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);
262 atomic_dec(&ctx->decrypt_pending);
267 static void tls_trim_both_msgs(struct sock *sk, int target_size)
269 struct tls_context *tls_ctx = tls_get_ctx(sk);
270 struct tls_prot_info *prot = &tls_ctx->prot_info;
271 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
272 struct tls_rec *rec = ctx->open_rec;
274 sk_msg_trim(sk, &rec->msg_plaintext, target_size);
276 target_size += prot->overhead_size;
277 sk_msg_trim(sk, &rec->msg_encrypted, target_size);
280 static int tls_alloc_encrypted_msg(struct sock *sk, int len)
282 struct tls_context *tls_ctx = tls_get_ctx(sk);
283 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
284 struct tls_rec *rec = ctx->open_rec;
285 struct sk_msg *msg_en = &rec->msg_encrypted;
287 return sk_msg_alloc(sk, msg_en, len, 0);
290 static int tls_clone_plaintext_msg(struct sock *sk, int required)
292 struct tls_context *tls_ctx = tls_get_ctx(sk);
293 struct tls_prot_info *prot = &tls_ctx->prot_info;
294 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
295 struct tls_rec *rec = ctx->open_rec;
296 struct sk_msg *msg_pl = &rec->msg_plaintext;
297 struct sk_msg *msg_en = &rec->msg_encrypted;
300 /* We add page references worth len bytes from encrypted sg
301 * at the end of plaintext sg. It is guaranteed that msg_en
302 * has enough required room (ensured by caller).
304 len = required - msg_pl->sg.size;
306 /* Skip initial bytes in msg_en's data to be able to use
307 * same offset of both plain and encrypted data.
309 skip = prot->prepend_size + msg_pl->sg.size;
311 return sk_msg_clone(sk, msg_pl, msg_en, skip, len);
314 static struct tls_rec *tls_get_rec(struct sock *sk)
316 struct tls_context *tls_ctx = tls_get_ctx(sk);
317 struct tls_prot_info *prot = &tls_ctx->prot_info;
318 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
319 struct sk_msg *msg_pl, *msg_en;
323 mem_size = sizeof(struct tls_rec) + crypto_aead_reqsize(ctx->aead_send);
325 rec = kzalloc(mem_size, sk->sk_allocation);
329 msg_pl = &rec->msg_plaintext;
330 msg_en = &rec->msg_encrypted;
335 sg_init_table(rec->sg_aead_in, 2);
336 sg_set_buf(&rec->sg_aead_in[0], rec->aad_space, prot->aad_size);
337 sg_unmark_end(&rec->sg_aead_in[1]);
339 sg_init_table(rec->sg_aead_out, 2);
340 sg_set_buf(&rec->sg_aead_out[0], rec->aad_space, prot->aad_size);
341 sg_unmark_end(&rec->sg_aead_out[1]);
346 static void tls_free_rec(struct sock *sk, struct tls_rec *rec)
348 sk_msg_free(sk, &rec->msg_encrypted);
349 sk_msg_free(sk, &rec->msg_plaintext);
353 static void tls_free_open_rec(struct sock *sk)
355 struct tls_context *tls_ctx = tls_get_ctx(sk);
356 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
357 struct tls_rec *rec = ctx->open_rec;
360 tls_free_rec(sk, rec);
361 ctx->open_rec = NULL;
365 int tls_tx_records(struct sock *sk, int flags)
367 struct tls_context *tls_ctx = tls_get_ctx(sk);
368 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
369 struct tls_rec *rec, *tmp;
370 struct sk_msg *msg_en;
371 int tx_flags, rc = 0;
373 if (tls_is_partially_sent_record(tls_ctx)) {
374 rec = list_first_entry(&ctx->tx_list,
375 struct tls_rec, list);
378 tx_flags = rec->tx_flags;
382 rc = tls_push_partial_record(sk, tls_ctx, tx_flags);
386 /* Full record has been transmitted.
387 * Remove the head of tx_list
389 list_del(&rec->list);
390 sk_msg_free(sk, &rec->msg_plaintext);
394 /* Tx all ready records */
395 list_for_each_entry_safe(rec, tmp, &ctx->tx_list, list) {
396 if (READ_ONCE(rec->tx_ready)) {
398 tx_flags = rec->tx_flags;
402 msg_en = &rec->msg_encrypted;
403 rc = tls_push_sg(sk, tls_ctx,
404 &msg_en->sg.data[msg_en->sg.curr],
409 list_del(&rec->list);
410 sk_msg_free(sk, &rec->msg_plaintext);
418 if (rc < 0 && rc != -EAGAIN)
419 tls_err_abort(sk, EBADMSG);
424 static void tls_encrypt_done(struct crypto_async_request *req, int err)
426 struct aead_request *aead_req = (struct aead_request *)req;
427 struct sock *sk = req->data;
428 struct tls_context *tls_ctx = tls_get_ctx(sk);
429 struct tls_prot_info *prot = &tls_ctx->prot_info;
430 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
431 struct scatterlist *sge;
432 struct sk_msg *msg_en;
437 rec = container_of(aead_req, struct tls_rec, aead_req);
438 msg_en = &rec->msg_encrypted;
440 sge = sk_msg_elem(msg_en, msg_en->sg.curr);
441 sge->offset -= prot->prepend_size;
442 sge->length += prot->prepend_size;
444 /* Check if error is previously set on socket */
445 if (err || sk->sk_err) {
448 /* If err is already set on socket, return the same code */
450 ctx->async_wait.err = sk->sk_err;
452 ctx->async_wait.err = err;
453 tls_err_abort(sk, err);
458 struct tls_rec *first_rec;
460 /* Mark the record as ready for transmission */
461 smp_store_mb(rec->tx_ready, true);
463 /* If received record is at head of tx_list, schedule tx */
464 first_rec = list_first_entry(&ctx->tx_list,
465 struct tls_rec, list);
466 if (rec == first_rec)
470 pending = atomic_dec_return(&ctx->encrypt_pending);
472 if (!pending && READ_ONCE(ctx->async_notify))
473 complete(&ctx->async_wait.completion);
478 /* Schedule the transmission */
479 if (!test_and_set_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask))
480 schedule_delayed_work(&ctx->tx_work.work, 1);
483 static int tls_do_encryption(struct sock *sk,
484 struct tls_context *tls_ctx,
485 struct tls_sw_context_tx *ctx,
486 struct aead_request *aead_req,
487 size_t data_len, u32 start)
489 struct tls_prot_info *prot = &tls_ctx->prot_info;
490 struct tls_rec *rec = ctx->open_rec;
491 struct sk_msg *msg_en = &rec->msg_encrypted;
492 struct scatterlist *sge = sk_msg_elem(msg_en, start);
493 int rc, iv_offset = 0;
495 /* For CCM based ciphers, first byte of IV is a constant */
496 if (prot->cipher_type == TLS_CIPHER_AES_CCM_128) {
497 rec->iv_data[0] = TLS_AES_CCM_IV_B0_BYTE;
501 memcpy(&rec->iv_data[iv_offset], tls_ctx->tx.iv,
502 prot->iv_size + prot->salt_size);
504 xor_iv_with_seq(prot->version, rec->iv_data, tls_ctx->tx.rec_seq);
506 sge->offset += prot->prepend_size;
507 sge->length -= prot->prepend_size;
509 msg_en->sg.curr = start;
511 aead_request_set_tfm(aead_req, ctx->aead_send);
512 aead_request_set_ad(aead_req, prot->aad_size);
513 aead_request_set_crypt(aead_req, rec->sg_aead_in,
515 data_len, rec->iv_data);
517 aead_request_set_callback(aead_req, CRYPTO_TFM_REQ_MAY_BACKLOG,
518 tls_encrypt_done, sk);
520 /* Add the record in tx_list */
521 list_add_tail((struct list_head *)&rec->list, &ctx->tx_list);
522 atomic_inc(&ctx->encrypt_pending);
524 rc = crypto_aead_encrypt(aead_req);
525 if (!rc || rc != -EINPROGRESS) {
526 atomic_dec(&ctx->encrypt_pending);
527 sge->offset -= prot->prepend_size;
528 sge->length += prot->prepend_size;
532 WRITE_ONCE(rec->tx_ready, true);
533 } else if (rc != -EINPROGRESS) {
534 list_del(&rec->list);
538 /* Unhook the record from context if encryption is not failure */
539 ctx->open_rec = NULL;
540 tls_advance_record_sn(sk, prot, &tls_ctx->tx);
544 static int tls_split_open_record(struct sock *sk, struct tls_rec *from,
545 struct tls_rec **to, struct sk_msg *msg_opl,
546 struct sk_msg *msg_oen, u32 split_point,
547 u32 tx_overhead_size, u32 *orig_end)
549 u32 i, j, bytes = 0, apply = msg_opl->apply_bytes;
550 struct scatterlist *sge, *osge, *nsge;
551 u32 orig_size = msg_opl->sg.size;
552 struct scatterlist tmp = { };
553 struct sk_msg *msg_npl;
557 new = tls_get_rec(sk);
560 ret = sk_msg_alloc(sk, &new->msg_encrypted, msg_opl->sg.size +
561 tx_overhead_size, 0);
563 tls_free_rec(sk, new);
567 *orig_end = msg_opl->sg.end;
568 i = msg_opl->sg.start;
569 sge = sk_msg_elem(msg_opl, i);
570 while (apply && sge->length) {
571 if (sge->length > apply) {
572 u32 len = sge->length - apply;
574 get_page(sg_page(sge));
575 sg_set_page(&tmp, sg_page(sge), len,
576 sge->offset + apply);
581 apply -= sge->length;
582 bytes += sge->length;
585 sk_msg_iter_var_next(i);
586 if (i == msg_opl->sg.end)
588 sge = sk_msg_elem(msg_opl, i);
592 msg_opl->sg.curr = i;
593 msg_opl->sg.copybreak = 0;
594 msg_opl->apply_bytes = 0;
595 msg_opl->sg.size = bytes;
597 msg_npl = &new->msg_plaintext;
598 msg_npl->apply_bytes = apply;
599 msg_npl->sg.size = orig_size - bytes;
601 j = msg_npl->sg.start;
602 nsge = sk_msg_elem(msg_npl, j);
604 memcpy(nsge, &tmp, sizeof(*nsge));
605 sk_msg_iter_var_next(j);
606 nsge = sk_msg_elem(msg_npl, j);
609 osge = sk_msg_elem(msg_opl, i);
610 while (osge->length) {
611 memcpy(nsge, osge, sizeof(*nsge));
613 sk_msg_iter_var_next(i);
614 sk_msg_iter_var_next(j);
617 osge = sk_msg_elem(msg_opl, i);
618 nsge = sk_msg_elem(msg_npl, j);
622 msg_npl->sg.curr = j;
623 msg_npl->sg.copybreak = 0;
629 static void tls_merge_open_record(struct sock *sk, struct tls_rec *to,
630 struct tls_rec *from, u32 orig_end)
632 struct sk_msg *msg_npl = &from->msg_plaintext;
633 struct sk_msg *msg_opl = &to->msg_plaintext;
634 struct scatterlist *osge, *nsge;
638 sk_msg_iter_var_prev(i);
639 j = msg_npl->sg.start;
641 osge = sk_msg_elem(msg_opl, i);
642 nsge = sk_msg_elem(msg_npl, j);
644 if (sg_page(osge) == sg_page(nsge) &&
645 osge->offset + osge->length == nsge->offset) {
646 osge->length += nsge->length;
647 put_page(sg_page(nsge));
650 msg_opl->sg.end = orig_end;
651 msg_opl->sg.curr = orig_end;
652 msg_opl->sg.copybreak = 0;
653 msg_opl->apply_bytes = msg_opl->sg.size + msg_npl->sg.size;
654 msg_opl->sg.size += msg_npl->sg.size;
656 sk_msg_free(sk, &to->msg_encrypted);
657 sk_msg_xfer_full(&to->msg_encrypted, &from->msg_encrypted);
662 static int tls_push_record(struct sock *sk, int flags,
663 unsigned char record_type)
665 struct tls_context *tls_ctx = tls_get_ctx(sk);
666 struct tls_prot_info *prot = &tls_ctx->prot_info;
667 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
668 struct tls_rec *rec = ctx->open_rec, *tmp = NULL;
669 u32 i, split_point, uninitialized_var(orig_end);
670 struct sk_msg *msg_pl, *msg_en;
671 struct aead_request *req;
678 msg_pl = &rec->msg_plaintext;
679 msg_en = &rec->msg_encrypted;
681 split_point = msg_pl->apply_bytes;
682 split = split_point && split_point < msg_pl->sg.size;
683 if (unlikely((!split &&
685 prot->overhead_size > msg_en->sg.size) ||
688 prot->overhead_size > msg_en->sg.size))) {
690 split_point = msg_en->sg.size;
693 rc = tls_split_open_record(sk, rec, &tmp, msg_pl, msg_en,
694 split_point, prot->overhead_size,
698 /* This can happen if above tls_split_open_record allocates
699 * a single large encryption buffer instead of two smaller
700 * ones. In this case adjust pointers and continue without
703 if (!msg_pl->sg.size) {
704 tls_merge_open_record(sk, rec, tmp, orig_end);
705 msg_pl = &rec->msg_plaintext;
706 msg_en = &rec->msg_encrypted;
709 sk_msg_trim(sk, msg_en, msg_pl->sg.size +
710 prot->overhead_size);
713 rec->tx_flags = flags;
714 req = &rec->aead_req;
717 sk_msg_iter_var_prev(i);
719 rec->content_type = record_type;
720 if (prot->version == TLS_1_3_VERSION) {
721 /* Add content type to end of message. No padding added */
722 sg_set_buf(&rec->sg_content_type, &rec->content_type, 1);
723 sg_mark_end(&rec->sg_content_type);
724 sg_chain(msg_pl->sg.data, msg_pl->sg.end + 1,
725 &rec->sg_content_type);
727 sg_mark_end(sk_msg_elem(msg_pl, i));
730 if (msg_pl->sg.end < msg_pl->sg.start) {
731 sg_chain(&msg_pl->sg.data[msg_pl->sg.start],
732 MAX_SKB_FRAGS - msg_pl->sg.start + 1,
736 i = msg_pl->sg.start;
737 sg_chain(rec->sg_aead_in, 2, &msg_pl->sg.data[i]);
740 sk_msg_iter_var_prev(i);
741 sg_mark_end(sk_msg_elem(msg_en, i));
743 i = msg_en->sg.start;
744 sg_chain(rec->sg_aead_out, 2, &msg_en->sg.data[i]);
746 tls_make_aad(rec->aad_space, msg_pl->sg.size + prot->tail_size,
747 tls_ctx->tx.rec_seq, prot->rec_seq_size,
748 record_type, prot->version);
750 tls_fill_prepend(tls_ctx,
751 page_address(sg_page(&msg_en->sg.data[i])) +
752 msg_en->sg.data[i].offset,
753 msg_pl->sg.size + prot->tail_size,
754 record_type, prot->version);
756 tls_ctx->pending_open_record_frags = false;
758 rc = tls_do_encryption(sk, tls_ctx, ctx, req,
759 msg_pl->sg.size + prot->tail_size, i);
761 if (rc != -EINPROGRESS) {
762 tls_err_abort(sk, EBADMSG);
764 tls_ctx->pending_open_record_frags = true;
765 tls_merge_open_record(sk, rec, tmp, orig_end);
768 ctx->async_capable = 1;
771 msg_pl = &tmp->msg_plaintext;
772 msg_en = &tmp->msg_encrypted;
773 sk_msg_trim(sk, msg_en, msg_pl->sg.size + prot->overhead_size);
774 tls_ctx->pending_open_record_frags = true;
778 return tls_tx_records(sk, flags);
781 static int bpf_exec_tx_verdict(struct sk_msg *msg, struct sock *sk,
782 bool full_record, u8 record_type,
783 size_t *copied, int flags)
785 struct tls_context *tls_ctx = tls_get_ctx(sk);
786 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
787 struct sk_msg msg_redir = { };
788 struct sk_psock *psock;
789 struct sock *sk_redir;
795 policy = !(flags & MSG_SENDPAGE_NOPOLICY);
796 psock = sk_psock_get(sk);
797 if (!psock || !policy) {
798 err = tls_push_record(sk, flags, record_type);
799 if (err && err != -EINPROGRESS) {
800 *copied -= sk_msg_free(sk, msg);
801 tls_free_open_rec(sk);
804 sk_psock_put(sk, psock);
808 enospc = sk_msg_full(msg);
809 if (psock->eval == __SK_NONE) {
810 delta = msg->sg.size;
811 psock->eval = sk_psock_msg_verdict(sk, psock, msg);
812 delta -= msg->sg.size;
814 if (msg->cork_bytes && msg->cork_bytes > msg->sg.size &&
815 !enospc && !full_record) {
821 if (msg->apply_bytes && msg->apply_bytes < send)
822 send = msg->apply_bytes;
824 switch (psock->eval) {
826 err = tls_push_record(sk, flags, record_type);
827 if (err && err != -EINPROGRESS) {
828 *copied -= sk_msg_free(sk, msg);
829 tls_free_open_rec(sk);
834 sk_redir = psock->sk_redir;
835 memcpy(&msg_redir, msg, sizeof(*msg));
836 if (msg->apply_bytes < send)
837 msg->apply_bytes = 0;
839 msg->apply_bytes -= send;
840 sk_msg_return_zero(sk, msg, send);
841 msg->sg.size -= send;
843 err = tcp_bpf_sendmsg_redir(sk_redir, &msg_redir, send, flags);
846 *copied -= sk_msg_free_nocharge(sk, &msg_redir);
849 if (msg->sg.size == 0)
850 tls_free_open_rec(sk);
854 sk_msg_free_partial(sk, msg, send);
855 if (msg->apply_bytes < send)
856 msg->apply_bytes = 0;
858 msg->apply_bytes -= send;
859 if (msg->sg.size == 0)
860 tls_free_open_rec(sk);
861 *copied -= (send + delta);
866 bool reset_eval = !ctx->open_rec;
870 msg = &rec->msg_plaintext;
871 if (!msg->apply_bytes)
875 psock->eval = __SK_NONE;
876 if (psock->sk_redir) {
877 sock_put(psock->sk_redir);
878 psock->sk_redir = NULL;
885 sk_psock_put(sk, psock);
889 static int tls_sw_push_pending_record(struct sock *sk, int flags)
891 struct tls_context *tls_ctx = tls_get_ctx(sk);
892 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
893 struct tls_rec *rec = ctx->open_rec;
894 struct sk_msg *msg_pl;
900 msg_pl = &rec->msg_plaintext;
901 copied = msg_pl->sg.size;
905 return bpf_exec_tx_verdict(msg_pl, sk, true, TLS_RECORD_TYPE_DATA,
909 int tls_sw_sendmsg(struct sock *sk, struct msghdr *msg, size_t size)
911 long timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
912 struct tls_context *tls_ctx = tls_get_ctx(sk);
913 struct tls_prot_info *prot = &tls_ctx->prot_info;
914 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
915 bool async_capable = ctx->async_capable;
916 unsigned char record_type = TLS_RECORD_TYPE_DATA;
917 bool is_kvec = iov_iter_is_kvec(&msg->msg_iter);
918 bool eor = !(msg->msg_flags & MSG_MORE);
919 size_t try_to_copy, copied = 0;
920 struct sk_msg *msg_pl, *msg_en;
930 if (msg->msg_flags & ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL))
933 mutex_lock(&tls_ctx->tx_lock);
936 if (unlikely(msg->msg_controllen)) {
937 ret = tls_proccess_cmsg(sk, msg, &record_type);
939 if (ret == -EINPROGRESS)
941 else if (ret != -EAGAIN)
946 while (msg_data_left(msg)) {
955 rec = ctx->open_rec = tls_get_rec(sk);
961 msg_pl = &rec->msg_plaintext;
962 msg_en = &rec->msg_encrypted;
964 orig_size = msg_pl->sg.size;
966 try_to_copy = msg_data_left(msg);
967 record_room = TLS_MAX_PAYLOAD_SIZE - msg_pl->sg.size;
968 if (try_to_copy >= record_room) {
969 try_to_copy = record_room;
973 required_size = msg_pl->sg.size + try_to_copy +
976 if (!sk_stream_memory_free(sk))
977 goto wait_for_sndbuf;
980 ret = tls_alloc_encrypted_msg(sk, required_size);
983 goto wait_for_memory;
985 /* Adjust try_to_copy according to the amount that was
986 * actually allocated. The difference is due
987 * to max sg elements limit
989 try_to_copy -= required_size - msg_en->sg.size;
993 if (!is_kvec && (full_record || eor) && !async_capable) {
994 u32 first = msg_pl->sg.end;
996 ret = sk_msg_zerocopy_from_iter(sk, &msg->msg_iter,
997 msg_pl, try_to_copy);
999 goto fallback_to_reg_send;
1002 copied += try_to_copy;
1004 sk_msg_sg_copy_set(msg_pl, first);
1005 ret = bpf_exec_tx_verdict(msg_pl, sk, full_record,
1006 record_type, &copied,
1009 if (ret == -EINPROGRESS)
1011 else if (ret == -ENOMEM)
1012 goto wait_for_memory;
1013 else if (ctx->open_rec && ret == -ENOSPC)
1015 else if (ret != -EAGAIN)
1020 copied -= try_to_copy;
1021 sk_msg_sg_copy_clear(msg_pl, first);
1022 iov_iter_revert(&msg->msg_iter,
1023 msg_pl->sg.size - orig_size);
1024 fallback_to_reg_send:
1025 sk_msg_trim(sk, msg_pl, orig_size);
1028 required_size = msg_pl->sg.size + try_to_copy;
1030 ret = tls_clone_plaintext_msg(sk, required_size);
1035 /* Adjust try_to_copy according to the amount that was
1036 * actually allocated. The difference is due
1037 * to max sg elements limit
1039 try_to_copy -= required_size - msg_pl->sg.size;
1041 sk_msg_trim(sk, msg_en,
1042 msg_pl->sg.size + prot->overhead_size);
1046 ret = sk_msg_memcopy_from_iter(sk, &msg->msg_iter,
1047 msg_pl, try_to_copy);
1052 /* Open records defined only if successfully copied, otherwise
1053 * we would trim the sg but not reset the open record frags.
1055 tls_ctx->pending_open_record_frags = true;
1056 copied += try_to_copy;
1057 if (full_record || eor) {
1058 ret = bpf_exec_tx_verdict(msg_pl, sk, full_record,
1059 record_type, &copied,
1062 if (ret == -EINPROGRESS)
1064 else if (ret == -ENOMEM)
1065 goto wait_for_memory;
1066 else if (ret != -EAGAIN) {
1077 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1079 ret = sk_stream_wait_memory(sk, &timeo);
1083 tls_trim_both_msgs(sk, orig_size);
1087 if (ctx->open_rec && msg_en->sg.size < required_size)
1088 goto alloc_encrypted;
1093 } else if (num_zc) {
1094 /* Wait for pending encryptions to get completed */
1095 smp_store_mb(ctx->async_notify, true);
1097 if (atomic_read(&ctx->encrypt_pending))
1098 crypto_wait_req(-EINPROGRESS, &ctx->async_wait);
1100 reinit_completion(&ctx->async_wait.completion);
1102 WRITE_ONCE(ctx->async_notify, false);
1104 if (ctx->async_wait.err) {
1105 ret = ctx->async_wait.err;
1110 /* Transmit if any encryptions have completed */
1111 if (test_and_clear_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask)) {
1112 cancel_delayed_work(&ctx->tx_work.work);
1113 tls_tx_records(sk, msg->msg_flags);
1117 ret = sk_stream_error(sk, msg->msg_flags, ret);
1120 mutex_unlock(&tls_ctx->tx_lock);
1121 return copied ? copied : ret;
1124 static int tls_sw_do_sendpage(struct sock *sk, struct page *page,
1125 int offset, size_t size, int flags)
1127 long timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT);
1128 struct tls_context *tls_ctx = tls_get_ctx(sk);
1129 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
1130 struct tls_prot_info *prot = &tls_ctx->prot_info;
1131 unsigned char record_type = TLS_RECORD_TYPE_DATA;
1132 struct sk_msg *msg_pl;
1133 struct tls_rec *rec;
1141 eor = !(flags & (MSG_MORE | MSG_SENDPAGE_NOTLAST));
1142 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
1144 /* Call the sk_stream functions to manage the sndbuf mem. */
1146 size_t copy, required_size;
1154 rec = ctx->open_rec;
1156 rec = ctx->open_rec = tls_get_rec(sk);
1162 msg_pl = &rec->msg_plaintext;
1164 full_record = false;
1165 record_room = TLS_MAX_PAYLOAD_SIZE - msg_pl->sg.size;
1167 if (copy >= record_room) {
1172 required_size = msg_pl->sg.size + copy + prot->overhead_size;
1174 if (!sk_stream_memory_free(sk))
1175 goto wait_for_sndbuf;
1177 ret = tls_alloc_encrypted_msg(sk, required_size);
1180 goto wait_for_memory;
1182 /* Adjust copy according to the amount that was
1183 * actually allocated. The difference is due
1184 * to max sg elements limit
1186 copy -= required_size - msg_pl->sg.size;
1190 sk_msg_page_add(msg_pl, page, copy, offset);
1191 sk_mem_charge(sk, copy);
1197 tls_ctx->pending_open_record_frags = true;
1198 if (full_record || eor || sk_msg_full(msg_pl)) {
1199 ret = bpf_exec_tx_verdict(msg_pl, sk, full_record,
1200 record_type, &copied, flags);
1202 if (ret == -EINPROGRESS)
1204 else if (ret == -ENOMEM)
1205 goto wait_for_memory;
1206 else if (ret != -EAGAIN) {
1215 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1217 ret = sk_stream_wait_memory(sk, &timeo);
1220 tls_trim_both_msgs(sk, msg_pl->sg.size);
1229 /* Transmit if any encryptions have completed */
1230 if (test_and_clear_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask)) {
1231 cancel_delayed_work(&ctx->tx_work.work);
1232 tls_tx_records(sk, flags);
1236 ret = sk_stream_error(sk, flags, ret);
1237 return copied ? copied : ret;
1240 int tls_sw_sendpage_locked(struct sock *sk, struct page *page,
1241 int offset, size_t size, int flags)
1243 if (flags & ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL |
1244 MSG_SENDPAGE_NOTLAST | MSG_SENDPAGE_NOPOLICY |
1245 MSG_NO_SHARED_FRAGS))
1248 return tls_sw_do_sendpage(sk, page, offset, size, flags);
1251 int tls_sw_sendpage(struct sock *sk, struct page *page,
1252 int offset, size_t size, int flags)
1254 struct tls_context *tls_ctx = tls_get_ctx(sk);
1257 if (flags & ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL |
1258 MSG_SENDPAGE_NOTLAST | MSG_SENDPAGE_NOPOLICY))
1261 mutex_lock(&tls_ctx->tx_lock);
1263 ret = tls_sw_do_sendpage(sk, page, offset, size, flags);
1265 mutex_unlock(&tls_ctx->tx_lock);
1269 static struct sk_buff *tls_wait_data(struct sock *sk, struct sk_psock *psock,
1270 int flags, long timeo, int *err)
1272 struct tls_context *tls_ctx = tls_get_ctx(sk);
1273 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1274 struct sk_buff *skb;
1275 DEFINE_WAIT_FUNC(wait, woken_wake_function);
1277 while (!(skb = ctx->recv_pkt) && sk_psock_queue_empty(psock)) {
1279 *err = sock_error(sk);
1283 if (sk->sk_shutdown & RCV_SHUTDOWN)
1286 if (sock_flag(sk, SOCK_DONE))
1289 if ((flags & MSG_DONTWAIT) || !timeo) {
1294 add_wait_queue(sk_sleep(sk), &wait);
1295 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
1296 sk_wait_event(sk, &timeo,
1297 ctx->recv_pkt != skb ||
1298 !sk_psock_queue_empty(psock),
1300 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
1301 remove_wait_queue(sk_sleep(sk), &wait);
1303 /* Handle signals */
1304 if (signal_pending(current)) {
1305 *err = sock_intr_errno(timeo);
1313 static int tls_setup_from_iter(struct sock *sk, struct iov_iter *from,
1314 int length, int *pages_used,
1315 unsigned int *size_used,
1316 struct scatterlist *to,
1319 int rc = 0, i = 0, num_elem = *pages_used, maxpages;
1320 struct page *pages[MAX_SKB_FRAGS];
1321 unsigned int size = *size_used;
1322 ssize_t copied, use;
1325 while (length > 0) {
1327 maxpages = to_max_pages - num_elem;
1328 if (maxpages == 0) {
1332 copied = iov_iter_get_pages(from, pages,
1340 iov_iter_advance(from, copied);
1345 use = min_t(int, copied, PAGE_SIZE - offset);
1347 sg_set_page(&to[num_elem],
1348 pages[i], use, offset);
1349 sg_unmark_end(&to[num_elem]);
1350 /* We do not uncharge memory from this API */
1359 /* Mark the end in the last sg entry if newly added */
1360 if (num_elem > *pages_used)
1361 sg_mark_end(&to[num_elem - 1]);
1364 iov_iter_revert(from, size - *size_used);
1366 *pages_used = num_elem;
1371 /* This function decrypts the input skb into either out_iov or in out_sg
1372 * or in skb buffers itself. The input parameter 'zc' indicates if
1373 * zero-copy mode needs to be tried or not. With zero-copy mode, either
1374 * out_iov or out_sg must be non-NULL. In case both out_iov and out_sg are
1375 * NULL, then the decryption happens inside skb buffers itself, i.e.
1376 * zero-copy gets disabled and 'zc' is updated.
1379 static int decrypt_internal(struct sock *sk, struct sk_buff *skb,
1380 struct iov_iter *out_iov,
1381 struct scatterlist *out_sg,
1382 int *chunk, bool *zc, bool async)
1384 struct tls_context *tls_ctx = tls_get_ctx(sk);
1385 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1386 struct tls_prot_info *prot = &tls_ctx->prot_info;
1387 struct strp_msg *rxm = strp_msg(skb);
1388 int n_sgin, n_sgout, nsg, mem_size, aead_size, err, pages = 0;
1389 struct aead_request *aead_req;
1390 struct sk_buff *unused;
1391 u8 *aad, *iv, *mem = NULL;
1392 struct scatterlist *sgin = NULL;
1393 struct scatterlist *sgout = NULL;
1394 const int data_len = rxm->full_len - prot->overhead_size +
1398 if (*zc && (out_iov || out_sg)) {
1400 n_sgout = iov_iter_npages(out_iov, INT_MAX) + 1;
1402 n_sgout = sg_nents(out_sg);
1403 n_sgin = skb_nsg(skb, rxm->offset + prot->prepend_size,
1404 rxm->full_len - prot->prepend_size);
1408 n_sgin = skb_cow_data(skb, 0, &unused);
1414 /* Increment to accommodate AAD */
1415 n_sgin = n_sgin + 1;
1417 nsg = n_sgin + n_sgout;
1419 aead_size = sizeof(*aead_req) + crypto_aead_reqsize(ctx->aead_recv);
1420 mem_size = aead_size + (nsg * sizeof(struct scatterlist));
1421 mem_size = mem_size + prot->aad_size;
1422 mem_size = mem_size + crypto_aead_ivsize(ctx->aead_recv);
1424 /* Allocate a single block of memory which contains
1425 * aead_req || sgin[] || sgout[] || aad || iv.
1426 * This order achieves correct alignment for aead_req, sgin, sgout.
1428 mem = kmalloc(mem_size, sk->sk_allocation);
1432 /* Segment the allocated memory */
1433 aead_req = (struct aead_request *)mem;
1434 sgin = (struct scatterlist *)(mem + aead_size);
1435 sgout = sgin + n_sgin;
1436 aad = (u8 *)(sgout + n_sgout);
1437 iv = aad + prot->aad_size;
1439 /* For CCM based ciphers, first byte of nonce+iv is always '2' */
1440 if (prot->cipher_type == TLS_CIPHER_AES_CCM_128) {
1446 err = skb_copy_bits(skb, rxm->offset + TLS_HEADER_SIZE,
1447 iv + iv_offset + prot->salt_size,
1453 if (prot->version == TLS_1_3_VERSION)
1454 memcpy(iv + iv_offset, tls_ctx->rx.iv,
1455 crypto_aead_ivsize(ctx->aead_recv));
1457 memcpy(iv + iv_offset, tls_ctx->rx.iv, prot->salt_size);
1459 xor_iv_with_seq(prot->version, iv, tls_ctx->rx.rec_seq);
1462 tls_make_aad(aad, rxm->full_len - prot->overhead_size +
1464 tls_ctx->rx.rec_seq, prot->rec_seq_size,
1465 ctx->control, prot->version);
1468 sg_init_table(sgin, n_sgin);
1469 sg_set_buf(&sgin[0], aad, prot->aad_size);
1470 err = skb_to_sgvec(skb, &sgin[1],
1471 rxm->offset + prot->prepend_size,
1472 rxm->full_len - prot->prepend_size);
1480 sg_init_table(sgout, n_sgout);
1481 sg_set_buf(&sgout[0], aad, prot->aad_size);
1484 err = tls_setup_from_iter(sk, out_iov, data_len,
1485 &pages, chunk, &sgout[1],
1488 goto fallback_to_reg_recv;
1489 } else if (out_sg) {
1490 memcpy(sgout, out_sg, n_sgout * sizeof(*sgout));
1492 goto fallback_to_reg_recv;
1495 fallback_to_reg_recv:
1502 /* Prepare and submit AEAD request */
1503 err = tls_do_decryption(sk, skb, sgin, sgout, iv,
1504 data_len, aead_req, async);
1505 if (err == -EINPROGRESS)
1508 /* Release the pages in case iov was mapped to pages */
1509 for (; pages > 0; pages--)
1510 put_page(sg_page(&sgout[pages]));
1516 static int decrypt_skb_update(struct sock *sk, struct sk_buff *skb,
1517 struct iov_iter *dest, int *chunk, bool *zc,
1520 struct tls_context *tls_ctx = tls_get_ctx(sk);
1521 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1522 struct tls_prot_info *prot = &tls_ctx->prot_info;
1523 struct strp_msg *rxm = strp_msg(skb);
1526 if (!ctx->decrypted) {
1527 if (tls_ctx->rx_conf == TLS_HW) {
1528 err = tls_device_decrypted(sk, tls_ctx, skb, rxm);
1533 /* Still not decrypted after tls_device */
1534 if (!ctx->decrypted) {
1535 err = decrypt_internal(sk, skb, dest, NULL, chunk, zc,
1538 if (err == -EINPROGRESS)
1539 tls_advance_record_sn(sk, prot,
1541 else if (err == -EBADMSG)
1542 TLS_INC_STATS(sock_net(sk),
1543 LINUX_MIB_TLSDECRYPTERROR);
1550 pad = padding_length(ctx, prot, skb);
1554 rxm->full_len -= pad;
1555 rxm->offset += prot->prepend_size;
1556 rxm->full_len -= prot->overhead_size;
1557 tls_advance_record_sn(sk, prot, &tls_ctx->rx);
1559 ctx->saved_data_ready(sk);
1567 int decrypt_skb(struct sock *sk, struct sk_buff *skb,
1568 struct scatterlist *sgout)
1573 return decrypt_internal(sk, skb, NULL, sgout, &chunk, &zc, false);
1576 static bool tls_sw_advance_skb(struct sock *sk, struct sk_buff *skb,
1579 struct tls_context *tls_ctx = tls_get_ctx(sk);
1580 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1583 struct strp_msg *rxm = strp_msg(skb);
1585 if (len < rxm->full_len) {
1587 rxm->full_len -= len;
1593 /* Finished with message */
1594 ctx->recv_pkt = NULL;
1595 __strp_unpause(&ctx->strp);
1600 /* This function traverses the rx_list in tls receive context to copies the
1601 * decrypted records into the buffer provided by caller zero copy is not
1602 * true. Further, the records are removed from the rx_list if it is not a peek
1603 * case and the record has been consumed completely.
1605 static int process_rx_list(struct tls_sw_context_rx *ctx,
1614 struct sk_buff *skb = skb_peek(&ctx->rx_list);
1617 struct tls_msg *tlm;
1620 /* Set the record type in 'control' if caller didn't pass it */
1623 ctrl = tlm->control;
1626 while (skip && skb) {
1627 struct strp_msg *rxm = strp_msg(skb);
1630 /* Cannot process a record of different type */
1631 if (ctrl != tlm->control)
1634 if (skip < rxm->full_len)
1637 skip = skip - rxm->full_len;
1638 skb = skb_peek_next(skb, &ctx->rx_list);
1641 while (len && skb) {
1642 struct sk_buff *next_skb;
1643 struct strp_msg *rxm = strp_msg(skb);
1644 int chunk = min_t(unsigned int, rxm->full_len - skip, len);
1648 /* Cannot process a record of different type */
1649 if (ctrl != tlm->control)
1652 /* Set record type if not already done. For a non-data record,
1653 * do not proceed if record type could not be copied.
1656 int cerr = put_cmsg(msg, SOL_TLS, TLS_GET_RECORD_TYPE,
1657 sizeof(ctrl), &ctrl);
1659 if (ctrl != TLS_RECORD_TYPE_DATA) {
1660 if (cerr || msg->msg_flags & MSG_CTRUNC)
1667 if (!zc || (rxm->full_len - skip) > len) {
1668 int err = skb_copy_datagram_msg(skb, rxm->offset + skip,
1675 copied = copied + chunk;
1677 /* Consume the data from record if it is non-peek case*/
1679 rxm->offset = rxm->offset + chunk;
1680 rxm->full_len = rxm->full_len - chunk;
1682 /* Return if there is unconsumed data in the record */
1683 if (rxm->full_len - skip)
1687 /* The remaining skip-bytes must lie in 1st record in rx_list.
1688 * So from the 2nd record, 'skip' should be 0.
1693 msg->msg_flags |= MSG_EOR;
1695 next_skb = skb_peek_next(skb, &ctx->rx_list);
1698 skb_unlink(skb, &ctx->rx_list);
1709 int tls_sw_recvmsg(struct sock *sk,
1716 struct tls_context *tls_ctx = tls_get_ctx(sk);
1717 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1718 struct tls_prot_info *prot = &tls_ctx->prot_info;
1719 struct sk_psock *psock;
1720 unsigned char control = 0;
1721 ssize_t decrypted = 0;
1722 struct strp_msg *rxm;
1723 struct tls_msg *tlm;
1724 struct sk_buff *skb;
1727 int target, err = 0;
1729 bool is_kvec = iov_iter_is_kvec(&msg->msg_iter);
1730 bool is_peek = flags & MSG_PEEK;
1735 if (unlikely(flags & MSG_ERRQUEUE))
1736 return sock_recv_errqueue(sk, msg, len, SOL_IP, IP_RECVERR);
1738 psock = sk_psock_get(sk);
1741 /* Process pending decrypted records. It must be non-zero-copy */
1742 err = process_rx_list(ctx, msg, &control, &cmsg, 0, len, false,
1745 tls_err_abort(sk, err);
1754 target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
1756 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1758 while (len && (decrypted + copied < target || ctx->recv_pkt)) {
1759 bool retain_skb = false;
1766 skb = tls_wait_data(sk, psock, flags, timeo, &err);
1769 int ret = __tcp_bpf_recvmsg(sk, psock,
1781 if (prot->version == TLS_1_3_VERSION)
1784 tlm->control = ctx->control;
1787 rxm = strp_msg(skb);
1789 to_decrypt = rxm->full_len - prot->overhead_size;
1791 if (to_decrypt <= len && !is_kvec && !is_peek &&
1792 ctx->control == TLS_RECORD_TYPE_DATA &&
1793 prot->version != TLS_1_3_VERSION)
1796 /* Do not use async mode if record is non-data */
1797 if (ctx->control == TLS_RECORD_TYPE_DATA)
1798 async_capable = ctx->async_capable;
1800 async_capable = false;
1802 err = decrypt_skb_update(sk, skb, &msg->msg_iter,
1803 &chunk, &zc, async_capable);
1804 if (err < 0 && err != -EINPROGRESS) {
1805 tls_err_abort(sk, EBADMSG);
1809 if (err == -EINPROGRESS) {
1812 } else if (prot->version == TLS_1_3_VERSION) {
1813 tlm->control = ctx->control;
1816 /* If the type of records being processed is not known yet,
1817 * set it to record type just dequeued. If it is already known,
1818 * but does not match the record type just dequeued, go to end.
1819 * We always get record type here since for tls1.2, record type
1820 * is known just after record is dequeued from stream parser.
1821 * For tls1.3, we disable async.
1825 control = tlm->control;
1826 else if (control != tlm->control)
1832 cerr = put_cmsg(msg, SOL_TLS, TLS_GET_RECORD_TYPE,
1833 sizeof(control), &control);
1835 if (control != TLS_RECORD_TYPE_DATA) {
1836 if (cerr || msg->msg_flags & MSG_CTRUNC) {
1844 goto pick_next_record;
1847 if (rxm->full_len > len) {
1851 chunk = rxm->full_len;
1854 err = skb_copy_datagram_msg(skb, rxm->offset,
1860 rxm->offset = rxm->offset + chunk;
1861 rxm->full_len = rxm->full_len - chunk;
1872 /* For async or peek case, queue the current skb */
1873 if (async || is_peek || retain_skb) {
1874 skb_queue_tail(&ctx->rx_list, skb);
1878 if (tls_sw_advance_skb(sk, skb, chunk)) {
1879 /* Return full control message to
1880 * userspace before trying to parse
1881 * another message type
1883 msg->msg_flags |= MSG_EOR;
1884 if (ctx->control != TLS_RECORD_TYPE_DATA)
1893 /* Wait for all previously submitted records to be decrypted */
1894 smp_store_mb(ctx->async_notify, true);
1895 if (atomic_read(&ctx->decrypt_pending)) {
1896 err = crypto_wait_req(-EINPROGRESS, &ctx->async_wait);
1898 /* one of async decrypt failed */
1899 tls_err_abort(sk, err);
1905 reinit_completion(&ctx->async_wait.completion);
1907 WRITE_ONCE(ctx->async_notify, false);
1909 /* Drain records from the rx_list & copy if required */
1910 if (is_peek || is_kvec)
1911 err = process_rx_list(ctx, msg, &control, &cmsg, copied,
1912 decrypted, false, is_peek);
1914 err = process_rx_list(ctx, msg, &control, &cmsg, 0,
1915 decrypted, true, is_peek);
1917 tls_err_abort(sk, err);
1923 copied += decrypted;
1928 sk_psock_put(sk, psock);
1929 return copied ? : err;
1932 ssize_t tls_sw_splice_read(struct socket *sock, loff_t *ppos,
1933 struct pipe_inode_info *pipe,
1934 size_t len, unsigned int flags)
1936 struct tls_context *tls_ctx = tls_get_ctx(sock->sk);
1937 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1938 struct strp_msg *rxm = NULL;
1939 struct sock *sk = sock->sk;
1940 struct sk_buff *skb;
1949 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1951 skb = tls_wait_data(sk, NULL, flags, timeo, &err);
1953 goto splice_read_end;
1955 if (!ctx->decrypted) {
1956 err = decrypt_skb_update(sk, skb, NULL, &chunk, &zc, false);
1958 /* splice does not support reading control messages */
1959 if (ctx->control != TLS_RECORD_TYPE_DATA) {
1961 goto splice_read_end;
1965 tls_err_abort(sk, EBADMSG);
1966 goto splice_read_end;
1970 rxm = strp_msg(skb);
1972 chunk = min_t(unsigned int, rxm->full_len, len);
1973 copied = skb_splice_bits(skb, sk, rxm->offset, pipe, chunk, flags);
1975 goto splice_read_end;
1977 if (likely(!(flags & MSG_PEEK)))
1978 tls_sw_advance_skb(sk, skb, copied);
1982 return copied ? : err;
1985 bool tls_sw_stream_read(const struct sock *sk)
1987 struct tls_context *tls_ctx = tls_get_ctx(sk);
1988 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1989 bool ingress_empty = true;
1990 struct sk_psock *psock;
1993 psock = sk_psock(sk);
1995 ingress_empty = list_empty(&psock->ingress_msg);
1998 return !ingress_empty || ctx->recv_pkt ||
1999 !skb_queue_empty(&ctx->rx_list);
2002 static int tls_read_size(struct strparser *strp, struct sk_buff *skb)
2004 struct tls_context *tls_ctx = tls_get_ctx(strp->sk);
2005 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2006 struct tls_prot_info *prot = &tls_ctx->prot_info;
2007 char header[TLS_HEADER_SIZE + MAX_IV_SIZE];
2008 struct strp_msg *rxm = strp_msg(skb);
2009 size_t cipher_overhead;
2010 size_t data_len = 0;
2013 /* Verify that we have a full TLS header, or wait for more data */
2014 if (rxm->offset + prot->prepend_size > skb->len)
2017 /* Sanity-check size of on-stack buffer. */
2018 if (WARN_ON(prot->prepend_size > sizeof(header))) {
2023 /* Linearize header to local buffer */
2024 ret = skb_copy_bits(skb, rxm->offset, header, prot->prepend_size);
2029 ctx->control = header[0];
2031 data_len = ((header[4] & 0xFF) | (header[3] << 8));
2033 cipher_overhead = prot->tag_size;
2034 if (prot->version != TLS_1_3_VERSION)
2035 cipher_overhead += prot->iv_size;
2037 if (data_len > TLS_MAX_PAYLOAD_SIZE + cipher_overhead +
2042 if (data_len < cipher_overhead) {
2047 /* Note that both TLS1.3 and TLS1.2 use TLS_1_2 version here */
2048 if (header[1] != TLS_1_2_VERSION_MINOR ||
2049 header[2] != TLS_1_2_VERSION_MAJOR) {
2054 tls_device_rx_resync_new_rec(strp->sk, data_len + TLS_HEADER_SIZE,
2055 TCP_SKB_CB(skb)->seq + rxm->offset);
2056 return data_len + TLS_HEADER_SIZE;
2059 tls_err_abort(strp->sk, ret);
2064 static void tls_queue(struct strparser *strp, struct sk_buff *skb)
2066 struct tls_context *tls_ctx = tls_get_ctx(strp->sk);
2067 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2071 ctx->recv_pkt = skb;
2074 ctx->saved_data_ready(strp->sk);
2077 static void tls_data_ready(struct sock *sk)
2079 struct tls_context *tls_ctx = tls_get_ctx(sk);
2080 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2081 struct sk_psock *psock;
2083 strp_data_ready(&ctx->strp);
2085 psock = sk_psock_get(sk);
2087 if (!list_empty(&psock->ingress_msg))
2088 ctx->saved_data_ready(sk);
2089 sk_psock_put(sk, psock);
2093 void tls_sw_cancel_work_tx(struct tls_context *tls_ctx)
2095 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
2097 set_bit(BIT_TX_CLOSING, &ctx->tx_bitmask);
2098 set_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask);
2099 cancel_delayed_work_sync(&ctx->tx_work.work);
2102 void tls_sw_release_resources_tx(struct sock *sk)
2104 struct tls_context *tls_ctx = tls_get_ctx(sk);
2105 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
2106 struct tls_rec *rec, *tmp;
2108 /* Wait for any pending async encryptions to complete */
2109 smp_store_mb(ctx->async_notify, true);
2110 if (atomic_read(&ctx->encrypt_pending))
2111 crypto_wait_req(-EINPROGRESS, &ctx->async_wait);
2113 tls_tx_records(sk, -1);
2115 /* Free up un-sent records in tx_list. First, free
2116 * the partially sent record if any at head of tx_list.
2118 if (tls_ctx->partially_sent_record) {
2119 tls_free_partial_record(sk, tls_ctx);
2120 rec = list_first_entry(&ctx->tx_list,
2121 struct tls_rec, list);
2122 list_del(&rec->list);
2123 sk_msg_free(sk, &rec->msg_plaintext);
2127 list_for_each_entry_safe(rec, tmp, &ctx->tx_list, list) {
2128 list_del(&rec->list);
2129 sk_msg_free(sk, &rec->msg_encrypted);
2130 sk_msg_free(sk, &rec->msg_plaintext);
2134 crypto_free_aead(ctx->aead_send);
2135 tls_free_open_rec(sk);
2138 void tls_sw_free_ctx_tx(struct tls_context *tls_ctx)
2140 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
2145 void tls_sw_release_resources_rx(struct sock *sk)
2147 struct tls_context *tls_ctx = tls_get_ctx(sk);
2148 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2150 kfree(tls_ctx->rx.rec_seq);
2151 kfree(tls_ctx->rx.iv);
2153 if (ctx->aead_recv) {
2154 kfree_skb(ctx->recv_pkt);
2155 ctx->recv_pkt = NULL;
2156 skb_queue_purge(&ctx->rx_list);
2157 crypto_free_aead(ctx->aead_recv);
2158 strp_stop(&ctx->strp);
2159 /* If tls_sw_strparser_arm() was not called (cleanup paths)
2160 * we still want to strp_stop(), but sk->sk_data_ready was
2163 if (ctx->saved_data_ready) {
2164 write_lock_bh(&sk->sk_callback_lock);
2165 sk->sk_data_ready = ctx->saved_data_ready;
2166 write_unlock_bh(&sk->sk_callback_lock);
2171 void tls_sw_strparser_done(struct tls_context *tls_ctx)
2173 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2175 strp_done(&ctx->strp);
2178 void tls_sw_free_ctx_rx(struct tls_context *tls_ctx)
2180 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2185 void tls_sw_free_resources_rx(struct sock *sk)
2187 struct tls_context *tls_ctx = tls_get_ctx(sk);
2189 tls_sw_release_resources_rx(sk);
2190 tls_sw_free_ctx_rx(tls_ctx);
2193 /* The work handler to transmitt the encrypted records in tx_list */
2194 static void tx_work_handler(struct work_struct *work)
2196 struct delayed_work *delayed_work = to_delayed_work(work);
2197 struct tx_work *tx_work = container_of(delayed_work,
2198 struct tx_work, work);
2199 struct sock *sk = tx_work->sk;
2200 struct tls_context *tls_ctx = tls_get_ctx(sk);
2201 struct tls_sw_context_tx *ctx;
2203 if (unlikely(!tls_ctx))
2206 ctx = tls_sw_ctx_tx(tls_ctx);
2207 if (test_bit(BIT_TX_CLOSING, &ctx->tx_bitmask))
2210 if (!test_and_clear_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask))
2212 mutex_lock(&tls_ctx->tx_lock);
2214 tls_tx_records(sk, -1);
2216 mutex_unlock(&tls_ctx->tx_lock);
2219 void tls_sw_write_space(struct sock *sk, struct tls_context *ctx)
2221 struct tls_sw_context_tx *tx_ctx = tls_sw_ctx_tx(ctx);
2223 /* Schedule the transmission if tx list is ready */
2224 if (is_tx_ready(tx_ctx) &&
2225 !test_and_set_bit(BIT_TX_SCHEDULED, &tx_ctx->tx_bitmask))
2226 schedule_delayed_work(&tx_ctx->tx_work.work, 0);
2229 void tls_sw_strparser_arm(struct sock *sk, struct tls_context *tls_ctx)
2231 struct tls_sw_context_rx *rx_ctx = tls_sw_ctx_rx(tls_ctx);
2233 write_lock_bh(&sk->sk_callback_lock);
2234 rx_ctx->saved_data_ready = sk->sk_data_ready;
2235 sk->sk_data_ready = tls_data_ready;
2236 write_unlock_bh(&sk->sk_callback_lock);
2238 strp_check_rcv(&rx_ctx->strp);
2241 int tls_set_sw_offload(struct sock *sk, struct tls_context *ctx, int tx)
2243 struct tls_context *tls_ctx = tls_get_ctx(sk);
2244 struct tls_prot_info *prot = &tls_ctx->prot_info;
2245 struct tls_crypto_info *crypto_info;
2246 struct tls12_crypto_info_aes_gcm_128 *gcm_128_info;
2247 struct tls12_crypto_info_aes_gcm_256 *gcm_256_info;
2248 struct tls12_crypto_info_aes_ccm_128 *ccm_128_info;
2249 struct tls_sw_context_tx *sw_ctx_tx = NULL;
2250 struct tls_sw_context_rx *sw_ctx_rx = NULL;
2251 struct cipher_context *cctx;
2252 struct crypto_aead **aead;
2253 struct strp_callbacks cb;
2254 u16 nonce_size, tag_size, iv_size, rec_seq_size, salt_size;
2255 struct crypto_tfm *tfm;
2256 char *iv, *rec_seq, *key, *salt, *cipher_name;
2266 if (!ctx->priv_ctx_tx) {
2267 sw_ctx_tx = kzalloc(sizeof(*sw_ctx_tx), GFP_KERNEL);
2272 ctx->priv_ctx_tx = sw_ctx_tx;
2275 (struct tls_sw_context_tx *)ctx->priv_ctx_tx;
2278 if (!ctx->priv_ctx_rx) {
2279 sw_ctx_rx = kzalloc(sizeof(*sw_ctx_rx), GFP_KERNEL);
2284 ctx->priv_ctx_rx = sw_ctx_rx;
2287 (struct tls_sw_context_rx *)ctx->priv_ctx_rx;
2292 crypto_init_wait(&sw_ctx_tx->async_wait);
2293 crypto_info = &ctx->crypto_send.info;
2295 aead = &sw_ctx_tx->aead_send;
2296 INIT_LIST_HEAD(&sw_ctx_tx->tx_list);
2297 INIT_DELAYED_WORK(&sw_ctx_tx->tx_work.work, tx_work_handler);
2298 sw_ctx_tx->tx_work.sk = sk;
2300 crypto_init_wait(&sw_ctx_rx->async_wait);
2301 crypto_info = &ctx->crypto_recv.info;
2303 skb_queue_head_init(&sw_ctx_rx->rx_list);
2304 aead = &sw_ctx_rx->aead_recv;
2307 switch (crypto_info->cipher_type) {
2308 case TLS_CIPHER_AES_GCM_128: {
2309 nonce_size = TLS_CIPHER_AES_GCM_128_IV_SIZE;
2310 tag_size = TLS_CIPHER_AES_GCM_128_TAG_SIZE;
2311 iv_size = TLS_CIPHER_AES_GCM_128_IV_SIZE;
2312 iv = ((struct tls12_crypto_info_aes_gcm_128 *)crypto_info)->iv;
2313 rec_seq_size = TLS_CIPHER_AES_GCM_128_REC_SEQ_SIZE;
2315 ((struct tls12_crypto_info_aes_gcm_128 *)crypto_info)->rec_seq;
2317 (struct tls12_crypto_info_aes_gcm_128 *)crypto_info;
2318 keysize = TLS_CIPHER_AES_GCM_128_KEY_SIZE;
2319 key = gcm_128_info->key;
2320 salt = gcm_128_info->salt;
2321 salt_size = TLS_CIPHER_AES_GCM_128_SALT_SIZE;
2322 cipher_name = "gcm(aes)";
2325 case TLS_CIPHER_AES_GCM_256: {
2326 nonce_size = TLS_CIPHER_AES_GCM_256_IV_SIZE;
2327 tag_size = TLS_CIPHER_AES_GCM_256_TAG_SIZE;
2328 iv_size = TLS_CIPHER_AES_GCM_256_IV_SIZE;
2329 iv = ((struct tls12_crypto_info_aes_gcm_256 *)crypto_info)->iv;
2330 rec_seq_size = TLS_CIPHER_AES_GCM_256_REC_SEQ_SIZE;
2332 ((struct tls12_crypto_info_aes_gcm_256 *)crypto_info)->rec_seq;
2334 (struct tls12_crypto_info_aes_gcm_256 *)crypto_info;
2335 keysize = TLS_CIPHER_AES_GCM_256_KEY_SIZE;
2336 key = gcm_256_info->key;
2337 salt = gcm_256_info->salt;
2338 salt_size = TLS_CIPHER_AES_GCM_256_SALT_SIZE;
2339 cipher_name = "gcm(aes)";
2342 case TLS_CIPHER_AES_CCM_128: {
2343 nonce_size = TLS_CIPHER_AES_CCM_128_IV_SIZE;
2344 tag_size = TLS_CIPHER_AES_CCM_128_TAG_SIZE;
2345 iv_size = TLS_CIPHER_AES_CCM_128_IV_SIZE;
2346 iv = ((struct tls12_crypto_info_aes_ccm_128 *)crypto_info)->iv;
2347 rec_seq_size = TLS_CIPHER_AES_CCM_128_REC_SEQ_SIZE;
2349 ((struct tls12_crypto_info_aes_ccm_128 *)crypto_info)->rec_seq;
2351 (struct tls12_crypto_info_aes_ccm_128 *)crypto_info;
2352 keysize = TLS_CIPHER_AES_CCM_128_KEY_SIZE;
2353 key = ccm_128_info->key;
2354 salt = ccm_128_info->salt;
2355 salt_size = TLS_CIPHER_AES_CCM_128_SALT_SIZE;
2356 cipher_name = "ccm(aes)";
2364 /* Sanity-check the sizes for stack allocations. */
2365 if (iv_size > MAX_IV_SIZE || nonce_size > MAX_IV_SIZE ||
2366 rec_seq_size > TLS_MAX_REC_SEQ_SIZE) {
2371 if (crypto_info->version == TLS_1_3_VERSION) {
2373 prot->aad_size = TLS_HEADER_SIZE;
2374 prot->tail_size = 1;
2376 prot->aad_size = TLS_AAD_SPACE_SIZE;
2377 prot->tail_size = 0;
2380 prot->version = crypto_info->version;
2381 prot->cipher_type = crypto_info->cipher_type;
2382 prot->prepend_size = TLS_HEADER_SIZE + nonce_size;
2383 prot->tag_size = tag_size;
2384 prot->overhead_size = prot->prepend_size +
2385 prot->tag_size + prot->tail_size;
2386 prot->iv_size = iv_size;
2387 prot->salt_size = salt_size;
2388 cctx->iv = kmalloc(iv_size + salt_size, GFP_KERNEL);
2393 /* Note: 128 & 256 bit salt are the same size */
2394 prot->rec_seq_size = rec_seq_size;
2395 memcpy(cctx->iv, salt, salt_size);
2396 memcpy(cctx->iv + salt_size, iv, iv_size);
2397 cctx->rec_seq = kmemdup(rec_seq, rec_seq_size, GFP_KERNEL);
2398 if (!cctx->rec_seq) {
2404 *aead = crypto_alloc_aead(cipher_name, 0, 0);
2405 if (IS_ERR(*aead)) {
2406 rc = PTR_ERR(*aead);
2412 ctx->push_pending_record = tls_sw_push_pending_record;
2414 rc = crypto_aead_setkey(*aead, key, keysize);
2419 rc = crypto_aead_setauthsize(*aead, prot->tag_size);
2424 tfm = crypto_aead_tfm(sw_ctx_rx->aead_recv);
2426 if (crypto_info->version == TLS_1_3_VERSION)
2427 sw_ctx_rx->async_capable = 0;
2429 sw_ctx_rx->async_capable =
2430 !!(tfm->__crt_alg->cra_flags &
2433 /* Set up strparser */
2434 memset(&cb, 0, sizeof(cb));
2435 cb.rcv_msg = tls_queue;
2436 cb.parse_msg = tls_read_size;
2438 strp_init(&sw_ctx_rx->strp, sk, &cb);
2444 crypto_free_aead(*aead);
2447 kfree(cctx->rec_seq);
2448 cctx->rec_seq = NULL;
2454 kfree(ctx->priv_ctx_tx);
2455 ctx->priv_ctx_tx = NULL;
2457 kfree(ctx->priv_ctx_rx);
2458 ctx->priv_ctx_rx = NULL;