priv_ctx_tx = sw_ctx_tx;
2252 (struct tls_sw_context_tx *)ctx->priv_ctx_tx;
2255 if (!ctx->priv_ctx_rx) {
2256 sw_ctx_rx = kzalloc(sizeof(*sw_ctx_rx), GFP_KERNEL);
2261 ctx->priv_ctx_rx = sw_ctx_rx;
2264 (struct tls_sw_context_rx *)ctx->priv_ctx_rx;
2269 crypto_init_wait(&sw_ctx_tx->async_wait);
2270 crypto_info = &ctx->crypto_send.info;
2272 aead = &sw_ctx_tx->aead_send;
2273 INIT_LIST_HEAD(&sw_ctx_tx->tx_list);
2274 INIT_DELAYED_WORK(&sw_ctx_tx->tx_work.work, tx_work_handler);
2275 sw_ctx_tx->tx_work.sk = sk;
2277 crypto_init_wait(&sw_ctx_rx->async_wait);
2278 crypto_info = &ctx->crypto_recv.info;
2280 skb_queue_head_init(&sw_ctx_rx->rx_list);
2281 aead = &sw_ctx_rx->aead_recv;
2284 switch (crypto_info->cipher_type) {
2285 case TLS_CIPHER_AES_GCM_128: {
2286 nonce_size = TLS_CIPHER_AES_GCM_128_IV_SIZE;
2287 tag_size = TLS_CIPHER_AES_GCM_128_TAG_SIZE;
2288 iv_size = TLS_CIPHER_AES_GCM_128_IV_SIZE;
2289 iv = ((struct tls12_crypto_info_aes_gcm_128 *)crypto_info)->iv;
2290 rec_seq_size = TLS_CIPHER_AES_GCM_128_REC_SEQ_SIZE;
2292 ((struct tls12_crypto_info_aes_gcm_128 *)crypto_info)->rec_seq;
2294 (struct tls12_crypto_info_aes_gcm_128 *)crypto_info;
2295 keysize = TLS_CIPHER_AES_GCM_128_KEY_SIZE;
2296 key = gcm_128_info->key;
2297 salt = gcm_128_info->salt;
2298 salt_size = TLS_CIPHER_AES_GCM_128_SALT_SIZE;
2299 cipher_name = "gcm(aes)";
2302 case TLS_CIPHER_AES_GCM_256: {
2303 nonce_size = TLS_CIPHER_AES_GCM_256_IV_SIZE;
2304 tag_size = TLS_CIPHER_AES_GCM_256_TAG_SIZE;
2305 iv_size = TLS_CIPHER_AES_GCM_256_IV_SIZE;
2306 iv = ((struct tls12_crypto_info_aes_gcm_256 *)crypto_info)->iv;
2307 rec_seq_size = TLS_CIPHER_AES_GCM_256_REC_SEQ_SIZE;
2309 ((struct tls12_crypto_info_aes_gcm_256 *)crypto_info)->rec_seq;
2311 (struct tls12_crypto_info_aes_gcm_256 *)crypto_info;
2312 keysize = TLS_CIPHER_AES_GCM_256_KEY_SIZE;
2313 key = gcm_256_info->key;
2314 salt = gcm_256_info->salt;
2315 salt_size = TLS_CIPHER_AES_GCM_256_SALT_SIZE;
2316 cipher_name = "gcm(aes)";
2319 case TLS_CIPHER_AES_CCM_128: {
2320 nonce_size = TLS_CIPHER_AES_CCM_128_IV_SIZE;
2321 tag_size = TLS_CIPHER_AES_CCM_128_TAG_SIZE;
2322 iv_size = TLS_CIPHER_AES_CCM_128_IV_SIZE;
2323 iv = ((struct tls12_crypto_info_aes_ccm_128 *)crypto_info)->iv;
2324 rec_seq_size = TLS_CIPHER_AES_CCM_128_REC_SEQ_SIZE;
2326 ((struct tls12_crypto_info_aes_ccm_128 *)crypto_info)->rec_seq;
2328 (struct tls12_crypto_info_aes_ccm_128 *)crypto_info;
2329 keysize = TLS_CIPHER_AES_CCM_128_KEY_SIZE;
2330 key = ccm_128_info->key;
2331 salt = ccm_128_info->salt;
2332 salt_size = TLS_CIPHER_AES_CCM_128_SALT_SIZE;
2333 cipher_name = "ccm(aes)";
2341 /* Sanity-check the sizes for stack allocations. */
2342 if (iv_size > MAX_IV_SIZE || nonce_size > MAX_IV_SIZE ||
2343 rec_seq_size > TLS_MAX_REC_SEQ_SIZE) {
2348 if (crypto_info->version == TLS_1_3_VERSION) {
2350 prot->aad_size = TLS_HEADER_SIZE;
2351 prot->tail_size = 1;
2353 prot->aad_size = TLS_AAD_SPACE_SIZE;
2354 prot->tail_size = 0;
2357 prot->version = crypto_info->version;
2358 prot->cipher_type = crypto_info->cipher_type;
2359 prot->prepend_size = TLS_HEADER_SIZE + nonce_size;
2360 prot->tag_size = tag_size;
2361 prot->overhead_size = prot->prepend_size +
2362 prot->tag_size + prot->tail_size;
2363 prot->iv_size = iv_size;
2364 prot->salt_size = salt_size;
2365 cctx->iv = kmalloc(iv_size + salt_size, GFP_KERNEL);
2370 /* Note: 128 & 256 bit salt are the same size */
2371 prot->rec_seq_size = rec_seq_size;
2372 memcpy(cctx->iv, salt, salt_size);
2373 memcpy(cctx->iv + salt_size, iv, iv_size);
2374 cctx->rec_seq = kmemdup(rec_seq, rec_seq_size, GFP_KERNEL);
2375 if (!cctx->rec_seq) {
2381 *aead = crypto_alloc_aead(cipher_name, 0, 0);
2382 if (IS_ERR(*aead)) {
2383 rc = PTR_ERR(*aead);
2389 ctx->push_pending_record = tls_sw_push_pending_record;
2391 rc = crypto_aead_setkey(*aead, key, keysize);
2396 rc = crypto_aead_setauthsize(*aead, prot->tag_size);
2401 tfm = crypto_aead_tfm(sw_ctx_rx->aead_recv);
2403 if (crypto_info->version == TLS_1_3_VERSION)
2404 sw_ctx_rx->async_capable = 0;
2406 sw_ctx_rx->async_capable =
2407 !!(tfm->__crt_alg->cra_flags &
2410 /* Set up strparser */
2411 memset(&cb, 0, sizeof(cb));
2412 cb.rcv_msg = tls_queue;
2413 cb.parse_msg = tls_read_size;
2415 strp_init(&sw_ctx_rx->strp, sk, &cb);
2421 crypto_free_aead(*aead);
2424 kfree(cctx->rec_seq);
2425 cctx->rec_seq = NULL;
2431 kfree(ctx->priv_ctx_tx);
2432 ctx->priv_ctx_tx = NULL;
2434 kfree(ctx->priv_ctx_rx);
2435 ctx->priv_ctx_rx = NULL;